diff options
Diffstat (limited to 'drivers/gpu/drm/vmwgfx')
23 files changed, 5746 insertions, 1532 deletions
diff --git a/drivers/gpu/drm/vmwgfx/Kconfig b/drivers/gpu/drm/vmwgfx/Kconfig index 30ad133..794ff67 100644 --- a/drivers/gpu/drm/vmwgfx/Kconfig +++ b/drivers/gpu/drm/vmwgfx/Kconfig @@ -7,7 +7,8 @@ config DRM_VMWGFX select FB_CFB_IMAGEBLIT select DRM_TTM help - KMS enabled DRM driver for SVGA2 virtual hardware. - - If unsure say n. The compiled module will be - called vmwgfx.ko + Choose this option if you would like to run 3D acceleration + in a VMware virtual machine. + This is a KMS enabled DRM driver for the VMware SVGA2 + virtual hardware. + The compiled module will be called "vmwgfx.ko". diff --git a/drivers/gpu/drm/vmwgfx/Makefile b/drivers/gpu/drm/vmwgfx/Makefile index c9281a1..586869c 100644 --- a/drivers/gpu/drm/vmwgfx/Makefile +++ b/drivers/gpu/drm/vmwgfx/Makefile @@ -4,6 +4,7 @@ ccflags-y := -Iinclude/drm vmwgfx-y := vmwgfx_execbuf.o vmwgfx_gmr.o vmwgfx_kms.o vmwgfx_drv.o \ vmwgfx_fb.o vmwgfx_ioctl.o vmwgfx_resource.o vmwgfx_buffer.o \ vmwgfx_fifo.o vmwgfx_irq.o vmwgfx_ldu.o vmwgfx_ttm_glue.o \ - vmwgfx_overlay.o vmwgfx_fence.o vmwgfx_gmrid_manager.o + vmwgfx_overlay.o vmwgfx_marker.o vmwgfx_gmrid_manager.o \ + vmwgfx_fence.o vmwgfx_dmabuf.o vmwgfx_scrn.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 77cb453..d0e085e 100644 --- a/drivers/gpu/drm/vmwgfx/svga3d_reg.h +++ b/drivers/gpu/drm/vmwgfx/svga3d_reg.h @@ -57,7 +57,8 @@ typedef enum { SVGA3D_HWVERSION_WS6_B1 = SVGA3D_MAKE_HWVERSION(1, 1), SVGA3D_HWVERSION_FUSION_11 = SVGA3D_MAKE_HWVERSION(1, 4), SVGA3D_HWVERSION_WS65_B1 = SVGA3D_MAKE_HWVERSION(2, 0), - SVGA3D_HWVERSION_CURRENT = SVGA3D_HWVERSION_WS65_B1, + SVGA3D_HWVERSION_WS8_B1 = SVGA3D_MAKE_HWVERSION(2, 1), + SVGA3D_HWVERSION_CURRENT = SVGA3D_HWVERSION_WS8_B1, } SVGA3dHardwareVersion; /* @@ -67,7 +68,8 @@ typedef enum { typedef uint32 SVGA3dBool; /* 32-bit Bool definition */ #define SVGA3D_NUM_CLIPPLANES 6 #define SVGA3D_MAX_SIMULTANEOUS_RENDER_TARGETS 8 - +#define SVGA3D_MAX_CONTEXT_IDS 256 +#define SVGA3D_MAX_SURFACE_IDS (32 * 1024) /* * Surface formats. @@ -79,76 +81,91 @@ typedef uint32 SVGA3dBool; /* 32-bit Bool definition */ */ typedef enum SVGA3dSurfaceFormat { - SVGA3D_FORMAT_INVALID = 0, + SVGA3D_FORMAT_INVALID = 0, - SVGA3D_X8R8G8B8 = 1, - SVGA3D_A8R8G8B8 = 2, + SVGA3D_X8R8G8B8 = 1, + SVGA3D_A8R8G8B8 = 2, - SVGA3D_R5G6B5 = 3, - SVGA3D_X1R5G5B5 = 4, - SVGA3D_A1R5G5B5 = 5, - SVGA3D_A4R4G4B4 = 6, + SVGA3D_R5G6B5 = 3, + SVGA3D_X1R5G5B5 = 4, + SVGA3D_A1R5G5B5 = 5, + SVGA3D_A4R4G4B4 = 6, - SVGA3D_Z_D32 = 7, - SVGA3D_Z_D16 = 8, - SVGA3D_Z_D24S8 = 9, - SVGA3D_Z_D15S1 = 10, + SVGA3D_Z_D32 = 7, + SVGA3D_Z_D16 = 8, + SVGA3D_Z_D24S8 = 9, + SVGA3D_Z_D15S1 = 10, - SVGA3D_LUMINANCE8 = 11, - SVGA3D_LUMINANCE4_ALPHA4 = 12, - SVGA3D_LUMINANCE16 = 13, - SVGA3D_LUMINANCE8_ALPHA8 = 14, + SVGA3D_LUMINANCE8 = 11, + SVGA3D_LUMINANCE4_ALPHA4 = 12, + SVGA3D_LUMINANCE16 = 13, + SVGA3D_LUMINANCE8_ALPHA8 = 14, - SVGA3D_DXT1 = 15, - SVGA3D_DXT2 = 16, - SVGA3D_DXT3 = 17, - SVGA3D_DXT4 = 18, - SVGA3D_DXT5 = 19, + SVGA3D_DXT1 = 15, + SVGA3D_DXT2 = 16, + SVGA3D_DXT3 = 17, + SVGA3D_DXT4 = 18, + SVGA3D_DXT5 = 19, - SVGA3D_BUMPU8V8 = 20, - SVGA3D_BUMPL6V5U5 = 21, - SVGA3D_BUMPX8L8V8U8 = 22, - SVGA3D_BUMPL8V8U8 = 23, + SVGA3D_BUMPU8V8 = 20, + SVGA3D_BUMPL6V5U5 = 21, + SVGA3D_BUMPX8L8V8U8 = 22, + SVGA3D_BUMPL8V8U8 = 23, - SVGA3D_ARGB_S10E5 = 24, /* 16-bit floating-point ARGB */ - SVGA3D_ARGB_S23E8 = 25, /* 32-bit floating-point ARGB */ + SVGA3D_ARGB_S10E5 = 24, /* 16-bit floating-point ARGB */ + SVGA3D_ARGB_S23E8 = 25, /* 32-bit floating-point ARGB */ - SVGA3D_A2R10G10B10 = 26, + SVGA3D_A2R10G10B10 = 26, /* signed formats */ - SVGA3D_V8U8 = 27, - SVGA3D_Q8W8V8U8 = 28, - SVGA3D_CxV8U8 = 29, + SVGA3D_V8U8 = 27, + SVGA3D_Q8W8V8U8 = 28, + SVGA3D_CxV8U8 = 29, /* mixed formats */ - SVGA3D_X8L8V8U8 = 30, - SVGA3D_A2W10V10U10 = 31, + SVGA3D_X8L8V8U8 = 30, + SVGA3D_A2W10V10U10 = 31, - SVGA3D_ALPHA8 = 32, + SVGA3D_ALPHA8 = 32, /* Single- and dual-component floating point formats */ - SVGA3D_R_S10E5 = 33, - SVGA3D_R_S23E8 = 34, - SVGA3D_RG_S10E5 = 35, - SVGA3D_RG_S23E8 = 36, + SVGA3D_R_S10E5 = 33, + SVGA3D_R_S23E8 = 34, + 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, + SVGA3D_BUFFER = 37, + + SVGA3D_Z_D24X8 = 38, - SVGA3D_V16U16 = 39, + SVGA3D_V16U16 = 39, - SVGA3D_G16R16 = 40, - SVGA3D_A16B16G16R16 = 41, + SVGA3D_G16R16 = 40, + SVGA3D_A16B16G16R16 = 41, /* Packed Video formats */ - SVGA3D_UYVY = 42, - SVGA3D_YUY2 = 43, + SVGA3D_UYVY = 42, + SVGA3D_YUY2 = 43, + + /* Planar video formats */ + SVGA3D_NV12 = 44, + + /* Video format with alpha */ + SVGA3D_AYUV = 45, + + SVGA3D_BC4_UNORM = 108, + SVGA3D_BC5_UNORM = 111, + + /* Advanced D3D9 depth formats. */ + SVGA3D_Z_DF16 = 118, + SVGA3D_Z_DF24 = 119, + SVGA3D_Z_D24S8_INT = 120, SVGA3D_FORMAT_MAX } SVGA3dSurfaceFormat; @@ -414,10 +431,20 @@ typedef enum { SVGA3D_RS_SRCBLENDALPHA = 94, /* SVGA3dBlendOp */ SVGA3D_RS_DSTBLENDALPHA = 95, /* SVGA3dBlendOp */ SVGA3D_RS_BLENDEQUATIONALPHA = 96, /* SVGA3dBlendEquation */ + SVGA3D_RS_TRANSPARENCYANTIALIAS = 97, /* SVGA3dTransparencyAntialiasType */ + SVGA3D_RS_LINEAA = 98, /* SVGA3dBool */ + SVGA3D_RS_LINEWIDTH = 99, /* float */ SVGA3D_RS_MAX } SVGA3dRenderStateName; typedef enum { + SVGA3D_TRANSPARENCYANTIALIAS_NORMAL = 0, + SVGA3D_TRANSPARENCYANTIALIAS_ALPHATOCOVERAGE = 1, + SVGA3D_TRANSPARENCYANTIALIAS_SUPERSAMPLE = 2, + SVGA3D_TRANSPARENCYANTIALIAS_MAX +} SVGA3dTransparencyAntialiasType; + +typedef enum { SVGA3D_VERTEXMATERIAL_NONE = 0, /* Use the value in the current material */ SVGA3D_VERTEXMATERIAL_DIFFUSE = 1, /* Use the value in the diffuse component */ SVGA3D_VERTEXMATERIAL_SPECULAR = 2, /* Use the value in the specular component */ @@ -728,10 +755,10 @@ typedef enum { SVGA3D_TEX_FILTER_NEAREST = 1, SVGA3D_TEX_FILTER_LINEAR = 2, SVGA3D_TEX_FILTER_ANISOTROPIC = 3, - SVGA3D_TEX_FILTER_FLATCUBIC = 4, // Deprecated, not implemented - SVGA3D_TEX_FILTER_GAUSSIANCUBIC = 5, // Deprecated, not implemented - SVGA3D_TEX_FILTER_PYRAMIDALQUAD = 6, // Not currently implemented - SVGA3D_TEX_FILTER_GAUSSIANQUAD = 7, // Not currently implemented + SVGA3D_TEX_FILTER_FLATCUBIC = 4, /* Deprecated, not implemented */ + SVGA3D_TEX_FILTER_GAUSSIANCUBIC = 5, /* Deprecated, not implemented */ + SVGA3D_TEX_FILTER_PYRAMIDALQUAD = 6, /* Not currently implemented */ + SVGA3D_TEX_FILTER_GAUSSIANQUAD = 7, /* Not currently implemented */ SVGA3D_TEX_FILTER_MAX } SVGA3dTextureFilter; @@ -799,19 +826,19 @@ typedef enum { typedef enum { SVGA3D_DECLUSAGE_POSITION = 0, - SVGA3D_DECLUSAGE_BLENDWEIGHT, // 1 - SVGA3D_DECLUSAGE_BLENDINDICES, // 2 - SVGA3D_DECLUSAGE_NORMAL, // 3 - SVGA3D_DECLUSAGE_PSIZE, // 4 - SVGA3D_DECLUSAGE_TEXCOORD, // 5 - SVGA3D_DECLUSAGE_TANGENT, // 6 - SVGA3D_DECLUSAGE_BINORMAL, // 7 - SVGA3D_DECLUSAGE_TESSFACTOR, // 8 - SVGA3D_DECLUSAGE_POSITIONT, // 9 - SVGA3D_DECLUSAGE_COLOR, // 10 - SVGA3D_DECLUSAGE_FOG, // 11 - SVGA3D_DECLUSAGE_DEPTH, // 12 - SVGA3D_DECLUSAGE_SAMPLE, // 13 + SVGA3D_DECLUSAGE_BLENDWEIGHT, /* 1 */ + SVGA3D_DECLUSAGE_BLENDINDICES, /* 2 */ + SVGA3D_DECLUSAGE_NORMAL, /* 3 */ + SVGA3D_DECLUSAGE_PSIZE, /* 4 */ + SVGA3D_DECLUSAGE_TEXCOORD, /* 5 */ + SVGA3D_DECLUSAGE_TANGENT, /* 6 */ + SVGA3D_DECLUSAGE_BINORMAL, /* 7 */ + SVGA3D_DECLUSAGE_TESSFACTOR, /* 8 */ + SVGA3D_DECLUSAGE_POSITIONT, /* 9 */ + SVGA3D_DECLUSAGE_COLOR, /* 10 */ + SVGA3D_DECLUSAGE_FOG, /* 11 */ + SVGA3D_DECLUSAGE_DEPTH, /* 12 */ + SVGA3D_DECLUSAGE_SAMPLE, /* 13 */ SVGA3D_DECLUSAGE_MAX } SVGA3dDeclUsage; @@ -819,10 +846,10 @@ typedef enum { SVGA3D_DECLMETHOD_DEFAULT = 0, SVGA3D_DECLMETHOD_PARTIALU, SVGA3D_DECLMETHOD_PARTIALV, - SVGA3D_DECLMETHOD_CROSSUV, // Normal + SVGA3D_DECLMETHOD_CROSSUV, /* Normal */ SVGA3D_DECLMETHOD_UV, - SVGA3D_DECLMETHOD_LOOKUP, // Lookup a displacement map - SVGA3D_DECLMETHOD_LOOKUPPRESAMPLED, // Lookup a pre-sampled displacement map + SVGA3D_DECLMETHOD_LOOKUP, /* Lookup a displacement map */ + SVGA3D_DECLMETHOD_LOOKUPPRESAMPLED, /* Lookup a pre-sampled displacement map */ } SVGA3dDeclMethod; typedef enum { @@ -930,7 +957,6 @@ typedef enum { } SVGA3dCubeFace; typedef enum { - SVGA3D_SHADERTYPE_COMPILED_DX8 = 0, SVGA3D_SHADERTYPE_VS = 1, SVGA3D_SHADERTYPE_PS = 2, SVGA3D_SHADERTYPE_MAX @@ -968,12 +994,18 @@ typedef enum { } SVGA3dTransferType; /* - * The maximum number vertex arrays we're guaranteed to support in + * The maximum number of vertex arrays we're guaranteed to support in * SVGA_3D_CMD_DRAWPRIMITIVES. */ #define SVGA3D_MAX_VERTEX_ARRAYS 32 /* + * The maximum number of primitive ranges we're guaranteed to support + * in SVGA_3D_CMD_DRAWPRIMITIVES. + */ +#define SVGA3D_MAX_DRAW_PRIMITIVE_RANGES 32 + +/* * Identifiers for commands in the command FIFO. * * IDs between 1000 and 1039 (inclusive) were used by obsolete versions of @@ -990,7 +1022,7 @@ typedef enum { #define SVGA_3D_CMD_LEGACY_BASE 1000 #define SVGA_3D_CMD_BASE 1040 -#define SVGA_3D_CMD_SURFACE_DEFINE SVGA_3D_CMD_BASE + 0 +#define SVGA_3D_CMD_SURFACE_DEFINE SVGA_3D_CMD_BASE + 0 /* Deprecated */ #define SVGA_3D_CMD_SURFACE_DESTROY SVGA_3D_CMD_BASE + 1 #define SVGA_3D_CMD_SURFACE_COPY SVGA_3D_CMD_BASE + 2 #define SVGA_3D_CMD_SURFACE_STRETCHBLT SVGA_3D_CMD_BASE + 3 @@ -1008,7 +1040,7 @@ typedef enum { #define SVGA_3D_CMD_SETVIEWPORT SVGA_3D_CMD_BASE + 15 #define SVGA_3D_CMD_SETCLIPPLANE SVGA_3D_CMD_BASE + 16 #define SVGA_3D_CMD_CLEAR SVGA_3D_CMD_BASE + 17 -#define SVGA_3D_CMD_PRESENT SVGA_3D_CMD_BASE + 18 // Deprecated +#define SVGA_3D_CMD_PRESENT SVGA_3D_CMD_BASE + 18 /* Deprecated */ #define SVGA_3D_CMD_SHADER_DEFINE SVGA_3D_CMD_BASE + 19 #define SVGA_3D_CMD_SHADER_DESTROY SVGA_3D_CMD_BASE + 20 #define SVGA_3D_CMD_SET_SHADER SVGA_3D_CMD_BASE + 21 @@ -1018,9 +1050,13 @@ typedef enum { #define SVGA_3D_CMD_BEGIN_QUERY SVGA_3D_CMD_BASE + 25 #define SVGA_3D_CMD_END_QUERY SVGA_3D_CMD_BASE + 26 #define SVGA_3D_CMD_WAIT_FOR_QUERY SVGA_3D_CMD_BASE + 27 -#define SVGA_3D_CMD_PRESENT_READBACK SVGA_3D_CMD_BASE + 28 // Deprecated +#define SVGA_3D_CMD_PRESENT_READBACK SVGA_3D_CMD_BASE + 28 /* Deprecated */ #define SVGA_3D_CMD_BLIT_SURFACE_TO_SCREEN SVGA_3D_CMD_BASE + 29 -#define SVGA_3D_CMD_MAX SVGA_3D_CMD_BASE + 30 +#define SVGA_3D_CMD_SURFACE_DEFINE_V2 SVGA_3D_CMD_BASE + 30 +#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 @@ -1031,9 +1067,9 @@ typedef enum { typedef struct { union { struct { - uint16 function; // SVGA3dFogFunction - uint8 type; // SVGA3dFogType - uint8 base; // SVGA3dFogBase + uint16 function; /* SVGA3dFogFunction */ + uint8 type; /* SVGA3dFogType */ + uint8 base; /* SVGA3dFogBase */ }; uint32 uintValue; }; @@ -1109,6 +1145,8 @@ typedef enum { SVGA3D_SURFACE_HINT_RENDERTARGET = (1 << 6), SVGA3D_SURFACE_HINT_DEPTHSTENCIL = (1 << 7), SVGA3D_SURFACE_HINT_WRITEONLY = (1 << 8), + SVGA3D_SURFACE_MASKABLE_ANTIALIAS = (1 << 9), + SVGA3D_SURFACE_AUTOGENMIPMAPS = (1 << 10), } SVGA3dSurfaceFlags; typedef @@ -1121,6 +1159,12 @@ struct { uint32 sid; SVGA3dSurfaceFlags surfaceFlags; SVGA3dSurfaceFormat format; + /* + * If surfaceFlags has SVGA3D_SURFACE_CUBEMAP bit set, all SVGA3dSurfaceFace + * structures must have the same value of numMipLevels field. + * Otherwise, all but the first SVGA3dSurfaceFace structures must have the + * numMipLevels set to 0. + */ SVGA3dSurfaceFace face[SVGA3D_MAX_SURFACE_FACES]; /* * Followed by an SVGA3dSize structure for each mip level in each face. @@ -1135,6 +1179,31 @@ struct { typedef struct { + uint32 sid; + SVGA3dSurfaceFlags surfaceFlags; + SVGA3dSurfaceFormat format; + /* + * If surfaceFlags has SVGA3D_SURFACE_CUBEMAP bit set, all SVGA3dSurfaceFace + * structures must have the same value of numMipLevels field. + * Otherwise, all but the first SVGA3dSurfaceFace structures must have the + * numMipLevels set to 0. + */ + SVGA3dSurfaceFace face[SVGA3D_MAX_SURFACE_FACES]; + uint32 multisampleCount; + SVGA3dTextureFilter autogenFilter; + /* + * Followed by an SVGA3dSize structure for each mip level in each face. + * + * A note on surface sizes: Sizes are always specified in pixels, + * even if the true surface size is not a multiple of the minimum + * block size of the surface's format. For example, a 3x3x1 DXT1 + * compressed texture would actually be stored as a 4x4x1 image in + * memory. + */ +} SVGA3dCmdDefineSurface_v2; /* SVGA_3D_CMD_SURFACE_DEFINE_V2 */ + +typedef +struct { uint32 sid; } SVGA3dCmdDestroySurface; /* SVGA_3D_CMD_SURFACE_DESTROY */ @@ -1474,10 +1543,12 @@ struct { * SVGA3dCmdDrawPrimitives structure. In order, * they are: * - * 1. SVGA3dVertexDecl, quantity 'numVertexDecls' - * 2. SVGA3dPrimitiveRange, quantity 'numRanges' + * 1. SVGA3dVertexDecl, quantity 'numVertexDecls', but no more than + * SVGA3D_MAX_VERTEX_ARRAYS; + * 2. SVGA3dPrimitiveRange, quantity 'numRanges', but no more than + * SVGA3D_MAX_DRAW_PRIMITIVE_RANGES; * 3. Optionally, SVGA3dVertexDivisor, quantity 'numVertexDecls' (contains - * the frequency divisor for this the corresponding vertex decl) + * the frequency divisor for the corresponding vertex decl). */ } SVGA3dCmdDrawPrimitives; /* SVGA_3D_CMD_DRAWPRIMITIVES */ @@ -1671,6 +1742,12 @@ struct { /* Clipping: zero or more SVGASignedRects follow */ } SVGA3dCmdBlitSurfaceToScreen; /* SVGA_3D_CMD_BLIT_SURFACE_TO_SCREEN */ +typedef +struct { + uint32 sid; + SVGA3dTextureFilter filter; +} SVGA3dCmdGenerateMipmaps; /* SVGA_3D_CMD_GENERATE_MIPMAPS */ + /* * Capability query index. @@ -1774,6 +1851,32 @@ typedef enum { SVGA3D_DEVCAP_SURFACEFMT_A16B16G16R16 = 67, SVGA3D_DEVCAP_SURFACEFMT_UYVY = 68, SVGA3D_DEVCAP_SURFACEFMT_YUY2 = 69, + SVGA3D_DEVCAP_MULTISAMPLE_NONMASKABLESAMPLES = 70, + SVGA3D_DEVCAP_MULTISAMPLE_MASKABLESAMPLES = 71, + SVGA3D_DEVCAP_ALPHATOCOVERAGE = 72, + SVGA3D_DEVCAP_SUPERSAMPLE = 73, + SVGA3D_DEVCAP_AUTOGENMIPMAPS = 74, + SVGA3D_DEVCAP_SURFACEFMT_NV12 = 75, + SVGA3D_DEVCAP_SURFACEFMT_AYUV = 76, + + /* + * This is the maximum number of SVGA context IDs that the guest + * can define using SVGA_3D_CMD_CONTEXT_DEFINE. + */ + SVGA3D_DEVCAP_MAX_CONTEXT_IDS = 77, + + /* + * This is the maximum number of SVGA surface IDs that the guest + * can define using SVGA_3D_CMD_SURFACE_DEFINE*. + */ + SVGA3D_DEVCAP_MAX_SURFACE_IDS = 78, + + SVGA3D_DEVCAP_SURFACEFMT_Z_DF16 = 79, + SVGA3D_DEVCAP_SURFACEFMT_Z_DF24 = 80, + SVGA3D_DEVCAP_SURFACEFMT_Z_D24S8_INT = 81, + + SVGA3D_DEVCAP_SURFACEFMT_BC4_UNORM = 82, + SVGA3D_DEVCAP_SURFACEFMT_BC5_UNORM = 83, /* * Don't add new caps into the previous section; the values in this diff --git a/drivers/gpu/drm/vmwgfx/svga_escape.h b/drivers/gpu/drm/vmwgfx/svga_escape.h index 7b85e9b..8e8d968 100644 --- a/drivers/gpu/drm/vmwgfx/svga_escape.h +++ b/drivers/gpu/drm/vmwgfx/svga_escape.h @@ -75,7 +75,7 @@ */ #define SVGA_ESCAPE_VMWARE_HINT 0x00030000 -#define SVGA_ESCAPE_VMWARE_HINT_FULLSCREEN 0x00030001 // Deprecated +#define SVGA_ESCAPE_VMWARE_HINT_FULLSCREEN 0x00030001 /* Deprecated */ typedef struct { diff --git a/drivers/gpu/drm/vmwgfx/svga_overlay.h b/drivers/gpu/drm/vmwgfx/svga_overlay.h index f753d73..f38416f 100644 --- a/drivers/gpu/drm/vmwgfx/svga_overlay.h +++ b/drivers/gpu/drm/vmwgfx/svga_overlay.h @@ -38,9 +38,9 @@ * Video formats we support */ -#define VMWARE_FOURCC_YV12 0x32315659 // 'Y' 'V' '1' '2' -#define VMWARE_FOURCC_YUY2 0x32595559 // 'Y' 'U' 'Y' '2' -#define VMWARE_FOURCC_UYVY 0x59565955 // 'U' 'Y' 'V' 'Y' +#define VMWARE_FOURCC_YV12 0x32315659 /* 'Y' 'V' '1' '2' */ +#define VMWARE_FOURCC_YUY2 0x32595559 /* 'Y' 'U' 'Y' '2' */ +#define VMWARE_FOURCC_UYVY 0x59565955 /* 'U' 'Y' 'V' 'Y' */ typedef enum { SVGA_OVERLAY_FORMAT_INVALID = 0, @@ -68,7 +68,7 @@ struct SVGAEscapeVideoSetRegs { uint32 streamId; } header; - // May include zero or more items. + /* May include zero or more items. */ struct { uint32 registerId; uint32 value; @@ -134,12 +134,12 @@ struct { */ static inline bool -VMwareVideoGetAttributes(const SVGAOverlayFormat format, // IN - uint32 *width, // IN / OUT - uint32 *height, // IN / OUT - uint32 *size, // OUT - uint32 *pitches, // OUT (optional) - uint32 *offsets) // OUT (optional) +VMwareVideoGetAttributes(const SVGAOverlayFormat format, /* IN */ + uint32 *width, /* IN / OUT */ + uint32 *height, /* IN / OUT */ + uint32 *size, /* OUT */ + uint32 *pitches, /* OUT (optional) */ + uint32 *offsets) /* OUT (optional) */ { int tmp; @@ -198,4 +198,4 @@ VMwareVideoGetAttributes(const SVGAOverlayFormat format, // IN return true; } -#endif // _SVGA_OVERLAY_H_ +#endif /* _SVGA_OVERLAY_H_ */ diff --git a/drivers/gpu/drm/vmwgfx/svga_reg.h b/drivers/gpu/drm/vmwgfx/svga_reg.h index 1b96c2e..01f63cb 100644 --- a/drivers/gpu/drm/vmwgfx/svga_reg.h +++ b/drivers/gpu/drm/vmwgfx/svga_reg.h @@ -39,6 +39,15 @@ #define PCI_DEVICE_ID_VMWARE_SVGA2 0x0405 /* + * SVGA_REG_ENABLE bit definitions. + */ +#define SVGA_REG_ENABLE_DISABLE 0 +#define SVGA_REG_ENABLE_ENABLE 1 +#define SVGA_REG_ENABLE_HIDE 2 +#define SVGA_REG_ENABLE_ENABLE_HIDE (SVGA_REG_ENABLE_ENABLE |\ + SVGA_REG_ENABLE_HIDE) + +/* * Legal values for the SVGA_REG_CURSOR_ON register in old-fashioned * cursor bypass mode. This is still supported, but no new guest * drivers should use it. @@ -158,7 +167,9 @@ enum { SVGA_REG_GMR_MAX_DESCRIPTOR_LENGTH = 44, SVGA_REG_TRACES = 45, /* Enable trace-based updates even when FIFO is on */ - SVGA_REG_TOP = 46, /* Must be 1 more than the last register */ + 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_PALETTE_BASE = 1024, /* Base of SVGA color map */ /* Next 768 (== 256*3) registers exist for colormap */ @@ -265,7 +276,7 @@ enum { * possible. */ #define SVGA_GMR_NULL ((uint32) -1) -#define SVGA_GMR_FRAMEBUFFER ((uint32) -2) // Guest Framebuffer (GFB) +#define SVGA_GMR_FRAMEBUFFER ((uint32) -2) /* Guest Framebuffer (GFB) */ typedef struct SVGAGuestMemDescriptor { @@ -306,13 +317,35 @@ struct SVGAGMRImageFormat { struct { uint32 bitsPerPixel : 8; uint32 colorDepth : 8; - uint32 reserved : 16; // Must be zero + uint32 reserved : 16; /* Must be zero */ }; uint32 value; }; } SVGAGMRImageFormat; +typedef +struct SVGAGuestImage { + SVGAGuestPtr ptr; + + /* + * A note on interpretation of pitch: This value of pitch is the + * number of bytes between vertically adjacent image + * blocks. Normally this is the number of bytes between the first + * pixel of two adjacent scanlines. With compressed textures, + * however, this may represent the number of bytes between + * compression blocks rather than between rows of pixels. + * + * XXX: Compressed textures currently must be tightly packed in guest memory. + * + * If the image is 1-dimensional, pitch is ignored. + * + * If 'pitch' is zero, the SVGA3D device calculates a pitch value + * assuming each row of blocks is tightly packed. + */ + uint32 pitch; +} SVGAGuestImage; + /* * SVGAColorBGRX -- * @@ -328,7 +361,7 @@ struct SVGAColorBGRX { uint32 b : 8; uint32 g : 8; uint32 r : 8; - uint32 x : 8; // Unused + uint32 x : 8; /* Unused */ }; uint32 value; @@ -370,23 +403,34 @@ struct SVGASignedPoint { * Note the holes in the bitfield. Missing bits have been deprecated, * and must not be reused. Those capabilities will never be reported * by new versions of the SVGA device. + * + * SVGA_CAP_GMR2 -- + * Provides asynchronous commands to define and remap guest memory + * regions. Adds device registers SVGA_REG_GMRS_MAX_PAGES and + * SVGA_REG_MEMORY_SIZE. + * + * SVGA_CAP_SCREEN_OBJECT_2 -- + * Allow screen object support, and require backing stores from the + * guest for each screen object. */ #define SVGA_CAP_NONE 0x00000000 #define SVGA_CAP_RECT_COPY 0x00000002 #define SVGA_CAP_CURSOR 0x00000020 -#define SVGA_CAP_CURSOR_BYPASS 0x00000040 // Legacy (Use Cursor Bypass 3 instead) -#define SVGA_CAP_CURSOR_BYPASS_2 0x00000080 // Legacy (Use Cursor Bypass 3 instead) +#define SVGA_CAP_CURSOR_BYPASS 0x00000040 /* Legacy (Use Cursor Bypass 3 instead) */ +#define SVGA_CAP_CURSOR_BYPASS_2 0x00000080 /* Legacy (Use Cursor Bypass 3 instead) */ #define SVGA_CAP_8BIT_EMULATION 0x00000100 #define SVGA_CAP_ALPHA_CURSOR 0x00000200 #define SVGA_CAP_3D 0x00004000 #define SVGA_CAP_EXTENDED_FIFO 0x00008000 -#define SVGA_CAP_MULTIMON 0x00010000 // Legacy multi-monitor support +#define SVGA_CAP_MULTIMON 0x00010000 /* Legacy multi-monitor support */ #define SVGA_CAP_PITCHLOCK 0x00020000 #define SVGA_CAP_IRQMASK 0x00040000 -#define SVGA_CAP_DISPLAY_TOPOLOGY 0x00080000 // Legacy multi-monitor support +#define SVGA_CAP_DISPLAY_TOPOLOGY 0x00080000 /* Legacy multi-monitor support */ #define SVGA_CAP_GMR 0x00100000 #define SVGA_CAP_TRACES 0x00200000 +#define SVGA_CAP_GMR2 0x00400000 +#define SVGA_CAP_SCREEN_OBJECT_2 0x00800000 /* @@ -431,7 +475,7 @@ enum { SVGA_FIFO_CAPABILITIES = 4, SVGA_FIFO_FLAGS, - // Valid with SVGA_FIFO_CAP_FENCE: + /* Valid with SVGA_FIFO_CAP_FENCE: */ SVGA_FIFO_FENCE, /* @@ -444,33 +488,47 @@ enum { * extended FIFO. */ - // Valid if exists (i.e. if extended FIFO enabled): + /* Valid if exists (i.e. if extended FIFO enabled): */ SVGA_FIFO_3D_HWVERSION, /* See SVGA3dHardwareVersion in svga3d_reg.h */ - // Valid with SVGA_FIFO_CAP_PITCHLOCK: + /* Valid with SVGA_FIFO_CAP_PITCHLOCK: */ SVGA_FIFO_PITCHLOCK, - // Valid with SVGA_FIFO_CAP_CURSOR_BYPASS_3: + /* Valid with SVGA_FIFO_CAP_CURSOR_BYPASS_3: */ SVGA_FIFO_CURSOR_ON, /* Cursor bypass 3 show/hide register */ SVGA_FIFO_CURSOR_X, /* Cursor bypass 3 x register */ SVGA_FIFO_CURSOR_Y, /* Cursor bypass 3 y register */ SVGA_FIFO_CURSOR_COUNT, /* Incremented when any of the other 3 change */ SVGA_FIFO_CURSOR_LAST_UPDATED,/* Last time the host updated the cursor */ - // Valid with SVGA_FIFO_CAP_RESERVE: + /* Valid with SVGA_FIFO_CAP_RESERVE: */ SVGA_FIFO_RESERVED, /* Bytes past NEXT_CMD with real contents */ /* - * Valid with SVGA_FIFO_CAP_SCREEN_OBJECT: + * Valid with SVGA_FIFO_CAP_SCREEN_OBJECT or SVGA_FIFO_CAP_SCREEN_OBJECT_2: * * By default this is SVGA_ID_INVALID, to indicate that the cursor * coordinates are specified relative to the virtual root. If this * is set to a specific screen ID, cursor position is reinterpreted - * as a signed offset relative to that screen's origin. This is the - * only way to place the cursor on a non-rooted screen. + * as a signed offset relative to that screen's origin. */ SVGA_FIFO_CURSOR_SCREEN_ID, /* + * Valid with SVGA_FIFO_CAP_DEAD + * + * An arbitrary value written by the host, drivers should not use it. + */ + SVGA_FIFO_DEAD, + + /* + * Valid with SVGA_FIFO_CAP_3D_HWVERSION_REVISED: + * + * Contains 3D HWVERSION (see SVGA3dHardwareVersion in svga3d_reg.h) + * on platforms that can enforce graphics resource limits. + */ + SVGA_FIFO_3D_HWVERSION_REVISED, + + /* * XXX: The gap here, up until SVGA_FIFO_3D_CAPS, can be used for new * registers, but this must be done carefully and with judicious use of * capability bits, since comparisons based on SVGA_FIFO_MIN aren't @@ -508,7 +566,7 @@ enum { * sets SVGA_FIFO_MIN high enough to leave room for them. */ - // Valid if register exists: + /* Valid if register exists: */ SVGA_FIFO_GUEST_3D_HWVERSION, /* Guest driver's 3D version */ SVGA_FIFO_FENCE_GOAL, /* Matching target for SVGA_IRQFLAG_FENCE_GOAL */ SVGA_FIFO_BUSY, /* See "FIFO Synchronization Registers" */ @@ -709,6 +767,37 @@ enum { * * - When a screen is resized, either using Screen Object commands or * legacy multimon registers, its contents are preserved. + * + * SVGA_FIFO_CAP_GMR2 -- + * + * Provides new commands to define and remap guest memory regions (GMR). + * + * New 2D commands: + * DEFINE_GMR2, REMAP_GMR2. + * + * SVGA_FIFO_CAP_3D_HWVERSION_REVISED -- + * + * Indicates new register SVGA_FIFO_3D_HWVERSION_REVISED exists. + * This register may replace SVGA_FIFO_3D_HWVERSION on platforms + * that enforce graphics resource limits. This allows the platform + * to clear SVGA_FIFO_3D_HWVERSION and disable 3D in legacy guest + * drivers that do not limit their resources. + * + * Note this is an alias to SVGA_FIFO_CAP_GMR2 because these indicators + * are codependent (and thus we use a single capability bit). + * + * SVGA_FIFO_CAP_SCREEN_OBJECT_2 -- + * + * Modifies the DEFINE_SCREEN command to include a guest provided + * backing store in GMR memory and the bytesPerLine for the backing + * store. This capability requires the use of a backing store when + * creating screen objects. However if SVGA_FIFO_CAP_SCREEN_OBJECT + * is present then backing stores are optional. + * + * SVGA_FIFO_CAP_DEAD -- + * + * Drivers should not use this cap bit. This cap bit can not be + * reused since some hosts already expose it. */ #define SVGA_FIFO_CAP_NONE 0 @@ -720,6 +809,10 @@ enum { #define SVGA_FIFO_CAP_ESCAPE (1<<5) #define SVGA_FIFO_CAP_RESERVE (1<<6) #define SVGA_FIFO_CAP_SCREEN_OBJECT (1<<7) +#define SVGA_FIFO_CAP_GMR2 (1<<8) +#define SVGA_FIFO_CAP_3D_HWVERSION_REVISED SVGA_FIFO_CAP_GMR2 +#define SVGA_FIFO_CAP_SCREEN_OBJECT_2 (1<<9) +#define SVGA_FIFO_CAP_DEAD (1<<10) /* @@ -730,7 +823,7 @@ enum { #define SVGA_FIFO_FLAG_NONE 0 #define SVGA_FIFO_FLAG_ACCELFRONT (1<<0) -#define SVGA_FIFO_FLAG_RESERVED (1<<31) // Internal use only +#define SVGA_FIFO_FLAG_RESERVED (1<<31) /* Internal use only */ /* * FIFO reservation sentinel value @@ -763,22 +856,22 @@ enum { SVGA_VIDEO_DATA_OFFSET, SVGA_VIDEO_FORMAT, SVGA_VIDEO_COLORKEY, - SVGA_VIDEO_SIZE, // Deprecated + SVGA_VIDEO_SIZE, /* Deprecated */ SVGA_VIDEO_WIDTH, SVGA_VIDEO_HEIGHT, SVGA_VIDEO_SRC_X, SVGA_VIDEO_SRC_Y, SVGA_VIDEO_SRC_WIDTH, SVGA_VIDEO_SRC_HEIGHT, - SVGA_VIDEO_DST_X, // Signed int32 - SVGA_VIDEO_DST_Y, // Signed int32 + SVGA_VIDEO_DST_X, /* Signed int32 */ + SVGA_VIDEO_DST_Y, /* Signed int32 */ SVGA_VIDEO_DST_WIDTH, SVGA_VIDEO_DST_HEIGHT, SVGA_VIDEO_PITCH_1, SVGA_VIDEO_PITCH_2, SVGA_VIDEO_PITCH_3, - SVGA_VIDEO_DATA_GMRID, // Optional, defaults to SVGA_GMR_FRAMEBUFFER - SVGA_VIDEO_DST_SCREEN_ID, // Optional, defaults to virtual coords (SVGA_ID_INVALID) + SVGA_VIDEO_DATA_GMRID, /* Optional, defaults to SVGA_GMR_FRAMEBUFFER */ + SVGA_VIDEO_DST_SCREEN_ID, /* Optional, defaults to virtual coords (SVGA_ID_INVALID) */ SVGA_VIDEO_NUM_REGS }; @@ -829,15 +922,51 @@ typedef struct SVGAOverlayUnit { * compatibility. New flags can be added, and the struct may grow, * but existing fields must retain their meaning. * + * Added with SVGA_FIFO_CAP_SCREEN_OBJECT_2 are required fields of + * a SVGAGuestPtr that is used to back the screen contents. This + * memory must come from the GFB. The guest is not allowed to + * access the memory and doing so will have undefined results. The + * backing store is required to be page aligned and the size is + * padded to the next page boundry. The number of pages is: + * (bytesPerLine * size.width * 4 + PAGE_SIZE - 1) / PAGE_SIZE + * + * The pitch in the backingStore is required to be at least large + * enough to hold a 32bbp scanline. It is recommended that the + * driver pad bytesPerLine for a potential performance win. + * + * The cloneCount field is treated as a hint from the guest that + * the user wants this display to be cloned, countCount times. A + * value of zero means no cloning should happen. + */ + +#define SVGA_SCREEN_MUST_BE_SET (1 << 0) /* Must be set or results undefined */ +#define SVGA_SCREEN_HAS_ROOT SVGA_SCREEN_MUST_BE_SET /* Deprecated */ +#define SVGA_SCREEN_IS_PRIMARY (1 << 1) /* Guest considers this screen to be 'primary' */ +#define SVGA_SCREEN_FULLSCREEN_HINT (1 << 2) /* Guest is running a fullscreen app here */ + +/* + * Added with SVGA_FIFO_CAP_SCREEN_OBJECT_2. When the screen is + * deactivated the base layer is defined to lose all contents and + * become black. When a screen is deactivated the backing store is + * optional. When set backingPtr and bytesPerLine will be ignored. */ +#define SVGA_SCREEN_DEACTIVATE (1 << 3) -#define SVGA_SCREEN_HAS_ROOT (1 << 0) // Screen is present in the virtual coord space -#define SVGA_SCREEN_IS_PRIMARY (1 << 1) // Guest considers this screen to be 'primary' -#define SVGA_SCREEN_FULLSCREEN_HINT (1 << 2) // Guest is running a fullscreen app here +/* + * Added with SVGA_FIFO_CAP_SCREEN_OBJECT_2. When this flag is set + * the screen contents will be outputted as all black to the user + * though the base layer contents is preserved. The screen base layer + * can still be read and written to like normal though the no visible + * effect will be seen by the user. When the flag is changed the + * screen will be blanked or redrawn to the current contents as needed + * without any extra commands from the driver. This flag only has an + * effect when the screen is not deactivated. + */ +#define SVGA_SCREEN_BLANKING (1 << 4) typedef struct SVGAScreenObject { - uint32 structSize; // sizeof(SVGAScreenObject) + uint32 structSize; /* sizeof(SVGAScreenObject) */ uint32 id; uint32 flags; struct { @@ -847,7 +976,14 @@ struct SVGAScreenObject { struct { int32 x; int32 y; - } root; // Only used if SVGA_SCREEN_HAS_ROOT is set. + } root; + + /* + * Added and required by SVGA_FIFO_CAP_SCREEN_OBJECT_2, optional + * with SVGA_FIFO_CAP_SCREEN_OBJECT. + */ + SVGAGuestImage backingStore; + uint32 cloneCount; } SVGAScreenObject; @@ -885,6 +1021,8 @@ typedef enum { SVGA_CMD_BLIT_SCREEN_TO_GMRFB = 38, SVGA_CMD_ANNOTATION_FILL = 39, SVGA_CMD_ANNOTATION_COPY = 40, + SVGA_CMD_DEFINE_GMR2 = 41, + SVGA_CMD_REMAP_GMR2 = 42, SVGA_CMD_MAX } SVGAFifoCmdId; @@ -920,7 +1058,7 @@ typedef enum { */ typedef -struct { +struct SVGAFifoCmdUpdate { uint32 x; uint32 y; uint32 width; @@ -939,7 +1077,7 @@ struct { */ typedef -struct { +struct SVGAFifoCmdRectCopy { uint32 srcX; uint32 srcY; uint32 destX; @@ -963,14 +1101,14 @@ struct { */ typedef -struct { - uint32 id; // Reserved, must be zero. +struct SVGAFifoCmdDefineCursor { + uint32 id; /* Reserved, must be zero. */ uint32 hotspotX; uint32 hotspotY; uint32 width; uint32 height; - uint32 andMaskDepth; // Value must be 1 or equal to BITS_PER_PIXEL - uint32 xorMaskDepth; // Value must be 1 or equal to BITS_PER_PIXEL + uint32 andMaskDepth; /* Value must be 1 or equal to BITS_PER_PIXEL */ + uint32 xorMaskDepth; /* Value must be 1 or equal to BITS_PER_PIXEL */ /* * Followed by scanline data for AND mask, then XOR mask. * Each scanline is padded to a 32-bit boundary. @@ -992,8 +1130,8 @@ struct { */ typedef -struct { - uint32 id; // Reserved, must be zero. +struct SVGAFifoCmdDefineAlphaCursor { + uint32 id; /* Reserved, must be zero. */ uint32 hotspotX; uint32 hotspotY; uint32 width; @@ -1015,7 +1153,7 @@ struct { */ typedef -struct { +struct SVGAFifoCmdUpdateVerbose { uint32 x; uint32 y; uint32 width; @@ -1040,13 +1178,13 @@ struct { #define SVGA_ROP_COPY 0x03 typedef -struct { - uint32 color; // In the same format as the GFB +struct SVGAFifoCmdFrontRopFill { + uint32 color; /* In the same format as the GFB */ uint32 x; uint32 y; uint32 width; uint32 height; - uint32 rop; // Must be SVGA_ROP_COPY + uint32 rop; /* Must be SVGA_ROP_COPY */ } SVGAFifoCmdFrontRopFill; @@ -1083,7 +1221,7 @@ struct { */ typedef -struct { +struct SVGAFifoCmdEscape { uint32 nsid; uint32 size; /* followed by 'size' bytes of data */ @@ -1113,12 +1251,12 @@ struct { * registers (SVGA_REG_NUM_GUEST_DISPLAYS, SVGA_REG_DISPLAY_*). * * Availability: - * SVGA_FIFO_CAP_SCREEN_OBJECT + * SVGA_FIFO_CAP_SCREEN_OBJECT or SVGA_FIFO_CAP_SCREEN_OBJECT_2 */ typedef struct { - SVGAScreenObject screen; // Variable-length according to version + SVGAScreenObject screen; /* Variable-length according to version */ } SVGAFifoCmdDefineScreen; @@ -1129,7 +1267,7 @@ struct { * re-use. * * Availability: - * SVGA_FIFO_CAP_SCREEN_OBJECT + * SVGA_FIFO_CAP_SCREEN_OBJECT or SVGA_FIFO_CAP_SCREEN_OBJECT_2 */ typedef @@ -1182,7 +1320,7 @@ struct { * GMRFB. * * Availability: - * SVGA_FIFO_CAP_SCREEN_OBJECT + * SVGA_FIFO_CAP_SCREEN_OBJECT or SVGA_FIFO_CAP_SCREEN_OBJECT_2 */ typedef @@ -1219,7 +1357,7 @@ struct { * SVGA_CMD_ANNOTATION_* commands for details. * * Availability: - * SVGA_FIFO_CAP_SCREEN_OBJECT + * SVGA_FIFO_CAP_SCREEN_OBJECT or SVGA_FIFO_CAP_SCREEN_OBJECT_2 */ typedef @@ -1267,7 +1405,7 @@ struct { * the time any subsequent FENCE commands are reached. * * Availability: - * SVGA_FIFO_CAP_SCREEN_OBJECT + * SVGA_FIFO_CAP_SCREEN_OBJECT or SVGA_FIFO_CAP_SCREEN_OBJECT_2 */ typedef @@ -1302,7 +1440,7 @@ struct { * user's display is being remoted over a network connection. * * Availability: - * SVGA_FIFO_CAP_SCREEN_OBJECT + * SVGA_FIFO_CAP_SCREEN_OBJECT or SVGA_FIFO_CAP_SCREEN_OBJECT_2 */ typedef @@ -1334,7 +1472,7 @@ struct { * undefined. * * Availability: - * SVGA_FIFO_CAP_SCREEN_OBJECT + * SVGA_FIFO_CAP_SCREEN_OBJECT or SVGA_FIFO_CAP_SCREEN_OBJECT_2 */ typedef @@ -1343,4 +1481,72 @@ struct { uint32 srcScreenId; } SVGAFifoCmdAnnotationCopy; + +/* + * SVGA_CMD_DEFINE_GMR2 -- + * + * Define guest memory region v2. See the description of GMRs above. + * + * Availability: + * SVGA_CAP_GMR2 + */ + +typedef +struct { + uint32 gmrId; + uint32 numPages; +} SVGAFifoCmdDefineGMR2; + + +/* + * SVGA_CMD_REMAP_GMR2 -- + * + * Remap guest memory region v2. See the description of GMRs above. + * + * This command allows guest to modify a portion of an existing GMR by + * invalidating it or reassigning it to different guest physical pages. + * The pages are identified by physical page number (PPN). The pages + * are assumed to be pinned and valid for DMA operations. + * + * Description of command flags: + * + * SVGA_REMAP_GMR2_VIA_GMR: If enabled, references a PPN list in a GMR. + * The PPN list must not overlap with the remap region (this can be + * handled trivially by referencing a separate GMR). If flag is + * disabled, PPN list is appended to SVGARemapGMR command. + * + * SVGA_REMAP_GMR2_PPN64: If set, PPN list is in PPN64 format, otherwise + * it is in PPN32 format. + * + * SVGA_REMAP_GMR2_SINGLE_PPN: If set, PPN list contains a single entry. + * A single PPN can be used to invalidate a portion of a GMR or + * map it to to a single guest scratch page. + * + * Availability: + * SVGA_CAP_GMR2 + */ + +typedef enum { + SVGA_REMAP_GMR2_PPN32 = 0, + SVGA_REMAP_GMR2_VIA_GMR = (1 << 0), + SVGA_REMAP_GMR2_PPN64 = (1 << 1), + SVGA_REMAP_GMR2_SINGLE_PPN = (1 << 2), +} SVGARemapGMR2Flags; + +typedef +struct { + uint32 gmrId; + SVGARemapGMR2Flags flags; + uint32 offsetPages; /* offset in pages to begin remap */ + uint32 numPages; /* number of pages to remap */ + /* + * Followed by additional data depending on SVGARemapGMR2Flags. + * + * If flag SVGA_REMAP_GMR2_VIA_GMR is set, single SVGAGuestPtr follows. + * Otherwise an array of page descriptors in PPN32 or PPN64 format + * (according to flag SVGA_REMAP_GMR2_PPN64) follows. If flag + * SVGA_REMAP_GMR2_SINGLE_PPN is set, array contains a single entry. + */ +} SVGAFifoCmdRemapGMR2; + #endif diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c index 87e43e0..5a72ed9 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c @@ -42,6 +42,10 @@ static uint32_t sys_placement_flags = TTM_PL_FLAG_SYSTEM | static uint32_t gmr_placement_flags = VMW_PL_FLAG_GMR | TTM_PL_FLAG_CACHED; +static uint32_t gmr_ne_placement_flags = VMW_PL_FLAG_GMR | + TTM_PL_FLAG_CACHED | + TTM_PL_FLAG_NO_EVICT; + struct ttm_placement vmw_vram_placement = { .fpfn = 0, .lpfn = 0, @@ -56,6 +60,11 @@ static uint32_t vram_gmr_placement_flags[] = { VMW_PL_FLAG_GMR | TTM_PL_FLAG_CACHED }; +static uint32_t gmr_vram_placement_flags[] = { + VMW_PL_FLAG_GMR | TTM_PL_FLAG_CACHED, + TTM_PL_FLAG_VRAM | TTM_PL_FLAG_CACHED +}; + struct ttm_placement vmw_vram_gmr_placement = { .fpfn = 0, .lpfn = 0, @@ -65,6 +74,20 @@ struct ttm_placement vmw_vram_gmr_placement = { .busy_placement = &gmr_placement_flags }; +static uint32_t vram_gmr_ne_placement_flags[] = { + TTM_PL_FLAG_VRAM | TTM_PL_FLAG_CACHED | TTM_PL_FLAG_NO_EVICT, + VMW_PL_FLAG_GMR | TTM_PL_FLAG_CACHED | TTM_PL_FLAG_NO_EVICT +}; + +struct ttm_placement vmw_vram_gmr_ne_placement = { + .fpfn = 0, + .lpfn = 0, + .num_placement = 2, + .placement = vram_gmr_ne_placement_flags, + .num_busy_placement = 1, + .busy_placement = &gmr_ne_placement_flags +}; + struct ttm_placement vmw_vram_sys_placement = { .fpfn = 0, .lpfn = 0, @@ -92,6 +115,30 @@ struct ttm_placement vmw_sys_placement = { .busy_placement = &sys_placement_flags }; +static uint32_t evictable_placement_flags[] = { + TTM_PL_FLAG_SYSTEM | TTM_PL_FLAG_CACHED, + TTM_PL_FLAG_VRAM | TTM_PL_FLAG_CACHED, + VMW_PL_FLAG_GMR | TTM_PL_FLAG_CACHED +}; + +struct ttm_placement vmw_evictable_placement = { + .fpfn = 0, + .lpfn = 0, + .num_placement = 3, + .placement = evictable_placement_flags, + .num_busy_placement = 1, + .busy_placement = &sys_placement_flags +}; + +struct ttm_placement vmw_srf_placement = { + .fpfn = 0, + .lpfn = 0, + .num_placement = 1, + .num_busy_placement = 2, + .placement = &gmr_placement_flags, + .busy_placement = gmr_vram_placement_flags +}; + struct vmw_ttm_backend { struct ttm_backend backend; struct page **pages; @@ -274,39 +321,39 @@ static int vmw_ttm_fault_reserve_notify(struct ttm_buffer_object *bo) static void *vmw_sync_obj_ref(void *sync_obj) { - return sync_obj; + + return (void *) + vmw_fence_obj_reference((struct vmw_fence_obj *) sync_obj); } static void vmw_sync_obj_unref(void **sync_obj) { - *sync_obj = NULL; + vmw_fence_obj_unreference((struct vmw_fence_obj **) sync_obj); } static int vmw_sync_obj_flush(void *sync_obj, void *sync_arg) { - struct vmw_private *dev_priv = (struct vmw_private *)sync_arg; - - mutex_lock(&dev_priv->hw_mutex); - vmw_write(dev_priv, SVGA_REG_SYNC, SVGA_SYNC_GENERIC); - mutex_unlock(&dev_priv->hw_mutex); + vmw_fence_obj_flush((struct vmw_fence_obj *) sync_obj); return 0; } static bool vmw_sync_obj_signaled(void *sync_obj, void *sync_arg) { - struct vmw_private *dev_priv = (struct vmw_private *)sync_arg; - uint32_t sequence = (unsigned long) sync_obj; + unsigned long flags = (unsigned long) sync_arg; + return vmw_fence_obj_signaled((struct vmw_fence_obj *) sync_obj, + (uint32_t) flags); - return vmw_fence_signaled(dev_priv, sequence); } static int vmw_sync_obj_wait(void *sync_obj, void *sync_arg, bool lazy, bool interruptible) { - struct vmw_private *dev_priv = (struct vmw_private *)sync_arg; - uint32_t sequence = (unsigned long) sync_obj; + unsigned long flags = (unsigned long) sync_arg; - return vmw_wait_fence(dev_priv, false, sequence, false, 3*HZ); + return vmw_fence_obj_wait((struct vmw_fence_obj *) sync_obj, + (uint32_t) flags, + lazy, interruptible, + VMW_FENCE_WAIT_TIMEOUT); } struct ttm_bo_driver vmw_bo_driver = { diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c index 55d4b29..f739fcf 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c @@ -24,6 +24,7 @@ * USE OR OTHER DEALINGS IN THE SOFTWARE. * **************************************************************************/ +#include <linux/module.h> #include "drmP.h" #include "vmwgfx_drv.h" @@ -82,17 +83,31 @@ #define DRM_IOCTL_VMW_EXECBUF \ DRM_IOW(DRM_COMMAND_BASE + DRM_VMW_EXECBUF, \ struct drm_vmw_execbuf_arg) -#define DRM_IOCTL_VMW_FIFO_DEBUG \ - DRM_IOWR(DRM_COMMAND_BASE + DRM_VMW_FIFO_DEBUG, \ - struct drm_vmw_fifo_debug_arg) +#define DRM_IOCTL_VMW_GET_3D_CAP \ + DRM_IOW(DRM_COMMAND_BASE + DRM_VMW_GET_3D_CAP, \ + struct drm_vmw_get_3d_cap_arg) #define DRM_IOCTL_VMW_FENCE_WAIT \ DRM_IOWR(DRM_COMMAND_BASE + DRM_VMW_FENCE_WAIT, \ struct drm_vmw_fence_wait_arg) +#define DRM_IOCTL_VMW_FENCE_SIGNALED \ + DRM_IOWR(DRM_COMMAND_BASE + DRM_VMW_FENCE_SIGNALED, \ + struct drm_vmw_fence_signaled_arg) +#define DRM_IOCTL_VMW_FENCE_UNREF \ + DRM_IOW(DRM_COMMAND_BASE + DRM_VMW_FENCE_UNREF, \ + struct drm_vmw_fence_arg) +#define DRM_IOCTL_VMW_FENCE_EVENT \ + DRM_IOW(DRM_COMMAND_BASE + DRM_VMW_FENCE_EVENT, \ + struct drm_vmw_fence_event_arg) +#define DRM_IOCTL_VMW_PRESENT \ + DRM_IOW(DRM_COMMAND_BASE + DRM_VMW_PRESENT, \ + struct drm_vmw_present_arg) +#define DRM_IOCTL_VMW_PRESENT_READBACK \ + DRM_IOW(DRM_COMMAND_BASE + DRM_VMW_PRESENT_READBACK, \ + struct drm_vmw_present_readback_arg) #define DRM_IOCTL_VMW_UPDATE_LAYOUT \ - DRM_IOWR(DRM_COMMAND_BASE + DRM_VMW_UPDATE_LAYOUT, \ + DRM_IOW(DRM_COMMAND_BASE + DRM_VMW_UPDATE_LAYOUT, \ struct drm_vmw_update_layout_arg) - /** * The core DRM version of this macro doesn't account for * DRM_COMMAND_BASE. @@ -135,12 +150,28 @@ static struct drm_ioctl_desc vmw_ioctls[] = { DRM_AUTH | DRM_UNLOCKED), VMW_IOCTL_DEF(VMW_EXECBUF, vmw_execbuf_ioctl, DRM_AUTH | DRM_UNLOCKED), - VMW_IOCTL_DEF(VMW_FIFO_DEBUG, vmw_fifo_debug_ioctl, - DRM_AUTH | DRM_ROOT_ONLY | DRM_MASTER | DRM_UNLOCKED), - VMW_IOCTL_DEF(VMW_FENCE_WAIT, vmw_fence_wait_ioctl, + VMW_IOCTL_DEF(VMW_FENCE_WAIT, vmw_fence_obj_wait_ioctl, + DRM_AUTH | DRM_UNLOCKED), + VMW_IOCTL_DEF(VMW_FENCE_SIGNALED, + vmw_fence_obj_signaled_ioctl, + DRM_AUTH | DRM_UNLOCKED), + 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), - VMW_IOCTL_DEF(VMW_UPDATE_LAYOUT, vmw_kms_update_layout_ioctl, - DRM_MASTER | DRM_CONTROL_ALLOW | DRM_UNLOCKED) + VMW_IOCTL_DEF(VMW_GET_3D_CAP, vmw_get_cap_3d_ioctl, + DRM_AUTH | DRM_UNLOCKED), + + /* these allow direct access to the framebuffers mark as master only */ + VMW_IOCTL_DEF(VMW_PRESENT, vmw_present_ioctl, + DRM_MASTER | DRM_AUTH | DRM_UNLOCKED), + VMW_IOCTL_DEF(VMW_PRESENT_READBACK, + vmw_present_readback_ioctl, + DRM_MASTER | DRM_AUTH | DRM_UNLOCKED), + VMW_IOCTL_DEF(VMW_UPDATE_LAYOUT, + vmw_kms_update_layout_ioctl, + DRM_MASTER | DRM_UNLOCKED), }; static struct pci_device_id vmw_pci_id_list[] = { @@ -190,8 +221,78 @@ static void vmw_print_capabilities(uint32_t capabilities) DRM_INFO(" GMR.\n"); if (capabilities & SVGA_CAP_TRACES) DRM_INFO(" Traces.\n"); + if (capabilities & SVGA_CAP_GMR2) + DRM_INFO(" GMR2.\n"); + if (capabilities & SVGA_CAP_SCREEN_OBJECT_2) + DRM_INFO(" Screen Object 2.\n"); } + +/** + * vmw_execbuf_prepare_dummy_query - Initialize a query result structure at + * the start of a buffer object. + * + * @dev_priv: The device private structure. + * + * This function will idle the buffer using an uninterruptible wait, then + * map the first page and initialize a pending occlusion query result structure, + * Finally it will unmap the buffer. + * + * TODO: Since we're only mapping a single page, we should optimize the map + * to use kmap_atomic / iomap_atomic. + */ +static void vmw_dummy_query_bo_prepare(struct vmw_private *dev_priv) +{ + struct ttm_bo_kmap_obj map; + volatile SVGA3dQueryResult *result; + bool dummy; + int ret; + struct ttm_bo_device *bdev = &dev_priv->bdev; + struct ttm_buffer_object *bo = dev_priv->dummy_query_bo; + + ttm_bo_reserve(bo, false, false, false, 0); + spin_lock(&bdev->fence_lock); + ret = ttm_bo_wait(bo, false, false, false); + spin_unlock(&bdev->fence_lock); + if (unlikely(ret != 0)) + (void) vmw_fallback_wait(dev_priv, false, true, 0, false, + 10*HZ); + + ret = ttm_bo_kmap(bo, 0, 1, &map); + if (likely(ret == 0)) { + result = ttm_kmap_obj_virtual(&map, &dummy); + result->totalSize = sizeof(*result); + result->state = SVGA3D_QUERYSTATE_PENDING; + result->result32 = 0xff; + ttm_bo_kunmap(&map); + } else + DRM_ERROR("Dummy query buffer map failed.\n"); + ttm_bo_unreserve(bo); +} + + +/** + * vmw_dummy_query_bo_create - create a bo to hold a dummy query result + * + * @dev_priv: A device private structure. + * + * This function creates a small buffer object that holds the query + * result for dummy queries emitted as query barriers. + * No interruptible waits are done within this function. + * + * Returns an error if bo creation fails. + */ +static int vmw_dummy_query_bo_create(struct vmw_private *dev_priv) +{ + return ttm_bo_create(&dev_priv->bdev, + PAGE_SIZE, + ttm_bo_type_device, + &vmw_vram_sys_placement, + 0, 0, false, NULL, + &dev_priv->dummy_query_bo); +} + + static int vmw_request_device(struct vmw_private *dev_priv) { int ret; @@ -201,16 +302,42 @@ static int vmw_request_device(struct vmw_private *dev_priv) DRM_ERROR("Unable to initialize FIFO.\n"); return ret; } + vmw_fence_fifo_up(dev_priv->fman); + ret = vmw_dummy_query_bo_create(dev_priv); + if (unlikely(ret != 0)) + goto out_no_query_bo; + vmw_dummy_query_bo_prepare(dev_priv); return 0; + +out_no_query_bo: + vmw_fence_fifo_down(dev_priv->fman); + vmw_fifo_release(dev_priv, &dev_priv->fifo); + return ret; } static void vmw_release_device(struct vmw_private *dev_priv) { + /* + * Previous destructions should've released + * the pinned bo. + */ + + BUG_ON(dev_priv->pinned_bo != NULL); + + ttm_bo_unref(&dev_priv->dummy_query_bo); + vmw_fence_fifo_down(dev_priv->fman); vmw_fifo_release(dev_priv, &dev_priv->fifo); } -int vmw_3d_resource_inc(struct vmw_private *dev_priv) +/** + * Increase the 3d resource refcount. + * If the count was prevously zero, initialize the fifo, switching to svga + * mode. Note that the master holds a ref as well, and may request an + * explicit switch to svga mode if fb is not running, using @unhide_svga. + */ +int vmw_3d_resource_inc(struct vmw_private *dev_priv, + bool unhide_svga) { int ret = 0; @@ -219,19 +346,42 @@ int vmw_3d_resource_inc(struct vmw_private *dev_priv) ret = vmw_request_device(dev_priv); if (unlikely(ret != 0)) --dev_priv->num_3d_resources; + } else if (unhide_svga) { + mutex_lock(&dev_priv->hw_mutex); + vmw_write(dev_priv, SVGA_REG_ENABLE, + vmw_read(dev_priv, SVGA_REG_ENABLE) & + ~SVGA_REG_ENABLE_HIDE); + mutex_unlock(&dev_priv->hw_mutex); } + mutex_unlock(&dev_priv->release_mutex); return ret; } - -void vmw_3d_resource_dec(struct vmw_private *dev_priv) +/** + * Decrease the 3d resource refcount. + * If the count reaches zero, disable the fifo, switching to vga mode. + * Note that the master holds a refcount as well, and may request an + * explicit switch to vga mode when it releases its refcount to account + * for the situation of an X server vt switch to VGA with 3d resources + * active. + */ +void vmw_3d_resource_dec(struct vmw_private *dev_priv, + bool hide_svga) { int32_t n3d; mutex_lock(&dev_priv->release_mutex); if (unlikely(--dev_priv->num_3d_resources == 0)) vmw_release_device(dev_priv); + else if (hide_svga) { + mutex_lock(&dev_priv->hw_mutex); + vmw_write(dev_priv, SVGA_REG_ENABLE, + vmw_read(dev_priv, SVGA_REG_ENABLE) | + SVGA_REG_ENABLE_HIDE); + mutex_unlock(&dev_priv->hw_mutex); + } + n3d = (int32_t) dev_priv->num_3d_resources; mutex_unlock(&dev_priv->release_mutex); @@ -253,7 +403,7 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset) dev_priv->dev = dev; dev_priv->vmw_chipset = chipset; - dev_priv->last_read_sequence = (uint32_t) -100; + dev_priv->last_read_seqno = (uint32_t) -100; mutex_init(&dev_priv->hw_mutex); mutex_init(&dev_priv->cmdbuf_mutex); mutex_init(&dev_priv->release_mutex); @@ -264,8 +414,10 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset) mutex_init(&dev_priv->init_mutex); init_waitqueue_head(&dev_priv->fence_queue); init_waitqueue_head(&dev_priv->fifo_queue); - atomic_set(&dev_priv->fence_queue_waiters, 0); + dev_priv->fence_queue_waiters = 0; atomic_set(&dev_priv->fifo_queue_waiters, 0); + INIT_LIST_HEAD(&dev_priv->surface_lru); + dev_priv->used_memory_size = 0; dev_priv->io_start = pci_resource_start(dev->pdev, 0); dev_priv->vram_start = pci_resource_start(dev->pdev, 1); @@ -286,6 +438,10 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset) dev_priv->capabilities = vmw_read(dev_priv, SVGA_REG_CAPABILITIES); + dev_priv->vram_size = vmw_read(dev_priv, SVGA_REG_VRAM_SIZE); + dev_priv->mmio_size = vmw_read(dev_priv, SVGA_REG_MEM_SIZE); + dev_priv->fb_max_width = vmw_read(dev_priv, SVGA_REG_MAX_WIDTH); + dev_priv->fb_max_height = vmw_read(dev_priv, SVGA_REG_MAX_HEIGHT); if (dev_priv->capabilities & SVGA_CAP_GMR) { dev_priv->max_gmr_descriptors = vmw_read(dev_priv, @@ -293,11 +449,19 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset) dev_priv->max_gmr_ids = vmw_read(dev_priv, SVGA_REG_GMR_MAX_IDS); } - - dev_priv->vram_size = vmw_read(dev_priv, SVGA_REG_VRAM_SIZE); - dev_priv->mmio_size = vmw_read(dev_priv, SVGA_REG_MEM_SIZE); - dev_priv->fb_max_width = vmw_read(dev_priv, SVGA_REG_MAX_WIDTH); - dev_priv->fb_max_height = vmw_read(dev_priv, SVGA_REG_MAX_HEIGHT); + if (dev_priv->capabilities & SVGA_CAP_GMR2) { + dev_priv->max_gmr_pages = + vmw_read(dev_priv, SVGA_REG_GMRS_MAX_PAGES); + dev_priv->memory_size = + vmw_read(dev_priv, SVGA_REG_MEMORY_SIZE); + dev_priv->memory_size -= dev_priv->vram_size; + } else { + /* + * An arbitrary limit of 512MiB on surface + * memory. But all HWV8 hardware supports GMR2. + */ + dev_priv->memory_size = 512*1024*1024; + } mutex_unlock(&dev_priv->hw_mutex); @@ -309,6 +473,12 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset) 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("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", @@ -333,21 +503,6 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset) goto out_err1; } - ret = ttm_bo_init_mm(&dev_priv->bdev, TTM_PL_VRAM, - (dev_priv->vram_size >> PAGE_SHIFT)); - if (unlikely(ret != 0)) { - DRM_ERROR("Failed initializing memory manager for VRAM.\n"); - goto out_err2; - } - - dev_priv->has_gmr = true; - if (ttm_bo_init_mm(&dev_priv->bdev, VMW_PL_GMR, - dev_priv->max_gmr_ids) != 0) { - DRM_INFO("No GMR memory available. " - "Graphics memory resources are very limited.\n"); - dev_priv->has_gmr = false; - } - dev_priv->mmio_mtrr = drm_mtrr_add(dev_priv->mmio_start, dev_priv->mmio_size, DRM_MTRR_WC); @@ -395,22 +550,50 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset) goto out_no_device; } } + + dev_priv->fman = vmw_fence_manager_init(dev_priv); + if (unlikely(dev_priv->fman == NULL)) + goto out_no_fman; + + + ret = ttm_bo_init_mm(&dev_priv->bdev, TTM_PL_VRAM, + (dev_priv->vram_size >> PAGE_SHIFT)); + if (unlikely(ret != 0)) { + DRM_ERROR("Failed initializing memory manager for VRAM.\n"); + goto out_no_vram; + } + + dev_priv->has_gmr = true; + if (ttm_bo_init_mm(&dev_priv->bdev, VMW_PL_GMR, + dev_priv->max_gmr_ids) != 0) { + DRM_INFO("No GMR memory available. " + "Graphics memory resources are very limited.\n"); + dev_priv->has_gmr = false; + } + + /* Need to start the fifo to check if we can do screen objects */ + ret = vmw_3d_resource_inc(dev_priv, true); + if (unlikely(ret != 0)) + goto out_no_fifo; + vmw_kms_save_vga(dev_priv); + + /* Start kms and overlay systems, needs fifo. */ ret = vmw_kms_init(dev_priv); if (unlikely(ret != 0)) goto out_no_kms; vmw_overlay_init(dev_priv); + + /* 3D Depends on Screen Objects being used. */ + DRM_INFO("Detected %sdevice 3D availability.\n", + vmw_fifo_have_3d(dev_priv) ? + "" : "no "); + + /* We might be done with the fifo now */ if (dev_priv->enable_fb) { - ret = vmw_3d_resource_inc(dev_priv); - if (unlikely(ret != 0)) - goto out_no_fifo; - vmw_kms_save_vga(dev_priv); vmw_fb_init(dev_priv); - DRM_INFO("%s", vmw_fifo_have_3d(dev_priv) ? - "Detected device 3D availability.\n" : - "Detected no device 3D availability.\n"); } else { - DRM_INFO("Delayed 3D detection since we're not " - "running the device in SVGA mode yet.\n"); + vmw_kms_restore_vga(dev_priv); + vmw_3d_resource_dec(dev_priv, true); } if (dev_priv->capabilities & SVGA_CAP_IRQMASK) { @@ -427,15 +610,23 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset) return 0; out_no_irq: - if (dev_priv->enable_fb) { + if (dev_priv->enable_fb) vmw_fb_close(dev_priv); - vmw_kms_restore_vga(dev_priv); - vmw_3d_resource_dec(dev_priv); - } -out_no_fifo: vmw_overlay_close(dev_priv); vmw_kms_close(dev_priv); out_no_kms: + /* We still have a 3D resource reference held */ + if (dev_priv->enable_fb) { + vmw_kms_restore_vga(dev_priv); + vmw_3d_resource_dec(dev_priv, false); + } +out_no_fifo: + 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); +out_no_vram: + vmw_fence_manager_takedown(dev_priv->fman); +out_no_fman: if (dev_priv->stealth) pci_release_region(dev->pdev, 2); else @@ -447,10 +638,6 @@ out_err4: out_err3: drm_mtrr_del(dev_priv->mmio_mtrr, dev_priv->mmio_start, dev_priv->mmio_size, DRM_MTRR_WC); - 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); -out_err2: (void)ttm_bo_device_release(&dev_priv->bdev); out_err1: vmw_ttm_global_release(dev_priv); @@ -468,15 +655,23 @@ static int vmw_driver_unload(struct drm_device *dev) unregister_pm_notifier(&dev_priv->pm_nb); + if (dev_priv->ctx.cmd_bounce) + vfree(dev_priv->ctx.cmd_bounce); if (dev_priv->capabilities & SVGA_CAP_IRQMASK) drm_irq_uninstall(dev_priv->dev); if (dev_priv->enable_fb) { vmw_fb_close(dev_priv); vmw_kms_restore_vga(dev_priv); - vmw_3d_resource_dec(dev_priv); + vmw_3d_resource_dec(dev_priv, false); } vmw_kms_close(dev_priv); vmw_overlay_close(dev_priv); + + 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); + + vmw_fence_manager_takedown(dev_priv->fman); if (dev_priv->stealth) pci_release_region(dev->pdev, 2); else @@ -486,9 +681,6 @@ static int vmw_driver_unload(struct drm_device *dev) iounmap(dev_priv->mmio_virt); drm_mtrr_del(dev_priv->mmio_mtrr, dev_priv->mmio_start, dev_priv->mmio_size, DRM_MTRR_WC); - 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); (void)ttm_bo_device_release(&dev_priv->bdev); vmw_ttm_global_release(dev_priv); idr_destroy(&dev_priv->surface_idr); @@ -647,7 +839,7 @@ static int vmw_master_set(struct drm_device *dev, int ret = 0; if (!dev_priv->enable_fb) { - ret = vmw_3d_resource_inc(dev_priv); + ret = vmw_3d_resource_inc(dev_priv, true); if (unlikely(ret != 0)) return ret; vmw_kms_save_vga(dev_priv); @@ -689,7 +881,7 @@ out_no_active_lock: vmw_write(dev_priv, SVGA_REG_TRACES, 1); mutex_unlock(&dev_priv->hw_mutex); vmw_kms_restore_vga(dev_priv); - vmw_3d_resource_dec(dev_priv); + vmw_3d_resource_dec(dev_priv, true); } return ret; } @@ -710,7 +902,7 @@ static void vmw_master_drop(struct drm_device *dev, vmw_fp->locked_master = drm_master_get(file_priv->master); ret = ttm_vt_lock(&vmaster->lock, false, vmw_fp->tfile); - vmw_kms_idle_workqueues(vmaster); + vmw_execbuf_release_pinned_bo(dev_priv, false, 0); if (unlikely((ret != 0))) { DRM_ERROR("Unable to lock TTM at VT switch.\n"); @@ -727,7 +919,7 @@ static void vmw_master_drop(struct drm_device *dev, vmw_write(dev_priv, SVGA_REG_TRACES, 1); mutex_unlock(&dev_priv->hw_mutex); vmw_kms_restore_vga(dev_priv); - vmw_3d_resource_dec(dev_priv); + vmw_3d_resource_dec(dev_priv, true); } dev_priv->active_master = &dev_priv->fbdev_master; @@ -762,6 +954,7 @@ static int vmwgfx_pm_notifier(struct notifier_block *nb, unsigned long val, * This empties VRAM and unbinds all GMR bindings. * Buffer contents is moved to swappable memory. */ + vmw_execbuf_release_pinned_bo(dev_priv, false, 0); ttm_bo_swapout_all(&dev_priv->bdev); break; @@ -836,7 +1029,7 @@ static int vmw_pm_prepare(struct device *kdev) */ dev_priv->suspended = true; if (dev_priv->enable_fb) - vmw_3d_resource_dec(dev_priv); + vmw_3d_resource_dec(dev_priv, true); if (dev_priv->num_3d_resources != 0) { @@ -844,7 +1037,7 @@ static int vmw_pm_prepare(struct device *kdev) "while 3D resources are active.\n"); if (dev_priv->enable_fb) - vmw_3d_resource_inc(dev_priv); + vmw_3d_resource_inc(dev_priv, true); dev_priv->suspended = false; return -EBUSY; } @@ -868,7 +1061,7 @@ static void vmw_pm_complete(struct device *kdev) * start fifo. */ if (dev_priv->enable_fb) - vmw_3d_resource_inc(dev_priv); + vmw_3d_resource_inc(dev_priv, false); dev_priv->suspended = false; } @@ -892,6 +1085,8 @@ static struct drm_driver driver = { .irq_uninstall = vmw_irq_uninstall, .irq_handler = vmw_irq_handler, .get_vblank_counter = vmw_get_vblank_counter, + .enable_vblank = vmw_enable_vblank, + .disable_vblank = vmw_disable_vblank, .reclaim_buffers_locked = NULL, .ioctls = vmw_ioctls, .num_ioctls = DRM_ARRAY_SIZE(vmw_ioctls), @@ -902,13 +1097,19 @@ static struct drm_driver driver = { .master_drop = vmw_master_drop, .open = vmw_driver_open, .postclose = vmw_postclose, + + .dumb_create = vmw_dumb_create, + .dumb_map_offset = vmw_dumb_map_offset, + .dumb_destroy = vmw_dumb_destroy, + .fops = { .owner = THIS_MODULE, .open = drm_open, .release = drm_release, .unlocked_ioctl = vmw_unlocked_ioctl, .mmap = vmw_mmap, - .poll = drm_poll, + .poll = vmw_fops_poll, + .read = vmw_fops_read, .fasync = drm_fasync, #if defined(CONFIG_COMPAT) .compat_ioctl = drm_compat_ioctl, diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h index 10fc01f..0e3fa7d 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h @@ -38,20 +38,27 @@ #include "ttm/ttm_lock.h" #include "ttm/ttm_execbuf_util.h" #include "ttm/ttm_module.h" +#include "vmwgfx_fence.h" -#define VMWGFX_DRIVER_DATE "20100927" -#define VMWGFX_DRIVER_MAJOR 1 -#define VMWGFX_DRIVER_MINOR 4 +#define VMWGFX_DRIVER_DATE "20111025" +#define VMWGFX_DRIVER_MAJOR 2 +#define VMWGFX_DRIVER_MINOR 3 #define VMWGFX_DRIVER_PATCHLEVEL 0 #define VMWGFX_FILE_PAGE_OFFSET 0x00100000 #define VMWGFX_FIFO_STATIC_SIZE (1024*1024) #define VMWGFX_MAX_RELOCATIONS 2048 -#define VMWGFX_MAX_GMRS 2048 +#define VMWGFX_MAX_VALIDATIONS 2048 #define VMWGFX_MAX_DISPLAYS 16 +#define VMWGFX_CMD_BOUNCE_INIT_SIZE 32768 #define VMW_PL_GMR TTM_PL_PRIV0 #define VMW_PL_FLAG_GMR TTM_PL_FLAG_PRIV0 +#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 + struct vmw_fpriv { struct drm_master *locked_master; struct ttm_object_file *tfile; @@ -72,9 +79,11 @@ struct vmw_resource { int id; enum ttm_object_type res_type; bool avail; + void (*remove_from_lists) (struct vmw_resource *res); void (*hw_destroy) (struct vmw_resource *res); void (*res_free) (struct vmw_resource *res); - + struct list_head validate_head; + struct list_head query_head; /* Protected by the cmdbuf mutex */ /* TODO is a generic snooper needed? */ #if 0 void (*snoop)(struct vmw_resource *res, @@ -90,8 +99,12 @@ struct vmw_cursor_snooper { uint32_t *image; }; +struct vmw_framebuffer; +struct vmw_surface_offset; + struct vmw_surface { struct vmw_resource res; + struct list_head lru_head; /* Protected by the resource lock */ uint32_t flags; uint32_t format; uint32_t mip_levels[DRM_VMW_MAX_SURFACE_FACES]; @@ -102,9 +115,12 @@ struct vmw_surface { /* TODO so far just a extra pointer */ struct vmw_cursor_snooper snooper; + struct ttm_buffer_object *backup; + struct vmw_surface_offset *offsets; + uint32_t backup_size; }; -struct vmw_fence_queue { +struct vmw_marker_queue { struct list_head head; struct timespec lag; struct timespec lag_time; @@ -115,16 +131,12 @@ struct vmw_fifo_state { unsigned long reserved_size; __le32 *dynamic_buffer; __le32 *static_buffer; - __le32 *last_buffer; - uint32_t last_data_size; - uint32_t last_buffer_size; - bool last_buffer_add; unsigned long static_buffer_size; bool using_bounce_buffer; uint32_t capabilities; struct mutex fifo_mutex; struct rw_semaphore rwsem; - struct vmw_fence_queue fence_queue; + struct vmw_marker_queue marker_queue; }; struct vmw_relocation { @@ -136,6 +148,8 @@ struct vmw_sw_context{ struct ida bo_list; uint32_t last_cid; bool cid_valid; + bool kernel; /**< is the called made from the kernel */ + struct vmw_resource *cur_ctx; uint32_t last_sid; uint32_t sid_translation; bool sid_valid; @@ -143,8 +157,16 @@ struct vmw_sw_context{ struct list_head validate_nodes; struct vmw_relocation relocs[VMWGFX_MAX_RELOCATIONS]; uint32_t cur_reloc; - struct ttm_validate_buffer val_bufs[VMWGFX_MAX_GMRS]; + struct ttm_validate_buffer val_bufs[VMWGFX_MAX_VALIDATIONS]; uint32_t cur_val_buf; + uint32_t *cmd_bounce; + uint32_t cmd_bounce_size; + struct list_head resource_list; + uint32_t fence_flags; + struct list_head query_list; + struct ttm_buffer_object *cur_query_bo; + uint32_t cur_query_cid; + bool query_cid_valid; }; struct vmw_legacy_display; @@ -185,6 +207,8 @@ struct vmw_private { uint32_t capabilities; uint32_t max_gmr_descriptors; uint32_t max_gmr_ids; + uint32_t max_gmr_pages; + uint32_t memory_size; bool has_gmr; struct mutex hw_mutex; @@ -195,12 +219,7 @@ struct vmw_private { struct vmw_vga_topology_state vga_save[VMWGFX_MAX_DISPLAYS]; uint32_t vga_width; uint32_t vga_height; - uint32_t vga_depth; uint32_t vga_bpp; - uint32_t vga_pseudo; - uint32_t vga_red_mask; - uint32_t vga_green_mask; - uint32_t vga_blue_mask; uint32_t vga_bpl; uint32_t vga_pitchlock; @@ -212,6 +231,7 @@ struct vmw_private { void *fb_info; struct vmw_legacy_display *ldu_priv; + struct vmw_screen_object_display *sou_priv; struct vmw_overlay *overlay_priv; /* @@ -240,13 +260,16 @@ struct vmw_private { * Fencing and IRQs. */ - atomic_t fence_seq; + atomic_t marker_seq; wait_queue_head_t fence_queue; wait_queue_head_t fifo_queue; - atomic_t fence_queue_waiters; + int fence_queue_waiters; /* Protected by hw_mutex */ + int goal_queue_waiters; /* Protected by hw_mutex */ atomic_t fifo_queue_waiters; - uint32_t last_read_sequence; + uint32_t last_read_seqno; spinlock_t irq_lock; + struct vmw_fence_manager *fman; + uint32_t irq_mask; /* * Device state @@ -285,6 +308,26 @@ struct vmw_private { struct mutex release_mutex; uint32_t num_3d_resources; + + /* + * Query processing. These members + * are protected by the cmdbuf mutex. + */ + + struct ttm_buffer_object *dummy_query_bo; + struct ttm_buffer_object *pinned_bo; + uint32_t query_cid; + bool dummy_query_bo_pinned; + + /* + * Surface swapping. The "surface_lru" list is protected by the + * resource lock in order to be able to destroy a surface and take + * it off the lru atomically. "used_memory_size" is currently + * protected by the cmdbuf mutex for simplicity. + */ + + struct list_head surface_lru; + uint32_t used_memory_size; }; static inline struct vmw_private *vmw_priv(struct drm_device *dev) @@ -319,8 +362,8 @@ static inline uint32_t vmw_read(struct vmw_private *dev_priv, return val; } -int vmw_3d_resource_inc(struct vmw_private *dev_priv); -void vmw_3d_resource_dec(struct vmw_private *dev_priv); +int vmw_3d_resource_inc(struct vmw_private *dev_priv, bool unhide_svga); +void vmw_3d_resource_dec(struct vmw_private *dev_priv, bool hide_svga); /** * GMR utilities - vmwgfx_gmr.c @@ -345,7 +388,13 @@ 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); + 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, + struct vmw_surface **out_surf, + struct vmw_dma_buffer **out_buf); extern void vmw_surface_res_free(struct vmw_resource *res); extern int vmw_surface_init(struct vmw_private *dev_priv, struct vmw_surface *srf, @@ -363,6 +412,8 @@ extern int vmw_surface_reference_ioctl(struct drm_device *dev, void *data, extern int vmw_surface_check(struct vmw_private *dev_priv, struct ttm_object_file *tfile, uint32_t handle, int *id); +extern int vmw_surface_validate(struct vmw_private *dev_priv, + struct vmw_surface *srf); extern void vmw_dmabuf_bo_free(struct ttm_buffer_object *bo); extern int vmw_dmabuf_init(struct vmw_private *dev_priv, struct vmw_dma_buffer *vmw_bo, @@ -378,10 +429,6 @@ extern uint32_t vmw_dmabuf_validate_node(struct ttm_buffer_object *bo, extern void vmw_dmabuf_validate_clear(struct ttm_buffer_object *bo); extern int vmw_user_dmabuf_lookup(struct ttm_object_file *tfile, uint32_t id, struct vmw_dma_buffer **out); -extern int vmw_dmabuf_to_start_of_vram(struct vmw_private *vmw_priv, - struct vmw_dma_buffer *bo); -extern int vmw_dmabuf_from_vram(struct vmw_private *vmw_priv, - struct vmw_dma_buffer *bo); extern int vmw_stream_claim_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv); extern int vmw_stream_unref_ioctl(struct drm_device *dev, void *data, @@ -390,7 +437,30 @@ extern int vmw_user_stream_lookup(struct vmw_private *dev_priv, struct ttm_object_file *tfile, uint32_t *inout_id, struct vmw_resource **out); +extern void vmw_resource_unreserve(struct list_head *list); +/** + * DMA buffer helper routines - vmwgfx_dmabuf.c + */ +extern int vmw_dmabuf_to_placement(struct vmw_private *vmw_priv, + struct vmw_dma_buffer *bo, + struct ttm_placement *placement, + bool interruptible); +extern int vmw_dmabuf_to_vram(struct vmw_private *dev_priv, + struct vmw_dma_buffer *buf, + bool pin, bool interruptible); +extern int vmw_dmabuf_to_vram_or_gmr(struct vmw_private *dev_priv, + struct vmw_dma_buffer *buf, + bool pin, bool interruptible); +extern int vmw_dmabuf_to_start_of_vram(struct vmw_private *vmw_priv, + struct vmw_dma_buffer *bo, + bool pin, bool interruptible); +extern int vmw_dmabuf_unpin(struct vmw_private *vmw_priv, + struct vmw_dma_buffer *bo, + bool interruptible); +extern void vmw_bo_get_guest_ptr(const struct ttm_buffer_object *buf, + SVGAGuestPtr *ptr); +extern void vmw_bo_pin(struct ttm_buffer_object *bo, bool pin); /** * Misc Ioctl functionality - vmwgfx_ioctl.c @@ -398,8 +468,16 @@ extern int vmw_user_stream_lookup(struct vmw_private *dev_priv, extern int vmw_getparam_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv); -extern int vmw_fifo_debug_ioctl(struct drm_device *dev, void *data, +extern int vmw_get_cap_3d_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv); +extern int vmw_present_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +extern int vmw_present_readback_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +extern unsigned int vmw_fops_poll(struct file *filp, + struct poll_table_struct *wait); +extern ssize_t vmw_fops_read(struct file *filp, char __user *buffer, + size_t count, loff_t *offset); /** * Fifo utilities - vmwgfx_fifo.c @@ -412,11 +490,12 @@ extern void vmw_fifo_release(struct vmw_private *dev_priv, extern void *vmw_fifo_reserve(struct vmw_private *dev_priv, uint32_t bytes); extern void vmw_fifo_commit(struct vmw_private *dev_priv, uint32_t bytes); extern int vmw_fifo_send_fence(struct vmw_private *dev_priv, - uint32_t *sequence); + uint32_t *seqno); extern void vmw_fifo_ping_host(struct vmw_private *dev_priv, uint32_t reason); -extern int vmw_fifo_mmap(struct file *filp, struct vm_area_struct *vma); extern bool vmw_fifo_have_3d(struct vmw_private *dev_priv); extern bool vmw_fifo_have_pitchlock(struct vmw_private *dev_priv); +extern int vmw_fifo_emit_dummy_query(struct vmw_private *dev_priv, + uint32_t cid); /** * TTM glue - vmwgfx_ttm_glue.c @@ -434,7 +513,10 @@ extern struct ttm_placement vmw_vram_placement; extern struct ttm_placement vmw_vram_ne_placement; extern struct ttm_placement vmw_vram_sys_placement; extern struct ttm_placement vmw_vram_gmr_placement; +extern struct ttm_placement vmw_vram_gmr_ne_placement; extern struct ttm_placement vmw_sys_placement; +extern struct ttm_placement vmw_evictable_placement; +extern struct ttm_placement vmw_srf_placement; extern struct ttm_bo_driver vmw_bo_driver; extern int vmw_dma_quiescent(struct drm_device *dev); @@ -444,45 +526,70 @@ extern int vmw_dma_quiescent(struct drm_device *dev); extern int vmw_execbuf_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv); +extern int vmw_execbuf_process(struct drm_file *file_priv, + struct vmw_private *dev_priv, + void __user *user_commands, + void *kernel_commands, + uint32_t command_size, + uint64_t throttle_us, + struct drm_vmw_fence_rep __user + *user_fence_rep); + +extern void +vmw_execbuf_release_pinned_bo(struct vmw_private *dev_priv, + bool only_on_cid_match, uint32_t cid); + +extern int vmw_execbuf_fence_commands(struct drm_file *file_priv, + struct vmw_private *dev_priv, + struct vmw_fence_obj **p_fence, + uint32_t *p_handle); +extern void vmw_execbuf_copy_fence_user(struct vmw_private *dev_priv, + struct vmw_fpriv *vmw_fp, + int ret, + struct drm_vmw_fence_rep __user + *user_fence_rep, + struct vmw_fence_obj *fence, + uint32_t fence_handle); /** * IRQs and wating - vmwgfx_irq.c */ extern irqreturn_t vmw_irq_handler(DRM_IRQ_ARGS); -extern int vmw_wait_fence(struct vmw_private *dev_priv, bool lazy, - uint32_t sequence, bool interruptible, - unsigned long timeout); +extern int vmw_wait_seqno(struct vmw_private *dev_priv, bool lazy, + uint32_t seqno, bool interruptible, + unsigned long timeout); extern void vmw_irq_preinstall(struct drm_device *dev); extern int vmw_irq_postinstall(struct drm_device *dev); extern void vmw_irq_uninstall(struct drm_device *dev); -extern bool vmw_fence_signaled(struct vmw_private *dev_priv, - uint32_t sequence); -extern int vmw_fence_wait_ioctl(struct drm_device *dev, void *data, - struct drm_file *file_priv); +extern bool vmw_seqno_passed(struct vmw_private *dev_priv, + uint32_t seqno); extern int vmw_fallback_wait(struct vmw_private *dev_priv, bool lazy, bool fifo_idle, - uint32_t sequence, + uint32_t seqno, bool interruptible, unsigned long timeout); -extern void vmw_update_sequence(struct vmw_private *dev_priv, +extern void vmw_update_seqno(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo_state); - +extern void vmw_seqno_waiter_add(struct vmw_private *dev_priv); +extern void vmw_seqno_waiter_remove(struct vmw_private *dev_priv); +extern void vmw_goal_waiter_add(struct vmw_private *dev_priv); +extern void vmw_goal_waiter_remove(struct vmw_private *dev_priv); /** - * Rudimentary fence objects currently used only for throttling - - * vmwgfx_fence.c + * Rudimentary fence-like objects currently used only for throttling - + * vmwgfx_marker.c */ -extern void vmw_fence_queue_init(struct vmw_fence_queue *queue); -extern void vmw_fence_queue_takedown(struct vmw_fence_queue *queue); -extern int vmw_fence_push(struct vmw_fence_queue *queue, - uint32_t sequence); -extern int vmw_fence_pull(struct vmw_fence_queue *queue, - uint32_t signaled_sequence); +extern void vmw_marker_queue_init(struct vmw_marker_queue *queue); +extern void vmw_marker_queue_takedown(struct vmw_marker_queue *queue); +extern int vmw_marker_push(struct vmw_marker_queue *queue, + uint32_t seqno); +extern int vmw_marker_pull(struct vmw_marker_queue *queue, + uint32_t signaled_seqno); extern int vmw_wait_lag(struct vmw_private *dev_priv, - struct vmw_fence_queue *queue, uint32_t us); + struct vmw_marker_queue *queue, uint32_t us); /** * Kernel framebuffer - vmwgfx_fb.c @@ -508,17 +615,42 @@ void vmw_kms_cursor_snoop(struct vmw_surface *srf, struct ttm_object_file *tfile, struct ttm_buffer_object *bo, SVGA3dCmdHeader *header); -void vmw_kms_write_svga(struct vmw_private *vmw_priv, - unsigned width, unsigned height, unsigned pitch, - unsigned bbp, unsigned depth); -int vmw_kms_update_layout_ioctl(struct drm_device *dev, void *data, - struct drm_file *file_priv); +int vmw_kms_write_svga(struct vmw_private *vmw_priv, + unsigned width, unsigned height, unsigned pitch, + unsigned bpp, unsigned depth); void vmw_kms_idle_workqueues(struct vmw_master *vmaster); bool vmw_kms_validate_mode_vram(struct vmw_private *dev_priv, uint32_t pitch, uint32_t height); u32 vmw_get_vblank_counter(struct drm_device *dev, int crtc); +int vmw_enable_vblank(struct drm_device *dev, int crtc); +void vmw_disable_vblank(struct drm_device *dev, int crtc); +int vmw_kms_present(struct vmw_private *dev_priv, + struct drm_file *file_priv, + struct vmw_framebuffer *vfb, + struct vmw_surface *surface, + uint32_t sid, int32_t destX, int32_t destY, + struct drm_vmw_rect *clips, + uint32_t num_clips); +int vmw_kms_readback(struct vmw_private *dev_priv, + struct drm_file *file_priv, + struct vmw_framebuffer *vfb, + struct drm_vmw_fence_rep __user *user_fence_rep, + struct drm_vmw_rect *clips, + uint32_t num_clips); +int vmw_kms_update_layout_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +int vmw_dumb_create(struct drm_file *file_priv, + struct drm_device *dev, + struct drm_mode_create_dumb *args); + +int vmw_dumb_map_offset(struct drm_file *file_priv, + struct drm_device *dev, uint32_t handle, + uint64_t *offset); +int vmw_dumb_destroy(struct drm_file *file_priv, + struct drm_device *dev, + uint32_t handle); /** * Overlay control - vmwgfx_overlay.c */ @@ -576,4 +708,8 @@ static inline struct vmw_dma_buffer *vmw_dmabuf_reference(struct vmw_dma_buffer return NULL; } +static inline struct ttm_mem_global *vmw_mem_glob(struct vmw_private *dev_priv) +{ + return (struct ttm_mem_global *) dev_priv->mem_global_ref.object; +} #endif diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c index 41b95ed..84ba033 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c @@ -44,10 +44,71 @@ static int vmw_cmd_ok(struct vmw_private *dev_priv, return 0; } +static void vmw_resource_to_validate_list(struct vmw_sw_context *sw_context, + struct vmw_resource **p_res) +{ + struct vmw_resource *res = *p_res; + + if (list_empty(&res->validate_head)) { + list_add_tail(&res->validate_head, &sw_context->resource_list); + *p_res = NULL; + } else + vmw_resource_unreference(p_res); +} + +/** + * vmw_bo_to_validate_list - add a bo to a validate list + * + * @sw_context: The software context used for this command submission batch. + * @bo: The buffer object to add. + * @fence_flags: Fence flags to be or'ed with any other fence flags for + * this buffer on this submission batch. + * @p_val_node: If non-NULL Will be updated with the validate node number + * on return. + * + * Returns -EINVAL if the limit of number of buffer objects per command + * submission is reached. + */ +static int vmw_bo_to_validate_list(struct vmw_sw_context *sw_context, + struct ttm_buffer_object *bo, + uint32_t fence_flags, + uint32_t *p_val_node) +{ + uint32_t val_node; + struct ttm_validate_buffer *val_buf; + + val_node = vmw_dmabuf_validate_node(bo, sw_context->cur_val_buf); + + if (unlikely(val_node >= VMWGFX_MAX_VALIDATIONS)) { + DRM_ERROR("Max number of DMA buffers per submission" + " exceeded.\n"); + return -EINVAL; + } + + val_buf = &sw_context->val_bufs[val_node]; + if (unlikely(val_node == sw_context->cur_val_buf)) { + val_buf->new_sync_obj_arg = NULL; + val_buf->bo = ttm_bo_reference(bo); + list_add_tail(&val_buf->head, &sw_context->validate_nodes); + ++sw_context->cur_val_buf; + } + + val_buf->new_sync_obj_arg = (void *) + ((unsigned long) val_buf->new_sync_obj_arg | fence_flags); + sw_context->fence_flags |= fence_flags; + + if (p_val_node) + *p_val_node = val_node; + + return 0; +} + static int vmw_cmd_cid_check(struct vmw_private *dev_priv, struct vmw_sw_context *sw_context, SVGA3dCmdHeader *header) { + struct vmw_resource *ctx; + struct vmw_cid_cmd { SVGA3dCmdHeader header; __le32 cid; @@ -58,7 +119,8 @@ static int vmw_cmd_cid_check(struct vmw_private *dev_priv, if (likely(sw_context->cid_valid && cmd->cid == sw_context->last_cid)) return 0; - ret = vmw_context_check(dev_priv, sw_context->tfile, cmd->cid); + ret = vmw_context_check(dev_priv, sw_context->tfile, cmd->cid, + &ctx); if (unlikely(ret != 0)) { DRM_ERROR("Could not find or use context %u\n", (unsigned) cmd->cid); @@ -67,6 +129,8 @@ static int vmw_cmd_cid_check(struct vmw_private *dev_priv, sw_context->last_cid = cmd->cid; sw_context->cid_valid = true; + sw_context->cur_ctx = ctx; + vmw_resource_to_validate_list(sw_context, &ctx); return 0; } @@ -75,29 +139,45 @@ static int vmw_cmd_sid_check(struct vmw_private *dev_priv, struct vmw_sw_context *sw_context, uint32_t *sid) { + struct vmw_surface *srf; + int ret; + struct vmw_resource *res; + if (*sid == SVGA3D_INVALID_ID) return 0; - if (unlikely((!sw_context->sid_valid || - *sid != sw_context->last_sid))) { - int real_id; - int ret = vmw_surface_check(dev_priv, sw_context->tfile, - *sid, &real_id); + if (likely((sw_context->sid_valid && + *sid == sw_context->last_sid))) { + *sid = sw_context->sid_translation; + return 0; + } - if (unlikely(ret != 0)) { - DRM_ERROR("Could ot find or use surface 0x%08x " - "address 0x%08lx\n", - (unsigned int) *sid, - (unsigned long) sid); - return ret; - } + ret = vmw_user_surface_lookup_handle(dev_priv, + sw_context->tfile, + *sid, &srf); + if (unlikely(ret != 0)) { + DRM_ERROR("Could ot find or use surface 0x%08x " + "address 0x%08lx\n", + (unsigned int) *sid, + (unsigned long) sid); + return ret; + } - sw_context->last_sid = *sid; - sw_context->sid_valid = true; - *sid = real_id; - sw_context->sid_translation = real_id; - } else - *sid = sw_context->sid_translation; + ret = vmw_surface_validate(dev_priv, srf); + if (unlikely(ret != 0)) { + if (ret != -ERESTARTSYS) + DRM_ERROR("Could not validate surface.\n"); + vmw_surface_unreference(&srf); + return ret; + } + + sw_context->last_sid = *sid; + sw_context->sid_valid = true; + sw_context->sid_translation = srf->res.id; + *sid = sw_context->sid_translation; + + res = &srf->res; + vmw_resource_to_validate_list(sw_context, &res); return 0; } @@ -166,6 +246,12 @@ static int vmw_cmd_blt_surf_screen_check(struct vmw_private *dev_priv, } *cmd; 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_sid_check(dev_priv, sw_context, &cmd->body.srcImage.sid); } @@ -178,10 +264,179 @@ static int vmw_cmd_present_check(struct vmw_private *dev_priv, SVGA3dCmdPresent body; } *cmd; + 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_sid_check(dev_priv, sw_context, &cmd->body.sid); } +/** + * vmw_query_bo_switch_prepare - Prepare to switch pinned buffer for queries. + * + * @dev_priv: The device private structure. + * @cid: The hardware context for the next query. + * @new_query_bo: The new buffer holding query results. + * @sw_context: The software context used for this command submission. + * + * This function checks whether @new_query_bo is suitable for holding + * query results, and if another buffer currently is pinned for query + * results. If so, the function prepares the state of @sw_context for + * switching pinned buffers after successful submission of the current + * command batch. It also checks whether we're using a new query context. + * In that case, it makes sure we emit a query barrier for the old + * context before the current query buffer is fenced. + */ +static int vmw_query_bo_switch_prepare(struct vmw_private *dev_priv, + uint32_t cid, + struct ttm_buffer_object *new_query_bo, + struct vmw_sw_context *sw_context) +{ + int ret; + bool add_cid = false; + uint32_t cid_to_add; + + if (unlikely(new_query_bo != sw_context->cur_query_bo)) { + + if (unlikely(new_query_bo->num_pages > 4)) { + DRM_ERROR("Query buffer too large.\n"); + return -EINVAL; + } + + if (unlikely(sw_context->cur_query_bo != NULL)) { + BUG_ON(!sw_context->query_cid_valid); + add_cid = true; + cid_to_add = sw_context->cur_query_cid; + ret = vmw_bo_to_validate_list(sw_context, + sw_context->cur_query_bo, + DRM_VMW_FENCE_FLAG_EXEC, + NULL); + if (unlikely(ret != 0)) + return ret; + } + sw_context->cur_query_bo = new_query_bo; + + ret = vmw_bo_to_validate_list(sw_context, + dev_priv->dummy_query_bo, + DRM_VMW_FENCE_FLAG_EXEC, + NULL); + if (unlikely(ret != 0)) + return ret; + + } + + if (unlikely(cid != sw_context->cur_query_cid && + sw_context->query_cid_valid)) { + add_cid = true; + cid_to_add = sw_context->cur_query_cid; + } + + sw_context->cur_query_cid = cid; + sw_context->query_cid_valid = true; + + if (add_cid) { + struct vmw_resource *ctx = sw_context->cur_ctx; + + if (list_empty(&ctx->query_head)) + list_add_tail(&ctx->query_head, + &sw_context->query_list); + ret = vmw_bo_to_validate_list(sw_context, + dev_priv->dummy_query_bo, + DRM_VMW_FENCE_FLAG_EXEC, + NULL); + if (unlikely(ret != 0)) + return ret; + } + return 0; +} + + +/** + * vmw_query_bo_switch_commit - Finalize switching pinned query buffer + * + * @dev_priv: The device private structure. + * @sw_context: The software context used for this command submission batch. + * + * This function will check if we're switching query buffers, and will then, + * if no other query waits are issued this command submission batch, + * issue a dummy occlusion query wait used as a query barrier. When the fence + * object following that query wait has signaled, we are sure that all + * preseding queries have finished, and the old query buffer can be unpinned. + * However, since both the new query buffer and the old one are fenced with + * that fence, we can do an asynchronus unpin now, and be sure that the + * old query buffer won't be moved until the fence has signaled. + * + * As mentioned above, both the new - and old query buffers need to be fenced + * using a sequence emitted *after* calling this function. + */ +static void vmw_query_bo_switch_commit(struct vmw_private *dev_priv, + struct vmw_sw_context *sw_context) +{ + + struct vmw_resource *ctx, *next_ctx; + int ret; + + /* + * The validate list should still hold references to all + * contexts here. + */ + + list_for_each_entry_safe(ctx, next_ctx, &sw_context->query_list, + query_head) { + list_del_init(&ctx->query_head); + + BUG_ON(list_empty(&ctx->validate_head)); + + ret = vmw_fifo_emit_dummy_query(dev_priv, ctx->id); + + if (unlikely(ret != 0)) + DRM_ERROR("Out of fifo space for dummy query.\n"); + } + + if (dev_priv->pinned_bo != sw_context->cur_query_bo) { + if (dev_priv->pinned_bo) { + vmw_bo_pin(dev_priv->pinned_bo, false); + ttm_bo_unref(&dev_priv->pinned_bo); + } + + vmw_bo_pin(sw_context->cur_query_bo, true); + + /* + * We pin also the dummy_query_bo buffer so that we + * don't need to validate it when emitting + * dummy queries in context destroy paths. + */ + + vmw_bo_pin(dev_priv->dummy_query_bo, true); + dev_priv->dummy_query_bo_pinned = true; + + dev_priv->query_cid = sw_context->cur_query_cid; + dev_priv->pinned_bo = + ttm_bo_reference(sw_context->cur_query_bo); + } +} + +/** + * vmw_query_switch_backoff - clear query barrier list + * @sw_context: The sw context used for this submission batch. + * + * This function is used as part of an error path, where a previously + * set up list of query barriers needs to be cleared. + * + */ +static void vmw_query_switch_backoff(struct vmw_sw_context *sw_context) +{ + struct list_head *list, *next; + + list_for_each_safe(list, next, &sw_context->query_list) { + list_del_init(list); + } +} + static int vmw_translate_guest_ptr(struct vmw_private *dev_priv, struct vmw_sw_context *sw_context, SVGAGuestPtr *ptr, @@ -191,8 +446,6 @@ static int vmw_translate_guest_ptr(struct vmw_private *dev_priv, struct ttm_buffer_object *bo; uint32_t handle = ptr->gmrId; struct vmw_relocation *reloc; - uint32_t cur_validate_node; - struct ttm_validate_buffer *val_buf; int ret; ret = vmw_user_dmabuf_lookup(sw_context->tfile, handle, &vmw_bo); @@ -212,22 +465,11 @@ static int vmw_translate_guest_ptr(struct vmw_private *dev_priv, reloc = &sw_context->relocs[sw_context->cur_reloc++]; reloc->location = ptr; - cur_validate_node = vmw_dmabuf_validate_node(bo, sw_context->cur_val_buf); - if (unlikely(cur_validate_node >= VMWGFX_MAX_GMRS)) { - DRM_ERROR("Max number of DMA buffers per submission" - " exceeded.\n"); - ret = -EINVAL; + ret = vmw_bo_to_validate_list(sw_context, bo, DRM_VMW_FENCE_FLAG_EXEC, + &reloc->index); + if (unlikely(ret != 0)) goto out_no_reloc; - } - reloc->index = cur_validate_node; - if (unlikely(cur_validate_node == sw_context->cur_val_buf)) { - val_buf = &sw_context->val_bufs[cur_validate_node]; - val_buf->bo = ttm_bo_reference(bo); - val_buf->new_sync_obj_arg = (void *) dev_priv; - list_add_tail(&val_buf->head, &sw_context->validate_nodes); - ++sw_context->cur_val_buf; - } *vmw_bo_p = vmw_bo; return 0; @@ -259,8 +501,11 @@ static int vmw_cmd_end_query(struct vmw_private *dev_priv, if (unlikely(ret != 0)) return ret; + ret = vmw_query_bo_switch_prepare(dev_priv, cmd->q.cid, + &vmw_bo->base, sw_context); + vmw_dmabuf_unreference(&vmw_bo); - return 0; + return ret; } static int vmw_cmd_wait_query(struct vmw_private *dev_priv, @@ -273,6 +518,7 @@ static int vmw_cmd_wait_query(struct vmw_private *dev_priv, SVGA3dCmdWaitForQuery q; } *cmd; int ret; + struct vmw_resource *ctx; cmd = container_of(header, struct vmw_query_cmd, header); ret = vmw_cmd_cid_check(dev_priv, sw_context, header); @@ -286,10 +532,19 @@ static int vmw_cmd_wait_query(struct vmw_private *dev_priv, return ret; vmw_dmabuf_unreference(&vmw_bo); + + /* + * This wait will act as a barrier for previous waits for this + * context. + */ + + ctx = sw_context->cur_ctx; + if (!list_empty(&ctx->query_head)) + list_del_init(&ctx->query_head); + return 0; } - static int vmw_cmd_dma(struct vmw_private *dev_priv, struct vmw_sw_context *sw_context, SVGA3dCmdHeader *header) @@ -302,14 +557,37 @@ static int vmw_cmd_dma(struct vmw_private *dev_priv, SVGA3dCmdSurfaceDMA dma; } *cmd; int ret; + struct vmw_resource *res; + 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; + bo = &vmw_bo->base; ret = vmw_user_surface_lookup_handle(dev_priv, sw_context->tfile, cmd->dma.host.sid, &srf); @@ -318,18 +596,28 @@ static int vmw_cmd_dma(struct vmw_private *dev_priv, goto out_no_reloc; } - /** + ret = vmw_surface_validate(dev_priv, srf); + if (unlikely(ret != 0)) { + if (ret != -ERESTARTSYS) + DRM_ERROR("Culd not validate surface.\n"); + goto out_no_validate; + } + + /* * Patch command stream with device SID. */ - cmd->dma.host.sid = srf->res.id; vmw_kms_cursor_snoop(srf, sw_context->tfile, bo, header); - /** - * FIXME: May deadlock here when called from the - * command parsing code. - */ - vmw_surface_unreference(&srf); + vmw_dmabuf_unreference(&vmw_bo); + + res = &srf->res; + vmw_resource_to_validate_list(sw_context, &res); + + return 0; + +out_no_validate: + vmw_surface_unreference(&srf); out_no_reloc: vmw_dmabuf_unreference(&vmw_bo); return ret; @@ -419,6 +707,71 @@ static int vmw_cmd_tex_state(struct vmw_private *dev_priv, return 0; } +static int vmw_cmd_check_define_gmrfb(struct vmw_private *dev_priv, + struct vmw_sw_context *sw_context, + void *buf) +{ + struct vmw_dma_buffer *vmw_bo; + int ret; + + struct { + uint32_t header; + SVGAFifoCmdDefineGMRFB body; + } *cmd = buf; + + ret = vmw_translate_guest_ptr(dev_priv, sw_context, + &cmd->body.ptr, + &vmw_bo); + if (unlikely(ret != 0)) + return ret; + + vmw_dmabuf_unreference(&vmw_bo); + + return ret; +} + +static int vmw_cmd_check_not_3d(struct vmw_private *dev_priv, + struct vmw_sw_context *sw_context, + void *buf, uint32_t *size) +{ + uint32_t size_remaining = *size; + uint32_t cmd_id; + + cmd_id = le32_to_cpu(((uint32_t *)buf)[0]); + switch (cmd_id) { + case SVGA_CMD_UPDATE: + *size = sizeof(uint32_t) + sizeof(SVGAFifoCmdUpdate); + break; + case SVGA_CMD_DEFINE_GMRFB: + *size = sizeof(uint32_t) + sizeof(SVGAFifoCmdDefineGMRFB); + break; + case SVGA_CMD_BLIT_GMRFB_TO_SCREEN: + *size = sizeof(uint32_t) + sizeof(SVGAFifoCmdBlitGMRFBToScreen); + break; + case SVGA_CMD_BLIT_SCREEN_TO_GMRFB: + *size = sizeof(uint32_t) + sizeof(SVGAFifoCmdBlitGMRFBToScreen); + break; + default: + DRM_ERROR("Unsupported SVGA command: %u.\n", cmd_id); + return -EINVAL; + } + + if (*size > size_remaining) { + DRM_ERROR("Invalid SVGA command (size mismatch):" + " %u.\n", cmd_id); + return -EINVAL; + } + + if (unlikely(!sw_context->kernel)) { + DRM_ERROR("Kernel only SVGA command: %u.\n", cmd_id); + return -EPERM; + } + + if (cmd_id == SVGA_CMD_DEFINE_GMRFB) + return vmw_cmd_check_define_gmrfb(dev_priv, sw_context, buf); + + return 0; +} typedef int (*vmw_cmd_func) (struct vmw_private *, struct vmw_sw_context *, @@ -471,11 +824,11 @@ static int vmw_cmd_check(struct vmw_private *dev_priv, SVGA3dCmdHeader *header = (SVGA3dCmdHeader *) buf; int ret; - cmd_id = ((uint32_t *)buf)[0]; - if (cmd_id == SVGA_CMD_UPDATE) { - *size = 5 << 2; - return 0; - } + cmd_id = le32_to_cpu(((uint32_t *)buf)[0]); + /* Handle any none 3D commands */ + if (unlikely(cmd_id < SVGA_CMD_MAX)) + return vmw_cmd_check_not_3d(dev_priv, sw_context, buf, size); + cmd_id = le32_to_cpu(header->id); *size = le32_to_cpu(header->size) + sizeof(SVGA3dCmdHeader); @@ -500,7 +853,8 @@ out_err: static int vmw_cmd_check_all(struct vmw_private *dev_priv, struct vmw_sw_context *sw_context, - void *buf, uint32_t size) + void *buf, + uint32_t size) { int32_t cur_size = size; int ret; @@ -550,7 +904,11 @@ static void vmw_apply_relocations(struct vmw_sw_context *sw_context) static void vmw_clear_validations(struct vmw_sw_context *sw_context) { struct ttm_validate_buffer *entry, *next; + struct vmw_resource *res, *res_next; + /* + * Drop references to DMA buffers held during command submission. + */ list_for_each_entry_safe(entry, next, &sw_context->validate_nodes, head) { list_del(&entry->head); @@ -559,6 +917,16 @@ static void vmw_clear_validations(struct vmw_sw_context *sw_context) sw_context->cur_val_buf--; } BUG_ON(sw_context->cur_val_buf != 0); + + /* + * Drop references to resources held during command submission. + */ + vmw_resource_unreserve(&sw_context->resource_list); + list_for_each_entry_safe(res, res_next, &sw_context->resource_list, + validate_head) { + list_del_init(&res->validate_head); + vmw_resource_unreference(&res); + } } static int vmw_validate_single_buffer(struct vmw_private *dev_priv, @@ -566,6 +934,16 @@ static int vmw_validate_single_buffer(struct vmw_private *dev_priv, { int ret; + + /* + * Don't validate pinned buffers. + */ + + if (bo == dev_priv->pinned_bo || + (bo == dev_priv->dummy_query_bo && + dev_priv->dummy_query_bo_pinned)) + return 0; + /** * 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, @@ -602,57 +980,208 @@ static int vmw_validate_buffers(struct vmw_private *dev_priv, return 0; } -int vmw_execbuf_ioctl(struct drm_device *dev, void *data, - struct drm_file *file_priv) +static int vmw_resize_cmd_bounce(struct vmw_sw_context *sw_context, + uint32_t size) +{ + if (likely(sw_context->cmd_bounce_size >= size)) + return 0; + + if (sw_context->cmd_bounce_size == 0) + sw_context->cmd_bounce_size = VMWGFX_CMD_BOUNCE_INIT_SIZE; + + while (sw_context->cmd_bounce_size < size) { + sw_context->cmd_bounce_size = + PAGE_ALIGN(sw_context->cmd_bounce_size + + (sw_context->cmd_bounce_size >> 1)); + } + + if (sw_context->cmd_bounce != NULL) + vfree(sw_context->cmd_bounce); + + sw_context->cmd_bounce = vmalloc(sw_context->cmd_bounce_size); + + if (sw_context->cmd_bounce == NULL) { + DRM_ERROR("Failed to allocate command bounce buffer.\n"); + sw_context->cmd_bounce_size = 0; + return -ENOMEM; + } + + return 0; +} + +/** + * vmw_execbuf_fence_commands - create and submit a command stream fence + * + * Creates a fence object and submits a command stream marker. + * If this fails for some reason, We sync the fifo and return NULL. + * It is then safe to fence buffers with a NULL pointer. + * + * If @p_handle is not NULL @file_priv must also not be NULL. Creates + * a userspace handle if @p_handle is not NULL, otherwise not. + */ + +int vmw_execbuf_fence_commands(struct drm_file *file_priv, + struct vmw_private *dev_priv, + struct vmw_fence_obj **p_fence, + uint32_t *p_handle) { - struct vmw_private *dev_priv = vmw_priv(dev); - struct drm_vmw_execbuf_arg *arg = (struct drm_vmw_execbuf_arg *)data; - struct drm_vmw_fence_rep fence_rep; - struct drm_vmw_fence_rep __user *user_fence_rep; - int ret; - void *user_cmd; - void *cmd; uint32_t sequence; - struct vmw_sw_context *sw_context = &dev_priv->ctx; - struct vmw_master *vmaster = vmw_master(file_priv->master); + int ret; + bool synced = false; - ret = ttm_read_lock(&vmaster->lock, true); - if (unlikely(ret != 0)) - return ret; + /* p_handle implies file_priv. */ + BUG_ON(p_handle != NULL && file_priv == NULL); - ret = mutex_lock_interruptible(&dev_priv->cmdbuf_mutex); + ret = vmw_fifo_send_fence(dev_priv, &sequence); if (unlikely(ret != 0)) { - ret = -ERESTARTSYS; - goto out_no_cmd_mutex; + DRM_ERROR("Fence submission error. Syncing.\n"); + synced = true; } - cmd = vmw_fifo_reserve(dev_priv, arg->command_size); - if (unlikely(cmd == NULL)) { - DRM_ERROR("Failed reserving fifo space for commands.\n"); - ret = -ENOMEM; - goto out_unlock; + if (p_handle != NULL) + ret = vmw_user_fence_create(file_priv, dev_priv->fman, + sequence, + DRM_VMW_FENCE_FLAG_EXEC, + p_fence, p_handle); + else + ret = vmw_fence_create(dev_priv->fman, sequence, + DRM_VMW_FENCE_FLAG_EXEC, + p_fence); + + if (unlikely(ret != 0 && !synced)) { + (void) vmw_fallback_wait(dev_priv, false, false, + sequence, false, + VMW_FENCE_WAIT_TIMEOUT); + *p_fence = NULL; } - user_cmd = (void __user *)(unsigned long)arg->commands; - ret = copy_from_user(cmd, user_cmd, arg->command_size); + return 0; +} + +/** + * vmw_execbuf_copy_fence_user - copy fence object information to + * user-space. + * + * @dev_priv: Pointer to a vmw_private struct. + * @vmw_fp: Pointer to the struct vmw_fpriv representing the calling file. + * @ret: Return value from fence object creation. + * @user_fence_rep: User space address of a struct drm_vmw_fence_rep to + * which the information should be copied. + * @fence: Pointer to the fenc object. + * @fence_handle: User-space fence handle. + * + * This function copies fence information to user-space. If copying fails, + * The user-space struct drm_vmw_fence_rep::error member is hopefully + * left untouched, and if it's preloaded with an -EFAULT by user-space, + * the error will hopefully be detected. + * Also if copying fails, user-space will be unable to signal the fence + * object so we wait for it immediately, and then unreference the + * user-space reference. + */ +void +vmw_execbuf_copy_fence_user(struct vmw_private *dev_priv, + struct vmw_fpriv *vmw_fp, + int ret, + struct drm_vmw_fence_rep __user *user_fence_rep, + struct vmw_fence_obj *fence, + uint32_t fence_handle) +{ + struct drm_vmw_fence_rep fence_rep; - if (unlikely(ret != 0)) { - ret = -EFAULT; - DRM_ERROR("Failed copying commands.\n"); - goto out_commit; + if (user_fence_rep == NULL) + return; + + memset(&fence_rep, 0, sizeof(fence_rep)); + + fence_rep.error = ret; + if (ret == 0) { + BUG_ON(fence == NULL); + + fence_rep.handle = fence_handle; + fence_rep.seqno = fence->seqno; + vmw_update_seqno(dev_priv, &dev_priv->fifo); + fence_rep.passed_seqno = dev_priv->last_read_seqno; + } + + /* + * copy_to_user errors will be detected by user space not + * seeing fence_rep::error filled in. Typically + * user-space would have pre-set that member to -EFAULT. + */ + ret = copy_to_user(user_fence_rep, &fence_rep, + sizeof(fence_rep)); + + /* + * User-space lost the fence object. We need to sync + * and unreference the handle. + */ + if (unlikely(ret != 0) && (fence_rep.error == 0)) { + ttm_ref_object_base_unref(vmw_fp->tfile, + fence_handle, TTM_REF_USAGE); + DRM_ERROR("Fence copy error. Syncing.\n"); + (void) vmw_fence_obj_wait(fence, fence->signal_mask, + false, false, + VMW_FENCE_WAIT_TIMEOUT); } +} + +int vmw_execbuf_process(struct drm_file *file_priv, + struct vmw_private *dev_priv, + void __user *user_commands, + void *kernel_commands, + uint32_t command_size, + uint64_t throttle_us, + struct drm_vmw_fence_rep __user *user_fence_rep) +{ + struct vmw_sw_context *sw_context = &dev_priv->ctx; + struct vmw_fence_obj *fence; + uint32_t handle; + void *cmd; + int ret; + + ret = mutex_lock_interruptible(&dev_priv->cmdbuf_mutex); + if (unlikely(ret != 0)) + return -ERESTARTSYS; + + if (kernel_commands == NULL) { + sw_context->kernel = false; + + ret = vmw_resize_cmd_bounce(sw_context, command_size); + if (unlikely(ret != 0)) + goto out_unlock; + + + ret = copy_from_user(sw_context->cmd_bounce, + user_commands, command_size); + + if (unlikely(ret != 0)) { + ret = -EFAULT; + DRM_ERROR("Failed copying commands.\n"); + goto out_unlock; + } + kernel_commands = sw_context->cmd_bounce; + } else + sw_context->kernel = true; sw_context->tfile = vmw_fpriv(file_priv)->tfile; sw_context->cid_valid = false; sw_context->sid_valid = false; sw_context->cur_reloc = 0; sw_context->cur_val_buf = 0; + sw_context->fence_flags = 0; + INIT_LIST_HEAD(&sw_context->query_list); + INIT_LIST_HEAD(&sw_context->resource_list); + sw_context->cur_query_bo = dev_priv->pinned_bo; + sw_context->cur_query_cid = dev_priv->query_cid; + sw_context->query_cid_valid = (dev_priv->pinned_bo != NULL); INIT_LIST_HEAD(&sw_context->validate_nodes); - ret = vmw_cmd_check_all(dev_priv, sw_context, cmd, arg->command_size); + ret = vmw_cmd_check_all(dev_priv, sw_context, kernel_commands, + command_size); if (unlikely(ret != 0)) goto out_err; + ret = ttm_eu_reserve_buffers(&sw_context->validate_nodes); if (unlikely(ret != 0)) goto out_err; @@ -663,57 +1192,206 @@ int vmw_execbuf_ioctl(struct drm_device *dev, void *data, vmw_apply_relocations(sw_context); - if (arg->throttle_us) { - ret = vmw_wait_lag(dev_priv, &dev_priv->fifo.fence_queue, - arg->throttle_us); + if (throttle_us) { + ret = vmw_wait_lag(dev_priv, &dev_priv->fifo.marker_queue, + throttle_us); if (unlikely(ret != 0)) - goto out_err; + goto out_throttle; } - vmw_fifo_commit(dev_priv, arg->command_size); + 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_throttle; + } - ret = vmw_fifo_send_fence(dev_priv, &sequence); - - ttm_eu_fence_buffer_objects(&sw_context->validate_nodes, - (void *)(unsigned long) sequence); - vmw_clear_validations(sw_context); - mutex_unlock(&dev_priv->cmdbuf_mutex); + memcpy(cmd, kernel_commands, command_size); + vmw_fifo_commit(dev_priv, command_size); + vmw_query_bo_switch_commit(dev_priv, sw_context); + ret = vmw_execbuf_fence_commands(file_priv, dev_priv, + &fence, + (user_fence_rep) ? &handle : NULL); /* * This error is harmless, because if fence submission fails, - * vmw_fifo_send_fence will sync. + * vmw_fifo_send_fence will sync. The error will be propagated to + * user-space in @fence_rep */ if (ret != 0) DRM_ERROR("Fence submission error. Syncing.\n"); - fence_rep.error = ret; - fence_rep.fence_seq = (uint64_t) sequence; - fence_rep.pad64 = 0; - - user_fence_rep = (struct drm_vmw_fence_rep __user *) - (unsigned long)arg->fence_rep; + ttm_eu_fence_buffer_objects(&sw_context->validate_nodes, + (void *) fence); - /* - * copy_to_user errors will be detected by user space not - * seeing fence_rep::error filled in. - */ + vmw_clear_validations(sw_context); + vmw_execbuf_copy_fence_user(dev_priv, vmw_fpriv(file_priv), ret, + user_fence_rep, fence, handle); - ret = copy_to_user(user_fence_rep, &fence_rep, sizeof(fence_rep)); + if (likely(fence != NULL)) + vmw_fence_obj_unreference(&fence); - vmw_kms_cursor_post_execbuf(dev_priv); - ttm_read_unlock(&vmaster->lock); + mutex_unlock(&dev_priv->cmdbuf_mutex); return 0; + out_err: vmw_free_relocations(sw_context); +out_throttle: + vmw_query_switch_backoff(sw_context); ttm_eu_backoff_reservation(&sw_context->validate_nodes); vmw_clear_validations(sw_context); -out_commit: - vmw_fifo_commit(dev_priv, 0); out_unlock: mutex_unlock(&dev_priv->cmdbuf_mutex); -out_no_cmd_mutex: + return ret; +} + +/** + * vmw_execbuf_unpin_panic - Idle the fifo and unpin the query buffer. + * + * @dev_priv: The device private structure. + * + * This function is called to idle the fifo and unpin the query buffer + * if the normal way to do this hits an error, which should typically be + * extremely rare. + */ +static void vmw_execbuf_unpin_panic(struct vmw_private *dev_priv) +{ + DRM_ERROR("Can't unpin query buffer. Trying to recover.\n"); + + (void) vmw_fallback_wait(dev_priv, false, true, 0, false, 10*HZ); + vmw_bo_pin(dev_priv->pinned_bo, false); + vmw_bo_pin(dev_priv->dummy_query_bo, false); + dev_priv->dummy_query_bo_pinned = false; +} + + +/** + * vmw_execbuf_release_pinned_bo - Flush queries and unpin the pinned + * query bo. + * + * @dev_priv: The device private structure. + * @only_on_cid_match: Only flush and unpin if the current active query cid + * matches @cid. + * @cid: Optional context id to match. + * + * This function should be used to unpin the pinned query bo, or + * as a query barrier when we need to make sure that all queries have + * finished before the next fifo command. (For example on hardware + * context destructions where the hardware may otherwise leak unfinished + * queries). + * + * This function does not return any failure codes, but make attempts + * to do safe unpinning in case of errors. + * + * The function will synchronize on the previous query barrier, and will + * thus not finish until that barrier has executed. + */ +void vmw_execbuf_release_pinned_bo(struct vmw_private *dev_priv, + bool only_on_cid_match, uint32_t cid) +{ + int ret = 0; + struct list_head validate_list; + struct ttm_validate_buffer pinned_val, query_val; + struct vmw_fence_obj *fence; + + mutex_lock(&dev_priv->cmdbuf_mutex); + + if (dev_priv->pinned_bo == NULL) + goto out_unlock; + + if (only_on_cid_match && cid != dev_priv->query_cid) + goto out_unlock; + + INIT_LIST_HEAD(&validate_list); + + pinned_val.new_sync_obj_arg = (void *)(unsigned long) + DRM_VMW_FENCE_FLAG_EXEC; + pinned_val.bo = ttm_bo_reference(dev_priv->pinned_bo); + list_add_tail(&pinned_val.head, &validate_list); + + query_val.new_sync_obj_arg = pinned_val.new_sync_obj_arg; + query_val.bo = ttm_bo_reference(dev_priv->dummy_query_bo); + list_add_tail(&query_val.head, &validate_list); + + do { + ret = ttm_eu_reserve_buffers(&validate_list); + } while (ret == -ERESTARTSYS); + + if (unlikely(ret != 0)) { + vmw_execbuf_unpin_panic(dev_priv); + goto out_no_reserve; + } + + ret = vmw_fifo_emit_dummy_query(dev_priv, dev_priv->query_cid); + if (unlikely(ret != 0)) { + vmw_execbuf_unpin_panic(dev_priv); + goto out_no_emit; + } + + vmw_bo_pin(dev_priv->pinned_bo, false); + vmw_bo_pin(dev_priv->dummy_query_bo, false); + dev_priv->dummy_query_bo_pinned = false; + + (void) vmw_execbuf_fence_commands(NULL, dev_priv, &fence, NULL); + ttm_eu_fence_buffer_objects(&validate_list, (void *) fence); + + ttm_bo_unref(&query_val.bo); + ttm_bo_unref(&pinned_val.bo); + ttm_bo_unref(&dev_priv->pinned_bo); + +out_unlock: + mutex_unlock(&dev_priv->cmdbuf_mutex); + return; + +out_no_emit: + ttm_eu_backoff_reservation(&validate_list); +out_no_reserve: + ttm_bo_unref(&query_val.bo); + ttm_bo_unref(&pinned_val.bo); + ttm_bo_unref(&dev_priv->pinned_bo); + mutex_unlock(&dev_priv->cmdbuf_mutex); +} + + +int vmw_execbuf_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + 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; + + /* + * This will allow us to extend the ioctl argument while + * maintaining backwards compatibility: + * We take different code paths depending on the value of + * arg->version. + */ + + if (unlikely(arg->version != DRM_VMW_EXECBUF_VERSION)) { + DRM_ERROR("Incorrect execbuf version.\n"); + DRM_ERROR("You're running outdated experimental " + "vmwgfx user-space drivers."); + return -EINVAL; + } + + ret = ttm_read_lock(&vmaster->lock, true); + if (unlikely(ret != 0)) + return ret; + + ret = vmw_execbuf_process(file_priv, dev_priv, + (void __user *)(unsigned long)arg->commands, + NULL, arg->command_size, arg->throttle_us, + (void __user *)(unsigned long)arg->fence_rep); + + if (unlikely(ret != 0)) + goto out_unlock; + + vmw_kms_cursor_post_execbuf(dev_priv); + +out_unlock: ttm_read_unlock(&vmaster->lock); return ret; } diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c index bfab60c..7f16ff2 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c @@ -26,6 +26,8 @@ * **************************************************************************/ +#include <linux/export.h> + #include "drmP.h" #include "vmwgfx_drv.h" @@ -145,7 +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; @@ -158,10 +160,16 @@ static int vmw_fb_set_par(struct fb_info *info) { struct vmw_fb_par *par = info->par; 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); + if (ret) + return ret; - vmw_kms_write_svga(vmw_priv, info->var.xres, info->var.yres, - info->fix.line_length, - par->bpp, par->depth); if (vmw_priv->capabilities & SVGA_CAP_DISPLAY_TOPOLOGY) { /* TODO check if pitch and offset changes */ vmw_write(vmw_priv, SVGA_REG_NUM_GUEST_DISPLAYS, 1); @@ -405,14 +413,14 @@ int vmw_fb_init(struct vmw_private *vmw_priv) struct fb_info *info; unsigned initial_width, initial_height; unsigned fb_width, fb_height; - unsigned fb_bbp, fb_depth, fb_offset, fb_pitch, fb_size; + unsigned fb_bpp, fb_depth, fb_offset, fb_pitch, fb_size; int ret; /* XXX These shouldn't be hardcoded. */ initial_width = 800; initial_height = 600; - fb_bbp = 32; + fb_bpp = 32; fb_depth = 24; /* XXX As shouldn't these be as well. */ @@ -422,7 +430,7 @@ int vmw_fb_init(struct vmw_private *vmw_priv) initial_width = min(fb_width, initial_width); initial_height = min(fb_height, initial_height); - fb_pitch = fb_width * fb_bbp / 8; + fb_pitch = fb_width * fb_bpp / 8; fb_size = fb_pitch * fb_height; fb_offset = vmw_read(vmw_priv, SVGA_REG_FB_OFFSET); @@ -437,7 +445,7 @@ int vmw_fb_init(struct vmw_private *vmw_priv) par = info->par; par->vmw_priv = vmw_priv; par->depth = fb_depth; - par->bpp = fb_bbp; + par->bpp = fb_bpp; par->vmalloc = NULL; par->max_width = fb_width; par->max_height = fb_height; @@ -588,58 +596,6 @@ int vmw_fb_close(struct vmw_private *vmw_priv) return 0; } -int vmw_dmabuf_from_vram(struct vmw_private *vmw_priv, - struct vmw_dma_buffer *vmw_bo) -{ - struct ttm_buffer_object *bo = &vmw_bo->base; - int ret = 0; - - ret = ttm_bo_reserve(bo, false, false, false, 0); - if (unlikely(ret != 0)) - return ret; - - ret = ttm_bo_validate(bo, &vmw_sys_placement, false, false, false); - ttm_bo_unreserve(bo); - - return ret; -} - -int vmw_dmabuf_to_start_of_vram(struct vmw_private *vmw_priv, - struct vmw_dma_buffer *vmw_bo) -{ - struct ttm_buffer_object *bo = &vmw_bo->base; - struct ttm_placement ne_placement = vmw_vram_ne_placement; - int ret = 0; - - ne_placement.lpfn = bo->num_pages; - - /* interuptable? */ - ret = ttm_write_lock(&vmw_priv->active_master->lock, false); - if (unlikely(ret != 0)) - return ret; - - ret = ttm_bo_reserve(bo, false, false, false, 0); - if (unlikely(ret != 0)) - goto err_unlock; - - if (bo->mem.mem_type == TTM_PL_VRAM && - bo->mem.start < bo->num_pages && - bo->mem.start > 0) - (void) ttm_bo_validate(bo, &vmw_sys_placement, false, - false, false); - - ret = ttm_bo_validate(bo, &ne_placement, false, false, false); - - /* Could probably bug on */ - WARN_ON(bo->offset != 0); - - ttm_bo_unreserve(bo); -err_unlock: - ttm_write_unlock(&vmw_priv->active_master->lock); - - return ret; -} - int vmw_fb_off(struct vmw_private *vmw_priv) { struct fb_info *info; @@ -661,7 +617,7 @@ int vmw_fb_off(struct vmw_private *vmw_priv) par->bo_ptr = NULL; ttm_bo_kunmap(&par->map); - vmw_dmabuf_from_vram(vmw_priv, par->vmw_bo); + vmw_dmabuf_unpin(vmw_priv, par->vmw_bo, false); return 0; } @@ -687,7 +643,7 @@ int vmw_fb_on(struct vmw_private *vmw_priv) /* Make sure that all overlays are stoped when we take over */ vmw_overlay_stop_all(vmw_priv); - ret = vmw_dmabuf_to_start_of_vram(vmw_priv, par->vmw_bo); + ret = vmw_dmabuf_to_start_of_vram(vmw_priv, par->vmw_bo, true, false); if (unlikely(ret != 0)) { DRM_ERROR("could not move buffer to start of VRAM\n"); goto err_no_buffer; diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c index 61eacc1..1ed5a1c 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c @@ -1,6 +1,6 @@ /************************************************************************** * - * Copyright (C) 2010 VMware, Inc., Palo Alto, CA., USA + * Copyright © 2011 VMware, Inc., Palo Alto, CA., USA * All Rights Reserved. * * Permission is hereby granted, free of charge, to any person obtaining a @@ -25,149 +25,1083 @@ * **************************************************************************/ - +#include "drmP.h" #include "vmwgfx_drv.h" -struct vmw_fence { - struct list_head head; - uint32_t sequence; - struct timespec submitted; +#define VMW_FENCE_WRAP (1 << 31) + +struct vmw_fence_manager { + int num_fence_objects; + struct vmw_private *dev_priv; + spinlock_t lock; + struct list_head fence_list; + struct work_struct work; + u32 user_fence_size; + u32 fence_size; + u32 event_fence_action_size; + bool fifo_down; + struct list_head cleanup_list; + uint32_t pending_actions[VMW_ACTION_MAX]; + struct mutex goal_irq_mutex; + bool goal_irq_on; /* Protected by @goal_irq_mutex */ + bool seqno_valid; /* Protected by @lock, and may not be set to true + without the @goal_irq_mutex held. */ }; -void vmw_fence_queue_init(struct vmw_fence_queue *queue) +struct vmw_user_fence { + struct ttm_base_object base; + struct vmw_fence_obj fence; +}; + +/** + * struct vmw_event_fence_action - fence action that delivers a drm event. + * + * @e: A struct drm_pending_event that controls the event delivery. + * @action: A struct vmw_fence_action to hook up to a fence. + * @fence: A referenced pointer to the fence to keep it alive while @action + * hangs on it. + * @dev: Pointer to a struct drm_device so we can access the event stuff. + * @kref: Both @e and @action has destructors, so we need to refcount. + * @size: Size accounted for this object. + * @tv_sec: If non-null, the variable pointed to will be assigned + * current time tv_sec val when the fence signals. + * @tv_usec: Must be set if @tv_sec is set, and the variable pointed to will + * be assigned the current time tv_usec val when the fence signals. + */ +struct vmw_event_fence_action { + struct drm_pending_event e; + struct vmw_fence_action action; + struct vmw_fence_obj *fence; + struct drm_device *dev; + struct kref kref; + uint32_t size; + uint32_t *tv_sec; + uint32_t *tv_usec; +}; + +/** + * Note on fencing subsystem usage of irqs: + * Typically the vmw_fences_update function is called + * + * a) When a new fence seqno has been submitted by the fifo code. + * b) On-demand when we have waiters. Sleeping waiters will switch on the + * ANY_FENCE irq and call vmw_fences_update function each time an ANY_FENCE + * irq is received. When the last fence waiter is gone, that IRQ is masked + * away. + * + * In situations where there are no waiters and we don't submit any new fences, + * fence objects may not be signaled. This is perfectly OK, since there are + * no consumers of the signaled data, but that is NOT ok when there are fence + * actions attached to a fence. The fencing subsystem then makes use of the + * FENCE_GOAL irq and sets the fence goal seqno to that of the next fence + * which has an action attached, and each time vmw_fences_update is called, + * the subsystem makes sure the fence goal seqno is updated. + * + * The fence goal seqno irq is on as long as there are unsignaled fence + * objects with actions attached to them. + */ + +static void vmw_fence_obj_destroy_locked(struct kref *kref) { - INIT_LIST_HEAD(&queue->head); - queue->lag = ns_to_timespec(0); - getrawmonotonic(&queue->lag_time); - spin_lock_init(&queue->lock); + struct vmw_fence_obj *fence = + container_of(kref, struct vmw_fence_obj, kref); + + struct vmw_fence_manager *fman = fence->fman; + unsigned int num_fences; + + list_del_init(&fence->head); + num_fences = --fman->num_fence_objects; + spin_unlock_irq(&fman->lock); + if (fence->destroy) + fence->destroy(fence); + else + kfree(fence); + + spin_lock_irq(&fman->lock); } -void vmw_fence_queue_takedown(struct vmw_fence_queue *queue) + +/** + * Execute signal actions on fences recently signaled. + * This is done from a workqueue so we don't have to execute + * signal actions from atomic context. + */ + +static void vmw_fence_work_func(struct work_struct *work) { - struct vmw_fence *fence, *next; + struct vmw_fence_manager *fman = + container_of(work, struct vmw_fence_manager, work); + struct list_head list; + struct vmw_fence_action *action, *next_action; + bool seqno_valid; - spin_lock(&queue->lock); - list_for_each_entry_safe(fence, next, &queue->head, head) { - kfree(fence); + do { + INIT_LIST_HEAD(&list); + mutex_lock(&fman->goal_irq_mutex); + + spin_lock_irq(&fman->lock); + list_splice_init(&fman->cleanup_list, &list); + seqno_valid = fman->seqno_valid; + spin_unlock_irq(&fman->lock); + + if (!seqno_valid && fman->goal_irq_on) { + fman->goal_irq_on = false; + vmw_goal_waiter_remove(fman->dev_priv); + } + mutex_unlock(&fman->goal_irq_mutex); + + if (list_empty(&list)) + return; + + /* + * At this point, only we should be able to manipulate the + * list heads of the actions we have on the private list. + * hence fman::lock not held. + */ + + list_for_each_entry_safe(action, next_action, &list, head) { + list_del_init(&action->head); + if (action->cleanup) + action->cleanup(action); + } + } while (1); +} + +struct vmw_fence_manager *vmw_fence_manager_init(struct vmw_private *dev_priv) +{ + struct vmw_fence_manager *fman = kzalloc(sizeof(*fman), GFP_KERNEL); + + if (unlikely(fman == NULL)) + return NULL; + + fman->dev_priv = dev_priv; + spin_lock_init(&fman->lock); + INIT_LIST_HEAD(&fman->fence_list); + INIT_LIST_HEAD(&fman->cleanup_list); + INIT_WORK(&fman->work, &vmw_fence_work_func); + fman->fifo_down = true; + fman->user_fence_size = ttm_round_pot(sizeof(struct vmw_user_fence)); + fman->fence_size = ttm_round_pot(sizeof(struct vmw_fence_obj)); + fman->event_fence_action_size = + ttm_round_pot(sizeof(struct vmw_event_fence_action)); + mutex_init(&fman->goal_irq_mutex); + + return fman; +} + +void vmw_fence_manager_takedown(struct vmw_fence_manager *fman) +{ + unsigned long irq_flags; + bool lists_empty; + + (void) cancel_work_sync(&fman->work); + + spin_lock_irqsave(&fman->lock, irq_flags); + lists_empty = list_empty(&fman->fence_list) && + list_empty(&fman->cleanup_list); + spin_unlock_irqrestore(&fman->lock, irq_flags); + + BUG_ON(!lists_empty); + kfree(fman); +} + +static int vmw_fence_obj_init(struct vmw_fence_manager *fman, + struct vmw_fence_obj *fence, + u32 seqno, + uint32_t mask, + void (*destroy) (struct vmw_fence_obj *fence)) +{ + unsigned long irq_flags; + unsigned int num_fences; + int ret = 0; + + fence->seqno = seqno; + INIT_LIST_HEAD(&fence->seq_passed_actions); + fence->fman = fman; + fence->signaled = 0; + fence->signal_mask = mask; + kref_init(&fence->kref); + fence->destroy = destroy; + init_waitqueue_head(&fence->queue); + + spin_lock_irqsave(&fman->lock, irq_flags); + if (unlikely(fman->fifo_down)) { + ret = -EBUSY; + goto out_unlock; + } + list_add_tail(&fence->head, &fman->fence_list); + num_fences = ++fman->num_fence_objects; + +out_unlock: + spin_unlock_irqrestore(&fman->lock, irq_flags); + return ret; + +} + +struct vmw_fence_obj *vmw_fence_obj_reference(struct vmw_fence_obj *fence) +{ + if (unlikely(fence == NULL)) + return NULL; + + kref_get(&fence->kref); + return fence; +} + +/** + * vmw_fence_obj_unreference + * + * Note that this function may not be entered with disabled irqs since + * it may re-enable them in the destroy function. + * + */ +void vmw_fence_obj_unreference(struct vmw_fence_obj **fence_p) +{ + struct vmw_fence_obj *fence = *fence_p; + struct vmw_fence_manager *fman; + + if (unlikely(fence == NULL)) + return; + + fman = fence->fman; + *fence_p = NULL; + spin_lock_irq(&fman->lock); + BUG_ON(atomic_read(&fence->kref.refcount) == 0); + kref_put(&fence->kref, vmw_fence_obj_destroy_locked); + spin_unlock_irq(&fman->lock); +} + +void vmw_fences_perform_actions(struct vmw_fence_manager *fman, + struct list_head *list) +{ + struct vmw_fence_action *action, *next_action; + + list_for_each_entry_safe(action, next_action, list, head) { + list_del_init(&action->head); + fman->pending_actions[action->type]--; + if (action->seq_passed != NULL) + action->seq_passed(action); + + /* + * Add the cleanup action to the cleanup list so that + * it will be performed by a worker task. + */ + + list_add_tail(&action->head, &fman->cleanup_list); + } +} + +/** + * vmw_fence_goal_new_locked - Figure out a new device fence goal + * seqno if needed. + * + * @fman: Pointer to a fence manager. + * @passed_seqno: The seqno the device currently signals as passed. + * + * This function should be called with the fence manager lock held. + * It is typically called when we have a new passed_seqno, and + * we might need to update the fence goal. It checks to see whether + * the current fence goal has already passed, and, in that case, + * scans through all unsignaled fences to get the next fence object with an + * action attached, and sets the seqno of that fence as a new fence goal. + * + * returns true if the device goal seqno was updated. False otherwise. + */ +static bool vmw_fence_goal_new_locked(struct vmw_fence_manager *fman, + u32 passed_seqno) +{ + u32 goal_seqno; + __le32 __iomem *fifo_mem; + struct vmw_fence_obj *fence; + + if (likely(!fman->seqno_valid)) + return false; + + fifo_mem = fman->dev_priv->mmio_virt; + goal_seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE_GOAL); + if (likely(passed_seqno - goal_seqno >= VMW_FENCE_WRAP)) + return false; + + fman->seqno_valid = false; + list_for_each_entry(fence, &fman->fence_list, head) { + if (!list_empty(&fence->seq_passed_actions)) { + fman->seqno_valid = true; + iowrite32(fence->seqno, + fifo_mem + SVGA_FIFO_FENCE_GOAL); + break; + } } - spin_unlock(&queue->lock); + + return true; +} + + +/** + * vmw_fence_goal_check_locked - Replace the device fence goal seqno if + * needed. + * + * @fence: Pointer to a struct vmw_fence_obj the seqno of which should be + * considered as a device fence goal. + * + * This function should be called with the fence manager lock held. + * It is typically called when an action has been attached to a fence to + * check whether the seqno of that fence should be used for a fence + * goal interrupt. This is typically needed if the current fence goal is + * invalid, or has a higher seqno than that of the current fence object. + * + * returns true if the device goal seqno was updated. False otherwise. + */ +static bool vmw_fence_goal_check_locked(struct vmw_fence_obj *fence) +{ + u32 goal_seqno; + __le32 __iomem *fifo_mem; + + if (fence->signaled & DRM_VMW_FENCE_FLAG_EXEC) + return false; + + fifo_mem = fence->fman->dev_priv->mmio_virt; + goal_seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE_GOAL); + if (likely(fence->fman->seqno_valid && + goal_seqno - fence->seqno < VMW_FENCE_WRAP)) + return false; + + iowrite32(fence->seqno, fifo_mem + SVGA_FIFO_FENCE_GOAL); + fence->fman->seqno_valid = true; + + return true; +} + +void vmw_fences_update(struct vmw_fence_manager *fman) +{ + unsigned long flags; + struct vmw_fence_obj *fence, *next_fence; + struct list_head action_list; + bool needs_rerun; + uint32_t seqno, new_seqno; + __le32 __iomem *fifo_mem = fman->dev_priv->mmio_virt; + + seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE); +rerun: + spin_lock_irqsave(&fman->lock, flags); + list_for_each_entry_safe(fence, next_fence, &fman->fence_list, head) { + if (seqno - fence->seqno < VMW_FENCE_WRAP) { + list_del_init(&fence->head); + fence->signaled |= DRM_VMW_FENCE_FLAG_EXEC; + INIT_LIST_HEAD(&action_list); + list_splice_init(&fence->seq_passed_actions, + &action_list); + vmw_fences_perform_actions(fman, &action_list); + wake_up_all(&fence->queue); + } else + break; + } + + needs_rerun = vmw_fence_goal_new_locked(fman, seqno); + + if (!list_empty(&fman->cleanup_list)) + (void) schedule_work(&fman->work); + spin_unlock_irqrestore(&fman->lock, flags); + + /* + * Rerun if the fence goal seqno was updated, and the + * hardware might have raced with that update, so that + * we missed a fence_goal irq. + */ + + if (unlikely(needs_rerun)) { + new_seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE); + if (new_seqno != seqno) { + seqno = new_seqno; + goto rerun; + } + } +} + +bool vmw_fence_obj_signaled(struct vmw_fence_obj *fence, + uint32_t flags) +{ + struct vmw_fence_manager *fman = fence->fman; + unsigned long irq_flags; + uint32_t signaled; + + spin_lock_irqsave(&fman->lock, irq_flags); + signaled = fence->signaled; + spin_unlock_irqrestore(&fman->lock, irq_flags); + + flags &= fence->signal_mask; + if ((signaled & flags) == flags) + return 1; + + if ((signaled & DRM_VMW_FENCE_FLAG_EXEC) == 0) + vmw_fences_update(fman); + + spin_lock_irqsave(&fman->lock, irq_flags); + signaled = fence->signaled; + spin_unlock_irqrestore(&fman->lock, irq_flags); + + return ((signaled & flags) == flags); } -int vmw_fence_push(struct vmw_fence_queue *queue, - uint32_t sequence) +int vmw_fence_obj_wait(struct vmw_fence_obj *fence, + uint32_t flags, bool lazy, + bool interruptible, unsigned long timeout) { - struct vmw_fence *fence = kmalloc(sizeof(*fence), GFP_KERNEL); + struct vmw_private *dev_priv = fence->fman->dev_priv; + long ret; + + if (likely(vmw_fence_obj_signaled(fence, flags))) + return 0; + + vmw_fifo_ping_host(dev_priv, SVGA_SYNC_GENERIC); + vmw_seqno_waiter_add(dev_priv); - if (unlikely(!fence)) + if (interruptible) + ret = wait_event_interruptible_timeout + (fence->queue, + vmw_fence_obj_signaled(fence, flags), + timeout); + else + ret = wait_event_timeout + (fence->queue, + vmw_fence_obj_signaled(fence, flags), + timeout); + + vmw_seqno_waiter_remove(dev_priv); + + if (unlikely(ret == 0)) + ret = -EBUSY; + else if (likely(ret > 0)) + ret = 0; + + return ret; +} + +void vmw_fence_obj_flush(struct vmw_fence_obj *fence) +{ + struct vmw_private *dev_priv = fence->fman->dev_priv; + + vmw_fifo_ping_host(dev_priv, SVGA_SYNC_GENERIC); +} + +static void vmw_fence_destroy(struct vmw_fence_obj *fence) +{ + kfree(fence); +} + +int vmw_fence_create(struct vmw_fence_manager *fman, + uint32_t seqno, + uint32_t mask, + struct vmw_fence_obj **p_fence) +{ + struct vmw_fence_obj *fence; + int ret; + + fence = kzalloc(sizeof(*fence), GFP_KERNEL); + if (unlikely(fence == NULL)) return -ENOMEM; - fence->sequence = sequence; - getrawmonotonic(&fence->submitted); - spin_lock(&queue->lock); - list_add_tail(&fence->head, &queue->head); - spin_unlock(&queue->lock); + ret = vmw_fence_obj_init(fman, fence, seqno, mask, + vmw_fence_destroy); + if (unlikely(ret != 0)) + goto out_err_init; + *p_fence = fence; return 0; + +out_err_init: + kfree(fence); + return ret; } -int vmw_fence_pull(struct vmw_fence_queue *queue, - uint32_t signaled_sequence) + +static void vmw_user_fence_destroy(struct vmw_fence_obj *fence) { - struct vmw_fence *fence, *next; - struct timespec now; - bool updated = false; + struct vmw_user_fence *ufence = + container_of(fence, struct vmw_user_fence, fence); + struct vmw_fence_manager *fman = fence->fman; - spin_lock(&queue->lock); - getrawmonotonic(&now); + kfree(ufence); + /* + * Free kernel space accounting. + */ + ttm_mem_global_free(vmw_mem_glob(fman->dev_priv), + fman->user_fence_size); +} - if (list_empty(&queue->head)) { - queue->lag = ns_to_timespec(0); - queue->lag_time = now; - updated = true; - goto out_unlock; +static void vmw_user_fence_base_release(struct ttm_base_object **p_base) +{ + struct ttm_base_object *base = *p_base; + struct vmw_user_fence *ufence = + container_of(base, struct vmw_user_fence, base); + struct vmw_fence_obj *fence = &ufence->fence; + + *p_base = NULL; + vmw_fence_obj_unreference(&fence); +} + +int vmw_user_fence_create(struct drm_file *file_priv, + struct vmw_fence_manager *fman, + uint32_t seqno, + uint32_t mask, + struct vmw_fence_obj **p_fence, + uint32_t *p_handle) +{ + struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile; + struct vmw_user_fence *ufence; + struct vmw_fence_obj *tmp; + struct ttm_mem_global *mem_glob = vmw_mem_glob(fman->dev_priv); + int ret; + + /* + * Kernel memory space accounting, since this object may + * be created by a user-space request. + */ + + ret = ttm_mem_global_alloc(mem_glob, fman->user_fence_size, + false, false); + if (unlikely(ret != 0)) + return ret; + + ufence = kzalloc(sizeof(*ufence), GFP_KERNEL); + if (unlikely(ufence == NULL)) { + ret = -ENOMEM; + goto out_no_object; } - list_for_each_entry_safe(fence, next, &queue->head, head) { - if (signaled_sequence - fence->sequence > (1 << 30)) - continue; + ret = vmw_fence_obj_init(fman, &ufence->fence, seqno, + mask, vmw_user_fence_destroy); + if (unlikely(ret != 0)) { + kfree(ufence); + goto out_no_object; + } - queue->lag = timespec_sub(now, fence->submitted); - queue->lag_time = now; - updated = true; - list_del(&fence->head); - kfree(fence); + /* + * The base object holds a reference which is freed in + * vmw_user_fence_base_release. + */ + tmp = vmw_fence_obj_reference(&ufence->fence); + ret = ttm_base_object_init(tfile, &ufence->base, false, + VMW_RES_FENCE, + &vmw_user_fence_base_release, NULL); + + + if (unlikely(ret != 0)) { + /* + * Free the base object's reference + */ + vmw_fence_obj_unreference(&tmp); + goto out_err; } -out_unlock: - spin_unlock(&queue->lock); + *p_fence = &ufence->fence; + *p_handle = ufence->base.hash.key; + + return 0; +out_err: + tmp = &ufence->fence; + vmw_fence_obj_unreference(&tmp); +out_no_object: + ttm_mem_global_free(mem_glob, fman->user_fence_size); + return ret; +} + + +/** + * vmw_fence_fifo_down - signal all unsignaled fence objects. + */ + +void vmw_fence_fifo_down(struct vmw_fence_manager *fman) +{ + unsigned long irq_flags; + struct list_head action_list; + int ret; + + /* + * The list may be altered while we traverse it, so always + * restart when we've released the fman->lock. + */ + + spin_lock_irqsave(&fman->lock, irq_flags); + fman->fifo_down = true; + while (!list_empty(&fman->fence_list)) { + struct vmw_fence_obj *fence = + list_entry(fman->fence_list.prev, struct vmw_fence_obj, + head); + kref_get(&fence->kref); + spin_unlock_irq(&fman->lock); + + ret = vmw_fence_obj_wait(fence, fence->signal_mask, + false, false, + VMW_FENCE_WAIT_TIMEOUT); + + if (unlikely(ret != 0)) { + list_del_init(&fence->head); + fence->signaled |= DRM_VMW_FENCE_FLAG_EXEC; + INIT_LIST_HEAD(&action_list); + list_splice_init(&fence->seq_passed_actions, + &action_list); + vmw_fences_perform_actions(fman, &action_list); + wake_up_all(&fence->queue); + } + + spin_lock_irq(&fman->lock); + + BUG_ON(!list_empty(&fence->head)); + kref_put(&fence->kref, vmw_fence_obj_destroy_locked); + } + spin_unlock_irqrestore(&fman->lock, irq_flags); +} + +void vmw_fence_fifo_up(struct vmw_fence_manager *fman) +{ + unsigned long irq_flags; + + spin_lock_irqsave(&fman->lock, irq_flags); + fman->fifo_down = false; + spin_unlock_irqrestore(&fman->lock, irq_flags); +} + + +int vmw_fence_obj_wait_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_vmw_fence_wait_arg *arg = + (struct drm_vmw_fence_wait_arg *)data; + unsigned long timeout; + struct ttm_base_object *base; + struct vmw_fence_obj *fence; + struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile; + int ret; + uint64_t wait_timeout = ((uint64_t)arg->timeout_us * HZ); + + /* + * 64-bit division not present on 32-bit systems, so do an + * approximation. (Divide by 1000000). + */ + + wait_timeout = (wait_timeout >> 20) + (wait_timeout >> 24) - + (wait_timeout >> 26); + + if (!arg->cookie_valid) { + arg->cookie_valid = 1; + arg->kernel_cookie = jiffies + wait_timeout; + } - return (updated) ? 0 : -EBUSY; + base = ttm_base_object_lookup(tfile, arg->handle); + if (unlikely(base == NULL)) { + printk(KERN_ERR "Wait invalid fence object handle " + "0x%08lx.\n", + (unsigned long)arg->handle); + return -EINVAL; + } + + fence = &(container_of(base, struct vmw_user_fence, base)->fence); + + timeout = jiffies; + if (time_after_eq(timeout, (unsigned long)arg->kernel_cookie)) { + ret = ((vmw_fence_obj_signaled(fence, arg->flags)) ? + 0 : -EBUSY); + goto out; + } + + timeout = (unsigned long)arg->kernel_cookie - timeout; + + ret = vmw_fence_obj_wait(fence, arg->flags, arg->lazy, true, timeout); + +out: + ttm_base_object_unref(&base); + + /* + * Optionally unref the fence object. + */ + + if (ret == 0 && (arg->wait_options & DRM_VMW_WAIT_OPTION_UNREF)) + return ttm_ref_object_base_unref(tfile, arg->handle, + TTM_REF_USAGE); + return ret; } -static struct timespec vmw_timespec_add(struct timespec t1, - struct timespec t2) +int vmw_fence_obj_signaled_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) { - t1.tv_sec += t2.tv_sec; - t1.tv_nsec += t2.tv_nsec; - if (t1.tv_nsec >= 1000000000L) { - t1.tv_sec += 1; - t1.tv_nsec -= 1000000000L; + struct drm_vmw_fence_signaled_arg *arg = + (struct drm_vmw_fence_signaled_arg *) data; + struct ttm_base_object *base; + struct vmw_fence_obj *fence; + struct vmw_fence_manager *fman; + struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile; + struct vmw_private *dev_priv = vmw_priv(dev); + + base = ttm_base_object_lookup(tfile, arg->handle); + if (unlikely(base == NULL)) { + printk(KERN_ERR "Fence signaled invalid fence object handle " + "0x%08lx.\n", + (unsigned long)arg->handle); + return -EINVAL; } - return t1; + fence = &(container_of(base, struct vmw_user_fence, base)->fence); + fman = fence->fman; + + arg->signaled = vmw_fence_obj_signaled(fence, arg->flags); + spin_lock_irq(&fman->lock); + + arg->signaled_flags = fence->signaled; + arg->passed_seqno = dev_priv->last_read_seqno; + spin_unlock_irq(&fman->lock); + + ttm_base_object_unref(&base); + + return 0; +} + + +int vmw_fence_obj_unref_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_vmw_fence_arg *arg = + (struct drm_vmw_fence_arg *) data; + + return ttm_ref_object_base_unref(vmw_fpriv(file_priv)->tfile, + arg->handle, + TTM_REF_USAGE); } -static struct timespec vmw_fifo_lag(struct vmw_fence_queue *queue) +/** + * vmw_event_fence_action_destroy + * + * @kref: The struct kref embedded in a struct vmw_event_fence_action. + * + * The vmw_event_fence_action destructor that may be called either after + * the fence action cleanup, or when the event is delivered. + * It frees both the vmw_event_fence_action struct and the actual + * event structure copied to user-space. + */ +static void vmw_event_fence_action_destroy(struct kref *kref) { - struct timespec now; + struct vmw_event_fence_action *eaction = + container_of(kref, struct vmw_event_fence_action, kref); + struct ttm_mem_global *mem_glob = + vmw_mem_glob(vmw_priv(eaction->dev)); + uint32_t size = eaction->size; - spin_lock(&queue->lock); - getrawmonotonic(&now); - queue->lag = vmw_timespec_add(queue->lag, - timespec_sub(now, queue->lag_time)); - queue->lag_time = now; - spin_unlock(&queue->lock); - return queue->lag; + kfree(eaction->e.event); + kfree(eaction); + ttm_mem_global_free(mem_glob, size); } -static bool vmw_lag_lt(struct vmw_fence_queue *queue, - uint32_t us) +/** + * vmw_event_fence_action_delivered + * + * @e: The struct drm_pending_event embedded in a struct + * vmw_event_fence_action. + * + * The struct drm_pending_event destructor that is called by drm + * once the event is delivered. Since we don't know whether this function + * will be called before or after the fence action destructor, we + * free a refcount and destroy if it becomes zero. + */ +static void vmw_event_fence_action_delivered(struct drm_pending_event *e) { - struct timespec lag, cond; + struct vmw_event_fence_action *eaction = + container_of(e, struct vmw_event_fence_action, e); - cond = ns_to_timespec((s64) us * 1000); - lag = vmw_fifo_lag(queue); - return (timespec_compare(&lag, &cond) < 1); + kref_put(&eaction->kref, vmw_event_fence_action_destroy); } -int vmw_wait_lag(struct vmw_private *dev_priv, - struct vmw_fence_queue *queue, uint32_t us) + +/** + * vmw_event_fence_action_seq_passed + * + * @action: The struct vmw_fence_action embedded in a struct + * vmw_event_fence_action. + * + * This function is called when the seqno of the fence where @action is + * attached has passed. It queues the event on the submitter's event list. + * This function is always called from atomic context, and may be called + * from irq context. It ups a refcount reflecting that we now have two + * destructors. + */ +static void vmw_event_fence_action_seq_passed(struct vmw_fence_action *action) { - struct vmw_fence *fence; - uint32_t sequence; - int ret; + struct vmw_event_fence_action *eaction = + container_of(action, struct vmw_event_fence_action, action); + struct drm_device *dev = eaction->dev; + struct drm_file *file_priv = eaction->e.file_priv; + unsigned long irq_flags; - while (!vmw_lag_lt(queue, us)) { - spin_lock(&queue->lock); - if (list_empty(&queue->head)) - sequence = atomic_read(&dev_priv->fence_seq); - else { - fence = list_first_entry(&queue->head, - struct vmw_fence, head); - sequence = fence->sequence; + kref_get(&eaction->kref); + spin_lock_irqsave(&dev->event_lock, irq_flags); + + if (likely(eaction->tv_sec != NULL)) { + struct timeval tv; + + do_gettimeofday(&tv); + *eaction->tv_sec = tv.tv_sec; + *eaction->tv_usec = tv.tv_usec; + } + + list_add_tail(&eaction->e.link, &file_priv->event_list); + wake_up_all(&file_priv->event_wait); + spin_unlock_irqrestore(&dev->event_lock, irq_flags); +} + +/** + * vmw_event_fence_action_cleanup + * + * @action: The struct vmw_fence_action embedded in a struct + * vmw_event_fence_action. + * + * This function is the struct vmw_fence_action destructor. It's typically + * called from a workqueue. + */ +static void vmw_event_fence_action_cleanup(struct vmw_fence_action *action) +{ + struct vmw_event_fence_action *eaction = + container_of(action, struct vmw_event_fence_action, action); + + vmw_fence_obj_unreference(&eaction->fence); + kref_put(&eaction->kref, vmw_event_fence_action_destroy); +} + + +/** + * vmw_fence_obj_add_action - Add an action to a fence object. + * + * @fence - The fence object. + * @action - The action to add. + * + * Note that the action callbacks may be executed before this function + * returns. + */ +void vmw_fence_obj_add_action(struct vmw_fence_obj *fence, + struct vmw_fence_action *action) +{ + struct vmw_fence_manager *fman = fence->fman; + unsigned long irq_flags; + bool run_update = false; + + mutex_lock(&fman->goal_irq_mutex); + spin_lock_irqsave(&fman->lock, irq_flags); + + fman->pending_actions[action->type]++; + if (fence->signaled & DRM_VMW_FENCE_FLAG_EXEC) { + struct list_head action_list; + + INIT_LIST_HEAD(&action_list); + list_add_tail(&action->head, &action_list); + vmw_fences_perform_actions(fman, &action_list); + } else { + list_add_tail(&action->head, &fence->seq_passed_actions); + + /* + * This function may set fman::seqno_valid, so it must + * be run with the goal_irq_mutex held. + */ + run_update = vmw_fence_goal_check_locked(fence); + } + + spin_unlock_irqrestore(&fman->lock, irq_flags); + + if (run_update) { + if (!fman->goal_irq_on) { + fman->goal_irq_on = true; + vmw_goal_waiter_add(fman->dev_priv); } - spin_unlock(&queue->lock); + vmw_fences_update(fman); + } + mutex_unlock(&fman->goal_irq_mutex); + +} - ret = vmw_wait_fence(dev_priv, false, sequence, true, - 3*HZ); +/** + * vmw_event_fence_action_create - Post an event for sending when a fence + * object seqno has passed. + * + * @file_priv: The file connection on which the event should be posted. + * @fence: The fence object on which to post the event. + * @event: Event to be posted. This event should've been alloced + * using k[mz]alloc, and should've been completely initialized. + * @interruptible: Interruptible waits if possible. + * + * As a side effect, the object pointed to by @event may have been + * freed when this function returns. If this function returns with + * an error code, the caller needs to free that object. + */ - if (unlikely(ret != 0)) - return ret; +int vmw_event_fence_action_create(struct drm_file *file_priv, + struct vmw_fence_obj *fence, + struct drm_event *event, + uint32_t *tv_sec, + uint32_t *tv_usec, + bool interruptible) +{ + struct vmw_event_fence_action *eaction; + struct ttm_mem_global *mem_glob = + vmw_mem_glob(fence->fman->dev_priv); + struct vmw_fence_manager *fman = fence->fman; + uint32_t size = fman->event_fence_action_size + + ttm_round_pot(event->length); + int ret; + + /* + * Account for internal structure size as well as the + * event size itself. + */ + + ret = ttm_mem_global_alloc(mem_glob, size, false, interruptible); + if (unlikely(ret != 0)) + return ret; - (void) vmw_fence_pull(queue, sequence); + eaction = kzalloc(sizeof(*eaction), GFP_KERNEL); + if (unlikely(eaction == NULL)) { + ttm_mem_global_free(mem_glob, size); + return -ENOMEM; } + + eaction->e.event = event; + eaction->e.file_priv = file_priv; + eaction->e.destroy = vmw_event_fence_action_delivered; + + eaction->action.seq_passed = vmw_event_fence_action_seq_passed; + eaction->action.cleanup = vmw_event_fence_action_cleanup; + eaction->action.type = VMW_ACTION_EVENT; + + eaction->fence = vmw_fence_obj_reference(fence); + eaction->dev = fman->dev_priv->dev; + eaction->size = size; + eaction->tv_sec = tv_sec; + eaction->tv_usec = tv_usec; + + kref_init(&eaction->kref); + vmw_fence_obj_add_action(fence, &eaction->action); + return 0; } +int vmw_fence_event_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct vmw_private *dev_priv = vmw_priv(dev); + struct drm_vmw_fence_event_arg *arg = + (struct drm_vmw_fence_event_arg *) data; + struct vmw_fence_obj *fence = NULL; + struct vmw_fpriv *vmw_fp = vmw_fpriv(file_priv); + struct drm_vmw_fence_rep __user *user_fence_rep = + (struct drm_vmw_fence_rep __user *)(unsigned long) + arg->fence_rep; + uint32_t handle; + unsigned long irq_flags; + struct drm_vmw_event_fence *event; + int ret; + + /* + * Look up an existing fence object, + * and if user-space wants a new reference, + * add one. + */ + if (arg->handle) { + struct ttm_base_object *base = + ttm_base_object_lookup(vmw_fp->tfile, arg->handle); + if (unlikely(base == NULL)) { + DRM_ERROR("Fence event invalid fence object handle " + "0x%08lx.\n", + (unsigned long)arg->handle); + return -EINVAL; + } + fence = &(container_of(base, struct vmw_user_fence, + base)->fence); + (void) vmw_fence_obj_reference(fence); + + if (user_fence_rep != NULL) { + bool existed; + + ret = ttm_ref_object_add(vmw_fp->tfile, base, + TTM_REF_USAGE, &existed); + if (unlikely(ret != 0)) { + DRM_ERROR("Failed to reference a fence " + "object.\n"); + goto out_no_ref_obj; + } + handle = base->hash.key; + } + ttm_base_object_unref(&base); + } + + /* + * Create a new fence object. + */ + if (!fence) { + ret = vmw_execbuf_fence_commands(file_priv, dev_priv, + &fence, + (user_fence_rep) ? + &handle : NULL); + if (unlikely(ret != 0)) { + DRM_ERROR("Fence event failed to create fence.\n"); + return ret; + } + } + + BUG_ON(fence == NULL); + + spin_lock_irqsave(&dev->event_lock, irq_flags); + + ret = (file_priv->event_space < sizeof(*event)) ? -EBUSY : 0; + if (likely(ret == 0)) + file_priv->event_space -= sizeof(*event); + + spin_unlock_irqrestore(&dev->event_lock, irq_flags); + + if (unlikely(ret != 0)) { + DRM_ERROR("Failed to allocate event space for this file.\n"); + goto out_no_event_space; + } + + event = kzalloc(sizeof(*event), GFP_KERNEL); + if (unlikely(event == NULL)) { + DRM_ERROR("Failed to allocate an event.\n"); + goto out_no_event; + } + + event->base.type = DRM_VMW_EVENT_FENCE_SIGNALED; + event->base.length = sizeof(*event); + event->user_data = arg->user_data; + + if (arg->flags & DRM_VMW_FE_FLAG_REQ_TIME) + ret = vmw_event_fence_action_create(file_priv, fence, + &event->base, + &event->tv_sec, + &event->tv_usec, + true); + else + ret = vmw_event_fence_action_create(file_priv, fence, + &event->base, + NULL, + NULL, + true); + + if (unlikely(ret != 0)) { + if (ret != -ERESTARTSYS) + DRM_ERROR("Failed to attach event to fence.\n"); + goto out_no_attach; + } + + vmw_execbuf_copy_fence_user(dev_priv, vmw_fp, 0, user_fence_rep, fence, + handle); + vmw_fence_obj_unreference(&fence); + return 0; +out_no_attach: + kfree(event); +out_no_event: + spin_lock_irqsave(&dev->event_lock, irq_flags); + file_priv->event_space += sizeof(*event); + spin_unlock_irqrestore(&dev->event_lock, irq_flags); +out_no_event_space: + if (user_fence_rep != NULL) + ttm_ref_object_base_unref(vmw_fpriv(file_priv)->tfile, + handle, TTM_REF_USAGE); +out_no_ref_obj: + vmw_fence_obj_unreference(&fence); + return ret; +} diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c index 635c0ff..decca82 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c @@ -33,6 +33,7 @@ bool vmw_fifo_have_3d(struct vmw_private *dev_priv) { __le32 __iomem *fifo_mem = dev_priv->mmio_virt; uint32_t fifo_min, hwversion; + const struct vmw_fifo_state *fifo = &dev_priv->fifo; if (!(dev_priv->capabilities & SVGA_CAP_EXTENDED_FIFO)) return false; @@ -41,11 +42,20 @@ bool vmw_fifo_have_3d(struct vmw_private *dev_priv) if (fifo_min <= SVGA_FIFO_3D_HWVERSION * sizeof(unsigned int)) return false; - hwversion = ioread32(fifo_mem + SVGA_FIFO_3D_HWVERSION); + hwversion = ioread32(fifo_mem + + ((fifo->capabilities & + SVGA_FIFO_CAP_3D_HWVERSION_REVISED) ? + SVGA_FIFO_3D_HWVERSION_REVISED : + SVGA_FIFO_3D_HWVERSION)); + if (hwversion == 0) return false; - if (hwversion < SVGA3D_HWVERSION_WS65_B1) + if (hwversion < SVGA3D_HWVERSION_WS8_B1) + return false; + + /* Non-Screen Object path does not support surfaces */ + if (!dev_priv->sou_priv) return false; return true; @@ -72,22 +82,12 @@ int vmw_fifo_init(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo) uint32_t max; uint32_t min; uint32_t dummy; - int ret; fifo->static_buffer_size = VMWGFX_FIFO_STATIC_SIZE; fifo->static_buffer = vmalloc(fifo->static_buffer_size); if (unlikely(fifo->static_buffer == NULL)) return -ENOMEM; - fifo->last_buffer_size = VMWGFX_FIFO_STATIC_SIZE; - fifo->last_data_size = 0; - fifo->last_buffer_add = false; - fifo->last_buffer = vmalloc(fifo->last_buffer_size); - if (unlikely(fifo->last_buffer == NULL)) { - ret = -ENOMEM; - goto out_err; - } - fifo->dynamic_buffer = NULL; fifo->reserved_size = 0; fifo->using_bounce_buffer = false; @@ -137,14 +137,10 @@ int vmw_fifo_init(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo) (unsigned int) min, (unsigned int) fifo->capabilities); - atomic_set(&dev_priv->fence_seq, dev_priv->last_read_sequence); - iowrite32(dev_priv->last_read_sequence, fifo_mem + SVGA_FIFO_FENCE); - vmw_fence_queue_init(&fifo->fence_queue); + atomic_set(&dev_priv->marker_seq, dev_priv->last_read_seqno); + iowrite32(dev_priv->last_read_seqno, fifo_mem + SVGA_FIFO_FENCE); + vmw_marker_queue_init(&fifo->marker_queue); return vmw_fifo_send_fence(dev_priv, &dummy); -out_err: - vfree(fifo->static_buffer); - fifo->static_buffer = NULL; - return ret; } void vmw_fifo_ping_host(struct vmw_private *dev_priv, uint32_t reason) @@ -167,10 +163,11 @@ void vmw_fifo_release(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo) mutex_lock(&dev_priv->hw_mutex); + vmw_write(dev_priv, SVGA_REG_SYNC, SVGA_SYNC_GENERIC); while (vmw_read(dev_priv, SVGA_REG_BUSY) != 0) - vmw_write(dev_priv, SVGA_REG_SYNC, SVGA_SYNC_GENERIC); + ; - dev_priv->last_read_sequence = ioread32(fifo_mem + SVGA_FIFO_FENCE); + dev_priv->last_read_seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE); vmw_write(dev_priv, SVGA_REG_CONFIG_DONE, dev_priv->config_done_state); @@ -180,12 +177,7 @@ void vmw_fifo_release(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo) dev_priv->traces_state); mutex_unlock(&dev_priv->hw_mutex); - vmw_fence_queue_takedown(&fifo->fence_queue); - - if (likely(fifo->last_buffer != NULL)) { - vfree(fifo->last_buffer); - fifo->last_buffer = NULL; - } + vmw_marker_queue_takedown(&fifo->marker_queue); if (likely(fifo->static_buffer != NULL)) { vfree(fifo->static_buffer); @@ -262,9 +254,8 @@ static int vmw_fifo_wait(struct vmw_private *dev_priv, spin_lock_irqsave(&dev_priv->irq_lock, irq_flags); outl(SVGA_IRQFLAG_FIFO_PROGRESS, dev_priv->io_start + VMWGFX_IRQSTATUS_PORT); - vmw_write(dev_priv, SVGA_REG_IRQMASK, - vmw_read(dev_priv, SVGA_REG_IRQMASK) | - SVGA_IRQFLAG_FIFO_PROGRESS); + dev_priv->irq_mask |= SVGA_IRQFLAG_FIFO_PROGRESS; + vmw_write(dev_priv, SVGA_REG_IRQMASK, dev_priv->irq_mask); spin_unlock_irqrestore(&dev_priv->irq_lock, irq_flags); } mutex_unlock(&dev_priv->hw_mutex); @@ -286,9 +277,8 @@ static int vmw_fifo_wait(struct vmw_private *dev_priv, mutex_lock(&dev_priv->hw_mutex); if (atomic_dec_and_test(&dev_priv->fifo_queue_waiters)) { spin_lock_irqsave(&dev_priv->irq_lock, irq_flags); - vmw_write(dev_priv, SVGA_REG_IRQMASK, - vmw_read(dev_priv, SVGA_REG_IRQMASK) & - ~SVGA_IRQFLAG_FIFO_PROGRESS); + dev_priv->irq_mask &= ~SVGA_IRQFLAG_FIFO_PROGRESS; + vmw_write(dev_priv, SVGA_REG_IRQMASK, dev_priv->irq_mask); spin_unlock_irqrestore(&dev_priv->irq_lock, irq_flags); } mutex_unlock(&dev_priv->hw_mutex); @@ -296,6 +286,16 @@ static int vmw_fifo_wait(struct vmw_private *dev_priv, return ret; } +/** + * Reserve @bytes number of bytes in the fifo. + * + * This function will return NULL (error) on two conditions: + * If it timeouts waiting for fifo space, or if @bytes is larger than the + * available fifo space. + * + * Returns: + * Pointer to the fifo, or null on error (possible hardware hang). + */ void *vmw_fifo_reserve(struct vmw_private *dev_priv, uint32_t bytes) { struct vmw_fifo_state *fifo_state = &dev_priv->fifo; @@ -466,7 +466,7 @@ void vmw_fifo_commit(struct vmw_private *dev_priv, uint32_t bytes) mutex_unlock(&fifo_state->fifo_mutex); } -int vmw_fifo_send_fence(struct vmw_private *dev_priv, uint32_t *sequence) +int vmw_fifo_send_fence(struct vmw_private *dev_priv, uint32_t *seqno) { struct vmw_fifo_state *fifo_state = &dev_priv->fifo; struct svga_fifo_cmd_fence *cmd_fence; @@ -476,16 +476,16 @@ int vmw_fifo_send_fence(struct vmw_private *dev_priv, uint32_t *sequence) fm = vmw_fifo_reserve(dev_priv, bytes); if (unlikely(fm == NULL)) { - *sequence = atomic_read(&dev_priv->fence_seq); + *seqno = atomic_read(&dev_priv->marker_seq); ret = -ENOMEM; - (void)vmw_fallback_wait(dev_priv, false, true, *sequence, + (void)vmw_fallback_wait(dev_priv, false, true, *seqno, false, 3*HZ); goto out_err; } do { - *sequence = atomic_add_return(1, &dev_priv->fence_seq); - } while (*sequence == 0); + *seqno = atomic_add_return(1, &dev_priv->marker_seq); + } while (*seqno == 0); if (!(fifo_state->capabilities & SVGA_FIFO_CAP_FENCE)) { @@ -502,61 +502,68 @@ int vmw_fifo_send_fence(struct vmw_private *dev_priv, uint32_t *sequence) cmd_fence = (struct svga_fifo_cmd_fence *) ((unsigned long)fm + sizeof(__le32)); - iowrite32(*sequence, &cmd_fence->fence); - fifo_state->last_buffer_add = true; + iowrite32(*seqno, &cmd_fence->fence); vmw_fifo_commit(dev_priv, bytes); - fifo_state->last_buffer_add = false; - (void) vmw_fence_push(&fifo_state->fence_queue, *sequence); - vmw_update_sequence(dev_priv, fifo_state); + (void) vmw_marker_push(&fifo_state->marker_queue, *seqno); + vmw_update_seqno(dev_priv, fifo_state); out_err: return ret; } /** - * Map the first page of the FIFO read-only to user-space. + * vmw_fifo_emit_dummy_query - emits a dummy query to the fifo. + * + * @dev_priv: The device private structure. + * @cid: The hardware context id used for the query. + * + * This function is used to emit a dummy occlusion query with + * no primitives rendered between query begin and query end. + * It's used to provide a query barrier, in order to know that when + * this query is finished, all preceding queries are also finished. + * + * A Query results structure should have been initialized at the start + * of the dev_priv->dummy_query_bo buffer object. And that buffer object + * must also be either reserved or pinned when this function is called. + * + * Returns -ENOMEM on failure to reserve fifo space. */ - -static int vmw_fifo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf) +int vmw_fifo_emit_dummy_query(struct vmw_private *dev_priv, + uint32_t cid) { - int ret; - unsigned long address = (unsigned long)vmf->virtual_address; + /* + * A query wait without a preceding query end will + * actually finish all queries for this cid + * without writing to the query result structure. + */ - if (address != vma->vm_start) - return VM_FAULT_SIGBUS; + struct ttm_buffer_object *bo = dev_priv->dummy_query_bo; + struct { + SVGA3dCmdHeader header; + SVGA3dCmdWaitForQuery body; + } *cmd; - ret = vm_insert_pfn(vma, address, vma->vm_pgoff); - if (likely(ret == -EBUSY || ret == 0)) - return VM_FAULT_NOPAGE; - else if (ret == -ENOMEM) - return VM_FAULT_OOM; + cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd)); - return VM_FAULT_SIGBUS; -} + if (unlikely(cmd == NULL)) { + DRM_ERROR("Out of fifo space for dummy query.\n"); + return -ENOMEM; + } -static struct vm_operations_struct vmw_fifo_vm_ops = { - .fault = vmw_fifo_vm_fault, - .open = NULL, - .close = NULL -}; + cmd->header.id = SVGA_3D_CMD_WAIT_FOR_QUERY; + cmd->header.size = sizeof(cmd->body); + cmd->body.cid = cid; + cmd->body.type = SVGA3D_QUERYTYPE_OCCLUSION; + + if (bo->mem.mem_type == TTM_PL_VRAM) { + cmd->body.guestResult.gmrId = SVGA_GMR_FRAMEBUFFER; + cmd->body.guestResult.offset = bo->offset; + } else { + cmd->body.guestResult.gmrId = bo->mem.start; + cmd->body.guestResult.offset = 0; + } + + vmw_fifo_commit(dev_priv, sizeof(*cmd)); -int vmw_fifo_mmap(struct file *filp, struct vm_area_struct *vma) -{ - struct drm_file *file_priv; - struct vmw_private *dev_priv; - - file_priv = filp->private_data; - dev_priv = vmw_priv(file_priv->minor->dev); - - if (vma->vm_pgoff != (dev_priv->mmio_start >> PAGE_SHIFT) || - (vma->vm_end - vma->vm_start) != PAGE_SIZE) - return -EINVAL; - - vma->vm_flags &= ~(VM_WRITE | VM_MAYWRITE); - vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_SHARED; - vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); - vma->vm_page_prot = ttm_io_prot(TTM_PL_FLAG_UNCACHED, - vma->vm_page_prot); - vma->vm_ops = &vmw_fifo_vm_ops; return 0; } diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_gmr.c b/drivers/gpu/drm/vmwgfx/vmwgfx_gmr.c index de0c594..2952249 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_gmr.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_gmr.c @@ -1,6 +1,6 @@ /************************************************************************** * - * Copyright © 2009 VMware, Inc., Palo Alto, CA., USA + * Copyright © 2009-2011 VMware, Inc., Palo Alto, CA., USA * All Rights Reserved. * * Permission is hereby granted, free of charge, to any person obtaining a @@ -29,6 +29,97 @@ #include "drmP.h" #include "ttm/ttm_bo_driver.h" +#define VMW_PPN_SIZE (sizeof(unsigned long)) +/* A future safe maximum remap size. */ +#define VMW_PPN_PER_REMAP ((31 * 1024) / VMW_PPN_SIZE) + +static int vmw_gmr2_bind(struct vmw_private *dev_priv, + struct page *pages[], + unsigned long num_pages, + int gmr_id) +{ + SVGAFifoCmdDefineGMR2 define_cmd; + SVGAFifoCmdRemapGMR2 remap_cmd; + uint32_t *cmd; + uint32_t *cmd_orig; + uint32_t define_size = sizeof(define_cmd) + sizeof(*cmd); + uint32_t remap_num = num_pages / VMW_PPN_PER_REMAP + ((num_pages % VMW_PPN_PER_REMAP) > 0); + uint32_t remap_size = VMW_PPN_SIZE * num_pages + (sizeof(remap_cmd) + sizeof(*cmd)) * remap_num; + uint32_t remap_pos = 0; + uint32_t cmd_size = define_size + remap_size; + uint32_t i; + + cmd_orig = cmd = vmw_fifo_reserve(dev_priv, cmd_size); + if (unlikely(cmd == NULL)) + return -ENOMEM; + + define_cmd.gmrId = gmr_id; + define_cmd.numPages = num_pages; + + *cmd++ = SVGA_CMD_DEFINE_GMR2; + memcpy(cmd, &define_cmd, sizeof(define_cmd)); + cmd += sizeof(define_cmd) / sizeof(*cmd); + + /* + * Need to split the command if there are too many + * pages that goes into the gmr. + */ + + remap_cmd.gmrId = gmr_id; + remap_cmd.flags = (VMW_PPN_SIZE > sizeof(*cmd)) ? + SVGA_REMAP_GMR2_PPN64 : SVGA_REMAP_GMR2_PPN32; + + while (num_pages > 0) { + unsigned long nr = min(num_pages, (unsigned long)VMW_PPN_PER_REMAP); + + remap_cmd.offsetPages = remap_pos; + remap_cmd.numPages = nr; + + *cmd++ = SVGA_CMD_REMAP_GMR2; + memcpy(cmd, &remap_cmd, sizeof(remap_cmd)); + cmd += sizeof(remap_cmd) / sizeof(*cmd); + + for (i = 0; i < nr; ++i) { + if (VMW_PPN_SIZE <= 4) + *cmd = page_to_pfn(*pages++); + else + *((uint64_t *)cmd) = page_to_pfn(*pages++); + + cmd += VMW_PPN_SIZE / sizeof(*cmd); + } + + num_pages -= nr; + remap_pos += nr; + } + + BUG_ON(cmd != cmd_orig + cmd_size / sizeof(*cmd)); + + vmw_fifo_commit(dev_priv, cmd_size); + + return 0; +} + +static void vmw_gmr2_unbind(struct vmw_private *dev_priv, + int gmr_id) +{ + SVGAFifoCmdDefineGMR2 define_cmd; + uint32_t define_size = sizeof(define_cmd) + 4; + uint32_t *cmd; + + cmd = vmw_fifo_reserve(dev_priv, define_size); + if (unlikely(cmd == NULL)) { + DRM_ERROR("GMR2 unbind failed.\n"); + return; + } + define_cmd.gmrId = gmr_id; + define_cmd.numPages = 0; + + *cmd++ = SVGA_CMD_DEFINE_GMR2; + memcpy(cmd, &define_cmd, sizeof(define_cmd)); + + vmw_fifo_commit(dev_priv, define_size); +} + /** * FIXME: Adjust to the ttm lowmem / highmem storage to minimize * the number of used descriptors. @@ -170,6 +261,9 @@ int vmw_gmr_bind(struct vmw_private *dev_priv, struct list_head desc_pages; int ret; + if (likely(dev_priv->capabilities & SVGA_CAP_GMR2)) + return vmw_gmr2_bind(dev_priv, pages, num_pages, gmr_id); + if (unlikely(!(dev_priv->capabilities & SVGA_CAP_GMR))) return -EINVAL; @@ -192,6 +286,11 @@ int vmw_gmr_bind(struct vmw_private *dev_priv, void vmw_gmr_unbind(struct vmw_private *dev_priv, int gmr_id) { + 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(); diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_gmrid_manager.c b/drivers/gpu/drm/vmwgfx/vmwgfx_gmrid_manager.c index ac6e0d1..5f71715 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_gmrid_manager.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_gmrid_manager.c @@ -40,6 +40,8 @@ struct vmwgfx_gmrid_man { spinlock_t lock; struct ida gmr_ida; uint32_t max_gmr_ids; + uint32_t max_gmr_pages; + uint32_t used_gmr_pages; }; static int vmw_gmrid_man_get_node(struct ttm_mem_type_manager *man, @@ -49,33 +51,50 @@ static int vmw_gmrid_man_get_node(struct ttm_mem_type_manager *man, { struct vmwgfx_gmrid_man *gman = (struct vmwgfx_gmrid_man *)man->priv; - int ret; + int ret = 0; int id; mem->mm_node = NULL; - do { - if (unlikely(ida_pre_get(&gman->gmr_ida, GFP_KERNEL) == 0)) - return -ENOMEM; + spin_lock(&gman->lock); + + if (gman->max_gmr_pages > 0) { + gman->used_gmr_pages += bo->num_pages; + if (unlikely(gman->used_gmr_pages > gman->max_gmr_pages)) + goto out_err_locked; + } + do { + spin_unlock(&gman->lock); + if (unlikely(ida_pre_get(&gman->gmr_ida, GFP_KERNEL) == 0)) { + ret = -ENOMEM; + goto out_err; + } spin_lock(&gman->lock); - ret = ida_get_new(&gman->gmr_ida, &id); + ret = ida_get_new(&gman->gmr_ida, &id); if (unlikely(ret == 0 && id >= gman->max_gmr_ids)) { ida_remove(&gman->gmr_ida, id); - spin_unlock(&gman->lock); - return 0; + ret = 0; + goto out_err_locked; } - - spin_unlock(&gman->lock); - } while (ret == -EAGAIN); if (likely(ret == 0)) { mem->mm_node = gman; mem->start = id; - } + mem->num_pages = bo->num_pages; + } else + goto out_err_locked; + + spin_unlock(&gman->lock); + return 0; +out_err: + spin_lock(&gman->lock); +out_err_locked: + gman->used_gmr_pages -= bo->num_pages; + spin_unlock(&gman->lock); return ret; } @@ -88,6 +107,7 @@ static void vmw_gmrid_man_put_node(struct ttm_mem_type_manager *man, if (mem->mm_node) { spin_lock(&gman->lock); ida_remove(&gman->gmr_ida, mem->start); + gman->used_gmr_pages -= mem->num_pages; spin_unlock(&gman->lock); mem->mm_node = NULL; } @@ -96,6 +116,8 @@ static void vmw_gmrid_man_put_node(struct ttm_mem_type_manager *man, static int vmw_gmrid_man_init(struct ttm_mem_type_manager *man, unsigned long p_size) { + struct vmw_private *dev_priv = + container_of(man->bdev, struct vmw_private, bdev); struct vmwgfx_gmrid_man *gman = kzalloc(sizeof(*gman), GFP_KERNEL); @@ -103,6 +125,8 @@ 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; man->priv = (void *) gman; diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c b/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c index 570d577..66917c6 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c @@ -27,6 +27,7 @@ #include "vmwgfx_drv.h" #include "vmwgfx_drm.h" +#include "vmwgfx_kms.h" int vmw_getparam_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv) @@ -45,9 +46,6 @@ int vmw_getparam_ioctl(struct drm_device *dev, void *data, case DRM_VMW_PARAM_3D: param->value = vmw_fifo_have_3d(dev_priv) ? 1 : 0; break; - case DRM_VMW_PARAM_FIFO_OFFSET: - param->value = dev_priv->mmio_start; - break; case DRM_VMW_PARAM_HW_CAPS: param->value = dev_priv->capabilities; break; @@ -57,6 +55,19 @@ int vmw_getparam_ioctl(struct drm_device *dev, void *data, case DRM_VMW_PARAM_MAX_FB_SIZE: param->value = dev_priv->vram_size; 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; + + param->value = + ioread32(fifo_mem + + ((fifo->capabilities & + SVGA_FIFO_CAP_3D_HWVERSION_REVISED) ? + SVGA_FIFO_3D_HWVERSION_REVISED : + SVGA_FIFO_3D_HWVERSION)); + break; + } default: DRM_ERROR("Illegal vmwgfx get param request: %d\n", param->param); @@ -66,25 +77,253 @@ int vmw_getparam_ioctl(struct drm_device *dev, void *data, return 0; } -int vmw_fifo_debug_ioctl(struct drm_device *dev, void *data, + +int vmw_get_cap_3d_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv) { + struct drm_vmw_get_3d_cap_arg *arg = + (struct drm_vmw_get_3d_cap_arg *) data; struct vmw_private *dev_priv = vmw_priv(dev); - struct vmw_fifo_state *fifo_state = &dev_priv->fifo; - struct drm_vmw_fifo_debug_arg *arg = - (struct drm_vmw_fifo_debug_arg *)data; - __le32 __user *buffer = (__le32 __user *) - (unsigned long)arg->debug_buffer; + uint32_t size; + __le32 __iomem *fifo_mem; + void __user *buffer = (void __user *)((unsigned long)(arg->buffer)); + void *bounce; + int ret; - if (unlikely(fifo_state->last_buffer == NULL)) + 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 (arg->max_size < size) + size = arg->max_size; + + bounce = vmalloc(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); + + ret = copy_to_user(buffer, bounce, size); + vfree(bounce); + + if (unlikely(ret != 0)) + DRM_ERROR("Failed to report 3D caps info.\n"); + + return ret; +} + +int vmw_present_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile; + struct vmw_private *dev_priv = vmw_priv(dev); + struct drm_vmw_present_arg *arg = + (struct drm_vmw_present_arg *)data; + struct vmw_surface *surface; + struct vmw_master *vmaster = vmw_master(file_priv->master); + struct drm_vmw_rect __user *clips_ptr; + struct drm_vmw_rect *clips = NULL; + struct drm_mode_object *obj; + struct vmw_framebuffer *vfb; + uint32_t num_clips; + int ret; + + num_clips = arg->num_clips; + clips_ptr = (struct drm_vmw_rect *)(unsigned long)arg->clips_ptr; + + if (unlikely(num_clips == 0)) + return 0; + + if (clips_ptr == NULL) { + DRM_ERROR("Variable clips_ptr must be specified.\n"); + ret = -EINVAL; + goto out_clips; + } + + clips = kcalloc(num_clips, sizeof(*clips), GFP_KERNEL); + if (clips == NULL) { + DRM_ERROR("Failed to allocate clip rect list.\n"); + ret = -ENOMEM; + goto out_clips; + } + + ret = copy_from_user(clips, clips_ptr, num_clips * sizeof(*clips)); + if (ret) { + DRM_ERROR("Failed to copy clip rects from userspace.\n"); + ret = -EFAULT; + goto out_no_copy; + } + + ret = mutex_lock_interruptible(&dev->mode_config.mutex); + if (unlikely(ret != 0)) { + ret = -ERESTARTSYS; + goto out_no_mode_mutex; + } + + obj = drm_mode_object_find(dev, arg->fb_id, DRM_MODE_OBJECT_FB); + if (!obj) { + DRM_ERROR("Invalid framebuffer id.\n"); + ret = -EINVAL; + goto out_no_fb; + } + vfb = vmw_framebuffer_to_vfb(obj_to_fb(obj)); + + ret = ttm_read_lock(&vmaster->lock, true); + if (unlikely(ret != 0)) + goto out_no_ttm_lock; + + ret = vmw_user_surface_lookup_handle(dev_priv, tfile, arg->sid, + &surface); + if (ret) + goto out_no_surface; + + ret = vmw_kms_present(dev_priv, file_priv, + vfb, surface, arg->sid, + arg->dest_x, arg->dest_y, + clips, num_clips); + + /* vmw_user_surface_lookup takes one ref so does new_fb */ + vmw_surface_unreference(&surface); - if (arg->debug_buffer_size < fifo_state->last_data_size) { - arg->used_size = arg->debug_buffer_size; - arg->did_not_fit = 1; - } else { - arg->used_size = fifo_state->last_data_size; - arg->did_not_fit = 0; +out_no_surface: + ttm_read_unlock(&vmaster->lock); +out_no_ttm_lock: +out_no_fb: + mutex_unlock(&dev->mode_config.mutex); +out_no_mode_mutex: +out_no_copy: + kfree(clips); +out_clips: + return ret; +} + +int vmw_present_readback_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct vmw_private *dev_priv = vmw_priv(dev); + struct drm_vmw_present_readback_arg *arg = + (struct drm_vmw_present_readback_arg *)data; + struct drm_vmw_fence_rep __user *user_fence_rep = + (struct drm_vmw_fence_rep __user *) + (unsigned long)arg->fence_rep; + struct vmw_master *vmaster = vmw_master(file_priv->master); + struct drm_vmw_rect __user *clips_ptr; + struct drm_vmw_rect *clips = NULL; + struct drm_mode_object *obj; + struct vmw_framebuffer *vfb; + uint32_t num_clips; + int ret; + + num_clips = arg->num_clips; + clips_ptr = (struct drm_vmw_rect *)(unsigned long)arg->clips_ptr; + + if (unlikely(num_clips == 0)) + return 0; + + if (clips_ptr == NULL) { + DRM_ERROR("Argument clips_ptr must be specified.\n"); + ret = -EINVAL; + goto out_clips; + } + + clips = kcalloc(num_clips, sizeof(*clips), GFP_KERNEL); + if (clips == NULL) { + DRM_ERROR("Failed to allocate clip rect list.\n"); + ret = -ENOMEM; + goto out_clips; + } + + ret = copy_from_user(clips, clips_ptr, num_clips * sizeof(*clips)); + if (ret) { + DRM_ERROR("Failed to copy clip rects from userspace.\n"); + ret = -EFAULT; + goto out_no_copy; + } + + ret = mutex_lock_interruptible(&dev->mode_config.mutex); + if (unlikely(ret != 0)) { + ret = -ERESTARTSYS; + goto out_no_mode_mutex; + } + + obj = drm_mode_object_find(dev, arg->fb_id, DRM_MODE_OBJECT_FB); + if (!obj) { + DRM_ERROR("Invalid framebuffer id.\n"); + ret = -EINVAL; + goto out_no_fb; + } + + vfb = vmw_framebuffer_to_vfb(obj_to_fb(obj)); + if (!vfb->dmabuf) { + DRM_ERROR("Framebuffer not dmabuf backed.\n"); + ret = -EINVAL; + goto out_no_fb; } - return copy_to_user(buffer, fifo_state->last_buffer, arg->used_size); + + ret = ttm_read_lock(&vmaster->lock, true); + if (unlikely(ret != 0)) + goto out_no_ttm_lock; + + ret = vmw_kms_readback(dev_priv, file_priv, + vfb, user_fence_rep, + clips, num_clips); + + ttm_read_unlock(&vmaster->lock); +out_no_ttm_lock: +out_no_fb: + mutex_unlock(&dev->mode_config.mutex); +out_no_mode_mutex: +out_no_copy: + kfree(clips); +out_clips: + return ret; +} + + +/** + * vmw_fops_poll - wrapper around the drm_poll function + * + * @filp: See the linux fops poll documentation. + * @wait: See the linux fops poll documentation. + * + * Wrapper around the drm_poll function that makes sure the device is + * processing the fifo if drm_poll decides to wait. + */ +unsigned int vmw_fops_poll(struct file *filp, struct poll_table_struct *wait) +{ + struct drm_file *file_priv = filp->private_data; + struct vmw_private *dev_priv = + vmw_priv(file_priv->minor->dev); + + vmw_fifo_ping_host(dev_priv, SVGA_SYNC_GENERIC); + return drm_poll(filp, wait); +} + + +/** + * vmw_fops_read - wrapper around the drm_read function + * + * @filp: See the linux fops read documentation. + * @buffer: See the linux fops read documentation. + * @count: See the linux fops read documentation. + * offset: See the linux fops read documentation. + * + * Wrapper around the drm_read function that makes sure the device is + * processing the fifo if drm_read decides to wait. + */ +ssize_t vmw_fops_read(struct file *filp, char __user *buffer, + size_t count, loff_t *offset) +{ + struct drm_file *file_priv = filp->private_data; + struct vmw_private *dev_priv = + vmw_priv(file_priv->minor->dev); + + vmw_fifo_ping_host(dev_priv, SVGA_SYNC_GENERIC); + return drm_read(filp, buffer, count, offset); } diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c b/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c index e92298a..cabc95f 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c @@ -34,26 +34,33 @@ irqreturn_t vmw_irq_handler(DRM_IRQ_ARGS) { struct drm_device *dev = (struct drm_device *)arg; struct vmw_private *dev_priv = vmw_priv(dev); - uint32_t status; + uint32_t status, masked_status; spin_lock(&dev_priv->irq_lock); status = inl(dev_priv->io_start + VMWGFX_IRQSTATUS_PORT); + masked_status = status & dev_priv->irq_mask; spin_unlock(&dev_priv->irq_lock); - if (status & SVGA_IRQFLAG_ANY_FENCE) + if (likely(status)) + outl(status, dev_priv->io_start + VMWGFX_IRQSTATUS_PORT); + + if (!masked_status) + return IRQ_NONE; + + if (masked_status & (SVGA_IRQFLAG_ANY_FENCE | + SVGA_IRQFLAG_FENCE_GOAL)) { + vmw_fences_update(dev_priv->fman); wake_up_all(&dev_priv->fence_queue); - if (status & SVGA_IRQFLAG_FIFO_PROGRESS) + } + + if (masked_status & SVGA_IRQFLAG_FIFO_PROGRESS) wake_up_all(&dev_priv->fifo_queue); - if (likely(status)) { - outl(status, dev_priv->io_start + VMWGFX_IRQSTATUS_PORT); - return IRQ_HANDLED; - } - return IRQ_NONE; + return IRQ_HANDLED; } -static bool vmw_fifo_idle(struct vmw_private *dev_priv, uint32_t sequence) +static bool vmw_fifo_idle(struct vmw_private *dev_priv, uint32_t seqno) { uint32_t busy; @@ -64,43 +71,43 @@ static bool vmw_fifo_idle(struct vmw_private *dev_priv, uint32_t sequence) return (busy == 0); } -void vmw_update_sequence(struct vmw_private *dev_priv, +void vmw_update_seqno(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo_state) { __le32 __iomem *fifo_mem = dev_priv->mmio_virt; + uint32_t seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE); - uint32_t sequence = ioread32(fifo_mem + SVGA_FIFO_FENCE); - - if (dev_priv->last_read_sequence != sequence) { - dev_priv->last_read_sequence = sequence; - vmw_fence_pull(&fifo_state->fence_queue, sequence); + if (dev_priv->last_read_seqno != seqno) { + dev_priv->last_read_seqno = seqno; + vmw_marker_pull(&fifo_state->marker_queue, seqno); + vmw_fences_update(dev_priv->fman); } } -bool vmw_fence_signaled(struct vmw_private *dev_priv, - uint32_t sequence) +bool vmw_seqno_passed(struct vmw_private *dev_priv, + uint32_t seqno) { struct vmw_fifo_state *fifo_state; bool ret; - if (likely(dev_priv->last_read_sequence - sequence < VMW_FENCE_WRAP)) + if (likely(dev_priv->last_read_seqno - seqno < VMW_FENCE_WRAP)) return true; fifo_state = &dev_priv->fifo; - vmw_update_sequence(dev_priv, fifo_state); - if (likely(dev_priv->last_read_sequence - sequence < VMW_FENCE_WRAP)) + vmw_update_seqno(dev_priv, fifo_state); + if (likely(dev_priv->last_read_seqno - seqno < VMW_FENCE_WRAP)) return true; if (!(fifo_state->capabilities & SVGA_FIFO_CAP_FENCE) && - vmw_fifo_idle(dev_priv, sequence)) + vmw_fifo_idle(dev_priv, seqno)) return true; /** - * Then check if the sequence is higher than what we've actually + * Then check if the seqno is higher than what we've actually * emitted. Then the fence is stale and signaled. */ - ret = ((atomic_read(&dev_priv->fence_seq) - sequence) + ret = ((atomic_read(&dev_priv->marker_seq) - seqno) > VMW_FENCE_WRAP); return ret; @@ -109,7 +116,7 @@ bool vmw_fence_signaled(struct vmw_private *dev_priv, int vmw_fallback_wait(struct vmw_private *dev_priv, bool lazy, bool fifo_idle, - uint32_t sequence, + uint32_t seqno, bool interruptible, unsigned long timeout) { @@ -123,7 +130,7 @@ int vmw_fallback_wait(struct vmw_private *dev_priv, DEFINE_WAIT(__wait); wait_condition = (fifo_idle) ? &vmw_fifo_idle : - &vmw_fence_signaled; + &vmw_seqno_passed; /** * Block command submission while waiting for idle. @@ -131,14 +138,14 @@ int vmw_fallback_wait(struct vmw_private *dev_priv, if (fifo_idle) down_read(&fifo_state->rwsem); - signal_seq = atomic_read(&dev_priv->fence_seq); + signal_seq = atomic_read(&dev_priv->marker_seq); ret = 0; for (;;) { prepare_to_wait(&dev_priv->fence_queue, &__wait, (interruptible) ? TASK_INTERRUPTIBLE : TASK_UNINTERRUPTIBLE); - if (wait_condition(dev_priv, sequence)) + if (wait_condition(dev_priv, seqno)) break; if (time_after_eq(jiffies, end_jiffies)) { DRM_ERROR("SVGA device lockup.\n"); @@ -175,68 +182,110 @@ int vmw_fallback_wait(struct vmw_private *dev_priv, return ret; } -int vmw_wait_fence(struct vmw_private *dev_priv, - bool lazy, uint32_t sequence, - bool interruptible, unsigned long timeout) +void vmw_seqno_waiter_add(struct vmw_private *dev_priv) +{ + mutex_lock(&dev_priv->hw_mutex); + if (dev_priv->fence_queue_waiters++ == 0) { + unsigned long irq_flags; + + spin_lock_irqsave(&dev_priv->irq_lock, irq_flags); + outl(SVGA_IRQFLAG_ANY_FENCE, + dev_priv->io_start + VMWGFX_IRQSTATUS_PORT); + dev_priv->irq_mask |= SVGA_IRQFLAG_ANY_FENCE; + vmw_write(dev_priv, SVGA_REG_IRQMASK, dev_priv->irq_mask); + spin_unlock_irqrestore(&dev_priv->irq_lock, irq_flags); + } + mutex_unlock(&dev_priv->hw_mutex); +} + +void vmw_seqno_waiter_remove(struct vmw_private *dev_priv) +{ + mutex_lock(&dev_priv->hw_mutex); + if (--dev_priv->fence_queue_waiters == 0) { + unsigned long irq_flags; + + spin_lock_irqsave(&dev_priv->irq_lock, irq_flags); + dev_priv->irq_mask &= ~SVGA_IRQFLAG_ANY_FENCE; + vmw_write(dev_priv, SVGA_REG_IRQMASK, dev_priv->irq_mask); + spin_unlock_irqrestore(&dev_priv->irq_lock, irq_flags); + } + mutex_unlock(&dev_priv->hw_mutex); +} + + +void vmw_goal_waiter_add(struct vmw_private *dev_priv) +{ + mutex_lock(&dev_priv->hw_mutex); + if (dev_priv->goal_queue_waiters++ == 0) { + unsigned long irq_flags; + + spin_lock_irqsave(&dev_priv->irq_lock, irq_flags); + outl(SVGA_IRQFLAG_FENCE_GOAL, + dev_priv->io_start + VMWGFX_IRQSTATUS_PORT); + dev_priv->irq_mask |= SVGA_IRQFLAG_FENCE_GOAL; + vmw_write(dev_priv, SVGA_REG_IRQMASK, dev_priv->irq_mask); + spin_unlock_irqrestore(&dev_priv->irq_lock, irq_flags); + } + mutex_unlock(&dev_priv->hw_mutex); +} + +void vmw_goal_waiter_remove(struct vmw_private *dev_priv) +{ + mutex_lock(&dev_priv->hw_mutex); + if (--dev_priv->goal_queue_waiters == 0) { + unsigned long irq_flags; + + spin_lock_irqsave(&dev_priv->irq_lock, irq_flags); + dev_priv->irq_mask &= ~SVGA_IRQFLAG_FENCE_GOAL; + vmw_write(dev_priv, SVGA_REG_IRQMASK, dev_priv->irq_mask); + spin_unlock_irqrestore(&dev_priv->irq_lock, irq_flags); + } + mutex_unlock(&dev_priv->hw_mutex); +} + +int vmw_wait_seqno(struct vmw_private *dev_priv, + bool lazy, uint32_t seqno, + bool interruptible, unsigned long timeout) { long ret; - unsigned long irq_flags; struct vmw_fifo_state *fifo = &dev_priv->fifo; - if (likely(dev_priv->last_read_sequence - sequence < VMW_FENCE_WRAP)) + if (likely(dev_priv->last_read_seqno - seqno < VMW_FENCE_WRAP)) return 0; - if (likely(vmw_fence_signaled(dev_priv, sequence))) + if (likely(vmw_seqno_passed(dev_priv, seqno))) return 0; vmw_fifo_ping_host(dev_priv, SVGA_SYNC_GENERIC); if (!(fifo->capabilities & SVGA_FIFO_CAP_FENCE)) - return vmw_fallback_wait(dev_priv, lazy, true, sequence, + return vmw_fallback_wait(dev_priv, lazy, true, seqno, interruptible, timeout); if (!(dev_priv->capabilities & SVGA_CAP_IRQMASK)) - return vmw_fallback_wait(dev_priv, lazy, false, sequence, + return vmw_fallback_wait(dev_priv, lazy, false, seqno, interruptible, timeout); - mutex_lock(&dev_priv->hw_mutex); - if (atomic_add_return(1, &dev_priv->fence_queue_waiters) > 0) { - spin_lock_irqsave(&dev_priv->irq_lock, irq_flags); - outl(SVGA_IRQFLAG_ANY_FENCE, - dev_priv->io_start + VMWGFX_IRQSTATUS_PORT); - vmw_write(dev_priv, SVGA_REG_IRQMASK, - vmw_read(dev_priv, SVGA_REG_IRQMASK) | - SVGA_IRQFLAG_ANY_FENCE); - spin_unlock_irqrestore(&dev_priv->irq_lock, irq_flags); - } - mutex_unlock(&dev_priv->hw_mutex); + vmw_seqno_waiter_add(dev_priv); if (interruptible) ret = wait_event_interruptible_timeout (dev_priv->fence_queue, - vmw_fence_signaled(dev_priv, sequence), + vmw_seqno_passed(dev_priv, seqno), timeout); else ret = wait_event_timeout (dev_priv->fence_queue, - vmw_fence_signaled(dev_priv, sequence), + vmw_seqno_passed(dev_priv, seqno), timeout); + vmw_seqno_waiter_remove(dev_priv); + if (unlikely(ret == 0)) ret = -EBUSY; else if (likely(ret > 0)) ret = 0; - mutex_lock(&dev_priv->hw_mutex); - if (atomic_dec_and_test(&dev_priv->fence_queue_waiters)) { - spin_lock_irqsave(&dev_priv->irq_lock, irq_flags); - vmw_write(dev_priv, SVGA_REG_IRQMASK, - vmw_read(dev_priv, SVGA_REG_IRQMASK) & - ~SVGA_IRQFLAG_ANY_FENCE); - spin_unlock_irqrestore(&dev_priv->irq_lock, irq_flags); - } - mutex_unlock(&dev_priv->hw_mutex); - return ret; } @@ -273,25 +322,3 @@ void vmw_irq_uninstall(struct drm_device *dev) status = inl(dev_priv->io_start + VMWGFX_IRQSTATUS_PORT); outl(status, dev_priv->io_start + VMWGFX_IRQSTATUS_PORT); } - -#define VMW_FENCE_WAIT_TIMEOUT 3*HZ; - -int vmw_fence_wait_ioctl(struct drm_device *dev, void *data, - struct drm_file *file_priv) -{ - struct drm_vmw_fence_wait_arg *arg = - (struct drm_vmw_fence_wait_arg *)data; - unsigned long timeout; - - if (!arg->cookie_valid) { - arg->cookie_valid = 1; - arg->kernel_cookie = jiffies + VMW_FENCE_WAIT_TIMEOUT; - } - - timeout = jiffies; - if (time_after_eq(timeout, (unsigned long)arg->kernel_cookie)) - return -EBUSY; - - timeout = (unsigned long)arg->kernel_cookie - timeout; - return vmw_wait_fence(vmw_priv(dev), true, arg->sequence, true, timeout); -} diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c index 8a38c91..eb9735e 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c @@ -27,11 +27,47 @@ #include "vmwgfx_kms.h" + /* Might need a hrtimer here? */ #define VMWGFX_PRESENT_RATE ((HZ / 60 > 0) ? HZ / 60 : 1) -static int vmw_surface_dmabuf_pin(struct vmw_framebuffer *vfb); -static int vmw_surface_dmabuf_unpin(struct vmw_framebuffer *vfb); + +struct vmw_clip_rect { + int x1, x2, y1, y2; +}; + +/** + * 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, + int num_rects, + struct vmw_clip_rect clip, + SVGASignedRect *out_rects, + int *out_num) +{ + int i, k; + + for (i = 0, k = 0; i < num_rects; i++) { + int x1 = max_t(int, clip.x1, rects[i].x1); + int y1 = max_t(int, clip.y1, rects[i].y1); + int x2 = min_t(int, clip.x2, rects[i].x2); + int y2 = min_t(int, clip.y2, rects[i].y2); + + if (x1 >= x2) + continue; + if (y1 >= y2) + continue; + + out_rects[k].left = x1; + out_rects[k].top = y1; + out_rects[k].right = x2; + out_rects[k].bottom = y2; + k++; + } + + *out_num = k; +} void vmw_display_unit_cleanup(struct vmw_display_unit *du) { @@ -84,6 +120,43 @@ int vmw_cursor_update_image(struct vmw_private *dev_priv, return 0; } +int vmw_cursor_update_dmabuf(struct vmw_private *dev_priv, + struct vmw_dma_buffer *dmabuf, + u32 width, u32 height, + u32 hotspotX, u32 hotspotY) +{ + struct ttm_bo_kmap_obj map; + unsigned long kmap_offset; + unsigned long kmap_num; + void *virtual; + bool dummy; + int ret; + + kmap_offset = 0; + kmap_num = (width*height*4 + PAGE_SIZE - 1) >> PAGE_SHIFT; + + ret = ttm_bo_reserve(&dmabuf->base, true, false, false, 0); + if (unlikely(ret != 0)) { + DRM_ERROR("reserve failed\n"); + return -EINVAL; + } + + ret = ttm_bo_kmap(&dmabuf->base, kmap_offset, kmap_num, &map); + if (unlikely(ret != 0)) + goto err_unreserve; + + virtual = ttm_kmap_obj_virtual(&map, &dummy); + ret = vmw_cursor_update_image(dev_priv, virtual, width, height, + hotspotX, hotspotY); + + ttm_bo_kunmap(&map); +err_unreserve: + ttm_bo_unreserve(&dmabuf->base); + + return ret; +} + + void vmw_cursor_update_position(struct vmw_private *dev_priv, bool show, int x, int y) { @@ -107,24 +180,26 @@ int vmw_du_crtc_cursor_set(struct drm_crtc *crtc, struct drm_file *file_priv, struct vmw_dma_buffer *dmabuf = NULL; int ret; + /* A lot of the code assumes this */ + if (handle && (width != 64 || height != 64)) + return -EINVAL; + if (handle) { - ret = vmw_user_surface_lookup_handle(dev_priv, tfile, - handle, &surface); - if (!ret) { - if (!surface->snooper.image) { - DRM_ERROR("surface not suitable for cursor\n"); - return -EINVAL; - } - } else { - ret = vmw_user_dmabuf_lookup(tfile, - handle, &dmabuf); - if (ret) { - DRM_ERROR("failed to find surface or dmabuf: %i\n", ret); - return -EINVAL; - } + ret = vmw_user_lookup_handle(dev_priv, tfile, + handle, &surface, &dmabuf); + if (ret) { + DRM_ERROR("failed to find surface or dmabuf: %i\n", ret); + return -EINVAL; } } + /* need to do this before taking down old image */ + if (surface && !surface->snooper.image) { + DRM_ERROR("surface not suitable for cursor\n"); + vmw_surface_unreference(&surface); + return -EINVAL; + } + /* takedown old cursor */ if (du->cursor_surface) { du->cursor_surface->snooper.crtc = NULL; @@ -143,42 +218,19 @@ int vmw_du_crtc_cursor_set(struct drm_crtc *crtc, struct drm_file *file_priv, vmw_cursor_update_image(dev_priv, surface->snooper.image, 64, 64, du->hotspot_x, du->hotspot_y); } else if (dmabuf) { - struct ttm_bo_kmap_obj map; - unsigned long kmap_offset; - unsigned long kmap_num; - void *virtual; - bool dummy; - /* vmw_user_surface_lookup takes one reference */ du->cursor_dmabuf = dmabuf; - kmap_offset = 0; - kmap_num = (64*64*4) >> PAGE_SHIFT; - - ret = ttm_bo_reserve(&dmabuf->base, true, false, false, 0); - if (unlikely(ret != 0)) { - DRM_ERROR("reserve failed\n"); - return -EINVAL; - } - - ret = ttm_bo_kmap(&dmabuf->base, kmap_offset, kmap_num, &map); - if (unlikely(ret != 0)) - goto err_unreserve; - - virtual = ttm_kmap_obj_virtual(&map, &dummy); - vmw_cursor_update_image(dev_priv, virtual, 64, 64, - du->hotspot_x, du->hotspot_y); - - ttm_bo_kunmap(&map); -err_unreserve: - ttm_bo_unreserve(&dmabuf->base); - + ret = vmw_cursor_update_dmabuf(dev_priv, dmabuf, width, height, + du->hotspot_x, du->hotspot_y); } else { vmw_cursor_update_position(dev_priv, false, 0, 0); return 0; } - vmw_cursor_update_position(dev_priv, true, du->cursor_x, du->cursor_y); + vmw_cursor_update_position(dev_priv, true, + du->cursor_x + du->hotspot_x, + du->cursor_y + du->hotspot_y); return 0; } @@ -193,7 +245,8 @@ int vmw_du_crtc_cursor_move(struct drm_crtc *crtc, int x, int y) du->cursor_y = y + crtc->y; vmw_cursor_update_position(dev_priv, shown, - du->cursor_x, du->cursor_y); + du->cursor_x + du->hotspot_x, + du->cursor_y + du->hotspot_y); return 0; } @@ -214,7 +267,7 @@ void vmw_kms_cursor_snoop(struct vmw_surface *srf, SVGA3dCmdHeader header; SVGA3dCmdSurfaceDMA dma; } *cmd; - int ret; + int i, ret; cmd = container_of(header, struct vmw_dma_cmd, header); @@ -236,16 +289,19 @@ void vmw_kms_cursor_snoop(struct vmw_surface *srf, box_count = (cmd->header.size - sizeof(SVGA3dCmdSurfaceDMA)) / sizeof(SVGA3dCopyBox); - if (cmd->dma.guest.pitch != (64 * 4) || - cmd->dma.guest.ptr.offset % PAGE_SIZE || + if (cmd->dma.guest.ptr.offset % PAGE_SIZE || box->x != 0 || box->y != 0 || box->z != 0 || box->srcx != 0 || box->srcy != 0 || box->srcz != 0 || - box->w != 64 || box->h != 64 || box->d != 1 || - box_count != 1) { + box->d != 1 || box_count != 1) { /* TODO handle none page aligned offsets */ - /* TODO handle partial uploads and pitch != 256 */ - /* TODO handle more then one copy (size != 64) */ - DRM_ERROR("lazy programmer, can't handle weird stuff\n"); + /* TODO handle more dst & src != 0 */ + /* TODO handle more then one copy */ + DRM_ERROR("Cant snoop dma request for cursor!\n"); + DRM_ERROR("(%u, %u, %u) (%u, %u, %u) (%ux%ux%u) %u %u\n", + box->srcx, box->srcy, box->srcz, + box->x, box->y, box->z, + box->w, box->h, box->d, box_count, + cmd->dma.guest.ptr.offset); return; } @@ -264,7 +320,16 @@ void vmw_kms_cursor_snoop(struct vmw_surface *srf, virtual = ttm_kmap_obj_virtual(&map, &dummy); - memcpy(srf->snooper.image, virtual, 64*64*4); + if (box->w == 64 && cmd->dma.guest.pitch == 64*4) { + memcpy(srf->snooper.image, virtual, 64*64*4); + } else { + /* Image is unsigned pointer. */ + for (i = 0; i < box->h; i++) + memcpy(srf->snooper.image + i * 64, + virtual + i * cmd->dma.guest.pitch, + box->w * 4); + } + srf->snooper.age++; /* we can't call this function from this function since execbuf has @@ -329,41 +394,10 @@ struct vmw_framebuffer_surface { struct vmw_framebuffer base; struct vmw_surface *surface; struct vmw_dma_buffer *buffer; - struct delayed_work d_work; - struct mutex work_lock; - bool present_fs; struct list_head head; struct drm_master *master; }; -/** - * vmw_kms_idle_workqueues - Flush workqueues on this master - * - * @vmaster - Pointer identifying the master, for the surfaces of which - * we idle the dirty work queues. - * - * This function should be called with the ttm lock held in exclusive mode - * to idle all dirty work queues before the fifo is taken down. - * - * The work task may actually requeue itself, but after the flush returns we're - * sure that there's nothing to present, since the ttm lock is held in - * exclusive mode, so the fifo will never get used. - */ - -void vmw_kms_idle_workqueues(struct vmw_master *vmaster) -{ - struct vmw_framebuffer_surface *entry; - - mutex_lock(&vmaster->fb_surf_mutex); - list_for_each_entry(entry, &vmaster->fb_surf, head) { - if (cancel_delayed_work_sync(&entry->d_work)) - (void) entry->d_work.work.func(&entry->d_work.work); - - (void) cancel_delayed_work_sync(&entry->d_work); - } - mutex_unlock(&vmaster->fb_surf_mutex); -} - void vmw_framebuffer_surface_destroy(struct drm_framebuffer *framebuffer) { struct vmw_framebuffer_surface *vfbs = @@ -375,65 +409,158 @@ void vmw_framebuffer_surface_destroy(struct drm_framebuffer *framebuffer) list_del(&vfbs->head); mutex_unlock(&vmaster->fb_surf_mutex); - cancel_delayed_work_sync(&vfbs->d_work); drm_master_put(&vfbs->master); drm_framebuffer_cleanup(framebuffer); vmw_surface_unreference(&vfbs->surface); + ttm_base_object_unref(&vfbs->base.user_obj); kfree(vfbs); } -static void vmw_framebuffer_present_fs_callback(struct work_struct *work) +static int do_surface_dirty_sou(struct vmw_private *dev_priv, + struct drm_file *file_priv, + struct vmw_framebuffer *framebuffer, + unsigned flags, unsigned color, + struct drm_clip_rect *clips, + unsigned num_clips, int inc) { - struct delayed_work *d_work = - container_of(work, struct delayed_work, work); - struct vmw_framebuffer_surface *vfbs = - container_of(d_work, struct vmw_framebuffer_surface, d_work); - struct vmw_surface *surf = vfbs->surface; - struct drm_framebuffer *framebuffer = &vfbs->base.base; - struct vmw_private *dev_priv = vmw_priv(framebuffer->dev); + struct vmw_display_unit *units[VMWGFX_NUM_DISPLAY_UNITS]; + struct drm_clip_rect *clips_ptr; + struct drm_clip_rect *tmp; + struct drm_crtc *crtc; + size_t fifo_size; + int i, num_units; + int ret = 0; /* silence warning */ + int left, right, top, bottom; struct { SVGA3dCmdHeader header; - SVGA3dCmdPresent body; - SVGA3dCopyRect cr; + SVGA3dCmdBlitSurfaceToScreen body; } *cmd; + SVGASignedRect *blits; - /** - * Strictly we should take the ttm_lock in read mode before accessing - * the fifo, to make sure the fifo is present and up. However, - * instead we flush all workqueues under the ttm lock in exclusive mode - * before taking down the fifo. - */ - mutex_lock(&vfbs->work_lock); - if (!vfbs->present_fs) - goto out_unlock; + num_units = 0; + list_for_each_entry(crtc, &dev_priv->dev->mode_config.crtc_list, + head) { + if (crtc->fb != &framebuffer->base) + continue; + units[num_units++] = vmw_crtc_to_du(crtc); + } - cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd)); - if (unlikely(cmd == NULL)) - goto out_resched; - - cmd->header.id = cpu_to_le32(SVGA_3D_CMD_PRESENT); - cmd->header.size = cpu_to_le32(sizeof(cmd->body) + sizeof(cmd->cr)); - cmd->body.sid = cpu_to_le32(surf->res.id); - cmd->cr.x = cpu_to_le32(0); - cmd->cr.y = cpu_to_le32(0); - cmd->cr.srcx = cmd->cr.x; - cmd->cr.srcy = cmd->cr.y; - cmd->cr.w = cpu_to_le32(framebuffer->width); - cmd->cr.h = cpu_to_le32(framebuffer->height); - vfbs->present_fs = false; - vmw_fifo_commit(dev_priv, sizeof(*cmd)); -out_resched: - /** - * Will not re-add if already pending. - */ - schedule_delayed_work(&vfbs->d_work, VMWGFX_PRESENT_RATE); -out_unlock: - mutex_unlock(&vfbs->work_lock); -} + BUG_ON(!clips || !num_clips); + + tmp = kzalloc(sizeof(*tmp) * num_clips, GFP_KERNEL); + if (unlikely(tmp == NULL)) { + DRM_ERROR("Temporary cliprect memory alloc failed.\n"); + return -ENOMEM; + } + + fifo_size = sizeof(*cmd) + sizeof(SVGASignedRect) * num_clips; + cmd = kzalloc(fifo_size, GFP_KERNEL); + if (unlikely(cmd == NULL)) { + DRM_ERROR("Temporary fifo memory alloc failed.\n"); + ret = -ENOMEM; + goto out_free_tmp; + } + + /* setup blits pointer */ + blits = (SVGASignedRect *)&cmd[1]; + + /* initial clip region */ + left = clips->x1; + right = clips->x2; + top = clips->y1; + bottom = clips->y2; + + /* skip the first clip rect */ + for (i = 1, clips_ptr = clips + inc; + i < num_clips; i++, clips_ptr += inc) { + left = min_t(int, left, (int)clips_ptr->x1); + right = max_t(int, right, (int)clips_ptr->x2); + top = min_t(int, top, (int)clips_ptr->y1); + bottom = max_t(int, bottom, (int)clips_ptr->y2); + } + + /* only need to do this once */ + memset(cmd, 0, fifo_size); + cmd->header.id = cpu_to_le32(SVGA_3D_CMD_BLIT_SURFACE_TO_SCREEN); + cmd->header.size = cpu_to_le32(fifo_size - sizeof(cmd->header)); + + cmd->body.srcRect.left = left; + cmd->body.srcRect.right = right; + cmd->body.srcRect.top = top; + cmd->body.srcRect.bottom = bottom; + + clips_ptr = clips; + for (i = 0; i < num_clips; i++, clips_ptr += inc) { + tmp[i].x1 = clips_ptr->x1 - left; + tmp[i].x2 = clips_ptr->x2 - left; + tmp[i].y1 = clips_ptr->y1 - top; + tmp[i].y2 = clips_ptr->y2 - top; + } + + /* do per unit writing, reuse fifo for each */ + for (i = 0; i < num_units; i++) { + struct vmw_display_unit *unit = units[i]; + struct vmw_clip_rect clip; + int num; + + clip.x1 = left - unit->crtc.x; + clip.y1 = top - unit->crtc.y; + clip.x2 = right - unit->crtc.x; + clip.y2 = bottom - unit->crtc.y; + + /* skip any crtcs that misses the clip region */ + if (clip.x1 >= unit->crtc.mode.hdisplay || + clip.y1 >= unit->crtc.mode.vdisplay || + clip.x2 <= 0 || clip.y2 <= 0) + continue; + + /* + * In order for the clip rects to be correctly scaled + * the src and dest rects needs to be the same size. + */ + cmd->body.destRect.left = clip.x1; + cmd->body.destRect.right = clip.x2; + cmd->body.destRect.top = clip.y1; + cmd->body.destRect.bottom = clip.y2; + + /* create a clip rect of the crtc in dest coords */ + clip.x2 = unit->crtc.mode.hdisplay - clip.x1; + clip.y2 = unit->crtc.mode.vdisplay - clip.y1; + clip.x1 = 0 - clip.x1; + clip.y1 = 0 - clip.y1; + + /* need to reset sid as it is changed by execbuf */ + cmd->body.srcImage.sid = cpu_to_le32(framebuffer->user_handle); + cmd->body.destScreenId = unit->unit; + + /* clip and write blits to cmd stream */ + vmw_clip_cliprects(tmp, num_clips, clip, blits, &num); + + /* if no cliprects hit skip this */ + if (num == 0) + continue; + /* recalculate package length */ + fifo_size = sizeof(*cmd) + sizeof(SVGASignedRect) * num; + cmd->header.size = cpu_to_le32(fifo_size - sizeof(cmd->header)); + ret = vmw_execbuf_process(file_priv, dev_priv, NULL, cmd, + fifo_size, 0, NULL); + + if (unlikely(ret != 0)) + break; + } + + + kfree(cmd); +out_free_tmp: + kfree(tmp); + + return ret; +} + int vmw_framebuffer_surface_dirty(struct drm_framebuffer *framebuffer, struct drm_file *file_priv, unsigned flags, unsigned color, @@ -444,44 +571,20 @@ int vmw_framebuffer_surface_dirty(struct drm_framebuffer *framebuffer, struct vmw_master *vmaster = vmw_master(file_priv->master); struct vmw_framebuffer_surface *vfbs = vmw_framebuffer_to_vfbs(framebuffer); - struct vmw_surface *surf = vfbs->surface; struct drm_clip_rect norect; - SVGA3dCopyRect *cr; - int i, inc = 1; - int ret; - - struct { - SVGA3dCmdHeader header; - SVGA3dCmdPresent body; - SVGA3dCopyRect cr; - } *cmd; + int ret, inc = 1; if (unlikely(vfbs->master != file_priv->master)) return -EINVAL; + /* Require ScreenObject support for 3D */ + if (!dev_priv->sou_priv) + return -EINVAL; + ret = ttm_read_lock(&vmaster->lock, true); if (unlikely(ret != 0)) return ret; - if (!num_clips || - !(dev_priv->fifo.capabilities & - SVGA_FIFO_CAP_SCREEN_OBJECT)) { - int ret; - - mutex_lock(&vfbs->work_lock); - vfbs->present_fs = true; - ret = schedule_delayed_work(&vfbs->d_work, VMWGFX_PRESENT_RATE); - mutex_unlock(&vfbs->work_lock); - if (ret) { - /** - * No work pending, Force immediate present. - */ - vmw_framebuffer_present_fs_callback(&vfbs->d_work.work); - } - ttm_read_unlock(&vmaster->lock); - return 0; - } - if (!num_clips) { num_clips = 1; clips = &norect; @@ -493,29 +596,10 @@ int vmw_framebuffer_surface_dirty(struct drm_framebuffer *framebuffer, inc = 2; /* skip source rects */ } - cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd) + (num_clips - 1) * sizeof(cmd->cr)); - if (unlikely(cmd == NULL)) { - DRM_ERROR("Fifo reserve failed.\n"); - ttm_read_unlock(&vmaster->lock); - return -ENOMEM; - } - - memset(cmd, 0, sizeof(*cmd)); - - cmd->header.id = cpu_to_le32(SVGA_3D_CMD_PRESENT); - cmd->header.size = cpu_to_le32(sizeof(cmd->body) + num_clips * sizeof(cmd->cr)); - cmd->body.sid = cpu_to_le32(surf->res.id); - - for (i = 0, cr = &cmd->cr; i < num_clips; i++, cr++, clips += inc) { - cr->x = cpu_to_le16(clips->x1); - cr->y = cpu_to_le16(clips->y1); - cr->srcx = cr->x; - cr->srcy = cr->y; - cr->w = cpu_to_le16(clips->x2 - clips->x1); - cr->h = cpu_to_le16(clips->y2 - clips->y1); - } + ret = do_surface_dirty_sou(dev_priv, file_priv, &vfbs->base, + flags, color, + clips, num_clips, inc); - vmw_fifo_commit(dev_priv, sizeof(*cmd) + (num_clips - 1) * sizeof(cmd->cr)); ttm_read_unlock(&vmaster->lock); return 0; } @@ -540,10 +624,18 @@ static int vmw_kms_new_framebuffer_surface(struct vmw_private *dev_priv, struct vmw_master *vmaster = vmw_master(file_priv->master); int ret; + /* 3D is only supported on HWv8 hosts which supports screen objects */ + if (!dev_priv->sou_priv) + return -ENOSYS; + /* * Sanity checks. */ + /* Surface must be marked as a scanout. */ + if (unlikely(!surface->scanout)) + return -EINVAL; + if (unlikely(surface->mip_levels[0] != 1 || surface->num_sizes != 1 || surface->sizes[0].width < mode_cmd->width || @@ -567,6 +659,9 @@ static int vmw_kms_new_framebuffer_surface(struct vmw_private *dev_priv, case 15: format = SVGA3D_A1R5G5B5; break; + case 8: + format = SVGA3D_LUMINANCE8; + break; default: DRM_ERROR("Invalid color depth: %d\n", mode_cmd->depth); return -EINVAL; @@ -599,14 +694,11 @@ static int vmw_kms_new_framebuffer_surface(struct vmw_private *dev_priv, vfbs->base.base.depth = mode_cmd->depth; vfbs->base.base.width = mode_cmd->width; vfbs->base.base.height = mode_cmd->height; - vfbs->base.pin = &vmw_surface_dmabuf_pin; - vfbs->base.unpin = &vmw_surface_dmabuf_unpin; vfbs->surface = surface; + vfbs->base.user_handle = mode_cmd->handle; vfbs->master = drm_master_get(file_priv->master); - mutex_init(&vfbs->work_lock); mutex_lock(&vmaster->fb_surf_mutex); - INIT_DELAYED_WORK(&vfbs->d_work, &vmw_framebuffer_present_fs_callback); list_add_tail(&vfbs->head, &vmaster->fb_surf); mutex_unlock(&vmaster->fb_surf_mutex); @@ -641,48 +733,33 @@ void vmw_framebuffer_dmabuf_destroy(struct drm_framebuffer *framebuffer) drm_framebuffer_cleanup(framebuffer); vmw_dmabuf_unreference(&vfbd->buffer); + ttm_base_object_unref(&vfbd->base.user_obj); kfree(vfbd); } -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) +static int do_dmabuf_dirty_ldu(struct vmw_private *dev_priv, + struct vmw_framebuffer *framebuffer, + unsigned flags, unsigned color, + struct drm_clip_rect *clips, + unsigned num_clips, int increment) { - struct vmw_private *dev_priv = vmw_priv(framebuffer->dev); - struct vmw_master *vmaster = vmw_master(file_priv->master); - struct drm_clip_rect norect; - int ret; + size_t fifo_size; + int i; + struct { uint32_t header; SVGAFifoCmdUpdate body; } *cmd; - int i, increment = 1; - - ret = ttm_read_lock(&vmaster->lock, true); - if (unlikely(ret != 0)) - return ret; - if (!num_clips) { - num_clips = 1; - clips = &norect; - norect.x1 = norect.y1 = 0; - norect.x2 = framebuffer->width; - norect.y2 = framebuffer->height; - } else if (flags & DRM_MODE_FB_DIRTY_ANNOTATE_COPY) { - num_clips /= 2; - increment = 2; - } - - cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd) * num_clips); + fifo_size = sizeof(*cmd) * num_clips; + cmd = vmw_fifo_reserve(dev_priv, fifo_size); if (unlikely(cmd == NULL)) { DRM_ERROR("Fifo reserve failed.\n"); - ttm_read_unlock(&vmaster->lock); return -ENOMEM; } + memset(cmd, 0, fifo_size); for (i = 0; i < num_clips; i++, clips += increment) { cmd[i].header = cpu_to_le32(SVGA_CMD_UPDATE); cmd[i].body.x = cpu_to_le32(clips->x1); @@ -691,57 +768,196 @@ int vmw_framebuffer_dmabuf_dirty(struct drm_framebuffer *framebuffer, cmd[i].body.height = cpu_to_le32(clips->y2 - clips->y1); } - vmw_fifo_commit(dev_priv, sizeof(*cmd) * num_clips); - ttm_read_unlock(&vmaster->lock); - + vmw_fifo_commit(dev_priv, fifo_size); return 0; } -static struct drm_framebuffer_funcs vmw_framebuffer_dmabuf_funcs = { - .destroy = vmw_framebuffer_dmabuf_destroy, - .dirty = vmw_framebuffer_dmabuf_dirty, - .create_handle = vmw_framebuffer_create_handle, -}; - -static int vmw_surface_dmabuf_pin(struct vmw_framebuffer *vfb) +static int do_dmabuf_define_gmrfb(struct drm_file *file_priv, + struct vmw_private *dev_priv, + struct vmw_framebuffer *framebuffer) { - struct vmw_private *dev_priv = vmw_priv(vfb->base.dev); - struct vmw_framebuffer_surface *vfbs = - vmw_framebuffer_to_vfbs(&vfb->base); - unsigned long size = vfbs->base.base.pitch * vfbs->base.base.height; + int depth = framebuffer->base.depth; + size_t fifo_size; int ret; - vfbs->buffer = kzalloc(sizeof(*vfbs->buffer), GFP_KERNEL); - if (unlikely(vfbs->buffer == NULL)) + struct { + uint32_t header; + SVGAFifoCmdDefineGMRFB body; + } *cmd; + + /* Emulate RGBA support, contrary to svga_reg.h this is not + * supported by hosts. This is only a problem if we are reading + * this value later and expecting what we uploaded back. + */ + if (depth == 32) + depth = 24; + + fifo_size = sizeof(*cmd); + cmd = kmalloc(fifo_size, GFP_KERNEL); + if (unlikely(cmd == NULL)) { + DRM_ERROR("Failed to allocate temporary cmd buffer.\n"); return -ENOMEM; + } - vmw_overlay_pause_all(dev_priv); - ret = vmw_dmabuf_init(dev_priv, vfbs->buffer, size, - &vmw_vram_ne_placement, - false, &vmw_dmabuf_bo_free); - vmw_overlay_resume_all(dev_priv); + memset(cmd, 0, fifo_size); + cmd->header = SVGA_CMD_DEFINE_GMRFB; + cmd->body.format.bitsPerPixel = framebuffer->base.bits_per_pixel; + cmd->body.format.colorDepth = depth; + cmd->body.format.reserved = 0; + cmd->body.bytesPerLine = framebuffer->base.pitch; + cmd->body.ptr.gmrId = framebuffer->user_handle; + cmd->body.ptr.offset = 0; + + ret = vmw_execbuf_process(file_priv, dev_priv, NULL, cmd, + fifo_size, 0, NULL); + + kfree(cmd); + + return ret; +} + +static int do_dmabuf_dirty_sou(struct drm_file *file_priv, + struct vmw_private *dev_priv, + struct vmw_framebuffer *framebuffer, + unsigned flags, unsigned color, + struct drm_clip_rect *clips, + unsigned num_clips, int increment) +{ + struct vmw_display_unit *units[VMWGFX_NUM_DISPLAY_UNITS]; + struct drm_clip_rect *clips_ptr; + int i, k, num_units, ret; + struct drm_crtc *crtc; + size_t fifo_size; + + struct { + uint32_t header; + SVGAFifoCmdBlitGMRFBToScreen body; + } *blits; + + ret = do_dmabuf_define_gmrfb(file_priv, dev_priv, framebuffer); if (unlikely(ret != 0)) - vfbs->buffer = NULL; + return ret; /* define_gmrfb prints warnings */ + + fifo_size = sizeof(*blits) * num_clips; + blits = kmalloc(fifo_size, GFP_KERNEL); + if (unlikely(blits == NULL)) { + DRM_ERROR("Failed to allocate temporary cmd buffer.\n"); + return -ENOMEM; + } + + num_units = 0; + list_for_each_entry(crtc, &dev_priv->dev->mode_config.crtc_list, head) { + if (crtc->fb != &framebuffer->base) + continue; + units[num_units++] = vmw_crtc_to_du(crtc); + } + + for (k = 0; k < num_units; k++) { + struct vmw_display_unit *unit = units[k]; + int hit_num = 0; + + clips_ptr = clips; + for (i = 0; i < num_clips; i++, clips_ptr += increment) { + int clip_x1 = clips_ptr->x1 - unit->crtc.x; + int clip_y1 = clips_ptr->y1 - unit->crtc.y; + int clip_x2 = clips_ptr->x2 - unit->crtc.x; + int clip_y2 = clips_ptr->y2 - unit->crtc.y; + int move_x, move_y; + + /* skip any crtcs that misses the clip region */ + if (clip_x1 >= unit->crtc.mode.hdisplay || + clip_y1 >= unit->crtc.mode.vdisplay || + clip_x2 <= 0 || clip_y2 <= 0) + continue; + + /* clip size to crtc size */ + clip_x2 = min_t(int, clip_x2, unit->crtc.mode.hdisplay); + clip_y2 = min_t(int, clip_y2, unit->crtc.mode.vdisplay); + + /* translate both src and dest to bring clip into screen */ + move_x = min_t(int, clip_x1, 0); + move_y = min_t(int, clip_y1, 0); + + /* actual translate done here */ + blits[hit_num].header = SVGA_CMD_BLIT_GMRFB_TO_SCREEN; + blits[hit_num].body.destScreenId = unit->unit; + blits[hit_num].body.srcOrigin.x = clips_ptr->x1 - move_x; + blits[hit_num].body.srcOrigin.y = clips_ptr->y1 - move_y; + blits[hit_num].body.destRect.left = clip_x1 - move_x; + blits[hit_num].body.destRect.top = clip_y1 - move_y; + blits[hit_num].body.destRect.right = clip_x2; + blits[hit_num].body.destRect.bottom = clip_y2; + hit_num++; + } + + /* no clips hit the crtc */ + if (hit_num == 0) + continue; + + fifo_size = sizeof(*blits) * hit_num; + ret = vmw_execbuf_process(file_priv, dev_priv, NULL, blits, + fifo_size, 0, NULL); + + if (unlikely(ret != 0)) + break; + } + + kfree(blits); return ret; } -static int vmw_surface_dmabuf_unpin(struct vmw_framebuffer *vfb) +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 ttm_buffer_object *bo; - struct vmw_framebuffer_surface *vfbs = - vmw_framebuffer_to_vfbs(&vfb->base); + 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; - if (unlikely(vfbs->buffer == NULL)) - return 0; + ret = ttm_read_lock(&vmaster->lock, true); + if (unlikely(ret != 0)) + return ret; - bo = &vfbs->buffer->base; - ttm_bo_unref(&bo); - vfbs->buffer = NULL; + if (!num_clips) { + num_clips = 1; + clips = &norect; + norect.x1 = norect.y1 = 0; + norect.x2 = framebuffer->width; + norect.y2 = framebuffer->height; + } else if (flags & DRM_MODE_FB_DIRTY_ANNOTATE_COPY) { + num_clips /= 2; + increment = 2; + } - return 0; + if (dev_priv->ldu_priv) { + ret = do_dmabuf_dirty_ldu(dev_priv, &vfbd->base, + flags, color, + clips, num_clips, increment); + } else { + ret = do_dmabuf_dirty_sou(file_priv, dev_priv, &vfbd->base, + flags, color, + clips, num_clips, increment); + } + + ttm_read_unlock(&vmaster->lock); + return ret; } +static struct drm_framebuffer_funcs vmw_framebuffer_dmabuf_funcs = { + .destroy = vmw_framebuffer_dmabuf_destroy, + .dirty = vmw_framebuffer_dmabuf_dirty, + .create_handle = vmw_framebuffer_create_handle, +}; + +/** + * Pin the dmabuffer to the start of vram. + */ static int vmw_framebuffer_dmabuf_pin(struct vmw_framebuffer *vfb) { struct vmw_private *dev_priv = vmw_priv(vfb->base.dev); @@ -749,10 +965,12 @@ static int vmw_framebuffer_dmabuf_pin(struct vmw_framebuffer *vfb) vmw_framebuffer_to_vfbd(&vfb->base); int ret; + /* This code should not be used with screen objects */ + BUG_ON(dev_priv->sou_priv); vmw_overlay_pause_all(dev_priv); - ret = vmw_dmabuf_to_start_of_vram(dev_priv, vfbd->buffer); + ret = vmw_dmabuf_to_start_of_vram(dev_priv, vfbd->buffer, true, false); vmw_overlay_resume_all(dev_priv); @@ -772,7 +990,7 @@ static int vmw_framebuffer_dmabuf_unpin(struct vmw_framebuffer *vfb) return 0; } - return vmw_dmabuf_from_vram(dev_priv, vfbd->buffer); + return vmw_dmabuf_unpin(dev_priv, vfbd->buffer, false); } static int vmw_kms_new_framebuffer_dmabuf(struct vmw_private *dev_priv, @@ -794,6 +1012,33 @@ static int vmw_kms_new_framebuffer_dmabuf(struct vmw_private *dev_priv, return -EINVAL; } + /* Limited framebuffer color depth support for screen objects */ + if (dev_priv->sou_priv) { + switch (mode_cmd->depth) { + case 32: + case 24: + /* Only support 32 bpp for 32 and 24 depth fbs */ + if (mode_cmd->bpp == 32) + break; + + DRM_ERROR("Invalid color depth/bbp: %d %d\n", + mode_cmd->depth, mode_cmd->bpp); + return -EINVAL; + case 16: + case 15: + /* Only support 16 bpp for 16 and 15 depth fbs */ + if (mode_cmd->bpp == 16) + break; + + DRM_ERROR("Invalid color depth/bbp: %d %d\n", + mode_cmd->depth, mode_cmd->bpp); + return -EINVAL; + default: + DRM_ERROR("Invalid color depth: %d\n", mode_cmd->depth); + return -EINVAL; + } + } + vfbd = kzalloc(sizeof(*vfbd), GFP_KERNEL); if (!vfbd) { ret = -ENOMEM; @@ -815,9 +1060,13 @@ static int vmw_kms_new_framebuffer_dmabuf(struct vmw_private *dev_priv, vfbd->base.base.depth = mode_cmd->depth; vfbd->base.base.width = mode_cmd->width; vfbd->base.base.height = mode_cmd->height; - vfbd->base.pin = vmw_framebuffer_dmabuf_pin; - vfbd->base.unpin = vmw_framebuffer_dmabuf_unpin; + if (!dev_priv->sou_priv) { + vfbd->base.pin = vmw_framebuffer_dmabuf_pin; + vfbd->base.unpin = vmw_framebuffer_dmabuf_unpin; + } + vfbd->base.dmabuf = true; vfbd->buffer = dmabuf; + vfbd->base.user_handle = mode_cmd->handle; *out = &vfbd->base; return 0; @@ -843,7 +1092,7 @@ static struct drm_framebuffer *vmw_kms_fb_create(struct drm_device *dev, struct vmw_framebuffer *vfb = NULL; struct vmw_surface *surface = NULL; struct vmw_dma_buffer *bo = NULL; - u64 required_size; + struct ttm_base_object *user_obj; int ret; /** @@ -852,69 +1101,308 @@ static struct drm_framebuffer *vmw_kms_fb_create(struct drm_device *dev, * requested framebuffer. */ - required_size = mode_cmd->pitch * mode_cmd->height; - if (unlikely(required_size > (u64) dev_priv->vram_size)) { + if (!vmw_kms_validate_mode_vram(dev_priv, + mode_cmd->pitch, + mode_cmd->height)) { DRM_ERROR("VRAM size is too small for requested mode.\n"); - return NULL; + return ERR_PTR(-ENOMEM); } - /** - * End conditioned code. + /* + * Take a reference on the user object of the resource + * backing the kms fb. This ensures that user-space handle + * lookups on that resource will always work as long as + * it's registered with a kms framebuffer. This is important, + * since vmw_execbuf_process identifies resources in the + * command stream using user-space handles. */ - ret = vmw_user_surface_lookup_handle(dev_priv, tfile, - mode_cmd->handle, &surface); - if (ret) - goto try_dmabuf; - - if (!surface->scanout) - goto err_not_scanout; - - ret = vmw_kms_new_framebuffer_surface(dev_priv, file_priv, surface, - &vfb, mode_cmd); + user_obj = ttm_base_object_lookup(tfile, mode_cmd->handle); + if (unlikely(user_obj == NULL)) { + DRM_ERROR("Could not locate requested kms frame buffer.\n"); + return ERR_PTR(-ENOENT); + } - /* vmw_user_surface_lookup takes one ref so does new_fb */ - vmw_surface_unreference(&surface); + /* returns either a dmabuf or surface */ + ret = vmw_user_lookup_handle(dev_priv, tfile, + mode_cmd->handle, + &surface, &bo); + if (ret) + goto err_out; + + /* Create the new framebuffer depending one what we got back */ + if (bo) + ret = vmw_kms_new_framebuffer_dmabuf(dev_priv, bo, &vfb, + mode_cmd); + else if (surface) + ret = vmw_kms_new_framebuffer_surface(dev_priv, file_priv, + surface, &vfb, mode_cmd); + else + BUG(); + +err_out: + /* vmw_user_lookup_handle takes one ref so does new_fb */ + if (bo) + vmw_dmabuf_unreference(&bo); + if (surface) + vmw_surface_unreference(&surface); if (ret) { DRM_ERROR("failed to create vmw_framebuffer: %i\n", ret); + ttm_base_object_unref(&user_obj); return ERR_PTR(ret); - } + } else + vfb->user_obj = user_obj; + return &vfb->base; +} -try_dmabuf: - DRM_INFO("%s: trying buffer\n", __func__); +static struct drm_mode_config_funcs vmw_kms_funcs = { + .fb_create = vmw_kms_fb_create, +}; - ret = vmw_user_dmabuf_lookup(tfile, mode_cmd->handle, &bo); - if (ret) { - DRM_ERROR("failed to find buffer: %i\n", ret); - return ERR_PTR(-ENOENT); +int vmw_kms_present(struct vmw_private *dev_priv, + struct drm_file *file_priv, + struct vmw_framebuffer *vfb, + struct vmw_surface *surface, + uint32_t sid, + int32_t destX, int32_t destY, + struct drm_vmw_rect *clips, + uint32_t num_clips) +{ + struct vmw_display_unit *units[VMWGFX_NUM_DISPLAY_UNITS]; + struct drm_clip_rect *tmp; + struct drm_crtc *crtc; + size_t fifo_size; + int i, k, num_units; + int ret = 0; /* silence warning */ + int left, right, top, bottom; + + struct { + SVGA3dCmdHeader header; + SVGA3dCmdBlitSurfaceToScreen body; + } *cmd; + SVGASignedRect *blits; + + num_units = 0; + list_for_each_entry(crtc, &dev_priv->dev->mode_config.crtc_list, head) { + if (crtc->fb != &vfb->base) + continue; + units[num_units++] = vmw_crtc_to_du(crtc); } - ret = vmw_kms_new_framebuffer_dmabuf(dev_priv, bo, &vfb, - mode_cmd); + BUG_ON(surface == NULL); + BUG_ON(!clips || !num_clips); - /* vmw_user_dmabuf_lookup takes one ref so does new_fb */ - vmw_dmabuf_unreference(&bo); + tmp = kzalloc(sizeof(*tmp) * num_clips, GFP_KERNEL); + if (unlikely(tmp == NULL)) { + DRM_ERROR("Temporary cliprect memory alloc failed.\n"); + return -ENOMEM; + } - if (ret) { - DRM_ERROR("failed to create vmw_framebuffer: %i\n", ret); - return ERR_PTR(ret); + fifo_size = sizeof(*cmd) + sizeof(SVGASignedRect) * num_clips; + cmd = kmalloc(fifo_size, GFP_KERNEL); + if (unlikely(cmd == NULL)) { + DRM_ERROR("Failed to allocate temporary fifo memory.\n"); + ret = -ENOMEM; + goto out_free_tmp; } - return &vfb->base; + left = clips->x; + right = clips->x + clips->w; + top = clips->y; + bottom = clips->y + clips->h; + + for (i = 1; i < num_clips; i++) { + left = min_t(int, left, (int)clips[i].x); + right = max_t(int, right, (int)clips[i].x + clips[i].w); + top = min_t(int, top, (int)clips[i].y); + bottom = max_t(int, bottom, (int)clips[i].y + clips[i].h); + } + + /* only need to do this once */ + memset(cmd, 0, fifo_size); + cmd->header.id = cpu_to_le32(SVGA_3D_CMD_BLIT_SURFACE_TO_SCREEN); + + blits = (SVGASignedRect *)&cmd[1]; -err_not_scanout: - DRM_ERROR("surface not marked as scanout\n"); - /* vmw_user_surface_lookup takes one ref */ - vmw_surface_unreference(&surface); + cmd->body.srcRect.left = left; + cmd->body.srcRect.right = right; + cmd->body.srcRect.top = top; + cmd->body.srcRect.bottom = bottom; - return ERR_PTR(-EINVAL); + for (i = 0; i < num_clips; i++) { + tmp[i].x1 = clips[i].x - left; + tmp[i].x2 = clips[i].x + clips[i].w - left; + tmp[i].y1 = clips[i].y - top; + tmp[i].y2 = clips[i].y + clips[i].h - top; + } + + for (k = 0; k < num_units; k++) { + struct vmw_display_unit *unit = units[k]; + struct vmw_clip_rect clip; + int num; + + clip.x1 = left + destX - unit->crtc.x; + clip.y1 = top + destY - unit->crtc.y; + clip.x2 = right + destX - unit->crtc.x; + clip.y2 = bottom + destY - unit->crtc.y; + + /* skip any crtcs that misses the clip region */ + if (clip.x1 >= unit->crtc.mode.hdisplay || + clip.y1 >= unit->crtc.mode.vdisplay || + clip.x2 <= 0 || clip.y2 <= 0) + continue; + + /* + * In order for the clip rects to be correctly scaled + * the src and dest rects needs to be the same size. + */ + cmd->body.destRect.left = clip.x1; + cmd->body.destRect.right = clip.x2; + cmd->body.destRect.top = clip.y1; + cmd->body.destRect.bottom = clip.y2; + + /* create a clip rect of the crtc in dest coords */ + clip.x2 = unit->crtc.mode.hdisplay - clip.x1; + clip.y2 = unit->crtc.mode.vdisplay - clip.y1; + clip.x1 = 0 - clip.x1; + clip.y1 = 0 - clip.y1; + + /* need to reset sid as it is changed by execbuf */ + cmd->body.srcImage.sid = sid; + cmd->body.destScreenId = unit->unit; + + /* clip and write blits to cmd stream */ + vmw_clip_cliprects(tmp, num_clips, clip, blits, &num); + + /* if no cliprects hit skip this */ + if (num == 0) + continue; + + /* recalculate package length */ + fifo_size = sizeof(*cmd) + sizeof(SVGASignedRect) * num; + cmd->header.size = cpu_to_le32(fifo_size - sizeof(cmd->header)); + ret = vmw_execbuf_process(file_priv, dev_priv, NULL, cmd, + fifo_size, 0, NULL); + + if (unlikely(ret != 0)) + break; + } + + kfree(cmd); +out_free_tmp: + kfree(tmp); + + return ret; } -static struct drm_mode_config_funcs vmw_kms_funcs = { - .fb_create = vmw_kms_fb_create, -}; +int vmw_kms_readback(struct vmw_private *dev_priv, + struct drm_file *file_priv, + struct vmw_framebuffer *vfb, + struct drm_vmw_fence_rep __user *user_fence_rep, + struct drm_vmw_rect *clips, + uint32_t num_clips) +{ + struct vmw_framebuffer_dmabuf *vfbd = + vmw_framebuffer_to_vfbd(&vfb->base); + struct vmw_dma_buffer *dmabuf = vfbd->buffer; + struct vmw_display_unit *units[VMWGFX_NUM_DISPLAY_UNITS]; + struct drm_crtc *crtc; + size_t fifo_size; + int i, k, ret, num_units, blits_pos; + + struct { + uint32_t header; + SVGAFifoCmdDefineGMRFB body; + } *cmd; + struct { + uint32_t header; + SVGAFifoCmdBlitScreenToGMRFB body; + } *blits; + + num_units = 0; + list_for_each_entry(crtc, &dev_priv->dev->mode_config.crtc_list, head) { + if (crtc->fb != &vfb->base) + continue; + units[num_units++] = vmw_crtc_to_du(crtc); + } + + BUG_ON(dmabuf == NULL); + BUG_ON(!clips || !num_clips); + + /* take a safe guess at fifo size */ + fifo_size = sizeof(*cmd) + sizeof(*blits) * num_clips * num_units; + cmd = kmalloc(fifo_size, GFP_KERNEL); + if (unlikely(cmd == NULL)) { + DRM_ERROR("Failed to allocate temporary fifo memory.\n"); + return -ENOMEM; + } + + memset(cmd, 0, fifo_size); + cmd->header = SVGA_CMD_DEFINE_GMRFB; + cmd->body.format.bitsPerPixel = vfb->base.bits_per_pixel; + cmd->body.format.colorDepth = vfb->base.depth; + cmd->body.format.reserved = 0; + cmd->body.bytesPerLine = vfb->base.pitch; + cmd->body.ptr.gmrId = vfb->user_handle; + cmd->body.ptr.offset = 0; + + blits = (void *)&cmd[1]; + blits_pos = 0; + for (i = 0; i < num_units; i++) { + struct drm_vmw_rect *c = clips; + for (k = 0; k < num_clips; k++, c++) { + /* transform clip coords to crtc origin based coords */ + int clip_x1 = c->x - units[i]->crtc.x; + int clip_x2 = c->x - units[i]->crtc.x + c->w; + int clip_y1 = c->y - units[i]->crtc.y; + int clip_y2 = c->y - units[i]->crtc.y + c->h; + int dest_x = c->x; + int dest_y = c->y; + + /* compensate for clipping, we negate + * a negative number and add that. + */ + if (clip_x1 < 0) + dest_x += -clip_x1; + if (clip_y1 < 0) + dest_y += -clip_y1; + + /* clip */ + clip_x1 = max(clip_x1, 0); + clip_y1 = max(clip_y1, 0); + clip_x2 = min(clip_x2, units[i]->crtc.mode.hdisplay); + clip_y2 = min(clip_y2, units[i]->crtc.mode.vdisplay); + + /* and cull any rects that misses the crtc */ + if (clip_x1 >= units[i]->crtc.mode.hdisplay || + clip_y1 >= units[i]->crtc.mode.vdisplay || + clip_x2 <= 0 || clip_y2 <= 0) + continue; + + blits[blits_pos].header = SVGA_CMD_BLIT_SCREEN_TO_GMRFB; + blits[blits_pos].body.srcScreenId = units[i]->unit; + blits[blits_pos].body.destOrigin.x = dest_x; + blits[blits_pos].body.destOrigin.y = dest_y; + + blits[blits_pos].body.srcRect.left = clip_x1; + blits[blits_pos].body.srcRect.top = clip_y1; + blits[blits_pos].body.srcRect.right = clip_x2; + blits[blits_pos].body.srcRect.bottom = clip_y2; + blits_pos++; + } + } + /* reset size here and use calculated exact size from loops */ + fifo_size = sizeof(*cmd) + sizeof(*blits) * blits_pos; + + ret = vmw_execbuf_process(file_priv, dev_priv, NULL, cmd, fifo_size, + 0, user_fence_rep); + + kfree(cmd); + + return ret; +} int vmw_kms_init(struct vmw_private *dev_priv) { @@ -929,7 +1417,9 @@ int vmw_kms_init(struct vmw_private *dev_priv) dev->mode_config.max_width = 8192; dev->mode_config.max_height = 8192; - ret = vmw_kms_init_legacy_display_system(dev_priv); + ret = vmw_kms_init_screen_object_display(dev_priv); + if (ret) /* Fallback */ + (void)vmw_kms_init_legacy_display_system(dev_priv); return 0; } @@ -942,7 +1432,10 @@ int vmw_kms_close(struct vmw_private *dev_priv) * drm_encoder_cleanup which takes the lock we deadlock. */ drm_mode_config_cleanup(dev_priv->dev); - vmw_kms_close_legacy_display_system(dev_priv); + if (dev_priv->sou_priv) + vmw_kms_close_screen_object_display(dev_priv); + else + vmw_kms_close_legacy_display_system(dev_priv); return 0; } @@ -987,9 +1480,9 @@ out: return ret; } -void vmw_kms_write_svga(struct vmw_private *vmw_priv, +int vmw_kms_write_svga(struct vmw_private *vmw_priv, unsigned width, unsigned height, unsigned pitch, - unsigned bbp, unsigned depth) + unsigned bpp, unsigned depth) { if (vmw_priv->capabilities & SVGA_CAP_PITCHLOCK) vmw_write(vmw_priv, SVGA_REG_PITCHLOCK, pitch); @@ -997,11 +1490,15 @@ void vmw_kms_write_svga(struct vmw_private *vmw_priv, iowrite32(pitch, vmw_priv->mmio_virt + SVGA_FIFO_PITCHLOCK); vmw_write(vmw_priv, SVGA_REG_WIDTH, width); vmw_write(vmw_priv, SVGA_REG_HEIGHT, height); - vmw_write(vmw_priv, SVGA_REG_BITS_PER_PIXEL, bbp); - vmw_write(vmw_priv, SVGA_REG_DEPTH, depth); - vmw_write(vmw_priv, SVGA_REG_RED_MASK, 0x00ff0000); - vmw_write(vmw_priv, SVGA_REG_GREEN_MASK, 0x0000ff00); - vmw_write(vmw_priv, SVGA_REG_BLUE_MASK, 0x000000ff); + vmw_write(vmw_priv, SVGA_REG_BITS_PER_PIXEL, bpp); + + if (vmw_read(vmw_priv, SVGA_REG_DEPTH) != depth) { + DRM_ERROR("Invalid depth %u for %u bpp, host expects %u\n", + depth, bpp, vmw_read(vmw_priv, SVGA_REG_DEPTH)); + return -EINVAL; + } + + return 0; } int vmw_kms_save_vga(struct vmw_private *vmw_priv) @@ -1011,12 +1508,7 @@ int vmw_kms_save_vga(struct vmw_private *vmw_priv) vmw_priv->vga_width = vmw_read(vmw_priv, SVGA_REG_WIDTH); vmw_priv->vga_height = vmw_read(vmw_priv, SVGA_REG_HEIGHT); - vmw_priv->vga_depth = vmw_read(vmw_priv, SVGA_REG_DEPTH); vmw_priv->vga_bpp = vmw_read(vmw_priv, SVGA_REG_BITS_PER_PIXEL); - vmw_priv->vga_pseudo = vmw_read(vmw_priv, SVGA_REG_PSEUDOCOLOR); - vmw_priv->vga_red_mask = vmw_read(vmw_priv, SVGA_REG_RED_MASK); - vmw_priv->vga_blue_mask = vmw_read(vmw_priv, SVGA_REG_BLUE_MASK); - vmw_priv->vga_green_mask = vmw_read(vmw_priv, SVGA_REG_GREEN_MASK); if (vmw_priv->capabilities & SVGA_CAP_PITCHLOCK) vmw_priv->vga_pitchlock = vmw_read(vmw_priv, SVGA_REG_PITCHLOCK); @@ -1065,12 +1557,7 @@ int vmw_kms_restore_vga(struct vmw_private *vmw_priv) vmw_write(vmw_priv, SVGA_REG_WIDTH, vmw_priv->vga_width); vmw_write(vmw_priv, SVGA_REG_HEIGHT, vmw_priv->vga_height); - vmw_write(vmw_priv, SVGA_REG_DEPTH, vmw_priv->vga_depth); vmw_write(vmw_priv, SVGA_REG_BITS_PER_PIXEL, vmw_priv->vga_bpp); - vmw_write(vmw_priv, SVGA_REG_PSEUDOCOLOR, vmw_priv->vga_pseudo); - vmw_write(vmw_priv, SVGA_REG_RED_MASK, vmw_priv->vga_red_mask); - vmw_write(vmw_priv, SVGA_REG_GREEN_MASK, vmw_priv->vga_green_mask); - vmw_write(vmw_priv, SVGA_REG_BLUE_MASK, vmw_priv->vga_blue_mask); if (vmw_priv->capabilities & SVGA_CAP_PITCHLOCK) vmw_write(vmw_priv, SVGA_REG_PITCHLOCK, vmw_priv->vga_pitchlock); @@ -1095,6 +1582,322 @@ int vmw_kms_restore_vga(struct vmw_private *vmw_priv) return 0; } +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; +} + + +/** + * Function called by DRM code called with vbl_lock held. + */ +u32 vmw_get_vblank_counter(struct drm_device *dev, int crtc) +{ + return 0; +} + +/** + * Function called by DRM code called with vbl_lock held. + */ +int vmw_enable_vblank(struct drm_device *dev, int crtc) +{ + return -ENOSYS; +} + +/** + * Function called by DRM code called with vbl_lock held. + */ +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, + struct drm_vmw_rect *rects) +{ + struct drm_device *dev = dev_priv->dev; + struct vmw_display_unit *du; + struct drm_connector *con; + + mutex_lock(&dev->mode_config.mutex); + +#if 0 + { + unsigned int i; + + DRM_INFO("%s: new layout ", __func__); + for (i = 0; i < num; i++) + DRM_INFO("(%i, %i %ux%u) ", rects[i].x, rects[i].y, + rects[i].w, rects[i].h); + DRM_INFO("\n"); + } +#endif + + list_for_each_entry(con, &dev->mode_config.connector_list, head) { + du = vmw_connector_to_du(con); + if (num > du->unit) { + du->pref_width = rects[du->unit].w; + du->pref_height = rects[du->unit].h; + du->pref_active = true; + du->gui_x = rects[du->unit].x; + du->gui_y = rects[du->unit].y; + } else { + du->pref_width = 800; + du->pref_height = 600; + du->pref_active = false; + } + con->status = vmw_du_connector_detect(con, true); + } + + mutex_unlock(&dev->mode_config.mutex); + + return 0; +} + +void vmw_du_crtc_save(struct drm_crtc *crtc) +{ +} + +void vmw_du_crtc_restore(struct drm_crtc *crtc) +{ +} + +void vmw_du_crtc_gamma_set(struct drm_crtc *crtc, + u16 *r, u16 *g, u16 *b, + uint32_t start, uint32_t size) +{ + struct vmw_private *dev_priv = vmw_priv(crtc->dev); + int i; + + for (i = 0; i < size; i++) { + DRM_DEBUG("%d r/g/b = 0x%04x / 0x%04x / 0x%04x\n", i, + r[i], g[i], b[i]); + vmw_write(dev_priv, SVGA_PALETTE_BASE + i * 3 + 0, r[i] >> 8); + vmw_write(dev_priv, SVGA_PALETTE_BASE + i * 3 + 1, g[i] >> 8); + vmw_write(dev_priv, SVGA_PALETTE_BASE + i * 3 + 2, b[i] >> 8); + } +} + +void vmw_du_connector_dpms(struct drm_connector *connector, int mode) +{ +} + +void vmw_du_connector_save(struct drm_connector *connector) +{ +} + +void vmw_du_connector_restore(struct drm_connector *connector) +{ +} + +enum drm_connector_status +vmw_du_connector_detect(struct drm_connector *connector, bool force) +{ + uint32_t num_displays; + struct drm_device *dev = connector->dev; + struct vmw_private *dev_priv = vmw_priv(dev); + struct vmw_display_unit *du = vmw_connector_to_du(connector); + + mutex_lock(&dev_priv->hw_mutex); + num_displays = vmw_read(dev_priv, SVGA_REG_NUM_DISPLAYS); + mutex_unlock(&dev_priv->hw_mutex); + + return ((vmw_connector_to_du(connector)->unit < num_displays && + du->pref_active) ? + connector_status_connected : connector_status_disconnected); +} + +static struct drm_display_mode vmw_kms_connector_builtin[] = { + /* 640x480@60Hz */ + { DRM_MODE("640x480", DRM_MODE_TYPE_DRIVER, 25175, 640, 656, + 752, 800, 0, 480, 489, 492, 525, 0, + DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_NVSYNC) }, + /* 800x600@60Hz */ + { DRM_MODE("800x600", DRM_MODE_TYPE_DRIVER, 40000, 800, 840, + 968, 1056, 0, 600, 601, 605, 628, 0, + DRM_MODE_FLAG_PHSYNC | DRM_MODE_FLAG_PVSYNC) }, + /* 1024x768@60Hz */ + { DRM_MODE("1024x768", DRM_MODE_TYPE_DRIVER, 65000, 1024, 1048, + 1184, 1344, 0, 768, 771, 777, 806, 0, + DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_NVSYNC) }, + /* 1152x864@75Hz */ + { DRM_MODE("1152x864", DRM_MODE_TYPE_DRIVER, 108000, 1152, 1216, + 1344, 1600, 0, 864, 865, 868, 900, 0, + DRM_MODE_FLAG_PHSYNC | DRM_MODE_FLAG_PVSYNC) }, + /* 1280x768@60Hz */ + { DRM_MODE("1280x768", DRM_MODE_TYPE_DRIVER, 79500, 1280, 1344, + 1472, 1664, 0, 768, 771, 778, 798, 0, + DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_PVSYNC) }, + /* 1280x800@60Hz */ + { DRM_MODE("1280x800", DRM_MODE_TYPE_DRIVER, 83500, 1280, 1352, + 1480, 1680, 0, 800, 803, 809, 831, 0, + DRM_MODE_FLAG_PHSYNC | DRM_MODE_FLAG_NVSYNC) }, + /* 1280x960@60Hz */ + { DRM_MODE("1280x960", DRM_MODE_TYPE_DRIVER, 108000, 1280, 1376, + 1488, 1800, 0, 960, 961, 964, 1000, 0, + DRM_MODE_FLAG_PHSYNC | DRM_MODE_FLAG_PVSYNC) }, + /* 1280x1024@60Hz */ + { DRM_MODE("1280x1024", DRM_MODE_TYPE_DRIVER, 108000, 1280, 1328, + 1440, 1688, 0, 1024, 1025, 1028, 1066, 0, + DRM_MODE_FLAG_PHSYNC | DRM_MODE_FLAG_PVSYNC) }, + /* 1360x768@60Hz */ + { DRM_MODE("1360x768", DRM_MODE_TYPE_DRIVER, 85500, 1360, 1424, + 1536, 1792, 0, 768, 771, 777, 795, 0, + DRM_MODE_FLAG_PHSYNC | DRM_MODE_FLAG_PVSYNC) }, + /* 1440x1050@60Hz */ + { DRM_MODE("1400x1050", DRM_MODE_TYPE_DRIVER, 121750, 1400, 1488, + 1632, 1864, 0, 1050, 1053, 1057, 1089, 0, + DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_PVSYNC) }, + /* 1440x900@60Hz */ + { DRM_MODE("1440x900", DRM_MODE_TYPE_DRIVER, 106500, 1440, 1520, + 1672, 1904, 0, 900, 903, 909, 934, 0, + DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_PVSYNC) }, + /* 1600x1200@60Hz */ + { DRM_MODE("1600x1200", DRM_MODE_TYPE_DRIVER, 162000, 1600, 1664, + 1856, 2160, 0, 1200, 1201, 1204, 1250, 0, + DRM_MODE_FLAG_PHSYNC | DRM_MODE_FLAG_PVSYNC) }, + /* 1680x1050@60Hz */ + { DRM_MODE("1680x1050", DRM_MODE_TYPE_DRIVER, 146250, 1680, 1784, + 1960, 2240, 0, 1050, 1053, 1059, 1089, 0, + DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_PVSYNC) }, + /* 1792x1344@60Hz */ + { DRM_MODE("1792x1344", DRM_MODE_TYPE_DRIVER, 204750, 1792, 1920, + 2120, 2448, 0, 1344, 1345, 1348, 1394, 0, + DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_PVSYNC) }, + /* 1853x1392@60Hz */ + { DRM_MODE("1856x1392", DRM_MODE_TYPE_DRIVER, 218250, 1856, 1952, + 2176, 2528, 0, 1392, 1393, 1396, 1439, 0, + DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_PVSYNC) }, + /* 1920x1200@60Hz */ + { DRM_MODE("1920x1200", DRM_MODE_TYPE_DRIVER, 193250, 1920, 2056, + 2256, 2592, 0, 1200, 1203, 1209, 1245, 0, + DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_PVSYNC) }, + /* 1920x1440@60Hz */ + { DRM_MODE("1920x1440", DRM_MODE_TYPE_DRIVER, 234000, 1920, 2048, + 2256, 2600, 0, 1440, 1441, 1444, 1500, 0, + DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_PVSYNC) }, + /* 2560x1600@60Hz */ + { DRM_MODE("2560x1600", DRM_MODE_TYPE_DRIVER, 348500, 2560, 2752, + 3032, 3504, 0, 1600, 1603, 1609, 1658, 0, + DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_PVSYNC) }, + /* Terminate */ + { DRM_MODE("", 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) }, +}; + +/** + * vmw_guess_mode_timing - Provide fake timings for a + * 60Hz vrefresh mode. + * + * @mode - Pointer to a struct drm_display_mode with hdisplay and vdisplay + * members filled in. + */ +static void vmw_guess_mode_timing(struct drm_display_mode *mode) +{ + mode->hsync_start = mode->hdisplay + 50; + mode->hsync_end = mode->hsync_start + 50; + mode->htotal = mode->hsync_end + 50; + + mode->vsync_start = mode->vdisplay + 50; + mode->vsync_end = mode->vsync_start + 50; + mode->vtotal = mode->vsync_end + 50; + + mode->clock = (u32)mode->htotal * (u32)mode->vtotal / 100 * 6; + mode->vrefresh = drm_mode_vrefresh(mode); +} + + +int vmw_du_connector_fill_modes(struct drm_connector *connector, + uint32_t max_width, uint32_t max_height) +{ + struct vmw_display_unit *du = vmw_connector_to_du(connector); + struct drm_device *dev = connector->dev; + struct vmw_private *dev_priv = vmw_priv(dev); + struct drm_display_mode *mode = NULL; + struct drm_display_mode *bmode; + struct drm_display_mode prefmode = { DRM_MODE("preferred", + DRM_MODE_TYPE_DRIVER | DRM_MODE_TYPE_PREFERRED, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_PVSYNC) + }; + int i; + u32 assumed_bpp = 2; + + /* + * If using screen objects, then assume 32-bpp because that's what the + * SVGA device is assuming + */ + if (dev_priv->sou_priv) + assumed_bpp = 4; + + /* Add preferred mode */ + { + mode = drm_mode_duplicate(dev, &prefmode); + if (!mode) + return 0; + mode->hdisplay = du->pref_width; + mode->vdisplay = du->pref_height; + vmw_guess_mode_timing(mode); + + if (vmw_kms_validate_mode_vram(dev_priv, + mode->hdisplay * assumed_bpp, + mode->vdisplay)) { + drm_mode_probed_add(connector, mode); + } else { + drm_mode_destroy(dev, mode); + mode = NULL; + } + + if (du->pref_mode) { + list_del_init(&du->pref_mode->head); + drm_mode_destroy(dev, du->pref_mode); + } + + /* mode might be null here, this is intended */ + du->pref_mode = mode; + } + + for (i = 0; vmw_kms_connector_builtin[i].type != 0; i++) { + bmode = &vmw_kms_connector_builtin[i]; + if (bmode->hdisplay > max_width || + bmode->vdisplay > max_height) + continue; + + if (!vmw_kms_validate_mode_vram(dev_priv, + bmode->hdisplay * assumed_bpp, + bmode->vdisplay)) + continue; + + mode = drm_mode_duplicate(dev, bmode); + if (!mode) + return 0; + mode->vrefresh = drm_mode_vrefresh(mode); + + drm_mode_probed_add(connector, mode); + } + + /* Move the prefered mode first, help apps pick the right mode. */ + if (du->pref_mode) + list_move(&du->pref_mode->head, &connector->probed_modes); + + drm_mode_connector_list_update(connector); + + return 1; +} + +int vmw_du_connector_set_property(struct drm_connector *connector, + struct drm_property *property, + uint64_t val) +{ + return 0; +} + + int vmw_kms_update_layout_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv) { @@ -1106,6 +1909,8 @@ int vmw_kms_update_layout_ioctl(struct drm_device *dev, void *data, struct drm_vmw_rect *rects; unsigned rects_size; int ret; + int i; + struct drm_mode_config *mode_config = &dev->mode_config; ret = ttm_read_lock(&vmaster->lock, true); if (unlikely(ret != 0)) @@ -1113,12 +1918,13 @@ int vmw_kms_update_layout_ioctl(struct drm_device *dev, void *data, if (!arg->num_outputs) { struct drm_vmw_rect def_rect = {0, 0, 800, 600}; - vmw_kms_ldu_update_layout(dev_priv, 1, &def_rect); + vmw_du_update_layout(dev_priv, 1, &def_rect); goto out_unlock; } rects_size = arg->num_outputs * sizeof(struct drm_vmw_rect); - rects = kzalloc(rects_size, GFP_KERNEL); + rects = kcalloc(arg->num_outputs, sizeof(struct drm_vmw_rect), + GFP_KERNEL); if (unlikely(!rects)) { ret = -ENOMEM; goto out_unlock; @@ -1132,7 +1938,18 @@ int vmw_kms_update_layout_ioctl(struct drm_device *dev, void *data, goto out_free; } - vmw_kms_ldu_update_layout(dev_priv, arg->num_outputs, rects); + for (i = 0; i < arg->num_outputs; ++i) { + if (rects[i].x < 0 || + rects[i].y < 0 || + rects[i].x + rects[i].w > mode_config->max_width || + rects[i].y + rects[i].h > mode_config->max_height) { + DRM_ERROR("Invalid GUI layout.\n"); + ret = -EINVAL; + goto out_free; + } + } + + vmw_du_update_layout(dev_priv, arg->num_outputs, rects); out_free: kfree(rects); @@ -1140,15 +1957,3 @@ out_unlock: ttm_read_unlock(&vmaster->lock); return ret; } - -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; -} - -u32 vmw_get_vblank_counter(struct drm_device *dev, int crtc) -{ - return 0; -} diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h index 8a398a0..e1cb855 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.h @@ -31,6 +31,8 @@ #include "drmP.h" #include "vmwgfx_drv.h" +#define VMWGFX_NUM_DISPLAY_UNITS 8 + #define vmw_framebuffer_to_vfb(x) \ container_of(x, struct vmw_framebuffer, base) @@ -45,6 +47,9 @@ struct vmw_framebuffer { struct drm_framebuffer base; int (*pin)(struct vmw_framebuffer *fb); int (*unpin)(struct vmw_framebuffer *fb); + bool dmabuf; + struct ttm_base_object *user_obj; + uint32_t user_handle; }; @@ -57,9 +62,14 @@ struct vmw_framebuffer { int vmw_cursor_update_image(struct vmw_private *dev_priv, u32 *image, u32 width, u32 height, u32 hotspotX, u32 hotspotY); +int vmw_cursor_update_dmabuf(struct vmw_private *dev_priv, + struct vmw_dma_buffer *dmabuf, + u32 width, u32 height, + u32 hotspotX, u32 hotspotY); void vmw_cursor_update_position(struct vmw_private *dev_priv, bool show, int x, int y); + /** * Base class display unit. * @@ -83,22 +93,65 @@ struct vmw_display_unit { int hotspot_y; unsigned unit; + + /* + * Prefered mode tracking. + */ + unsigned pref_width; + unsigned pref_height; + bool pref_active; + struct drm_display_mode *pref_mode; + + /* + * Gui positioning + */ + int gui_x; + int gui_y; + bool is_implicit; }; +#define vmw_crtc_to_du(x) \ + container_of(x, struct vmw_display_unit, crtc) +#define vmw_connector_to_du(x) \ + container_of(x, struct vmw_display_unit, connector) + + /* * Shared display unit functions - vmwgfx_kms.c */ void vmw_display_unit_cleanup(struct vmw_display_unit *du); +void vmw_du_crtc_save(struct drm_crtc *crtc); +void vmw_du_crtc_restore(struct drm_crtc *crtc); +void vmw_du_crtc_gamma_set(struct drm_crtc *crtc, + u16 *r, u16 *g, u16 *b, + uint32_t start, uint32_t size); int vmw_du_crtc_cursor_set(struct drm_crtc *crtc, struct drm_file *file_priv, uint32_t handle, uint32_t width, uint32_t height); int vmw_du_crtc_cursor_move(struct drm_crtc *crtc, int x, int y); +void vmw_du_connector_dpms(struct drm_connector *connector, int mode); +void vmw_du_connector_save(struct drm_connector *connector); +void vmw_du_connector_restore(struct drm_connector *connector); +enum drm_connector_status +vmw_du_connector_detect(struct drm_connector *connector, bool force); +int vmw_du_connector_fill_modes(struct drm_connector *connector, + uint32_t max_width, uint32_t max_height); +int vmw_du_connector_set_property(struct drm_connector *connector, + struct drm_property *property, + uint64_t val); + /* * Legacy display unit functions - vmwgfx_ldu.c */ int vmw_kms_init_legacy_display_system(struct vmw_private *dev_priv); int vmw_kms_close_legacy_display_system(struct vmw_private *dev_priv); -int vmw_kms_ldu_update_layout(struct vmw_private *dev_priv, unsigned num, + +/* + * Screen Objects display functions - vmwgfx_scrn.c + */ +int vmw_kms_init_screen_object_display(struct vmw_private *dev_priv); +int vmw_kms_close_screen_object_display(struct vmw_private *dev_priv); +int vmw_kms_sou_update_layout(struct vmw_private *dev_priv, unsigned num, struct drm_vmw_rect *rects); #endif diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c b/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c index b3a2cd5..8f8dbd4 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c @@ -27,7 +27,6 @@ #include "vmwgfx_kms.h" -#define VMWGFX_LDU_NUM_DU 8 #define vmw_crtc_to_ldu(x) \ container_of(x, struct vmw_legacy_display_unit, base.crtc) @@ -51,11 +50,6 @@ struct vmw_legacy_display { struct vmw_legacy_display_unit { struct vmw_display_unit base; - unsigned pref_width; - unsigned pref_height; - bool pref_active; - struct drm_display_mode *pref_mode; - struct list_head active; }; @@ -71,20 +65,6 @@ static void vmw_ldu_destroy(struct vmw_legacy_display_unit *ldu) * Legacy Display Unit CRTC functions */ -static void vmw_ldu_crtc_save(struct drm_crtc *crtc) -{ -} - -static void vmw_ldu_crtc_restore(struct drm_crtc *crtc) -{ -} - -static void vmw_ldu_crtc_gamma_set(struct drm_crtc *crtc, - u16 *r, u16 *g, u16 *b, - uint32_t start, uint32_t size) -{ -} - static void vmw_ldu_crtc_destroy(struct drm_crtc *crtc) { vmw_ldu_destroy(vmw_crtc_to_ldu(crtc)); @@ -94,9 +74,10 @@ static int vmw_ldu_commit_list(struct vmw_private *dev_priv) { struct vmw_legacy_display *lds = dev_priv->ldu_priv; struct vmw_legacy_display_unit *entry; + struct vmw_display_unit *du = NULL; struct drm_framebuffer *fb = NULL; struct drm_crtc *crtc = NULL; - int i = 0; + int i = 0, ret; /* If there is no display topology the host just assumes * that the guest will set the same layout as the host. @@ -114,10 +95,8 @@ static int vmw_ldu_commit_list(struct vmw_private *dev_priv) return 0; fb = entry->base.crtc.fb; - vmw_kms_write_svga(dev_priv, w, h, fb->pitch, - fb->bits_per_pixel, fb->depth); - - return 0; + return vmw_kms_write_svga(dev_priv, w, h, fb->pitch, + fb->bits_per_pixel, fb->depth); } if (!list_empty(&lds->active)) { @@ -151,6 +130,25 @@ static int vmw_ldu_commit_list(struct vmw_private *dev_priv) lds->last_num_active = lds->num_active; + + /* Find the first du with a cursor. */ + list_for_each_entry(entry, &lds->active, active) { + du = &entry->base; + + if (!du->cursor_dmabuf) + continue; + + ret = vmw_cursor_update_dmabuf(dev_priv, + du->cursor_dmabuf, + 64, 64, + du->hotspot_x, + du->hotspot_y); + if (ret == 0) + break; + + DRM_ERROR("Could not update cursor image\n"); + } + return 0; } @@ -265,9 +263,7 @@ static int vmw_ldu_crtc_set_config(struct drm_mode_set *set) vmw_ldu_del_active(dev_priv, ldu); - vmw_ldu_commit_list(dev_priv); - - return 0; + return vmw_ldu_commit_list(dev_priv); } @@ -292,21 +288,20 @@ static int vmw_ldu_crtc_set_config(struct drm_mode_set *set) vmw_ldu_add_active(dev_priv, ldu, vfb); - vmw_ldu_commit_list(dev_priv); - - return 0; + return vmw_ldu_commit_list(dev_priv); } static struct drm_crtc_funcs vmw_legacy_crtc_funcs = { - .save = vmw_ldu_crtc_save, - .restore = vmw_ldu_crtc_restore, + .save = vmw_du_crtc_save, + .restore = vmw_du_crtc_restore, .cursor_set = vmw_du_crtc_cursor_set, .cursor_move = vmw_du_crtc_cursor_move, - .gamma_set = vmw_ldu_crtc_gamma_set, + .gamma_set = vmw_du_crtc_gamma_set, .destroy = vmw_ldu_crtc_destroy, .set_config = vmw_ldu_crtc_set_config, }; + /* * Legacy Display Unit encoder functions */ @@ -324,183 +319,18 @@ static struct drm_encoder_funcs vmw_legacy_encoder_funcs = { * Legacy Display Unit connector functions */ -static void vmw_ldu_connector_dpms(struct drm_connector *connector, int mode) -{ -} - -static void vmw_ldu_connector_save(struct drm_connector *connector) -{ -} - -static void vmw_ldu_connector_restore(struct drm_connector *connector) -{ -} - -static enum drm_connector_status - vmw_ldu_connector_detect(struct drm_connector *connector, - bool force) -{ - if (vmw_connector_to_ldu(connector)->pref_active) - return connector_status_connected; - return connector_status_disconnected; -} - -static const struct drm_display_mode vmw_ldu_connector_builtin[] = { - /* 640x480@60Hz */ - { DRM_MODE("640x480", DRM_MODE_TYPE_DRIVER, 25175, 640, 656, - 752, 800, 0, 480, 489, 492, 525, 0, - DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_NVSYNC) }, - /* 800x600@60Hz */ - { DRM_MODE("800x600", DRM_MODE_TYPE_DRIVER, 40000, 800, 840, - 968, 1056, 0, 600, 601, 605, 628, 0, - DRM_MODE_FLAG_PHSYNC | DRM_MODE_FLAG_PVSYNC) }, - /* 1024x768@60Hz */ - { DRM_MODE("1024x768", DRM_MODE_TYPE_DRIVER, 65000, 1024, 1048, - 1184, 1344, 0, 768, 771, 777, 806, 0, - DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_NVSYNC) }, - /* 1152x864@75Hz */ - { DRM_MODE("1152x864", DRM_MODE_TYPE_DRIVER, 108000, 1152, 1216, - 1344, 1600, 0, 864, 865, 868, 900, 0, - DRM_MODE_FLAG_PHSYNC | DRM_MODE_FLAG_PVSYNC) }, - /* 1280x768@60Hz */ - { DRM_MODE("1280x768", DRM_MODE_TYPE_DRIVER, 79500, 1280, 1344, - 1472, 1664, 0, 768, 771, 778, 798, 0, - DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_PVSYNC) }, - /* 1280x800@60Hz */ - { DRM_MODE("1280x800", DRM_MODE_TYPE_DRIVER, 83500, 1280, 1352, - 1480, 1680, 0, 800, 803, 809, 831, 0, - DRM_MODE_FLAG_PHSYNC | DRM_MODE_FLAG_NVSYNC) }, - /* 1280x960@60Hz */ - { DRM_MODE("1280x960", DRM_MODE_TYPE_DRIVER, 108000, 1280, 1376, - 1488, 1800, 0, 960, 961, 964, 1000, 0, - DRM_MODE_FLAG_PHSYNC | DRM_MODE_FLAG_PVSYNC) }, - /* 1280x1024@60Hz */ - { DRM_MODE("1280x1024", DRM_MODE_TYPE_DRIVER, 108000, 1280, 1328, - 1440, 1688, 0, 1024, 1025, 1028, 1066, 0, - DRM_MODE_FLAG_PHSYNC | DRM_MODE_FLAG_PVSYNC) }, - /* 1360x768@60Hz */ - { DRM_MODE("1360x768", DRM_MODE_TYPE_DRIVER, 85500, 1360, 1424, - 1536, 1792, 0, 768, 771, 777, 795, 0, - DRM_MODE_FLAG_PHSYNC | DRM_MODE_FLAG_PVSYNC) }, - /* 1440x1050@60Hz */ - { DRM_MODE("1400x1050", DRM_MODE_TYPE_DRIVER, 121750, 1400, 1488, - 1632, 1864, 0, 1050, 1053, 1057, 1089, 0, - DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_PVSYNC) }, - /* 1440x900@60Hz */ - { DRM_MODE("1440x900", DRM_MODE_TYPE_DRIVER, 106500, 1440, 1520, - 1672, 1904, 0, 900, 903, 909, 934, 0, - DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_PVSYNC) }, - /* 1600x1200@60Hz */ - { DRM_MODE("1600x1200", DRM_MODE_TYPE_DRIVER, 162000, 1600, 1664, - 1856, 2160, 0, 1200, 1201, 1204, 1250, 0, - DRM_MODE_FLAG_PHSYNC | DRM_MODE_FLAG_PVSYNC) }, - /* 1680x1050@60Hz */ - { DRM_MODE("1680x1050", DRM_MODE_TYPE_DRIVER, 146250, 1680, 1784, - 1960, 2240, 0, 1050, 1053, 1059, 1089, 0, - DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_PVSYNC) }, - /* 1792x1344@60Hz */ - { DRM_MODE("1792x1344", DRM_MODE_TYPE_DRIVER, 204750, 1792, 1920, - 2120, 2448, 0, 1344, 1345, 1348, 1394, 0, - DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_PVSYNC) }, - /* 1853x1392@60Hz */ - { DRM_MODE("1856x1392", DRM_MODE_TYPE_DRIVER, 218250, 1856, 1952, - 2176, 2528, 0, 1392, 1393, 1396, 1439, 0, - DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_PVSYNC) }, - /* 1920x1200@60Hz */ - { DRM_MODE("1920x1200", DRM_MODE_TYPE_DRIVER, 193250, 1920, 2056, - 2256, 2592, 0, 1200, 1203, 1209, 1245, 0, - DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_PVSYNC) }, - /* 1920x1440@60Hz */ - { DRM_MODE("1920x1440", DRM_MODE_TYPE_DRIVER, 234000, 1920, 2048, - 2256, 2600, 0, 1440, 1441, 1444, 1500, 0, - DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_PVSYNC) }, - /* 2560x1600@60Hz */ - { DRM_MODE("2560x1600", DRM_MODE_TYPE_DRIVER, 348500, 2560, 2752, - 3032, 3504, 0, 1600, 1603, 1609, 1658, 0, - DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_PVSYNC) }, - /* Terminate */ - { DRM_MODE("", 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) }, -}; - -static int vmw_ldu_connector_fill_modes(struct drm_connector *connector, - uint32_t max_width, uint32_t max_height) -{ - struct vmw_legacy_display_unit *ldu = vmw_connector_to_ldu(connector); - struct drm_device *dev = connector->dev; - struct vmw_private *dev_priv = vmw_priv(dev); - struct drm_display_mode *mode = NULL; - struct drm_display_mode prefmode = { DRM_MODE("preferred", - DRM_MODE_TYPE_DRIVER | DRM_MODE_TYPE_PREFERRED, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_PVSYNC) - }; - int i; - - /* Add preferred mode */ - { - mode = drm_mode_duplicate(dev, &prefmode); - if (!mode) - return 0; - mode->hdisplay = ldu->pref_width; - mode->vdisplay = ldu->pref_height; - mode->vrefresh = drm_mode_vrefresh(mode); - if (vmw_kms_validate_mode_vram(dev_priv, mode->hdisplay * 2, - mode->vdisplay)) { - drm_mode_probed_add(connector, mode); - - if (ldu->pref_mode) { - list_del_init(&ldu->pref_mode->head); - drm_mode_destroy(dev, ldu->pref_mode); - } - - ldu->pref_mode = mode; - } - } - - for (i = 0; vmw_ldu_connector_builtin[i].type != 0; i++) { - const struct drm_display_mode *bmode; - - bmode = &vmw_ldu_connector_builtin[i]; - if (bmode->hdisplay > max_width || - bmode->vdisplay > max_height) - continue; - - if (!vmw_kms_validate_mode_vram(dev_priv, bmode->hdisplay * 2, - bmode->vdisplay)) - continue; - - mode = drm_mode_duplicate(dev, bmode); - if (!mode) - return 0; - mode->vrefresh = drm_mode_vrefresh(mode); - - drm_mode_probed_add(connector, mode); - } - - drm_mode_connector_list_update(connector); - - return 1; -} - -static int vmw_ldu_connector_set_property(struct drm_connector *connector, - struct drm_property *property, - uint64_t val) -{ - return 0; -} - static void vmw_ldu_connector_destroy(struct drm_connector *connector) { vmw_ldu_destroy(vmw_connector_to_ldu(connector)); } static struct drm_connector_funcs vmw_legacy_connector_funcs = { - .dpms = vmw_ldu_connector_dpms, - .save = vmw_ldu_connector_save, - .restore = vmw_ldu_connector_restore, - .detect = vmw_ldu_connector_detect, - .fill_modes = vmw_ldu_connector_fill_modes, - .set_property = vmw_ldu_connector_set_property, + .dpms = vmw_du_connector_dpms, + .save = vmw_du_connector_save, + .restore = vmw_du_connector_restore, + .detect = vmw_du_connector_detect, + .fill_modes = vmw_du_connector_fill_modes, + .set_property = vmw_du_connector_set_property, .destroy = vmw_ldu_connector_destroy, }; @@ -523,23 +353,26 @@ static int vmw_ldu_init(struct vmw_private *dev_priv, unsigned unit) INIT_LIST_HEAD(&ldu->active); - ldu->pref_active = (unit == 0); - ldu->pref_width = 800; - ldu->pref_height = 600; - ldu->pref_mode = NULL; + ldu->base.pref_active = (unit == 0); + ldu->base.pref_width = 800; + ldu->base.pref_height = 600; + ldu->base.pref_mode = NULL; + ldu->base.is_implicit = true; drm_connector_init(dev, connector, &vmw_legacy_connector_funcs, - DRM_MODE_CONNECTOR_LVDS); - connector->status = vmw_ldu_connector_detect(connector, true); + DRM_MODE_CONNECTOR_VIRTUAL); + connector->status = vmw_du_connector_detect(connector, true); drm_encoder_init(dev, encoder, &vmw_legacy_encoder_funcs, - DRM_MODE_ENCODER_LVDS); + DRM_MODE_ENCODER_VIRTUAL); drm_mode_connector_attach_encoder(connector, encoder); encoder->possible_crtcs = (1 << unit); encoder->possible_clones = 0; drm_crtc_init(dev, crtc, &vmw_legacy_crtc_funcs); + drm_mode_crtc_set_gamma_size(crtc, 256); + drm_connector_attach_property(connector, dev->mode_config.dirty_info_property, 1); @@ -550,8 +383,7 @@ static int vmw_ldu_init(struct vmw_private *dev_priv, unsigned unit) int vmw_kms_init_legacy_display_system(struct vmw_private *dev_priv) { struct drm_device *dev = dev_priv->dev; - int i; - int ret; + int i, ret; if (dev_priv->ldu_priv) { DRM_INFO("ldu system already on\n"); @@ -559,7 +391,6 @@ int vmw_kms_init_legacy_display_system(struct vmw_private *dev_priv) } dev_priv->ldu_priv = kmalloc(sizeof(*dev_priv->ldu_priv), GFP_KERNEL); - if (!dev_priv->ldu_priv) return -ENOMEM; @@ -568,18 +399,31 @@ int vmw_kms_init_legacy_display_system(struct vmw_private *dev_priv) dev_priv->ldu_priv->last_num_active = 0; dev_priv->ldu_priv->fb = NULL; - drm_mode_create_dirty_info_property(dev_priv->dev); + /* for old hardware without multimon only enable one display */ + if (dev_priv->capabilities & SVGA_CAP_MULTIMON) + ret = drm_vblank_init(dev, VMWGFX_NUM_DISPLAY_UNITS); + else + ret = drm_vblank_init(dev, 1); + if (ret != 0) + goto err_free; + + ret = drm_mode_create_dirty_info_property(dev); + if (ret != 0) + goto err_vblank_cleanup; - if (dev_priv->capabilities & SVGA_CAP_MULTIMON) { - for (i = 0; i < VMWGFX_LDU_NUM_DU; ++i) + if (dev_priv->capabilities & SVGA_CAP_MULTIMON) + for (i = 0; i < VMWGFX_NUM_DISPLAY_UNITS; ++i) vmw_ldu_init(dev_priv, i); - ret = drm_vblank_init(dev, VMWGFX_LDU_NUM_DU); - } else { - /* for old hardware without multimon only enable one display */ + else vmw_ldu_init(dev_priv, 0); - ret = drm_vblank_init(dev, 1); - } + return 0; + +err_vblank_cleanup: + drm_vblank_cleanup(dev); +err_free: + kfree(dev_priv->ldu_priv); + dev_priv->ldu_priv = NULL; return ret; } @@ -587,52 +431,14 @@ int vmw_kms_close_legacy_display_system(struct vmw_private *dev_priv) { struct drm_device *dev = dev_priv->dev; - drm_vblank_cleanup(dev); if (!dev_priv->ldu_priv) return -ENOSYS; + drm_vblank_cleanup(dev); + BUG_ON(!list_empty(&dev_priv->ldu_priv->active)); kfree(dev_priv->ldu_priv); return 0; } - -int vmw_kms_ldu_update_layout(struct vmw_private *dev_priv, unsigned num, - struct drm_vmw_rect *rects) -{ - struct drm_device *dev = dev_priv->dev; - struct vmw_legacy_display_unit *ldu; - struct drm_connector *con; - int i; - - mutex_lock(&dev->mode_config.mutex); - -#if 0 - DRM_INFO("%s: new layout ", __func__); - for (i = 0; i < (int)num; i++) - DRM_INFO("(%i, %i %ux%u) ", rects[i].x, rects[i].y, - rects[i].w, rects[i].h); - DRM_INFO("\n"); -#else - (void)i; -#endif - - list_for_each_entry(con, &dev->mode_config.connector_list, head) { - ldu = vmw_connector_to_ldu(con); - if (num > ldu->base.unit) { - ldu->pref_width = rects[ldu->base.unit].w; - ldu->pref_height = rects[ldu->base.unit].h; - ldu->pref_active = true; - } else { - ldu->pref_width = 800; - ldu->pref_height = 600; - ldu->pref_active = false; - } - con->status = vmw_ldu_connector_detect(con, true); - } - - mutex_unlock(&dev->mode_config.mutex); - - return 0; -} diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_overlay.c b/drivers/gpu/drm/vmwgfx/vmwgfx_overlay.c index f1a52f9..14399ee 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_overlay.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_overlay.c @@ -87,48 +87,6 @@ static inline void fill_flush(struct vmw_escape_video_flush *cmd, } /** - * Pin or unpin a buffer in vram. - * - * @dev_priv: Driver private. - * @buf: DMA buffer to pin or unpin. - * @pin: Pin buffer in vram if true. - * @interruptible: Use interruptible wait. - * - * Takes the current masters ttm lock in read. - * - * Returns - * -ERESTARTSYS if interrupted by a signal. - */ -static int vmw_dmabuf_pin_in_vram(struct vmw_private *dev_priv, - struct vmw_dma_buffer *buf, - bool pin, bool interruptible) -{ - struct ttm_buffer_object *bo = &buf->base; - struct ttm_placement *overlay_placement = &vmw_vram_placement; - int ret; - - ret = ttm_read_lock(&dev_priv->active_master->lock, interruptible); - if (unlikely(ret != 0)) - return ret; - - ret = ttm_bo_reserve(bo, interruptible, false, false, 0); - if (unlikely(ret != 0)) - goto err; - - if (pin) - overlay_placement = &vmw_vram_ne_placement; - - ret = ttm_bo_validate(bo, overlay_placement, interruptible, false, false); - - ttm_bo_unreserve(bo); - -err: - ttm_read_unlock(&dev_priv->active_master->lock); - - return ret; -} - -/** * Send put command to hw. * * Returns @@ -139,68 +97,80 @@ static int vmw_overlay_send_put(struct vmw_private *dev_priv, struct drm_vmw_control_stream_arg *arg, bool interruptible) { + struct vmw_escape_video_flush *flush; + size_t fifo_size; + bool have_so = dev_priv->sou_priv ? true : false; + int i, num_items; + SVGAGuestPtr ptr; + struct { struct vmw_escape_header escape; struct { - struct { - uint32_t cmdType; - uint32_t streamId; - } header; - struct { - uint32_t registerId; - uint32_t value; - } items[SVGA_VIDEO_PITCH_3 + 1]; - } body; - struct vmw_escape_video_flush flush; + uint32_t cmdType; + uint32_t streamId; + } header; } *cmds; - uint32_t offset; - int i, ret; + struct { + uint32_t registerId; + uint32_t value; + } *items; - for (;;) { - cmds = vmw_fifo_reserve(dev_priv, sizeof(*cmds)); - if (cmds) - break; + /* defines are a index needs + 1 */ + if (have_so) + num_items = SVGA_VIDEO_DST_SCREEN_ID + 1; + else + num_items = SVGA_VIDEO_PITCH_3 + 1; - ret = vmw_fallback_wait(dev_priv, false, true, 0, - interruptible, 3*HZ); - if (interruptible && ret == -ERESTARTSYS) - return ret; - else - BUG_ON(ret != 0); + fifo_size = sizeof(*cmds) + sizeof(*flush) + sizeof(*items) * num_items; + + cmds = vmw_fifo_reserve(dev_priv, fifo_size); + /* hardware has hung, can't do anything here */ + if (!cmds) + return -ENOMEM; + + items = (typeof(items))&cmds[1]; + flush = (struct vmw_escape_video_flush *)&items[num_items]; + + /* the size is header + number of items */ + fill_escape(&cmds->escape, sizeof(*items) * (num_items + 1)); + + cmds->header.cmdType = SVGA_ESCAPE_VMWARE_VIDEO_SET_REGS; + cmds->header.streamId = arg->stream_id; + + /* the IDs are neatly numbered */ + for (i = 0; i < num_items; i++) + items[i].registerId = i; + + vmw_bo_get_guest_ptr(&buf->base, &ptr); + ptr.offset += arg->offset; + + items[SVGA_VIDEO_ENABLED].value = true; + items[SVGA_VIDEO_FLAGS].value = arg->flags; + items[SVGA_VIDEO_DATA_OFFSET].value = ptr.offset; + items[SVGA_VIDEO_FORMAT].value = arg->format; + items[SVGA_VIDEO_COLORKEY].value = arg->color_key; + items[SVGA_VIDEO_SIZE].value = arg->size; + items[SVGA_VIDEO_WIDTH].value = arg->width; + items[SVGA_VIDEO_HEIGHT].value = arg->height; + items[SVGA_VIDEO_SRC_X].value = arg->src.x; + items[SVGA_VIDEO_SRC_Y].value = arg->src.y; + items[SVGA_VIDEO_SRC_WIDTH].value = arg->src.w; + items[SVGA_VIDEO_SRC_HEIGHT].value = arg->src.h; + items[SVGA_VIDEO_DST_X].value = arg->dst.x; + items[SVGA_VIDEO_DST_Y].value = arg->dst.y; + items[SVGA_VIDEO_DST_WIDTH].value = arg->dst.w; + items[SVGA_VIDEO_DST_HEIGHT].value = arg->dst.h; + items[SVGA_VIDEO_PITCH_1].value = arg->pitch[0]; + items[SVGA_VIDEO_PITCH_2].value = arg->pitch[1]; + items[SVGA_VIDEO_PITCH_3].value = arg->pitch[2]; + if (have_so) { + items[SVGA_VIDEO_DATA_GMRID].value = ptr.gmrId; + items[SVGA_VIDEO_DST_SCREEN_ID].value = SVGA_ID_INVALID; } - fill_escape(&cmds->escape, sizeof(cmds->body)); - cmds->body.header.cmdType = SVGA_ESCAPE_VMWARE_VIDEO_SET_REGS; - cmds->body.header.streamId = arg->stream_id; - - for (i = 0; i <= SVGA_VIDEO_PITCH_3; i++) - cmds->body.items[i].registerId = i; - - offset = buf->base.offset + arg->offset; - - cmds->body.items[SVGA_VIDEO_ENABLED].value = true; - cmds->body.items[SVGA_VIDEO_FLAGS].value = arg->flags; - cmds->body.items[SVGA_VIDEO_DATA_OFFSET].value = offset; - cmds->body.items[SVGA_VIDEO_FORMAT].value = arg->format; - cmds->body.items[SVGA_VIDEO_COLORKEY].value = arg->color_key; - cmds->body.items[SVGA_VIDEO_SIZE].value = arg->size; - cmds->body.items[SVGA_VIDEO_WIDTH].value = arg->width; - cmds->body.items[SVGA_VIDEO_HEIGHT].value = arg->height; - cmds->body.items[SVGA_VIDEO_SRC_X].value = arg->src.x; - cmds->body.items[SVGA_VIDEO_SRC_Y].value = arg->src.y; - cmds->body.items[SVGA_VIDEO_SRC_WIDTH].value = arg->src.w; - cmds->body.items[SVGA_VIDEO_SRC_HEIGHT].value = arg->src.h; - cmds->body.items[SVGA_VIDEO_DST_X].value = arg->dst.x; - cmds->body.items[SVGA_VIDEO_DST_Y].value = arg->dst.y; - cmds->body.items[SVGA_VIDEO_DST_WIDTH].value = arg->dst.w; - cmds->body.items[SVGA_VIDEO_DST_HEIGHT].value = arg->dst.h; - cmds->body.items[SVGA_VIDEO_PITCH_1].value = arg->pitch[0]; - cmds->body.items[SVGA_VIDEO_PITCH_2].value = arg->pitch[1]; - cmds->body.items[SVGA_VIDEO_PITCH_3].value = arg->pitch[2]; - - fill_flush(&cmds->flush, arg->stream_id); + fill_flush(flush, arg->stream_id); - vmw_fifo_commit(dev_priv, sizeof(*cmds)); + vmw_fifo_commit(dev_priv, fifo_size); return 0; } @@ -248,6 +218,25 @@ static int vmw_overlay_send_stop(struct vmw_private *dev_priv, } /** + * Move a buffer to vram or gmr if @pin is set, else unpin the buffer. + * + * With the introduction of screen objects buffers could now be + * used with GMRs instead of being locked to vram. + */ +static int vmw_overlay_move_buffer(struct vmw_private *dev_priv, + struct vmw_dma_buffer *buf, + bool pin, bool inter) +{ + if (!pin) + return vmw_dmabuf_unpin(dev_priv, buf, inter); + + if (!dev_priv->sou_priv) + return vmw_dmabuf_to_vram(dev_priv, buf, true, inter); + + return vmw_dmabuf_to_vram_or_gmr(dev_priv, buf, true, inter); +} + +/** * Stop or pause a stream. * * If the stream is paused the no evict flag is removed from the buffer @@ -279,8 +268,8 @@ static int vmw_overlay_stop(struct vmw_private *dev_priv, return ret; /* We just remove the NO_EVICT flag so no -ENOMEM */ - ret = vmw_dmabuf_pin_in_vram(dev_priv, stream->buf, false, - interruptible); + ret = vmw_overlay_move_buffer(dev_priv, stream->buf, false, + interruptible); if (interruptible && ret == -ERESTARTSYS) return ret; else @@ -342,7 +331,7 @@ static int vmw_overlay_update_stream(struct vmw_private *dev_priv, /* We don't start the old stream if we are interrupted. * Might return -ENOMEM if it can't fit the buffer in vram. */ - ret = vmw_dmabuf_pin_in_vram(dev_priv, buf, true, interruptible); + ret = vmw_overlay_move_buffer(dev_priv, buf, true, interruptible); if (ret) return ret; @@ -351,7 +340,8 @@ static int vmw_overlay_update_stream(struct vmw_private *dev_priv, /* This one needs to happen no matter what. We only remove * the NO_EVICT flag so this is safe from -ENOMEM. */ - BUG_ON(vmw_dmabuf_pin_in_vram(dev_priv, buf, false, false) != 0); + BUG_ON(vmw_overlay_move_buffer(dev_priv, buf, false, false) + != 0); return ret; } @@ -585,11 +575,10 @@ int vmw_overlay_init(struct vmw_private *dev_priv) return -ENOSYS; } - overlay = kmalloc(sizeof(*overlay), GFP_KERNEL); + overlay = kzalloc(sizeof(*overlay), GFP_KERNEL); if (!overlay) return -ENOMEM; - memset(overlay, 0, sizeof(*overlay)); mutex_init(&overlay->mutex); for (i = 0; i < VMW_MAX_NUM_STREAMS; i++) { overlay->stream[i].buf = NULL; diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c index 5408b1b..0795d17 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c @@ -31,10 +31,6 @@ #include "ttm/ttm_placement.h" #include "drmP.h" -#define VMW_RES_CONTEXT ttm_driver_type0 -#define VMW_RES_SURFACE ttm_driver_type1 -#define VMW_RES_STREAM ttm_driver_type2 - struct vmw_user_context { struct ttm_base_object base; struct vmw_resource res; @@ -43,6 +39,7 @@ struct vmw_user_context { struct vmw_user_surface { struct ttm_base_object base; struct vmw_surface srf; + uint32_t size; }; struct vmw_user_dma_buffer { @@ -65,6 +62,17 @@ struct vmw_user_stream { struct vmw_stream stream; }; +struct vmw_surface_offset { + uint32_t face; + uint32_t mip; + uint32_t bo_offset; +}; + + +static uint64_t vmw_user_context_size; +static uint64_t vmw_user_surface_size; +static uint64_t vmw_user_stream_size; + static inline struct vmw_dma_buffer * vmw_dma_buffer(struct ttm_buffer_object *bo) { @@ -84,13 +92,36 @@ struct vmw_resource *vmw_resource_reference(struct vmw_resource *res) return res; } + +/** + * vmw_resource_release_id - release a resource id to the id manager. + * + * @res: Pointer to the resource. + * + * Release the resource id to the resource id manager and set it to -1 + */ +static void vmw_resource_release_id(struct vmw_resource *res) +{ + struct vmw_private *dev_priv = res->dev_priv; + + write_lock(&dev_priv->resource_lock); + if (res->id != -1) + idr_remove(res->idr, res->id); + res->id = -1; + write_unlock(&dev_priv->resource_lock); +} + static void vmw_resource_release(struct kref *kref) { struct vmw_resource *res = container_of(kref, struct vmw_resource, kref); struct vmw_private *dev_priv = res->dev_priv; + int id = res->id; + struct idr *idr = res->idr; - idr_remove(res->idr, res->id); + res->avail = false; + if (res->remove_from_lists != NULL) + res->remove_from_lists(res); write_unlock(&dev_priv->resource_lock); if (likely(res->hw_destroy != NULL)) @@ -102,6 +133,9 @@ static void vmw_resource_release(struct kref *kref) kfree(res); write_lock(&dev_priv->resource_lock); + + if (id != -1) + idr_remove(idr, id); } void vmw_resource_unreference(struct vmw_resource **p_res) @@ -115,28 +149,29 @@ void vmw_resource_unreference(struct vmw_resource **p_res) write_unlock(&dev_priv->resource_lock); } -static int vmw_resource_init(struct vmw_private *dev_priv, - struct vmw_resource *res, - struct idr *idr, - enum ttm_object_type obj_type, - void (*res_free) (struct vmw_resource *res)) + +/** + * vmw_resource_alloc_id - release a resource id to the id manager. + * + * @dev_priv: Pointer to the device private structure. + * @res: Pointer to the resource. + * + * Allocate the lowest free resource from the resource manager, and set + * @res->id to that id. Returns 0 on success and -ENOMEM on failure. + */ +static int vmw_resource_alloc_id(struct vmw_private *dev_priv, + struct vmw_resource *res) { int ret; - kref_init(&res->kref); - res->hw_destroy = NULL; - res->res_free = res_free; - res->res_type = obj_type; - res->idr = idr; - res->avail = false; - res->dev_priv = dev_priv; + BUG_ON(res->id != -1); do { - if (unlikely(idr_pre_get(idr, GFP_KERNEL) == 0)) + if (unlikely(idr_pre_get(res->idr, GFP_KERNEL) == 0)) return -ENOMEM; write_lock(&dev_priv->resource_lock); - ret = idr_get_new_above(idr, res, 1, &res->id); + ret = idr_get_new_above(res->idr, res, 1, &res->id); write_unlock(&dev_priv->resource_lock); } while (ret == -EAGAIN); @@ -144,6 +179,33 @@ static int vmw_resource_init(struct vmw_private *dev_priv, return ret; } + +static int vmw_resource_init(struct vmw_private *dev_priv, + struct vmw_resource *res, + struct idr *idr, + enum ttm_object_type obj_type, + bool delay_id, + void (*res_free) (struct vmw_resource *res), + void (*remove_from_lists) + (struct vmw_resource *res)) +{ + kref_init(&res->kref); + res->hw_destroy = NULL; + res->res_free = res_free; + res->remove_from_lists = remove_from_lists; + res->res_type = obj_type; + res->idr = idr; + res->avail = false; + res->dev_priv = dev_priv; + INIT_LIST_HEAD(&res->query_head); + INIT_LIST_HEAD(&res->validate_head); + res->id = -1; + if (delay_id) + return 0; + else + return vmw_resource_alloc_id(dev_priv, res); +} + /** * vmw_resource_activate * @@ -198,8 +260,12 @@ static void vmw_hw_context_destroy(struct vmw_resource *res) struct { SVGA3dCmdHeader header; SVGA3dCmdDestroyContext body; - } *cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd)); + } *cmd; + + vmw_execbuf_release_pinned_bo(dev_priv, true, res->id); + + cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd)); if (unlikely(cmd == NULL)) { DRM_ERROR("Failed reserving FIFO space for surface " "destruction.\n"); @@ -211,7 +277,7 @@ static void vmw_hw_context_destroy(struct vmw_resource *res) cmd->body.cid = cpu_to_le32(res->id); vmw_fifo_commit(dev_priv, sizeof(*cmd)); - vmw_3d_resource_dec(dev_priv); + vmw_3d_resource_dec(dev_priv, false); } static int vmw_context_init(struct vmw_private *dev_priv, @@ -226,14 +292,17 @@ static int vmw_context_init(struct vmw_private *dev_priv, } *cmd; ret = vmw_resource_init(dev_priv, res, &dev_priv->context_idr, - VMW_RES_CONTEXT, res_free); + VMW_RES_CONTEXT, false, res_free, NULL); if (unlikely(ret != 0)) { - if (res_free == NULL) - kfree(res); - else - res_free(res); - return ret; + DRM_ERROR("Failed to allocate a resource id.\n"); + goto out_early; + } + + if (unlikely(res->id >= SVGA3D_MAX_CONTEXT_IDS)) { + DRM_ERROR("Out of hw context ids.\n"); + vmw_resource_unreference(&res); + return -ENOMEM; } cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd)); @@ -248,9 +317,16 @@ static int vmw_context_init(struct vmw_private *dev_priv, cmd->body.cid = cpu_to_le32(res->id); vmw_fifo_commit(dev_priv, sizeof(*cmd)); - (void) vmw_3d_resource_inc(dev_priv); + (void) vmw_3d_resource_inc(dev_priv, false); vmw_resource_activate(res, vmw_hw_context_destroy); return 0; + +out_early: + if (res_free == NULL) + kfree(res); + else + res_free(res); + return ret; } struct vmw_resource *vmw_context_alloc(struct vmw_private *dev_priv) @@ -273,8 +349,11 @@ static void vmw_user_context_free(struct vmw_resource *res) { struct vmw_user_context *ctx = container_of(res, struct vmw_user_context, res); + struct vmw_private *dev_priv = res->dev_priv; kfree(ctx); + ttm_mem_global_free(vmw_mem_glob(dev_priv), + vmw_user_context_size); } /** @@ -328,23 +407,56 @@ int vmw_context_define_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv) { struct vmw_private *dev_priv = vmw_priv(dev); - struct vmw_user_context *ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); + struct vmw_user_context *ctx; struct vmw_resource *res; 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; - if (unlikely(ctx == NULL)) - return -ENOMEM; + + /* + * Approximate idr memory usage with 128 bytes. It will be limited + * by maximum number_of contexts anyway. + */ + + if (unlikely(vmw_user_context_size == 0)) + vmw_user_context_size = ttm_round_pot(sizeof(*ctx)) + 128; + + ret = ttm_read_lock(&vmaster->lock, true); + if (unlikely(ret != 0)) + return ret; + + ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), + vmw_user_context_size, + false, true); + if (unlikely(ret != 0)) { + if (ret != -ERESTARTSYS) + DRM_ERROR("Out of graphics memory for context" + " creation.\n"); + goto out_unlock; + } + + ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); + if (unlikely(ctx == NULL)) { + ttm_mem_global_free(vmw_mem_glob(dev_priv), + vmw_user_context_size); + ret = -ENOMEM; + goto out_unlock; + } res = &ctx->res; ctx->base.shareable = false; ctx->base.tfile = NULL; + /* + * From here on, the destructor takes over resource freeing. + */ + ret = vmw_context_init(dev_priv, res, vmw_user_context_free); if (unlikely(ret != 0)) - return ret; + goto out_unlock; tmp = vmw_resource_reference(&ctx->res); ret = ttm_base_object_init(tfile, &ctx->base, false, VMW_RES_CONTEXT, @@ -358,13 +470,16 @@ int vmw_context_define_ioctl(struct drm_device *dev, void *data, arg->cid = res->id; out_err: vmw_resource_unreference(&res); +out_unlock: + ttm_read_unlock(&vmaster->lock); return ret; } int vmw_context_check(struct vmw_private *dev_priv, struct ttm_object_file *tfile, - int id) + int id, + struct vmw_resource **p_res) { struct vmw_resource *res; int ret = 0; @@ -376,6 +491,8 @@ int vmw_context_check(struct vmw_private *dev_priv, container_of(res, struct vmw_user_context, res); if (ctx->base.tfile != tfile && !ctx->base.shareable) ret = -EPERM; + if (p_res) + *p_res = vmw_resource_reference(res); } else ret = -EINVAL; read_unlock(&dev_priv->resource_lock); @@ -383,102 +500,638 @@ int vmw_context_check(struct vmw_private *dev_priv, return ret; } +struct vmw_bpp { + uint8_t bpp; + uint8_t s_bpp; +}; + +/* + * Size table for the supported SVGA3D surface formats. It consists of + * two values. The bpp value and the s_bpp value which is short for + * "stride bits per pixel" The values are given in such a way that the + * minimum stride for the image is calculated using + * + * min_stride = w*s_bpp + * + * and the total memory requirement for the image is + * + * h*min_stride*bpp/s_bpp + * + */ +static const struct vmw_bpp vmw_sf_bpp[] = { + [SVGA3D_FORMAT_INVALID] = {0, 0}, + [SVGA3D_X8R8G8B8] = {32, 32}, + [SVGA3D_A8R8G8B8] = {32, 32}, + [SVGA3D_R5G6B5] = {16, 16}, + [SVGA3D_X1R5G5B5] = {16, 16}, + [SVGA3D_A1R5G5B5] = {16, 16}, + [SVGA3D_A4R4G4B4] = {16, 16}, + [SVGA3D_Z_D32] = {32, 32}, + [SVGA3D_Z_D16] = {16, 16}, + [SVGA3D_Z_D24S8] = {32, 32}, + [SVGA3D_Z_D15S1] = {16, 16}, + [SVGA3D_LUMINANCE8] = {8, 8}, + [SVGA3D_LUMINANCE4_ALPHA4] = {8, 8}, + [SVGA3D_LUMINANCE16] = {16, 16}, + [SVGA3D_LUMINANCE8_ALPHA8] = {16, 16}, + [SVGA3D_DXT1] = {4, 16}, + [SVGA3D_DXT2] = {8, 32}, + [SVGA3D_DXT3] = {8, 32}, + [SVGA3D_DXT4] = {8, 32}, + [SVGA3D_DXT5] = {8, 32}, + [SVGA3D_BUMPU8V8] = {16, 16}, + [SVGA3D_BUMPL6V5U5] = {16, 16}, + [SVGA3D_BUMPX8L8V8U8] = {32, 32}, + [SVGA3D_ARGB_S10E5] = {16, 16}, + [SVGA3D_ARGB_S23E8] = {32, 32}, + [SVGA3D_A2R10G10B10] = {32, 32}, + [SVGA3D_V8U8] = {16, 16}, + [SVGA3D_Q8W8V8U8] = {32, 32}, + [SVGA3D_CxV8U8] = {16, 16}, + [SVGA3D_X8L8V8U8] = {32, 32}, + [SVGA3D_A2W10V10U10] = {32, 32}, + [SVGA3D_ALPHA8] = {8, 8}, + [SVGA3D_R_S10E5] = {16, 16}, + [SVGA3D_R_S23E8] = {32, 32}, + [SVGA3D_RG_S10E5] = {16, 16}, + [SVGA3D_RG_S23E8] = {32, 32}, + [SVGA3D_BUFFER] = {8, 8}, + [SVGA3D_Z_D24X8] = {32, 32}, + [SVGA3D_V16U16] = {32, 32}, + [SVGA3D_G16R16] = {32, 32}, + [SVGA3D_A16B16G16R16] = {64, 64}, + [SVGA3D_UYVY] = {12, 12}, + [SVGA3D_YUY2] = {12, 12}, + [SVGA3D_NV12] = {12, 8}, + [SVGA3D_AYUV] = {32, 32}, + [SVGA3D_BC4_UNORM] = {4, 16}, + [SVGA3D_BC5_UNORM] = {8, 32}, + [SVGA3D_Z_DF16] = {16, 16}, + [SVGA3D_Z_DF24] = {24, 24}, + [SVGA3D_Z_D24S8_INT] = {32, 32} +}; + /** * Surface management. */ +struct vmw_surface_dma { + SVGA3dCmdHeader header; + SVGA3dCmdSurfaceDMA body; + SVGA3dCopyBox cb; + SVGA3dCmdSurfaceDMASuffix suffix; +}; + +struct vmw_surface_define { + SVGA3dCmdHeader header; + SVGA3dCmdDefineSurface body; +}; + +struct vmw_surface_destroy { + SVGA3dCmdHeader header; + SVGA3dCmdDestroySurface body; +}; + + +/** + * vmw_surface_dma_size - Compute fifo size for a dma command. + * + * @srf: Pointer to a struct vmw_surface + * + * Computes the required size for a surface dma command for backup or + * restoration of the surface represented by @srf. + */ +static inline uint32_t vmw_surface_dma_size(const struct vmw_surface *srf) +{ + return srf->num_sizes * sizeof(struct vmw_surface_dma); +} + + +/** + * vmw_surface_define_size - Compute fifo size for a surface define command. + * + * @srf: Pointer to a struct vmw_surface + * + * Computes the required size for a surface define command for the definition + * of the surface represented by @srf. + */ +static inline uint32_t vmw_surface_define_size(const struct vmw_surface *srf) +{ + return sizeof(struct vmw_surface_define) + srf->num_sizes * + sizeof(SVGA3dSize); +} + + +/** + * vmw_surface_destroy_size - Compute fifo size for a surface destroy command. + * + * Computes the required size for a surface destroy command for the destruction + * of a hw surface. + */ +static inline uint32_t vmw_surface_destroy_size(void) +{ + return sizeof(struct vmw_surface_destroy); +} + +/** + * vmw_surface_destroy_encode - Encode a surface_destroy command. + * + * @id: The surface id + * @cmd_space: Pointer to memory area in which the commands should be encoded. + */ +static void vmw_surface_destroy_encode(uint32_t id, + void *cmd_space) +{ + struct vmw_surface_destroy *cmd = (struct vmw_surface_destroy *) + cmd_space; + + cmd->header.id = SVGA_3D_CMD_SURFACE_DESTROY; + cmd->header.size = sizeof(cmd->body); + cmd->body.sid = id; +} + +/** + * 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 void vmw_surface_define_encode(const struct vmw_surface *srf, + void *cmd_space) +{ + struct vmw_surface_define *cmd = (struct vmw_surface_define *) + cmd_space; + struct drm_vmw_size *src_size; + SVGA3dSize *cmd_size; + uint32_t cmd_len; + int i; + + cmd_len = sizeof(cmd->body) + srf->num_sizes * sizeof(SVGA3dSize); + + cmd->header.id = SVGA_3D_CMD_SURFACE_DEFINE; + cmd->header.size = cmd_len; + cmd->body.sid = srf->res.id; + cmd->body.surfaceFlags = srf->flags; + cmd->body.format = cpu_to_le32(srf->format); + for (i = 0; i < DRM_VMW_MAX_SURFACE_FACES; ++i) + cmd->body.face[i].numMipLevels = srf->mip_levels[i]; + + cmd += 1; + cmd_size = (SVGA3dSize *) cmd; + src_size = srf->sizes; + + for (i = 0; i < srf->num_sizes; ++i, cmd_size++, src_size++) { + cmd_size->width = src_size->width; + cmd_size->height = src_size->height; + cmd_size->depth = src_size->depth; + } +} + + +/** + * vmw_surface_dma_encode - Encode a surface_dma command. + * + * @srf: Pointer to a struct vmw_surface object. + * @cmd_space: Pointer to memory area in which the commands should be encoded. + * @ptr: Pointer to an SVGAGuestPtr indicating where the surface contents + * should be placed or read from. + * @to_surface: Boolean whether to DMA to the surface or from the surface. + */ +static void vmw_surface_dma_encode(struct vmw_surface *srf, + void *cmd_space, + const SVGAGuestPtr *ptr, + bool to_surface) +{ + uint32_t i; + uint32_t bpp = vmw_sf_bpp[srf->format].bpp; + uint32_t stride_bpp = vmw_sf_bpp[srf->format].s_bpp; + struct vmw_surface_dma *cmd = (struct vmw_surface_dma *)cmd_space; + + for (i = 0; i < srf->num_sizes; ++i) { + SVGA3dCmdHeader *header = &cmd->header; + SVGA3dCmdSurfaceDMA *body = &cmd->body; + SVGA3dCopyBox *cb = &cmd->cb; + SVGA3dCmdSurfaceDMASuffix *suffix = &cmd->suffix; + const struct vmw_surface_offset *cur_offset = &srf->offsets[i]; + const struct drm_vmw_size *cur_size = &srf->sizes[i]; + + header->id = SVGA_3D_CMD_SURFACE_DMA; + header->size = sizeof(*body) + sizeof(*cb) + sizeof(*suffix); + + body->guest.ptr = *ptr; + body->guest.ptr.offset += cur_offset->bo_offset; + body->guest.pitch = (cur_size->width * stride_bpp + 7) >> 3; + body->host.sid = srf->res.id; + body->host.face = cur_offset->face; + body->host.mipmap = cur_offset->mip; + body->transfer = ((to_surface) ? SVGA3D_WRITE_HOST_VRAM : + SVGA3D_READ_HOST_VRAM); + cb->x = 0; + cb->y = 0; + cb->z = 0; + cb->srcx = 0; + cb->srcy = 0; + cb->srcz = 0; + cb->w = cur_size->width; + cb->h = cur_size->height; + cb->d = cur_size->depth; + + suffix->suffixSize = sizeof(*suffix); + suffix->maximumOffset = body->guest.pitch*cur_size->height* + cur_size->depth*bpp / stride_bpp; + suffix->flags.discard = 0; + suffix->flags.unsynchronized = 0; + suffix->flags.reserved = 0; + ++cmd; + } +}; + + static void vmw_hw_surface_destroy(struct vmw_resource *res) { struct vmw_private *dev_priv = res->dev_priv; - struct { - SVGA3dCmdHeader header; - SVGA3dCmdDestroySurface body; - } *cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd)); + struct vmw_surface *srf; + void *cmd; - if (unlikely(cmd == NULL)) { - DRM_ERROR("Failed reserving FIFO space for surface " - "destruction.\n"); - return; - } + if (res->id != -1) { - cmd->header.id = cpu_to_le32(SVGA_3D_CMD_SURFACE_DESTROY); - cmd->header.size = cpu_to_le32(sizeof(cmd->body)); - cmd->body.sid = cpu_to_le32(res->id); + cmd = vmw_fifo_reserve(dev_priv, vmw_surface_destroy_size()); + if (unlikely(cmd == NULL)) { + DRM_ERROR("Failed reserving FIFO space for surface " + "destruction.\n"); + return; + } - vmw_fifo_commit(dev_priv, sizeof(*cmd)); - vmw_3d_resource_dec(dev_priv); + vmw_surface_destroy_encode(res->id, cmd); + vmw_fifo_commit(dev_priv, vmw_surface_destroy_size()); + + /* + * used_memory_size_atomic, or separate lock + * to avoid taking dev_priv::cmdbuf_mutex in + * the destroy path. + */ + + mutex_lock(&dev_priv->cmdbuf_mutex); + srf = container_of(res, struct vmw_surface, res); + dev_priv->used_memory_size -= srf->backup_size; + mutex_unlock(&dev_priv->cmdbuf_mutex); + + } + vmw_3d_resource_dec(dev_priv, false); } void vmw_surface_res_free(struct vmw_resource *res) { struct vmw_surface *srf = container_of(res, struct vmw_surface, res); + if (srf->backup) + ttm_bo_unref(&srf->backup); + kfree(srf->offsets); kfree(srf->sizes); kfree(srf->snooper.image); kfree(srf); } -int vmw_surface_init(struct vmw_private *dev_priv, - struct vmw_surface *srf, - void (*res_free) (struct vmw_resource *res)) + +/** + * vmw_surface_do_validate - make a surface available to the device. + * + * @dev_priv: Pointer to a device private struct. + * @srf: Pointer to a struct vmw_surface. + * + * If the surface doesn't have a hw id, allocate one, and optionally + * DMA the backed up surface contents to the device. + * + * Returns -EBUSY if there wasn't sufficient device resources to + * complete the validation. Retry after freeing up resources. + * + * May return other errors if the kernel is out of guest resources. + */ +int vmw_surface_do_validate(struct vmw_private *dev_priv, + struct vmw_surface *srf) { - int ret; - struct { - SVGA3dCmdHeader header; - SVGA3dCmdDefineSurface body; - } *cmd; - SVGA3dSize *cmd_size; struct vmw_resource *res = &srf->res; - struct drm_vmw_size *src_size; - size_t submit_size; - uint32_t cmd_len; - int i; + struct list_head val_list; + struct ttm_validate_buffer val_buf; + uint32_t submit_size; + uint8_t *cmd; + int ret; - BUG_ON(res_free == NULL); - ret = vmw_resource_init(dev_priv, res, &dev_priv->surface_idr, - VMW_RES_SURFACE, res_free); + if (likely(res->id != -1)) + return 0; + if (unlikely(dev_priv->used_memory_size + srf->backup_size >= + dev_priv->memory_size)) + return -EBUSY; + + /* + * Reserve- and validate the backup DMA bo. + */ + + if (srf->backup) { + INIT_LIST_HEAD(&val_list); + val_buf.bo = ttm_bo_reference(srf->backup); + val_buf.new_sync_obj_arg = (void *)((unsigned long) + DRM_VMW_FENCE_FLAG_EXEC); + list_add_tail(&val_buf.head, &val_list); + ret = ttm_eu_reserve_buffers(&val_list); + if (unlikely(ret != 0)) + goto out_no_reserve; + + ret = ttm_bo_validate(srf->backup, &vmw_srf_placement, + true, false, false); + if (unlikely(ret != 0)) + goto out_no_validate; + } + + /* + * Alloc id for the resource. + */ + + ret = vmw_resource_alloc_id(dev_priv, res); if (unlikely(ret != 0)) { - res_free(res); - return ret; + DRM_ERROR("Failed to allocate a surface id.\n"); + goto out_no_id; + } + if (unlikely(res->id >= SVGA3D_MAX_SURFACE_IDS)) { + ret = -EBUSY; + goto out_no_fifo; } - submit_size = sizeof(*cmd) + srf->num_sizes * sizeof(SVGA3dSize); - cmd_len = sizeof(cmd->body) + srf->num_sizes * sizeof(SVGA3dSize); + + /* + * Encode surface define- and dma commands. + */ + + submit_size = vmw_surface_define_size(srf); + if (srf->backup) + submit_size += vmw_surface_dma_size(srf); cmd = vmw_fifo_reserve(dev_priv, submit_size); if (unlikely(cmd == NULL)) { - DRM_ERROR("Fifo reserve failed for create surface.\n"); - vmw_resource_unreference(&res); - return -ENOMEM; + DRM_ERROR("Failed reserving FIFO space for surface " + "validation.\n"); + ret = -ENOMEM; + goto out_no_fifo; } - cmd->header.id = cpu_to_le32(SVGA_3D_CMD_SURFACE_DEFINE); - cmd->header.size = cpu_to_le32(cmd_len); - cmd->body.sid = cpu_to_le32(res->id); - cmd->body.surfaceFlags = cpu_to_le32(srf->flags); - cmd->body.format = cpu_to_le32(srf->format); - for (i = 0; i < DRM_VMW_MAX_SURFACE_FACES; ++i) { - cmd->body.face[i].numMipLevels = - cpu_to_le32(srf->mip_levels[i]); + vmw_surface_define_encode(srf, cmd); + if (srf->backup) { + SVGAGuestPtr ptr; + + cmd += vmw_surface_define_size(srf); + vmw_bo_get_guest_ptr(srf->backup, &ptr); + vmw_surface_dma_encode(srf, cmd, &ptr, true); } - cmd += 1; - cmd_size = (SVGA3dSize *) cmd; - src_size = srf->sizes; + vmw_fifo_commit(dev_priv, submit_size); - for (i = 0; i < srf->num_sizes; ++i, cmd_size++, src_size++) { - cmd_size->width = cpu_to_le32(src_size->width); - cmd_size->height = cpu_to_le32(src_size->height); - cmd_size->depth = cpu_to_le32(src_size->depth); + /* + * Create a fence object and fence the backup buffer. + */ + + if (srf->backup) { + struct vmw_fence_obj *fence; + + (void) vmw_execbuf_fence_commands(NULL, dev_priv, + &fence, NULL); + ttm_eu_fence_buffer_objects(&val_list, fence); + if (likely(fence != NULL)) + vmw_fence_obj_unreference(&fence); + ttm_bo_unref(&val_buf.bo); + ttm_bo_unref(&srf->backup); + } + + /* + * Surface memory usage accounting. + */ + + dev_priv->used_memory_size += srf->backup_size; + + return 0; + +out_no_fifo: + vmw_resource_release_id(res); +out_no_id: +out_no_validate: + if (srf->backup) + ttm_eu_backoff_reservation(&val_list); +out_no_reserve: + if (srf->backup) + ttm_bo_unref(&val_buf.bo); + return ret; +} + +/** + * vmw_surface_evict - Evict a hw surface. + * + * @dev_priv: Pointer to a device private struct. + * @srf: Pointer to a struct vmw_surface + * + * DMA the contents of a hw surface to a backup guest buffer object, + * and destroy the hw surface, releasing its id. + */ +int vmw_surface_evict(struct vmw_private *dev_priv, + struct vmw_surface *srf) +{ + struct vmw_resource *res = &srf->res; + struct list_head val_list; + struct ttm_validate_buffer val_buf; + uint32_t submit_size; + uint8_t *cmd; + int ret; + struct vmw_fence_obj *fence; + SVGAGuestPtr ptr; + + BUG_ON(res->id == -1); + + /* + * Create a surface backup buffer object. + */ + + if (!srf->backup) { + ret = ttm_bo_create(&dev_priv->bdev, srf->backup_size, + ttm_bo_type_device, + &vmw_srf_placement, 0, 0, true, + NULL, &srf->backup); + if (unlikely(ret != 0)) + return ret; + } + + /* + * Reserve- and validate the backup DMA bo. + */ + + INIT_LIST_HEAD(&val_list); + val_buf.bo = ttm_bo_reference(srf->backup); + val_buf.new_sync_obj_arg = (void *)(unsigned long) + DRM_VMW_FENCE_FLAG_EXEC; + list_add_tail(&val_buf.head, &val_list); + ret = ttm_eu_reserve_buffers(&val_list); + if (unlikely(ret != 0)) + goto out_no_reserve; + + ret = ttm_bo_validate(srf->backup, &vmw_srf_placement, + true, false, false); + if (unlikely(ret != 0)) + goto out_no_validate; + + + /* + * Encode the dma- and surface destroy commands. + */ + + submit_size = vmw_surface_dma_size(srf) + vmw_surface_destroy_size(); + cmd = vmw_fifo_reserve(dev_priv, submit_size); + if (unlikely(cmd == NULL)) { + DRM_ERROR("Failed reserving FIFO space for surface " + "eviction.\n"); + ret = -ENOMEM; + goto out_no_fifo; } + vmw_bo_get_guest_ptr(srf->backup, &ptr); + vmw_surface_dma_encode(srf, cmd, &ptr, false); + cmd += vmw_surface_dma_size(srf); + vmw_surface_destroy_encode(res->id, cmd); vmw_fifo_commit(dev_priv, submit_size); - (void) vmw_3d_resource_inc(dev_priv); - vmw_resource_activate(res, vmw_hw_surface_destroy); + + /* + * Surface memory usage accounting. + */ + + dev_priv->used_memory_size -= srf->backup_size; + + /* + * Create a fence object and fence the DMA buffer. + */ + + (void) vmw_execbuf_fence_commands(NULL, dev_priv, + &fence, NULL); + ttm_eu_fence_buffer_objects(&val_list, fence); + if (likely(fence != NULL)) + vmw_fence_obj_unreference(&fence); + ttm_bo_unref(&val_buf.bo); + + /* + * Release the surface ID. + */ + + vmw_resource_release_id(res); + return 0; + +out_no_fifo: +out_no_validate: + if (srf->backup) + ttm_eu_backoff_reservation(&val_list); +out_no_reserve: + ttm_bo_unref(&val_buf.bo); + ttm_bo_unref(&srf->backup); + return ret; +} + + +/** + * vmw_surface_validate - make a surface available to the device, evicting + * other surfaces if needed. + * + * @dev_priv: Pointer to a device private struct. + * @srf: Pointer to a struct vmw_surface. + * + * Try to validate a surface and if it fails due to limited device resources, + * repeatedly try to evict other surfaces until the request can be + * acommodated. + * + * May return errors if out of resources. + */ +int vmw_surface_validate(struct vmw_private *dev_priv, + struct vmw_surface *srf) +{ + int ret; + struct vmw_surface *evict_srf; + + do { + write_lock(&dev_priv->resource_lock); + list_del_init(&srf->lru_head); + write_unlock(&dev_priv->resource_lock); + + ret = vmw_surface_do_validate(dev_priv, srf); + if (likely(ret != -EBUSY)) + break; + + write_lock(&dev_priv->resource_lock); + if (list_empty(&dev_priv->surface_lru)) { + DRM_ERROR("Out of device memory for surfaces.\n"); + ret = -EBUSY; + write_unlock(&dev_priv->resource_lock); + break; + } + + evict_srf = vmw_surface_reference + (list_first_entry(&dev_priv->surface_lru, + struct vmw_surface, + lru_head)); + list_del_init(&evict_srf->lru_head); + + write_unlock(&dev_priv->resource_lock); + (void) vmw_surface_evict(dev_priv, evict_srf); + + vmw_surface_unreference(&evict_srf); + + } while (1); + + if (unlikely(ret != 0 && srf->res.id != -1)) { + write_lock(&dev_priv->resource_lock); + list_add_tail(&srf->lru_head, &dev_priv->surface_lru); + write_unlock(&dev_priv->resource_lock); + } + + return ret; +} + + +/** + * vmw_surface_remove_from_lists - Remove surface resources from lookup lists + * + * @res: Pointer to a struct vmw_resource embedded in a struct vmw_surface + * + * As part of the resource destruction, remove the surface from any + * lookup lists. + */ +static void vmw_surface_remove_from_lists(struct vmw_resource *res) +{ + struct vmw_surface *srf = container_of(res, struct vmw_surface, res); + + list_del_init(&srf->lru_head); +} + +int vmw_surface_init(struct vmw_private *dev_priv, + struct vmw_surface *srf, + void (*res_free) (struct vmw_resource *res)) +{ + int ret; + struct vmw_resource *res = &srf->res; + + BUG_ON(res_free == NULL); + INIT_LIST_HEAD(&srf->lru_head); + ret = vmw_resource_init(dev_priv, res, &dev_priv->surface_idr, + VMW_RES_SURFACE, true, res_free, + vmw_surface_remove_from_lists); + + if (unlikely(ret != 0)) + res_free(res); + + /* + * The surface won't be visible to hardware until a + * surface validate. + */ + + (void) vmw_3d_resource_inc(dev_priv, false); + vmw_resource_activate(res, vmw_hw_surface_destroy); + return ret; } static void vmw_user_surface_free(struct vmw_resource *res) @@ -486,12 +1139,81 @@ static void vmw_user_surface_free(struct vmw_resource *res) struct vmw_surface *srf = container_of(res, struct vmw_surface, res); struct vmw_user_surface *user_srf = container_of(srf, struct vmw_user_surface, srf); + struct vmw_private *dev_priv = srf->res.dev_priv; + uint32_t size = user_srf->size; + if (srf->backup) + ttm_bo_unref(&srf->backup); + kfree(srf->offsets); kfree(srf->sizes); kfree(srf->snooper.image); kfree(user_srf); + ttm_mem_global_free(vmw_mem_glob(dev_priv), size); +} + +/** + * vmw_resource_unreserve - unreserve resources previously reserved for + * command submission. + * + * @list_head: list of resources to unreserve. + * + * Currently only surfaces are considered, and unreserving a surface + * means putting it back on the device's surface lru list, + * so that it can be evicted if necessary. + * This function traverses the resource list and + * checks whether resources are surfaces, and in that case puts them back + * on the device's surface LRU list. + */ +void vmw_resource_unreserve(struct list_head *list) +{ + struct vmw_resource *res; + struct vmw_surface *srf; + rwlock_t *lock = NULL; + + list_for_each_entry(res, list, validate_head) { + + if (res->res_free != &vmw_surface_res_free && + res->res_free != &vmw_user_surface_free) + continue; + + if (unlikely(lock == NULL)) { + lock = &res->dev_priv->resource_lock; + write_lock(lock); + } + + srf = container_of(res, struct vmw_surface, res); + list_del_init(&srf->lru_head); + list_add_tail(&srf->lru_head, &res->dev_priv->surface_lru); + } + + if (lock != NULL) + write_unlock(lock); } +/** + * Helper function that looks either a surface or dmabuf. + * + * The pointer this pointed at by out_surf and out_buf needs to be null. + */ +int vmw_user_lookup_handle(struct vmw_private *dev_priv, + struct ttm_object_file *tfile, + uint32_t handle, + struct vmw_surface **out_surf, + struct vmw_dma_buffer **out_buf) +{ + int ret; + + BUG_ON(*out_surf || *out_buf); + + ret = vmw_user_surface_lookup_handle(dev_priv, tfile, handle, out_surf); + if (!ret) + return 0; + + ret = vmw_user_dmabuf_lookup(tfile, handle, out_buf); + return ret; +} + + int vmw_user_surface_lookup_handle(struct vmw_private *dev_priv, struct ttm_object_file *tfile, uint32_t handle, struct vmw_surface **out) @@ -556,8 +1278,7 @@ int vmw_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 = - kmalloc(sizeof(*user_srf), GFP_KERNEL); + struct vmw_user_surface *user_srf; struct vmw_surface *srf; struct vmw_resource *res; struct vmw_resource *tmp; @@ -568,10 +1289,51 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data, struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile; struct drm_vmw_size __user *user_sizes; int ret; - int i; + int i, j; + uint32_t cur_bo_offset; + struct drm_vmw_size *cur_size; + struct vmw_surface_offset *cur_offset; + uint32_t stride_bpp; + uint32_t bpp; + uint32_t num_sizes; + uint32_t size; + struct vmw_master *vmaster = vmw_master(file_priv->master); - if (unlikely(user_srf == NULL)) - return -ENOMEM; + if (unlikely(vmw_user_surface_size == 0)) + vmw_user_surface_size = ttm_round_pot(sizeof(*user_srf)) + + 128; + + num_sizes = 0; + for (i = 0; i < DRM_VMW_MAX_SURFACE_FACES; ++i) + num_sizes += req->mip_levels[i]; + + if (num_sizes > DRM_VMW_MAX_SURFACE_FACES * + DRM_VMW_MAX_MIP_LEVELS) + return -EINVAL; + + size = vmw_user_surface_size + 128 + + ttm_round_pot(num_sizes * sizeof(struct drm_vmw_size)) + + ttm_round_pot(num_sizes * sizeof(struct vmw_surface_offset)); + + + ret = ttm_read_lock(&vmaster->lock, true); + if (unlikely(ret != 0)) + return ret; + + ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), + size, false, true); + if (unlikely(ret != 0)) { + if (ret != -ERESTARTSYS) + DRM_ERROR("Out of graphics memory for surface" + " creation.\n"); + goto out_unlock; + } + + user_srf = kmalloc(sizeof(*user_srf), GFP_KERNEL); + if (unlikely(user_srf == NULL)) { + ret = -ENOMEM; + goto out_no_user_srf; + } srf = &user_srf->srf; res = &srf->res; @@ -579,21 +1341,22 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data, srf->flags = req->flags; srf->format = req->format; srf->scanout = req->scanout; - memcpy(srf->mip_levels, req->mip_levels, sizeof(srf->mip_levels)); - srf->num_sizes = 0; - for (i = 0; i < DRM_VMW_MAX_SURFACE_FACES; ++i) - srf->num_sizes += srf->mip_levels[i]; + srf->backup = NULL; - if (srf->num_sizes > DRM_VMW_MAX_SURFACE_FACES * - DRM_VMW_MAX_MIP_LEVELS) { - ret = -EINVAL; - goto out_err0; - } + memcpy(srf->mip_levels, req->mip_levels, sizeof(srf->mip_levels)); + srf->num_sizes = num_sizes; + user_srf->size = size; srf->sizes = kmalloc(srf->num_sizes * sizeof(*srf->sizes), GFP_KERNEL); if (unlikely(srf->sizes == NULL)) { ret = -ENOMEM; - goto out_err0; + goto out_no_sizes; + } + srf->offsets = kmalloc(srf->num_sizes * sizeof(*srf->offsets), + GFP_KERNEL); + if (unlikely(srf->sizes == NULL)) { + ret = -ENOMEM; + goto out_no_offsets; } user_sizes = (struct drm_vmw_size __user *)(unsigned long) @@ -603,23 +1366,44 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data, srf->num_sizes * sizeof(*srf->sizes)); if (unlikely(ret != 0)) { ret = -EFAULT; - goto out_err1; + goto out_no_copy; } + cur_bo_offset = 0; + cur_offset = srf->offsets; + cur_size = srf->sizes; + + bpp = vmw_sf_bpp[srf->format].bpp; + stride_bpp = vmw_sf_bpp[srf->format].s_bpp; + + for (i = 0; i < DRM_VMW_MAX_SURFACE_FACES; ++i) { + for (j = 0; j < srf->mip_levels[i]; ++j) { + uint32_t stride = + (cur_size->width * stride_bpp + 7) >> 3; + + cur_offset->face = i; + cur_offset->mip = j; + cur_offset->bo_offset = cur_bo_offset; + cur_bo_offset += stride * cur_size->height * + cur_size->depth * bpp / stride_bpp; + ++cur_offset; + ++cur_size; + } + } + srf->backup_size = cur_bo_offset; + if (srf->scanout && srf->num_sizes == 1 && srf->sizes[0].width == 64 && srf->sizes[0].height == 64 && srf->format == SVGA3D_A8R8G8B8) { - srf->snooper.image = kmalloc(64 * 64 * 4, GFP_KERNEL); - /* clear the image */ - if (srf->snooper.image) { - memset(srf->snooper.image, 0x00, 64 * 64 * 4); - } else { + /* allocate image area and clear it */ + srf->snooper.image = kzalloc(64 * 64 * 4, GFP_KERNEL); + if (!srf->snooper.image) { DRM_ERROR("Failed to allocate cursor_image\n"); ret = -ENOMEM; - goto out_err1; + goto out_no_copy; } } else { srf->snooper.image = NULL; @@ -636,7 +1420,7 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data, ret = vmw_surface_init(dev_priv, srf, vmw_user_surface_free); if (unlikely(ret != 0)) - return ret; + goto out_unlock; tmp = vmw_resource_reference(&srf->res); ret = ttm_base_object_init(tfile, &user_srf->base, @@ -646,7 +1430,7 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data, if (unlikely(ret != 0)) { vmw_resource_unreference(&tmp); vmw_resource_unreference(&res); - return ret; + goto out_unlock; } rep->sid = user_srf->base.hash.key; @@ -654,11 +1438,19 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data, DRM_ERROR("Created bad Surface ID.\n"); vmw_resource_unreference(&res); + + ttm_read_unlock(&vmaster->lock); return 0; -out_err1: +out_no_copy: + kfree(srf->offsets); +out_no_offsets: kfree(srf->sizes); -out_err0: +out_no_sizes: kfree(user_srf); +out_no_user_srf: + ttm_mem_global_free(vmw_mem_glob(dev_priv), size); +out_unlock: + ttm_read_unlock(&vmaster->lock); return ret; } @@ -972,7 +1764,7 @@ static int vmw_stream_init(struct vmw_private *dev_priv, int ret; ret = vmw_resource_init(dev_priv, res, &dev_priv->stream_idr, - VMW_RES_STREAM, res_free); + VMW_RES_STREAM, false, res_free, NULL); if (unlikely(ret != 0)) { if (res_free == NULL) @@ -1002,8 +1794,11 @@ static void vmw_user_stream_free(struct vmw_resource *res) { struct vmw_user_stream *stream = container_of(res, struct vmw_user_stream, stream.res); + struct vmw_private *dev_priv = res->dev_priv; kfree(stream); + ttm_mem_global_free(vmw_mem_glob(dev_priv), + vmw_user_stream_size); } /** @@ -1057,23 +1852,56 @@ int vmw_stream_claim_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv) { struct vmw_private *dev_priv = vmw_priv(dev); - struct vmw_user_stream *stream = kmalloc(sizeof(*stream), GFP_KERNEL); + struct vmw_user_stream *stream; struct vmw_resource *res; 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; - if (unlikely(stream == NULL)) - return -ENOMEM; + /* + * Approximate idr memory usage with 128 bytes. It will be limited + * by maximum number_of streams anyway? + */ + + if (unlikely(vmw_user_stream_size == 0)) + vmw_user_stream_size = ttm_round_pot(sizeof(*stream)) + 128; + + ret = ttm_read_lock(&vmaster->lock, true); + if (unlikely(ret != 0)) + return ret; + + ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), + vmw_user_stream_size, + false, true); + if (unlikely(ret != 0)) { + if (ret != -ERESTARTSYS) + DRM_ERROR("Out of graphics memory for stream" + " creation.\n"); + goto out_unlock; + } + + + stream = kmalloc(sizeof(*stream), GFP_KERNEL); + if (unlikely(stream == NULL)) { + ttm_mem_global_free(vmw_mem_glob(dev_priv), + vmw_user_stream_size); + ret = -ENOMEM; + goto out_unlock; + } res = &stream->stream.res; stream->base.shareable = false; stream->base.tfile = NULL; + /* + * From here on, the destructor takes over resource freeing. + */ + ret = vmw_stream_init(dev_priv, &stream->stream, vmw_user_stream_free); if (unlikely(ret != 0)) - return ret; + goto out_unlock; tmp = vmw_resource_reference(res); ret = ttm_base_object_init(tfile, &stream->base, false, VMW_RES_STREAM, @@ -1087,6 +1915,8 @@ int vmw_stream_claim_ioctl(struct drm_device *dev, void *data, arg->stream_id = res->id; out_err: vmw_resource_unreference(&res); +out_unlock: + ttm_read_unlock(&vmaster->lock); return ret; } @@ -1120,3 +1950,76 @@ err_ref: vmw_resource_unreference(&res); return ret; } + + +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; + 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); + 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); + if (unlikely(ret != 0)) + goto out_no_base_object; + + args->handle = vmw_user_bo->base.hash.key; + +out_no_base_object: + ttm_bo_unref(&tmp); +out_no_dmabuf: + ttm_read_unlock(&vmaster->lock); + return ret; +} + +int vmw_dumb_map_offset(struct drm_file *file_priv, + struct drm_device *dev, uint32_t handle, + uint64_t *offset) +{ + struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile; + struct vmw_dma_buffer *out_buf; + int ret; + + ret = vmw_user_dmabuf_lookup(tfile, handle, &out_buf); + if (ret != 0) + return -EINVAL; + + *offset = out_buf->base.addr_space_offset; + vmw_dmabuf_unreference(&out_buf); + return 0; +} + +int vmw_dumb_destroy(struct drm_file *file_priv, + struct drm_device *dev, + uint32_t handle) +{ + return ttm_ref_object_base_unref(vmw_fpriv(file_priv)->tfile, + handle, TTM_REF_USAGE); +} diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_glue.c b/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_glue.c index 1e8eedd..d3c11f5 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_glue.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_glue.c @@ -34,9 +34,8 @@ int vmw_mmap(struct file *filp, struct vm_area_struct *vma) struct vmw_private *dev_priv; if (unlikely(vma->vm_pgoff < VMWGFX_FILE_PAGE_OFFSET)) { - if (vmw_fifo_mmap(filp, vma) == 0) - return 0; - return drm_mmap(filp, vma); + DRM_ERROR("Illegal attempt to mmap old fifo space.\n"); + return -EINVAL; } file_priv = filp->private_data; |