drm/i915: Flush extra hard after writing relocations through the GTT

Recently discovered in commit bdae33b8 ("drm/i915: Use maximum write
flush for pwrite_gtt") was that we needed to our full write barrier
before changing the GGTT PTE to ensure that our indirect writes through
the GTT landed before the PTE changed (and the writes end up in a
different page). That also applies to our GGTT relocation path.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: stable@vger.kernel.org
Reviewed-by: Prathap Kumar Valsan <prathap.kumar.valsan@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190730112151.5633-4-chris@chris-wilson.co.uk

drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c

@@ -1014,11 +1014,12 @@ static void reloc_cache_reset(struct reloc_cache *cache)
 		kunmap_atomic(vaddr);
 		i915_gem_object_finish_access((struct drm_i915_gem_object *)cache->node.mm);
 	} else {
-		wmb();
+		struct i915_ggtt *ggtt = cache_to_ggtt(cache);
+
+		intel_gt_flush_ggtt_writes(ggtt->vm.gt);
 		io_mapping_unmap_atomic((void __iomem *)vaddr);
-		if (cache->node.allocated) {
-			struct i915_ggtt *ggtt = cache_to_ggtt(cache);
 
+		if (cache->node.allocated) {
 			ggtt->vm.clear_range(&ggtt->vm,
 					     cache->node.start,
 					     cache->node.size);
@@ -1073,6 +1074,7 @@ static void *reloc_iomap(struct drm_i915_gem_object *obj,
 	void *vaddr;
 
 	if (cache->vaddr) {
+		intel_gt_flush_ggtt_writes(ggtt->vm.gt);
 		io_mapping_unmap_atomic((void __force __iomem *) unmask_page(cache->vaddr));
 	} else {
 		struct i915_vma *vma;
@@ -1114,7 +1116,6 @@ static void *reloc_iomap(struct drm_i915_gem_object *obj,
 
 	offset = cache->node.start;
 	if (cache->node.allocated) {
-		wmb();
 		ggtt->vm.insert_page(&ggtt->vm,
 				     i915_gem_object_get_dma_address(obj, page),
 				     offset, I915_CACHE_NONE, 0);