Message-ID: <CAP-5=fUi-Y408hZmPosfejOW=SkXZZTTvRm326fBz-9zrHk8ew@mail.gmail.com>
Date: Wed, 16 Oct 2024 08:01:21 -0700
From: Ian Rogers <irogers@...gle.com>
To: James Clark <james.clark@...aro.org>
Cc: linux-perf-users@...r.kernel.org, acme@...nel.org, namhyung@...nel.org, 
	Peter Zijlstra <peterz@...radead.org>, Ingo Molnar <mingo@...hat.com>, 
	Mark Rutland <mark.rutland@....com>, 
	Alexander Shishkin <alexander.shishkin@...ux.intel.com>, Jiri Olsa <jolsa@...nel.org>, 
	Adrian Hunter <adrian.hunter@...el.com>, "Liang, Kan" <kan.liang@...ux.intel.com>, 
	linux-kernel@...r.kernel.org
Subject: Re: [RFC PATCH 1/1] libperf: evlist: Fix --cpu argument on hybrid platform

On Wed, Oct 16, 2024 at 1:29 AM James Clark <james.clark@...aro.org> wrote:
>
> On 15/10/2024 4:14 pm, Ian Rogers wrote:
> > On Tue, Oct 15, 2024 at 7:54 AM James Clark <james.clark@...aro.org> wrote:
> >>
> >> Since the linked fixes: commit, specifying a CPU on hybrid platforms
> >> results in an error because Perf tries to open an extended type event
> >> on "any" CPU which isn't valid. Extended type events can only be opened
> >> on CPUs that match the type.
> >>
> >> Before (working):
> >>
> >>    $ perf record --cpu 1 -- true
> >>    [ perf record: Woken up 1 times to write data ]
> >>    [ perf record: Captured and wrote 2.385 MB perf.data (7 samples) ]
> >>
> >> After (not working):
> >>
> >>    $ perf record -C 1 -- true
> >>    WARNING: A requested CPU in '1' is not supported by PMU 'cpu_atom' (CPUs 16-27) for event 'cycles:P'
> >>    Error:
> >>    The sys_perf_event_open() syscall returned with 22 (Invalid argument) for event (cpu_atom/cycles:P/).
> >>    /bin/dmesg | grep -i perf may provide additional information.
> >>
> >> (Ignore the warning message, that's expected and not particularly
> >> relevant to this issue).
> >>
> >> This is because perf_cpu_map__intersect() of the user specified CPU (1)
> >> and one of the PMU's CPUs (16-27) correctly results in an empty (NULL)
> >> CPU map. However for the purposes of opening an event, libperf converts
> >> empty CPU maps into an any CPU (-1) which the kernel rejects.
> >
> > Ugh. The cpumap API tries its best to confuse NULL == empty but empty
> > can give you dummy, dummy is also known as 'any' or -1, 'any' sounds a
> > lot like 'all' but sometimes 'all' is only online CPUs. I tried to
> > tidy up the naming a while ago, but there is still a mess.
> >
>
> I don't know if you think this is a good opportunity for me to have a go
> at finishing separating those? Or is it a dead end?

So cpumap (and threadmap) underpin a lot of things, we also used to
routinely confuse CPU numbers with cpumap indices that are used to
densely index xyarrays with file descriptors, etc. My thought was that
we may end up doing a proper Rust libperf that can be under a more
permissive license like libbpf - currently libperf is a source of GPL
infection. The rewrite would be a good time to clear these things up.
I believe someone at RedHat has looked at doing a Rust libperf.

> >> Fix it by deleting evsels with empty CPU maps in the specific case where
> >> user requested CPU maps are evaluated.
> >
> > If we delete evsels than the indices can be broken for certain things.
> > I'm guessing asan testing is clean but the large number of side data
> > structures that are indexed by things in another data structure makes
> > the whole code base brittle and I am nervous around this change.
> >
> >> Fixes: 251aa040244a ("perf parse-events: Wildcard most "numeric" events")
> >> Signed-off-by: James Clark <james.clark@...aro.org>
> >
> > Reviewed-by: Ian Rogers <irogers@...gle.com>
> >
> > Thanks,
> > Ian
> >
>
> Ok if we're not completely opposed to doing it this way I will dig a bit
> more and double check everything is working.

I think it is okay to do it this way (hence the reviewed-by tag) as
propagate maps should happen before the xyarrays are set up, it'd be
nice if these things were checked at runtime, or by the compiler...

Thanks,
Ian

> >> ---
> >>   tools/lib/perf/evlist.c | 11 +++++++++--
> >>   1 file changed, 9 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/tools/lib/perf/evlist.c b/tools/lib/perf/evlist.c
> >> index c6d67fc9e57e..8fae9a157a91 100644
> >> --- a/tools/lib/perf/evlist.c
> >> +++ b/tools/lib/perf/evlist.c
> >> @@ -47,6 +47,13 @@ static void __perf_evlist__propagate_maps(struct perf_evlist *evlist,
> >>                   */
> >>                  perf_cpu_map__put(evsel->cpus);
> >>                  evsel->cpus = perf_cpu_map__intersect(evlist->user_requested_cpus, evsel->own_cpus);
> >> +
> >> +               /*
> >> +                * Empty cpu lists would eventually get opened as "any" so remove
> >> +                * genuinely empty ones before they're opened in the wrong place.
> >> +                */
> >> +               if (perf_cpu_map__is_empty(evsel->cpus))
> >> +                       perf_evlist__remove(evlist, evsel);
> >>          } else if (!evsel->own_cpus || evlist->has_user_cpus ||
> >>                  (!evsel->requires_cpu && perf_cpu_map__has_any_cpu(evlist->user_requested_cpus))) {
> >>                  /*
> >> @@ -80,11 +87,11 @@ static void __perf_evlist__propagate_maps(struct perf_evlist *evlist,
> >>
> >>   static void perf_evlist__propagate_maps(struct perf_evlist *evlist)
> >>   {
> >> -       struct perf_evsel *evsel;
> >> +       struct perf_evsel *evsel, *n;
> >>
> >>          evlist->needs_map_propagation = true;
> >>
> >> -       perf_evlist__for_each_evsel(evlist, evsel)
> >> +       list_for_each_entry_safe(evsel, n, &evlist->entries, node)
> >>                  __perf_evlist__propagate_maps(evlist, evsel);
> >>   }
> >>
> >> --
> >> 2.34.1
> >>
>

Powered by blists - more mailing lists