lkml.org 
[lkml]   [2012]   [Aug]   [27]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [PATCH 3/4] perf: teach perf inject to merge sched_stat_* and sched_switch events (v2)
From
2012/8/25 Frederic Weisbecker <fweisbec@gmail.com>:
>> + if (!strcmp(evname, "sched_switch") || sched_process_exit) {
>> + list_for_each_entry(ent, &samples, node)
>> + if (sample->pid == ent->pid)
>
> I suspect what you're rather interested in is the sample tid.

Yes, you are right.

>
>> +
>> + } else if (!strncmp(evname, "sched_stat_", 11)) {
>> + u32 pid;
>> +
>> + pid = raw_field_value(evsel->tp_format,
>> + "pid", sample->raw_data);
>
> There you parse the pid from the trace content. That's fine because
> it's actually the tid that is saved on the trace event. But this one
> is not pid-namespace safe (it saves current->pid directly) while
> sample->tid is pid-namespace safe (it uses task_pid_nr_ns).
>
> So I suggest you to use sample->tid instead, plus that's going to be
> consistant with what you did above.

Here is a problem, because a pid from the trace content and
sample->tid are not the same.
A kernel wakes up a task1 and do it from a context of another task2.
In this case a pid from the trace content is a pid of the task1 and a
sample->tid is a pid of the task2.

DECLARE_EVENT_CLASS(sched_stat_template,
TP_PROTO(struct task_struct *tsk, u64 delay),
...
__entry->pid = tsk->pid;
...


Here is a patch, which allow us to get "foreign" events.
http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=e6dab5ffab59e910ec0e3355f4a6f29f7a7be474


\
 
 \ /
  Last update: 2012-08-27 10:01    [W:0.173 / U:0.304 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site