Merge branch 'vmwgfx-next' of git://people.freedesktop.org/~thomash/linux into drm-next

Changes for vmwgfx for 4.4. If there is time, I'll follow up with a series
to move to threaded irqs.

* 'vmwgfx-next' of git://people.freedesktop.org/~thomash/linux:
  drm/vmwgfx: Replace iowrite/ioread with volatile memory accesses
  drm/vmwgfx: Turn off support for multisample count != 0 v2
  drm/vmwgfx: switch from ioremap_cache to memremap

drivers/gpu/drm/vmwgfx/vmwgfx_drv.c

@@ -752,8 +752,8 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 	ttm_lock_set_kill(&dev_priv->fbdev_master.lock, false, SIGTERM);
 	dev_priv->active_master = &dev_priv->fbdev_master;
 
-	dev_priv->mmio_virt = ioremap_cache(dev_priv->mmio_start,
-					    dev_priv->mmio_size);
+	dev_priv->mmio_virt = memremap(dev_priv->mmio_start,
+				       dev_priv->mmio_size, MEMREMAP_WB);
 
 	if (unlikely(dev_priv->mmio_virt == NULL)) {
 		ret = -ENOMEM;
@@ -907,7 +907,7 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 out_no_device:
 	ttm_object_device_release(&dev_priv->tdev);
 out_err4:
-	iounmap(dev_priv->mmio_virt);
+	memunmap(dev_priv->mmio_virt);
 out_err3:
 	vmw_ttm_global_release(dev_priv);
 out_err0:
@@ -958,7 +958,7 @@ static int vmw_driver_unload(struct drm_device *dev)
 		pci_release_regions(dev->pdev);
 
 	ttm_object_device_release(&dev_priv->tdev);
-	iounmap(dev_priv->mmio_virt);
+	memunmap(dev_priv->mmio_virt);
 	if (dev_priv->ctx.staged_bindings)
 		vmw_binding_state_free(dev_priv->ctx.staged_bindings);
 	vmw_ttm_global_release(dev_priv);

drivers/gpu/drm/vmwgfx/vmwgfx_drv.h

@@ -375,7 +375,7 @@ struct vmw_private {
 	uint32_t stdu_max_height;
 	uint32_t initial_width;
 	uint32_t initial_height;
-	u32 __iomem *mmio_virt;
+	u32 *mmio_virt;
 	uint32_t capabilities;
 	uint32_t max_gmr_ids;
 	uint32_t max_gmr_pages;
@@ -1206,4 +1206,30 @@ static inline void vmw_fifo_resource_dec(struct vmw_private *dev_priv)
 {
 	atomic_dec(&dev_priv->num_fifo_resources);
 }
+
+/**
+ * vmw_mmio_read - Perform a MMIO read from volatile memory
+ *
+ * @addr: The address to read from
+ *
+ * This function is intended to be equivalent to ioread32() on
+ * memremap'd memory, but without byteswapping.
+ */
+static inline u32 vmw_mmio_read(u32 *addr)
+{
+	return READ_ONCE(*addr);
+}
+
+/**
+ * vmw_mmio_write - Perform a MMIO write to volatile memory
+ *
+ * @addr: The address to write to
+ *
+ * This function is intended to be equivalent to iowrite32 on
+ * memremap'd memory, but without byteswapping.
+ */
+static inline void vmw_mmio_write(u32 value, u32 *addr)
+{
+	WRITE_ONCE(*addr, value);
+}
 #endif

drivers/gpu/drm/vmwgfx/vmwgfx_fence.c

@@ -142,8 +142,8 @@ 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;
 
-	u32 __iomem *fifo_mem = dev_priv->mmio_virt;
-	u32 seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE);
+	u32 *fifo_mem = dev_priv->mmio_virt;
+	u32 seqno = vmw_mmio_read(fifo_mem + SVGA_FIFO_FENCE);
 	if (seqno - fence->base.seqno < VMW_FENCE_WRAP)
 		return false;
 
@@ -386,14 +386,14 @@ static bool vmw_fence_goal_new_locked(struct vmw_fence_manager *fman,
 				      u32 passed_seqno)
 {
 	u32 goal_seqno;
-	u32 __iomem *fifo_mem;
+	u32 *fifo_mem;
 	struct vmw_fence_obj *fence;
 
 	if (likely(!fman->seqno_valid))
 		return false;
 
 	fifo_mem = fman->dev_priv->mmio_virt;
-	goal_seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE_GOAL);
+	goal_seqno = vmw_mmio_read(fifo_mem + SVGA_FIFO_FENCE_GOAL);
 	if (likely(passed_seqno - goal_seqno >= VMW_FENCE_WRAP))
 		return false;
 
@@ -401,8 +401,8 @@ static bool vmw_fence_goal_new_locked(struct vmw_fence_manager *fman,
 	list_for_each_entry(fence, &fman->fence_list, head) {
 		if (!list_empty(&fence->seq_passed_actions)) {
 			fman->seqno_valid = true;
-			iowrite32(fence->base.seqno,
-				  fifo_mem + SVGA_FIFO_FENCE_GOAL);
+			vmw_mmio_write(fence->base.seqno,
+				       fifo_mem + SVGA_FIFO_FENCE_GOAL);
 			break;
 		}
 	}
@@ -430,18 +430,18 @@ static bool vmw_fence_goal_check_locked(struct vmw_fence_obj *fence)
 {
 	struct vmw_fence_manager *fman = fman_from_fence(fence);
 	u32 goal_seqno;
-	u32 __iomem *fifo_mem;
+	u32 *fifo_mem;
 
 	if (fence_is_signaled_locked(&fence->base))
 		return false;
 
 	fifo_mem = fman->dev_priv->mmio_virt;
-	goal_seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE_GOAL);
+	goal_seqno = vmw_mmio_read(fifo_mem + SVGA_FIFO_FENCE_GOAL);
 	if (likely(fman->seqno_valid &&
 		   goal_seqno - fence->base.seqno < VMW_FENCE_WRAP))
 		return false;
 
-	iowrite32(fence->base.seqno, fifo_mem + SVGA_FIFO_FENCE_GOAL);
+	vmw_mmio_write(fence->base.seqno, fifo_mem + SVGA_FIFO_FENCE_GOAL);
 	fman->seqno_valid = true;
 
 	return true;
@@ -453,9 +453,9 @@ static void __vmw_fences_update(struct vmw_fence_manager *fman)
 	struct list_head action_list;
 	bool needs_rerun;
 	uint32_t seqno, new_seqno;
-	u32 __iomem *fifo_mem = fman->dev_priv->mmio_virt;
+	u32 *fifo_mem = fman->dev_priv->mmio_virt;
 
-	seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE);
+	seqno = vmw_mmio_read(fifo_mem + SVGA_FIFO_FENCE);
 rerun:
 	list_for_each_entry_safe(fence, next_fence, &fman->fence_list, head) {
 		if (seqno - fence->base.seqno < VMW_FENCE_WRAP) {
@@ -477,7 +477,7 @@ static void __vmw_fences_update(struct vmw_fence_manager *fman)
 
 	needs_rerun = vmw_fence_goal_new_locked(fman, seqno);
 	if (unlikely(needs_rerun)) {
-		new_seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE);
+		new_seqno = vmw_mmio_read(fifo_mem + SVGA_FIFO_FENCE);
 		if (new_seqno != seqno) {
 			seqno = new_seqno;
 			goto rerun;

drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c

@@ -36,7 +36,7 @@ struct vmw_temp_set_context {
 
 bool vmw_fifo_have_3d(struct vmw_private *dev_priv)
 {
-	u32 __iomem *fifo_mem = dev_priv->mmio_virt;
+	u32 *fifo_mem = dev_priv->mmio_virt;
 	uint32_t fifo_min, hwversion;
 	const struct vmw_fifo_state *fifo = &dev_priv->fifo;
 
@@ -60,15 +60,15 @@ bool vmw_fifo_have_3d(struct vmw_private *dev_priv)
 	if (!(dev_priv->capabilities & SVGA_CAP_EXTENDED_FIFO))
 		return false;
 
-	fifo_min = ioread32(fifo_mem  + SVGA_FIFO_MIN);
+	fifo_min = vmw_mmio_read(fifo_mem  + SVGA_FIFO_MIN);
 	if (fifo_min <= SVGA_FIFO_3D_HWVERSION * sizeof(unsigned int))
 		return false;
 
-	hwversion = ioread32(fifo_mem +
-			     ((fifo->capabilities &
-			       SVGA_FIFO_CAP_3D_HWVERSION_REVISED) ?
-			      SVGA_FIFO_3D_HWVERSION_REVISED :
-			      SVGA_FIFO_3D_HWVERSION));
+	hwversion = vmw_mmio_read(fifo_mem +
+				  ((fifo->capabilities &
+				    SVGA_FIFO_CAP_3D_HWVERSION_REVISED) ?
+				   SVGA_FIFO_3D_HWVERSION_REVISED :
+				   SVGA_FIFO_3D_HWVERSION));
 
 	if (hwversion == 0)
 		return false;
@@ -85,13 +85,13 @@ bool vmw_fifo_have_3d(struct vmw_private *dev_priv)
 
 bool vmw_fifo_have_pitchlock(struct vmw_private *dev_priv)
 {
-	u32 __iomem *fifo_mem = dev_priv->mmio_virt;
+	u32  *fifo_mem = dev_priv->mmio_virt;
 	uint32_t caps;
 
 	if (!(dev_priv->capabilities & SVGA_CAP_EXTENDED_FIFO))
 		return false;
 
-	caps = ioread32(fifo_mem + SVGA_FIFO_CAPABILITIES);
+	caps = vmw_mmio_read(fifo_mem + SVGA_FIFO_CAPABILITIES);
 	if (caps & SVGA_FIFO_CAP_PITCHLOCK)
 		return true;
 
@@ -100,7 +100,7 @@ bool vmw_fifo_have_pitchlock(struct vmw_private *dev_priv)
 
 int vmw_fifo_init(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
 {
-	u32 __iomem *fifo_mem = dev_priv->mmio_virt;
+	u32  *fifo_mem = dev_priv->mmio_virt;
 	uint32_t max;
 	uint32_t min;
 
@@ -137,19 +137,19 @@ int vmw_fifo_init(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
 	if (min < PAGE_SIZE)
 		min = PAGE_SIZE;
 
-	iowrite32(min, fifo_mem + SVGA_FIFO_MIN);
-	iowrite32(dev_priv->mmio_size, fifo_mem + SVGA_FIFO_MAX);
+	vmw_mmio_write(min, fifo_mem + SVGA_FIFO_MIN);
+	vmw_mmio_write(dev_priv->mmio_size, fifo_mem + SVGA_FIFO_MAX);
 	wmb();
-	iowrite32(min,  fifo_mem + SVGA_FIFO_NEXT_CMD);
-	iowrite32(min,  fifo_mem + SVGA_FIFO_STOP);
-	iowrite32(0, fifo_mem + SVGA_FIFO_BUSY);
+	vmw_mmio_write(min,  fifo_mem + SVGA_FIFO_NEXT_CMD);
+	vmw_mmio_write(min,  fifo_mem + SVGA_FIFO_STOP);
+	vmw_mmio_write(0, fifo_mem + SVGA_FIFO_BUSY);
 	mb();
 
 	vmw_write(dev_priv, SVGA_REG_CONFIG_DONE, 1);
 
-	max = ioread32(fifo_mem + SVGA_FIFO_MAX);
-	min = ioread32(fifo_mem  + SVGA_FIFO_MIN);
-	fifo->capabilities = ioread32(fifo_mem + SVGA_FIFO_CAPABILITIES);
+	max = vmw_mmio_read(fifo_mem + SVGA_FIFO_MAX);
+	min = vmw_mmio_read(fifo_mem  + SVGA_FIFO_MIN);
+	fifo->capabilities = vmw_mmio_read(fifo_mem + SVGA_FIFO_CAPABILITIES);
 
 	DRM_INFO("Fifo max 0x%08x min 0x%08x cap 0x%08x\n",
 		 (unsigned int) max,
@@ -157,7 +157,7 @@ int vmw_fifo_init(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
 		 (unsigned int) fifo->capabilities);
 
 	atomic_set(&dev_priv->marker_seq, dev_priv->last_read_seqno);
-	iowrite32(dev_priv->last_read_seqno, fifo_mem + SVGA_FIFO_FENCE);
+	vmw_mmio_write(dev_priv->last_read_seqno, fifo_mem + SVGA_FIFO_FENCE);
 	vmw_marker_queue_init(&fifo->marker_queue);
 
 	return 0;
@@ -165,31 +165,23 @@ 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)
 {
-	u32 __iomem *fifo_mem = dev_priv->mmio_virt;
-	static DEFINE_SPINLOCK(ping_lock);
-	unsigned long irq_flags;
+	u32 *fifo_mem = dev_priv->mmio_virt;
 
-	/*
-	 * The ping_lock is needed because we don't have an atomic
-	 * test-and-set of the SVGA_FIFO_BUSY register.
-	 */
-	spin_lock_irqsave(&ping_lock, irq_flags);
-	if (unlikely(ioread32(fifo_mem + SVGA_FIFO_BUSY) == 0)) {
-		iowrite32(1, fifo_mem + SVGA_FIFO_BUSY);
+	preempt_disable();
+	if (cmpxchg(fifo_mem + SVGA_FIFO_BUSY, 0, 1) == 0)
 		vmw_write(dev_priv, SVGA_REG_SYNC, reason);
-	}
-	spin_unlock_irqrestore(&ping_lock, irq_flags);
+	preempt_enable();
 }
 
 void vmw_fifo_release(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
 {
-	u32 __iomem *fifo_mem = dev_priv->mmio_virt;
+	u32  *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)
 		;
 
-	dev_priv->last_read_seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE);
+	dev_priv->last_read_seqno = vmw_mmio_read(fifo_mem + SVGA_FIFO_FENCE);
 
 	vmw_write(dev_priv, SVGA_REG_CONFIG_DONE,
 		  dev_priv->config_done_state);
@@ -213,11 +205,11 @@ 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)
 {
-	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);
-	uint32_t stop = ioread32(fifo_mem + SVGA_FIFO_STOP);
+	u32  *fifo_mem = dev_priv->mmio_virt;
+	uint32_t max = vmw_mmio_read(fifo_mem + SVGA_FIFO_MAX);
+	uint32_t next_cmd = vmw_mmio_read(fifo_mem + SVGA_FIFO_NEXT_CMD);
+	uint32_t min = vmw_mmio_read(fifo_mem + SVGA_FIFO_MIN);
+	uint32_t stop = vmw_mmio_read(fifo_mem + SVGA_FIFO_STOP);
 
 	return ((max - next_cmd) + (stop - min) <= bytes);
 }
@@ -321,7 +313,7 @@ static void *vmw_local_fifo_reserve(struct vmw_private *dev_priv,
 				    uint32_t bytes)
 {
 	struct vmw_fifo_state *fifo_state = &dev_priv->fifo;
-	u32 __iomem *fifo_mem = dev_priv->mmio_virt;
+	u32  *fifo_mem = dev_priv->mmio_virt;
 	uint32_t max;
 	uint32_t min;
 	uint32_t next_cmd;
@@ -329,9 +321,9 @@ static void *vmw_local_fifo_reserve(struct vmw_private *dev_priv,
 	int ret;
 
 	mutex_lock(&fifo_state->fifo_mutex);
-	max = ioread32(fifo_mem + SVGA_FIFO_MAX);
-	min = ioread32(fifo_mem + SVGA_FIFO_MIN);
-	next_cmd = ioread32(fifo_mem + SVGA_FIFO_NEXT_CMD);
+	max = vmw_mmio_read(fifo_mem + SVGA_FIFO_MAX);
+	min = vmw_mmio_read(fifo_mem + SVGA_FIFO_MIN);
+	next_cmd = vmw_mmio_read(fifo_mem + SVGA_FIFO_NEXT_CMD);
 
 	if (unlikely(bytes >= (max - min)))
 		goto out_err;
@@ -342,7 +334,7 @@ static void *vmw_local_fifo_reserve(struct vmw_private *dev_priv,
 	fifo_state->reserved_size = bytes;
 
 	while (1) {
-		uint32_t stop = ioread32(fifo_mem + SVGA_FIFO_STOP);
+		uint32_t stop = vmw_mmio_read(fifo_mem + SVGA_FIFO_STOP);
 		bool need_bounce = false;
 		bool reserve_in_place = false;
 
@@ -376,8 +368,8 @@ static void *vmw_local_fifo_reserve(struct vmw_private *dev_priv,
 				fifo_state->using_bounce_buffer = false;
 
 				if (reserveable)
-					iowrite32(bytes, fifo_mem +
-						  SVGA_FIFO_RESERVED);
+					vmw_mmio_write(bytes, fifo_mem +
+						       SVGA_FIFO_RESERVED);
 				return (void __force *) (fifo_mem +
 							 (next_cmd >> 2));
 			} else {
@@ -427,7 +419,7 @@ void *vmw_fifo_reserve_dx(struct vmw_private *dev_priv, uint32_t bytes,
 }
 
 static void vmw_fifo_res_copy(struct vmw_fifo_state *fifo_state,
-			      u32 __iomem *fifo_mem,
+			      u32  *fifo_mem,
 			      uint32_t next_cmd,
 			      uint32_t max, uint32_t min, uint32_t bytes)
 {
@@ -439,17 +431,16 @@ static void vmw_fifo_res_copy(struct vmw_fifo_state *fifo_state,
 	if (bytes < chunk_size)
 		chunk_size = bytes;
 
-	iowrite32(bytes, fifo_mem + SVGA_FIFO_RESERVED);
+	vmw_mmio_write(bytes, fifo_mem + SVGA_FIFO_RESERVED);
 	mb();
-	memcpy_toio(fifo_mem + (next_cmd >> 2), buffer, chunk_size);
+	memcpy(fifo_mem + (next_cmd >> 2), buffer, chunk_size);
 	rest = bytes - chunk_size;
 	if (rest)
-		memcpy_toio(fifo_mem + (min >> 2), buffer + (chunk_size >> 2),
-			    rest);
+		memcpy(fifo_mem + (min >> 2), buffer + (chunk_size >> 2), rest);
 }
 
 static void vmw_fifo_slow_copy(struct vmw_fifo_state *fifo_state,
-			       u32 __iomem *fifo_mem,
+			       u32  *fifo_mem,
 			       uint32_t next_cmd,
 			       uint32_t max, uint32_t min, uint32_t bytes)
 {
@@ -457,12 +448,12 @@ static void vmw_fifo_slow_copy(struct vmw_fifo_state *fifo_state,
 	    fifo_state->dynamic_buffer : fifo_state->static_buffer;
 
 	while (bytes > 0) {
-		iowrite32(*buffer++, fifo_mem + (next_cmd >> 2));
+		vmw_mmio_write(*buffer++, fifo_mem + (next_cmd >> 2));
 		next_cmd += sizeof(uint32_t);
 		if (unlikely(next_cmd == max))
 			next_cmd = min;
 		mb();
-		iowrite32(next_cmd, fifo_mem + SVGA_FIFO_NEXT_CMD);
+		vmw_mmio_write(next_cmd, fifo_mem + SVGA_FIFO_NEXT_CMD);
 		mb();
 		bytes -= sizeof(uint32_t);
 	}
@@ -471,10 +462,10 @@ static void vmw_fifo_slow_copy(struct vmw_fifo_state *fifo_state,
 static void vmw_local_fifo_commit(struct vmw_private *dev_priv, uint32_t bytes)
 {
 	struct vmw_fifo_state *fifo_state = &dev_priv->fifo;
-	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);
+	u32  *fifo_mem = dev_priv->mmio_virt;
+	uint32_t next_cmd = vmw_mmio_read(fifo_mem + SVGA_FIFO_NEXT_CMD);
+	uint32_t max = vmw_mmio_read(fifo_mem + SVGA_FIFO_MAX);
+	uint32_t min = vmw_mmio_read(fifo_mem + SVGA_FIFO_MIN);
 	bool reserveable = fifo_state->capabilities & SVGA_FIFO_CAP_RESERVE;
 
 	if (fifo_state->dx)
@@ -507,11 +498,11 @@ static void vmw_local_fifo_commit(struct vmw_private *dev_priv, uint32_t bytes)
 		if (next_cmd >= max)
 			next_cmd -= max - min;
 		mb();
-		iowrite32(next_cmd, fifo_mem + SVGA_FIFO_NEXT_CMD);
+		vmw_mmio_write(next_cmd, fifo_mem + SVGA_FIFO_NEXT_CMD);
 	}
 
 	if (reserveable)
-		iowrite32(0, fifo_mem + SVGA_FIFO_RESERVED);
+		vmw_mmio_write(0, fifo_mem + SVGA_FIFO_RESERVED);
 	mb();
 	up_write(&fifo_state->rwsem);
 	vmw_fifo_ping_host(dev_priv, SVGA_SYNC_GENERIC);

drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c

@@ -64,7 +64,7 @@ int vmw_getparam_ioctl(struct drm_device *dev, void *data,
 		break;
 	case DRM_VMW_PARAM_FIFO_HW_VERSION:
 	{
-		u32 __iomem *fifo_mem = dev_priv->mmio_virt;
+		u32 *fifo_mem = dev_priv->mmio_virt;
 		const struct vmw_fifo_state *fifo = &dev_priv->fifo;
 
 		if ((dev_priv->capabilities & SVGA_CAP_GBOBJECTS)) {
@@ -73,11 +73,11 @@ int vmw_getparam_ioctl(struct drm_device *dev, void *data,
 		}
 
 		param->value =
-			ioread32(fifo_mem +
-				 ((fifo->capabilities &
-				   SVGA_FIFO_CAP_3D_HWVERSION_REVISED) ?
-				  SVGA_FIFO_3D_HWVERSION_REVISED :
-				  SVGA_FIFO_3D_HWVERSION));
+			vmw_mmio_read(fifo_mem +
+				      ((fifo->capabilities &
+					SVGA_FIFO_CAP_3D_HWVERSION_REVISED) ?
+				       SVGA_FIFO_3D_HWVERSION_REVISED :
+				       SVGA_FIFO_3D_HWVERSION));
 		break;
 	}
 	case DRM_VMW_PARAM_MAX_SURF_MEMORY:
@@ -122,6 +122,22 @@ int vmw_getparam_ioctl(struct drm_device *dev, void *data,
 	return 0;
 }
 
+static u32 vmw_mask_multisample(unsigned int cap, u32 fmt_value)
+{
+	/* If the header is updated, update the format test as well! */
+	BUILD_BUG_ON(SVGA3D_DEVCAP_DXFMT_BC5_UNORM + 1 != SVGA3D_DEVCAP_MAX);
+
+	if (cap >= SVGA3D_DEVCAP_DXFMT_X8R8G8B8 &&
+	    cap <= SVGA3D_DEVCAP_DXFMT_BC5_UNORM)
+		fmt_value &= ~(SVGADX_DXFMT_MULTISAMPLE_2 |
+			       SVGADX_DXFMT_MULTISAMPLE_4 |
+			       SVGADX_DXFMT_MULTISAMPLE_8);
+	else if (cap == SVGA3D_DEVCAP_MULTISAMPLE_MASKABLESAMPLES)
+		return 0;
+
+	return fmt_value;
+}
+
 static int vmw_fill_compat_cap(struct vmw_private *dev_priv, void *bounce,
 			       size_t size)
 {
@@ -147,7 +163,8 @@ static int vmw_fill_compat_cap(struct vmw_private *dev_priv, void *bounce,
 	for (i = 0; i < max_size; ++i) {
 		vmw_write(dev_priv, SVGA_REG_DEV_CAP, i);
 		compat_cap->pairs[i][0] = i;
-		compat_cap->pairs[i][1] = vmw_read(dev_priv, SVGA_REG_DEV_CAP);
+		compat_cap->pairs[i][1] = vmw_mask_multisample
+			(i, vmw_read(dev_priv, SVGA_REG_DEV_CAP));
 	}
 	spin_unlock(&dev_priv->cap_lock);
 
@@ -162,7 +179,7 @@ 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;
-	u32 __iomem *fifo_mem;
+	u32 *fifo_mem;
 	void __user *buffer = (void __user *)((unsigned long)(arg->buffer));
 	void *bounce;
 	int ret;
@@ -202,7 +219,8 @@ int vmw_get_cap_3d_ioctl(struct drm_device *dev, void *data,
 		spin_lock(&dev_priv->cap_lock);
 		for (i = 0; i < num; ++i) {
 			vmw_write(dev_priv, SVGA_REG_DEV_CAP, i);
-			*bounce32++ = vmw_read(dev_priv, SVGA_REG_DEV_CAP);
+			*bounce32++ = vmw_mask_multisample
+				(i, vmw_read(dev_priv, SVGA_REG_DEV_CAP));
 		}
 		spin_unlock(&dev_priv->cap_lock);
 	} else if (gb_objects) {
@@ -211,7 +229,7 @@ int vmw_get_cap_3d_ioctl(struct drm_device *dev, void *data,
 			goto out_err;
 	} else {
 		fifo_mem = dev_priv->mmio_virt;
-		memcpy_fromio(bounce, &fifo_mem[SVGA_FIFO_3D_CAPS], size);
+		memcpy(bounce, &fifo_mem[SVGA_FIFO_3D_CAPS], size);
 	}
 
 	ret = copy_to_user(buffer, bounce, size);

drivers/gpu/drm/vmwgfx/vmwgfx_irq.c

@@ -72,8 +72,8 @@ 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)
 {
-	u32 __iomem *fifo_mem = dev_priv->mmio_virt;
-	uint32_t seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE);
+	u32 *fifo_mem = dev_priv->mmio_virt;
+	uint32_t seqno = vmw_mmio_read(fifo_mem + SVGA_FIFO_FENCE);
 
 	if (dev_priv->last_read_seqno != seqno) {
 		dev_priv->last_read_seqno = seqno;
@@ -178,8 +178,9 @@ int vmw_fallback_wait(struct vmw_private *dev_priv,
 	}
 	finish_wait(&dev_priv->fence_queue, &__wait);
 	if (ret == 0 && fifo_idle) {
-		u32 __iomem *fifo_mem = dev_priv->mmio_virt;
-		iowrite32(signal_seq, fifo_mem + SVGA_FIFO_FENCE);
+		u32 *fifo_mem = dev_priv->mmio_virt;
+
+		vmw_mmio_write(signal_seq, fifo_mem + SVGA_FIFO_FENCE);
 	}
 	wake_up_all(&dev_priv->fence_queue);
 out_err:

drivers/gpu/drm/vmwgfx/vmwgfx_kms.c

@@ -123,14 +123,14 @@ int vmw_cursor_update_dmabuf(struct vmw_private *dev_priv,
 void vmw_cursor_update_position(struct vmw_private *dev_priv,
 				bool show, int x, int y)
 {
-	u32 __iomem *fifo_mem = dev_priv->mmio_virt;
+	u32 *fifo_mem = dev_priv->mmio_virt;
 	uint32_t count;
 
-	iowrite32(show ? 1 : 0, fifo_mem + SVGA_FIFO_CURSOR_ON);
-	iowrite32(x, fifo_mem + SVGA_FIFO_CURSOR_X);
-	iowrite32(y, fifo_mem + SVGA_FIFO_CURSOR_Y);
-	count = ioread32(fifo_mem + SVGA_FIFO_CURSOR_COUNT);
-	iowrite32(++count, fifo_mem + SVGA_FIFO_CURSOR_COUNT);
+	vmw_mmio_write(show ? 1 : 0, fifo_mem + SVGA_FIFO_CURSOR_ON);
+	vmw_mmio_write(x, fifo_mem + SVGA_FIFO_CURSOR_X);
+	vmw_mmio_write(y, fifo_mem + SVGA_FIFO_CURSOR_Y);
+	count = vmw_mmio_read(fifo_mem + SVGA_FIFO_CURSOR_COUNT);
+	vmw_mmio_write(++count, fifo_mem + SVGA_FIFO_CURSOR_COUNT);
 }
 
 int vmw_du_crtc_cursor_set(struct drm_crtc *crtc, struct drm_file *file_priv,
@@ -1155,7 +1155,8 @@ int vmw_kms_write_svga(struct vmw_private *vmw_priv,
 	if (vmw_priv->capabilities & SVGA_CAP_PITCHLOCK)
 		vmw_write(vmw_priv, SVGA_REG_PITCHLOCK, pitch);
 	else if (vmw_fifo_have_pitchlock(vmw_priv))
-		iowrite32(pitch, vmw_priv->mmio_virt + SVGA_FIFO_PITCHLOCK);
+		vmw_mmio_write(pitch, vmw_priv->mmio_virt +
+			       SVGA_FIFO_PITCHLOCK);
 	vmw_write(vmw_priv, SVGA_REG_WIDTH, width);
 	vmw_write(vmw_priv, SVGA_REG_HEIGHT, height);
 	vmw_write(vmw_priv, SVGA_REG_BITS_PER_PIXEL, bpp);
@@ -1181,8 +1182,8 @@ int vmw_kms_save_vga(struct vmw_private *vmw_priv)
 		vmw_priv->vga_pitchlock =
 		  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);
+		vmw_priv->vga_pitchlock = vmw_mmio_read(vmw_priv->mmio_virt +
+							SVGA_FIFO_PITCHLOCK);
 
 	if (!(vmw_priv->capabilities & SVGA_CAP_DISPLAY_TOPOLOGY))
 		return 0;
@@ -1230,8 +1231,8 @@ int vmw_kms_restore_vga(struct vmw_private *vmw_priv)
 		vmw_write(vmw_priv, SVGA_REG_PITCHLOCK,
 			  vmw_priv->vga_pitchlock);
 	else if (vmw_fifo_have_pitchlock(vmw_priv))
-		iowrite32(vmw_priv->vga_pitchlock,
-			  vmw_priv->mmio_virt + SVGA_FIFO_PITCHLOCK);
+		vmw_mmio_write(vmw_priv->vga_pitchlock,
+			       vmw_priv->mmio_virt + SVGA_FIFO_PITCHLOCK);
 
 	if (!(vmw_priv->capabilities & SVGA_CAP_DISPLAY_TOPOLOGY))
 		return 0;

drivers/gpu/drm/vmwgfx/vmwgfx_surface.c

@@ -1290,6 +1290,8 @@ int vmw_gb_surface_define_ioctl(struct drm_device *dev, void *data,
 	uint32_t size;
 	uint32_t backup_handle;
 
+	if (req->multisample_count != 0)
+		return -EINVAL;
 
 	if (unlikely(vmw_user_surface_size == 0))
 		vmw_user_surface_size = ttm_round_pot(sizeof(*user_srf)) +