| Message ID | 1477404024-2864-2-git-send-email-tvrtko.ursulin@linux.intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Tue, Oct 25, 2016 at 03:00:24PM +0100, Tvrtko Ursulin wrote: > From: Tvrtko Ursulin <tvrtko.ursulin@intel.com> > > Objects can have multiple VMAs used for display in which > case assertion that objects must not be pinned for display > more times than their single display related VMA is incorrect. But each of those vma are pinned, as many times as they are used by display. This explanation doesn't hold. -Chris
On 25/10/2016 15:09, Chris Wilson wrote: > On Tue, Oct 25, 2016 at 03:00:24PM +0100, Tvrtko Ursulin wrote: >> From: Tvrtko Ursulin <tvrtko.ursulin@intel.com> >> >> Objects can have multiple VMAs used for display in which >> case assertion that objects must not be pinned for display >> more times than their single display related VMA is incorrect. > > But each of those vma are pinned, as many times as they are used by > display. This explanation doesn't hold. We can have a normal and a rotated vma, each pinned once, object pinned therefore pinned to display twice. Regards, Tvrtko
On 25/10/2016 15:12, Tvrtko Ursulin wrote: > > On 25/10/2016 15:09, Chris Wilson wrote: >> On Tue, Oct 25, 2016 at 03:00:24PM +0100, Tvrtko Ursulin wrote: >>> From: Tvrtko Ursulin <tvrtko.ursulin@intel.com> >>> >>> Objects can have multiple VMAs used for display in which >>> case assertion that objects must not be pinned for display >>> more times than their single display related VMA is incorrect. >> >> But each of those vma are pinned, as many times as they are used by >> display. This explanation doesn't hold. > > We can have a normal and a rotated vma, each pinned once, object pinned > therefore pinned to display twice. P.S. I need a fixes tag for this as well. Looking for a suitable candidate in the VMA rewrite series.. suggestions welcome. :) Regards, Tvrtko
On Tue, Oct 25, 2016 at 03:12:51PM +0100, Tvrtko Ursulin wrote: > > On 25/10/2016 15:09, Chris Wilson wrote: > >On Tue, Oct 25, 2016 at 03:00:24PM +0100, Tvrtko Ursulin wrote: > >>From: Tvrtko Ursulin <tvrtko.ursulin@intel.com> > >> > >>Objects can have multiple VMAs used for display in which > >>case assertion that objects must not be pinned for display > >>more times than their single display related VMA is incorrect. > > > >But each of those vma are pinned, as many times as they are used by > >display. This explanation doesn't hold. > > We can have a normal and a rotated vma, each pinned once, object > pinned therefore pinned to display twice. Oh, vma_pin_count not obj_pin_count. Bleh. Is the warn useful to keep, and so we need to iterate over all vma? Probably not. Especially as we then start asking questions like, should we track pin_display on the vma as well for better accuracy in their assertion s/their single dislay related/the current/ Ok, I don't have a great idea for replacing it with some other early warning signal. With the slight tweak to the changelog, Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> -Chris
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c index 0e26ee96856e..96f48d2c971c 100644 --- a/drivers/gpu/drm/i915/i915_gem.c +++ b/drivers/gpu/drm/i915/i915_gem.c @@ -3595,8 +3595,6 @@ i915_gem_object_pin_to_display_plane(struct drm_i915_gem_object *obj, vma->display_alignment = max_t(u64, vma->display_alignment, alignment); - WARN_ON(obj->pin_display > i915_vma_pin_count(vma)); - i915_gem_object_flush_cpu_write_domain(obj); old_write_domain = obj->base.write_domain; @@ -3633,7 +3631,6 @@ i915_gem_object_unpin_from_display_plane(struct i915_vma *vma) list_move_tail(&vma->vm_link, &vma->vm->inactive_list); i915_vma_unpin(vma); - WARN_ON(vma->obj->pin_display > i915_vma_pin_count(vma)); } /**