2022-02-09 13:22:17

by Changbin Du

[permalink] [raw]
Subject: [PATCH] perf trace: Avoid early exit due SIGCHLD from non-workload processes

The function trace__symbols_init() runs "perf-read-vdso32" and that ends up
with a SIGCHLD delivered to 'perf'. And this SIGCHLD make perf exit early.

'perf trace' should exit only if the SIGCHLD is from our workload process.
So let's use sigaction() instead of signal() to match such condition.

Signed-off-by: Changbin Du <[email protected]>
---
tools/perf/builtin-trace.c | 21 ++++++++++++++++-----
1 file changed, 16 insertions(+), 5 deletions(-)

diff --git a/tools/perf/builtin-trace.c b/tools/perf/builtin-trace.c
index 32844d8a0ea5..d03556c14b0a 100644
--- a/tools/perf/builtin-trace.c
+++ b/tools/perf/builtin-trace.c
@@ -1536,13 +1536,20 @@ static size_t trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
return fprintf(fp, " ? ");
}

+static pid_t workload_pid = -1;
static bool done = false;
static bool interrupted = false;

-static void sig_handler(int sig)
+static void sighandler_interrupt(int sig __maybe_unused)
{
- done = true;
- interrupted = sig == SIGINT;
+ done = interrupted = true;
+}
+
+static void sighandler_chld(int sig __maybe_unused, siginfo_t *info,
+ void *context __maybe_unused)
+{
+ if (info->si_pid == workload_pid)
+ done = true;
}

static size_t trace__fprintf_comm_tid(struct trace *trace, struct thread *thread, FILE *fp)
@@ -3938,7 +3945,6 @@ static int trace__run(struct trace *trace, int argc, const char **argv)
bool draining = false;

trace->live = true;
- signal(SIGCHLD, sig_handler);

if (!trace->raw_augmented_syscalls) {
if (trace->trace_syscalls && trace__add_syscall_newtp(trace))
@@ -4018,6 +4024,7 @@ static int trace__run(struct trace *trace, int argc, const char **argv)
fprintf(trace->output, "Couldn't run the workload!\n");
goto out_delete_evlist;
}
+ workload_pid = evlist->workload.pid;
}

err = evlist__open(evlist);
@@ -4887,10 +4894,14 @@ int cmd_trace(int argc, const char **argv)
const char * const trace_subcommands[] = { "record", NULL };
int err = -1;
char bf[BUFSIZ];
+ struct sigaction sigchld_act = { 0 };

signal(SIGSEGV, sighandler_dump_stack);
signal(SIGFPE, sighandler_dump_stack);
- signal(SIGINT, sig_handler);
+ signal(SIGINT, sighandler_interrupt);
+ sigchld_act.sa_flags = SA_SIGINFO;
+ sigchld_act.sa_sigaction = sighandler_chld;
+ sigaction(SIGCHLD, &sigchld_act, NULL);

trace.evlist = evlist__new();
trace.sctbl = syscalltbl__new();
--
2.25.1



2022-02-09 16:27:04

by Arnaldo Carvalho de Melo

[permalink] [raw]
Subject: Re: [PATCH] perf trace: Avoid early exit due SIGCHLD from non-workload processes

Em Wed, Feb 09, 2022 at 04:01:33PM +0100, Jiri Olsa escreveu:
> On Tue, Feb 08, 2022 at 10:07:25PM +0800, Changbin Du wrote:
> > The function trace__symbols_init() runs "perf-read-vdso32" and that ends up
> > with a SIGCHLD delivered to 'perf'. And this SIGCHLD make perf exit early.
> >
> > 'perf trace' should exit only if the SIGCHLD is from our workload process.
> > So let's use sigaction() instead of signal() to match such condition.
> >
> > Signed-off-by: Changbin Du <[email protected]>
>
> good idea, I tested with the reproducer for:
> f06a82f9d31a perf trace: Avoid early exit due to running SIGCHLD handler before it makes sense to
>
> and it's working properly
>
> Acked-by: Jiri Olsa <[email protected]>

So can I replace this with the stronger:

Tested-by: Jiri Olsa <[email protected]>

?

- Arnaldo

> thanks,
> jirka
>
> > ---
> > tools/perf/builtin-trace.c | 21 ++++++++++++++++-----
> > 1 file changed, 16 insertions(+), 5 deletions(-)
> >
> > diff --git a/tools/perf/builtin-trace.c b/tools/perf/builtin-trace.c
> > index 32844d8a0ea5..d03556c14b0a 100644
> > --- a/tools/perf/builtin-trace.c
> > +++ b/tools/perf/builtin-trace.c
> > @@ -1536,13 +1536,20 @@ static size_t trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
> > return fprintf(fp, " ? ");
> > }
> >
> > +static pid_t workload_pid = -1;
> > static bool done = false;
> > static bool interrupted = false;
> >
> > -static void sig_handler(int sig)
> > +static void sighandler_interrupt(int sig __maybe_unused)
> > {
> > - done = true;
> > - interrupted = sig == SIGINT;
> > + done = interrupted = true;
> > +}
> > +
> > +static void sighandler_chld(int sig __maybe_unused, siginfo_t *info,
> > + void *context __maybe_unused)
> > +{
> > + if (info->si_pid == workload_pid)
> > + done = true;
> > }
> >
> > static size_t trace__fprintf_comm_tid(struct trace *trace, struct thread *thread, FILE *fp)
> > @@ -3938,7 +3945,6 @@ static int trace__run(struct trace *trace, int argc, const char **argv)
> > bool draining = false;
> >
> > trace->live = true;
> > - signal(SIGCHLD, sig_handler);
> >
> > if (!trace->raw_augmented_syscalls) {
> > if (trace->trace_syscalls && trace__add_syscall_newtp(trace))
> > @@ -4018,6 +4024,7 @@ static int trace__run(struct trace *trace, int argc, const char **argv)
> > fprintf(trace->output, "Couldn't run the workload!\n");
> > goto out_delete_evlist;
> > }
> > + workload_pid = evlist->workload.pid;
> > }
> >
> > err = evlist__open(evlist);
> > @@ -4887,10 +4894,14 @@ int cmd_trace(int argc, const char **argv)
> > const char * const trace_subcommands[] = { "record", NULL };
> > int err = -1;
> > char bf[BUFSIZ];
> > + struct sigaction sigchld_act = { 0 };
> >
> > signal(SIGSEGV, sighandler_dump_stack);
> > signal(SIGFPE, sighandler_dump_stack);
> > - signal(SIGINT, sig_handler);
> > + signal(SIGINT, sighandler_interrupt);
> > + sigchld_act.sa_flags = SA_SIGINFO;
> > + sigchld_act.sa_sigaction = sighandler_chld;
> > + sigaction(SIGCHLD, &sigchld_act, NULL);
> >
> > trace.evlist = evlist__new();
> > trace.sctbl = syscalltbl__new();
> > --
> > 2.25.1
> >

--

- Arnaldo

2022-02-09 18:35:05

by Jiri Olsa

[permalink] [raw]
Subject: Re: [PATCH] perf trace: Avoid early exit due SIGCHLD from non-workload processes

On Tue, Feb 08, 2022 at 10:07:25PM +0800, Changbin Du wrote:
> The function trace__symbols_init() runs "perf-read-vdso32" and that ends up
> with a SIGCHLD delivered to 'perf'. And this SIGCHLD make perf exit early.
>
> 'perf trace' should exit only if the SIGCHLD is from our workload process.
> So let's use sigaction() instead of signal() to match such condition.
>
> Signed-off-by: Changbin Du <[email protected]>

good idea, I tested with the reproducer for:
f06a82f9d31a perf trace: Avoid early exit due to running SIGCHLD handler before it makes sense to

and it's working properly

Acked-by: Jiri Olsa <[email protected]>

thanks,
jirka

> ---
> tools/perf/builtin-trace.c | 21 ++++++++++++++++-----
> 1 file changed, 16 insertions(+), 5 deletions(-)
>
> diff --git a/tools/perf/builtin-trace.c b/tools/perf/builtin-trace.c
> index 32844d8a0ea5..d03556c14b0a 100644
> --- a/tools/perf/builtin-trace.c
> +++ b/tools/perf/builtin-trace.c
> @@ -1536,13 +1536,20 @@ static size_t trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
> return fprintf(fp, " ? ");
> }
>
> +static pid_t workload_pid = -1;
> static bool done = false;
> static bool interrupted = false;
>
> -static void sig_handler(int sig)
> +static void sighandler_interrupt(int sig __maybe_unused)
> {
> - done = true;
> - interrupted = sig == SIGINT;
> + done = interrupted = true;
> +}
> +
> +static void sighandler_chld(int sig __maybe_unused, siginfo_t *info,
> + void *context __maybe_unused)
> +{
> + if (info->si_pid == workload_pid)
> + done = true;
> }
>
> static size_t trace__fprintf_comm_tid(struct trace *trace, struct thread *thread, FILE *fp)
> @@ -3938,7 +3945,6 @@ static int trace__run(struct trace *trace, int argc, const char **argv)
> bool draining = false;
>
> trace->live = true;
> - signal(SIGCHLD, sig_handler);
>
> if (!trace->raw_augmented_syscalls) {
> if (trace->trace_syscalls && trace__add_syscall_newtp(trace))
> @@ -4018,6 +4024,7 @@ static int trace__run(struct trace *trace, int argc, const char **argv)
> fprintf(trace->output, "Couldn't run the workload!\n");
> goto out_delete_evlist;
> }
> + workload_pid = evlist->workload.pid;
> }
>
> err = evlist__open(evlist);
> @@ -4887,10 +4894,14 @@ int cmd_trace(int argc, const char **argv)
> const char * const trace_subcommands[] = { "record", NULL };
> int err = -1;
> char bf[BUFSIZ];
> + struct sigaction sigchld_act = { 0 };
>
> signal(SIGSEGV, sighandler_dump_stack);
> signal(SIGFPE, sighandler_dump_stack);
> - signal(SIGINT, sig_handler);
> + signal(SIGINT, sighandler_interrupt);
> + sigchld_act.sa_flags = SA_SIGINFO;
> + sigchld_act.sa_sigaction = sighandler_chld;
> + sigaction(SIGCHLD, &sigchld_act, NULL);
>
> trace.evlist = evlist__new();
> trace.sctbl = syscalltbl__new();
> --
> 2.25.1
>

2022-02-09 19:36:18

by Jiri Olsa

[permalink] [raw]
Subject: Re: [PATCH] perf trace: Avoid early exit due SIGCHLD from non-workload processes

On Wed, Feb 09, 2022 at 12:50:07PM -0300, Arnaldo Carvalho de Melo wrote:
> Em Wed, Feb 09, 2022 at 04:01:33PM +0100, Jiri Olsa escreveu:
> > On Tue, Feb 08, 2022 at 10:07:25PM +0800, Changbin Du wrote:
> > > The function trace__symbols_init() runs "perf-read-vdso32" and that ends up
> > > with a SIGCHLD delivered to 'perf'. And this SIGCHLD make perf exit early.
> > >
> > > 'perf trace' should exit only if the SIGCHLD is from our workload process.
> > > So let's use sigaction() instead of signal() to match such condition.
> > >
> > > Signed-off-by: Changbin Du <[email protected]>
> >
> > good idea, I tested with the reproducer for:
> > f06a82f9d31a perf trace: Avoid early exit due to running SIGCHLD handler before it makes sense to
> >
> > and it's working properly
> >
> > Acked-by: Jiri Olsa <[email protected]>
>
> So can I replace this with the stronger:
>
> Tested-by: Jiri Olsa <[email protected]>

I did not see the case described in the changelog,
just checked the other one was fine..

jirka

>
> ?
>
> - Arnaldo
>
> > thanks,
> > jirka
> >
> > > ---
> > > tools/perf/builtin-trace.c | 21 ++++++++++++++++-----
> > > 1 file changed, 16 insertions(+), 5 deletions(-)
> > >
> > > diff --git a/tools/perf/builtin-trace.c b/tools/perf/builtin-trace.c
> > > index 32844d8a0ea5..d03556c14b0a 100644
> > > --- a/tools/perf/builtin-trace.c
> > > +++ b/tools/perf/builtin-trace.c
> > > @@ -1536,13 +1536,20 @@ static size_t trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
> > > return fprintf(fp, " ? ");
> > > }
> > >
> > > +static pid_t workload_pid = -1;
> > > static bool done = false;
> > > static bool interrupted = false;
> > >
> > > -static void sig_handler(int sig)
> > > +static void sighandler_interrupt(int sig __maybe_unused)
> > > {
> > > - done = true;
> > > - interrupted = sig == SIGINT;
> > > + done = interrupted = true;
> > > +}
> > > +
> > > +static void sighandler_chld(int sig __maybe_unused, siginfo_t *info,
> > > + void *context __maybe_unused)
> > > +{
> > > + if (info->si_pid == workload_pid)
> > > + done = true;
> > > }
> > >
> > > static size_t trace__fprintf_comm_tid(struct trace *trace, struct thread *thread, FILE *fp)
> > > @@ -3938,7 +3945,6 @@ static int trace__run(struct trace *trace, int argc, const char **argv)
> > > bool draining = false;
> > >
> > > trace->live = true;
> > > - signal(SIGCHLD, sig_handler);
> > >
> > > if (!trace->raw_augmented_syscalls) {
> > > if (trace->trace_syscalls && trace__add_syscall_newtp(trace))
> > > @@ -4018,6 +4024,7 @@ static int trace__run(struct trace *trace, int argc, const char **argv)
> > > fprintf(trace->output, "Couldn't run the workload!\n");
> > > goto out_delete_evlist;
> > > }
> > > + workload_pid = evlist->workload.pid;
> > > }
> > >
> > > err = evlist__open(evlist);
> > > @@ -4887,10 +4894,14 @@ int cmd_trace(int argc, const char **argv)
> > > const char * const trace_subcommands[] = { "record", NULL };
> > > int err = -1;
> > > char bf[BUFSIZ];
> > > + struct sigaction sigchld_act = { 0 };
> > >
> > > signal(SIGSEGV, sighandler_dump_stack);
> > > signal(SIGFPE, sighandler_dump_stack);
> > > - signal(SIGINT, sig_handler);
> > > + signal(SIGINT, sighandler_interrupt);
> > > + sigchld_act.sa_flags = SA_SIGINFO;
> > > + sigchld_act.sa_sigaction = sighandler_chld;
> > > + sigaction(SIGCHLD, &sigchld_act, NULL);
> > >
> > > trace.evlist = evlist__new();
> > > trace.sctbl = syscalltbl__new();
> > > --
> > > 2.25.1
> > >
>
> --
>
> - Arnaldo

2022-02-09 20:12:59

by Arnaldo Carvalho de Melo

[permalink] [raw]
Subject: Re: [PATCH] perf trace: Avoid early exit due SIGCHLD from non-workload processes



On February 9, 2022 3:53:55 PM GMT-03:00, Jiri Olsa <[email protected]> wrote:
>On Wed, Feb 09, 2022 at 12:50:07PM -0300, Arnaldo Carvalho de Melo wrote:
>> Em Wed, Feb 09, 2022 at 04:01:33PM +0100, Jiri Olsa escreveu:
>> > On Tue, Feb 08, 2022 at 10:07:25PM +0800, Changbin Du wrote:
>> > > The function trace__symbols_init() runs "perf-read-vdso32" and that ends up
>> > > with a SIGCHLD delivered to 'perf'. And this SIGCHLD make perf exit early.
>> > >
>> > > 'perf trace' should exit only if the SIGCHLD is from our workload process.
>> > > So let's use sigaction() instead of signal() to match such condition.
>> > >
>> > > Signed-off-by: Changbin Du <[email protected]>
>> >
>> > good idea, I tested with the reproducer for:
>> > f06a82f9d31a perf trace: Avoid early exit due to running SIGCHLD handler before it makes sense to
>> >
>> > and it's working properly
>> >
>> > Acked-by: Jiri Olsa <[email protected]>
>>
>> So can I replace this with the stronger:
>>
>> Tested-by: Jiri Olsa <[email protected]>
>
>I did not see the case described in the changelog,
>just checked the other one was fine..

Ok, I'll keep it as an acked-by then.

- Arnaldo
>
>jirka
>
>>
>> ?
>>
>> - Arnaldo
>>
>> > thanks,
>> > jirka
>> >
>> > > ---
>> > > tools/perf/builtin-trace.c | 21 ++++++++++++++++-----
>> > > 1 file changed, 16 insertions(+), 5 deletions(-)
>> > >
>> > > diff --git a/tools/perf/builtin-trace.c b/tools/perf/builtin-trace.c
>> > > index 32844d8a0ea5..d03556c14b0a 100644
>> > > --- a/tools/perf/builtin-trace.c
>> > > +++ b/tools/perf/builtin-trace.c
>> > > @@ -1536,13 +1536,20 @@ static size_t trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
>> > > return fprintf(fp, " ? ");
>> > > }
>> > >
>> > > +static pid_t workload_pid = -1;
>> > > static bool done = false;
>> > > static bool interrupted = false;
>> > >
>> > > -static void sig_handler(int sig)
>> > > +static void sighandler_interrupt(int sig __maybe_unused)
>> > > {
>> > > - done = true;
>> > > - interrupted = sig == SIGINT;
>> > > + done = interrupted = true;
>> > > +}
>> > > +
>> > > +static void sighandler_chld(int sig __maybe_unused, siginfo_t *info,
>> > > + void *context __maybe_unused)
>> > > +{
>> > > + if (info->si_pid == workload_pid)
>> > > + done = true;
>> > > }
>> > >
>> > > static size_t trace__fprintf_comm_tid(struct trace *trace, struct thread *thread, FILE *fp)
>> > > @@ -3938,7 +3945,6 @@ static int trace__run(struct trace *trace, int argc, const char **argv)
>> > > bool draining = false;
>> > >
>> > > trace->live = true;
>> > > - signal(SIGCHLD, sig_handler);
>> > >
>> > > if (!trace->raw_augmented_syscalls) {
>> > > if (trace->trace_syscalls && trace__add_syscall_newtp(trace))
>> > > @@ -4018,6 +4024,7 @@ static int trace__run(struct trace *trace, int argc, const char **argv)
>> > > fprintf(trace->output, "Couldn't run the workload!\n");
>> > > goto out_delete_evlist;
>> > > }
>> > > + workload_pid = evlist->workload.pid;
>> > > }
>> > >
>> > > err = evlist__open(evlist);
>> > > @@ -4887,10 +4894,14 @@ int cmd_trace(int argc, const char **argv)
>> > > const char * const trace_subcommands[] = { "record", NULL };
>> > > int err = -1;
>> > > char bf[BUFSIZ];
>> > > + struct sigaction sigchld_act = { 0 };
>> > >
>> > > signal(SIGSEGV, sighandler_dump_stack);
>> > > signal(SIGFPE, sighandler_dump_stack);
>> > > - signal(SIGINT, sig_handler);
>> > > + signal(SIGINT, sighandler_interrupt);
>> > > + sigchld_act.sa_flags = SA_SIGINFO;
>> > > + sigchld_act.sa_sigaction = sighandler_chld;
>> > > + sigaction(SIGCHLD, &sigchld_act, NULL);
>> > >
>> > > trace.evlist = evlist__new();
>> > > trace.sctbl = syscalltbl__new();
>> > > --
>> > > 2.25.1
>> > >
>>
>> --
>>
>> - Arnaldo