[Intel-gfx] [PATCH v2 0/3] drm/i915/gt: Expose per gt defaults in sysfs

[Dixit, Ashutosh]

On Thu, 02 Jun 2022 16:55:50 -0700, Matt Roper wrote:
>
> On Thu, Jun 02, 2022 at 04:36:02PM -0700, Dixit, Ashutosh wrote:
> > On Fri, 27 May 2022 16:41:28 -0700, Matt Roper wrote:
> > >
> > > On Thu, May 26, 2022 at 12:00:42PM -0700, Ashutosh Dixit wrote:
> > > > Create a gt/gtN/.defaults directory (similar to
> > > > engine/<engine-name>/.defaults) to expose default parameter values for each
> > > > gt in sysfs. This allows userspace to restore default parameter values
> > > > after they may have changed.
> > >
> > > Drive-by comment:  since this is a GT-related change, I think we're
> > > required to Cc: dri-devel as well.
> >
> > Sure.
> >
> > > And since this is new ABI, we'd need an open source userspace consumer
> > > or a very clear explanation why this is an exception to the rule (e.g.,
> > > if these are meant to be twiddled directly by the user rather than
> > > accessed by userspace software).
> >
> > This series is in response to the following oneAPI requirement ("setting to
> > -1 will return the frequency to the factory value"):
> >
> > https://spec.oneapi.io/level-zero/latest/sysman/api.html?highlight=zesfrequencysetrange#zesfrequencysetrange
> > https://spec.oneapi.io/level-zero/latest/sysman/api.html?highlight=zesfrequencysetrange#_CPPv416zes_freq_range_t
> >
> > So maybe an open-source oneAPI user mode driver will be acceptable
> > userspace? So yes not merely to be twiddled just by the user.
>
> oneapi/l0 is implemented at https://github.com/intel/compute-runtime,
> right?  I think that should be fine as long as we provide a link to the
> actual implementation code that's accessing the new sysfs directory.
> Since the userspace code probably won't fully land until after the
> kernel change lands, this would be something like a pull request
> against their repo that's reviewed and fully-ready to go.

Sure, thanks.