Received-SPF: pass (google.com: domain of linux-kernel+bounces-189303-linux.lists.archive=gmail.com@vger.kernel.org designates 139.178.88.99 as permitted sender) client-ip=139.178.88.99;
Precedence: bulk
MIME-Version: 1.0
References: <20240521192614.3937942-1-acme@kernel.org> <CAHk-=wiWvtFyedDNpoV7a8Fq_FpbB+F5KmWK2xPY3QoYseOf_A@mail.gmail.com>
 <ZlFE-wbwStOqt8Ra@x1> <ZlFGpxWGQskCTjeK@x1> <CAP-5=fXDdcjMmn8iBenjPmZZQdB=AX+gc4TYDsHXuwH9TYq4Ng@mail.gmail.com>
 <CAHk-=wheZptGieaObmQEsz6bocUjhQXNpWXFDmCK-TOKbOvO+Q@mail.gmail.com>
In-Reply-To: <CAHk-=wheZptGieaObmQEsz6bocUjhQXNpWXFDmCK-TOKbOvO+Q@mail.gmail.com>
From: Namhyung Kim <namhyung@kernel.org>
Date: Sat, 25 May 2024 00:37:57 -0700
Message-ID: <CAM9d7chXVsoNP6uYMCqy2MZOiWkt4GrFn+giYLHQjaJRsap1Cw@mail.gmail.com>
Subject: Re: [GIT PULL] perf tools changes for v6.10
To: Ian Rogers <irogers@google.com>, Linus Torvalds <torvalds@linux-foundation.org>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>, Leo Yan <leo.yan@linux.dev>, 
	Mark Rutland <mark.rutland@arm.com>, Ingo Molnar <mingo@kernel.org>, 
	Thomas Gleixner <tglx@linutronix.de>, Jiri Olsa <jolsa@kernel.org>, 
	Adrian Hunter <adrian.hunter@intel.com>, Clark Williams <williams@redhat.com>, 
	Kate Carcia <kcarcia@redhat.com>, linux-kernel@vger.kernel.org, 
	linux-perf-users@vger.kernel.org, Anne Macedo <retpolanne@posteo.net>, 
	Bhaskar Chowdhury <unixbhaskar@gmail.com>, Ethan Adams <j.ethan.adams@gmail.com>, 
	James Clark <james.clark@arm.com>, Kan Liang <kan.liang@linux.intel.com>, 
	Thomas Richter <tmricht@linux.ibm.com>, Tycho Andersen <tycho@tycho.pizza>, 
	Yang Jihong <yangjihong@bytedance.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

Hi guys,

On Fri, May 24, 2024 at 9:20=E2=80=AFPM Linus Torvalds
<torvalds@linux-foundation.org> wrote:
>
> On Fri, 24 May 2024 at 20:48, Ian Rogers <irogers@google.com> wrote:
> >
> > Thanks for the report. TL;DR try putting the PMU name with the event
> > name, so "cycles:pp" becomes "armv8_pmuv3_0/cycles/pp", where
> > armv8_pmuv3_0 is the name of the PMU from /sys/devices.
>
> What? No.
>
> If that is the new rule, then I'm just going to revert. I'm not going
> to use some random different PMU names across architectures when all I
> want is just "cycles".
>
> In fact, the "cycles:pp" is a complete red herring. Just doing a simple
>
>  $ perf record sleep 1
>
> with no explicit expression at all, results in that same
>
>   Error:
>   cycles:P: PMU Hardware doesn't support sampling/overflow-interrupts.
> Try 'perf stat'
>
> because perf *ITSELF* uses the sane format by default.

Yep, so I'm curious what makes it fail.  IIUC the commit in question
would convert "cycles" event to "${whatever_pmu}/cycles/" if the pmu
has "events/cycles" alias in sysfs or in JSON.  But it should work after
that too. :(

It seems my system doesn't have the alias (both in x86_64 and arm64)
at least in sysfs.  I think that's why I don't see the failure and maybe
there's a specific hardware issue - like a M1 macbook issue?  IIRC it
required the exclude_guest bit to be set, but I think we handled it already
so this must be a different issue.

The error message is for EOPNOTSUPP, and I don't think it will set any
special attributes for the default event.  So I have no clue yet..

Thanks,
Namhyung

>
> So ABSOLUTELY NO. We're not changing the rules to "You have to know
> some idiotic per-architecture JSON magic".
>
> I don't want any excuses like this. No "You are holding it wrong".
> This is a BUG. Treat it as such.
>
> And yes, "perf record -vv" shows that it apparently picked some insane
> arm_dsu_0 event. Which just shows that trying to parse the JSON rules
> instead of just having sane defaults is clearly not working.
>
> So here's the deal: I will revert your commit tomorrow unless you show
> that you are taking it seriously and have a sane fix.
>
> Because no, "You are holding it wrong" is not the answer. Never was,
> never will be. Things used to work, they don't work any more. That
> shit is not acceptable.
>
>                    Linus