This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
[PATCH] PR15693 - Fix spurious *running events, thread state, dprintf-style call
- From: Pedro Alves <palves at redhat dot com>
- To: Hui Zhu <teawater at gmail dot com>
- Cc: Tom Tromey <tromey at redhat dot com>, gdb-patches ml <gdb-patches at sourceware dot org>, Marc Khouzam <marc dot khouzam at ericsson dot com>
- Date: Fri, 16 May 2014 01:56:20 +0100
- Subject: [PATCH] PR15693 - Fix spurious *running events, thread state, dprintf-style call
- Authentication-results: sourceware.org; auth=none
- References: <CANFwon35XMxAGHxgW5ubhome5oLgxGNt7MCRYSb41nn2xh5dzQ at mail dot gmail dot com> <87ehahnp93 dot fsf at fleche dot redhat dot com> <CANFwon3vYcDDYsw99JtggMVD=kmHyyz9Jqby=Kikq03PdWzU0w at mail dot gmail dot com>
On 07/31/2013 09:15 AM, Hui Zhu wrote:
> On Tue, Jul 30, 2013 at 3:24 AM, Tom Tromey <tromey@redhat.com> wrote:
>>>>>>> "Hui" == Hui Zhu <teawater@gmail.com> writes:
>>
>> Hui> 2013-07-08 Hui Zhu <hui@codesourcery.com>
>> Hui> PR gdb/15693
>> Hui> infcall.c (run_inferior_call): Save value of call_thread->state
>> Hui> and set it back.
>>
>> I didn't see a response to this.
>> I think this needs a test case.
>>
>> Tom
>
> Hi Tom,
>
> I make a test for it.
Sorry that it took _so_ long to get this reviewed/fixed.
But better late than never...
I was going to approve this this morning, with a few tweaks
(the test as is is racy and doesn't actually fail with
an unfixed GDB), but then I realized that the fix isn't right
in case the inferior has multiple threads. If I run the test
in the patch below against your patch, I get this ...
-exec-continue
^running
*running,thread-id="all"
(gdb)
*running,thread-id="all"
*running,thread-id="all"
*running,thread-id="all"
*running,thread-id="all"
*running,thread-id="all"
*running,thread-id="all"
*running,thread-id="all"
*running,thread-id="all"
*running,thread-id="all"
*running,thread-id="all"
=breakpoint-modified,bkpt={number="3",type="breakpoint",disp="keep",enabled="y",addr="0x000000000040086a",func="test",file="../../../src/gdb/testsuite/gdb.mi/mi-condbreak-c
all-thr-state.c",fullname="/home/pedro/gdb/mygit/src/gdb/testsuite/gdb.mi/mi-condbreak-call-thr-state.c",line="32",thread-groups=["i1"],times="1",original-location="mi-cond
break-call-thr-state.c:32"}
*stopped,reason="breakpoint-hit",disp="keep",bkptno="3",frame={addr="0x000000000040086a",func="test",args=[],file="../../../src/gdb/testsuite/gdb.mi/mi-condbreak-call-thr-s
tate.c",fullname="/home/pedro/gdb/mygit/src/gdb/testsuite/gdb.mi/mi-condbreak-call-thr-state.c",line="32"},thread-id="1",stopped-threads="all",core="2"
FAIL: gdb.mi/mi-condbreak-call-thr-state.exp: mt: no spurious *running notifications
> 2013-07-31 Hui Zhu <hui@codesourcery.com>
>
> PR gdb/15693
> * infcall.c (run_inferior_call): Save value of call_thread->state
> and set it back.
... because this only saves/restores the state of _one_ thread.
So even though the current thread's state is reset to "running", the
other threads end up marked as stopped, and so the next resume
(because the breakpoint doesn't cause a stop) emits *running events
anyway...
So I thought about this a lot today, and come up with the
patch below.
Let me know what you think.
8<-------
Subject: [PATCH] PR15693 - Fix spurious *running events, thread state,
dprintf-style call
If one sets a breakpoint with a condition that involves calling a
function in the inferior, and then the condition evaluates false, GDB
outputs one *running event for each time the program hits the
breakpoint. E.g.,
$ gdb return-false -i=mi
(gdb)
start
...
(gdb)
b 14 if return_false ()
&"b 14 if return_false ()\n"
~"Breakpoint 2 at 0x4004eb: file return-false.c, line 14.\n"
...
^done
(gdb)
c
&"c\n"
~"Continuing.\n"
^running
*running,thread-id="1"
(gdb)
*running,thread-id="1"
*running,thread-id="1"
*running,thread-id="1"
*running,thread-id="1"
*running,thread-id="1"
*running,thread-id="1"
... repeat forever ...
An easy way a user can trip on this is with a dprintf with
dprintf-style set to call. In that case, dprintf calls a function in
the inferior, and the resumes, just like the case above.
If the breakpoint/dprintf is set in a loop, then these spurious events
can potentially slows down a frontend much, if it decides to refresh
its GUI whenever it sees this event.
When we run an infcall, we pretend we don't actually run the inferior.
This is already handled for the usual case of calling a function
directly from the CLI:
(gdb)
p return_false ()
&"p return_false ()\n"
~"$1 = 0"
~"\n"
^done
(gdb)
Note no *running, not *stopped events. That's handled by:
static void
mi_on_resume (ptid_t ptid)
{
...
/* Suppress output while calling an inferior function. */
if (tp->control.in_infcall)
return;
and equivalent code on normal_stop.
However, in the cases of the PR, after finishing the infcall there's
one more resume, and mi_on_resume doesn't know that that should be
suppressed too, somehow.
The "running/stopped" state is a high level user/frontend state.
Internal stops are invisible to the frontend. If follows from that
that we should be setting the thread to running at a higher level
where we still know the set of threads the user _intends_ to resume.
Currently we mark a thread as running from within target_resume, a low
level target operation. As consequence, today, if we resume a
multi-threaded program while stopped at a breakpoint, we see this:
-exec-continue
^running
*running,thread-id="1"
(gdb)
*running,thread-id="all"
The first *running was GDB stepping over the breakpoint, and the
second is GDB finally resuming everything.
Between those two *running's, threads other than "1" have bogus still
have their state set to stopped. That bogus -- in async mode, this
opens a tiny window between both resumes where the user might try to
run another execution command to threads other than thread 1, and very
much confuse GDB.
That is, the "step" below should fail the "step", complaining that the
thread is running:
(gdb) c -a &
(gdb) thread 2
(gdb) step
IOW, threads that GDB happens to not resume immediately (say, because
it needs to step over a breakpoint) shall still be marked as running.
Then, if we move marking threads as running to a higher layer,
decoupled from target_resume , and also skip marking threads as
running when running an infcall, the spurious *running events
disappear.
I think we might end up adding a new thread state -- THREAD_INFCALL or
some such, however since infcalls are always synchronous today, I
didn't find a need. There's no way to execute a CLI/MI command
directly from the prompt if some thread is running an infcall.
Tested on x86_64 Fedora 20.
gdb/
2014-05-16 Pedro Alves <palves@redhat.com>
PR PR15693
* infrun.c (resume): Determine how much to resume depending on
whether the caller wanted a step, not whether we can hardware step
the target. Mark all threads that we intend to run as running,
unless we're calling an inferior function.
(normal_stop): If the thread is running an infcall, don't finish
thread state.
* target.c (target_resume): Don't mark threads as running here.
gdb/testsuite/
2014-05-16 Pedro Alves <palves@redhat.com>
Hui Zhu <hui@codesourcery.com>
PR PR15693
* gdb.mi/mi-condbreak-call-thr-state-mt.c: New file.
* gdb.mi/mi-condbreak-call-thr-state-st.c: New file.
* gdb.mi/mi-condbreak-call-thr-state.c: New file.
* gdb.mi/mi-condbreak-call-thr-state.exp: New file.
---
gdb/infrun.c | 44 ++++++--
gdb/target.c | 3 +-
.../gdb.mi/mi-condbreak-call-thr-state-mt.c | 61 +++++++++++
.../gdb.mi/mi-condbreak-call-thr-state-st.c | 26 +++++
gdb/testsuite/gdb.mi/mi-condbreak-call-thr-state.c | 33 ++++++
.../gdb.mi/mi-condbreak-call-thr-state.exp | 116 +++++++++++++++++++++
6 files changed, 271 insertions(+), 12 deletions(-)
create mode 100644 gdb/testsuite/gdb.mi/mi-condbreak-call-thr-state-mt.c
create mode 100644 gdb/testsuite/gdb.mi/mi-condbreak-call-thr-state-st.c
create mode 100644 gdb/testsuite/gdb.mi/mi-condbreak-call-thr-state.c
create mode 100644 gdb/testsuite/gdb.mi/mi-condbreak-call-thr-state.exp
diff --git a/gdb/infrun.c b/gdb/infrun.c
index 597a188..2e44fef 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -1775,6 +1775,7 @@ resume (int step, enum gdb_signal sig)
CORE_ADDR pc = regcache_read_pc (regcache);
struct address_space *aspace = get_regcache_aspace (regcache);
ptid_t resume_ptid;
+ int hw_step = step;
QUIT;
@@ -1794,7 +1795,7 @@ resume (int step, enum gdb_signal sig)
if (debug_infrun)
fprintf_unfiltered (gdb_stdlog,
"infrun: resume : clear step\n");
- step = 0;
+ hw_step = 0;
}
if (debug_infrun)
@@ -1839,7 +1840,7 @@ a command like `return' or `jump' to continue execution."));
step software breakpoint. */
if (use_displaced_stepping (gdbarch)
&& (tp->control.trap_expected
- || (step && gdbarch_software_single_step_p (gdbarch)))
+ || (hw_step && gdbarch_software_single_step_p (gdbarch)))
&& sig == GDB_SIGNAL_0
&& !current_inferior ()->waiting_for_vfork_done)
{
@@ -1849,11 +1850,14 @@ a command like `return' or `jump' to continue execution."));
{
/* 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 frontend point of view,
- the thread is running. */
- set_running (inferior_ptid, 1);
+ 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 frontend
+ point of view, threads were set 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 (user_visible_resume_ptid (step), 1);
discard_cleanups (old_cleanups);
return;
}
@@ -1863,8 +1867,8 @@ a command like `return' or `jump' to continue execution."));
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);
+ hw_step = gdbarch_displaced_step_hw_singlestep (gdbarch,
+ displaced->step_closure);
}
/* Do we need to do it the hard way, w/temp breakpoints? */
@@ -1928,6 +1932,14 @@ a command like `return' or `jump' to continue execution."));
by applying increasingly restricting conditions. */
resume_ptid = user_visible_resume_ptid (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 || singlestep_breakpoints_inserted_p)
&& tp->control.trap_expected)
@@ -6172,8 +6184,18 @@ normal_stop (void)
if (has_stack_frames () && !stop_stack_dummy)
set_current_sal_from_frame (get_current_frame (), 1);
- /* Let the user/frontend see the threads as stopped. */
- do_cleanups (old_chain);
+ /* Let the user/frontend see the threads as stopped, but do nothing
+ if the thread was running an infcall. We may be e.g., evaluating
+ a breakpoint condition. In that case, the thread had state
+ THREAD_RUNNING before the infcall, and shall remain set to
+ running, all without informing the user/frontend about state
+ transition changes. If this is actually a call command, then the
+ thread was originally already stopped, so there's no state to
+ finish either. */
+ if (target_has_execution && inferior_thread ()->control.in_infcall)
+ discard_cleanups (old_chain);
+ else
+ do_cleanups (old_chain);
/* Look up the hook_stop and run it (CLI internally handles problem
of stop_command's pre-hook not existing). */
diff --git a/gdb/target.c b/gdb/target.c
index 1b48f79..c99b9c7 100644
--- a/gdb/target.c
+++ b/gdb/target.c
@@ -2149,8 +2149,9 @@ target_resume (ptid_t ptid, int step, enum gdb_signal signal)
gdb_signal_to_name (signal));
registers_changed_ptid (ptid);
+ /* We only set the internal executing state here. The user/frontend
+ running state is set at a higher level. */
set_executing (ptid, 1);
- set_running (ptid, 1);
clear_inline_frame_state (ptid);
}
diff --git a/gdb/testsuite/gdb.mi/mi-condbreak-call-thr-state-mt.c b/gdb/testsuite/gdb.mi/mi-condbreak-call-thr-state-mt.c
new file mode 100644
index 0000000..112a5cb
--- /dev/null
+++ b/gdb/testsuite/gdb.mi/mi-condbreak-call-thr-state-mt.c
@@ -0,0 +1,61 @@
+/* This testcase is part of GDB, the GNU debugger.
+
+ Copyright (C) 2014 Free Software Foundation, Inc.
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 3 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program. If not, see <http://www.gnu.org/licenses/>. */
+
+/* This is the multi-threaded driver for the real test. */
+
+#include <stdlib.h>
+#include <pthread.h>
+
+extern int test (void);
+
+#define NTHREADS 5
+pthread_barrier_t barrier;
+
+void *
+thread_func (void *arg)
+{
+ pthread_barrier_wait (&barrier);
+
+ while (1)
+ sleep (1);
+}
+
+void
+create_thread (void)
+{
+ pthread_t tid;
+
+ if (pthread_create (&tid, NULL, thread_func, NULL))
+ {
+ perror ("pthread_create");
+ exit (1);
+ }
+}
+
+int
+main (int argc, char *argv[])
+{
+ int i;
+
+ pthread_barrier_init (&barrier, NULL, NTHREADS + 1);
+
+ for (i = 0; i < NTHREADS; i++)
+ create_thread ();
+ pthread_barrier_wait (&barrier);
+
+ return test ();
+}
diff --git a/gdb/testsuite/gdb.mi/mi-condbreak-call-thr-state-st.c b/gdb/testsuite/gdb.mi/mi-condbreak-call-thr-state-st.c
new file mode 100644
index 0000000..66335dd
--- /dev/null
+++ b/gdb/testsuite/gdb.mi/mi-condbreak-call-thr-state-st.c
@@ -0,0 +1,26 @@
+/* This testcase is part of GDB, the GNU debugger.
+
+ Copyright (C) 2014 Free Software Foundation, Inc.
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 3 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program. If not, see <http://www.gnu.org/licenses/>. */
+
+/* This is single-threaded driver for the real test. */
+
+extern int test (void);
+
+int
+main ()
+{
+ return test ();
+}
diff --git a/gdb/testsuite/gdb.mi/mi-condbreak-call-thr-state.c b/gdb/testsuite/gdb.mi/mi-condbreak-call-thr-state.c
new file mode 100644
index 0000000..75d5601
--- /dev/null
+++ b/gdb/testsuite/gdb.mi/mi-condbreak-call-thr-state.c
@@ -0,0 +1,33 @@
+/* This testcase is part of GDB, the GNU debugger.
+
+ Copyright (C) 2013-2014 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/>. */
+
+int
+return_false (void)
+{
+ return 0;
+}
+
+int
+test (void)
+{
+ int a = 0;
+
+ while (a < 10)
+ a++; /* set breakpoint here */
+
+ return 0; /* set end breakpoint here */
+}
diff --git a/gdb/testsuite/gdb.mi/mi-condbreak-call-thr-state.exp b/gdb/testsuite/gdb.mi/mi-condbreak-call-thr-state.exp
new file mode 100644
index 0000000..82ca6cb
--- /dev/null
+++ b/gdb/testsuite/gdb.mi/mi-condbreak-call-thr-state.exp
@@ -0,0 +1,116 @@
+# Copyright (C) 2013-2014 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/>.
+
+# Regression test for PR15693. A breakpoint with a condition that
+# calls a function that evaluates false would result in a spurious
+# *running event sent to the frontend each time the breakpoint is hit
+# (and the target re-resumed). Like:
+#
+# -exec-continue
+# ^running
+# *running,thread-id="all"
+# (gdb)
+# *running,thread-id="1"
+# *running,thread-id="1"
+# *running,thread-id="1"
+# *running,thread-id="1"
+# *running,thread-id="1"
+# ...
+
+load_lib mi-support.exp
+set MIFLAGS "-i=mi"
+
+# Run either the multi-threaded or the single-threaded variant of the
+# test, as determined by VARIANT.
+proc test { variant } {
+ global gdb_test_file_name
+ global testfile srcdir subdir srcfile srcfile2 binfile
+ global mi_gdb_prompt async
+
+ with_test_prefix "$variant" {
+ gdb_exit
+ if [mi_gdb_start] {
+ continue
+ }
+
+ set options {debug}
+ if {$variant == "mt" } {
+ lappend options "pthreads"
+ }
+
+ # Don't use standard_testfile as we need a different binary
+ # for each variant.
+ set testfile $gdb_test_file_name-$variant
+ set binfile [standard_output_file ${testfile}]
+ set srcfile $testfile.c
+ set srcfile2 $gdb_test_file_name.c
+
+ if {[build_executable "failed to prepare" \
+ $testfile \
+ "${srcfile} ${srcfile2}" \
+ $options] == -1} {
+ return -1
+ }
+
+ mi_delete_breakpoints
+ mi_gdb_reinitialize_dir $srcdir/$subdir
+ mi_gdb_reinitialize_dir $srcdir/$subdir
+ mi_gdb_load ${binfile}
+
+ mi_runto test
+
+ # Leave the breakpoint at test set, on purpose. The next
+ # resume shall emit a single '*running,thread-id="all"', even
+ # if GDB needs to step over a breakpoint (that is, even if GDB
+ # needs to run only one thread for a little bit).
+
+ set bp_location [gdb_get_line_number "set breakpoint here" $srcfile2]
+ set bp_location_end [gdb_get_line_number "set end breakpoint here" $srcfile2]
+
+ mi_gdb_test "-break-insert -c return_false() $srcfile2:$bp_location" ".*" \
+ "insert conditional breakpoint"
+
+ mi_gdb_test "-break-insert $srcfile2:$bp_location_end" ".*" \
+ "insert end breakpoint"
+
+ set msg "no spurious *running notifications"
+ send_gdb "-exec-continue\n"
+ gdb_expect {
+ -re "\\*running.*\\*running.*\\*stopped" {
+ fail $msg
+ }
+ -re "\\^running\r\n\\*running,thread-id=\"all\"\r\n${mi_gdb_prompt}.*\\*stopped" {
+ pass $msg
+ }
+ timeout {
+ fail "$msg (timeout)"
+ }
+ }
+
+ # In sync mode, there's an extra prompt after *stopped. Consume it.
+ if {!$async} {
+ gdb_expect {
+ -re "$mi_gdb_prompt" {
+ }
+ }
+ }
+ }
+}
+
+# Single-threaded.
+test "st"
+
+# Multi-threaded.
+test "mt"
--
1.9.0