Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20220719223946.176299-1-irogers@google.com> <20220719223946.176299-2-irogers@google.com>
 <ef5125e0-8265-8c16-e904-c95c8ddc2754@maine.edu>
In-Reply-To: <ef5125e0-8265-8c16-e904-c95c8ddc2754@maine.edu>
From:   Ian Rogers <irogers@google.com>
Date:   Wed, 20 Jul 2022 08:18:04 -0700
Message-ID: <CAP-5=fVU17DxzdOvr0aob6G4HvjianSuWW4Q-QRWQV=Kd8_+jw@mail.gmail.com>
Subject: Re: [PATCH v3 1/3] perf: Align user space counter reading with code
To:     Vince Weaver <vincent.weaver@maine.edu>
Cc:     Peter Zijlstra <peterz@infradead.org>,
        Ingo Molnar <mingo@redhat.com>,
        Arnaldo Carvalho de Melo <acme@kernel.org>,
        Mark Rutland <mark.rutland@arm.com>,
        Alexander Shishkin <alexander.shishkin@linux.intel.com>,
        Jiri Olsa <jolsa@kernel.org>,
        Namhyung Kim <namhyung@kernel.org>,
        Kajol Jain <kjain@linux.ibm.com>,
        Andi Kleen <ak@linux.intel.com>,
        Adrian Hunter <adrian.hunter@intel.com>,
        Anshuman Khandual <anshuman.khandual@arm.com>,
        linux-kernel@vger.kernel.org, linux-perf-users@vger.kernel.org,
        Rob Herring <robh@kernel.org>,
        Stephane Eranian <eranian@google.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Wed, Jul 20, 2022 at 8:06 AM Vince Weaver <vincent.weaver@maine.edu> wrote:
>
> On Tue, 19 Jul 2022, Ian Rogers wrote:
>
> > Align the user space counter reading documentation with the code in
> > perf_mmap__read_self. Previously the documentation was based on the perf
> > rdpmc test, but now general purpose code is provided by libperf.
> >
> > Signed-off-by: Ian Rogers <irogers@google.com>
> > ---
> >  include/uapi/linux/perf_event.h       | 35 +++++++++++++++++----------
> >  tools/include/uapi/linux/perf_event.h | 35 +++++++++++++++++----------
> >  2 files changed, 44 insertions(+), 26 deletions(-)
> >
> > diff --git a/include/uapi/linux/perf_event.h b/include/uapi/linux/perf_event.h
> > index d37629dbad72..6826dabb7e03 100644
> > --- a/include/uapi/linux/perf_event.h
> > +++ b/include/uapi/linux/perf_event.h
> > @@ -538,9 +538,13 @@ struct perf_event_mmap_page {
> >        *
> >        *     if (pc->cap_usr_time && enabled != running) {
> >        *       cyc = rdtsc();
> > -      *       time_offset = pc->time_offset;
> >        *       time_mult   = pc->time_mult;
> >        *       time_shift  = pc->time_shift;
> > +      *       time_offset = pc->time_offset;
> > +      *       if (pc->cap_user_time_short) {
> > +      *         time_cycles = pc->time_cycles;
> > +      *         time_mask = pc->time_mask;
> > +      *       }
>
> From what I've been told, and from what perf_mmap__read_self() actually
> does, many of these MMAP fields need to be accessed by READ_ONCE()
> (a GPLv2 only interface) to be correct.
>
> Should we update perf_event.h to reflect this?  Otherwise it's confusing
> when the actual code and the documentation in the header don't match like
> this.  As an example, see the actual code snippets from
> perf_mmap__read_self()
>
>                 seq = READ_ONCE(pc->lock);
>                 barrier();
>
>                 count->ena = READ_ONCE(pc->time_enabled);
>                 count->run = READ_ONCE(pc->time_running);
>
>                 if (pc->cap_user_time && count->ena != count->run) {
>                         cyc = read_timestamp();
>                         time_mult = READ_ONCE(pc->time_mult);
>                         time_shift = READ_ONCE(pc->time_shift);
>                         time_offset = READ_ONCE(pc->time_offset);
>
>                         if (pc->cap_user_time_short) {
>                                 time_cycles = READ_ONCE(pc->time_cycles);
>                                 time_mask = READ_ONCE(pc->time_mask);
>                         }
>                 }
>
>                 idx = READ_ONCE(pc->index);
>                 cnt = READ_ONCE(pc->offset);
>
> ...

Thanks! So I think what this patch proposes is an improvement,
specifically it aligns it better with the code and deals with the
divide by zero. I think what you're asking for makes sense but as you
point out READ_ONCE probably isn't the correct API for something
outside the kernel. Given the kernel is now C11:
https://lwn.net/Articles/885941/
This opens the possibility of using stdatomic.h, so perhaps we can
move these variables to more correct atomic types. So, I think we can
land this and worry about the atomic API in a follow up.

Thanks,
Ian

> Vince