[PATCH v2] drm/xe/bo: add GPU memory trace points

[Lucas De Marchi]

On Wed, May 28, 2025 at 08:01:21PM +0000, Juston Li wrote:
>On Wed, 2025-05-28 at 18:34 +0000, Juston Li wrote:
>> On Thu, 2025-05-22 at 16:14 +0100, Tvrtko Ursulin wrote:
>> >
>> > On 22/05/2025 15:50, Lucas De Marchi wrote:
>> > > + dri-devel
>> > >
>> > > On Wed, May 21, 2025 at 10:42:35PM +0000, Juston Li wrote:
>> > > > Add tracepoints behind CONFIG_DRM_XE_GPU_MEM_TRACEPOINTS for
>> > > > tracking
>> > > > global and per-process GPU memory usage.
>> > > >
>> > > > These are required by VSR on Android 12+ for reporting GPU
>> > > > driver
>> > > > memory
>> > > > allocations.
>> > > >
>> > > > v2:
>> > > > - Use u64 as preferred by checkpatch (Tvrtko)
>> > > > - Fix errors in comments/Kconfig description (Tvrtko)
>> > > > - drop redundant "CONFIG_" in Kconfig
>> > > >
>> > > > Signed-off-by: Juston Li <justonli at chromium.org>
>> > > > Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin at igalia.com>
>> > > > ---
>> > > > drivers/gpu/drm/xe/Kconfig.debug     | 12 +++++++
>> > > > drivers/gpu/drm/xe/xe_bo.c           | 47
>> > > > ++++++++++++++++++++++++++++
>> > > > drivers/gpu/drm/xe/xe_device_types.h | 16 ++++++++++
>> > > > 3 files changed, 75 insertions(+)
>> > > >
>> > > > diff --git a/drivers/gpu/drm/xe/Kconfig.debug
>> > > > b/drivers/gpu/drm/xe/
>> > > > Kconfig.debug
>> > > > index 01735c6ece8ba..2371eeda0afd5 100644
>> > > > --- a/drivers/gpu/drm/xe/Kconfig.debug
>> > > > +++ b/drivers/gpu/drm/xe/Kconfig.debug
>> > > > @@ -111,3 +111,15 @@ config DRM_XE_USERPTR_INVAL_INJECT
>> > > >
>> > > >      Recommended for driver developers only.
>> > > >      If in doubt, say "N".
>> > > > +
>> > > > +config DRM_XE_GPU_MEM_TRACEPOINTS
>> > >
>> > > is there any particular reason to make this user-configurable per
>> > > driver?
>> > > Why aren't we making CONFIG_TRACE_GPU_MEM configurable (if
>> > > needed,
>> > > but
>> > > could just depend on CONFIG_TRACEPOINTS) and then drivers just
>> > > use
>> > > it.
>> >
>> > Could be done like that too. Msm does unconditional select
>> > TRACE_GPU_MEM
>> > which I thought wouldn't be acceptable so I suggested making it
>> > configurable.
>>
>> Ok yeah, I don't see a reason to make it per-driver either. I'll make
>> CONFIG_TRACE_GPU_MEM configurable since we don't have
>> CONFIG_TRACEPOINTS or CONFIG_TRACING enabled on GKI.
>>
>> Juston
>
>Scratch that, CONFIG_TRACEPOINTS is enabled on GKI so we could do the
>depend.
>
>But now that I think about it, CONFIG_TRACEPOINTS seems enabled so
>commonly I wonder if we might as well just unconditional select like
>Msm as Tvrtko mentioned and then not have to deal with the #ifdefs?

we are trying to remove the selects... they are too cumbersome.
See e.g. https://lore.kernel.org/intel-xe/704fd2b9-04da-4ec8-b854-22bc3ce9058e@app.fastmail.com

We should **depend** on other parts of the kernel, not select them.

>
>If that's not acceptable, I'm leaning on just making
>CONFIG_TRACE_GPU_MEM configurable.

sounds good to me.

thanks
Lucas De Marchi