Merge branch 'drm-vmware-next' into drm-core-next

* drm-vmware-next: (26 commits)
  vmwgfx: Minor cleanups
  vmwgfx: Bump driver minor to advertise support for new ioctls.
  vmwgfx: Be more strict with fb depths when using screen objects
  vmwgfx: Handle device surface memory limit
  vmwgfx: Make sure we always have a user-space handle to use for objects that are backing kms framebuffers.
  vmwgfx: Optimize the command submission resource list
  vmwgfx: Fix up query processing
  vmwgfx: Allow reference and unreference of NULL fence objects.
  vmwgfx: minor dmabuf utilities cleanup
  vmwgfx: Disallow user space to send present and readback commands
  vmwgfx: Add present and readback ioctls
  vmwgfx: Place overlays in GMR area if we can
  vmwgfx: Drop 3D Legacy Display Unit support
  vmwgfx: Require HWV8 for 3d support
  vmwgfx: Add screen object support
  vmwgfx: Add dmabuf helper functions for pinning
  vmwgfx: Refactor common display unit functions to shared file
  vmwgfx: Expand the command checker to cover screen object commands
  vmwgfx: Break out dirty submission code
  vmwgfx: Break out execbuf command processing
  ...

drivers/gpu/drm/ttm/ttm_bo.c

@@ -1295,6 +1295,7 @@ int ttm_bo_create(struct ttm_bo_device *bdev,
 
 	return ret;
 }
+EXPORT_SYMBOL(ttm_bo_create);
 
 static int ttm_bo_force_list_clean(struct ttm_bo_device *bdev,
 					unsigned mem_type, bool allow_errors)

drivers/gpu/drm/vmwgfx/Makefile

@@ -5,6 +5,6 @@ vmwgfx-y := vmwgfx_execbuf.o vmwgfx_gmr.o vmwgfx_kms.o vmwgfx_drv.o \
 	    vmwgfx_fb.o vmwgfx_ioctl.o vmwgfx_resource.o vmwgfx_buffer.o \
 	    vmwgfx_fifo.o vmwgfx_irq.o vmwgfx_ldu.o vmwgfx_ttm_glue.o \
 	    vmwgfx_overlay.o vmwgfx_marker.o vmwgfx_gmrid_manager.o \
-	    vmwgfx_fence.o
+	    vmwgfx_fence.o vmwgfx_dmabuf.o vmwgfx_scrn.o
 
 obj-$(CONFIG_DRM_VMWGFX) := vmwgfx.o

drivers/gpu/drm/vmwgfx/svga_escape.h

@@ -75,7 +75,7 @@
  */
 
 #define SVGA_ESCAPE_VMWARE_HINT               0x00030000
-#define SVGA_ESCAPE_VMWARE_HINT_FULLSCREEN    0x00030001  // Deprecated
+#define SVGA_ESCAPE_VMWARE_HINT_FULLSCREEN    0x00030001  /* Deprecated */
 
 typedef
 struct {

drivers/gpu/drm/vmwgfx/svga_overlay.h

@@ -38,9 +38,9 @@
  * Video formats we support
  */
 
-#define VMWARE_FOURCC_YV12 0x32315659 // 'Y' 'V' '1' '2'
-#define VMWARE_FOURCC_YUY2 0x32595559 // 'Y' 'U' 'Y' '2'
-#define VMWARE_FOURCC_UYVY 0x59565955 // 'U' 'Y' 'V' 'Y'
+#define VMWARE_FOURCC_YV12 0x32315659 /* 'Y' 'V' '1' '2' */
+#define VMWARE_FOURCC_YUY2 0x32595559 /* 'Y' 'U' 'Y' '2' */
+#define VMWARE_FOURCC_UYVY 0x59565955 /* 'U' 'Y' 'V' 'Y' */
 
 typedef enum {
    SVGA_OVERLAY_FORMAT_INVALID = 0,
@@ -68,7 +68,7 @@ struct SVGAEscapeVideoSetRegs {
       uint32 streamId;
    } header;
 
-   // May include zero or more items.
+   /* May include zero or more items. */
    struct {
       uint32 registerId;
       uint32 value;
@@ -134,12 +134,12 @@ struct {
  */
 
 static inline bool
-VMwareVideoGetAttributes(const SVGAOverlayFormat format,    // IN
-                         uint32 *width,                     // IN / OUT
-                         uint32 *height,                    // IN / OUT
-                         uint32 *size,                      // OUT
-                         uint32 *pitches,                   // OUT (optional)
-                         uint32 *offsets)                   // OUT (optional)
+VMwareVideoGetAttributes(const SVGAOverlayFormat format,    /* IN */
+                         uint32 *width,                     /* IN / OUT */
+                         uint32 *height,                    /* IN / OUT */
+                         uint32 *size,                      /* OUT */
+                         uint32 *pitches,                   /* OUT (optional) */
+                         uint32 *offsets)                   /* OUT (optional) */
 {
     int tmp;
 
@@ -198,4 +198,4 @@ VMwareVideoGetAttributes(const SVGAOverlayFormat format,    // IN
     return true;
 }
 
-#endif // _SVGA_OVERLAY_H_
+#endif /* _SVGA_OVERLAY_H_ */

drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c

@@ -42,6 +42,10 @@ static uint32_t sys_placement_flags = TTM_PL_FLAG_SYSTEM |
 static uint32_t gmr_placement_flags = VMW_PL_FLAG_GMR |
 	TTM_PL_FLAG_CACHED;
 
+static uint32_t gmr_ne_placement_flags = VMW_PL_FLAG_GMR |
+	TTM_PL_FLAG_CACHED |
+	TTM_PL_FLAG_NO_EVICT;
+
 struct ttm_placement vmw_vram_placement = {
 	.fpfn = 0,
 	.lpfn = 0,
@@ -56,6 +60,11 @@ static uint32_t vram_gmr_placement_flags[] = {
 	VMW_PL_FLAG_GMR | TTM_PL_FLAG_CACHED
 };
 
+static uint32_t gmr_vram_placement_flags[] = {
+	VMW_PL_FLAG_GMR | TTM_PL_FLAG_CACHED,
+	TTM_PL_FLAG_VRAM | TTM_PL_FLAG_CACHED
+};
+
 struct ttm_placement vmw_vram_gmr_placement = {
 	.fpfn = 0,
 	.lpfn = 0,
@@ -65,6 +74,20 @@ struct ttm_placement vmw_vram_gmr_placement = {
 	.busy_placement = &gmr_placement_flags
 };
 
+static uint32_t vram_gmr_ne_placement_flags[] = {
+	TTM_PL_FLAG_VRAM | TTM_PL_FLAG_CACHED | TTM_PL_FLAG_NO_EVICT,
+	VMW_PL_FLAG_GMR | TTM_PL_FLAG_CACHED | TTM_PL_FLAG_NO_EVICT
+};
+
+struct ttm_placement vmw_vram_gmr_ne_placement = {
+	.fpfn = 0,
+	.lpfn = 0,
+	.num_placement = 2,
+	.placement = vram_gmr_ne_placement_flags,
+	.num_busy_placement = 1,
+	.busy_placement = &gmr_ne_placement_flags
+};
+
 struct ttm_placement vmw_vram_sys_placement = {
 	.fpfn = 0,
 	.lpfn = 0,
@@ -92,6 +115,30 @@ struct ttm_placement vmw_sys_placement = {
 	.busy_placement = &sys_placement_flags
 };
 
+static uint32_t evictable_placement_flags[] = {
+	TTM_PL_FLAG_SYSTEM | TTM_PL_FLAG_CACHED,
+	TTM_PL_FLAG_VRAM | TTM_PL_FLAG_CACHED,
+	VMW_PL_FLAG_GMR | TTM_PL_FLAG_CACHED
+};
+
+struct ttm_placement vmw_evictable_placement = {
+	.fpfn = 0,
+	.lpfn = 0,
+	.num_placement = 3,
+	.placement = evictable_placement_flags,
+	.num_busy_placement = 1,
+	.busy_placement = &sys_placement_flags
+};
+
+struct ttm_placement vmw_srf_placement = {
+	.fpfn = 0,
+	.lpfn = 0,
+	.num_placement = 1,
+	.num_busy_placement = 2,
+	.placement = &gmr_placement_flags,
+	.busy_placement = gmr_vram_placement_flags
+};
+
 struct vmw_ttm_backend {
 	struct ttm_backend backend;
 	struct page **pages;

drivers/gpu/drm/vmwgfx/vmwgfx_dmabuf.c

+/**************************************************************************
+ *
+ * Copyright © 2011 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 "ttm/ttm_placement.h"
+
+#include "drmP.h"
+#include "vmwgfx_drv.h"
+
+
+/**
+ * vmw_dmabuf_to_placement - Validate a buffer to placement.
+ *
+ * @dev_priv:  Driver private.
+ * @buf:  DMA buffer to move.
+ * @pin:  Pin buffer if true.
+ * @interruptible:  Use interruptible wait.
+ *
+ * May only be called by the current master since it assumes that the
+ * master lock is the current master's lock.
+ * This function takes the master's lock in write mode.
+ * Flushes and unpins the query bo to avoid failures.
+ *
+ * Returns
+ *  -ERESTARTSYS if interrupted by a signal.
+ */
+int vmw_dmabuf_to_placement(struct vmw_private *dev_priv,
+			    struct vmw_dma_buffer *buf,
+			    struct ttm_placement *placement,
+			    bool interruptible)
+{
+	struct vmw_master *vmaster = dev_priv->active_master;
+	struct ttm_buffer_object *bo = &buf->base;
+	int ret;
+
+	ret = ttm_write_lock(&vmaster->lock, interruptible);
+	if (unlikely(ret != 0))
+		return ret;
+
+	vmw_execbuf_release_pinned_bo(dev_priv, false, 0);
+
+	ret = ttm_bo_reserve(bo, interruptible, false, false, 0);
+	if (unlikely(ret != 0))
+		goto err;
+
+	ret = ttm_bo_validate(bo, placement, interruptible, false, false);
+
+	ttm_bo_unreserve(bo);
+
+err:
+	ttm_write_unlock(&vmaster->lock);
+	return ret;
+}
+
+/**
+ * vmw_dmabuf_to_vram_or_gmr - Move a buffer to vram or gmr.
+ *
+ * May only be called by the current master since it assumes that the
+ * master lock is the current master's lock.
+ * This function takes the master's lock in write mode.
+ * Flushes and unpins the query bo if @pin == true to avoid failures.
+ *
+ * @dev_priv:  Driver private.
+ * @buf:  DMA buffer to move.
+ * @pin:  Pin buffer if true.
+ * @interruptible:  Use interruptible wait.
+ *
+ * Returns
+ * -ERESTARTSYS if interrupted by a signal.
+ */
+int vmw_dmabuf_to_vram_or_gmr(struct vmw_private *dev_priv,
+			      struct vmw_dma_buffer *buf,
+			      bool pin, bool interruptible)
+{
+	struct vmw_master *vmaster = dev_priv->active_master;
+	struct ttm_buffer_object *bo = &buf->base;
+	struct ttm_placement *placement;
+	int ret;
+
+	ret = ttm_write_lock(&vmaster->lock, interruptible);
+	if (unlikely(ret != 0))
+		return ret;
+
+	if (pin)
+		vmw_execbuf_release_pinned_bo(dev_priv, false, 0);
+
+	ret = ttm_bo_reserve(bo, interruptible, false, false, 0);
+	if (unlikely(ret != 0))
+		goto err;
+
+	/**
+	 * Put BO in VRAM if there is space, otherwise as a GMR.
+	 * If there is no space in VRAM and GMR ids are all used up,
+	 * start evicting GMRs to make room. If the DMA buffer can't be
+	 * used as a GMR, this will return -ENOMEM.
+	 */
+
+	if (pin)
+		placement = &vmw_vram_gmr_ne_placement;
+	else
+		placement = &vmw_vram_gmr_placement;
+
+	ret = ttm_bo_validate(bo, placement, interruptible, false, false);
+	if (likely(ret == 0) || ret == -ERESTARTSYS)
+		goto err_unreserve;
+
+
+	/**
+	 * If that failed, try VRAM again, this time evicting
+	 * previous contents.
+	 */
+
+	if (pin)
+		placement = &vmw_vram_ne_placement;
+	else
+		placement = &vmw_vram_placement;
+
+	ret = ttm_bo_validate(bo, placement, interruptible, false, false);
+
+err_unreserve:
+	ttm_bo_unreserve(bo);
+err:
+	ttm_write_unlock(&vmaster->lock);
+	return ret;
+}
+
+/**
+ * vmw_dmabuf_to_vram - Move a buffer to vram.
+ *
+ * May only be called by the current master since it assumes that the
+ * master lock is the current master's lock.
+ * This function takes the master's lock in write mode.
+ *
+ * @dev_priv:  Driver private.
+ * @buf:  DMA buffer to move.
+ * @pin:  Pin buffer in vram if true.
+ * @interruptible:  Use interruptible wait.
+ *
+ * Returns
+ * -ERESTARTSYS if interrupted by a signal.
+ */
+int vmw_dmabuf_to_vram(struct vmw_private *dev_priv,
+		       struct vmw_dma_buffer *buf,
+		       bool pin, bool interruptible)
+{
+	struct ttm_placement *placement;
+
+	if (pin)
+		placement = &vmw_vram_ne_placement;
+	else
+		placement = &vmw_vram_placement;
+
+	return vmw_dmabuf_to_placement(dev_priv, buf,
+				       placement,
+				       interruptible);
+}
+
+/**
+ * vmw_dmabuf_to_start_of_vram - Move a buffer to start of vram.
+ *
+ * May only be called by the current master since it assumes that the
+ * master lock is the current master's lock.
+ * This function takes the master's lock in write mode.
+ * Flushes and unpins the query bo if @pin == true to avoid failures.
+ *
+ * @dev_priv:  Driver private.
+ * @buf:  DMA buffer to move.
+ * @pin:  Pin buffer in vram if true.
+ * @interruptible:  Use interruptible wait.
+ *
+ * Returns
+ * -ERESTARTSYS if interrupted by a signal.
+ */
+int vmw_dmabuf_to_start_of_vram(struct vmw_private *dev_priv,
+				struct vmw_dma_buffer *buf,
+				bool pin, bool interruptible)
+{
+	struct vmw_master *vmaster = dev_priv->active_master;
+	struct ttm_buffer_object *bo = &buf->base;
+	struct ttm_placement placement;
+	int ret = 0;
+
+	if (pin)
+		placement = vmw_vram_ne_placement;
+	else
+		placement = vmw_vram_placement;
+	placement.lpfn = bo->num_pages;
+
+	ret = ttm_write_lock(&vmaster->lock, interruptible);
+	if (unlikely(ret != 0))
+		return ret;
+
+	if (pin)
+		vmw_execbuf_release_pinned_bo(dev_priv, false, 0);
+
+	ret = ttm_bo_reserve(bo, interruptible, false, false, 0);
+	if (unlikely(ret != 0))
+		goto err_unlock;
+
+	/* Is this buffer already in vram but not at the start of it? */
+	if (bo->mem.mem_type == TTM_PL_VRAM &&
+	    bo->mem.start < bo->num_pages &&
+	    bo->mem.start > 0)
+		(void) ttm_bo_validate(bo, &vmw_sys_placement, false,
+				       false, false);
+
+	ret = ttm_bo_validate(bo, &placement, interruptible, false, false);
+
+	/* For some reason we didn't up at the start of vram */
+	WARN_ON(ret == 0 && bo->offset != 0);
+
+	ttm_bo_unreserve(bo);
+err_unlock:
+	ttm_write_unlock(&vmaster->lock);
+
+	return ret;
+}
+
+
+/**
+ * vmw_dmabuf_upin - Unpin the buffer given buffer, does not move the buffer.
+ *
+ * May only be called by the current master since it assumes that the
+ * master lock is the current master's lock.
+ * This function takes the master's lock in write mode.
+ *
+ * @dev_priv:  Driver private.
+ * @buf:  DMA buffer to unpin.
+ * @interruptible:  Use interruptible wait.
+ *
+ * Returns
+ * -ERESTARTSYS if interrupted by a signal.
+ */
+int vmw_dmabuf_unpin(struct vmw_private *dev_priv,
+		     struct vmw_dma_buffer *buf,
+		     bool interruptible)
+{
+	/*
+	 * We could in theory early out if the buffer is
+	 * unpinned but we need to lock and reserve the buffer
+	 * anyways so we don't gain much by that.
+	 */
+	return vmw_dmabuf_to_placement(dev_priv, buf,
+				       &vmw_evictable_placement,
+				       interruptible);
+}
+
+
+/**
+ * vmw_bo_get_guest_ptr - Get the guest ptr representing the current placement
+ * of a buffer.
+ *
+ * @bo: Pointer to a struct ttm_buffer_object. Must be pinned or reserved.
+ * @ptr: SVGAGuestPtr returning the result.
+ */
+void vmw_bo_get_guest_ptr(const struct ttm_buffer_object *bo,
+			  SVGAGuestPtr *ptr)
+{
+	if (bo->mem.mem_type == TTM_PL_VRAM) {
+		ptr->gmrId = SVGA_GMR_FRAMEBUFFER;
+		ptr->offset = bo->offset;
+	} else {
+		ptr->gmrId = bo->mem.start;
+		ptr->offset = 0;
+	}
+}
+
+
+/**
+ * vmw_bo_pin - Pin or unpin a buffer object without moving it.
+ *
+ * @bo: The buffer object. Must be reserved, and present either in VRAM
+ * or GMR memory.
+ * @pin: Whether to pin or unpin.
+ *
+ */
+void vmw_bo_pin(struct ttm_buffer_object *bo, bool pin)
+{
+	uint32_t pl_flags;
+	struct ttm_placement placement;
+	uint32_t old_mem_type = bo->mem.mem_type;
+	int ret;
+
+	BUG_ON(!atomic_read(&bo->reserved));
+	BUG_ON(old_mem_type != TTM_PL_VRAM &&
+	       old_mem_type != VMW_PL_FLAG_GMR);
+
+	pl_flags = TTM_PL_FLAG_VRAM | VMW_PL_FLAG_GMR | TTM_PL_FLAG_CACHED;
+	if (pin)
+		pl_flags |= TTM_PL_FLAG_NO_EVICT;
+
+	memset(&placement, 0, sizeof(placement));
+	placement.num_placement = 1;
+	placement.placement = &pl_flags;
+
+	ret = ttm_bo_validate(bo, &placement, false, true, true);
+
+	BUG_ON(ret != 0 || bo->mem.mem_type != old_mem_type);
+}

drivers/gpu/drm/vmwgfx/vmwgfx_drv.c

@@ -94,6 +94,12 @@
 #define DRM_IOCTL_VMW_FENCE_UNREF				\
 	DRM_IOW(DRM_COMMAND_BASE + DRM_VMW_FENCE_UNREF,		\
 		 struct drm_vmw_fence_arg)
+#define DRM_IOCTL_VMW_PRESENT					\
+	DRM_IOW(DRM_COMMAND_BASE + DRM_VMW_PRESENT,		\
+		 struct drm_vmw_present_arg)
+#define DRM_IOCTL_VMW_PRESENT_READBACK				\
+	DRM_IOW(DRM_COMMAND_BASE + DRM_VMW_PRESENT_READBACK,	\
+		 struct drm_vmw_present_readback_arg)
 
 /**
  * The core DRM version of this macro doesn't account for
@@ -146,6 +152,13 @@ static struct drm_ioctl_desc vmw_ioctls[] = {
 		      DRM_AUTH | DRM_UNLOCKED),
 	VMW_IOCTL_DEF(VMW_GET_3D_CAP, vmw_get_cap_3d_ioctl,
 		      DRM_AUTH | DRM_UNLOCKED),
+
+	/* these allow direct access to the framebuffers mark as master only */
+	VMW_IOCTL_DEF(VMW_PRESENT, vmw_present_ioctl,
+		      DRM_MASTER | DRM_AUTH | DRM_UNLOCKED),
+	VMW_IOCTL_DEF(VMW_PRESENT_READBACK,
+		      vmw_present_readback_ioctl,
+		      DRM_MASTER | DRM_AUTH | DRM_UNLOCKED),
 };
 
 static struct pci_device_id vmw_pci_id_list[] = {
@@ -200,6 +213,72 @@ static void vmw_print_capabilities(uint32_t capabilities)
 		DRM_INFO("  Screen Object 2.\n");
 }
 
+
+/**
+ * vmw_execbuf_prepare_dummy_query - Initialize a query result structure at
+ * the start of a buffer object.
+ *
+ * @dev_priv: The device private structure.
+ *
+ * This function will idle the buffer using an uninterruptible wait, then
+ * map the first page and initialize a pending occlusion query result structure,
+ * Finally it will unmap the buffer.
+ *
+ * TODO: Since we're only mapping a single page, we should optimize the map
+ * to use kmap_atomic / iomap_atomic.
+ */
+static void vmw_dummy_query_bo_prepare(struct vmw_private *dev_priv)
+{
+	struct ttm_bo_kmap_obj map;
+	volatile SVGA3dQueryResult *result;
+	bool dummy;
+	int ret;
+	struct ttm_bo_device *bdev = &dev_priv->bdev;
+	struct ttm_buffer_object *bo = dev_priv->dummy_query_bo;
+
+	ttm_bo_reserve(bo, false, false, false, 0);
+	spin_lock(&bdev->fence_lock);
+	ret = ttm_bo_wait(bo, false, false, false, TTM_USAGE_READWRITE);
+	spin_unlock(&bdev->fence_lock);
+	if (unlikely(ret != 0))
+		(void) vmw_fallback_wait(dev_priv, false, true, 0, false,
+					 10*HZ);
+
+	ret = ttm_bo_kmap(bo, 0, 1, &map);
+	if (likely(ret == 0)) {
+		result = ttm_kmap_obj_virtual(&map, &dummy);
+		result->totalSize = sizeof(*result);
+		result->state = SVGA3D_QUERYSTATE_PENDING;
+		result->result32 = 0xff;
+		ttm_bo_kunmap(&map);
+	} else
+		DRM_ERROR("Dummy query buffer map failed.\n");
+	ttm_bo_unreserve(bo);
+}
+
+
+/**
+ * vmw_dummy_query_bo_create - create a bo to hold a dummy query result
+ *
+ * @dev_priv: A device private structure.
+ *
+ * This function creates a small buffer object that holds the query
+ * result for dummy queries emitted as query barriers.
+ * No interruptible waits are done within this function.
+ *
+ * Returns an error if bo creation fails.
+ */
+static int vmw_dummy_query_bo_create(struct vmw_private *dev_priv)
+{
+	return ttm_bo_create(&dev_priv->bdev,
+			     PAGE_SIZE,
+			     ttm_bo_type_device,
+			     &vmw_vram_sys_placement,
+			     0, 0, false, NULL,
+			     &dev_priv->dummy_query_bo);
+}
+
+
 static int vmw_request_device(struct vmw_private *dev_priv)
 {
 	int ret;
@@ -210,12 +289,29 @@ static int vmw_request_device(struct vmw_private *dev_priv)
 		return ret;
 	}
 	vmw_fence_fifo_up(dev_priv->fman);
+	ret = vmw_dummy_query_bo_create(dev_priv);
+	if (unlikely(ret != 0))
+		goto out_no_query_bo;
+	vmw_dummy_query_bo_prepare(dev_priv);
 
 	return 0;
+
+out_no_query_bo:
+	vmw_fence_fifo_down(dev_priv->fman);
+	vmw_fifo_release(dev_priv, &dev_priv->fifo);
+	return ret;
 }
 
 static void vmw_release_device(struct vmw_private *dev_priv)
 {
+	/*
+	 * Previous destructions should've released
+	 * the pinned bo.
+	 */
+
+	BUG_ON(dev_priv->pinned_bo != NULL);
+
+	ttm_bo_unref(&dev_priv->dummy_query_bo);
 	vmw_fence_fifo_down(dev_priv->fman);
 	vmw_fifo_release(dev_priv, &dev_priv->fifo);
 }
@@ -306,6 +402,8 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 	init_waitqueue_head(&dev_priv->fifo_queue);
 	dev_priv->fence_queue_waiters = 0;
 	atomic_set(&dev_priv->fifo_queue_waiters, 0);
+	INIT_LIST_HEAD(&dev_priv->surface_lru);
+	dev_priv->used_memory_size = 0;
 
 	dev_priv->io_start = pci_resource_start(dev->pdev, 0);
 	dev_priv->vram_start = pci_resource_start(dev->pdev, 1);
@@ -326,6 +424,10 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 
 	dev_priv->capabilities = vmw_read(dev_priv, SVGA_REG_CAPABILITIES);
 
+	dev_priv->vram_size = vmw_read(dev_priv, SVGA_REG_VRAM_SIZE);
+	dev_priv->mmio_size = vmw_read(dev_priv, SVGA_REG_MEM_SIZE);
+	dev_priv->fb_max_width = vmw_read(dev_priv, SVGA_REG_MAX_WIDTH);
+	dev_priv->fb_max_height = vmw_read(dev_priv, SVGA_REG_MAX_HEIGHT);
 	if (dev_priv->capabilities & SVGA_CAP_GMR) {
 		dev_priv->max_gmr_descriptors =
 			vmw_read(dev_priv,
@@ -338,13 +440,15 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 			vmw_read(dev_priv, SVGA_REG_GMRS_MAX_PAGES);
 		dev_priv->memory_size =
 			vmw_read(dev_priv, SVGA_REG_MEMORY_SIZE);
+		dev_priv->memory_size -= dev_priv->vram_size;
+	} else {
+		/*
+		 * An arbitrary limit of 512MiB on surface
+		 * memory. But all HWV8 hardware supports GMR2.
+		 */
+		dev_priv->memory_size = 512*1024*1024;
 	}
 
-	dev_priv->vram_size = vmw_read(dev_priv, SVGA_REG_VRAM_SIZE);
-	dev_priv->mmio_size = vmw_read(dev_priv, SVGA_REG_MEM_SIZE);
-	dev_priv->fb_max_width = vmw_read(dev_priv, SVGA_REG_MAX_WIDTH);
-	dev_priv->fb_max_height = vmw_read(dev_priv, SVGA_REG_MAX_HEIGHT);
-
 	mutex_unlock(&dev_priv->hw_mutex);
 
 	vmw_print_capabilities(dev_priv->capabilities);
@@ -358,8 +462,8 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 	if (dev_priv->capabilities & SVGA_CAP_GMR2) {
 		DRM_INFO("Max number of GMR pages is %u\n",
 			 (unsigned)dev_priv->max_gmr_pages);
-		DRM_INFO("Max dedicated hypervisor graphics memory is %u\n",
-			 (unsigned)dev_priv->memory_size);
+		DRM_INFO("Max dedicated hypervisor surface memory is %u kiB\n",
+			 (unsigned)dev_priv->memory_size / 1024);
 	}
 	DRM_INFO("VRAM at 0x%08x size is %u kiB\n",
 		 dev_priv->vram_start, dev_priv->vram_size / 1024);
@@ -451,22 +555,30 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 	dev_priv->fman = vmw_fence_manager_init(dev_priv);
 	if (unlikely(dev_priv->fman == NULL))
 		goto out_no_fman;
+
+	/* Need to start the fifo to check if we can do screen objects */
+	ret = vmw_3d_resource_inc(dev_priv, true);
+	if (unlikely(ret != 0))
+		goto out_no_fifo;
+	vmw_kms_save_vga(dev_priv);
+
+	/* Start kms and overlay systems, needs fifo. */
 	ret = vmw_kms_init(dev_priv);
 	if (unlikely(ret != 0))
 		goto out_no_kms;
 	vmw_overlay_init(dev_priv);
+
+	/* 3D Depends on Screen Objects being used. */
+	DRM_INFO("Detected %sdevice 3D availability.\n",
+		 vmw_fifo_have_3d(dev_priv) ?
+		 "" : "no ");
+
+	/* We might be done with the fifo now */
 	if (dev_priv->enable_fb) {
-		ret = vmw_3d_resource_inc(dev_priv, false);
-		if (unlikely(ret != 0))
-			goto out_no_fifo;
-		vmw_kms_save_vga(dev_priv);
 		vmw_fb_init(dev_priv);
-		DRM_INFO("%s", vmw_fifo_have_3d(dev_priv) ?
-			 "Detected device 3D availability.\n" :
-			 "Detected no device 3D availability.\n");
 	} else {
-		DRM_INFO("Delayed 3D detection since we're not "
-			 "running the device in SVGA mode yet.\n");
+		vmw_kms_restore_vga(dev_priv);
+		vmw_3d_resource_dec(dev_priv, true);
 	}
 
 	if (dev_priv->capabilities & SVGA_CAP_IRQMASK) {
@@ -483,15 +595,17 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 	return 0;
 
 out_no_irq:
-	if (dev_priv->enable_fb) {
+	if (dev_priv->enable_fb)
 		vmw_fb_close(dev_priv);
+	vmw_overlay_close(dev_priv);
+	vmw_kms_close(dev_priv);
+out_no_kms:
+	/* We still have a 3D resource reference held */
+	if (dev_priv->enable_fb) {
 		vmw_kms_restore_vga(dev_priv);
 		vmw_3d_resource_dec(dev_priv, false);
 	}
 out_no_fifo:
-	vmw_overlay_close(dev_priv);
-	vmw_kms_close(dev_priv);
-out_no_kms:
 	vmw_fence_manager_takedown(dev_priv->fman);
 out_no_fman:
 	if (dev_priv->stealth)
@@ -771,7 +885,7 @@ static void vmw_master_drop(struct drm_device *dev,
 
 	vmw_fp->locked_master = drm_master_get(file_priv->master);
 	ret = ttm_vt_lock(&vmaster->lock, false, vmw_fp->tfile);
-	vmw_kms_idle_workqueues(vmaster);
+	vmw_execbuf_release_pinned_bo(dev_priv, false, 0);
 
 	if (unlikely((ret != 0))) {
 		DRM_ERROR("Unable to lock TTM at VT switch.\n");
@@ -823,6 +937,7 @@ static int vmwgfx_pm_notifier(struct notifier_block *nb, unsigned long val,
 		 * This empties VRAM and unbinds all GMR bindings.
 		 * Buffer contents is moved to swappable memory.
 		 */
+		vmw_execbuf_release_pinned_bo(dev_priv, false, 0);
 		ttm_bo_swapout_all(&dev_priv->bdev);
 
 		break;

drivers/gpu/drm/vmwgfx/vmwgfx_drv.h

@@ -40,9 +40,9 @@
 #include "ttm/ttm_module.h"
 #include "vmwgfx_fence.h"
 
-#define VMWGFX_DRIVER_DATE "20110901"
+#define VMWGFX_DRIVER_DATE "20110927"
 #define VMWGFX_DRIVER_MAJOR 2
-#define VMWGFX_DRIVER_MINOR 0
+#define VMWGFX_DRIVER_MINOR 1
 #define VMWGFX_DRIVER_PATCHLEVEL 0
 #define VMWGFX_FILE_PAGE_OFFSET 0x00100000
 #define VMWGFX_FIFO_STATIC_SIZE (1024*1024)
@@ -79,9 +79,11 @@ struct vmw_resource {
 	int id;
 	enum ttm_object_type res_type;
 	bool avail;
+	void (*remove_from_lists) (struct vmw_resource *res);
 	void (*hw_destroy) (struct vmw_resource *res);
 	void (*res_free) (struct vmw_resource *res);
-	bool on_validate_list;
+	struct list_head validate_head;
+	struct list_head query_head; /* Protected by the cmdbuf mutex */
 	/* TODO is a generic snooper needed? */
 #if 0
 	void (*snoop)(struct vmw_resource *res,
@@ -97,8 +99,12 @@ struct vmw_cursor_snooper {
 	uint32_t *image;
 };
 
+struct vmw_framebuffer;
+struct vmw_surface_offset;
+
 struct vmw_surface {
 	struct vmw_resource res;
+	struct list_head lru_head; /* Protected by the resource lock */
 	uint32_t flags;
 	uint32_t format;
 	uint32_t mip_levels[DRM_VMW_MAX_SURFACE_FACES];
@@ -109,6 +115,9 @@ struct vmw_surface {
 
 	/* TODO so far just a extra pointer */
 	struct vmw_cursor_snooper snooper;
+	struct ttm_buffer_object *backup;
+	struct vmw_surface_offset *offsets;
+	uint32_t backup_size;
 };
 
 struct vmw_marker_queue {
@@ -139,6 +148,8 @@ struct vmw_sw_context{
 	struct ida bo_list;
 	uint32_t last_cid;
 	bool cid_valid;
+	bool kernel; /**< is the called made from the kernel */
+	struct vmw_resource *cur_ctx;
 	uint32_t last_sid;
 	uint32_t sid_translation;
 	bool sid_valid;
@@ -150,8 +161,12 @@ struct vmw_sw_context{
 	uint32_t cur_val_buf;
 	uint32_t *cmd_bounce;
 	uint32_t cmd_bounce_size;
-	struct vmw_resource *resources[VMWGFX_MAX_VALIDATIONS];
-	uint32_t num_ref_resources;
+	struct list_head resource_list;
+	uint32_t fence_flags;
+	struct list_head query_list;
+	struct ttm_buffer_object *cur_query_bo;
+	uint32_t cur_query_cid;
+	bool query_cid_valid;
 };
 
 struct vmw_legacy_display;
@@ -216,6 +231,7 @@ struct vmw_private {
 
 	void *fb_info;
 	struct vmw_legacy_display *ldu_priv;
+	struct vmw_screen_object_display *sou_priv;
 	struct vmw_overlay *overlay_priv;
 
 	/*
@@ -290,6 +306,26 @@ struct vmw_private {
 
 	struct mutex release_mutex;
 	uint32_t num_3d_resources;
+
+	/*
+	 * Query processing. These members
+	 * are protected by the cmdbuf mutex.
+	 */
+
+	struct ttm_buffer_object *dummy_query_bo;
+	struct ttm_buffer_object *pinned_bo;
+	uint32_t query_cid;
+	bool dummy_query_bo_pinned;
+
+	/*
+	 * Surface swapping. The "surface_lru" list is protected by the
+	 * resource lock in order to be able to destroy a surface and take
+	 * it off the lru atomically. "used_memory_size" is currently
+	 * protected by the cmdbuf mutex for simplicity.
+	 */
+
+	struct list_head surface_lru;
+	uint32_t used_memory_size;
 };
 
 static inline struct vmw_private *vmw_priv(struct drm_device *dev)
@@ -369,6 +405,8 @@ extern int vmw_surface_reference_ioctl(struct drm_device *dev, void *data,
 extern int vmw_surface_check(struct vmw_private *dev_priv,
 			     struct ttm_object_file *tfile,
 			     uint32_t handle, int *id);
+extern int vmw_surface_validate(struct vmw_private *dev_priv,
+				struct vmw_surface *srf);
 extern void vmw_dmabuf_bo_free(struct ttm_buffer_object *bo);
 extern int vmw_dmabuf_init(struct vmw_private *dev_priv,
 			   struct vmw_dma_buffer *vmw_bo,
@@ -384,10 +422,6 @@ extern uint32_t vmw_dmabuf_validate_node(struct ttm_buffer_object *bo,
 extern void vmw_dmabuf_validate_clear(struct ttm_buffer_object *bo);
 extern int vmw_user_dmabuf_lookup(struct ttm_object_file *tfile,
 				  uint32_t id, struct vmw_dma_buffer **out);
-extern int vmw_dmabuf_to_start_of_vram(struct vmw_private *vmw_priv,
-				       struct vmw_dma_buffer *bo);
-extern int vmw_dmabuf_from_vram(struct vmw_private *vmw_priv,
-				struct vmw_dma_buffer *bo);
 extern int vmw_stream_claim_ioctl(struct drm_device *dev, void *data,
 				  struct drm_file *file_priv);
 extern int vmw_stream_unref_ioctl(struct drm_device *dev, void *data,
@@ -396,7 +430,30 @@ extern int vmw_user_stream_lookup(struct vmw_private *dev_priv,
 				  struct ttm_object_file *tfile,
 				  uint32_t *inout_id,
 				  struct vmw_resource **out);
+extern void vmw_resource_unreserve(struct list_head *list);
 
+/**
+ * DMA buffer helper routines - vmwgfx_dmabuf.c
+ */
+extern int vmw_dmabuf_to_placement(struct vmw_private *vmw_priv,
+				   struct vmw_dma_buffer *bo,
+				   struct ttm_placement *placement,
+				   bool interruptible);
+extern int vmw_dmabuf_to_vram(struct vmw_private *dev_priv,
+			      struct vmw_dma_buffer *buf,
+			      bool pin, bool interruptible);
+extern int vmw_dmabuf_to_vram_or_gmr(struct vmw_private *dev_priv,
+				     struct vmw_dma_buffer *buf,
+				     bool pin, bool interruptible);
+extern int vmw_dmabuf_to_start_of_vram(struct vmw_private *vmw_priv,
+				       struct vmw_dma_buffer *bo,
+				       bool pin, bool interruptible);
+extern int vmw_dmabuf_unpin(struct vmw_private *vmw_priv,
+			    struct vmw_dma_buffer *bo,
+			    bool interruptible);
+extern void vmw_bo_get_guest_ptr(const struct ttm_buffer_object *buf,
+				 SVGAGuestPtr *ptr);
+extern void vmw_bo_pin(struct ttm_buffer_object *bo, bool pin);
 
 /**
  * Misc Ioctl functionality - vmwgfx_ioctl.c
@@ -406,6 +463,10 @@ extern int vmw_getparam_ioctl(struct drm_device *dev, void *data,
 			      struct drm_file *file_priv);
 extern int vmw_get_cap_3d_ioctl(struct drm_device *dev, void *data,
 				struct drm_file *file_priv);
+extern int vmw_present_ioctl(struct drm_device *dev, void *data,
+			     struct drm_file *file_priv);
+extern int vmw_present_readback_ioctl(struct drm_device *dev, void *data,
+				      struct drm_file *file_priv);
 
 /**
  * Fifo utilities - vmwgfx_fifo.c
@@ -422,6 +483,8 @@ extern int vmw_fifo_send_fence(struct vmw_private *dev_priv,
 extern void vmw_fifo_ping_host(struct vmw_private *dev_priv, uint32_t reason);
 extern bool vmw_fifo_have_3d(struct vmw_private *dev_priv);
 extern bool vmw_fifo_have_pitchlock(struct vmw_private *dev_priv);
+extern int vmw_fifo_emit_dummy_query(struct vmw_private *dev_priv,
+				     uint32_t cid);
 
 /**
  * TTM glue - vmwgfx_ttm_glue.c
@@ -439,7 +502,10 @@ extern struct ttm_placement vmw_vram_placement;
 extern struct ttm_placement vmw_vram_ne_placement;
 extern struct ttm_placement vmw_vram_sys_placement;
 extern struct ttm_placement vmw_vram_gmr_placement;
+extern struct ttm_placement vmw_vram_gmr_ne_placement;
 extern struct ttm_placement vmw_sys_placement;
+extern struct ttm_placement vmw_evictable_placement;
+extern struct ttm_placement vmw_srf_placement;
 extern struct ttm_bo_driver vmw_bo_driver;
 extern int vmw_dma_quiescent(struct drm_device *dev);
 
@@ -449,6 +515,24 @@ extern int vmw_dma_quiescent(struct drm_device *dev);
 
 extern int vmw_execbuf_ioctl(struct drm_device *dev, void *data,
 			     struct drm_file *file_priv);
+extern int vmw_execbuf_process(struct drm_file *file_priv,
+			       struct vmw_private *dev_priv,
+			       void __user *user_commands,
+			       void *kernel_commands,
+			       uint32_t command_size,
+			       uint64_t throttle_us,
+			       struct drm_vmw_fence_rep __user
+			       *user_fence_rep);
+
+extern void
+vmw_execbuf_release_pinned_bo(struct vmw_private *dev_priv,
+			      bool only_on_cid_match, uint32_t cid);
+
+extern int vmw_execbuf_fence_commands(struct drm_file *file_priv,
+				      struct vmw_private *dev_priv,
+				      struct vmw_fence_obj **p_fence,
+				      uint32_t *p_handle);
+
 
 /**
  * IRQs and wating - vmwgfx_irq.c
@@ -520,6 +604,19 @@ bool vmw_kms_validate_mode_vram(struct vmw_private *dev_priv,
 				uint32_t pitch,
 				uint32_t height);
 u32 vmw_get_vblank_counter(struct drm_device *dev, int crtc);
+int vmw_kms_present(struct vmw_private *dev_priv,
+		    struct drm_file *file_priv,
+		    struct vmw_framebuffer *vfb,
+		    struct vmw_surface *surface,
+		    uint32_t sid, int32_t destX, int32_t destY,
+		    struct drm_vmw_rect *clips,
+		    uint32_t num_clips);
+int vmw_kms_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 *clips,
+		     uint32_t num_clips);
 
 /**
  * Overlay control - vmwgfx_overlay.c

drivers/gpu/drm/vmwgfx/vmwgfx_fb.c

@@ -592,58 +592,6 @@ int vmw_fb_close(struct vmw_private *vmw_priv)
 	return 0;
 }
 
-int vmw_dmabuf_from_vram(struct vmw_private *vmw_priv,
-			 struct vmw_dma_buffer *vmw_bo)
-{
-	struct ttm_buffer_object *bo = &vmw_bo->base;
-	int ret = 0;
-
-	ret = ttm_bo_reserve(bo, false, false, false, 0);
-	if (unlikely(ret != 0))
-		return ret;
-
-	ret = ttm_bo_validate(bo, &vmw_sys_placement, false, false, false);
-	ttm_bo_unreserve(bo);
-
-	return ret;
-}
-
-int vmw_dmabuf_to_start_of_vram(struct vmw_private *vmw_priv,
-				struct vmw_dma_buffer *vmw_bo)
-{
-	struct ttm_buffer_object *bo = &vmw_bo->base;
-	struct ttm_placement ne_placement = vmw_vram_ne_placement;
-	int ret = 0;
-
-	ne_placement.lpfn = bo->num_pages;
-
-	/* interuptable? */
-	ret = ttm_write_lock(&vmw_priv->active_master->lock, false);
-	if (unlikely(ret != 0))
-		return ret;
-
-	ret = ttm_bo_reserve(bo, false, false, false, 0);
-	if (unlikely(ret != 0))
-		goto err_unlock;
-
-	if (bo->mem.mem_type == TTM_PL_VRAM &&
-	    bo->mem.start < bo->num_pages &&
-	    bo->mem.start > 0)
-		(void) ttm_bo_validate(bo, &vmw_sys_placement, false,
-				       false, false);
-
-	ret = ttm_bo_validate(bo, &ne_placement, false, false, false);
-
-	/* Could probably bug on */
-	WARN_ON(bo->offset != 0);
-
-	ttm_bo_unreserve(bo);
-err_unlock:
-	ttm_write_unlock(&vmw_priv->active_master->lock);
-
-	return ret;
-}
-
 int vmw_fb_off(struct vmw_private *vmw_priv)
 {
 	struct fb_info *info;
@@ -665,7 +613,7 @@ int vmw_fb_off(struct vmw_private *vmw_priv)
 	par->bo_ptr = NULL;
 	ttm_bo_kunmap(&par->map);
 
-	vmw_dmabuf_from_vram(vmw_priv, par->vmw_bo);
+	vmw_dmabuf_unpin(vmw_priv, par->vmw_bo, false);
 
 	return 0;
 }
@@ -691,7 +639,7 @@ int vmw_fb_on(struct vmw_private *vmw_priv)
 	/* Make sure that all overlays are stoped when we take over */
 	vmw_overlay_stop_all(vmw_priv);
 
-	ret = vmw_dmabuf_to_start_of_vram(vmw_priv, par->vmw_bo);
+	ret = vmw_dmabuf_to_start_of_vram(vmw_priv, par->vmw_bo, true, false);
 	if (unlikely(ret != 0)) {
 		DRM_ERROR("could not move buffer to start of VRAM\n");
 		goto err_no_buffer;

drivers/gpu/drm/vmwgfx/vmwgfx_fence.c

@@ -177,6 +177,9 @@ static int vmw_fence_obj_init(struct vmw_fence_manager *fman,
 
 struct vmw_fence_obj *vmw_fence_obj_reference(struct vmw_fence_obj *fence)
 {
+	if (unlikely(fence == NULL))
+		return NULL;
+
 	kref_get(&fence->kref);
 	return fence;
 }
@@ -191,8 +194,12 @@ struct vmw_fence_obj *vmw_fence_obj_reference(struct vmw_fence_obj *fence)
 void vmw_fence_obj_unreference(struct vmw_fence_obj **fence_p)
 {
 	struct vmw_fence_obj *fence = *fence_p;
-	struct vmw_fence_manager *fman = fence->fman;
+	struct vmw_fence_manager *fman;
+
+	if (unlikely(fence == NULL))
+		return;
 
+	fman = fence->fman;
 	*fence_p = NULL;
 	spin_lock_irq(&fman->lock);
 	BUG_ON(atomic_read(&fence->kref.refcount) == 0);

drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c

@@ -45,7 +45,11 @@ bool vmw_fifo_have_3d(struct vmw_private *dev_priv)
 	if (hwversion == 0)
 		return false;
 
-	if (hwversion < SVGA3D_HWVERSION_WS65_B1)
+	if (hwversion < SVGA3D_HWVERSION_WS8_B1)
+		return false;
+
+	/* Non-Screen Object path does not support surfaces */
+	if (!dev_priv->sou_priv)
 		return false;
 
 	return true;
@@ -277,6 +281,16 @@ static int vmw_fifo_wait(struct vmw_private *dev_priv,
 	return ret;
 }
 
+/**
+ * Reserve @bytes number of bytes in the fifo.
+ *
+ * This function will return NULL (error) on two conditions:
+ *  If it timeouts waiting for fifo space, or if @bytes is larger than the
+ *   available fifo space.
+ *
+ * Returns:
+ *   Pointer to the fifo, or null on error (possible hardware hang).
+ */
 void *vmw_fifo_reserve(struct vmw_private *dev_priv, uint32_t bytes)
 {
 	struct vmw_fifo_state *fifo_state = &dev_priv->fifo;
@@ -491,3 +505,60 @@ int vmw_fifo_send_fence(struct vmw_private *dev_priv, uint32_t *seqno)
 out_err:
 	return ret;
 }
+
+/**
+ * vmw_fifo_emit_dummy_query - emits a dummy query to the fifo.
+ *
+ * @dev_priv: The device private structure.
+ * @cid: The hardware context id used for the query.
+ *
+ * This function is used to emit a dummy occlusion query with
+ * no primitives rendered between query begin and query end.
+ * It's used to provide a query barrier, in order to know that when
+ * this query is finished, all preceding queries are also finished.
+ *
+ * A Query results structure should have been initialized at the start
+ * of the dev_priv->dummy_query_bo buffer object. And that buffer object
+ * must also be either reserved or pinned when this function is called.
+ *
+ * Returns -ENOMEM on failure to reserve fifo space.
+ */
+int vmw_fifo_emit_dummy_query(struct vmw_private *dev_priv,
+			      uint32_t cid)
+{
+	/*
+	 * A query wait without a preceding query end will
+	 * actually finish all queries for this cid
+	 * without writing to the query result structure.
+	 */
+
+	struct ttm_buffer_object *bo = dev_priv->dummy_query_bo;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdWaitForQuery body;
+	} *cmd;
+
+	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
+
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Out of fifo space for dummy query.\n");
+		return -ENOMEM;
+	}
+
+	cmd->header.id = SVGA_3D_CMD_WAIT_FOR_QUERY;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.cid = cid;
+	cmd->body.type = SVGA3D_QUERYTYPE_OCCLUSION;
+
+	if (bo->mem.mem_type == TTM_PL_VRAM) {
+		cmd->body.guestResult.gmrId = SVGA_GMR_FRAMEBUFFER;
+		cmd->body.guestResult.offset = bo->offset;
+	} else {
+		cmd->body.guestResult.gmrId = bo->mem.start;
+		cmd->body.guestResult.offset = 0;
+	}
+
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+
+	return 0;
+}

drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c

@@ -27,6 +27,7 @@
 
 #include "vmwgfx_drv.h"
 #include "vmwgfx_drm.h"
+#include "vmwgfx_kms.h"
 
 int vmw_getparam_ioctl(struct drm_device *dev, void *data,
 		       struct drm_file *file_priv)
@@ -110,3 +111,174 @@ int vmw_get_cap_3d_ioctl(struct drm_device *dev, void *data,
 
 	return ret;
 }
+
+int vmw_present_ioctl(struct drm_device *dev, void *data,
+		      struct drm_file *file_priv)
+{
+	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
+	struct vmw_private *dev_priv = vmw_priv(dev);
+	struct drm_vmw_present_arg *arg =
+		(struct drm_vmw_present_arg *)data;
+	struct vmw_surface *surface;
+	struct vmw_master *vmaster = vmw_master(file_priv->master);
+	struct drm_vmw_rect __user *clips_ptr;
+	struct drm_vmw_rect *clips = NULL;
+	struct drm_mode_object *obj;
+	struct vmw_framebuffer *vfb;
+	uint32_t num_clips;
+	int ret;
+
+	num_clips = arg->num_clips;
+	clips_ptr = (struct drm_vmw_rect *)(unsigned long)arg->clips_ptr;
+
+	if (unlikely(num_clips == 0))
+		return 0;
+
+	if (clips_ptr == NULL) {
+		DRM_ERROR("Variable clips_ptr must be specified.\n");
+		ret = -EINVAL;
+		goto out_clips;
+	}
+
+	clips = kzalloc(num_clips * sizeof(*clips), GFP_KERNEL);
+	if (clips == NULL) {
+		DRM_ERROR("Failed to allocate clip rect list.\n");
+		ret = -ENOMEM;
+		goto out_clips;
+	}
+
+	ret = copy_from_user(clips, clips_ptr, num_clips * sizeof(*clips));
+	if (ret) {
+		DRM_ERROR("Failed to copy clip rects from userspace.\n");
+		goto out_no_copy;
+	}
+
+	ret = mutex_lock_interruptible(&dev->mode_config.mutex);
+	if (unlikely(ret != 0)) {
+		ret = -ERESTARTSYS;
+		goto out_no_mode_mutex;
+	}
+
+	obj = drm_mode_object_find(dev, arg->fb_id, DRM_MODE_OBJECT_FB);
+	if (!obj) {
+		DRM_ERROR("Invalid framebuffer id.\n");
+		ret = -EINVAL;
+		goto out_no_fb;
+	}
+
+	vfb = vmw_framebuffer_to_vfb(obj_to_fb(obj));
+	if (!vfb->dmabuf) {
+		DRM_ERROR("Framebuffer not dmabuf backed.\n");
+		ret = -EINVAL;
+		goto out_no_fb;
+	}
+
+	ret = ttm_read_lock(&vmaster->lock, true);
+	if (unlikely(ret != 0))
+		goto out_no_ttm_lock;
+
+	ret = vmw_user_surface_lookup_handle(dev_priv, tfile, arg->sid,
+					     &surface);
+	if (ret)
+		goto out_no_surface;
+
+	ret = vmw_kms_present(dev_priv, file_priv,
+			      vfb, surface, arg->sid,
+			      arg->dest_x, arg->dest_y,
+			      clips, num_clips);
+
+	/* vmw_user_surface_lookup takes one ref so does new_fb */
+	vmw_surface_unreference(&surface);
+
+out_no_surface:
+	ttm_read_unlock(&vmaster->lock);
+out_no_ttm_lock:
+out_no_fb:
+	mutex_unlock(&dev->mode_config.mutex);
+out_no_mode_mutex:
+out_no_copy:
+	kfree(clips);
+out_clips:
+	return ret;
+}
+
+int vmw_present_readback_ioctl(struct drm_device *dev, void *data,
+			       struct drm_file *file_priv)
+{
+	struct vmw_private *dev_priv = vmw_priv(dev);
+	struct drm_vmw_present_readback_arg *arg =
+		(struct drm_vmw_present_readback_arg *)data;
+	struct drm_vmw_fence_rep __user *user_fence_rep =
+		(struct drm_vmw_fence_rep __user *)
+		(unsigned long)arg->fence_rep;
+	struct vmw_master *vmaster = vmw_master(file_priv->master);
+	struct drm_vmw_rect __user *clips_ptr;
+	struct drm_vmw_rect *clips = NULL;
+	struct drm_mode_object *obj;
+	struct vmw_framebuffer *vfb;
+	uint32_t num_clips;
+	int ret;
+
+	num_clips = arg->num_clips;
+	clips_ptr = (struct drm_vmw_rect *)(unsigned long)arg->clips_ptr;
+
+	if (unlikely(num_clips == 0))
+		return 0;
+
+	if (clips_ptr == NULL) {
+		DRM_ERROR("Argument clips_ptr must be specified.\n");
+		ret = -EINVAL;
+		goto out_clips;
+	}
+
+	clips = kzalloc(num_clips * sizeof(*clips), GFP_KERNEL);
+	if (clips == NULL) {
+		DRM_ERROR("Failed to allocate clip rect list.\n");
+		ret = -ENOMEM;
+		goto out_clips;
+	}
+
+	ret = copy_from_user(clips, clips_ptr, num_clips * sizeof(*clips));
+	if (ret) {
+		DRM_ERROR("Failed to copy clip rects from userspace.\n");
+		goto out_no_copy;
+	}
+
+	ret = mutex_lock_interruptible(&dev->mode_config.mutex);
+	if (unlikely(ret != 0)) {
+		ret = -ERESTARTSYS;
+		goto out_no_mode_mutex;
+	}
+
+	obj = drm_mode_object_find(dev, arg->fb_id, DRM_MODE_OBJECT_FB);
+	if (!obj) {
+		DRM_ERROR("Invalid framebuffer id.\n");
+		ret = -EINVAL;
+		goto out_no_fb;
+	}
+
+	vfb = vmw_framebuffer_to_vfb(obj_to_fb(obj));
+	if (!vfb->dmabuf) {
+		DRM_ERROR("Framebuffer not dmabuf backed.\n");
+		ret = -EINVAL;
+		goto out_no_fb;
+	}
+
+	ret = ttm_read_lock(&vmaster->lock, true);
+	if (unlikely(ret != 0))
+		goto out_no_ttm_lock;
+
+	ret = vmw_kms_readback(dev_priv, file_priv,
+			       vfb, user_fence_rep,
+			       clips, num_clips);
+
+	ttm_read_unlock(&vmaster->lock);
+out_no_ttm_lock:
+out_no_fb:
+	mutex_unlock(&dev->mode_config.mutex);
+out_no_mode_mutex:
+out_no_copy:
+	kfree(clips);
+out_clips:
+	return ret;
+}

drivers/gpu/drm/vmwgfx/vmwgfx_kms.h

@@ -31,6 +31,8 @@
 #include "drmP.h"
 #include "vmwgfx_drv.h"
 
+#define VMWGFX_NUM_DISPLAY_UNITS 8
+
 
 #define vmw_framebuffer_to_vfb(x) \
 	container_of(x, struct vmw_framebuffer, base)
@@ -45,6 +47,9 @@ struct vmw_framebuffer {
 	struct drm_framebuffer base;
 	int (*pin)(struct vmw_framebuffer *fb);
 	int (*unpin)(struct vmw_framebuffer *fb);
+	bool dmabuf;
+	struct ttm_base_object *user_obj;
+	uint32_t user_handle;
 };
 
 
@@ -83,22 +88,59 @@ struct vmw_display_unit {
 	int hotspot_y;
 
 	unsigned unit;
+
+	/*
+	 * Prefered mode tracking.
+	 */
+	unsigned pref_width;
+	unsigned pref_height;
+	bool pref_active;
+	struct drm_display_mode *pref_mode;
 };
 
+#define vmw_crtc_to_du(x) \
+	container_of(x, struct vmw_display_unit, crtc)
+#define vmw_connector_to_du(x) \
+	container_of(x, struct vmw_display_unit, connector)
+
+
 /*
  * Shared display unit functions - vmwgfx_kms.c
  */
 void vmw_display_unit_cleanup(struct vmw_display_unit *du);
+void vmw_du_crtc_save(struct drm_crtc *crtc);
+void vmw_du_crtc_restore(struct drm_crtc *crtc);
+void vmw_du_crtc_gamma_set(struct drm_crtc *crtc,
+			   u16 *r, u16 *g, u16 *b,
+			   uint32_t start, uint32_t size);
 int vmw_du_crtc_cursor_set(struct drm_crtc *crtc, struct drm_file *file_priv,
 			   uint32_t handle, uint32_t width, uint32_t height);
 int vmw_du_crtc_cursor_move(struct drm_crtc *crtc, int x, int y);
+void vmw_du_connector_dpms(struct drm_connector *connector, int mode);
+void vmw_du_connector_save(struct drm_connector *connector);
+void vmw_du_connector_restore(struct drm_connector *connector);
+enum drm_connector_status
+vmw_du_connector_detect(struct drm_connector *connector, bool force);
+int vmw_du_connector_fill_modes(struct drm_connector *connector,
+				uint32_t max_width, uint32_t max_height);
+int vmw_du_connector_set_property(struct drm_connector *connector,
+				  struct drm_property *property,
+				  uint64_t val);
+int vmw_du_update_layout(struct vmw_private *dev_priv, unsigned num,
+			 struct drm_vmw_rect *rects);
 
 /*
  * Legacy display unit functions - vmwgfx_ldu.c
  */
 int vmw_kms_init_legacy_display_system(struct vmw_private *dev_priv);
 int vmw_kms_close_legacy_display_system(struct vmw_private *dev_priv);
-int vmw_kms_ldu_update_layout(struct vmw_private *dev_priv, unsigned num,
+
+/*
+ * Screen Objects display functions - vmwgfx_scrn.c
+ */
+int vmw_kms_init_screen_object_display(struct vmw_private *dev_priv);
+int vmw_kms_close_screen_object_display(struct vmw_private *dev_priv);
+int vmw_kms_sou_update_layout(struct vmw_private *dev_priv, unsigned num,
 			      struct drm_vmw_rect *rects);
 
 #endif

drivers/gpu/drm/vmwgfx/vmwgfx_overlay.c

@@ -86,48 +86,6 @@ static inline void fill_flush(struct vmw_escape_video_flush *cmd,
 	cmd->flush.streamId = stream_id;
 }
 
-/**
- * Pin or unpin a buffer in vram.
- *
- * @dev_priv:  Driver private.
- * @buf:  DMA buffer to pin or unpin.
- * @pin:  Pin buffer in vram if true.
- * @interruptible:  Use interruptible wait.
- *
- * Takes the current masters ttm lock in read.
- *
- * Returns
- * -ERESTARTSYS if interrupted by a signal.
- */
-static int vmw_dmabuf_pin_in_vram(struct vmw_private *dev_priv,
-				  struct vmw_dma_buffer *buf,
-				  bool pin, bool interruptible)
-{
-	struct ttm_buffer_object *bo = &buf->base;
-	struct ttm_placement *overlay_placement = &vmw_vram_placement;
-	int ret;
-
-	ret = ttm_read_lock(&dev_priv->active_master->lock, interruptible);
-	if (unlikely(ret != 0))
-		return ret;
-
-	ret = ttm_bo_reserve(bo, interruptible, false, false, 0);
-	if (unlikely(ret != 0))
-		goto err;
-
-	if (pin)
-		overlay_placement = &vmw_vram_ne_placement;
-
-	ret = ttm_bo_validate(bo, overlay_placement, interruptible, false, false);
-
-	ttm_bo_unreserve(bo);
-
-err:
-	ttm_read_unlock(&dev_priv->active_master->lock);
-
-	return ret;
-}
-
 /**
  * Send put command to hw.
  *
@@ -139,68 +97,80 @@ static int vmw_overlay_send_put(struct vmw_private *dev_priv,
 				struct drm_vmw_control_stream_arg *arg,
 				bool interruptible)
 {
+	struct vmw_escape_video_flush *flush;
+	size_t fifo_size;
+	bool have_so = dev_priv->sou_priv ? true : false;
+	int i, num_items;
+	SVGAGuestPtr ptr;
+
 	struct {
 		struct vmw_escape_header escape;
 		struct {
-			struct {
-				uint32_t cmdType;
-				uint32_t streamId;
-			} header;
-			struct {
-				uint32_t registerId;
-				uint32_t value;
-			} items[SVGA_VIDEO_PITCH_3 + 1];
-		} body;
-		struct vmw_escape_video_flush flush;
+			uint32_t cmdType;
+			uint32_t streamId;
+		} header;
 	} *cmds;
-	uint32_t offset;
-	int i, ret;
+	struct {
+		uint32_t registerId;
+		uint32_t value;
+	} *items;
 
-	for (;;) {
-		cmds = vmw_fifo_reserve(dev_priv, sizeof(*cmds));
-		if (cmds)
-			break;
+	/* defines are a index needs + 1 */
+	if (have_so)
+		num_items = SVGA_VIDEO_DST_SCREEN_ID + 1;
+	else
+		num_items = SVGA_VIDEO_PITCH_3 + 1;
 
-		ret = vmw_fallback_wait(dev_priv, false, true, 0,
-					interruptible, 3*HZ);
-		if (interruptible && ret == -ERESTARTSYS)
-			return ret;
-		else
-			BUG_ON(ret != 0);
+	fifo_size = sizeof(*cmds) + sizeof(*flush) + sizeof(*items) * num_items;
+
+	cmds = vmw_fifo_reserve(dev_priv, fifo_size);
+	/* hardware has hung, can't do anything here */
+	if (!cmds)
+		return -ENOMEM;
+
+	items = (typeof(items))&cmds[1];
+	flush = (struct vmw_escape_video_flush *)&items[num_items];
+
+	/* the size is header + number of items */
+	fill_escape(&cmds->escape, sizeof(*items) * (num_items + 1));
+
+	cmds->header.cmdType = SVGA_ESCAPE_VMWARE_VIDEO_SET_REGS;
+	cmds->header.streamId = arg->stream_id;
+
+	/* the IDs are neatly numbered */
+	for (i = 0; i < num_items; i++)
+		items[i].registerId = i;
+
+	vmw_bo_get_guest_ptr(&buf->base, &ptr);
+	ptr.offset += arg->offset;
+
+	items[SVGA_VIDEO_ENABLED].value     = true;
+	items[SVGA_VIDEO_FLAGS].value       = arg->flags;
+	items[SVGA_VIDEO_DATA_OFFSET].value = ptr.offset;
+	items[SVGA_VIDEO_FORMAT].value      = arg->format;
+	items[SVGA_VIDEO_COLORKEY].value    = arg->color_key;
+	items[SVGA_VIDEO_SIZE].value        = arg->size;
+	items[SVGA_VIDEO_WIDTH].value       = arg->width;
+	items[SVGA_VIDEO_HEIGHT].value      = arg->height;
+	items[SVGA_VIDEO_SRC_X].value       = arg->src.x;
+	items[SVGA_VIDEO_SRC_Y].value       = arg->src.y;
+	items[SVGA_VIDEO_SRC_WIDTH].value   = arg->src.w;
+	items[SVGA_VIDEO_SRC_HEIGHT].value  = arg->src.h;
+	items[SVGA_VIDEO_DST_X].value       = arg->dst.x;
+	items[SVGA_VIDEO_DST_Y].value       = arg->dst.y;
+	items[SVGA_VIDEO_DST_WIDTH].value   = arg->dst.w;
+	items[SVGA_VIDEO_DST_HEIGHT].value  = arg->dst.h;
+	items[SVGA_VIDEO_PITCH_1].value     = arg->pitch[0];
+	items[SVGA_VIDEO_PITCH_2].value     = arg->pitch[1];
+	items[SVGA_VIDEO_PITCH_3].value     = arg->pitch[2];
+	if (have_so) {
+		items[SVGA_VIDEO_DATA_GMRID].value    = ptr.gmrId;
+		items[SVGA_VIDEO_DST_SCREEN_ID].value = SVGA_ID_INVALID;
 	}
 
-	fill_escape(&cmds->escape, sizeof(cmds->body));
-	cmds->body.header.cmdType = SVGA_ESCAPE_VMWARE_VIDEO_SET_REGS;
-	cmds->body.header.streamId = arg->stream_id;
-
-	for (i = 0; i <= SVGA_VIDEO_PITCH_3; i++)
-		cmds->body.items[i].registerId = i;
-
-	offset = buf->base.offset + arg->offset;
-
-	cmds->body.items[SVGA_VIDEO_ENABLED].value     = true;
-	cmds->body.items[SVGA_VIDEO_FLAGS].value       = arg->flags;
-	cmds->body.items[SVGA_VIDEO_DATA_OFFSET].value = offset;
-	cmds->body.items[SVGA_VIDEO_FORMAT].value      = arg->format;
-	cmds->body.items[SVGA_VIDEO_COLORKEY].value    = arg->color_key;
-	cmds->body.items[SVGA_VIDEO_SIZE].value        = arg->size;
-	cmds->body.items[SVGA_VIDEO_WIDTH].value       = arg->width;
-	cmds->body.items[SVGA_VIDEO_HEIGHT].value      = arg->height;
-	cmds->body.items[SVGA_VIDEO_SRC_X].value       = arg->src.x;
-	cmds->body.items[SVGA_VIDEO_SRC_Y].value       = arg->src.y;
-	cmds->body.items[SVGA_VIDEO_SRC_WIDTH].value   = arg->src.w;
-	cmds->body.items[SVGA_VIDEO_SRC_HEIGHT].value  = arg->src.h;
-	cmds->body.items[SVGA_VIDEO_DST_X].value       = arg->dst.x;
-	cmds->body.items[SVGA_VIDEO_DST_Y].value       = arg->dst.y;
-	cmds->body.items[SVGA_VIDEO_DST_WIDTH].value   = arg->dst.w;
-	cmds->body.items[SVGA_VIDEO_DST_HEIGHT].value  = arg->dst.h;
-	cmds->body.items[SVGA_VIDEO_PITCH_1].value     = arg->pitch[0];
-	cmds->body.items[SVGA_VIDEO_PITCH_2].value     = arg->pitch[1];
-	cmds->body.items[SVGA_VIDEO_PITCH_3].value     = arg->pitch[2];
-
-	fill_flush(&cmds->flush, arg->stream_id);
+	fill_flush(flush, arg->stream_id);
 
-	vmw_fifo_commit(dev_priv, sizeof(*cmds));
+	vmw_fifo_commit(dev_priv, fifo_size);
 
 	return 0;
 }
@@ -247,6 +217,25 @@ static int vmw_overlay_send_stop(struct vmw_private *dev_priv,
 	return 0;
 }
 
+/**
+ * Move a buffer to vram or gmr if @pin is set, else unpin the buffer.
+ *
+ * With the introduction of screen objects buffers could now be
+ * used with GMRs instead of being locked to vram.
+ */
+static int vmw_overlay_move_buffer(struct vmw_private *dev_priv,
+				   struct vmw_dma_buffer *buf,
+				   bool pin, bool inter)
+{
+	if (!pin)
+		return vmw_dmabuf_unpin(dev_priv, buf, inter);
+
+	if (!dev_priv->sou_priv)
+		return vmw_dmabuf_to_vram(dev_priv, buf, true, inter);
+
+	return vmw_dmabuf_to_vram_or_gmr(dev_priv, buf, true, inter);
+}
+
 /**
  * Stop or pause a stream.
  *
@@ -279,8 +268,8 @@ static int vmw_overlay_stop(struct vmw_private *dev_priv,
 			return ret;
 
 		/* We just remove the NO_EVICT flag so no -ENOMEM */
-		ret = vmw_dmabuf_pin_in_vram(dev_priv, stream->buf, false,
-					     interruptible);
+		ret = vmw_overlay_move_buffer(dev_priv, stream->buf, false,
+					      interruptible);
 		if (interruptible && ret == -ERESTARTSYS)
 			return ret;
 		else
@@ -342,7 +331,7 @@ static int vmw_overlay_update_stream(struct vmw_private *dev_priv,
 	/* We don't start the old stream if we are interrupted.
 	 * Might return -ENOMEM if it can't fit the buffer in vram.
 	 */
-	ret = vmw_dmabuf_pin_in_vram(dev_priv, buf, true, interruptible);
+	ret = vmw_overlay_move_buffer(dev_priv, buf, true, interruptible);
 	if (ret)
 		return ret;
 
@@ -351,7 +340,8 @@ static int vmw_overlay_update_stream(struct vmw_private *dev_priv,
 		/* This one needs to happen no matter what. We only remove
 		 * the NO_EVICT flag so this is safe from -ENOMEM.
 		 */
-		BUG_ON(vmw_dmabuf_pin_in_vram(dev_priv, buf, false, false) != 0);
+		BUG_ON(vmw_overlay_move_buffer(dev_priv, buf, false, false)
+		       != 0);
 		return ret;
 	}
 

include/drm/vmwgfx_drm.h

@@ -52,6 +52,8 @@
 #define DRM_VMW_FENCE_SIGNALED       15
 #define DRM_VMW_FENCE_UNREF          16
 #define DRM_VMW_FENCE_EVENT          17
+#define DRM_VMW_PRESENT              18
+#define DRM_VMW_PRESENT_READBACK     19
 
 
 /*************************************************************************/
@@ -681,5 +683,66 @@ struct drm_vmw_fence_arg {
 };
 
 
+/*************************************************************************/
+/**
+ * DRM_VMW_PRESENT
+ *
+ * Executes an SVGA present on a given fb for a given surface. The surface
+ * is placed on the framebuffer. Cliprects are given relative to the given
+ * point (the point disignated by dest_{x|y}).
+ *
+ */
+
+/**
+ * struct drm_vmw_present_arg
+ * @fb_id: framebuffer id to present / read back from.
+ * @sid: Surface id to present from.
+ * @dest_x: X placement coordinate for surface.
+ * @dest_y: Y placement coordinate for surface.
+ * @clips_ptr: Pointer to an array of clip rects cast to an uint64_t.
+ * @num_clips: Number of cliprects given relative to the framebuffer origin,
+ * in the same coordinate space as the frame buffer.
+ * @pad64: Unused 64-bit padding.
+ *
+ * Input argument to the DRM_VMW_PRESENT ioctl.
+ */
+
+struct drm_vmw_present_arg {
+	uint32_t fb_id;
+	uint32_t sid;
+	int32_t dest_x;
+	int32_t dest_y;
+	uint64_t clips_ptr;
+	uint32_t num_clips;
+	uint32_t pad64;
+};
+
+
+/*************************************************************************/
+/**
+ * DRM_VMW_PRESENT_READBACK
+ *
+ * Executes an SVGA present readback from a given fb to the dma buffer
+ * currently bound as the fb. If there is no dma buffer bound to the fb,
+ * an error will be returned.
+ *
+ */
+
+/**
+ * struct drm_vmw_present_arg
+ * @fb_id: fb_id to present / read back from.
+ * @num_clips: Number of cliprects.
+ * @clips_ptr: Pointer to an array of clip rects cast to an uint64_t.
+ * @fence_rep: Pointer to a struct drm_vmw_fence_rep, cast to an uint64_t.
+ * If this member is NULL, then the ioctl should not return a fence.
+ */
+
+struct drm_vmw_present_readback_arg {
+	 uint32_t fb_id;
+	 uint32_t num_clips;
+	 uint64_t clips_ptr;
+	 uint64_t fence_rep;
+};
+
 
 #endif