Re: [PATCH v2 1/2] perf record: Propagate exit status of a command line workload

[Namhyung Kim <namhyung@kernel.org>]

Hi Stephane,

2014-04-23 (수), 14:11 +0200, Stephane Eranian:
> On Wed, Apr 23, 2014 at 8:40 AM, Namhyung Kim <namhyung@kernel.org> wrote:
> > +out_child:
> > +       if (forks) {
> > +               int exit_status;
> >
> > -       return 0;
> > +               if (!child_finished)
> > +                       kill(rec->evlist->workload.pid, SIGTERM);
> > +
> > +               wait(&exit_status);
> > +
> > +               if (err < 0)
> Not quite this. Although this works with my test case with 'false'.
> It fails when I tried the opposite test case:
> 
> $ perf record true && echo yes || echo no
>   [ perf record: Woken up 1 times to write data ]
>   [ perf record: Captured and wrote 0.013 MB perf.data (~589 samples) ]
>   no
> 
> The return code in this case is 255. I tracked it down to err being last set
> by poll() which got interrupted by SIGCHLD. So I think the err value must
> be overridden somehow in this case. Maybe something like:
> 
>                         err = poll(evsel_list->pollfd, evsel_list->nr_fds, -1);
>                         if (err < 0 &&  errno == EINTR && forks && done)
>                                 err = 0;

Hmm.. just checking "err < 0 && errno == EINTR" isn't enough?

And I guess the same problem can be occurred during record__mmap_read()
as it calls record__write() which eventually calls ion() and it seems
not to handle the above case..  I'll cook a patch for it too.

Thanks,
Namhyung