Re: [PATCH v6 1/6] gpu: rfc: Proposal for a GPU cgroup controller

["T.J. Mercier" <tjmercier@google.com>]

On Wed, May 4, 2022 at 5:10 AM Michal Koutný <mkoutny@suse.com> wrote:
>
> Hello.
>
> On Mon, May 02, 2022 at 11:19:35PM +0000, "T.J. Mercier" <tjmercier@google.com> wrote:
> > --- a/Documentation/admin-guide/cgroup-v2.rst
> > +++ b/Documentation/admin-guide/cgroup-v2.rst
> > [...]
> > +     The device name string is set by a device driver when it registers with the
> > +     GPU cgroup controller to participate in resource accounting.
>
> Are these names available anywhere else for the user? (I.e. would
> drivers add respective sysfs attributes or similar?)
>
Hi, this sounds like it could be a good idea but it'd probably be best
to do this inside gpucg_register_bucket instead of requiring drivers
to perform this externally, possibly in a non-uniform way. Maybe a
sysfs file that prints each name of the gpucg_buckets elements?
However the only names that would result from this series are the
names of the dma-buf heaps, with "-heap" appended. So they are
predictable from the /dev/dma_heap/* names, and only the system and
cma heaps currently exist upstream.

For other future uses of this controller I thought we were headed in
the direction of "standardized" names which would be
predefined/hardcoded and documented, so these names wouldn't really
need to be made available to a user at runtime.
https://lore.kernel.org/lkml/CABdmKX3gTAohaOwkNccGrQyXN9tzT-oEVibO5ZPF+eP+Vq=AOg@mail.gmail.com/
>
> > +     Non-unique names will be rejected at the point of registration.
>
> This doesn't seem relevant to the cgroupfs user, does it?
> I think it should be mentioned at the respective API.
>
Yeah you're right. Thank you.

> HTH,
> Michal
>