[RFC 1/2] drm/amdgpu: Remove dynamic DRM scheduling priority override

Alex Deucher alexdeucher at gmail.com
Mon Sep 9 16:58:25 UTC 2024


On Mon, Sep 9, 2024 at 8:23 AM Christian König <christian.koenig at amd.com> wrote:
>
> Adding Leo as well.
>
> Am 06.09.24 um 20:06 schrieb Tvrtko Ursulin:
> > From: Tvrtko Ursulin <tvrtko.ursulin at igalia.com>
> >
> > According to Christian the dynamic DRM priority override was only
> > interesting before the hardware priority (dona via
> > drm_sched_entity_modify_sched()) existed. Furthermore, both
> > overrides also only work somewhat on paper while in reality they are only
> > effective if the entity is idle, which is something userspace is unaware
> > of when using the AMDGPU_SCHED_OP_*_PRIORITY_OVERRIDE uapi.
> >
> > Therefore follow Christian's advice and remove this call completely.
>
> The only potential use case I can come up with would be for multimedia
> engines since we never implemented the hardware priority handling for them.
>
> @Leo do we have any use case relying on that? As far as I know it
> doesn't work for UVD/VCE anyway because those engines can't switch loads
> and for VCN we should probably just implement different hw priorities.

VCN has moved away from multiple queues to a single queue; and the
next iteration is user queues.  On the older hardware, there was a
high priority queue, but IIRC, the firmware just prioritized fetches
from that queue; there is still only one engine that can process the
work.  As such, I'd expect it to operate similarly to the drm
scheduler.  For both GFX and VCN, having a high priority queue, even
if it was just in the drm scheduler, helped at least some workloads we
had in the past.

Alex

>
> Christian.
>
> >
> > Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin at igalia.com>
> > Cc: Christian König <christian.koenig at amd.com>
> > Cc: Alex Deucher <alexander.deucher at amd.com>
> > ---
> >   drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 4 ----
> >   1 file changed, 4 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
> > index c43d1b6e5d66..2480b3227dad 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
> > @@ -820,10 +820,6 @@ static void amdgpu_ctx_set_entity_priority(struct amdgpu_ctx *ctx,
> >       struct drm_gpu_scheduler **scheds = NULL;
> >       unsigned num_scheds;
> >
> > -     /* set sw priority */
> > -     drm_sched_entity_set_priority(&aentity->entity,
> > -                                   amdgpu_ctx_to_drm_sched_prio(priority));
> > -
> >       /* set hw priority */
> >       if (hw_ip == AMDGPU_HW_IP_COMPUTE || hw_ip == AMDGPU_HW_IP_GFX) {
> >               hw_prio = amdgpu_ctx_get_hw_prio(ctx, hw_ip);
>


More information about the amd-gfx mailing list