[PATCH v2 1/8] [ANDROID]: Add a new xe_user structure

[Dixit, Ashutosh]

On Fri, 22 Aug 2025 13:09:51 -0700, Rodrigo Vivi wrote:
>
> On Fri, Aug 22, 2025 at 10:01:25AM -0700, Matthew Brost wrote:
> > On Fri, Aug 22, 2025 at 11:48:54AM -0400, Rodrigo Vivi wrote:
> > > On Fri, Aug 22, 2025 at 08:59:23AM +0000, Aakash Deep Sarkar wrote:
> > > > For Android GPU work period event we need to track the runtime
> > > > on the GPU for each user id. This means we can have multiple
> > > > xe files opened by different processes/threads belonging to
> > > > the same user id. All these xe files need to be grouped together
> > > > so that one can easily identify these while calculating the
> > > > run time for the given user id.
> > > >
> > > > Currently, the xe driver doesn't record the user id of the
> > > > calling process. Also, all the xe files created using open
> > > > call are clubbed together inside the xe device structure
> > > > with no way to distinguish between them based on the user id
> > > > of the calling process.
> > >
> > > I thought I had already given this feedback, but I'm not sure if
> > > I forgot or if I was just ignored. I'm sorry either way.
> > >
> > > Android is not a justification. Please keep 'Android' mentions
> > > and ralated 'Android' justifications in the cover letter ONLY!
> > >
> > > The patch needs to make sense by itself. The patch needs to make
> > > sense in the currently linux upstream.
> > >
> >
> > So what are the rules here? There is another series [1] floating around
> > with a justification that Android needs this. Does that mean we can't
> > accept any Andriod only code upstream?
>
> I'm sorry for not being clear. Of course we can accept Android code
> upstream. I wish we had more (all?) Android code upstream ;)
>
> But we cannot add in xe, for instance, something with the namespace
> /sys/kernel/tracing/events/power/gpu_work_period where
> gpu_work_period is a definition in the Android tree only.
>
> We could add xe_work_period.

If Android user space consumes gpu_work_period, shouldn't gpu_work_period
be added at the drm level (rather than at xe level)?

I had a similar comment about gpu_frequency ftrace event here:
https://lore.kernel.org/intel-xe/87wm6vxedi.wl-ashutosh.dixit@intel.com/T/#me6005d00c09bf2198cc2a7465e780f50bd1ff291

Thanks.
--
Ashutosh