diff options
Diffstat (limited to 'drivers/gpu/drm/vmwgfx')
26 files changed, 6609 insertions, 650 deletions
diff --git a/drivers/gpu/drm/vmwgfx/Kconfig b/drivers/gpu/drm/vmwgfx/Kconfig index b71bcd0bfbb..67720f70fe2 100644 --- a/drivers/gpu/drm/vmwgfx/Kconfig +++ b/drivers/gpu/drm/vmwgfx/Kconfig @@ -1,11 +1,14 @@ config DRM_VMWGFX tristate "DRM driver for VMware Virtual GPU" - depends on DRM && PCI && FB + depends on DRM && PCI select FB_DEFERRED_IO select FB_CFB_FILLRECT select FB_CFB_COPYAREA select FB_CFB_IMAGEBLIT select DRM_TTM + # Only needed for the transitional use of drm_crtc_init - can be removed + # again once vmwgfx sets up the primary plane itself. + select DRM_KMS_HELPER help Choose this option if you would like to run 3D acceleration in a VMware virtual machine. @@ -14,7 +17,7 @@ config DRM_VMWGFX The compiled module will be called "vmwgfx.ko". config DRM_VMWGFX_FBCON - depends on DRM_VMWGFX + depends on DRM_VMWGFX && FB bool "Enable framebuffer console under vmwgfx by default" help Choose this option if you are shipping a new vmwgfx diff --git a/drivers/gpu/drm/vmwgfx/Makefile b/drivers/gpu/drm/vmwgfx/Makefile index 2cc6cd91ac1..458cdf6d81e 100644 --- a/drivers/gpu/drm/vmwgfx/Makefile +++ b/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_surface.o vmwgfx_prime.o vmwgfx_mob.o vmwgfx_shader.o obj-$(CONFIG_DRM_VMWGFX) := vmwgfx.o diff --git a/drivers/gpu/drm/vmwgfx/svga3d_reg.h b/drivers/gpu/drm/vmwgfx/svga3d_reg.h index d0e085ee824..f58dc7dd15c 100644 --- a/drivers/gpu/drm/vmwgfx/svga3d_reg.h +++ b/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 +159,109 @@ 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, + + SVGA3D_FORMAT_MAX = 122, } SVGA3dSurfaceFormat; typedef uint32 SVGA3dColor; /* a, r, g, b */ @@ -957,15 +1051,21 @@ 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 +1156,84 @@ 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_GB_SCREEN_DMA 1131 +#define SVGA_3D_CMD_BIND_GB_SURFACE_WITH_PITCH 1132 +#define SVGA_3D_CMD_GB_MOB_FENCE 1133 +#define SVGA_3D_CMD_DEFINE_GB_SURFACE_V2 1134 +#define SVGA_3D_CMD_DEFINE_GB_MOB64 1135 +#define SVGA_3D_CMD_REDEFINE_GB_MOB64 1136 +#define SVGA_3D_CMD_NOP_ERROR 1137 + +#define SVGA_3D_CMD_RESERVED1 1138 +#define SVGA_3D_CMD_RESERVED2 1139 +#define SVGA_3D_CMD_RESERVED3 1140 +#define SVGA_3D_CMD_RESERVED4 1141 +#define SVGA_3D_CMD_RESERVED5 1142 + +#define SVGA_3D_CMD_MAX 1142 +#define SVGA_3D_CMD_FUTURE_MAX 3000 /* * Common substructures used in multiple FIFO commands: @@ -1750,6 +1925,507 @@ struct { /* + * 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; +} __packed +SVGA3dCmdSetOTableBase; /* SVGA_3D_CMD_SET_OTABLE_BASE */ + +typedef +struct { + SVGAOTableType type; + PPN64 baseAddress; + uint32 sizeInBytes; + uint32 validSizeInBytes; + SVGAMobFormat ptDepth; +} __packed +SVGA3dCmdSetOTableBase64; /* SVGA_3D_CMD_SET_OTABLE_BASE64 */ + +typedef +struct { + SVGAOTableType type; +} __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; +} __packed +SVGA3dCmdDefineGBMob; /* SVGA_3D_CMD_DEFINE_GB_MOB */ + + +/* + * Destroys an object in the OTable. + */ + +typedef +struct SVGA3dCmdDestroyGBMob { + SVGAMobId mobid; +} __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; +} __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; +} __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; +} __packed +SVGA3dCmdRedefineGBMob64; /* SVGA_3D_CMD_REDEFINE_GB_MOB64 */ + +/* + * Notification that the page tables have been modified. + */ + +typedef +struct SVGA3dCmdUpdateGBMobMapping { + SVGAMobId mobid; +} __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; +} __packed +SVGA3dCmdDefineGBSurface; /* SVGA_3D_CMD_DEFINE_GB_SURFACE */ + +/* + * Destroy a guest-backed surface. + */ + +typedef +struct SVGA3dCmdDestroyGBSurface { + uint32 sid; +} __packed +SVGA3dCmdDestroyGBSurface; /* SVGA_3D_CMD_DESTROY_GB_SURFACE */ + +/* + * Bind a guest-backed surface to an object. + */ + +typedef +struct SVGA3dCmdBindGBSurface { + uint32 sid; + SVGAMobId mobid; +} __packed +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; +} __packed +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; +} __packed +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; +} __packed +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; +} __packed +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; +} __packed +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; +} __packed +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; +} __packed +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; +} __packed +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; +} __packed +SVGA3dCmdInvalidateGBImagePartial; /* SVGA_3D_CMD_INVALIDATE_GB_IMAGE_PARTIAL */ + +/* + * Define a guest-backed context. + */ + +typedef +struct SVGA3dCmdDefineGBContext { + uint32 cid; +} __packed +SVGA3dCmdDefineGBContext; /* SVGA_3D_CMD_DEFINE_GB_CONTEXT */ + +/* + * Destroy a guest-backed context. + */ + +typedef +struct SVGA3dCmdDestroyGBContext { + uint32 cid; +} __packed +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; +} __packed +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; +} __packed +SVGA3dCmdReadbackGBContext; /* SVGA_3D_CMD_READBACK_GB_CONTEXT */ + +/* + * Invalidate a guest-backed context. + */ +typedef +struct SVGA3dCmdInvalidateGBContext { + uint32 cid; +} __packed +SVGA3dCmdInvalidateGBContext; /* SVGA_3D_CMD_INVALIDATE_GB_CONTEXT */ + +/* + * Define a guest-backed shader. + */ + +typedef +struct SVGA3dCmdDefineGBShader { + uint32 shid; + SVGA3dShaderType type; + uint32 sizeInBytes; +} __packed +SVGA3dCmdDefineGBShader; /* SVGA_3D_CMD_DEFINE_GB_SHADER */ + +/* + * Bind a guest-backed shader. + */ + +typedef struct SVGA3dCmdBindGBShader { + uint32 shid; + SVGAMobId mobid; + uint32 offsetInBytes; +} __packed +SVGA3dCmdBindGBShader; /* SVGA_3D_CMD_BIND_GB_SHADER */ + +/* + * Destroy a guest-backed shader. + */ + +typedef struct SVGA3dCmdDestroyGBShader { + uint32 shid; +} __packed +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. + */ +} __packed +SVGA3dCmdSetGBShaderConstInline; +/* SVGA_3D_CMD_SET_GB_SHADERCONSTS_INLINE */ + +typedef +struct { + uint32 cid; + SVGA3dQueryType type; +} __packed +SVGA3dCmdBeginGBQuery; /* SVGA_3D_CMD_BEGIN_GB_QUERY */ + +typedef +struct { + uint32 cid; + SVGA3dQueryType type; + SVGAMobId mobid; + uint32 offset; +} __packed +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; +} __packed +SVGA3dCmdWaitForGBQuery; /* SVGA_3D_CMD_WAIT_FOR_GB_QUERY */ + +typedef +struct { + SVGAMobId mobid; + uint32 fbOffset; + uint32 initalized; +} __packed +SVGA3dCmdEnableGart; /* SVGA_3D_CMD_ENABLE_GART */ + +typedef +struct { + SVGAMobId mobid; + uint32 gartOffset; +} __packed +SVGA3dCmdMapMobIntoGart; /* SVGA_3D_CMD_MAP_MOB_INTO_GART */ + + +typedef +struct { + uint32 gartOffset; + uint32 numPages; +} __packed +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; +} __packed +SVGA3dCmdDefineGBScreenTarget; /* SVGA_3D_CMD_DEFINE_GB_SCREENTARGET */ + +typedef +struct { + uint32 stid; +} __packed +SVGA3dCmdDestroyGBScreenTarget; /* SVGA_3D_CMD_DESTROY_GB_SCREENTARGET */ + +typedef +struct { + uint32 stid; + SVGA3dSurfaceImageId image; +} __packed +SVGA3dCmdBindGBScreenTarget; /* SVGA_3D_CMD_BIND_GB_SCREENTARGET */ + +typedef +struct { + uint32 stid; + SVGA3dBox box; +} __packed +SVGA3dCmdUpdateGBScreenTarget; /* SVGA_3D_CMD_UPDATE_GB_SCREENTARGET */ + +/* * Capability query index. * * Notes: @@ -1879,10 +2555,41 @@ 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; @@ -1893,4 +2600,28 @@ typedef union { float f; } SVGA3dDevCapResult; +typedef enum { + SVGA3DCAPS_RECORD_UNKNOWN = 0, + SVGA3DCAPS_RECORD_DEVCAPS_MIN = 0x100, + SVGA3DCAPS_RECORD_DEVCAPS = 0x100, + SVGA3DCAPS_RECORD_DEVCAPS_MAX = 0x1ff, +} SVGA3dCapsRecordType; + +typedef +struct SVGA3dCapsRecordHeader { + uint32 length; + SVGA3dCapsRecordType type; +} +SVGA3dCapsRecordHeader; + +typedef +struct SVGA3dCapsRecord { + SVGA3dCapsRecordHeader header; + uint32 data[1]; +} +SVGA3dCapsRecord; + + +typedef uint32 SVGA3dCapPair[2]; + #endif /* _SVGA3D_REG_H_ */ diff --git a/drivers/gpu/drm/vmwgfx/svga3d_surfacedefs.h b/drivers/gpu/drm/vmwgfx/svga3d_surfacedefs.h index 8369c3ba10f..ef338509614 100644 --- a/drivers/gpu/drm/vmwgfx/svga3d_surfacedefs.h +++ b/drivers/gpu/drm/vmwgfx/svga3d_surfacedefs.h @@ -38,8 +38,11 @@ #define DIV_ROUND_UP(x, y) (((x) + (y) - 1) / (y)) #define max_t(type, x, y) ((x) > (y) ? (x) : (y)) +#define min_t(type, x, y) ((x) < (y) ? (x) : (y)) #define surf_size_struct SVGA3dSize #define u32 uint32 +#define u64 uint64_t +#define U32_MAX ((u32)~0U) #endif /* __KERNEL__ */ @@ -704,8 +707,8 @@ static const struct svga3d_surface_desc svga3d_surface_descs[] = { static inline u32 clamped_umul32(u32 a, u32 b) { - uint64_t tmp = (uint64_t) a*b; - return (tmp > (uint64_t) ((u32) -1)) ? (u32) -1 : tmp; + u64 tmp = (u64) a*b; + return (tmp > (u64) U32_MAX) ? U32_MAX : tmp; } static inline const struct svga3d_surface_desc * @@ -834,7 +837,7 @@ svga3dsurface_get_serialized_size(SVGA3dSurfaceFormat format, bool cubemap) { const struct svga3d_surface_desc *desc = svga3dsurface_get_desc(format); - u32 total_size = 0; + u64 total_size = 0; u32 mip; for (mip = 0; mip < num_mip_levels; mip++) { @@ -847,7 +850,7 @@ svga3dsurface_get_serialized_size(SVGA3dSurfaceFormat format, if (cubemap) total_size *= SVGA3D_MAX_SURFACE_FACES; - return total_size; + return (u32) min_t(u64, total_size, (u64) U32_MAX); } diff --git a/drivers/gpu/drm/vmwgfx/svga_reg.h b/drivers/gpu/drm/vmwgfx/svga_reg.h index 01f63cb4967..11323dd5196 100644 --- a/drivers/gpu/drm/vmwgfx/svga_reg.h +++ b/drivers/gpu/drm/vmwgfx/svga_reg.h @@ -169,7 +169,17 @@ 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_COMMAND_LOW = 48, /* Lower 32 bits and submits commands */ + SVGA_REG_COMMAND_HIGH = 49, /* Upper 32 bits of command buffer PA */ + 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_CMD_PREPEND_LOW = 53, + SVGA_REG_CMD_PREPEND_HIGH = 54, + SVGA_REG_SCREENTARGET_MAX_WIDTH = 55, + SVGA_REG_SCREENTARGET_MAX_HEIGHT = 56, + SVGA_REG_MOB_MAX_SIZE = 57, + SVGA_REG_TOP = 58, /* 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 +441,10 @@ 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. diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c index 96dc84dc34d..6327cfc3680 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c +++ b/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 @@ -140,71 +157,527 @@ 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_tt ttm; + 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; + bool mapped; }; +const size_t vmw_tt_size = sizeof(struct vmw_ttm_tt); + +/** + * Helper functions to advance a struct vmw_piter iterator. + * + * @viter: Pointer to the iterator. + * + * These functions return false if past the end of the list, + * true otherwise. Functions are selected depending on the current + * DMA mapping mode. + */ +static bool __vmw_piter_non_sg_next(struct vmw_piter *viter) +{ + return ++(viter->i) < viter->num_pages; +} + +static bool __vmw_piter_sg_next(struct vmw_piter *viter) +{ + return __sg_page_iter_next(&viter->iter); +} + + +/** + * Helper functions to return a pointer to the current page. + * + * @viter: Pointer to the iterator + * + * These functions return a pointer to the page currently + * pointed to by @viter. Functions are selected depending on the + * current mapping mode. + */ +static struct page *__vmw_piter_non_sg_page(struct vmw_piter *viter) +{ + return viter->pages[viter->i]; +} + +static struct page *__vmw_piter_sg_page(struct vmw_piter *viter) +{ + return sg_page_iter_page(&viter->iter); +} + + +/** + * Helper functions to return the DMA address of the current page. + * + * @viter: Pointer to the iterator + * + * These functions return the DMA address of the page currently + * pointed to by @viter. Functions are selected depending on the + * current mapping mode. + */ +static dma_addr_t __vmw_piter_phys_addr(struct vmw_piter *viter) +{ + return page_to_phys(viter->pages[viter->i]); +} + +static dma_addr_t __vmw_piter_dma_addr(struct vmw_piter *viter) +{ + return viter->addrs[viter->i]; +} + +static dma_addr_t __vmw_piter_sg_addr(struct vmw_piter *viter) +{ + return sg_page_iter_dma_address(&viter->iter); +} + + +/** + * vmw_piter_start - Initialize a struct vmw_piter. + * + * @viter: Pointer to the iterator to initialize + * @vsgt: Pointer to a struct vmw_sg_table to initialize from + * + * Note that we're following the convention of __sg_page_iter_start, so that + * the iterator doesn't point to a valid page after initialization; it has + * to be advanced one step first. + */ +void vmw_piter_start(struct vmw_piter *viter, const struct vmw_sg_table *vsgt, + unsigned long p_offset) +{ + viter->i = p_offset - 1; + viter->num_pages = vsgt->num_pages; + switch (vsgt->mode) { + case vmw_dma_phys: + viter->next = &__vmw_piter_non_sg_next; + viter->dma_address = &__vmw_piter_phys_addr; + viter->page = &__vmw_piter_non_sg_page; + viter->pages = vsgt->pages; + break; + case vmw_dma_alloc_coherent: + viter->next = &__vmw_piter_non_sg_next; + viter->dma_address = &__vmw_piter_dma_addr; + viter->page = &__vmw_piter_non_sg_page; + viter->addrs = vsgt->addrs; + viter->pages = vsgt->pages; + break; + case vmw_dma_map_populate: + case vmw_dma_map_bind: + viter->next = &__vmw_piter_sg_next; + viter->dma_address = &__vmw_piter_sg_addr; + viter->page = &__vmw_piter_sg_page; + __sg_page_iter_start(&viter->iter, vsgt->sgt->sgl, + vsgt->sgt->orig_nents, p_offset); + break; + default: + BUG(); + } +} + +/** + * vmw_ttm_unmap_from_dma - unmap device addresses previsouly mapped for + * TTM pages + * + * @vmw_tt: Pointer to a struct vmw_ttm_backend + * + * Used to free dma mappings previously mapped by vmw_ttm_map_for_dma. + */ +static void vmw_ttm_unmap_from_dma(struct vmw_ttm_tt *vmw_tt) +{ + struct device *dev = vmw_tt->dev_priv->dev->dev; + + dma_unmap_sg(dev, vmw_tt->sgt.sgl, vmw_tt->sgt.nents, + DMA_BIDIRECTIONAL); + vmw_tt->sgt.nents = vmw_tt->sgt.orig_nents; +} + +/** + * vmw_ttm_map_for_dma - map TTM pages to get device addresses + * + * @vmw_tt: Pointer to a struct vmw_ttm_backend + * + * This function is used to get device addresses from the kernel DMA layer. + * However, it's violating the DMA API in that when this operation has been + * performed, it's illegal for the CPU to write to the pages without first + * unmapping the DMA mappings, or calling dma_sync_sg_for_cpu(). It is + * therefore only legal to call this function if we know that the function + * dma_sync_sg_for_cpu() is a NOP, and dma_sync_sg_for_device() is at most + * a CPU write buffer flush. + */ +static int vmw_ttm_map_for_dma(struct vmw_ttm_tt *vmw_tt) +{ + struct device *dev = vmw_tt->dev_priv->dev->dev; + int ret; + + ret = dma_map_sg(dev, vmw_tt->sgt.sgl, vmw_tt->sgt.orig_nents, + DMA_BIDIRECTIONAL); + if (unlikely(ret == 0)) + return -ENOMEM; + + vmw_tt->sgt.nents = ret; + + return 0; +} + +/** + * vmw_ttm_map_dma - Make sure TTM pages are visible to the device + * + * @vmw_tt: Pointer to a struct vmw_ttm_tt + * + * Select the correct function for and make sure the TTM pages are + * visible to the device. Allocate storage for the device mappings. + * If a mapping has already been performed, indicated by the storage + * pointer being non NULL, the function returns success. + */ +static int vmw_ttm_map_dma(struct vmw_ttm_tt *vmw_tt) +{ + struct vmw_private *dev_priv = vmw_tt->dev_priv; + struct ttm_mem_global *glob = vmw_mem_glob(dev_priv); + struct vmw_sg_table *vsgt = &vmw_tt->vsgt; + struct vmw_piter iter; + dma_addr_t old; + int ret = 0; + static size_t sgl_size; + static size_t sgt_size; + + if (vmw_tt->mapped) + return 0; + + vsgt->mode = dev_priv->map_mode; + vsgt->pages = vmw_tt->dma_ttm.ttm.pages; + vsgt->num_pages = vmw_tt->dma_ttm.ttm.num_pages; + vsgt->addrs = vmw_tt->dma_ttm.dma_address; + vsgt->sgt = &vmw_tt->sgt; + + switch (dev_priv->map_mode) { + case vmw_dma_map_bind: + case vmw_dma_map_populate: + if (unlikely(!sgl_size)) { + sgl_size = ttm_round_pot(sizeof(struct scatterlist)); + sgt_size = ttm_round_pot(sizeof(struct sg_table)); + } + vmw_tt->sg_alloc_size = sgt_size + sgl_size * vsgt->num_pages; + ret = ttm_mem_global_alloc(glob, vmw_tt->sg_alloc_size, false, + true); + if (unlikely(ret != 0)) + return ret; + + ret = sg_alloc_table_from_pages(&vmw_tt->sgt, vsgt->pages, + vsgt->num_pages, 0, + (unsigned long) + vsgt->num_pages << PAGE_SHIFT, + GFP_KERNEL); + if (unlikely(ret != 0)) + goto out_sg_alloc_fail; + + if (vsgt->num_pages > vmw_tt->sgt.nents) { + uint64_t over_alloc = + sgl_size * (vsgt->num_pages - + vmw_tt->sgt.nents); + + ttm_mem_global_free(glob, over_alloc); + vmw_tt->sg_alloc_size -= over_alloc; + } + + ret = vmw_ttm_map_for_dma(vmw_tt); + if (unlikely(ret != 0)) + goto out_map_fail; + + break; + default: + break; + } + + old = ~((dma_addr_t) 0); + vmw_tt->vsgt.num_regions = 0; + for (vmw_piter_start(&iter, vsgt, 0); vmw_piter_next(&iter);) { + dma_addr_t cur = vmw_piter_dma_addr(&iter); + + if (cur != old + PAGE_SIZE) + vmw_tt->vsgt.num_regions++; + old = cur; + } + + vmw_tt->mapped = true; + return 0; + +out_map_fail: + sg_free_table(vmw_tt->vsgt.sgt); + vmw_tt->vsgt.sgt = NULL; +out_sg_alloc_fail: + ttm_mem_global_free(glob, vmw_tt->sg_alloc_size); + return ret; +} + +/** + * vmw_ttm_unmap_dma - Tear down any TTM page device mappings + * + * @vmw_tt: Pointer to a struct vmw_ttm_tt + * + * Tear down any previously set up device DMA mappings and free + * any storage space allocated for them. If there are no mappings set up, + * this function is a NOP. + */ +static void vmw_ttm_unmap_dma(struct vmw_ttm_tt *vmw_tt) +{ + struct vmw_private *dev_priv = vmw_tt->dev_priv; + + if (!vmw_tt->vsgt.sgt) + return; + + switch (dev_priv->map_mode) { + case vmw_dma_map_bind: + case vmw_dma_map_populate: + vmw_ttm_unmap_from_dma(vmw_tt); + sg_free_table(vmw_tt->vsgt.sgt); + vmw_tt->vsgt.sgt = NULL; + ttm_mem_global_free(vmw_mem_glob(dev_priv), + vmw_tt->sg_alloc_size); + break; + default: + break; + } + vmw_tt->mapped = false; +} + + +/** + * 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 = container_of(ttm, struct vmw_ttm_tt, ttm); + struct vmw_ttm_tt *vmw_be = + container_of(ttm, struct vmw_ttm_tt, dma_ttm.ttm); + int ret; + + ret = vmw_ttm_map_dma(vmw_be); + if (unlikely(ret != 0)) + return ret; vmw_be->gmr_id = bo_mem->start; + vmw_be->mem_type = bo_mem->mem_type; - return vmw_gmr_bind(vmw_be->dev_priv, ttm->pages, - ttm->num_pages, vmw_be->gmr_id); + 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_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) { - struct vmw_ttm_tt *vmw_be = container_of(ttm, struct vmw_ttm_tt, ttm); + struct vmw_ttm_tt *vmw_be = + container_of(ttm, struct vmw_ttm_tt, dma_ttm.ttm); + + 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); - vmw_gmr_unbind(vmw_be->dev_priv, vmw_be->gmr_id); return 0; } + static void vmw_ttm_destroy(struct ttm_tt *ttm) { - struct vmw_ttm_tt *vmw_be = container_of(ttm, struct vmw_ttm_tt, ttm); + struct vmw_ttm_tt *vmw_be = + container_of(ttm, struct vmw_ttm_tt, dma_ttm.ttm); + + vmw_ttm_unmap_dma(vmw_be); + if (vmw_be->dev_priv->map_mode == vmw_dma_alloc_coherent) + ttm_dma_tt_fini(&vmw_be->dma_ttm); + else + ttm_tt_fini(ttm); + + if (vmw_be->mob) + vmw_mob_destroy(vmw_be->mob); - ttm_tt_fini(ttm); kfree(vmw_be); } + +static int vmw_ttm_populate(struct ttm_tt *ttm) +{ + struct vmw_ttm_tt *vmw_tt = + container_of(ttm, struct vmw_ttm_tt, dma_ttm.ttm); + struct vmw_private *dev_priv = vmw_tt->dev_priv; + struct ttm_mem_global *glob = vmw_mem_glob(dev_priv); + int ret; + + if (ttm->state != tt_unpopulated) + return 0; + + if (dev_priv->map_mode == vmw_dma_alloc_coherent) { + size_t size = + ttm_round_pot(ttm->num_pages * sizeof(dma_addr_t)); + ret = ttm_mem_global_alloc(glob, size, false, true); + if (unlikely(ret != 0)) + return ret; + + ret = ttm_dma_populate(&vmw_tt->dma_ttm, dev_priv->dev->dev); + if (unlikely(ret != 0)) + ttm_mem_global_free(glob, size); + } else + ret = ttm_pool_populate(ttm); + + return ret; +} + +static void vmw_ttm_unpopulate(struct ttm_tt *ttm) +{ + struct vmw_ttm_tt *vmw_tt = container_of(ttm, struct vmw_ttm_tt, + dma_ttm.ttm); + struct vmw_private *dev_priv = vmw_tt->dev_priv; + struct ttm_mem_global *glob = vmw_mem_glob(dev_priv); + + + 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 = + ttm_round_pot(ttm->num_pages * sizeof(dma_addr_t)); + + ttm_dma_unpopulate(&vmw_tt->dma_ttm, dev_priv->dev->dev); + ttm_mem_global_free(glob, size); + } else + ttm_pool_unpopulate(ttm); +} + static struct ttm_backend_func vmw_ttm_func = { .bind = vmw_ttm_bind, .unbind = vmw_ttm_unbind, .destroy = vmw_ttm_destroy, }; -struct ttm_tt *vmw_ttm_tt_create(struct ttm_bo_device *bdev, +static struct ttm_tt *vmw_ttm_tt_create(struct ttm_bo_device *bdev, unsigned long size, uint32_t page_flags, struct page *dummy_read_page) { struct vmw_ttm_tt *vmw_be; + int ret; - vmw_be = kmalloc(sizeof(*vmw_be), GFP_KERNEL); + vmw_be = kzalloc(sizeof(*vmw_be), GFP_KERNEL); if (!vmw_be) return NULL; - vmw_be->ttm.func = &vmw_ttm_func; + vmw_be->dma_ttm.ttm.func = &vmw_ttm_func; vmw_be->dev_priv = container_of(bdev, struct vmw_private, bdev); - - if (ttm_tt_init(&vmw_be->ttm, bdev, size, page_flags, dummy_read_page)) { - kfree(vmw_be); - return NULL; - } - - return &vmw_be->ttm; + 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, + dummy_read_page); + else + ret = ttm_tt_init(&vmw_be->dma_ttm.ttm, bdev, size, page_flags, + dummy_read_page); + if (unlikely(ret != 0)) + goto out_no_init; + + return &vmw_be->dma_ttm.ttm; +out_no_init: + kfree(vmw_be); + return NULL; } -int vmw_invalidate_caches(struct ttm_bo_device *bdev, uint32_t flags) +static int vmw_invalidate_caches(struct ttm_bo_device *bdev, uint32_t flags) { return 0; } -int vmw_init_mem_type(struct ttm_bo_device *bdev, uint32_t type, +static int vmw_init_mem_type(struct ttm_bo_device *bdev, uint32_t type, struct ttm_mem_type_manager *man) { switch (type) { @@ -224,6 +697,7 @@ 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 @@ -242,7 +716,7 @@ int vmw_init_mem_type(struct ttm_bo_device *bdev, uint32_t type, return 0; } -void vmw_evict_flags(struct ttm_buffer_object *bo, +static void vmw_evict_flags(struct ttm_buffer_object *bo, struct ttm_placement *placement) { *placement = vmw_sys_placement; @@ -271,6 +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; @@ -330,10 +805,42 @@ 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 = &ttm_pool_populate, - .ttm_tt_unpopulate = &ttm_pool_unpopulate, + .ttm_tt_populate = &vmw_ttm_populate, + .ttm_tt_unpopulate = &vmw_ttm_unpopulate, .invalidate_caches = vmw_invalidate_caches, .init_mem_type = vmw_init_mem_type, .evict_flags = vmw_evict_flags, @@ -344,8 +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, diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_context.c b/drivers/gpu/drm/vmwgfx/vmwgfx_context.c index 00ae0925aca..8bb26dcd9ea 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_context.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_context.c @@ -32,12 +32,30 @@ 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 *, bool); + 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, bool rebind); +static int vmw_context_scrub_render_target(struct vmw_ctx_bindinfo *bi, + bool rebind); +static int vmw_context_scrub_texture(struct vmw_ctx_bindinfo *bi, bool rebind); +static void vmw_context_binding_state_scrub(struct vmw_ctx_binding_state *cbs); +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 +80,23 @@ 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 +111,20 @@ 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); + mutex_lock(&dev_priv->binding_mutex); + (void) vmw_context_binding_state_kill + (&container_of(res, struct vmw_user_context, res)->cbs); + (void) vmw_gb_context_destroy(res); + mutex_unlock(&dev_priv->binding_mutex); + 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 +141,33 @@ 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 +179,9 @@ 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 +233,180 @@ 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_scrub(&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; + + 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: */ @@ -209,7 +462,6 @@ int vmw_context_define_ioctl(struct drm_device *dev, void *data, struct vmw_resource *tmp; struct drm_vmw_context_arg *arg = (struct drm_vmw_context_arg *)data; struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile; - struct vmw_master *vmaster = vmw_master(file_priv->master); int ret; @@ -221,7 +473,7 @@ int vmw_context_define_ioctl(struct drm_device *dev, void *data, if (unlikely(vmw_user_context_size == 0)) vmw_user_context_size = ttm_round_pot(sizeof(*ctx)) + 128; - ret = ttm_read_lock(&vmaster->lock, true); + ret = ttm_read_lock(&dev_priv->reservation_sem, true); if (unlikely(ret != 0)) return ret; @@ -268,7 +520,384 @@ int vmw_context_define_ioctl(struct drm_device *dev, void *data, out_err: vmw_resource_unreference(&res); out_unlock: - ttm_read_unlock(&vmaster->lock); + ttm_read_unlock(&dev_priv->reservation_sem); return ret; } + +/** + * vmw_context_scrub_shader - scrub a shader binding from a context. + * + * @bi: single binding information. + * @rebind: Whether to issue a bind instead of scrub command. + */ +static int vmw_context_scrub_shader(struct vmw_ctx_bindinfo *bi, bool rebind) +{ + 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 = ((rebind) ? bi->res->id : 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. + * @rebind: Whether to issue a bind instead of scrub command. + */ +static int vmw_context_scrub_render_target(struct vmw_ctx_bindinfo *bi, + bool rebind) +{ + 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 = ((rebind) ? bi->res->id : 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. + * @rebind: Whether to issue a bind instead of scrub command. + * + * 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, + bool rebind) +{ + 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 = ((rebind) ? bi->res->id : 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; + loc->bi.scrubbed = false; + 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); + + if (bi->res != NULL) { + loc->bi = *bi; + list_add_tail(&loc->ctx_list, &cbs->list); + list_add_tail(&loc->res_list, &bi->res->binding_head); + } +} + +/** + * 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. + */ +static void vmw_context_binding_kill(struct vmw_ctx_binding *cb) +{ + if (!cb->bi.scrubbed) { + (void) vmw_scrub_funcs[cb->bi.bt](&cb->bi, false); + cb->bi.scrubbed = true; + } + 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_state_scrub - Scrub all bindings associated with a + * struct vmw_ctx_binding state structure. + * + * @cbs: Pointer to the context binding state tracker. + * + * Emits commands to scrub all bindings associated with the + * context binding state tracker. + */ +static void vmw_context_binding_state_scrub(struct vmw_ctx_binding_state *cbs) +{ + struct vmw_ctx_binding *entry; + + list_for_each_entry(entry, &cbs->list, ctx_list) { + if (!entry->bi.scrubbed) { + (void) vmw_scrub_funcs[entry->bi.bt](&entry->bi, false); + entry->bi.scrubbed = true; + } + } +} + +/** + * 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_res_list_scrub - Scrub all bindings on a + * resource binding list + * + * @head: list head of resource binding list + * + * Scrub all bindings associated with a specific resource. Typically + * called before the resource is evicted. + */ +void vmw_context_binding_res_list_scrub(struct list_head *head) +{ + struct vmw_ctx_binding *entry; + + list_for_each_entry(entry, head, res_list) { + if (!entry->bi.scrubbed) { + (void) vmw_scrub_funcs[entry->bi.bt](&entry->bi, false); + entry->bi.scrubbed = true; + } + } +} + +/** + * 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); +} + +/** + * vmw_context_rebind_all - Rebind all scrubbed bindings of a context + * + * @ctx: The context resource + * + * Walks through the context binding list and rebinds all scrubbed + * resources. + */ +int vmw_context_rebind_all(struct vmw_resource *ctx) +{ + struct vmw_ctx_binding *entry; + struct vmw_user_context *uctx = + container_of(ctx, struct vmw_user_context, res); + struct vmw_ctx_binding_state *cbs = &uctx->cbs; + int ret; + + list_for_each_entry(entry, &cbs->list, ctx_list) { + if (likely(!entry->bi.scrubbed)) + continue; + + if (WARN_ON(entry->bi.res == NULL || entry->bi.res->id == + SVGA3D_INVALID_ID)) + continue; + + ret = vmw_scrub_funcs[entry->bi.bt](&entry->bi, true); + if (unlikely(ret != 0)) + return ret; + + entry->bi.scrubbed = false; + } + + return 0; +} + +/** + * vmw_context_binding_list - Return a list of context bindings + * + * @ctx: The context resource + * + * Returns the current list of bindings of the given context. Note that + * this list becomes stale as soon as the dev_priv::binding_mutex is unlocked. + */ +struct list_head *vmw_context_binding_list(struct vmw_resource *ctx) +{ + return &(container_of(ctx, struct vmw_user_context, res)->cbs.list); +} diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_dmabuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_dmabuf.c index d4e54fcc0ac..70ddce8358b 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_dmabuf.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_dmabuf.c @@ -52,11 +52,10 @@ int vmw_dmabuf_to_placement(struct vmw_private *dev_priv, struct ttm_placement *placement, bool interruptible) { - struct vmw_master *vmaster = dev_priv->active_master; struct ttm_buffer_object *bo = &buf->base; int ret; - ret = ttm_write_lock(&vmaster->lock, interruptible); + ret = ttm_write_lock(&dev_priv->reservation_sem, interruptible); if (unlikely(ret != 0)) return ret; @@ -71,7 +70,7 @@ int vmw_dmabuf_to_placement(struct vmw_private *dev_priv, ttm_bo_unreserve(bo); err: - ttm_write_unlock(&vmaster->lock); + ttm_write_unlock(&dev_priv->reservation_sem); return ret; } @@ -95,12 +94,11 @@ int vmw_dmabuf_to_vram_or_gmr(struct vmw_private *dev_priv, struct vmw_dma_buffer *buf, bool pin, bool interruptible) { - struct vmw_master *vmaster = dev_priv->active_master; struct ttm_buffer_object *bo = &buf->base; struct ttm_placement *placement; int ret; - ret = ttm_write_lock(&vmaster->lock, interruptible); + ret = ttm_write_lock(&dev_priv->reservation_sem, interruptible); if (unlikely(ret != 0)) return ret; @@ -143,7 +141,7 @@ int vmw_dmabuf_to_vram_or_gmr(struct vmw_private *dev_priv, err_unreserve: ttm_bo_unreserve(bo); err: - ttm_write_unlock(&vmaster->lock); + ttm_write_unlock(&dev_priv->reservation_sem); return ret; } @@ -198,7 +196,6 @@ int vmw_dmabuf_to_start_of_vram(struct vmw_private *dev_priv, struct vmw_dma_buffer *buf, bool pin, bool interruptible) { - struct vmw_master *vmaster = dev_priv->active_master; struct ttm_buffer_object *bo = &buf->base; struct ttm_placement placement; int ret = 0; @@ -209,7 +206,7 @@ int vmw_dmabuf_to_start_of_vram(struct vmw_private *dev_priv, placement = vmw_vram_placement; placement.lpfn = bo->num_pages; - ret = ttm_write_lock(&vmaster->lock, interruptible); + ret = ttm_write_lock(&dev_priv->reservation_sem, interruptible); if (unlikely(ret != 0)) return ret; @@ -232,7 +229,7 @@ int vmw_dmabuf_to_start_of_vram(struct vmw_private *dev_priv, ttm_bo_unreserve(bo); err_unlock: - ttm_write_unlock(&vmaster->lock); + ttm_write_unlock(&dev_priv->reservation_sem); return ret; } @@ -290,8 +287,7 @@ 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 +299,9 @@ 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; diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c index 1a90f0a2f7e..246a62bab37 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c @@ -32,6 +32,7 @@ #include <drm/ttm/ttm_bo_driver.h> #include <drm/ttm/ttm_object.h> #include <drm/ttm/ttm_module.h> +#include <linux/dma_remapping.h> #define VMWGFX_DRIVER_NAME "vmwgfx" #define VMWGFX_DRIVER_DESC "Linux drm driver for VMware graphics devices" @@ -111,6 +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 @@ -126,11 +142,11 @@ static const struct drm_ioctl_desc vmw_ioctls[] = { VMW_IOCTL_DEF(VMW_GET_PARAM, vmw_getparam_ioctl, - DRM_AUTH | DRM_UNLOCKED), + DRM_AUTH | DRM_UNLOCKED | DRM_RENDER_ALLOW), VMW_IOCTL_DEF(VMW_ALLOC_DMABUF, vmw_dmabuf_alloc_ioctl, - DRM_AUTH | DRM_UNLOCKED), + DRM_AUTH | DRM_UNLOCKED | DRM_RENDER_ALLOW), VMW_IOCTL_DEF(VMW_UNREF_DMABUF, vmw_dmabuf_unref_ioctl, - DRM_AUTH | DRM_UNLOCKED), + DRM_UNLOCKED | DRM_RENDER_ALLOW), VMW_IOCTL_DEF(VMW_CURSOR_BYPASS, vmw_kms_cursor_bypass_ioctl, DRM_MASTER | DRM_CONTROL_ALLOW | DRM_UNLOCKED), @@ -143,29 +159,28 @@ static const struct drm_ioctl_desc vmw_ioctls[] = { DRM_MASTER | DRM_CONTROL_ALLOW | DRM_UNLOCKED), VMW_IOCTL_DEF(VMW_CREATE_CONTEXT, vmw_context_define_ioctl, - DRM_AUTH | DRM_UNLOCKED), + DRM_AUTH | DRM_UNLOCKED | DRM_RENDER_ALLOW), VMW_IOCTL_DEF(VMW_UNREF_CONTEXT, vmw_context_destroy_ioctl, - DRM_AUTH | DRM_UNLOCKED), + DRM_UNLOCKED | DRM_RENDER_ALLOW), VMW_IOCTL_DEF(VMW_CREATE_SURFACE, vmw_surface_define_ioctl, - DRM_AUTH | DRM_UNLOCKED), + DRM_AUTH | DRM_UNLOCKED | DRM_RENDER_ALLOW), VMW_IOCTL_DEF(VMW_UNREF_SURFACE, vmw_surface_destroy_ioctl, - DRM_AUTH | DRM_UNLOCKED), + DRM_UNLOCKED | DRM_RENDER_ALLOW), VMW_IOCTL_DEF(VMW_REF_SURFACE, vmw_surface_reference_ioctl, - DRM_AUTH | DRM_UNLOCKED), + DRM_AUTH | DRM_UNLOCKED | DRM_RENDER_ALLOW), VMW_IOCTL_DEF(VMW_EXECBUF, vmw_execbuf_ioctl, - DRM_AUTH | DRM_UNLOCKED), + DRM_AUTH | DRM_UNLOCKED | DRM_RENDER_ALLOW), VMW_IOCTL_DEF(VMW_FENCE_WAIT, vmw_fence_obj_wait_ioctl, - DRM_AUTH | DRM_UNLOCKED), + DRM_UNLOCKED | DRM_RENDER_ALLOW), VMW_IOCTL_DEF(VMW_FENCE_SIGNALED, vmw_fence_obj_signaled_ioctl, - DRM_AUTH | DRM_UNLOCKED), + DRM_UNLOCKED | DRM_RENDER_ALLOW), VMW_IOCTL_DEF(VMW_FENCE_UNREF, vmw_fence_obj_unref_ioctl, - DRM_AUTH | DRM_UNLOCKED), - VMW_IOCTL_DEF(VMW_FENCE_EVENT, - vmw_fence_event_ioctl, - DRM_AUTH | DRM_UNLOCKED), + DRM_UNLOCKED | DRM_RENDER_ALLOW), + VMW_IOCTL_DEF(VMW_FENCE_EVENT, vmw_fence_event_ioctl, + DRM_AUTH | DRM_UNLOCKED | DRM_RENDER_ALLOW), VMW_IOCTL_DEF(VMW_GET_3D_CAP, vmw_get_cap_3d_ioctl, - DRM_AUTH | DRM_UNLOCKED), + DRM_AUTH | DRM_UNLOCKED | DRM_RENDER_ALLOW), /* these allow direct access to the framebuffers mark as master only */ VMW_IOCTL_DEF(VMW_PRESENT, vmw_present_ioctl, @@ -176,6 +191,21 @@ 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 | DRM_RENDER_ALLOW), + VMW_IOCTL_DEF(VMW_UNREF_SHADER, + vmw_shader_destroy_ioctl, + DRM_UNLOCKED | DRM_RENDER_ALLOW), + VMW_IOCTL_DEF(VMW_GB_SURFACE_CREATE, + vmw_gb_surface_define_ioctl, + DRM_AUTH | DRM_UNLOCKED | DRM_RENDER_ALLOW), + VMW_IOCTL_DEF(VMW_GB_SURFACE_REF, + vmw_gb_surface_reference_ioctl, + DRM_AUTH | DRM_UNLOCKED | DRM_RENDER_ALLOW), + VMW_IOCTL_DEF(VMW_SYNCCPU, + vmw_user_dmabuf_synccpu_ioctl, + DRM_UNLOCKED | DRM_RENDER_ALLOW), }; static struct pci_device_id vmw_pci_id_list[] = { @@ -185,6 +215,10 @@ static struct pci_device_id vmw_pci_id_list[] = { MODULE_DEVICE_TABLE(pci, vmw_pci_id_list); static int enable_fbdev = IS_ENABLED(CONFIG_DRM_VMWGFX_FBCON); +static int vmw_force_iommu; +static int vmw_restrict_iommu; +static int vmw_force_coherent; +static int vmw_restrict_dma_mask; static int vmw_probe(struct pci_dev *, const struct pci_device_id *); static void vmw_master_init(struct vmw_master *); @@ -193,6 +227,15 @@ static int vmwgfx_pm_notifier(struct notifier_block *nb, unsigned long val, MODULE_PARM_DESC(enable_fbdev, "Enable vmwgfx fbdev"); module_param_named(enable_fbdev, enable_fbdev, int, 0600); +MODULE_PARM_DESC(force_dma_api, "Force using the DMA API for TTM pages"); +module_param_named(force_dma_api, vmw_force_iommu, int, 0600); +MODULE_PARM_DESC(restrict_iommu, "Try to limit IOMMU usage for TTM pages"); +module_param_named(restrict_iommu, vmw_restrict_iommu, int, 0600); +MODULE_PARM_DESC(force_coherent, "Force coherent TTM pages"); +module_param_named(force_coherent, vmw_force_coherent, int, 0600); +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) { @@ -229,38 +272,52 @@ 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)) { @@ -269,34 +326,19 @@ 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; @@ -307,14 +349,24 @@ 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; @@ -330,10 +382,13 @@ 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 @@ -427,12 +482,112 @@ static void vmw_get_initial_size(struct vmw_private *dev_priv) dev_priv->initial_height = height; } +/** + * vmw_dma_select_mode - Determine how DMA mappings should be set up for this + * system. + * + * @dev_priv: Pointer to a struct vmw_private + * + * This functions tries to determine the IOMMU setup and what actions + * need to be taken by the driver to make system pages visible to the + * device. + * If this function decides that DMA is not possible, it returns -EINVAL. + * The driver may then try to disable features of the device that require + * DMA. + */ +static int vmw_dma_select_mode(struct vmw_private *dev_priv) +{ + static const char *names[vmw_dma_map_max] = { + [vmw_dma_phys] = "Using physical TTM page addresses.", + [vmw_dma_alloc_coherent] = "Using coherent TTM pages.", + [vmw_dma_map_populate] = "Keeping DMA mappings.", + [vmw_dma_map_bind] = "Giving up DMA mappings early."}; +#ifdef CONFIG_X86 + const struct dma_map_ops *dma_ops = get_dma_ops(dev_priv->dev->dev); + +#ifdef CONFIG_INTEL_IOMMU + if (intel_iommu_enabled) { + dev_priv->map_mode = vmw_dma_map_populate; + goto out_fixup; + } +#endif + + if (!(vmw_force_iommu || vmw_force_coherent)) { + dev_priv->map_mode = vmw_dma_phys; + DRM_INFO("DMA map mode: %s\n", names[dev_priv->map_mode]); + return 0; + } + + dev_priv->map_mode = vmw_dma_map_populate; + + if (dma_ops->sync_single_for_cpu) + dev_priv->map_mode = vmw_dma_alloc_coherent; +#ifdef CONFIG_SWIOTLB + if (swiotlb_nr_tbl() == 0) + dev_priv->map_mode = vmw_dma_map_populate; +#endif + +#ifdef CONFIG_INTEL_IOMMU +out_fixup: +#endif + if (dev_priv->map_mode == vmw_dma_map_populate && + vmw_restrict_iommu) + dev_priv->map_mode = vmw_dma_map_bind; + + if (vmw_force_coherent) + dev_priv->map_mode = vmw_dma_alloc_coherent; + +#if !defined(CONFIG_SWIOTLB) && !defined(CONFIG_INTEL_IOMMU) + /* + * No coherent page pool + */ + if (dev_priv->map_mode == vmw_dma_alloc_coherent) + return -EINVAL; +#endif + +#else /* CONFIG_X86 */ + dev_priv->map_mode = vmw_dma_map_populate; +#endif /* CONFIG_X86 */ + + DRM_INFO("DMA map mode: %s\n", names[dev_priv->map_mode]); + + 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; int ret; uint32_t svga_id; enum vmw_res_type i; + bool refuse_dma = false; dev_priv = kzalloc(sizeof(*dev_priv), GFP_KERNEL); if (unlikely(dev_priv == NULL)) { @@ -448,7 +603,9 @@ 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); + ttm_lock_init(&dev_priv->reservation_sem); for (i = vmw_res_context; i < vmw_res_max; ++i) { idr_init(&dev_priv->res_idr[i]); @@ -481,6 +638,11 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset) } dev_priv->capabilities = vmw_read(dev_priv, SVGA_REG_CAPABILITIES); + ret = vmw_dma_select_mode(dev_priv); + if (unlikely(ret != 0)) { + DRM_INFO("Restricting capabilities due to IOMMU setup.\n"); + refuse_dma = true; + } dev_priv->vram_size = vmw_read(dev_priv, SVGA_REG_VRAM_SIZE); dev_priv->mmio_size = vmw_read(dev_priv, SVGA_REG_MEM_SIZE); @@ -489,14 +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 = @@ -509,23 +666,45 @@ 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; + dev_priv->max_mob_size = 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); + dev_priv->max_mob_size = + vmw_read(dev_priv, SVGA_REG_MOB_MAX_SIZE); + } else + dev_priv->prim_bb_mem = dev_priv->vram_size; + + ret = vmw_dma_masks(dev_priv); + if (unlikely(ret != 0)) { + mutex_unlock(&dev_priv->hw_mutex); + 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", @@ -543,7 +722,9 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset) ret = ttm_bo_device_init(&dev_priv->bdev, dev_priv->bo_global_ref.ref.object, - &vmw_bo_driver, VMWGFX_FILE_PAGE_OFFSET, + &vmw_bo_driver, + dev->anon_inode->i_mapping, + VMWGFX_FILE_PAGE_OFFSET, false); if (unlikely(ret != 0)) { DRM_ERROR("Failed initializing TTM buffer object driver.\n"); @@ -558,13 +739,24 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset) } dev_priv->has_gmr = true; - if (ttm_bo_init_mm(&dev_priv->bdev, VMW_PL_GMR, - dev_priv->max_gmr_ids) != 0) { + if (((dev_priv->capabilities & (SVGA_CAP_GMR | SVGA_CAP_GMR2)) == 0) || + refuse_dma || ttm_bo_init_mm(&dev_priv->bdev, VMW_PL_GMR, + 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); @@ -587,7 +779,7 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset) } dev_priv->tdev = ttm_object_device_init - (dev_priv->mem_global_ref.object, 12); + (dev_priv->mem_global_ref.object, 12, &vmw_prime_dmabuf_ops); if (unlikely(dev_priv->tdev == NULL)) { DRM_ERROR("Unable to initialize TTM object management.\n"); @@ -614,7 +806,7 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset) } if (dev_priv->capabilities & SVGA_CAP_IRQMASK) { - ret = drm_irq_install(dev); + ret = drm_irq_install(dev, dev->pdev->irq); if (ret != 0) { DRM_ERROR("Failed installing irq: %d\n", ret); goto out_no_irq; @@ -667,6 +859,8 @@ 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); @@ -711,6 +905,8 @@ 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); @@ -740,9 +936,18 @@ static void vmw_postclose(struct drm_device *dev, struct vmw_fpriv *vmw_fp; vmw_fp = vmw_fpriv(file_priv); - ttm_object_file_release(&vmw_fp->tfile); - if (vmw_fp->locked_master) + + if (vmw_fp->locked_master) { + struct vmw_master *vmaster = + vmw_master(vmw_fp->locked_master); + + ttm_lock_set_kill(&vmaster->lock, true, SIGTERM); + ttm_vt_unlock(&vmaster->lock); drm_master_put(&vmw_fp->locked_master); + } + + vmw_compat_shader_man_destroy(vmw_fp->shman); + ttm_object_file_release(&vmw_fp->tfile); kfree(vmw_fp); } @@ -761,22 +966,85 @@ static int vmw_driver_open(struct drm_device *dev, struct drm_file *file_priv) if (unlikely(vmw_fp->tfile == NULL)) goto out_no_tfile; + vmw_fp->shman = vmw_compat_shader_man_create(dev_priv); + if (IS_ERR(vmw_fp->shman)) + goto out_no_shman; + file_priv->driver_priv = vmw_fp; - dev_priv->bdev.dev_mapping = dev->dev_mapping; return 0; +out_no_shman: + ttm_object_file_release(&vmw_fp->tfile); out_no_tfile: kfree(vmw_fp); return ret; } -static long vmw_unlocked_ioctl(struct file *filp, unsigned int cmd, - unsigned long arg) +static struct vmw_master *vmw_master_check(struct drm_device *dev, + struct drm_file *file_priv, + unsigned int flags) +{ + int ret; + struct vmw_fpriv *vmw_fp = vmw_fpriv(file_priv); + struct vmw_master *vmaster; + + if (file_priv->minor->type != DRM_MINOR_LEGACY || + !(flags & DRM_AUTH)) + return NULL; + + ret = mutex_lock_interruptible(&dev->master_mutex); + if (unlikely(ret != 0)) + return ERR_PTR(-ERESTARTSYS); + + if (file_priv->is_master) { + mutex_unlock(&dev->master_mutex); + return NULL; + } + + /* + * Check if we were previously master, but now dropped. + */ + if (vmw_fp->locked_master) { + mutex_unlock(&dev->master_mutex); + DRM_ERROR("Dropped master trying to access ioctl that " + "requires authentication.\n"); + return ERR_PTR(-EACCES); + } + mutex_unlock(&dev->master_mutex); + + /* + * Taking the drm_global_mutex after the TTM lock might deadlock + */ + if (!(flags & DRM_UNLOCKED)) { + DRM_ERROR("Refusing locked ioctl access.\n"); + return ERR_PTR(-EDEADLK); + } + + /* + * Take the TTM lock. Possibly sleep waiting for the authenticating + * master to become master again, or for a SIGTERM if the + * authenticating master exits. + */ + vmaster = vmw_master(file_priv->master); + ret = ttm_read_lock(&vmaster->lock, true); + if (unlikely(ret != 0)) + vmaster = ERR_PTR(ret); + + return vmaster; +} + +static long vmw_generic_ioctl(struct file *filp, unsigned int cmd, + unsigned long arg, + long (*ioctl_func)(struct file *, unsigned int, + unsigned long)) { struct drm_file *file_priv = filp->private_data; struct drm_device *dev = file_priv->minor->dev; unsigned int nr = DRM_IOCTL_NR(cmd); + struct vmw_master *vmaster; + unsigned int flags; + long ret; /* * Do extra checking on driver private ioctls. @@ -785,18 +1053,44 @@ static long vmw_unlocked_ioctl(struct file *filp, unsigned int cmd, if ((nr >= DRM_COMMAND_BASE) && (nr < DRM_COMMAND_END) && (nr < DRM_COMMAND_BASE + dev->driver->num_ioctls)) { const struct drm_ioctl_desc *ioctl = - &vmw_ioctls[nr - DRM_COMMAND_BASE]; + &vmw_ioctls[nr - DRM_COMMAND_BASE]; if (unlikely(ioctl->cmd_drv != cmd)) { DRM_ERROR("Invalid command format, ioctl %d\n", nr - DRM_COMMAND_BASE); return -EINVAL; } + flags = ioctl->flags; + } else if (!drm_ioctl_flags(nr, &flags)) + return -EINVAL; + + vmaster = vmw_master_check(dev, file_priv, flags); + if (unlikely(IS_ERR(vmaster))) { + DRM_INFO("IOCTL ERROR %d\n", nr); + return PTR_ERR(vmaster); } - return drm_ioctl(filp, cmd, arg); + ret = ioctl_func(filp, cmd, arg); + if (vmaster) + ttm_read_unlock(&vmaster->lock); + + return ret; +} + +static long vmw_unlocked_ioctl(struct file *filp, unsigned int cmd, + unsigned long arg) +{ + return vmw_generic_ioctl(filp, cmd, arg, &drm_ioctl); } +#ifdef CONFIG_COMPAT +static long vmw_compat_ioctl(struct file *filp, unsigned int cmd, + unsigned long arg) +{ + return vmw_generic_ioctl(filp, cmd, arg, &drm_compat_ioctl); +} +#endif + static void vmw_lastclose(struct drm_device *dev) { struct drm_crtc *crtc; @@ -925,14 +1219,13 @@ static void vmw_master_drop(struct drm_device *dev, vmw_fp->locked_master = drm_master_get(file_priv->master); ret = ttm_vt_lock(&vmaster->lock, false, vmw_fp->tfile); - vmw_execbuf_release_pinned_bo(dev_priv); - if (unlikely((ret != 0))) { DRM_ERROR("Unable to lock TTM at VT switch.\n"); drm_master_put(&vmw_fp->locked_master); } - ttm_lock_set_kill(&vmaster->lock, true, SIGTERM); + ttm_lock_set_kill(&vmaster->lock, false, SIGTERM); + vmw_execbuf_release_pinned_bo(dev_priv); if (!dev_priv->enable_fb) { ret = ttm_bo_evict_mm(&dev_priv->bdev, TTM_PL_VRAM); @@ -966,12 +1259,11 @@ static int vmwgfx_pm_notifier(struct notifier_block *nb, unsigned long val, { struct vmw_private *dev_priv = container_of(nb, struct vmw_private, pm_nb); - struct vmw_master *vmaster = dev_priv->active_master; switch (val) { case PM_HIBERNATION_PREPARE: case PM_SUSPEND_PREPARE: - ttm_suspend_lock(&vmaster->lock); + ttm_suspend_lock(&dev_priv->reservation_sem); /** * This empties VRAM and unbinds all GMR bindings. @@ -985,7 +1277,7 @@ static int vmwgfx_pm_notifier(struct notifier_block *nb, unsigned long val, case PM_POST_HIBERNATION: case PM_POST_SUSPEND: case PM_POST_RESTORE: - ttm_suspend_unlock(&vmaster->lock); + ttm_suspend_unlock(&dev_priv->reservation_sem); break; case PM_RESTORE_PREPARE: @@ -1106,14 +1398,14 @@ static const struct file_operations vmwgfx_driver_fops = { .poll = vmw_fops_poll, .read = vmw_fops_read, #if defined(CONFIG_COMPAT) - .compat_ioctl = drm_compat_ioctl, + .compat_ioctl = vmw_compat_ioctl, #endif .llseek = noop_llseek, }; static struct drm_driver driver = { .driver_features = DRIVER_HAVE_IRQ | DRIVER_IRQ_SHARED | - DRIVER_MODESET, + DRIVER_MODESET | DRIVER_PRIME | DRIVER_RENDER, .load = vmw_driver_load, .unload = vmw_driver_unload, .lastclose = vmw_lastclose, @@ -1125,7 +1417,7 @@ static struct drm_driver driver = { .enable_vblank = vmw_enable_vblank, .disable_vblank = vmw_disable_vblank, .ioctls = vmw_ioctls, - .num_ioctls = DRM_ARRAY_SIZE(vmw_ioctls), + .num_ioctls = ARRAY_SIZE(vmw_ioctls), .master_create = vmw_master_create, .master_destroy = vmw_master_destroy, .master_set = vmw_master_set, @@ -1138,6 +1430,9 @@ static struct drm_driver driver = { .dumb_map_offset = vmw_dumb_map_offset, .dumb_destroy = vmw_dumb_destroy, + .prime_fd_to_handle = vmw_prime_fd_to_handle, + .prime_handle_to_fd = vmw_prime_handle_to_fd, + .fops = &vmwgfx_driver_fops, .name = VMWGFX_DRIVER_NAME, .desc = VMWGFX_DRIVER_DESC, diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h index 150ec64af61..6b252a887ae 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h +++ b/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 "20140325" #define VMWGFX_DRIVER_MAJOR 2 -#define VMWGFX_DRIVER_MINOR 4 +#define VMWGFX_DRIVER_MINOR 6 #define VMWGFX_DRIVER_PATCHLEVEL 0 #define VMWGFX_FILE_PAGE_OFFSET 0x00100000 #define VMWGFX_FIFO_STATIC_SIZE (1024*1024) @@ -50,19 +50,39 @@ #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_compat_shader_manager; struct vmw_fpriv { struct drm_master *locked_master; struct ttm_object_file *tfile; struct list_head fence_events; + bool gb_aware; + struct vmw_compat_shader_manager *shman; }; struct vmw_dma_buffer { @@ -82,6 +102,7 @@ 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 +119,7 @@ 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 +128,7 @@ enum vmw_res_type { vmw_res_context, vmw_res_surface, vmw_res_stream, + vmw_res_shader, vmw_res_max }; @@ -154,6 +177,7 @@ struct vmw_fifo_state { }; struct vmw_relocation { + SVGAMobId *mob_loc; SVGAGuestPtr *location; uint32_t index; }; @@ -177,11 +201,129 @@ struct vmw_res_cache_entry { struct vmw_resource_val_node *node; }; +/** + * enum vmw_dma_map_mode - indicate how to perform TTM page dma mappings. + */ +enum vmw_dma_map_mode { + vmw_dma_phys, /* Use physical page addresses */ + vmw_dma_alloc_coherent, /* Use TTM coherent pages */ + vmw_dma_map_populate, /* Unmap from DMA just after unpopulate */ + vmw_dma_map_bind, /* Unmap from DMA just before unbind */ + vmw_dma_map_max +}; + +/** + * struct vmw_sg_table - Scatter/gather table for binding, with additional + * device-specific information. + * + * @sgt: Pointer to a struct sg_table with binding information + * @num_regions: Number of regions with device-address contigous pages + */ +struct vmw_sg_table { + enum vmw_dma_map_mode mode; + struct page **pages; + const dma_addr_t *addrs; + struct sg_table *sgt; + unsigned long num_regions; + unsigned long num_pages; +}; + +/** + * struct vmw_piter - Page iterator that iterates over a list of pages + * and DMA addresses that could be either a scatter-gather list or + * arrays + * + * @pages: Array of page pointers to the pages. + * @addrs: DMA addresses to the pages if coherent pages are used. + * @iter: Scatter-gather page iterator. Current position in SG list. + * @i: Current position in arrays. + * @num_pages: Number of pages total. + * @next: Function to advance the iterator. Returns false if past the list + * of pages, true otherwise. + * @dma_address: Function to return the DMA address of the current page. + */ +struct vmw_piter { + struct page **pages; + const dma_addr_t *addrs; + struct sg_page_iter iter; + unsigned long i; + unsigned long num_pages; + bool (*next)(struct vmw_piter *); + dma_addr_t (*dma_address)(struct vmw_piter *); + struct page *(*page)(struct vmw_piter *); +}; + +/* + * 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; + bool scrubbed; + 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; bool kernel; /**< is the called made from the kernel */ - struct ttm_object_file *tfile; + struct vmw_fpriv *fp; struct list_head validate_nodes; struct vmw_relocation relocs[VMWGFX_MAX_RELOCATIONS]; uint32_t cur_reloc; @@ -198,6 +340,8 @@ 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 list_head staged_shaders; }; struct vmw_legacy_display; @@ -229,6 +373,7 @@ 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; @@ -238,11 +383,13 @@ 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 max_mob_size; uint32_t memory_size; bool has_gmr; + bool has_mob; struct mutex hw_mutex; /* @@ -318,6 +465,7 @@ struct vmw_private { struct vmw_sw_context ctx; struct mutex cmdbuf_mutex; + struct mutex binding_mutex; /** * Operating mode. @@ -339,6 +487,11 @@ struct vmw_private { uint32_t num_3d_resources; /* + * Replace this with an rwsem as soon as we have down_xx_interruptible() + */ + struct ttm_lock reservation_sem; + + /* * Query processing. These members * are protected by the cmdbuf mutex. */ @@ -358,6 +511,17 @@ struct vmw_private { struct list_head res_lru[vmw_res_max]; uint32_t used_memory_size; + + /* + * DMA mapping stuff. + */ + enum vmw_dma_map_mode map_mode; + + /* + * 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) @@ -405,7 +569,7 @@ void vmw_3d_resource_dec(struct vmw_private *dev_priv, bool hide_svga); */ extern int vmw_gmr_bind(struct vmw_private *dev_priv, - struct page *pages[], + const struct vmw_sg_table *vsgt, unsigned long num_pages, int gmr_id); extern void vmw_gmr_unbind(struct vmw_private *dev_priv, int gmr_id); @@ -414,23 +578,14 @@ 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 struct vmw_resource * +vmw_resource_reference_unless_doomed(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, @@ -442,18 +597,6 @@ 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, @@ -462,10 +605,21 @@ 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); @@ -558,16 +712,62 @@ extern int vmw_mmap(struct file *filp, struct vm_area_struct *vma); * TTM buffer object driver - vmwgfx_buffer.c */ +extern const size_t vmw_tt_size; extern struct ttm_placement vmw_vram_placement; extern struct ttm_placement vmw_vram_ne_placement; extern struct ttm_placement vmw_vram_sys_placement; extern struct ttm_placement vmw_vram_gmr_placement; extern struct ttm_placement vmw_vram_gmr_ne_placement; extern struct ttm_placement vmw_sys_placement; +extern struct ttm_placement vmw_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); + +/** + * vmw_piter_next - Advance the iterator one page. + * + * @viter: Pointer to the iterator to advance. + * + * Returns false if past the list of pages, true otherwise. + */ +static inline bool vmw_piter_next(struct vmw_piter *viter) +{ + return viter->next(viter); +} + +/** + * vmw_piter_dma_addr - Return the DMA address of the current page. + * + * @viter: Pointer to the iterator + * + * Returns the DMA address of the page pointed to by @viter. + */ +static inline dma_addr_t vmw_piter_dma_addr(struct vmw_piter *viter) +{ + return viter->dma_address(viter); +} + +/** + * vmw_piter_page - Return a pointer to the current page. + * + * @viter: Pointer to the iterator + * + * Returns the DMA address of the page pointed to by @viter. + */ +static inline struct page *vmw_piter_page(struct vmw_piter *viter) +{ + return viter->page(viter); +} /** * Command submission - vmwgfx_execbuf.c @@ -604,7 +804,7 @@ extern void vmw_execbuf_copy_fence_user(struct vmw_private *dev_priv, * IRQs and wating - vmwgfx_irq.c */ -extern irqreturn_t vmw_irq_handler(DRM_IRQ_ARGS); +extern irqreturn_t vmw_irq_handler(int irq, void *arg); extern int vmw_wait_seqno(struct vmw_private *dev_priv, bool lazy, uint32_t seqno, bool interruptible, unsigned long timeout); @@ -723,6 +923,115 @@ int vmw_overlay_num_free_overlays(struct vmw_private *dev_priv); extern const struct ttm_mem_type_manager_func vmw_gmrid_manager_func; /** + * Prime - vmwgfx_prime.c + */ + +extern const struct dma_buf_ops vmw_prime_dmabuf_ops; +extern int vmw_prime_fd_to_handle(struct drm_device *dev, + struct drm_file *file_priv, + int fd, u32 *handle); +extern int vmw_prime_handle_to_fd(struct drm_device *dev, + struct drm_file *file_priv, + 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); +extern void vmw_context_binding_res_list_scrub(struct list_head *head); +extern int vmw_context_rebind_all(struct vmw_resource *ctx); +extern struct list_head *vmw_context_binding_list(struct vmw_resource *ctx); + +/* + * 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); +extern int vmw_compat_shader_lookup(struct vmw_compat_shader_manager *man, + SVGA3dShaderType shader_type, + u32 *user_key); +extern void vmw_compat_shaders_commit(struct vmw_compat_shader_manager *man, + struct list_head *list); +extern void vmw_compat_shaders_revert(struct vmw_compat_shader_manager *man, + struct list_head *list); +extern int vmw_compat_shader_remove(struct vmw_compat_shader_manager *man, + u32 user_key, + SVGA3dShaderType shader_type, + struct list_head *list); +extern int vmw_compat_shader_add(struct vmw_compat_shader_manager *man, + u32 user_key, const void *bytecode, + SVGA3dShaderType shader_type, + size_t size, + struct ttm_object_file *tfile, + struct list_head *list); +extern struct vmw_compat_shader_manager * +vmw_compat_shader_man_create(struct vmw_private *dev_priv); +extern void +vmw_compat_shader_man_destroy(struct vmw_compat_shader_manager *man); + + +/** * Inline helper functions */ diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c index 599f6469a1e..87df0b3674f 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c +++ b/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. @@ -65,12 +67,32 @@ 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 +109,18 @@ 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)) { + if (!backoff) { + 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); @@ -146,6 +180,44 @@ static int vmw_resource_val_add(struct vmw_sw_context *sw_context, } /** + * vmw_resource_context_res_add - Put resources previously bound to a context on + * the validation list + * + * @dev_priv: Pointer to a device private structure + * @sw_context: Pointer to a software context used for this command submission + * @ctx: Pointer to the context resource + * + * This function puts all resources that were previously bound to @ctx on + * the resource validation list. This is part of the context state reemission + */ +static int vmw_resource_context_res_add(struct vmw_private *dev_priv, + struct vmw_sw_context *sw_context, + struct vmw_resource *ctx) +{ + struct list_head *binding_list; + struct vmw_ctx_binding *entry; + int ret = 0; + struct vmw_resource *res; + + mutex_lock(&dev_priv->binding_mutex); + binding_list = vmw_context_binding_list(ctx); + + list_for_each_entry(entry, binding_list, ctx_list) { + res = vmw_resource_reference_unless_doomed(entry->bi.res); + if (unlikely(res == NULL)) + continue; + + ret = vmw_resource_val_add(sw_context, entry->bi.res, NULL); + vmw_resource_unreference(&res); + if (unlikely(ret != 0)) + break; + } + + mutex_unlock(&dev_priv->binding_mutex); + return ret; +} + +/** * vmw_resource_relocation_add - Add a relocation to the relocation list * * @list: Pointer to head of relocation list. @@ -201,8 +273,12 @@ static void vmw_resource_relocations_apply(uint32_t *cb, { struct vmw_resource_relocation *rel; - list_for_each_entry(rel, list, head) - cb[rel->offset] = rel->res->id; + list_for_each_entry(rel, list, head) { + if (likely(rel->res != NULL)) + cb[rel->offset] = rel->res->id; + else + cb[rel->offset] = SVGA_3D_CMD_NOP; + } } static int vmw_cmd_invalid(struct vmw_private *dev_priv, @@ -224,6 +300,7 @@ 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 +309,7 @@ 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 +322,10 @@ 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 +348,7 @@ 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 +385,8 @@ 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; @@ -339,22 +423,27 @@ static int vmw_resources_validate(struct vmw_sw_context *sw_context) } /** - * vmw_cmd_res_check - Check that a resource is present and if so, put it + * vmw_cmd_compat_res_check - Check that a resource is present and if so, put it * on the resource validate list unless it's already there. * * @dev_priv: Pointer to a device private structure. * @sw_context: Pointer to the software context. * @res_type: Resource type. * @converter: User-space visisble type specific information. - * @id: Pointer to the location in the command buffer currently being + * @id: user-space resource id handle. + * @id_loc: Pointer to the location in the command buffer currently being * parsed from where the user-space resource id handle is located. + * @p_val: Pointer to pointer to resource validalidation node. Populated + * on exit. */ -static int vmw_cmd_res_check(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 *id, - struct vmw_resource_val_node **p_val) +static int +vmw_cmd_compat_res_check(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 id, + uint32_t *id_loc, + struct vmw_resource_val_node **p_val) { struct vmw_res_cache_entry *rcache = &sw_context->res_cache[res_type]; @@ -362,15 +451,22 @@ 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 * resource */ - if (likely(rcache->valid && *id == rcache->handle)) { + if (likely(rcache->valid && id == rcache->handle)) { const struct vmw_resource *res = rcache->res; rcache->node->first_usage = false; @@ -379,28 +475,28 @@ static int vmw_cmd_res_check(struct vmw_private *dev_priv, return vmw_resource_relocation_add (&sw_context->res_relocations, res, - id - sw_context->buf_start); + id_loc - sw_context->buf_start); } ret = vmw_user_resource_lookup_handle(dev_priv, - sw_context->tfile, - *id, + sw_context->fp->tfile, + id, converter, &res); if (unlikely(ret != 0)) { DRM_ERROR("Could not find or use resource 0x%08x.\n", - (unsigned) *id); + (unsigned) id); dump_stack(); return ret; } rcache->valid = true; rcache->res = res; - rcache->handle = *id; + rcache->handle = id; ret = vmw_resource_relocation_add(&sw_context->res_relocations, res, - id - sw_context->buf_start); + id_loc - sw_context->buf_start); if (unlikely(ret != 0)) goto out_no_reloc; @@ -411,6 +507,22 @@ static int vmw_cmd_res_check(struct vmw_private *dev_priv, rcache->node = node; if (p_val) *p_val = node; + + if (dev_priv->has_mob && node->first_usage && + res_type == vmw_res_context) { + ret = vmw_resource_context_res_add(dev_priv, sw_context, res); + if (unlikely(ret != 0)) + goto out_no_reloc; + 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; @@ -422,6 +534,59 @@ out_no_reloc: } /** + * vmw_cmd_res_check - Check that a resource is present and if so, put it + * on the resource validate list unless it's already there. + * + * @dev_priv: Pointer to a device private structure. + * @sw_context: Pointer to the software context. + * @res_type: Resource type. + * @converter: User-space visisble type specific information. + * @id_loc: Pointer to the location in the command buffer currently being + * parsed from where the user-space resource id handle is located. + * @p_val: Pointer to pointer to resource validalidation node. Populated + * on exit. + */ +static int +vmw_cmd_res_check(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 *id_loc, + struct vmw_resource_val_node **p_val) +{ + return vmw_cmd_compat_res_check(dev_priv, sw_context, res_type, + converter, *id_loc, id_loc, p_val); +} + +/** + * vmw_rebind_contexts - Rebind all resources previously bound to + * referenced contexts. + * + * @sw_context: Pointer to the software context. + * + * Rebind context binding points that have been scrubbed because of eviction. + */ +static int vmw_rebind_contexts(struct vmw_sw_context *sw_context) +{ + struct vmw_resource_val_node *val; + int ret; + + list_for_each_entry(val, &sw_context->resource_list, head) { + if (likely(!val->staged_bindings)) + continue; + + ret = vmw_context_rebind_all(val->res); + if (unlikely(ret != 0)) { + if (ret != -ERESTARTSYS) + DRM_ERROR("Failed to rebind context.\n"); + return ret; + } + } + + return 0; +} + +/** * vmw_cmd_cid_check - Check a command header for valid context information. * * @dev_priv: Pointer to a device private structure. @@ -437,7 +602,7 @@ static int vmw_cmd_cid_check(struct vmw_private *dev_priv, { struct vmw_cid_cmd { SVGA3dCmdHeader header; - __le32 cid; + uint32_t cid; } *cmd; cmd = container_of(header, struct vmw_cid_cmd, header); @@ -453,17 +618,35 @@ 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 +702,6 @@ 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 +719,6 @@ 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 +759,7 @@ 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 +767,7 @@ 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; @@ -672,6 +845,66 @@ 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->fp->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 * @@ -701,7 +934,7 @@ static int vmw_translate_guest_ptr(struct vmw_private *dev_priv, struct vmw_relocation *reloc; int ret; - ret = vmw_user_dmabuf_lookup(sw_context->tfile, handle, &vmw_bo); + ret = vmw_user_dmabuf_lookup(sw_context->fp->tfile, handle, &vmw_bo); if (unlikely(ret != 0)) { DRM_ERROR("Could not find or use GMR region.\n"); return -EINVAL; @@ -718,7 +951,7 @@ 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; @@ -732,6 +965,30 @@ out_no_reloc: } /** + * 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. * * @dev_priv: Pointer to a device private struct. @@ -750,12 +1007,64 @@ 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. * * @dev_priv: Pointer to a device private struct. @@ -774,6 +1083,25 @@ 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 +1118,40 @@ 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 +1170,25 @@ 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; @@ -834,14 +1214,36 @@ static int vmw_cmd_dma(struct vmw_private *dev_priv, SVGA3dCmdSurfaceDMA dma; } *cmd; int ret; + SVGA3dCmdSurfaceDMASuffix *suffix; + uint32_t bo_size; cmd = container_of(header, struct vmw_dma_cmd, header); + suffix = (SVGA3dCmdSurfaceDMASuffix *)((unsigned long) &cmd->dma + + header->size - sizeof(*suffix)); + + /* Make sure device and verifier stays in sync. */ + if (unlikely(suffix->suffixSize != sizeof(*suffix))) { + DRM_ERROR("Invalid DMA suffix size.\n"); + return -EINVAL; + } + ret = vmw_translate_guest_ptr(dev_priv, sw_context, &cmd->dma.guest.ptr, &vmw_bo); if (unlikely(ret != 0)) return ret; + /* Make sure DMA doesn't cross BO boundaries. */ + bo_size = vmw_bo->base.num_pages * PAGE_SIZE; + if (unlikely(cmd->dma.guest.ptr.offset > bo_size)) { + DRM_ERROR("Invalid DMA offset.\n"); + return -EINVAL; + } + + bo_size -= cmd->dma.guest.ptr.offset; + if (unlikely(suffix->maximumOffset > bo_size)) + suffix->maximumOffset = bo_size; + ret = vmw_cmd_res_check(dev_priv, sw_context, vmw_res_surface, user_surface_converter, &cmd->dma.host.sid, NULL); @@ -853,7 +1255,8 @@ static int vmw_cmd_dma(struct vmw_private *dev_priv, srf = vmw_res_to_srf(sw_context->res_cache[vmw_res_surface].res); - vmw_kms_cursor_snoop(srf, sw_context->tfile, &vmw_bo->base, header); + vmw_kms_cursor_snoop(srf, sw_context->fp->tfile, &vmw_bo->base, + header); out_no_surface: vmw_dmabuf_unreference(&vmw_bo); @@ -921,15 +1324,22 @@ 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 +1349,20 @@ 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; @@ -971,6 +1392,314 @@ static int vmw_cmd_check_define_gmrfb(struct vmw_private *dev_priv, } /** + * 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_shader_define - Validate an SVGA_3D_CMD_SHADER_DEFINE + * 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_shader_define(struct vmw_private *dev_priv, + struct vmw_sw_context *sw_context, + SVGA3dCmdHeader *header) +{ + struct vmw_shader_define_cmd { + SVGA3dCmdHeader header; + SVGA3dCmdDefineShader body; + } *cmd; + int ret; + size_t size; + + cmd = container_of(header, struct vmw_shader_define_cmd, + header); + + ret = vmw_cmd_res_check(dev_priv, sw_context, vmw_res_context, + user_context_converter, &cmd->body.cid, + NULL); + if (unlikely(ret != 0)) + return ret; + + if (unlikely(!dev_priv->has_mob)) + return 0; + + size = cmd->header.size - sizeof(cmd->body); + ret = vmw_compat_shader_add(sw_context->fp->shman, + cmd->body.shid, cmd + 1, + cmd->body.type, size, + sw_context->fp->tfile, + &sw_context->staged_shaders); + if (unlikely(ret != 0)) + return ret; + + return vmw_resource_relocation_add(&sw_context->res_relocations, + NULL, &cmd->header.id - + sw_context->buf_start); + + return 0; +} + +/** + * vmw_cmd_shader_destroy - Validate an SVGA_3D_CMD_SHADER_DESTROY + * 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_shader_destroy(struct vmw_private *dev_priv, + struct vmw_sw_context *sw_context, + SVGA3dCmdHeader *header) +{ + struct vmw_shader_destroy_cmd { + SVGA3dCmdHeader header; + SVGA3dCmdDestroyShader body; + } *cmd; + int ret; + + cmd = container_of(header, struct vmw_shader_destroy_cmd, + header); + + ret = vmw_cmd_res_check(dev_priv, sw_context, vmw_res_context, + user_context_converter, &cmd->body.cid, + NULL); + if (unlikely(ret != 0)) + return ret; + + if (unlikely(!dev_priv->has_mob)) + return 0; + + ret = vmw_compat_shader_remove(sw_context->fp->shman, + cmd->body.shid, + cmd->body.type, + &sw_context->staged_shaders); + if (unlikely(ret != 0)) + return ret; + + return vmw_resource_relocation_add(&sw_context->res_relocations, + NULL, &cmd->header.id - + sw_context->buf_start); + + return 0; +} + +/** * vmw_cmd_set_shader - Validate an SVGA_3D_CMD_SET_SHADER * command * @@ -986,18 +1715,105 @@ 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; + u32 shid = cmd->body.shid; + + if (shid != SVGA3D_INVALID_ID) + (void) vmw_compat_shader_lookup(sw_context->fp->shman, + cmd->body.type, + &shid); + + ret = vmw_cmd_compat_res_check(dev_priv, sw_context, + vmw_res_shader, + user_shader_converter, + shid, + &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_set_shader_const - Validate an SVGA_3D_CMD_SET_SHADER_CONST + * 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_set_shader_const(struct vmw_private *dev_priv, + struct vmw_sw_context *sw_context, + SVGA3dCmdHeader *header) +{ + struct vmw_set_shader_const_cmd { + SVGA3dCmdHeader header; + SVGA3dCmdSetShaderConst body; + } *cmd; + int ret; + + cmd = container_of(header, struct vmw_set_shader_const_cmd, + header); + + ret = vmw_cmd_res_check(dev_priv, sw_context, vmw_res_context, + user_context_converter, &cmd->body.cid, + NULL); + if (unlikely(ret != 0)) + return ret; + + if (dev_priv->has_mob) + header->id = SVGA_3D_CMD_SET_GB_SHADERCONSTS_INLINE; + + 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 +1857,173 @@ 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 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_shader_define, + true, false, false), + VMW_CMD_DEF(SVGA_3D_CMD_SHADER_DESTROY, &vmw_cmd_shader_destroy, + true, false, 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_set_shader_const, + true, false, 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 +2034,8 @@ 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 +2048,43 @@ 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->func)) + goto out_invalid; + + 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 +2140,9 @@ 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 +2167,8 @@ 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 +2195,8 @@ 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 +2210,9 @@ 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 +2234,6 @@ 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 +2241,8 @@ 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; } @@ -1461,7 +2436,7 @@ int vmw_execbuf_process(struct drm_file *file_priv, } else sw_context->kernel = true; - sw_context->tfile = vmw_fpriv(file_priv)->tfile; + sw_context->fp = vmw_fpriv(file_priv); sw_context->cur_reloc = 0; sw_context->cur_val_buf = 0; sw_context->fence_flags = 0; @@ -1478,16 +2453,17 @@ int vmw_execbuf_process(struct drm_file *file_priv, goto out_unlock; sw_context->res_ht_initialized = true; } + INIT_LIST_HEAD(&sw_context->staged_shaders); INIT_LIST_HEAD(&resource_list); ret = vmw_cmd_check_all(dev_priv, sw_context, kernel_commands, command_size); if (unlikely(ret != 0)) - goto out_err; + goto out_err_nores; ret = vmw_resources_reserve(sw_context); if (unlikely(ret != 0)) - goto out_err; + goto out_err_nores; ret = ttm_eu_reserve_buffers(&ticket, &sw_context->validate_nodes); if (unlikely(ret != 0)) @@ -1509,11 +2485,23 @@ 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; + } + + if (dev_priv->has_mob) { + ret = vmw_rebind_contexts(sw_context); + if (unlikely(ret != 0)) + goto out_unlock_binding; + } + 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 +2526,8 @@ 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); @@ -1558,6 +2548,8 @@ int vmw_execbuf_process(struct drm_file *file_priv, } list_splice_init(&sw_context->resource_list, &resource_list); + vmw_compat_shaders_commit(sw_context->fp->shman, + &sw_context->staged_shaders); mutex_unlock(&dev_priv->cmdbuf_mutex); /* @@ -1568,11 +2560,14 @@ 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); ttm_eu_backoff_reservation(&ticket, &sw_context->validate_nodes); +out_err_nores: vmw_resource_list_unreserve(&sw_context->resource_list, true); + vmw_resource_relocations_free(&sw_context->res_relocations); + vmw_free_relocations(sw_context); vmw_clear_validations(sw_context); if (unlikely(dev_priv->pinned_bo != NULL && !dev_priv->query_cid_valid)) @@ -1581,6 +2576,8 @@ out_unlock: list_splice_init(&sw_context->resource_list, &resource_list); error_resource = sw_context->error_resource; sw_context->error_resource = NULL; + vmw_compat_shaders_revert(sw_context->fp->shman, + &sw_context->staged_shaders); mutex_unlock(&dev_priv->cmdbuf_mutex); /* @@ -1737,7 +2734,6 @@ int vmw_execbuf_ioctl(struct drm_device *dev, void *data, { struct vmw_private *dev_priv = vmw_priv(dev); struct drm_vmw_execbuf_arg *arg = (struct drm_vmw_execbuf_arg *)data; - struct vmw_master *vmaster = vmw_master(file_priv->master); int ret; /* @@ -1754,7 +2750,7 @@ int vmw_execbuf_ioctl(struct drm_device *dev, void *data, return -EINVAL; } - ret = ttm_read_lock(&vmaster->lock, true); + ret = ttm_read_lock(&dev_priv->reservation_sem, true); if (unlikely(ret != 0)) return ret; @@ -1770,6 +2766,6 @@ int vmw_execbuf_ioctl(struct drm_device *dev, void *data, vmw_kms_cursor_post_execbuf(dev_priv); out_unlock: - ttm_read_unlock(&vmaster->lock); + ttm_read_unlock(&dev_priv->reservation_sem); return ret; } diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c index ed5ce2a41bb..b031b48dbb3 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c @@ -147,7 +147,7 @@ static int vmw_fb_check_var(struct fb_var_screeninfo *var, } if (!vmw_kms_validate_mode_vram(vmw_priv, - info->fix.line_length, + var->xres * var->bits_per_pixel/8, var->yoffset + var->yres)) { DRM_ERROR("Requested geom can not fit in framebuffer\n"); return -EINVAL; @@ -162,6 +162,8 @@ static int vmw_fb_set_par(struct fb_info *info) struct vmw_private *vmw_priv = par->vmw_priv; int ret; + info->fix.line_length = info->var.xres * info->var.bits_per_pixel/8; + ret = vmw_kms_write_svga(vmw_priv, info->var.xres, info->var.yres, info->fix.line_length, par->bpp, par->depth); @@ -377,14 +379,13 @@ static int vmw_fb_create_bo(struct vmw_private *vmw_priv, ne_placement.lpfn = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; - /* interuptable? */ - ret = ttm_write_lock(&vmw_priv->fbdev_master.lock, false); - if (unlikely(ret != 0)) - return ret; + (void) ttm_write_lock(&vmw_priv->reservation_sem, false); vmw_bo = kmalloc(sizeof(*vmw_bo), GFP_KERNEL); - if (!vmw_bo) + if (!vmw_bo) { + ret = -ENOMEM; goto err_unlock; + } ret = vmw_dmabuf_init(vmw_priv, vmw_bo, size, &ne_placement, diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c index c62d20e8a6f..436b013b423 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c @@ -271,7 +271,7 @@ void vmw_fence_obj_unreference(struct vmw_fence_obj **fence_p) spin_unlock_irq(&fman->lock); } -void vmw_fences_perform_actions(struct vmw_fence_manager *fman, +static void vmw_fences_perform_actions(struct vmw_fence_manager *fman, struct list_head *list) { struct vmw_fence_action *action, *next_action; @@ -897,7 +897,7 @@ static void vmw_event_fence_action_cleanup(struct vmw_fence_action *action) * Note that the action callbacks may be executed before this function * returns. */ -void vmw_fence_obj_add_action(struct vmw_fence_obj *fence, +static void vmw_fence_obj_add_action(struct vmw_fence_obj *fence, struct vmw_fence_action *action) { struct vmw_fence_manager *fman = fence->fman; @@ -993,7 +993,7 @@ struct vmw_event_fence_pending { struct drm_vmw_event_fence event; }; -int vmw_event_fence_action_create(struct drm_file *file_priv, +static int vmw_event_fence_action_create(struct drm_file *file_priv, struct vmw_fence_obj *fence, uint32_t flags, uint64_t user_data, @@ -1080,7 +1080,8 @@ int vmw_fence_event_ioctl(struct drm_device *dev, void *data, */ if (arg->handle) { struct ttm_base_object *base = - ttm_base_object_lookup(vmw_fp->tfile, arg->handle); + ttm_base_object_lookup_for_ref(dev_priv->tdev, + arg->handle); if (unlikely(base == NULL)) { DRM_ERROR("Fence event invalid fence object handle " diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c index 3eb148667d6..6ccd993e26b 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c +++ b/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); +} diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_gmr.c b/drivers/gpu/drm/vmwgfx/vmwgfx_gmr.c index 1a0bf07fe54..61d8d803199 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_gmr.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_gmr.c @@ -32,9 +32,11 @@ #define VMW_PPN_SIZE (sizeof(unsigned long)) /* A future safe maximum remap size. */ #define VMW_PPN_PER_REMAP ((31 * 1024) / VMW_PPN_SIZE) +#define DMA_ADDR_INVALID ((dma_addr_t) 0) +#define DMA_PAGE_INVALID 0UL static int vmw_gmr2_bind(struct vmw_private *dev_priv, - struct page *pages[], + struct vmw_piter *iter, unsigned long num_pages, int gmr_id) { @@ -81,11 +83,13 @@ static int vmw_gmr2_bind(struct vmw_private *dev_priv, for (i = 0; i < nr; ++i) { if (VMW_PPN_SIZE <= 4) - *cmd = page_to_pfn(*pages++); + *cmd = vmw_piter_dma_addr(iter) >> PAGE_SHIFT; else - *((uint64_t *)cmd) = page_to_pfn(*pages++); + *((uint64_t *)cmd) = vmw_piter_dma_addr(iter) >> + PAGE_SHIFT; cmd += VMW_PPN_SIZE / sizeof(*cmd); + vmw_piter_next(iter); } num_pages -= nr; @@ -120,181 +124,28 @@ static void vmw_gmr2_unbind(struct vmw_private *dev_priv, vmw_fifo_commit(dev_priv, define_size); } -/** - * FIXME: Adjust to the ttm lowmem / highmem storage to minimize - * the number of used descriptors. - */ - -static int vmw_gmr_build_descriptors(struct list_head *desc_pages, - struct page *pages[], - unsigned long num_pages) -{ - struct page *page, *next; - 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; - - 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); - - /* - * Point previous page terminating descriptor to this - * page before unmapping it. - */ - - if (likely(page_virtual != NULL)) { - desc_virtual->ppn = page_to_pfn(page); - kunmap_atomic(page_virtual); - } - - page_virtual = kmap_atomic(page); - desc_virtual = page_virtual - 1; - prev_pfn = ~(0UL); - - while (likely(num_pages != 0)) { - pfn = page_to_pfn(*pages); - - 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; - ++pages; - } - - (++desc_virtual)->ppn = cpu_to_le32(0); - desc_virtual->num_pages = cpu_to_le32(0); - } - - if (likely(page_virtual != NULL)) - kunmap_atomic(page_virtual); - - return 0; -out_err: - list_for_each_entry_safe(page, next, desc_pages, lru) { - list_del_init(&page->lru); - __free_page(page); - } - return ret; -} - -static inline void vmw_gmr_free_descriptors(struct list_head *desc_pages) -{ - struct page *page, *next; - - list_for_each_entry_safe(page, next, desc_pages, lru) { - list_del_init(&page->lru); - __free_page(page); - } -} - -static void vmw_gmr_fire_descriptors(struct vmw_private *dev_priv, - int gmr_id, struct list_head *desc_pages) -{ - struct page *page; - - if (unlikely(list_empty(desc_pages))) - return; - - page = list_entry(desc_pages->next, struct page, lru); - - mutex_lock(&dev_priv->hw_mutex); - - vmw_write(dev_priv, SVGA_REG_GMR_ID, gmr_id); - wmb(); - vmw_write(dev_priv, SVGA_REG_GMR_DESCRIPTOR, page_to_pfn(page)); - mb(); - - mutex_unlock(&dev_priv->hw_mutex); - -} - -/** - * FIXME: Adjust to the ttm lowmem / highmem storage to minimize - * the number of used descriptors. - */ - -static unsigned long vmw_gmr_count_descriptors(struct page *pages[], - unsigned long num_pages) -{ - unsigned long prev_pfn = ~(0UL); - unsigned long pfn; - unsigned long descriptors = 0; - - while (num_pages--) { - pfn = page_to_pfn(*pages++); - if (prev_pfn + 1 != pfn) - ++descriptors; - prev_pfn = pfn; - } - - return descriptors; -} int vmw_gmr_bind(struct vmw_private *dev_priv, - struct page *pages[], + const struct vmw_sg_table *vsgt, unsigned long num_pages, int gmr_id) { - struct list_head desc_pages; - int ret; + struct vmw_piter data_iter; - if (likely(dev_priv->capabilities & SVGA_CAP_GMR2)) - return vmw_gmr2_bind(dev_priv, pages, num_pages, gmr_id); + vmw_piter_start(&data_iter, vsgt, 0); - if (unlikely(!(dev_priv->capabilities & SVGA_CAP_GMR))) - return -EINVAL; + if (unlikely(!vmw_piter_next(&data_iter))) + return 0; - if (vmw_gmr_count_descriptors(pages, num_pages) > - 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(&desc_pages, pages, num_pages); - if (unlikely(ret != 0)) - return ret; - - vmw_gmr_fire_descriptors(dev_priv, gmr_id, &desc_pages); - vmw_gmr_free_descriptors(&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); } diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_gmrid_manager.c b/drivers/gpu/drm/vmwgfx/vmwgfx_gmrid_manager.c index c5c054ae905..b1273e8e9a6 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_gmrid_manager.c +++ b/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; } diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c b/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c index c509d40c489..37881ecf5d7 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c @@ -29,12 +29,18 @@ #include <drm/vmwgfx_drm.h> #include "vmwgfx_kms.h" +struct svga_3d_compat_cap { + SVGA3dCapsRecordHeader header; + SVGA3dCapPair pairs[SVGA3D_DEVCAP_MAX]; +}; + int vmw_getparam_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv) { struct vmw_private *dev_priv = vmw_priv(dev); struct drm_vmw_getparam_arg *param = (struct drm_vmw_getparam_arg *)data; + struct vmw_fpriv *vmw_fp = vmw_fpriv(file_priv); switch (param->param) { case DRM_VMW_PARAM_NUM_STREAMS: @@ -53,13 +59,18 @@ 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: { __le32 __iomem *fifo_mem = dev_priv->mmio_virt; const struct vmw_fifo_state *fifo = &dev_priv->fifo; + if ((dev_priv->capabilities & SVGA_CAP_GBOBJECTS)) { + param->value = SVGA3D_HWVERSION_WS8_B1; + break; + } + param->value = ioread32(fifo_mem + ((fifo->capabilities & @@ -68,6 +79,32 @@ int vmw_getparam_ioctl(struct drm_device *dev, void *data, SVGA_FIFO_3D_HWVERSION)); break; } + case DRM_VMW_PARAM_MAX_SURF_MEMORY: + if ((dev_priv->capabilities & SVGA_CAP_GBOBJECTS) && + !vmw_fp->gb_aware) + param->value = dev_priv->max_mob_pages * PAGE_SIZE / 2; + else + param->value = dev_priv->memory_size; + break; + case DRM_VMW_PARAM_3D_CAPS_SIZE: + if ((dev_priv->capabilities & SVGA_CAP_GBOBJECTS) && + vmw_fp->gb_aware) + param->value = SVGA3D_DEVCAP_MAX * sizeof(uint32_t); + else if (dev_priv->capabilities & SVGA_CAP_GBOBJECTS) + param->value = sizeof(struct svga_3d_compat_cap) + + sizeof(uint32_t); + else + param->value = (SVGA_FIFO_3D_CAPS_LAST - + SVGA_FIFO_3D_CAPS + 1) * + sizeof(uint32_t); + break; + case DRM_VMW_PARAM_MAX_MOB_MEMORY: + vmw_fp->gb_aware = true; + param->value = dev_priv->max_mob_pages * PAGE_SIZE; + break; + case DRM_VMW_PARAM_MAX_MOB_SIZE: + param->value = dev_priv->max_mob_size; + break; default: DRM_ERROR("Illegal vmwgfx get param request: %d\n", param->param); @@ -77,6 +114,38 @@ int vmw_getparam_ioctl(struct drm_device *dev, void *data, return 0; } +static int vmw_fill_compat_cap(struct vmw_private *dev_priv, void *bounce, + size_t size) +{ + struct svga_3d_compat_cap *compat_cap = + (struct svga_3d_compat_cap *) bounce; + unsigned int i; + size_t pair_offset = offsetof(struct svga_3d_compat_cap, pairs); + unsigned int max_size; + + if (size < pair_offset) + return -EINVAL; + + max_size = (size - pair_offset) / sizeof(SVGA3dCapPair); + + if (max_size > SVGA3D_DEVCAP_MAX) + max_size = SVGA3D_DEVCAP_MAX; + + compat_cap->header.length = + (pair_offset + max_size * sizeof(SVGA3dCapPair)) / sizeof(u32); + compat_cap->header.type = SVGA3DCAPS_RECORD_DEVCAPS; + + mutex_lock(&dev_priv->hw_mutex); + for (i = 0; i < max_size; ++i) { + vmw_write(dev_priv, SVGA_REG_DEV_CAP, i); + compat_cap->pairs[i][0] = i; + compat_cap->pairs[i][1] = vmw_read(dev_priv, SVGA_REG_DEV_CAP); + } + mutex_unlock(&dev_priv->hw_mutex); + + return 0; +} + int vmw_get_cap_3d_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv) @@ -89,29 +158,58 @@ 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); + struct vmw_fpriv *vmw_fp = vmw_fpriv(file_priv); 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 && vmw_fp->gb_aware) + size = SVGA3D_DEVCAP_MAX * sizeof(uint32_t); + else if (gb_objects) + size = sizeof(struct svga_3d_compat_cap) + sizeof(uint32_t); + else + size = (SVGA_FIFO_3D_CAPS_LAST - SVGA_FIFO_3D_CAPS + 1) * + sizeof(uint32_t); if (arg->max_size < size) size = arg->max_size; - bounce = vmalloc(size); + bounce = vzalloc(size); if (unlikely(bounce == NULL)) { DRM_ERROR("Failed to allocate bounce buffer for 3D caps.\n"); return -ENOMEM; } - fifo_mem = dev_priv->mmio_virt; - memcpy_fromio(bounce, &fifo_mem[SVGA_FIFO_3D_CAPS], size); + if (gb_objects && vmw_fp->gb_aware) { + int i, num; + uint32_t *bounce32 = (uint32_t *) bounce; + + num = size / sizeof(uint32_t); + if (num > SVGA3D_DEVCAP_MAX) + num = SVGA3D_DEVCAP_MAX; + + mutex_lock(&dev_priv->hw_mutex); + for (i = 0; i < num; ++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 if (gb_objects) { + ret = vmw_fill_compat_cap(dev_priv, bounce, size); + if (unlikely(ret != 0)) + goto out_err; + } 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) ret = -EFAULT; +out_err: vfree(bounce); if (unlikely(ret != 0)) @@ -128,7 +226,6 @@ int vmw_present_ioctl(struct drm_device *dev, void *data, struct drm_vmw_present_arg *arg = (struct drm_vmw_present_arg *)data; struct vmw_surface *surface; - struct vmw_master *vmaster = vmw_master(file_priv->master); struct drm_vmw_rect __user *clips_ptr; struct drm_vmw_rect *clips = NULL; struct drm_framebuffer *fb; @@ -168,12 +265,12 @@ int vmw_present_ioctl(struct drm_device *dev, void *data, fb = drm_framebuffer_lookup(dev, arg->fb_id); if (!fb) { DRM_ERROR("Invalid framebuffer id.\n"); - ret = -EINVAL; + ret = -ENOENT; goto out_no_fb; } vfb = vmw_framebuffer_to_vfb(fb); - ret = ttm_read_lock(&vmaster->lock, true); + ret = ttm_read_lock(&dev_priv->reservation_sem, true); if (unlikely(ret != 0)) goto out_no_ttm_lock; @@ -193,7 +290,7 @@ int vmw_present_ioctl(struct drm_device *dev, void *data, vmw_surface_unreference(&surface); out_no_surface: - ttm_read_unlock(&vmaster->lock); + ttm_read_unlock(&dev_priv->reservation_sem); out_no_ttm_lock: drm_framebuffer_unreference(fb); out_no_fb: @@ -213,7 +310,6 @@ int vmw_present_readback_ioctl(struct drm_device *dev, void *data, struct drm_vmw_fence_rep __user *user_fence_rep = (struct drm_vmw_fence_rep __user *) (unsigned long)arg->fence_rep; - struct vmw_master *vmaster = vmw_master(file_priv->master); struct drm_vmw_rect __user *clips_ptr; struct drm_vmw_rect *clips = NULL; struct drm_framebuffer *fb; @@ -252,7 +348,7 @@ int vmw_present_readback_ioctl(struct drm_device *dev, void *data, fb = drm_framebuffer_lookup(dev, arg->fb_id); if (!fb) { DRM_ERROR("Invalid framebuffer id.\n"); - ret = -EINVAL; + ret = -ENOENT; goto out_no_fb; } @@ -263,7 +359,7 @@ int vmw_present_readback_ioctl(struct drm_device *dev, void *data, goto out_no_ttm_lock; } - ret = ttm_read_lock(&vmaster->lock, true); + ret = ttm_read_lock(&dev_priv->reservation_sem, true); if (unlikely(ret != 0)) goto out_no_ttm_lock; @@ -271,7 +367,7 @@ int vmw_present_readback_ioctl(struct drm_device *dev, void *data, vfb, user_fence_rep, clips, num_clips); - ttm_read_unlock(&vmaster->lock); + ttm_read_unlock(&dev_priv->reservation_sem); out_no_ttm_lock: drm_framebuffer_unreference(fb); out_no_fb: diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c b/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c index 4640adbcaf9..0c423766c44 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c @@ -30,7 +30,7 @@ #define VMW_FENCE_WRAP (1 << 24) -irqreturn_t vmw_irq_handler(DRM_IRQ_ARGS) +irqreturn_t vmw_irq_handler(int irq, void *arg) { struct drm_device *dev = (struct drm_device *)arg; struct vmw_private *dev_priv = vmw_priv(dev); diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c index fc43c060123..8f3edc4710f 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c @@ -40,7 +40,7 @@ struct vmw_clip_rect { * Clip @num_rects number of @rects against @clip storing the * results in @out_rects and the number of passed rects in @out_num. */ -void vmw_clip_cliprects(struct drm_clip_rect *rects, +static void vmw_clip_cliprects(struct drm_clip_rect *rects, int num_rects, struct vmw_clip_rect clip, SVGASignedRect *out_rects, @@ -75,6 +75,7 @@ void vmw_display_unit_cleanup(struct vmw_display_unit *du) vmw_surface_unreference(&du->cursor_surface); if (du->cursor_dmabuf) vmw_dmabuf_unreference(&du->cursor_dmabuf); + drm_sysfs_connector_remove(&du->connector); drm_crtc_cleanup(&du->crtc); drm_encoder_cleanup(&du->encoder); drm_connector_cleanup(&du->connector); @@ -186,7 +187,7 @@ int vmw_du_crtc_cursor_set(struct drm_crtc *crtc, struct drm_file *file_priv, * can do this since the caller in the drm core doesn't check anything * which is protected by any looks. */ - mutex_unlock(&crtc->mutex); + drm_modeset_unlock(&crtc->mutex); drm_modeset_lock_all(dev_priv->dev); /* A lot of the code assumes this */ @@ -251,7 +252,7 @@ int vmw_du_crtc_cursor_set(struct drm_crtc *crtc, struct drm_file *file_priv, ret = 0; out: drm_modeset_unlock_all(dev_priv->dev); - mutex_lock(&crtc->mutex); + drm_modeset_lock(&crtc->mutex, NULL); return ret; } @@ -272,7 +273,7 @@ int vmw_du_crtc_cursor_move(struct drm_crtc *crtc, int x, int y) * can do this since the caller in the drm core doesn't check anything * which is protected by any looks. */ - mutex_unlock(&crtc->mutex); + drm_modeset_unlock(&crtc->mutex); drm_modeset_lock_all(dev_priv->dev); vmw_cursor_update_position(dev_priv, shown, @@ -280,7 +281,7 @@ int vmw_du_crtc_cursor_move(struct drm_crtc *crtc, int x, int y) du->cursor_y + du->hotspot_y); drm_modeset_unlock_all(dev_priv->dev); - mutex_lock(&crtc->mutex); + drm_modeset_lock(&crtc->mutex, NULL); return 0; } @@ -422,7 +423,7 @@ struct vmw_framebuffer_surface { struct drm_master *master; }; -void vmw_framebuffer_surface_destroy(struct drm_framebuffer *framebuffer) +static void vmw_framebuffer_surface_destroy(struct drm_framebuffer *framebuffer) { struct vmw_framebuffer_surface *vfbs = vmw_framebuffer_to_vfbs(framebuffer); @@ -467,7 +468,7 @@ static int do_surface_dirty_sou(struct vmw_private *dev_priv, num_units = 0; list_for_each_entry(crtc, &dev_priv->dev->mode_config.crtc_list, head) { - if (crtc->fb != &framebuffer->base) + if (crtc->primary->fb != &framebuffer->base) continue; units[num_units++] = vmw_crtc_to_du(crtc); } @@ -588,14 +589,13 @@ out_free_tmp: return ret; } -int vmw_framebuffer_surface_dirty(struct drm_framebuffer *framebuffer, +static int vmw_framebuffer_surface_dirty(struct drm_framebuffer *framebuffer, struct drm_file *file_priv, unsigned flags, unsigned color, struct drm_clip_rect *clips, unsigned num_clips) { struct vmw_private *dev_priv = vmw_priv(framebuffer->dev); - struct vmw_master *vmaster = vmw_master(file_priv->master); struct vmw_framebuffer_surface *vfbs = vmw_framebuffer_to_vfbs(framebuffer); struct drm_clip_rect norect; @@ -608,9 +608,13 @@ int vmw_framebuffer_surface_dirty(struct drm_framebuffer *framebuffer, if (!dev_priv->sou_priv) return -EINVAL; - ret = ttm_read_lock(&vmaster->lock, true); - if (unlikely(ret != 0)) + drm_modeset_lock_all(dev_priv->dev); + + ret = ttm_read_lock(&dev_priv->reservation_sem, true); + if (unlikely(ret != 0)) { + drm_modeset_unlock_all(dev_priv->dev); return ret; + } if (!num_clips) { num_clips = 1; @@ -627,7 +631,10 @@ int vmw_framebuffer_surface_dirty(struct drm_framebuffer *framebuffer, flags, color, clips, num_clips, inc, NULL); - ttm_read_unlock(&vmaster->lock); + ttm_read_unlock(&dev_priv->reservation_sem); + + drm_modeset_unlock_all(dev_priv->dev); + return 0; } @@ -664,9 +671,9 @@ 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; @@ -753,7 +760,7 @@ struct vmw_framebuffer_dmabuf { struct vmw_dma_buffer *buffer; }; -void vmw_framebuffer_dmabuf_destroy(struct drm_framebuffer *framebuffer) +static void vmw_framebuffer_dmabuf_destroy(struct drm_framebuffer *framebuffer) { struct vmw_framebuffer_dmabuf *vfbd = vmw_framebuffer_to_vfbd(framebuffer); @@ -875,7 +882,7 @@ static int do_dmabuf_dirty_sou(struct drm_file *file_priv, num_units = 0; list_for_each_entry(crtc, &dev_priv->dev->mode_config.crtc_list, head) { - if (crtc->fb != &framebuffer->base) + if (crtc->primary->fb != &framebuffer->base) continue; units[num_units++] = vmw_crtc_to_du(crtc); } @@ -939,22 +946,25 @@ static int do_dmabuf_dirty_sou(struct drm_file *file_priv, return ret; } -int vmw_framebuffer_dmabuf_dirty(struct drm_framebuffer *framebuffer, +static int vmw_framebuffer_dmabuf_dirty(struct drm_framebuffer *framebuffer, struct drm_file *file_priv, unsigned flags, unsigned color, struct drm_clip_rect *clips, unsigned num_clips) { struct vmw_private *dev_priv = vmw_priv(framebuffer->dev); - struct vmw_master *vmaster = vmw_master(file_priv->master); struct vmw_framebuffer_dmabuf *vfbd = vmw_framebuffer_to_vfbd(framebuffer); struct drm_clip_rect norect; int ret, increment = 1; - ret = ttm_read_lock(&vmaster->lock, true); - if (unlikely(ret != 0)) + drm_modeset_lock_all(dev_priv->dev); + + ret = ttm_read_lock(&dev_priv->reservation_sem, true); + if (unlikely(ret != 0)) { + drm_modeset_unlock_all(dev_priv->dev); return ret; + } if (!num_clips) { num_clips = 1; @@ -977,7 +987,10 @@ int vmw_framebuffer_dmabuf_dirty(struct drm_framebuffer *framebuffer, clips, num_clips, increment, NULL); } - ttm_read_unlock(&vmaster->lock); + ttm_read_unlock(&dev_priv->reservation_sem); + + drm_modeset_unlock_all(dev_priv->dev); + return ret; } @@ -1230,7 +1243,7 @@ int vmw_kms_present(struct vmw_private *dev_priv, num_units = 0; list_for_each_entry(crtc, &dev_priv->dev->mode_config.crtc_list, head) { - if (crtc->fb != &vfb->base) + if (crtc->primary->fb != &vfb->base) continue; units[num_units++] = vmw_crtc_to_du(crtc); } @@ -1367,7 +1380,7 @@ int vmw_kms_readback(struct vmw_private *dev_priv, num_units = 0; list_for_each_entry(crtc, &dev_priv->dev->mode_config.crtc_list, head) { - if (crtc->fb != &vfb->base) + if (crtc->primary->fb != &vfb->base) continue; units[num_units++] = vmw_crtc_to_du(crtc); } @@ -1508,7 +1521,7 @@ int vmw_kms_cursor_bypass_ioctl(struct drm_device *dev, void *data, obj = drm_mode_object_find(dev, arg->crtc_id, DRM_MODE_OBJECT_CRTC); if (!obj) { - ret = -EINVAL; + ret = -ENOENT; goto out; } @@ -1630,7 +1643,7 @@ 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; } @@ -1662,7 +1675,7 @@ void vmw_disable_vblank(struct drm_device *dev, int crtc) * Small shared kms functions. */ -int vmw_du_update_layout(struct vmw_private *dev_priv, unsigned num, +static int vmw_du_update_layout(struct vmw_private *dev_priv, unsigned num, struct drm_vmw_rect *rects) { struct drm_device *dev = dev_priv->dev; @@ -1710,7 +1723,7 @@ int vmw_du_page_flip(struct drm_crtc *crtc, uint32_t page_flip_flags) { struct vmw_private *dev_priv = vmw_priv(crtc->dev); - struct drm_framebuffer *old_fb = crtc->fb; + struct drm_framebuffer *old_fb = crtc->primary->fb; struct vmw_framebuffer *vfb = vmw_framebuffer_to_vfb(fb); struct drm_file *file_priv ; struct vmw_fence_obj *fence = NULL; @@ -1728,7 +1741,7 @@ int vmw_du_page_flip(struct drm_crtc *crtc, if (!vmw_kms_screen_object_flippable(dev_priv, crtc)) return -EINVAL; - crtc->fb = fb; + crtc->primary->fb = fb; /* do a full screen dirty update */ clips.x1 = clips.y1 = 0; @@ -1768,7 +1781,7 @@ int vmw_du_page_flip(struct drm_crtc *crtc, return ret; out_no_fence: - crtc->fb = old_fb; + crtc->primary->fb = old_fb; return ret; } @@ -1988,7 +2001,7 @@ int vmw_du_connector_fill_modes(struct drm_connector *connector, if (du->pref_mode) list_move(&du->pref_mode->head, &connector->probed_modes); - drm_mode_connector_list_update(connector); + drm_mode_connector_list_update(connector, true); return 1; } @@ -2007,7 +2020,6 @@ int vmw_kms_update_layout_ioctl(struct drm_device *dev, void *data, struct vmw_private *dev_priv = vmw_priv(dev); struct drm_vmw_update_layout_arg *arg = (struct drm_vmw_update_layout_arg *)data; - struct vmw_master *vmaster = vmw_master(file_priv->master); void __user *user_rects; struct drm_vmw_rect *rects; unsigned rects_size; @@ -2015,7 +2027,7 @@ int vmw_kms_update_layout_ioctl(struct drm_device *dev, void *data, int i; struct drm_mode_config *mode_config = &dev->mode_config; - ret = ttm_read_lock(&vmaster->lock, true); + ret = ttm_read_lock(&dev_priv->reservation_sem, true); if (unlikely(ret != 0)) return ret; @@ -2057,6 +2069,6 @@ int vmw_kms_update_layout_ioctl(struct drm_device *dev, void *data, out_free: kfree(rects); out_unlock: - ttm_read_unlock(&vmaster->lock); + ttm_read_unlock(&dev_priv->reservation_sem); return ret; } diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c b/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c index 79f7e8e6052..b2b9bd23aee 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c @@ -93,7 +93,7 @@ static int vmw_ldu_commit_list(struct vmw_private *dev_priv) if (crtc == NULL) return 0; - fb = entry->base.crtc.fb; + fb = entry->base.crtc.primary->fb; return vmw_kms_write_svga(dev_priv, w, h, fb->pitches[0], fb->bits_per_pixel, fb->depth); @@ -101,7 +101,7 @@ static int vmw_ldu_commit_list(struct vmw_private *dev_priv) if (!list_empty(&lds->active)) { entry = list_entry(lds->active.next, typeof(*entry), active); - fb = entry->base.crtc.fb; + fb = entry->base.crtc.primary->fb; vmw_kms_write_svga(dev_priv, fb->width, fb->height, fb->pitches[0], fb->bits_per_pixel, fb->depth); @@ -259,7 +259,8 @@ static int vmw_ldu_crtc_set_config(struct drm_mode_set *set) connector->encoder = NULL; encoder->crtc = NULL; - crtc->fb = NULL; + crtc->primary->fb = NULL; + crtc->enabled = false; vmw_ldu_del_active(dev_priv, ldu); @@ -279,12 +280,13 @@ static int vmw_ldu_crtc_set_config(struct drm_mode_set *set) vmw_fb_off(dev_priv); - crtc->fb = fb; + crtc->primary->fb = fb; encoder->crtc = crtc; connector->encoder = encoder; crtc->x = set->x; crtc->y = set->y; crtc->mode = *mode; + crtc->enabled = true; vmw_ldu_add_active(dev_priv, ldu, vfb); @@ -369,6 +371,8 @@ static int vmw_ldu_init(struct vmw_private *dev_priv, unsigned unit) encoder->possible_crtcs = (1 << unit); encoder->possible_clones = 0; + (void) drm_sysfs_connector_add(connector); + drm_crtc_init(dev, crtc, &vmw_legacy_crtc_funcs); drm_mode_crtc_set_gamma_size(crtc, 256); diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c b/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c new file mode 100644 index 00000000000..04a64b8cd3c --- /dev/null +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c @@ -0,0 +1,656 @@ +/************************************************************************** + * + * 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_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_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; + SVGA3dCmdSetOTableBase64 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"); + ret = -ENOMEM; + goto out_no_fifo; + } + + memset(cmd, 0, sizeof(*cmd)); + cmd->header.id = SVGA_3D_CMD_SET_OTABLE_BASE64; + cmd->header.size = sizeof(cmd->body); + cmd->body.type = type; + cmd->body.baseAddress = cpu_to_le64(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; + + if (otable->page_table == NULL) + return; + + bo = otable->page_table->pt_bo; + cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd)); + if (unlikely(cmd == NULL)) { + DRM_ERROR("Failed reserving FIFO space for OTable " + "takedown.\n"); + } else { + 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, NULL); + 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, NULL); + 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, NULL); + 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, NULL); + + 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(__le32 **addr, dma_addr_t val) +{ + *((__le64 *) *addr) = cpu_to_le64(val >> PAGE_SHIFT); + *addr += 2; +} +#else +static void vmw_mob_assign_ppn(__le32 **addr, dma_addr_t val) +{ + *(*addr)++ = cpu_to_le32(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; + __le32 *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, NULL); + 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, NULL); + /* + * 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"); + } else { + 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; + SVGA3dCmdDefineGBMob64 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 = SVGA_3D_CMD_DEFINE_GB_MOB64; + cmd->header.size = sizeof(cmd->body); + cmd->body.mobid = mob_id; + cmd->body.ptDepth = mob->pt_level; + cmd->body.base = cpu_to_le64(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; +} diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_prime.c b/drivers/gpu/drm/vmwgfx/vmwgfx_prime.c new file mode 100644 index 00000000000..31fe32d8d65 --- /dev/null +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_prime.c @@ -0,0 +1,137 @@ +/************************************************************************** + * + * Copyright © 2013 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. + * + **************************************************************************/ +/* + * Authors: + * Thomas Hellstrom <thellstrom@vmware.com> + * + */ + +#include "vmwgfx_drv.h" +#include <linux/dma-buf.h> +#include <drm/ttm/ttm_object.h> + +/* + * DMA-BUF attach- and mapping methods. No need to implement + * these until we have other virtual devices use them. + */ + +static int vmw_prime_map_attach(struct dma_buf *dma_buf, + struct device *target_dev, + struct dma_buf_attachment *attach) +{ + return -ENOSYS; +} + +static void vmw_prime_map_detach(struct dma_buf *dma_buf, + struct dma_buf_attachment *attach) +{ +} + +static struct sg_table *vmw_prime_map_dma_buf(struct dma_buf_attachment *attach, + enum dma_data_direction dir) +{ + return ERR_PTR(-ENOSYS); +} + +static void vmw_prime_unmap_dma_buf(struct dma_buf_attachment *attach, + struct sg_table *sgb, + enum dma_data_direction dir) +{ +} + +static void *vmw_prime_dmabuf_vmap(struct dma_buf *dma_buf) +{ + return NULL; +} + +static void vmw_prime_dmabuf_vunmap(struct dma_buf *dma_buf, void *vaddr) +{ +} + +static void *vmw_prime_dmabuf_kmap_atomic(struct dma_buf *dma_buf, + unsigned long page_num) +{ + return NULL; +} + +static void vmw_prime_dmabuf_kunmap_atomic(struct dma_buf *dma_buf, + unsigned long page_num, void *addr) +{ + +} +static void *vmw_prime_dmabuf_kmap(struct dma_buf *dma_buf, + unsigned long page_num) +{ + return NULL; +} + +static void vmw_prime_dmabuf_kunmap(struct dma_buf *dma_buf, + unsigned long page_num, void *addr) +{ + +} + +static int vmw_prime_dmabuf_mmap(struct dma_buf *dma_buf, + struct vm_area_struct *vma) +{ + WARN_ONCE(true, "Attempted use of dmabuf mmap. Bad.\n"); + return -ENOSYS; +} + +const struct dma_buf_ops vmw_prime_dmabuf_ops = { + .attach = vmw_prime_map_attach, + .detach = vmw_prime_map_detach, + .map_dma_buf = vmw_prime_map_dma_buf, + .unmap_dma_buf = vmw_prime_unmap_dma_buf, + .release = NULL, + .kmap = vmw_prime_dmabuf_kmap, + .kmap_atomic = vmw_prime_dmabuf_kmap_atomic, + .kunmap = vmw_prime_dmabuf_kunmap, + .kunmap_atomic = vmw_prime_dmabuf_kunmap_atomic, + .mmap = vmw_prime_dmabuf_mmap, + .vmap = vmw_prime_dmabuf_vmap, + .vunmap = vmw_prime_dmabuf_vunmap, +}; + +int vmw_prime_fd_to_handle(struct drm_device *dev, + struct drm_file *file_priv, + int fd, u32 *handle) +{ + struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile; + + return ttm_prime_fd_to_handle(tfile, fd, handle); +} + +int vmw_prime_handle_to_fd(struct drm_device *dev, + struct drm_file *file_priv, + uint32_t handle, uint32_t flags, + int *prime_fd) +{ + struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile; + + return ttm_prime_handle_to_fd(tfile, handle, flags, prime_fd); +} diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c index 0e67cf41065..01d68f0a69d 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c @@ -32,8 +32,10 @@ #include <drm/drmP.h> #include "vmwgfx_resource_priv.h" +#define VMW_RES_EVICT_ERR_COUNT 10 + struct vmw_user_dma_buffer { - struct ttm_base_object base; + struct ttm_prime_object prime; struct vmw_dma_buffer dma; }; @@ -86,6 +88,11 @@ struct vmw_resource *vmw_resource_reference(struct vmw_resource *res) return res; } +struct vmw_resource * +vmw_resource_reference_unless_doomed(struct vmw_resource *res) +{ + return kref_get_unless_zero(&res->kref) ? res : NULL; +} /** * vmw_resource_release_id - release a resource id to the id manager. @@ -134,8 +141,12 @@ static void vmw_resource_release(struct kref *kref) vmw_dmabuf_unreference(&res->backup); } - if (likely(res->hw_destroy != NULL)) + if (likely(res->hw_destroy != NULL)) { res->hw_destroy(res); + mutex_lock(&dev_priv->binding_mutex); + vmw_context_binding_res_list_kill(&res->binding_head); + mutex_unlock(&dev_priv->binding_mutex); + } id = res->id; if (res->res_free != NULL) @@ -213,6 +224,7 @@ 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; @@ -295,7 +307,7 @@ int vmw_user_resource_lookup_handle(struct vmw_private *dev_priv, if (unlikely(base == NULL)) return -EINVAL; - if (unlikely(base->object_type != converter->object_type)) + if (unlikely(ttm_base_object_type(base) != converter->object_type)) goto out_bad_resource; res = converter->base_obj_to_res(base); @@ -350,6 +362,38 @@ int vmw_user_lookup_handle(struct vmw_private *dev_priv, /** * Buffer management. */ + +/** + * vmw_dmabuf_acc_size - Calculate the pinned memory usage of buffers + * + * @dev_priv: Pointer to a struct vmw_private identifying the device. + * @size: The requested buffer size. + * @user: Whether this is an ordinary dma buffer or a user dma buffer. + */ +static size_t vmw_dmabuf_acc_size(struct vmw_private *dev_priv, size_t size, + bool user) +{ + static size_t struct_size, user_struct_size; + size_t num_pages = PAGE_ALIGN(size) >> PAGE_SHIFT; + size_t page_array_size = ttm_round_pot(num_pages * sizeof(void *)); + + if (unlikely(struct_size == 0)) { + size_t backend_size = ttm_round_pot(vmw_tt_size); + + struct_size = backend_size + + ttm_round_pot(sizeof(struct vmw_dma_buffer)); + user_struct_size = backend_size + + ttm_round_pot(sizeof(struct vmw_user_dma_buffer)); + } + + if (dev_priv->map_mode == vmw_dma_alloc_coherent) + page_array_size += + ttm_round_pot(num_pages * sizeof(dma_addr_t)); + + return ((user) ? user_struct_size : struct_size) + + page_array_size; +} + void vmw_dmabuf_bo_free(struct ttm_buffer_object *bo) { struct vmw_dma_buffer *vmw_bo = vmw_dma_buffer(bo); @@ -357,6 +401,13 @@ void vmw_dmabuf_bo_free(struct ttm_buffer_object *bo) kfree(vmw_bo); } +static void vmw_user_dmabuf_destroy(struct ttm_buffer_object *bo) +{ + struct vmw_user_dma_buffer *vmw_user_bo = vmw_user_dma_buffer(bo); + + ttm_prime_object_kfree(vmw_user_bo, prime); +} + int vmw_dmabuf_init(struct vmw_private *dev_priv, struct vmw_dma_buffer *vmw_bo, size_t size, struct ttm_placement *placement, @@ -366,10 +417,11 @@ int vmw_dmabuf_init(struct vmw_private *dev_priv, struct ttm_bo_device *bdev = &dev_priv->bdev; size_t acc_size; int ret; + bool user = (bo_free == &vmw_user_dmabuf_destroy); - BUG_ON(!bo_free); + BUG_ON(!bo_free && (!user && (bo_free != vmw_dmabuf_bo_free))); - acc_size = ttm_bo_acc_size(bdev, size, sizeof(struct vmw_dma_buffer)); + acc_size = vmw_dmabuf_acc_size(dev_priv, size, user); memset(vmw_bo, 0, sizeof(*vmw_bo)); INIT_LIST_HEAD(&vmw_bo->res_list); @@ -381,13 +433,6 @@ int vmw_dmabuf_init(struct vmw_private *dev_priv, return ret; } -static void vmw_user_dmabuf_destroy(struct ttm_buffer_object *bo) -{ - struct vmw_user_dma_buffer *vmw_user_bo = vmw_user_dma_buffer(bo); - - ttm_base_object_kfree(vmw_user_bo, base); -} - static void vmw_user_dmabuf_release(struct ttm_base_object **p_base) { struct vmw_user_dma_buffer *vmw_user_bo; @@ -399,11 +444,27 @@ static void vmw_user_dmabuf_release(struct ttm_base_object **p_base) if (unlikely(base == NULL)) return; - vmw_user_bo = container_of(base, struct vmw_user_dma_buffer, base); + vmw_user_bo = container_of(base, struct vmw_user_dma_buffer, + prime.base); bo = &vmw_user_bo->dma.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 * @@ -434,24 +495,28 @@ 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)) return ret; tmp = ttm_bo_reference(&user_bo->dma.base); - ret = ttm_base_object_init(tfile, - &user_bo->base, - shareable, - ttm_buffer_type, - &vmw_user_dmabuf_release, NULL); + ret = ttm_prime_object_init(tfile, + size, + &user_bo->prime, + shareable, + ttm_buffer_type, + &vmw_user_dmabuf_release, + &vmw_user_dmabuf_ref_obj_release); if (unlikely(ret != 0)) { ttm_bo_unref(&tmp); goto out_no_base_object; } *p_dma_buf = &user_bo->dma; - *handle = user_bo->base.hash.key; + *handle = user_bo->prime.base.hash.key; out_no_base_object: return ret; @@ -473,8 +538,137 @@ int vmw_user_dmabuf_verify_access(struct ttm_buffer_object *bo, return -EPERM; vmw_user_bo = vmw_user_dma_buffer(bo); - return (vmw_user_bo->base.tfile == tfile || - vmw_user_bo->base.shareable) ? 0 : -EPERM; + + /* Check that the caller has opened the object. */ + if (likely(ttm_ref_object_exists(tfile, &vmw_user_bo->prime.base))) + return 0; + + DRM_ERROR("Could not grant buffer access.\n"); + return -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, @@ -487,10 +681,9 @@ int vmw_dmabuf_alloc_ioctl(struct drm_device *dev, void *data, struct drm_vmw_dmabuf_rep *rep = &arg->rep; struct vmw_dma_buffer *dma_buf; uint32_t handle; - struct vmw_master *vmaster = vmw_master(file_priv->master); int ret; - ret = ttm_read_lock(&vmaster->lock, true); + ret = ttm_read_lock(&dev_priv->reservation_sem, true); if (unlikely(ret != 0)) return ret; @@ -507,7 +700,7 @@ int vmw_dmabuf_alloc_ioctl(struct drm_device *dev, void *data, vmw_dmabuf_unreference(&dma_buf); out_no_dmabuf: - ttm_read_unlock(&vmaster->lock); + ttm_read_unlock(&dev_priv->reservation_sem); return ret; } @@ -536,14 +729,15 @@ int vmw_user_dmabuf_lookup(struct ttm_object_file *tfile, return -ESRCH; } - if (unlikely(base->object_type != ttm_buffer_type)) { + if (unlikely(ttm_base_object_type(base) != ttm_buffer_type)) { ttm_base_object_unref(&base); printk(KERN_ERR "Invalid buffer object handle 0x%08lx.\n", (unsigned long)handle); return -EINVAL; } - vmw_user_bo = container_of(base, struct vmw_user_dma_buffer, base); + vmw_user_bo = container_of(base, struct vmw_user_dma_buffer, + prime.base); (void)ttm_bo_reference(&vmw_user_bo->dma.base); ttm_base_object_unref(&base); *out = &vmw_user_bo->dma; @@ -552,7 +746,8 @@ 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; @@ -560,7 +755,10 @@ int vmw_user_dmabuf_reference(struct ttm_object_file *tfile, return -EINVAL; user_bo = container_of(dma_buf, struct vmw_user_dma_buffer, dma); - return ttm_ref_object_add(tfile, &user_bo->base, TTM_REF_USAGE, NULL); + + *handle = user_bo->prime.base.hash.key; + return ttm_ref_object_add(tfile, &user_bo->prime.base, + TTM_REF_USAGE, NULL); } /* @@ -679,7 +877,6 @@ int vmw_stream_claim_ioctl(struct drm_device *dev, void *data, struct vmw_resource *tmp; struct drm_vmw_stream_arg *arg = (struct drm_vmw_stream_arg *)data; struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile; - struct vmw_master *vmaster = vmw_master(file_priv->master); int ret; /* @@ -690,7 +887,7 @@ int vmw_stream_claim_ioctl(struct drm_device *dev, void *data, if (unlikely(vmw_user_stream_size == 0)) vmw_user_stream_size = ttm_round_pot(sizeof(*stream)) + 128; - ret = ttm_read_lock(&vmaster->lock, true); + ret = ttm_read_lock(&dev_priv->reservation_sem, true); if (unlikely(ret != 0)) return ret; @@ -738,7 +935,7 @@ int vmw_stream_claim_ioctl(struct drm_device *dev, void *data, out_err: vmw_resource_unreference(&res); out_unlock: - ttm_read_unlock(&vmaster->lock); + ttm_read_unlock(&dev_priv->reservation_sem); return ret; } @@ -775,53 +972,54 @@ err_ref: } +/** + * vmw_dumb_create - Create a dumb kms buffer + * + * @file_priv: Pointer to a struct drm_file identifying the caller. + * @dev: Pointer to the drm device. + * @args: Pointer to a struct drm_mode_create_dumb structure + * + * This is a driver callback for the core drm create_dumb functionality. + * Note that this is very similar to the vmw_dmabuf_alloc ioctl, except + * that the arguments have a different format. + */ int vmw_dumb_create(struct drm_file *file_priv, struct drm_device *dev, struct drm_mode_create_dumb *args) { struct vmw_private *dev_priv = vmw_priv(dev); - struct vmw_master *vmaster = vmw_master(file_priv->master); - struct vmw_user_dma_buffer *vmw_user_bo; - struct ttm_buffer_object *tmp; + struct vmw_dma_buffer *dma_buf; int ret; args->pitch = args->width * ((args->bpp + 7) / 8); args->size = args->pitch * args->height; - vmw_user_bo = kzalloc(sizeof(*vmw_user_bo), GFP_KERNEL); - if (vmw_user_bo == NULL) - return -ENOMEM; - - ret = ttm_read_lock(&vmaster->lock, true); - if (ret != 0) { - kfree(vmw_user_bo); + ret = ttm_read_lock(&dev_priv->reservation_sem, true); + if (unlikely(ret != 0)) return ret; - } - ret = vmw_dmabuf_init(dev_priv, &vmw_user_bo->dma, args->size, - &vmw_vram_sys_placement, true, - &vmw_user_dmabuf_destroy); - if (ret != 0) - goto out_no_dmabuf; - - tmp = ttm_bo_reference(&vmw_user_bo->dma.base); - ret = ttm_base_object_init(vmw_fpriv(file_priv)->tfile, - &vmw_user_bo->base, - false, - ttm_buffer_type, - &vmw_user_dmabuf_release, NULL); + ret = vmw_user_dmabuf_alloc(dev_priv, vmw_fpriv(file_priv)->tfile, + args->size, false, &args->handle, + &dma_buf); if (unlikely(ret != 0)) - goto out_no_base_object; - - args->handle = vmw_user_bo->base.hash.key; + goto out_no_dmabuf; -out_no_base_object: - ttm_bo_unref(&tmp); + vmw_dmabuf_unreference(&dma_buf); out_no_dmabuf: - ttm_read_unlock(&vmaster->lock); + ttm_read_unlock(&dev_priv->reservation_sem); return ret; } +/** + * vmw_dumb_map_offset - Return the address space offset of a dumb buffer + * + * @file_priv: Pointer to a struct drm_file identifying the caller. + * @dev: Pointer to the drm device. + * @handle: Handle identifying the dumb buffer. + * @offset: The address space offset returned. + * + * This is a driver callback for the core drm dumb_map_offset functionality. + */ int vmw_dumb_map_offset(struct drm_file *file_priv, struct drm_device *dev, uint32_t handle, uint64_t *offset) @@ -839,6 +1037,15 @@ int vmw_dumb_map_offset(struct drm_file *file_priv, return 0; } +/** + * vmw_dumb_destroy - Destroy a dumb boffer + * + * @file_priv: Pointer to a struct drm_file identifying the caller. + * @dev: Pointer to the drm device. + * @handle: Handle identifying the dumb buffer. + * + * This is a driver callback for the core drm dumb_destroy functionality. + */ int vmw_dumb_destroy(struct drm_file *file_priv, struct drm_device *dev, uint32_t handle) @@ -970,7 +1177,7 @@ void vmw_resource_unreserve(struct vmw_resource *res, if (new_backup) res->backup_offset = new_backup_offset; - if (!res->func->may_evict) + if (!res->func->may_evict || res->id == -1) return; write_lock(&dev_priv->resource_lock); @@ -992,7 +1199,6 @@ void vmw_resource_unreserve(struct vmw_resource *res, */ static int vmw_resource_check_buffer(struct vmw_resource *res, - struct ww_acquire_ctx *ticket, bool interruptible, struct ttm_validate_buffer *val_buf) { @@ -1009,7 +1215,7 @@ vmw_resource_check_buffer(struct vmw_resource *res, INIT_LIST_HEAD(&val_list); val_buf->bo = ttm_bo_reference(&res->backup->base); list_add_tail(&val_buf->head, &val_list); - ret = ttm_eu_reserve_buffers(ticket, &val_list); + ret = ttm_eu_reserve_buffers(NULL, &val_list); if (unlikely(ret != 0)) goto out_no_reserve; @@ -1027,7 +1233,7 @@ vmw_resource_check_buffer(struct vmw_resource *res, return 0; out_no_validate: - ttm_eu_backoff_reservation(ticket, &val_list); + ttm_eu_backoff_reservation(NULL, &val_list); out_no_reserve: ttm_bo_unref(&val_buf->bo); if (backup_dirty) @@ -1072,8 +1278,7 @@ int vmw_resource_reserve(struct vmw_resource *res, bool no_backup) * @val_buf: Backup buffer information. */ static void -vmw_resource_backoff_reservation(struct ww_acquire_ctx *ticket, - struct ttm_validate_buffer *val_buf) +vmw_resource_backoff_reservation(struct ttm_validate_buffer *val_buf) { struct list_head val_list; @@ -1082,7 +1287,7 @@ vmw_resource_backoff_reservation(struct ww_acquire_ctx *ticket, INIT_LIST_HEAD(&val_list); list_add_tail(&val_buf->head, &val_list); - ttm_eu_backoff_reservation(ticket, &val_list); + ttm_eu_backoff_reservation(NULL, &val_list); ttm_bo_unref(&val_buf->bo); } @@ -1091,18 +1296,18 @@ vmw_resource_backoff_reservation(struct ww_acquire_ctx *ticket, * to a backup buffer. * * @res: The resource to evict. + * @interruptible: Whether to wait interruptible. */ -int vmw_resource_do_evict(struct vmw_resource *res) +int vmw_resource_do_evict(struct vmw_resource *res, bool interruptible) { struct ttm_validate_buffer val_buf; const struct vmw_res_func *func = res->func; - struct ww_acquire_ctx ticket; int ret; BUG_ON(!func->may_evict); val_buf.bo = NULL; - ret = vmw_resource_check_buffer(res, &ticket, true, &val_buf); + ret = vmw_resource_check_buffer(res, interruptible, &val_buf); if (unlikely(ret != 0)) return ret; @@ -1117,7 +1322,7 @@ int vmw_resource_do_evict(struct vmw_resource *res) res->backup_dirty = true; res->res_dirty = false; out_no_unbind: - vmw_resource_backoff_reservation(&ticket, &val_buf); + vmw_resource_backoff_reservation(&val_buf); return ret; } @@ -1141,6 +1346,7 @@ int vmw_resource_validate(struct vmw_resource *res) struct vmw_private *dev_priv = res->dev_priv; struct list_head *lru_list = &dev_priv->res_lru[res->func->res_type]; struct ttm_validate_buffer val_buf; + unsigned err_count = 0; if (likely(!res->func->may_evict)) return 0; @@ -1155,7 +1361,7 @@ int vmw_resource_validate(struct vmw_resource *res) write_lock(&dev_priv->resource_lock); if (list_empty(lru_list) || !res->func->may_evict) { - DRM_ERROR("Out of device device id entries " + DRM_ERROR("Out of device device resources " "for %s.\n", res->func->type_name); ret = -EBUSY; write_unlock(&dev_priv->resource_lock); @@ -1168,7 +1374,19 @@ int vmw_resource_validate(struct vmw_resource *res) list_del_init(&evict_res->lru_head); write_unlock(&dev_priv->resource_lock); - vmw_resource_do_evict(evict_res); + + ret = vmw_resource_do_evict(evict_res, true); + if (unlikely(ret != 0)) { + write_lock(&dev_priv->resource_lock); + list_add_tail(&evict_res->lru_head, lru_list); + write_unlock(&dev_priv->resource_lock); + if (ret == -ERESTARTSYS || + ++err_count > VMW_RES_EVICT_ERR_COUNT) { + vmw_resource_unreference(&evict_res); + goto out_no_validate; + } + } + vmw_resource_unreference(&evict_res); } while (1); @@ -1229,11 +1447,54 @@ 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); + } } /** @@ -1253,13 +1514,15 @@ bool vmw_resource_needs_backup(const struct vmw_resource *res) * @type: The resource type to evict * * To avoid thrashing starvation or as part of the hibernation sequence, - * evict all evictable resources of a specific type. + * try to evict all evictable resources of a specific type. */ static void vmw_resource_evict_type(struct vmw_private *dev_priv, enum vmw_res_type type) { struct list_head *lru_list = &dev_priv->res_lru[type]; struct vmw_resource *evict_res; + unsigned err_count = 0; + int ret; do { write_lock(&dev_priv->resource_lock); @@ -1272,7 +1535,18 @@ static void vmw_resource_evict_type(struct vmw_private *dev_priv, lru_head)); list_del_init(&evict_res->lru_head); write_unlock(&dev_priv->resource_lock); - vmw_resource_do_evict(evict_res); + + ret = vmw_resource_do_evict(evict_res, false); + if (unlikely(ret != 0)) { + write_lock(&dev_priv->resource_lock); + list_add_tail(&evict_res->lru_head, lru_list); + write_unlock(&dev_priv->resource_lock); + if (++err_count > VMW_RES_EVICT_ERR_COUNT) { + vmw_resource_unreference(&evict_res); + return; + } + } + vmw_resource_unreference(&evict_res); } while (1); diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c b/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c index 26387c3d5a2..a95d3a0cabe 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c @@ -307,9 +307,10 @@ static int vmw_sou_crtc_set_config(struct drm_mode_set *set) connector->encoder = NULL; encoder->crtc = NULL; - crtc->fb = NULL; + crtc->primary->fb = NULL; crtc->x = 0; crtc->y = 0; + crtc->enabled = false; vmw_sou_del_active(dev_priv, sou); @@ -367,9 +368,10 @@ static int vmw_sou_crtc_set_config(struct drm_mode_set *set) connector->encoder = NULL; encoder->crtc = NULL; - crtc->fb = NULL; + crtc->primary->fb = NULL; crtc->x = 0; crtc->y = 0; + crtc->enabled = false; return ret; } @@ -379,9 +381,10 @@ static int vmw_sou_crtc_set_config(struct drm_mode_set *set) connector->encoder = encoder; encoder->crtc = crtc; crtc->mode = *mode; - crtc->fb = fb; + crtc->primary->fb = fb; crtc->x = set->x; crtc->y = set->y; + crtc->enabled = true; return 0; } @@ -464,6 +467,8 @@ static int vmw_sou_init(struct vmw_private *dev_priv, unsigned unit) encoder->possible_crtcs = (1 << unit); encoder->possible_clones = 0; + (void) drm_sysfs_connector_add(connector); + drm_crtc_init(dev, crtc, &vmw_screen_object_crtc_funcs); drm_mode_crtc_set_gamma_size(crtc, 256); @@ -567,5 +572,5 @@ void vmw_kms_screen_object_update_implicit_fb(struct vmw_private *dev_priv, BUG_ON(!sou->base.is_implicit); dev_priv->sou_priv->implicit_fb = - vmw_framebuffer_to_vfb(sou->base.crtc.fb); + vmw_framebuffer_to_vfb(sou->base.crtc.primary->fb); } diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c b/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c new file mode 100644 index 00000000000..c1559eeaffe --- /dev/null +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c @@ -0,0 +1,811 @@ +/************************************************************************** + * + * 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" + +#define VMW_COMPAT_SHADER_HT_ORDER 12 + +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; +}; + +/** + * enum vmw_compat_shader_state - Staging state for compat shaders + */ +enum vmw_compat_shader_state { + VMW_COMPAT_COMMITED, + VMW_COMPAT_ADD, + VMW_COMPAT_DEL +}; + +/** + * struct vmw_compat_shader - Metadata for compat shaders. + * + * @handle: The TTM handle of the guest backed shader. + * @tfile: The struct ttm_object_file the guest backed shader is registered + * with. + * @hash: Hash item for lookup. + * @head: List head for staging lists or the compat shader manager list. + * @state: Staging state. + * + * The structure is protected by the cmdbuf lock. + */ +struct vmw_compat_shader { + u32 handle; + struct ttm_object_file *tfile; + struct drm_hash_item hash; + struct list_head head; + enum vmw_compat_shader_state state; +}; + +/** + * struct vmw_compat_shader_manager - Compat shader manager. + * + * @shaders: Hash table containing staged and commited compat shaders + * @list: List of commited shaders. + * @dev_priv: Pointer to a device private structure. + * + * @shaders and @list are protected by the cmdbuf mutex for now. + */ +struct vmw_compat_shader_manager { + struct drm_open_hash shaders; + struct list_head list; + struct vmw_private *dev_priv; +}; + +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_scrub(&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"); + mutex_unlock(&dev_priv->binding_mutex); + 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); +} + +static int vmw_shader_alloc(struct vmw_private *dev_priv, + struct vmw_dma_buffer *buffer, + size_t shader_size, + size_t offset, + SVGA3dShaderType shader_type, + struct ttm_object_file *tfile, + u32 *handle) +{ + struct vmw_user_shader *ushader; + struct vmw_resource *res, *tmp; + int ret; + + /* + * 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(struct vmw_user_shader)) + 128; + + 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; + } + + 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; + } + + 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, shader_size, + offset, shader_type, buffer, + vmw_user_shader_free); + if (unlikely(ret != 0)) + goto out; + + 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; + } + + if (handle) + *handle = ushader->base.hash.key; +out_err: + vmw_resource_unreference(&res); +out: + return ret; +} + + +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 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_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; + } + + ret = ttm_read_lock(&dev_priv->reservation_sem, true); + if (unlikely(ret != 0)) + goto out_bad_arg; + + ret = vmw_shader_alloc(dev_priv, buffer, arg->size, arg->offset, + shader_type, tfile, &arg->shader_handle); + + ttm_read_unlock(&dev_priv->reservation_sem); +out_bad_arg: + vmw_dmabuf_unreference(&buffer); + return ret; +} + +/** + * vmw_compat_shader_lookup - Look up a compat shader + * + * @man: Pointer to the compat shader manager. + * @shader_type: The shader type, that combined with the user_key identifies + * the shader. + * @user_key: On entry, this should be a pointer to the user_key. + * On successful exit, it will contain the guest-backed shader's TTM handle. + * + * Returns 0 on success. Non-zero on failure, in which case the value pointed + * to by @user_key is unmodified. + */ +int vmw_compat_shader_lookup(struct vmw_compat_shader_manager *man, + SVGA3dShaderType shader_type, + u32 *user_key) +{ + struct drm_hash_item *hash; + int ret; + unsigned long key = *user_key | (shader_type << 24); + + ret = drm_ht_find_item(&man->shaders, key, &hash); + if (unlikely(ret != 0)) + return ret; + + *user_key = drm_hash_entry(hash, struct vmw_compat_shader, + hash)->handle; + + return 0; +} + +/** + * vmw_compat_shader_free - Free a compat shader. + * + * @man: Pointer to the compat shader manager. + * @entry: Pointer to a struct vmw_compat_shader. + * + * Frees a struct vmw_compat_shder entry and drops its reference to the + * guest backed shader. + */ +static void vmw_compat_shader_free(struct vmw_compat_shader_manager *man, + struct vmw_compat_shader *entry) +{ + list_del(&entry->head); + WARN_ON(drm_ht_remove_item(&man->shaders, &entry->hash)); + WARN_ON(ttm_ref_object_base_unref(entry->tfile, entry->handle, + TTM_REF_USAGE)); + kfree(entry); +} + +/** + * vmw_compat_shaders_commit - Commit a list of compat shader actions. + * + * @man: Pointer to the compat shader manager. + * @list: Caller's list of compat shader actions. + * + * This function commits a list of compat shader additions or removals. + * It is typically called when the execbuf ioctl call triggering these + * actions has commited the fifo contents to the device. + */ +void vmw_compat_shaders_commit(struct vmw_compat_shader_manager *man, + struct list_head *list) +{ + struct vmw_compat_shader *entry, *next; + + list_for_each_entry_safe(entry, next, list, head) { + list_del(&entry->head); + switch (entry->state) { + case VMW_COMPAT_ADD: + entry->state = VMW_COMPAT_COMMITED; + list_add_tail(&entry->head, &man->list); + break; + case VMW_COMPAT_DEL: + ttm_ref_object_base_unref(entry->tfile, entry->handle, + TTM_REF_USAGE); + kfree(entry); + break; + default: + BUG(); + break; + } + } +} + +/** + * vmw_compat_shaders_revert - Revert a list of compat shader actions + * + * @man: Pointer to the compat shader manager. + * @list: Caller's list of compat shader actions. + * + * This function reverts a list of compat shader additions or removals. + * It is typically called when the execbuf ioctl call triggering these + * actions failed for some reason, and the command stream was never + * submitted. + */ +void vmw_compat_shaders_revert(struct vmw_compat_shader_manager *man, + struct list_head *list) +{ + struct vmw_compat_shader *entry, *next; + int ret; + + list_for_each_entry_safe(entry, next, list, head) { + switch (entry->state) { + case VMW_COMPAT_ADD: + vmw_compat_shader_free(man, entry); + break; + case VMW_COMPAT_DEL: + ret = drm_ht_insert_item(&man->shaders, &entry->hash); + list_del(&entry->head); + list_add_tail(&entry->head, &man->list); + entry->state = VMW_COMPAT_COMMITED; + break; + default: + BUG(); + break; + } + } +} + +/** + * vmw_compat_shader_remove - Stage a compat shader for removal. + * + * @man: Pointer to the compat shader manager + * @user_key: The key that is used to identify the shader. The key is + * unique to the shader type. + * @shader_type: Shader type. + * @list: Caller's list of staged shader actions. + * + * This function stages a compat shader for removal and removes the key from + * the shader manager's hash table. If the shader was previously only staged + * for addition it is completely removed (But the execbuf code may keep a + * reference if it was bound to a context between addition and removal). If + * it was previously commited to the manager, it is staged for removal. + */ +int vmw_compat_shader_remove(struct vmw_compat_shader_manager *man, + u32 user_key, SVGA3dShaderType shader_type, + struct list_head *list) +{ + struct vmw_compat_shader *entry; + struct drm_hash_item *hash; + int ret; + + ret = drm_ht_find_item(&man->shaders, user_key | (shader_type << 24), + &hash); + if (likely(ret != 0)) + return -EINVAL; + + entry = drm_hash_entry(hash, struct vmw_compat_shader, hash); + + switch (entry->state) { + case VMW_COMPAT_ADD: + vmw_compat_shader_free(man, entry); + break; + case VMW_COMPAT_COMMITED: + (void) drm_ht_remove_item(&man->shaders, &entry->hash); + list_del(&entry->head); + entry->state = VMW_COMPAT_DEL; + list_add_tail(&entry->head, list); + break; + default: + BUG(); + break; + } + + return 0; +} + +/** + * vmw_compat_shader_add - Create a compat shader and add the + * key to the manager + * + * @man: Pointer to the compat shader manager + * @user_key: The key that is used to identify the shader. The key is + * unique to the shader type. + * @bytecode: Pointer to the bytecode of the shader. + * @shader_type: Shader type. + * @tfile: Pointer to a struct ttm_object_file that the guest-backed shader is + * to be created with. + * @list: Caller's list of staged shader actions. + * + * Note that only the key is added to the shader manager's hash table. + * The shader is not yet added to the shader manager's list of shaders. + */ +int vmw_compat_shader_add(struct vmw_compat_shader_manager *man, + u32 user_key, const void *bytecode, + SVGA3dShaderType shader_type, + size_t size, + struct ttm_object_file *tfile, + struct list_head *list) +{ + struct vmw_dma_buffer *buf; + struct ttm_bo_kmap_obj map; + bool is_iomem; + struct vmw_compat_shader *compat; + u32 handle; + int ret; + + if (user_key > ((1 << 24) - 1) || (unsigned) shader_type > 16) + return -EINVAL; + + /* Allocate and pin a DMA buffer */ + buf = kzalloc(sizeof(*buf), GFP_KERNEL); + if (unlikely(buf == NULL)) + return -ENOMEM; + + ret = vmw_dmabuf_init(man->dev_priv, buf, size, &vmw_sys_ne_placement, + true, vmw_dmabuf_bo_free); + if (unlikely(ret != 0)) + goto out; + + ret = ttm_bo_reserve(&buf->base, false, true, false, NULL); + if (unlikely(ret != 0)) + goto no_reserve; + + /* Map and copy shader bytecode. */ + ret = ttm_bo_kmap(&buf->base, 0, PAGE_ALIGN(size) >> PAGE_SHIFT, + &map); + if (unlikely(ret != 0)) { + ttm_bo_unreserve(&buf->base); + goto no_reserve; + } + + memcpy(ttm_kmap_obj_virtual(&map, &is_iomem), bytecode, size); + WARN_ON(is_iomem); + + ttm_bo_kunmap(&map); + ret = ttm_bo_validate(&buf->base, &vmw_sys_placement, false, true); + WARN_ON(ret != 0); + ttm_bo_unreserve(&buf->base); + + /* Create a guest-backed shader container backed by the dma buffer */ + ret = vmw_shader_alloc(man->dev_priv, buf, size, 0, shader_type, + tfile, &handle); + vmw_dmabuf_unreference(&buf); + if (unlikely(ret != 0)) + goto no_reserve; + /* + * Create a compat shader structure and stage it for insertion + * in the manager + */ + compat = kzalloc(sizeof(*compat), GFP_KERNEL); + if (compat == NULL) + goto no_compat; + + compat->hash.key = user_key | (shader_type << 24); + ret = drm_ht_insert_item(&man->shaders, &compat->hash); + if (unlikely(ret != 0)) + goto out_invalid_key; + + compat->state = VMW_COMPAT_ADD; + compat->handle = handle; + compat->tfile = tfile; + list_add_tail(&compat->head, list); + + return 0; + +out_invalid_key: + kfree(compat); +no_compat: + ttm_ref_object_base_unref(tfile, handle, TTM_REF_USAGE); +no_reserve: +out: + return ret; +} + +/** + * vmw_compat_shader_man_create - Create a compat shader manager + * + * @dev_priv: Pointer to a device private structure. + * + * Typically done at file open time. If successful returns a pointer to a + * compat shader manager. Otherwise returns an error pointer. + */ +struct vmw_compat_shader_manager * +vmw_compat_shader_man_create(struct vmw_private *dev_priv) +{ + struct vmw_compat_shader_manager *man; + int ret; + + man = kzalloc(sizeof(*man), GFP_KERNEL); + if (man == NULL) + return ERR_PTR(-ENOMEM); + + man->dev_priv = dev_priv; + INIT_LIST_HEAD(&man->list); + ret = drm_ht_create(&man->shaders, VMW_COMPAT_SHADER_HT_ORDER); + if (ret == 0) + return man; + + kfree(man); + return ERR_PTR(ret); +} + +/** + * vmw_compat_shader_man_destroy - Destroy a compat shader manager + * + * @man: Pointer to the shader manager to destroy. + * + * Typically done at file close time. + */ +void vmw_compat_shader_man_destroy(struct vmw_compat_shader_manager *man) +{ + struct vmw_compat_shader *entry, *next; + + mutex_lock(&man->dev_priv->cmdbuf_mutex); + list_for_each_entry_safe(entry, next, &man->list, head) + vmw_compat_shader_free(man, entry); + + mutex_unlock(&man->dev_priv->cmdbuf_mutex); + kfree(man); +} diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c b/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c index 58281433974..4ecdbf3e59d 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c @@ -36,12 +36,13 @@ * @base: The TTM base object handling user-space visibility. * @srf: The surface metadata. * @size: TTM accounting size for the surface. + * @master: master of the creating client. Used for security check. */ struct vmw_user_surface { - struct ttm_base_object base; + struct ttm_prime_object prime; struct vmw_surface srf; uint32_t size; - uint32_t backup_handle; + struct drm_master *master; }; /** @@ -68,6 +69,14 @@ 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 +102,18 @@ 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 +312,11 @@ 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 +575,15 @@ 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; } @@ -580,7 +609,8 @@ static int vmw_surface_init(struct vmw_private *dev_priv, static struct vmw_resource * vmw_user_surface_base_to_res(struct ttm_base_object *base) { - return &(container_of(base, struct vmw_user_surface, base)->srf.res); + return &(container_of(base, struct vmw_user_surface, + prime.base)->srf.res); } /** @@ -596,10 +626,12 @@ static void vmw_user_surface_free(struct vmw_resource *res) struct vmw_private *dev_priv = srf->res.dev_priv; uint32_t size = user_srf->size; + if (user_srf->master) + drm_master_put(&user_srf->master); kfree(srf->offsets); kfree(srf->sizes); kfree(srf->snooper.image); - ttm_base_object_kfree(user_srf, base); + ttm_prime_object_kfree(user_srf, prime); ttm_mem_global_free(vmw_mem_glob(dev_priv), size); } @@ -616,7 +648,7 @@ static void vmw_user_surface_base_release(struct ttm_base_object **p_base) { struct ttm_base_object *base = *p_base; struct vmw_user_surface *user_srf = - container_of(base, struct vmw_user_surface, base); + container_of(base, struct vmw_user_surface, prime.base); struct vmw_resource *res = &user_srf->srf.res; *p_base = NULL; @@ -669,7 +701,6 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data, struct vmw_surface_offset *cur_offset; uint32_t num_sizes; uint32_t size; - struct vmw_master *vmaster = vmw_master(file_priv->master); const struct svga3d_surface_desc *desc; if (unlikely(vmw_user_surface_size == 0)) @@ -695,7 +726,7 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data, return -EINVAL; } - ret = ttm_read_lock(&vmaster->lock, true); + ret = ttm_read_lock(&dev_priv->reservation_sem, true); if (unlikely(ret != 0)) return ret; @@ -749,7 +780,7 @@ 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; @@ -790,8 +821,10 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data, } srf->snooper.crtc = NULL; - user_srf->base.shareable = false; - user_srf->base.tfile = NULL; + user_srf->prime.base.shareable = false; + user_srf->prime.base.tfile = NULL; + if (drm_is_primary_client(file_priv)) + user_srf->master = drm_master_get(file_priv->master); /** * From this point, the generic resource management functions @@ -802,10 +835,28 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data, if (unlikely(ret != 0)) goto out_unlock; + /* + * A gb-aware client referencing a shared surface will + * expect a backup buffer to be present. + */ + if (dev_priv->has_mob && req->shareable) { + uint32_t backup_handle; + + ret = vmw_user_dmabuf_alloc(dev_priv, tfile, + res->backup_size, + true, + &backup_handle, + &res->backup); + if (unlikely(ret != 0)) { + vmw_resource_unreference(&res); + goto out_unlock; + } + } + tmp = vmw_resource_reference(&srf->res); - ret = ttm_base_object_init(tfile, &user_srf->base, - req->shareable, VMW_RES_SURFACE, - &vmw_user_surface_base_release, NULL); + ret = ttm_prime_object_init(tfile, res->backup_size, &user_srf->prime, + req->shareable, VMW_RES_SURFACE, + &vmw_user_surface_base_release, NULL); if (unlikely(ret != 0)) { vmw_resource_unreference(&tmp); @@ -813,21 +864,95 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data, goto out_unlock; } - rep->sid = user_srf->base.hash.key; + rep->sid = user_srf->prime.base.hash.key; vmw_resource_unreference(&res); - ttm_read_unlock(&vmaster->lock); + ttm_read_unlock(&dev_priv->reservation_sem); return 0; out_no_copy: kfree(srf->offsets); out_no_offsets: kfree(srf->sizes); out_no_sizes: - ttm_base_object_kfree(user_srf, base); + ttm_prime_object_kfree(user_srf, prime); out_no_user_srf: ttm_mem_global_free(vmw_mem_glob(dev_priv), size); out_unlock: - ttm_read_unlock(&vmaster->lock); + ttm_read_unlock(&dev_priv->reservation_sem); + return ret; +} + + +static int +vmw_surface_handle_reference(struct vmw_private *dev_priv, + struct drm_file *file_priv, + uint32_t u_handle, + enum drm_vmw_handle_type handle_type, + struct ttm_base_object **base_p) +{ + struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile; + struct vmw_user_surface *user_srf; + uint32_t handle; + struct ttm_base_object *base; + int ret; + + if (handle_type == DRM_VMW_HANDLE_PRIME) { + ret = ttm_prime_fd_to_handle(tfile, u_handle, &handle); + if (unlikely(ret != 0)) + return ret; + } else { + if (unlikely(drm_is_render_client(file_priv))) { + DRM_ERROR("Render client refused legacy " + "surface reference.\n"); + return -EACCES; + } + handle = u_handle; + } + + ret = -EINVAL; + base = ttm_base_object_lookup_for_ref(dev_priv->tdev, handle); + if (unlikely(base == NULL)) { + DRM_ERROR("Could not find surface to reference.\n"); + goto out_no_lookup; + } + + if (unlikely(ttm_base_object_type(base) != VMW_RES_SURFACE)) { + DRM_ERROR("Referenced object is not a surface.\n"); + goto out_bad_resource; + } + + if (handle_type != DRM_VMW_HANDLE_PRIME) { + user_srf = container_of(base, struct vmw_user_surface, + prime.base); + + /* + * Make sure the surface creator has the same + * authenticating master. + */ + if (drm_is_primary_client(file_priv) && + user_srf->master != file_priv->master) { + DRM_ERROR("Trying to reference surface outside of" + " master domain.\n"); + ret = -EACCES; + goto out_bad_resource; + } + + ret = ttm_ref_object_add(tfile, base, TTM_REF_USAGE, NULL); + if (unlikely(ret != 0)) { + DRM_ERROR("Could not add a reference to a surface.\n"); + goto out_bad_resource; + } + } + + *base_p = base; + return 0; + +out_bad_resource: + ttm_base_object_unref(&base); +out_no_lookup: + if (handle_type == DRM_VMW_HANDLE_PRIME) + (void) ttm_ref_object_base_unref(tfile, handle, TTM_REF_USAGE); + return ret; } @@ -842,6 +967,7 @@ out_unlock: int vmw_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_surface_reference_arg *arg = (union drm_vmw_surface_reference_arg *)data; struct drm_vmw_surface_arg *req = &arg->req; @@ -851,26 +977,16 @@ int vmw_surface_reference_ioctl(struct drm_device *dev, void *data, struct vmw_user_surface *user_srf; struct drm_vmw_size __user *user_sizes; struct ttm_base_object *base; - int ret = -EINVAL; - - base = ttm_base_object_lookup(tfile, req->sid); - if (unlikely(base == NULL)) { - DRM_ERROR("Could not find surface to reference.\n"); - return -EINVAL; - } + int ret; - if (unlikely(base->object_type != VMW_RES_SURFACE)) - goto out_bad_resource; + ret = vmw_surface_handle_reference(dev_priv, file_priv, req->sid, + req->handle_type, &base); + if (unlikely(ret != 0)) + return ret; - user_srf = container_of(base, struct vmw_user_surface, base); + user_srf = container_of(base, struct vmw_user_surface, prime.base); srf = &user_srf->srf; - ret = ttm_ref_object_add(tfile, &user_srf->base, TTM_REF_USAGE, NULL); - if (unlikely(ret != 0)) { - DRM_ERROR("Could not add a reference to a surface.\n"); - goto out_no_reference; - } - rep->flags = srf->flags; rep->format = srf->format; memcpy(rep->mip_levels, srf->mip_levels, sizeof(srf->mip_levels)); @@ -878,15 +994,437 @@ int vmw_surface_reference_ioctl(struct drm_device *dev, void *data, rep->size_addr; if (user_sizes) - ret = copy_to_user(user_sizes, srf->sizes, - srf->num_sizes * sizeof(*srf->sizes)); + ret = copy_to_user(user_sizes, &srf->base_size, + sizeof(srf->base_size)); if (unlikely(ret != 0)) { DRM_ERROR("copy_to_user failed %p %u\n", user_sizes, srf->num_sizes); + ttm_ref_object_base_unref(tfile, base->hash.key, TTM_REF_USAGE); ret = -EFAULT; } + + ttm_base_object_unref(&base); + + 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_scrub(&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"); + mutex_unlock(&dev_priv->binding_mutex); + 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; + 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(&dev_priv->reservation_sem, 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; + if (drm_is_primary_client(file_priv)) + user_srf->master = drm_master_get(file_priv->master); + + /** + * 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(&dev_priv->reservation_sem); + return 0; +out_no_user_srf: + ttm_mem_global_free(vmw_mem_glob(dev_priv), size); +out_unlock: + ttm_read_unlock(&dev_priv->reservation_sem); + 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; + + ret = vmw_surface_handle_reference(dev_priv, file_priv, req->sid, + req->handle_type, &base); + if (unlikely(ret != 0)) + return ret; + + 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; + } + + 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(tfile, base->hash.key, + 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: -out_no_reference: ttm_base_object_unref(&base); return ret; |