drm/vmwgfx: Add support for streamoutput with mob commands

With SM5 capability a new version of streamoutput is supported by device
which need backing mob and a new field. With this change the new command
is supported in command buffer.

v2: Also track streamoutput context binding in binding manager.

v3: Track only one streamoutput as only one can be set to context.
v4: Fix comment typos

Signed-off-by: Deepak Rawat <drawat.floss@gmail.com>
Signed-off-by: Neha Bhende <bhenden@vmware.com>
Reviewed-by: Thomas Hellström (VMware) <thomas_os@shipmail.org>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Roland Scheidegger <sroland@vmware.com>

8 files changed, 630 insertions, 7 deletions

diff --git a/drivers/gpu/drm/vmwgfx/Makefile b/drivers/gpu/drm/vmwgfx/Makefile
index c877a21a0739..5c3515e8cce1 100644
--- a/drivers/gpu/drm/vmwgfx/Makefile
+++ b/drivers/gpu/drm/vmwgfx/Makefile
@@ -8,7 +8,7 @@ vmwgfx-y := vmwgfx_execbuf.o vmwgfx_gmr.o vmwgfx_kms.o vmwgfx_drv.o \
 	    vmwgfx_cmdbuf_res.o vmwgfx_cmdbuf.o vmwgfx_stdu.o \
 	    vmwgfx_cotable.o vmwgfx_so.o vmwgfx_binding.o vmwgfx_msg.o \
 	    vmwgfx_simple_resource.o vmwgfx_va.o vmwgfx_blit.o \
-	    vmwgfx_validation.o vmwgfx_page_dirty.o \
+	    vmwgfx_validation.o vmwgfx_page_dirty.o vmwgfx_streamoutput.o \
 	    ttm_object.o ttm_lock.o
 
 obj-$(CONFIG_DRM_VMWGFX) := vmwgfx.o
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c b/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c
index f923b3c7c152..f41550797970 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c
@@ -78,6 +78,7 @@
  * @index_buffer: Index buffer binding.
  * @per_shader: Per shader-type bindings.
  * @ua_views: UAV bindings.
+ * @so_state: StreamOutput bindings.
  * @dirty: Bitmap tracking per binding-type changes that have not yet
  * been emitted to the device.
  * @dirty_vb: Bitmap tracking individual vertex buffer binding changes that
@@ -103,6 +104,7 @@ struct vmw_ctx_binding_state {
 	struct vmw_ctx_bindinfo_ib index_buffer;
 	struct vmw_dx_shader_bindings per_shader[SVGA3D_NUM_SHADERTYPE];
 	struct vmw_ctx_bindinfo_uav ua_views[VMW_MAX_UAV_BIND_TYPE];
+	struct vmw_ctx_bindinfo_so so_state;
 
 	unsigned long dirty;
 	DECLARE_BITMAP(dirty_vb, SVGA3D_DX_MAX_VERTEXBUFFERS);
@@ -127,6 +129,7 @@ static int vmw_binding_scrub_ib(struct vmw_ctx_bindinfo *bi, bool rebind);
 static int vmw_binding_scrub_vb(struct vmw_ctx_bindinfo *bi, bool rebind);
 static int vmw_binding_scrub_uav(struct vmw_ctx_bindinfo *bi, bool rebind);
 static int vmw_binding_scrub_cs_uav(struct vmw_ctx_bindinfo *bi, bool rebind);
+static int vmw_binding_scrub_so(struct vmw_ctx_bindinfo *bi, bool rebind);
 
 static void vmw_binding_build_asserts(void) __attribute__ ((unused));
 
@@ -202,6 +205,9 @@ static const size_t vmw_binding_uav_offsets[] = {
 static const size_t vmw_binding_cs_uav_offsets[] = {
 	offsetof(struct vmw_ctx_binding_state, ua_views[1].views),
 };
+static const size_t vmw_binding_so_offsets[] = {
+	offsetof(struct vmw_ctx_binding_state, so_state),
+};
 
 static const struct vmw_binding_info vmw_binding_infos[] = {
 	[vmw_ctx_binding_shader] = {
@@ -256,6 +262,10 @@ static const struct vmw_binding_info vmw_binding_infos[] = {
 		.size = sizeof(struct vmw_ctx_bindinfo_view),
 		.offsets = vmw_binding_cs_uav_offsets,
 		.scrub_func = vmw_binding_scrub_cs_uav},
+	[vmw_ctx_binding_so] = {
+		.size = sizeof(struct vmw_ctx_bindinfo_so),
+		.offsets = vmw_binding_so_offsets,
+		.scrub_func = vmw_binding_scrub_so},
 };
 
 /**
@@ -1291,6 +1301,33 @@ static int vmw_binding_scrub_cs_uav(struct vmw_ctx_bindinfo *bi, bool rebind)
 }
 
 /**
+ * vmw_binding_scrub_so - Scrub a streamoutput binding from context.
+ * @bi: Single binding information.
+ * @rebind: Whether to issue a bind instead of scrub command.
+ */
+static int vmw_binding_scrub_so(struct vmw_ctx_bindinfo *bi, bool rebind)
+{
+	struct vmw_ctx_bindinfo_so *binding =
+		container_of(bi, typeof(*binding), bi);
+	struct vmw_private *dev_priv = bi->ctx->dev_priv;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdDXSetStreamOutput body;
+	} *cmd;
+
+	cmd = VMW_FIFO_RESERVE_DX(dev_priv, sizeof(*cmd), bi->ctx->id);
+	if (!cmd)
+		return -ENOMEM;
+
+	cmd->header.id = SVGA_3D_CMD_DX_SET_STREAMOUTPUT;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.soid = rebind ? bi->res->id : SVGA3D_INVALID_ID;
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+
+	return 0;
+}
+
+/**
  * vmw_binding_state_alloc - Allocate a struct vmw_ctx_binding_state with
  * memory accounting.
  *
@@ -1393,7 +1430,7 @@ u32 vmw_binding_dirtying(enum vmw_ctx_binding_type binding_type)
 	};
 
 	/* Review this function as new bindings are added. */
-	BUILD_BUG_ON(vmw_ctx_binding_max != 13);
+	BUILD_BUG_ON(vmw_ctx_binding_max != 14);
 	return is_binding_dirtying[binding_type];
 }
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_binding.h b/drivers/gpu/drm/vmwgfx/vmwgfx_binding.h
index 8d5f87aad490..dcb71fd0bb3b 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_binding.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_binding.h
@@ -55,6 +55,7 @@ enum vmw_ctx_binding_type {
 	vmw_ctx_binding_ib,
 	vmw_ctx_binding_uav,
 	vmw_ctx_binding_cs_uav,
+	vmw_ctx_binding_so,
 	vmw_ctx_binding_max
 };
 
@@ -203,6 +204,16 @@ struct vmw_ctx_bindinfo_uav {
 	uint32 index;
 };
 
+/**
+ * struct vmw_ctx_bindinfo_so - Stream output binding metadata.
+ * @bi: struct vmw_ctx_bindinfo we derive from.
+ * @slot: Device data used to reconstruct binding command.
+ */
+struct vmw_ctx_bindinfo_so {
+	struct vmw_ctx_bindinfo bi;
+	uint32 slot;
+};
+
 extern void vmw_binding_add(struct vmw_ctx_binding_state *cbs,
 			    const struct vmw_ctx_bindinfo *ci,
 			    u32 shader_slot, u32 slot);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c b/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c
index 08c5b2ccca98..65e8e7a97724 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c
@@ -80,7 +80,7 @@ static const struct vmw_cotable_info co_info[] = {
 	{1, sizeof(SVGACOTableDXDepthStencilEntry), NULL},
 	{1, sizeof(SVGACOTableDXRasterizerStateEntry), NULL},
 	{1, sizeof(SVGACOTableDXSamplerEntry), NULL},
-	{1, sizeof(SVGACOTableDXStreamOutputEntry), NULL},
+	{1, sizeof(SVGACOTableDXStreamOutputEntry), &vmw_dx_streamoutput_cotable_list_scrub},
 	{1, sizeof(SVGACOTableDXQueryEntry), NULL},
 	{1, sizeof(SVGACOTableDXShaderEntry), &vmw_dx_shader_cotable_list_scrub},
 	{1, sizeof(SVGACOTableDXUAViewEntry), &vmw_view_cotable_list_destroy}
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index 262e57623df4..cfd8eaf7973e 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -202,6 +202,7 @@ enum vmw_res_type {
 	vmw_res_dx_context,
 	vmw_res_cotable,
 	vmw_res_view,
+	vmw_res_streamoutput,
 	vmw_res_max
 };
 
@@ -210,7 +211,8 @@ enum vmw_res_type {
  */
 enum vmw_cmdbuf_res_type {
 	vmw_cmdbuf_res_shader,
-	vmw_cmdbuf_res_view
+	vmw_cmdbuf_res_view,
+	vmw_cmdbuf_res_streamoutput
 };
 
 struct vmw_cmdbuf_res_manager;
@@ -1308,6 +1310,24 @@ vmw_shader_lookup(struct vmw_cmdbuf_res_manager *man,
 		  u32 user_key, SVGA3dShaderType shader_type);
 
 /*
+ * Streamoutput management
+ */
+struct vmw_resource *
+vmw_dx_streamoutput_lookup(struct vmw_cmdbuf_res_manager *man,
+			   u32 user_key);
+int vmw_dx_streamoutput_add(struct vmw_cmdbuf_res_manager *man,
+			    struct vmw_resource *ctx,
+			    SVGA3dStreamOutputId user_key,
+			    struct list_head *list);
+void vmw_dx_streamoutput_set_size(struct vmw_resource *res, u32 size);
+int vmw_dx_streamoutput_remove(struct vmw_cmdbuf_res_manager *man,
+			       SVGA3dStreamOutputId user_key,
+			       struct list_head *list);
+void vmw_dx_streamoutput_cotable_list_scrub(struct vmw_private *dev_priv,
+					    struct list_head *list,
+					    bool readback);
+
+/*
  * Command buffer managed resources - vmwgfx_cmdbuf_res.c
  */
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
index d49169c68fba..367d5b87ee6a 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
@@ -2948,6 +2948,169 @@ static int vmw_cmd_set_cs_uav(struct vmw_private *dev_priv,
 	return ret;
 }
 
+static int vmw_cmd_dx_define_streamoutput(struct vmw_private *dev_priv,
+					  struct vmw_sw_context *sw_context,
+					  SVGA3dCmdHeader *header)
+{
+	struct vmw_ctx_validation_info *ctx_node = sw_context->dx_ctx_node;
+	struct vmw_resource *res;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdDXDefineStreamOutputWithMob body;
+	} *cmd = container_of(header, typeof(*cmd), header);
+	int ret;
+
+	if (!has_sm5_context(dev_priv))
+		return -EINVAL;
+
+	if (!ctx_node) {
+		DRM_ERROR("DX Context not set.\n");
+		return -EINVAL;
+	}
+
+	res = vmw_context_cotable(ctx_node->ctx, SVGA_COTABLE_STREAMOUTPUT);
+	ret = vmw_cotable_notify(res, cmd->body.soid);
+	if (ret)
+		return ret;
+
+	return vmw_dx_streamoutput_add(sw_context->man, ctx_node->ctx,
+				       cmd->body.soid,
+				       &sw_context->staged_cmd_res);
+}
+
+static int vmw_cmd_dx_destroy_streamoutput(struct vmw_private *dev_priv,
+					   struct vmw_sw_context *sw_context,
+					   SVGA3dCmdHeader *header)
+{
+	struct vmw_ctx_validation_info *ctx_node = sw_context->dx_ctx_node;
+	struct vmw_resource *res;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdDXDestroyStreamOutput body;
+	} *cmd = container_of(header, typeof(*cmd), header);
+
+	if (!ctx_node) {
+		DRM_ERROR("DX Context not set.\n");
+		return -EINVAL;
+	}
+
+	/*
+	 * When device does not support SM5 then streamoutput with mob command is
+	 * not available to user-space. Simply return in this case.
+	 */
+	if (!has_sm5_context(dev_priv))
+		return 0;
+
+	/*
+	 * With SM5 capable device if lookup fails then user-space probably used
+	 * old streamoutput define command. Return without an error.
+	 */
+	res = vmw_dx_streamoutput_lookup(vmw_context_res_man(ctx_node->ctx),
+					 cmd->body.soid);
+	if (IS_ERR(res))
+		return 0;
+
+	return vmw_dx_streamoutput_remove(sw_context->man, cmd->body.soid,
+					  &sw_context->staged_cmd_res);
+}
+
+static int vmw_cmd_dx_bind_streamoutput(struct vmw_private *dev_priv,
+					struct vmw_sw_context *sw_context,
+					SVGA3dCmdHeader *header)
+{
+	struct vmw_ctx_validation_info *ctx_node = sw_context->dx_ctx_node;
+	struct vmw_resource *res;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdDXBindStreamOutput body;
+	} *cmd = container_of(header, typeof(*cmd), header);
+	int ret;
+
+	if (!has_sm5_context(dev_priv))
+		return -EINVAL;
+
+	if (!ctx_node) {
+		DRM_ERROR("DX Context not set.\n");
+		return -EINVAL;
+	}
+
+	res = vmw_dx_streamoutput_lookup(vmw_context_res_man(ctx_node->ctx),
+					 cmd->body.soid);
+	if (IS_ERR(res)) {
+		DRM_ERROR("Cound not find streamoutput to bind.\n");
+		return PTR_ERR(res);
+	}
+
+	vmw_dx_streamoutput_set_size(res, cmd->body.sizeInBytes);
+
+	ret = vmw_execbuf_res_noctx_val_add(sw_context, res,
+					    VMW_RES_DIRTY_NONE);
+	if (ret) {
+		DRM_ERROR("Error creating resource validation node.\n");
+		return ret;
+	}
+
+	return vmw_cmd_res_switch_backup(dev_priv, sw_context, res,
+					 &cmd->body.mobid,
+					 cmd->body.offsetInBytes);
+}
+
+static int vmw_cmd_dx_set_streamoutput(struct vmw_private *dev_priv,
+				       struct vmw_sw_context *sw_context,
+				       SVGA3dCmdHeader *header)
+{
+	struct vmw_ctx_validation_info *ctx_node = sw_context->dx_ctx_node;
+	struct vmw_resource *res;
+	struct vmw_ctx_bindinfo_so binding;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdDXSetStreamOutput body;
+	} *cmd = container_of(header, typeof(*cmd), header);
+	int ret;
+
+	if (!ctx_node) {
+		DRM_ERROR("DX Context not set.\n");
+		return -EINVAL;
+	}
+
+	if (cmd->body.soid == SVGA3D_INVALID_ID)
+		return 0;
+
+	/*
+	 * When device does not support SM5 then streamoutput with mob command is
+	 * not available to user-space. Simply return in this case.
+	 */
+	if (!has_sm5_context(dev_priv))
+		return 0;
+
+	/*
+	 * With SM5 capable device if lookup fails then user-space probably used
+	 * old streamoutput define command. Return without an error.
+	 */
+	res = vmw_dx_streamoutput_lookup(vmw_context_res_man(ctx_node->ctx),
+					 cmd->body.soid);
+	if (IS_ERR(res)) {
+		return 0;
+	}
+
+	ret = vmw_execbuf_res_noctx_val_add(sw_context, res,
+					    VMW_RES_DIRTY_NONE);
+	if (ret) {
+		DRM_ERROR("Error creating resource validation node.\n");
+		return ret;
+	}
+
+	binding.bi.ctx = ctx_node->ctx;
+	binding.bi.res = res;
+	binding.bi.bt = vmw_ctx_binding_so;
+	binding.slot = 0; /* Only one SO set to context at a time. */
+
+	vmw_binding_add(sw_context->dx_ctx_node->staged, &binding.bi, 0,
+			binding.slot);
+
+	return ret;
+}
+
 static int vmw_cmd_indexed_instanced_indirect(struct vmw_private *dev_priv,
 					      struct vmw_sw_context *sw_context,
 					      SVGA3dCmdHeader *header)
@@ -3330,9 +3493,9 @@ static const struct vmw_cmd_entry vmw_cmd_entries[SVGA_3D_CMD_MAX] = {
 	VMW_CMD_DEF(SVGA_3D_CMD_DX_DEFINE_STREAMOUTPUT,
 		    &vmw_cmd_dx_so_define, true, false, true),
 	VMW_CMD_DEF(SVGA_3D_CMD_DX_DESTROY_STREAMOUTPUT,
-		    &vmw_cmd_dx_cid_check, true, false, true),
-	VMW_CMD_DEF(SVGA_3D_CMD_DX_SET_STREAMOUTPUT, &vmw_cmd_dx_cid_check,
-		    true, false, true),
+		    &vmw_cmd_dx_destroy_streamoutput, true, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DX_SET_STREAMOUTPUT,
+		    &vmw_cmd_dx_set_streamoutput, true, false, true),
 	VMW_CMD_DEF(SVGA_3D_CMD_DX_SET_SOTARGETS,
 		    &vmw_cmd_dx_set_so_targets, true, false, true),
 	VMW_CMD_DEF(SVGA_3D_CMD_DX_SET_INPUT_LAYOUT,
@@ -3375,6 +3538,10 @@ static const struct vmw_cmd_entry vmw_cmd_entries[SVGA_3D_CMD_MAX] = {
 		    false, true),
 	VMW_CMD_DEF(SVGA_3D_CMD_DX_DEFINE_DEPTHSTENCIL_VIEW_V2,
 		    &vmw_cmd_sm5_view_define, true, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DX_DEFINE_STREAMOUTPUT_WITH_MOB,
+		    &vmw_cmd_dx_define_streamoutput, true, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DX_BIND_STREAMOUTPUT,
+		    &vmw_cmd_dx_bind_streamoutput, true, false, true),
 };
 
 bool vmw_cmd_describe(const void *buf, u32 *size, char const **cmd)
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_so.h b/drivers/gpu/drm/vmwgfx/vmwgfx_so.h
index 22b4f5720908..f48b84bfeeac 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_so.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_so.h
@@ -129,6 +129,7 @@ static inline enum vmw_so_type vmw_so_cmd_to_type(u32 id)
 	case SVGA_3D_CMD_DX_DESTROY_SAMPLER_STATE:
 		return vmw_so_ss;
 	case SVGA_3D_CMD_DX_DEFINE_STREAMOUTPUT:
+	case SVGA_3D_CMD_DX_DEFINE_STREAMOUTPUT_WITH_MOB:
 	case SVGA_3D_CMD_DX_DESTROY_STREAMOUTPUT:
 		return vmw_so_so;
 	default:
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_streamoutput.c b/drivers/gpu/drm/vmwgfx/vmwgfx_streamoutput.c
new file mode 100644
index 000000000000..193192456663
--- /dev/null
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_streamoutput.c
@@ -0,0 +1,387 @@
+// SPDX-License-Identifier: GPL-2.0 OR MIT
+/**************************************************************************
+ *
+ * Copyright © 2018-2019 VMware, Inc., Palo Alto, CA., USA
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#include <drm/ttm/ttm_placement.h>
+
+#include "vmwgfx_drv.h"
+#include "vmwgfx_resource_priv.h"
+#include "vmwgfx_binding.h"
+
+/**
+ * struct vmw_dx_streamoutput - Streamoutput resource metadata.
+ * @res: Base resource struct.
+ * @ctx: Non-refcounted context to which @res belong.
+ * @cotable: Refcounted cotable holding this Streamoutput.
+ * @cotable_head: List head for cotable-so_res list.
+ * @id: User-space provided identifier.
+ * @size: User-space provided mob size.
+ * @committed: Whether streamoutput is actually created or pending creation.
+ */
+struct vmw_dx_streamoutput {
+	struct vmw_resource res;
+	struct vmw_resource *ctx;
+	struct vmw_resource *cotable;
+	struct list_head cotable_head;
+	u32 id;
+	u32 size;
+	bool committed;
+};
+
+static int vmw_dx_streamoutput_create(struct vmw_resource *res);
+static int vmw_dx_streamoutput_bind(struct vmw_resource *res,
+				    struct ttm_validate_buffer *val_buf);
+static int vmw_dx_streamoutput_unbind(struct vmw_resource *res, bool readback,
+				      struct ttm_validate_buffer *val_buf);
+static void vmw_dx_streamoutput_commit_notify(struct vmw_resource *res,
+					      enum vmw_cmdbuf_res_state state);
+
+static size_t vmw_streamoutput_size;
+
+static const struct vmw_res_func vmw_dx_streamoutput_func = {
+	.res_type = vmw_res_streamoutput,
+	.needs_backup = true,
+	.may_evict = false,
+	.type_name = "DX streamoutput",
+	.backup_placement = &vmw_mob_placement,
+	.create = vmw_dx_streamoutput_create,
+	.destroy = NULL, /* Command buffer managed resource. */
+	.bind = vmw_dx_streamoutput_bind,
+	.unbind = vmw_dx_streamoutput_unbind,
+	.commit_notify = vmw_dx_streamoutput_commit_notify,
+};
+
+static inline struct vmw_dx_streamoutput *
+vmw_res_to_dx_streamoutput(struct vmw_resource *res)
+{
+	return container_of(res, struct vmw_dx_streamoutput, res);
+}
+
+/**
+ * vmw_dx_streamoutput_unscrub - Reattach the MOB to streamoutput.
+ * @res: The streamoutput resource.
+ *
+ * Return: 0 on success, negative error code on failure.
+ */
+static int vmw_dx_streamoutput_unscrub(struct vmw_resource *res)
+{
+	struct vmw_dx_streamoutput *so = vmw_res_to_dx_streamoutput(res);
+	struct vmw_private *dev_priv = res->dev_priv;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdDXBindStreamOutput body;
+	} *cmd;
+
+	if (!list_empty(&so->cotable_head) || !so->committed )
+		return 0;
+
+	cmd = VMW_FIFO_RESERVE_DX(dev_priv, sizeof(*cmd), so->ctx->id);
+	if (!cmd)
+		return -ENOMEM;
+
+	cmd->header.id = SVGA_3D_CMD_DX_BIND_STREAMOUTPUT;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.soid = so->id;
+	cmd->body.mobid = res->backup->base.mem.start;
+	cmd->body.offsetInBytes = res->backup_offset;
+	cmd->body.sizeInBytes = so->size;
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+
+	vmw_cotable_add_resource(so->cotable, &so->cotable_head);
+
+	return 0;
+}
+
+static int vmw_dx_streamoutput_create(struct vmw_resource *res)
+{
+	struct vmw_private *dev_priv = res->dev_priv;
+	struct vmw_dx_streamoutput *so = vmw_res_to_dx_streamoutput(res);
+	int ret = 0;
+
+	WARN_ON_ONCE(!so->committed);
+
+	if (vmw_resource_mob_attached(res)) {
+		mutex_lock(&dev_priv->binding_mutex);
+		ret = vmw_dx_streamoutput_unscrub(res);
+		mutex_unlock(&dev_priv->binding_mutex);
+	}
+
+	res->id = so->id;
+
+	return ret;
+}
+
+static int vmw_dx_streamoutput_bind(struct vmw_resource *res,
+				    struct ttm_validate_buffer *val_buf)
+{
+	struct vmw_private *dev_priv = res->dev_priv;
+	struct ttm_buffer_object *bo = val_buf->bo;
+	int ret;
+
+	if (WARN_ON(bo->mem.mem_type != VMW_PL_MOB))
+		return -EINVAL;
+
+	mutex_lock(&dev_priv->binding_mutex);
+	ret = vmw_dx_streamoutput_unscrub(res);
+	mutex_unlock(&dev_priv->binding_mutex);
+
+	return ret;
+}
+
+/**
+ * vmw_dx_streamoutput_scrub - Unbind the MOB from streamoutput.
+ * @res: The streamoutput resource.
+ *
+ * Return: 0 on success, negative error code on failure.
+ */
+static int vmw_dx_streamoutput_scrub(struct vmw_resource *res)
+{
+	struct vmw_private *dev_priv = res->dev_priv;
+	struct vmw_dx_streamoutput *so = vmw_res_to_dx_streamoutput(res);
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdDXBindStreamOutput body;
+	} *cmd;
+
+	if (list_empty(&so->cotable_head))
+		return 0;
+
+	WARN_ON_ONCE(!so->committed);
+
+	cmd = VMW_FIFO_RESERVE_DX(dev_priv, sizeof(*cmd), so->ctx->id);
+	if (!cmd)
+		return -ENOMEM;
+
+	cmd->header.id = SVGA_3D_CMD_DX_BIND_STREAMOUTPUT;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.soid = res->id;
+	cmd->body.mobid = SVGA3D_INVALID_ID;
+	cmd->body.offsetInBytes = 0;
+	cmd->body.sizeInBytes = so->size;
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+
+	res->id = -1;
+	list_del_init(&so->cotable_head);
+
+	return 0;
+}
+
+static int vmw_dx_streamoutput_unbind(struct vmw_resource *res, bool readback,
+				      struct ttm_validate_buffer *val_buf)
+{
+	struct vmw_private *dev_priv = res->dev_priv;
+	struct vmw_fence_obj *fence;
+	int ret;
+
+	if (WARN_ON(res->backup->base.mem.mem_type != VMW_PL_MOB))
+		return -EINVAL;
+
+	mutex_lock(&dev_priv->binding_mutex);
+	ret = vmw_dx_streamoutput_scrub(res);
+	mutex_unlock(&dev_priv->binding_mutex);
+
+	if (ret)
+		return ret;
+
+	(void) vmw_execbuf_fence_commands(NULL, dev_priv, &fence, NULL);
+	vmw_bo_fence_single(val_buf->bo, fence);
+
+	if (fence != NULL)
+		vmw_fence_obj_unreference(&fence);
+
+	return 0;
+}
+
+static void vmw_dx_streamoutput_commit_notify(struct vmw_resource *res,
+					   enum vmw_cmdbuf_res_state state)
+{
+	struct vmw_private *dev_priv = res->dev_priv;
+	struct vmw_dx_streamoutput *so = vmw_res_to_dx_streamoutput(res);
+
+	if (state == VMW_CMDBUF_RES_ADD) {
+		mutex_lock(&dev_priv->binding_mutex);
+		vmw_cotable_add_resource(so->cotable, &so->cotable_head);
+		so->committed = true;
+		res->id = so->id;
+		mutex_unlock(&dev_priv->binding_mutex);
+	} else {
+		mutex_lock(&dev_priv->binding_mutex);
+		list_del_init(&so->cotable_head);
+		so->committed = false;
+		res->id = -1;
+		mutex_unlock(&dev_priv->binding_mutex);
+	}
+}
+
+/**
+ * vmw_dx_streamoutput_lookup - Do a streamoutput resource lookup by user key.
+ * @man: Command buffer managed resource manager for current context.
+ * @user_key: User-space identifier for lookup.
+ *
+ * Return: Valid refcounted vmw_resource on success, error pointer on failure.
+ */
+struct vmw_resource *
+vmw_dx_streamoutput_lookup(struct vmw_cmdbuf_res_manager *man,
+			   u32 user_key)
+{
+	return vmw_cmdbuf_res_lookup(man, vmw_cmdbuf_res_streamoutput,
+				     user_key);
+}
+
+static void vmw_dx_streamoutput_res_free(struct vmw_resource *res)
+{
+	struct vmw_private *dev_priv = res->dev_priv;
+	struct vmw_dx_streamoutput *so = vmw_res_to_dx_streamoutput(res);
+
+	vmw_resource_unreference(&so->cotable);
+	kfree(so);
+	ttm_mem_global_free(vmw_mem_glob(dev_priv), vmw_streamoutput_size);
+}
+
+static void vmw_dx_streamoutput_hw_destroy(struct vmw_resource *res)
+{
+	/* Destroyed by user-space cmd buf or as part of context takedown. */
+	res->id = -1;
+}
+
+/**
+ * vmw_dx_streamoutput_add - Add a streamoutput as a cmd buf managed resource.
+ * @man: Command buffer managed resource manager for current context.
+ * @ctx: Pointer to context resource.
+ * @user_key: The identifier for this streamoutput.
+ * @list: The list of staged command buffer managed resources.
+ *
+ * Return: 0 on success, negative error code on failure.
+ */
+int vmw_dx_streamoutput_add(struct vmw_cmdbuf_res_manager *man,
+			    struct vmw_resource *ctx, u32 user_key,
+			    struct list_head *list)
+{
+	struct vmw_dx_streamoutput *so;
+	struct vmw_resource *res;
+	struct vmw_private *dev_priv = ctx->dev_priv;
+	struct ttm_operation_ctx ttm_opt_ctx = {
+		.interruptible = true,
+		.no_wait_gpu = false
+	};
+	int ret;
+
+	if (!vmw_streamoutput_size)
+		vmw_streamoutput_size = ttm_round_pot(sizeof(*so));
+
+	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
+				   vmw_streamoutput_size, &ttm_opt_ctx);
+	if (ret) {
+		if (ret != -ERESTARTSYS)
+			DRM_ERROR("Out of graphics memory for streamout.\n");
+		return ret;
+	}
+
+	so = kmalloc(sizeof(*so), GFP_KERNEL);
+	if (!so) {
+		ttm_mem_global_free(vmw_mem_glob(dev_priv),
+				    vmw_streamoutput_size);
+		return -ENOMEM;
+	}
+
+	res = &so->res;
+	so->ctx = ctx;
+	so->cotable = vmw_resource_reference
+		(vmw_context_cotable(ctx, SVGA_COTABLE_STREAMOUTPUT));
+	so->id = user_key;
+	so->committed = false;
+	INIT_LIST_HEAD(&so->cotable_head);
+	ret = vmw_resource_init(dev_priv, res, true,
+				vmw_dx_streamoutput_res_free,
+				&vmw_dx_streamoutput_func);
+	if (ret)
+		goto out_resource_init;
+
+	ret = vmw_cmdbuf_res_add(man, vmw_cmdbuf_res_streamoutput, user_key,
+				 res, list);
+	if (ret)
+		goto out_resource_init;
+
+	res->id = so->id;
+	res->hw_destroy = vmw_dx_streamoutput_hw_destroy;
+
+out_resource_init:
+	vmw_resource_unreference(&res);
+
+	return ret;
+}
+
+/**
+ * vmw_dx_streamoutput_set_size - Sets streamoutput mob size in res struct.
+ * @res: The streamoutput res for which need to set size.
+ * @size: The size provided by user-space to set.
+ */
+void vmw_dx_streamoutput_set_size(struct vmw_resource *res, u32 size)
+{
+	struct vmw_dx_streamoutput *so = vmw_res_to_dx_streamoutput(res);
+
+	so->size = size;
+}
+
+/**
+ * vmw_dx_streamoutput_remove - Stage streamoutput for removal.
+ * @man: Command buffer managed resource manager for current context.
+ * @user_key: The identifier for this streamoutput.
+ * @list: The list of staged command buffer managed resources.
+ *
+ * Return: 0 on success, negative error code on failure.
+ */
+int vmw_dx_streamoutput_remove(struct vmw_cmdbuf_res_manager *man,
+			       u32 user_key,
+			       struct list_head *list)
+{
+	struct vmw_resource *r;
+
+	return vmw_cmdbuf_res_remove(man, vmw_cmdbuf_res_streamoutput,
+				     (u32)user_key, list, &r);
+}
+
+/**
+ * vmw_dx_streamoutput_cotable_list_scrub - cotable unbind_func callback.
+ * @dev_priv: Device private.
+ * @list: The list of cotable resources.
+ * @readback: Whether the call was part of a readback unbind.
+ */
+void vmw_dx_streamoutput_cotable_list_scrub(struct vmw_private *dev_priv,
+					    struct list_head *list,
+					    bool readback)
+{
+	struct vmw_dx_streamoutput *entry, *next;
+
+	lockdep_assert_held_once(&dev_priv->binding_mutex);
+
+	list_for_each_entry_safe(entry, next, list, cotable_head) {
+		WARN_ON(vmw_dx_streamoutput_scrub(&entry->res));
+		if (!readback)
+			entry->committed =false;
+	}
+}