drm/vmwgfx: Convert screen targets to new helpers v3

Also implements the missing readback function and
fixes page flip in case of no event.

v2:
- Adapt to the work done for screen targets for 2d, in particular
Handle proxy surface updates.
- Remove execbuf quirks since we now use fifo reserve / commit.
- Revert the initial placement of vmw dma buffers.

v3: Address review comments.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>

6 files changed, 534 insertions, 679 deletions

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index d60ae207136c..d6b247b1994a 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -343,9 +343,6 @@ enum vmw_display_unit_type {
 };
 
 
-#define VMW_QUIRK_DST_SID_OK (1U << 0)
-#define VMW_QUIRK_SRC_SID_OK (1U << 1)
-
 struct vmw_sw_context{
 	struct drm_open_hash res_ht;
 	bool res_ht_initialized;
@@ -368,7 +365,6 @@ struct vmw_sw_context{
 	struct vmw_resource *error_resource;
 	struct vmw_ctx_binding_state staged_bindings;
 	struct list_head staged_cmd_res;
-	uint32_t quirks;
 };
 
 struct vmw_legacy_display;
@@ -842,7 +838,6 @@ extern int vmw_execbuf_process(struct drm_file *file_priv,
 			       void *kernel_commands,
 			       uint32_t command_size,
 			       uint64_t throttle_us,
-			       uint32_t quirks,
 			       struct drm_vmw_fence_rep __user
 			       *user_fence_rep,
 			       struct vmw_fence_obj **out_fence);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
index 698a0e2add53..64dba53ca54c 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
@@ -675,16 +675,11 @@ static int vmw_cmd_surface_copy_check(struct vmw_private *dev_priv,
 
 	cmd = container_of(header, struct vmw_sid_cmd, header);
 
-	if (!(sw_context->quirks & VMW_QUIRK_SRC_SID_OK)) {
-		ret = vmw_cmd_res_check(dev_priv, sw_context, vmw_res_surface,
-					user_surface_converter,
-					&cmd->body.src.sid, NULL);
-		if (ret != 0)
-			return ret;
-	}
-
-	if (sw_context->quirks & VMW_QUIRK_DST_SID_OK)
-		return 0;
+	ret = vmw_cmd_res_check(dev_priv, sw_context, vmw_res_surface,
+				user_surface_converter,
+				&cmd->body.src.sid, NULL);
+	if (ret)
+		return ret;
 
 	return vmw_cmd_res_check(dev_priv, sw_context, vmw_res_surface,
 				 user_surface_converter,
@@ -1266,9 +1261,6 @@ static int vmw_cmd_dma(struct vmw_private *dev_priv,
 	if (unlikely(suffix->maximumOffset > bo_size))
 		suffix->maximumOffset = bo_size;
 
-	if (sw_context->quirks & VMW_QUIRK_DST_SID_OK)
-		goto out_no_surface;
-
 	ret = vmw_cmd_res_check(dev_priv, sw_context, vmw_res_surface,
 				user_surface_converter, &cmd->dma.host.sid,
 				NULL);
@@ -1507,9 +1499,6 @@ static int vmw_cmd_update_gb_image(struct vmw_private *dev_priv,
 
 	cmd = container_of(header, struct vmw_gb_surface_cmd, header);
 
-	if (sw_context->quirks & VMW_QUIRK_SRC_SID_OK)
-		return 0;
-
 	return vmw_cmd_res_check(dev_priv, sw_context, vmw_res_surface,
 				 user_surface_converter,
 				 &cmd->body.image.sid, NULL);
@@ -2554,7 +2543,6 @@ int vmw_execbuf_process(struct drm_file *file_priv,
 			void *kernel_commands,
 			uint32_t command_size,
 			uint64_t throttle_us,
-			uint32_t quirks,
 			struct drm_vmw_fence_rep __user *user_fence_rep,
 			struct vmw_fence_obj **out_fence)
 {
@@ -2609,7 +2597,6 @@ int vmw_execbuf_process(struct drm_file *file_priv,
 	sw_context->fp = vmw_fpriv(file_priv);
 	sw_context->cur_reloc = 0;
 	sw_context->cur_val_buf = 0;
-	sw_context->quirks = quirks;
 	INIT_LIST_HEAD(&sw_context->resource_list);
 	sw_context->cur_query_bo = dev_priv->pinned_bo;
 	sw_context->last_query_ctx = NULL;
@@ -2921,7 +2908,6 @@ int vmw_execbuf_ioctl(struct drm_device *dev, void *data,
 	ret = vmw_execbuf_process(file_priv, dev_priv,
 				  (void __user *)(unsigned long)arg->commands,
 				  NULL, arg->command_size, arg->throttle_us,
-				  0,
 				  (void __user *)(unsigned long)arg->fence_rep,
 				  NULL);
 	ttm_read_unlock(&dev_priv->reservation_sem);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
index 5901d32d0273..234a3cef1c25 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
@@ -427,10 +427,9 @@ static int vmw_framebuffer_surface_dirty(struct drm_framebuffer *framebuffer,
 						   clips, NULL, NULL, 0, 0,
 						   num_clips, inc, NULL);
 	else
-		ret = vmw_kms_stdu_do_surface_dirty(dev_priv, file_priv,
-						    &vfbs->base,
-						    clips, num_clips,
-						    inc);
+		ret = vmw_kms_stdu_surface_dirty(dev_priv, &vfbs->base,
+						 clips, NULL, NULL, 0, 0,
+						 num_clips, inc, NULL);
 
 	vmw_fifo_flush(dev_priv, false);
 	ttm_read_unlock(&dev_priv->reservation_sem);
@@ -467,10 +466,14 @@ int vmw_kms_readback(struct vmw_private *dev_priv,
 	case vmw_du_screen_object:
 		return vmw_kms_sou_readback(dev_priv, file_priv, vfb,
 					    user_fence_rep, vclips, num_clips);
+	case vmw_du_screen_target:
+		return vmw_kms_stdu_dma(dev_priv, file_priv, vfb,
+					user_fence_rep, NULL, vclips, num_clips,
+					1, false, true);
 	default:
 		WARN_ONCE(true,
 			  "Readback called with invalid display system.\n");
-	}
+}
 
 	return -ENOSYS;
 }
@@ -632,20 +635,23 @@ static int vmw_framebuffer_dmabuf_dirty(struct drm_framebuffer *framebuffer,
 		increment = 2;
 	}
 
-	if (dev_priv->ldu_priv) {
-		ret = vmw_kms_ldu_do_dmabuf_dirty(dev_priv, &vfbd->base,
-						  flags, color,
-						  clips, num_clips, increment);
-	} else if (dev_priv->active_display_unit == vmw_du_screen_object) {
+	switch (dev_priv->active_display_unit) {
+	case vmw_du_screen_target:
+		ret = vmw_kms_stdu_dma(dev_priv, NULL, &vfbd->base, NULL,
+				       clips, NULL, num_clips, increment,
+				       true, true);
+		break;
+	case vmw_du_screen_object:
 		ret = vmw_kms_sou_do_dmabuf_dirty(dev_priv, &vfbd->base,
 						  clips, num_clips, increment,
 						  true,
 						  NULL);
-	} else {
-		ret = vmw_kms_stdu_do_surface_dirty(dev_priv, file_priv,
-						    &vfbd->base,
-						    clips, num_clips,
-						    increment);
+		break;
+	default:
+		ret = -ENOSYS;
+		WARN_ONCE(true,
+			  "Dirty called with invalid display system.\n");
+		break;
 	}
 
 	vmw_fifo_flush(dev_priv, false);
@@ -721,9 +727,9 @@ static int vmw_create_dmabuf_proxy(struct drm_device *dev,
 {
 	uint32_t format;
 	struct drm_vmw_size content_base_size;
+	struct vmw_resource *res;
 	int ret;
 
-
 	switch (mode_cmd->depth) {
 	case 32:
 	case 24:
@@ -762,15 +768,18 @@ static int vmw_create_dmabuf_proxy(struct drm_device *dev,
 		return ret;
 	}
 
-	/* Use the same MOB backing for surface */
-	vmw_dmabuf_reference(dmabuf_mob);
-
-	(*srf_out)->res.backup = dmabuf_mob;
+	res = &(*srf_out)->res;
 
-	/* FIXME:  Waiting for fbdev rework to do a proper reserve/pin */
-	ret = vmw_resource_validate(&(*srf_out)->res);
+	/* Reserve and switch the backing mob. */
+	mutex_lock(&res->dev_priv->cmdbuf_mutex);
+	(void) vmw_resource_reserve(res, false, true);
+	vmw_dmabuf_unreference(&res->backup);
+	res->backup = vmw_dmabuf_reference(dmabuf_mob);
+	res->backup_offset = 0;
+	vmw_resource_unreserve(res, NULL, 0);
+	mutex_unlock(&res->dev_priv->cmdbuf_mutex);
 
-	return ret;
+	return 0;
 }
 
 
@@ -987,6 +996,7 @@ int vmw_kms_generic_present(struct vmw_private *dev_priv,
 					    num_clips, 1, NULL);
 }
 
+
 int vmw_kms_present(struct vmw_private *dev_priv,
 		    struct drm_file *file_priv,
 		    struct vmw_framebuffer *vfb,
@@ -998,13 +1008,23 @@ int vmw_kms_present(struct vmw_private *dev_priv,
 {
 	int ret;
 
-	if (dev_priv->active_display_unit == vmw_du_screen_target)
-		ret = vmw_kms_stdu_present(dev_priv, file_priv, vfb, sid,
-					   destX, destY, clips, num_clips);
-	else
-		ret = vmw_kms_generic_present(dev_priv, file_priv, vfb,
-					      surface, sid, destX, destY,
-					      clips, num_clips);
+	switch (dev_priv->active_display_unit) {
+	case vmw_du_screen_target:
+		ret = vmw_kms_stdu_surface_dirty(dev_priv, vfb, NULL, clips,
+						 &surface->res, destX, destY,
+						 num_clips, 1, NULL);
+		break;
+	case vmw_du_screen_object:
+		ret = vmw_kms_generic_present(dev_priv, file_priv, vfb, surface,
+					      sid, destX, destY, clips,
+					      num_clips);
+		break;
+	default:
+		WARN_ONCE(true,
+			  "Present called with invalid display system.\n");
+		ret = -ENOSYS;
+		break;
+	}
 	if (ret)
 		return ret;
 
@@ -1882,3 +1902,72 @@ void vmw_kms_helper_resource_finish(struct vmw_resource *res,
 	vmw_resource_unreserve(res, NULL, 0);
 	mutex_unlock(&res->dev_priv->cmdbuf_mutex);
 }
+
+/**
+ * vmw_kms_update_proxy - Helper function to update a proxy surface from
+ * its backing MOB.
+ *
+ * @res: Pointer to the surface resource
+ * @clips: Clip rects in framebuffer (surface) space.
+ * @num_clips: Number of clips in @clips.
+ * @increment: Integer with which to increment the clip counter when looping.
+ * Used to skip a predetermined number of clip rects.
+ *
+ * This function makes sure the proxy surface is updated from its backing MOB
+ * using the region given by @clips. The surface resource @res and its backing
+ * MOB needs to be reserved and validated on call.
+ */
+int vmw_kms_update_proxy(struct vmw_resource *res,
+			 const struct drm_clip_rect *clips,
+			 unsigned num_clips,
+			 int increment)
+{
+	struct vmw_private *dev_priv = res->dev_priv;
+	struct drm_vmw_size *size = &vmw_res_to_srf(res)->base_size;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdUpdateGBImage body;
+	} *cmd;
+	SVGA3dBox *box;
+	size_t copy_size = 0;
+	int i;
+
+	if (!clips)
+		return 0;
+
+	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd) * num_clips);
+	if (!cmd) {
+		DRM_ERROR("Couldn't reserve fifo space for proxy surface "
+			  "update.\n");
+		return -ENOMEM;
+	}
+
+	for (i = 0; i < num_clips; ++i, clips += increment, ++cmd) {
+		box = &cmd->body.box;
+
+		cmd->header.id = SVGA_3D_CMD_UPDATE_GB_IMAGE;
+		cmd->header.size = sizeof(cmd->body);
+		cmd->body.image.sid = res->id;
+		cmd->body.image.face = 0;
+		cmd->body.image.mipmap = 0;
+
+		if (clips->x1 > size->width || clips->x2 > size->width ||
+		    clips->y1 > size->height || clips->y2 > size->height) {
+			DRM_ERROR("Invalid clips outsize of framebuffer.\n");
+			return -EINVAL;
+		}
+
+		box->x = clips->x1;
+		box->y = clips->y1;
+		box->z = 0;
+		box->w = clips->x2 - clips->x1;
+		box->h = clips->y2 - clips->y1;
+		box->d = 1;
+
+		copy_size += sizeof(*cmd);
+	}
+
+	vmw_fifo_commit(dev_priv, copy_size);
+
+	return 0;
+}
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
index 8a8203c66adc..f941f92338a6 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
@@ -239,6 +239,7 @@ int vmw_kms_readback(struct vmw_private *dev_priv,
 		     struct drm_vmw_rect *vclips,
 		     uint32_t num_clips);
 
+
 /*
  * Legacy display unit functions - vmwgfx_ldu.c
  */
@@ -249,6 +250,10 @@ int vmw_kms_ldu_do_dmabuf_dirty(struct vmw_private *dev_priv,
 				unsigned flags, unsigned color,
 				struct drm_clip_rect *clips,
 				unsigned num_clips, int increment);
+int vmw_kms_update_proxy(struct vmw_resource *res,
+			 const struct drm_clip_rect *clips,
+			 unsigned num_clips,
+			 int increment);
 
 /*
  * Screen Objects display functions - vmwgfx_scrn.c
@@ -282,17 +287,25 @@ int vmw_kms_sou_readback(struct vmw_private *dev_priv,
  */
 int vmw_kms_stdu_init_display(struct vmw_private *dev_priv);
 int vmw_kms_stdu_close_display(struct vmw_private *dev_priv);
-int vmw_kms_stdu_do_surface_dirty(struct vmw_private *dev_priv,
-				  struct drm_file *file_priv,
-				  struct vmw_framebuffer *framebuffer,
-				  struct drm_clip_rect *clips,
-				  unsigned num_clips, int increment);
-int vmw_kms_stdu_present(struct vmw_private *dev_priv,
-			 struct drm_file *file_priv,
-			 struct vmw_framebuffer *vfb,
-			 uint32_t user_handle,
-			 int32_t dest_x, int32_t dest_y,
-			 struct drm_vmw_rect *clips,
-			 uint32_t num_clips);
+int vmw_kms_stdu_surface_dirty(struct vmw_private *dev_priv,
+			       struct vmw_framebuffer *framebuffer,
+			       struct drm_clip_rect *clips,
+			       struct drm_vmw_rect *vclips,
+			       struct vmw_resource *srf,
+			       s32 dest_x,
+			       s32 dest_y,
+			       unsigned num_clips, int inc,
+			       struct vmw_fence_obj **out_fence);
+int vmw_kms_stdu_dma(struct vmw_private *dev_priv,
+		     struct drm_file *file_priv,
+		     struct vmw_framebuffer *vfb,
+		     struct drm_vmw_fence_rep __user *user_fence_rep,
+		     struct drm_clip_rect *clips,
+		     struct drm_vmw_rect *vclips,
+		     uint32_t num_clips,
+		     int increment,
+		     bool to_surface,
+		     bool interruptible);
+
 
 #endif
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
index 521f1947b4e9..69b471af0130 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
@@ -497,7 +497,7 @@ int vmw_user_dmabuf_alloc(struct vmw_private *dev_priv,
 
 	ret = vmw_dmabuf_init(dev_priv, &user_bo->dma, size,
 			      (dev_priv->has_mob) ?
-			      &vmw_mob_placement :
+			      &vmw_sys_placement :
 			      &vmw_vram_sys_placement, true,
 			      &vmw_user_dmabuf_destroy);
 	if (unlikely(ret != 0))
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c b/drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c
index becf9650c228..493fcd1eb803 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c
@@ -1,6 +1,6 @@
 /******************************************************************************
  *
- * Copyright © 2014 VMware, Inc., Palo Alto, CA., USA
+ * COPYRIGHT © 2014 VMware, Inc., Palo Alto, CA., USA
  * All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -44,6 +44,47 @@ enum stdu_content_type {
 	SEPARATE_DMA
 };
 
+/**
+ * struct vmw_stdu_dirty - closure structure for the update functions
+ *
+ * @base: The base type we derive from. Used by vmw_kms_helper_dirty().
+ * @transfer: Transfer direction for DMA command.
+ * @left: Left side of bounding box.
+ * @right: Right side of bounding box.
+ * @top: Top side of bounding box.
+ * @bottom: Bottom side of bounding box.
+ * @buf: DMA buffer when DMA-ing between buffer and screen targets.
+ * @sid: Surface ID when copying between surface and screen targets.
+ */
+struct vmw_stdu_dirty {
+	struct vmw_kms_dirty base;
+	SVGA3dTransferType  transfer;
+	s32 left, right, top, bottom;
+	u32 pitch;
+	union {
+		struct vmw_dma_buffer *buf;
+		u32 sid;
+	};
+};
+
+/*
+ * SVGA commands that are used by this code. Please see the device headers
+ * for explanation.
+ */
+struct vmw_stdu_update {
+	SVGA3dCmdHeader header;
+	SVGA3dCmdUpdateGBScreenTarget body;
+};
+
+struct vmw_stdu_dma {
+	SVGA3dCmdHeader     header;
+	SVGA3dCmdSurfaceDMA body;
+};
+
+struct vmw_stdu_surface_copy {
+	SVGA3dCmdHeader      header;
+	SVGA3dCmdSurfaceCopy body;
+};
 
 
 /**
@@ -139,254 +180,6 @@ static void vmw_stdu_crtc_destroy(struct drm_crtc *crtc)
 	vmw_stdu_destroy(vmw_crtc_to_stdu(crtc));
 }
 
-
-
-/**
- * vmw_stdu_dma_update - Update DMA buf dirty region on the SVGA device
- *
- * @dev_priv:  VMW DRM device
- * @file_priv: Pointer to a drm file private structure
- * @vfbs: VMW framebuffer surface that may need a DMA buf update
- * @x: top/left corner of the content area to blit from
- * @y: top/left corner of the content area to blit from
- * @width: width of the blit area
- * @height: height of the blit area
- *
- * The SVGA device may have the DMA buf cached, so before letting the
- * device use it as the source image for a subsequent operation, we
- * update the cached copy.
- *
- * RETURNs:
- * 0 on success, error code on failure
- */
-static int vmw_stdu_dma_update(struct vmw_private *dev_priv,
-			       struct drm_file *file_priv,
-			       struct vmw_framebuffer_surface *vfbs,
-			       uint32_t x, uint32_t y,
-			       uint32_t width, uint32_t height)
-{
-	size_t fifo_size;
-	struct {
-		SVGA3dCmdHeader header;
-		SVGA3dCmdUpdateGBImage body;
-	} img_update_cmd;
-
-
-	/* Only need to do this if the surface is a DMA buf proxy */
-	if (!vfbs->is_dmabuf_proxy)
-		return 0;
-
-	fifo_size = sizeof(img_update_cmd);
-
-	memset(&img_update_cmd, 0, fifo_size);
-	img_update_cmd.header.id   = SVGA_3D_CMD_UPDATE_GB_IMAGE;
-	img_update_cmd.header.size = sizeof(img_update_cmd.body);
-
-	img_update_cmd.body.image.sid = vfbs->surface->res.id;
-
-	img_update_cmd.body.box.x = x;
-	img_update_cmd.body.box.y = y;
-	img_update_cmd.body.box.w = width;
-	img_update_cmd.body.box.h = height;
-	img_update_cmd.body.box.d = 1;
-
-	return vmw_execbuf_process(file_priv, dev_priv, NULL,
-				   (void *) &img_update_cmd,
-				   fifo_size, 0, VMW_QUIRK_SRC_SID_OK,
-				   NULL, NULL);
-}
-
-
-
-/**
- * vmw_stdu_content_copy - copies an area from the content to display surface
- *
- * @dev_priv:  VMW DRM device
- * @file_priv: Pointer to a drm file private structure
- * @stdu: STDU whose display surface will be blitted to
- * @content_x: top/left corner of the content area to blit from
- * @content_y: top/left corner of the content area to blit from
- * @width: width of the blit area
- * @height: height of the blit area
- * @display_x: top/left corner of the display area to blit to
- * @display_y: top/left corner of the display area to blit to
- *
- * Copies an area from the content surface to the display surface.
- *
- * RETURNs:
- * 0 on success, error code on failure
- */
-static int vmw_stdu_content_copy(struct vmw_private *dev_priv,
-				 struct drm_file *file_priv,
-				 struct vmw_screen_target_display_unit *stdu,
-				 uint32_t content_x, uint32_t content_y,
-				 uint32_t width, uint32_t height,
-				 uint32_t display_x, uint32_t display_y)
-{
-	struct vmw_framebuffer_surface *content_vfbs;
-	size_t fifo_size;	
-	int ret;
-	void *cmd;
-	u32 quirks = VMW_QUIRK_DST_SID_OK;
-
-	struct {
-		SVGA3dCmdHeader     header;
-		SVGA3dCmdSurfaceDMA body;
-		SVGA3dCopyBox       area;
-		SVGA3dCmdSurfaceDMASuffix suffix;
-	} surface_dma_cmd;
-
-	struct {
-		SVGA3dCmdHeader      header;
-		SVGA3dCmdSurfaceCopy body;
-		SVGA3dCopyBox        area;
-	} surface_cpy_cmd;
-
-
-	/*
-	 * Can only copy if content and display surfaces exist and are not
-	 * the same surface
-	 */
-	if (stdu->display_srf == NULL || stdu->content_fb == NULL ||
-	    stdu->content_fb_type == SAME_AS_DISPLAY) {
-		return -EINVAL;
-	}
-
-
-	if (stdu->content_fb_type == SEPARATE_DMA) {
-		struct vmw_framebuffer *content_vfb;
-		struct drm_vmw_size cur_size = {0};
-		const struct svga3d_surface_desc *desc;
-		enum SVGA3dSurfaceFormat format;
-		SVGA3dCmdSurfaceDMASuffix *suffix;
-		SVGAGuestPtr ptr;
-
-
-		content_vfb  = vmw_framebuffer_to_vfb(stdu->content_fb);
-
-		cur_size.width  = width;
-		cur_size.height = height;
-		cur_size.depth  = 1;
-
-		/* Derive a SVGA3dSurfaceFormat for the DMA buf */
-		switch (content_vfb->base.bits_per_pixel) {
-		case 32:
-			format = SVGA3D_A8R8G8B8;
-			break;
-		case 24:
-			format = SVGA3D_X8R8G8B8;
-			break;
-		case 16:
-			format = SVGA3D_R5G6B5;
-			break;
-		case 15:
-			format = SVGA3D_A1R5G5B5;
-			break;
-		default:
-			DRM_ERROR("Invalid color depth: %d\n",
-					content_vfb->base.depth);
-			return -EINVAL;
-		}
-
-		desc = svga3dsurface_get_desc(format);
-
-
-		fifo_size = sizeof(surface_dma_cmd);
-
-		memset(&surface_dma_cmd, 0, fifo_size);
-
-		ptr.gmrId  = content_vfb->user_handle;
-		ptr.offset = 0;
-
-		surface_dma_cmd.header.id   = SVGA_3D_CMD_SURFACE_DMA;
-		surface_dma_cmd.header.size = sizeof(surface_dma_cmd.body) +
-					      sizeof(surface_dma_cmd.area) +
-					      sizeof(surface_dma_cmd.suffix);
-
-		surface_dma_cmd.body.guest.ptr   = ptr;
-		surface_dma_cmd.body.guest.pitch = stdu->content_fb->pitches[0];
-		surface_dma_cmd.body.host.sid    = stdu->display_srf->res.id;
-		surface_dma_cmd.body.host.face   = 0;
-		surface_dma_cmd.body.host.mipmap = 0;
-		surface_dma_cmd.body.transfer    = SVGA3D_WRITE_HOST_VRAM;
-
-		surface_dma_cmd.area.srcx = content_x;
-		surface_dma_cmd.area.srcy = content_y;
-		surface_dma_cmd.area.x    = display_x;
-		surface_dma_cmd.area.y    = display_y;
-		surface_dma_cmd.area.d    = 1;
-		surface_dma_cmd.area.w    = width;
-		surface_dma_cmd.area.h    = height;
-
-		suffix = &surface_dma_cmd.suffix;
-
-		suffix->suffixSize    = sizeof(*suffix);
-		suffix->maximumOffset = svga3dsurface_get_image_buffer_size(
-						desc,
-						&cur_size,
-						stdu->content_fb->pitches[0]);
-
-		cmd = (void *) &surface_dma_cmd;
-	} else {
-		u32 src_id;
-
-
-		content_vfbs = vmw_framebuffer_to_vfbs(stdu->content_fb);
-
-		if (content_vfbs->is_dmabuf_proxy) {
-			ret = vmw_stdu_dma_update(dev_priv, file_priv,
-						  content_vfbs,
-						  content_x, content_y,
-						  width, height);
-
-			if (ret != 0) {
-				DRM_ERROR("Failed to update cached DMA buf\n");
-				return ret;
-			}
-
-			quirks |= VMW_QUIRK_SRC_SID_OK;
-			src_id = content_vfbs->surface->res.id;
-		} else {
-			struct vmw_framebuffer *content_vfb;
-
-			content_vfb = vmw_framebuffer_to_vfb(stdu->content_fb);
-			src_id = content_vfb->user_handle;
-		}
- 
-		fifo_size = sizeof(surface_cpy_cmd);
-
-		memset(&surface_cpy_cmd, 0, fifo_size);
-
-		surface_cpy_cmd.header.id   = SVGA_3D_CMD_SURFACE_COPY;
-		surface_cpy_cmd.header.size = sizeof(surface_cpy_cmd.body) +
-					      sizeof(surface_cpy_cmd.area);
-
-		surface_cpy_cmd.body.src.sid  = src_id;
-		surface_cpy_cmd.body.dest.sid = stdu->display_srf->res.id;
-
-		surface_cpy_cmd.area.srcx = content_x;
-		surface_cpy_cmd.area.srcy = content_y;
-		surface_cpy_cmd.area.x    = display_x;
-		surface_cpy_cmd.area.y    = display_y;
-		surface_cpy_cmd.area.d    = 1;
-		surface_cpy_cmd.area.w    = width;
-		surface_cpy_cmd.area.h    = height;
-
-		cmd = (void *) &surface_cpy_cmd;
-	}
-
-
-
-	ret = vmw_execbuf_process(file_priv, dev_priv, NULL,
-				  (void *) cmd,
-				  fifo_size, 0, quirks,
-				  NULL, NULL);
-
-	return ret;
-}
-
-
-
 /**
  * vmw_stdu_define_st - Defines a Screen Target
  *
@@ -487,108 +280,56 @@ static int vmw_stdu_bind_st(struct vmw_private *dev_priv,
 	return 0;
 }
 
+/**
+ * vmw_stdu_populate_update - populate an UPDATE_GB_SCREENTARGET command with a
+ * bounding box.
+ *
+ * @cmd: Pointer to command stream.
+ * @unit: Screen target unit.
+ * @left: Left side of bounding box.
+ * @right: Right side of bounding box.
+ * @top: Top side of bounding box.
+ * @bottom: Bottom side of bounding box.
+ */
+static void vmw_stdu_populate_update(void *cmd, int unit,
+				     s32 left, s32 right, s32 top, s32 bottom)
+{
+	struct vmw_stdu_update *update = cmd;
+
+	update->header.id   = SVGA_3D_CMD_UPDATE_GB_SCREENTARGET;
+	update->header.size = sizeof(update->body);
 
+	update->body.stid   = unit;
+	update->body.rect.x = left;
+	update->body.rect.y = top;
+	update->body.rect.w = right - left;
+	update->body.rect.h = bottom - top;
+}
 
 /**
- * vmw_stdu_update_st - Updates a Screen Target
+ * vmw_stdu_update_st - Full update of a Screen Target
  *
  * @dev_priv: VMW DRM device
- * @file_priv: Pointer to DRM file private structure.  Set to NULL when
- *             we want to blank display.
  * @stdu: display unit affected
- * @update_area: area that needs to be updated
  *
  * This function needs to be called whenever the content of a screen
- * target changes.
- * If the display and content buffers are different, then this function does
- * a blit first from the content buffer to the display buffer before issuing
- * the Screen Target update command.
+ * target has changed completely. Typically as a result of a backing
+ * surface change.
  *
  * RETURNS:
  * 0 on success, error code on failure
  */
 static int vmw_stdu_update_st(struct vmw_private *dev_priv,
-			      struct drm_file *file_priv,
-			      struct vmw_screen_target_display_unit *stdu,
-			      struct drm_clip_rect *update_area)
+			      struct vmw_screen_target_display_unit *stdu)
 {
-	u32 width, height;
-	u32 display_update_x, display_update_y;
-	unsigned short display_x1, display_y1, display_x2, display_y2;
-	int ret;
-
-	struct {
-		SVGA3dCmdHeader header;
-		SVGA3dCmdUpdateGBScreenTarget body;
-	} *cmd;
-
+	struct vmw_stdu_update *cmd;
+	struct drm_crtc *crtc = &stdu->base.crtc;
 
 	if (!stdu->defined) {
 		DRM_ERROR("No screen target defined");
 		return -EINVAL;
 	}
 
-	/* Display coordinates relative to its position in content surface */
-	display_x1 = stdu->base.crtc.x;
-	display_y1 = stdu->base.crtc.y;
-	display_x2 = display_x1 + stdu->display_srf->base_size.width;
-	display_y2 = display_y1 + stdu->display_srf->base_size.height;
-
-	/* Do nothing if the update area is outside of the display surface */
-	if (update_area->x2 <= display_x1 || update_area->x1 >= display_x2 ||
-	    update_area->y2 <= display_y1 || update_area->y1 >= display_y2)
-		return 0;
-
-	/* The top-left hand corner of the update area in display surface */
-	display_update_x = max(update_area->x1 - display_x1, 0);
-	display_update_y = max(update_area->y1 - display_y1, 0);
-
-	width  = min(update_area->x2, display_x2) -
-		 max(update_area->x1, display_x1);
-	height = min(update_area->y2, display_y2) -
-		 max(update_area->y1, display_y1);
-
-	/*
-	 * If content is on a separate surface, then copy the dirty area to
-	 * the display surface
-	 */
-	if (file_priv && stdu->content_fb_type != SAME_AS_DISPLAY) {
-
-		ret = vmw_stdu_content_copy(dev_priv, file_priv,
-					    stdu,
-					    max(update_area->x1, display_x1),
-					    max(update_area->y1, display_y1),
-					    width, height,
-					    display_update_x, display_update_y);
-		if (unlikely(ret != 0)) {
-			DRM_ERROR("Failed to blit content\n");
-			return ret;
-		}
-	}
-
-
-	/*
-	 * If the display surface is the same as the content surface, then
-	 * it may be backed by a DMA buf.  If it is then we need to update
-	 * the device's cached copy of the DMA buf before issuing the screen
-	 * target update.
-	 */
-	if (file_priv && stdu->content_fb_type == SAME_AS_DISPLAY) {
-		struct vmw_framebuffer_surface *vfbs;
-
-		vfbs = vmw_framebuffer_to_vfbs(stdu->content_fb);
-		ret = vmw_stdu_dma_update(dev_priv, file_priv,
-					  vfbs,
-					  max(update_area->x1, display_x1),
-					  max(update_area->y1, display_y1),
-					  width, height);
-
-		if (ret != 0) {
-			DRM_ERROR("Failed to update cached DMA buffer\n");
-			return ret;
-		}
-	}
-
 	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
 
 	if (unlikely(cmd == NULL)) {
@@ -596,14 +337,8 @@ static int vmw_stdu_update_st(struct vmw_private *dev_priv,
 		return -ENOMEM;
 	}
 
-	cmd->header.id   = SVGA_3D_CMD_UPDATE_GB_SCREENTARGET;
-	cmd->header.size = sizeof(cmd->body);
-
-	cmd->body.stid   = stdu->base.unit;
-	cmd->body.rect.x = display_update_x;
-	cmd->body.rect.y = display_update_y;
-	cmd->body.rect.w = width;
-	cmd->body.rect.h = height;
+	vmw_stdu_populate_update(cmd, stdu->base.unit, 0, crtc->mode.hdisplay,
+				 0, crtc->mode.vdisplay);
 
 	vmw_fifo_commit(dev_priv, sizeof(*cmd));
 
@@ -682,7 +417,6 @@ static int vmw_stdu_crtc_set_config(struct drm_mode_set *set)
 	struct drm_crtc      *crtc;
 	struct drm_encoder   *encoder;
 	struct drm_connector *connector;
-	struct drm_clip_rect update_area = {0};
 	int    ret;
 
 
@@ -728,12 +462,7 @@ static int vmw_stdu_crtc_set_config(struct drm_mode_set *set)
 
 		/* Update Screen Target, display will now be blank */
 		if (crtc->primary->fb) {
-			update_area.x2 = crtc->primary->fb->width;
-			update_area.y2 = crtc->primary->fb->height;
-
-			ret = vmw_stdu_update_st(dev_priv, NULL,
-						 stdu,
-						 &update_area);
+			vmw_stdu_update_st(dev_priv, stdu);
 			if (unlikely(ret != 0))
 				return ret;
 		}
@@ -852,7 +581,6 @@ static int vmw_stdu_crtc_set_config(struct drm_mode_set *set)
 		goto err_unref_content;
 	}
 
-	vmw_fb_off(dev_priv);
 	vmw_svga_enable(dev_priv);
 
 	/*
@@ -919,17 +647,8 @@ static int vmw_stdu_crtc_page_flip(struct drm_crtc *crtc,
 {
 	struct vmw_private *dev_priv = vmw_priv(crtc->dev);
 	struct vmw_screen_target_display_unit *stdu;
-	struct drm_file *file_priv;
-	struct drm_clip_rect update_area = {0};
 	int ret;
 
-	/*
-	 * Temporarily don't support event == NULL. We need the
-	 * @file_priv pointer!
-	 */
-	if (event == NULL)
-		return -EINVAL;
-
 	if (crtc == NULL)
 		return -EINVAL;
 
@@ -939,9 +658,6 @@ static int vmw_stdu_crtc_page_flip(struct drm_crtc *crtc,
 	stdu->content_fb  = new_fb;
 
 	if (stdu->display_srf) {
-		update_area.x2 = stdu->display_srf->base_size.width;
-		update_area.y2 = stdu->display_srf->base_size.height;
-
 		/*
 		 * If the display surface is the same as the content surface
 		 * then remove the reference
@@ -961,7 +677,7 @@ static int vmw_stdu_crtc_page_flip(struct drm_crtc *crtc,
 
 	if (!new_fb) {
 		/* Blanks the display */
-		(void) vmw_stdu_update_st(dev_priv, NULL, stdu, &update_area);
+		(void) vmw_stdu_update_st(dev_priv, stdu);
 
 		return 0;
 	}
@@ -982,16 +698,13 @@ static int vmw_stdu_crtc_page_flip(struct drm_crtc *crtc,
 	}
 
 	/* Update display surface: after this point everything is bound */
-	update_area.x2 = stdu->display_srf->base_size.width;
-	update_area.y2 = stdu->display_srf->base_size.height;
-
-	file_priv = event->base.file_priv;
-	ret = vmw_stdu_update_st(dev_priv, file_priv, stdu, &update_area);
+	ret = vmw_stdu_update_st(dev_priv, stdu);
 	if (unlikely(ret != 0))
 		return ret;
 
 	if (event) {
 		struct vmw_fence_obj *fence = NULL;
+		struct drm_file *file_priv = event->base.file_priv;
 
 		vmw_execbuf_fence_commands(NULL, dev_priv, &fence, NULL);
 		if (!fence)
@@ -1016,6 +729,310 @@ err_out:
 }
 
 
+/**
+ * vmw_stdu_dmabuf_clip - Callback to encode a suface DMA command cliprect
+ *
+ * @dirty: The closure structure.
+ *
+ * Encodes a surface DMA command cliprect and updates the bounding box
+ * for the DMA.
+ */
+static void vmw_stdu_dmabuf_clip(struct vmw_kms_dirty *dirty)
+{
+	struct vmw_stdu_dirty *ddirty =
+		container_of(dirty, struct vmw_stdu_dirty, base);
+	struct vmw_stdu_dma *cmd = dirty->cmd;
+	struct SVGA3dCopyBox *blit = (struct SVGA3dCopyBox *) &cmd[1];
+
+	blit += dirty->num_hits;
+	blit->srcx = dirty->fb_x;
+	blit->srcy = dirty->fb_y;
+	blit->x = dirty->unit_x1;
+	blit->y = dirty->unit_y1;
+	blit->d = 1;
+	blit->w = dirty->unit_x2 - dirty->unit_x1;
+	blit->h = dirty->unit_y2 - dirty->unit_y1;
+	dirty->num_hits++;
+
+	if (ddirty->transfer != SVGA3D_WRITE_HOST_VRAM)
+		return;
+
+	/* Destination bounding box */
+	ddirty->left = min_t(s32, ddirty->left, dirty->unit_x1);
+	ddirty->top = min_t(s32, ddirty->top, dirty->unit_y1);
+	ddirty->right = max_t(s32, ddirty->right, dirty->unit_x2);
+	ddirty->bottom = max_t(s32, ddirty->bottom, dirty->unit_y2);
+}
+
+/**
+ * vmw_stdu_dmabuf_fifo_commit - Callback to fill in and submit a DMA command.
+ *
+ * @dirty: The closure structure.
+ *
+ * Fills in the missing fields in a DMA command, and optionally encodes
+ * a screen target update command, depending on transfer direction.
+ */
+static void vmw_stdu_dmabuf_fifo_commit(struct vmw_kms_dirty *dirty)
+{
+	struct vmw_stdu_dirty *ddirty =
+		container_of(dirty, struct vmw_stdu_dirty, base);
+	struct vmw_screen_target_display_unit *stdu =
+		container_of(dirty->unit, typeof(*stdu), base);
+	struct vmw_stdu_dma *cmd = dirty->cmd;
+	struct SVGA3dCopyBox *blit = (struct SVGA3dCopyBox *) &cmd[1];
+	SVGA3dCmdSurfaceDMASuffix *suffix =
+		(SVGA3dCmdSurfaceDMASuffix *) &blit[dirty->num_hits];
+	size_t blit_size = sizeof(*blit) * dirty->num_hits + sizeof(*suffix);
+
+	if (!dirty->num_hits) {
+		vmw_fifo_commit(dirty->dev_priv, 0);
+		return;
+	}
+
+	cmd->header.id = SVGA_3D_CMD_SURFACE_DMA;
+	cmd->header.size = size