diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.h b/drivers/gpu/drm/nouveau/nouveau_drv.h
index 5445cef..1c15ef3 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drv.h
+++ b/drivers/gpu/drm/nouveau/nouveau_drv.h
@@ -583,6 +583,7 @@
 	uint64_t vm_end;
 	struct nouveau_gpuobj *vm_vram_pt[NV50_VM_VRAM_NR];
 	int vm_vram_pt_nr;
+	uint64_t vram_sys_base;
 
 	/* the mtrr covering the FB */
 	int fb_mtrr;
diff --git a/drivers/gpu/drm/nouveau/nouveau_mem.c b/drivers/gpu/drm/nouveau/nouveau_mem.c
index 8f3a12f..2dc09db 100644
--- a/drivers/gpu/drm/nouveau/nouveau_mem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_mem.c
@@ -285,53 +285,50 @@
 			uint32_t flags, uint64_t phys)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_gpuobj **pgt;
-	unsigned psz, pfl, pages;
+	struct nouveau_gpuobj *pgt;
+	unsigned block;
+	int i;
 
-	if (virt >= dev_priv->vm_gart_base &&
-	    (virt + size) < (dev_priv->vm_gart_base + dev_priv->vm_gart_size)) {
-		psz = 12;
-		pgt = &dev_priv->gart_info.sg_ctxdma;
-		pfl = 0x21;
-		virt -= dev_priv->vm_gart_base;
-	} else
-	if (virt >= dev_priv->vm_vram_base &&
-	    (virt + size) < (dev_priv->vm_vram_base + dev_priv->vm_vram_size)) {
-		psz = 16;
-		pgt = dev_priv->vm_vram_pt;
-		pfl = 0x01;
-		virt -= dev_priv->vm_vram_base;
-	} else {
-		NV_ERROR(dev, "Invalid address: 0x%16llx-0x%16llx\n",
-			 virt, virt + size - 1);
-		return -EINVAL;
+	virt = ((virt - dev_priv->vm_vram_base) >> 16) << 1;
+	size = (size >> 16) << 1;
+
+	phys |= ((uint64_t)flags << 32);
+	phys |= 1;
+	if (dev_priv->vram_sys_base) {
+		phys += dev_priv->vram_sys_base;
+		phys |= 0x30;
 	}
 
-	pages = size >> psz;
-
 	dev_priv->engine.instmem.prepare_access(dev, true);
-	if (flags & 0x80000000) {
-		while (pages--) {
-			struct nouveau_gpuobj *pt = pgt[virt >> 29];
-			unsigned pte = ((virt & 0x1fffffffULL) >> psz) << 1;
+	while (size) {
+		unsigned offset_h = upper_32_bits(phys);
+		unsigned offset_l = lower_32_bits(phys);
+		unsigned pte, end;
 
-			nv_wo32(dev, pt, pte++, 0x00000000);
-			nv_wo32(dev, pt, pte++, 0x00000000);
-
-			virt += (1 << psz);
+		for (i = 7; i >= 0; i--) {
+			block = 1 << (i + 1);
+			if (size >= block && !(virt & (block - 1)))
+				break;
 		}
-	} else {
-		while (pages--) {
-			struct nouveau_gpuobj *pt = pgt[virt >> 29];
-			unsigned pte = ((virt & 0x1fffffffULL) >> psz) << 1;
-			unsigned offset_h = upper_32_bits(phys) & 0xff;
-			unsigned offset_l = lower_32_bits(phys);
+		offset_l |= (i << 7);
 
-			nv_wo32(dev, pt, pte++, offset_l | pfl);
-			nv_wo32(dev, pt, pte++, offset_h | flags);
+		phys += block << 15;
+		size -= block;
 
-			phys += (1 << psz);
-			virt += (1 << psz);
+		while (block) {
+			pgt = dev_priv->vm_vram_pt[virt >> 14];
+			pte = virt & 0x3ffe;
+
+			end = pte + block;
+			if (end > 16384)
+				end = 16384;
+			block -= (end - pte);
+			virt  += (end - pte);
+
+			while (pte < end) {
+				nv_wo32(dev, pgt, pte++, offset_l);
+				nv_wo32(dev, pgt, pte++, offset_h);
+			}
 		}
 	}
 	dev_priv->engine.instmem.finish_access(dev);
@@ -356,7 +353,41 @@
 void
 nv50_mem_vm_unbind(struct drm_device *dev, uint64_t virt, uint32_t size)
 {
-	nv50_mem_vm_bind_linear(dev, virt, size, 0x80000000, 0);
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpuobj *pgt;
+	unsigned pages, pte, end;
+
+	virt -= dev_priv->vm_vram_base;
+	pages = (size >> 16) << 1;
+
+	dev_priv->engine.instmem.prepare_access(dev, true);
+	while (pages) {
+		pgt = dev_priv->vm_vram_pt[virt >> 29];
+		pte = (virt & 0x1ffe0000ULL) >> 15;
+
+		end = pte + pages;
+		if (end > 16384)
+			end = 16384;
+		pages -= (end - pte);
+		virt  += (end - pte) << 15;
+
+		while (pte < end)
+			nv_wo32(dev, pgt, pte++, 0);
+	}
+	dev_priv->engine.instmem.finish_access(dev);
+
+	nv_wr32(dev, 0x100c80, 0x00050001);
+	if (!nv_wait(0x100c80, 0x00000001, 0x00000000)) {
+		NV_ERROR(dev, "timeout: (0x100c80 & 1) == 0 (2)\n");
+		NV_ERROR(dev, "0x100c80 = 0x%08x\n", nv_rd32(dev, 0x100c80));
+		return;
+	}
+
+	nv_wr32(dev, 0x100c80, 0x00000001);
+	if (!nv_wait(0x100c80, 0x00000001, 0x00000000)) {
+		NV_ERROR(dev, "timeout: (0x100c80 & 1) == 0 (2)\n");
+		NV_ERROR(dev, "0x100c80 = 0x%08x\n", nv_rd32(dev, 0x100c80));
+	}
 }
 
 /*
diff --git a/drivers/gpu/drm/nouveau/nv04_dac.c b/drivers/gpu/drm/nouveau/nv04_dac.c
index d0e038d..1d73b15 100644
--- a/drivers/gpu/drm/nouveau/nv04_dac.c
+++ b/drivers/gpu/drm/nouveau/nv04_dac.c
@@ -119,7 +119,7 @@
 						 struct drm_connector *connector)
 {
 	struct drm_device *dev = encoder->dev;
-	uint8_t saved_seq1, saved_pi, saved_rpc1;
+	uint8_t saved_seq1, saved_pi, saved_rpc1, saved_cr_mode;
 	uint8_t saved_palette0[3], saved_palette_mask;
 	uint32_t saved_rtest_ctrl, saved_rgen_ctrl;
 	int i;
@@ -135,6 +135,9 @@
 		/* only implemented for head A for now */
 		NVSetOwner(dev, 0);
 
+	saved_cr_mode = NVReadVgaCrtc(dev, 0, NV_CIO_CR_MODE_INDEX);
+	NVWriteVgaCrtc(dev, 0, NV_CIO_CR_MODE_INDEX, saved_cr_mode | 0x80);
+
 	saved_seq1 = NVReadVgaSeq(dev, 0, NV_VIO_SR_CLOCK_INDEX);
 	NVWriteVgaSeq(dev, 0, NV_VIO_SR_CLOCK_INDEX, saved_seq1 & ~0x20);
 
@@ -203,6 +206,7 @@
 	NVWriteVgaCrtc(dev, 0, NV_CIO_CRE_PIXEL_INDEX, saved_pi);
 	NVWriteVgaCrtc(dev, 0, NV_CIO_CRE_RPC1_INDEX, saved_rpc1);
 	NVWriteVgaSeq(dev, 0, NV_VIO_SR_CLOCK_INDEX, saved_seq1);
+	NVWriteVgaCrtc(dev, 0, NV_CIO_CR_MODE_INDEX, saved_cr_mode);
 
 	if (blue == 0x18) {
 		NV_INFO(dev, "Load detected on head A\n");
diff --git a/drivers/gpu/drm/nouveau/nv50_instmem.c b/drivers/gpu/drm/nouveau/nv50_instmem.c
index 94400f7..f0dc4e3 100644
--- a/drivers/gpu/drm/nouveau/nv50_instmem.c
+++ b/drivers/gpu/drm/nouveau/nv50_instmem.c
@@ -76,6 +76,11 @@
 	for (i = 0x1700; i <= 0x1710; i += 4)
 		priv->save1700[(i-0x1700)/4] = nv_rd32(dev, i);
 
+	if (dev_priv->chipset == 0xaa || dev_priv->chipset == 0xac)
+		dev_priv->vram_sys_base = nv_rd32(dev, 0x100e10) << 12;
+	else
+		dev_priv->vram_sys_base = 0;
+
 	/* Reserve the last MiB of VRAM, we should probably try to avoid
 	 * setting up the below tables over the top of the VBIOS image at
 	 * some point.
@@ -172,16 +177,28 @@
 	 * We map the entire fake channel into the start of the PRAMIN BAR
 	 */
 	ret = nouveau_gpuobj_new_ref(dev, chan, NULL, 0, pt_size, 0x1000,
-							0, &priv->pramin_pt);
+				     0, &priv->pramin_pt);
 	if (ret)
 		return ret;
 
-	for (i = 0, v = c_offset; i < pt_size; i += 8, v += 0x1000) {
-		if (v < (c_offset + c_size))
-			BAR0_WI32(priv->pramin_pt->gpuobj, i + 0, v | 1);
-		else
-			BAR0_WI32(priv->pramin_pt->gpuobj, i + 0, 0x00000009);
+	v = c_offset | 1;
+	if (dev_priv->vram_sys_base) {
+		v += dev_priv->vram_sys_base;
+		v |= 0x30;
+	}
+
+	i = 0;
+	while (v < dev_priv->vram_sys_base + c_offset + c_size) {
+		BAR0_WI32(priv->pramin_pt->gpuobj, i + 0, v);
 		BAR0_WI32(priv->pramin_pt->gpuobj, i + 4, 0x00000000);
+		v += 0x1000;
+		i += 8;
+	}
+
+	while (i < pt_size) {
+		BAR0_WI32(priv->pramin_pt->gpuobj, i + 0, 0x00000000);
+		BAR0_WI32(priv->pramin_pt->gpuobj, i + 4, 0x00000000);
+		i += 8;
 	}
 
 	BAR0_WI32(chan->vm_pd, 0x00, priv->pramin_pt->instance | 0x63);
@@ -416,7 +433,9 @@
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nv50_instmem_priv *priv = dev_priv->engine.instmem.priv;
-	uint32_t pte, pte_end, vram;
+	struct nouveau_gpuobj *pramin_pt = priv->pramin_pt->gpuobj;
+	uint32_t pte, pte_end;
+	uint64_t vram;
 
 	if (!gpuobj->im_backing || !gpuobj->im_pramin || gpuobj->im_bound)
 		return -EINVAL;
@@ -424,20 +443,24 @@
 	NV_DEBUG(dev, "st=0x%0llx sz=0x%0llx\n",
 		 gpuobj->im_pramin->start, gpuobj->im_pramin->size);
 
-	pte     = (gpuobj->im_pramin->start >> 12) << 3;
-	pte_end = ((gpuobj->im_pramin->size >> 12) << 3) + pte;
+	pte     = (gpuobj->im_pramin->start >> 12) << 1;
+	pte_end = ((gpuobj->im_pramin->size >> 12) << 1) + pte;
 	vram    = gpuobj->im_backing_start;
 
 	NV_DEBUG(dev, "pramin=0x%llx, pte=%d, pte_end=%d\n",
 		 gpuobj->im_pramin->start, pte, pte_end);
 	NV_DEBUG(dev, "first vram page: 0x%08x\n", gpuobj->im_backing_start);
 
+	vram |= 1;
+	if (dev_priv->vram_sys_base) {
+		vram += dev_priv->vram_sys_base;
+		vram |= 0x30;
+	}
+
 	dev_priv->engine.instmem.prepare_access(dev, true);
 	while (pte < pte_end) {
-		nv_wo32(dev, priv->pramin_pt->gpuobj, (pte + 0)/4, vram | 1);
-		nv_wo32(dev, priv->pramin_pt->gpuobj, (pte + 4)/4, 0x00000000);
-
-		pte += 8;
+		nv_wo32(dev, pramin_pt, pte++, lower_32_bits(vram));
+		nv_wo32(dev, pramin_pt, pte++, upper_32_bits(vram));
 		vram += NV50_INSTMEM_PAGE_SIZE;
 	}
 	dev_priv->engine.instmem.finish_access(dev);
@@ -470,14 +493,13 @@
 	if (gpuobj->im_bound == 0)
 		return -EINVAL;
 
-	pte     = (gpuobj->im_pramin->start >> 12) << 3;
-	pte_end = ((gpuobj->im_pramin->size >> 12) << 3) + pte;
+	pte     = (gpuobj->im_pramin->start >> 12) << 1;
+	pte_end = ((gpuobj->im_pramin->size >> 12) << 1) + pte;
 
 	dev_priv->engine.instmem.prepare_access(dev, true);
 	while (pte < pte_end) {
-		nv_wo32(dev, priv->pramin_pt->gpuobj, (pte + 0)/4, 0x00000009);
-		nv_wo32(dev, priv->pramin_pt->gpuobj, (pte + 4)/4, 0x00000000);
-		pte += 8;
+		nv_wo32(dev, priv->pramin_pt->gpuobj, pte++, 0x00000000);
+		nv_wo32(dev, priv->pramin_pt->gpuobj, pte++, 0x00000000);
 	}
 	dev_priv->engine.instmem.finish_access(dev);
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
index d69caf9..0897359 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
@@ -182,25 +182,19 @@
 	return vmw_cmd_sid_check(dev_priv, sw_context, &cmd->body.sid);
 }
 
-static int vmw_cmd_dma(struct vmw_private *dev_priv,
-		       struct vmw_sw_context *sw_context,
-		       SVGA3dCmdHeader *header)
+static int vmw_translate_guest_ptr(struct vmw_private *dev_priv,
+				   struct vmw_sw_context *sw_context,
+				   SVGAGuestPtr *ptr,
+				   struct vmw_dma_buffer **vmw_bo_p)
 {
-	uint32_t handle;
 	struct vmw_dma_buffer *vmw_bo = NULL;
 	struct ttm_buffer_object *bo;
-	struct vmw_surface *srf = NULL;
-	struct vmw_dma_cmd {
-		SVGA3dCmdHeader header;
-		SVGA3dCmdSurfaceDMA dma;
-	} *cmd;
+	uint32_t handle = ptr->gmrId;
 	struct vmw_relocation *reloc;
-	int ret;
 	uint32_t cur_validate_node;
 	struct ttm_validate_buffer *val_buf;
+	int ret;
 
-	cmd = container_of(header, struct vmw_dma_cmd, header);
-	handle = cmd->dma.guest.ptr.gmrId;
 	ret = vmw_user_dmabuf_lookup(sw_context->tfile, handle, &vmw_bo);
 	if (unlikely(ret != 0)) {
 		DRM_ERROR("Could not find or use GMR region.\n");
@@ -209,14 +203,14 @@
 	bo = &vmw_bo->base;
 
 	if (unlikely(sw_context->cur_reloc >= VMWGFX_MAX_RELOCATIONS)) {
-		DRM_ERROR("Max number of DMA commands per submission"
+		DRM_ERROR("Max number relocations per submission"
 			  " exceeded\n");
 		ret = -EINVAL;
 		goto out_no_reloc;
 	}
 
 	reloc = &sw_context->relocs[sw_context->cur_reloc++];
-	reloc->location = &cmd->dma.guest.ptr;
+	reloc->location = ptr;
 
 	cur_validate_node = vmw_dmabuf_validate_node(bo, sw_context->cur_val_buf);
 	if (unlikely(cur_validate_node >= VMWGFX_MAX_GMRS)) {
@@ -234,7 +228,89 @@
 		list_add_tail(&val_buf->head, &sw_context->validate_nodes);
 		++sw_context->cur_val_buf;
 	}
+	*vmw_bo_p = vmw_bo;
+	return 0;
 
+out_no_reloc:
+	vmw_dmabuf_unreference(&vmw_bo);
+	vmw_bo_p = NULL;
+	return ret;
+}
+
+static int vmw_cmd_end_query(struct vmw_private *dev_priv,
+			     struct vmw_sw_context *sw_context,
+			     SVGA3dCmdHeader *header)
+{
+	struct vmw_dma_buffer *vmw_bo;
+	struct vmw_query_cmd {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdEndQuery q;
+	} *cmd;
+	int ret;
+
+	cmd = container_of(header, struct vmw_query_cmd, header);
+	ret = vmw_cmd_cid_check(dev_priv, sw_context, header);
+	if (unlikely(ret != 0))
+		return ret;
+
+	ret = vmw_translate_guest_ptr(dev_priv, sw_context,
+				      &cmd->q.guestResult,
+				      &vmw_bo);
+	if (unlikely(ret != 0))
+		return ret;
+
+	vmw_dmabuf_unreference(&vmw_bo);
+	return 0;
+}
+
+static int vmw_cmd_wait_query(struct vmw_private *dev_priv,
+			      struct vmw_sw_context *sw_context,
+			      SVGA3dCmdHeader *header)
+{
+	struct vmw_dma_buffer *vmw_bo;
+	struct vmw_query_cmd {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdWaitForQuery q;
+	} *cmd;
+	int ret;
+
+	cmd = container_of(header, struct vmw_query_cmd, header);
+	ret = vmw_cmd_cid_check(dev_priv, sw_context, header);
+	if (unlikely(ret != 0))
+		return ret;
+
+	ret = vmw_translate_guest_ptr(dev_priv, sw_context,
+				      &cmd->q.guestResult,
+				      &vmw_bo);
+	if (unlikely(ret != 0))
+		return ret;
+
+	vmw_dmabuf_unreference(&vmw_bo);
+	return 0;
+}
+
+
+static int vmw_cmd_dma(struct vmw_private *dev_priv,
+		       struct vmw_sw_context *sw_context,
+		       SVGA3dCmdHeader *header)
+{
+	struct vmw_dma_buffer *vmw_bo = NULL;
+	struct ttm_buffer_object *bo;
+	struct vmw_surface *srf = NULL;
+	struct vmw_dma_cmd {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdSurfaceDMA dma;
+	} *cmd;
+	int ret;
+
+	cmd = container_of(header, struct vmw_dma_cmd, header);
+	ret = vmw_translate_guest_ptr(dev_priv, sw_context,
+				      &cmd->dma.guest.ptr,
+				      &vmw_bo);
+	if (unlikely(ret != 0))
+		return ret;
+
+	bo = &vmw_bo->base;
 	ret = vmw_user_surface_lookup_handle(dev_priv, sw_context->tfile,
 					     cmd->dma.host.sid, &srf);
 	if (ret) {
@@ -379,8 +455,8 @@
 	VMW_CMD_DEF(SVGA_3D_CMD_DRAW_PRIMITIVES, &vmw_cmd_draw),
 	VMW_CMD_DEF(SVGA_3D_CMD_SETSCISSORRECT, &vmw_cmd_cid_check),
 	VMW_CMD_DEF(SVGA_3D_CMD_BEGIN_QUERY, &vmw_cmd_cid_check),
-	VMW_CMD_DEF(SVGA_3D_CMD_END_QUERY, &vmw_cmd_cid_check),
-	VMW_CMD_DEF(SVGA_3D_CMD_WAIT_FOR_QUERY, &vmw_cmd_cid_check),
+	VMW_CMD_DEF(SVGA_3D_CMD_END_QUERY, &vmw_cmd_end_query),
+	VMW_CMD_DEF(SVGA_3D_CMD_WAIT_FOR_QUERY, &vmw_cmd_wait_query),
 	VMW_CMD_DEF(SVGA_3D_CMD_PRESENT_READBACK, &vmw_cmd_ok),
 	VMW_CMD_DEF(SVGA_3D_CMD_BLIT_SURFACE_TO_SCREEN,
 		    &vmw_cmd_blt_surf_screen_check)