Merge branch 'vmwgfx-next-3.13' of git://people.freedesktop.org/~thomash/linux into drm-next

Pull request for vmwgfx. Currently just the DMA address stuff.

* 'vmwgfx-next-3.13' of git://people.freedesktop.org/~thomash/linux:
  drm/vmwgfx: Use the linux DMA api to get valid device addresses of pages
  drm/ttm: Enable the dma page pool also for intel IOMMUs

7 files changed, 634 insertions, 100 deletions

diff --git a/drivers/gpu/drm/ttm/Makefile b/drivers/gpu/drm/ttm/Makefile
index b2b33dde2afb..b433b9f040c9 100644
--- a/drivers/gpu/drm/ttm/Makefile
+++ b/drivers/gpu/drm/ttm/Makefile
@@ -5,10 +5,6 @@ ccflags-y := -Iinclude/drm
 ttm-y := ttm_agp_backend.o ttm_memory.o ttm_tt.o ttm_bo.o \
 	ttm_bo_util.o ttm_bo_vm.o ttm_module.o \
 	ttm_object.o ttm_lock.o ttm_execbuf_util.o ttm_page_alloc.o \
-	ttm_bo_manager.o
-
-ifeq ($(CONFIG_SWIOTLB),y)
-ttm-y += ttm_page_alloc_dma.o
-endif
+	ttm_bo_manager.o ttm_page_alloc_dma.o
 
 obj-$(CONFIG_DRM_TTM) += ttm.o
diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
index 7957beeeaf73..fb8259f69839 100644
--- a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
+++ b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
@@ -33,6 +33,7 @@
  *   when freed).
  */
 
+#if defined(CONFIG_SWIOTLB) || defined(CONFIG_INTEL_IOMMU)
 #define pr_fmt(fmt) "[TTM] " fmt
 
 #include <linux/dma-mapping.h>
@@ -1142,3 +1143,5 @@ int ttm_dma_page_alloc_debugfs(struct seq_file *m, void *data)
 	return 0;
 }
 EXPORT_SYMBOL_GPL(ttm_dma_page_alloc_debugfs);
+
+#endif
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
index 96dc84dc34d0..7776e6f0aef6 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
@@ -141,37 +141,374 @@ struct ttm_placement vmw_srf_placement = {
 };
 
 struct vmw_ttm_tt {
-	struct ttm_tt ttm;
+	struct ttm_dma_tt dma_ttm;
 	struct vmw_private *dev_priv;
 	int gmr_id;
+	struct sg_table sgt;
+	struct vmw_sg_table vsgt;
+	uint64_t sg_alloc_size;
+	bool mapped;
 };
 
+/**
+ * Helper functions to advance a struct vmw_piter iterator.
+ *
+ * @viter: Pointer to the iterator.
+ *
+ * These functions return false if past the end of the list,
+ * true otherwise. Functions are selected depending on the current
+ * DMA mapping mode.
+ */
+static bool __vmw_piter_non_sg_next(struct vmw_piter *viter)
+{
+	return ++(viter->i) < viter->num_pages;
+}
+
+static bool __vmw_piter_sg_next(struct vmw_piter *viter)
+{
+	return __sg_page_iter_next(&viter->iter);
+}
+
+
+/**
+ * Helper functions to return a pointer to the current page.
+ *
+ * @viter: Pointer to the iterator
+ *
+ * These functions return a pointer to the page currently
+ * pointed to by @viter. Functions are selected depending on the
+ * current mapping mode.
+ */
+static struct page *__vmw_piter_non_sg_page(struct vmw_piter *viter)
+{
+	return viter->pages[viter->i];
+}
+
+static struct page *__vmw_piter_sg_page(struct vmw_piter *viter)
+{
+	return sg_page_iter_page(&viter->iter);
+}
+
+
+/**
+ * Helper functions to return the DMA address of the current page.
+ *
+ * @viter: Pointer to the iterator
+ *
+ * These functions return the DMA address of the page currently
+ * pointed to by @viter. Functions are selected depending on the
+ * current mapping mode.
+ */
+static dma_addr_t __vmw_piter_phys_addr(struct vmw_piter *viter)
+{
+	return page_to_phys(viter->pages[viter->i]);
+}
+
+static dma_addr_t __vmw_piter_dma_addr(struct vmw_piter *viter)
+{
+	return viter->addrs[viter->i];
+}
+
+static dma_addr_t __vmw_piter_sg_addr(struct vmw_piter *viter)
+{
+	return sg_page_iter_dma_address(&viter->iter);
+}
+
+
+/**
+ * vmw_piter_start - Initialize a struct vmw_piter.
+ *
+ * @viter: Pointer to the iterator to initialize
+ * @vsgt: Pointer to a struct vmw_sg_table to initialize from
+ *
+ * Note that we're following the convention of __sg_page_iter_start, so that
+ * the iterator doesn't point to a valid page after initialization; it has
+ * to be advanced one step first.
+ */
+void vmw_piter_start(struct vmw_piter *viter, const struct vmw_sg_table *vsgt,
+		     unsigned long p_offset)
+{
+	viter->i = p_offset - 1;
+	viter->num_pages = vsgt->num_pages;
+	switch (vsgt->mode) {
+	case vmw_dma_phys:
+		viter->next = &__vmw_piter_non_sg_next;
+		viter->dma_address = &__vmw_piter_phys_addr;
+		viter->page = &__vmw_piter_non_sg_page;
+		viter->pages = vsgt->pages;
+		break;
+	case vmw_dma_alloc_coherent:
+		viter->next = &__vmw_piter_non_sg_next;
+		viter->dma_address = &__vmw_piter_dma_addr;
+		viter->page = &__vmw_piter_non_sg_page;
+		viter->addrs = vsgt->addrs;
+		break;
+	case vmw_dma_map_populate:
+	case vmw_dma_map_bind:
+		viter->next = &__vmw_piter_sg_next;
+		viter->dma_address = &__vmw_piter_sg_addr;
+		viter->page = &__vmw_piter_sg_page;
+		__sg_page_iter_start(&viter->iter, vsgt->sgt->sgl,
+				     vsgt->sgt->orig_nents, p_offset);
+		break;
+	default:
+		BUG();
+	}
+}
+
+/**
+ * vmw_ttm_unmap_from_dma - unmap  device addresses previsouly mapped for
+ * TTM pages
+ *
+ * @vmw_tt: Pointer to a struct vmw_ttm_backend
+ *
+ * Used to free dma mappings previously mapped by vmw_ttm_map_for_dma.
+ */
+static void vmw_ttm_unmap_from_dma(struct vmw_ttm_tt *vmw_tt)
+{
+	struct device *dev = vmw_tt->dev_priv->dev->dev;
+
+	dma_unmap_sg(dev, vmw_tt->sgt.sgl, vmw_tt->sgt.nents,
+		DMA_BIDIRECTIONAL);
+	vmw_tt->sgt.nents = vmw_tt->sgt.orig_nents;
+}
+
+/**
+ * vmw_ttm_map_for_dma - map TTM pages to get device addresses
+ *
+ * @vmw_tt: Pointer to a struct vmw_ttm_backend
+ *
+ * This function is used to get device addresses from the kernel DMA layer.
+ * However, it's violating the DMA API in that when this operation has been
+ * performed, it's illegal for the CPU to write to the pages without first
+ * unmapping the DMA mappings, or calling dma_sync_sg_for_cpu(). It is
+ * therefore only legal to call this function if we know that the function
+ * dma_sync_sg_for_cpu() is a NOP, and dma_sync_sg_for_device() is at most
+ * a CPU write buffer flush.
+ */
+static int vmw_ttm_map_for_dma(struct vmw_ttm_tt *vmw_tt)
+{
+	struct device *dev = vmw_tt->dev_priv->dev->dev;
+	int ret;
+
+	ret = dma_map_sg(dev, vmw_tt->sgt.sgl, vmw_tt->sgt.orig_nents,
+			 DMA_BIDIRECTIONAL);
+	if (unlikely(ret == 0))
+		return -ENOMEM;
+
+	vmw_tt->sgt.nents = ret;
+
+	return 0;
+}
+
+/**
+ * vmw_ttm_map_dma - Make sure TTM pages are visible to the device
+ *
+ * @vmw_tt: Pointer to a struct vmw_ttm_tt
+ *
+ * Select the correct function for and make sure the TTM pages are
+ * visible to the device. Allocate storage for the device mappings.
+ * If a mapping has already been performed, indicated by the storage
+ * pointer being non NULL, the function returns success.
+ */
+static int vmw_ttm_map_dma(struct vmw_ttm_tt *vmw_tt)
+{
+	struct vmw_private *dev_priv = vmw_tt->dev_priv;
+	struct ttm_mem_global *glob = vmw_mem_glob(dev_priv);
+	struct vmw_sg_table *vsgt = &vmw_tt->vsgt;
+	struct vmw_piter iter;
+	dma_addr_t old;
+	int ret = 0;
+	static size_t sgl_size;
+	static size_t sgt_size;
+
+	if (vmw_tt->mapped)
+		return 0;
+
+	vsgt->mode = dev_priv->map_mode;
+	vsgt->pages = vmw_tt->dma_ttm.ttm.pages;
+	vsgt->num_pages = vmw_tt->dma_ttm.ttm.num_pages;
+	vsgt->addrs = vmw_tt->dma_ttm.dma_address;
+	vsgt->sgt = &vmw_tt->sgt;
+
+	switch (dev_priv->map_mode) {
+	case vmw_dma_map_bind:
+	case vmw_dma_map_populate:
+		if (unlikely(!sgl_size)) {
+			sgl_size = ttm_round_pot(sizeof(struct scatterlist));
+			sgt_size = ttm_round_pot(sizeof(struct sg_table));
+		}
+		vmw_tt->sg_alloc_size = sgt_size + sgl_size * vsgt->num_pages;
+		ret = ttm_mem_global_alloc(glob, vmw_tt->sg_alloc_size, false,
+					   true);
+		if (unlikely(ret != 0))
+			return ret;
+
+		ret = sg_alloc_table_from_pages(&vmw_tt->sgt, vsgt->pages,
+						vsgt->num_pages, 0,
+						(unsigned long)
+						vsgt->num_pages << PAGE_SHIFT,
+						GFP_KERNEL);
+		if (unlikely(ret != 0))
+			goto out_sg_alloc_fail;
+
+		if (vsgt->num_pages > vmw_tt->sgt.nents) {
+			uint64_t over_alloc =
+				sgl_size * (vsgt->num_pages -
+					    vmw_tt->sgt.nents);
+
+			ttm_mem_global_free(glob, over_alloc);
+			vmw_tt->sg_alloc_size -= over_alloc;
+		}
+
+		ret = vmw_ttm_map_for_dma(vmw_tt);
+		if (unlikely(ret != 0))
+			goto out_map_fail;
+
+		break;
+	default:
+		break;
+	}
+
+	old = ~((dma_addr_t) 0);
+	vmw_tt->vsgt.num_regions = 0;
+	for (vmw_piter_start(&iter, vsgt, 0); vmw_piter_next(&iter);) {
+		dma_addr_t cur = vmw_piter_dma_addr(&iter);
+
+		if (cur != old + PAGE_SIZE)
+			vmw_tt->vsgt.num_regions++;
+		old = cur;
+	}
+
+	vmw_tt->mapped = true;
+	return 0;
+
+out_map_fail:
+	sg_free_table(vmw_tt->vsgt.sgt);
+	vmw_tt->vsgt.sgt = NULL;
+out_sg_alloc_fail:
+	ttm_mem_global_free(glob, vmw_tt->sg_alloc_size);
+	return ret;
+}
+
+/**
+ * vmw_ttm_unmap_dma - Tear down any TTM page device mappings
+ *
+ * @vmw_tt: Pointer to a struct vmw_ttm_tt
+ *
+ * Tear down any previously set up device DMA mappings and free
+ * any storage space allocated for them. If there are no mappings set up,
+ * this function is a NOP.
+ */
+static void vmw_ttm_unmap_dma(struct vmw_ttm_tt *vmw_tt)
+{
+	struct vmw_private *dev_priv = vmw_tt->dev_priv;
+
+	if (!vmw_tt->vsgt.sgt)
+		return;
+
+	switch (dev_priv->map_mode) {
+	case vmw_dma_map_bind:
+	case vmw_dma_map_populate:
+		vmw_ttm_unmap_from_dma(vmw_tt);
+		sg_free_table(vmw_tt->vsgt.sgt);
+		vmw_tt->vsgt.sgt = NULL;
+		ttm_mem_global_free(vmw_mem_glob(dev_priv),
+				    vmw_tt->sg_alloc_size);
+		break;
+	default:
+		break;
+	}
+	vmw_tt->mapped = false;
+}
+
 static int vmw_ttm_bind(struct ttm_tt *ttm, struct ttm_mem_reg *bo_mem)
 {
-	struct vmw_ttm_tt *vmw_be = container_of(ttm, struct vmw_ttm_tt, ttm);
+	struct vmw_ttm_tt *vmw_be =
+		container_of(ttm, struct vmw_ttm_tt, dma_ttm.ttm);
+	int ret;
+
+	ret = vmw_ttm_map_dma(vmw_be);
+	if (unlikely(ret != 0))
+		return ret;
 
 	vmw_be->gmr_id = bo_mem->start;
 
-	return vmw_gmr_bind(vmw_be->dev_priv, ttm->pages,
+	return vmw_gmr_bind(vmw_be->dev_priv, &vmw_be->vsgt,
 			    ttm->num_pages, vmw_be->gmr_id);
 }
 
 static int vmw_ttm_unbind(struct ttm_tt *ttm)
 {
-	struct vmw_ttm_tt *vmw_be = container_of(ttm, struct vmw_ttm_tt, ttm);
+	struct vmw_ttm_tt *vmw_be =
+		container_of(ttm, struct vmw_ttm_tt, dma_ttm.ttm);
 
 	vmw_gmr_unbind(vmw_be->dev_priv, vmw_be->gmr_id);
+
+	if (vmw_be->dev_priv->map_mode == vmw_dma_map_bind)
+		vmw_ttm_unmap_dma(vmw_be);
+
 	return 0;
 }
 
 static void vmw_ttm_destroy(struct ttm_tt *ttm)
 {
-	struct vmw_ttm_tt *vmw_be = container_of(ttm, struct vmw_ttm_tt, ttm);
-
-	ttm_tt_fini(ttm);
+	struct vmw_ttm_tt *vmw_be =
+		container_of(ttm, struct vmw_ttm_tt, dma_ttm.ttm);
+
+	vmw_ttm_unmap_dma(vmw_be);
+	if (vmw_be->dev_priv->map_mode == vmw_dma_alloc_coherent)
+		ttm_dma_tt_fini(&vmw_be->dma_ttm);
+	else
+		ttm_tt_fini(ttm);
 	kfree(vmw_be);
 }
 
+static int vmw_ttm_populate(struct ttm_tt *ttm)
+{
+	struct vmw_ttm_tt *vmw_tt =
+		container_of(ttm, struct vmw_ttm_tt, dma_ttm.ttm);
+	struct vmw_private *dev_priv = vmw_tt->dev_priv;
+	struct ttm_mem_global *glob = vmw_mem_glob(dev_priv);
+	int ret;
+
+	if (ttm->state != tt_unpopulated)
+		return 0;
+
+	if (dev_priv->map_mode == vmw_dma_alloc_coherent) {
+		size_t size =
+			ttm_round_pot(ttm->num_pages * sizeof(dma_addr_t));
+		ret = ttm_mem_global_alloc(glob, size, false, true);
+		if (unlikely(ret != 0))
+			return ret;
+
+		ret = ttm_dma_populate(&vmw_tt->dma_ttm, dev_priv->dev->dev);
+		if (unlikely(ret != 0))
+			ttm_mem_global_free(glob, size);
+	} else
+		ret = ttm_pool_populate(ttm);
+
+	return ret;
+}
+
+static void vmw_ttm_unpopulate(struct ttm_tt *ttm)
+{
+	struct vmw_ttm_tt *vmw_tt = container_of(ttm, struct vmw_ttm_tt,
+						 dma_ttm.ttm);
+	struct vmw_private *dev_priv = vmw_tt->dev_priv;
+	struct ttm_mem_global *glob = vmw_mem_glob(dev_priv);
+
+	vmw_ttm_unmap_dma(vmw_tt);
+	if (dev_priv->map_mode == vmw_dma_alloc_coherent) {
+		size_t size =
+			ttm_round_pot(ttm->num_pages * sizeof(dma_addr_t));
+
+		ttm_dma_unpopulate(&vmw_tt->dma_ttm, dev_priv->dev->dev);
+		ttm_mem_global_free(glob, size);
+	} else
+		ttm_pool_unpopulate(ttm);
+}
+
 static struct ttm_backend_func vmw_ttm_func = {
 	.bind = vmw_ttm_bind,
 	.unbind = vmw_ttm_unbind,
@@ -183,20 +520,28 @@ struct ttm_tt *vmw_ttm_tt_create(struct ttm_bo_device *bdev,
 				 struct page *dummy_read_page)
 {
 	struct vmw_ttm_tt *vmw_be;
+	int ret;
 
-	vmw_be = kmalloc(sizeof(*vmw_be), GFP_KERNEL);
+	vmw_be = kzalloc(sizeof(*vmw_be), GFP_KERNEL);
 	if (!vmw_be)
 		return NULL;
 
-	vmw_be->ttm.func = &vmw_ttm_func;
+	vmw_be->dma_ttm.ttm.func = &vmw_ttm_func;
 	vmw_be->dev_priv = container_of(bdev, struct vmw_private, bdev);
 
-	if (ttm_tt_init(&vmw_be->ttm, bdev, size, page_flags, dummy_read_page)) {
-		kfree(vmw_be);
-		return NULL;
-	}
-
-	return &vmw_be->ttm;
+	if (vmw_be->dev_priv->map_mode == vmw_dma_alloc_coherent)
+		ret = ttm_dma_tt_init(&vmw_be->dma_ttm, bdev, size, page_flags,
+				      dummy_read_page);
+	else
+		ret = ttm_tt_init(&vmw_be->dma_ttm.ttm, bdev, size, page_flags,
+				  dummy_read_page);
+	if (unlikely(ret != 0))
+		goto out_no_init;
+
+	return &vmw_be->dma_ttm.ttm;
+out_no_init:
+	kfree(vmw_be);
+	return NULL;
 }
 
 int vmw_invalidate_caches(struct ttm_bo_device *bdev, uint32_t flags)
@@ -332,8 +677,8 @@ static int vmw_sync_obj_wait(void *sync_obj, bool lazy, bool interruptible)
 
 struct ttm_bo_driver vmw_bo_driver = {
 	.ttm_tt_create = &vmw_ttm_tt_create,
-	.ttm_tt_populate = &ttm_pool_populate,
-	.ttm_tt_unpopulate = &ttm_pool_unpopulate,
+	.ttm_tt_populate = &vmw_ttm_populate,
+	.ttm_tt_unpopulate = &vmw_ttm_unpopulate,
 	.invalidate_caches = vmw_invalidate_caches,
 	.init_mem_type = vmw_init_mem_type,
 	.evict_flags = vmw_evict_flags,
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
index 1a90f0a2f7e5..0b5c7818ebfb 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
@@ -32,6 +32,7 @@
 #include <drm/ttm/ttm_bo_driver.h>
 #include <drm/ttm/ttm_object.h>
 #include <drm/ttm/ttm_module.h>
+#include <linux/dma_remapping.h>
 
 #define VMWGFX_DRIVER_NAME "vmwgfx"
 #define VMWGFX_DRIVER_DESC "Linux drm driver for VMware graphics devices"
@@ -185,6 +186,9 @@ static struct pci_device_id vmw_pci_id_list[] = {
 MODULE_DEVICE_TABLE(pci, vmw_pci_id_list);
 
 static int enable_fbdev = IS_ENABLED(CONFIG_DRM_VMWGFX_FBCON);
+static int vmw_force_iommu;
+static int vmw_restrict_iommu;
+static int vmw_force_coherent;
 
 static int vmw_probe(struct pci_dev *, const struct pci_device_id *);
 static void vmw_master_init(struct vmw_master *);
@@ -193,6 +197,13 @@ static int vmwgfx_pm_notifier(struct notifier_block *nb, unsigned long val,
 
 MODULE_PARM_DESC(enable_fbdev, "Enable vmwgfx fbdev");
 module_param_named(enable_fbdev, enable_fbdev, int, 0600);
+MODULE_PARM_DESC(force_dma_api, "Force using the DMA API for TTM pages");
+module_param_named(force_dma_api, vmw_force_iommu, int, 0600);
+MODULE_PARM_DESC(restrict_iommu, "Try to limit IOMMU usage for TTM pages");
+module_param_named(restrict_iommu, vmw_restrict_iommu, int, 0600);
+MODULE_PARM_DESC(force_coherent, "Force coherent TTM pages");
+module_param_named(force_coherent, vmw_force_coherent, int, 0600);
+
 
 static void vmw_print_capabilities(uint32_t capabilities)
 {
@@ -427,12 +438,78 @@ static void vmw_get_initial_size(struct vmw_private *dev_priv)
 	dev_priv->initial_height = height;
 }
 
+/**
+ * vmw_dma_select_mode - Determine how DMA mappings should be set up for this
+ * system.
+ *
+ * @dev_priv: Pointer to a struct vmw_private
+ *
+ * This functions tries to determine the IOMMU setup and what actions
+ * need to be taken by the driver to make system pages visible to the
+ * device.
+ * If this function decides that DMA is not possible, it returns -EINVAL.
+ * The driver may then try to disable features of the device that require
+ * DMA.
+ */
+static int vmw_dma_select_mode(struct vmw_private *dev_priv)
+{
+	const struct dma_map_ops *dma_ops = get_dma_ops(dev_priv->dev->dev);
+	static const char *names[vmw_dma_map_max] = {
+		[vmw_dma_phys] = "Using physical TTM page addresses.",
+		[vmw_dma_alloc_coherent] = "Using coherent TTM pages.",
+		[vmw_dma_map_populate] = "Keeping DMA mappings.",
+		[vmw_dma_map_bind] = "Giving up DMA mappings early."};
+
+#ifdef CONFIG_INTEL_IOMMU
+	if (intel_iommu_enabled) {
+		dev_priv->map_mode = vmw_dma_map_populate;
+		goto out_fixup;
+	}
+#endif
+
+	if (!(vmw_force_iommu || vmw_force_coherent)) {
+		dev_priv->map_mode = vmw_dma_phys;
+		DRM_INFO("DMA map mode: %s\n", names[dev_priv->map_mode]);
+		return 0;
+	}
+
+	dev_priv->map_mode = vmw_dma_map_populate;
+
+	if (dma_ops->sync_single_for_cpu)
+		dev_priv->map_mode = vmw_dma_alloc_coherent;
+#ifdef CONFIG_SWIOTLB
+	if (swiotlb_nr_tbl() == 0)
+		dev_priv->map_mode = vmw_dma_map_populate;
+#endif
+
+out_fixup:
+	if (dev_priv->map_mode == vmw_dma_map_populate &&
+	    vmw_restrict_iommu)
+		dev_priv->map_mode = vmw_dma_map_bind;
+
+	if (vmw_force_coherent)
+		dev_priv->map_mode = vmw_dma_alloc_coherent;
+
+#if !defined(CONFIG_SWIOTLB) && !defined(CONFIG_INTEL_IOMMU)
+	/*
+	 * No coherent page pool
+	 */
+	if (dev_priv->map_mode == vmw_dma_alloc_coherent)
+		return -EINVAL;
+#endif
+
+	DRM_INFO("DMA map mode: %s\n", names[dev_priv->map_mode]);
+
+	return 0;
+}
+
 static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 {
 	struct vmw_private *dev_priv;
 	int ret;
 	uint32_t svga_id;
 	enum vmw_res_type i;
+	bool refuse_dma = false;
 
 	dev_priv = kzalloc(sizeof(*dev_priv), GFP_KERNEL);
 	if (unlikely(dev_priv == NULL)) {
@@ -481,6 +558,11 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 	}
 
 	dev_priv->capabilities = vmw_read(dev_priv, SVGA_REG_CAPABILITIES);
+	ret = vmw_dma_select_mode(dev_priv);
+	if (unlikely(ret != 0)) {
+		DRM_INFO("Restricting capabilities due to IOMMU setup.\n");
+		refuse_dma = true;
+	}
 
 	dev_priv->vram_size = vmw_read(dev_priv, SVGA_REG_VRAM_SIZE);
 	dev_priv->mmio_size = vmw_read(dev_priv, SVGA_REG_MEM_SIZE);
@@ -558,8 +640,9 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 	}
 
 	dev_priv->has_gmr = true;
-	if (ttm_bo_init_mm(&dev_priv->bdev, VMW_PL_GMR,
-			   dev_priv->max_gmr_ids) != 0) {
+	if (((dev_priv->capabilities & (SVGA_CAP_GMR | SVGA_CAP_GMR2)) == 0) ||
+	    refuse_dma || ttm_bo_init_mm(&dev_priv->bdev, VMW_PL_GMR,
+					 dev_priv->max_gmr_ids) != 0) {
 		DRM_INFO("No GMR memory available. "
 			 "Graphics memory resources are very limited.\n");
 		dev_priv->has_gmr = false;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index 150ec64af617..e401d5dbcb96 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -177,6 +177,58 @@ struct vmw_res_cache_entry {
 	struct vmw_resource_val_node *node;
 };
 
+/**
+ * enum vmw_dma_map_mode - indicate how to perform TTM page dma mappings.
+ */
+enum vmw_dma_map_mode {
+	vmw_dma_phys,           /* Use physical page addresses */
+	vmw_dma_alloc_coherent, /* Use TTM coherent pages */
+	vmw_dma_map_populate,   /* Unmap from DMA just after unpopulate */
+	vmw_dma_map_bind,       /* Unmap from DMA just before unbind */
+	vmw_dma_map_max
+};
+
+/**
+ * struct vmw_sg_table - Scatter/gather table for binding, with additional
+ * device-specific information.
+ *
+ * @sgt: Pointer to a struct sg_table with binding information
+ * @num_regions: Number of regions with device-address contigous pages
+ */
+struct vmw_sg_table {
+	enum vmw_dma_map_mode mode;
+	struct page **pages;
+	const dma_addr_t *addrs;
+	struct sg_table *sgt;
+	unsigned long num_regions;
+	unsigned long num_pages;
+};
+
+/**
+ * struct vmw_piter - Page iterator that iterates over a list of pages
+ * and DMA addresses that could be either a scatter-gather list or
+ * arrays
+ *
+ * @pages: Array of page pointers to the pages.
+ * @addrs: DMA addresses to the pages if coherent pages are used.
+ * @iter: Scatter-gather page iterator. Current position in SG list.
+ * @i: Current position in arrays.
+ * @num_pages: Number of pages total.
+ * @next: Function to advance the iterator. Returns false if past the list
+ * of pages, true otherwise.
+ * @dma_address: Function to return the DMA address of the current page.
+ */
+struct vmw_piter {
+	struct page **pages;
+	const dma_addr_t *addrs;
+	struct sg_page_iter iter;
+	unsigned long i;
+	unsigned long num_pages;
+	bool (*next)(struct vmw_piter *);
+	dma_addr_t (*dma_address)(struct vmw_piter *);
+	struct page *(*page)(struct vmw_piter *);
+};
+
 struct vmw_sw_context{
 	struct drm_open_hash res_ht;
 	bool res_ht_initialized;
@@ -358,6 +410,11 @@ struct vmw_private {
 
 	struct list_head res_lru[vmw_res_max];
 	uint32_t used_memory_size;
+
+	/*
+	 * DMA mapping stuff.
+	 */
+	enum vmw_dma_map_mode map_mode;
 };
 
 static inline struct vmw_surface *vmw_res_to_srf(struct vmw_resource *res)
@@ -405,7 +462,7 @@ void vmw_3d_resource_dec(struct vmw_private *dev_priv, bool hide_svga);
  */
 
 extern int vmw_gmr_bind(struct vmw_private *dev_priv,
-			struct page *pages[],
+			const struct vmw_sg_table *vsgt,
 			unsigned long num_pages,
 			int gmr_id);
 extern void vmw_gmr_unbind(struct vmw_private *dev_priv, int gmr_id);
@@ -568,6 +625,45 @@ extern struct ttm_placement vmw_evictable_placement;
 extern struct ttm_placement vmw_srf_placement;
 extern struct ttm_bo_driver vmw_bo_driver;
 extern int vmw_dma_quiescent(struct drm_device *dev);
+extern void vmw_piter_start(struct vmw_piter *viter,
+			    const struct vmw_sg_table *vsgt,
+			    unsigned long p_offs);
+
+/**
+ * vmw_piter_next - Advance the iterator one page.
+ *
+ * @viter: Pointer to the iterator to advance.
+ *
+ * Returns false if past the list of pages, true otherwise.
+ */
+static inline bool vmw_piter_next(struct vmw_piter *viter)
+{
+	return viter->next(viter);
+}
+
+/**
+ * vmw_piter_dma_addr - Return the DMA address of the current page.
+ *
+ * @viter: Pointer to the iterator
+ *
+ * Returns the DMA address of the page pointed to by @viter.
+ */
+static inline dma_addr_t vmw_piter_dma_addr(struct vmw_piter *viter)
+{
+	return viter->dma_address(viter);
+}
+
+/**
+ * vmw_piter_page - Return a pointer to the current page.
+ *
+ * @viter: Pointer to the iterator
+ *
+ * Returns the DMA address of the page pointed to by @viter.
+ */
+static inline struct page *vmw_piter_page(struct vmw_piter *viter)
+{
+	return viter->page(viter);
+}
 
 /**
  * Command submission - vmwgfx_execbuf.c
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_gmr.c b/drivers/gpu/drm/vmwgfx/vmwgfx_gmr.c
index 1a0bf07fe54b..6d0952366f91 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_gmr.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_gmr.c
@@ -32,9 +32,11 @@
 #define VMW_PPN_SIZE (sizeof(unsigned long))
 /* A future safe maximum remap size. */
 #define VMW_PPN_PER_REMAP ((31 * 1024) / VMW_PPN_SIZE)
+#define DMA_ADDR_INVALID ((dma_addr_t) 0)
+#define DMA_PAGE_INVALID 0UL
 
 static int vmw_gmr2_bind(struct vmw_private *dev_priv,
-			 struct page *pages[],
+			 struct vmw_piter *iter,
 			 unsigned long num_pages,
 			 int gmr_id)
 {
@@ -81,11 +83,13 @@ static int vmw_gmr2_bind(struct vmw_private *dev_priv,
 
 		for (i = 0; i < nr; ++i) {
 			if (VMW_PPN_SIZE <= 4)
-				*cmd = page_to_pfn(*pages++);
+				*cmd = vmw_piter_dma_addr(iter) >> PAGE_SHIFT;
 			else
-				*((uint64_t *)cmd) = page_to_pfn(*pages++);
+				*((uint64_t *)cmd) = vmw_piter_dma_addr(iter) >>
+					PAGE_SHIFT;
 
 			cmd += VMW_PPN_SIZE / sizeof(*cmd);
+			vmw_piter_next(iter);
 		}
 
 		num_pages -= nr;
@@ -120,22 +124,54 @@ static void vmw_gmr2_unbind(struct vmw_private *dev_priv,
 	vmw_fifo_commit(dev_priv, define_size);
 }
 
+
+static void vmw_gmr_free_descriptors(struct device *dev, dma_addr_t desc_dma,
+				     struct list_head *desc_pages)
+{
+	struct page *page, *next;
+	struct svga_guest_mem_descriptor *page_virtual;
+	unsigned int desc_per_page = PAGE_SIZE /
+		sizeof(struct svga_guest_mem_descriptor) - 1;
+
+	if (list_empty(desc_pages))
+		return;
+
+	list_for_each_entry_safe(page, next, desc_pages, lru) {
+		list_del_init(&page->lru);
+
+		if (likely(desc_dma != DMA_ADDR_INVALID)) {
+			dma_unmap_page(dev, desc_dma, PAGE_SIZE,
+				       DMA_TO_DEVICE);
+		}
+
+		page_virtual = kmap_atomic(page);
+		desc_dma = page_virtual[desc_per_page].ppn << PAGE_SHIFT;
+		kunmap_atomic(page_virtual);
+
+		__free_page(page);
+	}
+}
+
 /**
  * FIXME: Adjust to the ttm lowmem / highmem storage to minimize
  * the number of used descriptors.
+ *
  */
 
-static int vmw_gmr_build_descriptors(struct list_head *desc_pages,
-				     struct page *pages[],
-				     unsigned long num_pages)
+static int vmw_gmr_build_descriptors(struct device *dev,
+				     struct list_head *desc_pages,
+				     struct vmw_piter *iter,
+				     unsigned long num_pages,
+				     dma_addr_t *first_dma)
 {
-	struct page *page, *next;
+	struct page *page;
 	struct svga_guest_mem_descriptor *page_virtual = NULL;
 	struct svga_guest_mem_descriptor *desc_virtual = NULL;
 	unsigned int desc_per_page;
 	unsigned long prev_pfn;
 	unsigned long pfn;
 	int ret;
+	dma_addr_t desc_dma;
 
 	desc_per_page = PAGE_SIZE /
 	    sizeof(struct svga_guest_mem_descriptor) - 1;
@@ -148,23 +184,12 @@ static int vmw_gmr_build_descriptors(struct list_head *desc_pages,
 		}
 
 		list_add_tail(&page->lru, desc_pages);
-
-		/*
-		 * Point previous page terminating descriptor to this
-		 * page before unmapping it.
-		 */
-
-		if (likely(page_virtual != NULL)) {
-			desc_virtual->ppn = page_to_pfn(page);
-			kunmap_atomic(page_virtual);
-		}
-
 		page_virtual = kmap_atomic(page);
 		desc_virtual = page_virtual - 1;
 		prev_pfn = ~(0UL);
 
 		while (likely(num_pages != 0)) {
-			pfn = page_to_pfn(*pages);
+			pfn = vmw_piter_dma_addr(iter) >> PAGE_SHIFT;
 
 			if (pfn != prev_pfn + 1) {
 
@@ -181,104 +206,81 @@ static int vmw_gmr_build_descriptors(struct list_head *desc_pages,
 			}
 			prev_pfn = pfn;
 			--num_pages;
-			++pages;
+			vmw_piter_next(iter);
 		}
 
-		(++desc_virtual)->ppn = cpu_to_le32(0);
+		(++desc_virtual)->ppn = DMA_PAGE_INVALID;
 		desc_virtual->num_pages = cpu_to_le32(0);
+		kunmap_atomic(page_virtual);
 	}
 
-	if (likely(page_virtual != NULL))
+	desc_dma = 0;
+	list_for_each_entry_reverse(page, desc_pages, lru) {
+		page_virtual = kmap_atomic(page);
+		page_virtual[desc_per_page].ppn = desc_dma >> PAGE_SHIFT;
 		kunmap_atomic(page_virtual);
+		desc_dma = dma_map_page(dev, page, 0, PAGE_SIZE,
+					DMA_TO_DEVICE);
+
+		if (unlikely(dma_mapping_error(dev, desc_dma)))
+			goto out_err;
+	}
+	*first_dma = desc_dma;
 
 	return 0;
 out_err:
-	list_for_each_entry_safe(page, next, desc_pages, lru) {
-		list_del_init(&page->lru);
-		__free_page(page);
-	}
+	vmw_gmr_free_descriptors(dev, DMA_ADDR_INVALID, desc_pages);
 	return ret;
 }
 
-static inline void vmw_gmr_free_descriptors(struct list_head *desc_pages)
-{
-	struct page *page, *next;
-
-	list_for_each_entry_safe(page, next, desc_pages, lru) {
-		list_del_init(&page->lru);
-		__free_page(page);
-	}
-}
-
 static void vmw_gmr_fire_descriptors(struct vmw_private *dev_priv,
-				     int gmr_id, struct list_head *desc_pages)
+				     int gmr_id, dma_addr_t desc_dma)
 {
-	struct page *page;
-
-	if (unlikely(list_empty(desc_pages)))
-		return;
-
-	page = list_entry(desc_pages->next, struct page, lru);
-
 	mutex_lock(&dev_priv->hw_mutex);
 
 	vmw_write(dev_priv, SVGA_REG_GMR_ID, gmr_id);
 	wmb();
-	vmw_write(dev_priv, SVGA_REG_GMR_DESCRIPTOR, page_to_pfn(page));
+	vmw_write(dev_priv, SVGA_REG_GMR_DESCRIPTOR, desc_dma >> PAGE_SHIFT);
 	mb();
 
 	mutex_unlock(&dev_priv->hw_mutex);
 
 }
 
-/**
- * FIXME: Adjust to the ttm lowmem / highmem storage to minimize
- * the number of used descriptors.
- */
-
-static unsigned long vmw_gmr_count_descriptors(struct page *pages[],
-					unsigned long num_pages)
-{
-	unsigned long prev_pfn = ~(0UL);
-	unsigned long pfn;
-	unsigned long descriptors = 0;
-
-	while (num_pages--) {
-		pfn = page_to_pfn(*pages++);
-		if (prev_pfn + 1 != pfn)
-			++descriptors;
-		prev_pfn = pfn;
-	}
-
-	return descriptors;
-}
-
 int vmw_gmr_bind(struct vmw_private *dev_priv,
-		 struct page *pages[],
+		 const struct vmw_sg_table *vsgt,
 		 unsigned long num_pages,
 		 int gmr_id)
 {
 	struct list_head desc_pages;
+	dma_addr_t desc_dma = 0;
+	struct device *dev = dev_priv->dev->dev;
+	struct vmw_piter data_iter;
 	int ret;
 
+	vmw_piter_start(&data_iter, vsgt, 0);
+
+	if (unlikely(!vmw_piter_next(&data_iter)))
+		return 0;
+
 	if (likely(dev_priv->capabilities & SVGA_CAP_GMR2))
-		return vmw_gmr2_bind(dev_priv, pages, num_pages, gmr_id);
+		return vmw_gmr2_bind(dev_priv, &data_iter, num_pages, gmr_id);
 
 	if (unlikely(!(dev_priv->capabilities & SVGA_CAP_GMR)))
 		return -EINVAL;
 
-	if (vmw_gmr_count_descriptors(pages, num_pages) >
-	    dev_priv->max_gmr_descriptors)
+	if (vsgt->num_regions > dev_priv->max_gmr_descriptors)
 		return -EINVAL;
 
 	INIT_LIST_HEAD(&desc_pages);
 
-	ret = vmw_gmr_build_descriptors(&desc_pages, pages, num_pages);
+	ret = vmw_gmr_build_descriptors(dev, &desc_pages, &data_iter,
+					num_pages, &desc_dma);
 	if (unlikely(ret != 0))
 		return ret;
 
-	vmw_gmr_fire_descriptors(dev_priv, gmr_id, &desc_pages);
-	vmw_gmr_free_descriptors(&desc_pages);
+	vmw_gmr_fire_descriptors(dev_priv, gmr_id, desc_dma);
+	vmw_gmr_free_descriptors(dev, desc_dma, &desc_pages);
 
 	return 0;
 }
diff --git a/include/drm/ttm/ttm_page_alloc.h b/include/drm/ttm/ttm_page_alloc.h
index 706b962c6467..d1f61bfe0ebe 100644
--- a/include/drm/ttm/ttm_page_alloc.h
+++ b/include/drm/ttm/ttm_page_alloc.h
@@ -62,7 +62,7 @@ extern void ttm_pool_unpopulate(struct ttm_tt *ttm);
 extern int ttm_page_alloc_debugfs(struct seq_file *m, void *data);
 
 
-#ifdef CONFIG_SWIOTLB
+#if defined(CONFIG_SWIOTLB) || defined(CONFIG_INTEL_IOMMU)
 /**
  * Initialize pool allocator.
  */
@@ -94,6 +94,15 @@ static inline int ttm_dma_page_alloc_debugfs(struct seq_file *m, void *data)
 {
 	return 0;
 }
+static inline int ttm_dma_populate(struct ttm_dma_tt *ttm_dma,
+				   struct device *dev)
+{
+	return -ENOMEM;
+}
+static inline void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma,
+				      struct device *dev)
+{
+}
 #endif
 
 #endif