* [PATCH] Fix reverse stepping multiple contiguous PC ranges over the line table.
@ 2023-04-27 20:59 Carl Love
2023-05-02 14:15 ` Bruno Larsen
2023-05-03 9:53 ` Bruno Larsen
0 siblings, 2 replies; 40+ messages in thread
From: Carl Love @ 2023-04-27 20:59 UTC (permalink / raw)
To: gdb-patches, Ulrich Weigand, pedro; +Cc: luis.machado, cel, blarsen
GDB maintainers:
The following patch fixes issues on PowerPC with the reverse-step and
reverse-next instructions when there are multiple assignment statements
on the same line and when there are multiple function calls on the same
line. The commit log below discusses these issues in further depth.
The discussion included what the correct operation should be for these
commands based on the GDB documentation. The proposed patch at that
time changed how the commands worked on other platforms such as X86 in
a way they no longer matched the documentation.
The issue is the line table contains multiple entries for the same
source line. The patch adds a function to search the line table to
find the address of the first instruction of a line. When setup up the
reverse stepping range, the function is called to make sure the start
of the range corresponds to the address of the first instruction for
the line.
The following patch fixes the execution of the reveres-step and
reverse-next commands for both senarios of multiple statements on the
same line for PowerPC and aarch64-linux. Unlike the previous patch, it
does not change the operation of the commands on other platforms, i.e.
X86. The patch adds new test cases for both scenarios to verify they
work correctly.
The patch has been tested on PowerPC, Intel X86 and aarch64-linux with
no new regression failures.
Please let me know if the patch is acceptable for mainline. Thanks.
Carl
---------------------------------------------------------------
Fix reverse stepping multiple contiguous PC ranges over the line table.
There are a couple of scenarios where the GDB reverse-step and reverse-next
commands do not work correctly. The first scenario consists of multiple
assignment statements on the same line. A patch was proposed to address the
issue by Luis Machado and briefly discussed on the mailing list in Feb 2021.
https://sourceware.org/pipermail/gdb-patches/2021-February/175678.html
The discussion was revived by Carl Love with regards to fixing the same
issue on PowerPC.
https://sourceware.org/pipermail/gdb-patches/2022-March/186463.html
The patch was not completed and has been on Carl's to do list for some time.
Discussion of a patch to change how the reverse-step and reverse-next
commands submitted by Carl Love was started in thread:
https://sourceware.org/pipermail/gdb-patches/2023-January/195563.html
The patch was withdrawn as it was pointed out the proposed patch would
change the intended behavior of the commands as documented in the GDB
manual. However, it was pointed out by Pedro Alves <pedro@palves.net>
that the reverse-step and reverse-next commands do not work when there
are multiple function calls on the same line. This is a second scenario
where the commands do not work correctly.
The following patch is an extended version of the original patch by
Luis Machado to fix the issues in scenario 1 to also address the issues in
scenario 2.
--------------------------------------------------------
Scenario 1 issue description by Luis Machado:
When running GDB's testsuite on aarch64-linux/Ubuntu 20.04 (also spotted on
the ppc backend), I noticed some failures in gdb.reverse/solib-precsave.exp
and gdb.reverse/solib-reverse.exp.
The failure happens around the following code:
38 b[1] = shr2(17); /* middle part two */
40 b[0] = 6; b[1] = 9; /* generic statement, end part two */
42 shr1 ("message 1\n"); /* shr1 one */
Normal execution:
- step from line 38 will land on line 40.
- step from line 40 will land on line 42.
Reverse execution:
- step from line 42 will land on line 40.
- step from line 40 will land on line 40.
- step from line 40 will land on line 38.
The problem here is that line 40 contains two contiguous but distinct
PC ranges in the line table, like so:
Line 40 - [0x7ec ~ 0x7f4]
Line 40 - [0x7f4 ~ 0x7fc]
The two distinct ranges are generated because GCC started outputting source
column information, which GDB doesn't take into account at the moment.
When stepping forward from line 40, we skip both of these ranges and land on
line 42. When stepping backward from line 42, we stop at the start PC of the
second (or first, going backwards) range of line 40.
This happens because we have this check in infrun.c:process_event_stop_test:
/* When stepping backward, stop at beginning of line range
(unless it's the function entry point, in which case
keep going back to the call point). */
CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
if (stop_pc == ecs->event_thread->control.step_range_start
&& stop_pc != ecs->stop_func_start
&& execution_direction == EXEC_REVERSE)
end_stepping_range (ecs);
else
keep_going (ecs);
Since we've reached ecs->event_thread->control.step_range_start, we stop
stepping backwards.
The right thing to do is to look for adjacent PC ranges for the same line,
until we notice a line change. Then we take that as the start PC of the
range.
Another solution I thought about is to merge the contiguous ranges when
we are reading the line tables. Though I'm not sure if we really want to
process that data as opposed to keeping it as the compiler created, and
then working around that.
The test case gdb.reverse/map-to-same-line.exp is added to test the fix
for the issues in scenario 1.
---------------------------------------------------------
Scenario 2 issue described by Pedro Alves:
The following explanation of the issue was taken from the gdb mailing list
discussion of the withdrawn patch to change the behavior of the reverse-step
and reverse-next commands. Specifically, message from Pedro Alves
<pedro@palves.net> where he demonstrates the issue where you have multiple
function calls on the same source code line:
https://sourceware.org/pipermail/gdb-patches/2023-January/196110.html
The source line looks like:
func1 (); func2 ();
so stepping backwards over that line should always stop at the first
instruction of the line, not in the middle. Let's simplify this.
Here's the full source code of my example:
(gdb) list 1
1 void func1 ()
2 {
3 }
4
5 void func2 ()
6 {
7 }
8
9 int main ()
10 {
11 func1 (); func2 ();
12 }
Compiled with:
$ gcc reverse.c -o reverse -g3 -O0
$ gcc -v
...
gcc version 11.3.0 (Ubuntu 11.3.0-1ubuntu1~22.04)
Now let's debug it with target record, using current gdb git master (f3d8ae90b236),
without your patch:
$ gdb ~/reverse
GNU gdb (GDB) 14.0.50.20230124-git
...
Reading symbols from /home/pedro/reverse...
(gdb) start
Temporary breakpoint 1 at 0x1147: file reverse.c, line 11.
Starting program: /home/pedro/reverse
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
Temporary breakpoint 1, main () at reverse.c:11
11 func1 (); func2 ();
(gdb) record
(gdb) disassemble /s
Dump of assembler code for function main:
reverse.c:
10 {
0x000055555555513f <+0>: endbr64
0x0000555555555143 <+4>: push %rbp
0x0000555555555144 <+5>: mov %rsp,%rbp
11 func1 (); func2 ();
=> 0x0000555555555147 <+8>: mov $0x0,%eax
0x000055555555514c <+13>: call 0x555555555129 <func1>
0x0000555555555151 <+18>: mov $0x0,%eax
0x0000555555555156 <+23>: call 0x555555555134 <func2>
0x000055555555515b <+28>: mov $0x0,%eax
12 }
0x0000555555555160 <+33>: pop %rbp
0x0000555555555161 <+34>: ret
End of assembler dump.
(gdb) n
12 }
So far so good, a "next" stepped over the whole of line 11 and stopped at line 12.
Let's confirm where we are now:
(gdb) disassemble /s
Dump of assembler code for function main:
reverse.c:
10 {
0x000055555555513f <+0>: endbr64
0x0000555555555143 <+4>: push %rbp
0x0000555555555144 <+5>: mov %rsp,%rbp
11 func1 (); func2 ();
0x0000555555555147 <+8>: mov $0x0,%eax
0x000055555555514c <+13>: call 0x555555555129 <func1>
0x0000555555555151 <+18>: mov $0x0,%eax
0x0000555555555156 <+23>: call 0x555555555134 <func2>
0x000055555555515b <+28>: mov $0x0,%eax
12 }
=> 0x0000555555555160 <+33>: pop %rbp
0x0000555555555161 <+34>: ret
End of assembler dump.
Good, we're at the first instruction of line 12.
Now let's undo the "next", with "reverse-next":
(gdb) reverse-next
11 func1 (); func2 ();
Seemingly stopped at line 11. Let's see exactly where:
(gdb) disassemble /s
Dump of assembler code for function main:
reverse.c:
10 {
0x000055555555513f <+0>: endbr64
0x0000555555555143 <+4>: push %rbp
0x0000555555555144 <+5>: mov %rsp,%rbp
11 func1 (); func2 ();
0x0000555555555147 <+8>: mov $0x0,%eax
0x000055555555514c <+13>: call 0x555555555129 <func1>
=> 0x0000555555555151 <+18>: mov $0x0,%eax
0x0000555555555156 <+23>: call 0x555555555134 <func2>
0x000055555555515b <+28>: mov $0x0,%eax
12 }
0x0000555555555160 <+33>: pop %rbp
0x0000555555555161 <+34>: ret
End of assembler dump.
(gdb)
And lo, we stopped in the middle of line 11! That is a bug, we should have
stepped back all the way to the beginning of the line. The "reverse-next"
should have fully undone the prior "next" command.
The test cases gdb.reverse/func-map-to-same-line-no-colum-info.exp and
gdb.reverse/func-map-to-same-line.exp were added to test the fix for scenario
2 when the binary was compiled with and without line table information.
bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28426
Co-authored-by: Luis Machado <luis.machado@arm.com>
Co-authored-by: Carl Love <cel@us.ibm.com>
---
gdb/infrun.c | 57 +++++++
gdb/symtab.c | 49 ++++++
gdb/symtab.h | 16 ++
.../func-map-to-same-line-no-column-info.exp | 135 ++++++++++++++++
.../gdb.reverse/func-map-to-same-line.c | 36 +++++
.../gdb.reverse/func-map-to-same-line.exp | 123 ++++++++++++++
gdb/testsuite/gdb.reverse/map-to-same-line.c | 58 +++++++
.../gdb.reverse/map-to-same-line.exp | 153 ++++++++++++++++++
8 files changed, 627 insertions(+)
create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-line-no-column-info.exp
create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-line.c
create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.c
create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.exp
diff --git a/gdb/infrun.c b/gdb/infrun.c
index 2f1c6cd694b..59374a05471 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -113,6 +113,9 @@ static struct async_event_handler *infrun_async_inferior_event_token;
Starts off as -1, indicating "never enabled/disabled". */
static int infrun_is_async = -1;
+static CORE_ADDR update_line_range_start (CORE_ADDR pc,
+ struct execution_control_state *ecs);
+
/* See infrun.h. */
void
@@ -6768,6 +6771,25 @@ handle_signal_stop (struct execution_control_state *ecs)
process_event_stop_test (ecs);
}
+CORE_ADDR
+update_line_range_start (CORE_ADDR pc, struct execution_control_state *ecs)
+{
+ /* The line table may have multiple entries for the same source code line.
+ Given the PC, check the line table and return the PC that corresponds
+ to the line table entry for the source line that PC is in. */
+ CORE_ADDR start_line_pc = ecs->event_thread->control.step_range_start;
+ gdb::optional<CORE_ADDR> real_range_start;
+
+ /* Call find_line_range_start to get smallest address in the
+ linetable for multiple Line X entries in the line table. */
+ real_range_start = find_line_range_start (pc);
+
+ if (real_range_start.has_value ())
+ start_line_pc = *real_range_start;
+
+ return start_line_pc;
+}
+
/* Come here when we've got some debug event / signal we can explain
(IOW, not a random signal), and test whether it should cause a
stop, or whether we should resume the inferior (transparently).
@@ -7569,6 +7591,28 @@ process_event_stop_test (struct execution_control_state *ecs)
if (stop_pc_sal.is_stmt)
{
+ if (execution_direction == EXEC_REVERSE)
+ {
+ /* We are stepping backwards make sure we have reached the
+ beginning of the line. */
+ CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
+ CORE_ADDR start_line_pc
+ = update_line_range_start (stop_pc, ecs);
+
+ if (stop_pc != start_line_pc)
+ {
+ /* Have not reached the beginning of the source code line.
+ Set a step range. Execution should stop in any function
+ calls we execute back into before reaching the beginning
+ of the line. */
+ ecs->event_thread->control.step_range_start = start_line_pc;
+ ecs->event_thread->control.step_range_end = stop_pc;
+ set_step_info (ecs->event_thread, frame, stop_pc_sal);
+ keep_going (ecs);
+ return;
+ }
+ }
+
/* We are at the start of a statement.
So stop. Note that we don't stop if we step into the middle of a
@@ -7631,6 +7675,19 @@ process_event_stop_test (struct execution_control_state *ecs)
set_step_info (ecs->event_thread, frame, stop_pc_sal);
infrun_debug_printf ("keep going");
+
+ if (execution_direction == EXEC_REVERSE)
+ {
+ CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
+
+ /* Make sure the stop_pc is set to the beginning of the line. */
+ if (stop_pc != ecs->event_thread->control.step_range_start)
+ {
+ stop_pc = update_line_range_start (stop_pc, ecs);
+ ecs->event_thread->control.step_range_start = stop_pc;
+ }
+ }
+
keep_going (ecs);
}
diff --git a/gdb/symtab.c b/gdb/symtab.c
index 27611a34ec4..91d35616eb9 100644
--- a/gdb/symtab.c
+++ b/gdb/symtab.c
@@ -3282,6 +3282,55 @@ find_pc_line (CORE_ADDR pc, int notcurrent)
return sal;
}
+/* Compare two symtab_and_line entries. Return true if both have
+ the same line number and the same symtab pointer. That means we
+ are dealing with two entries from the same line and from the same
+ source file.
+
+ Return false otherwise. */
+
+static bool
+sal_line_symtab_matches_p (const symtab_and_line &sal1,
+ const symtab_and_line &sal2)
+{
+ return (sal1.line == sal2.line && sal1.symtab == sal2.symtab);
+}
+
+/* See symtah.h. */
+
+gdb::optional<CORE_ADDR>
+find_line_range_start (CORE_ADDR pc)
+{
+ struct symtab_and_line current_sal = find_pc_line (pc, 0);
+
+ if (current_sal.line == 0)
+ return {};
+
+ struct symtab_and_line prev_sal = find_pc_line (current_sal.pc - 1, 0);
+
+ /* If the previous entry is for a different line, that means we are already
+ at the entry with the start PC for this line. */
+ if (!sal_line_symtab_matches_p (prev_sal, current_sal))
+ return current_sal.pc;
+
+ /* Otherwise, keep looking for entries for the same line but with
+ smaller PC's. */
+ bool done = false;
+ CORE_ADDR prev_pc;
+ while (!done)
+ {
+ prev_pc = prev_sal.pc;
+
+ prev_sal = find_pc_line (prev_pc - 1, 0);
+
+ /* Did we notice a line change? If so, we are done with the search. */
+ if (!sal_line_symtab_matches_p (prev_sal, current_sal))
+ done = true;
+ }
+
+ return prev_pc;
+}
+
/* See symtab.h. */
struct symtab *
diff --git a/gdb/symtab.h b/gdb/symtab.h
index 404d0ab30a8..f54305636da 100644
--- a/gdb/symtab.h
+++ b/gdb/symtab.h
@@ -2346,6 +2346,22 @@ extern struct symtab_and_line find_pc_line (CORE_ADDR, int);
extern struct symtab_and_line find_pc_sect_line (CORE_ADDR,
struct obj_section *, int);
+/* Given PC, and assuming it is part of a range of addresses that is part of a
+ line, go back through the linetable and find the starting PC of that
+ line.
+
+ For example, suppose we have 3 PC ranges for line X:
+
+ Line X - [0x0 - 0x8]
+ Line X - [0x8 - 0x10]
+ Line X - [0x10 - 0x18]
+
+ If we call the function with PC == 0x14, we want to return 0x0, as that is
+ the starting PC of line X, and the ranges are contiguous.
+*/
+
+extern gdb::optional<CORE_ADDR> find_line_range_start (CORE_ADDR pc);
+
/* Wrapper around find_pc_line to just return the symtab. */
extern struct symtab *find_pc_line_symtab (CORE_ADDR);
diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line-no-column-info.exp b/gdb/testsuite/gdb.reverse/func-map-to-same-line-no-column-info.exp
new file mode 100644
index 00000000000..20529c90fc2
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line-no-column-info.exp
@@ -0,0 +1,135 @@
+# Copyright 2008-2023 Free Software Foundation, Inc.
+
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program. If not, see <http://www.gnu.org/licenses/>. */
+
+# This file is part of the GDB testsuite. It tests reverse stepping.
+# Lots of code borrowed from "step-test.exp".
+
+# This test checks to make sure there is no regression failures for
+# the reverse-next command when stepping back over two functions in
+# the same line.
+
+if ![supports_reverse] {
+ return
+}
+
+# This test uses the gcc no-column-info command.
+if ![is_c_compiler_gcc] {
+ unsupported "gcc is required for this test"
+ return 0
+}
+
+set srcfile func-map-to-same-line.c
+set executable func-map-to-same-line
+
+set options [list debug additional_flags=-gno-column-info]
+
+if {[build_executable "failed to prepare" $executable $srcfile $options] == -1}\
+ {
+ return -1
+}
+
+clean_restart $executable
+
+runto_main
+set target_remote [gdb_is_target_remote]
+
+if [supports_process_record] {
+ # Activate process record/replay.
+ gdb_test_no_output "record" "turn on process record for test1"
+}
+
+# This regression test verifies the reverse-step and reverse-next commands
+# work properly when executing backwards thru a source line containing
+# two function calls on the same source line, i.e. func1 (); func2 ();
+# This test is compiled so the dwarf info not contain the line table
+# information.
+
+# Test 1, reverse-next command
+# Set breakpoint at the line after the function calls.
+set bp_start_reverse_test [gdb_get_line_number "START REVERSE TEST" $srcfile]
+gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
+
+# Continue to break point for reverse-next test.
+# Command definition: reverse-next [count]
+# Run backward to the beginning of the previous line executed in the current
+# (innermost) stack frame. If the line contains function calls, they will be
+# “un-executed” without stopping. Starting from the first line of a function,
+# reverse-next will take you back to the caller of that function, before the
+# function was called, just as the normal next command would take you from
+# the last line of a function back to its return to its caller 2 .
+
+gdb_continue_to_breakpoint \
+ "stopped at command reverse-next test start location" \
+ ".*$srcfile:$bp_start_reverse_test\r\n.*"
+
+# The reverse-next should step all the way back to the beginning of the line,
+# i.e. at the beginning of the func1 call.
+gdb_test "reverse-next" ".*func1 \\(\\); func2 \\(\\);.*" \
+ "reverse-next to line with two functions"
+
+# A reverse-step should step back and stop at the beginning
+# of the previous line b = 2, i.e. not in func1 ().
+gdb_test "reverse-step" ".*b = 2;.*" \
+ "reverse-step to previous line b = 2"
+
+
+# Setup for test 2
+# Go back to the start of the function
+gdb_test "reverse-continue" "a = 1;" "At start of main, setup for test 2"
+
+# Turn off record to clear logs and turn on again
+gdb_test "record stop" "Process record is stopped.*" \
+ "turn off process record for test1"
+gdb_test_no_output "record" "turn on process record for test2"
+
+# Delete all breakpoints and catchpoints.
+delete_breakpoints
+
+
+# Test 2, reverse-step command
+# Set breakpoint at the line after the function calls.
+gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
+
+# Continue to the start of the reverse-step test.
+# Command definition: reverse-step [count]
+# Run the program backward until control reaches the start of a
+# different source line; then stop it, and return control to gdb.
+# Like the step command, reverse-step will only stop at the beginning of a
+# source line. It “un-executes” the previously executed source line. If the
+# previous source line included calls to debuggable functions, reverse-step
+# will step (backward) into the called function, stopping at the beginning
+# of the last statement in the called function (typically a return
+# statement). Also, as with the step command, if non-debuggable functions
+# are called, reverse-step will run thru them backward without stopping.
+
+gdb_continue_to_breakpoint \
+ "stopped at command reverse-step test start location" \
+ ".*$srcfile:$bp_start_reverse_test\r\n.*"
+
+# The first reverse step should take us call of func2 ().
+gdb_test "reverse-step" ".*}.*" \
+ "reverse-step into func2 "
+
+# The second reverse step should take us into func1 ().
+gdb_test "reverse-step" ".*}.*" \
+ "reverse-step into func1 "
+
+# The third reverse step should take us call of func1 ().
+gdb_test "reverse-step" ".*func1 \\(\\); func2 \\(\\);.*" \
+ "reverse-step to line func1(); func2(), at call for func1 "
+
+# The fourth reverse step should take us to b = 2 ().
+gdb_test "reverse-step" ".*b = 2;.*" \
+ "reverse-step to line b = 2 "
diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.c b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
new file mode 100644
index 00000000000..e9787ef9ff5
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
@@ -0,0 +1,36 @@
+/* Copyright 2008-2023 Free Software Foundation, Inc.
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 3 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program. If not, see <http://www.gnu.org/licenses/>.
+
+ This test is used to test the reverse-step and reverse-next instruction
+ execution for a source line that contains multiple function calls. */
+
+void
+func1 ()
+{
+}
+
+void
+func2 ()
+{
+}
+
+int main ()
+{
+ int a, b;
+ a = 1;
+ b = 2;
+ func1 (); func2 ();
+ a = a + b; // START REVERSE TEST
+}
diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
new file mode 100644
index 00000000000..b632a236bbe
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
@@ -0,0 +1,123 @@
+# Copyright 2008-2023 Free Software Foundation, Inc.
+
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program. If not, see <http://www.gnu.org/licenses/>. */
+
+# This file is part of the GDB testsuite. It tests reverse stepping.
+# Lots of code borrowed from "step-test.exp".
+
+# This test checks to make sure there is no regression failures for
+# the reverse-next command when stepping back over two functions in
+# the same line.
+
+if ![supports_reverse] {
+ return
+}
+
+standard_testfile
+
+if { [prepare_for_testing "failed to prepare" $testfile $srcfile] } {
+ return -1
+}
+
+runto_main
+set target_remote [gdb_is_target_remote]
+
+if [supports_process_record] {
+ # Activate process record/replay.
+ gdb_test_no_output "record" "turn on process record for test1"
+}
+
+# This regression test verifies the reverse-step and reverse-next commands
+# work properly when executing backwards thru a source line containing
+# two function calls on the same source line, i.e. func1 (); func2 ();
+# The assumption for this test is the dwarf info contain the column
+# information.
+
+# Test 1, reverse-next command
+# Set breakpoint at the line after the function calls.
+set bp_start_reverse_test [gdb_get_line_number "START REVERSE TEST" $srcfile]
+gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
+
+# Continue to break point for reverse-next test.
+# Command definition: reverse-next [count]
+# Run backward to the beginning of the previous line executed in the current
+# (innermost) stack frame. If the line contains function calls, they will be
+# “un-executed” without stopping. Starting from the first line of a function,
+# reverse-next will take you back to the caller of that function, before the
+# function was called, just as the normal next command would take you from
+# the last line of a function back to its return to its caller 2 .
+
+gdb_continue_to_breakpoint \
+ "stopped at command reverse-next test start location" \
+ ".*$srcfile:$bp_start_reverse_test\r\n.*"
+
+# The reverse-next should step all the way back to the beginning of the line,
+# i.e. at the beginning of the func1 call.
+gdb_test "reverse-next" ".*func1 \\(\\); func2 \\(\\);.*" \
+ "reverse-next to line with two functions"
+
+# A reverse-step should step back and stop at the beginning
+# of the previous line b = 2, i.e. not in func1 ().
+gdb_test "reverse-step" ".*b = 2;.*" \
+ "reverse-step to previous line b = 2"
+
+
+# Setup for test 2
+# Go back to the start of the function
+gdb_test "reverse-continue" "a = 1;" "At start of main, setup for test 2"
+
+# Turn off record to clear logs and turn on again
+gdb_test "record stop" "Process record is stopped.*" \
+ "turn off process record for test1"
+gdb_test_no_output "record" "turn on process record for test2"
+
+# Delete all breakpoints and catchpoints.
+delete_breakpoints
+
+
+# Test 2, reverse-step command
+# Set breakpoint at the line after the function calls.
+gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
+
+# Continue to the start of the reverse-step test.
+# Command definition: reverse-step [count]
+# Run the program backward until control reaches the start of a
+# different source line; then stop it, and return control to gdb.
+# Like the step command, reverse-step will only stop at the beginning of a
+# source line. It “un-executes” the previously executed source line. If the
+# previous source line included calls to debuggable functions, reverse-step
+# will step (backward) into the called function, stopping at the beginning
+# of the last statement in the called function (typically a return
+# statement). Also, as with the step command, if non-debuggable functions
+# are called, reverse-step will run thru them backward without stopping.
+
+gdb_continue_to_breakpoint \
+ "stopped at command reverse-step test start location" \
+ ".*$srcfile:$bp_start_reverse_test\r\n.*"
+
+# The first reverse step should take us call of func2 ().
+gdb_test "reverse-step" ".*}.*" \
+ "reverse-step into func2 "
+
+# The second reverse step should take us into func1 ().
+gdb_test "reverse-step" ".*}.*" \
+ "reverse-step into func1 "
+
+# The third reverse step should take us call of func1 ().
+gdb_test "reverse-step" ".*func1 \\(\\); func2 \\(\\);.*" \
+ "reverse-step to line func1(); func2(), at call for func1 "
+
+# The fourth reverse step should take us to b = 2 ().
+gdb_test "reverse-step" ".*b = 2;.*" \
+ "reverse-step to line b = 2 "
diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.c b/gdb/testsuite/gdb.reverse/map-to-same-line.c
new file mode 100644
index 00000000000..f20d778f40e
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/map-to-same-line.c
@@ -0,0 +1,58 @@
+/* Copyright 2008-2023 Free Software Foundation, Inc.
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 3 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program. If not, see <http://www.gnu.org/licenses/ >. */
+
+/* The purpose of this test is to create a DWARF line table that contains two
+ or more entries for the same line. When stepping (forwards or backwards),
+ GDB should step over the entire line and not just a particular entry in the
+ line table. */
+
+int
+main ()
+{ /* TAG: main prologue */
+ asm ("main_label: .globl main_label");
+ int i = 1, j = 2, k;
+ float f1 = 2.0, f2 = 4.1, f3;
+ const char *str_1 = "foo", *str_2 = "bar", *str_3;
+
+ asm ("line1: .globl line1");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 1 */
+
+ asm ("line2: .globl line2");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 2 */
+
+ asm ("line3: .globl line3");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 3 */
+
+ asm ("line4: .globl line4");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 4 */
+
+ asm ("line5: .globl line5");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 5 */
+
+ asm ("line6: .globl line6");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 6 */
+
+ asm ("line7: .globl line7");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 7 */
+
+ asm ("line8: .globl line8");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 8 */
+
+ asm ("main_return: .globl main_return");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: main return */
+
+ asm ("end_of_sequence: .globl end_of_sequence");
+ return 0; /* TAG: main return */
+}
diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.exp b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
new file mode 100644
index 00000000000..a01579c2a8d
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
@@ -0,0 +1,153 @@
+# Copyright 2008-2023 Free Software Foundation, Inc.
+
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program. If not, see <http://www.gnu.org/licenses/ >.
+
+# When stepping (forwards or backwards), GDB should step over the entire line
+# and not just a particular entry in the line table. This test was added to
+# verify the find_line_range_start function properly sets the step range for a
+# line that consists of multiple statements, i.e. multiple entries in the line
+# table. This test creates a DWARF line table that contains two entries for
+# the same line to do the needed testing.
+
+load_lib dwarf.exp
+
+# This test can only be run on targets which support DWARF-2 and use gas.
+if {![dwarf2_support]} {
+ unsupported "dwarf2 support required for this test"
+ return 0
+}
+
+if [get_compiler_info] {
+ return -1
+}
+
+# The DWARF assembler requires the gcc compiler.
+if ![is_c_compiler_gcc] {
+ unsupported "gcc is required for this test"
+ return 0
+}
+
+# This test suitable only for process record-replay
+if ![supports_process_record] {
+ return
+}
+
+standard_testfile .c .S
+
+if { [prepare_for_testing "failed to prepare" ${testfile} ${srcfile}] } {
+ return -1
+}
+
+set asm_file [standard_output_file $srcfile2]
+Dwarf::assemble $asm_file {
+ global srcdir subdir srcfile
+ declare_labels integer_label L
+
+ # Find start address and length of program
+ lassign [function_range main [list ${srcdir}/${subdir}/$srcfile]] \
+ main_start main_len
+ set main_end "$main_start + $main_len"
+
+ cu {} {
+ compile_unit {
+ {language @DW_LANG_C}
+ {name map-to-same-line.c}
+ {stmt_list $L DW_FORM_sec_offset}
+ {low_pc 0 addr}
+ } {
+ subprogram {
+ {external 1 flag}
+ {name main}
+ {low_pc $main_start addr}
+ {high_pc $main_len DW_FORM_data4}
+ }
+ }
+ }
+
+ lines {version 2 default_is_stmt 1} L {
+ include_dir "${srcdir}/${subdir}"
+ file_name "$srcfile" 1
+
+ # Generate the line table program with distinct source lines being
+ # mapped to the same line entry. Line 1, 5 and 8 contain 1 statement
+ # each. Line 2 contains 2 statements. Line 3 contains 3 statements.
+ program {
+ DW_LNE_set_address $main_start
+ line [gdb_get_line_number "TAG: main prologue"]
+ DW_LNS_copy
+ DW_LNE_set_address line1
+ line [gdb_get_line_number "TAG: line 1" ]
+ DW_LNS_copy
+ DW_LNE_set_address line2
+ line [gdb_get_line_number "TAG: line 2" ]
+ DW_LNS_copy
+ DW_LNE_set_address line3
+ line [gdb_get_line_number "TAG: line 2" ]
+ DW_LNS_copy
+ DW_LNE_set_address line4
+ line [gdb_get_line_number "TAG: line 3" ]
+ DW_LNS_copy
+ DW_LNE_set_address line5
+ line [gdb_get_line_number "TAG: line 3" ]
+ DW_LNS_copy
+ DW_LNE_set_address line6
+ line [gdb_get_line_number "TAG: line 3" ]
+ DW_LNS_copy
+ DW_LNE_set_address line7
+ line [gdb_get_line_number "TAG: line 5" ]
+ DW_LNS_copy
+ DW_LNE_set_address line8
+ line [gdb_get_line_number "TAG: line 8" ]
+ DW_LNS_copy
+ DW_LNE_set_address main_return
+ line [gdb_get_line_number "TAG: main return"]
+ DW_LNS_copy
+ DW_LNE_set_address end_of_sequence
+ DW_LNE_end_sequence
+ }
+ }
+}
+
+if { [prepare_for_testing "failed to prepare" ${testfile} \
+ [list $srcfile $asm_file] {nodebug} ] } {
+ return -1
+}
+
+if ![runto_main] {
+ return -1
+}
+
+# Print the line table
+gdb_test_multiple "maint info line-table ${testfile}" "" {
+ -re "\r\n$decimal\[ \t\]+$decimal\[ \t\]+($hex)\[ \t\]+Y\[^\r\n\]*" {
+ lappend is_stmt $expect_out(1,string)
+ exp_continue
+ }
+ -re -wrap "" {
+ }
+}
+
+# Activate process record/replay
+gdb_test_no_output "record" "turn on process record"
+
+gdb_test "tbreak main_return" "Temporary breakpoint .*" "breakpoint at return"
+gdb_test "continue" "Temporary breakpoint .*" "run to end of main"
+gdb_test "display \$pc" ".*pc =.*" "display pc"
+
+# At this point, GDB has already recorded the execution up until the return
+# statement. Reverse-step and test if GDB transitions between lines in the
+# expected order. It should reverse-step across lines 8, 5, 3, 2 and 1.
+foreach line {8 5 3 2 1} {
+ gdb_test "reverse-step" ".*TAG: line $line.*" "reverse step to line $line"
+}
--
2.37.2
^ permalink raw reply [flat|nested] 40+ messages in thread
* Re: [PATCH] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-04-27 20:59 [PATCH] Fix reverse stepping multiple contiguous PC ranges over the line table Carl Love
@ 2023-05-02 14:15 ` Bruno Larsen
2023-05-02 15:40 ` Carl Love
2023-05-11 15:11 ` Simon Marchi
2023-05-03 9:53 ` Bruno Larsen
1 sibling, 2 replies; 40+ messages in thread
From: Bruno Larsen @ 2023-05-02 14:15 UTC (permalink / raw)
To: Carl Love, gdb-patches, Ulrich Weigand, pedro; +Cc: luis.machado
On 27/04/2023 22:59, Carl Love wrote:
> GDB maintainers:
>
> The following patch fixes issues on PowerPC with the reverse-step and
> reverse-next instructions when there are multiple assignment statements
> on the same line and when there are multiple function calls on the same
> line. The commit log below discusses these issues in further depth.
> The discussion included what the correct operation should be for these
> commands based on the GDB documentation. The proposed patch at that
> time changed how the commands worked on other platforms such as X86 in
> a way they no longer matched the documentation.
>
> The issue is the line table contains multiple entries for the same
> source line. The patch adds a function to search the line table to
> find the address of the first instruction of a line. When setup up the
> reverse stepping range, the function is called to make sure the start
> of the range corresponds to the address of the first instruction for
> the line.
>
> The following patch fixes the execution of the reveres-step and
> reverse-next commands for both senarios of multiple statements on the
> same line for PowerPC and aarch64-linux. Unlike the previous patch, it
> does not change the operation of the commands on other platforms, i.e.
> X86. The patch adds new test cases for both scenarios to verify they
> work correctly.
>
> The patch has been tested on PowerPC, Intel X86 and aarch64-linux with
> no new regression failures.
>
> Please let me know if the patch is acceptable for mainline. Thanks.
>
> Carl
>
> ---------------------------------------------------------------
> Fix reverse stepping multiple contiguous PC ranges over the line table.
>
> There are a couple of scenarios where the GDB reverse-step and reverse-next
> commands do not work correctly. The first scenario consists of multiple
> assignment statements on the same line. A patch was proposed to address the
> issue by Luis Machado and briefly discussed on the mailing list in Feb 2021.
>
> https://sourceware.org/pipermail/gdb-patches/2021-February/175678.html
>
> The discussion was revived by Carl Love with regards to fixing the same
> issue on PowerPC.
>
> https://sourceware.org/pipermail/gdb-patches/2022-March/186463.html
>
> The patch was not completed and has been on Carl's to do list for some time.
>
> Discussion of a patch to change how the reverse-step and reverse-next
> commands submitted by Carl Love was started in thread:
>
> https://sourceware.org/pipermail/gdb-patches/2023-January/195563.html
>
> The patch was withdrawn as it was pointed out the proposed patch would
> change the intended behavior of the commands as documented in the GDB
> manual. However, it was pointed out by Pedro Alves <pedro@palves.net>
> that the reverse-step and reverse-next commands do not work when there
> are multiple function calls on the same line. This is a second scenario
> where the commands do not work correctly.
>
> The following patch is an extended version of the original patch by
> Luis Machado to fix the issues in scenario 1 to also address the issues in
> scenario 2.
>
> --------------------------------------------------------
>
> Scenario 1 issue description by Luis Machado:
>
> When running GDB's testsuite on aarch64-linux/Ubuntu 20.04 (also spotted on
> the ppc backend), I noticed some failures in gdb.reverse/solib-precsave.exp
> and gdb.reverse/solib-reverse.exp.
>
> The failure happens around the following code:
>
> 38 b[1] = shr2(17); /* middle part two */
> 40 b[0] = 6; b[1] = 9; /* generic statement, end part two */
> 42 shr1 ("message 1\n"); /* shr1 one */
>
> Normal execution:
>
> - step from line 38 will land on line 40.
> - step from line 40 will land on line 42.
>
> Reverse execution:
> - step from line 42 will land on line 40.
> - step from line 40 will land on line 40.
> - step from line 40 will land on line 38.
>
> The problem here is that line 40 contains two contiguous but distinct
> PC ranges in the line table, like so:
>
> Line 40 - [0x7ec ~ 0x7f4]
> Line 40 - [0x7f4 ~ 0x7fc]
>
> The two distinct ranges are generated because GCC started outputting source
> column information, which GDB doesn't take into account at the moment.
>
> When stepping forward from line 40, we skip both of these ranges and land on
> line 42. When stepping backward from line 42, we stop at the start PC of the
> second (or first, going backwards) range of line 40.
>
> This happens because we have this check in infrun.c:process_event_stop_test:
>
> /* When stepping backward, stop at beginning of line range
> (unless it's the function entry point, in which case
> keep going back to the call point). */
> CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
> if (stop_pc == ecs->event_thread->control.step_range_start
> && stop_pc != ecs->stop_func_start
> && execution_direction == EXEC_REVERSE)
> end_stepping_range (ecs);
> else
> keep_going (ecs);
>
> Since we've reached ecs->event_thread->control.step_range_start, we stop
> stepping backwards.
>
> The right thing to do is to look for adjacent PC ranges for the same line,
> until we notice a line change. Then we take that as the start PC of the
> range.
>
> Another solution I thought about is to merge the contiguous ranges when
> we are reading the line tables. Though I'm not sure if we really want to
> process that data as opposed to keeping it as the compiler created, and
> then working around that.
>
> The test case gdb.reverse/map-to-same-line.exp is added to test the fix
> for the issues in scenario 1.
>
> ---------------------------------------------------------
>
> Scenario 2 issue described by Pedro Alves:
>
> The following explanation of the issue was taken from the gdb mailing list
> discussion of the withdrawn patch to change the behavior of the reverse-step
> and reverse-next commands. Specifically, message from Pedro Alves
> <pedro@palves.net> where he demonstrates the issue where you have multiple
> function calls on the same source code line:
>
> https://sourceware.org/pipermail/gdb-patches/2023-January/196110.html
>
> The source line looks like:
>
> func1 (); func2 ();
>
> so stepping backwards over that line should always stop at the first
> instruction of the line, not in the middle. Let's simplify this.
>
> Here's the full source code of my example:
>
> (gdb) list 1
> 1 void func1 ()
> 2 {
> 3 }
> 4
> 5 void func2 ()
> 6 {
> 7 }
> 8
> 9 int main ()
> 10 {
> 11 func1 (); func2 ();
> 12 }
>
> Compiled with:
>
> $ gcc reverse.c -o reverse -g3 -O0
> $ gcc -v
> ...
> gcc version 11.3.0 (Ubuntu 11.3.0-1ubuntu1~22.04)
>
> Now let's debug it with target record, using current gdb git master (f3d8ae90b236),
> without your patch:
>
> $ gdb ~/reverse
> GNU gdb (GDB) 14.0.50.20230124-git
> ...
> Reading symbols from /home/pedro/reverse...
> (gdb) start
> Temporary breakpoint 1 at 0x1147: file reverse.c, line 11.
> Starting program: /home/pedro/reverse
> [Thread debugging using libthread_db enabled]
> Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
>
> Temporary breakpoint 1, main () at reverse.c:11
> 11 func1 (); func2 ();
> (gdb) record
>
> (gdb) disassemble /s
> Dump of assembler code for function main:
> reverse.c:
> 10 {
> 0x000055555555513f <+0>: endbr64
> 0x0000555555555143 <+4>: push %rbp
> 0x0000555555555144 <+5>: mov %rsp,%rbp
>
> 11 func1 (); func2 ();
> => 0x0000555555555147 <+8>: mov $0x0,%eax
> 0x000055555555514c <+13>: call 0x555555555129 <func1>
> 0x0000555555555151 <+18>: mov $0x0,%eax
> 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> 0x000055555555515b <+28>: mov $0x0,%eax
>
> 12 }
> 0x0000555555555160 <+33>: pop %rbp
> 0x0000555555555161 <+34>: ret
> End of assembler dump.
>
> (gdb) n
> 12 }
>
> So far so good, a "next" stepped over the whole of line 11 and stopped at line 12.
>
> Let's confirm where we are now:
>
> (gdb) disassemble /s
> Dump of assembler code for function main:
> reverse.c:
> 10 {
> 0x000055555555513f <+0>: endbr64
> 0x0000555555555143 <+4>: push %rbp
> 0x0000555555555144 <+5>: mov %rsp,%rbp
>
> 11 func1 (); func2 ();
> 0x0000555555555147 <+8>: mov $0x0,%eax
> 0x000055555555514c <+13>: call 0x555555555129 <func1>
> 0x0000555555555151 <+18>: mov $0x0,%eax
> 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> 0x000055555555515b <+28>: mov $0x0,%eax
>
> 12 }
> => 0x0000555555555160 <+33>: pop %rbp
> 0x0000555555555161 <+34>: ret
> End of assembler dump.
>
> Good, we're at the first instruction of line 12.
>
> Now let's undo the "next", with "reverse-next":
>
> (gdb) reverse-next
> 11 func1 (); func2 ();
>
> Seemingly stopped at line 11. Let's see exactly where:
>
> (gdb) disassemble /s
> Dump of assembler code for function main:
> reverse.c:
> 10 {
> 0x000055555555513f <+0>: endbr64
> 0x0000555555555143 <+4>: push %rbp
> 0x0000555555555144 <+5>: mov %rsp,%rbp
>
> 11 func1 (); func2 ();
> 0x0000555555555147 <+8>: mov $0x0,%eax
> 0x000055555555514c <+13>: call 0x555555555129 <func1>
> => 0x0000555555555151 <+18>: mov $0x0,%eax
> 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> 0x000055555555515b <+28>: mov $0x0,%eax
>
> 12 }
> 0x0000555555555160 <+33>: pop %rbp
> 0x0000555555555161 <+34>: ret
> End of assembler dump.
> (gdb)
>
> And lo, we stopped in the middle of line 11! That is a bug, we should have
> stepped back all the way to the beginning of the line. The "reverse-next"
> should have fully undone the prior "next" command.
>
> The test cases gdb.reverse/func-map-to-same-line-no-colum-info.exp and
> gdb.reverse/func-map-to-same-line.exp were added to test the fix for scenario
> 2 when the binary was compiled with and without line table information.
>
> bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28426
>
> Co-authored-by: Luis Machado <luis.machado@arm.com>
> Co-authored-by: Carl Love <cel@us.ibm.com>
Hey Carl,
Thanks for working on this. I'm wondering which parts will be part of
the final commit messages and which is just context for the mailing
list, so some clarity would be nice, but that is not a huge deal.
I wanted to test this change, but it doesn't apply anymore on master,
and `git apply --3way` can't figure out how to do it. Which commit did
you use as base (or alternatively, can you rebase it)?
--
Cheers,
Bruno
> ---
> gdb/infrun.c | 57 +++++++
> gdb/symtab.c | 49 ++++++
> gdb/symtab.h | 16 ++
> .../func-map-to-same-line-no-column-info.exp | 135 ++++++++++++++++
> .../gdb.reverse/func-map-to-same-line.c | 36 +++++
> .../gdb.reverse/func-map-to-same-line.exp | 123 ++++++++++++++
> gdb/testsuite/gdb.reverse/map-to-same-line.c | 58 +++++++
> .../gdb.reverse/map-to-same-line.exp | 153 ++++++++++++++++++
> 8 files changed, 627 insertions(+)
> create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-line-no-column-info.exp
> create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.c
> create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.exp
>
> diff --git a/gdb/infrun.c b/gdb/infrun.c
> index 2f1c6cd694b..59374a05471 100644
> --- a/gdb/infrun.c
> +++ b/gdb/infrun.c
> @@ -113,6 +113,9 @@ static struct async_event_handler *infrun_async_inferior_event_token;
> Starts off as -1, indicating "never enabled/disabled". */
> static int infrun_is_async = -1;
>
> +static CORE_ADDR update_line_range_start (CORE_ADDR pc,
> + struct execution_control_state *ecs);
> +
> /* See infrun.h. */
>
> void
> @@ -6768,6 +6771,25 @@ handle_signal_stop (struct execution_control_state *ecs)
> process_event_stop_test (ecs);
> }
>
> +CORE_ADDR
> +update_line_range_start (CORE_ADDR pc, struct execution_control_state *ecs)
> +{
> + /* The line table may have multiple entries for the same source code line.
> + Given the PC, check the line table and return the PC that corresponds
> + to the line table entry for the source line that PC is in. */
> + CORE_ADDR start_line_pc = ecs->event_thread->control.step_range_start;
> + gdb::optional<CORE_ADDR> real_range_start;
> +
> + /* Call find_line_range_start to get smallest address in the
> + linetable for multiple Line X entries in the line table. */
> + real_range_start = find_line_range_start (pc);
> +
> + if (real_range_start.has_value ())
> + start_line_pc = *real_range_start;
> +
> + return start_line_pc;
> +}
> +
> /* Come here when we've got some debug event / signal we can explain
> (IOW, not a random signal), and test whether it should cause a
> stop, or whether we should resume the inferior (transparently).
> @@ -7569,6 +7591,28 @@ process_event_stop_test (struct execution_control_state *ecs)
>
> if (stop_pc_sal.is_stmt)
> {
> + if (execution_direction == EXEC_REVERSE)
> + {
> + /* We are stepping backwards make sure we have reached the
> + beginning of the line. */
> + CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
> + CORE_ADDR start_line_pc
> + = update_line_range_start (stop_pc, ecs);
> +
> + if (stop_pc != start_line_pc)
> + {
> + /* Have not reached the beginning of the source code line.
> + Set a step range. Execution should stop in any function
> + calls we execute back into before reaching the beginning
> + of the line. */
> + ecs->event_thread->control.step_range_start = start_line_pc;
> + ecs->event_thread->control.step_range_end = stop_pc;
> + set_step_info (ecs->event_thread, frame, stop_pc_sal);
> + keep_going (ecs);
> + return;
> + }
> + }
> +
> /* We are at the start of a statement.
>
> So stop. Note that we don't stop if we step into the middle of a
> @@ -7631,6 +7675,19 @@ process_event_stop_test (struct execution_control_state *ecs)
> set_step_info (ecs->event_thread, frame, stop_pc_sal);
>
> infrun_debug_printf ("keep going");
> +
> + if (execution_direction == EXEC_REVERSE)
> + {
> + CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
> +
> + /* Make sure the stop_pc is set to the beginning of the line. */
> + if (stop_pc != ecs->event_thread->control.step_range_start)
> + {
> + stop_pc = update_line_range_start (stop_pc, ecs);
> + ecs->event_thread->control.step_range_start = stop_pc;
> + }
> + }
> +
> keep_going (ecs);
> }
>
> diff --git a/gdb/symtab.c b/gdb/symtab.c
> index 27611a34ec4..91d35616eb9 100644
> --- a/gdb/symtab.c
> +++ b/gdb/symtab.c
> @@ -3282,6 +3282,55 @@ find_pc_line (CORE_ADDR pc, int notcurrent)
> return sal;
> }
>
> +/* Compare two symtab_and_line entries. Return true if both have
> + the same line number and the same symtab pointer. That means we
> + are dealing with two entries from the same line and from the same
> + source file.
> +
> + Return false otherwise. */
> +
> +static bool
> +sal_line_symtab_matches_p (const symtab_and_line &sal1,
> + const symtab_and_line &sal2)
> +{
> + return (sal1.line == sal2.line && sal1.symtab == sal2.symtab);
> +}
> +
> +/* See symtah.h. */
> +
> +gdb::optional<CORE_ADDR>
> +find_line_range_start (CORE_ADDR pc)
> +{
> + struct symtab_and_line current_sal = find_pc_line (pc, 0);
> +
> + if (current_sal.line == 0)
> + return {};
> +
> + struct symtab_and_line prev_sal = find_pc_line (current_sal.pc - 1, 0);
> +
> + /* If the previous entry is for a different line, that means we are already
> + at the entry with the start PC for this line. */
> + if (!sal_line_symtab_matches_p (prev_sal, current_sal))
> + return current_sal.pc;
> +
> + /* Otherwise, keep looking for entries for the same line but with
> + smaller PC's. */
> + bool done = false;
> + CORE_ADDR prev_pc;
> + while (!done)
> + {
> + prev_pc = prev_sal.pc;
> +
> + prev_sal = find_pc_line (prev_pc - 1, 0);
> +
> + /* Did we notice a line change? If so, we are done with the search. */
> + if (!sal_line_symtab_matches_p (prev_sal, current_sal))
> + done = true;
> + }
> +
> + return prev_pc;
> +}
> +
> /* See symtab.h. */
>
> struct symtab *
> diff --git a/gdb/symtab.h b/gdb/symtab.h
> index 404d0ab30a8..f54305636da 100644
> --- a/gdb/symtab.h
> +++ b/gdb/symtab.h
> @@ -2346,6 +2346,22 @@ extern struct symtab_and_line find_pc_line (CORE_ADDR, int);
> extern struct symtab_and_line find_pc_sect_line (CORE_ADDR,
> struct obj_section *, int);
>
> +/* Given PC, and assuming it is part of a range of addresses that is part of a
> + line, go back through the linetable and find the starting PC of that
> + line.
> +
> + For example, suppose we have 3 PC ranges for line X:
> +
> + Line X - [0x0 - 0x8]
> + Line X - [0x8 - 0x10]
> + Line X - [0x10 - 0x18]
> +
> + If we call the function with PC == 0x14, we want to return 0x0, as that is
> + the starting PC of line X, and the ranges are contiguous.
> +*/
> +
> +extern gdb::optional<CORE_ADDR> find_line_range_start (CORE_ADDR pc);
> +
> /* Wrapper around find_pc_line to just return the symtab. */
>
> extern struct symtab *find_pc_line_symtab (CORE_ADDR);
> diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line-no-column-info.exp b/gdb/testsuite/gdb.reverse/func-map-to-same-line-no-column-info.exp
> new file mode 100644
> index 00000000000..20529c90fc2
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line-no-column-info.exp
> @@ -0,0 +1,135 @@
> +# Copyright 2008-2023 Free Software Foundation, Inc.
> +
> +# This program is free software; you can redistribute it and/or modify
> +# it under the terms of the GNU General Public License as published by
> +# the Free Software Foundation; either version 3 of the License, or
> +# (at your option) any later version.
> +#
> +# This program is distributed in the hope that it will be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> +# GNU General Public License for more details.
> +#
> +# You should have received a copy of the GNU General Public License
> +# along with this program. If not, see <http://www.gnu.org/licenses/>. */
> +
> +# This file is part of the GDB testsuite. It tests reverse stepping.
> +# Lots of code borrowed from "step-test.exp".
> +
> +# This test checks to make sure there is no regression failures for
> +# the reverse-next command when stepping back over two functions in
> +# the same line.
> +
> +if ![supports_reverse] {
> + return
> +}
> +
> +# This test uses the gcc no-column-info command.
> +if ![is_c_compiler_gcc] {
> + unsupported "gcc is required for this test"
> + return 0
> +}
> +
> +set srcfile func-map-to-same-line.c
> +set executable func-map-to-same-line
> +
> +set options [list debug additional_flags=-gno-column-info]
> +
> +if {[build_executable "failed to prepare" $executable $srcfile $options] == -1}\
> + {
> + return -1
> +}
> +
> +clean_restart $executable
> +
> +runto_main
> +set target_remote [gdb_is_target_remote]
> +
> +if [supports_process_record] {
> + # Activate process record/replay.
> + gdb_test_no_output "record" "turn on process record for test1"
> +}
> +
> +# This regression test verifies the reverse-step and reverse-next commands
> +# work properly when executing backwards thru a source line containing
> +# two function calls on the same source line, i.e. func1 (); func2 ();
> +# This test is compiled so the dwarf info not contain the line table
> +# information.
> +
> +# Test 1, reverse-next command
> +# Set breakpoint at the line after the function calls.
> +set bp_start_reverse_test [gdb_get_line_number "START REVERSE TEST" $srcfile]
> +gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> +
> +# Continue to break point for reverse-next test.
> +# Command definition: reverse-next [count]
> +# Run backward to the beginning of the previous line executed in the current
> +# (innermost) stack frame. If the line contains function calls, they will be
> +# “un-executed” without stopping. Starting from the first line of a function,
> +# reverse-next will take you back to the caller of that function, before the
> +# function was called, just as the normal next command would take you from
> +# the last line of a function back to its return to its caller 2 .
> +
> +gdb_continue_to_breakpoint \
> + "stopped at command reverse-next test start location" \
> + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> +
> +# The reverse-next should step all the way back to the beginning of the line,
> +# i.e. at the beginning of the func1 call.
> +gdb_test "reverse-next" ".*func1 \\(\\); func2 \\(\\);.*" \
> + "reverse-next to line with two functions"
> +
> +# A reverse-step should step back and stop at the beginning
> +# of the previous line b = 2, i.e. not in func1 ().
> +gdb_test "reverse-step" ".*b = 2;.*" \
> + "reverse-step to previous line b = 2"
> +
> +
> +# Setup for test 2
> +# Go back to the start of the function
> +gdb_test "reverse-continue" "a = 1;" "At start of main, setup for test 2"
> +
> +# Turn off record to clear logs and turn on again
> +gdb_test "record stop" "Process record is stopped.*" \
> + "turn off process record for test1"
> +gdb_test_no_output "record" "turn on process record for test2"
> +
> +# Delete all breakpoints and catchpoints.
> +delete_breakpoints
> +
> +
> +# Test 2, reverse-step command
> +# Set breakpoint at the line after the function calls.
> +gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> +
> +# Continue to the start of the reverse-step test.
> +# Command definition: reverse-step [count]
> +# Run the program backward until control reaches the start of a
> +# different source line; then stop it, and return control to gdb.
> +# Like the step command, reverse-step will only stop at the beginning of a
> +# source line. It “un-executes” the previously executed source line. If the
> +# previous source line included calls to debuggable functions, reverse-step
> +# will step (backward) into the called function, stopping at the beginning
> +# of the last statement in the called function (typically a return
> +# statement). Also, as with the step command, if non-debuggable functions
> +# are called, reverse-step will run thru them backward without stopping.
> +
> +gdb_continue_to_breakpoint \
> + "stopped at command reverse-step test start location" \
> + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> +
> +# The first reverse step should take us call of func2 ().
> +gdb_test "reverse-step" ".*}.*" \
> + "reverse-step into func2 "
> +
> +# The second reverse step should take us into func1 ().
> +gdb_test "reverse-step" ".*}.*" \
> + "reverse-step into func1 "
> +
> +# The third reverse step should take us call of func1 ().
> +gdb_test "reverse-step" ".*func1 \\(\\); func2 \\(\\);.*" \
> + "reverse-step to line func1(); func2(), at call for func1 "
> +
> +# The fourth reverse step should take us to b = 2 ().
> +gdb_test "reverse-step" ".*b = 2;.*" \
> + "reverse-step to line b = 2 "
> diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.c b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> new file mode 100644
> index 00000000000..e9787ef9ff5
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> @@ -0,0 +1,36 @@
> +/* Copyright 2008-2023 Free Software Foundation, Inc.
> +
> + This program is free software; you can redistribute it and/or modify
> + it under the terms of the GNU General Public License as published by
> + the Free Software Foundation; either version 3 of the License, or
> + (at your option) any later version.
> +
> + This program is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> + GNU General Public License for more details.
> +
> + You should have received a copy of the GNU General Public License
> + along with this program. If not, see <http://www.gnu.org/licenses/>.
> +
> + This test is used to test the reverse-step and reverse-next instruction
> + execution for a source line that contains multiple function calls. */
> +
> +void
> +func1 ()
> +{
> +}
> +
> +void
> +func2 ()
> +{
> +}
> +
> +int main ()
> +{
> + int a, b;
> + a = 1;
> + b = 2;
> + func1 (); func2 ();
> + a = a + b; // START REVERSE TEST
> +}
> diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> new file mode 100644
> index 00000000000..b632a236bbe
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> @@ -0,0 +1,123 @@
> +# Copyright 2008-2023 Free Software Foundation, Inc.
> +
> +# This program is free software; you can redistribute it and/or modify
> +# it under the terms of the GNU General Public License as published by
> +# the Free Software Foundation; either version 3 of the License, or
> +# (at your option) any later version.
> +#
> +# This program is distributed in the hope that it will be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> +# GNU General Public License for more details.
> +#
> +# You should have received a copy of the GNU General Public License
> +# along with this program. If not, see <http://www.gnu.org/licenses/>. */
> +
> +# This file is part of the GDB testsuite. It tests reverse stepping.
> +# Lots of code borrowed from "step-test.exp".
> +
> +# This test checks to make sure there is no regression failures for
> +# the reverse-next command when stepping back over two functions in
> +# the same line.
> +
> +if ![supports_reverse] {
> + return
> +}
> +
> +standard_testfile
> +
> +if { [prepare_for_testing "failed to prepare" $testfile $srcfile] } {
> + return -1
> +}
> +
> +runto_main
> +set target_remote [gdb_is_target_remote]
> +
> +if [supports_process_record] {
> + # Activate process record/replay.
> + gdb_test_no_output "record" "turn on process record for test1"
> +}
> +
> +# This regression test verifies the reverse-step and reverse-next commands
> +# work properly when executing backwards thru a source line containing
> +# two function calls on the same source line, i.e. func1 (); func2 ();
> +# The assumption for this test is the dwarf info contain the column
> +# information.
> +
> +# Test 1, reverse-next command
> +# Set breakpoint at the line after the function calls.
> +set bp_start_reverse_test [gdb_get_line_number "START REVERSE TEST" $srcfile]
> +gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> +
> +# Continue to break point for reverse-next test.
> +# Command definition: reverse-next [count]
> +# Run backward to the beginning of the previous line executed in the current
> +# (innermost) stack frame. If the line contains function calls, they will be
> +# “un-executed” without stopping. Starting from the first line of a function,
> +# reverse-next will take you back to the caller of that function, before the
> +# function was called, just as the normal next command would take you from
> +# the last line of a function back to its return to its caller 2 .
> +
> +gdb_continue_to_breakpoint \
> + "stopped at command reverse-next test start location" \
> + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> +
> +# The reverse-next should step all the way back to the beginning of the line,
> +# i.e. at the beginning of the func1 call.
> +gdb_test "reverse-next" ".*func1 \\(\\); func2 \\(\\);.*" \
> + "reverse-next to line with two functions"
> +
> +# A reverse-step should step back and stop at the beginning
> +# of the previous line b = 2, i.e. not in func1 ().
> +gdb_test "reverse-step" ".*b = 2;.*" \
> + "reverse-step to previous line b = 2"
> +
> +
> +# Setup for test 2
> +# Go back to the start of the function
> +gdb_test "reverse-continue" "a = 1;" "At start of main, setup for test 2"
> +
> +# Turn off record to clear logs and turn on again
> +gdb_test "record stop" "Process record is stopped.*" \
> + "turn off process record for test1"
> +gdb_test_no_output "record" "turn on process record for test2"
> +
> +# Delete all breakpoints and catchpoints.
> +delete_breakpoints
> +
> +
> +# Test 2, reverse-step command
> +# Set breakpoint at the line after the function calls.
> +gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> +
> +# Continue to the start of the reverse-step test.
> +# Command definition: reverse-step [count]
> +# Run the program backward until control reaches the start of a
> +# different source line; then stop it, and return control to gdb.
> +# Like the step command, reverse-step will only stop at the beginning of a
> +# source line. It “un-executes” the previously executed source line. If the
> +# previous source line included calls to debuggable functions, reverse-step
> +# will step (backward) into the called function, stopping at the beginning
> +# of the last statement in the called function (typically a return
> +# statement). Also, as with the step command, if non-debuggable functions
> +# are called, reverse-step will run thru them backward without stopping.
> +
> +gdb_continue_to_breakpoint \
> + "stopped at command reverse-step test start location" \
> + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> +
> +# The first reverse step should take us call of func2 ().
> +gdb_test "reverse-step" ".*}.*" \
> + "reverse-step into func2 "
> +
> +# The second reverse step should take us into func1 ().
> +gdb_test "reverse-step" ".*}.*" \
> + "reverse-step into func1 "
> +
> +# The third reverse step should take us call of func1 ().
> +gdb_test "reverse-step" ".*func1 \\(\\); func2 \\(\\);.*" \
> + "reverse-step to line func1(); func2(), at call for func1 "
> +
> +# The fourth reverse step should take us to b = 2 ().
> +gdb_test "reverse-step" ".*b = 2;.*" \
> + "reverse-step to line b = 2 "
> diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.c b/gdb/testsuite/gdb.reverse/map-to-same-line.c
> new file mode 100644
> index 00000000000..f20d778f40e
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/map-to-same-line.c
> @@ -0,0 +1,58 @@
> +/* Copyright 2008-2023 Free Software Foundation, Inc.
> +
> + This program is free software; you can redistribute it and/or modify
> + it under the terms of the GNU General Public License as published by
> + the Free Software Foundation; either version 3 of the License, or
> + (at your option) any later version.
> +
> + This program is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> + GNU General Public License for more details.
> +
> + You should have received a copy of the GNU General Public License
> + along with this program. If not, see <http://www.gnu.org/licenses/ >. */
> +
> +/* The purpose of this test is to create a DWARF line table that contains two
> + or more entries for the same line. When stepping (forwards or backwards),
> + GDB should step over the entire line and not just a particular entry in the
> + line table. */
> +
> +int
> +main ()
> +{ /* TAG: main prologue */
> + asm ("main_label: .globl main_label");
> + int i = 1, j = 2, k;
> + float f1 = 2.0, f2 = 4.1, f3;
> + const char *str_1 = "foo", *str_2 = "bar", *str_3;
> +
> + asm ("line1: .globl line1");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 1 */
> +
> + asm ("line2: .globl line2");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 2 */
> +
> + asm ("line3: .globl line3");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 3 */
> +
> + asm ("line4: .globl line4");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 4 */
> +
> + asm ("line5: .globl line5");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 5 */
> +
> + asm ("line6: .globl line6");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 6 */
> +
> + asm ("line7: .globl line7");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 7 */
> +
> + asm ("line8: .globl line8");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 8 */
> +
> + asm ("main_return: .globl main_return");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: main return */
> +
> + asm ("end_of_sequence: .globl end_of_sequence");
> + return 0; /* TAG: main return */
> +}
> diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.exp b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> new file mode 100644
> index 00000000000..a01579c2a8d
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> @@ -0,0 +1,153 @@
> +# Copyright 2008-2023 Free Software Foundation, Inc.
> +
> +# This program is free software; you can redistribute it and/or modify
> +# it under the terms of the GNU General Public License as published by
> +# the Free Software Foundation; either version 3 of the License, or
> +# (at your option) any later version.
> +#
> +# This program is distributed in the hope that it will be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> +# GNU General Public License for more details.
> +#
> +# You should have received a copy of the GNU General Public License
> +# along with this program. If not, see <http://www.gnu.org/licenses/ >.
> +
> +# When stepping (forwards or backwards), GDB should step over the entire line
> +# and not just a particular entry in the line table. This test was added to
> +# verify the find_line_range_start function properly sets the step range for a
> +# line that consists of multiple statements, i.e. multiple entries in the line
> +# table. This test creates a DWARF line table that contains two entries for
> +# the same line to do the needed testing.
> +
> +load_lib dwarf.exp
> +
> +# This test can only be run on targets which support DWARF-2 and use gas.
> +if {![dwarf2_support]} {
> + unsupported "dwarf2 support required for this test"
> + return 0
> +}
> +
> +if [get_compiler_info] {
> + return -1
> +}
> +
> +# The DWARF assembler requires the gcc compiler.
> +if ![is_c_compiler_gcc] {
> + unsupported "gcc is required for this test"
> + return 0
> +}
> +
> +# This test suitable only for process record-replay
> +if ![supports_process_record] {
> + return
> +}
> +
> +standard_testfile .c .S
> +
> +if { [prepare_for_testing "failed to prepare" ${testfile} ${srcfile}] } {
> + return -1
> +}
> +
> +set asm_file [standard_output_file $srcfile2]
> +Dwarf::assemble $asm_file {
> + global srcdir subdir srcfile
> + declare_labels integer_label L
> +
> + # Find start address and length of program
> + lassign [function_range main [list ${srcdir}/${subdir}/$srcfile]] \
> + main_start main_len
> + set main_end "$main_start + $main_len"
> +
> + cu {} {
> + compile_unit {
> + {language @DW_LANG_C}
> + {name map-to-same-line.c}
> + {stmt_list $L DW_FORM_sec_offset}
> + {low_pc 0 addr}
> + } {
> + subprogram {
> + {external 1 flag}
> + {name main}
> + {low_pc $main_start addr}
> + {high_pc $main_len DW_FORM_data4}
> + }
> + }
> + }
> +
> + lines {version 2 default_is_stmt 1} L {
> + include_dir "${srcdir}/${subdir}"
> + file_name "$srcfile" 1
> +
> + # Generate the line table program with distinct source lines being
> + # mapped to the same line entry. Line 1, 5 and 8 contain 1 statement
> + # each. Line 2 contains 2 statements. Line 3 contains 3 statements.
> + program {
> + DW_LNE_set_address $main_start
> + line [gdb_get_line_number "TAG: main prologue"]
> + DW_LNS_copy
> + DW_LNE_set_address line1
> + line [gdb_get_line_number "TAG: line 1" ]
> + DW_LNS_copy
> + DW_LNE_set_address line2
> + line [gdb_get_line_number "TAG: line 2" ]
> + DW_LNS_copy
> + DW_LNE_set_address line3
> + line [gdb_get_line_number "TAG: line 2" ]
> + DW_LNS_copy
> + DW_LNE_set_address line4
> + line [gdb_get_line_number "TAG: line 3" ]
> + DW_LNS_copy
> + DW_LNE_set_address line5
> + line [gdb_get_line_number "TAG: line 3" ]
> + DW_LNS_copy
> + DW_LNE_set_address line6
> + line [gdb_get_line_number "TAG: line 3" ]
> + DW_LNS_copy
> + DW_LNE_set_address line7
> + line [gdb_get_line_number "TAG: line 5" ]
> + DW_LNS_copy
> + DW_LNE_set_address line8
> + line [gdb_get_line_number "TAG: line 8" ]
> + DW_LNS_copy
> + DW_LNE_set_address main_return
> + line [gdb_get_line_number "TAG: main return"]
> + DW_LNS_copy
> + DW_LNE_set_address end_of_sequence
> + DW_LNE_end_sequence
> + }
> + }
> +}
> +
> +if { [prepare_for_testing "failed to prepare" ${testfile} \
> + [list $srcfile $asm_file] {nodebug} ] } {
> + return -1
> +}
> +
> +if ![runto_main] {
> + return -1
> +}
> +
> +# Print the line table
> +gdb_test_multiple "maint info line-table ${testfile}" "" {
> + -re "\r\n$decimal\[ \t\]+$decimal\[ \t\]+($hex)\[ \t\]+Y\[^\r\n\]*" {
> + lappend is_stmt $expect_out(1,string)
> + exp_continue
> + }
> + -re -wrap "" {
> + }
> +}
> +
> +# Activate process record/replay
> +gdb_test_no_output "record" "turn on process record"
> +
> +gdb_test "tbreak main_return" "Temporary breakpoint .*" "breakpoint at return"
> +gdb_test "continue" "Temporary breakpoint .*" "run to end of main"
> +gdb_test "display \$pc" ".*pc =.*" "display pc"
> +
> +# At this point, GDB has already recorded the execution up until the return
> +# statement. Reverse-step and test if GDB transitions between lines in the
> +# expected order. It should reverse-step across lines 8, 5, 3, 2 and 1.
> +foreach line {8 5 3 2 1} {
> + gdb_test "reverse-step" ".*TAG: line $line.*" "reverse step to line $line"
> +}
^ permalink raw reply [flat|nested] 40+ messages in thread
* RE: [PATCH] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-05-02 14:15 ` Bruno Larsen
@ 2023-05-02 15:40 ` Carl Love
2023-05-02 15:42 ` Bruno Larsen
2023-05-11 15:11 ` Simon Marchi
1 sibling, 1 reply; 40+ messages in thread
From: Carl Love @ 2023-05-02 15:40 UTC (permalink / raw)
To: Bruno Larsen, gdb-patches, Ulrich Weigand, pedro; +Cc: luis.machado
Bruno:
On Tue, 2023-05-02 at 16:15 +0200, Bruno Larsen wrote:
I had intended everything from here down as part of the commit message.
It is admittedly a lot. It could be cut down my making the first part
could be moved to the mailing list context. The commit message would
then start with the two scenarios the patch fixes.
>
> > ---------------------------------------------------------------
> > Fix reverse stepping multiple contiguous PC ranges over the line
> > table.
> >
> > There are a couple of scenarios where the GDB reverse-step and
> > reverse-next
> > commands do not work correctly. The first scenario consists of
> > multiple
> > assignment statements on the same line. A patch was proposed to
> > address the
> > issue by Luis Machado and briefly discussed on the mailing list in
> > Feb 2021.
> >
> > https://sourceware.org/pipermail/gdb-patches/2021-February/175678.html
> >
> > The discussion was revived by Carl Love with regards to fixing the
> > same
> > issue on PowerPC.
> >
> > https://sourceware.org/pipermail/gdb-patches/2022-March/186463.html
> >
> > The patch was not completed and has been on Carl's to do list for
> > some time.
> >
> > Discussion of a patch to change how the reverse-step and reverse-
> > next
> > commands submitted by Carl Love was started in thread:
> >
> > https://sourceware.org/pipermail/gdb-patches/2023-January/195563.html
> >
> > The patch was withdrawn as it was pointed out the proposed patch
> > would
> > change the intended behavior of the commands as documented in the
> > GDB
> > manual. However, it was pointed out by Pedro Alves <
> > pedro@palves.net>
> > that the reverse-step and reverse-next commands do not work when
> > there
> > are multiple function calls on the same line. This is a second
> > scenario
> > where the commands do not work correctly.
> >
> > The following patch is an extended version of the original patch by
> > Luis Machado to fix the issues in scenario 1 to also address the
> > issues in
> > scenario 2.
> >
Move the above to the mailing list
Start here with the commit message
> > --------------------------------------------------------
> >
> > Scenario 1 issue description by Luis Machado:
> >
> > When running GDB's testsuite on aarch64-linux/Ubuntu 20.04 (also
> > spotted on
> > the ppc backend), I noticed some failures in gdb.reverse/solib-
> > precsave.exp
> > and gdb.reverse/solib-reverse.exp.
> >
> > The failure happens around the following code:
> >
> > 38 b[1] = shr2(17); /* middle part two */
> > 40 b[0] = 6; b[1] = 9; /* generic statement, end part two */
> > 42 shr1 ("message 1\n"); /* shr1 one */
> >
> > Normal execution:
> >
> > - step from line 38 will land on line 40.
> > - step from line 40 will land on line 42.
> >
> > Reverse execution:
> > - step from line 42 will land on line 40.
> > - step from line 40 will land on line 40.
> > - step from line 40 will land on line 38.
> >
> > The problem here is that line 40 contains two contiguous but
> > distinct
> > PC ranges in the line table, like so:
> >
> > Line 40 - [0x7ec ~ 0x7f4]
> > Line 40 - [0x7f4 ~ 0x7fc]
> >
> > The two distinct ranges are generated because GCC started
> > outputting source
> > column information, which GDB doesn't take into account at the
> > moment.
> >
> > When stepping forward from line 40, we skip both of these ranges
> > and land on
> > line 42. When stepping backward from line 42, we stop at the start
> > PC of the
> > second (or first, going backwards) range of line 40.
> >
> > This happens because we have this check in
> > infrun.c:process_event_stop_test:
> >
> > /* When stepping backward, stop at beginning of line range
> > (unless it's the function entry point, in which case
> > keep going back to the call point). */
> > CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
> > if (stop_pc == ecs->event_thread->control.step_range_start
> > && stop_pc != ecs->stop_func_start
> > && execution_direction == EXEC_REVERSE)
> > end_stepping_range (ecs);
> > else
> > keep_going (ecs);
> >
> > Since we've reached ecs->event_thread->control.step_range_start, we
> > stop
> > stepping backwards.
> >
> > The right thing to do is to look for adjacent PC ranges for the
> > same line,
> > until we notice a line change. Then we take that as the start PC of
> > the
> > range.
> >
> > Another solution I thought about is to merge the contiguous ranges
> > when
> > we are reading the line tables. Though I'm not sure if we really
> > want to
> > process that data as opposed to keeping it as the compiler created,
> > and
> > then working around that.
> >
> > The test case gdb.reverse/map-to-same-line.exp is added to test the
> > fix
> > for the issues in scenario 1.
> >
> > ---------------------------------------------------------
> >
> > Scenario 2 issue described by Pedro Alves:
> >
> > The following explanation of the issue was taken from the gdb
> > mailing list
> > discussion of the withdrawn patch to change the behavior of the
> > reverse-step
> > and reverse-next commands. Specifically, message from Pedro Alves
> > <pedro@palves.net> where he demonstrates the issue where you have
> > multiple
> > function calls on the same source code line:
> >
> > https://sourceware.org/pipermail/gdb-patches/2023-January/196110.html
> >
> > The source line looks like:
> >
> > func1 (); func2 ();
> >
> > so stepping backwards over that line should always stop at the
> > first
> > instruction of the line, not in the middle. Let's simplify this.
> >
> > Here's the full source code of my example:
> >
> > (gdb) list 1
> > 1 void func1 ()
> > 2 {
> > 3 }
> > 4
> > 5 void func2 ()
> > 6 {
> > 7 }
> > 8
> > 9 int main ()
> > 10 {
> > 11 func1 (); func2 ();
> > 12 }
> >
> > Compiled with:
> >
> > $ gcc reverse.c -o reverse -g3 -O0
> > $ gcc -v
> > ...
> > gcc version 11.3.0 (Ubuntu 11.3.0-1ubuntu1~22.04)
> >
> > Now let's debug it with target record, using current gdb git master
> > (f3d8ae90b236),
> > without your patch:
> >
> > $ gdb ~/reverse
> > GNU gdb (GDB) 14.0.50.20230124-git
> > ...
> > Reading symbols from /home/pedro/reverse...
> > (gdb) start
> > Temporary breakpoint 1 at 0x1147: file reverse.c, line 11.
> > Starting program: /home/pedro/reverse
> > [Thread debugging using libthread_db enabled]
> > Using host libthread_db library "/lib/x86_64-linux-
> > gnu/libthread_db.so.1".
> >
> > Temporary breakpoint 1, main () at reverse.c:11
> > 11 func1 (); func2 ();
> > (gdb) record
> >
> > (gdb) disassemble /s
> > Dump of assembler code for function main:
> > reverse.c:
> > 10 {
> > 0x000055555555513f <+0>: endbr64
> > 0x0000555555555143 <+4>: push %rbp
> > 0x0000555555555144 <+5>: mov %rsp,%rbp
> >
> > 11 func1 (); func2 ();
> > => 0x0000555555555147 <+8>: mov $0x0,%eax
> > 0x000055555555514c <+13>: call 0x555555555129 <func1>
> > 0x0000555555555151 <+18>: mov $0x0,%eax
> > 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> > 0x000055555555515b <+28>: mov $0x0,%eax
> >
> > 12 }
> > 0x0000555555555160 <+33>: pop %rbp
> > 0x0000555555555161 <+34>: ret
> > End of assembler dump.
> >
> > (gdb) n
> > 12 }
> >
> > So far so good, a "next" stepped over the whole of line 11 and
> > stopped at line 12.
> >
> > Let's confirm where we are now:
> >
> > (gdb) disassemble /s
> > Dump of assembler code for function main:
> > reverse.c:
> > 10 {
> > 0x000055555555513f <+0>: endbr64
> > 0x0000555555555143 <+4>: push %rbp
> > 0x0000555555555144 <+5>: mov %rsp,%rbp
> >
> > 11 func1 (); func2 ();
> > 0x0000555555555147 <+8>: mov $0x0,%eax
> > 0x000055555555514c <+13>: call 0x555555555129 <func1>
> > 0x0000555555555151 <+18>: mov $0x0,%eax
> > 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> > 0x000055555555515b <+28>: mov $0x0,%eax
> >
> > 12 }
> > => 0x0000555555555160 <+33>: pop %rbp
> > 0x0000555555555161 <+34>: ret
> > End of assembler dump.
> >
> > Good, we're at the first instruction of line 12.
> >
> > Now let's undo the "next", with "reverse-next":
> >
> > (gdb) reverse-next
> > 11 func1 (); func2 ();
> >
> > Seemingly stopped at line 11. Let's see exactly where:
> >
> > (gdb) disassemble /s
> > Dump of assembler code for function main:
> > reverse.c:
> > 10 {
> > 0x000055555555513f <+0>: endbr64
> > 0x0000555555555143 <+4>: push %rbp
> > 0x0000555555555144 <+5>: mov %rsp,%rbp
> >
> > 11 func1 (); func2 ();
> > 0x0000555555555147 <+8>: mov $0x0,%eax
> > 0x000055555555514c <+13>: call 0x555555555129 <func1>
> > => 0x0000555555555151 <+18>: mov $0x0,%eax
> > 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> > 0x000055555555515b <+28>: mov $0x0,%eax
> >
> > 12 }
> > 0x0000555555555160 <+33>: pop %rbp
> > 0x0000555555555161 <+34>: ret
> > End of assembler dump.
> > (gdb)
> >
> > And lo, we stopped in the middle of line 11! That is a bug, we
> > should have
> > stepped back all the way to the beginning of the line. The
> > "reverse-next"
> > should have fully undone the prior "next" command.
> >
> > The test cases gdb.reverse/func-map-to-same-line-no-colum-info.exp
> > and
> > gdb.reverse/func-map-to-same-line.exp were added to test the fix
> > for scenario
> > 2 when the binary was compiled with and without line table
> > information.
> >
> > bug:
> > https://sourceware.org/bugzilla/show_bug.cgi?id=28426
> >
> >
> > Co-authored-by: Luis Machado <luis.machado@arm.com>
> > Co-authored-by: Carl Love <cel@us.ibm.com>
End of commit message
>
> Hey Carl,
>
> Thanks for working on this. I'm wondering which parts will be part
> of
> the final commit messages and which is just context for the mailing
> list, so some clarity would be nice, but that is not a huge deal.
>
> I wanted to test this change, but it doesn't apply anymore on
> master,
> and `git apply --3way` can't figure out how to do it. Which commit
> did
> you use as base (or alternatively, can you rebase it)?
OK, let me rebase it and repost. Let me know if you think the
suggested commit message is still too much. Thanks.
Carl
>
^ permalink raw reply [flat|nested] 40+ messages in thread
* Re: [PATCH] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-05-02 15:40 ` Carl Love
@ 2023-05-02 15:42 ` Bruno Larsen
0 siblings, 0 replies; 40+ messages in thread
From: Bruno Larsen @ 2023-05-02 15:42 UTC (permalink / raw)
To: Carl Love, gdb-patches, Ulrich Weigand, pedro; +Cc: luis.machado
On 02/05/2023 17:40, Carl Love wrote:
> Bruno:
>
> On Tue, 2023-05-02 at 16:15 +0200, Bruno Larsen wrote:
>
> I had intended everything from here down as part of the commit message.
> It is admittedly a lot. It could be cut down my making the first part
> could be moved to the mailing list context. The commit message would
> then start with the two scenarios the patch fixes.
>>> ---------------------------------------------------------------
>>> Fix reverse stepping multiple contiguous PC ranges over the line
>>> table.
>>>
>>> There are a couple of scenarios where the GDB reverse-step and
>>> reverse-next
>>> commands do not work correctly. The first scenario consists of
>>> multiple
>>> assignment statements on the same line. A patch was proposed to
>>> address the
>>> issue by Luis Machado and briefly discussed on the mailing list in
>>> Feb 2021.
>>>
>>> https://sourceware.org/pipermail/gdb-patches/2021-February/175678.html
>>>
>>> The discussion was revived by Carl Love with regards to fixing the
>>> same
>>> issue on PowerPC.
>>>
>>> https://sourceware.org/pipermail/gdb-patches/2022-March/186463.html
>>>
>>> The patch was not completed and has been on Carl's to do list for
>>> some time.
>>>
>>> Discussion of a patch to change how the reverse-step and reverse-
>>> next
>>> commands submitted by Carl Love was started in thread:
>>>
>>> https://sourceware.org/pipermail/gdb-patches/2023-January/195563.html
>>>
>>> The patch was withdrawn as it was pointed out the proposed patch
>>> would
>>> change the intended behavior of the commands as documented in the
>>> GDB
>>> manual. However, it was pointed out by Pedro Alves <
>>> pedro@palves.net>
>>> that the reverse-step and reverse-next commands do not work when
>>> there
>>> are multiple function calls on the same line. This is a second
>>> scenario
>>> where the commands do not work correctly.
>>>
>>> The following patch is an extended version of the original patch by
>>> Luis Machado to fix the issues in scenario 1 to also address the
>>> issues in
>>> scenario 2.
>>>
> Move the above to the mailing list
>
> Start here with the commit message
>
>>> --------------------------------------------------------
>>>
>>> Scenario 1 issue description by Luis Machado:
>>>
>>> When running GDB's testsuite on aarch64-linux/Ubuntu 20.04 (also
>>> spotted on
>>> the ppc backend), I noticed some failures in gdb.reverse/solib-
>>> precsave.exp
>>> and gdb.reverse/solib-reverse.exp.
>>>
>>> The failure happens around the following code:
>>>
>>> 38 b[1] = shr2(17); /* middle part two */
>>> 40 b[0] = 6; b[1] = 9; /* generic statement, end part two */
>>> 42 shr1 ("message 1\n"); /* shr1 one */
>>>
>>> Normal execution:
>>>
>>> - step from line 38 will land on line 40.
>>> - step from line 40 will land on line 42.
>>>
>>> Reverse execution:
>>> - step from line 42 will land on line 40.
>>> - step from line 40 will land on line 40.
>>> - step from line 40 will land on line 38.
>>>
>>> The problem here is that line 40 contains two contiguous but
>>> distinct
>>> PC ranges in the line table, like so:
>>>
>>> Line 40 - [0x7ec ~ 0x7f4]
>>> Line 40 - [0x7f4 ~ 0x7fc]
>>>
>>> The two distinct ranges are generated because GCC started
>>> outputting source
>>> column information, which GDB doesn't take into account at the
>>> moment.
>>>
>>> When stepping forward from line 40, we skip both of these ranges
>>> and land on
>>> line 42. When stepping backward from line 42, we stop at the start
>>> PC of the
>>> second (or first, going backwards) range of line 40.
>>>
>>> This happens because we have this check in
>>> infrun.c:process_event_stop_test:
>>>
>>> /* When stepping backward, stop at beginning of line range
>>> (unless it's the function entry point, in which case
>>> keep going back to the call point). */
>>> CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
>>> if (stop_pc == ecs->event_thread->control.step_range_start
>>> && stop_pc != ecs->stop_func_start
>>> && execution_direction == EXEC_REVERSE)
>>> end_stepping_range (ecs);
>>> else
>>> keep_going (ecs);
>>>
>>> Since we've reached ecs->event_thread->control.step_range_start, we
>>> stop
>>> stepping backwards.
>>>
>>> The right thing to do is to look for adjacent PC ranges for the
>>> same line,
>>> until we notice a line change. Then we take that as the start PC of
>>> the
>>> range.
>>>
>>> Another solution I thought about is to merge the contiguous ranges
>>> when
>>> we are reading the line tables. Though I'm not sure if we really
>>> want to
>>> process that data as opposed to keeping it as the compiler created,
>>> and
>>> then working around that.
>>>
>>> The test case gdb.reverse/map-to-same-line.exp is added to test the
>>> fix
>>> for the issues in scenario 1.
>>>
>>> ---------------------------------------------------------
>>>
>>> Scenario 2 issue described by Pedro Alves:
>>>
>>> The following explanation of the issue was taken from the gdb
>>> mailing list
>>> discussion of the withdrawn patch to change the behavior of the
>>> reverse-step
>>> and reverse-next commands. Specifically, message from Pedro Alves
>>> <pedro@palves.net> where he demonstrates the issue where you have
>>> multiple
>>> function calls on the same source code line:
>>>
>>> https://sourceware.org/pipermail/gdb-patches/2023-January/196110.html
>>>
>>> The source line looks like:
>>>
>>> func1 (); func2 ();
>>>
>>> so stepping backwards over that line should always stop at the
>>> first
>>> instruction of the line, not in the middle. Let's simplify this.
>>>
>>> Here's the full source code of my example:
>>>
>>> (gdb) list 1
>>> 1 void func1 ()
>>> 2 {
>>> 3 }
>>> 4
>>> 5 void func2 ()
>>> 6 {
>>> 7 }
>>> 8
>>> 9 int main ()
>>> 10 {
>>> 11 func1 (); func2 ();
>>> 12 }
>>>
>>> Compiled with:
>>>
>>> $ gcc reverse.c -o reverse -g3 -O0
>>> $ gcc -v
>>> ...
>>> gcc version 11.3.0 (Ubuntu 11.3.0-1ubuntu1~22.04)
>>>
>>> Now let's debug it with target record, using current gdb git master
>>> (f3d8ae90b236),
>>> without your patch:
>>>
>>> $ gdb ~/reverse
>>> GNU gdb (GDB) 14.0.50.20230124-git
>>> ...
>>> Reading symbols from /home/pedro/reverse...
>>> (gdb) start
>>> Temporary breakpoint 1 at 0x1147: file reverse.c, line 11.
>>> Starting program: /home/pedro/reverse
>>> [Thread debugging using libthread_db enabled]
>>> Using host libthread_db library "/lib/x86_64-linux-
>>> gnu/libthread_db.so.1".
>>>
>>> Temporary breakpoint 1, main () at reverse.c:11
>>> 11 func1 (); func2 ();
>>> (gdb) record
>>>
>>> (gdb) disassemble /s
>>> Dump of assembler code for function main:
>>> reverse.c:
>>> 10 {
>>> 0x000055555555513f <+0>: endbr64
>>> 0x0000555555555143 <+4>: push %rbp
>>> 0x0000555555555144 <+5>: mov %rsp,%rbp
>>>
>>> 11 func1 (); func2 ();
>>> => 0x0000555555555147 <+8>: mov $0x0,%eax
>>> 0x000055555555514c <+13>: call 0x555555555129 <func1>
>>> 0x0000555555555151 <+18>: mov $0x0,%eax
>>> 0x0000555555555156 <+23>: call 0x555555555134 <func2>
>>> 0x000055555555515b <+28>: mov $0x0,%eax
>>>
>>> 12 }
>>> 0x0000555555555160 <+33>: pop %rbp
>>> 0x0000555555555161 <+34>: ret
>>> End of assembler dump.
>>>
>>> (gdb) n
>>> 12 }
>>>
>>> So far so good, a "next" stepped over the whole of line 11 and
>>> stopped at line 12.
>>>
>>> Let's confirm where we are now:
>>>
>>> (gdb) disassemble /s
>>> Dump of assembler code for function main:
>>> reverse.c:
>>> 10 {
>>> 0x000055555555513f <+0>: endbr64
>>> 0x0000555555555143 <+4>: push %rbp
>>> 0x0000555555555144 <+5>: mov %rsp,%rbp
>>>
>>> 11 func1 (); func2 ();
>>> 0x0000555555555147 <+8>: mov $0x0,%eax
>>> 0x000055555555514c <+13>: call 0x555555555129 <func1>
>>> 0x0000555555555151 <+18>: mov $0x0,%eax
>>> 0x0000555555555156 <+23>: call 0x555555555134 <func2>
>>> 0x000055555555515b <+28>: mov $0x0,%eax
>>>
>>> 12 }
>>> => 0x0000555555555160 <+33>: pop %rbp
>>> 0x0000555555555161 <+34>: ret
>>> End of assembler dump.
>>>
>>> Good, we're at the first instruction of line 12.
>>>
>>> Now let's undo the "next", with "reverse-next":
>>>
>>> (gdb) reverse-next
>>> 11 func1 (); func2 ();
>>>
>>> Seemingly stopped at line 11. Let's see exactly where:
>>>
>>> (gdb) disassemble /s
>>> Dump of assembler code for function main:
>>> reverse.c:
>>> 10 {
>>> 0x000055555555513f <+0>: endbr64
>>> 0x0000555555555143 <+4>: push %rbp
>>> 0x0000555555555144 <+5>: mov %rsp,%rbp
>>>
>>> 11 func1 (); func2 ();
>>> 0x0000555555555147 <+8>: mov $0x0,%eax
>>> 0x000055555555514c <+13>: call 0x555555555129 <func1>
>>> => 0x0000555555555151 <+18>: mov $0x0,%eax
>>> 0x0000555555555156 <+23>: call 0x555555555134 <func2>
>>> 0x000055555555515b <+28>: mov $0x0,%eax
>>>
>>> 12 }
>>> 0x0000555555555160 <+33>: pop %rbp
>>> 0x0000555555555161 <+34>: ret
>>> End of assembler dump.
>>> (gdb)
>>>
>>> And lo, we stopped in the middle of line 11! That is a bug, we
>>> should have
>>> stepped back all the way to the beginning of the line. The
>>> "reverse-next"
>>> should have fully undone the prior "next" command.
>>>
>>> The test cases gdb.reverse/func-map-to-same-line-no-colum-info.exp
>>> and
>>> gdb.reverse/func-map-to-same-line.exp were added to test the fix
>>> for scenario
>>> 2 when the binary was compiled with and without line table
>>> information.
>>>
>>> bug:
>>> https://sourceware.org/bugzilla/show_bug.cgi?id=28426
>>>
>>>
>>> Co-authored-by: Luis Machado <luis.machado@arm.com>
>>> Co-authored-by: Carl Love <cel@us.ibm.com>
> End of commit message
>> Hey Carl,
>>
>> Thanks for working on this. I'm wondering which parts will be part
>> of
>> the final commit messages and which is just context for the mailing
>> list, so some clarity would be nice, but that is not a huge deal.
>>
>> I wanted to test this change, but it doesn't apply anymore on
>> master,
>> and `git apply --3way` can't figure out how to do it. Which commit
>> did
>> you use as base (or alternatively, can you rebase it)?
> OK, let me rebase it and repost. Let me know if you think the
> suggested commit message is still too much. Thanks.
Hey sorry, this part was my bad, no need to rebase. I'll look at the
commit message in a bit
--
Cheers,
Bruno
>
> Carl
^ permalink raw reply [flat|nested] 40+ messages in thread
* Re: [PATCH] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-04-27 20:59 [PATCH] Fix reverse stepping multiple contiguous PC ranges over the line table Carl Love
2023-05-02 14:15 ` Bruno Larsen
@ 2023-05-03 9:53 ` Bruno Larsen
2023-05-04 2:55 ` Carl Love
2023-05-04 2:55 ` [PATCH v2] " Carl Love
1 sibling, 2 replies; 40+ messages in thread
From: Bruno Larsen @ 2023-05-03 9:53 UTC (permalink / raw)
To: Carl Love, gdb-patches, Ulrich Weigand, pedro; +Cc: luis.machado
On 27/04/2023 22:59, Carl Love wrote:
> GDB maintainers:
>
> The following patch fixes issues on PowerPC with the reverse-step and
> reverse-next instructions when there are multiple assignment statements
> on the same line and when there are multiple function calls on the same
> line. The commit log below discusses these issues in further depth.
> The discussion included what the correct operation should be for these
> commands based on the GDB documentation. The proposed patch at that
> time changed how the commands worked on other platforms such as X86 in
> a way they no longer matched the documentation.
>
> The issue is the line table contains multiple entries for the same
> source line. The patch adds a function to search the line table to
> find the address of the first instruction of a line. When setup up the
> reverse stepping range, the function is called to make sure the start
> of the range corresponds to the address of the first instruction for
> the line.
>
> The following patch fixes the execution of the reveres-step and
> reverse-next commands for both senarios of multiple statements on the
> same line for PowerPC and aarch64-linux. Unlike the previous patch, it
> does not change the operation of the commands on other platforms, i.e.
> X86. The patch adds new test cases for both scenarios to verify they
> work correctly.
>
> The patch has been tested on PowerPC, Intel X86 and aarch64-linux with
> no new regression failures.
>
> Please let me know if the patch is acceptable for mainline. Thanks.
>
> Carl
>
> ---------------------------------------------------------------
> Fix reverse stepping multiple contiguous PC ranges over the line table.
>
> There are a couple of scenarios where the GDB reverse-step and reverse-next
> commands do not work correctly. The first scenario consists of multiple
> assignment statements on the same line. A patch was proposed to address the
> issue by Luis Machado and briefly discussed on the mailing list in Feb 2021.
>
> https://sourceware.org/pipermail/gdb-patches/2021-February/175678.html
>
> The discussion was revived by Carl Love with regards to fixing the same
> issue on PowerPC.
>
> https://sourceware.org/pipermail/gdb-patches/2022-March/186463.html
>
> The patch was not completed and has been on Carl's to do list for some time.
>
> Discussion of a patch to change how the reverse-step and reverse-next
> commands submitted by Carl Love was started in thread:
>
> https://sourceware.org/pipermail/gdb-patches/2023-January/195563.html
>
> The patch was withdrawn as it was pointed out the proposed patch would
> change the intended behavior of the commands as documented in the GDB
> manual. However, it was pointed out by Pedro Alves <pedro@palves.net>
> that the reverse-step and reverse-next commands do not work when there
> are multiple function calls on the same line. This is a second scenario
> where the commands do not work correctly.
>
> The following patch is an extended version of the original patch by
> Luis Machado to fix the issues in scenario 1 to also address the issues in
> scenario 2.
>
> --------------------------------------------------------
Hi Carl, thanks for clarifying the intended commit message. I'm reacting
to it here because I also have some thoughts on the code, now that I
managed to apply it locally.
Starting on the commit message, it would be nice to have a 1-line
description of the problem before describing the scenarios in depth.
Taking the first line of the previous block is enough IMO.
> Scenario 1 issue description by Luis Machado:
>
> When running GDB's testsuite on aarch64-linux/Ubuntu 20.04 (also spotted on
> the ppc backend), I noticed some failures in gdb.reverse/solib-precsave.exp
> and gdb.reverse/solib-reverse.exp.
>
> The failure happens around the following code:
>
> 38 b[1] = shr2(17); /* middle part two */
> 40 b[0] = 6; b[1] = 9; /* generic statement, end part two */
> 42 shr1 ("message 1\n"); /* shr1 one */
>
> Normal execution:
>
> - step from line 38 will land on line 40.
> - step from line 40 will land on line 42.
>
> Reverse execution:
> - step from line 42 will land on line 40.
> - step from line 40 will land on line 40.
> - step from line 40 will land on line 38.
>
> The problem here is that line 40 contains two contiguous but distinct
> PC ranges in the line table, like so:
>
> Line 40 - [0x7ec ~ 0x7f4]
> Line 40 - [0x7f4 ~ 0x7fc]
>
> The two distinct ranges are generated because GCC started outputting source
> column information, which GDB doesn't take into account at the moment.
>
> When stepping forward from line 40, we skip both of these ranges and land on
> line 42. When stepping backward from line 42, we stop at the start PC of the
> second (or first, going backwards) range of line 40.
>
> This happens because we have this check in infrun.c:process_event_stop_test:
>
> /* When stepping backward, stop at beginning of line range
> (unless it's the function entry point, in which case
> keep going back to the call point). */
> CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
> if (stop_pc == ecs->event_thread->control.step_range_start
> && stop_pc != ecs->stop_func_start
> && execution_direction == EXEC_REVERSE)
> end_stepping_range (ecs);
> else
> keep_going (ecs);
>
> Since we've reached ecs->event_thread->control.step_range_start, we stop
> stepping backwards.
I think these last 3 paragraphs should be moved. I like to finish
commits with a description of the solution, rather than having it in the
middle of the text. Also, I think we like to avoid mentioning explicit
code in the commit text (though I might be mistaken).
> The right thing to do is to look for adjacent PC ranges for the same line,
> until we notice a line change. Then we take that as the start PC of the
> range.
>
> Another solution I thought about is to merge the contiguous ranges when
> we are reading the line tables. Though I'm not sure if we really want to
> process that data as opposed to keeping it as the compiler created, and
> then working around that.
This paragraph doesn't need to be here in the final commit message IMO.
It was nice context for the mailing list but is not necessary for future
reference, I don't think.
>
> The test case gdb.reverse/map-to-same-line.exp is added to test the fix
> for the issues in scenario 1.
>
> ---------------------------------------------------------
>
> Scenario 2 issue described by Pedro Alves:
>
> The following explanation of the issue was taken from the gdb mailing list
> discussion of the withdrawn patch to change the behavior of the reverse-step
> and reverse-next commands. Specifically, message from Pedro Alves
> <pedro@palves.net> where he demonstrates the issue where you have multiple
> function calls on the same source code line:
>
> https://sourceware.org/pipermail/gdb-patches/2023-January/196110.html
> The source line looks like:
>
> func1 (); func2 ();
>
> so stepping backwards over that line should always stop at the first
> instruction of the line, not in the middle. Let's simplify this.
>
> Here's the full source code of my example:
>
> (gdb) list 1
> 1 void func1 ()
> 2 {
> 3 }
> 4
> 5 void func2 ()
> 6 {
> 7 }
> 8
> 9 int main ()
> 10 {
> 11 func1 (); func2 ();
> 12 }
>
> Compiled with:
>
> $ gcc reverse.c -o reverse -g3 -O0
> $ gcc -v
> ...
> gcc version 11.3.0 (Ubuntu 11.3.0-1ubuntu1~22.04)
>
> Now let's debug it with target record, using current gdb git master (f3d8ae90b236),
> without your patch:
>
> $ gdb ~/reverse
> GNU gdb (GDB) 14.0.50.20230124-git
> ...
> Reading symbols from /home/pedro/reverse...
> (gdb) start
> Temporary breakpoint 1 at 0x1147: file reverse.c, line 11.
> Starting program: /home/pedro/reverse
> [Thread debugging using libthread_db enabled]
> Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
>
> Temporary breakpoint 1, main () at reverse.c:11
> 11 func1 (); func2 ();
> (gdb) record
>
> (gdb) disassemble /s
> Dump of assembler code for function main:
> reverse.c:
> 10 {
> 0x000055555555513f <+0>: endbr64
> 0x0000555555555143 <+4>: push %rbp
> 0x0000555555555144 <+5>: mov %rsp,%rbp
>
> 11 func1 (); func2 ();
> => 0x0000555555555147 <+8>: mov $0x0,%eax
> 0x000055555555514c <+13>: call 0x555555555129 <func1>
> 0x0000555555555151 <+18>: mov $0x0,%eax
> 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> 0x000055555555515b <+28>: mov $0x0,%eax
>
> 12 }
> 0x0000555555555160 <+33>: pop %rbp
> 0x0000555555555161 <+34>: ret
> End of assembler dump.
>
> (gdb) n
> 12 }
>
> So far so good, a "next" stepped over the whole of line 11 and stopped at line 12.
>
> Let's confirm where we are now:
>
> (gdb) disassemble /s
> Dump of assembler code for function main:
> reverse.c:
> 10 {
> 0x000055555555513f <+0>: endbr64
> 0x0000555555555143 <+4>: push %rbp
> 0x0000555555555144 <+5>: mov %rsp,%rbp
>
> 11 func1 (); func2 ();
> 0x0000555555555147 <+8>: mov $0x0,%eax
> 0x000055555555514c <+13>: call 0x555555555129 <func1>
> 0x0000555555555151 <+18>: mov $0x0,%eax
> 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> 0x000055555555515b <+28>: mov $0x0,%eax
>
> 12 }
> => 0x0000555555555160 <+33>: pop %rbp
> 0x0000555555555161 <+34>: ret
> End of assembler dump.
>
> Good, we're at the first instruction of line 12.
>
> Now let's undo the "next", with "reverse-next":
>
> (gdb) reverse-next
> 11 func1 (); func2 ();
>
> Seemingly stopped at line 11. Let's see exactly where:
>
> (gdb) disassemble /s
> Dump of assembler code for function main:
> reverse.c:
> 10 {
> 0x000055555555513f <+0>: endbr64
> 0x0000555555555143 <+4>: push %rbp
> 0x0000555555555144 <+5>: mov %rsp,%rbp
>
> 11 func1 (); func2 ();
> 0x0000555555555147 <+8>: mov $0x0,%eax
> 0x000055555555514c <+13>: call 0x555555555129 <func1>
> => 0x0000555555555151 <+18>: mov $0x0,%eax
> 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> 0x000055555555515b <+28>: mov $0x0,%eax
>
> 12 }
> 0x0000555555555160 <+33>: pop %rbp
> 0x0000555555555161 <+34>: ret
> End of assembler dump.
> (gdb)
>
> And lo, we stopped in the middle of line 11! That is a bug, we should have
> stepped back all the way to the beginning of the line. The "reverse-next"
> should have fully undone the prior "next" command.
>
> The test cases gdb.reverse/func-map-to-same-line-no-colum-info.exp and
> gdb.reverse/func-map-to-same-line.exp were added to test the fix for scenario
> 2 when the binary was compiled with and without line table information.
>
> bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28426
>
> Co-authored-by: Luis Machado <luis.machado@arm.com>
> Co-authored-by: Carl Love <cel@us.ibm.com>
> ---
> gdb/infrun.c | 57 +++++++
> gdb/symtab.c | 49 ++++++
> gdb/symtab.h | 16 ++
> .../func-map-to-same-line-no-column-info.exp | 135 ++++++++++++++++
> .../gdb.reverse/func-map-to-same-line.c | 36 +++++
> .../gdb.reverse/func-map-to-same-line.exp | 123 ++++++++++++++
> gdb/testsuite/gdb.reverse/map-to-same-line.c | 58 +++++++
> .../gdb.reverse/map-to-same-line.exp | 153 ++++++++++++++++++
> 8 files changed, 627 insertions(+)
> create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-line-no-column-info.exp
> create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.c
> create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.exp
>
> diff --git a/gdb/infrun.c b/gdb/infrun.c
> index 2f1c6cd694b..59374a05471 100644
> --- a/gdb/infrun.c
> +++ b/gdb/infrun.c
> @@ -113,6 +113,9 @@ static struct async_event_handler *infrun_async_inferior_event_token;
> Starts off as -1, indicating "never enabled/disabled". */
> static int infrun_is_async = -1;
>
> +static CORE_ADDR update_line_range_start (CORE_ADDR pc,
> + struct execution_control_state *ecs);
> +
> /* See infrun.h. */
>
> void
> @@ -6768,6 +6771,25 @@ handle_signal_stop (struct execution_control_state *ecs)
> process_event_stop_test (ecs);
> }
>
> +CORE_ADDR
> +update_line_range_start (CORE_ADDR pc, struct execution_control_state *ecs)
> +{
> + /* The line table may have multiple entries for the same source code line.
> + Given the PC, check the line table and return the PC that corresponds
> + to the line table entry for the source line that PC is in. */
> + CORE_ADDR start_line_pc = ecs->event_thread->control.step_range_start;
> + gdb::optional<CORE_ADDR> real_range_start;
> +
> + /* Call find_line_range_start to get smallest address in the
> + linetable for multiple Line X entries in the line table. */
> + real_range_start = find_line_range_start (pc);
> +
> + if (real_range_start.has_value ())
> + start_line_pc = *real_range_start;
> +
> + return start_line_pc;
> +}
> +
> /* Come here when we've got some debug event / signal we can explain
> (IOW, not a random signal), and test whether it should cause a
> stop, or whether we should resume the inferior (transparently).
> @@ -7569,6 +7591,28 @@ process_event_stop_test (struct execution_control_state *ecs)
>
> if (stop_pc_sal.is_stmt)
> {
> + if (execution_direction == EXEC_REVERSE)
> + {
> + /* We are stepping backwards make sure we have reached the
> + beginning of the line. */
> + CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
> + CORE_ADDR start_line_pc
> + = update_line_range_start (stop_pc, ecs);
> +
> + if (stop_pc != start_line_pc)
> + {
> + /* Have not reached the beginning of the source code line.
> + Set a step range. Execution should stop in any function
> + calls we execute back into before reaching the beginning
> + of the line. */
> + ecs->event_thread->control.step_range_start = start_line_pc;
> + ecs->event_thread->control.step_range_end = stop_pc;
> + set_step_info (ecs->event_thread, frame, stop_pc_sal);
> + keep_going (ecs);
> + return;
> + }
> + }
> +
> /* We are at the start of a statement.
>
> So stop. Note that we don't stop if we step into the middle of a
> @@ -7631,6 +7675,19 @@ process_event_stop_test (struct execution_control_state *ecs)
> set_step_info (ecs->event_thread, frame, stop_pc_sal);
>
> infrun_debug_printf ("keep going");
> +
> + if (execution_direction == EXEC_REVERSE)
> + {
> + CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
> +
> + /* Make sure the stop_pc is set to the beginning of the line. */
> + if (stop_pc != ecs->event_thread->control.step_range_start)
> + {
> + stop_pc = update_line_range_start (stop_pc, ecs);
> + ecs->event_thread->control.step_range_start = stop_pc;
> + }
> + }
> +
> keep_going (ecs);
> }
>
> diff --git a/gdb/symtab.c b/gdb/symtab.c
> index 27611a34ec4..91d35616eb9 100644
> --- a/gdb/symtab.c
> +++ b/gdb/symtab.c
> @@ -3282,6 +3282,55 @@ find_pc_line (CORE_ADDR pc, int notcurrent)
> return sal;
> }
>
> +/* Compare two symtab_and_line entries. Return true if both have
> + the same line number and the same symtab pointer. That means we
> + are dealing with two entries from the same line and from the same
> + source file.
> +
> + Return false otherwise. */
> +
> +static bool
> +sal_line_symtab_matches_p (const symtab_and_line &sal1,
> + const symtab_and_line &sal2)
> +{
> + return (sal1.line == sal2.line && sal1.symtab == sal2.symtab);
> +}
> +
> +/* See symtah.h. */
> +
> +gdb::optional<CORE_ADDR>
> +find_line_range_start (CORE_ADDR pc)
> +{
> + struct symtab_and_line current_sal = find_pc_line (pc, 0);
> +
> + if (current_sal.line == 0)
> + return {};
> +
> + struct symtab_and_line prev_sal = find_pc_line (current_sal.pc - 1, 0);
> +
> + /* If the previous entry is for a different line, that means we are already
> + at the entry with the start PC for this line. */
> + if (!sal_line_symtab_matches_p (prev_sal, current_sal))
> + return current_sal.pc;
> +
> + /* Otherwise, keep looking for entries for the same line but with
> + smaller PC's. */
> + bool done = false;
> + CORE_ADDR prev_pc;
> + while (!done)
> + {
> + prev_pc = prev_sal.pc;
> +
> + prev_sal = find_pc_line (prev_pc - 1, 0);
> +
> + /* Did we notice a line change? If so, we are done with the search. */
> + if (!sal_line_symtab_matches_p (prev_sal, current_sal))
> + done = true;
> + }
> +
> + return prev_pc;
> +}
> +
> /* See symtab.h. */
>
> struct symtab *
> diff --git a/gdb/symtab.h b/gdb/symtab.h
> index 404d0ab30a8..f54305636da 100644
> --- a/gdb/symtab.h
> +++ b/gdb/symtab.h
> @@ -2346,6 +2346,22 @@ extern struct symtab_and_line find_pc_line (CORE_ADDR, int);
> extern struct symtab_and_line find_pc_sect_line (CORE_ADDR,
> struct obj_section *, int);
>
> +/* Given PC, and assuming it is part of a range of addresses that is part of a
> + line, go back through the linetable and find the starting PC of that
> + line.
> +
> + For example, suppose we have 3 PC ranges for line X:
> +
> + Line X - [0x0 - 0x8]
> + Line X - [0x8 - 0x10]
> + Line X - [0x10 - 0x18]
> +
> + If we call the function with PC == 0x14, we want to return 0x0, as that is
> + the starting PC of line X, and the ranges are contiguous.
> +*/
> +
> +extern gdb::optional<CORE_ADDR> find_line_range_start (CORE_ADDR pc);
> +
> /* Wrapper around find_pc_line to just return the symtab. */
>
> extern struct symtab *find_pc_line_symtab (CORE_ADDR);
> diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line-no-column-info.exp b/gdb/testsuite/gdb.reverse/func-map-to-same-line-no-column-info.exp
> new file mode 100644
> index 00000000000..20529c90fc2
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line-no-column-info.exp
> @@ -0,0 +1,135 @@
> +# Copyright 2008-2023 Free Software Foundation, Inc.
> +
> +# This program is free software; you can redistribute it and/or modify
> +# it under the terms of the GNU General Public License as published by
> +# the Free Software Foundation; either version 3 of the License, or
> +# (at your option) any later version.
> +#
> +# This program is distributed in the hope that it will be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> +# GNU General Public License for more details.
> +#
> +# You should have received a copy of the GNU General Public License
> +# along with this program. If not, see <http://www.gnu.org/licenses/>. */
> +
> +# This file is part of the GDB testsuite. It tests reverse stepping.
> +# Lots of code borrowed from "step-test.exp".
> +
> +# This test checks to make sure there is no regression failures for
> +# the reverse-next command when stepping back over two functions in
> +# the same line.
> +
> +if ![supports_reverse] {
> + return
> +}
Nowadays you should use require instead of the if clause, like in
gdb.reverse/break-reverse.exp
> +
> +# This test uses the gcc no-column-info command.
Correct me if I'm wrong, but it seems to me that the other test is a
more generic version of this one, so this test could check for a gcc
recent enough to support this feature, instead of just generically gcc.
That said, gcc added it on version
7(https://gcc.gnu.org/git/gitweb.cgi?p=gcc.git;h=0029b929c9719a), is it
old enough that we don't care?
> +if ![is_c_compiler_gcc] {
> + unsupported "gcc is required for this test"
> + return 0
> +}
> +
> +set srcfile func-map-to-same-line.c
> +set executable func-map-to-same-line
> +
> +set options [list debug additional_flags=-gno-column-info]
> +
> +if {[build_executable "failed to prepare" $executable $srcfile $options] == -1}\
> + {
> + return -1
> +}
> +
> +clean_restart $executable
> +
> +runto_main
> +set target_remote [gdb_is_target_remote]
> +
> +if [supports_process_record] {
> + # Activate process record/replay.
> + gdb_test_no_output "record" "turn on process record for test1"
> +}
> +
> +# This regression test verifies the reverse-step and reverse-next commands
> +# work properly when executing backwards thru a source line containing
> +# two function calls on the same source line, i.e. func1 (); func2 ();
> +# This test is compiled so the dwarf info not contain the line table
> +# information.
> +
> +# Test 1, reverse-next command
> +# Set breakpoint at the line after the function calls.
> +set bp_start_reverse_test [gdb_get_line_number "START REVERSE TEST" $srcfile]
> +gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> +
> +# Continue to break point for reverse-next test.
> +# Command definition: reverse-next [count]
> +# Run backward to the beginning of the previous line executed in the current
> +# (innermost) stack frame. If the line contains function calls, they will be
> +# “un-executed” without stopping. Starting from the first line of a function,
> +# reverse-next will take you back to the caller of that function, before the
> +# function was called, just as the normal next command would take you from
> +# the last line of a function back to its return to its caller 2 .
> +
> +gdb_continue_to_breakpoint \
> + "stopped at command reverse-next test start location" \
> + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> +
> +# The reverse-next should step all the way back to the beginning of the line,
> +# i.e. at the beginning of the func1 call.
> +gdb_test "reverse-next" ".*func1 \\(\\); func2 \\(\\);.*" \
> + "reverse-next to line with two functions"
> +
> +# A reverse-step should step back and stop at the beginning
> +# of the previous line b = 2, i.e. not in func1 ().
> +gdb_test "reverse-step" ".*b = 2;.*" \
> + "reverse-step to previous line b = 2"
The point of this test is to confirm that we are at the very first
instruction of the line, right? So I would think it is better to do a
reverse-stepi here, to make sure that even walking a single instruction
we reach a different line. Either that or doing what Pedro did in his
email: save the PC before executing the line, then do and undo the line
and confirm that PCs match exactly.
> +
> +
> +# Setup for test 2
> +# Go back to the start of the function
> +gdb_test "reverse-continue" "a = 1;" "At start of main, setup for test 2"
> +
> +# Turn off record to clear logs and turn on again
> +gdb_test "record stop" "Process record is stopped.*" \
> + "turn off process record for test1"
> +gdb_test_no_output "record" "turn on process record for test2"
Since you don't require process record for this test, you can't assume
these to work. I think its better to clean restart and record if the
process supports recording, this way you're sure to reset history no
matter the inferior.
> +
> +# Delete all breakpoints and catchpoints.
> +delete_breakpoints
> +
> +
> +# Test 2, reverse-step command
> +# Set breakpoint at the line after the function calls.
> +gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> +
> +# Continue to the start of the reverse-step test.
> +# Command definition: reverse-step [count]
> +# Run the program backward until control reaches the start of a
> +# different source line; then stop it, and return control to gdb.
> +# Like the step command, reverse-step will only stop at the beginning of a
> +# source line. It “un-executes” the previously executed source line. If the
> +# previous source line included calls to debuggable functions, reverse-step
> +# will step (backward) into the called function, stopping at the beginning
> +# of the last statement in the called function (typically a return
> +# statement). Also, as with the step command, if non-debuggable functions
> +# are called, reverse-step will run thru them backward without stopping.
> +
> +gdb_continue_to_breakpoint \
> + "stopped at command reverse-step test start location" \
> + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> +
> +# The first reverse step should take us call of func2 ().
> +gdb_test "reverse-step" ".*}.*" \
> + "reverse-step into func2 "
> +
> +# The second reverse step should take us into func1 ().
> +gdb_test "reverse-step" ".*}.*" \
> + "reverse-step into func1 "
> +
> +# The third reverse step should take us call of func1 ().
> +gdb_test "reverse-step" ".*func1 \\(\\); func2 \\(\\);.*" \
> + "reverse-step to line func1(); func2(), at call for func1 "
> +
> +# The fourth reverse step should take us to b = 2 ().
> +gdb_test "reverse-step" ".*b = 2;.*" \
> + "reverse-step to line b = 2 "
Ditto from the other test like this. Also, I feel that, while the test
name for the last 2 gdb_test are different, they don't meaningfully
communicate which part of the test is failing. I think it would be
better if you differentiated them by adding "for step test" or "for test
2" at the end of the name would make it easier to understand where
things went wrong when looking at the sum file.
> diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.c b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> new file mode 100644
> index 00000000000..e9787ef9ff5
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> @@ -0,0 +1,36 @@
> +/* Copyright 2008-2023 Free Software Foundation, Inc.
> +
> + This program is free software; you can redistribute it and/or modify
> + it under the terms of the GNU General Public License as published by
> + the Free Software Foundation; either version 3 of the License, or
> + (at your option) any later version.
> +
> + This program is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> + GNU General Public License for more details.
> +
> + You should have received a copy of the GNU General Public License
> + along with this program. If not, see <http://www.gnu.org/licenses/>.
> +
> + This test is used to test the reverse-step and reverse-next instruction
> + execution for a source line that contains multiple function calls. */
> +
> +void
> +func1 ()
> +{
> +}
> +
> +void
> +func2 ()
> +{
> +}
> +
> +int main ()
> +{
> + int a, b;
> + a = 1;
> + b = 2;
> + func1 (); func2 ();
> + a = a + b; // START REVERSE TEST
> +}
> diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> new file mode 100644
> index 00000000000..b632a236bbe
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> @@ -0,0 +1,123 @@
> +# Copyright 2008-2023 Free Software Foundation, Inc.
> +
> +# This program is free software; you can redistribute it and/or modify
> +# it under the terms of the GNU General Public License as published by
> +# the Free Software Foundation; either version 3 of the License, or
> +# (at your option) any later version.
> +#
> +# This program is distributed in the hope that it will be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> +# GNU General Public License for more details.
> +#
> +# You should have received a copy of the GNU General Public License
> +# along with this program. If not, see <http://www.gnu.org/licenses/>. */
> +
> +# This file is part of the GDB testsuite. It tests reverse stepping.
> +# Lots of code borrowed from "step-test.exp".
> +
> +# This test checks to make sure there is no regression failures for
> +# the reverse-next command when stepping back over two functions in
> +# the same line.
> +
> +if ![supports_reverse] {
> + return
> +}
I'm not sure it's worth separating these 2 tests into separate files.
You could instead just have most of the test defined as a proc, and call
it twice, once after compiling the inferior with column info, the other
compiling without if gcc is used. This way it's less likely that the
tests will diverge over time.
> +
> +standard_testfile
> +
> +if { [prepare_for_testing "failed to prepare" $testfile $srcfile] } {
> + return -1
> +}
> +
> +runto_main
> +set target_remote [gdb_is_target_remote]
> +
> +if [supports_process_record] {
> + # Activate process record/replay.
> + gdb_test_no_output "record" "turn on process record for test1"
> +}
> +
> +# This regression test verifies the reverse-step and reverse-next commands
> +# work properly when executing backwards thru a source line containing
> +# two function calls on the same source line, i.e. func1 (); func2 ();
> +# The assumption for this test is the dwarf info contain the column
> +# information.
> +
> +# Test 1, reverse-next command
> +# Set breakpoint at the line after the function calls.
> +set bp_start_reverse_test [gdb_get_line_number "START REVERSE TEST" $srcfile]
> +gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> +
> +# Continue to break point for reverse-next test.
> +# Command definition: reverse-next [count]
> +# Run backward to the beginning of the previous line executed in the current
> +# (innermost) stack frame. If the line contains function calls, they will be
> +# “un-executed” without stopping. Starting from the first line of a function,
> +# reverse-next will take you back to the caller of that function, before the
> +# function was called, just as the normal next command would take you from
> +# the last line of a function back to its return to its caller 2 .
> +
> +gdb_continue_to_breakpoint \
> + "stopped at command reverse-next test start location" \
> + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> +
> +# The reverse-next should step all the way back to the beginning of the line,
> +# i.e. at the beginning of the func1 call.
> +gdb_test "reverse-next" ".*func1 \\(\\); func2 \\(\\);.*" \
> + "reverse-next to line with two functions"
> +
> +# A reverse-step should step back and stop at the beginning
> +# of the previous line b = 2, i.e. not in func1 ().
> +gdb_test "reverse-step" ".*b = 2;.*" \
> + "reverse-step to previous line b = 2"
> +
> +
> +# Setup for test 2
> +# Go back to the start of the function
> +gdb_test "reverse-continue" "a = 1;" "At start of main, setup for test 2"
> +
> +# Turn off record to clear logs and turn on again
> +gdb_test "record stop" "Process record is stopped.*" \
> + "turn off process record for test1"
> +gdb_test_no_output "record" "turn on process record for test2"
> +
> +# Delete all breakpoints and catchpoints.
> +delete_breakpoints
> +
> +
> +# Test 2, reverse-step command
> +# Set breakpoint at the line after the function calls.
> +gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> +
> +# Continue to the start of the reverse-step test.
> +# Command definition: reverse-step [count]
> +# Run the program backward until control reaches the start of a
> +# different source line; then stop it, and return control to gdb.
> +# Like the step command, reverse-step will only stop at the beginning of a
> +# source line. It “un-executes” the previously executed source line. If the
> +# previous source line included calls to debuggable functions, reverse-step
> +# will step (backward) into the called function, stopping at the beginning
> +# of the last statement in the called function (typically a return
> +# statement). Also, as with the step command, if non-debuggable functions
> +# are called, reverse-step will run thru them backward without stopping.
> +
> +gdb_continue_to_breakpoint \
> + "stopped at command reverse-step test start location" \
> + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> +
> +# The first reverse step should take us call of func2 ().
> +gdb_test "reverse-step" ".*}.*" \
> + "reverse-step into func2 "
> +
> +# The second reverse step should take us into func1 ().
> +gdb_test "reverse-step" ".*}.*" \
> + "reverse-step into func1 "
> +
> +# The third reverse step should take us call of func1 ().
> +gdb_test "reverse-step" ".*func1 \\(\\); func2 \\(\\);.*" \
> + "reverse-step to line func1(); func2(), at call for func1 "
> +
> +# The fourth reverse step should take us to b = 2 ().
> +gdb_test "reverse-step" ".*b = 2;.*" \
> + "reverse-step to line b = 2 "
> diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.c b/gdb/testsuite/gdb.reverse/map-to-same-line.c
> new file mode 100644
> index 00000000000..f20d778f40e
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/map-to-same-line.c
> @@ -0,0 +1,58 @@
> +/* Copyright 2008-2023 Free Software Foundation, Inc.
> +
> + This program is free software; you can redistribute it and/or modify
> + it under the terms of the GNU General Public License as published by
> + the Free Software Foundation; either version 3 of the License, or
> + (at your option) any later version.
> +
> + This program is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> + GNU General Public License for more details.
> +
> + You should have received a copy of the GNU General Public License
> + along with this program. If not, see <http://www.gnu.org/licenses/ >. */
> +
> +/* The purpose of this test is to create a DWARF line table that contains two
> + or more entries for the same line. When stepping (forwards or backwards),
> + GDB should step over the entire line and not just a particular entry in the
> + line table. */
> +
> +int
> +main ()
> +{ /* TAG: main prologue */
> + asm ("main_label: .globl main_label");
> + int i = 1, j = 2, k;
> + float f1 = 2.0, f2 = 4.1, f3;
> + const char *str_1 = "foo", *str_2 = "bar", *str_3;
> +
> + asm ("line1: .globl line1");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 1 */
> +
> + asm ("line2: .globl line2");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 2 */
> +
> + asm ("line3: .globl line3");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 3 */
> +
> + asm ("line4: .globl line4");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 4 */
> +
> + asm ("line5: .globl line5");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 5 */
> +
> + asm ("line6: .globl line6");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 6 */
> +
> + asm ("line7: .globl line7");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 7 */
> +
> + asm ("line8: .globl line8");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 8 */
> +
> + asm ("main_return: .globl main_return");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: main return */
> +
> + asm ("end_of_sequence: .globl end_of_sequence");
> + return 0; /* TAG: main return */
> +}
> diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.exp b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> new file mode 100644
> index 00000000000..a01579c2a8d
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> @@ -0,0 +1,153 @@
> +# Copyright 2008-2023 Free Software Foundation, Inc.
> +
> +# This program is free software; you can redistribute it and/or modify
> +# it under the terms of the GNU General Public License as published by
> +# the Free Software Foundation; either version 3 of the License, or
> +# (at your option) any later version.
> +#
> +# This program is distributed in the hope that it will be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> +# GNU General Public License for more details.
> +#
> +# You should have received a copy of the GNU General Public License
> +# along with this program. If not, see <http://www.gnu.org/licenses/ >.
> +
> +# When stepping (forwards or backwards), GDB should step over the entire line
> +# and not just a particular entry in the line table. This test was added to
> +# verify the find_line_range_start function properly sets the step range for a
> +# line that consists of multiple statements, i.e. multiple entries in the line
> +# table. This test creates a DWARF line table that contains two entries for
> +# the same line to do the needed testing.
> +
> +load_lib dwarf.exp
> +
> +# This test can only be run on targets which support DWARF-2 and use gas.
> +if {![dwarf2_support]} {
> + unsupported "dwarf2 support required for this test"
> + return 0
> +}
Again, the new way to check for these is "required". And IIUC, you can
add multiple requirements into a singe require call.
> +
> +if [get_compiler_info] {
> + return -1
> +}
> +
> +# The DWARF assembler requires the gcc compiler.
> +if ![is_c_compiler_gcc] {
> + unsupported "gcc is required for this test"
> + return 0
> +}
> +
> +# This test suitable only for process record-replay
> +if ![supports_process_record] {
> + return
> +}
> +
> +standard_testfile .c .S
> +
> +if { [prepare_for_testing "failed to prepare" ${testfile} ${srcfile}] } {
> + return -1
> +}
> +
> +set asm_file [standard_output_file $srcfile2]
> +Dwarf::assemble $asm_file {
> + global srcdir subdir srcfile
> + declare_labels integer_label L
> +
> + # Find start address and length of program
> + lassign [function_range main [list ${srcdir}/${subdir}/$srcfile]] \
> + main_start main_len
> + set main_end "$main_start + $main_len"
> +
> + cu {} {
> + compile_unit {
> + {language @DW_LANG_C}
> + {name map-to-same-line.c}
> + {stmt_list $L DW_FORM_sec_offset}
> + {low_pc 0 addr}
> + } {
> + subprogram {
> + {external 1 flag}
> + {name main}
> + {low_pc $main_start addr}
> + {high_pc $main_len DW_FORM_data4}
> + }
> + }
> + }
> +
> + lines {version 2 default_is_stmt 1} L {
> + include_dir "${srcdir}/${subdir}"
> + file_name "$srcfile" 1
> +
> + # Generate the line table program with distinct source lines being
> + # mapped to the same line entry. Line 1, 5 and 8 contain 1 statement
> + # each. Line 2 contains 2 statements. Line 3 contains 3 statements.
> + program {
> + DW_LNE_set_address $main_start
> + line [gdb_get_line_number "TAG: main prologue"]
> + DW_LNS_copy
> + DW_LNE_set_address line1
> + line [gdb_get_line_number "TAG: line 1" ]
> + DW_LNS_copy
> + DW_LNE_set_address line2
> + line [gdb_get_line_number "TAG: line 2" ]
> + DW_LNS_copy
> + DW_LNE_set_address line3
> + line [gdb_get_line_number "TAG: line 2" ]
> + DW_LNS_copy
> + DW_LNE_set_address line4
> + line [gdb_get_line_number "TAG: line 3" ]
> + DW_LNS_copy
> + DW_LNE_set_address line5
> + line [gdb_get_line_number "TAG: line 3" ]
> + DW_LNS_copy
> + DW_LNE_set_address line6
> + line [gdb_get_line_number "TAG: line 3" ]
> + DW_LNS_copy
> + DW_LNE_set_address line7
> + line [gdb_get_line_number "TAG: line 5" ]
> + DW_LNS_copy
> + DW_LNE_set_address line8
> + line [gdb_get_line_number "TAG: line 8" ]
> + DW_LNS_copy
> + DW_LNE_set_address main_return
> + line [gdb_get_line_number "TAG: main return"]
> + DW_LNS_copy
> + DW_LNE_set_address end_of_sequence
> + DW_LNE_end_sequence
> + }
> + }
> +}
> +
> +if { [prepare_for_testing "failed to prepare" ${testfile} \
> + [list $srcfile $asm_file] {nodebug} ] } {
> + return -1
> +}
> +
> +if ![runto_main] {
> + return -1
> +}
runto_main already errors out and leaves, I think, so no need for the if
clause.
> +
> +# Print the line table
> +gdb_test_multiple "maint info line-table ${testfile}" "" {
> + -re "\r\n$decimal\[ \t\]+$decimal\[ \t\]+($hex)\[ \t\]+Y\[^\r\n\]*" {
> + lappend is_stmt $expect_out(1,string)
> + exp_continue
> + }
> + -re -wrap "" {
> + }
> +}
> +
> +# Activate process record/replay
> +gdb_test_no_output "record" "turn on process record"
> +
> +gdb_test "tbreak main_return" "Temporary breakpoint .*" "breakpoint at return"
you can set a temporary breakpoint using gdb_breakpoint "..." temporary,
no need to manually call tbreak.
> +gdb_test "continue" "Temporary breakpoint .*" "run to end of main"
gdb_continue_to_breakpoint can handle temporary breakpoints as well.
> +gdb_test "display \$pc" ".*pc =.*" "display pc"
> +
> +# At this point, GDB has already recorded the execution up until the return
> +# statement. Reverse-step and test if GDB transitions between lines in the
> +# expected order. It should reverse-step across lines 8, 5, 3, 2 and 1.
> +foreach line {8 5 3 2 1} {
> + gdb_test "reverse-step" ".*TAG: line $line.*" "reverse step to line $line"
> +}
I'm not sure if it is needed, but I don't think it would hurt to also
test reverse-next in a separate foreach right after this one.
--
Cheers,
Bruno
^ permalink raw reply [flat|nested] 40+ messages in thread
* RE: [PATCH] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-05-03 9:53 ` Bruno Larsen
@ 2023-05-04 2:55 ` Carl Love
2023-05-04 9:24 ` Bruno Larsen
2023-05-04 2:55 ` [PATCH v2] " Carl Love
1 sibling, 1 reply; 40+ messages in thread
From: Carl Love @ 2023-05-04 2:55 UTC (permalink / raw)
To: Bruno Larsen, gdb-patches, Ulrich Weigand, pedro; +Cc: luis.machado, cel
Bruno:
On Wed, 2023-05-03 at 11:53 +0200, Bruno Larsen wrote:
> On 27/04/2023 22:59, Carl Love wrote:
<snip>
>
> Hi Carl, thanks for clarifying the intended commit message. I'm
> reacting
> to it here because I also have some thoughts on the code, now that I
> managed to apply it locally.
>
> Starting on the commit message, it would be nice to have a 1-line
> description of the problem before describing the scenarios in depth.
> Taking the first line of the previous block is enough IMO.
Yes, agreed. Kept the first line before the discussion of the
different failure scenarios.
>
> > Scenario 1 issue description by Luis Machado:
> >
> > When running GDB's testsuite on aarch64-linux/Ubuntu 20.04 (also
> > spotted on
> > the ppc backend), I noticed some failures in gdb.reverse/solib-
> > precsave.exp
> > and gdb.reverse/solib-reverse.exp.
> >
> > The failure happens around the following code:
> >
> > 38 b[1] = shr2(17); /* middle part two */
> > 40 b[0] = 6; b[1] = 9; /* generic statement, end part two */
> > 42 shr1 ("message 1\n"); /* shr1 one */
> >
> > Normal execution:
> >
> > - step from line 38 will land on line 40.
> > - step from line 40 will land on line 42.
> >
> > Reverse execution:
> > - step from line 42 will land on line 40.
> > - step from line 40 will land on line 40.
> > - step from line 40 will land on line 38.
> >
> > The problem here is that line 40 contains two contiguous but
> > distinct
> > PC ranges in the line table, like so:
> >
> > Line 40 - [0x7ec ~ 0x7f4]
> > Line 40 - [0x7f4 ~ 0x7fc]
> >
> > The two distinct ranges are generated because GCC started
> > outputting source
> > column information, which GDB doesn't take into account at the
> > moment.
> >
> > When stepping forward from line 40, we skip both of these ranges
> > and land on
> > line 42. When stepping backward from line 42, we stop at the start
> > PC of the
> > second (or first, going backwards) range of line 40.
> >
> > This happens because we have this check in
> > infrun.c:process_event_stop_test:
> >
> > /* When stepping backward, stop at beginning of line range
> > (unless it's the function entry point, in which case
> > keep going back to the call point). */
> > CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
> > if (stop_pc == ecs->event_thread->control.step_range_start
> > && stop_pc != ecs->stop_func_start
> > && execution_direction == EXEC_REVERSE)
> > end_stepping_range (ecs);
> > else
> > keep_going (ecs);
> >
> > Since we've reached ecs->event_thread->control.step_range_start, we
> > stop
> > stepping backwards.
> I think these last 3 paragraphs should be moved. I like to finish
> commits with a description of the solution, rather than having it in
> the
> middle of the text. Also, I think we like to avoid mentioning
> explicit
> code in the commit text (though I might be mistaken).
OK, I moved the fix discussion to the end. I also dropped the explicit
reference to infrun.c:process_event_stop_test.
> > The right thing to do is to look for adjacent PC ranges for the
> > same line,
> > until we notice a line change. Then we take that as the start PC of
> > the
> > range.
> >
> > Another solution I thought about is to merge the contiguous ranges
> > when
> > we are reading the line tables. Though I'm not sure if we really
> > want to
> > process that data as opposed to keeping it as the compiler created,
> > and
> > then working around that.
> This paragraph doesn't need to be here in the final commit message
> IMO.
> It was nice context for the mailing list but is not necessary for
> future
OK, removed it from the commit log and need to update the mailing list
message with this.
> reference, I don't think.
> > The test case gdb.reverse/map-to-same-line.exp is added to test the
> > fix
> > for the issues in scenario 1.
> >
> > ---------------------------------------------------------
> >
> > Scenario 2 issue described by Pedro Alves:
> >
> > The following explanation of the issue was taken from the gdb
> > mailing list
> > discussion of the withdrawn patch to change the behavior of the
> > reverse-step
> > and reverse-next commands. Specifically, message from Pedro Alves
> > <pedro@palves.net> where he demonstrates the issue where you have
> > multiple
> > function calls on the same source code line:
> >
> > https://sourceware.org/pipermail/gdb-patches/2023-January/196110.html
> > The source line looks like:
> >
> > func1 (); func2 ();
> >
> > so stepping backwards over that line should always stop at the
> > first
> > instruction of the line, not in the middle. Let's simplify this.
> >
> > Here's the full source code of my example:
> >
> > (gdb) list 1
> > 1 void func1 ()
> > 2 {
> > 3 }
> > 4
> > 5 void func2 ()
> > 6 {
> > 7 }
> > 8
> > 9 int main ()
> > 10 {
> > 11 func1 (); func2 ();
> > 12 }
> >
> > Compiled with:
> >
> > $ gcc reverse.c -o reverse -g3 -O0
> > $ gcc -v
> > ...
> > gcc version 11.3.0 (Ubuntu 11.3.0-1ubuntu1~22.04)
> >
> > Now let's debug it with target record, using current gdb git master
> > (f3d8ae90b236),
> > without your patch:
> >
> > $ gdb ~/reverse
> > GNU gdb (GDB) 14.0.50.20230124-git
> > ...
> > Reading symbols from /home/pedro/reverse...
> > (gdb) start
> > Temporary breakpoint 1 at 0x1147: file reverse.c, line 11.
> > Starting program: /home/pedro/reverse
> > [Thread debugging using libthread_db enabled]
> > Using host libthread_db library "/lib/x86_64-linux-
> > gnu/libthread_db.so.1".
> >
> > Temporary breakpoint 1, main () at reverse.c:11
> > 11 func1 (); func2 ();
> > (gdb) record
> >
> > (gdb) disassemble /s
> > Dump of assembler code for function main:
> > reverse.c:
> > 10 {
> > 0x000055555555513f <+0>: endbr64
> > 0x0000555555555143 <+4>: push %rbp
> > 0x0000555555555144 <+5>: mov %rsp,%rbp
> >
> > 11 func1 (); func2 ();
> > => 0x0000555555555147 <+8>: mov $0x0,%eax
> > 0x000055555555514c <+13>: call 0x555555555129 <func1>
> > 0x0000555555555151 <+18>: mov $0x0,%eax
> > 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> > 0x000055555555515b <+28>: mov $0x0,%eax
> >
> > 12 }
> > 0x0000555555555160 <+33>: pop %rbp
> > 0x0000555555555161 <+34>: ret
> > End of assembler dump.
> >
> > (gdb) n
> > 12 }
> >
> > So far so good, a "next" stepped over the whole of line 11 and
> > stopped at line 12.
> >
> > Let's confirm where we are now:
> >
> > (gdb) disassemble /s
> > Dump of assembler code for function main:
> > reverse.c:
> > 10 {
> > 0x000055555555513f <+0>: endbr64
> > 0x0000555555555143 <+4>: push %rbp
> > 0x0000555555555144 <+5>: mov %rsp,%rbp
> >
> > 11 func1 (); func2 ();
> > 0x0000555555555147 <+8>: mov $0x0,%eax
> > 0x000055555555514c <+13>: call 0x555555555129 <func1>
> > 0x0000555555555151 <+18>: mov $0x0,%eax
> > 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> > 0x000055555555515b <+28>: mov $0x0,%eax
> >
> > 12 }
> > => 0x0000555555555160 <+33>: pop %rbp
> > 0x0000555555555161 <+34>: ret
> > End of assembler dump.
> >
> > Good, we're at the first instruction of line 12.
> >
> > Now let's undo the "next", with "reverse-next":
> >
> > (gdb) reverse-next
> > 11 func1 (); func2 ();
> >
> > Seemingly stopped at line 11. Let's see exactly where:
> >
> > (gdb) disassemble /s
> > Dump of assembler code for function main:
> > reverse.c:
> > 10 {
> > 0x000055555555513f <+0>: endbr64
> > 0x0000555555555143 <+4>: push %rbp
> > 0x0000555555555144 <+5>: mov %rsp,%rbp
> >
> > 11 func1 (); func2 ();
> > 0x0000555555555147 <+8>: mov $0x0,%eax
> > 0x000055555555514c <+13>: call 0x555555555129 <func1>
> > => 0x0000555555555151 <+18>: mov $0x0,%eax
> > 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> > 0x000055555555515b <+28>: mov $0x0,%eax
> >
> > 12 }
> > 0x0000555555555160 <+33>: pop %rbp
> > 0x0000555555555161 <+34>: ret
> > End of assembler dump.
> > (gdb)
> >
> > And lo, we stopped in the middle of line 11! That is a bug, we
> > should have
> > stepped back all the way to the beginning of the line. The
> > "reverse-next"
> > should have fully undone the prior "next" command.
> >
> > The test cases gdb.reverse/func-map-to-same-line-no-colum-info.exp
> > and
> > gdb.reverse/func-map-to-same-line.exp were added to test the fix
> > for scenario
> > 2 when the binary was compiled with and without line table
> > information.
> >
> > bug:
> > https://sourceware.org/bugzilla/show_bug.cgi?id=28426
> >
> >
> > Co-authored-by: Luis Machado <luis.machado@arm.com>
> > Co-authored-by: Carl Love <cel@us.ibm.com>
> > ---
> > gdb/infrun.c | 57 +++++++
> > gdb/symtab.c | 49 ++++++
> > gdb/symtab.h | 16 ++
> > .../func-map-to-same-line-no-column-info.exp | 135
> > ++++++++++++++++
> > .../gdb.reverse/func-map-to-same-line.c | 36 +++++
> > .../gdb.reverse/func-map-to-same-line.exp | 123
> > ++++++++++++++
> > gdb/testsuite/gdb.reverse/map-to-same-line.c | 58 +++++++
> > .../gdb.reverse/map-to-same-line.exp | 153
> > ++++++++++++++++++
> > 8 files changed, 627 insertions(+)
> > create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-
> > line-no-column-info.exp
> > create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-
> > line.c
> > create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-
> > line.exp
> > create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.c
> > create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.exp
> >
> > diff --git a/gdb/infrun.c b/gdb/infrun.c
> > index 2f1c6cd694b..59374a05471 100644
> > --- a/gdb/infrun.c
> > +++ b/gdb/infrun.c
> > @@ -113,6 +113,9 @@ static struct async_event_handler
> > *infrun_async_inferior_event_token;
> > Starts off as -1, indicating "never enabled/disabled". */
> > static int infrun_is_async = -1;
> >
> > +static CORE_ADDR update_line_range_start (CORE_ADDR pc,
> > + struct
> > execution_control_state *ecs);
> > +
> > /* See infrun.h. */
> >
> > void
> > @@ -6768,6 +6771,25 @@ handle_signal_stop (struct
> > execution_control_state *ecs)
> > process_event_stop_test (ecs);
> > }
> >
> > +CORE_ADDR
> > +update_line_range_start (CORE_ADDR pc, struct
> > execution_control_state *ecs)
> > +{
> > + /* The line table may have multiple entries for the same source
> > code line.
> > + Given the PC, check the line table and return the PC that
> > corresponds
> > + to the line table entry for the source line that PC is
> > in. */
> > + CORE_ADDR start_line_pc = ecs->event_thread-
> > >control.step_range_start;
> > + gdb::optional<CORE_ADDR> real_range_start;
> > +
> > + /* Call find_line_range_start to get smallest address in the
> > + linetable for multiple Line X entries in the line table. */
> > + real_range_start = find_line_range_start (pc);
> > +
> > + if (real_range_start.has_value ())
> > + start_line_pc = *real_range_start;
> > +
> > + return start_line_pc;
> > +}
> > +
> > /* Come here when we've got some debug event / signal we can
> > explain
> > (IOW, not a random signal), and test whether it should cause a
> > stop, or whether we should resume the inferior
> > (transparently).
> > @@ -7569,6 +7591,28 @@ process_event_stop_test (struct
> > execution_control_state *ecs)
> >
> > if (stop_pc_sal.is_stmt)
> > {
> > + if (execution_direction == EXEC_REVERSE)
> > + {
> > + /* We are stepping backwards make sure we have reached
> > the
> > + beginning of the line. */
> > + CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
> > + CORE_ADDR start_line_pc
> > + = update_line_range_start (stop_pc, ecs);
> > +
> > + if (stop_pc != start_line_pc)
> > + {
> > + /* Have not reached the beginning of the source code
> > line.
> > + Set a step range. Execution should stop in any
> > function
> > + calls we execute back into before reaching the
> > beginning
> > + of the line. */
> > + ecs->event_thread->control.step_range_start =
> > start_line_pc;
> > + ecs->event_thread->control.step_range_end = stop_pc;
> > + set_step_info (ecs->event_thread, frame,
> > stop_pc_sal);
> > + keep_going (ecs);
> > + return;
> > + }
> > + }
> > +
> > /* We are at the start of a statement.
> >
> > So stop. Note that we don't stop if we step into the
> > middle of a
> > @@ -7631,6 +7675,19 @@ process_event_stop_test (struct
> > execution_control_state *ecs)
> > set_step_info (ecs->event_thread, frame, stop_pc_sal);
> >
> > infrun_debug_printf ("keep going");
> > +
> > + if (execution_direction == EXEC_REVERSE)
> > + {
> > + CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
> > +
> > + /* Make sure the stop_pc is set to the beginning of the
> > line. */
> > + if (stop_pc != ecs->event_thread->control.step_range_start)
> > + {
> > + stop_pc = update_line_range_start (stop_pc, ecs);
> > + ecs->event_thread->control.step_range_start = stop_pc;
> > + }
> > + }
> > +
> > keep_going (ecs);
> > }
> >
> > diff --git a/gdb/symtab.c b/gdb/symtab.c
> > index 27611a34ec4..91d35616eb9 100644
> > --- a/gdb/symtab.c
> > +++ b/gdb/symtab.c
> > @@ -3282,6 +3282,55 @@ find_pc_line (CORE_ADDR pc, int notcurrent)
> > return sal;
> > }
> >
> > +/* Compare two symtab_and_line entries. Return true if both have
> > + the same line number and the same symtab pointer. That means
> > we
> > + are dealing with two entries from the same line and from the
> > same
> > + source file.
> > +
> > + Return false otherwise. */
> > +
> > +static bool
> > +sal_line_symtab_matches_p (const symtab_and_line &sal1,
> > + const symtab_and_line &sal2)
> > +{
> > + return (sal1.line == sal2.line && sal1.symtab == sal2.symtab);
> > +}
> > +
> > +/* See symtah.h. */
> > +
> > +gdb::optional<CORE_ADDR>
> > +find_line_range_start (CORE_ADDR pc)
> > +{
> > + struct symtab_and_line current_sal = find_pc_line (pc, 0);
> > +
> > + if (current_sal.line == 0)
> > + return {};
> > +
> > + struct symtab_and_line prev_sal = find_pc_line (current_sal.pc -
> > 1, 0);
> > +
> > + /* If the previous entry is for a different line, that means we
> > are already
> > + at the entry with the start PC for this line. */
> > + if (!sal_line_symtab_matches_p (prev_sal, current_sal))
> > + return current_sal.pc;
> > +
> > + /* Otherwise, keep looking for entries for the same line but
> > with
> > + smaller PC's. */
> > + bool done = false;
> > + CORE_ADDR prev_pc;
> > + while (!done)
> > + {
> > + prev_pc = prev_sal.pc;
> > +
> > + prev_sal = find_pc_line (prev_pc - 1, 0);
> > +
> > + /* Did we notice a line change? If so, we are done with the
> > search. */
> > + if (!sal_line_symtab_matches_p (prev_sal, current_sal))
> > + done = true;
> > + }
> > +
> > + return prev_pc;
> > +}
> > +
> > /* See symtab.h. */
> >
> > struct symtab *
> > diff --git a/gdb/symtab.h b/gdb/symtab.h
> > index 404d0ab30a8..f54305636da 100644
> > --- a/gdb/symtab.h
> > +++ b/gdb/symtab.h
> > @@ -2346,6 +2346,22 @@ extern struct symtab_and_line find_pc_line
> > (CORE_ADDR, int);
> > extern struct symtab_and_line find_pc_sect_line (CORE_ADDR,
> > struct obj_section *,
> > int);
> >
> > +/* Given PC, and assuming it is part of a range of addresses that
> > is part of a
> > + line, go back through the linetable and find the starting PC of
> > that
> > + line.
> > +
> > + For example, suppose we have 3 PC ranges for line X:
> > +
> > + Line X - [0x0 - 0x8]
> > + Line X - [0x8 - 0x10]
> > + Line X - [0x10 - 0x18]
> > +
> > + If we call the function with PC == 0x14, we want to return 0x0,
> > as that is
> > + the starting PC of line X, and the ranges are contiguous.
> > +*/
> > +
> > +extern gdb::optional<CORE_ADDR> find_line_range_start (CORE_ADDR
> > pc);
> > +
> > /* Wrapper around find_pc_line to just return the symtab. */
> >
> > extern struct symtab *find_pc_line_symtab (CORE_ADDR);
> > diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line-no-
> > column-info.exp b/gdb/testsuite/gdb.reverse/func-map-to-same-line-
> > no-column-info.exp
> > new file mode 100644
> > index 00000000000..20529c90fc2
> > --- /dev/null
> > +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line-no-column-
> > info.exp
> > @@ -0,0 +1,135 @@
> > +# Copyright 2008-2023 Free Software Foundation, Inc.
> > +
> > +# This program is free software; you can redistribute it and/or
> > modify
> > +# it under the terms of the GNU General Public License as
> > published by
> > +# the Free Software Foundation; either version 3 of the License,
> > or
> > +# (at your option) any later version.
> > +#
> > +# This program is distributed in the hope that it will be useful,
> > +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> > +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> > +# GNU General Public License for more details.
> > +#
> > +# You should have received a copy of the GNU General Public
> > License
> > +# along with this program. If not, see <
> > http://www.gnu.org/licenses/
> > >. */
> > +
> > +# This file is part of the GDB testsuite. It tests reverse
> > stepping.
> > +# Lots of code borrowed from "step-test.exp".
> > +
> > +# This test checks to make sure there is no regression failures
> > for
> > +# the reverse-next command when stepping back over two functions
> > in
> > +# the same line.
> > +
> > +if ![supports_reverse] {
> > + return
> > +}
> Nowadays you should use require instead of the if clause, like in
> gdb.reverse/break-reverse.exp
OK, changed that
> > +
> > +# This test uses the gcc no-column-info command.
> Correct me if I'm wrong, but it seems to me that the other test is a
> more generic version of this one, so this test could check for a gcc
> recent enough to support this feature, instead of just generically
> gcc.
> That said, gcc added it on version
> 7(
> https://gcc.gnu.org/git/gitweb.cgi?p=gcc.git;h=0029b929c9719a
> ), is it
> old enough that we don't care?
GCC supports line tables, I don't know that clang or other compilers
do. So all we really need is to check for gcc. The line table stuff
was added a long time ago so not sure that we really need to check for
version 7 at this point. So just checked that we are using gcc. The
"other test" func-map-to-same-line.exp expects the line table so it
should probably also be checking that we are using gcc.
> > +if ![is_c_compiler_gcc] {
> > + unsupported "gcc is required for this test"
> > + return 0
> > +}
> > +
> > +set srcfile func-map-to-same-line.c
> > +set executable func-map-to-same-line
> > +
> > +set options [list debug additional_flags=-gno-column-info]
> > +
> > +if {[build_executable "failed to prepare" $executable $srcfile
> > $options] == -1}\
> > + {
> > + return -1
> > +}
> > +
> > +clean_restart $executable
> > +
> > +runto_main
> > +set target_remote [gdb_is_target_remote]
> > +
> > +if [supports_process_record] {
> > + # Activate process record/replay.
> > + gdb_test_no_output "record" "turn on process record for test1"
> > +}
> > +
> > +# This regression test verifies the reverse-step and reverse-next
> > commands
> > +# work properly when executing backwards thru a source line
> > containing
> > +# two function calls on the same source line, i.e. func1 (); func2
> > ();
> > +# This test is compiled so the dwarf info not contain the line
> > table
> > +# information.
> > +
> > +# Test 1, reverse-next command
> > +# Set breakpoint at the line after the function calls.
> > +set bp_start_reverse_test [gdb_get_line_number "START REVERSE
> > TEST" $srcfile]
> > +gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> > +
> > +# Continue to break point for reverse-next test.
> > +# Command definition: reverse-next [count]
> > +# Run backward to the beginning of the previous line executed in
> > the current
> > +# (innermost) stack frame. If the line contains function calls,
> > they will be
> > +# “un-executed” without stopping. Starting from the first line
> > of a function,
> > +# reverse-next will take you back to the caller of that
> > function, before the
> > +# function was called, just as the normal next command would
> > take you from
> > +# the last line of a function back to its return to its caller 2
> > .
> > +
> > +gdb_continue_to_breakpoint \
> > + "stopped at command reverse-next test start location" \
> > + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> > +
> > +# The reverse-next should step all the way back to the beginning
> > of the line,
> > +# i.e. at the beginning of the func1 call.
> > +gdb_test "reverse-next" ".*func1 \\(\\); func2 \\(\\);.*" \
> > + "reverse-next to line with two functions"
> > +
> > +# A reverse-step should step back and stop at the beginning
> > +# of the previous line b = 2, i.e. not in func1 ().
> > +gdb_test "reverse-step" ".*b = 2;.*" \
> > + "reverse-step to previous line b = 2"
> The point of this test is to confirm that we are at the very first
> instruction of the line, right? So I would think it is better to do
> a
> reverse-stepi here, to make sure that even walking a single
> instruction
> we reach a different line.
Yes, we should be on the first instruction of the line so stepi would
be a better test to prove we are really on the first instruction.
Changed the reverse-step to reverse-stepi.
> Either that or doing what Pedro did in his
> email: save the PC before executing the line, then do and undo the
> line
> and confirm that PCs match exactly.
> > +
> > +
> > +# Setup for test 2
> > +# Go back to the start of the function
> > +gdb_test "reverse-continue" "a = 1;" "At start of main, setup for
> > test 2"
> > +
> > +# Turn off record to clear logs and turn on again
> > +gdb_test "record stop" "Process record is stopped.*" \
> > + "turn off process record for test1"
> > +gdb_test_no_output "record" "turn on process record for test2"
> Since you don't require process record for this test, you can't
> assume
> these to work. I think its better to clean restart and record if the
> process supports recording, this way you're sure to reset history no
> matter the inferior.
No, the test requires process record. If record is not supported, we
can't do reverse execution. That said, doing a "clean restart" and
then record would be another way, probably better way, of clearing the
history. Put in a clean restart rather than turning off/on the
recording to clear the log file.
> > +
> > +# Delete all breakpoints and catchpoints.
> > +delete_breakpoints
> > +
> > +
> > +# Test 2, reverse-step command
> > +# Set breakpoint at the line after the function calls.
> > +gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> > +
> > +# Continue to the start of the reverse-step test.
> > +# Command definition: reverse-step [count]
> > +# Run the program backward until control reaches the start of a
> > +# different source line; then stop it, and return control to
> > gdb.
> > +# Like the step command, reverse-step will only stop at the
> > beginning of a
> > +# source line. It “un-executes” the previously executed source
> > line. If the
> > +# previous source line included calls to debuggable functions,
> > reverse-step
> > +# will step (backward) into the called function, stopping at
> > the beginning
> > +# of the last statement in the called function (typically a
> > return
> > +# statement). Also, as with the step command, if non-
> > debuggable functions
> > +# are called, reverse-step will run thru them backward without
> > stopping.
> > +
> > +gdb_continue_to_breakpoint \
> > + "stopped at command reverse-step test start location" \
> > + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> > +
> > +# The first reverse step should take us call of func2 ().
> > +gdb_test "reverse-step" ".*}.*" \
> > + "reverse-step into func2 "
> > +
> > +# The second reverse step should take us into func1 ().
> > +gdb_test "reverse-step" ".*}.*" \
> > + "reverse-step into func1 "
> > +
> > +# The third reverse step should take us call of func1 ().
> > +gdb_test "reverse-step" ".*func1 \\(\\); func2 \\(\\);.*" \
> > + "reverse-step to line func1(); func2(), at call for func1 "
> > +
> > +# The fourth reverse step should take us to b = 2 ().
> > +gdb_test "reverse-step" ".*b = 2;.*" \
> > + "reverse-step to line b = 2 "
> Ditto from the other test like this.
Yes, stepi is a better test to prove you were on the first instruction
in the line. Changed.
> Also, I feel that, while the test
> name for the last 2 gdb_test are different, they don't meaningfully
> communicate which part of the test is failing. I think it would be
> better if you differentiated them by adding "for step test" or "for
> test
> 2" at the end of the name would make it easier to understand where
> things went wrong when looking at the sum file.
OK, agreed. For the above tests I added "test#," where # is either 1
or 2 to the test names. For example:
+gdb_test "reverse-step" ".*b = 2;.*" \
+ "reverse-step to line b = 2 "
was changed to
+gdb_test "reverse-stepi" ".*b = 2;.*" \
+ "test2, reverse-step to line b = 2 "
I also updated the test to identify the closing } for func1 and func2
to make it clearer in the test which function we just steped back into.
> > diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> > b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> > new file mode 100644
> > index 00000000000..e9787ef9ff5
> > --- /dev/null
> > +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> > @@ -0,0 +1,36 @@
> > +/* Copyright 2008-2023 Free Software Foundation, Inc.
> > +
> > + This program is free software; you can redistribute it and/or
> > modify
> > + it under the terms of the GNU General Public License as
> > published by
> > + the Free Software Foundation; either version 3 of the License,
> > or
> > + (at your option) any later version.
> > +
> > + This program is distributed in the hope that it will be useful,
> > + but WITHOUT ANY WARRANTY; without even the implied warranty of
> > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> > + GNU General Public License for more details.
> > +
> > + You should have received a copy of the GNU General Public
> > License
> > + along with this program. If not, see <
> > http://www.gnu.org/licenses/
> > >.
> > +
> > + This test is used to test the reverse-step and reverse-next
> > instruction
> > + execution for a source line that contains multiple function
> > calls. */
> > +
> > +void
> > +func1 ()
> > +{
> > +}
> > +
> > +void
> > +func2 ()
> > +{
> > +}
> > +
> > +int main ()
> > +{
> > + int a, b;
> > + a = 1;
> > + b = 2;
> > + func1 (); func2 ();
> > + a = a + b; // START REVERSE TEST
> > +}
> > diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> > b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> > new file mode 100644
> > index 00000000000..b632a236bbe
> > --- /dev/null
> > +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> > @@ -0,0 +1,123 @@
> > +# Copyright 2008-2023 Free Software Foundation, Inc.
> > +
> > +# This program is free software; you can redistribute it and/or
> > modify
> > +# it under the terms of the GNU General Public License as
> > published by
> > +# the Free Software Foundation; either version 3 of the License,
> > or
> > +# (at your option) any later version.
> > +#
> > +# This program is distributed in the hope that it will be useful,
> > +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> > +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> > +# GNU General Public License for more details.
> > +#
> > +# You should have received a copy of the GNU General Public
> > License
> > +# along with this program. If not, see <
> > http://www.gnu.org/licenses/
> > >. */
> > +
> > +# This file is part of the GDB testsuite. It tests reverse
> > stepping.
> > +# Lots of code borrowed from "step-test.exp".
> > +
> > +# This test checks to make sure there is no regression failures
> > for
> > +# the reverse-next command when stepping back over two functions
> > in
> > +# the same line.
> > +
> > +if ![supports_reverse] {
> > + return
> > +}
>
> I'm not sure it's worth separating these 2 tests into separate
> files.
> You could instead just have most of the test defined as a proc, and
> call
> it twice, once after compiling the inferior with column info, the
> other
> compiling without if gcc is used. This way it's less likely that the
> tests will diverge over time.
OK, good point. I created a test proc which is then called with the
binary compiled with and without line information. The extra exp test
file was removed.
>
> > +
> > +standard_testfile
> > +
> > +if { [prepare_for_testing "failed to prepare" $testfile $srcfile]
> > } {
> > + return -1
> > +}
> > +
> > +runto_main
> > +set target_remote [gdb_is_target_remote]
> > +
> > +if [supports_process_record] {
> > + # Activate process record/replay.
> > + gdb_test_no_output "record" "turn on process record for test1"
> > +}
> > +
> > +# This regression test verifies the reverse-step and reverse-next
> > commands
> > +# work properly when executing backwards thru a source line
> > containing
> > +# two function calls on the same source line, i.e. func1 (); func2
> > ();
> > +# The assumption for this test is the dwarf info contain the
> > column
> > +# information.
> > +
> > +# Test 1, reverse-next command
> > +# Set breakpoint at the line after the function calls.
> > +set bp_start_reverse_test [gdb_get_line_number "START REVERSE
> > TEST" $srcfile]
> > +gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> > +
> > +# Continue to break point for reverse-next test.
> > +# Command definition: reverse-next [count]
> > +# Run backward to the beginning of the previous line executed in
> > the current
> > +# (innermost) stack frame. If the line contains function calls,
> > they will be
> > +# “un-executed” without stopping. Starting from the first line
> > of a function,
> > +# reverse-next will take you back to the caller of that
> > function, before the
> > +# function was called, just as the normal next command would
> > take you from
> > +# the last line of a function back to its return to its caller 2
> > .
> > +
> > +gdb_continue_to_breakpoint \
> > + "stopped at command reverse-next test start location" \
> > + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> > +
> > +# The reverse-next should step all the way back to the beginning
> > of the line,
> > +# i.e. at the beginning of the func1 call.
> > +gdb_test "reverse-next" ".*func1 \\(\\); func2 \\(\\);.*" \
> > + "reverse-next to line with two functions"
> > +
> > +# A reverse-step should step back and stop at the beginning
> > +# of the previous line b = 2, i.e. not in func1 ().
> > +gdb_test "reverse-step" ".*b = 2;.*" \
> > + "reverse-step to previous line b = 2"
> > +
> > +
> > +# Setup for test 2
> > +# Go back to the start of the function
> > +gdb_test "reverse-continue" "a = 1;" "At start of main, setup for
> > test 2"
> > +
> > +# Turn off record to clear logs and turn on again
> > +gdb_test "record stop" "Process record is stopped.*" \
> > + "turn off process record for test1"
> > +gdb_test_no_output "record" "turn on process record for test2"
> > +
> > +# Delete all breakpoints and catchpoints.
> > +delete_breakpoints
> > +
> > +
> > +# Test 2, reverse-step command
> > +# Set breakpoint at the line after the function calls.
> > +gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> > +
> > +# Continue to the start of the reverse-step test.
> > +# Command definition: reverse-step [count]
> > +# Run the program backward until control reaches the start of a
> > +# different source line; then stop it, and return control to
> > gdb.
> > +# Like the step command, reverse-step will only stop at the
> > beginning of a
> > +# source line. It “un-executes” the previously executed source
> > line. If the
> > +# previous source line included calls to debuggable functions,
> > reverse-step
> > +# will step (backward) into the called function, stopping at
> > the beginning
> > +# of the last statement in the called function (typically a
> > return
> > +# statement). Also, as with the step command, if non-
> > debuggable functions
> > +# are called, reverse-step will run thru them backward without
> > stopping.
> > +
> > +gdb_continue_to_breakpoint \
> > + "stopped at command reverse-step test start location" \
> > + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> > +
> > +# The first reverse step should take us call of func2 ().
> > +gdb_test "reverse-step" ".*}.*" \
> > + "reverse-step into func2 "
> > +
> > +# The second reverse step should take us into func1 ().
> > +gdb_test "reverse-step" ".*}.*" \
> > + "reverse-step into func1 "
> > +
> > +# The third reverse step should take us call of func1 ().
> > +gdb_test "reverse-step" ".*func1 \\(\\); func2 \\(\\);.*" \
> > + "reverse-step to line func1(); func2(), at call for func1 "
> > +
> > +# The fourth reverse step should take us to b = 2 ().
> > +gdb_test "reverse-step" ".*b = 2;.*" \
> > + "reverse-step to line b = 2 "
> > diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.c
> > b/gdb/testsuite/gdb.reverse/map-to-same-line.c
> > new file mode 100644
> > index 00000000000..f20d778f40e
> > --- /dev/null
> > +++ b/gdb/testsuite/gdb.reverse/map-to-same-line.c
> > @@ -0,0 +1,58 @@
> > +/* Copyright 2008-2023 Free Software Foundation, Inc.
> > +
> > + This program is free software; you can redistribute it and/or
> > modify
> > + it under the terms of the GNU General Public License as
> > published by
> > + the Free Software Foundation; either version 3 of the License,
> > or
> > + (at your option) any later version.
> > +
> > + This program is distributed in the hope that it will be useful,
> > + but WITHOUT ANY WARRANTY; without even the implied warranty of
> > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> > + GNU General Public License for more details.
> > +
> > + You should have received a copy of the GNU General Public
> > License
> > + along with this program. If not, see <
> > http://www.gnu.org/licenses/
> > >. */
> > +
> > +/* The purpose of this test is to create a DWARF line table that
> > contains two
> > + or more entries for the same line. When stepping (forwards or
> > backwards),
> > + GDB should step over the entire line and not just a particular
> > entry in the
> > + line table. */
> > +
> > +int
> > +main ()
> > +{ /* TAG: main prologue */
> > + asm ("main_label: .globl main_label");
> > + int i = 1, j = 2, k;
> > + float f1 = 2.0, f2 = 4.1, f3;
> > + const char *str_1 = "foo", *str_2 = "bar", *str_3;
> > +
> > + asm ("line1: .globl line1");
> > + k = i; f3 = f1; str_3 = str_1; /* TAG: line 1 */
> > +
> > + asm ("line2: .globl line2");
> > + k = j; f3 = f2; str_3 = str_2; /* TAG: line 2 */
> > +
> > + asm ("line3: .globl line3");
> > + k = i; f3 = f1; str_3 = str_1; /* TAG: line 3 */
> > +
> > + asm ("line4: .globl line4");
> > + k = j; f3 = f2; str_3 = str_2; /* TAG: line 4 */
> > +
> > + asm ("line5: .globl line5");
> > + k = i; f3 = f1; str_3 = str_1; /* TAG: line 5 */
> > +
> > + asm ("line6: .globl line6");
> > + k = j; f3 = f2; str_3 = str_2; /* TAG: line 6 */
> > +
> > + asm ("line7: .globl line7");
> > + k = i; f3 = f1; str_3 = str_1; /* TAG: line 7 */
> > +
> > + asm ("line8: .globl line8");
> > + k = j; f3 = f2; str_3 = str_2; /* TAG: line 8 */
> > +
> > + asm ("main_return: .globl main_return");
> > + k = j; f3 = f2; str_3 = str_2; /* TAG: main return */
> > +
> > + asm ("end_of_sequence: .globl end_of_sequence");
> > + return 0; /* TAG: main return */
> > +}
> > diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> > b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> > new file mode 100644
> > index 00000000000..a01579c2a8d
> > --- /dev/null
> > +++ b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> > @@ -0,0 +1,153 @@
> > +# Copyright 2008-2023 Free Software Foundation, Inc.
> > +
> > +# This program is free software; you can redistribute it and/or
> > modify
> > +# it under the terms of the GNU General Public License as
> > published by
> > +# the Free Software Foundation; either version 3 of the License,
> > or
> > +# (at your option) any later version.
> > +#
> > +# This program is distributed in the hope that it will be useful,
> > +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> > +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> > +# GNU General Public License for more details.
> > +#
> > +# You should have received a copy of the GNU General Public
> > License
> > +# along with this program. If not, see <
> > http://www.gnu.org/licenses/
> > = >.
> > +
> > +# When stepping (forwards or backwards), GDB should step over the
> > entire line
> > +# and not just a particular entry in the line table. This test was
> > added to
> > +# verify the find_line_range_start function properly sets the step
> > range for a
> > +# line that consists of multiple statements, i.e. multiple entries
> > in the line
> > +# table. This test creates a DWARF line table that contains two
> > entries for
> > +# the same line to do the needed testing.
> > +
> > +load_lib dwarf.exp
> > +
> > +# This test can only be run on targets which support DWARF-2 and
> > use gas.
> > +if {![dwarf2_support]} {
> > + unsupported "dwarf2 support required for this test"
> > + return 0
> > +}
> Again, the new way to check for these is "required". And IIUC, you
> can
> add multiple requirements into a singe require call.
OK, updated.
> > +
> > +if [get_compiler_info] {
> > + return -1
> > +}
> > +
> > +# The DWARF assembler requires the gcc compiler.
> > +if ![is_c_compiler_gcc] {
> > + unsupported "gcc is required for this test"
> > + return 0
> > +}
> > +
> > +# This test suitable only for process record-replay
> > +if ![supports_process_record] {
> > + return
> > +}
> > +
> > +standard_testfile .c .S
> > +
> > +if { [prepare_for_testing "failed to prepare" ${testfile}
> > ${srcfile}] } {
> > + return -1
> > +}
> > +
> > +set asm_file [standard_output_file $srcfile2]
> > +Dwarf::assemble $asm_file {
> > + global srcdir subdir srcfile
> > + declare_labels integer_label L
> > +
> > + # Find start address and length of program
> > + lassign [function_range main [list
> > ${srcdir}/${subdir}/$srcfile]] \
> > + main_start main_len
> > + set main_end "$main_start + $main_len"
> > +
> > + cu {} {
> > + compile_unit {
> > + {language @DW_LANG_C}
> > + {name map-to-same-line.c}
> > + {stmt_list $L DW_FORM_sec_offset}
> > + {low_pc 0 addr}
> > + } {
> > + subprogram {
> > + {external 1 flag}
> > + {name main}
> > + {low_pc $main_start addr}
> > + {high_pc $main_len DW_FORM_data4}
> > + }
> > + }
> > + }
> > +
> > + lines {version 2 default_is_stmt 1} L {
> > + include_dir "${srcdir}/${subdir}"
> > + file_name "$srcfile" 1
> > +
> > + # Generate the line table program with distinct source lines
> > being
> > + # mapped to the same line entry. Line 1, 5 and 8 contain 1
> > statement
> > + # each. Line 2 contains 2 statements. Line 3 contains 3
> > statements.
> > + program {
> > + DW_LNE_set_address $main_start
> > + line [gdb_get_line_number "TAG: main prologue"]
> > + DW_LNS_copy
> > + DW_LNE_set_address line1
> > + line [gdb_get_line_number "TAG: line 1" ]
> > + DW_LNS_copy
> > + DW_LNE_set_address line2
> > + line [gdb_get_line_number "TAG: line 2" ]
> > + DW_LNS_copy
> > + DW_LNE_set_address line3
> > + line [gdb_get_line_number "TAG: line 2" ]
> > + DW_LNS_copy
> > + DW_LNE_set_address line4
> > + line [gdb_get_line_number "TAG: line 3" ]
> > + DW_LNS_copy
> > + DW_LNE_set_address line5
> > + line [gdb_get_line_number "TAG: line 3" ]
> > + DW_LNS_copy
> > + DW_LNE_set_address line6
> > + line [gdb_get_line_number "TAG: line 3" ]
> > + DW_LNS_copy
> > + DW_LNE_set_address line7
> > + line [gdb_get_line_number "TAG: line 5" ]
> > + DW_LNS_copy
> > + DW_LNE_set_address line8
> > + line [gdb_get_line_number "TAG: line 8" ]
> > + DW_LNS_copy
> > + DW_LNE_set_address main_return
> > + line [gdb_get_line_number "TAG: main return"]
> > + DW_LNS_copy
> > + DW_LNE_set_address end_of_sequence
> > + DW_LNE_end_sequence
> > + }
> > + }
> > +}
> > +
> > +if { [prepare_for_testing "failed to prepare" ${testfile} \
> > + [list $srcfile $asm_file] {nodebug} ] } {
> > + return -1
> > +}
> > +
> > +if ![runto_main] {
> > + return -1
> > +}
> runto_main already errors out and leaves, I think, so no need for the
> if
> clause.
OK, updated.
> > +
> > +# Print the line table
> > +gdb_test_multiple "maint info line-table ${testfile}" "" {
> > + -re "\r\n$decimal\[ \t\]+$decimal\[ \t\]+($hex)\[
> > \t\]+Y\[^\r\n\]*" {
> > + lappend is_stmt $expect_out(1,string)
> > + exp_continue
> > + }
> > + -re -wrap "" {
> > + }
> > +}
> > +
> > +# Activate process record/replay
> > +gdb_test_no_output "record" "turn on process record"
> > +
> > +gdb_test "tbreak main_return" "Temporary breakpoint .*"
> > "breakpoint at return"
> you can set a temporary breakpoint using gdb_breakpoint "..."
> temporary,
> no need to manually call tbreak.
> > +gdb_test "continue" "Temporary breakpoint .*" "run to end of main"
> gdb_continue_to_breakpoint can handle temporary breakpoints as well.
OK, updated the break and continue statements.
> > +gdb_test "display \$pc" ".*pc =.*" "display pc"
> > +
> > +# At this point, GDB has already recorded the execution up until
> > the return
> > +# statement. Reverse-step and test if GDB transitions between
> > lines in the
> > +# expected order. It should reverse-step across lines 8, 5, 3, 2
> > and 1.
> > +foreach line {8 5 3 2 1} {
> > + gdb_test "reverse-step" ".*TAG: line $line.*" "reverse step to
> > line $line"
> > +}
>
> I'm not sure if it is needed, but I don't think it would hurt to
> also
> test reverse-next in a separate foreach right after this one.
OK, added a clean restart, run to the end of main then do reverse next
using a foreach line.
Carl
^ permalink raw reply [flat|nested] 40+ messages in thread
* [PATCH v2] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-05-03 9:53 ` Bruno Larsen
2023-05-04 2:55 ` Carl Love
@ 2023-05-04 2:55 ` Carl Love
2023-05-04 15:59 ` [PATCH v3] " Carl Love
1 sibling, 1 reply; 40+ messages in thread
From: Carl Love @ 2023-05-04 2:55 UTC (permalink / raw)
To: Bruno Larsen, gdb-patches, Ulrich Weigand, pedro; +Cc: luis.machado, cel
Bruno, GDB maintainers:
I believe I have addressed all of Bruno's comments on version 1.
The following patch fixes issues on PowerPC with the reverse-step and
reverse-next instructions when there are multiple assignment statements
on the same line and when there are multiple function calls on the same
line. The commit log below discusses these issues in further depth.
The discussion included what the correct operation should be for these
commands based on the GDB documentation. The proposed patch at that
time changed how the commands worked on other platforms such as X86 in
a way they no longer matched the documentation.
The issue is the line table contains multiple entries for the same
source line. The patch adds a function to search the line table to
find the address of the first instruction of a line. When setup up the
reverse stepping range, the function is called to make sure the start
of the range corresponds to the address of the first instruction for
the line. This approach was used. When Luis initially developed the
patch, he considered merging the contiguous ranges in the line table
when reading the line tables. He decided it was better to work with the
data directly in the line table rather than creating and using a
modified version of the line table.
The following patch fixes the execution of the reveres-step and
reverse-next commands for both senarios of multiple statements on the
same line for PowerPC and aarch64-linux. Unlike the previous patch, it
does not change the operation of the commands on other platforms, i.e.
X86. The patch adds new test cases for both scenarios to verify they
work correctly.
The patch has been tested on PowerPC, Intel X86 and aarch64-linux with
no new regression failures.
Please let me know if the patch is acceptable for mainline. Thanks.
Carl
---------------------------------------------------------------
Fix reverse stepping multiple contiguous PC ranges over the line table.
There are a couple of scenarios where the GDB reverse-step and reverse-next
commands do not work correctly.
Scenario 1 issue description by Luis Machado:
When running GDB's testsuite on aarch64-linux/Ubuntu 20.04 (also spotted on
the ppc backend), I noticed some failures in gdb.reverse/solib-precsave.exp
and gdb.reverse/solib-reverse.exp.
The failure happens around the following code:
38 b[1] = shr2(17); /* middle part two */
40 b[0] = 6; b[1] = 9; /* generic statement, end part two */
42 shr1 ("message 1\n"); /* shr1 one */
Normal execution:
- step from line 38 will land on line 40.
- step from line 40 will land on line 42.
Reverse execution:
- step from line 42 will land on line 40.
- step from line 40 will land on line 40.
- step from line 40 will land on line 38.
The problem here is that line 40 contains two contiguous but distinct
PC ranges in the line table, like so:
Line 40 - [0x7ec ~ 0x7f4]
Line 40 - [0x7f4 ~ 0x7fc]
The two distinct ranges are generated because GCC started outputting source
column information, which GDB doesn't take into account at the moment.
When stepping forward from line 40, we skip both of these ranges and land on
line 42. When stepping backward from line 42, we stop at the start PC of the
second (or first, going backwards) range of line 40.
Since we've reached ecs->event_thread->control.step_range_start, we stop
stepping backwards.
---------------------------------------------------------
Scenario 2 issue described by Pedro Alves:
The following explanation of the issue was taken from the gdb mailing list
discussion of the withdrawn patch to change the behavior of the reverse-step
and reverse-next commands. Specifically, message from Pedro Alves
<pedro@palves.net> where he demonstrates the issue where you have multiple
function calls on the same source code line:
https://sourceware.org/pipermail/gdb-patches/2023-January/196110.html
The source line looks like:
func1 (); func2 ();
so stepping backwards over that line should always stop at the first
instruction of the line, not in the middle. Let's simplify this.
Here's the full source code of my example:
(gdb) list 1
1 void func1 ()
2 {
3 }
4
5 void func2 ()
6 {
7 }
8
9 int main ()
10 {
11 func1 (); func2 ();
12 }
Compiled with:
$ gcc reverse.c -o reverse -g3 -O0
$ gcc -v
...
gcc version 11.3.0 (Ubuntu 11.3.0-1ubuntu1~22.04)
Now let's debug it with target record, using current gdb git master (f3d8ae90b236),
without your patch:
$ gdb ~/reverse
GNU gdb (GDB) 14.0.50.20230124-git
...
Reading symbols from /home/pedro/reverse...
(gdb) start
Temporary breakpoint 1 at 0x1147: file reverse.c, line 11.
Starting program: /home/pedro/reverse
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
Temporary breakpoint 1, main () at reverse.c:11
11 func1 (); func2 ();
(gdb) record
(gdb) disassemble /s
Dump of assembler code for function main:
reverse.c:
10 {
0x000055555555513f <+0>: endbr64
0x0000555555555143 <+4>: push %rbp
0x0000555555555144 <+5>: mov %rsp,%rbp
11 func1 (); func2 ();
=> 0x0000555555555147 <+8>: mov $0x0,%eax
0x000055555555514c <+13>: call 0x555555555129 <func1>
0x0000555555555151 <+18>: mov $0x0,%eax
0x0000555555555156 <+23>: call 0x555555555134 <func2>
0x000055555555515b <+28>: mov $0x0,%eax
12 }
0x0000555555555160 <+33>: pop %rbp
0x0000555555555161 <+34>: ret
End of assembler dump.
(gdb) n
12 }
So far so good, a "next" stepped over the whole of line 11 and stopped at line 12.
Let's confirm where we are now:
(gdb) disassemble /s
Dump of assembler code for function main:
reverse.c:
10 {
0x000055555555513f <+0>: endbr64
0x0000555555555143 <+4>: push %rbp
0x0000555555555144 <+5>: mov %rsp,%rbp
11 func1 (); func2 ();
0x0000555555555147 <+8>: mov $0x0,%eax
0x000055555555514c <+13>: call 0x555555555129 <func1>
0x0000555555555151 <+18>: mov $0x0,%eax
0x0000555555555156 <+23>: call 0x555555555134 <func2>
0x000055555555515b <+28>: mov $0x0,%eax
12 }
=> 0x0000555555555160 <+33>: pop %rbp
0x0000555555555161 <+34>: ret
End of assembler dump.
Good, we're at the first instruction of line 12.
Now let's undo the "next", with "reverse-next":
(gdb) reverse-next
11 func1 (); func2 ();
Seemingly stopped at line 11. Let's see exactly where:
(gdb) disassemble /s
Dump of assembler code for function main:
reverse.c:
10 {
0x000055555555513f <+0>: endbr64
0x0000555555555143 <+4>: push %rbp
0x0000555555555144 <+5>: mov %rsp,%rbp
11 func1 (); func2 ();
0x0000555555555147 <+8>: mov $0x0,%eax
0x000055555555514c <+13>: call 0x555555555129 <func1>
=> 0x0000555555555151 <+18>: mov $0x0,%eax
0x0000555555555156 <+23>: call 0x555555555134 <func2>
0x000055555555515b <+28>: mov $0x0,%eax
12 }
0x0000555555555160 <+33>: pop %rbp
0x0000555555555161 <+34>: ret
End of assembler dump.
(gdb)
And lo, we stopped in the middle of line 11! That is a bug, we should have
stepped back all the way to the beginning of the line. The "reverse-next"
should have fully undone the prior "next" command.
The above issues were fixed by introducing a new function that looks for
adjacent PC ranges for the same line, until we notice a line change. Then
we take that as the start PC of the range. The new start PC for the range
is used for the control.step_range_start when setting up a step range.
The test case gdb.reverse/map-to-same-line.exp is added to test the fix
for the issues in scenario 1.
The test case gdb.reverse/func-map-to-same-line.exp was added to test the
fix for scenario 2 when the binary was compiled with and without line
table information.
bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28426
Co-authored-by: Luis Machado <luis.machado@arm.com>
Co-authored-by: Carl Love <cel@us.ibm.com>
---
gdb/infrun.c | 57 ++++++
gdb/symtab.c | 49 ++++++
gdb/symtab.h | 16 ++
.../gdb.reverse/func-map-to-same-line.c | 36 ++++
.../gdb.reverse/func-map-to-same-line.exp | 155 ++++++++++++++++
gdb/testsuite/gdb.reverse/map-to-same-line.c | 58 ++++++
.../gdb.reverse/map-to-same-line.exp | 166 ++++++++++++++++++
7 files changed, 537 insertions(+)
create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-line.c
create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.c
create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.exp
diff --git a/gdb/infrun.c b/gdb/infrun.c
index efe2c00c489..8555e3c979f 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -114,6 +114,9 @@ static struct async_event_handler *infrun_async_inferior_event_token;
Starts off as -1, indicating "never enabled/disabled". */
static int infrun_is_async = -1;
+static CORE_ADDR update_line_range_start (CORE_ADDR pc,
+ struct execution_control_state *ecs);
+
/* See infrun.h. */
void
@@ -6769,6 +6772,25 @@ handle_signal_stop (struct execution_control_state *ecs)
process_event_stop_test (ecs);
}
+CORE_ADDR
+update_line_range_start (CORE_ADDR pc, struct execution_control_state *ecs)
+{
+ /* The line table may have multiple entries for the same source code line.
+ Given the PC, check the line table and return the PC that corresponds
+ to the line table entry for the source line that PC is in. */
+ CORE_ADDR start_line_pc = ecs->event_thread->control.step_range_start;
+ gdb::optional<CORE_ADDR> real_range_start;
+
+ /* Call find_line_range_start to get smallest address in the
+ linetable for multiple Line X entries in the line table. */
+ real_range_start = find_line_range_start (pc);
+
+ if (real_range_start.has_value ())
+ start_line_pc = *real_range_start;
+
+ return start_line_pc;
+}
+
/* Come here when we've got some debug event / signal we can explain
(IOW, not a random signal), and test whether it should cause a
stop, or whether we should resume the inferior (transparently).
@@ -7570,6 +7592,28 @@ process_event_stop_test (struct execution_control_state *ecs)
if (stop_pc_sal.is_stmt)
{
+ if (execution_direction == EXEC_REVERSE)
+ {
+ /* We are stepping backwards make sure we have reached the
+ beginning of the line. */
+ CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
+ CORE_ADDR start_line_pc
+ = update_line_range_start (stop_pc, ecs);
+
+ if (stop_pc != start_line_pc)
+ {
+ /* Have not reached the beginning of the source code line.
+ Set a step range. Execution should stop in any function
+ calls we execute back into before reaching the beginning
+ of the line. */
+ ecs->event_thread->control.step_range_start = start_line_pc;
+ ecs->event_thread->control.step_range_end = stop_pc;
+ set_step_info (ecs->event_thread, frame, stop_pc_sal);
+ keep_going (ecs);
+ return;
+ }
+ }
+
/* We are at the start of a statement.
So stop. Note that we don't stop if we step into the middle of a
@@ -7632,6 +7676,19 @@ process_event_stop_test (struct execution_control_state *ecs)
set_step_info (ecs->event_thread, frame, stop_pc_sal);
infrun_debug_printf ("keep going");
+
+ if (execution_direction == EXEC_REVERSE)
+ {
+ CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
+
+ /* Make sure the stop_pc is set to the beginning of the line. */
+ if (stop_pc != ecs->event_thread->control.step_range_start)
+ {
+ stop_pc = update_line_range_start (stop_pc, ecs);
+ ecs->event_thread->control.step_range_start = stop_pc;
+ }
+ }
+
keep_going (ecs);
}
diff --git a/gdb/symtab.c b/gdb/symtab.c
index 27611a34ec4..91d35616eb9 100644
--- a/gdb/symtab.c
+++ b/gdb/symtab.c
@@ -3282,6 +3282,55 @@ find_pc_line (CORE_ADDR pc, int notcurrent)
return sal;
}
+/* Compare two symtab_and_line entries. Return true if both have
+ the same line number and the same symtab pointer. That means we
+ are dealing with two entries from the same line and from the same
+ source file.
+
+ Return false otherwise. */
+
+static bool
+sal_line_symtab_matches_p (const symtab_and_line &sal1,
+ const symtab_and_line &sal2)
+{
+ return (sal1.line == sal2.line && sal1.symtab == sal2.symtab);
+}
+
+/* See symtah.h. */
+
+gdb::optional<CORE_ADDR>
+find_line_range_start (CORE_ADDR pc)
+{
+ struct symtab_and_line current_sal = find_pc_line (pc, 0);
+
+ if (current_sal.line == 0)
+ return {};
+
+ struct symtab_and_line prev_sal = find_pc_line (current_sal.pc - 1, 0);
+
+ /* If the previous entry is for a different line, that means we are already
+ at the entry with the start PC for this line. */
+ if (!sal_line_symtab_matches_p (prev_sal, current_sal))
+ return current_sal.pc;
+
+ /* Otherwise, keep looking for entries for the same line but with
+ smaller PC's. */
+ bool done = false;
+ CORE_ADDR prev_pc;
+ while (!done)
+ {
+ prev_pc = prev_sal.pc;
+
+ prev_sal = find_pc_line (prev_pc - 1, 0);
+
+ /* Did we notice a line change? If so, we are done with the search. */
+ if (!sal_line_symtab_matches_p (prev_sal, current_sal))
+ done = true;
+ }
+
+ return prev_pc;
+}
+
/* See symtab.h. */
struct symtab *
diff --git a/gdb/symtab.h b/gdb/symtab.h
index 404d0ab30a8..f54305636da 100644
--- a/gdb/symtab.h
+++ b/gdb/symtab.h
@@ -2346,6 +2346,22 @@ extern struct symtab_and_line find_pc_line (CORE_ADDR, int);
extern struct symtab_and_line find_pc_sect_line (CORE_ADDR,
struct obj_section *, int);
+/* Given PC, and assuming it is part of a range of addresses that is part of a
+ line, go back through the linetable and find the starting PC of that
+ line.
+
+ For example, suppose we have 3 PC ranges for line X:
+
+ Line X - [0x0 - 0x8]
+ Line X - [0x8 - 0x10]
+ Line X - [0x10 - 0x18]
+
+ If we call the function with PC == 0x14, we want to return 0x0, as that is
+ the starting PC of line X, and the ranges are contiguous.
+*/
+
+extern gdb::optional<CORE_ADDR> find_line_range_start (CORE_ADDR pc);
+
/* Wrapper around find_pc_line to just return the symtab. */
extern struct symtab *find_pc_line_symtab (CORE_ADDR);
diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.c b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
new file mode 100644
index 00000000000..412ab180943
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
@@ -0,0 +1,36 @@
+/* Copyright 2008-2023 Free Software Foundation, Inc.
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 3 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program. If not, see <http://www.gnu.org/licenses/>.
+
+ This test is used to test the reverse-step and reverse-next instruction
+ execution for a source line that contains multiple function calls. */
+
+void
+func1 ()
+{
+} // END FUNC1
+
+void
+func2 ()
+{
+} // END FUNC2
+
+int main ()
+{
+ int a, b;
+ a = 1;
+ b = 2;
+ func1 (); func2 ();
+ a = a + b; // START REVERSE TEST
+}
diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
new file mode 100644
index 00000000000..a75673dc99b
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
@@ -0,0 +1,155 @@
+# Copyright 2008-2023 Free Software Foundation, Inc.
+
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program. If not, see <http://www.gnu.org/licenses/>. */
+
+# This file is part of the GDB testsuite. It tests reverse stepping.
+# Lots of code borrowed from "step-test.exp".
+
+# This test checks to make sure there is no regression failures for
+# the reverse-next command when stepping back over two functions in
+# the same line.
+
+require supports_reverse
+
+# This test uses the gcc no-column-info command.
+if ![is_c_compiler_gcc] {
+ unsupported "gcc is required for this test"
+ return 0
+}
+
+proc run_tests {msg} {
+ global srcfile
+ global executable
+
+ runto_main
+ set target_remote [gdb_is_target_remote]
+
+ if [supports_process_record] {
+ # Activate process record/replay.
+ gdb_test_no_output "record" "$msg: turn on process record for test1"
+ }
+
+ # This regression test verifies the reverse-step and reverse-next commands
+ # work properly when executing backwards thru a source line containing
+ # two function calls on the same source line, i.e. func1 (); func2 ();
+ # This test is compiled so the dwarf info not contain the line table
+ # information.
+
+ # Test 1, reverse-next command
+ # Set breakpoint at the line after the function calls.
+ set bp_start_reverse_test [gdb_get_line_number "START REVERSE TEST" \
+ $srcfile]
+ gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
+
+ # Continue to break point for reverse-next test.
+ # Command definition: reverse-next [count]
+ # Run backward to the beginning of the previous line executed in the
+ # current (innermost) stack frame. If the line contains function calls,
+ # they will be “un-executed” without stopping. Starting from the first
+ # line of a function, reverse-next will take you back to the caller of
+ # that function, before the function was called, just as the normal next
+ # command would take you from the last line of a function back to its
+ # return to its caller 2 .
+ gdb_continue_to_breakpoint \
+ "$msg: test1: stopped at command reverse-next test start location" \
+ ".*$srcfile:$bp_start_reverse_test\r\n.*"
+
+ # The reverse-next should step all the way back to the beginning of the
+ # line, i.e. at the beginning of the func1 call.
+ gdb_test "reverse-next" ".*func1 \\(\\); func2 \\(\\);.*" \
+ "$msg: test1: reverse-next to line with two functions"
+
+ # We should be stopped at the first instruction of the line. A reverse-step
+ # should step back and stop at the beginning of the previous line b = 2,
+ # i.e. not in func1 ().
+ gdb_test "reverse-stepi" ".*b = 2;.*" \
+ "$msg: test1: reverse-stepi to previous line b = 2"
+
+
+ # Setup for test 2
+ clean_restart $executable
+ runto_main
+
+ if [supports_process_record] {
+ # Activate process record/replay.
+ gdb_test_no_output "record" "$msg: turn on process record for test2"
+ }
+
+ # Test 2, reverse-step command
+ # Set breakpoint at the line after the function calls.
+ gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
+
+ # Continue to the start of the reverse-step test.
+ # Command definition: reverse-step [count]
+ # Run the program backward until control reaches the start of a
+ # different source line; then stop it, and return control to gdb.
+ # Like the step command, reverse-step will only stop at the beginning
+ # of a source line. It “un-executes” the previously executed source
+ # line. If the previous source line included calls to debuggable
+ # functions, reverse-step will step (backward) into the called function,
+ # stopping at the beginning of the last statement in the called
+ # function (typically a return statement). Also, as with the step
+ # command, if non-debuggable functions are called, reverse-step will
+ # run thru them backward without stopping.
+
+ gdb_continue_to_breakpoint \
+ "$msg: test2: stopped at command reverse-step test start location" \
+ ".*$srcfile:$bp_start_reverse_test\r\n.*"
+
+ # The first reverse step should take us call of func2 ().
+ gdb_test "reverse-step" ".*END FUNC2.*" \
+ "$msg: test2: reverse-step into func2 "
+
+ # The second reverse step should take us into func1 ().
+ gdb_test "reverse-step" ".*END FUNC1.*" \
+ "$msg: test2: reverse-step into func1 "
+
+ # The third reverse step should take us call of func1 ().
+ gdb_test "reverse-step" ".*func1 \\(\\); func2 \\(\\);.*" \
+ "$msg: test2: reverse-step to line func1(); func2(), at call for func1 "
+
+ # We should be stopped at the first instruction of the line. A reverse
+ # stepi should take us to b = 2 ().
+ gdb_test "reverse-stepi" ".*b = 2;.*" \
+ "$msg: test2: reverse-stepi to line b = 2 "
+}
+
+set srcfile func-map-to-same-line.c
+set executable func-map-to-same-line
+
+# test with gcc column info enabled
+set options [list debug additional_flags=]
+
+if {[build_executable "failed to prepare" $executable $srcfile $options] == -1}\
+ {
+ return -1
+}
+
+clean_restart $executable
+
+run_tests {"with-column-info"}
+
+
+#test with gcc column info disabled
+set options [list debug additional_flags=-gno-column-info]
+
+if {[build_executable "failed to prepare" $executable $srcfile $options] == -1}\
+ {
+ return -1
+}
+
+set $executable executable_without_column_info
+clean_restart $executable
+
+run_tests {"no-column-info"}
diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.c b/gdb/testsuite/gdb.reverse/map-to-same-line.c
new file mode 100644
index 00000000000..f20d778f40e
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/map-to-same-line.c
@@ -0,0 +1,58 @@
+/* Copyright 2008-2023 Free Software Foundation, Inc.
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 3 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program. If not, see <http://www.gnu.org/licenses/ >. */
+
+/* The purpose of this test is to create a DWARF line table that contains two
+ or more entries for the same line. When stepping (forwards or backwards),
+ GDB should step over the entire line and not just a particular entry in the
+ line table. */
+
+int
+main ()
+{ /* TAG: main prologue */
+ asm ("main_label: .globl main_label");
+ int i = 1, j = 2, k;
+ float f1 = 2.0, f2 = 4.1, f3;
+ const char *str_1 = "foo", *str_2 = "bar", *str_3;
+
+ asm ("line1: .globl line1");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 1 */
+
+ asm ("line2: .globl line2");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 2 */
+
+ asm ("line3: .globl line3");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 3 */
+
+ asm ("line4: .globl line4");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 4 */
+
+ asm ("line5: .globl line5");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 5 */
+
+ asm ("line6: .globl line6");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 6 */
+
+ asm ("line7: .globl line7");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 7 */
+
+ asm ("line8: .globl line8");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 8 */
+
+ asm ("main_return: .globl main_return");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: main return */
+
+ asm ("end_of_sequence: .globl end_of_sequence");
+ return 0; /* TAG: main return */
+}
diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.exp b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
new file mode 100644
index 00000000000..97a3ba46fdd
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
@@ -0,0 +1,166 @@
+# Copyright 2008-2023 Free Software Foundation, Inc.
+
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program. If not, see <http://www.gnu.org/licenses/ >.
+
+# When stepping (forwards or backwards), GDB should step over the entire line
+# and not just a particular entry in the line table. This test was added to
+# verify the find_line_range_start function properly sets the step range for a
+# line that consists of multiple statements, i.e. multiple entries in the line
+# table. This test creates a DWARF line table that contains two entries for
+# the same line to do the needed testing.
+
+load_lib dwarf.exp
+
+# This test can only be run on targets which support DWARF-2 and use gas.
+require dwarf2_support
+
+if [get_compiler_info] {
+ return -1
+}
+
+# The DWARF assembler requires the gcc compiler.
+if ![is_c_compiler_gcc] {
+ unsupported "gcc is required for this test"
+ return 0
+}
+
+# This test suitable only for process record-replay
+if ![supports_process_record] {
+ return
+}
+
+standard_testfile .c .S
+
+if { [prepare_for_testing "failed to prepare" ${testfile} ${srcfile}] } {
+ return -1
+}
+
+set asm_file [standard_output_file $srcfile2]
+Dwarf::assemble $asm_file {
+ global srcdir subdir srcfile
+ declare_labels integer_label L
+
+ # Find start address and length of program
+ lassign [function_range main [list ${srcdir}/${subdir}/$srcfile]] \
+ main_start main_len
+ set main_end "$main_start + $main_len"
+
+ cu {} {
+ compile_unit {
+ {language @DW_LANG_C}
+ {name map-to-same-line.c}
+ {stmt_list $L DW_FORM_sec_offset}
+ {low_pc 0 addr}
+ } {
+ subprogram {
+ {external 1 flag}
+ {name main}
+ {low_pc $main_start addr}
+ {high_pc $main_len DW_FORM_data4}
+ }
+ }
+ }
+
+ lines {version 2 default_is_stmt 1} L {
+ include_dir "${srcdir}/${subdir}"
+ file_name "$srcfile" 1
+
+ # Generate the line table program with distinct source lines being
+ # mapped to the same line entry. Line 1, 5 and 8 contain 1 statement
+ # each. Line 2 contains 2 statements. Line 3 contains 3 statements.
+ program {
+ DW_LNE_set_address $main_start
+ line [gdb_get_line_number "TAG: main prologue"]
+ DW_LNS_copy
+ DW_LNE_set_address line1
+ line [gdb_get_line_number "TAG: line 1" ]
+ DW_LNS_copy
+ DW_LNE_set_address line2
+ line [gdb_get_line_number "TAG: line 2" ]
+ DW_LNS_copy
+ DW_LNE_set_address line3
+ line [gdb_get_line_number "TAG: line 2" ]
+ DW_LNS_copy
+ DW_LNE_set_address line4
+ line [gdb_get_line_number "TAG: line 3" ]
+ DW_LNS_copy
+ DW_LNE_set_address line5
+ line [gdb_get_line_number "TAG: line 3" ]
+ DW_LNS_copy
+ DW_LNE_set_address line6
+ line [gdb_get_line_number "TAG: line 3" ]
+ DW_LNS_copy
+ DW_LNE_set_address line7
+ line [gdb_get_line_number "TAG: line 5" ]
+ DW_LNS_copy
+ DW_LNE_set_address line8
+ line [gdb_get_line_number "TAG: line 8" ]
+ DW_LNS_copy
+ DW_LNE_set_address main_return
+ line [gdb_get_line_number "TAG: main return"]
+ DW_LNS_copy
+ DW_LNE_set_address end_of_sequence
+ DW_LNE_end_sequence
+ }
+ }
+}
+
+if { [prepare_for_testing "failed to prepare" ${testfile} \
+ [list $srcfile $asm_file] {nodebug} ] } {
+ return -1
+}
+
+runto_main
+
+# Print the line table
+gdb_test_multiple "maint info line-table ${testfile}" "" {
+ -re "\r\n$decimal\[ \t\]+$decimal\[ \t\]+($hex)\[ \t\]+Y\[^\r\n\]*" {
+ lappend is_stmt $expect_out(1,string)
+ exp_continue
+ }
+ -re -wrap "" {
+ }
+}
+
+# Do the reverse-step test
+gdb_test_no_output "record" "turn on process record"
+
+set bp_main_return [gdb_get_line_number "TAG: main return" $srcfile]
+gdb_breakpoint $srcfile:$bp_main_return
+gdb_continue_to_breakpoint "run to end of main, reverse-step test" ".*$srcfile:$bp_main_return.*"
+gdb_test "display \$pc" ".*pc =.*" "display pc, reverse-step test"
+
+# At this point, GDB has already recorded the execution up until the return
+# statement. Reverse-step and test if GDB transitions between lines in the
+# expected order. It should reverse-step across lines 8, 5, 3, 2 and 1.
+foreach line {8 5 3 2 1} {
+ gdb_test "reverse-step" ".*TAG: line $line.*" "reverse step to line $line"
+}
+
+## Clean restart, test reverse-next command
+clean_restart ${testfile}
+runto_main
+gdb_test_no_output "record" "turn on process record, reverst-next test"
+
+set bp_main_return [gdb_get_line_number "TAG: main return" $srcfile]
+gdb_breakpoint $srcfile:$bp_main_return
+gdb_continue_to_breakpoint "run to end of main, reverse-next test" ".*$srcfile:$bp_main_return.*"
+gdb_test "display \$pc" ".*pc =.*" "display pc, reverse-next test"
+
+# At this point, GDB has already recorded the execution up until the return
+# statement. Reverse-next and test if GDB transitions between lines in the
+# expected order. It should reverse-next across lines 8, 5, 3, 2 and 1.
+foreach line {8 5 3 2 1} {
+ gdb_test "reverse-next" ".*TAG: line $line.*" "reverse next to line $line"
+}
--
2.37.2
^ permalink raw reply [flat|nested] 40+ messages in thread
* Re: [PATCH] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-05-04 2:55 ` Carl Love
@ 2023-05-04 9:24 ` Bruno Larsen
2023-05-04 14:52 ` Carl Love
0 siblings, 1 reply; 40+ messages in thread
From: Bruno Larsen @ 2023-05-04 9:24 UTC (permalink / raw)
To: Carl Love, gdb-patches, Ulrich Weigand, pedro; +Cc: luis.machado
On 04/05/2023 04:55, Carl Love wrote:
> Bruno:
>
> On Wed, 2023-05-03 at 11:53 +0200, Bruno Larsen wrote:
>> On 27/04/2023 22:59, Carl Love wrote:
> <snip>
<snip>
>>> +
>>> +# This test uses the gcc no-column-info command.
>> Correct me if I'm wrong, but it seems to me that the other test is a
>> more generic version of this one, so this test could check for a gcc
>> recent enough to support this feature, instead of just generically
>> gcc.
>> That said, gcc added it on version
>> 7(
>> https://gcc.gnu.org/git/gitweb.cgi?p=gcc.git;h=0029b929c9719a
>> ), is it
>> old enough that we don't care?
> GCC supports line tables, I don't know that clang or other compilers
> do. So all we really need is to check for gcc. The line table stuff
> was added a long time ago so not sure that we really need to check for
> version 7 at this point. So just checked that we are using gcc. The
> "other test" func-map-to-same-line.exp expects the line table so it
> should probably also be checking that we are using gcc.
GCC 7.1 (first gcc 7 release) was on May 2nd 2017, almost exactly 6
years ago, and there was a gcc 6.8 release in october 2018. I don't know
if 5 years is long enough to assume that everyone has abandoned the old
version (especially seeing as we sometimes test for gcc 4 or 3, but that
might just be old cruft). That said, it's not a blocker for me, so /shrug
Also, clang will have line tables - otherwise almost nothing on our test
suite would work. It doesn't have column info, though, which is why I'm
fine with it being ignored in the test that uses -gno-column-info.
>
>>> +if ![is_c_compiler_gcc] {
>>> + unsupported "gcc is required for this test"
>>> + return 0
>>> +}
>>> +
>>> +set srcfile func-map-to-same-line.c
>>> +set executable func-map-to-same-line
>>> +
>>> +set options [list debug additional_flags=-gno-column-info]
>>> +
>>> +if {[build_executable "failed to prepare" $executable $srcfile
>>> $options] == -1}\
>>> + {
>>> + return -1
>>> +}
>>> +
>>> +clean_restart $executable
>>> +
>>> +runto_main
>>> +set target_remote [gdb_is_target_remote]
>>> +
>>> +if [supports_process_record] {
>>> + # Activate process record/replay.
>>> + gdb_test_no_output "record" "turn on process record for test1"
>>> +}
>>> +
>>> +# This regression test verifies the reverse-step and reverse-next
>>> commands
>>> +# work properly when executing backwards thru a source line
>>> containing
>>> +# two function calls on the same source line, i.e. func1 (); func2
>>> ();
>>> +# This test is compiled so the dwarf info not contain the line
>>> table
>>> +# information.
>>> +
>>> +# Test 1, reverse-next command
>>> +# Set breakpoint at the line after the function calls.
>>> +set bp_start_reverse_test [gdb_get_line_number "START REVERSE
>>> TEST" $srcfile]
>>> +gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
>>> +
>>> +# Continue to break point for reverse-next test.
>>> +# Command definition: reverse-next [count]
>>> +# Run backward to the beginning of the previous line executed in
>>> the current
>>> +# (innermost) stack frame. If the line contains function calls,
>>> they will be
>>> +# “un-executed” without stopping. Starting from the first line
>>> of a function,
>>> +# reverse-next will take you back to the caller of that
>>> function, before the
>>> +# function was called, just as the normal next command would
>>> take you from
>>> +# the last line of a function back to its return to its caller 2
>>> .
>>> +
>>> +gdb_continue_to_breakpoint \
>>> + "stopped at command reverse-next test start location" \
>>> + ".*$srcfile:$bp_start_reverse_test\r\n.*"
>>> +
>>> +# The reverse-next should step all the way back to the beginning
>>> of the line,
>>> +# i.e. at the beginning of the func1 call.
>>> +gdb_test "reverse-next" ".*func1 \\(\\); func2 \\(\\);.*" \
>>> + "reverse-next to line with two functions"
>>> +
>>> +# A reverse-step should step back and stop at the beginning
>>> +# of the previous line b = 2, i.e. not in func1 ().
>>> +gdb_test "reverse-step" ".*b = 2;.*" \
>>> + "reverse-step to previous line b = 2"
>> The point of this test is to confirm that we are at the very first
>> instruction of the line, right? So I would think it is better to do
>> a
>> reverse-stepi here, to make sure that even walking a single
>> instruction
>> we reach a different line.
> Yes, we should be on the first instruction of the line so stepi would
> be a better test to prove we are really on the first instruction.
> Changed the reverse-step to reverse-stepi.
>
>> Either that or doing what Pedro did in his
>> email: save the PC before executing the line, then do and undo the
>> line
>> and confirm that PCs match exactly.
>>> +
>>> +
>>> +# Setup for test 2
>>> +# Go back to the start of the function
>>> +gdb_test "reverse-continue" "a = 1;" "At start of main, setup for
>>> test 2"
>>> +
>>> +# Turn off record to clear logs and turn on again
>>> +gdb_test "record stop" "Process record is stopped.*" \
>>> + "turn off process record for test1"
>>> +gdb_test_no_output "record" "turn on process record for test2"
>> Since you don't require process record for this test, you can't
>> assume
>> these to work. I think its better to clean restart and record if the
>> process supports recording, this way you're sure to reset history no
>> matter the inferior.
> No, the test requires process record. If record is not supported, we
> can't do reverse execution. That said, doing a "clean restart" and
> then record would be another way, probably better way, of clearing the
> history. Put in a clean restart rather than turning off/on the
> recording to clear the log file.
From my understanding, you could have architectures or inferiors that
support reverse execution without needing to record, that's why
"supports_reverse" and "supports_process_record" are different.
However, if you want to restrict this to record-only, that's fine, I
just think it should be a requirement at the top of the test, not in the
middle of the execution.
--
Cheers,
Bruno
^ permalink raw reply [flat|nested] 40+ messages in thread
* RE: [PATCH] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-05-04 9:24 ` Bruno Larsen
@ 2023-05-04 14:52 ` Carl Love
0 siblings, 0 replies; 40+ messages in thread
From: Carl Love @ 2023-05-04 14:52 UTC (permalink / raw)
To: Bruno Larsen, gdb-patches, Ulrich Weigand, pedro; +Cc: luis.machado, cel
Bruno:
On Thu, 2023-05-04 at 11:24 +0200, Bruno Larsen wrote:
> On 04/05/2023 04:55, Carl Love wrote:
> > Bruno:
> >
> > On Wed, 2023-05-03 at 11:53 +0200, Bruno Larsen wrote:
> > > On 27/04/2023 22:59, Carl Love wrote:
> > <snip>
> <snip>
> > > > +
> > > > +# This test uses the gcc no-column-info command.
> > > Correct me if I'm wrong, but it seems to me that the other test
> > > is a
> > > more generic version of this one, so this test could check for a
> > > gcc
> > > recent enough to support this feature, instead of just
> > > generically
> > > gcc.
> > > That said, gcc added it on version
> > > 7(
> > > https://gcc.gnu.org/git/gitweb.cgi?p=gcc.git;h=0029b929c9719a
> > > ), is it
> > > old enough that we don't care?
> > GCC supports line tables, I don't know that clang or other
> > compilers
> > do. So all we really need is to check for gcc. The line table
> > stuff
> > was added a long time ago so not sure that we really need to check
> > for
> > version 7 at this point. So just checked that we are using
> > gcc. The
> > "other test" func-map-to-same-line.exp expects the line table so it
> > should probably also be checking that we are using gcc.
>
> GCC 7.1 (first gcc 7 release) was on May 2nd 2017, almost exactly 6
> years ago, and there was a gcc 6.8 release in october 2018. I don't
> know
> if 5 years is long enough to assume that everyone has abandoned the
> old
> version (especially seeing as we sometimes test for gcc 4 or 3, but
> that
> might just be old cruft). That said, it's not a blocker for me, so
> /shrug
OK, I would think that it has been there long enough. But it does
sound like there is the occasional test on an old gcc. So, OK lets
restrict it to GCC 7 and later.
>
> Also, clang will have line tables - otherwise almost nothing on our
> test
> suite would work. It doesn't have column info, though, which is why
> I'm
> fine with it being ignored in the test that uses -gno-column-info.
>
> > > > +if ![is_c_compiler_gcc] {
> > > > + unsupported "gcc is required for this test"
> > > > + return 0
> > > > +}
> > > > +
> > > >
<snip>
> From my understanding, you could have architectures or inferiors
> that
> support reverse execution without needing to record, that's why
> "supports_reverse" and "supports_process_record" are different.
>
> However, if you want to restrict this to record-only, that's fine, I
> just think it should be a requirement at the top of the test, not in
> the
> middle of the execution.
I can't imagine how you can do remote execution without recording but
if they can do that OK. The test really should be at the top, missed
that part. So, I will change this to "supports_reverse" and get it at
the top where it belongs.
Thanks for the clarification and help with this.
Carl
^ permalink raw reply [flat|nested] 40+ messages in thread
* [PATCH v3] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-05-04 2:55 ` [PATCH v2] " Carl Love
@ 2023-05-04 15:59 ` Carl Love
2023-05-05 14:59 ` Luis Machado
2023-05-10 13:47 ` Bruno Larsen
0 siblings, 2 replies; 40+ messages in thread
From: Carl Love @ 2023-05-04 15:59 UTC (permalink / raw)
To: Bruno Larsen, gdb-patches, UlrichWeigand, pedro; +Cc: luis.machado, cel
Bruno, GDB maintainers:
Version 3, added the gcc version check as discussed further from
version 2 of the patch. Also updated the tests to check for supporting
reverse execution rather than requiring recording. I also noticed
there were a couple more instances of a requirement check, i.e. if []
which I changed to "require" per the current style for checking on the
test requirements.
The following patch fixes issues on PowerPC with the reverse-step and
reverse-next instructions when there are multiple assignment statements
on the same line and when there are multiple function calls on the same
line. The commit log below discusses these issues in further depth.
The discussion included what the correct operation should be for these
commands based on the GDB documentation. The proposed patch at that
time changed how the commands worked on other platforms such as X86 in
a way they no longer matched the documentation.
The issue is the line table contains multiple entries for the same
source line. The patch adds a function to search the line table to
find the address of the first instruction of a line. When setup up the
reverse stepping range, the function is called to make sure the start
of the range corresponds to the address of the first instruction for
the line. This approach was used. When Luis initially developed the
patch, he considered merging the contiguous ranges in the line table
when reading the line tables. He decided it was better to work with the
data directly in the line table rather than creating and using a
modified version of the line table.
The following patch fixes the execution of the reveres-step and
reverse-next commands for both senarios of multiple statements on the
same line for PowerPC and aarch64-linux. Unlike the previous patch, it
does not change the operation of the commands on other platforms, i.e.
X86. The patch adds new test cases for both scenarios to verify they
work correctly.
The patch has been tested on PowerPC, Intel X86 and aarch64-linux with
no new regression failures.
Please let me know if the patch is acceptable for mainline. Thanks.
Carl
---------------------------------------------------------------
Fix reverse stepping multiple contiguous PC ranges over the line table.
There are a couple of scenarios where the GDB reverse-step and reverse-next
commands do not work correctly.
Scenario 1 issue description by Luis Machado:
When running GDB's testsuite on aarch64-linux/Ubuntu 20.04 (also spotted on
the ppc backend), I noticed some failures in gdb.reverse/solib-precsave.exp
and gdb.reverse/solib-reverse.exp.
The failure happens around the following code:
38 b[1] = shr2(17); /* middle part two */
40 b[0] = 6; b[1] = 9; /* generic statement, end part two */
42 shr1 ("message 1\n"); /* shr1 one */
Normal execution:
- step from line 38 will land on line 40.
- step from line 40 will land on line 42.
Reverse execution:
- step from line 42 will land on line 40.
- step from line 40 will land on line 40.
- step from line 40 will land on line 38.
The problem here is that line 40 contains two contiguous but distinct
PC ranges in the line table, like so:
Line 40 - [0x7ec ~ 0x7f4]
Line 40 - [0x7f4 ~ 0x7fc]
The two distinct ranges are generated because GCC started outputting source
column information, which GDB doesn't take into account at the moment.
When stepping forward from line 40, we skip both of these ranges and land on
line 42. When stepping backward from line 42, we stop at the start PC of the
second (or first, going backwards) range of line 40.
Since we've reached ecs->event_thread->control.step_range_start, we stop
stepping backwards.
---------------------------------------------------------
Scenario 2 issue described by Pedro Alves:
The following explanation of the issue was taken from the gdb mailing list
discussion of the withdrawn patch to change the behavior of the reverse-step
and reverse-next commands. Specifically, message from Pedro Alves
<pedro@palves.net> where he demonstrates the issue where you have multiple
function calls on the same source code line:
https://sourceware.org/pipermail/gdb-patches/2023-January/196110.html
The source line looks like:
func1 (); func2 ();
so stepping backwards over that line should always stop at the first
instruction of the line, not in the middle. Let's simplify this.
Here's the full source code of my example:
(gdb) list 1
1 void func1 ()
2 {
3 }
4
5 void func2 ()
6 {
7 }
8
9 int main ()
10 {
11 func1 (); func2 ();
12 }
Compiled with:
$ gcc reverse.c -o reverse -g3 -O0
$ gcc -v
...
gcc version 11.3.0 (Ubuntu 11.3.0-1ubuntu1~22.04)
Now let's debug it with target record, using current gdb git master (f3d8ae90b236),
without your patch:
$ gdb ~/reverse
GNU gdb (GDB) 14.0.50.20230124-git
...
Reading symbols from /home/pedro/reverse...
(gdb) start
Temporary breakpoint 1 at 0x1147: file reverse.c, line 11.
Starting program: /home/pedro/reverse
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
Temporary breakpoint 1, main () at reverse.c:11
11 func1 (); func2 ();
(gdb) record
(gdb) disassemble /s
Dump of assembler code for function main:
reverse.c:
10 {
0x000055555555513f <+0>: endbr64
0x0000555555555143 <+4>: push %rbp
0x0000555555555144 <+5>: mov %rsp,%rbp
11 func1 (); func2 ();
=> 0x0000555555555147 <+8>: mov $0x0,%eax
0x000055555555514c <+13>: call 0x555555555129 <func1>
0x0000555555555151 <+18>: mov $0x0,%eax
0x0000555555555156 <+23>: call 0x555555555134 <func2>
0x000055555555515b <+28>: mov $0x0,%eax
12 }
0x0000555555555160 <+33>: pop %rbp
0x0000555555555161 <+34>: ret
End of assembler dump.
(gdb) n
12 }
So far so good, a "next" stepped over the whole of line 11 and stopped at line 12.
Let's confirm where we are now:
(gdb) disassemble /s
Dump of assembler code for function main:
reverse.c:
10 {
0x000055555555513f <+0>: endbr64
0x0000555555555143 <+4>: push %rbp
0x0000555555555144 <+5>: mov %rsp,%rbp
11 func1 (); func2 ();
0x0000555555555147 <+8>: mov $0x0,%eax
0x000055555555514c <+13>: call 0x555555555129 <func1>
0x0000555555555151 <+18>: mov $0x0,%eax
0x0000555555555156 <+23>: call 0x555555555134 <func2>
0x000055555555515b <+28>: mov $0x0,%eax
12 }
=> 0x0000555555555160 <+33>: pop %rbp
0x0000555555555161 <+34>: ret
End of assembler dump.
Good, we're at the first instruction of line 12.
Now let's undo the "next", with "reverse-next":
(gdb) reverse-next
11 func1 (); func2 ();
Seemingly stopped at line 11. Let's see exactly where:
(gdb) disassemble /s
Dump of assembler code for function main:
reverse.c:
10 {
0x000055555555513f <+0>: endbr64
0x0000555555555143 <+4>: push %rbp
0x0000555555555144 <+5>: mov %rsp,%rbp
11 func1 (); func2 ();
0x0000555555555147 <+8>: mov $0x0,%eax
0x000055555555514c <+13>: call 0x555555555129 <func1>
=> 0x0000555555555151 <+18>: mov $0x0,%eax
0x0000555555555156 <+23>: call 0x555555555134 <func2>
0x000055555555515b <+28>: mov $0x0,%eax
12 }
0x0000555555555160 <+33>: pop %rbp
0x0000555555555161 <+34>: ret
End of assembler dump.
(gdb)
And lo, we stopped in the middle of line 11! That is a bug, we should have
stepped back all the way to the beginning of the line. The "reverse-next"
should have fully undone the prior "next" command.
The above issues were fixed by introducing a new function that looks for
adjacent PC ranges for the same line, until we notice a line change. Then
we take that as the start PC of the range. The new start PC for the range
is used for the control.step_range_start when setting up a step range.
The test case gdb.reverse/map-to-same-line.exp is added to test the fix
for the issues in scenario 1.
The test case gdb.reverse/func-map-to-same-line.exp was added to test the
fix for scenario 2 when the binary was compiled with and without line
table information.
bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28426
Co-authored-by: Luis Machado <luis.machado@arm.com>
Co-authored-by: Carl Love <cel@us.ibm.com>
---
gdb/infrun.c | 57 +++++++
gdb/symtab.c | 49 ++++++
gdb/symtab.h | 16 ++
.../gdb.reverse/func-map-to-same-line.c | 36 ++++
.../gdb.reverse/func-map-to-same-line.exp | 146 ++++++++++++++++
gdb/testsuite/gdb.reverse/map-to-same-line.c | 58 +++++++
.../gdb.reverse/map-to-same-line.exp | 156 ++++++++++++++++++
7 files changed, 518 insertions(+)
create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-line.c
create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.c
create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.exp
diff --git a/gdb/infrun.c b/gdb/infrun.c
index efe2c00c489..8555e3c979f 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -114,6 +114,9 @@ static struct async_event_handler *infrun_async_inferior_event_token;
Starts off as -1, indicating "never enabled/disabled". */
static int infrun_is_async = -1;
+static CORE_ADDR update_line_range_start (CORE_ADDR pc,
+ struct execution_control_state *ecs);
+
/* See infrun.h. */
void
@@ -6769,6 +6772,25 @@ handle_signal_stop (struct execution_control_state *ecs)
process_event_stop_test (ecs);
}
+CORE_ADDR
+update_line_range_start (CORE_ADDR pc, struct execution_control_state *ecs)
+{
+ /* The line table may have multiple entries for the same source code line.
+ Given the PC, check the line table and return the PC that corresponds
+ to the line table entry for the source line that PC is in. */
+ CORE_ADDR start_line_pc = ecs->event_thread->control.step_range_start;
+ gdb::optional<CORE_ADDR> real_range_start;
+
+ /* Call find_line_range_start to get smallest address in the
+ linetable for multiple Line X entries in the line table. */
+ real_range_start = find_line_range_start (pc);
+
+ if (real_range_start.has_value ())
+ start_line_pc = *real_range_start;
+
+ return start_line_pc;
+}
+
/* Come here when we've got some debug event / signal we can explain
(IOW, not a random signal), and test whether it should cause a
stop, or whether we should resume the inferior (transparently).
@@ -7570,6 +7592,28 @@ process_event_stop_test (struct execution_control_state *ecs)
if (stop_pc_sal.is_stmt)
{
+ if (execution_direction == EXEC_REVERSE)
+ {
+ /* We are stepping backwards make sure we have reached the
+ beginning of the line. */
+ CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
+ CORE_ADDR start_line_pc
+ = update_line_range_start (stop_pc, ecs);
+
+ if (stop_pc != start_line_pc)
+ {
+ /* Have not reached the beginning of the source code line.
+ Set a step range. Execution should stop in any function
+ calls we execute back into before reaching the beginning
+ of the line. */
+ ecs->event_thread->control.step_range_start = start_line_pc;
+ ecs->event_thread->control.step_range_end = stop_pc;
+ set_step_info (ecs->event_thread, frame, stop_pc_sal);
+ keep_going (ecs);
+ return;
+ }
+ }
+
/* We are at the start of a statement.
So stop. Note that we don't stop if we step into the middle of a
@@ -7632,6 +7676,19 @@ process_event_stop_test (struct execution_control_state *ecs)
set_step_info (ecs->event_thread, frame, stop_pc_sal);
infrun_debug_printf ("keep going");
+
+ if (execution_direction == EXEC_REVERSE)
+ {
+ CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
+
+ /* Make sure the stop_pc is set to the beginning of the line. */
+ if (stop_pc != ecs->event_thread->control.step_range_start)
+ {
+ stop_pc = update_line_range_start (stop_pc, ecs);
+ ecs->event_thread->control.step_range_start = stop_pc;
+ }
+ }
+
keep_going (ecs);
}
diff --git a/gdb/symtab.c b/gdb/symtab.c
index 27611a34ec4..91d35616eb9 100644
--- a/gdb/symtab.c
+++ b/gdb/symtab.c
@@ -3282,6 +3282,55 @@ find_pc_line (CORE_ADDR pc, int notcurrent)
return sal;
}
+/* Compare two symtab_and_line entries. Return true if both have
+ the same line number and the same symtab pointer. That means we
+ are dealing with two entries from the same line and from the same
+ source file.
+
+ Return false otherwise. */
+
+static bool
+sal_line_symtab_matches_p (const symtab_and_line &sal1,
+ const symtab_and_line &sal2)
+{
+ return (sal1.line == sal2.line && sal1.symtab == sal2.symtab);
+}
+
+/* See symtah.h. */
+
+gdb::optional<CORE_ADDR>
+find_line_range_start (CORE_ADDR pc)
+{
+ struct symtab_and_line current_sal = find_pc_line (pc, 0);
+
+ if (current_sal.line == 0)
+ return {};
+
+ struct symtab_and_line prev_sal = find_pc_line (current_sal.pc - 1, 0);
+
+ /* If the previous entry is for a different line, that means we are already
+ at the entry with the start PC for this line. */
+ if (!sal_line_symtab_matches_p (prev_sal, current_sal))
+ return current_sal.pc;
+
+ /* Otherwise, keep looking for entries for the same line but with
+ smaller PC's. */
+ bool done = false;
+ CORE_ADDR prev_pc;
+ while (!done)
+ {
+ prev_pc = prev_sal.pc;
+
+ prev_sal = find_pc_line (prev_pc - 1, 0);
+
+ /* Did we notice a line change? If so, we are done with the search. */
+ if (!sal_line_symtab_matches_p (prev_sal, current_sal))
+ done = true;
+ }
+
+ return prev_pc;
+}
+
/* See symtab.h. */
struct symtab *
diff --git a/gdb/symtab.h b/gdb/symtab.h
index 404d0ab30a8..f54305636da 100644
--- a/gdb/symtab.h
+++ b/gdb/symtab.h
@@ -2346,6 +2346,22 @@ extern struct symtab_and_line find_pc_line (CORE_ADDR, int);
extern struct symtab_and_line find_pc_sect_line (CORE_ADDR,
struct obj_section *, int);
+/* Given PC, and assuming it is part of a range of addresses that is part of a
+ line, go back through the linetable and find the starting PC of that
+ line.
+
+ For example, suppose we have 3 PC ranges for line X:
+
+ Line X - [0x0 - 0x8]
+ Line X - [0x8 - 0x10]
+ Line X - [0x10 - 0x18]
+
+ If we call the function with PC == 0x14, we want to return 0x0, as that is
+ the starting PC of line X, and the ranges are contiguous.
+*/
+
+extern gdb::optional<CORE_ADDR> find_line_range_start (CORE_ADDR pc);
+
/* Wrapper around find_pc_line to just return the symtab. */
extern struct symtab *find_pc_line_symtab (CORE_ADDR);
diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.c b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
new file mode 100644
index 00000000000..412ab180943
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
@@ -0,0 +1,36 @@
+/* Copyright 2008-2023 Free Software Foundation, Inc.
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 3 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program. If not, see <http://www.gnu.org/licenses/>.
+
+ This test is used to test the reverse-step and reverse-next instruction
+ execution for a source line that contains multiple function calls. */
+
+void
+func1 ()
+{
+} // END FUNC1
+
+void
+func2 ()
+{
+} // END FUNC2
+
+int main ()
+{
+ int a, b;
+ a = 1;
+ b = 2;
+ func1 (); func2 ();
+ a = a + b; // START REVERSE TEST
+}
diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
new file mode 100644
index 00000000000..4eae042a6bf
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
@@ -0,0 +1,146 @@
+# Copyright 2008-2023 Free Software Foundation, Inc.
+
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program. If not, see <http://www.gnu.org/licenses/>. */
+
+# This file is part of the GDB testsuite. It tests reverse stepping.
+# Lots of code borrowed from "step-test.exp".
+
+# This test checks to make sure there is no regression failures for
+# the reverse-next command when stepping back over two functions in
+# the same line.
+
+require supports_reverse
+
+# This test uses the gcc no-column-info command which was added in gcc 7.1.
+require get_compiler_info "gcc-7-*"
+
+proc run_tests {msg} {
+ global srcfile
+ global executable
+
+ runto_main
+ set target_remote [gdb_is_target_remote]
+
+ gdb_test_no_output "record" "turn on process record"
+
+ # This regression test verifies the reverse-step and reverse-next commands
+ # work properly when executing backwards thru a source line containing
+ # two function calls on the same source line, i.e. func1 (); func2 ();
+ # This test is compiled so the dwarf info not contain the line table
+ # information.
+
+ # Test 1, reverse-next command
+ # Set breakpoint at the line after the function calls.
+ set bp_start_reverse_test [gdb_get_line_number "START REVERSE TEST" \
+ $srcfile]
+ gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
+
+ # Continue to break point for reverse-next test.
+ # Command definition: reverse-next [count]
+ # Run backward to the beginning of the previous line executed in the
+ # current (innermost) stack frame. If the line contains function calls,
+ # they will be “un-executed” without stopping. Starting from the first
+ # line of a function, reverse-next will take you back to the caller of
+ # that function, before the function was called, just as the normal next
+ # command would take you from the last line of a function back to its
+ # return to its caller 2 .
+ gdb_continue_to_breakpoint \
+ "$msg: test1: stopped at command reverse-next test start location" \
+ ".*$srcfile:$bp_start_reverse_test\r\n.*"
+
+ # The reverse-next should step all the way back to the beginning of the
+ # line, i.e. at the beginning of the func1 call.
+ gdb_test "reverse-next" ".*func1 \\(\\); func2 \\(\\);.*" \
+ "$msg: test1: reverse-next to line with two functions"
+
+ # We should be stopped at the first instruction of the line. A reverse-step
+ # should step back and stop at the beginning of the previous line b = 2,
+ # i.e. not in func1 ().
+ gdb_test "reverse-stepi" ".*b = 2;.*" \
+ "$msg: test1: reverse-stepi to previous line b = 2"
+
+
+ # Setup for test 2
+ clean_restart $executable
+ runto_main
+
+ gdb_test_no_output "record" "turn on process record"
+
+ # Test 2, reverse-step command
+ # Set breakpoint at the line after the function calls.
+ gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
+
+ # Continue to the start of the reverse-step test.
+ # Command definition: reverse-step [count]
+ # Run the program backward until control reaches the start of a
+ # different source line; then stop it, and return control to gdb.
+ # Like the step command, reverse-step will only stop at the beginning
+ # of a source line. It “un-executes” the previously executed source
+ # line. If the previous source line included calls to debuggable
+ # functions, reverse-step will step (backward) into the called function,
+ # stopping at the beginning of the last statement in the called
+ # function (typically a return statement). Also, as with the step
+ # command, if non-debuggable functions are called, reverse-step will
+ # run thru them backward without stopping.
+
+ gdb_continue_to_breakpoint \
+ "$msg: test2: stopped at command reverse-step test start location" \
+ ".*$srcfile:$bp_start_reverse_test\r\n.*"
+
+ # The first reverse step should take us call of func2 ().
+ gdb_test "reverse-step" ".*END FUNC2.*" \
+ "$msg: test2: reverse-step into func2 "
+
+ # The second reverse step should take us into func1 ().
+ gdb_test "reverse-step" ".*END FUNC1.*" \
+ "$msg: test2: reverse-step into func1 "
+
+ # The third reverse step should take us call of func1 ().
+ gdb_test "reverse-step" ".*func1 \\(\\); func2 \\(\\);.*" \
+ "$msg: test2: reverse-step to line func1(); func2(), at call for func1 "
+
+ # We should be stopped at the first instruction of the line. A reverse
+ # stepi should take us to b = 2 ().
+ gdb_test "reverse-stepi" ".*b = 2;.*" \
+ "$msg: test2: reverse-stepi to line b = 2 "
+}
+
+set srcfile func-map-to-same-line.c
+set executable func-map-to-same-line
+
+# test with gcc column info enabled
+set options [list debug additional_flags=]
+
+if {[build_executable "failed to prepare" $executable $srcfile $options] == -1}\
+ {
+ return -1
+}
+
+clean_restart $executable
+
+run_tests {"with-column-info"}
+
+
+#test with gcc column info disabled
+set options [list debug additional_flags=-gno-column-info]
+
+if {[build_executable "failed to prepare" $executable $srcfile $options] == -1}\
+ {
+ return -1
+}
+
+set $executable executable_without_column_info
+clean_restart $executable
+
+run_tests {"no-column-info"}
diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.c b/gdb/testsuite/gdb.reverse/map-to-same-line.c
new file mode 100644
index 00000000000..f20d778f40e
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/map-to-same-line.c
@@ -0,0 +1,58 @@
+/* Copyright 2008-2023 Free Software Foundation, Inc.
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 3 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program. If not, see <http://www.gnu.org/licenses/ >. */
+
+/* The purpose of this test is to create a DWARF line table that contains two
+ or more entries for the same line. When stepping (forwards or backwards),
+ GDB should step over the entire line and not just a particular entry in the
+ line table. */
+
+int
+main ()
+{ /* TAG: main prologue */
+ asm ("main_label: .globl main_label");
+ int i = 1, j = 2, k;
+ float f1 = 2.0, f2 = 4.1, f3;
+ const char *str_1 = "foo", *str_2 = "bar", *str_3;
+
+ asm ("line1: .globl line1");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 1 */
+
+ asm ("line2: .globl line2");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 2 */
+
+ asm ("line3: .globl line3");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 3 */
+
+ asm ("line4: .globl line4");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 4 */
+
+ asm ("line5: .globl line5");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 5 */
+
+ asm ("line6: .globl line6");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 6 */
+
+ asm ("line7: .globl line7");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 7 */
+
+ asm ("line8: .globl line8");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 8 */
+
+ asm ("main_return: .globl main_return");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: main return */
+
+ asm ("end_of_sequence: .globl end_of_sequence");
+ return 0; /* TAG: main return */
+}
diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.exp b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
new file mode 100644
index 00000000000..02f3f4d8c9c
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
@@ -0,0 +1,156 @@
+# Copyright 2008-2023 Free Software Foundation, Inc.
+
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program. If not, see <http://www.gnu.org/licenses/ >.
+
+# When stepping (forwards or backwards), GDB should step over the entire line
+# and not just a particular entry in the line table. This test was added to
+# verify the find_line_range_start function properly sets the step range for a
+# line that consists of multiple statements, i.e. multiple entries in the line
+# table. This test creates a DWARF line table that contains two entries for
+# the same line to do the needed testing.
+
+# This test can only be run on targets which support DWARF-2 and use gas.
+require dwarf2_support
+load_lib dwarf.exp
+
+# The DWARF assembler requires the gcc compiler.
+require is_c_compiler_gcc
+
+# This test suitable only for process that can do reverse execution
+requires supports_reverse
+
+standard_testfile .c .S
+
+if { [prepare_for_testing "failed to prepare" ${testfile} ${srcfile}] } {
+ return -1
+}
+
+set asm_file [standard_output_file $srcfile2]
+Dwarf::assemble $asm_file {
+ global srcdir subdir srcfile
+ declare_labels integer_label L
+
+ # Find start address and length of program
+ lassign [function_range main [list ${srcdir}/${subdir}/$srcfile]] \
+ main_start main_len
+ set main_end "$main_start + $main_len"
+
+ cu {} {
+ compile_unit {
+ {language @DW_LANG_C}
+ {name map-to-same-line.c}
+ {stmt_list $L DW_FORM_sec_offset}
+ {low_pc 0 addr}
+ } {
+ subprogram {
+ {external 1 flag}
+ {name main}
+ {low_pc $main_start addr}
+ {high_pc $main_len DW_FORM_data4}
+ }
+ }
+ }
+
+ lines {version 2 default_is_stmt 1} L {
+ include_dir "${srcdir}/${subdir}"
+ file_name "$srcfile" 1
+
+ # Generate the line table program with distinct source lines being
+ # mapped to the same line entry. Line 1, 5 and 8 contain 1 statement
+ # each. Line 2 contains 2 statements. Line 3 contains 3 statements.
+ program {
+ DW_LNE_set_address $main_start
+ line [gdb_get_line_number "TAG: main prologue"]
+ DW_LNS_copy
+ DW_LNE_set_address line1
+ line [gdb_get_line_number "TAG: line 1" ]
+ DW_LNS_copy
+ DW_LNE_set_address line2
+ line [gdb_get_line_number "TAG: line 2" ]
+ DW_LNS_copy
+ DW_LNE_set_address line3
+ line [gdb_get_line_number "TAG: line 2" ]
+ DW_LNS_copy
+ DW_LNE_set_address line4
+ line [gdb_get_line_number "TAG: line 3" ]
+ DW_LNS_copy
+ DW_LNE_set_address line5
+ line [gdb_get_line_number "TAG: line 3" ]
+ DW_LNS_copy
+ DW_LNE_set_address line6
+ line [gdb_get_line_number "TAG: line 3" ]
+ DW_LNS_copy
+ DW_LNE_set_address line7
+ line [gdb_get_line_number "TAG: line 5" ]
+ DW_LNS_copy
+ DW_LNE_set_address line8
+ line [gdb_get_line_number "TAG: line 8" ]
+ DW_LNS_copy
+ DW_LNE_set_address main_return
+ line [gdb_get_line_number "TAG: main return"]
+ DW_LNS_copy
+ DW_LNE_set_address end_of_sequence
+ DW_LNE_end_sequence
+ }
+ }
+}
+
+if { [prepare_for_testing "failed to prepare" ${testfile} \
+ [list $srcfile $asm_file] {nodebug} ] } {
+ return -1
+}
+
+runto_main
+
+# Print the line table
+gdb_test_multiple "maint info line-table ${testfile}" "" {
+ -re "\r\n$decimal\[ \t\]+$decimal\[ \t\]+($hex)\[ \t\]+Y\[^\r\n\]*" {
+ lappend is_stmt $expect_out(1,string)
+ exp_continue
+ }
+ -re -wrap "" {
+ }
+}
+
+# Do the reverse-step test
+gdb_test_no_output "record" "turn on process record"
+
+set bp_main_return [gdb_get_line_number "TAG: main return" $srcfile]
+gdb_breakpoint $srcfile:$bp_main_return
+gdb_continue_to_breakpoint "run to end of main, reverse-step test" ".*$srcfile:$bp_main_return.*"
+gdb_test "display \$pc" ".*pc =.*" "display pc, reverse-step test"
+
+# At this point, GDB has already recorded the execution up until the return
+# statement. Reverse-step and test if GDB transitions between lines in the
+# expected order. It should reverse-step across lines 8, 5, 3, 2 and 1.
+foreach line {8 5 3 2 1} {
+ gdb_test "reverse-step" ".*TAG: line $line.*" "reverse step to line $line"
+}
+
+## Clean restart, test reverse-next command
+clean_restart ${testfile}
+runto_main
+gdb_test_no_output "record" "turn on process record, reverst-next test"
+
+set bp_main_return [gdb_get_line_number "TAG: main return" $srcfile]
+gdb_breakpoint $srcfile:$bp_main_return
+gdb_continue_to_breakpoint "run to end of main, reverse-next test" ".*$srcfile:$bp_main_return.*"
+gdb_test "display \$pc" ".*pc =.*" "display pc, reverse-next test"
+
+# At this point, GDB has already recorded the execution up until the return
+# statement. Reverse-next and test if GDB transitions between lines in the
+# expected order. It should reverse-next across lines 8, 5, 3, 2 and 1.
+foreach line {8 5 3 2 1} {
+ gdb_test "reverse-next" ".*TAG: line $line.*" "reverse next to line $line"
+}
--
2.37.2
^ permalink raw reply [flat|nested] 40+ messages in thread
* Re: [PATCH v3] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-05-04 15:59 ` [PATCH v3] " Carl Love
@ 2023-05-05 14:59 ` Luis Machado
2023-05-05 16:10 ` Carl Love
2023-05-10 13:47 ` Bruno Larsen
1 sibling, 1 reply; 40+ messages in thread
From: Luis Machado @ 2023-05-05 14:59 UTC (permalink / raw)
To: Carl Love, Bruno Larsen, gdb-patches, UlrichWeigand, pedro
On 5/4/23 16:59, Carl Love wrote:
>
> Bruno, GDB maintainers:
f>
> Version 3, added the gcc version check as discussed further from
> version 2 of the patch. Also updated the tests to check for supporting
> reverse execution rather than requiring recording. I also noticed
> there were a couple more instances of a requirement check, i.e. if []
> which I changed to "require" per the current style for checking on the
> test requirements.
>
>
> The following patch fixes issues on PowerPC with the reverse-step and
> reverse-next instructions when there are multiple assignment statements
> on the same line and when there are multiple function calls on the same
> line. The commit log below discusses these issues in further depth.
> The discussion included what the correct operation should be for these
> commands based on the GDB documentation. The proposed patch at that
> time changed how the commands worked on other platforms such as X86 in
> a way they no longer matched the documentation.
>
> The issue is the line table contains multiple entries for the same
> source line. The patch adds a function to search the line table to
> find the address of the first instruction of a line. When setup up the
> reverse stepping range, the function is called to make sure the start
> of the range corresponds to the address of the first instruction for
> the line. This approach was used. When Luis initially developed the
> patch, he considered merging the contiguous ranges in the line table
> when reading the line tables. He decided it was better to work with the
> data directly in the line table rather than creating and using a
> modified version of the line table.
>
> The following patch fixes the execution of the reveres-step and
> reverse-next commands for both senarios of multiple statements on the
> same line for PowerPC and aarch64-linux. Unlike the previous patch, it
> does not change the operation of the commands on other platforms, i.e.
> X86. The patch adds new test cases for both scenarios to verify they
> work correctly.
>
> The patch has been tested on PowerPC, Intel X86 and aarch64-linux with
> no new regression failures.
>
> Please let me know if the patch is acceptable for mainline. Thanks.
>
> Carl
>
> ---------------------------------------------------------------
> Fix reverse stepping multiple contiguous PC ranges over the line table.
>
> There are a couple of scenarios where the GDB reverse-step and reverse-next
> commands do not work correctly.
>
> Scenario 1 issue description by Luis Machado:
>
> When running GDB's testsuite on aarch64-linux/Ubuntu 20.04 (also spotted on
> the ppc backend), I noticed some failures in gdb.reverse/solib-precsave.exp
> and gdb.reverse/solib-reverse.exp.
>
> The failure happens around the following code:
>
> 38 b[1] = shr2(17); /* middle part two */
> 40 b[0] = 6; b[1] = 9; /* generic statement, end part two */
> 42 shr1 ("message 1\n"); /* shr1 one */
>
> Normal execution:
>
> - step from line 38 will land on line 40.
> - step from line 40 will land on line 42.
>
> Reverse execution:
> - step from line 42 will land on line 40.
> - step from line 40 will land on line 40.
> - step from line 40 will land on line 38.
>
> The problem here is that line 40 contains two contiguous but distinct
> PC ranges in the line table, like so:
>
> Line 40 - [0x7ec ~ 0x7f4]
> Line 40 - [0x7f4 ~ 0x7fc]
>
> The two distinct ranges are generated because GCC started outputting source
> column information, which GDB doesn't take into account at the moment.
>
> When stepping forward from line 40, we skip both of these ranges and land on
> line 42. When stepping backward from line 42, we stop at the start PC of the
> second (or first, going backwards) range of line 40.
>
> Since we've reached ecs->event_thread->control.step_range_start, we stop
> stepping backwards.
>
> ---------------------------------------------------------
>
> Scenario 2 issue described by Pedro Alves:
>
> The following explanation of the issue was taken from the gdb mailing list
> discussion of the withdrawn patch to change the behavior of the reverse-step
> and reverse-next commands. Specifically, message from Pedro Alves
> <pedro@palves.net> where he demonstrates the issue where you have multiple
> function calls on the same source code line:
>
> https://sourceware.org/pipermail/gdb-patches/2023-January/196110.html
>
> The source line looks like:
>
> func1 (); func2 ();
>
> so stepping backwards over that line should always stop at the first
> instruction of the line, not in the middle. Let's simplify this.
>
> Here's the full source code of my example:
>
> (gdb) list 1
> 1 void func1 ()
> 2 {
> 3 }
> 4
> 5 void func2 ()
> 6 {
> 7 }
> 8
> 9 int main ()
> 10 {
> 11 func1 (); func2 ();
> 12 }
>
> Compiled with:
>
> $ gcc reverse.c -o reverse -g3 -O0
> $ gcc -v
> ...
> gcc version 11.3.0 (Ubuntu 11.3.0-1ubuntu1~22.04)
>
> Now let's debug it with target record, using current gdb git master (f3d8ae90b236),
> without your patch:
>
> $ gdb ~/reverse
> GNU gdb (GDB) 14.0.50.20230124-git
> ...
> Reading symbols from /home/pedro/reverse...
> (gdb) start
> Temporary breakpoint 1 at 0x1147: file reverse.c, line 11.
> Starting program: /home/pedro/reverse
> [Thread debugging using libthread_db enabled]
> Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
>
> Temporary breakpoint 1, main () at reverse.c:11
> 11 func1 (); func2 ();
> (gdb) record
>
> (gdb) disassemble /s
> Dump of assembler code for function main:
> reverse.c:
> 10 {
> 0x000055555555513f <+0>: endbr64
> 0x0000555555555143 <+4>: push %rbp
> 0x0000555555555144 <+5>: mov %rsp,%rbp
>
> 11 func1 (); func2 ();
> => 0x0000555555555147 <+8>: mov $0x0,%eax
> 0x000055555555514c <+13>: call 0x555555555129 <func1>
> 0x0000555555555151 <+18>: mov $0x0,%eax
> 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> 0x000055555555515b <+28>: mov $0x0,%eax
>
> 12 }
> 0x0000555555555160 <+33>: pop %rbp
> 0x0000555555555161 <+34>: ret
> End of assembler dump.
>
> (gdb) n
> 12 }
>
> So far so good, a "next" stepped over the whole of line 11 and stopped at line 12.
>
> Let's confirm where we are now:
>
> (gdb) disassemble /s
> Dump of assembler code for function main:
> reverse.c:
> 10 {
> 0x000055555555513f <+0>: endbr64
> 0x0000555555555143 <+4>: push %rbp
> 0x0000555555555144 <+5>: mov %rsp,%rbp
>
> 11 func1 (); func2 ();
> 0x0000555555555147 <+8>: mov $0x0,%eax
> 0x000055555555514c <+13>: call 0x555555555129 <func1>
> 0x0000555555555151 <+18>: mov $0x0,%eax
> 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> 0x000055555555515b <+28>: mov $0x0,%eax
>
> 12 }
> => 0x0000555555555160 <+33>: pop %rbp
> 0x0000555555555161 <+34>: ret
> End of assembler dump.
>
> Good, we're at the first instruction of line 12.
>
> Now let's undo the "next", with "reverse-next":
>
> (gdb) reverse-next
> 11 func1 (); func2 ();
>
> Seemingly stopped at line 11. Let's see exactly where:
>
> (gdb) disassemble /s
> Dump of assembler code for function main:
> reverse.c:
> 10 {
> 0x000055555555513f <+0>: endbr64
> 0x0000555555555143 <+4>: push %rbp
> 0x0000555555555144 <+5>: mov %rsp,%rbp
>
> 11 func1 (); func2 ();
> 0x0000555555555147 <+8>: mov $0x0,%eax
> 0x000055555555514c <+13>: call 0x555555555129 <func1>
> => 0x0000555555555151 <+18>: mov $0x0,%eax
> 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> 0x000055555555515b <+28>: mov $0x0,%eax
>
> 12 }
> 0x0000555555555160 <+33>: pop %rbp
> 0x0000555555555161 <+34>: ret
> End of assembler dump.
> (gdb)
>
> And lo, we stopped in the middle of line 11! That is a bug, we should have
> stepped back all the way to the beginning of the line. The "reverse-next"
> should have fully undone the prior "next" command.
>
> The above issues were fixed by introducing a new function that looks for
> adjacent PC ranges for the same line, until we notice a line change. Then
> we take that as the start PC of the range. The new start PC for the range
> is used for the control.step_range_start when setting up a step range.
>
> The test case gdb.reverse/map-to-same-line.exp is added to test the fix
> for the issues in scenario 1.
>
> The test case gdb.reverse/func-map-to-same-line.exp was added to test the
> fix for scenario 2 when the binary was compiled with and without line
> table information.
>
> bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28426
>
> Co-authored-by: Luis Machado <luis.machado@arm.com>
> Co-authored-by: Carl Love <cel@us.ibm.com>
> ---
> gdb/infrun.c | 57 +++++++
> gdb/symtab.c | 49 ++++++
> gdb/symtab.h | 16 ++
> .../gdb.reverse/func-map-to-same-line.c | 36 ++++
> .../gdb.reverse/func-map-to-same-line.exp | 146 ++++++++++++++++
> gdb/testsuite/gdb.reverse/map-to-same-line.c | 58 +++++++
> .../gdb.reverse/map-to-same-line.exp | 156 ++++++++++++++++++
> 7 files changed, 518 insertions(+)
> create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.c
> create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.exp
>
> diff --git a/gdb/infrun.c b/gdb/infrun.c
> index efe2c00c489..8555e3c979f 100644
> --- a/gdb/infrun.c
> +++ b/gdb/infrun.c
> @@ -114,6 +114,9 @@ static struct async_event_handler *infrun_async_inferior_event_token;
> Starts off as -1, indicating "never enabled/disabled". */
> static int infrun_is_async = -1;
>
> +static CORE_ADDR update_line_range_start (CORE_ADDR pc,
> + struct execution_control_state *ecs);
> +
> /* See infrun.h. */
>
> void
> @@ -6769,6 +6772,25 @@ handle_signal_stop (struct execution_control_state *ecs)
> process_event_stop_test (ecs);
> }
>
> +CORE_ADDR
> +update_line_range_start (CORE_ADDR pc, struct execution_control_state *ecs)
> +{
> + /* The line table may have multiple entries for the same source code line.
> + Given the PC, check the line table and return the PC that corresponds
> + to the line table entry for the source line that PC is in. */
> + CORE_ADDR start_line_pc = ecs->event_thread->control.step_range_start;
> + gdb::optional<CORE_ADDR> real_range_start;
> +
> + /* Call find_line_range_start to get smallest address in the
s/smallest/the smallest
> + linetable for multiple Line X entries in the line table. */
> + real_range_start = find_line_range_start (pc);
> +
> + if (real_range_start.has_value ())
> + start_line_pc = *real_range_start;
> +
> + return start_line_pc;
> +}
> +
> /* Come here when we've got some debug event / signal we can explain
> (IOW, not a random signal), and test whether it should cause a
> stop, or whether we should resume the inferior (transparently).
> @@ -7570,6 +7592,28 @@ process_event_stop_test (struct execution_control_state *ecs)
>
> if (stop_pc_sal.is_stmt)
> {
> + if (execution_direction == EXEC_REVERSE)
> + {
> + /* We are stepping backwards make sure we have reached the
> + beginning of the line. */
> + CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
> + CORE_ADDR start_line_pc
> + = update_line_range_start (stop_pc, ecs);
> +
> + if (stop_pc != start_line_pc)
> + {
> + /* Have not reached the beginning of the source code line.
> + Set a step range. Execution should stop in any function
> + calls we execute back into before reaching the beginning
> + of the line. */
> + ecs->event_thread->control.step_range_start = start_line_pc;
> + ecs->event_thread->control.step_range_end = stop_pc;
> + set_step_info (ecs->event_thread, frame, stop_pc_sal);
> + keep_going (ecs);
> + return;
> + }
> + }
> +
> /* We are at the start of a statement.
>
> So stop. Note that we don't stop if we step into the middle of a
> @@ -7632,6 +7676,19 @@ process_event_stop_test (struct execution_control_state *ecs)
> set_step_info (ecs->event_thread, frame, stop_pc_sal);
>
> infrun_debug_printf ("keep going");
> +
> + if (execution_direction == EXEC_REVERSE)
> + {
> + CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
> +
> + /* Make sure the stop_pc is set to the beginning of the line. */
> + if (stop_pc != ecs->event_thread->control.step_range_start)
> + {
> + stop_pc = update_line_range_start (stop_pc, ecs);
> + ecs->event_thread->control.step_range_start = stop_pc;
> + }
> + }
> +
> keep_going (ecs);
> }
>
> diff --git a/gdb/symtab.c b/gdb/symtab.c
> index 27611a34ec4..91d35616eb9 100644
> --- a/gdb/symtab.c
> +++ b/gdb/symtab.c
> @@ -3282,6 +3282,55 @@ find_pc_line (CORE_ADDR pc, int notcurrent)
> return sal;
> }
>
> +/* Compare two symtab_and_line entries. Return true if both have
> + the same line number and the same symtab pointer. That means we
> + are dealing with two entries from the same line and from the same
> + source file.
> +
> + Return false otherwise. */
> +
> +static bool
> +sal_line_symtab_matches_p (const symtab_and_line &sal1,
> + const symtab_and_line &sal2)
> +{
> + return (sal1.line == sal2.line && sal1.symtab == sal2.symtab);
> +}
> +
> +/* See symtah.h. */
> +
> +gdb::optional<CORE_ADDR>
> +find_line_range_start (CORE_ADDR pc)
> +{
> + struct symtab_and_line current_sal = find_pc_line (pc, 0);
> +
> + if (current_sal.line == 0)
> + return {};
> +
> + struct symtab_and_line prev_sal = find_pc_line (current_sal.pc - 1, 0);
> +
> + /* If the previous entry is for a different line, that means we are already
> + at the entry with the start PC for this line. */
> + if (!sal_line_symtab_matches_p (prev_sal, current_sal))
> + return current_sal.pc;
> +
> + /* Otherwise, keep looking for entries for the same line but with
> + smaller PC's. */
> + bool done = false;
> + CORE_ADDR prev_pc;
> + while (!done)
> + {
> + prev_pc = prev_sal.pc;
> +
> + prev_sal = find_pc_line (prev_pc - 1, 0);
> +
> + /* Did we notice a line change? If so, we are done with the search. */
> + if (!sal_line_symtab_matches_p (prev_sal, current_sal))
> + done = true;
> + }
> +
> + return prev_pc;
> +}
> +
> /* See symtab.h. */
>
> struct symtab *
> diff --git a/gdb/symtab.h b/gdb/symtab.h
> index 404d0ab30a8..f54305636da 100644
> --- a/gdb/symtab.h
> +++ b/gdb/symtab.h
> @@ -2346,6 +2346,22 @@ extern struct symtab_and_line find_pc_line (CORE_ADDR, int);
> extern struct symtab_and_line find_pc_sect_line (CORE_ADDR,
> struct obj_section *, int);
>
> +/* Given PC, and assuming it is part of a range of addresses that is part of a
> + line, go back through the linetable and find the starting PC of that
> + line.
> +
> + For example, suppose we have 3 PC ranges for line X:
> +
> + Line X - [0x0 - 0x8]
> + Line X - [0x8 - 0x10]
> + Line X - [0x10 - 0x18]
> +
> + If we call the function with PC == 0x14, we want to return 0x0, as that is
> + the starting PC of line X, and the ranges are contiguous.
> +*/
> +
> +extern gdb::optional<CORE_ADDR> find_line_range_start (CORE_ADDR pc);
> +
> /* Wrapper around find_pc_line to just return the symtab. */
>
> extern struct symtab *find_pc_line_symtab (CORE_ADDR);
> diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.c b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> new file mode 100644
> index 00000000000..412ab180943
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> @@ -0,0 +1,36 @@
> +/* Copyright 2008-2023 Free Software Foundation, Inc.
> +
> + This program is free software; you can redistribute it and/or modify
> + it under the terms of the GNU General Public License as published by
> + the Free Software Foundation; either version 3 of the License, or
> + (at your option) any later version.
> +
> + This program is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> + GNU General Public License for more details.
> +
> + You should have received a copy of the GNU General Public License
> + along with this program. If not, see <http://www.gnu.org/licenses/>.
> +
> + This test is used to test the reverse-step and reverse-next instruction
> + execution for a source line that contains multiple function calls. */
> +
> +void
> +func1 ()
> +{
> +} // END FUNC1
> +
> +void
> +func2 ()
> +{
> +} // END FUNC2
> +
> +int main ()
> +{
> + int a, b;
> + a = 1;
> + b = 2;
> + func1 (); func2 ();
> + a = a + b; // START REVERSE TEST
> +}
> diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> new file mode 100644
> index 00000000000..4eae042a6bf
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> @@ -0,0 +1,146 @@
> +# Copyright 2008-2023 Free Software Foundation, Inc.
> +
> +# This program is free software; you can redistribute it and/or modify
> +# it under the terms of the GNU General Public License as published by
> +# the Free Software Foundation; either version 3 of the License, or
> +# (at your option) any later version.
> +#
> +# This program is distributed in the hope that it will be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> +# GNU General Public License for more details.
> +#
> +# You should have received a copy of the GNU General Public License
> +# along with this program. If not, see <http://www.gnu.org/licenses/>. */
> +
> +# This file is part of the GDB testsuite. It tests reverse stepping.
> +# Lots of code borrowed from "step-test.exp".
> +
> +# This test checks to make sure there is no regression failures for
> +# the reverse-next command when stepping back over two functions in
> +# the same line.
> +
> +require supports_reverse
> +
> +# This test uses the gcc no-column-info command which was added in gcc 7.1.
> +require get_compiler_info "gcc-7-*"
> +
> +proc run_tests {msg} {
> + global srcfile
> + global executable
> +
> + runto_main
> + set target_remote [gdb_is_target_remote]
> +
> + gdb_test_no_output "record" "turn on process record"
> +
> + # This regression test verifies the reverse-step and reverse-next commands
> + # work properly when executing backwards thru a source line containing
> + # two function calls on the same source line, i.e. func1 (); func2 ();
> + # This test is compiled so the dwarf info not contain the line table
> + # information.
> +
> + # Test 1, reverse-next command
> + # Set breakpoint at the line after the function calls.
> + set bp_start_reverse_test [gdb_get_line_number "START REVERSE TEST" \
> + $srcfile]
> + gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> +
> + # Continue to break point for reverse-next test.
> + # Command definition: reverse-next [count]
> + # Run backward to the beginning of the previous line executed in the
> + # current (innermost) stack frame. If the line contains function calls,
> + # they will be “un-executed” without stopping. Starting from the first
> + # line of a function, reverse-next will take you back to the caller of
> + # that function, before the function was called, just as the normal next
> + # command would take you from the last line of a function back to its
> + # return to its caller 2 .
> + gdb_continue_to_breakpoint \
> + "$msg: test1: stopped at command reverse-next test start location" \
> + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> +
> + # The reverse-next should step all the way back to the beginning of the
> + # line, i.e. at the beginning of the func1 call.
> + gdb_test "reverse-next" ".*func1 \\(\\); func2 \\(\\);.*" \
> + "$msg: test1: reverse-next to line with two functions"
> +
> + # We should be stopped at the first instruction of the line. A reverse-step
> + # should step back and stop at the beginning of the previous line b = 2,
> + # i.e. not in func1 ().
> + gdb_test "reverse-stepi" ".*b = 2;.*" \
> + "$msg: test1: reverse-stepi to previous line b = 2"
> +
> +
> + # Setup for test 2
> + clean_restart $executable
> + runto_main
> +
> + gdb_test_no_output "record" "turn on process record"
> +
> + # Test 2, reverse-step command
> + # Set breakpoint at the line after the function calls.
> + gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> +
> + # Continue to the start of the reverse-step test.
> + # Command definition: reverse-step [count]
> + # Run the program backward until control reaches the start of a
> + # different source line; then stop it, and return control to gdb.
> + # Like the step command, reverse-step will only stop at the beginning
> + # of a source line. It “un-executes” the previously executed source
> + # line. If the previous source line included calls to debuggable
> + # functions, reverse-step will step (backward) into the called function,
> + # stopping at the beginning of the last statement in the called
> + # function (typically a return statement). Also, as with the step
> + # command, if non-debuggable functions are called, reverse-step will
> + # run thru them backward without stopping.
> +
> + gdb_continue_to_breakpoint \
> + "$msg: test2: stopped at command reverse-step test start location" \
> + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> +
> + # The first reverse step should take us call of func2 ().
> + gdb_test "reverse-step" ".*END FUNC2.*" \
> + "$msg: test2: reverse-step into func2 "
> +
> + # The second reverse step should take us into func1 ().
> + gdb_test "reverse-step" ".*END FUNC1.*" \
> + "$msg: test2: reverse-step into func1 "
> +
> + # The third reverse step should take us call of func1 ().
> + gdb_test "reverse-step" ".*func1 \\(\\); func2 \\(\\);.*" \
> + "$msg: test2: reverse-step to line func1(); func2(), at call for func1 "
> +
> + # We should be stopped at the first instruction of the line. A reverse
> + # stepi should take us to b = 2 ().
> + gdb_test "reverse-stepi" ".*b = 2;.*" \
> + "$msg: test2: reverse-stepi to line b = 2 "
> +}
> +
> +set srcfile func-map-to-same-line.c
> +set executable func-map-to-same-line
> +
> +# test with gcc column info enabled
> +set options [list debug additional_flags=]
> +
> +if {[build_executable "failed to prepare" $executable $srcfile $options] == -1}\
> + {
> + return -1
> +}
> +
> +clean_restart $executable
> +
> +run_tests {"with-column-info"}
> +
> +
> +#test with gcc column info disabled
> +set options [list debug additional_flags=-gno-column-info]
> +
> +if {[build_executable "failed to prepare" $executable $srcfile $options] == -1}\
> + {
> + return -1
> +}
> +
> +set $executable executable_without_column_info
> +clean_restart $executable
> +
> +run_tests {"no-column-info"}
> diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.c b/gdb/testsuite/gdb.reverse/map-to-same-line.c
> new file mode 100644
> index 00000000000..f20d778f40e
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/map-to-same-line.c
> @@ -0,0 +1,58 @@
> +/* Copyright 2008-2023 Free Software Foundation, Inc.
> +
> + This program is free software; you can redistribute it and/or modify
> + it under the terms of the GNU General Public License as published by
> + the Free Software Foundation; either version 3 of the License, or
> + (at your option) any later version.
> +
> + This program is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> + GNU General Public License for more details.
> +
> + You should have received a copy of the GNU General Public License
> + along with this program. If not, see <http://www.gnu.org/licenses/ >. */
> +
> +/* The purpose of this test is to create a DWARF line table that contains two
> + or more entries for the same line. When stepping (forwards or backwards),
> + GDB should step over the entire line and not just a particular entry in the
> + line table. */
> +
> +int
> +main ()
> +{ /* TAG: main prologue */
> + asm ("main_label: .globl main_label");
> + int i = 1, j = 2, k;
> + float f1 = 2.0, f2 = 4.1, f3;
> + const char *str_1 = "foo", *str_2 = "bar", *str_3;
> +
> + asm ("line1: .globl line1");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 1 */
> +
> + asm ("line2: .globl line2");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 2 */
> +
> + asm ("line3: .globl line3");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 3 */
> +
> + asm ("line4: .globl line4");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 4 */
> +
> + asm ("line5: .globl line5");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 5 */
> +
> + asm ("line6: .globl line6");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 6 */
> +
> + asm ("line7: .globl line7");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 7 */
> +
> + asm ("line8: .globl line8");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 8 */
> +
> + asm ("main_return: .globl main_return");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: main return */
> +
> + asm ("end_of_sequence: .globl end_of_sequence");
> + return 0; /* TAG: main return */
> +}
> diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.exp b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> new file mode 100644
> index 00000000000..02f3f4d8c9c
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> @@ -0,0 +1,156 @@
> +# Copyright 2008-2023 Free Software Foundation, Inc.
> +
> +# This program is free software; you can redistribute it and/or modify
> +# it under the terms of the GNU General Public License as published by
> +# the Free Software Foundation; either version 3 of the License, or
> +# (at your option) any later version.
> +#
> +# This program is distributed in the hope that it will be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> +# GNU General Public License for more details.
> +#
> +# You should have received a copy of the GNU General Public License
> +# along with this program. If not, see <http://www.gnu.org/licenses/ >.
> +
> +# When stepping (forwards or backwards), GDB should step over the entire line
> +# and not just a particular entry in the line table. This test was added to
> +# verify the find_line_range_start function properly sets the step range for a
> +# line that consists of multiple statements, i.e. multiple entries in the line
> +# table. This test creates a DWARF line table that contains two entries for
> +# the same line to do the needed testing.
> +
> +# This test can only be run on targets which support DWARF-2 and use gas.
> +require dwarf2_support
> +load_lib dwarf.exp
> +
> +# The DWARF assembler requires the gcc compiler.
> +require is_c_compiler_gcc
> +
> +# This test suitable only for process that can do reverse execution
> +requires supports_reverse
> +
> +standard_testfile .c .S
> +
> +if { [prepare_for_testing "failed to prepare" ${testfile} ${srcfile}] } {
> + return -1
> +}
> +
> +set asm_file [standard_output_file $srcfile2]
> +Dwarf::assemble $asm_file {
> + global srcdir subdir srcfile
> + declare_labels integer_label L
> +
> + # Find start address and length of program
> + lassign [function_range main [list ${srcdir}/${subdir}/$srcfile]] \
> + main_start main_len
> + set main_end "$main_start + $main_len"
> +
> + cu {} {
> + compile_unit {
> + {language @DW_LANG_C}
> + {name map-to-same-line.c}
> + {stmt_list $L DW_FORM_sec_offset}
> + {low_pc 0 addr}
> + } {
> + subprogram {
> + {external 1 flag}
> + {name main}
> + {low_pc $main_start addr}
> + {high_pc $main_len DW_FORM_data4}
> + }
> + }
> + }
> +
> + lines {version 2 default_is_stmt 1} L {
> + include_dir "${srcdir}/${subdir}"
> + file_name "$srcfile" 1
> +
> + # Generate the line table program with distinct source lines being
> + # mapped to the same line entry. Line 1, 5 and 8 contain 1 statement
> + # each. Line 2 contains 2 statements. Line 3 contains 3 statements.
> + program {
> + DW_LNE_set_address $main_start
> + line [gdb_get_line_number "TAG: main prologue"]
> + DW_LNS_copy
> + DW_LNE_set_address line1
> + line [gdb_get_line_number "TAG: line 1" ]
> + DW_LNS_copy
> + DW_LNE_set_address line2
> + line [gdb_get_line_number "TAG: line 2" ]
> + DW_LNS_copy
> + DW_LNE_set_address line3
> + line [gdb_get_line_number "TAG: line 2" ]
> + DW_LNS_copy
> + DW_LNE_set_address line4
> + line [gdb_get_line_number "TAG: line 3" ]
> + DW_LNS_copy
> + DW_LNE_set_address line5
> + line [gdb_get_line_number "TAG: line 3" ]
> + DW_LNS_copy
> + DW_LNE_set_address line6
> + line [gdb_get_line_number "TAG: line 3" ]
> + DW_LNS_copy
> + DW_LNE_set_address line7
> + line [gdb_get_line_number "TAG: line 5" ]
> + DW_LNS_copy
> + DW_LNE_set_address line8
> + line [gdb_get_line_number "TAG: line 8" ]
> + DW_LNS_copy
> + DW_LNE_set_address main_return
> + line [gdb_get_line_number "TAG: main return"]
> + DW_LNS_copy
> + DW_LNE_set_address end_of_sequence
> + DW_LNE_end_sequence
> + }
> + }
> +}
> +
> +if { [prepare_for_testing "failed to prepare" ${testfile} \
> + [list $srcfile $asm_file] {nodebug} ] } {
> + return -1
> +}
> +
> +runto_main
> +
> +# Print the line table
> +gdb_test_multiple "maint info line-table ${testfile}" "" {
> + -re "\r\n$decimal\[ \t\]+$decimal\[ \t\]+($hex)\[ \t\]+Y\[^\r\n\]*" {
> + lappend is_stmt $expect_out(1,string)
> + exp_continue
> + }
> + -re -wrap "" {
> + }
> +}
> +
> +# Do the reverse-step test
> +gdb_test_no_output "record" "turn on process record"
> +
> +set bp_main_return [gdb_get_line_number "TAG: main return" $srcfile]
> +gdb_breakpoint $srcfile:$bp_main_return
> +gdb_continue_to_breakpoint "run to end of main, reverse-step test" ".*$srcfile:$bp_main_return.*"
> +gdb_test "display \$pc" ".*pc =.*" "display pc, reverse-step test"
> +
> +# At this point, GDB has already recorded the execution up until the return
> +# statement. Reverse-step and test if GDB transitions between lines in the
> +# expected order. It should reverse-step across lines 8, 5, 3, 2 and 1.
> +foreach line {8 5 3 2 1} {
> + gdb_test "reverse-step" ".*TAG: line $line.*" "reverse step to line $line"
> +}
> +
> +## Clean restart, test reverse-next command
> +clean_restart ${testfile}
> +runto_main
> +gdb_test_no_output "record" "turn on process record, reverst-next test"
> +
> +set bp_main_return [gdb_get_line_number "TAG: main return" $srcfile]
> +gdb_breakpoint $srcfile:$bp_main_return
> +gdb_continue_to_breakpoint "run to end of main, reverse-next test" ".*$srcfile:$bp_main_return.*"
> +gdb_test "display \$pc" ".*pc =.*" "display pc, reverse-next test"
> +
> +# At this point, GDB has already recorded the execution up until the return
> +# statement. Reverse-next and test if GDB transitions between lines in the
> +# expected order. It should reverse-next across lines 8, 5, 3, 2 and 1.
> +foreach line {8 5 3 2 1} {
> + gdb_test "reverse-next" ".*TAG: line $line.*" "reverse next to line $line"
> +}
Other than the nit, this LGTM. Thanks for picking this one up and improving it to fix other issues.
Reviewed-by: Luis Machado <luis.machado@arm.com>
^ permalink raw reply [flat|nested] 40+ messages in thread
* RE: [PATCH v3] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-05-05 14:59 ` Luis Machado
@ 2023-05-05 16:10 ` Carl Love
0 siblings, 0 replies; 40+ messages in thread
From: Carl Love @ 2023-05-05 16:10 UTC (permalink / raw)
To: Luis Machado, Bruno Larsen, gdb-patches, UlrichWeigand, pedro; +Cc: cel
Luis:
On Fri, 2023-05-05 at 15:59 +0100, Luis Machado wrote:
> > + /* Call find_line_range_start to get smallest address in the
>
> s/smallest/the smallest
>
> > + linetable for multiple Line X entries in the line table. */
> > + real_range_start = find_line_range_start (pc);
> > +
> > + if (real_range_start.has_value ())
<snip>
> Other than the nit, this LGTM. Thanks for picking this one up and
> improving it to fix other issues.
>
> Reviewed-by: Luis Machado <luis.machado@arm.com>
Thanks for the review. I updated the patch with the fix. Probably not
worth resending patch but will make sure it does get included in any
future posts or commits.
Carl
^ permalink raw reply [flat|nested] 40+ messages in thread
* Re: [PATCH v3] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-05-04 15:59 ` [PATCH v3] " Carl Love
2023-05-05 14:59 ` Luis Machado
@ 2023-05-10 13:47 ` Bruno Larsen
2023-05-10 17:16 ` Carl Love
1 sibling, 1 reply; 40+ messages in thread
From: Bruno Larsen @ 2023-05-10 13:47 UTC (permalink / raw)
To: Carl Love, gdb-patches, UlrichWeigand, pedro; +Cc: luis.machado
On 04/05/2023 17:59, Carl Love wrote:
> Bruno, GDB maintainers:
>
> Version 3, added the gcc version check as discussed further from
> version 2 of the patch. Also updated the tests to check for supporting
> reverse execution rather than requiring recording. I also noticed
> there were a couple more instances of a requirement check, i.e. if []
> which I changed to "require" per the current style for checking on the
> test requirements.
>
>
> The following patch fixes issues on PowerPC with the reverse-step and
> reverse-next instructions when there are multiple assignment statements
> on the same line and when there are multiple function calls on the same
> line. The commit log below discusses these issues in further depth.
> The discussion included what the correct operation should be for these
> commands based on the GDB documentation. The proposed patch at that
> time changed how the commands worked on other platforms such as X86 in
> a way they no longer matched the documentation.
>
> The issue is the line table contains multiple entries for the same
> source line. The patch adds a function to search the line table to
> find the address of the first instruction of a line. When setup up the
> reverse stepping range, the function is called to make sure the start
> of the range corresponds to the address of the first instruction for
> the line. This approach was used. When Luis initially developed the
> patch, he considered merging the contiguous ranges in the line table
> when reading the line tables. He decided it was better to work with the
> data directly in the line table rather than creating and using a
> modified version of the line table.
>
> The following patch fixes the execution of the reveres-step and
> reverse-next commands for both senarios of multiple statements on the
> same line for PowerPC and aarch64-linux. Unlike the previous patch, it
> does not change the operation of the commands on other platforms, i.e.
> X86. The patch adds new test cases for both scenarios to verify they
> work correctly.
>
> The patch has been tested on PowerPC, Intel X86 and aarch64-linux with
> no new regression failures.
>
> Please let me know if the patch is acceptable for mainline. Thanks.
>
> Carl
>
> ---------------------------------------------------------------
> Fix reverse stepping multiple contiguous PC ranges over the line table.
>
> There are a couple of scenarios where the GDB reverse-step and reverse-next
> commands do not work correctly.
>
> Scenario 1 issue description by Luis Machado:
>
> When running GDB's testsuite on aarch64-linux/Ubuntu 20.04 (also spotted on
> the ppc backend), I noticed some failures in gdb.reverse/solib-precsave.exp
> and gdb.reverse/solib-reverse.exp.
>
> The failure happens around the following code:
>
> 38 b[1] = shr2(17); /* middle part two */
> 40 b[0] = 6; b[1] = 9; /* generic statement, end part two */
> 42 shr1 ("message 1\n"); /* shr1 one */
>
> Normal execution:
>
> - step from line 38 will land on line 40.
> - step from line 40 will land on line 42.
>
> Reverse execution:
> - step from line 42 will land on line 40.
> - step from line 40 will land on line 40.
> - step from line 40 will land on line 38.
>
> The problem here is that line 40 contains two contiguous but distinct
> PC ranges in the line table, like so:
>
> Line 40 - [0x7ec ~ 0x7f4]
> Line 40 - [0x7f4 ~ 0x7fc]
>
> The two distinct ranges are generated because GCC started outputting source
> column information, which GDB doesn't take into account at the moment.
>
> When stepping forward from line 40, we skip both of these ranges and land on
> line 42. When stepping backward from line 42, we stop at the start PC of the
> second (or first, going backwards) range of line 40.
>
> Since we've reached ecs->event_thread->control.step_range_start, we stop
> stepping backwards.
>
> ---------------------------------------------------------
>
> Scenario 2 issue described by Pedro Alves:
>
> The following explanation of the issue was taken from the gdb mailing list
> discussion of the withdrawn patch to change the behavior of the reverse-step
> and reverse-next commands. Specifically, message from Pedro Alves
> <pedro@palves.net> where he demonstrates the issue where you have multiple
> function calls on the same source code line:
>
> https://sourceware.org/pipermail/gdb-patches/2023-January/196110.html
>
> The source line looks like:
>
> func1 (); func2 ();
>
> so stepping backwards over that line should always stop at the first
> instruction of the line, not in the middle. Let's simplify this.
>
> Here's the full source code of my example:
>
> (gdb) list 1
> 1 void func1 ()
> 2 {
> 3 }
> 4
> 5 void func2 ()
> 6 {
> 7 }
> 8
> 9 int main ()
> 10 {
> 11 func1 (); func2 ();
> 12 }
>
> Compiled with:
>
> $ gcc reverse.c -o reverse -g3 -O0
> $ gcc -v
> ...
> gcc version 11.3.0 (Ubuntu 11.3.0-1ubuntu1~22.04)
>
> Now let's debug it with target record, using current gdb git master (f3d8ae90b236),
> without your patch:
>
> $ gdb ~/reverse
> GNU gdb (GDB) 14.0.50.20230124-git
> ...
> Reading symbols from /home/pedro/reverse...
> (gdb) start
> Temporary breakpoint 1 at 0x1147: file reverse.c, line 11.
> Starting program: /home/pedro/reverse
> [Thread debugging using libthread_db enabled]
> Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
>
> Temporary breakpoint 1, main () at reverse.c:11
> 11 func1 (); func2 ();
> (gdb) record
>
> (gdb) disassemble /s
> Dump of assembler code for function main:
> reverse.c:
> 10 {
> 0x000055555555513f <+0>: endbr64
> 0x0000555555555143 <+4>: push %rbp
> 0x0000555555555144 <+5>: mov %rsp,%rbp
>
> 11 func1 (); func2 ();
> => 0x0000555555555147 <+8>: mov $0x0,%eax
> 0x000055555555514c <+13>: call 0x555555555129 <func1>
> 0x0000555555555151 <+18>: mov $0x0,%eax
> 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> 0x000055555555515b <+28>: mov $0x0,%eax
>
> 12 }
> 0x0000555555555160 <+33>: pop %rbp
> 0x0000555555555161 <+34>: ret
> End of assembler dump.
>
> (gdb) n
> 12 }
>
> So far so good, a "next" stepped over the whole of line 11 and stopped at line 12.
>
> Let's confirm where we are now:
>
> (gdb) disassemble /s
> Dump of assembler code for function main:
> reverse.c:
> 10 {
> 0x000055555555513f <+0>: endbr64
> 0x0000555555555143 <+4>: push %rbp
> 0x0000555555555144 <+5>: mov %rsp,%rbp
>
> 11 func1 (); func2 ();
> 0x0000555555555147 <+8>: mov $0x0,%eax
> 0x000055555555514c <+13>: call 0x555555555129 <func1>
> 0x0000555555555151 <+18>: mov $0x0,%eax
> 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> 0x000055555555515b <+28>: mov $0x0,%eax
>
> 12 }
> => 0x0000555555555160 <+33>: pop %rbp
> 0x0000555555555161 <+34>: ret
> End of assembler dump.
>
> Good, we're at the first instruction of line 12.
>
> Now let's undo the "next", with "reverse-next":
>
> (gdb) reverse-next
> 11 func1 (); func2 ();
>
> Seemingly stopped at line 11. Let's see exactly where:
>
> (gdb) disassemble /s
> Dump of assembler code for function main:
> reverse.c:
> 10 {
> 0x000055555555513f <+0>: endbr64
> 0x0000555555555143 <+4>: push %rbp
> 0x0000555555555144 <+5>: mov %rsp,%rbp
>
> 11 func1 (); func2 ();
> 0x0000555555555147 <+8>: mov $0x0,%eax
> 0x000055555555514c <+13>: call 0x555555555129 <func1>
> => 0x0000555555555151 <+18>: mov $0x0,%eax
> 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> 0x000055555555515b <+28>: mov $0x0,%eax
>
> 12 }
> 0x0000555555555160 <+33>: pop %rbp
> 0x0000555555555161 <+34>: ret
> End of assembler dump.
> (gdb)
>
> And lo, we stopped in the middle of line 11! That is a bug, we should have
> stepped back all the way to the beginning of the line. The "reverse-next"
> should have fully undone the prior "next" command.
>
> The above issues were fixed by introducing a new function that looks for
> adjacent PC ranges for the same line, until we notice a line change. Then
> we take that as the start PC of the range. The new start PC for the range
> is used for the control.step_range_start when setting up a step range.
>
> The test case gdb.reverse/map-to-same-line.exp is added to test the fix
> for the issues in scenario 1.
>
> The test case gdb.reverse/func-map-to-same-line.exp was added to test the
> fix for scenario 2 when the binary was compiled with and without line
> table information.
>
> bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28426
>
> Co-authored-by: Luis Machado <luis.machado@arm.com>
> Co-authored-by: Carl Love <cel@us.ibm.com>
> ---
> gdb/infrun.c | 57 +++++++
> gdb/symtab.c | 49 ++++++
> gdb/symtab.h | 16 ++
> .../gdb.reverse/func-map-to-same-line.c | 36 ++++
> .../gdb.reverse/func-map-to-same-line.exp | 146 ++++++++++++++++
> gdb/testsuite/gdb.reverse/map-to-same-line.c | 58 +++++++
> .../gdb.reverse/map-to-same-line.exp | 156 ++++++++++++++++++
> 7 files changed, 518 insertions(+)
> create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.c
> create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.exp
>
> diff --git a/gdb/infrun.c b/gdb/infrun.c
> index efe2c00c489..8555e3c979f 100644
> --- a/gdb/infrun.c
> +++ b/gdb/infrun.c
> @@ -114,6 +114,9 @@ static struct async_event_handler *infrun_async_inferior_event_token;
> Starts off as -1, indicating "never enabled/disabled". */
> static int infrun_is_async = -1;
>
> +static CORE_ADDR update_line_range_start (CORE_ADDR pc,
> + struct execution_control_state *ecs);
> +
> /* See infrun.h. */
>
> void
> @@ -6769,6 +6772,25 @@ handle_signal_stop (struct execution_control_state *ecs)
> process_event_stop_test (ecs);
> }
>
> +CORE_ADDR
> +update_line_range_start (CORE_ADDR pc, struct execution_control_state *ecs)
> +{
> + /* The line table may have multiple entries for the same source code line.
> + Given the PC, check the line table and return the PC that corresponds
> + to the line table entry for the source line that PC is in. */
> + CORE_ADDR start_line_pc = ecs->event_thread->control.step_range_start;
> + gdb::optional<CORE_ADDR> real_range_start;
> +
> + /* Call find_line_range_start to get smallest address in the
> + linetable for multiple Line X entries in the line table. */
> + real_range_start = find_line_range_start (pc);
> +
> + if (real_range_start.has_value ())
> + start_line_pc = *real_range_start;
> +
> + return start_line_pc;
> +}
> +
> /* Come here when we've got some debug event / signal we can explain
> (IOW, not a random signal), and test whether it should cause a
> stop, or whether we should resume the inferior (transparently).
> @@ -7570,6 +7592,28 @@ process_event_stop_test (struct execution_control_state *ecs)
>
> if (stop_pc_sal.is_stmt)
> {
> + if (execution_direction == EXEC_REVERSE)
> + {
> + /* We are stepping backwards make sure we have reached the
> + beginning of the line. */
> + CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
> + CORE_ADDR start_line_pc
> + = update_line_range_start (stop_pc, ecs);
> +
> + if (stop_pc != start_line_pc)
> + {
> + /* Have not reached the beginning of the source code line.
> + Set a step range. Execution should stop in any function
> + calls we execute back into before reaching the beginning
> + of the line. */
> + ecs->event_thread->control.step_range_start = start_line_pc;
> + ecs->event_thread->control.step_range_end = stop_pc;
> + set_step_info (ecs->event_thread, frame, stop_pc_sal);
> + keep_going (ecs);
> + return;
> + }
> + }
> +
> /* We are at the start of a statement.
>
> So stop. Note that we don't stop if we step into the middle of a
> @@ -7632,6 +7676,19 @@ process_event_stop_test (struct execution_control_state *ecs)
> set_step_info (ecs->event_thread, frame, stop_pc_sal);
>
> infrun_debug_printf ("keep going");
> +
> + if (execution_direction == EXEC_REVERSE)
> + {
> + CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
> +
> + /* Make sure the stop_pc is set to the beginning of the line. */
> + if (stop_pc != ecs->event_thread->control.step_range_start)
> + {
> + stop_pc = update_line_range_start (stop_pc, ecs);
> + ecs->event_thread->control.step_range_start = stop_pc;
> + }
> + }
> +
> keep_going (ecs);
> }
>
> diff --git a/gdb/symtab.c b/gdb/symtab.c
> index 27611a34ec4..91d35616eb9 100644
> --- a/gdb/symtab.c
> +++ b/gdb/symtab.c
> @@ -3282,6 +3282,55 @@ find_pc_line (CORE_ADDR pc, int notcurrent)
> return sal;
> }
>
> +/* Compare two symtab_and_line entries. Return true if both have
> + the same line number and the same symtab pointer. That means we
> + are dealing with two entries from the same line and from the same
> + source file.
> +
> + Return false otherwise. */
> +
> +static bool
> +sal_line_symtab_matches_p (const symtab_and_line &sal1,
> + const symtab_and_line &sal2)
> +{
> + return (sal1.line == sal2.line && sal1.symtab == sal2.symtab);
> +}
> +
> +/* See symtah.h. */
> +
> +gdb::optional<CORE_ADDR>
> +find_line_range_start (CORE_ADDR pc)
> +{
> + struct symtab_and_line current_sal = find_pc_line (pc, 0);
> +
> + if (current_sal.line == 0)
> + return {};
> +
> + struct symtab_and_line prev_sal = find_pc_line (current_sal.pc - 1, 0);
> +
> + /* If the previous entry is for a different line, that means we are already
> + at the entry with the start PC for this line. */
> + if (!sal_line_symtab_matches_p (prev_sal, current_sal))
> + return current_sal.pc;
> +
> + /* Otherwise, keep looking for entries for the same line but with
> + smaller PC's. */
> + bool done = false;
> + CORE_ADDR prev_pc;
> + while (!done)
> + {
> + prev_pc = prev_sal.pc;
> +
> + prev_sal = find_pc_line (prev_pc - 1, 0);
> +
> + /* Did we notice a line change? If so, we are done with the search. */
> + if (!sal_line_symtab_matches_p (prev_sal, current_sal))
> + done = true;
> + }
> +
> + return prev_pc;
> +}
> +
> /* See symtab.h. */
>
> struct symtab *
> diff --git a/gdb/symtab.h b/gdb/symtab.h
> index 404d0ab30a8..f54305636da 100644
> --- a/gdb/symtab.h
> +++ b/gdb/symtab.h
> @@ -2346,6 +2346,22 @@ extern struct symtab_and_line find_pc_line (CORE_ADDR, int);
> extern struct symtab_and_line find_pc_sect_line (CORE_ADDR,
> struct obj_section *, int);
>
> +/* Given PC, and assuming it is part of a range of addresses that is part of a
> + line, go back through the linetable and find the starting PC of that
> + line.
> +
> + For example, suppose we have 3 PC ranges for line X:
> +
> + Line X - [0x0 - 0x8]
> + Line X - [0x8 - 0x10]
> + Line X - [0x10 - 0x18]
> +
> + If we call the function with PC == 0x14, we want to return 0x0, as that is
> + the starting PC of line X, and the ranges are contiguous.
> +*/
> +
> +extern gdb::optional<CORE_ADDR> find_line_range_start (CORE_ADDR pc);
> +
> /* Wrapper around find_pc_line to just return the symtab. */
>
> extern struct symtab *find_pc_line_symtab (CORE_ADDR);
> diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.c b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> new file mode 100644
> index 00000000000..412ab180943
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> @@ -0,0 +1,36 @@
> +/* Copyright 2008-2023 Free Software Foundation, Inc.
> +
> + This program is free software; you can redistribute it and/or modify
> + it under the terms of the GNU General Public License as published by
> + the Free Software Foundation; either version 3 of the License, or
> + (at your option) any later version.
> +
> + This program is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> + GNU General Public License for more details.
> +
> + You should have received a copy of the GNU General Public License
> + along with this program. If not, see <http://www.gnu.org/licenses/>.
> +
> + This test is used to test the reverse-step and reverse-next instruction
> + execution for a source line that contains multiple function calls. */
> +
> +void
> +func1 ()
> +{
> +} // END FUNC1
> +
> +void
> +func2 ()
> +{
> +} // END FUNC2
> +
> +int main ()
> +{
> + int a, b;
> + a = 1;
> + b = 2;
> + func1 (); func2 ();
> + a = a + b; // START REVERSE TEST
> +}
> diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> new file mode 100644
> index 00000000000..4eae042a6bf
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> @@ -0,0 +1,146 @@
> +# Copyright 2008-2023 Free Software Foundation, Inc.
> +
> +# This program is free software; you can redistribute it and/or modify
> +# it under the terms of the GNU General Public License as published by
> +# the Free Software Foundation; either version 3 of the License, or
> +# (at your option) any later version.
> +#
> +# This program is distributed in the hope that it will be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> +# GNU General Public License for more details.
> +#
> +# You should have received a copy of the GNU General Public License
> +# along with this program. If not, see <http://www.gnu.org/licenses/>. */
> +
> +# This file is part of the GDB testsuite. It tests reverse stepping.
> +# Lots of code borrowed from "step-test.exp".
> +
> +# This test checks to make sure there is no regression failures for
> +# the reverse-next command when stepping back over two functions in
> +# the same line.
> +
> +require supports_reverse
> +
> +# This test uses the gcc no-column-info command which was added in gcc 7.1.
> +require get_compiler_info "gcc-7-*"
By constructing your regex like this, you are only allowing this test to
be run on gcc 7. Anything later is also not accepted. I would do
something like (Warning, untested)
require get_compiler_info "gcc"
require !get_compiler_info "gcc-[1-6]-*"
Which requires gcc, but does not allow versions 1 to 6. There is
probably a way to do it with a single require line, but I'm not the best
with regexes.
> +
> +proc run_tests {msg} {
> + global srcfile
> + global executable
> +
> + runto_main
> + set target_remote [gdb_is_target_remote]
When probing for target remote, GDB will emit pass/fails with hardcoded
names, so the current proc setup gives us some duplicated test names.
I would suggest that, instead of passing a message as a parameter, you
wrapped all function calls in a with_test_prefix scope, like:
with_test_prefix "with-column-info" {
run_test
}
> +
> + gdb_test_no_output "record" "turn on process record"
> +
> + # This regression test verifies the reverse-step and reverse-next commands
> + # work properly when executing backwards thru a source line containing
> + # two function calls on the same source line, i.e. func1 (); func2 ();
> + # This test is compiled so the dwarf info not contain the line table
> + # information.
> +
> + # Test 1, reverse-next command
> + # Set breakpoint at the line after the function calls.
> + set bp_start_reverse_test [gdb_get_line_number "START REVERSE TEST" \
> + $srcfile]
> + gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> +
> + # Continue to break point for reverse-next test.
> + # Command definition: reverse-next [count]
> + # Run backward to the beginning of the previous line executed in the
> + # current (innermost) stack frame. If the line contains function calls,
> + # they will be “un-executed” without stopping. Starting from the first
> + # line of a function, reverse-next will take you back to the caller of
> + # that function, before the function was called, just as the normal next
> + # command would take you from the last line of a function back to its
> + # return to its caller 2 .
> + gdb_continue_to_breakpoint \
> + "$msg: test1: stopped at command reverse-next test start location" \
> + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> +
> + # The reverse-next should step all the way back to the beginning of the
> + # line, i.e. at the beginning of the func1 call.
> + gdb_test "reverse-next" ".*func1 \\(\\); func2 \\(\\);.*" \
> + "$msg: test1: reverse-next to line with two functions"
> +
> + # We should be stopped at the first instruction of the line. A reverse-step
> + # should step back and stop at the beginning of the previous line b = 2,
> + # i.e. not in func1 ().
> + gdb_test "reverse-stepi" ".*b = 2;.*" \
> + "$msg: test1: reverse-stepi to previous line b = 2"
> +
> +
> + # Setup for test 2
> + clean_restart $executable
> + runto_main
> +
> + gdb_test_no_output "record" "turn on process record"
This gives a duplicate test name from setting up for the first test.
Adding "test 2:(...)" in here solves it.
> +
> + # Test 2, reverse-step command
> + # Set breakpoint at the line after the function calls.
> + gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> +
> + # Continue to the start of the reverse-step test.
> + # Command definition: reverse-step [count]
> + # Run the program backward until control reaches the start of a
> + # different source line; then stop it, and return control to gdb.
> + # Like the step command, reverse-step will only stop at the beginning
> + # of a source line. It “un-executes” the previously executed source
> + # line. If the previous source line included calls to debuggable
> + # functions, reverse-step will step (backward) into the called function,
> + # stopping at the beginning of the last statement in the called
> + # function (typically a return statement). Also, as with the step
> + # command, if non-debuggable functions are called, reverse-step will
> + # run thru them backward without stopping.
> +
> + gdb_continue_to_breakpoint \
> + "$msg: test2: stopped at command reverse-step test start location" \
> + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> +
> + # The first reverse step should take us call of func2 ().
> + gdb_test "reverse-step" ".*END FUNC2.*" \
> + "$msg: test2: reverse-step into func2 "
> +
> + # The second reverse step should take us into func1 ().
> + gdb_test "reverse-step" ".*END FUNC1.*" \
> + "$msg: test2: reverse-step into func1 "
> +
> + # The third reverse step should take us call of func1 ().
> + gdb_test "reverse-step" ".*func1 \\(\\); func2 \\(\\);.*" \
> + "$msg: test2: reverse-step to line func1(); func2(), at call for func1 "
> +
> + # We should be stopped at the first instruction of the line. A reverse
> + # stepi should take us to b = 2 ().
> + gdb_test "reverse-stepi" ".*b = 2;.*" \
> + "$msg: test2: reverse-stepi to line b = 2 "
> +}
> +
> +set srcfile func-map-to-same-line.c
> +set executable func-map-to-same-line
> +
> +# test with gcc column info enabled
> +set options [list debug additional_flags=]
> +
> +if {[build_executable "failed to prepare" $executable $srcfile $options] == -1}\
> + {
> + return -1
> +}
> +
> +clean_restart $executable
> +
> +run_tests {"with-column-info"}
> +
> +
> +#test with gcc column info disabled
> +set options [list debug additional_flags=-gno-column-info]
> +
> +if {[build_executable "failed to prepare" $executable $srcfile $options] == -1}\
> + {
> + return -1
> +}
> +
> +set $executable executable_without_column_info
> +clean_restart $executable
> +
> +run_tests {"no-column-info"}
> diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.c b/gdb/testsuite/gdb.reverse/map-to-same-line.c
> new file mode 100644
> index 00000000000..f20d778f40e
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/map-to-same-line.c
> @@ -0,0 +1,58 @@
> +/* Copyright 2008-2023 Free Software Foundation, Inc.
> +
> + This program is free software; you can redistribute it and/or modify
> + it under the terms of the GNU General Public License as published by
> + the Free Software Foundation; either version 3 of the License, or
> + (at your option) any later version.
> +
> + This program is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> + GNU General Public License for more details.
> +
> + You should have received a copy of the GNU General Public License
> + along with this program. If not, see <http://www.gnu.org/licenses/ >. */
> +
> +/* The purpose of this test is to create a DWARF line table that contains two
> + or more entries for the same line. When stepping (forwards or backwards),
> + GDB should step over the entire line and not just a particular entry in the
> + line table. */
> +
> +int
> +main ()
> +{ /* TAG: main prologue */
> + asm ("main_label: .globl main_label");
> + int i = 1, j = 2, k;
> + float f1 = 2.0, f2 = 4.1, f3;
> + const char *str_1 = "foo", *str_2 = "bar", *str_3;
> +
> + asm ("line1: .globl line1");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 1 */
> +
> + asm ("line2: .globl line2");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 2 */
> +
> + asm ("line3: .globl line3");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 3 */
> +
> + asm ("line4: .globl line4");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 4 */
> +
> + asm ("line5: .globl line5");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 5 */
> +
> + asm ("line6: .globl line6");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 6 */
> +
> + asm ("line7: .globl line7");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 7 */
> +
> + asm ("line8: .globl line8");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 8 */
> +
> + asm ("main_return: .globl main_return");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: main return */
> +
> + asm ("end_of_sequence: .globl end_of_sequence");
> + return 0; /* TAG: main return */
> +}
> diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.exp b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> new file mode 100644
> index 00000000000..02f3f4d8c9c
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> @@ -0,0 +1,156 @@
> +# Copyright 2008-2023 Free Software Foundation, Inc.
> +
> +# This program is free software; you can redistribute it and/or modify
> +# it under the terms of the GNU General Public License as published by
> +# the Free Software Foundation; either version 3 of the License, or
> +# (at your option) any later version.
> +#
> +# This program is distributed in the hope that it will be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> +# GNU General Public License for more details.
> +#
> +# You should have received a copy of the GNU General Public License
> +# along with this program. If not, see <http://www.gnu.org/licenses/ >.
> +
> +# When stepping (forwards or backwards), GDB should step over the entire line
> +# and not just a particular entry in the line table. This test was added to
> +# verify the find_line_range_start function properly sets the step range for a
> +# line that consists of multiple statements, i.e. multiple entries in the line
> +# table. This test creates a DWARF line table that contains two entries for
> +# the same line to do the needed testing.
> +
> +# This test can only be run on targets which support DWARF-2 and use gas.
> +require dwarf2_support
> +load_lib dwarf.exp
the library has to be imported before the "require"
> +
> +# The DWARF assembler requires the gcc compiler.
> +require is_c_compiler_gcc
> +
> +# This test suitable only for process that can do reverse execution
> +requires supports_reverse
s/requires/require
With these nits fixed, you can add my tag too!
Reviewed-By: Bruno Larsen <blarsen@redhat.com>
--
Cheers,
Bruno
> +
> +standard_testfile .c .S
> +
> +if { [prepare_for_testing "failed to prepare" ${testfile} ${srcfile}] } {
> + return -1
> +}
> +
> +set asm_file [standard_output_file $srcfile2]
> +Dwarf::assemble $asm_file {
> + global srcdir subdir srcfile
> + declare_labels integer_label L
> +
> + # Find start address and length of program
> + lassign [function_range main [list ${srcdir}/${subdir}/$srcfile]] \
> + main_start main_len
> + set main_end "$main_start + $main_len"
> +
> + cu {} {
> + compile_unit {
> + {language @DW_LANG_C}
> + {name map-to-same-line.c}
> + {stmt_list $L DW_FORM_sec_offset}
> + {low_pc 0 addr}
> + } {
> + subprogram {
> + {external 1 flag}
> + {name main}
> + {low_pc $main_start addr}
> + {high_pc $main_len DW_FORM_data4}
> + }
> + }
> + }
> +
> + lines {version 2 default_is_stmt 1} L {
> + include_dir "${srcdir}/${subdir}"
> + file_name "$srcfile" 1
> +
> + # Generate the line table program with distinct source lines being
> + # mapped to the same line entry. Line 1, 5 and 8 contain 1 statement
> + # each. Line 2 contains 2 statements. Line 3 contains 3 statements.
> + program {
> + DW_LNE_set_address $main_start
> + line [gdb_get_line_number "TAG: main prologue"]
> + DW_LNS_copy
> + DW_LNE_set_address line1
> + line [gdb_get_line_number "TAG: line 1" ]
> + DW_LNS_copy
> + DW_LNE_set_address line2
> + line [gdb_get_line_number "TAG: line 2" ]
> + DW_LNS_copy
> + DW_LNE_set_address line3
> + line [gdb_get_line_number "TAG: line 2" ]
> + DW_LNS_copy
> + DW_LNE_set_address line4
> + line [gdb_get_line_number "TAG: line 3" ]
> + DW_LNS_copy
> + DW_LNE_set_address line5
> + line [gdb_get_line_number "TAG: line 3" ]
> + DW_LNS_copy
> + DW_LNE_set_address line6
> + line [gdb_get_line_number "TAG: line 3" ]
> + DW_LNS_copy
> + DW_LNE_set_address line7
> + line [gdb_get_line_number "TAG: line 5" ]
> + DW_LNS_copy
> + DW_LNE_set_address line8
> + line [gdb_get_line_number "TAG: line 8" ]
> + DW_LNS_copy
> + DW_LNE_set_address main_return
> + line [gdb_get_line_number "TAG: main return"]
> + DW_LNS_copy
> + DW_LNE_set_address end_of_sequence
> + DW_LNE_end_sequence
> + }
> + }
> +}
> +
> +if { [prepare_for_testing "failed to prepare" ${testfile} \
> + [list $srcfile $asm_file] {nodebug} ] } {
> + return -1
> +}
> +
> +runto_main
> +
> +# Print the line table
> +gdb_test_multiple "maint info line-table ${testfile}" "" {
> + -re "\r\n$decimal\[ \t\]+$decimal\[ \t\]+($hex)\[ \t\]+Y\[^\r\n\]*" {
> + lappend is_stmt $expect_out(1,string)
> + exp_continue
> + }
> + -re -wrap "" {
> + }
> +}
> +
> +# Do the reverse-step test
> +gdb_test_no_output "record" "turn on process record"
> +
> +set bp_main_return [gdb_get_line_number "TAG: main return" $srcfile]
> +gdb_breakpoint $srcfile:$bp_main_return
> +gdb_continue_to_breakpoint "run to end of main, reverse-step test" ".*$srcfile:$bp_main_return.*"
> +gdb_test "display \$pc" ".*pc =.*" "display pc, reverse-step test"
> +
> +# At this point, GDB has already recorded the execution up until the return
> +# statement. Reverse-step and test if GDB transitions between lines in the
> +# expected order. It should reverse-step across lines 8, 5, 3, 2 and 1.
> +foreach line {8 5 3 2 1} {
> + gdb_test "reverse-step" ".*TAG: line $line.*" "reverse step to line $line"
> +}
> +
> +## Clean restart, test reverse-next command
> +clean_restart ${testfile}
> +runto_main
> +gdb_test_no_output "record" "turn on process record, reverst-next test"
> +
> +set bp_main_return [gdb_get_line_number "TAG: main return" $srcfile]
> +gdb_breakpoint $srcfile:$bp_main_return
> +gdb_continue_to_breakpoint "run to end of main, reverse-next test" ".*$srcfile:$bp_main_return.*"
> +gdb_test "display \$pc" ".*pc =.*" "display pc, reverse-next test"
> +
> +# At this point, GDB has already recorded the execution up until the return
> +# statement. Reverse-next and test if GDB transitions between lines in the
> +# expected order. It should reverse-next across lines 8, 5, 3, 2 and 1.
> +foreach line {8 5 3 2 1} {
> + gdb_test "reverse-next" ".*TAG: line $line.*" "reverse next to line $line"
> +}
^ permalink raw reply [flat|nested] 40+ messages in thread
* RE: [PATCH v3] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-05-10 13:47 ` Bruno Larsen
@ 2023-05-10 17:16 ` Carl Love
2023-05-10 17:32 ` [PATCH v4] " Carl Love
2023-05-11 7:52 ` [PATCH v3] " Bruno Larsen
0 siblings, 2 replies; 40+ messages in thread
From: Carl Love @ 2023-05-10 17:16 UTC (permalink / raw)
To: Bruno Larsen, gdb-patches, UlrichWeigand, pedro; +Cc: luis.machado, cel
Bruno:
Thanks for the review. I addressed your comments as mentioned below.
I will post version 4 with the changes.
FYI, I will be out of the office from May 11 thru May 15. So will
reply to any additional comments when I return.
On Wed, 2023-05-10 at 15:47 +0200, Bruno Larsen wrote:
> On 04/05/2023 17:59, Carl Love wrote:
> > Bruno, GDB maintainers:
> >
> > Version 3, added the gcc version check as discussed further from
> > version 2 of the patch. Also updated the tests to check for
> > supporting
> >
<snip>
> > extern struct symtab *find_pc_line_symtab (CORE_ADDR);
> > diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> > b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> > new file mode 100644
> > index 00000000000..412ab180943
> > --- /dev/null
> > +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> > @@ -0,0 +1,36 @@
> > +/* Copyright 2008-2023 Free Software Foundation, Inc.
> > +
> > + This program is free software; you can redistribute it and/or
> > modify
> > + it under the terms of the GNU General Public License as
> > published by
> > + the Free Software Foundation; either version 3 of the License,
> > or
> > + (at your option) any later version.
> > +
> > + This program is distributed in the hope that it will be useful,
> > + but WITHOUT ANY WARRANTY; without even the implied warranty of
> > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> > + GNU General Public L
<snip>
> > +
> > +require supports_reverse
> > +
> > +# This test uses the gcc no-column-info command which was added in
> > gcc 7.1.
> > +require get_compiler_info "gcc-7-*"
I put the compiler check in last. When I ran it, I obviously didn't
double check gdb/testsuite/gdb.log to make sure it really worked. I
normally try to make a point of double checking the log file. I have
been burned before thinking it was OK when there were no errors visible
on the command line. The above command fails if you check the log file.
>
> By constructing your regex like this, you are only allowing this test
> to
> be run on gcc 7. Anything later is also not accepted. I would do
> something like (Warning, untested)
>
> require get_compiler_info "gcc"
> require !get_compiler_info "gcc-[1-6]-*"
I couldn't get require to work like that. The get_compiler_info
doesn't seem take "gcc" as an argument.
I changed the test to:
if {![test_compiler_info {gcc-*}]
|| [test_compiler_info {gcc-[1-6]-*}]} {
return
}
With this, I do see the correct number of passes in
gdb/testsuite/gdb.log.
>
> Which requires gcc, but does not allow versions 1 to 6. There is
> probably a way to do it with a single require line, but I'm not the
> best
> with regexes.
>
> > +
> > +proc run_tests {msg} {
> > + global srcfile
> > + global executable
> > +
> > + runto_main
> > + set target_remote [gdb_is_target_remote]
>
> When probing for target remote, GDB will emit pass/fails with
> hardcoded
> names, so the current proc setup gives us some duplicated test names.
>
> I would suggest that, instead of passing a message as a parameter,
> you
> wrapped all function calls in a with_test_prefix scope, like:
>
> with_test_prefix "with-column-info" {
> run_test
> }
OK, I changed from passing in an argument and did the wrapped calls to
run_test instead. Note, this still didn't fix the duplicate test names
for turning on record.
>
> > +
> > + gdb_test_no_output "record" "turn on process record"
> > +
> > + # This regression test verifies the reverse-step and reverse-
> > next commands
> > + # work properly when executing backwards thru a source line
> > containing
> > + # two function calls on the same source line, i.e. func1 ();
> > func2 ();
> > + # This test is compiled so the dwarf info not contain the line
> > table
> > + # information.
> > +
> > + # Test 1, reverse-next command
> > + # Set breakpoint at the line after the function calls.
> > + set bp_start_reverse_test [gdb_get_line_number "START REVERSE
> > TEST" \
> > + $srcfile]
> > + gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> > +
> > + # Continue to break point for reverse-next test.
> > + # Command definition: reverse-next [count]
> > + # Run backward to the beginning of the previous line
> > executed in the
> > + # current (innermost) stack frame. If the line contains
> > function calls,
> > + # they will be “un-executed” without stopping. Starting from
> > the first
> > + # line of a function, reverse-next will take you back to the
> > caller of
> > + # that function, before the function was called, just as the
> > normal next
> > + # command would take you from the last line of a function
> > back to its
> > + # return to its caller 2 .
> > + gdb_continue_to_breakpoint \
> > + "$msg: test1: stopped at command reverse-next test start
> > location" \
> > + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> > +
> > + # The reverse-next should step all the way back to the
> > beginning of the
> > + # line, i.e. at the beginning of the func1 call.
> > + gdb_test "reverse-next" ".*func1 \\(\\); func2 \\(\\);.*" \
> > + "$msg: test1: reverse-next to line with two functions"
> > +
> > + # We should be stopped at the first instruction of the line. A
> > reverse-step
> > + # should step back and stop at the beginning of the previous
> > line b = 2,
> > + # i.e. not in func1 ().
> > + gdb_test "reverse-stepi" ".*b = 2;.*" \
> > + "$msg: test1: reverse-stepi to previous line b = 2"
> > +
> > +
> > + # Setup for test 2
> > + clean_restart $executable
> > + runto_main
> > +
> > + gdb_test_no_output "record" "turn on process record"
>
> This gives a duplicate test name from setting up for the first test.
> Adding "test 2:(...)" in here solves it.
I couldn't figure out how to get the above syntax to work. So I used
the with_test_prefix instead, i.e.
with_test_prefix "test2" {
gdb_test_no_output "record" "turn on process record"
}
That fixed the duplicate names. I also wrapped the first record with
"test1" for consistency.
>
> > +
> > + # Test 2, reverse-step command
> > + # Set breakpoint at the line after the function calls.
> > + gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> > +
> >
<snip>
> > +# the same line to do the needed testing.
> > +
> > +# This test can only be run on targets which support DWARF-2 and
> > use gas.
> > +require dwarf2_support
> > +load_lib dwarf.exp
> the library has to be imported before the "require"
OK, switched the order of the lines.
> > +
> > +# The DWARF assembler requires the gcc compiler.
> > +require is_c_compiler_gcc
> > +
> > +# This test suitable only for process that can do reverse
> > execution
> > +requires supports_reverse
>
> s/requires/require
Fixed.
Carl
^ permalink raw reply [flat|nested] 40+ messages in thread
* [PATCH v4] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-05-10 17:16 ` Carl Love
@ 2023-05-10 17:32 ` Carl Love
2023-05-11 16:01 ` Simon Marchi
2023-05-11 7:52 ` [PATCH v3] " Bruno Larsen
1 sibling, 1 reply; 40+ messages in thread
From: Carl Love @ 2023-05-10 17:32 UTC (permalink / raw)
To: Bruno Larsen, gdb-patches, UlrichWeigand, pedro; +Cc: luis.machado, cel, cel
Bruno, GDB maintainers:
Version 4, additional fixes for gcc version check, wrap function calls
using "with_test_prefix", move load_lib dwarf.exe. Fixed typo noted by
Luis.
Version 3, added the gcc version check as discussed further from
version 2 of the patch. Also updated the tests to check for supporting
reverse execution rather than requiring recording. I also noticed
there were a couple more instances of a requirement check, i.e. if []
which I changed to "require" per the current style for checking on the
test requirements.
The following patch fixes issues on PowerPC with the reverse-step and
reverse-next instructions when there are multiple assignment statements
on the same line and when there are multiple function calls on the same
line. The commit log below discusses these issues in further depth.
The discussion included what the correct operation should be for these
commands based on the GDB documentation. The proposed patch at that
time changed how the commands worked on other platforms such as X86 in
a way they no longer matched the documentation.
The issue is the line table contains multiple entries for the same
source line. The patch adds a function to search the line table to
find the address of the first instruction of a line. When setup up the
reverse stepping range, the function is called to make sure the start
of the range corresponds to the address of the first instruction for
the line. This approach was used. When Luis initially developed the
patch, he considered merging the contiguous ranges in the line table
when reading the line tables. He decided it was better to work with the
data directly in the line table rather than creating and using a
modified version of the line table.
The following patch fixes the execution of the reveres-step and
reverse-next commands for both senarios of multiple statements on the
same line for PowerPC and aarch64-linux. Unlike the previous patch, it
does not change the operation of the commands on other platforms, i.e.
X86. The patch adds new test cases for both scenarios to verify they
work correctly.
The patch has been tested on PowerPC, Intel X86 and aarch64-linux with
no new regression failures.
Please let me know if the patch is acceptable for mainline. Thanks.
Carl
---------------------------------------------
Fix reverse stepping multiple contiguous PC ranges over the line table.
There are a couple of scenarios where the GDB reverse-step and reverse-next
commands do not work correctly.
Scenario 1 issue description by Luis Machado:
When running GDB's testsuite on aarch64-linux/Ubuntu 20.04 (also spotted on
the ppc backend), I noticed some failures in gdb.reverse/solib-precsave.exp
and gdb.reverse/solib-reverse.exp.
The failure happens around the following code:
38 b[1] = shr2(17); /* middle part two */
40 b[0] = 6; b[1] = 9; /* generic statement, end part two */
42 shr1 ("message 1\n"); /* shr1 one */
Normal execution:
- step from line 38 will land on line 40.
- step from line 40 will land on line 42.
Reverse execution:
- step from line 42 will land on line 40.
- step from line 40 will land on line 40.
- step from line 40 will land on line 38.
The problem here is that line 40 contains two contiguous but distinct
PC ranges in the line table, like so:
Line 40 - [0x7ec ~ 0x7f4]
Line 40 - [0x7f4 ~ 0x7fc]
The two distinct ranges are generated because GCC started outputting source
column information, which GDB doesn't take into account at the moment.
When stepping forward from line 40, we skip both of these ranges and land on
line 42. When stepping backward from line 42, we stop at the start PC of the
second (or first, going backwards) range of line 40.
Since we've reached ecs->event_thread->control.step_range_start, we stop
stepping backwards.
---------------------------------------------------------
Scenario 2 issue described by Pedro Alves:
The following explanation of the issue was taken from the gdb mailing list
discussion of the withdrawn patch to change the behavior of the reverse-step
and reverse-next commands. Specifically, message from Pedro Alves
<pedro@palves.net> where he demonstrates the issue where you have multiple
function calls on the same source code line:
https://sourceware.org/pipermail/gdb-patches/2023-January/196110.html
The source line looks like:
func1 (); func2 ();
so stepping backwards over that line should always stop at the first
instruction of the line, not in the middle. Let's simplify this.
Here's the full source code of my example:
(gdb) list 1
1 void func1 ()
2 {
3 }
4
5 void func2 ()
6 {
7 }
8
9 int main ()
10 {
11 func1 (); func2 ();
12 }
Compiled with:
$ gcc reverse.c -o reverse -g3 -O0
$ gcc -v
...
gcc version 11.3.0 (Ubuntu 11.3.0-1ubuntu1~22.04)
Now let's debug it with target record, using current gdb git master (f3d8ae90b236),
without your patch:
$ gdb ~/reverse
GNU gdb (GDB) 14.0.50.20230124-git
...
Reading symbols from /home/pedro/reverse...
(gdb) start
Temporary breakpoint 1 at 0x1147: file reverse.c, line 11.
Starting program: /home/pedro/reverse
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
Temporary breakpoint 1, main () at reverse.c:11
11 func1 (); func2 ();
(gdb) record
(gdb) disassemble /s
Dump of assembler code for function main:
reverse.c:
10 {
0x000055555555513f <+0>: endbr64
0x0000555555555143 <+4>: push %rbp
0x0000555555555144 <+5>: mov %rsp,%rbp
11 func1 (); func2 ();
=> 0x0000555555555147 <+8>: mov $0x0,%eax
0x000055555555514c <+13>: call 0x555555555129 <func1>
0x0000555555555151 <+18>: mov $0x0,%eax
0x0000555555555156 <+23>: call 0x555555555134 <func2>
0x000055555555515b <+28>: mov $0x0,%eax
12 }
0x0000555555555160 <+33>: pop %rbp
0x0000555555555161 <+34>: ret
End of assembler dump.
(gdb) n
12 }
So far so good, a "next" stepped over the whole of line 11 and stopped at line 12.
Let's confirm where we are now:
(gdb) disassemble /s
Dump of assembler code for function main:
reverse.c:
10 {
0x000055555555513f <+0>: endbr64
0x0000555555555143 <+4>: push %rbp
0x0000555555555144 <+5>: mov %rsp,%rbp
11 func1 (); func2 ();
0x0000555555555147 <+8>: mov $0x0,%eax
0x000055555555514c <+13>: call 0x555555555129 <func1>
0x0000555555555151 <+18>: mov $0x0,%eax
0x0000555555555156 <+23>: call 0x555555555134 <func2>
0x000055555555515b <+28>: mov $0x0,%eax
12 }
=> 0x0000555555555160 <+33>: pop %rbp
0x0000555555555161 <+34>: ret
End of assembler dump.
Good, we're at the first instruction of line 12.
Now let's undo the "next", with "reverse-next":
(gdb) reverse-next
11 func1 (); func2 ();
Seemingly stopped at line 11. Let's see exactly where:
(gdb) disassemble /s
Dump of assembler code for function main:
reverse.c:
10 {
0x000055555555513f <+0>: endbr64
0x0000555555555143 <+4>: push %rbp
0x0000555555555144 <+5>: mov %rsp,%rbp
11 func1 (); func2 ();
0x0000555555555147 <+8>: mov $0x0,%eax
0x000055555555514c <+13>: call 0x555555555129 <func1>
=> 0x0000555555555151 <+18>: mov $0x0,%eax
0x0000555555555156 <+23>: call 0x555555555134 <func2>
0x000055555555515b <+28>: mov $0x0,%eax
12 }
0x0000555555555160 <+33>: pop %rbp
0x0000555555555161 <+34>: ret
End of assembler dump.
(gdb)
And lo, we stopped in the middle of line 11! That is a bug, we should have
stepped back all the way to the beginning of the line. The "reverse-next"
should have fully undone the prior "next" command.
The above issues were fixed by introducing a new function that looks for
adjacent PC ranges for the same line, until we notice a line change. Then
we take that as the start PC of the range. The new start PC for the range
is used for the control.step_range_start when setting up a step range.
The test case gdb.reverse/map-to-same-line.exp is added to test the fix
for the issues in scenario 1.
The test case gdb.reverse/func-map-to-same-line.exp was added to test the
fix for scenario 2 when the binary was compiled with and without line
table information.
bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28426
Co-authored-by: Luis Machado <luis.machado@arm.com>
Co-authored-by: Carl Love <cel@us.ibm.com>
Reviewed-By: Bruno Larsen <blarsen@redhat.com>
---
gdb/infrun.c | 57 +++++++
gdb/symtab.c | 49 ++++++
gdb/symtab.h | 16 ++
.../gdb.reverse/func-map-to-same-line.c | 36 ++++
.../gdb.reverse/func-map-to-same-line.exp | 156 ++++++++++++++++++
gdb/testsuite/gdb.reverse/map-to-same-line.c | 58 +++++++
.../gdb.reverse/map-to-same-line.exp | 156 ++++++++++++++++++
7 files changed, 528 insertions(+)
create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-line.c
create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.c
create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.exp
diff --git a/gdb/infrun.c b/gdb/infrun.c
index efe2c00c489..31cd817c733 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -114,6 +114,9 @@ static struct async_event_handler *infrun_async_inferior_event_token;
Starts off as -1, indicating "never enabled/disabled". */
static int infrun_is_async = -1;
+static CORE_ADDR update_line_range_start (CORE_ADDR pc,
+ struct execution_control_state *ecs);
+
/* See infrun.h. */
void
@@ -6769,6 +6772,25 @@ handle_signal_stop (struct execution_control_state *ecs)
process_event_stop_test (ecs);
}
+CORE_ADDR
+update_line_range_start (CORE_ADDR pc, struct execution_control_state *ecs)
+{
+ /* The line table may have multiple entries for the same source code line.
+ Given the PC, check the line table and return the PC that corresponds
+ to the line table entry for the source line that PC is in. */
+ CORE_ADDR start_line_pc = ecs->event_thread->control.step_range_start;
+ gdb::optional<CORE_ADDR> real_range_start;
+
+ /* Call find_line_range_start to get the smallest address in the
+ linetable for multiple Line X entries in the line table. */
+ real_range_start = find_line_range_start (pc);
+
+ if (real_range_start.has_value ())
+ start_line_pc = *real_range_start;
+
+ return start_line_pc;
+}
+
/* Come here when we've got some debug event / signal we can explain
(IOW, not a random signal), and test whether it should cause a
stop, or whether we should resume the inferior (transparently).
@@ -7570,6 +7592,28 @@ process_event_stop_test (struct execution_control_state *ecs)
if (stop_pc_sal.is_stmt)
{
+ if (execution_direction == EXEC_REVERSE)
+ {
+ /* We are stepping backwards make sure we have reached the
+ beginning of the line. */
+ CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
+ CORE_ADDR start_line_pc
+ = update_line_range_start (stop_pc, ecs);
+
+ if (stop_pc != start_line_pc)
+ {
+ /* Have not reached the beginning of the source code line.
+ Set a step range. Execution should stop in any function
+ calls we execute back into before reaching the beginning
+ of the line. */
+ ecs->event_thread->control.step_range_start = start_line_pc;
+ ecs->event_thread->control.step_range_end = stop_pc;
+ set_step_info (ecs->event_thread, frame, stop_pc_sal);
+ keep_going (ecs);
+ return;
+ }
+ }
+
/* We are at the start of a statement.
So stop. Note that we don't stop if we step into the middle of a
@@ -7632,6 +7676,19 @@ process_event_stop_test (struct execution_control_state *ecs)
set_step_info (ecs->event_thread, frame, stop_pc_sal);
infrun_debug_printf ("keep going");
+
+ if (execution_direction == EXEC_REVERSE)
+ {
+ CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
+
+ /* Make sure the stop_pc is set to the beginning of the line. */
+ if (stop_pc != ecs->event_thread->control.step_range_start)
+ {
+ stop_pc = update_line_range_start (stop_pc, ecs);
+ ecs->event_thread->control.step_range_start = stop_pc;
+ }
+ }
+
keep_going (ecs);
}
diff --git a/gdb/symtab.c b/gdb/symtab.c
index 27611a34ec4..91d35616eb9 100644
--- a/gdb/symtab.c
+++ b/gdb/symtab.c
@@ -3282,6 +3282,55 @@ find_pc_line (CORE_ADDR pc, int notcurrent)
return sal;
}
+/* Compare two symtab_and_line entries. Return true if both have
+ the same line number and the same symtab pointer. That means we
+ are dealing with two entries from the same line and from the same
+ source file.
+
+ Return false otherwise. */
+
+static bool
+sal_line_symtab_matches_p (const symtab_and_line &sal1,
+ const symtab_and_line &sal2)
+{
+ return (sal1.line == sal2.line && sal1.symtab == sal2.symtab);
+}
+
+/* See symtah.h. */
+
+gdb::optional<CORE_ADDR>
+find_line_range_start (CORE_ADDR pc)
+{
+ struct symtab_and_line current_sal = find_pc_line (pc, 0);
+
+ if (current_sal.line == 0)
+ return {};
+
+ struct symtab_and_line prev_sal = find_pc_line (current_sal.pc - 1, 0);
+
+ /* If the previous entry is for a different line, that means we are already
+ at the entry with the start PC for this line. */
+ if (!sal_line_symtab_matches_p (prev_sal, current_sal))
+ return current_sal.pc;
+
+ /* Otherwise, keep looking for entries for the same line but with
+ smaller PC's. */
+ bool done = false;
+ CORE_ADDR prev_pc;
+ while (!done)
+ {
+ prev_pc = prev_sal.pc;
+
+ prev_sal = find_pc_line (prev_pc - 1, 0);
+
+ /* Did we notice a line change? If so, we are done with the search. */
+ if (!sal_line_symtab_matches_p (prev_sal, current_sal))
+ done = true;
+ }
+
+ return prev_pc;
+}
+
/* See symtab.h. */
struct symtab *
diff --git a/gdb/symtab.h b/gdb/symtab.h
index 404d0ab30a8..f54305636da 100644
--- a/gdb/symtab.h
+++ b/gdb/symtab.h
@@ -2346,6 +2346,22 @@ extern struct symtab_and_line find_pc_line (CORE_ADDR, int);
extern struct symtab_and_line find_pc_sect_line (CORE_ADDR,
struct obj_section *, int);
+/* Given PC, and assuming it is part of a range of addresses that is part of a
+ line, go back through the linetable and find the starting PC of that
+ line.
+
+ For example, suppose we have 3 PC ranges for line X:
+
+ Line X - [0x0 - 0x8]
+ Line X - [0x8 - 0x10]
+ Line X - [0x10 - 0x18]
+
+ If we call the function with PC == 0x14, we want to return 0x0, as that is
+ the starting PC of line X, and the ranges are contiguous.
+*/
+
+extern gdb::optional<CORE_ADDR> find_line_range_start (CORE_ADDR pc);
+
/* Wrapper around find_pc_line to just return the symtab. */
extern struct symtab *find_pc_line_symtab (CORE_ADDR);
diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.c b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
new file mode 100644
index 00000000000..412ab180943
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
@@ -0,0 +1,36 @@
+/* Copyright 2008-2023 Free Software Foundation, Inc.
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 3 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program. If not, see <http://www.gnu.org/licenses/>.
+
+ This test is used to test the reverse-step and reverse-next instruction
+ execution for a source line that contains multiple function calls. */
+
+void
+func1 ()
+{
+} // END FUNC1
+
+void
+func2 ()
+{
+} // END FUNC2
+
+int main ()
+{
+ int a, b;
+ a = 1;
+ b = 2;
+ func1 (); func2 ();
+ a = a + b; // START REVERSE TEST
+}
diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
new file mode 100644
index 00000000000..da5ee282053
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
@@ -0,0 +1,156 @@
+# Copyright 2008-2023 Free Software Foundation, Inc.
+
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program. If not, see <http://www.gnu.org/licenses/>. */
+
+# This file is part of the GDB testsuite. It tests reverse stepping.
+# Lots of code borrowed from "step-test.exp".
+
+# This test checks to make sure there is no regression failures for
+# the reverse-next command when stepping back over two functions in
+# the same line.
+
+require supports_reverse
+
+# This test uses the gcc no-column-info command which was added in gcc 7.1.
+if {![test_compiler_info {gcc-*}]
+ || [test_compiler_info {gcc-[1-6]-*}]} {
+ return
+}
+
+proc run_tests {} {
+ global srcfile
+ global executable
+
+ runto_main
+ set target_remote [gdb_is_target_remote]
+
+ with_test_prefix "test1" {
+ gdb_test_no_output "record" "turn on process record"
+ }
+
+ # This regression test verifies the reverse-step and reverse-next commands
+ # work properly when executing backwards thru a source line containing
+ # two function calls on the same source line, i.e. func1 (); func2 ();
+ # This test is compiled so the dwarf info not contain the line table
+ # information.
+
+ # Test 1, reverse-next command
+ # Set breakpoint at the line after the function calls.
+ set bp_start_reverse_test [gdb_get_line_number "START REVERSE TEST" \
+ $srcfile]
+ gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
+
+ # Continue to break point for reverse-next test.
+ # Command definition: reverse-next [count]
+ # Run backward to the beginning of the previous line executed in the
+ # current (innermost) stack frame. If the line contains function calls,
+ # they will be “un-executed” without stopping. Starting from the first
+ # line of a function, reverse-next will take you back to the caller of
+ # that function, before the function was called, just as the normal next
+ # command would take you from the last line of a function back to its
+ # return to its caller 2 .
+ gdb_continue_to_breakpoint \
+ "test1: stopped at command reverse-next test start location" \
+ ".*$srcfile:$bp_start_reverse_test\r\n.*"
+
+ # The reverse-next should step all the way back to the beginning of the
+ # line, i.e. at the beginning of the func1 call.
+ gdb_test "reverse-next" ".*func1 \\(\\); func2 \\(\\);.*" \
+ "test1: reverse-next to line with two functions"
+
+ # We should be stopped at the first instruction of the line. A reverse-step
+ # should step back and stop at the beginning of the previous line b = 2,
+ # i.e. not in func1 ().
+ gdb_test "reverse-stepi" ".*b = 2;.*" \
+ "test1: reverse-stepi to previous line b = 2"
+
+
+ # Setup for test 2
+ clean_restart $executable
+ runto_main
+
+ with_test_prefix "test2" {
+ gdb_test_no_output "record" "turn on process record"
+ }
+
+ # Test 2, reverse-step command
+ # Set breakpoint at the line after the function calls.
+ gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
+
+ # Continue to the start of the reverse-step test.
+ # Command definition: reverse-step [count]
+ # Run the program backward until control reaches the start of a
+ # different source line; then stop it, and return control to gdb.
+ # Like the step command, reverse-step will only stop at the beginning
+ # of a source line. It “un-executes” the previously executed source
+ # line. If the previous source line included calls to debuggable
+ # functions, reverse-step will step (backward) into the called function,
+ # stopping at the beginning of the last statement in the called
+ # function (typically a return statement). Also, as with the step
+ # command, if non-debuggable functions are called, reverse-step will
+ # run thru them backward without stopping.
+
+ gdb_continue_to_breakpoint \
+ "test2: stopped at command reverse-step test start location" \
+ ".*$srcfile:$bp_start_reverse_test\r\n.*"
+
+ # The first reverse step should take us call of func2 ().
+ gdb_test "reverse-step" ".*END FUNC2.*" \
+ "test2: reverse-step into func2 "
+
+ # The second reverse step should take us into func1 ().
+ gdb_test "reverse-step" ".*END FUNC1.*" \
+ "test2: reverse-step into func1 "
+
+ # The third reverse step should take us call of func1 ().
+ gdb_test "reverse-step" ".*func1 \\(\\); func2 \\(\\);.*" \
+ "test2: reverse-step to line func1(); func2(), at call for func1 "
+
+ # We should be stopped at the first instruction of the line. A reverse
+ # stepi should take us to b = 2 ().
+ gdb_test "reverse-stepi" ".*b = 2;.*" \
+ "test2: reverse-stepi to line b = 2 "
+}
+
+set srcfile func-map-to-same-line.c
+set executable func-map-to-same-line
+
+# test with gcc column info enabled
+set options [list debug additional_flags=]
+
+if {[build_executable "failed to prepare" $executable $srcfile $options] == -1}\
+ {
+ return -1
+}
+
+clean_restart $executable
+
+with_test_prefix "with-column-info" {
+ run_tests
+}
+
+#test with gcc column info disabled
+set options [list debug additional_flags=-gno-column-info]
+
+if {[build_executable "failed to prepare" $executable $srcfile $options] == -1}\
+ {
+ return -1
+}
+
+set $executable executable_without_column_info
+clean_restart $executable
+
+with_test_prefix "no-column-info" {
+ run_tests
+}
diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.c b/gdb/testsuite/gdb.reverse/map-to-same-line.c
new file mode 100644
index 00000000000..f20d778f40e
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/map-to-same-line.c
@@ -0,0 +1,58 @@
+/* Copyright 2008-2023 Free Software Foundation, Inc.
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 3 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program. If not, see <http://www.gnu.org/licenses/ >. */
+
+/* The purpose of this test is to create a DWARF line table that contains two
+ or more entries for the same line. When stepping (forwards or backwards),
+ GDB should step over the entire line and not just a particular entry in the
+ line table. */
+
+int
+main ()
+{ /* TAG: main prologue */
+ asm ("main_label: .globl main_label");
+ int i = 1, j = 2, k;
+ float f1 = 2.0, f2 = 4.1, f3;
+ const char *str_1 = "foo", *str_2 = "bar", *str_3;
+
+ asm ("line1: .globl line1");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 1 */
+
+ asm ("line2: .globl line2");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 2 */
+
+ asm ("line3: .globl line3");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 3 */
+
+ asm ("line4: .globl line4");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 4 */
+
+ asm ("line5: .globl line5");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 5 */
+
+ asm ("line6: .globl line6");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 6 */
+
+ asm ("line7: .globl line7");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 7 */
+
+ asm ("line8: .globl line8");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 8 */
+
+ asm ("main_return: .globl main_return");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: main return */
+
+ asm ("end_of_sequence: .globl end_of_sequence");
+ return 0; /* TAG: main return */
+}
diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.exp b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
new file mode 100644
index 00000000000..16a359d90ec
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
@@ -0,0 +1,156 @@
+# Copyright 2008-2023 Free Software Foundation, Inc.
+
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program. If not, see <http://www.gnu.org/licenses/ >.
+
+# When stepping (forwards or backwards), GDB should step over the entire line
+# and not just a particular entry in the line table. This test was added to
+# verify the find_line_range_start function properly sets the step range for a
+# line that consists of multiple statements, i.e. multiple entries in the line
+# table. This test creates a DWARF line table that contains two entries for
+# the same line to do the needed testing.
+
+# This test can only be run on targets which support DWARF-2 and use gas.
+load_lib dwarf.exp
+require dwarf2_support
+
+# The DWARF assembler requires the gcc compiler.
+require is_c_compiler_gcc
+
+# This test suitable only for process that can do reverse execution
+require supports_reverse
+
+standard_testfile .c .S
+
+if { [prepare_for_testing "failed to prepare" ${testfile} ${srcfile}] } {
+ return -1
+}
+
+set asm_file [standard_output_file $srcfile2]
+Dwarf::assemble $asm_file {
+ global srcdir subdir srcfile
+ declare_labels integer_label L
+
+ # Find start address and length of program
+ lassign [function_range main [list ${srcdir}/${subdir}/$srcfile]] \
+ main_start main_len
+ set main_end "$main_start + $main_len"
+
+ cu {} {
+ compile_unit {
+ {language @DW_LANG_C}
+ {name map-to-same-line.c}
+ {stmt_list $L DW_FORM_sec_offset}
+ {low_pc 0 addr}
+ } {
+ subprogram {
+ {external 1 flag}
+ {name main}
+ {low_pc $main_start addr}
+ {high_pc $main_len DW_FORM_data4}
+ }
+ }
+ }
+
+ lines {version 2 default_is_stmt 1} L {
+ include_dir "${srcdir}/${subdir}"
+ file_name "$srcfile" 1
+
+ # Generate the line table program with distinct source lines being
+ # mapped to the same line entry. Line 1, 5 and 8 contain 1 statement
+ # each. Line 2 contains 2 statements. Line 3 contains 3 statements.
+ program {
+ DW_LNE_set_address $main_start
+ line [gdb_get_line_number "TAG: main prologue"]
+ DW_LNS_copy
+ DW_LNE_set_address line1
+ line [gdb_get_line_number "TAG: line 1" ]
+ DW_LNS_copy
+ DW_LNE_set_address line2
+ line [gdb_get_line_number "TAG: line 2" ]
+ DW_LNS_copy
+ DW_LNE_set_address line3
+ line [gdb_get_line_number "TAG: line 2" ]
+ DW_LNS_copy
+ DW_LNE_set_address line4
+ line [gdb_get_line_number "TAG: line 3" ]
+ DW_LNS_copy
+ DW_LNE_set_address line5
+ line [gdb_get_line_number "TAG: line 3" ]
+ DW_LNS_copy
+ DW_LNE_set_address line6
+ line [gdb_get_line_number "TAG: line 3" ]
+ DW_LNS_copy
+ DW_LNE_set_address line7
+ line [gdb_get_line_number "TAG: line 5" ]
+ DW_LNS_copy
+ DW_LNE_set_address line8
+ line [gdb_get_line_number "TAG: line 8" ]
+ DW_LNS_copy
+ DW_LNE_set_address main_return
+ line [gdb_get_line_number "TAG: main return"]
+ DW_LNS_copy
+ DW_LNE_set_address end_of_sequence
+ DW_LNE_end_sequence
+ }
+ }
+}
+
+if { [prepare_for_testing "failed to prepare" ${testfile} \
+ [list $srcfile $asm_file] {nodebug} ] } {
+ return -1
+}
+
+runto_main
+
+# Print the line table
+gdb_test_multiple "maint info line-table ${testfile}" "" {
+ -re "\r\n$decimal\[ \t\]+$decimal\[ \t\]+($hex)\[ \t\]+Y\[^\r\n\]*" {
+ lappend is_stmt $expect_out(1,string)
+ exp_continue
+ }
+ -re -wrap "" {
+ }
+}
+
+# Do the reverse-step test
+gdb_test_no_output "record" "turn on process record"
+
+set bp_main_return [gdb_get_line_number "TAG: main return" $srcfile]
+gdb_breakpoint $srcfile:$bp_main_return
+gdb_continue_to_breakpoint "run to end of main, reverse-step test" ".*$srcfile:$bp_main_return.*"
+gdb_test "display \$pc" ".*pc =.*" "display pc, reverse-step test"
+
+# At this point, GDB has already recorded the execution up until the return
+# statement. Reverse-step and test if GDB transitions between lines in the
+# expected order. It should reverse-step across lines 8, 5, 3, 2 and 1.
+foreach line {8 5 3 2 1} {
+ gdb_test "reverse-step" ".*TAG: line $line.*" "reverse step to line $line"
+}
+
+## Clean restart, test reverse-next command
+clean_restart ${testfile}
+runto_main
+gdb_test_no_output "record" "turn on process record, reverst-next test"
+
+set bp_main_return [gdb_get_line_number "TAG: main return" $srcfile]
+gdb_breakpoint $srcfile:$bp_main_return
+gdb_continue_to_breakpoint "run to end of main, reverse-next test" ".*$srcfile:$bp_main_return.*"
+gdb_test "display \$pc" ".*pc =.*" "display pc, reverse-next test"
+
+# At this point, GDB has already recorded the execution up until the return
+# statement. Reverse-next and test if GDB transitions between lines in the
+# expected order. It should reverse-next across lines 8, 5, 3, 2 and 1.
+foreach line {8 5 3 2 1} {
+ gdb_test "reverse-next" ".*TAG: line $line.*" "reverse next to line $line"
+}
--
2.37.2
^ permalink raw reply [flat|nested] 40+ messages in thread
* Re: [PATCH v3] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-05-10 17:16 ` Carl Love
2023-05-10 17:32 ` [PATCH v4] " Carl Love
@ 2023-05-11 7:52 ` Bruno Larsen
1 sibling, 0 replies; 40+ messages in thread
From: Bruno Larsen @ 2023-05-11 7:52 UTC (permalink / raw)
To: Carl Love, gdb-patches, UlrichWeigand, pedro; +Cc: luis.machado
On 10/05/2023 19:16, Carl Love wrote:
>>> +
>>> +require supports_reverse
>>> +
>>> +# This test uses the gcc no-column-info command which was added in
>>> gcc 7.1.
>>> +require get_compiler_info "gcc-7-*"
> I put the compiler check in last. When I ran it, I obviously didn't
> double check gdb/testsuite/gdb.log to make sure it really worked. I
> normally try to make a point of double checking the log file. I have
> been burned before thinking it was OK when there were no errors visible
> on the command line. The above command fails if you check the log file.
>> By constructing your regex like this, you are only allowing this test
>> to
>> be run on gcc 7. Anything later is also not accepted. I would do
>> something like (Warning, untested)
>>
>> require get_compiler_info "gcc"
>> require !get_compiler_info "gcc-[1-6]-*"
> I couldn't get require to work like that. The get_compiler_info
> doesn't seem take "gcc" as an argument.
>
> I changed the test to:
>
> if {![test_compiler_info {gcc-*}]
> || [test_compiler_info {gcc-[1-6]-*}]} {
> return
> }
>
> With this, I do see the correct number of passes in
> gdb/testsuite/gdb.log.
Yeah, this is perfectly fine IMO, since this is a more complicated use
of "require".
>
>> Which requires gcc, but does not allow versions 1 to 6. There is
>> probably a way to do it with a single require line, but I'm not the
>> best
>> with regexes.
>>
>>> +
>>> +proc run_tests {msg} {
>>> + global srcfile
>>> + global executable
>>> +
>>> + runto_main
>>> + set target_remote [gdb_is_target_remote]
>> When probing for target remote, GDB will emit pass/fails with
>> hardcoded
>> names, so the current proc setup gives us some duplicated test names.
>>
>> I would suggest that, instead of passing a message as a parameter,
>> you
>> wrapped all function calls in a with_test_prefix scope, like:
>>
>> with_test_prefix "with-column-info" {
>> run_test
>> }
> OK, I changed from passing in an argument and did the wrapped calls to
> run_test instead. Note, this still didn't fix the duplicate test names
> for turning on record.
>
>>> +
>>> + gdb_test_no_output "record" "turn on process record"
>>> +
>>> + # This regression test verifies the reverse-step and reverse-
>>> next commands
>>> + # work properly when executing backwards thru a source line
>>> containing
>>> + # two function calls on the same source line, i.e. func1 ();
>>> func2 ();
>>> + # This test is compiled so the dwarf info not contain the line
>>> table
>>> + # information.
>>> +
>>> + # Test 1, reverse-next command
>>> + # Set breakpoint at the line after the function calls.
>>> + set bp_start_reverse_test [gdb_get_line_number "START REVERSE
>>> TEST" \
>>> + $srcfile]
>>> + gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
>>> +
>>> + # Continue to break point for reverse-next test.
>>> + # Command definition: reverse-next [count]
>>> + # Run backward to the beginning of the previous line
>>> executed in the
>>> + # current (innermost) stack frame. If the line contains
>>> function calls,
>>> + # they will be “un-executed” without stopping. Starting from
>>> the first
>>> + # line of a function, reverse-next will take you back to the
>>> caller of
>>> + # that function, before the function was called, just as the
>>> normal next
>>> + # command would take you from the last line of a function
>>> back to its
>>> + # return to its caller 2 .
>>> + gdb_continue_to_breakpoint \
>>> + "$msg: test1: stopped at command reverse-next test start
>>> location" \
>>> + ".*$srcfile:$bp_start_reverse_test\r\n.*"
>>> +
>>> + # The reverse-next should step all the way back to the
>>> beginning of the
>>> + # line, i.e. at the beginning of the func1 call.
>>> + gdb_test "reverse-next" ".*func1 \\(\\); func2 \\(\\);.*" \
>>> + "$msg: test1: reverse-next to line with two functions"
>>> +
>>> + # We should be stopped at the first instruction of the line. A
>>> reverse-step
>>> + # should step back and stop at the beginning of the previous
>>> line b = 2,
>>> + # i.e. not in func1 ().
>>> + gdb_test "reverse-stepi" ".*b = 2;.*" \
>>> + "$msg: test1: reverse-stepi to previous line b = 2"
>>> +
>>> +
>>> + # Setup for test 2
>>> + clean_restart $executable
>>> + runto_main
>>> +
>>> + gdb_test_no_output "record" "turn on process record"
>> This gives a duplicate test name from setting up for the first test.
>> Adding "test 2:(...)" in here solves it.
> I couldn't figure out how to get the above syntax to work. So I used
> the with_test_prefix instead, i.e.
oops, sorry, that's what I get for sending a patch review at the end of
a long work day. That wasn't some arcane magic of TCL, it was just me
being too lazy to type out the whole message. Sorry for the wild goose
chase.
>
> with_test_prefix "test2" {
> gdb_test_no_output "record" "turn on process record"
> }
>
> That fixed the duplicate names. I also wrapped the first record with
> "test1" for consistency.
>
--
Cheers,
Bruno
^ permalink raw reply [flat|nested] 40+ messages in thread
* Re: [PATCH] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-05-02 14:15 ` Bruno Larsen
2023-05-02 15:40 ` Carl Love
@ 2023-05-11 15:11 ` Simon Marchi
1 sibling, 0 replies; 40+ messages in thread
From: Simon Marchi @ 2023-05-11 15:11 UTC (permalink / raw)
To: Bruno Larsen, Carl Love, gdb-patches, Ulrich Weigand, pedro; +Cc: luis.machado
> I wanted to test this change, but it doesn't apply anymore on master,
> and `git apply --3way` can't figure out how to do it. Which commit did
> you use as base (or alternatively, can you rebase it)?
To help with this, you can consider setting the `format.useAutoBase`
configuration option in git (documented in the git-format-patch man
page).
Simon
^ permalink raw reply [flat|nested] 40+ messages in thread
* Re: [PATCH v4] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-05-10 17:32 ` [PATCH v4] " Carl Love
@ 2023-05-11 16:01 ` Simon Marchi
2023-05-11 16:23 ` Bruno Larsen
2023-05-16 22:54 ` [PATCH v4] " Carl Love
0 siblings, 2 replies; 40+ messages in thread
From: Simon Marchi @ 2023-05-11 16:01 UTC (permalink / raw)
To: Carl Love, Bruno Larsen, gdb-patches, UlrichWeigand, pedro; +Cc: luis.machado
I'd like to help reviewing this, but I don't have much time at the
moment, so just a few comments on one test to start with.
> diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.c b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> new file mode 100644
> index 00000000000..412ab180943
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> @@ -0,0 +1,36 @@
> +/* Copyright 2008-2023 Free Software Foundation, Inc.
> +
> + This program is free software; you can redistribute it and/or modify
> + it under the terms of the GNU General Public License as published by
> + the Free Software Foundation; either version 3 of the License, or
> + (at your option) any later version.
> +
> + This program is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> + GNU General Public License for more details.
> +
> + You should have received a copy of the GNU General Public License
> + along with this program. If not, see <http://www.gnu.org/licenses/>.
> +
> + This test is used to test the reverse-step and reverse-next instruction
> + execution for a source line that contains multiple function calls. */
> +
> +void
> +func1 ()
> +{
> +} // END FUNC1
Use /* */ for comments, for consistency with the rest of the code base.
> +
> +void
> +func2 ()
> +{
> +} // END FUNC2
> +
> +int main ()
> +{
> + int a, b;
> + a = 1;
> + b = 2;
> + func1 (); func2 ();
> + a = a + b; // START REVERSE TEST
> +}
> diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> new file mode 100644
> index 00000000000..da5ee282053
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> @@ -0,0 +1,156 @@
> +# Copyright 2008-2023 Free Software Foundation, Inc.
> +
> +# This program is free software; you can redistribute it and/or modify
> +# it under the terms of the GNU General Public License as published by
> +# the Free Software Foundation; either version 3 of the License, or
> +# (at your option) any later version.
> +#
> +# This program is distributed in the hope that it will be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> +# GNU General Public License for more details.
> +#
> +# You should have received a copy of the GNU General Public License
> +# along with this program. If not, see <http://www.gnu.org/licenses/>. */
> +
> +# This file is part of the GDB testsuite. It tests reverse stepping.
> +# Lots of code borrowed from "step-test.exp".
> +
> +# This test checks to make sure there is no regression failures for
> +# the reverse-next command when stepping back over two functions in
> +# the same line.
> +
> +require supports_reverse
> +
> +# This test uses the gcc no-column-info command which was added in gcc 7.1.
> +if {![test_compiler_info {gcc-*}]
> + || [test_compiler_info {gcc-[1-6]-*}]} {
> + return
> +}
I would prefer not to filter out by compiler explicitly like that.
It would be useful for the test to run with other compilers too.
> +
> +proc run_tests {} {
> + global srcfile
> + global executable
> +
> + runto_main
> + set target_remote [gdb_is_target_remote]
> +
> + with_test_prefix "test1" {
> + gdb_test_no_output "record" "turn on process record"
> + }
> +
> + # This regression test verifies the reverse-step and reverse-next commands
> + # work properly when executing backwards thru a source line containing
> + # two function calls on the same source line, i.e. func1 (); func2 ();
> + # This test is compiled so the dwarf info not contain the line table
> + # information.
> +
> + # Test 1, reverse-next command
> + # Set breakpoint at the line after the function calls.
> + set bp_start_reverse_test [gdb_get_line_number "START REVERSE TEST" \
> + $srcfile]
> + gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> +
> + # Continue to break point for reverse-next test.
> + # Command definition: reverse-next [count]
> + # Run backward to the beginning of the previous line executed in the
> + # current (innermost) stack frame. If the line contains function calls,
> + # they will be “un-executed” without stopping. Starting from the first
> + # line of a function, reverse-next will take you back to the caller of
> + # that function, before the function was called, just as the normal next
> + # command would take you from the last line of a function back to its
> + # return to its caller 2 .
> + gdb_continue_to_breakpoint \
> + "test1: stopped at command reverse-next test start location" \
> + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> +
> + # The reverse-next should step all the way back to the beginning of the
> + # line, i.e. at the beginning of the func1 call.
> + gdb_test "reverse-next" ".*func1 \\(\\); func2 \\(\\);.*" \
> + "test1: reverse-next to line with two functions"
> +
> + # We should be stopped at the first instruction of the line. A reverse-step
> + # should step back and stop at the beginning of the previous line b = 2,
> + # i.e. not in func1 ().
> + gdb_test "reverse-stepi" ".*b = 2;.*" \
> + "test1: reverse-stepi to previous line b = 2"
> +
> +
> + # Setup for test 2
> + clean_restart $executable
> + runto_main
> +
> + with_test_prefix "test2" {
> + gdb_test_no_output "record" "turn on process record"
> + }
> +
> + # Test 2, reverse-step command
> + # Set breakpoint at the line after the function calls.
> + gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> +
> + # Continue to the start of the reverse-step test.
> + # Command definition: reverse-step [count]
> + # Run the program backward until control reaches the start of a
> + # different source line; then stop it, and return control to gdb.
> + # Like the step command, reverse-step will only stop at the beginning
> + # of a source line. It “un-executes” the previously executed source
> + # line. If the previous source line included calls to debuggable
> + # functions, reverse-step will step (backward) into the called function,
> + # stopping at the beginning of the last statement in the called
> + # function (typically a return statement). Also, as with the step
> + # command, if non-debuggable functions are called, reverse-step will
> + # run thru them backward without stopping.
> +
> + gdb_continue_to_breakpoint \
> + "test2: stopped at command reverse-step test start location" \
> + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> +
> + # The first reverse step should take us call of func2 ().
> + gdb_test "reverse-step" ".*END FUNC2.*" \
> + "test2: reverse-step into func2 "
> +
> + # The second reverse step should take us into func1 ().
> + gdb_test "reverse-step" ".*END FUNC1.*" \
> + "test2: reverse-step into func1 "
> +
> + # The third reverse step should take us call of func1 ().
> + gdb_test "reverse-step" ".*func1 \\(\\); func2 \\(\\);.*" \
> + "test2: reverse-step to line func1(); func2(), at call for func1 "
> +
> + # We should be stopped at the first instruction of the line. A reverse
> + # stepi should take us to b = 2 ().
> + gdb_test "reverse-stepi" ".*b = 2;.*" \
> + "test2: reverse-stepi to line b = 2 "
> +}
> +
> +set srcfile func-map-to-same-line.c
> +set executable func-map-to-same-line
> +
> +# test with gcc column info enabled
> +set options [list debug additional_flags=]
> +
> +if {[build_executable "failed to prepare" $executable $srcfile $options] == -1}\
> + {
> + return -1
> +}
> +
> +clean_restart $executable
> +
> +with_test_prefix "with-column-info" {
> + run_tests
> +}
So, the above assumes that the compiler generates column-info by
default, which has not historically been the case for GCC (it started to
emit columns by default with version 8, according to my tests). Other
compilers may choose to not emit them by default.
I think it would make sense to make gdb_compile recognize the new
"column-info" and "no-column-info" options, which would translate to the
right flags for the given compiler. gdb_compile already handles the
nitty gritty details of choosing compiler flags for specific compiler
versions. This way, individual tests don't contain compiler flags that
are possibly compiler-specific.
> +
> +#test with gcc column info disabled
> +set options [list debug additional_flags=-gno-column-info]
> +
> +if {[build_executable "failed to prepare" $executable $srcfile $options] == -1}\
> + {
> + return -1
> +}
> +
> +set $executable executable_without_column_info
> +clean_restart $executable
> +
> +with_test_prefix "no-column-info" {
> + run_tests
> +}
This would probably be a good use for foreach_with_prefix (if you can
make it work), to make things more compact:
foreach_with_prefix with_column_info {yes no} {
}
... or something like that.
Simon
^ permalink raw reply [flat|nested] 40+ messages in thread
* Re: [PATCH v4] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-05-11 16:01 ` Simon Marchi
@ 2023-05-11 16:23 ` Bruno Larsen
2023-05-11 17:28 ` Simon Marchi
2023-05-16 22:54 ` [PATCH v4] " Carl Love
1 sibling, 1 reply; 40+ messages in thread
From: Bruno Larsen @ 2023-05-11 16:23 UTC (permalink / raw)
To: Simon Marchi, Carl Love, gdb-patches, UlrichWeigand, pedro; +Cc: luis.machado
On 11/05/2023 18:01, Simon Marchi wrote:
>> +with_test_prefix "with-column-info" {
>> + run_tests
>> +}
> So, the above assumes that the compiler generates column-info by
> default, which has not historically been the case for GCC (it started to
> emit columns by default with version 8, according to my tests). Other
> compilers may choose to not emit them by default.
Yes, column info started being generated in gcc7 and was made default in
gcc 8
>
> I think it would make sense to make gdb_compile recognize the new
> "column-info" and "no-column-info" options, which would translate to the
> right flags for the given compiler. gdb_compile already handles the
> nitty gritty details of choosing compiler flags for specific compiler
> versions. This way, individual tests don't contain compiler flags that
> are possibly compiler-specific.
>
I was going to suggest something similar in an earlier revision, but
when I tried to look for how to control it in clang, I couldn't see it
at all, that's why I thought it was OK to restrict it to gcc only. Can
clang (or other compilers for that matter) emit this information?
Also, how would gdb_compile handle if the current compiler doesn't
support a given option, but the others do? Should it loudly fail, or
silently ignore the "broken" option? If the second, I guess there is no
harm in allowing clang to run these tests and testing the same scenario
twice
--
Cheers,
Bruno
^ permalink raw reply [flat|nested] 40+ messages in thread
* Re: [PATCH v4] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-05-11 16:23 ` Bruno Larsen
@ 2023-05-11 17:28 ` Simon Marchi
2023-05-16 22:54 ` [PATCH 1/2] " Carl Love
2023-05-16 22:54 ` [PATCH 2/2 v5] " Carl Love
0 siblings, 2 replies; 40+ messages in thread
From: Simon Marchi @ 2023-05-11 17:28 UTC (permalink / raw)
To: Bruno Larsen, Carl Love, gdb-patches, UlrichWeigand, pedro; +Cc: luis.machado
> I was going to suggest something similar in an earlier revision, but
> when I tried to look for how to control it in clang, I couldn't see it
> at all, that's why I thought it was OK to restrict it to gcc only. Can
> clang (or other compilers for that matter) emit this information?clang does, yes, with the same flags as gcc:
https://clang.llvm.org/docs/ClangCommandLineReference.html#cmdoption-clang-gcolumn-info
> Also, how would gdb_compile handle if the current compiler doesn't
> support a given option, but the others do? Should it loudly fail, or
> silently ignore the "broken" option? If the second, I guess there is
> no harm in allowing clang to run these tests and testing the same
> scenario twice
I'm not sure, you'd have to look at how other options are handled.
But intuitively, if gdb_compile isn't able to fulfill your request, then
it should fail. For instance, if you used no-column-info with gcc 6
(which doesn't support column info at all), gdb_compile should succeed,
even if there isn't an option to disable column info with that compiler.
If you used column-info with gcc 6, gdb_compile would fail.
If there exists some compiler that always emits column info, with no
option to turn it off, then column-info would work with that compiler,
but no-column-info wouldn't.
To start with, I think it's fine to make column-info and no-column-info
map to -gcolumn-info and -gno-column-info. And then, you can probably
teach gdb_compile about older gccs and older clangs that don't support
column info.
If people test with other compilers that don't support -gcolumn-info or
-gno-column-info, the test won't compile. They can then contribute
support for the column-info / no-column-info options for that compiler.
Simon
^ permalink raw reply [flat|nested] 40+ messages in thread
* RE: [PATCH v4] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-05-11 16:01 ` Simon Marchi
2023-05-11 16:23 ` Bruno Larsen
@ 2023-05-16 22:54 ` Carl Love
1 sibling, 0 replies; 40+ messages in thread
From: Carl Love @ 2023-05-16 22:54 UTC (permalink / raw)
To: Simon Marchi, Bruno Larsen, gdb-patches, UlrichWeigand, pedro
Cc: luis.machado, cel
Simon:
On Thu, 2023-05-11 at 12:01 -0400, Simon Marchi wrote:
> >
<snip>
> I'd like to help reviewing this, but I don't have much time at the
> moment, so just a few comments on one test to start with.
>
> > + This test is used to test the reverse-step and reverse-next
> > instruction
> > + execution for a source line that contains multiple function
> > calls. */
> > +
> > +void
> > +func1 ()
> > +{
> > +} // END FUNC1
>
> Use /* */ for comments, for consistency with the rest of the code
> base.
>
OK, changed all instances of // comment.
> > +
> > +void
> > +func2 ()
> > +{
> > +} // END FUNC2
> > +
> > +int main ()
> > +{
> > + int a, b;
> > + a = 1;
> > + b = 2;
> > + func1 (); func2 ();
> > + a = a + b; // START REVERSE TEST
> > +}
> >
<snip>
> > +
> > +require supports_reverse
> > +
> > +# This test uses the gcc no-column-info command which was added in
> > gcc 7.1.
> > +if {![test_compiler_info {gcc-*}]
> > + || [test_compiler_info {gcc-[1-6]-*}]} {
> > + return
> > +}
>
> I would prefer not to filter out by compiler explicitly like that.
> It would be useful for the test to run with other compilers too.
OK, implemented the gdb_compile options. Moved the specific gcc
version test there.
>
> > +
> > +proc run_tests {} {
> > + global srcfile
<snip>
> > +
> > +set srcfile func-map-to-same-line.c
> > +set executable func-map-to-same-line
> > +
> > +# test with gcc column info enabled
> > +set options [list debug additional_flags=]
> > +
> > +if {[build_executable "failed to prepare" $executable $srcfile
> > $options] == -1}\
> > + {
> > + return -1
> > +}
> > +
> > +clean_restart $executable
> > +
> > +with_test_prefix "with-column-info" {
> > + run_tests
> > +}
>
> So, the above assumes that the compiler generates column-info by
> default, which has not historically been the case for GCC (it started
> to
> emit columns by default with version 8, according to my
> tests). Other
> compilers may choose to not emit them by default.
>
> I think it would make sense to make gdb_compile recognize the new
> "column-info" and "no-column-info" options, which would translate to
> the
> right flags for the given compiler. gdb_compile already handles the
> nitty gritty details of choosing compiler flags for specific compiler
> versions. This way, individual tests don't contain compiler flags
> that
> are possibly compiler-specific.
OK, I think I have this implemented as suggested. It does seem to
work. I added both gcc and clang support. Per the link to clang which
does have line info options.
I put the new gdb_compile option support in a separate patch, followed
by the updated "Fix reverse stepping multiple.... " patch.
I tested the test case with:
make check RUNTESTFLAGS='CC_FOR_TARGET=clang GDB=.../gdb gdb.reverse/func-map-to-same-line.exp' > out
and
make check RUNTESTFLAGS='GDB=.../gdb gdb.reverse/func-map-to-same-line.exp' > out
I looked in the gdb/testsuite/gdb.log file to verify that the compiler
line (gcc or clang) explicitly has the -gcolumn-info for the first
test and -gno-column-info for the second test. The expected number
of success were seen for the test with gcc and clang. It all looks
like it works.
>
> > +
> > +#test with gcc column info disabled
> > +set options [list debug additional_flags=-gno-column-info]
> > +
> > +if {[build_executable "failed to prepare" $executable $srcfile
> > $options] == -1}\
> > + {
> > + return -1
> > +}
> > +
> > +set $executable executable_without_column_info
> > +clean_restart $executable
> > +
> > +with_test_prefix "no-column-info" {
> > + run_tests
> > +}
>
> This would probably be a good use for foreach_with_prefix (if you can
> make it work), to make things more compact:
>
> foreach_with_prefix with_column_info {yes no} {
> }
>
> ... or something like that.
Changed to the foreach_with_prefix.
Thanks for the review.
Will post version 5 as a series of two patches.
Carl
^ permalink raw reply [flat|nested] 40+ messages in thread
* [PATCH 1/2] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-05-11 17:28 ` Simon Marchi
@ 2023-05-16 22:54 ` Carl Love
2023-06-19 17:11 ` Simon Marchi
2023-05-16 22:54 ` [PATCH 2/2 v5] " Carl Love
1 sibling, 1 reply; 40+ messages in thread
From: Carl Love @ 2023-05-16 22:54 UTC (permalink / raw)
To: Simon Marchi, Bruno Larsen, gdb-patches, UlrichWeigand, pedro
Cc: luis.machado, cel
Simon, GDB maintainers:
Per the comments on version 4 for the gdb.reverse/func-map-to-same-
line.exp, I have added support to proc gdb_compile to enable or disable
generating line information as part of the debug information. The two
new options are column-info and no-column-info.
This patch implements the new options for gdb_compile.
These options have been tested with patch 2 of 2 on PowerPC with the
GCC and clang compilers.
Please let me know if the patch is acceptable for mainline. Thanks.
Carl
-------------------------------------------------------------
Add gdb_compile options column-info and no-column-info
This patch adds two new options to gdb_compile to specify if the compile
should or should not generate the line table information. The
options are supported on clang and gcc version 7 and newer.
Patch has been tested on PowerPC with both gcc and clang.
---
gdb/testsuite/lib/gdb.exp | 30 ++++++++++++++++++++++++++++++
1 file changed, 30 insertions(+)
diff --git a/gdb/testsuite/lib/gdb.exp b/gdb/testsuite/lib/gdb.exp
index aed7e2d043c..e993fddf4c7 100644
--- a/gdb/testsuite/lib/gdb.exp
+++ b/gdb/testsuite/lib/gdb.exp
@@ -4794,6 +4794,8 @@ proc quote_for_host { args } {
# debug information
# - text_segment=addr: Tell the linker to place the text segment at ADDR.
# - build-id: Ensure the final binary includes a build-id.
+# - no-column-info: Disable generation of column table information.
+# - column-info: Enable generation of column table information.
#
# And here are some of the not too obscure options understood by DejaGnu that
# influence the compilation:
@@ -5003,6 +5005,34 @@ proc gdb_compile {source dest type options} {
} else {
error "Don't know how to handle text_segment option."
}
+ } elseif { $opt == "column-info" } {
+ if {[test_compiler_info {gcc-*}]} {
+ if {[test_compiler_info {gcc-[1-6]-*}]} {
+ error "gdb_compile option no-column-info not supported."
+ }
+ lappend new_options "additional_flags=-gcolumn-info"
+
+ } elseif {[test_compiler_info {clang-*}]} {
+ lappend new_options "additional_flags=-gcolumn-info"
+
+ } else {
+ error "Don't know how to handle gcolumn-info option."
+ }
+
+ } elseif { $opt == "no-column-info" } {
+ if {[test_compiler_info {gcc-*}]} {
+ if {[test_compiler_info {gcc-[1-6]-*}]} {
+ error "gdb_compile option no-column-info not supported."
+ }
+ lappend new_options "additional_flags=-gno-column-info"
+
+ } elseif {[test_compiler_info {clang-*}]} {
+ lappend new_options "additional_flags=-gno-column-info"
+
+ } else {
+ error "Don't know how to handle gno-column-info option."
+ }
+
} else {
lappend new_options $opt
}
--
2.37.2
^ permalink raw reply [flat|nested] 40+ messages in thread
* [PATCH 2/2 v5] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-05-11 17:28 ` Simon Marchi
2023-05-16 22:54 ` [PATCH 1/2] " Carl Love
@ 2023-05-16 22:54 ` Carl Love
2023-05-25 15:08 ` Carl Love
2023-06-19 17:58 ` Simon Marchi
1 sibling, 2 replies; 40+ messages in thread
From: Carl Love @ 2023-05-16 22:54 UTC (permalink / raw)
To: Simon Marchi, Bruno Larsen, gdb-patches, UlrichWeigand, pedro
Cc: luis.machado, cel
Bruno, Simon, GDB maintainers:
Version 5, changed comments in test case func-map-to-same-line.c.
Patch 1/2 implemented the new options for gdb_compile. Updated the
call to proc run_tests to use the new gdb_compile options in a
foreach_with_prefix loop.
Version 4, additional fixes for gcc version check, wrap function calls
using "with_test_prefix", move load_lib dwarf.exe. Fixed typo noted by
Luis.
Version 3, added the gcc version check as discussed further from
version 2 of the patch. Also updated the tests to check for supporting
reverse execution rather than requiring recording. I also noticed
there were a couple more instances of a requirement check, i.e. if []
which I changed to "require" per the current style for checking on the
test requirements.
The following patch fixes issues on PowerPC with the reverse-step and
reverse-next instructions when there are multiple assignment statements
on the same line and when there are multiple function calls on the same
line. The commit log below discusses these issues in further depth.
The discussion included what the correct operation should be for these
commands based on the GDB documentation. The proposed patch at that
time changed how the commands worked on other platforms such as X86 in
a way they no longer matched the documentation.
The issue is the line table contains multiple entries for the same
source line. The patch adds a function to search the line table to
find the address of the first instruction of a line. When setup up the
reverse stepping range, the function is called to make sure the start
of the range corresponds to the address of the first instruction for
the line. This approach was used. When Luis initially developed the
patch, he considered merging the contiguous ranges in the line table
when reading the line tables. He decided it was better to work with the
data directly in the line table rather than creating and using a
modified version of the line table.
The following patch fixes the execution of the reveres-step and
reverse-next commands for both senarios of multiple statements on the
same line for PowerPC and aarch64-linux. Unlike the previous patch, it
does not change the operation of the commands on other platforms, i.e.
X86. The patch adds new test cases for both scenarios to verify they
work correctly.
The patch has been tested on PowerPC, Intel X86 and aarch64-linux with
no new regression failures.
Please let me know if the patch is acceptable for mainline. Thanks.
Carl
---------------------------------------------
Fix reverse stepping multiple contiguous PC ranges over the line table.
There are a couple of scenarios where the GDB reverse-step and reverse-next
commands do not work correctly.
Scenario 1 issue description by Luis Machado:
When running GDB's testsuite on aarch64-linux/Ubuntu 20.04 (also spotted on
the ppc backend), I noticed some failures in gdb.reverse/solib-precsave.exp
and gdb.reverse/solib-reverse.exp.
The failure happens around the following code:
38 b[1] = shr2(17); /* middle part two */
40 b[0] = 6; b[1] = 9; /* generic statement, end part two */
42 shr1 ("message 1\n"); /* shr1 one */
Normal execution:
- step from line 38 will land on line 40.
- step from line 40 will land on line 42.
Reverse execution:
- step from line 42 will land on line 40.
- step from line 40 will land on line 40.
- step from line 40 will land on line 38.
The problem here is that line 40 contains two contiguous but distinct
PC ranges in the line table, like so:
Line 40 - [0x7ec ~ 0x7f4]
Line 40 - [0x7f4 ~ 0x7fc]
The two distinct ranges are generated because GCC started outputting source
column information, which GDB doesn't take into account at the moment.
When stepping forward from line 40, we skip both of these ranges and land on
line 42. When stepping backward from line 42, we stop at the start PC of the
second (or first, going backwards) range of line 40.
Since we've reached ecs->event_thread->control.step_range_start, we stop
stepping backwards.
---------------------------------------------------------
Scenario 2 issue described by Pedro Alves:
The following explanation of the issue was taken from the gdb mailing list
discussion of the withdrawn patch to change the behavior of the reverse-step
and reverse-next commands. Specifically, message from Pedro Alves
<pedro@palves.net> where he demonstrates the issue where you have multiple
function calls on the same source code line:
https://sourceware.org/pipermail/gdb-patches/2023-January/196110.html
The source line looks like:
func1 (); func2 ();
so stepping backwards over that line should always stop at the first
instruction of the line, not in the middle. Let's simplify this.
Here's the full source code of my example:
(gdb) list 1
1 void func1 ()
2 {
3 }
4
5 void func2 ()
6 {
7 }
8
9 int main ()
10 {
11 func1 (); func2 ();
12 }
Compiled with:
$ gcc reverse.c -o reverse -g3 -O0
$ gcc -v
...
gcc version 11.3.0 (Ubuntu 11.3.0-1ubuntu1~22.04)
Now let's debug it with target record, using current gdb git master (f3d8ae90b236),
without your patch:
$ gdb ~/reverse
GNU gdb (GDB) 14.0.50.20230124-git
...
Reading symbols from /home/pedro/reverse...
(gdb) start
Temporary breakpoint 1 at 0x1147: file reverse.c, line 11.
Starting program: /home/pedro/reverse
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
Temporary breakpoint 1, main () at reverse.c:11
11 func1 (); func2 ();
(gdb) record
(gdb) disassemble /s
Dump of assembler code for function main:
reverse.c:
10 {
0x000055555555513f <+0>: endbr64
0x0000555555555143 <+4>: push %rbp
0x0000555555555144 <+5>: mov %rsp,%rbp
11 func1 (); func2 ();
=> 0x0000555555555147 <+8>: mov $0x0,%eax
0x000055555555514c <+13>: call 0x555555555129 <func1>
0x0000555555555151 <+18>: mov $0x0,%eax
0x0000555555555156 <+23>: call 0x555555555134 <func2>
0x000055555555515b <+28>: mov $0x0,%eax
12 }
0x0000555555555160 <+33>: pop %rbp
0x0000555555555161 <+34>: ret
End of assembler dump.
(gdb) n
12 }
So far so good, a "next" stepped over the whole of line 11 and stopped at line 12.
Let's confirm where we are now:
(gdb) disassemble /s
Dump of assembler code for function main:
reverse.c:
10 {
0x000055555555513f <+0>: endbr64
0x0000555555555143 <+4>: push %rbp
0x0000555555555144 <+5>: mov %rsp,%rbp
11 func1 (); func2 ();
0x0000555555555147 <+8>: mov $0x0,%eax
0x000055555555514c <+13>: call 0x555555555129 <func1>
0x0000555555555151 <+18>: mov $0x0,%eax
0x0000555555555156 <+23>: call 0x555555555134 <func2>
0x000055555555515b <+28>: mov $0x0,%eax
12 }
=> 0x0000555555555160 <+33>: pop %rbp
0x0000555555555161 <+34>: ret
End of assembler dump.
Good, we're at the first instruction of line 12.
Now let's undo the "next", with "reverse-next":
(gdb) reverse-next
11 func1 (); func2 ();
Seemingly stopped at line 11. Let's see exactly where:
(gdb) disassemble /s
Dump of assembler code for function main:
reverse.c:
10 {
0x000055555555513f <+0>: endbr64
0x0000555555555143 <+4>: push %rbp
0x0000555555555144 <+5>: mov %rsp,%rbp
11 func1 (); func2 ();
0x0000555555555147 <+8>: mov $0x0,%eax
0x000055555555514c <+13>: call 0x555555555129 <func1>
=> 0x0000555555555151 <+18>: mov $0x0,%eax
0x0000555555555156 <+23>: call 0x555555555134 <func2>
0x000055555555515b <+28>: mov $0x0,%eax
12 }
0x0000555555555160 <+33>: pop %rbp
0x0000555555555161 <+34>: ret
End of assembler dump.
(gdb)
And lo, we stopped in the middle of line 11! That is a bug, we should have
stepped back all the way to the beginning of the line. The "reverse-next"
should have fully undone the prior "next" command.
The above issues were fixed by introducing a new function that looks for
adjacent PC ranges for the same line, until we notice a line change. Then
we take that as the start PC of the range. The new start PC for the range
is used for the control.step_range_start when setting up a step range.
The test case gdb.reverse/map-to-same-line.exp is added to test the fix
for the issues in scenario 1.
The test case gdb.reverse/func-map-to-same-line.exp was added to test the
fix for scenario 2 when the binary was compiled with and without line
table information.
bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28426
Co-authored-by: Luis Machado <luis.machado@arm.com>
Co-authored-by: Carl Love <cel@us.ibm.com>
Reviewed-By: Bruno Larsen <blarsen@redhat.com>
---
gdb/infrun.c | 57 +++++++
gdb/symtab.c | 49 ++++++
gdb/symtab.h | 16 ++
.../gdb.reverse/func-map-to-same-line.c | 36 ++++
.../gdb.reverse/func-map-to-same-line.exp | 140 ++++++++++++++++
gdb/testsuite/gdb.reverse/map-to-same-line.c | 58 +++++++
.../gdb.reverse/map-to-same-line.exp | 156 ++++++++++++++++++
7 files changed, 512 insertions(+)
create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-line.c
create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.c
create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.exp
diff --git a/gdb/infrun.c b/gdb/infrun.c
index efe2c00c489..31cd817c733 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -114,6 +114,9 @@ static struct async_event_handler *infrun_async_inferior_event_token;
Starts off as -1, indicating "never enabled/disabled". */
static int infrun_is_async = -1;
+static CORE_ADDR update_line_range_start (CORE_ADDR pc,
+ struct execution_control_state *ecs);
+
/* See infrun.h. */
void
@@ -6769,6 +6772,25 @@ handle_signal_stop (struct execution_control_state *ecs)
process_event_stop_test (ecs);
}
+CORE_ADDR
+update_line_range_start (CORE_ADDR pc, struct execution_control_state *ecs)
+{
+ /* The line table may have multiple entries for the same source code line.
+ Given the PC, check the line table and return the PC that corresponds
+ to the line table entry for the source line that PC is in. */
+ CORE_ADDR start_line_pc = ecs->event_thread->control.step_range_start;
+ gdb::optional<CORE_ADDR> real_range_start;
+
+ /* Call find_line_range_start to get the smallest address in the
+ linetable for multiple Line X entries in the line table. */
+ real_range_start = find_line_range_start (pc);
+
+ if (real_range_start.has_value ())
+ start_line_pc = *real_range_start;
+
+ return start_line_pc;
+}
+
/* Come here when we've got some debug event / signal we can explain
(IOW, not a random signal), and test whether it should cause a
stop, or whether we should resume the inferior (transparently).
@@ -7570,6 +7592,28 @@ process_event_stop_test (struct execution_control_state *ecs)
if (stop_pc_sal.is_stmt)
{
+ if (execution_direction == EXEC_REVERSE)
+ {
+ /* We are stepping backwards make sure we have reached the
+ beginning of the line. */
+ CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
+ CORE_ADDR start_line_pc
+ = update_line_range_start (stop_pc, ecs);
+
+ if (stop_pc != start_line_pc)
+ {
+ /* Have not reached the beginning of the source code line.
+ Set a step range. Execution should stop in any function
+ calls we execute back into before reaching the beginning
+ of the line. */
+ ecs->event_thread->control.step_range_start = start_line_pc;
+ ecs->event_thread->control.step_range_end = stop_pc;
+ set_step_info (ecs->event_thread, frame, stop_pc_sal);
+ keep_going (ecs);
+ return;
+ }
+ }
+
/* We are at the start of a statement.
So stop. Note that we don't stop if we step into the middle of a
@@ -7632,6 +7676,19 @@ process_event_stop_test (struct execution_control_state *ecs)
set_step_info (ecs->event_thread, frame, stop_pc_sal);
infrun_debug_printf ("keep going");
+
+ if (execution_direction == EXEC_REVERSE)
+ {
+ CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
+
+ /* Make sure the stop_pc is set to the beginning of the line. */
+ if (stop_pc != ecs->event_thread->control.step_range_start)
+ {
+ stop_pc = update_line_range_start (stop_pc, ecs);
+ ecs->event_thread->control.step_range_start = stop_pc;
+ }
+ }
+
keep_going (ecs);
}
diff --git a/gdb/symtab.c b/gdb/symtab.c
index 27611a34ec4..91d35616eb9 100644
--- a/gdb/symtab.c
+++ b/gdb/symtab.c
@@ -3282,6 +3282,55 @@ find_pc_line (CORE_ADDR pc, int notcurrent)
return sal;
}
+/* Compare two symtab_and_line entries. Return true if both have
+ the same line number and the same symtab pointer. That means we
+ are dealing with two entries from the same line and from the same
+ source file.
+
+ Return false otherwise. */
+
+static bool
+sal_line_symtab_matches_p (const symtab_and_line &sal1,
+ const symtab_and_line &sal2)
+{
+ return (sal1.line == sal2.line && sal1.symtab == sal2.symtab);
+}
+
+/* See symtah.h. */
+
+gdb::optional<CORE_ADDR>
+find_line_range_start (CORE_ADDR pc)
+{
+ struct symtab_and_line current_sal = find_pc_line (pc, 0);
+
+ if (current_sal.line == 0)
+ return {};
+
+ struct symtab_and_line prev_sal = find_pc_line (current_sal.pc - 1, 0);
+
+ /* If the previous entry is for a different line, that means we are already
+ at the entry with the start PC for this line. */
+ if (!sal_line_symtab_matches_p (prev_sal, current_sal))
+ return current_sal.pc;
+
+ /* Otherwise, keep looking for entries for the same line but with
+ smaller PC's. */
+ bool done = false;
+ CORE_ADDR prev_pc;
+ while (!done)
+ {
+ prev_pc = prev_sal.pc;
+
+ prev_sal = find_pc_line (prev_pc - 1, 0);
+
+ /* Did we notice a line change? If so, we are done with the search. */
+ if (!sal_line_symtab_matches_p (prev_sal, current_sal))
+ done = true;
+ }
+
+ return prev_pc;
+}
+
/* See symtab.h. */
struct symtab *
diff --git a/gdb/symtab.h b/gdb/symtab.h
index 404d0ab30a8..f54305636da 100644
--- a/gdb/symtab.h
+++ b/gdb/symtab.h
@@ -2346,6 +2346,22 @@ extern struct symtab_and_line find_pc_line (CORE_ADDR, int);
extern struct symtab_and_line find_pc_sect_line (CORE_ADDR,
struct obj_section *, int);
+/* Given PC, and assuming it is part of a range of addresses that is part of a
+ line, go back through the linetable and find the starting PC of that
+ line.
+
+ For example, suppose we have 3 PC ranges for line X:
+
+ Line X - [0x0 - 0x8]
+ Line X - [0x8 - 0x10]
+ Line X - [0x10 - 0x18]
+
+ If we call the function with PC == 0x14, we want to return 0x0, as that is
+ the starting PC of line X, and the ranges are contiguous.
+*/
+
+extern gdb::optional<CORE_ADDR> find_line_range_start (CORE_ADDR pc);
+
/* Wrapper around find_pc_line to just return the symtab. */
extern struct symtab *find_pc_line_symtab (CORE_ADDR);
diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.c b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
new file mode 100644
index 00000000000..da944874e86
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
@@ -0,0 +1,36 @@
+/* Copyright 2008-2023 Free Software Foundation, Inc.
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 3 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program. If not, see <http://www.gnu.org/licenses/>.
+
+ This test is used to test the reverse-step and reverse-next instruction
+ execution for a source line that contains multiple function calls. */
+
+void
+func1 ()
+{
+} /* END FUNC1 */
+
+void
+func2 ()
+{
+} /* END FUNC2 */
+
+int main ()
+{
+ int a, b;
+ a = 1;
+ b = 2;
+ func1 (); func2 ();
+ a = a + b; /* START REVERSE TEST */
+}
diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
new file mode 100644
index 00000000000..89e226b0f84
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
@@ -0,0 +1,140 @@
+# Copyright 2008-2023 Free Software Foundation, Inc.
+
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program. If not, see <http://www.gnu.org/licenses/>. */
+
+# This file is part of the GDB testsuite. It tests reverse stepping.
+# Lots of code borrowed from "step-test.exp".
+
+# This test checks to make sure there is no regression failures for
+# the reverse-next command when stepping back over two functions in
+# the same line.
+
+require supports_reverse
+
+# This test uses the gcc no-column-info command which was added in gcc 7.1.
+
+proc run_tests {} {
+ global srcfile
+ global executable
+
+ runto_main
+ set target_remote [gdb_is_target_remote]
+
+ with_test_prefix "test1" {
+ gdb_test_no_output "record" "turn on process record"
+ }
+
+ # This regression test verifies the reverse-step and reverse-next commands
+ # work properly when executing backwards thru a source line containing
+ # two function calls on the same source line, i.e. func1 (); func2 ();
+ # This test is compiled so the dwarf info not contain the line table
+ # information.
+
+ # Test 1, reverse-next command
+ # Set breakpoint at the line after the function calls.
+ set bp_start_reverse_test [gdb_get_line_number "START REVERSE TEST" \
+ $srcfile]
+ gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
+
+ # Continue to break point for reverse-next test.
+ # Command definition: reverse-next [count]
+ # Run backward to the beginning of the previous line executed in the
+ # current (innermost) stack frame. If the line contains function calls,
+ # they will be “un-executed” without stopping. Starting from the first
+ # line of a function, reverse-next will take you back to the caller of
+ # that function, before the function was called, just as the normal next
+ # command would take you from the last line of a function back to its
+ # return to its caller 2 .
+ gdb_continue_to_breakpoint \
+ "test1: stopped at command reverse-next test start location" \
+ ".*$srcfile:$bp_start_reverse_test\r\n.*"
+
+ # The reverse-next should step all the way back to the beginning of the
+ # line, i.e. at the beginning of the func1 call.
+ gdb_test "reverse-next" ".*func1 \\(\\); func2 \\(\\);.*" \
+ "test1: reverse-next to line with two functions"
+
+ # We should be stopped at the first instruction of the line. A reverse-step
+ # should step back and stop at the beginning of the previous line b = 2,
+ # i.e. not in func1 ().
+ gdb_test "reverse-stepi" ".*b = 2;.*" \
+ "test1: reverse-stepi to previous line b = 2"
+
+
+ # Setup for test 2
+ clean_restart $executable
+ runto_main
+
+ with_test_prefix "test2" {
+ gdb_test_no_output "record" "turn on process record"
+ }
+
+ # Test 2, reverse-step command
+ # Set breakpoint at the line after the function calls.
+ gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
+
+ # Continue to the start of the reverse-step test.
+ # Command definition: reverse-step [count]
+ # Run the program backward until control reaches the start of a
+ # different source line; then stop it, and return control to gdb.
+ # Like the step command, reverse-step will only stop at the beginning
+ # of a source line. It “un-executes” the previously executed source
+ # line. If the previous source line included calls to debuggable
+ # functions, reverse-step will step (backward) into the called function,
+ # stopping at the beginning of the last statement in the called
+ # function (typically a return statement). Also, as with the step
+ # command, if non-debuggable functions are called, reverse-step will
+ # run thru them backward without stopping.
+
+ gdb_continue_to_breakpoint \
+ "test2: stopped at command reverse-step test start location" \
+ ".*$srcfile:$bp_start_reverse_test\r\n.*"
+
+ # The first reverse step should take us call of func2 ().
+ gdb_test "reverse-step" ".*END FUNC2.*" \
+ "test2: reverse-step into func2 "
+
+ # The second reverse step should take us into func1 ().
+ gdb_test "reverse-step" ".*END FUNC1.*" \
+ "test2: reverse-step into func1 "
+
+ # The third reverse step should take us call of func1 ().
+ gdb_test "reverse-step" ".*func1 \\(\\); func2 \\(\\);.*" \
+ "test2: reverse-step to line func1(); func2(), at call for func1 "
+
+ # We should be stopped at the first instruction of the line. A reverse
+ # stepi should take us to b = 2 ().
+ gdb_test "reverse-stepi" ".*b = 2;.*" \
+ "test2: reverse-stepi to line b = 2 "
+}
+
+set srcfile func-map-to-same-line.c
+set executable func-map-to-same-line
+
+# test with and without gcc column info enabled
+foreach_with_prefix with_column_info {yes no} {
+ if {$with_column_info == "yes"} {
+ set options [list debug column-info]
+ } else {
+ set options [list debug no-column-info]
+ }
+
+ if {[build_executable "failed to prepare" $executable $srcfile \
+ $options] == -1} {
+ return -1
+ }
+
+ clean_restart $executable
+ run_tests
+}
diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.c b/gdb/testsuite/gdb.reverse/map-to-same-line.c
new file mode 100644
index 00000000000..f20d778f40e
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/map-to-same-line.c
@@ -0,0 +1,58 @@
+/* Copyright 2008-2023 Free Software Foundation, Inc.
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 3 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program. If not, see <http://www.gnu.org/licenses/ >. */
+
+/* The purpose of this test is to create a DWARF line table that contains two
+ or more entries for the same line. When stepping (forwards or backwards),
+ GDB should step over the entire line and not just a particular entry in the
+ line table. */
+
+int
+main ()
+{ /* TAG: main prologue */
+ asm ("main_label: .globl main_label");
+ int i = 1, j = 2, k;
+ float f1 = 2.0, f2 = 4.1, f3;
+ const char *str_1 = "foo", *str_2 = "bar", *str_3;
+
+ asm ("line1: .globl line1");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 1 */
+
+ asm ("line2: .globl line2");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 2 */
+
+ asm ("line3: .globl line3");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 3 */
+
+ asm ("line4: .globl line4");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 4 */
+
+ asm ("line5: .globl line5");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 5 */
+
+ asm ("line6: .globl line6");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 6 */
+
+ asm ("line7: .globl line7");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 7 */
+
+ asm ("line8: .globl line8");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 8 */
+
+ asm ("main_return: .globl main_return");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: main return */
+
+ asm ("end_of_sequence: .globl end_of_sequence");
+ return 0; /* TAG: main return */
+}
diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.exp b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
new file mode 100644
index 00000000000..16a359d90ec
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
@@ -0,0 +1,156 @@
+# Copyright 2008-2023 Free Software Foundation, Inc.
+
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program. If not, see <http://www.gnu.org/licenses/ >.
+
+# When stepping (forwards or backwards), GDB should step over the entire line
+# and not just a particular entry in the line table. This test was added to
+# verify the find_line_range_start function properly sets the step range for a
+# line that consists of multiple statements, i.e. multiple entries in the line
+# table. This test creates a DWARF line table that contains two entries for
+# the same line to do the needed testing.
+
+# This test can only be run on targets which support DWARF-2 and use gas.
+load_lib dwarf.exp
+require dwarf2_support
+
+# The DWARF assembler requires the gcc compiler.
+require is_c_compiler_gcc
+
+# This test suitable only for process that can do reverse execution
+require supports_reverse
+
+standard_testfile .c .S
+
+if { [prepare_for_testing "failed to prepare" ${testfile} ${srcfile}] } {
+ return -1
+}
+
+set asm_file [standard_output_file $srcfile2]
+Dwarf::assemble $asm_file {
+ global srcdir subdir srcfile
+ declare_labels integer_label L
+
+ # Find start address and length of program
+ lassign [function_range main [list ${srcdir}/${subdir}/$srcfile]] \
+ main_start main_len
+ set main_end "$main_start + $main_len"
+
+ cu {} {
+ compile_unit {
+ {language @DW_LANG_C}
+ {name map-to-same-line.c}
+ {stmt_list $L DW_FORM_sec_offset}
+ {low_pc 0 addr}
+ } {
+ subprogram {
+ {external 1 flag}
+ {name main}
+ {low_pc $main_start addr}
+ {high_pc $main_len DW_FORM_data4}
+ }
+ }
+ }
+
+ lines {version 2 default_is_stmt 1} L {
+ include_dir "${srcdir}/${subdir}"
+ file_name "$srcfile" 1
+
+ # Generate the line table program with distinct source lines being
+ # mapped to the same line entry. Line 1, 5 and 8 contain 1 statement
+ # each. Line 2 contains 2 statements. Line 3 contains 3 statements.
+ program {
+ DW_LNE_set_address $main_start
+ line [gdb_get_line_number "TAG: main prologue"]
+ DW_LNS_copy
+ DW_LNE_set_address line1
+ line [gdb_get_line_number "TAG: line 1" ]
+ DW_LNS_copy
+ DW_LNE_set_address line2
+ line [gdb_get_line_number "TAG: line 2" ]
+ DW_LNS_copy
+ DW_LNE_set_address line3
+ line [gdb_get_line_number "TAG: line 2" ]
+ DW_LNS_copy
+ DW_LNE_set_address line4
+ line [gdb_get_line_number "TAG: line 3" ]
+ DW_LNS_copy
+ DW_LNE_set_address line5
+ line [gdb_get_line_number "TAG: line 3" ]
+ DW_LNS_copy
+ DW_LNE_set_address line6
+ line [gdb_get_line_number "TAG: line 3" ]
+ DW_LNS_copy
+ DW_LNE_set_address line7
+ line [gdb_get_line_number "TAG: line 5" ]
+ DW_LNS_copy
+ DW_LNE_set_address line8
+ line [gdb_get_line_number "TAG: line 8" ]
+ DW_LNS_copy
+ DW_LNE_set_address main_return
+ line [gdb_get_line_number "TAG: main return"]
+ DW_LNS_copy
+ DW_LNE_set_address end_of_sequence
+ DW_LNE_end_sequence
+ }
+ }
+}
+
+if { [prepare_for_testing "failed to prepare" ${testfile} \
+ [list $srcfile $asm_file] {nodebug} ] } {
+ return -1
+}
+
+runto_main
+
+# Print the line table
+gdb_test_multiple "maint info line-table ${testfile}" "" {
+ -re "\r\n$decimal\[ \t\]+$decimal\[ \t\]+($hex)\[ \t\]+Y\[^\r\n\]*" {
+ lappend is_stmt $expect_out(1,string)
+ exp_continue
+ }
+ -re -wrap "" {
+ }
+}
+
+# Do the reverse-step test
+gdb_test_no_output "record" "turn on process record"
+
+set bp_main_return [gdb_get_line_number "TAG: main return" $srcfile]
+gdb_breakpoint $srcfile:$bp_main_return
+gdb_continue_to_breakpoint "run to end of main, reverse-step test" ".*$srcfile:$bp_main_return.*"
+gdb_test "display \$pc" ".*pc =.*" "display pc, reverse-step test"
+
+# At this point, GDB has already recorded the execution up until the return
+# statement. Reverse-step and test if GDB transitions between lines in the
+# expected order. It should reverse-step across lines 8, 5, 3, 2 and 1.
+foreach line {8 5 3 2 1} {
+ gdb_test "reverse-step" ".*TAG: line $line.*" "reverse step to line $line"
+}
+
+## Clean restart, test reverse-next command
+clean_restart ${testfile}
+runto_main
+gdb_test_no_output "record" "turn on process record, reverst-next test"
+
+set bp_main_return [gdb_get_line_number "TAG: main return" $srcfile]
+gdb_breakpoint $srcfile:$bp_main_return
+gdb_continue_to_breakpoint "run to end of main, reverse-next test" ".*$srcfile:$bp_main_return.*"
+gdb_test "display \$pc" ".*pc =.*" "display pc, reverse-next test"
+
+# At this point, GDB has already recorded the execution up until the return
+# statement. Reverse-next and test if GDB transitions between lines in the
+# expected order. It should reverse-next across lines 8, 5, 3, 2 and 1.
+foreach line {8 5 3 2 1} {
+ gdb_test "reverse-next" ".*TAG: line $line.*" "reverse next to line $line"
+}
--
2.37.2
^ permalink raw reply [flat|nested] 40+ messages in thread
* Re: [PATCH 2/2 v5] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-05-16 22:54 ` [PATCH 2/2 v5] " Carl Love
@ 2023-05-25 15:08 ` Carl Love
2023-06-08 16:36 ` Carl Love
2023-06-19 17:58 ` Simon Marchi
1 sibling, 1 reply; 40+ messages in thread
From: Carl Love @ 2023-05-25 15:08 UTC (permalink / raw)
To: Simon Marchi, Bruno Larsen, gdb-patches, UlrichWeigand, pedro
Cc: luis.machado, cel
Ping.
Simon:
I know Simon said he was busy but wanted to look more at this patch.
Just wondering if you have found any time to review the patch further.
Thanks
Carl
On Tue, 2023-05-16 at 15:54 -0700, Carl Love wrote:
> Bruno, Simon, GDB maintainers:
>
> Version 5, changed comments in test case func-map-to-same-line.c.
> Patch 1/2 implemented the new options for gdb_compile. Updated the
> call to proc run_tests to use the new gdb_compile options in a
> foreach_with_prefix loop.
>
> Version 4, additional fixes for gcc version check, wrap function
> calls
> using "with_test_prefix", move load_lib dwarf.exe. Fixed typo noted
> by
> Luis.
>
> Version 3, added the gcc version check as discussed further from
> version 2 of the patch. Also updated the tests to check for
> supporting
> reverse execution rather than requiring recording. I also noticed
> there were a couple more instances of a requirement check, i.e. if []
> which I changed to "require" per the current style for checking on
> the
> test requirements.
>
>
> The following patch fixes issues on PowerPC with the reverse-step and
> reverse-next instructions when there are multiple assignment
> statements
> on the same line and when there are multiple function calls on the
> same
> line. The commit log below discusses these issues in further depth.
> The discussion included what the correct operation should be for
> these
> commands based on the GDB documentation. The proposed patch at that
> time changed how the commands worked on other platforms such as X86
> in
> a way they no longer matched the documentation.
>
> The issue is the line table contains multiple entries for the same
> source line. The patch adds a function to search the line table to
> find the address of the first instruction of a line. When setup up
> the
> reverse stepping range, the function is called to make sure the start
> of the range corresponds to the address of the first instruction for
> the line. This approach was used. When Luis initially developed the
> patch, he considered merging the contiguous ranges in the line table
> when reading the line tables. He decided it was better to work with
> the
> data directly in the line table rather than creating and using a
> modified version of the line table.
>
> The following patch fixes the execution of the reveres-step and
> reverse-next commands for both senarios of multiple statements on the
> same line for PowerPC and aarch64-linux. Unlike the previous patch,
> it
> does not change the operation of the commands on other platforms,
> i.e.
> X86. The patch adds new test cases for both scenarios to verify they
> work correctly.
>
> The patch has been tested on PowerPC, Intel X86 and aarch64-linux
> with
> no new regression failures.
>
> Please let me know if the patch is acceptable for mainline. Thanks.
>
> Carl
>
> ---------------------------------------------
> Fix reverse stepping multiple contiguous PC ranges over the line
> table.
>
> There are a couple of scenarios where the GDB reverse-step and
> reverse-next
> commands do not work correctly.
>
> Scenario 1 issue description by Luis Machado:
>
> When running GDB's testsuite on aarch64-linux/Ubuntu 20.04 (also
> spotted on
> the ppc backend), I noticed some failures in gdb.reverse/solib-
> precsave.exp
> and gdb.reverse/solib-reverse.exp.
>
> The failure happens around the following code:
>
> 38 b[1] = shr2(17); /* middle part two */
> 40 b[0] = 6; b[1] = 9; /* generic statement, end part two */
> 42 shr1 ("message 1\n"); /* shr1 one */
>
> Normal execution:
>
> - step from line 38 will land on line 40.
> - step from line 40 will land on line 42.
>
> Reverse execution:
> - step from line 42 will land on line 40.
> - step from line 40 will land on line 40.
> - step from line 40 will land on line 38.
>
> The problem here is that line 40 contains two contiguous but distinct
> PC ranges in the line table, like so:
>
> Line 40 - [0x7ec ~ 0x7f4]
> Line 40 - [0x7f4 ~ 0x7fc]
>
> The two distinct ranges are generated because GCC started outputting
> source
> column information, which GDB doesn't take into account at the
> moment.
>
> When stepping forward from line 40, we skip both of these ranges and
> land on
> line 42. When stepping backward from line 42, we stop at the start PC
> of the
> second (or first, going backwards) range of line 40.
>
> Since we've reached ecs->event_thread->control.step_range_start, we
> stop
> stepping backwards.
>
> ---------------------------------------------------------
>
> Scenario 2 issue described by Pedro Alves:
>
> The following explanation of the issue was taken from the gdb mailing
> list
> discussion of the withdrawn patch to change the behavior of the
> reverse-step
> and reverse-next commands. Specifically, message from Pedro Alves
> <pedro@palves.net> where he demonstrates the issue where you have
> multiple
> function calls on the same source code line:
>
> https://sourceware.org/pipermail/gdb-patches/2023-January/196110.html
>
> The source line looks like:
>
> func1 (); func2 ();
>
> so stepping backwards over that line should always stop at the first
> instruction of the line, not in the middle. Let's simplify this.
>
> Here's the full source code of my example:
>
> (gdb) list 1
> 1 void func1 ()
> 2 {
> 3 }
> 4
> 5 void func2 ()
> 6 {
> 7 }
> 8
> 9 int main ()
> 10 {
> 11 func1 (); func2 ();
> 12 }
>
> Compiled with:
>
> $ gcc reverse.c -o reverse -g3 -O0
> $ gcc -v
> ...
> gcc version 11.3.0 (Ubuntu 11.3.0-1ubuntu1~22.04)
>
> Now let's debug it with target record, using current gdb git master
> (f3d8ae90b236),
> without your patch:
>
> $ gdb ~/reverse
> GNU gdb (GDB) 14.0.50.20230124-git
> ...
> Reading symbols from /home/pedro/reverse...
> (gdb) start
> Temporary breakpoint 1 at 0x1147: file reverse.c, line 11.
> Starting program: /home/pedro/reverse
> [Thread debugging using libthread_db enabled]
> Using host libthread_db library "/lib/x86_64-linux-
> gnu/libthread_db.so.1".
>
> Temporary breakpoint 1, main () at reverse.c:11
> 11 func1 (); func2 ();
> (gdb) record
>
> (gdb) disassemble /s
> Dump of assembler code for function main:
> reverse.c:
> 10 {
> 0x000055555555513f <+0>: endbr64
> 0x0000555555555143 <+4>: push %rbp
> 0x0000555555555144 <+5>: mov %rsp,%rbp
>
> 11 func1 (); func2 ();
> => 0x0000555555555147 <+8>: mov $0x0,%eax
> 0x000055555555514c <+13>: call 0x555555555129 <func1>
> 0x0000555555555151 <+18>: mov $0x0,%eax
> 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> 0x000055555555515b <+28>: mov $0x0,%eax
>
> 12 }
> 0x0000555555555160 <+33>: pop %rbp
> 0x0000555555555161 <+34>: ret
> End of assembler dump.
>
> (gdb) n
> 12 }
>
> So far so good, a "next" stepped over the whole of line 11 and
> stopped at line 12.
>
> Let's confirm where we are now:
>
> (gdb) disassemble /s
> Dump of assembler code for function main:
> reverse.c:
> 10 {
> 0x000055555555513f <+0>: endbr64
> 0x0000555555555143 <+4>: push %rbp
> 0x0000555555555144 <+5>: mov %rsp,%rbp
>
> 11 func1 (); func2 ();
> 0x0000555555555147 <+8>: mov $0x0,%eax
> 0x000055555555514c <+13>: call 0x555555555129 <func1>
> 0x0000555555555151 <+18>: mov $0x0,%eax
> 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> 0x000055555555515b <+28>: mov $0x0,%eax
>
> 12 }
> => 0x0000555555555160 <+33>: pop %rbp
> 0x0000555555555161 <+34>: ret
> End of assembler dump.
>
> Good, we're at the first instruction of line 12.
>
> Now let's undo the "next", with "reverse-next":
>
> (gdb) reverse-next
> 11 func1 (); func2 ();
>
> Seemingly stopped at line 11. Let's see exactly where:
>
> (gdb) disassemble /s
> Dump of assembler code for function main:
> reverse.c:
> 10 {
> 0x000055555555513f <+0>: endbr64
> 0x0000555555555143 <+4>: push %rbp
> 0x0000555555555144 <+5>: mov %rsp,%rbp
>
> 11 func1 (); func2 ();
> 0x0000555555555147 <+8>: mov $0x0,%eax
> 0x000055555555514c <+13>: call 0x555555555129 <func1>
> => 0x0000555555555151 <+18>: mov $0x0,%eax
> 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> 0x000055555555515b <+28>: mov $0x0,%eax
>
> 12 }
> 0x0000555555555160 <+33>: pop %rbp
> 0x0000555555555161 <+34>: ret
> End of assembler dump.
> (gdb)
>
> And lo, we stopped in the middle of line 11! That is a bug, we
> should have
> stepped back all the way to the beginning of the line. The "reverse-
> next"
> should have fully undone the prior "next" command.
>
> The above issues were fixed by introducing a new function that looks
> for
> adjacent PC ranges for the same line, until we notice a line change.
> Then
> we take that as the start PC of the range. The new start PC for the
> range
> is used for the control.step_range_start when setting up a step
> range.
>
> The test case gdb.reverse/map-to-same-line.exp is added to test the
> fix
> for the issues in scenario 1.
>
> The test case gdb.reverse/func-map-to-same-line.exp was added to test
> the
> fix for scenario 2 when the binary was compiled with and without line
> table information.
>
> bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28426
>
> Co-authored-by: Luis Machado <luis.machado@arm.com>
> Co-authored-by: Carl Love <cel@us.ibm.com>
> Reviewed-By: Bruno Larsen <blarsen@redhat.com>
> ---
> gdb/infrun.c | 57 +++++++
> gdb/symtab.c | 49 ++++++
> gdb/symtab.h | 16 ++
> .../gdb.reverse/func-map-to-same-line.c | 36 ++++
> .../gdb.reverse/func-map-to-same-line.exp | 140 ++++++++++++++++
> gdb/testsuite/gdb.reverse/map-to-same-line.c | 58 +++++++
> .../gdb.reverse/map-to-same-line.exp | 156
> ++++++++++++++++++
> 7 files changed, 512 insertions(+)
> create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-
> line.exp
> create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.c
> create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.exp
>
> diff --git a/gdb/infrun.c b/gdb/infrun.c
> index efe2c00c489..31cd817c733 100644
> --- a/gdb/infrun.c
> +++ b/gdb/infrun.c
> @@ -114,6 +114,9 @@ static struct async_event_handler
> *infrun_async_inferior_event_token;
> Starts off as -1, indicating "never enabled/disabled". */
> static int infrun_is_async = -1;
>
> +static CORE_ADDR update_line_range_start (CORE_ADDR pc,
> + struct
> execution_control_state *ecs);
> +
> /* See infrun.h. */
>
> void
> @@ -6769,6 +6772,25 @@ handle_signal_stop (struct
> execution_control_state *ecs)
> process_event_stop_test (ecs);
> }
>
> +CORE_ADDR
> +update_line_range_start (CORE_ADDR pc, struct
> execution_control_state *ecs)
> +{
> + /* The line table may have multiple entries for the same source
> code line.
> + Given the PC, check the line table and return the PC that
> corresponds
> + to the line table entry for the source line that PC is in. */
> + CORE_ADDR start_line_pc = ecs->event_thread-
> >control.step_range_start;
> + gdb::optional<CORE_ADDR> real_range_start;
> +
> + /* Call find_line_range_start to get the smallest address in the
> + linetable for multiple Line X entries in the line table. */
> + real_range_start = find_line_range_start (pc);
> +
> + if (real_range_start.has_value ())
> + start_line_pc = *real_range_start;
> +
> + return start_line_pc;
> +}
> +
> /* Come here when we've got some debug event / signal we can explain
> (IOW, not a random signal), and test whether it should cause a
> stop, or whether we should resume the inferior (transparently).
> @@ -7570,6 +7592,28 @@ process_event_stop_test (struct
> execution_control_state *ecs)
>
> if (stop_pc_sal.is_stmt)
> {
> + if (execution_direction == EXEC_REVERSE)
> + {
> + /* We are stepping backwards make sure we have reached
> the
> + beginning of the line. */
> + CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
> + CORE_ADDR start_line_pc
> + = update_line_range_start (stop_pc, ecs);
> +
> + if (stop_pc != start_line_pc)
> + {
> + /* Have not reached the beginning of the source code
> line.
> + Set a step range. Execution should stop in any
> function
> + calls we execute back into before reaching the
> beginning
> + of the line. */
> + ecs->event_thread->control.step_range_start =
> start_line_pc;
> + ecs->event_thread->control.step_range_end = stop_pc;
> + set_step_info (ecs->event_thread, frame,
> stop_pc_sal);
> + keep_going (ecs);
> + return;
> + }
> + }
> +
> /* We are at the start of a statement.
>
> So stop. Note that we don't stop if we step into the
> middle of a
> @@ -7632,6 +7676,19 @@ process_event_stop_test (struct
> execution_control_state *ecs)
> set_step_info (ecs->event_thread, frame, stop_pc_sal);
>
> infrun_debug_printf ("keep going");
> +
> + if (execution_direction == EXEC_REVERSE)
> + {
> + CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
> +
> + /* Make sure the stop_pc is set to the beginning of the
> line. */
> + if (stop_pc != ecs->event_thread->control.step_range_start)
> + {
> + stop_pc = update_line_range_start (stop_pc, ecs);
> + ecs->event_thread->control.step_range_start = stop_pc;
> + }
> + }
> +
> keep_going (ecs);
> }
>
> diff --git a/gdb/symtab.c b/gdb/symtab.c
> index 27611a34ec4..91d35616eb9 100644
> --- a/gdb/symtab.c
> +++ b/gdb/symtab.c
> @@ -3282,6 +3282,55 @@ find_pc_line (CORE_ADDR pc, int notcurrent)
> return sal;
> }
>
> +/* Compare two symtab_and_line entries. Return true if both have
> + the same line number and the same symtab pointer. That means we
> + are dealing with two entries from the same line and from the same
> + source file.
> +
> + Return false otherwise. */
> +
> +static bool
> +sal_line_symtab_matches_p (const symtab_and_line &sal1,
> + const symtab_and_line &sal2)
> +{
> + return (sal1.line == sal2.line && sal1.symtab == sal2.symtab);
> +}
> +
> +/* See symtah.h. */
> +
> +gdb::optional<CORE_ADDR>
> +find_line_range_start (CORE_ADDR pc)
> +{
> + struct symtab_and_line current_sal = find_pc_line (pc, 0);
> +
> + if (current_sal.line == 0)
> + return {};
> +
> + struct symtab_and_line prev_sal = find_pc_line (current_sal.pc -
> 1, 0);
> +
> + /* If the previous entry is for a different line, that means we
> are already
> + at the entry with the start PC for this line. */
> + if (!sal_line_symtab_matches_p (prev_sal, current_sal))
> + return current_sal.pc;
> +
> + /* Otherwise, keep looking for entries for the same line but with
> + smaller PC's. */
> + bool done = false;
> + CORE_ADDR prev_pc;
> + while (!done)
> + {
> + prev_pc = prev_sal.pc;
> +
> + prev_sal = find_pc_line (prev_pc - 1, 0);
> +
> + /* Did we notice a line change? If so, we are done with the
> search. */
> + if (!sal_line_symtab_matches_p (prev_sal, current_sal))
> + done = true;
> + }
> +
> + return prev_pc;
> +}
> +
> /* See symtab.h. */
>
> struct symtab *
> diff --git a/gdb/symtab.h b/gdb/symtab.h
> index 404d0ab30a8..f54305636da 100644
> --- a/gdb/symtab.h
> +++ b/gdb/symtab.h
> @@ -2346,6 +2346,22 @@ extern struct symtab_and_line find_pc_line
> (CORE_ADDR, int);
> extern struct symtab_and_line find_pc_sect_line (CORE_ADDR,
> struct obj_section *,
> int);
>
> +/* Given PC, and assuming it is part of a range of addresses that is
> part of a
> + line, go back through the linetable and find the starting PC of
> that
> + line.
> +
> + For example, suppose we have 3 PC ranges for line X:
> +
> + Line X - [0x0 - 0x8]
> + Line X - [0x8 - 0x10]
> + Line X - [0x10 - 0x18]
> +
> + If we call the function with PC == 0x14, we want to return 0x0,
> as that is
> + the starting PC of line X, and the ranges are contiguous.
> +*/
> +
> +extern gdb::optional<CORE_ADDR> find_line_range_start (CORE_ADDR
> pc);
> +
> /* Wrapper around find_pc_line to just return the symtab. */
>
> extern struct symtab *find_pc_line_symtab (CORE_ADDR);
> diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> new file mode 100644
> index 00000000000..da944874e86
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> @@ -0,0 +1,36 @@
> +/* Copyright 2008-2023 Free Software Foundation, Inc.
> +
> + This program is free software; you can redistribute it and/or
> modify
> + it under the terms of the GNU General Public License as published
> by
> + the Free Software Foundation; either version 3 of the License, or
> + (at your option) any later version.
> +
> + This program is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> + GNU General Public License for more details.
> +
> + You should have received a copy of the GNU General Public License
> + along with this program. If not, see <
> http://www.gnu.org/licenses/>.
> +
> + This test is used to test the reverse-step and reverse-next
> instruction
> + execution for a source line that contains multiple function
> calls. */
> +
> +void
> +func1 ()
> +{
> +} /* END FUNC1 */
> +
> +void
> +func2 ()
> +{
> +} /* END FUNC2 */
> +
> +int main ()
> +{
> + int a, b;
> + a = 1;
> + b = 2;
> + func1 (); func2 ();
> + a = a + b; /* START REVERSE TEST */
> +}
> diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> new file mode 100644
> index 00000000000..89e226b0f84
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> @@ -0,0 +1,140 @@
> +# Copyright 2008-2023 Free Software Foundation, Inc.
> +
> +# This program is free software; you can redistribute it and/or
> modify
> +# it under the terms of the GNU General Public License as published
> by
> +# the Free Software Foundation; either version 3 of the License, or
> +# (at your option) any later version.
> +#
> +# This program is distributed in the hope that it will be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> +# GNU General Public License for more details.
> +#
> +# You should have received a copy of the GNU General Public License
> +# along with this program. If not, see <
> http://www.gnu.org/licenses/>. */
> +
> +# This file is part of the GDB testsuite. It tests reverse
> stepping.
> +# Lots of code borrowed from "step-test.exp".
> +
> +# This test checks to make sure there is no regression failures for
> +# the reverse-next command when stepping back over two functions in
> +# the same line.
> +
> +require supports_reverse
> +
> +# This test uses the gcc no-column-info command which was added in
> gcc 7.1.
> +
> +proc run_tests {} {
> + global srcfile
> + global executable
> +
> + runto_main
> + set target_remote [gdb_is_target_remote]
> +
> + with_test_prefix "test1" {
> + gdb_test_no_output "record" "turn on process record"
> + }
> +
> + # This regression test verifies the reverse-step and reverse-
> next commands
> + # work properly when executing backwards thru a source line
> containing
> + # two function calls on the same source line, i.e. func1 ();
> func2 ();
> + # This test is compiled so the dwarf info not contain the line
> table
> + # information.
> +
> + # Test 1, reverse-next command
> + # Set breakpoint at the line after the function calls.
> + set bp_start_reverse_test [gdb_get_line_number "START REVERSE
> TEST" \
> + $srcfile]
> + gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> +
> + # Continue to break point for reverse-next test.
> + # Command definition: reverse-next [count]
> + # Run backward to the beginning of the previous line executed
> in the
> + # current (innermost) stack frame. If the line contains
> function calls,
> + # they will be “un-executed” without stopping. Starting from
> the first
> + # line of a function, reverse-next will take you back to the
> caller of
> + # that function, before the function was called, just as the
> normal next
> + # command would take you from the last line of a function back
> to its
> + # return to its caller 2 .
> + gdb_continue_to_breakpoint \
> + "test1: stopped at command reverse-next test start location" \
> + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> +
> + # The reverse-next should step all the way back to the beginning
> of the
> + # line, i.e. at the beginning of the func1 call.
> + gdb_test "reverse-next" ".*func1 \\(\\); func2 \\(\\);.*" \
> + "test1: reverse-next to line with two functions"
> +
> + # We should be stopped at the first instruction of the line. A
> reverse-step
> + # should step back and stop at the beginning of the previous
> line b = 2,
> + # i.e. not in func1 ().
> + gdb_test "reverse-stepi" ".*b = 2;.*" \
> + "test1: reverse-stepi to previous line b = 2"
> +
> +
> + # Setup for test 2
> + clean_restart $executable
> + runto_main
> +
> + with_test_prefix "test2" {
> + gdb_test_no_output "record" "turn on process record"
> + }
> +
> + # Test 2, reverse-step command
> + # Set breakpoint at the line after the function calls.
> + gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> +
> + # Continue to the start of the reverse-step test.
> + # Command definition: reverse-step [count]
> + # Run the program backward until control reaches the start of
> a
> + # different source line; then stop it, and return control to
> gdb.
> + # Like the step command, reverse-step will only stop at the
> beginning
> + # of a source line. It “un-executes” the previously executed
> source
> + # line. If the previous source line included calls to
> debuggable
> + # functions, reverse-step will step (backward) into the
> called function,
> + # stopping at the beginning of the last statement in the
> called
> + # function (typically a return statement). Also, as with the
> step
> + # command, if non-debuggable functions are called, reverse-
> step will
> + # run thru them backward without stopping.
> +
> + gdb_continue_to_breakpoint \
> + "test2: stopped at command reverse-step test start location" \
> + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> +
> + # The first reverse step should take us call of func2 ().
> + gdb_test "reverse-step" ".*END FUNC2.*" \
> + "test2: reverse-step into func2 "
> +
> + # The second reverse step should take us into func1 ().
> + gdb_test "reverse-step" ".*END FUNC1.*" \
> + "test2: reverse-step into func1 "
> +
> + # The third reverse step should take us call of func1 ().
> + gdb_test "reverse-step" ".*func1 \\(\\); func2 \\(\\);.*" \
> + "test2: reverse-step to line func1(); func2(), at call for
> func1 "
> +
> + # We should be stopped at the first instruction of the line. A
> reverse
> + # stepi should take us to b = 2 ().
> + gdb_test "reverse-stepi" ".*b = 2;.*" \
> + "test2: reverse-stepi to line b = 2 "
> +}
> +
> +set srcfile func-map-to-same-line.c
> +set executable func-map-to-same-line
> +
> +# test with and without gcc column info enabled
> +foreach_with_prefix with_column_info {yes no} {
> + if {$with_column_info == "yes"} {
> + set options [list debug column-info]
> + } else {
> + set options [list debug no-column-info]
> + }
> +
> + if {[build_executable "failed to prepare" $executable $srcfile \
> + $options] == -1} {
> + return -1
> + }
> +
> + clean_restart $executable
> + run_tests
> +}
> diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.c
> b/gdb/testsuite/gdb.reverse/map-to-same-line.c
> new file mode 100644
> index 00000000000..f20d778f40e
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/map-to-same-line.c
> @@ -0,0 +1,58 @@
> +/* Copyright 2008-2023 Free Software Foundation, Inc.
> +
> + This program is free software; you can redistribute it and/or
> modify
> + it under the terms of the GNU General Public License as published
> by
> + the Free Software Foundation; either version 3 of the License, or
> + (at your option) any later version.
> +
> + This program is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> + GNU General Public License for more details.
> +
> + You should have received a copy of the GNU General Public License
> + along with this program. If not, see <
> http://www.gnu.org/licenses/
> >. */
> +
> +/* The purpose of this test is to create a DWARF line table that
> contains two
> + or more entries for the same line. When stepping (forwards or
> backwards),
> + GDB should step over the entire line and not just a particular
> entry in the
> + line table. */
> +
> +int
> +main ()
> +{ /* TAG: main prologue */
> + asm ("main_label: .globl main_label");
> + int i = 1, j = 2, k;
> + float f1 = 2.0, f2 = 4.1, f3;
> + const char *str_1 = "foo", *str_2 = "bar", *str_3;
> +
> + asm ("line1: .globl line1");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 1 */
> +
> + asm ("line2: .globl line2");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 2 */
> +
> + asm ("line3: .globl line3");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 3 */
> +
> + asm ("line4: .globl line4");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 4 */
> +
> + asm ("line5: .globl line5");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 5 */
> +
> + asm ("line6: .globl line6");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 6 */
> +
> + asm ("line7: .globl line7");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 7 */
> +
> + asm ("line8: .globl line8");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 8 */
> +
> + asm ("main_return: .globl main_return");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: main return */
> +
> + asm ("end_of_sequence: .globl end_of_sequence");
> + return 0; /* TAG: main return */
> +}
> diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> new file mode 100644
> index 00000000000..16a359d90ec
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> @@ -0,0 +1,156 @@
> +# Copyright 2008-2023 Free Software Foundation, Inc.
> +
> +# This program is free software; you can redistribute it and/or
> modify
> +# it under the terms of the GNU General Public License as published
> by
> +# the Free Software Foundation; either version 3 of the License, or
> +# (at your option) any later version.
> +#
> +# This program is distributed in the hope that it will be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> +# GNU General Public License for more details.
> +#
> +# You should have received a copy of the GNU General Public License
> +# along with this program. If not, see <
> http://www.gnu.org/licenses/
> >.
> +
> +# When stepping (forwards or backwards), GDB should step over the
> entire line
> +# and not just a particular entry in the line table. This test was
> added to
> +# verify the find_line_range_start function properly sets the step
> range for a
> +# line that consists of multiple statements, i.e. multiple entries
> in the line
> +# table. This test creates a DWARF line table that contains two
> entries for
> +# the same line to do the needed testing.
> +
> +# This test can only be run on targets which support DWARF-2 and use
> gas.
> +load_lib dwarf.exp
> +require dwarf2_support
> +
> +# The DWARF assembler requires the gcc compiler.
> +require is_c_compiler_gcc
> +
> +# This test suitable only for process that can do reverse execution
> +require supports_reverse
> +
> +standard_testfile .c .S
> +
> +if { [prepare_for_testing "failed to prepare" ${testfile}
> ${srcfile}] } {
> + return -1
> +}
> +
> +set asm_file [standard_output_file $srcfile2]
> +Dwarf::assemble $asm_file {
> + global srcdir subdir srcfile
> + declare_labels integer_label L
> +
> + # Find start address and length of program
> + lassign [function_range main [list
> ${srcdir}/${subdir}/$srcfile]] \
> + main_start main_len
> + set main_end "$main_start + $main_len"
> +
> + cu {} {
> + compile_unit {
> + {language @DW_LANG_C}
> + {name map-to-same-line.c}
> + {stmt_list $L DW_FORM_sec_offset}
> + {low_pc 0 addr}
> + } {
> + subprogram {
> + {external 1 flag}
> + {name main}
> + {low_pc $main_start addr}
> + {high_pc $main_len DW_FORM_data4}
> + }
> + }
> + }
> +
> + lines {version 2 default_is_stmt 1} L {
> + include_dir "${srcdir}/${subdir}"
> + file_name "$srcfile" 1
> +
> + # Generate the line table program with distinct source lines
> being
> + # mapped to the same line entry. Line 1, 5 and 8 contain 1
> statement
> + # each. Line 2 contains 2 statements. Line 3 contains 3
> statements.
> + program {
> + DW_LNE_set_address $main_start
> + line [gdb_get_line_number "TAG: main prologue"]
> + DW_LNS_copy
> + DW_LNE_set_address line1
> + line [gdb_get_line_number "TAG: line 1" ]
> + DW_LNS_copy
> + DW_LNE_set_address line2
> + line [gdb_get_line_number "TAG: line 2" ]
> + DW_LNS_copy
> + DW_LNE_set_address line3
> + line [gdb_get_line_number "TAG: line 2" ]
> + DW_LNS_copy
> + DW_LNE_set_address line4
> + line [gdb_get_line_number "TAG: line 3" ]
> + DW_LNS_copy
> + DW_LNE_set_address line5
> + line [gdb_get_line_number "TAG: line 3" ]
> + DW_LNS_copy
> + DW_LNE_set_address line6
> + line [gdb_get_line_number "TAG: line 3" ]
> + DW_LNS_copy
> + DW_LNE_set_address line7
> + line [gdb_get_line_number "TAG: line 5" ]
> + DW_LNS_copy
> + DW_LNE_set_address line8
> + line [gdb_get_line_number "TAG: line 8" ]
> + DW_LNS_copy
> + DW_LNE_set_address main_return
> + line [gdb_get_line_number "TAG: main return"]
> + DW_LNS_copy
> + DW_LNE_set_address end_of_sequence
> + DW_LNE_end_sequence
> + }
> + }
> +}
> +
> +if { [prepare_for_testing "failed to prepare" ${testfile} \
> + [list $srcfile $asm_file] {nodebug} ] } {
> + return -1
> +}
> +
> +runto_main
> +
> +# Print the line table
> +gdb_test_multiple "maint info line-table ${testfile}" "" {
> + -re "\r\n$decimal\[ \t\]+$decimal\[ \t\]+($hex)\[
> \t\]+Y\[^\r\n\]*" {
> + lappend is_stmt $expect_out(1,string)
> + exp_continue
> + }
> + -re -wrap "" {
> + }
> +}
> +
> +# Do the reverse-step test
> +gdb_test_no_output "record" "turn on process record"
> +
> +set bp_main_return [gdb_get_line_number "TAG: main return" $srcfile]
> +gdb_breakpoint $srcfile:$bp_main_return
> +gdb_continue_to_breakpoint "run to end of main, reverse-step test"
> ".*$srcfile:$bp_main_return.*"
> +gdb_test "display \$pc" ".*pc =.*" "display pc, reverse-step test"
> +
> +# At this point, GDB has already recorded the execution up until the
> return
> +# statement. Reverse-step and test if GDB transitions between lines
> in the
> +# expected order. It should reverse-step across lines 8, 5, 3, 2
> and 1.
> +foreach line {8 5 3 2 1} {
> + gdb_test "reverse-step" ".*TAG: line $line.*" "reverse step to
> line $line"
> +}
> +
> +## Clean restart, test reverse-next command
> +clean_restart ${testfile}
> +runto_main
> +gdb_test_no_output "record" "turn on process record, reverst-next
> test"
> +
> +set bp_main_return [gdb_get_line_number "TAG: main return" $srcfile]
> +gdb_breakpoint $srcfile:$bp_main_return
> +gdb_continue_to_breakpoint "run to end of main, reverse-next test"
> ".*$srcfile:$bp_main_return.*"
> +gdb_test "display \$pc" ".*pc =.*" "display pc, reverse-next test"
> +
> +# At this point, GDB has already recorded the execution up until the
> return
> +# statement. Reverse-next and test if GDB transitions between lines
> in the
> +# expected order. It should reverse-next across lines 8, 5, 3, 2
> and 1.
> +foreach line {8 5 3 2 1} {
> + gdb_test "reverse-next" ".*TAG: line $line.*" "reverse next to
> line $line"
> +}
^ permalink raw reply [flat|nested] 40+ messages in thread
* Re: [PATCH 2/2 v5] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-05-25 15:08 ` Carl Love
@ 2023-06-08 16:36 ` Carl Love
0 siblings, 0 replies; 40+ messages in thread
From: Carl Love @ 2023-06-08 16:36 UTC (permalink / raw)
To: Simon Marchi, Bruno Larsen, gdb-patches, UlrichWeigand, pedro
Cc: luis.machado, cel
Ping. Hoping Simon can take a look at the patch. Thanks.
Carl
On Thu, 2023-05-25 at 08:08 -0700, Carl Love wrote:
> Ping.
>
> Simon:
>
> I know Simon said he was busy but wanted to look more at this patch.
> Just wondering if you have found any time to review the patch
> further.
>
> Thanks
>
> Carl
>
> On Tue, 2023-05-16 at 15:54 -0700, Carl Love wrote:
> > Bruno, Simon, GDB maintainers:
> >
> > Version 5, changed comments in test case func-map-to-same-line.c.
> > Patch 1/2 implemented the new options for gdb_compile. Updated the
> > call to proc run_tests to use the new gdb_compile options in a
> > foreach_with_prefix loop.
> >
> > Version 4, additional fixes for gcc version check, wrap function
> > calls
> > using "with_test_prefix", move load_lib dwarf.exe. Fixed typo noted
> > by
> > Luis.
> >
> > Version 3, added the gcc version check as discussed further from
> > version 2 of the patch. Also updated the tests to check for
> > supporting
> > reverse execution rather than requiring recording. I also noticed
> > there were a couple more instances of a requirement check, i.e. if
> > []
> > which I changed to "require" per the current style for checking on
> > the
> > test requirements.
> >
> >
> > The following patch fixes issues on PowerPC with the reverse-step
> > and
> > reverse-next instructions when there are multiple assignment
> > statements
> > on the same line and when there are multiple function calls on the
> > same
> > line. The commit log below discusses these issues in further
> > depth.
> > The discussion included what the correct operation should be for
> > these
> > commands based on the GDB documentation. The proposed patch at
> > that
> > time changed how the commands worked on other platforms such as X86
> > in
> > a way they no longer matched the documentation.
> >
> > The issue is the line table contains multiple entries for the same
> > source line. The patch adds a function to search the line table to
> > find the address of the first instruction of a line. When setup up
> > the
> > reverse stepping range, the function is called to make sure the
> > start
> > of the range corresponds to the address of the first instruction
> > for
> > the line. This approach was used. When Luis initially developed
> > the
> > patch, he considered merging the contiguous ranges in the line
> > table
> > when reading the line tables. He decided it was better to work with
> > the
> > data directly in the line table rather than creating and using a
> > modified version of the line table.
> >
> > The following patch fixes the execution of the reveres-step and
> > reverse-next commands for both senarios of multiple statements on
> > the
> > same line for PowerPC and aarch64-linux. Unlike the previous
> > patch,
> > it
> > does not change the operation of the commands on other platforms,
> > i.e.
> > X86. The patch adds new test cases for both scenarios to verify
> > they
> > work correctly.
> >
> > The patch has been tested on PowerPC, Intel X86 and aarch64-linux
> > with
> > no new regression failures.
> >
> > Please let me know if the patch is acceptable for
> > mainline. Thanks.
> >
> > Carl
> >
> > ---------------------------------------------
> > Fix reverse stepping multiple contiguous PC ranges over the line
> > table.
> >
> > There are a couple of scenarios where the GDB reverse-step and
> > reverse-next
> > commands do not work correctly.
> >
> > Scenario 1 issue description by Luis Machado:
> >
> > When running GDB's testsuite on aarch64-linux/Ubuntu 20.04 (also
> > spotted on
> > the ppc backend), I noticed some failures in gdb.reverse/solib-
> > precsave.exp
> > and gdb.reverse/solib-reverse.exp.
> >
> > The failure happens around the following code:
> >
> > 38 b[1] = shr2(17); /* middle part two */
> > 40 b[0] = 6; b[1] = 9; /* generic statement, end part two */
> > 42 shr1 ("message 1\n"); /* shr1 one */
> >
> > Normal execution:
> >
> > - step from line 38 will land on line 40.
> > - step from line 40 will land on line 42.
> >
> > Reverse execution:
> > - step from line 42 will land on line 40.
> > - step from line 40 will land on line 40.
> > - step from line 40 will land on line 38.
> >
> > The problem here is that line 40 contains two contiguous but
> > distinct
> > PC ranges in the line table, like so:
> >
> > Line 40 - [0x7ec ~ 0x7f4]
> > Line 40 - [0x7f4 ~ 0x7fc]
> >
> > The two distinct ranges are generated because GCC started
> > outputting
> > source
> > column information, which GDB doesn't take into account at the
> > moment.
> >
> > When stepping forward from line 40, we skip both of these ranges
> > and
> > land on
> > line 42. When stepping backward from line 42, we stop at the start
> > PC
> > of the
> > second (or first, going backwards) range of line 40.
> >
> > Since we've reached ecs->event_thread->control.step_range_start, we
> > stop
> > stepping backwards.
> >
> > ---------------------------------------------------------
> >
> > Scenario 2 issue described by Pedro Alves:
> >
> > The following explanation of the issue was taken from the gdb
> > mailing
> > list
> > discussion of the withdrawn patch to change the behavior of the
> > reverse-step
> > and reverse-next commands. Specifically, message from Pedro Alves
> > <pedro@palves.net> where he demonstrates the issue where you have
> > multiple
> > function calls on the same source code line:
> >
> > https://sourceware.org/pipermail/gdb-patches/2023-January/196110.html
> >
> > The source line looks like:
> >
> > func1 (); func2 ();
> >
> > so stepping backwards over that line should always stop at the
> > first
> > instruction of the line, not in the middle. Let's simplify this.
> >
> > Here's the full source code of my example:
> >
> > (gdb) list 1
> > 1 void func1 ()
> > 2 {
> > 3 }
> > 4
> > 5 void func2 ()
> > 6 {
> > 7 }
> > 8
> > 9 int main ()
> > 10 {
> > 11 func1 (); func2 ();
> > 12 }
> >
> > Compiled with:
> >
> > $ gcc reverse.c -o reverse -g3 -O0
> > $ gcc -v
> > ...
> > gcc version 11.3.0 (Ubuntu 11.3.0-1ubuntu1~22.04)
> >
> > Now let's debug it with target record, using current gdb git master
> > (f3d8ae90b236),
> > without your patch:
> >
> > $ gdb ~/reverse
> > GNU gdb (GDB) 14.0.50.20230124-git
> > ...
> > Reading symbols from /home/pedro/reverse...
> > (gdb) start
> > Temporary breakpoint 1 at 0x1147: file reverse.c, line 11.
> > Starting program: /home/pedro/reverse
> > [Thread debugging using libthread_db enabled]
> > Using host libthread_db library "/lib/x86_64-linux-
> > gnu/libthread_db.so.1".
> >
> > Temporary breakpoint 1, main () at reverse.c:11
> > 11 func1 (); func2 ();
> > (gdb) record
> >
> > (gdb) disassemble /s
> > Dump of assembler code for function main:
> > reverse.c:
> > 10 {
> > 0x000055555555513f <+0>: endbr64
> > 0x0000555555555143 <+4>: push %rbp
> > 0x0000555555555144 <+5>: mov %rsp,%rbp
> >
> > 11 func1 (); func2 ();
> > => 0x0000555555555147 <+8>: mov $0x0,%eax
> > 0x000055555555514c <+13>: call 0x555555555129 <func1>
> > 0x0000555555555151 <+18>: mov $0x0,%eax
> > 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> > 0x000055555555515b <+28>: mov $0x0,%eax
> >
> > 12 }
> > 0x0000555555555160 <+33>: pop %rbp
> > 0x0000555555555161 <+34>: ret
> > End of assembler dump.
> >
> > (gdb) n
> > 12 }
> >
> > So far so good, a "next" stepped over the whole of line 11 and
> > stopped at line 12.
> >
> > Let's confirm where we are now:
> >
> > (gdb) disassemble /s
> > Dump of assembler code for function main:
> > reverse.c:
> > 10 {
> > 0x000055555555513f <+0>: endbr64
> > 0x0000555555555143 <+4>: push %rbp
> > 0x0000555555555144 <+5>: mov %rsp,%rbp
> >
> > 11 func1 (); func2 ();
> > 0x0000555555555147 <+8>: mov $0x0,%eax
> > 0x000055555555514c <+13>: call 0x555555555129 <func1>
> > 0x0000555555555151 <+18>: mov $0x0,%eax
> > 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> > 0x000055555555515b <+28>: mov $0x0,%eax
> >
> > 12 }
> > => 0x0000555555555160 <+33>: pop %rbp
> > 0x0000555555555161 <+34>: ret
> > End of assembler dump.
> >
> > Good, we're at the first instruction of line 12.
> >
> > Now let's undo the "next", with "reverse-next":
> >
> > (gdb) reverse-next
> > 11 func1 (); func2 ();
> >
> > Seemingly stopped at line 11. Let's see exactly where:
> >
> > (gdb) disassemble /s
> > Dump of assembler code for function main:
> > reverse.c:
> > 10 {
> > 0x000055555555513f <+0>: endbr64
> > 0x0000555555555143 <+4>: push %rbp
> > 0x0000555555555144 <+5>: mov %rsp,%rbp
> >
> > 11 func1 (); func2 ();
> > 0x0000555555555147 <+8>: mov $0x0,%eax
> > 0x000055555555514c <+13>: call 0x555555555129 <func1>
> > => 0x0000555555555151 <+18>: mov $0x0,%eax
> > 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> > 0x000055555555515b <+28>: mov $0x0,%eax
> >
> > 12 }
> > 0x0000555555555160 <+33>: pop %rbp
> > 0x0000555555555161 <+34>: ret
> > End of assembler dump.
> > (gdb)
> >
> > And lo, we stopped in the middle of line 11! That is a bug, we
> > should have
> > stepped back all the way to the beginning of the line. The
> > "reverse-
> > next"
> > should have fully undone the prior "next" command.
> >
> > The above issues were fixed by introducing a new function that
> > looks
> > for
> > adjacent PC ranges for the same line, until we notice a line
> > change.
> > Then
> > we take that as the start PC of the range. The new start PC for
> > the
> > range
> > is used for the control.step_range_start when setting up a step
> > range.
> >
> > The test case gdb.reverse/map-to-same-line.exp is added to test the
> > fix
> > for the issues in scenario 1.
> >
> > The test case gdb.reverse/func-map-to-same-line.exp was added to
> > test
> > the
> > fix for scenario 2 when the binary was compiled with and without
> > line
> > table information.
> >
> > bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28426
> >
> > Co-authored-by: Luis Machado <luis.machado@arm.com>
> > Co-authored-by: Carl Love <cel@us.ibm.com>
> > Reviewed-By: Bruno Larsen <blarsen@redhat.com>
> > ---
> > gdb/infrun.c | 57 +++++++
> > gdb/symtab.c | 49 ++++++
> > gdb/symtab.h | 16 ++
> > .../gdb.reverse/func-map-to-same-line.c | 36 ++++
> > .../gdb.reverse/func-map-to-same-line.exp | 140
> > ++++++++++++++++
> > gdb/testsuite/gdb.reverse/map-to-same-line.c | 58 +++++++
> > .../gdb.reverse/map-to-same-line.exp | 156
> > ++++++++++++++++++
> > 7 files changed, 512 insertions(+)
> > create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-
> > line.c
> > create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-
> > line.exp
> > create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.c
> > create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.exp
> >
> > diff --git a/gdb/infrun.c b/gdb/infrun.c
> > index efe2c00c489..31cd817c733 100644
> > --- a/gdb/infrun.c
> > +++ b/gdb/infrun.c
> > @@ -114,6 +114,9 @@ static struct async_event_handler
> > *infrun_async_inferior_event_token;
> > Starts off as -1, indicating "never enabled/disabled". */
> > static int infrun_is_async = -1;
> >
> > +static CORE_ADDR update_line_range_start (CORE_ADDR pc,
> > + struct
> > execution_control_state *ecs);
> > +
> > /* See infrun.h. */
> >
> > void
> > @@ -6769,6 +6772,25 @@ handle_signal_stop (struct
> > execution_control_state *ecs)
> > process_event_stop_test (ecs);
> > }
> >
> > +CORE_ADDR
> > +update_line_range_start (CORE_ADDR pc, struct
> > execution_control_state *ecs)
> > +{
> > + /* The line table may have multiple entries for the same source
> > code line.
> > + Given the PC, check the line table and return the PC that
> > corresponds
> > + to the line table entry for the source line that PC is
> > in. */
> > + CORE_ADDR start_line_pc = ecs->event_thread-
> > > control.step_range_start;
> > + gdb::optional<CORE_ADDR> real_range_start;
> > +
> > + /* Call find_line_range_start to get the smallest address in the
> > + linetable for multiple Line X entries in the line table. */
> > + real_range_start = find_line_range_start (pc);
> > +
> > + if (real_range_start.has_value ())
> > + start_line_pc = *real_range_start;
> > +
> > + return start_line_pc;
> > +}
> > +
> > /* Come here when we've got some debug event / signal we can
> > explain
> > (IOW, not a random signal), and test whether it should cause a
> > stop, or whether we should resume the inferior (transparently).
> > @@ -7570,6 +7592,28 @@ process_event_stop_test (struct
> > execution_control_state *ecs)
> >
> > if (stop_pc_sal.is_stmt)
> > {
> > + if (execution_direction == EXEC_REVERSE)
> > + {
> > + /* We are stepping backwards make sure we have reached
> > the
> > + beginning of the line. */
> > + CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
> > + CORE_ADDR start_line_pc
> > + = update_line_range_start (stop_pc, ecs);
> > +
> > + if (stop_pc != start_line_pc)
> > + {
> > + /* Have not reached the beginning of the source code
> > line.
> > + Set a step range. Execution should stop in any
> > function
> > + calls we execute back into before reaching the
> > beginning
> > + of the line. */
> > + ecs->event_thread->control.step_range_start =
> > start_line_pc;
> > + ecs->event_thread->control.step_range_end = stop_pc;
> > + set_step_info (ecs->event_thread, frame,
> > stop_pc_sal);
> > + keep_going (ecs);
> > + return;
> > + }
> > + }
> > +
> > /* We are at the start of a statement.
> >
> > So stop. Note that we don't stop if we step into the
> > middle of a
> > @@ -7632,6 +7676,19 @@ process_event_stop_test (struct
> > execution_control_state *ecs)
> > set_step_info (ecs->event_thread, frame, stop_pc_sal);
> >
> > infrun_debug_printf ("keep going");
> > +
> > + if (execution_direction == EXEC_REVERSE)
> > + {
> > + CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
> > +
> > + /* Make sure the stop_pc is set to the beginning of the
> > line. */
> > + if (stop_pc != ecs->event_thread->control.step_range_start)
> > + {
> > + stop_pc = update_line_range_start (stop_pc, ecs);
> > + ecs->event_thread->control.step_range_start = stop_pc;
> > + }
> > + }
> > +
> > keep_going (ecs);
> > }
> >
> > diff --git a/gdb/symtab.c b/gdb/symtab.c
> > index 27611a34ec4..91d35616eb9 100644
> > --- a/gdb/symtab.c
> > +++ b/gdb/symtab.c
> > @@ -3282,6 +3282,55 @@ find_pc_line (CORE_ADDR pc, int notcurrent)
> > return sal;
> > }
> >
> > +/* Compare two symtab_and_line entries. Return true if both have
> > + the same line number and the same symtab pointer. That means
> > we
> > + are dealing with two entries from the same line and from the
> > same
> > + source file.
> > +
> > + Return false otherwise. */
> > +
> > +static bool
> > +sal_line_symtab_matches_p (const symtab_and_line &sal1,
> > + const symtab_and_line &sal2)
> > +{
> > + return (sal1.line == sal2.line && sal1.symtab == sal2.symtab);
> > +}
> > +
> > +/* See symtah.h. */
> > +
> > +gdb::optional<CORE_ADDR>
> > +find_line_range_start (CORE_ADDR pc)
> > +{
> > + struct symtab_and_line current_sal = find_pc_line (pc, 0);
> > +
> > + if (current_sal.line == 0)
> > + return {};
> > +
> > + struct symtab_and_line prev_sal = find_pc_line (current_sal.pc -
> > 1, 0);
> > +
> > + /* If the previous entry is for a different line, that means we
> > are already
> > + at the entry with the start PC for this line. */
> > + if (!sal_line_symtab_matches_p (prev_sal, current_sal))
> > + return current_sal.pc;
> > +
> > + /* Otherwise, keep looking for entries for the same line but
> > with
> > + smaller PC's. */
> > + bool done = false;
> > + CORE_ADDR prev_pc;
> > + while (!done)
> > + {
> > + prev_pc = prev_sal.pc;
> > +
> > + prev_sal = find_pc_line (prev_pc - 1, 0);
> > +
> > + /* Did we notice a line change? If so, we are done with the
> > search. */
> > + if (!sal_line_symtab_matches_p (prev_sal, current_sal))
> > + done = true;
> > + }
> > +
> > + return prev_pc;
> > +}
> > +
> > /* See symtab.h. */
> >
> > struct symtab *
> > diff --git a/gdb/symtab.h b/gdb/symtab.h
> > index 404d0ab30a8..f54305636da 100644
> > --- a/gdb/symtab.h
> > +++ b/gdb/symtab.h
> > @@ -2346,6 +2346,22 @@ extern struct symtab_and_line find_pc_line
> > (CORE_ADDR, int);
> > extern struct symtab_and_line find_pc_sect_line (CORE_ADDR,
> > struct obj_section *,
> > int);
> >
> > +/* Given PC, and assuming it is part of a range of addresses that
> > is
> > part of a
> > + line, go back through the linetable and find the starting PC of
> > that
> > + line.
> > +
> > + For example, suppose we have 3 PC ranges for line X:
> > +
> > + Line X - [0x0 - 0x8]
> > + Line X - [0x8 - 0x10]
> > + Line X - [0x10 - 0x18]
> > +
> > + If we call the function with PC == 0x14, we want to return 0x0,
> > as that is
> > + the starting PC of line X, and the ranges are contiguous.
> > +*/
> > +
> > +extern gdb::optional<CORE_ADDR> find_line_range_start (CORE_ADDR
> > pc);
> > +
> > /* Wrapper around find_pc_line to just return the symtab. */
> >
> > extern struct symtab *find_pc_line_symtab (CORE_ADDR);
> > diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> > b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> > new file mode 100644
> > index 00000000000..da944874e86
> > --- /dev/null
> > +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> > @@ -0,0 +1,36 @@
> > +/* Copyright 2008-2023 Free Software Foundation, Inc.
> > +
> > + This program is free software; you can redistribute it and/or
> > modify
> > + it under the terms of the GNU General Public License as
> > published
> > by
> > + the Free Software Foundation; either version 3 of the License,
> > or
> > + (at your option) any later version.
> > +
> > + This program is distributed in the hope that it will be useful,
> > + but WITHOUT ANY WARRANTY; without even the implied warranty of
> > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> > + GNU General Public License for more details.
> > +
> > + You should have received a copy of the GNU General Public
> > License
> > + along with this program. If not, see <
> > http://www.gnu.org/licenses/>;.
> > +
> > + This test is used to test the reverse-step and reverse-next
> > instruction
> > + execution for a source line that contains multiple function
> > calls. */
> > +
> > +void
> > +func1 ()
> > +{
> > +} /* END FUNC1 */
> > +
> > +void
> > +func2 ()
> > +{
> > +} /* END FUNC2 */
> > +
> > +int main ()
> > +{
> > + int a, b;
> > + a = 1;
> > + b = 2;
> > + func1 (); func2 ();
> > + a = a + b; /* START REVERSE TEST */
> > +}
> > diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> > b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> > new file mode 100644
> > index 00000000000..89e226b0f84
> > --- /dev/null
> > +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> > @@ -0,0 +1,140 @@
> > +# Copyright 2008-2023 Free Software Foundation, Inc.
> > +
> > +# This program is free software; you can redistribute it and/or
> > modify
> > +# it under the terms of the GNU General Public License as
> > published
> > by
> > +# the Free Software Foundation; either version 3 of the License,
> > or
> > +# (at your option) any later version.
> > +#
> > +# This program is distributed in the hope that it will be useful,
> > +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> > +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> > +# GNU General Public License for more details.
> > +#
> > +# You should have received a copy of the GNU General Public
> > License
> > +# along with this program. If not, see <
> > http://www.gnu.org/licenses/>. */
> > +
> > +# This file is part of the GDB testsuite. It tests reverse
> > stepping.
> > +# Lots of code borrowed from "step-test.exp".
> > +
> > +# This test checks to make sure there is no regression failures
> > for
> > +# the reverse-next command when stepping back over two functions
> > in
> > +# the same line.
> > +
> > +require supports_reverse
> > +
> > +# This test uses the gcc no-column-info command which was added in
> > gcc 7.1.
> > +
> > +proc run_tests {} {
> > + global srcfile
> > + global executable
> > +
> > + runto_main
> > + set target_remote [gdb_is_target_remote]
> > +
> > + with_test_prefix "test1" {
> > + gdb_test_no_output "record" "turn on process record"
> > + }
> > +
> > + # This regression test verifies the reverse-step and reverse-
> > next commands
> > + # work properly when executing backwards thru a source line
> > containing
> > + # two function calls on the same source line, i.e. func1 ();
> > func2 ();
> > + # This test is compiled so the dwarf info not contain the line
> > table
> > + # information.
> > +
> > + # Test 1, reverse-next command
> > + # Set breakpoint at the line after the function calls.
> > + set bp_start_reverse_test [gdb_get_line_number "START REVERSE
> > TEST" \
> > + $srcfile]
> > + gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> > +
> > + # Continue to break point for reverse-next test.
> > + # Command definition: reverse-next [count]
> > + # Run backward to the beginning of the previous line
> > executed
> > in the
> > + # current (innermost) stack frame. If the line contains
> > function calls,
> > + # they will be “un-executed” without stopping. Starting from
> > the first
> > + # line of a function, reverse-next will take you back to the
> > caller of
> > + # that function, before the function was called, just as the
> > normal next
> > + # command would take you from the last line of a function
> > back
> > to its
> > + # return to its caller 2 .
> > + gdb_continue_to_breakpoint \
> > + "test1: stopped at command reverse-next test start location" \
> > + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> > +
> > + # The reverse-next should step all the way back to the
> > beginning
> > of the
> > + # line, i.e. at the beginning of the func1 call.
> > + gdb_test "reverse-next" ".*func1 \\(\\); func2 \\(\\);.*" \
> > + "test1: reverse-next to line with two functions"
> > +
> > + # We should be stopped at the first instruction of the line. A
> > reverse-step
> > + # should step back and stop at the beginning of the previous
> > line b = 2,
> > + # i.e. not in func1 ().
> > + gdb_test "reverse-stepi" ".*b = 2;.*" \
> > + "test1: reverse-stepi to previous line b = 2"
> > +
> > +
> > + # Setup for test 2
> > + clean_restart $executable
> > + runto_main
> > +
> > + with_test_prefix "test2" {
> > + gdb_test_no_output "record" "turn on process record"
> > + }
> > +
> > + # Test 2, reverse-step command
> > + # Set breakpoint at the line after the function calls.
> > + gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> > +
> > + # Continue to the start of the reverse-step test.
> > + # Command definition: reverse-step [count]
> > + # Run the program backward until control reaches the start
> > of
> > a
> > + # different source line; then stop it, and return control
> > to
> > gdb.
> > + # Like the step command, reverse-step will only stop at the
> > beginning
> > + # of a source line. It “un-executes” the previously
> > executed
> > source
> > + # line. If the previous source line included calls to
> > debuggable
> > + # functions, reverse-step will step (backward) into the
> > called function,
> > + # stopping at the beginning of the last statement in the
> > called
> > + # function (typically a return statement). Also, as with
> > the
> > step
> > + # command, if non-debuggable functions are called, reverse-
> > step will
> > + # run thru them backward without stopping.
> > +
> > + gdb_continue_to_breakpoint \
> > + "test2: stopped at command reverse-step test start location" \
> > + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> > +
> > + # The first reverse step should take us call of func2 ().
> > + gdb_test "reverse-step" ".*END FUNC2.*" \
> > + "test2: reverse-step into func2 "
> > +
> > + # The second reverse step should take us into func1 ().
> > + gdb_test "reverse-step" ".*END FUNC1.*" \
> > + "test2: reverse-step into func1 "
> > +
> > + # The third reverse step should take us call of func1 ().
> > + gdb_test "reverse-step" ".*func1 \\(\\); func2 \\(\\);.*" \
> > + "test2: reverse-step to line func1(); func2(), at call for
> > func1 "
> > +
> > + # We should be stopped at the first instruction of the line. A
> > reverse
> > + # stepi should take us to b = 2 ().
> > + gdb_test "reverse-stepi" ".*b = 2;.*" \
> > + "test2: reverse-stepi to line b = 2 "
> > +}
> > +
> > +set srcfile func-map-to-same-line.c
> > +set executable func-map-to-same-line
> > +
> > +# test with and without gcc column info enabled
> > +foreach_with_prefix with_column_info {yes no} {
> > + if {$with_column_info == "yes"} {
> > + set options [list debug column-info]
> > + } else {
> > + set options [list debug no-column-info]
> > + }
> > +
> > + if {[build_executable "failed to prepare" $executable $srcfile
> > \
> > + $options] == -1} {
> > + return -1
> > + }
> > +
> > + clean_restart $executable
> > + run_tests
> > +}
> > diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.c
> > b/gdb/testsuite/gdb.reverse/map-to-same-line.c
> > new file mode 100644
> > index 00000000000..f20d778f40e
> > --- /dev/null
> > +++ b/gdb/testsuite/gdb.reverse/map-to-same-line.c
> > @@ -0,0 +1,58 @@
> > +/* Copyright 2008-2023 Free Software Foundation, Inc.
> > +
> > + This program is free software; you can redistribute it and/or
> > modify
> > + it under the terms of the GNU General Public License as
> > published
> > by
> > + the Free Software Foundation; either version 3 of the License,
> > or
> > + (at your option) any later version.
> > +
> > + This program is distributed in the hope that it will be useful,
> > + but WITHOUT ANY WARRANTY; without even the implied warranty of
> > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> > + GNU General Public License for more details.
> > +
> > + You should have received a copy of the GNU General Public
> > License
> > + along with this program. If not, see <
> > http://www.gnu.org/licenses/
> > >. */
> > +
> > +/* The purpose of this test is to create a DWARF line table that
> > contains two
> > + or more entries for the same line. When stepping (forwards or
> > backwards),
> > + GDB should step over the entire line and not just a particular
> > entry in the
> > + line table. */
> > +
> > +int
> > +main ()
> > +{ /* TAG: main prologue */
> > + asm ("main_label: .globl main_label");
> > + int i = 1, j = 2, k;
> > + float f1 = 2.0, f2 = 4.1, f3;
> > + const char *str_1 = "foo", *str_2 = "bar", *str_3;
> > +
> > + asm ("line1: .globl line1");
> > + k = i; f3 = f1; str_3 = str_1; /* TAG: line 1 */
> > +
> > + asm ("line2: .globl line2");
> > + k = j; f3 = f2; str_3 = str_2; /* TAG: line 2 */
> > +
> > + asm ("line3: .globl line3");
> > + k = i; f3 = f1; str_3 = str_1; /* TAG: line 3 */
> > +
> > + asm ("line4: .globl line4");
> > + k = j; f3 = f2; str_3 = str_2; /* TAG: line 4 */
> > +
> > + asm ("line5: .globl line5");
> > + k = i; f3 = f1; str_3 = str_1; /* TAG: line 5 */
> > +
> > + asm ("line6: .globl line6");
> > + k = j; f3 = f2; str_3 = str_2; /* TAG: line 6 */
> > +
> > + asm ("line7: .globl line7");
> > + k = i; f3 = f1; str_3 = str_1; /* TAG: line 7 */
> > +
> > + asm ("line8: .globl line8");
> > + k = j; f3 = f2; str_3 = str_2; /* TAG: line 8 */
> > +
> > + asm ("main_return: .globl main_return");
> > + k = j; f3 = f2; str_3 = str_2; /* TAG: main return */
> > +
> > + asm ("end_of_sequence: .globl end_of_sequence");
> > + return 0; /* TAG: main return */
> > +}
> > diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> > b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> > new file mode 100644
> > index 00000000000..16a359d90ec
> > --- /dev/null
> > +++ b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> > @@ -0,0 +1,156 @@
> > +# Copyright 2008-2023 Free Software Foundation, Inc.
> > +
> > +# This program is free software; you can redistribute it and/or
> > modify
> > +# it under the terms of the GNU General Public License as
> > published
> > by
> > +# the Free Software Foundation; either version 3 of the License,
> > or
> > +# (at your option) any later version.
> > +#
> > +# This program is distributed in the hope that it will be useful,
> > +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> > +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> > +# GNU General Public License for more details.
> > +#
> > +# You should have received a copy of the GNU General Public
> > License
> > +# along with this program. If not, see <
> > http://www.gnu.org/licenses/
> > >.
> > +
> > +# When stepping (forwards or backwards), GDB should step over the
> > entire line
> > +# and not just a particular entry in the line table. This test was
> > added to
> > +# verify the find_line_range_start function properly sets the step
> > range for a
> > +# line that consists of multiple statements, i.e. multiple entries
> > in the line
> > +# table. This test creates a DWARF line table that contains two
> > entries for
> > +# the same line to do the needed testing.
> > +
> > +# This test can only be run on targets which support DWARF-2 and
> > use
> > gas.
> > +load_lib dwarf.exp
> > +require dwarf2_support
> > +
> > +# The DWARF assembler requires the gcc compiler.
> > +require is_c_compiler_gcc
> > +
> > +# This test suitable only for process that can do reverse
> > execution
> > +require supports_reverse
> > +
> > +standard_testfile .c .S
> > +
> > +if { [prepare_for_testing "failed to prepare" ${testfile}
> > ${srcfile}] } {
> > + return -1
> > +}
> > +
> > +set asm_file [standard_output_file $srcfile2]
> > +Dwarf::assemble $asm_file {
> > + global srcdir subdir srcfile
> > + declare_labels integer_label L
> > +
> > + # Find start address and length of program
> > + lassign [function_range main [list
> > ${srcdir}/${subdir}/$srcfile]] \
> > + main_start main_len
> > + set main_end "$main_start + $main_len"
> > +
> > + cu {} {
> > + compile_unit {
> > + {language @DW_LANG_C}
> > + {name map-to-same-line.c}
> > + {stmt_list $L DW_FORM_sec_offset}
> > + {low_pc 0 addr}
> > + } {
> > + subprogram {
> > + {external 1 flag}
> > + {name main}
> > + {low_pc $main_start addr}
> > + {high_pc $main_len DW_FORM_data4}
> > + }
> > + }
> > + }
> > +
> > + lines {version 2 default_is_stmt 1} L {
> > + include_dir "${srcdir}/${subdir}"
> > + file_name "$srcfile" 1
> > +
> > + # Generate the line table program with distinct source lines
> > being
> > + # mapped to the same line entry. Line 1, 5 and 8 contain 1
> > statement
> > + # each. Line 2 contains 2 statements. Line 3 contains 3
> > statements.
> > + program {
> > + DW_LNE_set_address $main_start
> > + line [gdb_get_line_number "TAG: main prologue"]
> > + DW_LNS_copy
> > + DW_LNE_set_address line1
> > + line [gdb_get_line_number "TAG: line 1" ]
> > + DW_LNS_copy
> > + DW_LNE_set_address line2
> > + line [gdb_get_line_number "TAG: line 2" ]
> > + DW_LNS_copy
> > + DW_LNE_set_address line3
> > + line [gdb_get_line_number "TAG: line 2" ]
> > + DW_LNS_copy
> > + DW_LNE_set_address line4
> > + line [gdb_get_line_number "TAG: line 3" ]
> > + DW_LNS_copy
> > + DW_LNE_set_address line5
> > + line [gdb_get_line_number "TAG: line 3" ]
> > + DW_LNS_copy
> > + DW_LNE_set_address line6
> > + line [gdb_get_line_number "TAG: line 3" ]
> > + DW_LNS_copy
> > + DW_LNE_set_address line7
> > + line [gdb_get_line_number "TAG: line 5" ]
> > + DW_LNS_copy
> > + DW_LNE_set_address line8
> > + line [gdb_get_line_number "TAG: line 8" ]
> > + DW_LNS_copy
> > + DW_LNE_set_address main_return
> > + line [gdb_get_line_number "TAG: main return"]
> > + DW_LNS_copy
> > + DW_LNE_set_address end_of_sequence
> > + DW_LNE_end_sequence
> > + }
> > + }
> > +}
> > +
> > +if { [prepare_for_testing "failed to prepare" ${testfile} \
> > + [list $srcfile $asm_file] {nodebug} ] } {
> > + return -1
> > +}
> > +
> > +runto_main
> > +
> > +# Print the line table
> > +gdb_test_multiple "maint info line-table ${testfile}" "" {
> > + -re "\r\n$decimal\[ \t\]+$decimal\[ \t\]+($hex)\[
> > \t\]+Y\[^\r\n\]*" {
> > + lappend is_stmt $expect_out(1,string)
> > + exp_continue
> > + }
> > + -re -wrap "" {
> > + }
> > +}
> > +
> > +# Do the reverse-step test
> > +gdb_test_no_output "record" "turn on process record"
> > +
> > +set bp_main_return [gdb_get_line_number "TAG: main return"
> > $srcfile]
> > +gdb_breakpoint $srcfile:$bp_main_return
> > +gdb_continue_to_breakpoint "run to end of main, reverse-step
> > test"
> > ".*$srcfile:$bp_main_return.*"
> > +gdb_test "display \$pc" ".*pc =.*" "display pc, reverse-step test"
> > +
> > +# At this point, GDB has already recorded the execution up until
> > the
> > return
> > +# statement. Reverse-step and test if GDB transitions between
> > lines
> > in the
> > +# expected order. It should reverse-step across lines 8, 5, 3, 2
> > and 1.
> > +foreach line {8 5 3 2 1} {
> > + gdb_test "reverse-step" ".*TAG: line $line.*" "reverse step to
> > line $line"
> > +}
> > +
> > +## Clean restart, test reverse-next command
> > +clean_restart ${testfile}
> > +runto_main
> > +gdb_test_no_output "record" "turn on process record, reverst-next
> > test"
> > +
> > +set bp_main_return [gdb_get_line_number "TAG: main return"
> > $srcfile]
> > +gdb_breakpoint $srcfile:$bp_main_return
> > +gdb_continue_to_breakpoint "run to end of main, reverse-next
> > test"
> > ".*$srcfile:$bp_main_return.*"
> > +gdb_test "display \$pc" ".*pc =.*" "display pc, reverse-next test"
> > +
> > +# At this point, GDB has already recorded the execution up until
> > the
> > return
> > +# statement. Reverse-next and test if GDB transitions between
> > lines
> > in the
> > +# expected order. It should reverse-next across lines 8, 5, 3, 2
> > and 1.
> > +foreach line {8 5 3 2 1} {
> > + gdb_test "reverse-next" ".*TAG: line $line.*" "reverse next to
> > line $line"
> > +}
>
>
^ permalink raw reply [flat|nested] 40+ messages in thread
* Re: [PATCH 1/2] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-05-16 22:54 ` [PATCH 1/2] " Carl Love
@ 2023-06-19 17:11 ` Simon Marchi
2023-06-22 16:52 ` Carl Love
0 siblings, 1 reply; 40+ messages in thread
From: Simon Marchi @ 2023-06-19 17:11 UTC (permalink / raw)
To: Carl Love, Bruno Larsen, gdb-patches, UlrichWeigand, pedro; +Cc: luis.machado
On 5/16/23 18:54, Carl Love wrote:
> Simon, GDB maintainers:
>
> Per the comments on version 4 for the gdb.reverse/func-map-to-same-
> line.exp, I have added support to proc gdb_compile to enable or disable
> generating line information as part of the debug information. The two
> new options are column-info and no-column-info.
>
> This patch implements the new options for gdb_compile.
>
> These options have been tested with patch 2 of 2 on PowerPC with the
> GCC and clang compilers.
>
> Please let me know if the patch is acceptable for mainline. Thanks.
>
> Carl
>
> -------------------------------------------------------------
>
> Add gdb_compile options column-info and no-column-info
>
> This patch adds two new options to gdb_compile to specify if the compile
> should or should not generate the line table information. The
> options are supported on clang and gcc version 7 and newer.
>
> Patch has been tested on PowerPC with both gcc and clang.
> ---
> gdb/testsuite/lib/gdb.exp | 30 ++++++++++++++++++++++++++++++
> 1 file changed, 30 insertions(+)
>
> diff --git a/gdb/testsuite/lib/gdb.exp b/gdb/testsuite/lib/gdb.exp
> index aed7e2d043c..e993fddf4c7 100644
> --- a/gdb/testsuite/lib/gdb.exp
> +++ b/gdb/testsuite/lib/gdb.exp
> @@ -4794,6 +4794,8 @@ proc quote_for_host { args } {
> # debug information
> # - text_segment=addr: Tell the linker to place the text segment at ADDR.
> # - build-id: Ensure the final binary includes a build-id.
> +# - no-column-info: Disable generation of column table information.
> +# - column-info: Enable generation of column table information.
> #
> # And here are some of the not too obscure options understood by DejaGnu that
> # influence the compilation:
> @@ -5003,6 +5005,34 @@ proc gdb_compile {source dest type options} {
> } else {
> error "Don't know how to handle text_segment option."
> }
> + } elseif { $opt == "column-info" } {
> + if {[test_compiler_info {gcc-*}]} {
> + if {[test_compiler_info {gcc-[1-6]-*}]} {
> + error "gdb_compile option no-column-info not supported."
I think this path should return the equivalent of "failed to compile",
instead of throwing an error. Control will go back to the test, which
will generally skip the portion of the test that requires that
binary.
> + }
> + lappend new_options "additional_flags=-gcolumn-info"
> +
> + } elseif {[test_compiler_info {clang-*}]} {
> + lappend new_options "additional_flags=-gcolumn-info"
> +
> + } else {
> + error "Don't know how to handle gcolumn-info option."
I think it's ok to throw an error in this path. If you are testing
against a compiler that we don't know about, it will produce errors that
are easy to spot, and you'll be able to add support for your compiler
here.
Simon
^ permalink raw reply [flat|nested] 40+ messages in thread
* Re: [PATCH 2/2 v5] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-05-16 22:54 ` [PATCH 2/2 v5] " Carl Love
2023-05-25 15:08 ` Carl Love
@ 2023-06-19 17:58 ` Simon Marchi
2023-06-22 20:38 ` Carl Love
` (2 more replies)
1 sibling, 3 replies; 40+ messages in thread
From: Simon Marchi @ 2023-06-19 17:58 UTC (permalink / raw)
To: Carl Love, Bruno Larsen, gdb-patches, UlrichWeigand, pedro; +Cc: luis.machado
On 5/16/23 18:54, Carl Love wrote:
> Bruno, Simon, GDB maintainers:
>
> Version 5, changed comments in test case func-map-to-same-line.c.
> Patch 1/2 implemented the new options for gdb_compile. Updated the
> call to proc run_tests to use the new gdb_compile options in a
> foreach_with_prefix loop.
>
> Version 4, additional fixes for gcc version check, wrap function calls
> using "with_test_prefix", move load_lib dwarf.exe. Fixed typo noted by
> Luis.
>
> Version 3, added the gcc version check as discussed further from
> version 2 of the patch. Also updated the tests to check for supporting
> reverse execution rather than requiring recording. I also noticed
> there were a couple more instances of a requirement check, i.e. if []
> which I changed to "require" per the current style for checking on the
> test requirements.
>
>
> The following patch fixes issues on PowerPC with the reverse-step and
> reverse-next instructions when there are multiple assignment statements
> on the same line and when there are multiple function calls on the same
> line. The commit log below discusses these issues in further depth.
> The discussion included what the correct operation should be for these
> commands based on the GDB documentation. The proposed patch at that
> time changed how the commands worked on other platforms such as X86 in
> a way they no longer matched the documentation.
>
> The issue is the line table contains multiple entries for the same
> source line. The patch adds a function to search the line table to
> find the address of the first instruction of a line. When setup up the
> reverse stepping range, the function is called to make sure the start
> of the range corresponds to the address of the first instruction for
> the line. This approach was used. When Luis initially developed the
> patch, he considered merging the contiguous ranges in the line table
> when reading the line tables. He decided it was better to work with the
> data directly in the line table rather than creating and using a
> modified version of the line table.
>
> The following patch fixes the execution of the reveres-step and
> reverse-next commands for both senarios of multiple statements on the
> same line for PowerPC and aarch64-linux. Unlike the previous patch, it
> does not change the operation of the commands on other platforms, i.e.
> X86. The patch adds new test cases for both scenarios to verify they
> work correctly.
>
> The patch has been tested on PowerPC, Intel X86 and aarch64-linux with
> no new regression failures.
>
> Please let me know if the patch is acceptable for mainline. Thanks.
>
> Carl
>
> ---------------------------------------------
> Fix reverse stepping multiple contiguous PC ranges over the line table.
>
> There are a couple of scenarios where the GDB reverse-step and reverse-next
> commands do not work correctly.
>
> Scenario 1 issue description by Luis Machado:
>
> When running GDB's testsuite on aarch64-linux/Ubuntu 20.04 (also spotted on
> the ppc backend), I noticed some failures in gdb.reverse/solib-precsave.exp
> and gdb.reverse/solib-reverse.exp.
>
> The failure happens around the following code:
>
> 38 b[1] = shr2(17); /* middle part two */
> 40 b[0] = 6; b[1] = 9; /* generic statement, end part two */
> 42 shr1 ("message 1\n"); /* shr1 one */
>
> Normal execution:
>
> - step from line 38 will land on line 40.
> - step from line 40 will land on line 42.
>
> Reverse execution:
> - step from line 42 will land on line 40.
> - step from line 40 will land on line 40.
> - step from line 40 will land on line 38.
>
> The problem here is that line 40 contains two contiguous but distinct
> PC ranges in the line table, like so:
>
> Line 40 - [0x7ec ~ 0x7f4]
> Line 40 - [0x7f4 ~ 0x7fc]
>
> The two distinct ranges are generated because GCC started outputting source
> column information, which GDB doesn't take into account at the moment.
>
> When stepping forward from line 40, we skip both of these ranges and land on
> line 42. When stepping backward from line 42, we stop at the start PC of the
> second (or first, going backwards) range of line 40.
>
> Since we've reached ecs->event_thread->control.step_range_start, we stop
> stepping backwards.
>
> ---------------------------------------------------------
>
> Scenario 2 issue described by Pedro Alves:
>
> The following explanation of the issue was taken from the gdb mailing list
> discussion of the withdrawn patch to change the behavior of the reverse-step
> and reverse-next commands. Specifically, message from Pedro Alves
> <pedro@palves.net> where he demonstrates the issue where you have multiple
> function calls on the same source code line:
>
> https://sourceware.org/pipermail/gdb-patches/2023-January/196110.html
>
> The source line looks like:
>
> func1 (); func2 ();
>
> so stepping backwards over that line should always stop at the first
> instruction of the line, not in the middle. Let's simplify this.
>
> Here's the full source code of my example:
>
> (gdb) list 1
> 1 void func1 ()
> 2 {
> 3 }
> 4
> 5 void func2 ()
> 6 {
> 7 }
> 8
> 9 int main ()
> 10 {
> 11 func1 (); func2 ();
> 12 }
>
> Compiled with:
>
> $ gcc reverse.c -o reverse -g3 -O0
> $ gcc -v
> ...
> gcc version 11.3.0 (Ubuntu 11.3.0-1ubuntu1~22.04)
>
> Now let's debug it with target record, using current gdb git master (f3d8ae90b236),
> without your patch:
>
> $ gdb ~/reverse
> GNU gdb (GDB) 14.0.50.20230124-git
> ...
> Reading symbols from /home/pedro/reverse...
> (gdb) start
> Temporary breakpoint 1 at 0x1147: file reverse.c, line 11.
> Starting program: /home/pedro/reverse
> [Thread debugging using libthread_db enabled]
> Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
>
> Temporary breakpoint 1, main () at reverse.c:11
> 11 func1 (); func2 ();
> (gdb) record
>
> (gdb) disassemble /s
> Dump of assembler code for function main:
> reverse.c:
> 10 {
> 0x000055555555513f <+0>: endbr64
> 0x0000555555555143 <+4>: push %rbp
> 0x0000555555555144 <+5>: mov %rsp,%rbp
>
> 11 func1 (); func2 ();
> => 0x0000555555555147 <+8>: mov $0x0,%eax
> 0x000055555555514c <+13>: call 0x555555555129 <func1>
> 0x0000555555555151 <+18>: mov $0x0,%eax
> 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> 0x000055555555515b <+28>: mov $0x0,%eax
>
> 12 }
> 0x0000555555555160 <+33>: pop %rbp
> 0x0000555555555161 <+34>: ret
> End of assembler dump.
>
> (gdb) n
> 12 }
>
> So far so good, a "next" stepped over the whole of line 11 and stopped at line 12.
>
> Let's confirm where we are now:
>
> (gdb) disassemble /s
> Dump of assembler code for function main:
> reverse.c:
> 10 {
> 0x000055555555513f <+0>: endbr64
> 0x0000555555555143 <+4>: push %rbp
> 0x0000555555555144 <+5>: mov %rsp,%rbp
>
> 11 func1 (); func2 ();
> 0x0000555555555147 <+8>: mov $0x0,%eax
> 0x000055555555514c <+13>: call 0x555555555129 <func1>
> 0x0000555555555151 <+18>: mov $0x0,%eax
> 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> 0x000055555555515b <+28>: mov $0x0,%eax
>
> 12 }
> => 0x0000555555555160 <+33>: pop %rbp
> 0x0000555555555161 <+34>: ret
> End of assembler dump.
>
> Good, we're at the first instruction of line 12.
>
> Now let's undo the "next", with "reverse-next":
>
> (gdb) reverse-next
> 11 func1 (); func2 ();
>
> Seemingly stopped at line 11. Let's see exactly where:
>
> (gdb) disassemble /s
> Dump of assembler code for function main:
> reverse.c:
> 10 {
> 0x000055555555513f <+0>: endbr64
> 0x0000555555555143 <+4>: push %rbp
> 0x0000555555555144 <+5>: mov %rsp,%rbp
>
> 11 func1 (); func2 ();
> 0x0000555555555147 <+8>: mov $0x0,%eax
> 0x000055555555514c <+13>: call 0x555555555129 <func1>
> => 0x0000555555555151 <+18>: mov $0x0,%eax
> 0x0000555555555156 <+23>: call 0x555555555134 <func2>
> 0x000055555555515b <+28>: mov $0x0,%eax
>
> 12 }
> 0x0000555555555160 <+33>: pop %rbp
> 0x0000555555555161 <+34>: ret
> End of assembler dump.
> (gdb)
>
> And lo, we stopped in the middle of line 11! That is a bug, we should have
> stepped back all the way to the beginning of the line. The "reverse-next"
> should have fully undone the prior "next" command.
>
> The above issues were fixed by introducing a new function that looks for
> adjacent PC ranges for the same line, until we notice a line change. Then
> we take that as the start PC of the range. The new start PC for the range
> is used for the control.step_range_start when setting up a step range.
>
> The test case gdb.reverse/map-to-same-line.exp is added to test the fix
> for the issues in scenario 1.
>
> The test case gdb.reverse/func-map-to-same-line.exp was added to test the
> fix for scenario 2 when the binary was compiled with and without line
> table information.
>
> bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28426
>
> Co-authored-by: Luis Machado <luis.machado@arm.com>
> Co-authored-by: Carl Love <cel@us.ibm.com>
> Reviewed-By: Bruno Larsen <blarsen@redhat.com>
> ---
> gdb/infrun.c | 57 +++++++
> gdb/symtab.c | 49 ++++++
> gdb/symtab.h | 16 ++
> .../gdb.reverse/func-map-to-same-line.c | 36 ++++
> .../gdb.reverse/func-map-to-same-line.exp | 140 ++++++++++++++++
> gdb/testsuite/gdb.reverse/map-to-same-line.c | 58 +++++++
> .../gdb.reverse/map-to-same-line.exp | 156 ++++++++++++++++++
> 7 files changed, 512 insertions(+)
> create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.c
> create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.exp
>
> diff --git a/gdb/infrun.c b/gdb/infrun.c
> index efe2c00c489..31cd817c733 100644
> --- a/gdb/infrun.c
> +++ b/gdb/infrun.c
> @@ -114,6 +114,9 @@ static struct async_event_handler *infrun_async_inferior_event_token;
> Starts off as -1, indicating "never enabled/disabled". */
> static int infrun_is_async = -1;
>
> +static CORE_ADDR update_line_range_start (CORE_ADDR pc,
> + struct execution_control_state *ecs);
> +
This forward-declaration is not needed.
> /* See infrun.h. */
>
> void
> @@ -6769,6 +6772,25 @@ handle_signal_stop (struct execution_control_state *ecs)
> process_event_stop_test (ecs);
> }
>
> +CORE_ADDR
> +update_line_range_start (CORE_ADDR pc, struct execution_control_state *ecs)
Please add a comment for the function.
> +{
> + /* The line table may have multiple entries for the same source code line.
> + Given the PC, check the line table and return the PC that corresponds
> + to the line table entry for the source line that PC is in. */
> + CORE_ADDR start_line_pc = ecs->event_thread->control.step_range_start;
> + gdb::optional<CORE_ADDR> real_range_start;
> +
> + /* Call find_line_range_start to get the smallest address in the
> + linetable for multiple Line X entries in the line table. */
> + real_range_start = find_line_range_start (pc);
> +
> + if (real_range_start.has_value ())
> + start_line_pc = *real_range_start;
> +
> + return start_line_pc;
When I read this, I wonder: why was control.step_range_start not set to
the "real" range start in the first place (not only in the context of
reverse execution, every time it is set)? It would seem more robust
than patching it afterwards in some very specific spots.
I could see some benefits for range-stepping uses cases too (relevant
when debugging remotely). Using your example here:
Line X - [0x0 - 0x8]
Line X - [0x8 - 0x10]
Line X - [0x10 - 0x18]
Imagine we are stopped at 0x14, and we type "next", and 0x14 is a
conditional jump to 0x5. It seems like current GDB would send a "range
step" request to GDBserver, to step in the [0x10, 0x18[ range. When
reaching 0x5, execution would stop, and GDB would resume it again with
the [0x0,0x8[ range. When reaching 0x8, it would stop again, GDB would
resume it with [0x8,0x10[, and so on. If GDB could send a "range step"
request with the [0x0,0x18[ range, it would avoid those unnecessary
intermediary stop.
> +}
> +
> /* Come here when we've got some debug event / signal we can explain
> (IOW, not a random signal), and test whether it should cause a
> stop, or whether we should resume the inferior (transparently).
> @@ -7570,6 +7592,28 @@ process_event_stop_test (struct execution_control_state *ecs)
>
> if (stop_pc_sal.is_stmt)
> {
> + if (execution_direction == EXEC_REVERSE)
> + {
> + /* We are stepping backwards make sure we have reached the
> + beginning of the line. */
> + CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
> + CORE_ADDR start_line_pc
> + = update_line_range_start (stop_pc, ecs);
> +
> + if (stop_pc != start_line_pc)
> + {
> + /* Have not reached the beginning of the source code line.
> + Set a step range. Execution should stop in any function
> + calls we execute back into before reaching the beginning
> + of the line. */
> + ecs->event_thread->control.step_range_start = start_line_pc;
> + ecs->event_thread->control.step_range_end = stop_pc;
> + set_step_info (ecs->event_thread, frame, stop_pc_sal);
> + keep_going (ecs);
> + return;
> + }
> + }
> +
> /* We are at the start of a statement.
>
> So stop. Note that we don't stop if we step into the middle of a
> @@ -7632,6 +7676,19 @@ process_event_stop_test (struct execution_control_state *ecs)
> set_step_info (ecs->event_thread, frame, stop_pc_sal);
>
> infrun_debug_printf ("keep going");
> +
> + if (execution_direction == EXEC_REVERSE)
> + {
> + CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
> +
> + /* Make sure the stop_pc is set to the beginning of the line. */
> + if (stop_pc != ecs->event_thread->control.step_range_start)
> + {
> + stop_pc = update_line_range_start (stop_pc, ecs);
> + ecs->event_thread->control.step_range_start = stop_pc;
> + }
> + }
> +
> keep_going (ecs);
> }
>
> diff --git a/gdb/symtab.c b/gdb/symtab.c
> index 27611a34ec4..91d35616eb9 100644
> --- a/gdb/symtab.c
> +++ b/gdb/symtab.c
> @@ -3282,6 +3282,55 @@ find_pc_line (CORE_ADDR pc, int notcurrent)
> return sal;
> }
>
> +/* Compare two symtab_and_line entries. Return true if both have
> + the same line number and the same symtab pointer. That means we
> + are dealing with two entries from the same line and from the same
> + source file.
> +
> + Return false otherwise. */
> +
> +static bool
> +sal_line_symtab_matches_p (const symtab_and_line &sal1,
> + const symtab_and_line &sal2)
> +{
> + return (sal1.line == sal2.line && sal1.symtab == sal2.symtab);
Unnecessary parenthesis.
> +}
> +
> +/* See symtah.h. */
> +
> +gdb::optional<CORE_ADDR>
> +find_line_range_start (CORE_ADDR pc)
> +{
> + struct symtab_and_line current_sal = find_pc_line (pc, 0);
> +
> + if (current_sal.line == 0)
> + return {};
> +
> + struct symtab_and_line prev_sal = find_pc_line (current_sal.pc - 1, 0);
> +
> + /* If the previous entry is for a different line, that means we are already
> + at the entry with the start PC for this line. */
> + if (!sal_line_symtab_matches_p (prev_sal, current_sal))
> + return current_sal.pc;
> +
> + /* Otherwise, keep looking for entries for the same line but with
> + smaller PC's. */
> + bool done = false;
> + CORE_ADDR prev_pc;
> + while (!done)
> + {
> + prev_pc = prev_sal.pc;
> +
> + prev_sal = find_pc_line (prev_pc - 1, 0);
> +
> + /* Did we notice a line change? If so, we are done with the search. */
> + if (!sal_line_symtab_matches_p (prev_sal, current_sal))
> + done = true;
> + }
> +
> + return prev_pc;
Algorithmic complexity question: given that line tables are sorted by
address, would it work to start at the current line table item, and go
look at the previous ones until we find one that is no longer
contiguous and same line? find_pc_line is somewhat heavy, so if we
don't need to do it repeatedly...
> +}
> +
> /* See symtab.h. */
>
> struct symtab *
> diff --git a/gdb/symtab.h b/gdb/symtab.h
> index 404d0ab30a8..f54305636da 100644
> --- a/gdb/symtab.h
> +++ b/gdb/symtab.h
> @@ -2346,6 +2346,22 @@ extern struct symtab_and_line find_pc_line (CORE_ADDR, int);
> extern struct symtab_and_line find_pc_sect_line (CORE_ADDR,
> struct obj_section *, int);
>
> +/* Given PC, and assuming it is part of a range of addresses that is part of a
> + line, go back through the linetable and find the starting PC of that
> + line.
> +
> + For example, suppose we have 3 PC ranges for line X:
> +
> + Line X - [0x0 - 0x8]
> + Line X - [0x8 - 0x10]
> + Line X - [0x10 - 0x18]
> +
> + If we call the function with PC == 0x14, we want to return 0x0, as that is
> + the starting PC of line X, and the ranges are contiguous.
I think that putting this example in the comment is great. It makes it
much more obvious what the function specifically does.
> +*/
> +
> +extern gdb::optional<CORE_ADDR> find_line_range_start (CORE_ADDR pc);
> +
> /* Wrapper around find_pc_line to just return the symtab. */
>
> extern struct symtab *find_pc_line_symtab (CORE_ADDR);
> diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.c b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> new file mode 100644
> index 00000000000..da944874e86
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> @@ -0,0 +1,36 @@
> +/* Copyright 2008-2023 Free Software Foundation, Inc.
> +
> + This program is free software; you can redistribute it and/or modify
> + it under the terms of the GNU General Public License as published by
> + the Free Software Foundation; either version 3 of the License, or
> + (at your option) any later version.
> +
> + This program is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> + GNU General Public License for more details.
> +
> + You should have received a copy of the GNU General Public License
> + along with this program. If not, see <http://www.gnu.org/licenses/>.
> +
> + This test is used to test the reverse-step and reverse-next instruction
> + execution for a source line that contains multiple function calls. */
> +
> +void
> +func1 ()
> +{
> +} /* END FUNC1 */
> +
> +void
> +func2 ()
> +{
> +} /* END FUNC2 */
> +
> +int main ()
int
main (void)
> +{
> + int a, b;
> + a = 1;
> + b = 2;
> + func1 (); func2 ();
> + a = a + b; /* START REVERSE TEST */
> +}
> diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> new file mode 100644
> index 00000000000..89e226b0f84
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> @@ -0,0 +1,140 @@
> +# Copyright 2008-2023 Free Software Foundation, Inc.
> +
> +# This program is free software; you can redistribute it and/or modify
> +# it under the terms of the GNU General Public License as published by
> +# the Free Software Foundation; either version 3 of the License, or
> +# (at your option) any later version.
> +#
> +# This program is distributed in the hope that it will be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> +# GNU General Public License for more details.
> +#
> +# You should have received a copy of the GNU General Public License
> +# along with this program. If not, see <http://www.gnu.org/licenses/>. */
> +
> +# This file is part of the GDB testsuite. It tests reverse stepping.
> +# Lots of code borrowed from "step-test.exp".
> +
> +# This test checks to make sure there is no regression failures for
> +# the reverse-next command when stepping back over two functions in
> +# the same line.
> +
> +require supports_reverse
> +
> +# This test uses the gcc no-column-info command which was added in gcc 7.1.
> +
> +proc run_tests {} {
> + global srcfile
> + global executable
> +
> + runto_main
We typically check for runto_main's success:
if { ![runto_main] } {
return
}
runto_main logs a FAIL on failure. There are a few runto_mains in the
patch.
> + set target_remote [gdb_is_target_remote]
target_remote seems unused
> +
> + with_test_prefix "test1" {
> + gdb_test_no_output "record" "turn on process record"
> + }
with_test_prefix with a single test in it is really just the same as:
gdb_test_no_output "record" "test1: turn on process record"
In fact, you have some other tests with the "test1:" or "test2:" prefix,
I think they should be moved to the with_test_prefix. And maybe use
"next" and "step" instead of "test1" and "test2".
> +
> + # This regression test verifies the reverse-step and reverse-next commands
> + # work properly when executing backwards thru a source line containing
> + # two function calls on the same source line, i.e. func1 (); func2 ();
> + # This test is compiled so the dwarf info not contain the line table
> + # information.
> +
> + # Test 1, reverse-next command
> + # Set breakpoint at the line after the function calls.
> + set bp_start_reverse_test [gdb_get_line_number "START REVERSE TEST" \
> + $srcfile]
> + gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> +
> + # Continue to break point for reverse-next test.
> + # Command definition: reverse-next [count]
> + # Run backward to the beginning of the previous line executed in the
> + # current (innermost) stack frame. If the line contains function calls,
> + # they will be “un-executed” without stopping. Starting from the first
> + # line of a function, reverse-next will take you back to the caller of
> + # that function, before the function was called, just as the normal next
> + # command would take you from the last line of a function back to its
> + # return to its caller 2 .
> + gdb_continue_to_breakpoint \
> + "test1: stopped at command reverse-next test start location" \
> + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> +
> + # The reverse-next should step all the way back to the beginning of the
> + # line, i.e. at the beginning of the func1 call.
> + gdb_test "reverse-next" ".*func1 \\(\\); func2 \\(\\);.*" \
> + "test1: reverse-next to line with two functions"
> +
> + # We should be stopped at the first instruction of the line. A reverse-step
> + # should step back and stop at the beginning of the previous line b = 2,
> + # i.e. not in func1 ().
> + gdb_test "reverse-stepi" ".*b = 2;.*" \
> + "test1: reverse-stepi to previous line b = 2"
> +
> +
> + # Setup for test 2
> + clean_restart $executable
> + runto_main
> +
> + with_test_prefix "test2" {
> + gdb_test_no_output "record" "turn on process record"
> + }
> +
> + # Test 2, reverse-step command
> + # Set breakpoint at the line after the function calls.
> + gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> +
> + # Continue to the start of the reverse-step test.
> + # Command definition: reverse-step [count]
> + # Run the program backward until control reaches the start of a
> + # different source line; then stop it, and return control to gdb.
> + # Like the step command, reverse-step will only stop at the beginning
> + # of a source line. It “un-executes” the previously executed source
> + # line. If the previous source line included calls to debuggable
> + # functions, reverse-step will step (backward) into the called function,
> + # stopping at the beginning of the last statement in the called
> + # function (typically a return statement). Also, as with the step
> + # command, if non-debuggable functions are called, reverse-step will
> + # run thru them backward without stopping.
> +
> + gdb_continue_to_breakpoint \
> + "test2: stopped at command reverse-step test start location" \
> + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> +
> + # The first reverse step should take us call of func2 ().
> + gdb_test "reverse-step" ".*END FUNC2.*" \
> + "test2: reverse-step into func2 "
> +
> + # The second reverse step should take us into func1 ().
> + gdb_test "reverse-step" ".*END FUNC1.*" \
> + "test2: reverse-step into func1 "
> +
> + # The third reverse step should take us call of func1 ().
> + gdb_test "reverse-step" ".*func1 \\(\\); func2 \\(\\);.*" \
> + "test2: reverse-step to line func1(); func2(), at call for func1 "
> +
> + # We should be stopped at the first instruction of the line. A reverse
> + # stepi should take us to b = 2 ().
> + gdb_test "reverse-stepi" ".*b = 2;.*" \
> + "test2: reverse-stepi to line b = 2 "
> +}
> +
> +set srcfile func-map-to-same-line.c
> +set executable func-map-to-same-line
Wondering if this test should use standard_testfile (like almost every
other tests) to set these.
> +
> +# test with and without gcc column info enabled
> +foreach_with_prefix with_column_info {yes no} {
> + if {$with_column_info == "yes"} {
> + set options [list debug column-info]
> + } else {
> + set options [list debug no-column-info]
> + }
I didn't think of this when proposing the foreach_with_prefix, but you
could perhaps use:
foreach_with_prefix column_info_flag {column-info no-column-info}
... to avoid this boilerplate. You can then use $column_info_flag
directly when setting options.
> +
> + if {[build_executable "failed to prepare" $executable $srcfile \
> + $options] == -1} {
> + return -1
> + }
> +
> + clean_restart $executable
clean_restart can go in run_tests.
> + run_tests
> +}
> diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.c b/gdb/testsuite/gdb.reverse/map-to-same-line.c
> new file mode 100644
> index 00000000000..f20d778f40e
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/map-to-same-line.c
> @@ -0,0 +1,58 @@
> +/* Copyright 2008-2023 Free Software Foundation, Inc.
Just wondering if the copyright years are right.
> +
> + This program is free software; you can redistribute it and/or modify
> + it under the terms of the GNU General Public License as published by
> + the Free Software Foundation; either version 3 of the License, or
> + (at your option) any later version.
> +
> + This program is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> + GNU General Public License for more details.
> +
> + You should have received a copy of the GNU General Public License
> + along with this program. If not, see <http://www.gnu.org/licenses/ >. */
> +
> +/* The purpose of this test is to create a DWARF line table that contains two
> + or more entries for the same line. When stepping (forwards or backwards),
> + GDB should step over the entire line and not just a particular entry in the
> + line table. */
> +
> +int
> +main ()
void in the parenthesis
> +{ /* TAG: main prologue */
> + asm ("main_label: .globl main_label");
> + int i = 1, j = 2, k;
> + float f1 = 2.0, f2 = 4.1, f3;
> + const char *str_1 = "foo", *str_2 = "bar", *str_3;
> +
> + asm ("line1: .globl line1");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 1 */
> +
> + asm ("line2: .globl line2");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 2 */
> +
> + asm ("line3: .globl line3");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 3 */
> +
> + asm ("line4: .globl line4");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 4 */
> +
> + asm ("line5: .globl line5");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 5 */
> +
> + asm ("line6: .globl line6");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 6 */
> +
> + asm ("line7: .globl line7");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 7 */
> +
> + asm ("line8: .globl line8");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 8 */
> +
> + asm ("main_return: .globl main_return");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: main return */
> +
> + asm ("end_of_sequence: .globl end_of_sequence");
> + return 0; /* TAG: main return */
> +}
> diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.exp b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> new file mode 100644
> index 00000000000..16a359d90ec
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> @@ -0,0 +1,156 @@
> +# Copyright 2008-2023 Free Software Foundation, Inc.
> +
> +# This program is free software; you can redistribute it and/or modify
> +# it under the terms of the GNU General Public License as published by
> +# the Free Software Foundation; either version 3 of the License, or
> +# (at your option) any later version.
> +#
> +# This program is distributed in the hope that it will be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> +# GNU General Public License for more details.
> +#
> +# You should have received a copy of the GNU General Public License
> +# along with this program. If not, see <http://www.gnu.org/licenses/ >.
> +
> +# When stepping (forwards or backwards), GDB should step over the entire line
> +# and not just a particular entry in the line table. This test was added to
> +# verify the find_line_range_start function properly sets the step range for a
> +# line that consists of multiple statements, i.e. multiple entries in the line
> +# table. This test creates a DWARF line table that contains two entries for
> +# the same line to do the needed testing.
> +
> +# This test can only be run on targets which support DWARF-2 and use gas.
> +load_lib dwarf.exp
> +require dwarf2_support
> +
> +# The DWARF assembler requires the gcc compiler.
> +require is_c_compiler_gcc
> +
> +# This test suitable only for process that can do reverse execution
> +require supports_reverse
> +
> +standard_testfile .c .S
> +
> +if { [prepare_for_testing "failed to prepare" ${testfile} ${srcfile}] } {
> + return -1
> +}
> +
> +set asm_file [standard_output_file $srcfile2]
> +Dwarf::assemble $asm_file {
> + global srcdir subdir srcfile
> + declare_labels integer_label L
> +
> + # Find start address and length of program
> + lassign [function_range main [list ${srcdir}/${subdir}/$srcfile]] \
> + main_start main_len
> + set main_end "$main_start + $main_len"
> +
> + cu {} {
> + compile_unit {
> + {language @DW_LANG_C}
> + {name map-to-same-line.c}
> + {stmt_list $L DW_FORM_sec_offset}
> + {low_pc 0 addr}
> + } {
> + subprogram {
> + {external 1 flag}
> + {name main}
> + {low_pc $main_start addr}
> + {high_pc $main_len DW_FORM_data4}
> + }
> + }
> + }
> +
> + lines {version 2 default_is_stmt 1} L {
> + include_dir "${srcdir}/${subdir}"
> + file_name "$srcfile" 1
> +
> + # Generate the line table program with distinct source lines being
> + # mapped to the same line entry. Line 1, 5 and 8 contain 1 statement
> + # each. Line 2 contains 2 statements. Line 3 contains 3 statements.
> + program {
> + DW_LNE_set_address $main_start
> + line [gdb_get_line_number "TAG: main prologue"]
> + DW_LNS_copy
> + DW_LNE_set_address line1
> + line [gdb_get_line_number "TAG: line 1" ]
> + DW_LNS_copy
> + DW_LNE_set_address line2
> + line [gdb_get_line_number "TAG: line 2" ]
> + DW_LNS_copy
> + DW_LNE_set_address line3
> + line [gdb_get_line_number "TAG: line 2" ]
> + DW_LNS_copy
> + DW_LNE_set_address line4
> + line [gdb_get_line_number "TAG: line 3" ]
> + DW_LNS_copy
> + DW_LNE_set_address line5
> + line [gdb_get_line_number "TAG: line 3" ]
> + DW_LNS_copy
> + DW_LNE_set_address line6
> + line [gdb_get_line_number "TAG: line 3" ]
> + DW_LNS_copy
> + DW_LNE_set_address line7
> + line [gdb_get_line_number "TAG: line 5" ]
> + DW_LNS_copy
> + DW_LNE_set_address line8
> + line [gdb_get_line_number "TAG: line 8" ]
> + DW_LNS_copy
> + DW_LNE_set_address main_return
> + line [gdb_get_line_number "TAG: main return"]
> + DW_LNS_copy
> + DW_LNE_set_address end_of_sequence
> + DW_LNE_end_sequence
> + }
> + }
> +}
> +
> +if { [prepare_for_testing "failed to prepare" ${testfile} \
> + [list $srcfile $asm_file] {nodebug} ] } {
> + return -1
> +}
> +
> +runto_main
> +
> +# Print the line table
> +gdb_test_multiple "maint info line-table ${testfile}" "" {
> + -re "\r\n$decimal\[ \t\]+$decimal\[ \t\]+($hex)\[ \t\]+Y\[^\r\n\]*" {
> + lappend is_stmt $expect_out(1,string)
> + exp_continue
> + }
> + -re -wrap "" {
> + }
> +}
> +
> +# Do the reverse-step test
> +gdb_test_no_output "record" "turn on process record"
> +
> +set bp_main_return [gdb_get_line_number "TAG: main return" $srcfile]
> +gdb_breakpoint $srcfile:$bp_main_return
> +gdb_continue_to_breakpoint "run to end of main, reverse-step test" ".*$srcfile:$bp_main_return.*"
> +gdb_test "display \$pc" ".*pc =.*" "display pc, reverse-step test"
> +
> +# At this point, GDB has already recorded the execution up until the return
> +# statement. Reverse-step and test if GDB transitions between lines in the
> +# expected order. It should reverse-step across lines 8, 5, 3, 2 and 1.
> +foreach line {8 5 3 2 1} {
> + gdb_test "reverse-step" ".*TAG: line $line.*" "reverse step to line $line"
> +}
> +
> +## Clean restart, test reverse-next command
> +clean_restart ${testfile}
> +runto_main
> +gdb_test_no_output "record" "turn on process record, reverst-next test"
> +
> +set bp_main_return [gdb_get_line_number "TAG: main return" $srcfile]
> +gdb_breakpoint $srcfile:$bp_main_return
> +gdb_continue_to_breakpoint "run to end of main, reverse-next test" ".*$srcfile:$bp_main_return.*"
> +gdb_test "display \$pc" ".*pc =.*" "display pc, reverse-next test"
> +
> +# At this point, GDB has already recorded the execution up until the return
> +# statement. Reverse-next and test if GDB transitions between lines in the
> +# expected order. It should reverse-next across lines 8, 5, 3, 2 and 1.
> +foreach line {8 5 3 2 1} {
> + gdb_test "reverse-next" ".*TAG: line $line.*" "reverse next to line $line"
> +}
It seems like the step and next tests are identical, so I guess it could
be factored out using:
foreach_with_prefix method {step next} {
...
}
?
Simon
^ permalink raw reply [flat|nested] 40+ messages in thread
* Re: [PATCH 1/2] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-06-19 17:11 ` Simon Marchi
@ 2023-06-22 16:52 ` Carl Love
2023-06-23 17:44 ` Simon Marchi
0 siblings, 1 reply; 40+ messages in thread
From: Carl Love @ 2023-06-22 16:52 UTC (permalink / raw)
To: Simon Marchi, Bruno Larsen, gdb-patches, UlrichWeigand, pedro
Cc: luis.machado, cel
Simon:
On Mon, 2023-06-19 at 13:11 -0400, Simon Marchi wrote:
> > --- a/gdb/testsuite/lib/gdb.exp
> > +++ b/gdb/testsuite/lib/gdb.exp
> > @@ -4794,6 +4794,8 @@ proc quote_for_host { args } {
> > # debug information
> > # - text_segment=addr: Tell the linker to place the text
> > segment at ADDR.
> > # - build-id: Ensure the final binary includes a build-id.
> > +# - no-column-info: Disable generation of column table
> > information.
> > +# - column-info: Enable generation of column table information.
> > #
> > # And here are some of the not too obscure options understood by
> > DejaGnu that
> > # influence the compilation:
> > @@ -5003,6 +5005,34 @@ proc gdb_compile {source dest type options}
> > {
> > } else {
> > error "Don't know how to handle text_segment
> > option."
> > }
> > + } elseif { $opt == "column-info" } {
> > + if {[test_compiler_info {gcc-*}]} {
> > + if {[test_compiler_info {gcc-[1-6]-*}]} {
> > + error "gdb_compile option no-column-info not
> > supported."
>
> I think this path should return the equivalent of "failed to
> compile",
> instead of throwing an error. Control will go back to the test,
> which
> will generally skip the portion of the test that requires that
> binary.
Not entirely sure how to accomplish what you are looking for.
I change:
error "gdb_compile option no-column-info not supported."
to
set result "option no-column-info not supported."
clone_output "gdb compile failed, $result"
return 1
When I force the if {[test_compiler_info...]} tp be true to test this,
I get:
get_compiler_info: gcc-12-2-1
gdb compile failed, option no-column-info not supported.
UNTESTED: gdb.reverse/func-map-to-same-line.exp:
with_column_info=yes: failed t\
o prepare
testcase /home/carll/GDB/build-reverse-multiple-
contiguous/gdb/testsuite/../../\
../binutils-gdb-reverse-multiple-
contiguous/gdb/testsuite/gdb.reverse/func-map-\
to-same-line.exp completed in 0 seconds
=== gdb Summary ===
# of untested testcases 1
The test case doesn't have any part of the test that doesn't require
compiling so it is not clear if that would work with this fix. Anyway,
wanted to run that by you to see if this is an appropriate fix? I am
really not sure about it. Thanks.
Carl
^ permalink raw reply [flat|nested] 40+ messages in thread
* RE: [PATCH 2/2 v5] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-06-19 17:58 ` Simon Marchi
@ 2023-06-22 20:38 ` Carl Love
2023-06-22 20:39 ` Carl Love
2023-06-23 17:49 ` Simon Marchi
2023-06-23 20:04 ` Carl Love
2023-06-23 20:04 ` [PATCH 2/2 v6] " Carl Love
2 siblings, 2 replies; 40+ messages in thread
From: Carl Love @ 2023-06-22 20:38 UTC (permalink / raw)
To: Simon Marchi, Bruno Larsen, gdb-patches, UlrichWeigand, pedro
Cc: luis.machado
Simon:
On Mon, 2023-06-19 at 13:58 -0400, Simon Marchi wrote:
<snip>
>
100644
> > --- a/gdb/infrun.c
> > +++ b/gdb/infrun.c
> > @@ -114,6 +114,9 @@ static struct async_event_handler
> > *infrun_async_inferior_event_token;
> > Starts off as -1, indicating "never enabled/disabled". */
> > static int infrun_is_async = -1;
> >
> > +static CORE_ADDR update_line_range_start (CORE_ADDR pc,
> > + struct
> > execution_control_state *ecs);
> > +
>
> This forward-declaration is not needed.
I tried removing the forward-declaration and the compile fails with the
message:
../../binutils-gdb-reverse-multiple-contiguous/gdb/infrun.c:6773:1:
error: no previous declaration for ‘CORE_ADDR
update_line_range_start(CORE_ADDR, execution_control_state*)’ [-
Werror=missing-declarations]
6773 | update_line_range_start (CORE_ADDR pc, struct
execution_control_state *ecs)
| ^~~~~~~~~~~~~~~~~~~~~~~
cc1plus: all warnings being treated as errors
make[2]: *** [Makefile:1922: infrun.o] Error 1
make[2]: Leaving directory '/home/carll/GDB/build-reverse-multiple-
contiguous/gdb'
make[1]: *** [Makefile:13569: all-gdb] Error 2
make[1]: Leaving directory '/home/carll/GDB/build-reverse-multiple-
contiguous'
make: *** [Makefile:1005: all] Error 2
Leaving the forward declaration in the code.
>
> > /* See infrun.h. */
> >
> > void
> > @@ -6769,6 +6772,25 @@ handle_signal_stop (struct
> > execution_control_state *ecs)
> > process_event_stop_test (ecs);
> > }
> >
> > +CORE_ADDR
> > +update_line_range_start (CORE_ADDR pc, struct
> > execution_control_state *ecs)
>
> Please add a comment for the function.
Done.
>
> > +{
> > + /* The line table may have multiple entries for the same source
> > code line.
> > + Given the PC, check the line table and return the PC that
> > corresponds
> > + to the line table entry for the source line that PC is
> > in. */
> > + CORE_ADDR start_line_pc = ecs->event_thread-
> > >control.step_range_start;
> > + gdb::optional<CORE_ADDR> real_range_start;
> > +
> > + /* Call find_line_range_start to get the smallest address in the
> > + linetable for multiple Line X entries in the line table. */
> > + real_range_start = find_line_range_start (pc);
> > +
> > + if (real_range_start.has_value ())
> > + start_line_pc = *real_range_start;
> > +
> > + return start_line_pc;
>
> When I read this, I wonder: why was control.step_range_start not set
> to
> the "real" range start in the first place (not only in the context of
> reverse execution, every time it is set)? It would seem more robust
> than patching it afterwards in some very specific spots.
>
> I could see some benefits for range-stepping uses cases too (relevant
> when debugging remotely). Using your example here:
>
> Line X - [0x0 - 0x8]
> Line X - [0x8 - 0x10]
> Line X - [0x10 - 0x18]
>
> Imagine we are stopped at 0x14, and we type "next", and 0x14 is a
> conditional jump to 0x5. It seems like current GDB would send a
> "range
> step" request to GDBserver, to step in the [0x10, 0x18[ range. When
> reaching 0x5, execution would stop, and GDB would resume it again
> with
> the [0x0,0x8[ range. When reaching 0x8, it would stop again, GDB
> would
> resume it with [0x8,0x10[, and so on. If GDB could send a "range
> step"
> request with the [0x0,0x18[ range, it would avoid those unnecessary
> intermediary stop.
>
> > +}
> > +
> > /* Come here when we've got some debug event / signal we can
> > explain
> > (IOW, not a random signal), and test whether it should cause a
> > stop, or whether we should resume the inferior (transparently).
> > @@ -7570,6 +7592,28 @@ process_event_stop_test (struct
> > execution_control_state *ecs)
> >
> > if (stop_pc_sal.is_stmt)
> > {
> > + if (execution_direction == EXEC_REVERSE)
> > + {
> > + /* We are stepping backwards make sure we have reached
> > the
> > + beginning of the line. */
> > + CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
> > + CORE_ADDR start_line_pc
> > + = update_line_range_start (stop_pc, ecs);
> > +
> > + if (stop_pc != start_line_pc)
> > + {
> > + /* Have not reached the beginning of the source code
> > line.
> > + Set a step range. Execution should stop in any
> > function
> > + calls we execute back into before reaching the
> > beginning
> > + of the line. */
> > + ecs->event_thread->control.step_range_start =
> > start_line_pc;
> > + ecs->event_thread->control.step_range_end = stop_pc;
> > + set_step_info (ecs->event_thread, frame,
> > stop_pc_sal);
> > + keep_going (ecs);
> > + return;
> > + }
> > + }
> > +
> > /* We are at the start of a statement.
> >
> > So stop. Note that we don't stop if we step into the
> > middle of a
> > @@ -7632,6 +7676,19 @@ process_event_stop_test (struct
> > execution_control_state *ecs)
> > set_step_info (ecs->event_thread, frame, stop_pc_sal);
> >
> > infrun_debug_printf ("keep going");
> > +
> > + if (execution_direction == EXEC_REVERSE)
> > + {
> > + CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
> > +
> > + /* Make sure the stop_pc is set to the beginning of the
> > line. */
> > + if (stop_pc != ecs->event_thread->control.step_range_start)
> > + {
> > + stop_pc = update_line_range_start (stop_pc, ecs);
> > + ecs->event_thread->control.step_range_start = stop_pc;
> > + }
> > + }
> > +
> > keep_going (ecs);
> > }
> >
> > diff --git a/gdb/symtab.c b/gdb/symtab.c
> > index 27611a34ec4..91d35616eb9 100644
> > --- a/gdb/symtab.c
> > +++ b/gdb/symtab.c
> > @@ -3282,6 +3282,55 @@ find_pc_line (CORE_ADDR pc, int notcurrent)
> > return sal;
> > }
> >
> > +/* Compare two symtab_and_line entries. Return true if both have
> > + the same line number and the same symtab pointer. That means
> > we
> > + are dealing with two entries from the same line and from the
> > same
> > + source file.
> > +
> > + Return false otherwise. */
> > +
> > +static bool
> > +sal_line_symtab_matches_p (const symtab_and_line &sal1,
> > + const symtab_and_line &sal2)
> > +{
> > + return (sal1.line == sal2.line && sal1.symtab == sal2.symtab);
>
> Unnecessary parenthesis.
Removed unnecessary parenthesis.
>
> > +}
> > +
> > +/* See symtah.h. */
> > +
> > +gdb::optional<CORE_ADDR>
> > +find_line_range_start (CORE_ADDR pc)
> > +{
> > + struct symtab_and_line current_sal = find_pc_line (pc, 0);
> > +
> > + if (current_sal.line == 0)
> > + return {};
> > +
> > + struct symtab_and_line prev_sal = find_pc_line (current_sal.pc -
> > 1, 0);
> > +
> > + /* If the previous entry is for a different line, that means we
> > are already
> > + at the entry with the start PC for this line. */
> > + if (!sal_line_symtab_matches_p (prev_sal, current_sal))
> > + return current_sal.pc;
> > +
> > + /* Otherwise, keep looking for entries for the same line but
> > with
> > + smaller PC's. */
> > + bool done = false;
> > + CORE_ADDR prev_pc;
> > + while (!done)
> > + {
> > + prev_pc = prev_sal.pc;
> > +
> > + prev_sal = find_pc_line (prev_pc - 1, 0);
> > +
> > + /* Did we notice a line change? If so, we are done with the
> > search. */
> > + if (!sal_line_symtab_matches_p (prev_sal, current_sal))
> > + done = true;
> > + }
> > +
> > + return prev_pc;
>
> Algorithmic complexity question: given that line tables are sorted by
> address, would it work to start at the current line table item, and
> go
> look at the previous ones until we find one that is no longer
> contiguous and same line? find_pc_line is somewhat heavy, so if we
> don't need to do it repeatedly...
>
> > +}
> > +
> > /* See symtab.h. */
> >
> > struct symtab *
> > diff --git a/gdb/symtab.h b/gdb/symtab.h
> > index 404d0ab30a8..f54305636da 100644
> > --- a/gdb/symtab.h
> > +++ b/gdb/symtab.h
> > @@ -2346,6 +2346,22 @@ extern struct symtab_and_line find_pc_line
> > (CORE_ADDR, int);
> > extern struct symtab_and_line find_pc_sect_line (CORE_ADDR,
> > struct obj_section *,
> > int);
> >
> > +/* Given PC, and assuming it is part of a range of addresses that
> > is part of a
> > + line, go back through the linetable and find the starting PC of
> > that
> > + line.
> > +
> > + For example, suppose we have 3 PC ranges for line X:
> > +
> > + Line X - [0x0 - 0x8]
> > + Line X - [0x8 - 0x10]
> > + Line X - [0x10 - 0x18]
> > +
> > + If we call the function with PC == 0x14, we want to return 0x0,
> > as that is
> > + the starting PC of line X, and the ranges are contiguous.
>
> I think that putting this example in the comment is great. It makes
> it
> much more obvious what the function specifically does.
>
> > +*/
> > +
> > +extern gdb::optional<CORE_ADDR> find_line_range_start (CORE_ADDR
> > pc);
> > +
> > /* Wrapper around find_pc_line to just return the symtab. */
> >
> > extern struct symtab *find_pc_line_symtab (CORE_ADDR);
> > diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> > b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> > new file mode 100644
> > index 00000000000..da944874e86
> > --- /dev/null
> > +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> > @@ -0,0 +1,36 @@
> > +/* Copyright 2008-2023 Free Software Foundation, Inc.
> > +
> > + This program is free software; you can redistribute it and/or
> > modify
> > + it under the terms of the GNU General Public License as
> > published by
> > + the Free Software Foundation; either version 3 of the License,
> > or
> > + (at your option) any later version.
> > +
> > + This program is distributed in the hope that it will be useful,
> > + but WITHOUT ANY WARRANTY; without even the implied warranty of
> > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> > + GNU General Public License for more details.
> > +
> > + You should have received a copy of the GNU General Public
> > License
> > + along with this program. If not, see <
> > http://www.gnu.org/licenses/
> > >.
> > +
> > + This test is used to test the reverse-step and reverse-next
> > instruction
> > + execution for a source line that contains multiple function
> > calls. */
> > +
> > +void
> > +func1 ()
> > +{
> > +} /* END FUNC1 */
> > +
> > +void
> > +func2 ()
> > +{
> > +} /* END FUNC2 */
> > +
> > +int main ()
>
> int
> main (void)
>
Fixed.
> > +{
> > + int a, b;
> > + a = 1;
> > + b = 2;
> > + func1 (); func2 ();
> > + a = a + b; /* START REVERSE TEST */
> > +}
> > diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> > b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> > new file mode 100644
> > index 00000000000..89e226b0f84
> > --- /dev/null
> > +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> > @@ -0,0 +1,140 @@
> > +# Copyright 2008-2023 Free Software Foundation, Inc.
> > +
> > +# This program is free software; you can redistribute it and/or
> > modify
> > +# it under the terms of the GNU General Public License as
> > published by
> > +# the Free Software Foundation; either version 3 of the License,
> > or
> > +# (at your option) any later version.
> > +#
> > +# This program is distributed in the hope that it will be useful,
> > +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> > +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> > +# GNU General Public License for more details.
> > +#
> > +# You should have received a copy of the GNU General Public
> > License
> > +# along with this program. If not, see <
> > http://www.gnu.org/licenses/
> > >. */
> > +
> > +# This file is part of the GDB testsuite. It tests reverse
> > stepping.
> > +# Lots of code borrowed from "step-test.exp".
> > +
> > +# This test checks to make sure there is no regression failures
> > for
> > +# the reverse-next command when stepping back over two functions
> > in
> > +# the same line.
> > +
> > +require supports_reverse
> > +
> > +# This test uses the gcc no-column-info command which was added in
> > gcc 7.1.
> > +
> > +proc run_tests {} {
> > + global srcfile
> > + global executable
> > +
> > + runto_main
>
> We typically check for runto_main's success:
Fixed two instances of runto_main in this test case and two in the
other test case.
>
> if { ![runto_main] } {
> return
> }
>
> runto_main logs a FAIL on failure. There are a few runto_mains in
> the
> patch.
>
> > + set target_remote [gdb_is_target_remote]
>
> target_remote seems unused
Removed.
>
> > +
> > + with_test_prefix "test1" {
> > + gdb_test_no_output "record" "turn on process record"
> > + }
>
> with_test_prefix with a single test in it is really just the same as:
>
> gdb_test_no_output "record" "test1: turn on process record"
>
> In fact, you have some other tests with the "test1:" or "test2:"
> prefix,
> I think they should be moved to the with_test_prefix. And maybe use
> "next" and "step" instead of "test1" and "test2".
Yup, cleaner to have the with_test_prefix cover the whole test.
Changed test1 to next-test and test2 to step-next.
>
> > +
> > + # This regression test verifies the reverse-step and reverse-
> > next commands
> > + # work properly when executing backwards thru a source line
> > containing
> > + # two function calls on the same source line, i.e. func1 ();
> > func2 ();
> > + # This test is compiled so the dwarf info not contain the line
> > table
> > + # information.
> > +
> > + # Test 1, reverse-next command
> > + # Set breakpoint at the line after the function calls.
> > + set bp_start_reverse_test [gdb_get_line_number "START REVERSE
> > TEST" \
> > + $srcfile]
> > + gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> > +
> > + # Continue to break point for reverse-next test.
> > + # Command definition: reverse-next [count]
> > + # Run backward to the beginning of the previous line
> > executed in the
> > + # current (innermost) stack frame. If the line contains
> > function calls,
> > + # they will be “un-executed” without stopping. Starting from
> > the first
> > + # line of a function, reverse-next will take you back to the
> > caller of
> > + # that function, before the function was called, just as the
> > normal next
> > + # command would take you from the last line of a function
> > back to its
> > + # return to its caller 2 .
> > + gdb_continue_to_breakpoint \
> > + "test1: stopped at command reverse-next test start location" \
> > + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> > +
> > + # The reverse-next should step all the way back to the
> > beginning of the
> > + # line, i.e. at the beginning of the func1 call.
> > + gdb_test "reverse-next" ".*func1 \\(\\); func2 \\(\\);.*" \
> > + "test1: reverse-next to line with two functions"
> > +
> > + # We should be stopped at the first instruction of the line. A
> > reverse-step
> > + # should step back and stop at the beginning of the previous
> > line b = 2,
> > + # i.e. not in func1 ().
> > + gdb_test "reverse-stepi" ".*b = 2;.*" \
> > + "test1: reverse-stepi to previous line b = 2"
> > +
> > +
> > + # Setup for test 2
> > + clean_restart $executable
> > + runto_main
> > +
> > + with_test_prefix "test2" {
> > + gdb_test_no_output "record" "turn on process record"
> > + }
> > +
> > + # Test 2, reverse-step command
> > + # Set breakpoint at the line after the function calls.
> > + gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> > +
> > + # Continue to the start of the reverse-step test.
> > + # Command definition: reverse-step [count]
> > + # Run the program backward until control reaches the start
> > of a
> > + # different source line; then stop it, and return control
> > to gdb.
> > + # Like the step command, reverse-step will only stop at the
> > beginning
> > + # of a source line. It “un-executes” the previously
> > executed source
> > + # line. If the previous source line included calls to
> > debuggable
> > + # functions, reverse-step will step (backward) into the
> > called function,
> > + # stopping at the beginning of the last statement in the
> > called
> > + # function (typically a return statement). Also, as with
> > the step
> > + # command, if non-debuggable functions are called, reverse-
> > step will
> > + # run thru them backward without stopping.
> > +
> > + gdb_continue_to_breakpoint \
> > + "test2: stopped at command reverse-step test start location" \
> > + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> > +
> > + # The first reverse step should take us call of func2 ().
> > + gdb_test "reverse-step" ".*END FUNC2.*" \
> > + "test2: reverse-step into func2 "
> > +
> > + # The second reverse step should take us into func1 ().
> > + gdb_test "reverse-step" ".*END FUNC1.*" \
> > + "test2: reverse-step into func1 "
> > +
> > + # The third reverse step should take us call of func1 ().
> > + gdb_test "reverse-step" ".*func1 \\(\\); func2 \\(\\);.*" \
> > + "test2: reverse-step to line func1(); func2(), at call for
> > func1 "
> > +
> > + # We should be stopped at the first instruction of the line. A
> > reverse
> > + # stepi should take us to b = 2 ().
> > + gdb_test "reverse-stepi" ".*b = 2;.*" \
> > + "test2: reverse-stepi to line b = 2 "
> > +}
> > +
> > +set srcfile func-map-to-same-line.c
> > +set executable func-map-to-same-line
>
> Wondering if this test should use standard_testfile (like almost
> every
> other tests) to set these.
OK, changed to use the standard_testfile.
>
> > +
> > +# test with and without gcc column info enabled
> > +foreach_with_prefix with_column_info {yes no} {
> > + if {$with_column_info == "yes"} {
> > + set options [list debug column-info]
> > + } else {
> > + set options [list debug no-column-info]
> > + }
>
> I didn't think of this when proposing the foreach_with_prefix, but
> you
> could perhaps use:
>
> foreach_with_prefix column_info_flag {column-info no-column-info}
>
> ... to avoid this boilerplate. You can then use $column_info_flag
> directly when setting options.
>
OK, that cleans things up a bit. Changed.
+
+ if {[build_executable "failed to prepare" $executable $srcfile
\
+ $options] == -1} {
+ return -1
+ }
+
+ clean_restart $executable
clean_restart can go in run_tests.
+ run_tests
+}
diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.c
b/gdb/testsuite/gdb.reverse/map-to-same-line.c
new file mode 100644
index 00000000000..f20d778f40e
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/map-to-same-line.c
@@ -0,0 +1,58 @@
+/* Copyright 2008-2023 Free Software Foundation, Inc.
Just wondering if the copyright years are right.
New files so yea, should just start with 2023.
>
+
+ This program is free software; you can redistribute it and/or
modify
+ it under the terms of the GNU General Public License as
published by
+ the Free Software Foundation; either version 3 of the License,
or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public
License
+ along with this program. If not, see <
http://www.gnu.org/licenses/
>. */
+
+/* The purpose of this test is to create a DWARF line table that
contains two
+ or more entries for the same line. When stepping (forwards or
backwards),
+ GDB should step over the entire line and not just a particular
entry in the
+ line table. */
+
+int
+main ()
void in the parenthesis
Fixed in both test files.
>
+{ /* TAG: main prologue */
+ asm ("main_label: .globl main_label");
+ int i = 1, j = 2, k;
+ float f1 = 2.0, f2 = 4.1, f3;
+ const char *str_1 = "foo", *str_2 = "bar", *str_3;
+
+ asm ("line1: .globl line1");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 1 */
+
+ asm ("line2: .globl line2");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 2 */
+
+ asm ("line3: .globl line3");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 3 */
+
+ asm ("line4: .globl line4");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 4 */
+
+ asm ("line5: .globl line5");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 5 */
+
+ asm ("line6: .globl line6");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 6 */
+
+ asm ("line7: .globl line7");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 7 */
+
+ asm ("line8: .globl line8");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 8 */
+
+ asm ("main_return: .globl main_return");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: main return */
+
+ asm ("end_of_sequence: .globl end_of_sequence");
+ return 0; /* TAG: main return */
+}
diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.exp
b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
new file mode 100644
index 00000000000..16a359d90ec
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
@@ -0,0 +1,156 @@
+# Copyright 2008-2023 Free Software Foundation, Inc.
+
+# This program is free software; you can redistribute it and/or
modify
+# it under the terms of the GNU General Public License as
published by
+# the Free Software Foundation; either version 3 of the License,
or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public
License
+# along with this program. If not, see <
http://www.gnu.org/licenses/
= >.
+
+# When stepping (forwards or backwards), GDB should step over the
entire line
+# and not just a particular entry in the line table. This test was
added to
+# verify the find_line_range_start function properly sets the step
range for a
+# line that consists of multiple statements, i.e. multiple entries
in the line
+# table. This test creates a DWARF line table that contains two
entries for
+# the same line to do the needed testing.
+
+# This test can only be run on targets which support DWARF-2 and
use gas.
+load_lib dwarf.exp
+require dwarf2_support
+
+# The DWARF assembler requires the gcc compiler.
+require is_c_compiler_gcc
+
+# This test suitable only for process that can do reverse
execution
+require supports_reverse
+
+standard_testfile .c .S
+
+if { [prepare_for_testing "failed to prepare" ${testfile}
${srcfile}] } {
+ return -1
+}
+
+set asm_file [standard_output_file $srcfile2]
+Dwarf::assemble $asm_file {
+ global srcdir subdir srcfile
+ declare_labels integer_label L
+
+ # Find start address and length of program
+ lassign [function_range main [list
${srcdir}/${subdir}/$srcfile]] \
+ main_start main_len
+ set main_end "$main_start + $main_len"
+
+ cu {} {
+ compile_unit {
+ {language @DW_LANG_C}
+ {name map-to-same-line.c}
+ {stmt_list $L DW_FORM_sec_offset}
+ {low_pc 0 addr}
+ } {
+ subprogram {
+ {external 1 flag}
+ {name main}
+ {low_pc $main_start addr}
+ {high_pc $main_len DW_FORM_data4}
+ }
+ }
+ }
+
+ lines {version 2 default_is_stmt 1} L {
+ include_dir "${srcdir}/${subdir}"
+ file_name "$srcfile" 1
+
+ # Generate the line table program with distinct source lines
being
+ # mapped to the same line entry. Line 1, 5 and 8 contain 1
statement
+ # each. Line 2 contains 2 statements. Line 3 contains 3
statements.
+ program {
+ DW_LNE_set_address $main_start
+ line [gdb_get_line_number "TAG: main prologue"]
+ DW_LNS_copy
+ DW_LNE_set_address line1
+ line [gdb_get_line_number "TAG: line 1" ]
+ DW_LNS_copy
+ DW_LNE_set_address line2
+ line [gdb_get_line_number "TAG: line 2" ]
+ DW_LNS_copy
+ DW_LNE_set_address line3
+ line [gdb_get_line_number "TAG: line 2" ]
+ DW_LNS_copy
+ DW_LNE_set_address line4
+ line [gdb_get_line_number "TAG: line 3" ]
+ DW_LNS_copy
+ DW_LNE_set_address line5
+ line [gdb_get_line_number "TAG: line 3" ]
+ DW_LNS_copy
+ DW_LNE_set_address line6
+ line [gdb_get_line_number "TAG: line 3" ]
+ DW_LNS_copy
+ DW_LNE_set_address line7
+ line [gdb_get_line_number "TAG: line 5" ]
+ DW_LNS_copy
+ DW_LNE_set_address line8
+ line [gdb_get_line_number "TAG: line 8" ]
+ DW_LNS_copy
+ DW_LNE_set_address main_return
+ line [gdb_get_line_number "TAG: main return"]
+ DW_LNS_copy
+ DW_LNE_set_address end_of_sequence
+ DW_LNE_end_sequence
+ }
+ }
+}
+
+if { [prepare_for_testing "failed to prepare" ${testfile} \
+ [list $srcfile $asm_file] {nodebug} ] } {
+ return -1
+}
+
+runto_main
+
+# Print the line table
+gdb_test_multiple "maint info line-table ${testfile}" "" {
+ -re "\r\n$decimal\[ \t\]+$decimal\[ \t\]+($hex)\[
\t\]+Y\[^\r\n\]*" {
+ lappend is_stmt $expect_out(1,string)
+ exp_continue
+ }
+ -re -wrap "" {
+ }
+}
+
+# Do the reverse-step test
+gdb_test_no_output "record" "turn on process record"
+
+set bp_main_return [gdb_get_line_number "TAG: main return"
$srcfile]
+gdb_breakpoint $srcfile:$bp_main_return
+gdb_continue_to_breakpoint "run to end of main, reverse-step
test" ".*$srcfile:$bp_main_return.*"
+gdb_test "display \$pc" ".*pc =.*" "display pc, reverse-step test"
+
+# At this point, GDB has already recorded the execution up until
the return
+# statement. Reverse-step and test if GDB transitions between
lines in the
+# expected order. It should reverse-step across lines 8, 5, 3, 2
and 1.
+foreach line {8 5 3 2 1} {
+ gdb_test "reverse-step" ".*TAG: line $line.*" "reverse step to
line $line"
+}
+
+## Clean restart, test reverse-next command
+clean_restart ${testfile}
+runto_main
+gdb_test_no_output "record" "turn on process record, reverst-next
test"
+
+set bp_main_return [gdb_get_line_number "TAG: main return"
$srcfile]
+gdb_breakpoint $srcfile:$bp_main_return
+gdb_continue_to_breakpoint "run to end of main, reverse-next
test" ".*$srcfile:$bp_main_return.*"
+gdb_test "display \$pc" ".*pc =.*" "display pc, reverse-next test"
+
+# At this point, GDB has already recorded the execution up until
the return
+# statement. Reverse-next and test if GDB transitions between
lines in the
+# expected order. It should reverse-next across lines 8, 5, 3, 2
and 1.
+foreach line {8 5 3 2 1} {
+ gdb_test "reverse-next" ".*TAG: line $line.*" "reverse next to
line $line"
+}
It seems like the step and next tests are identical, so I guess it
could
be factored out using:
foreach_with_prefix method {step next} {
...
}
?
Yup, redid the code using the foreach_with_prefix.
>
Simon
> yR
^ permalink raw reply [flat|nested] 40+ messages in thread
* RE: [PATCH 2/2 v5] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-06-22 20:38 ` Carl Love
@ 2023-06-22 20:39 ` Carl Love
2023-06-23 17:49 ` Simon Marchi
1 sibling, 0 replies; 40+ messages in thread
From: Carl Love @ 2023-06-22 20:39 UTC (permalink / raw)
To: Simon Marchi, Bruno Larsen, gdb-patches, UlrichWeigand, pedro
Cc: luis.machado
Sorry, ignore this message. Accidentally hit send before I was done
with the message.
Carl
--------------------------------------------------
On Thu, 2023-06-22 at 13:38 -0700, Carl Love wrote:
> Simon:
>
> On Mon, 2023-06-19 at 13:58 -0400, Simon Marchi wrote:
> <snip>
>
> 100644
> > > --- a/gdb/infrun.c
> > > +++ b/gdb/infrun.c
> > > @@ -114,6 +114,9 @@ static struct async_event_handler
> > > *infrun_async_inferior_event_token;
> > > Starts off as -1, indicating "never enabled/disabled". */
> > > static int infrun_is_async = -1;
> > >
> > > +static CORE_ADDR update_line_range_start (CORE_ADDR pc,
> > > + struct
> > > execution_control_state *ecs);
> > > +
> >
> > This forward-declaration is not needed.
>
> I tried removing the forward-declaration and the compile fails with
> the
> message:
>
> ../../binutils-gdb-reverse-multiple-
> contiguous/gdb/infrun.c:6773:1:
> error: no previous declaration for ‘CORE_ADDR
> update_line_range_start(CORE_ADDR, execution_control_state*)’ [-
> Werror=missing-declarations]
> 6773 | update_line_range_start (CORE_ADDR pc, struct
> execution_control_state *ecs)
> | ^~~~~~~~~~~~~~~~~~~~~~~
> cc1plus: all warnings being treated as errors
> make[2]: *** [Makefile:1922: infrun.o] Error 1
> make[2]: Leaving directory '/home/carll/GDB/build-reverse-
> multiple-
> contiguous/gdb'
> make[1]: *** [Makefile:13569: all-gdb] Error 2
> make[1]: Leaving directory '/home/carll/GDB/build-reverse-
> multiple-
> contiguous'
> make: *** [Makefile:1005: all] Error 2
>
> Leaving the forward declaration in the code.
>
> >
> > > /* See infrun.h. */
> > >
> > > void
> > > @@ -6769,6 +6772,25 @@ handle_signal_stop (struct
> > > execution_control_state *ecs)
> > > process_event_stop_test (ecs);
> > > }
> > >
> > > +CORE_ADDR
> > > +update_line_range_start (CORE_ADDR pc, struct
> > > execution_control_state *ecs)
> >
> > Please add a comment for the function.
>
> Done.
>
> > > +{
> > > + /* The line table may have multiple entries for the same
> > > source
> > > code line.
> > > + Given the PC, check the line table and return the PC that
> > > corresponds
> > > + to the line table entry for the source line that PC is
> > > in. */
> > > + CORE_ADDR start_line_pc = ecs->event_thread-
> > > > control.step_range_start;
> > > + gdb::optional<CORE_ADDR> real_range_start;
> > > +
> > > + /* Call find_line_range_start to get the smallest address in
> > > the
> > > + linetable for multiple Line X entries in the line
> > > table. */
> > > + real_range_start = find_line_range_start (pc);
> > > +
> > > + if (real_range_start.has_value ())
> > > + start_line_pc = *real_range_start;
> > > +
> > > + return start_line_pc;
> >
> > When I read this, I wonder: why was control.step_range_start not
> > set
> > to
> > the "real" range start in the first place (not only in the context
> > of
> > reverse execution, every time it is set)? It would seem more
> > robust
> > than patching it afterwards in some very specific spots.
> >
> > I could see some benefits for range-stepping uses cases too
> > (relevant
> > when debugging remotely). Using your example here:
> >
> > Line X - [0x0 - 0x8]
> > Line X - [0x8 - 0x10]
> > Line X - [0x10 - 0x18]
> >
> > Imagine we are stopped at 0x14, and we type "next", and 0x14 is a
> > conditional jump to 0x5. It seems like current GDB would send a
> > "range
> > step" request to GDBserver, to step in the [0x10, 0x18[
> > range. When
> > reaching 0x5, execution would stop, and GDB would resume it again
> > with
> > the [0x0,0x8[ range. When reaching 0x8, it would stop again, GDB
> > would
> > resume it with [0x8,0x10[, and so on. If GDB could send a "range
> > step"
> > request with the [0x0,0x18[ range, it would avoid those unnecessary
> > intermediary stop.
> >
> > > +}
> > > +
> > > /* Come here when we've got some debug event / signal we can
> > > explain
> > > (IOW, not a random signal), and test whether it should cause
> > > a
> > > stop, or whether we should resume the inferior
> > > (transparently).
> > > @@ -7570,6 +7592,28 @@ process_event_stop_test (struct
> > > execution_control_state *ecs)
> > >
> > > if (stop_pc_sal.is_stmt)
> > > {
> > > + if (execution_direction == EXEC_REVERSE)
> > > + {
> > > + /* We are stepping backwards make sure we have reached
> > > the
> > > + beginning of the line. */
> > > + CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
> > > + CORE_ADDR start_line_pc
> > > + = update_line_range_start (stop_pc, ecs);
> > > +
> > > + if (stop_pc != start_line_pc)
> > > + {
> > > + /* Have not reached the beginning of the source code
> > > line.
> > > + Set a step range. Execution should stop in any
> > > function
> > > + calls we execute back into before reaching the
> > > beginning
> > > + of the line. */
> > > + ecs->event_thread->control.step_range_start =
> > > start_line_pc;
> > > + ecs->event_thread->control.step_range_end = stop_pc;
> > > + set_step_info (ecs->event_thread, frame,
> > > stop_pc_sal);
> > > + keep_going (ecs);
> > > + return;
> > > + }
> > > + }
> > > +
> > > /* We are at the start of a statement.
> > >
> > > So stop. Note that we don't stop if we step into the
> > > middle of a
> > > @@ -7632,6 +7676,19 @@ process_event_stop_test (struct
> > > execution_control_state *ecs)
> > > set_step_info (ecs->event_thread, frame, stop_pc_sal);
> > >
> > > infrun_debug_printf ("keep going");
> > > +
> > > + if (execution_direction == EXEC_REVERSE)
> > > + {
> > > + CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
> > > +
> > > + /* Make sure the stop_pc is set to the beginning of the
> > > line. */
> > > + if (stop_pc != ecs->event_thread-
> > > >control.step_range_start)
> > > + {
> > > + stop_pc = update_line_range_start (stop_pc, ecs);
> > > + ecs->event_thread->control.step_range_start = stop_pc;
> > > + }
> > > + }
> > > +
> > > keep_going (ecs);
> > > }
> > >
> > > diff --git a/gdb/symtab.c b/gdb/symtab.c
> > > index 27611a34ec4..91d35616eb9 100644
> > > --- a/gdb/symtab.c
> > > +++ b/gdb/symtab.c
> > > @@ -3282,6 +3282,55 @@ find_pc_line (CORE_ADDR pc, int
> > > notcurrent)
> > > return sal;
> > > }
> > >
> > > +/* Compare two symtab_and_line entries. Return true if both
> > > have
> > > + the same line number and the same symtab pointer. That means
> > > we
> > > + are dealing with two entries from the same line and from the
> > > same
> > > + source file.
> > > +
> > > + Return false otherwise. */
> > > +
> > > +static bool
> > > +sal_line_symtab_matches_p (const symtab_and_line &sal1,
> > > + const symtab_and_line &sal2)
> > > +{
> > > + return (sal1.line == sal2.line && sal1.symtab == sal2.symtab);
> >
> > Unnecessary parenthesis.
>
> Removed unnecessary parenthesis.
>
> > > +}
> > > +
> > > +/* See symtah.h. */
> > > +
> > > +gdb::optional<CORE_ADDR>
> > > +find_line_range_start (CORE_ADDR pc)
> > > +{
> > > + struct symtab_and_line current_sal = find_pc_line (pc, 0);
> > > +
> > > + if (current_sal.line == 0)
> > > + return {};
> > > +
> > > + struct symtab_and_line prev_sal = find_pc_line (current_sal.pc
> > > -
> > > 1, 0);
> > > +
> > > + /* If the previous entry is for a different line, that means
> > > we
> > > are already
> > > + at the entry with the start PC for this line. */
> > > + if (!sal_line_symtab_matches_p (prev_sal, current_sal))
> > > + return current_sal.pc;
> > > +
> > > + /* Otherwise, keep looking for entries for the same line but
> > > with
> > > + smaller PC's. */
> > > + bool done = false;
> > > + CORE_ADDR prev_pc;
> > > + while (!done)
> > > + {
> > > + prev_pc = prev_sal.pc;
> > > +
> > > + prev_sal = find_pc_line (prev_pc - 1, 0);
> > > +
> > > + /* Did we notice a line change? If so, we are done with
> > > the
> > > search. */
> > > + if (!sal_line_symtab_matches_p (prev_sal, current_sal))
> > > + done = true;
> > > + }
> > > +
> > > + return prev_pc;
> >
> > Algorithmic complexity question: given that line tables are sorted
> > by
> > address, would it work to start at the current line table item, and
> > go
> > look at the previous ones until we find one that is no longer
> > contiguous and same line? find_pc_line is somewhat heavy, so if we
> > don't need to do it repeatedly...
> >
> > > +}
> > > +
> > > /* See symtab.h. */
> > >
> > > struct symtab *
> > > diff --git a/gdb/symtab.h b/gdb/symtab.h
> > > index 404d0ab30a8..f54305636da 100644
> > > --- a/gdb/symtab.h
> > > +++ b/gdb/symtab.h
> > > @@ -2346,6 +2346,22 @@ extern struct symtab_and_line find_pc_line
> > > (CORE_ADDR, int);
> > > extern struct symtab_and_line find_pc_sect_line (CORE_ADDR,
> > > struct obj_section *,
> > > int);
> > >
> > > +/* Given PC, and assuming it is part of a range of addresses
> > > that
> > > is part of a
> > > + line, go back through the linetable and find the starting PC
> > > of
> > > that
> > > + line.
> > > +
> > > + For example, suppose we have 3 PC ranges for line X:
> > > +
> > > + Line X - [0x0 - 0x8]
> > > + Line X - [0x8 - 0x10]
> > > + Line X - [0x10 - 0x18]
> > > +
> > > + If we call the function with PC == 0x14, we want to return
> > > 0x0,
> > > as that is
> > > + the starting PC of line X, and the ranges are contiguous.
> >
> > I think that putting this example in the comment is great. It
> > makes
> > it
> > much more obvious what the function specifically does.
> >
> > > +*/
> > > +
> > > +extern gdb::optional<CORE_ADDR> find_line_range_start (CORE_ADDR
> > > pc);
> > > +
> > > /* Wrapper around find_pc_line to just return the symtab. */
> > >
> > > extern struct symtab *find_pc_line_symtab (CORE_ADDR);
> > > diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> > > b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> > > new file mode 100644
> > > index 00000000000..da944874e86
> > > --- /dev/null
> > > +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> > > @@ -0,0 +1,36 @@
> > > +/* Copyright 2008-2023 Free Software Foundation, Inc.
> > > +
> > > + This program is free software; you can redistribute it and/or
> > > modify
> > > + it under the terms of the GNU General Public License as
> > > published by
> > > + the Free Software Foundation; either version 3 of the
> > > License,
> > > or
> > > + (at your option) any later version.
> > > +
> > > + This program is distributed in the hope that it will be
> > > useful,
> > > + but WITHOUT ANY WARRANTY; without even the implied warranty
> > > of
> > > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> > > + GNU General Public License for more details.
> > > +
> > > + You should have received a copy of the GNU General Public
> > > License
> > > + along with this program. If not, see <
> > > http://www.gnu.org/licenses/
> > > >.
> > > +
> > > + This test is used to test the reverse-step and reverse-next
> > > instruction
> > > + execution for a source line that contains multiple function
> > > calls. */
> > > +
> > > +void
> > > +func1 ()
> > > +{
> > > +} /* END FUNC1 */
> > > +
> > > +void
> > > +func2 ()
> > > +{
> > > +} /* END FUNC2 */
> > > +
> > > +int main ()
> >
> > int
> > main (void)
> >
>
> Fixed.
>
> > > +{
> > > + int a, b;
> > > + a = 1;
> > > + b = 2;
> > > + func1 (); func2 ();
> > > + a = a + b; /* START REVERSE TEST */
> > > +}
> > > diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> > > b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> > > new file mode 100644
> > > index 00000000000..89e226b0f84
> > > --- /dev/null
> > > +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> > > @@ -0,0 +1,140 @@
> > > +# Copyright 2008-2023 Free Software Foundation, Inc.
> > > +
> > > +# This program is free software; you can redistribute it and/or
> > > modify
> > > +# it under the terms of the GNU General Public License as
> > > published by
> > > +# the Free Software Foundation; either version 3 of the License,
> > > or
> > > +# (at your option) any later version.
> > > +#
> > > +# This program is distributed in the hope that it will be
> > > useful,
> > > +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> > > +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> > > +# GNU General Public License for more details.
> > > +#
> > > +# You should have received a copy of the GNU General Public
> > > License
> > > +# along with this program. If not, see <
> > > http://www.gnu.org/licenses/
> > > >. */
> > > +
> > > +# This file is part of the GDB testsuite. It tests reverse
> > > stepping.
> > > +# Lots of code borrowed from "step-test.exp".
> > > +
> > > +# This test checks to make sure there is no regression failures
> > > for
> > > +# the reverse-next command when stepping back over two functions
> > > in
> > > +# the same line.
> > > +
> > > +require supports_reverse
> > > +
> > > +# This test uses the gcc no-column-info command which was added
> > > in
> > > gcc 7.1.
> > > +
> > > +proc run_tests {} {
> > > + global srcfile
> > > + global executable
> > > +
> > > + runto_main
> >
> > We typically check for runto_main's success:
>
> Fixed two instances of runto_main in this test case and two in the
> other test case.
>
> > if { ![runto_main] } {
> > return
> > }
> >
> > runto_main logs a FAIL on failure. There are a few runto_mains in
> > the
> > patch.
> >
> > > + set target_remote [gdb_is_target_remote]
> >
> > target_remote seems unused
>
> Removed.
>
> > > +
> > > + with_test_prefix "test1" {
> > > + gdb_test_no_output "record" "turn on process record"
> > > + }
> >
> > with_test_prefix with a single test in it is really just the same
> > as:
> >
> > gdb_test_no_output "record" "test1: turn on process record"
> >
> > In fact, you have some other tests with the "test1:" or "test2:"
> > prefix,
> > I think they should be moved to the with_test_prefix. And maybe
> > use
> > "next" and "step" instead of "test1" and "test2".
>
> Yup, cleaner to have the with_test_prefix cover the whole test.
> Changed test1 to next-test and test2 to step-next.
>
> > > +
> > > + # This regression test verifies the reverse-step and
> > > reverse-
> > > next commands
> > > + # work properly when executing backwards thru a source line
> > > containing
> > > + # two function calls on the same source line, i.e. func1 ();
> > > func2 ();
> > > + # This test is compiled so the dwarf info not contain the
> > > line
> > > table
> > > + # information.
> > > +
> > > + # Test 1, reverse-next command
> > > + # Set breakpoint at the line after the function calls.
> > > + set bp_start_reverse_test [gdb_get_line_number "START
> > > REVERSE
> > > TEST" \
> > > + $srcfile]
> > > + gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> > > +
> > > + # Continue to break point for reverse-next test.
> > > + # Command definition: reverse-next [count]
> > > + # Run backward to the beginning of the previous line
> > > executed in the
> > > + # current (innermost) stack frame. If the line contains
> > > function calls,
> > > + # they will be “un-executed” without stopping. Starting
> > > from
> > > the first
> > > + # line of a function, reverse-next will take you back to
> > > the
> > > caller of
> > > + # that function, before the function was called, just as
> > > the
> > > normal next
> > > + # command would take you from the last line of a function
> > > back to its
> > > + # return to its caller 2 .
> > > + gdb_continue_to_breakpoint \
> > > + "test1: stopped at command reverse-next test start location" \
> > > + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> > > +
> > > + # The reverse-next should step all the way back to the
> > > beginning of the
> > > + # line, i.e. at the beginning of the func1 call.
> > > + gdb_test "reverse-next" ".*func1 \\(\\); func2 \\(\\);.*" \
> > > + "test1: reverse-next to line with two functions"
> > > +
> > > + # We should be stopped at the first instruction of the line.
> > > A
> > > reverse-step
> > > + # should step back and stop at the beginning of the previous
> > > line b = 2,
> > > + # i.e. not in func1 ().
> > > + gdb_test "reverse-stepi" ".*b = 2;.*" \
> > > + "test1: reverse-stepi to previous line b = 2"
> > > +
> > > +
> > > + # Setup for test 2
> > > + clean_restart $executable
> > > + runto_main
> > > +
> > > + with_test_prefix "test2" {
> > > + gdb_test_no_output "record" "turn on process record"
> > > + }
> > > +
> > > + # Test 2, reverse-step command
> > > + # Set breakpoint at the line after the function calls.
> > > + gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> > > +
> > > + # Continue to the start of the reverse-step test.
> > > + # Command definition: reverse-step [count]
> > > + # Run the program backward until control reaches the
> > > start
> > > of a
> > > + # different source line; then stop it, and return control
> > > to gdb.
> > > + # Like the step command, reverse-step will only stop at
> > > the
> > > beginning
> > > + # of a source line. It “un-executes” the previously
> > > executed source
> > > + # line. If the previous source line included calls to
> > > debuggable
> > > + # functions, reverse-step will step (backward) into the
> > > called function,
> > > + # stopping at the beginning of the last statement in the
> > > called
> > > + # function (typically a return statement). Also, as with
> > > the step
> > > + # command, if non-debuggable functions are called,
> > > reverse-
> > > step will
> > > + # run thru them backward without stopping.
> > > +
> > > + gdb_continue_to_breakpoint \
> > > + "test2: stopped at command reverse-step test start location" \
> > > + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> > > +
> > > + # The first reverse step should take us call of func2 ().
> > > + gdb_test "reverse-step" ".*END FUNC2.*" \
> > > + "test2: reverse-step into func2 "
> > > +
> > > + # The second reverse step should take us into func1 ().
> > > + gdb_test "reverse-step" ".*END FUNC1.*" \
> > > + "test2: reverse-step into func1 "
> > > +
> > > + # The third reverse step should take us call of func1 ().
> > > + gdb_test "reverse-step" ".*func1 \\(\\); func2 \\(\\);.*" \
> > > + "test2: reverse-step to line func1(); func2(), at call for
> > > func1 "
> > > +
> > > + # We should be stopped at the first instruction of the line.
> > > A
> > > reverse
> > > + # stepi should take us to b = 2 ().
> > > + gdb_test "reverse-stepi" ".*b = 2;.*" \
> > > + "test2: reverse-stepi to line b = 2 "
> > > +}
> > > +
> > > +set srcfile func-map-to-same-line.c
> > > +set executable func-map-to-same-line
> >
> > Wondering if this test should use standard_testfile (like almost
> > every
> > other tests) to set these.
>
> OK, changed to use the standard_testfile.
>
> > > +
> > > +# test with and without gcc column info enabled
> > > +foreach_with_prefix with_column_info {yes no} {
> > > + if {$with_column_info == "yes"} {
> > > + set options [list debug column-info]
> > > + } else {
> > > + set options [list debug no-column-info]
> > > + }
> >
> > I didn't think of this when proposing the foreach_with_prefix, but
> > you
> > could perhaps use:
> >
> > foreach_with_prefix column_info_flag {column-info no-column-info}
> >
> > ... to avoid this boilerplate. You can then use $column_info_flag
> > directly when setting options.
> >
> OK, that cleans things up a bit. Changed.
> +
> + if {[build_executable "failed to prepare" $executable $srcfile
> \
> + $options] == -1} {
> + return -1
> + }
> +
> + clean_restart $executable
>
> clean_restart can go in run_tests.
>
>
> + run_tests
> +}
> diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.c
> b/gdb/testsuite/gdb.reverse/map-to-same-line.c
> new file mode 100644
> index 00000000000..f20d778f40e
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/map-to-same-line.c
> @@ -0,0 +1,58 @@
> +/* Copyright 2008-2023 Free Software Foundation, Inc.
>
> Just wondering if the copyright years are right.
>
> New files so yea, should just start with 2023.
> +
> + This program is free software; you can redistribute it and/or
> modify
> + it under the terms of the GNU General Public License as
> published by
> + the Free Software Foundation; either version 3 of the License,
> or
> + (at your option) any later version.
> +
> + This program is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> + GNU General Public License for more details.
> +
> + You should have received a copy of the GNU General Public
> License
> + along with this program. If not, see <
> http://www.gnu.org/licenses/
> >. */
> +
> +/* The purpose of this test is to create a DWARF line table that
> contains two
> + or more entries for the same line. When stepping (forwards or
> backwards),
> + GDB should step over the entire line and not just a particular
> entry in the
> + line table. */
> +
> +int
> +main ()
>
> void in the parenthesis
>
> Fixed in both test files.
>
> +{ /* TAG: main prologue */
> + asm ("main_label: .globl main_label");
> + int i = 1, j = 2, k;
> + float f1 = 2.0, f2 = 4.1, f3;
> + const char *str_1 = "foo", *str_2 = "bar", *str_3;
> +
> + asm ("line1: .globl line1");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 1 */
> +
> + asm ("line2: .globl line2");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 2 */
> +
> + asm ("line3: .globl line3");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 3 */
> +
> + asm ("line4: .globl line4");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 4 */
> +
> + asm ("line5: .globl line5");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 5 */
> +
> + asm ("line6: .globl line6");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 6 */
> +
> + asm ("line7: .globl line7");
> + k = i; f3 = f1; str_3 = str_1; /* TAG: line 7 */
> +
> + asm ("line8: .globl line8");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: line 8 */
> +
> + asm ("main_return: .globl main_return");
> + k = j; f3 = f2; str_3 = str_2; /* TAG: main return */
> +
> + asm ("end_of_sequence: .globl end_of_sequence");
> + return 0; /* TAG: main return */
> +}
> diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> new file mode 100644
> index 00000000000..16a359d90ec
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> @@ -0,0 +1,156 @@
> +# Copyright 2008-2023 Free Software Foundation, Inc.
> +
> +# This program is free software; you can redistribute it and/or
> modify
> +# it under the terms of the GNU General Public License as
> published by
> +# the Free Software Foundation; either version 3 of the License,
> or
> +# (at your option) any later version.
> +#
> +# This program is distributed in the hope that it will be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> +# GNU General Public License for more details.
> +#
> +# You should have received a copy of the GNU General Public
> License
> +# along with this program. If not, see <
> http://www.gnu.org/licenses/
> = >.
> +
> +# When stepping (forwards or backwards), GDB should step over the
> entire line
> +# and not just a particular entry in the line table. This test was
> added to
> +# verify the find_line_range_start function properly sets the step
> range for a
> +# line that consists of multiple statements, i.e. multiple entries
> in the line
> +# table. This test creates a DWARF line table that contains two
> entries for
> +# the same line to do the needed testing.
> +
> +# This test can only be run on targets which support DWARF-2 and
> use gas.
> +load_lib dwarf.exp
> +require dwarf2_support
> +
> +# The DWARF assembler requires the gcc compiler.
> +require is_c_compiler_gcc
> +
> +# This test suitable only for process that can do reverse
> execution
> +require supports_reverse
> +
> +standard_testfile .c .S
> +
> +if { [prepare_for_testing "failed to prepare" ${testfile}
> ${srcfile}] } {
> + return -1
> +}
> +
> +set asm_file [standard_output_file $srcfile2]
> +Dwarf::assemble $asm_file {
> + global srcdir subdir srcfile
> + declare_labels integer_label L
> +
> + # Find start address and length of program
> + lassign [function_range main [list
> ${srcdir}/${subdir}/$srcfile]] \
> + main_start main_len
> + set main_end "$main_start + $main_len"
> +
> + cu {} {
> + compile_unit {
> + {language @DW_LANG_C}
> + {name map-to-same-line.c}
> + {stmt_list $L DW_FORM_sec_offset}
> + {low_pc 0 addr}
> + } {
> + subprogram {
> + {external 1 flag}
> + {name main}
> + {low_pc $main_start addr}
> + {high_pc $main_len DW_FORM_data4}
> + }
> + }
> + }
> +
> + lines {version 2 default_is_stmt 1} L {
> + include_dir "${srcdir}/${subdir}"
> + file_name "$srcfile" 1
> +
> + # Generate the line table program with distinct source lines
> being
> + # mapped to the same line entry. Line 1, 5 and 8 contain 1
> statement
> + # each. Line 2 contains 2 statements. Line 3 contains 3
> statements.
> + program {
> + DW_LNE_set_address $main_start
> + line [gdb_get_line_number "TAG: main prologue"]
> + DW_LNS_copy
> + DW_LNE_set_address line1
> + line [gdb_get_line_number "TAG: line 1" ]
> + DW_LNS_copy
> + DW_LNE_set_address line2
> + line [gdb_get_line_number "TAG: line 2" ]
> + DW_LNS_copy
> + DW_LNE_set_address line3
> + line [gdb_get_line_number "TAG: line 2" ]
> + DW_LNS_copy
> + DW_LNE_set_address line4
> + line [gdb_get_line_number "TAG: line 3" ]
> + DW_LNS_copy
> + DW_LNE_set_address line5
> + line [gdb_get_line_number "TAG: line 3" ]
> + DW_LNS_copy
> + DW_LNE_set_address line6
> + line [gdb_get_line_number "TAG: line 3" ]
> + DW_LNS_copy
> + DW_LNE_set_address line7
> + line [gdb_get_line_number "TAG: line 5" ]
> + DW_LNS_copy
> + DW_LNE_set_address line8
> + line [gdb_get_line_number "TAG: line 8" ]
> + DW_LNS_copy
> + DW_LNE_set_address main_return
> + line [gdb_get_line_number "TAG: main return"]
> + DW_LNS_copy
> + DW_LNE_set_address end_of_sequence
> + DW_LNE_end_sequence
> + }
> + }
> +}
> +
> +if { [prepare_for_testing "failed to prepare" ${testfile} \
> + [list $srcfile $asm_file] {nodebug} ] } {
> + return -1
> +}
> +
> +runto_main
> +
> +# Print the line table
> +gdb_test_multiple "maint info line-table ${testfile}" "" {
> + -re "\r\n$decimal\[ \t\]+$decimal\[ \t\]+($hex)\[
> \t\]+Y\[^\r\n\]*" {
> + lappend is_stmt $expect_out(1,string)
> + exp_continue
> + }
> + -re -wrap "" {
> + }
> +}
> +
> +# Do the reverse-step test
> +gdb_test_no_output "record" "turn on process record"
> +
> +set bp_main_return [gdb_get_line_number "TAG: main return"
> $srcfile]
> +gdb_breakpoint $srcfile:$bp_main_return
> +gdb_continue_to_breakpoint "run to end of main, reverse-step
> test" ".*$srcfile:$bp_main_return.*"
> +gdb_test "display \$pc" ".*pc =.*" "display pc, reverse-step test"
> +
> +# At this point, GDB has already recorded the execution up until
> the return
> +# statement. Reverse-step and test if GDB transitions between
> lines in the
> +# expected order. It should reverse-step across lines 8, 5, 3, 2
> and 1.
> +foreach line {8 5 3 2 1} {
> + gdb_test "reverse-step" ".*TAG: line $line.*" "reverse step to
> line $line"
> +}
> +
> +## Clean restart, test reverse-next command
> +clean_restart ${testfile}
> +runto_main
> +gdb_test_no_output "record" "turn on process record, reverst-next
> test"
> +
> +set bp_main_return [gdb_get_line_number "TAG: main return"
> $srcfile]
> +gdb_breakpoint $srcfile:$bp_main_return
> +gdb_continue_to_breakpoint "run to end of main, reverse-next
> test" ".*$srcfile:$bp_main_return.*"
> +gdb_test "display \$pc" ".*pc =.*" "display pc, reverse-next test"
> +
> +# At this point, GDB has already recorded the execution up until
> the return
> +# statement. Reverse-next and test if GDB transitions between
> lines in the
> +# expected order. It should reverse-next across lines 8, 5, 3, 2
> and 1.
> +foreach line {8 5 3 2 1} {
> + gdb_test "reverse-next" ".*TAG: line $line.*" "reverse next to
> line $line"
> +}
>
> It seems like the step and next tests are identical, so I guess it
> could
> be factored out using:
>
> foreach_with_prefix method {step next} {
> ...
> }
>
> ?
>
> Yup, redid the code using the foreach_with_prefix.
>
> Simon
> > yR
^ permalink raw reply [flat|nested] 40+ messages in thread
* Re: [PATCH 1/2] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-06-22 16:52 ` Carl Love
@ 2023-06-23 17:44 ` Simon Marchi
2023-06-23 19:41 ` Carl Love
2023-06-23 20:04 ` [PATCH 1/2 ver 2] " Carl Love
0 siblings, 2 replies; 40+ messages in thread
From: Simon Marchi @ 2023-06-23 17:44 UTC (permalink / raw)
To: Carl Love, Bruno Larsen, gdb-patches, UlrichWeigand, pedro; +Cc: luis.machado
On 6/22/23 12:52, Carl Love wrote:
>
> Simon:
>
> On Mon, 2023-06-19 at 13:11 -0400, Simon Marchi wrote:
>>> --- a/gdb/testsuite/lib/gdb.exp
>>> +++ b/gdb/testsuite/lib/gdb.exp
>>> @@ -4794,6 +4794,8 @@ proc quote_for_host { args } {
>>> # debug information
>>> # - text_segment=addr: Tell the linker to place the text
>>> segment at ADDR.
>>> # - build-id: Ensure the final binary includes a build-id.
>>> +# - no-column-info: Disable generation of column table
>>> information.
>>> +# - column-info: Enable generation of column table information.
>>> #
>>> # And here are some of the not too obscure options understood by
>>> DejaGnu that
>>> # influence the compilation:
>>> @@ -5003,6 +5005,34 @@ proc gdb_compile {source dest type options}
>>> {
>>> } else {
>>> error "Don't know how to handle text_segment
>>> option."
>>> }
>>> + } elseif { $opt == "column-info" } {
>>> + if {[test_compiler_info {gcc-*}]} {
>>> + if {[test_compiler_info {gcc-[1-6]-*}]} {
>>> + error "gdb_compile option no-column-info not
>>> supported."
>>
>> I think this path should return the equivalent of "failed to
>> compile",
>> instead of throwing an error. Control will go back to the test,
>> which
>> will generally skip the portion of the test that requires that
>> binary.
>
> Not entirely sure how to accomplish what you are looking for.
>
> I change:
> error "gdb_compile option no-column-info not supported."
> to
> set result "option no-column-info not supported."
> clone_output "gdb compile failed, $result"
> return 1
>
> When I force the if {[test_compiler_info...]} tp be true to test this,
> I get:
>
> get_compiler_info: gcc-12-2-1
> gdb compile failed, option no-column-info not supported.
> UNTESTED: gdb.reverse/func-map-to-same-line.exp:
> with_column_info=yes: failed t\
> o prepare
> testcase /home/carll/GDB/build-reverse-multiple-
> contiguous/gdb/testsuite/../../\
> ../binutils-gdb-reverse-multiple-
> contiguous/gdb/testsuite/gdb.reverse/func-map-\
> to-same-line.exp completed in 0 seconds
>
> === gdb Summary ===
>
> # of untested testcases 1
>
> The test case doesn't have any part of the test that doesn't require
> compiling so it is not clear if that would work with this fix. Anyway,
> wanted to run that by you to see if this is an appropriate fix? I am
> really not sure about it. Thanks.
I think that's the expected behavior. The UNTESTED is emitted by
build_executable_from_specs, I think. If the test used gdb_compile, I
think we wouldn't see an UNTESTED. But as far as your addition is
concerned, I think it's fine.
I just thought of a simpler alternative though. Just remove the version
check. If we build with an older gcc, there will simply be a message
that says that the flag is not recognized, and the result should be just
the same. I just hacked it locally and changed the flag name to be
wrong (I don't have a gcc <= 6 on hand to test). It looks like:
Executing on host: gcc -fno-stack-protector -fdiagnostics-color=never -gcolumn-info-foo -c -g -o /home/simark/build/binutils-gdb/gdb/testsuite/outputs/gdb.reverse/func-map-to-same-line/func-map-to-same-line0.o /home/simark/src/binutils-gdb/gdb/testsuite/gdb.reverse/func-map-to-same-line.c (timeout = 300)
builtin_spawn -ignore SIGHUP gcc -fno-stack-protector -fdiagnostics-color=never -gcolumn-info-foo -c -g -o /home/simark/build/binutils-gdb/gdb/testsuite/outputs/gdb.reverse/func-map-to-same-line/func-map-to-same-line0.o /home/simark/src/binutils-gdb/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
gcc: error: unrecognized debug output level 'column-info-foo'
compiler exited with status 1
output is:
gcc: error: unrecognized debug output level 'column-info-foo'
gdb compile failed, gcc: error: unrecognized debug output level 'column-info-foo'
UNTESTED: gdb.reverse/func-map-to-same-line.exp: with_column_info=yes: failed to prepare
I then thought about the "no-column-info" case. Currently, you error
out for gccs <= 6. However, shouldn't we just compile without any
special flag in that case, since there just wasn't any support for
column-info back then?
Simon
^ permalink raw reply [flat|nested] 40+ messages in thread
* Re: [PATCH 2/2 v5] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-06-22 20:38 ` Carl Love
2023-06-22 20:39 ` Carl Love
@ 2023-06-23 17:49 ` Simon Marchi
1 sibling, 0 replies; 40+ messages in thread
From: Simon Marchi @ 2023-06-23 17:49 UTC (permalink / raw)
To: Carl Love, Bruno Larsen, gdb-patches, UlrichWeigand, pedro; +Cc: luis.machado
On 6/22/23 16:38, Carl Love wrote:
> Simon:
>
> On Mon, 2023-06-19 at 13:58 -0400, Simon Marchi wrote:
> <snip>
>
>>
> 100644
>>> --- a/gdb/infrun.c
>>> +++ b/gdb/infrun.c
>>> @@ -114,6 +114,9 @@ static struct async_event_handler
>>> *infrun_async_inferior_event_token;
>>> Starts off as -1, indicating "never enabled/disabled". */
>>> static int infrun_is_async = -1;
>>>
>>> +static CORE_ADDR update_line_range_start (CORE_ADDR pc,
>>> + struct
>>> execution_control_state *ecs);
>>> +
>>
>> This forward-declaration is not needed.
>
> I tried removing the forward-declaration and the compile fails with the
> message:
>
> ../../binutils-gdb-reverse-multiple-contiguous/gdb/infrun.c:6773:1:
> error: no previous declaration for ‘CORE_ADDR
> update_line_range_start(CORE_ADDR, execution_control_state*)’ [-
> Werror=missing-declarations]
> 6773 | update_line_range_start (CORE_ADDR pc, struct
> execution_control_state *ecs)
> | ^~~~~~~~~~~~~~~~~~~~~~~
> cc1plus: all warnings being treated as errors
> make[2]: *** [Makefile:1922: infrun.o] Error 1
> make[2]: Leaving directory '/home/carll/GDB/build-reverse-multiple-
> contiguous/gdb'
> make[1]: *** [Makefile:13569: all-gdb] Error 2
> make[1]: Leaving directory '/home/carll/GDB/build-reverse-multiple-
> contiguous'
> make: *** [Makefile:1005: all] Error 2
>
> Leaving the forward declaration in the code.
Because you need to put "static" at the other place (line 6813).
Simon
^ permalink raw reply [flat|nested] 40+ messages in thread
* RE: [PATCH 1/2] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-06-23 17:44 ` Simon Marchi
@ 2023-06-23 19:41 ` Carl Love
2023-06-23 20:04 ` [PATCH 1/2 ver 2] " Carl Love
1 sibling, 0 replies; 40+ messages in thread
From: Carl Love @ 2023-06-23 19:41 UTC (permalink / raw)
To: Simon Marchi, Bruno Larsen, gdb-patches, UlrichWeigand, pedro
Cc: luis.machado, cel
On Fri, 2023-06-23 at 13:44 -0400, Simon Marchi wrote:
>
> On 6/22/23 12:52, Carl Love wrote:
> > Simon:
> >
> > On Mon, 2023-06-19 at 13:11 -0400, Simon Marchi wrote:
> > > > --- a/gdb/testsuite/lib/gdb.exp
> > > > +++ b/gdb/testsuite/lib/gdb.exp
> > > > @@ -4794,6 +4794,8 @@ proc quote_for_host { args } {
> > > > # debug information
> > > > # - text_segment=addr: Tell the linker to place the text
> > > > segment at ADDR.
> > > > # - build-id: Ensure the final binary includes a build-id.
> > > > +# - no-column-info: Disable generation of column table
> > > > information.
> > > > +# - column-info: Enable generation of column table
> > > > information.
> > > > #
> > > > # And here are some of the not too obscure options understood
> > > > by
> > > > DejaGnu that
> > > > # influence the compilation:
> > > > @@ -5003,6 +5005,34 @@ proc gdb_compile {source dest type
> > > > options}
> > > > {
> > > > } else {
> > > > error "Don't know how to handle text_segment
> > > > option."
> > > > }
> > > > + } elseif { $opt == "column-info" } {
> > > > + if {[test_compiler_info {gcc-*}]} {
> > > > + if {[test_compiler_info {gcc-[1-6]-*}]} {
> > > > + error "gdb_compile option no-column-info not
> > > > supported."
> > >
> > > I think this path should return the equivalent of "failed to
> > > compile",
> > > instead of throwing an error. Control will go back to the test,
> > > which
> > > will generally skip the portion of the test that requires that
> > > binary.
> >
> > Not entirely sure how to accomplish what you are looking for.
> >
> > I change:
> > error "gdb_compile option no-column-info not supported."
> > to
> > set result "option no-column-info not supported."
> > clone_output "gdb compile failed, $result"
> > return 1
> >
> > When I force the if {[test_compiler_info...]} tp be true to test
> > this,
> > I get:
> >
> > get_compiler_info: gcc-12-2-1
> > gdb compile failed, option no-column-info not supported.
> > UNTESTED: gdb.reverse/func-map-to-same-line.exp:
> > with_column_info=yes: failed t\
> > o prepare
> > testcase /home/carll/GDB/build-reverse-multiple-
> > contiguous/gdb/testsuite/../../\
> > ../binutils-gdb-reverse-multiple-
> > contiguous/gdb/testsuite/gdb.reverse/func-map-\
> > to-same-line.exp completed in 0 seconds
> >
> > === gdb Summary ===
> >
> > # of untested testcases 1
> >
> > The test case doesn't have any part of the test that doesn't
> > require
> > compiling so it is not clear if that would work with this
> > fix. Anyway,
> > wanted to run that by you to see if this is an appropriate fix? I
> > am
> > really not sure about it. Thanks.
>
> I think that's the expected behavior. The UNTESTED is emitted by
> build_executable_from_specs, I think. If the test used gdb_compile,
> I
> think we wouldn't see an UNTESTED. But as far as your addition is
> concerned, I think it's fine.
>
> I just thought of a simpler alternative though. Just remove the
> version
> check. If we build with an older gcc, there will simply be a message
> that says that the flag is not recognized, and the result should be
> just
> the same. I just hacked it locally and changed the flag name to be
> wrong (I don't have a gcc <= 6 on hand to test). It looks like:
Yea, hacking the if {[test_compiler_info {gcc-[...]} is how I have
been testing it as well. :-)
>
> Executing on host: gcc -fno-stack-protector -fdiagnostics-
> color=never -gcolumn-info-foo -c -g -o /home/simark/build/binutils-
> gdb/gdb/testsuite/outputs/gdb.reverse/func-map-to-same-line/func-map-
> to-same-line0.o /home/simark/src/binutils-
> gdb/gdb/testsuite/gdb.reverse/func-map-to-same-line.c (timeout =
> 300)
> builtin_spawn -ignore SIGHUP gcc -fno-stack-protector
> -fdiagnostics-color=never -gcolumn-info-foo -c -g -o
> /home/simark/build/binutils-
> gdb/gdb/testsuite/outputs/gdb.reverse/func-map-to-same-line/func-map-
> to-same-line0.o /home/simark/src/binutils-
> gdb/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
>
> gcc: error: unrecognized debug output level 'column-info-foo'
>
> compiler exited with status 1
> output is:
> gcc: error: unrecognized debug output level 'column-info-foo'
>
>
> gdb compile failed, gcc: error: unrecognized debug output level
> 'column-info-foo'
> UNTESTED: gdb.reverse/func-map-to-same-line.exp:
> with_column_info=yes: failed to prepare
Yes, that seems to give us the desired result.
>
> I then thought about the "no-column-info" case. Currently, you error
> out for gccs <= 6. However, shouldn't we just compile without any
> special flag in that case, since there just wasn't any support for
> column-info back then?
OK, but seems like we should also warn the user that the option is not
supported and we are ignoring it. I put
# In this case, don't add the compile line option and
# the result will be the same as using no-column-info
# on a version that supports the option.
warning "gdb_compile option no-column-info not supported, ignoring."
in for this case.
Carl
^ permalink raw reply [flat|nested] 40+ messages in thread
* [PATCH 1/2 ver 2] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-06-23 17:44 ` Simon Marchi
2023-06-23 19:41 ` Carl Love
@ 2023-06-23 20:04 ` Carl Love
2023-07-06 15:07 ` Carl Love
1 sibling, 1 reply; 40+ messages in thread
From: Carl Love @ 2023-06-23 20:04 UTC (permalink / raw)
To: Simon Marchi, Bruno Larsen, gdb-patches, UlrichWeigand, pedro
Cc: luis.machado, cel
Simon, GDB maintainers:
Version 2, updated the compiler check and handling for gcc version 6
and earlier. Retested on Power 10.
Per the comments on version 4 for the gdb.reverse/func-map-to-same-
line.exp, I have added support to proc gdb_compile to enable or disable
generating line information as part of the debug information. The two
new options are column-info and no-column-info.
This patch implements the new options for gdb_compile.
These options have been tested with patch 2 of 2 on PowerPC with the
GCC and clang compilers.
Please let me know if the patch is acceptable for mainline. Thanks.
Carl
-----------------------------
Add gdb_compile options column-info and no-column-info
This patch adds two new options to gdb_compile to specify if the compile
should or should not generate the line table information. The
options are supported on clang and gcc version 7 and newer.
Patch has been tested on PowerPC with both gcc and clang.
---
gdb/testsuite/lib/gdb.exp | 34 ++++++++++++++++++++++++++++++++++
1 file changed, 34 insertions(+)
diff --git a/gdb/testsuite/lib/gdb.exp b/gdb/testsuite/lib/gdb.exp
index aed7e2d043c..5857c59a47a 100644
--- a/gdb/testsuite/lib/gdb.exp
+++ b/gdb/testsuite/lib/gdb.exp
@@ -4794,6 +4794,8 @@ proc quote_for_host { args } {
# debug information
# - text_segment=addr: Tell the linker to place the text segment at ADDR.
# - build-id: Ensure the final binary includes a build-id.
+# - no-column-info: Disable generation of column table information.
+# - column-info: Enable generation of column table information.
#
# And here are some of the not too obscure options understood by DejaGnu that
# influence the compilation:
@@ -5003,6 +5005,38 @@ proc gdb_compile {source dest type options} {
} else {
error "Don't know how to handle text_segment option."
}
+ } elseif { $opt == "column-info" } {
+ # If GCC or clang does not support column-info, compilation
+ # will fail and the usupported column-info option will be
+ # reported as such.
+ if {[test_compiler_info {gcc-*}]} {
+ lappend new_options "additional_flags=-gcolumn-info"
+
+ } elseif {[test_compiler_info {clang-*}]} {
+ lappend new_options "additional_flags=-gcolumn-info"
+
+ } else {
+ error "Don't know how to handle gcolumn-info option."
+ }
+
+ } elseif { $opt == "no-column-info" } {
+ if {[test_compiler_info {gcc-*}]} {
+ if {[test_compiler_info {gcc-[1-6]-*}]} {
+ # In this case, don't add the compile line option and
+ # the result will be the same as using no-column-info
+ # on a version that supports the option.
+ warning "gdb_compile option no-column-info not supported, ignoring."
+ } else {
+ lappend new_options "additional_flags=-gno-column-info"
+ }
+
+ } elseif {[test_compiler_info {clang-*}]} {
+ lappend new_options "additional_flags=-gno-column-info"
+
+ } else {
+ error "Don't know how to handle gno-column-info option."
+ }
+
} else {
lappend new_options $opt
}
--
2.37.2
^ permalink raw reply [flat|nested] 40+ messages in thread
* RE: [PATCH 2/2 v5] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-06-19 17:58 ` Simon Marchi
2023-06-22 20:38 ` Carl Love
@ 2023-06-23 20:04 ` Carl Love
2023-06-23 20:04 ` [PATCH 2/2 v6] " Carl Love
2 siblings, 0 replies; 40+ messages in thread
From: Carl Love @ 2023-06-23 20:04 UTC (permalink / raw)
To: Bruno Larsen, gdb-patches, UlrichWeigand, pedro; +Cc: luis.machado, cel
Simon:
Thanks for the reply to my previous unintended send of this note before
it was done. I did fix the forward declaration as you said. The issue
with removing it was making sure the actual declaration is static.
On Mon, 2023-06-19 at 13:58 -0400, Simon Marchi wrote:
> On 5/16/23 18:54, Carl Love wrote:
<snip>
> >
> > diff --git a/gdb/infrun.c b/gdb/infrun.c
> > index efe2c00c489..31cd817c733 100644
> > --- a/gdb/infrun.c
> > +++ b/gdb/infrun.c
> > @@ -114,6 +114,9 @@ static struct async_event_handler
> > *infrun_async_inferior_event_token;
> > Starts off as -1, indicating "never enabled/disabled". */
> > static int infrun_is_async = -1;
> >
> > +static CORE_ADDR update_line_range_start (CORE_ADDR pc,
> > + struct
> > execution_control_state *ecs);
> > +
>
> This forward-declaration is not needed.
Per second email, removed the forward declaration and made the actual
declaration static.
>
> > /* See infrun.h. */
> >
> > void
> > @@ -6769,6 +6772,25 @@ handle_signal_stop (struct
> > execution_control_state *ecs)
> > process_event_stop_test (ecs);
> > }
> >
> > +CORE_ADDR
> > +update_line_range_start (CORE_ADDR pc, struct
> > execution_control_state *ecs)
>
> Please add a comment for the function.
OK, added comment.
>
> > +{
> > + /* The line table may have multiple entries for the same source
> > code line.
> > + Given the PC, check the line table and return the PC that
> > corresponds
> > + to the line table entry for the source line that PC is
> > in. */
> > + CORE_ADDR start_line_pc = ecs->event_thread-
> > >control.step_range_start;
> > + gdb::optional<CORE_ADDR> real_range_start;
> > +
> > + /* Call find_line_range_start to get the smallest address in the
> > + linetable for multiple Line X entries in the line table. */
> > + real_range_start = find_line_range_start (pc);
> > +
> > + if (real_range_start.has_value ())
> > + start_line_pc = *real_range_start;
> > +
> > + return start_line_pc;
>
> When I read this, I wonder: why was control.step_range_start not set
> to
> the "real" range start in the first place (not only in the context of
> reverse execution, every time it is set)? It would seem more robust
> than patching it afterwards in some very specific spots.
>
> I could see some benefits for range-stepping uses cases too (relevant
> when debugging remotely). Using your example here:
>
> Line X - [0x0 - 0x8]
> Line X - [0x8 - 0x10]
> Line X - [0x10 - 0x18]
>
> Imagine we are stopped at 0x14, and we type "next", and 0x14 is a
> conditional jump to 0x5. It seems like current GDB would send a
> "range
> step" request to GDBserver, to step in the [0x10, 0x18[ range. When
> reaching 0x5, execution would stop, and GDB would resume it again
> with
> the [0x0,0x8[ range. When reaching 0x8, it would stop again, GDB
> would
> resume it with [0x8,0x10[, and so on. If GDB could send a "range
> step"
> request with the [0x0,0x18[ range, it would avoid those unnecessary
> intermediary stop.
>
We looked at trying to set control.step_range_start to the real start
range initially. Ulrich brought this point up in our internal review
of the patch.
So, when I am in function finish_backward() in infcmd.c I have no way
to determine what the previous PC was. If I assume it was the previous
value, i.e. pc - 4byes (on PowerPC). I get a gdb internal error. It
seems that I am not allowed to change the line range to something that
does not include the current pc value.
../../binutils-gdb-reverse-multiple-contiguous/gdb/infrun.c:2740:
internal-error: resume_1:
Assertion `pc_in_thread_step_range (pc, tp)' failed.
In order to make that work, we concluded that it would probably entail
a much bigger change to how reverse execution works which would be
beyond the scope of what this patch is trying to fix. So, being able
to do what I believe you want to do is in theory possible but it would
require a larger, independent change to what this patch is trying to
fix.
> > +}
> > +
> > /* Come here when we've got some debug event / signal we can
> > explain
> > (IOW, not a random signal), and test whether it should cause a
> > stop, or whether we should resume the inferior (transparently).
> > @@ -7570,6 +7592,28 @@ process_event_stop_test (struct
> > execution_control_state *ecs)
> >
> > if (stop_pc_sal.is_stmt)
> > {
> > + if (execution_direction == EXEC_REVERSE)
> > + {
> > + /* We are stepping backwards make sure we have reached
> > the
> > + beginning of the line. */
> > + CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
> > + CORE_ADDR start_line_pc
> > + = update_line_range_start (stop_pc, ecs);
> > +
> > + if (stop_pc != start_line_pc)
> > + {
> > + /* Have not reached the beginning of the source code
> > line.
> > + Set a step range. Execution should stop in any
> > function
> > + calls we execute back into before reaching the
> > beginning
> > + of the line. */
> > + ecs->event_thread->control.step_range_start =
> > start_line_pc;
> > + ecs->event_thread->control.step_range_end = stop_pc;
> > + set_step_info (ecs->event_thread, frame,
> > stop_pc_sal);
> > + keep_going (ecs);
> > + return;
> > + }
> > + }
> > +
> > /* We are at the start of a statement.
> >
> > So stop. Note that we don't stop if we step into the
> > middle of a
> > @@ -7632,6 +7676,19 @@ process_event_stop_test (struct
> > execution_control_state *ecs)
> > set_step_info (ecs->event_thread, frame, stop_pc_sal);
> >
> > infrun_debug_printf ("keep going");
> > +
> > + if (execution_direction == EXEC_REVERSE)
> > + {
> > + CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
> > +
> > + /* Make sure the stop_pc is set to the beginning of the
> > line. */
> > + if (stop_pc != ecs->event_thread->control.step_range_start)
> > + {
> > + stop_pc = update_line_range_start (stop_pc, ecs);
> > + ecs->event_thread->control.step_range_start = stop_pc;
> > + }
> > + }
> > +
> > keep_going (ecs);
> > }
> >
> > diff --git a/gdb/symtab.c b/gdb/symtab.c
> > index 27611a34ec4..91d35616eb9 100644
> > --- a/gdb/symtab.c
> > +++ b/gdb/symtab.c
> > @@ -3282,6 +3282,55 @@ find_pc_line (CORE_ADDR pc, int notcurrent)
> > return sal;
> > }
> >
> > +/* Compare two symtab_and_line entries. Return true if both have
> > + the same line number and the same symtab pointer. That means
> > we
> > + are dealing with two entries from the same line and from the
> > same
> > + source file.
> > +
> > + Return false otherwise. */
> > +
> > +static bool
> > +sal_line_symtab_matches_p (const symtab_and_line &sal1,
> > + const symtab_and_line &sal2)
> > +{
> > + return (sal1.line == sal2.line && sal1.symtab == sal2.symtab);
>
> Unnecessary parenthesis.
Fixed.
>
> > +}
> > +
> > +/* See symtah.h. */
> > +
> > +gdb::optional<CORE_ADDR>
> > +find_line_range_start (CORE_ADDR pc)
> > +{
> > + struct symtab_and_line current_sal = find_pc_line (pc, 0);
> > +
> > + if (current_sal.line == 0)
> > + return {};
> > +
> > + struct symtab_and_line prev_sal = find_pc_line (current_sal.pc -
> > 1, 0);
> > +
> > + /* If the previous entry is for a different line, that means we
> > are already
> > + at the entry with the start PC for this line. */
> > + if (!sal_line_symtab_matches_p (prev_sal, current_sal))
> > + return current_sal.pc;
> > +
> > + /* Otherwise, keep looking for entries for the same line but
> > with
> > + smaller PC's. */
> > + bool done = false;
> > + CORE_ADDR prev_pc;
> > + while (!done)
> > + {
> > + prev_pc = prev_sal.pc;
> > +
> > + prev_sal = find_pc_line (prev_pc - 1, 0);
> > +
> > + /* Did we notice a line change? If so, we are done with the
> > search. */
> > + if (!sal_line_symtab_matches_p (prev_sal, current_sal))
> > + done = true;
> > + }
> > +
> > + return prev_pc;
>
> Algorithmic complexity question: given that line tables are sorted by
> address, would it work to start at the current line table item, and
> go
> look at the previous ones until we find one that is no longer
> contiguous and same line? find_pc_line is somewhat heavy, so if we
> don't need to do it repeatedly...
Yes, line tables are sorted by address. The issue with just looking
back thru the current line table is branches/function calls. Suppose
we arrived at the current the source code line for a given entry in the
line table via a branch or a function call. If we looked at the source
code line number for the previous entry in the line table, it may be
the same in this case if we jumped into the line. The pc for the
previous entry in the line table would not for the line we came from
because we arrived via a branch/function call. The address we came
from could even be in another symbol table. We need to walk back one
pc at a time, to see where we came from, and checking to see if the
symbol table entry changed or the line number changed to determine
where the beginning of the line is. So, we need to call
find_pc_line(), to find out if the previous PC is in a different line
or if it corresponds to a different symbol table.
Just searching back in the "current" line table for a line number
change to figure out the first pc in the line is not sufficient. I
don't see anyway to avoid calling find_pc_line() to determine if the
symbol table for the previous PC changed or not.
Note, I did play with just walking back thru the line table and that
does work for the test cases. But I don't believe it will work in
general. The pointer to the base of the linetable for the symbol table
and the maximum number of lines in the linetable are accessible in
function find_pc_sect_line. Given that information, then we search the
line table for the first linetable entry that corresponds to the line
number we want and then return the pc for that linetable entry. We
would need to do a binary search since we don't know the actual offset
into the table for the current pc. The time (log base 2 of the
linetable size) to do the search is obviously a function of the size of
the linetable. If we have function find_pc_sect_line (which is called
by our intial call to find_pc_line()) record the offset into the line
table where the current line number occurs and store it in struct
symtab, then we can very efficiently start at the correct linetable
entry and walk back as you suggested. But what we can't tell is if the
execution history actually came in via the pc values in the earlier
linetable entries. If execution didn't come in via an earlier entry in
the linetable, the returned pc for the first linetable entry of the
desired source code line may never be reached during our reverse
execution. Obviously in that case, gdb will not stop at the correct
location.
>
> > +}
> > +
> > /* See symtab.h. */
> >
> > struct symtab *
> > diff --git a/gdb/symtab.h b/gdb/symtab.h
> > index 404d0ab30a8..f54305636da 100644
> > --- a/gdb/symtab.h
> > +++ b/gdb/symtab.h
> > @@ -2346,6 +2346,22 @@ extern struct symtab_and_line find_pc_line
> > (CORE_ADDR, int);
> > extern struct symtab_and_line find_pc_sect_line (CORE_ADDR,
> > struct obj_section *,
> > int);
> >
> > +/* Given PC, and assuming it is part of a range of addresses that
> > is part of a
> > + line, go back through the linetable and find the starting PC of
> > that
> > + line.
> > +
> > + For example, suppose we have 3 PC ranges for line X:
> > +
> > + Line X - [0x0 - 0x8]
> > + Line X - [0x8 - 0x10]
> > + Line X - [0x10 - 0x18]
> > +
> > + If we call the function with PC == 0x14, we want to return 0x0,
> > as that is
> > + the starting PC of line X, and the ranges are contiguous.
>
> I think that putting this example in the comment is great. It makes
> it
> much more obvious what the function specifically does.
>
> > +*/
> > +
> > +extern gdb::optional<CORE_ADDR> find_line_range_start (CORE_ADDR
> > pc);
> > +
> > /* Wrapper around find_pc_line to just return the symtab. */
> >
> > extern struct symtab *find_pc_line_symtab (CORE_ADDR);
> > diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> > b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> > new file mode 100644
> > index 00000000000..da944874e86
> > --- /dev/null
> > +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
> > @@ -0,0 +1,36 @@
> > +/* Copyright 2008-2023 Free Software Foundation, Inc.
> > +
> > + This program is free software; you can redistribute it and/or
> > modify
> > + it under the terms of the GNU General Public License as
> > published by
> > + the Free Software Foundation; either version 3 of the License,
> > or
> > + (at your option) any later version.
> > +
> > + This program is distributed in the hope that it will be useful,
> > + but WITHOUT ANY WARRANTY; without even the implied warranty of
> > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> > + GNU General Public License for more details.
> > +
> > + You should have received a copy of the GNU General Public
> > License
> > + along with this program. If not, see <
> > http://www.gnu.org/licenses/
> > >.
> > +
> > + This test is used to test the reverse-step and reverse-next
> > instruction
> > + execution for a source line that contains multiple function
> > calls. */
> > +
> > +void
> > +func1 ()
> > +{
> > +} /* END FUNC1 */
> > +
> > +void
> > +func2 ()
> > +{
> > +} /* END FUNC2 */
> > +
> > +int main ()
>
> int
> main (void)
>
Fixed.
> > +{
> > + int a, b;
> > + a = 1;
> > + b = 2;
> > + func1 (); func2 ();
> > + a = a + b; /* START REVERSE TEST */
> > +}
> > diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> > b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> > new file mode 100644
> > index 00000000000..89e226b0f84
> > --- /dev/null
> > +++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
> > @@ -0,0 +1,140 @@
> > +# Copyright 2008-2023 Free Software Foundation, Inc.
> > +
> > +# This program is free software; you can redistribute it and/or
> > modify
> > +# it under the terms of the GNU General Public License as
> > published by
> > +# the Free Software Foundation; either version 3 of the License,
> > or
> > +# (at your option) any later version.
> > +#
> > +# This program is distributed in the hope that it will be useful,
> > +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> > +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> > +# GNU General Public License for more details.
> > +#
> > +# You should have received a copy of the GNU General Public
> > License
> > +# along with this program. If not, see <
> > http://www.gnu.org/licenses/
> > >. */
> > +
> > +# This file is part of the GDB testsuite. It tests reverse
> > stepping.
> > +# Lots of code borrowed from "step-test.exp".
> > +
> > +# This test checks to make sure there is no regression failures
> > for
> > +# the reverse-next command when stepping back over two functions
> > in
> > +# the same line.
> > +
> > +require supports_reverse
> > +
> > +# This test uses the gcc no-column-info command which was added in
> > gcc 7.1.
> > +
> > +proc run_tests {} {
> > + global srcfile
> > + global executable
> > +
> > + runto_main
>
> We typically check for runto_main's success:
Fixed a couple instances of runto_main in this test case and in the
other test case.
>
> if { ![runto_main] } {
> return
> }
>
> runto_main logs a FAIL on failure. There are a few runto_mains in
> the
> patch.
>
> > + set target_remote [gdb_is_target_remote]
>
> target_remote seems unused
Removed.
>
> > +
> > + with_test_prefix "test1" {
> > + gdb_test_no_output "record" "turn on process record"
> > + }
>
> with_test_prefix with a single test in it is really just the same as:
>
> gdb_test_no_output "record" "test1: turn on process record"
>
> In fact, you have some other tests with the "test1:" or "test2:"
> prefix,
> I think they should be moved to the with_test_prefix. And maybe use
> "next" and "step" instead of "test1" and "test2".
Yup, cleaner to have the with_test_prefix cover the whole test.
Changed test1 to next-test and test2 to step-next.
>
> > +
> > + # This regression test verifies the reverse-step and reverse-
> > next commands
> > + # work properly when executing backwards thru a source line
> > containing
> > + # two function calls on the same source line, i.e. func1 ();
> > func2 ();
> > + # This test is compiled so the dwarf info not contain the line
> > table
> > + # information.
> > +
> > + # Test 1, reverse-next command
> > + # Set breakpoint at the line after the function calls.
> > + set bp_start_reverse_test [gdb_get_line_number "START REVERSE
> > TEST" \
> > + $srcfile]
> > + gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> > +
> > + # Continue to break point for reverse-next test.
> > + # Command definition: reverse-next [count]
> > + # Run backward to the beginning of the previous line
> > executed in the
> > + # current (innermost) stack frame. If the line contains
> > function calls,
> > + # they will be “un-executed” without stopping. Starting from
> > the first
> > + # line of a function, reverse-next will take you back to the
> > caller of
> > + # that function, before the function was called, just as the
> > normal next
> > + # command would take you from the last line of a function
> > back to its
> > + # return to its caller 2 .
> > + gdb_continue_to_breakpoint \
> > + "test1: stopped at command reverse-next test start location" \
> > + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> > +
> > + # The reverse-next should step all the way back to the
> > beginning of the
> > + # line, i.e. at the beginning of the func1 call.
> > + gdb_test "reverse-next" ".*func1 \\(\\); func2 \\(\\);.*" \
> > + "test1: reverse-next to line with two functions"
> > +
> > + # We should be stopped at the first instruction of the line. A
> > reverse-step
> > + # should step back and stop at the beginning of the previous
> > line b = 2,
> > + # i.e. not in func1 ().
> > + gdb_test "reverse-stepi" ".*b = 2;.*" \
> > + "test1: reverse-stepi to previous line b = 2"
> > +
> > +
> > + # Setup for test 2
> > + clean_restart $executable
> > + runto_main
> > +
> > + with_test_prefix "test2" {
> > + gdb_test_no_output "record" "turn on process record"
> > + }
> > +
> > + # Test 2, reverse-step command
> > + # Set breakpoint at the line after the function calls.
> > + gdb_breakpoint $srcfile:$bp_start_reverse_test temporary
> > +
> > + # Continue to the start of the reverse-step test.
> > + # Command definition: reverse-step [count]
> > + # Run the program backward until control reaches the start
> > of a
> > + # different source line; then stop it, and return control
> > to gdb.
> > + # Like the step command, reverse-step will only stop at the
> > beginning
> > + # of a source line. It “un-executes” the previously
> > executed source
> > + # line. If the previous source line included calls to
> > debuggable
> > + # functions, reverse-step will step (backward) into the
> > called function,
> > + # stopping at the beginning of the last statement in the
> > called
> > + # function (typically a return statement). Also, as with
> > the step
> > + # command, if non-debuggable functions are called, reverse-
> > step will
> > + # run thru them backward without stopping.
> > +
> > + gdb_continue_to_breakpoint \
> > + "test2: stopped at command reverse-step test start location" \
> > + ".*$srcfile:$bp_start_reverse_test\r\n.*"
> > +
> > + # The first reverse step should take us call of func2 ().
> > + gdb_test "reverse-step" ".*END FUNC2.*" \
> > + "test2: reverse-step into func2 "
> > +
> > + # The second reverse step should take us into func1 ().
> > + gdb_test "reverse-step" ".*END FUNC1.*" \
> > + "test2: reverse-step into func1 "
> > +
> > + # The third reverse step should take us call of func1 ().
> > + gdb_test "reverse-step" ".*func1 \\(\\); func2 \\(\\);.*" \
> > + "test2: reverse-step to line func1(); func2(), at call for
> > func1 "
> > +
> > + # We should be stopped at the first instruction of the line. A
> > reverse
> > + # stepi should take us to b = 2 ().
> > + gdb_test "reverse-stepi" ".*b = 2;.*" \
> > + "test2: reverse-stepi to line b = 2 "
> > +}
> > +
> > +set srcfile func-map-to-same-line.c
> > +set executable func-map-to-same-line
>
> Wondering if this test should use standard_testfile (like almost
> every
> other tests) to set these.
OK, changed to use the standard_testfile.
>
> > +
> > +# test with and without gcc column info enabled
> > +foreach_with_prefix with_column_info {yes no} {
> > + if {$with_column_info == "yes"} {
> > + set options [list debug column-info]
> > + } else {
> > + set options [list debug no-column-info]
> > + }
>
> I didn't think of this when proposing the foreach_with_prefix, but
> you
> could perhaps use:
>
> foreach_with_prefix column_info_flag {column-info no-column-info}
>
> ... to avoid this boilerplate. You can then use $column_info_flag
> directly when setting options.
Yes, that makes things a bit cleaner. Changed.
>
> > +
> > + if {[build_executable "failed to prepare" $executable $srcfile
> > \
> > + $options] == -1} {
> > + return -1
> > + }
> > +
> > + clean_restart $executable
>
> clean_restart can go in run_tests.
Moved.
>
> > + run_tests
> > +}
> > diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.c
> > b/gdb/testsuite/gdb.reverse/map-to-same-line.c
> > new file mode 100644
> > index 00000000000..f20d778f40e
> > --- /dev/null
> > +++ b/gdb/testsuite/gdb.reverse/map-to-same-line.c
> > @@ -0,0 +1,58 @@
> > +/* Copyright 2008-2023 Free Software Foundation, Inc.
>
> Just wondering if the copyright years are right.
These are new test files so, yea, should start with 2023.
>
> > +
> > + This program is free software; you can redistribute it and/or
> > modify
> > + it under the terms of the GNU General Public License as
> > published by
> > + the Free Software Foundation; either version 3 of the License,
> > or
> > + (at your option) any later version.
> > +
> > + This program is distributed in the hope that it will be useful,
> > + but WITHOUT ANY WARRANTY; without even the implied warranty of
> > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> > + GNU General Public License for more details.
> > +
> > + You should have received a copy of the GNU General Public
> > License
> > + along with this program. If not, see <
> > http://www.gnu.org/licenses/
> > >. */
> > +
> > +/* The purpose of this test is to create a DWARF line table that
> > contains two
> > + or more entries for the same line. When stepping (forwards or
> > backwards),
> > + GDB should step over the entire line and not just a particular
> > entry in the
> > + line table. */
> > +
> > +int
> > +main ()
>
> void in the parenthesis
Fixed in both tests.
>
> > +{ /* TAG: main prologue */
> > + asm ("main_label: .globl main_label");
> > + int i = 1, j = 2, k;
> > + float f1 = 2.0, f2 = 4.1, f3;
> > + const char *str_1 = "foo", *str_2 = "bar", *str_3;
> > +
> > + asm ("line1: .globl line1");
> > + k = i; f3 = f1; str_3 = str_1; /* TAG: line 1 */
> > +
> > + asm ("line2: .globl line2");
> > + k = j; f3 = f2; str_3 = str_2; /* TAG: line 2 */
> > +
> > + asm ("line3: .globl line3");
> > + k = i; f3 = f1; str_3 = str_1; /* TAG: line 3 */
> > +
> > + asm ("line4: .globl line4");
> > + k = j; f3 = f2; str_3 = str_2; /* TAG: line 4 */
> > +
> > + asm ("line5: .globl line5");
> > + k = i; f3 = f1; str_3 = str_1; /* TAG: line 5 */
> > +
> > + asm ("line6: .globl line6");
> > + k = j; f3 = f2; str_3 = str_2; /* TAG: line 6 */
> > +
> > + asm ("line7: .globl line7");
> > + k = i; f3 = f1; str_3 = str_1; /* TAG: line 7 */
> > +
> > + asm ("line8: .globl line8");
> > + k = j; f3 = f2; str_3 = str_2; /* TAG: line 8 */
> > +
> > + asm ("main_return: .globl main_return");
> > + k = j; f3 = f2; str_3 = str_2; /* TAG: main return */
> > +
> > + asm ("end_of_sequence: .globl end_of_sequence");
> > + return 0; /* TAG: main return */
> > +}
> > diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> > b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> > new file mode 100644
> > index 00000000000..16a359d90ec
> > --- /dev/null
> > +++ b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> > @@ -0,0 +1,156 @@
> > +# Copyright 2008-2023 Free Software Foundation, Inc.
> > +
> > +# This program is free software; you can redistribute it and/or
> > modify
> > +# it under the terms of the GNU General Public License as
> > published by
> > +# the Free Software Foundation; either version 3 of the License,
> > or
> > +# (at your option) any later version.
> > +#
> > +# This program is distributed in the hope that it will be useful,
> > +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> > +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> > +# GNU General Public License for more details.
> > +#
> > +# You should have received a copy of the GNU General Public
> > License
> > +# along with this program. If not, see <
> > http://www.gnu.org/licenses/
> > = >.
> > +
> > +# When stepping (forwards or backwards), GDB should step over the
> > entire line
> > +# and not just a particular entry in the line table. This test was
> > added to
> > +# verify the find_line_range_start function properly sets the step
> > range for a
> > +# line that consists of multiple statements, i.e. multiple entries
> > in the line
> > +# table. This test creates a DWARF line table that contains two
> > entries for
> > +# the same line to do the needed testing.
> > +
> > +# This test can only be run on targets which support DWARF-2 and
> > use gas.
> > +load_lib dwarf.exp
> > +require dwarf2_support
> > +
> > +# The DWARF assembler requires the gcc compiler.
> > +require is_c_compiler_gcc
> > +
> > +# This test suitable only for process that can do reverse
> > execution
> > +require supports_reverse
> > +
> > +standard_testfile .c .S
> > +
> > +if { [prepare_for_testing "failed to prepare" ${testfile}
> > ${srcfile}] } {
> > + return -1
> > +}
> > +
> > +set asm_file [standard_output_file $srcfile2]
> > +Dwarf::assemble $asm_file {
> > + global srcdir subdir srcfile
> > + declare_labels integer_label L
> > +
> > + # Find start address and length of program
> > + lassign [function_range main [list
> > ${srcdir}/${subdir}/$srcfile]] \
> > + main_start main_len
> > + set main_end "$main_start + $main_len"
> > +
> > + cu {} {
> > + compile_unit {
> > + {language @DW_LANG_C}
> > + {name map-to-same-line.c}
> > + {stmt_list $L DW_FORM_sec_offset}
> > + {low_pc 0 addr}
> > + } {
> > + subprogram {
> > + {external 1 flag}
> > + {name main}
> > + {low_pc $main_start addr}
> > + {high_pc $main_len DW_FORM_data4}
> > + }
> > + }
> > + }
> > +
> > + lines {version 2 default_is_stmt 1} L {
> > + include_dir "${srcdir}/${subdir}"
> > + file_name "$srcfile" 1
> > +
> > + # Generate the line table program with distinct source lines
> > being
> > + # mapped to the same line entry. Line 1, 5 and 8 contain 1
> > statement
> > + # each. Line 2 contains 2 statements. Line 3 contains 3
> > statements.
> > + program {
> > + DW_LNE_set_address $main_start
> > + line [gdb_get_line_number "TAG: main prologue"]
> > + DW_LNS_copy
> > + DW_LNE_set_address line1
> > + line [gdb_get_line_number "TAG: line 1" ]
> > + DW_LNS_copy
> > + DW_LNE_set_address line2
> > + line [gdb_get_line_number "TAG: line 2" ]
> > + DW_LNS_copy
> > + DW_LNE_set_address line3
> > + line [gdb_get_line_number "TAG: line 2" ]
> > + DW_LNS_copy
> > + DW_LNE_set_address line4
> > + line [gdb_get_line_number "TAG: line 3" ]
> > + DW_LNS_copy
> > + DW_LNE_set_address line5
> > + line [gdb_get_line_number "TAG: line 3" ]
> > + DW_LNS_copy
> > + DW_LNE_set_address line6
> > + line [gdb_get_line_number "TAG: line 3" ]
> > + DW_LNS_copy
> > + DW_LNE_set_address line7
> > + line [gdb_get_line_number "TAG: line 5" ]
> > + DW_LNS_copy
> > + DW_LNE_set_address line8
> > + line [gdb_get_line_number "TAG: line 8" ]
> > + DW_LNS_copy
> > + DW_LNE_set_address main_return
> > + line [gdb_get_line_number "TAG: main return"]
> > + DW_LNS_copy
> > + DW_LNE_set_address end_of_sequence
> > + DW_LNE_end_sequence
> > + }
> > + }
> > +}
> > +
> > +if { [prepare_for_testing "failed to prepare" ${testfile} \
> > + [list $srcfile $asm_file] {nodebug} ] } {
> > + return -1
> > +}
> > +
> > +runto_main
> > +
> > +# Print the line table
> > +gdb_test_multiple "maint info line-table ${testfile}" "" {
> > + -re "\r\n$decimal\[ \t\]+$decimal\[ \t\]+($hex)\[
> > \t\]+Y\[^\r\n\]*" {
> > + lappend is_stmt $expect_out(1,string)
> > + exp_continue
> > + }
> > + -re -wrap "" {
> > + }
> > +}
> > +
> > +# Do the reverse-step test
> > +gdb_test_no_output "record" "turn on process record"
> > +
> > +set bp_main_return [gdb_get_line_number "TAG: main return"
> > $srcfile]
> > +gdb_breakpoint $srcfile:$bp_main_return
> > +gdb_continue_to_breakpoint "run to end of main, reverse-step
> > test" ".*$srcfile:$bp_main_return.*"
> > +gdb_test "display \$pc" ".*pc =.*" "display pc, reverse-step test"
> > +
> > +# At this point, GDB has already recorded the execution up until
> > the return
> > +# statement. Reverse-step and test if GDB transitions between
> > lines in the
> > +# expected order. It should reverse-step across lines 8, 5, 3, 2
> > and 1.
> > +foreach line {8 5 3 2 1} {
> > + gdb_test "reverse-step" ".*TAG: line $line.*" "reverse step to
> > line $line"
> > +}
> > +
> > +## Clean restart, test reverse-next command
> > +clean_restart ${testfile}
> > +runto_main
> > +gdb_test_no_output "record" "turn on process record, reverst-next
> > test"
> > +
> > +set bp_main_return [gdb_get_line_number "TAG: main return"
> > $srcfile]
> > +gdb_breakpoint $srcfile:$bp_main_return
> > +gdb_continue_to_breakpoint "run to end of main, reverse-next
> > test" ".*$srcfile:$bp_main_return.*"
> > +gdb_test "display \$pc" ".*pc =.*" "display pc, reverse-next test"
> > +
> > +# At this point, GDB has already recorded the execution up until
> > the return
> > +# statement. Reverse-next and test if GDB transitions between
> > lines in the
> > +# expected order. It should reverse-next across lines 8, 5, 3, 2
> > and 1.
> > +foreach line {8 5 3 2 1} {
> > + gdb_test "reverse-next" ".*TAG: line $line.*" "reverse next to
> > line $line"
> > +}
>
> It seems like the step and next tests are identical, so I guess it
> could
> be factored out using:
>
> foreach_with_prefix method {step next} {
> ...
> }
>
Yes, changed to use foreach_with_prefix.
Carl
^ permalink raw reply [flat|nested] 40+ messages in thread
* [PATCH 2/2 v6] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-06-19 17:58 ` Simon Marchi
2023-06-22 20:38 ` Carl Love
2023-06-23 20:04 ` Carl Love
@ 2023-06-23 20:04 ` Carl Love
2 siblings, 0 replies; 40+ messages in thread
From: Carl Love @ 2023-06-23 20:04 UTC (permalink / raw)
To: Simon Marchi, Bruno Larsen, gdb-patches, UlrichWeigand, pedro
Cc: luis.machado, cel
Bruno, Simon, GDB maintainers:
Version 6, fixed various code style issues in the GDB source. The
testcases were updated to use with_test_prefix for each gdb test in the
step and next test cases, switch using the standard_testfile, use
foreach_with_prefix to combine otherwise identical tests. Retested on
Power 10.
Version 5, changed comments in test case func-map-to-same-line.c.
Patch 1/2 implemented the new options for gdb_compile. Updated the
call to proc run_tests to use the new gdb_compile options in a
foreach_with_prefix loop.
Version 4, additional fixes for gcc version check, wrap function calls
using "with_test_prefix", move load_lib dwarf.exe. Fixed typo noted by
Luis.
Version 3, added the gcc version check as discussed further from
version 2 of the patch. Also updated the tests to check for supporting
reverse execution rather than requiring recording. I also noticed
there were a couple more instances of a requirement check, i.e. if []
which I changed to "require" per the current style for checking on the
test requirements.
The following patch fixes issues on PowerPC with the reverse-step and
reverse-next instructions when there are multiple assignment statements
on the same line and when there are multiple function calls on the same
line. The commit log below discusses these issues in further depth.
The discussion included what the correct operation should be for these
commands based on the GDB documentation. The proposed patch at that
time changed how the commands worked on other platforms such as X86 in
a way they no longer matched the documentation.
The issue is the line table contains multiple entries for the same
source line. The patch adds a function to search the line table to
find the address of the first instruction of a line. When setup up the
reverse stepping range, the function is called to make sure the start
of the range corresponds to the address of the first instruction for
the line. This approach was used. When Luis initially developed the
patch, he considered merging the contiguous ranges in the line table
when reading the line tables. He decided it was better to work with the
data directly in the line table rather than creating and using a
modified version of the line table.
The following patch fixes the execution of the reveres-step and
reverse-next commands for both senarios of multiple statements on the
same line for PowerPC and aarch64-linux. Unlike the previous patch, it
does not change the operation of the commands on other platforms, i.e.
X86. The patch adds new test cases for both scenarios to verify they
work correctly.
The patch has been tested on PowerPC, Intel X86 and aarch64-linux with
no new regression failures.
Please let me know if the patch is acceptable for mainline. Thanks.
Carl
--------------------------
Fix reverse stepping multiple contiguous PC ranges over the line table.
There are a couple of scenarios where the GDB reverse-step and reverse-next
commands do not work correctly.
Scenario 1 issue description by Luis Machado:
When running GDB's testsuite on aarch64-linux/Ubuntu 20.04 (also spotted on
the ppc backend), I noticed some failures in gdb.reverse/solib-precsave.exp
and gdb.reverse/solib-reverse.exp.
The failure happens around the following code:
38 b[1] = shr2(17); /* middle part two */
40 b[0] = 6; b[1] = 9; /* generic statement, end part two */
42 shr1 ("message 1\n"); /* shr1 one */
Normal execution:
- step from line 38 will land on line 40.
- step from line 40 will land on line 42.
Reverse execution:
- step from line 42 will land on line 40.
- step from line 40 will land on line 40.
- step from line 40 will land on line 38.
The problem here is that line 40 contains two contiguous but distinct
PC ranges in the line table, like so:
Line 40 - [0x7ec ~ 0x7f4]
Line 40 - [0x7f4 ~ 0x7fc]
The two distinct ranges are generated because GCC started outputting source
column information, which GDB doesn't take into account at the moment.
When stepping forward from line 40, we skip both of these ranges and land on
line 42. When stepping backward from line 42, we stop at the start PC of the
second (or first, going backwards) range of line 40.
Since we've reached ecs->event_thread->control.step_range_start, we stop
stepping backwards.
---------------------------------------------------------
Scenario 2 issue described by Pedro Alves:
The following explanation of the issue was taken from the gdb mailing list
discussion of the withdrawn patch to change the behavior of the reverse-step
and reverse-next commands. Specifically, message from Pedro Alves
<pedro@palves.net> where he demonstrates the issue where you have multiple
function calls on the same source code line:
https://sourceware.org/pipermail/gdb-patches/2023-January/196110.html
The source line looks like:
func1 (); func2 ();
so stepping backwards over that line should always stop at the first
instruction of the line, not in the middle. Let's simplify this.
Here's the full source code of my example:
(gdb) list 1
1 void func1 ()
2 {
3 }
4
5 void func2 ()
6 {
7 }
8
9 int main ()
10 {
11 func1 (); func2 ();
12 }
Compiled with:
$ gcc reverse.c -o reverse -g3 -O0
$ gcc -v
...
gcc version 11.3.0 (Ubuntu 11.3.0-1ubuntu1~22.04)
Now let's debug it with target record, using current gdb git master (f3d8ae90b236),
without your patch:
$ gdb ~/reverse
GNU gdb (GDB) 14.0.50.20230124-git
...
Reading symbols from /home/pedro/reverse...
(gdb) start
Temporary breakpoint 1 at 0x1147: file reverse.c, line 11.
Starting program: /home/pedro/reverse
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
Temporary breakpoint 1, main () at reverse.c:11
11 func1 (); func2 ();
(gdb) record
(gdb) disassemble /s
Dump of assembler code for function main:
reverse.c:
10 {
0x000055555555513f <+0>: endbr64
0x0000555555555143 <+4>: push %rbp
0x0000555555555144 <+5>: mov %rsp,%rbp
11 func1 (); func2 ();
=> 0x0000555555555147 <+8>: mov $0x0,%eax
0x000055555555514c <+13>: call 0x555555555129 <func1>
0x0000555555555151 <+18>: mov $0x0,%eax
0x0000555555555156 <+23>: call 0x555555555134 <func2>
0x000055555555515b <+28>: mov $0x0,%eax
12 }
0x0000555555555160 <+33>: pop %rbp
0x0000555555555161 <+34>: ret
End of assembler dump.
(gdb) n
12 }
So far so good, a "next" stepped over the whole of line 11 and stopped at line 12.
Let's confirm where we are now:
(gdb) disassemble /s
Dump of assembler code for function main:
reverse.c:
10 {
0x000055555555513f <+0>: endbr64
0x0000555555555143 <+4>: push %rbp
0x0000555555555144 <+5>: mov %rsp,%rbp
11 func1 (); func2 ();
0x0000555555555147 <+8>: mov $0x0,%eax
0x000055555555514c <+13>: call 0x555555555129 <func1>
0x0000555555555151 <+18>: mov $0x0,%eax
0x0000555555555156 <+23>: call 0x555555555134 <func2>
0x000055555555515b <+28>: mov $0x0,%eax
12 }
=> 0x0000555555555160 <+33>: pop %rbp
0x0000555555555161 <+34>: ret
End of assembler dump.
Good, we're at the first instruction of line 12.
Now let's undo the "next", with "reverse-next":
(gdb) reverse-next
11 func1 (); func2 ();
Seemingly stopped at line 11. Let's see exactly where:
(gdb) disassemble /s
Dump of assembler code for function main:
reverse.c:
10 {
0x000055555555513f <+0>: endbr64
0x0000555555555143 <+4>: push %rbp
0x0000555555555144 <+5>: mov %rsp,%rbp
11 func1 (); func2 ();
0x0000555555555147 <+8>: mov $0x0,%eax
0x000055555555514c <+13>: call 0x555555555129 <func1>
=> 0x0000555555555151 <+18>: mov $0x0,%eax
0x0000555555555156 <+23>: call 0x555555555134 <func2>
0x000055555555515b <+28>: mov $0x0,%eax
12 }
0x0000555555555160 <+33>: pop %rbp
0x0000555555555161 <+34>: ret
End of assembler dump.
(gdb)
And lo, we stopped in the middle of line 11! That is a bug, we should have
stepped back all the way to the beginning of the line. The "reverse-next"
should have fully undone the prior "next" command.
The above issues were fixed by introducing a new function that looks for
adjacent PC ranges for the same line, until we notice a line change. Then
we take that as the start PC of the range. The new start PC for the range
is used for the control.step_range_start when setting up a step range.
The test case gdb.reverse/map-to-same-line.exp is added to test the fix
for the issues in scenario 1.
The test case gdb.reverse/func-map-to-same-line.exp was added to test the
fix for scenario 2 when the binary was compiled with and without line
table information.
bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28426
Co-authored-by: Luis Machado <luis.machado@arm.com>
Co-authored-by: Carl Love <cel@us.ibm.com>
Reviewed-By: Bruno Larsen <blarsen@redhat.com>
---
gdb/infrun.c | 59 +++++++
gdb/symtab.c | 49 ++++++
gdb/symtab.h | 16 ++
.../gdb.reverse/func-map-to-same-line.c | 37 +++++
.../gdb.reverse/func-map-to-same-line.exp | 139 ++++++++++++++++
gdb/testsuite/gdb.reverse/map-to-same-line.c | 58 +++++++
.../gdb.reverse/map-to-same-line.exp | 153 ++++++++++++++++++
7 files changed, 511 insertions(+)
create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-line.c
create mode 100644 gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.c
create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.exp
diff --git a/gdb/infrun.c b/gdb/infrun.c
index efe2c00c489..b06d6969ab8 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -114,6 +114,9 @@ static struct async_event_handler *infrun_async_inferior_event_token;
Starts off as -1, indicating "never enabled/disabled". */
static int infrun_is_async = -1;
+static CORE_ADDR update_line_range_start (CORE_ADDR pc,
+ struct execution_control_state *ecs);
+
/* See infrun.h. */
void
@@ -6769,6 +6772,27 @@ handle_signal_stop (struct execution_control_state *ecs)
process_event_stop_test (ecs);
}
+/* Return the address for the beginning of the line. */
+
+CORE_ADDR
+update_line_range_start (CORE_ADDR pc, struct execution_control_state *ecs)
+{
+ /* The line table may have multiple entries for the same source code line.
+ Given the PC, check the line table and return the PC that corresponds
+ to the line table entry for the source line that PC is in. */
+ CORE_ADDR start_line_pc = ecs->event_thread->control.step_range_start;
+ gdb::optional<CORE_ADDR> real_range_start;
+
+ /* Call find_line_range_start to get the smallest address in the
+ linetable for multiple Line X entries in the line table. */
+ real_range_start = find_line_range_start (pc);
+
+ if (real_range_start.has_value ())
+ start_line_pc = *real_range_start;
+
+ return start_line_pc;
+}
+
/* Come here when we've got some debug event / signal we can explain
(IOW, not a random signal), and test whether it should cause a
stop, or whether we should resume the inferior (transparently).
@@ -7570,6 +7594,28 @@ process_event_stop_test (struct execution_control_state *ecs)
if (stop_pc_sal.is_stmt)
{
+ if (execution_direction == EXEC_REVERSE)
+ {
+ /* We are stepping backwards make sure we have reached the
+ beginning of the line. */
+ CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
+ CORE_ADDR start_line_pc
+ = update_line_range_start (stop_pc, ecs);
+
+ if (stop_pc != start_line_pc)
+ {
+ /* Have not reached the beginning of the source code line.
+ Set a step range. Execution should stop in any function
+ calls we execute back into before reaching the beginning
+ of the line. */
+ ecs->event_thread->control.step_range_start = start_line_pc;
+ ecs->event_thread->control.step_range_end = stop_pc;
+ set_step_info (ecs->event_thread, frame, stop_pc_sal);
+ keep_going (ecs);
+ return;
+ }
+ }
+
/* We are at the start of a statement.
So stop. Note that we don't stop if we step into the middle of a
@@ -7632,6 +7678,19 @@ process_event_stop_test (struct execution_control_state *ecs)
set_step_info (ecs->event_thread, frame, stop_pc_sal);
infrun_debug_printf ("keep going");
+
+ if (execution_direction == EXEC_REVERSE)
+ {
+ CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
+
+ /* Make sure the stop_pc is set to the beginning of the line. */
+ if (stop_pc != ecs->event_thread->control.step_range_start)
+ {
+ stop_pc = update_line_range_start (stop_pc, ecs);
+ ecs->event_thread->control.step_range_start = stop_pc;
+ }
+ }
+
keep_going (ecs);
}
diff --git a/gdb/symtab.c b/gdb/symtab.c
index 27611a34ec4..fa8a892e530 100644
--- a/gdb/symtab.c
+++ b/gdb/symtab.c
@@ -3282,6 +3282,55 @@ find_pc_line (CORE_ADDR pc, int notcurrent)
return sal;
}
+/* Compare two symtab_and_line entries. Return true if both have
+ the same line number and the same symtab pointer. That means we
+ are dealing with two entries from the same line and from the same
+ source file.
+
+ Return false otherwise. */
+
+static bool
+sal_line_symtab_matches_p (const symtab_and_line &sal1,
+ const symtab_and_line &sal2)
+{
+ return sal1.line == sal2.line && sal1.symtab == sal2.symtab;
+}
+
+/* See symtah.h. */
+
+gdb::optional<CORE_ADDR>
+find_line_range_start (CORE_ADDR pc)
+{
+ struct symtab_and_line current_sal = find_pc_line (pc, 0);
+
+ if (current_sal.line == 0)
+ return {};
+
+ struct symtab_and_line prev_sal = find_pc_line (current_sal.pc - 1, 0);
+
+ /* If the previous entry is for a different line, that means we are already
+ at the entry with the start PC for this line. */
+ if (!sal_line_symtab_matches_p (prev_sal, current_sal))
+ return current_sal.pc;
+
+ /* Otherwise, keep looking for entries for the same line but with
+ smaller PC's. */
+ bool done = false;
+ CORE_ADDR prev_pc;
+ while (!done)
+ {
+ prev_pc = prev_sal.pc;
+
+ prev_sal = find_pc_line (prev_pc - 1, 0);
+
+ /* Did we notice a line change? If so, we are done with the search. */
+ if (!sal_line_symtab_matches_p (prev_sal, current_sal))
+ done = true;
+ }
+
+ return prev_pc;
+}
+
/* See symtab.h. */
struct symtab *
diff --git a/gdb/symtab.h b/gdb/symtab.h
index 404d0ab30a8..f54305636da 100644
--- a/gdb/symtab.h
+++ b/gdb/symtab.h
@@ -2346,6 +2346,22 @@ extern struct symtab_and_line find_pc_line (CORE_ADDR, int);
extern struct symtab_and_line find_pc_sect_line (CORE_ADDR,
struct obj_section *, int);
+/* Given PC, and assuming it is part of a range of addresses that is part of a
+ line, go back through the linetable and find the starting PC of that
+ line.
+
+ For example, suppose we have 3 PC ranges for line X:
+
+ Line X - [0x0 - 0x8]
+ Line X - [0x8 - 0x10]
+ Line X - [0x10 - 0x18]
+
+ If we call the function with PC == 0x14, we want to return 0x0, as that is
+ the starting PC of line X, and the ranges are contiguous.
+*/
+
+extern gdb::optional<CORE_ADDR> find_line_range_start (CORE_ADDR pc);
+
/* Wrapper around find_pc_line to just return the symtab. */
extern struct symtab *find_pc_line_symtab (CORE_ADDR);
diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.c b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
new file mode 100644
index 00000000000..17fe17af267
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.c
@@ -0,0 +1,37 @@
+/* Copyright 2023 Free Software Foundation, Inc.
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 3 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program. If not, see <http://www.gnu.org/licenses/>.
+
+ This test is used to test the reverse-step and reverse-next instruction
+ execution for a source line that contains multiple function calls. */
+
+void
+func1 (void)
+{
+} /* END FUNC1 */
+
+void
+func2 (void)
+{
+} /* END FUNC2 */
+
+int
+main (void)
+{
+ int a, b;
+ a = 1;
+ b = 2;
+ func1 (); func2 ();
+ a = a + b; /* START REVERSE TEST */
+}
diff --git a/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
new file mode 100644
index 00000000000..2890b5b1a70
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/func-map-to-same-line.exp
@@ -0,0 +1,139 @@
+# Copyright 2023 Free Software Foundation, Inc.
+
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program. If not, see <http://www.gnu.org/licenses/>. */
+
+# This file is part of the GDB testsuite. It tests reverse stepping.
+# Lots of code borrowed from "step-test.exp".
+
+# This test checks to make sure there is no regression failures for
+# the reverse-next command when stepping back over two functions in
+# the same line.
+
+require supports_reverse
+
+# This test uses the gcc no-column-info command which was added in gcc 7.1.
+
+proc run_tests {} {
+ global testfile
+
+ clean_restart ${testfile}
+
+ if { ![runto_main] } {
+ return
+ }
+
+ with_test_prefix "next-test" {
+ gdb_test_no_output "record" "turn on process record"
+
+ # This regression test verifies the reverse-step and reverse-next
+ # commands work properly when executing backwards thru a source line
+ # containing two function calls on the same source line, i.e. func1 ();
+ # func2 ();. This test is compiled so the dwarf info not contain the
+ # line table information.
+
+ # Test 1, reverse-next command
+ # Set breakpoint at the line after the function calls.
+ set bp_start_reverse_test [gdb_get_line_number "START REVERSE TEST"]
+
+ gdb_breakpoint $bp_start_reverse_test temporary
+
+ # Continue to break point for reverse-next test.
+ # Command definition: reverse-next [count]
+ # Run backward to the beginning of the previous line executed in the
+ # current (innermost) stack frame. If the line contains function
+ # calls,they will be “un-executed” without stopping. Starting from
+ # the first line of a function, reverse-next will take you back to
+ # the caller of that function, before the function was called, just
+ # as the normal next command would take you from the last line of a
+ # function back to its return to its caller 2 .
+
+ gdb_continue_to_breakpoint \
+ "stopped at command reverse-next test start location" \
+ ".*$bp_start_reverse_test\r\n.*"
+
+ # The reverse-next should step all the way back to the beginning of the
+ # line, i.e. at the beginning of the func1 call.
+ gdb_test "reverse-next" ".*func1 \\(\\); func2 \\(\\);.*" \
+ " reverse-next to line with two functions"
+
+ # We should be stopped at the first instruction of the line. A
+ # reverse-step should step back and stop at the beginning of the
+ # previous line b = 2, i.e. not in func1 ().
+ gdb_test "reverse-stepi" ".*b = 2;.*" \
+ "reverse-stepi to previous line b = 2"
+ }
+
+ # Setup for test 2
+ clean_restart ${testfile}
+
+ if { ![runto_main] } {
+ return
+ }
+
+ with_test_prefix "step-test" {
+ gdb_test_no_output "record" "turn on process record"
+
+ # Test 2, reverse-step command
+ # Set breakpoint at the line after the function calls.
+ gdb_breakpoint $bp_start_reverse_test temporary
+
+ # Continue to the start of the reverse-step test.
+ # Command definition: reverse-step [count]
+ # Run the program backward until control reaches the start of a
+ # different source line; then stop it, and return control to gdb.
+ # Like the step command, reverse-step will only stop at the beginning
+ # of a source line. It “un-executes” the previously executed source
+ # line. If the previous source line included calls to debuggable
+ # functions, reverse-step will step (backward) into the called
+ # function, stopping at the beginning of the last statement in the
+ # called function (typically a return statement). Also, as with the
+ # step command, if non-debuggable functions are called, reverse-step
+ # will run thru them backward without stopping.
+
+ gdb_continue_to_breakpoint \
+ "stopped at command reverse-step test start location" \
+ ".*$bp_start_reverse_test\r\n.*"
+
+ # The first reverse step should take us call of func2 ().
+ gdb_test "reverse-step" ".*END FUNC2.*" \
+ "reverse-step into func2 "
+
+ # The second reverse step should take us into func1 ().
+ gdb_test "reverse-step" ".*END FUNC1.*" \
+ "reverse-step into func1 "
+
+ # The third reverse step should take us call of func1 ().
+ gdb_test "reverse-step" ".*func1 \\(\\); func2 \\(\\);.*" \
+ "reverse-step to line func1(); func2(), at call for func1 "
+
+ # We should be stopped at the first instruction of the line. A reverse
+ # stepi should take us to b = 2 ().
+ gdb_test "reverse-stepi" ".*b = 2;.*" \
+ "reverse-stepi to line b = 2 "
+ }
+}
+
+standard_testfile .c
+
+# test with and without gcc column info enabled
+foreach_with_prefix column_info_flag {column-info no-column-info} {
+ set options [list debug $column_info_flag]
+
+ if {[prepare_for_testing "failed to prepare" ${testfile} ${srcfile} \
+ $options]} {
+ return -1
+ }
+
+ run_tests
+}
diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.c b/gdb/testsuite/gdb.reverse/map-to-same-line.c
new file mode 100644
index 00000000000..5ae0a89b329
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/map-to-same-line.c
@@ -0,0 +1,58 @@
+/* Copyright 2023 Free Software Foundation, Inc.
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 3 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program. If not, see <http://www.gnu.org/licenses/ >. */
+
+/* The purpose of this test is to create a DWARF line table that contains two
+ or more entries for the same line. When stepping (forwards or backwards),
+ GDB should step over the entire line and not just a particular entry in the
+ line table. */
+
+int
+main (void)
+{ /* TAG: main prologue */
+ asm ("main_label: .globl main_label");
+ int i = 1, j = 2, k;
+ float f1 = 2.0, f2 = 4.1, f3;
+ const char *str_1 = "foo", *str_2 = "bar", *str_3;
+
+ asm ("line1: .globl line1");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 1 */
+
+ asm ("line2: .globl line2");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 2 */
+
+ asm ("line3: .globl line3");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 3 */
+
+ asm ("line4: .globl line4");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 4 */
+
+ asm ("line5: .globl line5");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 5 */
+
+ asm ("line6: .globl line6");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 6 */
+
+ asm ("line7: .globl line7");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 7 */
+
+ asm ("line8: .globl line8");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 8 */
+
+ asm ("main_return: .globl main_return");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: main return */
+
+ asm ("end_of_sequence: .globl end_of_sequence");
+ return 0; /* TAG: main return */
+}
diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.exp b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
new file mode 100644
index 00000000000..63f8c9c76b3
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
@@ -0,0 +1,153 @@
+# Copyright 2023 Free Software Foundation, Inc.
+
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program. If not, see <http://www.gnu.org/licenses/ >.
+
+# When stepping (forwards or backwards), GDB should step over the entire line
+# and not just a particular entry in the line table. This test was added to
+# verify the find_line_range_start function properly sets the step range for a
+# line that consists of multiple statements, i.e. multiple entries in the line
+# table. This test creates a DWARF line table that contains two entries for
+# the same line to do the needed testing.
+
+# This test can only be run on targets which support DWARF-2 and use gas.
+load_lib dwarf.exp
+require dwarf2_support
+
+# The DWARF assembler requires the gcc compiler.
+require is_c_compiler_gcc
+
+# This test suitable only for process that can do reverse execution
+require supports_reverse
+
+standard_testfile .c .S
+
+if { [prepare_for_testing "failed to prepare" ${testfile} ${srcfile}] } {
+ return -1
+}
+
+set asm_file [standard_output_file $srcfile2]
+Dwarf::assemble $asm_file {
+ global srcdir subdir srcfile
+ declare_labels integer_label L
+
+ # Find start address and length of program
+ lassign [function_range main [list ${srcdir}/${subdir}/$srcfile]] \
+ main_start main_len
+ set main_end "$main_start + $main_len"
+
+ cu {} {
+ compile_unit {
+ {language @DW_LANG_C}
+ {name map-to-same-line.c}
+ {stmt_list $L DW_FORM_sec_offset}
+ {low_pc 0 addr}
+ } {
+ subprogram {
+ {external 1 flag}
+ {name main}
+ {low_pc $main_start addr}
+ {high_pc $main_len DW_FORM_data4}
+ }
+ }
+ }
+
+ lines {version 2 default_is_stmt 1} L {
+ include_dir "${srcdir}/${subdir}"
+ file_name "$srcfile" 1
+
+ # Generate the line table program with distinct source lines being
+ # mapped to the same line entry. Line 1, 5 and 8 contain 1 statement
+ # each. Line 2 contains 2 statements. Line 3 contains 3 statements.
+ program {
+ DW_LNE_set_address $main_start
+ line [gdb_get_line_number "TAG: main prologue"]
+ DW_LNS_copy
+ DW_LNE_set_address line1
+ line [gdb_get_line_number "TAG: line 1" ]
+ DW_LNS_copy
+ DW_LNE_set_address line2
+ line [gdb_get_line_number "TAG: line 2" ]
+ DW_LNS_copy
+ DW_LNE_set_address line3
+ line [gdb_get_line_number "TAG: line 2" ]
+ DW_LNS_copy
+ DW_LNE_set_address line4
+ line [gdb_get_line_number "TAG: line 3" ]
+ DW_LNS_copy
+ DW_LNE_set_address line5
+ line [gdb_get_line_number "TAG: line 3" ]
+ DW_LNS_copy
+ DW_LNE_set_address line6
+ line [gdb_get_line_number "TAG: line 3" ]
+ DW_LNS_copy
+ DW_LNE_set_address line7
+ line [gdb_get_line_number "TAG: line 5" ]
+ DW_LNS_copy
+ DW_LNE_set_address line8
+ line [gdb_get_line_number "TAG: line 8" ]
+ DW_LNS_copy
+ DW_LNE_set_address main_return
+ line [gdb_get_line_number "TAG: main return"]
+ DW_LNS_copy
+ DW_LNE_set_address end_of_sequence
+ DW_LNE_end_sequence
+ }
+ }
+}
+
+if { [prepare_for_testing "failed to prepare" ${testfile} \
+ [list $srcfile $asm_file] {nodebug} ] } {
+ return -1
+}
+
+if { ![runto_main] } {
+ return
+}
+
+# Print the line table
+gdb_test_multiple "maint info line-table ${testfile}" "" {
+ -re "\r\n$decimal\[ \t\]+$decimal\[ \t\]+($hex)\[ \t\]+Y\[^\r\n\]*" {
+ lappend is_stmt $expect_out(1,string)
+ exp_continue
+ }
+ -re -wrap "" {
+ }
+}
+
+# Do the reverse-step and reverse-next tests
+foreach_with_prefix cmd {step next} {
+ gdb_test_no_output "record" "turn on process record, test $cmd"
+
+ set bp_main_return [gdb_get_line_number "TAG: main return" $srcfile]
+ gdb_breakpoint $srcfile:$bp_main_return
+ gdb_continue_to_breakpoint "run to end of main, reverse-$cmd test" ".*$srcfile:$bp_main_return.*"
+ gdb_test "display \$pc" ".*pc =.*" "display pc, reverse-$cmd test"
+
+ # At this point, GDB has already recorded the execution up until the return
+ # statement. Reverse and test if GDB transitions between lines in the
+ # expected order. It should reverse-step or reverse-next across lines 8,
+ # 5, 3, 2 and 1.
+ foreach line {8 5 3 2 1} {
+ gdb_test "reverse-$cmd" ".*TAG: line $line.*" "reverse $cmd to line $line"
+ }
+
+ if {$cmd =="step"} {
+ ## Clean restart, test reverse-next command
+ clean_restart ${testfile}
+
+ if { ![runto_main] } {
+ return
+ }
+ }
+}
--
2.37.2
^ permalink raw reply [flat|nested] 40+ messages in thread
* Re: [PATCH 1/2 ver 2] Fix reverse stepping multiple contiguous PC ranges over the line table.
2023-06-23 20:04 ` [PATCH 1/2 ver 2] " Carl Love
@ 2023-07-06 15:07 ` Carl Love
0 siblings, 0 replies; 40+ messages in thread
From: Carl Love @ 2023-07-06 15:07 UTC (permalink / raw)
To: Simon Marchi, Bruno Larsen, gdb-patches, UlrichWeigand, pedro
Cc: luis.machado, cel
GDB maintainers:
I believe Simon is on parental leave now. He mentioned in a private
email before he sent the last set of comments that he was expecting to
go on parental leave shortly.
Just wondering if anyone else would be willing to take a look at my
feedback on his questions and changes per his comments to see if we can
move these two patches forward. Thanks
Carl
-------------------------------------------
On Fri, 2023-06-23 at 13:04 -0700, Carl Love wrote:
> Simon, GDB maintainers:
>
> Version 2, updated the compiler check and handling for gcc version 6
> and earlier. Retested on Power 10.
>
> Per the comments on version 4 for the gdb.reverse/func-map-to-same-
> line.exp, I have added support to proc gdb_compile to enable or
> disable
> generating line information as part of the debug information. The
> two
> new options are column-info and no-column-info.
>
> This patch implements the new options for gdb_compile.
>
> These options have been tested with patch 2 of 2 on PowerPC with the
> GCC and clang compilers.
>
> Please let me know if the patch is acceptable for mainline. Thanks.
>
> Carl
>
>
>
>
> -----------------------------
> Add gdb_compile options column-info and no-column-info
>
> This patch adds two new options to gdb_compile to specify if the
> compile
> should or should not generate the line table information. The
> options are supported on clang and gcc version 7 and newer.
>
> Patch has been tested on PowerPC with both gcc and clang.
> ---
> gdb/testsuite/lib/gdb.exp | 34 ++++++++++++++++++++++++++++++++++
> 1 file changed, 34 insertions(+)
>
> diff --git a/gdb/testsuite/lib/gdb.exp b/gdb/testsuite/lib/gdb.exp
> index aed7e2d043c..5857c59a47a 100644
> --- a/gdb/testsuite/lib/gdb.exp
> +++ b/gdb/testsuite/lib/gdb.exp
> @@ -4794,6 +4794,8 @@ proc quote_for_host { args } {
> # debug information
> # - text_segment=addr: Tell the linker to place the text segment
> at ADDR.
> # - build-id: Ensure the final binary includes a build-id.
> +# - no-column-info: Disable generation of column table
> information.
> +# - column-info: Enable generation of column table information.
> #
> # And here are some of the not too obscure options understood by
> DejaGnu that
> # influence the compilation:
> @@ -5003,6 +5005,38 @@ proc gdb_compile {source dest type options} {
> } else {
> error "Don't know how to handle text_segment
> option."
> }
> + } elseif { $opt == "column-info" } {
> + # If GCC or clang does not support column-info, compilation
> + # will fail and the usupported column-info option will be
> + # reported as such.
> + if {[test_compiler_info {gcc-*}]} {
> + lappend new_options "additional_flags=-gcolumn-info"
> +
> + } elseif {[test_compiler_info {clang-*}]} {
> + lappend new_options "additional_flags=-gcolumn-info"
> +
> + } else {
> + error "Don't know how to handle gcolumn-info option."
> + }
> +
> + } elseif { $opt == "no-column-info" } {
> + if {[test_compiler_info {gcc-*}]} {
> + if {[test_compiler_info {gcc-[1-6]-*}]} {
> + # In this case, don't add the compile line option
> and
> + # the result will be the same as using no-column-
> info
> + # on a version that supports the option.
> + warning "gdb_compile option no-column-info not
> supported, ignoring."
> + } else {
> + lappend new_options "additional_flags=-gno-column-
> info"
> + }
> +
> + } elseif {[test_compiler_info {clang-*}]} {
> + lappend new_options "additional_flags=-gno-column-info"
> +
> + } else {
> + error "Don't know how to handle gno-column-info
> option."
> + }
> +
> } else {
> lappend new_options $opt
> }
^ permalink raw reply [flat|nested] 40+ messages in thread
* RE: [PATCH, v4] Fix reverse stepping multiple contiguous PC ranges over the line table
2022-05-06 15:04 ` [PATCH,v4] " Bruno Larsen
@ 2022-05-06 16:46 ` Carl Love
0 siblings, 0 replies; 40+ messages in thread
From: Carl Love @ 2022-05-06 16:46 UTC (permalink / raw)
To: Bruno Larsen, Luis Machado, gdb-patches; +Cc: rogealve, will_schmidt
Bruno:
On Fri, 2022-05-06 at 12:04 -0300, Bruno Larsen wrote:
<snip>
> > >.
> > +
> > +# The purpose of this test is to create a DWARF line table that
> > contains two
> > +# or more entries for the same line. When stepping (forwards or
> > backwards),
> > +# GDB should step over the entire line and not just a particular
> > entry in the
> > +# line table.
> > +
>
> Could you add a comment here after the copyright blurb explaining why
> this testcase exists? Something similar to the find_line_range_start
> comment should be enough, jsut so the next unrelated person who looks
> at this code in 2 years has some context as to what is going on.
OK, updated the comment to say the purpose is to test the
find_line_range_start functionality. I did reorganize the comment a
bit to make things flow better. Specifically, started with the
expected behaviour of GDB followed by testing find_line_range_start to
ensure GDB works as expected.
>
> > +load_lib dwarf.exp
> > +
> > +# This test can only be run on targets which support DWARF-2 and
> > use gas.
> > +if {![dwarf2_support]} {
> > + unsupported "dwarf2 support required for this test"
> > + return 0
> > +}
> > +
> > +if [get_compiler_info] {
> > + return -1
> > +}
> > +
> > +# The DWARF assembler requires the gcc compiler.
> > +if {!$gcc_compiled} {
> > + unsupported "gcc is required for this test"
> > + return 0
> > +}
>
> There should probably be a test here for supports_reverse. Or
> supports_process_record. I'm not sure what the difference is between
> these 2 checks.
Yes, I added if ![supports_process_record] test here to make sure the
system supports the record and replay.
<snip>
>
> > +
> > +if { [prepare_for_testing "failed to prepare" ${testfile} \
> > + [list $srcfile $asm_file] {nodebug} ] } {
> > + return -1
> > +}
> > +
> > +if ![runto_main] {
> > + return -1
> > +}
> > +
> > +if [supports_process_record] {
> > + # Activate process record/replay
> > + gdb_test_no_output "record" "turn on process record"
> > +}
> since we would have tested above if the target supports
> recording/reversing, we should be good to just use record here,
> without the if clause. I suggest doing it this way because there is
> no point in this whole test if the target doesn't support reverse
> execution.
Yes, removed the supports_process_record test.
> >
Will post v5 of the patch for review.
Thanks for the input.
Carl Love
^ permalink raw reply [flat|nested] 40+ messages in thread
* Re: [PATCH,v4] Fix reverse stepping multiple contiguous PC ranges over the line table
2022-05-06 8:55 [PATCH, v4] " Luis Machado
@ 2022-05-06 15:04 ` Bruno Larsen
2022-05-06 16:46 ` [PATCH, v4] " Carl Love
0 siblings, 1 reply; 40+ messages in thread
From: Bruno Larsen @ 2022-05-06 15:04 UTC (permalink / raw)
To: Luis Machado, gdb-patches; +Cc: cel, rogealve, will_schmidt
Hi Luis and Carl!
Sorry about not answering about this earlier, I had a few rushed weeks, but I have some nits about the .exp file
On 5/6/22 05:55, Luis Machado wrote:
> v4:
> - Updated testcase to make it a bit longer so it can exercise reverse-stepping
> multiple times.
> - Cleaned up debugging prints.
>
<snip>
> diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.exp b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> new file mode 100644
> index 00000000000..efaa60a957f
> --- /dev/null
> +++ b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
> @@ -0,0 +1,136 @@
> +# Copyright 2022 Free Software Foundation, Inc.
> +
> +# This program is free software; you can redistribute it and/or modify
> +# it under the terms of the GNU General Public License as published by
> +# the Free Software Foundation; either version 3 of the License, or
> +# (at your option) any later version.
> +#
> +# This program is distributed in the hope that it will be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> +# GNU General Public License for more details.
> +#
> +# You should have received a copy of the GNU General Public License
> +# along with this program. If not, see <http://www.gnu.org/licenses/>.
> +
> +# The purpose of this test is to create a DWARF line table that contains two
> +# or more entries for the same line. When stepping (forwards or backwards),
> +# GDB should step over the entire line and not just a particular entry in the
> +# line table.
> +
Could you add a comment here after the copyright blurb explaining why this testcase exists? Something similar to the find_line_range_start comment should be enough, jsut so the next unrelated person who looks at this code in 2 years has some context as to what is going on.
> +load_lib dwarf.exp
> +
> +# This test can only be run on targets which support DWARF-2 and use gas.
> +if {![dwarf2_support]} {
> + unsupported "dwarf2 support required for this test"
> + return 0
> +}
> +
> +if [get_compiler_info] {
> + return -1
> +}
> +
> +# The DWARF assembler requires the gcc compiler.
> +if {!$gcc_compiled} {
> + unsupported "gcc is required for this test"
> + return 0
> +}
There should probably be a test here for supports_reverse. Or supports_process_record. I'm not sure what the difference is between these 2 checks.
> +
> +standard_testfile .c .S
> +
> +if { [prepare_for_testing "failed to prepare" ${testfile} ${srcfile}] } {
> + return -1
> +}
> +
> +set asm_file [standard_output_file $srcfile2]
> +Dwarf::assemble $asm_file {
> + global srcdir subdir srcfile
> + declare_labels integer_label L
> +
> + # Find start address and length of program
> + lassign [function_range main [list ${srcdir}/${subdir}/$srcfile]] \
> + main_start main_len
> + set main_end "$main_start + $main_len"
> +
> + cu {} {
> + compile_unit {
> + {language @DW_LANG_C}
> + {name map-to-same-line.c}
> + {stmt_list $L DW_FORM_sec_offset}
> + {low_pc 0 addr}
> + } {
> + subprogram {
> + {external 1 flag}
> + {name main}
> + {low_pc $main_start addr}
> + {high_pc $main_len DW_FORM_data4}
> + }
> + }
> + }
> +
> + lines {version 2 default_is_stmt 1} L {
> + include_dir "${srcdir}/${subdir}"
> + file_name "$srcfile" 1
> +
> + # Generate the line table program with distinct source lines being
> + # mapped to the same line entry. Line 1, 5 and 8 contain 1 statement
> + # each. Line 2 contains 2 statements. Line 3 contains 3 statements.
> + program {
> + DW_LNE_set_address $main_start
> + line [gdb_get_line_number "TAG: main prologue"]
> + DW_LNS_copy
> + DW_LNE_set_address line1
> + line [gdb_get_line_number "TAG: line 1" ]
> + DW_LNS_copy
> + DW_LNE_set_address line2
> + line [gdb_get_line_number "TAG: line 2" ]
> + DW_LNS_copy
> + DW_LNE_set_address line3
> + line [gdb_get_line_number "TAG: line 2" ]
> + DW_LNS_copy
> + DW_LNE_set_address line4
> + line [gdb_get_line_number "TAG: line 3" ]
> + DW_LNS_copy
> + DW_LNE_set_address line5
> + line [gdb_get_line_number "TAG: line 3" ]
> + DW_LNS_copy
> + DW_LNE_set_address line6
> + line [gdb_get_line_number "TAG: line 3" ]
> + DW_LNS_copy
> + DW_LNE_set_address line7
> + line [gdb_get_line_number "TAG: line 5" ]
> + DW_LNS_copy
> + DW_LNE_set_address line8
> + line [gdb_get_line_number "TAG: line 8" ]
> + DW_LNS_copy
> + DW_LNE_set_address main_return
> + line [gdb_get_line_number "TAG: main return"]
> + DW_LNS_copy
> + DW_LNE_end_sequence
> + }
> + }
> +}
> +
> +if { [prepare_for_testing "failed to prepare" ${testfile} \
> + [list $srcfile $asm_file] {nodebug} ] } {
> + return -1
> +}
> +
> +if ![runto_main] {
> + return -1
> +}
> +
> +if [supports_process_record] {
> + # Activate process record/replay
> + gdb_test_no_output "record" "turn on process record"
> +}
since we would have tested above if the target supports recording/reversing, we should be good to just use record here, without the if clause. I suggest doing it this way because there is no point in this whole test if the target doesn't support reverse execution.
> +
> +gdb_test "tbreak main_return" "Temporary breakpoint .*" "breakpoint at return"
> +gdb_test "continue" "Temporary breakpoint .*" "run to end of main"
> +
> +# At this point, GDB has already recorded the execution up until the return
> +# statement. Reverse-step and test if GDB transitions between lines in the
> +# expected order. It should reverse-step across lines 8, 5, 3, 2 and 1.
> +foreach line {8 5 3 2 1} {
> + gdb_test "reverse-step" ".*TAG: line $line.*" "reverse step to line $line"
> +}
Cheers!
Bruno Larsen
^ permalink raw reply [flat|nested] 40+ messages in thread
* [PATCH, v4] Fix reverse stepping multiple contiguous PC ranges over the line table
@ 2022-05-06 8:55 Luis Machado
2022-05-06 15:04 ` [PATCH,v4] " Bruno Larsen
0 siblings, 1 reply; 40+ messages in thread
From: Luis Machado @ 2022-05-06 8:55 UTC (permalink / raw)
To: gdb-patches; +Cc: cel, rogealve, will_schmidt, blarsen
v4:
- Updated testcase to make it a bit longer so it can exercise reverse-stepping
multiple times.
- Cleaned up debugging prints.
v3:
- Updated testcase. The format for writing the DWARF program body in the
testcase expect file changed.
See commit gdb/testsuite/dwarf: simplify line number program syntax
(commit d4c4a2298cad06ca71cfef725f5248f68205f0be)
v2:
- Check if both the line and symtab match for a particular line table entry.
--
When running GDB's testsuite on aarch64-linux/Ubuntu 20.04 (also spotted on
the ppc backend), I noticed some failures in gdb.reverse/solib-precsave.exp
and gdb.reverse/solib-reverse.exp.
The failure happens around the following code:
38 b[1] = shr2(17); /* middle part two */
40 b[0] = 6; b[1] = 9; /* generic statement, end part two */
42 shr1 ("message 1\n"); /* shr1 one */
Normal execution:
- step from line 38 will land on line 40.
- step from line 40 will land on line 42.
Reverse execution:
- step from line 42 will land on line 40.
- step from line 40 will land on line 40.
- step from line 40 will land on line 38.
The problem here is that line 40 contains two contiguous but distinct
PC ranges in the line table, like so:
Line 40 - [0x7ec ~ 0x7f4]
Line 40 - [0x7f4 ~ 0x7fc]
The two distinct ranges are generated because GCC started outputting source
column information, which GDB doesn't take into account at the moment.
When stepping forward from line 40, we skip both of these ranges and land on
line 42. When stepping backward from line 42, we stop at the start PC of the
second (or first, going backwards) range of line 40.
This happens because we have this check in infrun.c:process_event_stop_test:
/* When stepping backward, stop at beginning of line range
(unless it's the function entry point, in which case
keep going back to the call point). */
CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
if (stop_pc == ecs->event_thread->control.step_range_start
&& stop_pc != ecs->stop_func_start
&& execution_direction == EXEC_REVERSE)
end_stepping_range (ecs);
else
keep_going (ecs);
Since we've reached ecs->event_thread->control.step_range_start, we stop
stepping backwards.
The right thing to do is to look for adjacent PC ranges for the same line,
until we notice a line change. Then we take that as the start PC of the
range.
Another solution I thought about is to merge the contiguous ranges when
we are reading the line tables. Though I'm not sure if we really want to process
that data as opposed to keeping it as the compiler created, and then working
around that.
In any case, the following patch addresses this problem.
Validated on aarch64-linux and x86_64/Ubuntu 20.04/18.04. Carl Love has
verified that it does fix a similar issue on ppc.
Ubuntu 18.04 doesn't actually run into these failures because the compiler
doesn't generate distinct PC ranges for the same line.
I see similar failures on x86_64 in the gdb.reverse tests
(gdb.reverse/step-reverse.exp and gdb.reverse/step-reverse.exp). Those are
also fixed by this patch.
The included testcase (based on a test Carl wrote) exercises this problem for
Arm, ppc and x86. It shows full passes with the patch applied.
Co-authored-by: Carl Love <cel@us.ibm.com>
---
gdb/infrun.c | 22 ++-
gdb/symtab.c | 49 +++++++
gdb/symtab.h | 16 +++
gdb/testsuite/gdb.reverse/map-to-same-line.c | 55 +++++++
.../gdb.reverse/map-to-same-line.exp | 136 ++++++++++++++++++
5 files changed, 277 insertions(+), 1 deletion(-)
create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.c
create mode 100644 gdb/testsuite/gdb.reverse/map-to-same-line.exp
diff --git a/gdb/infrun.c b/gdb/infrun.c
index 02c98b50c8c..e9e14e58745 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -6917,11 +6917,31 @@ process_event_stop_test (struct execution_control_state *ecs)
have software watchpoints). */
ecs->event_thread->control.may_range_step = 1;
+ /* When we are stepping inside a particular line range, in reverse,
+ and we are sitting at the first address of that range, we need to
+ check if this address also shows up in another line range as the
+ end address.
+
+ If so, we need to check what line such a step range points to.
+ If it points to the same line as the current step range, that
+ means we need to keep going in order to reach the first address
+ of the line range. We repeat this until we eventually get to the
+ first address of a particular line we're stepping through. */
+ CORE_ADDR range_start = ecs->event_thread->control.step_range_start;
+ if (execution_direction == EXEC_REVERSE)
+ {
+ gdb::optional<CORE_ADDR> real_range_start
+ = find_line_range_start (ecs->event_thread->stop_pc ());
+
+ if (real_range_start.has_value ())
+ range_start = *real_range_start;
+ }
+
/* When stepping backward, stop at beginning of line range
(unless it's the function entry point, in which case
keep going back to the call point). */
CORE_ADDR stop_pc = ecs->event_thread->stop_pc ();
- if (stop_pc == ecs->event_thread->control.step_range_start
+ if (stop_pc == range_start
&& stop_pc != ecs->stop_func_start
&& execution_direction == EXEC_REVERSE)
end_stepping_range (ecs);
diff --git a/gdb/symtab.c b/gdb/symtab.c
index 4b33d6c91af..de4cb5dd0eb 100644
--- a/gdb/symtab.c
+++ b/gdb/symtab.c
@@ -3433,6 +3433,55 @@ find_pc_line (CORE_ADDR pc, int notcurrent)
return sal;
}
+/* Compare two symtab_and_line entries. Return true if both have
+ the same line number and the same symtab pointer. That means we
+ are dealing with two entries from the same line and from the same
+ source file.
+
+ Return false otherwise. */
+
+static bool
+sal_line_symtab_matches_p (const symtab_and_line &sal1,
+ const symtab_and_line &sal2)
+{
+ return (sal1.line == sal2.line && sal1.symtab == sal2.symtab);
+}
+
+/* See symtah.h. */
+
+gdb::optional<CORE_ADDR>
+find_line_range_start (CORE_ADDR pc)
+{
+ struct symtab_and_line current_sal = find_pc_line (pc, 0);
+
+ if (current_sal.line == 0)
+ return {};
+
+ struct symtab_and_line prev_sal = find_pc_line (current_sal.pc - 1, 0);
+
+ /* If the previous entry is for a different line, that means we are already
+ at the entry with the start PC for this line. */
+ if (!sal_line_symtab_matches_p (prev_sal, current_sal))
+ return current_sal.pc;
+
+ /* Otherwise, keep looking for entries for the same line but with
+ smaller PC's. */
+ bool done = false;
+ CORE_ADDR prev_pc;
+ while (!done)
+ {
+ prev_pc = prev_sal.pc;
+
+ prev_sal = find_pc_line (prev_pc - 1, 0);
+
+ /* Did we notice a line change? If so, we are done with the search. */
+ if (!sal_line_symtab_matches_p (prev_sal, current_sal))
+ done = true;
+ }
+
+ return prev_pc;
+}
+
/* See symtab.h. */
struct symtab *
diff --git a/gdb/symtab.h b/gdb/symtab.h
index b1cf84f756f..226fe8803db 100644
--- a/gdb/symtab.h
+++ b/gdb/symtab.h
@@ -2285,6 +2285,22 @@ extern struct symtab_and_line find_pc_line (CORE_ADDR, int);
extern struct symtab_and_line find_pc_sect_line (CORE_ADDR,
struct obj_section *, int);
+/* Given PC, and assuming it is part of a range of addresses that is part of a
+ line, go back through the linetable and find the starting PC of that
+ line.
+
+ For example, suppose we have 3 PC ranges for line X:
+
+ Line X - [0x0 - 0x8]
+ Line X - [0x8 - 0x10]
+ Line X - [0x10 - 0x18]
+
+ If we call the function with PC == 0x14, we want to return 0x0, as that is
+ the starting PC of line X, and the ranges are contiguous.
+*/
+
+extern gdb::optional<CORE_ADDR> find_line_range_start (CORE_ADDR pc);
+
/* Wrapper around find_pc_line to just return the symtab. */
extern struct symtab *find_pc_line_symtab (CORE_ADDR);
diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.c b/gdb/testsuite/gdb.reverse/map-to-same-line.c
new file mode 100644
index 00000000000..fa751ffe842
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/map-to-same-line.c
@@ -0,0 +1,55 @@
+/* Copyright 2022 Free Software Foundation, Inc.
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 3 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program. If not, see <http://www.gnu.org/licenses/>. */
+
+/* The purpose of this test is to create a DWARF line table that contains two
+ or more entries for the same line. When stepping (forwards or backwards),
+ GDB should step over the entire line and not just a particular entry in the
+ line table. */
+
+int
+main ()
+{ /* TAG: main prologue */
+ asm ("main_label: .globl main_label");
+ int i = 1, j = 2, k;
+ float f1 = 2.0, f2 = 4.1, f3;
+ const char *str_1 = "foo", *str_2 = "bar", *str_3;
+
+ asm ("line1: .globl line1");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 1 */
+
+ asm ("line2: .globl line2");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 2 */
+
+ asm ("line3: .globl line3");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 3 */
+
+ asm ("line4: .globl line4");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 4 */
+
+ asm ("line5: .globl line5");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 5 */
+
+ asm ("line6: .globl line6");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 6 */
+
+ asm ("line7: .globl line7");
+ k = i; f3 = f1; str_3 = str_1; /* TAG: line 7 */
+
+ asm ("line8: .globl line8");
+ k = j; f3 = f2; str_3 = str_2; /* TAG: line 8 */
+
+ asm ("main_return: .globl main_return");
+ return 0; /* TAG: main return */
+}
diff --git a/gdb/testsuite/gdb.reverse/map-to-same-line.exp b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
new file mode 100644
index 00000000000..efaa60a957f
--- /dev/null
+++ b/gdb/testsuite/gdb.reverse/map-to-same-line.exp
@@ -0,0 +1,136 @@
+# Copyright 2022 Free Software Foundation, Inc.
+
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program. If not, see <http://www.gnu.org/licenses/>.
+
+# The purpose of this test is to create a DWARF line table that contains two
+# or more entries for the same line. When stepping (forwards or backwards),
+# GDB should step over the entire line and not just a particular entry in the
+# line table.
+
+load_lib dwarf.exp
+
+# This test can only be run on targets which support DWARF-2 and use gas.
+if {![dwarf2_support]} {
+ unsupported "dwarf2 support required for this test"
+ return 0
+}
+
+if [get_compiler_info] {
+ return -1
+}
+
+# The DWARF assembler requires the gcc compiler.
+if {!$gcc_compiled} {
+ unsupported "gcc is required for this test"
+ return 0
+}
+
+standard_testfile .c .S
+
+if { [prepare_for_testing "failed to prepare" ${testfile} ${srcfile}] } {
+ return -1
+}
+
+set asm_file [standard_output_file $srcfile2]
+Dwarf::assemble $asm_file {
+ global srcdir subdir srcfile
+ declare_labels integer_label L
+
+ # Find start address and length of program
+ lassign [function_range main [list ${srcdir}/${subdir}/$srcfile]] \
+ main_start main_len
+ set main_end "$main_start + $main_len"
+
+ cu {} {
+ compile_unit {
+ {language @DW_LANG_C}
+ {name map-to-same-line.c}
+ {stmt_list $L DW_FORM_sec_offset}
+ {low_pc 0 addr}
+ } {
+ subprogram {
+ {external 1 flag}
+ {name main}
+ {low_pc $main_start addr}
+ {high_pc $main_len DW_FORM_data4}
+ }
+ }
+ }
+
+ lines {version 2 default_is_stmt 1} L {
+ include_dir "${srcdir}/${subdir}"
+ file_name "$srcfile" 1
+
+ # Generate the line table program with distinct source lines being
+ # mapped to the same line entry. Line 1, 5 and 8 contain 1 statement
+ # each. Line 2 contains 2 statements. Line 3 contains 3 statements.
+ program {
+ DW_LNE_set_address $main_start
+ line [gdb_get_line_number "TAG: main prologue"]
+ DW_LNS_copy
+ DW_LNE_set_address line1
+ line [gdb_get_line_number "TAG: line 1" ]
+ DW_LNS_copy
+ DW_LNE_set_address line2
+ line [gdb_get_line_number "TAG: line 2" ]
+ DW_LNS_copy
+ DW_LNE_set_address line3
+ line [gdb_get_line_number "TAG: line 2" ]
+ DW_LNS_copy
+ DW_LNE_set_address line4
+ line [gdb_get_line_number "TAG: line 3" ]
+ DW_LNS_copy
+ DW_LNE_set_address line5
+ line [gdb_get_line_number "TAG: line 3" ]
+ DW_LNS_copy
+ DW_LNE_set_address line6
+ line [gdb_get_line_number "TAG: line 3" ]
+ DW_LNS_copy
+ DW_LNE_set_address line7
+ line [gdb_get_line_number "TAG: line 5" ]
+ DW_LNS_copy
+ DW_LNE_set_address line8
+ line [gdb_get_line_number "TAG: line 8" ]
+ DW_LNS_copy
+ DW_LNE_set_address main_return
+ line [gdb_get_line_number "TAG: main return"]
+ DW_LNS_copy
+ DW_LNE_end_sequence
+ }
+ }
+}
+
+if { [prepare_for_testing "failed to prepare" ${testfile} \
+ [list $srcfile $asm_file] {nodebug} ] } {
+ return -1
+}
+
+if ![runto_main] {
+ return -1
+}
+
+if [supports_process_record] {
+ # Activate process record/replay
+ gdb_test_no_output "record" "turn on process record"
+}
+
+gdb_test "tbreak main_return" "Temporary breakpoint .*" "breakpoint at return"
+gdb_test "continue" "Temporary breakpoint .*" "run to end of main"
+
+# At this point, GDB has already recorded the execution up until the return
+# statement. Reverse-step and test if GDB transitions between lines in the
+# expected order. It should reverse-step across lines 8, 5, 3, 2 and 1.
+foreach line {8 5 3 2 1} {
+ gdb_test "reverse-step" ".*TAG: line $line.*" "reverse step to line $line"
+}
--
2.25.1
^ permalink raw reply [flat|nested] 40+ messages in thread
end of thread, other threads:[~2023-07-06 15:07 UTC | newest]
Thread overview: 40+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-04-27 20:59 [PATCH] Fix reverse stepping multiple contiguous PC ranges over the line table Carl Love
2023-05-02 14:15 ` Bruno Larsen
2023-05-02 15:40 ` Carl Love
2023-05-02 15:42 ` Bruno Larsen
2023-05-11 15:11 ` Simon Marchi
2023-05-03 9:53 ` Bruno Larsen
2023-05-04 2:55 ` Carl Love
2023-05-04 9:24 ` Bruno Larsen
2023-05-04 14:52 ` Carl Love
2023-05-04 2:55 ` [PATCH v2] " Carl Love
2023-05-04 15:59 ` [PATCH v3] " Carl Love
2023-05-05 14:59 ` Luis Machado
2023-05-05 16:10 ` Carl Love
2023-05-10 13:47 ` Bruno Larsen
2023-05-10 17:16 ` Carl Love
2023-05-10 17:32 ` [PATCH v4] " Carl Love
2023-05-11 16:01 ` Simon Marchi
2023-05-11 16:23 ` Bruno Larsen
2023-05-11 17:28 ` Simon Marchi
2023-05-16 22:54 ` [PATCH 1/2] " Carl Love
2023-06-19 17:11 ` Simon Marchi
2023-06-22 16:52 ` Carl Love
2023-06-23 17:44 ` Simon Marchi
2023-06-23 19:41 ` Carl Love
2023-06-23 20:04 ` [PATCH 1/2 ver 2] " Carl Love
2023-07-06 15:07 ` Carl Love
2023-05-16 22:54 ` [PATCH 2/2 v5] " Carl Love
2023-05-25 15:08 ` Carl Love
2023-06-08 16:36 ` Carl Love
2023-06-19 17:58 ` Simon Marchi
2023-06-22 20:38 ` Carl Love
2023-06-22 20:39 ` Carl Love
2023-06-23 17:49 ` Simon Marchi
2023-06-23 20:04 ` Carl Love
2023-06-23 20:04 ` [PATCH 2/2 v6] " Carl Love
2023-05-16 22:54 ` [PATCH v4] " Carl Love
2023-05-11 7:52 ` [PATCH v3] " Bruno Larsen
-- strict thread matches above, loose matches on Subject: below --
2022-05-06 8:55 [PATCH, v4] " Luis Machado
2022-05-06 15:04 ` [PATCH,v4] " Bruno Larsen
2022-05-06 16:46 ` [PATCH, v4] " Carl Love
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).