Merge tag 'vmwgfx-next-2014-01-17' of git://people.freedesktop.org/~thomash/linux into drm-next

Pull request of 2014-01-17

Pull request for 3.14. One not so urgent fix, One huge device update.

The pull request corresponds to the patches sent out on dri-devel, except:
[PATCH 02/33], review tag typo pointed out by Matt Turner.
[PATCH 04/33], dropped. The new surface formats are never used.

The upcoming vmware svga2 hardware version 11 will introduce the concept
of "guest backed objects" or -resources. The device will in principle
get all
of its memory from the guest, which has big advantages from the device
point of view.

This means that vmwgfx contexts, shaders and surfaces need to be backed
by guest memory in the form of buffer objects called MOBs, presumably
short for MemoryOBjects, which are bound to the device in a special way.

This patch series introduces guest backed object support. Some new IOCTLs
are added to allocate these new guest backed object, and to optionally
provide
them with a backing MOB.

There is an update to the gallium driver that comes with this update, and
it will be pushed in the near timeframe presumably to a separate mesa branch
before merged to master.

* tag 'vmwgfx-next-2014-01-17' of git://people.freedesktop.org/~thomash/linux: (33 commits)
  drm/vmwgfx: Invalidate surface on non-readback unbind
  drm/vmwgfx: Silence the device command verifier
  drm/vmwgfx: Implement 64-bit Otable- and MOB binding v2
  drm/vmwgfx: Fix surface framebuffer check for guest-backed surfaces
  drm/vmwgfx: Update otable definitions
  drm/vmwgfx: Use the linux DMA api also for MOBs
  drm/vmwgfx: Ditch the vmw_dummy_query_bo_prepare function
  drm/vmwgfx: Persistent tracking of context bindings
  drm/vmwgfx: Track context bindings and scrub them upon exiting execbuf
  drm/vmwgfx: Block the BIND_SHADERCONSTS command
  drm/vmwgfx: Add a parameter to get max MOB memory size
  drm/vmwgfx: Implement a buffer object synccpu ioctl.
  drm/vmwgfx: Make sure that the multisampling is off
  drm/vmwgfx: Extend the command verifier to handle guest-backed on / off
  drm/vmwgfx: Fix up the vmwgfx_drv.h header for new files
  drm/vmwgfx: Enable 3D for new hardware version
  drm/vmwgfx: Add new unused (by user-space) commands to the verifier
  drm/vmwgfx: Validate guest-backed shader const commands
  drm/vmwgfx: Add guest-backed shaders
  drm/vmwgfx: Hook up guest-backed surfaces
  ...

drivers/gpu/drm/vmwgfx/Makefile

@@ -6,6 +6,6 @@ vmwgfx-y := vmwgfx_execbuf.o vmwgfx_gmr.o vmwgfx_kms.o vmwgfx_drv.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_dmabuf.o vmwgfx_scrn.o vmwgfx_context.o \
-	    vmwgfx_surface.o vmwgfx_prime.o
+	    vmwgfx_surface.o vmwgfx_prime.o vmwgfx_mob.o vmwgfx_shader.o
 
 obj-$(CONFIG_DRM_VMWGFX) := vmwgfx.o

drivers/gpu/drm/vmwgfx/svga3d_reg.h

@@ -34,6 +34,8 @@
 
 #include "svga_reg.h"
 
+typedef uint32 PPN;
+typedef __le64 PPN64;
 
 /*
  * 3D Hardware Version
@@ -71,6 +73,9 @@ typedef uint32 SVGA3dBool; /* 32-bit Bool definition */
 #define SVGA3D_MAX_CONTEXT_IDS                  256
 #define SVGA3D_MAX_SURFACE_IDS                  (32 * 1024)
 
+#define SVGA3D_NUM_TEXTURE_UNITS                32
+#define SVGA3D_NUM_LIGHTS                       8
+
 /*
  * Surface formats.
  *
@@ -81,6 +86,7 @@ typedef uint32 SVGA3dBool; /* 32-bit Bool definition */
  */
 
 typedef enum SVGA3dSurfaceFormat {
+   SVGA3D_FORMAT_MIN                   = 0,
    SVGA3D_FORMAT_INVALID               = 0,
 
    SVGA3D_X8R8G8B8                     = 1,
@@ -134,12 +140,6 @@ typedef enum SVGA3dSurfaceFormat {
    SVGA3D_RG_S10E5                     = 35,
    SVGA3D_RG_S23E8                     = 36,
 
-   /*
-    * Any surface can be used as a buffer object, but SVGA3D_BUFFER is
-    * the most efficient format to use when creating new surfaces
-    * expressly for index or vertex data.
-    */
-
    SVGA3D_BUFFER                       = 37,
 
    SVGA3D_Z_D24X8                      = 38,
@@ -159,15 +159,114 @@ typedef enum SVGA3dSurfaceFormat {
    /* Video format with alpha */
    SVGA3D_AYUV                         = 45,
 
+   SVGA3D_R32G32B32A32_TYPELESS        = 46,
+   SVGA3D_R32G32B32A32_FLOAT           = 25,
+   SVGA3D_R32G32B32A32_UINT            = 47,
+   SVGA3D_R32G32B32A32_SINT            = 48,
+   SVGA3D_R32G32B32_TYPELESS           = 49,
+   SVGA3D_R32G32B32_FLOAT              = 50,
+   SVGA3D_R32G32B32_UINT               = 51,
+   SVGA3D_R32G32B32_SINT               = 52,
+   SVGA3D_R16G16B16A16_TYPELESS        = 53,
+   SVGA3D_R16G16B16A16_FLOAT           = 24,
+   SVGA3D_R16G16B16A16_UNORM           = 41,
+   SVGA3D_R16G16B16A16_UINT            = 54,
+   SVGA3D_R16G16B16A16_SNORM           = 55,
+   SVGA3D_R16G16B16A16_SINT            = 56,
+   SVGA3D_R32G32_TYPELESS              = 57,
+   SVGA3D_R32G32_FLOAT                 = 36,
+   SVGA3D_R32G32_UINT                  = 58,
+   SVGA3D_R32G32_SINT                  = 59,
+   SVGA3D_R32G8X24_TYPELESS            = 60,
+   SVGA3D_D32_FLOAT_S8X24_UINT         = 61,
+   SVGA3D_R32_FLOAT_X8X24_TYPELESS     = 62,
+   SVGA3D_X32_TYPELESS_G8X24_UINT      = 63,
+   SVGA3D_R10G10B10A2_TYPELESS         = 64,
+   SVGA3D_R10G10B10A2_UNORM            = 26,
+   SVGA3D_R10G10B10A2_UINT             = 65,
+   SVGA3D_R11G11B10_FLOAT              = 66,
+   SVGA3D_R8G8B8A8_TYPELESS            = 67,
+   SVGA3D_R8G8B8A8_UNORM               = 68,
+   SVGA3D_R8G8B8A8_UNORM_SRGB          = 69,
+   SVGA3D_R8G8B8A8_UINT                = 70,
+   SVGA3D_R8G8B8A8_SNORM               = 28,
+   SVGA3D_R8G8B8A8_SINT                = 71,
+   SVGA3D_R16G16_TYPELESS              = 72,
+   SVGA3D_R16G16_FLOAT                 = 35,
+   SVGA3D_R16G16_UNORM                 = 40,
+   SVGA3D_R16G16_UINT                  = 73,
+   SVGA3D_R16G16_SNORM                 = 39,
+   SVGA3D_R16G16_SINT                  = 74,
+   SVGA3D_R32_TYPELESS                 = 75,
+   SVGA3D_D32_FLOAT                    = 76,
+   SVGA3D_R32_FLOAT                    = 34,
+   SVGA3D_R32_UINT                     = 77,
+   SVGA3D_R32_SINT                     = 78,
+   SVGA3D_R24G8_TYPELESS               = 79,
+   SVGA3D_D24_UNORM_S8_UINT            = 80,
+   SVGA3D_R24_UNORM_X8_TYPELESS        = 81,
+   SVGA3D_X24_TYPELESS_G8_UINT         = 82,
+   SVGA3D_R8G8_TYPELESS                = 83,
+   SVGA3D_R8G8_UNORM                   = 84,
+   SVGA3D_R8G8_UINT                    = 85,
+   SVGA3D_R8G8_SNORM                   = 27,
+   SVGA3D_R8G8_SINT                    = 86,
+   SVGA3D_R16_TYPELESS                 = 87,
+   SVGA3D_R16_FLOAT                    = 33,
+   SVGA3D_D16_UNORM                    = 8,
+   SVGA3D_R16_UNORM                    = 88,
+   SVGA3D_R16_UINT                     = 89,
+   SVGA3D_R16_SNORM                    = 90,
+   SVGA3D_R16_SINT                     = 91,
+   SVGA3D_R8_TYPELESS                  = 92,
+   SVGA3D_R8_UNORM                     = 93,
+   SVGA3D_R8_UINT                      = 94,
+   SVGA3D_R8_SNORM                     = 95,
+   SVGA3D_R8_SINT                      = 96,
+   SVGA3D_A8_UNORM                     = 32,
+   SVGA3D_R1_UNORM                     = 97,
+   SVGA3D_R9G9B9E5_SHAREDEXP           = 98,
+   SVGA3D_R8G8_B8G8_UNORM              = 99,
+   SVGA3D_G8R8_G8B8_UNORM              = 100,
+   SVGA3D_BC1_TYPELESS                 = 101,
+   SVGA3D_BC1_UNORM                    = 15,
+   SVGA3D_BC1_UNORM_SRGB               = 102,
+   SVGA3D_BC2_TYPELESS                 = 103,
+   SVGA3D_BC2_UNORM                    = 17,
+   SVGA3D_BC2_UNORM_SRGB               = 104,
+   SVGA3D_BC3_TYPELESS                 = 105,
+   SVGA3D_BC3_UNORM                    = 19,
+   SVGA3D_BC3_UNORM_SRGB               = 106,
+   SVGA3D_BC4_TYPELESS                 = 107,
    SVGA3D_BC4_UNORM                    = 108,
+   SVGA3D_BC4_SNORM                    = 109,
+   SVGA3D_BC5_TYPELESS                 = 110,
    SVGA3D_BC5_UNORM                    = 111,
+   SVGA3D_BC5_SNORM                    = 112,
+   SVGA3D_B5G6R5_UNORM                 = 3,
+   SVGA3D_B5G5R5A1_UNORM               = 5,
+   SVGA3D_B8G8R8A8_UNORM               = 2,
+   SVGA3D_B8G8R8X8_UNORM               = 1,
+   SVGA3D_R10G10B10_XR_BIAS_A2_UNORM   = 113,
+   SVGA3D_B8G8R8A8_TYPELESS            = 114,
+   SVGA3D_B8G8R8A8_UNORM_SRGB          = 115,
+   SVGA3D_B8G8R8X8_TYPELESS            = 116,
+   SVGA3D_B8G8R8X8_UNORM_SRGB          = 117,
 
    /* Advanced D3D9 depth formats. */
    SVGA3D_Z_DF16                       = 118,
    SVGA3D_Z_DF24                       = 119,
    SVGA3D_Z_D24S8_INT                  = 120,
 
-   SVGA3D_FORMAT_MAX
+   /* Planar video formats. */
+   SVGA3D_YV12                         = 121,
+
+   /* Shader constant formats. */
+   SVGA3D_SURFACE_SHADERCONST_FLOAT    = 122,
+   SVGA3D_SURFACE_SHADERCONST_INT      = 123,
+   SVGA3D_SURFACE_SHADERCONST_BOOL     = 124,
+
+   SVGA3D_FORMAT_MAX                   = 125,
 } SVGA3dSurfaceFormat;
 
 typedef uint32 SVGA3dColor; /* a, r, g, b */
@@ -957,15 +1056,21 @@ typedef enum {
 } SVGA3dCubeFace;
 
 typedef enum {
+   SVGA3D_SHADERTYPE_INVALID                    = 0,
+   SVGA3D_SHADERTYPE_MIN                        = 1,
    SVGA3D_SHADERTYPE_VS                         = 1,
    SVGA3D_SHADERTYPE_PS                         = 2,
-   SVGA3D_SHADERTYPE_MAX
+   SVGA3D_SHADERTYPE_MAX                        = 3,
+   SVGA3D_SHADERTYPE_GS                         = 3,
 } SVGA3dShaderType;
 
+#define SVGA3D_NUM_SHADERTYPE (SVGA3D_SHADERTYPE_MAX - SVGA3D_SHADERTYPE_MIN)
+
 typedef enum {
    SVGA3D_CONST_TYPE_FLOAT                      = 0,
    SVGA3D_CONST_TYPE_INT                        = 1,
    SVGA3D_CONST_TYPE_BOOL                       = 2,
+   SVGA3D_CONST_TYPE_MAX
 } SVGA3dShaderConstType;
 
 #define SVGA3D_MAX_SURFACE_FACES                6
@@ -1056,9 +1161,74 @@ typedef enum {
 #define SVGA_3D_CMD_GENERATE_MIPMAPS       SVGA_3D_CMD_BASE + 31
 #define SVGA_3D_CMD_ACTIVATE_SURFACE       SVGA_3D_CMD_BASE + 40
 #define SVGA_3D_CMD_DEACTIVATE_SURFACE     SVGA_3D_CMD_BASE + 41
-#define SVGA_3D_CMD_MAX                    SVGA_3D_CMD_BASE + 42
-
-#define SVGA_3D_CMD_FUTURE_MAX             2000
+#define SVGA_3D_CMD_SCREEN_DMA               1082
+#define SVGA_3D_CMD_SET_UNITY_SURFACE_COOKIE 1083
+#define SVGA_3D_CMD_OPEN_CONTEXT_SURFACE     1084
+
+#define SVGA_3D_CMD_LOGICOPS_BITBLT          1085
+#define SVGA_3D_CMD_LOGICOPS_TRANSBLT        1086
+#define SVGA_3D_CMD_LOGICOPS_STRETCHBLT      1087
+#define SVGA_3D_CMD_LOGICOPS_COLORFILL       1088
+#define SVGA_3D_CMD_LOGICOPS_ALPHABLEND      1089
+#define SVGA_3D_CMD_LOGICOPS_CLEARTYPEBLEND  1090
+
+#define SVGA_3D_CMD_SET_OTABLE_BASE          1091
+#define SVGA_3D_CMD_READBACK_OTABLE          1092
+
+#define SVGA_3D_CMD_DEFINE_GB_MOB            1093
+#define SVGA_3D_CMD_DESTROY_GB_MOB           1094
+#define SVGA_3D_CMD_REDEFINE_GB_MOB          1095
+#define SVGA_3D_CMD_UPDATE_GB_MOB_MAPPING    1096
+
+#define SVGA_3D_CMD_DEFINE_GB_SURFACE        1097
+#define SVGA_3D_CMD_DESTROY_GB_SURFACE       1098
+#define SVGA_3D_CMD_BIND_GB_SURFACE          1099
+#define SVGA_3D_CMD_COND_BIND_GB_SURFACE     1100
+#define SVGA_3D_CMD_UPDATE_GB_IMAGE          1101
+#define SVGA_3D_CMD_UPDATE_GB_SURFACE        1102
+#define SVGA_3D_CMD_READBACK_GB_IMAGE        1103
+#define SVGA_3D_CMD_READBACK_GB_SURFACE      1104
+#define SVGA_3D_CMD_INVALIDATE_GB_IMAGE      1105
+#define SVGA_3D_CMD_INVALIDATE_GB_SURFACE    1106
+
+#define SVGA_3D_CMD_DEFINE_GB_CONTEXT        1107
+#define SVGA_3D_CMD_DESTROY_GB_CONTEXT       1108
+#define SVGA_3D_CMD_BIND_GB_CONTEXT          1109
+#define SVGA_3D_CMD_READBACK_GB_CONTEXT      1110
+#define SVGA_3D_CMD_INVALIDATE_GB_CONTEXT    1111
+
+#define SVGA_3D_CMD_DEFINE_GB_SHADER         1112
+#define SVGA_3D_CMD_DESTROY_GB_SHADER        1113
+#define SVGA_3D_CMD_BIND_GB_SHADER           1114
+
+#define SVGA_3D_CMD_SET_OTABLE_BASE64        1115
+
+#define SVGA_3D_CMD_BEGIN_GB_QUERY           1116
+#define SVGA_3D_CMD_END_GB_QUERY             1117
+#define SVGA_3D_CMD_WAIT_FOR_GB_QUERY        1118
+
+#define SVGA_3D_CMD_NOP                      1119
+
+#define SVGA_3D_CMD_ENABLE_GART              1120
+#define SVGA_3D_CMD_DISABLE_GART             1121
+#define SVGA_3D_CMD_MAP_MOB_INTO_GART        1122
+#define SVGA_3D_CMD_UNMAP_GART_RANGE         1123
+
+#define SVGA_3D_CMD_DEFINE_GB_SCREENTARGET   1124
+#define SVGA_3D_CMD_DESTROY_GB_SCREENTARGET  1125
+#define SVGA_3D_CMD_BIND_GB_SCREENTARGET     1126
+#define SVGA_3D_CMD_UPDATE_GB_SCREENTARGET   1127
+
+#define SVGA_3D_CMD_READBACK_GB_IMAGE_PARTIAL   1128
+#define SVGA_3D_CMD_INVALIDATE_GB_IMAGE_PARTIAL 1129
+
+#define SVGA_3D_CMD_SET_GB_SHADERCONSTS_INLINE  1130
+
+#define SVGA_3D_CMD_DEFINE_GB_MOB64          1135
+#define SVGA_3D_CMD_REDEFINE_GB_MOB64        1136
+
+#define SVGA_3D_CMD_MAX                      1142
+#define SVGA_3D_CMD_FUTURE_MAX               3000
 
 /*
  * Common substructures used in multiple FIFO commands:
@@ -1749,6 +1919,495 @@ struct {
 } SVGA3dCmdGenerateMipmaps;             /* SVGA_3D_CMD_GENERATE_MIPMAPS */
 
 
+/*
+ * Guest-backed surface definitions.
+ */
+
+typedef uint32 SVGAMobId;
+
+typedef enum SVGAMobFormat {
+   SVGA3D_MOBFMT_INVALID = SVGA3D_INVALID_ID,
+   SVGA3D_MOBFMT_PTDEPTH_0 = 0,
+   SVGA3D_MOBFMT_PTDEPTH_1 = 1,
+   SVGA3D_MOBFMT_PTDEPTH_2 = 2,
+   SVGA3D_MOBFMT_RANGE     = 3,
+   SVGA3D_MOBFMT_PTDEPTH64_0 = 4,
+   SVGA3D_MOBFMT_PTDEPTH64_1 = 5,
+   SVGA3D_MOBFMT_PTDEPTH64_2 = 6,
+   SVGA3D_MOBFMT_MAX,
+} SVGAMobFormat;
+
+/*
+ * Sizes of opaque types.
+ */
+
+#define SVGA3D_OTABLE_MOB_ENTRY_SIZE 16
+#define SVGA3D_OTABLE_CONTEXT_ENTRY_SIZE 8
+#define SVGA3D_OTABLE_SURFACE_ENTRY_SIZE 64
+#define SVGA3D_OTABLE_SHADER_ENTRY_SIZE 16
+#define SVGA3D_OTABLE_SCREEN_TARGET_ENTRY_SIZE 64
+#define SVGA3D_CONTEXT_DATA_SIZE 16384
+
+/*
+ * SVGA3dCmdSetOTableBase --
+ *
+ * This command allows the guest to specify the base PPN of the
+ * specified object table.
+ */
+
+typedef enum {
+   SVGA_OTABLE_MOB           = 0,
+   SVGA_OTABLE_MIN           = 0,
+   SVGA_OTABLE_SURFACE       = 1,
+   SVGA_OTABLE_CONTEXT       = 2,
+   SVGA_OTABLE_SHADER        = 3,
+   SVGA_OTABLE_SCREEN_TARGET = 4,
+   SVGA_OTABLE_DX9_MAX       = 5,
+   SVGA_OTABLE_MAX           = 8
+} SVGAOTableType;
+
+typedef
+struct {
+   SVGAOTableType type;
+   PPN baseAddress;
+   uint32 sizeInBytes;
+   uint32 validSizeInBytes;
+   SVGAMobFormat ptDepth;
+}
+__attribute__((__packed__))
+SVGA3dCmdSetOTableBase;  /* SVGA_3D_CMD_SET_OTABLE_BASE */
+
+typedef
+struct {
+   SVGAOTableType type;
+   PPN64 baseAddress;
+   uint32 sizeInBytes;
+   uint32 validSizeInBytes;
+   SVGAMobFormat ptDepth;
+}
+__attribute__((__packed__))
+SVGA3dCmdSetOTableBase64;  /* SVGA_3D_CMD_SET_OTABLE_BASE64 */
+
+typedef
+struct {
+   SVGAOTableType type;
+}
+__attribute__((__packed__))
+SVGA3dCmdReadbackOTable;  /* SVGA_3D_CMD_READBACK_OTABLE */
+
+/*
+ * Define a memory object (Mob) in the OTable.
+ */
+
+typedef
+struct SVGA3dCmdDefineGBMob {
+   SVGAMobId mobid;
+   SVGAMobFormat ptDepth;
+   PPN base;
+   uint32 sizeInBytes;
+}
+__attribute__((__packed__))
+SVGA3dCmdDefineGBMob;   /* SVGA_3D_CMD_DEFINE_GB_MOB */
+
+
+/*
+ * Destroys an object in the OTable.
+ */
+
+typedef
+struct SVGA3dCmdDestroyGBMob {
+   SVGAMobId mobid;
+}
+__attribute__((__packed__))
+SVGA3dCmdDestroyGBMob;   /* SVGA_3D_CMD_DESTROY_GB_MOB */
+
+/*
+ * Redefine an object in the OTable.
+ */
+
+typedef
+struct SVGA3dCmdRedefineGBMob {
+   SVGAMobId mobid;
+   SVGAMobFormat ptDepth;
+   PPN base;
+   uint32 sizeInBytes;
+}
+__attribute__((__packed__))
+SVGA3dCmdRedefineGBMob;   /* SVGA_3D_CMD_REDEFINE_GB_MOB */
+
+/*
+ * Define a memory object (Mob) in the OTable with a PPN64 base.
+ */
+
+typedef
+struct SVGA3dCmdDefineGBMob64 {
+   SVGAMobId mobid;
+   SVGAMobFormat ptDepth;
+   PPN64 base;
+   uint32 sizeInBytes;
+}
+__attribute__((__packed__))
+SVGA3dCmdDefineGBMob64;   /* SVGA_3D_CMD_DEFINE_GB_MOB64 */
+
+/*
+ * Redefine an object in the OTable with PPN64 base.
+ */
+
+typedef
+struct SVGA3dCmdRedefineGBMob64 {
+   SVGAMobId mobid;
+   SVGAMobFormat ptDepth;
+   PPN64 base;
+   uint32 sizeInBytes;
+}
+__attribute__((__packed__))
+SVGA3dCmdRedefineGBMob64;   /* SVGA_3D_CMD_REDEFINE_GB_MOB64 */
+
+/*
+ * Notification that the page tables have been modified.
+ */
+
+typedef
+struct SVGA3dCmdUpdateGBMobMapping {
+   SVGAMobId mobid;
+}
+__attribute__((__packed__))
+SVGA3dCmdUpdateGBMobMapping;   /* SVGA_3D_CMD_UPDATE_GB_MOB_MAPPING */
+
+/*
+ * Define a guest-backed surface.
+ */
+
+typedef
+struct SVGA3dCmdDefineGBSurface {
+   uint32 sid;
+   SVGA3dSurfaceFlags surfaceFlags;
+   SVGA3dSurfaceFormat format;
+   uint32 numMipLevels;
+   uint32 multisampleCount;
+   SVGA3dTextureFilter autogenFilter;
+   SVGA3dSize size;
+} SVGA3dCmdDefineGBSurface;   /* SVGA_3D_CMD_DEFINE_GB_SURFACE */
+
+/*
+ * Destroy a guest-backed surface.
+ */
+
+typedef
+struct SVGA3dCmdDestroyGBSurface {
+   uint32 sid;
+} SVGA3dCmdDestroyGBSurface;   /* SVGA_3D_CMD_DESTROY_GB_SURFACE */
+
+/*
+ * Bind a guest-backed surface to an object.
+ */
+
+typedef
+struct SVGA3dCmdBindGBSurface {
+   uint32 sid;
+   SVGAMobId mobid;
+} SVGA3dCmdBindGBSurface;   /* SVGA_3D_CMD_BIND_GB_SURFACE */
+
+/*
+ * Conditionally bind a mob to a guest backed surface if testMobid
+ * matches the currently bound mob.  Optionally issue a readback on
+ * the surface while it is still bound to the old mobid if the mobid
+ * is changed by this command.
+ */
+
+#define SVGA3D_COND_BIND_GB_SURFACE_FLAG_READBACK (1 << 0)
+
+typedef
+struct{
+   uint32 sid;
+   SVGAMobId testMobid;
+   SVGAMobId mobid;
+   uint32 flags;
+}
+SVGA3dCmdCondBindGBSurface;          /* SVGA_3D_CMD_COND_BIND_GB_SURFACE */
+
+/*
+ * Update an image in a guest-backed surface.
+ * (Inform the device that the guest-contents have been updated.)
+ */
+
+typedef
+struct SVGA3dCmdUpdateGBImage {
+   SVGA3dSurfaceImageId image;
+   SVGA3dBox box;
+} SVGA3dCmdUpdateGBImage;   /* SVGA_3D_CMD_UPDATE_GB_IMAGE */
+
+/*
+ * Update an entire guest-backed surface.
+ * (Inform the device that the guest-contents have been updated.)
+ */
+
+typedef
+struct SVGA3dCmdUpdateGBSurface {
+   uint32 sid;
+} SVGA3dCmdUpdateGBSurface;   /* SVGA_3D_CMD_UPDATE_GB_SURFACE */
+
+/*
+ * Readback an image in a guest-backed surface.
+ * (Request the device to flush the dirty contents into the guest.)
+ */
+
+typedef
+struct SVGA3dCmdReadbackGBImage {
+   SVGA3dSurfaceImageId image;
+} SVGA3dCmdReadbackGBImage;   /* SVGA_3D_CMD_READBACK_GB_IMAGE*/
+
+/*
+ * Readback an entire guest-backed surface.
+ * (Request the device to flush the dirty contents into the guest.)
+ */
+
+typedef
+struct SVGA3dCmdReadbackGBSurface {
+   uint32 sid;
+} SVGA3dCmdReadbackGBSurface;   /* SVGA_3D_CMD_READBACK_GB_SURFACE */
+
+/*
+ * Readback a sub rect of an image in a guest-backed surface.  After
+ * issuing this command the driver is required to issue an update call
+ * of the same region before issuing any other commands that reference
+ * this surface or rendering is not guaranteed.
+ */
+
+typedef
+struct SVGA3dCmdReadbackGBImagePartial {
+   SVGA3dSurfaceImageId image;
+   SVGA3dBox box;
+   uint32 invertBox;
+}
+SVGA3dCmdReadbackGBImagePartial; /* SVGA_3D_CMD_READBACK_GB_IMAGE_PARTIAL */
+
+/*
+ * Invalidate an image in a guest-backed surface.
+ * (Notify the device that the contents can be lost.)
+ */
+
+typedef
+struct SVGA3dCmdInvalidateGBImage {
+   SVGA3dSurfaceImageId image;
+} SVGA3dCmdInvalidateGBImage;   /* SVGA_3D_CMD_INVALIDATE_GB_IMAGE */
+
+/*
+ * Invalidate an entire guest-backed surface.
+ * (Notify the device that the contents if all images can be lost.)
+ */
+
+typedef
+struct SVGA3dCmdInvalidateGBSurface {
+   uint32 sid;
+} SVGA3dCmdInvalidateGBSurface; /* SVGA_3D_CMD_INVALIDATE_GB_SURFACE */
+
+/*
+ * Invalidate a sub rect of an image in a guest-backed surface.  After
+ * issuing this command the driver is required to issue an update call
+ * of the same region before issuing any other commands that reference
+ * this surface or rendering is not guaranteed.
+ */
+
+typedef
+struct SVGA3dCmdInvalidateGBImagePartial {
+   SVGA3dSurfaceImageId image;
+   SVGA3dBox box;
+   uint32 invertBox;
+}
+SVGA3dCmdInvalidateGBImagePartial; /* SVGA_3D_CMD_INVALIDATE_GB_IMAGE_PARTIAL */
+
+/*
+ * Define a guest-backed context.
+ */
+
+typedef
+struct SVGA3dCmdDefineGBContext {
+   uint32 cid;
+} SVGA3dCmdDefineGBContext;   /* SVGA_3D_CMD_DEFINE_GB_CONTEXT */
+
+/*
+ * Destroy a guest-backed context.
+ */
+
+typedef
+struct SVGA3dCmdDestroyGBContext {
+   uint32 cid;
+} SVGA3dCmdDestroyGBContext;   /* SVGA_3D_CMD_DESTROY_GB_CONTEXT */
+
+/*
+ * Bind a guest-backed context.
+ *
+ * validContents should be set to 0 for new contexts,
+ * and 1 if this is an old context which is getting paged
+ * back on to the device.
+ *
+ * For new contexts, it is recommended that the driver
+ * issue commands to initialize all interesting state
+ * prior to rendering.
+ */
+
+typedef
+struct SVGA3dCmdBindGBContext {
+   uint32 cid;
+   SVGAMobId mobid;
+   uint32 validContents;
+} SVGA3dCmdBindGBContext;   /* SVGA_3D_CMD_BIND_GB_CONTEXT */
+
+/*
+ * Readback a guest-backed context.
+ * (Request that the device flush the contents back into guest memory.)
+ */
+
+typedef
+struct SVGA3dCmdReadbackGBContext {
+   uint32 cid;
+} SVGA3dCmdReadbackGBContext;   /* SVGA_3D_CMD_READBACK_GB_CONTEXT */
+
+/*
+ * Invalidate a guest-backed context.
+ */
+typedef
+struct SVGA3dCmdInvalidateGBContext {
+   uint32 cid;
+} SVGA3dCmdInvalidateGBContext;   /* SVGA_3D_CMD_INVALIDATE_GB_CONTEXT */
+
+/*
+ * Define a guest-backed shader.
+ */
+
+typedef
+struct SVGA3dCmdDefineGBShader {
+   uint32 shid;
+   SVGA3dShaderType type;
+   uint32 sizeInBytes;
+} SVGA3dCmdDefineGBShader;   /* SVGA_3D_CMD_DEFINE_GB_SHADER */
+
+/*
+ * Bind a guest-backed shader.
+ */
+
+typedef struct SVGA3dCmdBindGBShader {
+   uint32 shid;
+   SVGAMobId mobid;
+   uint32 offsetInBytes;
+} SVGA3dCmdBindGBShader;   /* SVGA_3D_CMD_BIND_GB_SHADER */
+
+/*
+ * Destroy a guest-backed shader.
+ */
+
+typedef struct SVGA3dCmdDestroyGBShader {
+   uint32 shid;
+} SVGA3dCmdDestroyGBShader;   /* SVGA_3D_CMD_DESTROY_GB_SHADER */
+
+typedef
+struct {
+   uint32                  cid;
+   uint32                  regStart;
+   SVGA3dShaderType        shaderType;
+   SVGA3dShaderConstType   constType;
+
+   /*
+    * Followed by a variable number of shader constants.
+    *
+    * Note that FLOAT and INT constants are 4-dwords in length, while
+    * BOOL constants are 1-dword in length.
+    */
+} SVGA3dCmdSetGBShaderConstInline;
+/* SVGA_3D_CMD_SET_GB_SHADERCONSTS_INLINE */
+
+typedef
+struct {
+   uint32               cid;
+   SVGA3dQueryType      type;
+} SVGA3dCmdBeginGBQuery;           /* SVGA_3D_CMD_BEGIN_GB_QUERY */
+
+typedef
+struct {
+   uint32               cid;
+   SVGA3dQueryType      type;
+   SVGAMobId mobid;
+   uint32 offset;
+} SVGA3dCmdEndGBQuery;                  /* SVGA_3D_CMD_END_GB_QUERY */
+
+
+/*
+ * SVGA_3D_CMD_WAIT_FOR_GB_QUERY --
+ *
+ *    The semantics of this command are identical to the
+ *    SVGA_3D_CMD_WAIT_FOR_QUERY except that the results are written
+ *    to a Mob instead of a GMR.
+ */
+
+typedef
+struct {
+   uint32               cid;
+   SVGA3dQueryType      type;
+   SVGAMobId mobid;
+   uint32 offset;
+} SVGA3dCmdWaitForGBQuery;          /* SVGA_3D_CMD_WAIT_FOR_GB_QUERY */
+
+typedef
+struct {
+   SVGAMobId mobid;
+   uint32 fbOffset;
+   uint32 initalized;
+}
+SVGA3dCmdEnableGart;              /* SVGA_3D_CMD_ENABLE_GART */
+
+typedef
+struct {
+   SVGAMobId mobid;
+   uint32 gartOffset;
+}
+SVGA3dCmdMapMobIntoGart;          /* SVGA_3D_CMD_MAP_MOB_INTO_GART */
+
+
+typedef
+struct {
+   uint32 gartOffset;
+   uint32 numPages;
+}
+SVGA3dCmdUnmapGartRange;          /* SVGA_3D_CMD_UNMAP_GART_RANGE */
+
+
+/*
+ * Screen Targets
+ */
+#define SVGA_STFLAG_PRIMARY (1 << 0)
+
+typedef
+struct {
+   uint32 stid;
+   uint32 width;
+   uint32 height;
+   int32 xRoot;
+   int32 yRoot;
+   uint32 flags;
+}
+SVGA3dCmdDefineGBScreenTarget;    /* SVGA_3D_CMD_DEFINE_GB_SCREENTARGET */
+
+typedef
+struct {
+   uint32 stid;
+}
+SVGA3dCmdDestroyGBScreenTarget;  /* SVGA_3D_CMD_DESTROY_GB_SCREENTARGET */
+
+typedef
+struct {
+   uint32 stid;
+   SVGA3dSurfaceImageId image;
+}
+SVGA3dCmdBindGBScreenTarget;  /* SVGA_3D_CMD_BIND_GB_SCREENTARGET */
+
+typedef
+struct {
+   uint32 stid;
+   SVGA3dBox box;
+}
+SVGA3dCmdUpdateGBScreenTarget;  /* SVGA_3D_CMD_UPDATE_GB_SCREENTARGET */
+
 /*
  * Capability query index.
  *
@@ -1879,10 +2538,41 @@ typedef enum {
    SVGA3D_DEVCAP_SURFACEFMT_BC5_UNORM              = 83,
 
    /*
-    * Don't add new caps into the previous section; the values in this
-    * enumeration must not change. You can put new values right before
-    * SVGA3D_DEVCAP_MAX.
+    * Deprecated.
     */
+   SVGA3D_DEVCAP_VGPU10                            = 84,
+
+   /*
+    * This contains several SVGA_3D_CAPS_VIDEO_DECODE elements
+    * ored together, one for every type of video decoding supported.
+    */
+   SVGA3D_DEVCAP_VIDEO_DECODE                      = 85,
+
+   /*
+    * This contains several SVGA_3D_CAPS_VIDEO_PROCESS elements
+    * ored together, one for every type of video processing supported.
+    */
+   SVGA3D_DEVCAP_VIDEO_PROCESS                     = 86,
+
+   SVGA3D_DEVCAP_LINE_AA                           = 87,  /* boolean */
+   SVGA3D_DEVCAP_LINE_STIPPLE                      = 88,  /* boolean */
+   SVGA3D_DEVCAP_MAX_LINE_WIDTH                    = 89,  /* float */
+   SVGA3D_DEVCAP_MAX_AA_LINE_WIDTH                 = 90,  /* float */
+
+   SVGA3D_DEVCAP_SURFACEFMT_YV12                   = 91,
+
+   /*
+    * Does the host support the SVGA logic ops commands?
+    */
+   SVGA3D_DEVCAP_LOGICOPS                          = 92,
+
+   /*
+    * What support does the host have for screen targets?
+    *
+    * See the SVGA3D_SCREENTARGET_CAP bits below.
+    */
+   SVGA3D_DEVCAP_SCREENTARGETS                     = 93,
+
    SVGA3D_DEVCAP_MAX                                  /* This must be the last index. */
 } SVGA3dDevCapIndex;
 

drivers/gpu/drm/vmwgfx/svga_reg.h

@@ -169,7 +169,10 @@ enum {
    SVGA_REG_TRACES = 45,            /* Enable trace-based updates even when FIFO is on */
    SVGA_REG_GMRS_MAX_PAGES = 46,    /* Maximum number of 4KB pages for all GMRs */
    SVGA_REG_MEMORY_SIZE = 47,       /* Total dedicated device memory excluding FIFO */
-   SVGA_REG_TOP = 48,               /* Must be 1 more than the last register */
+   SVGA_REG_MAX_PRIMARY_BOUNDING_BOX_MEM = 50,   /* Max primary memory */
+   SVGA_REG_SUGGESTED_GBOBJECT_MEM_SIZE_KB = 51, /* Suggested limit on mob mem */
+   SVGA_REG_DEV_CAP = 52,           /* Write dev cap index, read value */
+   SVGA_REG_TOP = 53,               /* Must be 1 more than the last register */
 
    SVGA_PALETTE_BASE = 1024,        /* Base of SVGA color map */
    /* Next 768 (== 256*3) registers exist for colormap */
@@ -431,7 +434,10 @@ struct SVGASignedPoint {
 #define SVGA_CAP_TRACES             0x00200000
 #define SVGA_CAP_GMR2               0x00400000
 #define SVGA_CAP_SCREEN_OBJECT_2    0x00800000
-
+#define SVGA_CAP_COMMAND_BUFFERS    0x01000000
+#define SVGA_CAP_DEAD1              0x02000000
+#define SVGA_CAP_CMD_BUFFERS_2      0x04000000
+#define SVGA_CAP_GBOBJECTS          0x08000000
 
 /*
  * FIFO register indices.

drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c

@@ -40,6 +40,10 @@ static uint32_t vram_ne_placement_flags = TTM_PL_FLAG_VRAM |
 static uint32_t sys_placement_flags = TTM_PL_FLAG_SYSTEM |
 	TTM_PL_FLAG_CACHED;
 
+static uint32_t sys_ne_placement_flags = TTM_PL_FLAG_SYSTEM |
+	TTM_PL_FLAG_CACHED |
+	TTM_PL_FLAG_NO_EVICT;
+
 static uint32_t gmr_placement_flags = VMW_PL_FLAG_GMR |
 	TTM_PL_FLAG_CACHED;
 
@@ -47,6 +51,9 @@ static uint32_t gmr_ne_placement_flags = VMW_PL_FLAG_GMR |
 	TTM_PL_FLAG_CACHED |
 	TTM_PL_FLAG_NO_EVICT;
 
+static uint32_t mob_placement_flags = VMW_PL_FLAG_MOB |
+	TTM_PL_FLAG_CACHED;
+
 struct ttm_placement vmw_vram_placement = {
 	.fpfn = 0,
 	.lpfn = 0,
@@ -116,16 +123,26 @@ struct ttm_placement vmw_sys_placement = {
 	.busy_placement = &sys_placement_flags
 };
 
+struct ttm_placement vmw_sys_ne_placement = {
+	.fpfn = 0,
+	.lpfn = 0,
+	.num_placement = 1,
+	.placement = &sys_ne_placement_flags,
+	.num_busy_placement = 1,
+	.busy_placement = &sys_ne_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
+	VMW_PL_FLAG_GMR | TTM_PL_FLAG_CACHED,
+	VMW_PL_FLAG_MOB | TTM_PL_FLAG_CACHED
 };
 
 struct ttm_placement vmw_evictable_placement = {
 	.fpfn = 0,
 	.lpfn = 0,
-	.num_placement = 3,
+	.num_placement = 4,
 	.placement = evictable_placement_flags,
 	.num_busy_placement = 1,
 	.busy_placement = &sys_placement_flags
@@ -140,10 +157,21 @@ struct ttm_placement vmw_srf_placement = {
 	.busy_placement = gmr_vram_placement_flags
 };
 
+struct ttm_placement vmw_mob_placement = {
+	.fpfn = 0,
+	.lpfn = 0,
+	.num_placement = 1,
+	.num_busy_placement = 1,
+	.placement = &mob_placement_flags,
+	.busy_placement = &mob_placement_flags
+};
+
 struct vmw_ttm_tt {
 	struct ttm_dma_tt dma_ttm;
 	struct vmw_private *dev_priv;
 	int gmr_id;
+	struct vmw_mob *mob;
+	int mem_type;
 	struct sg_table sgt;
 	struct vmw_sg_table vsgt;
 	uint64_t sg_alloc_size;
@@ -244,6 +272,7 @@ void vmw_piter_start(struct vmw_piter *viter, const struct vmw_sg_table *vsgt,
 		viter->dma_address = &__vmw_piter_dma_addr;
 		viter->page = &__vmw_piter_non_sg_page;
 		viter->addrs = vsgt->addrs;
+		viter->pages = vsgt->pages;
 		break;
 	case vmw_dma_map_populate:
 	case vmw_dma_map_bind:
@@ -424,6 +453,63 @@ static void vmw_ttm_unmap_dma(struct vmw_ttm_tt *vmw_tt)
 	vmw_tt->mapped = false;
 }
 
+
+/**
+ * vmw_bo_map_dma - Make sure buffer object pages are visible to the device
+ *
+ * @bo: Pointer to a struct ttm_buffer_object
+ *
+ * Wrapper around vmw_ttm_map_dma, that takes a TTM buffer object pointer
+ * instead of a pointer to a struct vmw_ttm_backend as argument.
+ * Note that the buffer object must be either pinned or reserved before
+ * calling this function.
+ */
+int vmw_bo_map_dma(struct ttm_buffer_object *bo)
+{
+	struct vmw_ttm_tt *vmw_tt =
+		container_of(bo->ttm, struct vmw_ttm_tt, dma_ttm.ttm);
+
+	return vmw_ttm_map_dma(vmw_tt);
+}
+
+
+/**
+ * vmw_bo_unmap_dma - Make sure buffer object pages are visible to the device
+ *
+ * @bo: Pointer to a struct ttm_buffer_object
+ *
+ * Wrapper around vmw_ttm_unmap_dma, that takes a TTM buffer object pointer
+ * instead of a pointer to a struct vmw_ttm_backend as argument.
+ */
+void vmw_bo_unmap_dma(struct ttm_buffer_object *bo)
+{
+	struct vmw_ttm_tt *vmw_tt =
+		container_of(bo->ttm, struct vmw_ttm_tt, dma_ttm.ttm);
+
+	vmw_ttm_unmap_dma(vmw_tt);
+}
+
+
+/**
+ * vmw_bo_sg_table - Return a struct vmw_sg_table object for a
+ * TTM buffer object
+ *
+ * @bo: Pointer to a struct ttm_buffer_object
+ *
+ * Returns a pointer to a struct vmw_sg_table object. The object should
+ * not be freed after use.
+ * Note that for the device addresses to be valid, the buffer object must
+ * either be reserved or pinned.
+ */
+const struct vmw_sg_table *vmw_bo_sg_table(struct ttm_buffer_object *bo)
+{
+	struct vmw_ttm_tt *vmw_tt =
+		container_of(bo->ttm, struct vmw_ttm_tt, dma_ttm.ttm);
+
+	return &vmw_tt->vsgt;
+}
+
+
 static int vmw_ttm_bind(struct ttm_tt *ttm, struct ttm_mem_reg *bo_mem)
 {
 	struct vmw_ttm_tt *vmw_be =
@@ -435,9 +521,27 @@ static int vmw_ttm_bind(struct ttm_tt *ttm, struct ttm_mem_reg *bo_mem)
 		return ret;
 
 	vmw_be->gmr_id = bo_mem->start;
+	vmw_be->mem_type = bo_mem->mem_type;
+
+	switch (bo_mem->mem_type) {
+	case VMW_PL_GMR:
+		return vmw_gmr_bind(vmw_be->dev_priv, &vmw_be->vsgt,
+				    ttm->num_pages, vmw_be->gmr_id);
+	case VMW_PL_MOB:
+		if (unlikely(vmw_be->mob == NULL)) {
+			vmw_be->mob =
+				vmw_mob_create(ttm->num_pages);
+			if (unlikely(vmw_be->mob == NULL))
+				return -ENOMEM;
+		}
 
-	return vmw_gmr_bind(vmw_be->dev_priv, &vmw_be->vsgt,
-			    ttm->num_pages, vmw_be->gmr_id);
+		return vmw_mob_bind(vmw_be->dev_priv, vmw_be->mob,
+				    &vmw_be->vsgt, ttm->num_pages,
+				    vmw_be->gmr_id);
+	default:
+		BUG();
+	}
+	return 0;
 }
 
 static int vmw_ttm_unbind(struct ttm_tt *ttm)
@@ -445,7 +549,16 @@ static int vmw_ttm_unbind(struct ttm_tt *ttm)
 	struct vmw_ttm_tt *vmw_be =
 		container_of(ttm, struct vmw_ttm_tt, dma_ttm.ttm);
 
-	vmw_gmr_unbind(vmw_be->dev_priv, vmw_be->gmr_id);
+	switch (vmw_be->mem_type) {
+	case VMW_PL_GMR:
+		vmw_gmr_unbind(vmw_be->dev_priv, vmw_be->gmr_id);
+		break;
+	case VMW_PL_MOB:
+		vmw_mob_unbind(vmw_be->dev_priv, vmw_be->mob);
+		break;
+	default:
+		BUG();
+	}
 
 	if (vmw_be->dev_priv->map_mode == vmw_dma_map_bind)
 		vmw_ttm_unmap_dma(vmw_be);
@@ -453,6 +566,7 @@ static int vmw_ttm_unbind(struct ttm_tt *ttm)
 	return 0;
 }
 
+
 static void vmw_ttm_destroy(struct ttm_tt *ttm)
 {
 	struct vmw_ttm_tt *vmw_be =
@@ -463,9 +577,14 @@ static void vmw_ttm_destroy(struct ttm_tt *ttm)
 		ttm_dma_tt_fini(&vmw_be->dma_ttm);
 	else
 		ttm_tt_fini(ttm);
+
+	if (vmw_be->mob)
+		vmw_mob_destroy(vmw_be->mob);
+
 	kfree(vmw_be);
 }
 
+
 static int vmw_ttm_populate(struct ttm_tt *ttm)
 {
 	struct vmw_ttm_tt *vmw_tt =
@@ -500,6 +619,12 @@ static void vmw_ttm_unpopulate(struct ttm_tt *ttm)
 	struct vmw_private *dev_priv = vmw_tt->dev_priv;
 	struct ttm_mem_global *glob = vmw_mem_glob(dev_priv);
 
+
+	if (vmw_tt->mob) {
+		vmw_mob_destroy(vmw_tt->mob);
+		vmw_tt->mob = NULL;
+	}
+
 	vmw_ttm_unmap_dma(vmw_tt);
 	if (dev_priv->map_mode == vmw_dma_alloc_coherent) {
 		size_t size =
@@ -530,6 +655,7 @@ static struct ttm_tt *vmw_ttm_tt_create(struct ttm_bo_device *bdev,
 
 	vmw_be->dma_ttm.ttm.func = &vmw_ttm_func;
 	vmw_be->dev_priv = container_of(bdev, struct vmw_private, bdev);
+	vmw_be->mob = NULL;
 
 	if (vmw_be->dev_priv->map_mode == vmw_dma_alloc_coherent)
 		ret = ttm_dma_tt_init(&vmw_be->dma_ttm, bdev, size, page_flags,
@@ -571,6 +697,7 @@ static int vmw_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 		man->default_caching = TTM_PL_FLAG_CACHED;
 		break;
 	case VMW_PL_GMR:
+	case VMW_PL_MOB:
 		/*
 		 * "Guest Memory Regions" is an aperture like feature with
 		 *  one slot per bo. There is an upper limit of the number of
@@ -618,6 +745,7 @@ static int vmw_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg
 	switch (mem->mem_type) {
 	case TTM_PL_SYSTEM:
 	case VMW_PL_GMR:
+	case VMW_PL_MOB:
 		return 0;
 	case TTM_PL_VRAM:
 		mem->bus.offset = mem->start << PAGE_SHIFT;
@@ -677,6 +805,38 @@ static int vmw_sync_obj_wait(void *sync_obj, bool lazy, bool interruptible)
 				  VMW_FENCE_WAIT_TIMEOUT);
 }
 
+/**
+ * vmw_move_notify - TTM move_notify_callback
+ *
+ * @bo:             The TTM buffer object about to move.
+ * @mem:            The truct ttm_mem_reg indicating to what memory
+ *                  region the move is taking place.
+ *
+ * Calls move_notify for all subsystems needing it.
+ * (currently only resources).
+ */
+static void vmw_move_notify(struct ttm_buffer_object *bo,
+			    struct ttm_mem_reg *mem)
+{
+	vmw_resource_move_notify(bo, mem);
+}
+
+
+/**
+ * vmw_swap_notify - TTM move_notify_callback
+ *
+ * @bo:             The TTM buffer object about to be swapped out.
+ */
+static void vmw_swap_notify(struct ttm_buffer_object *bo)
+{
+	struct ttm_bo_device *bdev = bo->bdev;
+
+	spin_lock(&bdev->fence_lock);
+	ttm_bo_wait(bo, false, false, false);
+	spin_unlock(&bdev->fence_lock);
+}
+
+
 struct ttm_bo_driver vmw_bo_driver = {
 	.ttm_tt_create = &vmw_ttm_tt_create,
 	.ttm_tt_populate = &vmw_ttm_populate,
@@ -691,8 +851,8 @@ struct ttm_bo_driver vmw_bo_driver = {
 	.sync_obj_flush = vmw_sync_obj_flush,
 	.sync_obj_unref = vmw_sync_obj_unref,
 	.sync_obj_ref = vmw_sync_obj_ref,
-	.move_notify = NULL,
-	.swap_notify = NULL,
+	.move_notify = vmw_move_notify,
+	.swap_notify = vmw_swap_notify,
 	.fault_reserve_notify = &vmw_ttm_fault_reserve_notify,
 	.io_mem_reserve = &vmw_ttm_io_mem_reserve,
 	.io_mem_free = &vmw_ttm_io_mem_free,

drivers/gpu/drm/vmwgfx/vmwgfx_context.c

@@ -32,12 +32,28 @@
 struct vmw_user_context {
 	struct ttm_base_object base;
 	struct vmw_resource res;
+	struct vmw_ctx_binding_state cbs;
 };
 
+
+
+typedef int (*vmw_scrub_func)(struct vmw_ctx_bindinfo *);
+
 static void vmw_user_context_free(struct vmw_resource *res);
 static struct vmw_resource *
 vmw_user_context_base_to_res(struct ttm_base_object *base);
 
+static int vmw_gb_context_create(struct vmw_resource *res);
+static int vmw_gb_context_bind(struct vmw_resource *res,
+			       struct ttm_validate_buffer *val_buf);
+static int vmw_gb_context_unbind(struct vmw_resource *res,
+				 bool readback,
+				 struct ttm_validate_buffer *val_buf);
+static int vmw_gb_context_destroy(struct vmw_resource *res);
+static int vmw_context_scrub_shader(struct vmw_ctx_bindinfo *bi);
+static int vmw_context_scrub_render_target(struct vmw_ctx_bindinfo *bi);
+static int vmw_context_scrub_texture(struct vmw_ctx_bindinfo *bi);
+static void vmw_context_binding_state_kill(struct vmw_ctx_binding_state *cbs);
 static uint64_t vmw_user_context_size;
 
 static const struct vmw_user_resource_conv user_context_conv = {
@@ -62,6 +78,23 @@ static const struct vmw_res_func vmw_legacy_context_func = {
 	.unbind = NULL
 };
 
+static const struct vmw_res_func vmw_gb_context_func = {
+	.res_type = vmw_res_context,
+	.needs_backup = true,
+	.may_evict = true,
+	.type_name = "guest backed contexts",
+	.backup_placement = &vmw_mob_placement,
+	.create = vmw_gb_context_create,
+	.destroy = vmw_gb_context_destroy,
+	.bind = vmw_gb_context_bind,
+	.unbind = vmw_gb_context_unbind
+};
+
+static const vmw_scrub_func vmw_scrub_funcs[vmw_ctx_binding_max] = {
+	[vmw_ctx_binding_shader] = vmw_context_scrub_shader,
+	[vmw_ctx_binding_rt] = vmw_context_scrub_render_target,
+	[vmw_ctx_binding_tex] = vmw_context_scrub_texture };
+
 /**
  * Context management:
  */
@@ -76,6 +109,16 @@ static void vmw_hw_context_destroy(struct vmw_resource *res)
 	} *cmd;
 
 
+	if (res->func->destroy == vmw_gb_context_destroy) {
+		mutex_lock(&dev_priv->cmdbuf_mutex);
+		(void) vmw_gb_context_destroy(res);
+		if (dev_priv->pinned_bo != NULL &&
+		    !dev_priv->query_cid_valid)
+			__vmw_execbuf_release_pinned_bo(dev_priv, NULL);
+		mutex_unlock(&dev_priv->cmdbuf_mutex);
+		return;
+	}
+
 	vmw_execbuf_release_pinned_bo(dev_priv);
 	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
 	if (unlikely(cmd == NULL)) {
@@ -92,6 +135,33 @@ static void vmw_hw_context_destroy(struct vmw_resource *res)
 	vmw_3d_resource_dec(dev_priv, false);
 }
 
+static int vmw_gb_context_init(struct vmw_private *dev_priv,
+			       struct vmw_resource *res,
+			       void (*res_free) (struct vmw_resource *res))
+{
+	int ret;
+	struct vmw_user_context *uctx =
+		container_of(res, struct vmw_user_context, res);
+
+	ret = vmw_resource_init(dev_priv, res, true,
+				res_free, &vmw_gb_context_func);
+	res->backup_size = SVGA3D_CONTEXT_DATA_SIZE;
+
+	if (unlikely(ret != 0)) {
+		if (res_free)
+			res_free(res);
+		else
+			kfree(res);
+		return ret;
+	}
+
+	memset(&uctx->cbs, 0, sizeof(uctx->cbs));
+	INIT_LIST_HEAD(&uctx->cbs.list);
+
+	vmw_resource_activate(res, vmw_hw_context_destroy);
+	return 0;
+}
+
 static int vmw_context_init(struct vmw_private *dev_priv,
 			    struct vmw_resource *res,
 			    void (*res_free) (struct vmw_resource *res))
@@ -103,6 +173,9 @@ static int vmw_context_init(struct vmw_private *dev_priv,
 		SVGA3dCmdDefineContext body;
 	} *cmd;
 
+	if (dev_priv->has_mob)
+		return vmw_gb_context_init(dev_priv, res, res_free);
+
 	ret = vmw_resource_init(dev_priv, res, false,
 				res_free, &vmw_legacy_context_func);
 
@@ -154,6 +227,184 @@ struct vmw_resource *vmw_context_alloc(struct vmw_private *dev_priv)
 	return (ret == 0) ? res : NULL;
 }
 
+
+static int vmw_gb_context_create(struct vmw_resource *res)
+{
+	struct vmw_private *dev_priv = res->dev_priv;
+	int ret;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdDefineGBContext body;
+	} *cmd;
+
+	if (likely(res->id != -1))
+		return 0;
+
+	ret = vmw_resource_alloc_id(res);
+	if (unlikely(ret != 0)) {
+		DRM_ERROR("Failed to allocate a context id.\n");
+		goto out_no_id;
+	}
+
+	if (unlikely(res->id >= VMWGFX_NUM_GB_CONTEXT)) {
+		ret = -EBUSY;
+		goto out_no_fifo;
+	}
+
+	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Failed reserving FIFO space for context "
+			  "creation.\n");
+		ret = -ENOMEM;
+		goto out_no_fifo;
+	}
+
+	cmd->header.id = SVGA_3D_CMD_DEFINE_GB_CONTEXT;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.cid = res->id;
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+	(void) vmw_3d_resource_inc(dev_priv, false);
+
+	return 0;
+
+out_no_fifo:
+	vmw_resource_release_id(res);
+out_no_id:
+	return ret;
+}
+
+static int vmw_gb_context_bind(struct vmw_resource *res,
+			       struct ttm_validate_buffer *val_buf)
+{
+	struct vmw_private *dev_priv = res->dev_priv;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdBindGBContext body;
+	} *cmd;
+	struct ttm_buffer_object *bo = val_buf->bo;
+
+	BUG_ON(bo->mem.mem_type != VMW_PL_MOB);
+
+	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Failed reserving FIFO space for context "
+			  "binding.\n");
+		return -ENOMEM;
+	}
+
+	cmd->header.id = SVGA_3D_CMD_BIND_GB_CONTEXT;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.cid = res->id;
+	cmd->body.mobid = bo->mem.start;
+	cmd->body.validContents = res->backup_dirty;
+	res->backup_dirty = false;
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+
+	return 0;
+}
+
+static int vmw_gb_context_unbind(struct vmw_resource *res,
+				 bool readback,
+				 struct ttm_validate_buffer *val_buf)
+{
+	struct vmw_private *dev_priv = res->dev_priv;
+	struct ttm_buffer_object *bo = val_buf->bo;
+	struct vmw_fence_obj *fence;
+	struct vmw_user_context *uctx =
+		container_of(res, struct vmw_user_context, res);
+
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdReadbackGBContext body;
+	} *cmd1;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdBindGBContext body;
+	} *cmd2;
+	uint32_t submit_size;
+	uint8_t *cmd;
+
+
+	BUG_ON(bo->mem.mem_type != VMW_PL_MOB);
+
+	mutex_lock(&dev_priv->binding_mutex);
+	vmw_context_binding_state_kill(&uctx->cbs);
+
+	submit_size = sizeof(*cmd2) + (readback ? sizeof(*cmd1) : 0);
+
+	cmd = vmw_fifo_reserve(dev_priv, submit_size);
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Failed reserving FIFO space for context "
+			  "unbinding.\n");
+		mutex_unlock(&dev_priv->binding_mutex);
+		return -ENOMEM;
+	}
+
+	cmd2 = (void *) cmd;
+	if (readback) {
+		cmd1 = (void *) cmd;
+		cmd1->header.id = SVGA_3D_CMD_READBACK_GB_CONTEXT;
+		cmd1->header.size = sizeof(cmd1->body);
+		cmd1->body.cid = res->id;
+		cmd2 = (void *) (&cmd1[1]);
+	}
+	cmd2->header.id = SVGA_3D_CMD_BIND_GB_CONTEXT;
+	cmd2->header.size = sizeof(cmd2->body);
+	cmd2->body.cid = res->id;
+	cmd2->body.mobid = SVGA3D_INVALID_ID;
+
+	vmw_fifo_commit(dev_priv, submit_size);
+	mutex_unlock(&dev_priv->binding_mutex);
+
+	/*
+	 * Create a fence object and fence the backup buffer.
+	 */
+
+	(void) vmw_execbuf_fence_commands(NULL, dev_priv,
+					  &fence, NULL);
+
+	vmw_fence_single_bo(bo, fence);
+
+	if (likely(fence != NULL))
+		vmw_fence_obj_unreference(&fence);
+
+	return 0;
+}
+
+static int vmw_gb_context_destroy(struct vmw_resource *res)
+{
+	struct vmw_private *dev_priv = res->dev_priv;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdDestroyGBContext body;
+	} *cmd;
+	struct vmw_user_context *uctx =
+		container_of(res, struct vmw_user_context, res);
+
+	BUG_ON(!list_empty(&uctx->cbs.list));
+
+	if (likely(res->id == -1))
+		return 0;
+
+	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Failed reserving FIFO space for context "
+			  "destruction.\n");
+		return -ENOMEM;
+	}
+
+	cmd->header.id = SVGA_3D_CMD_DESTROY_GB_CONTEXT;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.cid = res->id;
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+	if (dev_priv->query_cid == res->id)
+		dev_priv->query_cid_valid = false;
+	vmw_resource_release_id(res);
+	vmw_3d_resource_dec(dev_priv, false);
+
+	return 0;
+}
+
 /**
  * User-space context management:
  */
@@ -272,3 +523,283 @@ out_unlock:
 	return ret;
 
 }
+
+/**
+ * vmw_context_scrub_shader - scrub a shader binding from a context.
+ *
+ * @bi: single binding information.
+ */
+static int vmw_context_scrub_shader(struct vmw_ctx_bindinfo *bi)
+{
+	struct vmw_private *dev_priv = bi->ctx->dev_priv;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdSetShader body;
+	} *cmd;
+
+	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Failed reserving FIFO space for shader "
+			  "unbinding.\n");
+		return -ENOMEM;
+	}
+
+	cmd->header.id = SVGA_3D_CMD_SET_SHADER;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.cid = bi->ctx->id;
+	cmd->body.type = bi->i1.shader_type;
+	cmd->body.shid = SVGA3D_INVALID_ID;
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+
+	return 0;
+}
+
+/**
+ * vmw_context_scrub_render_target - scrub a render target binding
+ * from a context.
+ *
+ * @bi: single binding information.
+ */
+static int vmw_context_scrub_render_target(struct vmw_ctx_bindinfo *bi)
+{
+	struct vmw_private *dev_priv = bi->ctx->dev_priv;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdSetRenderTarget body;
+	} *cmd;
+
+	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Failed reserving FIFO space for render target "
+			  "unbinding.\n");
+		return -ENOMEM;
+	}
+
+	cmd->header.id = SVGA_3D_CMD_SETRENDERTARGET;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.cid = bi->ctx->id;
+	cmd->body.type = bi->i1.rt_type;
+	cmd->body.target.sid = SVGA3D_INVALID_ID;
+	cmd->body.target.face = 0;
+	cmd->body.target.mipmap = 0;
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+
+	return 0;
+}
+
+/**
+ * vmw_context_scrub_texture - scrub a texture binding from a context.
+ *
+ * @bi: single binding information.
+ *
+ * TODO: Possibly complement this function with a function that takes
+ * a list of texture bindings and combines them to a single command.
+ */
+static int vmw_context_scrub_texture(struct vmw_ctx_bindinfo *bi)
+{
+	struct vmw_private *dev_priv = bi->ctx->dev_priv;
+	struct {
+		SVGA3dCmdHeader header;
+		struct {
+			SVGA3dCmdSetTextureState c;
+			SVGA3dTextureState s1;
+		} body;
+	} *cmd;
+
+	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Failed reserving FIFO space for texture "
+			  "unbinding.\n");
+		return -ENOMEM;
+	}
+
+
+	cmd->header.id = SVGA_3D_CMD_SETTEXTURESTATE;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.c.cid = bi->ctx->id;
+	cmd->body.s1.stage = bi->i1.texture_stage;
+	cmd->body.s1.name = SVGA3D_TS_BIND_TEXTURE;
+	cmd->body.s1.value = (uint32) SVGA3D_INVALID_ID;
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+
+	return 0;
+}
+
+/**
+ * vmw_context_binding_drop: Stop tracking a context binding
+ *
+ * @cb: Pointer to binding tracker storage.
+ *
+ * Stops tracking a context binding, and re-initializes its storage.
+ * Typically used when the context binding is replaced with a binding to
+ * another (or the same, for that matter) resource.
+ */
+static void vmw_context_binding_drop(struct vmw_ctx_binding *cb)
+{
+	list_del(&cb->ctx_list);
+	if (!list_empty(&cb->res_list))
+		list_del(&cb->res_list);
+	cb->bi.ctx = NULL;
+}
+
+/**
+ * vmw_context_binding_add: Start tracking a context binding
+ *
+ * @cbs: Pointer to the context binding state tracker.
+ * @bi: Information about the binding to track.
+ *
+ * Performs basic checks on the binding to make sure arguments are within
+ * bounds and then starts tracking the binding in the context binding
+ * state structure @cbs.
+ */
+int vmw_context_binding_add(struct vmw_ctx_binding_state *cbs,
+			    const struct vmw_ctx_bindinfo *bi)
+{
+	struct vmw_ctx_binding *loc;
+
+	switch (bi->bt) {
+	case vmw_ctx_binding_rt:
+		if (unlikely((unsigned)bi->i1.rt_type >= SVGA3D_RT_MAX)) {
+			DRM_ERROR("Illegal render target type %u.\n",
+				  (unsigned) bi->i1.rt_type);
+			return -EINVAL;
+		}
+		loc = &cbs->render_targets[bi->i1.rt_type];
+		break;
+	case vmw_ctx_binding_tex:
+		if (unlikely((unsigned)bi->i1.texture_stage >=
+			     SVGA3D_NUM_TEXTURE_UNITS)) {
+			DRM_ERROR("Illegal texture/sampler unit %u.\n",
+				  (unsigned) bi->i1.texture_stage);
+			return -EINVAL;
+		}
+		loc = &cbs->texture_units[bi->i1.texture_stage];
+		break;
+	case vmw_ctx_binding_shader:
+		if (unlikely((unsigned)bi->i1.shader_type >=
+			     SVGA3D_SHADERTYPE_MAX)) {
+			DRM_ERROR("Illegal shader type %u.\n",
+				  (unsigned) bi->i1.shader_type);
+			return -EINVAL;
+		}
+		loc = &cbs->shaders[bi->i1.shader_type];
+		break;
+	default:
+		BUG();
+	}
+
+	if (loc->bi.ctx != NULL)
+		vmw_context_binding_drop(loc);
+
+	loc->bi = *bi;
+	list_add_tail(&loc->ctx_list, &cbs->list);
+	INIT_LIST_HEAD(&loc->res_list);
+
+	return 0;
+}
+
+/**
+ * vmw_context_binding_transfer: Transfer a context binding tracking entry.
+ *
+ * @cbs: Pointer to the persistent context binding state tracker.
+ * @bi: Information about the binding to track.
+ *
+ */
+static void vmw_context_binding_transfer(struct vmw_ctx_binding_state *cbs,
+					 const struct vmw_ctx_bindinfo *bi)
+{
+	struct vmw_ctx_binding *loc;
+
+	switch (bi->bt) {
+	case vmw_ctx_binding_rt:
+		loc = &cbs->render_targets[bi->i1.rt_type];
+		break;
+	case vmw_ctx_binding_tex:
+		loc = &cbs->texture_units[bi->i1.texture_stage];
+		break;
+	case vmw_ctx_binding_shader:
+		loc = &cbs->shaders[bi->i1.shader_type];
+		break;
+	default:
+		BUG();
+	}
+
+	if (loc->bi.ctx != NULL)
+		vmw_context_binding_drop(loc);
+
+	loc->bi = *bi;
+	list_add_tail(&loc->ctx_list, &cbs->list);
+	if (bi->res != NULL)
+		list_add_tail(&loc->res_list, &bi->res->binding_head);
+	else
+		INIT_LIST_HEAD(&loc->res_list);
+}
+
+/**
+ * vmw_context_binding_kill - Kill a binding on the device
+ * and stop tracking it.
+ *
+ * @cb: Pointer to binding tracker storage.
+ *
+ * Emits FIFO commands to scrub a binding represented by @cb.
+ * Then stops tracking the binding and re-initializes its storage.
+ */
+void vmw_context_binding_kill(struct vmw_ctx_binding *cb)
+{
+	(void) vmw_scrub_funcs[cb->bi.bt](&cb->bi);
+	vmw_context_binding_drop(cb);
+}
+
+/**
+ * vmw_context_binding_state_kill - Kill all bindings associated with a
+ * struct vmw_ctx_binding state structure, and re-initialize the structure.
+ *
+ * @cbs: Pointer to the context binding state tracker.
+ *
+ * Emits commands to scrub all bindings associated with the
+ * context binding state tracker. Then re-initializes the whole structure.
+ */
+static void vmw_context_binding_state_kill(struct vmw_ctx_binding_state *cbs)
+{
+	struct vmw_ctx_binding *entry, *next;
+
+	list_for_each_entry_safe(entry, next, &cbs->list, ctx_list)
+		vmw_context_binding_kill(entry);
+}
+
+/**
+ * vmw_context_binding_res_list_kill - Kill all bindings on a
+ * resource binding list
+ *
+ * @head: list head of resource binding list
+ *
+ * Kills all bindings associated with a specific resource. Typically
+ * called before the resource is destroyed.
+ */
+void vmw_context_binding_res_list_kill(struct list_head *head)
+{
+	struct vmw_ctx_binding *entry, *next;
+
+	list_for_each_entry_safe(entry, next, head, res_list)
+		vmw_context_binding_kill(entry);
+}
+
+/**
+ * vmw_context_binding_state_transfer - Commit staged binding info
+ *
+ * @ctx: Pointer to context to commit the staged binding info to.
+ * @from: Staged binding info built during execbuf.
+ *
+ * Transfers binding info from a temporary structure to the persistent
+ * structure in the context. This can be done once commands
+ */
+void vmw_context_binding_state_transfer(struct vmw_resource *ctx,
+					struct vmw_ctx_binding_state *from)
+{
+	struct vmw_user_context *uctx =
+		container_of(ctx, struct vmw_user_context, res);
+	struct vmw_ctx_binding *entry, *next;
+
+	list_for_each_entry_safe(entry, next, &from->list, ctx_list)
+		vmw_context_binding_transfer(&uctx->cbs, &entry->bi);
+}

drivers/gpu/drm/vmwgfx/vmwgfx_dmabuf.c

@@ -290,8 +290,7 @@ void vmw_bo_get_guest_ptr(const struct ttm_buffer_object *bo,
 /**
  * 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.
+ * @bo: The buffer object. Must be reserved.
  * @pin: Whether to pin or unpin.
  *
  */
@@ -303,10 +302,9 @@ void vmw_bo_pin(struct ttm_buffer_object *bo, bool pin)
 	int ret;
 
 	lockdep_assert_held(&bo->resv->lock.base);
-	BUG_ON(old_mem_type != TTM_PL_VRAM &&
-	       old_mem_type != VMW_PL_GMR);
 
-	pl_flags = TTM_PL_FLAG_VRAM | VMW_PL_FLAG_GMR | TTM_PL_FLAG_CACHED;
+	pl_flags = TTM_PL_FLAG_VRAM | VMW_PL_FLAG_GMR | VMW_PL_FLAG_MOB
+		| TTM_PL_FLAG_SYSTEM | TTM_PL_FLAG_CACHED;
 	if (pin)
 		pl_flags |= TTM_PL_FLAG_NO_EVICT;
 

drivers/gpu/drm/vmwgfx/vmwgfx_drv.c

@@ -112,6 +112,21 @@
 #define DRM_IOCTL_VMW_UPDATE_LAYOUT				\
 	DRM_IOW(DRM_COMMAND_BASE + DRM_VMW_UPDATE_LAYOUT,	\
 		 struct drm_vmw_update_layout_arg)
+#define DRM_IOCTL_VMW_CREATE_SHADER				\
+	DRM_IOWR(DRM_COMMAND_BASE + DRM_VMW_CREATE_SHADER,	\
+		 struct drm_vmw_shader_create_arg)
+#define DRM_IOCTL_VMW_UNREF_SHADER				\
+	DRM_IOW(DRM_COMMAND_BASE + DRM_VMW_UNREF_SHADER,	\
+		 struct drm_vmw_shader_arg)
+#define DRM_IOCTL_VMW_GB_SURFACE_CREATE				\
+	DRM_IOWR(DRM_COMMAND_BASE + DRM_VMW_GB_SURFACE_CREATE,	\
+		 union drm_vmw_gb_surface_create_arg)
+#define DRM_IOCTL_VMW_GB_SURFACE_REF				\
+	DRM_IOWR(DRM_COMMAND_BASE + DRM_VMW_GB_SURFACE_REF,	\
+		 union drm_vmw_gb_surface_reference_arg)
+#define DRM_IOCTL_VMW_SYNCCPU					\
+	DRM_IOW(DRM_COMMAND_BASE + DRM_VMW_SYNCCPU,		\
+		 struct drm_vmw_synccpu_arg)
 
 /**
  * The core DRM version of this macro doesn't account for
@@ -177,6 +192,21 @@ static const struct drm_ioctl_desc vmw_ioctls[] = {
 	VMW_IOCTL_DEF(VMW_UPDATE_LAYOUT,
 		      vmw_kms_update_layout_ioctl,
 		      DRM_MASTER | DRM_UNLOCKED),
+	VMW_IOCTL_DEF(VMW_CREATE_SHADER,
+		      vmw_shader_define_ioctl,
+		      DRM_AUTH | DRM_UNLOCKED),
+	VMW_IOCTL_DEF(VMW_UNREF_SHADER,
+		      vmw_shader_destroy_ioctl,
+		      DRM_AUTH | DRM_UNLOCKED),
+	VMW_IOCTL_DEF(VMW_GB_SURFACE_CREATE,
+		      vmw_gb_surface_define_ioctl,
+		      DRM_AUTH | DRM_UNLOCKED),
+	VMW_IOCTL_DEF(VMW_GB_SURFACE_REF,
+		      vmw_gb_surface_reference_ioctl,
+		      DRM_AUTH | DRM_UNLOCKED),
+	VMW_IOCTL_DEF(VMW_SYNCCPU,
+		      vmw_user_dmabuf_synccpu_ioctl,
+		      DRM_AUTH | DRM_UNLOCKED),
 };
 
 static struct pci_device_id vmw_pci_id_list[] = {
@@ -189,6 +219,7 @@ static int enable_fbdev = IS_ENABLED(CONFIG_DRM_VMWGFX_FBCON);
 static int vmw_force_iommu;
 static int vmw_restrict_iommu;
 static int vmw_force_coherent;
+static int vmw_restrict_dma_mask;
 
 static int vmw_probe(struct pci_dev *, const struct pci_device_id *);
 static void vmw_master_init(struct vmw_master *);
@@ -203,6 +234,8 @@ MODULE_PARM_DESC(restrict_iommu, "Try to limit IOMMU usage for TTM pages");
 module_param_named(restrict_iommu, vmw_restrict_iommu, int, 0600);
 MODULE_PARM_DESC(force_coherent, "Force coherent TTM pages");
 module_param_named(force_coherent, vmw_force_coherent, int, 0600);
+MODULE_PARM_DESC(restrict_dma_mask, "Restrict DMA mask to 44 bits with IOMMU");
+module_param_named(restrict_dma_mask, vmw_restrict_dma_mask, int, 0600);
 
 
 static void vmw_print_capabilities(uint32_t capabilities)
@@ -240,38 +273,52 @@ static void vmw_print_capabilities(uint32_t capabilities)
 		DRM_INFO("  GMR2.\n");
 	if (capabilities & SVGA_CAP_SCREEN_OBJECT_2)
 		DRM_INFO("  Screen Object 2.\n");
+	if (capabilities & SVGA_CAP_COMMAND_BUFFERS)
+		DRM_INFO("  Command Buffers.\n");
+	if (capabilities & SVGA_CAP_CMD_BUFFERS_2)
+		DRM_INFO("  Command Buffers 2.\n");
+	if (capabilities & SVGA_CAP_GBOBJECTS)
+		DRM_INFO("  Guest Backed Resources.\n");
 }
 
-
 /**
- * vmw_execbuf_prepare_dummy_query - Initialize a query result structure at
- * the start of a buffer object.
+ * vmw_dummy_query_bo_create - create a bo to hold a dummy query result
  *
- * @dev_priv: The device private structure.
+ * @dev_priv: A 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.
+ * This function creates a small buffer object that holds the query
+ * result for dummy queries emitted as query barriers.
+ * The function will then map the first page and initialize a pending
+ * occlusion query result structure, Finally it will unmap the buffer.
+ * No interruptible waits are done within this function.
  *
- * TODO: Since we're only mapping a single page, we should optimize the map
- * to use kmap_atomic / iomap_atomic.
+ * Returns an error if bo creation or initialization fails.
  */
-static void vmw_dummy_query_bo_prepare(struct vmw_private *dev_priv)
+static int vmw_dummy_query_bo_create(struct vmw_private *dev_priv)
 {
+	int ret;
+	struct ttm_buffer_object *bo;
 	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);
-	spin_unlock(&bdev->fence_lock);
+	/*
+	 * Create the bo as pinned, so that a tryreserve will
+	 * immediately succeed. This is because we're the only
+	 * user of the bo currently.
+	 */
+	ret = ttm_bo_create(&dev_priv->bdev,
+			    PAGE_SIZE,
+			    ttm_bo_type_device,
+			    &vmw_sys_ne_placement,
+			    0, false, NULL,
+			    &bo);
+
 	if (unlikely(ret != 0))
-		(void) vmw_fallback_wait(dev_priv, false, true, 0, false,
-					 10*HZ);
+		return ret;
+
+	ret = ttm_bo_reserve(bo, false, true, false, 0);
+	BUG_ON(ret != 0);
 
 	ret = ttm_bo_kmap(bo, 0, 1, &map);
 	if (likely(ret == 0)) {
@@ -280,34 +327,19 @@ static void vmw_dummy_query_bo_prepare(struct vmw_private *dev_priv)
 		result->state = SVGA3D_QUERYSTATE_PENDING;
 		result->result32 = 0xff;
 		ttm_bo_kunmap(&map);
-	} else
-		DRM_ERROR("Dummy query buffer map failed.\n");
+	}
+	vmw_bo_pin(bo, false);
 	ttm_bo_unreserve(bo);
-}
 
+	if (unlikely(ret != 0)) {
+		DRM_ERROR("Dummy query buffer map failed.\n");
+		ttm_bo_unref(&bo);
+	} else
+		dev_priv->dummy_query_bo = 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, false, NULL,
-			     &dev_priv->dummy_query_bo);
+	return ret;
 }
 
-
 static int vmw_request_device(struct vmw_private *dev_priv)
 {
 	int ret;
@@ -318,14 +350,24 @@ static int vmw_request_device(struct vmw_private *dev_priv)
 		return ret;
 	}
 	vmw_fence_fifo_up(dev_priv->fman);
+	if (dev_priv->has_mob) {
+		ret = vmw_otables_setup(dev_priv);
+		if (unlikely(ret != 0)) {
+			DRM_ERROR("Unable to initialize "
+				  "guest Memory OBjects.\n");
+			goto out_no_mob;
+		}
+	}
 	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:
+	if (dev_priv->has_mob)
+		vmw_otables_takedown(dev_priv);
+out_no_mob:
 	vmw_fence_fifo_down(dev_priv->fman);
 	vmw_fifo_release(dev_priv, &dev_priv->fifo);
 	return ret;
@@ -341,10 +383,13 @@ static void vmw_release_device(struct vmw_private *dev_priv)
 	BUG_ON(dev_priv->pinned_bo != NULL);
 
 	ttm_bo_unref(&dev_priv->dummy_query_bo);
+	if (dev_priv->has_mob)
+		vmw_otables_takedown(dev_priv);
 	vmw_fence_fifo_down(dev_priv->fman);
 	vmw_fifo_release(dev_priv, &dev_priv->fifo);
 }
 
+
 /**
  * Increase the 3d resource refcount.
  * If the count was prevously zero, initialize the fifo, switching to svga
@@ -510,6 +555,33 @@ out_fixup:
 	return 0;
 }
 
+/**
+ * vmw_dma_masks - set required page- and dma masks
+ *
+ * @dev: Pointer to struct drm-device
+ *
+ * With 32-bit we can only handle 32 bit PFNs. Optionally set that
+ * restriction also for 64-bit systems.
+ */
+#ifdef CONFIG_INTEL_IOMMU
+static int vmw_dma_masks(struct vmw_private *dev_priv)
+{
+	struct drm_device *dev = dev_priv->dev;
+
+	if (intel_iommu_enabled &&
+	    (sizeof(unsigned long) == 4 || vmw_restrict_dma_mask)) {
+		DRM_INFO("Restricting DMA addresses to 44 bits.\n");
+		return dma_set_mask(dev->dev, DMA_BIT_MASK(44));
+	}
+	return 0;
+}
+#else
+static int vmw_dma_masks(struct vmw_private *dev_priv)
+{
+	return 0;
+}
+#endif
+
 static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 {
 	struct vmw_private *dev_priv;
@@ -532,6 +604,7 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 	mutex_init(&dev_priv->hw_mutex);
 	mutex_init(&dev_priv->cmdbuf_mutex);
 	mutex_init(&dev_priv->release_mutex);
+	mutex_init(&dev_priv->binding_mutex);
 	rwlock_init(&dev_priv->resource_lock);
 
 	for (i = vmw_res_context; i < vmw_res_max; ++i) {
@@ -578,14 +651,9 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 
 	vmw_get_initial_size(dev_priv);
 
-	if (dev_priv->capabilities & SVGA_CAP_GMR) {
-		dev_priv->max_gmr_descriptors =
-			vmw_read(dev_priv,
-				 SVGA_REG_GMR_MAX_DESCRIPTOR_LENGTH);
+	if (dev_priv->capabilities & SVGA_CAP_GMR2) {
 		dev_priv->max_gmr_ids =
 			vmw_read(dev_priv, SVGA_REG_GMR_MAX_IDS);
-	}
-	if (dev_priv->capabilities & SVGA_CAP_GMR2) {
 		dev_priv->max_gmr_pages =
 			vmw_read(dev_priv, SVGA_REG_GMRS_MAX_PAGES);
 		dev_priv->memory_size =
@@ -598,23 +666,40 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 		 */
 		dev_priv->memory_size = 512*1024*1024;
 	}
+	dev_priv->max_mob_pages = 0;
+	if (dev_priv->capabilities & SVGA_CAP_GBOBJECTS) {
+		uint64_t mem_size =
+			vmw_read(dev_priv,
+				 SVGA_REG_SUGGESTED_GBOBJECT_MEM_SIZE_KB);
+
+		dev_priv->max_mob_pages = mem_size * 1024 / PAGE_SIZE;
+		dev_priv->prim_bb_mem =
+			vmw_read(dev_priv,
+				 SVGA_REG_MAX_PRIMARY_BOUNDING_BOX_MEM);
+	} else
+		dev_priv->prim_bb_mem = dev_priv->vram_size;
+
+	ret = vmw_dma_masks(dev_priv);
+	if (unlikely(ret != 0))
+		goto out_err0;
+
+	if (unlikely(dev_priv->prim_bb_mem < dev_priv->vram_size))
+		dev_priv->prim_bb_mem = dev_priv->vram_size;
 
 	mutex_unlock(&dev_priv->hw_mutex);
 
 	vmw_print_capabilities(dev_priv->capabilities);
 
-	if (dev_priv->capabilities & SVGA_CAP_GMR) {
+	if (dev_priv->capabilities & SVGA_CAP_GMR2) {
 		DRM_INFO("Max GMR ids is %u\n",
 			 (unsigned)dev_priv->max_gmr_ids);
-		DRM_INFO("Max GMR descriptors is %u\n",
-			 (unsigned)dev_priv->max_gmr_descriptors);
-	}
-	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 surface memory is %u kiB\n",
 			 (unsigned)dev_priv->memory_size / 1024);
 	}
+	DRM_INFO("Maximum display memory size is %u kiB\n",
+		 dev_priv->prim_bb_mem / 1024);
 	DRM_INFO("VRAM at 0x%08x size is %u kiB\n",
 		 dev_priv->vram_start, dev_priv->vram_size / 1024);
 	DRM_INFO("MMIO at 0x%08x size is %u kiB\n",
@@ -649,12 +734,22 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 	dev_priv->has_gmr = true;
 	if (((dev_priv->capabilities & (SVGA_CAP_GMR | SVGA_CAP_GMR2)) == 0) ||
 	    refuse_dma || ttm_bo_init_mm(&dev_priv->bdev, VMW_PL_GMR,
-					 dev_priv->max_gmr_ids) != 0) {
+					 VMW_PL_GMR) != 0) {
 		DRM_INFO("No GMR memory available. "
 			 "Graphics memory resources are very limited.\n");
 		dev_priv->has_gmr = false;
 	}
 
+	if (dev_priv->capabilities & SVGA_CAP_GBOBJECTS) {
+		dev_priv->has_mob = true;
+		if (ttm_bo_init_mm(&dev_priv->bdev, VMW_PL_MOB,
+				   VMW_PL_MOB) != 0) {
+			DRM_INFO("No MOB memory available. "
+				 "3D will be disabled.\n");
+			dev_priv->has_mob = false;
+		}
+	}
+
 	dev_priv->mmio_mtrr = arch_phys_wc_add(dev_priv->mmio_start,
 					       dev_priv->mmio_size);
 
@@ -757,6 +852,8 @@ out_err4:
 	iounmap(dev_priv->mmio_virt);
 out_err3:
 	arch_phys_wc_del(dev_priv->mmio_mtrr);
+	if (dev_priv->has_mob)
+		(void) ttm_bo_clean_mm(&dev_priv->bdev, VMW_PL_MOB);
 	if (dev_priv->has_gmr)
 		(void) ttm_bo_clean_mm(&dev_priv->bdev, VMW_PL_GMR);
 	(void)ttm_bo_clean_mm(&dev_priv->bdev, TTM_PL_VRAM);
@@ -801,6 +898,8 @@ static int vmw_driver_unload(struct drm_device *dev)
 	ttm_object_device_release(&dev_priv->tdev);
 	iounmap(dev_priv->mmio_virt);
 	arch_phys_wc_del(dev_priv->mmio_mtrr);
+	if (dev_priv->has_mob)
+		(void) ttm_bo_clean_mm(&dev_priv->bdev, VMW_PL_MOB);
 	if (dev_priv->has_gmr)
 		(void)ttm_bo_clean_mm(&dev_priv->bdev, VMW_PL_GMR);
 	(void)ttm_bo_clean_mm(&dev_priv->bdev, TTM_PL_VRAM);

drivers/gpu/drm/vmwgfx/vmwgfx_drv.h

@@ -40,9 +40,9 @@
 #include <drm/ttm/ttm_module.h>
 #include "vmwgfx_fence.h"
 
-#define VMWGFX_DRIVER_DATE "20120209"
+#define VMWGFX_DRIVER_DATE "20121114"
 #define VMWGFX_DRIVER_MAJOR 2
-#define VMWGFX_DRIVER_MINOR 4
+#define VMWGFX_DRIVER_MINOR 5
 #define VMWGFX_DRIVER_PATCHLEVEL 0
 #define VMWGFX_FILE_PAGE_OFFSET 0x00100000
 #define VMWGFX_FIFO_STATIC_SIZE (1024*1024)
@@ -50,14 +50,30 @@
 #define VMWGFX_MAX_VALIDATIONS 2048
 #define VMWGFX_MAX_DISPLAYS 16
 #define VMWGFX_CMD_BOUNCE_INIT_SIZE 32768
+#define VMWGFX_ENABLE_SCREEN_TARGET_OTABLE 0
+
+/*
+ * Perhaps we should have sysfs entries for these.
+ */
+#define VMWGFX_NUM_GB_CONTEXT 256
+#define VMWGFX_NUM_GB_SHADER 20000
+#define VMWGFX_NUM_GB_SURFACE 32768
+#define VMWGFX_NUM_GB_SCREEN_TARGET VMWGFX_MAX_DISPLAYS
+#define VMWGFX_NUM_MOB (VMWGFX_NUM_GB_CONTEXT +\
+			VMWGFX_NUM_GB_SHADER +\
+			VMWGFX_NUM_GB_SURFACE +\
+			VMWGFX_NUM_GB_SCREEN_TARGET)
 
 #define VMW_PL_GMR TTM_PL_PRIV0
 #define VMW_PL_FLAG_GMR TTM_PL_FLAG_PRIV0
+#define VMW_PL_MOB TTM_PL_PRIV1
+#define VMW_PL_FLAG_MOB TTM_PL_FLAG_PRIV1
 
 #define VMW_RES_CONTEXT ttm_driver_type0
 #define VMW_RES_SURFACE ttm_driver_type1
 #define VMW_RES_STREAM ttm_driver_type2
 #define VMW_RES_FENCE ttm_driver_type3
+#define VMW_RES_SHADER ttm_driver_type4
 
 struct vmw_fpriv {
 	struct drm_master *locked_master;
@@ -82,6 +98,7 @@ struct vmw_dma_buffer {
 struct vmw_validate_buffer {
 	struct ttm_validate_buffer base;
 	struct drm_hash_item hash;
+	bool validate_as_mob;
 };
 
 struct vmw_res_func;
@@ -98,6 +115,7 @@ struct vmw_resource {
 	const struct vmw_res_func *func;
 	struct list_head lru_head; /* Protected by the resource lock */
 	struct list_head mob_head; /* Protected by @backup reserved */
+	struct list_head binding_head; /* Protected by binding_mutex */
 	void (*res_free) (struct vmw_resource *res);
 	void (*hw_destroy) (struct vmw_resource *res);
 };
@@ -106,6 +124,7 @@ enum vmw_res_type {
 	vmw_res_context,
 	vmw_res_surface,
 	vmw_res_stream,
+	vmw_res_shader,
 	vmw_res_max
 };
 
@@ -154,6 +173,7 @@ struct vmw_fifo_state {
 };
 
 struct vmw_relocation {
+	SVGAMobId *mob_loc;
 	SVGAGuestPtr *location;
 	uint32_t index;
 };
@@ -229,6 +249,71 @@ struct vmw_piter {
 	struct page *(*page)(struct vmw_piter *);
 };
 
+/*
+ * enum vmw_ctx_binding_type - abstract resource to context binding types
+ */
+enum vmw_ctx_binding_type {
+	vmw_ctx_binding_shader,
+	vmw_ctx_binding_rt,
+	vmw_ctx_binding_tex,
+	vmw_ctx_binding_max
+};
+
+/**
+ * struct vmw_ctx_bindinfo - structure representing a single context binding
+ *
+ * @ctx: Pointer to the context structure. NULL means the binding is not
+ * active.
+ * @res: Non ref-counted pointer to the bound resource.
+ * @bt: The binding type.
+ * @i1: Union of information needed to unbind.
+ */
+struct vmw_ctx_bindinfo {
+	struct vmw_resource *ctx;
+	struct vmw_resource *res;
+	enum vmw_ctx_binding_type bt;
+	union {
+		SVGA3dShaderType shader_type;
+		SVGA3dRenderTargetType rt_type;
+		uint32 texture_stage;
+	} i1;
+};
+
+/**
+ * struct vmw_ctx_binding - structure representing a single context binding
+ *                        - suitable for tracking in a context
+ *
+ * @ctx_list: List head for context.
+ * @res_list: List head for bound resource.
+ * @bi: Binding info
+ */
+struct vmw_ctx_binding {
+	struct list_head ctx_list;
+	struct list_head res_list;
+	struct vmw_ctx_bindinfo bi;
+};
+
+
+/**
+ * struct vmw_ctx_binding_state - context binding state
+ *
+ * @list: linked list of individual bindings.
+ * @render_targets: Render target bindings.
+ * @texture_units: Texture units/samplers bindings.
+ * @shaders: Shader bindings.
+ *
+ * Note that this structure also provides storage space for the individual
+ * struct vmw_ctx_binding objects, so that no dynamic allocation is needed
+ * for individual bindings.
+ *
+ */
+struct vmw_ctx_binding_state {
+	struct list_head list;
+	struct vmw_ctx_binding render_targets[SVGA3D_RT_MAX];
+	struct vmw_ctx_binding texture_units[SVGA3D_NUM_TEXTURE_UNITS];
+	struct vmw_ctx_binding shaders[SVGA3D_SHADERTYPE_MAX];
+};
+
 struct vmw_sw_context{
 	struct drm_open_hash res_ht;
 	bool res_ht_initialized;
@@ -250,6 +335,7 @@ struct vmw_sw_context{
 	struct vmw_resource *last_query_ctx;
 	bool needs_post_query_barrier;
 	struct vmw_resource *error_resource;
+	struct vmw_ctx_binding_state staged_bindings;
 };
 
 struct vmw_legacy_display;
@@ -281,6 +367,7 @@ struct vmw_private {
 	unsigned int io_start;
 	uint32_t vram_start;
 	uint32_t vram_size;
+	uint32_t prim_bb_mem;
 	uint32_t mmio_start;
 	uint32_t mmio_size;
 	uint32_t fb_max_width;
@@ -290,11 +377,12 @@ struct vmw_private {
 	__le32 __iomem *mmio_virt;
 	int mmio_mtrr;
 	uint32_t capabilities;
-	uint32_t max_gmr_descriptors;
 	uint32_t max_gmr_ids;
 	uint32_t max_gmr_pages;
+	uint32_t max_mob_pages;
 	uint32_t memory_size;
 	bool has_gmr;
+	bool has_mob;
 	struct mutex hw_mutex;
 
 	/*
@@ -370,6 +458,7 @@ struct vmw_private {
 
 	struct vmw_sw_context ctx;
 	struct mutex cmdbuf_mutex;
+	struct mutex binding_mutex;
 
 	/**
 	 * Operating mode.
@@ -415,6 +504,12 @@ struct vmw_private {
 	 * DMA mapping stuff.
 	 */
 	enum vmw_dma_map_mode map_mode;
+
+	/*
+	 * Guest Backed stuff
+	 */
+	struct ttm_buffer_object *otable_bo;
+	struct vmw_otable *otables;
 };
 
 static inline struct vmw_surface *vmw_res_to_srf(struct vmw_resource *res)
@@ -471,23 +566,12 @@ extern void vmw_gmr_unbind(struct vmw_private *dev_priv, int gmr_id);
  * Resource utilities - vmwgfx_resource.c
  */
 struct vmw_user_resource_conv;
-extern const struct vmw_user_resource_conv *user_surface_converter;
-extern const struct vmw_user_resource_conv *user_context_converter;
 
-extern struct vmw_resource *vmw_context_alloc(struct vmw_private *dev_priv);
 extern void vmw_resource_unreference(struct vmw_resource **p_res);
 extern struct vmw_resource *vmw_resource_reference(struct vmw_resource *res);
 extern int vmw_resource_validate(struct vmw_resource *res);
 extern int vmw_resource_reserve(struct vmw_resource *res, bool no_backup);
 extern bool vmw_resource_needs_backup(const struct vmw_resource *res);
-extern int vmw_context_destroy_ioctl(struct drm_device *dev, void *data,
-				     struct drm_file *file_priv);
-extern int vmw_context_define_ioctl(struct drm_device *dev, void *data,
-				    struct drm_file *file_priv);
-extern int vmw_context_check(struct vmw_private *dev_priv,
-			     struct ttm_object_file *tfile,
-			     int id,
-			     struct vmw_resource **p_res);
 extern int vmw_user_lookup_handle(struct vmw_private *dev_priv,
 				  struct ttm_object_file *tfile,
 				  uint32_t handle,
@@ -499,18 +583,6 @@ extern int vmw_user_resource_lookup_handle(
 	uint32_t handle,
 	const struct vmw_user_resource_conv *converter,
 	struct vmw_resource **p_res);
-extern void vmw_surface_res_free(struct vmw_resource *res);
-extern int vmw_surface_destroy_ioctl(struct drm_device *dev, void *data,
-				     struct drm_file *file_priv);
-extern int vmw_surface_define_ioctl(struct drm_device *dev, void *data,
-				    struct drm_file *file_priv);
-extern int vmw_surface_reference_ioctl(struct drm_device *dev, void *data,
-				       struct drm_file *file_priv);
-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,
@@ -519,10 +591,21 @@ extern int vmw_dmabuf_init(struct vmw_private *dev_priv,
 			   void (*bo_free) (struct ttm_buffer_object *bo));
 extern int vmw_user_dmabuf_verify_access(struct ttm_buffer_object *bo,
 				  struct ttm_object_file *tfile);
+extern int vmw_user_dmabuf_alloc(struct vmw_private *dev_priv,
+				 struct ttm_object_file *tfile,
+				 uint32_t size,
+				 bool shareable,
+				 uint32_t *handle,
+				 struct vmw_dma_buffer **p_dma_buf);
+extern int vmw_user_dmabuf_reference(struct ttm_object_file *tfile,
+				     struct vmw_dma_buffer *dma_buf,
+				     uint32_t *handle);
 extern int vmw_dmabuf_alloc_ioctl(struct drm_device *dev, void *data,
 				  struct drm_file *file_priv);
 extern int vmw_dmabuf_unref_ioctl(struct drm_device *dev, void *data,
 				  struct drm_file *file_priv);
+extern int vmw_user_dmabuf_synccpu_ioctl(struct drm_device *dev, void *data,
+					 struct drm_file *file_priv);
 extern uint32_t vmw_dmabuf_validate_node(struct ttm_buffer_object *bo,
 					 uint32_t cur_validate_node);
 extern void vmw_dmabuf_validate_clear(struct ttm_buffer_object *bo);
@@ -622,10 +705,16 @@ 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_sys_ne_placement;
 extern struct ttm_placement vmw_evictable_placement;
 extern struct ttm_placement vmw_srf_placement;
+extern struct ttm_placement vmw_mob_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);
+extern void vmw_bo_unmap_dma(struct ttm_buffer_object *bo);
+extern const struct vmw_sg_table *
+vmw_bo_sg_table(struct ttm_buffer_object *bo);
 extern void vmw_piter_start(struct vmw_piter *viter,
 			    const struct vmw_sg_table *vsgt,
 			    unsigned long p_offs);
@@ -832,6 +921,76 @@ extern int vmw_prime_handle_to_fd(struct drm_device *dev,
 				  uint32_t handle, uint32_t flags,
 				  int *prime_fd);
 
+/*
+ * MemoryOBject management -  vmwgfx_mob.c
+ */
+struct vmw_mob;
+extern int vmw_mob_bind(struct vmw_private *dev_priv, struct vmw_mob *mob,
+			const struct vmw_sg_table *vsgt,
+			unsigned long num_data_pages, int32_t mob_id);
+extern void vmw_mob_unbind(struct vmw_private *dev_priv,
+			   struct vmw_mob *mob);
+extern void vmw_mob_destroy(struct vmw_mob *mob);
+extern struct vmw_mob *vmw_mob_create(unsigned long data_pages);
+extern int vmw_otables_setup(struct vmw_private *dev_priv);
+extern void vmw_otables_takedown(struct vmw_private *dev_priv);
+
+/*
+ * Context management - vmwgfx_context.c
+ */
+
+extern const struct vmw_user_resource_conv *user_context_converter;
+
+extern struct vmw_resource *vmw_context_alloc(struct vmw_private *dev_priv);
+
+extern int vmw_context_check(struct vmw_private *dev_priv,
+			     struct ttm_object_file *tfile,
+			     int id,
+			     struct vmw_resource **p_res);
+extern int vmw_context_define_ioctl(struct drm_device *dev, void *data,
+				    struct drm_file *file_priv);
+extern int vmw_context_destroy_ioctl(struct drm_device *dev, void *data,
+				     struct drm_file *file_priv);
+extern int vmw_context_binding_add(struct vmw_ctx_binding_state *cbs,
+				   const struct vmw_ctx_bindinfo *ci);
+extern void
+vmw_context_binding_state_transfer(struct vmw_resource *res,
+				   struct vmw_ctx_binding_state *cbs);
+extern void vmw_context_binding_res_list_kill(struct list_head *head);
+
+/*
+ * Surface management - vmwgfx_surface.c
+ */
+
+extern const struct vmw_user_resource_conv *user_surface_converter;
+
+extern void vmw_surface_res_free(struct vmw_resource *res);
+extern int vmw_surface_destroy_ioctl(struct drm_device *dev, void *data,
+				     struct drm_file *file_priv);
+extern int vmw_surface_define_ioctl(struct drm_device *dev, void *data,
+				    struct drm_file *file_priv);
+extern int vmw_surface_reference_ioctl(struct drm_device *dev, void *data,
+				       struct drm_file *file_priv);
+extern int vmw_gb_surface_define_ioctl(struct drm_device *dev, void *data,
+				       struct drm_file *file_priv);
+extern int vmw_gb_surface_reference_ioctl(struct drm_device *dev, void *data,
+					  struct drm_file *file_priv);
+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);
+
+/*
+ * Shader management - vmwgfx_shader.c
+ */
+
+extern const struct vmw_user_resource_conv *user_shader_converter;
+
+extern int vmw_shader_define_ioctl(struct drm_device *dev, void *data,
+				   struct drm_file *file_priv);
+extern int vmw_shader_destroy_ioctl(struct drm_device *dev, void *data,
+				    struct drm_file *file_priv);
 
 /**
  * Inline helper functions

drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c

@@ -54,6 +54,8 @@ struct vmw_resource_relocation {
  * @res: Ref-counted pointer to the resource.
  * @switch_backup: Boolean whether to switch backup buffer on unreserve.
  * @new_backup: Refcounted pointer to the new backup buffer.
+ * @staged_bindings: If @res is a context, tracks bindings set up during
+ * the command batch. Otherwise NULL.
  * @new_backup_offset: New backup buffer offset if @new_backup is non-NUll.
  * @first_usage: Set to true the first time the resource is referenced in
  * the command stream.
@@ -65,11 +67,31 @@ struct vmw_resource_val_node {
 	struct drm_hash_item hash;
 	struct vmw_resource *res;
 	struct vmw_dma_buffer *new_backup;
+	struct vmw_ctx_binding_state *staged_bindings;
 	unsigned long new_backup_offset;
 	bool first_usage;
 	bool no_buffer_needed;
 };
 
+/**
+ * struct vmw_cmd_entry - Describe a command for the verifier
+ *
+ * @user_allow: Whether allowed from the execbuf ioctl.
+ * @gb_disable: Whether disabled if guest-backed objects are available.
+ * @gb_enable: Whether enabled iff guest-backed objects are available.
+ */
+struct vmw_cmd_entry {
+	int (*func) (struct vmw_private *, struct vmw_sw_context *,
+		     SVGA3dCmdHeader *);
+	bool user_allow;
+	bool gb_disable;
+	bool gb_enable;
+};
+
+#define VMW_CMD_DEF(_cmd, _func, _user_allow, _gb_disable, _gb_enable)	\
+	[(_cmd) - SVGA_3D_CMD_BASE] = {(_func), (_user_allow),\
+				       (_gb_disable), (_gb_enable)}
+
 /**
  * vmw_resource_unreserve - unreserve resources previously reserved for
  * command submission.
@@ -87,6 +109,16 @@ static void vmw_resource_list_unreserve(struct list_head *list,
 		struct vmw_dma_buffer *new_backup =
 			backoff ? NULL : val->new_backup;
 
+		/*
+		 * Transfer staged context bindings to the
+		 * persistent context binding tracker.
+		 */
+		if (unlikely(val->staged_bindings)) {
+			vmw_context_binding_state_transfer
+				(val->res, val->staged_bindings);
+			kfree(val->staged_bindings);
+			val->staged_bindings = NULL;
+		}
 		vmw_resource_unreserve(res, new_backup,
 			val->new_backup_offset);
 		vmw_dmabuf_unreference(&val->new_backup);
@@ -224,6 +256,7 @@ static int vmw_cmd_ok(struct vmw_private *dev_priv,
  *
  * @sw_context: The software context used for this command submission batch.
  * @bo: The buffer object to add.
+ * @validate_as_mob: Validate this buffer as a MOB.
  * @p_val_node: If non-NULL Will be updated with the validate node number
  * on return.
  *
@@ -232,6 +265,7 @@ static int vmw_cmd_ok(struct vmw_private *dev_priv,
  */
 static int vmw_bo_to_validate_list(struct vmw_sw_context *sw_context,
 				   struct ttm_buffer_object *bo,
+				   bool validate_as_mob,
 				   uint32_t *p_val_node)
 {
 	uint32_t val_node;
@@ -244,6 +278,10 @@ static int vmw_bo_to_validate_list(struct vmw_sw_context *sw_context,
 				    &hash) == 0)) {
 		vval_buf = container_of(hash, struct vmw_validate_buffer,
 					hash);
+		if (unlikely(vval_buf->validate_as_mob != validate_as_mob)) {
+			DRM_ERROR("Inconsistent buffer usage.\n");
+			return -EINVAL;
+		}
 		val_buf = &vval_buf->base;
 		val_node = vval_buf - sw_context->val_bufs;
 	} else {
@@ -266,6 +304,7 @@ static int vmw_bo_to_validate_list(struct vmw_sw_context *sw_context,
 		val_buf->bo = ttm_bo_reference(bo);
 		val_buf->reserved = false;
 		list_add_tail(&val_buf->head, &sw_context->validate_nodes);
+		vval_buf->validate_as_mob = validate_as_mob;
 	}
 
 	sw_context->fence_flags |= DRM_VMW_FENCE_FLAG_EXEC;
@@ -302,7 +341,8 @@ static int vmw_resources_reserve(struct vmw_sw_context *sw_context)
 			struct ttm_buffer_object *bo = &res->backup->base;
 
 			ret = vmw_bo_to_validate_list
-				(sw_context, bo, NULL);
+				(sw_context, bo,
+				 vmw_resource_needs_backup(res), NULL);
 
 			if (unlikely(ret != 0))
 				return ret;
@@ -362,8 +402,15 @@ static int vmw_cmd_res_check(struct vmw_private *dev_priv,
 	struct vmw_resource_val_node *node;
 	int ret;
 
-	if (*id == SVGA3D_INVALID_ID)
+	if (*id == SVGA3D_INVALID_ID) {
+		if (p_val)
+			*p_val = NULL;
+		if (res_type == vmw_res_context) {
+			DRM_ERROR("Illegal context invalid id.\n");
+			return -EINVAL;
+		}
 		return 0;
+	}
 
 	/*
 	 * Fastpath in case of repeated commands referencing the same
@@ -411,6 +458,18 @@ static int vmw_cmd_res_check(struct vmw_private *dev_priv,
 	rcache->node = node;
 	if (p_val)
 		*p_val = node;
+
+	if (node->first_usage && res_type == vmw_res_context) {
+		node->staged_bindings =
+			kzalloc(sizeof(*node->staged_bindings), GFP_KERNEL);
+		if (node->staged_bindings == NULL) {
+			DRM_ERROR("Failed to allocate context binding "
+				  "information.\n");
+			goto out_no_reloc;
+		}
+		INIT_LIST_HEAD(&node->staged_bindings->list);
+	}
+
 	vmw_resource_unreference(&res);
 	return 0;
 
@@ -453,17 +512,35 @@ static int vmw_cmd_set_render_target_check(struct vmw_private *dev_priv,
 		SVGA3dCmdHeader header;
 		SVGA3dCmdSetRenderTarget body;
 	} *cmd;
+	struct vmw_resource_val_node *ctx_node;
+	struct vmw_resource_val_node *res_node;
 	int ret;
 
-	ret = vmw_cmd_cid_check(dev_priv, sw_context, header);
+	cmd = container_of(header, struct vmw_sid_cmd, header);
+
+	ret = vmw_cmd_res_check(dev_priv, sw_context, vmw_res_context,
+				user_context_converter, &cmd->body.cid,
+				&ctx_node);
 	if (unlikely(ret != 0))
 		return ret;
 
-	cmd = container_of(header, struct vmw_sid_cmd, header);
 	ret = vmw_cmd_res_check(dev_priv, sw_context, vmw_res_surface,
 				user_surface_converter,
-				&cmd->body.target.sid, NULL);
-	return ret;
+				&cmd->body.target.sid, &res_node);
+	if (unlikely(ret != 0))
+		return ret;
+
+	if (dev_priv->has_mob) {
+		struct vmw_ctx_bindinfo bi;
+
+		bi.ctx = ctx_node->res;
+		bi.res = res_node ? res_node->res : NULL;
+		bi.bt = vmw_ctx_binding_rt;
+		bi.i1.rt_type = cmd->body.type;
+		return vmw_context_binding_add(ctx_node->staged_bindings, &bi);
+	}
+
+	return 0;
 }
 
 static int vmw_cmd_surface_copy_check(struct vmw_private *dev_priv,
@@ -519,11 +596,6 @@ static int vmw_cmd_blt_surf_screen_check(struct vmw_private *dev_priv,
 
 	cmd = container_of(header, struct vmw_sid_cmd, header);
 
-	if (unlikely(!sw_context->kernel)) {
-		DRM_ERROR("Kernel only SVGA3d command: %u.\n", cmd->header.id);
-		return -EPERM;
-	}
-
 	return vmw_cmd_res_check(dev_priv, sw_context, vmw_res_surface,
 				 user_surface_converter,
 				 &cmd->body.srcImage.sid, NULL);
@@ -541,11 +613,6 @@ static int vmw_cmd_present_check(struct vmw_private *dev_priv,
 
 	cmd = container_of(header, struct vmw_sid_cmd, header);
 
-	if (unlikely(!sw_context->kernel)) {
-		DRM_ERROR("Kernel only SVGA3d command: %u.\n", cmd->header.id);
-		return -EPERM;
-	}
-
 	return vmw_cmd_res_check(dev_priv, sw_context, vmw_res_surface,
 				 user_surface_converter, &cmd->body.sid,
 				 NULL);
@@ -586,7 +653,7 @@ static int vmw_query_bo_switch_prepare(struct vmw_private *dev_priv,
 			sw_context->needs_post_query_barrier = true;
 			ret = vmw_bo_to_validate_list(sw_context,
 						      sw_context->cur_query_bo,
-						      NULL);
+						      dev_priv->has_mob, NULL);
 			if (unlikely(ret != 0))
 				return ret;
 		}
@@ -594,7 +661,7 @@ static int vmw_query_bo_switch_prepare(struct vmw_private *dev_priv,
 
 		ret = vmw_bo_to_validate_list(sw_context,
 					      dev_priv->dummy_query_bo,
-					      NULL);
+					      dev_priv->has_mob, NULL);
 		if (unlikely(ret != 0))
 			return ret;
 
@@ -671,6 +738,66 @@ static void vmw_query_bo_switch_commit(struct vmw_private *dev_priv,
 	}
 }
 
+/**
+ * vmw_translate_mob_pointer - Prepare to translate a user-space buffer
+ * handle to a MOB id.
+ *
+ * @dev_priv: Pointer to a device private structure.
+ * @sw_context: The software context used for this command batch validation.
+ * @id: Pointer to the user-space handle to be translated.
+ * @vmw_bo_p: Points to a location that, on successful return will carry
+ * a reference-counted pointer to the DMA buffer identified by the
+ * user-space handle in @id.
+ *
+ * This function saves information needed to translate a user-space buffer
+ * handle to a MOB id. The translation does not take place immediately, but
+ * during a call to vmw_apply_relocations(). This function builds a relocation
+ * list and a list of buffers to validate. The former needs to be freed using
+ * either vmw_apply_relocations() or vmw_free_relocations(). The latter
+ * needs to be freed using vmw_clear_validations.
+ */
+static int vmw_translate_mob_ptr(struct vmw_private *dev_priv,
+				 struct vmw_sw_context *sw_context,
+				 SVGAMobId *id,
+				 struct vmw_dma_buffer **vmw_bo_p)
+{
+	struct vmw_dma_buffer *vmw_bo = NULL;
+	struct ttm_buffer_object *bo;
+	uint32_t handle = *id;
+	struct vmw_relocation *reloc;
+	int ret;
+
+	ret = vmw_user_dmabuf_lookup(sw_context->tfile, handle, &vmw_bo);
+	if (unlikely(ret != 0)) {
+		DRM_ERROR("Could not find or use MOB buffer.\n");
+		return -EINVAL;
+	}
+	bo = &vmw_bo->base;
+
+	if (unlikely(sw_context->cur_reloc >= VMWGFX_MAX_RELOCATIONS)) {
+		DRM_ERROR("Max number relocations per submission"
+			  " exceeded\n");
+		ret = -EINVAL;
+		goto out_no_reloc;
+	}
+
+	reloc = &sw_context->relocs[sw_context->cur_reloc++];
+	reloc->mob_loc = id;
+	reloc->location = NULL;
+
+	ret = vmw_bo_to_validate_list(sw_context, bo, true, &reloc->index);
+	if (unlikely(ret != 0))
+		goto out_no_reloc;
+
+	*vmw_bo_p = vmw_bo;
+	return 0;
+
+out_no_reloc:
+	vmw_dmabuf_unreference(&vmw_bo);
+	vmw_bo_p = NULL;
+	return ret;
+}
+
 /**
  * vmw_translate_guest_pointer - Prepare to translate a user-space buffer
  * handle to a valid SVGAGuestPtr
@@ -718,7 +845,7 @@ static int vmw_translate_guest_ptr(struct vmw_private *dev_priv,
 	reloc = &sw_context->relocs[sw_context->cur_reloc++];
 	reloc->location = ptr;
 
-	ret = vmw_bo_to_validate_list(sw_context, bo, &reloc->index);
+	ret = vmw_bo_to_validate_list(sw_context, bo, false, &reloc->index);
 	if (unlikely(ret != 0))
 		goto out_no_reloc;
 
@@ -731,6 +858,30 @@ out_no_reloc:
 	return ret;
 }
 
+/**
+ * vmw_cmd_begin_gb_query - validate a  SVGA_3D_CMD_BEGIN_GB_QUERY command.
+ *
+ * @dev_priv: Pointer to a device private struct.
+ * @sw_context: The software context used for this command submission.
+ * @header: Pointer to the command header in the command stream.
+ */
+static int vmw_cmd_begin_gb_query(struct vmw_private *dev_priv,
+				  struct vmw_sw_context *sw_context,
+				  SVGA3dCmdHeader *header)
+{
+	struct vmw_begin_gb_query_cmd {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdBeginGBQuery q;
+	} *cmd;
+
+	cmd = container_of(header, struct vmw_begin_gb_query_cmd,
+			   header);
+
+	return vmw_cmd_res_check(dev_priv, sw_context, vmw_res_context,
+				 user_context_converter, &cmd->q.cid,
+				 NULL);
+}
+
 /**
  * vmw_cmd_begin_query - validate a  SVGA_3D_CMD_BEGIN_QUERY command.
  *
@@ -750,11 +901,63 @@ static int vmw_cmd_begin_query(struct vmw_private *dev_priv,
 	cmd = container_of(header, struct vmw_begin_query_cmd,
 			   header);
 
+	if (unlikely(dev_priv->has_mob)) {
+		struct {
+			SVGA3dCmdHeader header;
+			SVGA3dCmdBeginGBQuery q;
+		} gb_cmd;
+
+		BUG_ON(sizeof(gb_cmd) != sizeof(*cmd));
+
+		gb_cmd.header.id = SVGA_3D_CMD_BEGIN_GB_QUERY;
+		gb_cmd.header.size = cmd->header.size;
+		gb_cmd.q.cid = cmd->q.cid;
+		gb_cmd.q.type = cmd->q.type;
+
+		memcpy(cmd, &gb_cmd, sizeof(*cmd));
+		return vmw_cmd_begin_gb_query(dev_priv, sw_context, header);
+	}
+
 	return vmw_cmd_res_check(dev_priv, sw_context, vmw_res_context,
 				 user_context_converter, &cmd->q.cid,
 				 NULL);
 }
 
+/**
+ * vmw_cmd_end_gb_query - validate a  SVGA_3D_CMD_END_GB_QUERY command.
+ *
+ * @dev_priv: Pointer to a device private struct.
+ * @sw_context: The software context used for this command submission.
+ * @header: Pointer to the command header in the command stream.
+ */
+static int vmw_cmd_end_gb_query(struct vmw_private *dev_priv,
+				struct vmw_sw_context *sw_context,
+				SVGA3dCmdHeader *header)
+{
+	struct vmw_dma_buffer *vmw_bo;
+	struct vmw_query_cmd {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdEndGBQuery q;
+	} *cmd;
+	int ret;
+
+	cmd = container_of(header, struct vmw_query_cmd, header);
+	ret = vmw_cmd_cid_check(dev_priv, sw_context, header);
+	if (unlikely(ret != 0))
+		return ret;
+
+	ret = vmw_translate_mob_ptr(dev_priv, sw_context,
+				    &cmd->q.mobid,
+				    &vmw_bo);
+	if (unlikely(ret != 0))
+		return ret;
+
+	ret = vmw_query_bo_switch_prepare(dev_priv, &vmw_bo->base, sw_context);
+
+	vmw_dmabuf_unreference(&vmw_bo);
+	return ret;
+}
+
 /**
  * vmw_cmd_end_query - validate a  SVGA_3D_CMD_END_QUERY command.
  *
@@ -774,6 +977,25 @@ static int vmw_cmd_end_query(struct vmw_private *dev_priv,
 	int ret;
 
 	cmd = container_of(header, struct vmw_query_cmd, header);
+	if (dev_priv->has_mob) {
+		struct {
+			SVGA3dCmdHeader header;
+			SVGA3dCmdEndGBQuery q;
+		} gb_cmd;
+
+		BUG_ON(sizeof(gb_cmd) != sizeof(*cmd));
+
+		gb_cmd.header.id = SVGA_3D_CMD_END_GB_QUERY;
+		gb_cmd.header.size = cmd->header.size;
+		gb_cmd.q.cid = cmd->q.cid;
+		gb_cmd.q.type = cmd->q.type;
+		gb_cmd.q.mobid = cmd->q.guestResult.gmrId;
+		gb_cmd.q.offset = cmd->q.guestResult.offset;
+
+		memcpy(cmd, &gb_cmd, sizeof(*cmd));
+		return vmw_cmd_end_gb_query(dev_priv, sw_context, header);
+	}
+
 	ret = vmw_cmd_cid_check(dev_priv, sw_context, header);
 	if (unlikely(ret != 0))
 		return ret;
@@ -790,7 +1012,40 @@ static int vmw_cmd_end_query(struct vmw_private *dev_priv,
 	return ret;
 }
 
-/*
+/**
+ * vmw_cmd_wait_gb_query - validate a  SVGA_3D_CMD_WAIT_GB_QUERY command.
+ *
+ * @dev_priv: Pointer to a device private struct.
+ * @sw_context: The software context used for this command submission.
+ * @header: Pointer to the command header in the command stream.
+ */
+static int vmw_cmd_wait_gb_query(struct vmw_private *dev_priv,
+				 struct vmw_sw_context *sw_context,
+				 SVGA3dCmdHeader *header)
+{
+	struct vmw_dma_buffer *vmw_bo;
+	struct vmw_query_cmd {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdWaitForGBQuery q;
+	} *cmd;
+	int ret;
+
+	cmd = container_of(header, struct vmw_query_cmd, header);
+	ret = vmw_cmd_cid_check(dev_priv, sw_context, header);
+	if (unlikely(ret != 0))
+		return ret;
+
+	ret = vmw_translate_mob_ptr(dev_priv, sw_context,
+				    &cmd->q.mobid,
+				    &vmw_bo);
+	if (unlikely(ret != 0))
+		return ret;
+
+	vmw_dmabuf_unreference(&vmw_bo);
+	return 0;
+}
+
+/**
  * vmw_cmd_wait_query - validate a  SVGA_3D_CMD_WAIT_QUERY command.
  *
  * @dev_priv: Pointer to a device private struct.
@@ -809,6 +1064,25 @@ static int vmw_cmd_wait_query(struct vmw_private *dev_priv,
 	int ret;
 
 	cmd = container_of(header, struct vmw_query_cmd, header);
+	if (dev_priv->has_mob) {
+		struct {
+			SVGA3dCmdHeader header;
+			SVGA3dCmdWaitForGBQuery q;
+		} gb_cmd;
+
+		BUG_ON(sizeof(gb_cmd) != sizeof(*cmd));
+
+		gb_cmd.header.id = SVGA_3D_CMD_WAIT_FOR_GB_QUERY;
+		gb_cmd.header.size = cmd->header.size;
+		gb_cmd.q.cid = cmd->q.cid;
+		gb_cmd.q.type = cmd->q.type;
+		gb_cmd.q.mobid = cmd->q.guestResult.gmrId;
+		gb_cmd.q.offset = cmd->q.guestResult.offset;
+
+		memcpy(cmd, &gb_cmd, sizeof(*cmd));
+		return vmw_cmd_wait_gb_query(dev_priv, sw_context, header);
+	}
+
 	ret = vmw_cmd_cid_check(dev_priv, sw_context, header);
 	if (unlikely(ret != 0))
 		return ret;
@@ -921,15 +1195,22 @@ static int vmw_cmd_tex_state(struct vmw_private *dev_priv,
 	struct vmw_tex_state_cmd {
 		SVGA3dCmdHeader header;
 		SVGA3dCmdSetTextureState state;
-	};
+	} *cmd;
 
 	SVGA3dTextureState *last_state = (SVGA3dTextureState *)
 	  ((unsigned long) header + header->size + sizeof(header));
 	SVGA3dTextureState *cur_state = (SVGA3dTextureState *)
 		((unsigned long) header + sizeof(struct vmw_tex_state_cmd));
+	struct vmw_resource_val_node *ctx_node;
+	struct vmw_resource_val_node *res_node;
 	int ret;
 
-	ret = vmw_cmd_cid_check(dev_priv, sw_context, header);
+	cmd = container_of(header, struct vmw_tex_state_cmd,
+			   header);
+
+	ret = vmw_cmd_res_check(dev_priv, sw_context, vmw_res_context,
+				user_context_converter, &cmd->state.cid,
+				&ctx_node);
 	if (unlikely(ret != 0))
 		return ret;
 
@@ -939,9 +1220,20 @@ static int vmw_cmd_tex_state(struct vmw_private *dev_priv,
 
 		ret = vmw_cmd_res_check(dev_priv, sw_context, vmw_res_surface,
 					user_surface_converter,
-					&cur_state->value, NULL);
+					&cur_state->value, &res_node);
 		if (unlikely(ret != 0))
 			return ret;
+
+		if (dev_priv->has_mob) {
+			struct vmw_ctx_bindinfo bi;
+
+			bi.ctx = ctx_node->res;
+			bi.res = res_node ? res_node->res : NULL;
+			bi.bt = vmw_ctx_binding_tex;
+			bi.i1.texture_stage = cur_state->stage;
+			vmw_context_binding_add(ctx_node->staged_bindings,
+						&bi);
+		}
 	}
 
 	return 0;
@@ -970,6 +1262,222 @@ static int vmw_cmd_check_define_gmrfb(struct vmw_private *dev_priv,
 	return ret;
 }
 
+/**
+ * vmw_cmd_switch_backup - Utility function to handle backup buffer switching
+ *
+ * @dev_priv: Pointer to a device private struct.
+ * @sw_context: The software context being used for this batch.
+ * @res_type: The resource type.
+ * @converter: Information about user-space binding for this resource type.
+ * @res_id: Pointer to the user-space resource handle in the command stream.
+ * @buf_id: Pointer to the user-space backup buffer handle in the command
+ * stream.
+ * @backup_offset: Offset of backup into MOB.
+ *
+ * This function prepares for registering a switch of backup buffers
+ * in the resource metadata just prior to unreserving.
+ */
+static int vmw_cmd_switch_backup(struct vmw_private *dev_priv,
+				 struct vmw_sw_context *sw_context,
+				 enum vmw_res_type res_type,
+				 const struct vmw_user_resource_conv
+				 *converter,
+				 uint32_t *res_id,
+				 uint32_t *buf_id,
+				 unsigned long backup_offset)
+{
+	int ret;
+	struct vmw_dma_buffer *dma_buf;
+	struct vmw_resource_val_node *val_node;
+
+	ret = vmw_cmd_res_check(dev_priv, sw_context, res_type,
+				converter, res_id, &val_node);
+	if (unlikely(ret != 0))
+		return ret;
+
+	ret = vmw_translate_mob_ptr(dev_priv, sw_context, buf_id, &dma_buf);
+	if (unlikely(ret != 0))
+		return ret;
+
+	if (val_node->first_usage)
+		val_node->no_buffer_needed = true;
+
+	vmw_dmabuf_unreference(&val_node->new_backup);
+	val_node->new_backup = dma_buf;
+	val_node->new_backup_offset = backup_offset;
+
+	return 0;
+}
+
+/**
+ * vmw_cmd_bind_gb_surface - Validate an SVGA_3D_CMD_BIND_GB_SURFACE
+ * command
+ *
+ * @dev_priv: Pointer to a device private struct.
+ * @sw_context: The software context being used for this batch.
+ * @header: Pointer to the command header in the command stream.
+ */
+static int vmw_cmd_bind_gb_surface(struct vmw_private *dev_priv,
+				   struct vmw_sw_context *sw_context,
+				   SVGA3dCmdHeader *header)
+{
+	struct vmw_bind_gb_surface_cmd {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdBindGBSurface body;
+	} *cmd;
+
+	cmd = container_of(header, struct vmw_bind_gb_surface_cmd, header);
+
+	return vmw_cmd_switch_backup(dev_priv, sw_context, vmw_res_surface,
+				     user_surface_converter,
+				     &cmd->body.sid, &cmd->body.mobid,
+				     0);
+}
+
+/**
+ * vmw_cmd_update_gb_image - Validate an SVGA_3D_CMD_UPDATE_GB_IMAGE
+ * command
+ *
+ * @dev_priv: Pointer to a device private struct.
+ * @sw_context: The software context being used for this batch.
+ * @header: Pointer to the command header in the command stream.
+ */
+static int vmw_cmd_update_gb_image(struct vmw_private *dev_priv,
+				   struct vmw_sw_context *sw_context,
+				   SVGA3dCmdHeader *header)
+{
+	struct vmw_gb_surface_cmd {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdUpdateGBImage body;
+	} *cmd;
+
+	cmd = container_of(header, struct vmw_gb_surface_cmd, header);
+
+	return vmw_cmd_res_check(dev_priv, sw_context, vmw_res_surface,
+				 user_surface_converter,
+				 &cmd->body.image.sid, NULL);
+}
+
+/**
+ * vmw_cmd_update_gb_surface - Validate an SVGA_3D_CMD_UPDATE_GB_SURFACE
+ * command
+ *
+ * @dev_priv: Pointer to a device private struct.
+ * @sw_context: The software context being used for this batch.
+ * @header: Pointer to the command header in the command stream.
+ */
+static int vmw_cmd_update_gb_surface(struct vmw_private *dev_priv,
+				     struct vmw_sw_context *sw_context,
+				     SVGA3dCmdHeader *header)
+{
+	struct vmw_gb_surface_cmd {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdUpdateGBSurface body;
+	} *cmd;
+
+	cmd = container_of(header, struct vmw_gb_surface_cmd, header);
+
+	return vmw_cmd_res_check(dev_priv, sw_context, vmw_res_surface,
+				 user_surface_converter,
+				 &cmd->body.sid, NULL);
+}
+
+/**
+ * vmw_cmd_readback_gb_image - Validate an SVGA_3D_CMD_READBACK_GB_IMAGE
+ * command
+ *
+ * @dev_priv: Pointer to a device private struct.
+ * @sw_context: The software context being used for this batch.
+ * @header: Pointer to the command header in the command stream.
+ */
+static int vmw_cmd_readback_gb_image(struct vmw_private *dev_priv,
+				     struct vmw_sw_context *sw_context,
+				     SVGA3dCmdHeader *header)
+{
+	struct vmw_gb_surface_cmd {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdReadbackGBImage body;
+	} *cmd;
+
+	cmd = container_of(header, struct vmw_gb_surface_cmd, header);
+
+	return vmw_cmd_res_check(dev_priv, sw_context, vmw_res_surface,
+				 user_surface_converter,
+				 &cmd->body.image.sid, NULL);
+}
+
+/**
+ * vmw_cmd_readback_gb_surface - Validate an SVGA_3D_CMD_READBACK_GB_SURFACE
+ * command
+ *
+ * @dev_priv: Pointer to a device private struct.
+ * @sw_context: The software context being used for this batch.
+ * @header: Pointer to the command header in the command stream.
+ */
+static int vmw_cmd_readback_gb_surface(struct vmw_private *dev_priv,
+				       struct vmw_sw_context *sw_context,
+				       SVGA3dCmdHeader *header)
+{
+	struct vmw_gb_surface_cmd {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdReadbackGBSurface body;
+	} *cmd;
+
+	cmd = container_of(header, struct vmw_gb_surface_cmd, header);
+
+	return vmw_cmd_res_check(dev_priv, sw_context, vmw_res_surface,
+				 user_surface_converter,
+				 &cmd->body.sid, NULL);
+}
+
+/**
+ * vmw_cmd_invalidate_gb_image - Validate an SVGA_3D_CMD_INVALIDATE_GB_IMAGE
+ * command
+ *
+ * @dev_priv: Pointer to a device private struct.
+ * @sw_context: The software context being used for this batch.
+ * @header: Pointer to the command header in the command stream.
+ */
+static int vmw_cmd_invalidate_gb_image(struct vmw_private *dev_priv,
+				       struct vmw_sw_context *sw_context,
+				       SVGA3dCmdHeader *header)
+{
+	struct vmw_gb_surface_cmd {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdInvalidateGBImage body;
+	} *cmd;
+
+	cmd = container_of(header, struct vmw_gb_surface_cmd, header);
+
+	return vmw_cmd_res_check(dev_priv, sw_context, vmw_res_surface,
+				 user_surface_converter,
+				 &cmd->body.image.sid, NULL);
+}
+
+/**
+ * vmw_cmd_invalidate_gb_surface - Validate an
+ * SVGA_3D_CMD_INVALIDATE_GB_SURFACE command
+ *
+ * @dev_priv: Pointer to a device private struct.
+ * @sw_context: The software context being used for this batch.
+ * @header: Pointer to the command header in the command stream.
+ */
+static int vmw_cmd_invalidate_gb_surface(struct vmw_private *dev_priv,
+					 struct vmw_sw_context *sw_context,
+					 SVGA3dCmdHeader *header)
+{
+	struct vmw_gb_surface_cmd {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdInvalidateGBSurface body;
+	} *cmd;
+
+	cmd = container_of(header, struct vmw_gb_surface_cmd, header);
+
+	return vmw_cmd_res_check(dev_priv, sw_context, vmw_res_surface,
+				 user_surface_converter,
+				 &cmd->body.sid, NULL);
+}
+
 /**
  * vmw_cmd_set_shader - Validate an SVGA_3D_CMD_SET_SHADER
  * command
@@ -986,18 +1494,64 @@ static int vmw_cmd_set_shader(struct vmw_private *dev_priv,
 		SVGA3dCmdHeader header;
 		SVGA3dCmdSetShader body;
 	} *cmd;
+	struct vmw_resource_val_node *ctx_node;
 	int ret;
 
 	cmd = container_of(header, struct vmw_set_shader_cmd,
 			   header);
 
-	ret = vmw_cmd_cid_check(dev_priv, sw_context, header);
+	ret = vmw_cmd_res_check(dev_priv, sw_context, vmw_res_context,
+				user_context_converter, &cmd->body.cid,
+				&ctx_node);
 	if (unlikely(ret != 0))
 		return ret;
 
+	if (dev_priv->has_mob) {
+		struct vmw_ctx_bindinfo bi;
+		struct vmw_resource_val_node *res_node;
+
+		ret = vmw_cmd_res_check(dev_priv, sw_context, vmw_res_shader,
+					user_shader_converter,
+					&cmd->body.shid, &res_node);
+		if (unlikely(ret != 0))
+			return ret;
+
+		bi.ctx = ctx_node->res;
+		bi.res = res_node ? res_node->res : NULL;
+		bi.bt = vmw_ctx_binding_shader;
+		bi.i1.shader_type = cmd->body.type;
+		return vmw_context_binding_add(ctx_node->staged_bindings, &bi);
+	}
+
 	return 0;
 }
 
+/**
+ * vmw_cmd_bind_gb_shader - Validate an SVGA_3D_CMD_BIND_GB_SHADER
+ * command
+ *
+ * @dev_priv: Pointer to a device private struct.
+ * @sw_context: The software context being used for this batch.
+ * @header: Pointer to the command header in the command stream.
+ */
+static int vmw_cmd_bind_gb_shader(struct vmw_private *dev_priv,
+				  struct vmw_sw_context *sw_context,
+				  SVGA3dCmdHeader *header)
+{
+	struct vmw_bind_gb_shader_cmd {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdBindGBShader body;
+	} *cmd;
+
+	cmd = container_of(header, struct vmw_bind_gb_shader_cmd,
+			   header);
+
+	return vmw_cmd_switch_backup(dev_priv, sw_context, vmw_res_shader,
+				     user_shader_converter,
+				     &cmd->body.shid, &cmd->body.mobid,
+				     cmd->body.offsetInBytes);
+}
+
 static int vmw_cmd_check_not_3d(struct vmw_private *dev_priv,
 				struct vmw_sw_context *sw_context,
 				void *buf, uint32_t *size)
@@ -1041,50 +1595,173 @@ static int vmw_cmd_check_not_3d(struct vmw_private *dev_priv,
 	return 0;
 }
 
-typedef int (*vmw_cmd_func) (struct vmw_private *,
-			     struct vmw_sw_context *,
-			     SVGA3dCmdHeader *);
-
-#define VMW_CMD_DEF(cmd, func) \
-	[cmd - SVGA_3D_CMD_BASE] = func
-
-static vmw_cmd_func vmw_cmd_funcs[SVGA_3D_CMD_MAX] = {
-	VMW_CMD_DEF(SVGA_3D_CMD_SURFACE_DEFINE, &vmw_cmd_invalid),
-	VMW_CMD_DEF(SVGA_3D_CMD_SURFACE_DESTROY, &vmw_cmd_invalid),
-	VMW_CMD_DEF(SVGA_3D_CMD_SURFACE_COPY, &vmw_cmd_surface_copy_check),
-	VMW_CMD_DEF(SVGA_3D_CMD_SURFACE_STRETCHBLT, &vmw_cmd_stretch_blt_check),
-	VMW_CMD_DEF(SVGA_3D_CMD_SURFACE_DMA, &vmw_cmd_dma),
-	VMW_CMD_DEF(SVGA_3D_CMD_CONTEXT_DEFINE, &vmw_cmd_invalid),
-	VMW_CMD_DEF(SVGA_3D_CMD_CONTEXT_DESTROY, &vmw_cmd_invalid),
-	VMW_CMD_DEF(SVGA_3D_CMD_SETTRANSFORM, &vmw_cmd_cid_check),
-	VMW_CMD_DEF(SVGA_3D_CMD_SETZRANGE, &vmw_cmd_cid_check),
-	VMW_CMD_DEF(SVGA_3D_CMD_SETRENDERSTATE, &vmw_cmd_cid_check),
+static const struct vmw_cmd_entry const vmw_cmd_entries[SVGA_3D_CMD_MAX] = {
+	VMW_CMD_DEF(SVGA_3D_CMD_SURFACE_DEFINE, &vmw_cmd_invalid,
+		    false, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_SURFACE_DESTROY, &vmw_cmd_invalid,
+		    false, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_SURFACE_COPY, &vmw_cmd_surface_copy_check,
+		    true, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_SURFACE_STRETCHBLT, &vmw_cmd_stretch_blt_check,
+		    true, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_SURFACE_DMA, &vmw_cmd_dma,
+		    true, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_CONTEXT_DEFINE, &vmw_cmd_invalid,
+		    false, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_CONTEXT_DESTROY, &vmw_cmd_invalid,
+		    false, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_SETTRANSFORM, &vmw_cmd_cid_check,
+		    true, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_SETZRANGE, &vmw_cmd_cid_check,
+		    true, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_SETRENDERSTATE, &vmw_cmd_cid_check,
+		    true, false, false),
 	VMW_CMD_DEF(SVGA_3D_CMD_SETRENDERTARGET,
-		    &vmw_cmd_set_render_target_check),
-	VMW_CMD_DEF(SVGA_3D_CMD_SETTEXTURESTATE, &vmw_cmd_tex_state),
-	VMW_CMD_DEF(SVGA_3D_CMD_SETMATERIAL, &vmw_cmd_cid_check),
-	VMW_CMD_DEF(SVGA_3D_CMD_SETLIGHTDATA, &vmw_cmd_cid_check),
-	VMW_CMD_DEF(SVGA_3D_CMD_SETLIGHTENABLED, &vmw_cmd_cid_check),
-	VMW_CMD_DEF(SVGA_3D_CMD_SETVIEWPORT, &vmw_cmd_cid_check),
-	VMW_CMD_DEF(SVGA_3D_CMD_SETCLIPPLANE, &vmw_cmd_cid_check),
-	VMW_CMD_DEF(SVGA_3D_CMD_CLEAR, &vmw_cmd_cid_check),
-	VMW_CMD_DEF(SVGA_3D_CMD_PRESENT, &vmw_cmd_present_check),
-	VMW_CMD_DEF(SVGA_3D_CMD_SHADER_DEFINE, &vmw_cmd_cid_check),
-	VMW_CMD_DEF(SVGA_3D_CMD_SHADER_DESTROY, &vmw_cmd_cid_check),
-	VMW_CMD_DEF(SVGA_3D_CMD_SET_SHADER, &vmw_cmd_set_shader),
-	VMW_CMD_DEF(SVGA_3D_CMD_SET_SHADER_CONST, &vmw_cmd_cid_check),
-	VMW_CMD_DEF(SVGA_3D_CMD_DRAW_PRIMITIVES, &vmw_cmd_draw),
-	VMW_CMD_DEF(SVGA_3D_CMD_SETSCISSORRECT, &vmw_cmd_cid_check),
-	VMW_CMD_DEF(SVGA_3D_CMD_BEGIN_QUERY, &vmw_cmd_begin_query),
-	VMW_CMD_DEF(SVGA_3D_CMD_END_QUERY, &vmw_cmd_end_query),
-	VMW_CMD_DEF(SVGA_3D_CMD_WAIT_FOR_QUERY, &vmw_cmd_wait_query),
-	VMW_CMD_DEF(SVGA_3D_CMD_PRESENT_READBACK, &vmw_cmd_ok),
+		    &vmw_cmd_set_render_target_check, true, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_SETTEXTURESTATE, &vmw_cmd_tex_state,
+		    true, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_SETMATERIAL, &vmw_cmd_cid_check,
+		    true, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_SETLIGHTDATA, &vmw_cmd_cid_check,
+		    true, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_SETLIGHTENABLED, &vmw_cmd_cid_check,
+		    true, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_SETVIEWPORT, &vmw_cmd_cid_check,
+		    true, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_SETCLIPPLANE, &vmw_cmd_cid_check,
+		    true, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_CLEAR, &vmw_cmd_cid_check,
+		    true, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_PRESENT, &vmw_cmd_present_check,
+		    false, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_SHADER_DEFINE, &vmw_cmd_cid_check,
+		    true, true, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_SHADER_DESTROY, &vmw_cmd_cid_check,
+		    true, true, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_SET_SHADER, &vmw_cmd_set_shader,
+		    true, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_SET_SHADER_CONST, &vmw_cmd_cid_check,
+		    true, true, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_DRAW_PRIMITIVES, &vmw_cmd_draw,
+		    true, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_SETSCISSORRECT, &vmw_cmd_cid_check,
+		    true, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_BEGIN_QUERY, &vmw_cmd_begin_query,
+		    true, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_END_QUERY, &vmw_cmd_end_query,
+		    true, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_WAIT_FOR_QUERY, &vmw_cmd_wait_query,
+		    true, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_PRESENT_READBACK, &vmw_cmd_ok,
+		    true, false, false),
 	VMW_CMD_DEF(SVGA_3D_CMD_BLIT_SURFACE_TO_SCREEN,
-		    &vmw_cmd_blt_surf_screen_check),
-	VMW_CMD_DEF(SVGA_3D_CMD_SURFACE_DEFINE_V2, &vmw_cmd_invalid),
-	VMW_CMD_DEF(SVGA_3D_CMD_GENERATE_MIPMAPS, &vmw_cmd_invalid),
-	VMW_CMD_DEF(SVGA_3D_CMD_ACTIVATE_SURFACE, &vmw_cmd_invalid),
-	VMW_CMD_DEF(SVGA_3D_CMD_DEACTIVATE_SURFACE, &vmw_cmd_invalid),
+		    &vmw_cmd_blt_surf_screen_check, false, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_SURFACE_DEFINE_V2, &vmw_cmd_invalid,
+		    false, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_GENERATE_MIPMAPS, &vmw_cmd_invalid,
+		    false, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_ACTIVATE_SURFACE, &vmw_cmd_invalid,
+		    false, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_DEACTIVATE_SURFACE, &vmw_cmd_invalid,
+		    false, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_SCREEN_DMA, &vmw_cmd_invalid,
+		    false, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_SET_UNITY_SURFACE_COOKIE, &vmw_cmd_invalid,
+		    false, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_OPEN_CONTEXT_SURFACE, &vmw_cmd_invalid,
+		    false, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_LOGICOPS_BITBLT, &vmw_cmd_invalid,
+		    false, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_LOGICOPS_TRANSBLT, &vmw_cmd_invalid,
+		    false, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_LOGICOPS_STRETCHBLT, &vmw_cmd_invalid,
+		    false, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_LOGICOPS_COLORFILL, &vmw_cmd_invalid,
+		    false, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_LOGICOPS_ALPHABLEND, &vmw_cmd_invalid,
+		    false, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_LOGICOPS_CLEARTYPEBLEND, &vmw_cmd_invalid,
+		    false, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_SET_OTABLE_BASE, &vmw_cmd_invalid,
+		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_READBACK_OTABLE, &vmw_cmd_invalid,
+		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DEFINE_GB_MOB, &vmw_cmd_invalid,
+		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DESTROY_GB_MOB, &vmw_cmd_invalid,
+		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_REDEFINE_GB_MOB, &vmw_cmd_invalid,
+		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_UPDATE_GB_MOB_MAPPING, &vmw_cmd_invalid,
+		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DEFINE_GB_SURFACE, &vmw_cmd_invalid,
+		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DESTROY_GB_SURFACE, &vmw_cmd_invalid,
+		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_BIND_GB_SURFACE, &vmw_cmd_bind_gb_surface,
+		    true, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_COND_BIND_GB_SURFACE, &vmw_cmd_invalid,
+		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_UPDATE_GB_IMAGE, &vmw_cmd_update_gb_image,
+		    true, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_UPDATE_GB_SURFACE,
+		    &vmw_cmd_update_gb_surface, true, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_READBACK_GB_IMAGE,
+		    &vmw_cmd_readback_gb_image, true, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_READBACK_GB_SURFACE,
+		    &vmw_cmd_readback_gb_surface, true, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_INVALIDATE_GB_IMAGE,
+		    &vmw_cmd_invalidate_gb_image, true, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_INVALIDATE_GB_SURFACE,
+		    &vmw_cmd_invalidate_gb_surface, true, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DEFINE_GB_CONTEXT, &vmw_cmd_invalid,
+		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DESTROY_GB_CONTEXT, &vmw_cmd_invalid,
+		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_BIND_GB_CONTEXT, &vmw_cmd_invalid,
+		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_READBACK_GB_CONTEXT, &vmw_cmd_invalid,
+		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_INVALIDATE_GB_CONTEXT, &vmw_cmd_invalid,
+		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DEFINE_GB_SHADER, &vmw_cmd_invalid,
+		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_BIND_GB_SHADER, &vmw_cmd_bind_gb_shader,
+		    true, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DESTROY_GB_SHADER, &vmw_cmd_invalid,
+		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_SET_OTABLE_BASE64, &vmw_cmd_invalid,
+		    false, false, false),
+	VMW_CMD_DEF(SVGA_3D_CMD_BEGIN_GB_QUERY, &vmw_cmd_begin_gb_query,
+		    true, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_END_GB_QUERY, &vmw_cmd_end_gb_query,
+		    true, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_WAIT_FOR_GB_QUERY, &vmw_cmd_wait_gb_query,
+		    true, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_NOP, &vmw_cmd_ok,
+		    true, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_ENABLE_GART, &vmw_cmd_invalid,
+		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DISABLE_GART, &vmw_cmd_invalid,
+		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_MAP_MOB_INTO_GART, &vmw_cmd_invalid,
+		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_UNMAP_GART_RANGE, &vmw_cmd_invalid,
+		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DEFINE_GB_SCREENTARGET, &vmw_cmd_invalid,
+		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DESTROY_GB_SCREENTARGET, &vmw_cmd_invalid,
+		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_BIND_GB_SCREENTARGET, &vmw_cmd_invalid,
+		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_UPDATE_GB_SCREENTARGET, &vmw_cmd_invalid,
+		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_READBACK_GB_IMAGE_PARTIAL, &vmw_cmd_invalid,
+		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_INVALIDATE_GB_IMAGE_PARTIAL, &vmw_cmd_invalid,
+		    false, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_SET_GB_SHADERCONSTS_INLINE, &vmw_cmd_cid_check,
+		    true, false, true)
 };
 
 static int vmw_cmd_check(struct vmw_private *dev_priv,
@@ -1095,6 +1772,8 @@ static int vmw_cmd_check(struct vmw_private *dev_priv,
 	uint32_t size_remaining = *size;
 	SVGA3dCmdHeader *header = (SVGA3dCmdHeader *) buf;
 	int ret;
+	const struct vmw_cmd_entry *entry;
+	bool gb = dev_priv->capabilities & SVGA_CAP_GBOBJECTS;
 
 	cmd_id = le32_to_cpu(((uint32_t *)buf)[0]);
 	/* Handle any none 3D commands */
@@ -1107,18 +1786,40 @@ static int vmw_cmd_check(struct vmw_private *dev_priv,
 
 	cmd_id -= SVGA_3D_CMD_BASE;
 	if (unlikely(*size > size_remaining))
-		goto out_err;
+		goto out_invalid;
 
 	if (unlikely(cmd_id >= SVGA_3D_CMD_MAX - SVGA_3D_CMD_BASE))
-		goto out_err;
+		goto out_invalid;
+
+	entry = &vmw_cmd_entries[cmd_id];
+	if (unlikely(!entry->user_allow && !sw_context->kernel))
+		goto out_privileged;
 
-	ret = vmw_cmd_funcs[cmd_id](dev_priv, sw_context, header);
+	if (unlikely(entry->gb_disable && gb))
+		goto out_old;
+
+	if (unlikely(entry->gb_enable && !gb))
+		goto out_new;
+
+	ret = entry->func(dev_priv, sw_context, header);
 	if (unlikely(ret != 0))
-		goto out_err;
+		goto out_invalid;
 
 	return 0;
-out_err:
-	DRM_ERROR("Illegal / Invalid SVGA3D command: %d\n",
+out_invalid:
+	DRM_ERROR("Invalid SVGA3D command: %d\n",
+		  cmd_id + SVGA_3D_CMD_BASE);
+	return -EINVAL;
+out_privileged:
+	DRM_ERROR("Privileged SVGA3D command: %d\n",
+		  cmd_id + SVGA_3D_CMD_BASE);
+	return -EPERM;
+out_old:
+	DRM_ERROR("Deprecated (disallowed) SVGA3D command: %d\n",
+		  cmd_id + SVGA_3D_CMD_BASE);
+	return -EINVAL;
+out_new:
+	DRM_ERROR("SVGA3D command: %d not supported by virtual hardware.\n",
 		  cmd_id + SVGA_3D_CMD_BASE);
 	return -EINVAL;
 }
@@ -1174,6 +1875,9 @@ static void vmw_apply_relocations(struct vmw_sw_context *sw_context)
 		case VMW_PL_GMR:
 			reloc->location->gmrId = bo->mem.start;
 			break;
+		case VMW_PL_MOB:
+			*reloc->mob_loc = bo->mem.start;
+			break;
 		default:
 			BUG();
 		}
@@ -1198,6 +1902,8 @@ static void vmw_resource_list_unreference(struct list_head *list)
 	list_for_each_entry_safe(val, val_next, list, head) {
 		list_del_init(&val->head);
 		vmw_resource_unreference(&val->res);
+		if (unlikely(val->staged_bindings))
+			kfree(val->staged_bindings);
 		kfree(val);
 	}
 }
@@ -1224,7 +1930,8 @@ static void vmw_clear_validations(struct vmw_sw_context *sw_context)
 }
 
 static int vmw_validate_single_buffer(struct vmw_private *dev_priv,
-				      struct ttm_buffer_object *bo)
+				      struct ttm_buffer_object *bo,
+				      bool validate_as_mob)
 {
 	int ret;
 
@@ -1238,6 +1945,9 @@ static int vmw_validate_single_buffer(struct vmw_private *dev_priv,
 	     dev_priv->dummy_query_bo_pinned))
 		return 0;
 
+	if (validate_as_mob)
+		return ttm_bo_validate(bo, &vmw_mob_placement, true, false);
+
 	/**
 	 * 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,
@@ -1259,7 +1969,6 @@ static int vmw_validate_single_buffer(struct vmw_private *dev_priv,
 	return ret;
 }
 
-
 static int vmw_validate_buffers(struct vmw_private *dev_priv,
 				struct vmw_sw_context *sw_context)
 {
@@ -1267,7 +1976,8 @@ static int vmw_validate_buffers(struct vmw_private *dev_priv,
 	int ret;
 
 	list_for_each_entry(entry, &sw_context->validate_nodes, base.head) {
-		ret = vmw_validate_single_buffer(dev_priv, entry->base.bo);
+		ret = vmw_validate_single_buffer(dev_priv, entry->base.bo,
+						 entry->validate_as_mob);
 		if (unlikely(ret != 0))
 			return ret;
 	}
@@ -1509,11 +2219,17 @@ int vmw_execbuf_process(struct drm_file *file_priv,
 			goto out_err;
 	}
 
+	ret = mutex_lock_interruptible(&dev_priv->binding_mutex);
+	if (unlikely(ret != 0)) {
+		ret = -ERESTARTSYS;
+		goto out_err;
+	}
+
 	cmd = vmw_fifo_reserve(dev_priv, command_size);
 	if (unlikely(cmd == NULL)) {
 		DRM_ERROR("Failed reserving fifo space for commands.\n");
 		ret = -ENOMEM;
-		goto out_err;
+		goto out_unlock_binding;
 	}
 
 	vmw_apply_relocations(sw_context);
@@ -1538,6 +2254,8 @@ int vmw_execbuf_process(struct drm_file *file_priv,
 		DRM_ERROR("Fence submission error. Syncing.\n");
 
 	vmw_resource_list_unreserve(&sw_context->resource_list, false);
+	mutex_unlock(&dev_priv->binding_mutex);
+
 	ttm_eu_fence_buffer_objects(&ticket, &sw_context->validate_nodes,
 				    (void *) fence);
 
@@ -1568,6 +2286,8 @@ int vmw_execbuf_process(struct drm_file *file_priv,
 
 	return 0;
 
+out_unlock_binding:
+	mutex_unlock(&dev_priv->binding_mutex);
 out_err:
 	vmw_resource_relocations_free(&sw_context->res_relocations);
 	vmw_free_relocations(sw_context);

drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c

@@ -35,6 +35,23 @@ bool vmw_fifo_have_3d(struct vmw_private *dev_priv)
 	uint32_t fifo_min, hwversion;
 	const struct vmw_fifo_state *fifo = &dev_priv->fifo;
 
+	if (!(dev_priv->capabilities & SVGA_CAP_3D))
+		return false;
+
+	if (dev_priv->capabilities & SVGA_CAP_GBOBJECTS) {
+		uint32_t result;
+
+		if (!dev_priv->has_mob)
+			return false;
+
+		mutex_lock(&dev_priv->hw_mutex);
+		vmw_write(dev_priv, SVGA_REG_DEV_CAP, SVGA3D_DEVCAP_3D);
+		result = vmw_read(dev_priv, SVGA_REG_DEV_CAP);
+		mutex_unlock(&dev_priv->hw_mutex);
+
+		return (result != 0);
+	}
+
 	if (!(dev_priv->capabilities & SVGA_CAP_EXTENDED_FIFO))
 		return false;
 
@@ -511,24 +528,16 @@ out_err:
 }
 
 /**
- * vmw_fifo_emit_dummy_query - emits a dummy query to the fifo.
+ * vmw_fifo_emit_dummy_legacy_query - emits a dummy query to the fifo using
+ * legacy query commands.
  *
  * @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.
+ * See the vmw_fifo_emit_dummy_query documentation.
  */
-int vmw_fifo_emit_dummy_query(struct vmw_private *dev_priv,
-			      uint32_t cid)
+static int vmw_fifo_emit_dummy_legacy_query(struct vmw_private *dev_priv,
+					    uint32_t cid)
 {
 	/*
 	 * A query wait without a preceding query end will
@@ -566,3 +575,75 @@ int vmw_fifo_emit_dummy_query(struct vmw_private *dev_priv,
 
 	return 0;
 }
+
+/**
+ * vmw_fifo_emit_dummy_gb_query - emits a dummy query to the fifo using
+ * guest-backed resource query commands.
+ *
+ * @dev_priv: The device private structure.
+ * @cid: The hardware context id used for the query.
+ *
+ * See the vmw_fifo_emit_dummy_query documentation.
+ */
+static int vmw_fifo_emit_dummy_gb_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;
+		SVGA3dCmdWaitForGBQuery 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_GB_QUERY;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.cid = cid;
+	cmd->body.type = SVGA3D_QUERYTYPE_OCCLUSION;
+	BUG_ON(bo->mem.mem_type != VMW_PL_MOB);
+	cmd->body.mobid = bo->mem.start;
+	cmd->body.offset = 0;
+
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+
+	return 0;
+}
+
+
+/**
+ * vmw_fifo_emit_dummy_gb_query - emits a dummy query to the fifo using
+ * appropriate resource query commands.
+ *
+ * @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)
+{
+	if (dev_priv->has_mob)
+		return vmw_fifo_emit_dummy_gb_query(dev_priv, cid);
+
+	return vmw_fifo_emit_dummy_legacy_query(dev_priv, cid);
+}

drivers/gpu/drm/vmwgfx/vmwgfx_gmr.c

@@ -125,181 +125,27 @@ static void vmw_gmr2_unbind(struct vmw_private *dev_priv,
 }
 
 
-static void vmw_gmr_free_descriptors(struct device *dev, dma_addr_t desc_dma,
-				     struct list_head *desc_pages)
-{
-	struct page *page, *next;
-	struct svga_guest_mem_descriptor *page_virtual;
-	unsigned int desc_per_page = PAGE_SIZE /
-		sizeof(struct svga_guest_mem_descriptor) - 1;
-
-	if (list_empty(desc_pages))
-		return;
-
-	list_for_each_entry_safe(page, next, desc_pages, lru) {
-		list_del_init(&page->lru);
-
-		if (likely(desc_dma != DMA_ADDR_INVALID)) {
-			dma_unmap_page(dev, desc_dma, PAGE_SIZE,
-				       DMA_TO_DEVICE);
-		}
-
-		page_virtual = kmap_atomic(page);
-		desc_dma = (dma_addr_t)
-			le32_to_cpu(page_virtual[desc_per_page].ppn) <<
-			PAGE_SHIFT;
-		kunmap_atomic(page_virtual);
-
-		__free_page(page);
-	}
-}
-
-/**
- * FIXME: Adjust to the ttm lowmem / highmem storage to minimize
- * the number of used descriptors.
- *
- */
-
-static int vmw_gmr_build_descriptors(struct device *dev,
-				     struct list_head *desc_pages,
-				     struct vmw_piter *iter,
-				     unsigned long num_pages,
-				     dma_addr_t *first_dma)
-{
-	struct page *page;
-	struct svga_guest_mem_descriptor *page_virtual = NULL;
-	struct svga_guest_mem_descriptor *desc_virtual = NULL;
-	unsigned int desc_per_page;
-	unsigned long prev_pfn;
-	unsigned long pfn;
-	int ret;
-	dma_addr_t desc_dma;
-
-	desc_per_page = PAGE_SIZE /
-	    sizeof(struct svga_guest_mem_descriptor) - 1;
-
-	while (likely(num_pages != 0)) {
-		page = alloc_page(__GFP_HIGHMEM);
-		if (unlikely(page == NULL)) {
-			ret = -ENOMEM;
-			goto out_err;
-		}
-
-		list_add_tail(&page->lru, desc_pages);
-		page_virtual = kmap_atomic(page);
-		desc_virtual = page_virtual - 1;
-		prev_pfn = ~(0UL);
-
-		while (likely(num_pages != 0)) {
-			pfn = vmw_piter_dma_addr(iter) >> PAGE_SHIFT;
-
-			if (pfn != prev_pfn + 1) {
-
-				if (desc_virtual - page_virtual ==
-				    desc_per_page - 1)
-					break;
-
-				(++desc_virtual)->ppn = cpu_to_le32(pfn);
-				desc_virtual->num_pages = cpu_to_le32(1);
-			} else {
-				uint32_t tmp =
-				    le32_to_cpu(desc_virtual->num_pages);
-				desc_virtual->num_pages = cpu_to_le32(tmp + 1);
-			}
-			prev_pfn = pfn;
-			--num_pages;
-			vmw_piter_next(iter);
-		}
-
-		(++desc_virtual)->ppn = DMA_PAGE_INVALID;
-		desc_virtual->num_pages = cpu_to_le32(0);
-		kunmap_atomic(page_virtual);
-	}
-
-	desc_dma = 0;
-	list_for_each_entry_reverse(page, desc_pages, lru) {
-		page_virtual = kmap_atomic(page);
-		page_virtual[desc_per_page].ppn = cpu_to_le32
-			(desc_dma >> PAGE_SHIFT);
-		kunmap_atomic(page_virtual);
-		desc_dma = dma_map_page(dev, page, 0, PAGE_SIZE,
-					DMA_TO_DEVICE);
-
-		if (unlikely(dma_mapping_error(dev, desc_dma)))
-			goto out_err;
-	}
-	*first_dma = desc_dma;
-
-	return 0;
-out_err:
-	vmw_gmr_free_descriptors(dev, DMA_ADDR_INVALID, desc_pages);
-	return ret;
-}
-
-static void vmw_gmr_fire_descriptors(struct vmw_private *dev_priv,
-				     int gmr_id, dma_addr_t desc_dma)
-{
-	mutex_lock(&dev_priv->hw_mutex);
-
-	vmw_write(dev_priv, SVGA_REG_GMR_ID, gmr_id);
-	wmb();
-	vmw_write(dev_priv, SVGA_REG_GMR_DESCRIPTOR, desc_dma >> PAGE_SHIFT);
-	mb();
-
-	mutex_unlock(&dev_priv->hw_mutex);
-
-}
-
 int vmw_gmr_bind(struct vmw_private *dev_priv,
 		 const struct vmw_sg_table *vsgt,
 		 unsigned long num_pages,
 		 int gmr_id)
 {
-	struct list_head desc_pages;
-	dma_addr_t desc_dma = 0;
-	struct device *dev = dev_priv->dev->dev;
 	struct vmw_piter data_iter;
-	int ret;
 
 	vmw_piter_start(&data_iter, vsgt, 0);
 
 	if (unlikely(!vmw_piter_next(&data_iter)))
 		return 0;
 
-	if (likely(dev_priv->capabilities & SVGA_CAP_GMR2))
-		return vmw_gmr2_bind(dev_priv, &data_iter, num_pages, gmr_id);
-
-	if (unlikely(!(dev_priv->capabilities & SVGA_CAP_GMR)))
-		return -EINVAL;
-
-	if (vsgt->num_regions > dev_priv->max_gmr_descriptors)
+	if (unlikely(!(dev_priv->capabilities & SVGA_CAP_GMR2)))
 		return -EINVAL;
 
-	INIT_LIST_HEAD(&desc_pages);
-
-	ret = vmw_gmr_build_descriptors(dev, &desc_pages, &data_iter,
-					num_pages, &desc_dma);
-	if (unlikely(ret != 0))
-		return ret;
-
-	vmw_gmr_fire_descriptors(dev_priv, gmr_id, desc_dma);
-	vmw_gmr_free_descriptors(dev, desc_dma, &desc_pages);
-
-	return 0;
+	return vmw_gmr2_bind(dev_priv, &data_iter, num_pages, gmr_id);
 }
 
 
 void vmw_gmr_unbind(struct vmw_private *dev_priv, int gmr_id)
 {
-	if (likely(dev_priv->capabilities & SVGA_CAP_GMR2)) {
+	if (likely(dev_priv->capabilities & SVGA_CAP_GMR2))
 		vmw_gmr2_unbind(dev_priv, gmr_id);
-		return;
-	}
-
-	mutex_lock(&dev_priv->hw_mutex);
-	vmw_write(dev_priv, SVGA_REG_GMR_ID, gmr_id);
-	wmb();
-	vmw_write(dev_priv, SVGA_REG_GMR_DESCRIPTOR, 0);
-	mb();
-	mutex_unlock(&dev_priv->hw_mutex);
 }

drivers/gpu/drm/vmwgfx/vmwgfx_gmrid_manager.c

@@ -125,10 +125,21 @@ static int vmw_gmrid_man_init(struct ttm_mem_type_manager *man,
 		return -ENOMEM;
 
 	spin_lock_init(&gman->lock);
-	gman->max_gmr_pages = dev_priv->max_gmr_pages;
 	gman->used_gmr_pages = 0;
 	ida_init(&gman->gmr_ida);
-	gman->max_gmr_ids = p_size;
+
+	switch (p_size) {
+	case VMW_PL_GMR:
+		gman->max_gmr_ids = dev_priv->max_gmr_ids;
+		gman->max_gmr_pages = dev_priv->max_gmr_pages;
+		break;
+	case VMW_PL_MOB:
+		gman->max_gmr_ids = VMWGFX_NUM_MOB;
+		gman->max_gmr_pages = dev_priv->max_mob_pages;
+		break;
+	default:
+		BUG();
+	}
 	man->priv = (void *) gman;
 	return 0;
 }

drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c

@@ -53,7 +53,7 @@ int vmw_getparam_ioctl(struct drm_device *dev, void *data,
 		param->value = dev_priv->fifo.capabilities;
 		break;
 	case DRM_VMW_PARAM_MAX_FB_SIZE:
-		param->value = dev_priv->vram_size;
+		param->value = dev_priv->prim_bb_mem;
 		break;
 	case DRM_VMW_PARAM_FIFO_HW_VERSION:
 	{
@@ -68,6 +68,20 @@ int vmw_getparam_ioctl(struct drm_device *dev, void *data,
 				  SVGA_FIFO_3D_HWVERSION));
 		break;
 	}
+	case DRM_VMW_PARAM_MAX_SURF_MEMORY:
+		param->value = dev_priv->memory_size;
+		break;
+	case DRM_VMW_PARAM_3D_CAPS_SIZE:
+		if (dev_priv->capabilities & SVGA_CAP_GBOBJECTS)
+			param->value = SVGA3D_DEVCAP_MAX;
+		else
+			param->value = (SVGA_FIFO_3D_CAPS_LAST -
+					SVGA_FIFO_3D_CAPS + 1);
+		param->value *= sizeof(uint32_t);
+		break;
+	case DRM_VMW_PARAM_MAX_MOB_MEMORY:
+		param->value = dev_priv->max_mob_pages * PAGE_SIZE;
+		break;
 	default:
 		DRM_ERROR("Illegal vmwgfx get param request: %d\n",
 			  param->param);
@@ -89,13 +103,19 @@ int vmw_get_cap_3d_ioctl(struct drm_device *dev, void *data,
 	void __user *buffer = (void __user *)((unsigned long)(arg->buffer));
 	void *bounce;
 	int ret;
+	bool gb_objects = !!(dev_priv->capabilities & SVGA_CAP_GBOBJECTS);
 
 	if (unlikely(arg->pad64 != 0)) {
 		DRM_ERROR("Illegal GET_3D_CAP argument.\n");
 		return -EINVAL;
 	}
 
-	size = (SVGA_FIFO_3D_CAPS_LAST - SVGA_FIFO_3D_CAPS + 1) << 2;
+	if (gb_objects)
+		size = SVGA3D_DEVCAP_MAX;
+	else
+		size = (SVGA_FIFO_3D_CAPS_LAST - SVGA_FIFO_3D_CAPS + 1);
+
+	size *= sizeof(uint32_t);
 
 	if (arg->max_size < size)
 		size = arg->max_size;
@@ -106,8 +126,22 @@ int vmw_get_cap_3d_ioctl(struct drm_device *dev, void *data,
 		return -ENOMEM;
 	}
 
-	fifo_mem = dev_priv->mmio_virt;
-	memcpy_fromio(bounce, &fifo_mem[SVGA_FIFO_3D_CAPS], size);
+	if (gb_objects) {
+		int i;
+		uint32_t *bounce32 = (uint32_t *) bounce;
+
+		mutex_lock(&dev_priv->hw_mutex);
+		for (i = 0; i < SVGA3D_DEVCAP_MAX; ++i) {
+			vmw_write(dev_priv, SVGA_REG_DEV_CAP, i);
+			*bounce32++ = vmw_read(dev_priv, SVGA_REG_DEV_CAP);
+		}
+		mutex_unlock(&dev_priv->hw_mutex);
+
+	} else {
+
+		fifo_mem = dev_priv->mmio_virt;
+		memcpy_fromio(bounce, &fifo_mem[SVGA_FIFO_3D_CAPS], size);
+	}
 
 	ret = copy_to_user(buffer, bounce, size);
 	if (ret)

drivers/gpu/drm/vmwgfx/vmwgfx_kms.c

@@ -672,9 +672,9 @@ static int vmw_kms_new_framebuffer_surface(struct vmw_private *dev_priv,
 
 	if (unlikely(surface->mip_levels[0] != 1 ||
 		     surface->num_sizes != 1 ||
-		     surface->sizes[0].width < mode_cmd->width ||
-		     surface->sizes[0].height < mode_cmd->height ||
-		     surface->sizes[0].depth != 1)) {
+		     surface->base_size.width < mode_cmd->width ||
+		     surface->base_size.height < mode_cmd->height ||
+		     surface->base_size.depth != 1)) {
 		DRM_ERROR("Incompatible surface dimensions "
 			  "for requested mode.\n");
 		return -EINVAL;
@@ -1645,7 +1645,7 @@ bool vmw_kms_validate_mode_vram(struct vmw_private *dev_priv,
 				uint32_t pitch,
 				uint32_t height)
 {
-	return ((u64) pitch * (u64) height) < (u64) dev_priv->vram_size;
+	return ((u64) pitch * (u64) height) < (u64) dev_priv->prim_bb_mem;
 }
 
 

drivers/gpu/drm/vmwgfx/vmwgfx_mob.c

@@ -0,0 +1,659 @@
+/**************************************************************************
+ *
+ * Copyright © 2012 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"
+
+/*
+ * If we set up the screen target otable, screen objects stop working.
+ */
+
+#define VMW_OTABLE_SETUP_SUB ((VMWGFX_ENABLE_SCREEN_TARGET_OTABLE) ? 0 : 1)
+
+#ifdef CONFIG_64BIT
+#define VMW_PPN_SIZE 8
+#define vmw_cmd_set_otable_base SVGA3dCmdSetOTableBase64
+#define VMW_ID_SET_OTABLE_BASE SVGA_3D_CMD_SET_OTABLE_BASE64
+#define vmw_cmd_define_gb_mob SVGA3dCmdDefineGBMob64
+#define VMW_ID_DEFINE_GB_MOB SVGA_3D_CMD_DEFINE_GB_MOB64
+#define VMW_MOBFMT_PTDEPTH_0 SVGA3D_MOBFMT_PTDEPTH64_0
+#define VMW_MOBFMT_PTDEPTH_1 SVGA3D_MOBFMT_PTDEPTH64_1
+#define VMW_MOBFMT_PTDEPTH_2 SVGA3D_MOBFMT_PTDEPTH64_2
+#else
+#define VMW_PPN_SIZE 4
+#define vmw_cmd_set_otable_base SVGA3dCmdSetOTableBase
+#define VMW_ID_SET_OTABLE_BASE SVGA_3D_CMD_SET_OTABLE_BASE
+#define vmw_cmd_define_gb_mob SVGA3dCmdDefineGBMob
+#define VMW_ID_DEFINE_GB_MOB SVGA_3D_CMD_DEFINE_GB_MOB
+#define VMW_MOBFMT_PTDEPTH_0 SVGA3D_MOBFMT_PTDEPTH_0
+#define VMW_MOBFMT_PTDEPTH_1 SVGA3D_MOBFMT_PTDEPTH_1
+#define VMW_MOBFMT_PTDEPTH_2 SVGA3D_MOBFMT_PTDEPTH_2
+#endif
+
+/*
+ * struct vmw_mob - Structure containing page table and metadata for a
+ * Guest Memory OBject.
+ *
+ * @num_pages       Number of pages that make up the page table.
+ * @pt_level        The indirection level of the page table. 0-2.
+ * @pt_root_page    DMA address of the level 0 page of the page table.
+ */
+struct vmw_mob {
+	struct ttm_buffer_object *pt_bo;
+	unsigned long num_pages;
+	unsigned pt_level;
+	dma_addr_t pt_root_page;
+	uint32_t id;
+};
+
+/*
+ * struct vmw_otable - Guest Memory OBject table metadata
+ *
+ * @size:           Size of the table (page-aligned).
+ * @page_table:     Pointer to a struct vmw_mob holding the page table.
+ */
+struct vmw_otable {
+	unsigned long size;
+	struct vmw_mob *page_table;
+};
+
+static int vmw_mob_pt_populate(struct vmw_private *dev_priv,
+			       struct vmw_mob *mob);
+static void vmw_mob_pt_setup(struct vmw_mob *mob,
+			     struct vmw_piter data_iter,
+			     unsigned long num_data_pages);
+
+/*
+ * vmw_setup_otable_base - Issue an object table base setup command to
+ * the device
+ *
+ * @dev_priv:       Pointer to a device private structure
+ * @type:           Type of object table base
+ * @offset          Start of table offset into dev_priv::otable_bo
+ * @otable          Pointer to otable metadata;
+ *
+ * This function returns -ENOMEM if it fails to reserve fifo space,
+ * and may block waiting for fifo space.
+ */
+static int vmw_setup_otable_base(struct vmw_private *dev_priv,
+				 SVGAOTableType type,
+				 unsigned long offset,
+				 struct vmw_otable *otable)
+{
+	struct {
+		SVGA3dCmdHeader header;
+		vmw_cmd_set_otable_base body;
+	} *cmd;
+	struct vmw_mob *mob;
+	const struct vmw_sg_table *vsgt;
+	struct vmw_piter iter;
+	int ret;
+
+	BUG_ON(otable->page_table != NULL);
+
+	vsgt = vmw_bo_sg_table(dev_priv->otable_bo);
+	vmw_piter_start(&iter, vsgt, offset >> PAGE_SHIFT);
+	WARN_ON(!vmw_piter_next(&iter));
+
+	mob = vmw_mob_create(otable->size >> PAGE_SHIFT);
+	if (unlikely(mob == NULL)) {
+		DRM_ERROR("Failed creating OTable page table.\n");
+		return -ENOMEM;
+	}
+
+	if (otable->size <= PAGE_SIZE) {
+		mob->pt_level = VMW_MOBFMT_PTDEPTH_0;
+		mob->pt_root_page = vmw_piter_dma_addr(&iter);
+	} else if (vsgt->num_regions == 1) {
+		mob->pt_level = SVGA3D_MOBFMT_RANGE;
+		mob->pt_root_page = vmw_piter_dma_addr(&iter);
+	} else {
+		ret = vmw_mob_pt_populate(dev_priv, mob);
+		if (unlikely(ret != 0))
+			goto out_no_populate;
+
+		vmw_mob_pt_setup(mob, iter, otable->size >> PAGE_SHIFT);
+		mob->pt_level += VMW_MOBFMT_PTDEPTH_1 - SVGA3D_MOBFMT_PTDEPTH_1;
+	}
+
+	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Failed reserving FIFO space for OTable setup.\n");
+		goto out_no_fifo;
+	}
+
+	memset(cmd, 0, sizeof(*cmd));
+	cmd->header.id = VMW_ID_SET_OTABLE_BASE;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.type = type;
+	cmd->body.baseAddress = mob->pt_root_page >> PAGE_SHIFT;
+	cmd->body.sizeInBytes = otable->size;
+	cmd->body.validSizeInBytes = 0;
+	cmd->body.ptDepth = mob->pt_level;
+
+	/*
+	 * The device doesn't support this, But the otable size is
+	 * determined at compile-time, so this BUG shouldn't trigger
+	 * randomly.
+	 */
+	BUG_ON(mob->pt_level == VMW_MOBFMT_PTDEPTH_2);
+
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+	otable->page_table = mob;
+
+	return 0;
+
+out_no_fifo:
+out_no_populate:
+	vmw_mob_destroy(mob);
+	return ret;
+}
+
+/*
+ * vmw_takedown_otable_base - Issue an object table base takedown command
+ * to the device
+ *
+ * @dev_priv:       Pointer to a device private structure
+ * @type:           Type of object table base
+ *
+ */
+static void vmw_takedown_otable_base(struct vmw_private *dev_priv,
+				     SVGAOTableType type,
+				     struct vmw_otable *otable)
+{
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdSetOTableBase body;
+	} *cmd;
+	struct ttm_buffer_object *bo = otable->page_table->pt_bo;
+
+	if (otable->page_table == NULL)
+		return;
+
+	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
+	if (unlikely(cmd == NULL))
+		DRM_ERROR("Failed reserving FIFO space for OTable setup.\n");
+
+	memset(cmd, 0, sizeof(*cmd));
+	cmd->header.id = SVGA_3D_CMD_SET_OTABLE_BASE;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.type = type;
+	cmd->body.baseAddress = 0;
+	cmd->body.sizeInBytes = 0;
+	cmd->body.validSizeInBytes = 0;
+	cmd->body.ptDepth = SVGA3D_MOBFMT_INVALID;
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+
+	if (bo) {
+		int ret;
+
+		ret = ttm_bo_reserve(bo, false, true, false, false);
+		BUG_ON(ret != 0);
+
+		vmw_fence_single_bo(bo, NULL);
+		ttm_bo_unreserve(bo);
+	}
+
+	vmw_mob_destroy(otable->page_table);
+	otable->page_table = NULL;
+}
+
+/*
+ * vmw_otables_setup - Set up guest backed memory object tables
+ *
+ * @dev_priv:       Pointer to a device private structure
+ *
+ * Takes care of the device guest backed surface
+ * initialization, by setting up the guest backed memory object tables.
+ * Returns 0 on success and various error codes on failure. A succesful return
+ * means the object tables can be taken down using the vmw_otables_takedown
+ * function.
+ */
+int vmw_otables_setup(struct vmw_private *dev_priv)
+{
+	unsigned long offset;
+	unsigned long bo_size;
+	struct vmw_otable *otables;
+	SVGAOTableType i;
+	int ret;
+
+	otables = kzalloc(SVGA_OTABLE_DX9_MAX * sizeof(*otables),
+			  GFP_KERNEL);
+	if (unlikely(otables == NULL)) {
+		DRM_ERROR("Failed to allocate space for otable "
+			  "metadata.\n");
+		return -ENOMEM;
+	}
+
+	otables[SVGA_OTABLE_MOB].size =
+		VMWGFX_NUM_MOB * SVGA3D_OTABLE_MOB_ENTRY_SIZE;
+	otables[SVGA_OTABLE_SURFACE].size =
+		VMWGFX_NUM_GB_SURFACE * SVGA3D_OTABLE_SURFACE_ENTRY_SIZE;
+	otables[SVGA_OTABLE_CONTEXT].size =
+		VMWGFX_NUM_GB_CONTEXT * SVGA3D_OTABLE_CONTEXT_ENTRY_SIZE;
+	otables[SVGA_OTABLE_SHADER].size =
+		VMWGFX_NUM_GB_SHADER * SVGA3D_OTABLE_SHADER_ENTRY_SIZE;
+	otables[SVGA_OTABLE_SCREEN_TARGET].size =
+		VMWGFX_NUM_GB_SCREEN_TARGET *
+		SVGA3D_OTABLE_SCREEN_TARGET_ENTRY_SIZE;
+
+	bo_size = 0;
+	for (i = 0; i < SVGA_OTABLE_DX9_MAX; ++i) {
+		otables[i].size =
+			(otables[i].size + PAGE_SIZE - 1) & PAGE_MASK;
+		bo_size += otables[i].size;
+	}
+
+	ret = ttm_bo_create(&dev_priv->bdev, bo_size,
+			    ttm_bo_type_device,
+			    &vmw_sys_ne_placement,
+			    0, false, NULL,
+			    &dev_priv->otable_bo);
+
+	if (unlikely(ret != 0))
+		goto out_no_bo;
+
+	ret = ttm_bo_reserve(dev_priv->otable_bo, false, true, false, false);
+	BUG_ON(ret != 0);
+	ret = vmw_bo_driver.ttm_tt_populate(dev_priv->otable_bo->ttm);
+	if (unlikely(ret != 0))
+		goto out_unreserve;
+	ret = vmw_bo_map_dma(dev_priv->otable_bo);
+	if (unlikely(ret != 0))
+		goto out_unreserve;
+
+	ttm_bo_unreserve(dev_priv->otable_bo);
+
+	offset = 0;
+	for (i = 0; i < SVGA_OTABLE_DX9_MAX - VMW_OTABLE_SETUP_SUB; ++i) {
+		ret = vmw_setup_otable_base(dev_priv, i, offset,
+					    &otables[i]);
+		if (unlikely(ret != 0))
+			goto out_no_setup;
+		offset += otables[i].size;
+	}
+
+	dev_priv->otables = otables;
+	return 0;
+
+out_unreserve:
+	ttm_bo_unreserve(dev_priv->otable_bo);
+out_no_setup:
+	for (i = 0; i < SVGA_OTABLE_DX9_MAX - VMW_OTABLE_SETUP_SUB; ++i)
+		vmw_takedown_otable_base(dev_priv, i, &otables[i]);
+
+	ttm_bo_unref(&dev_priv->otable_bo);
+out_no_bo:
+	kfree(otables);
+	return ret;
+}
+
+
+/*
+ * vmw_otables_takedown - Take down guest backed memory object tables
+ *
+ * @dev_priv:       Pointer to a device private structure
+ *
+ * Take down the Guest Memory Object tables.
+ */
+void vmw_otables_takedown(struct vmw_private *dev_priv)
+{
+	SVGAOTableType i;
+	struct ttm_buffer_object *bo = dev_priv->otable_bo;
+	int ret;
+
+	for (i = 0; i < SVGA_OTABLE_DX9_MAX - VMW_OTABLE_SETUP_SUB; ++i)
+		vmw_takedown_otable_base(dev_priv, i,
+					 &dev_priv->otables[i]);
+
+	ret = ttm_bo_reserve(bo, false, true, false, false);
+	BUG_ON(ret != 0);
+
+	vmw_fence_single_bo(bo, NULL);
+	ttm_bo_unreserve(bo);
+
+	ttm_bo_unref(&dev_priv->otable_bo);
+	kfree(dev_priv->otables);
+	dev_priv->otables = NULL;
+}
+
+
+/*
+ * vmw_mob_calculate_pt_pages - Calculate the number of page table pages
+ * needed for a guest backed memory object.
+ *
+ * @data_pages:  Number of data pages in the memory object buffer.
+ */
+static unsigned long vmw_mob_calculate_pt_pages(unsigned long data_pages)
+{
+	unsigned long data_size = data_pages * PAGE_SIZE;
+	unsigned long tot_size = 0;
+
+	while (likely(data_size > PAGE_SIZE)) {
+		data_size = DIV_ROUND_UP(data_size, PAGE_SIZE);
+		data_size *= VMW_PPN_SIZE;
+		tot_size += (data_size + PAGE_SIZE - 1) & PAGE_MASK;
+	}
+
+	return tot_size >> PAGE_SHIFT;
+}
+
+/*
+ * vmw_mob_create - Create a mob, but don't populate it.
+ *
+ * @data_pages:  Number of data pages of the underlying buffer object.
+ */
+struct vmw_mob *vmw_mob_create(unsigned long data_pages)
+{
+	struct vmw_mob *mob = kzalloc(sizeof(*mob), GFP_KERNEL);
+
+	if (unlikely(mob == NULL))
+		return NULL;
+
+	mob->num_pages = vmw_mob_calculate_pt_pages(data_pages);
+
+	return mob;
+}
+
+/*
+ * vmw_mob_pt_populate - Populate the mob pagetable
+ *
+ * @mob:         Pointer to the mob the pagetable of which we want to
+ *               populate.
+ *
+ * This function allocates memory to be used for the pagetable, and
+ * adjusts TTM memory accounting accordingly. Returns ENOMEM if
+ * memory resources aren't sufficient and may cause TTM buffer objects
+ * to be swapped out by using the TTM memory accounting function.
+ */
+static int vmw_mob_pt_populate(struct vmw_private *dev_priv,
+			       struct vmw_mob *mob)
+{
+	int ret;
+	BUG_ON(mob->pt_bo != NULL);
+
+	ret = ttm_bo_create(&dev_priv->bdev, mob->num_pages * PAGE_SIZE,
+			    ttm_bo_type_device,
+			    &vmw_sys_ne_placement,
+			    0, false, NULL, &mob->pt_bo);
+	if (unlikely(ret != 0))
+		return ret;
+
+	ret = ttm_bo_reserve(mob->pt_bo, false, true, false, false);
+
+	BUG_ON(ret != 0);
+	ret = vmw_bo_driver.ttm_tt_populate(mob->pt_bo->ttm);
+	if (unlikely(ret != 0))
+		goto out_unreserve;
+	ret = vmw_bo_map_dma(mob->pt_bo);
+	if (unlikely(ret != 0))
+		goto out_unreserve;
+
+	ttm_bo_unreserve(mob->pt_bo);
+	
+	return 0;
+
+out_unreserve:
+	ttm_bo_unreserve(mob->pt_bo);
+	ttm_bo_unref(&mob->pt_bo);
+
+	return ret;
+}
+
+/**
+ * vmw_mob_assign_ppn - Assign a value to a page table entry
+ *
+ * @addr: Pointer to pointer to page table entry.
+ * @val: The page table entry
+ *
+ * Assigns a value to a page table entry pointed to by *@addr and increments
+ * *@addr according to the page table entry size.
+ */
+#if (VMW_PPN_SIZE == 8)
+static void vmw_mob_assign_ppn(uint32_t **addr, dma_addr_t val)
+{
+	*((uint64_t *) *addr) = val >> PAGE_SHIFT;
+	*addr += 2;
+}
+#else
+static void vmw_mob_assign_ppn(uint32_t **addr, dma_addr_t val)
+{
+	*(*addr)++ = val >> PAGE_SHIFT;
+}
+#endif
+
+/*
+ * vmw_mob_build_pt - Build a pagetable
+ *
+ * @data_addr:      Array of DMA addresses to the underlying buffer
+ *                  object's data pages.
+ * @num_data_pages: Number of buffer object data pages.
+ * @pt_pages:       Array of page pointers to the page table pages.
+ *
+ * Returns the number of page table pages actually used.
+ * Uses atomic kmaps of highmem pages to avoid TLB thrashing.
+ */
+static unsigned long vmw_mob_build_pt(struct vmw_piter *data_iter,
+				      unsigned long num_data_pages,
+				      struct vmw_piter *pt_iter)
+{
+	unsigned long pt_size = num_data_pages * VMW_PPN_SIZE;
+	unsigned long num_pt_pages = DIV_ROUND_UP(pt_size, PAGE_SIZE);
+	unsigned long pt_page;
+	uint32_t *addr, *save_addr;
+	unsigned long i;
+	struct page *page;
+
+	for (pt_page = 0; pt_page < num_pt_pages; ++pt_page) {
+		page = vmw_piter_page(pt_iter);
+
+		save_addr = addr = kmap_atomic(page);
+
+		for (i = 0; i < PAGE_SIZE / VMW_PPN_SIZE; ++i) {
+			vmw_mob_assign_ppn(&addr,
+					   vmw_piter_dma_addr(data_iter));
+			if (unlikely(--num_data_pages == 0))
+				break;
+			WARN_ON(!vmw_piter_next(data_iter));
+		}
+		kunmap_atomic(save_addr);
+		vmw_piter_next(pt_iter);
+	}
+
+	return num_pt_pages;
+}
+
+/*
+ * vmw_mob_build_pt - Set up a multilevel mob pagetable
+ *
+ * @mob:            Pointer to a mob whose page table needs setting up.
+ * @data_addr       Array of DMA addresses to the buffer object's data
+ *                  pages.
+ * @num_data_pages: Number of buffer object data pages.
+ *
+ * Uses tail recursion to set up a multilevel mob page table.
+ */
+static void vmw_mob_pt_setup(struct vmw_mob *mob,
+			     struct vmw_piter data_iter,
+			     unsigned long num_data_pages)
+{
+	unsigned long num_pt_pages = 0;
+	struct ttm_buffer_object *bo = mob->pt_bo;
+	struct vmw_piter save_pt_iter;
+	struct vmw_piter pt_iter;
+	const struct vmw_sg_table *vsgt;
+	int ret;
+
+	ret = ttm_bo_reserve(bo, false, true, false, 0);
+	BUG_ON(ret != 0);
+
+	vsgt = vmw_bo_sg_table(bo);
+	vmw_piter_start(&pt_iter, vsgt, 0);
+	BUG_ON(!vmw_piter_next(&pt_iter));
+	mob->pt_level = 0;
+	while (likely(num_data_pages > 1)) {
+		++mob->pt_level;
+		BUG_ON(mob->pt_level > 2);
+		save_pt_iter = pt_iter;
+		num_pt_pages = vmw_mob_build_pt(&data_iter, num_data_pages,
+						&pt_iter);
+		data_iter = save_pt_iter;
+		num_data_pages = num_pt_pages;
+	}
+
+	mob->pt_root_page = vmw_piter_dma_addr(&save_pt_iter);
+	ttm_bo_unreserve(bo);
+}
+
+/*
+ * vmw_mob_destroy - Destroy a mob, unpopulating first if necessary.
+ *
+ * @mob:            Pointer to a mob to destroy.
+ */
+void vmw_mob_destroy(struct vmw_mob *mob)
+{
+	if (mob->pt_bo)
+		ttm_bo_unref(&mob->pt_bo);
+	kfree(mob);
+}
+
+/*
+ * vmw_mob_unbind - Hide a mob from the device.
+ *
+ * @dev_priv:       Pointer to a device private.
+ * @mob_id:         Device id of the mob to unbind.
+ */
+void vmw_mob_unbind(struct vmw_private *dev_priv,
+		    struct vmw_mob *mob)
+{
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdDestroyGBMob body;
+	} *cmd;
+	int ret;
+	struct ttm_buffer_object *bo = mob->pt_bo;
+
+	if (bo) {
+		ret = ttm_bo_reserve(bo, false, true, false, 0);
+		/*
+		 * Noone else should be using this buffer.
+		 */
+		BUG_ON(ret != 0);
+	}
+
+	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Failed reserving FIFO space for Memory "
+			  "Object unbinding.\n");
+	}
+	cmd->header.id = SVGA_3D_CMD_DESTROY_GB_MOB;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.mobid = mob->id;
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+	if (bo) {
+		vmw_fence_single_bo(bo, NULL);
+		ttm_bo_unreserve(bo);
+	}
+	vmw_3d_resource_dec(dev_priv, false);
+}
+
+/*
+ * vmw_mob_bind - Make a mob visible to the device after first
+ *                populating it if necessary.
+ *
+ * @dev_priv:       Pointer to a device private.
+ * @mob:            Pointer to the mob we're making visible.
+ * @data_addr:      Array of DMA addresses to the data pages of the underlying
+ *                  buffer object.
+ * @num_data_pages: Number of data pages of the underlying buffer
+ *                  object.
+ * @mob_id:         Device id of the mob to bind
+ *
+ * This function is intended to be interfaced with the ttm_tt backend
+ * code.
+ */
+int vmw_mob_bind(struct vmw_private *dev_priv,
+		 struct vmw_mob *mob,
+		 const struct vmw_sg_table *vsgt,
+		 unsigned long num_data_pages,
+		 int32_t mob_id)
+{
+	int ret;
+	bool pt_set_up = false;
+	struct vmw_piter data_iter;
+	struct {
+		SVGA3dCmdHeader header;
+		vmw_cmd_define_gb_mob body;
+	} *cmd;
+
+	mob->id = mob_id;
+	vmw_piter_start(&data_iter, vsgt, 0);
+	if (unlikely(!vmw_piter_next(&data_iter)))
+		return 0;
+
+	if (likely(num_data_pages == 1)) {
+		mob->pt_level = VMW_MOBFMT_PTDEPTH_0;
+		mob->pt_root_page = vmw_piter_dma_addr(&data_iter);
+	} else if (vsgt->num_regions == 1) {
+		mob->pt_level = SVGA3D_MOBFMT_RANGE;
+		mob->pt_root_page = vmw_piter_dma_addr(&data_iter);
+	} else if (unlikely(mob->pt_bo == NULL)) {
+		ret = vmw_mob_pt_populate(dev_priv, mob);
+		if (unlikely(ret != 0))
+			return ret;
+
+		vmw_mob_pt_setup(mob, data_iter, num_data_pages);
+		pt_set_up = true;
+		mob->pt_level += VMW_MOBFMT_PTDEPTH_1 - SVGA3D_MOBFMT_PTDEPTH_1;
+	}
+
+	(void) vmw_3d_resource_inc(dev_priv, false);
+
+	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Failed reserving FIFO space for Memory "
+			  "Object binding.\n");
+		goto out_no_cmd_space;
+	}
+
+	cmd->header.id = VMW_ID_DEFINE_GB_MOB;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.mobid = mob_id;
+	cmd->body.ptDepth = mob->pt_level;
+	cmd->body.base = mob->pt_root_page >> PAGE_SHIFT;
+	cmd->body.sizeInBytes = num_data_pages * PAGE_SIZE;
+
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+
+	return 0;
+
+out_no_cmd_space:
+	vmw_3d_resource_dec(dev_priv, false);
+	if (pt_set_up)
+		ttm_bo_unref(&mob->pt_bo);
+
+	return -ENOMEM;
+}

drivers/gpu/drm/vmwgfx/vmwgfx_resource.c

@@ -215,6 +215,7 @@ int vmw_resource_init(struct vmw_private *dev_priv, struct vmw_resource *res,
 	res->func = func;
 	INIT_LIST_HEAD(&res->lru_head);
 	INIT_LIST_HEAD(&res->mob_head);
+	INIT_LIST_HEAD(&res->binding_head);
 	res->id = -1;
 	res->backup = NULL;
 	res->backup_offset = 0;
@@ -441,6 +442,21 @@ static void vmw_user_dmabuf_release(struct ttm_base_object **p_base)
 	ttm_bo_unref(&bo);
 }
 
+static void vmw_user_dmabuf_ref_obj_release(struct ttm_base_object *base,
+					    enum ttm_ref_type ref_type)
+{
+	struct vmw_user_dma_buffer *user_bo;
+	user_bo = container_of(base, struct vmw_user_dma_buffer, prime.base);
+
+	switch (ref_type) {
+	case TTM_REF_SYNCCPU_WRITE:
+		ttm_bo_synccpu_write_release(&user_bo->dma.base);
+		break;
+	default:
+		BUG();
+	}
+}
+
 /**
  * vmw_user_dmabuf_alloc - Allocate a user dma buffer
  *
@@ -471,6 +487,8 @@ int vmw_user_dmabuf_alloc(struct vmw_private *dev_priv,
 	}
 
 	ret = vmw_dmabuf_init(dev_priv, &user_bo->dma, size,
+			      (dev_priv->has_mob) ?
+			      &vmw_sys_placement :
 			      &vmw_vram_sys_placement, true,
 			      &vmw_user_dmabuf_destroy);
 	if (unlikely(ret != 0))
@@ -482,7 +500,8 @@ int vmw_user_dmabuf_alloc(struct vmw_private *dev_priv,
 				    &user_bo->prime,
 				    shareable,
 				    ttm_buffer_type,
-				    &vmw_user_dmabuf_release, NULL);
+				    &vmw_user_dmabuf_release,
+				    &vmw_user_dmabuf_ref_obj_release);
 	if (unlikely(ret != 0)) {
 		ttm_bo_unref(&tmp);
 		goto out_no_base_object;
@@ -515,6 +534,130 @@ int vmw_user_dmabuf_verify_access(struct ttm_buffer_object *bo,
 		vmw_user_bo->prime.base.shareable) ? 0 : -EPERM;
 }
 
+/**
+ * vmw_user_dmabuf_synccpu_grab - Grab a struct vmw_user_dma_buffer for cpu
+ * access, idling previous GPU operations on the buffer and optionally
+ * blocking it for further command submissions.
+ *
+ * @user_bo: Pointer to the buffer object being grabbed for CPU access
+ * @tfile: Identifying the caller.
+ * @flags: Flags indicating how the grab should be performed.
+ *
+ * A blocking grab will be automatically released when @tfile is closed.
+ */
+static int vmw_user_dmabuf_synccpu_grab(struct vmw_user_dma_buffer *user_bo,
+					struct ttm_object_file *tfile,
+					uint32_t flags)
+{
+	struct ttm_buffer_object *bo = &user_bo->dma.base;
+	bool existed;
+	int ret;
+
+	if (flags & drm_vmw_synccpu_allow_cs) {
+		struct ttm_bo_device *bdev = bo->bdev;
+
+		spin_lock(&bdev->fence_lock);
+		ret = ttm_bo_wait(bo, false, true,
+				  !!(flags & drm_vmw_synccpu_dontblock));
+		spin_unlock(&bdev->fence_lock);
+		return ret;
+	}
+
+	ret = ttm_bo_synccpu_write_grab
+		(bo, !!(flags & drm_vmw_synccpu_dontblock));
+	if (unlikely(ret != 0))
+		return ret;
+
+	ret = ttm_ref_object_add(tfile, &user_bo->prime.base,
+				 TTM_REF_SYNCCPU_WRITE, &existed);
+	if (ret != 0 || existed)
+		ttm_bo_synccpu_write_release(&user_bo->dma.base);
+
+	return ret;
+}
+
+/**
+ * vmw_user_dmabuf_synccpu_release - Release a previous grab for CPU access,
+ * and unblock command submission on the buffer if blocked.
+ *
+ * @handle: Handle identifying the buffer object.
+ * @tfile: Identifying the caller.
+ * @flags: Flags indicating the type of release.
+ */
+static int vmw_user_dmabuf_synccpu_release(uint32_t handle,
+					   struct ttm_object_file *tfile,
+					   uint32_t flags)
+{
+	if (!(flags & drm_vmw_synccpu_allow_cs))
+		return ttm_ref_object_base_unref(tfile, handle,
+						 TTM_REF_SYNCCPU_WRITE);
+
+	return 0;
+}
+
+/**
+ * vmw_user_dmabuf_synccpu_release - ioctl function implementing the synccpu
+ * functionality.
+ *
+ * @dev: Identifies the drm device.
+ * @data: Pointer to the ioctl argument.
+ * @file_priv: Identifies the caller.
+ *
+ * This function checks the ioctl arguments for validity and calls the
+ * relevant synccpu functions.
+ */
+int vmw_user_dmabuf_synccpu_ioctl(struct drm_device *dev, void *data,
+				  struct drm_file *file_priv)
+{
+	struct drm_vmw_synccpu_arg *arg =
+		(struct drm_vmw_synccpu_arg *) data;
+	struct vmw_dma_buffer *dma_buf;
+	struct vmw_user_dma_buffer *user_bo;
+	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
+	int ret;
+
+	if ((arg->flags & (drm_vmw_synccpu_read | drm_vmw_synccpu_write)) == 0
+	    || (arg->flags & ~(drm_vmw_synccpu_read | drm_vmw_synccpu_write |
+			       drm_vmw_synccpu_dontblock |
+			       drm_vmw_synccpu_allow_cs)) != 0) {
+		DRM_ERROR("Illegal synccpu flags.\n");
+		return -EINVAL;
+	}
+
+	switch (arg->op) {
+	case drm_vmw_synccpu_grab:
+		ret = vmw_user_dmabuf_lookup(tfile, arg->handle, &dma_buf);
+		if (unlikely(ret != 0))
+			return ret;
+
+		user_bo = container_of(dma_buf, struct vmw_user_dma_buffer,
+				       dma);
+		ret = vmw_user_dmabuf_synccpu_grab(user_bo, tfile, arg->flags);
+		vmw_dmabuf_unreference(&dma_buf);
+		if (unlikely(ret != 0 && ret != -ERESTARTSYS &&
+			     ret != -EBUSY)) {
+			DRM_ERROR("Failed synccpu grab on handle 0x%08x.\n",
+				  (unsigned int) arg->handle);
+			return ret;
+		}
+		break;
+	case drm_vmw_synccpu_release:
+		ret = vmw_user_dmabuf_synccpu_release(arg->handle, tfile,
+						      arg->flags);
+		if (unlikely(ret != 0)) {
+			DRM_ERROR("Failed synccpu release on handle 0x%08x.\n",
+				  (unsigned int) arg->handle);
+			return ret;
+		}
+		break;
+	default:
+		DRM_ERROR("Invalid synccpu operation.\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
 int vmw_dmabuf_alloc_ioctl(struct drm_device *dev, void *data,
 			   struct drm_file *file_priv)
 {
@@ -591,7 +734,8 @@ int vmw_user_dmabuf_lookup(struct ttm_object_file *tfile,
 }
 
 int vmw_user_dmabuf_reference(struct ttm_object_file *tfile,
-			      struct vmw_dma_buffer *dma_buf)
+			      struct vmw_dma_buffer *dma_buf,
+			      uint32_t *handle)
 {
 	struct vmw_user_dma_buffer *user_bo;
 
@@ -599,6 +743,8 @@ int vmw_user_dmabuf_reference(struct ttm_object_file *tfile,
 		return -EINVAL;
 
 	user_bo = container_of(dma_buf, struct vmw_user_dma_buffer, dma);
+
+	*handle = user_bo->prime.base.hash.key;
 	return ttm_ref_object_add(tfile, &user_bo->prime.base,
 				  TTM_REF_USAGE, NULL);
 }
@@ -1291,11 +1437,54 @@ void vmw_fence_single_bo(struct ttm_buffer_object *bo,
  * @mem:            The truct ttm_mem_reg indicating to what memory
  *                  region the move is taking place.
  *
- * For now does nothing.
+ * Evicts the Guest Backed hardware resource if the backup
+ * buffer is being moved out of MOB memory.
+ * Note that this function should not race with the resource
+ * validation code as long as it accesses only members of struct
+ * resource that remain static while bo::res is !NULL and
+ * while we have @bo reserved. struct resource::backup is *not* a
+ * static member. The resource validation code will take care
+ * to set @bo::res to NULL, while having @bo reserved when the
+ * buffer is no longer bound to the resource, so @bo:res can be
+ * used to determine whether there is a need to unbind and whether
+ * it is safe to unbind.
  */
 void vmw_resource_move_notify(struct ttm_buffer_object *bo,
 			      struct ttm_mem_reg *mem)
 {
+	struct vmw_dma_buffer *dma_buf;
+
+	if (mem == NULL)
+		return;
+
+	if (bo->destroy != vmw_dmabuf_bo_free &&
+	    bo->destroy != vmw_user_dmabuf_destroy)
+		return;
+
+	dma_buf = container_of(bo, struct vmw_dma_buffer, base);
+
+	if (mem->mem_type != VMW_PL_MOB) {
+		struct vmw_resource *res, *n;
+		struct ttm_bo_device *bdev = bo->bdev;
+		struct ttm_validate_buffer val_buf;
+
+		val_buf.bo = bo;
+
+		list_for_each_entry_safe(res, n, &dma_buf->res_list, mob_head) {
+
+			if (unlikely(res->func->unbind == NULL))
+				continue;
+
+			(void) res->func->unbind(res, true, &val_buf);
+			res->backup_dirty = true;
+			res->res_dirty = false;
+			list_del_init(&res->mob_head);
+		}
+
+		spin_lock(&bdev->fence_lock);
+		(void) ttm_bo_wait(bo, false, false, false);
+		spin_unlock(&bdev->fence_lock);
+	}
 }
 
 /**

drivers/gpu/drm/vmwgfx/vmwgfx_shader.c

@@ -0,0 +1,440 @@
+/**************************************************************************
+ *
+ * Copyright © 2009-2012 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"
+#include "vmwgfx_resource_priv.h"
+#include "ttm/ttm_placement.h"
+
+struct vmw_shader {
+	struct vmw_resource res;
+	SVGA3dShaderType type;
+	uint32_t size;
+};
+
+struct vmw_user_shader {
+	struct ttm_base_object base;
+	struct vmw_shader shader;
+};
+
+static void vmw_user_shader_free(struct vmw_resource *res);
+static struct vmw_resource *
+vmw_user_shader_base_to_res(struct ttm_base_object *base);
+
+static int vmw_gb_shader_create(struct vmw_resource *res);
+static int vmw_gb_shader_bind(struct vmw_resource *res,
+			       struct ttm_validate_buffer *val_buf);
+static int vmw_gb_shader_unbind(struct vmw_resource *res,
+				 bool readback,
+				 struct ttm_validate_buffer *val_buf);
+static int vmw_gb_shader_destroy(struct vmw_resource *res);
+
+static uint64_t vmw_user_shader_size;
+
+static const struct vmw_user_resource_conv user_shader_conv = {
+	.object_type = VMW_RES_SHADER,
+	.base_obj_to_res = vmw_user_shader_base_to_res,
+	.res_free = vmw_user_shader_free
+};
+
+const struct vmw_user_resource_conv *user_shader_converter =
+	&user_shader_conv;
+
+
+static const struct vmw_res_func vmw_gb_shader_func = {
+	.res_type = vmw_res_shader,
+	.needs_backup = true,
+	.may_evict = true,
+	.type_name = "guest backed shaders",
+	.backup_placement = &vmw_mob_placement,
+	.create = vmw_gb_shader_create,
+	.destroy = vmw_gb_shader_destroy,
+	.bind = vmw_gb_shader_bind,
+	.unbind = vmw_gb_shader_unbind
+};
+
+/**
+ * Shader management:
+ */
+
+static inline struct vmw_shader *
+vmw_res_to_shader(struct vmw_resource *res)
+{
+	return container_of(res, struct vmw_shader, res);
+}
+
+static void vmw_hw_shader_destroy(struct vmw_resource *res)
+{
+	(void) vmw_gb_shader_destroy(res);
+}
+
+static int vmw_gb_shader_init(struct vmw_private *dev_priv,
+			      struct vmw_resource *res,
+			      uint32_t size,
+			      uint64_t offset,
+			      SVGA3dShaderType type,
+			      struct vmw_dma_buffer *byte_code,
+			      void (*res_free) (struct vmw_resource *res))
+{
+	struct vmw_shader *shader = vmw_res_to_shader(res);
+	int ret;
+
+	ret = vmw_resource_init(dev_priv, res, true,
+				res_free, &vmw_gb_shader_func);
+
+
+	if (unlikely(ret != 0)) {
+		if (res_free)
+			res_free(res);
+		else
+			kfree(res);
+		return ret;
+	}
+
+	res->backup_size = size;
+	if (byte_code) {
+		res->backup = vmw_dmabuf_reference(byte_code);
+		res->backup_offset = offset;
+	}
+	shader->size = size;
+	shader->type = type;
+
+	vmw_resource_activate(res, vmw_hw_shader_destroy);
+	return 0;
+}
+
+static int vmw_gb_shader_create(struct vmw_resource *res)
+{
+	struct vmw_private *dev_priv = res->dev_priv;
+	struct vmw_shader *shader = vmw_res_to_shader(res);
+	int ret;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdDefineGBShader body;
+	} *cmd;
+
+	if (likely(res->id != -1))
+		return 0;
+
+	ret = vmw_resource_alloc_id(res);
+	if (unlikely(ret != 0)) {
+		DRM_ERROR("Failed to allocate a shader id.\n");
+		goto out_no_id;
+	}
+
+	if (unlikely(res->id >= VMWGFX_NUM_GB_SHADER)) {
+		ret = -EBUSY;
+		goto out_no_fifo;
+	}
+
+	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Failed reserving FIFO space for shader "
+			  "creation.\n");
+		ret = -ENOMEM;
+		goto out_no_fifo;
+	}
+
+	cmd->header.id = SVGA_3D_CMD_DEFINE_GB_SHADER;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.shid = res->id;
+	cmd->body.type = shader->type;
+	cmd->body.sizeInBytes = shader->size;
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+	(void) vmw_3d_resource_inc(dev_priv, false);
+
+	return 0;
+
+out_no_fifo:
+	vmw_resource_release_id(res);
+out_no_id:
+	return ret;
+}
+
+static int vmw_gb_shader_bind(struct vmw_resource *res,
+			      struct ttm_validate_buffer *val_buf)
+{
+	struct vmw_private *dev_priv = res->dev_priv;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdBindGBShader body;
+	} *cmd;
+	struct ttm_buffer_object *bo = val_buf->bo;
+
+	BUG_ON(bo->mem.mem_type != VMW_PL_MOB);
+
+	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Failed reserving FIFO space for shader "
+			  "binding.\n");
+		return -ENOMEM;
+	}
+
+	cmd->header.id = SVGA_3D_CMD_BIND_GB_SHADER;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.shid = res->id;
+	cmd->body.mobid = bo->mem.start;
+	cmd->body.offsetInBytes = 0;
+	res->backup_dirty = false;
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+
+	return 0;
+}
+
+static int vmw_gb_shader_unbind(struct vmw_resource *res,
+				bool readback,
+				struct ttm_validate_buffer *val_buf)
+{
+	struct vmw_private *dev_priv = res->dev_priv;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdBindGBShader body;
+	} *cmd;
+	struct vmw_fence_obj *fence;
+
+	BUG_ON(res->backup->base.mem.mem_type != VMW_PL_MOB);
+
+	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Failed reserving FIFO space for shader "
+			  "unbinding.\n");
+		return -ENOMEM;
+	}
+
+	cmd->header.id = SVGA_3D_CMD_BIND_GB_SHADER;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.shid = res->id;
+	cmd->body.mobid = SVGA3D_INVALID_ID;
+	cmd->body.offsetInBytes = 0;
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+
+	/*
+	 * Create a fence object and fence the backup buffer.
+	 */
+
+	(void) vmw_execbuf_fence_commands(NULL, dev_priv,
+					  &fence, NULL);
+
+	vmw_fence_single_bo(val_buf->bo, fence);
+
+	if (likely(fence != NULL))
+		vmw_fence_obj_unreference(&fence);
+
+	return 0;
+}
+
+static int vmw_gb_shader_destroy(struct vmw_resource *res)
+{
+	struct vmw_private *dev_priv = res->dev_priv;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdDestroyGBShader body;
+	} *cmd;
+
+	if (likely(res->id == -1))
+		return 0;
+
+	mutex_lock(&dev_priv->binding_mutex);
+	vmw_context_binding_res_list_kill(&res->binding_head);
+
+	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Failed reserving FIFO space for shader "
+			  "destruction.\n");
+		return -ENOMEM;
+	}
+
+	cmd->header.id = SVGA_3D_CMD_DESTROY_GB_SHADER;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.shid = res->id;
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+	mutex_unlock(&dev_priv->binding_mutex);
+	vmw_resource_release_id(res);
+	vmw_3d_resource_dec(dev_priv, false);
+
+	return 0;
+}
+
+/**
+ * User-space shader management:
+ */
+
+static struct vmw_resource *
+vmw_user_shader_base_to_res(struct ttm_base_object *base)
+{
+	return &(container_of(base, struct vmw_user_shader, base)->
+		 shader.res);
+}
+
+static void vmw_user_shader_free(struct vmw_resource *res)
+{
+	struct vmw_user_shader *ushader =
+		container_of(res, struct vmw_user_shader, shader.res);
+	struct vmw_private *dev_priv = res->dev_priv;
+
+	ttm_base_object_kfree(ushader, base);
+	ttm_mem_global_free(vmw_mem_glob(dev_priv),
+			    vmw_user_shader_size);
+}
+
+/**
+ * This function is called when user space has no more references on the
+ * base object. It releases the base-object's reference on the resource object.
+ */
+
+static void vmw_user_shader_base_release(struct ttm_base_object **p_base)
+{
+	struct ttm_base_object *base = *p_base;
+	struct vmw_resource *res = vmw_user_shader_base_to_res(base);
+
+	*p_base = NULL;
+	vmw_resource_unreference(&res);
+}
+
+int vmw_shader_destroy_ioctl(struct drm_device *dev, void *data,
+			      struct drm_file *file_priv)
+{
+	struct drm_vmw_shader_arg *arg = (struct drm_vmw_shader_arg *)data;
+	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
+
+	return ttm_ref_object_base_unref(tfile, arg->handle,
+					 TTM_REF_USAGE);
+}
+
+int vmw_shader_define_ioctl(struct drm_device *dev, void *data,
+			     struct drm_file *file_priv)
+{
+	struct vmw_private *dev_priv = vmw_priv(dev);
+	struct vmw_user_shader *ushader;
+	struct vmw_resource *res;
+	struct vmw_resource *tmp;
+	struct drm_vmw_shader_create_arg *arg =
+		(struct drm_vmw_shader_create_arg *)data;
+	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
+	struct vmw_master *vmaster = vmw_master(file_priv->master);
+	struct vmw_dma_buffer *buffer = NULL;
+	SVGA3dShaderType shader_type;
+	int ret;
+
+	if (arg->buffer_handle != SVGA3D_INVALID_ID) {
+		ret = vmw_user_dmabuf_lookup(tfile, arg->buffer_handle,
+					     &buffer);
+		if (unlikely(ret != 0)) {
+			DRM_ERROR("Could not find buffer for shader "
+				  "creation.\n");
+			return ret;
+		}
+
+		if ((u64)buffer->base.num_pages * PAGE_SIZE <
+		    (u64)arg->size + (u64)arg->offset) {
+			DRM_ERROR("Illegal buffer- or shader size.\n");
+			ret = -EINVAL;
+			goto out_bad_arg;
+		}
+	}
+
+	switch (arg->shader_type) {
+	case drm_vmw_shader_type_vs:
+		shader_type = SVGA3D_SHADERTYPE_VS;
+		break;
+	case drm_vmw_shader_type_ps:
+		shader_type = SVGA3D_SHADERTYPE_PS;
+		break;
+	case drm_vmw_shader_type_gs:
+		shader_type = SVGA3D_SHADERTYPE_GS;
+		break;
+	default:
+		DRM_ERROR("Illegal shader type.\n");
+		ret = -EINVAL;
+		goto out_bad_arg;
+	}
+
+	/*
+	 * Approximate idr memory usage with 128 bytes. It will be limited
+	 * by maximum number_of shaders anyway.
+	 */
+
+	if (unlikely(vmw_user_shader_size == 0))
+		vmw_user_shader_size = ttm_round_pot(sizeof(*ushader))
+			+ 128;
+
+	ret = ttm_read_lock(&vmaster->lock, true);
+	if (unlikely(ret != 0))
+		return ret;
+
+	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
+				   vmw_user_shader_size,
+				   false, true);
+	if (unlikely(ret != 0)) {
+		if (ret != -ERESTARTSYS)
+			DRM_ERROR("Out of graphics memory for shader"
+				  " creation.\n");
+		goto out_unlock;
+	}
+
+	ushader = kzalloc(sizeof(*ushader), GFP_KERNEL);
+	if (unlikely(ushader == NULL)) {
+		ttm_mem_global_free(vmw_mem_glob(dev_priv),
+				    vmw_user_shader_size);
+		ret = -ENOMEM;
+		goto out_unlock;
+	}
+
+	res = &ushader->shader.res;
+	ushader->base.shareable = false;
+	ushader->base.tfile = NULL;
+
+	/*
+	 * From here on, the destructor takes over resource freeing.
+	 */
+
+	ret = vmw_gb_shader_init(dev_priv, res, arg->size,
+				 arg->offset, shader_type, buffer,
+				 vmw_user_shader_free);
+	if (unlikely(ret != 0))
+		goto out_unlock;
+
+	tmp = vmw_resource_reference(res);
+	ret = ttm_base_object_init(tfile, &ushader->base, false,
+				   VMW_RES_SHADER,
+				   &vmw_user_shader_base_release, NULL);
+
+	if (unlikely(ret != 0)) {
+		vmw_resource_unreference(&tmp);
+		goto out_err;
+	}
+
+	arg->shader_handle = ushader->base.hash.key;
+out_err:
+	vmw_resource_unreference(&res);
+out_unlock:
+	ttm_read_unlock(&vmaster->lock);
+out_bad_arg:
+	vmw_dmabuf_unreference(&buffer);
+
+	return ret;
+
+}

drivers/gpu/drm/vmwgfx/vmwgfx_surface.c

@@ -41,7 +41,6 @@ struct vmw_user_surface {
 	struct ttm_prime_object prime;
 	struct vmw_surface srf;
 	uint32_t size;
-	uint32_t backup_handle;
 };
 
 /**
@@ -68,6 +67,14 @@ static int vmw_legacy_srf_unbind(struct vmw_resource *res,
 				 struct ttm_validate_buffer *val_buf);
 static int vmw_legacy_srf_create(struct vmw_resource *res);
 static int vmw_legacy_srf_destroy(struct vmw_resource *res);
+static int vmw_gb_surface_create(struct vmw_resource *res);
+static int vmw_gb_surface_bind(struct vmw_resource *res,
+			       struct ttm_validate_buffer *val_buf);
+static int vmw_gb_surface_unbind(struct vmw_resource *res,
+				 bool readback,
+				 struct ttm_validate_buffer *val_buf);
+static int vmw_gb_surface_destroy(struct vmw_resource *res);
+
 
 static const struct vmw_user_resource_conv user_surface_conv = {
 	.object_type = VMW_RES_SURFACE,
@@ -93,6 +100,18 @@ static const struct vmw_res_func vmw_legacy_surface_func = {
 	.unbind = &vmw_legacy_srf_unbind
 };
 
+static const struct vmw_res_func vmw_gb_surface_func = {
+	.res_type = vmw_res_surface,
+	.needs_backup = true,
+	.may_evict = true,
+	.type_name = "guest backed surfaces",
+	.backup_placement = &vmw_mob_placement,
+	.create = vmw_gb_surface_create,
+	.destroy = vmw_gb_surface_destroy,
+	.bind = vmw_gb_surface_bind,
+	.unbind = vmw_gb_surface_unbind
+};
+
 /**
  * struct vmw_surface_dma - SVGA3D DMA command
  */
@@ -291,6 +310,11 @@ static void vmw_hw_surface_destroy(struct vmw_resource *res)
 	struct vmw_surface *srf;
 	void *cmd;
 
+	if (res->func->destroy == vmw_gb_surface_destroy) {
+		(void) vmw_gb_surface_destroy(res);
+		return;
+	}
+
 	if (res->id != -1) {
 
 		cmd = vmw_fifo_reserve(dev_priv, vmw_surface_destroy_size());
@@ -549,12 +573,15 @@ static int vmw_surface_init(struct vmw_private *dev_priv,
 	struct vmw_resource *res = &srf->res;
 
 	BUG_ON(res_free == NULL);
-	(void) vmw_3d_resource_inc(dev_priv, false);
+	if (!dev_priv->has_mob)
+		(void) vmw_3d_resource_inc(dev_priv, false);
 	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)) {
-		vmw_3d_resource_dec(dev_priv, false);
+		if (!dev_priv->has_mob)
+			vmw_3d_resource_dec(dev_priv, false);
 		res_free(res);
 		return ret;
 	}
@@ -750,7 +777,7 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data,
 
 	srf->base_size = *srf->sizes;
 	srf->autogen_filter = SVGA3D_TEX_FILTER_NONE;
-	srf->multisample_count = 1;
+	srf->multisample_count = 0;
 
 	cur_bo_offset = 0;
 	cur_offset = srf->offsets;
@@ -894,3 +921,435 @@ out_no_reference:
 
 	return ret;
 }
+
+/**
+ * vmw_surface_define_encode - Encode a surface_define command.
+ *
+ * @srf: Pointer to a struct vmw_surface object.
+ * @cmd_space: Pointer to memory area in which the commands should be encoded.
+ */
+static int vmw_gb_surface_create(struct vmw_resource *res)
+{
+	struct vmw_private *dev_priv = res->dev_priv;
+	struct vmw_surface *srf = vmw_res_to_srf(res);
+	uint32_t cmd_len, submit_len;
+	int ret;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdDefineGBSurface body;
+	} *cmd;
+
+	if (likely(res->id != -1))
+		return 0;
+
+	(void) vmw_3d_resource_inc(dev_priv, false);
+	ret = vmw_resource_alloc_id(res);
+	if (unlikely(ret != 0)) {
+		DRM_ERROR("Failed to allocate a surface id.\n");
+		goto out_no_id;
+	}
+
+	if (unlikely(res->id >= VMWGFX_NUM_GB_SURFACE)) {
+		ret = -EBUSY;
+		goto out_no_fifo;
+	}
+
+	cmd_len = sizeof(cmd->body);
+	submit_len = sizeof(*cmd);
+	cmd = vmw_fifo_reserve(dev_priv, submit_len);
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Failed reserving FIFO space for surface "
+			  "creation.\n");
+		ret = -ENOMEM;
+		goto out_no_fifo;
+	}
+
+	cmd->header.id = SVGA_3D_CMD_DEFINE_GB_SURFACE;
+	cmd->header.size = cmd_len;
+	cmd->body.sid = srf->res.id;
+	cmd->body.surfaceFlags = srf->flags;
+	cmd->body.format = cpu_to_le32(srf->format);
+	cmd->body.numMipLevels = srf->mip_levels[0];
+	cmd->body.multisampleCount = srf->multisample_count;
+	cmd->body.autogenFilter = srf->autogen_filter;
+	cmd->body.size.width = srf->base_size.width;
+	cmd->body.size.height = srf->base_size.height;
+	cmd->body.size.depth = srf->base_size.depth;
+	vmw_fifo_commit(dev_priv, submit_len);
+
+	return 0;
+
+out_no_fifo:
+	vmw_resource_release_id(res);
+out_no_id:
+	vmw_3d_resource_dec(dev_priv, false);
+	return ret;
+}
+
+
+static int vmw_gb_surface_bind(struct vmw_resource *res,
+			       struct ttm_validate_buffer *val_buf)
+{
+	struct vmw_private *dev_priv = res->dev_priv;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdBindGBSurface body;
+	} *cmd1;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdUpdateGBSurface body;
+	} *cmd2;
+	uint32_t submit_size;
+	struct ttm_buffer_object *bo = val_buf->bo;
+
+	BUG_ON(bo->mem.mem_type != VMW_PL_MOB);
+
+	submit_size = sizeof(*cmd1) + (res->backup_dirty ? sizeof(*cmd2) : 0);
+
+	cmd1 = vmw_fifo_reserve(dev_priv, submit_size);
+	if (unlikely(cmd1 == NULL)) {
+		DRM_ERROR("Failed reserving FIFO space for surface "
+			  "binding.\n");
+		return -ENOMEM;
+	}
+
+	cmd1->header.id = SVGA_3D_CMD_BIND_GB_SURFACE;
+	cmd1->header.size = sizeof(cmd1->body);
+	cmd1->body.sid = res->id;
+	cmd1->body.mobid = bo->mem.start;
+	if (res->backup_dirty) {
+		cmd2 = (void *) &cmd1[1];
+		cmd2->header.id = SVGA_3D_CMD_UPDATE_GB_SURFACE;
+		cmd2->header.size = sizeof(cmd2->body);
+		cmd2->body.sid = res->id;
+		res->backup_dirty = false;
+	}
+	vmw_fifo_commit(dev_priv, submit_size);
+
+	return 0;
+}
+
+static int vmw_gb_surface_unbind(struct vmw_resource *res,
+				 bool readback,
+				 struct ttm_validate_buffer *val_buf)
+{
+	struct vmw_private *dev_priv = res->dev_priv;
+	struct ttm_buffer_object *bo = val_buf->bo;
+	struct vmw_fence_obj *fence;
+
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdReadbackGBSurface body;
+	} *cmd1;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdInvalidateGBSurface body;
+	} *cmd2;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdBindGBSurface body;
+	} *cmd3;
+	uint32_t submit_size;
+	uint8_t *cmd;
+
+
+	BUG_ON(bo->mem.mem_type != VMW_PL_MOB);
+
+	submit_size = sizeof(*cmd3) + (readback ? sizeof(*cmd1) : sizeof(*cmd2));
+	cmd = vmw_fifo_reserve(dev_priv, submit_size);
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Failed reserving FIFO space for surface "
+			  "unbinding.\n");
+		return -ENOMEM;
+	}
+
+	if (readback) {
+		cmd1 = (void *) cmd;
+		cmd1->header.id = SVGA_3D_CMD_READBACK_GB_SURFACE;
+		cmd1->header.size = sizeof(cmd1->body);
+		cmd1->body.sid = res->id;
+		cmd3 = (void *) &cmd1[1];
+	} else {
+		cmd2 = (void *) cmd;
+		cmd2->header.id = SVGA_3D_CMD_INVALIDATE_GB_SURFACE;
+		cmd2->header.size = sizeof(cmd2->body);
+		cmd2->body.sid = res->id;
+		cmd3 = (void *) &cmd2[1];
+	}
+
+	cmd3->header.id = SVGA_3D_CMD_BIND_GB_SURFACE;
+	cmd3->header.size = sizeof(cmd3->body);
+	cmd3->body.sid = res->id;
+	cmd3->body.mobid = SVGA3D_INVALID_ID;
+
+	vmw_fifo_commit(dev_priv, submit_size);
+
+	/*
+	 * Create a fence object and fence the backup buffer.
+	 */
+
+	(void) vmw_execbuf_fence_commands(NULL, dev_priv,
+					  &fence, NULL);
+
+	vmw_fence_single_bo(val_buf->bo, fence);
+
+	if (likely(fence != NULL))
+		vmw_fence_obj_unreference(&fence);
+
+	return 0;
+}
+
+static int vmw_gb_surface_destroy(struct vmw_resource *res)
+{
+	struct vmw_private *dev_priv = res->dev_priv;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdDestroyGBSurface body;
+	} *cmd;
+
+	if (likely(res->id == -1))
+		return 0;
+
+	mutex_lock(&dev_priv->binding_mutex);
+	vmw_context_binding_res_list_kill(&res->binding_head);
+
+	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Failed reserving FIFO space for surface "
+			  "destruction.\n");
+		return -ENOMEM;
+	}
+
+	cmd->header.id = SVGA_3D_CMD_DESTROY_GB_SURFACE;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.sid = res->id;
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+	mutex_unlock(&dev_priv->binding_mutex);
+	vmw_resource_release_id(res);
+	vmw_3d_resource_dec(dev_priv, false);
+
+	return 0;
+}
+
+/**
+ * vmw_gb_surface_define_ioctl - Ioctl function implementing
+ *                               the user surface define functionality.
+ *
+ * @dev:            Pointer to a struct drm_device.
+ * @data:           Pointer to data copied from / to user-space.
+ * @file_priv:      Pointer to a drm file private structure.
+ */
+int vmw_gb_surface_define_ioctl(struct drm_device *dev, void *data,
+				struct drm_file *file_priv)
+{
+	struct vmw_private *dev_priv = vmw_priv(dev);
+	struct vmw_user_surface *user_srf;
+	struct vmw_surface *srf;
+	struct vmw_resource *res;
+	struct vmw_resource *tmp;
+	union drm_vmw_gb_surface_create_arg *arg =
+	    (union drm_vmw_gb_surface_create_arg *)data;
+	struct drm_vmw_gb_surface_create_req *req = &arg->req;
+	struct drm_vmw_gb_surface_create_rep *rep = &arg->rep;
+	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
+	int ret;
+	uint32_t size;
+	struct vmw_master *vmaster = vmw_master(file_priv->master);
+	const struct svga3d_surface_desc *desc;
+	uint32_t backup_handle;
+
+	if (unlikely(vmw_user_surface_size == 0))
+		vmw_user_surface_size = ttm_round_pot(sizeof(*user_srf)) +
+			128;
+
+	size = vmw_user_surface_size + 128;
+
+	desc = svga3dsurface_get_desc(req->format);
+	if (unlikely(desc->block_desc == SVGA3DBLOCKDESC_NONE)) {
+		DRM_ERROR("Invalid surface format for surface creation.\n");
+		return -EINVAL;
+	}
+
+	ret = ttm_read_lock(&vmaster->lock, true);
+	if (unlikely(ret != 0))
+		return ret;
+
+	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
+				   size, false, true);
+	if (unlikely(ret != 0)) {
+		if (ret != -ERESTARTSYS)
+			DRM_ERROR("Out of graphics memory for surface"
+				  " creation.\n");
+		goto out_unlock;
+	}
+
+	user_srf = kzalloc(sizeof(*user_srf), GFP_KERNEL);
+	if (unlikely(user_srf == NULL)) {
+		ret = -ENOMEM;
+		goto out_no_user_srf;
+	}
+
+	srf = &user_srf->srf;
+	res = &srf->res;
+
+	srf->flags = req->svga3d_flags;
+	srf->format = req->format;
+	srf->scanout = req->drm_surface_flags & drm_vmw_surface_flag_scanout;
+	srf->mip_levels[0] = req->mip_levels;
+	srf->num_sizes = 1;
+	srf->sizes = NULL;
+	srf->offsets = NULL;
+	user_srf->size = size;
+	srf->base_size = req->base_size;
+	srf->autogen_filter = SVGA3D_TEX_FILTER_NONE;
+	srf->multisample_count = req->multisample_count;
+	res->backup_size = svga3dsurface_get_serialized_size
+	  (srf->format, srf->base_size, srf->mip_levels[0],
+	   srf->flags & SVGA3D_SURFACE_CUBEMAP);
+
+	user_srf->prime.base.shareable = false;
+	user_srf->prime.base.tfile = NULL;
+
+	/**
+	 * From this point, the generic resource management functions
+	 * destroy the object on failure.
+	 */
+
+	ret = vmw_surface_init(dev_priv, srf, vmw_user_surface_free);
+	if (unlikely(ret != 0))
+		goto out_unlock;
+
+	if (req->buffer_handle != SVGA3D_INVALID_ID) {
+		ret = vmw_user_dmabuf_lookup(tfile, req->buffer_handle,
+					     &res->backup);
+	} else if (req->drm_surface_flags &
+		   drm_vmw_surface_flag_create_buffer)
+		ret = vmw_user_dmabuf_alloc(dev_priv, tfile,
+					    res->backup_size,
+					    req->drm_surface_flags &
+					    drm_vmw_surface_flag_shareable,
+					    &backup_handle,
+					    &res->backup);
+
+	if (unlikely(ret != 0)) {
+		vmw_resource_unreference(&res);
+		goto out_unlock;
+	}
+
+	tmp = vmw_resource_reference(&srf->res);
+	ret = ttm_prime_object_init(tfile, res->backup_size, &user_srf->prime,
+				    req->drm_surface_flags &
+				    drm_vmw_surface_flag_shareable,
+				    VMW_RES_SURFACE,
+				    &vmw_user_surface_base_release, NULL);
+
+	if (unlikely(ret != 0)) {
+		vmw_resource_unreference(&tmp);
+		vmw_resource_unreference(&res);
+		goto out_unlock;
+	}
+
+	rep->handle = user_srf->prime.base.hash.key;
+	rep->backup_size = res->backup_size;
+	if (res->backup) {
+		rep->buffer_map_handle =
+			drm_vma_node_offset_addr(&res->backup->base.vma_node);
+		rep->buffer_size = res->backup->base.num_pages * PAGE_SIZE;
+		rep->buffer_handle = backup_handle;
+	} else {
+		rep->buffer_map_handle = 0;
+		rep->buffer_size = 0;
+		rep->buffer_handle = SVGA3D_INVALID_ID;
+	}
+
+	vmw_resource_unreference(&res);
+
+	ttm_read_unlock(&vmaster->lock);
+	return 0;
+out_no_user_srf:
+	ttm_mem_global_free(vmw_mem_glob(dev_priv), size);
+out_unlock:
+	ttm_read_unlock(&vmaster->lock);
+	return ret;
+}
+
+/**
+ * vmw_gb_surface_reference_ioctl - Ioctl function implementing
+ *                                  the user surface reference functionality.
+ *
+ * @dev:            Pointer to a struct drm_device.
+ * @data:           Pointer to data copied from / to user-space.
+ * @file_priv:      Pointer to a drm file private structure.
+ */
+int vmw_gb_surface_reference_ioctl(struct drm_device *dev, void *data,
+				   struct drm_file *file_priv)
+{
+	struct vmw_private *dev_priv = vmw_priv(dev);
+	union drm_vmw_gb_surface_reference_arg *arg =
+	    (union drm_vmw_gb_surface_reference_arg *)data;
+	struct drm_vmw_surface_arg *req = &arg->req;
+	struct drm_vmw_gb_surface_ref_rep *rep = &arg->rep;
+	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
+	struct vmw_surface *srf;
+	struct vmw_user_surface *user_srf;
+	struct ttm_base_object *base;
+	uint32_t backup_handle;
+	int ret = -EINVAL;
+
+	base = ttm_base_object_lookup_for_ref(dev_priv->tdev, req->sid);
+	if (unlikely(base == NULL)) {
+		DRM_ERROR("Could not find surface to reference.\n");
+		return -EINVAL;
+	}
+
+	if (unlikely(ttm_base_object_type(base) != VMW_RES_SURFACE))
+		goto out_bad_resource;
+
+	user_srf = container_of(base, struct vmw_user_surface, prime.base);
+	srf = &user_srf->srf;
+	if (srf->res.backup == NULL) {
+		DRM_ERROR("Shared GB surface is missing a backup buffer.\n");
+		goto out_bad_resource;
+	}
+
+	ret = ttm_ref_object_add(tfile, &user_srf->prime.base,
+				 TTM_REF_USAGE, NULL);
+	if (unlikely(ret != 0)) {
+		DRM_ERROR("Could not add a reference to a GB surface.\n");
+		goto out_bad_resource;
+	}
+
+	mutex_lock(&dev_priv->cmdbuf_mutex); /* Protect res->backup */
+	ret = vmw_user_dmabuf_reference(tfile, srf->res.backup,
+					&backup_handle);
+	mutex_unlock(&dev_priv->cmdbuf_mutex);
+
+	if (unlikely(ret != 0)) {
+		DRM_ERROR("Could not add a reference to a GB surface "
+			  "backup buffer.\n");
+		(void) ttm_ref_object_base_unref(vmw_fpriv(file_priv)->tfile,
+						 req->sid,
+						 TTM_REF_USAGE);
+		goto out_bad_resource;
+	}
+
+	rep->creq.svga3d_flags = srf->flags;
+	rep->creq.format = srf->format;
+	rep->creq.mip_levels = srf->mip_levels[0];
+	rep->creq.drm_surface_flags = 0;
+	rep->creq.multisample_count = srf->multisample_count;
+	rep->creq.autogen_filter = srf->autogen_filter;
+	rep->creq.buffer_handle = backup_handle;
+	rep->creq.base_size = srf->base_size;
+	rep->crep.handle = user_srf->prime.base.hash.key;
+	rep->crep.backup_size = srf->res.backup_size;
+	rep->crep.buffer_handle = backup_handle;
+	rep->crep.buffer_map_handle =
+		drm_vma_node_offset_addr(&srf->res.backup->base.vma_node);
+	rep->crep.buffer_size = srf->res.backup->base.num_pages * PAGE_SIZE;
+
+out_bad_resource:
+	ttm_base_object_unref(&base);
+
+	return ret;
+}

include/uapi/drm/vmwgfx_drm.h

@@ -28,6 +28,10 @@
 #ifndef __VMWGFX_DRM_H__
 #define __VMWGFX_DRM_H__
 
+#ifndef __KERNEL__
+#include <drm.h>
+#endif
+
 #define DRM_VMW_MAX_SURFACE_FACES 6
 #define DRM_VMW_MAX_MIP_LEVELS 24
 
@@ -55,6 +59,11 @@
 #define DRM_VMW_PRESENT              18
 #define DRM_VMW_PRESENT_READBACK     19
 #define DRM_VMW_UPDATE_LAYOUT        20
+#define DRM_VMW_CREATE_SHADER        21
+#define DRM_VMW_UNREF_SHADER         22
+#define DRM_VMW_GB_SURFACE_CREATE    23
+#define DRM_VMW_GB_SURFACE_REF       24
+#define DRM_VMW_SYNCCPU              25
 
 /*************************************************************************/
 /**
@@ -75,6 +84,9 @@
 #define DRM_VMW_PARAM_FIFO_CAPS        4
 #define DRM_VMW_PARAM_MAX_FB_SIZE      5
 #define DRM_VMW_PARAM_FIFO_HW_VERSION  6
+#define DRM_VMW_PARAM_MAX_SURF_MEMORY  7
+#define DRM_VMW_PARAM_3D_CAPS_SIZE     8
+#define DRM_VMW_PARAM_MAX_MOB_MEMORY   9
 
 /**
  * struct drm_vmw_getparam_arg
@@ -787,4 +799,253 @@ struct drm_vmw_update_layout_arg {
 	uint64_t rects;
 };
 
+
+/*************************************************************************/
+/**
+ * DRM_VMW_CREATE_SHADER - Create shader
+ *
+ * Creates a shader and optionally binds it to a dma buffer containing
+ * the shader byte-code.
+ */
+
+/**
+ * enum drm_vmw_shader_type - Shader types
+ */
+enum drm_vmw_shader_type {
+	drm_vmw_shader_type_vs = 0,
+	drm_vmw_shader_type_ps,
+	drm_vmw_shader_type_gs
+};
+
+
+/**
+ * struct drm_vmw_shader_create_arg
+ *
+ * @shader_type: Shader type of the shader to create.
+ * @size: Size of the byte-code in bytes.
+ * where the shader byte-code starts
+ * @buffer_handle: Buffer handle identifying the buffer containing the
+ * shader byte-code
+ * @shader_handle: On successful completion contains a handle that
+ * can be used to subsequently identify the shader.
+ * @offset: Offset in bytes into the buffer given by @buffer_handle,
+ *
+ * Input / Output argument to the DRM_VMW_CREATE_SHADER Ioctl.
+ */
+struct drm_vmw_shader_create_arg {
+	enum drm_vmw_shader_type shader_type;
+	uint32_t size;
+	uint32_t buffer_handle;
+	uint32_t shader_handle;
+	uint64_t offset;
+};
+
+/*************************************************************************/
+/**
+ * DRM_VMW_UNREF_SHADER - Unreferences a shader
+ *
+ * Destroys a user-space reference to a shader, optionally destroying
+ * it.
+ */
+
+/**
+ * struct drm_vmw_shader_arg
+ *
+ * @handle: Handle identifying the shader to destroy.
+ *
+ * Input argument to the DRM_VMW_UNREF_SHADER ioctl.
+ */
+struct drm_vmw_shader_arg {
+	uint32_t handle;
+	uint32_t pad64;
+};
+
+/*************************************************************************/
+/**
+ * DRM_VMW_GB_SURFACE_CREATE - Create a host guest-backed surface.
+ *
+ * Allocates a surface handle and queues a create surface command
+ * for the host on the first use of the surface. The surface ID can
+ * be used as the surface ID in commands referencing the surface.
+ */
+
+/**
+ * enum drm_vmw_surface_flags
+ *
+ * @drm_vmw_surface_flag_shareable:     Whether the surface is shareable
+ * @drm_vmw_surface_flag_scanout:       Whether the surface is a scanout
+ *                                      surface.
+ * @drm_vmw_surface_flag_create_buffer: Create a backup buffer if none is
+ *                                      given.
+ */
+enum drm_vmw_surface_flags {
+	drm_vmw_surface_flag_shareable = (1 << 0),
+	drm_vmw_surface_flag_scanout = (1 << 1),
+	drm_vmw_surface_flag_create_buffer = (1 << 2)
+};
+
+/**
+ * struct drm_vmw_gb_surface_create_req
+ *
+ * @svga3d_flags:     SVGA3d surface flags for the device.
+ * @format:           SVGA3d format.
+ * @mip_level:        Number of mip levels for all faces.
+ * @drm_surface_flags Flags as described above.
+ * @multisample_count Future use. Set to 0.
+ * @autogen_filter    Future use. Set to 0.
+ * @buffer_handle     Buffer handle of backup buffer. SVGA3D_INVALID_ID
+ *                    if none.
+ * @base_size         Size of the base mip level for all faces.
+ *
+ * Input argument to the  DRM_VMW_GB_SURFACE_CREATE Ioctl.
+ * Part of output argument for the DRM_VMW_GB_SURFACE_REF Ioctl.
+ */
+struct drm_vmw_gb_surface_create_req {
+	uint32_t svga3d_flags;
+	uint32_t format;
+	uint32_t mip_levels;
+	enum drm_vmw_surface_flags drm_surface_flags;
+	uint32_t multisample_count;
+	uint32_t autogen_filter;
+	uint32_t buffer_handle;
+	uint32_t pad64;
+	struct drm_vmw_size base_size;
+};
+
+/**
+ * struct drm_vmw_gb_surface_create_rep
+ *
+ * @handle:            Surface handle.
+ * @backup_size:       Size of backup buffers for this surface.
+ * @buffer_handle:     Handle of backup buffer. SVGA3D_INVALID_ID if none.
+ * @buffer_size:       Actual size of the buffer identified by
+ *                     @buffer_handle
+ * @buffer_map_handle: Offset into device address space for the buffer
+ *                     identified by @buffer_handle.
+ *
+ * Part of output argument for the DRM_VMW_GB_SURFACE_REF ioctl.
+ * Output argument for the DRM_VMW_GB_SURFACE_CREATE ioctl.
+ */
+struct drm_vmw_gb_surface_create_rep {
+	uint32_t handle;
+	uint32_t backup_size;
+	uint32_t buffer_handle;
+	uint32_t buffer_size;
+	uint64_t buffer_map_handle;
+};
+
+/**
+ * union drm_vmw_gb_surface_create_arg
+ *
+ * @req: Input argument as described above.
+ * @rep: Output argument as described above.
+ *
+ * Argument to the DRM_VMW_GB_SURFACE_CREATE ioctl.
+ */
+union drm_vmw_gb_surface_create_arg {
+	struct drm_vmw_gb_surface_create_rep rep;
+	struct drm_vmw_gb_surface_create_req req;
+};
+
+/*************************************************************************/
+/**
+ * DRM_VMW_GB_SURFACE_REF - Reference a host surface.
+ *
+ * Puts a reference on a host surface with a given handle, as previously
+ * returned by the DRM_VMW_GB_SURFACE_CREATE ioctl.
+ * A reference will make sure the surface isn't destroyed while we hold
+ * it and will allow the calling client to use the surface handle in
+ * the command stream.
+ *
+ * On successful return, the Ioctl returns the surface information given
+ * to and returned from the DRM_VMW_GB_SURFACE_CREATE ioctl.
+ */
+
+/**
+ * struct drm_vmw_gb_surface_reference_arg
+ *
+ * @creq: The data used as input when the surface was created, as described
+ *        above at "struct drm_vmw_gb_surface_create_req"
+ * @crep: Additional data output when the surface was created, as described
+ *        above at "struct drm_vmw_gb_surface_create_rep"
+ *
+ * Output Argument to the DRM_VMW_GB_SURFACE_REF ioctl.
+ */
+struct drm_vmw_gb_surface_ref_rep {
+	struct drm_vmw_gb_surface_create_req creq;
+	struct drm_vmw_gb_surface_create_rep crep;
+};
+
+/**
+ * union drm_vmw_gb_surface_reference_arg
+ *
+ * @req: Input data as described above at "struct drm_vmw_surface_arg"
+ * @rep: Output data as described above at "struct drm_vmw_gb_surface_ref_rep"
+ *
+ * Argument to the DRM_VMW_GB_SURFACE_REF Ioctl.
+ */
+union drm_vmw_gb_surface_reference_arg {
+	struct drm_vmw_gb_surface_ref_rep rep;
+	struct drm_vmw_surface_arg req;
+};
+
+
+/*************************************************************************/
+/**
+ * DRM_VMW_SYNCCPU - Sync a DMA buffer / MOB for CPU access.
+ *
+ * Idles any previously submitted GPU operations on the buffer and
+ * by default blocks command submissions that reference the buffer.
+ * If the file descriptor used to grab a blocking CPU sync is closed, the
+ * cpu sync is released.
+ * The flags argument indicates how the grab / release operation should be
+ * performed:
+ */
+
+/**
+ * enum drm_vmw_synccpu_flags - Synccpu flags:
+ *
+ * @drm_vmw_synccpu_read: Sync for read. If sync is done for read only, it's a
+ * hint to the kernel to allow command submissions that references the buffer
+ * for read-only.
+ * @drm_vmw_synccpu_write: Sync for write. Block all command submissions
+ * referencing this buffer.
+ * @drm_vmw_synccpu_dontblock: Dont wait for GPU idle, but rather return
+ * -EBUSY should the buffer be busy.
+ * @drm_vmw_synccpu_allow_cs: Allow command submission that touches the buffer
+ * while the buffer is synced for CPU. This is similar to the GEM bo idle
+ * behavior.
+ */
+enum drm_vmw_synccpu_flags {
+	drm_vmw_synccpu_read = (1 << 0),
+	drm_vmw_synccpu_write = (1 << 1),
+	drm_vmw_synccpu_dontblock = (1 << 2),
+	drm_vmw_synccpu_allow_cs = (1 << 3)
+};
+
+/**
+ * enum drm_vmw_synccpu_op - Synccpu operations:
+ *
+ * @drm_vmw_synccpu_grab:    Grab the buffer for CPU operations
+ * @drm_vmw_synccpu_release: Release a previous grab.
+ */
+enum drm_vmw_synccpu_op {
+	drm_vmw_synccpu_grab,
+	drm_vmw_synccpu_release
+};
+
+/**
+ * struct drm_vmw_synccpu_arg
+ *
+ * @op:			     The synccpu operation as described above.
+ * @handle:		     Handle identifying the buffer object.
+ * @flags:		     Flags as described above.
+ */
+struct drm_vmw_synccpu_arg {
+	enum drm_vmw_synccpu_op op;
+	enum drm_vmw_synccpu_flags flags;
+	uint32_t handle;
+	uint32_t pad64;
+};
+
 #endif