1 files changed, 61 insertions, 101 deletions

diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
index f627a8c47c58..3f0c612d42e7 100644
--- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
@@ -66,6 +66,15 @@ enum {
 #define __I915_EXEC_ILLEGAL_FLAGS \
 	(__I915_EXEC_UNKNOWN_FLAGS | I915_EXEC_CONSTANTS_MASK)
 
+/* Catch emission of unexpected errors for CI! */
+#if IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM)
+#undef EINVAL
+#define EINVAL ({ \
+	DRM_DEBUG_DRIVER("EINVAL at %s:%d\n", __func__, __LINE__); \
+	22; \
+})
+#endif
+
 /**
  * DOC: User command execution
  *
@@ -489,7 +498,9 @@ eb_validate_vma(struct i915_execbuffer *eb,
 }
 
 static int
-eb_add_vma(struct i915_execbuffer *eb, unsigned int i, struct i915_vma *vma)
+eb_add_vma(struct i915_execbuffer *eb,
+	   unsigned int i, unsigned batch_idx,
+	   struct i915_vma *vma)
 {
 	struct drm_i915_gem_exec_object2 *entry = &eb->exec[i];
 	int err;
@@ -522,6 +533,25 @@ eb_add_vma(struct i915_execbuffer *eb, unsigned int i, struct i915_vma *vma)
 	eb->flags[i] = entry->flags;
 	vma->exec_flags = &eb->flags[i];
 
+	/*
+	 * SNA is doing fancy tricks with compressing batch buffers, which leads
+	 * to negative relocation deltas. Usually that works out ok since the
+	 * relocate address is still positive, except when the batch is placed
+	 * very low in the GTT. Ensure this doesn't happen.
+	 *
+	 * Note that actual hangs have only been observed on gen7, but for
+	 * paranoia do it everywhere.
+	 */
+	if (i == batch_idx) {
+		if (entry->relocation_count &&
+		    !(eb->flags[i] & EXEC_OBJECT_PINNED))
+			eb->flags[i] |= __EXEC_OBJECT_NEEDS_BIAS;
+		if (eb->reloc_cache.has_fence)
+			eb->flags[i] |= EXEC_OBJECT_NEEDS_FENCE;
+
+		eb->batch = vma;
+	}
+
 	err = 0;
 	if (eb_pin_vma(eb, entry, vma)) {
 		if (entry->offset != vma->node.start) {
@@ -703,7 +733,7 @@ static int eb_select_context(struct i915_execbuffer *eb)
 		return -ENOENT;
 
 	eb->ctx = ctx;
-	eb->vm = ctx->ppgtt ? &ctx->ppgtt->base : &eb->i915->ggtt.base;
+	eb->vm = ctx->ppgtt ? &ctx->ppgtt->vm : &eb->i915->ggtt.vm;
 
 	eb->context_flags = 0;
 	if (ctx->flags & CONTEXT_NO_ZEROMAP)
@@ -716,7 +746,7 @@ static int eb_lookup_vmas(struct i915_execbuffer *eb)
 {
 	struct radix_tree_root *handles_vma = &eb->ctx->handles_vma;
 	struct drm_i915_gem_object *obj;
-	unsigned int i;
+	unsigned int i, batch;
 	int err;
 
 	if (unlikely(i915_gem_context_is_closed(eb->ctx)))
@@ -728,6 +758,8 @@ static int eb_lookup_vmas(struct i915_execbuffer *eb)
 	INIT_LIST_HEAD(&eb->relocs);
 	INIT_LIST_HEAD(&eb->unbound);
 
+	batch = eb_batch_index(eb);
+
 	for (i = 0; i < eb->buffer_count; i++) {
 		u32 handle = eb->exec[i].handle;
 		struct i915_lut_handle *lut;
@@ -770,33 +802,16 @@ static int eb_lookup_vmas(struct i915_execbuffer *eb)
 		lut->handle = handle;
 
 add_vma:
-		err = eb_add_vma(eb, i, vma);
+		err = eb_add_vma(eb, i, batch, vma);
 		if (unlikely(err))
 			goto err_vma;
 
 		GEM_BUG_ON(vma != eb->vma[i]);
 		GEM_BUG_ON(vma->exec_flags != &eb->flags[i]);
+		GEM_BUG_ON(drm_mm_node_allocated(&vma->node) &&
+			   eb_vma_misplaced(&eb->exec[i], vma, eb->flags[i]));
 	}
 
-	/* take note of the batch buffer before we might reorder the lists */
-	i = eb_batch_index(eb);
-	eb->batch = eb->vma[i];
-	GEM_BUG_ON(eb->batch->exec_flags != &eb->flags[i]);
-
-	/*
-	 * SNA is doing fancy tricks with compressing batch buffers, which leads
-	 * to negative relocation deltas. Usually that works out ok since the
-	 * relocate address is still positive, except when the batch is placed
-	 * very low in the GTT. Ensure this doesn't happen.
-	 *
-	 * Note that actual hangs have only been observed on gen7, but for
-	 * paranoia do it everywhere.
-	 */
-	if (!(eb->flags[i] & EXEC_OBJECT_PINNED))
-		eb->flags[i] |= __EXEC_OBJECT_NEEDS_BIAS;
-	if (eb->reloc_cache.has_fence)
-		eb->flags[i] |= EXEC_OBJECT_NEEDS_FENCE;
-
 	eb->args->flags |= __EXEC_VALIDATED;
 	return eb_reserve(eb);
 
@@ -916,7 +931,7 @@ static void reloc_gpu_flush(struct reloc_cache *cache)
 	i915_gem_object_unpin_map(cache->rq->batch->obj);
 	i915_gem_chipset_flush(cache->rq->i915);
 
-	__i915_request_add(cache->rq, true);
+	i915_request_add(cache->rq);
 	cache->rq = NULL;
 }
 
@@ -943,9 +958,9 @@ static void reloc_cache_reset(struct reloc_cache *cache)
 		if (cache->node.allocated) {
 			struct i915_ggtt *ggtt = cache_to_ggtt(cache);
 
-			ggtt->base.clear_range(&ggtt->base,
-					       cache->node.start,
-					       cache->node.size);
+			ggtt->vm.clear_range(&ggtt->vm,
+					     cache->node.start,
+					     cache->node.size);
 			drm_mm_remove_node(&cache->node);
 		} else {
 			i915_vma_unpin((struct i915_vma *)cache->node.mm);
@@ -1016,7 +1031,7 @@ static void *reloc_iomap(struct drm_i915_gem_object *obj,
 		if (IS_ERR(vma)) {
 			memset(&cache->node, 0, sizeof(cache->node));
 			err = drm_mm_insert_node_in_range
-				(&ggtt->base.mm, &cache->node,
+				(&ggtt->vm.mm, &cache->node,
 				 PAGE_SIZE, 0, I915_COLOR_UNEVICTABLE,
 				 0, ggtt->mappable_end,
 				 DRM_MM_INSERT_LOW);
@@ -1037,9 +1052,9 @@ static void *reloc_iomap(struct drm_i915_gem_object *obj,
 	offset = cache->node.start;
 	if (cache->node.allocated) {
 		wmb();
-		ggtt->base.insert_page(&ggtt->base,
-				       i915_gem_object_get_dma_address(obj, page),
-				       offset, I915_CACHE_NONE, 0);
+		ggtt->vm.insert_page(&ggtt->vm,
+				     i915_gem_object_get_dma_address(obj, page),
+				     offset, I915_CACHE_NONE, 0);
 	} else {
 		offset += page << PAGE_SHIFT;
 	}
@@ -1150,18 +1165,16 @@ static int __reloc_gpu_alloc(struct i915_execbuffer *eb,
 		goto err_request;
 
 	GEM_BUG_ON(!reservation_object_test_signaled_rcu(batch->resv, true));
-	i915_vma_move_to_active(batch, rq, 0);
-	reservation_object_lock(batch->resv, NULL);
-	reservation_object_add_excl_fence(batch->resv, &rq->fence);
-	reservation_object_unlock(batch->resv);
-	i915_vma_unpin(batch);
+	err = i915_vma_move_to_active(batch, rq, 0);
+	if (err)
+		goto skip_request;
 
-	i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
-	reservation_object_lock(vma->resv, NULL);
-	reservation_object_add_excl_fence(vma->resv, &rq->fence);
-	reservation_object_unlock(vma->resv);
+	err = i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
+	if (err)
+		goto skip_request;
 
 	rq->batch = batch;
+	i915_vma_unpin(batch);
 
 	cache->rq = rq;
 	cache->rq_cmd = cmd;
@@ -1170,6 +1183,8 @@ static int __reloc_gpu_alloc(struct i915_execbuffer *eb,
 	/* Return with batch mapping (cmd) still pinned */
 	return 0;
 
+skip_request:
+	i915_request_skip(rq, err);
 err_request:
 	i915_request_add(rq);
 err_unpin:
@@ -1756,25 +1771,6 @@ slow:
 	return eb_relocate_slow(eb);
 }
 
-static void eb_export_fence(struct i915_vma *vma,
-			    struct i915_request *rq,
-			    unsigned int flags)
-{
-	struct reservation_object *resv = vma->resv;
-
-	/*
-	 * Ignore errors from failing to allocate the new fence, we can't
-	 * handle an error right now. Worst case should be missed
-	 * synchronisation leading to rendering corruption.
-	 */
-	reservation_object_lock(resv, NULL);
-	if (flags & EXEC_OBJECT_WRITE)
-		reservation_object_add_excl_fence(resv, &rq->fence);
-	else if (reservation_object_reserve_shared(resv) == 0)
-		reservation_object_add_shared_fence(resv, &rq->fence);
-	reservation_object_unlock(resv);
-}
-
 static int eb_move_to_gpu(struct i915_execbuffer *eb)
 {
 	const unsigned int count = eb->buffer_count;
@@ -1828,8 +1824,11 @@ static int eb_move_to_gpu(struct i915_execbuffer *eb)
 		unsigned int flags = eb->flags[i];
 		struct i915_vma *vma = eb->vma[i];
 
-		i915_vma_move_to_active(vma, eb->request, flags);
-		eb_export_fence(vma, eb->request, flags);
+		err = i915_vma_move_to_active(vma, eb->request, flags);
+		if (unlikely(err)) {
+			i915_request_skip(eb->request, err);
+			return err;
+		}
 
 		__eb_unreserve_vma(vma, flags);
 		vma->exec_flags = NULL;
@@ -1869,45 +1868,6 @@ static bool i915_gem_check_execbuffer(struct drm_i915_gem_execbuffer2 *exec)
 	return true;
 }
 
-void i915_vma_move_to_active(struct i915_vma *vma,
-			     struct i915_request *rq,
-			     unsigned int flags)
-{
-	struct drm_i915_gem_object *obj = vma->obj;
-	const unsigned int idx = rq->engine->id;
-
-	lockdep_assert_held(&rq->i915->drm.struct_mutex);
-	GEM_BUG_ON(!drm_mm_node_allocated(&vma->node));
-
-	/*
-	 * Add a reference if we're newly entering the active list.
-	 * The order in which we add operations to the retirement queue is
-	 * vital here: mark_active adds to the start of the callback list,
-	 * such that subsequent callbacks are called first. Therefore we
-	 * add the active reference first and queue for it to be dropped
-	 * *last*.
-	 */
-	if (!i915_vma_is_active(vma))
-		obj->active_count++;
-	i915_vma_set_active(vma, idx);
-	i915_gem_active_set(&vma->last_read[idx], rq);
-	list_move_tail(&vma->vm_link, &vma->vm->active_list);
-
-	obj->write_domain = 0;
-	if (flags & EXEC_OBJECT_WRITE) {
-		obj->write_domain = I915_GEM_DOMAIN_RENDER;
-
-		if (intel_fb_obj_invalidate(obj, ORIGIN_CS))
-			i915_gem_active_set(&obj->frontbuffer_write, rq);
-
-		obj->read_domains = 0;
-	}
-	obj->read_domains |= I915_GEM_GPU_DOMAINS;
-
-	if (flags & EXEC_OBJECT_NEEDS_FENCE)
-		i915_gem_active_set(&vma->last_fence, rq);
-}
-
 static int i915_reset_gen7_sol_offsets(struct i915_request *rq)
 {
 	u32 *cs;
@@ -2433,7 +2393,7 @@ i915_gem_do_execbuffer(struct drm_device *dev,
 	trace_i915_request_queue(eb.request, eb.batch_flags);
 	err = eb_submit(&eb);
 err_request:
-	__i915_request_add(eb.request, err == 0);
+	i915_request_add(eb.request);
 	add_to_client(eb.request, file);
 
 	if (fences)