drm/vmwgfx: Add DX query support. Various fixes.

Add support for vgpu10 queries. Functional- and formatting fixes.
Signed-off-by: Sinclair Yeh <syeh@vmware.com>
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>

drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c

@@ -818,7 +818,7 @@ static int vmw_ttm_fault_reserve_notify(struct ttm_buffer_object *bo)
  * vmw_move_notify - TTM move_notify_callback
  *
  * @bo: The TTM buffer object about to move.
- * @mem:            The truct ttm_mem_reg indicating to what memory
+ * @mem: The struct ttm_mem_reg indicating to what memory
  *       region the move is taking place.
  *
  * Calls move_notify for all subsystems needing it.
@@ -828,6 +828,7 @@ static void vmw_move_notify(struct ttm_buffer_object *bo,
 			    struct ttm_mem_reg *mem)
 {
 	vmw_resource_move_notify(bo, mem);
+	vmw_query_move_notify(bo, mem);
 }
 
 

drivers/gpu/drm/vmwgfx/vmwgfx_context.c

@@ -121,6 +121,8 @@ static void vmw_context_cotables_unref(struct vmw_user_context *uctx)
 		res = uctx->cotables[i];
 		uctx->cotables[i] = NULL;
 		spin_unlock(&uctx->cotable_lock);
+
+		if (res)
 			vmw_resource_unreference(&res);
 	}
 }
@@ -585,6 +587,8 @@ static int vmw_dx_context_unbind(struct vmw_resource *res,
 	struct vmw_private *dev_priv = res->dev_priv;
 	struct ttm_buffer_object *bo = val_buf->bo;
 	struct vmw_fence_obj *fence;
+	struct vmw_user_context *uctx =
+		container_of(res, struct vmw_user_context, res);
 
 	struct {
 		SVGA3dCmdHeader header;
@@ -603,6 +607,13 @@ static int vmw_dx_context_unbind(struct vmw_resource *res,
 	mutex_lock(&dev_priv->binding_mutex);
 	vmw_dx_context_scrub_cotables(res, readback);
 
+	if (uctx->dx_query_mob && uctx->dx_query_mob->dx_query_ctx &&
+	    readback) {
+		WARN_ON(uctx->dx_query_mob->dx_query_ctx != res);
+		if (vmw_query_readback_all(uctx->dx_query_mob))
+			DRM_ERROR("Failed to read back query states\n");
+	}
+
 	submit_size = sizeof(*cmd2) + (readback ? sizeof(*cmd1) : 0);
 
 	cmd = vmw_fifo_reserve(dev_priv, submit_size);
@@ -692,6 +703,9 @@ static void vmw_user_context_free(struct vmw_resource *res)
 
 	if (ctx->cbs)
 		vmw_binding_state_free(ctx->cbs);
+
+	(void) vmw_context_bind_dx_query(res, NULL);
+
 	ttm_base_object_kfree(ctx, base);
 	ttm_mem_global_free(vmw_mem_glob(dev_priv),
 			    vmw_user_context_size);
@@ -867,3 +881,57 @@ vmw_context_binding_state(struct vmw_resource *ctx)
 {
 	return container_of(ctx, struct vmw_user_context, res)->cbs;
 }
+
+/**
+ * vmw_context_bind_dx_query -
+ * Sets query MOB for the context.  If @mob is NULL, then this function will
+ * remove the association between the MOB and the context.  This function
+ * assumes the binding_mutex is held.
+ *
+ * @ctx_res: The context resource
+ * @mob: a reference to the query MOB
+ *
+ * Returns -EINVAL if a MOB has already been set and does not match the one
+ * specified in the parameter.  0 otherwise.
+ */
+int vmw_context_bind_dx_query(struct vmw_resource *ctx_res,
+			      struct vmw_dma_buffer *mob)
+{
+	struct vmw_user_context *uctx =
+		container_of(ctx_res, struct vmw_user_context, res);
+
+	if (mob == NULL) {
+		if (uctx->dx_query_mob) {
+			uctx->dx_query_mob->dx_query_ctx = NULL;
+			vmw_dmabuf_unreference(&uctx->dx_query_mob);
+			uctx->dx_query_mob = NULL;
+		}
+
+		return 0;
+	}
+
+	/* Can only have one MOB per context for queries */
+	if (uctx->dx_query_mob && uctx->dx_query_mob != mob)
+		return -EINVAL;
+
+	mob->dx_query_ctx  = ctx_res;
+
+	if (!uctx->dx_query_mob)
+		uctx->dx_query_mob = vmw_dmabuf_reference(mob);
+
+	return 0;
+}
+
+/**
+ * vmw_context_get_dx_query_mob - Returns non-counted reference to DX query mob
+ *
+ * @ctx_res: The context resource
+ */
+struct vmw_dma_buffer *
+vmw_context_get_dx_query_mob(struct vmw_resource *ctx_res)
+{
+	struct vmw_user_context *uctx =
+		container_of(ctx_res, struct vmw_user_context, res);
+
+	return uctx->dx_query_mob;
+}

drivers/gpu/drm/vmwgfx/vmwgfx_drv.h

@@ -88,6 +88,8 @@ struct vmw_dma_buffer {
 	struct ttm_buffer_object base;
 	struct list_head res_list;
 	s32 pin_count;
+	/* Not ref-counted.  Protected by binding_mutex */
+	struct vmw_resource *dx_query_ctx;
 };
 
 /**
@@ -658,6 +660,9 @@ extern void vmw_resource_unreserve(struct vmw_resource *res,
 				   unsigned long new_backup_offset);
 extern void vmw_resource_move_notify(struct ttm_buffer_object *bo,
 				     struct ttm_mem_reg *mem);
+extern void vmw_query_move_notify(struct ttm_buffer_object *bo,
+				  struct ttm_mem_reg *mem);
+extern int vmw_query_readback_all(struct vmw_dma_buffer *dx_query_mob);
 extern void vmw_fence_single_bo(struct ttm_buffer_object *bo,
 				struct vmw_fence_obj *fence);
 extern void vmw_resource_evict_all(struct vmw_private *dev_priv);
@@ -1011,6 +1016,11 @@ extern struct vmw_ctx_binding_state *
 vmw_context_binding_state(struct vmw_resource *ctx);
 extern void vmw_dx_context_scrub_cotables(struct vmw_resource *ctx,
 					  bool readback);
+extern int vmw_context_bind_dx_query(struct vmw_resource *ctx_res,
+				     struct vmw_dma_buffer *mob);
+extern struct vmw_dma_buffer *
+vmw_context_get_dx_query_mob(struct vmw_resource *ctx_res);
+
 
 /*
  * Surface management - vmwgfx_surface.c

drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c

 /**************************************************************************
  *
- * Copyright © 2009 VMware, Inc., Palo Alto, CA., USA
+ * Copyright © 2009 - 2015 VMware, Inc., Palo Alto, CA., USA
  * All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -101,19 +101,32 @@ struct vmw_cmd_entry {
 static int vmw_resource_context_res_add(struct vmw_private *dev_priv,
 					struct vmw_sw_context *sw_context,
 					struct vmw_resource *ctx);
+static int vmw_translate_mob_ptr(struct vmw_private *dev_priv,
+				 struct vmw_sw_context *sw_context,
+				 SVGAMobId *id,
+				 struct vmw_dma_buffer **vmw_bo_p);
+static int vmw_bo_to_validate_list(struct vmw_sw_context *sw_context,
+				   struct vmw_dma_buffer *vbo,
+				   bool validate_as_mob,
+				   uint32_t *p_val_node);
+
 
 /**
- * vmw_resource_unreserve - unreserve resources previously reserved for
+ * vmw_resources_unreserve - unreserve resources previously reserved for
  * command submission.
  *
- * @list_head: list of resources to unreserve.
+ * @sw_context: pointer to the software context
  * @backoff: Whether command submission failed.
  */
-static void vmw_resource_list_unreserve(struct vmw_sw_context *sw_context,
-					struct list_head *list,
+static void vmw_resources_unreserve(struct vmw_sw_context *sw_context,
 				    bool backoff)
 {
 	struct vmw_resource_val_node *val;
+	struct list_head *list = &sw_context->resource_list;
+
+	if (sw_context->dx_query_mob && !backoff)
+		vmw_context_bind_dx_query(sw_context->dx_query_ctx,
+					  sw_context->dx_query_mob);
 
 	list_for_each_entry(val, list, head) {
 		struct vmw_resource *res = val->res;
@@ -376,6 +389,16 @@ static int vmw_resource_context_res_add(struct vmw_private *dev_priv,
 			break;
 	}
 
+	if (dev_priv->has_dx && vmw_res_type(ctx) == vmw_res_dx_context) {
+		struct vmw_dma_buffer *dx_query_mob;
+
+		dx_query_mob = vmw_context_get_dx_query_mob(ctx);
+		if (dx_query_mob)
+			ret = vmw_bo_to_validate_list(sw_context,
+						      dx_query_mob,
+						      true, NULL);
+	}
+
 	mutex_unlock(&dev_priv->binding_mutex);
 	return ret;
 }
@@ -533,7 +556,7 @@ static int vmw_bo_to_validate_list(struct vmw_sw_context *sw_context,
 static int vmw_resources_reserve(struct vmw_sw_context *sw_context)
 {
 	struct vmw_resource_val_node *val;
-	int ret;
+	int ret = 0;
 
 	list_for_each_entry(val, &sw_context->resource_list, head) {
 		struct vmw_resource *res = val->res;
@@ -554,7 +577,18 @@ static int vmw_resources_reserve(struct vmw_sw_context *sw_context)
 		}
 	}
 
-	return 0;
+	if (sw_context->dx_query_mob) {
+		struct vmw_dma_buffer *expected_dx_query_mob;
+
+		expected_dx_query_mob =
+			vmw_context_get_dx_query_mob(sw_context->dx_query_ctx);
+		if (expected_dx_query_mob &&
+		    expected_dx_query_mob != sw_context->dx_query_mob) {
+			ret = -EINVAL;
+		}
+	}
+
+	return ret;
 }
 
 /**
@@ -724,6 +758,46 @@ vmw_cmd_res_check(struct vmw_private *dev_priv,
 	return ret;
 }
 
+/**
+ * vmw_rebind_dx_query - Rebind DX query associated with the context
+ *
+ * @ctx_res: context the query belongs to
+ *
+ * This function assumes binding_mutex is held.
+ */
+static int vmw_rebind_all_dx_query(struct vmw_resource *ctx_res)
+{
+	struct vmw_private *dev_priv = ctx_res->dev_priv;
+	struct vmw_dma_buffer *dx_query_mob;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdDXBindAllQuery body;
+	} *cmd;
+
+
+	dx_query_mob = vmw_context_get_dx_query_mob(ctx_res);
+
+	if (!dx_query_mob || dx_query_mob->dx_query_ctx)
+		return 0;
+
+	cmd = vmw_fifo_reserve_dx(dev_priv, sizeof(*cmd), ctx_res->id);
+
+	if (cmd == NULL) {
+		DRM_ERROR("Failed to rebind queries.\n");
+		return -ENOMEM;
+	}
+
+	cmd->header.id = SVGA_3D_CMD_DX_BIND_ALL_QUERY;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.cid = ctx_res->id;
+	cmd->body.mobid = dx_query_mob->base.mem.start;
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+
+	vmw_context_bind_dx_query(ctx_res, dx_query_mob);
+
+	return 0;
+}
+
 /**
  * vmw_rebind_contexts - Rebind all resources previously bound to
  * referenced contexts.
@@ -748,6 +822,10 @@ static int vmw_rebind_contexts(struct vmw_sw_context *sw_context)
 				DRM_ERROR("Failed to rebind context.\n");
 			return ret;
 		}
+
+		ret = vmw_rebind_all_dx_query(val->res);
+		if (ret != 0)
+			return ret;
 	}
 
 	return 0;
@@ -1248,6 +1326,98 @@ static int vmw_translate_guest_ptr(struct vmw_private *dev_priv,
 	return ret;
 }
 
+
+
+/**
+ * vmw_cmd_dx_define_query - validate a SVGA_3D_CMD_DX_DEFINE_QUERY command.
+ *
+ * @dev_priv: Pointer to a device private struct.
+ * @sw_context: The software context used for this command submission.
+ * @header: Pointer to the command header in the command stream.
+ *
+ * This function adds the new query into the query COTABLE
+ */
+static int vmw_cmd_dx_define_query(struct vmw_private *dev_priv,
+				   struct vmw_sw_context *sw_context,
+				   SVGA3dCmdHeader *header)
+{
+	struct vmw_dx_define_query_cmd {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdDXDefineQuery q;
+	} *cmd;
+
+	int    ret;
+	struct vmw_resource_val_node *ctx_node = sw_context->dx_ctx_node;
+	struct vmw_resource *cotable_res;
+
+
+	if (ctx_node == NULL) {
+		DRM_ERROR("DX Context not set for query.\n");
+		return -EINVAL;
+	}
+
+	cmd = container_of(header, struct vmw_dx_define_query_cmd, header);
+
+	if (cmd->q.type <  SVGA3D_QUERYTYPE_MIN ||
+	    cmd->q.type >= SVGA3D_QUERYTYPE_MAX)
+		return -EINVAL;
+
+	cotable_res = vmw_context_cotable(ctx_node->res, SVGA_COTABLE_DXQUERY);
+	ret = vmw_cotable_notify(cotable_res, cmd->q.queryId);
+	vmw_resource_unreference(&cotable_res);
+
+	return ret;
+}
+
+
+
+/**
+ * vmw_cmd_dx_bind_query - validate a SVGA_3D_CMD_DX_BIND_QUERY command.
+ *
+ * @dev_priv: Pointer to a device private struct.
+ * @sw_context: The software context used for this command submission.
+ * @header: Pointer to the command header in the command stream.
+ *
+ * The query bind operation will eventually associate the query ID
+ * with its backing MOB.  In this function, we take the user mode
+ * MOB ID and use vmw_translate_mob_ptr() to translate it to its
+ * kernel mode equivalent.
+ */
+static int vmw_cmd_dx_bind_query(struct vmw_private *dev_priv,
+				 struct vmw_sw_context *sw_context,
+				 SVGA3dCmdHeader *header)
+{
+	struct vmw_dx_bind_query_cmd {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdDXBindQuery q;
+	} *cmd;
+
+	struct vmw_dma_buffer *vmw_bo;
+	int    ret;
+
+
+	cmd = container_of(header, struct vmw_dx_bind_query_cmd, header);
+
+	/*
+	 * Look up the buffer pointed to by q.mobid, put it on the relocation
+	 * list so its kernel mode MOB ID can be filled in later
+	 */
+	ret = vmw_translate_mob_ptr(dev_priv, sw_context, &cmd->q.mobid,
+				    &vmw_bo);
+
+	if (ret != 0)
+		return ret;
+
+	sw_context->dx_query_mob = vmw_bo;
+	sw_context->dx_query_ctx = sw_context->dx_ctx_node->res;
+
+	vmw_dmabuf_unreference(&vmw_bo);
+
+	return ret;
+}
+
+
+
 /**
  * vmw_cmd_begin_gb_query - validate a  SVGA_3D_CMD_BEGIN_GB_QUERY command.
  *
@@ -2975,6 +3145,8 @@ static const struct vmw_cmd_entry vmw_cmd_entries[SVGA_3D_CMD_MAX] = {
 		    false, false, true),
 	VMW_CMD_DEF(SVGA_3D_CMD_DESTROY_GB_MOB, &vmw_cmd_invalid,
 		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_REDEFINE_GB_MOB64, &vmw_cmd_invalid,
+		    false, false, true),
 	VMW_CMD_DEF(SVGA_3D_CMD_UPDATE_GB_MOB_MAPPING, &vmw_cmd_invalid,
 		    false, false, true),
 	VMW_CMD_DEF(SVGA_3D_CMD_DEFINE_GB_SURFACE, &vmw_cmd_invalid,
@@ -3097,15 +3269,17 @@ static const struct vmw_cmd_entry vmw_cmd_entries[SVGA_3D_CMD_MAX] = {
 		    &vmw_cmd_dx_cid_check, true, false, true),
 	VMW_CMD_DEF(SVGA_3D_CMD_DX_SET_RASTERIZER_STATE,
 		    &vmw_cmd_dx_cid_check, true, false, true),
-	VMW_CMD_DEF(SVGA_3D_CMD_DX_DEFINE_QUERY, &vmw_cmd_invalid,
+	VMW_CMD_DEF(SVGA_3D_CMD_DX_DEFINE_QUERY, &vmw_cmd_dx_define_query,
 		    true, false, true),
-	VMW_CMD_DEF(SVGA_3D_CMD_DX_DESTROY_QUERY, &vmw_cmd_invalid,
+	VMW_CMD_DEF(SVGA_3D_CMD_DX_DESTROY_QUERY, &vmw_cmd_ok,
 		    true, false, true),
-	VMW_CMD_DEF(SVGA_3D_CMD_DX_BIND_QUERY, &vmw_cmd_invalid,
+	VMW_CMD_DEF(SVGA_3D_CMD_DX_BIND_QUERY, &vmw_cmd_dx_bind_query,
 		    true, false, true),
-	VMW_CMD_DEF(SVGA_3D_CMD_DX_BEGIN_QUERY, &vmw_cmd_invalid,
+	VMW_CMD_DEF(SVGA_3D_CMD_DX_SET_QUERY_OFFSET,
+		    &vmw_cmd_ok, true, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DX_BEGIN_QUERY, &vmw_cmd_ok,
 		    true, false, true),
-	VMW_CMD_DEF(SVGA_3D_CMD_DX_END_QUERY, &vmw_cmd_invalid,
+	VMW_CMD_DEF(SVGA_3D_CMD_DX_END_QUERY, &vmw_cmd_ok,
 		    true, false, true),
 	VMW_CMD_DEF(SVGA_3D_CMD_DX_READBACK_QUERY, &vmw_cmd_invalid,
 		    true, false, true),
@@ -3780,6 +3954,8 @@ int vmw_execbuf_process(struct drm_file *file_priv,
 	sw_context->last_query_ctx = NULL;
 	sw_context->needs_post_query_barrier = false;
 	sw_context->dx_ctx_node = NULL;
+	sw_context->dx_query_mob = NULL;
+	sw_context->dx_query_ctx = NULL;
 	memset(sw_context->res_cache, 0, sizeof(sw_context->res_cache));
 	INIT_LIST_HEAD(&sw_context->validate_nodes);
 	INIT_LIST_HEAD(&sw_context->res_relocations);
@@ -3803,7 +3979,6 @@ int vmw_execbuf_process(struct drm_file *file_priv,
 
 	ret = vmw_cmd_check_all(dev_priv, sw_context, kernel_commands,
 				command_size);
-
 	/*
 	 * Merge the resource lists before checking the return status
 	 * from vmd_cmd_check_all so that all the open hashtabs will
@@ -3869,8 +4044,7 @@ int vmw_execbuf_process(struct drm_file *file_priv,
 	if (ret != 0)
 		DRM_ERROR("Fence submission error. Syncing.\n");
 
-	vmw_resource_list_unreserve(sw_context, &sw_context->resource_list,
-				    false);
+	vmw_resources_unreserve(sw_context, false);
 
 	ttm_eu_fence_buffer_objects(&ticket, &sw_context->validate_nodes,
 				    (void *) fence);
@@ -3908,8 +4082,7 @@ int vmw_execbuf_process(struct drm_file *file_priv,
 out_err:
 	ttm_eu_backoff_reservation(&ticket, &sw_context->validate_nodes);
 out_err_nores:
-	vmw_resource_list_unreserve(sw_context, &sw_context->resource_list,
-				    true);
+	vmw_resources_unreserve(sw_context, true);
 	vmw_resource_relocations_free(&sw_context->res_relocations);
 	vmw_free_relocations(sw_context);
 	vmw_clear_validations(sw_context);

drivers/gpu/drm/vmwgfx/vmwgfx_resource.c

@@ -1452,7 +1452,7 @@ void vmw_fence_single_bo(struct ttm_buffer_object *bo,
  * vmw_resource_move_notify - TTM move_notify_callback
  *
  * @bo: The TTM buffer object about to move.
- * @mem:            The truct ttm_mem_reg indicating to what memory
+ * @mem: The struct ttm_mem_reg indicating to what memory
  *       region the move is taking place.
  *
  * Evicts the Guest Backed hardware resource if the backup
@@ -1503,6 +1503,101 @@ void vmw_resource_move_notify(struct ttm_buffer_object *bo,
 	}
 }
 
+
+
+/**
+ * vmw_query_readback_all - Read back cached query states
+ *
+ * @dx_query_mob: Buffer containing the DX query MOB
+ *
+ * Read back cached states from the device if they exist.  This function
+ * assumings binding_mutex is held.
+ */
+int vmw_query_readback_all(struct vmw_dma_buffer *dx_query_mob)
+{
+	struct vmw_resource *dx_query_ctx;
+	struct vmw_private *dev_priv;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdDXReadbackAllQuery body;
+	} *cmd;
+
+
+	/* No query bound, so do nothing */
+	if (!dx_query_mob || !dx_query_mob->dx_query_ctx)
+		return 0;
+
+	dx_query_ctx = dx_query_mob->dx_query_ctx;
+	dev_priv     = dx_query_ctx->dev_priv;
+
+	cmd = vmw_fifo_reserve_dx(dev_priv, sizeof(*cmd), dx_query_ctx->id);
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Failed reserving FIFO space for "
+			  "query MOB read back.\n");
+		return -ENOMEM;
+	}
+
+	cmd->header.id   = SVGA_3D_CMD_DX_READBACK_ALL_QUERY;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.cid    = dx_query_ctx->id;
+
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+
+	/* Triggers a rebind the next time affected context is bound */
+	dx_query_mob->dx_query_ctx = NULL;
+
+	return 0;
+}
+
+
+
+/**
+ * vmw_query_move_notify - Read back cached query states
+ *
+ * @bo: The TTM buffer object about to move.
+ * @mem: The memory region @bo is moving to.
+ *
+ * Called before the query MOB is swapped out to read back cached query
+ * states from the device.
+ */
+void vmw_query_move_notify(struct ttm_buffer_object *bo,
+			   struct ttm_mem_reg *mem)
+{
+	struct vmw_dma_buffer *dx_query_mob;
+	struct ttm_bo_device *bdev = bo->bdev;
+	struct vmw_private *dev_priv;
+
+
+	dev_priv = container_of(bdev, struct vmw_private, bdev);
+
+	mutex_lock(&dev_priv->binding_mutex);
+
+	dx_query_mob = container_of(bo, struct vmw_dma_buffer, base);
+	if (mem == NULL || !dx_query_mob || !dx_query_mob->dx_query_ctx) {
+		mutex_unlock(&dev_priv->binding_mutex);
+		return;
+	}
+
+	/* If BO is being moved from MOB to system memory */
+	if (mem->mem_type == TTM_PL_SYSTEM && bo->mem.mem_type == VMW_PL_MOB) {
+		struct vmw_fence_obj *fence;
+
+		(void) vmw_query_readback_all(dx_query_mob);
+		mutex_unlock(&dev_priv->binding_mutex);
+
+		/* Create a fence and attach the BO to it */
+		(void) vmw_execbuf_fence_commands(NULL, dev_priv, &fence, NULL);
+		vmw_fence_single_bo(bo, fence);
+
+		if (fence != NULL)
+			vmw_fence_obj_unreference(&fence);
+
+		(void) ttm_bo_wait(bo, false, false, false);
+	} else
+		mutex_unlock(&dev_priv->binding_mutex);
+
+}
+
 /**
  * vmw_resource_needs_backup - Return whether a resource needs a backup buffer.
  *