Re: [PATCH] perf record: Fix continue profiling after draining the buffer

[Arnaldo Carvalho de Melo]

Em Fri, Feb 05, 2021 at 11:46:12AM +0100, Jiri Olsa escreveu:
> On Fri, Feb 05, 2021 at 07:35:22PM +0900, Namhyung Kim wrote:
> > Hello,
> > 
> > On Fri, Feb 5, 2021 at 3:50 PM Yang Jihong <yangjihong1@xxxxxxxxxx> wrote:
> > >
> > > commit da231338ec9c098707c8a1e4d8a50e2400e2fe17 uses eventfd to solve rare race
> > > where the setting and checking of 'done' which add done_fd to pollfd.
> > > When draining buffer, revents of done_fd is 0 and evlist__filter_pollfd
> > > function returns a non-zero value.
> > > As a result, perf record does not stop profiling.
> > >
> > > The following simple scenarios can trigger this condition:
> > >
> > > sleep 10 &
> > > perf record -p $!
> > >
> > > After the sleep process exits, perf record should stop profiling and exit.
> > > However, perf record keeps running.
> > >
> > > If pollfd revents contains only POLLERR or POLLHUP,
> > > perf record indicates that buffer is draining and need to stop profiling.
> > > Use fdarray_flag__nonfilterable to set done eventfd to nonfilterable objects,
> > > so that evlist__filter_pollfd does not filter and check done eventfd.
> > >
> > > Fixes: da231338ec9c (perf record: Use an eventfd to wakeup when done)
> > > Signed-off-by: Yang Jihong <yangjihong1@xxxxxxxxxx>
> > > ---
> > >  tools/perf/builtin-record.c | 2 +-
> > >  tools/perf/util/evlist.c    | 8 ++++++++
> > >  tools/perf/util/evlist.h    | 4 ++++
> > >  3 files changed, 13 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
> > > index fd3911650612..51e593e896ea 100644
> > > --- a/tools/perf/builtin-record.c
> > > +++ b/tools/perf/builtin-record.c
> > > @@ -1663,7 +1663,7 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)
> > >                 status = -1;
> > >                 goto out_delete_session;
> > >         }
> > > -       err = evlist__add_pollfd(rec->evlist, done_fd);
> > > +       err = evlist__add_wakeup_eventfd(rec->evlist, done_fd);
> > >         if (err < 0) {
> > >                 pr_err("Failed to add wakeup eventfd to poll list\n");
> > >                 status = err;
> > > diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
> > > index 05363a7247c4..fea4c1e8010d 100644
> > > --- a/tools/perf/util/evlist.c
> > > +++ b/tools/perf/util/evlist.c
> > > @@ -572,6 +572,14 @@ int evlist__filter_pollfd(struct evlist *evlist, short revents_and_mask)
> > >         return perf_evlist__filter_pollfd(&evlist->core, revents_and_mask);
> > >  }
> > >
> > > +#ifdef HAVE_EVENTFD_SUPPORT
> > > +int evlist__add_wakeup_eventfd(struct evlist *evlist, int fd)
> > > +{
> > > +       return perf_evlist__add_pollfd(&evlist->core, fd, NULL, POLLIN,
> > > +                                      fdarray_flag__nonfilterable);
> > > +}
> > > +#endif
> > 
> > Does it build when HAVE_EVENTFD_SUPPORT is not defined?
> 
> yea, I was wondering the same.. but it's called only from
> code within HAVE_EVENTFD_SUPPORT ifdef

Yes, this can't work on systems without eventfd, it will simply not
build, and why do we have to make the definition of this function
conditional on HAVE_EVENTFD_SUPPORT?

I'm missing something :-\

Yeah, this whole call to evlist__add_pollfd is already surrounded by
#ifdef HAVE_EVENTFD_SUPPORT:

1656         if (zstd_init(&session->zstd_data, rec->opts.comp_level) < 0) {
1657                 pr_err("Compression initialization failed.\n");
1658                 return -1;
1659         }
1660 #ifdef HAVE_EVENTFD_SUPPORT
1661         done_fd = eventfd(0, EFD_NONBLOCK);
1662         if (done_fd < 0) {
1663                 pr_err("Failed to create wakeup eventfd, error: %m\n");
1664                 status = -1;
1665                 goto out_delete_session;
1666         }
1667         err = evlist__add_pollfd(rec->evlist, done_fd);
1668         if (err < 0) {
1669                 pr_err("Failed to add wakeup eventfd to poll list\n");
1670                 status = err;
1671                 goto out_delete_session;
1672         }
1673 #endif // HAVE_EVENTFD_SUPPORT
1674 
1675         session->header.env.comp_type  = PERF_COMP_ZSTD;
1676         session->header.env.comp_level = rec->opts.comp_level;

Jiri, does your Acked-by stands? Namhyung?

- Arnaldo
 
> jirka
> 
> > 
> > Thanks,
> > Namhyung
> > 
> > 
> > > +
> > >  int evlist__poll(struct evlist *evlist, int timeout)
> > >  {
> > >         return perf_evlist__poll(&evlist->core, timeout);
> > > diff --git a/tools/perf/util/evlist.h b/tools/perf/util/evlist.h
> > > index 1aae75895dea..6d4d62151bc8 100644
> > > --- a/tools/perf/util/evlist.h
> > > +++ b/tools/perf/util/evlist.h
> > > @@ -142,6 +142,10 @@ struct evsel *evlist__find_tracepoint_by_name(struct evlist *evlist, const char
> > >  int evlist__add_pollfd(struct evlist *evlist, int fd);
> > >  int evlist__filter_pollfd(struct evlist *evlist, short revents_and_mask);
> > >
> > > +#ifdef HAVE_EVENTFD_SUPPORT
> > > +int evlist__add_wakeup_eventfd(struct evlist *evlist, int fd);
> > > +#endif
> > > +
> > >  int evlist__poll(struct evlist *evlist, int timeout);
> > >
> > >  struct evsel *evlist__id2evsel(struct evlist *evlist, u64 id);
> > > --
> > > 2.17.1
> > >
> > 
> 

-- 

- Arnaldo