[PATCH 00/28] vmwgfx preparation for GL3

[PATCH 00/28] vmwgfx preparation for GL3

[Thomas Hellstrom]

This patch series does some reorganization and preparation for the upcoming
GL3 functionality.

* Reworks device initialization
* Adds command buffer support, replaces the old FIFO ring
* Adds screen target support, a new interface to assign "scanouts"
* Adds a simple implementation of fbdev on kms. At some point we should
  however consider moving the KMS code to use the helpers.

I've already sent out a pull request for a part of this patchset.


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 01/28] drm/vmwgfx: Fix an fb unlocking bug

[Thomas Hellstrom]

A regression introduced when the master ttm lock was split into two.

Reported-and-tested-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_fb.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
index 0a474f3..e2d40eb 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
@@ -402,12 +402,12 @@ static int vmw_fb_create_bo(struct vmw_private *vmw_priv,
 
 	*out = vmw_bo;
 
-	ttm_write_unlock(&vmw_priv->fbdev_master.lock);
+	ttm_write_unlock(&vmw_priv->reservation_sem);
 
 	return 0;
 
 err_unlock:
-	ttm_write_unlock(&vmw_priv->fbdev_master.lock);
+	ttm_write_unlock(&vmw_priv->reservation_sem);
 	return ret;
 }
 
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 02/28] vmwgfx: Rework device initialization

[Thomas Hellstrom]

This commit reworks device initialization so that we always enable the
FIFO at driver load, deferring SVGA enable until either first modeset
or fbdev enable.
This should always leave the fifo properly enabled for render- and
control nodes.
In addition,
*) We disable the use of VRAM when SVGA is not enabled.
*) We simplify PM support so that we only throw out resources on hibernate,
not on suspend, since the device keeps its state on suspend.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_context.c |   8 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.c     | 337 ++++++++++++++++++--------------
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.h     |  19 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_fb.c      |   4 +
 drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c    |  12 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c     |   1 +
 drivers/gpu/drm/vmwgfx/vmwgfx_mob.c     |   6 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c    |   1 +
 drivers/gpu/drm/vmwgfx/vmwgfx_shader.c  |   4 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_surface.c |  12 +-
 10 files changed, 230 insertions(+), 174 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_context.c b/drivers/gpu/drm/vmwgfx/vmwgfx_context.c
index 5ac9287..a8e370a 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_context.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_context.c
@@ -140,7 +140,7 @@ static void vmw_hw_context_destroy(struct vmw_resource *res)
 	cmd->body.cid = cpu_to_le32(res->id);
 
 	vmw_fifo_commit(dev_priv, sizeof(*cmd));
-	vmw_3d_resource_dec(dev_priv, false);
+	vmw_fifo_resource_dec(dev_priv);
 }
 
 static int vmw_gb_context_init(struct vmw_private *dev_priv,
@@ -220,7 +220,7 @@ static int vmw_context_init(struct vmw_private *dev_priv,
 	cmd->body.cid = cpu_to_le32(res->id);
 
 	vmw_fifo_commit(dev_priv, sizeof(*cmd));
-	(void) vmw_3d_resource_inc(dev_priv, false);
+	vmw_fifo_resource_inc(dev_priv);
 	vmw_resource_activate(res, vmw_hw_context_destroy);
 	return 0;
 
@@ -281,7 +281,7 @@ static int vmw_gb_context_create(struct vmw_resource *res)
 	cmd->header.size = sizeof(cmd->body);
 	cmd->body.cid = res->id;
 	vmw_fifo_commit(dev_priv, sizeof(*cmd));
-	(void) vmw_3d_resource_inc(dev_priv, false);
+	vmw_fifo_resource_inc(dev_priv);
 
 	return 0;
 
@@ -414,7 +414,7 @@ static int vmw_gb_context_destroy(struct vmw_resource *res)
 	if (dev_priv->query_cid == res->id)
 		dev_priv->query_cid_valid = false;
 	vmw_resource_release_id(res);
-	vmw_3d_resource_dec(dev_priv, false);
+	vmw_fifo_resource_dec(dev_priv);
 
 	return 0;
 }
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
index 620bb5c..a4766ac 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
@@ -339,24 +339,47 @@ static int vmw_dummy_query_bo_create(struct vmw_private *dev_priv)
 	return ret;
 }
 
-static int vmw_request_device(struct vmw_private *dev_priv)
+/**
+ * vmw_request_device_late - Perform late device setup
+ *
+ * @dev_priv: Pointer to device private.
+ *
+ * This function performs setup of otables and enables large command
+ * buffer submission. These tasks are split out to a separate function
+ * because it reverts vmw_release_device_early and is intended to be used
+ * by an error path in the hibernation code.
+ */
+static int vmw_request_device_late(struct vmw_private *dev_priv)
 {
 	int ret;
 
-	ret = vmw_fifo_init(dev_priv, &dev_priv->fifo);
-	if (unlikely(ret != 0)) {
-		DRM_ERROR("Unable to initialize FIFO.\n");
-		return ret;
-	}
-	vmw_fence_fifo_up(dev_priv->fman);
 	if (dev_priv->has_mob) {
 		ret = vmw_otables_setup(dev_priv);
 		if (unlikely(ret != 0)) {
 			DRM_ERROR("Unable to initialize "
 				  "guest Memory OBjects.\n");
-			goto out_no_mob;
+			return ret;
 		}
 	}
+
+	return 0;
+}
+
+static int vmw_request_device(struct vmw_private *dev_priv)
+{
+	int ret;
+
+	ret = vmw_fifo_init(dev_priv, &dev_priv->fifo);
+	if (unlikely(ret != 0)) {
+		DRM_ERROR("Unable to initialize FIFO.\n");
+		return ret;
+	}
+	vmw_fence_fifo_up(dev_priv->fman);
+
+	ret = vmw_request_device_late(dev_priv);
+	if (ret)
+		goto out_no_mob;
+
 	ret = vmw_dummy_query_bo_create(dev_priv);
 	if (unlikely(ret != 0))
 		goto out_no_query_bo;
@@ -364,15 +387,25 @@ static int vmw_request_device(struct vmw_private *dev_priv)
 	return 0;
 
 out_no_query_bo:
-	if (dev_priv->has_mob)
+	if (dev_priv->has_mob) {
+		(void) ttm_bo_evict_mm(&dev_priv->bdev, VMW_PL_MOB);
 		vmw_otables_takedown(dev_priv);
+	}
 out_no_mob:
 	vmw_fence_fifo_down(dev_priv->fman);
 	vmw_fifo_release(dev_priv, &dev_priv->fifo);
 	return ret;
 }
 
-static void vmw_release_device(struct vmw_private *dev_priv)
+/**
+ * vmw_release_device_early - Early part of fifo takedown.
+ *
+ * @dev_priv: Pointer to device private struct.
+ *
+ * This is the first part of command submission takedown, to be called before
+ * buffer management is taken down.
+ */
+static void vmw_release_device_early(struct vmw_private *dev_priv)
 {
 	/*
 	 * Previous destructions should've released
@@ -382,64 +415,24 @@ static void vmw_release_device(struct vmw_private *dev_priv)
 	BUG_ON(dev_priv->pinned_bo != NULL);
 
 	ttm_bo_unref(&dev_priv->dummy_query_bo);
-	if (dev_priv->has_mob)
+	if (dev_priv->has_mob) {
+		ttm_bo_evict_mm(&dev_priv->bdev, VMW_PL_MOB);
 		vmw_otables_takedown(dev_priv);
-	vmw_fence_fifo_down(dev_priv->fman);
-	vmw_fifo_release(dev_priv, &dev_priv->fifo);
-}
-
-
-/**
- * Increase the 3d resource refcount.
- * If the count was prevously zero, initialize the fifo, switching to svga
- * mode. Note that the master holds a ref as well, and may request an
- * explicit switch to svga mode if fb is not running, using @unhide_svga.
- */
-int vmw_3d_resource_inc(struct vmw_private *dev_priv,
-			bool unhide_svga)
-{
-	int ret = 0;
-
-	mutex_lock(&dev_priv->release_mutex);
-	if (unlikely(dev_priv->num_3d_resources++ == 0)) {
-		ret = vmw_request_device(dev_priv);
-		if (unlikely(ret != 0))
-			--dev_priv->num_3d_resources;
-	} else if (unhide_svga) {
-		vmw_write(dev_priv, SVGA_REG_ENABLE,
-			  vmw_read(dev_priv, SVGA_REG_ENABLE) &
-			  ~SVGA_REG_ENABLE_HIDE);
 	}
-
-	mutex_unlock(&dev_priv->release_mutex);
-	return ret;
 }
 
 /**
- * Decrease the 3d resource refcount.
- * If the count reaches zero, disable the fifo, switching to vga mode.
- * Note that the master holds a refcount as well, and may request an
- * explicit switch to vga mode when it releases its refcount to account
- * for the situation of an X server vt switch to VGA with 3d resources
- * active.
+ * vmw_release_device_late - Late part of fifo takedown.
+ *
+ * @dev_priv: Pointer to device private struct.
+ *
+ * This is the last part of the command submission takedown, to be called when
+ * command submission is no longer needed. It may wait on pending fences.
  */
-void vmw_3d_resource_dec(struct vmw_private *dev_priv,
-			 bool hide_svga)
+static void vmw_release_device_late(struct vmw_private *dev_priv)
 {
-	int32_t n3d;
-
-	mutex_lock(&dev_priv->release_mutex);
-	if (unlikely(--dev_priv->num_3d_resources == 0))
-		vmw_release_device(dev_priv);
-	else if (hide_svga)
-		vmw_write(dev_priv, SVGA_REG_ENABLE,
-			  vmw_read(dev_priv, SVGA_REG_ENABLE) |
-			  SVGA_REG_ENABLE_HIDE);
-
-	n3d = (int32_t) dev_priv->num_3d_resources;
-	mutex_unlock(&dev_priv->release_mutex);
-
-	BUG_ON(n3d < 0);
+	vmw_fence_fifo_down(dev_priv->fman);
+	vmw_fifo_release(dev_priv, &dev_priv->fifo);
 }
 
 /**
@@ -603,6 +596,7 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 	spin_lock_init(&dev_priv->hw_lock);
 	spin_lock_init(&dev_priv->waiter_lock);
 	spin_lock_init(&dev_priv->cap_lock);
+	spin_lock_init(&dev_priv->svga_lock);
 
 	for (i = vmw_res_context; i < vmw_res_max; ++i) {
 		idr_init(&dev_priv->res_idr[i]);
@@ -714,17 +708,6 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 	dev_priv->active_master = &dev_priv->fbdev_master;
 
 
-	ret = ttm_bo_device_init(&dev_priv->bdev,
-				 dev_priv->bo_global_ref.ref.object,
-				 &vmw_bo_driver,
-				 dev->anon_inode->i_mapping,
-				 VMWGFX_FILE_PAGE_OFFSET,
-				 false);
-	if (unlikely(ret != 0)) {
-		DRM_ERROR("Failed initializing TTM buffer object driver.\n");
-		goto out_err1;
-	}
-
 	dev_priv->mmio_mtrr = arch_phys_wc_add(dev_priv->mmio_start,
 					       dev_priv->mmio_size);
 
@@ -787,13 +770,28 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 		goto out_no_fman;
 	}
 
+	ret = ttm_bo_device_init(&dev_priv->bdev,
+				 dev_priv->bo_global_ref.ref.object,
+				 &vmw_bo_driver,
+				 dev->anon_inode->i_mapping,
+				 VMWGFX_FILE_PAGE_OFFSET,
+				 false);
+	if (unlikely(ret != 0)) {
+		DRM_ERROR("Failed initializing TTM buffer object driver.\n");
+		goto out_no_bdev;
+	}
 
+	/*
+	 * Enable VRAM, but initially don't use it until SVGA is enabled and
+	 * unhidden.
+	 */
 	ret = ttm_bo_init_mm(&dev_priv->bdev, TTM_PL_VRAM,
 			     (dev_priv->vram_size >> PAGE_SHIFT));
 	if (unlikely(ret != 0)) {
 		DRM_ERROR("Failed initializing memory manager for VRAM.\n");
 		goto out_no_vram;
 	}
+	dev_priv->bdev.man[TTM_PL_VRAM].use_type = false;
 
 	dev_priv->has_gmr = true;
 	if (((dev_priv->capabilities & (SVGA_CAP_GMR | SVGA_CAP_GMR2)) == 0) ||
@@ -814,18 +812,18 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 		}
 	}
 
-	vmw_kms_save_vga(dev_priv);
-
-	/* Start kms and overlay systems, needs fifo. */
 	ret = vmw_kms_init(dev_priv);
 	if (unlikely(ret != 0))
 		goto out_no_kms;
 	vmw_overlay_init(dev_priv);
 
+	ret = vmw_request_device(dev_priv);
+	if (ret)
+		goto out_no_fifo;
+
 	if (dev_priv->enable_fb) {
-		ret = vmw_3d_resource_inc(dev_priv, true);
-		if (unlikely(ret != 0))
-			goto out_no_fifo;
+		vmw_fifo_resource_inc(dev_priv);
+		vmw_svga_enable(dev_priv);
 		vmw_fb_init(dev_priv);
 	}
 
@@ -838,13 +836,14 @@ out_no_fifo:
 	vmw_overlay_close(dev_priv);
 	vmw_kms_close(dev_priv);
 out_no_kms:
-	vmw_kms_restore_vga(dev_priv);
 	if (dev_priv->has_mob)
 		(void) ttm_bo_clean_mm(&dev_priv->bdev, VMW_PL_MOB);
 	if (dev_priv->has_gmr)
 		(void) ttm_bo_clean_mm(&dev_priv->bdev, VMW_PL_GMR);
 	(void)ttm_bo_clean_mm(&dev_priv->bdev, TTM_PL_VRAM);
 out_no_vram:
+	(void)ttm_bo_device_release(&dev_priv->bdev);
+out_no_bdev:
 	vmw_fence_manager_takedown(dev_priv->fman);
 out_no_fman:
 	if (dev_priv->capabilities & SVGA_CAP_IRQMASK)
@@ -860,8 +859,6 @@ out_err4:
 	iounmap(dev_priv->mmio_virt);
 out_err3:
 	arch_phys_wc_del(dev_priv->mmio_mtrr);
-	(void)ttm_bo_device_release(&dev_priv->bdev);
-out_err1:
 	vmw_ttm_global_release(dev_priv);
 out_err0:
 	for (i = vmw_res_context; i < vmw_res_max; ++i)
@@ -883,18 +880,22 @@ static int vmw_driver_unload(struct drm_device *dev)
 	vfree(dev_priv->ctx.cmd_bounce);
 	if (dev_priv->enable_fb) {
 		vmw_fb_close(dev_priv);
-		vmw_kms_restore_vga(dev_priv);
-		vmw_3d_resource_dec(dev_priv, false);
+		vmw_fifo_resource_dec(dev_priv);
+		vmw_svga_disable(dev_priv);
 	}
+
 	vmw_kms_close(dev_priv);
 	vmw_overlay_close(dev_priv);
 
-	if (dev_priv->has_mob)
-		(void) ttm_bo_clean_mm(&dev_priv->bdev, VMW_PL_MOB);
 	if (dev_priv->has_gmr)
 		(void)ttm_bo_clean_mm(&dev_priv->bdev, VMW_PL_GMR);
 	(void)ttm_bo_clean_mm(&dev_priv->bdev, TTM_PL_VRAM);
 
+	vmw_release_device_early(dev_priv);
+	if (dev_priv->has_mob)
+		(void) ttm_bo_clean_mm(&dev_priv->bdev, VMW_PL_MOB);
+	(void) ttm_bo_device_release(&dev_priv->bdev);
+	vmw_release_device_late(dev_priv);
 	vmw_fence_manager_takedown(dev_priv->fman);
 	if (dev_priv->capabilities & SVGA_CAP_IRQMASK)
 		drm_irq_uninstall(dev_priv->dev);
@@ -1148,27 +1149,13 @@ static int vmw_master_set(struct drm_device *dev,
 	struct vmw_master *vmaster = vmw_master(file_priv->master);
 	int ret = 0;
 
-	if (!dev_priv->enable_fb) {
-		ret = vmw_3d_resource_inc(dev_priv, true);
-		if (unlikely(ret != 0))
-			return ret;
-		vmw_kms_save_vga(dev_priv);
-		vmw_write(dev_priv, SVGA_REG_TRACES, 0);
-	}
-
 	if (active) {
 		BUG_ON(active != &dev_priv->fbdev_master);
 		ret = ttm_vt_lock(&active->lock, false, vmw_fp->tfile);
 		if (unlikely(ret != 0))
-			goto out_no_active_lock;
+			return ret;
 
 		ttm_lock_set_kill(&active->lock, true, SIGTERM);
-		ret = ttm_bo_evict_mm(&dev_priv->bdev, TTM_PL_VRAM);
-		if (unlikely(ret != 0)) {
-			DRM_ERROR("Unable to clean VRAM on "
-				  "master drop.\n");
-		}
-
 		dev_priv->active_master = NULL;
 	}
 
@@ -1182,14 +1169,6 @@ static int vmw_master_set(struct drm_device *dev,
 	dev_priv->active_master = vmaster;
 
 	return 0;
-
-out_no_active_lock:
-	if (!dev_priv->enable_fb) {
-		vmw_kms_restore_vga(dev_priv);
-		vmw_3d_resource_dec(dev_priv, true);
-		vmw_write(dev_priv, SVGA_REG_TRACES, 1);
-	}
-	return ret;
 }
 
 static void vmw_master_drop(struct drm_device *dev,
@@ -1214,16 +1193,9 @@ static void vmw_master_drop(struct drm_device *dev,
 	}
 
 	ttm_lock_set_kill(&vmaster->lock, false, SIGTERM);
-	vmw_execbuf_release_pinned_bo(dev_priv);
 
-	if (!dev_priv->enable_fb) {
-		ret = ttm_bo_evict_mm(&dev_priv->bdev, TTM_PL_VRAM);
-		if (unlikely(ret != 0))
-			DRM_ERROR("Unable to clean VRAM on master drop.\n");
-		vmw_kms_restore_vga(dev_priv);
-		vmw_3d_resource_dec(dev_priv, true);
-		vmw_write(dev_priv, SVGA_REG_TRACES, 1);
-	}
+	if (!dev_priv->enable_fb)
+		vmw_svga_disable(dev_priv);
 
 	dev_priv->active_master = &dev_priv->fbdev_master;
 	ttm_lock_set_kill(&dev_priv->fbdev_master.lock, false, SIGTERM);
@@ -1233,6 +1205,74 @@ static void vmw_master_drop(struct drm_device *dev,
 		vmw_fb_on(dev_priv);
 }
 
+/**
+ * __vmw_svga_enable - Enable SVGA mode, FIFO and use of VRAM.
+ *
+ * @dev_priv: Pointer to device private struct.
+ * Needs the reservation sem to be held in non-exclusive mode.
+ */
+void __vmw_svga_enable(struct vmw_private *dev_priv)
+{
+	spin_lock(&dev_priv->svga_lock);
+	if (!dev_priv->bdev.man[TTM_PL_VRAM].use_type) {
+		vmw_write(dev_priv, SVGA_REG_ENABLE, SVGA_REG_ENABLE);
+		dev_priv->bdev.man[TTM_PL_VRAM].use_type = true;
+	}
+	spin_unlock(&dev_priv->svga_lock);
+}
+
+/**
+ * vmw_svga_enable - Enable SVGA mode, FIFO and use of VRAM.
+ *
+ * @dev_priv: Pointer to device private struct.
+ */
+void vmw_svga_enable(struct vmw_private *dev_priv)
+{
+	ttm_read_lock(&dev_priv->reservation_sem, false);
+	__vmw_svga_enable(dev_priv);
+	ttm_read_unlock(&dev_priv->reservation_sem);
+}
+
+/**
+ * __vmw_svga_disable - Disable SVGA mode and use of VRAM.
+ *
+ * @dev_priv: Pointer to device private struct.
+ * Needs the reservation sem to be held in exclusive mode.
+ * Will not empty VRAM. VRAM must be emptied by caller.
+ */
+void __vmw_svga_disable(struct vmw_private *dev_priv)
+{
+	spin_lock(&dev_priv->svga_lock);
+	if (dev_priv->bdev.man[TTM_PL_VRAM].use_type) {
+		dev_priv->bdev.man[TTM_PL_VRAM].use_type = false;
+		vmw_write(dev_priv, SVGA_REG_ENABLE,
+			  SVGA_REG_ENABLE_ENABLE_HIDE);
+	}
+	spin_unlock(&dev_priv->svga_lock);
+}
+
+/**
+ * vmw_svga_disable - Disable SVGA_MODE, and use of VRAM. Keep the fifo
+ * running.
+ *
+ * @dev_priv: Pointer to device private struct.
+ * Will empty VRAM.
+ */
+void vmw_svga_disable(struct vmw_private *dev_priv)
+{
+	ttm_write_lock(&dev_priv->reservation_sem, false);
+	spin_lock(&dev_priv->svga_lock);
+	if (dev_priv->bdev.man[TTM_PL_VRAM].use_type) {
+		dev_priv->bdev.man[TTM_PL_VRAM].use_type = false;
+		vmw_write(dev_priv, SVGA_REG_ENABLE,
+			  SVGA_REG_ENABLE_ENABLE_HIDE);
+		spin_unlock(&dev_priv->svga_lock);
+		if (ttm_bo_evict_mm(&dev_priv->bdev, TTM_PL_VRAM))
+			DRM_ERROR("Failed evicting VRAM buffers.\n");
+	} else
+		spin_unlock(&dev_priv->svga_lock);
+	ttm_write_unlock(&dev_priv->reservation_sem);
+}
 
 static void vmw_remove(struct pci_dev *pdev)
 {
@@ -1250,21 +1290,21 @@ static int vmwgfx_pm_notifier(struct notifier_block *nb, unsigned long val,
 
 	switch (val) {
 	case PM_HIBERNATION_PREPARE:
-	case PM_SUSPEND_PREPARE:
 		ttm_suspend_lock(&dev_priv->reservation_sem);
 
-		/**
+		/*
 		 * This empties VRAM and unbinds all GMR bindings.
 		 * Buffer contents is moved to swappable memory.
 		 */
 		vmw_execbuf_release_pinned_bo(dev_priv);
 		vmw_resource_evict_all(dev_priv);
+		vmw_release_device_early(dev_priv);
 		ttm_bo_swapout_all(&dev_priv->bdev);
-
+		vmw_fence_fifo_down(dev_priv->fman);
 		break;
 	case PM_POST_HIBERNATION:
-	case PM_POST_SUSPEND:
 	case PM_POST_RESTORE:
+		vmw_fence_fifo_up(dev_priv->fman);
 		ttm_suspend_unlock(&dev_priv->reservation_sem);
 
 		break;
@@ -1276,20 +1316,13 @@ static int vmwgfx_pm_notifier(struct notifier_block *nb, unsigned long val,
 	return 0;
 }
 
-/**
- * These might not be needed with the virtual SVGA device.
- */
-
 static int vmw_pci_suspend(struct pci_dev *pdev, pm_message_t state)
 {
 	struct drm_device *dev = pci_get_drvdata(pdev);
 	struct vmw_private *dev_priv = vmw_priv(dev);
 
-	if (dev_priv->num_3d_resources != 0) {
-		DRM_INFO("Can't suspend or hibernate "
-			 "while 3D resources are active.\n");
+	if (dev_priv->refuse_hibernation)
 		return -EBUSY;
-	}
 
 	pci_save_state(pdev);
 	pci_disable_device(pdev);
@@ -1321,56 +1354,62 @@ static int vmw_pm_resume(struct device *kdev)
 	return vmw_pci_resume(pdev);
 }
 
-static int vmw_pm_prepare(struct device *kdev)
+static int vmw_pm_freeze(struct device *kdev)
 {
 	struct pci_dev *pdev = to_pci_dev(kdev);
 	struct drm_device *dev = pci_get_drvdata(pdev);
 	struct vmw_private *dev_priv = vmw_priv(dev);
 
-	/**
-	 * Release 3d reference held by fbdev and potentially
-	 * stop fifo.
-	 */
 	dev_priv->suspended = true;
 	if (dev_priv->enable_fb)
-			vmw_3d_resource_dec(dev_priv, true);
-
-	if (dev_priv->num_3d_resources != 0) {
-
-		DRM_INFO("Can't suspend or hibernate "
-			 "while 3D resources are active.\n");
+		vmw_fifo_resource_dec(dev_priv);
 
+	if (atomic_read(&dev_priv->num_fifo_resources) != 0) {
+		DRM_ERROR("Can't hibernate while 3D resources are active.\n");
 		if (dev_priv->enable_fb)
-			vmw_3d_resource_inc(dev_priv, true);
+			vmw_fifo_resource_inc(dev_priv);
+		WARN_ON(vmw_request_device_late(dev_priv));
 		dev_priv->suspended = false;
 		return -EBUSY;
 	}
 
+	if (dev_priv->enable_fb)
+		__vmw_svga_disable(dev_priv);
+	
+	vmw_release_device_late(dev_priv);
+
 	return 0;
 }
 
-static void vmw_pm_complete(struct device *kdev)
+static int vmw_pm_restore(struct device *kdev)
 {
 	struct pci_dev *pdev = to_pci_dev(kdev);
 	struct drm_device *dev = pci_get_drvdata(pdev);
 	struct vmw_private *dev_priv = vmw_priv(dev);
+	int ret;
 
 	vmw_write(dev_priv, SVGA_REG_ID, SVGA_ID_2);
 	(void) vmw_read(dev_priv, SVGA_REG_ID);
 
-	/**
-	 * Reclaim 3d reference held by fbdev and potentially
-	 * start fifo.
-	 */
 	if (dev_priv->enable_fb)
-			vmw_3d_resource_inc(dev_priv, false);
+		vmw_fifo_resource_inc(dev_priv);
+
+	ret = vmw_request_device(dev_priv);
+	if (ret)
+		return ret;
+
+	if (dev_priv->enable_fb)
+		__vmw_svga_enable(dev_priv);
 
 	dev_priv->suspended = false;
+
+	return 0;
 }
 
 static const struct dev_pm_ops vmw_pm_ops = {
-	.prepare = vmw_pm_prepare,
-	.complete = vmw_pm_complete,
+	.freeze = vmw_pm_freeze,
+	.thaw = vmw_pm_restore,
+	.restore = vmw_pm_restore,
 	.suspend = vmw_pm_suspend,
 	.resume = vmw_pm_resume,
 };
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index d26a6da..a5f221e 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -484,6 +484,7 @@ struct vmw_private {
 
 	bool stealth;
 	bool enable_fb;
+	spinlock_t svga_lock;
 
 	/**
 	 * Master management.
@@ -493,9 +494,10 @@ struct vmw_private {
 	struct vmw_master fbdev_master;
 	struct notifier_block pm_nb;
 	bool suspended;
+	bool refuse_hibernation;
 
 	struct mutex release_mutex;
-	uint32_t num_3d_resources;
+	atomic_t num_fifo_resources;
 
 	/*
 	 * Replace this with an rwsem as soon as we have down_xx_interruptible()
@@ -587,8 +589,9 @@ static inline uint32_t vmw_read(struct vmw_private *dev_priv,
 	return val;
 }
 
-int vmw_3d_resource_inc(struct vmw_private *dev_priv, bool unhide_svga);
-void vmw_3d_resource_dec(struct vmw_private *dev_priv, bool hide_svga);
+extern void vmw_svga_enable(struct vmw_private *dev_priv);
+extern void vmw_svga_disable(struct vmw_private *dev_priv);
+
 
 /**
  * GMR utilities - vmwgfx_gmr.c
@@ -1116,4 +1119,14 @@ static inline struct ttm_mem_global *vmw_mem_glob(struct vmw_private *dev_priv)
 {
 	return (struct ttm_mem_global *) dev_priv->mem_global_ref.object;
 }
+
+static inline void vmw_fifo_resource_inc(struct vmw_private *dev_priv)
+{
+	atomic_inc(&dev_priv->num_fifo_resources);
+}
+
+static inline void vmw_fifo_resource_dec(struct vmw_private *dev_priv)
+{
+	atomic_dec(&dev_priv->num_fifo_resources);
+}
 #endif
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
index e2d40eb..ecdc8d9 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
@@ -596,7 +596,10 @@ int vmw_fb_off(struct vmw_private *vmw_priv)
 
 	info = vmw_priv->fb_info;
 	par = info->par;
+	if (!par->bo_ptr)
+		return 0;
 
+	vmw_kms_save_vga(vmw_priv);
 	spin_lock_irqsave(&par->dirty.lock, flags);
 	par->dirty.active = false;
 	spin_unlock_irqrestore(&par->dirty.lock, flags);
@@ -648,6 +651,7 @@ int vmw_fb_on(struct vmw_private *vmw_priv)
 	spin_lock_irqsave(&par->dirty.lock, flags);
 	par->dirty.active = true;
 	spin_unlock_irqrestore(&par->dirty.lock, flags);
+	vmw_kms_restore_vga(vmw_priv);
 
 err_no_buffer:
 	vmw_fb_set_par(info);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
index 39f2b03..cd5d9f3 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
@@ -98,7 +98,6 @@ int vmw_fifo_init(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
 	__le32 __iomem *fifo_mem = dev_priv->mmio_virt;
 	uint32_t max;
 	uint32_t min;
-	uint32_t dummy;
 
 	fifo->static_buffer_size = VMWGFX_FIFO_STATIC_SIZE;
 	fifo->static_buffer = vmalloc(fifo->static_buffer_size);
@@ -112,10 +111,6 @@ int vmw_fifo_init(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
 	mutex_init(&fifo->fifo_mutex);
 	init_rwsem(&fifo->rwsem);
 
-	/*
-	 * Allow mapping the first page read-only to user-space.
-	 */
-
 	DRM_INFO("width %d\n", vmw_read(dev_priv, SVGA_REG_WIDTH));
 	DRM_INFO("height %d\n", vmw_read(dev_priv, SVGA_REG_HEIGHT));
 	DRM_INFO("bpp %d\n", vmw_read(dev_priv, SVGA_REG_BITS_PER_PIXEL));
@@ -123,7 +118,9 @@ int vmw_fifo_init(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
 	dev_priv->enable_state = vmw_read(dev_priv, SVGA_REG_ENABLE);
 	dev_priv->config_done_state = vmw_read(dev_priv, SVGA_REG_CONFIG_DONE);
 	dev_priv->traces_state = vmw_read(dev_priv, SVGA_REG_TRACES);
-	vmw_write(dev_priv, SVGA_REG_ENABLE, 1);
+
+	vmw_write(dev_priv, SVGA_REG_ENABLE, SVGA_REG_ENABLE_ENABLE_HIDE);
+	vmw_write(dev_priv, SVGA_REG_TRACES, 0);
 
 	min = 4;
 	if (dev_priv->capabilities & SVGA_CAP_EXTENDED_FIFO)
@@ -155,7 +152,8 @@ int vmw_fifo_init(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
 	atomic_set(&dev_priv->marker_seq, dev_priv->last_read_seqno);
 	iowrite32(dev_priv->last_read_seqno, fifo_mem + SVGA_FIFO_FENCE);
 	vmw_marker_queue_init(&fifo->marker_queue);
-	return vmw_fifo_send_fence(dev_priv, &dummy);
+
+	return 0;
 }
 
 void vmw_fifo_ping_host(struct vmw_private *dev_priv, uint32_t reason)
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c b/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c
index 5c289f7..53579f2 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c
@@ -280,6 +280,7 @@ static int vmw_ldu_crtc_set_config(struct drm_mode_set *set)
 	}
 
 	vmw_fb_off(dev_priv);
+	vmw_svga_enable(dev_priv);
 
 	crtc->primary->fb = fb;
 	encoder->crtc = crtc;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c b/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c
index 04a64b8..f06d60f 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c
@@ -574,7 +574,7 @@ void vmw_mob_unbind(struct vmw_private *dev_priv,
 		vmw_fence_single_bo(bo, NULL);
 		ttm_bo_unreserve(bo);
 	}
-	vmw_3d_resource_dec(dev_priv, false);
+	vmw_fifo_resource_dec(dev_priv);
 }
 
 /*
@@ -627,7 +627,7 @@ int vmw_mob_bind(struct vmw_private *dev_priv,
 		mob->pt_level += VMW_MOBFMT_PTDEPTH_1 - SVGA3D_MOBFMT_PTDEPTH_1;
 	}
 
-	(void) vmw_3d_resource_inc(dev_priv, false);
+	vmw_fifo_resource_inc(dev_priv);
 
 	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
 	if (unlikely(cmd == NULL)) {
@@ -648,7 +648,7 @@ int vmw_mob_bind(struct vmw_private *dev_priv,
 	return 0;
 
 out_no_cmd_space:
-	vmw_3d_resource_dec(dev_priv, false);
+	vmw_fifo_resource_dec(dev_priv);
 	if (pt_set_up)
 		ttm_bo_unref(&mob->pt_bo);
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c b/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c
index 7dc591d..9e8eb36 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c
@@ -332,6 +332,7 @@ static int vmw_sou_crtc_set_config(struct drm_mode_set *set)
 	}
 
 	vmw_fb_off(dev_priv);
+	vmw_svga_enable(dev_priv);
 
 	if (mode->hdisplay != crtc->mode.hdisplay ||
 	    mode->vdisplay != crtc->mode.vdisplay) {
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c b/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c
index 6a4584a..6110a43 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c
@@ -165,7 +165,7 @@ static int vmw_gb_shader_create(struct vmw_resource *res)
 	cmd->body.type = shader->type;
 	cmd->body.sizeInBytes = shader->size;
 	vmw_fifo_commit(dev_priv, sizeof(*cmd));
-	(void) vmw_3d_resource_inc(dev_priv, false);
+	vmw_fifo_resource_inc(dev_priv);
 
 	return 0;
 
@@ -275,7 +275,7 @@ static int vmw_gb_shader_destroy(struct vmw_resource *res)
 	vmw_fifo_commit(dev_priv, sizeof(*cmd));
 	mutex_unlock(&dev_priv->binding_mutex);
 	vmw_resource_release_id(res);
-	vmw_3d_resource_dec(dev_priv, false);
+	vmw_fifo_resource_dec(dev_priv);
 
 	return 0;
 }
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c b/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
index 4ecdbf3..4d0c98e 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
@@ -340,7 +340,7 @@ static void vmw_hw_surface_destroy(struct vmw_resource *res)
 		dev_priv->used_memory_size -= res->backup_size;
 		mutex_unlock(&dev_priv->cmdbuf_mutex);
 	}
-	vmw_3d_resource_dec(dev_priv, false);
+	vmw_fifo_resource_dec(dev_priv);
 }
 
 /**
@@ -576,14 +576,14 @@ static int vmw_surface_init(struct vmw_private *dev_priv,
 
 	BUG_ON(res_free == NULL);
 	if (!dev_priv->has_mob)
-		(void) vmw_3d_resource_inc(dev_priv, false);
+		vmw_fifo_resource_inc(dev_priv);
 	ret = vmw_resource_init(dev_priv, res, true, res_free,
 				(dev_priv->has_mob) ? &vmw_gb_surface_func :
 				&vmw_legacy_surface_func);
 
 	if (unlikely(ret != 0)) {
 		if (!dev_priv->has_mob)
-			vmw_3d_resource_dec(dev_priv, false);
+			vmw_fifo_resource_dec(dev_priv);
 		res_free(res);
 		return ret;
 	}
@@ -1028,7 +1028,7 @@ static int vmw_gb_surface_create(struct vmw_resource *res)
 	if (likely(res->id != -1))
 		return 0;
 
-	(void) vmw_3d_resource_inc(dev_priv, false);
+	vmw_fifo_resource_inc(dev_priv);
 	ret = vmw_resource_alloc_id(res);
 	if (unlikely(ret != 0)) {
 		DRM_ERROR("Failed to allocate a surface id.\n");
@@ -1068,7 +1068,7 @@ static int vmw_gb_surface_create(struct vmw_resource *res)
 out_no_fifo:
 	vmw_resource_release_id(res);
 out_no_id:
-	vmw_3d_resource_dec(dev_priv, false);
+	vmw_fifo_resource_dec(dev_priv);
 	return ret;
 }
 
@@ -1213,7 +1213,7 @@ static int vmw_gb_surface_destroy(struct vmw_resource *res)
 	vmw_fifo_commit(dev_priv, sizeof(*cmd));
 	mutex_unlock(&dev_priv->binding_mutex);
 	vmw_resource_release_id(res);
-	vmw_3d_resource_dec(dev_priv, false);
+	vmw_fifo_resource_dec(dev_priv);
 
 	return 0;
 }
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 03/28] drm/vmwgfx: Fix OTABLE takedown

[Thomas Hellstrom]

Don't fence and free the BO if command submission fails.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_mob.c | 21 +++++++++++----------
 1 file changed, 11 insertions(+), 10 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c b/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c
index f06d60f..46f975e 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c
@@ -191,17 +191,18 @@ static void vmw_takedown_otable_base(struct vmw_private *dev_priv,
 	if (unlikely(cmd == NULL)) {
 		DRM_ERROR("Failed reserving FIFO space for OTable "
 			  "takedown.\n");
-	} else {
-		memset(cmd, 0, sizeof(*cmd));
-		cmd->header.id = SVGA_3D_CMD_SET_OTABLE_BASE;
-		cmd->header.size = sizeof(cmd->body);
-		cmd->body.type = type;
-		cmd->body.baseAddress = 0;
-		cmd->body.sizeInBytes = 0;
-		cmd->body.validSizeInBytes = 0;
-		cmd->body.ptDepth = SVGA3D_MOBFMT_INVALID;
-		vmw_fifo_commit(dev_priv, sizeof(*cmd));
+		return;
 	}
+ 
+	memset(cmd, 0, sizeof(*cmd));
+	cmd->header.id = SVGA_3D_CMD_SET_OTABLE_BASE;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.type = type;
+	cmd->body.baseAddress = 0;
+	cmd->body.sizeInBytes = 0;
+	cmd->body.validSizeInBytes = 0;
+	cmd->body.ptDepth = SVGA3D_MOBFMT_INVALID;
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
 
 	if (bo) {
 		int ret;
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 04/28] vmwgfx: Update device headers for command buffers.

[Thomas Hellstrom]

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
---
 drivers/gpu/drm/vmwgfx/svga3d_reg.h |   2 -
 drivers/gpu/drm/vmwgfx/svga_reg.h   | 187 ++++++++++++++++++++++++++++++++++++
 drivers/gpu/drm/vmwgfx/svga_types.h |   3 +
 3 files changed, 190 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/svga3d_reg.h b/drivers/gpu/drm/vmwgfx/svga3d_reg.h
index f58dc7d..e50d20c 100644
--- a/drivers/gpu/drm/vmwgfx/svga3d_reg.h
+++ b/drivers/gpu/drm/vmwgfx/svga3d_reg.h
@@ -1928,8 +1928,6 @@ struct {
  * Guest-backed surface definitions.
  */
 
-typedef uint32 SVGAMobId;
-
 typedef enum SVGAMobFormat {
    SVGA3D_MOBFMT_INVALID = SVGA3D_INVALID_ID,
    SVGA3D_MOBFMT_PTDEPTH_0 = 0,
diff --git a/drivers/gpu/drm/vmwgfx/svga_reg.h b/drivers/gpu/drm/vmwgfx/svga_reg.h
index e4259c2..3763d5b 100644
--- a/drivers/gpu/drm/vmwgfx/svga_reg.h
+++ b/drivers/gpu/drm/vmwgfx/svga_reg.h
@@ -106,6 +106,8 @@
 #define SVGA_IRQFLAG_ANY_FENCE            0x1    /* Any fence was passed */
 #define SVGA_IRQFLAG_FIFO_PROGRESS        0x2    /* Made forward progress in the FIFO */
 #define SVGA_IRQFLAG_FENCE_GOAL           0x4    /* SVGA_FIFO_FENCE_GOAL reached */
+#define SVGA_IRQFLAG_COMMAND_BUFFER       0x8    /* Command buffer completed */
+#define SVGA_IRQFLAG_ERROR                0x10   /* Error while processing commands */
 
 /*
  * Registers
@@ -299,6 +301,190 @@ struct SVGAGuestPtr {
    uint32 offset;
 } SVGAGuestPtr;
 
+/*
+ * Register based command buffers --
+ *
+ * Provide an SVGA device interface that allows the guest to submit
+ * command buffers to the SVGA device through an SVGA device register.
+ * The metadata for each command buffer is contained in the
+ * SVGACBHeader structure along with the return status codes.
+ *
+ * The SVGA device supports command buffers if
+ * SVGA_CAP_COMMAND_BUFFERS is set in the device caps register.  The
+ * fifo must be enabled for command buffers to be submitted.
+ *
+ * Command buffers are submitted when the guest writing the 64 byte
+ * aligned physical address into the SVGA_REG_COMMAND_LOW and
+ * SVGA_REG_COMMAND_HIGH.  SVGA_REG_COMMAND_HIGH contains the upper 32
+ * bits of the physical address.  SVGA_REG_COMMAND_LOW contains the
+ * lower 32 bits of the physical address, since the command buffer
+ * headers are required to be 64 byte aligned the lower 6 bits are
+ * used for the SVGACBContext value.  Writing to SVGA_REG_COMMAND_LOW
+ * submits the command buffer to the device and queues it for
+ * execution.  The SVGA device supports at least
+ * SVGA_CB_MAX_QUEUED_PER_CONTEXT command buffers that can be queued
+ * per context and if that limit is reached the device will write the
+ * status SVGA_CB_STATUS_QUEUE_FULL to the status value of the command
+ * buffer header synchronously and not raise any IRQs.
+ *
+ * It is invalid to submit a command buffer without a valid physical
+ * address and results are undefined.
+ *
+ * The device guarantees that command buffers of size SVGA_CB_MAX_SIZE
+ * will be supported.  If a larger command buffer is submitted results
+ * are unspecified and the device will either complete the command
+ * buffer or return an error.
+ *
+ * The device guarantees that any individual command in a command
+ * buffer can be up to SVGA_CB_MAX_COMMAND_SIZE in size which is
+ * enough to fit a 64x64 color-cursor definition.  If the command is
+ * too large the device is allowed to process the command or return an
+ * error.
+ *
+ * The device context is a special SVGACBContext that allows for
+ * synchronous register like accesses with the flexibility of
+ * commands.  There is a different command set defined by
+ * SVGADeviceContextCmdId.  The commands in each command buffer is not
+ * allowed to straddle physical pages.
+ *
+ * The offset field which is available starting with the
+ * SVGA_CAP_CMD_BUFFERS_2 cap bit can be set by the guest to bias the
+ * start of command processing into the buffer.  If an error is
+ * encountered the errorOffset will still be relative to the specific
+ * PA, not biased by the offset.  When the command buffer is finished
+ * the guest should not read the offset field as there is no guarantee
+ * what it will set to.
+ */
+
+#define SVGA_CB_MAX_SIZE (512 * 1024)  // 512 KB
+#define SVGA_CB_MAX_QUEUED_PER_CONTEXT 32
+#define SVGA_CB_MAX_COMMAND_SIZE (32 * 1024) // 32 KB
+
+#define SVGA_CB_CONTEXT_MASK 0x3f
+typedef enum {
+   SVGA_CB_CONTEXT_DEVICE = 0x3f,
+   SVGA_CB_CONTEXT_0      = 0x0,
+   SVGA_CB_CONTEXT_MAX    = 0x1,
+} SVGACBContext;
+
+
+typedef enum {
+   /*
+    * The guest is supposed to write SVGA_CB_STATUS_NONE to the status
+    * field before submitting the command buffer header, the host will
+    * change the value when it is done with the command buffer.
+    */
+   SVGA_CB_STATUS_NONE             = 0,
+
+   /*
+    * Written by the host when a command buffer completes successfully.
+    * The device raises an IRQ with SVGA_IRQFLAG_COMMAND_BUFFER unless
+    * the SVGA_CB_FLAG_NO_IRQ flag is set.
+    */
+   SVGA_CB_STATUS_COMPLETED        = 1,
+
+   /*
+    * Written by the host synchronously with the command buffer
+    * submission to indicate the command buffer was not submitted.  No
+    * IRQ is raised.
+    */
+   SVGA_CB_STATUS_QUEUE_FULL       = 2,
+
+   /*
+    * Written by the host when an error was detected parsing a command
+    * in the command buffer, errorOffset is written to contain the
+    * offset to the first byte of the failing command.  The device
+    * raises the IRQ with both SVGA_IRQFLAG_ERROR and
+    * SVGA_IRQFLAG_COMMAND_BUFFER.  Some of the commands may have been
+    * processed.
+    */
+   SVGA_CB_STATUS_COMMAND_ERROR    = 3,
+
+   /*
+    * Written by the host if there is an error parsing the command
+    * buffer header.  The device raises the IRQ with both
+    * SVGA_IRQFLAG_ERROR and SVGA_IRQFLAG_COMMAND_BUFFER.  The device
+    * did not processes any of the command buffer.
+    */
+   SVGA_CB_STATUS_CB_HEADER_ERROR  = 4,
+
+   /*
+    * Written by the host if the guest requested the host to preempt
+    * the command buffer.  The device will not raise any IRQs and the
+    * command buffer was not processed.
+    */
+   SVGA_CB_STATUS_PREEMPTED        = 5,
+
+   /*
+    * Written by the host synchronously with the command buffer
+    * submission to indicate the the command buffer was not submitted
+    * due to an error.  No IRQ is raised.
+    */
+   SVGA_CB_STATUS_SUBMISSION_ERROR = 6,
+} SVGACBStatus;
+
+typedef enum {
+   SVGA_CB_FLAG_NONE       = 0,
+   SVGA_CB_FLAG_NO_IRQ     = 1 << 0,
+   SVGA_CB_FLAG_DX_CONTEXT = 1 << 1,
+   SVGA_CB_FLAG_MOB        = 1 << 2,
+} SVGACBFlags;
+
+typedef
+struct {
+   volatile SVGACBStatus status; /* Modified by device. */
+   volatile uint32 errorOffset;  /* Modified by device. */
+   uint64 id;
+   SVGACBFlags flags;
+   uint32 length;
+   union {
+      PA pa;
+      struct {
+         SVGAMobId mobid;
+         uint32 mobOffset;
+      } mob;
+   } ptr;
+   uint32 offset; /* Valid if CMD_BUFFERS_2 cap set, must be zero otherwise,
+                   * modified by device.
+                   */
+   uint32 dxContext; /* Valid if DX_CONTEXT flag set, must be zero otherwise */
+   uint32 mustBeZero[6];
+}
+__attribute__((__packed__))
+SVGACBHeader;
+
+typedef enum {
+   SVGA_DC_CMD_NOP                   = 0,
+   SVGA_DC_CMD_START_STOP_CONTEXT    = 1,
+   SVGA_DC_CMD_PREEMPT               = 2,
+   SVGA_DC_CMD_MAX                   = 3,
+} SVGADeviceContextCmdId;
+
+
+typedef struct {
+   uint32 enable;
+   SVGACBContext context;
+} SVGADCCmdStartStop;
+
+/*
+ * SVGADCCmdPreempt --
+ *
+ * This command allows the guest to request that all command buffers
+ * on the specified context be preempted that can be.  After execution
+ * of this command all command buffers that were preempted will
+ * already have SVGA_CB_STATUS_PREEMPTED written into the status
+ * field.  The device might still be processing a command buffer,
+ * assuming execution of it started before the preemption request was
+ * received.  Specifying the ignoreIDZero flag to TRUE will cause the
+ * device to not preempt command buffers with the id field in the
+ * command buffer header set to zero.
+ */
+
+typedef struct {
+   SVGACBContext context;
+   uint32 ignoreIDZero;
+} SVGADCCmdPreempt;
+
 
 /*
  * SVGAGMRImageFormat --
@@ -444,6 +630,7 @@ struct SVGASignedPoint {
 #define SVGA_CAP_DEAD1              0x02000000
 #define SVGA_CAP_CMD_BUFFERS_2      0x04000000
 #define SVGA_CAP_GBOBJECTS          0x08000000
+#define SVGA_CAP_CMD_BUFFERS_3      0x10000000
 
 /*
  * FIFO register indices.
diff --git a/drivers/gpu/drm/vmwgfx/svga_types.h b/drivers/gpu/drm/vmwgfx/svga_types.h
index 55836de..1186898 100644
--- a/drivers/gpu/drm/vmwgfx/svga_types.h
+++ b/drivers/gpu/drm/vmwgfx/svga_types.h
@@ -40,6 +40,9 @@ typedef uint16_t uint16;
 typedef uint32_t uint32;
 typedef uint8_t uint8;
 typedef int32_t int32;
+typedef uint64_t uint64;
 typedef bool Bool;
+typedef uint64 PA;
+typedef uint32 SVGAMobId;
 
 #endif
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 05/28] drm/vmwgfx: Add command buffer support v3

[Thomas Hellstrom]

Add command buffer support.
Currently we don't implement preemption or fancy error handling.
Tested with a couple of mesa-demos, compiz/unity and viewperf maya-03.

v2:
- Synchronize with pending work at command buffer manager takedown.
- Add an interface to flush the current command buffer for latency-critical
  command batches and apply it to framebuffer dirtying.

v3:
- Minor fixes of definitions and typos to address reviews.
- Removed new or moved branch predictor hints.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
---
 drivers/gpu/drm/vmwgfx/Makefile         |    2 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c  |   13 +
 drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c  | 1315 +++++++++++++++++++++++++++++++
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.c     |   26 +
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.h     |   40 +
 drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c |  180 ++++-
 drivers/gpu/drm/vmwgfx/vmwgfx_fb.c      |    1 +
 drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c    |   67 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_irq.c     |   41 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_kms.c     |    4 +
 10 files changed, 1656 insertions(+), 33 deletions(-)
 create mode 100644 drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c

diff --git a/drivers/gpu/drm/vmwgfx/Makefile b/drivers/gpu/drm/vmwgfx/Makefile
index ce0ab95..529bc72 100644
--- a/drivers/gpu/drm/vmwgfx/Makefile
+++ b/drivers/gpu/drm/vmwgfx/Makefile
@@ -7,6 +7,6 @@ vmwgfx-y := vmwgfx_execbuf.o vmwgfx_gmr.o vmwgfx_kms.o vmwgfx_drv.o \
 	    vmwgfx_overlay.o vmwgfx_marker.o vmwgfx_gmrid_manager.o \
 	    vmwgfx_fence.o vmwgfx_dmabuf.o vmwgfx_scrn.o vmwgfx_context.o \
 	    vmwgfx_surface.o vmwgfx_prime.o vmwgfx_mob.o vmwgfx_shader.o \
-	    vmwgfx_cmdbuf_res.o \
+	    vmwgfx_cmdbuf_res.o vmwgfx_cmdbuf.o \
 
 obj-$(CONFIG_DRM_VMWGFX) := vmwgfx.o
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
index cff2bf9..3b349fd 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
@@ -72,6 +72,12 @@ static struct ttm_place mob_placement_flags = {
 	.flags = VMW_PL_FLAG_MOB | TTM_PL_FLAG_CACHED
 };
 
+static struct ttm_place mob_ne_placement_flags = {
+	.fpfn = 0,
+	.lpfn = 0,
+	.flags = VMW_PL_FLAG_MOB | TTM_PL_FLAG_CACHED | TTM_PL_FLAG_NO_EVICT
+};
+
 struct ttm_placement vmw_vram_placement = {
 	.num_placement = 1,
 	.placement = &vram_placement_flags,
@@ -200,6 +206,13 @@ struct ttm_placement vmw_mob_placement = {
 	.busy_placement = &mob_placement_flags
 };
 
+struct ttm_placement vmw_mob_ne_placement = {
+	.num_placement = 1,
+	.num_busy_placement = 1,
+	.placement = &mob_ne_placement_flags,
+	.busy_placement = &mob_ne_placement_flags
+};
+
 struct vmw_ttm_tt {
 	struct ttm_dma_tt dma_ttm;
 	struct vmw_private *dev_priv;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c
new file mode 100644
index 0000000..b044bf5
--- /dev/null
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c
@@ -0,0 +1,1315 @@
+/**************************************************************************
+ *
+ * Copyright © 2015 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 "vmwgfx_drv.h"
+#include "ttm/ttm_bo_api.h"
+
+/*
+ * Size of inline command buffers. Try to make sure that a page size is a
+ * multiple of the DMA pool allocation size.
+ */
+#define VMW_CMDBUF_INLINE_ALIGN 64
+#define VMW_CMDBUF_INLINE_SIZE (1024 - VMW_CMDBUF_INLINE_ALIGN)
+
+/**
+ * struct vmw_cmdbuf_context - Command buffer context queues
+ *
+ * @submitted: List of command buffers that have been submitted to the
+ * manager but not yet submitted to hardware.
+ * @hw_submitted: List of command buffers submitted to hardware.
+ * @preempted: List of preempted command buffers.
+ * @num_hw_submitted: Number of buffers currently being processed by hardware
+ */
+struct vmw_cmdbuf_context {
+	struct list_head submitted;
+	struct list_head hw_submitted;
+	struct list_head preempted;
+	unsigned num_hw_submitted;
+};
+
+/**
+ * struct vmw_cmdbuf_man: - Command buffer manager
+ *
+ * @cur_mutex: Mutex protecting the command buffer used for incremental small
+ * kernel command submissions, @cur.
+ * @space_mutex: Mutex to protect against starvation when we allocate
+ * main pool buffer space.
+ * @work: A struct work_struct implementeing command buffer error handling.
+ * Immutable.
+ * @dev_priv: Pointer to the device private struct. Immutable.
+ * @ctx: Array of command buffer context queues. The queues and the context
+ * data is protected by @lock.
+ * @error: List of command buffers that have caused device errors.
+ * Protected by @lock.
+ * @mm: Range manager for the command buffer space. Manager allocations and
+ * frees are protected by @lock.
+ * @cmd_space: Buffer object for the command buffer space, unless we were
+ * able to make a contigous coherent DMA memory allocation, @handle. Immutable.
+ * @map_obj: Mapping state for @cmd_space. Immutable.
+ * @map: Pointer to command buffer space. May be a mapped buffer object or
+ * a contigous coherent DMA memory allocation. Immutable.
+ * @cur: Command buffer for small kernel command submissions. Protected by
+ * the @cur_mutex.
+ * @cur_pos: Space already used in @cur. Protected by @cur_mutex.
+ * @default_size: Default size for the @cur command buffer. Immutable.
+ * @max_hw_submitted: Max number of in-flight command buffers the device can
+ * handle. Immutable.
+ * @lock: Spinlock protecting command submission queues.
+ * @header: Pool of DMA memory for device command buffer headers.
+ * Internal protection.
+ * @dheaders: Pool of DMA memory for device command buffer headers with trailing
+ * space for inline data. Internal protection.
+ * @tasklet: Tasklet struct for irq processing. Immutable.
+ * @alloc_queue: Wait queue for processes waiting to allocate command buffer
+ * space.
+ * @idle_queue: Wait queue for processes waiting for command buffer idle.
+ * @irq_on: Whether the process function has requested irq to be turned on.
+ * Protected by @lock.
+ * @using_mob: Whether the command buffer space is a MOB or a contigous DMA
+ * allocation. Immutable.
+ * @has_pool: Has a large pool of DMA memory which allows larger allocations.
+ * Typically this is false only during bootstrap.
+ * @handle: DMA address handle for the command buffer space if @using_mob is
+ * false. Immutable.
+ * @size: The size of the command buffer space. Immutable.
+ */
+struct vmw_cmdbuf_man {
+	struct mutex cur_mutex;
+	struct mutex space_mutex;
+	struct work_struct work;
+	struct vmw_private *dev_priv;
+	struct vmw_cmdbuf_context ctx[SVGA_CB_CONTEXT_MAX];
+	struct list_head error;
+	struct drm_mm mm;
+	struct ttm_buffer_object *cmd_space;
+	struct ttm_bo_kmap_obj map_obj;
+	u8 *map;
+	struct vmw_cmdbuf_header *cur;
+	size_t cur_pos;
+	size_t default_size;
+	unsigned max_hw_submitted;
+	spinlock_t lock;
+	struct dma_pool *headers;
+	struct dma_pool *dheaders;
+	struct tasklet_struct tasklet;
+	wait_queue_head_t alloc_queue;
+	wait_queue_head_t idle_queue;
+	bool irq_on;
+	bool using_mob;
+	bool has_pool;
+	dma_addr_t handle;
+	size_t size;
+};
+
+/**
+ * struct vmw_cmdbuf_header - Command buffer metadata
+ *
+ * @man: The command buffer manager.
+ * @cb_header: Device command buffer header, allocated from a DMA pool.
+ * @cb_context: The device command buffer context.
+ * @list: List head for attaching to the manager lists.
+ * @node: The range manager node.
+ * @handle. The DMA address of @cb_header. Handed to the device on command
+ * buffer submission.
+ * @cmd: Pointer to the command buffer space of this buffer.
+ * @size: Size of the command buffer space of this buffer.
+ * @reserved: Reserved space of this buffer.
+ * @inline_space: Whether inline command buffer space is used.
+ */
+struct vmw_cmdbuf_header {
+	struct vmw_cmdbuf_man *man;
+	SVGACBHeader *cb_header;
+	SVGACBContext cb_context;
+	struct list_head list;
+	struct drm_mm_node *node;
+	dma_addr_t handle;
+	u8 *cmd;
+	size_t size;
+	size_t reserved;
+	bool inline_space;
+};
+
+/**
+ * struct vmw_cmdbuf_dheader - Device command buffer header with inline
+ * command buffer space.
+ *
+ * @cb_header: Device command buffer header.
+ * @cmd: Inline command buffer space.
+ */
+struct vmw_cmdbuf_dheader {
+	SVGACBHeader cb_header;
+	u8 cmd[VMW_CMDBUF_INLINE_SIZE] __aligned(VMW_CMDBUF_INLINE_ALIGN);
+};
+
+/**
+ * struct vmw_cmdbuf_alloc_info - Command buffer space allocation metadata
+ *
+ * @page_size: Size of requested command buffer space in pages.
+ * @node: The range manager node if allocation succeeded.
+ * @ret: Error code if failure. Otherwise 0.
+ */
+struct vmw_cmdbuf_alloc_info {
+	size_t page_size;
+	struct drm_mm_node *node;
+	int ret;
+};
+
+/* Loop over each context in the command buffer manager. */
+#define for_each_cmdbuf_ctx(_man, _i, _ctx) \
+	for (_i = 0, _ctx = &(_man)->ctx[0]; (_i) < SVGA_CB_CONTEXT_MAX; \
+	     ++(_i), ++(_ctx))
+
+static int vmw_cmdbuf_startstop(struct vmw_cmdbuf_man *man, bool enable);
+
+
+/**
+ * vmw_cmdbuf_cur_lock - Helper to lock the cur_mutex.
+ *
+ * @man: The range manager.
+ * @interruptible: Whether to wait interruptible when locking.
+ */
+static int vmw_cmdbuf_cur_lock(struct vmw_cmdbuf_man *man, bool interruptible)
+{
+	if (interruptible) {
+		if (mutex_lock_interruptible(&man->cur_mutex))
+			return -ERESTARTSYS;
+	} else {
+		mutex_lock(&man->cur_mutex);
+	}
+
+	return 0;
+}
+
+/**
+ * vmw_cmdbuf_cur_unlock - Helper to unlock the cur_mutex.
+ *
+ * @man: The range manager.
+ */
+static void vmw_cmdbuf_cur_unlock(struct vmw_cmdbuf_man *man)
+{
+	mutex_unlock(&man->cur_mutex);
+}
+
+/**
+ * vmw_cmdbuf_header_inline_free - Free a struct vmw_cmdbuf_header that has
+ * been used for the device context with inline command buffers.
+ * Need not be called locked.
+ *
+ * @header: Pointer to the header to free.
+ */
+static void vmw_cmdbuf_header_inline_free(struct vmw_cmdbuf_header *header)
+{
+	struct vmw_cmdbuf_dheader *dheader;
+
+	if (WARN_ON_ONCE(!header->inline_space))
+		return;
+
+	dheader = container_of(header->cb_header, struct vmw_cmdbuf_dheader,
+			       cb_header);
+	dma_pool_free(header->man->dheaders, dheader, header->handle);
+	kfree(header);
+}
+
+/**
+ * __vmw_cmdbuf_header_free - Free a struct vmw_cmdbuf_header  and its
+ * associated structures.
+ *
+ * header: Pointer to the header to free.
+ *
+ * For internal use. Must be called with man::lock held.
+ */
+static void __vmw_cmdbuf_header_free(struct vmw_cmdbuf_header *header)
+{
+	struct vmw_cmdbuf_man *man = header->man;
+
+	BUG_ON(!spin_is_locked(&man->lock));
+
+	if (header->inline_space) {
+		vmw_cmdbuf_header_inline_free(header);
+		return;
+	}
+
+	drm_mm_remove_node(header->node);
+	kfree(header->node);
+	header->node = NULL;
+	wake_up_all(&man->alloc_queue);
+	if (header->cb_header)
+		dma_pool_free(man->headers, header->cb_header,
+			      header->handle);
+	kfree(header);
+}
+
+/**
+ * vmw_cmdbuf_header_free - Free a struct vmw_cmdbuf_header  and its
+ * associated structures.
+ *
+ * @header: Pointer to the header to free.
+ */
+void vmw_cmdbuf_header_free(struct vmw_cmdbuf_header *header)
+{
+	struct vmw_cmdbuf_man *man = header->man;
+
+	/* Avoid locking if inline_space */
+	if (header->inline_space) {
+		vmw_cmdbuf_header_inline_free(header);
+		return;
+	}
+	spin_lock_bh(&man->lock);
+	__vmw_cmdbuf_header_free(header);
+	spin_unlock_bh(&man->lock);
+}
+
+
+/**
+ * vmw_cmbuf_header_submit: Submit a command buffer to hardware.
+ *
+ * @header: The header of the buffer to submit.
+ */
+static int vmw_cmdbuf_header_submit(struct vmw_cmdbuf_header *header)
+{
+	struct vmw_cmdbuf_man *man = header->man;
+	u32 val;
+
+	val = (header->handle >> 32);
+	vmw_write(man->dev_priv, SVGA_REG_COMMAND_HIGH, val);
+	val = (header->handle & 0xFFFFFFFFULL);
+	val |= header->cb_context & SVGA_CB_CONTEXT_MASK;
+	vmw_write(man->dev_priv, SVGA_REG_COMMAND_LOW, val);
+
+	return header->cb_header->status;
+}
+
+/**
+ * vmw_cmdbuf_ctx_init: Initialize a command buffer context.
+ *
+ * @ctx: The command buffer context to initialize
+ */
+static void vmw_cmdbuf_ctx_init(struct vmw_cmdbuf_context *ctx)
+{
+	INIT_LIST_HEAD(&ctx->hw_submitted);
+	INIT_LIST_HEAD(&ctx->submitted);
+	INIT_LIST_HEAD(&ctx->preempted);
+	ctx->num_hw_submitted = 0;
+}
+
+/**
+ * vmw_cmdbuf_ctx_submit: Submit command buffers from a command buffer
+ * context.
+ *
+ * @man: The command buffer manager.
+ * @ctx: The command buffer context.
+ *
+ * Submits command buffers to hardware until there are no more command
+ * buffers to submit or the hardware can't handle more command buffers.
+ */
+static void vmw_cmdbuf_ctx_submit(struct vmw_cmdbuf_man *man,
+				  struct vmw_cmdbuf_context *ctx)
+{
+	while (ctx->num_hw_submitted < man->max_hw_submitted &&
+	      !list_empty(&ctx->submitted)) {
+		struct vmw_cmdbuf_header *entry;
+		SVGACBStatus status;
+
+		entry = list_first_entry(&ctx->submitted,
+					 struct vmw_cmdbuf_header,
+					 list);
+
+		status = vmw_cmdbuf_header_submit(entry);
+
+		/* This should never happen */
+		if (WARN_ON_ONCE(status == SVGA_CB_STATUS_QUEUE_FULL)) {
+			entry->cb_header->status = SVGA_CB_STATUS_NONE;
+			break;
+		}
+
+		list_del(&entry->list);
+		list_add_tail(&entry->list, &ctx->hw_submitted);
+		ctx->num_hw_submitted++;
+	}
+
+}
+
+/**
+ * vmw_cmdbuf_ctx_submit: Process a command buffer context.
+ *
+ * @man: The command buffer manager.
+ * @ctx: The command buffer context.
+ *
+ * Submit command buffers to hardware if possible, and process finished
+ * buffers. Typically freeing them, but on preemption or error take
+ * appropriate action. Wake up waiters if appropriate.
+ */
+static void vmw_cmdbuf_ctx_process(struct vmw_cmdbuf_man *man,
+				   struct vmw_cmdbuf_context *ctx,
+				   int *notempty)
+{
+	struct vmw_cmdbuf_header *entry, *next;
+
+	vmw_cmdbuf_ctx_submit(man, ctx);
+
+	list_for_each_entry_safe(entry, next, &ctx->hw_submitted, list) {
+		SVGACBStatus status = entry->cb_header->status;
+
+		if (status == SVGA_CB_STATUS_NONE)
+			break;
+
+		list_del(&entry->list);
+		wake_up_all(&man->idle_queue);
+		ctx->num_hw_submitted--;
+		switch (status) {
+		case SVGA_CB_STATUS_COMPLETED:
+			__vmw_cmdbuf_header_free(entry);
+			break;
+		case SVGA_CB_STATUS_COMMAND_ERROR:
+		case SVGA_CB_STATUS_CB_HEADER_ERROR:
+			list_add_tail(&entry->list, &man->error);
+			schedule_work(&man->work);
+			break;
+		case SVGA_CB_STATUS_PREEMPTED:
+			list_add(&entry->list, &ctx->preempted);
+			break;
+		default:
+			WARN_ONCE(true, "Undefined command buffer status.\n");
+			__vmw_cmdbuf_header_free(entry);
+			break;
+		}
+	}
+
+	vmw_cmdbuf_ctx_submit(man, ctx);
+	if (!list_empty(&ctx->submitted))
+		(*notempty)++;
+}
+
+/**
+ * vmw_cmdbuf_man_process - Process all command buffer contexts and
+ * switch on and off irqs as appropriate.
+ *
+ * @man: The command buffer manager.
+ *
+ * Calls vmw_cmdbuf_ctx_process() on all contexts. If any context has
+ * command buffers left that are not submitted to hardware, Make sure
+ * IRQ handling is turned on. Otherwise, make sure it's turned off. This
+ * function may return -EAGAIN to indicate it should be rerun due to
+ * possibly missed IRQs if IRQs has just been turned on.
+ */
+static int vmw_cmdbuf_man_process(struct vmw_cmdbuf_man *man)
+{
+	int notempty = 0;
+	struct vmw_cmdbuf_context *ctx;
+	int i;
+
+	for_each_cmdbuf_ctx(man, i, ctx)
+		vmw_cmdbuf_ctx_process(man, ctx, &notempty);
+
+	if (man->irq_on && !notempty) {
+		vmw_generic_waiter_remove(man->dev_priv,
+					  SVGA_IRQFLAG_COMMAND_BUFFER,
+					  &man->dev_priv->cmdbuf_waiters);
+		man->irq_on = false;
+	} else if (!man->irq_on && notempty) {
+		vmw_generic_waiter_add(man->dev_priv,
+				       SVGA_IRQFLAG_COMMAND_BUFFER,
+				       &man->dev_priv->cmdbuf_waiters);
+		man->irq_on = true;
+
+		/* Rerun in case we just missed an irq. */
+		return -EAGAIN;
+	}
+
+	return 0;
+}
+
+/**
+ * vmw_cmdbuf_ctx_add - Schedule a command buffer for submission on a
+ * command buffer context
+ *
+ * @man: The command buffer manager.
+ * @header: The header of the buffer to submit.
+ * @cb_context: The command buffer context to use.
+ *
+ * This function adds @header to the "submitted" queue of the command
+ * buffer context identified by @cb_context. It then calls the command buffer
+ * manager processing to potentially submit the buffer to hardware.
+ * @man->lock needs to be held when calling this function.
+ */
+static void vmw_cmdbuf_ctx_add(struct vmw_cmdbuf_man *man,
+			       struct vmw_cmdbuf_header *header,
+			       SVGACBContext cb_context)
+{
+	if (!(header->cb_header->flags & SVGA_CB_FLAG_DX_CONTEXT))
+		header->cb_header->dxContext = 0;
+	header->cb_context = cb_context;
+	list_add_tail(&header->list, &man->ctx[cb_context].submitted);
+
+	if (vmw_cmdbuf_man_process(man) == -EAGAIN)
+		vmw_cmdbuf_man_process(man);
+}
+
+/**
+ * vmw_cmdbuf_man_tasklet - The main part of the command buffer interrupt
+ * handler implemented as a tasklet.
+ *
+ * @data: Tasklet closure. A pointer to the command buffer manager cast to
+ * an unsigned long.
+ *
+ * The bottom half (tasklet) of the interrupt handler simply calls into the
+ * command buffer processor to free finished buffers and submit any
+ * queued buffers to hardware.
+ */
+static void vmw_cmdbuf_man_tasklet(unsigned long data)
+{
+	struct vmw_cmdbuf_man *man = (struct vmw_cmdbuf_man *) data;
+
+	spin_lock(&man->lock);
+	if (vmw_cmdbuf_man_process(man) == -EAGAIN)
+		(void) vmw_cmdbuf_man_process(man);
+	spin_unlock(&man->lock);
+}
+
+/**
+ * vmw_cmdbuf_work_func - The deferred work function that handles
+ * command buffer errors.
+ *
+ * @work: The work func closure argument.
+ *
+ * Restarting the command buffer context after an error requires process
+ * context, so it is deferred to this work function.
+ */
+static void vmw_cmdbuf_work_func(struct work_struct *work)
+{
+	struct vmw_cmdbuf_man *man =
+		container_of(work, struct vmw_cmdbuf_man, work);
+	struct vmw_cmdbuf_header *entry, *next;
+	bool restart;
+
+	spin_lock_bh(&man->lock);
+	list_for_each_entry_safe(entry, next, &man->error, list) {
+		restart = true;
+		DRM_ERROR("Command buffer error.\n");
+
+		list_del(&entry->list);
+		__vmw_cmdbuf_header_free(entry);
+		wake_up_all(&man->idle_queue);
+	}
+	spin_unlock_bh(&man->lock);
+
+	if (restart && vmw_cmdbuf_startstop(man, true))
+		DRM_ERROR("Failed restarting command buffer context 0.\n");
+
+}
+
+/**
+ * vmw_cmdbuf_man idle - Check whether the command buffer manager is idle.
+ *
+ * @man: The command buffer manager.
+ * @check_preempted: Check also the preempted queue for pending command buffers.
+ *
+ */
+static bool vmw_cmdbuf_man_idle(struct vmw_cmdbuf_man *man,
+				bool check_preempted)
+{
+	struct vmw_cmdbuf_context *ctx;
+	bool idle = false;
+	int i;
+
+	spin_lock_bh(&man->lock);
+	vmw_cmdbuf_man_process(man);
+	for_each_cmdbuf_ctx(man, i, ctx) {
+		if (!list_empty(&ctx->submitted) ||
+		    !list_empty(&ctx->hw_submitted) ||
+		    (check_preempted && !list_empty(&ctx->preempted)))
+			goto out_unlock;
+	}
+
+	idle = list_empty(&man->error);
+
+out_unlock:
+	spin_unlock_bh(&man->lock);
+
+	return idle;
+}
+
+/**
+ * __vmw_cmdbuf_cur_flush - Flush the current command buffer for small kernel
+ * command submissions
+ *
+ * @man: The command buffer manager.
+ *
+ * Flushes the current command buffer without allocating a new one. A new one
+ * is automatically allocated when needed. Call with @man->cur_mutex held.
+ */
+static void __vmw_cmdbuf_cur_flush(struct vmw_cmdbuf_man *man)
+{
+	struct vmw_cmdbuf_header *cur = man->cur;
+
+	WARN_ON(!mutex_is_locked(&man->cur_mutex));
+
+	if (!cur)
+		return;
+
+	spin_lock_bh(&man->lock);
+	if (man->cur_pos == 0) {
+		__vmw_cmdbuf_header_free(cur);
+		goto out_unlock;
+	}
+
+	man->cur->cb_header->length = man->cur_pos;
+	vmw_cmdbuf_ctx_add(man, man->cur, SVGA_CB_CONTEXT_0);
+out_unlock:
+	spin_unlock_bh(&man->lock);
+	man->cur = NULL;
+	man->cur_pos = 0;
+}
+
+/**
+ * vmw_cmdbuf_cur_flush - Flush the current command buffer for small kernel
+ * command submissions
+ *
+ * @man: The command buffer manager.
+ * @interruptible: Whether to sleep interruptible when sleeping.
+ *
+ * Flushes the current command buffer without allocating a new one. A new one
+ * is automatically allocated when needed.
+ */
+int vmw_cmdbuf_cur_flush(struct vmw_cmdbuf_man *man,
+			 bool interruptible)
+{
+	int ret = vmw_cmdbuf_cur_lock(man, interruptible);
+
+	if (ret)
+		return ret;
+
+	__vmw_cmdbuf_cur_flush(man);
+	vmw_cmdbuf_cur_unlock(man);
+
+	return 0;
+}
+
+/**
+ * vmw_cmdbuf_idle - Wait for command buffer manager idle.
+ *
+ * @man: The command buffer manager.
+ * @interruptible: Sleep interruptible while waiting.
+ * @timeout: Time out after this many ticks.
+ *
+ * Wait until the command buffer manager has processed all command buffers,
+ * or until a timeout occurs. If a timeout occurs, the function will return
+ * -EBUSY.
+ */
+int vmw_cmdbuf_idle(struct vmw_cmdbuf_man *man, bool interruptible,
+		    unsigned long timeout)
+{
+	int ret;
+
+	ret = vmw_cmdbuf_cur_flush(man, interruptible);
+	vmw_generic_waiter_add(man->dev_priv,
+			       SVGA_IRQFLAG_COMMAND_BUFFER,
+			       &man->dev_priv->cmdbuf_waiters);
+
+	if (interruptible) {
+		ret = wait_event_interruptible_timeout
+			(man->idle_queue, vmw_cmdbuf_man_idle(man, true),
+			 timeout);
+	} else {
+		ret = wait_event_timeout
+			(man->idle_queue, vmw_cmdbuf_man_idle(man, true),
+			 timeout);
+	}
+	vmw_generic_waiter_remove(man->dev_priv,
+				  SVGA_IRQFLAG_COMMAND_BUFFER,
+				  &man->dev_priv->cmdbuf_waiters);
+	if (ret == 0) {
+		if (!vmw_cmdbuf_man_idle(man, true))
+			ret = -EBUSY;
+		else
+			ret = 0;
+	}
+	if (ret > 0)
+		ret = 0;
+
+	return ret;
+}
+
+/**
+ * vmw_cmdbuf_try_alloc - Try to allocate buffer space from the main pool.
+ *
+ * @man: The command buffer manager.
+ * @info: Allocation info. Will hold the size on entry and allocated mm node
+ * on successful return.
+ *
+ * Try to allocate buffer space from the main pool. Returns true if succeeded.
+ * If a fatal error was hit, the error code is returned in @info->ret.
+ */
+static bool vmw_cmdbuf_try_alloc(struct vmw_cmdbuf_man *man,
+				 struct vmw_cmdbuf_alloc_info *info)
+{
+	int ret;
+
+	if (info->node)
+		return true;
+
+	info->node = kzalloc(sizeof(*info->node), GFP_KERNEL);
+	if (!info->node) {
+		info->ret = -ENOMEM;
+		return true;
+	}
+
+	spin_lock_bh(&man->lock);
+	ret = drm_mm_insert_node_generic(&man->mm, info->node, info->page_size, 0, 0,
+					 DRM_MM_SEARCH_DEFAULT,
+					 DRM_MM_CREATE_DEFAULT);
+	spin_unlock_bh(&man->lock);
+	if (ret) {
+		kfree(info->node);
+		info->node = NULL;
+	}
+
+	return !!info->node;
+}
+
+/**
+ * vmw_cmdbuf_alloc_space - Allocate buffer space from the main pool.
+ *
+ * @man: The command buffer manager.
+ * @size: The size of the allocation.
+ * @interruptible: Whether to sleep interruptible while waiting for space.
+ *
+ * This function allocates buffer space from the main pool, and if there is
+ * no space available ATM, it turns on IRQ handling and sleeps waiting for it to
+ * become available.
+ */
+static struct drm_mm_node *vmw_cmdbuf_alloc_space(struct vmw_cmdbuf_man *man,
+						  size_t size,
+						  bool interruptible)
+{
+	struct vmw_cmdbuf_alloc_info info;
+
+	info.page_size = PAGE_ALIGN(size) >> PAGE_SHIFT;
+	info.node = NULL;
+	info.ret = 0;
+
+	/*
+	 * To prevent starvation of large requests, only one allocating call
+	 * at a time waiting for space.
+	 */
+	if (interruptible) {
+		if (mutex_lock_interruptible(&man->space_mutex))
+			return ERR_PTR(-ERESTARTSYS);
+	} else {
+		mutex_lock(&man->space_mutex);
+	}
+
+	/* Try to allocate space without waiting. */
+	(void) vmw_cmdbuf_try_alloc(man, &info);
+	if (info.ret && !info.node) {
+		mutex_unlock(&man->space_mutex);
+		return ERR_PTR(info.ret);
+	}
+
+	if (info.node) {
+		mutex_unlock(&man->space_mutex);
+		return info.node;
+	}
+
+	vmw_generic_waiter_add(man->dev_priv,
+			       SVGA_IRQFLAG_COMMAND_BUFFER,
+			       &man->dev_priv->cmdbuf_waiters);
+
+	if (interruptible) {
+		int ret;
+
+		ret = wait_event_interruptible
+			(man->alloc_queue, vmw_cmdbuf_try_alloc(man, &info));
+		if (ret) {
+			vmw_generic_waiter_remove
+				(man->dev_priv, SVGA_IRQFLAG_COMMAND_BUFFER,
+				 &man->dev_priv->cmdbuf_waiters);
+			mutex_unlock(&man->space_mutex);
+			return ERR_PTR(ret);
+		}
+	} else {
+		wait_event(man->alloc_queue, vmw_cmdbuf_try_alloc(man, &info));
+	}
+	vmw_generic_waiter_remove(man->dev_priv,
+				  SVGA_IRQFLAG_COMMAND_BUFFER,
+				  &man->dev_priv->cmdbuf_waiters);
+	mutex_unlock(&man->space_mutex);
+	if (info.ret && !info.node)
+		return ERR_PTR(info.ret);
+
+	return info.node;
+}
+
+/**
+ * vmw_cmdbuf_space_pool - Set up a command buffer header with command buffer
+ * space from the main pool.
+ *
+ * @man: The command buffer manager.
+ * @header: Pointer to the header to set up.
+ * @size: The requested size of the buffer space.
+ * @interruptible: Whether to sleep interruptible while waiting for space.
+ */
+static int vmw_cmdbuf_space_pool(struct vmw_cmdbuf_man *man,
+				 struct vmw_cmdbuf_header *header,
+				 size_t size,
+				 bool interruptible)
+{
+	SVGACBHeader *cb_hdr;
+	size_t offset;
+	int ret;
+
+	if (!man->has_pool)
+		return -ENOMEM;
+
+	header->node = vmw_cmdbuf_alloc_space(man, size, interruptible);
+
+	if (IS_ERR(header->node))
+		return PTR_ERR(header->node);
+
+	header->cb_header = dma_pool_alloc(man->headers, GFP_KERNEL,
+					   &header->handle);
+	if (!header->cb_header) {
+		ret = -ENOMEM;
+		goto out_no_cb_header;
+	}
+
+	header->size = header->node->size << PAGE_SHIFT;
+	cb_hdr = header->cb_header;
+	offset = header->node->start << PAGE_SHIFT;
+	header->cmd = man->map + offset;
+	memset(cb_hdr, 0, sizeof(*cb_hdr));
+	if (man->using_mob) {
+		cb_hdr->flags = SVGA_CB_FLAG_MOB;
+		cb_hdr->ptr.mob.mobid = man->cmd_space->mem.start;
+		cb_hdr->ptr.mob.mobOffset = offset;
+	} else {
+		cb_hdr->ptr.pa = (u64)man->handle + (u64)offset;
+	}
+
+	return 0;
+
+out_no_cb_header:
+	spin_lock_bh(&man->lock);
+	drm_mm_remove_node(header->node);
+	spin_unlock_bh(&man->lock);
+	kfree(header->node);
+
+	return ret;
+}
+
+/**
+ * vmw_cmdbuf_space_inline - Set up a command buffer header with
+ * inline command buffer space.
+ *
+ * @man: The command buffer manager.
+ * @header: Pointer to the header to set up.
+ * @size: The requested size of the buffer space.
+ */
+static int vmw_cmdbuf_space_inline(struct vmw_cmdbuf_man *man,
+				   struct vmw_cmdbuf_header *header,
+				   int size)
+{
+	struct vmw_cmdbuf_dheader *dheader;
+	SVGACBHeader *cb_hdr;
+
+	if (WARN_ON_ONCE(size > VMW_CMDBUF_INLINE_SIZE))
+		return -ENOMEM;
+
+	dheader = dma_pool_alloc(man->dheaders, GFP_KERNEL,
+				 &header->handle);
+	if (!dheader)
+		return -ENOMEM;
+
+	header->inline_space = true;
+	header->size = VMW_CMDBUF_INLINE_SIZE;
+	cb_hdr = &dheader->cb_header;
+	header->cb_header = cb_hdr;
+	header->cmd = dheader->cmd;
+	memset(dheader, 0, sizeof(*dheader));
+	cb_hdr->status = SVGA_CB_STATUS_NONE;
+	cb_hdr->flags = SVGA_CB_FLAG_NONE;
+	cb_hdr->ptr.pa = (u64)header->handle +
+		(u64)offsetof(struct vmw_cmdbuf_dheader, cmd);
+
+	return 0;
+}
+
+/**
+ * vmw_cmdbuf_alloc - Allocate a command buffer header complete with
+ * command buffer space.
+ *
+ * @man: The command buffer manager.
+ * @size: The requested size of the buffer space.
+ * @interruptible: Whether to sleep interruptible while waiting for space.
+ * @p_header: points to a header pointer to populate on successful return.
+ *
+ * Returns a pointer to command buffer space if successful. Otherwise
+ * returns an error pointer. The header pointer returned in @p_header should
+ * be used for upcoming calls to vmw_cmdbuf_reserve() and vmw_cmdbuf_commit().
+ */
+void *vmw_cmdbuf_alloc(struct vmw_cmdbuf_man *man,
+		       size_t size, bool interruptible,
+		       struct vmw_cmdbuf_header **p_header)
+{
+	struct vmw_cmdbuf_header *header;
+	int ret = 0;
+
+	*p_header = NULL;
+
+	header = kzalloc(sizeof(*header), GFP_KERNEL);
+	if (!header)
+		return ERR_PTR(-ENOMEM);
+
+	if (size <= VMW_CMDBUF_INLINE_SIZE)
+		ret = vmw_cmdbuf_space_inline(man, header, size);
+	else
+		ret = vmw_cmdbuf_space_pool(man, header, size, interruptible);
+
+	if (ret) {
+		kfree(header);
+		return ERR_PTR(ret);
+	}
+
+	header->man = man;
+	INIT_LIST_HEAD(&header->list);
+	header->cb_header->status = SVGA_CB_STATUS_NONE;
+	*p_header = header;
+
+	return header->cmd;
+}
+
+/**
+ * vmw_cmdbuf_reserve_cur - Reserve space for commands in the current
+ * command buffer.
+ *
+ * @man: The command buffer manager.
+ * @size: The requested size of the commands.
+ * @ctx_id: The context id if any. Otherwise set to SVGA3D_REG_INVALID.
+ * @interruptible: Whether to sleep interruptible while waiting for space.
+ *
+ * Returns a pointer to command buffer space if successful. Otherwise
+ * returns an error pointer.
+ */
+static void *vmw_cmdbuf_reserve_cur(struct vmw_cmdbuf_man *man,
+				    size_t size,
+				    int ctx_id,
+				    bool interruptible)
+{
+	struct vmw_cmdbuf_header *cur;
+	void *ret;
+
+	if (vmw_cmdbuf_cur_lock(man, interruptible))
+		return ERR_PTR(-ERESTARTSYS);
+
+	cur = man->cur;
+	if (cur && (size + man->cur_pos > cur->size ||
+	    (ctx_id != SVGA3D_INVALID_ID &&
+	     (cur->cb_header->flags & SVGA_CB_FLAG_DX_CONTEXT) &&
+	     ctx_id != cur->cb_header->dxContext)))
+		__vmw_cmdbuf_cur_flush(man);
+
+	if (!man->cur) {
+		ret = vmw_cmdbuf_alloc(man,
+				       max_t(size_t, size, man->default_size),
+				       interruptible, &man->cur);
+		if (IS_ERR(ret)) {
+			vmw_cmdbuf_cur_unlock(man);
+			return ret;
+		}
+
+		cur = man->cur;
+	}
+
+	if (ctx_id != SVGA3D_INVALID_ID) {
+		cur->cb_header->flags |= SVGA_CB_FLAG_DX_CONTEXT;
+		cur->cb_header->dxContext = ctx_id;
+	}
+
+	cur->reserved = size;
+
+	return (void *) (man->cur->cmd + man->cur_pos);
+}
+
+/**
+ * vmw_cmdbuf_commit_cur - Commit commands in the current command buffer.
+ *
+ * @man: The command buffer manager.
+ * @size: The size of the commands actually written.
+ * @flush: Whether to flush the command buffer immediately.
+ */
+static void vmw_cmdbuf_commit_cur(struct vmw_cmdbuf_man *man,
+				  size_t size, bool flush)
+{
+	struct vmw_cmdbuf_header *cur = man->cur;
+
+	WARN_ON(!mutex_is_locked(&man->cur_mutex));
+
+	WARN_ON(size > cur->reserved);
+	man->cur_pos += size;
+	if (!size)
+		cur->cb_header->flags &= ~SVGA_CB_FLAG_DX_CONTEXT;
+	if (flush)
+		__vmw_cmdbuf_cur_flush(man);
+	vmw_cmdbuf_cur_unlock(man);
+}
+
+/**
+ * vmw_cmdbuf_reserve - Reserve space for commands in a command buffer.
+ *
+ * @man: The command buffer manager.
+ * @size: The requested size of the commands.
+ * @ctx_id: The context id if any. Otherwise set to SVGA3D_REG_INVALID.
+ * @interruptible: Whether to sleep interruptible while waiting for space.
+ * @header: Header of the command buffer. NULL if the current command buffer
+ * should be used.
+ *
+ * Returns a pointer to command buffer space if successful. Otherwise
+ * returns an error pointer.
+ */
+void *vmw_cmdbuf_reserve(struct vmw_cmdbuf_man *man, size_t size,
+			 int ctx_id, bool interruptible,
+			 struct vmw_cmdbuf_header *header)
+{
+	if (!header)
+		return vmw_cmdbuf_reserve_cur(man, size, ctx_id, interruptible);
+
+	if (size > header->size)
+		return ERR_PTR(-EINVAL);
+
+	if (ctx_id != SVGA3D_INVALID_ID) {
+		header->cb_header->flags |= SVGA_CB_FLAG_DX_CONTEXT;
+		header->cb_header->dxContext = ctx_id;
+	}
+
+	header->reserved = size;
+	return header->cmd;
+}
+
+/**
+ * vmw_cmdbuf_commit - Commit commands in a command buffer.
+ *
+ * @man: The command buffer manager.
+ * @size: The size of the commands actually written.
+ * @header: Header of the command buffer. NULL if the current command buffer
+ * should be used.
+ * @flush: Whether to flush the command buffer immediately.
+ */
+void vmw_cmdbuf_commit(struct vmw_cmdbuf_man *man, size_t size,
+		       struct vmw_cmdbuf_header *header, bool flush)
+{
+	if (!header) {
+		vmw_cmdbuf_commit_cur(man, size, flush);
+		return;
+	}
+
+	(void) vmw_cmdbuf_cur_lock(man, false);
+	__vmw_cmdbuf_cur_flush(man);
+	WARN_ON(size > header->reserved);
+	man->cur = header;
+	man->cur_pos = size;
+	if (!size)
+		header->cb_header->flags &= ~SVGA_CB_FLAG_DX_CONTEXT;
+	if (flush)
+		__vmw_cmdbuf_cur_flush(man);
+	vmw_cmdbuf_cur_unlock(man);
+}
+
+/**
+ * vmw_cmdbuf_tasklet_schedule - Schedule the interrupt handler bottom half.
+ *
+ * @man: The command buffer manager.
+ */
+void vmw_cmdbuf_tasklet_schedule(struct vmw_cmdbuf_man *man)
+{
+	if (!man)
+		return;
+
+	tasklet_schedule(&man->tasklet);
+}
+
+/**
+ * vmw_cmdbuf_send_device_command - Send a command through the device context.
+ *
+ * @man: The command buffer manager.
+ * @command: Pointer to the command to send.
+ * @size: Size of the command.
+ *
+ * Synchronously sends a device context command.
+ */
+static int vmw_cmdbuf_send_device_command(struct vmw_cmdbuf_man *man,
+					  const void *command,
+					  size_t size)
+{
+	struct vmw_cmdbuf_header *header;
+	int status;
+	void *cmd = vmw_cmdbuf_alloc(man, size, false, &header);
+
+	if (IS_ERR(cmd))
+		return PTR_ERR(cmd);
+
+	memcpy(cmd, command, size);
+	header->cb_header->length = size;
+	header->cb_context = SVGA_CB_CONTEXT_DEVICE;
+	spin_lock_bh(&man->lock);
+	status = vmw_cmdbuf_header_submit(header);
+	spin_unlock_bh(&man->lock);
+	vmw_cmdbuf_header_free(header);
+
+	if (status != SVGA_CB_STATUS_COMPLETED) {
+		DRM_ERROR("Device context command failed with status %d\n",
+			  status);
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+/**
+ * vmw_cmdbuf_startstop - Send a start / stop command through the device
+ * context.
+ *
+ * @man: The command buffer manager.
+ * @enable: Whether to enable or disable the context.
+ *
+ * Synchronously sends a device start / stop context command.
+ */
+static int vmw_cmdbuf_startstop(struct vmw_cmdbuf_man *man,
+				bool enable)
+{
+	struct {
+		uint32 id;
+		SVGADCCmdStartStop body;
+	} __packed cmd;
+
+	cmd.id = SVGA_DC_CMD_START_STOP_CONTEXT;
+	cmd.body.enable = (enable) ? 1 : 0;
+	cmd.body.context = SVGA_CB_CONTEXT_0;
+
+	return vmw_cmdbuf_send_device_command(man, &cmd, sizeof(cmd));
+}
+
+/**
+ * vmw_cmdbuf_set_pool_size - Set command buffer manager sizes
+ *
+ * @man: The command buffer manager.
+ * @size: The size of the main space pool.
+ * @default_size: The default size of the command buffer for small kernel
+ * submissions.
+ *
+ * Set the size and allocate the main command buffer space pool,
+ * as well as the default size of the command buffer for
+ * small kernel submissions. If successful, this enables large command
+ * submissions. Note that this function requires that rudimentary command
+ * submission is already available and that the MOB memory manager is alive.
+ * Returns 0 on success. Negative error code on failure.
+ */
+int vmw_cmdbuf_set_pool_size(struct vmw_cmdbuf_man *man,
+			     size_t size, size_t default_size)
+{
+	struct vmw_private *dev_priv = man->dev_priv;
+	bool dummy;
+	int ret;
+
+	if (man->has_pool)
+		return -EINVAL;
+
+	/* First, try to allocate a huge chunk of DMA memory */
+	size = PAGE_ALIGN(size);
+	man->map = dma_alloc_coherent(&dev_priv->dev->pdev->dev, size,
+				      &man->handle, GFP_KERNEL);
+	if (man->map) {
+		man->using_mob = false;
+	} else {
+		/*
+		 * DMA memory failed. If we can have command buffers in a
+		 * MOB, try to use that instead. Note that this will
+		 * actually call into the already enabled manager, when
+		 * binding the MOB.
+		 */
+		if (!(dev_priv->capabilities & SVGA_CAP_CMD_BUFFERS_3))
+			return -ENOMEM;
+
+		ret = ttm_bo_create(&dev_priv->bdev, size, ttm_bo_type_device,
+				    &vmw_mob_ne_placement, 0, false, NULL,
+				    &man->cmd_space);
+		if (ret)
+			return ret;
+
+		man->using_mob = true;
+		ret = ttm_bo_kmap(man->cmd_space, 0, size >> PAGE_SHIFT,
+				  &man->map_obj);
+		if (ret)
+			goto out_no_map;
+
+		man->map = ttm_kmap_obj_virtual(&man->map_obj, &dummy);
+	}
+
+	man->size = size;
+	drm_mm_init(&man->mm, 0, size >> PAGE_SHIFT);
+
+	man->has_pool = true;
+	man->default_size = default_size;
+	DRM_INFO("Using command buffers with %s pool.\n",
+		 (man->using_mob) ? "MOB" : "DMA");
+
+	return 0;
+
+out_no_map:
+	if (man->using_mob)
+		ttm_bo_unref(&man->cmd_space);
+
+	return ret;
+}
+
+/**
+ * vmw_cmdbuf_man_create: Create a command buffer manager and enable it for
+ * inline command buffer submissions only.
+ *
+ * @dev_priv: Pointer to device private structure.
+ *
+ * Returns a pointer to a cummand buffer manager to success or error pointer
+ * on failure. The command buffer manager will be enabled for submissions of
+ * size VMW_CMDBUF_INLINE_SIZE only.
+ */
+struct vmw_cmdbuf_man *vmw_cmdbuf_man_create(struct vmw_private *dev_priv)
+{
+	struct vmw_cmdbuf_man *man;
+	struct vmw_cmdbuf_context *ctx;
+	int i;
+	int ret;
+
+	if (!(dev_priv->capabilities & SVGA_CAP_COMMAND_BUFFERS))
+		return ERR_PTR(-ENOSYS);
+
+	man = kzalloc(sizeof(*man), GFP_KERNEL);
+	if (!man)
+		return ERR_PTR(-ENOMEM);
+
+	man->headers = dma_pool_create("vmwgfx cmdbuf",
+				       &dev_priv->dev->pdev->dev,
+				       sizeof(SVGACBHeader),
+				       64, PAGE_SIZE);
+	if (!man->headers) {
+		ret = -ENOMEM;
+		goto out_no_pool;
+	}
+
+	man->dheaders = dma_pool_create("vmwgfx inline cmdbuf",
+					&dev_priv->dev->pdev->dev,
+					sizeof(struct vmw_cmdbuf_dheader),
+					64, PAGE_SIZE);
+	if (!man->dheaders) {
+		ret = -ENOMEM;
+		goto out_no_dpool;
+	}
+
+	for_each_cmdbuf_ctx(man, i, ctx)
+		vmw_cmdbuf_ctx_init(ctx);
+
+	INIT_LIST_HEAD(&man->error);
+	spin_lock_init(&man->lock);
+	mutex_init(&man->cur_mutex);
+	mutex_init(&man->space_mutex);
+	tasklet_init(&man->tasklet, vmw_cmdbuf_man_tasklet,
+		     (unsigned long) man);
+	man->default_size = VMW_CMDBUF_INLINE_SIZE;
+	init_waitqueue_head(&man->alloc_queue);
+	init_waitqueue_head(&man->idle_queue);
+	man->dev_priv = dev_priv;
+	man->max_hw_submitted = SVGA_CB_MAX_QUEUED_PER_CONTEXT - 1;
+	INIT_WORK(&man->work, &vmw_cmdbuf_work_func);
+	vmw_generic_waiter_add(dev_priv, SVGA_IRQFLAG_ERROR,
+			       &dev_priv->error_waiters);
+	ret = vmw_cmdbuf_startstop(man, true);
+	if (ret) {
+		DRM_ERROR("Failed starting command buffer context 0.\n");
+		vmw_cmdbuf_man_destroy(man);
+		return ERR_PTR(ret);
+	}
+
+	return man;
+
+out_no_dpool:
+	dma_pool_destroy(man->headers);
+out_no_pool:
+	kfree(man);
+
+	return ERR_PTR(ret);
+}
+
+/**
+ * vmw_cmdbuf_remove_pool - Take down the main buffer space pool.
+ *
+ * @man: Pointer to a command buffer manager.
+ *
+ * This function removes the main buffer space pool, and should be called
+ * before MOB memory management is removed. When this function has been called,
+ * only small command buffer submissions of size VMW_CMDBUF_INLINE_SIZE or
+ * less are allowed, and the default size of the command buffer for small kernel
+ * submissions is also set to this size.
+ */
+void vmw_cmdbuf_remove_pool(struct vmw_cmdbuf_man *man)
+{
+	if (!man->has_pool)
+		return;
+
+	man->has_pool = false;
+	man->default_size = VMW_CMDBUF_INLINE_SIZE;
+	(void) vmw_cmdbuf_idle(man, false, 10*HZ);
+	if (man->using_mob) {
+		(void) ttm_bo_kunmap(&man->map_obj);
+		ttm_bo_unref(&man->cmd_space);
+	} else {
+		dma_free_coherent(&man->dev_priv->dev->pdev->dev,
+				  man->size, man->map, man->handle);
+	}
+}
+
+/**
+ * vmw_cmdbuf_man_destroy - Take down a command buffer manager.
+ *
+ * @man: Pointer to a command buffer manager.
+ *
+ * This function idles and then destroys a command buffer manager.
+ */
+void vmw_cmdbuf_man_destroy(struct vmw_cmdbuf_man *man)
+{
+	WARN_ON_ONCE(man->has_pool);
+	(void) vmw_cmdbuf_idle(man, false, 10*HZ);
+	if (vmw_cmdbuf_startstop(man, false))
+		DRM_ERROR("Failed stopping command buffer context 0.\n");
+
+	vmw_generic_waiter_remove(man->dev_priv, SVGA_IRQFLAG_ERROR,
+				  &man->dev_priv->error_waiters);
+	tasklet_kill(&man->tasklet);
+	(void) cancel_work_sync(&man->work);
+	dma_pool_destroy(man->dheaders);
+	dma_pool_destroy(man->headers);
+	mutex_destroy(&man->cur_mutex);
+	mutex_destroy(&man->space_mutex);
+	kfree(man);
+}
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
index a4766ac..7e2b3c8 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
@@ -278,6 +278,8 @@ static void vmw_print_capabilities(uint32_t capabilities)
 		DRM_INFO("  Command Buffers 2.\n");
 	if (capabilities & SVGA_CAP_GBOBJECTS)
 		DRM_INFO("  Guest Backed Resources.\n");
+	if (capabilities & SVGA_CAP_CMD_BUFFERS_3)
+		DRM_INFO("  Command Buffers 3.\n");
 }
 
 /**
@@ -362,6 +364,17 @@ static int vmw_request_device_late(struct vmw_private *dev_priv)
 		}
 	}
 
+	if (dev_priv->cman) {
+		ret = vmw_cmdbuf_set_pool_size(dev_priv->cman,
+					       256*4096, 2*4096);
+		if (ret) {
+			struct vmw_cmdbuf_man *man = dev_priv->cman;
+
+			dev_priv->cman = NULL;
+			vmw_cmdbuf_man_destroy(man);
+		}
+	}
+
 	return 0;
 }
 
@@ -375,6 +388,9 @@ static int vmw_request_device(struct vmw_private *dev_priv)
 		return ret;
 	}
 	vmw_fence_fifo_up(dev_priv->fman);
+	dev_priv->cman = vmw_cmdbuf_man_create(dev_priv);
+	if (IS_ERR(dev_priv->cman))
+		dev_priv->cman = NULL;
 
 	ret = vmw_request_device_late(dev_priv);
 	if (ret)
@@ -387,10 +403,14 @@ static int vmw_request_device(struct vmw_private *dev_priv)
 	return 0;
 
 out_no_query_bo:
+	if (dev_priv->cman)
+		vmw_cmdbuf_remove_pool(dev_priv->cman);
 	if (dev_priv->has_mob) {
 		(void) ttm_bo_evict_mm(&dev_priv->bdev, VMW_PL_MOB);
 		vmw_otables_takedown(dev_priv);
 	}
+	if (dev_priv->cman)
+		vmw_cmdbuf_man_destroy(dev_priv->cman);
 out_no_mob:
 	vmw_fence_fifo_down(dev_priv->fman);
 	vmw_fifo_release(dev_priv, &dev_priv->fifo);
@@ -415,6 +435,9 @@ static void vmw_release_device_early(struct vmw_private *dev_priv)
 	BUG_ON(dev_priv->pinned_bo != NULL);
 
 	ttm_bo_unref(&dev_priv->dummy_query_bo);
+	if (dev_priv->cman)
+		vmw_cmdbuf_remove_pool(dev_priv->cman);
+
 	if (dev_priv->has_mob) {
 		ttm_bo_evict_mm(&dev_priv->bdev, VMW_PL_MOB);
 		vmw_otables_takedown(dev_priv);
@@ -432,6 +455,9 @@ static void vmw_release_device_early(struct vmw_private *dev_priv)
 static void vmw_release_device_late(struct vmw_private *dev_priv)
 {
 	vmw_fence_fifo_down(dev_priv->fman);
+	if (dev_priv->cman)
+		vmw_cmdbuf_man_destroy(dev_priv->cman);
+
 	vmw_fifo_release(dev_priv, &dev_priv->fifo);
 }
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index a5f221e..8fd40c6 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -453,6 +453,8 @@ struct vmw_private {
 	spinlock_t waiter_lock;
 	int fence_queue_waiters; /* Protected by waiter_lock */
 	int goal_queue_waiters; /* Protected by waiter_lock */
+	int cmdbuf_waiters; /* Protected by irq_lock */
+	int error_waiters; /* Protected by irq_lock */
 	atomic_t fifo_queue_waiters;
 	uint32_t last_read_seqno;
 	spinlock_t irq_lock;
@@ -535,6 +537,8 @@ struct vmw_private {
 	 */
 	struct ttm_buffer_object *otable_bo;
 	struct vmw_otable *otables;
+
+	struct vmw_cmdbuf_man *cman;
 };
 
 static inline struct vmw_surface *vmw_res_to_srf(struct vmw_resource *res)
@@ -729,6 +733,8 @@ extern bool vmw_fifo_have_3d(struct vmw_private *dev_priv);
 extern bool vmw_fifo_have_pitchlock(struct vmw_private *dev_priv);
 extern int vmw_fifo_emit_dummy_query(struct vmw_private *dev_priv,
 				     uint32_t cid);
+extern int vmw_fifo_flush(struct vmw_private *dev_priv,
+			  bool interruptible);
 
 /**
  * TTM glue - vmwgfx_ttm_glue.c
@@ -753,6 +759,7 @@ extern struct ttm_placement vmw_sys_ne_placement;
 extern struct ttm_placement vmw_evictable_placement;
 extern struct ttm_placement vmw_srf_placement;
 extern struct ttm_placement vmw_mob_placement;
+extern struct ttm_placement vmw_mob_ne_placement;
 extern struct ttm_bo_driver vmw_bo_driver;
 extern int vmw_dma_quiescent(struct drm_device *dev);
 extern int vmw_bo_map_dma(struct ttm_buffer_object *bo);
@@ -855,6 +862,10 @@ extern void vmw_seqno_waiter_add(struct vmw_private *dev_priv);
 extern void vmw_seqno_waiter_remove(struct vmw_private *dev_priv);
 extern void vmw_goal_waiter_add(struct vmw_private *dev_priv);
 extern void vmw_goal_waiter_remove(struct vmw_private *dev_priv);
+extern void vmw_generic_waiter_add(struct vmw_private *dev_priv, u32 flag,
+				   int *waiter_count);
+extern void vmw_generic_waiter_remove(struct vmw_private *dev_priv,
+				      u32 flag, int *waiter_count);
 
 /**
  * Rudimentary fence-like objects currently used only for throttling -
@@ -1077,6 +1088,35 @@ extern int vmw_cmdbuf_res_remove(struct vmw_cmdbuf_res_manager *man,
 				 struct list_head *list);
 
 
+/*
+ * Command buffer managerment vmwgfx_cmdbuf.c
+ */
+struct vmw_cmdbuf_man;
+struct vmw_cmdbuf_header;
+
+extern struct vmw_cmdbuf_man *
+vmw_cmdbuf_man_create(struct vmw_private *dev_priv);
+extern int vmw_cmdbuf_set_pool_size(struct vmw_cmdbuf_man *man,
+				    size_t size, size_t default_size);
+extern void vmw_cmdbuf_remove_pool(struct vmw_cmdbuf_man *man);
+extern void vmw_cmdbuf_man_destroy(struct vmw_cmdbuf_man *man);
+extern int vmw_cmdbuf_idle(struct vmw_cmdbuf_man *man, bool interruptible,
+			   unsigned long timeout);
+extern void *vmw_cmdbuf_reserve(struct vmw_cmdbuf_man *man, size_t size,
+				int ctx_id, bool interruptible,
+				struct vmw_cmdbuf_header *header);
+extern void vmw_cmdbuf_commit(struct vmw_cmdbuf_man *man, size_t size,
+			      struct vmw_cmdbuf_header *header,
+			      bool flush);
+extern void vmw_cmdbuf_tasklet_schedule(struct vmw_cmdbuf_man *man);
+extern void *vmw_cmdbuf_alloc(struct vmw_cmdbuf_man *man,
+			      size_t size, bool interruptible,
+			      struct vmw_cmdbuf_header **p_header);
+extern void vmw_cmdbuf_header_free(struct vmw_cmdbuf_header *header);
+extern int vmw_cmdbuf_cur_flush(struct vmw_cmdbuf_man *man,
+				bool interruptible);
+
+
 /**
  * Inline helper functions
  */
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
index 654c8da..0792d8d 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
@@ -2417,7 +2417,126 @@ vmw_execbuf_copy_fence_user(struct vmw_private *dev_priv,
 	}
 }
 
+/**
+ * vmw_execbuf_submit_fifo - Patch a command batch and submit it using
+ * the fifo.
+ *
+ * @dev_priv: Pointer to a device private structure.
+ * @kernel_commands: Pointer to the unpatched command batch.
+ * @command_size: Size of the unpatched command batch.
+ * @sw_context: Structure holding the relocation lists.
+ *
+ * Side effects: If this function returns 0, then the command batch
+ * pointed to by @kernel_commands will have been modified.
+ */
+static int vmw_execbuf_submit_fifo(struct vmw_private *dev_priv,
+				   void *kernel_commands,
+				   u32 command_size,
+				   struct vmw_sw_context *sw_context)
+{
+	void *cmd = vmw_fifo_reserve(dev_priv, command_size);
+
+	if (!cmd) {
+		DRM_ERROR("Failed reserving fifo space for commands.\n");
+		return -ENOMEM;
+	}
+
+	vmw_apply_relocations(sw_context);
+	memcpy(cmd, kernel_commands, command_size);
+	vmw_resource_relocations_apply(cmd, &sw_context->res_relocations);
+	vmw_resource_relocations_free(&sw_context->res_relocations);
+	vmw_fifo_commit(dev_priv, command_size);
+
+	return 0;
+}
 
+/**
+ * vmw_execbuf_submit_cmdbuf - Patch a command batch and submit it using
+ * the command buffer manager.
+ *
+ * @dev_priv: Pointer to a device private structure.
+ * @header: Opaque handle to the command buffer allocation.
+ * @command_size: Size of the unpatched command batch.
+ * @sw_context: Structure holding the relocation lists.
+ *
+ * Side effects: If this function returns 0, then the command buffer
+ * represented by @header will have been modified.
+ */
+static int vmw_execbuf_submit_cmdbuf(struct vmw_private *dev_priv,
+				     struct vmw_cmdbuf_header *header,
+				     u32 command_size,
+				     struct vmw_sw_context *sw_context)
+{
+	void *cmd = vmw_cmdbuf_reserve(dev_priv->cman, command_size,
+				       SVGA3D_INVALID_ID, false, header);
+
+	vmw_apply_relocations(sw_context);
+	vmw_resource_relocations_apply(cmd, &sw_context->res_relocations);
+	vmw_resource_relocations_free(&sw_context->res_relocations);
+	vmw_cmdbuf_commit(dev_priv->cman, command_size, header, false);
+
+	return 0;
+}
+
+/**
+ * vmw_execbuf_cmdbuf - Prepare, if possible, a user-space command batch for
+ * submission using a command buffer.
+ *
+ * @dev_priv: Pointer to a device private structure.
+ * @user_commands: User-space pointer to the commands to be submitted.
+ * @command_size: Size of the unpatched command batch.
+ * @header: Out parameter returning the opaque pointer to the command buffer.
+ *
+ * This function checks whether we can use the command buffer manager for
+ * submission and if so, creates a command buffer of suitable size and
+ * copies the user data into that buffer.
+ *
+ * On successful return, the function returns a pointer to the data in the
+ * command buffer and *@header is set to non-NULL.
+ * If command buffers could not be used, the function will return the value
+ * of @kernel_commands on function call. That value may be NULL. In that case,
+ * the value of *@header will be set to NULL.
+ * If an error is encountered, the function will return a pointer error value.
+ * If the function is interrupted by a signal while sleeping, it will return
+ * -ERESTARTSYS casted to a pointer error value.
+ */
+void *vmw_execbuf_cmdbuf(struct vmw_private *dev_priv,
+			 void __user *user_commands,
+			 void *kernel_commands,
+			 u32 command_size,
+			 struct vmw_cmdbuf_header **header)
+{
+	size_t cmdbuf_size;
+	int ret;
+
+	*header = NULL;
+	if (!dev_priv->cman || kernel_commands)
+		return kernel_commands;
+
+	if (command_size > SVGA_CB_MAX_SIZE) {
+		DRM_ERROR("Command buffer is too large.\n");
+		return ERR_PTR(-EINVAL);
+	}
+
+	/* If possible, add a little space for fencing. */
+	cmdbuf_size = command_size + 512;
+	cmdbuf_size = min_t(size_t, cmdbuf_size, SVGA_CB_MAX_SIZE);
+	kernel_commands = vmw_cmdbuf_alloc(dev_priv->cman, cmdbuf_size,
+					   true, header);
+	if (IS_ERR(kernel_commands))
+		return kernel_commands;
+
+	ret = copy_from_user(kernel_commands, user_commands,
+			     command_size);
+	if (ret) {
+		DRM_ERROR("Failed copying commands.\n");
+		vmw_cmdbuf_header_free(*header);
+		*header = NULL;
+		return ERR_PTR(-EFAULT);
+	}
+
+	return kernel_commands;
+}
 
 int vmw_execbuf_process(struct drm_file *file_priv,
 			struct vmw_private *dev_priv,
@@ -2432,18 +2551,33 @@ int vmw_execbuf_process(struct drm_file *file_priv,
 	struct vmw_fence_obj *fence = NULL;
 	struct vmw_resource *error_resource;
 	struct list_head resource_list;
+	struct vmw_cmdbuf_header *header;
 	struct ww_acquire_ctx ticket;
 	uint32_t handle;
-	void *cmd;
 	int ret;
 
+     	if (throttle_us) {
+		ret = vmw_wait_lag(dev_priv, &dev_priv->fifo.marker_queue,
+				   throttle_us);
+		
+		if (ret)
+			return ret;
+	}
+	
+	kernel_commands = vmw_execbuf_cmdbuf(dev_priv, user_commands,
+					     kernel_commands, command_size,
+					     &header);
+	if (IS_ERR(kernel_commands))
+		return PTR_ERR(kernel_commands);
+
 	ret = mutex_lock_interruptible(&dev_priv->cmdbuf_mutex);
-	if (unlikely(ret != 0))
-		return -ERESTARTSYS;
+	if (ret) {
+		ret = -ERESTARTSYS;
+		goto out_free_header;
+	}
 
+	sw_context->kernel = false;
 	if (kernel_commands == NULL) {
-		sw_context->kernel = false;
-
 		ret = vmw_resize_cmd_bounce(sw_context, command_size);
 		if (unlikely(ret != 0))
 			goto out_unlock;
@@ -2458,7 +2592,7 @@ int vmw_execbuf_process(struct drm_file *file_priv,
 			goto out_unlock;
 		}
 		kernel_commands = sw_context->cmd_bounce;
-	} else
+	} else if (!header)
 		sw_context->kernel = true;
 
 	sw_context->fp = vmw_fpriv(file_priv);
@@ -2478,7 +2612,6 @@ int vmw_execbuf_process(struct drm_file *file_priv,
 		sw_context->res_ht_initialized = true;
 	}
 	INIT_LIST_HEAD(&sw_context->staged_cmd_res);
-
 	INIT_LIST_HEAD(&resource_list);
 	ret = vmw_cmd_check_all(dev_priv, sw_context, kernel_commands,
 				command_size);
@@ -2502,14 +2635,6 @@ int vmw_execbuf_process(struct drm_file *file_priv,
 	if (unlikely(ret != 0))
 		goto out_err;
 
-	if (throttle_us) {
-		ret = vmw_wait_lag(dev_priv, &dev_priv->fifo.marker_queue,
-				   throttle_us);
-
-		if (unlikely(ret != 0))
-			goto out_err;
-	}
-
 	ret = mutex_lock_interruptible(&dev_priv->binding_mutex);
 	if (unlikely(ret != 0)) {
 		ret = -ERESTARTSYS;
@@ -2522,20 +2647,16 @@ int vmw_execbuf_process(struct drm_file *file_priv,
 			goto out_unlock_binding;
 	}
 
-	cmd = vmw_fifo_reserve(dev_priv, command_size);
-	if (unlikely(cmd == NULL)) {
-		DRM_ERROR("Failed reserving fifo space for commands.\n");
-		ret = -ENOMEM;
-		goto out_unlock_binding;
+	if (!header) {
+		ret = vmw_execbuf_submit_fifo(dev_priv, kernel_commands,
+					      command_size, sw_context);
+	} else {
+		ret = vmw_execbuf_submit_cmdbuf(dev_priv, header, command_size,
+						sw_context);
+		header = NULL;
 	}
-
-	vmw_apply_relocations(sw_context);
-	memcpy(cmd, kernel_commands, command_size);
-
-	vmw_resource_relocations_apply(cmd, &sw_context->res_relocations);
-	vmw_resource_relocations_free(&sw_context->res_relocations);
-
-	vmw_fifo_commit(dev_priv, command_size);
+	if (ret)
+		goto out_unlock_binding;
 
 	vmw_query_bo_switch_commit(dev_priv, sw_context);
 	ret = vmw_execbuf_fence_commands(file_priv, dev_priv,
@@ -2610,6 +2731,9 @@ out_unlock:
 	vmw_resource_list_unreference(&resource_list);
 	if (unlikely(error_resource != NULL))
 		vmw_resource_unreference(&error_resource);
+out_free_header:
+	if (header)
+		vmw_cmdbuf_header_free(header);
 
 	return ret;
 }
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
index ecdc8d9..d0a3bcf 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
@@ -257,6 +257,7 @@ static void vmw_fb_dirty_flush(struct vmw_fb_par *par)
 	cmd->body.width = cpu_to_le32(w);
 	cmd->body.height = cpu_to_le32(h);
 	vmw_fifo_commit(vmw_priv, sizeof(*cmd));
+	vmw_fifo_flush(vmw_priv, false);
 }
 
 static void vmw_fb_dirty_mark(struct vmw_fb_par *par,
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
index cd5d9f3..189102d 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
@@ -310,7 +310,8 @@ static int vmw_fifo_wait(struct vmw_private *dev_priv,
  * Returns:
  *   Pointer to the fifo, or null on error (possible hardware hang).
  */
-void *vmw_fifo_reserve(struct vmw_private *dev_priv, uint32_t bytes)
+static void *vmw_local_fifo_reserve(struct vmw_private *dev_priv,
+				    uint32_t bytes)
 {
 	struct vmw_fifo_state *fifo_state = &dev_priv->fifo;
 	__le32 __iomem *fifo_mem = dev_priv->mmio_virt;
@@ -389,9 +390,29 @@ void *vmw_fifo_reserve(struct vmw_private *dev_priv, uint32_t bytes)
 out_err:
 	fifo_state->reserved_size = 0;
 	mutex_unlock(&fifo_state->fifo_mutex);
+
 	return NULL;
 }
 
+void *vmw_fifo_reserve(struct vmw_private *dev_priv, uint32_t bytes)
+{
+	void *ret;
+
+	if (dev_priv->cman)
+		ret = vmw_cmdbuf_reserve(dev_priv->cman, bytes,
+					 SVGA3D_INVALID_ID, false, NULL);
+	else
+		ret = vmw_local_fifo_reserve(dev_priv, bytes);
+	if (IS_ERR_OR_NULL(ret)) {
+		DRM_ERROR("Fifo reserve failure of %u bytes.\n",
+			  (unsigned) bytes);
+		dump_stack();
+		return NULL;
+	}
+
+	return ret;
+}
+
 static void vmw_fifo_res_copy(struct vmw_fifo_state *fifo_state,
 			      __le32 __iomem *fifo_mem,
 			      uint32_t next_cmd,
@@ -434,7 +455,7 @@ static void vmw_fifo_slow_copy(struct vmw_fifo_state *fifo_state,
 	}
 }
 
-void vmw_fifo_commit(struct vmw_private *dev_priv, uint32_t bytes)
+void vmw_local_fifo_commit(struct vmw_private *dev_priv, uint32_t bytes)
 {
 	struct vmw_fifo_state *fifo_state = &dev_priv->fifo;
 	__le32 __iomem *fifo_mem = dev_priv->mmio_virt;
@@ -480,6 +501,46 @@ void vmw_fifo_commit(struct vmw_private *dev_priv, uint32_t bytes)
 	mutex_unlock(&fifo_state->fifo_mutex);
 }
 
+void vmw_fifo_commit(struct vmw_private *dev_priv, uint32_t bytes)
+{
+	if (dev_priv->cman)
+		vmw_cmdbuf_commit(dev_priv->cman, bytes, NULL, false);
+	else
+		vmw_local_fifo_commit(dev_priv, bytes);
+}
+
+
+/**
+ * vmw_fifo_commit_flush - Commit fifo space and flush any buffered commands.
+ *
+ * @dev_priv: Pointer to device private structure.
+ * @bytes: Number of bytes to commit.
+ */
+static void vmw_fifo_commit_flush(struct vmw_private *dev_priv, uint32_t bytes)
+{
+	if (dev_priv->cman)
+		vmw_cmdbuf_commit(dev_priv->cman, bytes, NULL, true);
+	else
+		vmw_local_fifo_commit(dev_priv, bytes);
+}
+
+/**
+ * vmw_fifo_flush - Flush any buffered commands and make sure command processing
+ * starts.
+ *
+ * @dev_priv: Pointer to device private structure.
+ * @interruptible: Whether to wait interruptible if function needs to sleep.
+ */
+int vmw_fifo_flush(struct vmw_private *dev_priv, bool interruptible)
+{
+	might_sleep();
+
+	if (dev_priv->cman)
+		return vmw_cmdbuf_cur_flush(dev_priv->cman, interruptible);
+	else
+		return 0;
+}
+
 int vmw_fifo_send_fence(struct vmw_private *dev_priv, uint32_t *seqno)
 {
 	struct vmw_fifo_state *fifo_state = &dev_priv->fifo;
@@ -517,7 +578,7 @@ int vmw_fifo_send_fence(struct vmw_private *dev_priv, uint32_t *seqno)
 	    ((unsigned long)fm + sizeof(__le32));
 
 	iowrite32(*seqno, &cmd_fence->fence);
-	vmw_fifo_commit(dev_priv, bytes);
+	vmw_fifo_commit_flush(dev_priv, bytes);
 	(void) vmw_marker_push(&fifo_state->marker_queue, *seqno);
 	vmw_update_seqno(dev_priv, fifo_state);
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c b/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c
index 9fe9827..87964bb 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c
@@ -56,6 +56,9 @@ irqreturn_t vmw_irq_handler(int irq, void *arg)
 	if (masked_status & SVGA_IRQFLAG_FIFO_PROGRESS)
 		wake_up_all(&dev_priv->fifo_queue);
 
+	if (masked_status & (SVGA_IRQFLAG_COMMAND_BUFFER |
+			     SVGA_IRQFLAG_ERROR))
+		vmw_cmdbuf_tasklet_schedule(dev_priv->cman);
 
 	return IRQ_HANDLED;
 }
@@ -131,8 +134,16 @@ int vmw_fallback_wait(struct vmw_private *dev_priv,
 	 * Block command submission while waiting for idle.
 	 */
 
-	if (fifo_idle)
+	if (fifo_idle) {
 		down_read(&fifo_state->rwsem);
+		if (dev_priv->cman) {
+			ret = vmw_cmdbuf_idle(dev_priv->cman, interruptible,
+					      10*HZ);
+			if (ret)
+				goto out_err;
+		}
+	}
+
 	signal_seq = atomic_read(&dev_priv->marker_seq);
 	ret = 0;
 
@@ -171,6 +182,7 @@ int vmw_fallback_wait(struct vmw_private *dev_priv,
 		iowrite32(signal_seq, fifo_mem + SVGA_FIFO_FENCE);
 	}
 	wake_up_all(&dev_priv->fence_queue);
+out_err:
 	if (fifo_idle)
 		up_read(&fifo_state->rwsem);
 
@@ -315,3 +327,30 @@ void vmw_irq_uninstall(struct drm_device *dev)
 	status = inl(dev_priv->io_start + VMWGFX_IRQSTATUS_PORT);
 	outl(status, dev_priv->io_start + VMWGFX_IRQSTATUS_PORT);
 }
+
+void vmw_generic_waiter_add(struct vmw_private *dev_priv,
+			    u32 flag, int *waiter_count)
+{
+	unsigned long irq_flags;
+
+	spin_lock_irqsave(&dev_priv->irq_lock, irq_flags);
+	if ((*waiter_count)++ == 0) {
+		outl(flag, dev_priv->io_start + VMWGFX_IRQSTATUS_PORT);
+		dev_priv->irq_mask |= flag;
+		vmw_write(dev_priv, SVGA_REG_IRQMASK, dev_priv->irq_mask);
+	}
+	spin_unlock_irqrestore(&dev_priv->irq_lock, irq_flags);
+}
+
+void vmw_generic_waiter_remove(struct vmw_private *dev_priv,
+			       u32 flag, int *waiter_count)
+{
+	unsigned long irq_flags;
+
+	spin_lock_irqsave(&dev_priv->irq_lock, irq_flags);
+	if (--(*waiter_count) == 0) {
+		dev_priv->irq_mask &= ~flag;
+		vmw_write(dev_priv, SVGA_REG_IRQMASK, dev_priv->irq_mask);
+	}
+	spin_unlock_irqrestore(&dev_priv->irq_lock, irq_flags);
+}
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
index 07cda8c..b5632c2 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
@@ -631,6 +631,7 @@ static int vmw_framebuffer_surface_dirty(struct drm_framebuffer *framebuffer,
 				   flags, color,
 				   clips, num_clips, inc, NULL);
 
+	vmw_fifo_flush(dev_priv, false);
 	ttm_read_unlock(&dev_priv->reservation_sem);
 
 	drm_modeset_unlock_all(dev_priv->dev);
@@ -987,6 +988,7 @@ static int vmw_framebuffer_dmabuf_dirty(struct drm_framebuffer *framebuffer,
 					  clips, num_clips, increment, NULL);
 	}
 
+	vmw_fifo_flush(dev_priv, false);
 	ttm_read_unlock(&dev_priv->reservation_sem);
 
 	drm_modeset_unlock_all(dev_priv->dev);
@@ -1347,6 +1349,8 @@ int vmw_kms_present(struct vmw_private *dev_priv,
 			break;
 	}
 
+	vmw_fifo_flush(dev_priv, false);
+
 	kfree(cmd);
 out_free_tmp:
 	kfree(tmp);
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 06/28] drm/vmwgfx: Fix an overlay lockdep error

[Thomas Hellstrom]

Fix a circular locking dependency between
struct vmw_overlay::mutex and
struct vmw_private::reservation_sem

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_resource.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
index 210ef15..3fd8070 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
@@ -900,20 +900,21 @@ int vmw_stream_claim_ioctl(struct drm_device *dev, void *data,
 	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
 				   vmw_user_stream_size,
 				   false, true);
+	ttm_read_unlock(&dev_priv->reservation_sem);
 	if (unlikely(ret != 0)) {
 		if (ret != -ERESTARTSYS)
 			DRM_ERROR("Out of graphics memory for stream"
 				  " creation.\n");
-		goto out_unlock;
-	}
 
+		goto out_ret;
+	}
 
 	stream = kmalloc(sizeof(*stream), GFP_KERNEL);
 	if (unlikely(stream == NULL)) {
 		ttm_mem_global_free(vmw_mem_glob(dev_priv),
 				    vmw_user_stream_size);
 		ret = -ENOMEM;
-		goto out_unlock;
+		goto out_ret;
 	}
 
 	res = &stream->stream.res;
@@ -926,7 +927,7 @@ int vmw_stream_claim_ioctl(struct drm_device *dev, void *data,
 
 	ret = vmw_stream_init(dev_priv, &stream->stream, vmw_user_stream_free);
 	if (unlikely(ret != 0))
-		goto out_unlock;
+		goto out_ret;
 
 	tmp = vmw_resource_reference(res);
 	ret = ttm_base_object_init(tfile, &stream->base, false, VMW_RES_STREAM,
@@ -940,8 +941,7 @@ int vmw_stream_claim_ioctl(struct drm_device *dev, void *data,
 	arg->stream_id = res->id;
 out_err:
 	vmw_resource_unreference(&res);
-out_unlock:
-	ttm_read_unlock(&dev_priv->reservation_sem);
+out_ret:
 	return ret;
 }
 
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 07/28] drm/vmwgfx: Add an interface to pin a resource v3

[Thomas Hellstrom]

For screen targets it appears we need to pin surfaces while they are bound
as screen targets, so add a small interface to do that.

v2: Always increase pin_count on pin.
v3: Add missing reservation sem.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.h      |  4 ++
 drivers/gpu/drm/vmwgfx/vmwgfx_resource.c | 91 +++++++++++++++++++++++++++++++-
 2 files changed, 94 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index 8fd40c6..338dce3 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -113,6 +113,7 @@ struct vmw_resource {
 	bool backup_dirty; /* Protected by backup buffer reserved */
 	struct vmw_dma_buffer *backup;
 	unsigned long backup_offset;
+	unsigned long pin_count; /* Protected by resource reserved */
 	const struct vmw_res_func *func;
 	struct list_head lru_head; /* Protected by the resource lock */
 	struct list_head mob_head; /* Protected by @backup reserved */
@@ -941,6 +942,9 @@ int vmw_dumb_map_offset(struct drm_file *file_priv,
 int vmw_dumb_destroy(struct drm_file *file_priv,
 		     struct drm_device *dev,
 		     uint32_t handle);
+extern int vmw_resource_pin(struct vmw_resource *res);
+extern void vmw_resource_unpin(struct vmw_resource *res);
+
 /**
  * Overlay control - vmwgfx_overlay.c
  */
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
index 3fd8070..6738c1e 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
@@ -1183,7 +1183,7 @@ void vmw_resource_unreserve(struct vmw_resource *res,
 	if (new_backup)
 		res->backup_offset = new_backup_offset;
 
-	if (!res->func->may_evict || res->id == -1)
+	if (!res->func->may_evict || res->id == -1 || res->pin_count)
 		return;
 
 	write_lock(&dev_priv->resource_lock);
@@ -1573,3 +1573,92 @@ void vmw_resource_evict_all(struct vmw_private *dev_priv)
 
 	mutex_unlock(&dev_priv->cmdbuf_mutex);
 }
+
+/**
+ * vmw_resource_pin - Add a pin reference on a resource
+ *
+ * @res: The resource to add a pin reference on
+ *
+ * This function adds a pin reference, and if needed validates the resource.
+ * Having a pin reference means that the resource can never be evicted, and
+ * its id will never change as long as there is a pin reference.
+ * This function returns 0 on success and a negative error code on failure.
+ */
+int vmw_resource_pin(struct vmw_resource *res)
+{
+	struct vmw_private *dev_priv = res->dev_priv;
+	int ret;
+
+	ttm_write_lock(&dev_priv->reservation_sem, false);
+	mutex_lock(&dev_priv->cmdbuf_mutex);
+	ret = vmw_resource_reserve(res, false);
+	if (ret)
+		goto out_no_reserve;
+
+	if (res->pin_count == 0) {
+		struct ttm_buffer_object *bo = NULL;
+
+		if (res->backup) {
+			bo = &res->backup->base;
+
+			ttm_bo_reserve(bo, false, false, false, NULL);
+			ret = ttm_bo_validate(bo, res->func->backup_placement,
+					      false, false);
+			if (ret) {
+				ttm_bo_unreserve(bo);
+				goto out_no_validate;
+			}
+
+			/* Do we really need to pin the MOB as well? */
+			vmw_bo_pin(bo, true);
+		}
+		ret = vmw_resource_validate(res);
+		if (bo)
+			ttm_bo_unreserve(bo);
+		if (ret)
+			goto out_no_validate;
+	}
+	res->pin_count++;
+
+out_no_validate:
+	vmw_resource_unreserve(res, NULL, 0UL);
+out_no_reserve:
+	mutex_unlock(&dev_priv->cmdbuf_mutex);
+	ttm_write_unlock(&dev_priv->reservation_sem);
+
+	return ret;
+}
+
+/**
+ * vmw_resource_unpin - Remove a pin reference from a resource
+ *
+ * @res: The resource to remove a pin reference from
+ *
+ * Having a pin reference means that the resource can never be evicted, and
+ * its id will never change as long as there is a pin reference.
+ */
+void vmw_resource_unpin(struct vmw_resource *res)
+{
+	struct vmw_private *dev_priv = res->dev_priv;
+	int ret;
+
+	ttm_read_lock(&dev_priv->reservation_sem, false);
+	mutex_lock(&dev_priv->cmdbuf_mutex);
+
+	ret = vmw_resource_reserve(res, true);
+	WARN_ON(ret);
+
+	WARN_ON(res->pin_count == 0);
+	if (--res->pin_count == 0 && res->backup) {
+		struct ttm_buffer_object *bo = &res->backup->base;
+
+		ttm_bo_reserve(bo, false, false, false, NULL);
+		vmw_bo_pin(bo, false);
+		ttm_bo_unreserve(bo);
+	}
+
+	vmw_resource_unreserve(res, NULL, 0UL);
+
+	mutex_unlock(&dev_priv->cmdbuf_mutex);
+	ttm_read_unlock(&dev_priv->reservation_sem);
+}
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 08/28] drm/vmwgfx: SVGA device definition update

[Thomas Hellstrom]

From: Sinclair Yeh <syeh@vmware.com>

Update device definition headers to support screen targets.

Signed-off-by: Sinclair Yeh <syeh@vmware.com>
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
---
 drivers/gpu/drm/vmwgfx/svga3d_reg.h         | 56 ++++++++++++++++++++++--
 drivers/gpu/drm/vmwgfx/svga3d_surfacedefs.h | 67 +++++++++++++++++++++++++++--
 2 files changed, 117 insertions(+), 6 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/svga3d_reg.h b/drivers/gpu/drm/vmwgfx/svga3d_reg.h
index e50d20c..c9a595a 100644
--- a/drivers/gpu/drm/vmwgfx/svga3d_reg.h
+++ b/drivers/gpu/drm/vmwgfx/svga3d_reg.h
@@ -1,5 +1,5 @@
 /**********************************************************
- * Copyright 1998-2009 VMware, Inc.  All rights reserved.
+ * Copyright 1998-2014 VMware, Inc.  All rights reserved.
  *
  * Permission is hereby granted, free of charge, to any person
  * obtaining a copy of this software and associated documentation
@@ -224,7 +224,7 @@ typedef enum SVGA3dSurfaceFormat {
    SVGA3D_R8_SNORM                     = 95,
    SVGA3D_R8_SINT                      = 96,
    SVGA3D_A8_UNORM                     = 32,
-   SVGA3D_R1_UNORM                     = 97,
+   SVGA3D_P8                           = 97,
    SVGA3D_R9G9B9E5_SHAREDEXP           = 98,
    SVGA3D_R8G8_B8G8_UNORM              = 99,
    SVGA3D_G8R8_G8B8_UNORM              = 100,
@@ -1312,6 +1312,11 @@ struct {
 
 typedef enum {
    SVGA3D_SURFACE_CUBEMAP              = (1 << 0),
+
+   /*
+    * HINT flags are not enforced by the device but are useful for
+    * performance.
+    */
    SVGA3D_SURFACE_HINT_STATIC          = (1 << 1),
    SVGA3D_SURFACE_HINT_DYNAMIC         = (1 << 2),
    SVGA3D_SURFACE_HINT_INDEXBUFFER     = (1 << 3),
@@ -1322,6 +1327,50 @@ typedef enum {
    SVGA3D_SURFACE_HINT_WRITEONLY       = (1 << 8),
    SVGA3D_SURFACE_MASKABLE_ANTIALIAS   = (1 << 9),
    SVGA3D_SURFACE_AUTOGENMIPMAPS       = (1 << 10),
+   SVGA3D_SURFACE_DECODE_RENDERTARGET   = (1 << 11),
+
+   /*
+    * Is this surface using a base-level pitch for it's mob backing?
+    *
+    * This flag is not intended to be set by guest-drivers, but is instead
+    * set by the device when the surface is bound to a mob with a specified
+    * pitch.
+    */
+   SVGA3D_SURFACE_MOB_PITCH             = (1 << 12),
+
+   SVGA3D_SURFACE_INACTIVE              = (1 << 13),
+   SVGA3D_SURFACE_HINT_RT_LOCKABLE      = (1 << 14),
+   SVGA3D_SURFACE_VOLUME                = (1 << 15),
+
+   /*
+    * Required to be set on a surface to bind it to a screen target.
+    */
+   SVGA3D_SURFACE_SCREENTARGET          = (1 << 16),
+
+   /*
+    * Align images in the guest-backing mob to 16-bytes.
+    */
+   SVGA3D_SURFACE_ALIGN16               = (1 << 17),
+
+   SVGA3D_SURFACE_1D                    = (1 << 18),
+   SVGA3D_SURFACE_ARRAY                 = (1 << 19),
+
+   /*
+    * Bind flags.
+    * These are enforced for any surface defined with DefineGBSurface_v2.
+    */
+   SVGA3D_SURFACE_BIND_VERTEX_BUFFER    = (1 << 20),
+   SVGA3D_SURFACE_BIND_INDEX_BUFFER     = (1 << 21),
+   SVGA3D_SURFACE_BIND_CONSTANT_BUFFER  = (1 << 22),
+   SVGA3D_SURFACE_BIND_SHADER_RESOURCE  = (1 << 23),
+   SVGA3D_SURFACE_BIND_RENDER_TARGET    = (1 << 24),
+   SVGA3D_SURFACE_BIND_DEPTH_STENCIL    = (1 << 25),
+   SVGA3D_SURFACE_BIND_STREAM_OUTPUT    = (1 << 26),
+
+   /*
+    * Marker for the last defined bit.
+    */
+   SVGA3D_SURFACE_FLAG_MAX              = (1 << 27),
 } SVGA3dSurfaceFlags;
 
 typedef
@@ -2400,6 +2449,7 @@ struct {
    int32 xRoot;
    int32 yRoot;
    uint32 flags;
+   uint32 dpi;
 } __packed
 SVGA3dCmdDefineGBScreenTarget;    /* SVGA_3D_CMD_DEFINE_GB_SCREENTARGET */
 
@@ -2419,7 +2469,7 @@ SVGA3dCmdBindGBScreenTarget;  /* SVGA_3D_CMD_BIND_GB_SCREENTARGET */
 typedef
 struct {
    uint32 stid;
-   SVGA3dBox box;
+   SVGA3dRect rect;
 } __packed
 SVGA3dCmdUpdateGBScreenTarget;  /* SVGA_3D_CMD_UPDATE_GB_SCREENTARGET */
 
diff --git a/drivers/gpu/drm/vmwgfx/svga3d_surfacedefs.h b/drivers/gpu/drm/vmwgfx/svga3d_surfacedefs.h
index ef33850..d55ab01 100644
--- a/drivers/gpu/drm/vmwgfx/svga3d_surfacedefs.h
+++ b/drivers/gpu/drm/vmwgfx/svga3d_surfacedefs.h
@@ -608,9 +608,9 @@ static const struct svga3d_surface_desc svga3d_surface_descs[] = {
 	 {1, 1, 1},  1, 1, {8, {{0}, {0}, {8}, {0} } },
 	 {{{0}, {0}, {0}, {0} } } },     /* SVGA3D_R8_SINT */
 
-	{SVGA3DBLOCKDESC_RED,
-	 {8, 1, 1},  1, 1, {8, {{0}, {0}, {8}, {0} } },
-	 {{{0}, {0}, {0}, {0} } } },     /* SVGA3D_R1_UNORM */
+        {SVGA3DBLOCKDESC_NONE,
+         {1, 1, 1},  1, 1, {8, {{0}, {0}, {8}, {0} } },
+         {{{0}, {0}, {0}, {0} } } },     /* SVGA3D_P8 */
 
 	{SVGA3DBLOCKDESC_RGBE,
 	 {1, 1, 1},  4, 4, {32, {{9}, {9}, {9}, {5} } },
@@ -910,3 +910,64 @@ svga3dsurface_get_image_offset(SVGA3dSurfaceFormat format,
 
 	return offset;
 }
+
+
+/**
+ * svga3dsurface_is_gb_screen_target_format - Is the specified format usable as
+ *                                            a ScreenTarget?
+ *                                            (with just the GBObjects cap-bit
+ *                                             set)
+ * @format: format to queried
+ *
+ * RETURNS:
+ * true if queried format is valid for screen targets
+ */
+static inline bool
+svga3dsurface_is_gb_screen_target_format(SVGA3dSurfaceFormat format)
+{
+	return (format == SVGA3D_X8R8G8B8 ||
+		format == SVGA3D_A8R8G8B8 ||
+		format == SVGA3D_R5G6B5   ||
+		format == SVGA3D_X1R5G5B5 ||
+		format == SVGA3D_A1R5G5B5 ||
+		format == SVGA3D_P8);
+}
+
+
+/**
+ * svga3dsurface_is_dx_screen_target_format - Is the specified format usable as
+ *                                            a ScreenTarget?
+ *                                            (with DX10 enabled)
+ *
+ * @format: format to queried
+ *
+ * Results:
+ * true if queried format is valid for screen targets
+ */
+static inline bool
+svga3dsurface_is_dx_screen_target_format(SVGA3dSurfaceFormat format)
+{
+	return (format == SVGA3D_R8G8B8A8_UNORM ||
+		format == SVGA3D_B8G8R8A8_UNORM ||
+		format == SVGA3D_B8G8R8X8_UNORM);
+}
+
+
+/**
+ * svga3dsurface_is_screen_target_format - Is the specified format usable as a
+ *                                         ScreenTarget?
+ *                                         (for some combination of caps)
+ *
+ * @format: format to queried
+ *
+ * Results:
+ * true if queried format is valid for screen targets
+ */
+static inline bool
+svga3dsurface_is_screen_target_format(SVGA3dSurfaceFormat format)
+{
+	if (svga3dsurface_is_gb_screen_target_format(format)) {
+		return true;
+	}
+	return svga3dsurface_is_dx_screen_target_format(format);
+}
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 09/28] drm/vmwgfx: Refactor vmw_gb_surface_define_ioctl()

[Thomas Hellstrom]

From: Sinclair Yeh <syeh@vmware.com>

Refactored vmw_gb_surface_define_ioctl() and made the surface
definition part a separate function.  This way other parts of vmwgfx
can use it to allocate kernel-visible GB surfaces.

Signed-off-by: Sinclair Yeh <syeh@vmware.com>
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.h     |   9 ++
 drivers/gpu/drm/vmwgfx/vmwgfx_surface.c | 195 ++++++++++++++++++++++----------
 2 files changed, 143 insertions(+), 61 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index 338dce3..c300a0a 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -1043,6 +1043,15 @@ extern int vmw_surface_check(struct vmw_private *dev_priv,
 			     uint32_t handle, int *id);
 extern int vmw_surface_validate(struct vmw_private *dev_priv,
 				struct vmw_surface *srf);
+int vmw_surface_gb_priv_define(struct drm_device *dev,
+			       uint32_t user_accounting_size,
+			       uint32_t svga3d_flags,
+			       SVGA3dSurfaceFormat format,
+			       bool for_scanout,
+			       uint32_t num_mip_levels,
+			       uint32_t multisample_count,
+			       struct drm_vmw_size size,
+			       struct vmw_surface **srf_out);
 
 /*
  * Shader management - vmwgfx_shader.c
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c b/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
index 4d0c98e..fb54ccd 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
@@ -1,6 +1,6 @@
 /**************************************************************************
  *
- * Copyright © 2009-2012 VMware, Inc., Palo Alto, CA., USA
+ * Copyright © 2009-2014 VMware, Inc., Palo Alto, CA., USA
  * All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -36,7 +36,7 @@
  * @base:           The TTM base object handling user-space visibility.
  * @srf:            The surface metadata.
  * @size:           TTM accounting size for the surface.
- * @master:         master of the creating client. Used for security check.
+ * @master: master of the creating client. Used for security check.
  */
 struct vmw_user_surface {
 	struct ttm_prime_object prime;
@@ -1218,6 +1218,7 @@ static int vmw_gb_surface_destroy(struct vmw_resource *res)
 	return 0;
 }
 
+
 /**
  * vmw_gb_surface_define_ioctl - Ioctl function implementing
  *                               the user surface define functionality.
@@ -1241,77 +1242,43 @@ int vmw_gb_surface_define_ioctl(struct drm_device *dev, void *data,
 	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
 	int ret;
 	uint32_t size;
-	const struct svga3d_surface_desc *desc;
 	uint32_t backup_handle;
 
+
 	if (unlikely(vmw_user_surface_size == 0))
 		vmw_user_surface_size = ttm_round_pot(sizeof(*user_srf)) +
 			128;
 
 	size = vmw_user_surface_size + 128;
 
-	desc = svga3dsurface_get_desc(req->format);
-	if (unlikely(desc->block_desc == SVGA3DBLOCKDESC_NONE)) {
-		DRM_ERROR("Invalid surface format for surface creation.\n");
-		return -EINVAL;
-	}
-
-	ret = ttm_read_lock(&dev_priv->reservation_sem, true);
+	/* Define a surface based on the parameters. */
+	ret = vmw_surface_gb_priv_define(dev,
+			size,
+			req->svga3d_flags,
+			req->format,
+			req->drm_surface_flags & drm_vmw_surface_flag_scanout,
+			req->mip_levels,
+			req->multisample_count,
+			req->base_size,
+			&srf);
 	if (unlikely(ret != 0))
 		return ret;
 
-	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
-				   size, false, true);
-	if (unlikely(ret != 0)) {
-		if (ret != -ERESTARTSYS)
-			DRM_ERROR("Out of graphics memory for surface"
-				  " creation.\n");
-		goto out_unlock;
-	}
-
-	user_srf = kzalloc(sizeof(*user_srf), GFP_KERNEL);
-	if (unlikely(user_srf == NULL)) {
-		ret = -ENOMEM;
-		goto out_no_user_srf;
-	}
-
-	srf = &user_srf->srf;
-	res = &srf->res;
-
-	srf->flags = req->svga3d_flags;
-	srf->format = req->format;
-	srf->scanout = req->drm_surface_flags & drm_vmw_surface_flag_scanout;
-	srf->mip_levels[0] = req->mip_levels;
-	srf->num_sizes = 1;
-	srf->sizes = NULL;
-	srf->offsets = NULL;
-	user_srf->size = size;
-	srf->base_size = req->base_size;
-	srf->autogen_filter = SVGA3D_TEX_FILTER_NONE;
-	srf->multisample_count = req->multisample_count;
-	res->backup_size = svga3dsurface_get_serialized_size
-	  (srf->format, srf->base_size, srf->mip_levels[0],
-	   srf->flags & SVGA3D_SURFACE_CUBEMAP);
-
-	user_srf->prime.base.shareable = false;
-	user_srf->prime.base.tfile = NULL;
+	user_srf = container_of(srf, struct vmw_user_surface, srf);
 	if (drm_is_primary_client(file_priv))
 		user_srf->master = drm_master_get(file_priv->master);
 
-	/**
-	 * From this point, the generic resource management functions
-	 * destroy the object on failure.
-	 */
-
-	ret = vmw_surface_init(dev_priv, srf, vmw_user_surface_free);
+	ret = ttm_read_lock(&dev_priv->reservation_sem, true);
 	if (unlikely(ret != 0))
-		goto out_unlock;
+		return ret;
+
+	res = &user_srf->srf.res;
 
-	if (req->buffer_handle != SVGA3D_INVALID_ID) {
+
+	if (req->buffer_handle != SVGA3D_INVALID_ID)
 		ret = vmw_user_dmabuf_lookup(tfile, req->buffer_handle,
 					     &res->backup);
-	} else if (req->drm_surface_flags &
-		   drm_vmw_surface_flag_create_buffer)
+	else if (req->drm_surface_flags & drm_vmw_surface_flag_create_buffer)
 		ret = vmw_user_dmabuf_alloc(dev_priv, tfile,
 					    res->backup_size,
 					    req->drm_surface_flags &
@@ -1324,7 +1291,7 @@ int vmw_gb_surface_define_ioctl(struct drm_device *dev, void *data,
 		goto out_unlock;
 	}
 
-	tmp = vmw_resource_reference(&srf->res);
+	tmp = vmw_resource_reference(res);
 	ret = ttm_prime_object_init(tfile, res->backup_size, &user_srf->prime,
 				    req->drm_surface_flags &
 				    drm_vmw_surface_flag_shareable,
@@ -1337,7 +1304,7 @@ int vmw_gb_surface_define_ioctl(struct drm_device *dev, void *data,
 		goto out_unlock;
 	}
 
-	rep->handle = user_srf->prime.base.hash.key;
+	rep->handle      = user_srf->prime.base.hash.key;
 	rep->backup_size = res->backup_size;
 	if (res->backup) {
 		rep->buffer_map_handle =
@@ -1352,10 +1319,6 @@ int vmw_gb_surface_define_ioctl(struct drm_device *dev, void *data,
 
 	vmw_resource_unreference(&res);
 
-	ttm_read_unlock(&dev_priv->reservation_sem);
-	return 0;
-out_no_user_srf:
-	ttm_mem_global_free(vmw_mem_glob(dev_priv), size);
 out_unlock:
 	ttm_read_unlock(&dev_priv->reservation_sem);
 	return ret;
@@ -1429,3 +1392,113 @@ out_bad_resource:
 
 	return ret;
 }
+
+/**
+ * vmw_surface_gb_priv_define - Define a private GB surface
+ *
+ * @dev:  Pointer to a struct drm_device
+ * @user_accounting_size:  Used to track user-space memory usage, set
+ *                         to 0 for kernel mode only memory
+ * @svga3d_flags: SVGA3d surface flags for the device
+ * @format: requested surface format
+ * @for_scanout: true if inteded to be used for scanout buffer
+ * @num_mip_levels:  number of MIP levels
+ * @multisample_count:
+ * @size: width, heigh, depth of the surface requested
+ * @user_srf_out: allocated user_srf.  Set to NULL on failure.
+ *
+ * GB surfaces allocated by this function will not have a user mode handle, and
+ * thus will only be visible to vmwgfx.  For optimization reasons the
+ * surface may later be given a user mode handle by another function to make
+ * it available to user mode drivers.
+ */
+int vmw_surface_gb_priv_define(struct drm_device *dev,
+			       uint32_t user_accounting_size,
+			       uint32_t svga3d_flags,
+			       SVGA3dSurfaceFormat format,
+			       bool for_scanout,
+			       uint32_t num_mip_levels,
+			       uint32_t multisample_count,
+			       struct drm_vmw_size size,
+			       struct vmw_surface **srf_out)
+{
+	struct vmw_private *dev_priv = vmw_priv(dev);
+	struct vmw_user_surface *user_srf;
+	struct vmw_surface *srf;
+	int ret;
+
+
+	*srf_out = NULL;
+
+	if (for_scanout) {
+		if (!svga3dsurface_is_screen_target_format(format)) {
+			DRM_ERROR("Invalid Screen Target surface format.");
+			return -EINVAL;
+		}
+	} else {
+		const struct svga3d_surface_desc *desc;
+
+		desc = svga3dsurface_get_desc(format);
+		if (unlikely(desc->block_desc == SVGA3DBLOCKDESC_NONE)) {
+			DRM_ERROR("Invalid surface format.\n");
+			return -EINVAL;
+		}
+	}
+
+	ret = ttm_read_lock(&dev_priv->reservation_sem, true);
+	if (unlikely(ret != 0))
+		return ret;
+
+	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
+				   user_accounting_size, false, true);
+	if (unlikely(ret != 0)) {
+		if (ret != -ERESTARTSYS)
+			DRM_ERROR("Out of graphics memory for surface"
+				  " creation.\n");
+		goto out_unlock;
+	}
+
+	user_srf = kzalloc(sizeof(*user_srf), GFP_KERNEL);
+	if (unlikely(user_srf == NULL)) {
+		ret = -ENOMEM;
+		goto out_no_user_srf;
+	}
+
+	*srf_out  = &user_srf->srf;
+	user_srf->size = user_accounting_size;
+	user_srf->prime.base.shareable = false;
+	user_srf->prime.base.tfile     = NULL;
+
+	srf = &user_srf->srf;
+	srf->flags             = svga3d_flags;
+	srf->format            = format;
+	srf->scanout           = for_scanout;
+	srf->mip_levels[0]     = num_mip_levels;
+	srf->num_sizes         = 1;
+	srf->sizes             = NULL;
+	srf->offsets           = NULL;
+	srf->base_size         = size;
+	srf->autogen_filter    = SVGA3D_TEX_FILTER_NONE;
+	srf->multisample_count = multisample_count;
+
+	srf->res.backup_size   = svga3dsurface_get_serialized_size(srf->format,
+					srf->base_size,
+					srf->mip_levels[0],
+					srf->flags & SVGA3D_SURFACE_CUBEMAP);
+
+	/*
+	 * From this point, the generic resource management functions
+	 * destroy the object on failure.
+	 */
+	ret = vmw_surface_init(dev_priv, srf, vmw_user_surface_free);
+
+	ttm_read_unlock(&dev_priv->reservation_sem);
+	return ret;
+
+out_no_user_srf:
+	ttm_mem_global_free(vmw_mem_glob(dev_priv), user_accounting_size);
+
+out_unlock:
+	ttm_read_unlock(&dev_priv->reservation_sem);
+	return ret;
+}
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 10/28] vmwgfx: Major KMS refactoring / cleanup in preparation of screen targets

[Thomas Hellstrom]

From: Sinclair Yeh <syeh@vmware.com>

Signed-off-by: Sinclair Yeh <syeh@vmware.com>
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.h     |  22 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c    |   4 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_kms.c     | 562 +++++---------------------------
 drivers/gpu/drm/vmwgfx/vmwgfx_kms.h     |  80 +++--
 drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c     |  45 ++-
 drivers/gpu/drm/vmwgfx/vmwgfx_overlay.c |   6 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c    | 424 ++++++++++++++++++++++--
 7 files changed, 597 insertions(+), 546 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index c300a0a..b65eb02 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -1,6 +1,6 @@
 /**************************************************************************
  *
- * Copyright © 2009 VMware, Inc., Palo Alto, CA., USA
+ * Copyright © 2009-2014 VMware, Inc., Palo Alto, CA., USA
  * All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -330,6 +330,17 @@ struct vmw_ctx_binding_state {
 	struct vmw_ctx_binding shaders[SVGA3D_SHADERTYPE_MAX];
 };
 
+
+/*
+ * enum vmw_display_unit_type - Describes the display unit
+ */
+enum vmw_display_unit_type {
+	vmw_du_invalid = 0,
+	vmw_du_legacy,
+	vmw_du_screen_object
+};
+
+
 struct vmw_sw_context{
 	struct drm_open_hash res_ht;
 	bool res_ht_initialized;
@@ -421,6 +432,7 @@ struct vmw_private {
 	 */
 
 	void *fb_info;
+	enum vmw_display_unit_type active_display_unit;
 	struct vmw_legacy_display *ldu_priv;
 	struct vmw_screen_object_display *sou_priv;
 	struct vmw_overlay *overlay_priv;
@@ -844,8 +856,8 @@ extern void vmw_execbuf_copy_fence_user(struct vmw_private *dev_priv,
 
 extern irqreturn_t vmw_irq_handler(int irq, void *arg);
 extern int vmw_wait_seqno(struct vmw_private *dev_priv, bool lazy,
-			     uint32_t seqno, bool interruptible,
-			     unsigned long timeout);
+			  uint32_t seqno, bool interruptible,
+			  unsigned long timeout);
 extern void vmw_irq_preinstall(struct drm_device *dev);
 extern int vmw_irq_postinstall(struct drm_device *dev);
 extern void vmw_irq_uninstall(struct drm_device *dev);
@@ -876,9 +888,9 @@ extern void vmw_generic_waiter_remove(struct vmw_private *dev_priv,
 extern void vmw_marker_queue_init(struct vmw_marker_queue *queue);
 extern void vmw_marker_queue_takedown(struct vmw_marker_queue *queue);
 extern int vmw_marker_push(struct vmw_marker_queue *queue,
-			  uint32_t seqno);
+			   uint32_t seqno);
 extern int vmw_marker_pull(struct vmw_marker_queue *queue,
-			  uint32_t signaled_seqno);
+			   uint32_t signaled_seqno);
 extern int vmw_wait_lag(struct vmw_private *dev_priv,
 			struct vmw_marker_queue *queue, uint32_t us);
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
index 189102d..239815c 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
@@ -71,8 +71,8 @@ bool vmw_fifo_have_3d(struct vmw_private *dev_priv)
 	if (hwversion < SVGA3D_HWVERSION_WS8_B1)
 		return false;
 
-	/* Non-Screen Object path does not support surfaces */
-	if (!dev_priv->sou_priv)
+	/* Legacy Display Unit does not support surfaces */
+	if (dev_priv->active_display_unit == vmw_du_legacy)
 		return false;
 
 	return true;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
index b5632c2..cac17c2 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
@@ -1,6 +1,6 @@
 /**************************************************************************
  *
- * Copyright © 2009 VMware, Inc., Palo Alto, CA., USA
+ * Copyright © 2009-2014 VMware, Inc., Palo Alto, CA., USA
  * All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -32,15 +32,12 @@
 #define VMWGFX_PRESENT_RATE ((HZ / 60 > 0) ? HZ / 60 : 1)
 
 
-struct vmw_clip_rect {
-	int x1, x2, y1, y2;
-};
 
 /**
  * Clip @num_rects number of @rects against @clip storing the
  * results in @out_rects and the number of passed rects in @out_num.
  */
-static void vmw_clip_cliprects(struct drm_clip_rect *rects,
+void vmw_clip_cliprects(struct drm_clip_rect *rects,
 			int num_rects,
 			struct vmw_clip_rect clip,
 			SVGASignedRect *out_rects,
@@ -69,7 +66,7 @@ static void vmw_clip_cliprects(struct drm_clip_rect *rects,
 	*out_num = k;
 }
 
-void vmw_display_unit_cleanup(struct vmw_display_unit *du)
+void vmw_du_cleanup(struct vmw_display_unit *du)
 {
 	if (du->cursor_surface)
 		vmw_surface_unreference(&du->cursor_surface);
@@ -367,15 +364,6 @@ void vmw_kms_cursor_snoop(struct vmw_surface *srf,
 
 	srf->snooper.age++;
 
-	/* we can't call this function from this function since execbuf has
-	 * reserved fifo space.
-	 *
-	 * if (srf->snooper.crtc)
-	 *	vmw_ldu_crtc_cursor_update_image(dev_priv,
-	 *					 srf->snooper.image, 64, 64,
-	 *					 du->hotspot_x, du->hotspot_y);
-	 */
-
 	ttm_bo_kunmap(&map);
 err_unreserve:
 	ttm_bo_unreserve(bo);
@@ -412,17 +400,6 @@ void vmw_kms_cursor_post_execbuf(struct vmw_private *dev_priv)
  * Surface framebuffer code
  */
 
-#define vmw_framebuffer_to_vfbs(x) \
-	container_of(x, struct vmw_framebuffer_surface, base.base)
-
-struct vmw_framebuffer_surface {
-	struct vmw_framebuffer base;
-	struct vmw_surface *surface;
-	struct vmw_dma_buffer *buffer;
-	struct list_head head;
-	struct drm_master *master;
-};
-
 static void vmw_framebuffer_surface_destroy(struct drm_framebuffer *framebuffer)
 {
 	struct vmw_framebuffer_surface *vfbs =
@@ -442,153 +419,6 @@ static void vmw_framebuffer_surface_destroy(struct drm_framebuffer *framebuffer)
 	kfree(vfbs);
 }
 
-static int do_surface_dirty_sou(struct vmw_private *dev_priv,
-				struct drm_file *file_priv,
-				struct vmw_framebuffer *framebuffer,
-				unsigned flags, unsigned color,
-				struct drm_clip_rect *clips,
-				unsigned num_clips, int inc,
-				struct vmw_fence_obj **out_fence)
-{
-	struct vmw_display_unit *units[VMWGFX_NUM_DISPLAY_UNITS];
-	struct drm_clip_rect *clips_ptr;
-	struct drm_clip_rect *tmp;
-	struct drm_crtc *crtc;
-	size_t fifo_size;
-	int i, num_units;
-	int ret = 0; /* silence warning */
-	int left, right, top, bottom;
-
-	struct {
-		SVGA3dCmdHeader header;
-		SVGA3dCmdBlitSurfaceToScreen body;
-	} *cmd;
-	SVGASignedRect *blits;
-
-	num_units = 0;
-	list_for_each_entry(crtc, &dev_priv->dev->mode_config.crtc_list,
-			    head) {
-		if (crtc->primary->fb != &framebuffer->base)
-			continue;
-		units[num_units++] = vmw_crtc_to_du(crtc);
-	}
-
-	BUG_ON(!clips || !num_clips);
-
-	tmp = kzalloc(sizeof(*tmp) * num_clips, GFP_KERNEL);
-	if (unlikely(tmp == NULL)) {
-		DRM_ERROR("Temporary cliprect memory alloc failed.\n");
-		return -ENOMEM;
-	}
-
-	fifo_size = sizeof(*cmd) + sizeof(SVGASignedRect) * num_clips;
-	cmd = kzalloc(fifo_size, GFP_KERNEL);
-	if (unlikely(cmd == NULL)) {
-		DRM_ERROR("Temporary fifo memory alloc failed.\n");
-		ret = -ENOMEM;
-		goto out_free_tmp;
-	}
-
-	/* setup blits pointer */
-	blits = (SVGASignedRect *)&cmd[1];
-
-	/* initial clip region */
-	left = clips->x1;
-	right = clips->x2;
-	top = clips->y1;
-	bottom = clips->y2;
-
-	/* skip the first clip rect */
-	for (i = 1, clips_ptr = clips + inc;
-	     i < num_clips; i++, clips_ptr += inc) {
-		left = min_t(int, left, (int)clips_ptr->x1);
-		right = max_t(int, right, (int)clips_ptr->x2);
-		top = min_t(int, top, (int)clips_ptr->y1);
-		bottom = max_t(int, bottom, (int)clips_ptr->y2);
-	}
-
-	/* only need to do this once */
-	cmd->header.id = cpu_to_le32(SVGA_3D_CMD_BLIT_SURFACE_TO_SCREEN);
-	cmd->header.size = cpu_to_le32(fifo_size - sizeof(cmd->header));
-
-	cmd->body.srcRect.left = left;
-	cmd->body.srcRect.right = right;
-	cmd->body.srcRect.top = top;
-	cmd->body.srcRect.bottom = bottom;
-
-	clips_ptr = clips;
-	for (i = 0; i < num_clips; i++, clips_ptr += inc) {
-		tmp[i].x1 = clips_ptr->x1 - left;
-		tmp[i].x2 = clips_ptr->x2 - left;
-		tmp[i].y1 = clips_ptr->y1 - top;
-		tmp[i].y2 = clips_ptr->y2 - top;
-	}
-
-	/* do per unit writing, reuse fifo for each */
-	for (i = 0; i < num_units; i++) {
-		struct vmw_display_unit *unit = units[i];
-		struct vmw_clip_rect clip;
-		int num;
-
-		clip.x1 = left - unit->crtc.x;
-		clip.y1 = top - unit->crtc.y;
-		clip.x2 = right - unit->crtc.x;
-		clip.y2 = bottom - unit->crtc.y;
-
-		/* skip any crtcs that misses the clip region */
-		if (clip.x1 >= unit->crtc.mode.hdisplay ||
-		    clip.y1 >= unit->crtc.mode.vdisplay ||
-		    clip.x2 <= 0 || clip.y2 <= 0)
-			continue;
-
-		/*
-		 * In order for the clip rects to be correctly scaled
-		 * the src and dest rects needs to be the same size.
-		 */
-		cmd->body.destRect.left = clip.x1;
-		cmd->body.destRect.right = clip.x2;
-		cmd->body.destRect.top = clip.y1;
-		cmd->body.destRect.bottom = clip.y2;
-
-		/* create a clip rect of the crtc in dest coords */
-		clip.x2 = unit->crtc.mode.hdisplay - clip.x1;
-		clip.y2 = unit->crtc.mode.vdisplay - clip.y1;
-		clip.x1 = 0 - clip.x1;
-		clip.y1 = 0 - clip.y1;
-
-		/* need to reset sid as it is changed by execbuf */
-		cmd->body.srcImage.sid = cpu_to_le32(framebuffer->user_handle);
-		cmd->body.destScreenId = unit->unit;
-
-		/* clip and write blits to cmd stream */
-		vmw_clip_cliprects(tmp, num_clips, clip, blits, &num);
-
-		/* if no cliprects hit skip this */
-		if (num == 0)
-			continue;
-
-		/* only return the last fence */
-		if (out_fence && *out_fence)
-			vmw_fence_obj_unreference(out_fence);
-
-		/* recalculate package length */
-		fifo_size = sizeof(*cmd) + sizeof(SVGASignedRect) * num;
-		cmd->header.size = cpu_to_le32(fifo_size - sizeof(cmd->header));
-		ret = vmw_execbuf_process(file_priv, dev_priv, NULL, cmd,
-					  fifo_size, 0, NULL, out_fence);
-
-		if (unlikely(ret != 0))
-			break;
-	}
-
-
-	kfree(cmd);
-out_free_tmp:
-	kfree(tmp);
-
-	return ret;
-}
-
 static int vmw_framebuffer_surface_dirty(struct drm_framebuffer *framebuffer,
 				  struct drm_file *file_priv,
 				  unsigned flags, unsigned color,
@@ -604,8 +434,8 @@ static int vmw_framebuffer_surface_dirty(struct drm_framebuffer *framebuffer,
 	if (unlikely(vfbs->master != file_priv->master))
 		return -EINVAL;
 
-	/* Require ScreenObject support for 3D */
-	if (!dev_priv->sou_priv)
+	/* Legacy Display Unit does not support 3D */
+	if (dev_priv->active_display_unit == vmw_du_legacy)
 		return -EINVAL;
 
 	drm_modeset_lock_all(dev_priv->dev);
@@ -627,9 +457,12 @@ static int vmw_framebuffer_surface_dirty(struct drm_framebuffer *framebuffer,
 		inc = 2; /* skip source rects */
 	}
 
-	ret = do_surface_dirty_sou(dev_priv, file_priv, &vfbs->base,
-				   flags, color,
-				   clips, num_clips, inc, NULL);
+	if (dev_priv->active_display_unit == vmw_du_screen_object)
+		ret = vmw_kms_sou_do_surface_dirty(dev_priv, file_priv,
+						   &vfbs->base,
+						   flags, color,
+						   clips, num_clips,
+						   inc, NULL);
 
 	vmw_fifo_flush(dev_priv, false);
 	ttm_read_unlock(&dev_priv->reservation_sem);
@@ -658,8 +491,8 @@ static int vmw_kms_new_framebuffer_surface(struct vmw_private *dev_priv,
 	struct vmw_master *vmaster = vmw_master(file_priv->master);
 	int ret;
 
-	/* 3D is only supported on HWv8 hosts which supports screen objects */
-	if (!dev_priv->sou_priv)
+	/* 3D is only supported on HWv8 and newer hosts */
+	if (dev_priv->active_display_unit == vmw_du_legacy)
 		return -ENOSYS;
 
 	/*
@@ -693,9 +526,6 @@ static int vmw_kms_new_framebuffer_surface(struct vmw_private *dev_priv,
 	case 15:
 		format = SVGA3D_A1R5G5B5;
 		break;
-	case 8:
-		format = SVGA3D_LUMINANCE8;
-		break;
 	default:
 		DRM_ERROR("Invalid color depth: %d\n", mode_cmd->depth);
 		return -EINVAL;
@@ -753,14 +583,6 @@ out_err1:
  * Dmabuf framebuffer code
  */
 
-#define vmw_framebuffer_to_vfbd(x) \
-	container_of(x, struct vmw_framebuffer_dmabuf, base.base)
-
-struct vmw_framebuffer_dmabuf {
-	struct vmw_framebuffer base;
-	struct vmw_dma_buffer *buffer;
-};
-
 static void vmw_framebuffer_dmabuf_destroy(struct drm_framebuffer *framebuffer)
 {
 	struct vmw_framebuffer_dmabuf *vfbd =
@@ -773,180 +595,6 @@ static void vmw_framebuffer_dmabuf_destroy(struct drm_framebuffer *framebuffer)
 	kfree(vfbd);
 }
 
-static int do_dmabuf_dirty_ldu(struct vmw_private *dev_priv,
-			       struct vmw_framebuffer *framebuffer,
-			       unsigned flags, unsigned color,
-			       struct drm_clip_rect *clips,
-			       unsigned num_clips, int increment)
-{
-	size_t fifo_size;
-	int i;
-
-	struct {
-		uint32_t header;
-		SVGAFifoCmdUpdate body;
-	} *cmd;
-
-	fifo_size = sizeof(*cmd) * num_clips;
-	cmd = vmw_fifo_reserve(dev_priv, fifo_size);
-	if (unlikely(cmd == NULL)) {
-		DRM_ERROR("Fifo reserve failed.\n");
-		return -ENOMEM;
-	}
-
-	memset(cmd, 0, fifo_size);
-	for (i = 0; i < num_clips; i++, clips += increment) {
-		cmd[i].header = cpu_to_le32(SVGA_CMD_UPDATE);
-		cmd[i].body.x = cpu_to_le32(clips->x1);
-		cmd[i].body.y = cpu_to_le32(clips->y1);
-		cmd[i].body.width = cpu_to_le32(clips->x2 - clips->x1);
-		cmd[i].body.height = cpu_to_le32(clips->y2 - clips->y1);
-	}
-
-	vmw_fifo_commit(dev_priv, fifo_size);
-	return 0;
-}
-
-static int do_dmabuf_define_gmrfb(struct drm_file *file_priv,
-				  struct vmw_private *dev_priv,
-				  struct vmw_framebuffer *framebuffer)
-{
-	int depth = framebuffer->base.depth;
-	size_t fifo_size;
-	int ret;
-
-	struct {
-		uint32_t header;
-		SVGAFifoCmdDefineGMRFB body;
-	} *cmd;
-
-	/* Emulate RGBA support, contrary to svga_reg.h this is not
-	 * supported by hosts. This is only a problem if we are reading
-	 * this value later and expecting what we uploaded back.
-	 */
-	if (depth == 32)
-		depth = 24;
-
-	fifo_size = sizeof(*cmd);
-	cmd = kmalloc(fifo_size, GFP_KERNEL);
-	if (unlikely(cmd == NULL)) {
-		DRM_ERROR("Failed to allocate temporary cmd buffer.\n");
-		return -ENOMEM;
-	}
-
-	memset(cmd, 0, fifo_size);
-	cmd->header = SVGA_CMD_DEFINE_GMRFB;
-	cmd->body.format.bitsPerPixel = framebuffer->base.bits_per_pixel;
-	cmd->body.format.colorDepth = depth;
-	cmd->body.format.reserved = 0;
-	cmd->body.bytesPerLine = framebuffer->base.pitches[0];
-	cmd->body.ptr.gmrId = framebuffer->user_handle;
-	cmd->body.ptr.offset = 0;
-
-	ret = vmw_execbuf_process(file_priv, dev_priv, NULL, cmd,
-				  fifo_size, 0, NULL, NULL);
-
-	kfree(cmd);
-
-	return ret;
-}
-
-static int do_dmabuf_dirty_sou(struct drm_file *file_priv,
-			       struct vmw_private *dev_priv,
-			       struct vmw_framebuffer *framebuffer,
-			       unsigned flags, unsigned color,
-			       struct drm_clip_rect *clips,
-			       unsigned num_clips, int increment,
-			       struct vmw_fence_obj **out_fence)
-{
-	struct vmw_display_unit *units[VMWGFX_NUM_DISPLAY_UNITS];
-	struct drm_clip_rect *clips_ptr;
-	int i, k, num_units, ret;
-	struct drm_crtc *crtc;
-	size_t fifo_size;
-
-	struct {
-		uint32_t header;
-		SVGAFifoCmdBlitGMRFBToScreen body;
-	} *blits;
-
-	ret = do_dmabuf_define_gmrfb(file_priv, dev_priv, framebuffer);
-	if (unlikely(ret != 0))
-		return ret; /* define_gmrfb prints warnings */
-
-	fifo_size = sizeof(*blits) * num_clips;
-	blits = kmalloc(fifo_size, GFP_KERNEL);
-	if (unlikely(blits == NULL)) {
-		DRM_ERROR("Failed to allocate temporary cmd buffer.\n");
-		return -ENOMEM;
-	}
-
-	num_units = 0;
-	list_for_each_entry(crtc, &dev_priv->dev->mode_config.crtc_list, head) {
-		if (crtc->primary->fb != &framebuffer->base)
-			continue;
-		units[num_units++] = vmw_crtc_to_du(crtc);
-	}
-
-	for (k = 0; k < num_units; k++) {
-		struct vmw_display_unit *unit = units[k];
-		int hit_num = 0;
-
-		clips_ptr = clips;
-		for (i = 0; i < num_clips; i++, clips_ptr += increment) {
-			int clip_x1 = clips_ptr->x1 - unit->crtc.x;
-			int clip_y1 = clips_ptr->y1 - unit->crtc.y;
-			int clip_x2 = clips_ptr->x2 - unit->crtc.x;
-			int clip_y2 = clips_ptr->y2 - unit->crtc.y;
-			int move_x, move_y;
-
-			/* skip any crtcs that misses the clip region */
-			if (clip_x1 >= unit->crtc.mode.hdisplay ||
-			    clip_y1 >= unit->crtc.mode.vdisplay ||
-			    clip_x2 <= 0 || clip_y2 <= 0)
-				continue;
-
-			/* clip size to crtc size */
-			clip_x2 = min_t(int, clip_x2, unit->crtc.mode.hdisplay);
-			clip_y2 = min_t(int, clip_y2, unit->crtc.mode.vdisplay);
-
-			/* translate both src and dest to bring clip into screen */
-			move_x = min_t(int, clip_x1, 0);
-			move_y = min_t(int, clip_y1, 0);
-
-			/* actual translate done here */
-			blits[hit_num].header = SVGA_CMD_BLIT_GMRFB_TO_SCREEN;
-			blits[hit_num].body.destScreenId = unit->unit;
-			blits[hit_num].body.srcOrigin.x = clips_ptr->x1 - move_x;
-			blits[hit_num].body.srcOrigin.y = clips_ptr->y1 - move_y;
-			blits[hit_num].body.destRect.left = clip_x1 - move_x;
-			blits[hit_num].body.destRect.top = clip_y1 - move_y;
-			blits[hit_num].body.destRect.right = clip_x2;
-			blits[hit_num].body.destRect.bottom = clip_y2;
-			hit_num++;
-		}
-
-		/* no clips hit the crtc */
-		if (hit_num == 0)
-			continue;
-
-		/* only return the last fence */
-		if (out_fence && *out_fence)
-			vmw_fence_obj_unreference(out_fence);
-
-		fifo_size = sizeof(*blits) * hit_num;
-		ret = vmw_execbuf_process(file_priv, dev_priv, NULL, blits,
-					  fifo_size, 0, NULL, out_fence);
-
-		if (unlikely(ret != 0))
-			break;
-	}
-
-	kfree(blits);
-
-	return ret;
-}
-
 static int vmw_framebuffer_dmabuf_dirty(struct drm_framebuffer *framebuffer,
 				 struct drm_file *file_priv,
 				 unsigned flags, unsigned color,
@@ -979,13 +627,15 @@ static int vmw_framebuffer_dmabuf_dirty(struct drm_framebuffer *framebuffer,
 	}
 
 	if (dev_priv->ldu_priv) {
-		ret = do_dmabuf_dirty_ldu(dev_priv, &vfbd->base,
-					  flags, color,
-					  clips, num_clips, increment);
-	} else {
-		ret = do_dmabuf_dirty_sou(file_priv, dev_priv, &vfbd->base,
-					  flags, color,
-					  clips, num_clips, increment, NULL);
+		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) {
+		ret = vmw_kms_sou_do_dmabuf_dirty(file_priv, dev_priv,
+						  &vfbd->base,
+						  flags, color,
+						  clips, num_clips, increment,
+						  NULL);
 	}
 
 	vmw_fifo_flush(dev_priv, false);
@@ -1011,8 +661,8 @@ static int vmw_framebuffer_dmabuf_pin(struct vmw_framebuffer *vfb)
 		vmw_framebuffer_to_vfbd(&vfb->base);
 	int ret;
 
-	/* This code should not be used with screen objects */
-	BUG_ON(dev_priv->sou_priv);
+	/* This code should only be used with Legacy Display Unit */
+	BUG_ON(dev_priv->active_display_unit != vmw_du_legacy);
 
 	vmw_overlay_pause_all(dev_priv);
 
@@ -1059,7 +709,7 @@ static int vmw_kms_new_framebuffer_dmabuf(struct vmw_private *dev_priv,
 	}
 
 	/* Limited framebuffer color depth support for screen objects */
-	if (dev_priv->sou_priv) {
+	if (dev_priv->active_display_unit == vmw_du_screen_object) {
 		switch (mode_cmd->depth) {
 		case 32:
 		case 24:
@@ -1102,7 +752,7 @@ static int vmw_kms_new_framebuffer_dmabuf(struct vmw_private *dev_priv,
 	vfbd->base.base.depth = mode_cmd->depth;
 	vfbd->base.base.width = mode_cmd->width;
 	vfbd->base.base.height = mode_cmd->height;
-	if (!dev_priv->sou_priv) {
+	if (dev_priv->active_display_unit == vmw_du_legacy) {
 		vfbd->base.pin = vmw_framebuffer_dmabuf_pin;
 		vfbd->base.unpin = vmw_framebuffer_dmabuf_unpin;
 	}
@@ -1159,7 +809,7 @@ static struct drm_framebuffer *vmw_kms_fb_create(struct drm_device *dev,
 	if (!vmw_kms_validate_mode_vram(dev_priv,
 					mode_cmd.pitch,
 					mode_cmd.height)) {
-		DRM_ERROR("VRAM size is too small for requested mode.\n");
+		DRM_ERROR("Requested mode exceed bounding box limit.\n");
 		return ERR_PTR(-ENOMEM);
 	}
 
@@ -1220,7 +870,7 @@ static const struct drm_mode_config_funcs vmw_kms_funcs = {
 	.fb_create = vmw_kms_fb_create,
 };
 
-int vmw_kms_present(struct vmw_private *dev_priv,
+int vmw_kms_generic_present(struct vmw_private *dev_priv,
 		    struct drm_file *file_priv,
 		    struct vmw_framebuffer *vfb,
 		    struct vmw_surface *surface,
@@ -1358,6 +1008,19 @@ out_free_tmp:
 	return ret;
 }
 
+int vmw_kms_present(struct vmw_private *dev_priv,
+		    struct drm_file *file_priv,
+		    struct vmw_framebuffer *vfb,
+		    struct vmw_surface *surface,
+		    uint32_t sid,
+		    int32_t destX, int32_t destY,
+		    struct drm_vmw_rect *clips,
+		    uint32_t num_clips)
+{
+	return vmw_kms_generic_present(dev_priv, file_priv, vfb, surface, sid,
+				       destX, destY, clips, num_clips);
+}
+
 int vmw_kms_readback(struct vmw_private *dev_priv,
 		     struct drm_file *file_priv,
 		     struct vmw_framebuffer *vfb,
@@ -1478,26 +1141,29 @@ int vmw_kms_init(struct vmw_private *dev_priv)
 	dev->mode_config.max_width = 8192;
 	dev->mode_config.max_height = 8192;
 
-	ret = vmw_kms_init_screen_object_display(dev_priv);
+	ret = vmw_kms_sou_init_display(dev_priv);
 	if (ret) /* Fallback */
-		(void)vmw_kms_init_legacy_display_system(dev_priv);
+		ret = vmw_kms_ldu_init_display(dev_priv);
 
-	return 0;
+	return ret;
 }
 
 int vmw_kms_close(struct vmw_private *dev_priv)
 {
+	int ret;
+
 	/*
 	 * Docs says we should take the lock before calling this function
 	 * but since it destroys encoders and our destructor calls
 	 * drm_encoder_cleanup which takes the lock we deadlock.
 	 */
 	drm_mode_config_cleanup(dev_priv->dev);
-	if (dev_priv->sou_priv)
-		vmw_kms_close_screen_object_display(dev_priv);
+	if (dev_priv->active_display_unit == vmw_du_screen_object)
+		ret = vmw_kms_sou_close_display(dev_priv);
 	else
-		vmw_kms_close_legacy_display_system(dev_priv);
-	return 0;
+		ret = vmw_kms_ldu_close_display(dev_priv);
+
+	return ret;
 }
 
 int vmw_kms_cursor_bypass_ioctl(struct drm_device *dev, void *data,
@@ -1573,7 +1239,7 @@ int vmw_kms_save_vga(struct vmw_private *vmw_priv)
 		  vmw_read(vmw_priv, SVGA_REG_PITCHLOCK);
 	else if (vmw_fifo_have_pitchlock(vmw_priv))
 		vmw_priv->vga_pitchlock = ioread32(vmw_priv->mmio_virt +
-						       SVGA_FIFO_PITCHLOCK);
+						   SVGA_FIFO_PITCHLOCK);
 
 	if (!(vmw_priv->capabilities & SVGA_CAP_DISPLAY_TOPOLOGY))
 		return 0;
@@ -1719,75 +1385,6 @@ static int vmw_du_update_layout(struct vmw_private *dev_priv, unsigned num,
 	return 0;
 }
 
-int vmw_du_page_flip(struct drm_crtc *crtc,
-		     struct drm_framebuffer *fb,
-		     struct drm_pending_vblank_event *event,
-		     uint32_t page_flip_flags)
-{
-	struct vmw_private *dev_priv = vmw_priv(crtc->dev);
-	struct drm_framebuffer *old_fb = crtc->primary->fb;
-	struct vmw_framebuffer *vfb = vmw_framebuffer_to_vfb(fb);
-	struct drm_file *file_priv ;
-	struct vmw_fence_obj *fence = NULL;
-	struct drm_clip_rect clips;
-	int ret;
-
-	if (event == NULL)
-		return -EINVAL;
-
-	/* require ScreenObject support for page flipping */
-	if (!dev_priv->sou_priv)
-		return -ENOSYS;
-
-	file_priv = event->base.file_priv;
-	if (!vmw_kms_screen_object_flippable(dev_priv, crtc))
-		return -EINVAL;
-
-	crtc->primary->fb = fb;
-
-	/* do a full screen dirty update */
-	clips.x1 = clips.y1 = 0;
-	clips.x2 = fb->width;
-	clips.y2 = fb->height;
-
-	if (vfb->dmabuf)
-		ret = do_dmabuf_dirty_sou(file_priv, dev_priv, vfb,
-					  0, 0, &clips, 1, 1, &fence);
-	else
-		ret = do_surface_dirty_sou(dev_priv, file_priv, vfb,
-					   0, 0, &clips, 1, 1, &fence);
-
-
-	if (ret != 0)
-		goto out_no_fence;
-	if (!fence) {
-		ret = -EINVAL;
-		goto out_no_fence;
-	}
-
-	ret = vmw_event_fence_action_queue(file_priv, fence,
-					   &event->base,
-					   &event->event.tv_sec,
-					   &event->event.tv_usec,
-					   true);
-
-	/*
-	 * No need to hold on to this now. The only cleanup
-	 * we need to do if we fail is unref the fence.
-	 */
-	vmw_fence_obj_unreference(&fence);
-
-	if (vmw_crtc_to_du(crtc)->is_implicit)
-		vmw_kms_screen_object_update_implicit_fb(dev_priv, crtc);
-
-	return ret;
-
-out_no_fence:
-	crtc->primary->fb = old_fb;
-	return ret;
-}
-
-
 void vmw_du_crtc_save(struct drm_crtc *crtc)
 {
 }
@@ -1958,36 +1555,34 @@ int vmw_du_connector_fill_modes(struct drm_connector *connector,
 	 * If using screen objects, then assume 32-bpp because that's what the
 	 * SVGA device is assuming
 	 */
-	if (dev_priv->sou_priv)
+	if (dev_priv->active_display_unit == vmw_du_screen_object)
 		assumed_bpp = 4;
 
 	/* Add preferred mode */
-	{
-		mode = drm_mode_duplicate(dev, &prefmode);
-		if (!mode)
-			return 0;
-		mode->hdisplay = du->pref_width;
-		mode->vdisplay = du->pref_height;
-		vmw_guess_mode_timing(mode);
-
-		if (vmw_kms_validate_mode_vram(dev_priv,
-						mode->hdisplay * assumed_bpp,
-						mode->vdisplay)) {
-			drm_mode_probed_add(connector, mode);
-		} else {
-			drm_mode_destroy(dev, mode);
-			mode = NULL;
-		}
+	mode = drm_mode_duplicate(dev, &prefmode);
+	if (!mode)
+		return 0;
+	mode->hdisplay = du->pref_width;
+	mode->vdisplay = du->pref_height;
+	vmw_guess_mode_timing(mode);
 
-		if (du->pref_mode) {
-			list_del_init(&du->pref_mode->head);
-			drm_mode_destroy(dev, du->pref_mode);
-		}
+	if (vmw_kms_validate_mode_vram(dev_priv,
+					mode->hdisplay * assumed_bpp,
+					mode->vdisplay)) {
+		drm_mode_probed_add(connector, mode);
+	} else {
+		drm_mode_destroy(dev, mode);
+		mode = NULL;
+	}
 
-		/* mode might be null here, this is intended */
-		du->pref_mode = mode;
+	if (du->pref_mode) {
+		list_del_init(&du->pref_mode->head);
+		drm_mode_destroy(dev, du->pref_mode);
 	}
 
+	/* mode might be null here, this is intended */
+	du->pref_mode = mode;
+
 	for (i = 0; vmw_kms_connector_builtin[i].type != 0; i++) {
 		bmode = &vmw_kms_connector_builtin[i];
 		if (bmode->hdisplay > max_width ||
@@ -2036,6 +1631,7 @@ int vmw_kms_update_layout_ioctl(struct drm_device *dev, void *data,
 	int ret;
 	int i;
 	struct drm_mode_config *mode_config = &dev->mode_config;
+	struct drm_vmw_rect bounding_box = {0};
 
 	if (!arg->num_outputs) {
 		struct drm_vmw_rect def_rect = {0, 0, 800, 600};
@@ -2066,6 +1662,16 @@ int vmw_kms_update_layout_ioctl(struct drm_device *dev, void *data,
 			ret = -EINVAL;
 			goto out_free;
 		}
+
+		/*
+		 * bounding_box.w and bunding_box.h are used as
+		 * lower-right coordinates
+		 */
+		if (rects[i].x + rects[i].w > bounding_box.w)
+			bounding_box.w = rects[i].x + rects[i].w;
+
+		if (rects[i].y + rects[i].h > bounding_box.h)
+			bounding_box.h = rects[i].y + rects[i].h;
 	}
 
 	vmw_du_update_layout(dev_priv, arg->num_outputs, rects);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
index 8d038c3..0f2c291 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
@@ -1,6 +1,6 @@
 /**************************************************************************
  *
- * Copyright © 2009 VMware, Inc., Palo Alto, CA., USA
+ * Copyright © 2009-2014 VMware, Inc., Palo Alto, CA., USA
  * All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -32,11 +32,17 @@
 #include <drm/drm_crtc_helper.h>
 #include "vmwgfx_drv.h"
 
+
+
 #define VMWGFX_NUM_DISPLAY_UNITS 8
 
 
 #define vmw_framebuffer_to_vfb(x) \
 	container_of(x, struct vmw_framebuffer, base)
+#define vmw_framebuffer_to_vfbs(x) \
+	container_of(x, struct vmw_framebuffer_surface, base.base)
+#define vmw_framebuffer_to_vfbd(x) \
+	container_of(x, struct vmw_framebuffer_dmabuf, base.base)
 
 /**
  * Base class for framebuffers
@@ -53,9 +59,36 @@ struct vmw_framebuffer {
 	uint32_t user_handle;
 };
 
+/*
+ * Clip rectangle
+ */
+struct vmw_clip_rect {
+	int x1, x2, y1, y2;
+};
+
+struct vmw_framebuffer_surface {
+	struct vmw_framebuffer base;
+	struct vmw_surface *surface;
+	struct vmw_dma_buffer *buffer;
+	struct list_head head;
+	struct drm_master *master;
+};
 
-#define vmw_crtc_to_du(x) \
-	container_of(x, struct vmw_display_unit, crtc)
+
+struct vmw_framebuffer_dmabuf {
+	struct vmw_framebuffer base;
+	struct vmw_dma_buffer *buffer;
+};
+
+
+/*
+ * Basic clip rect manipulation
+ */
+void vmw_clip_cliprects(struct drm_clip_rect *rects,
+			int num_rects,
+			struct vmw_clip_rect clip,
+			SVGASignedRect *out_rects,
+			int *out_num);
 
 /*
  * Basic cursor manipulation
@@ -120,11 +153,7 @@ struct vmw_display_unit {
 /*
  * Shared display unit functions - vmwgfx_kms.c
  */
-void vmw_display_unit_cleanup(struct vmw_display_unit *du);
-int vmw_du_page_flip(struct drm_crtc *crtc,
-		     struct drm_framebuffer *fb,
-		     struct drm_pending_vblank_event *event,
-		     uint32_t page_flip_flags);
+void vmw_du_cleanup(struct vmw_display_unit *du);
 void vmw_du_crtc_save(struct drm_crtc *crtc);
 void vmw_du_crtc_restore(struct drm_crtc *crtc);
 void vmw_du_crtc_gamma_set(struct drm_crtc *crtc,
@@ -148,20 +177,31 @@ int vmw_du_connector_set_property(struct drm_connector *connector,
 /*
  * Legacy display unit functions - vmwgfx_ldu.c
  */
-int vmw_kms_init_legacy_display_system(struct vmw_private *dev_priv);
-int vmw_kms_close_legacy_display_system(struct vmw_private *dev_priv);
+int vmw_kms_ldu_init_display(struct vmw_private *dev_priv);
+int vmw_kms_ldu_close_display(struct vmw_private *dev_priv);
+int vmw_kms_ldu_do_dmabuf_dirty(struct vmw_private *dev_priv,
+				struct vmw_framebuffer *framebuffer,
+				unsigned flags, unsigned color,
+				struct drm_clip_rect *clips,
+				unsigned num_clips, int increment);
 
 /*
  * Screen Objects display functions - vmwgfx_scrn.c
  */
-int vmw_kms_init_screen_object_display(struct vmw_private *dev_priv);
-int vmw_kms_close_screen_object_display(struct vmw_private *dev_priv);
-int vmw_kms_sou_update_layout(struct vmw_private *dev_priv, unsigned num,
-			      struct drm_vmw_rect *rects);
-bool vmw_kms_screen_object_flippable(struct vmw_private *dev_priv,
-				     struct drm_crtc *crtc);
-void vmw_kms_screen_object_update_implicit_fb(struct vmw_private *dev_priv,
-					      struct drm_crtc *crtc);
-
-
+int vmw_kms_sou_init_display(struct vmw_private *dev_priv);
+int vmw_kms_sou_close_display(struct vmw_private *dev_priv);
+int vmw_kms_sou_do_surface_dirty(struct vmw_private *dev_priv,
+				 struct drm_file *file_priv,
+				 struct vmw_framebuffer *framebuffer,
+				 unsigned flags, unsigned color,
+				 struct drm_clip_rect *clips,
+				 unsigned num_clips, int inc,
+				 struct vmw_fence_obj **out_fence);
+int vmw_kms_sou_do_dmabuf_dirty(struct drm_file *file_priv,
+				struct vmw_private *dev_priv,
+				struct vmw_framebuffer *framebuffer,
+				unsigned flags, unsigned color,
+				struct drm_clip_rect *clips,
+				unsigned num_clips, int increment,
+				struct vmw_fence_obj **out_fence);
 #endif
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c b/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c
index 53579f2..f0fd565 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c
@@ -57,7 +57,7 @@ struct vmw_legacy_display_unit {
 static void vmw_ldu_destroy(struct vmw_legacy_display_unit *ldu)
 {
 	list_del_init(&ldu->active);
-	vmw_display_unit_cleanup(&ldu->base);
+	vmw_du_cleanup(&ldu->base);
 	kfree(ldu);
 }
 
@@ -386,7 +386,7 @@ static int vmw_ldu_init(struct vmw_private *dev_priv, unsigned unit)
 	return 0;
 }
 
-int vmw_kms_init_legacy_display_system(struct vmw_private *dev_priv)
+int vmw_kms_ldu_init_display(struct vmw_private *dev_priv)
 {
 	struct drm_device *dev = dev_priv->dev;
 	int i, ret;
@@ -423,6 +423,10 @@ int vmw_kms_init_legacy_display_system(struct vmw_private *dev_priv)
 	else
 		vmw_ldu_init(dev_priv, 0);
 
+	dev_priv->active_display_unit = vmw_du_legacy;
+
+	DRM_INFO("Legacy Display Unit initialized\n");
+
 	return 0;
 
 err_vblank_cleanup:
@@ -433,7 +437,7 @@ err_free:
 	return ret;
 }
 
-int vmw_kms_close_legacy_display_system(struct vmw_private *dev_priv)
+int vmw_kms_ldu_close_display(struct vmw_private *dev_priv)
 {
 	struct drm_device *dev = dev_priv->dev;
 
@@ -448,3 +452,38 @@ int vmw_kms_close_legacy_display_system(struct vmw_private *dev_priv)
 
 	return 0;
 }
+
+
+int vmw_kms_ldu_do_dmabuf_dirty(struct vmw_private *dev_priv,
+				struct vmw_framebuffer *framebuffer,
+				unsigned flags, unsigned color,
+				struct drm_clip_rect *clips,
+				unsigned num_clips, int increment)
+{
+	size_t fifo_size;
+	int i;
+
+	struct {
+		uint32_t header;
+		SVGAFifoCmdUpdate body;
+	} *cmd;
+
+	fifo_size = sizeof(*cmd) * num_clips;
+	cmd = vmw_fifo_reserve(dev_priv, fifo_size);
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Fifo reserve failed.\n");
+		return -ENOMEM;
+	}
+
+	memset(cmd, 0, fifo_size);
+	for (i = 0; i < num_clips; i++, clips += increment) {
+		cmd[i].header = cpu_to_le32(SVGA_CMD_UPDATE);
+		cmd[i].body.x = cpu_to_le32(clips->x1);
+		cmd[i].body.y = cpu_to_le32(clips->y1);
+		cmd[i].body.width = cpu_to_le32(clips->x2 - clips->x1);
+		cmd[i].body.height = cpu_to_le32(clips->y2 - clips->y1);
+	}
+
+	vmw_fifo_commit(dev_priv, fifo_size);
+	return 0;
+}
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_overlay.c b/drivers/gpu/drm/vmwgfx/vmwgfx_overlay.c
index 87e39f6..7f4b2f0 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_overlay.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_overlay.c
@@ -100,7 +100,7 @@ static int vmw_overlay_send_put(struct vmw_private *dev_priv,
 {
 	struct vmw_escape_video_flush *flush;
 	size_t fifo_size;
-	bool have_so = dev_priv->sou_priv ? true : false;
+	bool have_so = (dev_priv->active_display_unit == vmw_du_screen_object);
 	int i, num_items;
 	SVGAGuestPtr ptr;
 
@@ -231,7 +231,7 @@ static int vmw_overlay_move_buffer(struct vmw_private *dev_priv,
 	if (!pin)
 		return vmw_dmabuf_unpin(dev_priv, buf, inter);
 
-	if (!dev_priv->sou_priv)
+	if (dev_priv->active_display_unit == vmw_du_legacy)
 		return vmw_dmabuf_to_vram(dev_priv, buf, true, inter);
 
 	return vmw_dmabuf_to_vram_or_gmr(dev_priv, buf, true, inter);
@@ -453,7 +453,7 @@ int vmw_overlay_pause_all(struct vmw_private *dev_priv)
 
 static bool vmw_overlay_available(const struct vmw_private *dev_priv)
 {
-	return (dev_priv->overlay_priv != NULL && 
+	return (dev_priv->overlay_priv != NULL &&
 		((dev_priv->fifo.capabilities & VMW_OVERLAY_CAP_MASK) ==
 		 VMW_OVERLAY_CAP_MASK));
 }
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c b/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c
index 9e8eb36..807fc87 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c
@@ -1,6 +1,6 @@
 /**************************************************************************
  *
- * Copyright © 2011 VMware, Inc., Palo Alto, CA., USA
+ * Copyright © 2011-2014 VMware, Inc., Palo Alto, CA., USA
  * All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -57,7 +57,7 @@ struct vmw_screen_object_unit {
 
 static void vmw_sou_destroy(struct vmw_screen_object_unit *sou)
 {
-	vmw_display_unit_cleanup(&sou->base);
+	vmw_du_cleanup(&sou->base);
 	kfree(sou);
 }
 
@@ -72,7 +72,7 @@ static void vmw_sou_crtc_destroy(struct drm_crtc *crtc)
 }
 
 static void vmw_sou_del_active(struct vmw_private *vmw_priv,
-			      struct vmw_screen_object_unit *sou)
+			       struct vmw_screen_object_unit *sou)
 {
 	struct vmw_screen_object_display *ld = vmw_priv->sou_priv;
 
@@ -84,8 +84,8 @@ static void vmw_sou_del_active(struct vmw_private *vmw_priv,
 }
 
 static void vmw_sou_add_active(struct vmw_private *vmw_priv,
-			      struct vmw_screen_object_unit *sou,
-			      struct vmw_framebuffer *vfb)
+			       struct vmw_screen_object_unit *sou,
+			       struct vmw_framebuffer *vfb)
 {
 	struct vmw_screen_object_display *ld = vmw_priv->sou_priv;
 
@@ -274,13 +274,13 @@ static int vmw_sou_crtc_set_config(struct drm_mode_set *set)
 	dev_priv = vmw_priv(crtc->dev);
 
 	if (set->num_connectors > 1) {
-		DRM_ERROR("to many connectors\n");
+		DRM_ERROR("Too many connectors\n");
 		return -EINVAL;
 	}
 
 	if (set->num_connectors == 1 &&
 	    set->connectors[0] != &sou->base.connector) {
-		DRM_ERROR("connector doesn't match %p %p\n",
+		DRM_ERROR("Connector doesn't match %p %p\n",
 			set->connectors[0], &sou->base.connector);
 		return -EINVAL;
 	}
@@ -391,6 +391,250 @@ static int vmw_sou_crtc_set_config(struct drm_mode_set *set)
 	return 0;
 }
 
+/**
+ * Returns if this unit can be page flipped.
+ * Must be called with the mode_config mutex held.
+ */
+static bool vmw_sou_screen_object_flippable(struct vmw_private *dev_priv,
+					    struct drm_crtc *crtc)
+{
+	struct vmw_screen_object_unit *sou = vmw_crtc_to_sou(crtc);
+
+	if (!sou->base.is_implicit)
+		return true;
+
+	if (dev_priv->sou_priv->num_implicit != 1)
+		return false;
+
+	return true;
+}
+
+/**
+ * Update the implicit fb to the current fb of this crtc.
+ * Must be called with the mode_config mutex held.
+ */
+void vmw_sou_update_implicit_fb(struct vmw_private *dev_priv,
+				struct drm_crtc *crtc)
+{
+	struct vmw_screen_object_unit *sou = vmw_crtc_to_sou(crtc);
+
+	BUG_ON(!sou->base.is_implicit);
+
+	dev_priv->sou_priv->implicit_fb =
+		vmw_framebuffer_to_vfb(sou->base.crtc.primary->fb);
+}
+
+static int vmw_sou_crtc_page_flip(struct drm_crtc *crtc,
+				  struct drm_framebuffer *fb,
+				  struct drm_pending_vblank_event *event,
+				  uint32_t flags)
+{
+	struct vmw_private *dev_priv = vmw_priv(crtc->dev);
+	struct drm_framebuffer *old_fb = crtc->primary->fb;
+	struct vmw_framebuffer *vfb = vmw_framebuffer_to_vfb(fb);
+	struct drm_file *file_priv = event->base.file_priv;
+	struct vmw_fence_obj *fence = NULL;
+	struct drm_clip_rect clips;
+	int ret;
+
+	/* require ScreenObject support for page flipping */
+	if (!dev_priv->sou_priv)
+		return -ENOSYS;
+
+	if (!vmw_sou_screen_object_flippable(dev_priv, crtc))
+		return -EINVAL;
+
+	crtc->primary->fb = fb;
+
+	/* do a full screen dirty update */
+	clips.x1 = clips.y1 = 0;
+	clips.x2 = fb->width;
+	clips.y2 = fb->height;
+
+	if (vfb->dmabuf)
+		ret = vmw_kms_sou_do_dmabuf_dirty(file_priv, dev_priv, vfb,
+						  0, 0, &clips, 1, 1, &fence);
+	else
+		ret = vmw_kms_sou_do_surface_dirty(dev_priv, file_priv, vfb,
+						   0, 0, &clips, 1, 1, &fence);
+
+
+	if (ret != 0)
+		goto out_no_fence;
+	if (!fence) {
+		ret = -EINVAL;
+		goto out_no_fence;
+	}
+
+	ret = vmw_event_fence_action_queue(file_priv, fence,
+					   &event->base,
+					   &event->event.tv_sec,
+					   &event->event.tv_usec,
+					   true);
+
+	/*
+	 * No need to hold on to this now. The only cleanup
+	 * we need to do if we fail is unref the fence.
+	 */
+	vmw_fence_obj_unreference(&fence);
+
+	if (vmw_crtc_to_du(crtc)->is_implicit)
+		vmw_sou_update_implicit_fb(dev_priv, crtc);
+
+	return ret;
+
+out_no_fence:
+	crtc->primary->fb = old_fb;
+	return ret;
+}
+
+int vmw_kms_sou_do_surface_dirty(struct vmw_private *dev_priv,
+				 struct drm_file *file_priv,
+				 struct vmw_framebuffer *framebuffer,
+				 unsigned flags, unsigned color,
+				 struct drm_clip_rect *clips,
+				 unsigned num_clips, int inc,
+				 struct vmw_fence_obj **out_fence)
+{
+	struct vmw_display_unit *units[VMWGFX_NUM_DISPLAY_UNITS];
+	struct drm_clip_rect *clips_ptr;
+	struct drm_clip_rect *tmp;
+	struct drm_crtc *crtc;
+	size_t fifo_size;
+	int i, num_units;
+	int ret = 0; /* silence warning */
+	int left, right, top, bottom;
+
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdBlitSurfaceToScreen body;
+	} *cmd;
+	SVGASignedRect *blits;
+
+	num_units = 0;
+	list_for_each_entry(crtc, &dev_priv->dev->mode_config.crtc_list,
+			    head) {
+		if (crtc->primary->fb != &framebuffer->base)
+			continue;
+		units[num_units++] = vmw_crtc_to_du(crtc);
+	}
+
+	BUG_ON(!clips || !num_clips);
+
+	tmp = kzalloc(sizeof(*tmp) * num_clips, GFP_KERNEL);
+	if (unlikely(tmp == NULL)) {
+		DRM_ERROR("Temporary cliprect memory alloc failed.\n");
+		return -ENOMEM;
+	}
+
+	fifo_size = sizeof(*cmd) + sizeof(SVGASignedRect) * num_clips;
+	cmd = kzalloc(fifo_size, GFP_KERNEL);
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Temporary fifo memory alloc failed.\n");
+		ret = -ENOMEM;
+		goto out_free_tmp;
+	}
+
+	/* setup blits pointer */
+	blits = (SVGASignedRect *)&cmd[1];
+
+	/* initial clip region */
+	left = clips->x1;
+	right = clips->x2;
+	top = clips->y1;
+	bottom = clips->y2;
+
+	/* skip the first clip rect */
+	for (i = 1, clips_ptr = clips + inc;
+	     i < num_clips; i++, clips_ptr += inc) {
+		left = min_t(int, left, (int)clips_ptr->x1);
+		right = max_t(int, right, (int)clips_ptr->x2);
+		top = min_t(int, top, (int)clips_ptr->y1);
+		bottom = max_t(int, bottom, (int)clips_ptr->y2);
+	}
+
+	/* only need to do this once */
+	cmd->header.id = cpu_to_le32(SVGA_3D_CMD_BLIT_SURFACE_TO_SCREEN);
+	cmd->header.size = cpu_to_le32(fifo_size - sizeof(cmd->header));
+
+	cmd->body.srcRect.left = left;
+	cmd->body.srcRect.right = right;
+	cmd->body.srcRect.top = top;
+	cmd->body.srcRect.bottom = bottom;
+
+	clips_ptr = clips;
+	for (i = 0; i < num_clips; i++, clips_ptr += inc) {
+		tmp[i].x1 = clips_ptr->x1 - left;
+		tmp[i].x2 = clips_ptr->x2 - left;
+		tmp[i].y1 = clips_ptr->y1 - top;
+		tmp[i].y2 = clips_ptr->y2 - top;
+	}
+
+	/* do per unit writing, reuse fifo for each */
+	for (i = 0; i < num_units; i++) {
+		struct vmw_display_unit *unit = units[i];
+		struct vmw_clip_rect clip;
+		int num;
+
+		clip.x1 = left - unit->crtc.x;
+		clip.y1 = top - unit->crtc.y;
+		clip.x2 = right - unit->crtc.x;
+		clip.y2 = bottom - unit->crtc.y;
+
+		/* skip any crtcs that misses the clip region */
+		if (clip.x1 >= unit->crtc.mode.hdisplay ||
+		    clip.y1 >= unit->crtc.mode.vdisplay ||
+		    clip.x2 <= 0 || clip.y2 <= 0)
+			continue;
+
+		/*
+		 * In order for the clip rects to be correctly scaled
+		 * the src and dest rects needs to be the same size.
+		 */
+		cmd->body.destRect.left = clip.x1;
+		cmd->body.destRect.right = clip.x2;
+		cmd->body.destRect.top = clip.y1;
+		cmd->body.destRect.bottom = clip.y2;
+
+		/* create a clip rect of the crtc in dest coords */
+		clip.x2 = unit->crtc.mode.hdisplay - clip.x1;
+		clip.y2 = unit->crtc.mode.vdisplay - clip.y1;
+		clip.x1 = 0 - clip.x1;
+		clip.y1 = 0 - clip.y1;
+
+		/* need to reset sid as it is changed by execbuf */
+		cmd->body.srcImage.sid = cpu_to_le32(framebuffer->user_handle);
+		cmd->body.destScreenId = unit->unit;
+
+		/* clip and write blits to cmd stream */
+		vmw_clip_cliprects(tmp, num_clips, clip, blits, &num);
+
+		/* if no cliprects hit skip this */
+		if (num == 0)
+			continue;
+
+		/* only return the last fence */
+		if (out_fence && *out_fence)
+			vmw_fence_obj_unreference(out_fence);
+
+		/* recalculate package length */
+		fifo_size = sizeof(*cmd) + sizeof(SVGASignedRect) * num;
+		cmd->header.size = cpu_to_le32(fifo_size - sizeof(cmd->header));
+		ret = vmw_execbuf_process(file_priv, dev_priv, NULL, cmd,
+					  fifo_size, 0, NULL, out_fence);
+
+		if (unlikely(ret != 0))
+			break;
+	}
+
+
+	kfree(cmd);
+out_free_tmp:
+	kfree(tmp);
+
+	return ret;
+}
+
 static struct drm_crtc_funcs vmw_screen_object_crtc_funcs = {
 	.save = vmw_du_crtc_save,
 	.restore = vmw_du_crtc_restore,
@@ -399,7 +643,7 @@ static struct drm_crtc_funcs vmw_screen_object_crtc_funcs = {
 	.gamma_set = vmw_du_crtc_gamma_set,
 	.destroy = vmw_sou_crtc_destroy,
 	.set_config = vmw_sou_crtc_set_config,
-	.page_flip = vmw_du_page_flip,
+	.page_flip = vmw_sou_crtc_page_flip,
 };
 
 /*
@@ -424,7 +668,7 @@ static void vmw_sou_connector_destroy(struct drm_connector *connector)
 	vmw_sou_destroy(vmw_connector_to_sou(connector));
 }
 
-static struct drm_connector_funcs vmw_legacy_connector_funcs = {
+static struct drm_connector_funcs vmw_sou_connector_funcs = {
 	.dpms = vmw_du_connector_dpms,
 	.save = vmw_du_connector_save,
 	.restore = vmw_du_connector_restore,
@@ -459,7 +703,7 @@ static int vmw_sou_init(struct vmw_private *dev_priv, unsigned unit)
 	sou->base.pref_mode = NULL;
 	sou->base.is_implicit = true;
 
-	drm_connector_init(dev, connector, &vmw_legacy_connector_funcs,
+	drm_connector_init(dev, connector, &vmw_sou_connector_funcs,
 			   DRM_MODE_CONNECTOR_VIRTUAL);
 	connector->status = vmw_du_connector_detect(connector, true);
 
@@ -482,7 +726,7 @@ static int vmw_sou_init(struct vmw_private *dev_priv, unsigned unit)
 	return 0;
 }
 
-int vmw_kms_init_screen_object_display(struct vmw_private *dev_priv)
+int vmw_kms_sou_init_display(struct vmw_private *dev_priv)
 {
 	struct drm_device *dev = dev_priv->dev;
 	int i, ret;
@@ -517,7 +761,9 @@ int vmw_kms_init_screen_object_display(struct vmw_private *dev_priv)
 	for (i = 0; i < VMWGFX_NUM_DISPLAY_UNITS; ++i)
 		vmw_sou_init(dev_priv, i);
 
-	DRM_INFO("Screen objects system initialized\n");
+	dev_priv->active_display_unit = vmw_du_screen_object;
+
+	DRM_INFO("Screen Objects Display Unit initialized\n");
 
 	return 0;
 
@@ -530,7 +776,7 @@ err_no_mem:
 	return ret;
 }
 
-int vmw_kms_close_screen_object_display(struct vmw_private *dev_priv)
+int vmw_kms_sou_close_display(struct vmw_private *dev_priv)
 {
 	struct drm_device *dev = dev_priv->dev;
 
@@ -544,35 +790,143 @@ int vmw_kms_close_screen_object_display(struct vmw_private *dev_priv)
 	return 0;
 }
 
-/**
- * Returns if this unit can be page flipped.
- * Must be called with the mode_config mutex held.
- */
-bool vmw_kms_screen_object_flippable(struct vmw_private *dev_priv,
-				     struct drm_crtc *crtc)
+static int do_dmabuf_define_gmrfb(struct drm_file *file_priv,
+				  struct vmw_private *dev_priv,
+				  struct vmw_framebuffer *framebuffer)
 {
-	struct vmw_screen_object_unit *sou = vmw_crtc_to_sou(crtc);
+	int depth = framebuffer->base.depth;
+	size_t fifo_size;
+	int ret;
 
-	if (!sou->base.is_implicit)
-		return true;
+	struct {
+		uint32_t header;
+		SVGAFifoCmdDefineGMRFB body;
+	} *cmd;
 
-	if (dev_priv->sou_priv->num_implicit != 1)
-		return false;
+	/* Emulate RGBA support, contrary to svga_reg.h this is not
+	 * supported by hosts. This is only a problem if we are reading
+	 * this value later and expecting what we uploaded back.
+	 */
+	if (depth == 32)
+		depth = 24;
 
-	return true;
+	fifo_size = sizeof(*cmd);
+	cmd = kmalloc(fifo_size, GFP_KERNEL);
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Failed to allocate temporary cmd buffer.\n");
+		return -ENOMEM;
+	}
+
+	memset(cmd, 0, fifo_size);
+	cmd->header = SVGA_CMD_DEFINE_GMRFB;
+	cmd->body.format.bitsPerPixel = framebuffer->base.bits_per_pixel;
+	cmd->body.format.colorDepth = depth;
+	cmd->body.format.reserved = 0;
+	cmd->body.bytesPerLine = framebuffer->base.pitches[0];
+	cmd->body.ptr.gmrId = framebuffer->user_handle;
+	cmd->body.ptr.offset = 0;
+
+	ret = vmw_execbuf_process(file_priv, dev_priv, NULL, cmd,
+				  fifo_size, 0, NULL, NULL);
+
+	kfree(cmd);
+
+	return ret;
 }
 
-/**
- * Update the implicit fb to the current fb of this crtc.
- * Must be called with the mode_config mutex held.
- */
-void vmw_kms_screen_object_update_implicit_fb(struct vmw_private *dev_priv,
-					      struct drm_crtc *crtc)
+int vmw_kms_sou_do_dmabuf_dirty(struct drm_file *file_priv,
+				struct vmw_private *dev_priv,
+				struct vmw_framebuffer *framebuffer,
+				unsigned flags, unsigned color,
+				struct drm_clip_rect *clips,
+				unsigned num_clips, int increment,
+				struct vmw_fence_obj **out_fence)
 {
-	struct vmw_screen_object_unit *sou = vmw_crtc_to_sou(crtc);
+	struct vmw_display_unit *units[VMWGFX_NUM_DISPLAY_UNITS];
+	struct drm_clip_rect *clips_ptr;
+	int i, k, num_units, ret;
+	struct drm_crtc *crtc;
+	size_t fifo_size;
 
-	BUG_ON(!sou->base.is_implicit);
+	struct {
+		uint32_t header;
+		SVGAFifoCmdBlitGMRFBToScreen body;
+	} *blits;
 
-	dev_priv->sou_priv->implicit_fb =
-		vmw_framebuffer_to_vfb(sou->base.crtc.primary->fb);
+	ret = do_dmabuf_define_gmrfb(file_priv, dev_priv, framebuffer);
+	if (unlikely(ret != 0))
+		return ret; /* define_gmrfb prints warnings */
+
+	fifo_size = sizeof(*blits) * num_clips;
+	blits = kmalloc(fifo_size, GFP_KERNEL);
+	if (unlikely(blits == NULL)) {
+		DRM_ERROR("Failed to allocate temporary cmd buffer.\n");
+		return -ENOMEM;
+	}
+
+	num_units = 0;
+	list_for_each_entry(crtc, &dev_priv->dev->mode_config.crtc_list, head) {
+		if (crtc->primary->fb != &framebuffer->base)
+			continue;
+		units[num_units++] = vmw_crtc_to_du(crtc);
+	}
+
+	for (k = 0; k < num_units; k++) {
+		struct vmw_display_unit *unit = units[k];
+		int hit_num = 0;
+
+		clips_ptr = clips;
+		for (i = 0; i < num_clips; i++, clips_ptr += increment) {
+			int clip_x1 = clips_ptr->x1 - unit->crtc.x;
+			int clip_y1 = clips_ptr->y1 - unit->crtc.y;
+			int clip_x2 = clips_ptr->x2 - unit->crtc.x;
+			int clip_y2 = clips_ptr->y2 - unit->crtc.y;
+			int move_x, move_y;
+
+			/* skip any crtcs that misses the clip region */
+			if (clip_x1 >= unit->crtc.mode.hdisplay ||
+			    clip_y1 >= unit->crtc.mode.vdisplay ||
+			    clip_x2 <= 0 || clip_y2 <= 0)
+				continue;
+
+			/* clip size to crtc size */
+			clip_x2 = min_t(int, clip_x2, unit->crtc.mode.hdisplay);
+			clip_y2 = min_t(int, clip_y2, unit->crtc.mode.vdisplay);
+
+			/* translate both src and dest to bring clip into screen */
+			move_x = min_t(int, clip_x1, 0);
+			move_y = min_t(int, clip_y1, 0);
+
+			/* actual translate done here */
+			blits[hit_num].header = SVGA_CMD_BLIT_GMRFB_TO_SCREEN;
+			blits[hit_num].body.destScreenId = unit->unit;
+			blits[hit_num].body.srcOrigin.x = clips_ptr->x1 - move_x;
+			blits[hit_num].body.srcOrigin.y = clips_ptr->y1 - move_y;
+			blits[hit_num].body.destRect.left = clip_x1 - move_x;
+			blits[hit_num].body.destRect.top = clip_y1 - move_y;
+			blits[hit_num].body.destRect.right = clip_x2;
+			blits[hit_num].body.destRect.bottom = clip_y2;
+			hit_num++;
+		}
+
+		/* no clips hit the crtc */
+		if (hit_num == 0)
+			continue;
+
+		/* only return the last fence */
+		if (out_fence && *out_fence)
+			vmw_fence_obj_unreference(out_fence);
+
+		fifo_size = sizeof(*blits) * hit_num;
+		ret = vmw_execbuf_process(file_priv, dev_priv, NULL, blits,
+					  fifo_size, 0, NULL, out_fence);
+
+		if (unlikely(ret != 0))
+			break;
+	}
+
+	kfree(blits);
+
+	return ret;
 }
+
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 11/28] drm/vmwgfx: Add "quirk" to handling command verification exceptions

[Thomas Hellstrom]

For certain surface copies, we don't have a user space handle for
the destination surface.  In such cases, we are going to trust that
our caller is giving us the right surface ID.

To do this case, we created a quirk flag that may be useful
in the future for handling other cases.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Sinclair Yeh <syeh@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.h     |  4 ++++
 drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c | 10 ++++++++++
 drivers/gpu/drm/vmwgfx/vmwgfx_kms.c     |  4 ++--
 drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c    |  6 +++---
 4 files changed, 19 insertions(+), 5 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index b65eb02..c3f8fc9 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -341,6 +341,8 @@ enum vmw_display_unit_type {
 };
 
 
+#define VMW_QUIRK_SCREENTARGET (1U << 0)
+
 struct vmw_sw_context{
 	struct drm_open_hash res_ht;
 	bool res_ht_initialized;
@@ -363,6 +365,7 @@ 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;
@@ -831,6 +834,7 @@ 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 0792d8d..497ad6a 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
@@ -679,6 +679,10 @@ static int vmw_cmd_surface_copy_check(struct vmw_private *dev_priv,
 				&cmd->body.src.sid, NULL);
 	if (unlikely(ret != 0))
 		return ret;
+
+	if (sw_context->quirks & VMW_QUIRK_SCREENTARGET)
+		return 0;
+
 	return vmw_cmd_res_check(dev_priv, sw_context, vmw_res_surface,
 				 user_surface_converter,
 				 &cmd->body.dest.sid, NULL);
@@ -1260,6 +1264,9 @@ 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_SCREENTARGET)
+		goto out_no_surface;
+
 	ret = vmw_cmd_res_check(dev_priv, sw_context, vmw_res_surface,
 				user_surface_converter, &cmd->dma.host.sid,
 				NULL);
@@ -2544,6 +2551,7 @@ 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)
 {
@@ -2598,6 +2606,7 @@ 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;
@@ -2904,6 +2913,7 @@ 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 cac17c2..7566a5a 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
@@ -993,7 +993,7 @@ int vmw_kms_generic_present(struct vmw_private *dev_priv,
 		fifo_size = sizeof(*cmd) + sizeof(SVGASignedRect) * num;
 		cmd->header.size = cpu_to_le32(fifo_size - sizeof(cmd->header));
 		ret = vmw_execbuf_process(file_priv, dev_priv, NULL, cmd,
-					  fifo_size, 0, NULL, NULL);
+					  fifo_size, 0, 0, NULL, NULL);
 
 		if (unlikely(ret != 0))
 			break;
@@ -1121,7 +1121,7 @@ int vmw_kms_readback(struct vmw_private *dev_priv,
 	fifo_size = sizeof(*cmd) + sizeof(*blits) * blits_pos;
 
 	ret = vmw_execbuf_process(file_priv, dev_priv, NULL, cmd, fifo_size,
-				  0, user_fence_rep, NULL);
+				  0, 0, user_fence_rep, NULL);
 
 	kfree(cmd);
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c b/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c
index 807fc87..0d06d86 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c
@@ -621,7 +621,7 @@ int vmw_kms_sou_do_surface_dirty(struct vmw_private *dev_priv,
 		fifo_size = sizeof(*cmd) + sizeof(SVGASignedRect) * num;
 		cmd->header.size = cpu_to_le32(fifo_size - sizeof(cmd->header));
 		ret = vmw_execbuf_process(file_priv, dev_priv, NULL, cmd,
-					  fifo_size, 0, NULL, out_fence);
+					  fifo_size, 0, 0, NULL, out_fence);
 
 		if (unlikely(ret != 0))
 			break;
@@ -827,7 +827,7 @@ static int do_dmabuf_define_gmrfb(struct drm_file *file_priv,
 	cmd->body.ptr.offset = 0;
 
 	ret = vmw_execbuf_process(file_priv, dev_priv, NULL, cmd,
-				  fifo_size, 0, NULL, NULL);
+				  fifo_size, 0, 0, NULL, NULL);
 
 	kfree(cmd);
 
@@ -919,7 +919,7 @@ int vmw_kms_sou_do_dmabuf_dirty(struct drm_file *file_priv,
 
 		fifo_size = sizeof(*blits) * hit_num;
 		ret = vmw_execbuf_process(file_priv, dev_priv, NULL, blits,
-					  fifo_size, 0, NULL, out_fence);
+					  fifo_size, 0, 0, NULL, out_fence);
 
 		if (unlikely(ret != 0))
 			break;
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 12/28] drm/vmwgfx: Implement screen targets

[Thomas Hellstrom]

From: Sinclair Yeh <syeh@vmware.com>

Add support for the screen target device interface.
Add a getparam parameter and bump minor to signal availability.

Signed-off-by: Sinclair Yeh <syeh@vmware.com>
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
---
 drivers/gpu/drm/vmwgfx/Makefile         |    2 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.c     |   24 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.h     |   15 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c   |    4 +
 drivers/gpu/drm/vmwgfx/vmwgfx_kms.c     |   62 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_kms.h     |   20 +
 drivers/gpu/drm/vmwgfx/vmwgfx_mob.c     |    3 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c    | 1364 +++++++++++++++++++++++++++++++
 drivers/gpu/drm/vmwgfx/vmwgfx_surface.c |    4 +
 include/uapi/drm/vmwgfx_drm.h           |    1 +
 10 files changed, 1475 insertions(+), 24 deletions(-)
 create mode 100644 drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c

diff --git a/drivers/gpu/drm/vmwgfx/Makefile b/drivers/gpu/drm/vmwgfx/Makefile
index 529bc72..4840939 100644
--- a/drivers/gpu/drm/vmwgfx/Makefile
+++ b/drivers/gpu/drm/vmwgfx/Makefile
@@ -7,6 +7,6 @@ vmwgfx-y := vmwgfx_execbuf.o vmwgfx_gmr.o vmwgfx_kms.o vmwgfx_drv.o \
 	    vmwgfx_overlay.o vmwgfx_marker.o vmwgfx_gmrid_manager.o \
 	    vmwgfx_fence.o vmwgfx_dmabuf.o vmwgfx_scrn.o vmwgfx_context.o \
 	    vmwgfx_surface.o vmwgfx_prime.o vmwgfx_mob.o vmwgfx_shader.o \
-	    vmwgfx_cmdbuf_res.o vmwgfx_cmdbuf.o \
+	    vmwgfx_cmdbuf_res.o vmwgfx_cmdbuf.o vmwgfx_stdu.o \
 
 obj-$(CONFIG_DRM_VMWGFX) := vmwgfx.o
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
index 7e2b3c8..ab1b70c 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
@@ -693,22 +693,28 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 				 SVGA_REG_MAX_PRIMARY_BOUNDING_BOX_MEM);
 		dev_priv->max_mob_size =
 			vmw_read(dev_priv, SVGA_REG_MOB_MAX_SIZE);
+		dev_priv->stdu_max_width =
+			vmw_read(dev_priv, SVGA_REG_SCREENTARGET_MAX_WIDTH);
+		dev_priv->stdu_max_height =
+			vmw_read(dev_priv, SVGA_REG_SCREENTARGET_MAX_HEIGHT);
+
+		vmw_write(dev_priv, SVGA_REG_DEV_CAP,
+			  SVGA3D_DEVCAP_MAX_TEXTURE_WIDTH);
+		dev_priv->texture_max_width = vmw_read(dev_priv,
+						       SVGA_REG_DEV_CAP);
+		vmw_write(dev_priv, SVGA_REG_DEV_CAP,
+			  SVGA3D_DEVCAP_MAX_TEXTURE_HEIGHT);
+		dev_priv->texture_max_height = vmw_read(dev_priv,
+							SVGA_REG_DEV_CAP);
 	} else
 		dev_priv->prim_bb_mem = dev_priv->vram_size;
+ 
+	vmw_print_capabilities(dev_priv->capabilities);
 
 	ret = vmw_dma_masks(dev_priv);
 	if (unlikely(ret != 0))
 		goto out_err0;
 
-	/*
-	 * Limit back buffer size to VRAM size.  Remove this once
-	 * screen targets are implemented.
-	 */
-	if (dev_priv->prim_bb_mem > dev_priv->vram_size)
-		dev_priv->prim_bb_mem = dev_priv->vram_size;
-
-	vmw_print_capabilities(dev_priv->capabilities);
-
 	if (dev_priv->capabilities & SVGA_CAP_GMR2) {
 		DRM_INFO("Max GMR ids is %u\n",
 			 (unsigned)dev_priv->max_gmr_ids);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index c3f8fc9..04f8bf2 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -40,17 +40,17 @@
 #include <drm/ttm/ttm_module.h>
 #include "vmwgfx_fence.h"
 
-#define VMWGFX_DRIVER_DATE "20140704"
+#define VMWGFX_DRIVER_DATE "20150626"
 #define VMWGFX_DRIVER_MAJOR 2
-#define VMWGFX_DRIVER_MINOR 6
-#define VMWGFX_DRIVER_PATCHLEVEL 1
+#define VMWGFX_DRIVER_MINOR 7
+#define VMWGFX_DRIVER_PATCHLEVEL 0
 #define VMWGFX_FILE_PAGE_OFFSET 0x00100000
 #define VMWGFX_FIFO_STATIC_SIZE (1024*1024)
 #define VMWGFX_MAX_RELOCATIONS 2048
 #define VMWGFX_MAX_VALIDATIONS 2048
 #define VMWGFX_MAX_DISPLAYS 16
 #define VMWGFX_CMD_BOUNCE_INIT_SIZE 32768
-#define VMWGFX_ENABLE_SCREEN_TARGET_OTABLE 0
+#define VMWGFX_ENABLE_SCREEN_TARGET_OTABLE 1
 
 /*
  * Perhaps we should have sysfs entries for these.
@@ -337,7 +337,8 @@ struct vmw_ctx_binding_state {
 enum vmw_display_unit_type {
 	vmw_du_invalid = 0,
 	vmw_du_legacy,
-	vmw_du_screen_object
+	vmw_du_screen_object,
+	vmw_du_screen_target
 };
 
 
@@ -402,6 +403,10 @@ struct vmw_private {
 	uint32_t mmio_size;
 	uint32_t fb_max_width;
 	uint32_t fb_max_height;
+	uint32_t texture_max_width;
+	uint32_t texture_max_height;
+	uint32_t stdu_max_width;
+	uint32_t stdu_max_height;
 	uint32_t initial_width;
 	uint32_t initial_height;
 	__le32 __iomem *mmio_virt;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c b/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c
index 69c8ce2..55940bc 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c
@@ -105,6 +105,10 @@ int vmw_getparam_ioctl(struct drm_device *dev, void *data,
 	case DRM_VMW_PARAM_MAX_MOB_SIZE:
 		param->value = dev_priv->max_mob_size;
 		break;
+	case DRM_VMW_PARAM_SCREEN_TARGET:
+		param->value =
+			(dev_priv->active_display_unit == vmw_du_screen_target);
+		break;
 	default:
 		DRM_ERROR("Illegal vmwgfx get param request: %d\n",
 			  param->param);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
index 7566a5a..6680aa6 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
@@ -463,6 +463,11 @@ static int vmw_framebuffer_surface_dirty(struct drm_framebuffer *framebuffer,
 						   flags, color,
 						   clips, num_clips,
 						   inc, NULL);
+	else
+		ret = vmw_kms_stdu_do_surface_dirty(dev_priv, file_priv,
+						    &vfbs->base,
+						    clips, num_clips,
+						    inc);
 
 	vmw_fifo_flush(dev_priv, false);
 	ttm_read_unlock(&dev_priv->reservation_sem);
@@ -636,6 +641,11 @@ static int vmw_framebuffer_dmabuf_dirty(struct drm_framebuffer *framebuffer,
 						  flags, color,
 						  clips, num_clips, increment,
 						  NULL);
+	} else {
+		ret = vmw_kms_stdu_do_surface_dirty(dev_priv, file_priv,
+						    &vfbd->base,
+						    clips, num_clips,
+						    increment);
 	}
 
 	vmw_fifo_flush(dev_priv, false);
@@ -999,8 +1009,6 @@ int vmw_kms_generic_present(struct vmw_private *dev_priv,
 			break;
 	}
 
-	vmw_fifo_flush(dev_priv, false);
-
 	kfree(cmd);
 out_free_tmp:
 	kfree(tmp);
@@ -1017,8 +1025,21 @@ int vmw_kms_present(struct vmw_private *dev_priv,
 		    struct drm_vmw_rect *clips,
 		    uint32_t num_clips)
 {
-	return vmw_kms_generic_present(dev_priv, file_priv, vfb, surface, sid,
-				       destX, destY, clips, num_clips);
+	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);
+	if (ret)
+		return ret;
+
+	vmw_fifo_flush(dev_priv, false);
+
+	return 0;
 }
 
 int vmw_kms_readback(struct vmw_private *dev_priv,
@@ -1141,9 +1162,12 @@ int vmw_kms_init(struct vmw_private *dev_priv)
 	dev->mode_config.max_width = 8192;
 	dev->mode_config.max_height = 8192;
 
-	ret = vmw_kms_sou_init_display(dev_priv);
-	if (ret) /* Fallback */
-		ret = vmw_kms_ldu_init_display(dev_priv);
+	ret = vmw_kms_stdu_init_display(dev_priv);
+	if (ret) {
+		ret = vmw_kms_sou_init_display(dev_priv);
+		if (ret) /* Fallback */
+			ret = vmw_kms_ldu_init_display(dev_priv);
+	}
 
 	return ret;
 }
@@ -1160,6 +1184,8 @@ int vmw_kms_close(struct vmw_private *dev_priv)
 	drm_mode_config_cleanup(dev_priv->dev);
 	if (dev_priv->active_display_unit == vmw_du_screen_object)
 		ret = vmw_kms_sou_close_display(dev_priv);
+	else if (dev_priv->active_display_unit == vmw_du_screen_target)
+		ret = vmw_kms_stdu_close_display(dev_priv);
 	else
 		ret = vmw_kms_ldu_close_display(dev_priv);
 
@@ -1311,7 +1337,9 @@ bool vmw_kms_validate_mode_vram(struct vmw_private *dev_priv,
 				uint32_t pitch,
 				uint32_t height)
 {
-	return ((u64) pitch * (u64) height) < (u64) dev_priv->prim_bb_mem;
+	return ((u64) pitch * (u64) height) < (u64)
+		((dev_priv->active_display_unit == vmw_du_screen_target) ?
+		 dev_priv->prim_bb_mem : dev_priv->vram_size);
 }
 
 
@@ -1558,6 +1586,11 @@ int vmw_du_connector_fill_modes(struct drm_connector *connector,
 	if (dev_priv->active_display_unit == vmw_du_screen_object)
 		assumed_bpp = 4;
 
+	if (dev_priv->active_display_unit == vmw_du_screen_target) {
+		max_width  = min(max_width,  dev_priv->stdu_max_width);
+		max_height = min(max_height, dev_priv->stdu_max_height);
+	}
+
 	/* Add preferred mode */
 	mode = drm_mode_duplicate(dev, &prefmode);
 	if (!mode)
@@ -1674,6 +1707,19 @@ int vmw_kms_update_layout_ioctl(struct drm_device *dev, void *data,
 			bounding_box.h = rects[i].y + rects[i].h;
 	}
 
+	/*
+	 * For Screen Target Display Unit, all the displays must fit
+	 * inside of maximum texture size.
+	 */
+	if (dev_priv->active_display_unit == vmw_du_screen_target)
+		if (bounding_box.w > dev_priv->texture_max_width ||
+		    bounding_box.h > dev_priv->texture_max_height) {
+			DRM_ERROR("Layout exceeds maximum texture size\n");
+			ret = -EINVAL;
+			goto out_free;
+		}
+
+
 	vmw_du_update_layout(dev_priv, arg->num_outputs, rects);
 
 out_free:
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
index 0f2c291..548fa87 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
@@ -204,4 +204,24 @@ int vmw_kms_sou_do_dmabuf_dirty(struct drm_file *file_priv,
 				struct drm_clip_rect *clips,
 				unsigned num_clips, int increment,
 				struct vmw_fence_obj **out_fence);
+
+
+/*
+ * Screen Target Display Unit functions - vmwgfx_stdu.c
+ */
+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);
+
 #endif
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c b/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c
index 46f975e..0feac56 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c
@@ -31,7 +31,8 @@
  * If we set up the screen target otable, screen objects stop working.
  */
 
-#define VMW_OTABLE_SETUP_SUB ((VMWGFX_ENABLE_SCREEN_TARGET_OTABLE) ? 0 : 1)
+#define VMW_OTABLE_SETUP_SUB ((VMWGFX_ENABLE_SCREEN_TARGET_OTABLE &&	\
+			       (dev_priv->capabilities & SVGA_CAP_3D)) ? 0 : 1)
 
 #ifdef CONFIG_64BIT
 #define VMW_PPN_SIZE 8
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c b/drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c
new file mode 100644
index 0000000..3b8235c
--- /dev/null
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c
@@ -0,0 +1,1364 @@
+/******************************************************************************
+ *
+ * Copyright © 2014 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 "vmwgfx_kms.h"
+#include "svga3d_surfacedefs.h"
+#include <drm/drm_plane_helper.h>
+
+#define vmw_crtc_to_stdu(x) \
+	container_of(x, struct vmw_screen_target_display_unit, base.crtc)
+#define vmw_encoder_to_stdu(x) \
+	container_of(x, struct vmw_screen_target_display_unit, base.encoder)
+#define vmw_connector_to_stdu(x) \
+	container_of(x, struct vmw_screen_target_display_unit, base.connector)
+
+
+
+enum stdu_content_type {
+	SAME_AS_DISPLAY = 0,
+	SEPARATE_SURFACE,
+	SEPARATE_DMA
+};
+
+
+
+/**
+ * struct vmw_screen_target_display_unit
+ *
+ * @base: VMW specific DU structure
+ * @display_srf: surface to be displayed.  The dimension of this will always
+ *               match the display mode.  If the display mode matches
+ *               content_vfbs dimensions, then this is a pointer into the
+ *               corresponding field in content_vfbs.  If not, then this
+ *               is a separate buffer to which content_vfbs will blit to.
+ * @content_fb: holds the rendered content, can be a surface or DMA buffer
+ * @content_type:  content_fb type
+ * @defined:  true if the current display unit has been initialized
+ */
+struct vmw_screen_target_display_unit {
+	struct vmw_display_unit base;
+
+	struct vmw_surface     *display_srf;
+	struct drm_framebuffer *content_fb;
+
+	enum stdu_content_type content_fb_type;
+
+	bool defined;
+};
+
+
+
+static void vmw_stdu_destroy(struct vmw_screen_target_display_unit *stdu);
+
+
+
+/******************************************************************************
+ * Screen Target Display Unit helper Functions
+ *****************************************************************************/
+
+/**
+ * vmw_stdu_pin_display - pins the resource associated with the display surface
+ *
+ * @stdu: contains the display surface
+ *
+ * Since the display surface can either be a private surface allocated by us,
+ * or it can point to the content surface, we use this function to not pin the
+ * same resource twice.
+ */
+static int vmw_stdu_pin_display(struct vmw_screen_target_display_unit *stdu)
+{
+	return vmw_resource_pin(&stdu->display_srf->res);
+}
+
+
+
+/**
+ * vmw_stdu_unpin_display - unpins the resource associated with display surface
+ *
+ * @stdu: contains the display surface
+ *
+ * If the display surface was privatedly allocated by
+ * vmw_surface_gb_priv_define() and not registered as a framebuffer, then it
+ * won't be automatically cleaned up when all the framebuffers are freed.  As
+ * such, we have to explicitly call vmw_resource_unreference() to get it freed.
+ */
+static void vmw_stdu_unpin_display(struct vmw_screen_target_display_unit *stdu)
+{
+	if (stdu->display_srf) {
+		struct vmw_resource *res = &stdu->display_srf->res;
+
+		vmw_resource_unpin(res);
+
+		if (stdu->content_fb_type != SAME_AS_DISPLAY) {
+			vmw_resource_unreference(&res);
+			stdu->content_fb_type = SAME_AS_DISPLAY;
+		}
+
+		stdu->display_srf = NULL;
+	}
+}
+
+
+
+/******************************************************************************
+ * Screen Target Display Unit CRTC Functions
+ *****************************************************************************/
+
+
+/**
+ * vmw_stdu_crtc_destroy - cleans up the STDU
+ *
+ * @crtc: used to get a reference to the containing STDU
+ */
+static void vmw_stdu_crtc_destroy(struct drm_crtc *crtc)
+{
+	vmw_stdu_destroy(vmw_crtc_to_stdu(crtc));
+}
+
+
+
+/**
+ * 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)
+{
+	size_t fifo_size;
+	int ret;
+	void *cmd;
+
+	struct vmw_surface_dma {
+		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 vmw_framebuffer_dmabuf *content_vfbd;
+		struct vmw_framebuffer_surface *content_vfbs;
+		struct drm_vmw_size cur_size = {0};
+		const struct svga3d_surface_desc *desc;
+		SVGA3dCmdSurfaceDMASuffix *suffix;
+		SVGAGuestPtr ptr;
+
+		content_vfb  = vmw_framebuffer_to_vfb(stdu->content_fb);
+		content_vfbd = vmw_framebuffer_to_vfbd(stdu->content_fb);
+		content_vfbs = vmw_framebuffer_to_vfbs(stdu->content_fb);
+
+		cur_size.width  = width;
+		cur_size.height = height;
+		cur_size.depth  = 1;
+
+		desc = svga3dsurface_get_desc(content_vfbs->surface->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 {
+		struct vmw_framebuffer *content_vfb;
+
+		content_vfb = vmw_framebuffer_to_vfb(stdu->content_fb);
+
+		fifo_size = sizeof(surface_cpy_cmd);
+
+		memset(&surface_cpy_cmd, 0, sizeof(surface_cpy_cmd));
+
+		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  = content_vfb->user_handle;
+		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, cmd,
+				  fifo_size, 0, VMW_QUIRK_SCREENTARGET,
+				  NULL, NULL);
+
+	return ret;
+}
+
+
+
+/**
+ * vmw_stdu_define_st - Defines a Screen Target
+ *
+ * @dev_priv:  VMW DRM device
+ * @stdu: display unit to create a Screen Target for
+ *
+ * Creates a STDU that we can used later.  This function is called whenever the
+ * framebuffer size changes.
+ *
+ * RETURNs:
+ * 0 on success, error code on failure
+ */
+static int vmw_stdu_define_st(struct vmw_private *dev_priv,
+			      struct vmw_screen_target_display_unit *stdu)
+{
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdDefineGBScreenTarget body;
+	} *cmd;
+
+	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
+
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Out of FIFO space defining Screen Target\n");
+		return -ENOMEM;
+	}
+
+	cmd->header.id   = SVGA_3D_CMD_DEFINE_GB_SCREENTARGET;
+	cmd->header.size = sizeof(cmd->body);
+
+	cmd->body.stid   = stdu->base.unit;
+	cmd->body.width  = stdu->display_srf->base_size.width;
+	cmd->body.height = stdu->display_srf->base_size.height;
+	cmd->body.flags  = (0 == cmd->body.stid) ? SVGA_STFLAG_PRIMARY : 0;
+	cmd->body.dpi    = 0;
+	cmd->body.xRoot  = stdu->base.crtc.x;
+	cmd->body.yRoot  = stdu->base.crtc.y;
+
+	if (!stdu->base.is_implicit) {
+		cmd->body.xRoot  = stdu->base.gui_x;
+		cmd->body.yRoot  = stdu->base.gui_y;
+	}
+
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+
+	stdu->defined = true;
+
+	return 0;
+}
+
+
+
+/**
+ * vmw_stdu_bind_st - Binds a surface to a Screen Target
+ *
+ * @dev_priv: VMW DRM device
+ * @stdu: display unit affected
+ * @res: Buffer to bind to the screen target.  Set to NULL to blank screen.
+ *
+ * Binding a surface to a Screen Target the same as flipping
+ */
+static int vmw_stdu_bind_st(struct vmw_private *dev_priv,
+			    struct vmw_screen_target_display_unit *stdu,
+			    struct vmw_resource *res)
+{
+	SVGA3dSurfaceImageId image;
+
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdBindGBScreenTarget body;
+	} *cmd;
+
+
+	if (!stdu->defined) {
+		DRM_ERROR("No screen target defined\n");
+		return -EINVAL;
+	}
+
+	/* Set up image using information in vfb */
+	memset(&image, 0, sizeof(image));
+	image.sid = res ? res->id : SVGA3D_INVALID_ID;
+
+	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
+
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Out of FIFO space binding a screen target\n");
+		return -ENOMEM;
+	}
+
+	cmd->header.id   = SVGA_3D_CMD_BIND_GB_SCREENTARGET;
+	cmd->header.size = sizeof(cmd->body);
+
+	cmd->body.stid   = stdu->base.unit;
+	cmd->body.image  = image;
+
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+
+	return 0;
+}
+
+
+
+/**
+ * vmw_stdu_update_st - Updates a Screen Target
+ *
+ * @dev_priv: VMW DRM device
+ * @file_priv: Pointer to a drm file private structure
+ * @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.
+ *
+ * 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)
+{
+	u32 width, height;
+	u32 display_update_x, display_update_y;
+	unsigned short display_x1, display_y1, display_x2, display_y2;
+
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdUpdateGBScreenTarget body;
+	} *cmd;
+
+
+	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 (file_priv && stdu->content_fb_type != SAME_AS_DISPLAY) {
+		int ret;
+
+		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;
+		}
+	}
+
+	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
+
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Out of FIFO space updating a Screen Target\n");
+		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_fifo_commit(dev_priv, sizeof(*cmd));
+
+	return 0;
+}
+
+
+
+/**
+ * vmw_stdu_destroy_st - Destroy a Screen Target
+ *
+ * @dev_priv:  VMW DRM device
+ * @stdu: display unit to destroy
+ */
+static int vmw_stdu_destroy_st(struct vmw_private *dev_priv,
+			       struct vmw_screen_target_display_unit *stdu)
+{
+	int    ret;
+
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdDestroyGBScreenTarget body;
+	} *cmd;
+
+
+	/* Nothing to do if not successfully defined */
+	if (unlikely(!stdu->defined))
+		return 0;
+
+	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
+
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Out of FIFO space, screen target not destroyed\n");
+		return -ENOMEM;
+	}
+
+	cmd->header.id   = SVGA_3D_CMD_DESTROY_GB_SCREENTARGET;
+	cmd->header.size = sizeof(cmd->body);
+
+	cmd->body.stid   = stdu->base.unit;
+
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+
+	/* Force sync */
+	ret = vmw_fallback_wait(dev_priv, false, true, 0, false, 3*HZ);
+	if (unlikely(ret != 0))
+		DRM_ERROR("Failed to sync with HW");
+
+	stdu->defined = false;
+
+	return ret;
+}
+
+
+
+/**
+ * vmw_stdu_crtc_set_config - Sets a mode
+ *
+ * @set:  mode parameters
+ *
+ * This function is the device-specific portion of the DRM CRTC mode set.
+ * For the SVGA device, we do this by defining a Screen Target, binding a
+ * GB Surface to that target, and finally update the screen target.
+ *
+ * RETURNS:
+ * 0 on success, error code otherwise
+ */
+static int vmw_stdu_crtc_set_config(struct drm_mode_set *set)
+{
+	struct vmw_private *dev_priv;
+	struct vmw_screen_target_display_unit *stdu;
+	struct vmw_framebuffer *vfb;
+	struct vmw_framebuffer_surface *new_vfbs;
+	struct drm_display_mode *mode;
+	struct drm_framebuffer  *new_fb;
+	struct drm_crtc      *crtc;
+	struct drm_encoder   *encoder;
+	struct drm_connector *connector;
+	struct drm_clip_rect update_area = {0};
+	int    ret;
+
+
+	if (!set || !set->crtc)
+		return -EINVAL;
+
+	crtc     = set->crtc;
+	crtc->x  = set->x;
+	crtc->y  = set->y;
+	stdu     = vmw_crtc_to_stdu(crtc);
+	mode     = set->mode;
+	new_fb   = set->fb;
+	dev_priv = vmw_priv(crtc->dev);
+
+
+	if (set->num_connectors > 1) {
+		DRM_ERROR("Too many connectors\n");
+		return -EINVAL;
+	}
+
+	if (set->num_connectors == 1 &&
+	    set->connectors[0] != &stdu->base.connector) {
+		DRM_ERROR("Connectors don't match %p %p\n",
+			set->connectors[0], &stdu->base.connector);
+		return -EINVAL;
+	}
+
+
+	/* Since they always map one to one these are safe */
+	connector = &stdu->base.connector;
+	encoder   = &stdu->base.encoder;
+
+
+	/*
+	 * After this point the CRTC will be considered off unless a new fb
+	 * is bound
+	 */
+	if (stdu->defined) {
+		/* Unbind current surface by binding an invalid one */
+		ret = vmw_stdu_bind_st(dev_priv, stdu, NULL);
+		if (unlikely(ret != 0))
+			return ret;
+
+		/* 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);
+			if (unlikely(ret != 0))
+				return ret;
+		}
+
+		crtc->primary->fb  = NULL;
+		crtc->enabled      = false;
+		encoder->crtc      = NULL;
+		connector->encoder = NULL;
+
+		vmw_stdu_unpin_display(stdu);
+		stdu->content_fb      = NULL;
+		stdu->content_fb_type = SAME_AS_DISPLAY;
+
+		ret = vmw_stdu_destroy_st(dev_priv, stdu);
+		/* The hardware is hung, give up */
+		if (unlikely(ret != 0))
+			return ret;
+	}
+
+
+	/* Any of these conditions means the caller wants CRTC off */
+	if (set->num_connectors == 0 || !mode || !new_fb)
+		return 0;
+
+
+	if (set->x + mode->hdisplay > new_fb->width ||
+	    set->y + mode->vdisplay > new_fb->height) {
+		DRM_ERROR("Set outside of framebuffer\n");
+		return -EINVAL;
+	}
+
+	stdu->content_fb = new_fb;
+	vfb = vmw_framebuffer_to_vfb(stdu->content_fb);
+
+	if (vfb->dmabuf)
+		stdu->content_fb_type = SEPARATE_DMA;
+
+	/*
+	 * If the requested mode is different than the width and height
+	 * of the FB or if the content buffer is a DMA buf, then allocate
+	 * a display FB that matches the dimension of the mode
+	 */
+	if (mode->hdisplay != new_fb->width  ||
+	    mode->vdisplay != new_fb->height ||
+	    stdu->content_fb_type != SAME_AS_DISPLAY) {
+		struct vmw_surface content_srf;
+		struct drm_vmw_size display_base_size = {0};
+		struct vmw_surface *display_srf;
+
+
+		display_base_size.width  = mode->hdisplay;
+		display_base_size.height = mode->vdisplay;
+		display_base_size.depth  = 1;
+
+		/*
+		 * If content buffer is a DMA buf, then we have to construct
+		 * surface info
+		 */
+		if (stdu->content_fb_type == SEPARATE_DMA) {
+
+			switch (new_fb->bits_per_pixel) {
+			case 32:
+				content_srf.format = SVGA3D_X8R8G8B8;
+				break;
+
+			case 16:
+				content_srf.format = SVGA3D_R5G6B5;
+				break;
+
+			case 8:
+				content_srf.format = SVGA3D_P8;
+				break;
+
+			default:
+				DRM_ERROR("Invalid format\n");
+				ret = -EINVAL;
+				goto err_unref_content;
+			}
+
+			content_srf.flags             = 0;
+			content_srf.mip_levels[0]     = 1;
+			content_srf.multisample_count = 0;
+		} else {
+
+			stdu->content_fb_type = SEPARATE_SURFACE;
+
+			new_vfbs = vmw_framebuffer_to_vfbs(new_fb);
+			content_srf = *new_vfbs->surface;
+		}
+
+
+		ret = vmw_surface_gb_priv_define(crtc->dev,
+				0, /* because kernel visible only */
+				content_srf.flags,
+				content_srf.format,
+				true, /* a scanout buffer */
+				content_srf.mip_levels[0],
+				content_srf.multisample_count,
+				display_base_size,
+				&display_srf);
+		if (unlikely(ret != 0)) {
+			DRM_ERROR("Cannot allocate a display FB.\n");
+			goto err_unref_content;
+		}
+
+		stdu->display_srf = display_srf;
+	} else {
+		new_vfbs = vmw_framebuffer_to_vfbs(new_fb);
+		stdu->display_srf = new_vfbs->surface;
+	}
+
+
+	ret = vmw_stdu_pin_display(stdu);
+	if (unlikely(ret != 0)) {
+		stdu->display_srf = NULL;
+		goto err_unref_content;
+	}
+
+	vmw_fb_off(dev_priv);
+	vmw_svga_enable(dev_priv);
+
+	/*
+	 * Steps to displaying a surface, assume surface is already
+	 * bound:
+	 *   1.  define a screen target
+	 *   2.  bind a fb to the screen target
+	 *   3.  update that screen target (this is done later by
+	 *       vmw_kms_stdu_do_surface_dirty_or_present)
+	 */
+	ret = vmw_stdu_define_st(dev_priv, stdu);
+	if (unlikely(ret != 0))
+		goto err_unpin_display_and_content;
+
+	ret = vmw_stdu_bind_st(dev_priv, stdu, &stdu->display_srf->res);
+	if (unlikely(ret != 0))
+		goto err_unpin_destroy_st;
+
+
+	connector->encoder = encoder;
+	encoder->crtc      = crtc;
+
+	crtc->mode    = *mode;
+	crtc->primary->fb = new_fb;
+	crtc->enabled = true;
+
+	return ret;
+
+err_unpin_destroy_st:
+	vmw_stdu_destroy_st(dev_priv, stdu);
+err_unpin_display_and_content:
+	vmw_stdu_unpin_display(stdu);
+err_unref_content:
+	stdu->content_fb = NULL;
+	return ret;
+}
+
+
+
+/**
+ * vmw_stdu_crtc_page_flip - Binds a buffer to a screen target
+ *
+ * @crtc: CRTC to attach FB to
+ * @fb: FB to attach
+ * @event: Event to be posted. This event should've been alloced
+ *         using k[mz]alloc, and should've been completely initialized.
+ * @page_flip_flags: Input flags.
+ *
+ * If the STDU uses the same display and content buffers, i.e. a true flip,
+ * this function will replace the existing display buffer with the new content
+ * buffer.
+ *
+ * If the STDU uses different display and content buffers, i.e. a blit, then
+ * only the content buffer will be updated.
+ *
+ * RETURNS:
+ * 0 on success, error code on failure
+ */
+static int vmw_stdu_crtc_page_flip(struct drm_crtc *crtc,
+				   struct drm_framebuffer *new_fb,
+				   struct drm_pending_vblank_event *event,
+				   uint32_t flags)
+
+{
+	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;
+
+	dev_priv          = vmw_priv(crtc->dev);
+	stdu              = vmw_crtc_to_stdu(crtc);
+	crtc->primary->fb = new_fb;
+	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
+		 */
+		if (stdu->content_fb_type == SAME_AS_DISPLAY) {
+			if (stdu->defined) {
+				/* Unbind the current surface */
+				ret = vmw_stdu_bind_st(dev_priv, stdu, NULL);
+				if (unlikely(ret != 0))
+					goto err_out;
+			}
+			vmw_stdu_unpin_display(stdu);
+			stdu->display_srf = NULL;
+		}
+	}
+
+
+	if (!new_fb) {
+		/* Blanks the display */
+		(void) vmw_stdu_update_st(dev_priv, NULL, stdu, &update_area);
+
+		return 0;
+	}
+
+
+	if (stdu->content_fb_type == SAME_AS_DISPLAY) {
+		stdu->display_srf = vmw_framebuffer_to_vfbs(new_fb)->surface;
+		ret = vmw_stdu_pin_display(stdu);
+		if (ret) {
+			stdu->display_srf = NULL;
+			goto err_out;
+		}
+
+		/* Bind display surface */
+		ret = vmw_stdu_bind_st(dev_priv, stdu, &stdu->display_srf->res);
+		if (unlikely(ret != 0))
+			goto err_unpin_display_and_content;
+	}
+
+	/* 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);
+	if (unlikely(ret != 0))
+		return ret;
+
+	if (event) {
+		struct vmw_fence_obj *fence = NULL;
+
+		vmw_execbuf_fence_commands(NULL, dev_priv, &fence, NULL);
+		if (!fence)
+			return -ENOMEM;
+
+		ret = vmw_event_fence_action_queue(file_priv, fence,
+						   &event->base,
+						   &event->event.tv_sec,
+						   &event->event.tv_usec,
+						   true);
+		vmw_fence_obj_unreference(&fence);
+	}
+
+	return ret;
+
+err_unpin_display_and_content:
+	vmw_stdu_unpin_display(stdu);
+err_out:
+	crtc->primary->fb = NULL;
+	stdu->content_fb = NULL;
+	return ret;
+}
+
+
+
+/*
+ *  Screen Target CRTC dispatch table
+ */
+static struct drm_crtc_funcs vmw_stdu_crtc_funcs = {
+	.save = vmw_du_crtc_save,
+	.restore = vmw_du_crtc_restore,
+	.cursor_set = vmw_du_crtc_cursor_set,
+	.cursor_move = vmw_du_crtc_cursor_move,
+	.gamma_set = vmw_du_crtc_gamma_set,
+	.destroy = vmw_stdu_crtc_destroy,
+	.set_config = vmw_stdu_crtc_set_config,
+	.page_flip = vmw_stdu_crtc_page_flip,
+};
+
+
+
+/******************************************************************************
+ * Screen Target Display Unit Encoder Functions
+ *****************************************************************************/
+
+/**
+ * vmw_stdu_encoder_destroy - cleans up the STDU
+ *
+ * @encoder: used the get the containing STDU
+ *
+ * vmwgfx cleans up crtc/encoder/connector all at the same time so technically
+ * this can be a no-op.  Nevertheless, it doesn't hurt of have this in case
+ * the common KMS code changes and somehow vmw_stdu_crtc_destroy() doesn't
+ * get called.
+ */
+static void vmw_stdu_encoder_destroy(struct drm_encoder *encoder)
+{
+	vmw_stdu_destroy(vmw_encoder_to_stdu(encoder));
+}
+
+static struct drm_encoder_funcs vmw_stdu_encoder_funcs = {
+	.destroy = vmw_stdu_encoder_destroy,
+};
+
+
+
+/******************************************************************************
+ * Screen Target Display Unit Connector Functions
+ *****************************************************************************/
+
+/**
+ * vmw_stdu_connector_destroy - cleans up the STDU
+ *
+ * @connector: used to get the containing STDU
+ *
+ * vmwgfx cleans up crtc/encoder/connector all at the same time so technically
+ * this can be a no-op.  Nevertheless, it doesn't hurt of have this in case
+ * the common KMS code changes and somehow vmw_stdu_crtc_destroy() doesn't
+ * get called.
+ */
+static void vmw_stdu_connector_destroy(struct drm_connector *connector)
+{
+	vmw_stdu_destroy(vmw_connector_to_stdu(connector));
+}
+
+
+
+static struct drm_connector_funcs vmw_stdu_connector_funcs = {
+	.dpms = vmw_du_connector_dpms,
+	.save = vmw_du_connector_save,
+	.restore = vmw_du_connector_restore,
+	.detect = vmw_du_connector_detect,
+	.fill_modes = vmw_du_connector_fill_modes,
+	.set_property = vmw_du_connector_set_property,
+	.destroy = vmw_stdu_connector_destroy,
+};
+
+
+
+/**
+ * vmw_stdu_init - Sets up a Screen Target Display Unit
+ *
+ * @dev_priv: VMW DRM device
+ * @unit: unit number range from 0 to VMWGFX_NUM_DISPLAY_UNITS
+ *
+ * This function is called once per CRTC, and allocates one Screen Target
+ * display unit to represent that CRTC.  Since the SVGA device does not separate
+ * out encoder and connector, they are represented as part of the STDU as well.
+ */
+static int vmw_stdu_init(struct vmw_private *dev_priv, unsigned unit)
+{
+	struct vmw_screen_target_display_unit *stdu;
+	struct drm_device *dev = dev_priv->dev;
+	struct drm_connector *connector;
+	struct drm_encoder *encoder;
+	struct drm_crtc *crtc;
+
+
+	stdu = kzalloc(sizeof(*stdu), GFP_KERNEL);
+	if (!stdu)
+		return -ENOMEM;
+
+	stdu->base.unit = unit;
+	crtc = &stdu->base.crtc;
+	encoder = &stdu->base.encoder;
+	connector = &stdu->base.connector;
+
+	stdu->base.pref_active = (unit == 0);
+	stdu->base.pref_width  = dev_priv->initial_width;
+	stdu->base.pref_height = dev_priv->initial_height;
+	stdu->base.pref_mode   = NULL;
+	stdu->base.is_implicit = true;
+
+	drm_connector_init(dev, connector, &vmw_stdu_connector_funcs,
+			   DRM_MODE_CONNECTOR_VIRTUAL);
+	connector->status = vmw_du_connector_detect(connector, false);
+
+	drm_encoder_init(dev, encoder, &vmw_stdu_encoder_funcs,
+			 DRM_MODE_ENCODER_VIRTUAL);
+	drm_mode_connector_attach_encoder(connector, encoder);
+	encoder->possible_crtcs = (1 << unit);
+	encoder->possible_clones = 0;
+
+	(void) drm_connector_register(connector);
+
+	drm_crtc_init(dev, crtc, &vmw_stdu_crtc_funcs);
+
+	drm_mode_crtc_set_gamma_size(crtc, 256);
+
+	drm_object_attach_property(&connector->base,
+				   dev->mode_config.dirty_info_property,
+				   1);
+
+	return 0;
+}
+
+
+
+/**
+ *  vmw_stdu_destroy - Cleans up a vmw_screen_target_display_unit
+ *
+ *  @stdu:  Screen Target Display Unit to be destroyed
+ *
+ *  Clean up after vmw_stdu_init
+ */
+static void vmw_stdu_destroy(struct vmw_screen_target_display_unit *stdu)
+{
+	vmw_stdu_unpin_display(stdu);
+
+	vmw_du_cleanup(&stdu->base);
+	kfree(stdu);
+}
+
+
+
+/******************************************************************************
+ * Screen Target Display KMS Functions
+ *
+ * These functions are called by the common KMS code in vmwgfx_kms.c
+ *****************************************************************************/
+
+/**
+ * vmw_kms_stdu_init_display - Initializes a Screen Target based display
+ *
+ * @dev_priv: VMW DRM device
+ *
+ * This function initialize a Screen Target based display device.  It checks
+ * the capability bits to make sure the underlying hardware can support
+ * screen targets, and then creates the maximum number of CRTCs, a.k.a Display
+ * Units, as supported by the display hardware.
+ *
+ * RETURNS:
+ * 0 on success, error code otherwise
+ */
+int vmw_kms_stdu_init_display(struct vmw_private *dev_priv)
+{
+	struct drm_device *dev = dev_priv->dev;
+	int i, ret;
+
+
+	/* Do nothing if Screen Target support is turned off */
+	if (!VMWGFX_ENABLE_SCREEN_TARGET_OTABLE)
+		return -ENOSYS;
+
+	if (!(dev_priv->capabilities & SVGA_CAP_GBOBJECTS) ||
+	    !(dev_priv->capabilities & SVGA_CAP_3D))
+		return -ENOSYS;
+
+	ret = drm_vblank_init(dev, VMWGFX_NUM_DISPLAY_UNITS);
+	if (unlikely(ret != 0))
+		return ret;
+
+	ret = drm_mode_create_dirty_info_property(dev);
+	if (unlikely(ret != 0))
+		goto err_vblank_cleanup;
+
+	for (i = 0; i < VMWGFX_NUM_DISPLAY_UNITS; ++i) {
+		ret = vmw_stdu_init(dev_priv, i);
+
+		if (unlikely(ret != 0)) {
+			DRM_ERROR("Failed to initialize STDU %d", i);
+			goto err_vblank_cleanup;
+		}
+	}
+
+	dev_priv->active_display_unit = vmw_du_screen_target;
+
+	DRM_INFO("Screen Target Display device initialized\n");
+
+	return 0;
+
+err_vblank_cleanup:
+	drm_vblank_cleanup(dev);
+	return ret;
+}
+
+
+
+/**
+ * vmw_kms_stdu_close_display - Cleans up after vmw_kms_stdu_init_display
+ *
+ * @dev_priv: VMW DRM device
+ *
+ * Frees up any resources allocated by vmw_kms_stdu_init_display
+ *
+ * RETURNS:
+ * 0 on success
+ */
+int vmw_kms_stdu_close_display(struct vmw_private *dev_priv)
+{
+	struct drm_device *dev = dev_priv->dev;
+
+	drm_vblank_cleanup(dev);
+
+	return 0;
+}
+
+
+
+/**
+ * vmw_kms_stdu_do_surface_dirty - updates a dirty rectange to SVGA device
+ *
+ * @dev_priv: VMW DRM device
+ * @file_priv: Pointer to a drm file private structure
+ * @framebuffer: FB with the new content to be copied to SVGA device
+ * @clip_rects: array of dirty rectanges
+ * @num_of_clip_rects: number of rectanges in @clips
+ * @increment: increment to the next dirty rect in @clips
+ *
+ * This function sends an Update command to the SVGA device.  This will notify
+ * the device that a region needs to be copied to the screen.  At this time
+ * we are not coalescing clip rects into one large clip rect because the SVGA
+ * device will do it for us.
+ *
+ * RETURNS:
+ * 0 on success, error code otherwise
+ */
+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 *clip_rects,
+				  unsigned num_of_clip_rects, int increment)
+{
+	struct vmw_screen_target_display_unit *stdu[VMWGFX_NUM_DISPLAY_UNITS];
+	struct drm_clip_rect *cur_rect;
+	struct drm_crtc *crtc;
+
+	unsigned num_of_du = 0, cur_du, count = 0;
+	int      ret = 0;
+
+
+	BUG_ON(!clip_rects || !num_of_clip_rects);
+
+	/* Figure out all the DU affected by this surface */
+	list_for_each_entry(crtc, &dev_priv->dev->mode_config.crtc_list,
+			    head) {
+		if (crtc->primary->fb != &framebuffer->base)
+			continue;
+
+		stdu[num_of_du++] = vmw_crtc_to_stdu(crtc);
+	}
+
+	for (cur_du = 0; cur_du < num_of_du; cur_du++)
+		for (cur_rect = clip_rects, count = 0;
+		     count < num_of_clip_rects && ret == 0;
+		     cur_rect += increment, count++) {
+			ret = vmw_stdu_update_st(dev_priv, file_priv,
+						 stdu[cur_du],
+						 cur_rect);
+		}
+
+	return ret;
+}
+
+
+
+/**
+ * vmw_kms_stdu_present - present a surface to the display surface
+ *
+ * @dev_priv: VMW DRM device
+ * @file_priv: Pointer to a drm file private structure
+ * @vfb: Used to pick which STDU(s) is affected
+ * @user_handle: user handle for the source surface
+ * @dest_x: top/left corner of the display area to blit to
+ * @dest_y: top/left corner of the display area to blit to
+ * @clip_rects: array of dirty rectanges
+ * @num_of_clip_rects: number of rectanges in @clips
+ *
+ * This function copies a surface onto the display surface, and
+ * updates the screen target.  Strech blit is currently not
+ * supported.
+ *
+ * RETURNS:
+ * 0 on success, error code otherwise
+ */
+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 *clip_rects,
+			 uint32_t num_of_clip_rects)
+{
+	struct vmw_screen_target_display_unit *stdu[VMWGFX_NUM_DISPLAY_UNITS];
+	struct drm_clip_rect *update_area;
+	struct drm_crtc *crtc;
+	size_t fifo_size;
+	int num_of_du = 0, cur_du, i;
+	int ret = 0;
+	struct vmw_clip_rect src_bb;
+
+	struct {
+		SVGA3dCmdHeader      header;
+		SVGA3dCmdSurfaceCopy body;
+	} *cmd;
+	SVGA3dCopyBox *blits;
+
+
+	BUG_ON(!clip_rects || !num_of_clip_rects);
+
+	list_for_each_entry(crtc, &dev_priv->dev->mode_config.crtc_list, head) {
+		if (crtc->primary->fb != &vfb->base)
+			continue;
+
+		stdu[num_of_du++] = vmw_crtc_to_stdu(crtc);
+	}
+
+
+	update_area = kcalloc(num_of_clip_rects, sizeof(*update_area),
+			      GFP_KERNEL);
+	if (unlikely(update_area == NULL)) {
+		DRM_ERROR("Temporary clip rect memory alloc failed.\n");
+		return -ENOMEM;
+	}
+
+
+	fifo_size = sizeof(*cmd) + sizeof(SVGA3dCopyBox) * num_of_clip_rects;
+
+	cmd = kmalloc(fifo_size, GFP_KERNEL);
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Failed to allocate memory for surface copy.\n");
+		ret = -ENOMEM;
+		goto out_free_update_area;
+	}
+
+	memset(cmd, 0, fifo_size);
+	cmd->header.id = SVGA_3D_CMD_SURFACE_COPY;
+
+	blits = (SVGA3dCopyBox *)&cmd[1];
+
+
+	/* Figure out the source bounding box */
+	src_bb.x1 = clip_rects->x;
+	src_bb.y1 = clip_rects->y;
+	src_bb.x2 = clip_rects->x + clip_rects->w;
+	src_bb.y2 = clip_rects->y + clip_rects->h;
+
+	for (i = 1; i < num_of_clip_rects; i++) {
+		src_bb.x1 = min_t(int, src_bb.x1, clip_rects[i].x);
+		src_bb.x2 = max_t(int, src_bb.x2,
+				  clip_rects[i].x + (int) clip_rects[i].w);
+		src_bb.y1 = min_t(int, src_bb.y1, clip_rects[i].y);
+		src_bb.y2 = max_t(int, src_bb.y2,
+				  clip_rects[i].y + (int) clip_rects[i].h);
+	}
+
+	for (i = 0; i < num_of_clip_rects; i++) {
+		update_area[i].x1 = clip_rects[i].x - src_bb.x1;
+		update_area[i].x2 = update_area[i].x1 + clip_rects[i].w;
+		update_area[i].y1 = clip_rects[i].y - src_bb.y1;
+		update_area[i].y2 = update_area[i].y1 + clip_rects[i].h;
+	}
+
+
+	for (cur_du = 0; cur_du < num_of_du; cur_du++) {
+		struct vmw_clip_rect dest_bb;
+		int num_of_blits;
+
+		crtc = &stdu[cur_du]->base.crtc;
+
+		dest_bb.x1 = src_bb.x1 + dest_x - crtc->x;
+		dest_bb.y1 = src_bb.y1 + dest_y - crtc->y;
+		dest_bb.x2 = src_bb.x2 + dest_x - crtc->x;
+		dest_bb.y2 = src_bb.y2 + dest_y - crtc->y;
+
+		/* Skip any STDU outside of the destination bounding box */
+		if (dest_bb.x1 >= crtc->mode.hdisplay ||
+		    dest_bb.y1 >= crtc->mode.vdisplay ||
+		    dest_bb.x2 <= 0 || dest_bb.y2 <= 0)
+			continue;
+
+		/* Normalize to top-left of src bounding box in dest coord */
+		dest_bb.x2 = crtc->mode.hdisplay - dest_bb.x1;
+		dest_bb.y2 = crtc->mode.vdisplay - dest_bb.y1;
+		dest_bb.x1 = 0 - dest_bb.x1;
+		dest_bb.y1 = 0 - dest_bb.y1;
+
+		for (i = 0, num_of_blits = 0; i < num_of_clip_rects; i++) {
+			int x1 = max_t(int, dest_bb.x1, (int)update_area[i].x1);
+			int y1 = max_t(int, dest_bb.y1, (int)update_area[i].y1);
+			int x2 = min_t(int, dest_bb.x2, (int)update_area[i].x2);
+			int y2 = min_t(int, dest_bb.y2, (int)update_area[i].y2);
+
+			if (x1 >= x2)
+				continue;
+
+			if (y1 >= y2)
+				continue;
+
+			blits[num_of_blits].srcx =  src_bb.x1  + x1;
+			blits[num_of_blits].srcy =  src_bb.y1  + y1;
+			blits[num_of_blits].x    = -dest_bb.x1 + x1;
+			blits[num_of_blits].y    = -dest_bb.y1 + y1;
+			blits[num_of_blits].d    = 1;
+			blits[num_of_blits].w    = x2 - x1;
+			blits[num_of_blits].h    = y2 - y1;
+			num_of_blits++;
+		}
+
+		if (num_of_blits == 0)
+			continue;
+
+		/* Calculate new command size */
+		fifo_size = sizeof(*cmd) + sizeof(SVGA3dCopyBox) * num_of_blits;
+
+		cmd->header.size = cpu_to_le32(fifo_size - sizeof(cmd->header));
+
+		cmd->body.src.sid  = user_handle;
+		cmd->body.dest.sid = stdu[cur_du]->display_srf->res.id;
+
+		ret = vmw_execbuf_process(file_priv, dev_priv, NULL, cmd,
+					  fifo_size, 0, VMW_QUIRK_SCREENTARGET,
+					  NULL, NULL);
+
+		if (unlikely(ret != 0))
+			break;
+
+		for (i = 0; i < num_of_blits; i++) {
+			struct drm_clip_rect blit_area;
+
+			/*
+			 * Add crtc offset because vmw_stdu_update_st expects
+			 * desktop coordinates
+			 */
+			blit_area.x1 = blits[i].x + crtc->x;
+			blit_area.x2 = blit_area.x1 + blits[i].w;
+			blit_area.y1 = blits[i].y + crtc->y;
+			blit_area.y2 = blit_area.y1 + blits[i].h;
+			(void) vmw_stdu_update_st(dev_priv, NULL, stdu[cur_du],
+						  &blit_area);
+		}
+	}
+
+	kfree(cmd);
+
+out_free_update_area:
+	kfree(update_area);
+
+	return ret;
+}
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c b/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
index fb54ccd..835f343 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
@@ -1486,6 +1486,10 @@ int vmw_surface_gb_priv_define(struct drm_device *dev,
 					srf->mip_levels[0],
 					srf->flags & SVGA3D_SURFACE_CUBEMAP);
 
+	if (dev_priv->active_display_unit == vmw_du_screen_target &&
+	    for_scanout)
+		srf->flags |= SVGA3D_SURFACE_SCREENTARGET;
+
 	/*
 	 * From this point, the generic resource management functions
 	 * destroy the object on failure.
diff --git a/include/uapi/drm/vmwgfx_drm.h b/include/uapi/drm/vmwgfx_drm.h
index c472bedb..c8a8631 100644
--- a/include/uapi/drm/vmwgfx_drm.h
+++ b/include/uapi/drm/vmwgfx_drm.h
@@ -88,6 +88,7 @@
 #define DRM_VMW_PARAM_3D_CAPS_SIZE     8
 #define DRM_VMW_PARAM_MAX_MOB_MEMORY   9
 #define DRM_VMW_PARAM_MAX_MOB_SIZE     10
+#define DRM_VMW_PARAM_SCREEN_TARGET    11
 
 /**
  * enum drm_vmw_handle_type - handle type for ref ioctls
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 13/28] drm/vmwgfx: Replace SurfaceDMA usage with SurfaceCopy in 2D VMs

[Thomas Hellstrom]

From: Sinclair Yeh <syeh@vmware.com>

This patch address the following underlying issues with SurfaceDMA

* SurfaceDMA command does not work in a 2D VM, but we can wrap a
  proxy surface around the same DMA buffer and use the SurfaceCopy
  command which does work in a 2D VM.

* Wrapping a DMA buffer with a proxy surface also gives us an
  added optimization path for the case when the DMA buf
  dimensions match the mode.  In this case, the DMA buf can
  be pinned as the display surface, saving an extra copy.
  This only works in a 2D VM because we won't be doing any
  rendering operations directly to the display surface.

v2
* Moved is_dmabuf_proxy field to vmw_framebuffer_surface
* Undone coding style changes
* Addressed other issues from review

Signed-off-by: Sinclair Yeh <syeh@vmware.com>
Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.h      |   3 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c  |  20 ++--
 drivers/gpu/drm/vmwgfx/vmwgfx_kms.c      | 107 ++++++++++++++++++--
 drivers/gpu/drm/vmwgfx/vmwgfx_kms.h      |   1 +
 drivers/gpu/drm/vmwgfx/vmwgfx_mob.c      |   3 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_resource.c |   2 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c     | 165 +++++++++++++++++++++++++++----
 7 files changed, 266 insertions(+), 35 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index 04f8bf2..5d04859 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -342,7 +342,8 @@ enum vmw_display_unit_type {
 };
 
 
-#define VMW_QUIRK_SCREENTARGET (1U << 0)
+#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;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
index 497ad6a..0ec5fd6 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
@@ -674,13 +674,16 @@ static int vmw_cmd_surface_copy_check(struct vmw_private *dev_priv,
 	int ret;
 
 	cmd = container_of(header, struct vmw_sid_cmd, header);
-	ret = vmw_cmd_res_check(dev_priv, sw_context, vmw_res_surface,
-				user_surface_converter,
-				&cmd->body.src.sid, NULL);
-	if (unlikely(ret != 0))
-		return ret;
 
-	if (sw_context->quirks & VMW_QUIRK_SCREENTARGET)
+	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;
 
 	return vmw_cmd_res_check(dev_priv, sw_context, vmw_res_surface,
@@ -1264,7 +1267,7 @@ 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_SCREENTARGET)
+	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,
@@ -1505,6 +1508,9 @@ 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);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
index 6680aa6..615ff6cfc 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
@@ -487,7 +487,8 @@ static int vmw_kms_new_framebuffer_surface(struct vmw_private *dev_priv,
 					   struct vmw_surface *surface,
 					   struct vmw_framebuffer **out,
 					   const struct drm_mode_fb_cmd
-					   *mode_cmd)
+					   *mode_cmd,
+					   bool is_dmabuf_proxy)
 
 {
 	struct drm_device *dev = dev_priv->dev;
@@ -562,6 +563,7 @@ static int vmw_kms_new_framebuffer_surface(struct vmw_private *dev_priv,
 	vfbs->surface = surface;
 	vfbs->base.user_handle = mode_cmd->handle;
 	vfbs->master = drm_master_get(file_priv->master);
+	vfbs->is_dmabuf_proxy = is_dmabuf_proxy;
 
 	mutex_lock(&vmaster->fb_surf_mutex);
 	list_add_tail(&vfbs->head, &vmaster->fb_surf);
@@ -699,6 +701,82 @@ static int vmw_framebuffer_dmabuf_unpin(struct vmw_framebuffer *vfb)
 	return vmw_dmabuf_unpin(dev_priv, vfbd->buffer, false);
 }
 
+/**
+ * vmw_create_dmabuf_proxy - create a proxy surface for the DMA buf
+ *
+ * @dev: DRM device
+ * @mode_cmd: parameters for the new surface
+ * @dmabuf_mob: MOB backing the DMA buf
+ * @srf_out: newly created surface
+ *
+ * When the content FB is a DMA buf, we create a surface as a proxy to the
+ * same buffer.  This way we can do a surface copy rather than a surface DMA.
+ * This is a more efficient approach
+ *
+ * RETURNS:
+ * 0 on success, error code otherwise
+ */
+static int vmw_create_dmabuf_proxy(struct drm_device *dev,
+				   struct drm_mode_fb_cmd *mode_cmd,
+				   struct vmw_dma_buffer *dmabuf_mob,
+				   struct vmw_surface **srf_out)
+{
+	uint32_t format;
+	struct drm_vmw_size content_base_size;
+	int ret;
+
+
+	switch (mode_cmd->depth) {
+	case 32:
+	case 24:
+		format = SVGA3D_X8R8G8B8;
+		break;
+
+	case 16:
+	case 15:
+		format = SVGA3D_R5G6B5;
+		break;
+
+	case 8:
+		format = SVGA3D_P8;
+		break;
+
+	default:
+		DRM_ERROR("Invalid framebuffer format %d\n", mode_cmd->depth);
+		return -EINVAL;
+	}
+
+	content_base_size.width  = mode_cmd->width;
+	content_base_size.height = mode_cmd->height;
+	content_base_size.depth  = 1;
+
+	ret = vmw_surface_gb_priv_define(dev,
+			0, /* kernel visible only */
+			0, /* flags */
+			format,
+			true, /* can be a scanout buffer */
+			1, /* num of mip levels */
+			0,
+			content_base_size,
+			srf_out);
+	if (ret) {
+		DRM_ERROR("Failed to allocate proxy content buffer\n");
+		return ret;
+	}
+
+	/* Use the same MOB backing for surface */
+	vmw_dmabuf_reference(dmabuf_mob);
+
+	(*srf_out)->res.backup = dmabuf_mob;
+
+	/* FIXME:  Waiting for fbdev rework to do a proper reserve/pin */
+	ret = vmw_resource_validate(&(*srf_out)->res);
+
+	return ret;
+}
+
+
+
 static int vmw_kms_new_framebuffer_dmabuf(struct vmw_private *dev_priv,
 					  struct vmw_dma_buffer *dmabuf,
 					  struct vmw_framebuffer **out,
@@ -801,6 +879,7 @@ static struct drm_framebuffer *vmw_kms_fb_create(struct drm_device *dev,
 	struct vmw_dma_buffer *bo = NULL;
 	struct ttm_base_object *user_obj;
 	struct drm_mode_fb_cmd mode_cmd;
+	bool is_dmabuf_proxy = false;
 	int ret;
 
 	mode_cmd.width = mode_cmd2->width;
@@ -849,13 +928,29 @@ static struct drm_framebuffer *vmw_kms_fb_create(struct drm_device *dev,
 	if (ret)
 		goto err_out;
 
-	/* Create the new framebuffer depending one what we got back */
-	if (bo)
+	/*
+	 * We cannot use the SurfaceDMA command in an non-accelerated VM,
+	 * therefore, wrap the DMA buf in a surface so we can use the
+	 * SurfaceCopy command.
+	 */
+	if (bo && !(dev_priv->capabilities & SVGA_CAP_3D) &&
+	    dev_priv->active_display_unit == vmw_du_screen_target) {
+		ret = vmw_create_dmabuf_proxy(dev_priv->dev, &mode_cmd, bo,
+			&surface);
+		if (ret)
+			goto err_out;
+
+		is_dmabuf_proxy = true;
+	}
+
+	/* Create the new framebuffer depending one what we have */
+	if (surface)
+		ret = vmw_kms_new_framebuffer_surface(dev_priv, file_priv,
+						      surface, &vfb, &mode_cmd,
+						      is_dmabuf_proxy);
+	else if (bo)
 		ret = vmw_kms_new_framebuffer_dmabuf(dev_priv, bo, &vfb,
 						     &mode_cmd);
-	else if (surface)
-		ret = vmw_kms_new_framebuffer_surface(dev_priv, file_priv,
-						      surface, &vfb, &mode_cmd);
 	else
 		BUG();
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
index 548fa87..db8ae94 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
@@ -72,6 +72,7 @@ struct vmw_framebuffer_surface {
 	struct vmw_dma_buffer *buffer;
 	struct list_head head;
 	struct drm_master *master;
+	bool is_dmabuf_proxy;  /* true if this is proxy surface for DMA buf */
 };
 
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c b/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c
index 0feac56..e0fc248 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c
@@ -31,8 +31,7 @@
  * If we set up the screen target otable, screen objects stop working.
  */
 
-#define VMW_OTABLE_SETUP_SUB ((VMWGFX_ENABLE_SCREEN_TARGET_OTABLE &&	\
-			       (dev_priv->capabilities & SVGA_CAP_3D)) ? 0 : 1)
+#define VMW_OTABLE_SETUP_SUB ((VMWGFX_ENABLE_SCREEN_TARGET_OTABLE ? 0 : 1))
 
 #ifdef CONFIG_64BIT
 #define VMW_PPN_SIZE 8
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
index 6738c1e..9dcbe8b 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_sys_placement :
+			      &vmw_mob_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 3b8235c..ef99df7 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c
@@ -142,6 +142,63 @@ static void vmw_stdu_crtc_destroy(struct drm_crtc *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
@@ -166,11 +223,13 @@ static int vmw_stdu_content_copy(struct vmw_private *dev_priv,
 				 uint32_t width, uint32_t height,
 				 uint32_t display_x, uint32_t display_y)
 {
-	size_t fifo_size;
+	struct vmw_framebuffer_surface *content_vfbs;
+	size_t fifo_size;	
 	int ret;
 	void *cmd;
+	u32 quirks = VMW_QUIRK_DST_SID_OK;
 
-	struct vmw_surface_dma {
+	struct {
 		SVGA3dCmdHeader     header;
 		SVGA3dCmdSurfaceDMA body;
 		SVGA3dCopyBox       area;
@@ -193,24 +252,43 @@ static int vmw_stdu_content_copy(struct vmw_private *dev_priv,
 		return -EINVAL;
 	}
 
+
 	if (stdu->content_fb_type == SEPARATE_DMA) {
 		struct vmw_framebuffer *content_vfb;
-		struct vmw_framebuffer_dmabuf *content_vfbd;
-		struct vmw_framebuffer_surface *content_vfbs;
 		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);
-		content_vfbd = vmw_framebuffer_to_vfbd(stdu->content_fb);
-		content_vfbs = vmw_framebuffer_to_vfbs(stdu->content_fb);
 
 		cur_size.width  = width;
 		cur_size.height = height;
 		cur_size.depth  = 1;
 
-		desc = svga3dsurface_get_desc(content_vfbs->surface->format);
+		/* 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);
@@ -250,19 +328,40 @@ static int vmw_stdu_content_copy(struct vmw_private *dev_priv,
 
 		cmd = (void *) &surface_dma_cmd;
 	} else {
-		struct vmw_framebuffer *content_vfb;
+		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;
+			}
 
-		content_vfb = vmw_framebuffer_to_vfb(stdu->content_fb);
+			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, 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  = content_vfb->user_handle;
+		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;
@@ -276,8 +375,11 @@ static int vmw_stdu_content_copy(struct vmw_private *dev_priv,
 		cmd = (void *) &surface_cpy_cmd;
 	}
 
-	ret = vmw_execbuf_process(file_priv, dev_priv, NULL, cmd,
-				  fifo_size, 0, VMW_QUIRK_SCREENTARGET,
+
+
+	ret = vmw_execbuf_process(file_priv, dev_priv, NULL,
+				  (void *) cmd,
+				  fifo_size, 0, quirks,
 				  NULL, NULL);
 
 	return ret;
@@ -391,7 +493,8 @@ static int vmw_stdu_bind_st(struct vmw_private *dev_priv,
  * vmw_stdu_update_st - Updates a Screen Target
  *
  * @dev_priv: VMW DRM device
- * @file_priv: Pointer to a drm file private structure
+ * @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
  *
@@ -412,6 +515,7 @@ static int vmw_stdu_update_st(struct vmw_private *dev_priv,
 	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;
@@ -444,8 +548,11 @@ static int vmw_stdu_update_st(struct vmw_private *dev_priv,
 	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) {
-		int ret;
 
 		ret = vmw_stdu_content_copy(dev_priv, file_priv,
 					    stdu,
@@ -459,6 +566,29 @@ static int vmw_stdu_update_st(struct vmw_private *dev_priv,
 		}
 	}
 
+
+	/*
+	 * 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)) {
@@ -1066,8 +1196,7 @@ int vmw_kms_stdu_init_display(struct vmw_private *dev_priv)
 	if (!VMWGFX_ENABLE_SCREEN_TARGET_OTABLE)
 		return -ENOSYS;
 
-	if (!(dev_priv->capabilities & SVGA_CAP_GBOBJECTS) ||
-	    !(dev_priv->capabilities & SVGA_CAP_3D))
+	if (!(dev_priv->capabilities & SVGA_CAP_GBOBJECTS))
 		return -ENOSYS;
 
 	ret = drm_vblank_init(dev, VMWGFX_NUM_DISPLAY_UNITS);
@@ -1333,7 +1462,7 @@ int vmw_kms_stdu_present(struct vmw_private *dev_priv,
 		cmd->body.dest.sid = stdu[cur_du]->display_srf->res.id;
 
 		ret = vmw_execbuf_process(file_priv, dev_priv, NULL, cmd,
-					  fifo_size, 0, VMW_QUIRK_SCREENTARGET,
+					  fifo_size, 0, VMW_QUIRK_DST_SID_OK,
 					  NULL, NULL);
 
 		if (unlikely(ret != 0))
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 14/28] drm/vmwgfx: Introduce a pin count to allow for recursive pinning v2

[Thomas Hellstrom]

v2: Fix dma buffer validation on resource pinning.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_dmabuf.c   | 182 ++++++++++++++-----------------
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.c      |  31 +++---
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.h      |  33 +++---
 drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c  |  81 +++++++-------
 drivers/gpu/drm/vmwgfx/vmwgfx_fb.c       |   2 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c     |   4 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_kms.c      |   2 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_overlay.c  |   4 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_resource.c |  36 +++---
 9 files changed, 179 insertions(+), 196 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_dmabuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_dmabuf.c
index 914b375..4b9344d 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_dmabuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_dmabuf.c
@@ -32,25 +32,20 @@
 
 
 /**
- * vmw_dmabuf_to_placement - Validate a buffer to placement.
+ * vmw_dmabuf_pin_in_placement - Validate a buffer to placement.
  *
  * @dev_priv:  Driver private.
  * @buf:  DMA buffer to move.
- * @pin:  Pin buffer if true.
+ * @placement:  The placement to pin it.
  * @interruptible:  Use interruptible wait.
  *
- * May only be called by the current master since it assumes that the
- * master lock is the current master's lock.
- * This function takes the master's lock in write mode.
- * Flushes and unpins the query bo to avoid failures.
- *
  * Returns
  *  -ERESTARTSYS if interrupted by a signal.
  */
-int vmw_dmabuf_to_placement(struct vmw_private *dev_priv,
-			    struct vmw_dma_buffer *buf,
-			    struct ttm_placement *placement,
-			    bool interruptible)
+int vmw_dmabuf_pin_in_placement(struct vmw_private *dev_priv,
+				struct vmw_dma_buffer *buf,
+				struct ttm_placement *placement,
+				bool interruptible)
 {
 	struct ttm_buffer_object *bo = &buf->base;
 	int ret;
@@ -66,6 +61,8 @@ int vmw_dmabuf_to_placement(struct vmw_private *dev_priv,
 		goto err;
 
 	ret = ttm_bo_validate(bo, placement, interruptible, false);
+	if (!ret)
+		vmw_bo_pin_reserved(buf, true);
 
 	ttm_bo_unreserve(bo);
 
@@ -75,12 +72,10 @@ err:
 }
 
 /**
- * vmw_dmabuf_to_vram_or_gmr - Move a buffer to vram or gmr.
+ * vmw_dmabuf_pin_in_vram_or_gmr - Move a buffer to vram or gmr.
  *
- * May only be called by the current master since it assumes that the
- * master lock is the current master's lock.
- * This function takes the master's lock in write mode.
- * Flushes and unpins the query bo if @pin == true to avoid failures.
+ * This function takes the reservation_sem in write mode.
+ * Flushes and unpins the query bo to avoid failures.
  *
  * @dev_priv:  Driver private.
  * @buf:  DMA buffer to move.
@@ -90,55 +85,34 @@ err:
  * Returns
  * -ERESTARTSYS if interrupted by a signal.
  */
-int vmw_dmabuf_to_vram_or_gmr(struct vmw_private *dev_priv,
-			      struct vmw_dma_buffer *buf,
-			      bool pin, bool interruptible)
+int vmw_dmabuf_pin_in_vram_or_gmr(struct vmw_private *dev_priv,
+				  struct vmw_dma_buffer *buf,
+				  bool interruptible)
 {
 	struct ttm_buffer_object *bo = &buf->base;
-	struct ttm_placement *placement;
 	int ret;
 
 	ret = ttm_write_lock(&dev_priv->reservation_sem, interruptible);
 	if (unlikely(ret != 0))
 		return ret;
 
-	if (pin)
-		vmw_execbuf_release_pinned_bo(dev_priv);
+	vmw_execbuf_release_pinned_bo(dev_priv);
 
 	ret = ttm_bo_reserve(bo, interruptible, false, false, NULL);
 	if (unlikely(ret != 0))
 		goto err;
 
-	/**
-	 * Put BO in VRAM if there is space, otherwise as a GMR.
-	 * If there is no space in VRAM and GMR ids are all used up,
-	 * start evicting GMRs to make room. If the DMA buffer can't be
-	 * used as a GMR, this will return -ENOMEM.
-	 */
-
-	if (pin)
-		placement = &vmw_vram_gmr_ne_placement;
-	else
-		placement = &vmw_vram_gmr_placement;
-
-	ret = ttm_bo_validate(bo, placement, interruptible, false);
+	ret = ttm_bo_validate(bo, &vmw_vram_gmr_placement, interruptible,
+			      false);
 	if (likely(ret == 0) || ret == -ERESTARTSYS)
-		goto err_unreserve;
-
+		goto out_unreserve;
 
-	/**
-	 * If that failed, try VRAM again, this time evicting
-	 * previous contents.
-	 */
-
-	if (pin)
-		placement = &vmw_vram_ne_placement;
-	else
-		placement = &vmw_vram_placement;
+	ret = ttm_bo_validate(bo, &vmw_vram_placement, interruptible, false);
 
-	ret = ttm_bo_validate(bo, placement, interruptible, false);
+out_unreserve:
+	if (!ret)
+		vmw_bo_pin_reserved(buf, true);
 
-err_unreserve:
 	ttm_bo_unreserve(bo);
 err:
 	ttm_write_unlock(&dev_priv->reservation_sem);
@@ -146,67 +120,50 @@ err:
 }
 
 /**
- * vmw_dmabuf_to_vram - Move a buffer to vram.
+ * vmw_dmabuf_pin_in_vram - Move a buffer to vram.
  *
- * May only be called by the current master since it assumes that the
- * master lock is the current master's lock.
- * This function takes the master's lock in write mode.
+ * This function takes the reservation_sem in write mode.
+ * Flushes and unpins the query bo to avoid failures.
  *
  * @dev_priv:  Driver private.
  * @buf:  DMA buffer to move.
- * @pin:  Pin buffer in vram if true.
  * @interruptible:  Use interruptible wait.
  *
  * Returns
  * -ERESTARTSYS if interrupted by a signal.
  */
-int vmw_dmabuf_to_vram(struct vmw_private *dev_priv,
-		       struct vmw_dma_buffer *buf,
-		       bool pin, bool interruptible)
+int vmw_dmabuf_pin_in_vram(struct vmw_private *dev_priv,
+			   struct vmw_dma_buffer *buf,
+			   bool interruptible)
 {
-	struct ttm_placement *placement;
-
-	if (pin)
-		placement = &vmw_vram_ne_placement;
-	else
-		placement = &vmw_vram_placement;
-
-	return vmw_dmabuf_to_placement(dev_priv, buf,
-				       placement,
-				       interruptible);
+	return vmw_dmabuf_pin_in_placement(dev_priv, buf, &vmw_vram_placement,
+					   interruptible);
 }
 
 /**
- * vmw_dmabuf_to_start_of_vram - Move a buffer to start of vram.
+ * vmw_dmabuf_pin_in_start_of_vram - Move a buffer to start of vram.
  *
- * May only be called by the current master since it assumes that the
- * master lock is the current master's lock.
- * This function takes the master's lock in write mode.
- * Flushes and unpins the query bo if @pin == true to avoid failures.
+ * This function takes the reservation_sem in write mode.
+ * Flushes and unpins the query bo to avoid failures.
  *
  * @dev_priv:  Driver private.
- * @buf:  DMA buffer to move.
- * @pin:  Pin buffer in vram if true.
+ * @buf:  DMA buffer to pin.
  * @interruptible:  Use interruptible wait.
  *
  * Returns
  * -ERESTARTSYS if interrupted by a signal.
  */
-int vmw_dmabuf_to_start_of_vram(struct vmw_private *dev_priv,
-				struct vmw_dma_buffer *buf,
-				bool pin, bool interruptible)
+int vmw_dmabuf_pin_in_start_of_vram(struct vmw_private *dev_priv,
+				    struct vmw_dma_buffer *buf,
+				    bool interruptible)
 {
 	struct ttm_buffer_object *bo = &buf->base;
 	struct ttm_placement placement;
 	struct ttm_place place;
 	int ret = 0;
 
-	if (pin)
-		place = vmw_vram_ne_placement.placement[0];
-	else
-		place = vmw_vram_placement.placement[0];
+	place = vmw_vram_placement.placement[0];
 	place.lpfn = bo->num_pages;
-
 	placement.num_placement = 1;
 	placement.placement = &place;
 	placement.num_busy_placement = 1;
@@ -216,13 +173,16 @@ int vmw_dmabuf_to_start_of_vram(struct vmw_private *dev_priv,
 	if (unlikely(ret != 0))
 		return ret;
 
-	if (pin)
-		vmw_execbuf_release_pinned_bo(dev_priv);
+	vmw_execbuf_release_pinned_bo(dev_priv);
 	ret = ttm_bo_reserve(bo, interruptible, false, false, NULL);
 	if (unlikely(ret != 0))
 		goto err_unlock;
 
-	/* Is this buffer already in vram but not at the start of it? */
+	/*
+	 * Is this buffer already in vram but not at the start of it?
+	 * In that case, evict it first because TTM isn't good at handling
+	 * that situation.
+	 */
 	if (bo->mem.mem_type == TTM_PL_VRAM &&
 	    bo->mem.start < bo->num_pages &&
 	    bo->mem.start > 0)
@@ -230,8 +190,10 @@ int vmw_dmabuf_to_start_of_vram(struct vmw_private *dev_priv,
 
 	ret = ttm_bo_validate(bo, &placement, interruptible, false);
 
-	/* For some reason we didn't up at the start of vram */
+	/* For some reason we didn't end up at the start of vram */
 	WARN_ON(ret == 0 && bo->offset != 0);
+	if (!ret)
+		vmw_bo_pin_reserved(buf, true);
 
 	ttm_bo_unreserve(bo);
 err_unlock:
@@ -240,13 +202,10 @@ err_unlock:
 	return ret;
 }
 
-
 /**
- * vmw_dmabuf_upin - Unpin the buffer given buffer, does not move the buffer.
+ * vmw_dmabuf_unpin - Unpin the buffer given buffer, does not move the buffer.
  *
- * May only be called by the current master since it assumes that the
- * master lock is the current master's lock.
- * This function takes the master's lock in write mode.
+ * This function takes the reservation_sem in write mode.
  *
  * @dev_priv:  Driver private.
  * @buf:  DMA buffer to unpin.
@@ -259,16 +218,25 @@ int vmw_dmabuf_unpin(struct vmw_private *dev_priv,
 		     struct vmw_dma_buffer *buf,
 		     bool interruptible)
 {
-	/*
-	 * We could in theory early out if the buffer is
-	 * unpinned but we need to lock and reserve the buffer
-	 * anyways so we don't gain much by that.
-	 */
-	return vmw_dmabuf_to_placement(dev_priv, buf,
-				       &vmw_evictable_placement,
-				       interruptible);
-}
+	struct ttm_buffer_object *bo = &buf->base;
+	int ret;
 
+	ret = ttm_read_lock(&dev_priv->reservation_sem, interruptible);
+	if (unlikely(ret != 0))
+		return ret;
+
+	ret = ttm_bo_reserve(bo, interruptible, false, false, 0);
+	if (unlikely(ret != 0))
+		goto err;
+
+	vmw_bo_pin_reserved(buf, false);
+
+	ttm_bo_unreserve(bo);
+
+err:
+	ttm_read_unlock(&dev_priv->reservation_sem);
+	return ret;
+}
 
 /**
  * vmw_bo_get_guest_ptr - Get the guest ptr representing the current placement
@@ -291,21 +259,31 @@ void vmw_bo_get_guest_ptr(const struct ttm_buffer_object *bo,
 
 
 /**
- * vmw_bo_pin - Pin or unpin a buffer object without moving it.
+ * vmw_bo_pin_reserved - Pin or unpin a buffer object without moving it.
  *
- * @bo: The buffer object. Must be reserved.
+ * @vbo: The buffer object. Must be reserved.
  * @pin: Whether to pin or unpin.
  *
  */
-void vmw_bo_pin(struct ttm_buffer_object *bo, bool pin)
+void vmw_bo_pin_reserved(struct vmw_dma_buffer *vbo, bool pin)
 {
 	struct ttm_place pl;
 	struct ttm_placement placement;
+	struct ttm_buffer_object *bo = &vbo->base;
 	uint32_t old_mem_type = bo->mem.mem_type;
 	int ret;
 
 	lockdep_assert_held(&bo->resv->lock.base);
 
+	if (pin) {
+		if (vbo->pin_count++ > 0)
+			return;
+	} else {
+		WARN_ON(vbo->pin_count <= 0);
+		if (--vbo->pin_count > 0)
+			return;
+	}
+
 	pl.fpfn = 0;
 	pl.lpfn = 0;
 	pl.flags = TTM_PL_FLAG_VRAM | VMW_PL_FLAG_GMR | VMW_PL_FLAG_MOB
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
index ab1b70c..e55db3f 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
@@ -298,30 +298,31 @@ static void vmw_print_capabilities(uint32_t capabilities)
 static int vmw_dummy_query_bo_create(struct vmw_private *dev_priv)
 {
 	int ret;
-	struct ttm_buffer_object *bo;
+	struct vmw_dma_buffer *vbo;
 	struct ttm_bo_kmap_obj map;
 	volatile SVGA3dQueryResult *result;
 	bool dummy;
 
 	/*
-	 * Create the bo as pinned, so that a tryreserve will
+	 * Create the vbo as pinned, so that a tryreserve will
 	 * immediately succeed. This is because we're the only
 	 * user of the bo currently.
 	 */
-	ret = ttm_bo_create(&dev_priv->bdev,
-			    PAGE_SIZE,
-			    ttm_bo_type_device,
-			    &vmw_sys_ne_placement,
-			    0, false, NULL,
-			    &bo);
+	vbo = kzalloc(sizeof(*vbo), GFP_KERNEL);
+	if (!vbo)
+		return -ENOMEM;
 
+	ret = vmw_dmabuf_init(dev_priv, vbo, PAGE_SIZE,
+			      &vmw_sys_ne_placement, false,
+			      &vmw_dmabuf_bo_free);
 	if (unlikely(ret != 0))
 		return ret;
 
-	ret = ttm_bo_reserve(bo, false, true, false, NULL);
+	ret = ttm_bo_reserve(&vbo->base, false, true, false, NULL);
 	BUG_ON(ret != 0);
+	vmw_bo_pin_reserved(vbo, true);
 
-	ret = ttm_bo_kmap(bo, 0, 1, &map);
+	ret = ttm_bo_kmap(&vbo->base, 0, 1, &map);
 	if (likely(ret == 0)) {
 		result = ttm_kmap_obj_virtual(&map, &dummy);
 		result->totalSize = sizeof(*result);
@@ -329,14 +330,14 @@ static int vmw_dummy_query_bo_create(struct vmw_private *dev_priv)
 		result->result32 = 0xff;
 		ttm_bo_kunmap(&map);
 	}
-	vmw_bo_pin(bo, false);
-	ttm_bo_unreserve(bo);
+	vmw_bo_pin_reserved(vbo, false);
+	ttm_bo_unreserve(&vbo->base);
 
 	if (unlikely(ret != 0)) {
 		DRM_ERROR("Dummy query buffer map failed.\n");
-		ttm_bo_unref(&bo);
+		vmw_dmabuf_unreference(&vbo);
 	} else
-		dev_priv->dummy_query_bo = bo;
+		dev_priv->dummy_query_bo = vbo;
 
 	return ret;
 }
@@ -434,7 +435,7 @@ static void vmw_release_device_early(struct vmw_private *dev_priv)
 
 	BUG_ON(dev_priv->pinned_bo != NULL);
 
-	ttm_bo_unref(&dev_priv->dummy_query_bo);
+	vmw_dmabuf_unreference(&dev_priv->dummy_query_bo);
 	if (dev_priv->cman)
 		vmw_cmdbuf_remove_pool(dev_priv->cman);
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index 5d04859..12eaa6c 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -85,6 +85,7 @@ struct vmw_fpriv {
 struct vmw_dma_buffer {
 	struct ttm_buffer_object base;
 	struct list_head res_list;
+	s32 pin_count;
 };
 
 /**
@@ -358,7 +359,7 @@ struct vmw_sw_context{
 	uint32_t *cmd_bounce;
 	uint32_t cmd_bounce_size;
 	struct list_head resource_list;
-	struct ttm_buffer_object *cur_query_bo;
+	struct vmw_dma_buffer *cur_query_bo;
 	struct list_head res_relocations;
 	uint32_t *buf_start;
 	struct vmw_res_cache_entry res_cache[vmw_res_max];
@@ -533,8 +534,8 @@ struct vmw_private {
 	 * are protected by the cmdbuf mutex.
 	 */
 
-	struct ttm_buffer_object *dummy_query_bo;
-	struct ttm_buffer_object *pinned_bo;
+	struct vmw_dma_buffer *dummy_query_bo;
+	struct vmw_dma_buffer *pinned_bo;
 	uint32_t query_cid;
 	uint32_t query_cid_valid;
 	bool dummy_query_bo_pinned;
@@ -700,25 +701,25 @@ extern void vmw_resource_evict_all(struct vmw_private *dev_priv);
 /**
  * DMA buffer helper routines - vmwgfx_dmabuf.c
  */
-extern int vmw_dmabuf_to_placement(struct vmw_private *vmw_priv,
-				   struct vmw_dma_buffer *bo,
-				   struct ttm_placement *placement,
-				   bool interruptible);
-extern int vmw_dmabuf_to_vram(struct vmw_private *dev_priv,
-			      struct vmw_dma_buffer *buf,
-			      bool pin, bool interruptible);
-extern int vmw_dmabuf_to_vram_or_gmr(struct vmw_private *dev_priv,
-				     struct vmw_dma_buffer *buf,
-				     bool pin, bool interruptible);
-extern int vmw_dmabuf_to_start_of_vram(struct vmw_private *vmw_priv,
+extern int vmw_dmabuf_pin_in_placement(struct vmw_private *vmw_priv,
 				       struct vmw_dma_buffer *bo,
-				       bool pin, bool interruptible);
+				       struct ttm_placement *placement,
+				       bool interruptible);
+extern int vmw_dmabuf_pin_in_vram(struct vmw_private *dev_priv,
+				  struct vmw_dma_buffer *buf,
+				  bool interruptible);
+extern int vmw_dmabuf_pin_in_vram_or_gmr(struct vmw_private *dev_priv,
+					 struct vmw_dma_buffer *buf,
+					 bool interruptible);
+extern int vmw_dmabuf_pin_in_start_of_vram(struct vmw_private *vmw_priv,
+					   struct vmw_dma_buffer *bo,
+					   bool interruptible);
 extern int vmw_dmabuf_unpin(struct vmw_private *vmw_priv,
 			    struct vmw_dma_buffer *bo,
 			    bool interruptible);
 extern void vmw_bo_get_guest_ptr(const struct ttm_buffer_object *buf,
 				 SVGAGuestPtr *ptr);
-extern void vmw_bo_pin(struct ttm_buffer_object *bo, bool pin);
+extern void vmw_bo_pin_reserved(struct vmw_dma_buffer *bo, bool pin);
 
 /**
  * Misc Ioctl functionality - vmwgfx_ioctl.c
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
index 0ec5fd6..92e8998 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
@@ -308,7 +308,7 @@ static int vmw_cmd_ok(struct vmw_private *dev_priv,
  * submission is reached.
  */
 static int vmw_bo_to_validate_list(struct vmw_sw_context *sw_context,
-				   struct ttm_buffer_object *bo,
+				   struct vmw_dma_buffer *vbo,
 				   bool validate_as_mob,
 				   uint32_t *p_val_node)
 {
@@ -318,7 +318,7 @@ static int vmw_bo_to_validate_list(struct vmw_sw_context *sw_context,
 	struct drm_hash_item *hash;
 	int ret;
 
-	if (likely(drm_ht_find_item(&sw_context->res_ht, (unsigned long) bo,
+	if (likely(drm_ht_find_item(&sw_context->res_ht, (unsigned long) vbo,
 				    &hash) == 0)) {
 		vval_buf = container_of(hash, struct vmw_validate_buffer,
 					hash);
@@ -336,7 +336,7 @@ static int vmw_bo_to_validate_list(struct vmw_sw_context *sw_context,
 			return -EINVAL;
 		}
 		vval_buf = &sw_context->val_bufs[val_node];
-		vval_buf->hash.key = (unsigned long) bo;
+		vval_buf->hash.key = (unsigned long) vbo;
 		ret = drm_ht_insert_item(&sw_context->res_ht, &vval_buf->hash);
 		if (unlikely(ret != 0)) {
 			DRM_ERROR("Failed to initialize a buffer validation "
@@ -345,7 +345,7 @@ static int vmw_bo_to_validate_list(struct vmw_sw_context *sw_context,
 		}
 		++sw_context->cur_val_buf;
 		val_buf = &vval_buf->base;
-		val_buf->bo = ttm_bo_reference(bo);
+		val_buf->bo = ttm_bo_reference(&vbo->base);
 		val_buf->shared = false;
 		list_add_tail(&val_buf->head, &sw_context->validate_nodes);
 		vval_buf->validate_as_mob = validate_as_mob;
@@ -380,10 +380,10 @@ static int vmw_resources_reserve(struct vmw_sw_context *sw_context)
 			return ret;
 
 		if (res->backup) {
-			struct ttm_buffer_object *bo = &res->backup->base;
+			struct vmw_dma_buffer *vbo = res->backup;
 
 			ret = vmw_bo_to_validate_list
-				(sw_context, bo,
+				(sw_context, vbo,
 				 vmw_resource_needs_backup(res), NULL);
 
 			if (unlikely(ret != 0))
@@ -759,7 +759,7 @@ static int vmw_cmd_present_check(struct vmw_private *dev_priv,
  * command batch.
  */
 static int vmw_query_bo_switch_prepare(struct vmw_private *dev_priv,
-				       struct ttm_buffer_object *new_query_bo,
+				       struct vmw_dma_buffer *new_query_bo,
 				       struct vmw_sw_context *sw_context)
 {
 	struct vmw_res_cache_entry *ctx_entry =
@@ -771,7 +771,7 @@ static int vmw_query_bo_switch_prepare(struct vmw_private *dev_priv,
 
 	if (unlikely(new_query_bo != sw_context->cur_query_bo)) {
 
-		if (unlikely(new_query_bo->num_pages > 4)) {
+		if (unlikely(new_query_bo->base.num_pages > 4)) {
 			DRM_ERROR("Query buffer too large.\n");
 			return -EINVAL;
 		}
@@ -840,12 +840,12 @@ static void vmw_query_bo_switch_commit(struct vmw_private *dev_priv,
 
 	if (dev_priv->pinned_bo != sw_context->cur_query_bo) {
 		if (dev_priv->pinned_bo) {
-			vmw_bo_pin(dev_priv->pinned_bo, false);
-			ttm_bo_unref(&dev_priv->pinned_bo);
+			vmw_bo_pin_reserved(dev_priv->pinned_bo, false);
+			vmw_dmabuf_unreference(&dev_priv->pinned_bo);
 		}
 
 		if (!sw_context->needs_post_query_barrier) {
-			vmw_bo_pin(sw_context->cur_query_bo, true);
+			vmw_bo_pin_reserved(sw_context->cur_query_bo, true);
 
 			/*
 			 * We pin also the dummy_query_bo buffer so that we
@@ -853,14 +853,17 @@ static void vmw_query_bo_switch_commit(struct vmw_private *dev_priv,
 			 * dummy queries in context destroy paths.
 			 */
 
-			vmw_bo_pin(dev_priv->dummy_query_bo, true);
-			dev_priv->dummy_query_bo_pinned = true;
+			if (!dev_priv->dummy_query_bo_pinned) {
+				vmw_bo_pin_reserved(dev_priv->dummy_query_bo,
+						    true);
+				dev_priv->dummy_query_bo_pinned = true;
+			}
 
 			BUG_ON(sw_context->last_query_ctx == NULL);
 			dev_priv->query_cid = sw_context->last_query_ctx->id;
 			dev_priv->query_cid_valid = true;
 			dev_priv->pinned_bo =
-				ttm_bo_reference(sw_context->cur_query_bo);
+				vmw_dmabuf_reference(sw_context->cur_query_bo);
 		}
 	}
 }
@@ -889,7 +892,6 @@ static int vmw_translate_mob_ptr(struct vmw_private *dev_priv,
 				 struct vmw_dma_buffer **vmw_bo_p)
 {
 	struct vmw_dma_buffer *vmw_bo = NULL;
-	struct ttm_buffer_object *bo;
 	uint32_t handle = *id;
 	struct vmw_relocation *reloc;
 	int ret;
@@ -900,7 +902,6 @@ static int vmw_translate_mob_ptr(struct vmw_private *dev_priv,
 		ret = -EINVAL;
 		goto out_no_reloc;
 	}
-	bo = &vmw_bo->base;
 
 	if (unlikely(sw_context->cur_reloc >= VMWGFX_MAX_RELOCATIONS)) {
 		DRM_ERROR("Max number relocations per submission"
@@ -913,7 +914,7 @@ static int vmw_translate_mob_ptr(struct vmw_private *dev_priv,
 	reloc->mob_loc = id;
 	reloc->location = NULL;
 
-	ret = vmw_bo_to_validate_list(sw_context, bo, true, &reloc->index);
+	ret = vmw_bo_to_validate_list(sw_context, vmw_bo, true, &reloc->index);
 	if (unlikely(ret != 0))
 		goto out_no_reloc;
 
@@ -951,7 +952,6 @@ static int vmw_translate_guest_ptr(struct vmw_private *dev_priv,
 				   struct vmw_dma_buffer **vmw_bo_p)
 {
 	struct vmw_dma_buffer *vmw_bo = NULL;
-	struct ttm_buffer_object *bo;
 	uint32_t handle = ptr->gmrId;
 	struct vmw_relocation *reloc;
 	int ret;
@@ -962,7 +962,6 @@ static int vmw_translate_guest_ptr(struct vmw_private *dev_priv,
 		ret = -EINVAL;
 		goto out_no_reloc;
 	}
-	bo = &vmw_bo->base;
 
 	if (unlikely(sw_context->cur_reloc >= VMWGFX_MAX_RELOCATIONS)) {
 		DRM_ERROR("Max number relocations per submission"
@@ -974,7 +973,7 @@ static int vmw_translate_guest_ptr(struct vmw_private *dev_priv,
 	reloc = &sw_context->relocs[sw_context->cur_reloc++];
 	reloc->location = ptr;
 
-	ret = vmw_bo_to_validate_list(sw_context, bo, false, &reloc->index);
+	ret = vmw_bo_to_validate_list(sw_context, vmw_bo, false, &reloc->index);
 	if (unlikely(ret != 0))
 		goto out_no_reloc;
 
@@ -1081,7 +1080,7 @@ static int vmw_cmd_end_gb_query(struct vmw_private *dev_priv,
 	if (unlikely(ret != 0))
 		return ret;
 
-	ret = vmw_query_bo_switch_prepare(dev_priv, &vmw_bo->base, sw_context);
+	ret = vmw_query_bo_switch_prepare(dev_priv, vmw_bo, sw_context);
 
 	vmw_dmabuf_unreference(&vmw_bo);
 	return ret;
@@ -1135,7 +1134,7 @@ static int vmw_cmd_end_query(struct vmw_private *dev_priv,
 	if (unlikely(ret != 0))
 		return ret;
 
-	ret = vmw_query_bo_switch_prepare(dev_priv, &vmw_bo->base, sw_context);
+	ret = vmw_query_bo_switch_prepare(dev_priv, vmw_bo, sw_context);
 
 	vmw_dmabuf_unreference(&vmw_bo);
 	return ret;
@@ -2239,16 +2238,11 @@ static int vmw_validate_single_buffer(struct vmw_private *dev_priv,
 				      struct ttm_buffer_object *bo,
 				      bool validate_as_mob)
 {
+	struct vmw_dma_buffer *vbo = container_of(bo, struct vmw_dma_buffer,
+						  base);
 	int ret;
 
-
-	/*
-	 * Don't validate pinned buffers.
-	 */
-
-	if (bo == dev_priv->pinned_bo ||
-	    (bo == dev_priv->dummy_query_bo &&
-	     dev_priv->dummy_query_bo_pinned))
+	if (vbo->pin_count > 0)
 		return 0;
 
 	if (validate_as_mob)
@@ -2767,9 +2761,11 @@ static void vmw_execbuf_unpin_panic(struct vmw_private *dev_priv)
 	DRM_ERROR("Can't unpin query buffer. Trying to recover.\n");
 
 	(void) vmw_fallback_wait(dev_priv, false, true, 0, false, 10*HZ);
-	vmw_bo_pin(dev_priv->pinned_bo, false);
-	vmw_bo_pin(dev_priv->dummy_query_bo, false);
-	dev_priv->dummy_query_bo_pinned = false;
+	vmw_bo_pin_reserved(dev_priv->pinned_bo, false);
+	if (dev_priv->dummy_query_bo_pinned) {
+		vmw_bo_pin_reserved(dev_priv->dummy_query_bo, false);
+		dev_priv->dummy_query_bo_pinned = false;
+	}
 }
 
 
@@ -2811,11 +2807,11 @@ void __vmw_execbuf_release_pinned_bo(struct vmw_private *dev_priv,
 
 	INIT_LIST_HEAD(&validate_list);
 
-	pinned_val.bo = ttm_bo_reference(dev_priv->pinned_bo);
+	pinned_val.bo = ttm_bo_reference(&dev_priv->pinned_bo->base);
 	pinned_val.shared = false;
 	list_add_tail(&pinned_val.head, &validate_list);
 
-	query_val.bo = ttm_bo_reference(dev_priv->dummy_query_bo);
+	query_val.bo = ttm_bo_reference(&dev_priv->dummy_query_bo->base);
 	query_val.shared = false;
 	list_add_tail(&query_val.head, &validate_list);
 
@@ -2836,10 +2832,11 @@ void __vmw_execbuf_release_pinned_bo(struct vmw_private *dev_priv,
 		dev_priv->query_cid_valid = false;
 	}
 
-	vmw_bo_pin(dev_priv->pinned_bo, false);
-	vmw_bo_pin(dev_priv->dummy_query_bo, false);
-	dev_priv->dummy_query_bo_pinned = false;
-
+	vmw_bo_pin_reserved(dev_priv->pinned_bo, false);
+	if (dev_priv->dummy_query_bo_pinned) {
+		vmw_bo_pin_reserved(dev_priv->dummy_query_bo, false);
+		dev_priv->dummy_query_bo_pinned = false;
+	}
 	if (fence == NULL) {
 		(void) vmw_execbuf_fence_commands(NULL, dev_priv, &lfence,
 						  NULL);
@@ -2851,7 +2848,9 @@ void __vmw_execbuf_release_pinned_bo(struct vmw_private *dev_priv,
 
 	ttm_bo_unref(&query_val.bo);
 	ttm_bo_unref(&pinned_val.bo);
-	ttm_bo_unref(&dev_priv->pinned_bo);
+	vmw_dmabuf_unreference(&dev_priv->pinned_bo);
+	DRM_INFO("Dummy query bo pin count: %d\n",
+		 dev_priv->dummy_query_bo->pin_count);
 
 out_unlock:
 	return;
@@ -2861,7 +2860,7 @@ out_no_emit:
 out_no_reserve:
 	ttm_bo_unref(&query_val.bo);
 	ttm_bo_unref(&pinned_val.bo);
-	ttm_bo_unref(&dev_priv->pinned_bo);
+	vmw_dmabuf_unreference(&dev_priv->pinned_bo);
 }
 
 /**
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
index d0a3bcf..b54d99bc 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
@@ -636,7 +636,7 @@ int vmw_fb_on(struct vmw_private *vmw_priv)
 	/* Make sure that all overlays are stoped when we take over */
 	vmw_overlay_stop_all(vmw_priv);
 
-	ret = vmw_dmabuf_to_start_of_vram(vmw_priv, par->vmw_bo, true, false);
+	ret = vmw_dmabuf_pin_in_start_of_vram(vmw_priv, par->vmw_bo, false);
 	if (unlikely(ret != 0)) {
 		DRM_ERROR("could not move buffer to start of VRAM\n");
 		goto err_no_buffer;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
index 239815c..9b8b09f 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
@@ -604,7 +604,7 @@ static int vmw_fifo_emit_dummy_legacy_query(struct vmw_private *dev_priv,
 	 * without writing to the query result structure.
 	 */
 
-	struct ttm_buffer_object *bo = dev_priv->dummy_query_bo;
+	struct ttm_buffer_object *bo = &dev_priv->dummy_query_bo->base;
 	struct {
 		SVGA3dCmdHeader header;
 		SVGA3dCmdWaitForQuery body;
@@ -653,7 +653,7 @@ static int vmw_fifo_emit_dummy_gb_query(struct vmw_private *dev_priv,
 	 * without writing to the query result structure.
 	 */
 
-	struct ttm_buffer_object *bo = dev_priv->dummy_query_bo;
+	struct ttm_buffer_object *bo = &dev_priv->dummy_query_bo->base;
 	struct {
 		SVGA3dCmdHeader header;
 		SVGA3dCmdWaitForGBQuery body;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
index 615ff6cfc..99e2f5b 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
@@ -678,7 +678,7 @@ static int vmw_framebuffer_dmabuf_pin(struct vmw_framebuffer *vfb)
 
 	vmw_overlay_pause_all(dev_priv);
 
-	ret = vmw_dmabuf_to_start_of_vram(dev_priv, vfbd->buffer, true, false);
+	ret = vmw_dmabuf_pin_in_start_of_vram(dev_priv, vfbd->buffer, false);
 
 	vmw_overlay_resume_all(dev_priv);
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_overlay.c b/drivers/gpu/drm/vmwgfx/vmwgfx_overlay.c
index 7f4b2f0..d839051 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_overlay.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_overlay.c
@@ -232,9 +232,9 @@ static int vmw_overlay_move_buffer(struct vmw_private *dev_priv,
 		return vmw_dmabuf_unpin(dev_priv, buf, inter);
 
 	if (dev_priv->active_display_unit == vmw_du_legacy)
-		return vmw_dmabuf_to_vram(dev_priv, buf, true, inter);
+		return vmw_dmabuf_pin_in_vram(dev_priv, buf, inter);
 
-	return vmw_dmabuf_to_vram_or_gmr(dev_priv, buf, true, inter);
+	return vmw_dmabuf_pin_in_vram_or_gmr(dev_priv, buf, inter);
 }
 
 /**
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
index 9dcbe8b..271bc90 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
@@ -1596,25 +1596,29 @@ int vmw_resource_pin(struct vmw_resource *res)
 		goto out_no_reserve;
 
 	if (res->pin_count == 0) {
-		struct ttm_buffer_object *bo = NULL;
+		struct vmw_dma_buffer *vbo = NULL;
 
 		if (res->backup) {
-			bo = &res->backup->base;
-
-			ttm_bo_reserve(bo, false, false, false, NULL);
-			ret = ttm_bo_validate(bo, res->func->backup_placement,
-					      false, false);
-			if (ret) {
-				ttm_bo_unreserve(bo);
-				goto out_no_validate;
+			vbo = res->backup;
+
+			ttm_bo_reserve(&vbo->base, false, false, false, NULL);
+			if (!vbo->pin_count) {
+				ret = ttm_bo_validate
+					(&vbo->base,
+					 res->func->backup_placement,
+					 false, false);
+				if (ret) {
+					ttm_bo_unreserve(&vbo->base);
+					goto out_no_validate;
+				}
 			}
 
 			/* Do we really need to pin the MOB as well? */
-			vmw_bo_pin(bo, true);
+			vmw_bo_pin_reserved(vbo, true);
 		}
 		ret = vmw_resource_validate(res);
-		if (bo)
-			ttm_bo_unreserve(bo);
+		if (vbo)
+			ttm_bo_unreserve(&vbo->base);
 		if (ret)
 			goto out_no_validate;
 	}
@@ -1650,11 +1654,11 @@ void vmw_resource_unpin(struct vmw_resource *res)
 
 	WARN_ON(res->pin_count == 0);
 	if (--res->pin_count == 0 && res->backup) {
-		struct ttm_buffer_object *bo = &res->backup->base;
+		struct vmw_dma_buffer *vbo = res->backup;
 
-		ttm_bo_reserve(bo, false, false, false, NULL);
-		vmw_bo_pin(bo, false);
-		ttm_bo_unreserve(bo);
+		ttm_bo_reserve(&vbo->base, false, false, false, NULL);
+		vmw_bo_pin_reserved(vbo, false);
+		ttm_bo_unreserve(&vbo->base);
 	}
 
 	vmw_resource_unreserve(res, NULL, 0UL);
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 15/28] drm/vmwgfx: Add kms helpers for dirty- and readback functions

[Thomas Hellstrom]

We need to make the dirty- and readback functions callable without a struct
drm_file pointer. We also need to unify the handling of dirty- and readback
cliprects that are now implemented in various places across the kms system,
som add helpers to facilitate this.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.h      |  10 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c  |  19 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_kms.c      | 292 +++++++++++++++++++++++++++++++
 drivers/gpu/drm/vmwgfx/vmwgfx_kms.h      |  71 +++++++-
 drivers/gpu/drm/vmwgfx/vmwgfx_resource.c |  18 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c     |   2 +-
 6 files changed, 391 insertions(+), 21 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index 12eaa6c..7504f92 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -640,7 +640,8 @@ extern struct vmw_resource *vmw_resource_reference(struct vmw_resource *res);
 extern struct vmw_resource *
 vmw_resource_reference_unless_doomed(struct vmw_resource *res);
 extern int vmw_resource_validate(struct vmw_resource *res);
-extern int vmw_resource_reserve(struct vmw_resource *res, bool no_backup);
+extern int vmw_resource_reserve(struct vmw_resource *res, bool interruptible,
+				bool no_backup);
 extern bool vmw_resource_needs_backup(const struct vmw_resource *res);
 extern int vmw_user_lookup_handle(struct vmw_private *dev_priv,
 				  struct ttm_object_file *tfile,
@@ -860,6 +861,11 @@ extern void vmw_execbuf_copy_fence_user(struct vmw_private *dev_priv,
 					*user_fence_rep,
 					struct vmw_fence_obj *fence,
 					uint32_t fence_handle);
+extern int vmw_validate_single_buffer(struct vmw_private *dev_priv,
+				      struct ttm_buffer_object *bo,
+				      bool interruptible,
+				      bool validate_as_mob);
+
 
 /**
  * IRQs and wating - vmwgfx_irq.c
@@ -965,7 +971,7 @@ int vmw_dumb_map_offset(struct drm_file *file_priv,
 int vmw_dumb_destroy(struct drm_file *file_priv,
 		     struct drm_device *dev,
 		     uint32_t handle);
-extern int vmw_resource_pin(struct vmw_resource *res);
+extern int vmw_resource_pin(struct vmw_resource *res, bool interruptible);
 extern void vmw_resource_unpin(struct vmw_resource *res);
 
 /**
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
index 92e8998..698a0e2 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
@@ -375,7 +375,7 @@ static int vmw_resources_reserve(struct vmw_sw_context *sw_context)
 	list_for_each_entry(val, &sw_context->resource_list, head) {
 		struct vmw_resource *res = val->res;
 
-		ret = vmw_resource_reserve(res, val->no_buffer_needed);
+		ret = vmw_resource_reserve(res, true, val->no_buffer_needed);
 		if (unlikely(ret != 0))
 			return ret;
 
@@ -2234,9 +2234,10 @@ static void vmw_clear_validations(struct vmw_sw_context *sw_context)
 		(void) drm_ht_remove_item(&sw_context->res_ht, &val->hash);
 }
 
-static int vmw_validate_single_buffer(struct vmw_private *dev_priv,
-				      struct ttm_buffer_object *bo,
-				      bool validate_as_mob)
+int vmw_validate_single_buffer(struct vmw_private *dev_priv,
+			       struct ttm_buffer_object *bo,
+			       bool interruptible,
+			       bool validate_as_mob)
 {
 	struct vmw_dma_buffer *vbo = container_of(bo, struct vmw_dma_buffer,
 						  base);
@@ -2246,7 +2247,8 @@ static int vmw_validate_single_buffer(struct vmw_private *dev_priv,
 		return 0;
 
 	if (validate_as_mob)
-		return ttm_bo_validate(bo, &vmw_mob_placement, true, false);
+		return ttm_bo_validate(bo, &vmw_mob_placement, interruptible,
+				       false);
 
 	/**
 	 * Put BO in VRAM if there is space, otherwise as a GMR.
@@ -2255,7 +2257,8 @@ static int vmw_validate_single_buffer(struct vmw_private *dev_priv,
 	 * used as a GMR, this will return -ENOMEM.
 	 */
 
-	ret = ttm_bo_validate(bo, &vmw_vram_gmr_placement, true, false);
+	ret = ttm_bo_validate(bo, &vmw_vram_gmr_placement, interruptible,
+			      false);
 	if (likely(ret == 0 || ret == -ERESTARTSYS))
 		return ret;
 
@@ -2264,8 +2267,7 @@ static int vmw_validate_single_buffer(struct vmw_private *dev_priv,
 	 * previous contents.
 	 */
 
-	DRM_INFO("Falling through to VRAM.\n");
-	ret = ttm_bo_validate(bo, &vmw_vram_placement, true, false);
+	ret = ttm_bo_validate(bo, &vmw_vram_placement, interruptible, false);
 	return ret;
 }
 
@@ -2277,6 +2279,7 @@ static int vmw_validate_buffers(struct vmw_private *dev_priv,
 
 	list_for_each_entry(entry, &sw_context->validate_nodes, base.head) {
 		ret = vmw_validate_single_buffer(dev_priv, entry->base.bo,
+						 true,
 						 entry->validate_as_mob);
 		if (unlikely(ret != 0))
 			return ret;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
index 99e2f5b..c46c688 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
@@ -1821,3 +1821,295 @@ out_free:
 	kfree(rects);
 	return ret;
 }
+
+/**
+ * vmw_kms_helper_dirty - Helper to build commands and perform actions based
+ * on a set of cliprects and a set of display units.
+ *
+ * @dev_priv: Pointer to a device private structure.
+ * @framebuffer: Pointer to the framebuffer on which to perform the actions.
+ * @clips: A set of struct drm_clip_rect. Either this os @vclips must be NULL.
+ * Cliprects are given in framebuffer coordinates.
+ * @vclips: A set of struct drm_vmw_rect cliprects. Either this or @clips must
+ * be NULL. Cliprects are given in source coordinates.
+ * @dest_x: X coordinate offset for the crtc / destination clip rects.
+ * @dest_y: Y coordinate offset for the crtc / destination clip rects.
+ * @num_clips: Number of cliprects in the @clips or @vclips array.
+ * @increment: Integer with which to increment the clip counter when looping.
+ * Used to skip a predetermined number of clip rects.
+ * @dirty: Closure structure. See the description of struct vmw_kms_dirty.
+ */
+int vmw_kms_helper_dirty(struct vmw_private *dev_priv,
+			 struct vmw_framebuffer *framebuffer,
+			 const struct drm_clip_rect *clips,
+			 const struct drm_vmw_rect *vclips,
+			 s32 dest_x, s32 dest_y,
+			 int num_clips,
+			 int increment,
+			 struct vmw_kms_dirty *dirty)
+{
+	struct vmw_display_unit *units[VMWGFX_NUM_DISPLAY_UNITS];
+	struct drm_crtc *crtc;
+	u32 num_units = 0;
+	u32 i, k;
+	int ret;
+
+	dirty->dev_priv = dev_priv;
+
+	list_for_each_entry(crtc, &dev_priv->dev->mode_config.crtc_list, head) {
+		if (crtc->primary->fb != &framebuffer->base)
+			continue;
+		units[num_units++] = vmw_crtc_to_du(crtc);
+	}
+
+	for (k = 0; k < num_units; k++) {
+		struct vmw_display_unit *unit = units[k];
+		s32 crtc_x = unit->crtc.x;
+		s32 crtc_y = unit->crtc.y;
+		s32 crtc_width = unit->crtc.mode.hdisplay;
+		s32 crtc_height = unit->crtc.mode.vdisplay;
+		const struct drm_clip_rect *clips_ptr = clips;
+		const struct drm_vmw_rect *vclips_ptr = vclips;
+
+		dirty->unit = unit;
+		if (dirty->fifo_reserve_size > 0) {
+			dirty->cmd = vmw_fifo_reserve(dev_priv,
+						      dirty->fifo_reserve_size);
+			if (!dirty->cmd) {
+				DRM_ERROR("Couldn't reserve fifo space "
+					  "for dirty blits.\n");
+				return ret;
+			}
+			memset(dirty->cmd, 0, dirty->fifo_reserve_size);
+		}
+		dirty->num_hits = 0;
+		for (i = 0; i < num_clips; i++, clips_ptr += increment,
+		       vclips_ptr += increment) {
+			s32 clip_left;
+			s32 clip_top;
+
+			/*
+			 * Select clip array type. Note that integer type
+			 * in @clips is unsigned short, whereas in @vclips
+			 * it's 32-bit.
+			 */
+			if (clips) {
+				dirty->fb_x = (s32) clips_ptr->x1;
+				dirty->fb_y = (s32) clips_ptr->y1;
+				dirty->unit_x2 = (s32) clips_ptr->x2 + dest_x -
+					crtc_x;
+				dirty->unit_y2 = (s32) clips_ptr->y2 + dest_y -
+					crtc_y;
+			} else {
+				dirty->fb_x = vclips_ptr->x;
+				dirty->fb_y = vclips_ptr->y;
+				dirty->unit_x2 = dirty->fb_x + vclips_ptr->w +
+					dest_x - crtc_x;
+				dirty->unit_y2 = dirty->fb_y + vclips_ptr->h +
+					dest_y - crtc_y;
+			}
+
+			dirty->unit_x1 = dirty->fb_x + dest_x - crtc_x;
+			dirty->unit_y1 = dirty->fb_y + dest_y - crtc_y;
+
+			/* Skip this clip if it's outside the crtc region */
+			if (dirty->unit_x1 >= crtc_width ||
+			    dirty->unit_y1 >= crtc_height ||
+			    dirty->unit_x2 <= 0 || dirty->unit_y2 <= 0)
+				continue;
+
+			/* Clip right and bottom to crtc limits */
+			dirty->unit_x2 = min_t(s32, dirty->unit_x2,
+					       crtc_width);
+			dirty->unit_y2 = min_t(s32, dirty->unit_y2,
+					       crtc_height);
+
+			/* Clip left and top to crtc limits */
+			clip_left = min_t(s32, dirty->unit_x1, 0);
+			clip_top = min_t(s32, dirty->unit_y1, 0);
+			dirty->unit_x1 -= clip_left;
+			dirty->unit_y1 -= clip_top;
+			dirty->fb_x -= clip_left;
+			dirty->fb_y -= clip_top;
+
+			dirty->clip(dirty);
+		}
+
+		dirty->fifo_commit(dirty);
+	}
+
+	return 0;
+}
+
+/**
+ * vmw_kms_helper_buffer_prepare - Reserve and validate a buffer object before
+ * command submission.
+ *
+ * @dev_priv. Pointer to a device private structure.
+ * @buf: The buffer object
+ * @interruptible: Whether to perform waits as interruptible.
+ * @validate_as_mob: Whether the buffer should be validated as a MOB. If false,
+ * The buffer will be validated as a GMR. Already pinned buffers will not be
+ * validated.
+ *
+ * Returns 0 on success, negative error code on failure, -ERESTARTSYS if
+ * interrupted by a signal.
+ */
+int vmw_kms_helper_buffer_prepare(struct vmw_private *dev_priv,
+				  struct vmw_dma_buffer *buf,
+				  bool interruptible,
+				  bool validate_as_mob)
+{
+	struct ttm_buffer_object *bo = &buf->base;
+	int ret;
+
+	ttm_bo_reserve(bo, false, false, interruptible, 0);
+	ret = vmw_validate_single_buffer(dev_priv, bo, interruptible,
+					 validate_as_mob);
+	if (ret)
+		ttm_bo_unreserve(bo);
+
+	return ret;
+}
+
+/**
+ * vmw_kms_helper_buffer_revert - Undo the actions of
+ * vmw_kms_helper_buffer_prepare.
+ *
+ * @res: Pointer to the buffer object.
+ *
+ * Helper to be used if an error forces the caller to undo the actions of
+ * vmw_kms_helper_buffer_prepare.
+ */
+void vmw_kms_helper_buffer_revert(struct vmw_dma_buffer *buf)
+{
+	if (buf)
+		ttm_bo_unreserve(&buf->base);
+}
+
+/**
+ * vmw_kms_helper_buffer_finish - Unreserve and fence a buffer object after
+ * kms command submission.
+ *
+ * @dev_priv: Pointer to a device private structure.
+ * @file_priv: Pointer to a struct drm_file representing the caller's
+ * connection. Must be set to NULL if @user_fence_rep is NULL, and conversely
+ * if non-NULL, @user_fence_rep must be non-NULL.
+ * @buf: The buffer object.
+ * @out_fence:  Optional pointer to a fence pointer. If non-NULL, a
+ * ref-counted fence pointer is returned here.
+ * @user_fence_rep: Optional pointer to a user-space provided struct
+ * drm_vmw_fence_rep. If provided, @file_priv must also be provided and the
+ * function copies fence data to user-space in a fail-safe manner.
+ */
+void vmw_kms_helper_buffer_finish(struct vmw_private *dev_priv,
+				  struct drm_file *file_priv,
+				  struct vmw_dma_buffer *buf,
+				  struct vmw_fence_obj **out_fence,
+				  struct drm_vmw_fence_rep __user *
+				  user_fence_rep)
+{
+	struct vmw_fence_obj *fence;
+	uint32_t handle;
+	int ret;
+
+	ret = vmw_execbuf_fence_commands(file_priv, dev_priv, &fence,
+					 file_priv ? &handle : NULL);
+	if (buf)
+		vmw_fence_single_bo(&buf->base, fence);
+	if (file_priv)
+		vmw_execbuf_copy_fence_user(dev_priv, vmw_fpriv(file_priv),
+					    ret, user_fence_rep, fence,
+					    handle);
+	if (out_fence)
+		*out_fence = fence;
+	else
+		vmw_fence_obj_unreference(&fence);
+
+	vmw_kms_helper_buffer_revert(buf);
+}
+
+
+/**
+ * vmw_kms_helper_resource_revert - Undo the actions of
+ * vmw_kms_helper_resource_prepare.
+ *
+ * @res: Pointer to the resource. Typically a surface.
+ *
+ * Helper to be used if an error forces the caller to undo the actions of
+ * vmw_kms_helper_resource_prepare.
+ */
+void vmw_kms_helper_resource_revert(struct vmw_resource *res)
+{
+	vmw_kms_helper_buffer_revert(res->backup);
+	vmw_resource_unreserve(res, NULL, 0);
+	mutex_unlock(&res->dev_priv->cmdbuf_mutex);
+}
+
+/**
+ * vmw_kms_helper_resource_prepare - Reserve and validate a resource before
+ * command submission.
+ *
+ * @res: Pointer to the resource. Typically a surface.
+ * @interruptible: Whether to perform waits as interruptible.
+ *
+ * Reserves and validates also the backup buffer if a guest-backed resource.
+ * Returns 0 on success, negative error code on failure. -ERESTARTSYS if
+ * interrupted by a signal.
+ */
+int vmw_kms_helper_resource_prepare(struct vmw_resource *res,
+				    bool interruptible)
+{
+	int ret = 0;
+
+	if (interruptible)
+		ret = mutex_lock_interruptible(&res->dev_priv->cmdbuf_mutex);
+	else
+		mutex_lock(&res->dev_priv->cmdbuf_mutex);
+
+	if (unlikely(ret != 0))
+		return -ERESTARTSYS;
+
+	ret = vmw_resource_reserve(res, interruptible, false);
+	if (ret)
+		goto out_unlock;
+
+	if (res->backup) {
+		ret = vmw_kms_helper_buffer_prepare(res->dev_priv, res->backup,
+						    interruptible,
+						    res->dev_priv->has_mob);
+		if (ret)
+			goto out_unreserve;
+	}
+	ret = vmw_resource_validate(res);
+	if (ret)
+		goto out_revert;
+	return 0;
+
+out_revert:
+	vmw_kms_helper_buffer_revert(res->backup);
+out_unreserve:
+	vmw_resource_unreserve(res, NULL, 0);
+out_unlock:
+	mutex_unlock(&res->dev_priv->cmdbuf_mutex);
+	return ret;
+}
+
+/**
+ * vmw_kms_helper_resource_finish - Unreserve and fence a resource after
+ * kms command submission.
+ *
+ * @res: Pointer to the resource. Typically a surface.
+ * @out_fence: Optional pointer to a fence pointer. If non-NULL, a
+ * ref-counted fence pointer is returned here.
+ */
+void vmw_kms_helper_resource_finish(struct vmw_resource *res,
+			     struct vmw_fence_obj **out_fence)
+{
+	if (res->backup || out_fence)
+		vmw_kms_helper_buffer_finish(res->dev_priv, NULL, res->backup,
+					     out_fence, NULL);
+
+	vmw_resource_unreserve(res, NULL, 0);
+	mutex_unlock(&res->dev_priv->cmdbuf_mutex);
+}
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
index db8ae94..c19a515 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
@@ -32,7 +32,50 @@
 #include <drm/drm_crtc_helper.h>
 #include "vmwgfx_drv.h"
 
-
+/**
+ * struct vmw_kms_dirty - closure structure for the vmw_kms_helper_dirty
+ * function.
+ *
+ * @fifo_commit: Callback that is called once for each display unit after
+ * all clip rects. This function must commit the fifo space reserved by the
+ * helper. Set up by the caller.
+ * @clip: Callback that is called for each cliprect on each display unit.
+ * Set up by the caller.
+ * @fifo_reserve_size: Fifo size that the helper should try to allocat for
+ * each display unit. Set up by the caller.
+ * @dev_priv: Pointer to the device private. Set up by the helper.
+ * @unit: The current display unit. Set up by the helper before a call to @clip.
+ * @cmd: The allocated fifo space. Set up by the helper before the first @clip
+ * call.
+ * @num_hits: Number of clip rect commands for this display unit.
+ * Cleared by the helper before the first @clip call. Updated by the @clip
+ * callback.
+ * @fb_x: Clip rect left side in framebuffer coordinates.
+ * @fb_y: Clip rect right side in framebuffer coordinates.
+ * @unit_x1: Clip rect left side in crtc coordinates.
+ * @unit_y1: Clip rect top side in crtc coordinates.
+ * @unit_x2: Clip rect right side in crtc coordinates.
+ * @unit_y2: Clip rect bottom side in crtc coordinates.
+ *
+ * The clip rect coordinates are updated by the helper for each @clip call.
+ * Note that this may be derived from if more info needs to be passed between
+ * helper caller and helper callbacks.
+ */
+struct vmw_kms_dirty {
+	void (*fifo_commit)(struct vmw_kms_dirty *);
+	void (*clip)(struct vmw_kms_dirty *);
+	size_t fifo_reserve_size;
+	struct vmw_private *dev_priv;
+	struct vmw_display_unit *unit;
+	void *cmd;
+	u32 num_hits;
+	s32 fb_x;
+	s32 fb_y;
+	s32 unit_x1;
+	s32 unit_y1;
+	s32 unit_x2;
+	s32 unit_y2;
+};
 
 #define VMWGFX_NUM_DISPLAY_UNITS 8
 
@@ -173,7 +216,31 @@ int vmw_du_connector_fill_modes(struct drm_connector *connector,
 int vmw_du_connector_set_property(struct drm_connector *connector,
 				  struct drm_property *property,
 				  uint64_t val);
-
+int vmw_kms_helper_dirty(struct vmw_private *dev_priv,
+			 struct vmw_framebuffer *framebuffer,
+			 const struct drm_clip_rect *clips,
+			 const struct drm_vmw_rect *vclips,
+			 s32 dest_x, s32 dest_y,
+			 int num_clips,
+			 int increment,
+			 struct vmw_kms_dirty *dirty);
+
+int vmw_kms_helper_buffer_prepare(struct vmw_private *dev_priv,
+				  struct vmw_dma_buffer *buf,
+				  bool interruptible,
+				  bool validate_as_mob);
+void vmw_kms_helper_buffer_revert(struct vmw_dma_buffer *buf);
+void vmw_kms_helper_buffer_finish(struct vmw_private *dev_priv,
+				  struct drm_file *file_priv,
+				  struct vmw_dma_buffer *buf,
+				  struct vmw_fence_obj **out_fence,
+				  struct drm_vmw_fence_rep __user *
+				  user_fence_rep);
+int vmw_kms_helper_resource_prepare(struct vmw_resource *res,
+				    bool interruptible);
+void vmw_kms_helper_resource_revert(struct vmw_resource *res);
+void vmw_kms_helper_resource_finish(struct vmw_resource *res,
+				    struct vmw_fence_obj **out_fence);
 
 /*
  * Legacy display unit functions - vmwgfx_ldu.c
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
index 271bc90..521f194 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
@@ -1259,7 +1259,8 @@ out_no_reserve:
  * the buffer may not be bound to the resource at this point.
  *
  */
-int vmw_resource_reserve(struct vmw_resource *res, bool no_backup)
+int vmw_resource_reserve(struct vmw_resource *res, bool interruptible,
+			 bool no_backup)
 {
 	struct vmw_private *dev_priv = res->dev_priv;
 	int ret;
@@ -1270,7 +1271,7 @@ int vmw_resource_reserve(struct vmw_resource *res, bool no_backup)
 
 	if (res->func->needs_backup && res->backup == NULL &&
 	    !no_backup) {
-		ret = vmw_resource_buf_alloc(res, true);
+		ret = vmw_resource_buf_alloc(res, interruptible);
 		if (unlikely(ret != 0))
 			return ret;
 	}
@@ -1584,14 +1585,14 @@ void vmw_resource_evict_all(struct vmw_private *dev_priv)
  * its id will never change as long as there is a pin reference.
  * This function returns 0 on success and a negative error code on failure.
  */
-int vmw_resource_pin(struct vmw_resource *res)
+int vmw_resource_pin(struct vmw_resource *res, bool interruptible)
 {
 	struct vmw_private *dev_priv = res->dev_priv;
 	int ret;
 
-	ttm_write_lock(&dev_priv->reservation_sem, false);
+	ttm_write_lock(&dev_priv->reservation_sem, interruptible);
 	mutex_lock(&dev_priv->cmdbuf_mutex);
-	ret = vmw_resource_reserve(res, false);
+	ret = vmw_resource_reserve(res, interruptible, false);
 	if (ret)
 		goto out_no_reserve;
 
@@ -1601,12 +1602,13 @@ int vmw_resource_pin(struct vmw_resource *res)
 		if (res->backup) {
 			vbo = res->backup;
 
-			ttm_bo_reserve(&vbo->base, false, false, false, NULL);
+			ttm_bo_reserve(&vbo->base, interruptible, false, false,
+				       NULL);
 			if (!vbo->pin_count) {
 				ret = ttm_bo_validate
 					(&vbo->base,
 					 res->func->backup_placement,
-					 false, false);
+					 interruptible, false);
 				if (ret) {
 					ttm_bo_unreserve(&vbo->base);
 					goto out_no_validate;
@@ -1649,7 +1651,7 @@ void vmw_resource_unpin(struct vmw_resource *res)
 	ttm_read_lock(&dev_priv->reservation_sem, false);
 	mutex_lock(&dev_priv->cmdbuf_mutex);
 
-	ret = vmw_resource_reserve(res, true);
+	ret = vmw_resource_reserve(res, false, true);
 	WARN_ON(ret);
 
 	WARN_ON(res->pin_count == 0);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c b/drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c
index ef99df7..becf965 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c
@@ -91,7 +91,7 @@ static void vmw_stdu_destroy(struct vmw_screen_target_display_unit *stdu);
  */
 static int vmw_stdu_pin_display(struct vmw_screen_target_display_unit *stdu)
 {
-	return vmw_resource_pin(&stdu->display_srf->res);
+	return vmw_resource_pin(&stdu->display_srf->res, false);
 }
 
 
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 16/28] drm/vmwgfx: Convert screen objects to the new helpers

[Thomas Hellstrom]

This makes it possible to use the same function for surface dirty and
present. Also fixes page flip without events.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.h  |   6 -
 drivers/gpu/drm/vmwgfx/vmwgfx_kms.c  | 319 +++--------------
 drivers/gpu/drm/vmwgfx/vmwgfx_kms.h  |  33 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c | 642 +++++++++++++++++++++--------------
 4 files changed, 444 insertions(+), 556 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index 7504f92..d60ae20 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -952,12 +952,6 @@ int vmw_kms_present(struct vmw_private *dev_priv,
 		    uint32_t sid, int32_t destX, int32_t destY,
 		    struct drm_vmw_rect *clips,
 		    uint32_t num_clips);
-int vmw_kms_readback(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_vmw_rect *clips,
-		     uint32_t num_clips);
 int vmw_kms_update_layout_ioctl(struct drm_device *dev, void *data,
 				struct drm_file *file_priv);
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
index c46c688..5901d32 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
@@ -31,41 +31,6 @@
 /* Might need a hrtimer here? */
 #define VMWGFX_PRESENT_RATE ((HZ / 60 > 0) ? HZ / 60 : 1)
 
-
-
-/**
- * Clip @num_rects number of @rects against @clip storing the
- * results in @out_rects and the number of passed rects in @out_num.
- */
-void vmw_clip_cliprects(struct drm_clip_rect *rects,
-			int num_rects,
-			struct vmw_clip_rect clip,
-			SVGASignedRect *out_rects,
-			int *out_num)
-{
-	int i, k;
-
-	for (i = 0, k = 0; i < num_rects; i++) {
-		int x1 = max_t(int, clip.x1, rects[i].x1);
-		int y1 = max_t(int, clip.y1, rects[i].y1);
-		int x2 = min_t(int, clip.x2, rects[i].x2);
-		int y2 = min_t(int, clip.y2, rects[i].y2);
-
-		if (x1 >= x2)
-			continue;
-		if (y1 >= y2)
-			continue;
-
-		out_rects[k].left   = x1;
-		out_rects[k].top    = y1;
-		out_rects[k].right  = x2;
-		out_rects[k].bottom = y2;
-		k++;
-	}
-
-	*out_num = k;
-}
-
 void vmw_du_cleanup(struct vmw_display_unit *du)
 {
 	if (du->cursor_surface)
@@ -458,11 +423,9 @@ static int vmw_framebuffer_surface_dirty(struct drm_framebuffer *framebuffer,
 	}
 
 	if (dev_priv->active_display_unit == vmw_du_screen_object)
-		ret = vmw_kms_sou_do_surface_dirty(dev_priv, file_priv,
-						   &vfbs->base,
-						   flags, color,
-						   clips, num_clips,
-						   inc, NULL);
+		ret = vmw_kms_sou_do_surface_dirty(dev_priv, &vfbs->base,
+						   clips, NULL, NULL, 0, 0,
+						   num_clips, inc, NULL);
 	else
 		ret = vmw_kms_stdu_do_surface_dirty(dev_priv, file_priv,
 						    &vfbs->base,
@@ -477,6 +440,42 @@ static int vmw_framebuffer_surface_dirty(struct drm_framebuffer *framebuffer,
 	return 0;
 }
 
+/**
+ * vmw_kms_readback - Perform a readback from the screen system to
+ * a dma-buffer backed framebuffer.
+ *
+ * @dev_priv: Pointer to the device private structure.
+ * @file_priv: Pointer to a struct drm_file identifying the caller.
+ * Must be set to NULL if @user_fence_rep is NULL.
+ * @vfb: Pointer to the dma-buffer backed framebuffer.
+ * @user_fence_rep: User-space provided structure for fence information.
+ * Must be set to non-NULL if @file_priv is non-NULL.
+ * @vclips: Array of clip rects.
+ * @num_clips: Number of clip rects in @vclips.
+ *
+ * Returns 0 on success, negative error code on failure. -ERESTARTSYS if
+ * interrupted.
+ */
+int vmw_kms_readback(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_vmw_rect *vclips,
+		     uint32_t num_clips)
+{
+	switch (dev_priv->active_display_unit) {
+	case vmw_du_screen_object:
+		return vmw_kms_sou_readback(dev_priv, file_priv, vfb,
+					    user_fence_rep, vclips, num_clips);
+	default:
+		WARN_ONCE(true,
+			  "Readback called with invalid display system.\n");
+	}
+
+	return -ENOSYS;
+}
+
+
 static struct drm_framebuffer_funcs vmw_framebuffer_surface_funcs = {
 	.destroy = vmw_framebuffer_surface_destroy,
 	.dirty = vmw_framebuffer_surface_dirty,
@@ -638,10 +637,9 @@ static int vmw_framebuffer_dmabuf_dirty(struct drm_framebuffer *framebuffer,
 						  flags, color,
 						  clips, num_clips, increment);
 	} else if (dev_priv->active_display_unit == vmw_du_screen_object) {
-		ret = vmw_kms_sou_do_dmabuf_dirty(file_priv, dev_priv,
-						  &vfbd->base,
-						  flags, color,
+		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,
@@ -984,131 +982,9 @@ int vmw_kms_generic_present(struct vmw_private *dev_priv,
 		    struct drm_vmw_rect *clips,
 		    uint32_t num_clips)
 {
-	struct vmw_display_unit *units[VMWGFX_NUM_DISPLAY_UNITS];
-	struct drm_clip_rect *tmp;
-	struct drm_crtc *crtc;
-	size_t fifo_size;
-	int i, k, num_units;
-	int ret = 0; /* silence warning */
-	int left, right, top, bottom;
-
-	struct {
-		SVGA3dCmdHeader header;
-		SVGA3dCmdBlitSurfaceToScreen body;
-	} *cmd;
-	SVGASignedRect *blits;
-
-	num_units = 0;
-	list_for_each_entry(crtc, &dev_priv->dev->mode_config.crtc_list, head) {
-		if (crtc->primary->fb != &vfb->base)
-			continue;
-		units[num_units++] = vmw_crtc_to_du(crtc);
-	}
-
-	BUG_ON(surface == NULL);
-	BUG_ON(!clips || !num_clips);
-
-	tmp = kzalloc(sizeof(*tmp) * num_clips, GFP_KERNEL);
-	if (unlikely(tmp == NULL)) {
-		DRM_ERROR("Temporary cliprect memory alloc failed.\n");
-		return -ENOMEM;
-	}
-
-	fifo_size = sizeof(*cmd) + sizeof(SVGASignedRect) * num_clips;
-	cmd = kmalloc(fifo_size, GFP_KERNEL);
-	if (unlikely(cmd == NULL)) {
-		DRM_ERROR("Failed to allocate temporary fifo memory.\n");
-		ret = -ENOMEM;
-		goto out_free_tmp;
-	}
-
-	left = clips->x;
-	right = clips->x + clips->w;
-	top = clips->y;
-	bottom = clips->y + clips->h;
-
-	for (i = 1; i < num_clips; i++) {
-		left = min_t(int, left, (int)clips[i].x);
-		right = max_t(int, right, (int)clips[i].x + clips[i].w);
-		top = min_t(int, top, (int)clips[i].y);
-		bottom = max_t(int, bottom, (int)clips[i].y + clips[i].h);
-	}
-
-	/* only need to do this once */
-	memset(cmd, 0, fifo_size);
-	cmd->header.id = cpu_to_le32(SVGA_3D_CMD_BLIT_SURFACE_TO_SCREEN);
-
-	blits = (SVGASignedRect *)&cmd[1];
-
-	cmd->body.srcRect.left = left;
-	cmd->body.srcRect.right = right;
-	cmd->body.srcRect.top = top;
-	cmd->body.srcRect.bottom = bottom;
-
-	for (i = 0; i < num_clips; i++) {
-		tmp[i].x1 = clips[i].x - left;
-		tmp[i].x2 = clips[i].x + clips[i].w - left;
-		tmp[i].y1 = clips[i].y - top;
-		tmp[i].y2 = clips[i].y + clips[i].h - top;
-	}
-
-	for (k = 0; k < num_units; k++) {
-		struct vmw_display_unit *unit = units[k];
-		struct vmw_clip_rect clip;
-		int num;
-
-		clip.x1 = left + destX - unit->crtc.x;
-		clip.y1 = top + destY - unit->crtc.y;
-		clip.x2 = right + destX - unit->crtc.x;
-		clip.y2 = bottom + destY - unit->crtc.y;
-
-		/* skip any crtcs that misses the clip region */
-		if (clip.x1 >= unit->crtc.mode.hdisplay ||
-		    clip.y1 >= unit->crtc.mode.vdisplay ||
-		    clip.x2 <= 0 || clip.y2 <= 0)
-			continue;
-
-		/*
-		 * In order for the clip rects to be correctly scaled
-		 * the src and dest rects needs to be the same size.
-		 */
-		cmd->body.destRect.left = clip.x1;
-		cmd->body.destRect.right = clip.x2;
-		cmd->body.destRect.top = clip.y1;
-		cmd->body.destRect.bottom = clip.y2;
-
-		/* create a clip rect of the crtc in dest coords */
-		clip.x2 = unit->crtc.mode.hdisplay - clip.x1;
-		clip.y2 = unit->crtc.mode.vdisplay - clip.y1;
-		clip.x1 = 0 - clip.x1;
-		clip.y1 = 0 - clip.y1;
-
-		/* need to reset sid as it is changed by execbuf */
-		cmd->body.srcImage.sid = sid;
-		cmd->body.destScreenId = unit->unit;
-
-		/* clip and write blits to cmd stream */
-		vmw_clip_cliprects(tmp, num_clips, clip, blits, &num);
-
-		/* if no cliprects hit skip this */
-		if (num == 0)
-			continue;
-
-		/* recalculate package length */
-		fifo_size = sizeof(*cmd) + sizeof(SVGASignedRect) * num;
-		cmd->header.size = cpu_to_le32(fifo_size - sizeof(cmd->header));
-		ret = vmw_execbuf_process(file_priv, dev_priv, NULL, cmd,
-					  fifo_size, 0, 0, NULL, NULL);
-
-		if (unlikely(ret != 0))
-			break;
-	}
-
-	kfree(cmd);
-out_free_tmp:
-	kfree(tmp);
-
-	return ret;
+	return vmw_kms_sou_do_surface_dirty(dev_priv, vfb, NULL, clips,
+					    &surface->res, destX, destY,
+					    num_clips, 1, NULL);
 }
 
 int vmw_kms_present(struct vmw_private *dev_priv,
@@ -1137,113 +1013,6 @@ int vmw_kms_present(struct vmw_private *dev_priv,
 	return 0;
 }
 
-int vmw_kms_readback(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_vmw_rect *clips,
-		     uint32_t num_clips)
-{
-	struct vmw_framebuffer_dmabuf *vfbd =
-		vmw_framebuffer_to_vfbd(&vfb->base);
-	struct vmw_dma_buffer *dmabuf = vfbd->buffer;
-	struct vmw_display_unit *units[VMWGFX_NUM_DISPLAY_UNITS];
-	struct drm_crtc *crtc;
-	size_t fifo_size;
-	int i, k, ret, num_units, blits_pos;
-
-	struct {
-		uint32_t header;
-		SVGAFifoCmdDefineGMRFB body;
-	} *cmd;
-	struct {
-		uint32_t header;
-		SVGAFifoCmdBlitScreenToGMRFB body;
-	} *blits;
-
-	num_units = 0;
-	list_for_each_entry(crtc, &dev_priv->dev->mode_config.crtc_list, head) {
-		if (crtc->primary->fb != &vfb->base)
-			continue;
-		units[num_units++] = vmw_crtc_to_du(crtc);
-	}
-
-	BUG_ON(dmabuf == NULL);
-	BUG_ON(!clips || !num_clips);
-
-	/* take a safe guess at fifo size */
-	fifo_size = sizeof(*cmd) + sizeof(*blits) * num_clips * num_units;
-	cmd = kmalloc(fifo_size, GFP_KERNEL);
-	if (unlikely(cmd == NULL)) {
-		DRM_ERROR("Failed to allocate temporary fifo memory.\n");
-		return -ENOMEM;
-	}
-
-	memset(cmd, 0, fifo_size);
-	cmd->header = SVGA_CMD_DEFINE_GMRFB;
-	cmd->body.format.bitsPerPixel = vfb->base.bits_per_pixel;
-	cmd->body.format.colorDepth = vfb->base.depth;
-	cmd->body.format.reserved = 0;
-	cmd->body.bytesPerLine = vfb->base.pitches[0];
-	cmd->body.ptr.gmrId = vfb->user_handle;
-	cmd->body.ptr.offset = 0;
-
-	blits = (void *)&cmd[1];
-	blits_pos = 0;
-	for (i = 0; i < num_units; i++) {
-		struct drm_vmw_rect *c = clips;
-		for (k = 0; k < num_clips; k++, c++) {
-			/* transform clip coords to crtc origin based coords */
-			int clip_x1 = c->x - units[i]->crtc.x;
-			int clip_x2 = c->x - units[i]->crtc.x + c->w;
-			int clip_y1 = c->y - units[i]->crtc.y;
-			int clip_y2 = c->y - units[i]->crtc.y + c->h;
-			int dest_x = c->x;
-			int dest_y = c->y;
-
-			/* compensate for clipping, we negate
-			 * a negative number and add that.
-			 */
-			if (clip_x1 < 0)
-				dest_x += -clip_x1;
-			if (clip_y1 < 0)
-				dest_y += -clip_y1;
-
-			/* clip */
-			clip_x1 = max(clip_x1, 0);
-			clip_y1 = max(clip_y1, 0);
-			clip_x2 = min(clip_x2, units[i]->crtc.mode.hdisplay);
-			clip_y2 = min(clip_y2, units[i]->crtc.mode.vdisplay);
-
-			/* and cull any rects that misses the crtc */
-			if (clip_x1 >= units[i]->crtc.mode.hdisplay ||
-			    clip_y1 >= units[i]->crtc.mode.vdisplay ||
-			    clip_x2 <= 0 || clip_y2 <= 0)
-				continue;
-
-			blits[blits_pos].header = SVGA_CMD_BLIT_SCREEN_TO_GMRFB;
-			blits[blits_pos].body.srcScreenId = units[i]->unit;
-			blits[blits_pos].body.destOrigin.x = dest_x;
-			blits[blits_pos].body.destOrigin.y = dest_y;
-
-			blits[blits_pos].body.srcRect.left = clip_x1;
-			blits[blits_pos].body.srcRect.top = clip_y1;
-			blits[blits_pos].body.srcRect.right = clip_x2;
-			blits[blits_pos].body.srcRect.bottom = clip_y2;
-			blits_pos++;
-		}
-	}
-	/* reset size here and use calculated exact size from loops */
-	fifo_size = sizeof(*cmd) + sizeof(*blits) * blits_pos;
-
-	ret = vmw_execbuf_process(file_priv, dev_priv, NULL, cmd, fifo_size,
-				  0, 0, user_fence_rep, NULL);
-
-	kfree(cmd);
-
-	return ret;
-}
-
 int vmw_kms_init(struct vmw_private *dev_priv)
 {
 	struct drm_device *dev = dev_priv->dev;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
index c19a515..8a8203c 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
@@ -126,15 +126,6 @@ struct vmw_framebuffer_dmabuf {
 
 
 /*
- * Basic clip rect manipulation
- */
-void vmw_clip_cliprects(struct drm_clip_rect *rects,
-			int num_rects,
-			struct vmw_clip_rect clip,
-			SVGASignedRect *out_rects,
-			int *out_num);
-
-/*
  * Basic cursor manipulation
  */
 int vmw_cursor_update_image(struct vmw_private *dev_priv,
@@ -241,6 +232,12 @@ int vmw_kms_helper_resource_prepare(struct vmw_resource *res,
 void vmw_kms_helper_resource_revert(struct vmw_resource *res);
 void vmw_kms_helper_resource_finish(struct vmw_resource *res,
 				    struct vmw_fence_obj **out_fence);
+int vmw_kms_readback(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_vmw_rect *vclips,
+		     uint32_t num_clips);
 
 /*
  * Legacy display unit functions - vmwgfx_ldu.c
@@ -259,20 +256,26 @@ int vmw_kms_ldu_do_dmabuf_dirty(struct vmw_private *dev_priv,
 int vmw_kms_sou_init_display(struct vmw_private *dev_priv);
 int vmw_kms_sou_close_display(struct vmw_private *dev_priv);
 int vmw_kms_sou_do_surface_dirty(struct vmw_private *dev_priv,
-				 struct drm_file *file_priv,
 				 struct vmw_framebuffer *framebuffer,
-				 unsigned flags, unsigned color,
 				 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_sou_do_dmabuf_dirty(struct drm_file *file_priv,
-				struct vmw_private *dev_priv,
+int vmw_kms_sou_do_dmabuf_dirty(struct vmw_private *dev_priv,
 				struct vmw_framebuffer *framebuffer,
-				unsigned flags, unsigned color,
 				struct drm_clip_rect *clips,
 				unsigned num_clips, int increment,
+				bool interruptible,
 				struct vmw_fence_obj **out_fence);
-
+int vmw_kms_sou_readback(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_vmw_rect *vclips,
+			 uint32_t num_clips);
 
 /*
  * Screen Target Display Unit functions - vmwgfx_stdu.c
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c b/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c
index 0d06d86..73fe20e 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c
@@ -36,10 +36,55 @@
 #define vmw_connector_to_sou(x) \
 	container_of(x, struct vmw_screen_object_unit, base.connector)
 
+/**
+ * struct vmw_kms_sou_surface_dirty - Closure structure for
+ * blit surface to screen command.
+ * @base: The base type we derive from. Used by vmw_kms_helper_dirty().
+ * @left: Left side of bounding box.
+ * @right: Right side of bounding box.
+ * @top: Top side of bounding box.
+ * @bottom: Bottom side of bounding box.
+ * @dst_x: Difference between source clip rects and framebuffer coordinates.
+ * @dst_y: Difference between source clip rects and framebuffer coordinates.
+ * @sid: Surface id of surface to copy from.
+ */
+struct vmw_kms_sou_surface_dirty {
+	struct vmw_kms_dirty base;
+	s32 left, right, top, bottom;
+	s32 dst_x, dst_y;
+	u32 sid;
+};
+
+/*
+ * SVGA commands that are used by this code. Please see the device headers
+ * for explanation.
+ */
+struct vmw_kms_sou_readback_blit {
+	uint32 header;
+	SVGAFifoCmdBlitScreenToGMRFB body;
+};
+
+struct vmw_kms_sou_dmabuf_blit {
+	uint32 header;
+	SVGAFifoCmdBlitGMRFBToScreen body;
+};
+
+struct vmw_kms_sou_dirty_cmd {
+	SVGA3dCmdHeader header;
+	SVGA3dCmdBlitSurfaceToScreen body;
+};
+
+
+/*
+ * Other structs.
+ */
+
 struct vmw_screen_object_display {
 	unsigned num_implicit;
 
 	struct vmw_framebuffer *implicit_fb;
+	SVGAFifoCmdDefineGMRFB cur;
+	struct vmw_dma_buffer *pinned_gmrfb;
 };
 
 /**
@@ -202,14 +247,7 @@ static int vmw_sou_fifo_destroy(struct vmw_private *dev_priv,
 static void vmw_sou_backing_free(struct vmw_private *dev_priv,
 				 struct vmw_screen_object_unit *sou)
 {
-	struct ttm_buffer_object *bo;
-
-	if (unlikely(sou->buffer == NULL))
-		return;
-
-	bo = &sou->buffer->base;
-	ttm_bo_unref(&bo);
-	sou->buffer = NULL;
+	vmw_dmabuf_unreference(&sou->buffer);
 	sou->buffer_size = 0;
 }
 
@@ -432,7 +470,6 @@ static int vmw_sou_crtc_page_flip(struct drm_crtc *crtc,
 	struct vmw_private *dev_priv = vmw_priv(crtc->dev);
 	struct drm_framebuffer *old_fb = crtc->primary->fb;
 	struct vmw_framebuffer *vfb = vmw_framebuffer_to_vfb(fb);
-	struct drm_file *file_priv = event->base.file_priv;
 	struct vmw_fence_obj *fence = NULL;
 	struct drm_clip_rect clips;
 	int ret;
@@ -452,11 +489,13 @@ static int vmw_sou_crtc_page_flip(struct drm_crtc *crtc,
 	clips.y2 = fb->height;
 
 	if (vfb->dmabuf)
-		ret = vmw_kms_sou_do_dmabuf_dirty(file_priv, dev_priv, vfb,
-						  0, 0, &clips, 1, 1, &fence);
+		ret = vmw_kms_sou_do_dmabuf_dirty(dev_priv, vfb,
+						  &clips, 1, 1,
+						  true, &fence);
 	else
-		ret = vmw_kms_sou_do_surface_dirty(dev_priv, file_priv, vfb,
-						   0, 0, &clips, 1, 1, &fence);
+		ret = vmw_kms_sou_do_surface_dirty(dev_priv, vfb,
+						   &clips, NULL, NULL,
+						   0, 0, 1, 1, &fence);
 
 
 	if (ret != 0)
@@ -466,11 +505,15 @@ static int vmw_sou_crtc_page_flip(struct drm_crtc *crtc,
 		goto out_no_fence;
 	}
 
-	ret = vmw_event_fence_action_queue(file_priv, fence,
-					   &event->base,
-					   &event->event.tv_sec,
-					   &event->event.tv_usec,
-					   true);
+	if (event) {
+		struct drm_file *file_priv = event->base.file_priv;
+
+		ret = vmw_event_fence_action_queue(file_priv, fence,
+						   &event->base,
+						   &event->event.tv_sec,
+						   &event->event.tv_usec,
+						   true);
+	}
 
 	/*
 	 * No need to hold on to this now. The only cleanup
@@ -488,153 +531,6 @@ out_no_fence:
 	return ret;
 }
 
-int vmw_kms_sou_do_surface_dirty(struct vmw_private *dev_priv,
-				 struct drm_file *file_priv,
-				 struct vmw_framebuffer *framebuffer,
-				 unsigned flags, unsigned color,
-				 struct drm_clip_rect *clips,
-				 unsigned num_clips, int inc,
-				 struct vmw_fence_obj **out_fence)
-{
-	struct vmw_display_unit *units[VMWGFX_NUM_DISPLAY_UNITS];
-	struct drm_clip_rect *clips_ptr;
-	struct drm_clip_rect *tmp;
-	struct drm_crtc *crtc;
-	size_t fifo_size;
-	int i, num_units;
-	int ret = 0; /* silence warning */
-	int left, right, top, bottom;
-
-	struct {
-		SVGA3dCmdHeader header;
-		SVGA3dCmdBlitSurfaceToScreen body;
-	} *cmd;
-	SVGASignedRect *blits;
-
-	num_units = 0;
-	list_for_each_entry(crtc, &dev_priv->dev->mode_config.crtc_list,
-			    head) {
-		if (crtc->primary->fb != &framebuffer->base)
-			continue;
-		units[num_units++] = vmw_crtc_to_du(crtc);
-	}
-
-	BUG_ON(!clips || !num_clips);
-
-	tmp = kzalloc(sizeof(*tmp) * num_clips, GFP_KERNEL);
-	if (unlikely(tmp == NULL)) {
-		DRM_ERROR("Temporary cliprect memory alloc failed.\n");
-		return -ENOMEM;
-	}
-
-	fifo_size = sizeof(*cmd) + sizeof(SVGASignedRect) * num_clips;
-	cmd = kzalloc(fifo_size, GFP_KERNEL);
-	if (unlikely(cmd == NULL)) {
-		DRM_ERROR("Temporary fifo memory alloc failed.\n");
-		ret = -ENOMEM;
-		goto out_free_tmp;
-	}
-
-	/* setup blits pointer */
-	blits = (SVGASignedRect *)&cmd[1];
-
-	/* initial clip region */
-	left = clips->x1;
-	right = clips->x2;
-	top = clips->y1;
-	bottom = clips->y2;
-
-	/* skip the first clip rect */
-	for (i = 1, clips_ptr = clips + inc;
-	     i < num_clips; i++, clips_ptr += inc) {
-		left = min_t(int, left, (int)clips_ptr->x1);
-		right = max_t(int, right, (int)clips_ptr->x2);
-		top = min_t(int, top, (int)clips_ptr->y1);
-		bottom = max_t(int, bottom, (int)clips_ptr->y2);
-	}
-
-	/* only need to do this once */
-	cmd->header.id = cpu_to_le32(SVGA_3D_CMD_BLIT_SURFACE_TO_SCREEN);
-	cmd->header.size = cpu_to_le32(fifo_size - sizeof(cmd->header));
-
-	cmd->body.srcRect.left = left;
-	cmd->body.srcRect.right = right;
-	cmd->body.srcRect.top = top;
-	cmd->body.srcRect.bottom = bottom;
-
-	clips_ptr = clips;
-	for (i = 0; i < num_clips; i++, clips_ptr += inc) {
-		tmp[i].x1 = clips_ptr->x1 - left;
-		tmp[i].x2 = clips_ptr->x2 - left;
-		tmp[i].y1 = clips_ptr->y1 - top;
-		tmp[i].y2 = clips_ptr->y2 - top;
-	}
-
-	/* do per unit writing, reuse fifo for each */
-	for (i = 0; i < num_units; i++) {
-		struct vmw_display_unit *unit = units[i];
-		struct vmw_clip_rect clip;
-		int num;
-
-		clip.x1 = left - unit->crtc.x;
-		clip.y1 = top - unit->crtc.y;
-		clip.x2 = right - unit->crtc.x;
-		clip.y2 = bottom - unit->crtc.y;
-
-		/* skip any crtcs that misses the clip region */
-		if (clip.x1 >= unit->crtc.mode.hdisplay ||
-		    clip.y1 >= unit->crtc.mode.vdisplay ||
-		    clip.x2 <= 0 || clip.y2 <= 0)
-			continue;
-
-		/*
-		 * In order for the clip rects to be correctly scaled
-		 * the src and dest rects needs to be the same size.
-		 */
-		cmd->body.destRect.left = clip.x1;
-		cmd->body.destRect.right = clip.x2;
-		cmd->body.destRect.top = clip.y1;
-		cmd->body.destRect.bottom = clip.y2;
-
-		/* create a clip rect of the crtc in dest coords */
-		clip.x2 = unit->crtc.mode.hdisplay - clip.x1;
-		clip.y2 = unit->crtc.mode.vdisplay - clip.y1;
-		clip.x1 = 0 - clip.x1;
-		clip.y1 = 0 - clip.y1;
-
-		/* need to reset sid as it is changed by execbuf */
-		cmd->body.srcImage.sid = cpu_to_le32(framebuffer->user_handle);
-		cmd->body.destScreenId = unit->unit;
-
-		/* clip and write blits to cmd stream */
-		vmw_clip_cliprects(tmp, num_clips, clip, blits, &num);
-
-		/* if no cliprects hit skip this */
-		if (num == 0)
-			continue;
-
-		/* only return the last fence */
-		if (out_fence && *out_fence)
-			vmw_fence_obj_unreference(out_fence);
-
-		/* recalculate package length */
-		fifo_size = sizeof(*cmd) + sizeof(SVGASignedRect) * num;
-		cmd->header.size = cpu_to_le32(fifo_size - sizeof(cmd->header));
-		ret = vmw_execbuf_process(file_priv, dev_priv, NULL, cmd,
-					  fifo_size, 0, 0, NULL, out_fence);
-
-		if (unlikely(ret != 0))
-			break;
-	}
-
-
-	kfree(cmd);
-out_free_tmp:
-	kfree(tmp);
-
-	return ret;
-}
-
 static struct drm_crtc_funcs vmw_screen_object_crtc_funcs = {
 	.save = vmw_du_crtc_save,
 	.restore = vmw_du_crtc_restore,
@@ -790,14 +686,13 @@ int vmw_kms_sou_close_display(struct vmw_private *dev_priv)
 	return 0;
 }
 
-static int do_dmabuf_define_gmrfb(struct drm_file *file_priv,
-				  struct vmw_private *dev_priv,
+static int do_dmabuf_define_gmrfb(struct vmw_private *dev_priv,
 				  struct vmw_framebuffer *framebuffer)
 {
+	struct vmw_dma_buffer *buf =
+		container_of(framebuffer, struct vmw_framebuffer_dmabuf,
+			     base)->buffer;
 	int depth = framebuffer->base.depth;
-	size_t fifo_size;
-	int ret;
-
 	struct {
 		uint32_t header;
 		SVGAFifoCmdDefineGMRFB body;
@@ -810,123 +705,350 @@ static int do_dmabuf_define_gmrfb(struct drm_file *file_priv,
 	if (depth == 32)
 		depth = 24;
 
-	fifo_size = sizeof(*cmd);
-	cmd = kmalloc(fifo_size, GFP_KERNEL);
-	if (unlikely(cmd == NULL)) {
-		DRM_ERROR("Failed to allocate temporary cmd buffer.\n");
+	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
+	if (!cmd) {
+		DRM_ERROR("Out of fifo space for dirty framebuffer command.\n");
 		return -ENOMEM;
 	}
 
-	memset(cmd, 0, fifo_size);
 	cmd->header = SVGA_CMD_DEFINE_GMRFB;
 	cmd->body.format.bitsPerPixel = framebuffer->base.bits_per_pixel;
 	cmd->body.format.colorDepth = depth;
 	cmd->body.format.reserved = 0;
 	cmd->body.bytesPerLine = framebuffer->base.pitches[0];
-	cmd->body.ptr.gmrId = framebuffer->user_handle;
-	cmd->body.ptr.offset = 0;
+	/* Buffer is reserved in vram or GMR */
+	vmw_bo_get_guest_ptr(&buf->base, &cmd->body.ptr);
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+
+	return 0;
+}
+
+/**
+ * vmw_sou_surface_fifo_commit - Callback to fill in and submit a
+ * blit surface to screen command.
+ *
+ * @dirty: The closure structure.
+ *
+ * Fills in the missing fields in the command, and translates the cliprects
+ * to match the destination bounding box encoded.
+ */
+static void vmw_sou_surface_fifo_commit(struct vmw_kms_dirty *dirty)
+{
+	struct vmw_kms_sou_surface_dirty *sdirty =
+		container_of(dirty, typeof(*sdirty), base);
+	struct vmw_kms_sou_dirty_cmd *cmd = dirty->cmd;
+	s32 trans_x = dirty->unit->crtc.x - sdirty->dst_x;
+	s32 trans_y = dirty->unit->crtc.y - sdirty->dst_y;
+	size_t region_size = dirty->num_hits * sizeof(SVGASignedRect);
+	SVGASignedRect *blit = (SVGASignedRect *) &cmd[1];
+	int i;
+
+	cmd->header.id = SVGA_3D_CMD_BLIT_SURFACE_TO_SCREEN;
+	cmd->header.size = sizeof(cmd->body) + region_size;
+
+	/*
+	 * Use the destination bounding box to specify destination - and
+	 * source bounding regions.
+	 */
+	cmd->body.destRect.left = sdirty->left;
+	cmd->body.destRect.right = sdirty->right;
+	cmd->body.destRect.top = sdirty->top;
+	cmd->body.destRect.bottom = sdirty->bottom;
+
+	cmd->body.srcRect.left = sdirty->left + trans_x;
+	cmd->body.srcRect.right = sdirty->right + trans_x;
+	cmd->body.srcRect.top = sdirty->top + trans_y;
+	cmd->body.srcRect.bottom = sdirty->bottom + trans_y;
+
+	cmd->body.srcImage.sid = sdirty->sid;
+	cmd->body.destScreenId = dirty->unit->unit;
+
+	/* Blits are relative to the destination rect. Translate. */
+	for (i = 0; i < dirty->num_hits; ++i, ++blit) {
+		blit->left -= sdirty->left;
+		blit->right -= sdirty->left;
+		blit->top -= sdirty->top;
+		blit->bottom -= sdirty->top;
+	}
+
+	vmw_fifo_commit(dirty->dev_priv, region_size + sizeof(*cmd));
+
+	sdirty->left = sdirty->top = S32_MAX;
+	sdirty->right = sdirty->bottom = S32_MIN;
+}
+
+/**
+ * vmw_sou_surface_clip - Callback to encode a blit surface to screen cliprect.
+ *
+ * @dirty: The closure structure
+ *
+ * Encodes a SVGASignedRect cliprect and updates the bounding box of the
+ * BLIT_SURFACE_TO_SCREEN command.
+ */
+static void vmw_sou_surface_clip(struct vmw_kms_dirty *dirty)
+{
+	struct vmw_kms_sou_surface_dirty *sdirty =
+		container_of(dirty, typeof(*sdirty), base);
+	struct vmw_kms_sou_dirty_cmd *cmd = dirty->cmd;
+	SVGASignedRect *blit = (SVGASignedRect *) &cmd[1];
+
+	/* Destination rect. */
+	blit += dirty->num_hits;
+	blit->left = dirty->unit_x1;
+	blit->top = dirty->unit_y1;
+	blit->right = dirty->unit_x2;
+	blit->bottom = dirty->unit_y2;
+
+	/* Destination bounding box */
+	sdirty->left = min_t(s32, sdirty->left, dirty->unit_x1);
+	sdirty->top = min_t(s32, sdirty->top, dirty->unit_y1);
+	sdirty->right = max_t(s32, sdirty->right, dirty->unit_x2);
+	sdirty->bottom = max_t(s32, sdirty->bottom, dirty->unit_y2);
+
+	dirty->num_hits++;
+}
+
+/**
+ * vmw_kms_sou_do_surface_dirty - Dirty part of a surface backed framebuffer
+ *
+ * @dev_priv: Pointer to the device private structure.
+ * @framebuffer: Pointer to the surface-buffer backed framebuffer.
+ * @clips: Array of clip rects. Either @clips or @vclips must be NULL.
+ * @vclips: Alternate array of clip rects. Either @clips or @vclips must
+ * be NULL.
+ * @srf: Pointer to surface to blit from. If NULL, the surface attached
+ * to @framebuffer will be used.
+ * @dest_x: X coordinate offset to align @srf with framebuffer coordinates.
+ * @dest_y: Y coordinate offset to align @srf with framebuffer coordinates.
+ * @num_clips: Number of clip rects in @clips.
+ * @inc: Increment to use when looping over @clips.
+ * @out_fence: If non-NULL, will return a ref-counted pointer to a
+ * struct vmw_fence_obj. The returned fence pointer may be NULL in which
+ * case the device has already synchronized.
+ *
+ * Returns 0 on success, negative error code on failure. -ERESTARTSYS if
+ * interrupted.
+ */
+int vmw_kms_sou_do_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)
+{
+	struct vmw_framebuffer_surface *vfbs =
+		container_of(framebuffer, typeof(*vfbs), base);
+	struct vmw_kms_sou_surface_dirty sdirty;
+	int ret;
+
+	if (!srf)
+		srf = &vfbs->surface->res;
+
+	ret = vmw_kms_helper_resource_prepare(srf, true);
+	if (ret)
+		return ret;
+
+	sdirty.base.fifo_commit = vmw_sou_surface_fifo_commit;
+	sdirty.base.clip = vmw_sou_surface_clip;
+	sdirty.base.dev_priv = dev_priv;
+	sdirty.base.fifo_reserve_size = sizeof(struct vmw_kms_sou_dirty_cmd) +
+	  sizeof(SVGASignedRect) * num_clips;
 
-	ret = vmw_execbuf_process(file_priv, dev_priv, NULL, cmd,
-				  fifo_size, 0, 0, NULL, NULL);
+	sdirty.sid = srf->id;
+	sdirty.left = sdirty.top = S32_MAX;
+	sdirty.right = sdirty.bottom = S32_MIN;
+	sdirty.dst_x = dest_x;
+	sdirty.dst_y = dest_y;
 
-	kfree(cmd);
+	ret = vmw_kms_helper_dirty(dev_priv, framebuffer, clips, vclips,
+				   dest_x, dest_y, num_clips, inc,
+				   &sdirty.base);
+	vmw_kms_helper_resource_finish(srf, out_fence);
 
 	return ret;
 }
 
-int vmw_kms_sou_do_dmabuf_dirty(struct drm_file *file_priv,
-				struct vmw_private *dev_priv,
+/**
+ * vmw_sou_dmabuf_fifo_commit - Callback to submit a set of readback clips.
+ *
+ * @dirty: The closure structure.
+ *
+ * Commits a previously built command buffer of readback clips.
+ */
+static void vmw_sou_dmabuf_fifo_commit(struct vmw_kms_dirty *dirty)
+{
+	vmw_fifo_commit(dirty->dev_priv,
+			sizeof(struct vmw_kms_sou_dmabuf_blit) *
+			dirty->num_hits);
+}
+
+/**
+ * vmw_sou_dmabuf_clip - Callback to encode a readback cliprect.
+ *
+ * @dirty: The closure structure
+ *
+ * Encodes a BLIT_GMRFB_TO_SCREEN cliprect.
+ */
+static void vmw_sou_dmabuf_clip(struct vmw_kms_dirty *dirty)
+{
+	struct vmw_kms_sou_dmabuf_blit *blit = dirty->cmd;
+
+	blit += dirty->num_hits;
+	blit->header = SVGA_CMD_BLIT_GMRFB_TO_SCREEN;
+	blit->body.destScreenId = dirty->unit->unit;
+	blit->body.srcOrigin.x = dirty->fb_x;
+	blit->body.srcOrigin.y = dirty->fb_y;
+	blit->body.destRect.left = dirty->unit_x1;
+	blit->body.destRect.top = dirty->unit_y1;
+	blit->body.destRect.right = dirty->unit_x2;
+	blit->body.destRect.bottom = dirty->unit_y2;
+	dirty->num_hits++;
+}
+
+/**
+ * vmw_kms_do_dmabuf_dirty - Dirty part of a dma-buffer backed framebuffer
+ *
+ * @dev_priv: Pointer to the device private structure.
+ * @framebuffer: Pointer to the dma-buffer backed framebuffer.
+ * @clips: Array of clip rects.
+ * @num_clips: Number of clip rects in @clips.
+ * @increment: Increment to use when looping over @clips.
+ * @interruptible: Whether to perform waits interruptible if possible.
+ * @out_fence: If non-NULL, will return a ref-counted pointer to a
+ * struct vmw_fence_obj. The returned fence pointer may be NULL in which
+ * case the device has already synchronized.
+ *
+ * Returns 0 on success, negative error code on failure. -ERESTARTSYS if
+ * interrupted.
+ */
+int vmw_kms_sou_do_dmabuf_dirty(struct vmw_private *dev_priv,
 				struct vmw_framebuffer *framebuffer,
-				unsigned flags, unsigned color,
 				struct drm_clip_rect *clips,
 				unsigned num_clips, int increment,
+				bool interruptible,
 				struct vmw_fence_obj **out_fence)
 {
-	struct vmw_display_unit *units[VMWGFX_NUM_DISPLAY_UNITS];
-	struct drm_clip_rect *clips_ptr;
-	int i, k, num_units, ret;
-	struct drm_crtc *crtc;
-	size_t fifo_size;
+	struct vmw_dma_buffer *buf =
+		container_of(framebuffer, struct vmw_framebuffer_dmabuf,
+			     base)->buffer;
+	struct vmw_kms_dirty dirty;
+	int ret;
 
-	struct {
-		uint32_t header;
-		SVGAFifoCmdBlitGMRFBToScreen body;
-	} *blits;
+	ret = vmw_kms_helper_buffer_prepare(dev_priv, buf, interruptible,
+					    false);
+	if (ret)
+		return ret;
 
-	ret = do_dmabuf_define_gmrfb(file_priv, dev_priv, framebuffer);
+	ret = do_dmabuf_define_gmrfb(dev_priv, framebuffer);
 	if (unlikely(ret != 0))
-		return ret; /* define_gmrfb prints warnings */
+		goto out_revert;
 
-	fifo_size = sizeof(*blits) * num_clips;
-	blits = kmalloc(fifo_size, GFP_KERNEL);
-	if (unlikely(blits == NULL)) {
-		DRM_ERROR("Failed to allocate temporary cmd buffer.\n");
-		return -ENOMEM;
-	}
+	dirty.fifo_commit = vmw_sou_dmabuf_fifo_commit;
+	dirty.clip = vmw_sou_dmabuf_clip;
+	dirty.fifo_reserve_size = sizeof(struct vmw_kms_sou_dmabuf_blit) *
+		num_clips;
+	ret = vmw_kms_helper_dirty(dev_priv, framebuffer, clips, NULL,
+				   0, 0, num_clips, increment, &dirty);
+	vmw_kms_helper_buffer_finish(dev_priv, NULL, buf, out_fence, NULL);
 
-	num_units = 0;
-	list_for_each_entry(crtc, &dev_priv->dev->mode_config.crtc_list, head) {
-		if (crtc->primary->fb != &framebuffer->base)
-			continue;
-		units[num_units++] = vmw_crtc_to_du(crtc);
-	}
+	return ret;
 
-	for (k = 0; k < num_units; k++) {
-		struct vmw_display_unit *unit = units[k];
-		int hit_num = 0;
-
-		clips_ptr = clips;
-		for (i = 0; i < num_clips; i++, clips_ptr += increment) {
-			int clip_x1 = clips_ptr->x1 - unit->crtc.x;
-			int clip_y1 = clips_ptr->y1 - unit->crtc.y;
-			int clip_x2 = clips_ptr->x2 - unit->crtc.x;
-			int clip_y2 = clips_ptr->y2 - unit->crtc.y;
-			int move_x, move_y;
-
-			/* skip any crtcs that misses the clip region */
-			if (clip_x1 >= unit->crtc.mode.hdisplay ||
-			    clip_y1 >= unit->crtc.mode.vdisplay ||
-			    clip_x2 <= 0 || clip_y2 <= 0)
-				continue;
-
-			/* clip size to crtc size */
-			clip_x2 = min_t(int, clip_x2, unit->crtc.mode.hdisplay);
-			clip_y2 = min_t(int, clip_y2, unit->crtc.mode.vdisplay);
-
-			/* translate both src and dest to bring clip into screen */
-			move_x = min_t(int, clip_x1, 0);
-			move_y = min_t(int, clip_y1, 0);
-
-			/* actual translate done here */
-			blits[hit_num].header = SVGA_CMD_BLIT_GMRFB_TO_SCREEN;
-			blits[hit_num].body.destScreenId = unit->unit;
-			blits[hit_num].body.srcOrigin.x = clips_ptr->x1 - move_x;
-			blits[hit_num].body.srcOrigin.y = clips_ptr->y1 - move_y;
-			blits[hit_num].body.destRect.left = clip_x1 - move_x;
-			blits[hit_num].body.destRect.top = clip_y1 - move_y;
-			blits[hit_num].body.destRect.right = clip_x2;
-			blits[hit_num].body.destRect.bottom = clip_y2;
-			hit_num++;
-		}
-
-		/* no clips hit the crtc */
-		if (hit_num == 0)
-			continue;
-
-		/* only return the last fence */
-		if (out_fence && *out_fence)
-			vmw_fence_obj_unreference(out_fence);
-
-		fifo_size = sizeof(*blits) * hit_num;
-		ret = vmw_execbuf_process(file_priv, dev_priv, NULL, blits,
-					  fifo_size, 0, 0, NULL, out_fence);
+out_revert:
+	vmw_kms_helper_buffer_revert(buf);
 
-		if (unlikely(ret != 0))
-			break;
-	}
+	return ret;
+}
 
-	kfree(blits);
 
-	return ret;
+/**
+ * vmw_sou_readback_fifo_commit - Callback to submit a set of readback clips.
+ *
+ * @dirty: The closure structure.
+ *
+ * Commits a previously built command buffer of readback clips.
+ */
+static void vmw_sou_readback_fifo_commit(struct vmw_kms_dirty *dirty)
+{
+	vmw_fifo_commit(dirty->dev_priv,
+			sizeof(struct vmw_kms_sou_readback_blit) *
+			dirty->num_hits);
 }
 
+/**
+ * vmw_sou_readback_clip - Callback to encode a readback cliprect.
+ *
+ * @dirty: The closure structure
+ *
+ * Encodes a BLIT_SCREEN_TO_GMRFB cliprect.
+ */
+static void vmw_sou_readback_clip(struct vmw_kms_dirty *dirty)
+{
+	struct vmw_kms_sou_readback_blit *blit = dirty->cmd;
+
+	blit += dirty->num_hits;
+	blit->header = SVGA_CMD_BLIT_SCREEN_TO_GMRFB;
+	blit->body.srcScreenId = dirty->unit->unit;
+	blit->body.destOrigin.x = dirty->fb_x;
+	blit->body.destOrigin.y = dirty->fb_y;
+	blit->body.srcRect.left = dirty->unit_x1;
+	blit->body.srcRect.top = dirty->unit_y1;
+	blit->body.srcRect.right = dirty->unit_x2;
+	blit->body.srcRect.bottom = dirty->unit_y2;
+	dirty->num_hits++;
+}
+
+/**
+ * vmw_kms_sou_readback - Perform a readback from the screen object system to
+ * a dma-buffer backed framebuffer.
+ *
+ * @dev_priv: Pointer to the device private structure.
+ * @file_priv: Pointer to a struct drm_file identifying the caller.
+ * Must be set to NULL if @user_fence_rep is NULL.
+ * @vfb: Pointer to the dma-buffer backed framebuffer.
+ * @user_fence_rep: User-space provided structure for fence information.
+ * Must be set to non-NULL if @file_priv is non-NULL.
+ * @vclips: Array of clip rects.
+ * @num_clips: Number of clip rects in @vclips.
+ *
+ * Returns 0 on success, negative error code on failure. -ERESTARTSYS if
+ * interrupted.
+ */
+int vmw_kms_sou_readback(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_vmw_rect *vclips,
+			 uint32_t num_clips)
+{
+	struct vmw_dma_buffer *buf =
+		container_of(vfb, struct vmw_framebuffer_dmabuf, base)->buffer;
+	struct vmw_kms_dirty dirty;
+	int ret;
+
+	ret = vmw_kms_helper_buffer_prepare(dev_priv, buf, true, false);
+	if (ret)
+		return ret;
+
+	ret = do_dmabuf_define_gmrfb(dev_priv, vfb);
+	if (unlikely(ret != 0))
+		goto out_revert;
+
+	dirty.fifo_commit = vmw_sou_readback_fifo_commit;
+	dirty.clip = vmw_sou_readback_clip;
+	dirty.fifo_reserve_size = sizeof(struct vmw_kms_sou_readback_blit) *
+		num_clips;
+	ret = vmw_kms_helper_dirty(dev_priv, vfb, NULL, vclips,
+				   0, 0, num_clips, 1, &dirty);
+	vmw_kms_helper_buffer_finish(dev_priv, file_priv, buf, NULL,
+				     user_fence_rep);
+
+	return ret;
+
+out_revert:
+	vmw_kms_helper_buffer_revert(buf);
+
+	return ret;
+}
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 17/28] drm/vmwgfx: Convert screen targets to new helpers v3

[Thomas Hellstrom]

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>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.h      |   5 -
 drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c  |  24 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_kms.c      | 149 ++++-
 drivers/gpu/drm/vmwgfx/vmwgfx_kms.h      |  37 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_resource.c |   2 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c     | 996 ++++++++++++-------------------
 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 d60ae20..d6b247b 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 698a0e2..64dba53 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 5901d32..234a3ce 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 8a8203c..f941f92 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 521f194..69b471a 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 becf965..493fcd1 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 = sizeof(cmd->body) + blit_size;
+	vmw_bo_get_guest_ptr(&ddirty->buf->base, &cmd->body.guest.ptr);
+	cmd->body.guest.pitch = ddirty->pitch;
+	cmd->body.host.sid = stdu->display_srf->res.id;
+	cmd->body.host.face = 0;
+	cmd->body.host.mipmap = 0;
+	cmd->body.transfer = ddirty->transfer;
+	suffix->suffixSize = sizeof(*suffix);
+	suffix->maximumOffset = ddirty->buf->base.num_pages * PAGE_SIZE;
+
+	if (ddirty->transfer == SVGA3D_WRITE_HOST_VRAM) {
+		blit_size += sizeof(struct vmw_stdu_update);
+
+		vmw_stdu_populate_update(&suffix[1], stdu->base.unit,
+					 ddirty->left, ddirty->right,
+					 ddirty->top, ddirty->bottom);
+	}
+
+	vmw_fifo_commit(dirty->dev_priv, sizeof(*cmd) + blit_size);
+
+	ddirty->left = ddirty->top = S32_MAX;
+	ddirty->right = ddirty->bottom = S32_MIN;
+}
+
+/**
+ * vmw_kms_stdu_dma - Perform a DMA transfer between a dma-buffer backed
+ * framebuffer and the screen target system.
+ *
+ * @dev_priv: Pointer to the device private structure.
+ * @file_priv: Pointer to a struct drm-file identifying the caller. May be
+ * set to NULL, but then @user_fence_rep must also be set to NULL.
+ * @vfb: Pointer to the dma-buffer backed framebuffer.
+ * @clips: Array of clip rects. Either @clips or @vclips must be NULL.
+ * @vclips: Alternate array of clip rects. Either @clips or @vclips must
+ * be NULL.
+ * @num_clips: Number of clip rects in @clips or @vclips.
+ * @increment: Increment to use when looping over @clips or @vclips.
+ * @to_surface: Whether to DMA to the screen target system as opposed to
+ * from the screen target system.
+ * @interruptible: Whether to perform waits interruptible if possible.
+ *
+ * If DMA-ing till the screen target system, the function will also notify
+ * the screen target system that a bounding box of the cliprects has been
+ * updated.
+ * Returns 0 on success, negative error code on failure. -ERESTARTSYS if
+ * interrupted.
+ */
+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)
+{
+	struct vmw_dma_buffer *buf =
+		container_of(vfb, struct vmw_framebuffer_dmabuf, base)->buffer;
+	struct vmw_stdu_dirty ddirty;
+	int ret;
+
+	ret = vmw_kms_helper_buffer_prepare(dev_priv, buf, interruptible,
+					    false);
+	if (ret)
+		return ret;
+
+	ddirty.transfer = (to_surface) ? SVGA3D_WRITE_HOST_VRAM :
+		SVGA3D_READ_HOST_VRAM;
+	ddirty.left = ddirty.top = S32_MAX;
+	ddirty.right = ddirty.bottom = S32_MIN;
+	ddirty.pitch = vfb->base.pitches[0];
+	ddirty.buf = buf;
+	ddirty.base.fifo_commit = vmw_stdu_dmabuf_fifo_commit;
+	ddirty.base.clip = vmw_stdu_dmabuf_clip;
+	ddirty.base.fifo_reserve_size = sizeof(struct vmw_stdu_dma) +
+		num_clips * sizeof(SVGA3dCopyBox) +
+		sizeof(SVGA3dCmdSurfaceDMASuffix);
+	if (to_surface)
+		ddirty.base.fifo_reserve_size += sizeof(struct vmw_stdu_update);
+
+	ret = vmw_kms_helper_dirty(dev_priv, vfb, clips, vclips,
+				   0, 0, num_clips, increment, &ddirty.base);
+	vmw_kms_helper_buffer_finish(dev_priv, file_priv, buf, NULL,
+				     user_fence_rep);
+
+	return ret;
+}
+
+/**
+ * vmw_stdu_surface_clip - Callback to encode a surface copy command cliprect
+ *
+ * @dirty: The closure structure.
+ *
+ * Encodes a surface copy command cliprect and updates the bounding box
+ * for the copy.
+ */
+static void vmw_kms_stdu_surface_clip(struct vmw_kms_dirty *dirty)
+{
+	struct vmw_stdu_dirty *sdirty =
+		container_of(dirty, struct vmw_stdu_dirty, base);
+	struct vmw_stdu_surface_copy *cmd = dirty->cmd;
+	struct vmw_screen_target_display_unit *stdu =
+		container_of(dirty->unit, typeof(*stdu), base);
+
+	if (sdirty->sid != stdu->display_srf->res.id) {
+		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++;
+
+	/* Destination bounding box */
+	sdirty->left = min_t(s32, sdirty->left, dirty->unit_x1);
+	sdirty->top = min_t(s32, sdirty->top, dirty->unit_y1);
+	sdirty->right = max_t(s32, sdirty->right, dirty->unit_x2);
+	sdirty->bottom = max_t(s32, sdirty->bottom, dirty->unit_y2);
+}
+
+/**
+ * vmw_stdu_surface_fifo_commit - Callback to fill in and submit a surface
+ * copy command.
+ *
+ * @dirty: The closure structure.
+ *
+ * Fills in the missing fields in a surface copy command, and encodes a screen
+ * target update command.
+ */
+static void vmw_kms_stdu_surface_fifo_commit(struct vmw_kms_dirty *dirty)
+{
+	struct vmw_stdu_dirty *sdirty =
+		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_surface_copy *cmd = dirty->cmd;
+	struct vmw_stdu_update *update;
+	size_t blit_size = sizeof(SVGA3dCopyBox) * dirty->num_hits;
+	size_t commit_size;
+
+	if (!dirty->num_hits) {
+		vmw_fifo_commit(dirty->dev_priv, 0);
+		return;
+	}
+
+	if (sdirty->sid != stdu->display_srf->res.id) {
+		struct SVGA3dCopyBox *blit = (struct SVGA3dCopyBox *) &cmd[1];
+
+		cmd->header.id = SVGA_3D_CMD_SURFACE_COPY;
+		cmd->header.size = sizeof(cmd->body) + blit_size;
+		cmd->body.src.sid = sdirty->sid;
+		cmd->body.dest.sid = stdu->display_srf->res.id;
+		update = (struct vmw_stdu_update *) &blit[dirty->num_hits];
+		commit_size = sizeof(*cmd) + blit_size + sizeof(*update);
+	} else {
+		update = dirty->cmd;
+		commit_size = sizeof(*update);
+	}
+
+	vmw_stdu_populate_update(update, stdu->base.unit, sdirty->left,
+				 sdirty->right, sdirty->top, sdirty->bottom);
+
+	vmw_fifo_commit(dirty->dev_priv, commit_size);
+
+	sdirty->left = sdirty->top = S32_MAX;
+	sdirty->right = sdirty->bottom = S32_MIN;
+}
+
+/**
+ * vmw_kms_stdu_surface_dirty - Dirty part of a surface backed framebuffer
+ *
+ * @dev_priv: Pointer to the device private structure.
+ * @framebuffer: Pointer to the surface-buffer backed framebuffer.
+ * @clips: Array of clip rects. Either @clips or @vclips must be NULL.
+ * @vclips: Alternate array of clip rects. Either @clips or @vclips must
+ * be NULL.
+ * @srf: Pointer to surface to blit from. If NULL, the surface attached
+ * to @framebuffer will be used.
+ * @dest_x: X coordinate offset to align @srf with framebuffer coordinates.
+ * @dest_y: Y coordinate offset to align @srf with framebuffer coordinates.
+ * @num_clips: Number of clip rects in @clips.
+ * @inc: Increment to use when looping over @clips.
+ * @out_fence: If non-NULL, will return a ref-counted pointer to a
+ * struct vmw_fence_obj. The returned fence pointer may be NULL in which
+ * case the device has already synchronized.
+ *
+ * Returns 0 on success, negative error code on failure. -ERESTARTSYS if
+ * interrupted.
+ */
+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)
+{
+	struct vmw_framebuffer_surface *vfbs =
+		container_of(framebuffer, typeof(*vfbs), base);
+	struct vmw_stdu_dirty sdirty;
+	int ret;
+
+	if (!srf)
+		srf = &vfbs->surface->res;
+
+	ret = vmw_kms_helper_resource_prepare(srf, true);
+	if (ret)
+		return ret;
+
+	if (vfbs->is_dmabuf_proxy) {
+		ret = vmw_kms_update_proxy(srf, clips, num_clips, inc);
+		if (ret)
+			goto out_finish;
+	}
+
+	sdirty.base.fifo_commit = vmw_kms_stdu_surface_fifo_commit;
+	sdirty.base.clip = vmw_kms_stdu_surface_clip;
+	sdirty.base.fifo_reserve_size = sizeof(struct vmw_stdu_surface_copy) +
+		sizeof(SVGA3dCopyBox) * num_clips +
+		sizeof(struct vmw_stdu_update);
+	sdirty.sid = srf->id;
+	sdirty.left = sdirty.top = S32_MAX;
+	sdirty.right = sdirty.bottom = S32_MIN;
+
+	ret = vmw_kms_helper_dirty(dev_priv, framebuffer, clips, vclips,
+				   dest_x, dest_y, num_clips, inc,
+				   &sdirty.base);
+out_finish:
+	vmw_kms_helper_resource_finish(srf, out_fence);
+
+	return ret;
+}
+
 
 /*
  *  Screen Target CRTC dispatch table
@@ -1122,7 +1139,6 @@ static int vmw_stdu_init(struct vmw_private *dev_priv, unsigned unit)
 	stdu->base.pref_active = (unit == 0);
 	stdu->base.pref_width  = dev_priv->initial_width;
 	stdu->base.pref_height = dev_priv->initial_height;
-	stdu->base.pref_mode   = NULL;
 	stdu->base.is_implicit = true;
 
 	drm_connector_init(dev, connector, &vmw_stdu_connector_funcs,
@@ -1207,6 +1223,8 @@ int vmw_kms_stdu_init_display(struct vmw_private *dev_priv)
 	if (unlikely(ret != 0))
 		goto err_vblank_cleanup;
 
+	dev_priv->active_display_unit = vmw_du_screen_target;
+
 	for (i = 0; i < VMWGFX_NUM_DISPLAY_UNITS; ++i) {
 		ret = vmw_stdu_init(dev_priv, i);
 
@@ -1216,8 +1234,6 @@ int vmw_kms_stdu_init_display(struct vmw_private *dev_priv)
 		}
 	}
 
-	dev_priv->active_display_unit = vmw_du_screen_target;
-
 	DRM_INFO("Screen Target Display device initialized\n");
 
 	return 0;
@@ -1247,247 +1263,3 @@ int vmw_kms_stdu_close_display(struct vmw_private *dev_priv)
 
 	return 0;
 }
-
-
-
-/**
- * vmw_kms_stdu_do_surface_dirty - updates a dirty rectange to SVGA device
- *
- * @dev_priv: VMW DRM device
- * @file_priv: Pointer to a drm file private structure
- * @framebuffer: FB with the new content to be copied to SVGA device
- * @clip_rects: array of dirty rectanges
- * @num_of_clip_rects: number of rectanges in @clips
- * @increment: increment to the next dirty rect in @clips
- *
- * This function sends an Update command to the SVGA device.  This will notify
- * the device that a region needs to be copied to the screen.  At this time
- * we are not coalescing clip rects into one large clip rect because the SVGA
- * device will do it for us.
- *
- * RETURNS:
- * 0 on success, error code otherwise
- */
-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 *clip_rects,
-				  unsigned num_of_clip_rects, int increment)
-{
-	struct vmw_screen_target_display_unit *stdu[VMWGFX_NUM_DISPLAY_UNITS];
-	struct drm_clip_rect *cur_rect;
-	struct drm_crtc *crtc;
-
-	unsigned num_of_du = 0, cur_du, count = 0;
-	int      ret = 0;
-
-
-	BUG_ON(!clip_rects || !num_of_clip_rects);
-
-	/* Figure out all the DU affected by this surface */
-	list_for_each_entry(crtc, &dev_priv->dev->mode_config.crtc_list,
-			    head) {
-		if (crtc->primary->fb != &framebuffer->base)
-			continue;
-
-		stdu[num_of_du++] = vmw_crtc_to_stdu(crtc);
-	}
-
-	for (cur_du = 0; cur_du < num_of_du; cur_du++)
-		for (cur_rect = clip_rects, count = 0;
-		     count < num_of_clip_rects && ret == 0;
-		     cur_rect += increment, count++) {
-			ret = vmw_stdu_update_st(dev_priv, file_priv,
-						 stdu[cur_du],
-						 cur_rect);
-		}
-
-	return ret;
-}
-
-
-
-/**
- * vmw_kms_stdu_present - present a surface to the display surface
- *
- * @dev_priv: VMW DRM device
- * @file_priv: Pointer to a drm file private structure
- * @vfb: Used to pick which STDU(s) is affected
- * @user_handle: user handle for the source surface
- * @dest_x: top/left corner of the display area to blit to
- * @dest_y: top/left corner of the display area to blit to
- * @clip_rects: array of dirty rectanges
- * @num_of_clip_rects: number of rectanges in @clips
- *
- * This function copies a surface onto the display surface, and
- * updates the screen target.  Strech blit is currently not
- * supported.
- *
- * RETURNS:
- * 0 on success, error code otherwise
- */
-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 *clip_rects,
-			 uint32_t num_of_clip_rects)
-{
-	struct vmw_screen_target_display_unit *stdu[VMWGFX_NUM_DISPLAY_UNITS];
-	struct drm_clip_rect *update_area;
-	struct drm_crtc *crtc;
-	size_t fifo_size;
-	int num_of_du = 0, cur_du, i;
-	int ret = 0;
-	struct vmw_clip_rect src_bb;
-
-	struct {
-		SVGA3dCmdHeader      header;
-		SVGA3dCmdSurfaceCopy body;
-	} *cmd;
-	SVGA3dCopyBox *blits;
-
-
-	BUG_ON(!clip_rects || !num_of_clip_rects);
-
-	list_for_each_entry(crtc, &dev_priv->dev->mode_config.crtc_list, head) {
-		if (crtc->primary->fb != &vfb->base)
-			continue;
-
-		stdu[num_of_du++] = vmw_crtc_to_stdu(crtc);
-	}
-
-
-	update_area = kcalloc(num_of_clip_rects, sizeof(*update_area),
-			      GFP_KERNEL);
-	if (unlikely(update_area == NULL)) {
-		DRM_ERROR("Temporary clip rect memory alloc failed.\n");
-		return -ENOMEM;
-	}
-
-
-	fifo_size = sizeof(*cmd) + sizeof(SVGA3dCopyBox) * num_of_clip_rects;
-
-	cmd = kmalloc(fifo_size, GFP_KERNEL);
-	if (unlikely(cmd == NULL)) {
-		DRM_ERROR("Failed to allocate memory for surface copy.\n");
-		ret = -ENOMEM;
-		goto out_free_update_area;
-	}
-
-	memset(cmd, 0, fifo_size);
-	cmd->header.id = SVGA_3D_CMD_SURFACE_COPY;
-
-	blits = (SVGA3dCopyBox *)&cmd[1];
-
-
-	/* Figure out the source bounding box */
-	src_bb.x1 = clip_rects->x;
-	src_bb.y1 = clip_rects->y;
-	src_bb.x2 = clip_rects->x + clip_rects->w;
-	src_bb.y2 = clip_rects->y + clip_rects->h;
-
-	for (i = 1; i < num_of_clip_rects; i++) {
-		src_bb.x1 = min_t(int, src_bb.x1, clip_rects[i].x);
-		src_bb.x2 = max_t(int, src_bb.x2,
-				  clip_rects[i].x + (int) clip_rects[i].w);
-		src_bb.y1 = min_t(int, src_bb.y1, clip_rects[i].y);
-		src_bb.y2 = max_t(int, src_bb.y2,
-				  clip_rects[i].y + (int) clip_rects[i].h);
-	}
-
-	for (i = 0; i < num_of_clip_rects; i++) {
-		update_area[i].x1 = clip_rects[i].x - src_bb.x1;
-		update_area[i].x2 = update_area[i].x1 + clip_rects[i].w;
-		update_area[i].y1 = clip_rects[i].y - src_bb.y1;
-		update_area[i].y2 = update_area[i].y1 + clip_rects[i].h;
-	}
-
-
-	for (cur_du = 0; cur_du < num_of_du; cur_du++) {
-		struct vmw_clip_rect dest_bb;
-		int num_of_blits;
-
-		crtc = &stdu[cur_du]->base.crtc;
-
-		dest_bb.x1 = src_bb.x1 + dest_x - crtc->x;
-		dest_bb.y1 = src_bb.y1 + dest_y - crtc->y;
-		dest_bb.x2 = src_bb.x2 + dest_x - crtc->x;
-		dest_bb.y2 = src_bb.y2 + dest_y - crtc->y;
-
-		/* Skip any STDU outside of the destination bounding box */
-		if (dest_bb.x1 >= crtc->mode.hdisplay ||
-		    dest_bb.y1 >= crtc->mode.vdisplay ||
-		    dest_bb.x2 <= 0 || dest_bb.y2 <= 0)
-			continue;
-
-		/* Normalize to top-left of src bounding box in dest coord */
-		dest_bb.x2 = crtc->mode.hdisplay - dest_bb.x1;
-		dest_bb.y2 = crtc->mode.vdisplay - dest_bb.y1;
-		dest_bb.x1 = 0 - dest_bb.x1;
-		dest_bb.y1 = 0 - dest_bb.y1;
-
-		for (i = 0, num_of_blits = 0; i < num_of_clip_rects; i++) {
-			int x1 = max_t(int, dest_bb.x1, (int)update_area[i].x1);
-			int y1 = max_t(int, dest_bb.y1, (int)update_area[i].y1);
-			int x2 = min_t(int, dest_bb.x2, (int)update_area[i].x2);
-			int y2 = min_t(int, dest_bb.y2, (int)update_area[i].y2);
-
-			if (x1 >= x2)
-				continue;
-
-			if (y1 >= y2)
-				continue;
-
-			blits[num_of_blits].srcx =  src_bb.x1  + x1;
-			blits[num_of_blits].srcy =  src_bb.y1  + y1;
-			blits[num_of_blits].x    = -dest_bb.x1 + x1;
-			blits[num_of_blits].y    = -dest_bb.y1 + y1;
-			blits[num_of_blits].d    = 1;
-			blits[num_of_blits].w    = x2 - x1;
-			blits[num_of_blits].h    = y2 - y1;
-			num_of_blits++;
-		}
-
-		if (num_of_blits == 0)
-			continue;
-
-		/* Calculate new command size */
-		fifo_size = sizeof(*cmd) + sizeof(SVGA3dCopyBox) * num_of_blits;
-
-		cmd->header.size = cpu_to_le32(fifo_size - sizeof(cmd->header));
-
-		cmd->body.src.sid  = user_handle;
-		cmd->body.dest.sid = stdu[cur_du]->display_srf->res.id;
-
-		ret = vmw_execbuf_process(file_priv, dev_priv, NULL, cmd,
-					  fifo_size, 0, VMW_QUIRK_DST_SID_OK,
-					  NULL, NULL);
-
-		if (unlikely(ret != 0))
-			break;
-
-		for (i = 0; i < num_of_blits; i++) {
-			struct drm_clip_rect blit_area;
-
-			/*
-			 * Add crtc offset because vmw_stdu_update_st expects
-			 * desktop coordinates
-			 */
-			blit_area.x1 = blits[i].x + crtc->x;
-			blit_area.x2 = blit_area.x1 + blits[i].w;
-			blit_area.y1 = blits[i].y + crtc->y;
-			blit_area.y2 = blit_area.y1 + blits[i].h;
-			(void) vmw_stdu_update_st(dev_priv, NULL, stdu[cur_du],
-						  &blit_area);
-		}
-	}
-
-	kfree(cmd);
-
-out_free_update_area:
-	kfree(update_area);
-
-	return ret;
-}
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 18/28] drm/vmwgfx: Avoid cmdbuf alloc sleeping if !TASK_RUNNING

[Thomas Hellstrom]

If the command buffer pool is out of space, the code waits until space is
available. However since the condition code tries to allocate a range manager
node while !TASK_RUNNING we get a kernel warning.

Avoid this by pre-allocating the mm node. This will also probably be more
efficient.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c | 83 ++++++++++++++--------------------
 1 file changed, 34 insertions(+), 49 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c
index b044bf5..e94feb3 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c
@@ -33,7 +33,8 @@
  * multiple of the DMA pool allocation size.
  */
 #define VMW_CMDBUF_INLINE_ALIGN 64
-#define VMW_CMDBUF_INLINE_SIZE (1024 - VMW_CMDBUF_INLINE_ALIGN)
+#define VMW_CMDBUF_INLINE_SIZE \
+	(1024 - ALIGN(sizeof(SVGACBHeader), VMW_CMDBUF_INLINE_ALIGN))
 
 /**
  * struct vmw_cmdbuf_context - Command buffer context queues
@@ -145,7 +146,7 @@ struct vmw_cmdbuf_header {
 	SVGACBHeader *cb_header;
 	SVGACBContext cb_context;
 	struct list_head list;
-	struct drm_mm_node *node;
+	struct drm_mm_node node;
 	dma_addr_t handle;
 	u8 *cmd;
 	size_t size;
@@ -169,13 +170,13 @@ struct vmw_cmdbuf_dheader {
  * struct vmw_cmdbuf_alloc_info - Command buffer space allocation metadata
  *
  * @page_size: Size of requested command buffer space in pages.
- * @node: The range manager node if allocation succeeded.
- * @ret: Error code if failure. Otherwise 0.
+ * @node: Pointer to the range manager node.
+ * @done: True if this allocation has succeeded.
  */
 struct vmw_cmdbuf_alloc_info {
 	size_t page_size;
 	struct drm_mm_node *node;
-	int ret;
+	bool done;
 };
 
 /* Loop over each context in the command buffer manager. */
@@ -253,9 +254,7 @@ static void __vmw_cmdbuf_header_free(struct vmw_cmdbuf_header *header)
 		return;
 	}
 
-	drm_mm_remove_node(header->node);
-	kfree(header->node);
-	header->node = NULL;
+	drm_mm_remove_node(&header->node);
 	wake_up_all(&man->alloc_queue);
 	if (header->cb_header)
 		dma_pool_free(man->headers, header->cb_header,
@@ -669,32 +668,26 @@ static bool vmw_cmdbuf_try_alloc(struct vmw_cmdbuf_man *man,
 {
 	int ret;
 
-	if (info->node)
+	if (info->done)
 		return true;
-
-	info->node = kzalloc(sizeof(*info->node), GFP_KERNEL);
-	if (!info->node) {
-		info->ret = -ENOMEM;
-		return true;
-	}
-
+ 
+	memset(info->node, 0, sizeof(*info->node));
 	spin_lock_bh(&man->lock);
-	ret = drm_mm_insert_node_generic(&man->mm, info->node, info->page_size, 0, 0,
+	ret = drm_mm_insert_node_generic(&man->mm, info->node, info->page_size,
+					 0, 0,
 					 DRM_MM_SEARCH_DEFAULT,
 					 DRM_MM_CREATE_DEFAULT);
 	spin_unlock_bh(&man->lock);
-	if (ret) {
-		kfree(info->node);
-		info->node = NULL;
-	}
+	info->done = !ret;
 
-	return !!info->node;
+	return info->done;
 }
 
 /**
  * vmw_cmdbuf_alloc_space - Allocate buffer space from the main pool.
  *
  * @man: The command buffer manager.
+ * @node: Pointer to pre-allocated range-manager node.
  * @size: The size of the allocation.
  * @interruptible: Whether to sleep interruptible while waiting for space.
  *
@@ -702,15 +695,16 @@ static bool vmw_cmdbuf_try_alloc(struct vmw_cmdbuf_man *man,
  * no space available ATM, it turns on IRQ handling and sleeps waiting for it to
  * become available.
  */
-static struct drm_mm_node *vmw_cmdbuf_alloc_space(struct vmw_cmdbuf_man *man,
-						  size_t size,
-						  bool interruptible)
+int vmw_cmdbuf_alloc_space(struct vmw_cmdbuf_man *man,
+			   struct drm_mm_node *node,
+			   size_t size,
+			   bool interruptible)
 {
 	struct vmw_cmdbuf_alloc_info info;
 
 	info.page_size = PAGE_ALIGN(size) >> PAGE_SHIFT;
-	info.node = NULL;
-	info.ret = 0;
+	info.node = node;
+	info.done = false;
 
 	/*
 	 * To prevent starvation of large requests, only one allocating call
@@ -718,22 +712,14 @@ static struct drm_mm_node *vmw_cmdbuf_alloc_space(struct vmw_cmdbuf_man *man,
 	 */
 	if (interruptible) {
 		if (mutex_lock_interruptible(&man->space_mutex))
-			return ERR_PTR(-ERESTARTSYS);
+			return -ERESTARTSYS;
 	} else {
 		mutex_lock(&man->space_mutex);
 	}
 
 	/* Try to allocate space without waiting. */
-	(void) vmw_cmdbuf_try_alloc(man, &info);
-	if (info.ret && !info.node) {
-		mutex_unlock(&man->space_mutex);
-		return ERR_PTR(info.ret);
-	}
-
-	if (info.node) {
-		mutex_unlock(&man->space_mutex);
-		return info.node;
-	}
+	if (vmw_cmdbuf_try_alloc(man, &info))
+		goto out_unlock;
 
 	vmw_generic_waiter_add(man->dev_priv,
 			       SVGA_IRQFLAG_COMMAND_BUFFER,
@@ -749,7 +735,7 @@ static struct drm_mm_node *vmw_cmdbuf_alloc_space(struct vmw_cmdbuf_man *man,
 				(man->dev_priv, SVGA_IRQFLAG_COMMAND_BUFFER,
 				 &man->dev_priv->cmdbuf_waiters);
 			mutex_unlock(&man->space_mutex);
-			return ERR_PTR(ret);
+			return ret;
 		}
 	} else {
 		wait_event(man->alloc_queue, vmw_cmdbuf_try_alloc(man, &info));
@@ -757,11 +743,11 @@ static struct drm_mm_node *vmw_cmdbuf_alloc_space(struct vmw_cmdbuf_man *man,
 	vmw_generic_waiter_remove(man->dev_priv,
 				  SVGA_IRQFLAG_COMMAND_BUFFER,
 				  &man->dev_priv->cmdbuf_waiters);
+
+out_unlock:
 	mutex_unlock(&man->space_mutex);
-	if (info.ret && !info.node)
-		return ERR_PTR(info.ret);
 
-	return info.node;
+	return 0;
 }
 
 /**
@@ -785,10 +771,10 @@ static int vmw_cmdbuf_space_pool(struct vmw_cmdbuf_man *man,
 	if (!man->has_pool)
 		return -ENOMEM;
 
-	header->node = vmw_cmdbuf_alloc_space(man, size, interruptible);
+	ret = vmw_cmdbuf_alloc_space(man, &header->node,  size, interruptible);
 
-	if (IS_ERR(header->node))
-		return PTR_ERR(header->node);
+	if (ret)
+		return ret;
 
 	header->cb_header = dma_pool_alloc(man->headers, GFP_KERNEL,
 					   &header->handle);
@@ -797,9 +783,9 @@ static int vmw_cmdbuf_space_pool(struct vmw_cmdbuf_man *man,
 		goto out_no_cb_header;
 	}
 
-	header->size = header->node->size << PAGE_SHIFT;
+	header->size = header->node.size << PAGE_SHIFT;
 	cb_hdr = header->cb_header;
-	offset = header->node->start << PAGE_SHIFT;
+	offset = header->node.start << PAGE_SHIFT;
 	header->cmd = man->map + offset;
 	memset(cb_hdr, 0, sizeof(*cb_hdr));
 	if (man->using_mob) {
@@ -814,9 +800,8 @@ static int vmw_cmdbuf_space_pool(struct vmw_cmdbuf_man *man,
 
 out_no_cb_header:
 	spin_lock_bh(&man->lock);
-	drm_mm_remove_node(header->node);
+	drm_mm_remove_node(&header->node);
 	spin_unlock_bh(&man->lock);
-	kfree(header->node);
 
 	return ret;
 }
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 19/28] drm/vmwgfx: Add a kernel interface to create a framebuffer v2

[Thomas Hellstrom]

The kernel interface is needed for fbdev, and needs to be free from
a file_priv member. To accomplish this, remove the fb surface mutex
and list which isn't used anymore, anyway.

Finally, make the pin() and unpin() pin the framebuffer for all display
system backends, so that fbdev can pin its framebuffer before mapping it.

v2: Address review comments:
- Fix vmw_framebuffer_unpin() to handle also the surface framebuffer case.
- Fix vmw_kms_new_framebuffer() to actually use the only_2d parameter.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.c |   2 -
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.h |   2 -
 drivers/gpu/drm/vmwgfx/vmwgfx_kms.c | 162 +++++++++++++++++++++---------------
 drivers/gpu/drm/vmwgfx/vmwgfx_kms.h |   8 +-
 4 files changed, 102 insertions(+), 72 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
index e55db3f..bcf1962 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
@@ -1142,8 +1142,6 @@ static void vmw_lastclose(struct drm_device *dev)
 static void vmw_master_init(struct vmw_master *vmaster)
 {
 	ttm_lock_init(&vmaster->lock);
-	INIT_LIST_HEAD(&vmaster->fb_surf);
-	mutex_init(&vmaster->fb_surf_mutex);
 }
 
 static int vmw_master_create(struct drm_device *dev,
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index d6b247b..9ae5736 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -372,8 +372,6 @@ struct vmw_overlay;
 
 struct vmw_master {
 	struct ttm_lock lock;
-	struct mutex fb_surf_mutex;
-	struct list_head fb_surf;
 };
 
 struct vmw_vga_topology_state {
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
index 234a3ce..dc9f7d0 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
@@ -369,14 +369,7 @@ static void vmw_framebuffer_surface_destroy(struct drm_framebuffer *framebuffer)
 {
 	struct vmw_framebuffer_surface *vfbs =
 		vmw_framebuffer_to_vfbs(framebuffer);
-	struct vmw_master *vmaster = vmw_master(vfbs->master);
 
-
-	mutex_lock(&vmaster->fb_surf_mutex);
-	list_del(&vfbs->head);
-	mutex_unlock(&vmaster->fb_surf_mutex);
-
-	drm_master_put(&vfbs->master);
 	drm_framebuffer_cleanup(framebuffer);
 	vmw_surface_unreference(&vfbs->surface);
 	ttm_base_object_unref(&vfbs->base.user_obj);
@@ -396,9 +389,6 @@ static int vmw_framebuffer_surface_dirty(struct drm_framebuffer *framebuffer,
 	struct drm_clip_rect norect;
 	int ret, inc = 1;
 
-	if (unlikely(vfbs->master != file_priv->master))
-		return -EINVAL;
-
 	/* Legacy Display Unit does not support 3D */
 	if (dev_priv->active_display_unit == vmw_du_legacy)
 		return -EINVAL;
@@ -485,7 +475,6 @@ static struct drm_framebuffer_funcs vmw_framebuffer_surface_funcs = {
 };
 
 static int vmw_kms_new_framebuffer_surface(struct vmw_private *dev_priv,
-					   struct drm_file *file_priv,
 					   struct vmw_surface *surface,
 					   struct vmw_framebuffer **out,
 					   const struct drm_mode_fb_cmd
@@ -496,7 +485,6 @@ static int vmw_kms_new_framebuffer_surface(struct vmw_private *dev_priv,
 	struct drm_device *dev = dev_priv->dev;
 	struct vmw_framebuffer_surface *vfbs;
 	enum SVGA3dSurfaceFormat format;
-	struct vmw_master *vmaster = vmw_master(file_priv->master);
 	int ret;
 
 	/* 3D is only supported on HWv8 and newer hosts */
@@ -564,13 +552,8 @@ static int vmw_kms_new_framebuffer_surface(struct vmw_private *dev_priv,
 	vfbs->base.base.height = mode_cmd->height;
 	vfbs->surface = surface;
 	vfbs->base.user_handle = mode_cmd->handle;
-	vfbs->master = drm_master_get(file_priv->master);
 	vfbs->is_dmabuf_proxy = is_dmabuf_proxy;
 
-	mutex_lock(&vmaster->fb_surf_mutex);
-	list_add_tail(&vfbs->head, &vmaster->fb_surf);
-	mutex_unlock(&vmaster->fb_surf_mutex);
-
 	*out = &vfbs->base;
 
 	ret = drm_framebuffer_init(dev, &vfbs->base.base,
@@ -670,39 +653,51 @@ static struct drm_framebuffer_funcs vmw_framebuffer_dmabuf_funcs = {
 /**
  * Pin the dmabuffer to the start of vram.
  */
-static int vmw_framebuffer_dmabuf_pin(struct vmw_framebuffer *vfb)
+static int vmw_framebuffer_pin(struct vmw_framebuffer *vfb)
 {
 	struct vmw_private *dev_priv = vmw_priv(vfb->base.dev);
-	struct vmw_framebuffer_dmabuf *vfbd =
-		vmw_framebuffer_to_vfbd(&vfb->base);
+	struct vmw_dma_buffer *buf;
 	int ret;
 
-	/* This code should only be used with Legacy Display Unit */
-	BUG_ON(dev_priv->active_display_unit != vmw_du_legacy);
-
-	vmw_overlay_pause_all(dev_priv);
+	buf = vfb->dmabuf ?  vmw_framebuffer_to_vfbd(&vfb->base)->buffer :
+		vmw_framebuffer_to_vfbs(&vfb->base)->surface->res.backup;
 
-	ret = vmw_dmabuf_pin_in_start_of_vram(dev_priv, vfbd->buffer, false);
+	if (!buf)
+		return 0;
 
-	vmw_overlay_resume_all(dev_priv);
+	switch (dev_priv->active_display_unit) {
+	case vmw_du_legacy:
+		vmw_overlay_pause_all(dev_priv);
+		ret = vmw_dmabuf_pin_in_start_of_vram(dev_priv, buf, false);
+		vmw_overlay_resume_all(dev_priv);
+		break;
+	case vmw_du_screen_object:
+	case vmw_du_screen_target:
+		if (vfb->dmabuf)
+			return vmw_dmabuf_pin_in_vram_or_gmr(dev_priv, buf,
+							     false);
 
-	WARN_ON(ret != 0);
+		return vmw_dmabuf_pin_in_placement(dev_priv, buf,
+						   &vmw_mob_placement, false);
+	default:
+		return -EINVAL;
+	}
 
-	return 0;
+	return ret;
 }
 
-static int vmw_framebuffer_dmabuf_unpin(struct vmw_framebuffer *vfb)
+static int vmw_framebuffer_unpin(struct vmw_framebuffer *vfb)
 {
 	struct vmw_private *dev_priv = vmw_priv(vfb->base.dev);
-	struct vmw_framebuffer_dmabuf *vfbd =
-		vmw_framebuffer_to_vfbd(&vfb->base);
+	struct vmw_dma_buffer *buf;
 
-	if (!vfbd->buffer) {
-		WARN_ON(!vfbd->buffer);
+	buf = vfb->dmabuf ?  vmw_framebuffer_to_vfbd(&vfb->base)->buffer :
+		vmw_framebuffer_to_vfbs(&vfb->base)->surface->res.backup;
+
+	if (WARN_ON(!buf))
 		return 0;
-	}
 
-	return vmw_dmabuf_unpin(dev_priv, vfbd->buffer, false);
+	return vmw_dmabuf_unpin(dev_priv, buf, false);
 }
 
 /**
@@ -721,7 +716,7 @@ static int vmw_framebuffer_dmabuf_unpin(struct vmw_framebuffer *vfb)
  * 0 on success, error code otherwise
  */
 static int vmw_create_dmabuf_proxy(struct drm_device *dev,
-				   struct drm_mode_fb_cmd *mode_cmd,
+				   const struct drm_mode_fb_cmd *mode_cmd,
 				   struct vmw_dma_buffer *dmabuf_mob,
 				   struct vmw_surface **srf_out)
 {
@@ -847,10 +842,6 @@ static int vmw_kms_new_framebuffer_dmabuf(struct vmw_private *dev_priv,
 	vfbd->base.base.depth = mode_cmd->depth;
 	vfbd->base.base.width = mode_cmd->width;
 	vfbd->base.base.height = mode_cmd->height;
-	if (dev_priv->active_display_unit == vmw_du_legacy) {
-		vfbd->base.pin = vmw_framebuffer_dmabuf_pin;
-		vfbd->base.unpin = vmw_framebuffer_dmabuf_unpin;
-	}
 	vfbd->base.dmabuf = true;
 	vfbd->buffer = dmabuf;
 	vfbd->base.user_handle = mode_cmd->handle;
@@ -871,6 +862,64 @@ out_err1:
 	return ret;
 }
 
+/**
+ * vmw_kms_new_framebuffer - Create a new framebuffer.
+ *
+ * @dev_priv: Pointer to device private struct.
+ * @dmabuf: Pointer to dma buffer to wrap the kms framebuffer around.
+ * Either @dmabuf or @surface must be NULL.
+ * @surface: Pointer to a surface to wrap the kms framebuffer around.
+ * Either @dmabuf or @surface must be NULL.
+ * @only_2d: No presents will occur to this dma buffer based framebuffer. This
+ * Helps the code to do some important optimizations.
+ * @mode_cmd: Frame-buffer metadata.
+ */
+struct vmw_framebuffer *
+vmw_kms_new_framebuffer(struct vmw_private *dev_priv,
+			struct vmw_dma_buffer *dmabuf,
+			struct vmw_surface *surface,
+			bool only_2d,
+			const struct drm_mode_fb_cmd *mode_cmd)
+{
+	struct vmw_framebuffer *vfb;
+	bool is_dmabuf_proxy = false;
+	int ret;
+
+	/*
+	 * We cannot use the SurfaceDMA command in an non-accelerated VM,
+	 * therefore, wrap the DMA buf in a surface so we can use the
+	 * SurfaceCopy command.
+	 */
+	if (dmabuf && only_2d &&
+	    dev_priv->active_display_unit == vmw_du_screen_target) {
+		ret = vmw_create_dmabuf_proxy(dev_priv->dev, mode_cmd,
+					      dmabuf, &surface);
+		if (ret)
+			return ERR_PTR(ret);
+
+		is_dmabuf_proxy = true;
+	}
+
+	/* Create the new framebuffer depending one what we have */
+	if (surface)
+		ret = vmw_kms_new_framebuffer_surface(dev_priv, surface, &vfb,
+						      mode_cmd,
+						      is_dmabuf_proxy);
+	else if (dmabuf)
+		ret = vmw_kms_new_framebuffer_dmabuf(dev_priv, dmabuf, &vfb,
+						     mode_cmd);
+	else
+		BUG();
+
+	if (ret)
+		return ERR_PTR(ret);
+
+	vfb->pin = vmw_framebuffer_pin;
+	vfb->unpin = vmw_framebuffer_unpin;
+
+	return vfb;
+}
+
 /*
  * Generic Kernel modesetting functions
  */
@@ -886,7 +935,6 @@ static struct drm_framebuffer *vmw_kms_fb_create(struct drm_device *dev,
 	struct vmw_dma_buffer *bo = NULL;
 	struct ttm_base_object *user_obj;
 	struct drm_mode_fb_cmd mode_cmd;
-	bool is_dmabuf_proxy = false;
 	int ret;
 
 	mode_cmd.width = mode_cmd2->width;
@@ -935,31 +983,13 @@ static struct drm_framebuffer *vmw_kms_fb_create(struct drm_device *dev,
 	if (ret)
 		goto err_out;
 
-	/*
-	 * We cannot use the SurfaceDMA command in an non-accelerated VM,
-	 * therefore, wrap the DMA buf in a surface so we can use the
-	 * SurfaceCopy command.
-	 */
-	if (bo && !(dev_priv->capabilities & SVGA_CAP_3D) &&
-	    dev_priv->active_display_unit == vmw_du_screen_target) {
-		ret = vmw_create_dmabuf_proxy(dev_priv->dev, &mode_cmd, bo,
-			&surface);
-		if (ret)
-			goto err_out;
-
-		is_dmabuf_proxy = true;
-	}
-
-	/* Create the new framebuffer depending one what we have */
-	if (surface)
-		ret = vmw_kms_new_framebuffer_surface(dev_priv, file_priv,
-						      surface, &vfb, &mode_cmd,
-						      is_dmabuf_proxy);
-	else if (bo)
-		ret = vmw_kms_new_framebuffer_dmabuf(dev_priv, bo, &vfb,
-						     &mode_cmd);
-	else
-		BUG();
+	vfb = vmw_kms_new_framebuffer(dev_priv, bo, surface,
+				      !(dev_priv->capabilities & SVGA_CAP_3D),
+				      &mode_cmd);
+	if (IS_ERR(vfb)) {
+		ret = PTR_ERR(vfb);
+		goto err_out;
+ 	}
 
 err_out:
 	/* vmw_user_lookup_handle takes one ref so does new_fb */
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
index f941f92..311effc 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
@@ -114,7 +114,6 @@ struct vmw_framebuffer_surface {
 	struct vmw_surface *surface;
 	struct vmw_dma_buffer *buffer;
 	struct list_head head;
-	struct drm_master *master;
 	bool is_dmabuf_proxy;  /* true if this is proxy surface for DMA buf */
 };
 
@@ -238,7 +237,12 @@ int vmw_kms_readback(struct vmw_private *dev_priv,
 		     struct drm_vmw_fence_rep __user *user_fence_rep,
 		     struct drm_vmw_rect *vclips,
 		     uint32_t num_clips);
-
+struct vmw_framebuffer *
+vmw_kms_new_framebuffer(struct vmw_private *dev_priv,
+			struct vmw_dma_buffer *dmabuf,
+			struct vmw_surface *surface,
+			bool only_2d,
+			const struct drm_mode_fb_cmd *mode_cmd);
 
 /*
  * Legacy display unit functions - vmwgfx_ldu.c
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 20/28] drm/vmwgfx: Implement fbdev on kms v2

[Thomas Hellstrom]

With screen targets the old legacy display system fbdev doesn't work
satisfactory anymore. At best the resolution is severely restricted.
Therefore implement fbdev on top of the kms system. With this change, fbdev
will be using whatever KMS backend is chosen.

There are helpers available for this, so in the future we'd probably want
to implement the helper callbacks instead of calling into our KMS
implementation directly.

v2: Make sure we take the mode_config mutex around modesetting,
Also clear the initial framebuffer using vzalloc instead of vmalloc.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.c  |  22 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_fb.c   | 555 +++++++++++++++++++++++------------
 drivers/gpu/drm/vmwgfx/vmwgfx_kms.c  |  63 +++-
 drivers/gpu/drm/vmwgfx/vmwgfx_kms.h  |   8 +
 drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c  |   1 -
 drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c |   1 -
 6 files changed, 436 insertions(+), 214 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
index bcf1962..1892144 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
@@ -1120,23 +1120,6 @@ static long vmw_compat_ioctl(struct file *filp, unsigned int cmd,
 
 static void vmw_lastclose(struct drm_device *dev)
 {
-	struct drm_crtc *crtc;
-	struct drm_mode_set set;
-	int ret;
-
-	set.x = 0;
-	set.y = 0;
-	set.fb = NULL;
-	set.mode = NULL;
-	set.connectors = NULL;
-	set.num_connectors = 0;
-
-	list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
-		set.crtc = crtc;
-		ret = drm_mode_set_config_internal(&set);
-		WARN_ON(ret != 0);
-	}
-
 }
 
 static void vmw_master_init(struct vmw_master *vmaster)
@@ -1321,6 +1304,8 @@ static int vmwgfx_pm_notifier(struct notifier_block *nb, unsigned long val,
 
 	switch (val) {
 	case PM_HIBERNATION_PREPARE:
+		if (dev_priv->enable_fb)
+			vmw_fb_off(dev_priv);
 		ttm_suspend_lock(&dev_priv->reservation_sem);
 
 		/*
@@ -1337,7 +1322,8 @@ static int vmwgfx_pm_notifier(struct notifier_block *nb, unsigned long val,
 	case PM_POST_RESTORE:
 		vmw_fence_fifo_up(dev_priv->fman);
 		ttm_suspend_unlock(&dev_priv->reservation_sem);
-
+		if (dev_priv->enable_fb)
+			vmw_fb_on(dev_priv);
 		break;
 	case PM_RESTORE_PREPARE:
 		break;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
index b54d99bc..9dbb203 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
@@ -30,6 +30,7 @@
 
 #include <drm/drmP.h>
 #include "vmwgfx_drv.h"
+#include "vmwgfx_kms.h"
 
 #include <drm/ttm/ttm_placement.h>
 
@@ -40,21 +41,22 @@ struct vmw_fb_par {
 
 	void *vmalloc;
 
+	struct mutex bo_mutex;
 	struct vmw_dma_buffer *vmw_bo;
 	struct ttm_bo_kmap_obj map;
+	void *bo_ptr;
+	unsigned bo_size;
+	struct drm_framebuffer *set_fb;
+	struct drm_display_mode *set_mode;
+	u32 fb_x;
+	u32 fb_y;
+	bool bo_iowrite;
 
 	u32 pseudo_palette[17];
 
-	unsigned depth;
-	unsigned bpp;
-
 	unsigned max_width;
 	unsigned max_height;
 
-	void *bo_ptr;
-	unsigned bo_size;
-	bool bo_iowrite;
-
 	struct {
 		spinlock_t lock;
 		bool active;
@@ -63,6 +65,11 @@ struct vmw_fb_par {
 		unsigned x2;
 		unsigned y2;
 	} dirty;
+
+	struct drm_crtc *crtc;
+	struct drm_connector *con;
+
+	bool local_mode;
 };
 
 static int vmw_fb_setcolreg(unsigned regno, unsigned red, unsigned green,
@@ -77,7 +84,7 @@ static int vmw_fb_setcolreg(unsigned regno, unsigned red, unsigned green,
 		return 1;
 	}
 
-	switch (par->depth) {
+	switch (par->set_fb->depth) {
 	case 24:
 	case 32:
 		pal[regno] = ((red & 0xff00) << 8) |
@@ -85,7 +92,8 @@ static int vmw_fb_setcolreg(unsigned regno, unsigned red, unsigned green,
 			     ((blue  & 0xff00) >> 8);
 		break;
 	default:
-		DRM_ERROR("Bad depth %u, bpp %u.\n", par->depth, par->bpp);
+		DRM_ERROR("Bad depth %u, bpp %u.\n", par->set_fb->depth,
+			  par->set_fb->bits_per_pixel);
 		return 1;
 	}
 
@@ -134,12 +142,6 @@ static int vmw_fb_check_var(struct fb_var_screeninfo *var,
 		return -EINVAL;
 	}
 
-	if (!(vmw_priv->capabilities & SVGA_CAP_DISPLAY_TOPOLOGY) &&
-	    (var->xoffset != 0 || var->yoffset != 0)) {
-		DRM_ERROR("Can not handle panning without display topology\n");
-		return -EINVAL;
-	}
-
 	if ((var->xoffset + var->xres) > par->max_width ||
 	    (var->yoffset + var->yres) > par->max_height) {
 		DRM_ERROR("Requested geom can not fit in framebuffer\n");
@@ -156,46 +158,6 @@ static int vmw_fb_check_var(struct fb_var_screeninfo *var,
 	return 0;
 }
 
-static int vmw_fb_set_par(struct fb_info *info)
-{
-	struct vmw_fb_par *par = info->par;
-	struct vmw_private *vmw_priv = par->vmw_priv;
-	int ret;
-
-	info->fix.line_length = info->var.xres * info->var.bits_per_pixel/8;
-
-	ret = vmw_kms_write_svga(vmw_priv, info->var.xres, info->var.yres,
-				 info->fix.line_length,
-				 par->bpp, par->depth);
-	if (ret)
-		return ret;
-
-	if (vmw_priv->capabilities & SVGA_CAP_DISPLAY_TOPOLOGY) {
-		/* TODO check if pitch and offset changes */
-		vmw_write(vmw_priv, SVGA_REG_NUM_GUEST_DISPLAYS, 1);
-		vmw_write(vmw_priv, SVGA_REG_DISPLAY_ID, 0);
-		vmw_write(vmw_priv, SVGA_REG_DISPLAY_IS_PRIMARY, true);
-		vmw_write(vmw_priv, SVGA_REG_DISPLAY_POSITION_X, info->var.xoffset);
-		vmw_write(vmw_priv, SVGA_REG_DISPLAY_POSITION_Y, info->var.yoffset);
-		vmw_write(vmw_priv, SVGA_REG_DISPLAY_WIDTH, info->var.xres);
-		vmw_write(vmw_priv, SVGA_REG_DISPLAY_HEIGHT, info->var.yres);
-		vmw_write(vmw_priv, SVGA_REG_DISPLAY_ID, SVGA_ID_INVALID);
-	}
-
-	/* This is really helpful since if this fails the user
-	 * can probably not see anything on the screen.
-	 */
-	WARN_ON(vmw_read(vmw_priv, SVGA_REG_FB_OFFSET) != 0);
-
-	return 0;
-}
-
-static int vmw_fb_pan_display(struct fb_var_screeninfo *var,
-			      struct fb_info *info)
-{
-	return 0;
-}
-
 static int vmw_fb_blank(int blank, struct fb_info *info)
 {
 	return 0;
@@ -209,55 +171,77 @@ static void vmw_fb_dirty_flush(struct vmw_fb_par *par)
 {
 	struct vmw_private *vmw_priv = par->vmw_priv;
 	struct fb_info *info = vmw_priv->fb_info;
-	int stride = (info->fix.line_length / 4);
-	int *src = (int *)info->screen_base;
-	__le32 __iomem *vram_mem = par->bo_ptr;
-	unsigned long flags;
-	unsigned x, y, w, h;
-	int i, k;
-	struct {
-		uint32_t header;
-		SVGAFifoCmdUpdate body;
-	} *cmd;
+	unsigned long irq_flags;
+	s32 dst_x1, dst_x2, dst_y1, dst_y2, w, h;
+	u32 cpp, max_x, max_y;
+	struct drm_clip_rect clip;
+	struct drm_framebuffer *cur_fb;
+	u8 *src_ptr, *dst_ptr;
 
 	if (vmw_priv->suspended)
 		return;
 
-	spin_lock_irqsave(&par->dirty.lock, flags);
-	if (!par->dirty.active) {
-		spin_unlock_irqrestore(&par->dirty.lock, flags);
-		return;
-	}
-	x = par->dirty.x1;
-	y = par->dirty.y1;
-	w = min(par->dirty.x2, info->var.xres) - x;
-	h = min(par->dirty.y2, info->var.yres) - y;
-	par->dirty.x1 = par->dirty.x2 = 0;
-	par->dirty.y1 = par->dirty.y2 = 0;
-	spin_unlock_irqrestore(&par->dirty.lock, flags);
+	mutex_lock(&par->bo_mutex);
+	cur_fb = par->set_fb;
+	if (!cur_fb)
+		goto out_unlock;
 
-	for (i = y * stride; i < info->fix.smem_len / 4; i += stride) {
-		for (k = i+x; k < i+x+w && k < info->fix.smem_len / 4; k++)
-			iowrite32(src[k], vram_mem + k);
+	spin_lock_irqsave(&par->dirty.lock, irq_flags);
+	if (!par->dirty.active) {
+		spin_unlock_irqrestore(&par->dirty.lock, irq_flags);
+		goto out_unlock;
 	}
 
-#if 0
-	DRM_INFO("%s, (%u, %u) (%ux%u)\n", __func__, x, y, w, h);
-#endif
+	/*
+	 * Handle panning when copying from vmalloc to framebuffer.
+	 * Clip dirty area to framebuffer.
+	 */
+	cpp = (cur_fb->bits_per_pixel + 7) / 8;
+	max_x = par->fb_x + cur_fb->width;
+	max_y = par->fb_y + cur_fb->height;
+
+	dst_x1 = par->dirty.x1 - par->fb_x;
+	dst_y1 = par->dirty.y1 - par->fb_y;
+	dst_x1 = max_t(s32, dst_x1, 0);
+	dst_y1 = max_t(s32, dst_y1, 0);
+
+	dst_x2 = par->dirty.x2 - par->fb_x;
+	dst_y2 = par->dirty.y2 - par->fb_y;
+	dst_x2 = min_t(s32, dst_x2, max_x);
+	dst_y2 = min_t(s32, dst_y2, max_y);
+	w = dst_x2 - dst_x1;
+	h = dst_y2 - dst_y1;
+	w = max_t(s32, 0, w);
+	h = max_t(s32, 0, h);
 
-	cmd = vmw_fifo_reserve(vmw_priv, sizeof(*cmd));
-	if (unlikely(cmd == NULL)) {
-		DRM_ERROR("Fifo reserve failed.\n");
-		return;
+	par->dirty.x1 = par->dirty.x2 = 0;
+	par->dirty.y1 = par->dirty.y2 = 0;
+	spin_unlock_irqrestore(&par->dirty.lock, irq_flags);
+
+	if (w && h) {
+		dst_ptr = (u8 *)par->bo_ptr  +
+			(dst_y1 * par->set_fb->pitches[0] + dst_x1 * cpp);
+		src_ptr = (u8 *)par->vmalloc +
+			((dst_y1 + par->fb_y) * info->fix.line_length +
+			 (dst_x1 + par->fb_x) * cpp);
+
+		while (h-- > 0) {
+			memcpy(dst_ptr, src_ptr, w*cpp);
+			dst_ptr += par->set_fb->pitches[0];
+			src_ptr += info->fix.line_length;
+		}
+
+		clip.x1 = dst_x1;
+		clip.x2 = dst_x2;
+		clip.y1 = dst_y1;
+		clip.y2 = dst_y2;
+
+		WARN_ON_ONCE(par->set_fb->funcs->dirty(cur_fb, NULL, 0, 0,
+						       &clip, 1));
+		vmw_fifo_flush(vmw_priv, false);
 	}
-
-	cmd->header = cpu_to_le32(SVGA_CMD_UPDATE);
-	cmd->body.x = cpu_to_le32(x);
-	cmd->body.y = cpu_to_le32(y);
-	cmd->body.width = cpu_to_le32(w);
-	cmd->body.height = cpu_to_le32(h);
-	vmw_fifo_commit(vmw_priv, sizeof(*cmd));
-	vmw_fifo_flush(vmw_priv, false);
+out_unlock:
+	mutex_unlock(&par->bo_mutex);
 }
 
 static void vmw_fb_dirty_mark(struct vmw_fb_par *par,
@@ -292,6 +276,28 @@ static void vmw_fb_dirty_mark(struct vmw_fb_par *par,
 	spin_unlock_irqrestore(&par->dirty.lock, flags);
 }
 
+static int vmw_fb_pan_display(struct fb_var_screeninfo *var,
+			      struct fb_info *info)
+{
+	struct vmw_fb_par *par = info->par;
+
+	if ((var->xoffset + var->xres) > var->xres_virtual ||
+	    (var->yoffset + var->yres) > var->yres_virtual) {
+		DRM_ERROR("Requested panning can not fit in framebuffer\n");
+		return -EINVAL;
+	}
+
+	mutex_lock(&par->bo_mutex);
+	par->fb_x = var->xoffset;
+	par->fb_y = var->yoffset;
+	if (par->set_fb)
+		vmw_fb_dirty_mark(par, par->fb_x, par->fb_y, par->set_fb->width,
+				  par->set_fb->height);
+	mutex_unlock(&par->bo_mutex);
+
+	return 0;
+}
+
 static void vmw_deferred_io(struct fb_info *info,
 			    struct list_head *pagelist)
 {
@@ -359,33 +365,12 @@ static void vmw_fb_imageblit(struct fb_info *info, const struct fb_image *image)
  * Bring up code
  */
 
-static struct fb_ops vmw_fb_ops = {
-	.owner = THIS_MODULE,
-	.fb_check_var = vmw_fb_check_var,
-	.fb_set_par = vmw_fb_set_par,
-	.fb_setcolreg = vmw_fb_setcolreg,
-	.fb_fillrect = vmw_fb_fillrect,
-	.fb_copyarea = vmw_fb_copyarea,
-	.fb_imageblit = vmw_fb_imageblit,
-	.fb_pan_display = vmw_fb_pan_display,
-	.fb_blank = vmw_fb_blank,
-};
-
 static int vmw_fb_create_bo(struct vmw_private *vmw_priv,
 			    size_t size, struct vmw_dma_buffer **out)
 {
 	struct vmw_dma_buffer *vmw_bo;
-	struct ttm_place ne_place = vmw_vram_ne_placement.placement[0];
-	struct ttm_placement ne_placement;
 	int ret;
 
-	ne_placement.num_placement = 1;
-	ne_placement.placement = &ne_place;
-	ne_placement.num_busy_placement = 1;
-	ne_placement.busy_placement = &ne_place;
-
-	ne_place.lpfn = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
-
 	(void) ttm_write_lock(&vmw_priv->reservation_sem, false);
 
 	vmw_bo = kmalloc(sizeof(*vmw_bo), GFP_KERNEL);
@@ -395,14 +380,13 @@ static int vmw_fb_create_bo(struct vmw_private *vmw_priv,
 	}
 
 	ret = vmw_dmabuf_init(vmw_priv, vmw_bo, size,
-			      &ne_placement,
+			      &vmw_sys_placement,
 			      false,
 			      &vmw_dmabuf_bo_free);
 	if (unlikely(ret != 0))
 		goto err_unlock; /* init frees the buffer on failure */
 
 	*out = vmw_bo;
-
 	ttm_write_unlock(&vmw_priv->reservation_sem);
 
 	return 0;
@@ -412,14 +396,249 @@ err_unlock:
 	return ret;
 }
 
+static int vmw_fb_compute_depth(struct fb_var_screeninfo *var,
+				int *depth)
+{
+	switch (var->bits_per_pixel) {
+	case 32:
+		*depth = (var->transp.length > 0) ? 32 : 24;
+		break;
+	default:
+		DRM_ERROR("Bad bpp %u.\n", var->bits_per_pixel);
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int vmw_fb_kms_detach(struct vmw_fb_par *par,
+			     bool detach_bo,
+			     bool unref_bo)
+{
+	struct drm_framebuffer *cur_fb = par->set_fb;
+	int ret;
+
+	/* Detach the KMS framebuffer from crtcs */
+	if (par->set_mode) {
+		struct drm_mode_set set;
+
+		set.crtc = par->crtc;
+		set.x = 0;
+		set.y = 0;
+		set.mode = NULL;
+		set.fb = NULL;
+		set.num_connectors = 1;
+		set.connectors = &par->con;
+		ret = drm_mode_set_config_internal(&set);
+		if (ret) {
+			DRM_ERROR("Could not unset a mode.\n");
+			return ret;
+		}
+		drm_mode_destroy(par->vmw_priv->dev, par->set_mode);
+		par->set_mode = NULL;
+	}
+
+	if (cur_fb) {
+		drm_framebuffer_unreference(cur_fb);
+		par->set_fb = NULL;
+	}
+
+	if (par->vmw_bo && detach_bo) {
+		if (par->bo_ptr) {
+			ttm_bo_kunmap(&par->map);
+			par->bo_ptr = NULL;
+		}
+		if (unref_bo)
+			vmw_dmabuf_unreference(&par->vmw_bo);
+		else
+			vmw_dmabuf_unpin(par->vmw_priv, par->vmw_bo, false);
+	}
+
+	return 0;
+}
+
+static int vmw_fb_kms_framebuffer(struct fb_info *info)
+{
+	struct drm_mode_fb_cmd mode_cmd;
+	struct vmw_fb_par *par = info->par;
+	struct fb_var_screeninfo *var = &info->var;
+	struct drm_framebuffer *cur_fb;
+	struct vmw_framebuffer *vfb;
+	int ret = 0;
+	size_t new_bo_size;
+
+	ret = vmw_fb_compute_depth(var, &mode_cmd.depth);
+	if (ret)
+		return ret;
+
+	mode_cmd.width = var->xres;
+	mode_cmd.height = var->yres;
+	mode_cmd.bpp = var->bits_per_pixel;
+	mode_cmd.pitch = ((mode_cmd.bpp + 7) / 8) * mode_cmd.width;
+
+	cur_fb = par->set_fb;
+	if (cur_fb && cur_fb->width == mode_cmd.width &&
+	    cur_fb->height == mode_cmd.height &&
+	    cur_fb->bits_per_pixel == mode_cmd.bpp &&
+	    cur_fb->depth == mode_cmd.depth &&
+	    cur_fb->pitches[0] == mode_cmd.pitch)
+		return 0;
+
+	/* Need new buffer object ? */
+	new_bo_size = (size_t) mode_cmd.pitch * (size_t) mode_cmd.height;
+	ret = vmw_fb_kms_detach(par,
+				par->bo_size < new_bo_size ||
+				par->bo_size > 2*new_bo_size,
+				true);
+	if (ret)
+		return ret;
+
+	if (!par->vmw_bo) {
+		ret = vmw_fb_create_bo(par->vmw_priv, new_bo_size,
+				       &par->vmw_bo);
+		if (ret) {
+			DRM_ERROR("Failed creating a buffer object for "
+				  "fbdev.\n");
+			return ret;
+		}
+		par->bo_size = new_bo_size;
+	}
+
+	vfb = vmw_kms_new_framebuffer(par->vmw_priv, par->vmw_bo, NULL,
+				      true, &mode_cmd);
+	if (IS_ERR(vfb))
+		return PTR_ERR(vfb);
+
+	par->set_fb = &vfb->base;
+
+	if (!par->bo_ptr) {
+		/*
+		 * Pin before mapping. Since we don't know in what placement
+		 * to pin, call into KMS to do it for us.
+		 */
+		ret = vfb->pin(vfb);
+		if (ret) {
+			DRM_ERROR("Could not pin the fbdev framebuffer.\n");
+			return ret;
+		}
+
+		ret = ttm_bo_kmap(&par->vmw_bo->base, 0,
+				  par->vmw_bo->base.num_pages, &par->map);
+		if (ret) {
+			vfb->unpin(vfb);
+			DRM_ERROR("Could not map the fbdev framebuffer.\n");
+			return ret;
+		}
+
+		par->bo_ptr = ttm_kmap_obj_virtual(&par->map, &par->bo_iowrite);
+	}
+
+	return 0;
+}
+
+static int vmw_fb_set_par(struct fb_info *info)
+{
+	struct vmw_fb_par *par = info->par;
+	struct vmw_private *vmw_priv = par->vmw_priv;
+	struct drm_mode_set set;
+	struct fb_var_screeninfo *var = &info->var;
+	struct drm_display_mode new_mode = { DRM_MODE("fb_mode",
+		DRM_MODE_TYPE_DRIVER,
+		0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+		DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_PVSYNC)
+	};
+	struct drm_display_mode *old_mode;
+	struct drm_display_mode *mode;
+	int ret;
+
+	old_mode = par->set_mode;
+	mode = drm_mode_duplicate(vmw_priv->dev, &new_mode);
+	if (!mode) {
+		DRM_ERROR("Could not create new fb mode.\n");
+		return -ENOMEM;
+	}
+
+	mode->hdisplay = var->xres;
+	mode->vdisplay = var->yres;
+	vmw_guess_mode_timing(mode);
+
+	if (old_mode && drm_mode_equal(old_mode, mode)) {
+		drm_mode_destroy(vmw_priv->dev, mode);
+		mode = old_mode;
+		old_mode = NULL;
+	} else if (!vmw_kms_validate_mode_vram(vmw_priv,
+					       mode->hdisplay *
+					       (var->bits_per_pixel + 7) / 8,
+					       mode->vdisplay)) {
+		drm_mode_destroy(vmw_priv->dev, mode);
+		return -EINVAL;
+	}
+
+	mutex_lock(&par->bo_mutex);
+	drm_modeset_lock_all(vmw_priv->dev);
+	ret = vmw_fb_kms_framebuffer(info);
+	if (ret)
+		goto out_unlock;
+
+	par->fb_x = var->xoffset;
+	par->fb_y = var->yoffset;
+
+	set.crtc = par->crtc;
+	set.x = 0;
+	set.y = 0;
+	set.mode = mode;
+	set.fb = par->set_fb;
+	set.num_connectors = 1;
+	set.connectors = &par->con;
+
+	ret = drm_mode_set_config_internal(&set);
+	if (ret)
+		goto out_unlock;
+
+	vmw_fb_dirty_mark(par, par->fb_x, par->fb_y,
+			  par->set_fb->width, par->set_fb->height);
+
+	/* If there already was stuff dirty we wont
+	 * schedule a new work, so lets do it now */
+
+#if (defined(VMWGFX_STANDALONE) && defined(VMWGFX_FB_DEFERRED))
+	schedule_delayed_work(&par->def_par.deferred_work, 0);
+#else
+	schedule_delayed_work(&info->deferred_work, 0);
+#endif
+
+out_unlock:
+	if (old_mode)
+		drm_mode_destroy(vmw_priv->dev, old_mode);
+	par->set_mode = mode;
+
+	drm_modeset_unlock_all(vmw_priv->dev);
+	mutex_unlock(&par->bo_mutex);
+
+	return ret;
+}
+
+
+static struct fb_ops vmw_fb_ops = {
+	.owner = THIS_MODULE,
+	.fb_check_var = vmw_fb_check_var,
+	.fb_set_par = vmw_fb_set_par,
+	.fb_setcolreg = vmw_fb_setcolreg,
+	.fb_fillrect = vmw_fb_fillrect,
+	.fb_copyarea = vmw_fb_copyarea,
+	.fb_imageblit = vmw_fb_imageblit,
+	.fb_pan_display = vmw_fb_pan_display,
+	.fb_blank = vmw_fb_blank,
+};
+
 int vmw_fb_init(struct vmw_private *vmw_priv)
 {
 	struct device *device = &vmw_priv->dev->pdev->dev;
 	struct vmw_fb_par *par;
 	struct fb_info *info;
-	unsigned initial_width, initial_height;
 	unsigned fb_width, fb_height;
 	unsigned fb_bpp, fb_depth, fb_offset, fb_pitch, fb_size;
+	struct drm_display_mode *init_mode;
 	int ret;
 
 	fb_bpp = 32;
@@ -429,9 +648,6 @@ int vmw_fb_init(struct vmw_private *vmw_priv)
 	fb_width = min(vmw_priv->fb_max_width, (unsigned)2048);
 	fb_height = min(vmw_priv->fb_max_height, (unsigned)2048);
 
-	initial_width = min(vmw_priv->initial_width, fb_width);
-	initial_height = min(vmw_priv->initial_height, fb_height);
-
 	fb_pitch = fb_width * fb_bpp / 8;
 	fb_size = fb_pitch * fb_height;
 	fb_offset = vmw_read(vmw_priv, SVGA_REG_FB_OFFSET);
@@ -445,35 +661,34 @@ int vmw_fb_init(struct vmw_private *vmw_priv)
 	 */
 	vmw_priv->fb_info = info;
 	par = info->par;
+	memset(par, 0, sizeof(*par));
 	par->vmw_priv = vmw_priv;
-	par->depth = fb_depth;
-	par->bpp = fb_bpp;
 	par->vmalloc = NULL;
 	par->max_width = fb_width;
 	par->max_height = fb_height;
 
+	drm_modeset_lock_all(vmw_priv->dev);
+	ret = vmw_kms_fbdev_init_data(vmw_priv, 0, par->max_width,
+				      par->max_height, &par->con,
+				      &par->crtc, &init_mode);
+	if (ret) {
+		drm_modeset_unlock_all(vmw_priv->dev);
+		goto err_kms;
+	}
+
+	info->var.xres = init_mode->hdisplay;
+	info->var.yres = init_mode->vdisplay;
+	drm_modeset_unlock_all(vmw_priv->dev);
+
 	/*
 	 * Create buffers and alloc memory
 	 */
-	par->vmalloc = vmalloc(fb_size);
+	par->vmalloc = vzalloc(fb_size);
 	if (unlikely(par->vmalloc == NULL)) {
 		ret = -ENOMEM;
 		goto err_free;
 	}
 
-	ret = vmw_fb_create_bo(vmw_priv, fb_size, &par->vmw_bo);
-	if (unlikely(ret != 0))
-		goto err_free;
-
-	ret = ttm_bo_kmap(&par->vmw_bo->base,
-			  0,
-			  par->vmw_bo->base.num_pages,
-			  &par->map);
-	if (unlikely(ret != 0))
-		goto err_unref;
-	par->bo_ptr = ttm_kmap_obj_virtual(&par->map, &par->bo_iowrite);
-	par->bo_size = fb_size;
-
 	/*
 	 * Fixed and var
 	 */
@@ -509,18 +724,14 @@ int vmw_fb_init(struct vmw_private *vmw_priv)
 
 	info->var.xres_virtual = fb_width;
 	info->var.yres_virtual = fb_height;
-	info->var.bits_per_pixel = par->bpp;
+	info->var.bits_per_pixel = fb_bpp;
 	info->var.xoffset = 0;
 	info->var.yoffset = 0;
 	info->var.activate = FB_ACTIVATE_NOW;
 	info->var.height = -1;
 	info->var.width = -1;
 
-	info->var.xres = initial_width;
-	info->var.yres = initial_height;
-
 	/* Use default scratch pixmap (info->pixmap.flags = FB_PIXMAP_SYSTEM) */
-
 	info->apertures = alloc_apertures(1);
 	if (!info->apertures) {
 		ret = -ENOMEM;
@@ -536,6 +747,7 @@ int vmw_fb_init(struct vmw_private *vmw_priv)
 	par->dirty.y1 = par->dirty.y2 = 0;
 	par->dirty.active = true;
 	spin_lock_init(&par->dirty.lock);
+	mutex_init(&par->bo_mutex);
 	info->fbdefio = &vmw_defio;
 	fb_deferred_io_init(info);
 
@@ -543,16 +755,16 @@ int vmw_fb_init(struct vmw_private *vmw_priv)
 	if (unlikely(ret != 0))
 		goto err_defio;
 
+	vmw_fb_set_par(info);
+
 	return 0;
 
 err_defio:
 	fb_deferred_io_cleanup(info);
 err_aper:
-	ttm_bo_kunmap(&par->map);
-err_unref:
-	ttm_bo_unref((struct ttm_buffer_object **)&par->vmw_bo);
 err_free:
 	vfree(par->vmalloc);
+err_kms:
 	framebuffer_release(info);
 	vmw_priv->fb_info = NULL;
 
@@ -563,22 +775,18 @@ int vmw_fb_close(struct vmw_private *vmw_priv)
 {
 	struct fb_info *info;
 	struct vmw_fb_par *par;
-	struct ttm_buffer_object *bo;
 
 	if (!vmw_priv->fb_info)
 		return 0;
 
 	info = vmw_priv->fb_info;
 	par = info->par;
-	bo = &par->vmw_bo->base;
-	par->vmw_bo = NULL;
 
 	/* ??? order */
 	fb_deferred_io_cleanup(info);
 	unregister_framebuffer(info);
 
-	ttm_bo_kunmap(&par->map);
-	ttm_bo_unref(&bo);
+	(void) vmw_fb_kms_detach(par, true, true);
 
 	vfree(par->vmalloc);
 	framebuffer_release(info);
@@ -597,20 +805,16 @@ int vmw_fb_off(struct vmw_private *vmw_priv)
 
 	info = vmw_priv->fb_info;
 	par = info->par;
-	if (!par->bo_ptr)
-		return 0;
 
-	vmw_kms_save_vga(vmw_priv);
 	spin_lock_irqsave(&par->dirty.lock, flags);
 	par->dirty.active = false;
 	spin_unlock_irqrestore(&par->dirty.lock, flags);
 
 	flush_delayed_work(&info->deferred_work);
 
-	par->bo_ptr = NULL;
-	ttm_bo_kunmap(&par->map);
-
-	vmw_dmabuf_unpin(vmw_priv, par->vmw_bo, false);
+	mutex_lock(&par->bo_mutex);
+	(void) vmw_fb_kms_detach(par, true, false);
+	mutex_unlock(&par->bo_mutex);
 
 	return 0;
 }
@@ -620,8 +824,6 @@ int vmw_fb_on(struct vmw_private *vmw_priv)
 	struct fb_info *info;
 	struct vmw_fb_par *par;
 	unsigned long flags;
-	bool dummy;
-	int ret;
 
 	if (!vmw_priv->fb_info)
 		return -EINVAL;
@@ -629,39 +831,10 @@ int vmw_fb_on(struct vmw_private *vmw_priv)
 	info = vmw_priv->fb_info;
 	par = info->par;
 
-	/* we are already active */
-	if (par->bo_ptr != NULL)
-		return 0;
-
-	/* Make sure that all overlays are stoped when we take over */
-	vmw_overlay_stop_all(vmw_priv);
-
-	ret = vmw_dmabuf_pin_in_start_of_vram(vmw_priv, par->vmw_bo, false);
-	if (unlikely(ret != 0)) {
-		DRM_ERROR("could not move buffer to start of VRAM\n");
-		goto err_no_buffer;
-	}
-
-	ret = ttm_bo_kmap(&par->vmw_bo->base,
-			  0,
-			  par->vmw_bo->base.num_pages,
-			  &par->map);
-	BUG_ON(ret != 0);
-	par->bo_ptr = ttm_kmap_obj_virtual(&par->map, &dummy);
-
+	vmw_fb_set_par(info);
 	spin_lock_irqsave(&par->dirty.lock, flags);
 	par->dirty.active = true;
 	spin_unlock_irqrestore(&par->dirty.lock, flags);
-	vmw_kms_restore_vga(vmw_priv);
-
-err_no_buffer:
-	vmw_fb_set_par(info);
-
-	vmw_fb_dirty_mark(par, 0, 0, info->var.xres, info->var.yres);
-
-	/* If there already was stuff dirty we wont
-	 * schedule a new work, so lets do it now */
-	schedule_delayed_work(&info->deferred_work, 0);
-
+ 
 	return 0;
 }
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
index dc9f7d0..06ff7c8 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
@@ -372,7 +372,8 @@ static void vmw_framebuffer_surface_destroy(struct drm_framebuffer *framebuffer)
 
 	drm_framebuffer_cleanup(framebuffer);
 	vmw_surface_unreference(&vfbs->surface);
-	ttm_base_object_unref(&vfbs->base.user_obj);
+	if (vfbs->base.user_obj)
+		ttm_base_object_unref(&vfbs->base.user_obj);
 
 	kfree(vfbs);
 }
@@ -582,7 +583,8 @@ static void vmw_framebuffer_dmabuf_destroy(struct drm_framebuffer *framebuffer)
 
 	drm_framebuffer_cleanup(framebuffer);
 	vmw_dmabuf_unreference(&vfbd->buffer);
-	ttm_base_object_unref(&vfbd->base.user_obj);
+	if (vfbd->base.user_obj)
+		ttm_base_object_unref(&vfbd->base.user_obj);
 
 	kfree(vfbd);
 }
@@ -1462,7 +1464,7 @@ static struct drm_display_mode vmw_kms_connector_builtin[] = {
  * @mode - Pointer to a struct drm_display_mode with hdisplay and vdisplay
  * members filled in.
  */
-static void vmw_guess_mode_timing(struct drm_display_mode *mode)
+void vmw_guess_mode_timing(struct drm_display_mode *mode)
 {
 	mode->hsync_start = mode->hdisplay + 50;
 	mode->hsync_end = mode->hsync_start + 50;
@@ -2001,3 +2003,58 @@ int vmw_kms_update_proxy(struct vmw_resource *res,
 
 	return 0;
 }
+
+int vmw_kms_fbdev_init_data(struct vmw_private *dev_priv,
+			    unsigned unit,
+			    u32 max_width,
+			    u32 max_height,
+			    struct drm_connector **p_con,
+			    struct drm_crtc **p_crtc,
+			    struct drm_display_mode **p_mode)
+{
+	struct drm_connector *con;
+	struct vmw_display_unit *du;
+	struct drm_display_mode *mode;
+	int i = 0;
+
+	list_for_each_entry(con, &dev_priv->dev->mode_config.connector_list,
+			    head) {
+		if (i == unit)
+			break;
+
+		++i;
+	}
+
+	if (i != unit) {
+		DRM_ERROR("Could not find initial display unit.\n");
+		return -EINVAL;
+	}
+
+	if (list_empty(&con->modes))
+		(void) vmw_du_connector_fill_modes(con, max_width, max_height);
+
+	if (list_empty(&con->modes)) {
+		DRM_ERROR("Could not find initial display mode.\n");
+		return -EINVAL;
+	}
+
+	du = vmw_connector_to_du(con);
+	*p_con = con;
+	*p_crtc = &du->crtc;
+
+	list_for_each_entry(mode, &con->modes, head) {
+		if (mode->type & DRM_MODE_TYPE_PREFERRED)
+			break;
+	}
+
+	if (mode->type & DRM_MODE_TYPE_PREFERRED)
+		*p_mode = mode;
+	else {
+		WARN_ONCE(true, "Could not find initial preferred mode.\n");
+		*p_mode = list_first_entry(&con->modes,
+					   struct drm_display_mode,
+					   head);
+	}
+
+	return 0;
+}
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
index 311effc..eb6c853 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
@@ -243,6 +243,14 @@ vmw_kms_new_framebuffer(struct vmw_private *dev_priv,
 			struct vmw_surface *surface,
 			bool only_2d,
 			const struct drm_mode_fb_cmd *mode_cmd);
+int vmw_kms_fbdev_init_data(struct vmw_private *dev_priv,
+			    unsigned unit,
+			    u32 max_width,
+			    u32 max_height,
+			    struct drm_connector **p_con,
+			    struct drm_crtc **p_crtc,
+			    struct drm_display_mode **p_mode);
+void vmw_guess_mode_timing(struct drm_display_mode *mode);
 
 /*
  * Legacy display unit functions - vmwgfx_ldu.c
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c b/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c
index f0fd565..51721c3 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c
@@ -279,7 +279,6 @@ static int vmw_ldu_crtc_set_config(struct drm_mode_set *set)
 		return -EINVAL;
 	}
 
-	vmw_fb_off(dev_priv);
 	vmw_svga_enable(dev_priv);
 
 	crtc->primary->fb = fb;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c b/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c
index 73fe20e..8b5bc17 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c
@@ -369,7 +369,6 @@ static int vmw_sou_crtc_set_config(struct drm_mode_set *set)
 		return -EINVAL;
 	}
 
-	vmw_fb_off(dev_priv);
 	vmw_svga_enable(dev_priv);
 
 	if (mode->hdisplay != crtc->mode.hdisplay ||
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 21/28] drm/vmwgfx: Reinstate the legacy display system dirty callback

[Thomas Hellstrom]

It somehow got lost in a rewrite.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_kms.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
index 06ff7c8..ae87e7e 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
@@ -632,10 +632,13 @@ static int vmw_framebuffer_dmabuf_dirty(struct drm_framebuffer *framebuffer,
 						  true,
 						  NULL);
 		break;
+	case vmw_du_legacy:
+		ret = vmw_kms_ldu_do_dmabuf_dirty(dev_priv, &vfbd->base, 0, 0,
+						  clips, num_clips, increment);
+		break;
 	default:
-		ret = -ENOSYS;
-		WARN_ONCE(true,
-			  "Dirty called with invalid display system.\n");
+		ret = -EINVAL;
+		WARN_ONCE(true, "Dirty called with invalid display system.\n");
 		break;
 	}
 
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 22/28] drm/vmwgfx: Fix kms preferred mode sorting

[Thomas Hellstrom]

The preferred mode typically didn't end up first, since the function
drm_mode_connector_list_update() reordered the modes.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_kms.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
index ae87e7e..ef605b6 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
@@ -1554,11 +1554,9 @@ int vmw_du_connector_fill_modes(struct drm_connector *connector,
 		drm_mode_probed_add(connector, mode);
 	}
 
-	/* Move the prefered mode first, help apps pick the right mode. */
-	if (du->pref_mode)
-		list_move(&du->pref_mode->head, &connector->probed_modes);
-
 	drm_mode_connector_list_update(connector, true);
+	/* Move the prefered mode first, help apps pick the right mode. */
+	drm_mode_sort(&connector->modes);
 
 	return 1;
 }
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 23/28] drm/vmwgfx: Kill a bunch of sparse warnings

[Thomas Hellstrom]

We're giving up all attempts to keep cpu- and device byte ordering separate.

This silences sparse when compiled using
make C=2 CF="-D__CHECK_ENDIAN__"

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
---
 drivers/gpu/drm/vmwgfx/svga3d_reg.h      |  2 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c   |  8 +++----
 drivers/gpu/drm/vmwgfx/vmwgfx_context.c  | 12 +++++------
 drivers/gpu/drm/vmwgfx/vmwgfx_dmabuf.c   |  2 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.c      |  4 ++--
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.h      |  6 +++---
 drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c  | 18 ++++++++--------
 drivers/gpu/drm/vmwgfx/vmwgfx_fb.c       |  4 ++--
 drivers/gpu/drm/vmwgfx/vmwgfx_fence.c    |  8 +++----
 drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c     | 37 ++++++++++++++++----------------
 drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c    |  8 +++----
 drivers/gpu/drm/vmwgfx/vmwgfx_irq.c      |  4 ++--
 drivers/gpu/drm/vmwgfx/vmwgfx_kms.c      | 32 +++++++++++++--------------
 drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c      | 10 ++++-----
 drivers/gpu/drm/vmwgfx/vmwgfx_mob.c      | 14 ++++++------
 drivers/gpu/drm/vmwgfx/vmwgfx_reg.h      |  6 +++---
 drivers/gpu/drm/vmwgfx/vmwgfx_resource.c | 17 ++++++---------
 drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c     |  4 ++--
 drivers/gpu/drm/vmwgfx/vmwgfx_shader.c   | 10 ++++-----
 drivers/gpu/drm/vmwgfx/vmwgfx_surface.c  |  4 ++--
 20 files changed, 103 insertions(+), 107 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/svga3d_reg.h b/drivers/gpu/drm/vmwgfx/svga3d_reg.h
index c9a595a..f4af9f1 100644
--- a/drivers/gpu/drm/vmwgfx/svga3d_reg.h
+++ b/drivers/gpu/drm/vmwgfx/svga3d_reg.h
@@ -35,7 +35,7 @@
 #include "svga_reg.h"
 
 typedef uint32 PPN;
-typedef __le64 PPN64;
+typedef u64 PPN64;
 
 /*
  * 3D Hardware Version
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c
index e94feb3..32ec52e 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c
@@ -695,10 +695,10 @@ static bool vmw_cmdbuf_try_alloc(struct vmw_cmdbuf_man *man,
  * no space available ATM, it turns on IRQ handling and sleeps waiting for it to
  * become available.
  */
-int vmw_cmdbuf_alloc_space(struct vmw_cmdbuf_man *man,
-			   struct drm_mm_node *node,
-			   size_t size,
-			   bool interruptible)
+static int vmw_cmdbuf_alloc_space(struct vmw_cmdbuf_man *man,
+				  struct drm_mm_node *node,
+				  size_t size,
+				  bool interruptible)
 {
 	struct vmw_cmdbuf_alloc_info info;
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_context.c b/drivers/gpu/drm/vmwgfx/vmwgfx_context.c
index a8e370a..2aa8bb8 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_context.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_context.c
@@ -135,9 +135,9 @@ static void vmw_hw_context_destroy(struct vmw_resource *res)
 		return;
 	}
 
-	cmd->header.id = cpu_to_le32(SVGA_3D_CMD_CONTEXT_DESTROY);
-	cmd->header.size = cpu_to_le32(sizeof(cmd->body));
-	cmd->body.cid = cpu_to_le32(res->id);
+	cmd->header.id = SVGA_3D_CMD_CONTEXT_DESTROY;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.cid = res->id;
 
 	vmw_fifo_commit(dev_priv, sizeof(*cmd));
 	vmw_fifo_resource_dec(dev_priv);
@@ -215,9 +215,9 @@ static int vmw_context_init(struct vmw_private *dev_priv,
 		return -ENOMEM;
 	}
 
-	cmd->header.id = cpu_to_le32(SVGA_3D_CMD_CONTEXT_DEFINE);
-	cmd->header.size = cpu_to_le32(sizeof(cmd->body));
-	cmd->body.cid = cpu_to_le32(res->id);
+	cmd->header.id = SVGA_3D_CMD_CONTEXT_DEFINE;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.cid = res->id;
 
 	vmw_fifo_commit(dev_priv, sizeof(*cmd));
 	vmw_fifo_resource_inc(dev_priv);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_dmabuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_dmabuf.c
index 4b9344d..9b4f093 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_dmabuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_dmabuf.c
@@ -225,7 +225,7 @@ int vmw_dmabuf_unpin(struct vmw_private *dev_priv,
 	if (unlikely(ret != 0))
 		return ret;
 
-	ret = ttm_bo_reserve(bo, interruptible, false, false, 0);
+	ret = ttm_bo_reserve(bo, interruptible, false, false, NULL);
 	if (unlikely(ret != 0))
 		goto err;
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
index 1892144..ab67d2a 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
@@ -1225,7 +1225,7 @@ static void vmw_master_drop(struct drm_device *dev,
  * @dev_priv: Pointer to device private struct.
  * Needs the reservation sem to be held in non-exclusive mode.
  */
-void __vmw_svga_enable(struct vmw_private *dev_priv)
+static void __vmw_svga_enable(struct vmw_private *dev_priv)
 {
 	spin_lock(&dev_priv->svga_lock);
 	if (!dev_priv->bdev.man[TTM_PL_VRAM].use_type) {
@@ -1254,7 +1254,7 @@ void vmw_svga_enable(struct vmw_private *dev_priv)
  * Needs the reservation sem to be held in exclusive mode.
  * Will not empty VRAM. VRAM must be emptied by caller.
  */
-void __vmw_svga_disable(struct vmw_private *dev_priv)
+static void __vmw_svga_disable(struct vmw_private *dev_priv)
 {
 	spin_lock(&dev_priv->svga_lock);
 	if (dev_priv->bdev.man[TTM_PL_VRAM].use_type) {
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index 9ae5736..c9ea9b1 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -178,8 +178,8 @@ struct vmw_marker_queue {
 
 struct vmw_fifo_state {
 	unsigned long reserved_size;
-	__le32 *dynamic_buffer;
-	__le32 *static_buffer;
+	u32 *dynamic_buffer;
+	u32 *static_buffer;
 	unsigned long static_buffer_size;
 	bool using_bounce_buffer;
 	uint32_t capabilities;
@@ -405,7 +405,7 @@ struct vmw_private {
 	uint32_t stdu_max_height;
 	uint32_t initial_width;
 	uint32_t initial_height;
-	__le32 __iomem *mmio_virt;
+	u32 __iomem *mmio_virt;
 	int mmio_mtrr;
 	uint32_t capabilities;
 	uint32_t max_gmr_ids;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
index 64dba53..40fdd02 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
@@ -1850,7 +1850,7 @@ static int vmw_cmd_check_not_3d(struct vmw_private *dev_priv,
 	uint32_t size_remaining = *size;
 	uint32_t cmd_id;
 
-	cmd_id = le32_to_cpu(((uint32_t *)buf)[0]);
+	cmd_id = ((uint32_t *)buf)[0];
 	switch (cmd_id) {
 	case SVGA_CMD_UPDATE:
 		*size = sizeof(uint32_t) + sizeof(SVGAFifoCmdUpdate);
@@ -2066,14 +2066,14 @@ static int vmw_cmd_check(struct vmw_private *dev_priv,
 	const struct vmw_cmd_entry *entry;
 	bool gb = dev_priv->capabilities & SVGA_CAP_GBOBJECTS;
 
-	cmd_id = le32_to_cpu(((uint32_t *)buf)[0]);
+	cmd_id = ((uint32_t *)buf)[0];
 	/* Handle any none 3D commands */
 	if (unlikely(cmd_id < SVGA_CMD_MAX))
 		return vmw_cmd_check_not_3d(dev_priv, sw_context, buf, size);
 
 
-	cmd_id = le32_to_cpu(header->id);
-	*size = le32_to_cpu(header->size) + sizeof(SVGA3dCmdHeader);
+	cmd_id = header->id;
+	*size = header->size + sizeof(SVGA3dCmdHeader);
 
 	cmd_id -= SVGA_3D_CMD_BASE;
 	if (unlikely(*size > size_remaining))
@@ -2499,11 +2499,11 @@ static int vmw_execbuf_submit_cmdbuf(struct vmw_private *dev_priv,
  * If the function is interrupted by a signal while sleeping, it will return
  * -ERESTARTSYS casted to a pointer error value.
  */
-void *vmw_execbuf_cmdbuf(struct vmw_private *dev_priv,
-			 void __user *user_commands,
-			 void *kernel_commands,
-			 u32 command_size,
-			 struct vmw_cmdbuf_header **header)
+static void *vmw_execbuf_cmdbuf(struct vmw_private *dev_priv,
+				void __user *user_commands,
+				void *kernel_commands,
+				u32 command_size,
+				struct vmw_cmdbuf_header **header)
 {
 	size_t cmdbuf_size;
 	int ret;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
index 9dbb203..9856803 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
@@ -331,7 +331,7 @@ static void vmw_deferred_io(struct fb_info *info,
 	vmw_fb_dirty_flush(par);
 };
 
-struct fb_deferred_io vmw_defio = {
+static struct fb_deferred_io vmw_defio = {
 	.delay		= VMW_DIRTY_DELAY,
 	.deferred_io	= vmw_deferred_io,
 };
@@ -706,7 +706,7 @@ int vmw_fb_init(struct vmw_private *vmw_priv)
 	info->fix.smem_len = fb_size;
 
 	info->pseudo_palette = par->pseudo_palette;
-	info->screen_base = par->vmalloc;
+	info->screen_base = (char __iomem *)par->vmalloc;
 	info->screen_size = fb_size;
 
 	info->flags = FBINFO_DEFAULT;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
index 945f1e0..75d6222 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
@@ -142,7 +142,7 @@ static bool vmw_fence_enable_signaling(struct fence *f)
 	struct vmw_fence_manager *fman = fman_from_fence(fence);
 	struct vmw_private *dev_priv = fman->dev_priv;
 
-	__le32 __iomem *fifo_mem = dev_priv->mmio_virt;
+	u32 __iomem *fifo_mem = dev_priv->mmio_virt;
 	u32 seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE);
 	if (seqno - fence->base.seqno < VMW_FENCE_WRAP)
 		return false;
@@ -386,7 +386,7 @@ static bool vmw_fence_goal_new_locked(struct vmw_fence_manager *fman,
 				      u32 passed_seqno)
 {
 	u32 goal_seqno;
-	__le32 __iomem *fifo_mem;
+	u32 __iomem *fifo_mem;
 	struct vmw_fence_obj *fence;
 
 	if (likely(!fman->seqno_valid))
@@ -430,7 +430,7 @@ static bool vmw_fence_goal_check_locked(struct vmw_fence_obj *fence)
 {
 	struct vmw_fence_manager *fman = fman_from_fence(fence);
 	u32 goal_seqno;
-	__le32 __iomem *fifo_mem;
+	u32 __iomem *fifo_mem;
 
 	if (fence_is_signaled_locked(&fence->base))
 		return false;
@@ -453,7 +453,7 @@ static void __vmw_fences_update(struct vmw_fence_manager *fman)
 	struct list_head action_list;
 	bool needs_rerun;
 	uint32_t seqno, new_seqno;
-	__le32 __iomem *fifo_mem = fman->dev_priv->mmio_virt;
+	u32 __iomem *fifo_mem = fman->dev_priv->mmio_virt;
 
 	seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE);
 rerun:
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
index 9b8b09f..7a6cf17 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
@@ -31,7 +31,7 @@
 
 bool vmw_fifo_have_3d(struct vmw_private *dev_priv)
 {
-	__le32 __iomem *fifo_mem = dev_priv->mmio_virt;
+	u32 __iomem *fifo_mem = dev_priv->mmio_virt;
 	uint32_t fifo_min, hwversion;
 	const struct vmw_fifo_state *fifo = &dev_priv->fifo;
 
@@ -80,7 +80,7 @@ bool vmw_fifo_have_3d(struct vmw_private *dev_priv)
 
 bool vmw_fifo_have_pitchlock(struct vmw_private *dev_priv)
 {
-	__le32 __iomem *fifo_mem = dev_priv->mmio_virt;
+	u32 __iomem *fifo_mem = dev_priv->mmio_virt;
 	uint32_t caps;
 
 	if (!(dev_priv->capabilities & SVGA_CAP_EXTENDED_FIFO))
@@ -95,7 +95,7 @@ bool vmw_fifo_have_pitchlock(struct vmw_private *dev_priv)
 
 int vmw_fifo_init(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
 {
-	__le32 __iomem *fifo_mem = dev_priv->mmio_virt;
+	u32 __iomem *fifo_mem = dev_priv->mmio_virt;
 	uint32_t max;
 	uint32_t min;
 
@@ -158,7 +158,7 @@ int vmw_fifo_init(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
 
 void vmw_fifo_ping_host(struct vmw_private *dev_priv, uint32_t reason)
 {
-	__le32 __iomem *fifo_mem = dev_priv->mmio_virt;
+	u32 __iomem *fifo_mem = dev_priv->mmio_virt;
 	static DEFINE_SPINLOCK(ping_lock);
 	unsigned long irq_flags;
 
@@ -176,7 +176,7 @@ void vmw_fifo_ping_host(struct vmw_private *dev_priv, uint32_t reason)
 
 void vmw_fifo_release(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
 {
-	__le32 __iomem *fifo_mem = dev_priv->mmio_virt;
+	u32 __iomem *fifo_mem = dev_priv->mmio_virt;
 
 	vmw_write(dev_priv, SVGA_REG_SYNC, SVGA_SYNC_GENERIC);
 	while (vmw_read(dev_priv, SVGA_REG_BUSY) != 0)
@@ -206,7 +206,7 @@ void vmw_fifo_release(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
 
 static bool vmw_fifo_is_full(struct vmw_private *dev_priv, uint32_t bytes)
 {
-	__le32 __iomem *fifo_mem = dev_priv->mmio_virt;
+	u32 __iomem *fifo_mem = dev_priv->mmio_virt;
 	uint32_t max = ioread32(fifo_mem + SVGA_FIFO_MAX);
 	uint32_t next_cmd = ioread32(fifo_mem + SVGA_FIFO_NEXT_CMD);
 	uint32_t min = ioread32(fifo_mem + SVGA_FIFO_MIN);
@@ -314,7 +314,7 @@ static void *vmw_local_fifo_reserve(struct vmw_private *dev_priv,
 				    uint32_t bytes)
 {
 	struct vmw_fifo_state *fifo_state = &dev_priv->fifo;
-	__le32 __iomem *fifo_mem = dev_priv->mmio_virt;
+	u32 __iomem *fifo_mem = dev_priv->mmio_virt;
 	uint32_t max;
 	uint32_t min;
 	uint32_t next_cmd;
@@ -371,7 +371,8 @@ static void *vmw_local_fifo_reserve(struct vmw_private *dev_priv,
 				if (reserveable)
 					iowrite32(bytes, fifo_mem +
 						  SVGA_FIFO_RESERVED);
-				return fifo_mem + (next_cmd >> 2);
+				return (void __force *) (fifo_mem +
+							 (next_cmd >> 2));
 			} else {
 				need_bounce = true;
 			}
@@ -414,7 +415,7 @@ void *vmw_fifo_reserve(struct vmw_private *dev_priv, uint32_t bytes)
 }
 
 static void vmw_fifo_res_copy(struct vmw_fifo_state *fifo_state,
-			      __le32 __iomem *fifo_mem,
+			      u32 __iomem *fifo_mem,
 			      uint32_t next_cmd,
 			      uint32_t max, uint32_t min, uint32_t bytes)
 {
@@ -436,7 +437,7 @@ static void vmw_fifo_res_copy(struct vmw_fifo_state *fifo_state,
 }
 
 static void vmw_fifo_slow_copy(struct vmw_fifo_state *fifo_state,
-			       __le32 __iomem *fifo_mem,
+			       u32 __iomem *fifo_mem,
 			       uint32_t next_cmd,
 			       uint32_t max, uint32_t min, uint32_t bytes)
 {
@@ -455,10 +456,10 @@ static void vmw_fifo_slow_copy(struct vmw_fifo_state *fifo_state,
 	}
 }
 
-void vmw_local_fifo_commit(struct vmw_private *dev_priv, uint32_t bytes)
+static void vmw_local_fifo_commit(struct vmw_private *dev_priv, uint32_t bytes)
 {
 	struct vmw_fifo_state *fifo_state = &dev_priv->fifo;
-	__le32 __iomem *fifo_mem = dev_priv->mmio_virt;
+	u32 __iomem *fifo_mem = dev_priv->mmio_virt;
 	uint32_t next_cmd = ioread32(fifo_mem + SVGA_FIFO_NEXT_CMD);
 	uint32_t max = ioread32(fifo_mem + SVGA_FIFO_MAX);
 	uint32_t min = ioread32(fifo_mem + SVGA_FIFO_MIN);
@@ -545,9 +546,9 @@ int vmw_fifo_send_fence(struct vmw_private *dev_priv, uint32_t *seqno)
 {
 	struct vmw_fifo_state *fifo_state = &dev_priv->fifo;
 	struct svga_fifo_cmd_fence *cmd_fence;
-	void *fm;
+	u32 *fm;
 	int ret = 0;
-	uint32_t bytes = sizeof(__le32) + sizeof(*cmd_fence);
+	uint32_t bytes = sizeof(u32) + sizeof(*cmd_fence);
 
 	fm = vmw_fifo_reserve(dev_priv, bytes);
 	if (unlikely(fm == NULL)) {
@@ -573,11 +574,9 @@ int vmw_fifo_send_fence(struct vmw_private *dev_priv, uint32_t *seqno)
 		return 0;
 	}
 
-	*(__le32 *) fm = cpu_to_le32(SVGA_CMD_FENCE);
-	cmd_fence = (struct svga_fifo_cmd_fence *)
-	    ((unsigned long)fm + sizeof(__le32));
-
-	iowrite32(*seqno, &cmd_fence->fence);
+	*fm++ = SVGA_CMD_FENCE;
+	cmd_fence = (struct svga_fifo_cmd_fence *) fm;
+	cmd_fence->fence = *seqno;
 	vmw_fifo_commit_flush(dev_priv, bytes);
 	(void) vmw_marker_push(&fifo_state->marker_queue, *seqno);
 	vmw_update_seqno(dev_priv, fifo_state);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c b/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c
index 55940bc..6db9828 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c
@@ -63,7 +63,7 @@ int vmw_getparam_ioctl(struct drm_device *dev, void *data,
 		break;
 	case DRM_VMW_PARAM_FIFO_HW_VERSION:
 	{
-		__le32 __iomem *fifo_mem = dev_priv->mmio_virt;
+		u32 __iomem *fifo_mem = dev_priv->mmio_virt;
 		const struct vmw_fifo_state *fifo = &dev_priv->fifo;
 
 		if ((dev_priv->capabilities & SVGA_CAP_GBOBJECTS)) {
@@ -158,7 +158,7 @@ int vmw_get_cap_3d_ioctl(struct drm_device *dev, void *data,
 		(struct drm_vmw_get_3d_cap_arg *) data;
 	struct vmw_private *dev_priv = vmw_priv(dev);
 	uint32_t size;
-	__le32 __iomem *fifo_mem;
+	u32 __iomem *fifo_mem;
 	void __user *buffer = (void __user *)((unsigned long)(arg->buffer));
 	void *bounce;
 	int ret;
@@ -239,7 +239,7 @@ int vmw_present_ioctl(struct drm_device *dev, void *data,
 	int ret;
 
 	num_clips = arg->num_clips;
-	clips_ptr = (struct drm_vmw_rect *)(unsigned long)arg->clips_ptr;
+	clips_ptr = (struct drm_vmw_rect __user *)(unsigned long)arg->clips_ptr;
 
 	if (unlikely(num_clips == 0))
 		return 0;
@@ -322,7 +322,7 @@ int vmw_present_readback_ioctl(struct drm_device *dev, void *data,
 	int ret;
 
 	num_clips = arg->num_clips;
-	clips_ptr = (struct drm_vmw_rect *)(unsigned long)arg->clips_ptr;
+	clips_ptr = (struct drm_vmw_rect __user *)(unsigned long)arg->clips_ptr;
 
 	if (unlikely(num_clips == 0))
 		return 0;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c b/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c
index 87964bb..2c2bac4 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c
@@ -72,7 +72,7 @@ static bool vmw_fifo_idle(struct vmw_private *dev_priv, uint32_t seqno)
 void vmw_update_seqno(struct vmw_private *dev_priv,
 			 struct vmw_fifo_state *fifo_state)
 {
-	__le32 __iomem *fifo_mem = dev_priv->mmio_virt;
+	u32 __iomem *fifo_mem = dev_priv->mmio_virt;
 	uint32_t seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE);
 
 	if (dev_priv->last_read_seqno != seqno) {
@@ -178,7 +178,7 @@ int vmw_fallback_wait(struct vmw_private *dev_priv,
 	}
 	finish_wait(&dev_priv->fence_queue, &__wait);
 	if (ret == 0 && fifo_idle) {
-		__le32 __iomem *fifo_mem = dev_priv->mmio_virt;
+		u32 __iomem *fifo_mem = dev_priv->mmio_virt;
 		iowrite32(signal_seq, fifo_mem + SVGA_FIFO_FENCE);
 	}
 	wake_up_all(&dev_priv->fence_queue);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
index ef605b6..ca69ed4 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
@@ -71,12 +71,12 @@ int vmw_cursor_update_image(struct vmw_private *dev_priv,
 
 	memcpy(&cmd[1], image, image_size);
 
-	cmd->cmd = cpu_to_le32(SVGA_CMD_DEFINE_ALPHA_CURSOR);
-	cmd->cursor.id = cpu_to_le32(0);
-	cmd->cursor.width = cpu_to_le32(width);
-	cmd->cursor.height = cpu_to_le32(height);
-	cmd->cursor.hotspotX = cpu_to_le32(hotspotX);
-	cmd->cursor.hotspotY = cpu_to_le32(hotspotY);
+	cmd->cmd = SVGA_CMD_DEFINE_ALPHA_CURSOR;
+	cmd->cursor.id = 0;
+	cmd->cursor.width = width;
+	cmd->cursor.height = height;
+	cmd->cursor.hotspotX = hotspotX;
+	cmd->cursor.hotspotY = hotspotY;
 
 	vmw_fifo_commit(dev_priv, cmd_size);
 
@@ -123,7 +123,7 @@ err_unreserve:
 void vmw_cursor_update_position(struct vmw_private *dev_priv,
 				bool show, int x, int y)
 {
-	__le32 __iomem *fifo_mem = dev_priv->mmio_virt;
+	u32 __iomem *fifo_mem = dev_priv->mmio_virt;
 	uint32_t count;
 
 	iowrite32(show ? 1 : 0, fifo_mem + SVGA_FIFO_CURSOR_ON);
@@ -1017,14 +1017,14 @@ static const struct drm_mode_config_funcs vmw_kms_funcs = {
 	.fb_create = vmw_kms_fb_create,
 };
 
-int vmw_kms_generic_present(struct vmw_private *dev_priv,
-		    struct drm_file *file_priv,
-		    struct vmw_framebuffer *vfb,
-		    struct vmw_surface *surface,
-		    uint32_t sid,
-		    int32_t destX, int32_t destY,
-		    struct drm_vmw_rect *clips,
-		    uint32_t num_clips)
+static int vmw_kms_generic_present(struct vmw_private *dev_priv,
+				   struct drm_file *file_priv,
+				   struct vmw_framebuffer *vfb,
+				   struct vmw_surface *surface,
+				   uint32_t sid,
+				   int32_t destX, int32_t destY,
+				   struct drm_vmw_rect *clips,
+				   uint32_t num_clips)
 {
 	return vmw_kms_sou_do_surface_dirty(dev_priv, vfb, NULL, clips,
 					    &surface->res, destX, destY,
@@ -1785,7 +1785,7 @@ int vmw_kms_helper_buffer_prepare(struct vmw_private *dev_priv,
 	struct ttm_buffer_object *bo = &buf->base;
 	int ret;
 
-	ttm_bo_reserve(bo, false, false, interruptible, 0);
+	ttm_bo_reserve(bo, false, false, interruptible, NULL);
 	ret = vmw_validate_single_buffer(dev_priv, bo, interruptible,
 					 validate_as_mob);
 	if (ret)
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c b/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c
index 51721c3..5503845 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c
@@ -476,11 +476,11 @@ int vmw_kms_ldu_do_dmabuf_dirty(struct vmw_private *dev_priv,
 
 	memset(cmd, 0, fifo_size);
 	for (i = 0; i < num_clips; i++, clips += increment) {
-		cmd[i].header = cpu_to_le32(SVGA_CMD_UPDATE);
-		cmd[i].body.x = cpu_to_le32(clips->x1);
-		cmd[i].body.y = cpu_to_le32(clips->y1);
-		cmd[i].body.width = cpu_to_le32(clips->x2 - clips->x1);
-		cmd[i].body.height = cpu_to_le32(clips->y2 - clips->y1);
+		cmd[i].header = SVGA_CMD_UPDATE;
+		cmd[i].body.x = clips->x1;
+		cmd[i].body.y = clips->y1;
+		cmd[i].body.width = clips->x2 - clips->x1;
+		cmd[i].body.height = clips->y2 - clips->y1;
 	}
 
 	vmw_fifo_commit(dev_priv, fifo_size);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c b/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c
index e0fc248..c5897cb 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c
@@ -142,7 +142,7 @@ static int vmw_setup_otable_base(struct vmw_private *dev_priv,
 	cmd->header.id = SVGA_3D_CMD_SET_OTABLE_BASE64;
 	cmd->header.size = sizeof(cmd->body);
 	cmd->body.type = type;
-	cmd->body.baseAddress = cpu_to_le64(mob->pt_root_page >> PAGE_SHIFT);
+	cmd->body.baseAddress = mob->pt_root_page >> PAGE_SHIFT;
 	cmd->body.sizeInBytes = otable->size;
 	cmd->body.validSizeInBytes = 0;
 	cmd->body.ptDepth = mob->pt_level;
@@ -430,15 +430,15 @@ out_unreserve:
  * *@addr according to the page table entry size.
  */
 #if (VMW_PPN_SIZE == 8)
-static void vmw_mob_assign_ppn(__le32 **addr, dma_addr_t val)
+static void vmw_mob_assign_ppn(u32 **addr, dma_addr_t val)
 {
-	*((__le64 *) *addr) = cpu_to_le64(val >> PAGE_SHIFT);
+	*((u64 *) *addr) = val >> PAGE_SHIFT;
 	*addr += 2;
 }
 #else
-static void vmw_mob_assign_ppn(__le32 **addr, dma_addr_t val)
+static void vmw_mob_assign_ppn(u32 **addr, dma_addr_t val)
 {
-	*(*addr)++ = cpu_to_le32(val >> PAGE_SHIFT);
+	*(*addr)++ = val >> PAGE_SHIFT;
 }
 #endif
 
@@ -460,7 +460,7 @@ static unsigned long vmw_mob_build_pt(struct vmw_piter *data_iter,
 	unsigned long pt_size = num_data_pages * VMW_PPN_SIZE;
 	unsigned long num_pt_pages = DIV_ROUND_UP(pt_size, PAGE_SIZE);
 	unsigned long pt_page;
-	__le32 *addr, *save_addr;
+	u32 *addr, *save_addr;
 	unsigned long i;
 	struct page *page;
 
@@ -641,7 +641,7 @@ int vmw_mob_bind(struct vmw_private *dev_priv,
 	cmd->header.size = sizeof(cmd->body);
 	cmd->body.mobid = mob_id;
 	cmd->body.ptDepth = mob->pt_level;
-	cmd->body.base = cpu_to_le64(mob->pt_root_page >> PAGE_SHIFT);
+	cmd->body.base = mob->pt_root_page >> PAGE_SHIFT;
 	cmd->body.sizeInBytes = num_data_pages * PAGE_SIZE;
 
 	vmw_fifo_commit(dev_priv, sizeof(*cmd));
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_reg.h b/drivers/gpu/drm/vmwgfx/vmwgfx_reg.h
index 9d0dd3a..29d06a4 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_reg.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_reg.h
@@ -39,12 +39,12 @@
 #define VMWGFX_IRQSTATUS_PORT 0x8
 
 struct svga_guest_mem_descriptor {
-	__le32 ppn;
-	__le32 num_pages;
+	u32 ppn;
+	u32 num_pages;
 };
 
 struct svga_fifo_cmd_fence {
-	__le32 fence;
+	u32 fence;
 };
 
 #define SVGA_SYNC_GENERIC         1
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
index 69b471a..be2809a 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
@@ -121,6 +121,7 @@ static void vmw_resource_release(struct kref *kref)
 	int id;
 	struct idr *idr = &dev_priv->res_idr[res->func->res_type];
 
+	write_lock(&dev_priv->resource_lock);
 	res->avail = false;
 	list_del_init(&res->lru_head);
 	write_unlock(&dev_priv->resource_lock);
@@ -156,20 +157,17 @@ static void vmw_resource_release(struct kref *kref)
 		kfree(res);
 
 	write_lock(&dev_priv->resource_lock);
-
 	if (id != -1)
 		idr_remove(idr, id);
+	write_unlock(&dev_priv->resource_lock);
 }
 
 void vmw_resource_unreference(struct vmw_resource **p_res)
 {
 	struct vmw_resource *res = *p_res;
-	struct vmw_private *dev_priv = res->dev_priv;
 
 	*p_res = NULL;
-	write_lock(&dev_priv->resource_lock);
 	kref_put(&res->kref, vmw_resource_release);
-	write_unlock(&dev_priv->resource_lock);
 }
 
 
@@ -260,17 +258,16 @@ void vmw_resource_activate(struct vmw_resource *res,
 	write_unlock(&dev_priv->resource_lock);
 }
 
-struct vmw_resource *vmw_resource_lookup(struct vmw_private *dev_priv,
-					 struct idr *idr, int id)
+static struct vmw_resource *vmw_resource_lookup(struct vmw_private *dev_priv,
+						struct idr *idr, int id)
 {
 	struct vmw_resource *res;
 
 	read_lock(&dev_priv->resource_lock);
 	res = idr_find(idr, id);
-	if (res && res->avail)
-		kref_get(&res->kref);
-	else
+	if (!res || !res->avail || !kref_get_unless_zero(&res->kref))
 		res = NULL;
+
 	read_unlock(&dev_priv->resource_lock);
 
 	if (unlikely(res == NULL))
@@ -1306,7 +1303,7 @@ vmw_resource_backoff_reservation(struct ttm_validate_buffer *val_buf)
  * @res:            The resource to evict.
  * @interruptible:  Whether to wait interruptible.
  */
-int vmw_resource_do_evict(struct vmw_resource *res, bool interruptible)
+static int vmw_resource_do_evict(struct vmw_resource *res, bool interruptible)
 {
 	struct ttm_validate_buffer val_buf;
 	const struct vmw_res_func *func = res->func;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c b/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c
index 8b5bc17..2af3fa1 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c
@@ -450,8 +450,8 @@ static bool vmw_sou_screen_object_flippable(struct vmw_private *dev_priv,
  * Update the implicit fb to the current fb of this crtc.
  * Must be called with the mode_config mutex held.
  */
-void vmw_sou_update_implicit_fb(struct vmw_private *dev_priv,
-				struct drm_crtc *crtc)
+static void vmw_sou_update_implicit_fb(struct vmw_private *dev_priv,
+				       struct drm_crtc *crtc)
 {
 	struct vmw_screen_object_unit *sou = vmw_crtc_to_sou(crtc);
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c b/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c
index 6110a43..11bc60c 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c
@@ -407,11 +407,11 @@ out:
 }
 
 
-struct vmw_resource *vmw_shader_alloc(struct vmw_private *dev_priv,
-				      struct vmw_dma_buffer *buffer,
-				      size_t shader_size,
-				      size_t offset,
-				      SVGA3dShaderType shader_type)
+static struct vmw_resource *vmw_shader_alloc(struct vmw_private *dev_priv,
+					     struct vmw_dma_buffer *buffer,
+					     size_t shader_size,
+					     size_t offset,
+					     SVGA3dShaderType shader_type)
 {
 	struct vmw_shader *shader;
 	struct vmw_resource *res;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c b/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
index 835f343..843d7e0 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
@@ -220,7 +220,7 @@ static void vmw_surface_define_encode(const struct vmw_surface *srf,
 	cmd->header.size = cmd_len;
 	cmd->body.sid = srf->res.id;
 	cmd->body.surfaceFlags = srf->flags;
-	cmd->body.format = cpu_to_le32(srf->format);
+	cmd->body.format = srf->format;
 	for (i = 0; i < DRM_VMW_MAX_SURFACE_FACES; ++i)
 		cmd->body.face[i].numMipLevels = srf->mip_levels[i];
 
@@ -1054,7 +1054,7 @@ static int vmw_gb_surface_create(struct vmw_resource *res)
 	cmd->header.size = cmd_len;
 	cmd->body.sid = srf->res.id;
 	cmd->body.surfaceFlags = srf->flags;
-	cmd->body.format = cpu_to_le32(srf->format);
+	cmd->body.format = srf->format;
 	cmd->body.numMipLevels = srf->mip_levels[0];
 	cmd->body.multisampleCount = srf->multisample_count;
 	cmd->body.autogenFilter = srf->autogen_filter;
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 24/28] drm/vmwgfx: Fix compiler warning with 32-bit dma_addr_t

[Thomas Hellstrom]

When the size of dma_addr_t was 32 bits, the compiler warned
about the size of the 32 bit shift being larger than the size
of the data type.

Reported by Intel's kbuild robot.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c
index 32ec52e..afc6d1d 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c
@@ -293,8 +293,12 @@ static int vmw_cmdbuf_header_submit(struct vmw_cmdbuf_header *header)
 	struct vmw_cmdbuf_man *man = header->man;
 	u32 val;
 
-	val = (header->handle >> 32);
+	if (sizeof(header->handle) > 4)
+		val = (header->handle >> 32);
+	else
+		val = 0;
 	vmw_write(man->dev_priv, SVGA_REG_COMMAND_HIGH, val);
+
 	val = (header->handle & 0xFFFFFFFFULL);
 	val |= header->cb_context & SVGA_CB_CONTEXT_MASK;
 	vmw_write(man->dev_priv, SVGA_REG_COMMAND_LOW, val);
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 25/28] drm/vmwgfx: Fix an uninitialized value

[Thomas Hellstrom]

Reported by Intel's kbuild robot.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c
index afc6d1d..5667c13 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c
@@ -507,7 +507,7 @@ static void vmw_cmdbuf_work_func(struct work_struct *work)
 	struct vmw_cmdbuf_man *man =
 		container_of(work, struct vmw_cmdbuf_man, work);
 	struct vmw_cmdbuf_header *entry, *next;
-	bool restart;
+	bool restart = false;
 
 	spin_lock_bh(&man->lock);
 	list_for_each_entry_safe(entry, next, &man->error, list) {
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 26/28] drm/vmwgfx: Fixed topology boundary checking for Screen Targets

[Thomas Hellstrom]

From: Sinclair Yeh <syeh@vmware.com>

For a Screen Target capable display device, the display topology is
limited by SVGA_REG_MAX_PRIMARY_BOUNDING_BOX_MEM.  Two values are
checked against this limit:
  1.  Size of the bounding box enclosing all the displays, and
  2.  Size of the total number of displays, e.g. framebuffers

The limitations above mean we do not have exact max width and
height for the topology.  The best current option is to set those to
the maximum texture width/height.

Signed-off-by: Sinclair Yeh <syeh@vmware.com>
Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_kms.c | 33 ++++++++++++++++++++++-----------
 1 file changed, 22 insertions(+), 11 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
index ca69ed4..8b8ae20 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
@@ -1077,9 +1077,8 @@ int vmw_kms_init(struct vmw_private *dev_priv)
 	dev->mode_config.funcs = &vmw_kms_funcs;
 	dev->mode_config.min_width = 1;
 	dev->mode_config.min_height = 1;
-	/* assumed largest fb size */
-	dev->mode_config.max_width = 8192;
-	dev->mode_config.max_height = 8192;
+	dev->mode_config.max_width = dev_priv->texture_max_width;
+	dev->mode_config.max_height = dev_priv->texture_max_height;
 
 	ret = vmw_kms_stdu_init_display(dev_priv);
 	if (ret) {
@@ -1580,6 +1579,7 @@ int vmw_kms_update_layout_ioctl(struct drm_device *dev, void *data,
 	unsigned rects_size;
 	int ret;
 	int i;
+	u64 total_pixels = 0;
 	struct drm_mode_config *mode_config = &dev->mode_config;
 	struct drm_vmw_rect bounding_box = {0};
 
@@ -1622,20 +1622,31 @@ int vmw_kms_update_layout_ioctl(struct drm_device *dev, void *data,
 
 		if (rects[i].y + rects[i].h > bounding_box.h)
 			bounding_box.h = rects[i].y + rects[i].h;
+
+		total_pixels += (u64) rects[i].w * (u64) rects[i].h;
 	}
 
-	/*
-	 * For Screen Target Display Unit, all the displays must fit
-	 * inside of maximum texture size.
-	 */
-	if (dev_priv->active_display_unit == vmw_du_screen_target)
-		if (bounding_box.w > dev_priv->texture_max_width ||
-		    bounding_box.h > dev_priv->texture_max_height) {
-			DRM_ERROR("Layout exceeds maximum texture size\n");
+	if (dev_priv->active_display_unit == vmw_du_screen_target) {
+		/*
+		 * For Screen Targets, the limits for a toplogy are:
+		 *	1. Bounding box (assuming 32bpp) must be < prim_bb_mem
+		 *      2. Total pixels (assuming 32bpp) must be < prim_bb_mem
+		 */
+		u64 bb_mem    = bounding_box.w * bounding_box.h * 4;
+		u64 pixel_mem = total_pixels * 4;
+
+		if (bb_mem > dev_priv->prim_bb_mem) {
+			DRM_ERROR("Topology is beyond supported limits.\n");
 			ret = -EINVAL;
 			goto out_free;
 		}
 
+		if (pixel_mem > dev_priv->prim_bb_mem) {
+			DRM_ERROR("Combined output size too large\n");
+			ret = -EINVAL;
+			goto out_free;
+		}
+	}
 
 	vmw_du_update_layout(dev_priv, arg->num_outputs, rects);
 
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 27/28] drm/vmwgfx: Fix framebuffer creation on older hardware

[Thomas Hellstrom]

On older hardware, texture max width and height is not available, so set
it to something reasonable, like 8192.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
index ab67d2a..bc4235f 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
@@ -707,9 +707,12 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 			  SVGA3D_DEVCAP_MAX_TEXTURE_HEIGHT);
 		dev_priv->texture_max_height = vmw_read(dev_priv,
 							SVGA_REG_DEV_CAP);
-	} else
+	} else {
+		dev_priv->texture_max_width = 8192;
+		dev_priv->texture_max_height = 8192;
 		dev_priv->prim_bb_mem = dev_priv->vram_size;
- 
+	}
+
 	vmw_print_capabilities(dev_priv->capabilities);
 
 	ret = vmw_dma_masks(dev_priv);
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 28/28] drm/vmwgfx: Fix crash when unloading vmwgfx v2

[Thomas Hellstrom]

From: Sinclair Yeh <syeh@vmware.com>

This patch fixes two issues.  One, when a surface is a proxy for a DMA
buffer, it holds an extra reference that needs to be cleared.

Two, when fbdev is enabled, we need to unpin the framebuffer before
unloading the driver.  This is done by a call to vmw_fb_off().

v2
Moved unreferencing surface to from vmw_framebuffer_surface_destroy()
to vmw_kms_new_framebuffer()

Added "struct vmw_framebuffer *vfb = NULL;" to silence a compiler
warning.

Removed error checking after calling vmw_surface/dmabuf_reference()

Signed-off-by: Sinclair Yeh <syeh@vmware.com>
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.c |  1 +
 drivers/gpu/drm/vmwgfx/vmwgfx_kms.c | 42 ++++++++++++++++---------------------
 2 files changed, 19 insertions(+), 24 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
index bc4235f..65e3565 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
@@ -915,6 +915,7 @@ static int vmw_driver_unload(struct drm_device *dev)
 		drm_ht_remove(&dev_priv->ctx.res_ht);
 	vfree(dev_priv->ctx.cmd_bounce);
 	if (dev_priv->enable_fb) {
+		vmw_fb_off(dev_priv);
 		vmw_fb_close(dev_priv);
 		vmw_fifo_resource_dec(dev_priv);
 		vmw_svga_disable(dev_priv);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
index 8b8ae20..34d04bf 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
@@ -539,19 +539,13 @@ static int vmw_kms_new_framebuffer_surface(struct vmw_private *dev_priv,
 		goto out_err1;
 	}
 
-	if (!vmw_surface_reference(surface)) {
-		DRM_ERROR("failed to reference surface %p\n", surface);
-		ret = -EINVAL;
-		goto out_err2;
-	}
-
 	/* XXX get the first 3 from the surface info */
 	vfbs->base.base.bits_per_pixel = mode_cmd->bpp;
 	vfbs->base.base.pitches[0] = mode_cmd->pitch;
 	vfbs->base.base.depth = mode_cmd->depth;
 	vfbs->base.base.width = mode_cmd->width;
 	vfbs->base.base.height = mode_cmd->height;
-	vfbs->surface = surface;
+	vfbs->surface = vmw_surface_reference(surface);
 	vfbs->base.user_handle = mode_cmd->handle;
 	vfbs->is_dmabuf_proxy = is_dmabuf_proxy;
 
@@ -560,13 +554,12 @@ static int vmw_kms_new_framebuffer_surface(struct vmw_private *dev_priv,
 	ret = drm_framebuffer_init(dev, &vfbs->base.base,
 				   &vmw_framebuffer_surface_funcs);
 	if (ret)
-		goto out_err3;
+		goto out_err2;
 
 	return 0;
 
-out_err3:
-	vmw_surface_unreference(&surface);
 out_err2:
+	vmw_surface_unreference(&surface);
 	kfree(vfbs);
 out_err1:
 	return ret;
@@ -836,32 +829,25 @@ static int vmw_kms_new_framebuffer_dmabuf(struct vmw_private *dev_priv,
 		goto out_err1;
 	}
 
-	if (!vmw_dmabuf_reference(dmabuf)) {
-		DRM_ERROR("failed to reference dmabuf %p\n", dmabuf);
-		ret = -EINVAL;
-		goto out_err2;
-	}
-
 	vfbd->base.base.bits_per_pixel = mode_cmd->bpp;
 	vfbd->base.base.pitches[0] = mode_cmd->pitch;
 	vfbd->base.base.depth = mode_cmd->depth;
 	vfbd->base.base.width = mode_cmd->width;
 	vfbd->base.base.height = mode_cmd->height;
 	vfbd->base.dmabuf = true;
-	vfbd->buffer = dmabuf;
+	vfbd->buffer = vmw_dmabuf_reference(dmabuf);
 	vfbd->base.user_handle = mode_cmd->handle;
 	*out = &vfbd->base;
 
 	ret = drm_framebuffer_init(dev, &vfbd->base.base,
 				   &vmw_framebuffer_dmabuf_funcs);
 	if (ret)
-		goto out_err3;
+		goto out_err2;
 
 	return 0;
 
-out_err3:
-	vmw_dmabuf_unreference(&dmabuf);
 out_err2:
+	vmw_dmabuf_unreference(&dmabuf);
 	kfree(vfbd);
 out_err1:
 	return ret;
@@ -886,7 +872,7 @@ vmw_kms_new_framebuffer(struct vmw_private *dev_priv,
 			bool only_2d,
 			const struct drm_mode_fb_cmd *mode_cmd)
 {
-	struct vmw_framebuffer *vfb;
+	struct vmw_framebuffer *vfb = NULL;
 	bool is_dmabuf_proxy = false;
 	int ret;
 
@@ -906,15 +892,23 @@ vmw_kms_new_framebuffer(struct vmw_private *dev_priv,
 	}
 
 	/* Create the new framebuffer depending one what we have */
-	if (surface)
+	if (surface) {
 		ret = vmw_kms_new_framebuffer_surface(dev_priv, surface, &vfb,
 						      mode_cmd,
 						      is_dmabuf_proxy);
-	else if (dmabuf)
+
+		/*
+		 * vmw_create_dmabuf_proxy() adds a reference that is no longer
+		 * needed
+		 */
+		if (is_dmabuf_proxy)
+			vmw_surface_unreference(&surface);
+	} else if (dmabuf) {
 		ret = vmw_kms_new_framebuffer_dmabuf(dev_priv, dmabuf, &vfb,
 						     mode_cmd);
-	else
+	} else {
 		BUG();
+	}
 
 	if (ret)
 		return ERR_PTR(ret);
-- 
2.1.0


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel