Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <20200211011631.7619-1-zzyiwei@google.com> <20200210211951.1633c7d0@rorschach.local.home>
 <CAKT=dDm+UKqa7j744iTsvYs+jqrdOHpTqdksRUjDe-6vqkigew@mail.gmail.com> <20200210221521.59928416@rorschach.local.home>
In-Reply-To: <20200210221521.59928416@rorschach.local.home>
From:   Yiwei Zhang <zzyiwei@google.com>
Date:   Wed, 12 Feb 2020 11:26:08 -0800
Message-ID: <CAKT=dDk+CiMQ_-f6Daa_ea2FOW=De6PKmcyiGrm4KEkVbH2fDQ@mail.gmail.com>
Subject: Re: [PATCH] Add gpu memory tracepoints
To:     Steven Rostedt <rostedt@goodmis.org>
Cc:     mingo@redhat.com, linux-kernel@vger.kernel.org,
        Prahlad Kilambi <prahladk@google.com>,
        Joel Fernandes <joelaf@google.com>,
        android-kernel <android-kernel@google.com>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Hi Steven,

I can move the stuff out from the kernel/trace. Then can we still
leave include/trace/events/gpu_mem.h where it is right now? Or do we
have to move that out as well? Because we would need a non-drm common
header place for the tracepoint so that downstream drivers can find
the tracepoint definition.

Best,
Yiwei

On Mon, Feb 10, 2020 at 7:15 PM Steven Rostedt <rostedt@goodmis.org> wrote:
>
> On Mon, 10 Feb 2020 19:05:35 -0800
> Yiwei Zhang <zzyiwei@google.com> wrote:
>
> > Thanks for the prompt reply!
> >
> > The tracepoint proposed here is for tracking global gpu memory usage
> > total counter and per-process gpu memory usage total counter. The
> > tracepoint is for gfx drivers who have implemented gpu memory tracking
> > system. The tracepoint expects the de-duplication of the shared memory
> > is done inside the tracking system.
> >
> > On Android, the graphics driver has implemented gpu memory tracking.
> > First, we'd like to profiler GPU memory with this tracepoint. Second,
> > we implement eBPF programs and attach to this tracepoint for tracking
> > GPU memory at runtime on production devices. However, the tracepoint +
> > eBPF approach requires the tracepoint to be upstreamed so that it's
> > considered a stable interface which Android common kernel can carry it
> > forever.
>
>
> Then it needs to live in the drivers/gpu directory. kernel/trace is for
> tracing infrastructure and not for adding trace points.
>
> -- Steve