* [PATCH v3 15/17] PPC64: Fix gdb.arch/ppc64-atomic-inst.exp with displaced stepping
2015-04-17 10:47 [PATCH v3 00/23] All-stop on top of non-stop Pedro Alves
@ 2015-04-17 10:45 ` Pedro Alves
2015-04-21 11:21 ` Yao Qi
2015-04-17 10:45 ` [PATCH v3 11/17] Fix signal-while-stepping-over-bp-other-thread.exp on targets always in non-stop Pedro Alves
` (17 subsequent siblings)
18 siblings, 1 reply; 62+ messages in thread
From: Pedro Alves @ 2015-04-17 10:45 UTC (permalink / raw)
To: gdb-patches
The ppc64 displaced step code can't handle atomic sequences. Fallback
to stepping over the breakpoint in-line if we detect one.
gdb/ChangeLog:
2015-04-17 Pedro Alves <palves@redhat.com>
* infrun.c (displaced_step_prepare): Return -1 if
gdbarch_displaced_step_copy_insn returns NULL.
(resume): When displaced stepping doesn't work, try software
single-stepping.
* rs6000-tdep.c (LWARX_MASK, LWARX_INSTRUCTION, LDARX_INSTRUCTION)
(STWCX_MASK, STWCX_INSTRUCTION, STDCX_INSTRUCTION): Move higher up
in file.
(ppc_displaced_step_copy_insn): New function.
(ppc_displaced_step_fixup): Update comment.
(rs6000_gdbarch_init): Install ppc_displaced_step_copy_insn as
gdbarch_displaced_step_copy_insn hook.
* gdbarch.sh (displaced_step_copy_insn): Document what happens on
NULL return.
* gdbarch.h: Regenerate.
gdb/testsuite/ChangeLog:
2015-04-17 Pedro Alves <palves@redhat.com>
* gdb.arch/ppc64-atomic-inst.exp (do_test): New procedure, move
tests here.
(top level): Run do_test with and without displaced stepping.
v3:
- Document that gdbarch_displaced_step_copy_insn can return NULL.
---
gdb/gdbarch.h | 6 ++-
gdb/gdbarch.sh | 4 ++
gdb/infrun.c | 11 +++--
gdb/rs6000-tdep.c | 68 +++++++++++++++++++++++-----
gdb/testsuite/gdb.arch/ppc64-atomic-inst.exp | 50 ++++++++++++--------
5 files changed, 106 insertions(+), 33 deletions(-)
diff --git a/gdb/gdbarch.h b/gdb/gdbarch.h
index c94c19c..b02225a 100644
--- a/gdb/gdbarch.h
+++ b/gdb/gdbarch.h
@@ -923,7 +923,11 @@ extern void set_gdbarch_max_insn_length (struct gdbarch *gdbarch, ULONGEST max_i
If your architecture doesn't need to adjust instructions before
single-stepping them, consider using simple_displaced_step_copy_insn
- here. */
+ here.
+
+ If the instruction cannot execute out of line, return NULL. The
+ core falls back to stepping past the instruction in-line instead in
+ that case. */
extern int gdbarch_displaced_step_copy_insn_p (struct gdbarch *gdbarch);
diff --git a/gdb/gdbarch.sh b/gdb/gdbarch.sh
index 0f303a4..749cbda 100755
--- a/gdb/gdbarch.sh
+++ b/gdb/gdbarch.sh
@@ -763,6 +763,10 @@ V:ULONGEST:max_insn_length:::0:0
# If your architecture doesn't need to adjust instructions before
# single-stepping them, consider using simple_displaced_step_copy_insn
# here.
+#
+# If the instruction cannot execute out of line, return NULL. The
+# core falls back to stepping past the instruction in-line instead in
+# that case.
M:struct displaced_step_closure *:displaced_step_copy_insn:CORE_ADDR from, CORE_ADDR to, struct regcache *regs:from, to, regs
# Return true if GDB should use hardware single-stepping to execute
diff --git a/gdb/infrun.c b/gdb/infrun.c
index be186a7..a89eefc 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -1797,9 +1797,14 @@ displaced_step_prepare_throw (ptid_t ptid)
closure = gdbarch_displaced_step_copy_insn (gdbarch,
original, copy, regcache);
-
- /* We don't support the fully-simulated case at present. */
- gdb_assert (closure);
+ if (closure == NULL)
+ {
+ /* The architecture doesn't know how or want to displaced step
+ this instruction or instruction sequence. Fallback to
+ stepping over the breakpoint in-line. */
+ do_cleanups (old_cleanups);
+ return -1;
+ }
/* Save the information we need to fix things up if the step
succeeds. */
diff --git a/gdb/rs6000-tdep.c b/gdb/rs6000-tdep.c
index a125000..6106029 100644
--- a/gdb/rs6000-tdep.c
+++ b/gdb/rs6000-tdep.c
@@ -974,6 +974,61 @@ rs6000_breakpoint_from_pc (struct gdbarch *gdbarch, CORE_ADDR *bp_addr,
#define BXL_INSN 0x4c000000
#define BP_INSN 0x7C000008
+/* Instruction masks used during single-stepping of atomic
+ sequences. */
+#define LWARX_MASK 0xfc0007fe
+#define LWARX_INSTRUCTION 0x7c000028
+#define LDARX_INSTRUCTION 0x7c0000A8
+#define STWCX_MASK 0xfc0007ff
+#define STWCX_INSTRUCTION 0x7c00012d
+#define STDCX_INSTRUCTION 0x7c0001ad
+
+/* We can't displaces step atomic sequenes. Otherwise this is just
+ simple_displaced_step_copy_insn. */
+
+static struct displaced_step_closure *
+ppc_displaced_step_copy_insn (struct gdbarch *gdbarch,
+ CORE_ADDR from, CORE_ADDR to,
+ struct regcache *regs)
+{
+ size_t len = gdbarch_max_insn_length (gdbarch);
+ gdb_byte *buf = xmalloc (len);
+ struct cleanup *old_chain = make_cleanup (xfree, buf);
+ enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
+ int insn;
+
+ read_memory (from, buf, len);
+
+ insn = extract_signed_integer (buf, PPC_INSN_SIZE, byte_order);
+
+ /* Assume all atomic sequences start with a lwarx/ldarx instruction. */
+ if ((insn & LWARX_MASK) == LWARX_INSTRUCTION
+ || (insn & LWARX_MASK) == LDARX_INSTRUCTION)
+ {
+ if (debug_displaced)
+ {
+ fprintf_unfiltered (gdb_stdlog,
+ "displaced: can't displaced step "
+ "atomic sequence at %s\n",
+ paddress (gdbarch, from));
+ }
+ do_cleanups (old_chain);
+ return NULL;
+ }
+
+ write_memory (to, buf, len);
+
+ if (debug_displaced)
+ {
+ fprintf_unfiltered (gdb_stdlog, "displaced: copy %s->%s: ",
+ paddress (gdbarch, from), paddress (gdbarch, to));
+ displaced_step_dump_bytes (gdb_stdlog, buf, len);
+ }
+
+ discard_cleanups (old_chain);
+ return (struct displaced_step_closure *) buf;
+}
+
/* Fix up the state of registers and memory after having single-stepped
a displaced instruction. */
static void
@@ -983,8 +1038,7 @@ ppc_displaced_step_fixup (struct gdbarch *gdbarch,
struct regcache *regs)
{
enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
- /* Since we use simple_displaced_step_copy_insn, our closure is a
- copy of the instruction. */
+ /* Our closure is a copy of the instruction. */
ULONGEST insn = extract_unsigned_integer ((gdb_byte *) closure,
PPC_INSN_SIZE, byte_order);
ULONGEST opcode = 0;
@@ -1077,14 +1131,6 @@ ppc_displaced_step_hw_singlestep (struct gdbarch *gdbarch,
return 1;
}
-/* Instruction masks used during single-stepping of atomic sequences. */
-#define LWARX_MASK 0xfc0007fe
-#define LWARX_INSTRUCTION 0x7c000028
-#define LDARX_INSTRUCTION 0x7c0000A8
-#define STWCX_MASK 0xfc0007ff
-#define STWCX_INSTRUCTION 0x7c00012d
-#define STDCX_INSTRUCTION 0x7c0001ad
-
/* Checks for an atomic sequence of instructions beginning with a LWARX/LDARX
instruction and ending with a STWCX/STDCX instruction. If such a sequence
is found, attempt to step through it. A breakpoint is placed at the end of
@@ -5923,7 +5969,7 @@ rs6000_gdbarch_init (struct gdbarch_info info, struct gdbarch_list *arches)
/* Setup displaced stepping. */
set_gdbarch_displaced_step_copy_insn (gdbarch,
- simple_displaced_step_copy_insn);
+ ppc_displaced_step_copy_insn);
set_gdbarch_displaced_step_hw_singlestep (gdbarch,
ppc_displaced_step_hw_singlestep);
set_gdbarch_displaced_step_fixup (gdbarch, ppc_displaced_step_fixup);
diff --git a/gdb/testsuite/gdb.arch/ppc64-atomic-inst.exp b/gdb/testsuite/gdb.arch/ppc64-atomic-inst.exp
index d251425..08518fb 100644
--- a/gdb/testsuite/gdb.arch/ppc64-atomic-inst.exp
+++ b/gdb/testsuite/gdb.arch/ppc64-atomic-inst.exp
@@ -32,27 +32,41 @@ if { [prepare_for_testing ${testfile}.exp ${testfile} ${srcfile} {debug quiet}]
return -1
}
-if ![runto_main] then {
- untested "could not run to main"
- return -1
-}
+# The test proper. DISPLACED is true if we should try with displaced
+# stepping.
+proc do_test { displaced } {
+ global decimal hex
+
+ if ![runto_main] then {
+ untested "could not run to main"
+ return -1
+ }
+
+ gdb_test_no_output "set displaced-stepping $displaced"
-set bp1 [gdb_get_line_number "lwarx"]
-gdb_breakpoint "$bp1" "Breakpoint $decimal at $hex" \
- "Set the breakpoint at the start of the lwarx/stwcx sequence"
+ set bp1 [gdb_get_line_number "lwarx"]
+ gdb_breakpoint "$bp1" "Breakpoint $decimal at $hex" \
+ "Set the breakpoint at the start of the lwarx/stwcx sequence"
-set bp2 [gdb_get_line_number "ldarx"]
-gdb_breakpoint "$bp2" "Breakpoint $decimal at $hex" \
- "Set the breakpoint at the start of the ldarx/stdcx sequence"
+ set bp2 [gdb_get_line_number "ldarx"]
+ gdb_breakpoint "$bp2" "Breakpoint $decimal at $hex" \
+ "Set the breakpoint at the start of the ldarx/stdcx sequence"
-gdb_test continue "Continuing.*Breakpoint $decimal.*" \
- "Continue until lwarx/stwcx start breakpoint"
+ gdb_test continue "Continuing.*Breakpoint $decimal.*" \
+ "Continue until lwarx/stwcx start breakpoint"
-gdb_test nexti "bne.*1b" \
- "Step through the lwarx/stwcx sequence"
+ gdb_test nexti "bne.*1b" \
+ "Step through the lwarx/stwcx sequence"
-gdb_test continue "Continuing.*Breakpoint $decimal.*" \
- "Continue until ldarx/stdcx start breakpoint"
+ gdb_test continue "Continuing.*Breakpoint $decimal.*" \
+ "Continue until ldarx/stdcx start breakpoint"
-gdb_test nexti "bne.*1b" \
- "Step through the ldarx/stdcx sequence"
+ gdb_test nexti "bne.*1b" \
+ "Step through the ldarx/stdcx sequence"
+}
+
+foreach displaced { "off" "on" } {
+ with_test_prefix "displaced=$displaced" {
+ do_test $displaced
+ }
+}
--
1.9.3
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 15/17] PPC64: Fix gdb.arch/ppc64-atomic-inst.exp with displaced stepping
2015-04-17 10:45 ` [PATCH v3 15/17] PPC64: Fix gdb.arch/ppc64-atomic-inst.exp with displaced stepping Pedro Alves
@ 2015-04-21 11:21 ` Yao Qi
2015-04-22 20:04 ` Pedro Alves
0 siblings, 1 reply; 62+ messages in thread
From: Yao Qi @ 2015-04-21 11:21 UTC (permalink / raw)
To: Pedro Alves; +Cc: gdb-patches
Pedro Alves <palves@redhat.com> writes:
> The ppc64 displaced step code can't handle atomic sequences. Fallback
> to stepping over the breakpoint in-line if we detect one.
Probably arm atom sequences instructions can't be executed out-of-line
either, I'll write a test for this.
>
> gdb/ChangeLog:
> 2015-04-17 Pedro Alves <palves@redhat.com>
>
> * infrun.c (displaced_step_prepare): Return -1 if
> gdbarch_displaced_step_copy_insn returns NULL.
> (resume): When displaced stepping doesn't work, try software
> single-stepping.
This chunk isn't shown in the patch below, has been committed already?
> diff --git a/gdb/infrun.c b/gdb/infrun.c
> index be186a7..a89eefc 100644
> --- a/gdb/infrun.c
> +++ b/gdb/infrun.c
> @@ -1797,9 +1797,14 @@ displaced_step_prepare_throw (ptid_t ptid)
>
> closure = gdbarch_displaced_step_copy_insn (gdbarch,
> original, copy, regcache);
> -
> - /* We don't support the fully-simulated case at present. */
> - gdb_assert (closure);
> + if (closure == NULL)
> + {
> + /* The architecture doesn't know how or want to displaced step
> + this instruction or instruction sequence. Fallback to
> + stepping over the breakpoint in-line. */
> + do_cleanups (old_cleanups);
> + return -1;
> + }
>
> /* Save the information we need to fix things up if the step
> succeeds. */
We need to update the comments to displaced_step_prepare_throw about
return value -1.
--
Yao (齐尧)
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 15/17] PPC64: Fix gdb.arch/ppc64-atomic-inst.exp with displaced stepping
2015-04-21 11:21 ` Yao Qi
@ 2015-04-22 20:04 ` Pedro Alves
0 siblings, 0 replies; 62+ messages in thread
From: Pedro Alves @ 2015-04-22 20:04 UTC (permalink / raw)
To: Yao Qi; +Cc: gdb-patches
On 04/21/2015 12:20 PM, Yao Qi wrote:
> Pedro Alves <palves@redhat.com> writes:
>
>> The ppc64 displaced step code can't handle atomic sequences. Fallback
>> to stepping over the breakpoint in-line if we detect one.
>
> Probably arm atom sequences instructions can't be executed out-of-line
> either, I'll write a test for this.
Thanks.
>
>>
>> gdb/ChangeLog:
>> 2015-04-17 Pedro Alves <palves@redhat.com>
>>
>> * infrun.c (displaced_step_prepare): Return -1 if
>> gdbarch_displaced_step_copy_insn returns NULL.
>> (resume): When displaced stepping doesn't work, try software
>> single-stepping.
>
> This chunk isn't shown in the patch below, has been committed already?
Sorry about that. Double snafu. It was supposed to be "try stepping
over the breakpoint in-line", not software single-step. And then, that
"resume" change ended up moved to the previous patch.
>
>> diff --git a/gdb/infrun.c b/gdb/infrun.c
>> index be186a7..a89eefc 100644
>> --- a/gdb/infrun.c
>> +++ b/gdb/infrun.c
>> @@ -1797,9 +1797,14 @@ displaced_step_prepare_throw (ptid_t ptid)
>>
>> closure = gdbarch_displaced_step_copy_insn (gdbarch,
>> original, copy, regcache);
>> -
>> - /* We don't support the fully-simulated case at present. */
>> - gdb_assert (closure);
>> + if (closure == NULL)
>> + {
>> + /* The architecture doesn't know how or want to displaced step
>> + this instruction or instruction sequence. Fallback to
>> + stepping over the breakpoint in-line. */
>> + do_cleanups (old_cleanups);
>> + return -1;
>> + }
>>
>> /* Save the information we need to fix things up if the step
>> succeeds. */
>
> We need to update the comments to displaced_step_prepare_throw about
> return value -1.
>
I've added this to the patch:
diff --git i/gdb/infrun.c w/gdb/infrun.c
index f627094..aaa543b 100644
--- i/gdb/infrun.c
+++ w/gdb/infrun.c
@@ -1713,7 +1713,9 @@ displaced_step_dump_bytes (struct ui_file *file,
explain how we handle this case instead.
Returns 1 if preparing was successful -- this thread is going to be
- stepped now; or 0 if displaced stepping this thread got queued. */
+ stepped now; 0 if displaced stepping this thread got queued; or -1
+ if this instruction can't be displaced stepped. */
+
static int
displaced_step_prepare_throw (ptid_t ptid)
{
Thanks,
Pedro Alves
^ permalink raw reply [flat|nested] 62+ messages in thread
* [PATCH v3 11/17] Fix signal-while-stepping-over-bp-other-thread.exp on targets always in non-stop
2015-04-17 10:47 [PATCH v3 00/23] All-stop on top of non-stop Pedro Alves
2015-04-17 10:45 ` [PATCH v3 15/17] PPC64: Fix gdb.arch/ppc64-atomic-inst.exp with displaced stepping Pedro Alves
@ 2015-04-17 10:45 ` Pedro Alves
2015-04-17 10:45 ` [PATCH v3 08/17] Factor out code to re-resume stepped thread Pedro Alves
` (16 subsequent siblings)
18 siblings, 0 replies; 62+ messages in thread
From: Pedro Alves @ 2015-04-17 10:45 UTC (permalink / raw)
To: gdb-patches
With "maint set target-non-stop on" we get:
-PASS: gdb.threads/signal-while-stepping-over-bp-other-thread.exp: step
+FAIL: gdb.threads/signal-while-stepping-over-bp-other-thread.exp: step
The issue is simply that switch_back_to_stepped_thread is not used in
non-stop mode, thus infrun doesn't output the expected "switching back
to stepped thread" log.
gdb/testsuite/ChangeLog:
2015-04-17 Pedro Alves <palves@redhat.com>
* signal-while-stepping-over-bp-other-thread.exp: Expect "restart
threads" as alternative to "switching back to stepped thread".
v3:
- Cleanups were moved to a separate patch and pushed already. Now a
one-liner. Also, v2 mentioned the wrong output by silly
mistake....
---
.../gdb.threads/signal-while-stepping-over-bp-other-thread.exp | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/gdb/testsuite/gdb.threads/signal-while-stepping-over-bp-other-thread.exp b/gdb/testsuite/gdb.threads/signal-while-stepping-over-bp-other-thread.exp
index 3201b1f..3713951 100644
--- a/gdb/testsuite/gdb.threads/signal-while-stepping-over-bp-other-thread.exp
+++ b/gdb/testsuite/gdb.threads/signal-while-stepping-over-bp-other-thread.exp
@@ -100,7 +100,7 @@ gdb_test_sequence $test $test {
"need to step-over"
"resume \\(step=1"
"signal arrived while stepping over breakpoint"
- "switching back to stepped thread"
+ "(restart threads|switching back to stepped thread)"
"stepped to a different line"
"callme"
}
--
1.9.3
^ permalink raw reply [flat|nested] 62+ messages in thread* [PATCH v3 08/17] Factor out code to re-resume stepped thread
2015-04-17 10:47 [PATCH v3 00/23] All-stop on top of non-stop Pedro Alves
2015-04-17 10:45 ` [PATCH v3 15/17] PPC64: Fix gdb.arch/ppc64-atomic-inst.exp with displaced stepping Pedro Alves
2015-04-17 10:45 ` [PATCH v3 11/17] Fix signal-while-stepping-over-bp-other-thread.exp on targets always in non-stop Pedro Alves
@ 2015-04-17 10:45 ` Pedro Alves
2015-04-17 10:45 ` [PATCH v3 05/17] Embed the pending step-over chain in thread_info objects Pedro Alves
` (15 subsequent siblings)
18 siblings, 0 replies; 62+ messages in thread
From: Pedro Alves @ 2015-04-17 10:45 UTC (permalink / raw)
To: gdb-patches
Just a code refactor, no funcionality change intended.
gdb/ChangeLog:
2015-04-17 Pedro Alves <palves@redhat.com>
* infrun.c (keep_going_stepped_thread): New function, factored out
from ...
(switch_back_to_stepped_thread): ... here.
---
gdb/infrun.c | 198 ++++++++++++++++++++++++++++++++---------------------------
1 file changed, 106 insertions(+), 92 deletions(-)
diff --git a/gdb/infrun.c b/gdb/infrun.c
index ce2be31..6415fa2 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -1879,6 +1879,7 @@ reset_ecs (struct execution_control_state *ecs, struct thread_info *tp)
static void keep_going_pass (struct execution_control_state *ecs);
static void prepare_to_wait (struct execution_control_state *ecs);
+static int keep_going_stepped_thread (struct thread_info *tp);
static int thread_still_needs_step_over (struct thread_info *tp);
/* Are there any pending step-over requests? If so, run one now and
@@ -5733,110 +5734,123 @@ switch_back_to_stepped_thread (struct execution_control_state *ecs)
if (stepping_thread != NULL)
{
- struct frame_info *frame;
- struct gdbarch *gdbarch;
-
- tp = stepping_thread;
-
- /* If the stepping thread exited, then don't try to switch
- back and resume it, which could fail in several different
- ways depending on the target. Instead, just keep going.
-
- We can find a stepping dead thread in the thread list in
- two cases:
-
- - The target supports thread exit events, and when the
- target tries to delete the thread from the thread list,
- inferior_ptid pointed at the exiting thread. In such
- case, calling delete_thread does not really remove the
- thread from the list; instead, the thread is left listed,
- with 'exited' state.
-
- - The target's debug interface does not support thread
- exit events, and so we have no idea whatsoever if the
- previously stepping thread is still alive. For that
- reason, we need to synchronously query the target
- now. */
- if (is_exited (tp->ptid)
- || !target_thread_alive (tp->ptid))
- {
- if (debug_infrun)
- fprintf_unfiltered (gdb_stdlog,
- "infrun: not switching back to "
- "stepped thread, it has vanished\n");
-
- delete_thread (tp->ptid);
- return 0;
- }
-
if (debug_infrun)
fprintf_unfiltered (gdb_stdlog,
"infrun: switching back to stepped thread\n");
- reset_ecs (ecs, tp);
- switch_to_thread (tp->ptid);
+ if (keep_going_stepped_thread (stepping_thread))
+ {
+ prepare_to_wait (ecs);
+ return 1;
+ }
+ }
+ }
- stop_pc = regcache_read_pc (get_thread_regcache (tp->ptid));
- frame = get_current_frame ();
- gdbarch = get_frame_arch (frame);
+ return 0;
+}
- /* If the PC of the thread we were trying to single-step has
- changed, then that thread has trapped or been signaled,
- but the event has not been reported to GDB yet. Re-poll
- the target looking for this particular thread's event
- (i.e. temporarily enable schedlock) by:
+/* Set a previously stepped thread back to stepping. Returns true on
+ success, false if the resume is not possible (e.g., the thread
+ vanished). */
+
+static int
+keep_going_stepped_thread (struct thread_info *tp)
+{
+ struct frame_info *frame;
+ struct gdbarch *gdbarch;
+ struct execution_control_state ecss;
+ struct execution_control_state *ecs = &ecss;
- - setting a break at the current PC
- - resuming that particular thread, only (by setting
- trap expected)
+ /* If the stepping thread exited, then don't try to switch back and
+ resume it, which could fail in several different ways depending
+ on the target. Instead, just keep going.
- This prevents us continuously moving the single-step
- breakpoint forward, one instruction at a time,
- overstepping. */
+ We can find a stepping dead thread in the thread list in two
+ cases:
- if (stop_pc != tp->prev_pc)
- {
- ptid_t resume_ptid;
+ - The target supports thread exit events, and when the target
+ tries to delete the thread from the thread list, inferior_ptid
+ pointed at the exiting thread. In such case, calling
+ delete_thread does not really remove the thread from the list;
+ instead, the thread is left listed, with 'exited' state.
- if (debug_infrun)
- fprintf_unfiltered (gdb_stdlog,
- "infrun: expected thread advanced also "
- "(%s -> %s)\n",
- paddress (target_gdbarch (), tp->prev_pc),
- paddress (target_gdbarch (), stop_pc));
-
- /* Clear the info of the previous step-over, as it's no
- longer valid (if the thread was trying to step over a
- breakpoint, it has already succeeded). It's what
- keep_going would do too, if we called it. Do this
- before trying to insert the sss breakpoint, otherwise
- if we were previously trying to step over this exact
- address in another thread, the breakpoint is
- skipped. */
- clear_step_over_info ();
- tp->control.trap_expected = 0;
-
- insert_single_step_breakpoint (get_frame_arch (frame),
- get_frame_address_space (frame),
- stop_pc);
-
- resume_ptid = user_visible_resume_ptid (tp->control.stepping_command);
- do_target_resume (resume_ptid, 0, GDB_SIGNAL_0);
- prepare_to_wait (ecs);
- }
- else
- {
- if (debug_infrun)
- fprintf_unfiltered (gdb_stdlog,
- "infrun: expected thread still "
- "hasn't advanced\n");
- keep_going_pass (ecs);
- }
+ - The target's debug interface does not support thread exit
+ events, and so we have no idea whatsoever if the previously
+ stepping thread is still alive. For that reason, we need to
+ synchronously query the target now. */
- return 1;
- }
+ if (is_exited (tp->ptid)
+ || !target_thread_alive (tp->ptid))
+ {
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: not resuming previously "
+ "stepped thread, it has vanished\n");
+
+ delete_thread (tp->ptid);
+ return 0;
}
- return 0;
+
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: resuming previously stepped thread\n");
+
+ reset_ecs (ecs, tp);
+ switch_to_thread (tp->ptid);
+
+ stop_pc = regcache_read_pc (get_thread_regcache (tp->ptid));
+ frame = get_current_frame ();
+ gdbarch = get_frame_arch (frame);
+
+ /* If the PC of the thread we were trying to single-step has
+ changed, then that thread has trapped or been signaled, but the
+ event has not been reported to GDB yet. Re-poll the target
+ looking for this particular thread's event (i.e. temporarily
+ enable schedlock) by:
+
+ - setting a break at the current PC
+ - resuming that particular thread, only (by setting trap
+ expected)
+
+ This prevents us continuously moving the single-step breakpoint
+ forward, one instruction at a time, overstepping. */
+
+ if (stop_pc != tp->prev_pc)
+ {
+ ptid_t resume_ptid;
+
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: expected thread advanced also (%s -> %s)\n",
+ paddress (target_gdbarch (), tp->prev_pc),
+ paddress (target_gdbarch (), stop_pc));
+
+ /* Clear the info of the previous step-over, as it's no longer
+ valid (if the thread was trying to step over a breakpoint, it
+ has already succeeded). It's what keep_going would do too,
+ if we called it. Do this before trying to insert the sss
+ breakpoint, otherwise if we were previously trying to step
+ over this exact address in another thread, the breakpoint is
+ skipped. */
+ clear_step_over_info ();
+ tp->control.trap_expected = 0;
+
+ insert_single_step_breakpoint (get_frame_arch (frame),
+ get_frame_address_space (frame),
+ stop_pc);
+
+ resume_ptid = user_visible_resume_ptid (tp->control.stepping_command);
+ do_target_resume (resume_ptid, 0, GDB_SIGNAL_0);
+ }
+ else
+ {
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: expected thread still hasn't advanced\n");
+
+ keep_going_pass (ecs);
+ }
+ return 1;
}
/* Is thread TP in the middle of single-stepping? */
--
1.9.3
^ permalink raw reply [flat|nested] 62+ messages in thread* [PATCH v3 05/17] Embed the pending step-over chain in thread_info objects
2015-04-17 10:47 [PATCH v3 00/23] All-stop on top of non-stop Pedro Alves
` (2 preceding siblings ...)
2015-04-17 10:45 ` [PATCH v3 08/17] Factor out code to re-resume stepped thread Pedro Alves
@ 2015-04-17 10:45 ` Pedro Alves
2015-04-21 8:28 ` Yao Qi
` (2 more replies)
2015-04-17 10:45 ` [PATCH v3 13/17] Fix step-over-{trips-on-watchpoint|lands-on-breakpoint}.exp race Pedro Alves
` (14 subsequent siblings)
18 siblings, 3 replies; 62+ messages in thread
From: Pedro Alves @ 2015-04-17 10:45 UTC (permalink / raw)
To: gdb-patches
In order to teach non-stop mode to do in-line step-overs (pause all
threads, remove breakpoint, single-step, reinsert breakpoint, restart
threads), we'll need to be able to queue in-line step over requests,
much like we queue displaced stepping (out-of-line) requests.
Actually, the queue should be the same -- threads wait for their turn
to step past something (breakpoint, watchpoint), doesn't matter what
technique we end up using when the step over actually starts.
I found that the queue management ends up simpler and more efficient
if embedded in the thread objects themselves. This commit converts
the existing displaced stepping queue to that. Later patches will
make the in-line step-overs code paths use it too.
gdb/ChangeLog:
2015-04-17 Pedro Alves <palves@redhat.com>
* gdbthread.h (struct thread_info) <step_over_prev,
step_over_next>: New fields.
(thread_step_over_chain_enqueue, thread_step_over_chain_remove)
(thread_step_over_chain_next, thread_is_in_step_over_chain): New
declarations.
* infrun.c (struct displaced_step_request): Delete.
(struct displaced_step_inferior_state) <step_request_queue>:
Delete field.
(displaced_step_in_progress): New function.
(displaced_step_prepare): Assert that trap_expected is set. Use
thread_step_over_chain_enqueue. Split starting a new displaced
step to ...
(start_step_over): ... this new function.
(resume): Assert the thread isn't waiting for a step over already.
(proceed): Assert the thread isn't waiting for a step over
already.
(infrun_thread_stop_requested): Adjust to remove threads from the
embedded step-over chain.
(handle_inferior_event) <fork/vfork>: Call start_step_over after
displaced_step_fixup.
(handle_signal_stop): Call start_step_over after
displaced_step_fixup.
* infrun.h (step_over_queue_head): New declaration.
* thread.c (step_over_chain_enqueue, step_over_chain_remove)
(thread_step_over_chain_next, thread_is_in_step_over_chain)
(thread_step_over_chain_enqueue)
(thread_step_over_chain_remove): New functions.
(delete_thread_1): Remove thread from the step-over chain.
v3:
More comments. The step-over chain is now a global instead of
being per-inferior. Previous versions had actually broken
multiple-processes displaced stepping at the same time. Added new
thread_is_in_step_over_chain predicate, and new
thread_step_over_chain_next helper function. Should make reading
the code a bit easier (following patches adjusted too).
---
gdb/gdbthread.h | 23 ++++++++++
gdb/infrun.c | 133 +++++++++++++++++++++++++++-----------------------------
gdb/infrun.h | 4 ++
gdb/thread.c | 84 +++++++++++++++++++++++++++++++++++
4 files changed, 176 insertions(+), 68 deletions(-)
diff --git a/gdb/gdbthread.h b/gdb/gdbthread.h
index ff7cec2..2c871a2 100644
--- a/gdb/gdbthread.h
+++ b/gdb/gdbthread.h
@@ -285,6 +285,12 @@ struct thread_info
/* Values that are stored as temporaries on stack while evaluating
expressions. */
value_vec *stack_temporaries;
+
+ /* Step-over chain. A thread is in the step-over queue if these are
+ non-NULL. If only a single thread is in the chain, then these
+ fields point to self. */
+ struct thread_info *step_over_prev;
+ struct thread_info *step_over_next;
};
/* Create an empty thread list, or empty the existing one. */
@@ -502,6 +508,23 @@ extern struct value *get_last_thread_stack_temporary (ptid_t);
extern int value_in_thread_stack_temporaries (struct value *, ptid_t);
+/* Add TP to the end of its inferior's pending step-over chain. */
+
+extern void thread_step_over_chain_enqueue (struct thread_info *tp);
+
+/* Remove TP from its inferior's pending step-over chain. */
+
+extern void thread_step_over_chain_remove (struct thread_info *tp);
+
+/* Return the next thread in the step-over chain starting at TP. NULL
+ if TP is the last entry in the chain. */
+
+extern struct thread_info *thread_step_over_chain_next (struct thread_info *tp);
+
+/* Return true if TP is in the step-over chain. */
+
+extern int thread_is_in_step_over_chain (struct thread_info *tp);
+
extern struct thread_info *thread_list;
#endif /* GDBTHREAD_H */
diff --git a/gdb/infrun.c b/gdb/infrun.c
index 534ecef..f325a53 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -1223,6 +1223,15 @@ follow_exec (ptid_t ptid, char *execd_pathname)
matically get reset there in the new process.). */
}
+/* The queue of threads that need to do a step-over operation to get
+ past e.g., a breakpoint. What technique is used to step over the
+ breakpoint/watchpoint does not matter -- all threads end up in the
+ same queue, to maintain rough temporal order of execution, in order
+ to avoid starvation, otherwise, we could e.g., find ourselves
+ constantly stepping the same couple threads past their breakpoints
+ over and over, if the single-step finish fast enough. */
+struct thread_info *step_over_queue_head;
+
/* Bit flags indicating what the thread needs to step over. */
enum step_over_what
@@ -1419,12 +1428,6 @@ step_over_info_valid_p (void)
displaced step operation on it. See displaced_step_prepare and
displaced_step_fixup for details. */
-struct displaced_step_request
-{
- ptid_t ptid;
- struct displaced_step_request *next;
-};
-
/* Per-inferior displaced stepping state. */
struct displaced_step_inferior_state
{
@@ -1434,10 +1437,6 @@ struct displaced_step_inferior_state
/* The process this displaced step state refers to. */
int pid;
- /* A queue of pending displaced stepping requests. One entry per
- thread that needs to do a displaced step. */
- struct displaced_step_request *step_request_queue;
-
/* If this is not null_ptid, this is the thread carrying out a
displaced single-step in process PID. This thread's state will
require fixing up once it has completed its step. */
@@ -1669,6 +1668,9 @@ displaced_step_prepare (ptid_t ptid)
support displaced stepping. */
gdb_assert (gdbarch_displaced_step_copy_insn_p (gdbarch));
+ /* Nor if the thread isn't meant to step over a breakpoint. */
+ gdb_assert (tp->control.trap_expected);
+
/* Disable range stepping while executing in the scratch pad. We
want a single-step even if executing the displaced instruction in
the scratch buffer lands within the stepping range (e.g., a
@@ -1684,28 +1686,13 @@ displaced_step_prepare (ptid_t ptid)
{
/* Already waiting for a displaced step to finish. Defer this
request and place in queue. */
- struct displaced_step_request *req, *new_req;
if (debug_displaced)
fprintf_unfiltered (gdb_stdlog,
- "displaced: defering step of %s\n",
+ "displaced: deferring step of %s\n",
target_pid_to_str (ptid));
- new_req = xmalloc (sizeof (*new_req));
- new_req->ptid = ptid;
- new_req->next = NULL;
-
- if (displaced->step_request_queue)
- {
- for (req = displaced->step_request_queue;
- req && req->next;
- req = req->next)
- ;
- req->next = new_req;
- }
- else
- displaced->step_request_queue = new_req;
-
+ thread_step_over_chain_enqueue (tp);
return 0;
}
else
@@ -1855,24 +1842,44 @@ displaced_step_fixup (ptid_t event_ptid, enum gdb_signal signal)
do_cleanups (old_cleanups);
displaced->step_ptid = null_ptid;
+}
- /* Are there any pending displaced stepping requests? If so, run
- one now. Leave the state object around, since we're likely to
- need it again soon. */
- while (displaced->step_request_queue)
+/* Are there any pending step-over requests? If so, run one now. */
+
+static void
+start_step_over (void)
+{
+ struct thread_info *tp, *next;
+
+ for (tp = step_over_queue_head; tp != NULL; tp = next)
{
- struct displaced_step_request *head;
ptid_t ptid;
+ struct displaced_step_inferior_state *displaced;
struct regcache *regcache;
struct gdbarch *gdbarch;
CORE_ADDR actual_pc;
struct address_space *aspace;
+ struct inferior *inf = find_inferior_ptid (tp->ptid);
+
+ next = thread_step_over_chain_next (tp);
- head = displaced->step_request_queue;
- ptid = head->ptid;
- displaced->step_request_queue = head->next;
- xfree (head);
+ displaced = get_displaced_stepping_state (inf->pid);
+ /* If this inferior already has a displaced step in process,
+ don't start a new one. */
+ if (!ptid_equal (displaced->step_ptid, null_ptid))
+ continue;
+
+ thread_step_over_chain_remove (tp);
+
+ if (step_over_queue_head == NULL)
+ {
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: step-over queue now empty\n");
+ }
+
+ ptid = tp->ptid;
context_switch (ptid);
regcache = get_thread_regcache (ptid);
@@ -1935,6 +1942,10 @@ displaced_step_fixup (ptid_t event_ptid, enum gdb_signal signal)
/* This request was discarded. See if there's any other
thread waiting for its turn. */
}
+
+ /* A new displaced stepping sequence started. Maybe we can
+ start a displaced step on a thread of other process.
+ Continue looking. */
}
}
@@ -1955,10 +1966,6 @@ infrun_thread_ptid_changed (ptid_t old_ptid, ptid_t new_ptid)
{
if (ptid_equal (displaced->step_ptid, old_ptid))
displaced->step_ptid = new_ptid;
-
- for (it = displaced->step_request_queue; it; it = it->next)
- if (ptid_equal (it->ptid, old_ptid))
- it->ptid = new_ptid;
}
}
@@ -2137,6 +2144,8 @@ resume (enum gdb_signal sig)
tp->stepped_breakpoint = 0;
+ gdb_assert (!thread_is_in_step_over_chain (tp));
+
QUIT;
/* Depends on stepped_breakpoint. */
@@ -2667,6 +2676,8 @@ proceed (CORE_ADDR addr, enum gdb_signal siggnal)
/* Fill in with reasonable starting values. */
init_thread_stepping_state (tp);
+ gdb_assert (!thread_is_in_step_over_chain (tp));
+
if (addr == (CORE_ADDR) -1)
{
if (pc == stop_pc
@@ -2972,35 +2983,17 @@ infrun_thread_stop_requested_callback (struct thread_info *info, void *arg)
static void
infrun_thread_stop_requested (ptid_t ptid)
{
- struct displaced_step_inferior_state *displaced;
-
- /* PTID was requested to stop. Remove it from the displaced
- stepping queue, so we don't try to resume it automatically. */
-
- for (displaced = displaced_step_inferior_states;
- displaced;
- displaced = displaced->next)
- {
- struct displaced_step_request *it, **prev_next_p;
-
- it = displaced->step_request_queue;
- prev_next_p = &displaced->step_request_queue;
- while (it)
- {
- if (ptid_match (it->ptid, ptid))
- {
- *prev_next_p = it->next;
- it->next = NULL;
- xfree (it);
- }
- else
- {
- prev_next_p = &it->next;
- }
+ struct thread_info *tp;
- it = *prev_next_p;
- }
- }
+ /* PTID was requested to stop. Remove matching threads from the
+ step-over queue, so we don't try to resume them
+ automatically. */
+ ALL_NON_EXITED_THREADS (tp)
+ if (ptid_match (tp->ptid, ptid))
+ {
+ if (thread_is_in_step_over_chain (tp))
+ thread_step_over_chain_remove (tp);
+ }
iterate_over_threads (infrun_thread_stop_requested_callback, &ptid);
}
@@ -4051,6 +4044,9 @@ Cannot fill $_exitsignal with the correct signal number.\n"));
that this operation also cleans up the child process for vfork,
because their pages are shared. */
displaced_step_fixup (ecs->ptid, GDB_SIGNAL_TRAP);
+ /* Start a new step-over in another thread if there's one
+ that needs it. */
+ start_step_over ();
if (ecs->ws.kind == TARGET_WAITKIND_FORKED)
{
@@ -4283,6 +4279,7 @@ handle_signal_stop (struct execution_control_state *ecs)
the PC, so do it here, before we set stop_pc.) */
displaced_step_fixup (ecs->ptid,
ecs->event_thread->suspend.stop_signal);
+ start_step_over ();
/* If we either finished a single-step or hit a breakpoint, but
the user wanted this thread to be stopped, pretend we got a
diff --git a/gdb/infrun.h b/gdb/infrun.h
index 1f09e41..bfce810 100644
--- a/gdb/infrun.h
+++ b/gdb/infrun.h
@@ -196,4 +196,8 @@ extern void signal_catch_update (const unsigned int *);
systems. Use of symbolic signal names is strongly encouraged. */
enum gdb_signal gdb_signal_from_command (int num);
+/* The global queue of threads that need to do a step-over operation
+ to get past e.g., a breakpoint. */
+extern struct thread_info *step_over_queue_head;
+
#endif /* INFRUN_H */
diff --git a/gdb/thread.c b/gdb/thread.c
index 46b5947..28e5ef8 100644
--- a/gdb/thread.c
+++ b/gdb/thread.c
@@ -307,6 +307,86 @@ add_thread (ptid_t ptid)
return add_thread_with_info (ptid, NULL);
}
+/* Add TP to the end of the step-over chain LIST_P. */
+
+static void
+step_over_chain_enqueue (struct thread_info **list_p, struct thread_info *tp)
+{
+ gdb_assert (tp->step_over_next == NULL);
+ gdb_assert (tp->step_over_prev == NULL);
+
+ if (*list_p == NULL)
+ {
+ *list_p = tp;
+ tp->step_over_prev = tp->step_over_next = tp;
+ }
+ else
+ {
+ struct thread_info *head = *list_p;
+ struct thread_info *tail = head->step_over_prev;
+
+ tp->step_over_prev = tail;
+ tp->step_over_next = head;
+ head->step_over_prev = tp;
+ tail->step_over_next = tp;
+ }
+}
+
+/* Remove TP from step-over chain LIST_P. */
+
+static void
+step_over_chain_remove (struct thread_info **list_p, struct thread_info *tp)
+{
+ gdb_assert (tp->step_over_next != NULL);
+ gdb_assert (tp->step_over_prev != NULL);
+
+ if (*list_p == tp)
+ {
+ if (tp == tp->step_over_next)
+ *list_p = NULL;
+ else
+ *list_p = tp->step_over_next;
+ }
+
+ tp->step_over_prev->step_over_next = tp->step_over_next;
+ tp->step_over_next->step_over_prev = tp->step_over_prev;
+ tp->step_over_prev = tp->step_over_next = NULL;
+}
+
+/* See gdbthread.h. */
+
+struct thread_info *
+thread_step_over_chain_next (struct thread_info *tp)
+{
+ struct thread_info *next = tp->step_over_next;
+
+ return (next == step_over_queue_head ? NULL : next);
+}
+
+/* See gdbthread.h. */
+
+int
+thread_is_in_step_over_chain (struct thread_info *tp)
+{
+ return (tp->step_over_next != NULL);
+}
+
+/* See gdbthread.h. */
+
+void
+thread_step_over_chain_enqueue (struct thread_info *tp)
+{
+ step_over_chain_enqueue (&step_over_queue_head, tp);
+}
+
+/* See gdbthread.h. */
+
+void
+thread_step_over_chain_remove (struct thread_info *tp)
+{
+ step_over_chain_remove (&step_over_queue_head, tp);
+}
+
/* Delete thread PTID. If SILENT, don't notify the observer of this
exit. */
static void
@@ -323,6 +403,10 @@ delete_thread_1 (ptid_t ptid, int silent)
if (!tp)
return;
+ /* Dead threads don't need to step-over. Remove from queue. */
+ if (tp->step_over_next != NULL)
+ thread_step_over_chain_remove (tp);
+
/* If this is the current thread, or there's code out there that
relies on it existing (refcount > 0) we can't delete yet. Mark
it as exited, and notify it. */
--
1.9.3
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 05/17] Embed the pending step-over chain in thread_info objects
2015-04-17 10:45 ` [PATCH v3 05/17] Embed the pending step-over chain in thread_info objects Pedro Alves
@ 2015-04-21 8:28 ` Yao Qi
2015-04-22 20:14 ` Pedro Alves
2015-04-21 9:53 ` Yao Qi
2015-04-22 4:25 ` Doug Evans
2 siblings, 1 reply; 62+ messages in thread
From: Yao Qi @ 2015-04-21 8:28 UTC (permalink / raw)
To: Pedro Alves; +Cc: gdb-patches
Pedro Alves <palves@redhat.com> writes:
Hi Pedro,
This patch looks good to me, some questions below.
> (displaced_step_prepare): Assert that trap_expected is set. Use
> thread_step_over_chain_enqueue. Split starting a new displaced
> step to ...
> (start_step_over): ... this new function.
If I read this patch correctly, start_step_over is moved from
displaced_step_fixup. That is why we call start_step_over after each
displaced_step_fixup.
> v3:
>
> More comments. The step-over chain is now a global instead of
> being per-inferior. Previous versions had actually broken
> multiple-processes displaced stepping at the same time. Added new
How does per-inferior step-over chain (or displaced stepping queue)
break multi-process displaced stepping?
> @@ -2972,35 +2983,17 @@ infrun_thread_stop_requested_callback (struct thread_info *info, void *arg)
> static void
> infrun_thread_stop_requested (ptid_t ptid)
> {
> - struct displaced_step_inferior_state *displaced;
> -
> - /* PTID was requested to stop. Remove it from the displaced
> - stepping queue, so we don't try to resume it automatically. */
> -
> - for (displaced = displaced_step_inferior_states;
> - displaced;
> - displaced = displaced->next)
> - {
> - struct displaced_step_request *it, **prev_next_p;
> -
> - it = displaced->step_request_queue;
> - prev_next_p = &displaced->step_request_queue;
> - while (it)
> - {
> - if (ptid_match (it->ptid, ptid))
> - {
> - *prev_next_p = it->next;
> - it->next = NULL;
> - xfree (it);
> - }
> - else
> - {
> - prev_next_p = &it->next;
> - }
> + struct thread_info *tp;
>
> - it = *prev_next_p;
> - }
> - }
> + /* PTID was requested to stop. Remove matching threads from the
> + step-over queue, so we don't try to resume them
> + automatically. */
I can understand the code below, except the comment "we don't try to
resume them automatically". It looks not necessary here.
> + ALL_NON_EXITED_THREADS (tp)
> + if (ptid_match (tp->ptid, ptid))
> + {
> + if (thread_is_in_step_over_chain (tp))
> + thread_step_over_chain_remove (tp);
> + }
>
> iterate_over_threads (infrun_thread_stop_requested_callback, &ptid);
> }
> @@ -4051,6 +4044,9 @@ Cannot fill $_exitsignal with the correct signal number.\n"));
> that this operation also cleans up the child process for vfork,
> because their pages are shared. */
> displaced_step_fixup (ecs->ptid, GDB_SIGNAL_TRAP);
> + /* Start a new step-over in another thread if there's one
> + that needs it. */
> + start_step_over ();
The comment is confusing to me, especially the "one" and the "it". Do
you mean "in another thread if there is one thread that needs step-over"?
> @@ -323,6 +403,10 @@ delete_thread_1 (ptid_t ptid, int silent)
> if (!tp)
> return;
>
> + /* Dead threads don't need to step-over. Remove from queue. */
> + if (tp->step_over_next != NULL)
> + thread_step_over_chain_remove (tp);
> +
I am wondering how this can happen? A thread needs step-over becomes dead?
> /* If this is the current thread, or there's code out there that
> relies on it existing (refcount > 0) we can't delete yet. Mark
> it as exited, and notify it. */
--
Yao (齐尧)
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 05/17] Embed the pending step-over chain in thread_info objects
2015-04-21 8:28 ` Yao Qi
@ 2015-04-22 20:14 ` Pedro Alves
0 siblings, 0 replies; 62+ messages in thread
From: Pedro Alves @ 2015-04-22 20:14 UTC (permalink / raw)
To: Yao Qi; +Cc: gdb-patches
On 04/21/2015 09:28 AM, Yao Qi wrote:
> Pedro Alves <palves@redhat.com> writes:
>
> Hi Pedro,
> This patch looks good to me, some questions below.
>
>> (displaced_step_prepare): Assert that trap_expected is set. Use
>> thread_step_over_chain_enqueue. Split starting a new displaced
>> step to ...
>> (start_step_over): ... this new function.
>
> If I read this patch correctly, start_step_over is moved from
> displaced_step_fixup. That is why we call start_step_over after each
> displaced_step_fixup.
Correct.
>
>> v3:
>>
>> More comments. The step-over chain is now a global instead of
>> being per-inferior. Previous versions had actually broken
>> multiple-processes displaced stepping at the same time. Added new
>
> How does per-inferior step-over chain (or displaced stepping queue)
> break multi-process displaced stepping?
v1 and v2 put the head of the step-over chain in the inferior
structure. start_step_over would look up the inferior structure of
the thread that just finished the step over, and try to
start a step-over of another thread of that inferior.
And if we had just finished an in-line step over, and the step-over that we
could start now is a displaced-step of _another_ inferior, in v2,
we wouldn't start it (because start_step_over_inferior wouldn't see that
thread).
And if we could start a displaced-step for a thread of the event
inferior, start_step_over would return immediately, instead
of trying to start a displaced step in another inferior too.
Even in-line step-overs were broken. E.g., say you have two inferiors,
each with one thread. Everything is stopped at a breakpoint that
must be stepped over. sched-multi is on, and the user does "continue"
to continue both inferiors. We'd start an in-line step-over for
inferior 1. Once that is done, we'd try starting a new step over
in the same inferior, and we'd miss that the other inferior
has a thread to step over too.
I went a bit in circles a trying to address that. The fact that for
in-line step overs we must stop all threads currently (we should be
able to stop only threads sharing the stepped thread's address space
instead, but we're not there yet), but not for displaced stepping
started making it too complicated. Related I also considered that we
could have more that one displaced step scratch pad slot per
inferior (e.g., "reserve" a few more bytes around the entry point).
Another thing I realized is that per-inferior queue breaks the
forward-progress-guarantee ordering, as we'd be giving priority
to start step overs on threads of the same inferior that had
just finished a step over, potentially starving threads of other
inferiors. The ordering issue is that there was no ordering
between the step-over chains of the multiple inferiors, so if we
left the chain per-inferior, we wouldn't know which inferior's
chain had the thread that was waiting for a step-over for the
longest time.
All in all, I realized that a single list is simpler and
more flexible.
>
>> @@ -2972,35 +2983,17 @@ infrun_thread_stop_requested_callback (struct thread_info *info, void *arg)
>> static void
>> infrun_thread_stop_requested (ptid_t ptid)
>> {
>> - struct displaced_step_inferior_state *displaced;
>> -
>> - /* PTID was requested to stop. Remove it from the displaced
>> - stepping queue, so we don't try to resume it automatically. */
>> -
>> - for (displaced = displaced_step_inferior_states;
>> - displaced;
>> - displaced = displaced->next)
>> - {
>> - struct displaced_step_request *it, **prev_next_p;
>> -
>> - it = displaced->step_request_queue;
>> - prev_next_p = &displaced->step_request_queue;
>> - while (it)
>> - {
>> - if (ptid_match (it->ptid, ptid))
>> - {
>> - *prev_next_p = it->next;
>> - it->next = NULL;
>> - xfree (it);
>> - }
>> - else
>> - {
>> - prev_next_p = &it->next;
>> - }
>> + struct thread_info *tp;
>>
>> - it = *prev_next_p;
>> - }
>> - }
>> + /* PTID was requested to stop. Remove matching threads from the
>> + step-over queue, so we don't try to resume them
>> + automatically. */
>
> I can understand the code below, except the comment "we don't try to
> resume them automatically". It looks not necessary here.
By "resumed automatically" I meant that if thread A is left in
the step-over chain (or currently in mainline in the displaced step queue)
it ends up stepped when all others threads in the step over queue
are done with their steps, even if thread A is supposed to be stopped.
But I agree it's not really necessary. I'll remove it.
>
>> + ALL_NON_EXITED_THREADS (tp)
>> + if (ptid_match (tp->ptid, ptid))
>> + {
>> + if (thread_is_in_step_over_chain (tp))
>> + thread_step_over_chain_remove (tp);
>> + }
>>
>> iterate_over_threads (infrun_thread_stop_requested_callback, &ptid);
>> }
>> @@ -4051,6 +4044,9 @@ Cannot fill $_exitsignal with the correct signal number.\n"));
>> that this operation also cleans up the child process for vfork,
>> because their pages are shared. */
>> displaced_step_fixup (ecs->ptid, GDB_SIGNAL_TRAP);
>> + /* Start a new step-over in another thread if there's one
>> + that needs it. */
>> + start_step_over ();
>
> The comment is confusing to me, especially the "one" and the "it". Do
> you mean "in another thread if there is one thread that needs step-over"?
>
>> @@ -323,6 +403,10 @@ delete_thread_1 (ptid_t ptid, int silent)
>> if (!tp)
>> return;
>>
>> + /* Dead threads don't need to step-over. Remove from queue. */
>> + if (tp->step_over_next != NULL)
>> + thread_step_over_chain_remove (tp);
>> +
>
> I am wondering how this can happen? A thread needs step-over becomes dead?
It can happen if the process exits or disappears (target is closed, etc.)
while the thread was waiting for its turn.
Thanks,
Pedro Alves
^ permalink raw reply [flat|nested] 62+ messages in thread
* Re: [PATCH v3 05/17] Embed the pending step-over chain in thread_info objects
2015-04-17 10:45 ` [PATCH v3 05/17] Embed the pending step-over chain in thread_info objects Pedro Alves
2015-04-21 8:28 ` Yao Qi
@ 2015-04-21 9:53 ` Yao Qi
2015-04-22 19:07 ` Pedro Alves
2015-04-22 4:25 ` Doug Evans
2 siblings, 1 reply; 62+ messages in thread
From: Yao Qi @ 2015-04-21 9:53 UTC (permalink / raw)
To: Pedro Alves; +Cc: gdb-patches
Pedro Alves <palves@redhat.com> writes:
> (displaced_step_in_progress): New function.
This changelog entry can be removed as this function has been in FSF
tree already.
--
Yao (齐尧)
^ permalink raw reply [flat|nested] 62+ messages in thread
* Re: [PATCH v3 05/17] Embed the pending step-over chain in thread_info objects
2015-04-21 9:53 ` Yao Qi
@ 2015-04-22 19:07 ` Pedro Alves
0 siblings, 0 replies; 62+ messages in thread
From: Pedro Alves @ 2015-04-22 19:07 UTC (permalink / raw)
To: Yao Qi; +Cc: gdb-patches
On 04/21/2015 10:52 AM, Yao Qi wrote:
> Pedro Alves <palves@redhat.com> writes:
>
>> (displaced_step_in_progress): New function.
>
> This changelog entry can be removed as this function has been in FSF
> tree already.
Good catch. Fixed.
Thanks,
Pedro Alves
^ permalink raw reply [flat|nested] 62+ messages in thread
* Re: [PATCH v3 05/17] Embed the pending step-over chain in thread_info objects
2015-04-17 10:45 ` [PATCH v3 05/17] Embed the pending step-over chain in thread_info objects Pedro Alves
2015-04-21 8:28 ` Yao Qi
2015-04-21 9:53 ` Yao Qi
@ 2015-04-22 4:25 ` Doug Evans
2015-04-22 22:19 ` Pedro Alves
2 siblings, 1 reply; 62+ messages in thread
From: Doug Evans @ 2015-04-22 4:25 UTC (permalink / raw)
To: Pedro Alves; +Cc: gdb-patches
Pedro Alves writes:
> In order to teach non-stop mode to do in-line step-overs (pause all
> threads, remove breakpoint, single-step, reinsert breakpoint, restart
> threads), we'll need to be able to queue in-line step over requests,
> much like we queue displaced stepping (out-of-line) requests.
> Actually, the queue should be the same -- threads wait for their turn
> to step past something (breakpoint, watchpoint), doesn't matter what
> technique we end up using when the step over actually starts.
>
> I found that the queue management ends up simpler and more efficient
> if embedded in the thread objects themselves. This commit converts
> the existing displaced stepping queue to that. Later patches will
> make the in-line step-overs code paths use it too.
>
> gdb/ChangeLog:
> 2015-04-17 Pedro Alves <palves@redhat.com>
>
> * gdbthread.h (struct thread_info) <step_over_prev,
> step_over_next>: New fields.
> (thread_step_over_chain_enqueue, thread_step_over_chain_remove)
> (thread_step_over_chain_next, thread_is_in_step_over_chain): New
> declarations.
> * infrun.c (struct displaced_step_request): Delete.
> (struct displaced_step_inferior_state) <step_request_queue>:
> Delete field.
> (displaced_step_in_progress): New function.
> (displaced_step_prepare): Assert that trap_expected is set. Use
> thread_step_over_chain_enqueue. Split starting a new displaced
> step to ...
> (start_step_over): ... this new function.
> (resume): Assert the thread isn't waiting for a step over already.
> (proceed): Assert the thread isn't waiting for a step over
> already.
> (infrun_thread_stop_requested): Adjust to remove threads from the
> embedded step-over chain.
> (handle_inferior_event) <fork/vfork>: Call start_step_over after
> displaced_step_fixup.
> (handle_signal_stop): Call start_step_over after
> displaced_step_fixup.
> * infrun.h (step_over_queue_head): New declaration.
> * thread.c (step_over_chain_enqueue, step_over_chain_remove)
> (thread_step_over_chain_next, thread_is_in_step_over_chain)
> (thread_step_over_chain_enqueue)
> (thread_step_over_chain_remove): New functions.
> (delete_thread_1): Remove thread from the step-over chain.
>
> v3:
>
> More comments. The step-over chain is now a global instead of
> being per-inferior. Previous versions had actually broken
> multiple-processes displaced stepping at the same time. Added new
> thread_is_in_step_over_chain predicate, and new
> thread_step_over_chain_next helper function. Should make reading
> the code a bit easier (following patches adjusted too).
>...
>
> diff --git a/gdb/infrun.c b/gdb/infrun.c
> index 534ecef..f325a53 100644
> --- a/gdb/infrun.c
> +++ b/gdb/infrun.c
>...
> @@ -1855,24 +1842,44 @@ displaced_step_fixup (ptid_t event_ptid, enum gdb_signal signal)
> do_cleanups (old_cleanups);
>
> displaced->step_ptid = null_ptid;
> +}
>
> - /* Are there any pending displaced stepping requests? If so, run
> - one now. Leave the state object around, since we're likely to
> - need it again soon. */
> - while (displaced->step_request_queue)
> +/* Are there any pending step-over requests? If so, run one now. */
Hi.
Nit: IIUC "run one now" should read "run all we can now".
> +
> +static void
> +start_step_over (void)
> +{
>...
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 05/17] Embed the pending step-over chain in thread_info objects
2015-04-22 4:25 ` Doug Evans
@ 2015-04-22 22:19 ` Pedro Alves
0 siblings, 0 replies; 62+ messages in thread
From: Pedro Alves @ 2015-04-22 22:19 UTC (permalink / raw)
To: Doug Evans; +Cc: gdb-patches
On 04/22/2015 05:24 AM, Doug Evans wrote:
> > --- a/gdb/infrun.c
> > +++ b/gdb/infrun.c
> >...
> > @@ -1855,24 +1842,44 @@ displaced_step_fixup (ptid_t event_ptid, enum gdb_signal signal)
> > do_cleanups (old_cleanups);
> >
> > displaced->step_ptid = null_ptid;
> > +}
> >
> > - /* Are there any pending displaced stepping requests? If so, run
> > - one now. Leave the state object around, since we're likely to
> > - need it again soon. */
> > - while (displaced->step_request_queue)
> > +/* Are there any pending step-over requests? If so, run one now. */
>
> Hi.
>
> Nit: IIUC "run one now" should read "run all we can now".
Thanks, I did that change.
--
Pedro Alves
^ permalink raw reply [flat|nested] 62+ messages in thread
* [PATCH v3 13/17] Fix step-over-{trips-on-watchpoint|lands-on-breakpoint}.exp race
2015-04-17 10:47 [PATCH v3 00/23] All-stop on top of non-stop Pedro Alves
` (3 preceding siblings ...)
2015-04-17 10:45 ` [PATCH v3 05/17] Embed the pending step-over chain in thread_info objects Pedro Alves
@ 2015-04-17 10:45 ` Pedro Alves
2015-04-17 10:45 ` [PATCH v3 03/17] remote.c/all-stop: Implement TARGET_WAITKIND_NO_RESUMED and TARGET_WNOHANG Pedro Alves
` (13 subsequent siblings)
18 siblings, 0 replies; 62+ messages in thread
From: Pedro Alves @ 2015-04-17 10:45 UTC (permalink / raw)
To: gdb-patches
On a target that is both always in non-stop mode and can do displaced
stepping (such as native x86_64 GNU/Linux, with "maint set
target-non-stop on"), the step-over-trips-on-watchpoint.exp test
sometimes fails like this:
(gdb) PASS: gdb.threads/step-over-trips-on-watchpoint.exp: no thread-specific bp: step: thread 1
set scheduler-locking off
(gdb) PASS: gdb.threads/step-over-trips-on-watchpoint.exp: no thread-specific bp: step: set scheduler-locking off
step
-[Switching to Thread 0x7ffff7fc0700 (LWP 11782)]
-Hardware watchpoint 4: watch_me
-
-Old value = 0
-New value = 1
-child_function (arg=0x0) at /home/pedro/gdb/mygit/src/gdb/testsuite/gdb.threads/step-over-trips-on-watchpoint.c:39
-39 other = 1; /* set thread-specific breakpoint here */
-(gdb) PASS: gdb.threads/step-over-trips-on-watchpoint.exp: no thread-specific bp: step: step
+wait_threads () at /home/pedro/gdb/mygit/src/gdb/testsuite/gdb.threads/step-over-trips-on-watchpoint.c:49
+49 return 1; /* in wait_threads */
+(gdb) FAIL: gdb.threads/step-over-trips-on-watchpoint.exp: no thread-specific bp: step: step
Note "scheduler-locking" was set off. The problem is that on such
targets, the step-over of thread 2 and the "step" of thread 1 can be
set to run simultaneously (since with displaced stepping the
breakpoint isn't ever removed from the target), and sometimes, the
"step" of thread 1 finishes first, so it'd take another resume to see
the watchpoint trigger. Fix this by replacing the wait_threads
function with a one-line infinite loop that doesn't call any function,
so that the "step" of thread 1 never finishes.
gdb/testsuite/ChangeLog:
2015-04-17 Pedro Alves <palves@redhat.com>
* gdb.threads/step-over-lands-on-breakpoint.c (wait_threads):
Delete function.
(main): Add alarm. Run an infinite loop instead of calling
wait_threads.
* gdb.threads/step-over-lands-on-breakpoint.exp (do_test): Change
comment.
* gdb.threads/step-over-trips-on-watchpoint.c (wait_threads):
Delete function.
(main): Add alarm. Run an infinite loop instead of calling
wait_threads.
* gdb.threads/step-over-trips-on-watchpoint.exp (do_test): Change
comment.
v3:
- Rebased.
---
.../gdb.threads/step-over-lands-on-breakpoint.c | 17 ++++++++++-------
.../gdb.threads/step-over-lands-on-breakpoint.exp | 6 +++---
.../gdb.threads/step-over-trips-on-watchpoint.c | 17 ++++++++++-------
.../gdb.threads/step-over-trips-on-watchpoint.exp | 9 ++++-----
4 files changed, 27 insertions(+), 22 deletions(-)
diff --git a/gdb/testsuite/gdb.threads/step-over-lands-on-breakpoint.c b/gdb/testsuite/gdb.threads/step-over-lands-on-breakpoint.c
index 0a6ed8f..2480164 100644
--- a/gdb/testsuite/gdb.threads/step-over-lands-on-breakpoint.c
+++ b/gdb/testsuite/gdb.threads/step-over-lands-on-breakpoint.c
@@ -41,23 +41,26 @@ child_function (void *arg)
pthread_exit (NULL);
}
-static int
-wait_threads (void)
-{
- return 1; /* in wait_threads */
-}
-
int
main ()
{
int res;
long i;
+ alarm (300);
+
pthread_barrier_init (&barrier, NULL, 2);
res = pthread_create (&child_thread, NULL, child_function, NULL);
pthread_barrier_wait (&barrier);
- wait_threads (); /* set wait-thread breakpoint here */
+
+ /* Use an infinite loop with no function calls so that "step" over
+ this line never finishes before the breakpoint in the other
+ thread triggers. That can happen if the step-over of thread 2 is
+ done with displaced stepping on a target that is always in
+ non-stop mode, as in that case GDB runs both threads
+ simultaneously. */
+ while (1); /* set wait-thread breakpoint here */
pthread_join (child_thread, NULL);
diff --git a/gdb/testsuite/gdb.threads/step-over-lands-on-breakpoint.exp b/gdb/testsuite/gdb.threads/step-over-lands-on-breakpoint.exp
index 52b59ec..b38f23b 100644
--- a/gdb/testsuite/gdb.threads/step-over-lands-on-breakpoint.exp
+++ b/gdb/testsuite/gdb.threads/step-over-lands-on-breakpoint.exp
@@ -59,9 +59,9 @@ proc do_test {displaced command} {
gdb_test_no_output "set scheduler-locking off"
# Thread 2 is still stopped at a breakpoint that needs to be
- # stepped over before proceeding thread 1. However, right
- # where the step-over lands there's another breakpoint
- # installed, which should trap and be reported to the user.
+ # stepped over. However, right where the step-over lands
+ # there's another breakpoint installed, which should trap and
+ # be reported to the user.
gdb_test "$command" "step-over here.*"
}
}
diff --git a/gdb/testsuite/gdb.threads/step-over-trips-on-watchpoint.c b/gdb/testsuite/gdb.threads/step-over-trips-on-watchpoint.c
index 6cf97fb..34ba079 100644
--- a/gdb/testsuite/gdb.threads/step-over-trips-on-watchpoint.c
+++ b/gdb/testsuite/gdb.threads/step-over-trips-on-watchpoint.c
@@ -43,23 +43,26 @@ child_function (void *arg)
pthread_exit (NULL);
}
-static int
-wait_threads (void)
-{
- return 1; /* in wait_threads */
-}
-
int
main ()
{
int res;
long i;
+ alarm (300);
+
pthread_barrier_init (&barrier, NULL, 2);
res = pthread_create (&child_thread, NULL, child_function, NULL);
pthread_barrier_wait (&barrier);
- wait_threads (); /* set wait-thread breakpoint here */
+
+ /* Use an infinite loop with no function calls so that "step" over
+ this line never finishes before the watchpoint in the other
+ thread triggers. That can happen if the step-over of thread 2 is
+ done with displaced stepping on a target that is always in
+ non-stop mode, as in that case GDB runs both threads
+ simultaneously. */
+ while (1); /* set wait-thread breakpoint here */
pthread_join (child_thread, NULL);
diff --git a/gdb/testsuite/gdb.threads/step-over-trips-on-watchpoint.exp b/gdb/testsuite/gdb.threads/step-over-trips-on-watchpoint.exp
index 89b66e5..7e27f97 100644
--- a/gdb/testsuite/gdb.threads/step-over-trips-on-watchpoint.exp
+++ b/gdb/testsuite/gdb.threads/step-over-trips-on-watchpoint.exp
@@ -115,11 +115,10 @@ proc do_test { displaced with_bp } {
gdb_test "thread 1" "Switching to .*"
gdb_test_no_output "set scheduler-locking off"
- # Thread 2 is still stopped at a breakpoint that needs to be
- # stepped over before proceeding thread 1. However, the
- # instruction that is under the breakpoint triggers a
- # watchpoint, which should trap and be reported to the
- # user.
+ # Thread 2 is still stopped at a breakpoint that needs
+ # to be stepped over. However, the instruction that
+ # is under the breakpoint triggers a watchpoint, which
+ # should trap and be reported to the user.
gdb_test "$command" "Hardware watchpoint.*: watch_me.*New value = 1.*"
}
}
--
1.9.3
^ permalink raw reply [flat|nested] 62+ messages in thread* [PATCH v3 03/17] remote.c/all-stop: Implement TARGET_WAITKIND_NO_RESUMED and TARGET_WNOHANG
2015-04-17 10:47 [PATCH v3 00/23] All-stop on top of non-stop Pedro Alves
` (4 preceding siblings ...)
2015-04-17 10:45 ` [PATCH v3 13/17] Fix step-over-{trips-on-watchpoint|lands-on-breakpoint}.exp race Pedro Alves
@ 2015-04-17 10:45 ` Pedro Alves
2015-04-17 10:45 ` [PATCH v3 04/17] Make thread_still_needs_step_over consider stepping_over_watchpoint too Pedro Alves
` (12 subsequent siblings)
18 siblings, 0 replies; 62+ messages in thread
From: Pedro Alves @ 2015-04-17 10:45 UTC (permalink / raw)
To: gdb-patches
Even though "target remote" supports target-async, the all-stop
target_wait implementation ignores TARGET_WNOHANG. If the core
happens to poll for events and we've already read the stop reply out
of the serial/socket, remote_wait_as hangs forever instead of
returning an indication that there are no events to process. This
can't happen currently, but later changes will trigger this.
gdb/ChangeLog:
2015-04-17 Pedro Alves <palves@redhat.com>
* remote.c (remote_wait_as): If not waiting for a stop reply,
return TARGET_WAITKIND_NO_RESUMED. If TARGET_WNOHANG is
requested, don't block waiting forever.
v3:
No changes.
---
gdb/remote.c | 13 ++++++++++++-
1 file changed, 12 insertions(+), 1 deletion(-)
diff --git a/gdb/remote.c b/gdb/remote.c
index 3b2325f..18492bf 100644
--- a/gdb/remote.c
+++ b/gdb/remote.c
@@ -5964,6 +5964,14 @@ remote_wait_as (ptid_t ptid, struct target_waitstatus *status, int options)
{
int ret;
int is_notif;
+ int forever = ((options & TARGET_WNOHANG) == 0
+ && wait_forever_enabled_p);
+
+ if (!rs->waiting_for_stop_reply)
+ {
+ status->kind = TARGET_WAITKIND_NO_RESUMED;
+ return minus_one_ptid;
+ }
if (!target_is_async_p ())
{
@@ -5982,7 +5990,7 @@ remote_wait_as (ptid_t ptid, struct target_waitstatus *status, int options)
However, before we do that we need to ensure that the caller
knows how to take the target into/out of async mode. */
ret = getpkt_or_notif_sane (&rs->buf, &rs->buf_size,
- wait_forever_enabled_p, &is_notif);
+ forever, &is_notif);
if (!target_is_async_p ())
signal (SIGINT, ofunc);
@@ -5991,6 +5999,9 @@ remote_wait_as (ptid_t ptid, struct target_waitstatus *status, int options)
not interesting. */
if (ret != -1 && is_notif)
return minus_one_ptid;
+
+ if (ret == -1 && (options & TARGET_WNOHANG) != 0)
+ return minus_one_ptid;
}
buf = rs->buf;
--
1.9.3
^ permalink raw reply [flat|nested] 62+ messages in thread* [PATCH v3 04/17] Make thread_still_needs_step_over consider stepping_over_watchpoint too
2015-04-17 10:47 [PATCH v3 00/23] All-stop on top of non-stop Pedro Alves
` (5 preceding siblings ...)
2015-04-17 10:45 ` [PATCH v3 03/17] remote.c/all-stop: Implement TARGET_WAITKIND_NO_RESUMED and TARGET_WNOHANG Pedro Alves
@ 2015-04-17 10:45 ` Pedro Alves
2015-04-17 10:45 ` [PATCH v3 02/17] Change adjust_pc_after_break's prototype Pedro Alves
` (11 subsequent siblings)
18 siblings, 0 replies; 62+ messages in thread
From: Pedro Alves @ 2015-04-17 10:45 UTC (permalink / raw)
To: gdb-patches
I noticed that even though keep_going knows to start a step over for a
watchpoint, thread_still_needs_step_over forgets it.
gdb/ChangeLog:
2015-04-17 Pedro Alves <palves@redhat.com>
* infrun.c (thread_still_needs_step_over): Rename to ...
(thread_still_needs_step_over_bp): ... this.
(enum step_over_what): New.
(thread_still_needs_step_over): Reimplement.
v3:
Add comments to enum values. Changed commit log a bit.
---
gdb/infrun.c | 43 +++++++++++++++++++++++++++++++++++++++----
1 file changed, 39 insertions(+), 4 deletions(-)
diff --git a/gdb/infrun.c b/gdb/infrun.c
index a45877e..534ecef 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -1223,6 +1223,19 @@ follow_exec (ptid_t ptid, char *execd_pathname)
matically get reset there in the new process.). */
}
+/* Bit flags indicating what the thread needs to step over. */
+
+enum step_over_what
+ {
+ /* Step over a breakpoint. */
+ STEP_OVER_BREAKPOINT = 1,
+
+ /* Step past a non-continuable watchpoint, in order to let the
+ instruction execute so we can evaluate the watchpoint
+ expression. */
+ STEP_OVER_WATCHPOINT = 2
+ };
+
/* Info about an instruction that is being stepped over. */
struct step_over_info
@@ -2519,7 +2532,7 @@ clear_proceed_status (int step)
meanwhile, we can skip the whole step-over dance. */
static int
-thread_still_needs_step_over (struct thread_info *tp)
+thread_still_needs_step_over_bp (struct thread_info *tp)
{
if (tp->stepping_over_breakpoint)
{
@@ -2536,6 +2549,26 @@ thread_still_needs_step_over (struct thread_info *tp)
return 0;
}
+/* Check whether thread TP still needs to start a step-over in order
+ to make progress when resumed. Returns an bitwise or of enum
+ step_over_what bits, indicating what needs to be stepped over. */
+
+static int
+thread_still_needs_step_over (struct thread_info *tp)
+{
+ struct inferior *inf = find_inferior_ptid (tp->ptid);
+ int what = 0;
+
+ if (thread_still_needs_step_over_bp (tp))
+ what |= STEP_OVER_BREAKPOINT;
+
+ if (tp->stepping_over_watchpoint
+ && !target_have_steppable_watchpoint)
+ what |= STEP_OVER_WATCHPOINT;
+
+ return what;
+}
+
/* Returns true if scheduler locking applies. STEP indicates whether
we're about to do a step/next-like command to a thread. */
@@ -6270,6 +6303,7 @@ keep_going (struct execution_control_state *ecs)
struct regcache *regcache = get_current_regcache ();
int remove_bp;
int remove_wps;
+ enum step_over_what step_what;
/* Either the trap was not expected, but we are continuing
anyway (if we got a signal, the user asked it be passed to
@@ -6290,10 +6324,11 @@ keep_going (struct execution_control_state *ecs)
instruction, and then re-insert the breakpoint when that step
is finished. */
+ step_what = thread_still_needs_step_over (ecs->event_thread);
+
remove_bp = (ecs->hit_singlestep_breakpoint
- || thread_still_needs_step_over (ecs->event_thread));
- remove_wps = (ecs->event_thread->stepping_over_watchpoint
- && !target_have_steppable_watchpoint);
+ || (step_what & STEP_OVER_BREAKPOINT));
+ remove_wps = (step_what & STEP_OVER_WATCHPOINT);
/* We can't use displaced stepping if we need to step past a
watchpoint. The instruction copied to the scratch pad would
--
1.9.3
^ permalink raw reply [flat|nested] 62+ messages in thread* [PATCH v3 02/17] Change adjust_pc_after_break's prototype
2015-04-17 10:47 [PATCH v3 00/23] All-stop on top of non-stop Pedro Alves
` (6 preceding siblings ...)
2015-04-17 10:45 ` [PATCH v3 04/17] Make thread_still_needs_step_over consider stepping_over_watchpoint too Pedro Alves
@ 2015-04-17 10:45 ` Pedro Alves
2015-04-17 10:45 ` [PATCH v3 06/17] Use keep_going in proceed and start_step_over too Pedro Alves
` (10 subsequent siblings)
18 siblings, 0 replies; 62+ messages in thread
From: Pedro Alves @ 2015-04-17 10:45 UTC (permalink / raw)
To: gdb-patches
Prepare to use it in contexts without an ecs handy. Follow up patches
will make use of this.
gdb/ChangeLog:
2015-04-17 Pedro Alves <pedro@codesourcery.com>
* infrun.c (adjust_pc_after_break): Now takes thread_info and
waitstatus pointers instead of an ecs. Adjust.
(handle_inferior_event): Adjust caller.
v3:
No changes.
---
gdb/infrun.c | 24 +++++++++++++++---------
1 file changed, 15 insertions(+), 9 deletions(-)
diff --git a/gdb/infrun.c b/gdb/infrun.c
index 9c1fdc5..a45877e 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -3446,8 +3446,14 @@ context_switch (ptid_t ptid)
switch_to_thread (ptid);
}
+/* If the target can't tell whether we've hit breakpoints
+ (target_supports_stopped_by_sw_breakpoint), and we got a SIGTRAP,
+ check whether that could have been caused by a breakpoint. If so,
+ adjust the PC, per gdbarch_decr_pc_after_break. */
+
static void
-adjust_pc_after_break (struct execution_control_state *ecs)
+adjust_pc_after_break (struct thread_info *thread,
+ struct target_waitstatus *ws)
{
struct regcache *regcache;
struct gdbarch *gdbarch;
@@ -3475,10 +3481,10 @@ adjust_pc_after_break (struct execution_control_state *ecs)
target with both of these set in GDB history, and it seems unlikely to be
correct, so gdbarch_have_nonsteppable_watchpoint is not checked here. */
- if (ecs->ws.kind != TARGET_WAITKIND_STOPPED)
+ if (ws->kind != TARGET_WAITKIND_STOPPED)
return;
- if (ecs->ws.value.sig != GDB_SIGNAL_TRAP)
+ if (ws->value.sig != GDB_SIGNAL_TRAP)
return;
/* In reverse execution, when a breakpoint is hit, the instruction
@@ -3524,7 +3530,7 @@ adjust_pc_after_break (struct execution_control_state *ecs)
/* If this target does not decrement the PC after breakpoints, then
we have nothing to do. */
- regcache = get_thread_regcache (ecs->ptid);
+ regcache = get_thread_regcache (thread->ptid);
gdbarch = get_regcache_arch (regcache);
decr_pc = gdbarch_decr_pc_after_break (gdbarch);
@@ -3578,10 +3584,10 @@ adjust_pc_after_break (struct execution_control_state *ecs)
software breakpoint. In this case (prev_pc == breakpoint_pc),
we also need to back up to the breakpoint address. */
- if (thread_has_single_step_breakpoints_set (ecs->event_thread)
- || !currently_stepping (ecs->event_thread)
- || (ecs->event_thread->stepped_breakpoint
- && ecs->event_thread->prev_pc == breakpoint_pc))
+ if (thread_has_single_step_breakpoints_set (thread)
+ || !currently_stepping (thread)
+ || (thread->stepped_breakpoint
+ && thread->prev_pc == breakpoint_pc))
regcache_write_pc (regcache, breakpoint_pc);
do_cleanups (old_cleanups);
@@ -3762,7 +3768,7 @@ handle_inferior_event (struct execution_control_state *ecs)
}
/* Dependent on valid ECS->EVENT_THREAD. */
- adjust_pc_after_break (ecs);
+ adjust_pc_after_break (ecs->event_thread, &ecs->ws);
/* Dependent on the current PC value modified by adjust_pc_after_break. */
reinit_frame_cache ();
--
1.9.3
^ permalink raw reply [flat|nested] 62+ messages in thread* [PATCH v3 06/17] Use keep_going in proceed and start_step_over too
2015-04-17 10:47 [PATCH v3 00/23] All-stop on top of non-stop Pedro Alves
` (7 preceding siblings ...)
2015-04-17 10:45 ` [PATCH v3 02/17] Change adjust_pc_after_break's prototype Pedro Alves
@ 2015-04-17 10:45 ` Pedro Alves
2015-04-22 5:09 ` Doug Evans
2015-04-17 10:47 ` [PATCH v3 17/17] native Linux: enable always non-stop by default Pedro Alves
` (9 subsequent siblings)
18 siblings, 1 reply; 62+ messages in thread
From: Pedro Alves @ 2015-04-17 10:45 UTC (permalink / raw)
To: gdb-patches
The main motivation of this patch is sharing more code between the
proceed (starting the inferior for the first time) and keep_going
(restarting the inferior after handling an event) paths and using the
step_over_chain queue now embedded in the thread_info object for
pending in-line step-overs too (instead of just for displaced
stepping).
So this commit:
- splits out a new keep_going_pass function out of keep_going that is
just like keep_going except for the bits that clear the signal to
pass if the signal is set to "handle nopass".
- makes proceed use keep_going too.
- Makes start_step_over use keep_going_pass instead of lower level
displaced stepping things.
One user visible change: if inserting breakpoints while trying to
proceed fails, we now get:
(gdb) si
Warning:
Could not insert hardware watchpoint 7.
Could not insert hardware breakpoints:
You may have requested too many hardware breakpoints/watchpoints.
Command aborted.
(gdb)
while before we only saw warnings with no indication that the command
was cancelled:
(gdb) si
Warning:
Could not insert hardware watchpoint 7.
Could not insert hardware breakpoints:
You may have requested too many hardware breakpoints/watchpoints.
(gdb)
Tested on x86_64-linux-gnu, ppc64-linux-gnu and s390-linux-gnu.
gdb/ChangeLog:
2015-04-17 Pedro Alves <palves@redhat.com>
* gdbthread.h (struct thread_info) <prev_pc>: Extend comment.
* infrun.c (struct execution_control_state): Move higher up in the
file.
(reset_ecs): New function.
(start_step_over): Now returns int. Rewrite to use
keep_going_pass instead of manually starting a displaced step.
(resume): Don't call set_running here. If displaced stepping
can't start now, clear trap_expected.
(find_thread_needs_step_over): Delete function.
(proceed): Set up finish_thread_state_cleanup. Call set_running.
If the current thread needs a step over, push it in the step-over
chain. Don't set insert breakpoints nor call resume directly
here. Instead rewrite to use start_step_over and keep_going_pass.
(finish_step_over): New function.
(handle_signal_stop): Call finish_step_over instead of
start_step_over.
(switch_back_to_stepped_thread): If the event thread needs another
step-over do that first. Use start_step_over.
(keep_going_pass): New function, factored out from ...
(keep_going): ... here.
(_initialize_infrun): Comment moved here.
* thread.c (set_running_thread): New function.
(set_running, finish_thread_state): Use set_running_thread.
v3:
- Adjusted to step_over_chain changes. Fixed gdb internal error
caught by signal-sigtrap.exp on ARM (software single-step target).
v2:
- Testing on x86_64 with software single-step revealed v1 missed
setting prev_pc on proceed, for switch_back_to_stepped_thread.
Rewrite tail end of proceed to fix that. This ends up simplifying
later patches in the series.
---
gdb/gdbthread.h | 6 +-
gdb/infrun.c | 579 ++++++++++++++++++++++++++++----------------------------
gdb/thread.c | 52 +++--
3 files changed, 324 insertions(+), 313 deletions(-)
diff --git a/gdb/gdbthread.h b/gdb/gdbthread.h
index 2c871a2..7052ee1 100644
--- a/gdb/gdbthread.h
+++ b/gdb/gdbthread.h
@@ -208,8 +208,10 @@ struct thread_info
/* Internal stepping state. */
- /* Record the pc of the thread the last time it stopped. This is
- maintained by proceed and keep_going, and used in
+ /* Record the pc of the thread the last time it was resumed. (It
+ can't be done on stop as the PC may change since the last stop,
+ e.g., "return" command, or "p $pc = 0xf000"). This is maintained
+ by proceed and keep_going, and among other things, it's used in
adjust_pc_after_break to distinguish a hardware single-step
SIGTRAP from a breakpoint SIGTRAP. */
CORE_ADDR prev_pc;
diff --git a/gdb/infrun.c b/gdb/infrun.c
index f325a53..beb9ea2 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -1844,30 +1844,61 @@ displaced_step_fixup (ptid_t event_ptid, enum gdb_signal signal)
displaced->step_ptid = null_ptid;
}
-/* Are there any pending step-over requests? If so, run one now. */
+/* Data to be passed around while handling an event. This data is
+ discarded between events. */
+struct execution_control_state
+{
+ ptid_t ptid;
+ /* The thread that got the event, if this was a thread event; NULL
+ otherwise. */
+ struct thread_info *event_thread;
+
+ struct target_waitstatus ws;
+ int stop_func_filled_in;
+ CORE_ADDR stop_func_start;
+ CORE_ADDR stop_func_end;
+ const char *stop_func_name;
+ int wait_some_more;
+
+ /* True if the event thread hit the single-step breakpoint of
+ another thread. Thus the event doesn't cause a stop, the thread
+ needs to be single-stepped past the single-step breakpoint before
+ we can switch back to the original stepping thread. */
+ int hit_singlestep_breakpoint;
+};
+
+/* Clear ECS and set it to point at TP. */
static void
+reset_ecs (struct execution_control_state *ecs, struct thread_info *tp)
+{
+ memset (ecs, 0, sizeof (*ecs));
+ ecs->event_thread = tp;
+ ecs->ptid = tp->ptid;
+}
+
+static void keep_going_pass (struct execution_control_state *ecs);
+static void prepare_to_wait (struct execution_control_state *ecs);
+static int thread_still_needs_step_over (struct thread_info *tp);
+
+/* Are there any pending step-over requests? If so, run one now and
+ return true. Otherwise, return false. */
+
+static int
start_step_over (void)
{
struct thread_info *tp, *next;
for (tp = step_over_queue_head; tp != NULL; tp = next)
{
- ptid_t ptid;
- struct displaced_step_inferior_state *displaced;
- struct regcache *regcache;
- struct gdbarch *gdbarch;
- CORE_ADDR actual_pc;
- struct address_space *aspace;
- struct inferior *inf = find_inferior_ptid (tp->ptid);
+ struct execution_control_state ecss;
+ struct execution_control_state *ecs = &ecss;
next = thread_step_over_chain_next (tp);
- displaced = get_displaced_stepping_state (inf->pid);
-
/* If this inferior already has a displaced step in process,
don't start a new one. */
- if (!ptid_equal (displaced->step_ptid, null_ptid))
+ if (displaced_step_in_progress (ptid_get_pid (tp->ptid)))
continue;
thread_step_over_chain_remove (tp);
@@ -1879,74 +1910,57 @@ start_step_over (void)
"infrun: step-over queue now empty\n");
}
- ptid = tp->ptid;
- context_switch (ptid);
-
- regcache = get_thread_regcache (ptid);
- actual_pc = regcache_read_pc (regcache);
- aspace = get_regcache_aspace (regcache);
- gdbarch = get_regcache_arch (regcache);
-
- if (breakpoint_here_p (aspace, actual_pc))
+ if (tp->control.trap_expected || tp->executing)
{
- if (debug_displaced)
- fprintf_unfiltered (gdb_stdlog,
- "displaced: stepping queued %s now\n",
- target_pid_to_str (ptid));
-
- displaced_step_prepare (ptid);
-
- if (debug_displaced)
- {
- CORE_ADDR actual_pc = regcache_read_pc (regcache);
- gdb_byte buf[4];
-
- fprintf_unfiltered (gdb_stdlog, "displaced: run %s: ",
- paddress (gdbarch, actual_pc));
- read_memory (actual_pc, buf, sizeof (buf));
- displaced_step_dump_bytes (gdb_stdlog, buf, sizeof (buf));
- }
-
- if (gdbarch_displaced_step_hw_singlestep (gdbarch,
- displaced->step_closure))
- target_resume (ptid, 1, GDB_SIGNAL_0);
- else
- target_resume (ptid, 0, GDB_SIGNAL_0);
-
- /* Done, we're stepping a thread. */
- break;
+ internal_error (__FILE__, __LINE__,
+ "[%s] has inconsistent state: "
+ "trap_expected=%d, executing=%d\n",
+ target_pid_to_str (tp->ptid),
+ tp->control.trap_expected,
+ tp->executing);
}
- else
- {
- int step;
- struct thread_info *tp = inferior_thread ();
-
- /* The breakpoint we were sitting under has since been
- removed. */
- tp->control.trap_expected = 0;
- /* Go back to what we were trying to do. */
- step = currently_stepping (tp);
-
- if (step)
- step = maybe_software_singlestep (gdbarch, actual_pc);
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: resuming [%s] for step-over\n",
+ target_pid_to_str (tp->ptid));
+
+ /* keep_going_pass skips the step-over if the breakpoint is no
+ longer inserted. In all-stop, we want to keep looking for a
+ that needs a step-over instead of resuming TP, because we
+ wouldn't be able to resume anything else until the target
+ stops again. In non-stop, the resume always resumes only TP,
+ so it's OK to let the thread resume freely. */
+ if (!non_stop && !thread_still_needs_step_over (tp))
+ continue;
- if (debug_displaced)
- fprintf_unfiltered (gdb_stdlog,
- "displaced: breakpoint is gone: %s, step(%d)\n",
- target_pid_to_str (tp->ptid), step);
+ switch_to_thread (tp->ptid);
+ reset_ecs (ecs, tp);
+ keep_going_pass (ecs);
- target_resume (ptid, step, GDB_SIGNAL_0);
- tp->suspend.stop_signal = GDB_SIGNAL_0;
+ if (!ecs->wait_some_more)
+ error (_("Command aborted."));
- /* This request was discarded. See if there's any other
- thread waiting for its turn. */
+ if (!non_stop)
+ {
+ /* On all-stop, shouldn't have resumed unless we needed a
+ step over. */
+ gdb_assert (tp->control.trap_expected
+ || tp->step_after_step_resume_breakpoint);
+
+ /* With remote targets (at least), in all-stop, we can't
+ issue any further remote commands until the program stops
+ again. */
+ return 1;
}
- /* A new displaced stepping sequence started. Maybe we can
- start a displaced step on a thread of other process.
- Continue looking. */
+ /* Either the thread no longer needed a step-over, or a new
+ displaced stepping sequence started. Even in the latter
+ case, continue looking. Maybe we can also start another
+ displaced step on a thread of other process. */
}
+
+ return 0;
}
/* Update global variables holding ptids to hold NEW_PTID if they were
@@ -2283,16 +2297,11 @@ resume (enum gdb_signal sig)
if (!displaced_step_prepare (inferior_ptid))
{
- /* Got placed in displaced stepping queue. Will be resumed
- later when all the currently queued displaced stepping
- requests finish. The thread is not executing at this
- point, and the call to set_executing will be made later.
- But we need to call set_running here, since from the
- user/frontend's point of view, threads were set running.
- Unless we're calling an inferior function, as in that
- case we pretend the inferior doesn't run at all. */
- if (!tp->control.in_infcall)
- set_running (user_visible_resume_ptid (user_step), 1);
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "Got placed in displaced stepping queue\n");
+
+ tp->control.trap_expected = 0;
discard_cleanups (old_cleanups);
return;
}
@@ -2367,14 +2376,6 @@ resume (enum gdb_signal sig)
by applying increasingly restricting conditions. */
resume_ptid = user_visible_resume_ptid (user_step);
- /* Even if RESUME_PTID is a wildcard, and we end up resuming less
- (e.g., we might need to step over a breakpoint), from the
- user/frontend's point of view, all threads in RESUME_PTID are now
- running. Unless we're calling an inferior function, as in that
- case pretend we inferior doesn't run at all. */
- if (!tp->control.in_infcall)
- set_running (resume_ptid, 1);
-
/* Maybe resume a single thread after all. */
if ((step || thread_has_single_step_breakpoints_set (tp))
&& tp->control.trap_expected)
@@ -2589,48 +2590,6 @@ schedlock_applies (struct thread_info *tp)
&& tp->control.stepping_command));
}
-/* Look a thread other than EXCEPT that has previously reported a
- breakpoint event, and thus needs a step-over in order to make
- progress. Returns NULL is none is found. */
-
-static struct thread_info *
-find_thread_needs_step_over (struct thread_info *except)
-{
- struct thread_info *tp, *current;
-
- /* With non-stop mode on, threads are always handled individually. */
- gdb_assert (! non_stop);
-
- current = inferior_thread ();
-
- /* If scheduler locking applies, we can avoid iterating over all
- threads. */
- if (schedlock_applies (except))
- {
- if (except != current
- && thread_still_needs_step_over (current))
- return current;
-
- return NULL;
- }
-
- ALL_NON_EXITED_THREADS (tp)
- {
- /* Ignore the EXCEPT thread. */
- if (tp == except)
- continue;
- /* Ignore threads of processes we're not resuming. */
- if (!sched_multi
- && ptid_get_pid (tp->ptid) != ptid_get_pid (inferior_ptid))
- continue;
-
- if (thread_still_needs_step_over (tp))
- return tp;
- }
-
- return NULL;
-}
-
/* Basic routine for continuing the program in various fashions.
ADDR is the address to resume at, or -1 for resume where stopped.
@@ -2651,6 +2610,11 @@ proceed (CORE_ADDR addr, enum gdb_signal siggnal)
struct thread_info *tp;
CORE_ADDR pc;
struct address_space *aspace;
+ ptid_t resume_ptid;
+ struct execution_control_state ecss;
+ struct execution_control_state *ecs = &ecss;
+ struct cleanup *old_chain;
+ int started;
/* If we're stopped at a fork/vfork, follow the branch set by the
"set follow-fork-mode" command; otherwise, we'll just proceed
@@ -2713,7 +2677,23 @@ proceed (CORE_ADDR addr, enum gdb_signal siggnal)
(next/step/etc.), we'll want to print stop event output to the MI
console channel (the stepped-to line, etc.), as if the user
entered the execution command on a real GDB console. */
- inferior_thread ()->control.command_interp = command_interp ();
+ tp->control.command_interp = command_interp ();
+
+ resume_ptid = user_visible_resume_ptid (tp->control.stepping_command);
+
+ /* If an exception is thrown from this point on, make sure to
+ propagate GDB's knowledge of the executing state to the
+ frontend/user running state. */
+ old_chain = make_cleanup (finish_thread_state_cleanup, &resume_ptid);
+
+ /* Even if RESUME_PTID is a wildcard, and we end up resuming fewer
+ threads (e.g., we might need to set threads stepping over
+ breakpoints first), from the user/frontend's point of view, all
+ threads in RESUME_PTID are now running. Unless we're calling an
+ inferior function, as in that case we pretend the inferior
+ doesn't run at all. */
+ if (!tp->control.in_infcall)
+ set_running (resume_ptid, 1);
if (debug_infrun)
fprintf_unfiltered (gdb_stdlog,
@@ -2721,91 +2701,92 @@ proceed (CORE_ADDR addr, enum gdb_signal siggnal)
paddress (gdbarch, addr),
gdb_signal_to_symbol_string (siggnal));
- if (non_stop)
- /* In non-stop, each thread is handled individually. The context
- must already be set to the right thread here. */
- ;
- else
+ annotate_starting ();
+
+ /* Make sure that output from GDB appears before output from the
+ inferior. */
+ gdb_flush (gdb_stdout);
+
+ /* In a multi-threaded task we may select another thread and
+ then continue or step.
+
+ But if a thread that we're resuming had stopped at a breakpoint,
+ it will immediately cause another breakpoint stop without any
+ execution (i.e. it will report a breakpoint hit incorrectly). So
+ we must step over it first.
+
+ Look for threads other than the current (TP) that reported a
+ breakpoint hit and haven't been resumed yet since. */
+
+ /* If scheduler locking applies, we can avoid iterating over all
+ threads. */
+ if (!non_stop && !schedlock_applies (tp))
{
- struct thread_info *step_over;
+ struct thread_info *current = tp;
+
+ ALL_NON_EXITED_THREADS (tp)
+ {
+ /* Ignore the current thread here. It's handled
+ afterwards. */
+ if (tp == current)
+ continue;
- /* In a multi-threaded task we may select another thread and
- then continue or step.
+ /* Ignore threads of processes we're not resuming. */
+ if (!ptid_match (tp->ptid, resume_ptid))
+ continue;
- But if the old thread was stopped at a breakpoint, it will
- immediately cause another breakpoint stop without any
- execution (i.e. it will report a breakpoint hit incorrectly).
- So we must step over it first.
+ if (!thread_still_needs_step_over (tp))
+ continue;
+
+ gdb_assert (!thread_is_in_step_over_chain (tp));
- Look for a thread other than the current (TP) that reported a
- breakpoint hit and hasn't been resumed yet since. */
- step_over = find_thread_needs_step_over (tp);
- if (step_over != NULL)
- {
if (debug_infrun)
fprintf_unfiltered (gdb_stdlog,
"infrun: need to step-over [%s] first\n",
- target_pid_to_str (step_over->ptid));
+ target_pid_to_str (tp->ptid));
- /* Store the prev_pc for the stepping thread too, needed by
- switch_back_to_stepped_thread. */
- tp->prev_pc = regcache_read_pc (get_current_regcache ());
- switch_to_thread (step_over->ptid);
- tp = step_over;
+ thread_step_over_chain_enqueue (tp);
}
- }
-
- /* If we need to step over a breakpoint, and we're not using
- displaced stepping to do so, insert all breakpoints (watchpoints,
- etc.) but the one we're stepping over, step one instruction, and
- then re-insert the breakpoint when that step is finished. */
- if (tp->stepping_over_breakpoint && !use_displaced_stepping (gdbarch))
- {
- struct regcache *regcache = get_current_regcache ();
- set_step_over_info (get_regcache_aspace (regcache),
- regcache_read_pc (regcache), 0);
+ tp = current;
}
- else
- clear_step_over_info ();
- insert_breakpoints ();
+ /* Enqueue the current thread last, so that we move all other
+ threads over their breakpoints first. */
+ if (tp->stepping_over_breakpoint)
+ thread_step_over_chain_enqueue (tp);
- tp->control.trap_expected = tp->stepping_over_breakpoint;
+ /* If the thread isn't started, we'll still need to set its prev_pc,
+ so that switch_back_to_stepped_thread knows the thread hasn't
+ advanced. Must do this before resuming any thread, as in
+ all-stop/remote, once we resume we can't send any other packet
+ until the target stops again. */
+ tp->prev_pc = regcache_read_pc (regcache);
- annotate_starting ();
+ started = start_step_over ();
- /* Make sure that output from GDB appears before output from the
- inferior. */
- gdb_flush (gdb_stdout);
+ if (step_over_info_valid_p ())
+ {
+ /* Either this thread started a new in-line step over, or some
+ other thread was already doing one. In either case, don't
+ resume anything else until the step-over is finished. */
+ }
+ else if (started && !non_stop)
+ {
+ /* A new displaced stepping sequence was started. In all-stop,
+ we can't talk to the target anymore until it next stops. */
+ }
+ else if (!tp->executing && !thread_is_in_step_over_chain (tp))
+ {
+ /* The thread wasn't started, and isn't queued, run it now. */
+ reset_ecs (ecs, tp);
+ switch_to_thread (tp->ptid);
+ keep_going_pass (ecs);
+ if (!ecs->wait_some_more)
+ error ("Command aborted.");
+ }
- /* Refresh prev_pc value just prior to resuming. This used to be
- done in stop_waiting, however, setting prev_pc there did not handle
- scenarios such as inferior function calls or returning from
- a function via the return command. In those cases, the prev_pc
- value was not set properly for subsequent commands. The prev_pc value
- is used to initialize the starting line number in the ecs. With an
- invalid value, the gdb next command ends up stopping at the position
- represented by the next line table entry past our start position.
- On platforms that generate one line table entry per line, this
- is not a problem. However, on the ia64, the compiler generates
- extraneous line table entries that do not increase the line number.
- When we issue the gdb next command on the ia64 after an inferior call
- or a return command, we often end up a few instructions forward, still
- within the original line we started.
-
- An attempt was made to refresh the prev_pc at the same time the
- execution_control_state is initialized (for instance, just before
- waiting for an inferior event). But this approach did not work
- because of platforms that use ptrace, where the pc register cannot
- be read unless the inferior is stopped. At that point, we are not
- guaranteed the inferior is stopped and so the regcache_read_pc() call
- can fail. Setting the prev_pc value here ensures the value is updated
- correctly when the inferior is stopped. */
- tp->prev_pc = regcache_read_pc (get_current_regcache ());
-
- /* Resume inferior. */
- resume (tp->suspend.stop_signal);
+ discard_cleanups (old_chain);
/* Wait for it to stop (if not standalone)
and in any case decode why it stopped, and act accordingly. */
@@ -2873,28 +2854,6 @@ init_wait_for_inferior (void)
}
\f
-/* Data to be passed around while handling an event. This data is
- discarded between events. */
-struct execution_control_state
-{
- ptid_t ptid;
- /* The thread that got the event, if this was a thread event; NULL
- otherwise. */
- struct thread_info *event_thread;
-
- struct target_waitstatus ws;
- int stop_func_filled_in;
- CORE_ADDR stop_func_start;
- CORE_ADDR stop_func_end;
- const char *stop_func_name;
- int wait_some_more;
-
- /* True if the event thread hit the single-step breakpoint of
- another thread. Thus the event doesn't cause a stop, the thread
- needs to be single-stepped past the single-step breakpoint before
- we can switch back to the original stepping thread. */
- int hit_singlestep_breakpoint;
-};
static void handle_inferior_event (struct execution_control_state *ecs);
@@ -2908,7 +2867,6 @@ static void check_exception_resume (struct execution_control_state *,
static void end_stepping_range (struct execution_control_state *ecs);
static void stop_waiting (struct execution_control_state *ecs);
-static void prepare_to_wait (struct execution_control_state *ecs);
static void keep_going (struct execution_control_state *ecs);
static void process_event_stop_test (struct execution_control_state *ecs);
static int switch_back_to_stepped_thread (struct execution_control_state *ecs);
@@ -4261,6 +4219,34 @@ Cannot fill $_exitsignal with the correct signal number.\n"));
}
}
+/* Called when we get an event that may finish an in-line or
+ out-of-line (displaced stepping) step-over started previously. */
+
+static void
+finish_step_over (struct execution_control_state *ecs)
+{
+ displaced_step_fixup (ecs->ptid,
+ ecs->event_thread->suspend.stop_signal);
+
+ if (step_over_info_valid_p ())
+ {
+ /* If we're stepping over a breakpoint with all threads locked,
+ then only the thread that was stepped should be reporting
+ back an event. */
+ gdb_assert (ecs->event_thread->control.trap_expected);
+
+ if (ecs->event_thread->suspend.stop_signal == GDB_SIGNAL_TRAP)
+ clear_step_over_info ();
+ }
+
+ if (!non_stop)
+ return;
+
+ /* Start a new step-over in another thread if there's one that
+ needs it. */
+ start_step_over ();
+}
+
/* Come here when the program has stopped with a signal. */
static void
@@ -4277,9 +4263,7 @@ handle_signal_stop (struct execution_control_state *ecs)
/* Do we need to clean up the state of a thread that has
completed a displaced single-step? (Doing so usually affects
the PC, so do it here, before we set stop_pc.) */
- displaced_step_fixup (ecs->ptid,
- ecs->event_thread->suspend.stop_signal);
- start_step_over ();
+ finish_step_over (ecs);
/* If we either finished a single-step or hit a breakpoint, but
the user wanted this thread to be stopped, pretend we got a
@@ -5629,7 +5613,6 @@ switch_back_to_stepped_thread (struct execution_control_state *ecs)
{
struct thread_info *tp;
struct thread_info *stepping_thread;
- struct thread_info *step_over;
/* If any thread is blocked on some internal breakpoint, and we
simply need to step over that breakpoint to get it going
@@ -5672,14 +5655,20 @@ switch_back_to_stepped_thread (struct execution_control_state *ecs)
return 1;
}
- /* Otherwise, we no longer expect a trap in the current thread.
- Clear the trap_expected flag before switching back -- this is
- what keep_going does as well, if we call it. */
- ecs->event_thread->control.trap_expected = 0;
-
- /* Likewise, clear the signal if it should not be passed. */
- if (!signal_program[ecs->event_thread->suspend.stop_signal])
- ecs->event_thread->suspend.stop_signal = GDB_SIGNAL_0;
+ /* If this thread needs yet another step-over (e.g., stepping
+ through a delay slot), do it first before moving on to
+ another thread. */
+ if (thread_still_needs_step_over (ecs->event_thread))
+ {
+ if (debug_infrun)
+ {
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: thread [%s] still needs step-over\n",
+ target_pid_to_str (ecs->event_thread->ptid));
+ }
+ keep_going (ecs);
+ return 1;
+ }
/* If scheduler locking applies even if not stepping, there's no
need to walk over threads. Above we've checked whether the
@@ -5689,12 +5678,26 @@ switch_back_to_stepped_thread (struct execution_control_state *ecs)
if (schedlock_applies (ecs->event_thread))
return 0;
- /* Look for the stepping/nexting thread, and check if any other
- thread other than the stepping thread needs to start a
- step-over. Do all step-overs before actually proceeding with
+ /* Otherwise, we no longer expect a trap in the current thread.
+ Clear the trap_expected flag before switching back -- this is
+ what keep_going does as well, if we call it. */
+ ecs->event_thread->control.trap_expected = 0;
+
+ /* Likewise, clear the signal if it should not be passed. */
+ if (!signal_program[ecs->event_thread->suspend.stop_signal])
+ ecs->event_thread->suspend.stop_signal = GDB_SIGNAL_0;
+
+ /* Do all pending step-overs before actually proceeding with
step/next/etc. */
+ if (start_step_over ())
+ {
+ prepare_to_wait (ecs);
+ return 1;
+ }
+
+ /* Look for the stepping/nexting thread. */
stepping_thread = NULL;
- step_over = NULL;
+
ALL_NON_EXITED_THREADS (tp)
{
/* Ignore threads of processes we're not resuming. */
@@ -5726,37 +5729,6 @@ switch_back_to_stepped_thread (struct execution_control_state *ecs)
stepping_thread = tp;
}
- else if (thread_still_needs_step_over (tp))
- {
- step_over = tp;
-
- /* At the top we've returned early if the event thread
- is stepping. If some other thread not the event
- thread is stepping, then scheduler locking can't be
- in effect, and we can resume this thread. No need to
- keep looking for the stepping thread then. */
- break;
- }
- }
-
- if (step_over != NULL)
- {
- tp = step_over;
- if (debug_infrun)
- {
- fprintf_unfiltered (gdb_stdlog,
- "infrun: need to step-over [%s]\n",
- target_pid_to_str (tp->ptid));
- }
-
- /* Only the stepping thread should have this set. */
- gdb_assert (tp->control.step_range_end == 0);
-
- ecs->ptid = tp->ptid;
- ecs->event_thread = tp;
- switch_to_thread (ecs->ptid);
- keep_going (ecs);
- return 1;
}
if (stepping_thread != NULL)
@@ -5855,7 +5827,7 @@ switch_back_to_stepped_thread (struct execution_control_state *ecs)
fprintf_unfiltered (gdb_stdlog,
"infrun: expected thread still "
"hasn't advanced\n");
- keep_going (ecs);
+ keep_going_pass (ecs);
}
return 1;
@@ -6270,24 +6242,32 @@ stop_waiting (struct execution_control_state *ecs)
ecs->wait_some_more = 0;
}
-/* Called when we should continue running the inferior, because the
- current event doesn't cause a user visible stop. This does the
- resuming part; waiting for the next event is done elsewhere. */
+/* Like keep_going, but passes the signal to the inferior, even if the
+ signal is set to nopass. */
static void
-keep_going (struct execution_control_state *ecs)
+keep_going_pass (struct execution_control_state *ecs)
{
/* Make sure normal_stop is called if we get a QUIT handled before
reaching resume. */
struct cleanup *old_cleanups = make_cleanup (resume_cleanups, 0);
+ gdb_assert (ptid_equal (ecs->event_thread->ptid, inferior_ptid));
+
/* Save the pc before execution, to compare with pc after stop. */
ecs->event_thread->prev_pc
= regcache_read_pc (get_thread_regcache (ecs->ptid));
- if (ecs->event_thread->control.trap_expected
- && ecs->event_thread->suspend.stop_signal != GDB_SIGNAL_TRAP)
+ if (ecs->event_thread->control.trap_expected)
{
+ struct thread_info *tp = ecs->event_thread;
+
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: %s has trap_expected set, "
+ "resuming to collect trap\n",
+ target_pid_to_str (tp->ptid));
+
/* We haven't yet gotten our trap, and either: intercepted a
non-signal event (e.g., a fork); or took a signal which we
are supposed to pass through to the inferior. Simply
@@ -6358,20 +6338,6 @@ keep_going (struct execution_control_state *ecs)
ecs->event_thread->control.trap_expected = (remove_bp || remove_wps);
- /* Do not deliver GDB_SIGNAL_TRAP (except when the user
- explicitly specifies that such a signal should be delivered
- to the target program). Typically, that would occur when a
- user is debugging a target monitor on a simulator: the target
- monitor sets a breakpoint; the simulator encounters this
- breakpoint and halts the simulation handing control to GDB;
- GDB, noting that the stop address doesn't map to any known
- breakpoint, returns control back to the simulator; the
- simulator then delivers the hardware equivalent of a
- GDB_SIGNAL_TRAP to the program being debugged. */
- if (ecs->event_thread->suspend.stop_signal == GDB_SIGNAL_TRAP
- && !signal_program[ecs->event_thread->suspend.stop_signal])
- ecs->event_thread->suspend.stop_signal = GDB_SIGNAL_0;
-
discard_cleanups (old_cleanups);
resume (ecs->event_thread->suspend.stop_signal);
}
@@ -6379,6 +6345,22 @@ keep_going (struct execution_control_state *ecs)
prepare_to_wait (ecs);
}
+/* Called when we should continue running the inferior, because the
+ current event doesn't cause a user visible stop. This does the
+ resuming part; waiting for the next event is done elsewhere. */
+
+static void
+keep_going (struct execution_control_state *ecs)
+{
+ if (ecs->event_thread->control.trap_expected
+ && ecs->event_thread->suspend.stop_signal == GDB_SIGNAL_TRAP)
+ ecs->event_thread->control.trap_expected = 0;
+
+ if (!signal_program[ecs->event_thread->suspend.stop_signal])
+ ecs->event_thread->suspend.stop_signal = GDB_SIGNAL_0;
+ keep_going_pass (ecs);
+}
+
/* This function normally comes after a resume, before
handle_inferior_event exits. It takes care of any last bits of
housekeeping, and sets the all-important wait_some_more flag. */
@@ -7806,8 +7788,19 @@ leave it stopped or free to run as needed."),
signal_catch[i] = 0;
}
- /* Signals caused by debugger's own actions
- should not be given to the program afterwards. */
+ /* Signals caused by debugger's own actions should not be given to
+ the program afterwards.
+
+ Do not deliver GDB_SIGNAL_TRAP by default, except when the user
+ explicitly specifies that it should be delivered to the target
+ program. Typically, that would occur when a user is debugging a
+ target monitor on a simulator: the target monitor sets a
+ breakpoint; the simulator encounters this breakpoint and halts
+ the simulation handing control to GDB; GDB, noting that the stop
+ address doesn't map to any known breakpoint, returns control back
+ to the simulator; the simulator then delivers the hardware
+ equivalent of a GDB_SIGNAL_TRAP to the program being
+ debugged. */
signal_program[GDB_SIGNAL_TRAP] = 0;
signal_program[GDB_SIGNAL_INT] = 0;
diff --git a/gdb/thread.c b/gdb/thread.c
index 28e5ef8..3e3f419 100644
--- a/gdb/thread.c
+++ b/gdb/thread.c
@@ -852,44 +852,62 @@ thread_change_ptid (ptid_t old_ptid, ptid_t new_ptid)
observer_notify_thread_ptid_changed (old_ptid, new_ptid);
}
+/* Helper for set_running, that marks one thread either running or
+ stopped. */
+
+static int
+set_running_thread (struct thread_info *tp, int running)
+{
+ int started = 0;
+
+ if (running && tp->state == THREAD_STOPPED)
+ started = 1;
+ tp->state = running ? THREAD_RUNNING : THREAD_STOPPED;
+
+ if (!running)
+ {
+ /* If the thread is now marked stopped, remove it from
+ the step-over queue, so that we don't try to resume
+ it until the user wants it to. */
+ if (tp->step_over_next != NULL)
+ thread_step_over_chain_remove (tp);
+ }
+
+ return started;
+}
+
void
set_running (ptid_t ptid, int running)
{
struct thread_info *tp;
int all = ptid_equal (ptid, minus_one_ptid);
+ int any_started = 0;
/* We try not to notify the observer if no thread has actually changed
the running state -- merely to reduce the number of messages to
frontend. Frontend is supposed to handle multiple *running just fine. */
if (all || ptid_is_pid (ptid))
{
- int any_started = 0;
-
for (tp = thread_list; tp; tp = tp->next)
if (all || ptid_get_pid (tp->ptid) == ptid_get_pid (ptid))
{
if (tp->state == THREAD_EXITED)
continue;
- if (running && tp->state == THREAD_STOPPED)
+
+ if (set_running_thread (tp, running))
any_started = 1;
- tp->state = running ? THREAD_RUNNING : THREAD_STOPPED;
}
- if (any_started)
- observer_notify_target_resumed (ptid);
}
else
{
- int started = 0;
-
tp = find_thread_ptid (ptid);
- gdb_assert (tp);
+ gdb_assert (tp != NULL);
gdb_assert (tp->state != THREAD_EXITED);
- if (running && tp->state == THREAD_STOPPED)
- started = 1;
- tp->state = running ? THREAD_RUNNING : THREAD_STOPPED;
- if (started)
- observer_notify_target_resumed (ptid);
+ if (set_running_thread (tp, running))
+ any_started = 1;
}
+ if (any_started)
+ observer_notify_target_resumed (ptid);
}
static int
@@ -1008,9 +1026,8 @@ finish_thread_state (ptid_t ptid)
continue;
if (all || ptid_get_pid (ptid) == ptid_get_pid (tp->ptid))
{
- if (tp->executing && tp->state == THREAD_STOPPED)
+ if (set_running_thread (tp, tp->executing))
any_started = 1;
- tp->state = tp->executing ? THREAD_RUNNING : THREAD_STOPPED;
}
}
}
@@ -1020,9 +1037,8 @@ finish_thread_state (ptid_t ptid)
gdb_assert (tp);
if (tp->state != THREAD_EXITED)
{
- if (tp->executing && tp->state == THREAD_STOPPED)
+ if (set_running_thread (tp, tp->executing))
any_started = 1;
- tp->state = tp->executing ? THREAD_RUNNING : THREAD_STOPPED;
}
}
--
1.9.3
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 06/17] Use keep_going in proceed and start_step_over too
2015-04-17 10:45 ` [PATCH v3 06/17] Use keep_going in proceed and start_step_over too Pedro Alves
@ 2015-04-22 5:09 ` Doug Evans
2015-04-22 22:22 ` Pedro Alves
0 siblings, 1 reply; 62+ messages in thread
From: Doug Evans @ 2015-04-22 5:09 UTC (permalink / raw)
To: Pedro Alves; +Cc: gdb-patches
Pedro Alves writes:
> The main motivation of this patch is sharing more code between the
> proceed (starting the inferior for the first time) and keep_going
> (restarting the inferior after handling an event) paths and using the
> step_over_chain queue now embedded in the thread_info object for
> pending in-line step-overs too (instead of just for displaced
> stepping).
Hi.
A couple of nits inline.
grep for ====
>
> So this commit:
>
> - splits out a new keep_going_pass function out of keep_going that is
> just like keep_going except for the bits that clear the signal to
> pass if the signal is set to "handle nopass".
>
> - makes proceed use keep_going too.
>
> - Makes start_step_over use keep_going_pass instead of lower level
> displaced stepping things.
>
> One user visible change: if inserting breakpoints while trying to
> proceed fails, we now get:
>
> (gdb) si
> Warning:
> Could not insert hardware watchpoint 7.
> Could not insert hardware breakpoints:
> You may have requested too many hardware breakpoints/watchpoints.
>
> Command aborted.
> (gdb)
>
> while before we only saw warnings with no indication that the command
> was cancelled:
>
> (gdb) si
> Warning:
> Could not insert hardware watchpoint 7.
> Could not insert hardware breakpoints:
> You may have requested too many hardware breakpoints/watchpoints.
>
> (gdb)
>
> Tested on x86_64-linux-gnu, ppc64-linux-gnu and s390-linux-gnu.
>
> gdb/ChangeLog:
> 2015-04-17 Pedro Alves <palves@redhat.com>
>
> * gdbthread.h (struct thread_info) <prev_pc>: Extend comment.
> * infrun.c (struct execution_control_state): Move higher up in the
> file.
> (reset_ecs): New function.
> (start_step_over): Now returns int. Rewrite to use
> keep_going_pass instead of manually starting a displaced step.
> (resume): Don't call set_running here. If displaced stepping
> can't start now, clear trap_expected.
> (find_thread_needs_step_over): Delete function.
> (proceed): Set up finish_thread_state_cleanup. Call set_running.
> If the current thread needs a step over, push it in the step-over
> chain. Don't set insert breakpoints nor call resume directly
> here. Instead rewrite to use start_step_over and keep_going_pass.
> (finish_step_over): New function.
> (handle_signal_stop): Call finish_step_over instead of
> start_step_over.
> (switch_back_to_stepped_thread): If the event thread needs another
> step-over do that first. Use start_step_over.
> (keep_going_pass): New function, factored out from ...
> (keep_going): ... here.
> (_initialize_infrun): Comment moved here.
> * thread.c (set_running_thread): New function.
> (set_running, finish_thread_state): Use set_running_thread.
>
> v3:
>
> - Adjusted to step_over_chain changes. Fixed gdb internal error
> caught by signal-sigtrap.exp on ARM (software single-step target).
>
> v2:
>
> - Testing on x86_64 with software single-step revealed v1 missed
> setting prev_pc on proceed, for switch_back_to_stepped_thread.
> Rewrite tail end of proceed to fix that. This ends up simplifying
> later patches in the series.
>...
>
> diff --git a/gdb/gdbthread.h b/gdb/gdbthread.h
> index 2c871a2..7052ee1 100644
> --- a/gdb/gdbthread.h
> +++ b/gdb/gdbthread.h
> @@ -208,8 +208,10 @@ struct thread_info
>
> /* Internal stepping state. */
>
> - /* Record the pc of the thread the last time it stopped. This is
> - maintained by proceed and keep_going, and used in
> + /* Record the pc of the thread the last time it was resumed. (It
> + can't be done on stop as the PC may change since the last stop,
> + e.g., "return" command, or "p $pc = 0xf000"). This is maintained
> + by proceed and keep_going, and among other things, it's used in
> adjust_pc_after_break to distinguish a hardware single-step
> SIGTRAP from a breakpoint SIGTRAP. */
> CORE_ADDR prev_pc;
> diff --git a/gdb/infrun.c b/gdb/infrun.c
> index f325a53..beb9ea2 100644
> --- a/gdb/infrun.c
> +++ b/gdb/infrun.c
> @@ -1844,30 +1844,61 @@ displaced_step_fixup (ptid_t event_ptid, enum gdb_signal signal)
> displaced->step_ptid = null_ptid;
> }
>
> -/* Are there any pending step-over requests? If so, run one now. */
> +/* Data to be passed around while handling an event. This data is
> + discarded between events. */
> +struct execution_control_state
> +{
> + ptid_t ptid;
> + /* The thread that got the event, if this was a thread event; NULL
> + otherwise. */
> + struct thread_info *event_thread;
> +
> + struct target_waitstatus ws;
> + int stop_func_filled_in;
> + CORE_ADDR stop_func_start;
> + CORE_ADDR stop_func_end;
> + const char *stop_func_name;
> + int wait_some_more;
> +
> + /* True if the event thread hit the single-step breakpoint of
> + another thread. Thus the event doesn't cause a stop, the thread
> + needs to be single-stepped past the single-step breakpoint before
> + we can switch back to the original stepping thread. */
> + int hit_singlestep_breakpoint;
> +};
> +
> +/* Clear ECS and set it to point at TP. */
>
> static void
> +reset_ecs (struct execution_control_state *ecs, struct thread_info *tp)
> +{
> + memset (ecs, 0, sizeof (*ecs));
> + ecs->event_thread = tp;
> + ecs->ptid = tp->ptid;
> +}
> +
> +static void keep_going_pass (struct execution_control_state *ecs);
> +static void prepare_to_wait (struct execution_control_state *ecs);
> +static int thread_still_needs_step_over (struct thread_info *tp);
> +
> +/* Are there any pending step-over requests? If so, run one now and
> + return true. Otherwise, return false. */
> +
> +static int
> start_step_over (void)
> {
> struct thread_info *tp, *next;
>
> for (tp = step_over_queue_head; tp != NULL; tp = next)
> {
> - ptid_t ptid;
> - struct displaced_step_inferior_state *displaced;
> - struct regcache *regcache;
> - struct gdbarch *gdbarch;
> - CORE_ADDR actual_pc;
> - struct address_space *aspace;
> - struct inferior *inf = find_inferior_ptid (tp->ptid);
> + struct execution_control_state ecss;
> + struct execution_control_state *ecs = &ecss;
>
> next = thread_step_over_chain_next (tp);
>
> - displaced = get_displaced_stepping_state (inf->pid);
> -
> /* If this inferior already has a displaced step in process,
> don't start a new one. */
> - if (!ptid_equal (displaced->step_ptid, null_ptid))
> + if (displaced_step_in_progress (ptid_get_pid (tp->ptid)))
> continue;
>
> thread_step_over_chain_remove (tp);
> @@ -1879,74 +1910,57 @@ start_step_over (void)
> "infrun: step-over queue now empty\n");
> }
>
> - ptid = tp->ptid;
> - context_switch (ptid);
> -
> - regcache = get_thread_regcache (ptid);
> - actual_pc = regcache_read_pc (regcache);
> - aspace = get_regcache_aspace (regcache);
> - gdbarch = get_regcache_arch (regcache);
> -
> - if (breakpoint_here_p (aspace, actual_pc))
> + if (tp->control.trap_expected || tp->executing)
> {
> - if (debug_displaced)
> - fprintf_unfiltered (gdb_stdlog,
> - "displaced: stepping queued %s now\n",
> - target_pid_to_str (ptid));
> -
> - displaced_step_prepare (ptid);
> -
> - if (debug_displaced)
> - {
> - CORE_ADDR actual_pc = regcache_read_pc (regcache);
> - gdb_byte buf[4];
> -
> - fprintf_unfiltered (gdb_stdlog, "displaced: run %s: ",
> - paddress (gdbarch, actual_pc));
> - read_memory (actual_pc, buf, sizeof (buf));
> - displaced_step_dump_bytes (gdb_stdlog, buf, sizeof (buf));
> - }
> -
> - if (gdbarch_displaced_step_hw_singlestep (gdbarch,
> - displaced->step_closure))
> - target_resume (ptid, 1, GDB_SIGNAL_0);
> - else
> - target_resume (ptid, 0, GDB_SIGNAL_0);
> -
> - /* Done, we're stepping a thread. */
> - break;
> + internal_error (__FILE__, __LINE__,
> + "[%s] has inconsistent state: "
> + "trap_expected=%d, executing=%d\n",
> + target_pid_to_str (tp->ptid),
> + tp->control.trap_expected,
> + tp->executing);
> }
> - else
> - {
> - int step;
> - struct thread_info *tp = inferior_thread ();
> -
> - /* The breakpoint we were sitting under has since been
> - removed. */
> - tp->control.trap_expected = 0;
>
> - /* Go back to what we were trying to do. */
> - step = currently_stepping (tp);
> -
> - if (step)
> - step = maybe_software_singlestep (gdbarch, actual_pc);
> + if (debug_infrun)
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: resuming [%s] for step-over\n",
> + target_pid_to_str (tp->ptid));
> +
> + /* keep_going_pass skips the step-over if the breakpoint is no
> + longer inserted. In all-stop, we want to keep looking for a
====
missing word? "... keep looking for a thread" ?
> + that needs a step-over instead of resuming TP, because we
> + wouldn't be able to resume anything else until the target
> + stops again. In non-stop, the resume always resumes only TP,
> + so it's OK to let the thread resume freely. */
> + if (!non_stop && !thread_still_needs_step_over (tp))
> + continue;
>
> - if (debug_displaced)
> - fprintf_unfiltered (gdb_stdlog,
> - "displaced: breakpoint is gone: %s, step(%d)\n",
> - target_pid_to_str (tp->ptid), step);
> + switch_to_thread (tp->ptid);
> + reset_ecs (ecs, tp);
> + keep_going_pass (ecs);
>
> - target_resume (ptid, step, GDB_SIGNAL_0);
> - tp->suspend.stop_signal = GDB_SIGNAL_0;
> + if (!ecs->wait_some_more)
> + error (_("Command aborted."));
>
> - /* This request was discarded. See if there's any other
> - thread waiting for its turn. */
> + if (!non_stop)
> + {
> + /* On all-stop, shouldn't have resumed unless we needed a
> + step over. */
> + gdb_assert (tp->control.trap_expected
> + || tp->step_after_step_resume_breakpoint);
> +
> + /* With remote targets (at least), in all-stop, we can't
> + issue any further remote commands until the program stops
> + again. */
> + return 1;
> }
>
> - /* A new displaced stepping sequence started. Maybe we can
> - start a displaced step on a thread of other process.
> - Continue looking. */
> + /* Either the thread no longer needed a step-over, or a new
> + displaced stepping sequence started. Even in the latter
> + case, continue looking. Maybe we can also start another
> + displaced step on a thread of other process. */
> }
> +
> + return 0;
> }
>
> /* Update global variables holding ptids to hold NEW_PTID if they were
> @@ -2283,16 +2297,11 @@ resume (enum gdb_signal sig)
>
> if (!displaced_step_prepare (inferior_ptid))
> {
> - /* Got placed in displaced stepping queue. Will be resumed
> - later when all the currently queued displaced stepping
> - requests finish. The thread is not executing at this
> - point, and the call to set_executing will be made later.
> - But we need to call set_running here, since from the
> - user/frontend's point of view, threads were set running.
> - Unless we're calling an inferior function, as in that
> - case we pretend the inferior doesn't run at all. */
> - if (!tp->control.in_infcall)
> - set_running (user_visible_resume_ptid (user_step), 1);
> + if (debug_infrun)
> + fprintf_unfiltered (gdb_stdlog,
> + "Got placed in displaced stepping queue\n");
====
Nit: Suggest rewording "displaced stepping queue".
IIUC there is no displaced-stepping specific queue anymore.
> +
> + tp->control.trap_expected = 0;
> discard_cleanups (old_cleanups);
> return;
> }
> @@ -2367,14 +2376,6 @@ resume (enum gdb_signal sig)
> by applying increasingly restricting conditions. */
> resume_ptid = user_visible_resume_ptid (user_step);
>
> - /* Even if RESUME_PTID is a wildcard, and we end up resuming less
> - (e.g., we might need to step over a breakpoint), from the
> - user/frontend's point of view, all threads in RESUME_PTID are now
> - running. Unless we're calling an inferior function, as in that
> - case pretend we inferior doesn't run at all. */
> - if (!tp->control.in_infcall)
> - set_running (resume_ptid, 1);
> -
> /* Maybe resume a single thread after all. */
> if ((step || thread_has_single_step_breakpoints_set (tp))
> && tp->control.trap_expected)
> @@ -2589,48 +2590,6 @@ schedlock_applies (struct thread_info *tp)
> && tp->control.stepping_command));
> }
>
> -/* Look a thread other than EXCEPT that has previously reported a
> - breakpoint event, and thus needs a step-over in order to make
> - progress. Returns NULL is none is found. */
> -
> -static struct thread_info *
> -find_thread_needs_step_over (struct thread_info *except)
> -{
> - struct thread_info *tp, *current;
> -
> - /* With non-stop mode on, threads are always handled individually. */
> - gdb_assert (! non_stop);
> -
> - current = inferior_thread ();
> -
> - /* If scheduler locking applies, we can avoid iterating over all
> - threads. */
> - if (schedlock_applies (except))
> - {
> - if (except != current
> - && thread_still_needs_step_over (current))
> - return current;
> -
> - return NULL;
> - }
> -
> - ALL_NON_EXITED_THREADS (tp)
> - {
> - /* Ignore the EXCEPT thread. */
> - if (tp == except)
> - continue;
> - /* Ignore threads of processes we're not resuming. */
> - if (!sched_multi
> - && ptid_get_pid (tp->ptid) != ptid_get_pid (inferior_ptid))
> - continue;
> -
> - if (thread_still_needs_step_over (tp))
> - return tp;
> - }
> -
> - return NULL;
> -}
> -
> /* Basic routine for continuing the program in various fashions.
>
> ADDR is the address to resume at, or -1 for resume where stopped.
> @@ -2651,6 +2610,11 @@ proceed (CORE_ADDR addr, enum gdb_signal siggnal)
> struct thread_info *tp;
> CORE_ADDR pc;
> struct address_space *aspace;
> + ptid_t resume_ptid;
> + struct execution_control_state ecss;
> + struct execution_control_state *ecs = &ecss;
> + struct cleanup *old_chain;
> + int started;
>
> /* If we're stopped at a fork/vfork, follow the branch set by the
> "set follow-fork-mode" command; otherwise, we'll just proceed
> @@ -2713,7 +2677,23 @@ proceed (CORE_ADDR addr, enum gdb_signal siggnal)
> (next/step/etc.), we'll want to print stop event output to the MI
> console channel (the stepped-to line, etc.), as if the user
> entered the execution command on a real GDB console. */
> - inferior_thread ()->control.command_interp = command_interp ();
> + tp->control.command_interp = command_interp ();
> +
> + resume_ptid = user_visible_resume_ptid (tp->control.stepping_command);
> +
> + /* If an exception is thrown from this point on, make sure to
> + propagate GDB's knowledge of the executing state to the
> + frontend/user running state. */
> + old_chain = make_cleanup (finish_thread_state_cleanup, &resume_ptid);
> +
> + /* Even if RESUME_PTID is a wildcard, and we end up resuming fewer
> + threads (e.g., we might need to set threads stepping over
> + breakpoints first), from the user/frontend's point of view, all
> + threads in RESUME_PTID are now running. Unless we're calling an
> + inferior function, as in that case we pretend the inferior
> + doesn't run at all. */
> + if (!tp->control.in_infcall)
> + set_running (resume_ptid, 1);
>
> if (debug_infrun)
> fprintf_unfiltered (gdb_stdlog,
> @@ -2721,91 +2701,92 @@ proceed (CORE_ADDR addr, enum gdb_signal siggnal)
> paddress (gdbarch, addr),
> gdb_signal_to_symbol_string (siggnal));
>
> - if (non_stop)
> - /* In non-stop, each thread is handled individually. The context
> - must already be set to the right thread here. */
> - ;
> - else
> + annotate_starting ();
> +
> + /* Make sure that output from GDB appears before output from the
> + inferior. */
> + gdb_flush (gdb_stdout);
> +
> + /* In a multi-threaded task we may select another thread and
> + then continue or step.
> +
> + But if a thread that we're resuming had stopped at a breakpoint,
> + it will immediately cause another breakpoint stop without any
> + execution (i.e. it will report a breakpoint hit incorrectly). So
> + we must step over it first.
> +
> + Look for threads other than the current (TP) that reported a
> + breakpoint hit and haven't been resumed yet since. */
> +
> + /* If scheduler locking applies, we can avoid iterating over all
> + threads. */
> + if (!non_stop && !schedlock_applies (tp))
> {
> - struct thread_info *step_over;
> + struct thread_info *current = tp;
> +
> + ALL_NON_EXITED_THREADS (tp)
> + {
> + /* Ignore the current thread here. It's handled
> + afterwards. */
> + if (tp == current)
> + continue;
>
> - /* In a multi-threaded task we may select another thread and
> - then continue or step.
> + /* Ignore threads of processes we're not resuming. */
> + if (!ptid_match (tp->ptid, resume_ptid))
> + continue;
>
> - But if the old thread was stopped at a breakpoint, it will
> - immediately cause another breakpoint stop without any
> - execution (i.e. it will report a breakpoint hit incorrectly).
> - So we must step over it first.
> + if (!thread_still_needs_step_over (tp))
> + continue;
> +
> + gdb_assert (!thread_is_in_step_over_chain (tp));
>
> - Look for a thread other than the current (TP) that reported a
> - breakpoint hit and hasn't been resumed yet since. */
> - step_over = find_thread_needs_step_over (tp);
> - if (step_over != NULL)
> - {
> if (debug_infrun)
> fprintf_unfiltered (gdb_stdlog,
> "infrun: need to step-over [%s] first\n",
> - target_pid_to_str (step_over->ptid));
> + target_pid_to_str (tp->ptid));
>
> - /* Store the prev_pc for the stepping thread too, needed by
> - switch_back_to_stepped_thread. */
> - tp->prev_pc = regcache_read_pc (get_current_regcache ());
> - switch_to_thread (step_over->ptid);
> - tp = step_over;
> + thread_step_over_chain_enqueue (tp);
> }
> - }
> -
> - /* If we need to step over a breakpoint, and we're not using
> - displaced stepping to do so, insert all breakpoints (watchpoints,
> - etc.) but the one we're stepping over, step one instruction, and
> - then re-insert the breakpoint when that step is finished. */
> - if (tp->stepping_over_breakpoint && !use_displaced_stepping (gdbarch))
> - {
> - struct regcache *regcache = get_current_regcache ();
>
> - set_step_over_info (get_regcache_aspace (regcache),
> - regcache_read_pc (regcache), 0);
> + tp = current;
> }
> - else
> - clear_step_over_info ();
>
> - insert_breakpoints ();
> + /* Enqueue the current thread last, so that we move all other
> + threads over their breakpoints first. */
> + if (tp->stepping_over_breakpoint)
> + thread_step_over_chain_enqueue (tp);
>
> - tp->control.trap_expected = tp->stepping_over_breakpoint;
> + /* If the thread isn't started, we'll still need to set its prev_pc,
> + so that switch_back_to_stepped_thread knows the thread hasn't
> + advanced. Must do this before resuming any thread, as in
> + all-stop/remote, once we resume we can't send any other packet
> + until the target stops again. */
> + tp->prev_pc = regcache_read_pc (regcache);
>
> - annotate_starting ();
> + started = start_step_over ();
>
> - /* Make sure that output from GDB appears before output from the
> - inferior. */
> - gdb_flush (gdb_stdout);
> + if (step_over_info_valid_p ())
> + {
> + /* Either this thread started a new in-line step over, or some
> + other thread was already doing one. In either case, don't
> + resume anything else until the step-over is finished. */
> + }
> + else if (started && !non_stop)
> + {
> + /* A new displaced stepping sequence was started. In all-stop,
> + we can't talk to the target anymore until it next stops. */
> + }
> + else if (!tp->executing && !thread_is_in_step_over_chain (tp))
> + {
> + /* The thread wasn't started, and isn't queued, run it now. */
> + reset_ecs (ecs, tp);
> + switch_to_thread (tp->ptid);
> + keep_going_pass (ecs);
> + if (!ecs->wait_some_more)
> + error ("Command aborted.");
> + }
>
> - /* Refresh prev_pc value just prior to resuming. This used to be
> - done in stop_waiting, however, setting prev_pc there did not handle
> - scenarios such as inferior function calls or returning from
> - a function via the return command. In those cases, the prev_pc
> - value was not set properly for subsequent commands. The prev_pc value
> - is used to initialize the starting line number in the ecs. With an
> - invalid value, the gdb next command ends up stopping at the position
> - represented by the next line table entry past our start position.
> - On platforms that generate one line table entry per line, this
> - is not a problem. However, on the ia64, the compiler generates
> - extraneous line table entries that do not increase the line number.
> - When we issue the gdb next command on the ia64 after an inferior call
> - or a return command, we often end up a few instructions forward, still
> - within the original line we started.
> -
> - An attempt was made to refresh the prev_pc at the same time the
> - execution_control_state is initialized (for instance, just before
> - waiting for an inferior event). But this approach did not work
> - because of platforms that use ptrace, where the pc register cannot
> - be read unless the inferior is stopped. At that point, we are not
> - guaranteed the inferior is stopped and so the regcache_read_pc() call
> - can fail. Setting the prev_pc value here ensures the value is updated
> - correctly when the inferior is stopped. */
> - tp->prev_pc = regcache_read_pc (get_current_regcache ());
> -
> - /* Resume inferior. */
> - resume (tp->suspend.stop_signal);
> + discard_cleanups (old_chain);
>
> /* Wait for it to stop (if not standalone)
> and in any case decode why it stopped, and act accordingly. */
> @@ -2873,28 +2854,6 @@ init_wait_for_inferior (void)
> }
>
> \f
> -/* Data to be passed around while handling an event. This data is
> - discarded between events. */
> -struct execution_control_state
> -{
> - ptid_t ptid;
> - /* The thread that got the event, if this was a thread event; NULL
> - otherwise. */
> - struct thread_info *event_thread;
> -
> - struct target_waitstatus ws;
> - int stop_func_filled_in;
> - CORE_ADDR stop_func_start;
> - CORE_ADDR stop_func_end;
> - const char *stop_func_name;
> - int wait_some_more;
> -
> - /* True if the event thread hit the single-step breakpoint of
> - another thread. Thus the event doesn't cause a stop, the thread
> - needs to be single-stepped past the single-step breakpoint before
> - we can switch back to the original stepping thread. */
> - int hit_singlestep_breakpoint;
> -};
>
> static void handle_inferior_event (struct execution_control_state *ecs);
>
> @@ -2908,7 +2867,6 @@ static void check_exception_resume (struct execution_control_state *,
>
> static void end_stepping_range (struct execution_control_state *ecs);
> static void stop_waiting (struct execution_control_state *ecs);
> -static void prepare_to_wait (struct execution_control_state *ecs);
> static void keep_going (struct execution_control_state *ecs);
> static void process_event_stop_test (struct execution_control_state *ecs);
> static int switch_back_to_stepped_thread (struct execution_control_state *ecs);
> @@ -4261,6 +4219,34 @@ Cannot fill $_exitsignal with the correct signal number.\n"));
> }
> }
>
> +/* Called when we get an event that may finish an in-line or
> + out-of-line (displaced stepping) step-over started previously. */
> +
> +static void
> +finish_step_over (struct execution_control_state *ecs)
> +{
> + displaced_step_fixup (ecs->ptid,
> + ecs->event_thread->suspend.stop_signal);
> +
> + if (step_over_info_valid_p ())
> + {
> + /* If we're stepping over a breakpoint with all threads locked,
> + then only the thread that was stepped should be reporting
> + back an event. */
> + gdb_assert (ecs->event_thread->control.trap_expected);
> +
> + if (ecs->event_thread->suspend.stop_signal == GDB_SIGNAL_TRAP)
> + clear_step_over_info ();
> + }
> +
> + if (!non_stop)
> + return;
> +
> + /* Start a new step-over in another thread if there's one that
> + needs it. */
> + start_step_over ();
> +}
> +
> /* Come here when the program has stopped with a signal. */
>
> static void
> @@ -4277,9 +4263,7 @@ handle_signal_stop (struct execution_control_state *ecs)
> /* Do we need to clean up the state of a thread that has
> completed a displaced single-step? (Doing so usually affects
> the PC, so do it here, before we set stop_pc.) */
> - displaced_step_fixup (ecs->ptid,
> - ecs->event_thread->suspend.stop_signal);
> - start_step_over ();
> + finish_step_over (ecs);
>
> /* If we either finished a single-step or hit a breakpoint, but
> the user wanted this thread to be stopped, pretend we got a
> @@ -5629,7 +5613,6 @@ switch_back_to_stepped_thread (struct execution_control_state *ecs)
> {
> struct thread_info *tp;
> struct thread_info *stepping_thread;
> - struct thread_info *step_over;
>
> /* If any thread is blocked on some internal breakpoint, and we
> simply need to step over that breakpoint to get it going
> @@ -5672,14 +5655,20 @@ switch_back_to_stepped_thread (struct execution_control_state *ecs)
> return 1;
> }
>
> - /* Otherwise, we no longer expect a trap in the current thread.
> - Clear the trap_expected flag before switching back -- this is
> - what keep_going does as well, if we call it. */
> - ecs->event_thread->control.trap_expected = 0;
> -
> - /* Likewise, clear the signal if it should not be passed. */
> - if (!signal_program[ecs->event_thread->suspend.stop_signal])
> - ecs->event_thread->suspend.stop_signal = GDB_SIGNAL_0;
> + /* If this thread needs yet another step-over (e.g., stepping
> + through a delay slot), do it first before moving on to
> + another thread. */
> + if (thread_still_needs_step_over (ecs->event_thread))
> + {
> + if (debug_infrun)
> + {
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: thread [%s] still needs step-over\n",
> + target_pid_to_str (ecs->event_thread->ptid));
> + }
> + keep_going (ecs);
> + return 1;
> + }
>
> /* If scheduler locking applies even if not stepping, there's no
> need to walk over threads. Above we've checked whether the
> @@ -5689,12 +5678,26 @@ switch_back_to_stepped_thread (struct execution_control_state *ecs)
> if (schedlock_applies (ecs->event_thread))
> return 0;
>
> - /* Look for the stepping/nexting thread, and check if any other
> - thread other than the stepping thread needs to start a
> - step-over. Do all step-overs before actually proceeding with
> + /* Otherwise, we no longer expect a trap in the current thread.
> + Clear the trap_expected flag before switching back -- this is
> + what keep_going does as well, if we call it. */
> + ecs->event_thread->control.trap_expected = 0;
> +
> + /* Likewise, clear the signal if it should not be passed. */
> + if (!signal_program[ecs->event_thread->suspend.stop_signal])
> + ecs->event_thread->suspend.stop_signal = GDB_SIGNAL_0;
> +
> + /* Do all pending step-overs before actually proceeding with
> step/next/etc. */
> + if (start_step_over ())
> + {
> + prepare_to_wait (ecs);
> + return 1;
> + }
> +
> + /* Look for the stepping/nexting thread. */
> stepping_thread = NULL;
> - step_over = NULL;
> +
> ALL_NON_EXITED_THREADS (tp)
> {
> /* Ignore threads of processes we're not resuming. */
> @@ -5726,37 +5729,6 @@ switch_back_to_stepped_thread (struct execution_control_state *ecs)
>
> stepping_thread = tp;
> }
> - else if (thread_still_needs_step_over (tp))
> - {
> - step_over = tp;
> -
> - /* At the top we've returned early if the event thread
> - is stepping. If some other thread not the event
> - thread is stepping, then scheduler locking can't be
> - in effect, and we can resume this thread. No need to
> - keep looking for the stepping thread then. */
> - break;
> - }
> - }
> -
> - if (step_over != NULL)
> - {
> - tp = step_over;
> - if (debug_infrun)
> - {
> - fprintf_unfiltered (gdb_stdlog,
> - "infrun: need to step-over [%s]\n",
> - target_pid_to_str (tp->ptid));
> - }
> -
> - /* Only the stepping thread should have this set. */
> - gdb_assert (tp->control.step_range_end == 0);
> -
> - ecs->ptid = tp->ptid;
> - ecs->event_thread = tp;
> - switch_to_thread (ecs->ptid);
> - keep_going (ecs);
> - return 1;
> }
>
> if (stepping_thread != NULL)
> @@ -5855,7 +5827,7 @@ switch_back_to_stepped_thread (struct execution_control_state *ecs)
> fprintf_unfiltered (gdb_stdlog,
> "infrun: expected thread still "
> "hasn't advanced\n");
> - keep_going (ecs);
> + keep_going_pass (ecs);
> }
>
> return 1;
> @@ -6270,24 +6242,32 @@ stop_waiting (struct execution_control_state *ecs)
> ecs->wait_some_more = 0;
> }
>
> -/* Called when we should continue running the inferior, because the
> - current event doesn't cause a user visible stop. This does the
> - resuming part; waiting for the next event is done elsewhere. */
> +/* Like keep_going, but passes the signal to the inferior, even if the
> + signal is set to nopass. */
>
> static void
> -keep_going (struct execution_control_state *ecs)
> +keep_going_pass (struct execution_control_state *ecs)
====
For whatever weird reasons, "keep_going_pass" doesn't read very well to me.
"pass" as in "pass/fail"?
"pass" as in one of several passes, and this is the "keep going" one?
Can you rename this to keep_going_pass_signal?
TIA
> {
> /* Make sure normal_stop is called if we get a QUIT handled before
> reaching resume. */
> struct cleanup *old_cleanups = make_cleanup (resume_cleanups, 0);
>
> + gdb_assert (ptid_equal (ecs->event_thread->ptid, inferior_ptid));
> +
> /* Save the pc before execution, to compare with pc after stop. */
> ecs->event_thread->prev_pc
> = regcache_read_pc (get_thread_regcache (ecs->ptid));
>
> - if (ecs->event_thread->control.trap_expected
> - && ecs->event_thread->suspend.stop_signal != GDB_SIGNAL_TRAP)
> + if (ecs->event_thread->control.trap_expected)
> {
> + struct thread_info *tp = ecs->event_thread;
> +
> + if (debug_infrun)
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: %s has trap_expected set, "
> + "resuming to collect trap\n",
> + target_pid_to_str (tp->ptid));
> +
> /* We haven't yet gotten our trap, and either: intercepted a
> non-signal event (e.g., a fork); or took a signal which we
> are supposed to pass through to the inferior. Simply
> @@ -6358,20 +6338,6 @@ keep_going (struct execution_control_state *ecs)
>
> ecs->event_thread->control.trap_expected = (remove_bp || remove_wps);
>
> - /* Do not deliver GDB_SIGNAL_TRAP (except when the user
> - explicitly specifies that such a signal should be delivered
> - to the target program). Typically, that would occur when a
> - user is debugging a target monitor on a simulator: the target
> - monitor sets a breakpoint; the simulator encounters this
> - breakpoint and halts the simulation handing control to GDB;
> - GDB, noting that the stop address doesn't map to any known
> - breakpoint, returns control back to the simulator; the
> - simulator then delivers the hardware equivalent of a
> - GDB_SIGNAL_TRAP to the program being debugged. */
> - if (ecs->event_thread->suspend.stop_signal == GDB_SIGNAL_TRAP
> - && !signal_program[ecs->event_thread->suspend.stop_signal])
> - ecs->event_thread->suspend.stop_signal = GDB_SIGNAL_0;
> -
> discard_cleanups (old_cleanups);
> resume (ecs->event_thread->suspend.stop_signal);
> }
> @@ -6379,6 +6345,22 @@ keep_going (struct execution_control_state *ecs)
> prepare_to_wait (ecs);
> }
>
> +/* Called when we should continue running the inferior, because the
> + current event doesn't cause a user visible stop. This does the
> + resuming part; waiting for the next event is done elsewhere. */
> +
> +static void
> +keep_going (struct execution_control_state *ecs)
> +{
> + if (ecs->event_thread->control.trap_expected
> + && ecs->event_thread->suspend.stop_signal == GDB_SIGNAL_TRAP)
> + ecs->event_thread->control.trap_expected = 0;
> +
> + if (!signal_program[ecs->event_thread->suspend.stop_signal])
> + ecs->event_thread->suspend.stop_signal = GDB_SIGNAL_0;
> + keep_going_pass (ecs);
> +}
> +
> /* This function normally comes after a resume, before
> handle_inferior_event exits. It takes care of any last bits of
> housekeeping, and sets the all-important wait_some_more flag. */
> @@ -7806,8 +7788,19 @@ leave it stopped or free to run as needed."),
> signal_catch[i] = 0;
> }
>
> - /* Signals caused by debugger's own actions
> - should not be given to the program afterwards. */
> + /* Signals caused by debugger's own actions should not be given to
> + the program afterwards.
> +
> + Do not deliver GDB_SIGNAL_TRAP by default, except when the user
> + explicitly specifies that it should be delivered to the target
> + program. Typically, that would occur when a user is debugging a
> + target monitor on a simulator: the target monitor sets a
> + breakpoint; the simulator encounters this breakpoint and halts
> + the simulation handing control to GDB; GDB, noting that the stop
> + address doesn't map to any known breakpoint, returns control back
> + to the simulator; the simulator then delivers the hardware
> + equivalent of a GDB_SIGNAL_TRAP to the program being
> + debugged. */
> signal_program[GDB_SIGNAL_TRAP] = 0;
> signal_program[GDB_SIGNAL_INT] = 0;
>
> diff --git a/gdb/thread.c b/gdb/thread.c
> index 28e5ef8..3e3f419 100644
> --- a/gdb/thread.c
> +++ b/gdb/thread.c
> @@ -852,44 +852,62 @@ thread_change_ptid (ptid_t old_ptid, ptid_t new_ptid)
> observer_notify_thread_ptid_changed (old_ptid, new_ptid);
> }
>
> +/* Helper for set_running, that marks one thread either running or
> + stopped. */
> +
> +static int
> +set_running_thread (struct thread_info *tp, int running)
> +{
> + int started = 0;
> +
> + if (running && tp->state == THREAD_STOPPED)
> + started = 1;
> + tp->state = running ? THREAD_RUNNING : THREAD_STOPPED;
> +
> + if (!running)
> + {
> + /* If the thread is now marked stopped, remove it from
> + the step-over queue, so that we don't try to resume
> + it until the user wants it to. */
> + if (tp->step_over_next != NULL)
> + thread_step_over_chain_remove (tp);
> + }
> +
> + return started;
> +}
> +
> void
> set_running (ptid_t ptid, int running)
> {
> struct thread_info *tp;
> int all = ptid_equal (ptid, minus_one_ptid);
> + int any_started = 0;
>
> /* We try not to notify the observer if no thread has actually changed
> the running state -- merely to reduce the number of messages to
> frontend. Frontend is supposed to handle multiple *running just fine. */
> if (all || ptid_is_pid (ptid))
> {
> - int any_started = 0;
> -
> for (tp = thread_list; tp; tp = tp->next)
> if (all || ptid_get_pid (tp->ptid) == ptid_get_pid (ptid))
> {
> if (tp->state == THREAD_EXITED)
> continue;
> - if (running && tp->state == THREAD_STOPPED)
> +
> + if (set_running_thread (tp, running))
> any_started = 1;
> - tp->state = running ? THREAD_RUNNING : THREAD_STOPPED;
> }
> - if (any_started)
> - observer_notify_target_resumed (ptid);
> }
> else
> {
> - int started = 0;
> -
> tp = find_thread_ptid (ptid);
> - gdb_assert (tp);
> + gdb_assert (tp != NULL);
> gdb_assert (tp->state != THREAD_EXITED);
> - if (running && tp->state == THREAD_STOPPED)
> - started = 1;
> - tp->state = running ? THREAD_RUNNING : THREAD_STOPPED;
> - if (started)
> - observer_notify_target_resumed (ptid);
> + if (set_running_thread (tp, running))
> + any_started = 1;
> }
> + if (any_started)
> + observer_notify_target_resumed (ptid);
> }
>
> static int
> @@ -1008,9 +1026,8 @@ finish_thread_state (ptid_t ptid)
> continue;
> if (all || ptid_get_pid (ptid) == ptid_get_pid (tp->ptid))
> {
> - if (tp->executing && tp->state == THREAD_STOPPED)
> + if (set_running_thread (tp, tp->executing))
> any_started = 1;
> - tp->state = tp->executing ? THREAD_RUNNING : THREAD_STOPPED;
> }
> }
> }
> @@ -1020,9 +1037,8 @@ finish_thread_state (ptid_t ptid)
> gdb_assert (tp);
> if (tp->state != THREAD_EXITED)
> {
> - if (tp->executing && tp->state == THREAD_STOPPED)
> + if (set_running_thread (tp, tp->executing))
> any_started = 1;
> - tp->state = tp->executing ? THREAD_RUNNING : THREAD_STOPPED;
> }
> }
>
> --
> 1.9.3
>
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 06/17] Use keep_going in proceed and start_step_over too
2015-04-22 5:09 ` Doug Evans
@ 2015-04-22 22:22 ` Pedro Alves
0 siblings, 0 replies; 62+ messages in thread
From: Pedro Alves @ 2015-04-22 22:22 UTC (permalink / raw)
To: Doug Evans; +Cc: gdb-patches
On 04/22/2015 06:08 AM, Doug Evans wrote:
> Pedro Alves writes:
> > The main motivation of this patch is sharing more code between the
> > proceed (starting the inferior for the first time) and keep_going
> > (restarting the inferior after handling an event) paths and using the
> > step_over_chain queue now embedded in the thread_info object for
> > pending in-line step-overs too (instead of just for displaced
> > stepping).
>
> Hi.
> A couple of nits inline.
> grep for ====
>
Thanks.
> > + /* keep_going_pass skips the step-over if the breakpoint is no
> > + longer inserted. In all-stop, we want to keep looking for a
>
> ====
> missing word? "... keep looking for a thread" ?
Indeed.
> > - if (!tp->control.in_infcall)
> > - set_running (user_visible_resume_ptid (user_step), 1);
> > + if (debug_infrun)
> > + fprintf_unfiltered (gdb_stdlog,
> > + "Got placed in displaced stepping queue\n");
>
> ====
> Nit: Suggest rewording "displaced stepping queue".
> IIUC there is no displaced-stepping specific queue anymore.
Fixed:
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "Got placed in step-over queue\n");
> > -/* Called when we should continue running the inferior, because the
> > - current event doesn't cause a user visible stop. This does the
> > - resuming part; waiting for the next event is done elsewhere. */
> > +/* Like keep_going, but passes the signal to the inferior, even if the
> > + signal is set to nopass. */
> >
> > static void
> > -keep_going (struct execution_control_state *ecs)
> > +keep_going_pass (struct execution_control_state *ecs)
>
> ====
> For whatever weird reasons, "keep_going_pass" doesn't read very well to me.
> "pass" as in "pass/fail"?
> "pass" as in one of several passes, and this is the "keep going" one?
>
> Can you rename this to keep_going_pass_signal?
No problem. Done that.
Thanks,
Pedro Alves
^ permalink raw reply [flat|nested] 62+ messages in thread
* [PATCH v3 17/17] native Linux: enable always non-stop by default
2015-04-17 10:47 [PATCH v3 00/23] All-stop on top of non-stop Pedro Alves
` (8 preceding siblings ...)
2015-04-17 10:45 ` [PATCH v3 06/17] Use keep_going in proceed and start_step_over too Pedro Alves
@ 2015-04-17 10:47 ` Pedro Alves
2015-04-17 10:47 ` [PATCH v3 07/17] Misc switch_back_to_stepped_thread cleanups Pedro Alves
` (8 subsequent siblings)
18 siblings, 0 replies; 62+ messages in thread
From: Pedro Alves @ 2015-04-17 10:47 UTC (permalink / raw)
To: gdb-patches
The testsuite shows no regressions with this forced on, on:
- Native x86_64 Fedora 20, with and output "set displaced off".
- Native x86_64 Fedora 20, on top of x86 software single-step series.
- PPC64 Fedora 18.
- S/390 RHEL 7.1.
Let's try making it the default.
gdb/ChangeLog:
2015-04-17 Pedro Alves <palves@redhat.com>
* linux-nat.c (linux_nat_always_non_stop_p): Return 1.
---
gdb/linux-nat.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/gdb/linux-nat.c b/gdb/linux-nat.c
index d6fd40d..51ecdbc 100644
--- a/gdb/linux-nat.c
+++ b/gdb/linux-nat.c
@@ -4565,7 +4565,7 @@ linux_nat_supports_non_stop (struct target_ops *self)
static int
linux_nat_always_non_stop_p (struct target_ops *self)
{
- return 0;
+ return 1;
}
/* True if we want to support multi-process. To be removed when GDB
--
1.9.3
^ permalink raw reply [flat|nested] 62+ messages in thread* [PATCH v3 07/17] Misc switch_back_to_stepped_thread cleanups
2015-04-17 10:47 [PATCH v3 00/23] All-stop on top of non-stop Pedro Alves
` (9 preceding siblings ...)
2015-04-17 10:47 ` [PATCH v3 17/17] native Linux: enable always non-stop by default Pedro Alves
@ 2015-04-17 10:47 ` Pedro Alves
2015-04-21 9:50 ` Yao Qi
2015-04-22 5:23 ` Doug Evans
2015-04-17 10:47 ` [PATCH v3 01/17] Fix and test "checkpoint" in non-stop mode Pedro Alves
` (7 subsequent siblings)
18 siblings, 2 replies; 62+ messages in thread
From: Pedro Alves @ 2015-04-17 10:47 UTC (permalink / raw)
To: gdb-patches
Several misc cleanups that prepare the tail end of this function, the
part that actually re-resumes the stepped thread.
The most non-obvious would be the currently_stepping change, I guess.
That's because it isn't ever correct to pass step=1 to target_resume
on software single-step targets, and currently_stepping works at a
conceptual higher level, it returns step=true even on software step
targets. It doesn't really matter on hardware step targets, as the
breakpoint will be hit immediately, but it's just wrong on software
step targets. I tested it against my x86 software single-step branch,
and it indeed fixes failed assertions (that catch spurious
PTRACE_SINGLESTEP requests) there.
gdb/ChangeLog:
2015-04-17 Pedro Alves <palves@redhat.com>
* infrun.c (switch_back_to_stepped_thread): Use ecs->ptid instead
pf inferior_ptid. If the stepped thread vanished, return 0
instead of resuming here. Use reset_ecs. Print the prev_pc and
the current stop_pc in log message. Clear trap_expected if the
thread advanced. Don't pass currently_stepping to
do_target_resume.
---
gdb/infrun.c | 33 ++++++++++++++++++---------------
1 file changed, 18 insertions(+), 15 deletions(-)
diff --git a/gdb/infrun.c b/gdb/infrun.c
index beb9ea2..ce2be31 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -5702,7 +5702,7 @@ switch_back_to_stepped_thread (struct execution_control_state *ecs)
{
/* Ignore threads of processes we're not resuming. */
if (!sched_multi
- && ptid_get_pid (tp->ptid) != ptid_get_pid (inferior_ptid))
+ && ptid_get_pid (tp->ptid) != ptid_get_pid (ecs->ptid))
continue;
/* When stepping over a breakpoint, we lock all threads
@@ -5766,19 +5766,17 @@ switch_back_to_stepped_thread (struct execution_control_state *ecs)
"stepped thread, it has vanished\n");
delete_thread (tp->ptid);
- keep_going (ecs);
- return 1;
+ return 0;
}
if (debug_infrun)
fprintf_unfiltered (gdb_stdlog,
"infrun: switching back to stepped thread\n");
- ecs->event_thread = tp;
- ecs->ptid = tp->ptid;
- context_switch (ecs->ptid);
+ reset_ecs (ecs, tp);
+ switch_to_thread (tp->ptid);
- stop_pc = regcache_read_pc (get_thread_regcache (ecs->ptid));
+ stop_pc = regcache_read_pc (get_thread_regcache (tp->ptid));
frame = get_current_frame ();
gdbarch = get_frame_arch (frame);
@@ -5802,23 +5800,28 @@ switch_back_to_stepped_thread (struct execution_control_state *ecs)
if (debug_infrun)
fprintf_unfiltered (gdb_stdlog,
- "infrun: expected thread advanced also\n");
+ "infrun: expected thread advanced also "
+ "(%s -> %s)\n",
+ paddress (target_gdbarch (), tp->prev_pc),
+ paddress (target_gdbarch (), stop_pc));
/* Clear the info of the previous step-over, as it's no
- longer valid. It's what keep_going would do too, if
- we called it. Must do this before trying to insert
- the sss breakpoint, otherwise if we were previously
- trying to step over this exact address in another
- thread, the breakpoint ends up not installed. */
+ longer valid (if the thread was trying to step over a
+ breakpoint, it has already succeeded). It's what
+ keep_going would do too, if we called it. Do this
+ before trying to insert the sss breakpoint, otherwise
+ if we were previously trying to step over this exact
+ address in another thread, the breakpoint is
+ skipped. */
clear_step_over_info ();
+ tp->control.trap_expected = 0;
insert_single_step_breakpoint (get_frame_arch (frame),
get_frame_address_space (frame),
stop_pc);
resume_ptid = user_visible_resume_ptid (tp->control.stepping_command);
- do_target_resume (resume_ptid,
- currently_stepping (tp), GDB_SIGNAL_0);
+ do_target_resume (resume_ptid, 0, GDB_SIGNAL_0);
prepare_to_wait (ecs);
}
else
--
1.9.3
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 07/17] Misc switch_back_to_stepped_thread cleanups
2015-04-17 10:47 ` [PATCH v3 07/17] Misc switch_back_to_stepped_thread cleanups Pedro Alves
@ 2015-04-21 9:50 ` Yao Qi
2015-04-22 20:04 ` Pedro Alves
2015-04-22 5:23 ` Doug Evans
1 sibling, 1 reply; 62+ messages in thread
From: Yao Qi @ 2015-04-21 9:50 UTC (permalink / raw)
To: Pedro Alves; +Cc: gdb-patches
Pedro Alves <palves@redhat.com> writes:
> Several misc cleanups that prepare the tail end of this function, the
> part that actually re-resumes the stepped thread.
>
> The most non-obvious would be the currently_stepping change, I guess.
> That's because it isn't ever correct to pass step=1 to target_resume
> on software single-step targets, and currently_stepping works at a
> conceptual higher level, it returns step=true even on software step
> targets. It doesn't really matter on hardware step targets, as the
> breakpoint will be hit immediately, but it's just wrong on software
> step targets. I tested it against my x86 software single-step branch,
> and it indeed fixes failed assertions (that catch spurious
> PTRACE_SINGLESTEP requests) there.
>
> gdb/ChangeLog:
> 2015-04-17 Pedro Alves <palves@redhat.com>
>
> * infrun.c (switch_back_to_stepped_thread): Use ecs->ptid instead
> pf inferior_ptid. If the stepped thread vanished, return 0
^^ of
> instead of resuming here. Use reset_ecs. Print the prev_pc and
> the current stop_pc in log message. Clear trap_expected if the
> thread advanced. Don't pass currently_stepping to
> do_target_resume.
--
Yao (齐尧)
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 07/17] Misc switch_back_to_stepped_thread cleanups
2015-04-21 9:50 ` Yao Qi
@ 2015-04-22 20:04 ` Pedro Alves
0 siblings, 0 replies; 62+ messages in thread
From: Pedro Alves @ 2015-04-22 20:04 UTC (permalink / raw)
To: Yao Qi; +Cc: gdb-patches
On 04/21/2015 10:50 AM, Yao Qi wrote:
>> gdb/ChangeLog:
>> 2015-04-17 Pedro Alves <palves@redhat.com>
>>
>> * infrun.c (switch_back_to_stepped_thread): Use ecs->ptid instead
>> pf inferior_ptid. If the stepped thread vanished, return 0
> ^^ of
Thanks, fixed.
--
Pedro Alves
^ permalink raw reply [flat|nested] 62+ messages in thread
* Re: [PATCH v3 07/17] Misc switch_back_to_stepped_thread cleanups
2015-04-17 10:47 ` [PATCH v3 07/17] Misc switch_back_to_stepped_thread cleanups Pedro Alves
2015-04-21 9:50 ` Yao Qi
@ 2015-04-22 5:23 ` Doug Evans
2015-04-22 20:05 ` Pedro Alves
1 sibling, 1 reply; 62+ messages in thread
From: Doug Evans @ 2015-04-22 5:23 UTC (permalink / raw)
To: Pedro Alves; +Cc: gdb-patches
Pedro Alves writes:
> [...] it isn't ever correct to pass step=1 to target_resume
> on software single-step targets [...]
Sounds like a good thing to document in a comment or assert.
Can do?
> , and currently_stepping works at a
> conceptual higher level, it returns step=true even on software step
> targets. It doesn't really matter on hardware step targets, as the
> breakpoint will be hit immediately, but it's just wrong on software
> step targets. I tested it against my x86 software single-step branch,
> and it indeed fixes failed assertions (that catch spurious
> PTRACE_SINGLESTEP requests) there.
>
> gdb/ChangeLog:
> 2015-04-17 Pedro Alves <palves@redhat.com>
>
> * infrun.c (switch_back_to_stepped_thread): Use ecs->ptid instead
> pf inferior_ptid. If the stepped thread vanished, return 0
> instead of resuming here. Use reset_ecs. Print the prev_pc and
> the current stop_pc in log message. Clear trap_expected if the
> thread advanced. Don't pass currently_stepping to
> do_target_resume.
>...
^ permalink raw reply [flat|nested] 62+ messages in thread
* Re: [PATCH v3 07/17] Misc switch_back_to_stepped_thread cleanups
2015-04-22 5:23 ` Doug Evans
@ 2015-04-22 20:05 ` Pedro Alves
2015-04-28 20:28 ` Doug Evans
0 siblings, 1 reply; 62+ messages in thread
From: Pedro Alves @ 2015-04-22 20:05 UTC (permalink / raw)
To: Doug Evans; +Cc: gdb-patches
On 04/22/2015 06:21 AM, Doug Evans wrote:
> Pedro Alves writes:
> > [...] it isn't ever correct to pass step=1 to target_resume
> > on software single-step targets [...]
>
> Sounds like a good thing to document in a comment or assert.
Yeah, we were discussing adding an assertion on the gdbserver
side here:
https://sourceware.org/ml/gdb-patches/2015-04/msg00232.html
An assertion on the gdb side is complicated, and I'd rather that be
done separately. gdbarch_software_single_step_p() returning true
does not mean that we can't use hardware step. E.g., ppc installs
that hook for stepping past atomic regions. We can't assert based on
software single-step breakpoints inserted, as infrun.c uses those
even on targets that don't implement gdbarch_software_single_step_p.
We could add a new target_can_hardware_single_step method, which
ideally for remote targets would be based on the reply to the "vCont?"
probe packet, but gdbserver always includes "s" actions in the reply
even on targets that can't hardware step, and we can't just make it
not do that, since remote.c:remote_vcont_probe has:
/* If s, S, c, and C are not all supported, we can't use vCont. Clearing
BUF will make packet_ok disable the packet. */
if (!support_s || !support_S || !support_c || !support_C)
buf[0] = 0;
so even if we fixed mainline, newer gdbserver against older gdb
would be broken... So we need something else or in addition (probably
based on qSupported).
For comments, how about this?
diff --git i/gdb/infrun.c w/gdb/infrun.c
index 0bf1274..fbe12ab 100644
--- i/gdb/infrun.c
+++ w/gdb/infrun.c
@@ -5839,7 +5839,9 @@ switch_back_to_stepped_thread (struct execution_control_state *ecs)
return 0;
}
-/* Is thread TP in the middle of single-stepping? */
+/* Is thread TP in the middle of (software or hardware)
+ single-stepping? (Note the result of this function must never be
+ passed directly as target_resume's STEP parameter.) */
static int
currently_stepping (struct thread_info *tp)
diff --git i/gdb/target.h w/gdb/target.h
index 66bf91e..283f56f 100644
--- i/gdb/target.h
+++ w/gdb/target.h
@@ -1254,8 +1254,8 @@ extern void target_detach (const char *, int);
extern void target_disconnect (const char *, int);
/* Resume execution of the target process PTID (or a group of
- threads). STEP says whether to single-step or to run free; SIGGNAL
- is the signal to be given to the target, or GDB_SIGNAL_0 for no
+ threads). STEP says whether to hardware single-step or to run free;
+ SIGGNAL is the signal to be given to the target, or GDB_SIGNAL_0 for no
signal. The caller may not pass GDB_SIGNAL_DEFAULT. A specific
PTID means `step/resume only this process id'. A wildcard PTID
(all threads, or all threads of process) means `step/resume
Thanks,
Pedro Alves
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 07/17] Misc switch_back_to_stepped_thread cleanups
2015-04-22 20:05 ` Pedro Alves
@ 2015-04-28 20:28 ` Doug Evans
0 siblings, 0 replies; 62+ messages in thread
From: Doug Evans @ 2015-04-28 20:28 UTC (permalink / raw)
To: Pedro Alves; +Cc: gdb-patches
Pedro Alves writes:
> On 04/22/2015 06:21 AM, Doug Evans wrote:
> > Pedro Alves writes:
> > > [...] it isn't ever correct to pass step=1 to target_resume
> > > on software single-step targets [...]
> >
> > Sounds like a good thing to document in a comment or assert.
>
> Yeah, we were discussing adding an assertion on the gdbserver
> side here:
>
> https://sourceware.org/ml/gdb-patches/2015-04/msg00232.html
>
> An assertion on the gdb side is complicated, and I'd rather that be
> done separately. gdbarch_software_single_step_p() returning true
> does not mean that we can't use hardware step. E.g., ppc installs
> that hook for stepping past atomic regions. We can't assert based on
> software single-step breakpoints inserted, as infrun.c uses those
> even on targets that don't implement gdbarch_software_single_step_p.
> We could add a new target_can_hardware_single_step method, which
> ideally for remote targets would be based on the reply to the "vCont?"
> probe packet, but gdbserver always includes "s" actions in the reply
> even on targets that can't hardware step, and we can't just make it
> not do that, since remote.c:remote_vcont_probe has:
>
> /* If s, S, c, and C are not all supported, we can't use vCont. Clearing
> BUF will make packet_ok disable the packet. */
> if (!support_s || !support_S || !support_c || !support_C)
> buf[0] = 0;
>
> so even if we fixed mainline, newer gdbserver against older gdb
> would be broken... So we need something else or in addition (probably
> based on qSupported).
>
> For comments, how about this?
>
> diff --git i/gdb/infrun.c w/gdb/infrun.c
> index 0bf1274..fbe12ab 100644
> --- i/gdb/infrun.c
> +++ w/gdb/infrun.c
> @@ -5839,7 +5839,9 @@ switch_back_to_stepped_thread (struct execution_control_state *ecs)
> return 0;
> }
>
> -/* Is thread TP in the middle of single-stepping? */
> +/* Is thread TP in the middle of (software or hardware)
> + single-stepping? (Note the result of this function must never be
> + passed directly as target_resume's STEP parameter.) */
>
> static int
> currently_stepping (struct thread_info *tp)
> diff --git i/gdb/target.h w/gdb/target.h
> index 66bf91e..283f56f 100644
> --- i/gdb/target.h
> +++ w/gdb/target.h
> @@ -1254,8 +1254,8 @@ extern void target_detach (const char *, int);
> extern void target_disconnect (const char *, int);
>
> /* Resume execution of the target process PTID (or a group of
> - threads). STEP says whether to single-step or to run free; SIGGNAL
> - is the signal to be given to the target, or GDB_SIGNAL_0 for no
> + threads). STEP says whether to hardware single-step or to run free;
> + SIGGNAL is the signal to be given to the target, or GDB_SIGNAL_0 for no
> signal. The caller may not pass GDB_SIGNAL_DEFAULT. A specific
> PTID means `step/resume only this process id'. A wildcard PTID
> (all threads, or all threads of process) means `step/resume
>
> Thanks,
> Pedro Alves
Works for me.
Thanks!
^ permalink raw reply [flat|nested] 62+ messages in thread
* [PATCH v3 01/17] Fix and test "checkpoint" in non-stop mode
2015-04-17 10:47 [PATCH v3 00/23] All-stop on top of non-stop Pedro Alves
` (10 preceding siblings ...)
2015-04-17 10:47 ` [PATCH v3 07/17] Misc switch_back_to_stepped_thread cleanups Pedro Alves
@ 2015-04-17 10:47 ` Pedro Alves
2015-04-21 2:36 ` Doug Evans
2015-04-17 10:52 ` [PATCH v3 12/17] Fix interrupt-noterm.exp on targets always in non-stop Pedro Alves
` (6 subsequent siblings)
18 siblings, 1 reply; 62+ messages in thread
From: Pedro Alves @ 2015-04-17 10:47 UTC (permalink / raw)
To: gdb-patches
Letting a "checkpoint" run to exit with "set non-stop on" behaves
differently compared to the default all-stop mode ("set non-stop
off").
Currently, in non-stop mode:
(gdb) start
Temporary breakpoint 1 at 0x40086b: file src/gdb/testsuite/gdb.base/checkpoint.c, line 28.
Starting program: build/gdb/testsuite/gdb.base/checkpoint
Temporary breakpoint 1, main () at src/gdb/testsuite/gdb.base/checkpoint.c:28
28 char *tmp = &linebuf[0];
(gdb) checkpoint
checkpoint 1: fork returned pid 24948.
(gdb) c
Continuing.
Copy complete.
Deleting copy.
[Inferior 1 (process 24944) exited normally]
[Switching to process 24948]
(gdb) info threads
Id Target Id Frame
1 process 24948 "checkpoint" (running)
No selected thread. See `help thread'.
(gdb) c
The program is not being run.
(gdb)
Two issues above:
1. Thread 1 got stuck in "(running)" state (it isn't really running)
2. While checkpoints try to preserve the illusion that the thread is
still the same when the process exits, GDB switched to "No thread
selected." instead of staying with thread 1 selected.
Problem #1 is caused by handle_inferior_event and normal_stop not
considering that when a
TARGET_WAITKIND_SIGNALLED/TARGET_WAITKIND_EXITED event is reported,
and the inferior is mourned, the target may still have execution.
Problem #2 is caused by the make_cleanup_restore_current_thread
cleanup installed by fetch_inferior_event not being able to find the
original thread 1's ptid in the thread list, thus not being able to
restore thread 1 as selected thread. The fix is to make the cleanup
installed by make_cleanup_restore_current_thread aware of thread ptid
changes, by installing a thread_ptid_changed observer that adjusts the
cleanup's data.
After the patch, we get the same in all-stop and non-stop modes:
(gdb) c
Continuing.
Copy complete.
Deleting copy.
[Inferior 1 (process 25109) exited normally]
[Switching to process 25113]
(gdb) info threads
Id Target Id Frame
* 1 process 25113 "checkpoint" main () at src/gdb/testsuite/gdb.base/checkpoint.c:28
(gdb)
Turns out the whole checkpoints.exp file can run in non-stop mode
unmodified. I thought of moving most of the test file's contents to a
procedure that can be called twice, once in non-stop mode and another
in all-stop mode. But then, the test already takes over 30 seconds to
run on my machine, so I thought it'd be nicer to run all-stop and
non-stop mode in parallel. Thus I added a new checkpoint-ns.exp file
that just sources checkpoint.exp, and sets a knob that checkpoint.exp
reads to know it should test non-stop mode. No other test in the tree
currently uses this mechanism, but I can't see a reason we shouldn't
do this.
gdb/ChangeLog:
2015-04-17 Pedro Alves <palves@redhat.com>
* infrun.c (handle_inferior_event): If we get
TARGET_WAITKIND_SIGNALLED or TARGET_WAITKIND_EXITED in non-stop
mode, mark all threads of the exiting process as not-executing.
(normal_stop): If we get TARGET_WAITKIND_SIGNALLED or
TARGET_WAITKIND_EXITED in non-stop mode, finish all threads of the
exiting process, if inferior_ptid still points at a process.
* thread.c (struct current_thread_cleanup) <next>: New field.
(current_thread_cleanup_chain): New global.
(restore_current_thread_ptid_changed): New function.
(restore_current_thread_cleanup_dtor): Remove the cleanup from the
current_thread_cleanup_chain list.
(make_cleanup_restore_current_thread): Add the cleanup data to the
current_thread_cleanup_chain list.
(_initialize_thread): Install restore_current_thread_ptid_changed
as thread_ptid_changed observer.
gdb/testsuite/ChangeLog:
2015-04-17 Pedro Alves <palves@redhat.com>
* gdb.base/checkpoint-ns.exp: New file.
* gdb.base/checkpoint.exp (checkpoint_non_stop): New input
variable. If set, test in non-stop mode. Pass explicit
"checkpoint.c" to standard_testfile.
v3:
No changes.
---
gdb/infrun.c | 31 +++++++++++++++++++++-----
gdb/testsuite/gdb.base/checkpoint-ns.exp | 27 +++++++++++++++++++++++
gdb/testsuite/gdb.base/checkpoint.exp | 31 +++++++++++++++++++++-----
gdb/thread.c | 38 ++++++++++++++++++++++++++++++++
4 files changed, 117 insertions(+), 10 deletions(-)
create mode 100644 gdb/testsuite/gdb.base/checkpoint-ns.exp
diff --git a/gdb/infrun.c b/gdb/infrun.c
index 7870f70..9c1fdc5 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -3804,8 +3804,18 @@ handle_inferior_event (struct execution_control_state *ecs)
any other process were left running. */
if (!non_stop)
set_executing (minus_one_ptid, 0);
- else if (ecs->ws.kind != TARGET_WAITKIND_SIGNALLED
- && ecs->ws.kind != TARGET_WAITKIND_EXITED)
+ else if (ecs->ws.kind == TARGET_WAITKIND_SIGNALLED
+ && ecs->ws.kind == TARGET_WAITKIND_EXITED)
+ {
+ ptid_t pid_ptid;
+
+ /* Some targets still have execution when a process exits.
+ E.g., for "checkpoint", when when a fork exits and is
+ mourned, linux-fork.c switches to another fork. */
+ pid_ptid = pid_to_ptid (ptid_get_pid (ecs->ptid));
+ set_executing (pid_ptid, 0);
+ }
+ else
set_executing (ecs->ptid, 0);
switch (ecs->ws.kind)
@@ -6550,6 +6560,7 @@ normal_stop (void)
struct target_waitstatus last;
ptid_t last_ptid;
struct cleanup *old_chain = make_cleanup (null_cleanup, NULL);
+ ptid_t pid_ptid;
get_last_target_status (&last_ptid, &last);
@@ -6559,9 +6570,19 @@ normal_stop (void)
here, so do this before any filtered output. */
if (!non_stop)
make_cleanup (finish_thread_state_cleanup, &minus_one_ptid);
- else if (last.kind != TARGET_WAITKIND_SIGNALLED
- && last.kind != TARGET_WAITKIND_EXITED
- && last.kind != TARGET_WAITKIND_NO_RESUMED)
+ else if (last.kind == TARGET_WAITKIND_SIGNALLED
+ || last.kind == TARGET_WAITKIND_EXITED)
+ {
+ /* Some targets still have execution when a process exits.
+ E.g., for "checkpoint", when when a fork exits and is
+ mourned, linux-fork.c switches to another fork. */
+ if (!ptid_equal (inferior_ptid, null_ptid))
+ {
+ pid_ptid = pid_to_ptid (ptid_get_pid (inferior_ptid));
+ make_cleanup (finish_thread_state_cleanup, &pid_ptid);
+ }
+ }
+ else if (last.kind != TARGET_WAITKIND_NO_RESUMED)
make_cleanup (finish_thread_state_cleanup, &inferior_ptid);
/* As we're presenting a stop, and potentially removing breakpoints,
diff --git a/gdb/testsuite/gdb.base/checkpoint-ns.exp b/gdb/testsuite/gdb.base/checkpoint-ns.exp
new file mode 100644
index 0000000..03a1747
--- /dev/null
+++ b/gdb/testsuite/gdb.base/checkpoint-ns.exp
@@ -0,0 +1,27 @@
+# Copyright 2015 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 checkpoint and restart in non-stop mode.
+
+# We drive non-stop mode from a separate file because the whole test
+# takes a while to run. This way, we can test both modes in parallel.
+
+# checkpoint.exp reads this variable.
+set checkpoint_non_stop 1
+
+source $srcdir/$subdir/checkpoint.exp
+
+# Unset to avoid problems with non-parallel mode.
+unset checkpoint_non_stop
diff --git a/gdb/testsuite/gdb.base/checkpoint.exp b/gdb/testsuite/gdb.base/checkpoint.exp
index 6d94ab6..c297b50 100644
--- a/gdb/testsuite/gdb.base/checkpoint.exp
+++ b/gdb/testsuite/gdb.base/checkpoint.exp
@@ -13,6 +13,20 @@
# 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 tests gdb checkpoint and restart.
+#
+
+# Note this file is also sourced by checkpoint-ns.exp to run all the
+# tests below in non-stop mode. That's driven by a separate file so
+# testing both all-stop and non-stop can be done in parallel.
+
+# This is the variable checkpoint-ns.exp sets to request non-stop
+# mode. When not set, default to all-stop mode.
+if {![info exists checkpoint_non_stop]} {
+ set checkpoint_non_stop 0
+}
+
if { [is_remote target] || ![isnative] } then {
continue
}
@@ -24,8 +38,10 @@ if {![istarget "*-*-linux*"]} then {
continue
}
-
-standard_testfile .c
+# Must name the source file explicitly, otherwise when driven by
+# checkpoints-ns.exp, we'd try compiling checkpoints-ns.c, which
+# doesn't exist.
+standard_testfile checkpoint.c
set pi_txt [gdb_remote_download host ${srcdir}/${subdir}/pi.txt]
if {[is_remote host]} {
@@ -42,9 +58,9 @@ if {[prepare_for_testing ${testfile}.exp $testfile $srcfile \
global gdb_prompt
-#
-# This tests gdb checkpoint and restart.
-#
+if {$checkpoint_non_stop} {
+ gdb_test_no_output "set non-stop on"
+}
runto_main
set break1_loc [gdb_get_line_number "breakpoint 1"]
@@ -327,6 +343,11 @@ gdb_start
gdb_reinitialize_dir $srcdir/$subdir
gdb_load ${binfile}
+if {$checkpoint_non_stop} {
+ gdb_test_no_output "set non-stop on" \
+ "set non-stop on, for many checkpoints"
+}
+
runto_main
gdb_breakpoint $break1_loc
diff --git a/gdb/thread.c b/gdb/thread.c
index 23dfcc9..46b5947 100644
--- a/gdb/thread.c
+++ b/gdb/thread.c
@@ -1279,8 +1279,16 @@ restore_selected_frame (struct frame_id a_frame_id, int frame_level)
}
}
+/* Data used by the cleanup installed by
+ 'make_cleanup_restore_current_thread'. */
+
struct current_thread_cleanup
{
+ /* Next in list of currently installed 'struct
+ current_thread_cleanup' cleanups. See
+ 'current_thread_cleanup_chain' below. */
+ struct current_thread_cleanup *next;
+
ptid_t inferior_ptid;
struct frame_id selected_frame_id;
int selected_frame_level;
@@ -1289,6 +1297,29 @@ struct current_thread_cleanup
int was_removable;
};
+/* A chain of currently installed 'struct current_thread_cleanup'
+ cleanups. Restoring the previously selected thread looks up the
+ old thread in the thread list by ptid. If the thread changes ptid,
+ we need to update the cleanup's thread structure so the look up
+ succeeds. */
+static struct current_thread_cleanup *current_thread_cleanup_chain;
+
+/* A thread_ptid_changed observer. Update all currently installed
+ current_thread_cleanup cleanups that want to switch back to
+ OLD_PTID to switch back to NEW_PTID instead. */
+
+static void
+restore_current_thread_ptid_changed (ptid_t old_ptid, ptid_t new_ptid)
+{
+ struct current_thread_cleanup *it;
+
+ for (it = current_thread_cleanup_chain; it != NULL; it = it->next)
+ {
+ if (ptid_equal (it->inferior_ptid, old_ptid))
+ it->inferior_ptid = new_ptid;
+ }
+}
+
static void
do_restore_current_thread_cleanup (void *arg)
{
@@ -1329,6 +1360,8 @@ restore_current_thread_cleanup_dtor (void *arg)
struct thread_info *tp;
struct inferior *inf;
+ current_thread_cleanup_chain = current_thread_cleanup_chain->next;
+
tp = find_thread_ptid (old->inferior_ptid);
if (tp)
tp->refcount--;
@@ -1362,6 +1395,9 @@ make_cleanup_restore_current_thread (void)
old->inf_id = current_inferior ()->num;
old->was_removable = current_inferior ()->removable;
+ old->next = current_thread_cleanup_chain;
+ current_thread_cleanup_chain = old;
+
if (!ptid_equal (inferior_ptid, null_ptid))
{
old->was_stopped = is_stopped (inferior_ptid);
@@ -1815,4 +1851,6 @@ Show printing of thread events (such as thread start and exit)."), NULL,
&setprintlist, &showprintlist);
create_internalvar_type_lazy ("_thread", &thread_funcs, NULL);
+
+ observer_attach_thread_ptid_changed (restore_current_thread_ptid_changed);
}
--
1.9.3
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 01/17] Fix and test "checkpoint" in non-stop mode
2015-04-17 10:47 ` [PATCH v3 01/17] Fix and test "checkpoint" in non-stop mode Pedro Alves
@ 2015-04-21 2:36 ` Doug Evans
2015-04-22 17:48 ` Pedro Alves
0 siblings, 1 reply; 62+ messages in thread
From: Doug Evans @ 2015-04-21 2:36 UTC (permalink / raw)
To: Pedro Alves; +Cc: gdb-patches
Pedro Alves writes:
> Letting a "checkpoint" run to exit with "set non-stop on" behaves
> differently compared to the default all-stop mode ("set non-stop
> off").
>
> Currently, in non-stop mode:
>
> (gdb) start
> Temporary breakpoint 1 at 0x40086b: file src/gdb/testsuite/gdb.base/checkpoint.c, line 28.
> Starting program: build/gdb/testsuite/gdb.base/checkpoint
>
> Temporary breakpoint 1, main () at src/gdb/testsuite/gdb.base/checkpoint.c:28
> 28 char *tmp = &linebuf[0];
> (gdb) checkpoint
> checkpoint 1: fork returned pid 24948.
> (gdb) c
> Continuing.
> Copy complete.
> Deleting copy.
> [Inferior 1 (process 24944) exited normally]
> [Switching to process 24948]
> (gdb) info threads
> Id Target Id Frame
> 1 process 24948 "checkpoint" (running)
>
> No selected thread. See `help thread'.
> (gdb) c
> The program is not being run.
> (gdb)
>
> Two issues above:
>
> 1. Thread 1 got stuck in "(running)" state (it isn't really running)
>
> 2. While checkpoints try to preserve the illusion that the thread is
> still the same when the process exits, GDB switched to "No thread
> selected." instead of staying with thread 1 selected.
>
> Problem #1 is caused by handle_inferior_event and normal_stop not
> considering that when a
> TARGET_WAITKIND_SIGNALLED/TARGET_WAITKIND_EXITED event is reported,
> and the inferior is mourned, the target may still have execution.
Hi.
What does "target may still have execution" mean here?
Is it that there can be another inferior running?
> Problem #2 is caused by the make_cleanup_restore_current_thread
> cleanup installed by fetch_inferior_event not being able to find the
> original thread 1's ptid in the thread list, thus not being able to
> restore thread 1 as selected thread. The fix is to make the cleanup
> installed by make_cleanup_restore_current_thread aware of thread ptid
> changes, by installing a thread_ptid_changed observer that adjusts the
> cleanup's data.
I'm guessing this is less hacky than it sounds :-),
but it does make me want to understand why this only comes up
now since threads/inferiors have always been able to "go away"
while gdb is doing something.
> After the patch, we get the same in all-stop and non-stop modes:
>
> (gdb) c
> Continuing.
> Copy complete.
> Deleting copy.
> [Inferior 1 (process 25109) exited normally]
> [Switching to process 25113]
> (gdb) info threads
> Id Target Id Frame
> * 1 process 25113 "checkpoint" main () at src/gdb/testsuite/gdb.base/checkpoint.c:28
> (gdb)
>
> Turns out the whole checkpoints.exp file can run in non-stop mode
> unmodified. I thought of moving most of the test file's contents to a
> procedure that can be called twice, once in non-stop mode and another
> in all-stop mode. But then, the test already takes over 30 seconds to
> run on my machine, so I thought it'd be nicer to run all-stop and
> non-stop mode in parallel. Thus I added a new checkpoint-ns.exp file
> that just sources checkpoint.exp, and sets a knob that checkpoint.exp
> reads to know it should test non-stop mode. No other test in the tree
> currently uses this mechanism, but I can't see a reason we shouldn't
> do this.
Re: checkpoint-ns.exp:
If we're going to have one "make check" involve both all-stop
and non-stop testing (as opposed to doing one "make check" with all-stop
and another one with non-stop), I'd rather see a more table-driven approach
(the details don't matter to me much other than I'd rather have one file
than a proliferation of foo-ns.exp files).
Parallelizing that may involve an extra step, but that's ok.
>
> gdb/ChangeLog:
> 2015-04-17 Pedro Alves <palves@redhat.com>
>
> * infrun.c (handle_inferior_event): If we get
> TARGET_WAITKIND_SIGNALLED or TARGET_WAITKIND_EXITED in non-stop
> mode, mark all threads of the exiting process as not-executing.
> (normal_stop): If we get TARGET_WAITKIND_SIGNALLED or
> TARGET_WAITKIND_EXITED in non-stop mode, finish all threads of the
> exiting process, if inferior_ptid still points at a process.
> * thread.c (struct current_thread_cleanup) <next>: New field.
> (current_thread_cleanup_chain): New global.
> (restore_current_thread_ptid_changed): New function.
> (restore_current_thread_cleanup_dtor): Remove the cleanup from the
> current_thread_cleanup_chain list.
> (make_cleanup_restore_current_thread): Add the cleanup data to the
> current_thread_cleanup_chain list.
> (_initialize_thread): Install restore_current_thread_ptid_changed
> as thread_ptid_changed observer.
>
> gdb/testsuite/ChangeLog:
> 2015-04-17 Pedro Alves <palves@redhat.com>
>
> * gdb.base/checkpoint-ns.exp: New file.
> * gdb.base/checkpoint.exp (checkpoint_non_stop): New input
> variable. If set, test in non-stop mode. Pass explicit
> "checkpoint.c" to standard_testfile.
>
> v3:
> No changes.
> ---
> gdb/infrun.c | 31 +++++++++++++++++++++-----
> gdb/testsuite/gdb.base/checkpoint-ns.exp | 27 +++++++++++++++++++++++
> gdb/testsuite/gdb.base/checkpoint.exp | 31 +++++++++++++++++++++-----
> gdb/thread.c | 38 ++++++++++++++++++++++++++++++++
> 4 files changed, 117 insertions(+), 10 deletions(-)
> create mode 100644 gdb/testsuite/gdb.base/checkpoint-ns.exp
>
> diff --git a/gdb/infrun.c b/gdb/infrun.c
> index 7870f70..9c1fdc5 100644
> --- a/gdb/infrun.c
> +++ b/gdb/infrun.c
> @@ -3804,8 +3804,18 @@ handle_inferior_event (struct execution_control_state *ecs)
> any other process were left running. */
> if (!non_stop)
> set_executing (minus_one_ptid, 0);
> - else if (ecs->ws.kind != TARGET_WAITKIND_SIGNALLED
> - && ecs->ws.kind != TARGET_WAITKIND_EXITED)
> + else if (ecs->ws.kind == TARGET_WAITKIND_SIGNALLED
> + && ecs->ws.kind == TARGET_WAITKIND_EXITED)
pasto: s/&&/||/
> + {
> + ptid_t pid_ptid;
> +
> + /* Some targets still have execution when a process exits.
> + E.g., for "checkpoint", when when a fork exits and is
s/when when/when/
> + mourned, linux-fork.c switches to another fork. */
> + pid_ptid = pid_to_ptid (ptid_get_pid (ecs->ptid));
> + set_executing (pid_ptid, 0);
This is a bit confusing. I'm guessing ecs->ptid is for the inferior
that just exited/signalled, but we're not changing pids here.
I'm guessing we'll mourn the inferior later, but IWBN to elaborate
on the comment a little, e.g., to say we always need to call set_executing
here, even if the inferior exited/signalled (which the code didn't
previously do). But it's not clear why "switches to another fork" comes
into play here.
> + }
> + else
> set_executing (ecs->ptid, 0);
>
> switch (ecs->ws.kind)
> @@ -6550,6 +6560,7 @@ normal_stop (void)
> struct target_waitstatus last;
> ptid_t last_ptid;
> struct cleanup *old_chain = make_cleanup (null_cleanup, NULL);
> + ptid_t pid_ptid;
>
> get_last_target_status (&last_ptid, &last);
>
> @@ -6559,9 +6570,19 @@ normal_stop (void)
> here, so do this before any filtered output. */
> if (!non_stop)
> make_cleanup (finish_thread_state_cleanup, &minus_one_ptid);
> - else if (last.kind != TARGET_WAITKIND_SIGNALLED
> - && last.kind != TARGET_WAITKIND_EXITED
> - && last.kind != TARGET_WAITKIND_NO_RESUMED)
> + else if (last.kind == TARGET_WAITKIND_SIGNALLED
> + || last.kind == TARGET_WAITKIND_EXITED)
> + {
> + /* Some targets still have execution when a process exits.
> + E.g., for "checkpoint", when when a fork exits and is
s/when when/when/
> + mourned, linux-fork.c switches to another fork. */
> + if (!ptid_equal (inferior_ptid, null_ptid))
> + {
> + pid_ptid = pid_to_ptid (ptid_get_pid (inferior_ptid));
> + make_cleanup (finish_thread_state_cleanup, &pid_ptid);
> + }
> + }
> + else if (last.kind != TARGET_WAITKIND_NO_RESUMED)
> make_cleanup (finish_thread_state_cleanup, &inferior_ptid);
>
> /* As we're presenting a stop, and potentially removing breakpoints,
> diff --git a/gdb/testsuite/gdb.base/checkpoint-ns.exp b/gdb/testsuite/gdb.base/checkpoint-ns.exp
> new file mode 100644
> index 0000000..03a1747
> --- /dev/null
> +++ b/gdb/testsuite/gdb.base/checkpoint-ns.exp
> @@ -0,0 +1,27 @@
> +# Copyright 2015 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 checkpoint and restart in non-stop mode.
> +
> +# We drive non-stop mode from a separate file because the whole test
> +# takes a while to run. This way, we can test both modes in parallel.
> +
> +# checkpoint.exp reads this variable.
> +set checkpoint_non_stop 1
> +
> +source $srcdir/$subdir/checkpoint.exp
> +
> +# Unset to avoid problems with non-parallel mode.
> +unset checkpoint_non_stop
> diff --git a/gdb/testsuite/gdb.base/checkpoint.exp b/gdb/testsuite/gdb.base/checkpoint.exp
> index 6d94ab6..c297b50 100644
> --- a/gdb/testsuite/gdb.base/checkpoint.exp
> +++ b/gdb/testsuite/gdb.base/checkpoint.exp
> @@ -13,6 +13,20 @@
> # 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 tests gdb checkpoint and restart.
> +#
> +
> +# Note this file is also sourced by checkpoint-ns.exp to run all the
> +# tests below in non-stop mode. That's driven by a separate file so
> +# testing both all-stop and non-stop can be done in parallel.
> +
> +# This is the variable checkpoint-ns.exp sets to request non-stop
> +# mode. When not set, default to all-stop mode.
> +if {![info exists checkpoint_non_stop]} {
> + set checkpoint_non_stop 0
> +}
> +
> if { [is_remote target] || ![isnative] } then {
> continue
> }
> @@ -24,8 +38,10 @@ if {![istarget "*-*-linux*"]} then {
> continue
> }
>
> -
> -standard_testfile .c
> +# Must name the source file explicitly, otherwise when driven by
> +# checkpoints-ns.exp, we'd try compiling checkpoints-ns.c, which
> +# doesn't exist.
> +standard_testfile checkpoint.c
>
> set pi_txt [gdb_remote_download host ${srcdir}/${subdir}/pi.txt]
> if {[is_remote host]} {
> @@ -42,9 +58,9 @@ if {[prepare_for_testing ${testfile}.exp $testfile $srcfile \
>
> global gdb_prompt
>
> -#
> -# This tests gdb checkpoint and restart.
> -#
> +if {$checkpoint_non_stop} {
> + gdb_test_no_output "set non-stop on"
> +}
>
> runto_main
> set break1_loc [gdb_get_line_number "breakpoint 1"]
> @@ -327,6 +343,11 @@ gdb_start
> gdb_reinitialize_dir $srcdir/$subdir
> gdb_load ${binfile}
>
> +if {$checkpoint_non_stop} {
> + gdb_test_no_output "set non-stop on" \
> + "set non-stop on, for many checkpoints"
> +}
> +
> runto_main
> gdb_breakpoint $break1_loc
>
> diff --git a/gdb/thread.c b/gdb/thread.c
> index 23dfcc9..46b5947 100644
> --- a/gdb/thread.c
> +++ b/gdb/thread.c
> @@ -1279,8 +1279,16 @@ restore_selected_frame (struct frame_id a_frame_id, int frame_level)
> }
> }
>
> +/* Data used by the cleanup installed by
> + 'make_cleanup_restore_current_thread'. */
> +
> struct current_thread_cleanup
> {
> + /* Next in list of currently installed 'struct
> + current_thread_cleanup' cleanups. See
> + 'current_thread_cleanup_chain' below. */
> + struct current_thread_cleanup *next;
> +
> ptid_t inferior_ptid;
> struct frame_id selected_frame_id;
> int selected_frame_level;
> @@ -1289,6 +1297,29 @@ struct current_thread_cleanup
> int was_removable;
> };
>
> +/* A chain of currently installed 'struct current_thread_cleanup'
> + cleanups. Restoring the previously selected thread looks up the
> + old thread in the thread list by ptid. If the thread changes ptid,
> + we need to update the cleanup's thread structure so the look up
> + succeeds. */
> +static struct current_thread_cleanup *current_thread_cleanup_chain;
> +
> +/* A thread_ptid_changed observer. Update all currently installed
> + current_thread_cleanup cleanups that want to switch back to
> + OLD_PTID to switch back to NEW_PTID instead. */
> +
> +static void
> +restore_current_thread_ptid_changed (ptid_t old_ptid, ptid_t new_ptid)
> +{
> + struct current_thread_cleanup *it;
> +
> + for (it = current_thread_cleanup_chain; it != NULL; it = it->next)
> + {
> + if (ptid_equal (it->inferior_ptid, old_ptid))
> + it->inferior_ptid = new_ptid;
> + }
> +}
> +
> static void
> do_restore_current_thread_cleanup (void *arg)
> {
> @@ -1329,6 +1360,8 @@ restore_current_thread_cleanup_dtor (void *arg)
> struct thread_info *tp;
> struct inferior *inf;
>
> + current_thread_cleanup_chain = current_thread_cleanup_chain->next;
> +
> tp = find_thread_ptid (old->inferior_ptid);
> if (tp)
> tp->refcount--;
> @@ -1362,6 +1395,9 @@ make_cleanup_restore_current_thread (void)
> old->inf_id = current_inferior ()->num;
> old->was_removable = current_inferior ()->removable;
>
> + old->next = current_thread_cleanup_chain;
> + current_thread_cleanup_chain = old;
> +
> if (!ptid_equal (inferior_ptid, null_ptid))
> {
> old->was_stopped = is_stopped (inferior_ptid);
> @@ -1815,4 +1851,6 @@ Show printing of thread events (such as thread start and exit)."), NULL,
> &setprintlist, &showprintlist);
>
> create_internalvar_type_lazy ("_thread", &thread_funcs, NULL);
> +
> + observer_attach_thread_ptid_changed (restore_current_thread_ptid_changed);
> }
> --
> 1.9.3
>
--
/dje
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 01/17] Fix and test "checkpoint" in non-stop mode
2015-04-21 2:36 ` Doug Evans
@ 2015-04-22 17:48 ` Pedro Alves
2015-04-28 18:18 ` Doug Evans
0 siblings, 1 reply; 62+ messages in thread
From: Pedro Alves @ 2015-04-22 17:48 UTC (permalink / raw)
To: Doug Evans; +Cc: gdb-patches
Hi Doug,
Thanks for the review.
On 04/21/2015 03:35 AM, Doug Evans wrote:
> Pedro Alves writes:
> > Letting a "checkpoint" run to exit with "set non-stop on" behaves
> > differently compared to the default all-stop mode ("set non-stop
> > off").
> >
> > Currently, in non-stop mode:
> >
> > (gdb) start
> > Temporary breakpoint 1 at 0x40086b: file src/gdb/testsuite/gdb.base/checkpoint.c, line 28.
> > Starting program: build/gdb/testsuite/gdb.base/checkpoint
> >
> > Temporary breakpoint 1, main () at src/gdb/testsuite/gdb.base/checkpoint.c:28
> > 28 char *tmp = &linebuf[0];
> > (gdb) checkpoint
> > checkpoint 1: fork returned pid 24948.
> > (gdb) c
> > Continuing.
> > Copy complete.
> > Deleting copy.
> > [Inferior 1 (process 24944) exited normally]
> > [Switching to process 24948]
> > (gdb) info threads
> > Id Target Id Frame
> > 1 process 24948 "checkpoint" (running)
> >
> > No selected thread. See `help thread'.
> > (gdb) c
> > The program is not being run.
> > (gdb)
> >
> > Two issues above:
> >
> > 1. Thread 1 got stuck in "(running)" state (it isn't really running)
> >
> > 2. While checkpoints try to preserve the illusion that the thread is
> > still the same when the process exits, GDB switched to "No thread
> > selected." instead of staying with thread 1 selected.
> >
> > Problem #1 is caused by handle_inferior_event and normal_stop not
> > considering that when a
> > TARGET_WAITKIND_SIGNALLED/TARGET_WAITKIND_EXITED event is reported,
> > and the inferior is mourned, the target may still have execution.
>
> Hi.
>
> What does "target may still have execution" mean here?
That "target_has_execution" returns true. (But I
extended/changed the comment, see below.)
> Is it that there can be another inferior running?
It's the same inferior, but a different fork/process.
The checkpoints that the "checkpoint" code makes are really
forked processes. This is done by linux-fork.c. Each fork
naturally is a different process, but that code tries to hide
that from the user and from core gdb. When the user switches
to a different checkpoint, the inferior is still the same,
though behind the scenes, the process/thread associated with
the inferior changed:
(top-gdb) checkpoint
checkpoint 1: fork returned pid 19228.
(top-gdb) info checkpoints
1 process 19228 at 0x4677bf, file /home/pedro/gdb/mygit/src/gdb/gdb.c, line 28
* 0 Thread 0x7ffff7c2c7c0 (LWP 19224) (main process) at 0x4677bf, file /home/pedro/gdb/mygit/src/gdb/gdb.c, line 28
(top-gdb) info inferiors
Num Description Executable
* 1 process 19224 /home/pedro/gdb/mygit/build/gdb/gdb
(top-gdb) restart 1
Switching to Thread 0x7ffff7c2c7c0 (LWP 19228)
#0 main (argc=1, argv=0x7fffffffd8f8) at /home/pedro/gdb/mygit/src/gdb/gdb.c:28
28 memset (&args, 0, sizeof args);
(top-gdb) info inferiors
Num Description Executable
* 1 process 19228 /home/pedro/gdb/mygit/build/gdb/gdb
(top-gdb)
So if you let one of the checkpoints run to exit, gdb switches
to another checkpoint:
(gdb) info checkpoints
1 process 19265 at 0x4004cf, file main.c, line 5
* 0 process 19264 (main process) at 0x4004cf, file main.c, line 5
(gdb) c
Continuing.
[Inferior 1 (process 19264) exited normally]
[Switching to process 19265]
(gdb)
(gdb) info threads
Id Target Id Frame
* 1 process 19265 "main" main (argc=1, argv=0x7fffffffd908) at main.c:5
again, it's still the same gdb thread/inferior object.
That is done at target_mourn_inferior time, in
linux-fork.c:linux_fork_mourn_inferior:
/* The current inferior_ptid has exited, but there are other viable
forks to debug. Delete the exiting one and context-switch to the
first available. */
void
linux_fork_mourn_inferior (void)
{
which ends up calling:
void
linux_nat_switch_fork (ptid_t new_ptid)
{
struct lwp_info *lp;
purge_lwp_list (ptid_get_pid (inferior_ptid));
lp = add_lwp (new_ptid);
lp->stopped = 1;
/* This changes the thread's ptid while preserving the gdb thread
num. Also changes the inferior pid, while preserving the
inferior num. */
thread_change_ptid (inferior_ptid, new_ptid);
/* We've just told GDB core that the thread changed target id, but,
in fact, it really is a different thread, with different register
contents. */
registers_changed ();
}
Note this is whole checkpoints feature is old code, that predates
the multi-inferior support for a long while. It has many
limitations (e.g., relies on fork, so can't use it on multi-threaded
programs), but I/we've kept it limping along, at least in
the single-inferior case.
>
> > Problem #2 is caused by the make_cleanup_restore_current_thread
> > cleanup installed by fetch_inferior_event not being able to find the
> > original thread 1's ptid in the thread list, thus not being able to
> > restore thread 1 as selected thread. The fix is to make the cleanup
> > installed by make_cleanup_restore_current_thread aware of thread ptid
> > changes, by installing a thread_ptid_changed observer that adjusts the
> > cleanup's data.
>
> I'm guessing this is less hacky than it sounds :-),
:-) I'd say this is just using the thread_changed_ptid observer
hook as intended.
> but it does make me want to understand why this only comes up
> now since threads/inferiors have always been able to "go away"
> while gdb is doing something.
See above. It's just that non-stop mode was never adjusted
to provide the same illusion that all-stop always provided.
I first noticed it by forcing the target always in non-stop
mode, but the problem exists in current master, with "set non-stop on".
This issue wasn't visible in all-stop, because:
- in all-stop we always call set_executing(..., 0) / set_running (..., 0)
with minus_one_ptid (all threads), even on process exits.
- fetch_inferior_event does not try to preserve the selected
thread in all-stop. (although as discussed a while ago, it should,
if the command was a background command, but that's another story).
>
> > After the patch, we get the same in all-stop and non-stop modes:
> >
> > (gdb) c
> > Continuing.
> > Copy complete.
> > Deleting copy.
> > [Inferior 1 (process 25109) exited normally]
> > [Switching to process 25113]
> > (gdb) info threads
> > Id Target Id Frame
> > * 1 process 25113 "checkpoint" main () at src/gdb/testsuite/gdb.base/checkpoint.c:28
> > (gdb)
> >
> > Turns out the whole checkpoints.exp file can run in non-stop mode
> > unmodified. I thought of moving most of the test file's contents to a
> > procedure that can be called twice, once in non-stop mode and another
> > in all-stop mode. But then, the test already takes over 30 seconds to
> > run on my machine, so I thought it'd be nicer to run all-stop and
> > non-stop mode in parallel. Thus I added a new checkpoint-ns.exp file
> > that just sources checkpoint.exp, and sets a knob that checkpoint.exp
> > reads to know it should test non-stop mode. No other test in the tree
> > currently uses this mechanism, but I can't see a reason we shouldn't
> > do this.
>
> Re: checkpoint-ns.exp:
> If we're going to have one "make check" involve both all-stop
> and non-stop testing (as opposed to doing one "make check" with all-stop
> and another one with non-stop), I'd rather see a more table-driven approach
> (the details don't matter to me much other than I'd rather have one file
> than a proliferation of foo-ns.exp files).
> Parallelizing that may involve an extra step, but that's ok.
Yes, it sounds like something we may want to get at at some point,
but I can't invest time right now on designing something
more elaborate. I suggest letting this first example go in,
and see where it takes us, what patterns start emerging.
For example, I realized that we don't really need the extra
variable in checkpoint/checkpoint-ns.exp; we can use GDBFLAGS
instead, though I'm not sure that will/would be sufficient for
other uses.
If I understand your "extra step" correctly, you're suggesting a generated
file. So there's still an extra file anyway? I just think of the separate
file as moving the test's body to a helper "library" instead of a
procedure, with the bonus that we can call either variant from the command line,
instead of being forced to run both variants in sequence, as in
the foreach+procedure way that we've been using in many files.
Whether separate file or foreach, it seems to be that it's the same thing
wrt to table-driven vs procedural, so I can't see how an extra file makes
things any worse. I will personally keep tending to pick the "foreach
variant" route as first choice, as done in numerous tests now. It's just
that checkpoint.exp is longish in both line number and time it takes to run.
> > --- a/gdb/infrun.c
> > +++ b/gdb/infrun.c
> > @@ -3804,8 +3804,18 @@ handle_inferior_event (struct execution_control_state *ecs)
> > any other process were left running. */
> > if (!non_stop)
> > set_executing (minus_one_ptid, 0);
> > - else if (ecs->ws.kind != TARGET_WAITKIND_SIGNALLED
> > - && ecs->ws.kind != TARGET_WAITKIND_EXITED)
> > + else if (ecs->ws.kind == TARGET_WAITKIND_SIGNALLED
> > + && ecs->ws.kind == TARGET_WAITKIND_EXITED)
>
> pasto: s/&&/||/
>
Whoops, fixed. Checkpoints being single-threaded, the else
branch ended up working...
> > + {
> > + ptid_t pid_ptid;
> > +
> > + /* Some targets still have execution when a process exits.
> > + E.g., for "checkpoint", when when a fork exits and is
>
> s/when when/when/
>
> > + mourned, linux-fork.c switches to another fork. */
> > + pid_ptid = pid_to_ptid (ptid_get_pid (ecs->ptid));
> > + set_executing (pid_ptid, 0);
>
> This is a bit confusing. I'm guessing ecs->ptid is for the inferior
> that just exited/signalled, but we're not changing pids here.
> I'm guessing we'll mourn the inferior later, but IWBN to elaborate
> on the comment a little, e.g., to say we always need to call set_executing
> here, even if the inferior exited/signalled (which the code didn't
> previously do). But it's not clear why "switches to another fork" comes
> into play here.
I've changed it this way:
diff --git c/gdb/infrun.c w/gdb/infrun.c
index 9c1fdc5..caf1ff5 100644
--- c/gdb/infrun.c
+++ w/gdb/infrun.c
@@ -3798,20 +3798,27 @@ handle_inferior_event (struct execution_control_state *ecs)
/* Mark the non-executing threads accordingly. In all-stop, all
threads of all processes are stopped when we get any event
- reported. In non-stop mode, only the event thread stops. If
- we're handling a process exit in non-stop mode, there's nothing
- to do, as threads of the dead process are gone, and threads of
- any other process were left running. */
+ reported. In non-stop mode, only the event thread stops. */
if (!non_stop)
set_executing (minus_one_ptid, 0);
else if (ecs->ws.kind == TARGET_WAITKIND_SIGNALLED
- && ecs->ws.kind == TARGET_WAITKIND_EXITED)
+ || ecs->ws.kind == TARGET_WAITKIND_EXITED)
{
ptid_t pid_ptid;
- /* Some targets still have execution when a process exits.
- E.g., for "checkpoint", when when a fork exits and is
- mourned, linux-fork.c switches to another fork. */
+ /* If we're handling a process exit in non-stop mode, even
+ though threads haven't been deleted yet, one would think that
+ there is nothing to do, as threads of the dead process will
+ be soon deleted, and threads of any other process were left
+ running. However, on some targets, threads survive a process
+ exit event. E.g., for the "checkpoint" command, when the
+ current checkpoint/fork exits, linux-fork.c automatically
+ switches to another fork from within target_mourn_inferior,
+ by associating the same inferior/thread to another fork. We
+ haven't mourned yet at this point, but we must mark any
+ threads left in the process as not-executing so that
+ finish_thread_state marks them stopped (in the user's
+ perspective) if/when we present the stop to the user. */
pid_ptid = pid_to_ptid (ptid_get_pid (ecs->ptid));
set_executing (pid_ptid, 0);
}
@@ -6573,9 +6580,11 @@ normal_stop (void)
else if (last.kind == TARGET_WAITKIND_SIGNALLED
|| last.kind == TARGET_WAITKIND_EXITED)
{
- /* Some targets still have execution when a process exits.
- E.g., for "checkpoint", when when a fork exits and is
- mourned, linux-fork.c switches to another fork. */
+ /* On some targets, we may still have live threads in the
+ inferior when we get a process exit event. E.g., for
+ "checkpoint", when the current checkpoint/fork exits,
+ linux-fork.c automatically switches to another fork from
+ within target_mourn_inferior. */
if (!ptid_equal (inferior_ptid, null_ptid))
{
pid_ptid = pid_to_ptid (ptid_get_pid (inferior_ptid));
Below's the resulting patch
WDYT?
From 5932f834e3b4feb2f0771d7bb344ad8f07ed98e1 Mon Sep 17 00:00:00 2001
From: Pedro Alves <palves@redhat.com>
Date: Wed, 22 Apr 2015 18:25:29 +0100
Subject: [PATCH] Fix and test "checkpoint" in non-stop mode
Letting a "checkpoint" run to exit with "set non-stop on" behaves
differently compared to the default all-stop mode ("set non-stop
off").
Currently, in non-stop mode:
(gdb) start
Temporary breakpoint 1 at 0x40086b: file src/gdb/testsuite/gdb.base/checkpoint.c, line 28.
Starting program: build/gdb/testsuite/gdb.base/checkpoint
Temporary breakpoint 1, main () at src/gdb/testsuite/gdb.base/checkpoint.c:28
28 char *tmp = &linebuf[0];
(gdb) checkpoint
checkpoint 1: fork returned pid 24948.
(gdb) c
Continuing.
Copy complete.
Deleting copy.
[Inferior 1 (process 24944) exited normally]
[Switching to process 24948]
(gdb) info threads
Id Target Id Frame
1 process 24948 "checkpoint" (running)
No selected thread. See `help thread'.
(gdb) c
The program is not being run.
(gdb)
Two issues above:
1. Thread 1 got stuck in "(running)" state (it isn't really running)
2. While checkpoints try to preserve the illusion that the thread is
still the same when the process exits, GDB switched to "No thread
selected." instead of staying with thread 1 selected.
Problem #1 is caused by handle_inferior_event and normal_stop not
considering that when a
TARGET_WAITKIND_SIGNALLED/TARGET_WAITKIND_EXITED event is reported,
and the inferior is mourned, the target may still have execution.
Problem #2 is caused by the make_cleanup_restore_current_thread
cleanup installed by fetch_inferior_event not being able to find the
original thread 1's ptid in the thread list, thus not being able to
restore thread 1 as selected thread. The fix is to make the cleanup
installed by make_cleanup_restore_current_thread aware of thread ptid
changes, by installing a thread_ptid_changed observer that adjusts the
cleanup's data.
After the patch, we get the same in all-stop and non-stop modes:
(gdb) c
Continuing.
Copy complete.
Deleting copy.
[Inferior 1 (process 25109) exited normally]
[Switching to process 25113]
(gdb) info threads
Id Target Id Frame
* 1 process 25113 "checkpoint" main () at src/gdb/testsuite/gdb.base/checkpoint.c:28
(gdb)
Turns out the whole checkpoints.exp file can run in non-stop mode
unmodified. I thought of moving most of the test file's contents to a
procedure that can be called twice, once in non-stop mode and another
in all-stop mode. But then, the test already takes over 30 seconds to
run on my machine, so I thought it'd be nicer to run all-stop and
non-stop mode in parallel. Thus I added a new checkpoint-ns.exp file
that just appends "set non-stop on" to GDBFLAGS and sources
checkpoint.exp.
gdb/ChangeLog:
2015-04-22 Pedro Alves <palves@redhat.com>
* infrun.c (handle_inferior_event): If we get
TARGET_WAITKIND_SIGNALLED or TARGET_WAITKIND_EXITED in non-stop
mode, mark all threads of the exiting process as not-executing.
(normal_stop): If we get TARGET_WAITKIND_SIGNALLED or
TARGET_WAITKIND_EXITED in non-stop mode, finish all threads of the
exiting process, if inferior_ptid still points at a process.
* thread.c (struct current_thread_cleanup) <next>: New field.
(current_thread_cleanup_chain): New global.
(restore_current_thread_ptid_changed): New function.
(restore_current_thread_cleanup_dtor): Remove the cleanup from the
current_thread_cleanup_chain list.
(make_cleanup_restore_current_thread): Add the cleanup data to the
current_thread_cleanup_chain list.
(_initialize_thread): Install restore_current_thread_ptid_changed
as thread_ptid_changed observer.
gdb/testsuite/ChangeLog:
2015-04-22 Pedro Alves <palves@redhat.com>
* gdb.base/checkpoint-ns.exp: New file.
* gdb.base/checkpoint.exp: Pass explicit "checkpoint.c" to
standard_testfile.
v3.1:
Extend comments.
In test, use GDBFLAGS instead of new variable.
v3:
No changes.
---
gdb/infrun.c | 48 ++++++++++++++++++++++++++------
gdb/testsuite/gdb.base/checkpoint-ns.exp | 26 +++++++++++++++++
gdb/testsuite/gdb.base/checkpoint.exp | 6 ++--
gdb/thread.c | 38 +++++++++++++++++++++++++
4 files changed, 107 insertions(+), 11 deletions(-)
create mode 100644 gdb/testsuite/gdb.base/checkpoint-ns.exp
diff --git a/gdb/infrun.c b/gdb/infrun.c
index 7870f70..caf1ff5 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -3798,14 +3798,31 @@ handle_inferior_event (struct execution_control_state *ecs)
/* Mark the non-executing threads accordingly. In all-stop, all
threads of all processes are stopped when we get any event
- reported. In non-stop mode, only the event thread stops. If
- we're handling a process exit in non-stop mode, there's nothing
- to do, as threads of the dead process are gone, and threads of
- any other process were left running. */
+ reported. In non-stop mode, only the event thread stops. */
if (!non_stop)
set_executing (minus_one_ptid, 0);
- else if (ecs->ws.kind != TARGET_WAITKIND_SIGNALLED
- && ecs->ws.kind != TARGET_WAITKIND_EXITED)
+ else if (ecs->ws.kind == TARGET_WAITKIND_SIGNALLED
+ || ecs->ws.kind == TARGET_WAITKIND_EXITED)
+ {
+ ptid_t pid_ptid;
+
+ /* If we're handling a process exit in non-stop mode, even
+ though threads haven't been deleted yet, one would think that
+ there is nothing to do, as threads of the dead process will
+ be soon deleted, and threads of any other process were left
+ running. However, on some targets, threads survive a process
+ exit event. E.g., for the "checkpoint" command, when the
+ current checkpoint/fork exits, linux-fork.c automatically
+ switches to another fork from within target_mourn_inferior,
+ by associating the same inferior/thread to another fork. We
+ haven't mourned yet at this point, but we must mark any
+ threads left in the process as not-executing so that
+ finish_thread_state marks them stopped (in the user's
+ perspective) if/when we present the stop to the user. */
+ pid_ptid = pid_to_ptid (ptid_get_pid (ecs->ptid));
+ set_executing (pid_ptid, 0);
+ }
+ else
set_executing (ecs->ptid, 0);
switch (ecs->ws.kind)
@@ -6550,6 +6567,7 @@ normal_stop (void)
struct target_waitstatus last;
ptid_t last_ptid;
struct cleanup *old_chain = make_cleanup (null_cleanup, NULL);
+ ptid_t pid_ptid;
get_last_target_status (&last_ptid, &last);
@@ -6559,9 +6577,21 @@ normal_stop (void)
here, so do this before any filtered output. */
if (!non_stop)
make_cleanup (finish_thread_state_cleanup, &minus_one_ptid);
- else if (last.kind != TARGET_WAITKIND_SIGNALLED
- && last.kind != TARGET_WAITKIND_EXITED
- && last.kind != TARGET_WAITKIND_NO_RESUMED)
+ else if (last.kind == TARGET_WAITKIND_SIGNALLED
+ || last.kind == TARGET_WAITKIND_EXITED)
+ {
+ /* On some targets, we may still have live threads in the
+ inferior when we get a process exit event. E.g., for
+ "checkpoint", when the current checkpoint/fork exits,
+ linux-fork.c automatically switches to another fork from
+ within target_mourn_inferior. */
+ if (!ptid_equal (inferior_ptid, null_ptid))
+ {
+ pid_ptid = pid_to_ptid (ptid_get_pid (inferior_ptid));
+ make_cleanup (finish_thread_state_cleanup, &pid_ptid);
+ }
+ }
+ else if (last.kind != TARGET_WAITKIND_NO_RESUMED)
make_cleanup (finish_thread_state_cleanup, &inferior_ptid);
/* As we're presenting a stop, and potentially removing breakpoints,
diff --git a/gdb/testsuite/gdb.base/checkpoint-ns.exp b/gdb/testsuite/gdb.base/checkpoint-ns.exp
new file mode 100644
index 0000000..d3698ba
--- /dev/null
+++ b/gdb/testsuite/gdb.base/checkpoint-ns.exp
@@ -0,0 +1,26 @@
+# Copyright 2015 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 checkpoint and restart in non-stop mode.
+
+# We drive non-stop mode from a separate file because the whole test
+# takes a while to run. This way, we can test both modes in parallel.
+
+set saved_gdbflags $GDBFLAGS
+append GDBFLAGS " -ex \"set non-stop on\""
+
+source $srcdir/$subdir/checkpoint.exp
+
+set GDBFLAGS $saved_gdbflags
diff --git a/gdb/testsuite/gdb.base/checkpoint.exp b/gdb/testsuite/gdb.base/checkpoint.exp
index 6d94ab6..4a476be 100644
--- a/gdb/testsuite/gdb.base/checkpoint.exp
+++ b/gdb/testsuite/gdb.base/checkpoint.exp
@@ -24,8 +24,10 @@ if {![istarget "*-*-linux*"]} then {
continue
}
-
-standard_testfile .c
+# Must name the source file explicitly, otherwise when driven by
+# checkpoints-ns.exp, we'd try compiling checkpoints-ns.c, which
+# doesn't exist.
+standard_testfile checkpoint.c
set pi_txt [gdb_remote_download host ${srcdir}/${subdir}/pi.txt]
if {[is_remote host]} {
diff --git a/gdb/thread.c b/gdb/thread.c
index 23dfcc9..46b5947 100644
--- a/gdb/thread.c
+++ b/gdb/thread.c
@@ -1279,8 +1279,16 @@ restore_selected_frame (struct frame_id a_frame_id, int frame_level)
}
}
+/* Data used by the cleanup installed by
+ 'make_cleanup_restore_current_thread'. */
+
struct current_thread_cleanup
{
+ /* Next in list of currently installed 'struct
+ current_thread_cleanup' cleanups. See
+ 'current_thread_cleanup_chain' below. */
+ struct current_thread_cleanup *next;
+
ptid_t inferior_ptid;
struct frame_id selected_frame_id;
int selected_frame_level;
@@ -1289,6 +1297,29 @@ struct current_thread_cleanup
int was_removable;
};
+/* A chain of currently installed 'struct current_thread_cleanup'
+ cleanups. Restoring the previously selected thread looks up the
+ old thread in the thread list by ptid. If the thread changes ptid,
+ we need to update the cleanup's thread structure so the look up
+ succeeds. */
+static struct current_thread_cleanup *current_thread_cleanup_chain;
+
+/* A thread_ptid_changed observer. Update all currently installed
+ current_thread_cleanup cleanups that want to switch back to
+ OLD_PTID to switch back to NEW_PTID instead. */
+
+static void
+restore_current_thread_ptid_changed (ptid_t old_ptid, ptid_t new_ptid)
+{
+ struct current_thread_cleanup *it;
+
+ for (it = current_thread_cleanup_chain; it != NULL; it = it->next)
+ {
+ if (ptid_equal (it->inferior_ptid, old_ptid))
+ it->inferior_ptid = new_ptid;
+ }
+}
+
static void
do_restore_current_thread_cleanup (void *arg)
{
@@ -1329,6 +1360,8 @@ restore_current_thread_cleanup_dtor (void *arg)
struct thread_info *tp;
struct inferior *inf;
+ current_thread_cleanup_chain = current_thread_cleanup_chain->next;
+
tp = find_thread_ptid (old->inferior_ptid);
if (tp)
tp->refcount--;
@@ -1362,6 +1395,9 @@ make_cleanup_restore_current_thread (void)
old->inf_id = current_inferior ()->num;
old->was_removable = current_inferior ()->removable;
+ old->next = current_thread_cleanup_chain;
+ current_thread_cleanup_chain = old;
+
if (!ptid_equal (inferior_ptid, null_ptid))
{
old->was_stopped = is_stopped (inferior_ptid);
@@ -1815,4 +1851,6 @@ Show printing of thread events (such as thread start and exit)."), NULL,
&setprintlist, &showprintlist);
create_internalvar_type_lazy ("_thread", &thread_funcs, NULL);
+
+ observer_attach_thread_ptid_changed (restore_current_thread_ptid_changed);
}
--
1.9.3
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 01/17] Fix and test "checkpoint" in non-stop mode
2015-04-22 17:48 ` Pedro Alves
@ 2015-04-28 18:18 ` Doug Evans
2015-04-29 4:56 ` Doug Evans
0 siblings, 1 reply; 62+ messages in thread
From: Doug Evans @ 2015-04-28 18:18 UTC (permalink / raw)
To: Pedro Alves; +Cc: gdb-patches
Pedro Alves writes:
> Note this is whole checkpoints feature is old code, that predates
> the multi-inferior support for a long while. It has many
> limitations (e.g., relies on fork, so can't use it on multi-threaded
> programs), but I/we've kept it limping along, at least in
> the single-inferior case.
Hmmm. Got it.
> > Re: checkpoint-ns.exp:
> > If we're going to have one "make check" involve both all-stop
> > and non-stop testing (as opposed to doing one "make check" with all-stop
> > and another one with non-stop), I'd rather see a more table-driven approach
> > (the details don't matter to me much other than I'd rather have one file
> > than a proliferation of foo-ns.exp files).
> > Parallelizing that may involve an extra step, but that's ok.
>
> Yes, it sounds like something we may want to get at at some point,
> but I can't invest time right now on designing something
> more elaborate. I suggest letting this first example go in,
> and see where it takes us, what patterns start emerging.
> For example, I realized that we don't really need the extra
> variable in checkpoint/checkpoint-ns.exp; we can use GDBFLAGS
> instead, though I'm not sure that will/would be sufficient for
> other uses.
>
> If I understand your "extra step" correctly, you're suggesting a generated
> file. So there's still an extra file anyway? I just think of the separate
> file as moving the test's body to a helper "library" instead of a
> procedure, with the bonus that we can call either variant from the command line,
> instead of being forced to run both variants in sequence, as in
> the foreach+procedure way that we've been using in many files.
>
> Whether separate file or foreach, it seems to be that it's the same thing
> wrt to table-driven vs procedural, so I can't see how an extra file makes
> things any worse. I will personally keep tending to pick the "foreach
> variant" route as first choice, as done in numerous tests now. It's just
> that checkpoint.exp is longish in both line number and time it takes to run.
Ah.
The axis on which I'm worried is whether this will turn out to be
1 extra file or 1000. One extra file per test doesn't scale.
And trying to reign in 100 files, or whatever the threshold is
before it's clear that we need to change course, will be more
painful than having a table-driven approach that can be simple
now and still grow as needed. (*1)
Another way to go would be really just have 1 extra file (or 1 extra
file per gdb.foo subdir or some such), and this file
documents all the tests intended to be run in different modes
within one "make check". One could even extend
this to be more than just all-stop/non-stop.
Another way, though, is to have two "make check" runs:
One with all-stop and one with non-stop.
Tests that can't handle one or the other should be marked
as such anyway. How a test chooses to handle the choice
is an internal implementation decision regardless of which
route is chosen. Yeah, running some tests twice this way
won't enhance coverage (e.g., no point in running help.exp in
all-stop and non-stop), but we should be almost within epsilon
of being able to do this today (I know I've done it in the past),
and complete test runs only take a few minutes on any reasonably
beefy system (at least they do on mine).
(*1): A hybrid table driven approach could be to (effectively)
run the test twice using whatever mechanism that running
"make check" twice would use. This way only the specified
set of tests would be run twice (instead of running the whole testsuite
twice), and one can still manually run individual tests in whatever
mode (all-stop/non-stop/whatever) one chooses.
My counter-proposal would be to see the extent to which we can
just run the testsuite in all-stop and non-stop, and do that for now
while we work towards something that scales better.
Someone could get a basic table-driven scheme going in a day.
> > > + {
> > > + ptid_t pid_ptid;
> > > +
> > > + /* Some targets still have execution when a process exits.
> > > + E.g., for "checkpoint", when when a fork exits and is
> >
> > s/when when/when/
> >
> > > + mourned, linux-fork.c switches to another fork. */
> > > + pid_ptid = pid_to_ptid (ptid_get_pid (ecs->ptid));
> > > + set_executing (pid_ptid, 0);
> >
> > This is a bit confusing. I'm guessing ecs->ptid is for the inferior
> > that just exited/signalled, but we're not changing pids here.
> > I'm guessing we'll mourn the inferior later, but IWBN to elaborate
> > on the comment a little, e.g., to say we always need to call set_executing
> > here, even if the inferior exited/signalled (which the code didn't
> > previously do). But it's not clear why "switches to another fork" comes
> > into play here.
>
> I've changed it this way:
>
> diff --git c/gdb/infrun.c w/gdb/infrun.c
> index 9c1fdc5..caf1ff5 100644
> --- c/gdb/infrun.c
> +++ w/gdb/infrun.c
> @@ -3798,20 +3798,27 @@ handle_inferior_event (struct execution_control_state *ecs)
>
> /* Mark the non-executing threads accordingly. In all-stop, all
> threads of all processes are stopped when we get any event
> - reported. In non-stop mode, only the event thread stops. If
> - we're handling a process exit in non-stop mode, there's nothing
> - to do, as threads of the dead process are gone, and threads of
> - any other process were left running. */
> + reported. In non-stop mode, only the event thread stops. */
> if (!non_stop)
> set_executing (minus_one_ptid, 0);
> else if (ecs->ws.kind == TARGET_WAITKIND_SIGNALLED
> - && ecs->ws.kind == TARGET_WAITKIND_EXITED)
> + || ecs->ws.kind == TARGET_WAITKIND_EXITED)
> {
> ptid_t pid_ptid;
>
> - /* Some targets still have execution when a process exits.
> - E.g., for "checkpoint", when when a fork exits and is
> - mourned, linux-fork.c switches to another fork. */
> + /* If we're handling a process exit in non-stop mode, even
> + though threads haven't been deleted yet, one would think that
> + there is nothing to do, as threads of the dead process will
> + be soon deleted, and threads of any other process were left
> + running. However, on some targets, threads survive a process
> + exit event. E.g., for the "checkpoint" command, when the
> + current checkpoint/fork exits, linux-fork.c automatically
> + switches to another fork from within target_mourn_inferior,
> + by associating the same inferior/thread to another fork. We
> + haven't mourned yet at this point, but we must mark any
> + threads left in the process as not-executing so that
> + finish_thread_state marks them stopped (in the user's
> + perspective) if/when we present the stop to the user. */
> pid_ptid = pid_to_ptid (ptid_get_pid (ecs->ptid));
> set_executing (pid_ptid, 0);
> }
Works for me.
Thanks!
> >From 5932f834e3b4feb2f0771d7bb344ad8f07ed98e1 Mon Sep 17 00:00:00 2001
> From: Pedro Alves <palves@redhat.com>
> Date: Wed, 22 Apr 2015 18:25:29 +0100
> Subject: [PATCH] Fix and test "checkpoint" in non-stop mode
>
> Letting a "checkpoint" run to exit with "set non-stop on" behaves
> differently compared to the default all-stop mode ("set non-stop
> off").
>
> Currently, in non-stop mode:
>
> (gdb) start
> Temporary breakpoint 1 at 0x40086b: file src/gdb/testsuite/gdb.base/checkpoint.c, line 28.
> Starting program: build/gdb/testsuite/gdb.base/checkpoint
>
> Temporary breakpoint 1, main () at src/gdb/testsuite/gdb.base/checkpoint.c:28
> 28 char *tmp = &linebuf[0];
> (gdb) checkpoint
> checkpoint 1: fork returned pid 24948.
> (gdb) c
> Continuing.
> Copy complete.
> Deleting copy.
> [Inferior 1 (process 24944) exited normally]
> [Switching to process 24948]
> (gdb) info threads
> Id Target Id Frame
> 1 process 24948 "checkpoint" (running)
>
> No selected thread. See `help thread'.
> (gdb) c
> The program is not being run.
> (gdb)
>
> Two issues above:
>
> 1. Thread 1 got stuck in "(running)" state (it isn't really running)
>
> 2. While checkpoints try to preserve the illusion that the thread is
> still the same when the process exits, GDB switched to "No thread
> selected." instead of staying with thread 1 selected.
>
> Problem #1 is caused by handle_inferior_event and normal_stop not
> considering that when a
> TARGET_WAITKIND_SIGNALLED/TARGET_WAITKIND_EXITED event is reported,
> and the inferior is mourned, the target may still have execution.
>
> Problem #2 is caused by the make_cleanup_restore_current_thread
> cleanup installed by fetch_inferior_event not being able to find the
> original thread 1's ptid in the thread list, thus not being able to
> restore thread 1 as selected thread. The fix is to make the cleanup
> installed by make_cleanup_restore_current_thread aware of thread ptid
> changes, by installing a thread_ptid_changed observer that adjusts the
> cleanup's data.
>
> After the patch, we get the same in all-stop and non-stop modes:
>
> (gdb) c
> Continuing.
> Copy complete.
> Deleting copy.
> [Inferior 1 (process 25109) exited normally]
> [Switching to process 25113]
> (gdb) info threads
> Id Target Id Frame
> * 1 process 25113 "checkpoint" main () at src/gdb/testsuite/gdb.base/checkpoint.c:28
> (gdb)
>
> Turns out the whole checkpoints.exp file can run in non-stop mode
> unmodified. I thought of moving most of the test file's contents to a
> procedure that can be called twice, once in non-stop mode and another
> in all-stop mode. But then, the test already takes over 30 seconds to
> run on my machine, so I thought it'd be nicer to run all-stop and
> non-stop mode in parallel. Thus I added a new checkpoint-ns.exp file
> that just appends "set non-stop on" to GDBFLAGS and sources
> checkpoint.exp.
>
> gdb/ChangeLog:
> 2015-04-22 Pedro Alves <palves@redhat.com>
>
> * infrun.c (handle_inferior_event): If we get
> TARGET_WAITKIND_SIGNALLED or TARGET_WAITKIND_EXITED in non-stop
> mode, mark all threads of the exiting process as not-executing.
> (normal_stop): If we get TARGET_WAITKIND_SIGNALLED or
> TARGET_WAITKIND_EXITED in non-stop mode, finish all threads of the
> exiting process, if inferior_ptid still points at a process.
> * thread.c (struct current_thread_cleanup) <next>: New field.
> (current_thread_cleanup_chain): New global.
> (restore_current_thread_ptid_changed): New function.
> (restore_current_thread_cleanup_dtor): Remove the cleanup from the
> current_thread_cleanup_chain list.
> (make_cleanup_restore_current_thread): Add the cleanup data to the
> current_thread_cleanup_chain list.
> (_initialize_thread): Install restore_current_thread_ptid_changed
> as thread_ptid_changed observer.
>
> gdb/testsuite/ChangeLog:
> 2015-04-22 Pedro Alves <palves@redhat.com>
>
> * gdb.base/checkpoint-ns.exp: New file.
> * gdb.base/checkpoint.exp: Pass explicit "checkpoint.c" to
> standard_testfile.
>
> v3.1:
>
> Extend comments.
> In test, use GDBFLAGS instead of new variable.
>
> v3:
> No changes.
> ---
> gdb/infrun.c | 48 ++++++++++++++++++++++++++------
> gdb/testsuite/gdb.base/checkpoint-ns.exp | 26 +++++++++++++++++
> gdb/testsuite/gdb.base/checkpoint.exp | 6 ++--
> gdb/thread.c | 38 +++++++++++++++++++++++++
> 4 files changed, 107 insertions(+), 11 deletions(-)
> create mode 100644 gdb/testsuite/gdb.base/checkpoint-ns.exp
>
> diff --git a/gdb/infrun.c b/gdb/infrun.c
> index 7870f70..caf1ff5 100644
> --- a/gdb/infrun.c
> +++ b/gdb/infrun.c
> @@ -3798,14 +3798,31 @@ handle_inferior_event (struct execution_control_state *ecs)
>
> /* Mark the non-executing threads accordingly. In all-stop, all
> threads of all processes are stopped when we get any event
> - reported. In non-stop mode, only the event thread stops. If
> - we're handling a process exit in non-stop mode, there's nothing
> - to do, as threads of the dead process are gone, and threads of
> - any other process were left running. */
> + reported. In non-stop mode, only the event thread stops. */
> if (!non_stop)
> set_executing (minus_one_ptid, 0);
> - else if (ecs->ws.kind != TARGET_WAITKIND_SIGNALLED
> - && ecs->ws.kind != TARGET_WAITKIND_EXITED)
> + else if (ecs->ws.kind == TARGET_WAITKIND_SIGNALLED
> + || ecs->ws.kind == TARGET_WAITKIND_EXITED)
> + {
> + ptid_t pid_ptid;
> +
> + /* If we're handling a process exit in non-stop mode, even
> + though threads haven't been deleted yet, one would think that
> + there is nothing to do, as threads of the dead process will
> + be soon deleted, and threads of any other process were left
> + running. However, on some targets, threads survive a process
> + exit event. E.g., for the "checkpoint" command, when the
> + current checkpoint/fork exits, linux-fork.c automatically
> + switches to another fork from within target_mourn_inferior,
> + by associating the same inferior/thread to another fork. We
> + haven't mourned yet at this point, but we must mark any
> + threads left in the process as not-executing so that
> + finish_thread_state marks them stopped (in the user's
> + perspective) if/when we present the stop to the user. */
> + pid_ptid = pid_to_ptid (ptid_get_pid (ecs->ptid));
> + set_executing (pid_ptid, 0);
> + }
> + else
> set_executing (ecs->ptid, 0);
>
> switch (ecs->ws.kind)
> @@ -6550,6 +6567,7 @@ normal_stop (void)
> struct target_waitstatus last;
> ptid_t last_ptid;
> struct cleanup *old_chain = make_cleanup (null_cleanup, NULL);
> + ptid_t pid_ptid;
>
> get_last_target_status (&last_ptid, &last);
>
> @@ -6559,9 +6577,21 @@ normal_stop (void)
> here, so do this before any filtered output. */
> if (!non_stop)
> make_cleanup (finish_thread_state_cleanup, &minus_one_ptid);
> - else if (last.kind != TARGET_WAITKIND_SIGNALLED
> - && last.kind != TARGET_WAITKIND_EXITED
> - && last.kind != TARGET_WAITKIND_NO_RESUMED)
> + else if (last.kind == TARGET_WAITKIND_SIGNALLED
> + || last.kind == TARGET_WAITKIND_EXITED)
> + {
> + /* On some targets, we may still have live threads in the
> + inferior when we get a process exit event. E.g., for
> + "checkpoint", when the current checkpoint/fork exits,
> + linux-fork.c automatically switches to another fork from
> + within target_mourn_inferior. */
> + if (!ptid_equal (inferior_ptid, null_ptid))
> + {
> + pid_ptid = pid_to_ptid (ptid_get_pid (inferior_ptid));
> + make_cleanup (finish_thread_state_cleanup, &pid_ptid);
> + }
> + }
> + else if (last.kind != TARGET_WAITKIND_NO_RESUMED)
> make_cleanup (finish_thread_state_cleanup, &inferior_ptid);
>
> /* As we're presenting a stop, and potentially removing breakpoints,
> diff --git a/gdb/testsuite/gdb.base/checkpoint-ns.exp b/gdb/testsuite/gdb.base/checkpoint-ns.exp
> new file mode 100644
> index 0000000..d3698ba
> --- /dev/null
> +++ b/gdb/testsuite/gdb.base/checkpoint-ns.exp
> @@ -0,0 +1,26 @@
> +# Copyright 2015 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 checkpoint and restart in non-stop mode.
> +
> +# We drive non-stop mode from a separate file because the whole test
> +# takes a while to run. This way, we can test both modes in parallel.
> +
> +set saved_gdbflags $GDBFLAGS
> +append GDBFLAGS " -ex \"set non-stop on\""
> +
> +source $srcdir/$subdir/checkpoint.exp
> +
> +set GDBFLAGS $saved_gdbflags
> diff --git a/gdb/testsuite/gdb.base/checkpoint.exp b/gdb/testsuite/gdb.base/checkpoint.exp
> index 6d94ab6..4a476be 100644
> --- a/gdb/testsuite/gdb.base/checkpoint.exp
> +++ b/gdb/testsuite/gdb.base/checkpoint.exp
> @@ -24,8 +24,10 @@ if {![istarget "*-*-linux*"]} then {
> continue
> }
>
> -
> -standard_testfile .c
> +# Must name the source file explicitly, otherwise when driven by
> +# checkpoints-ns.exp, we'd try compiling checkpoints-ns.c, which
> +# doesn't exist.
> +standard_testfile checkpoint.c
>
> set pi_txt [gdb_remote_download host ${srcdir}/${subdir}/pi.txt]
> if {[is_remote host]} {
> diff --git a/gdb/thread.c b/gdb/thread.c
> index 23dfcc9..46b5947 100644
> --- a/gdb/thread.c
> +++ b/gdb/thread.c
> @@ -1279,8 +1279,16 @@ restore_selected_frame (struct frame_id a_frame_id, int frame_level)
> }
> }
>
> +/* Data used by the cleanup installed by
> + 'make_cleanup_restore_current_thread'. */
> +
> struct current_thread_cleanup
> {
> + /* Next in list of currently installed 'struct
> + current_thread_cleanup' cleanups. See
> + 'current_thread_cleanup_chain' below. */
> + struct current_thread_cleanup *next;
> +
> ptid_t inferior_ptid;
> struct frame_id selected_frame_id;
> int selected_frame_level;
> @@ -1289,6 +1297,29 @@ struct current_thread_cleanup
> int was_removable;
> };
>
> +/* A chain of currently installed 'struct current_thread_cleanup'
> + cleanups. Restoring the previously selected thread looks up the
> + old thread in the thread list by ptid. If the thread changes ptid,
> + we need to update the cleanup's thread structure so the look up
> + succeeds. */
> +static struct current_thread_cleanup *current_thread_cleanup_chain;
> +
> +/* A thread_ptid_changed observer. Update all currently installed
> + current_thread_cleanup cleanups that want to switch back to
> + OLD_PTID to switch back to NEW_PTID instead. */
> +
> +static void
> +restore_current_thread_ptid_changed (ptid_t old_ptid, ptid_t new_ptid)
> +{
> + struct current_thread_cleanup *it;
> +
> + for (it = current_thread_cleanup_chain; it != NULL; it = it->next)
> + {
> + if (ptid_equal (it->inferior_ptid, old_ptid))
> + it->inferior_ptid = new_ptid;
> + }
> +}
> +
> static void
> do_restore_current_thread_cleanup (void *arg)
> {
> @@ -1329,6 +1360,8 @@ restore_current_thread_cleanup_dtor (void *arg)
> struct thread_info *tp;
> struct inferior *inf;
>
> + current_thread_cleanup_chain = current_thread_cleanup_chain->next;
> +
> tp = find_thread_ptid (old->inferior_ptid);
> if (tp)
> tp->refcount--;
> @@ -1362,6 +1395,9 @@ make_cleanup_restore_current_thread (void)
> old->inf_id = current_inferior ()->num;
> old->was_removable = current_inferior ()->removable;
>
> + old->next = current_thread_cleanup_chain;
> + current_thread_cleanup_chain = old;
> +
> if (!ptid_equal (inferior_ptid, null_ptid))
> {
> old->was_stopped = is_stopped (inferior_ptid);
> @@ -1815,4 +1851,6 @@ Show printing of thread events (such as thread start and exit)."), NULL,
> &setprintlist, &showprintlist);
>
> create_internalvar_type_lazy ("_thread", &thread_funcs, NULL);
> +
> + observer_attach_thread_ptid_changed (restore_current_thread_ptid_changed);
> }
> --
> 1.9.3
>
>
--
/dje
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 01/17] Fix and test "checkpoint" in non-stop mode
2015-04-28 18:18 ` Doug Evans
@ 2015-04-29 4:56 ` Doug Evans
2015-05-19 18:08 ` Pedro Alves
0 siblings, 1 reply; 62+ messages in thread
From: Doug Evans @ 2015-04-29 4:56 UTC (permalink / raw)
To: Pedro Alves; +Cc: gdb-patches
On Tue, Apr 28, 2015 at 9:06 AM, Doug Evans <dje@google.com> wrote:
> The axis on which I'm worried is whether this will turn out to be
> 1 extra file or 1000. One extra file per test doesn't scale.
> And trying to reign in 100 files, or whatever the threshold is
> before it's clear that we need to change course, will be more
> painful than having a table-driven approach that can be simple
> now and still grow as needed. (*1)
>
> Another way to go would be really just have 1 extra file (or 1 extra
> file per gdb.foo subdir or some such), and this file
> documents all the tests intended to be run in different modes
> within one "make check". One could even extend
> this to be more than just all-stop/non-stop.
>
> Another way, though, is to have two "make check" runs:
> One with all-stop and one with non-stop.
> Tests that can't handle one or the other should be marked
> as such anyway. How a test chooses to handle the choice
> is an internal implementation decision regardless of which
> route is chosen. Yeah, running some tests twice this way
> won't enhance coverage (e.g., no point in running help.exp in
> all-stop and non-stop), but we should be almost within epsilon
> of being able to do this today (I know I've done it in the past),
> and complete test runs only take a few minutes on any reasonably
> beefy system (at least they do on mine).
>
> (*1): A hybrid table driven approach could be to (effectively)
> run the test twice using whatever mechanism that running
> "make check" twice would use. This way only the specified
> set of tests would be run twice (instead of running the whole testsuite
> twice), and one can still manually run individual tests in whatever
> mode (all-stop/non-stop/whatever) one chooses.
>
> My counter-proposal would be to see the extent to which we can
> just run the testsuite in all-stop and non-stop, and do that for now
> while we work towards something that scales better.
> Someone could get a basic table-driven scheme going in a day.
[Sorry for the resend ... cursed text/html.]
Another thought was that if this is just a one-off and you're manually
splitting up non-stop from all-stop to maintain parallelizability of the tests
then I might not mind the new file.
IOW, tests that want to handle both all-stop and non-stop in one
"make check" can do so in one file. Any loss in parallelization
is probably minimal (modulo really big tests).
Btw, checkpoint.exp completes in 9 seconds on my system.
I suspect doubling that and doing both all-stop and non-stop in
the one file won't increase "make check-parallel" noticeably.
^ permalink raw reply [flat|nested] 62+ messages in thread
* Re: [PATCH v3 01/17] Fix and test "checkpoint" in non-stop mode
2015-04-29 4:56 ` Doug Evans
@ 2015-05-19 18:08 ` Pedro Alves
0 siblings, 0 replies; 62+ messages in thread
From: Pedro Alves @ 2015-05-19 18:08 UTC (permalink / raw)
To: Doug Evans; +Cc: gdb-patches
On 04/28/2015 10:50 PM, Doug Evans wrote:
> Another thought was that if this is just a one-off and you're manually
> splitting up non-stop from all-stop to maintain parallelizability of the tests
> then I might not mind the new file.
I don't have plans to do this in any other test, so for now at least,
this is really just a one-off. It's special because "checkpoint"
only really works with single-threaded programs (because it uses
fork), while non-stop/all-stop mostly matters about threaded programs.
>
> IOW, tests that want to handle both all-stop and non-stop in one
> "make check" can do so in one file. Any loss in parallelization
> is probably minimal (modulo really big tests).
Yes, and we already do that in some tests.
>
> Btw, checkpoint.exp completes in 9 seconds on my system.
> I suspect doubling that and doing both all-stop and non-stop in
> the one file won't increase "make check-parallel" noticeably.
It makes a larger difference for me:
$ time make check RUNTESTFLAGS="checkpoint.exp"
real 0m24.716s
user 0m22.676s
sys 0m2.168s
$ time make check RUNTESTFLAGS="checkpoint.exp checkpoint-ns.exp"
real 0m49.241s
user 0m45.183s
sys 0m4.306s
$ time make check TESTS="gdb.base/checkpoint.exp gdb.base/checkpoint-ns.exp" -j
real 0m28.606s
user 0m49.684s
sys 0m5.590s
Thanks,
Pedro Alves
^ permalink raw reply [flat|nested] 62+ messages in thread
* [PATCH v3 12/17] Fix interrupt-noterm.exp on targets always in non-stop
2015-04-17 10:47 [PATCH v3 00/23] All-stop on top of non-stop Pedro Alves
` (11 preceding siblings ...)
2015-04-17 10:47 ` [PATCH v3 01/17] Fix and test "checkpoint" in non-stop mode Pedro Alves
@ 2015-04-17 10:52 ` Pedro Alves
2015-04-21 11:40 ` Yao Qi
2015-04-17 10:52 ` [PATCH v3 09/17] Teach non-stop to do in-line step-overs (stop all, step, restart) Pedro Alves
` (5 subsequent siblings)
18 siblings, 1 reply; 62+ messages in thread
From: Pedro Alves @ 2015-04-17 10:52 UTC (permalink / raw)
To: gdb-patches
With "maint set target-non-stop on" we get:
@@ -66,13 +66,16 @@ Continuing.
interrupt
(gdb) PASS: gdb.base/interrupt-noterm.exp: interrupt
-Program received signal SIGINT, Interrupt.
-PASS: gdb.base/interrupt-noterm.exp: inferior received SIGINT
-testcase src/gdb/testsuite/gdb.base/interrupt-noterm.exp completed in 0 seconds
+[process 12119] #1 stopped.
+0x0000003615ebc6d0 in __nanosleep_nocancel () at ../sysdeps/unix/syscall-template.S:81
+81 T_PSEUDO (SYSCALL_SYMBOL, SYSCALL_NAME, SYSCALL_NARGS)
+FAIL: gdb.base/interrupt-noterm.exp: inferior received SIGINT (timeout)
+testcase src/gdb/testsuite/gdb.base/interrupt-noterm.exp completed in 10 seconds
That is, we get "[$thread] #1 stopped" instead of SIGINT.
The issue is that we don't currently distinguish send
"interrupt/ctrl-c" to target terminal vs "stop/pause" thread well;
both cases go through "target_stop".
And then, the native Linux backend (linux-nat.c) implements
target_stop with SIGSTOP in non-stop mode, and SIGINT in all-stop
mode. Since "maint set target-non-stop on" forces the backend to be
always running in non-stop mode, even though the user-visible behavior
is "set non-stop" is "off", "interrupt" causes a SIGSTOP instead of
the SIGINT the test expects.
Fix this by introducing a target_interrupt method to use in the
"interrupt/ctrl-c" case, so "set non-stop off" can always work the
same irrespective of "maint set target-non-stop on/off". I'm
explictly considering changing the "set non-stop on" behavior as out
of scope here.
Most of the patch is an across-the-board rename of to_stop hook
implementations to to_interrupt. The only targets where something
more than a rename is being done are linux-nat.c and remote.c, which
are the only targets that support async, and thus are the only ones
the core side calls target_stop on.
gdb/ChangeLog:
2015-04-17 Pedro Alves <palves@redhat.com>
* darwin-nat.c (darwin_stop): Rename to ...
(darwin_interrupt): ... this.
(_initialize_darwin_inferior): Adjust.
* gnu-nat.c (gnu_stop): Delete.
(gnu_thread_alive): Don't install gnu_stop.
* inf-ptrace.c (inf_ptrace_stop): Rename to ...
(inf_ptrace_interrupt): ... this.
(inf_ptrace_target): Adjust.
* infcmd.c (interrupt_target_1): Use target_interrupt instead of
target_stop.
* linux-nat (linux_nat_stop): Rename to ...
(linux_nat_interrupt): ... this.
(linux_nat_stop): Reimplement.
(linux_nat_add_target): Install linux_nat_interrupt.
* monitor.c (monitor_stop): Rename to ...
(monitor_interrupt): ... this.
(init_base_monitor_ops): Adjust.
* nto-procfs.c (nto_interrupt_twice): Rename to ...
(nto_handle_sigint_twice): ... this.
(nto_interrupt): Rename to ...
(nto_handle_sigint): ... this. Call target_interrupt instead of
target_stop.
(procfs_wait, init_procfs_targets): Adjust.
* procfs.c (procfs_target): Adjust.
(procfs_stop): Rename to ...
(procfs_interrupt): ... this.
* remote-m32r-sdi.c (m32r_stop): Rename to ...
(m32r_interrupt): ... this.
(init_m32r_ops): Adjust.
* remote-sim.c (gdbsim_stop_inferior): Rename to ...
(gdbsim_interrupt_inferior): ... this.
(gdbsim_stop): Rename to ...
(gdbsim_interrupt): ... this.
(gdbsim_cntrl_c): Adjust.
(init_gdbsim_ops): Adjust.
* remote.c (sync_remote_interrupt): Adjust comments.
(remote_stop_as): Rename to ...
(remote_interrupt_as): ... this.
(remote_stop): Adjust comment.
(remote_interrupt): New function.
(init_remote_ops): Install remote_interrupt.
* target.c (target_interrupt): New function.
* target.h (struct target_ops) <to_interrupt>: New field.
(target_interrupt): New declaration.
* windows-nat.c (windows_stop): Rename to ...
(windows_interrupt): ... this.
* target-delegates.c: Regenerate.
v3, v2:
- No changes.
---
gdb/darwin-nat.c | 6 +++---
gdb/gnu-nat.c | 7 -------
gdb/inf-ptrace.c | 6 +++---
gdb/infcmd.c | 2 +-
gdb/linux-nat.c | 12 +++++++++---
gdb/monitor.c | 4 ++--
gdb/nto-procfs.c | 14 +++++++-------
gdb/procfs.c | 4 ++--
gdb/remote-m32r-sdi.c | 6 +++---
gdb/remote-sim.c | 20 ++++++++++----------
gdb/remote.c | 44 +++++++++++++++++++++++++++++++++-----------
gdb/target-delegates.c | 28 ++++++++++++++++++++++++++++
gdb/target.c | 12 ++++++++++++
gdb/target.h | 8 ++++++++
gdb/windows-nat.c | 6 +++---
15 files changed, 124 insertions(+), 55 deletions(-)
diff --git a/gdb/darwin-nat.c b/gdb/darwin-nat.c
index e1acc05..30e968f 100644
--- a/gdb/darwin-nat.c
+++ b/gdb/darwin-nat.c
@@ -83,7 +83,7 @@
#define PTRACE(CMD, PID, ADDR, SIG) \
darwin_ptrace(#CMD, CMD, (PID), (ADDR), (SIG))
-static void darwin_stop (struct target_ops *self, ptid_t);
+static void darwin_interrupt (struct target_ops *self, ptid_t);
static void darwin_resume_to (struct target_ops *ops, ptid_t ptid, int step,
enum gdb_signal signal);
@@ -1198,7 +1198,7 @@ darwin_wait_to (struct target_ops *ops,
}
static void
-darwin_stop (struct target_ops *self, ptid_t t)
+darwin_interrupt (struct target_ops *self, ptid_t t)
{
struct inferior *inf = current_inferior ();
@@ -2156,7 +2156,7 @@ _initialize_darwin_inferior (void)
darwin_ops->to_wait = darwin_wait_to;
darwin_ops->to_mourn_inferior = darwin_mourn_inferior;
darwin_ops->to_kill = darwin_kill_inferior;
- darwin_ops->to_stop = darwin_stop;
+ darwin_ops->to_interrupt = darwin_interrupt;
darwin_ops->to_resume = darwin_resume_to;
darwin_ops->to_thread_alive = darwin_thread_alive;
darwin_ops->to_pid_to_str = darwin_pid_to_str;
diff --git a/gdb/gnu-nat.c b/gdb/gnu-nat.c
index d830773..4753634 100644
--- a/gdb/gnu-nat.c
+++ b/gdb/gnu-nat.c
@@ -2271,12 +2271,6 @@ gnu_terminal_init (struct target_ops *self)
child_terminal_init_with_pgrp (gnu_current_inf->pid);
}
-static void
-gnu_stop (struct target_ops *self, ptid_t ptid)
-{
- error (_("to_stop target function not implemented"));
-}
-
static int
gnu_thread_alive (struct target_ops *ops, ptid_t ptid)
{
@@ -2686,7 +2680,6 @@ gnu_target (void)
t->to_mourn_inferior = gnu_mourn_inferior;
t->to_thread_alive = gnu_thread_alive;
t->to_pid_to_str = gnu_pid_to_str;
- t->to_stop = gnu_stop;
return t;
}
diff --git a/gdb/inf-ptrace.c b/gdb/inf-ptrace.c
index cd58dfb..17ba903 100644
--- a/gdb/inf-ptrace.c
+++ b/gdb/inf-ptrace.c
@@ -288,10 +288,10 @@ inf_ptrace_kill (struct target_ops *ops)
target_mourn_inferior ();
}
-/* Stop the inferior. */
+/* Interrupt the inferior. */
static void
-inf_ptrace_stop (struct target_ops *self, ptid_t ptid)
+inf_ptrace_interrupt (struct target_ops *self, ptid_t ptid)
{
/* Send a SIGINT to the process group. This acts just like the user
typed a ^C on the controlling terminal. Note that using a
@@ -686,7 +686,7 @@ inf_ptrace_target (void)
t->to_mourn_inferior = inf_ptrace_mourn_inferior;
t->to_thread_alive = inf_ptrace_thread_alive;
t->to_pid_to_str = inf_ptrace_pid_to_str;
- t->to_stop = inf_ptrace_stop;
+ t->to_interrupt = inf_ptrace_interrupt;
t->to_xfer_partial = inf_ptrace_xfer_partial;
#if defined (PT_IO) && defined (PIOD_READ_AUXV)
t->to_auxv_parse = inf_ptrace_auxv_parse;
diff --git a/gdb/infcmd.c b/gdb/infcmd.c
index 3505b1b..0a3c515 100644
--- a/gdb/infcmd.c
+++ b/gdb/infcmd.c
@@ -2805,7 +2805,7 @@ interrupt_target_1 (int all_threads)
ptid = minus_one_ptid;
else
ptid = inferior_ptid;
- target_stop (ptid);
+ target_interrupt (ptid);
/* Tag the thread as having been explicitly requested to stop, so
other parts of gdb know not to resume this thread automatically,
diff --git a/gdb/linux-nat.c b/gdb/linux-nat.c
index b449485..d6fd40d 100644
--- a/gdb/linux-nat.c
+++ b/gdb/linux-nat.c
@@ -4787,10 +4787,16 @@ linux_nat_stop_lwp (struct lwp_info *lwp, void *data)
static void
linux_nat_stop (struct target_ops *self, ptid_t ptid)
{
- if (target_is_non_stop_p ())
+ iterate_over_lwps (ptid, linux_nat_stop_lwp, NULL);
+}
+
+static void
+linux_nat_interrupt (struct target_ops *self, ptid_t ptid)
+{
+ if (non_stop)
iterate_over_lwps (ptid, linux_nat_stop_lwp, NULL);
else
- linux_ops->to_stop (linux_ops, ptid);
+ linux_ops->to_interrupt (linux_ops, ptid);
}
static void
@@ -4894,8 +4900,8 @@ linux_nat_add_target (struct target_ops *t)
super_close = t->to_close;
t->to_close = linux_nat_close;
- /* Methods for non-stop support. */
t->to_stop = linux_nat_stop;
+ t->to_interrupt = linux_nat_interrupt;
t->to_supports_multi_process = linux_nat_supports_multi_process;
diff --git a/gdb/monitor.c b/gdb/monitor.c
index 548dae3..e47d479 100644
--- a/gdb/monitor.c
+++ b/gdb/monitor.c
@@ -2267,7 +2267,7 @@ monitor_load (struct target_ops *self, const char *args, int from_tty)
}
static void
-monitor_stop (struct target_ops *self, ptid_t ptid)
+monitor_interrupt (struct target_ops *self, ptid_t ptid)
{
monitor_debug ("MON stop\n");
if ((current_monitor->flags & MO_SEND_BREAK_ON_STOP) != 0)
@@ -2375,7 +2375,7 @@ init_base_monitor_ops (void)
monitor_ops.to_load = monitor_load;
monitor_ops.to_create_inferior = monitor_create_inferior;
monitor_ops.to_mourn_inferior = monitor_mourn_inferior;
- monitor_ops.to_stop = monitor_stop;
+ monitor_ops.to_interrupt = monitor_interrupt;
monitor_ops.to_rcmd = monitor_rcmd;
monitor_ops.to_log_command = serial_log_command;
monitor_ops.to_thread_alive = monitor_thread_alive;
diff --git a/gdb/nto-procfs.c b/gdb/nto-procfs.c
index 7312060..c318f19 100644
--- a/gdb/nto-procfs.c
+++ b/gdb/nto-procfs.c
@@ -692,20 +692,20 @@ Give up (and stop debugging it)? ")))
/* The user typed ^C twice. */
static void
-nto_interrupt_twice (int signo)
+nto_handle_sigint_twice (int signo)
{
signal (signo, ofunc);
interrupt_query ();
- signal (signo, nto_interrupt_twice);
+ signal (signo, nto_handle_sigint_twice);
}
static void
-nto_interrupt (int signo)
+nto_handle_sigint (int signo)
{
/* If this doesn't work, try more severe steps. */
- signal (signo, nto_interrupt_twice);
+ signal (signo, nto_handle_sigint_twice);
- target_stop (inferior_ptid);
+ target_interrupt (inferior_ptid);
}
static ptid_t
@@ -733,7 +733,7 @@ procfs_wait (struct target_ops *ops,
devctl (ctl_fd, DCMD_PROC_STATUS, &status, sizeof (status), 0);
while (!(status.flags & _DEBUG_FLAG_ISTOP))
{
- ofunc = (void (*)()) signal (SIGINT, nto_interrupt);
+ ofunc = (void (*)()) signal (SIGINT, nto_handle_sigint);
sigwaitinfo (&set, &info);
signal (SIGINT, ofunc);
devctl (ctl_fd, DCMD_PROC_STATUS, &status, sizeof (status), 0);
@@ -1444,7 +1444,7 @@ init_procfs_targets (void)
t->to_thread_alive = procfs_thread_alive;
t->to_update_thread_list = procfs_update_thread_list;
t->to_pid_to_str = procfs_pid_to_str;
- t->to_stop = procfs_stop;
+ t->to_interrupt = procfs_interrupt;
t->to_have_continuable_watchpoint = 1;
t->to_extra_thread_info = nto_extra_thread_info;
diff --git a/gdb/procfs.c b/gdb/procfs.c
index b62539f..e540e2a 100644
--- a/gdb/procfs.c
+++ b/gdb/procfs.c
@@ -194,7 +194,7 @@ procfs_target (void)
t->to_xfer_partial = procfs_xfer_partial;
t->to_pass_signals = procfs_pass_signals;
t->to_files_info = procfs_files_info;
- t->to_stop = procfs_stop;
+ t->to_interrupt = procfs_interrupt;
t->to_update_thread_list = procfs_update_thread_list;
t->to_thread_alive = procfs_thread_alive;
@@ -4204,7 +4204,7 @@ procfs_files_info (struct target_ops *ignore)
kill(SIGINT) to the child's process group. */
static void
-procfs_stop (struct target_ops *self, ptid_t ptid)
+procfs_interrupt (struct target_ops *self, ptid_t ptid)
{
kill (-inferior_process_group (), SIGINT);
}
diff --git a/gdb/remote-m32r-sdi.c b/gdb/remote-m32r-sdi.c
index 01cb5b6..b246cd9 100644
--- a/gdb/remote-m32r-sdi.c
+++ b/gdb/remote-m32r-sdi.c
@@ -1407,10 +1407,10 @@ m32r_load (struct target_ops *self, const char *args, int from_tty)
}
static void
-m32r_stop (struct target_ops *self, ptid_t ptid)
+m32r_interrupt (struct target_ops *self, ptid_t ptid)
{
if (remote_debug)
- fprintf_unfiltered (gdb_stdlog, "m32r_stop()\n");
+ fprintf_unfiltered (gdb_stdlog, "m32r_interrupt()\n");
send_cmd (SDI_STOP_CPU);
@@ -1664,7 +1664,7 @@ init_m32r_ops (void)
m32r_ops.to_load = m32r_load;
m32r_ops.to_create_inferior = m32r_create_inferior;
m32r_ops.to_mourn_inferior = m32r_mourn_inferior;
- m32r_ops.to_stop = m32r_stop;
+ m32r_ops.to_interrupt = m32r_interrupt;
m32r_ops.to_log_command = serial_log_command;
m32r_ops.to_thread_alive = m32r_thread_alive;
m32r_ops.to_pid_to_str = m32r_pid_to_str;
diff --git a/gdb/remote-sim.c b/gdb/remote-sim.c
index fd2fd58..766356a 100644
--- a/gdb/remote-sim.c
+++ b/gdb/remote-sim.c
@@ -87,7 +87,7 @@ static void gdbsim_files_info (struct target_ops *target);
static void gdbsim_mourn_inferior (struct target_ops *target);
-static void gdbsim_stop (struct target_ops *self, ptid_t ptid);
+static void gdbsim_interrupt (struct target_ops *self, ptid_t ptid);
void simulator_command (char *args, int from_tty);
@@ -888,17 +888,17 @@ gdbsim_resume (struct target_ops *ops,
error (_("The program is not being run."));
}
-/* Notify the simulator of an asynchronous request to stop.
+/* Notify the simulator of an asynchronous request to interrupt.
- The simulator shall ensure that the stop request is eventually
+ The simulator shall ensure that the interrupt request is eventually
delivered to the simulator. If the call is made while the
- simulator is not running then the stop request is processed when
+ simulator is not running then the interrupt request is processed when
the simulator is next resumed.
For simulators that do not support this operation, just abort. */
static int
-gdbsim_stop_inferior (struct inferior *inf, void *arg)
+gdbsim_interrupt_inferior (struct inferior *inf, void *arg)
{
struct sim_inferior_data *sim_data
= get_sim_inferior_data (inf, SIM_INSTANCE_NEEDED);
@@ -918,13 +918,13 @@ gdbsim_stop_inferior (struct inferior *inf, void *arg)
}
static void
-gdbsim_stop (struct target_ops *self, ptid_t ptid)
+gdbsim_interrupt (struct target_ops *self, ptid_t ptid)
{
struct sim_inferior_data *sim_data;
if (ptid_equal (ptid, minus_one_ptid))
{
- iterate_over_inferiors (gdbsim_stop_inferior, NULL);
+ iterate_over_inferiors (gdbsim_interrupt_inferior, NULL);
}
else
{
@@ -934,7 +934,7 @@ gdbsim_stop (struct target_ops *self, ptid_t ptid)
error (_("Can't stop pid %d. No inferior found."),
ptid_get_pid (ptid));
- gdbsim_stop_inferior (inf, NULL);
+ gdbsim_interrupt_inferior (inf, NULL);
}
}
@@ -962,7 +962,7 @@ gdb_os_poll_quit (host_callback *p)
static void
gdbsim_cntrl_c (int signo)
{
- gdbsim_stop (NULL, minus_one_ptid);
+ gdbsim_interrupt (NULL, minus_one_ptid);
}
static ptid_t
@@ -1320,7 +1320,7 @@ init_gdbsim_ops (void)
gdbsim_ops.to_load = gdbsim_load;
gdbsim_ops.to_create_inferior = gdbsim_create_inferior;
gdbsim_ops.to_mourn_inferior = gdbsim_mourn_inferior;
- gdbsim_ops.to_stop = gdbsim_stop;
+ gdbsim_ops.to_interrupt = gdbsim_interrupt;
gdbsim_ops.to_thread_alive = gdbsim_thread_alive;
gdbsim_ops.to_pid_to_str = gdbsim_pid_to_str;
gdbsim_ops.to_stratum = process_stratum;
diff --git a/gdb/remote.c b/gdb/remote.c
index 18492bf..3d61869 100644
--- a/gdb/remote.c
+++ b/gdb/remote.c
@@ -5002,11 +5002,12 @@ async_cleanup_sigint_signal_handler (void *dummy)
packet. */
static void (*ofunc) (int);
-/* The command line interface's stop routine. This function is installed
- as a signal handler for SIGINT. The first time a user requests a
- stop, we call remote_stop to send a break or ^C. If there is no
+/* The command line interface's interrupt routine. This function is installed
+ as a signal handler for SIGINT. The first time a user requests an
+ interrupt, we call remote_interrupt to send a break or ^C. If there is no
response from the target (it didn't stop when the user requested it),
we ask the user if he'd like to detach from the target. */
+
static void
sync_remote_interrupt (int signo)
{
@@ -5077,12 +5078,12 @@ remote_stop_ns (ptid_t ptid)
error (_("Stopping %s failed: %s"), target_pid_to_str (ptid), rs->buf);
}
-/* All-stop version of target_stop. Sends a break or a ^C to stop the
- remote target. It is undefined which thread of which process
- reports the stop. */
+/* All-stop version of target_interrupt. Sends a break or a ^C to
+ interrupt the remote target. It is undefined which thread of which
+ process reports the interrupt. */
static void
-remote_stop_as (ptid_t ptid)
+remote_interrupt_as (ptid_t ptid)
{
struct remote_state *rs = get_remote_state ();
@@ -5098,9 +5099,7 @@ remote_stop_as (ptid_t ptid)
send_interrupt_sequence ();
}
-/* This is the generic stop called via the target vector. When a target
- interrupt is requested, either by the command line or the GUI, we
- will eventually end up here. */
+/* Implement the to_stop function for the remote targets. */
static void
remote_stop (struct target_ops *self, ptid_t ptid)
@@ -5111,7 +5110,29 @@ remote_stop (struct target_ops *self, ptid_t ptid)
if (non_stop)
remote_stop_ns (ptid);
else
- remote_stop_as (ptid);
+ {
+ /* We don't currently have a way to transparently pause the
+ remote target in all-stop mode. Interrupt it instead. */
+ remote_interrupt_as (ptid);
+ }
+}
+
+/* Implement the to_interrupt function for the remote targets. */
+
+static void
+remote_interrupt (struct target_ops *self, ptid_t ptid)
+{
+ if (remote_debug)
+ fprintf_unfiltered (gdb_stdlog, "remote_interrupt called\n");
+
+ if (non_stop)
+ {
+ /* We don't currently have a way to ^C the remote target in
+ non-stop mode. Stop it (with no signal) instead. */
+ remote_stop_ns (ptid);
+ }
+ else
+ remote_interrupt_as (ptid);
}
/* Ask the user what to do when an interrupt is received. */
@@ -11739,6 +11760,7 @@ Specify the serial device it is connected to\n\
remote_ops.to_extra_thread_info = remote_threads_extra_info;
remote_ops.to_get_ada_task_ptid = remote_get_ada_task_ptid;
remote_ops.to_stop = remote_stop;
+ remote_ops.to_interrupt = remote_interrupt;
remote_ops.to_xfer_partial = remote_xfer_partial;
remote_ops.to_rcmd = remote_rcmd;
remote_ops.to_pid_to_exec_file = remote_pid_to_exec_file;
diff --git a/gdb/target-delegates.c b/gdb/target-delegates.c
index 8a92acf..054ad32 100644
--- a/gdb/target-delegates.c
+++ b/gdb/target-delegates.c
@@ -1537,6 +1537,30 @@ debug_stop (struct target_ops *self, ptid_t arg1)
}
static void
+delegate_interrupt (struct target_ops *self, ptid_t arg1)
+{
+ self = self->beneath;
+ self->to_interrupt (self, arg1);
+}
+
+static void
+tdefault_interrupt (struct target_ops *self, ptid_t arg1)
+{
+}
+
+static void
+debug_interrupt (struct target_ops *self, ptid_t arg1)
+{
+ fprintf_unfiltered (gdb_stdlog, "-> %s->to_interrupt (...)\n", debug_target.to_shortname);
+ debug_target.to_interrupt (&debug_target, arg1);
+ fprintf_unfiltered (gdb_stdlog, "<- %s->to_interrupt (", debug_target.to_shortname);
+ target_debug_print_struct_target_ops_p (&debug_target);
+ fputs_unfiltered (", ", gdb_stdlog);
+ target_debug_print_ptid_t (arg1);
+ fputs_unfiltered (")\n", gdb_stdlog);
+}
+
+static void
delegate_rcmd (struct target_ops *self, const char *arg1, struct ui_file *arg2)
{
self = self->beneath;
@@ -4016,6 +4040,8 @@ install_delegators (struct target_ops *ops)
ops->to_thread_name = delegate_thread_name;
if (ops->to_stop == NULL)
ops->to_stop = delegate_stop;
+ if (ops->to_interrupt == NULL)
+ ops->to_interrupt = delegate_interrupt;
if (ops->to_rcmd == NULL)
ops->to_rcmd = delegate_rcmd;
if (ops->to_pid_to_exec_file == NULL)
@@ -4253,6 +4279,7 @@ install_dummy_methods (struct target_ops *ops)
ops->to_extra_thread_info = tdefault_extra_thread_info;
ops->to_thread_name = tdefault_thread_name;
ops->to_stop = tdefault_stop;
+ ops->to_interrupt = tdefault_interrupt;
ops->to_rcmd = default_rcmd;
ops->to_pid_to_exec_file = tdefault_pid_to_exec_file;
ops->to_log_command = tdefault_log_command;
@@ -4402,6 +4429,7 @@ init_debug_target (struct target_ops *ops)
ops->to_extra_thread_info = debug_extra_thread_info;
ops->to_thread_name = debug_thread_name;
ops->to_stop = debug_stop;
+ ops->to_interrupt = debug_interrupt;
ops->to_rcmd = debug_rcmd;
ops->to_pid_to_exec_file = debug_pid_to_exec_file;
ops->to_log_command = debug_log_command;
diff --git a/gdb/target.c b/gdb/target.c
index ab6c977..2262131 100644
--- a/gdb/target.c
+++ b/gdb/target.c
@@ -3244,6 +3244,18 @@ target_stop (ptid_t ptid)
(*current_target.to_stop) (¤t_target, ptid);
}
+void
+target_interrupt (ptid_t ptid)
+{
+ if (!may_stop)
+ {
+ warning (_("May not interrupt or stop the target, ignoring attempt"));
+ return;
+ }
+
+ (*current_target.to_interrupt) (¤t_target, ptid);
+}
+
/* See target/target.h. */
void
diff --git a/gdb/target.h b/gdb/target.h
index 17dee43..276b5e2 100644
--- a/gdb/target.h
+++ b/gdb/target.h
@@ -611,6 +611,8 @@ struct target_ops
TARGET_DEFAULT_RETURN (NULL);
void (*to_stop) (struct target_ops *, ptid_t)
TARGET_DEFAULT_IGNORE ();
+ void (*to_interrupt) (struct target_ops *, ptid_t)
+ TARGET_DEFAULT_IGNORE ();
void (*to_rcmd) (struct target_ops *,
const char *command, struct ui_file *output)
TARGET_DEFAULT_FUNC (default_rcmd);
@@ -1635,6 +1637,12 @@ extern void target_update_thread_list (void);
extern void target_stop (ptid_t ptid);
+/* Interrupt the target just like the user typed a ^C on the
+ inferior's controlling terminal. (For instance, under Unix, this
+ should act like SIGINT). This function is asynchronous. */
+
+extern void target_interrupt (ptid_t ptid);
+
/* Send the specified COMMAND to the target's monitor
(shell,interpreter) for execution. The result of the query is
placed in OUTBUF. */
diff --git a/gdb/windows-nat.c b/gdb/windows-nat.c
index fd31083..46a378c 100644
--- a/gdb/windows-nat.c
+++ b/gdb/windows-nat.c
@@ -162,7 +162,7 @@ static int windows_initialization_done;
#define DEBUG_MEM(x) if (debug_memory) printf_unfiltered x
#define DEBUG_EXCEPT(x) if (debug_exceptions) printf_unfiltered x
-static void windows_stop (struct target_ops *self, ptid_t);
+static void windows_interrupt (struct target_ops *self, ptid_t);
static int windows_thread_alive (struct target_ops *, ptid_t);
static void windows_kill_inferior (struct target_ops *);
@@ -2294,7 +2294,7 @@ windows_mourn_inferior (struct target_ops *ops)
^C on the controlling terminal. */
static void
-windows_stop (struct target_ops *self, ptid_t ptid)
+windows_interrupt (struct target_ops *self, ptid_t ptid)
{
DEBUG_EVENTS (("gdb: GenerateConsoleCtrlEvent (CTRLC_EVENT, 0)\n"));
CHECK (GenerateConsoleCtrlEvent (CTRL_C_EVENT, current_event.dwProcessId));
@@ -2488,7 +2488,7 @@ windows_target (void)
t->to_mourn_inferior = windows_mourn_inferior;
t->to_thread_alive = windows_thread_alive;
t->to_pid_to_str = windows_pid_to_str;
- t->to_stop = windows_stop;
+ t->to_interrupt = windows_interrupt;
t->to_pid_to_exec_file = windows_pid_to_exec_file;
t->to_get_ada_task_ptid = windows_get_ada_task_ptid;
t->to_get_tib_address = windows_get_tib_address;
--
1.9.3
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 12/17] Fix interrupt-noterm.exp on targets always in non-stop
2015-04-17 10:52 ` [PATCH v3 12/17] Fix interrupt-noterm.exp on targets always in non-stop Pedro Alves
@ 2015-04-21 11:40 ` Yao Qi
2015-04-22 20:03 ` Pedro Alves
0 siblings, 1 reply; 62+ messages in thread
From: Yao Qi @ 2015-04-21 11:40 UTC (permalink / raw)
To: Pedro Alves; +Cc: gdb-patches
Pedro Alves <palves@redhat.com> writes:
> Most of the patch is an across-the-board rename of to_stop hook
> implementations to to_interrupt. The only targets where something
except on gnu-nat.c,
> more than a rename is being done are linux-nat.c and remote.c, which
> are the only targets that support async, and thus are the only ones
> the core side calls target_stop on.
>
> gdb/ChangeLog:
> 2015-04-17 Pedro Alves <palves@redhat.com>
>
> * darwin-nat.c (darwin_stop): Rename to ...
> (darwin_interrupt): ... this.
> (_initialize_darwin_inferior): Adjust.
> * gnu-nat.c (gnu_stop): Delete.
> (gnu_thread_alive): Don't install gnu_stop.
^^^^^^^^^^^^^^^^^ gnu_target
> diff --git a/gdb/gnu-nat.c b/gdb/gnu-nat.c
> index d830773..4753634 100644
> --- a/gdb/gnu-nat.c
> +++ b/gdb/gnu-nat.c
> @@ -2271,12 +2271,6 @@ gnu_terminal_init (struct target_ops *self)
> child_terminal_init_with_pgrp (gnu_current_inf->pid);
> }
>
> -static void
> -gnu_stop (struct target_ops *self, ptid_t ptid)
> -{
> - error (_("to_stop target function not implemented"));
> -}
> -
> static int
> gnu_thread_alive (struct target_ops *ops, ptid_t ptid)
> {
> @@ -2686,7 +2680,6 @@ gnu_target (void)
> t->to_mourn_inferior = gnu_mourn_inferior;
> t->to_thread_alive = gnu_thread_alive;
> t->to_pid_to_str = gnu_pid_to_str;
> - t->to_stop = gnu_stop;
>
> return t;
> }
Any reason you don't install to_interrupt for gnu-nat?
--
Yao (齐尧)
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 12/17] Fix interrupt-noterm.exp on targets always in non-stop
2015-04-21 11:40 ` Yao Qi
@ 2015-04-22 20:03 ` Pedro Alves
0 siblings, 0 replies; 62+ messages in thread
From: Pedro Alves @ 2015-04-22 20:03 UTC (permalink / raw)
To: Yao Qi; +Cc: gdb-patches
On 04/21/2015 12:40 PM, Yao Qi wrote:
> Pedro Alves <palves@redhat.com> writes:
>
>> Most of the patch is an across-the-board rename of to_stop hook
>> implementations to to_interrupt. The only targets where something
>
> except on gnu-nat.c,
Ah, yes.
>
>> more than a rename is being done are linux-nat.c and remote.c, which
>> are the only targets that support async, and thus are the only ones
>> the core side calls target_stop on.
>>
>> gdb/ChangeLog:
>> 2015-04-17 Pedro Alves <palves@redhat.com>
>>
>> * darwin-nat.c (darwin_stop): Rename to ...
>> (darwin_interrupt): ... this.
>> (_initialize_darwin_inferior): Adjust.
>> * gnu-nat.c (gnu_stop): Delete.
>> (gnu_thread_alive): Don't install gnu_stop.
> ^^^^^^^^^^^^^^^^^ gnu_target
Whoops, fixed.
>
>> diff --git a/gdb/gnu-nat.c b/gdb/gnu-nat.c
>> index d830773..4753634 100644
>> --- a/gdb/gnu-nat.c
>> +++ b/gdb/gnu-nat.c
>> @@ -2271,12 +2271,6 @@ gnu_terminal_init (struct target_ops *self)
>> child_terminal_init_with_pgrp (gnu_current_inf->pid);
>> }
>>
>> -static void
>> -gnu_stop (struct target_ops *self, ptid_t ptid)
>> -{
>> - error (_("to_stop target function not implemented"));
>> -}
>> -
>> static int
>> gnu_thread_alive (struct target_ops *ops, ptid_t ptid)
>> {
>> @@ -2686,7 +2680,6 @@ gnu_target (void)
>> t->to_mourn_inferior = gnu_mourn_inferior;
>> t->to_thread_alive = gnu_thread_alive;
>> t->to_pid_to_str = gnu_pid_to_str;
>> - t->to_stop = gnu_stop;
>>
>> return t;
>> }
>
> Any reason you don't install to_interrupt for gnu-nat?
I'll sent the gnu-nat change as a separate patch, along with
the rationale.
--
Pedro Alves
^ permalink raw reply [flat|nested] 62+ messages in thread
* [PATCH v3 09/17] Teach non-stop to do in-line step-overs (stop all, step, restart)
2015-04-17 10:47 [PATCH v3 00/23] All-stop on top of non-stop Pedro Alves
` (12 preceding siblings ...)
2015-04-17 10:52 ` [PATCH v3 12/17] Fix interrupt-noterm.exp on targets always in non-stop Pedro Alves
@ 2015-04-17 10:52 ` Pedro Alves
2015-04-17 11:01 ` Pedro Alves
` (2 more replies)
2015-04-17 10:56 ` [PATCH v3 14/17] Disable displaced stepping if trying it fails Pedro Alves
` (4 subsequent siblings)
18 siblings, 3 replies; 62+ messages in thread
From: Pedro Alves @ 2015-04-17 10:52 UTC (permalink / raw)
To: gdb-patches
That is, step past breakpoints by:
- pausing all threads
- removing breakpoint at PC
- single-step
- reinsert breakpoint
- restart threads
similarly to all-stop (with displaced stepping disabled). This allows
non-stop to work on targets/architectures without displaced stepping
support. That is, it makes displaced stepping an optimization instead
of a requirement. For example, in principle, all GNU/Linux ports
support non-stop mode at the target_ops level, but not all
corresponding gdbarch's implement displaced stepping. This should
make non-stop work for all (albeit, not as efficiently). And then
there are scenarios where even if the architecture supports displaced
stepping, we can't use it, because we e.g., don't find a usable
address to use as displaced step scratch pad. It should also fix
stepping past watchpoints on targets that have non-continuable
watchpoints in non-stop mode (e.g., PPC, untested). Running the
instruction out of line in the displaced stepping scratch pad doesn't
help that case, as the copied instruction reads/writes the same
watched memory... We can fix that too by teaching GDB to only remove
the watchpoint from the thread that we want to move past the
watchpoint (currently, removing a watchpoint always removes it from
all threads), but again, that can be considered an optimization; not
all targets would support it.
For those familiar with the gdb and gdbserver Linux target_ops
backends, the implementation should look similar, except it is done on
the core side. When we pause threads, we may find they stop with an
interesting event that should be handled later when the thread is
re-resumed, thus we store such events in the thread object, and mark
the event as pending. We should only consume pending events if the
thread is indeed resumed, thus we add a new "resumed" flag to the
thread object. At a later stage, we might add new target methods to
accelerate some of this, like "pause all threads", with corresponding
RSP packets, but we'd still need a fallback method for remote targets
that don't support such packets, so, again, that can be deferred as
optimization.
My _real_ motivation here is making it possible to reimplement
all-stop mode on top of the target always working on non-stop mode, so
that e.g., we can send RSP packets to a remote target even while the
target is running -- can't do that in the all-stop RSP variant, by
design).
Tested on x86_64 Fedora 20, with and without "set displaced off"
forced. The latter forces the new code paths whenever GDB needs to
step past a breakpoint.
gdb/ChangeLog:
2015-04-17 Pedro Alves <pedro@codesourcery.com>
* breakpoint.c (breakpoints_should_be_inserted_now): If any thread
has a pending status, return true.
* gdbthread.h: Include target/waitstatus.h.
(struct thread_suspend_state) <stop_reason, waitstatus_pending_p,
stop_pc>: New fields.
(struct thread_info) <resumed>: New field.
(set_resumed): Declare.
* infrun.c: Include "event-loop.h".
(infrun_async_inferior_event_token, infrun_is_async): New globals.
(infrun_async): New function.
(clear_step_over_info): Add debug output.
(displaced_step_in_progress_any_inferior): New function.
(displaced_step_fixup): New returns int.
(start_step_over): Handle in-line step-overs too. Assert the
thread is marked resumed.
(resume_cleanups): Clear the thread's resumed flag.
(resume): Set the thread's resumed flag. Return early if the
thread has a pending status. Allow stepping a breakpoint with no
signal.
(proceed): Adjust to check 'resumed' instead of 'executing'.
(clear_proceed_status_thread): If the thread has a pending status,
and that status is a finished step, discard the pending status.
(clear_proceed_status): Don't clear step_over_info here.
(random_pending_event_thread, do_target_wait): New functions.
(prepare_for_detach, wait_for_inferior, fetch_inferior_event): Use
do_target_wait.
(wait_one): New function.
(THREAD_STOPPED_BY): New macro.
(thread_stopped_by_watchpoint, thread_stopped_by_sw_breakpoint)
(thread_stopped_by_hw_breakpoint): New functions.
(switch_to_thread_cleanup, save_waitstatus, stop_all_threads): New
functions.
(handle_inferior_event): Also call set_resumed(false) on all
threads implicitly stopped by the event.
(restart_threads, resumed_thread_with_pending_status): New
functions.
(finish_step_over): If we were doing an in-line step-over before,
and no longer are after trying to start a new step-over, restart
all threads. If we have multiple threads with pending events,
save the current event and go through the event loop again.
(handle_signal_stop): Return early if finish_step_over returns
false.
<random signal>: Clear step_over_info before delivering
the signal.
(keep_going_stepped_thread): Use internal_error instead of
gdb_assert. Mark the thread as resumed.
(keep_going_pass): Assert the thread isn't already resumed. If
some other thread is doing an in-line step-over, defer the resume.
If we just started a new in-line step-over, stop all threads.
Don't clear step_over_info.
(infrun_async_inferior_event_handler): New function.
(_initialize_infrun): Create async event handler with
infrun_async_inferior_event_handler as callback.
(infrun_async): New declaration.
* target.c (target_async): New function.
* target.h (target_async): Declare macro and readd as function
declaration.
* target/waitstatus.h (enum target_stop_reason)
<TARGET_STOPPED_BY_SINGLE_STEP>: New value.
* thread.c (new_thread): Clear the new waitstatus field.
(set_resumed): New function.
v3:
- pending status handing moved from keep_going_stepped_thread to
restart_threads. Fixes an assert triggered while testing with sss
and no displaced stepping.
- proceed and restart_threads now check "resumed" instead of
"executing", because the thread that was started might have a
pending status, and thus ends resumed, but not executing. Caught
by an assertion in keep_going_stepped_thread.
- do_target_wait was missing a call to mark the infrun event source
in case we are returning a pending wait status and there are other
pending events to process. Caught by testing with gdbserver + no
displaced stepping + sss.
- had finish_step_over detect that there are multiple pending events
that could be consumed after an in-line step-over finishes, and
pick one at random. Fixes non-stop-fair-events.exp timeout due to
event starvation on targets without displaced stepping.
- start_step_over now waits for all displaced steps in progress
before starting an in-line step over (avoid starvation/bouncing
effects caused by pausing threads doing the displaced step, thus
cancelling the displaced step, only to restart it again later).
- resume now only sets the resume flag on success, instead of
undoing it on error. avoids the need for resume_cleanups change
in v2.
- Adjust to use the new thread_is_in_step_over_chain.
v2:
- `proceed' changes dropped. no longer needed after fix to previous patch.
- "infrun.c" -> "infrun" in debug output.
- clear trap_expected when displaced_step_fixup returns -1, when
stopping threads (as the step-over was cancelled). Testing on
PPC64 caught this was missing ("inconsistent state" internal error
in start_step_over_inferior).
---
gdb/breakpoint.c | 9 +
gdb/gdbthread.h | 31 ++
gdb/infrun.c | 1058 ++++++++++++++++++++++++++++++++++++++++++++---
gdb/infrun.h | 3 +
gdb/target.c | 9 +
gdb/target.h | 3 +-
gdb/target/waitstatus.h | 5 +-
gdb/thread.c | 23 ++
8 files changed, 1085 insertions(+), 56 deletions(-)
diff --git a/gdb/breakpoint.c b/gdb/breakpoint.c
index 31869f1..e23d223 100644
--- a/gdb/breakpoint.c
+++ b/gdb/breakpoint.c
@@ -468,6 +468,8 @@ breakpoints_should_be_inserted_now (void)
}
else if (target_has_execution)
{
+ struct thread_info *tp;
+
if (always_inserted_mode)
{
/* The user wants breakpoints inserted even if all threads
@@ -477,6 +479,13 @@ breakpoints_should_be_inserted_now (void)
if (threads_are_executing ())
return 1;
+
+ /* Don't remove breakpoints yet if, even though all threads are
+ stopped, we still have events to process. */
+ ALL_NON_EXITED_THREADS (tp)
+ if (tp->resumed
+ && tp->suspend.waitstatus_pending_p)
+ return 1;
}
return 0;
}
diff --git a/gdb/gdbthread.h b/gdb/gdbthread.h
index 7052ee1..22c8eb2 100644
--- a/gdb/gdbthread.h
+++ b/gdb/gdbthread.h
@@ -29,6 +29,7 @@ struct symtab;
#include "inferior.h"
#include "btrace.h"
#include "common/vec.h"
+#include "target/waitstatus.h"
/* Frontend view of the thread state. Possible extensions: stepping,
finishing, until(ling),... */
@@ -159,6 +160,23 @@ struct thread_suspend_state
should be suppressed, the core will take care of clearing this
before the target is resumed. */
enum gdb_signal stop_signal;
+
+ /* The reason the thread last stopped, if we need to track it
+ (breakpoint, watchpoint, etc.) */
+ enum target_stop_reason stop_reason;
+
+ /* The waitstatus for this thread's last event. */
+ struct target_waitstatus waitstatus;
+ /* If true WAITSTATUS hasn't been handled yet. */
+ int waitstatus_pending_p;
+
+ /* Record the pc of the thread the last time it stopped. (This is
+ not the current thread's PC as that may have changed since the
+ last stop, e.g., "return" command, or "p $pc = 0xf000"). This
+ used in coordination with stop_reason and waitstatus_pending_p:
+ if the thread's PC is changed since it last stopped, a pending
+ breakpoint waitstatus is discarded. */
+ CORE_ADDR stop_pc;
};
typedef struct value *value_ptr;
@@ -183,6 +201,14 @@ struct thread_info
thread is off and running. */
int executing;
+ /* Non-zero if this thread will be/has been resumed. Note that a
+ thread can be marked both as not-executing and resumed at the
+ same time. This happens if we try to resume a thread that has a
+ wait status pending. We shouldn't let the thread run until that
+ wait status has been processed, but we should not process that
+ wait status if we didn't try to let the thread run. */
+ int resumed;
+
/* Frontend view of the thread state. Note that the THREAD_RUNNING/
THREAD_STOPPED states are different from EXECUTING. When the
thread is stopped internally while handling an internal event,
@@ -400,6 +426,11 @@ extern int thread_count (void);
/* Switch from one thread to another. */
extern void switch_to_thread (ptid_t ptid);
+/* Marks or clears thread(s) PTID as resumed. If PTID is
+ MINUS_ONE_PTID, applies to all threads. If ptid_is_pid(PTID) is
+ true, applies to all threads of the process pointed at by PTID. */
+extern void set_resumed (ptid_t ptid, int resumed);
+
/* Marks thread PTID is running, or stopped.
If PTID is minus_one_ptid, marks all threads. */
extern void set_running (ptid_t ptid, int running);
diff --git a/gdb/infrun.c b/gdb/infrun.c
index 6415fa2..8859b9f 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -60,6 +60,36 @@
#include "target-descriptions.h"
#include "target-dcache.h"
#include "terminal.h"
+#include "event-loop.h"
+
+/* Asynchronous signal handler registered as event loop source for
+ when we have pending events ready to be passed to the core. */
+static struct async_event_handler *infrun_async_inferior_event_token;
+
+/* Stores whether infrun_async was previously enabled or disabled.
+ Starts off as -1, indicating "never enabled/disabled". */
+static int infrun_is_async = -1;
+
+/* See infrun.h. */
+
+void
+infrun_async (int enable)
+{
+ if (infrun_is_async != enable)
+ {
+ infrun_is_async = enable;
+
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: infrun_async(%d)\n",
+ enable);
+
+ if (enable)
+ mark_async_event_handler (infrun_async_inferior_event_token);
+ else
+ clear_async_event_handler (infrun_async_inferior_event_token);
+ }
+}
/* Prototypes for local functions */
@@ -1305,6 +1335,9 @@ set_step_over_info (struct address_space *aspace, CORE_ADDR address,
static void
clear_step_over_info (void)
{
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: clear_step_over_info\n");
step_over_info.aspace = NULL;
step_over_info.address = 0;
step_over_info.nonsteppable_watchpoint_p = 0;
@@ -1477,6 +1510,23 @@ get_displaced_stepping_state (int pid)
return NULL;
}
+/* Returns true if any inferior has a thread doing a displaced
+ step. */
+
+static int
+displaced_step_in_progress_any_inferior (void)
+{
+ struct displaced_step_inferior_state *state;
+
+ for (state = displaced_step_inferior_states;
+ state != NULL;
+ state = state->next)
+ if (!ptid_equal (state->step_ptid, null_ptid))
+ return 1;
+
+ return 0;
+}
+
/* Return true if process PID has a thread doing a displaced step. */
static int
@@ -1790,21 +1840,28 @@ displaced_step_restore (struct displaced_step_inferior_state *displaced,
displaced->step_copy));
}
-static void
+/* If we displaced stepped an instruction successfully, adjust
+ registers and memory to yield the same effect the instruction would
+ have had if we had executed it at its original address, and return
+ 1. If the instruction didn't complete, relocate the PC and return
+ -1. If the thread wasn't displaced stepping, return 0. */
+
+static int
displaced_step_fixup (ptid_t event_ptid, enum gdb_signal signal)
{
struct cleanup *old_cleanups;
struct displaced_step_inferior_state *displaced
= get_displaced_stepping_state (ptid_get_pid (event_ptid));
+ int ret;
/* Was any thread of this process doing a displaced step? */
if (displaced == NULL)
- return;
+ return 0;
/* Was this event for the pid we displaced? */
if (ptid_equal (displaced->step_ptid, null_ptid)
|| ! ptid_equal (displaced->step_ptid, event_ptid))
- return;
+ return 0;
old_cleanups = make_cleanup (displaced_step_clear_cleanup, displaced);
@@ -1827,6 +1884,7 @@ displaced_step_fixup (ptid_t event_ptid, enum gdb_signal signal)
displaced->step_original,
displaced->step_copy,
get_thread_regcache (displaced->step_ptid));
+ ret = 1;
}
else
{
@@ -1837,11 +1895,14 @@ displaced_step_fixup (ptid_t event_ptid, enum gdb_signal signal)
pc = displaced->step_original + (pc - displaced->step_copy);
regcache_write_pc (regcache, pc);
+ ret = -1;
}
do_cleanups (old_cleanups);
displaced->step_ptid = null_ptid;
+
+ return ret;
}
/* Data to be passed around while handling an event. This data is
@@ -1890,10 +1951,19 @@ start_step_over (void)
{
struct thread_info *tp, *next;
+ /* Don't start a new step-over if we already have a step-over
+ operation ongoing. */
+ if (step_over_info_valid_p ())
+ return 0;
+
for (tp = step_over_queue_head; tp != NULL; tp = next)
{
struct execution_control_state ecss;
struct execution_control_state *ecs = &ecss;
+ enum step_over_what step_what;
+ int must_be_in_line;
+ struct regcache *regcache = get_thread_regcache (tp->ptid);
+ struct gdbarch *gdbarch = get_regcache_arch (regcache);
next = thread_step_over_chain_next (tp);
@@ -1902,6 +1972,17 @@ start_step_over (void)
if (displaced_step_in_progress (ptid_get_pid (tp->ptid)))
continue;
+ step_what = thread_still_needs_step_over (tp);
+ must_be_in_line = ((step_what & STEP_OVER_WATCHPOINT)
+ || ((step_what & STEP_OVER_BREAKPOINT)
+ && !use_displaced_stepping (gdbarch)));
+
+ /* We currently stop all threads of all processes to step-over
+ in-line. If we need to start a new in-line step-over, let
+ any pending displace steps finish first. */
+ if (must_be_in_line && displaced_step_in_progress_any_inferior ())
+ return 0;
+
thread_step_over_chain_remove (tp);
if (step_over_queue_head == NULL)
@@ -1911,13 +1992,16 @@ start_step_over (void)
"infrun: step-over queue now empty\n");
}
- if (tp->control.trap_expected || tp->executing)
+ if (tp->control.trap_expected
+ || tp->resumed
+ || tp->executing)
{
internal_error (__FILE__, __LINE__,
"[%s] has inconsistent state: "
- "trap_expected=%d, executing=%d\n",
+ "trap_expected=%d, resumed=%d, executing=%d\n",
target_pid_to_str (tp->ptid),
tp->control.trap_expected,
+ tp->resumed,
tp->executing);
}
@@ -1932,7 +2016,7 @@ start_step_over (void)
wouldn't be able to resume anything else until the target
stops again. In non-stop, the resume always resumes only TP,
so it's OK to let the thread resume freely. */
- if (!non_stop && !thread_still_needs_step_over (tp))
+ if (!non_stop && !step_what)
continue;
switch_to_thread (tp->ptid);
@@ -1942,6 +2026,15 @@ start_step_over (void)
if (!ecs->wait_some_more)
error (_("Command aborted."));
+ gdb_assert (tp->resumed);
+
+ /* If we started a new in-line step-over, we're done. */
+ if (step_over_info_valid_p ())
+ {
+ gdb_assert (tp->control.trap_expected);
+ return 1;
+ }
+
if (!non_stop)
{
/* On all-stop, shouldn't have resumed unless we needed a
@@ -2157,12 +2250,38 @@ resume (enum gdb_signal sig)
single-step). */
int step;
- tp->stepped_breakpoint = 0;
-
gdb_assert (!thread_is_in_step_over_chain (tp));
QUIT;
+ if (tp->suspend.waitstatus_pending_p)
+ {
+ if (debug_infrun)
+ {
+ char *statstr;
+
+ statstr = target_waitstatus_to_string (&tp->suspend.waitstatus);
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: resume: thread %s has pending wait status %s "
+ "(currently_stepping=%d).\n",
+ target_pid_to_str (tp->ptid), statstr,
+ currently_stepping (tp));
+ xfree (statstr);
+ }
+
+ tp->resumed = 1;
+ /* Avoid confusing the next resume, if the next stop/resume
+ happens to apply to another thread. */
+ tp->suspend.stop_signal = GDB_SIGNAL_0;
+ discard_cleanups (old_cleanups);
+
+ if (target_can_async_p ())
+ target_async (1);
+ return;
+ }
+
+ tp->stepped_breakpoint = 0;
+
/* Depends on stepped_breakpoint. */
step = currently_stepping (tp);
@@ -2267,6 +2386,7 @@ resume (enum gdb_signal sig)
resume_ptid = user_visible_resume_ptid (user_step);
do_target_resume (resume_ptid, 0, GDB_SIGNAL_0);
discard_cleanups (old_cleanups);
+ tp->resumed = 1;
return;
}
}
@@ -2392,11 +2512,12 @@ resume (enum gdb_signal sig)
if (execution_direction != EXEC_REVERSE
&& step && breakpoint_inserted_here_p (aspace, pc))
{
- /* The only case we currently need to step a breakpoint
- instruction is when we have a signal to deliver. See
- handle_signal_stop where we handle random signals that could
- take out us out of the stepping range. Normally, in that
- case we end up continuing (instead of stepping) over the
+ /* There are two cases where we currently need to step a
+ breakpoint instruction when we have a signal to deliver:
+
+ - See handle_signal_stop where we handle random signals that
+ could take out us out of the stepping range. Normally, in
+ that case we end up continuing (instead of stepping) over the
signal handler with a breakpoint at PC, but there are cases
where we should _always_ single-step, even if we have a
step-resume breakpoint, like when a software watchpoint is
@@ -2409,8 +2530,20 @@ resume (enum gdb_signal sig)
recurses and executes PC again, it'll miss the breakpoint.
So we leave the breakpoint inserted anyway, but we need to
record that we tried to step a breakpoint instruction, so
- that adjust_pc_after_break doesn't end up confused. */
- gdb_assert (sig != GDB_SIGNAL_0);
+ that adjust_pc_after_break doesn't end up confused.
+
+ - In non-stop if we insert a breakpoint (e.g., a step-resume)
+ in one thread after another thread that was stepping had been
+ momentarily paused for a step-over. When we re-resume the
+ stepping thread, it may be resumed from that address with a
+ breakpoint that hasn't trapped yet. Seen with
+ gdb.threads/non-stop-fair-events.exp, on targets that don't
+ do displaced stepping. */
+
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: resume: [%s] stepped breakpoint\n",
+ target_pid_to_str (tp->ptid));
tp->stepped_breakpoint = 1;
@@ -2448,6 +2581,7 @@ resume (enum gdb_signal sig)
}
do_target_resume (resume_ptid, step, sig);
+ tp->resumed = 1;
discard_cleanups (old_cleanups);
}
\f
@@ -2464,6 +2598,37 @@ clear_proceed_status_thread (struct thread_info *tp)
"infrun: clear_proceed_status_thread (%s)\n",
target_pid_to_str (tp->ptid));
+ /* If we're starting a new sequence, then the previous finished
+ single-step is no longer relevant. */
+ if (tp->suspend.waitstatus_pending_p)
+ {
+ if (tp->suspend.stop_reason == TARGET_STOPPED_BY_SINGLE_STEP)
+ {
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: clear_proceed_status: pending "
+ "event of %s was a finished step. "
+ "Discarding.\n",
+ target_pid_to_str (tp->ptid));
+
+ tp->suspend.waitstatus_pending_p = 0;
+ tp->suspend.stop_reason = TARGET_STOPPED_BY_NO_REASON;
+ }
+ else if (debug_infrun)
+ {
+ char *statstr;
+
+ statstr = target_waitstatus_to_string (&tp->suspend.waitstatus);
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: clear_proceed_status_thread: thread %s "
+ "has pending wait status %s "
+ "(currently_stepping=%d).\n",
+ target_pid_to_str (tp->ptid), statstr,
+ currently_stepping (tp));
+ xfree (statstr);
+ }
+ }
+
/* If this signal should not be seen by program, give it zero.
Used for debugging signals. */
if (!signal_pass_state (tp->suspend.stop_signal))
@@ -2527,8 +2692,6 @@ clear_proceed_status (int step)
stop_after_trap = 0;
- clear_step_over_info ();
-
observer_notify_about_to_proceed ();
if (stop_registers)
@@ -2777,7 +2940,7 @@ proceed (CORE_ADDR addr, enum gdb_signal siggnal)
/* A new displaced stepping sequence was started. In all-stop,
we can't talk to the target anymore until it next stops. */
}
- else if (!tp->executing && !thread_is_in_step_over_chain (tp))
+ else if (!tp->resumed && !thread_is_in_step_over_chain (tp))
{
/* The thread wasn't started, and isn't queued, run it now. */
reset_ecs (ecs, tp);
@@ -3076,6 +3239,179 @@ print_target_wait_results (ptid_t waiton_ptid, ptid_t result_ptid,
ui_file_delete (tmp_stream);
}
+/* Select a thread at random, out of those which are resumed and have
+ had events. */
+
+static struct thread_info *
+random_pending_event_thread (ptid_t waiton_ptid)
+{
+ struct thread_info *event_tp;
+ int num_events = 0;
+ int random_selector;
+
+ /* First see how many events we have. Count only resumed threads
+ that have an event pending. */
+ ALL_NON_EXITED_THREADS (event_tp)
+ if (ptid_match (event_tp->ptid, waiton_ptid)
+ && event_tp->resumed
+ && event_tp->suspend.waitstatus_pending_p)
+ num_events++;
+
+ if (num_events == 0)
+ return NULL;
+
+ /* Now randomly pick a thread out of those that have had events. */
+ random_selector = (int)
+ ((num_events * (double) rand ()) / (RAND_MAX + 1.0));
+
+ if (debug_infrun && num_events > 1)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: Found %d events, selecting #%d\n",
+ num_events, random_selector);
+
+ /* Select the Nth thread that has had an event. */
+ ALL_NON_EXITED_THREADS (event_tp)
+ if (ptid_match (event_tp->ptid, waiton_ptid)
+ && event_tp->resumed
+ && event_tp->suspend.waitstatus_pending_p)
+ if (random_selector-- == 0)
+ break;
+
+ return event_tp;
+}
+
+/* Wrapper for target_wait that first checks whether threads have
+ pending status to report before actually asking the target for more
+ events. */
+
+static ptid_t
+do_target_wait (ptid_t ptid, struct target_waitstatus *status, int options)
+{
+ ptid_t event_ptid;
+ struct thread_info *tp;
+
+ /* First check if there is a resumed thread with a wait status
+ pending. */
+ if (ptid_equal (ptid, minus_one_ptid) || ptid_is_pid (ptid))
+ {
+ tp = random_pending_event_thread (ptid);
+ }
+ else
+ {
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: Waiting for specific thread %s.\n",
+ target_pid_to_str (ptid));
+
+ /* We have a specific thread to check. */
+ tp = find_thread_ptid (ptid);
+ gdb_assert (tp != NULL);
+ if (!tp->suspend.waitstatus_pending_p)
+ tp = NULL;
+ }
+
+ if (tp != NULL
+ && (tp->suspend.stop_reason == TARGET_STOPPED_BY_SW_BREAKPOINT
+ || tp->suspend.stop_reason == TARGET_STOPPED_BY_HW_BREAKPOINT))
+ {
+ struct regcache *regcache = get_thread_regcache (tp->ptid);
+ struct gdbarch *gdbarch = get_regcache_arch (regcache);
+ CORE_ADDR pc;
+ int discard = 0;
+
+ pc = regcache_read_pc (regcache);
+
+ if (pc != tp->suspend.stop_pc)
+ {
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: PC of %s changed. was=%s, now=%s\n",
+ target_pid_to_str (tp->ptid),
+ paddress (target_gdbarch (), tp->prev_pc),
+ paddress (target_gdbarch (), pc));
+ discard = 1;
+ }
+ else if (!breakpoint_inserted_here_p (get_regcache_aspace (regcache), pc))
+ {
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: previous breakpoint of %s, at %s gone\n",
+ target_pid_to_str (tp->ptid),
+ paddress (target_gdbarch (), pc));
+
+ discard = 1;
+ }
+
+ if (discard)
+ {
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: pending event of %s cancelled.\n",
+ target_pid_to_str (tp->ptid));
+
+ tp->suspend.waitstatus.kind = TARGET_WAITKIND_SPURIOUS;
+ tp->suspend.stop_reason = TARGET_STOPPED_BY_NO_REASON;
+ }
+ }
+
+ if (tp != NULL)
+ {
+ if (debug_infrun)
+ {
+ char *statstr;
+
+ statstr = target_waitstatus_to_string (&tp->suspend.waitstatus);
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: Using pending wait status %s for %s.\n",
+ statstr,
+ target_pid_to_str (tp->ptid));
+ xfree (statstr);
+ }
+
+ /* Now that we've selected our final event LWP, un-adjust its PC
+ if it was a software breakpoint (and the target doesn't
+ always adjust the PC itself). */
+ if (tp->suspend.stop_reason == TARGET_STOPPED_BY_SW_BREAKPOINT
+ && !target_supports_stopped_by_sw_breakpoint ())
+ {
+ struct regcache *regcache;
+ struct gdbarch *gdbarch;
+ int decr_pc;
+
+ regcache = get_thread_regcache (tp->ptid);
+ gdbarch = get_regcache_arch (regcache);
+
+ decr_pc = gdbarch_decr_pc_after_break (gdbarch);
+ if (decr_pc != 0)
+ {
+ CORE_ADDR pc;
+
+ pc = regcache_read_pc (regcache);
+ regcache_write_pc (regcache, pc + decr_pc);
+ }
+ }
+
+ tp->suspend.stop_reason = TARGET_STOPPED_BY_NO_REASON;
+ *status = tp->suspend.waitstatus;
+ tp->suspend.waitstatus_pending_p = 0;
+
+ /* Wake up the event loop again, until all pending events are
+ processed. */
+ if (target_is_async_p ())
+ mark_async_event_handler (infrun_async_inferior_event_token);
+ return tp->ptid;
+ }
+
+ /* But if we don't find one, we'll have to wait. */
+
+ if (deprecated_target_wait_hook)
+ event_ptid = deprecated_target_wait_hook (ptid, status, options);
+ else
+ event_ptid = target_wait (ptid, status, options);
+
+ return event_ptid;
+}
+
/* Prepare and stabilize the inferior for detaching it. E.g.,
detaching while a thread is displaced stepping is a recipe for
crashing it, as nothing would readjust the PC out of the scratch
@@ -3119,10 +3455,7 @@ prepare_for_detach (void)
don't get any event. */
target_dcache_invalidate ();
- if (deprecated_target_wait_hook)
- ecs->ptid = deprecated_target_wait_hook (pid_ptid, &ecs->ws, 0);
- else
- ecs->ptid = target_wait (pid_ptid, &ecs->ws, 0);
+ ecs->ptid = do_target_wait (pid_ptid, &ecs->ws, 0);
if (debug_infrun)
print_target_wait_results (pid_ptid, ecs->ptid, &ecs->ws);
@@ -3194,10 +3527,7 @@ wait_for_inferior (void)
don't get any event. */
target_dcache_invalidate ();
- if (deprecated_target_wait_hook)
- ecs->ptid = deprecated_target_wait_hook (waiton_ptid, &ecs->ws, 0);
- else
- ecs->ptid = target_wait (waiton_ptid, &ecs->ws, 0);
+ ecs->ptid = do_target_wait (waiton_ptid, &ecs->ws, 0);
if (debug_infrun)
print_target_wait_results (waiton_ptid, ecs->ptid, &ecs->ws);
@@ -3294,11 +3624,7 @@ fetch_inferior_event (void *client_data)
make_cleanup_restore_integer (&execution_direction);
execution_direction = target_execution_direction ();
- if (deprecated_target_wait_hook)
- ecs->ptid =
- deprecated_target_wait_hook (waiton_ptid, &ecs->ws, TARGET_WNOHANG);
- else
- ecs->ptid = target_wait (waiton_ptid, &ecs->ws, TARGET_WNOHANG);
+ ecs->ptid = do_target_wait (waiton_ptid, &ecs->ws, TARGET_WNOHANG);
if (debug_infrun)
print_target_wait_results (waiton_ptid, ecs->ptid, &ecs->ws);
@@ -3671,6 +3997,349 @@ get_inferior_stop_soon (ptid_t ptid)
return inf->control.stop_soon;
}
+/* Wait for one event. Store the resulting waitstatus in WS, and
+ return the event ptid. */
+
+static ptid_t
+wait_one (struct target_waitstatus *ws)
+{
+ ptid_t event_ptid;
+ ptid_t wait_ptid = minus_one_ptid;
+
+ overlay_cache_invalid = 1;
+
+ /* Flush target cache before starting to handle each event.
+ Target was running and cache could be stale. This is just a
+ heuristic. Running threads may modify target memory, but we
+ don't get any event. */
+ target_dcache_invalidate ();
+
+ if (deprecated_target_wait_hook)
+ event_ptid = deprecated_target_wait_hook (wait_ptid, ws, 0);
+ else
+ event_ptid = target_wait (wait_ptid, ws, 0);
+
+ if (debug_infrun)
+ print_target_wait_results (wait_ptid, event_ptid, ws);
+
+ if (ws->kind == TARGET_WAITKIND_SYSCALL_ENTRY
+ || ws->kind == TARGET_WAITKIND_SYSCALL_RETURN)
+ ws->value.syscall_number = UNKNOWN_SYSCALL;
+
+ return event_ptid;
+}
+
+/* Generate a wrapper for target_stopped_by_REASON that works on PTID
+ instead of the current thread. */
+#define THREAD_STOPPED_BY(REASON) \
+static int \
+thread_stopped_by_ ## REASON (ptid_t ptid) \
+{ \
+ struct cleanup *old_chain; \
+ int res; \
+ \
+ old_chain = save_inferior_ptid (); \
+ inferior_ptid = ptid; \
+ \
+ res = target_stopped_by_ ## REASON (); \
+ \
+ do_cleanups (old_chain); \
+ \
+ return res; \
+}
+
+/* Generate thread_stopped_by_watchpoint. */
+THREAD_STOPPED_BY (watchpoint)
+/* Generate thread_stopped_by_sw_breakpoint. */
+THREAD_STOPPED_BY (sw_breakpoint)
+/* Generate thread_stopped_by_hw_breakpoint. */
+THREAD_STOPPED_BY (hw_breakpoint)
+
+/* Cleanups that switches to the PTID pointed at by PTID_P. */
+
+static void
+switch_to_thread_cleanup (void *ptid_p)
+{
+ ptid_t ptid = *(ptid_t *) ptid_p;
+
+ switch_to_thread (ptid);
+}
+
+/* Save the thread's event and stop reason to process it later. */
+
+static void
+save_waitstatus (struct thread_info *tp, struct target_waitstatus *ws)
+{
+ struct regcache *regcache;
+ struct address_space *aspace;
+
+ if (debug_infrun)
+ {
+ char *statstr;
+
+ statstr = target_waitstatus_to_string (ws);
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: saving status %s for %d.%ld.%ld\n",
+ statstr,
+ ptid_get_pid (tp->ptid),
+ ptid_get_lwp (tp->ptid),
+ ptid_get_tid (tp->ptid));
+ xfree (statstr);
+ }
+
+ /* Record for later. */
+ tp->suspend.waitstatus = *ws;
+ tp->suspend.waitstatus_pending_p = 1;
+
+ regcache = get_thread_regcache (tp->ptid);
+ aspace = get_regcache_aspace (regcache);
+
+ if (ws->kind == TARGET_WAITKIND_STOPPED
+ && ws->value.sig == GDB_SIGNAL_TRAP)
+ {
+ CORE_ADDR pc = regcache_read_pc (regcache);
+
+ adjust_pc_after_break (tp, &tp->suspend.waitstatus);
+
+ if (thread_stopped_by_watchpoint (tp->ptid))
+ {
+ tp->suspend.stop_reason
+ = TARGET_STOPPED_BY_WATCHPOINT;
+ }
+ else if (target_supports_stopped_by_sw_breakpoint ()
+ && thread_stopped_by_sw_breakpoint (tp->ptid))
+ {
+ tp->suspend.stop_reason
+ = TARGET_STOPPED_BY_SW_BREAKPOINT;
+ }
+ else if (target_supports_stopped_by_hw_breakpoint ()
+ && thread_stopped_by_hw_breakpoint (tp->ptid))
+ {
+ tp->suspend.stop_reason
+ = TARGET_STOPPED_BY_HW_BREAKPOINT;
+ }
+ else if (!target_supports_stopped_by_hw_breakpoint ()
+ && hardware_breakpoint_inserted_here_p (aspace,
+ pc))
+ {
+ tp->suspend.stop_reason
+ = TARGET_STOPPED_BY_HW_BREAKPOINT;
+ }
+ else if (!target_supports_stopped_by_sw_breakpoint ()
+ && software_breakpoint_inserted_here_p (aspace,
+ pc))
+ {
+ tp->suspend.stop_reason
+ = TARGET_STOPPED_BY_SW_BREAKPOINT;
+ }
+ else if (!thread_has_single_step_breakpoints_set (tp)
+ && currently_stepping (tp))
+ {
+ tp->suspend.stop_reason
+ = TARGET_STOPPED_BY_SINGLE_STEP;
+ }
+ }
+}
+
+/* Stop all threads. */
+
+static void
+stop_all_threads (void)
+{
+ /* We may need multiple passes to discover all threads. */
+ int pass;
+ int iterations = 0;
+ ptid_t entry_ptid;
+ struct cleanup *old_chain;
+
+ gdb_assert (non_stop);
+
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog, "infrun: stop_all_threads\n");
+
+ entry_ptid = inferior_ptid;
+ old_chain = make_cleanup (switch_to_thread_cleanup, &entry_ptid);
+
+ /* Stop threads in two passes since threads could be spawning as we
+ go through the first pass. In the second pass, we will stop such
+ spawned threads. */
+ for (pass = 0; pass < 2; pass++, iterations++)
+ {
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: stop_all_threads, pass=%d, "
+ "iterations=%d\n", pass, iterations);
+ while (1)
+ {
+ ptid_t event_ptid;
+ struct target_waitstatus ws;
+ int need_wait = 0;
+ struct thread_info *t;
+
+ update_thread_list ();
+
+ /* Go through all threads looking for threads that we need
+ to tell the target to stop. */
+ ALL_NON_EXITED_THREADS (t)
+ {
+ if (t->executing)
+ {
+ /* If already stopping, don't request a stop again.
+ We just haven't seen the notification yet. */
+ if (!t->stop_requested)
+ {
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: %s executing, "
+ "need stop\n",
+ target_pid_to_str (t->ptid));
+ target_stop (t->ptid);
+ t->stop_requested = 1;
+ }
+ else
+ {
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: %s executing, "
+ "already stopping\n",
+ target_pid_to_str (t->ptid));
+ }
+
+ if (t->stop_requested)
+ need_wait = 1;
+ }
+ else
+ {
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: %s not executing\n",
+ target_pid_to_str (t->ptid));
+
+ /* The thread may be not executing, but still be
+ resumed with a pending status to process. */
+ t->resumed = 0;
+ }
+ }
+
+ if (!need_wait)
+ break;
+
+ /* If we find new threads on the second iteration, restart
+ over. We want to see two iterations in a row with all
+ threads stopped. */
+ if (pass > 0)
+ pass = -1;
+
+ event_ptid = wait_one (&ws);
+ if (ws.kind == TARGET_WAITKIND_NO_RESUMED)
+ /* All resumed threads exited. */
+ ;
+ else if (ws.kind == TARGET_WAITKIND_EXITED
+ || ws.kind == TARGET_WAITKIND_SIGNALLED)
+ {
+ if (debug_infrun)
+ {
+ ptid_t ptid = pid_to_ptid (ws.value.integer);
+
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: %s exited while "
+ "stopping threads\n",
+ target_pid_to_str (ptid));
+ }
+ }
+ else
+ {
+ if (!in_thread_list (event_ptid))
+ t = add_thread (event_ptid);
+ else
+ t = find_thread_ptid (event_ptid);
+
+ t->stop_requested = 0;
+ t->executing = 0;
+ t->resumed = 0;
+ t->control.may_range_step = 0;
+
+ if (ws.kind == TARGET_WAITKIND_STOPPED
+ && ws.value.sig == GDB_SIGNAL_0)
+ {
+ /* We caught the event that we intended to catch, so
+ there's no event pending. */
+ t->suspend.waitstatus.kind = TARGET_WAITKIND_IGNORE;
+ t->suspend.waitstatus_pending_p = 0;
+
+ if (displaced_step_fixup (t->ptid, GDB_SIGNAL_0) < 0)
+ {
+ /* Add it back to the step-over queue. */
+ if (debug_infrun)
+ {
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: displaced-step of %s "
+ "canceled: adding back to the "
+ "step-over queue\n",
+ target_pid_to_str (t->ptid));
+ }
+ t->control.trap_expected = 0;
+ thread_step_over_chain_enqueue (t);
+ }
+ }
+ else
+ {
+ enum gdb_signal sig;
+ struct regcache *regcache;
+ struct address_space *aspace;
+
+ if (debug_infrun)
+ {
+ char *statstr;
+
+ statstr = target_waitstatus_to_string (&ws);
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: target_wait %s, saving "
+ "status for %d.%ld.%ld\n",
+ statstr,
+ ptid_get_pid (t->ptid),
+ ptid_get_lwp (t->ptid),
+ ptid_get_tid (t->ptid));
+ xfree (statstr);
+ }
+
+ /* Record for later. */
+ save_waitstatus (t, &ws);
+
+ sig = (ws.kind == TARGET_WAITKIND_STOPPED
+ ? ws.value.sig : GDB_SIGNAL_0);
+
+ if (displaced_step_fixup (t->ptid, sig) < 0)
+ {
+ /* Add it back to the step-over queue. */
+ t->control.trap_expected = 0;
+ thread_step_over_chain_enqueue (t);
+ }
+
+ regcache = get_thread_regcache (t->ptid);
+ t->suspend.stop_pc = regcache_read_pc (regcache);
+
+ if (debug_infrun)
+ {
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: saved stop_pc=%s for %s "
+ "(currently_stepping=%d)\n",
+ paddress (target_gdbarch (),
+ t->suspend.stop_pc),
+ target_pid_to_str (t->ptid),
+ currently_stepping (t));
+ }
+ }
+ }
+ }
+ }
+
+ do_cleanups (old_chain);
+
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog, "infrun: stop_all_threads done\n");
+}
+
/* Given an execution control state that has been freshly filled in by
an event from the inferior, figure out what it means and take
appropriate action.
@@ -3793,21 +4462,28 @@ handle_inferior_event (struct execution_control_state *ecs)
we're handling a process exit in non-stop mode, there's nothing
to do, as threads of the dead process are gone, and threads of
any other process were left running. */
- if (!non_stop)
- set_executing (minus_one_ptid, 0);
- else if (ecs->ws.kind == TARGET_WAITKIND_SIGNALLED
- && ecs->ws.kind == TARGET_WAITKIND_EXITED)
- {
- ptid_t pid_ptid;
- /* Some targets still have execution when a process exits.
- E.g., for "checkpoint", when when a fork exits and is
- mourned, linux-fork.c switches to another fork. */
- pid_ptid = pid_to_ptid (ptid_get_pid (ecs->ptid));
- set_executing (pid_ptid, 0);
- }
- else
- set_executing (ecs->ptid, 0);
+ {
+ ptid_t mark_ptid;
+
+ if (!non_stop)
+ mark_ptid = minus_one_ptid;
+ else if (ecs->ws.kind == TARGET_WAITKIND_SIGNALLED
+ && ecs->ws.kind == TARGET_WAITKIND_EXITED)
+ {
+ /* Some targets still have execution when a process exits.
+ E.g., for "checkpoint", when when a fork exits and is
+ mourned, linux-fork.c switches to another fork. */
+ mark_ptid = pid_to_ptid (ptid_get_pid (ecs->ptid));
+ }
+ else
+ mark_ptid = ecs->ptid;
+
+ set_executing (mark_ptid, 0);
+
+ /* Likewise the resumed flag. */
+ set_resumed (mark_ptid, 0);
+ }
switch (ecs->ws.kind)
{
@@ -4220,16 +4896,137 @@ Cannot fill $_exitsignal with the correct signal number.\n"));
}
}
-/* Called when we get an event that may finish an in-line or
- out-of-line (displaced stepping) step-over started previously. */
+/* Restart threads back to what they were trying to do back when we
+ paused them for an in-line step-over. The EVENT_THREAD thread is
+ ignored. */
static void
+restart_threads (struct thread_info *event_thread)
+{
+ struct thread_info *tp;
+ struct thread_info *step_over = NULL;
+
+ /* In case the instruction just stepped spawned a new thread. */
+ update_thread_list ();
+
+ ALL_NON_EXITED_THREADS (tp)
+ {
+ struct execution_control_state ecss;
+ struct execution_control_state *ecs = &ecss;
+
+ if (tp == event_thread)
+ {
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: restart threads: "
+ "[%s] is event thread\n",
+ target_pid_to_str (tp->ptid));
+ continue;
+ }
+
+ if (!(tp->state == THREAD_RUNNING || tp->control.in_infcall))
+ {
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: restart threads: "
+ "[%s] not meant to be running\n",
+ target_pid_to_str (tp->ptid));
+ continue;
+ }
+
+ if (tp->resumed)
+ {
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: restart threads: [%s] resumed\n",
+ target_pid_to_str (tp->ptid));
+ gdb_assert (tp->executing || tp->suspend.waitstatus_pending_p);
+ continue;
+ }
+
+ if (thread_is_in_step_over_chain (tp))
+ {
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: restart threads: "
+ "[%s] needs step-over\n",
+ target_pid_to_str (tp->ptid));
+ gdb_assert (!tp->resumed);
+ continue;
+ }
+
+
+ if (tp->suspend.waitstatus_pending_p)
+ {
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: restart threads: "
+ "[%s] has pending status\n",
+ target_pid_to_str (tp->ptid));
+ tp->resumed = 1;
+ continue;
+ }
+
+ /* If some thread needs to start a step-over at this point, it
+ should still be in the step-over queue, and thus skipped
+ above. */
+ if (thread_still_needs_step_over (tp))
+ {
+ internal_error (__FILE__, __LINE__,
+ "thread [%s] needs a step-over, but not in "
+ "step-over queue\n",
+ target_pid_to_str (tp->ptid));
+ }
+
+ if (currently_stepping (tp))
+ {
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: restart threads: [%s] was stepping\n",
+ target_pid_to_str (tp->ptid));
+ keep_going_stepped_thread (tp);
+ }
+ else
+ {
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: restart threads: [%s] continuing\n",
+ target_pid_to_str (tp->ptid));
+ reset_ecs (ecs, tp);
+ switch_to_thread (tp->ptid);
+ keep_going_pass (ecs);
+ }
+ }
+}
+
+/* Callback for iterate_over_threads. Find a resumed thread that has
+ a pending waitstatus. */
+
+static int
+resumed_thread_with_pending_status (struct thread_info *tp,
+ void *arg)
+{
+ return (tp->resumed
+ && tp->suspend.waitstatus_pending_p);
+}
+
+/* Called when we get an event that may finish an in-line or
+ out-of-line (displaced stepping) step-over started previously.
+ Return true if the event is processed and we should go back to the
+ event loop; false if the caller should continue processing the
+ event. */
+
+static int
finish_step_over (struct execution_control_state *ecs)
{
+ int had_step_over_info;
+
displaced_step_fixup (ecs->ptid,
ecs->event_thread->suspend.stop_signal);
- if (step_over_info_valid_p ())
+ had_step_over_info = step_over_info_valid_p ();
+
+ if (had_step_over_info)
{
/* If we're stepping over a breakpoint with all threads locked,
then only the thread that was stepped should be reporting
@@ -4241,11 +5038,99 @@ finish_step_over (struct execution_control_state *ecs)
}
if (!non_stop)
- return;
+ return 0;
/* Start a new step-over in another thread if there's one that
needs it. */
start_step_over ();
+
+ /* If we were stepping over a breakpoint before, and haven't started
+ a new in-line step-over sequence, then restart all other threads
+ (except the event thread). We can't do this in all-stop, as then
+ e.g., we wouldn't be able to issue any other remote packet until
+ these other threads stop. */
+ if (had_step_over_info && !step_over_info_valid_p ())
+ {
+ struct thread_info *pending;
+
+ /* If we only have threads with pending statuses, the restart
+ below won't restart any thread and so nothing re-inserts the
+ breakpoint we just stepped over. But we need it inserted
+ when we later process the pending events, otherwise if
+ another thread has a pending event for this breakpoint too,
+ we'd discard its event (because the breakpoint that
+ originally caused the event was no longer inserted). */
+ context_switch (ecs->ptid);
+ insert_breakpoints ();
+
+ restart_threads (ecs->event_thread);
+
+ /* If we have events pending, go through handle_inferior_event
+ again, picking up a pending event at random. This avoids
+ thread starvation. */
+
+ /* But not if we just stepped over a watchpoint in order to let
+ the instruction execute so we can evaluate its expression.
+ The set of watchpoints that triggered is recorded in the
+ breakpoint objects themselves (see bp->watchpoint_triggered).
+ If we processed another event first, that other event could
+ clobber this info. */
+ if (ecs->event_thread->stepping_over_watchpoint)
+ return 0;
+
+ pending = iterate_over_threads (resumed_thread_with_pending_status,
+ NULL);
+ if (pending != NULL)
+ {
+ struct thread_info *tp = ecs->event_thread;
+ struct regcache *regcache;
+
+ if (debug_infrun)
+ {
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: found resumed threads with "
+ "pending events, saving status\n");
+ }
+
+ gdb_assert (pending != tp);
+
+ /* Record the event thread's event for later. */
+ save_waitstatus (tp, &ecs->ws);
+ /* This was cleared early, by handle_inferior_event. Set it
+ so this pending event is considered by
+ do_target_wait. */
+ tp->resumed = 1;
+
+ gdb_assert (!tp->executing);
+
+ regcache = get_thread_regcache (tp->ptid);
+ tp->suspend.stop_pc = regcache_read_pc (regcache);
+
+ if (debug_infrun)
+ {
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: saved stop_pc=%s for %s "
+ "(currently_stepping=%d)\n",
+ paddress (target_gdbarch (),
+ tp->suspend.stop_pc),
+ target_pid_to_str (tp->ptid),
+ currently_stepping (tp));
+ }
+
+ /* This in-line step-over finished; clear this so we won't
+ start a new one. This is what handle_signal_stop would
+ do, if we returned false. */
+ tp->stepping_over_breakpoint = 0;
+
+ /* Wake up the event loop again. */
+ mark_async_event_handler (infrun_async_inferior_event_token);
+
+ prepare_to_wait (ecs);
+ return 1;
+ }
+ }
+
+ return 0;
}
/* Come here when the program has stopped with a signal. */
@@ -4264,7 +5149,8 @@ handle_signal_stop (struct execution_control_state *ecs)
/* Do we need to clean up the state of a thread that has
completed a displaced single-step? (Doing so usually affects
the PC, so do it here, before we set stop_pc.) */
- finish_step_over (ecs);
+ if (finish_step_over (ecs))
+ return;
/* If we either finished a single-step or hit a breakpoint, but
the user wanted this thread to be stopped, pretend we got a
@@ -4722,6 +5608,7 @@ handle_signal_stop (struct execution_control_state *ecs)
"infrun: signal arrived while stepping over "
"breakpoint\n");
+ clear_step_over_info ();
insert_hp_step_resume_breakpoint_at_frame (frame);
ecs->event_thread->step_after_step_resume_breakpoint = 1;
/* Reset trap_expected to ensure breakpoints are re-inserted. */
@@ -4755,6 +5642,7 @@ handle_signal_stop (struct execution_control_state *ecs)
"infrun: signal may take us out of "
"single-step range\n");
+ clear_step_over_info ();
insert_hp_step_resume_breakpoint_at_frame (frame);
ecs->event_thread->step_after_step_resume_breakpoint = 1;
/* Reset trap_expected to ensure breakpoints are re-inserted. */
@@ -5710,7 +6598,14 @@ switch_back_to_stepped_thread (struct execution_control_state *ecs)
except the one that needs to move past the breakpoint.
If a non-event thread has this set, the "incomplete
step-over" check above should have caught it earlier. */
- gdb_assert (!tp->control.trap_expected);
+ if (tp->control.trap_expected)
+ {
+ internal_error (__FILE__, __LINE__,
+ "[%s] has inconsistent state: "
+ "trap_expected=%d\n",
+ target_pid_to_str (tp->ptid),
+ tp->control.trap_expected);
+ }
/* Did we find the stepping thread? */
if (tp->control.step_range_end)
@@ -5839,6 +6734,7 @@ keep_going_stepped_thread (struct thread_info *tp)
get_frame_address_space (frame),
stop_pc);
+ tp->resumed = 1;
resume_ptid = user_visible_resume_ptid (tp->control.stepping_command);
do_target_resume (resume_ptid, 0, GDB_SIGNAL_0);
}
@@ -6270,6 +7166,7 @@ keep_going_pass (struct execution_control_state *ecs)
struct cleanup *old_cleanups = make_cleanup (resume_cleanups, 0);
gdb_assert (ptid_equal (ecs->event_thread->ptid, inferior_ptid));
+ gdb_assert (!ecs->event_thread->resumed);
/* Save the pc before execution, to compare with pc after stop. */
ecs->event_thread->prev_pc
@@ -6292,6 +7189,34 @@ keep_going_pass (struct execution_control_state *ecs)
discard_cleanups (old_cleanups);
resume (ecs->event_thread->suspend.stop_signal);
}
+ else if (step_over_info_valid_p ())
+ {
+ /* Another thread is stepping over a breakpoint in-line. If
+ this thread needs a step-over too, queue the request. In
+ either case, this resume must be deferred for later. */
+ struct thread_info *tp = ecs->event_thread;
+
+ if (ecs->hit_singlestep_breakpoint
+ || thread_still_needs_step_over (tp))
+ {
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: step-over already in progress: "
+ "step-over for %s deferred\n",
+ target_pid_to_str (tp->ptid));
+ thread_step_over_chain_enqueue (tp);
+ }
+ else
+ {
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: step-over in progress: "
+ "resume of %s deferred\n",
+ target_pid_to_str (tp->ptid));
+ }
+
+ discard_cleanups (old_cleanups);
+ }
else
{
struct regcache *regcache = get_current_regcache ();
@@ -6336,8 +7261,14 @@ keep_going_pass (struct execution_control_state *ecs)
}
else if (remove_wps)
set_step_over_info (NULL, 0, remove_wps);
- else
- clear_step_over_info ();
+
+ /* If we now need to do an in-line step-over, we need to stop
+ all other threads. Note this must be done before
+ insert_breakpoints below, because that removes the breakpoint
+ we're about to step over, otherwise other threads could miss
+ it. */
+ if (step_over_info_valid_p () && non_stop)
+ stop_all_threads ();
/* Stop stepping if inserting breakpoints fails. */
TRY
@@ -7705,6 +8636,23 @@ static const struct internalvar_funcs siginfo_funcs =
NULL
};
+/* Callback for infrun's target events source. This is marked when a
+ thread has a pending status to process. */
+
+static void
+infrun_async_inferior_event_handler (gdb_client_data data)
+{
+ /* If the target is closed while this event source is marked, we
+ will reach here without execution, or a target to call
+ target_wait on, which is an error. Instead of tracking whether
+ the target has been popped already, or whether we do have threads
+ with pending statutes, simply ignore the event. */
+ if (!target_is_async_p ())
+ return;
+
+ inferior_event_handler (INF_REG_EVENT, NULL);
+}
+
void
_initialize_infrun (void)
{
@@ -7712,6 +8660,10 @@ _initialize_infrun (void)
int numsigs;
struct cmd_list_element *c;
+ /* Register extra event sources in the event loop. */
+ infrun_async_inferior_event_token
+ = create_async_event_handler (infrun_async_inferior_event_handler, NULL);
+
add_info ("signals", signals_info, _("\
What debugger does when program gets various signals.\n\
Specify a signal as argument to print info on that signal only."));
diff --git a/gdb/infrun.h b/gdb/infrun.h
index bfce810..2dc55a9 100644
--- a/gdb/infrun.h
+++ b/gdb/infrun.h
@@ -196,6 +196,9 @@ extern void signal_catch_update (const unsigned int *);
systems. Use of symbolic signal names is strongly encouraged. */
enum gdb_signal gdb_signal_from_command (int num);
+/* Enables/disables infrun's async event source in the event loop. */
+extern void infrun_async (int enable);
+
/* The global queue of threads that need to do a step-over operation
to get past e.g., a breakpoint. */
extern struct thread_info *step_over_queue_head;
diff --git a/gdb/target.c b/gdb/target.c
index 306c21d..e992a35 100644
--- a/gdb/target.c
+++ b/gdb/target.c
@@ -3712,6 +3712,15 @@ maintenance_print_target_stack (char *cmd, int from_tty)
}
}
+/* See target.h. */
+
+void
+target_async (int enable)
+{
+ infrun_async (enable);
+ current_target.to_async (¤t_target, enable);
+}
+
/* Controls if targets can report that they can/are async. This is
just for maintainers to use when debugging gdb. */
int target_async_permitted = 1;
diff --git a/gdb/target.h b/gdb/target.h
index 66bf91e..3f1a8b0 100644
--- a/gdb/target.h
+++ b/gdb/target.h
@@ -1704,8 +1704,7 @@ extern int target_async_permitted;
#define target_is_async_p() (current_target.to_is_async_p (¤t_target))
/* Enables/disabled async target events. */
-#define target_async(ENABLE) \
- (current_target.to_async (¤t_target, (ENABLE)))
+extern void target_async (int enable);
#define target_execution_direction() \
(current_target.to_execution_direction (¤t_target))
diff --git a/gdb/target/waitstatus.h b/gdb/target/waitstatus.h
index d4ef3b8..ffaddc1 100644
--- a/gdb/target/waitstatus.h
+++ b/gdb/target/waitstatus.h
@@ -131,7 +131,10 @@ enum target_stop_reason
TARGET_STOPPED_BY_HW_BREAKPOINT,
/* Stopped by a watchpoint. */
- TARGET_STOPPED_BY_WATCHPOINT
+ TARGET_STOPPED_BY_WATCHPOINT,
+
+ /* Stopped by a single step finishing. */
+ TARGET_STOPPED_BY_SINGLE_STEP
};
/* Prototypes */
diff --git a/gdb/thread.c b/gdb/thread.c
index 3e3f419..4dde722 100644
--- a/gdb/thread.c
+++ b/gdb/thread.c
@@ -232,6 +232,7 @@ new_thread (ptid_t ptid)
/* Nothing to follow yet. */
tp->pending_follow.kind = TARGET_WAITKIND_SPURIOUS;
tp->state = THREAD_STOPPED;
+ tp->suspend.waitstatus.kind = TARGET_WAITKIND_IGNORE;
return tp;
}
@@ -852,6 +853,28 @@ thread_change_ptid (ptid_t old_ptid, ptid_t new_ptid)
observer_notify_thread_ptid_changed (old_ptid, new_ptid);
}
+/* See gdbthread.h. */
+
+void
+set_resumed (ptid_t ptid, int resumed)
+{
+ struct thread_info *tp;
+ int all = ptid_equal (ptid, minus_one_ptid);
+
+ if (all || ptid_is_pid (ptid))
+ {
+ for (tp = thread_list; tp; tp = tp->next)
+ if (all || ptid_get_pid (tp->ptid) == ptid_get_pid (ptid))
+ tp->resumed = resumed;
+ }
+ else
+ {
+ tp = find_thread_ptid (ptid);
+ gdb_assert (tp != NULL);
+ tp->resumed = resumed;
+ }
+}
+
/* Helper for set_running, that marks one thread either running or
stopped. */
--
1.9.3
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 09/17] Teach non-stop to do in-line step-overs (stop all, step, restart)
2015-04-17 10:52 ` [PATCH v3 09/17] Teach non-stop to do in-line step-overs (stop all, step, restart) Pedro Alves
@ 2015-04-17 11:01 ` Pedro Alves
2015-04-21 15:01 ` Yao Qi
2015-04-27 20:17 ` Doug Evans
2 siblings, 0 replies; 62+ messages in thread
From: Pedro Alves @ 2015-04-17 11:01 UTC (permalink / raw)
To: gdb-patches
On 04/17/2015 11:45 AM, Pedro Alves wrote:
> make non-stop work for all (albeit, not as efficiently). And then
> there are scenarios where even if the architecture supports displaced
> stepping, we can't use it, because we e.g., don't find a usable
> address to use as displaced step scratch pad. It should also fix
> stepping past watchpoints on targets that have non-continuable
> watchpoints in non-stop mode (e.g., PPC, untested).
Reading back, I notice that wrote that "should fix" and "untested"
back before v1, and forgot to update it since. This definitely
fixes that scenario.
Since:
https://sourceware.org/ml/gdb-patches/2015-04/msg00384.html
PPC64: Fix step-over-trips-on-watchpoint.exp with displaced stepping on
GDB now does not start a displaced step if it need to step
over a watchpoint, instead if steps the instruction at the
original address. What was missing was pausing all threads
in order to be able to remove the watchpoint without other
threads missing it -- that's what's fixed by this patch.
> Running the
> instruction out of line in the displaced stepping scratch pad doesn't
> help that case, as the copied instruction reads/writes the same
> watched memory... We can fix that too by teaching GDB to only remove
> the watchpoint from the thread that we want to move past the
> watchpoint (currently, removing a watchpoint always removes it from
> all threads), but again, that can be considered an optimization; not
> all targets would support it.
I'll update the commit log before pushing.
Thanks,
Pedro Alves
^ permalink raw reply [flat|nested] 62+ messages in thread
* Re: [PATCH v3 09/17] Teach non-stop to do in-line step-overs (stop all, step, restart)
2015-04-17 10:52 ` [PATCH v3 09/17] Teach non-stop to do in-line step-overs (stop all, step, restart) Pedro Alves
2015-04-17 11:01 ` Pedro Alves
@ 2015-04-21 15:01 ` Yao Qi
2015-04-22 20:03 ` Pedro Alves
2015-04-27 20:17 ` Doug Evans
2 siblings, 1 reply; 62+ messages in thread
From: Yao Qi @ 2015-04-21 15:01 UTC (permalink / raw)
To: Pedro Alves; +Cc: gdb-patches
Pedro Alves <palves@redhat.com> writes:
> - that adjust_pc_after_break doesn't end up confused. */
> - gdb_assert (sig != GDB_SIGNAL_0);
> + that adjust_pc_after_break doesn't end up confused.
> +
> + - In non-stop if we insert a breakpoint (e.g., a step-resume)
> + in one thread after another thread that was stepping had been
> + momentarily paused for a step-over. When we re-resume the
> + stepping thread, it may be resumed from that address with a
> + breakpoint that hasn't trapped yet. Seen with
> + gdb.threads/non-stop-fair-events.exp, on targets that don't
> + do displaced stepping. */
Does this comment mean the sequence like this?
thread B is paused,
a breakpoint is inserted at address PC for thread C,
thread B may be resumed from address PC
>
> +/* Select a thread at random, out of those which are resumed and have
> + had events. */
> +
> +static struct thread_info *
> +random_pending_event_thread (ptid_t waiton_ptid)
> +{
If GDB core knows select events at random, then we don't have to do
something similar in linux-nat.c:select_event_lwp, right?
--
Yao (齐尧)
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 09/17] Teach non-stop to do in-line step-overs (stop all, step, restart)
2015-04-21 15:01 ` Yao Qi
@ 2015-04-22 20:03 ` Pedro Alves
2015-04-24 9:06 ` Yao Qi
0 siblings, 1 reply; 62+ messages in thread
From: Pedro Alves @ 2015-04-22 20:03 UTC (permalink / raw)
To: Yao Qi; +Cc: gdb-patches
On 04/21/2015 04:01 PM, Yao Qi wrote:
> Pedro Alves <palves@redhat.com> writes:
>
>> - that adjust_pc_after_break doesn't end up confused. */
>> - gdb_assert (sig != GDB_SIGNAL_0);
>> + that adjust_pc_after_break doesn't end up confused.
>> +
>> + - In non-stop if we insert a breakpoint (e.g., a step-resume)
>> + in one thread after another thread that was stepping had been
>> + momentarily paused for a step-over. When we re-resume the
>> + stepping thread, it may be resumed from that address with a
>> + breakpoint that hasn't trapped yet. Seen with
>> + gdb.threads/non-stop-fair-events.exp, on targets that don't
>> + do displaced stepping. */
>
> Does this comment mean the sequence like this?
>
> thread B is paused,
> a breakpoint is inserted at address PC for thread C,
> thread B may be resumed from address PC
Yep. In that case, we should _not_ step over the breakpoint,
as thread B has not reported that breakpoint event. Detecting the
case and reporting the breakpoint immediately without actually
resuming isn't correct, as in that case we'd lose the signal.
It's just simpler to let the breakpoint trap.
>
>>
>> +/* Select a thread at random, out of those which are resumed and have
>> + had events. */
>> +
>> +static struct thread_info *
>> +random_pending_event_thread (ptid_t waiton_ptid)
>> +{
>
> If GDB core knows select events at random, then we don't have to do
> something similar in linux-nat.c:select_event_lwp, right?
>
I think so, yes. Though I'd rather leave it be for a while,
so that the "maint set target-non-stop off" fallback option
is fully functional as backup plan. In the long term, if we manage
to merge linux-nat.c and gdbserver/linux-low.c, the resulting
code will still need it as long as we'd like to support gdbserver
connected to a gdb that doesn't force non-stop mode though, and,
probably as long as gdbserver's own step-over-breakpoints/tracepoints
implementation is baked in the backend too. :-/
Thanks,
Pedro Alves
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 09/17] Teach non-stop to do in-line step-overs (stop all, step, restart)
2015-04-22 20:03 ` Pedro Alves
@ 2015-04-24 9:06 ` Yao Qi
0 siblings, 0 replies; 62+ messages in thread
From: Yao Qi @ 2015-04-24 9:06 UTC (permalink / raw)
To: Pedro Alves; +Cc: Yao Qi, gdb-patches
Pedro Alves <palves@redhat.com> writes:
> I think so, yes. Though I'd rather leave it be for a while,
> so that the "maint set target-non-stop off" fallback option
> is fully functional as backup plan. In the long term, if we manage
> to merge linux-nat.c and gdbserver/linux-low.c, the resulting
> code will still need it as long as we'd like to support gdbserver
> connected to a gdb that doesn't force non-stop mode though, and,
> probably as long as gdbserver's own step-over-breakpoints/tracepoints
> implementation is baked in the backend too. :-/
Hmm, that is OK.
--
Yao (齐尧)
^ permalink raw reply [flat|nested] 62+ messages in thread
* Re: [PATCH v3 09/17] Teach non-stop to do in-line step-overs (stop all, step, restart)
2015-04-17 10:52 ` [PATCH v3 09/17] Teach non-stop to do in-line step-overs (stop all, step, restart) Pedro Alves
2015-04-17 11:01 ` Pedro Alves
2015-04-21 15:01 ` Yao Qi
@ 2015-04-27 20:17 ` Doug Evans
2015-05-19 18:09 ` Pedro Alves
2 siblings, 1 reply; 62+ messages in thread
From: Doug Evans @ 2015-04-27 20:17 UTC (permalink / raw)
To: Pedro Alves; +Cc: gdb-patches
Pedro Alves writes:
> ...
> gdb/ChangeLog:
> 2015-04-17 Pedro Alves <pedro@codesourcery.com>
>
> * breakpoint.c (breakpoints_should_be_inserted_now): If any thread
> has a pending status, return true.
> * gdbthread.h: Include target/waitstatus.h.
> (struct thread_suspend_state) <stop_reason, waitstatus_pending_p,
> stop_pc>: New fields.
> (struct thread_info) <resumed>: New field.
> (set_resumed): Declare.
> * infrun.c: Include "event-loop.h".
> (infrun_async_inferior_event_token, infrun_is_async): New globals.
> (infrun_async): New function.
> (clear_step_over_info): Add debug output.
> (displaced_step_in_progress_any_inferior): New function.
> (displaced_step_fixup): New returns int.
> (start_step_over): Handle in-line step-overs too. Assert the
> thread is marked resumed.
> (resume_cleanups): Clear the thread's resumed flag.
> (resume): Set the thread's resumed flag. Return early if the
> thread has a pending status. Allow stepping a breakpoint with no
> signal.
> (proceed): Adjust to check 'resumed' instead of 'executing'.
> (clear_proceed_status_thread): If the thread has a pending status,
> and that status is a finished step, discard the pending status.
> (clear_proceed_status): Don't clear step_over_info here.
> (random_pending_event_thread, do_target_wait): New functions.
> (prepare_for_detach, wait_for_inferior, fetch_inferior_event): Use
> do_target_wait.
> (wait_one): New function.
> (THREAD_STOPPED_BY): New macro.
> (thread_stopped_by_watchpoint, thread_stopped_by_sw_breakpoint)
> (thread_stopped_by_hw_breakpoint): New functions.
> (switch_to_thread_cleanup, save_waitstatus, stop_all_threads): New
> functions.
> (handle_inferior_event): Also call set_resumed(false) on all
> threads implicitly stopped by the event.
> (restart_threads, resumed_thread_with_pending_status): New
> functions.
> (finish_step_over): If we were doing an in-line step-over before,
> and no longer are after trying to start a new step-over, restart
> all threads. If we have multiple threads with pending events,
> save the current event and go through the event loop again.
> (handle_signal_stop): Return early if finish_step_over returns
> false.
> <random signal>: Clear step_over_info before delivering
> the signal.
> (keep_going_stepped_thread): Use internal_error instead of
> gdb_assert. Mark the thread as resumed.
> (keep_going_pass): Assert the thread isn't already resumed. If
> some other thread is doing an in-line step-over, defer the resume.
> If we just started a new in-line step-over, stop all threads.
> Don't clear step_over_info.
> (infrun_async_inferior_event_handler): New function.
> (_initialize_infrun): Create async event handler with
> infrun_async_inferior_event_handler as callback.
> (infrun_async): New declaration.
> * target.c (target_async): New function.
> * target.h (target_async): Declare macro and readd as function
> declaration.
> * target/waitstatus.h (enum target_stop_reason)
> <TARGET_STOPPED_BY_SINGLE_STEP>: New value.
> * thread.c (new_thread): Clear the new waitstatus field.
> (set_resumed): New function.
Hi.
A few comments inline.
> ...
> diff --git a/gdb/breakpoint.c b/gdb/breakpoint.c
> index 31869f1..e23d223 100644
> --- a/gdb/breakpoint.c
> +++ b/gdb/breakpoint.c
> @@ -468,6 +468,8 @@ breakpoints_should_be_inserted_now (void)
> }
> else if (target_has_execution)
> {
> + struct thread_info *tp;
> +
> if (always_inserted_mode)
> {
> /* The user wants breakpoints inserted even if all threads
> @@ -477,6 +479,13 @@ breakpoints_should_be_inserted_now (void)
>
> if (threads_are_executing ())
> return 1;
Not a problem introduced by this patch, but as an fyi, the terminology
employed here is a bit confusing.
Why would we want to insert breakpoints into executing threads,
or when threads are executing? That's what a simple reading of this
code says is happening. This reading can't be correct of course. :-)
> +
> + /* Don't remove breakpoints yet if, even though all threads are
> + stopped, we still have events to process. */
> + ALL_NON_EXITED_THREADS (tp)
> + if (tp->resumed
> + && tp->suspend.waitstatus_pending_p)
> + return 1;
Plus, this function is named "breakpoints_should_be_inserted_now"
but the comment is talking about whether breakpoints should be removed.
Can you elaborate on how to interpret the name of this function?
Guessing at how I'm supposed to interpret what this function is for,
is a better name
"breakpoints_should_have_been_inserted_by_now_or_should_remain_inserted"?
[Not that that's my recommendation :-). Just trying to understand how
to read this function.]
> }
> return 0;
> }
> diff --git a/gdb/gdbthread.h b/gdb/gdbthread.h
> index 7052ee1..22c8eb2 100644
> --- a/gdb/gdbthread.h
> +++ b/gdb/gdbthread.h
> @@ -29,6 +29,7 @@ struct symtab;
> #include "inferior.h"
> #include "btrace.h"
> #include "common/vec.h"
> +#include "target/waitstatus.h"
>
> /* Frontend view of the thread state. Possible extensions: stepping,
> finishing, until(ling),... */
> @@ -159,6 +160,23 @@ struct thread_suspend_state
> should be suppressed, the core will take care of clearing this
> before the target is resumed. */
> enum gdb_signal stop_signal;
> +
> + /* The reason the thread last stopped, if we need to track it
> + (breakpoint, watchpoint, etc.) */
> + enum target_stop_reason stop_reason;
> +
> + /* The waitstatus for this thread's last event. */
> + struct target_waitstatus waitstatus;
> + /* If true WAITSTATUS hasn't been handled yet. */
> + int waitstatus_pending_p;
> +
> + /* Record the pc of the thread the last time it stopped. (This is
> + not the current thread's PC as that may have changed since the
> + last stop, e.g., "return" command, or "p $pc = 0xf000"). This
> + used in coordination with stop_reason and waitstatus_pending_p:
s/used/is used/ ?
> + if the thread's PC is changed since it last stopped, a pending
> + breakpoint waitstatus is discarded. */
> + CORE_ADDR stop_pc;
> };
>
> typedef struct value *value_ptr;
> @@ -183,6 +201,14 @@ struct thread_info
> thread is off and running. */
> int executing;
>
> + /* Non-zero if this thread will be/has been resumed. Note that a
> + thread can be marked both as not-executing and resumed at the
> + same time. This happens if we try to resume a thread that has a
> + wait status pending. We shouldn't let the thread run until that
> + wait status has been processed, but we should not process that
> + wait status if we didn't try to let the thread run. */
> + int resumed;
I suspect this will be another source of confusion, but I don't have
a good suggestion at the moment for how to improve it.
The "will be" in the comment speaks in future tense, but the name
"resumed" is past tense. Maybe (though I'd have to spend more time
reading the code to be sure) it would make sense to have this be
multi-state: not-resumed, to-be-resumed, and resumed; thus splitting up
"will be" resumed from "has been" resumed.
> +
> /* Frontend view of the thread state. Note that the THREAD_RUNNING/
> THREAD_STOPPED states are different from EXECUTING. When the
> thread is stopped internally while handling an internal event,
> @@ -400,6 +426,11 @@ extern int thread_count (void);
> /* Switch from one thread to another. */
> extern void switch_to_thread (ptid_t ptid);
>
> +/* Marks or clears thread(s) PTID as resumed. If PTID is
> + MINUS_ONE_PTID, applies to all threads. If ptid_is_pid(PTID) is
> + true, applies to all threads of the process pointed at by PTID. */
> +extern void set_resumed (ptid_t ptid, int resumed);
> +
> /* Marks thread PTID is running, or stopped.
> If PTID is minus_one_ptid, marks all threads. */
> extern void set_running (ptid_t ptid, int running);
> diff --git a/gdb/infrun.c b/gdb/infrun.c
> index 6415fa2..8859b9f 100644
> --- a/gdb/infrun.c
> +++ b/gdb/infrun.c
> @@ -60,6 +60,36 @@
> #include "target-descriptions.h"
> #include "target-dcache.h"
> #include "terminal.h"
> +#include "event-loop.h"
> +
> +/* Asynchronous signal handler registered as event loop source for
> + when we have pending events ready to be passed to the core. */
> +static struct async_event_handler *infrun_async_inferior_event_token;
> +
> +/* Stores whether infrun_async was previously enabled or disabled.
> + Starts off as -1, indicating "never enabled/disabled". */
> +static int infrun_is_async = -1;
> +
> +/* See infrun.h. */
> +
> +void
> +infrun_async (int enable)
> +{
> + if (infrun_is_async != enable)
> + {
> + infrun_is_async = enable;
> +
> + if (debug_infrun)
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: infrun_async(%d)\n",
> + enable);
> +
> + if (enable)
> + mark_async_event_handler (infrun_async_inferior_event_token);
> + else
> + clear_async_event_handler (infrun_async_inferior_event_token);
> + }
> +}
>
> /* Prototypes for local functions */
>
> @@ -1305,6 +1335,9 @@ set_step_over_info (struct address_space *aspace, CORE_ADDR address,
> static void
> clear_step_over_info (void)
> {
> + if (debug_infrun)
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: clear_step_over_info\n");
> step_over_info.aspace = NULL;
> step_over_info.address = 0;
> step_over_info.nonsteppable_watchpoint_p = 0;
> @@ -1477,6 +1510,23 @@ get_displaced_stepping_state (int pid)
> return NULL;
> }
>
> +/* Returns true if any inferior has a thread doing a displaced
> + step. */
> +
> +static int
> +displaced_step_in_progress_any_inferior (void)
> +{
> + struct displaced_step_inferior_state *state;
> +
> + for (state = displaced_step_inferior_states;
> + state != NULL;
> + state = state->next)
> + if (!ptid_equal (state->step_ptid, null_ptid))
> + return 1;
> +
> + return 0;
> +}
> +
> /* Return true if process PID has a thread doing a displaced step. */
>
> static int
> @@ -1790,21 +1840,28 @@ displaced_step_restore (struct displaced_step_inferior_state *displaced,
> displaced->step_copy));
> }
>
> -static void
> +/* If we displaced stepped an instruction successfully, adjust
> + registers and memory to yield the same effect the instruction would
> + have had if we had executed it at its original address, and return
> + 1. If the instruction didn't complete, relocate the PC and return
> + -1. If the thread wasn't displaced stepping, return 0. */
> +
> +static int
> displaced_step_fixup (ptid_t event_ptid, enum gdb_signal signal)
> {
> struct cleanup *old_cleanups;
> struct displaced_step_inferior_state *displaced
> = get_displaced_stepping_state (ptid_get_pid (event_ptid));
> + int ret;
>
> /* Was any thread of this process doing a displaced step? */
> if (displaced == NULL)
> - return;
> + return 0;
>
> /* Was this event for the pid we displaced? */
> if (ptid_equal (displaced->step_ptid, null_ptid)
> || ! ptid_equal (displaced->step_ptid, event_ptid))
> - return;
> + return 0;
>
> old_cleanups = make_cleanup (displaced_step_clear_cleanup, displaced);
>
> @@ -1827,6 +1884,7 @@ displaced_step_fixup (ptid_t event_ptid, enum gdb_signal signal)
> displaced->step_original,
> displaced->step_copy,
> get_thread_regcache (displaced->step_ptid));
> + ret = 1;
> }
> else
> {
> @@ -1837,11 +1895,14 @@ displaced_step_fixup (ptid_t event_ptid, enum gdb_signal signal)
>
> pc = displaced->step_original + (pc - displaced->step_copy);
> regcache_write_pc (regcache, pc);
> + ret = -1;
> }
>
> do_cleanups (old_cleanups);
>
> displaced->step_ptid = null_ptid;
> +
> + return ret;
> }
>
> /* Data to be passed around while handling an event. This data is
> @@ -1890,10 +1951,19 @@ start_step_over (void)
> {
> struct thread_info *tp, *next;
>
> + /* Don't start a new step-over if we already have a step-over
> + operation ongoing. */
> + if (step_over_info_valid_p ())
> + return 0;
> +
> for (tp = step_over_queue_head; tp != NULL; tp = next)
> {
> struct execution_control_state ecss;
> struct execution_control_state *ecs = &ecss;
> + enum step_over_what step_what;
> + int must_be_in_line;
> + struct regcache *regcache = get_thread_regcache (tp->ptid);
> + struct gdbarch *gdbarch = get_regcache_arch (regcache);
>
> next = thread_step_over_chain_next (tp);
>
> @@ -1902,6 +1972,17 @@ start_step_over (void)
> if (displaced_step_in_progress (ptid_get_pid (tp->ptid)))
> continue;
>
> + step_what = thread_still_needs_step_over (tp);
> + must_be_in_line = ((step_what & STEP_OVER_WATCHPOINT)
> + || ((step_what & STEP_OVER_BREAKPOINT)
> + && !use_displaced_stepping (gdbarch)));
> +
> + /* We currently stop all threads of all processes to step-over
> + in-line. If we need to start a new in-line step-over, let
> + any pending displace steps finish first. */
> + if (must_be_in_line && displaced_step_in_progress_any_inferior ())
> + return 0;
> +
> thread_step_over_chain_remove (tp);
>
> if (step_over_queue_head == NULL)
> @@ -1911,13 +1992,16 @@ start_step_over (void)
> "infrun: step-over queue now empty\n");
> }
>
> - if (tp->control.trap_expected || tp->executing)
> + if (tp->control.trap_expected
> + || tp->resumed
> + || tp->executing)
> {
> internal_error (__FILE__, __LINE__,
> "[%s] has inconsistent state: "
> - "trap_expected=%d, executing=%d\n",
> + "trap_expected=%d, resumed=%d, executing=%d\n",
> target_pid_to_str (tp->ptid),
> tp->control.trap_expected,
> + tp->resumed,
> tp->executing);
> }
>
> @@ -1932,7 +2016,7 @@ start_step_over (void)
> wouldn't be able to resume anything else until the target
> stops again. In non-stop, the resume always resumes only TP,
> so it's OK to let the thread resume freely. */
> - if (!non_stop && !thread_still_needs_step_over (tp))
> + if (!non_stop && !step_what)
> continue;
>
> switch_to_thread (tp->ptid);
> @@ -1942,6 +2026,15 @@ start_step_over (void)
> if (!ecs->wait_some_more)
> error (_("Command aborted."));
>
> + gdb_assert (tp->resumed);
> +
> + /* If we started a new in-line step-over, we're done. */
> + if (step_over_info_valid_p ())
> + {
> + gdb_assert (tp->control.trap_expected);
> + return 1;
> + }
> +
> if (!non_stop)
> {
> /* On all-stop, shouldn't have resumed unless we needed a
> @@ -2157,12 +2250,38 @@ resume (enum gdb_signal sig)
> single-step). */
> int step;
>
> - tp->stepped_breakpoint = 0;
> -
> gdb_assert (!thread_is_in_step_over_chain (tp));
>
> QUIT;
>
> + if (tp->suspend.waitstatus_pending_p)
> + {
> + if (debug_infrun)
> + {
> + char *statstr;
> +
> + statstr = target_waitstatus_to_string (&tp->suspend.waitstatus);
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: resume: thread %s has pending wait status %s "
> + "(currently_stepping=%d).\n",
> + target_pid_to_str (tp->ptid), statstr,
> + currently_stepping (tp));
> + xfree (statstr);
Not something that has to be done with this patch of course,
but it's nice that we don't have to track the memory of target_pid_to_str;
IWBN to be able to do the same for target_waitstatus_to_string.
[In C++ it could just return a string, and we *could* just wait for C++.
Just a thought.]
> + }
> +
> + tp->resumed = 1;
> + /* Avoid confusing the next resume, if the next stop/resume
> + happens to apply to another thread. */
> + tp->suspend.stop_signal = GDB_SIGNAL_0;
This is a bit confusing. How will the value of stop_signal in one
thread affect stop/resume in another thread?
Plus, the reader is left worried that we just clobbered the signal
that we need to resume thread tp with. Can you elaborate on what's
happening here?
> + discard_cleanups (old_cleanups);
> +
> + if (target_can_async_p ())
> + target_async (1);
> + return;
> + }
> +
> + tp->stepped_breakpoint = 0;
> +
> /* Depends on stepped_breakpoint. */
> step = currently_stepping (tp);
>
> @@ -2267,6 +2386,7 @@ resume (enum gdb_signal sig)
> resume_ptid = user_visible_resume_ptid (user_step);
> do_target_resume (resume_ptid, 0, GDB_SIGNAL_0);
> discard_cleanups (old_cleanups);
> + tp->resumed = 1;
> return;
> }
> }
> @@ -2392,11 +2512,12 @@ resume (enum gdb_signal sig)
> if (execution_direction != EXEC_REVERSE
> && step && breakpoint_inserted_here_p (aspace, pc))
> {
> - /* The only case we currently need to step a breakpoint
> - instruction is when we have a signal to deliver. See
> - handle_signal_stop where we handle random signals that could
> - take out us out of the stepping range. Normally, in that
> - case we end up continuing (instead of stepping) over the
> + /* There are two cases where we currently need to step a
> + breakpoint instruction when we have a signal to deliver:
> +
> + - See handle_signal_stop where we handle random signals that
> + could take out us out of the stepping range. Normally, in
> + that case we end up continuing (instead of stepping) over the
> signal handler with a breakpoint at PC, but there are cases
> where we should _always_ single-step, even if we have a
> step-resume breakpoint, like when a software watchpoint is
> @@ -2409,8 +2530,20 @@ resume (enum gdb_signal sig)
> recurses and executes PC again, it'll miss the breakpoint.
> So we leave the breakpoint inserted anyway, but we need to
> record that we tried to step a breakpoint instruction, so
> - that adjust_pc_after_break doesn't end up confused. */
> - gdb_assert (sig != GDB_SIGNAL_0);
> + that adjust_pc_after_break doesn't end up confused.
> +
> + - In non-stop if we insert a breakpoint (e.g., a step-resume)
> + in one thread after another thread that was stepping had been
> + momentarily paused for a step-over. When we re-resume the
> + stepping thread, it may be resumed from that address with a
> + breakpoint that hasn't trapped yet. Seen with
> + gdb.threads/non-stop-fair-events.exp, on targets that don't
> + do displaced stepping. */
> +
> + if (debug_infrun)
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: resume: [%s] stepped breakpoint\n",
> + target_pid_to_str (tp->ptid));
>
> tp->stepped_breakpoint = 1;
>
> @@ -2448,6 +2581,7 @@ resume (enum gdb_signal sig)
> }
>
> do_target_resume (resume_ptid, step, sig);
> + tp->resumed = 1;
> discard_cleanups (old_cleanups);
> }
> \f
> @@ -2464,6 +2598,37 @@ clear_proceed_status_thread (struct thread_info *tp)
> "infrun: clear_proceed_status_thread (%s)\n",
> target_pid_to_str (tp->ptid));
>
> + /* If we're starting a new sequence, then the previous finished
> + single-step is no longer relevant. */
> + if (tp->suspend.waitstatus_pending_p)
> + {
> + if (tp->suspend.stop_reason == TARGET_STOPPED_BY_SINGLE_STEP)
> + {
> + if (debug_infrun)
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: clear_proceed_status: pending "
> + "event of %s was a finished step. "
> + "Discarding.\n",
> + target_pid_to_str (tp->ptid));
> +
> + tp->suspend.waitstatus_pending_p = 0;
> + tp->suspend.stop_reason = TARGET_STOPPED_BY_NO_REASON;
> + }
> + else if (debug_infrun)
> + {
> + char *statstr;
> +
> + statstr = target_waitstatus_to_string (&tp->suspend.waitstatus);
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: clear_proceed_status_thread: thread %s "
> + "has pending wait status %s "
> + "(currently_stepping=%d).\n",
> + target_pid_to_str (tp->ptid), statstr,
> + currently_stepping (tp));
> + xfree (statstr);
> + }
> + }
> +
> /* If this signal should not be seen by program, give it zero.
> Used for debugging signals. */
> if (!signal_pass_state (tp->suspend.stop_signal))
> @@ -2527,8 +2692,6 @@ clear_proceed_status (int step)
>
> stop_after_trap = 0;
>
> - clear_step_over_info ();
> -
> observer_notify_about_to_proceed ();
>
> if (stop_registers)
> @@ -2777,7 +2940,7 @@ proceed (CORE_ADDR addr, enum gdb_signal siggnal)
> /* A new displaced stepping sequence was started. In all-stop,
> we can't talk to the target anymore until it next stops. */
> }
> - else if (!tp->executing && !thread_is_in_step_over_chain (tp))
> + else if (!tp->resumed && !thread_is_in_step_over_chain (tp))
> {
> /* The thread wasn't started, and isn't queued, run it now. */
> reset_ecs (ecs, tp);
> @@ -3076,6 +3239,179 @@ print_target_wait_results (ptid_t waiton_ptid, ptid_t result_ptid,
> ui_file_delete (tmp_stream);
> }
>
> +/* Select a thread at random, out of those which are resumed and have
> + had events. */
> +
> +static struct thread_info *
> +random_pending_event_thread (ptid_t waiton_ptid)
> +{
> + struct thread_info *event_tp;
> + int num_events = 0;
> + int random_selector;
> +
> + /* First see how many events we have. Count only resumed threads
> + that have an event pending. */
> + ALL_NON_EXITED_THREADS (event_tp)
> + if (ptid_match (event_tp->ptid, waiton_ptid)
> + && event_tp->resumed
> + && event_tp->suspend.waitstatus_pending_p)
> + num_events++;
> +
> + if (num_events == 0)
> + return NULL;
> +
> + /* Now randomly pick a thread out of those that have had events. */
> + random_selector = (int)
> + ((num_events * (double) rand ()) / (RAND_MAX + 1.0));
> +
> + if (debug_infrun && num_events > 1)
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: Found %d events, selecting #%d\n",
> + num_events, random_selector);
> +
> + /* Select the Nth thread that has had an event. */
> + ALL_NON_EXITED_THREADS (event_tp)
> + if (ptid_match (event_tp->ptid, waiton_ptid)
> + && event_tp->resumed
> + && event_tp->suspend.waitstatus_pending_p)
> + if (random_selector-- == 0)
> + break;
> +
> + return event_tp;
> +}
> +
> +/* Wrapper for target_wait that first checks whether threads have
> + pending status to report before actually asking the target for more
> + events. */
> +
> +static ptid_t
> +do_target_wait (ptid_t ptid, struct target_waitstatus *status, int options)
> +{
> + ptid_t event_ptid;
> + struct thread_info *tp;
> +
> + /* First check if there is a resumed thread with a wait status
> + pending. */
> + if (ptid_equal (ptid, minus_one_ptid) || ptid_is_pid (ptid))
> + {
> + tp = random_pending_event_thread (ptid);
> + }
> + else
> + {
> + if (debug_infrun)
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: Waiting for specific thread %s.\n",
> + target_pid_to_str (ptid));
> +
> + /* We have a specific thread to check. */
> + tp = find_thread_ptid (ptid);
> + gdb_assert (tp != NULL);
> + if (!tp->suspend.waitstatus_pending_p)
> + tp = NULL;
> + }
> +
> + if (tp != NULL
> + && (tp->suspend.stop_reason == TARGET_STOPPED_BY_SW_BREAKPOINT
> + || tp->suspend.stop_reason == TARGET_STOPPED_BY_HW_BREAKPOINT))
> + {
> + struct regcache *regcache = get_thread_regcache (tp->ptid);
> + struct gdbarch *gdbarch = get_regcache_arch (regcache);
> + CORE_ADDR pc;
> + int discard = 0;
> +
> + pc = regcache_read_pc (regcache);
> +
> + if (pc != tp->suspend.stop_pc)
> + {
> + if (debug_infrun)
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: PC of %s changed. was=%s, now=%s\n",
> + target_pid_to_str (tp->ptid),
> + paddress (target_gdbarch (), tp->prev_pc),
> + paddress (target_gdbarch (), pc));
s/target_gdbarch ()/gdbarch/ ?
> + discard = 1;
> + }
> + else if (!breakpoint_inserted_here_p (get_regcache_aspace (regcache), pc))
> + {
> + if (debug_infrun)
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: previous breakpoint of %s, at %s gone\n",
> + target_pid_to_str (tp->ptid),
> + paddress (target_gdbarch (), pc));
> +
> + discard = 1;
> + }
> +
> + if (discard)
> + {
> + if (debug_infrun)
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: pending event of %s cancelled.\n",
> + target_pid_to_str (tp->ptid));
> +
> + tp->suspend.waitstatus.kind = TARGET_WAITKIND_SPURIOUS;
> + tp->suspend.stop_reason = TARGET_STOPPED_BY_NO_REASON;
> + }
> + }
> +
> + if (tp != NULL)
> + {
> + if (debug_infrun)
> + {
> + char *statstr;
> +
> + statstr = target_waitstatus_to_string (&tp->suspend.waitstatus);
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: Using pending wait status %s for %s.\n",
> + statstr,
> + target_pid_to_str (tp->ptid));
> + xfree (statstr);
> + }
> +
> + /* Now that we've selected our final event LWP, un-adjust its PC
> + if it was a software breakpoint (and the target doesn't
> + always adjust the PC itself). */
> + if (tp->suspend.stop_reason == TARGET_STOPPED_BY_SW_BREAKPOINT
> + && !target_supports_stopped_by_sw_breakpoint ())
> + {
> + struct regcache *regcache;
> + struct gdbarch *gdbarch;
> + int decr_pc;
> +
> + regcache = get_thread_regcache (tp->ptid);
> + gdbarch = get_regcache_arch (regcache);
> +
> + decr_pc = gdbarch_decr_pc_after_break (gdbarch);
> + if (decr_pc != 0)
> + {
> + CORE_ADDR pc;
> +
> + pc = regcache_read_pc (regcache);
> + regcache_write_pc (regcache, pc + decr_pc);
> + }
> + }
> +
> + tp->suspend.stop_reason = TARGET_STOPPED_BY_NO_REASON;
> + *status = tp->suspend.waitstatus;
> + tp->suspend.waitstatus_pending_p = 0;
> +
> + /* Wake up the event loop again, until all pending events are
> + processed. */
> + if (target_is_async_p ())
> + mark_async_event_handler (infrun_async_inferior_event_token);
> + return tp->ptid;
> + }
> +
> + /* But if we don't find one, we'll have to wait. */
> +
> + if (deprecated_target_wait_hook)
> + event_ptid = deprecated_target_wait_hook (ptid, status, options);
> + else
> + event_ptid = target_wait (ptid, status, options);
> +
> + return event_ptid;
> +}
> +
> /* Prepare and stabilize the inferior for detaching it. E.g.,
> detaching while a thread is displaced stepping is a recipe for
> crashing it, as nothing would readjust the PC out of the scratch
> @@ -3119,10 +3455,7 @@ prepare_for_detach (void)
> don't get any event. */
> target_dcache_invalidate ();
>
> - if (deprecated_target_wait_hook)
> - ecs->ptid = deprecated_target_wait_hook (pid_ptid, &ecs->ws, 0);
> - else
> - ecs->ptid = target_wait (pid_ptid, &ecs->ws, 0);
> + ecs->ptid = do_target_wait (pid_ptid, &ecs->ws, 0);
>
> if (debug_infrun)
> print_target_wait_results (pid_ptid, ecs->ptid, &ecs->ws);
> @@ -3194,10 +3527,7 @@ wait_for_inferior (void)
> don't get any event. */
> target_dcache_invalidate ();
>
> - if (deprecated_target_wait_hook)
> - ecs->ptid = deprecated_target_wait_hook (waiton_ptid, &ecs->ws, 0);
> - else
> - ecs->ptid = target_wait (waiton_ptid, &ecs->ws, 0);
> + ecs->ptid = do_target_wait (waiton_ptid, &ecs->ws, 0);
>
> if (debug_infrun)
> print_target_wait_results (waiton_ptid, ecs->ptid, &ecs->ws);
> @@ -3294,11 +3624,7 @@ fetch_inferior_event (void *client_data)
> make_cleanup_restore_integer (&execution_direction);
> execution_direction = target_execution_direction ();
>
> - if (deprecated_target_wait_hook)
> - ecs->ptid =
> - deprecated_target_wait_hook (waiton_ptid, &ecs->ws, TARGET_WNOHANG);
> - else
> - ecs->ptid = target_wait (waiton_ptid, &ecs->ws, TARGET_WNOHANG);
> + ecs->ptid = do_target_wait (waiton_ptid, &ecs->ws, TARGET_WNOHANG);
>
> if (debug_infrun)
> print_target_wait_results (waiton_ptid, ecs->ptid, &ecs->ws);
> @@ -3671,6 +3997,349 @@ get_inferior_stop_soon (ptid_t ptid)
> return inf->control.stop_soon;
> }
>
> +/* Wait for one event. Store the resulting waitstatus in WS, and
> + return the event ptid. */
> +
> +static ptid_t
> +wait_one (struct target_waitstatus *ws)
> +{
> + ptid_t event_ptid;
> + ptid_t wait_ptid = minus_one_ptid;
> +
> + overlay_cache_invalid = 1;
> +
> + /* Flush target cache before starting to handle each event.
> + Target was running and cache could be stale. This is just a
> + heuristic. Running threads may modify target memory, but we
> + don't get any event. */
> + target_dcache_invalidate ();
> +
> + if (deprecated_target_wait_hook)
> + event_ptid = deprecated_target_wait_hook (wait_ptid, ws, 0);
> + else
> + event_ptid = target_wait (wait_ptid, ws, 0);
> +
> + if (debug_infrun)
> + print_target_wait_results (wait_ptid, event_ptid, ws);
> +
> + if (ws->kind == TARGET_WAITKIND_SYSCALL_ENTRY
> + || ws->kind == TARGET_WAITKIND_SYSCALL_RETURN)
> + ws->value.syscall_number = UNKNOWN_SYSCALL;
IWBN to have a comment explaining why we set UNKNOWN_SYSCALL here.
> +
> + return event_ptid;
> +}
> +
> +/* Generate a wrapper for target_stopped_by_REASON that works on PTID
> + instead of the current thread. */
> +#define THREAD_STOPPED_BY(REASON) \
> +static int \
> +thread_stopped_by_ ## REASON (ptid_t ptid) \
> +{ \
> + struct cleanup *old_chain; \
> + int res; \
> + \
> + old_chain = save_inferior_ptid (); \
> + inferior_ptid = ptid; \
> + \
> + res = target_stopped_by_ ## REASON (); \
> + \
> + do_cleanups (old_chain); \
> + \
> + return res; \
> +}
> +
> +/* Generate thread_stopped_by_watchpoint. */
> +THREAD_STOPPED_BY (watchpoint)
> +/* Generate thread_stopped_by_sw_breakpoint. */
> +THREAD_STOPPED_BY (sw_breakpoint)
> +/* Generate thread_stopped_by_hw_breakpoint. */
> +THREAD_STOPPED_BY (hw_breakpoint)
People always nitpick when I do stuff like this. :-)
Don't change this on my account though.
> +
> +/* Cleanups that switches to the PTID pointed at by PTID_P. */
> +
> +static void
> +switch_to_thread_cleanup (void *ptid_p)
> +{
> + ptid_t ptid = *(ptid_t *) ptid_p;
> +
> + switch_to_thread (ptid);
> +}
> +
> +/* Save the thread's event and stop reason to process it later. */
> +
> +static void
> +save_waitstatus (struct thread_info *tp, struct target_waitstatus *ws)
> +{
> + struct regcache *regcache;
> + struct address_space *aspace;
> +
> + if (debug_infrun)
> + {
> + char *statstr;
> +
> + statstr = target_waitstatus_to_string (ws);
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: saving status %s for %d.%ld.%ld\n",
> + statstr,
> + ptid_get_pid (tp->ptid),
> + ptid_get_lwp (tp->ptid),
> + ptid_get_tid (tp->ptid));
> + xfree (statstr);
> + }
> +
> + /* Record for later. */
> + tp->suspend.waitstatus = *ws;
> + tp->suspend.waitstatus_pending_p = 1;
> +
> + regcache = get_thread_regcache (tp->ptid);
> + aspace = get_regcache_aspace (regcache);
> +
> + if (ws->kind == TARGET_WAITKIND_STOPPED
> + && ws->value.sig == GDB_SIGNAL_TRAP)
> + {
> + CORE_ADDR pc = regcache_read_pc (regcache);
> +
> + adjust_pc_after_break (tp, &tp->suspend.waitstatus);
> +
> + if (thread_stopped_by_watchpoint (tp->ptid))
> + {
> + tp->suspend.stop_reason
> + = TARGET_STOPPED_BY_WATCHPOINT;
> + }
> + else if (target_supports_stopped_by_sw_breakpoint ()
> + && thread_stopped_by_sw_breakpoint (tp->ptid))
> + {
> + tp->suspend.stop_reason
> + = TARGET_STOPPED_BY_SW_BREAKPOINT;
> + }
> + else if (target_supports_stopped_by_hw_breakpoint ()
> + && thread_stopped_by_hw_breakpoint (tp->ptid))
> + {
> + tp->suspend.stop_reason
> + = TARGET_STOPPED_BY_HW_BREAKPOINT;
> + }
> + else if (!target_supports_stopped_by_hw_breakpoint ()
> + && hardware_breakpoint_inserted_here_p (aspace,
> + pc))
> + {
> + tp->suspend.stop_reason
> + = TARGET_STOPPED_BY_HW_BREAKPOINT;
> + }
> + else if (!target_supports_stopped_by_sw_breakpoint ()
> + && software_breakpoint_inserted_here_p (aspace,
> + pc))
> + {
> + tp->suspend.stop_reason
> + = TARGET_STOPPED_BY_SW_BREAKPOINT;
> + }
> + else if (!thread_has_single_step_breakpoints_set (tp)
> + && currently_stepping (tp))
> + {
> + tp->suspend.stop_reason
> + = TARGET_STOPPED_BY_SINGLE_STEP;
> + }
> + }
> +}
> +
> +/* Stop all threads. */
> +
> +static void
> +stop_all_threads (void)
> +{
> + /* We may need multiple passes to discover all threads. */
> + int pass;
> + int iterations = 0;
> + ptid_t entry_ptid;
> + struct cleanup *old_chain;
> +
> + gdb_assert (non_stop);
> +
> + if (debug_infrun)
> + fprintf_unfiltered (gdb_stdlog, "infrun: stop_all_threads\n");
> +
> + entry_ptid = inferior_ptid;
> + old_chain = make_cleanup (switch_to_thread_cleanup, &entry_ptid);
> +
> + /* Stop threads in two passes since threads could be spawning as we
> + go through the first pass. In the second pass, we will stop such
> + spawned threads. */
> + for (pass = 0; pass < 2; pass++, iterations++)
Can you rephrase the "Stop threads in two passes ... In the second pass ..."
comment? What's happening here is that we keep iterating until two passes
find no new threads (IIUC).
> + {
> + if (debug_infrun)
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: stop_all_threads, pass=%d, "
> + "iterations=%d\n", pass, iterations);
> + while (1)
> + {
> + ptid_t event_ptid;
> + struct target_waitstatus ws;
> + int need_wait = 0;
> + struct thread_info *t;
> +
> + update_thread_list ();
> +
> + /* Go through all threads looking for threads that we need
> + to tell the target to stop. */
> + ALL_NON_EXITED_THREADS (t)
> + {
> + if (t->executing)
> + {
> + /* If already stopping, don't request a stop again.
> + We just haven't seen the notification yet. */
> + if (!t->stop_requested)
> + {
> + if (debug_infrun)
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: %s executing, "
> + "need stop\n",
> + target_pid_to_str (t->ptid));
> + target_stop (t->ptid);
> + t->stop_requested = 1;
> + }
> + else
> + {
> + if (debug_infrun)
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: %s executing, "
> + "already stopping\n",
> + target_pid_to_str (t->ptid));
> + }
> +
> + if (t->stop_requested)
> + need_wait = 1;
> + }
> + else
> + {
> + if (debug_infrun)
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: %s not executing\n",
> + target_pid_to_str (t->ptid));
> +
> + /* The thread may be not executing, but still be
> + resumed with a pending status to process. */
> + t->resumed = 0;
> + }
> + }
> +
> + if (!need_wait)
> + break;
> +
> + /* If we find new threads on the second iteration, restart
> + over. We want to see two iterations in a row with all
> + threads stopped. */
> + if (pass > 0)
> + pass = -1;
> +
> + event_ptid = wait_one (&ws);
> + if (ws.kind == TARGET_WAITKIND_NO_RESUMED)
> + /* All resumed threads exited. */
> + ;
> + else if (ws.kind == TARGET_WAITKIND_EXITED
> + || ws.kind == TARGET_WAITKIND_SIGNALLED)
> + {
> + if (debug_infrun)
> + {
> + ptid_t ptid = pid_to_ptid (ws.value.integer);
> +
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: %s exited while "
> + "stopping threads\n",
> + target_pid_to_str (ptid));
> + }
> + }
> + else
> + {
> + if (!in_thread_list (event_ptid))
> + t = add_thread (event_ptid);
> + else
> + t = find_thread_ptid (event_ptid);
> +
> + t->stop_requested = 0;
> + t->executing = 0;
> + t->resumed = 0;
> + t->control.may_range_step = 0;
> +
> + if (ws.kind == TARGET_WAITKIND_STOPPED
> + && ws.value.sig == GDB_SIGNAL_0)
> + {
> + /* We caught the event that we intended to catch, so
> + there's no event pending. */
> + t->suspend.waitstatus.kind = TARGET_WAITKIND_IGNORE;
> + t->suspend.waitstatus_pending_p = 0;
> +
> + if (displaced_step_fixup (t->ptid, GDB_SIGNAL_0) < 0)
> + {
> + /* Add it back to the step-over queue. */
> + if (debug_infrun)
> + {
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: displaced-step of %s "
> + "canceled: adding back to the "
> + "step-over queue\n",
> + target_pid_to_str (t->ptid));
> + }
> + t->control.trap_expected = 0;
> + thread_step_over_chain_enqueue (t);
> + }
> + }
> + else
> + {
> + enum gdb_signal sig;
> + struct regcache *regcache;
> + struct address_space *aspace;
> +
> + if (debug_infrun)
> + {
> + char *statstr;
> +
> + statstr = target_waitstatus_to_string (&ws);
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: target_wait %s, saving "
> + "status for %d.%ld.%ld\n",
> + statstr,
> + ptid_get_pid (t->ptid),
> + ptid_get_lwp (t->ptid),
> + ptid_get_tid (t->ptid));
> + xfree (statstr);
> + }
> +
> + /* Record for later. */
> + save_waitstatus (t, &ws);
> +
> + sig = (ws.kind == TARGET_WAITKIND_STOPPED
> + ? ws.value.sig : GDB_SIGNAL_0);
> +
> + if (displaced_step_fixup (t->ptid, sig) < 0)
> + {
> + /* Add it back to the step-over queue. */
> + t->control.trap_expected = 0;
> + thread_step_over_chain_enqueue (t);
> + }
> +
> + regcache = get_thread_regcache (t->ptid);
> + t->suspend.stop_pc = regcache_read_pc (regcache);
> +
> + if (debug_infrun)
> + {
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: saved stop_pc=%s for %s "
> + "(currently_stepping=%d)\n",
> + paddress (target_gdbarch (),
> + t->suspend.stop_pc),
> + target_pid_to_str (t->ptid),
> + currently_stepping (t));
> + }
> + }
> + }
> + }
> + }
> +
> + do_cleanups (old_chain);
> +
> + if (debug_infrun)
> + fprintf_unfiltered (gdb_stdlog, "infrun: stop_all_threads done\n");
> +}
> +
> /* Given an execution control state that has been freshly filled in by
> an event from the inferior, figure out what it means and take
> appropriate action.
> @@ -3793,21 +4462,28 @@ handle_inferior_event (struct execution_control_state *ecs)
> we're handling a process exit in non-stop mode, there's nothing
> to do, as threads of the dead process are gone, and threads of
> any other process were left running. */
> - if (!non_stop)
> - set_executing (minus_one_ptid, 0);
> - else if (ecs->ws.kind == TARGET_WAITKIND_SIGNALLED
> - && ecs->ws.kind == TARGET_WAITKIND_EXITED)
> - {
> - ptid_t pid_ptid;
>
> - /* Some targets still have execution when a process exits.
> - E.g., for "checkpoint", when when a fork exits and is
> - mourned, linux-fork.c switches to another fork. */
> - pid_ptid = pid_to_ptid (ptid_get_pid (ecs->ptid));
> - set_executing (pid_ptid, 0);
> - }
> - else
> - set_executing (ecs->ptid, 0);
> + {
> + ptid_t mark_ptid;
> +
> + if (!non_stop)
> + mark_ptid = minus_one_ptid;
> + else if (ecs->ws.kind == TARGET_WAITKIND_SIGNALLED
> + && ecs->ws.kind == TARGET_WAITKIND_EXITED)
> + {
> + /* Some targets still have execution when a process exits.
> + E.g., for "checkpoint", when when a fork exits and is
> + mourned, linux-fork.c switches to another fork. */
> + mark_ptid = pid_to_ptid (ptid_get_pid (ecs->ptid));
> + }
> + else
> + mark_ptid = ecs->ptid;
> +
> + set_executing (mark_ptid, 0);
> +
> + /* Likewise the resumed flag. */
> + set_resumed (mark_ptid, 0);
> + }
>
> switch (ecs->ws.kind)
> {
> @@ -4220,16 +4896,137 @@ Cannot fill $_exitsignal with the correct signal number.\n"));
> }
> }
>
> -/* Called when we get an event that may finish an in-line or
> - out-of-line (displaced stepping) step-over started previously. */
> +/* Restart threads back to what they were trying to do back when we
> + paused them for an in-line step-over. The EVENT_THREAD thread is
> + ignored. */
>
> static void
> +restart_threads (struct thread_info *event_thread)
> +{
> + struct thread_info *tp;
> + struct thread_info *step_over = NULL;
> +
> + /* In case the instruction just stepped spawned a new thread. */
> + update_thread_list ();
> +
> + ALL_NON_EXITED_THREADS (tp)
> + {
> + struct execution_control_state ecss;
> + struct execution_control_state *ecs = &ecss;
> +
> + if (tp == event_thread)
> + {
> + if (debug_infrun)
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: restart threads: "
> + "[%s] is event thread\n",
> + target_pid_to_str (tp->ptid));
> + continue;
> + }
> +
> + if (!(tp->state == THREAD_RUNNING || tp->control.in_infcall))
> + {
> + if (debug_infrun)
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: restart threads: "
> + "[%s] not meant to be running\n",
> + target_pid_to_str (tp->ptid));
> + continue;
> + }
> +
> + if (tp->resumed)
> + {
> + if (debug_infrun)
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: restart threads: [%s] resumed\n",
> + target_pid_to_str (tp->ptid));
> + gdb_assert (tp->executing || tp->suspend.waitstatus_pending_p);
> + continue;
> + }
> +
> + if (thread_is_in_step_over_chain (tp))
> + {
> + if (debug_infrun)
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: restart threads: "
> + "[%s] needs step-over\n",
> + target_pid_to_str (tp->ptid));
> + gdb_assert (!tp->resumed);
> + continue;
> + }
> +
> +
> + if (tp->suspend.waitstatus_pending_p)
> + {
> + if (debug_infrun)
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: restart threads: "
> + "[%s] has pending status\n",
> + target_pid_to_str (tp->ptid));
> + tp->resumed = 1;
> + continue;
> + }
> +
> + /* If some thread needs to start a step-over at this point, it
> + should still be in the step-over queue, and thus skipped
> + above. */
> + if (thread_still_needs_step_over (tp))
> + {
> + internal_error (__FILE__, __LINE__,
> + "thread [%s] needs a step-over, but not in "
> + "step-over queue\n",
> + target_pid_to_str (tp->ptid));
> + }
> +
> + if (currently_stepping (tp))
> + {
> + if (debug_infrun)
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: restart threads: [%s] was stepping\n",
> + target_pid_to_str (tp->ptid));
> + keep_going_stepped_thread (tp);
> + }
> + else
> + {
> + if (debug_infrun)
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: restart threads: [%s] continuing\n",
> + target_pid_to_str (tp->ptid));
> + reset_ecs (ecs, tp);
> + switch_to_thread (tp->ptid);
> + keep_going_pass (ecs);
> + }
> + }
> +}
> +
> +/* Callback for iterate_over_threads. Find a resumed thread that has
> + a pending waitstatus. */
> +
> +static int
> +resumed_thread_with_pending_status (struct thread_info *tp,
> + void *arg)
> +{
> + return (tp->resumed
> + && tp->suspend.waitstatus_pending_p);
> +}
> +
> +/* Called when we get an event that may finish an in-line or
> + out-of-line (displaced stepping) step-over started previously.
> + Return true if the event is processed and we should go back to the
> + event loop; false if the caller should continue processing the
> + event. */
> +
> +static int
> finish_step_over (struct execution_control_state *ecs)
> {
> + int had_step_over_info;
> +
> displaced_step_fixup (ecs->ptid,
> ecs->event_thread->suspend.stop_signal);
>
> - if (step_over_info_valid_p ())
> + had_step_over_info = step_over_info_valid_p ();
> +
> + if (had_step_over_info)
> {
> /* If we're stepping over a breakpoint with all threads locked,
> then only the thread that was stepped should be reporting
> @@ -4241,11 +5038,99 @@ finish_step_over (struct execution_control_state *ecs)
> }
>
> if (!non_stop)
> - return;
> + return 0;
>
> /* Start a new step-over in another thread if there's one that
> needs it. */
> start_step_over ();
> +
> + /* If we were stepping over a breakpoint before, and haven't started
> + a new in-line step-over sequence, then restart all other threads
> + (except the event thread). We can't do this in all-stop, as then
> + e.g., we wouldn't be able to issue any other remote packet until
> + these other threads stop. */
> + if (had_step_over_info && !step_over_info_valid_p ())
> + {
> + struct thread_info *pending;
> +
> + /* If we only have threads with pending statuses, the restart
> + below won't restart any thread and so nothing re-inserts the
> + breakpoint we just stepped over. But we need it inserted
> + when we later process the pending events, otherwise if
> + another thread has a pending event for this breakpoint too,
> + we'd discard its event (because the breakpoint that
> + originally caused the event was no longer inserted). */
> + context_switch (ecs->ptid);
> + insert_breakpoints ();
> +
> + restart_threads (ecs->event_thread);
> +
> + /* If we have events pending, go through handle_inferior_event
> + again, picking up a pending event at random. This avoids
> + thread starvation. */
> +
> + /* But not if we just stepped over a watchpoint in order to let
> + the instruction execute so we can evaluate its expression.
> + The set of watchpoints that triggered is recorded in the
> + breakpoint objects themselves (see bp->watchpoint_triggered).
> + If we processed another event first, that other event could
> + clobber this info. */
> + if (ecs->event_thread->stepping_over_watchpoint)
> + return 0;
> +
> + pending = iterate_over_threads (resumed_thread_with_pending_status,
> + NULL);
> + if (pending != NULL)
> + {
> + struct thread_info *tp = ecs->event_thread;
> + struct regcache *regcache;
> +
> + if (debug_infrun)
> + {
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: found resumed threads with "
> + "pending events, saving status\n");
> + }
> +
> + gdb_assert (pending != tp);
> +
> + /* Record the event thread's event for later. */
> + save_waitstatus (tp, &ecs->ws);
> + /* This was cleared early, by handle_inferior_event. Set it
> + so this pending event is considered by
> + do_target_wait. */
> + tp->resumed = 1;
> +
> + gdb_assert (!tp->executing);
> +
> + regcache = get_thread_regcache (tp->ptid);
> + tp->suspend.stop_pc = regcache_read_pc (regcache);
> +
> + if (debug_infrun)
> + {
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: saved stop_pc=%s for %s "
> + "(currently_stepping=%d)\n",
> + paddress (target_gdbarch (),
> + tp->suspend.stop_pc),
> + target_pid_to_str (tp->ptid),
> + currently_stepping (tp));
> + }
> +
> + /* This in-line step-over finished; clear this so we won't
> + start a new one. This is what handle_signal_stop would
> + do, if we returned false. */
> + tp->stepping_over_breakpoint = 0;
> +
> + /* Wake up the event loop again. */
> + mark_async_event_handler (infrun_async_inferior_event_token);
> +
> + prepare_to_wait (ecs);
> + return 1;
> + }
> + }
> +
> + return 0;
> }
>
> /* Come here when the program has stopped with a signal. */
> @@ -4264,7 +5149,8 @@ handle_signal_stop (struct execution_control_state *ecs)
> /* Do we need to clean up the state of a thread that has
> completed a displaced single-step? (Doing so usually affects
> the PC, so do it here, before we set stop_pc.) */
> - finish_step_over (ecs);
> + if (finish_step_over (ecs))
> + return;
>
> /* If we either finished a single-step or hit a breakpoint, but
> the user wanted this thread to be stopped, pretend we got a
> @@ -4722,6 +5608,7 @@ handle_signal_stop (struct execution_control_state *ecs)
> "infrun: signal arrived while stepping over "
> "breakpoint\n");
>
> + clear_step_over_info ();
> insert_hp_step_resume_breakpoint_at_frame (frame);
> ecs->event_thread->step_after_step_resume_breakpoint = 1;
> /* Reset trap_expected to ensure breakpoints are re-inserted. */
> @@ -4755,6 +5642,7 @@ handle_signal_stop (struct execution_control_state *ecs)
> "infrun: signal may take us out of "
> "single-step range\n");
>
> + clear_step_over_info ();
> insert_hp_step_resume_breakpoint_at_frame (frame);
> ecs->event_thread->step_after_step_resume_breakpoint = 1;
> /* Reset trap_expected to ensure breakpoints are re-inserted. */
> @@ -5710,7 +6598,14 @@ switch_back_to_stepped_thread (struct execution_control_state *ecs)
> except the one that needs to move past the breakpoint.
> If a non-event thread has this set, the "incomplete
> step-over" check above should have caught it earlier. */
> - gdb_assert (!tp->control.trap_expected);
> + if (tp->control.trap_expected)
> + {
> + internal_error (__FILE__, __LINE__,
> + "[%s] has inconsistent state: "
> + "trap_expected=%d\n",
> + target_pid_to_str (tp->ptid),
> + tp->control.trap_expected);
> + }
>
> /* Did we find the stepping thread? */
> if (tp->control.step_range_end)
> @@ -5839,6 +6734,7 @@ keep_going_stepped_thread (struct thread_info *tp)
> get_frame_address_space (frame),
> stop_pc);
>
> + tp->resumed = 1;
> resume_ptid = user_visible_resume_ptid (tp->control.stepping_command);
> do_target_resume (resume_ptid, 0, GDB_SIGNAL_0);
> }
> @@ -6270,6 +7166,7 @@ keep_going_pass (struct execution_control_state *ecs)
> struct cleanup *old_cleanups = make_cleanup (resume_cleanups, 0);
>
> gdb_assert (ptid_equal (ecs->event_thread->ptid, inferior_ptid));
> + gdb_assert (!ecs->event_thread->resumed);
>
> /* Save the pc before execution, to compare with pc after stop. */
> ecs->event_thread->prev_pc
> @@ -6292,6 +7189,34 @@ keep_going_pass (struct execution_control_state *ecs)
> discard_cleanups (old_cleanups);
> resume (ecs->event_thread->suspend.stop_signal);
> }
> + else if (step_over_info_valid_p ())
> + {
> + /* Another thread is stepping over a breakpoint in-line. If
> + this thread needs a step-over too, queue the request. In
> + either case, this resume must be deferred for later. */
> + struct thread_info *tp = ecs->event_thread;
> +
> + if (ecs->hit_singlestep_breakpoint
> + || thread_still_needs_step_over (tp))
> + {
> + if (debug_infrun)
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: step-over already in progress: "
> + "step-over for %s deferred\n",
> + target_pid_to_str (tp->ptid));
> + thread_step_over_chain_enqueue (tp);
> + }
> + else
> + {
> + if (debug_infrun)
> + fprintf_unfiltered (gdb_stdlog,
> + "infrun: step-over in progress: "
> + "resume of %s deferred\n",
> + target_pid_to_str (tp->ptid));
> + }
> +
> + discard_cleanups (old_cleanups);
> + }
> else
> {
> struct regcache *regcache = get_current_regcache ();
> @@ -6336,8 +7261,14 @@ keep_going_pass (struct execution_control_state *ecs)
> }
> else if (remove_wps)
> set_step_over_info (NULL, 0, remove_wps);
> - else
> - clear_step_over_info ();
> +
> + /* If we now need to do an in-line step-over, we need to stop
> + all other threads. Note this must be done before
> + insert_breakpoints below, because that removes the breakpoint
> + we're about to step over, otherwise other threads could miss
> + it. */
> + if (step_over_info_valid_p () && non_stop)
> + stop_all_threads ();
>
> /* Stop stepping if inserting breakpoints fails. */
> TRY
> @@ -7705,6 +8636,23 @@ static const struct internalvar_funcs siginfo_funcs =
> NULL
> };
>
> +/* Callback for infrun's target events source. This is marked when a
> + thread has a pending status to process. */
> +
> +static void
> +infrun_async_inferior_event_handler (gdb_client_data data)
> +{
> + /* If the target is closed while this event source is marked, we
> + will reach here without execution, or a target to call
> + target_wait on, which is an error. Instead of tracking whether
> + the target has been popped already, or whether we do have threads
> + with pending statutes, simply ignore the event. */
> + if (!target_is_async_p ())
> + return;
> +
> + inferior_event_handler (INF_REG_EVENT, NULL);
> +}
> +
> void
> _initialize_infrun (void)
> {
> @@ -7712,6 +8660,10 @@ _initialize_infrun (void)
> int numsigs;
> struct cmd_list_element *c;
>
> + /* Register extra event sources in the event loop. */
> + infrun_async_inferior_event_token
> + = create_async_event_handler (infrun_async_inferior_event_handler, NULL);
> +
> add_info ("signals", signals_info, _("\
> What debugger does when program gets various signals.\n\
> Specify a signal as argument to print info on that signal only."));
> diff --git a/gdb/infrun.h b/gdb/infrun.h
> index bfce810..2dc55a9 100644
> --- a/gdb/infrun.h
> +++ b/gdb/infrun.h
> @@ -196,6 +196,9 @@ extern void signal_catch_update (const unsigned int *);
> systems. Use of symbolic signal names is strongly encouraged. */
> enum gdb_signal gdb_signal_from_command (int num);
>
> +/* Enables/disables infrun's async event source in the event loop. */
> +extern void infrun_async (int enable);
> +
> /* The global queue of threads that need to do a step-over operation
> to get past e.g., a breakpoint. */
> extern struct thread_info *step_over_queue_head;
> diff --git a/gdb/target.c b/gdb/target.c
> index 306c21d..e992a35 100644
> --- a/gdb/target.c
> +++ b/gdb/target.c
> @@ -3712,6 +3712,15 @@ maintenance_print_target_stack (char *cmd, int from_tty)
> }
> }
>
> +/* See target.h. */
> +
> +void
> +target_async (int enable)
> +{
> + infrun_async (enable);
> + current_target.to_async (¤t_target, enable);
> +}
> +
> /* Controls if targets can report that they can/are async. This is
> just for maintainers to use when debugging gdb. */
> int target_async_permitted = 1;
> diff --git a/gdb/target.h b/gdb/target.h
> index 66bf91e..3f1a8b0 100644
> --- a/gdb/target.h
> +++ b/gdb/target.h
> @@ -1704,8 +1704,7 @@ extern int target_async_permitted;
> #define target_is_async_p() (current_target.to_is_async_p (¤t_target))
>
> /* Enables/disabled async target events. */
> -#define target_async(ENABLE) \
> - (current_target.to_async (¤t_target, (ENABLE)))
> +extern void target_async (int enable);
>
> #define target_execution_direction() \
> (current_target.to_execution_direction (¤t_target))
> diff --git a/gdb/target/waitstatus.h b/gdb/target/waitstatus.h
> index d4ef3b8..ffaddc1 100644
> --- a/gdb/target/waitstatus.h
> +++ b/gdb/target/waitstatus.h
> @@ -131,7 +131,10 @@ enum target_stop_reason
> TARGET_STOPPED_BY_HW_BREAKPOINT,
>
> /* Stopped by a watchpoint. */
> - TARGET_STOPPED_BY_WATCHPOINT
> + TARGET_STOPPED_BY_WATCHPOINT,
> +
> + /* Stopped by a single step finishing. */
> + TARGET_STOPPED_BY_SINGLE_STEP
> };
>
> /* Prototypes */
> diff --git a/gdb/thread.c b/gdb/thread.c
> index 3e3f419..4dde722 100644
> --- a/gdb/thread.c
> +++ b/gdb/thread.c
> @@ -232,6 +232,7 @@ new_thread (ptid_t ptid)
> /* Nothing to follow yet. */
> tp->pending_follow.kind = TARGET_WAITKIND_SPURIOUS;
> tp->state = THREAD_STOPPED;
> + tp->suspend.waitstatus.kind = TARGET_WAITKIND_IGNORE;
>
> return tp;
> }
> @@ -852,6 +853,28 @@ thread_change_ptid (ptid_t old_ptid, ptid_t new_ptid)
> observer_notify_thread_ptid_changed (old_ptid, new_ptid);
> }
>
> +/* See gdbthread.h. */
> +
> +void
> +set_resumed (ptid_t ptid, int resumed)
> +{
> + struct thread_info *tp;
> + int all = ptid_equal (ptid, minus_one_ptid);
> +
> + if (all || ptid_is_pid (ptid))
> + {
> + for (tp = thread_list; tp; tp = tp->next)
> + if (all || ptid_get_pid (tp->ptid) == ptid_get_pid (ptid))
> + tp->resumed = resumed;
> + }
> + else
> + {
> + tp = find_thread_ptid (ptid);
> + gdb_assert (tp != NULL);
> + tp->resumed = resumed;
> + }
> +}
> +
> /* Helper for set_running, that marks one thread either running or
> stopped. */
>
> --
> 1.9.3
>
--
/dje
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 09/17] Teach non-stop to do in-line step-overs (stop all, step, restart)
2015-04-27 20:17 ` Doug Evans
@ 2015-05-19 18:09 ` Pedro Alves
2015-05-19 18:49 ` Pedro Alves
0 siblings, 1 reply; 62+ messages in thread
From: Pedro Alves @ 2015-05-19 18:09 UTC (permalink / raw)
To: Doug Evans; +Cc: gdb-patches
Hi Doug,
Thanks for your comments, and sorry for the delay in getting
back to answering them.
On 04/27/2015 08:27 PM, Doug Evans wrote:
> > diff --git a/gdb/breakpoint.c b/gdb/breakpoint.c
> > index 31869f1..e23d223 100644
> > --- a/gdb/breakpoint.c
> > +++ b/gdb/breakpoint.c
> > @@ -468,6 +468,8 @@ breakpoints_should_be_inserted_now (void)
> > }
> > else if (target_has_execution)
> > {
> > + struct thread_info *tp;
> > +
> > if (always_inserted_mode)
> > {
> > /* The user wants breakpoints inserted even if all threads
> > @@ -477,6 +479,13 @@ breakpoints_should_be_inserted_now (void)
> >
> > if (threads_are_executing ())
> > return 1;
>
> Not a problem introduced by this patch, but as an fyi, the terminology
> employed here is a bit confusing.
> Why would we want to insert breakpoints into executing threads,
> or when threads are executing? That's what a simple reading of this
> code says is happening. This reading can't be correct of course. :-)
Right. :-)
>
> > +
> > + /* Don't remove breakpoints yet if, even though all threads are
> > + stopped, we still have events to process. */
> > + ALL_NON_EXITED_THREADS (tp)
> > + if (tp->resumed
> > + && tp->suspend.waitstatus_pending_p)
> > + return 1;
>
> Plus, this function is named "breakpoints_should_be_inserted_now"
> but the comment is talking about whether breakpoints should be removed.
Yeah, because if breakpoints are inserted now, and the function returns
false (meaning, they shouldn't be inserted now", they'll be removed.
>
> Can you elaborate on how to interpret the name of this function?
> Guessing at how I'm supposed to interpret what this function is for,
> is a better name
> "breakpoints_should_have_been_inserted_by_now_or_should_remain_inserted"?
> [Not that that's my recommendation :-). Just trying to understand how
> to read this function.]
You got it right, but I'm afraid I lack the English skills to come
up with a better name. "be inserted" is the state we want to reach, not
an action. Maybe "should_breakpoints_be_inserted_now" is little clearer,
but it still doesn't distinguish "state" vs "action". Because
state("be inserted"=false) is clearly "no breakpoints on target",
while action("be inserted"=false) could mean that whatever
breakpoint is already inserted remains inserted.
>
> > }
> > return 0;
> > }
> > diff --git a/gdb/gdbthread.h b/gdb/gdbthread.h
> > index 7052ee1..22c8eb2 100644
> > --- a/gdb/gdbthread.h
> > +++ b/gdb/gdbthread.h
> > @@ -29,6 +29,7 @@ struct symtab;
> > #include "inferior.h"
> > #include "btrace.h"
> > #include "common/vec.h"
> > +#include "target/waitstatus.h"
> >
> > /* Frontend view of the thread state. Possible extensions: stepping,
> > finishing, until(ling),... */
> > @@ -159,6 +160,23 @@ struct thread_suspend_state
> > should be suppressed, the core will take care of clearing this
> > before the target is resumed. */
> > enum gdb_signal stop_signal;
> > +
> > + /* The reason the thread last stopped, if we need to track it
> > + (breakpoint, watchpoint, etc.) */
> > + enum target_stop_reason stop_reason;
> > +
> > + /* The waitstatus for this thread's last event. */
> > + struct target_waitstatus waitstatus;
> > + /* If true WAITSTATUS hasn't been handled yet. */
> > + int waitstatus_pending_p;
> > +
> > + /* Record the pc of the thread the last time it stopped. (This is
> > + not the current thread's PC as that may have changed since the
> > + last stop, e.g., "return" command, or "p $pc = 0xf000"). This
> > + used in coordination with stop_reason and waitstatus_pending_p:
>
> s/used/is used/ ?
Fixed.
> > typedef struct value *value_ptr;
> > @@ -183,6 +201,14 @@ struct thread_info
> > thread is off and running. */
> > int executing;
> >
> > + /* Non-zero if this thread will be/has been resumed. Note that a
> > + thread can be marked both as not-executing and resumed at the
> > + same time. This happens if we try to resume a thread that has a
> > + wait status pending. We shouldn't let the thread run until that
> > + wait status has been processed, but we should not process that
> > + wait status if we didn't try to let the thread run. */
> > + int resumed;
>
> I suspect this will be another source of confusion, but I don't have
> a good suggestion at the moment for how to improve it.
> The "will be" in the comment speaks in future tense, but the name
> "resumed" is past tense. Maybe (though I'd have to spend more time
> reading the code to be sure) it would make sense to have this be
> multi-state: not-resumed, to-be-resumed, and resumed; thus splitting up
> "will be" resumed from "has been" resumed.
I've changed this to:
/* Non-zero if this thread is resumed from infrun's perspective.
Note that a thread can be marked both as not-executing and
resumed at the same time. This happens if we try to resume a
thread that has a wait status pending. We shouldn't let the
thread really run until that wait status has been processed, but
we should not process that wait status if we didn't try to let
the thread run. */
int resumed;
> > + if (tp->suspend.waitstatus_pending_p)
> > + {
> > + if (debug_infrun)
> > + {
> > + char *statstr;
> > +
> > + statstr = target_waitstatus_to_string (&tp->suspend.waitstatus);
> > + fprintf_unfiltered (gdb_stdlog,
> > + "infrun: resume: thread %s has pending wait status %s "
> > + "(currently_stepping=%d).\n",
> > + target_pid_to_str (tp->ptid), statstr,
> > + currently_stepping (tp));
> > + xfree (statstr);
>
> Not something that has to be done with this patch of course,
> but it's nice that we don't have to track the memory of target_pid_to_str;
> IWBN to be able to do the same for target_waitstatus_to_string.
> [In C++ it could just return a string, and we *could* just wait for C++.
> Just a thought.]
I'm just going with the flow you yourself created. ;-)
>
> > + }
> > +
> > + tp->resumed = 1;
> > + /* Avoid confusing the next resume, if the next stop/resume
> > + happens to apply to another thread. */
> > + tp->suspend.stop_signal = GDB_SIGNAL_0;
>
> This is a bit confusing. How will the value of stop_signal in one
> thread affect stop/resume in another thread?
The wonders of copy/paste led to that comment percolating all the way
up here. I first added that comment in 2020b7abd8 elsewhere, when
stop_signal was converted to be per-thread instead of a global.
With all the 7.9 changes to make gdb pass signals to the right
threads, I think this comment no longer makes sense. I'll remove it.
> Plus, the reader is left worried that we just clobbered the signal
> that we need to resume thread tp with. Can you elaborate on what's
> happening here?
Yeah good point. I recall wanting to mention something about
this, but forgot. This is the same situation we already have here
in linux-nat.c's target_resume:
/* If we have a pending wait status for this thread, there is no
point in resuming the process. But first make sure that
linux_nat_wait won't preemptively handle the event - we
should never take this short-circuit if we are going to
leave LP running, since we have skipped resuming all the
other threads. This bit of code needs to be synchronized
with linux_nat_wait. */
if (lp->status && WIFSTOPPED (lp->status))
{
if (!lp->step
&& WSTOPSIG (lp->status)
&& sigismember (&pass_mask, WSTOPSIG (lp->status)))
{
if (debug_linux_nat)
fprintf_unfiltered (gdb_stdlog,
"LLR: Not short circuiting for ignored "
"status 0x%x\n", lp->status);
/* FIXME: What should we do if we are supposed to continue
this thread with a signal? */
gdb_assert (signo == GDB_SIGNAL_0);
signo = gdb_signal_from_host (WSTOPSIG (lp->status));
lp->status = 0;
}
}
Probably the only thing we can do is queue the new signal
to deliver later, somehow. gdbserver's Linux backend handles
that with linux-low.c's lwp->pending_signals list.
Since Mark added that FIXME back in 2000 already, and this
is only enabled on native Linux for now, and this is only
reachable with gdbserver + non-stop on cases we're mishandle
before anyway, we're not losing anything by not handling
this yet.
So I think we should be fine with adding a warning for now:
/* FIXME: What should we do if we are supposed to resume this
thread with a signal? Maybe we should maintain a queue of
pending signals to deliver. */
if (sig != GDB_SIGNAL_0)
{
warning (_("Couldn't deliver signal %s to %s.\n"),
gdb_signal_to_name (sig), target_pid_to_str (tp->ptid));
}
I ran this against the testsuite and nothing exercises this today,
as expected; but I know there's a PR about the assert above.
> > + if (pc != tp->suspend.stop_pc)
> > + {
> > + if (debug_infrun)
> > + fprintf_unfiltered (gdb_stdlog,
> > + "infrun: PC of %s changed. was=%s, now=%s\n",
> > + target_pid_to_str (tp->ptid),
> > + paddress (target_gdbarch (), tp->prev_pc),
> > + paddress (target_gdbarch (), pc));
>
> s/target_gdbarch ()/gdbarch/ ?
Fixed. Below as well.
> > +wait_one (struct target_waitstatus *ws)
> > +{
> > + ptid_t event_ptid;
> > + ptid_t wait_ptid = minus_one_ptid;
> > +
> > + overlay_cache_invalid = 1;
> > +
> > + /* Flush target cache before starting to handle each event.
> > + Target was running and cache could be stale. This is just a
> > + heuristic. Running threads may modify target memory, but we
> > + don't get any event. */
> > + target_dcache_invalidate ();
> > +
> > + if (deprecated_target_wait_hook)
> > + event_ptid = deprecated_target_wait_hook (wait_ptid, ws, 0);
> > + else
> > + event_ptid = target_wait (wait_ptid, ws, 0);
> > +
> > + if (debug_infrun)
> > + print_target_wait_results (wait_ptid, event_ptid, ws);
> > +
> > + if (ws->kind == TARGET_WAITKIND_SYSCALL_ENTRY
> > + || ws->kind == TARGET_WAITKIND_SYSCALL_RETURN)
> > + ws->value.syscall_number = UNKNOWN_SYSCALL;
>
> IWBN to have a comment explaining why we set UNKNOWN_SYSCALL here.
Good question. I honestly don't recall. I ran the testsuite now
with that removed, and nothing broke. Dropped.
> > +/* Stop all threads. */
> > +
> > +static void
> > +stop_all_threads (void)
> > +{
> > + /* We may need multiple passes to discover all threads. */
> > + int pass;
> > + int iterations = 0;
> > + ptid_t entry_ptid;
> > + struct cleanup *old_chain;
> > +
> > + gdb_assert (non_stop);
> > +
> > + if (debug_infrun)
> > + fprintf_unfiltered (gdb_stdlog, "infrun: stop_all_threads\n");
> > +
> > + entry_ptid = inferior_ptid;
> > + old_chain = make_cleanup (switch_to_thread_cleanup, &entry_ptid);
> > +
> > + /* Stop threads in two passes since threads could be spawning as we
> > + go through the first pass. In the second pass, we will stop such
> > + spawned threads. */
> > + for (pass = 0; pass < 2; pass++, iterations++)
>
> Can you rephrase the "Stop threads in two passes ... In the second pass ..."
> comment? What's happening here is that we keep iterating until two passes
> find no new threads (IIUC).
You're right. Here's what I got now:
/* Request threads to stop, and then wait for the stops. Because
threads we already know about can spawn more threads while we're
trying to stop them, and we only learn about new threads when we
update the thread list, do this in a loop, and keep iterating
until two passes find no threads that need to be stopped. */
for (pass = 0; pass < 2; pass++, iterations++)
{
Here's what I'm squashing to the original patch. Let me know
what you think.
gdb/infrun.c | 30 ++++++++++++++++++------------
1 file changed, 18 insertions(+), 12 deletions(-)
diff --git a/gdb/infrun.c b/gdb/infrun.c
index 9f3da09..540ca87 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -2262,8 +2262,16 @@ resume (enum gdb_signal sig)
}
tp->resumed = 1;
- /* Avoid confusing the next resume, if the next stop/resume
- happens to apply to another thread. */
+
+ /* FIXME: What should we do if we are supposed to resume this
+ thread with a signal? Maybe we should maintain a queue of
+ pending signals to deliver. */
+ if (sig != GDB_SIGNAL_0)
+ {
+ warning (_("Couldn't deliver signal %s to %s.\n"),
+ gdb_signal_to_name (sig), target_pid_to_str (tp->ptid));
+ }
+
tp->suspend.stop_signal = GDB_SIGNAL_0;
discard_cleanups (old_cleanups);
@@ -3313,8 +3321,8 @@ do_target_wait (ptid_t ptid, struct target_waitstatus *status, int options)
fprintf_unfiltered (gdb_stdlog,
"infrun: PC of %s changed. was=%s, now=%s\n",
target_pid_to_str (tp->ptid),
- paddress (target_gdbarch (), tp->prev_pc),
- paddress (target_gdbarch (), pc));
+ paddress (gdbarch, tp->prev_pc),
+ paddress (gdbarch, pc));
discard = 1;
}
else if (!breakpoint_inserted_here_p (get_regcache_aspace (regcache), pc))
@@ -3323,7 +3331,7 @@ do_target_wait (ptid_t ptid, struct target_waitstatus *status, int options)
fprintf_unfiltered (gdb_stdlog,
"infrun: previous breakpoint of %s, at %s gone\n",
target_pid_to_str (tp->ptid),
- paddress (target_gdbarch (), pc));
+ paddress (gdbarch, pc));
discard = 1;
}
@@ -4008,10 +4016,6 @@ wait_one (struct target_waitstatus *ws)
if (debug_infrun)
print_target_wait_results (wait_ptid, event_ptid, ws);
- if (ws->kind == TARGET_WAITKIND_SYSCALL_ENTRY
- || ws->kind == TARGET_WAITKIND_SYSCALL_RETURN)
- ws->value.syscall_number = UNKNOWN_SYSCALL;
-
return event_ptid;
}
@@ -4146,9 +4150,11 @@ stop_all_threads (void)
entry_ptid = inferior_ptid;
old_chain = make_cleanup (switch_to_thread_cleanup, &entry_ptid);
- /* Stop threads in two passes since threads could be spawning as we
- go through the first pass. In the second pass, we will stop such
- spawned threads. */
+ /* Request threads to stop, and then wait for the stops. Because
+ threads we already know about can spawn more threads while we're
+ trying to stop them, and we only learn about new threads when we
+ update the thread list, do this in a loop, and keep iterating
+ until two passes find no threads that need to be stopped. */
for (pass = 0; pass < 2; pass++, iterations++)
{
if (debug_infrun)
--
1.9.3
^ permalink raw reply [flat|nested] 62+ messages in thread
* [PATCH v3 14/17] Disable displaced stepping if trying it fails
2015-04-17 10:47 [PATCH v3 00/23] All-stop on top of non-stop Pedro Alves
` (13 preceding siblings ...)
2015-04-17 10:52 ` [PATCH v3 09/17] Teach non-stop to do in-line step-overs (stop all, step, restart) Pedro Alves
@ 2015-04-17 10:56 ` Pedro Alves
2015-04-17 11:06 ` [PATCH v3 16/17] S/390: displaced stepping and PC-relative RIL-b/RIL-c instructions Pedro Alves
` (3 subsequent siblings)
18 siblings, 0 replies; 62+ messages in thread
From: Pedro Alves @ 2015-04-17 10:56 UTC (permalink / raw)
To: gdb-patches
Running the testsuite with "maint set target-non-stop on" shows:
(gdb) PASS: gdb.base/valgrind-infcall.exp: continue #98 (false warning)
continue
Continuing.
dl_main (phdr=<optimized out>..., auxv=<optimized out>) at rtld.c:2302
2302 LIBC_PROBE (init_complete, 2, LM_ID_BASE, r);
Cannot access memory at address 0x400532
(gdb) PASS: gdb.base/valgrind-infcall.exp: continue #99 (false warning)
p gdb_test_infcall ()
$1 = 1
(gdb) FAIL: gdb.base/valgrind-infcall.exp: p gdb_test_infcall ()
Even though that was a native GNU/Linux test run, this test spawns
Valgrind and connects to it with "target remote". The error above is
actually orthogonal to target-non-stop. The real issue is that that
enables displaced stepping, and displaced stepping doesn't work with
Valgrind, because we can't write to the inferior memory (thus can't
copy the instruction to the scratch pad area).
I'm sure there will be other targets with the same issue, so trying to
identify Valgrind wouldn't be sufficient. The fix is to try setting
up the displaced step anyway. If we get a MEMORY_ERROR, we disable
displaced stepping for that inferior, and fall back to doing an
in-line step-over. If "set displaced-stepping" is "on" (as opposed to
"auto), GDB warns displaced stepping failed ("on" is mainly useful for
the testsuite, not for users).
Tested on x86_64 Fedora 20.
gdb/ChangeLog:
2015-04-17 Pedro Alves <palves@redhat.com>
* inferior.h (struct inferior) <displaced_stepping_failed>: New
field.
* infrun.c (use_displaced_stepping_now_p): New parameter 'inf'.
Return false if dispaced stepping failed before.
(resume): Pass the current inferior to
use_displaced_stepping_now_p. Wrap displaced_step_prepare in
TRY/CATCH. If we get a MEMORY_ERROR, set the inferior's
displaced_stepping_failed flag, and fall back to an in-line
step-over.
gdb/testsuite/ChangeLog:
2015-04-17 Pedro Alves <palves@redhat.com>
* gdb.base/valgrind-disp-step.c: New file.
* gdb.base/valgrind-disp-step.exp: New file.
v3:
- displaced_stepping_failed flag moved from "struct inferior" to
"struct displaced_step_inferior_state", and renamed "failed_before".
- use_displaced_stepping_now_p in v2 eliminated. Simply change the
prototype of use_displaced_stepping to take a thread pointer
instead.
- TRY/CATCH around displaced_step_prepare moved to separate
function.
v2:
- warn if displaced stepping fails, and "set displaced-stepping" is
"on" (but not "auto").
---
gdb/infrun.c | 112 ++++++++++++++++++----
gdb/testsuite/gdb.base/valgrind-disp-step.c | 32 +++++++
gdb/testsuite/gdb.base/valgrind-disp-step.exp | 131 ++++++++++++++++++++++++++
3 files changed, 255 insertions(+), 20 deletions(-)
create mode 100644 gdb/testsuite/gdb.base/valgrind-disp-step.c
create mode 100644 gdb/testsuite/gdb.base/valgrind-disp-step.exp
diff --git a/gdb/infrun.c b/gdb/infrun.c
index f6d3e07..be186a7 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -1470,6 +1470,10 @@ struct displaced_step_inferior_state
/* The process this displaced step state refers to. */
int pid;
+ /* True if preparing a displaced step ever failed. If so, we won't
+ try displaced stepping for this inferior again. */
+ int failed_before;
+
/* If this is not null_ptid, this is the thread carrying out a
displaced single-step in process PID. This thread's state will
require fixing up once it has completed its step. */
@@ -1640,16 +1644,24 @@ show_can_use_displaced_stepping (struct ui_file *file, int from_tty,
}
/* Return non-zero if displaced stepping can/should be used to step
- over breakpoints. */
+ over breakpoints of thread TP. */
static int
-use_displaced_stepping (struct gdbarch *gdbarch)
+use_displaced_stepping (struct thread_info *tp)
{
+ struct regcache *regcache = get_thread_regcache (tp->ptid);
+ struct gdbarch *gdbarch = get_regcache_arch (regcache);
+ struct displaced_step_inferior_state *displaced_state;
+
+ displaced_state = get_displaced_stepping_state (ptid_get_pid (tp->ptid));
+
return (((can_use_displaced_stepping == AUTO_BOOLEAN_AUTO
&& target_is_non_stop_p ())
|| can_use_displaced_stepping == AUTO_BOOLEAN_TRUE)
&& gdbarch_displaced_step_copy_insn_p (gdbarch)
- && find_record_target () == NULL);
+ && find_record_target () == NULL
+ && (displaced_state == NULL
+ || !displaced_state->failed_before));
}
/* Clean out any stray displaced stepping state. */
@@ -1703,7 +1715,7 @@ displaced_step_dump_bytes (struct ui_file *file,
Returns 1 if preparing was successful -- this thread is going to be
stepped now; or 0 if displaced stepping this thread got queued. */
static int
-displaced_step_prepare (ptid_t ptid)
+displaced_step_prepare_throw (ptid_t ptid)
{
struct cleanup *old_cleanups, *ignore_cleanups;
struct thread_info *tp = find_thread_ptid (ptid);
@@ -1813,6 +1825,50 @@ displaced_step_prepare (ptid_t ptid)
return 1;
}
+/* Wrapper for displaced_step_prepare_throw that disabled further
+ attempts at displaced stepping if we get a memory error. */
+
+static int
+displaced_step_prepare (ptid_t ptid)
+{
+ int prepared = -1;
+
+ TRY
+ {
+ prepared = displaced_step_prepare_throw (ptid);
+ }
+ CATCH (ex, RETURN_MASK_ERROR)
+ {
+ struct displaced_step_inferior_state *displaced_state;
+
+ if (ex.error != MEMORY_ERROR)
+ throw_exception (ex);
+
+ if (debug_infrun)
+ {
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: disabling displaced stepping: %s\n",
+ ex.message);
+ }
+
+ /* Be verbose if "set displaced-stepping" is "on", silent if
+ "auto". */
+ if (can_use_displaced_stepping == AUTO_BOOLEAN_TRUE)
+ {
+ warning (_("disabling displaced stepping: %s\n"),
+ ex.message);
+ }
+
+ /* Disable further displaced stepping attempts. */
+ displaced_state
+ = get_displaced_stepping_state (ptid_get_pid (ptid));
+ displaced_state->failed_before = 1;
+ }
+ END_CATCH
+
+ return prepared;
+}
+
static void
write_memory_ptid (ptid_t ptid, CORE_ADDR memaddr,
const gdb_byte *myaddr, int len)
@@ -1943,6 +1999,7 @@ static void keep_going_pass (struct execution_control_state *ecs);
static void prepare_to_wait (struct execution_control_state *ecs);
static int keep_going_stepped_thread (struct thread_info *tp);
static int thread_still_needs_step_over (struct thread_info *tp);
+static void stop_all_threads (void);
/* Are there any pending step-over requests? If so, run one now and
return true. Otherwise, return false. */
@@ -1963,8 +2020,6 @@ start_step_over (void)
struct execution_control_state *ecs = &ecss;
enum step_over_what step_what;
int must_be_in_line;
- struct regcache *regcache = get_thread_regcache (tp->ptid);
- struct gdbarch *gdbarch = get_regcache_arch (regcache);
next = thread_step_over_chain_next (tp);
@@ -1976,7 +2031,7 @@ start_step_over (void)
step_what = thread_still_needs_step_over (tp);
must_be_in_line = ((step_what & STEP_OVER_WATCHPOINT)
|| ((step_what & STEP_OVER_BREAKPOINT)
- && !use_displaced_stepping (gdbarch)));
+ && !use_displaced_stepping (tp)));
/* We currently stop all threads of all processes to step-over
in-line. If we need to start a new in-line step-over, let
@@ -2412,15 +2467,15 @@ resume (enum gdb_signal sig)
We can't use displaced stepping when we are waiting for vfork_done
event, displaced stepping breaks the vfork child similarly as single
step software breakpoint. */
- if (use_displaced_stepping (gdbarch)
- && tp->control.trap_expected
+ if (tp->control.trap_expected
+ && use_displaced_stepping (tp)
&& !step_over_info_valid_p ()
&& sig == GDB_SIGNAL_0
&& !current_inferior ()->waiting_for_vfork_done)
{
- struct displaced_step_inferior_state *displaced;
+ int prepared = displaced_step_prepare (inferior_ptid);
- if (!displaced_step_prepare (inferior_ptid))
+ if (prepared == 0)
{
if (debug_infrun)
fprintf_unfiltered (gdb_stdlog,
@@ -2430,14 +2485,32 @@ resume (enum gdb_signal sig)
discard_cleanups (old_cleanups);
return;
}
+ else if (prepared < 0)
+ {
+ /* Fallback to stepping over the breakpoint in-line. */
- /* Update pc to reflect the new address from which we will execute
- instructions due to displaced stepping. */
- pc = regcache_read_pc (get_thread_regcache (inferior_ptid));
+ if (target_is_non_stop_p ())
+ stop_all_threads ();
- displaced = get_displaced_stepping_state (ptid_get_pid (inferior_ptid));
- step = gdbarch_displaced_step_hw_singlestep (gdbarch,
- displaced->step_closure);
+ set_step_over_info (get_regcache_aspace (regcache),
+ regcache_read_pc (regcache), 0);
+
+ step = maybe_software_singlestep (gdbarch, pc);
+
+ insert_breakpoints ();
+ }
+ else if (prepared > 0)
+ {
+ struct displaced_step_inferior_state *displaced;
+
+ /* Update pc to reflect the new address from which we will
+ execute instructions due to displaced stepping. */
+ pc = regcache_read_pc (get_thread_regcache (inferior_ptid));
+
+ displaced = get_displaced_stepping_state (ptid_get_pid (inferior_ptid));
+ step = gdbarch_displaced_step_hw_singlestep (gdbarch,
+ displaced->step_closure);
+ }
}
/* Do we need to do it the hard way, w/temp breakpoints? */
@@ -2565,8 +2638,8 @@ resume (enum gdb_signal sig)
}
if (debug_displaced
- && use_displaced_stepping (gdbarch)
&& tp->control.trap_expected
+ && use_displaced_stepping (tp)
&& !step_over_info_valid_p ())
{
struct regcache *resume_regcache = get_thread_regcache (tp->ptid);
@@ -7325,8 +7398,7 @@ keep_going_pass (struct execution_control_state *ecs)
watchpoint. The instruction copied to the scratch pad would
still trigger the watchpoint. */
if (remove_bp
- && (remove_wps
- || !use_displaced_stepping (get_regcache_arch (regcache))))
+ && (remove_wps || !use_displaced_stepping (ecs->event_thread)))
{
set_step_over_info (get_regcache_aspace (regcache),
regcache_read_pc (regcache), remove_wps);
diff --git a/gdb/testsuite/gdb.base/valgrind-disp-step.c b/gdb/testsuite/gdb.base/valgrind-disp-step.c
new file mode 100644
index 0000000..baba74e
--- /dev/null
+++ b/gdb/testsuite/gdb.base/valgrind-disp-step.c
@@ -0,0 +1,32 @@
+/* This testcase is part of GDB, the GNU debugger.
+
+ Copyright 2015 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/>. */
+
+static int
+foo (void)
+{
+}
+
+int
+main (void)
+{
+ foo (); /* stop 0 */
+ foo (); /* stop 1 */
+ foo (); /* stop 2 */
+ foo (); /* stop 3 */
+ foo (); /* stop 4 */
+ return 0;
+}
diff --git a/gdb/testsuite/gdb.base/valgrind-disp-step.exp b/gdb/testsuite/gdb.base/valgrind-disp-step.exp
new file mode 100644
index 0000000..7e6803b
--- /dev/null
+++ b/gdb/testsuite/gdb.base/valgrind-disp-step.exp
@@ -0,0 +1,131 @@
+# Copyright 2012-2015 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/>.
+
+# Step over breakpoints with displaced stepping on, against Valgrind.
+# We can't really use displaced stepping with Valgrind, so what this
+# really tests is that GDB falls back to in-line stepping
+# automatically instead of getting stuck or crashing.
+
+if [is_remote target] {
+ # The test always runs locally.
+ return 0
+}
+
+standard_testfile .c
+if {[build_executable "failed to build" $testfile $srcfile {debug}] == -1} {
+ return -1
+}
+
+set test "spawn valgrind"
+set cmd "valgrind --vgdb-error=0 $binfile"
+set res [remote_spawn host $cmd]
+if { $res < 0 || $res == "" } {
+ verbose -log "Spawning $cmd failed."
+ unsupported $test
+ return -1
+}
+pass $test
+# Declare GDB now as running.
+set gdb_spawn_id $res
+
+# GDB started by vgdb stops already after the startup is executed, like with
+# non-extended gdbserver. It is also not correct to run/attach the inferior.
+set use_gdb_stub 1
+
+set test "valgrind started"
+# The trailing '.' differs for different memcheck versions.
+gdb_test_multiple "" $test {
+ -re "Memcheck, a memory error detector\\.?\r\n" {
+ pass $test
+ }
+ -re "valgrind: failed to start tool 'memcheck' for platform '.*': No such file or directory" {
+ unsupported $test
+ return -1
+ }
+ -re "valgrind: wrong ELF executable class" {
+ unsupported $test
+ return -1
+ }
+ -re "command not found" {
+ # The spawn succeeded, but then valgrind was not found - e.g. if
+ # we spawned SSH to a remote system.
+ unsupported $test
+ return -1
+ }
+ -re "valgrind: Bad option.*--vgdb-error=0" {
+ # valgrind is not >= 3.7.0.
+ unsupported $test
+ return -1
+ }
+}
+
+set test "vgdb prompt"
+# The trailing '.' differs for different memcheck versions.
+gdb_test_multiple "" $test {
+ -re " (target remote | \[^\r\n\]*/vgdb \[^\r\n\]*)\r\n" {
+ set vgdbcmd $expect_out(1,string)
+ pass $test
+ }
+}
+
+# Do not kill valgrind.
+set valgrind_pid [exp_pid -i [board_info host fileid]]
+unset gdb_spawn_id
+set board [host_info name]
+unset_board_info fileid
+
+clean_restart $testfile
+
+gdb_test "$vgdbcmd" " in \\.?_start .*" "target remote for vgdb"
+
+gdb_test "monitor v.set gdb_output" "valgrind output will go to gdb.*"
+
+gdb_test_no_output "set displaced-stepping off"
+gdb_breakpoint "main" "breakpoint at main"
+gdb_test "continue" " stop 0 .*" "continue to main"
+delete_breakpoints
+
+set curr_stop 0
+foreach displaced { "off" "on" } {
+ with_test_prefix "displaced $displaced" {
+
+ gdb_test_no_output "set displaced-stepping $displaced"
+
+ foreach go { "once" "twice" } {
+ with_test_prefix $go {
+ gdb_test "break" "Breakpoint .* at .*" "set breakpoint"
+
+ # Whether we should see a warning.
+ set should_warn [expr {$go == "once" && $displaced == "on"}]
+
+ incr curr_stop
+
+ set msg "step over breakpoint"
+ set pattern " stop $curr_stop .*$gdb_prompt $"
+ gdb_test_multiple "next" $msg {
+ -re "warning: disabling displaced stepping.*$pattern" {
+ gdb_assert $should_warn $msg
+ }
+ -re "$pattern" {
+ gdb_assert !$should_warn $msg
+ }
+ }
+ }
+ }
+ }
+}
+
+# Only if valgrind got stuck.
+remote_exec host "kill -9 ${valgrind_pid}"
--
1.9.3
^ permalink raw reply [flat|nested] 62+ messages in thread* [PATCH v3 16/17] S/390: displaced stepping and PC-relative RIL-b/RIL-c instructions
2015-04-17 10:47 [PATCH v3 00/23] All-stop on top of non-stop Pedro Alves
` (14 preceding siblings ...)
2015-04-17 10:56 ` [PATCH v3 14/17] Disable displaced stepping if trying it fails Pedro Alves
@ 2015-04-17 11:06 ` Pedro Alves
2015-04-17 11:38 ` [PATCH v3 10/17] Implement all-stop on top of a target running non-stop mode Pedro Alves
` (2 subsequent siblings)
18 siblings, 0 replies; 62+ messages in thread
From: Pedro Alves @ 2015-04-17 11:06 UTC (permalink / raw)
To: gdb-patches
This adds displaced stepping support for the General-Instruction
Extension Facility instructions, which have a PC-relative displacement
(RIL-b/RIL-c). We already handle RIL branches, but not others.
Currently, displaced stepping a breakpoint put on any of these
instructions results in the inferior crashing when or after the
instruction is executed out-of-line in the scratch pad.
This patch takes the easy route of patching the displacement in the
copy of the instruction in the scratch pad. As the displacement is a
signed 32-bit field, it's possible that the stratch pad ends too far
that the needed displacement doesn't fit in the adjusted instruction,
as e.g., if stepping over a breakpoint in a shared library (the
scratch pad is around the main program's entry point). That case is
detected and GDB falls back to stepping over the breakpoint in-line
(which involves pausing all threads momentarily).
(We could probably do something smarter, but I don't plan on doing it
myself. This was already sufficient to get "maint set target-non-stop
on" working regression free on S/390.)
Tested on S/390 RHEL 7.1, where it fixes a few hundred FAILs when
testing with displaced stepping force-enabled, with the end result
being no regressions compared to a test run that doesn't force
displaced stepping. Fixes the non-stop tests compared to mainline
too; most are crashing due to this on the machine I run tests on.
gdb/ChangeLog:
2015-04-17 Pedro Alves <palves@redhat.com>
* s390-linux-tdep.c (is_non_branch_ril)
(s390_displaced_step_copy_insn): New functions.
(s390_displaced_step_fixup): Update comment.
(s390_gdbarch_init): Install s390_displaced_step_copy_insn as
gdbarch_displaced_step_copy_insn hook.
v3:
- No changes.
---
gdb/s390-linux-tdep.c | 115 ++++++++++++++++++++++++++++++++++++++++++++++++--
1 file changed, 112 insertions(+), 3 deletions(-)
diff --git a/gdb/s390-linux-tdep.c b/gdb/s390-linux-tdep.c
index b83a264..41790a4 100644
--- a/gdb/s390-linux-tdep.c
+++ b/gdb/s390-linux-tdep.c
@@ -1529,6 +1529,116 @@ s390_in_function_epilogue_p (struct gdbarch *gdbarch, CORE_ADDR pc)
/* Displaced stepping. */
+/* Return true if INSN is a non-branch RIL-b or RIL-c format
+ instruction. */
+
+static int
+is_non_branch_ril (gdb_byte *insn)
+{
+ gdb_byte op1 = insn[0];
+
+ if (op1 == 0xc4)
+ {
+ gdb_byte op2 = insn[1] & 0x0f;
+
+ switch (op2)
+ {
+ case 0x02: /* llhrl */
+ case 0x04: /* lghrl */
+ case 0x05: /* lhrl */
+ case 0x06: /* llghrl */
+ case 0x07: /* sthrl */
+ case 0x08: /* lgrl */
+ case 0x0b: /* stgrl */
+ case 0x0c: /* lgfrl */
+ case 0x0d: /* lrl */
+ case 0x0e: /* llgfrl */
+ case 0x0f: /* strl */
+ return 1;
+ }
+ }
+ else if (op1 == 0xc6)
+ {
+ gdb_byte op2 = insn[1] & 0x0f;
+
+ switch (op2)
+ {
+ case 0x00: /* exrl */
+ case 0x02: /* pfdrl */
+ case 0x04: /* cghrl */
+ case 0x05: /* chrl */
+ case 0x06: /* clghrl */
+ case 0x07: /* clhrl */
+ case 0x08: /* cgrl */
+ case 0x0a: /* clgrl */
+ case 0x0c: /* cgfrl */
+ case 0x0d: /* crl */
+ case 0x0e: /* clgfrl */
+ case 0x0f: /* clrl */
+ return 1;
+ }
+ }
+
+ return 0;
+}
+
+/* Implementation of gdbarch_displaced_step_copy_insn. */
+
+static struct displaced_step_closure *
+s390_displaced_step_copy_insn (struct gdbarch *gdbarch,
+ CORE_ADDR from, CORE_ADDR to,
+ struct regcache *regs)
+{
+ size_t len = gdbarch_max_insn_length (gdbarch);
+ gdb_byte *buf = xmalloc (len);
+ struct cleanup *old_chain = make_cleanup (xfree, buf);
+
+ read_memory (from, buf, len);
+
+ /* Adjust the displacement field of PC-relative RIL instructions,
+ except branches. The latter are handled in the fixup hook. */
+ if (is_non_branch_ril (buf))
+ {
+ LONGEST offset;
+
+ offset = extract_signed_integer (buf + 2, 4, BFD_ENDIAN_BIG);
+ offset = (from - to + offset * 2) / 2;
+
+ /* If the instruction is too far from the jump pad, punt. This
+ will usually happen with instructions in shared libraries.
+ We could probably support these by rewriting them to be
+ absolute or fully emulating them. */
+ if (offset < INT32_MIN || offset > INT32_MAX)
+ {
+ /* Let the core fall back to stepping over the breakpoint
+ in-line. */
+ if (debug_displaced)
+ {
+ fprintf_unfiltered (gdb_stdlog,
+ "displaced: can't displaced step "
+ "RIL instruction: offset %s out of range\n",
+ plongest (offset));
+ }
+ do_cleanups (old_chain);
+ return NULL;
+ }
+
+ store_signed_integer (buf + 2, 4, BFD_ENDIAN_BIG, offset);
+ }
+
+ write_memory (to, buf, len);
+
+ if (debug_displaced)
+ {
+ fprintf_unfiltered (gdb_stdlog, "displaced: copy %s->%s: ",
+ paddress (gdbarch, from), paddress (gdbarch, to));
+ displaced_step_dump_bytes (gdb_stdlog, buf, len);
+ }
+
+ discard_cleanups (old_chain);
+ return (struct displaced_step_closure *) buf;
+}
+
/* Fix up the state of registers and memory after having single-stepped
a displaced instruction. */
static void
@@ -1537,8 +1647,7 @@ s390_displaced_step_fixup (struct gdbarch *gdbarch,
CORE_ADDR from, CORE_ADDR to,
struct regcache *regs)
{
- /* Since we use simple_displaced_step_copy_insn, our closure is a
- copy of the instruction. */
+ /* Our closure is a copy of the instruction. */
gdb_byte *insn = (gdb_byte *) closure;
static int s390_instrlen[] = { 2, 4, 4, 6 };
int insnlen = s390_instrlen[insn[0] >> 6];
@@ -3291,7 +3400,7 @@ s390_gdbarch_init (struct gdbarch_info info, struct gdbarch_list *arches)
/* Displaced stepping. */
set_gdbarch_displaced_step_copy_insn (gdbarch,
- simple_displaced_step_copy_insn);
+ s390_displaced_step_copy_insn);
set_gdbarch_displaced_step_fixup (gdbarch, s390_displaced_step_fixup);
set_gdbarch_displaced_step_free_closure (gdbarch,
simple_displaced_step_free_closure);
--
1.9.3
^ permalink raw reply [flat|nested] 62+ messages in thread* [PATCH v3 10/17] Implement all-stop on top of a target running non-stop mode
2015-04-17 10:47 [PATCH v3 00/23] All-stop on top of non-stop Pedro Alves
` (15 preceding siblings ...)
2015-04-17 11:06 ` [PATCH v3 16/17] S/390: displaced stepping and PC-relative RIL-b/RIL-c instructions Pedro Alves
@ 2015-04-17 11:38 ` Pedro Alves
2015-04-21 11:09 ` Yao Qi
2015-04-20 12:02 ` [PATCH v3 00/23] All-stop on top of non-stop Yao Qi
2015-04-20 17:35 ` Simon Marchi
18 siblings, 1 reply; 62+ messages in thread
From: Pedro Alves @ 2015-04-17 11:38 UTC (permalink / raw)
To: gdb-patches
This finally implements user-visible all-stop mode running with the
target_ops backend always in non-stop mode. This is a stepping stone
towards finer-grained control of threads, being able to do interesting
things like thread groups, associating groups with breakpoints, etc.
From the user's perspective, all-stop mode is really just a special
case of being able to stop and resume specific sets of threads, so it
makes sense to do this step first.
With this, even in all-stop, the target is no longer in charge of
stopping all threads before reporting an event to the core -- the core
takes care of it when it sees fit. For example, when "next"- or
"step"-ing, we can avoid stopping and resuming all threads at each
internal single-step, and instead only stop all threads when we're
about to present the stop to the user.
The implementation is almost straight forward, as the heavy lifting
has been done already in previous patches. Basically, we replace
checks for "set non-stop on/off" (the non_stop global), with calls to
a new target_is_non_stop_p function. In a few places, if "set
non-stop off", we stop all threads explicitly, and in a few other
places we resume all threads explicitly, making use of existing
methods that were added for teaching non-stop to step over breakpoints
without displaced stepping.
This adds a new "maint set target-non-stop on/off/auto" knob that
allows both disabling the feature if we find problems, and
force-enable it for development (useful when teaching a target about
this. The default is "auto", which means the feature is enabled if a
new target method says it should be enabled. The patch implements the
method in linux-nat.c, just for illustration, because it still returns
false. We'll need a few follow up fixes before turning it on by
default. This is a separate target method from indicating regular
non-stop support, because e.g., while e.g., native linux-nat.c is
close to regression free with all-stop-non-stop (with following
patches will fixing the remaining regressions), remote.c+gdbserver
will still need more fixing, even though it supports "set non-stop
on".
Tested on x86_64 Fedora 20, native, with and without "set displaced
off", and with and without "maint set target-non-stop on"; and also
against gdbserver.
gdb/ChangeLog:
2015-04-17 Pedro Alves <palves@redhat.com>
* NEWS: Mention "maint set/show target-non-stop".
* breakpoint.c (update_global_location_list): Check
target_is_non_stop_p instead of non_stop.
* infcmd.c (attach_command_post_wait, attach_command): Likewise.
* infrun.c (show_can_use_displaced_stepping)
(can_use_displaced_stepping_p, start_step_over_inferior):
Likewise.
(resume): Always resume a single thread if the target is in
non-stop mode.
(proceed): Check target_is_non_stop_p instead of non_stop. If in
all-mode but the target is always in non-stop mode, start all the
other threads that are implicitly resumed too.
(for_each_just_stopped_thread, fetch_inferior_event)
(adjust_pc_after_break, stop_all_threads): Check
target_is_non_stop_p instead of non_stop.
(handle_inferior_event): Likewise. Handle detach-fork in all-stop
with the target always in non-stop mode.
(handle_signal_stop) <random signal>: If we get a signal while
stepping over a breakpoint, and the target is always in non-stop
mode, restart all threads.
(switch_back_to_stepped_thread): Check target_is_non_stop_p
instead of non_stop.
(keep_going_stepped_thread): Always resume a single thread if the
target is in non-stop mode.
(stop_waiting): If in all-stop mode, and the target is in non-stop
mode, stop all threads.
(keep_going_pass): Likewise, when starting a new in-line step-over
sequence.
* linux-nat.c (get_pending_status, select_event_lwp)
(linux_nat_filter_event, linux_nat_wait_1, linux_nat_wait): Check
target_is_non_stop_p instead of non_stop.
(linux_nat_always_non_stop_p): New function.
(linux_nat_stop): Check target_is_non_stop_p instead of non_stop.
(linux_nat_add_target): Install linux_nat_always_non_stop_p.
* target-delegates.c: Regenerate.
* target.c (target_is_non_stop_p): New function.
(target_non_stop_enabled, target_non_stop_enabled_1): New globals.
(maint_set_target_non_stop_command)
(maint_show_target_non_stop_command): New functions.
(_initilize_target): Install "maint set/show target-non-stop"
commands.
* target.h (struct target_ops) <to_always_non_stop_p>: New field.
(target_non_stop_enabled): New declaration.
(target_is_non_stop_p): New declaration.
gdb/doc/ChangeLog:
2015-04-07 Pedro Alves <palves@redhat.com>
* gdb.texinfo (Maintenance Commands): Document "maint set/show
target-non-stop".
v3:
Adjustments as per previous patches:
- Change 'resumed' instead of 'executing' in 'proceed' to avoid
resuming threads with pending statuses.
- Adjust to use thread_is_in_step_over_chain.
v2:
- Documentation adjusted per Eli's review. NEWS entry added.
- `proceed' change simplified, thanks to fixes in v2 of previous
patches in the series.
---
gdb/NEWS | 6 +++
gdb/breakpoint.c | 2 +-
gdb/doc/gdb.texinfo | 24 +++++++++++
gdb/infcmd.c | 4 +-
gdb/infrun.c | 110 ++++++++++++++++++++++++++++++++++++++++---------
gdb/linux-nat.c | 25 +++++++----
gdb/target-delegates.c | 31 ++++++++++++++
gdb/target.c | 62 ++++++++++++++++++++++++++++
gdb/target.h | 13 ++++++
9 files changed, 247 insertions(+), 30 deletions(-)
diff --git a/gdb/NEWS b/gdb/NEWS
index 62cbdcb..7b412e6 100644
--- a/gdb/NEWS
+++ b/gdb/NEWS
@@ -60,6 +60,12 @@ maint print symbol-cache-statistics
maint flush-symbol-cache
Flush the contents of the symbol cache.
+maint set target-non-stop (on|off|auto)
+maint show target-non-stop
+ Control whether GDB targets always operate in non-stop mode even if
+ "set non-stop" is "off". The default is "auto", meaning non-stop
+ mode is enabled if supported by the target.
+
record btrace bts
record bts
Start branch trace recording using Branch Trace Store (BTS) format.
diff --git a/gdb/breakpoint.c b/gdb/breakpoint.c
index e23d223..96d2a14 100644
--- a/gdb/breakpoint.c
+++ b/gdb/breakpoint.c
@@ -12345,7 +12345,7 @@ update_global_location_list (enum ugll_insert_mode insert_mode)
if (!found_object)
{
- if (removed && non_stop
+ if (removed && target_is_non_stop_p ()
&& need_moribund_for_location_type (old_loc))
{
/* This location was removed from the target. In
diff --git a/gdb/doc/gdb.texinfo b/gdb/doc/gdb.texinfo
index 0410702..97f7f78 100644
--- a/gdb/doc/gdb.texinfo
+++ b/gdb/doc/gdb.texinfo
@@ -34192,6 +34192,30 @@ asynchronous mode (@pxref{Background Execution}). Normally the
default is asynchronous, if it is available; but this can be changed
to more easily debug problems occurring only in synchronous mode.
+@kindex maint set target-non-stop @var{mode} [on|off|auto]
+@kindex maint show target-non-stop
+@item maint set target-non-stop
+@itemx maint show target-non-stop
+
+This controls whether @value{GDBN} targets always operate in non-stop
+mode even if @code{set non-stop} is @code{off} (@pxref{Non-Stop
+Mode}). The default is @code{auto}, meaning non-stop mode is enabled
+if supported by the target.
+
+@table @code
+@item maint set target-non-stop auto
+This is the default mode. @value{GDBN} controls the target in
+non-stop mode if the target supports it.
+
+@item maint set target-non-stop on
+@value{GDBN} controls the target in non-stop mode even if the target
+does not indicate support.
+
+@item maint set target-non-stop off
+@value{GDBN} does not control the target in non-stop mode even if the
+target supports it.
+@end table
+
@kindex maint set per-command
@kindex maint show per-command
@item maint set per-command
diff --git a/gdb/infcmd.c b/gdb/infcmd.c
index 7e2484b..3505b1b 100644
--- a/gdb/infcmd.c
+++ b/gdb/infcmd.c
@@ -2517,7 +2517,7 @@ attach_command_post_wait (char *args, int from_tty, int async_exec)
selected thread is stopped, others may still be executing.
Be sure to explicitly stop all threads of the process. This
should have no effect on already stopped threads. */
- if (non_stop)
+ if (target_is_non_stop_p ())
target_stop (pid_to_ptid (inferior->pid));
/* Tell the user/frontend where we're stopped. */
@@ -2622,7 +2622,7 @@ attach_command (char *args, int from_tty)
init_wait_for_inferior ();
clear_proceed_status (0);
- if (non_stop)
+ if (target_is_non_stop_p ())
{
/* If we find that the current thread isn't stopped, explicitly
do so now, because we're going to install breakpoints and
diff --git a/gdb/infrun.c b/gdb/infrun.c
index 8859b9f..f6d3e07 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -1632,7 +1632,7 @@ show_can_use_displaced_stepping (struct ui_file *file, int from_tty,
fprintf_filtered (file,
_("Debugger's willingness to use displaced stepping "
"to step over breakpoints is %s (currently %s).\n"),
- value, non_stop ? "on" : "off");
+ value, target_is_non_stop_p () ? "on" : "off");
else
fprintf_filtered (file,
_("Debugger's willingness to use displaced stepping "
@@ -1645,7 +1645,8 @@ show_can_use_displaced_stepping (struct ui_file *file, int from_tty,
static int
use_displaced_stepping (struct gdbarch *gdbarch)
{
- return (((can_use_displaced_stepping == AUTO_BOOLEAN_AUTO && non_stop)
+ return (((can_use_displaced_stepping == AUTO_BOOLEAN_AUTO
+ && target_is_non_stop_p ())
|| can_use_displaced_stepping == AUTO_BOOLEAN_TRUE)
&& gdbarch_displaced_step_copy_insn_p (gdbarch)
&& find_record_target () == NULL);
@@ -2016,7 +2017,7 @@ start_step_over (void)
wouldn't be able to resume anything else until the target
stops again. In non-stop, the resume always resumes only TP,
so it's OK to let the thread resume freely. */
- if (!non_stop && !step_what)
+ if (!target_is_non_stop_p () && !step_what)
continue;
switch_to_thread (tp->ptid);
@@ -2035,7 +2036,7 @@ start_step_over (void)
return 1;
}
- if (!non_stop)
+ if (!target_is_non_stop_p ())
{
/* On all-stop, shouldn't have resumed unless we needed a
step over. */
@@ -2383,7 +2384,10 @@ resume (enum gdb_signal sig)
insert_single_step_breakpoint (gdbarch, aspace, pc);
insert_breakpoints ();
- resume_ptid = user_visible_resume_ptid (user_step);
+ if (target_is_non_stop_p ())
+ resume_ptid = inferior_ptid;
+ else
+ resume_ptid = user_visible_resume_ptid (user_step);
do_target_resume (resume_ptid, 0, GDB_SIGNAL_0);
discard_cleanups (old_cleanups);
tp->resumed = 1;
@@ -2498,8 +2502,14 @@ resume (enum gdb_signal sig)
resume_ptid = user_visible_resume_ptid (user_step);
/* Maybe resume a single thread after all. */
- if ((step || thread_has_single_step_breakpoints_set (tp))
- && tp->control.trap_expected)
+ if (target_is_non_stop_p ())
+ {
+ /* If non-stop mode, threads are always controlled
+ individually. */
+ resume_ptid = inferior_ptid;
+ }
+ else if ((step || thread_has_single_step_breakpoints_set (tp))
+ && tp->control.trap_expected)
{
/* We're allowing a thread to run past a breakpoint it has
hit, by single-stepping the thread with the breakpoint
@@ -2935,11 +2945,51 @@ proceed (CORE_ADDR addr, enum gdb_signal siggnal)
other thread was already doing one. In either case, don't
resume anything else until the step-over is finished. */
}
- else if (started && !non_stop)
+ else if (started && !target_is_non_stop_p ())
{
/* A new displaced stepping sequence was started. In all-stop,
we can't talk to the target anymore until it next stops. */
}
+ else if (!non_stop && target_is_non_stop_p ())
+ {
+ /* Start all other threads that are implicitly resumed too. */
+ ALL_NON_EXITED_THREADS (tp)
+ {
+ /* Ignore threads of processes we're not resuming. */
+ if (!ptid_match (tp->ptid, resume_ptid))
+ continue;
+
+ if (tp->resumed)
+ {
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: proceed: [%s] resumed\n",
+ target_pid_to_str (tp->ptid));
+ gdb_assert (tp->executing || tp->suspend.waitstatus_pending_p);
+ continue;
+ }
+
+ if (thread_is_in_step_over_chain (tp))
+ {
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: proceed: [%s] needs step-over\n",
+ target_pid_to_str (tp->ptid));
+ continue;
+ }
+
+ if (debug_infrun)
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: proceed: resuming %s\n",
+ target_pid_to_str (tp->ptid));
+
+ reset_ecs (ecs, tp);
+ switch_to_thread (tp->ptid);
+ keep_going_pass (ecs);
+ if (!ecs->wait_some_more)
+ error ("Command aborted.");
+ }
+ }
else if (!tp->resumed && !thread_is_in_step_over_chain (tp))
{
/* The thread wasn't started, and isn't queued, run it now. */
@@ -3151,7 +3201,7 @@ for_each_just_stopped_thread (for_each_just_stopped_thread_callback_func func)
if (!target_has_execution || ptid_equal (inferior_ptid, null_ptid))
return;
- if (non_stop)
+ if (target_is_non_stop_p ())
{
/* If in non-stop mode, only the current thread stopped. */
func (inferior_thread ());
@@ -3632,7 +3682,7 @@ fetch_inferior_event (void *client_data)
/* If an error happens while handling the event, propagate GDB's
knowledge of the executing state to the frontend/user running
state. */
- if (!non_stop)
+ if (!target_is_non_stop_p ())
ts_old_chain = make_cleanup (finish_thread_state_cleanup, &minus_one_ptid);
else
ts_old_chain = make_cleanup (finish_thread_state_cleanup, &ecs->ptid);
@@ -3871,7 +3921,8 @@ adjust_pc_after_break (struct thread_info *thread,
to get the "stopped by SW BP and needs adjustment" info out of
the target/kernel (and thus never reach here; see above). */
if (software_breakpoint_inserted_here_p (aspace, breakpoint_pc)
- || (non_stop && moribund_breakpoint_here_p (aspace, breakpoint_pc)))
+ || (target_is_non_stop_p ()
+ && moribund_breakpoint_here_p (aspace, breakpoint_pc)))
{
struct cleanup *old_cleanups = make_cleanup (null_cleanup, NULL);
@@ -4152,7 +4203,7 @@ stop_all_threads (void)
ptid_t entry_ptid;
struct cleanup *old_chain;
- gdb_assert (non_stop);
+ gdb_assert (target_is_non_stop_p ());
if (debug_infrun)
fprintf_unfiltered (gdb_stdlog, "infrun: stop_all_threads\n");
@@ -4466,7 +4517,7 @@ handle_inferior_event (struct execution_control_state *ecs)
{
ptid_t mark_ptid;
- if (!non_stop)
+ if (!target_is_non_stop_p ())
mark_ptid = minus_one_ptid;
else if (ecs->ws.kind == TARGET_WAITKIND_SIGNALLED
&& ecs->ws.kind == TARGET_WAITKIND_EXITED)
@@ -4769,7 +4820,8 @@ Cannot fill $_exitsignal with the correct signal number.\n"));
child = ecs->ws.value.related_pid;
/* In non-stop mode, also resume the other branch. */
- if (non_stop && !detach_fork)
+ if (!detach_fork && (non_stop
+ || (sched_multi && target_is_non_stop_p ())))
{
if (follow_child)
switch_to_thread (parent);
@@ -5037,7 +5089,7 @@ finish_step_over (struct execution_control_state *ecs)
clear_step_over_info ();
}
- if (!non_stop)
+ if (!target_is_non_stop_p ())
return 0;
/* Start a new step-over in another thread if there's one that
@@ -5614,6 +5666,17 @@ handle_signal_stop (struct execution_control_state *ecs)
/* Reset trap_expected to ensure breakpoints are re-inserted. */
ecs->event_thread->control.trap_expected = 0;
+ if (!non_stop && target_is_non_stop_p ())
+ {
+ keep_going (ecs);
+
+ /* We've canceled the step-over temporarily while the
+ signal handler executes. Let other threads run,
+ according to schedlock. */
+ restart_threads (ecs->event_thread);
+ return;
+ }
+
/* If we were nexting/stepping some other thread, switch to
it, so that we don't continue it, losing control. */
if (!switch_back_to_stepped_thread (ecs))
@@ -6498,7 +6561,7 @@ process_event_stop_test (struct execution_control_state *ecs)
static int
switch_back_to_stepped_thread (struct execution_control_state *ecs)
{
- if (!non_stop)
+ if (!target_is_non_stop_p ())
{
struct thread_info *tp;
struct thread_info *stepping_thread;
@@ -6589,7 +6652,8 @@ switch_back_to_stepped_thread (struct execution_control_state *ecs)
ALL_NON_EXITED_THREADS (tp)
{
- /* Ignore threads of processes we're not resuming. */
+ /* Ignore threads of processes the caller is not
+ resuming. */
if (!sched_multi
&& ptid_get_pid (tp->ptid) != ptid_get_pid (ecs->ptid))
continue;
@@ -6735,7 +6799,10 @@ keep_going_stepped_thread (struct thread_info *tp)
stop_pc);
tp->resumed = 1;
- resume_ptid = user_visible_resume_ptid (tp->control.stepping_command);
+ if (target_is_non_stop_p ())
+ resume_ptid = inferior_ptid;
+ else
+ resume_ptid = user_visible_resume_ptid (tp->control.stepping_command);
do_target_resume (resume_ptid, 0, GDB_SIGNAL_0);
}
else
@@ -7153,6 +7220,11 @@ stop_waiting (struct execution_control_state *ecs)
/* Let callers know we don't want to wait for the inferior anymore. */
ecs->wait_some_more = 0;
+
+ /* If all-stop, but the target is always in non-stop mode, stop all
+ threads now that we're presenting the stop to the user. */
+ if (!non_stop && target_is_non_stop_p ())
+ stop_all_threads ();
}
/* Like keep_going, but passes the signal to the inferior, even if the
@@ -7267,7 +7339,7 @@ keep_going_pass (struct execution_control_state *ecs)
insert_breakpoints below, because that removes the breakpoint
we're about to step over, otherwise other threads could miss
it. */
- if (step_over_info_valid_p () && non_stop)
+ if (step_over_info_valid_p () && target_is_non_stop_p ())
stop_all_threads ();
/* Stop stepping if inserting breakpoints fails. */
diff --git a/gdb/linux-nat.c b/gdb/linux-nat.c
index b04aa68..b449485 100644
--- a/gdb/linux-nat.c
+++ b/gdb/linux-nat.c
@@ -1375,13 +1375,13 @@ get_pending_status (struct lwp_info *lp, int *status)
signo = GDB_SIGNAL_0; /* a pending ptrace event, not a real signal. */
else if (lp->status)
signo = gdb_signal_from_host (WSTOPSIG (lp->status));
- else if (non_stop && !is_executing (lp->ptid))
+ else if (target_is_non_stop_p () && !is_executing (lp->ptid))
{
struct thread_info *tp = find_thread_ptid (lp->ptid);
signo = tp->suspend.stop_signal;
}
- else if (!non_stop)
+ else if (!target_is_non_stop_p ())
{
struct target_waitstatus last;
ptid_t last_ptid;
@@ -2931,7 +2931,7 @@ select_event_lwp (ptid_t filter, struct lwp_info **orig_lp, int *status)
having stepped the thread, wouldn't understand what the trap was
for, and therefore would report it to the user as a random
signal. */
- if (!non_stop)
+ if (!target_is_non_stop_p ())
{
event_lp = iterate_over_lwps (filter,
select_singlestep_lwp_callback, NULL);
@@ -3278,7 +3278,7 @@ linux_nat_filter_event (int lwpid, int status)
{
enum gdb_signal signo = gdb_signal_from_host (WSTOPSIG (status));
- if (!non_stop)
+ if (!target_is_non_stop_p ())
{
/* Only do the below in all-stop, as we currently use SIGSTOP
to implement target_stop (see linux_nat_stop) in
@@ -3544,7 +3544,7 @@ linux_nat_wait_1 (struct target_ops *ops,
status = lp->status;
lp->status = 0;
- if (!non_stop)
+ if (!target_is_non_stop_p ())
{
/* Now stop all other LWP's ... */
iterate_over_lwps (minus_one_ptid, stop_callback, NULL);
@@ -3586,7 +3586,7 @@ linux_nat_wait_1 (struct target_ops *ops,
clears it. */
last_resume_kind = lp->last_resume_kind;
- if (!non_stop)
+ if (!target_is_non_stop_p ())
{
/* In all-stop, from the core's perspective, all LWPs are now
stopped until a new resume action is sent over. */
@@ -3719,7 +3719,7 @@ linux_nat_wait (struct target_ops *ops,
specific_process, for example, see linux_nat_wait_1), and
meanwhile the event became uninteresting. Don't bother resuming
LWPs we're not going to wait for if they'd stop immediately. */
- if (non_stop)
+ if (target_is_non_stop_p ())
iterate_over_lwps (minus_one_ptid, resume_stopped_resumed_lwps, &ptid);
event_ptid = linux_nat_wait_1 (ops, ptid, ourstatus, target_options);
@@ -4560,6 +4560,14 @@ linux_nat_supports_non_stop (struct target_ops *self)
return 1;
}
+/* to_always_non_stop_p implementation. */
+
+static int
+linux_nat_always_non_stop_p (struct target_ops *self)
+{
+ return 0;
+}
+
/* True if we want to support multi-process. To be removed when GDB
supports multi-exec. */
@@ -4779,7 +4787,7 @@ linux_nat_stop_lwp (struct lwp_info *lwp, void *data)
static void
linux_nat_stop (struct target_ops *self, ptid_t ptid)
{
- if (non_stop)
+ if (target_is_non_stop_p ())
iterate_over_lwps (ptid, linux_nat_stop_lwp, NULL);
else
linux_ops->to_stop (linux_ops, ptid);
@@ -4878,6 +4886,7 @@ linux_nat_add_target (struct target_ops *t)
t->to_can_async_p = linux_nat_can_async_p;
t->to_is_async_p = linux_nat_is_async_p;
t->to_supports_non_stop = linux_nat_supports_non_stop;
+ t->to_always_non_stop_p = linux_nat_always_non_stop_p;
t->to_async = linux_nat_async;
t->to_terminal_inferior = linux_nat_terminal_inferior;
t->to_terminal_ours = linux_nat_terminal_ours;
diff --git a/gdb/target-delegates.c b/gdb/target-delegates.c
index 36eacbf..8a92acf 100644
--- a/gdb/target-delegates.c
+++ b/gdb/target-delegates.c
@@ -1744,6 +1744,33 @@ debug_supports_non_stop (struct target_ops *self)
}
static int
+delegate_always_non_stop_p (struct target_ops *self)
+{
+ self = self->beneath;
+ return self->to_always_non_stop_p (self);
+}
+
+static int
+tdefault_always_non_stop_p (struct target_ops *self)
+{
+ return 0;
+}
+
+static int
+debug_always_non_stop_p (struct target_ops *self)
+{
+ int result;
+ fprintf_unfiltered (gdb_stdlog, "-> %s->to_always_non_stop_p (...)\n", debug_target.to_shortname);
+ result = debug_target.to_always_non_stop_p (&debug_target);
+ fprintf_unfiltered (gdb_stdlog, "<- %s->to_always_non_stop_p (", debug_target.to_shortname);
+ target_debug_print_struct_target_ops_p (&debug_target);
+ fputs_unfiltered (") = ", gdb_stdlog);
+ target_debug_print_int (result);
+ fputs_unfiltered ("\n", gdb_stdlog);
+ return result;
+}
+
+static int
delegate_find_memory_regions (struct target_ops *self, find_memory_region_ftype arg1, void *arg2)
{
self = self->beneath;
@@ -4005,6 +4032,8 @@ install_delegators (struct target_ops *ops)
ops->to_async = delegate_async;
if (ops->to_supports_non_stop == NULL)
ops->to_supports_non_stop = delegate_supports_non_stop;
+ if (ops->to_always_non_stop_p == NULL)
+ ops->to_always_non_stop_p = delegate_always_non_stop_p;
if (ops->to_find_memory_regions == NULL)
ops->to_find_memory_regions = delegate_find_memory_regions;
if (ops->to_make_corefile_notes == NULL)
@@ -4232,6 +4261,7 @@ install_dummy_methods (struct target_ops *ops)
ops->to_is_async_p = tdefault_is_async_p;
ops->to_async = tdefault_async;
ops->to_supports_non_stop = tdefault_supports_non_stop;
+ ops->to_always_non_stop_p = tdefault_always_non_stop_p;
ops->to_find_memory_regions = dummy_find_memory_regions;
ops->to_make_corefile_notes = dummy_make_corefile_notes;
ops->to_get_bookmark = tdefault_get_bookmark;
@@ -4380,6 +4410,7 @@ init_debug_target (struct target_ops *ops)
ops->to_is_async_p = debug_is_async_p;
ops->to_async = debug_async;
ops->to_supports_non_stop = debug_supports_non_stop;
+ ops->to_always_non_stop_p = debug_always_non_stop_p;
ops->to_find_memory_regions = debug_find_memory_regions;
ops->to_make_corefile_notes = debug_make_corefile_notes;
ops->to_get_bookmark = debug_get_bookmark;
diff --git a/gdb/target.c b/gdb/target.c
index e992a35..ab6c977 100644
--- a/gdb/target.c
+++ b/gdb/target.c
@@ -3752,6 +3752,58 @@ maint_show_target_async_command (struct ui_file *file, int from_tty,
"asynchronous mode is %s.\n"), value);
}
+/* See target.h. */
+
+int
+target_is_non_stop_p (void)
+{
+ return (non_stop
+ || target_non_stop_enabled == AUTO_BOOLEAN_TRUE
+ || (target_non_stop_enabled == AUTO_BOOLEAN_AUTO
+ && current_target.to_always_non_stop_p (¤t_target)));
+}
+
+/* Controls if targets can report that they always run in non-stop
+ mode. This is just for maintainers to use when debugging gdb. */
+enum auto_boolean target_non_stop_enabled = AUTO_BOOLEAN_AUTO;
+
+/* The set command writes to this variable. If the inferior is
+ executing, target_non_stop_enabled is *not* updated. */
+static enum auto_boolean target_non_stop_enabled_1 = AUTO_BOOLEAN_AUTO;
+
+/* Implementation of "maint set target-non-stop". */
+
+static void
+maint_set_target_non_stop_command (char *args, int from_tty,
+ struct cmd_list_element *c)
+{
+ if (have_live_inferiors ())
+ {
+ target_non_stop_enabled_1 = target_non_stop_enabled;
+ error (_("Cannot change this setting while the inferior is running."));
+ }
+
+ target_non_stop_enabled = target_non_stop_enabled_1;
+}
+
+/* Implementation of "maint show target-non-stop". */
+
+static void
+maint_show_target_non_stop_command (struct ui_file *file, int from_tty,
+ struct cmd_list_element *c,
+ const char *value)
+{
+ if (target_non_stop_enabled == AUTO_BOOLEAN_AUTO)
+ fprintf_filtered (file,
+ _("Whether the target is always in non-stop mode "
+ "is %s (currently %s).\n"), value,
+ target_is_non_stop_p () ? "on" : "off");
+ else
+ fprintf_filtered (file,
+ _("Whether the target is always in non-stop mode "
+ "is %s.\n"), value);
+}
+
/* Temporary copies of permission settings. */
static int may_write_registers_1 = 1;
@@ -3854,6 +3906,16 @@ Tells gdb whether to control the inferior in asynchronous mode."),
&maintenance_set_cmdlist,
&maintenance_show_cmdlist);
+ add_setshow_auto_boolean_cmd ("target-non-stop", no_class,
+ &target_non_stop_enabled_1, _("\
+Set whether gdb always controls the inferior in non-stop mode."), _("\
+Show whether gdb always controls the inferior in non-stop mode."), _("\
+Tells gdb whether to control the inferior in non-stop mode."),
+ maint_set_target_non_stop_command,
+ maint_show_target_non_stop_command,
+ &maintenance_set_cmdlist,
+ &maintenance_show_cmdlist);
+
add_setshow_boolean_cmd ("may-write-registers", class_support,
&may_write_registers_1, _("\
Set permission to write into registers."), _("\
diff --git a/gdb/target.h b/gdb/target.h
index 3f1a8b0..17dee43 100644
--- a/gdb/target.h
+++ b/gdb/target.h
@@ -640,6 +640,10 @@ struct target_ops
comment on 'to_can_run'. */
int (*to_supports_non_stop) (struct target_ops *)
TARGET_DEFAULT_RETURN (0);
+ /* Return true if the target operates in non-stop mode even when
+ "set non-stop off". */
+ int (*to_always_non_stop_p) (struct target_ops *)
+ TARGET_DEFAULT_RETURN (0);
/* find_memory_regions support method for gcore */
int (*to_find_memory_regions) (struct target_ops *,
find_memory_region_ftype func, void *data)
@@ -1706,6 +1710,15 @@ extern int target_async_permitted;
/* Enables/disabled async target events. */
extern void target_async (int enable);
+/* Whether support for controlling the target backends always in
+ non-stop mode is enabled. */
+extern enum auto_boolean target_non_stop_enabled;
+
+/* Is the target in non-stop mode? Some targets control the inferior
+ in non-stop mode even with "set non-stop off". Always true if "set
+ non-stop" is on. */
+extern int target_is_non_stop_p (void);
+
#define target_execution_direction() \
(current_target.to_execution_direction (¤t_target))
--
1.9.3
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 10/17] Implement all-stop on top of a target running non-stop mode
2015-04-17 11:38 ` [PATCH v3 10/17] Implement all-stop on top of a target running non-stop mode Pedro Alves
@ 2015-04-21 11:09 ` Yao Qi
2015-04-22 20:16 ` Pedro Alves
0 siblings, 1 reply; 62+ messages in thread
From: Yao Qi @ 2015-04-21 11:09 UTC (permalink / raw)
To: Pedro Alves; +Cc: gdb-patches
Pedro Alves <palves@redhat.com> writes:
> gdb/ChangeLog:
> 2015-04-17 Pedro Alves <palves@redhat.com>
>
> * NEWS: Mention "maint set/show target-non-stop".
> * breakpoint.c (update_global_location_list): Check
> target_is_non_stop_p instead of non_stop.
> * infcmd.c (attach_command_post_wait, attach_command): Likewise.
> * infrun.c (show_can_use_displaced_stepping)
> (can_use_displaced_stepping_p, start_step_over_inferior):
> Likewise.
> (resume): Always resume a single thread if the target is in
> non-stop mode.
> (proceed): Check target_is_non_stop_p instead of non_stop. If in
> all-mode but the target is always in non-stop mode, start all the
^^^^^^^^
all-stop mode.
> (handle_signal_stop) <random signal>: If we get a signal while
> stepping over a breakpoint, and the target is always in non-stop
> mode, restart all threads.
> @@ -5614,6 +5666,17 @@ handle_signal_stop (struct execution_control_state *ecs)
> /* Reset trap_expected to ensure breakpoints are re-inserted. */
> ecs->event_thread->control.trap_expected = 0;
>
> + if (!non_stop && target_is_non_stop_p ())
> + {
This path is about the case that a signal is got while in in-line
stepping, isn't? If so, non_stop should be an invariant false. We
don't need to check it.
> + keep_going (ecs);
> +
> + /* We've canceled the step-over temporarily while the
> + signal handler executes. Let other threads run,
> + according to schedlock. */
IMO, we don't need to run other threads according to schedlock. GDB
resumes some threads here and operations should be invisible to user,
schedlock, as a user visible option, shouldn't affect what threads
should be resumed. On the other hand, restart_threads is to undo the
changes done by stop_all_threads, so no user preference (schedlock)
should be considered here.
> + restart_threads (ecs->event_thread);
> + return;
> + }
> +
--
Yao (齐尧)
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 10/17] Implement all-stop on top of a target running non-stop mode
2015-04-21 11:09 ` Yao Qi
@ 2015-04-22 20:16 ` Pedro Alves
2015-04-24 7:39 ` Yao Qi
0 siblings, 1 reply; 62+ messages in thread
From: Pedro Alves @ 2015-04-22 20:16 UTC (permalink / raw)
To: Yao Qi; +Cc: gdb-patches
On 04/21/2015 12:08 PM, Yao Qi wrote:
> Pedro Alves <palves@redhat.com> writes:
>
>> gdb/ChangeLog:
>> 2015-04-17 Pedro Alves <palves@redhat.com>
>>
>> * NEWS: Mention "maint set/show target-non-stop".
>> * breakpoint.c (update_global_location_list): Check
>> target_is_non_stop_p instead of non_stop.
>> * infcmd.c (attach_command_post_wait, attach_command): Likewise.
>> * infrun.c (show_can_use_displaced_stepping)
>> (can_use_displaced_stepping_p, start_step_over_inferior):
>> Likewise.
>> (resume): Always resume a single thread if the target is in
>> non-stop mode.
>> (proceed): Check target_is_non_stop_p instead of non_stop. If in
>> all-mode but the target is always in non-stop mode, start all the
> ^^^^^^^^
>
> all-stop mode.
Thanks, fixed.
>
>> (handle_signal_stop) <random signal>: If we get a signal while
>> stepping over a breakpoint, and the target is always in non-stop
>> mode, restart all threads.
>
>> @@ -5614,6 +5666,17 @@ handle_signal_stop (struct execution_control_state *ecs)
>> /* Reset trap_expected to ensure breakpoints are re-inserted. */
>> ecs->event_thread->control.trap_expected = 0;
>>
>> + if (!non_stop && target_is_non_stop_p ())
>> + {
>
> This path is about the case that a signal is got while in in-line
> stepping, isn't? If so, non_stop should be an invariant false. We
> don't need to check it.
Hmm, not sure what you mean:
- We need to do this with displaced stepping too, because we can't
deliver signals while doing a displaced step. See comments at the
top of displaced_step_prepare and in do_target_resume.
- We can certainly get a signal while doing an in-line step-over.
The simplest would be, trying to step-over a breakpoint here:
*(volatile int *)0 = 1;
which usually results SIGSEGV.
- We can step past breakpoints in-line in non-stop mode too.
However, I agree there's something wrong here. If we get here
with "set non-stop on", and we were doing an in-line step-over,
then we also need to restart threads, not just
in all-stop + "target always-non-stop". I've applied this change
on top now, and it tested OK on x86-64, native with
both displaced on/off, and gdbserver.
diff --git a/gdb/infrun.c b/gdb/infrun.c
index e236510..3e60313 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -5732,6 +5732,8 @@ handle_signal_stop (struct execution_control_state *ecs)
&& ecs->event_thread->control.trap_expected
&& ecs->event_thread->control.step_resume_breakpoint == NULL)
{
+ int was_in_line;
+
/* We were just starting a new sequence, attempting to
single-step off of a breakpoint and expecting a SIGTRAP.
Instead this signal arrives. This signal will take us out
@@ -5747,20 +5749,29 @@ handle_signal_stop (struct execution_control_state *ecs)
"infrun: signal arrived while stepping over "
"breakpoint\n");
+ was_in_line = step_over_info_valid_p ();
clear_step_over_info ();
insert_hp_step_resume_breakpoint_at_frame (frame);
ecs->event_thread->step_after_step_resume_breakpoint = 1;
/* Reset trap_expected to ensure breakpoints are re-inserted. */
ecs->event_thread->control.trap_expected = 0;
- if (!non_stop && target_is_non_stop_p ())
+ if (target_is_non_stop_p ())
{
keep_going (ecs);
- /* We've canceled the step-over temporarily while the
- signal handler executes. Let other threads run,
- according to schedlock. */
- restart_threads (ecs->event_thread);
+ /* The step-over has been canceled temporarily while the
+ signal handler executes. */
+ if (was_in_line)
+ {
+ /* We had paused all threads, restart them. */
+ restart_threads (ecs->event_thread);
+ }
+ else if (debug_infrun)
+ {
+ fprintf_unfiltered (gdb_stdlog,
+ "infrun: no need to restart threads\n");
+ }
return;
}
--
1.9.3
I've now folded that into the patch that teaches non-stop about
in-line step overs.
>> + keep_going (ecs);
>> +
>> + /* We've canceled the step-over temporarily while the
>> + signal handler executes. Let other threads run,
>> + according to schedlock. */
>
> IMO, we don't need to run other threads according to schedlock. GDB
> resumes some threads here and operations should be invisible to user,
> schedlock, as a user visible option, shouldn't affect what threads
> should be resumed. On the other hand, restart_threads is to undo the
> changes done by stop_all_threads, so no user preference (schedlock)
> should be considered here.
Yes, you're right, thanks for noticing this. The code in restart_threads does
not in fact look at schedlock (through e.g., user_visible_resume_ptid) at
all, it's the comment that is stale from a previous attempt, from
before https://sourceware.org/ml/gdb-patches/2015-03/msg00293.html.
It was issues like that led to that schedlock patch, even.
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 10/17] Implement all-stop on top of a target running non-stop mode
2015-04-22 20:16 ` Pedro Alves
@ 2015-04-24 7:39 ` Yao Qi
2015-05-19 18:08 ` Pedro Alves
0 siblings, 1 reply; 62+ messages in thread
From: Yao Qi @ 2015-04-24 7:39 UTC (permalink / raw)
To: Pedro Alves; +Cc: Yao Qi, gdb-patches
Pedro Alves <palves@redhat.com> writes:
>> This path is about the case that a signal is got while in in-line
>> stepping, isn't? If so, non_stop should be an invariant false. We
>> don't need to check it.
>
> Hmm, not sure what you mean:
>
Let me ask it in another way, when we get here, it means a signal
arrived, the code some lines above is:
if (debug_infrun)
fprintf_unfiltered (gdb_stdlog,
"infrun: signal arrived while stepping over "
"breakpoint\n");
GDB just did the breakpoint step over, via in-line stepping or
out-of-line stepping, right? as your patch below shows.
> - We need to do this with displaced stepping too, because we can't
> deliver signals while doing a displaced step. See comments at the
> top of displaced_step_prepare and in do_target_resume.
The first sentence is contradictory, or you mean we *can* do either
out-of-line stepping or in-line stepping, but we can't deliver a signal
while doing a displaced stepping...
>
> - We can certainly get a signal while doing an in-line step-over.
> The simplest would be, trying to step-over a breakpoint here:
>
> *(volatile int *)0 = 1;
>
> which usually results SIGSEGV.
... while we can deliver a signal in in-line stepping. Is it correct?
--
Yao (齐尧)
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 10/17] Implement all-stop on top of a target running non-stop mode
2015-04-24 7:39 ` Yao Qi
@ 2015-05-19 18:08 ` Pedro Alves
2015-05-21 9:17 ` Yao Qi
0 siblings, 1 reply; 62+ messages in thread
From: Pedro Alves @ 2015-05-19 18:08 UTC (permalink / raw)
To: Yao Qi; +Cc: gdb-patches
Hi Yao,
I had not realized I missed answering this.
On 04/24/2015 08:39 AM, Yao Qi wrote:
> Pedro Alves <palves@redhat.com> writes:
>
>>> This path is about the case that a signal is got while in in-line
>>> stepping, isn't? If so, non_stop should be an invariant false. We
>>> don't need to check it.
>>
>> Hmm, not sure what you mean:
>>
>
> Let me ask it in another way, when we get here, it means a signal
> arrived, the code some lines above is:
>
> if (debug_infrun)
> fprintf_unfiltered (gdb_stdlog,
> "infrun: signal arrived while stepping over "
> "breakpoint\n");
>
> GDB just did the breakpoint step over, via in-line stepping or
> out-of-line stepping, right? as your patch below shows.
>
>> - We need to do this with displaced stepping too, because we can't
>> deliver signals while doing a displaced step. See comments at the
>> top of displaced_step_prepare and in do_target_resume.
>
> The first sentence is contradictory, or you mean we *can* do either
> out-of-line stepping or in-line stepping, but we can't deliver a signal
> while doing a displaced stepping...
By "need to do this" I meant that we need to cancel the step over in
progress, and insert the step-resume at the signal handlers return.
>
>>
>> - We can certainly get a signal while doing an in-line step-over.
>> The simplest would be, trying to step-over a breakpoint here:
>>
>> *(volatile int *)0 = 1;
>>
>> which usually results SIGSEGV.
>
> ... while we can deliver a signal in in-line stepping. Is it correct?
>
Not really.
We can't deliver a signal while a displaced step is in
progress, because we wouldn't be able to tell whether the thread stepped
to the handler or to main line code, so we wouldn't know whether to apply
the displaced step fixup. Also, if the thread stops for some reason while
in the handler, we'd end up with the scratch pad still in the frame chain,
so later when the thread is re-resumed, it'd return to the scratch pad, but
the scratch pad would have already have been overwritten.
And we can't deliver a signal while stepping over a breakpoint in-line
either, because that requires removing the breakpoint, and the signal handler
can recurse and call the same code that had the breakpoint we were
stepping over -- which would mean we'd miss trapping on that breakpoint.
(this is tested by signest.exp).
So the only difference between stepping over a breakpoint in-line or
out-of-line here, is that with the former, threads had been previously
paused, so we re-resume them while the signal handler runs (to avoid
debugger-induced inferior deadlock).
Let me know if that answered your question.
Thanks,
Pedro Alves
^ permalink raw reply [flat|nested] 62+ messages in thread
* Re: [PATCH v3 10/17] Implement all-stop on top of a target running non-stop mode
2015-05-19 18:08 ` Pedro Alves
@ 2015-05-21 9:17 ` Yao Qi
0 siblings, 0 replies; 62+ messages in thread
From: Yao Qi @ 2015-05-21 9:17 UTC (permalink / raw)
To: Pedro Alves; +Cc: Yao Qi, gdb-patches
Pedro Alves <palves@redhat.com> writes:
> So the only difference between stepping over a breakpoint in-line or
> out-of-line here, is that with the former, threads had been previously
> paused, so we re-resume them while the signal handler runs (to avoid
> debugger-induced inferior deadlock).
Looks in-line stepping and out-of-line stepping behaves similarly under
this context.
>
> Let me know if that answered your question.
Yes, that is what I want to know, thanks.
--
Yao (齐尧)
^ permalink raw reply [flat|nested] 62+ messages in thread
* Re: [PATCH v3 00/23] All-stop on top of non-stop
2015-04-17 10:47 [PATCH v3 00/23] All-stop on top of non-stop Pedro Alves
` (16 preceding siblings ...)
2015-04-17 11:38 ` [PATCH v3 10/17] Implement all-stop on top of a target running non-stop mode Pedro Alves
@ 2015-04-20 12:02 ` Yao Qi
2015-04-20 16:54 ` Sergio Durigan Junior
2015-04-20 17:35 ` Simon Marchi
18 siblings, 1 reply; 62+ messages in thread
From: Yao Qi @ 2015-04-20 12:02 UTC (permalink / raw)
To: Pedro Alves, gdb-patches
On 17/04/15 11:45, Pedro Alves wrote:
> x86-64 GNU/Linux
> x86-64 GNU/Linux on top of software single-step branch
> PPC64 GNU/Linux
> S/390 GNU/Linux
Hi Pedro,
I tested V3 on arm GNU/Linux and aarch64 GNU/Linux in both native and
gdbserver, there is no regression on arm GNU/Linux. There is one
regression on aarch64 GNU/Linux native, but I don't see how is this fail
related to your V3 patches. Note that there is no regression either on
aarch64 GNU/Linux with gdbserver.
(gdb) info threads^M
Id Target Id Frame ^M
...
30 Thread 0x2019dff1f0 (LWP 28308) "joinable" 0x0000002000121a68
in clone () from /lib/aarch64-linux-gnu/libc.so.6^
...
GPT: lwp Thread 0x2019dff1f0 (LWP 28308) has no pending signal^M
Can't detach Thread 0x2019dff1f0 (LWP 28308): No such process^
(gdb) FAIL: gdb.threads/attach-many-short-lived-threads.exp: iter 1:
detach
Another race problem on attach/detach?
--
Yao (é½å°§)
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 00/23] All-stop on top of non-stop
2015-04-20 12:02 ` [PATCH v3 00/23] All-stop on top of non-stop Yao Qi
@ 2015-04-20 16:54 ` Sergio Durigan Junior
2015-04-20 16:43 ` Pedro Alves
0 siblings, 1 reply; 62+ messages in thread
From: Sergio Durigan Junior @ 2015-04-20 16:54 UTC (permalink / raw)
To: Yao Qi; +Cc: Pedro Alves, gdb-patches
On Monday, April 20 2015, Yao Qi wrote:
> Hi Pedro,
> I tested V3 on arm GNU/Linux and aarch64 GNU/Linux in both native and
> gdbserver, there is no regression on arm GNU/Linux. There is one
> regression on aarch64 GNU/Linux native, but I don't see how is this fail
> related to your V3 patches. Note that there is no regression either on
> aarch64 GNU/Linux with gdbserver.
>
> (gdb) info threads^M
> Id Target Id Frame ^M
> ...
> 30 Thread 0x2019dff1f0 (LWP 28308) "joinable" 0x0000002000121a68
> in clone () from /lib/aarch64-linux-gnu/libc.so.6^
> ...
> GPT: lwp Thread 0x2019dff1f0 (LWP 28308) has no pending signal^M
> Can't detach Thread 0x2019dff1f0 (LWP 28308): No such process^
> (gdb) FAIL: gdb.threads/attach-many-short-lived-threads.exp: iter 1:
> detach
>
> Another race problem on attach/detach?
FWIW, gdb.threads/attach-many-short-lived-threads.exp is one of the
villains... It always shows up on our BuildBot logs.
--
Sergio
GPG key ID: 237A 54B1 0287 28BF 00EF 31F4 D0EB 7628 65FC 5E36
Please send encrypted e-mail if possible
http://sergiodj.net/
^ permalink raw reply [flat|nested] 62+ messages in thread
* Re: [PATCH v3 00/23] All-stop on top of non-stop
2015-04-20 16:54 ` Sergio Durigan Junior
@ 2015-04-20 16:43 ` Pedro Alves
2015-04-21 7:48 ` Yao Qi
0 siblings, 1 reply; 62+ messages in thread
From: Pedro Alves @ 2015-04-20 16:43 UTC (permalink / raw)
To: Sergio Durigan Junior, Yao Qi; +Cc: gdb-patches
On 04/20/2015 04:56 PM, Sergio Durigan Junior wrote:
> On Monday, April 20 2015, Yao Qi wrote:
>
>> Hi Pedro,
>> I tested V3 on arm GNU/Linux and aarch64 GNU/Linux in both native and
>> gdbserver, there is no regression on arm GNU/Linux.
Thanks! That's great news.
>> There is one
>> regression on aarch64 GNU/Linux native, but I don't see how is this fail
>> related to your V3 patches.
I assume you mean you saw this once, and if that if you rerun that
test in particular, the fail goes away.
> Note that there is no regression either on
>> aarch64 GNU/Linux with gdbserver.
Awesome.
>>
>> (gdb) info threads^M
>> Id Target Id Frame ^M
>> ...
>> 30 Thread 0x2019dff1f0 (LWP 28308) "joinable" 0x0000002000121a68
>> in clone () from /lib/aarch64-linux-gnu/libc.so.6^
>> ...
>> GPT: lwp Thread 0x2019dff1f0 (LWP 28308) has no pending signal^M
>> Can't detach Thread 0x2019dff1f0 (LWP 28308): No such process^
>> (gdb) FAIL: gdb.threads/attach-many-short-lived-threads.exp: iter 1:
>> detach
>>
>> Another race problem on attach/detach?
I think so, yes. The same race that was fixed for regular
resumes, here:
https://sourceware.org/ml/gdb-patches/2015-03/msg00597.html
can easily happen when detaching. Say, you have 2 threads in your
program. GDB must detach from each thread individually. So it
starts by detaching from thread 1. And just before detaching
from thread 2, thread 1 causes the whole process to exit. So
the PTRACE_DETACH for thread 2 errors with ESRCH (No such process.)
Now I don't know is why you would see that on iteration 1 ("iter 1").
One possibility could be that a thread that has already been
detached crashed the process, because e.g., GDB left breakpoints
inserted in memory by mistake, or something is wrong with
infrun.c:prepare_for_detach, or something else along those lines.
In that case, assuming "ulimit -c unlimited", you'll probably find you have
a core dump for the crashing inferior.
>
> FWIW, gdb.threads/attach-many-short-lived-threads.exp is one of the
> villains... It always shows up on our BuildBot logs.
Yes, thought I don't think those have been showing a problem with the
test, but rather that the test is still catching bugs.
Thanks,
Pedro Alves
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 00/23] All-stop on top of non-stop
2015-04-20 16:43 ` Pedro Alves
@ 2015-04-21 7:48 ` Yao Qi
2015-04-21 15:05 ` Yao Qi
0 siblings, 1 reply; 62+ messages in thread
From: Yao Qi @ 2015-04-21 7:48 UTC (permalink / raw)
To: Pedro Alves; +Cc: Sergio Durigan Junior, Yao Qi, gdb-patches
Pedro Alves <palves@redhat.com> writes:
> I assume you mean you saw this once, and if that if you rerun that
> test in particular, the fail goes away.
No, I saw these fails very time I run this single test.
>
>> Note that there is no regression either on
>>> aarch64 GNU/Linux with gdbserver.
>
> Awesome.
>
>>>
>>> (gdb) info threads^M
>>> Id Target Id Frame ^M
>>> ...
>>> 30 Thread 0x2019dff1f0 (LWP 28308) "joinable" 0x0000002000121a68
>>> in clone () from /lib/aarch64-linux-gnu/libc.so.6^
>>> ...
>>> GPT: lwp Thread 0x2019dff1f0 (LWP 28308) has no pending signal^M
>>> Can't detach Thread 0x2019dff1f0 (LWP 28308): No such process^
>>> (gdb) FAIL: gdb.threads/attach-many-short-lived-threads.exp: iter 1:
>>> detach
>>>
>>> Another race problem on attach/detach?
>
> I think so, yes. The same race that was fixed for regular
> resumes, here:
>
> https://sourceware.org/ml/gdb-patches/2015-03/msg00597.html
>
> can easily happen when detaching. Say, you have 2 threads in your
> program. GDB must detach from each thread individually. So it
> starts by detaching from thread 1. And just before detaching
> from thread 2, thread 1 causes the whole process to exit. So
> the PTRACE_DETACH for thread 2 errors with ESRCH (No such process.)
>
> Now I don't know is why you would see that on iteration 1 ("iter 1").
> One possibility could be that a thread that has already been
> detached crashed the process, because e.g., GDB left breakpoints
> inserted in memory by mistake, or something is wrong with
> infrun.c:prepare_for_detach, or something else along those lines.
> In that case, assuming "ulimit -c unlimited", you'll probably find you have
> a core dump for the crashing inferior.
Thanks for the clue. I'll dig it deeper. I'll read these patches
today, if I don't see any new problems or questions, these patches are
fine to me.
--
Yao (齐尧)
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCH v3 00/23] All-stop on top of non-stop
2015-04-21 7:48 ` Yao Qi
@ 2015-04-21 15:05 ` Yao Qi
2015-04-22 22:27 ` Pedro Alves
0 siblings, 1 reply; 62+ messages in thread
From: Yao Qi @ 2015-04-21 15:05 UTC (permalink / raw)
To: Pedro Alves; +Cc: Sergio Durigan Junior, gdb-patches
On 21/04/15 08:47, Yao Qi wrote:
> Thanks for the clue. I'll dig it deeper. I'll read these patches
> today, if I don't see any new problems or questions, these patches are
> fine to me.
Hi Pedro,
I've look at all this patch series, and sent out my questions to them.
Although I don't fully understand them, they are good to me in general,
given the test results we've go on different archs.
--
Yao (é½å°§)
^ permalink raw reply [flat|nested] 62+ messages in thread
* Re: [PATCH v3 00/23] All-stop on top of non-stop
2015-04-21 15:05 ` Yao Qi
@ 2015-04-22 22:27 ` Pedro Alves
0 siblings, 0 replies; 62+ messages in thread
From: Pedro Alves @ 2015-04-22 22:27 UTC (permalink / raw)
To: Yao Qi; +Cc: gdb-patches
On 04/21/2015 04:05 PM, Yao Qi wrote:
> On 21/04/15 08:47, Yao Qi wrote:
>> Thanks for the clue. I'll dig it deeper. I'll read these patches
>> today, if I don't see any new problems or questions, these patches are
>> fine to me.
>
> Hi Pedro,
> I've look at all this patch series, and sent out my questions to them.
> Although I don't fully understand them, they are good to me in general,
> given the test results we've go on different archs.
Thanks Yao. I think I replied to all your questions. I'll be glad to
try to clarify anything else that still might need it.
I think I addressed all the comments I got so far, and I've tested on
x86. Out of paranoia, I'm retesting on PPC64 and S/390 as well, though
it's probably unnecessary as nothing major changed.
Meanwhile, I pushed the changes to users/palves/all-stop-non-stop.
Thanks,
Pedro Alves
^ permalink raw reply [flat|nested] 62+ messages in thread
* Re: [PATCH v3 00/23] All-stop on top of non-stop
2015-04-17 10:47 [PATCH v3 00/23] All-stop on top of non-stop Pedro Alves
` (17 preceding siblings ...)
2015-04-20 12:02 ` [PATCH v3 00/23] All-stop on top of non-stop Yao Qi
@ 2015-04-20 17:35 ` Simon Marchi
2015-05-19 18:14 ` Pedro Alves
18 siblings, 1 reply; 62+ messages in thread
From: Simon Marchi @ 2015-04-20 17:35 UTC (permalink / raw)
To: Pedro Alves, gdb-patches
On 15-04-17 06:45 AM, Pedro Alves wrote:
> This is v3 of the series first posted at:
> https://sourceware.org/ml/gdb-patches/2015-04/msg00073.html
> v2 here:
> https://sourceware.org/ml/gdb-patches/2015-04/msg00198.html
>
> Compared to v2, this version:
>
> - Fixes problems on software single-step targets, first exposed by
> Yao's ARM testing.
>
> - Makes the per-inferior displaced stepping queue work correctly
> again. v1 and v2 has borked that.
>
> - General cleanups.
>
> A few patches in v2 were pushed to master meanwhile. This one in v2
> was dropped in v3:
>
> https://sourceware.org/ml/gdb-patches/2015-04/msg00207.html
> PPC64: symbol-file + exec-file results in broken displaced stepping
>
> because we now don't trip on that issue anymore, since this went in:
>
> https://sourceware.org/ml/gdb-patches/2015-04/msg00341.html
> [PATCH][AT_ENTRY version] PR13858 - Can't do displaced stepping with no symbols
>
> The bug is still present, but is now latent, so I'll probably resend
> that one out separately at some point.
>
> I tried to mention the differences to v2 in each patch that had
> changes, if any (scroll past ChangeLog).
>
> Documentation changes have not changed, and have already been
> approved.
>
> Note most of these patches can go in without the final bits that
> actually do all-stop-on-top-of-non-stop, as they fix problems that
> exist e.g., with "set non-stop on" or "set displaced-stepping on" too.
>
> I've (force) pushed this to users/palves/all-stop-non-stop for review
> and testing convenience.
>
> Tested on:
>
> x86-64 GNU/Linux
> x86-64 GNU/Linux on top of software single-step branch
> PPC64 GNU/Linux
> S/390 GNU/Linux
>
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> Series-intro proper follows:
>
> This series implements user-visible all-stop mode running with the
> target_ops backend always in non-stop mode. This is a stepping stone
> towards finer-grained control of threads, being able to do interesting
> things like inferior and thread groups/sets (itsets), associating
> groups with breakpoints (which threads cause a stop, which threads are
> implicitly paused when a breakpoint triggers), etc. From the user's
> perspective, all-stop mode is really just a special case of being able
> to stop and resume specific sets of threads, so it makes sense to do
> this step first.
>
> After this series, even in all-stop mode ("set non-stop off", the
> default) the target is no longer in charge of stopping all threads
> before reporting an event to the core -- the core takes care of it
> when it sees fit. For example, when "next"- or "step"-ing, we can
> avoid stopping and resuming all threads at each internal single-step,
> and instead only stop all threads when we're about to present the stop
> to the user.
>
> In order to get there, the series teaches GDB to do non-stop mode even
> without displaced stepping (stop all threads, step over breakpoint,
> restart threads), as displaced stepping isn't implemented everywhere,
> and doesn't work in some scenarios (as in, we can't use it, not that
> it's buggy).
>
> And then it fixes all testsuite regressions (on native x86-64, ppc64
> and s390 GNU/Linux) this new mode causes, compared to all-stop ("set
> non-stop off") with the target backend in all-stop mode too (i.e., the
> current default). Making "target remote" work in always non-stop mode
> is deferred for later (this will be even more useful for remote as the
> all-stop mode RSP variant can't really do asynchronous debugging).
>
> Tested on x86_64 Fedora 20, native, with and without "set displaced
> off", and with and without "maint set target-non-stop on"; and also
> against gdbserver. Tested on x86-64 GNU/Linux native on top of x86
> software single-step branch. Tested on PPC64 Fedora 18. Tested on
> S/390 RHEL 7.1.
>
> Pedro Alves (17):
> Fix and test "checkpoint" in non-stop mode
> Change adjust_pc_after_break's prototype
> remote.c/all-stop: Implement TARGET_WAITKIND_NO_RESUMED and
> TARGET_WNOHANG
> Make thread_still_needs_step_over consider stepping_over_watchpoint
> too
> Embed the pending step-over chain in thread_info objects
> Use keep_going in proceed and start_step_over too
> Misc switch_back_to_stepped_thread cleanups
> Factor out code to re-resume stepped thread
> Teach non-stop to do in-line step-overs (stop all, step, restart)
> Implement all-stop on top of a target running non-stop mode
> Fix signal-while-stepping-over-bp-other-thread.exp on targets always
> in non-stop
> Fix interrupt-noterm.exp on targets always in non-stop
> Fix step-over-{trips-on-watchpoint|lands-on-breakpoint}.exp race
> Disable displaced stepping if trying it fails
> PPC64: Fix gdb.arch/ppc64-atomic-inst.exp with displaced stepping
> S/390: displaced stepping and PC-relative RIL-b/RIL-c instructions
> native Linux: enable always non-stop by default
>
> gdb/NEWS | 6 +
> gdb/breakpoint.c | 11 +-
> gdb/darwin-nat.c | 6 +-
> gdb/doc/gdb.texinfo | 24 +
> gdb/gdbarch.h | 6 +-
> gdb/gdbarch.sh | 4 +
> gdb/gdbthread.h | 60 +-
> gdb/gnu-nat.c | 7 -
> gdb/inf-ptrace.c | 6 +-
> gdb/infcmd.c | 6 +-
> gdb/infrun.c | 2290 +++++++++++++++-----
> gdb/infrun.h | 7 +
> gdb/linux-nat.c | 33 +-
> gdb/monitor.c | 4 +-
> gdb/nto-procfs.c | 14 +-
> gdb/procfs.c | 4 +-
> gdb/remote-m32r-sdi.c | 6 +-
> gdb/remote-sim.c | 20 +-
> gdb/remote.c | 57 +-
> gdb/rs6000-tdep.c | 68 +-
> gdb/s390-linux-tdep.c | 115 +-
> gdb/target-delegates.c | 59 +
> gdb/target.c | 83 +
> gdb/target.h | 24 +-
> gdb/target/waitstatus.h | 5 +-
> gdb/testsuite/gdb.arch/ppc64-atomic-inst.exp | 50 +-
> gdb/testsuite/gdb.base/checkpoint-ns.exp | 27 +
> gdb/testsuite/gdb.base/checkpoint.exp | 31 +-
> gdb/testsuite/gdb.base/valgrind-disp-step.c | 32 +
> gdb/testsuite/gdb.base/valgrind-disp-step.exp | 131 ++
> .../signal-while-stepping-over-bp-other-thread.exp | 2 +-
> .../gdb.threads/step-over-lands-on-breakpoint.c | 17 +-
> .../gdb.threads/step-over-lands-on-breakpoint.exp | 6 +-
> .../gdb.threads/step-over-trips-on-watchpoint.c | 17 +-
> .../gdb.threads/step-over-trips-on-watchpoint.exp | 9 +-
> gdb/thread.c | 197 +-
> gdb/windows-nat.c | 6 +-
> 37 files changed, 2741 insertions(+), 709 deletions(-)
> create mode 100644 gdb/testsuite/gdb.base/checkpoint-ns.exp
> create mode 100644 gdb/testsuite/gdb.base/valgrind-disp-step.c
> create mode 100644 gdb/testsuite/gdb.base/valgrind-disp-step.exp
FYI, I ran the Eclipse CDT testsuite using this branch and it's all green. :)
Simon
^ permalink raw reply [flat|nested] 62+ messages in thread