[PATCH 0/3] drm/panfrost: Expose HW counters to userspace

[Rob Clark]

On Tue, Apr 30, 2019 at 5:42 AM Boris Brezillon
<boris.brezillon at collabora.com> wrote:
>
> +Rob, Eric, Mark and more
>
> Hi,
>
> On Fri, 5 Apr 2019 16:20:45 +0100
> Steven Price <steven.price at arm.com> wrote:
>
> > On 04/04/2019 16:20, Boris Brezillon wrote:
> > > Hello,
> > >
> > > This patch adds new ioctls to expose GPU counters to userspace.
> > > These will be used by the mesa driver (should be posted soon).
> > >
> > > A few words about the implementation: I followed the VC4/Etnaviv model
> > > where perf counters are retrieved on a per-job basis. This allows one
> > > to have get accurate results when there are users using the GPU
> > > concurrently.
> > > AFAICT, the mali kbase is using a different approach where several
> > > users can register a performance monitor but with no way to have fined
> > > grained control over what job/GPU-context to track.
> >
> > mali_kbase submits overlapping jobs. The jobs on slot 0 and slot 1 can
> > be from different contexts (address spaces), and mali_kbase also fully
> > uses the _NEXT registers. So there can be a job from one context
> > executing on slot 0 and a job from a different context waiting in the
> > _NEXT registers. (And the same for slot 1). This means that there's no
> > (visible) gap between the first job finishing and the second job
> > starting. Early versions of the driver even had a throttle to avoid
> > interrupt storms (see JOB_IRQ_THROTTLE) which would further delay the
> > IRQ - but thankfully that's gone.
> >
> > The upshot is that it's basically impossible to measure "per-job"
> > counters when running at full speed. Because multiple jobs are running
> > and the driver doesn't actually know when one ends and the next starts.
> >
> > Since one of the primary use cases is to draw pretty graphs of the
> > system load [1], this "per-job" information isn't all that relevant (and
> > minimal performance overhead is important). And if you want to monitor
> > just one application it is usually easiest to ensure that it is the only
> > thing running.
> >
> > [1]
> > https://developer.arm.com/tools-and-software/embedded/arm-development-studio/components/streamline-performance-analyzer
> >
> > > This design choice comes at a cost: every time the perfmon context
> > > changes (the perfmon context is the list of currently active
> > > perfmons), the driver has to add a fence to prevent new jobs from
> > > corrupting counters that will be dumped by previous jobs.
> > >
> > > Let me know if that's an issue and if you think we should approach
> > > things differently.
> >
> > It depends what you expect to do with the counters. Per-job counters are
> > certainly useful sometimes. But serialising all jobs can mess up the
> > thing you are trying to measure the performance of.
>
> I finally found some time to work on v2 this morning, and it turns out
> implementing global perf monitors as done in mali_kbase means rewriting
> almost everything (apart from the perfcnt layout stuff). I'm not against
> doing that, but I'd like to be sure this is really what we want.
>
> Eric, Rob, any opinion on that? Is it acceptable to expose counters
> through the pipe_query/AMD_perfmon interface if we don't have this
> job (or at least draw call) granularity? If not, should we keep the
> solution I'm proposing here to make sure counters values are accurate,
> or should we expose perf counters through a non-standard API?

I think if you can't do per-draw level granularity, then you should
not try to implement AMD_perfmon..  instead the use case is more for a
sort of "gpu top" app (as opposed to something like frameretrace which
is taking per-draw-call level measurements from within the app.
Things that use AMD_perfmon are going to, I think, expect to query
values between individual glDraw calls, and you probably don't want to
flush tile passes 500 times per frame.

(Although, I suppose if there are multiple GPUs where perfcntrs work
this way, it might be an interesting exercise to think about coming up
w/ a standardized API (debugfs perhaps?) to monitor the counters.. so
you could have a single userspace tool that works across several
different drivers.)

BR,
-R

>
> BTW, I'd like to remind you that serialization (waiting on the perfcnt
> fence) only happens if we have a perfmon context change between 2
> consecutive jobs, which only happens when
> * 2 applications are running in // and at least one of them is
>   monitored
> * or when userspace decides to stop monitoring things and dump counter
>   values
>
> That means that, for the usual case (all perfmons disabled), there's
> almost zero overhead (just a few more checks in the submit job code).
> That also means that, if we ever decide to support global perfmon (perf
> monitors that track things globably) on top of the current approach,
> and only global perfmons are enabled, things won't be serialized as
> with the per-job approach, because everyone will share the same perfmon
> ctx (the same set of perfmons).
>
> I'd appreciate any feedback from people that have used perf counters
> (or implemented a way to dump them) on their platform.
>
> Thanks,
>
> Boris