drm: merge Linux master into HEAD

Conflicts:
	drivers/gpu/drm/drm_info.c
	drivers/gpu/drm/drm_proc.c
	drivers/gpu/drm/i915/i915_gem_debugfs.c

14 files changed, 1375 insertions, 353 deletions

diff --git a/drivers/gpu/drm/drm_fops.c b/drivers/gpu/drm/drm_fops.c
index f52663ebe01..e13cb62bbae 100644
--- a/drivers/gpu/drm/drm_fops.c
+++ b/drivers/gpu/drm/drm_fops.c
@@ -337,14 +337,10 @@ int drm_fasync(int fd, struct file *filp, int on)
 {
 	struct drm_file *priv = filp->private_data;
 	struct drm_device *dev = priv->minor->dev;
-	int retcode;
 
 	DRM_DEBUG("fd = %d, device = 0x%lx\n", fd,
 		  (long)old_encode_dev(priv->minor->device));
-	retcode = fasync_helper(fd, filp, on, &dev->buf_async);
-	if (retcode < 0)
-		return retcode;
-	return 0;
+	return fasync_helper(fd, filp, on, &dev->buf_async);
 }
 EXPORT_SYMBOL(drm_fasync);
 
diff --git a/drivers/gpu/drm/drm_info.c b/drivers/gpu/drm/drm_info.c
index 60a1b6cb376..f0f6c6b93f3 100644
--- a/drivers/gpu/drm/drm_info.c
+++ b/drivers/gpu/drm/drm_info.c
@@ -286,9 +286,9 @@ int drm_vma_info(struct seq_file *m, void *data)
 #endif
 
 	mutex_lock(&dev->struct_mutex);
-	seq_printf(m, "vma use count: %d, high_memory = %p, 0x%08lx\n",
+	seq_printf(m, "vma use count: %d, high_memory = %p, 0x%08llx\n",
 		   atomic_read(&dev->vma_count),
-		   high_memory, virt_to_phys(high_memory));
+		   high_memory, (u64)virt_to_phys(high_memory));
 
 	list_for_each_entry(pt, &dev->vmalist, head) {
 		vma = pt->vma;
diff --git a/drivers/gpu/drm/drm_sysfs.c b/drivers/gpu/drm/drm_sysfs.c
index f7510a8f0eb..5de573a981c 100644
--- a/drivers/gpu/drm/drm_sysfs.c
+++ b/drivers/gpu/drm/drm_sysfs.c
@@ -349,8 +349,8 @@ int drm_sysfs_connector_add(struct drm_connector *connector)
 	DRM_DEBUG("adding \"%s\" to sysfs\n",
 		  drm_get_connector_name(connector));
 
-	snprintf(connector->kdev.bus_id, BUS_ID_SIZE, "card%d-%s",
-		 dev->primary->index, drm_get_connector_name(connector));
+	dev_set_name(&connector->kdev, "card%d-%s",
+		     dev->primary->index, drm_get_connector_name(connector));
 	ret = device_register(&connector->kdev);
 
 	if (ret) {
diff --git a/drivers/gpu/drm/i915/i915_dma.c b/drivers/gpu/drm/i915/i915_dma.c
index 4d9f5c6818c..85549f615b1 100644
--- a/drivers/gpu/drm/i915/i915_dma.c
+++ b/drivers/gpu/drm/i915/i915_dma.c
@@ -41,7 +41,6 @@
 int i915_wait_ring(struct drm_device * dev, int n, const char *caller)
 {
 	drm_i915_private_t *dev_priv = dev->dev_private;
-	struct drm_i915_master_private *master_priv = dev->primary->master->driver_priv;
 	drm_i915_ring_buffer_t *ring = &(dev_priv->ring);
 	u32 acthd_reg = IS_I965G(dev) ? ACTHD_I965 : ACTHD;
 	u32 last_acthd = I915_READ(acthd_reg);
@@ -58,8 +57,12 @@ int i915_wait_ring(struct drm_device * dev, int n, const char *caller)
 		if (ring->space >= n)
 			return 0;
 
-		if (master_priv->sarea_priv)
-			master_priv->sarea_priv->perf_boxes |= I915_BOX_WAIT;
+		if (dev->primary->master) {
+			struct drm_i915_master_private *master_priv = dev->primary->master->driver_priv;
+			if (master_priv->sarea_priv)
+				master_priv->sarea_priv->perf_boxes |= I915_BOX_WAIT;
+		}
+
 
 		if (ring->head != last_head)
 			i = 0;
@@ -356,7 +359,7 @@ static int validate_cmd(int cmd)
 	return ret;
 }
 
-static int i915_emit_cmds(struct drm_device * dev, int __user * buffer, int dwords)
+static int i915_emit_cmds(struct drm_device * dev, int *buffer, int dwords)
 {
 	drm_i915_private_t *dev_priv = dev->dev_private;
 	int i;
@@ -370,8 +373,7 @@ static int i915_emit_cmds(struct drm_device * dev, int __user * buffer, int dwor
 	for (i = 0; i < dwords;) {
 		int cmd, sz;
 
-		if (DRM_COPY_FROM_USER_UNCHECKED(&cmd, &buffer[i], sizeof(cmd)))
-			return -EINVAL;
+		cmd = buffer[i];
 
 		if ((sz = validate_cmd(cmd)) == 0 || i + sz > dwords)
 			return -EINVAL;
@@ -379,11 +381,7 @@ static int i915_emit_cmds(struct drm_device * dev, int __user * buffer, int dwor
 		OUT_RING(cmd);
 
 		while (++i, --sz) {
-			if (DRM_COPY_FROM_USER_UNCHECKED(&cmd, &buffer[i],
-							 sizeof(cmd))) {
-				return -EINVAL;
-			}
-			OUT_RING(cmd);
+			OUT_RING(buffer[i]);
 		}
 	}
 
@@ -397,17 +395,13 @@ static int i915_emit_cmds(struct drm_device * dev, int __user * buffer, int dwor
 
 int
 i915_emit_box(struct drm_device *dev,
-	      struct drm_clip_rect __user *boxes,
+	      struct drm_clip_rect *boxes,
 	      int i, int DR1, int DR4)
 {
 	drm_i915_private_t *dev_priv = dev->dev_private;
-	struct drm_clip_rect box;
+	struct drm_clip_rect box = boxes[i];
 	RING_LOCALS;
 
-	if (DRM_COPY_FROM_USER_UNCHECKED(&box, &boxes[i], sizeof(box))) {
-		return -EFAULT;
-	}
-
 	if (box.y2 <= box.y1 || box.x2 <= box.x1 || box.y2 <= 0 || box.x2 <= 0) {
 		DRM_ERROR("Bad box %d,%d..%d,%d\n",
 			  box.x1, box.y1, box.x2, box.y2);
@@ -460,7 +454,9 @@ static void i915_emit_breadcrumb(struct drm_device *dev)
 }
 
 static int i915_dispatch_cmdbuffer(struct drm_device * dev,
-				   drm_i915_cmdbuffer_t * cmd)
+				   drm_i915_cmdbuffer_t *cmd,
+				   struct drm_clip_rect *cliprects,
+				   void *cmdbuf)
 {
 	int nbox = cmd->num_cliprects;
 	int i = 0, count, ret;
@@ -476,13 +472,13 @@ static int i915_dispatch_cmdbuffer(struct drm_device * dev,
 
 	for (i = 0; i < count; i++) {
 		if (i < nbox) {
-			ret = i915_emit_box(dev, cmd->cliprects, i,
+			ret = i915_emit_box(dev, cliprects, i,
 					    cmd->DR1, cmd->DR4);
 			if (ret)
 				return ret;
 		}
 
-		ret = i915_emit_cmds(dev, (int __user *)cmd->buf, cmd->sz / 4);
+		ret = i915_emit_cmds(dev, cmdbuf, cmd->sz / 4);
 		if (ret)
 			return ret;
 	}
@@ -492,10 +488,10 @@ static int i915_dispatch_cmdbuffer(struct drm_device * dev,
 }
 
 static int i915_dispatch_batchbuffer(struct drm_device * dev,
-				     drm_i915_batchbuffer_t * batch)
+				     drm_i915_batchbuffer_t * batch,
+				     struct drm_clip_rect *cliprects)
 {
 	drm_i915_private_t *dev_priv = dev->dev_private;
-	struct drm_clip_rect __user *boxes = batch->cliprects;
 	int nbox = batch->num_cliprects;
 	int i = 0, count;
 	RING_LOCALS;
@@ -511,7 +507,7 @@ static int i915_dispatch_batchbuffer(struct drm_device * dev,
 
 	for (i = 0; i < count; i++) {
 		if (i < nbox) {
-			int ret = i915_emit_box(dev, boxes, i,
+			int ret = i915_emit_box(dev, cliprects, i,
 						batch->DR1, batch->DR4);
 			if (ret)
 				return ret;
@@ -626,6 +622,7 @@ static int i915_batchbuffer(struct drm_device *dev, void *data,
 	    master_priv->sarea_priv;
 	drm_i915_batchbuffer_t *batch = data;
 	int ret;
+	struct drm_clip_rect *cliprects = NULL;
 
 	if (!dev_priv->allow_batchbuffer) {
 		DRM_ERROR("Batchbuffer ioctl disabled\n");
@@ -637,17 +634,35 @@ static int i915_batchbuffer(struct drm_device *dev, void *data,
 
 	RING_LOCK_TEST_WITH_RETURN(dev, file_priv);
 
-	if (batch->num_cliprects && DRM_VERIFYAREA_READ(batch->cliprects,
-						       batch->num_cliprects *
-						       sizeof(struct drm_clip_rect)))
-		return -EFAULT;
+	if (batch->num_cliprects < 0)
+		return -EINVAL;
+
+	if (batch->num_cliprects) {
+		cliprects = drm_calloc(batch->num_cliprects,
+				       sizeof(struct drm_clip_rect),
+				       DRM_MEM_DRIVER);
+		if (cliprects == NULL)
+			return -ENOMEM;
+
+		ret = copy_from_user(cliprects, batch->cliprects,
+				     batch->num_cliprects *
+				     sizeof(struct drm_clip_rect));
+		if (ret != 0)
+			goto fail_free;
+	}
 
 	mutex_lock(&dev->struct_mutex);
-	ret = i915_dispatch_batchbuffer(dev, batch);
+	ret = i915_dispatch_batchbuffer(dev, batch, cliprects);
 	mutex_unlock(&dev->struct_mutex);
 
 	if (sarea_priv)
 		sarea_priv->last_dispatch = READ_BREADCRUMB(dev_priv);
+
+fail_free:
+	drm_free(cliprects,
+		 batch->num_cliprects * sizeof(struct drm_clip_rect),
+		 DRM_MEM_DRIVER);
+
 	return ret;
 }
 
@@ -659,6 +674,8 @@ static int i915_cmdbuffer(struct drm_device *dev, void *data,
 	drm_i915_sarea_t *sarea_priv = (drm_i915_sarea_t *)
 	    master_priv->sarea_priv;
 	drm_i915_cmdbuffer_t *cmdbuf = data;
+	struct drm_clip_rect *cliprects = NULL;
+	void *batch_data;
 	int ret;
 
 	DRM_DEBUG("i915 cmdbuffer, buf %p sz %d cliprects %d\n",
@@ -666,25 +683,50 @@ static int i915_cmdbuffer(struct drm_device *dev, void *data,
 
 	RING_LOCK_TEST_WITH_RETURN(dev, file_priv);
 
-	if (cmdbuf->num_cliprects &&
-	    DRM_VERIFYAREA_READ(cmdbuf->cliprects,
-				cmdbuf->num_cliprects *
-				sizeof(struct drm_clip_rect))) {
-		DRM_ERROR("Fault accessing cliprects\n");
-		return -EFAULT;
+	if (cmdbuf->num_cliprects < 0)
+		return -EINVAL;
+
+	batch_data = drm_alloc(cmdbuf->sz, DRM_MEM_DRIVER);
+	if (batch_data == NULL)
+		return -ENOMEM;
+
+	ret = copy_from_user(batch_data, cmdbuf->buf, cmdbuf->sz);
+	if (ret != 0)
+		goto fail_batch_free;
+
+	if (cmdbuf->num_cliprects) {
+		cliprects = drm_calloc(cmdbuf->num_cliprects,
+				       sizeof(struct drm_clip_rect),
+				       DRM_MEM_DRIVER);
+		if (cliprects == NULL)
+			goto fail_batch_free;
+
+		ret = copy_from_user(cliprects, cmdbuf->cliprects,
+				     cmdbuf->num_cliprects *
+				     sizeof(struct drm_clip_rect));
+		if (ret != 0)
+			goto fail_clip_free;
 	}
 
 	mutex_lock(&dev->struct_mutex);
-	ret = i915_dispatch_cmdbuffer(dev, cmdbuf);
+	ret = i915_dispatch_cmdbuffer(dev, cmdbuf, cliprects, batch_data);
 	mutex_unlock(&dev->struct_mutex);
 	if (ret) {
 		DRM_ERROR("i915_dispatch_cmdbuffer failed\n");
-		return ret;
+		goto fail_batch_free;
 	}
 
 	if (sarea_priv)
 		sarea_priv->last_dispatch = READ_BREADCRUMB(dev_priv);
-	return 0;
+
+fail_batch_free:
+	drm_free(batch_data, cmdbuf->sz, DRM_MEM_DRIVER);
+fail_clip_free:
+	drm_free(cliprects,
+		 cmdbuf->num_cliprects * sizeof(struct drm_clip_rect),
+		 DRM_MEM_DRIVER);
+
+	return ret;
 }
 
 static int i915_flip_bufs(struct drm_device *dev, void *data,
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 1bc45a78ffc..c1685d0c704 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -404,7 +404,8 @@ struct drm_i915_gem_object {
 	/** AGP memory structure for our GTT binding. */
 	DRM_AGP_MEM *agp_mem;
 
-	struct page **page_list;
+	struct page **pages;
+	int pages_refcount;
 
 	/**
 	 * Current offset of the object in GTT space.
@@ -519,7 +520,7 @@ extern int i915_driver_device_is_agp(struct drm_device * dev);
 extern long i915_compat_ioctl(struct file *filp, unsigned int cmd,
 			      unsigned long arg);
 extern int i915_emit_box(struct drm_device *dev,
-			 struct drm_clip_rect __user *boxes,
+			 struct drm_clip_rect *boxes,
 			 int i, int DR1, int DR4);
 
 /* i915_irq.c */
@@ -786,15 +787,21 @@ extern int i915_wait_ring(struct drm_device * dev, int n, const char *caller);
 		     (dev)->pci_device == 0x2E22 || \
 		     IS_GM45(dev))
 
+#define IS_IGDG(dev) ((dev)->pci_device == 0xa001)
+#define IS_IGDGM(dev) ((dev)->pci_device == 0xa011)
+#define IS_IGD(dev) (IS_IGDG(dev) || IS_IGDGM(dev))
+
 #define IS_G33(dev)    ((dev)->pci_device == 0x29C2 ||	\
 			(dev)->pci_device == 0x29B2 ||	\
-			(dev)->pci_device == 0x29D2)
+			(dev)->pci_device == 0x29D2 ||  \
+			(IS_IGD(dev)))
 
 #define IS_I9XX(dev) (IS_I915G(dev) || IS_I915GM(dev) || IS_I945G(dev) || \
 		      IS_I945GM(dev) || IS_I965G(dev) || IS_G33(dev))
 
 #define IS_MOBILE(dev) (IS_I830(dev) || IS_I85X(dev) || IS_I915GM(dev) || \
-			IS_I945GM(dev) || IS_I965GM(dev) || IS_GM45(dev))
+			IS_I945GM(dev) || IS_I965GM(dev) || IS_GM45(dev) || \
+			IS_IGD(dev))
 
 #define I915_NEED_GFX_HWS(dev) (IS_G33(dev) || IS_GM45(dev) || IS_G4X(dev))
 /* With the 945 and later, Y tiling got adjusted so that it was 32 128-byte
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 8d5ec5fd525..e5d2bdf2cc9 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -43,8 +43,8 @@ static int i915_gem_object_set_cpu_read_domain_range(struct drm_gem_object *obj,
 						     uint64_t offset,
 						     uint64_t size);
 static void i915_gem_object_set_to_full_cpu_read_domain(struct drm_gem_object *obj);
-static int i915_gem_object_get_page_list(struct drm_gem_object *obj);
-static void i915_gem_object_free_page_list(struct drm_gem_object *obj);
+static int i915_gem_object_get_pages(struct drm_gem_object *obj);
+static void i915_gem_object_put_pages(struct drm_gem_object *obj);
 static int i915_gem_object_wait_rendering(struct drm_gem_object *obj);
 static int i915_gem_object_bind_to_gtt(struct drm_gem_object *obj,
 					   unsigned alignment);
@@ -136,6 +136,224 @@ i915_gem_create_ioctl(struct drm_device *dev, void *data,
 	return 0;
 }
 
+static inline int
+fast_shmem_read(struct page **pages,
+		loff_t page_base, int page_offset,
+		char __user *data,
+		int length)
+{
+	char __iomem *vaddr;
+	int ret;
+
+	vaddr = kmap_atomic(pages[page_base >> PAGE_SHIFT], KM_USER0);
+	if (vaddr == NULL)
+		return -ENOMEM;
+	ret = __copy_to_user_inatomic(data, vaddr + page_offset, length);
+	kunmap_atomic(vaddr, KM_USER0);
+
+	return ret;
+}
+
+static inline int
+slow_shmem_copy(struct page *dst_page,
+		int dst_offset,
+		struct page *src_page,
+		int src_offset,
+		int length)
+{
+	char *dst_vaddr, *src_vaddr;
+
+	dst_vaddr = kmap_atomic(dst_page, KM_USER0);
+	if (dst_vaddr == NULL)
+		return -ENOMEM;
+
+	src_vaddr = kmap_atomic(src_page, KM_USER1);
+	if (src_vaddr == NULL) {
+		kunmap_atomic(dst_vaddr, KM_USER0);
+		return -ENOMEM;
+	}
+
+	memcpy(dst_vaddr + dst_offset, src_vaddr + src_offset, length);
+
+	kunmap_atomic(src_vaddr, KM_USER1);
+	kunmap_atomic(dst_vaddr, KM_USER0);
+
+	return 0;
+}
+
+/**
+ * This is the fast shmem pread path, which attempts to copy_from_user directly
+ * from the backing pages of the object to the user's address space.  On a
+ * fault, it fails so we can fall back to i915_gem_shmem_pwrite_slow().
+ */
+static int
+i915_gem_shmem_pread_fast(struct drm_device *dev, struct drm_gem_object *obj,
+			  struct drm_i915_gem_pread *args,
+			  struct drm_file *file_priv)
+{
+	struct drm_i915_gem_object *obj_priv = obj->driver_private;
+	ssize_t remain;
+	loff_t offset, page_base;
+	char __user *user_data;
+	int page_offset, page_length;
+	int ret;
+
+	user_data = (char __user *) (uintptr_t) args->data_ptr;
+	remain = args->size;
+
+	mutex_lock(&dev->struct_mutex);
+
+	ret = i915_gem_object_get_pages(obj);
+	if (ret != 0)
+		goto fail_unlock;
+
+	ret = i915_gem_object_set_cpu_read_domain_range(obj, args->offset,
+							args->size);
+	if (ret != 0)
+		goto fail_put_pages;
+
+	obj_priv = obj->driver_private;
+	offset = args->offset;
+
+	while (remain > 0) {
+		/* Operation in this page
+		 *
+		 * page_base = page offset within aperture
+		 * page_offset = offset within page
+		 * page_length = bytes to copy for this page
+		 */
+		page_base = (offset & ~(PAGE_SIZE-1));
+		page_offset = offset & (PAGE_SIZE-1);
+		page_length = remain;
+		if ((page_offset + remain) > PAGE_SIZE)
+			page_length = PAGE_SIZE - page_offset;
+
+		ret = fast_shmem_read(obj_priv->pages,
+				      page_base, page_offset,
+				      user_data, page_length);
+		if (ret)
+			goto fail_put_pages;
+
+		remain -= page_length;
+		user_data += page_length;
+		offset += page_length;
+	}
+
+fail_put_pages:
+	i915_gem_object_put_pages(obj);
+fail_unlock:
+	mutex_unlock(&dev->struct_mutex);
+
+	return ret;
+}
+
+/**
+ * This is the fallback shmem pread path, which allocates temporary storage
+ * in kernel space to copy_to_user into outside of the struct_mutex, so we
+ * can copy out of the object's backing pages while holding the struct mutex
+ * and not take page faults.
+ */
+static int
+i915_gem_shmem_pread_slow(struct drm_device *dev, struct drm_gem_object *obj,
+			  struct drm_i915_gem_pread *args,
+			  struct drm_file *file_priv)
+{
+	struct drm_i915_gem_object *obj_priv = obj->driver_private;
+	struct mm_struct *mm = current->mm;
+	struct page **user_pages;
+	ssize_t remain;
+	loff_t offset, pinned_pages, i;
+	loff_t first_data_page, last_data_page, num_pages;
+	int shmem_page_index, shmem_page_offset;
+	int data_page_index,  data_page_offset;
+	int page_length;
+	int ret;
+	uint64_t data_ptr = args->data_ptr;
+
+	remain = args->size;
+
+	/* Pin the user pages containing the data.  We can't fault while
+	 * holding the struct mutex, yet we want to hold it while
+	 * dereferencing the user data.
+	 */
+	first_data_page = data_ptr / PAGE_SIZE;
+	last_data_page = (data_ptr + args->size - 1) / PAGE_SIZE;
+	num_pages = last_data_page - first_data_page + 1;
+
+	user_pages = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
+	if (user_pages == NULL)
+		return -ENOMEM;
+
+	down_read(&mm->mmap_sem);
+	pinned_pages = get_user_pages(current, mm, (uintptr_t)args->data_ptr,
+				      num_pages, 0, 0, user_pages, NULL);
+	up_read(&mm->mmap_sem);
+	if (pinned_pages < num_pages) {
+		ret = -EFAULT;
+		goto fail_put_user_pages;
+	}
+
+	mutex_lock(&dev->struct_mutex);
+
+	ret = i915_gem_object_get_pages(obj);
+	if (ret != 0)
+		goto fail_unlock;
+
+	ret = i915_gem_object_set_cpu_read_domain_range(obj, args->offset,
+							args->size);
+	if (ret != 0)
+		goto fail_put_pages;
+
+	obj_priv = obj->driver_private;
+	offset = args->offset;
+
+	while (remain > 0) {
+		/* Operation in this page
+		 *
+		 * shmem_page_index = page number within shmem file
+		 * shmem_page_offset = offset within page in shmem file
+		 * data_page_index = page number in get_user_pages return
+		 * data_page_offset = offset with data_page_index page.
+		 * page_length = bytes to copy for this page
+		 */
+		shmem_page_index = offset / PAGE_SIZE;
+		shmem_page_offset = offset & ~PAGE_MASK;
+		data_page_index = data_ptr / PAGE_SIZE - first_data_page;
+		data_page_offset = data_ptr & ~PAGE_MASK;
+
+		page_length = remain;
+		if ((shmem_page_offset + page_length) > PAGE_SIZE)
+			page_length = PAGE_SIZE - shmem_page_offset;
+		if ((data_page_offset + page_length) > PAGE_SIZE)
+			page_length = PAGE_SIZE - data_page_offset;
+
+		ret = slow_shmem_copy(user_pages[data_page_index],
+				      data_page_offset,
+				      obj_priv->pages[shmem_page_index],
+				      shmem_page_offset,
+				      page_length);
+		if (ret)
+			goto fail_put_pages;
+
+		remain -= page_length;
+		data_ptr += page_length;
+		offset += page_length;
+	}
+
+fail_put_pages:
+	i915_gem_object_put_pages(obj);
+fail_unlock:
+	mutex_unlock(&dev->struct_mutex);
+fail_put_user_pages:
+	for (i = 0; i < pinned_pages; i++) {
+		SetPageDirty(user_pages[i]);
+		page_cache_release(user_pages[i]);
+	}
+	kfree(user_pages);
+
+	return ret;
+}
+
 /**
  * Reads data from the object referenced by handle.
  *
@@ -148,8 +366,6 @@ i915_gem_pread_ioctl(struct drm_device *dev, void *data,
 	struct drm_i915_gem_pread *args = data;
 	struct drm_gem_object *obj;
 	struct drm_i915_gem_object *obj_priv;
-	ssize_t read;
-	loff_t offset;
 	int ret;
 
 	obj = drm_gem_object_lookup(dev, file_priv, args->handle);
@@ -167,33 +383,13 @@ i915_gem_pread_ioctl(struct drm_device *dev, void *data,
 		return -EINVAL;
 	}
 
-	mutex_lock(&dev->struct_mutex);
-
-	ret = i915_gem_object_set_cpu_read_domain_range(obj, args->offset,
-							args->size);
-	if (ret != 0) {
-		drm_gem_object_unreference(obj);
-		mutex_unlock(&dev->struct_mutex);
-		return ret;
-	}
-
-	offset = args->offset;
-
-	read = vfs_read(obj->filp, (char __user *)(uintptr_t)args->data_ptr,
-			args->size, &offset);
-	if (read != args->size) {
-		drm_gem_object_unreference(obj);
-		mutex_unlock(&dev->struct_mutex);
-		if (read < 0)
-			return read;
-		else
-			return -EINVAL;
-	}
+	ret = i915_gem_shmem_pread_fast(dev, obj, args, file_priv);
+	if (ret != 0)
+		ret = i915_gem_shmem_pread_slow(dev, obj, args, file_priv);
 
 	drm_gem_object_unreference(obj);
-	mutex_unlock(&dev->struct_mutex);
 
-	return 0;
+	return ret;
 }
 
 /* This is the fast write path which cannot handle
@@ -223,29 +419,51 @@ fast_user_write(struct io_mapping *mapping,
  */
 
 static inline int
-slow_user_write(struct io_mapping *mapping,
-		loff_t page_base, int page_offset,
-		char __user *user_data,
-		int length)
+slow_kernel_write(struct io_mapping *mapping,
+		  loff_t gtt_base, int gtt_offset,
+		  struct page *user_page, int user_offset,
+		  int length)
 {
-	char __iomem *vaddr;
+	char *src_vaddr, *dst_vaddr;
 	unsigned long unwritten;
 
-	vaddr = io_mapping_map_wc(mapping, page_base);
-	if (vaddr == NULL)
-		return -EFAULT;
-	unwritten = __copy_from_user(vaddr + page_offset,
-				     user_data, length);
-	io_mapping_unmap(vaddr);
+	dst_vaddr = io_mapping_map_atomic_wc(mapping, gtt_base);
+	src_vaddr = kmap_atomic(user_page, KM_USER1);
+	unwritten = __copy_from_user_inatomic_nocache(dst_vaddr + gtt_offset,
+						      src_vaddr + user_offset,
+						      length);
+	kunmap_atomic(src_vaddr, KM_USER1);
+	io_mapping_unmap_atomic(dst_vaddr);
 	if (unwritten)
 		return -EFAULT;
 	return 0;
 }
 
+static inline int
+fast_shmem_write(struct page **pages,
+		 loff_t page_base, int page_offset,
+		 char __user *data,
+		 int length)
+{
+	char __iomem *vaddr;
+
+	vaddr = kmap_atomic(pages[page_base >> PAGE_SHIFT], KM_USER0);
+	if (vaddr == NULL)
+		return -ENOMEM;
+	__copy_from_user_inatomic(vaddr + page_offset, data, length);
+	kunmap_atomic(vaddr, KM_USER0);
+
+	return 0;
+}
+
+/**
+ * This is the fast pwrite path, where we copy the data directly from the
+ * user into the GTT, uncached.
+ */
 static int
-i915_gem_gtt_pwrite(struct drm_device *dev, struct drm_gem_object *obj,
-		    struct drm_i915_gem_pwrite *args,
-		    struct drm_file *file_priv)
+i915_gem_gtt_pwrite_fast(struct drm_device *dev, struct drm_gem_object *obj,
+			 struct drm_i915_gem_pwrite *args,
+			 struct drm_file *file_priv)
 {
 	struct drm_i915_gem_object *obj_priv = obj->driver_private;
 	drm_i915_private_t *dev_priv = dev->dev_private;
@@ -273,7 +491,6 @@ i915_gem_gtt_pwrite(struct drm_device *dev, struct drm_gem_object *obj,
 
 	obj_priv = obj->driver_private;
 	offset = obj_priv->gtt_offset + args->offset;
-	obj_priv->dirty = 1;
 
 	while (remain > 0) {
 		/* Operation in this page
@@ -292,16 +509,11 @@ i915_gem_gtt_pwrite(struct drm_device *dev, struct drm_gem_object *obj,
 				       page_offset, user_data, page_length);
 
 		/* If we get a fault while copying data, then (presumably) our
-		 * source page isn't available. In this case, use the
-		 * non-atomic function
+		 * source page isn't available.  Return the error and we'll
+		 * retry in the slow path.
 		 */
-		if (ret) {
-			ret = slow_user_write (dev_priv->mm.gtt_mapping,
-					       page_base, page_offset,
-					       user_data, page_length);
-			if (ret)
-				goto fail;
-		}
+		if (ret)
+			goto fail;
 
 		remain -= page_length;
 		user_data += page_length;
@@ -315,39 +527,284 @@ fail:
 	return ret;
 }
 
+/**
+ * This is the fallback GTT pwrite path, which uses get_user_pages to pin
+ * the memory and maps it using kmap_atomic for copying.
+ *
+ * This code resulted in x11perf -rgb10text consuming about 10% more CPU
+ * than using i915_gem_gtt_pwrite_fast on a G45 (32-bit).
+ */
 static int
-i915_gem_shmem_pwrite(struct drm_device *dev, struct drm_gem_object *obj,
-		      struct drm_i915_gem_pwrite *args,
-		      struct drm_file *file_priv)
+i915_gem_gtt_pwrite_slow(struct drm_device *dev, struct drm_gem_object *obj,
+			 struct drm_i915_gem_pwrite *args,
+			 struct drm_file *file_priv)
 {
+	struct drm_i915_gem_object *obj_priv = obj->driver_private;
+	drm_i915_private_t *dev_priv = dev->dev_private;
+	ssize_t remain;
+	loff_t gtt_page_base, offset;
+	loff_t first_data_page, last_data_page, num_pages;
+	loff_t pinned_pages, i;
+	struct page **user_pages;
+	struct mm_struct *mm = current->mm;
+	int gtt_page_offset, data_page_offset, data_page_index, page_length;
 	int ret;
-	loff_t offset;
-	ssize_t written;
+	uint64_t data_ptr = args->data_ptr;
+
+	remain = args->size;
+
+	/* Pin the user pages containing the data.  We can't fault while
+	 * holding the struct mutex, and all of the pwrite implementations
+	 * want to hold it while dereferencing the user data.
+	 */
+	first_data_page = data_ptr / PAGE_SIZE;
+	last_data_page = (data_ptr + args->size - 1) / PAGE_SIZE;
+	num_pages = last_data_page - first_data_page + 1;
+
+	user_pages = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
+	if (user_pages == NULL)
+		return -ENOMEM;
+
+	down_read(&mm->mmap_sem);
+	pinned_pages = get_user_pages(current, mm, (uintptr_t)args->data_ptr,
+				      num_pages, 0, 0, user_pages, NULL);
+	up_read(&mm->mmap_sem);
+	if (pinned_pages < num_pages) {
+		ret = -EFAULT;
+		goto out_unpin_pages;
+	}
 
 	mutex_lock(&dev->struct_mutex);
+	ret = i915_gem_object_pin(obj, 0);
+	if (ret)
+		goto out_unlock;
+
+	ret = i915_gem_object_set_to_gtt_domain(obj, 1);
+	if (ret)
+		goto out_unpin_object;
+
+	obj_priv = obj->driver_private;
+	offset = obj_priv->gtt_offset + args->offset;
+
+	while (remain > 0) {
+		/* Operation in this page
+		 *
+		 * gtt_page_base = page offset within aperture
+		 * gtt_page_offset = offset within page in aperture
+		 * data_page_index = page number in get_user_pages return
+		 * data_page_offset = offset with data_page_index page.
+		 * page_length = bytes to copy for this page
+		 */
+		gtt_page_base = offset & PAGE_MASK;
+		gtt_page_offset = offset & ~PAGE_MASK;
+		data_page_index = data_ptr / PAGE_SIZE - first_data_page;
+		data_page_offset = data_ptr & ~PAGE_MASK;
+
+		page_length = remain;
+		if ((gtt_page_offset + page_length) > PAGE_SIZE)
+			page_length = PAGE_SIZE - gtt_page_offset;
+		if ((data_page_offset + page_length) > PAGE_SIZE)
+			page_length = PAGE_SIZE - data_page_offset;
+
+		ret = slow_kernel_write(dev_priv->mm.gtt_mapping,
+					gtt_page_base, gtt_page_offset,
+					user_pages[data_page_index],
+					data_page_offset,
+					page_length);
+
+		/* If we get a fault while copying data, then (presumably) our
+		 * source page isn't available.  Return the error and we'll
+		 * retry in the slow path.
+		 */
+		if (ret)
+			goto out_unpin_object;
+
+		remain -= page_length;
+		offset += page_length;
+		data_ptr += page_length;
+	}
+
+out_unpin_object:
+	i915_gem_object_unpin(obj);
+out_unlock:
+	mutex_unlock(&dev->struct_mutex);
+out_unpin_pages:
+	for (i = 0; i < pinned_pages; i++)
+		page_cache_release(user_pages[i]);
+	kfree(user_pages);
+
+	return ret;
+}
+
+/**
+ * This is the fast shmem pwrite path, which attempts to directly
+ * copy_from_user into the kmapped pages backing the object.
+ */
+static int
+i915_gem_shmem_pwrite_fast(struct drm_device *dev, struct drm_gem_object *obj,
+			   struct drm_i915_gem_pwrite *args,
+			   struct drm_file *file_priv)
+{
+	struct drm_i915_gem_object *obj_priv = obj->driver_private;
+	ssize_t remain;
+	loff_t offset, page_base;
+	char __user *user_data;
+	int page_offset, page_length;
+	int ret;
+
+	user_data = (char __user *) (uintptr_t) args->data_ptr;
+	remain = args->size;
+
+	mutex_lock(&dev->struct_mutex);
+
+	ret = i915_gem_object_get_pages(obj);
+	if (ret != 0)
+		goto fail_unlock;
 
 	ret = i915_gem_object_set_to_cpu_domain(obj, 1);
-	if (ret) {
-		mutex_unlock(&dev->struct_mutex);
-		return ret;
+	if (ret != 0)
+		goto fail_put_pages;
+
+	obj_priv = obj->driver_private;
+	offset = args->offset;
+	obj_priv->dirty = 1;
+
+	while (remain > 0) {
+		/* Operation in this page
+		 *
+		 * page_base = page offset within aperture
+		 * page_offset = offset within page
+		 * page_length = bytes to copy for this page
+		 */
+		page_base = (offset & ~(PAGE_SIZE-1));
+		page_offset = offset & (PAGE_SIZE-1);
+		page_length = remain;
+		if ((page_offset + remain) > PAGE_SIZE)
+			page_length = PAGE_SIZE - page_offset;
+
+		ret = fast_shmem_write(obj_priv->pages,
+				       page_base, page_offset,
+				       user_data, page_length);
+		if (ret)
+			goto fail_put_pages;
+
+		remain -= page_length;
+		user_data += page_length;
+		offset += page_length;
+	}
+
+fail_put_pages:
+	i915_gem_object_put_pages(obj);
+fail_unlock:
+	mutex_unlock(&dev->struct_mutex);
+
+	return ret;
+}
+
+/**
+ * This is the fallback shmem pwrite path, which uses get_user_pages to pin
+ * the memory and maps it using kmap_atomic for copying.
+ *
+ * This avoids taking mmap_sem for faulting on the user's address while the
+ * struct_mutex is held.
+ */
+static int
+i915_gem_shmem_pwrite_slow(struct drm_device *dev, struct drm_gem_object *obj,
+			   struct drm_i915_gem_pwrite *args,
+			   struct drm_file *file_priv)
+{
+	struct drm_i915_gem_object *obj_priv = obj->driver_private;
+	struct mm_struct *mm = current->mm;
+	struct page **user_pages;
+	ssize_t remain;
+	loff_t offset, pinned_pages, i;
+	loff_t first_data_page, last_data_page, num_pages;
+	int shmem_page_index, shmem_page_offset;
+	int data_page_index,  data_page_offset;
+	int page_length;
+	int ret;
+	uint64_t data_ptr = args->data_ptr;
+
+	remain = args->size;
+
+	/* Pin the user pages containing the data.  We can't fault while
+	 * holding the struct mutex, and all of the pwrite implementations
+	 * want to hold it while dereferencing the user data.
+	 */
+	first_data_page = data_ptr / PAGE_SIZE;
+	last_data_page = (data_ptr + args->size - 1) / PAGE_SIZE;
+	num_pages = last_data_page - first_data_page + 1;
+
+	user_pages = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
+	if (user_pages == NULL)
+		return -ENOMEM;
+
+	down_read(&mm->mmap_sem);
+	pinned_pages = get_user_pages(current, mm, (uintptr_t)args->data_ptr,
+				      num_pages, 0, 0, user_pages, NULL);
+	up_read(&mm->mmap_sem);
+	if (pinned_pages < num_pages) {
+		ret = -EFAULT;
+		goto fail_put_user_pages;
 	}
 
+	mutex_lock(&dev->struct_mutex);
+
+	ret = i915_gem_object_get_pages(obj);
+	if (ret != 0)
+		goto fail_unlock;
+
+	ret = i915_gem_object_set_to_cpu_domain(obj, 1);
+	if (ret != 0)
+		goto fail_put_pages;
+
+	obj_priv = obj->driver_private;
 	offset = args->offset;
+	obj_priv->dirty = 1;
 
-	written = vfs_write(obj->filp,
-			    (char __user *)(uintptr_t) args->data_ptr,
-			    args->siz