lkml.org 
[lkml]   [2013]   [May]   [5]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
From
SubjectRe: [PATCH 09/14] perf ftrace: Add 'record' sub-command
Date
Hi Jiri,

Sorry for late reply. I've been busy these days.

On Thu, 25 Apr 2013 15:24:18 +0900, Namhyung Kim wrote:
> On Wed, 24 Apr 2013 16:12:41 +0200, Jiri Olsa wrote:
>> I got following lockup for record command:
>>
>> # ./perf --no-pager ftrace record ls
>> ...
>> hangs
>>
>> in other terminal:
>>
>> # pstack 14237
>> Thread 2 (Thread 0x7f3f1aa1d700 (LWP 14241)):
>> #0 0x0000003cec20b595 in pthread_cond_wait@@GLIBC_2.3.2 () from /lib64/libpthread.so.0
>> #1 0x0000000000446ebe in record_ftrace_raw_buffer (arg=0x21f9ce0) at builtin-ftrace.c:451
>> #2 0x0000003cec207d14 in start_thread () from /lib64/libpthread.so.0
>> #3 0x0000003cebaf168d in clone () from /lib64/libc.so.6
>> Thread 1 (Thread 0x7f3f1ba209c0 (LWP 14237)):
>> #0 0x0000003cec208e60 in pthread_join () from /lib64/libpthread.so.0
>> #1 0x0000000000449c8c in do_ftrace_record (ftrace=ftrace@entry=0x7fff1377d240) at builtin-ftrace.c:682
>> #2 0x000000000044a57b in __cmd_ftrace_record (argv=0x7fff1377d818, argc=<optimized out>, ftrace=0x7fff1377d240) at builtin-ftrace.c:1535
>> #3 cmd_ftrace (argc=2, argv=0x7fff1377d818, prefix=<optimized out>) at builtin-ftrace.c:1655
>> #4 0x000000000041a763 in run_builtin (p=p@entry=0x7d58a0, argc=argc@entry=3, argv=argv@entry=0x7fff1377d818) at perf.c:320
>> #5 0x0000000000419faf in handle_internal_command (argv=0x7fff1377d818, argc=3) at perf.c:377
>> #6 run_argv (argv=0x7fff1377d600, argcp=0x7fff1377d60c) at perf.c:421
>> #7 main (argc=3, argv=0x7fff1377d818) at perf.c:522
>>
>
> How often can you reproduce it? I can't. :(
>
> But it looks there's a race between cond_wait() and cond_broatcast().
> I'll take a look at that.

I think the problem is that a recorder can be blocked if it received a
broadcast signal before calling cond_wait(). So I modified the code to
place signaling ready_cond and waiting start_cond in the same cs.

Does the patch below fix your problem?

Thanks,
Namhyung


diff --git a/tools/perf/builtin-ftrace.c b/tools/perf/builtin-ftrace.c
index 4296948197de..f1ecfa207822 100644
--- a/tools/perf/builtin-ftrace.c
+++ b/tools/perf/builtin-ftrace.c
@@ -428,12 +428,6 @@ static void *record_ftrace_raw_buffer(void *arg)

fra->state = RECORD_STATE__READY;

- /* Now I'm ready */
- pthread_mutex_lock(&recorder_mutex);
- recorder_count++;
- pthread_cond_signal(&recorder_ready_cond);
- pthread_mutex_unlock(&recorder_mutex);
-
/*
* block all signals but SIGUSR2.
* It'll be used to unblock a recorder to finish.
@@ -447,7 +441,10 @@ static void *record_ftrace_raw_buffer(void *arg)
fcntl(trace_fd, F_SETFL, O_NONBLOCK);
pollfd.fd = trace_fd;

+ /* Now I'm ready */
pthread_mutex_lock(&recorder_mutex);
+ recorder_count++;
+ pthread_cond_signal(&recorder_ready_cond);
pthread_cond_wait(&recorder_start_cond, &recorder_mutex);
pthread_mutex_unlock(&recorder_mutex);


\
 
 \ /
  Last update: 2013-05-06 04:21    [W:0.905 / U:0.220 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site