drm/i915: Move GEM object domain management from struct_mutex to local

Use the per-object local lock to control the cache domain of the
individual GEM objects, not struct_mutex. This is a huge leap forward
for us in terms of object-level synchronisation; execbuffers are
coordinated using the ww_mutex and pread/pwrite is finally fully
serialised again.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190528092956.14910-10-chris@chris-wilson.co.uk

drivers/gpu/drm/i915/gem/selftests/huge_pages.c

@@ -960,10 +960,6 @@ static int gpu_write(struct i915_vma *vma,
 
 	GEM_BUG_ON(!intel_engine_can_store_dword(engine));
 
-	err = i915_gem_object_set_to_gtt_domain(vma->obj, true);
-	if (err)
-		return err;
-
 	batch = gpu_write_dw(vma, dword * sizeof(u32), value);
 	if (IS_ERR(batch))
 		return PTR_ERR(batch);
@@ -974,13 +970,19 @@ static int gpu_write(struct i915_vma *vma,
 		goto err_batch;
 	}
 
+	i915_vma_lock(batch);
 	err = i915_vma_move_to_active(batch, rq, 0);
+	i915_vma_unlock(batch);
 	if (err)
 		goto err_request;
 
 	i915_gem_object_set_active_reference(batch->obj);
 
-	err = i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
+	i915_vma_lock(vma);
+	err = i915_gem_object_set_to_gtt_domain(vma->obj, false);
+	if (err == 0)
+		err = i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
+	i915_vma_unlock(vma);
 	if (err)
 		goto err_request;
 

drivers/gpu/drm/i915/gem/selftests/i915_gem_coherency.c

@@ -78,7 +78,9 @@ static int gtt_set(struct drm_i915_gem_object *obj,
 	u32 __iomem *map;
 	int err;
 
+	i915_gem_object_lock(obj);
 	err = i915_gem_object_set_to_gtt_domain(obj, true);
+	i915_gem_object_unlock(obj);
 	if (err)
 		return err;
 
@@ -105,7 +107,9 @@ static int gtt_get(struct drm_i915_gem_object *obj,
 	u32 __iomem *map;
 	int err;
 
+	i915_gem_object_lock(obj);
 	err = i915_gem_object_set_to_gtt_domain(obj, false);
+	i915_gem_object_unlock(obj);
 	if (err)
 		return err;
 
@@ -131,7 +135,9 @@ static int wc_set(struct drm_i915_gem_object *obj,
 	u32 *map;
 	int err;
 
+	i915_gem_object_lock(obj);
 	err = i915_gem_object_set_to_wc_domain(obj, true);
+	i915_gem_object_unlock(obj);
 	if (err)
 		return err;
 
@@ -152,7 +158,9 @@ static int wc_get(struct drm_i915_gem_object *obj,
 	u32 *map;
 	int err;
 
+	i915_gem_object_lock(obj);
 	err = i915_gem_object_set_to_wc_domain(obj, false);
+	i915_gem_object_unlock(obj);
 	if (err)
 		return err;
 
@@ -176,7 +184,9 @@ static int gpu_set(struct drm_i915_gem_object *obj,
 	u32 *cs;
 	int err;
 
+	i915_gem_object_lock(obj);
 	err = i915_gem_object_set_to_gtt_domain(obj, true);
+	i915_gem_object_unlock(obj);
 	if (err)
 		return err;
 
@@ -215,7 +225,9 @@ static int gpu_set(struct drm_i915_gem_object *obj,
 	}
 	intel_ring_advance(rq, cs);
 
+	i915_vma_lock(vma);
 	err = i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
+	i915_vma_unlock(vma);
 	i915_vma_unpin(vma);
 
 	i915_request_add(rq);

drivers/gpu/drm/i915/gem/selftests/i915_gem_context.c

@@ -209,7 +209,9 @@ gpu_fill_dw(struct i915_vma *vma, u64 offset, unsigned long count, u32 value)
 	i915_gem_object_flush_map(obj);
 	i915_gem_object_unpin_map(obj);
 
+	i915_gem_object_lock(obj);
 	err = i915_gem_object_set_to_gtt_domain(obj, false);
+	i915_gem_object_unlock(obj);
 	if (err)
 		goto err;
 
@@ -261,7 +263,9 @@ static int gpu_fill(struct drm_i915_gem_object *obj,
 	if (IS_ERR(vma))
 		return PTR_ERR(vma);
 
+	i915_gem_object_lock(obj);
 	err = i915_gem_object_set_to_gtt_domain(obj, false);
+	i915_gem_object_unlock(obj);
 	if (err)
 		return err;
 
@@ -302,11 +306,15 @@ static int gpu_fill(struct drm_i915_gem_object *obj,
 	if (err)
 		goto err_request;
 
+	i915_vma_lock(batch);
 	err = i915_vma_move_to_active(batch, rq, 0);
+	i915_vma_unlock(batch);
 	if (err)
 		goto skip_request;
 
+	i915_vma_lock(vma);
 	err = i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
+	i915_vma_unlock(vma);
 	if (err)
 		goto skip_request;
 
@@ -754,7 +762,9 @@ emit_rpcs_query(struct drm_i915_gem_object *obj,
 	if (IS_ERR(vma))
 		return PTR_ERR(vma);
 
+	i915_gem_object_lock(obj);
 	err = i915_gem_object_set_to_gtt_domain(obj, false);
+	i915_gem_object_unlock(obj);
 	if (err)
 		return err;
 
@@ -780,11 +790,15 @@ emit_rpcs_query(struct drm_i915_gem_object *obj,
 	if (err)
 		goto err_request;
 
+	i915_vma_lock(batch);
 	err = i915_vma_move_to_active(batch, rq, 0);
+	i915_vma_unlock(batch);
 	if (err)
 		goto skip_request;
 
+	i915_vma_lock(vma);
 	err = i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
+	i915_vma_unlock(vma);
 	if (err)
 		goto skip_request;
 
@@ -1345,7 +1359,9 @@ static int write_to_scratch(struct i915_gem_context *ctx,
 	if (err)
 		goto err_request;
 
+	i915_vma_lock(vma);
 	err = i915_vma_move_to_active(vma, rq, 0);
+	i915_vma_unlock(vma);
 	if (err)
 		goto skip_request;
 
@@ -1440,7 +1456,9 @@ static int read_from_scratch(struct i915_gem_context *ctx,
 	if (err)
 		goto err_request;
 
+	i915_vma_lock(vma);
 	err = i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
+	i915_vma_unlock(vma);
 	if (err)
 		goto skip_request;
 
@@ -1449,7 +1467,9 @@ static int read_from_scratch(struct i915_gem_context *ctx,
 
 	i915_request_add(rq);
 
+	i915_gem_object_lock(obj);
 	err = i915_gem_object_set_to_cpu_domain(obj, false);
+	i915_gem_object_unlock(obj);
 	if (err)
 		goto err;
 

drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c

@@ -110,7 +110,9 @@ static int check_partial_mapping(struct drm_i915_gem_object *obj,
 		GEM_BUG_ON(view.partial.size > nreal);
 		cond_resched();
 
+		i915_gem_object_lock(obj);
 		err = i915_gem_object_set_to_gtt_domain(obj, true);
+		i915_gem_object_unlock(obj);
 		if (err) {
 			pr_err("Failed to flush to GTT write domain; err=%d\n",
 			       err);
@@ -142,7 +144,9 @@ static int check_partial_mapping(struct drm_i915_gem_object *obj,
 		if (offset >= obj->base.size)
 			continue;
 
+		i915_gem_object_lock(obj);
 		i915_gem_object_flush_write_domain(obj, ~I915_GEM_DOMAIN_CPU);
+		i915_gem_object_unlock(obj);
 
 		p = i915_gem_object_get_page(obj, offset >> PAGE_SHIFT);
 		cpu = kmap(p) + offset_in_page(offset);
@@ -344,7 +348,9 @@ static int make_obj_busy(struct drm_i915_gem_object *obj)
 		return PTR_ERR(rq);
 	}
 
+	i915_vma_lock(vma);
 	err = i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
+	i915_vma_unlock(vma);
 
 	i915_request_add(rq);
 

drivers/gpu/drm/i915/gem/selftests/i915_gem_phys.c

@@ -46,9 +46,9 @@ static int mock_phys_object(void *arg)
 	}
 
 	/* Make the object dirty so that put_pages must do copy back the data */
-	mutex_lock(&i915->drm.struct_mutex);
+	i915_gem_object_lock(obj);
 	err = i915_gem_object_set_to_gtt_domain(obj, true);
-	mutex_unlock(&i915->drm.struct_mutex);
+	i915_gem_object_unlock(obj);
 	if (err) {
 		pr_err("i915_gem_object_set_to_gtt_domain failed with err=%d\n",
 		       err);