Currently, we may skip the clflush on an object if the object has not
yet allocated any pages. However, the object may still be polluting the
CPU cache and may now be in a coherent uncached domain - and so no
longer generating clflushing.
Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
---
drivers/gpu/drm/i915/i915_gem.c | 21 +++++++++++++--------
1 file changed, 13 insertions(+), 8 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 6845cf03287c..e43fad3ccf24 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -3182,20 +3182,25 @@ i915_vma_insert(struct i915_vma *vma, u64 size, u64 alignment, u64 flags)
void i915_gem_clflush_object(struct drm_i915_gem_object *obj,
bool force)
{
- /* If we don't have a page list set up, then we're not pinned
- * to GPU, and we can ignore the cache flush because it'll happen
- * again at bind time.
- */
- if (!obj->mm.pages)
- return;
-
/*
* Stolen memory is always coherent with the GPU as it is explicitly
* marked as wc by the system, or the system is cache-coherent.
+ * Similarly, we only access struct pages through the CPU cache, so
+ * anything not backed by physical memory we consider to be always
+ * coherent and not need clflushing.
*/
- if (obj->stolen || obj->phys_handle)
+ if (!i915_gem_object_has_struct_page(obj))
return;
+ /* If we don't have a page list set up, then we're not pinned
+ * to GPU, and we can ignore the cache flush because it'll happen
+ * again at bind time.
+ */
+ if (!obj->mm.pages) {
+ obj->cache_dirty = true;
+ return;
+ }
+
/* If the GPU is snooping the contents of the CPU cache,
* we do not need to manually clear the CPU cache lines. However,
* the caches are only snooped when the render cache is
--
2.10.2
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
