drm/i915: Remove check for conflicting relocation write-domains

Simply use the last write-domain set for the object in the batch,
trusting userspace to have correctly flushed the caches between usage as
a write target. This check dates back from the golden age of having only
a single operation per batch with the kernel repeating it for each
cliprect, and conflicts both with userspace trying to efficiently batch
multiple operations and with reducing the kernel overhead of relocation
processing.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drivers/gpu/drm/i915/i915_gem_execbuffer.c

@@ -150,17 +150,6 @@ i915_gem_execbuffer_relocate_entry(struct drm_i915_gem_object *obj,
 			  reloc->write_domain);
 		return ret;
 	}
-	if (unlikely(reloc->write_domain && target_obj->pending_write_domain &&
-		     reloc->write_domain != target_obj->pending_write_domain)) {
-		DRM_DEBUG("Write domain conflict: "
-			  "obj %p target %d offset %d "
-			  "new %08x old %08x\n",
-			  obj, reloc->target_handle,
-			  (int) reloc->offset,
-			  reloc->write_domain,
-			  target_obj->pending_write_domain);
-		return ret;
-	}
 
 	target_obj->pending_read_domains |= reloc->read_domains;
 	target_obj->pending_write_domain |= reloc->write_domain;