Whenever we want to unbind a vma, we must wait on all GPU activity to complete first. (This is what gives us the ability to do fine grained eviction and purging by only having to wait on the VMA that we need to unbind to proceed; though if pushed we can make it a rule that we are only allowed to unbind already idle VMA and move the burden of the work and organising the sleep onto the caller.) Currently, we might only sleep if the vma is still active on the GPU, but in principle i915_vma_unbind() always implies a sleep, so mark it up with a might_sleep(). Suggested-by: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx> References: https://bugs.freedesktop.org/show_bug.cgi?id=103638 Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> Cc: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx> Cc: Joonas Lahtinen <joonas.lahtinen@xxxxxxxxxxxxxxx> Cc: Matthew Auld <matthew.william.auld@xxxxxxxxx> Cc: Daniel Vetter <daniel.vetter@xxxxxxxx> Reviewed-by: Joonas Lahtinen <joonas.lahtinen@xxxxxxxxxxxxxxx> --- drivers/gpu/drm/i915/i915_vma.c | 1 + 1 file changed, 1 insertion(+) diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c index b7ac84db1d2e..bf6d8d1eaabe 100644 --- a/drivers/gpu/drm/i915/i915_vma.c +++ b/drivers/gpu/drm/i915/i915_vma.c @@ -743,6 +743,7 @@ int i915_vma_unbind(struct i915_vma *vma) /* First wait upon any activity as retiring the request may * have side-effects such as unpinning or even unbinding this vma. */ + might_sleep(); active = i915_vma_get_active(vma); if (active) { int idx; -- 2.15.0 _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
