drm/i915/clflush: fixup handling of cache_dirty
authorMatthew Auld <matthew.auld@intel.com>
Wed, 27 Oct 2021 16:18:10 +0000 (17:18 +0100)
committerMatthew Auld <matthew.auld@intel.com>
Tue, 2 Nov 2021 09:43:10 +0000 (09:43 +0000)
In theory if clflush_work_create() somehow fails here, and we don't yet
have mm.pages populated then we end up resetting cache_dirty, which is
likely wrong, since that will potentially skip the flush-on-acquire, if
it was needed.

It looks like intel_user_framebuffer_dirty() can arrive here before the
pages are populated.

v2(Thomas):
  - Move setting cache_dirty out of the async portion, also add a
    comment for why that should still be safe.
v3:
  - Add Thomas' irc r-b

Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20211027161813.3094681-1-matthew.auld@intel.com
drivers/gpu/drm/i915/gem/i915_gem_clflush.c

index f0435c6feb68b579ee7086b12407748635628ebe..47586a8a1b7327cfcaaa7606c3165159f942898c 100644 (file)
@@ -109,12 +109,20 @@ bool i915_gem_clflush_object(struct drm_i915_gem_object *obj,
                                                I915_FENCE_GFP);
                dma_resv_add_excl_fence(obj->base.resv, &clflush->base.dma);
                dma_fence_work_commit(&clflush->base);
+               /*
+                * We must have successfully populated the pages(since we are
+                * holding a pin on the pages as per the flush worker) to reach
+                * this point, which must mean we have already done the required
+                * flush-on-acquire, hence resetting cache_dirty here should be
+                * safe.
+                */
+               obj->cache_dirty = false;
        } else if (obj->mm.pages) {
                __do_clflush(obj);
+               obj->cache_dirty = false;
        } else {
                GEM_BUG_ON(obj->write_domain != I915_GEM_DOMAIN_CPU);
        }
 
-       obj->cache_dirty = false;
        return true;
 }