[Intel-xe] [PATCH] drm/xe : Fix GT looping for standalone media

[Matt Roper]

On Tue, Jun 13, 2023 at 02:39:45PM -0700, Matt Roper wrote:
> On Tue, Jun 13, 2023 at 03:12:32PM +0530, Riana Tauro wrote:
> > Currently the id of primary gt is set using gt_count and not
> > the media gt.
> > 
> > set gt->info.id of media gt using gt_count
> > 
> > Signed-off-by: Riana Tauro <riana.tauro at intel.com>
> > ---
> >  drivers/gpu/drm/xe/xe_pci.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/drivers/gpu/drm/xe/xe_pci.c b/drivers/gpu/drm/xe/xe_pci.c
> > index bd1f59b49928..2991cf5365d5 100644
> > --- a/drivers/gpu/drm/xe/xe_pci.c
> > +++ b/drivers/gpu/drm/xe/xe_pci.c
> > @@ -590,7 +590,7 @@ static int xe_info_init(struct xe_device *xe,
> >  		 * up with platforms that support both together.
> >  		 */
> >  		drm_WARN_ON(&xe->drm, id != 0);
> > -		gt->info.id = 1;
> > +		gt->info.id = xe->info.gt_count++;
> 
> 1 is always correct today since MTL is the only platform with a
> standalone media GT, and it's a single-tile platform (so the GTs are
> 0/1).  This will also be true for any future single-tile platforms with
> standalone media.
> 
> If/when we eventually have a platform that's both multi-tile _and_
> multi-gt, we need to decide how we're going to identify GTs.  Since no
> such platform exists today, we also don't know for sure whether every
> tile will have both GTs, or whether only some of the tiles will have
> multiple GTs.  It's not clear that using gt_count++ will necessarily be
> the right thing to do; maybe we want to make primary GTs 2*tile and
> media GTs 2*tile+1 (i.e., intentionally skip IDs if some of the tiles
> have both and others do not).  We may also want to adjust how we
> approach GT identification in the uapi in general --- maybe we don't
> even want to give GTs a global ID anymore and instead want to move
> toward identifying GT with a (tile_id, gt_id) tuple where gt_id is an
> intra-tile number (i.e., 0=primary, 1=media).
> 
> The current GT and tile series aimed to just preserve the existing uapi
> without changes, and using an ID of 1 for media (with all the FIXME
> comments in the code) does that for now.  But sometime soon we do need
> to make a formal decision on exactly how we want to work with tiles and
> GTs in the uapi, going forward and that needs to be something that we
> loop in our userspace partners on; we can't just make the decision
> ourselves on the kernel side.

So looking at this patch again, the important part isn't actually the ID
of the GT (that's already being set correctly); what's important is that
we need to bump the GT count again when initializing the media GT,
otherwise gt_count will stay "1" forever, even when it should be 2.  So
with an updated commit message that focuses on the gt_count rather than
the ID,

Reviewed-by: Matt Roper <matthew.d.roper at intel.com>

I can make that tweak to the commit message while applying; no need to
re-send.


Matt

> 
> 
> Matt
> 
> >  	}
> >  
> >  	return 0;
> > -- 
> > 2.40.0
> > 
> 
> -- 
> Matt Roper
> Graphics Software Engineer
> Linux GPU Platform Enablement
> Intel Corporation

-- 
Matt Roper
Graphics Software Engineer
Linux GPU Platform Enablement
Intel Corporation