Re: [PATCH v4 4/9] libperf: Add libperf_evsel__mmap()

[Rob Herring]

On Wed, Oct 21, 2020 at 6:24 AM Jiri Olsa <jolsa@xxxxxxxxxx> wrote:
>
> On Tue, Oct 20, 2020 at 12:11:47PM -0500, Rob Herring wrote:
>
> SNIP
>
> > > > > >
> > > > > > The mmapped read will actually fail and then we fallback here. My main
> > > > > > concern though is adding more overhead on a feature that's meant to be
> > > > > > low overhead (granted, it's not much). Maybe we could add checks on
> > > > > > the mmap that we've opened the event with pid == 0 and cpu == -1 (so
> > > > > > only 1 FD)?
> > > > >
> > > > > but then you limit this just for single fd.. having mmap as xyarray
> > > > > would not be that bad and perf_evsel__mmap will call perf_mmap__mmap
> > > > > for each defined cpu/thread .. so it depends on user how fast this
> > > > > will be - how many maps needs to be created/mmaped
> > > >
> > > > Given userspace access fails for anything other than the calling
> > > > thread and all cpus, how would more than 1 mmap be useful here?
> > >
> > > I'm not sure what you mean by fail in here.. you need mmap for each
> > > event fd you want to read from
> >
> > Yes, but that's one mmap per event (evsel) which is different than per
> > cpu/thread.
>
> right, and you need mmap per fd IIUC
>
> >
> > > in the example below we read stats from all cpus via perf_evsel__read,
> > > if we insert this call after perf_evsel__open:
> > >
> > >   perf_evsel__mmap(cpus, NULL);
> > >
> > > that maps page for each event, then perf_evsel__read
> > > could go through the fast code, no?
> >
> > No, because we're not self-monitoring (pid == 0 and cpu == -1). With
> > the following change:
> >
> > diff --git a/tools/lib/perf/tests/test-evsel.c
> > b/tools/lib/perf/tests/test-evsel.c
> > index eeca8203d73d..1fca9c121f7c 100644
> > --- a/tools/lib/perf/tests/test-evsel.c
> > +++ b/tools/lib/perf/tests/test-evsel.c
> > @@ -17,6 +17,7 @@ static int test_stat_cpu(void)
> >  {
> >         struct perf_cpu_map *cpus;
> >         struct perf_evsel *evsel;
> > +       struct perf_event_mmap_page *pc;
> >         struct perf_event_attr attr = {
> >                 .type   = PERF_TYPE_SOFTWARE,
> >                 .config = PERF_COUNT_SW_CPU_CLOCK,
> > @@ -32,6 +33,15 @@ static int test_stat_cpu(void)
> >         err = perf_evsel__open(evsel, cpus, NULL);
> >         __T("failed to open evsel", err == 0);
> >
> > +       pc = perf_evsel__mmap(evsel, 0);
> > +       __T("failed to mmap evsel", pc);
> > +
> > +#if defined(__i386__) || defined(__x86_64__) || defined(__aarch64__)
> > +       __T("userspace counter access not supported", pc->cap_user_rdpmc);
> > +       __T("userspace counter access not enabled", pc->index);
> > +       __T("userspace counter width not set", pc->pmc_width >= 32);
> > +#endif
>
> I'll need to check, I'm surprised this would depend on the way
> you open the event

Any more thoughts on this?

Rob