[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAP-5=fUBsuOau6scssB_RL6uGczGfe8GHL2Cx-ZE1Oj7ZTZ4KA@mail.gmail.com>
Date: Wed, 12 Jun 2024 13:59:22 -0700
From: Ian Rogers <irogers@...gle.com>
To: Leo Yan <leo.yan@....com>
Cc: Suzuki K Poulose <suzuki.poulose@....com>, Mike Leach <mike.leach@...aro.org>,
James Clark <james.clark@....com>, John Garry <john.g.garry@...cle.com>,
Will Deacon <will@...nel.org>, Leo Yan <leo.yan@...ux.dev>,
Peter Zijlstra <peterz@...radead.org>, Ingo Molnar <mingo@...hat.com>,
Arnaldo Carvalho de Melo <acme@...nel.org>, Namhyung Kim <namhyung@...nel.org>,
Mark Rutland <mark.rutland@....com>,
Alexander Shishkin <alexander.shishkin@...ux.intel.com>, Jiri Olsa <jolsa@...nel.org>,
Adrian Hunter <adrian.hunter@...el.com>, Kan Liang <kan.liang@...ux.intel.com>,
coresight@...ts.linaro.org, linux-arm-kernel@...ts.infradead.org,
linux-perf-users@...r.kernel.org, linux-kernel@...r.kernel.org,
Yicong Yang <yangyicong@...wei.com>
Subject: Re: [PATCH v1] perf arm: Workaround ARM PMUs cpu maps having offline cpus
On Wed, Jun 12, 2024 at 1:32 PM Leo Yan <leo.yan@....com> wrote:
[...]
> >
> > Renaming of the cycles event in arm_dsu_pmu.c - I'd say this is a top
> > priority issue right now.
>
> I cannot promise this. The main reason is that I still believe the
> 'cycles' event (or, generally speaking, all events) should be managed by
> the tool rather than by the uncore PMU drivers. Additionally, the perf
> tool currently has handled these symbolic events effectively.
I don't understand this.
1) the PMU advertises an event called 'cycles' - nothing to do with the tool
2) perf without a PMU matches events on all PMUs. I don't know people
involved in the perf tool development who think cycles shouldn't match
on an uncore PMU, which was Linus' stand point and reason for
rejecting my fix in favor of a revert.
3) matching all PMUs wasn't the case for legacy events, like cycles,
but now we want it to be to fix the Apple M? PMU issues where legacy
events fail to work. The whole reason we changed the priority of
legacy events as ARM requested. It also makes things more uniform with
BIG.little/hybrid.
We can update perf record to warn (not fail) about not opening events.
ARM should do this as (1) and (3) were caused by ARM - I have a WIP
patch but I'm in no mood to finish/send it. Even with this there will
be a warning for "perf record -e cycles .." and fixing the event name
is the only way to clear that up without special case code or making
warnings verbose only - something that makes me and others
uncomfortable.
Thanks,
Ian
Powered by blists - more mailing lists