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

A relative large set of various improvements for vmwgfx. Some of them
have been around for a while, some are relatively new, but functionality
should have been tested in our standalone repo.

* 'vmwgfx-next' of git://people.freedesktop.org/~thomash/linux:
  drm/vmwgfx: Bump version patchlevel and date
  drm/vmwgfx: use monotonic event timestamps
  drm/vmwgfx: Unpin the screen object backup buffer when not used
  drm/vmwgfx: Stricter count of legacy surface device resources
  drm/vmwgfx: Use kasprintf
  drm/vmwgfx: Get rid of the device-private suspended member
  drm/vmwgfx: Improve on hibernation
  drm/vmwgfx: Avoid pinning fbdev framebuffers
  drm/vmwgfx: Fix multiple command buffer context use
  drm/vmwgfx: Use the cpu blit utility for framebuffer to screen target blits
  drm/vmwgfx: Add a cpu blit utility that can be used for page-backed bos
  drm/ttm: Export the ttm_k[un]map_atomic_prot API.
  drm/ttm: Clean up kmap_atomic_prot selection code
  drm/vmwgfx: Cursor update fixes
  drm/vmwgfx: Send the correct nonblock option for atomic_commit
  drm/vmwgfx: Move the stdu vblank event to atomic function
  drm/vmwgfx: Move screen object page flip to atomic function
  drm/vmwgfx: Remove drm_crtc_arm_vblank_event from atomic flush
  drm/vmwgfx: Move surface copy cmd to atomic function
  drm/vmwgfx: Avoid iterating over display unit if crtc is available

18 files changed, 1223 insertions, 456 deletions

diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c b/drivers/gpu/drm/ttm/ttm_bo_util.c
index 1f730b3..2ebbae6 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_util.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_util.c
@@ -255,6 +255,54 @@ static int ttm_copy_io_page(void *dst, void *src, unsigned long page)
 	return 0;
 }
 
+#ifdef CONFIG_X86
+#define __ttm_kmap_atomic_prot(__page, __prot) kmap_atomic_prot(__page, __prot)
+#define __ttm_kunmap_atomic(__addr) kunmap_atomic(__addr)
+#else
+#define __ttm_kmap_atomic_prot(__page, __prot) vmap(&__page, 1, 0,  __prot)
+#define __ttm_kunmap_atomic(__addr) vunmap(__addr)
+#endif
+
+
+/**
+ * ttm_kmap_atomic_prot - Efficient kernel map of a single page with
+ * specified page protection.
+ *
+ * @page: The page to map.
+ * @prot: The page protection.
+ *
+ * This function maps a TTM page using the kmap_atomic api if available,
+ * otherwise falls back to vmap. The user must make sure that the
+ * specified page does not have an aliased mapping with a different caching
+ * policy unless the architecture explicitly allows it. Also mapping and
+ * unmapping using this api must be correctly nested. Unmapping should
+ * occur in the reverse order of mapping.
+ */
+void *ttm_kmap_atomic_prot(struct page *page, pgprot_t prot)
+{
+	if (pgprot_val(prot) == pgprot_val(PAGE_KERNEL))
+		return kmap_atomic(page);
+	else
+		return __ttm_kmap_atomic_prot(page, prot);
+}
+EXPORT_SYMBOL(ttm_kmap_atomic_prot);
+
+/**
+ * ttm_kunmap_atomic_prot - Unmap a page that was mapped using
+ * ttm_kmap_atomic_prot.
+ *
+ * @addr: The virtual address from the map.
+ * @prot: The page protection.
+ */
+void ttm_kunmap_atomic_prot(void *addr, pgprot_t prot)
+{
+	if (pgprot_val(prot) == pgprot_val(PAGE_KERNEL))
+		kunmap_atomic(addr);
+	else
+		__ttm_kunmap_atomic(addr);
+}
+EXPORT_SYMBOL(ttm_kunmap_atomic_prot);
+
 static int ttm_copy_io_ttm_page(struct ttm_tt *ttm, void *src,
 				unsigned long page,
 				pgprot_t prot)
@@ -266,28 +314,13 @@ static int ttm_copy_io_ttm_page(struct ttm_tt *ttm, void *src,
 		return -ENOMEM;
 
 	src = (void *)((unsigned long)src + (page << PAGE_SHIFT));
-
-#ifdef CONFIG_X86
-	dst = kmap_atomic_prot(d, prot);
-#else
-	if (pgprot_val(prot) != pgprot_val(PAGE_KERNEL))
-		dst = vmap(&d, 1, 0, prot);
-	else
-		dst = kmap(d);
-#endif
+	dst = ttm_kmap_atomic_prot(d, prot);
 	if (!dst)
 		return -ENOMEM;
 
 	memcpy_fromio(dst, src, PAGE_SIZE);
 
-#ifdef CONFIG_X86
-	kunmap_atomic(dst);
-#else
-	if (pgprot_val(prot) != pgprot_val(PAGE_KERNEL))
-		vunmap(dst);
-	else
-		kunmap(d);
-#endif
+	ttm_kunmap_atomic_prot(dst, prot);
 
 	return 0;
 }
@@ -303,27 +336,13 @@ static int ttm_copy_ttm_io_page(struct ttm_tt *ttm, void *dst,
 		return -ENOMEM;
 
 	dst = (void *)((unsigned long)dst + (page << PAGE_SHIFT));
-#ifdef CONFIG_X86
-	src = kmap_atomic_prot(s, prot);
-#else
-	if (pgprot_val(prot) != pgprot_val(PAGE_KERNEL))
-		src = vmap(&s, 1, 0, prot);
-	else
-		src = kmap(s);
-#endif
+	src = ttm_kmap_atomic_prot(s, prot);
 	if (!src)
 		return -ENOMEM;
 
 	memcpy_toio(dst, src, PAGE_SIZE);
 
-#ifdef CONFIG_X86
-	kunmap_atomic(src);
-#else
-	if (pgprot_val(prot) != pgprot_val(PAGE_KERNEL))
-		vunmap(src);
-	else
-		kunmap(s);
-#endif
+	ttm_kunmap_atomic_prot(src, prot);
 
 	return 0;
 }
diff --git a/drivers/gpu/drm/vmwgfx/Makefile b/drivers/gpu/drm/vmwgfx/Makefile
index ad80211..794cc9d 100644
--- a/drivers/gpu/drm/vmwgfx/Makefile
+++ b/drivers/gpu/drm/vmwgfx/Makefile
@@ -7,6 +7,6 @@ vmwgfx-y := vmwgfx_execbuf.o vmwgfx_gmr.o vmwgfx_kms.o vmwgfx_drv.o \
 	    vmwgfx_surface.o vmwgfx_prime.o vmwgfx_mob.o vmwgfx_shader.o \
 	    vmwgfx_cmdbuf_res.o vmwgfx_cmdbuf.o vmwgfx_stdu.o \
 	    vmwgfx_cotable.o vmwgfx_so.o vmwgfx_binding.o vmwgfx_msg.o \
-	    vmwgfx_simple_resource.o vmwgfx_va.o
+	    vmwgfx_simple_resource.o vmwgfx_va.o vmwgfx_blit.o
 
 obj-$(CONFIG_DRM_VMWGFX) := vmwgfx.o
diff --git a/drivers/gpu/drm/vmwgfx/device_include/svga_reg.h b/drivers/gpu/drm/vmwgfx/device_include/svga_reg.h
index 6e0ccb7..88e72bf 100644
--- a/drivers/gpu/drm/vmwgfx/device_include/svga_reg.h
+++ b/drivers/gpu/drm/vmwgfx/device_include/svga_reg.h
@@ -372,6 +372,14 @@ SVGAGuestPtr;
  * PA, not biased by the offset.  When the command buffer is finished
  * the guest should not read the offset field as there is no guarantee
  * what it will set to.
+ *
+ * When the SVGA_CAP_HP_CMD_QUEUE cap bit is set a new command queue
+ * SVGA_CB_CONTEXT_1 is available.  Commands submitted to this queue
+ * will be executed as quickly as possible by the SVGA device
+ * potentially before already queued commands on SVGA_CB_CONTEXT_0.
+ * The SVGA device guarantees that any command buffers submitted to
+ * SVGA_CB_CONTEXT_0 will be executed after any _already_ submitted
+ * command buffers to SVGA_CB_CONTEXT_1.
  */
 
 #define SVGA_CB_MAX_SIZE (512 * 1024)  /* 512 KB */
@@ -382,7 +390,8 @@ SVGAGuestPtr;
 typedef enum {
    SVGA_CB_CONTEXT_DEVICE = 0x3f,
    SVGA_CB_CONTEXT_0      = 0x0,
-   SVGA_CB_CONTEXT_MAX    = 0x1,
+   SVGA_CB_CONTEXT_1      = 0x1, /* Supported with SVGA_CAP_HP_CMD_QUEUE */
+   SVGA_CB_CONTEXT_MAX    = 0x2,
 } SVGACBContext;
 
 
@@ -689,6 +698,7 @@ SVGASignedPoint;
 #define SVGA_CAP_CMD_BUFFERS_2      0x04000000
 #define SVGA_CAP_GBOBJECTS          0x08000000
 #define SVGA_CAP_DX                 0x10000000
+#define SVGA_CAP_HP_CMD_QUEUE       0x20000000
 
 #define SVGA_CAP_CMD_RESERVED       0x80000000
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_blit.c b/drivers/gpu/drm/vmwgfx/vmwgfx_blit.c
new file mode 100644
index 0000000..e8c94b1
--- /dev/null
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_blit.c
@@ -0,0 +1,506 @@
+/**************************************************************************
+ *
+ * Copyright © 2017 VMware, Inc., Palo Alto, CA., USA
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#include "vmwgfx_drv.h"
+
+/*
+ * Template that implements find_first_diff() for a generic
+ * unsigned integer type. @size and return value are in bytes.
+ */
+#define VMW_FIND_FIRST_DIFF(_type)			 \
+static size_t vmw_find_first_diff_ ## _type		 \
+	(const _type * dst, const _type * src, size_t size)\
+{							 \
+	size_t i;					 \
+							 \
+	for (i = 0; i < size; i += sizeof(_type)) {	 \
+		if (*dst++ != *src++)			 \
+			break;				 \
+	}						 \
+							 \
+	return i;					 \
+}
+
+
+/*
+ * Template that implements find_last_diff() for a generic
+ * unsigned integer type. Pointers point to the item following the
+ * *end* of the area to be examined. @size and return value are in
+ * bytes.
+ */
+#define VMW_FIND_LAST_DIFF(_type)					\
+static ssize_t vmw_find_last_diff_ ## _type(				\
+	const _type * dst, const _type * src, size_t size)		\
+{									\
+	while (size) {							\
+		if (*--dst != *--src)					\
+			break;						\
+									\
+		size -= sizeof(_type);					\
+	}								\
+	return size;							\
+}
+
+
+/*
+ * Instantiate find diff functions for relevant unsigned integer sizes,
+ * assuming that wider integers are faster (including aligning) up to the
+ * architecture native width, which is assumed to be 32 bit unless
+ * CONFIG_64BIT is defined.
+ */
+VMW_FIND_FIRST_DIFF(u8);
+VMW_FIND_LAST_DIFF(u8);
+
+VMW_FIND_FIRST_DIFF(u16);
+VMW_FIND_LAST_DIFF(u16);
+
+VMW_FIND_FIRST_DIFF(u32);
+VMW_FIND_LAST_DIFF(u32);
+
+#ifdef CONFIG_64BIT
+VMW_FIND_FIRST_DIFF(u64);
+VMW_FIND_LAST_DIFF(u64);
+#endif
+
+
+/* We use size aligned copies. This computes (addr - align(addr)) */
+#define SPILL(_var, _type) ((unsigned long) _var & (sizeof(_type) - 1))
+
+
+/*
+ * Template to compute find_first_diff() for a certain integer type
+ * including a head copy for alignment, and adjustment of parameters
+ * for tail find or increased resolution find using an unsigned integer find
+ * of smaller width. If finding is complete, and resolution is sufficient,
+ * the macro executes a return statement. Otherwise it falls through.
+ */
+#define VMW_TRY_FIND_FIRST_DIFF(_type)					\
+do {									\
+	unsigned int spill = SPILL(dst, _type);				\
+	size_t diff_offs;						\
+									\
+	if (spill && spill == SPILL(src, _type) &&			\
+	    sizeof(_type) - spill <= size) {				\
+		spill = sizeof(_type) - spill;				\
+		diff_offs = vmw_find_first_diff_u8(dst, src, spill);	\
+		if (diff_offs < spill)					\
+			return round_down(offset + diff_offs, granularity); \
+									\
+		dst += spill;						\
+		src += spill;						\
+		size -= spill;						\
+		offset += spill;					\
+		spill = 0;						\
+	}								\
+	if (!spill && !SPILL(src, _type)) {				\
+		size_t to_copy = size &	 ~(sizeof(_type) - 1);		\
+									\
+		diff_offs = vmw_find_first_diff_ ## _type		\
+			((_type *) dst, (_type *) src, to_copy);	\
+		if (diff_offs >= size || granularity == sizeof(_type))	\
+			return (offset + diff_offs);			\
+									\
+		dst += diff_offs;					\
+		src += diff_offs;					\
+		size -= diff_offs;					\
+		offset += diff_offs;					\
+	}								\
+} while (0)								\
+
+
+/**
+ * vmw_find_first_diff - find the first difference between dst and src
+ *
+ * @dst: The destination address
+ * @src: The source address
+ * @size: Number of bytes to compare
+ * @granularity: The granularity needed for the return value in bytes.
+ * return: The offset from find start where the first difference was
+ * encountered in bytes. If no difference was found, the function returns
+ * a value >= @size.
+ */
+static size_t vmw_find_first_diff(const u8 *dst, const u8 *src, size_t size,
+				  size_t granularity)
+{
+	size_t offset = 0;
+
+	/*
+	 * Try finding with large integers if alignment allows, or we can
+	 * fix it. Fall through if we need better resolution or alignment
+	 * was bad.
+	 */
+#ifdef CONFIG_64BIT
+	VMW_TRY_FIND_FIRST_DIFF(u64);
+#endif
+	VMW_TRY_FIND_FIRST_DIFF(u32);
+	VMW_TRY_FIND_FIRST_DIFF(u16);
+
+	return round_down(offset + vmw_find_first_diff_u8(dst, src, size),
+			  granularity);
+}
+
+
+/*
+ * Template to compute find_last_diff() for a certain integer type
+ * including a tail copy for alignment, and adjustment of parameters
+ * for head find or increased resolution find using an unsigned integer find
+ * of smaller width. If finding is complete, and resolution is sufficient,
+ * the macro executes a return statement. Otherwise it falls through.
+ */
+#define VMW_TRY_FIND_LAST_DIFF(_type)					\
+do {									\
+	unsigned int spill = SPILL(dst, _type);				\
+	ssize_t location;						\
+	ssize_t diff_offs;						\
+									\
+	if (spill && spill <= size && spill == SPILL(src, _type)) {	\
+		diff_offs = vmw_find_last_diff_u8(dst, src, spill);	\
+		if (diff_offs) {					\
+			location = size - spill + diff_offs - 1;	\
+			return round_down(location, granularity);	\
+		}							\
+									\
+		dst -= spill;						\
+		src -= spill;						\
+		size -= spill;						\
+		spill = 0;						\
+	}								\
+	if (!spill && !SPILL(src, _type)) {				\
+		size_t to_copy = round_down(size, sizeof(_type));	\
+									\
+		diff_offs = vmw_find_last_diff_ ## _type		\
+			((_type *) dst, (_type *) src, to_copy);	\
+		location = size - to_copy + diff_offs - sizeof(_type);	\
+		if (location < 0 || granularity == sizeof(_type))	\
+			return location;				\
+									\
+		dst -= to_copy - diff_offs;				\
+		src -= to_copy - diff_offs;				\
+		size -= to_copy - diff_offs;				\
+	}								\
+} while (0)
+
+
+/**
+ * vmw_find_last_diff - find the last difference between dst and src
+ *
+ * @dst: The destination address
+ * @src: The source address
+ * @size: Number of bytes to compare
+ * @granularity: The granularity needed for the return value in bytes.
+ * return: The offset from find start where the last difference was
+ * encountered in bytes, or a negative value if no difference was found.
+ */
+static ssize_t vmw_find_last_diff(const u8 *dst, const u8 *src, size_t size,
+				  size_t granularity)
+{
+	dst += size;
+	src += size;
+
+#ifdef CONFIG_64BIT
+	VMW_TRY_FIND_LAST_DIFF(u64);
+#endif
+	VMW_TRY_FIND_LAST_DIFF(u32);
+	VMW_TRY_FIND_LAST_DIFF(u16);
+
+	return round_down(vmw_find_last_diff_u8(dst, src, size) - 1,
+			  granularity);
+}
+
+
+/**
+ * vmw_memcpy - A wrapper around kernel memcpy with allowing to plug it into a
+ * struct vmw_diff_cpy.
+ *
+ * @diff: The struct vmw_diff_cpy closure argument (unused).
+ * @dest: The copy destination.
+ * @src: The copy source.
+ * @n: Number of bytes to copy.
+ */
+void vmw_memcpy(struct vmw_diff_cpy *diff, u8 *dest, const u8 *src, size_t n)
+{
+	memcpy(dest, src, n);
+}
+
+
+/**
+ * vmw_adjust_rect - Adjust rectangle coordinates for newly found difference
+ *
+ * @diff: The struct vmw_diff_cpy used to track the modified bounding box.
+ * @diff_offs: The offset from @diff->line_offset where the difference was
+ * found.
+ */
+static void vmw_adjust_rect(struct vmw_diff_cpy *diff, size_t diff_offs)
+{
+	size_t offs = (diff_offs + diff->line_offset) / diff->cpp;
+	struct drm_rect *rect = &diff->rect;
+
+	rect->x1 = min_t(int, rect->x1, offs);
+	rect->x2 = max_t(int, rect->x2, offs + 1);
+	rect->y1 = min_t(int, rect->y1, diff->line);
+	rect->y2 = max_t(int, rect->y2, diff->line + 1);
+}
+
+/**
+ * vmw_diff_memcpy - memcpy that creates a bounding box of modified content.
+ *
+ * @diff: The struct vmw_diff_cpy used to track the modified bounding box.
+ * @dest: The copy destination.
+ * @src: The copy source.
+ * @n: Number of bytes to copy.
+ *
+ * In order to correctly track the modified content, the field @diff->line must
+ * be pre-loaded with the current line number, the field @diff->line_offset must
+ * be pre-loaded with the line offset in bytes where the copy starts, and
+ * finally the field @diff->cpp need to be preloaded with the number of bytes
+ * per unit in the horizontal direction of the area we're examining.
+ * Typically bytes per pixel.
+ * This is needed to know the needed granularity of the difference computing
+ * operations. A higher cpp generally leads to faster execution at the cost of
+ * bounding box width precision.
+ */
+void vmw_diff_memcpy(struct vmw_diff_cpy *diff, u8 *dest, const u8 *src,
+		     size_t n)
+{
+	ssize_t csize, byte_len;
+
+	if (WARN_ON_ONCE(round_down(n, diff->cpp) != n))
+		return;
+
+	/* TODO: Possibly use a single vmw_find_first_diff per line? */
+	csize = vmw_find_first_diff(dest, src, n, diff->cpp);
+	if (csize < n) {
+		vmw_adjust_rect(diff, csize);
+		byte_len = diff->cpp;
+
+		/*
+		 * Starting from where first difference was found, find
+		 * location of last difference, and then copy.
+		 */
+		diff->line_offset += csize;
+		dest += csize;
+		src += csize;
+		n -= csize;
+		csize = vmw_find_last_diff(dest, src, n, diff->cpp);
+		if (csize >= 0) {
+			byte_len += csize;
+			vmw_adjust_rect(diff, csize);
+		}
+		memcpy(dest, src, byte_len);
+	}
+	diff->line_offset += n;
+}
+
+/**
+ * struct vmw_bo_blit_line_data - Convenience argument to vmw_bo_cpu_blit_line
+ *
+ * @mapped_dst: Already mapped destination page index in @dst_pages.
+ * @dst_addr: Kernel virtual address of mapped destination page.
+ * @dst_pages: Array of destination bo pages.
+ * @dst_num_pages: Number of destination bo pages.
+ * @dst_prot: Destination bo page protection.
+ * @mapped_src: Already mapped source page index in @dst_pages.
+ * @src_addr: Kernel virtual address of mapped source page.
+ * @src_pages: Array of source bo pages.
+ * @src_num_pages: Number of source bo pages.
+ * @src_prot: Source bo page protection.
+ * @diff: Struct vmw_diff_cpy, in the end forwarded to the memcpy routine.
+ */
+struct vmw_bo_blit_line_data {
+	u32 mapped_dst;
+	u8 *dst_addr;
+	struct page **dst_pages;
+	u32 dst_num_pages;
+	pgprot_t dst_prot;
+	u32 mapped_src;
+	u8 *src_addr;
+	struct page **src_pages;
+	u32 src_num_pages;
+	pgprot_t src_prot;
+	struct vmw_diff_cpy *diff;
+};
+
+/**
+ * vmw_bo_cpu_blit_line - Blit part of a line from one bo to another.
+ *
+ * @d: Blit data as described above.
+ * @dst_offset: Destination copy start offset from start of bo.
+ * @src_offset: Source copy start offset from start of bo.
+ * @bytes_to_copy: Number of bytes to copy in this line.
+ */
+static int vmw_bo_cpu_blit_line(struct vmw_bo_blit_line_data *d,
+				u32 dst_offset,
+				u32 src_offset,
+				u32 bytes_to_copy)
+{
+	struct vmw_diff_cpy *diff = d->diff;
+
+	while (bytes_to_copy) {
+		u32 copy_size = bytes_to_copy;
+		u32 dst_page = dst_offset >> PAGE_SHIFT;
+		u32 src_page = src_offset >> PAGE_SHIFT;
+		u32 dst_page_offset = dst_offset & ~PAGE_MASK;
+		u32 src_page_offset = src_offset & ~PAGE_MASK;
+		bool unmap_dst = d->dst_addr && dst_page != d->mapped_dst;
+		bool unmap_src = d->src_addr && (src_page != d->mapped_src ||
+						 unmap_dst);
+
+		copy_size = min_t(u32, copy_size, PAGE_SIZE - dst_page_offset);
+		copy_size = min_t(u32, copy_size, PAGE_SIZE - src_page_offset);
+
+		if (unmap_src) {
+			ttm_kunmap_atomic_prot(d->src_addr, d->src_prot);
+			d->src_addr = NULL;
+		}
+
+		if (unmap_dst) {
+			ttm_kunmap_atomic_prot(d->dst_addr, d->dst_prot);
+			d->dst_addr = NULL;
+		}
+
+		if (!d->dst_addr) {
+			if (WARN_ON_ONCE(dst_page >= d->dst_num_pages))
+				return -EINVAL;
+
+			d->dst_addr =
+				ttm_kmap_atomic_prot(d->dst_pages[dst_page],
+						     d->dst_prot);
+			if (!d->dst_addr)
+				return -ENOMEM;
+
+			d->mapped_dst = dst_page;
+		}
+
+		if (!d->src_addr) {
+			if (WARN_ON_ONCE(src_page >= d->src_num_pages))
+				return -EINVAL;
+
+			d->src_addr =
+				ttm_kmap_atomic_prot(d->src_pages[src_page],
+						     d->src_prot);
+			if (!d->src_addr)
+				return -ENOMEM;
+
+			d->mapped_src = src_page;
+		}
+		diff->do_cpy(diff, d->dst_addr + dst_page_offset,
+			     d->src_addr + src_page_offset, copy_size);
+
+		bytes_to_copy -= copy_size;
+		dst_offset += copy_size;
+		src_offset += copy_size;
+	}
+
+	return 0;
+}
+
+/**
+ * ttm_bo_cpu_blit - in-kernel cpu blit.
+ *
+ * @dst: Destination buffer object.
+ * @dst_offset: Destination offset of blit start in bytes.
+ * @dst_stride: Destination stride in bytes.
+ * @src: Source buffer object.
+ * @src_offset: Source offset of blit start in bytes.
+ * @src_stride: Source stride in bytes.
+ * @w: Width of blit.
+ * @h: Height of blit.
+ * return: Zero on success. Negative error value on failure. Will print out
+ * kernel warnings on caller bugs.
+ *
+ * Performs a CPU blit from one buffer object to another avoiding a full
+ * bo vmap which may exhaust- or fragment vmalloc space.
+ * On supported architectures (x86), we're using kmap_atomic which avoids
+ * cross-processor TLB- and cache flushes and may, on non-HIGHMEM systems
+ * reference already set-up mappings.
+ *
+ * Neither of the buffer objects may be placed in PCI memory
+ * (Fixed memory in TTM terminology) when using this function.
+ */
+int vmw_bo_cpu_blit(struct ttm_buffer_object *dst,
+		    u32 dst_offset, u32 dst_stride,
+		    struct ttm_buffer_object *src,
+		    u32 src_offset, u32 src_stride,
+		    u32 w, u32 h,
+		    struct vmw_diff_cpy *diff)
+{
+	struct ttm_operation_ctx ctx = {
+		.interruptible = false,
+		.no_wait_gpu = false
+	};
+	u32 j, initial_line = dst_offset / dst_stride;
+	struct vmw_bo_blit_line_data d;
+	int ret = 0;
+
+	/* Buffer objects need to be either pinned or reserved: */
+	if (!(dst->mem.placement & TTM_PL_FLAG_NO_EVICT))
+		lockdep_assert_held(&dst->resv->lock.base);
+	if (!(src->mem.placement & TTM_PL_FLAG_NO_EVICT))
+		lockdep_assert_held(&src->resv->lock.base);
+
+	if (dst->ttm->state == tt_unpopulated) {
+		ret = dst->ttm->bdev->driver->ttm_tt_populate(dst->ttm, &ctx);
+		if (ret)
+			return ret;
+	}
+
+	if (src->ttm->state == tt_unpopulated) {
+		ret = src->ttm->bdev->driver->ttm_tt_populate(src->ttm, &ctx);
+		if (ret)
+			return ret;
+	}
+
+	d.mapped_dst = 0;
+	d.mapped_src = 0;
+	d.dst_addr = NULL;
+	d.src_addr = NULL;
+	d.dst_pages = dst->ttm->pages;
+	d.src_pages = src->ttm->pages;
+	d.dst_num_pages = dst->num_pages;
+	d.src_num_pages = src->num_pages;
+	d.dst_prot = ttm_io_prot(dst->mem.placement, PAGE_KERNEL);
+	d.src_prot = ttm_io_prot(src->mem.placement, PAGE_KERNEL);
+	d.diff = diff;
+
+	for (j = 0; j < h; ++j) {
+		diff->line = j + initial_line;
+		diff->line_offset = dst_offset % dst_stride;
+		ret = vmw_bo_cpu_blit_line(&d, dst_offset, src_offset, w);
+		if (ret)
+			goto out;
+
+		dst_offset += dst_stride;
+		src_offset += src_stride;
+	}
+out:
+	if (d.src_addr)
+		ttm_kunmap_atomic_prot(d.src_addr, d.src_prot);
+	if (d.dst_addr)
+		ttm_kunmap_atomic_prot(d.dst_addr, d.dst_prot);
+
+	return ret;
+}
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
index 7177eec..21111fd 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
@@ -185,6 +185,22 @@ static const struct ttm_place evictable_placement_flags[] = {
 	}
 };
 
+static const struct ttm_place nonfixed_placement_flags[] = {
+	{
+		.fpfn = 0,
+		.lpfn = 0,
+		.flags = TTM_PL_FLAG_SYSTEM | TTM_PL_FLAG_CACHED
+	}, {
+		.fpfn = 0,
+		.lpfn = 0,
+		.flags = VMW_PL_FLAG_GMR | TTM_PL_FLAG_CACHED
+	}, {
+		.fpfn = 0,
+		.lpfn = 0,
+		.flags = VMW_PL_FLAG_MOB | TTM_PL_FLAG_CACHED
+	}
+};
+
 struct ttm_placement vmw_evictable_placement = {
 	.num_placement = 4,
 	.placement = evictable_placement_flags,
@@ -213,6 +229,13 @@ struct ttm_placement vmw_mob_ne_placement = {
 	.busy_placement = &mob_ne_placement_flags
 };
 
+struct ttm_placement vmw_nonfixed_placement = {
+	.num_placement = 3,
+	.placement = nonfixed_placement_flags,
+	.num_busy_placement = 1,
+	.busy_placement = &sys_placement_flags
+};
+
 struct vmw_ttm_tt {
 	struct ttm_dma_tt dma_ttm;
 	struct vmw_private *dev_priv;
@@ -841,6 +864,7 @@ static void vmw_move_notify(struct ttm_buffer_object *bo,
  */
 static void vmw_swap_notify(struct ttm_buffer_object *bo)
 {
+	vmw_resource_swap_notify(bo);
 	(void) ttm_bo_wait(bo, false, false);
 }
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c
index f283324..9f45d50 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c
@@ -101,6 +101,7 @@ struct vmw_cmdbuf_context {
  * @handle: DMA address handle for the command buffer space if @using_mob is
  * false. Immutable.
  * @size: The size of the command buffer space. Immutable.
+ * @num_contexts: Number of contexts actually enabled.
  */
 struct vmw_cmdbuf_man {
 	struct mutex cur_mutex;
@@ -128,6 +129,7 @@ struct vmw_cmdbuf_man {
 	bool has_pool;
 	dma_addr_t handle;
 	size_t size;
+	u32 num_contexts;
 };
 
 /**
@@ -185,7 +187,7 @@ struct vmw_cmdbuf_alloc_info {
 
 /* Loop over each context in the command buffer manager. */
 #define for_each_cmdbuf_ctx(_man, _i, _ctx)				\
-	for (_i = 0, _ctx = &(_man)->ctx[0]; (_i) < SVGA_CB_CONTEXT_MAX; \
+	for (_i = 0, _ctx = &(_man)->ctx[0]; (_i) < (_man)->num_contexts; \
 	     ++(_i), ++(_ctx))
 
 static int vmw_cmdbuf_startstop(struct vmw_cmdbuf_man *man, u32 context,
@@ -514,6 +516,7 @@ static void vmw_cmdbuf_work_func(struct work_struct *work)
 	struct list_head restart_head[SVGA_CB_CONTEXT_MAX];
 	int i;
 	struct vmw_cmdbuf_context *ctx;
+	bool global_block = false;
 
 	for_each_cmdbuf_ctx(man, i, ctx) {
 		INIT_LIST_HEAD(&restart_head[i]);
@@ -531,6 +534,7 @@ static void vmw_cmdbuf_work_func(struct work_struct *work)
 
 		list_del_init(&entry->list);
 		restart[entry->cb_context] = true;
+		global_block = true;
 
 		if (!vmw_cmd_describe(header, &error_cmd_size, &cmd_name)) {
 			DRM_ERROR("Unknown command causing device error.\n");
@@ -564,23 +568,21 @@ static void vmw_cmdbuf_work_func(struct work_struct *work)
 		cb_hdr->length -= new_start_offset;
 		cb_hdr->errorOffset = 0;
 		cb_hdr->offset = 0;
+
 		list_add_tail(&entry->list, &restart_head[entry->cb_context]);
-		man->ctx[entry->cb_context].block_submission = true;
 	}
+
+	for_each_cmdbuf_ctx(man, i, ctx)
+		man->ctx[i].block_submission = true;
+
 	spin_unlock(&man->lock);
 
-	/* Preempt all contexts with errors */
-	for_each_cmdbuf_ctx(man, i, ctx) {
-		if (ctx->block_submission && vmw_cmdbuf_preempt(man, i))
-			DRM_ERROR("Failed preempting command buffer "
-				  "context %u.\n", i);
-	}
+	/* Preempt all contexts */
+	if (global_block && vmw_cmdbuf_preempt(man, 0))
+		DRM_ERROR("Failed preempting command buffer contexts\n");
 
 	spin_lock(&man->lock);
 	for_each_cmdbuf_ctx(man, i, ctx) {
-		if (!ctx->block_submission)
-			continue;
-
 		/* Move preempted command buffers to the preempted queue. */
 		vmw_cmdbuf_ctx_process(man, ctx, &dummy);
 
@@ -594,19 +596,16 @@ static void vmw_cmdbuf_work_func(struct work_struct *work)
 		 * Finally add all command buffers first in the submitted
 		 * queue, to rerun them.
 		 */
-		list_splice_init(&restart_head[i], &ctx->submitted);
 
 		ctx->block_submission = false;
+		list_splice_init(&restart_head[i], &ctx->submitted);
 	}
 
 	vmw_cmdbuf_man_process(man);
 	spin_unlock(&man->lock);
 
-	for_each_cmdbuf_ctx(man, i, ctx) {
-		if (restart[i] && vmw_cmdbuf_startstop(man, i, true))
-			DRM_ERROR("Failed restarting command buffer "
-				  "context %u.\n", i);
-	}
+	if (global_block && vmw_cmdbuf_startstop(man, 0, true))
+		DRM_ERROR("Failed restarting command buffer contexts\n");
 
 	/* Send a new fence in case one was removed */
 	if (send_fence) {
@@ -1307,6 +1306,8 @@ struct vmw_cmdbuf_man *vmw_cmdbuf_man_create(struct vmw_private *dev_priv)
 	if (!man)
 		return ERR_PTR(-ENOMEM);
 
+	man->num_contexts = (dev_priv->capabilities & SVGA_CAP_HP_CMD_QUEUE) ?
+		2 : 1;
 	man->headers = dma_pool_create("vmwgfx cmdbuf",
 				       &dev_priv->dev->pdev->dev,
 				       sizeof(SVGACBHeader),
@@ -1341,14 +1342,11 @@ struct vmw_cmdbuf_man *vmw_cmdbuf_man_create(struct vmw_private *dev_priv)
 	INIT_WORK(&man->work, &vmw_cmdbuf_work_func);
 	vmw_generic_waiter_add(dev_priv, SVGA_IRQFLAG_ERROR,
 			       &dev_priv->error_waiters);
-	for_each_cmdbuf_ctx(man, i, ctx) {
-		ret = vmw_cmdbuf_startstop(man, i, true);
-		if (ret) {
-			DRM_ERROR("Failed starting command buffer "
-				  "context %u.\n", i);
-			vmw_cmdbuf_man_destroy(man);
-			return ERR_PTR(ret);
-		}
+	ret = vmw_cmdbuf_startstop(man, 0, true);
+	if (ret) {
+		DRM_ERROR("Failed starting command buffer contexts\n");
+		vmw_cmdbuf_man_destroy(man);
+		return ERR_PTR(ret);
 	}
 
 	return man;
@@ -1398,16 +1396,11 @@ void vmw_cmdbuf_remove_pool(struct vmw_cmdbuf_man *man)
  */
 void vmw_cmdbuf_man_destroy(struct vmw_cmdbuf_man *man)
 {
-	struct vmw_cmdbuf_context *ctx;
-	unsigned int i;
-
 	WARN_ON_ONCE(man->has_pool);
 	(void) vmw_cmdbuf_idle(man, false, 10*HZ);
 
-	for_each_cmdbuf_ctx(man, i, ctx)
-		if (vmw_cmdbuf_startstop(man, i, false))
-			DRM_ERROR("Failed stopping command buffer "
-				  "context %u.\n", i);
+	if (vmw_cmdbuf_startstop(man, 0, false))
+		DRM_ERROR("Failed stopping command buffer contexts.\n");
 
 	vmw_generic_waiter_remove(man->dev_priv, SVGA_IRQFLAG_ERROR,
 				  &man->dev_priv->error_waiters);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_dmabuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_dmabuf.c
index d45d2ca..d59d9dd 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_dmabuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_dmabuf.c
@@ -323,3 +323,54 @@ void vmw_bo_pin_reserved(struct vmw_dma_buffer *vbo, bool pin)
 
 	BUG_ON(ret != 0 || bo->mem.mem_type != old_mem_type);
 }
+
+
+/*
+ * vmw_dma_buffer_unmap - Tear down a cached buffer object map.
+ *
+ * @vbo: The buffer object whose map we are tearing down.
+ *
+ * This function tears down a cached map set up using
+ * vmw_dma_buffer_map_and_cache().
+ */
+void vmw_dma_buffer_unmap(struct vmw_dma_buffer *vbo)
+{
+	if (vbo->map.bo == NULL)
+		return;
+
+	ttm_bo_kunmap(&vbo->map);
+}
+
+
+/*
+ * vmw_dma_buffer_map_and_cache - Map a buffer object and cache the map
+ *
+ * @vbo: The buffer object to map
+ * Return: A kernel virtual address or NULL if mapping failed.
+ *
+ * This function maps a buffer object into the kernel address space, or
+ * returns the virtual kernel address of an already existing map. The virtual
+ * address remains valid as long as the buffer object is pinned or reserved.
+ * The cached map is torn down on either
+ * 1) Buffer object move
+ * 2) Buffer object swapout
+ * 3) Buffer object destruction
+ *
+ */
+void *vmw_dma_buffer_map_and_cache(struct vmw_dma_buffer *vbo)
+{
+	struct ttm_buffer_object *bo = &vbo->base;
+	bool not_used;
+	void *virtual;
+	int ret;
+
+	virtual = ttm_kmap_obj_virtual(&vbo->map, &not_used);
+	if (virtual)
+		return virtual;
+
+	ret = ttm_bo_kmap(bo, 0, bo->num_pages, &vbo->map);
+	if (ret)
+		DRM_ERROR("Buffer object map failed: %d.\n", ret);
+
+	return ttm_kmap_obj_virtual(&vbo->map, &not_used);
+}
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
index 184340d..61a03ac 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
@@ -301,6 +301,8 @@ static void vmw_print_capabilities(uint32_t capabilities)
 		DRM_INFO("  Guest Backed Resources.\n");
 	if (capabilities & SVGA_CAP_DX)
 		DRM_INFO("  DX Features.\n");
+	if (capabilities & SVGA_CAP_HP_CMD_QUEUE)
+		DRM_INFO("  HP Command Queue.\n");
 }
 
 /**
@@ -1277,8 +1279,7 @@ static void vmw_master_drop(struct drm_device *dev,
 	ttm_lock_set_kill(&dev_priv->fbdev_master.lock, false, SIGTERM);
 	ttm_vt_unlock(&dev_priv->fbdev_master.lock);
 
-	if (dev_priv->enable_fb)
-		vmw_fb_on(dev_priv);
+	vmw_fb_refresh(dev_priv);
 }
 
 /**
@@ -1368,28 +1369,23 @@ static int vmwgfx_pm_notifier(struct notifier_block *nb, unsigned long val,
 
 	switch (val) {
 	case PM_HIBERNATION_PREPARE:
-		if (dev_priv->enable_fb)
-			vmw_fb_off(dev_priv);
-		ttm_suspend_lock(&dev_priv->reservation_sem);
-
 		/*
-		 * This empties VRAM and unbinds all GMR bindings.
-		 * Buffer contents is moved to swappable memory.
+		 * Take the reservation sem in write mode, which will make sure
+		 * there are no other processes holding a buffer object
+		 * reservation, meaning we should be able to evict all buffer
+		 * objects if needed.
+		 * Once user-space processes have been frozen, we can release
+		 * the lock again.
 		 */
-		vmw_execbuf_release_pinned_bo(dev_priv);
-		vmw_resource_evict_all(dev_priv);
-		vmw_release_device_early(dev_priv);
-		ttm_bo_swapout_all(&dev_priv->bdev);
-		vmw_fence_fifo_down(dev_priv->fman);
+		ttm_suspend_lock(&dev_priv->reservation_sem);
+		dev_priv->suspend_locked = true;
 		break;
 	case PM_POST_HIBERNATION:
 	case PM_POST_RESTORE:
-		vmw_fence_fifo_up(dev_priv->fman);
-		ttm_suspend_unlock(&dev_priv->reservation_sem);
-		if (dev_priv->enable_fb)
-			vmw_fb_on(dev_priv);
-		break;
-	case PM_RESTORE_PREPARE:
+		if (READ_ONCE(dev_priv->suspend_locked)) {
+			dev_priv->suspend_locked = false;
+			ttm_suspend_unlock(&dev_priv->reservation_sem);
+		}
 		break;
 	default:
 		break;
@@ -1440,25 +1436,48 @@ static int vmw_pm_freeze(struct device *kdev)
 	struct pci_dev *pdev = to_pci_dev(kdev);
 	struct drm_device *dev = pci_get_drvdata(pdev);
 	struct vmw_private *dev_priv = vmw_priv(dev);
+	int ret;
 
-	dev_priv->suspended = true;
+	/*
+	 * Unlock for vmw_kms_suspend.
+	 * No user-space processes should be running now.
+	 */
+	ttm_suspend_unlock(&dev_priv->reservation_sem);
+	ret = vmw_kms_suspend(dev_priv->dev);
+	if (ret) {
+		ttm_suspend_lock(&dev_priv->reservation_sem);
+		DRM_ERROR("Failed to freeze modesetting.\n");
+		return ret;
+	}
 	if (dev_priv->enable_fb)
-		vmw_fifo_resource_dec(dev_priv);
+		vmw_fb_off(dev_priv);
 
+	ttm_suspend_lock(&dev_priv->reservation_sem);
+	vmw_execbuf_release_pinned_bo(dev_priv);
+	vmw_resource_evict_all(dev_priv);
+	vmw_release_device_early(dev_priv);
+	ttm_bo_swapout_all(&dev_priv->bdev);
+	if (dev_priv->enable_fb)
+		vmw_fifo_resource_dec(dev_priv);
 	if (atomic_read(&dev_priv->num_fifo_resources) != 0) {
 		DRM_ERROR("Can't hibernate while 3D resources are active.\n");
 		if (dev_priv->enable_fb)
 			vmw_fifo_resource_inc(dev_priv);
 		WARN_ON(vmw_request_device_late(dev_priv));
-		dev_priv->suspended = false;
+		dev_priv->suspend_locked = false;
+		ttm_suspend_unlock(&dev_priv->reservation_sem);
+		if (dev_priv->suspend_state)
+			vmw_kms_resume(dev);
+		if (dev_priv->enable_fb)
+			vmw_fb_on(dev_priv);
+		vmw_fb_refresh(dev_priv);
 		return -EBUSY;
 	}
 
-	if (dev_priv->enable_fb)
-		__vmw_svga_disable(dev_priv);
+	vmw_fence_fifo_down(dev_priv->fman);
+	__vmw_svga_disable(dev_priv);
 	
 	vmw_release_device_late(dev_priv);
-
 	return 0;
 }
 
@@ -1482,7 +1501,16 @@ static int vmw_pm_restore(struct device *kdev)
 	if (dev_priv->enable_fb)
 		__vmw_svga_enable(dev_priv);
 
-	dev_priv->suspended = false;
+	vmw_fence_fifo_up(dev_priv->fman);
+	dev_priv->suspend_locked = false;
+	ttm_suspend_unlock(&dev_priv->reservation_sem);
+	if (dev_priv->suspend_state)
+		vmw_kms_resume(dev_priv->dev);
+
+	if (dev_priv->enable_fb)
+		vmw_fb_on(dev_priv);
+
+	vmw_fb_refresh(dev_priv);
 
 	return 0;
 }
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index d08753e..9e60de9 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -43,10 +43,10 @@
 #include <linux/sync_file.h>
 
 #define VMWGFX_DRIVER_NAME "vmwgfx"
-#define VMWGFX_DRIVER_DATE "20170612"
+#define VMWGFX_DRIVER_DATE "20180322"
 #define VMWGFX_DRIVER_MAJOR 2
 #define VMWGFX_DRIVER_MINOR 14
-#define VMWGFX_DRIVER_PATCHLEVEL 0
+#define VMWGFX_DRIVER_PATCHLEVEL 1
 #define VMWGFX_FILE_PAGE_OFFSET 0x00100000
 #define VMWGFX_FIFO_STATIC_SIZE (1024*1024)
 #define VMWGFX_MAX_RELOCATIONS 2048
@@ -92,6 +92,8 @@ struct vmw_dma_buffer {
 	s32 pin_count;
 	/* Not ref-counted.  Protected by binding_mutex */
 	struct vmw_resource *dx_query_ctx;
+	/* Protected by reservation */
+	struct ttm_bo_kmap_obj map;
 };
 
 /**
@@ -423,6 +425,7 @@ struct vmw_private {
 	struct vmw_framebuffer *implicit_fb;
 	struct mutex global_kms_state_mutex;
 	spinlock_t cursor_lock;
+	struct drm_atomic_state *suspend_state;
 
 	/*
 	 * Context and surface management.
@@ -494,8 +497,8 @@ struct vmw_private {
 	struct vmw_master *active_master;
 	struct vmw_master fbdev_master;
 	struct notifier_block pm_nb;
-	bool suspended;
 	bool refuse_hibernation;
+	bool suspend_locked;
 
 	struct mutex release_mutex;
 	atomic_t num_fifo_resources;
@@ -673,11 +676,13 @@ extern void vmw_resource_move_notify(struct ttm_buffer_object *bo,
 				     struct ttm_mem_reg *mem);
 extern void vmw_query_move_notify(struct ttm_buffer_object *bo,
 				  struct ttm_mem_reg *mem);
+extern void vmw_resource_swap_notify(struct ttm_buffer_object *bo);
 extern int vmw_query_readback_all(struct vmw_dma_buffer *dx_query_mob);
 extern void vmw_fence_single_bo(struct ttm_buffer_object *bo,
 				struct vmw_fence_obj *fence);
 extern void vmw_resource_evict_all(struct vmw_private *dev_priv);
 
+
 /**
  * DMA buffer helper routines - vmwgfx_dmabuf.c
  */
@@ -700,6 +705,8 @@ extern int vmw_dmabuf_unpin(struct vmw_private *vmw_priv,
 extern void vmw_bo_get_guest_ptr(const struct ttm_buffer_object *buf,
 				 SVGAGuestPtr *ptr);
 extern void vmw_bo_pin_reserved(struct vmw_dma_buffer *bo, bool pin);
+extern void *vmw_dma_buffer_map_and_cache(struct vmw_dma_buffer *vbo);
+extern void vmw_dma_buffer_unmap(struct vmw_dma_buffer *vbo);
 
 /**
  * Misc Ioctl functionality - vmwgfx_ioctl.c
@@ -766,6 +773,7 @@ extern struct ttm_placement vmw_evictable_placement;
 extern struct ttm_placement vmw_srf_placement;
 extern struct ttm_placement vmw_mob_placement;
 extern struct ttm_placement vmw_mob_ne_placement;
+extern struct ttm_placement vmw_nonfixed_placement;
 extern struct ttm_bo_driver vmw_bo_driver;
 extern int vmw_dma_quiescent(struct drm_device *dev);
 extern int vmw_bo_map_dma(struct ttm_buffer_object *bo);
@@ -902,6 +910,7 @@ int vmw_fb_init(struct vmw_private *vmw_priv);
 int vmw_fb_close(struct vmw_private *dev_priv);
 int vmw_fb_off(struct vmw_private *vmw_priv);
 int vmw_fb_on(struct vmw_private *vmw_priv);
+void vmw_fb_refresh(struct vmw_private *vmw_priv);
 
 /**
  * Kernel modesetting - vmwgfx_kms.c
@@ -938,6 +947,8 @@ int vmw_kms_present(struct vmw_private *dev_priv,
 int vmw_kms_update_layout_ioctl(struct drm_device *dev, void *data,
 				struct drm_file *file_priv);
 void vmw_kms_legacy_hotspot_clear(struct vmw_private *dev_priv);
+int vmw_kms_suspend(struct drm_device *dev);
+int vmw_kms_resume(struct drm_device *dev);
 
 int vmw_dumb_create(struct drm_file *file_priv,
 		    struct drm_device *dev,
@@ -1165,6 +1176,53 @@ extern int vmw_cmdbuf_cur_flush(struct vmw_cmdbuf_man *man,
 				bool interruptible);
 extern void vmw_cmdbuf_irqthread(struct vmw_cmdbuf_man *man);
 
+/* CPU blit utilities - vmwgfx_blit.c */
+
+/**
+ * struct vmw_diff_cpy - CPU blit information structure
+ *
+ * @rect: The output bounding box rectangle.
+ * @line: The current line of the blit.
+ * @line_offset: Offset of the current line segment.
+ * @cpp: Bytes per pixel (granularity information).
+ * @memcpy: Which memcpy function to use.
+ */
+struct vmw_diff_cpy {
+	struct drm_rect rect;
+	size_t line;
+	size_t line_offset;
+	int cpp;
+	void (*do_cpy)(struct vmw_diff_cpy *diff, u8 *dest, const u8 *src,
+		       size_t n);
+};
+
+#define VMW_CPU_BLIT_INITIALIZER {	\
+	.do_cpy = vmw_memcpy,		\
+}
+
+#define VMW_CPU_BLIT_DIFF_INITIALIZER(_cpp) {	  \
+	.line = 0,				  \
+	.line_offset = 0,			  \
+	.rect = { .x1 = INT_MAX/2,		  \
+		  .y1 = INT_MAX/2,		  \
+		  .x2 = INT_MIN/2,		  \
+		  .y2 = INT_MIN/2		  \
+	},					  \
+	.cpp = _cpp,				  \
+	.do_cpy = vmw_diff_memcpy,		  \
+}
+
+void vmw_diff_memcpy(struct vmw_diff_cpy *diff, u8 *dest, const u8 *src,
+		     size_t n);
+
+void vmw_memcpy(struct vmw_diff_cpy *diff, u8 *dest, const u8 *src, size_t n);
+
+int vmw_bo_cpu_blit(struct ttm_buffer_object *dst,
+		    u32 dst_offset, u32 dst_stride,
+		    struct ttm_buffer_object *src,
+		    u32 src_offset, u32 src_stride,
+		    u32 w, u32 h,
+		    struct vmw_diff_cpy *diff);
 
 /**
  * Inline helper functions
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
index be7d7fb..2582ffd 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
@@ -43,8 +43,6 @@ struct vmw_fb_par {
 
 	struct mutex bo_mutex;
 	struct vmw_dma_buffer *vmw_bo;
-	struct ttm_bo_kmap_obj map;
-	void *bo_ptr;
 	unsigned bo_size;
 	struct drm_framebuffer *set_fb;
 	struct drm_display_mode *set_mode;
@@ -163,10 +161,17 @@ static int vmw_fb_blank(int blank, struct fb_info *info)
 	return 0;
 }
 
-/*
- * Dirty code
+/**
+ * vmw_fb_dirty_flush - flush dirty regions to the kms framebuffer
+ *
+ * @work: The struct work_struct associated with this task.
+ *
+ * This function flushes the dirty regions of the vmalloc framebuffer to the
+ * kms framebuffer, and if the kms framebuffer is visible, also updated the
+ * corresponding displays. Note that this function runs even if the kms
+ * framebuffer is not bound to a crtc and thus not visible, but it's turned
+ * off during hibernation using the par->dirty.active bool.
  */
-
 static void vmw_fb_dirty_flush(struct work_struct *work)
 {
 	struct vmw_fb_par *par = container_of(work, struct vmw_fb_par,
@@ -174,13 +179,15 @@ static void vmw_fb_dirty_flush(struct work_struct *work)
 	struct vmw_private *vmw_priv = par->vmw_priv;
 	struct fb_info *info = vmw_priv->fb_info;
 	unsigned long irq_flags;
-	s32 dst_x1, dst_x2, dst_y1, dst_y2, w, h;
+	s32 dst_x1, dst_x2, dst_y1, dst_y2, w = 0, h = 0;
 	u32 cpp, max_x, max_y;
 	struct drm_clip_rect clip;
 	struct drm_framebuffer *cur_fb;
 	u8 *src_ptr, *dst_ptr;
+	struct vmw_dma_buffer *vbo = par->vmw_bo;
+	void *virtual;
 
-	if (vmw_priv->suspended)
+	if (!READ_ONCE(par->dirty.active))
 		return;
 
 	mutex_lock(&par->bo_mutex);
@@ -188,10 +195,16 @@ static void vmw_fb_dirty_flush(struct work_struct *work)
 	if (!cur_fb)
 		goto out_unlock;
 
+	(void) ttm_read_lock(&vmw_priv->reservation_sem, false);
+	(void) ttm_bo_reserve(&vbo->base, false, false, NULL);
+	virtual = vmw_dma_buffer_map_and_cache(vbo);
+	if (!virtual)
+		goto out_unreserve;
+
 	spin_lock_irqsave(&par->dirty.lock, irq_flags);
 	if (!par->dirty.active) {
 		spin_unlock_irqrestore(&par->dirty.lock, irq_flags);
-		goto out_unlock;
+		goto out_unreserve;
 	}
 
 	/*
@@ -221,7 +234,7 @@ static void vmw_fb_dirty_flush(struct work_struct *work)
 	spin_unlock_irqrestore(&par->dirty.lock, irq_flags);
 
 	if (w && h) {
-		dst_ptr = (u8 *)par->bo_ptr  +
+		dst_ptr = (u8 *)virtual  +
 			(dst_y1 * par->set_fb->pitches[0] + dst_x1 * cpp);
 		src_ptr = (u8 *)par->vmalloc +
 			((dst_y1 + par->fb_y) * info->fix.line_length +
@@ -237,7 +250,12 @@ static void vmw_fb_dirty_flush(struct work_struct *work)
 		clip.x2 = dst_x2;
 		clip.y1 = dst_y1;
 		clip.y2 = dst_y2;
+	}
 
+out_unreserve:
+	ttm_bo_unreserve(&vbo->base);
+	ttm_read_unlock(&vmw_priv->reservation_sem);
+	if (w && h) {
 		WARN_ON_ONCE(par->set_fb->funcs->dirty(cur_fb, NULL, 0, 0,
 						       &clip, 1));
 		vmw_fifo_flush(vmw_priv, false);
@@ -504,18 +522,8 @@ static int vmw_fb_kms_detach(struct vmw_fb_par *par,
 		par->set_fb = NULL;
 	}
 
-	if (par->vmw_bo && detach_bo) {
-		struct vmw_private *vmw_priv = par->vmw_priv;
-
-		if (par->bo_ptr) {
-			ttm_bo_kunmap(&par->map);
-			par->bo_ptr = NULL;
-		}
-		if (unref_bo)
-			vmw_dmabuf_unreference(&par->vmw_bo);
-		else if (vmw_priv->active_display_unit != vmw_du_legacy)
-			vmw_dmabuf_unpin(par->vmw_priv, par->vmw_bo, false);
-	}
+	if (par->vmw_bo && detach_bo && unref_bo)
+		vmw_dmabuf_unreference(&par->vmw_bo);
 
 	return 0;
 }
@@ -636,38 +644,6 @@ static int vmw_fb_set_par(struct fb_info *info)
 	if (ret)
 		goto out_unlock;
 
-	if (!par->bo_ptr) {
-		struct vmw_framebuffer *vfb = vmw_framebuffer_to_vfb(set.fb);
-
-		/*
-		 * Pin before mapping. Since we don't know in what placement
-		 * to pin, call into KMS to do it for us.  LDU doesn't require
-		 * additional pinning because set_config() would've pinned
-		 * it already
-		 */
-		if (vmw_priv->active_display_unit != vmw_du_legacy) {
-			ret = vfb->pin(vfb);
-			if (ret) {
-				DRM_ERROR("Could not pin the fbdev "
-					  "framebuffer.\n");
-				goto out_unlock;
-			}
-		}
-
-		ret = ttm_bo_kmap(&par->vmw_bo->base, 0,
-				  par->vmw_bo->base.num_pages, &par->map);
-		if (ret) {
-			if (vmw_priv->active_display_unit != vmw_du_legacy)
-				vfb->unpin(vfb);
-
-			DRM_ERROR("Could not map the fbdev framebuffer.\n");
-			goto out_unlock;
-		}
-
-		par->bo_ptr = ttm_kmap_obj_virtual(&par->map, &par->bo_iowrite);
-	}
-
-
 	vmw_fb_dirty_mark(par, par->fb_x, par->fb_y,
 			  par->set_fb->width, par->set_fb->height);
 
@@ -883,12 +859,6 @@ int vmw_fb_off(struct vmw_private *vmw_priv)
 	flush_delayed_work(&info->deferred_work);
 	flush_delayed_work(&par->local_work);
 
-	mutex_lock(&par->bo_mutex);
-	drm_modeset_lock_all(vmw_priv->dev);
-	(void) vmw_fb_kms_detach(par, true, false);
-	drm_modeset_unlock_all(vmw_priv->dev);
-	mutex_unlock(&par->bo_mutex);
-
 	return 0;
 }
 
@@ -904,10 +874,24 @@ int vmw_fb_on(struct vmw_private *vmw_priv)
 	info = vmw_priv->fb_info;
 	par = info->par;
 
-	vmw_fb_set_par(info);
 	spin_lock_irqsave(&par->dirty.lock, flags);
 	par->dirty.active = true;
 	spin_unlock_irqrestore(&par->dirty.lock, flags);
  
 	return 0;
 }
+
+/**
+ * vmw_fb_refresh - Refresh fb display
+ *
+ * @vmw_priv: Pointer to device private
+ *
+ * Call into kms to show the fbdev display(s).
+ */
+void vmw_fb_refresh(struct vmw_private *vmw_priv)
+{
+	if (!vmw_priv->fb_info)
+		return;
+
+	vmw_fb_set_par(vmw_priv->fb_info);
+}
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
index 6c5c75c..9ed544f8 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
@@ -901,11 +901,12 @@ static void vmw_event_fence_action_seq_passed(struct vmw_fence_action *action)
 	spin_lock_irq(&dev->event_lock);
 
 	if (likely(eaction->tv_sec != NULL)) {
-		struct timeval tv;
+		struct timespec64 ts;
 
-		do_gettimeofday(&tv);
-		*eaction->tv_sec = tv.tv_sec;
-		*eaction->tv_usec = tv.tv_usec;
+		ktime_get_ts64(&ts);
+		/* monotonic time, so no y2038 overflow */
+		*eaction->tv_sec = ts.tv_sec;
+		*eaction->tv_usec = ts.tv_nsec / NSEC_PER_USEC;
 	}
 
 	drm_send_event_locked(dev, eaction->event);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
index 34ecc27..3628a9f 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
@@ -393,13 +393,13 @@ vmw_du_cursor_plane_atomic_update(struct drm_plane *plane,
 	du->cursor_surface = vps->surf;
 	du->cursor_dmabuf = vps->dmabuf;
 
-	/* setup new image */
 	if (vps->surf) {
 		du->cursor_age = du->cursor_surface->snooper.age;
 
 		ret = vmw_cursor_update_image(dev_priv,
 					      vps->surf->snooper.image,
-					      64, 64, hotspot_x, hotspot_y);
+					      64, 64, hotspot_x,
+					      hotspot_y);
 	} else if (vps->dmabuf) {
 		ret = vmw_cursor_update_dmabuf(dev_priv, vps->dmabuf,
 					       plane->state->crtc_w,
@@ -497,11 +497,22 @@ int vmw_du_cursor_plane_atomic_check(struct drm_plane *plane,
 	struct vmw_surface *surface = NULL;
 	struct drm_framebuffer *fb = new_state->fb;
 
+	struct drm_rect src = drm_plane_state_src(new_state);
+	struct drm_rect dest = drm_plane_state_dest(new_state);
 
 	/* Turning off */
 	if (!fb)
 		return ret;
 
+	ret = drm_plane_helper_check_update(plane, new_state->crtc, fb,
+					    &src, &dest,
+					    DRM_MODE_ROTATE_0,
+					    DRM_PLANE_HELPER_NO_SCALING,
+					    DRM_PLANE_HELPER_NO_SCALING,
+					    true, true, &new_state->visible);
+	if (!ret)
+		return ret;
+
 	/* A lot of the code assumes this */
 	if (new_state->crtc_w != 64 || new_state->crtc_h != 64) {
 		DRM_ERROR("Invalid cursor dimensions (%d, %d)\n",
@@ -566,13 +577,9 @@ void vmw_du_crtc_atomic_flush(struct drm_crtc *crtc,
 		crtc->state->event = NULL;
 
 		spin_lock_irq(&crtc->dev->event_lock);
-		if (drm_crtc_vblank_get(crtc) == 0)
-			drm_crtc_arm_vblank_event(crtc, event);
-		else
-			drm_crtc_send_vblank_event(crtc, event);
+		drm_crtc_send_vblank_event(crtc, event);
 		spin_unlock_irq(&crtc->dev->event_lock);
 	}
-
 }
 
 
@@ -675,9 +682,6 @@ vmw_du_plane_duplicate_state(struct drm_plane *plane)
 		return NULL;
 
 	vps->pinned = 0;
-
-	/* Mapping is managed by prepare_fb/cleanup_fb */
-	memset(&vps->host_map, 0, sizeof(vps->host_map));
 	vps->cpp = 0;
 
 	/* Each ref counted resource needs to be acquired again */
@@ -739,11 +743,6 @@ vmw_du_plane_destroy_state(struct drm_plane *plane,
 
 
 	/* Should have been freed by cleanup_fb */
-	if (vps->host_map.virtual) {
-		DRM_ERROR("Host mapping not freed\n");
-		ttm_bo_kunmap(&vps->host_map);
-	}
-
 	if (vps->surf)
 		vmw_surface_unreference(&vps->surf);
 
@@ -888,11 +887,11 @@ static int vmw_framebuffer_surface_dirty(struct drm_framebuffer *framebuffer,
 	if (dev_priv->active_display_unit == vmw_du_screen_object)
 		ret = vmw_kms_sou_do_surface_dirty(dev_priv, &vfbs->base,
 						   clips, NULL, NULL, 0, 0,
-						   num_clips, inc, NULL);
+						   num_clips, inc, NULL, NULL);
 	else
 		ret = vmw_kms_stdu_surface_dirty(dev_priv, &vfbs->base,
 						 clips, NULL, NULL, 0, 0,
-						 num_clips, inc, NULL);
+						 num_clips, inc, NULL, NULL);
 
 	vmw_fifo_flush(dev_priv, false);
 	ttm_read_unlock(&dev_priv->reservation_sem);
@@ -928,11 +927,12 @@ int vmw_kms_readback(struct vmw_private *dev_priv,
 	switch (dev_priv->active_display_unit) {
 	case vmw_du_screen_object:
 		return vmw_kms_sou_readback(dev_priv, file_priv, vfb,
-					    user_fence_rep, vclips, num_clips);
+					    user_fence_rep, vclips, num_clips,
+					    NULL);
 	case vmw_du_screen_target:
 		return vmw_kms_stdu_dma(dev_priv, file_priv, vfb,
 					user_fence_rep, NULL, vclips, num_clips,
-					1, false, true);
+					1, false, true, NULL);
 	default:
 		WARN_ONCE(true,
 			  "Readback called with invalid display system.\n");
@@ -1090,12 +1090,12 @@ static int vmw_framebuffer_dmabuf_dirty(struct drm_framebuffer *framebuffer,
 	case vmw_du_screen_target:
 		ret = vmw_kms_stdu_dma(dev_priv, NULL, &vfbd->base, NULL,
 				       clips, NULL, num_clips, increment,
-				       true, true);
+				       true, true, NULL);
 		break;
 	case vmw_du_screen_object:
 		ret = vmw_kms_sou_do_dmabuf_dirty(dev_priv, &vfbd->base,
 						  clips, NULL, num_clips,
-						  increment, true, NULL);
+						  increment, true, NULL, NULL);
 		break;
 	case vmw_du_legacy:
 		ret = vmw_kms_ldu_do_dmabuf_dirty(dev_priv, &vfbd->base, 0, 0,
@@ -1121,12 +1121,14 @@ static const struct drm_framebuffer_funcs vmw_framebuffer_dmabuf_funcs = {
 };
 
 /**
- * Pin the dmabuffer to the start of vram.
+ * Pin the dmabuffer in a location suitable for access by the
+ * display system.
  */
 static int vmw_framebuffer_pin(struct vmw_framebuffer *vfb)
 {
 	struct vmw_private *dev_priv = vmw_priv(vfb->base.dev);
 	struct vmw_dma_buffer *buf;
+	struct ttm_placement *placement;
 	int ret;
 
 	buf = vfb->dmabuf ?  vmw_framebuffer_to_vfbd(&vfb->base)->buffer :
@@ -1143,12 +1145,24 @@ static int vmw_framebuffer_pin(struct vmw_framebuffer *vfb)
 		break;
 	case vmw_du_screen_object:
 	case vmw_du_screen_target:
-		if (vfb->dmabuf)
-			return vmw_dmabuf_pin_in_vram_or_gmr(dev_priv, buf,
-							     false);
+		if (vfb->dmabuf) {
+			if (dev_priv->capabilities & SVGA_CAP_3D) {
+				/*
+				 * Use surface DMA to get content to
+				 * sreen target surface.
+				 */
+				placement = &vmw_vram_gmr_placement;
+			} else {
+				/* Use CPU blit. */
+				placement = &vmw_sys_placement;
+			}
+		} else {
+			/* Use surface / image update */
+			placement = &vmw_mob_placement;
+		}
 
-		return vmw_dmabuf_pin_in_placement(dev_priv, buf,
-						   &vmw_mob_placement, false);
+		return vmw_dmabuf_pin_in_placement(dev_priv, buf, placement,
+						   false);
 	default:
 		return -EINVAL;
 	}
@@ -1539,35 +1553,10 @@ vmw_kms_atomic_check_modeset(struct drm_device *dev,
 	return drm_atomic_helper_check(dev, state);
 }
 
-
-/**
- * vmw_kms_atomic_commit - Perform an atomic state commit
- *
- * @dev: DRM device
- * @state: the driver state object
- * @nonblock: Whether nonblocking behaviour is requested
- *
- * This is a simple wrapper around drm_atomic_helper_commit() for
- * us to clear the nonblocking value.
- *
- * Nonblocking commits currently cause synchronization issues
- * for vmwgfx.
- *
- * RETURNS
- * Zero for success or negative error code on failure.
- */
-int vmw_kms_atomic_commit(struct drm_device *dev,
-			  struct drm_atomic_state *state,
-			  bool nonblock)
-{
-	return drm_atomic_helper_commit(dev, state, false);
-}
-
-
 static const struct drm_mode_config_funcs vmw_kms_funcs = {
 	.fb_create = vmw_kms_fb_create,
 	.atomic_check = vmw_kms_atomic_check_modeset,
-	.atomic_commit = vmw_kms_atomic_commit,
+	.atomic_commit = drm_atomic_helper_commit,
 };
 
 static int vmw_kms_generic_present(struct vmw_private *dev_priv,
@@ -1581,7 +1570,7 @@ static int vmw_kms_generic_present(struct vmw_private *dev_priv,
 {
 	return vmw_kms_sou_do_surface_dirty(dev_priv, vfb, NULL, clips,
 					    &surface->res, destX, destY,
-					    num_clips, 1, NULL);
+					    num_clips, 1, NULL, NULL);
 }
 
 
@@ -1600,7 +1589,7 @@ int vmw_kms_present(struct vmw_private *dev_priv,
 	case vmw_du_screen_target:
 		ret = vmw_kms_stdu_surface_dirty(dev_priv, vfb, NULL, clips,
 						 &surface->res, destX, destY,
-						 num_clips, 1, NULL);
+						 num_clips, 1, NULL, NULL);
 		break;
 	case vmw_du_screen_object:
 		ret = vmw_kms_generic_present(dev_priv, file_priv, vfb, surface,
@@ -2328,10 +2317,16 @@ int vmw_kms_helper_dirty(struct vmw_private *dev_priv,
 
 	dirty->dev_priv = dev_priv;
 
-	list_for_each_entry(crtc, &dev_priv->dev->mode_config.crtc_list, head) {
-		if (crtc->primary->fb != &framebuffer->base)
-			continue;
-		units[num_units++] = vmw_crtc_to_du(crtc);
+	/* If crtc is passed, no need to iterate over other display units */
+	if (dirty->crtc) {
+		units[num_units++] = vmw_crtc_to_du(dirty->crtc);
+	} else {
+		list_for_each_entry(crtc, &dev_priv->dev->mode_config.crtc_list,
+				    head) {
+			if (crtc->primary->fb != &framebuffer->base)
+				continue;
+			units[num_units++] = vmw_crtc_to_du(crtc);
+		}
 	}
 
 	for (k = 0; k < num_units; k++) {
@@ -2430,14 +2425,21 @@ int vmw_kms_helper_dirty(struct vmw_private *dev_priv,
 int vmw_kms_helper_buffer_prepare(struct vmw_private *dev_priv,
 				  struct vmw_dma_buffer *buf,
 				  bool interruptible,
-				  bool validate_as_mob)
+				  bool validate_as_mob,
+				  bool for_cpu_blit)
 {
+	struct ttm_operation_ctx ctx = {
+		.interruptible = interruptible,
+		.no_wait_gpu = false};
 	struct ttm_buffer_object *bo = &buf->base;
 	int ret;
 
 	ttm_bo_reserve(bo, false, false, NULL);
-	ret = vmw_validate_single_buffer(dev_priv, bo, interruptible,
-					 validate_as_mob);
+	if (for_cpu_blit)
+		ret = ttm_bo_validate(bo, &vmw_nonfixed_placement, &ctx);
+	else
+		ret = vmw_validate_single_buffer(dev_priv, bo, interruptible,
+						 validate_as_mob);
 	if (ret)
 		ttm_bo_unreserve(bo);
 
@@ -2549,7 +2551,8 @@ int vmw_kms_helper_resource_prepare(struct vmw_resource *res,
 	if (res->backup) {
 		ret = vmw_kms_helper_buffer_prepare(res->dev_priv, res->backup,
 						    interruptible,
-						    res->dev_priv->has_mob);
+						    res->dev_priv->has_mob,
+						    false);
 		if (ret)
 			goto out_unreserve;
 	}
@@ -2845,3 +2848,51 @@ int vmw_kms_set_config(struct drm_mode_set *set,
 
 	return drm_atomic_helper_set_config(set, ctx);
 }
+
+
+/**
+ * vmw_kms_suspend - Save modesetting state and turn modesetting off.
+ *
+ * @dev: Pointer to the drm device
+ * Return: 0 on success. Negative error code on failure.
+ */
+int vmw_kms_suspend(struct drm_device *dev)
+{
+	struct vmw_private *dev_priv = vmw_priv(dev);
+
+	dev_priv->suspend_state = drm_atomic_helper_suspend(dev);
+	if (IS_ERR(dev_priv->suspend_state)) {
+		int ret = PTR_ERR(dev_priv->suspend_state);
+
+		DRM_ERROR("Failed kms suspend: %d\n", ret);
+		dev_priv->suspend_state = NULL;
+
+		return ret;
+	}
+
+	return 0;
+}
+
+
+/**
+ * vmw_kms_resume - Re-enable modesetting and restore state
+ *
+ * @dev: Pointer to the drm device
+ * Return: 0 on success. Negative error code on failure.
+ *
+ * State is resumed from a previous vmw_kms_suspend(). It's illegal
+ * to call this function without a previous vmw_kms_suspend().
+ */
+int vmw_kms_resume(struct drm_device *dev)
+{
+	struct vmw_private *dev_priv = vmw_priv(dev);
+	int ret;
+
+	if (WARN_ON(!dev_priv->suspend_state))
+		return 0;
+
+	ret = drm_atomic_helper_resume(dev, dev_priv->suspend_state);
+	dev_priv->suspend_state = NULL;
+
+	return ret;
+}
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
index cd9da2d..4e8749a 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
@@ -50,6 +50,7 @@
  * @unit: The current display unit. Set up by the helper before a call to @clip.
  * @cmd: The allocated fifo space. Set up by the helper before the first @clip
  * call.
+ * @crtc: The crtc for which to build dirty commands.
  * @num_hits: Number of clip rect commands for this display unit.
  * Cleared by the helper before the first @clip call. Updated by the @clip
  * callback.
@@ -71,6 +72,7 @@ struct vmw_kms_dirty {
 	struct vmw_private *dev_priv;
 	struct vmw_display_unit *unit;
 	void *cmd;
+	struct drm_crtc *crtc;
 	u32 num_hits;
 	s32 fb_x;
 	s32 fb_y;
@@ -175,7 +177,6 @@ struct vmw_plane_state {
 	int pinned;
 
 	/* For CPU Blit */
-	struct ttm_bo_kmap_obj host_map;
 	unsigned int cpp;
 };
 
@@ -287,7 +288,8 @@ int vmw_kms_helper_dirty(struct vmw_private *dev_priv,
 int vmw_kms_helper_buffer_prepare(struct vmw_private *dev_priv,
 				  struct vmw_dma_buffer *buf,
 				  bool interruptible,
-				  bool validate_as_mob);
+				  bool validate_as_mob,
+				  bool for_cpu_blit);
 void vmw_kms_helper_buffer_revert(struct vmw_dma_buffer *buf);
 void vmw_kms_helper_buffer_finish(struct vmw_private *dev_priv,
 				  struct drm_file *file_priv,
@@ -398,20 +400,23 @@ int vmw_kms_sou_do_surface_dirty(struct vmw_private *dev_priv,
 				 s32 dest_x,
 				 s32 dest_y,
 				 unsigned num_clips, int inc,
-				 struct vmw_fence_obj **out_fence);
+				 struct vmw_fence_obj **out_fence,
+				 struct drm_crtc *crtc);
 int vmw_kms_sou_do_dmabuf_dirty(struct vmw_private *dev_priv,
 				struct vmw_framebuffer *framebuffer,
 				struct drm_clip_rect *clips,
 				struct drm_vmw_rect *vclips,
 				unsigned num_clips, int increment,
 				bool interruptible,
-				struct vmw_fence_obj **out_fence);
+				struct vmw_fence_obj **out_fence,
+				struct drm_crtc *crtc);
 int vmw_kms_sou_readback(struct vmw_private *dev_priv,
 			 struct drm_file *file_priv,
 			 struct vmw_framebuffer *vfb,
 			 struct drm_vmw_fence_rep __user *user_fence_rep,
 			 struct drm_vmw_rect *vclips,
-			 uint32_t num_clips);
+			 uint32_t num_clips,
+			 struct drm_crtc *crtc);
 
 /*
  * Screen Target Display Unit functions - vmwgfx_stdu.c
@@ -425,7 +430,8 @@ int vmw_kms_stdu_surface_dirty(struct vmw_private *dev_priv,
 			       s32 dest_x,
 			       s32 dest_y,
 			       unsigned num_clips, int inc,
-			       struct vmw_fence_obj **out_fence);
+			       struct vmw_fence_obj **out_fence,
+			       struct drm_crtc *crtc);
 int vmw_kms_stdu_dma(struct vmw_private *dev_priv,
 		     struct drm_file *file_priv,
 		     struct vmw_framebuffer *vfb,
@@ -435,7 +441,8 @@ int vmw_kms_stdu_dma(struct vmw_private *dev_priv,
 		     uint32_t num_clips,
 		     int increment,
 		     bool to_surface,
-		     bool interruptible);
+		     bool interruptible,
+		     struct drm_crtc *crtc);
 
 int vmw_kms_set_config(struct drm_mode_set *set,
 		       struct drm_modeset_acquire_ctx *ctx);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_msg.c b/drivers/gpu/drm/vmwgfx/vmwgfx_msg.c
index 9700099..cdff992 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_msg.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_msg.c
@@ -328,7 +328,7 @@ int vmw_host_get_guestinfo(const char *guest_info_param,
 {
 	struct rpc_channel channel;
 	char *msg, *reply = NULL;
-	size_t msg_len, reply_len = 0;
+	size_t reply_len = 0;
 	int ret = 0;
 
 
@@ -338,15 +338,12 @@ int vmw_host_get_guestinfo(const char *guest_info_param,
 	if (!guest_info_param || !length)
 		return -EINVAL;
 
-	msg_len = strlen(guest_info_param) + strlen("info-get ") + 1;
-	msg = kzalloc(msg_len, GFP_KERNEL);
+	msg = kasprintf(GFP_KERNEL, "info-get %s", guest_info_param);
 	if (!msg) {
 		DRM_ERROR("Cannot allocate memory to get %s", guest_info_param);
 		return -ENOMEM;
 	}
 
-	sprintf(msg, "info-get %s", guest_info_param);
-
 	if (vmw_open_channel(&channel, RPCI_PROTOCOL_NUM) ||
 	    vmw_send_msg(&channel, msg) ||
 	    vmw_recv_msg(&channel, (void *) &reply, &reply_len) ||
@@ -388,7 +385,6 @@ int vmw_host_log(const char *log)
 {
 	struct rpc_channel channel;
 	char *msg;
-	int msg_len;
 	int ret = 0;
 
 
@@ -398,15 +394,12 @@ int vmw_host_log(const char *log)
 	if (!log)
 		return ret;
 
-	msg_len = strlen(log) + strlen("log ") + 1;
-	msg = kzalloc(msg_len, GFP_KERNEL);
+	msg = kasprintf(GFP_KERNEL, "log %s", log);
 	if (!msg) {
 		DRM_ERROR("Cannot allocate memory for log message\n");
 		return -ENOMEM;
 	}
 
-	sprintf(msg, "log %s", log);
-
 	if (vmw_open_channel(&channel, RPCI_PROTOCOL_NUM) ||
 	    vmw_send_msg(&channel, msg) ||
 	    vmw_close_channel(&channel)) {
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
index 9e10145..6b3a942 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
@@ -354,6 +354,7 @@ void vmw_dmabuf_bo_free(struct ttm_buffer_object *bo)
 {
 	struct vmw_dma_buffer *vmw_bo = vmw_dma_buffer(bo);
 
+	vmw_dma_buffer_unmap(vmw_bo);
 	kfree(vmw_bo);
 }
 
@@ -361,6 +362,7 @@ static void vmw_user_dmabuf_destroy(struct ttm_buffer_object *bo)
 {
 	struct vmw_user_dma_buffer *vmw_user_bo = vmw_user_dma_buffer(bo);
 
+	vmw_dma_buffer_unmap(&vmw_user_bo->dma);
 	ttm_prime_object_kfree(vmw_user_bo, prime);
 }
 
@@ -1239,6 +1241,12 @@ void vmw_resource_move_notify(struct ttm_buffer_object *bo,
 
 	dma_buf = container_of(bo, struct vmw_dma_buffer, base);
 
+	/*
+	 * Kill any cached kernel maps before move. An optimization could
+	 * be to do this iff source or destination memory type is VRAM.
+	 */
+	vmw_dma_buffer_unmap(dma_buf);
+
 	if (mem->mem_type != VMW_PL_MOB) {
 		struct vmw_resource *res, *n;
 		struct ttm_validate_buffer val_buf;
@@ -1262,6 +1270,21 @@ void vmw_resource_move_notify(struct ttm_buffer_object *bo,
 }
 
 
+/**
+ * vmw_resource_swap_notify - swapout notify callback.
+ *
+ * @bo: The buffer object to be swapped out.
+ */
+void vmw_resource_swap_notify(struct ttm_buffer_object *bo)
+{
+	if (bo->destroy != vmw_dmabuf_bo_free &&
+	    bo->destroy != vmw_user_dmabuf_destroy)
+		return;
+
+	/* Kill any cached kernel maps before swapout */
+	vmw_dma_buffer_unmap(vmw_dma_buffer(bo));
+}
+
 
 /**
  * vmw_query_readback_all - Read back cached query states
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c b/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c
index 63a4cd7..419185f 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c
@@ -316,69 +316,21 @@ static int vmw_sou_crtc_page_flip(struct drm_crtc *crtc,
 				  struct drm_modeset_acquire_ctx *ctx)
 {
 	struct vmw_private *dev_priv = vmw_priv(crtc->dev);
-	struct drm_framebuffer *old_fb = crtc->primary->fb;
-	struct vmw_framebuffer *vfb = vmw_framebuffer_to_vfb(new_fb);
-	struct vmw_fence_obj *fence = NULL;
-	struct drm_vmw_rect vclips;
 	int ret;
 
 	if (!vmw_kms_crtc_flippable(dev_priv, crtc))
 		return -EINVAL;
 
-	flags &= ~DRM_MODE_PAGE_FLIP_ASYNC;
-	ret = drm_atomic_helper_page_flip(crtc, new_fb, NULL, flags, ctx);
+	ret = drm_atomic_helper_page_flip(crtc, new_fb, event, flags, ctx);
 	if (ret) {
 		DRM_ERROR("Page flip error %d.\n", ret);
 		return ret;
 	}
 
-	/* do a full screen dirty update */
-	vclips.x = crtc->x;
-	vclips.y = crtc->y;
-	vclips.w = crtc->mode.hdisplay;
-	vclips.h = crtc->mode.vdisplay;
-
-	if (vfb->dmabuf)
-		ret = vmw_kms_sou_do_dmabuf_dirty(dev_priv, vfb,
-						  NULL, &vclips, 1, 1,
-						  true, &fence);
-	else
-		ret = vmw_kms_sou_do_surface_dirty(dev_priv, vfb,
-						   NULL, &vclips, NULL,
-						   0, 0, 1, 1, &fence);
-
-
-	if (ret != 0)
-		goto out_no_fence;
-	if (!fence) {
-		ret = -EINVAL;
-		goto out_no_fence;
-	}
-
-	if (event) {
-		struct drm_file *file_priv = event->base.file_priv;
-
-		ret = vmw_event_fence_action_queue(file_priv, fence,
-						   &event->base,
-						   &event->event.vbl.tv_sec,
-						   &event->event.vbl.tv_usec,
-						   true);
-	}
-
-	/*
-	 * No need to hold on to this now. The only cleanup
-	 * we need to do if we fail is unref the fence.
-	 */
-	vmw_fence_obj_unreference(&fence);
-
 	if (vmw_crtc_to_du(crtc)->is_implicit)
 		vmw_kms_update_implicit_fb(dev_priv, crtc);
 
 	return ret;
-
-out_no_fence:
-	drm_atomic_set_fb_for_plane(crtc->primary->state, old_fb);
-	return ret;
 }
 
 static const struct drm_crtc_funcs vmw_screen_object_crtc_funcs = {
@@ -453,7 +405,11 @@ vmw_sou_primary_plane_cleanup_fb(struct drm_plane *plane,
 				 struct drm_plane_state *old_state)
 {
 	struct vmw_plane_state *vps = vmw_plane_state_to_vps(old_state);
+	struct drm_crtc *crtc = plane->state->crtc ?
+		plane->state->crtc : old_state->crtc;
 
+	if (vps->dmabuf)
+		vmw_dmabuf_unpin(vmw_priv(crtc->dev), vps->dmabuf, false);
 	vmw_dmabuf_unreference(&vps->dmabuf);
 	vps->dmabuf_size = 0;
 
@@ -491,10 +447,17 @@ vmw_sou_primary_plane_prepare_fb(struct drm_plane *plane,
 	}
 
 	size = new_state->crtc_w * new_state->crtc_h * 4;
+	dev_priv = vmw_priv(crtc->dev);
 
 	if (vps->dmabuf) {
-		if (vps->dmabuf_size == size)
-			return 0;
+		if (vps->dmabuf_size == size) {
+			/*
+			 * Note that this might temporarily up the pin-count
+			 * to 2, until cleanup_fb() is called.
+			 */
+			return vmw_dmabuf_pin_in_vram(dev_priv, vps->dmabuf,
+						      true);
+		}
 
 		vmw_dmabuf_unreference(&vps->dmabuf);
 		vps->dmabuf_size = 0;
@@ -504,7 +467,6 @@ vmw_sou_primary_plane_prepare_fb(struct drm_plane *plane,
 	if (!vps->dmabuf)
 		return -ENOMEM;
 
-	dev_priv = vmw_priv(crtc->dev);
 	vmw_svga_enable(dev_priv);
 
 	/* After we have alloced the backing store might not be able to
@@ -515,13 +477,16 @@ vmw_sou_primary_plane_prepare_fb(struct drm_plane *plane,
 			      &vmw_vram_ne_placement,
 			      false, &vmw_dmabuf_bo_free);
 	vmw_overlay_resume_all(dev_priv);
-
-	if (ret != 0)
+	if (ret) {
 		vps->dmabuf = NULL; /* vmw_dmabuf_init frees on error */
-	else
-		vps->dmabuf_size = size;
+		return ret;
+	}
 
-	return ret;
+	/*
+	 * TTM already thinks the buffer is pinned, but make sure the
+	 * pin_count is upped.
+	 */
+	return vmw_dmabuf_pin_in_vram(dev_priv, vps->dmabuf, true);
 }
 
 
@@ -530,9 +495,71 @@ vmw_sou_primary_plane_atomic_update(struct drm_plane *plane,
 				    struct drm_plane_state *old_state)
 {
 	struct drm_crtc *crtc = plane->state->crtc;
+	struct drm_pending_vblank_event *event = NULL;
+	struct vmw_fence_obj *fence = NULL;
+	int ret;
+
+	if (crtc && plane->state->fb) {
+		struct vmw_private *dev_priv = vmw_priv(crtc->dev);
+		struct vmw_framebuffer *vfb =
+			vmw_framebuffer_to_vfb(plane->state->fb);
+		struct drm_vmw_rect vclips;
+
+		vclips.x = crtc->x;
+		vclips.y = crtc->y;
+		vclips.w = crtc->mode.hdisplay;
+		vclips.h = crtc->mode.vdisplay;
+
+		if (vfb->dmabuf)
+			ret = vmw_kms_sou_do_dmabuf_dirty(dev_priv, vfb, NULL,
+							  &vclips, 1, 1, true,
+							  &fence, crtc);
+		else
+			ret = vmw_kms_sou_do_surface_dirty(dev_priv, vfb, NULL,
+							   &vclips, NULL, 0, 0,
+							   1, 1, &fence, crtc);
+
+		/*
+		 * We cannot really fail this function, so if we do, then output
+		 * an error and maintain consistent atomic state.
+		 */
+		if (ret != 0)
+			DRM_ERROR("Failed to update screen.\n");
 
-	if (crtc)
 		crtc->primary->fb = plane->state->fb;
+	} else {
+		/*
+		 * When disabling a plane, CRTC and FB should always be NULL
+		 * together, otherwise it's an error.
+		 * Here primary plane is being disable so should really blank
+		 * the screen object display unit, if not already done.
+		 */
+		return;
+	}
+
+	event = crtc->state->event;
+	/*
+	 * In case of failure and other cases, vblank event will be sent in
+	 * vmw_du_crtc_atomic_flush.
+	 */
+	if (event && fence) {
+		struct drm_file *file_priv = event->base.file_priv;
+
+		ret = vmw_event_fence_action_queue(file_priv,
+						   fence,
+						   &event->base,
+						   &event->event.vbl.tv_sec,
+						   &event->event.vbl.tv_usec,
+						   true);
+
+		if (unlikely(ret != 0))
+			DRM_ERROR("Failed to queue event on fence.\n");
+		else
+			crtc->state->event = NULL;
+	}
+
+	if (fence)
+		vmw_fence_obj_unreference(&fence);
 }
 
 
@@ -892,6 +919,7 @@ static void vmw_sou_surface_clip(struct vmw_kms_dirty *dirty)
  * @out_fence: If non-NULL, will return a ref-counted pointer to a
  * struct vmw_fence_obj. The returned fence pointer may be NULL in which
  * case the device has already synchronized.
+ * @crtc: If crtc is passed, perform surface dirty on that crtc only.
  *
  * Returns 0 on success, negative error code on failure. -ERESTARTSYS if
  * interrupted.
@@ -904,7 +932,8 @@ int vmw_kms_sou_do_surface_dirty(struct vmw_private *dev_priv,
 				 s32 dest_x,
 				 s32 dest_y,
 				 unsigned num_clips, int inc,
-				 struct vmw_fence_obj **out_fence)
+				 struct vmw_fence_obj **out_fence,
+				 struct drm_crtc *crtc)
 {
 	struct vmw_framebuffer_surface *vfbs =
 		container_of(framebuffer, typeof(*vfbs), base);
@@ -923,6 +952,7 @@ int vmw_kms_sou_do_surface_dirty(struct vmw_private *dev_priv,
 	sdirty.base.dev_priv = dev_priv;
 	sdirty.base.fifo_reserve_size = sizeof(struct vmw_kms_sou_dirty_cmd) +
 	  sizeof(SVGASignedRect) * num_clips;
+	sdirty.base.crtc = crtc;
 
 	sdirty.sid = srf->id;
 	sdirty.left = sdirty.top = S32_MAX;
@@ -994,6 +1024,7 @@ static void vmw_sou_dmabuf_clip(struct vmw_kms_dirty *dirty)
  * @out_fence: If non-NULL, will return a ref-counted pointer to a
  * struct vmw_fence_obj. The returned fence pointer may be NULL in which
  * case the device has already synchronized.
+ * @crtc: If crtc is passed, perform dmabuf dirty on that crtc only.
  *
  * Returns 0 on success, negative error code on failure. -ERESTARTSYS if
  * interrupted.
@@ -1004,7 +1035,8 @@ int vmw_kms_sou_do_dmabuf_dirty(struct vmw_private *dev_priv,
 				struct drm_vmw_rect *vclips,
 				unsigned num_clips, int increment,
 				bool interruptible,
-				struct vmw_fence_obj **out_fence)
+				struct vmw_fence_obj **out_fence,
+				struct drm_crtc *crtc)
 {
 	struct vmw_dma_buffer *buf =
 		container_of(framebuffer, struct vmw_framebuffer_dmabuf,
@@ -1013,7 +1045,7 @@ int vmw_kms_sou_do_dmabuf_dirty(struct vmw_private *dev_priv,
 	int ret;
 
 	ret = vmw_kms_helper_buffer_prepare(dev_priv, buf, interruptible,
-					    false);
+					    false, false);
 	if (ret)
 		return ret;
 
@@ -1021,6 +1053,7 @@ int vmw_kms_sou_do_dmabuf_dirty(struct vmw_private *dev_priv,
 	if (unlikely(ret != 0))
 		goto out_revert;
 
+	dirty.crtc = crtc;
 	dirty.fifo_commit = vmw_sou_dmabuf_fifo_commit;
 	dirty.clip = vmw_sou_dmabuf_clip;
 	dirty.fifo_reserve_size = sizeof(struct vmw_kms_sou_dmabuf_blit) *
@@ -1092,6 +1125,7 @@ static void vmw_sou_readback_clip(struct vmw_kms_dirty *dirty)
  * Must be set to non-NULL if @file_priv is non-NULL.
  * @vclips: Array of clip rects.
  * @num_clips: Number of clip rects in @vclips.
+ * @crtc: If crtc is passed, readback on that crtc only.
  *
  * Returns 0 on success, negative error code on failure. -ERESTARTSYS if
  * interrupted.
@@ -1101,14 +1135,16 @@ int vmw_kms_sou_readback(struct vmw_private *dev_priv,
 			 struct vmw_framebuffer *vfb,
 			 struct drm_vmw_fence_rep __user *user_fence_rep,
 			 struct drm_vmw_rect *vclips,
-			 uint32_t num_clips)
+			 uint32_t num_clips,
+			 struct drm_crtc *crtc)
 {
 	struct vmw_dma_buffer *buf =
 		container_of(vfb, struct vmw_framebuffer_dmabuf, base)->buffer;
 	struct vmw_kms_dirty dirty;
 	int ret;
 
-	ret = vmw_kms_helper_buffer_prepare(dev_priv, buf, true, false);
+	ret = vmw_kms_helper_buffer_prepare(dev_priv, buf, true, false,
+					    false);
 	if (ret)
 		return ret;
 
@@ -1116,6 +1152,7 @@ int vmw_kms_sou_readback(struct vmw_private *dev_priv,
 	if (unlikely(ret != 0))
 		goto out_revert;
 
+	dirty.crtc = crtc;
 	dirty.fifo_commit = vmw_sou_readback_fifo_commit;
 	dirty.clip = vmw_sou_readback_clip;
 	dirty.fifo_reserve_size = sizeof(struct vmw_kms_sou_readback_blit) *
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c b/drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c
index b68d748..8eec889 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c
@@ -114,7 +114,6 @@ struct vmw_screen_target_display_unit {
 	bool defined;
 
 	/* For CPU Blit */
-	struct ttm_bo_kmap_obj host_map;
 	unsigned int cpp;
 };
 
@@ -492,71 +491,17 @@ static int vmw_stdu_crtc_page_flip(struct drm_crtc *crtc,
 {
 	struct vmw_private *dev_priv = vmw_priv(crtc->dev);
 	struct vmw_screen_target_display_unit *stdu = vmw_crtc_to_stdu(crtc);
-	struct vmw_framebuffer *vfb = vmw_framebuffer_to_vfb(new_fb);
-	struct drm_vmw_rect vclips;
 	int ret;
 
-	dev_priv          = vmw_priv(crtc->dev);
-	stdu              = vmw_crtc_to_stdu(crtc);
-
 	if (!stdu->defined || !vmw_kms_crtc_flippable(dev_priv, crtc))
 		return -EINVAL;
 
-	/*
-	 * We're always async, but the helper doesn't know how to set async
-	 * so lie to the helper. Also, the helper expects someone
-	 * to pick the event up from the crtc state, and if nobody does,
-	 * it will free it. Since we handle the event in this function,
-	 * don't hand it to the helper.
-	 */
-	flags &= ~DRM_MODE_PAGE_FLIP_ASYNC;
-	ret = drm_atomic_helper_page_flip(crtc, new_fb, NULL, flags, ctx);
+	ret = drm_atomic_helper_page_flip(crtc, new_fb, event, flags, ctx);
 	if (ret) {
 		DRM_ERROR("Page flip error %d.\n", ret);
 		return ret;
 	}
 
-	if (stdu->base.is_implicit)
-		vmw_kms_update_implicit_fb(dev_priv, crtc);
-
-	/*
-	 * Now that we've bound a new surface to the screen target,
-	 * update the contents.
-	 */
-	vclips.x = crtc->x;
-	vclips.y = crtc->y;
-	vclips.w = crtc->mode.hdisplay;
-	vclips.h = crtc->mode.vdisplay;
-
-	if (vfb->dmabuf)
-		ret = vmw_kms_stdu_dma(dev_priv, NULL, vfb, NULL, NULL, &vclips,
-				       1, 1, true, false);
-	else
-		ret = vmw_kms_stdu_surface_dirty(dev_priv, vfb, NULL, &vclips,
-						 NULL, 0, 0, 1, 1, NULL);
-	if (ret) {
-		DRM_ERROR("Page flip update error %d.\n", ret);
-		return ret;
-	}
-
-	if (event) {
-		struct vmw_fence_obj *fence = NULL;
-		struct drm_file *file_priv = event->base.file_priv;
-
-		vmw_execbuf_fence_commands(NULL, dev_priv, &fence, NULL);
-		if (!fence)
-			return -ENOMEM;
-
-		ret = vmw_event_fence_action_queue(file_priv, fence,
-						   &event->base,
-						   &event->event.vbl.tv_sec,
-						   &event->event.vbl.tv_usec,
-						   true);
-		vmw_fence_obj_unreference(&fence);
-	} else {
-		(void) vmw_fifo_flush(dev_priv, false);
-	}
-
 	return 0;
 }
 
@@ -693,10 +638,9 @@ static void vmw_stdu_dmabuf_cpu_commit(struct vmw_kms_dirty *dirty)
 		container_of(dirty->unit, typeof(*stdu), base);
 	s32 width, height;
 	s32 src_pitch, dst_pitch;
-	u8 *src, *dst;
-	bool not_used;
-	struct ttm_bo_kmap_obj guest_map;
-	int ret;
+	struct ttm_buffer_object *src_bo, *dst_bo;
+	u32 src_offset, dst_offset;
+	struct vmw_diff_cpy diff = VMW_CPU_BLIT_DIFF_INITIALIZER(stdu->cpp);
 
 	if (!dirty->num_hits)
 		return;
@@ -707,57 +651,38 @@ static void vmw_stdu_dmabuf_cpu_commit(struct vmw_kms_dirty *dirty)
 	if (width == 0 || height == 0)
 		return;
 
-	ret = ttm_bo_kmap(&ddirty->buf->base, 0, ddirty->buf->base.num_pages,
-			  &guest_map);
-	if (ret) {
-		DRM_ERROR("Failed mapping framebuffer for blit: %d\n",
-			  ret);
-		goto out_cleanup;
-	}
-
-	/* Assume we are blitting from Host (display_srf) to Guest (dmabuf) */
-	src_pitch = stdu->display_srf->base_size.width * stdu->cpp;
-	src = ttm_kmap_obj_virtual(&stdu->host_map, &not_used);
-	src += ddirty->top * src_pitch + ddirty->left * stdu->cpp;
-
-	dst_pitch = ddirty->pitch;
-	dst = ttm_kmap_obj_virtual(&guest_map, &not_used);
-	dst += ddirty->fb_top * dst_pitch + ddirty->fb_left * stdu->cpp;
-
-
-	/* Figure out the real direction */
-	if (ddirty->transfer == SVGA3D_WRITE_HOST_VRAM) {
-		u8 *tmp;
-		s32 tmp_pitch;
-
-		tmp = src;
-		tmp_pitch = src_pitch;
+	/* Assume we are blitting from Guest (dmabuf) to Host (display_srf) */
+	dst_pitch = stdu->display_srf->base_size.width * stdu->cpp;
+	dst_bo = &stdu->display_srf->res.backup->base;
+	dst_offset = ddirty->top * dst_pitch + ddirty->left * stdu->cpp;
 
-		src = dst;
-		src_pitch = dst_pitch;
+	src_pitch = ddirty->pitch;
+	src_bo = &ddirty->buf->base;
+	src_offset = ddirty->fb_top * src_pitch + ddirty->fb_left * stdu->cpp;
 
-		dst = tmp;
-		dst_pitch = tmp_pitch;
+	/* Swap src and dst if the assumption was wrong. */
+	if (ddirty->transfer != SVGA3D_WRITE_HOST_VRAM) {
+		swap(dst_pitch, src_pitch);
+		swap(dst_bo, src_bo);
+		swap(src_offset, dst_offset);
 	}
 
-	/* CPU Blit */
-	while (height-- > 0) {
-		memcpy(dst, src, width * stdu->cpp);
-		dst += dst_pitch;
-		src += src_pitch;
-	}
+	(void) vmw_bo_cpu_blit(dst_bo, dst_offset, dst_pitch,
+			       src_bo, src_offset, src_pitch,
+			       width * stdu->cpp, height, &diff);
 
-	if (ddirty->transfer == SVGA3D_WRITE_HOST_VRAM) {
+	if (ddirty->transfer == SVGA3D_WRITE_HOST_VRAM &&
+	    drm_rect_visible(&diff.rect)) {
 		struct vmw_private *dev_priv;
 		struct vmw_stdu_update *cmd;
 		struct drm_clip_rect region;
 		int ret;
 
 		/* We are updating the actual surface, not a proxy */
-		region.x1 = ddirty->left;
-		region.x2 = ddirty->right;
-		region.y1 = ddirty->top;
-		region.y2 = ddirty->bottom;
+		region.x1 = diff.rect.x1;
+		region.x2 = diff.rect.x2;
+		region.y1 = diff.rect.y1;
+		region.y2 = diff.rect.y2;
 		ret = vmw_kms_update_proxy(
 			(struct vmw_resource *) &stdu->display_srf->res,
 			(const struct drm_clip_rect *) &region, 1, 1);
@@ -774,13 +699,12 @@ static void vmw_stdu_dmabuf_cpu_commit(struct vmw_kms_dirty *dirty)
 		}
 
 		vmw_stdu_populate_update(cmd, stdu->base.unit,
-					 ddirty->left, ddirty->right,
-					 ddirty->top, ddirty->bottom);
+					 region.x1, region.x2,
+					 region.y1, region.y2);
 
 		vmw_fifo_commit(dev_priv, sizeof(*cmd));
 	}
 
-	ttm_bo_kunmap(&guest_map);
 out_cleanup:
 	ddirty->left = ddirty->top = ddirty->fb_left = ddirty->fb_top = S32_MAX;
 	ddirty->right = ddirty->bottom = S32_MIN;
@@ -802,6 +726,7 @@ out_cleanup:
  * @to_surface: Whether to DMA to the screen target system as opposed to
  * from the screen target system.
  * @interruptible: Whether to perform waits interruptible if possible.
+ * @crtc: If crtc is passed, perform stdu dma on that crtc only.
  *
  * If DMA-ing till the screen target system, the function will also notify
  * the screen target system that a bounding box of the cliprects has been
@@ -818,15 +743,22 @@ int vmw_kms_stdu_dma(struct vmw_private *dev_priv,
 		     uint32_t num_clips,
 		     int increment,
 		     bool to_surface,
-		     bool interruptible)
+		     bool interruptible,
+		     struct drm_crtc *crtc)
 {
 	struct vmw_dma_buffer *buf =
 		container_of(vfb, struct vmw_framebuffer_dmabuf, base)->buffer;
 	struct vmw_stdu_dirty ddirty;
 	int ret;
+	bool cpu_blit = !(dev_priv->capabilities & SVGA_CAP_3D);
 
+	/*
+	 * VMs without 3D support don't have the surface DMA command and
+	 * we'll be using a CPU blit, and the framebuffer should be moved out
+	 * of VRAM.
+	 */
 	ret = vmw_kms_helper_buffer_prepare(dev_priv, buf, interruptible,
-					    false);
+					    false, cpu_blit);
 	if (ret)
 		return ret;
 
@@ -845,13 +777,15 @@ int vmw_kms_stdu_dma(struct vmw_private *dev_priv,
 	if (to_surface)
 		ddirty.base.fifo_reserve_size += sizeof(struct vmw_stdu_update);
 
-	/* 2D VMs cannot use SVGA_3D_CMD_SURFACE_DMA so do CPU blit instead */
-	if (!(dev_priv->capabilities & SVGA_CAP_3D)) {
+
+	if (cpu_blit) {
 		ddirty.base.fifo_commit = vmw_stdu_dmabuf_cpu_commit;
 		ddirty.base.clip = vmw_stdu_dmabuf_cpu_clip;
 		ddirty.base.fifo_reserve_size = 0;
 	}
 
+	ddirty.base.crtc = crtc;
+
 	ret = vmw_kms_helper_dirty(dev_priv, vfb, clips, vclips,
 				   0, 0, num_clips, increment, &ddirty.base);
 	vmw_kms_helper_buffer_finish(dev_priv, file_priv, buf, NULL,
@@ -963,6 +897,7 @@ static void vmw_kms_stdu_surface_fifo_commit(struct vmw_kms_dirty *dirty)
  * @out_fence: If non-NULL, will return a ref-counted pointer to a
  * struct vmw_fence_obj. The returned fence pointer may be NULL in which
  * case the device has already synchronized.
+ * @crtc: If crtc is passed, perform surface dirty on that crtc only.
  *
  * Returns 0 on success, negative error code on failure. -ERESTARTSYS if
  * interrupted.
@@ -975,7 +910,8 @@ int vmw_kms_stdu_surface_dirty(struct vmw_private *dev_priv,
 			       s32 dest_x,
 			       s32 dest_y,
 			       unsigned num_clips, int inc,
-			       struct vmw_fence_obj **out_fence)
+			       struct vmw_fence_obj **out_fence,
+			       struct drm_crtc *crtc)
 {
 	struct vmw_framebuffer_surface *vfbs =
 		container_of(framebuffer, typeof(*vfbs), base);
@@ -1000,6 +936,7 @@ int vmw_kms_stdu_surface_dirty(struct vmw_private *dev_priv,
 	sdirty.base.fifo_reserve_size = sizeof(struct vmw_stdu_surface_copy) +
 		sizeof(SVGA3dCopyBox) * num_clips +
 		sizeof(struct vmw_stdu_update);
+	sdirty.base.crtc = crtc;
 	sdirty.sid = srf->id;
 	sdirty.left = sdirty.top = S32_MAX;
 	sdirty.right = sdirty.bottom = S32_MIN;
@@ -1118,9 +1055,6 @@ vmw_stdu_primary_plane_cleanup_fb(struct drm_plane *plane,
 {
 	struct vmw_plane_state *vps = vmw_plane_state_to_vps(old_state);
 
-	if (vps->host_map.virtual)
-		ttm_bo_kunmap(&vps->host_map);
-
 	if (vps->surf)
 		WARN_ON(!vps->pinned);
 
@@ -1282,24 +1216,11 @@ vmw_stdu_primary_plane_prepare_fb(struct drm_plane *plane,
 	 * so cache these mappings
 	 */
 	if (vps->content_fb_type == SEPARATE_DMA &&
-	    !(dev_priv->capabilities & SVGA_CAP_3D)) {
-		ret = ttm_bo_kmap(&vps->surf->res.backup->base, 0,
-				  vps->surf->res.backup->base.num_pages,
-				  &vps->host_map);
-		if (ret) {
-			DRM_ERROR("Failed to map display buffer to CPU\n");
-			goto out_srf_unpin;
-		}
-
+	    !(dev_priv->capabilities & SVGA_CAP_3D))
 		vps->cpp = new_fb->pitches[0] / new_fb->width;
-	}
 
 	return 0;
 
-out_srf_unpin:
-	vmw_resource_unpin(&vps->surf->res);
-	vps->pinned--;
-
 out_srf_unref:
 	vmw_surface_unreference(&vps->surf);
 	return ret;
@@ -1322,41 +1243,104 @@ static void
 vmw_stdu_primary_plane_atomic_update(struct drm_plane *plane,
 				     struct drm_plane_state *old_state)
 {
-	struct vmw_private *dev_priv;
-	struct vmw_screen_target_display_unit *stdu;
 	struct vmw_plane_state *vps = vmw_plane_state_to_vps(plane->state);
-	struct drm_crtc *crtc = plane->state->crtc ?: old_state->crtc;
+	struct drm_crtc *crtc = plane->state->crtc;
+	struct vmw_screen_target_display_unit *stdu;
+	struct drm_pending_vblank_event *event;
+	struct vmw_private *dev_priv;
 	int ret;
 
-	stdu     = vmw_crtc_to_stdu(crtc);
-	dev_priv = vmw_priv(crtc->dev);
+	/*
+	 * We cannot really fail this function, so if we do, then output an
+	 * error and maintain consistent atomic state.
+	 */
+	if (crtc && plane->state->fb) {
+		struct vmw_framebuffer *vfb =
+			vmw_framebuffer_to_vfb(plane->state->fb);
+		struct drm_vmw_rect vclips;
+		stdu = vmw_crtc_to_stdu(crtc);
+		dev_priv = vmw_priv(crtc->dev);
+
+		stdu->display_srf = vps->surf;
+		stdu->content_fb_type = vps->content_fb_type;
+		stdu->cpp = vps->cpp;
+
+		vclips.x = crtc->x;
+		vclips.y = crtc->y;
+		vclips.w = crtc->mode.hdisplay;
+		vclips.h = crtc->mode.vdisplay;
+
+		ret = vmw_stdu_bind_st(dev_priv, stdu, &stdu->display_srf->res);
+		if (ret)
+			DRM_ERROR("Failed to bind surface to STDU.\n");
+
+		if (vfb->dmabuf)
+			ret = vmw_kms_stdu_dma(dev_priv, NULL, vfb, NULL, NULL,
+					       &vclips, 1, 1, true, false,
+					       crtc);
+		else
+			ret = vmw_kms_stdu_surface_dirty(dev_priv, vfb, NULL,
+							 &vclips, NULL, 0, 0,
+							 1, 1, NULL, crtc);
+		if (ret)
+			DRM_ERROR("Failed to update STDU.\n");
 
-	stdu->display_srf = vps->surf;
-	stdu->content_fb_type = vps->content_fb_type;
-	stdu->cpp = vps->cpp;
-	memcpy(&stdu->host_map, &vps->host_map, sizeof(vps->host_map));
+		crtc->primary->fb = plane->state->fb;
+	} else {
+		crtc = old_state->crtc;
+		stdu = vmw_crtc_to_stdu(crtc);
+		dev_priv = vmw_priv(crtc->dev);
 
-	if (!stdu->defined)
-		return;
+		/*
+		 * When disabling a plane, CRTC and FB should always be NULL
+		 * together, otherwise it's an error.
+		 * Here primary plane is being disable so blank the screen
+		 * target display unit, if not already done.
+		 */
+		if (!stdu->defined)
+			return;
 
-	if (plane->state->fb)
-		ret = vmw_stdu_bind_st(dev_priv, stdu, &stdu->display_srf->res);
-	else
 		ret = vmw_stdu_bind_st(dev_priv, stdu, NULL);
+		if (ret)
+			DRM_ERROR("Failed to blank STDU\n");
+
+		ret = vmw_stdu_update_st(dev_priv, stdu);
+		if (ret)
+			DRM_ERROR("Failed to update STDU.\n");
+
+		return;
+	}
 
+	event = crtc->state->event;
 	/*
-	 * We cannot really fail this function, so if we do, then output an
-	 * error and quit
+	 * In case of failure and other cases, vblank event will be sent in
+	 * vmw_du_crtc_atomic_flush.
 	 */
-	if (ret)
-		DRM_ERROR("Failed to bind surface to STDU.\n");
-	else
-		crtc->primary->fb = plane->state->fb;
+	if (event && (ret == 0)) {
+		struct vmw_fence_obj *fence = NULL;
+		struct drm_file *file_priv = event->base.file_priv;
 
-	ret = vmw_stdu_update_st(dev_priv, stdu);
+		vmw_execbuf_fence_commands(NULL, dev_priv, &fence, NULL);
 
-	if (ret)
-		DRM_ERROR("Failed to update STDU.\n");
+		/*
+		 * If fence is NULL, then already sync.
+		 */
+		if (fence) {
+			ret = vmw_event_fence_action_queue(
+				file_priv, fence, &event->base,
+				&event->event.vbl.tv_sec,
+				&event->event.vbl.tv_usec,
+				true);
+			if (ret)
+				DRM_ERROR("Failed to queue event on fence.\n");
+			else
+				crtc->state->event = NULL;
+
+			vmw_fence_obj_unreference(&fence);
+		}
+	} else {
+		(void) vmw_fifo_flush(dev_priv, false);
+	}
 }
 
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c b/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
index db1bb16..b236c48 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
@@ -345,7 +345,6 @@ static void vmw_hw_surface_destroy(struct vmw_resource *res)
 		dev_priv->used_memory_size -= res->backup_size;
 		mutex_unlock(&dev_priv->cmdbuf_mutex);
 	}
-	vmw_fifo_resource_dec(dev_priv);
 }
 
 /**
@@ -407,6 +406,8 @@ static int vmw_legacy_srf_create(struct vmw_resource *res)
 
 	vmw_surface_define_encode(srf, cmd);
 	vmw_fifo_commit(dev_priv, submit_size);
+	vmw_fifo_resource_inc(dev_priv);
+
 	/*
 	 * Surface memory usage accounting.
 	 */
@@ -558,6 +559,7 @@ static int vmw_legacy_srf_destroy(struct vmw_resource *res)
 	 */
 
 	vmw_resource_release_id(res);
+	vmw_fifo_resource_dec(dev_priv);
 
 	return 0;
 }
@@ -579,15 +581,11 @@ static int vmw_surface_init(struct vmw_private *dev_priv,
 	struct vmw_resource *res = &srf->res;
 
 	BUG_ON(!res_free);
-	if (!dev_priv->has_mob)
-		vmw_fifo_resource_inc(dev_priv);
 	ret = vmw_resource_init(dev_priv, res, true, res_free,
 				(dev_priv->has_mob) ? &vmw_gb_surface_func :
 				&vmw_legacy_surface_func);
 
 	if (unlikely(ret != 0)) {
-		if (!dev_priv->has_mob)
-			vmw_fifo_resource_dec(dev_priv);
 		res_free(res);
 		return ret;
 	}