This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
[PATCH] Do not respawn signals, take 2.
- From: Pedro Alves <palves at redhat dot com>
- To: gdb-patches at sourceware dot org
- Cc: jan dot kratochvil at redhat dot com
- Date: Fri, 22 Jun 2012 15:55:25 +0100
- Subject: [PATCH] Do not respawn signals, take 2.
Long story short, a test I wrote for something else was failing
occasionally. Eventually, after a long, LONG, session of head
banging, I traced it to this bit in glibc (nptl/nptl-init.c):
/* For asynchronous cancellation we use a signal. This is the handler. */
static void
sigcancel_handler (int sig, siginfo_t *si, void *ctx)
{
...
/* Safety check. It would be possible to call this function for
other signals and send a signal from another process. This is not
correct and might even be a security problem. Try to catch as
many incorrect invocations as possible. */
if (sig != SIGCANCEL
#ifdef __ASSUME_CORRECT_SI_PID
/* Kernels before 2.5.75 stored the thread ID and not the process
ID in si_pid so we skip this test. */
|| si->si_pid != pid
#endif
|| si->si_code != SI_TKILL)
return;
The "si->si_pid != pid" check is what triggered the issue. In that
new test, a thread happens to cancel another with a SIGCANCEL (glibc
internals) as GDB is trying to stop all threads. GDB catches the
SIGCANCEL in stop_wait_callback, sets it aside, and keeps fetching
signals until it sees the SIGSTOP. Later, when GDB finaly re-delivers
the SIGCANCEL to the inferior, the signal's si_pid is set to GDB's pid
instead of the inferior's.
We had discussed this back in 2010, around:
http://sourceware.org/ml/gdb-patches/2010-09/msg00360.html
I took that patch, and noticed that the new sigstep-threads.exp test
failed on gdbserver, because it didn't have that new "fetch SIGTRAP if
thread was stepping" bit in the 2010 patch.
Looking further at the "[rfc] linux-nat: Never PTRACE_CONT a stepping
thread" follow up to that series,
http://sourceware.org/ml/gdb-patches/2010-09/msg00362.html
that can't work as is for software single-step targets, as on those we
can't just set "lwp->step" to make an lwp software single-step.
So I took a step back and re-analyzed the issue.
The main issue is that when infrun.c sees a spurious signal that is in
"pass" state, it needs to re-forward that signal back to the inferior.
This is the end of the
/* For the program's own signals, act according to
the signal handling tables. */
if (ecs->random_signal)
block. And it does that with a keep_going call. If scheduler-locking
is not in effect, that delivers the signal to the inferior, but, also
continues all other threads, even if one of them should be stepped
instead, because we were "step"ing or "next"ing, etc. some other
thread.
(This could be seen as a limitation of the target_resume interface.
There's no way to step one thread, deliver a signal to another, and
resume all others with a single call. We can't issue multiple
target_resume calls without an intervening target_wait, due to how the
regular remote protocol is designed. The remote target, with vCont,
does support this though, allowing multiple actions in the same
packet.)
So let's tackly the lost step issue first. This change:
diff --git a/gdb/infrun.c b/gdb/infrun.c
index 53db335..3e8dbc8 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -4363,10 +4363,8 @@ process_event_stop_test:
(leaving the inferior at the step-resume-breakpoint without
actually executing it). Either way continue until the
breakpoint is really hit. */
- keep_going (ecs);
- return;
}
-
+ else
/* Handle cases caused by hitting a breakpoint. */
{
(I'll reindent this before checking in)
means that instead of calling keep_going immediately when we have a
spurious signal to forward to the inferior, we reach the
> /* In all-stop mode, if we're currently stepping but have stopped in
> some other thread, we need to switch back to the stepped thread. */
> if (!non_stop)
part, which makes sure we don't PTRACE_CONT a thread that was meant to
be stepping, and do all the necessary handling of software-single
step. All good, we unify the "spurious signal", and "breakpoint that
doesn't cause stop" cases.
However, if we do just that, then we lose the signal that we were
meant to pass back to the inferior, when ultimately we call keep_going
with the stepping thread selected as current. The way to address this
that is not a complete overhaul is to make the target backend
responsible for forwarding a thread's last stop signal back to the
thread, unless it has been cleared or is in "no pass" state. This
isn't actually that alien. It is actually quite natural, as it is
largelly how 'proceed/keep_going' track the signal to be forwarded as
well. (I started out with a new field in struct thread_info, until I
realized this.)
This way, the extra SIGTRAP handling is not necessary, and gdbserver
doesn't need to change either - only remote.c is taught to send more
actions in the vCont (which is valid, and defined) packet. If gdb's
target_resume interface is ever made more like gdbserver's, then those
bits that handle stop_signal in both linux-nat.c and remote.c could be
factored out to common code.
The tests are pretty much unmodified. I only did some cosmetic fixes,
and moved the "signal" call to "main" in sigstep-threads.c, as signal
handlers are per process.
No regressions on x86_64 Fedora 16, native and gdbserver. New tests
pass. Fixes my unsubmitted test as well.
Comments?
gdb/
2012-06-22 Jan Kratochvil <jan.kratochvil@redhat.com>
Pedro Alves <palves@redhat.com>
* gdbthread.h (ALL_THREADS): New macro.
(thread_list): Declare.
* infrun.c (handle_inferior_event) <spurious signal>: Don't keep
going, but instead fall through to the stepping handling.
* linux-nat.c (resume_lwp): New parameter 'signo'. Resume with
the passed in signal. Adjust debug output.
(resume_callback): Rename to ...
(linux_nat_resume_callback): ... this. Pass the thread's last
stop signal, if in "pass" state.
(linux_nat_resume): Adjust to rename.
(stop_wait_callback): New assertion. Don't respawn signals;
instead let the LWP remain with SIGNALLED set.
(linux_nat_wait_1): Remove flushing of pending SIGSTOPs.
gdb/testsuite/
2012-06-22 Jan Kratochvil <jan.kratochvil@redhat.com>
Pedro Alves <palves@redhat.com>
* gdb.threads/siginfo-threads.exp: New file.
* gdb.threads/siginfo-threads.c: New file.
* gdb.threads/sigstep-threads.exp: New file.
* gdb.threads/sigstep-threads.c: New file.
---
gdb/gdbthread.h | 7
gdb/infrun.c | 4
gdb/linux-nat.c | 208 +++--------
gdb/remote.c | 23 +
gdb/target.h | 14 +
gdb/testsuite/gdb.threads/siginfo-threads.c | 457 +++++++++++++++++++++++++
gdb/testsuite/gdb.threads/siginfo-threads.exp | 99 +++++
gdb/testsuite/gdb.threads/sigstep-threads.c | 54 +++
gdb/testsuite/gdb.threads/sigstep-threads.exp | 75 ++++
gdb/thread.c | 2
10 files changed, 782 insertions(+), 161 deletions(-)
create mode 100644 gdb/testsuite/gdb.threads/siginfo-threads.c
create mode 100644 gdb/testsuite/gdb.threads/siginfo-threads.exp
create mode 100644 gdb/testsuite/gdb.threads/sigstep-threads.c
create mode 100644 gdb/testsuite/gdb.threads/sigstep-threads.exp
diff --git a/gdb/gdbthread.h b/gdb/gdbthread.h
index 7cd66b6..0250555 100644
--- a/gdb/gdbthread.h
+++ b/gdb/gdbthread.h
@@ -301,6 +301,11 @@ void thread_change_ptid (ptid_t old_ptid, ptid_t new_ptid);
typedef int (*thread_callback_func) (struct thread_info *, void *);
extern struct thread_info *iterate_over_threads (thread_callback_func, void *);
+/* Traverse all threads. */
+
+#define ALL_THREADS(T) \
+ for (T = thread_list; T; T = T->next)
+
extern int thread_count (void);
/* Switch from one thread to another. */
@@ -391,4 +396,6 @@ extern struct thread_info* inferior_thread (void);
extern void update_thread_list (void);
+extern struct thread_info *thread_list;
+
#endif /* GDBTHREAD_H */
diff --git a/gdb/infrun.c b/gdb/infrun.c
index 53db335..3e8dbc8 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -4363,10 +4363,8 @@ process_event_stop_test:
(leaving the inferior at the step-resume-breakpoint without
actually executing it). Either way continue until the
breakpoint is really hit. */
- keep_going (ecs);
- return;
}
-
+ else
/* Handle cases caused by hitting a breakpoint. */
{
CORE_ADDR jmp_buf_pc;
diff --git a/gdb/linux-nat.c b/gdb/linux-nat.c
index 33b13fa..c0c9584 100644
--- a/gdb/linux-nat.c
+++ b/gdb/linux-nat.c
@@ -1901,7 +1901,7 @@ linux_nat_detach (struct target_ops *ops, char *args, int from_tty)
/* Resume LP. */
static void
-resume_lwp (struct lwp_info *lp, int step)
+resume_lwp (struct lwp_info *lp, int step, enum gdb_signal signo)
{
if (lp->stopped)
{
@@ -1919,14 +1919,18 @@ resume_lwp (struct lwp_info *lp, int step)
{
if (debug_linux_nat)
fprintf_unfiltered (gdb_stdlog,
- "RC: PTRACE_CONT %s, 0, 0 (resuming sibling)\n",
- target_pid_to_str (lp->ptid));
+ "RC: Resuming sibling %s, %s, %s\n",
+ target_pid_to_str (lp->ptid),
+ (signo != GDB_SIGNAL_0
+ ? strsignal (gdb_signal_to_host (signo))
+ : "0"),
+ step ? "step" : "resume");
if (linux_nat_prepare_to_resume != NULL)
linux_nat_prepare_to_resume (lp);
linux_ops->to_resume (linux_ops,
pid_to_ptid (GET_LWP (lp->ptid)),
- step, GDB_SIGNAL_0);
+ step, signo);
lp->stopped = 0;
lp->step = step;
memset (&lp->siginfo, 0, sizeof (lp->siginfo));
@@ -1950,9 +1954,24 @@ resume_lwp (struct lwp_info *lp, int step)
}
static int
-resume_callback (struct lwp_info *lp, void *data)
+linux_nat_resume_callback (struct lwp_info *lp, void *data)
{
- resume_lwp (lp, 0);
+ enum gdb_signal signo = GDB_SIGNAL_0;
+
+ if (lp->stopped)
+ {
+ struct thread_info *thread;
+
+ thread = find_thread_ptid (lp->ptid);
+ if (thread != NULL)
+ {
+ if (signal_pass_state (thread->suspend.stop_signal))
+ signo = thread->suspend.stop_signal;
+ thread->suspend.stop_signal = GDB_SIGNAL_0;
+ }
+ }
+
+ resume_lwp (lp, 0, signo);
return 0;
}
@@ -2059,11 +2078,11 @@ linux_nat_resume (struct target_ops *ops,
}
/* Mark LWP as not stopped to prevent it from being continued by
- resume_callback. */
+ linux_nat_resume_callback. */
lp->stopped = 0;
if (resume_many)
- iterate_over_lwps (ptid, resume_callback, NULL);
+ iterate_over_lwps (ptid, linux_nat_resume_callback, NULL);
/* Convert to something the lower layer understands. */
ptid = pid_to_ptid (GET_LWP (lp->ptid));
@@ -2856,6 +2875,8 @@ stop_wait_callback (struct lwp_info *lp, void *data)
{
int status;
+ gdb_assert (lp->resumed);
+
status = wait_lwp (lp);
if (status == 0)
return 0;
@@ -2881,110 +2902,39 @@ stop_wait_callback (struct lwp_info *lp, void *data)
if (WSTOPSIG (status) != SIGSTOP)
{
- if (linux_nat_status_is_event (status))
- {
- /* If a LWP other than the LWP that we're reporting an
- event for has hit a GDB breakpoint (as opposed to
- some random trap signal), then just arrange for it to
- hit it again later. We don't keep the SIGTRAP status
- and don't forward the SIGTRAP signal to the LWP. We
- will handle the current event, eventually we will
- resume all LWPs, and this one will get its breakpoint
- trap again.
-
- If we do not do this, then we run the risk that the
- user will delete or disable the breakpoint, but the
- thread will have already tripped on it. */
-
- /* Save the trap's siginfo in case we need it later. */
- save_siginfo (lp);
-
- save_sigtrap (lp);
-
- /* Now resume this LWP and get the SIGSTOP event. */
- errno = 0;
- ptrace (PTRACE_CONT, GET_LWP (lp->ptid), 0, 0);
- if (debug_linux_nat)
- {
- fprintf_unfiltered (gdb_stdlog,
- "PTRACE_CONT %s, 0, 0 (%s)\n",
- target_pid_to_str (lp->ptid),
- errno ? safe_strerror (errno) : "OK");
-
- fprintf_unfiltered (gdb_stdlog,
- "SWC: Candidate SIGTRAP event in %s\n",
- target_pid_to_str (lp->ptid));
- }
- /* Hold this event/waitstatus while we check to see if
- there are any more (we still want to get that SIGSTOP). */
- stop_wait_callback (lp, NULL);
-
- /* Hold the SIGTRAP for handling by linux_nat_wait. If
- there's another event, throw it back into the
- queue. */
- if (lp->status)
- {
- if (debug_linux_nat)
- fprintf_unfiltered (gdb_stdlog,
- "SWC: kill %s, %s\n",
- target_pid_to_str (lp->ptid),
- status_to_str ((int) status));
- kill_lwp (GET_LWP (lp->ptid), WSTOPSIG (lp->status));
- }
+ /* The thread was stopped with a signal other than SIGSTOP. */
- /* Save the sigtrap event. */
- lp->status = status;
- return 0;
- }
- else
- {
- /* The thread was stopped with a signal other than
- SIGSTOP, and didn't accidentally trip a breakpoint. */
+ /* Save the trap's siginfo in case we need it later. */
+ save_siginfo (lp);
- if (debug_linux_nat)
- {
- fprintf_unfiltered (gdb_stdlog,
- "SWC: Pending event %s in %s\n",
- status_to_str ((int) status),
- target_pid_to_str (lp->ptid));
- }
- /* Now resume this LWP and get the SIGSTOP event. */
- errno = 0;
- ptrace (PTRACE_CONT, GET_LWP (lp->ptid), 0, 0);
- if (debug_linux_nat)
- fprintf_unfiltered (gdb_stdlog,
- "SWC: PTRACE_CONT %s, 0, 0 (%s)\n",
- target_pid_to_str (lp->ptid),
- errno ? safe_strerror (errno) : "OK");
-
- /* Hold this event/waitstatus while we check to see if
- there are any more (we still want to get that SIGSTOP). */
- stop_wait_callback (lp, NULL);
-
- /* If the lp->status field is still empty, use it to
- hold this event. If not, then this event must be
- returned to the event queue of the LWP. */
- if (lp->status)
- {
- if (debug_linux_nat)
- {
- fprintf_unfiltered (gdb_stdlog,
- "SWC: kill %s, %s\n",
- target_pid_to_str (lp->ptid),
- status_to_str ((int) status));
- }
- kill_lwp (GET_LWP (lp->ptid), WSTOPSIG (status));
- }
- else
- lp->status = status;
- return 0;
- }
+ save_sigtrap (lp);
+
+ if (debug_linux_nat)
+ fprintf_unfiltered (gdb_stdlog,
+ "SWC: Pending event %s in %s\n",
+ status_to_str ((int) status),
+ target_pid_to_str (lp->ptid));
+
+ /* Save the sigtrap event. */
+ lp->status = status;
+ gdb_assert (!lp->stopped);
+ gdb_assert (lp->signalled);
+ lp->stopped = 1;
}
else
{
/* We caught the SIGSTOP that we intended to catch, so
there's no SIGSTOP pending. */
+
+ if (debug_linux_nat)
+ fprintf_unfiltered (gdb_stdlog,
+ "SWC: Delayed SIGSTOP caught for %s.\n",
+ target_pid_to_str (lp->ptid));
+
lp->stopped = 1;
+
+ /* Reset SIGNALLED only after the stop_wait_callback call
+ above as it does gdb_assert on SIGNALLED. */
lp->signalled = 0;
}
}
@@ -3238,7 +3188,7 @@ stop_and_resume_callback (struct lwp_info *lp, void *data)
fprintf_unfiltered (gdb_stdlog,
"SARC: re-resuming LWP %ld\n",
GET_LWP (lp->ptid));
- resume_lwp (lp, lp->step);
+ resume_lwp (lp, lp->step, GDB_SIGNAL_0);
}
else
{
@@ -3612,54 +3562,6 @@ retry:
lp = NULL;
}
- if (lp && lp->signalled && lp->last_resume_kind != resume_stop)
- {
- /* A pending SIGSTOP may interfere with the normal stream of
- events. In a typical case where interference is a problem,
- we have a SIGSTOP signal pending for LWP A while
- single-stepping it, encounter an event in LWP B, and take the
- pending SIGSTOP while trying to stop LWP A. After processing
- the event in LWP B, LWP A is continued, and we'll never see
- the SIGTRAP associated with the last time we were
- single-stepping LWP A. */
-
- /* Resume the thread. It should halt immediately returning the
- pending SIGSTOP. */
- registers_changed ();
- if (linux_nat_prepare_to_resume != NULL)
- linux_nat_prepare_to_resume (lp);
- linux_ops->to_resume (linux_ops, pid_to_ptid (GET_LWP (lp->ptid)),
- lp->step, GDB_SIGNAL_0);
- if (debug_linux_nat)
- fprintf_unfiltered (gdb_stdlog,
- "LLW: %s %s, 0, 0 (expect SIGSTOP)\n",
- lp->step ? "PTRACE_SINGLESTEP" : "PTRACE_CONT",
- target_pid_to_str (lp->ptid));
- lp->stopped = 0;
- gdb_assert (lp->resumed);
-
- /* Catch the pending SIGSTOP. */
- status = lp->status;
- lp->status = 0;
-
- stop_wait_callback (lp, NULL);
-
- /* If the lp->status field isn't empty, we caught another signal
- while flushing the SIGSTOP. Return it back to the event
- queue of the LWP, as we already have an event to handle. */
- if (lp->status)
- {
- if (debug_linux_nat)
- fprintf_unfiltered (gdb_stdlog,
- "LLW: kill %s, %s\n",
- target_pid_to_str (lp->ptid),
- status_to_str (lp->status));
- kill_lwp (GET_LWP (lp->ptid), WSTOPSIG (lp->status));
- }
-
- lp->status = status;
- }
-
if (!target_can_async_p ())
{
/* Causes SIGINT to be passed on to the attached process. */
diff --git a/gdb/remote.c b/gdb/remote.c
index 38ecd08..a440153 100644
--- a/gdb/remote.c
+++ b/gdb/remote.c
@@ -4644,6 +4644,25 @@ append_resumption (char *p, char *endp,
return p;
}
+/* Send in resumptions recorded in the threads themselves. */
+
+static char *
+append_pending_thread_resumptions (char *p, char *endp, ptid_t ptid)
+{
+ struct thread_info *thread;
+
+ ALL_THREADS (thread)
+ if (ptid_match (thread->ptid, ptid)
+ && thread->suspend.stop_signal != GDB_SIGNAL_0)
+ {
+ p = append_resumption (p, endp, thread->ptid,
+ 0, thread->suspend.stop_signal);
+ thread->suspend.stop_signal = GDB_SIGNAL_0;
+ }
+
+ return p;
+}
+
/* Resume the remote inferior by using a "vCont" packet. The thread
to be resumed is PTID; STEP and SIGGNAL indicate whether the
resumed thread should be single-stepped and/or signalled. If PTID
@@ -4696,6 +4715,10 @@ remote_vcont_resume (ptid_t ptid, int step, enum gdb_signal siggnal)
p = append_resumption (p, endp, inferior_ptid, step, siggnal);
}
+ /* Also pass down any pending resumptions for other threads not
+ the current. */
+ p = append_pending_thread_resumptions (p, endp, ptid);
+
/* And continue others without a signal. */
append_resumption (p, endp, ptid, /*step=*/ 0, GDB_SIGNAL_0);
}
diff --git a/gdb/target.h b/gdb/target.h
index 233d355..e5bdf4d 100644
--- a/gdb/target.h
+++ b/gdb/target.h
@@ -920,10 +920,16 @@ extern void target_detach (char *, int);
extern void target_disconnect (char *, int);
-/* Resume execution of the target process PTID. 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 signal. The caller may not
- pass GDB_SIGNAL_DEFAULT. */
+/* 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
+ 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
+ INFERIOR_PTID, and let other threads (for which the wildcard PTID
+ matches) resume with their 'thread->suspend.stop_signal' signal
+ (usually GDB_SIGNAL_0) if it is in "pass" state, or with no signal
+ if in "no pass" state. */
extern void target_resume (ptid_t ptid, int step, enum gdb_signal signal);
diff --git a/gdb/testsuite/gdb.threads/siginfo-threads.c b/gdb/testsuite/gdb.threads/siginfo-threads.c
new file mode 100644
index 0000000..5859088
--- /dev/null
+++ b/gdb/testsuite/gdb.threads/siginfo-threads.c
@@ -0,0 +1,457 @@
+/* This testcase is part of GDB, the GNU debugger.
+
+ Copyright 2010-2012 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/>. */
+
+#define _GNU_SOURCE
+#include <pthread.h>
+#include <stdio.h>
+#include <limits.h>
+#include <errno.h>
+#include <stdlib.h>
+#include <string.h>
+#include <assert.h>
+#include <sys/types.h>
+#include <signal.h>
+#include <unistd.h>
+#include <asm/unistd.h>
+
+#define gettid() syscall (__NR_gettid)
+#define tgkill(tgid, tid, sig) syscall (__NR_tgkill, tgid, tid, sig)
+
+/* Terminate always in the main task. It can lock up with SIGSTOPped
+ GDB otherwise. */
+#define TIMEOUT (gettid () == getpid() ? 10 : 15)
+
+static pid_t thread1_tid;
+static pthread_cond_t thread1_tid_cond
+ = PTHREAD_COND_INITIALIZER;
+static pthread_mutex_t thread1_tid_mutex
+ = PTHREAD_ERRORCHECK_MUTEX_INITIALIZER_NP;
+static int thread1_sigusr1_hit;
+static int thread1_sigusr2_hit;
+
+static pid_t thread2_tid;
+static pthread_cond_t thread2_tid_cond
+ = PTHREAD_COND_INITIALIZER;
+static pthread_mutex_t thread2_tid_mutex
+ = PTHREAD_ERRORCHECK_MUTEX_INITIALIZER_NP;
+static int thread2_sigusr1_hit;
+static int thread2_sigusr2_hit;
+
+static pthread_mutex_t terminate_mutex
+ = PTHREAD_ERRORCHECK_MUTEX_INITIALIZER_NP;
+
+/* Do not use alarm as it would create a ptrace event which would hang
+ us up if we are being traced by GDB, which we stopped
+ ourselves. */
+
+static void
+timed_mutex_lock (pthread_mutex_t *mutex)
+{
+ int i;
+ struct timespec start, now;
+
+ i = clock_gettime (CLOCK_MONOTONIC, &start);
+ assert (i == 0);
+
+ do
+ {
+ i = pthread_mutex_trylock (mutex);
+ if (i == 0)
+ return;
+ assert (i == EBUSY);
+
+ i = clock_gettime (CLOCK_MONOTONIC, &now);
+ assert (i == 0);
+ assert (now.tv_sec >= start.tv_sec);
+ }
+ while (now.tv_sec - start.tv_sec < TIMEOUT);
+
+ fprintf (stderr, "Timed out waiting for internal lock!\n");
+ exit (EXIT_FAILURE);
+}
+
+static void
+handler (int signo, siginfo_t *siginfo, void *exception)
+{
+ int *varp;
+
+ assert (siginfo->si_signo == signo);
+ assert (siginfo->si_code == SI_TKILL);
+ assert (siginfo->si_pid == getpid ());
+
+ if (gettid () == thread1_tid)
+ {
+ if (signo == SIGUSR1)
+ varp = &thread1_sigusr1_hit;
+ else if (signo == SIGUSR2)
+ varp = &thread1_sigusr2_hit;
+ else
+ assert (0);
+ }
+ else if (gettid () == thread2_tid)
+ {
+ if (signo == SIGUSR1)
+ varp = &thread2_sigusr1_hit;
+ else if (signo == SIGUSR2)
+ varp = &thread2_sigusr2_hit;
+ else
+ assert (0);
+ }
+ else
+ assert (0);
+
+ if (*varp)
+ {
+ fprintf (stderr, "Signal %d for TID %lu has been already hit!\n", signo,
+ (unsigned long) gettid ());
+ exit (EXIT_FAILURE);
+ }
+ *varp = 1;
+}
+
+static void *
+thread1_func (void *unused)
+{
+ int i;
+
+ timed_mutex_lock (&thread1_tid_mutex);
+
+ /* THREAD1_TID_MUTEX must be already locked to avoid a race. */
+ thread1_tid = gettid ();
+
+ i = pthread_cond_signal (&thread1_tid_cond);
+ assert (i == 0);
+ i = pthread_mutex_unlock (&thread1_tid_mutex);
+ assert (i == 0);
+
+ /* Be sure the "t (tracing stop)" test can proceed for both
+ threads. */
+ timed_mutex_lock (&terminate_mutex);
+ i = pthread_mutex_unlock (&terminate_mutex);
+ assert (i == 0);
+
+ if (!thread1_sigusr1_hit)
+ {
+ fprintf (stderr, "Thread 1 signal SIGUSR1 not hit!\n");
+ exit (EXIT_FAILURE);
+ }
+ if (!thread1_sigusr2_hit)
+ {
+ fprintf (stderr, "Thread 1 signal SIGUSR2 not hit!\n");
+ exit (EXIT_FAILURE);
+ }
+
+ return NULL;
+}
+
+static void *
+thread2_func (void *unused)
+{
+ int i;
+
+ timed_mutex_lock (&thread2_tid_mutex);
+
+ /* THREAD2_TID_MUTEX must be already locked to avoid a race. */
+ thread2_tid = gettid ();
+
+ i = pthread_cond_signal (&thread2_tid_cond);
+ assert (i == 0);
+ i = pthread_mutex_unlock (&thread2_tid_mutex);
+ assert (i == 0);
+
+ /* Be sure the "t (tracing stop)" test can proceed for both
+ threads. */
+ timed_mutex_lock (&terminate_mutex);
+ i = pthread_mutex_unlock (&terminate_mutex);
+ assert (i == 0);
+
+ if (!thread2_sigusr1_hit)
+ {
+ fprintf (stderr, "Thread 2 signal SIGUSR1 not hit!\n");
+ exit (EXIT_FAILURE);
+ }
+ if (!thread2_sigusr2_hit)
+ {
+ fprintf (stderr, "Thread 2 signal SIGUSR2 not hit!\n");
+ exit (EXIT_FAILURE);
+ }
+
+ return NULL;
+}
+
+static const char *
+proc_string (const char *filename, const char *line)
+{
+ FILE *f;
+ static char buf[LINE_MAX];
+ size_t line_len = strlen (line);
+
+ f = fopen (filename, "r");
+ if (f == NULL)
+ {
+ fprintf (stderr, "fopen (\"%s\") for \"%s\": %s\n", filename, line,
+ strerror (errno));
+ exit (EXIT_FAILURE);
+ }
+ while (errno = 0, fgets (buf, sizeof (buf), f))
+ {
+ char *s;
+
+ s = strchr (buf, '\n');
+ assert (s != NULL);
+ *s = 0;
+
+ if (strncmp (buf, line, line_len) != 0)
+ continue;
+
+ if (fclose (f))
+ {
+ fprintf (stderr, "fclose (\"%s\") for \"%s\": %s\n", filename, line,
+ strerror (errno));
+ exit (EXIT_FAILURE);
+ }
+
+ return &buf[line_len];
+ }
+ if (errno != 0)
+ {
+ fprintf (stderr, "fgets (\"%s\": %s\n", filename, strerror (errno));
+ exit (EXIT_FAILURE);
+ }
+ fprintf (stderr, "\"%s\": No line \"%s\" found.\n", filename, line);
+ exit (EXIT_FAILURE);
+}
+
+static unsigned long
+proc_ulong (const char *filename, const char *line)
+{
+ const char *s = proc_string (filename, line);
+ long retval;
+ char *end;
+
+ errno = 0;
+ retval = strtol (s, &end, 10);
+ if (retval < 0 || retval >= LONG_MAX || (end && *end))
+ {
+ fprintf (stderr, "\"%s\":\"%s\": %ld, %s\n", filename, line, retval,
+ strerror (errno));
+ exit (EXIT_FAILURE);
+ }
+ return retval;
+}
+
+static void
+state_wait (pid_t process, const char *wanted)
+{
+ char *filename;
+ int i;
+ struct timespec start, now;
+ const char *state;
+
+ i = asprintf (&filename, "/proc/%lu/status", (unsigned long) process);
+ assert (i > 0);
+
+ i = clock_gettime (CLOCK_MONOTONIC, &start);
+ assert (i == 0);
+
+ do
+ {
+ state = proc_string (filename, "State:\t");
+
+ /* torvalds/linux-2.6.git 464763cf1c6df632dccc8f2f4c7e50163154a2c0
+ has changed "T (tracing stop)" to "t (tracing stop)". Make the GDB
+ testcase backward compatible with older Linux kernels. */
+ if (strcmp (state, "T (tracing stop)") == 0)
+ state = "t (tracing stop)";
+
+ if (strcmp (state, wanted) == 0)
+ {
+ free (filename);
+ return;
+ }
+
+ if (sched_yield ())
+ {
+ perror ("sched_yield()");
+ exit (EXIT_FAILURE);
+ }
+
+ i = clock_gettime (CLOCK_MONOTONIC, &now);
+ assert (i == 0);
+ assert (now.tv_sec >= start.tv_sec);
+ }
+ while (now.tv_sec - start.tv_sec < TIMEOUT);
+
+ fprintf (stderr, "Timed out waiting for PID %lu \"%s\" (now it is \"%s\")!\n",
+ (unsigned long) process, wanted, state);
+ exit (EXIT_FAILURE);
+}
+
+static volatile pid_t tracer = 0;
+static pthread_t thread1, thread2;
+
+static void
+cleanup (void)
+{
+ printf ("Resuming GDB PID %lu.\n", (unsigned long) tracer);
+
+ if (tracer)
+ {
+ int i;
+ int tracer_save = tracer;
+
+ tracer = 0;
+
+ i = kill (tracer_save, SIGCONT);
+ assert (i == 0);
+ }
+}
+
+int
+main (int argc, char **argv)
+{
+ int i;
+ int standalone = 0;
+ struct sigaction act;
+
+ if (argc == 2 && strcmp (argv[1], "-s") == 0)
+ standalone = 1;
+ else
+ assert (argc == 1);
+
+ setbuf (stdout, NULL);
+
+ timed_mutex_lock (&thread1_tid_mutex);
+ timed_mutex_lock (&thread2_tid_mutex);
+
+ timed_mutex_lock (&terminate_mutex);
+
+ errno = 0;
+ memset (&act, 0, sizeof (act));
+ act.sa_sigaction = handler;
+ act.sa_flags = SA_RESTART | SA_SIGINFO;
+ i = sigemptyset (&act.sa_mask);
+ assert_perror (errno);
+ assert (i == 0);
+ i = sigaction (SIGUSR1, &act, NULL);
+ assert_perror (errno);
+ assert (i == 0);
+ i = sigaction (SIGUSR2, &act, NULL);
+ assert_perror (errno);
+ assert (i == 0);
+
+ i = pthread_create (&thread1, NULL, thread1_func, NULL);
+ assert (i == 0);
+
+ i = pthread_create (&thread2, NULL, thread2_func, NULL);
+ assert (i == 0);
+
+ if (!standalone)
+ {
+ tracer = proc_ulong ("/proc/self/status", "TracerPid:\t");
+ if (tracer == 0)
+ {
+ fprintf (stderr, "The testcase must be run by GDB!\n");
+ exit (EXIT_FAILURE);
+ }
+ if (tracer != getppid ())
+ {
+ fprintf (stderr, "The testcase parent must be our GDB tracer!\n");
+ exit (EXIT_FAILURE);
+ }
+ }
+
+ /* SIGCONT our debugger in the case of our crash as we would deadlock
+ otherwise. */
+
+ atexit (cleanup);
+
+ printf ("Stopping GDB PID %lu.\n", (unsigned long) tracer);
+
+ if (tracer)
+ {
+ i = kill (tracer, SIGSTOP);
+ assert (i == 0);
+ state_wait (tracer, "T (stopped)");
+ }
+
+ /* Threads are now waiting at timed_mutex_lock (thread1_tid_mutex)
+ and so they could not trigger the signals before GDB is unstopped
+ later. Threads get resumed by the pthread_cond_wait below. Use
+ `while' loops for protection against spurious pthread_cond_wait
+ wakeups. */
+
+ printf ("Waiting till the threads initialize their TIDs.\n");
+
+ while (thread1_tid == 0)
+ {
+ i = pthread_cond_wait (&thread1_tid_cond, &thread1_tid_mutex);
+ assert (i == 0);
+ }
+
+ while (thread2_tid == 0)
+ {
+ i = pthread_cond_wait (&thread2_tid_cond, &thread2_tid_mutex);
+ assert (i == 0);
+ }
+
+ printf ("Thread 1 TID = %lu, thread 2 TID = %lu, PID = %lu.\n",
+ (unsigned long) thread1_tid, (unsigned long) thread2_tid,
+ (unsigned long) getpid ());
+
+ errno = 0;
+ i = tgkill (getpid (), thread1_tid, SIGUSR1);
+ assert_perror (errno);
+ assert (i == 0);
+ i = tgkill (getpid (), thread1_tid, SIGUSR2);
+ assert_perror (errno);
+ assert (i == 0);
+ i = tgkill (getpid (), thread2_tid, SIGUSR1);
+ assert_perror (errno);
+ assert (i == 0);
+ i = tgkill (getpid (), thread2_tid, SIGUSR2);
+ assert_perror (errno);
+ assert (i == 0);
+
+ printf ("Waiting till the threads are trapped by the signals.\n");
+
+ if (tracer)
+ {
+ /* s390x-unknown-linux-gnu will fail with "R (running)". */
+
+ state_wait (thread1_tid, "t (tracing stop)");
+
+ state_wait (thread2_tid, "t (tracing stop)");
+ }
+
+ cleanup ();
+
+ printf ("Joining the threads.\n");
+
+ i = pthread_mutex_unlock (&terminate_mutex);
+ assert (i == 0);
+
+ i = pthread_join (thread1, NULL);
+ assert (i == 0);
+
+ i = pthread_join (thread2, NULL);
+ assert (i == 0);
+
+ printf ("Exiting.\n"); /* break-at-exit */
+
+ return EXIT_SUCCESS;
+}
diff --git a/gdb/testsuite/gdb.threads/siginfo-threads.exp b/gdb/testsuite/gdb.threads/siginfo-threads.exp
new file mode 100644
index 0000000..b4b605f
--- /dev/null
+++ b/gdb/testsuite/gdb.threads/siginfo-threads.exp
@@ -0,0 +1,99 @@
+# Copyright 2010-2012 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/>.
+
+set testfile "siginfo-threads"
+set srcfile ${testfile}.c
+set binfile ${objdir}/${subdir}/${testfile}
+if {[gdb_compile_pthreads "${srcdir}/${subdir}/${srcfile}" \
+ ${binfile} \
+ executable \
+ [list debug additional_flags=-lrt]] != "" } {
+ return -1
+}
+
+clean_restart $testfile
+
+if ![runto_main] {
+ return -1
+}
+
+# `nostop noprint pass' could in some cases report false PASS due to
+# the (preempt 'handle') code path in linux-nat.c.
+
+gdb_test "handle SIGUSR1 stop print pass" \
+ "Signal\[ \t\]+Stop\[ \t\]+Print\[ \t\]+Pass to program\[ \t\]+Description\r\nSIGUSR1\[ \t\]+Yes\[ \t\]+Yes\[ \t\]+Yes\[ \t\].*"
+gdb_test "handle SIGUSR2 stop print pass" \
+ "Signal\[ \t\]+Stop\[ \t\]+Print\[ \t\]+Pass to program\[ \t\]+Description\r\nSIGUSR2\[ \t\]+Yes\[ \t\]+Yes\[ \t\]+Yes\[ \t\].*"
+
+gdb_breakpoint [gdb_get_line_number "break-at-exit"]
+
+set test "get pid"
+gdb_test_multiple "p getpid ()" $test {
+ -re " = (\[0-9\]+)\r\n$gdb_prompt $" {
+ set pid $expect_out(1,string)
+ pass $test
+ }
+}
+
+for {set sigcount 0} {$sigcount < 4} {incr sigcount} {
+ set test "catch signal $sigcount"
+ set sigusr ""
+ gdb_test_multiple "continue" $test {
+ -re "Program received signal SIGUSR(\[12\]), User defined signal \[12\]\\.\r\n.*\r\n$gdb_prompt $" {
+ set sigusr $expect_out(1,string)
+ pass $test
+ }
+ }
+ if {$sigusr == ""} {
+ return -1
+ }
+
+ set test "signal $sigcount si_signo"
+ if {$sigusr == 1} {
+ set signo 10
+ } else {
+ set signo 12
+ }
+ gdb_test_multiple {p $_siginfo.si_signo} $test {
+ -re " = $signo\r\n$gdb_prompt $" {
+ pass $test
+ }
+ -re "Attempt to extract a component of a value that is not a structure\\.\r\n$gdb_prompt $" {
+ unsupported $test
+ }
+ }
+
+ set test "signal $sigcount si_code is SI_TKILL"
+ gdb_test_multiple {p $_siginfo.si_code} $test {
+ -re " = -6\r\n$gdb_prompt $" {
+ pass $test
+ }
+ -re "Attempt to extract a component of a value that is not a structure\\.\r\n$gdb_prompt $" {
+ unsupported $test
+ }
+ }
+
+ set test "signal $sigcount si_pid"
+ gdb_test_multiple {p $_siginfo._sifields._kill.si_pid} $test {
+ -re " = $pid\r\n$gdb_prompt $" {
+ pass $test
+ }
+ -re "Attempt to extract a component of a value that is not a structure\\.\r\n$gdb_prompt $" {
+ unsupported $test
+ }
+ }
+}
+
+gdb_continue_to_breakpoint break-at-exit ".*break-at-exit.*"
diff --git a/gdb/testsuite/gdb.threads/sigstep-threads.c b/gdb/testsuite/gdb.threads/sigstep-threads.c
new file mode 100644
index 0000000..ca77fe4
--- /dev/null
+++ b/gdb/testsuite/gdb.threads/sigstep-threads.c
@@ -0,0 +1,54 @@
+/* This testcase is part of GDB, the GNU debugger.
+
+ Copyright 2010-2012 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/>. */
+
+#include <pthread.h>
+#include <assert.h>
+#include <signal.h>
+
+#include <asm/unistd.h>
+#include <unistd.h>
+#define tgkill(tgid, tid, sig) syscall (__NR_tgkill, (tgid), (tid), (sig))
+#define gettid() syscall (__NR_gettid)
+
+static volatile int var;
+
+static void
+handler (int signo) /* step-0 */
+{ /* step-0 */
+ var++; /* step-1 */
+ tgkill (getpid (), gettid (), SIGUSR1); /* step-2 */
+}
+
+static void *
+start (void *arg)
+{
+ tgkill (getpid (), gettid (), SIGUSR1);
+ assert (0);
+ return NULL;
+}
+
+int
+main (void)
+{
+ pthread_t thread;
+
+ signal (SIGUSR1, handler);
+
+ pthread_create (&thread, NULL, start, NULL);
+ start (NULL); /* main-start */
+ return 0;
+}
diff --git a/gdb/testsuite/gdb.threads/sigstep-threads.exp b/gdb/testsuite/gdb.threads/sigstep-threads.exp
new file mode 100644
index 0000000..313869c
--- /dev/null
+++ b/gdb/testsuite/gdb.threads/sigstep-threads.exp
@@ -0,0 +1,75 @@
+# Copyright 2010-2012 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/>.
+
+set testfile sigstep-threads
+set srcfile ${testfile}.c
+set executable ${testfile}
+set binfile ${objdir}/${subdir}/${executable}
+
+if { [gdb_compile_pthreads "${srcdir}/${subdir}/${srcfile}" "${binfile}" executable {debug}] != "" } {
+ untested ${testfile}.exp
+ return -1
+}
+
+clean_restart $executable
+
+if ![runto_main] {
+ return -1;
+}
+
+# `noprint' would not test the full logic of GDB.
+gdb_test "handle SIGUSR1 nostop print pass" \
+ "\r\nSIGUSR1\[ \t\]+No\[ \t\]+Yes\[ \t\]+Yes\[ \t\].*"
+
+gdb_test_no_output "set scheduler-locking off"
+
+gdb_breakpoint [gdb_get_line_number "step-1"]
+gdb_test_no_output {set $step1=$bpnum}
+gdb_continue_to_breakpoint "step-1" ".* step-1 .*"
+gdb_test_no_output {disable $step1}
+
+# 1 as we are now stopped at the `step-1' label.
+set step_at 1
+for {set i 0} {$i < 100} {incr i} {
+ set test "step $i"
+ # Presume this step failed - as in the case of a timeout.
+ set failed 1
+ gdb_test_multiple "step" $test {
+ -re "\r\nProgram received signal SIGUSR1, User defined signal 1.\r\n" {
+ exp_continue -continue_timer
+ }
+ -re "step-(\[012\]).*\r\n$gdb_prompt $" {
+ set now $expect_out(1,string)
+ if {$step_at == 2 && $now == 1} {
+ set failed 0
+ } elseif {$step_at == 1 && $now == 2} {
+ set failed 0
+ # Continue over the re-signalling back to the handle entry.
+ gdb_test_no_output {enable $step1} ""
+ gdb_test "continue" " step-1 .*" ""
+ set now 1
+ gdb_test_no_output {disable $step1} ""
+ } else {
+ fail $test
+ }
+ set step_at $now
+ }
+ }
+ if $failed {
+ return
+ }
+}
+# We can never reliably say the racy problematic case has been tested.
+pass "step"
diff --git a/gdb/thread.c b/gdb/thread.c
index d361dd8..ff5220d 100644
--- a/gdb/thread.c
+++ b/gdb/thread.c
@@ -54,7 +54,7 @@ void _initialize_thread (void);
/* Prototypes for local functions. */
-static struct thread_info *thread_list = NULL;
+struct thread_info *thread_list = NULL;
static int highest_thread_num;
static void thread_command (char *tidstr, int from_tty);