From: "Maciej W. Rozycki" <macro@embecosm.com>
To: gdb-patches@sourceware.org
Cc: Andrew Burgess <aburgess@redhat.com>, Tom Tromey <tom@tromey.com>,
Richard Bunt <Richard.Bunt@arm.com>
Subject: [PATCH v3 3/5] GDB: Only make data actually retrieved into value history available
Date: Mon, 23 Jan 2023 23:14:06 +0000 (GMT) [thread overview]
Message-ID: <alpine.DEB.2.20.2301211715160.7841@tpp.orcam.me.uk> (raw)
In-Reply-To: <alpine.DEB.2.20.2301210444350.7841@tpp.orcam.me.uk>
While it makes sense to allow accessing out-of-bounds elements in the
debuggee and see whatever there might happen to be there in memory (we
are a debugger and not a programming rules enforcement facility and we
want to make people's life easier in chasing bugs), e.g.:
(gdb) print one_hundred[-1]
$1 = 0
(gdb) print one_hundred[100]
$2 = 0
(gdb)
we shouldn't really pretend that we have any meaningful data around
values recorded in history (what these commands really retrieve are
current debuggee memory contents outside the original data accessed,
really confusing in my opinion). Add code to mark data around values
recorded in history unavailable then:
(gdb) print one_hundred[-1]
$1 = <unavailable>
(gdb) print one_hundred[100]
$2 = <unavailable>
Adjust `value_entirely_available' accordingly; also zero-length types
need special handling in `value_entirely_covered_by_range_vector'.
Add a suitable test case, which also covers integer overflows in data
location calculation.
---
Hi,
While at it I noticed that while values recorded in the history are just
as one would expect immutable, one can poke at the original data object
data in history has originated from by writing to the value in the
history, e.g.:
(gdb) print -elements 3 -- one_hundred
$1 = {0, 1, 2...}
(gdb) print -elements 3 -- $1
$2 = {0, 1, 2...}
(gdb) set $1[0] = -1
(gdb) print -elements 3 -- $1
$3 = {0, 1, 2...}
(gdb) print -elements 3 -- one_hundred
$4 = {-1, 1, 2...}
(gdb)
This is undocumented and I find it surprising to say the least. Shall I
file a bug for this phenomenon?
LONG_MIN is not used, but I have thought we can have it anyway for
consistency and so that people do not try to invent local variants.
Maciej
Changes from v2:
- Correct the copyright year in new files.
New change in v2.
---
gdb/defs.h | 8 ++
gdb/testsuite/gdb.base/value-history-unavailable.c | 29 +++++++
gdb/testsuite/gdb.base/value-history-unavailable.exp | 73 +++++++++++++++++++
gdb/valarith.c | 15 +++
gdb/value.c | 24 ++++--
5 files changed, 144 insertions(+), 5 deletions(-)
gdb-value-history-unavailable.diff
Index: src/gdb/defs.h
===================================================================
--- src.orig/gdb/defs.h
+++ src/gdb/defs.h
@@ -469,6 +469,10 @@ enum val_prettyformat
#define LONG_MAX ((long)(ULONG_MAX >> 1)) /* 0x7FFFFFFF for 32-bits */
#endif
+#if !defined (LONG_MIN) /* 0x80000000 for 32-bits */
+#define LONG_MIN ((long) (~(long) 0 ^ LONG_MAX))
+#endif
+
#if !defined (ULONGEST_MAX)
#define ULONGEST_MAX (~(ULONGEST)0) /* 0xFFFFFFFFFFFFFFFF for 64-bits */
#endif
@@ -477,6 +481,10 @@ enum val_prettyformat
#define LONGEST_MAX ((LONGEST)(ULONGEST_MAX >> 1))
#endif
+#if !defined (LONGEST_MIN) /* 0x8000000000000000 for 64-bits */
+#define LONGEST_MIN ((LONGEST) (~(LONGEST) 0 ^ LONGEST_MAX))
+#endif
+
/* * Convert a LONGEST to an int. This is used in contexts (e.g. number of
arguments to a function, number in a value history, register number, etc.)
where the value must not be larger than can fit in an int. */
Index: src/gdb/testsuite/gdb.base/value-history-unavailable.c
===================================================================
--- /dev/null
+++ src/gdb/testsuite/gdb.base/value-history-unavailable.c
@@ -0,0 +1,29 @@
+/* This testcase is part of GDB, the GNU debugger.
+
+ Copyright (C) 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/>. */
+
+struct
+{
+ unsigned char x[1024];
+ unsigned char a[1024];
+ unsigned char y[1024];
+} a = { {-1} };
+
+int
+main ()
+{
+ return 0;
+}
Index: src/gdb/testsuite/gdb.base/value-history-unavailable.exp
===================================================================
--- /dev/null
+++ src/gdb/testsuite/gdb.base/value-history-unavailable.exp
@@ -0,0 +1,73 @@
+# Copyright (C) 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/>.
+
+# Test GDB's value availability ranges.
+
+standard_testfile
+
+if { [prepare_for_testing "failed to prepare" $testfile $srcfile] } {
+ return -1
+}
+
+if ![runto_main] then {
+ perror "couldn't run to breakpoint"
+ continue
+}
+
+set target_char_mask [get_valueof "/u" "a.x\[0]" "255" "get target char mask"]
+set target_char_bit 0
+for {set i $target_char_mask} {$i > 0} {set i [expr $i >> 1]} {
+ incr target_char_bit
+}
+set target_char_rank -1
+for {set i $target_char_bit} {$i > 0} {set i [expr $i >> 1]} {
+ incr target_char_rank
+}
+
+# Verify accesses to original inferior data.
+gdb_test "print a.a" "\\\$2 = '\\\\000' <repeats 1023 times>"
+gdb_test "print a.a\[-1\]" "\\\$3 = 0 '\\\\000'"
+gdb_test "print a.a\[1024\]" "\\\$4 = 0 '\\\\000'"
+
+# Verify in-range value history accesses.
+gdb_test "print \$2" "\\\$5 = '\\\\000' <repeats 1023 times>"
+gdb_test "print \$2\[0\]" "\\\$6 = 0 '\\\\000'"
+gdb_test "print \$2\[1023\]" "\\\$7 = 0 '\\\\000'"
+
+# Values outside the array recorded will have not been retrieved.
+gdb_test "print \$2\[-1\]" "\\\$8 = <unavailable>"
+gdb_test "print \$2\[1024\]" "\\\$9 = <unavailable>"
+gdb_test "print \$2\[-1LL << 63 - $target_char_rank\]" \
+ "\\\$10 = <unavailable>"
+gdb_test "print \$2\[(1LL << 63 - $target_char_rank) - 1\]" \
+ "\\\$11 = <unavailable>"
+
+# Accesses through pointers in history go straight to the inferior though.
+gdb_test "print \$2\[0\]@1" "\\\$12 = \"\""
+gdb_test "print \$2\[-1\]@1" "\\\$13 = \"\""
+gdb_test "print \$2\[1024\]@1" "\\\$14 = \"\""
+
+# Verify out-of-range value history accesses.
+gdb_test "print \$2\[(-1LL << 63 - $target_char_rank) - 1\]" \
+ "Integer overflow in data location calculation"
+gdb_test "print \$2\[(1LL << 63 - $target_char_rank)\]" \
+ "Integer overflow in data location calculation"
+gdb_test "print \$2\[-1LL << 63\]" \
+ "Integer overflow in data location calculation"
+gdb_test "print \$2\[(1ULL << 63) - 1\]" \
+ "Integer overflow in data location calculation"
+
+# Sanity-check a copy of an unavailable value.
+gdb_test "print \$11" "\\\$15 = <unavailable>"
Index: src/gdb/valarith.c
===================================================================
--- src.orig/gdb/valarith.c
+++ src/gdb/valarith.c
@@ -182,6 +182,21 @@ value_subscript (struct value *array, LO
}
index -= *lowerbound;
+
+ /* Do not try to dereference a pointer to an unavailable value.
+ Instead mock up a new one and give it the original address. */
+ struct type *elt_type = check_typedef (tarray->target_type ());
+ LONGEST elt_size = type_length_units (elt_type);
+ if (!value_lazy (array)
+ && !value_bytes_available (array, elt_size * index, elt_size))
+ {
+ struct value *val = allocate_value (elt_type);
+ mark_value_bytes_unavailable (val, 0, elt_size);
+ VALUE_LVAL (val) = lval_memory;
+ set_value_address (val, value_address (array) + elt_size * index);
+ return val;
+ }
+
array = value_coerce_array (array);
}
Index: src/gdb/value.c
===================================================================
--- src.orig/gdb/value.c
+++ src/gdb/value.c
@@ -421,7 +421,9 @@ value_entirely_available (struct value *
if (value->lazy)
value_fetch_lazy (value);
- if (value->unavailable.empty ())
+ if (value->unavailable.empty ()
+ || value_bytes_available (value, 0,
+ value_enclosing_type (value)->length ()))
return 1;
return 0;
}
@@ -441,11 +443,12 @@ value_entirely_covered_by_range_vector (
if (ranges.size () == 1)
{
+ ULONGEST length = value_enclosing_type (value)->length ();
const struct range &t = ranges[0];
- if (t.offset == 0
- && t.length == (TARGET_CHAR_BIT
- * value_enclosing_type (value)->length ()))
+ if (length == 0)
+ return t.offset == 0 && t.length == 0;
+ if (t.offset <= 0 && t.offset + t.length >= TARGET_CHAR_BIT * length)
return 1;
}
@@ -1277,7 +1280,9 @@ require_not_optimized_out (const struct
static void
require_available (const struct value *value)
{
- if (!value->unavailable.empty ())
+ if (!value->unavailable.empty ()
+ && !value_bytes_available (value, 0,
+ value_enclosing_type (value)->length ()))
throw_error (NOT_AVAILABLE_ERROR, _("value is not available"));
}
@@ -1950,6 +1955,15 @@ record_latest_value (struct value *val)
a value on the value history never changes. */
if (value_lazy (val))
value_fetch_lazy (val);
+
+ /* Don't pretend we have anything available there in the history beyond
+ the boundaries of the value recorded. It's not like inferior memory
+ where there is actual stuff underneath. */
+ ULONGEST length = value_enclosing_type (val)->length ();
+ mark_value_bits_unavailable (val, LONGEST_MIN, 0 ^ LONGEST_MIN);
+ mark_value_bits_unavailable (val, length * TARGET_CHAR_BIT,
+ LONGEST_MAX - length * TARGET_CHAR_BIT);
+
/* We preserve VALUE_LVAL so that the user can find out where it was fetched
from. This is a bit dubious, because then *&$1 does not just return $1
but the current contents of that location. c'est la vie... */
next prev parent reply other threads:[~2023-01-23 23:14 UTC|newest]
Thread overview: 18+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-01-23 23:00 [PATCH v3 0/5] gdb: introduce limited array lengths while printing values Maciej W. Rozycki
2023-01-23 23:13 ` [PATCH v3 1/5] GDB: Ignore `max-value-size' setting with value history accesses Maciej W. Rozycki
2023-01-31 17:58 ` Tom Tromey
2023-02-10 14:17 ` Maciej W. Rozycki
2023-01-23 23:13 ` [PATCH v3 2/5] GDB: Fix the mess with value byte/bit range types Maciej W. Rozycki
2023-01-31 18:09 ` Tom Tromey
2023-02-10 14:18 ` Maciej W. Rozycki
2023-02-10 14:49 ` Tom Tromey
2023-01-23 23:14 ` Maciej W. Rozycki [this message]
2023-01-31 18:47 ` [PATCH v3 3/5] GDB: Only make data actually retrieved into value history available Tom Tromey
2023-02-10 14:18 ` Maciej W. Rozycki
2023-02-10 21:11 ` Tom Tromey
2023-01-23 23:14 ` [PATCH v3 4/5] GDB/testsuite: Add `-nonl' option to `gdb_test' Maciej W. Rozycki
2023-01-31 19:02 ` Tom Tromey
2023-01-23 23:14 ` [PATCH v3 5/5] GDB: Introduce limited array lengths while printing values Maciej W. Rozycki
2023-01-24 12:59 ` Eli Zaretskii
2023-01-31 20:49 ` Tom Tromey
2023-02-10 14:18 ` Maciej W. Rozycki
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=alpine.DEB.2.20.2301211715160.7841@tpp.orcam.me.uk \
--to=macro@embecosm.com \
--cc=Richard.Bunt@arm.com \
--cc=aburgess@redhat.com \
--cc=gdb-patches@sourceware.org \
--cc=tom@tromey.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
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).