diff options
Diffstat (limited to 'drivers/gpu/drm/vmwgfx')
-rw-r--r-- | drivers/gpu/drm/vmwgfx/svga3d_reg.h | 153 | ||||
-rw-r--r-- | drivers/gpu/drm/vmwgfx/svga3d_surfacedefs.h | 11 | ||||
-rw-r--r-- | drivers/gpu/drm/vmwgfx/svga_reg.h | 9 | ||||
-rw-r--r-- | drivers/gpu/drm/vmwgfx/vmwgfx_context.c | 141 | ||||
-rw-r--r-- | drivers/gpu/drm/vmwgfx/vmwgfx_drv.c | 10 | ||||
-rw-r--r-- | drivers/gpu/drm/vmwgfx/vmwgfx_drv.h | 38 | ||||
-rw-r--r-- | drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c | 337 | ||||
-rw-r--r-- | drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c | 96 | ||||
-rw-r--r-- | drivers/gpu/drm/vmwgfx/vmwgfx_mob.c | 36 | ||||
-rw-r--r-- | drivers/gpu/drm/vmwgfx/vmwgfx_resource.c | 14 | ||||
-rw-r--r-- | drivers/gpu/drm/vmwgfx/vmwgfx_shader.c | 469 | ||||
-rw-r--r-- | drivers/gpu/drm/vmwgfx/vmwgfx_surface.c | 6 |
12 files changed, 1108 insertions, 212 deletions
diff --git a/drivers/gpu/drm/vmwgfx/svga3d_reg.h b/drivers/gpu/drm/vmwgfx/svga3d_reg.h index d95335c..f58dc7d 100644 --- a/drivers/gpu/drm/vmwgfx/svga3d_reg.h +++ b/drivers/gpu/drm/vmwgfx/svga3d_reg.h @@ -261,12 +261,7 @@ typedef enum SVGA3dSurfaceFormat { /* Planar video formats. */ SVGA3D_YV12 = 121, - /* Shader constant formats. */ - SVGA3D_SURFACE_SHADERCONST_FLOAT = 122, - SVGA3D_SURFACE_SHADERCONST_INT = 123, - SVGA3D_SURFACE_SHADERCONST_BOOL = 124, - - SVGA3D_FORMAT_MAX = 125, + SVGA3D_FORMAT_MAX = 122, } SVGA3dSurfaceFormat; typedef uint32 SVGA3dColor; /* a, r, g, b */ @@ -1223,9 +1218,19 @@ typedef enum { #define SVGA_3D_CMD_INVALIDATE_GB_IMAGE_PARTIAL 1129 #define SVGA_3D_CMD_SET_GB_SHADERCONSTS_INLINE 1130 - +#define SVGA_3D_CMD_GB_SCREEN_DMA 1131 +#define SVGA_3D_CMD_BIND_GB_SURFACE_WITH_PITCH 1132 +#define SVGA_3D_CMD_GB_MOB_FENCE 1133 +#define SVGA_3D_CMD_DEFINE_GB_SURFACE_V2 1134 #define SVGA_3D_CMD_DEFINE_GB_MOB64 1135 #define SVGA_3D_CMD_REDEFINE_GB_MOB64 1136 +#define SVGA_3D_CMD_NOP_ERROR 1137 + +#define SVGA_3D_CMD_RESERVED1 1138 +#define SVGA_3D_CMD_RESERVED2 1139 +#define SVGA_3D_CMD_RESERVED3 1140 +#define SVGA_3D_CMD_RESERVED4 1141 +#define SVGA_3D_CMD_RESERVED5 1142 #define SVGA_3D_CMD_MAX 1142 #define SVGA_3D_CMD_FUTURE_MAX 3000 @@ -1973,8 +1978,7 @@ struct { uint32 sizeInBytes; uint32 validSizeInBytes; SVGAMobFormat ptDepth; -} -__attribute__((__packed__)) +} __packed SVGA3dCmdSetOTableBase; /* SVGA_3D_CMD_SET_OTABLE_BASE */ typedef @@ -1984,15 +1988,13 @@ struct { uint32 sizeInBytes; uint32 validSizeInBytes; SVGAMobFormat ptDepth; -} -__attribute__((__packed__)) +} __packed SVGA3dCmdSetOTableBase64; /* SVGA_3D_CMD_SET_OTABLE_BASE64 */ typedef struct { SVGAOTableType type; -} -__attribute__((__packed__)) +} __packed SVGA3dCmdReadbackOTable; /* SVGA_3D_CMD_READBACK_OTABLE */ /* @@ -2005,8 +2007,7 @@ struct SVGA3dCmdDefineGBMob { SVGAMobFormat ptDepth; PPN base; uint32 sizeInBytes; -} -__attribute__((__packed__)) +} __packed SVGA3dCmdDefineGBMob; /* SVGA_3D_CMD_DEFINE_GB_MOB */ @@ -2017,8 +2018,7 @@ SVGA3dCmdDefineGBMob; /* SVGA_3D_CMD_DEFINE_GB_MOB */ typedef struct SVGA3dCmdDestroyGBMob { SVGAMobId mobid; -} -__attribute__((__packed__)) +} __packed SVGA3dCmdDestroyGBMob; /* SVGA_3D_CMD_DESTROY_GB_MOB */ /* @@ -2031,8 +2031,7 @@ struct SVGA3dCmdRedefineGBMob { SVGAMobFormat ptDepth; PPN base; uint32 sizeInBytes; -} -__attribute__((__packed__)) +} __packed SVGA3dCmdRedefineGBMob; /* SVGA_3D_CMD_REDEFINE_GB_MOB */ /* @@ -2045,8 +2044,7 @@ struct SVGA3dCmdDefineGBMob64 { SVGAMobFormat ptDepth; PPN64 base; uint32 sizeInBytes; -} -__attribute__((__packed__)) +} __packed SVGA3dCmdDefineGBMob64; /* SVGA_3D_CMD_DEFINE_GB_MOB64 */ /* @@ -2059,8 +2057,7 @@ struct SVGA3dCmdRedefineGBMob64 { SVGAMobFormat ptDepth; PPN64 base; uint32 sizeInBytes; -} -__attribute__((__packed__)) +} __packed SVGA3dCmdRedefineGBMob64; /* SVGA_3D_CMD_REDEFINE_GB_MOB64 */ /* @@ -2070,8 +2067,7 @@ SVGA3dCmdRedefineGBMob64; /* SVGA_3D_CMD_REDEFINE_GB_MOB64 */ typedef struct SVGA3dCmdUpdateGBMobMapping { SVGAMobId mobid; -} -__attribute__((__packed__)) +} __packed SVGA3dCmdUpdateGBMobMapping; /* SVGA_3D_CMD_UPDATE_GB_MOB_MAPPING */ /* @@ -2087,7 +2083,8 @@ struct SVGA3dCmdDefineGBSurface { uint32 multisampleCount; SVGA3dTextureFilter autogenFilter; SVGA3dSize size; -} SVGA3dCmdDefineGBSurface; /* SVGA_3D_CMD_DEFINE_GB_SURFACE */ +} __packed +SVGA3dCmdDefineGBSurface; /* SVGA_3D_CMD_DEFINE_GB_SURFACE */ /* * Destroy a guest-backed surface. @@ -2096,7 +2093,8 @@ struct SVGA3dCmdDefineGBSurface { typedef struct SVGA3dCmdDestroyGBSurface { uint32 sid; -} SVGA3dCmdDestroyGBSurface; /* SVGA_3D_CMD_DESTROY_GB_SURFACE */ +} __packed +SVGA3dCmdDestroyGBSurface; /* SVGA_3D_CMD_DESTROY_GB_SURFACE */ /* * Bind a guest-backed surface to an object. @@ -2106,7 +2104,8 @@ typedef struct SVGA3dCmdBindGBSurface { uint32 sid; SVGAMobId mobid; -} SVGA3dCmdBindGBSurface; /* SVGA_3D_CMD_BIND_GB_SURFACE */ +} __packed +SVGA3dCmdBindGBSurface; /* SVGA_3D_CMD_BIND_GB_SURFACE */ /* * Conditionally bind a mob to a guest backed surface if testMobid @@ -2123,7 +2122,7 @@ struct{ SVGAMobId testMobid; SVGAMobId mobid; uint32 flags; -} +} __packed SVGA3dCmdCondBindGBSurface; /* SVGA_3D_CMD_COND_BIND_GB_SURFACE */ /* @@ -2135,7 +2134,8 @@ typedef struct SVGA3dCmdUpdateGBImage { SVGA3dSurfaceImageId image; SVGA3dBox box; -} SVGA3dCmdUpdateGBImage; /* SVGA_3D_CMD_UPDATE_GB_IMAGE */ +} __packed +SVGA3dCmdUpdateGBImage; /* SVGA_3D_CMD_UPDATE_GB_IMAGE */ /* * Update an entire guest-backed surface. @@ -2145,7 +2145,8 @@ struct SVGA3dCmdUpdateGBImage { typedef struct SVGA3dCmdUpdateGBSurface { uint32 sid; -} SVGA3dCmdUpdateGBSurface; /* SVGA_3D_CMD_UPDATE_GB_SURFACE */ +} __packed +SVGA3dCmdUpdateGBSurface; /* SVGA_3D_CMD_UPDATE_GB_SURFACE */ /* * Readback an image in a guest-backed surface. @@ -2155,7 +2156,8 @@ struct SVGA3dCmdUpdateGBSurface { typedef struct SVGA3dCmdReadbackGBImage { SVGA3dSurfaceImageId image; -} SVGA3dCmdReadbackGBImage; /* SVGA_3D_CMD_READBACK_GB_IMAGE*/ +} __packed +SVGA3dCmdReadbackGBImage; /* SVGA_3D_CMD_READBACK_GB_IMAGE*/ /* * Readback an entire guest-backed surface. @@ -2165,7 +2167,8 @@ struct SVGA3dCmdReadbackGBImage { typedef struct SVGA3dCmdReadbackGBSurface { uint32 sid; -} SVGA3dCmdReadbackGBSurface; /* SVGA_3D_CMD_READBACK_GB_SURFACE */ +} __packed +SVGA3dCmdReadbackGBSurface; /* SVGA_3D_CMD_READBACK_GB_SURFACE */ /* * Readback a sub rect of an image in a guest-backed surface. After @@ -2179,7 +2182,7 @@ struct SVGA3dCmdReadbackGBImagePartial { SVGA3dSurfaceImageId image; SVGA3dBox box; uint32 invertBox; -} +} __packed SVGA3dCmdReadbackGBImagePartial; /* SVGA_3D_CMD_READBACK_GB_IMAGE_PARTIAL */ /* @@ -2190,7 +2193,8 @@ SVGA3dCmdReadbackGBImagePartial; /* SVGA_3D_CMD_READBACK_GB_IMAGE_PARTIAL */ typedef struct SVGA3dCmdInvalidateGBImage { SVGA3dSurfaceImageId image; -} SVGA3dCmdInvalidateGBImage; /* SVGA_3D_CMD_INVALIDATE_GB_IMAGE */ +} __packed +SVGA3dCmdInvalidateGBImage; /* SVGA_3D_CMD_INVALIDATE_GB_IMAGE */ /* * Invalidate an entire guest-backed surface. @@ -2200,7 +2204,8 @@ struct SVGA3dCmdInvalidateGBImage { typedef struct SVGA3dCmdInvalidateGBSurface { uint32 sid; -} SVGA3dCmdInvalidateGBSurface; /* SVGA_3D_CMD_INVALIDATE_GB_SURFACE */ +} __packed +SVGA3dCmdInvalidateGBSurface; /* SVGA_3D_CMD_INVALIDATE_GB_SURFACE */ /* * Invalidate a sub rect of an image in a guest-backed surface. After @@ -2214,7 +2219,7 @@ struct SVGA3dCmdInvalidateGBImagePartial { SVGA3dSurfaceImageId image; SVGA3dBox box; uint32 invertBox; -} +} __packed SVGA3dCmdInvalidateGBImagePartial; /* SVGA_3D_CMD_INVALIDATE_GB_IMAGE_PARTIAL */ /* @@ -2224,7 +2229,8 @@ SVGA3dCmdInvalidateGBImagePartial; /* SVGA_3D_CMD_INVALIDATE_GB_IMAGE_PARTIAL */ typedef struct SVGA3dCmdDefineGBContext { uint32 cid; -} SVGA3dCmdDefineGBContext; /* SVGA_3D_CMD_DEFINE_GB_CONTEXT */ +} __packed +SVGA3dCmdDefineGBContext; /* SVGA_3D_CMD_DEFINE_GB_CONTEXT */ /* * Destroy a guest-backed context. @@ -2233,7 +2239,8 @@ struct SVGA3dCmdDefineGBContext { typedef struct SVGA3dCmdDestroyGBContext { uint32 cid; -} SVGA3dCmdDestroyGBContext; /* SVGA_3D_CMD_DESTROY_GB_CONTEXT */ +} __packed +SVGA3dCmdDestroyGBContext; /* SVGA_3D_CMD_DESTROY_GB_CONTEXT */ /* * Bind a guest-backed context. @@ -2252,7 +2259,8 @@ struct SVGA3dCmdBindGBContext { uint32 cid; SVGAMobId mobid; uint32 validContents; -} SVGA3dCmdBindGBContext; /* SVGA_3D_CMD_BIND_GB_CONTEXT */ +} __packed +SVGA3dCmdBindGBContext; /* SVGA_3D_CMD_BIND_GB_CONTEXT */ /* * Readback a guest-backed context. @@ -2262,7 +2270,8 @@ struct SVGA3dCmdBindGBContext { typedef struct SVGA3dCmdReadbackGBContext { uint32 cid; -} SVGA3dCmdReadbackGBContext; /* SVGA_3D_CMD_READBACK_GB_CONTEXT */ +} __packed +SVGA3dCmdReadbackGBContext; /* SVGA_3D_CMD_READBACK_GB_CONTEXT */ /* * Invalidate a guest-backed context. @@ -2270,7 +2279,8 @@ struct SVGA3dCmdReadbackGBContext { typedef struct SVGA3dCmdInvalidateGBContext { uint32 cid; -} SVGA3dCmdInvalidateGBContext; /* SVGA_3D_CMD_INVALIDATE_GB_CONTEXT */ +} __packed +SVGA3dCmdInvalidateGBContext; /* SVGA_3D_CMD_INVALIDATE_GB_CONTEXT */ /* * Define a guest-backed shader. @@ -2281,7 +2291,8 @@ struct SVGA3dCmdDefineGBShader { uint32 shid; SVGA3dShaderType type; uint32 sizeInBytes; -} SVGA3dCmdDefineGBShader; /* SVGA_3D_CMD_DEFINE_GB_SHADER */ +} __packed +SVGA3dCmdDefineGBShader; /* SVGA_3D_CMD_DEFINE_GB_SHADER */ /* * Bind a guest-backed shader. @@ -2291,7 +2302,8 @@ typedef struct SVGA3dCmdBindGBShader { uint32 shid; SVGAMobId mobid; uint32 offsetInBytes; -} SVGA3dCmdBindGBShader; /* SVGA_3D_CMD_BIND_GB_SHADER */ +} __packed +SVGA3dCmdBindGBShader; /* SVGA_3D_CMD_BIND_GB_SHADER */ /* * Destroy a guest-backed shader. @@ -2299,7 +2311,8 @@ typedef struct SVGA3dCmdBindGBShader { typedef struct SVGA3dCmdDestroyGBShader { uint32 shid; -} SVGA3dCmdDestroyGBShader; /* SVGA_3D_CMD_DESTROY_GB_SHADER */ +} __packed +SVGA3dCmdDestroyGBShader; /* SVGA_3D_CMD_DESTROY_GB_SHADER */ typedef struct { @@ -2314,14 +2327,16 @@ struct { * Note that FLOAT and INT constants are 4-dwords in length, while * BOOL constants are 1-dword in length. */ -} SVGA3dCmdSetGBShaderConstInline; +} __packed +SVGA3dCmdSetGBShaderConstInline; /* SVGA_3D_CMD_SET_GB_SHADERCONSTS_INLINE */ typedef struct { uint32 cid; SVGA3dQueryType type; -} SVGA3dCmdBeginGBQuery; /* SVGA_3D_CMD_BEGIN_GB_QUERY */ +} __packed +SVGA3dCmdBeginGBQuery; /* SVGA_3D_CMD_BEGIN_GB_QUERY */ typedef struct { @@ -2329,7 +2344,8 @@ struct { SVGA3dQueryType type; SVGAMobId mobid; uint32 offset; -} SVGA3dCmdEndGBQuery; /* SVGA_3D_CMD_END_GB_QUERY */ +} __packed +SVGA3dCmdEndGBQuery; /* SVGA_3D_CMD_END_GB_QUERY */ /* @@ -2346,21 +2362,22 @@ struct { SVGA3dQueryType type; SVGAMobId mobid; uint32 offset; -} SVGA3dCmdWaitForGBQuery; /* SVGA_3D_CMD_WAIT_FOR_GB_QUERY */ +} __packed +SVGA3dCmdWaitForGBQuery; /* SVGA_3D_CMD_WAIT_FOR_GB_QUERY */ typedef struct { SVGAMobId mobid; uint32 fbOffset; uint32 initalized; -} +} __packed SVGA3dCmdEnableGart; /* SVGA_3D_CMD_ENABLE_GART */ typedef struct { SVGAMobId mobid; uint32 gartOffset; -} +} __packed SVGA3dCmdMapMobIntoGart; /* SVGA_3D_CMD_MAP_MOB_INTO_GART */ @@ -2368,7 +2385,7 @@ typedef struct { uint32 gartOffset; uint32 numPages; -} +} __packed SVGA3dCmdUnmapGartRange; /* SVGA_3D_CMD_UNMAP_GART_RANGE */ @@ -2385,27 +2402,27 @@ struct { int32 xRoot; int32 yRoot; uint32 flags; -} +} __packed SVGA3dCmdDefineGBScreenTarget; /* SVGA_3D_CMD_DEFINE_GB_SCREENTARGET */ typedef struct { uint32 stid; -} +} __packed SVGA3dCmdDestroyGBScreenTarget; /* SVGA_3D_CMD_DESTROY_GB_SCREENTARGET */ typedef struct { uint32 stid; SVGA3dSurfaceImageId image; -} +} __packed SVGA3dCmdBindGBScreenTarget; /* SVGA_3D_CMD_BIND_GB_SCREENTARGET */ typedef struct { uint32 stid; SVGA3dBox box; -} +} __packed SVGA3dCmdUpdateGBScreenTarget; /* SVGA_3D_CMD_UPDATE_GB_SCREENTARGET */ /* @@ -2583,4 +2600,28 @@ typedef union { float f; } SVGA3dDevCapResult; +typedef enum { + SVGA3DCAPS_RECORD_UNKNOWN = 0, + SVGA3DCAPS_RECORD_DEVCAPS_MIN = 0x100, + SVGA3DCAPS_RECORD_DEVCAPS = 0x100, + SVGA3DCAPS_RECORD_DEVCAPS_MAX = 0x1ff, +} SVGA3dCapsRecordType; + +typedef +struct SVGA3dCapsRecordHeader { + uint32 length; + SVGA3dCapsRecordType type; +} +SVGA3dCapsRecordHeader; + +typedef +struct SVGA3dCapsRecord { + SVGA3dCapsRecordHeader header; + uint32 data[1]; +} +SVGA3dCapsRecord; + + +typedef uint32 SVGA3dCapPair[2]; + #endif /* _SVGA3D_REG_H_ */ diff --git a/drivers/gpu/drm/vmwgfx/svga3d_surfacedefs.h b/drivers/gpu/drm/vmwgfx/svga3d_surfacedefs.h index 8369c3b..ef33850 100644 --- a/drivers/gpu/drm/vmwgfx/svga3d_surfacedefs.h +++ b/drivers/gpu/drm/vmwgfx/svga3d_surfacedefs.h @@ -38,8 +38,11 @@ #define DIV_ROUND_UP(x, y) (((x) + (y) - 1) / (y)) #define max_t(type, x, y) ((x) > (y) ? (x) : (y)) +#define min_t(type, x, y) ((x) < (y) ? (x) : (y)) #define surf_size_struct SVGA3dSize #define u32 uint32 +#define u64 uint64_t +#define U32_MAX ((u32)~0U) #endif /* __KERNEL__ */ @@ -704,8 +707,8 @@ static const struct svga3d_surface_desc svga3d_surface_descs[] = { static inline u32 clamped_umul32(u32 a, u32 b) { - uint64_t tmp = (uint64_t) a*b; - return (tmp > (uint64_t) ((u32) -1)) ? (u32) -1 : tmp; + u64 tmp = (u64) a*b; + return (tmp > (u64) U32_MAX) ? U32_MAX : tmp; } static inline const struct svga3d_surface_desc * @@ -834,7 +837,7 @@ svga3dsurface_get_serialized_size(SVGA3dSurfaceFormat format, bool cubemap) { const struct svga3d_surface_desc *desc = svga3dsurface_get_desc(format); - u32 total_size = 0; + u64 total_size = 0; u32 mip; for (mip = 0; mip < num_mip_levels; mip++) { @@ -847,7 +850,7 @@ svga3dsurface_get_serialized_size(SVGA3dSurfaceFormat format, if (cubemap) total_size *= SVGA3D_MAX_SURFACE_FACES; - return total_size; + return (u32) min_t(u64, total_size, (u64) U32_MAX); } diff --git a/drivers/gpu/drm/vmwgfx/svga_reg.h b/drivers/gpu/drm/vmwgfx/svga_reg.h index 71defa4..11323dd 100644 --- a/drivers/gpu/drm/vmwgfx/svga_reg.h +++ b/drivers/gpu/drm/vmwgfx/svga_reg.h @@ -169,10 +169,17 @@ enum { SVGA_REG_TRACES = 45, /* Enable trace-based updates even when FIFO is on */ SVGA_REG_GMRS_MAX_PAGES = 46, /* Maximum number of 4KB pages for all GMRs */ SVGA_REG_MEMORY_SIZE = 47, /* Total dedicated device memory excluding FIFO */ + SVGA_REG_COMMAND_LOW = 48, /* Lower 32 bits and submits commands */ + SVGA_REG_COMMAND_HIGH = 49, /* Upper 32 bits of command buffer PA */ SVGA_REG_MAX_PRIMARY_BOUNDING_BOX_MEM = 50, /* Max primary memory */ SVGA_REG_SUGGESTED_GBOBJECT_MEM_SIZE_KB = 51, /* Suggested limit on mob mem */ SVGA_REG_DEV_CAP = 52, /* Write dev cap index, read value */ - SVGA_REG_TOP = 53, /* Must be 1 more than the last register */ + SVGA_REG_CMD_PREPEND_LOW = 53, + SVGA_REG_CMD_PREPEND_HIGH = 54, + SVGA_REG_SCREENTARGET_MAX_WIDTH = 55, + SVGA_REG_SCREENTARGET_MAX_HEIGHT = 56, + SVGA_REG_MOB_MAX_SIZE = 57, + SVGA_REG_TOP = 58, /* Must be 1 more than the last register */ SVGA_PALETTE_BASE = 1024, /* Base of SVGA color map */ /* Next 768 (== 256*3) registers exist for colormap */ diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_context.c b/drivers/gpu/drm/vmwgfx/vmwgfx_context.c index 82c41da..1e80152 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_context.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_context.c @@ -37,7 +37,7 @@ struct vmw_user_context { -typedef int (*vmw_scrub_func)(struct vmw_ctx_bindinfo *); +typedef int (*vmw_scrub_func)(struct vmw_ctx_bindinfo *, bool); static void vmw_user_context_free(struct vmw_resource *res); static struct vmw_resource * @@ -50,9 +50,11 @@ static int vmw_gb_context_unbind(struct vmw_resource *res, bool readback, struct ttm_validate_buffer *val_buf); static int vmw_gb_context_destroy(struct vmw_resource *res); -static int vmw_context_scrub_shader(struct vmw_ctx_bindinfo *bi); -static int vmw_context_scrub_render_target(struct vmw_ctx_bindinfo *bi); -static int vmw_context_scrub_texture(struct vmw_ctx_bindinfo *bi); +static int vmw_context_scrub_shader(struct vmw_ctx_bindinfo *bi, bool rebind); +static int vmw_context_scrub_render_target(struct vmw_ctx_bindinfo *bi, + bool rebind); +static int vmw_context_scrub_texture(struct vmw_ctx_bindinfo *bi, bool rebind); +static void vmw_context_binding_state_scrub(struct vmw_ctx_binding_state *cbs); static void vmw_context_binding_state_kill(struct vmw_ctx_binding_state *cbs); static uint64_t vmw_user_context_size; @@ -111,10 +113,14 @@ static void vmw_hw_context_destroy(struct vmw_resource *res) if (res->func->destroy == vmw_gb_context_destroy) { mutex_lock(&dev_priv->cmdbuf_mutex); + mutex_lock(&dev_priv->binding_mutex); + (void) vmw_context_binding_state_kill + (&container_of(res, struct vmw_user_context, res)->cbs); (void) vmw_gb_context_destroy(res); if (dev_priv->pinned_bo != NULL && !dev_priv->query_cid_valid) __vmw_execbuf_release_pinned_bo(dev_priv, NULL); + mutex_unlock(&dev_priv->binding_mutex); mutex_unlock(&dev_priv->cmdbuf_mutex); return; } @@ -328,7 +334,7 @@ static int vmw_gb_context_unbind(struct vmw_resource *res, BUG_ON(bo->mem.mem_type != VMW_PL_MOB); mutex_lock(&dev_priv->binding_mutex); - vmw_context_binding_state_kill(&uctx->cbs); + vmw_context_binding_state_scrub(&uctx->cbs); submit_size = sizeof(*cmd2) + (readback ? sizeof(*cmd1) : 0); @@ -378,10 +384,6 @@ static int vmw_gb_context_destroy(struct vmw_resource *res) SVGA3dCmdHeader header; SVGA3dCmdDestroyGBContext body; } *cmd; - struct vmw_user_context *uctx = - container_of(res, struct vmw_user_context, res); - - BUG_ON(!list_empty(&uctx->cbs.list)); if (likely(res->id == -1)) return 0; @@ -528,8 +530,9 @@ out_unlock: * vmw_context_scrub_shader - scrub a shader binding from a context. * * @bi: single binding information. + * @rebind: Whether to issue a bind instead of scrub command. */ -static int vmw_context_scrub_shader(struct vmw_ctx_bindinfo *bi) +static int vmw_context_scrub_shader(struct vmw_ctx_bindinfo *bi, bool rebind) { struct vmw_private *dev_priv = bi->ctx->dev_priv; struct { @@ -548,7 +551,7 @@ static int vmw_context_scrub_shader(struct vmw_ctx_bindinfo *bi) cmd->header.size = sizeof(cmd->body); cmd->body.cid = bi->ctx->id; cmd->body.type = bi->i1.shader_type; - cmd->body.shid = SVGA3D_INVALID_ID; + cmd->body.shid = ((rebind) ? bi->res->id : SVGA3D_INVALID_ID); vmw_fifo_commit(dev_priv, sizeof(*cmd)); return 0; @@ -559,8 +562,10 @@ static int vmw_context_scrub_shader(struct vmw_ctx_bindinfo *bi) * from a context. * * @bi: single binding information. + * @rebind: Whether to issue a bind instead of scrub command. */ -static int vmw_context_scrub_render_target(struct vmw_ctx_bindinfo *bi) +static int vmw_context_scrub_render_target(struct vmw_ctx_bindinfo *bi, + bool rebind) { struct vmw_private *dev_priv = bi->ctx->dev_priv; struct { @@ -579,7 +584,7 @@ static int vmw_context_scrub_render_target(struct vmw_ctx_bindinfo *bi) cmd->header.size = sizeof(cmd->body); cmd->body.cid = bi->ctx->id; cmd->body.type = bi->i1.rt_type; - cmd->body.target.sid = SVGA3D_INVALID_ID; + cmd->body.target.sid = ((rebind) ? bi->res->id : SVGA3D_INVALID_ID); cmd->body.target.face = 0; cmd->body.target.mipmap = 0; vmw_fifo_commit(dev_priv, sizeof(*cmd)); @@ -591,11 +596,13 @@ static int vmw_context_scrub_render_target(struct vmw_ctx_bindinfo *bi) * vmw_context_scrub_texture - scrub a texture binding from a context. * * @bi: single binding information. + * @rebind: Whether to issue a bind instead of scrub command. * * TODO: Possibly complement this function with a function that takes * a list of texture bindings and combines them to a single command. */ -static int vmw_context_scrub_texture(struct vmw_ctx_bindinfo *bi) +static int vmw_context_scrub_texture(struct vmw_ctx_bindinfo *bi, + bool rebind) { struct vmw_private *dev_priv = bi->ctx->dev_priv; struct { @@ -619,7 +626,7 @@ static int vmw_context_scrub_texture(struct vmw_ctx_bindinfo *bi) cmd->body.c.cid = bi->ctx->id; cmd->body.s1.stage = bi->i1.texture_stage; cmd->body.s1.name = SVGA3D_TS_BIND_TEXTURE; - cmd->body.s1.value = (uint32) SVGA3D_INVALID_ID; + cmd->body.s1.value = ((rebind) ? bi->res->id : SVGA3D_INVALID_ID); vmw_fifo_commit(dev_priv, sizeof(*cmd)); return 0; @@ -692,6 +699,7 @@ int vmw_context_binding_add(struct vmw_ctx_binding_state *cbs, vmw_context_binding_drop(loc); loc->bi = *bi; + loc->bi.scrubbed = false; list_add_tail(&loc->ctx_list, &cbs->list); INIT_LIST_HEAD(&loc->res_list); @@ -727,12 +735,11 @@ static void vmw_context_binding_transfer(struct vmw_ctx_binding_state *cbs, if (loc->bi.ctx != NULL) vmw_context_binding_drop(loc); - loc->bi = *bi; - list_add_tail(&loc->ctx_list, &cbs->list); - if (bi->res != NULL) + if (bi->res != NULL) { + loc->bi = *bi; + list_add_tail(&loc->ctx_list, &cbs->list); list_add_tail(&loc->res_list, &bi->res->binding_head); - else - INIT_LIST_HEAD(&loc->res_list); + } } /** @@ -746,7 +753,10 @@ static void vmw_context_binding_transfer(struct vmw_ctx_binding_state *cbs, */ static void vmw_context_binding_kill(struct vmw_ctx_binding *cb) { - (void) vmw_scrub_funcs[cb->bi.bt](&cb->bi); + if (!cb->bi.scrubbed) { + (void) vmw_scrub_funcs[cb->bi.bt](&cb->bi, false); + cb->bi.scrubbed = true; + } vmw_context_binding_drop(cb); } @@ -768,6 +778,27 @@ static void vmw_context_binding_state_kill(struct vmw_ctx_binding_state *cbs) } /** + * vmw_context_binding_state_scrub - Scrub all bindings associated with a + * struct vmw_ctx_binding state structure. + * + * @cbs: Pointer to the context binding state tracker. + * + * Emits commands to scrub all bindings associated with the + * context binding state tracker. + */ +static void vmw_context_binding_state_scrub(struct vmw_ctx_binding_state *cbs) +{ + struct vmw_ctx_binding *entry; + + list_for_each_entry(entry, &cbs->list, ctx_list) { + if (!entry->bi.scrubbed) { + (void) vmw_scrub_funcs[entry->bi.bt](&entry->bi, false); + entry->bi.scrubbed = true; + } + } +} + +/** * vmw_context_binding_res_list_kill - Kill all bindings on a * resource binding list * @@ -785,6 +816,27 @@ void vmw_context_binding_res_list_kill(struct list_head *head) } /** + * vmw_context_binding_res_list_scrub - Scrub all bindings on a + * resource binding list + * + * @head: list head of resource binding list + * + * Scrub all bindings associated with a specific resource. Typically + * called before the resource is evicted. + */ +void vmw_context_binding_res_list_scrub(struct list_head *head) +{ + struct vmw_ctx_binding *entry; + + list_for_each_entry(entry, head, res_list) { + if (!entry->bi.scrubbed) { + (void) vmw_scrub_funcs[entry->bi.bt](&entry->bi, false); + entry->bi.scrubbed = true; + } + } +} + +/** * vmw_context_binding_state_transfer - Commit staged binding info * * @ctx: Pointer to context to commit the staged binding info to. @@ -803,3 +855,50 @@ void vmw_context_binding_state_transfer(struct vmw_resource *ctx, list_for_each_entry_safe(entry, next, &from->list, ctx_list) vmw_context_binding_transfer(&uctx->cbs, &entry->bi); } + +/** + * vmw_context_rebind_all - Rebind all scrubbed bindings of a context + * + * @ctx: The context resource + * + * Walks through the context binding list and rebinds all scrubbed + * resources. + */ +int vmw_context_rebind_all(struct vmw_resource *ctx) +{ + struct vmw_ctx_binding *entry; + struct vmw_user_context *uctx = + container_of(ctx, struct vmw_user_context, res); + struct vmw_ctx_binding_state *cbs = &uctx->cbs; + int ret; + + list_for_each_entry(entry, &cbs->list, ctx_list) { + if (likely(!entry->bi.scrubbed)) + continue; + + if (WARN_ON(entry->bi.res == NULL || entry->bi.res->id == + SVGA3D_INVALID_ID)) + continue; + + ret = vmw_scrub_funcs[entry->bi.bt](&entry->bi, true); + if (unlikely(ret != 0)) + return ret; + + entry->bi.scrubbed = false; + } + + return 0; +} + +/** + * vmw_context_binding_list - Return a list of context bindings + * + * @ctx: The context resource + * + * Returns the current list of bindings of the given context. Note that + * this list becomes stale as soon as the dev_priv::binding_mutex is unlocked. + */ +struct list_head *vmw_context_binding_list(struct vmw_resource *ctx) +{ + return &(container_of(ctx, struct vmw_user_context, res)->cbs.list); +} diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c index 9893328..0083cbf 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c @@ -667,6 +667,7 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset) dev_priv->memory_size = 512*1024*1024; } dev_priv->max_mob_pages = 0; + dev_priv->max_mob_size = 0; if (dev_priv->capabilities & SVGA_CAP_GBOBJECTS) { uint64_t mem_size = vmw_read(dev_priv, @@ -676,6 +677,8 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset) dev_priv->prim_bb_mem = vmw_read(dev_priv, SVGA_REG_MAX_PRIMARY_BOUNDING_BOX_MEM); + dev_priv->max_mob_size = + vmw_read(dev_priv, SVGA_REG_MOB_MAX_SIZE); } else dev_priv->prim_bb_mem = dev_priv->vram_size; @@ -941,6 +944,7 @@ static void vmw_postclose(struct drm_device *dev, drm_master_put(&vmw_fp->locked_master); } + vmw_compat_shader_man_destroy(vmw_fp->shman); ttm_object_file_release(&vmw_fp->tfile); kfree(vmw_fp); } @@ -960,11 +964,17 @@ static int vmw_driver_open(struct drm_device *dev, struct drm_file *file_priv) if (unlikely(vmw_fp->tfile == NULL)) goto out_no_tfile; + vmw_fp->shman = vmw_compat_shader_man_create(dev_priv); + if (IS_ERR(vmw_fp->shman)) + goto out_no_shman; + file_priv->driver_priv = vmw_fp; dev_priv->bdev.dev_mapping = dev->dev_mapping; return 0; +out_no_shman: + ttm_object_file_release(&vmw_fp->tfile); out_no_tfile: kfree(vmw_fp); return ret; diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h index 554e7fa..0783155 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h @@ -40,7 +40,7 @@ #include <drm/ttm/ttm_module.h> #include "vmwgfx_fence.h" -#define VMWGFX_DRIVER_DATE "20121114" +#define VMWGFX_DRIVER_DATE "20140228" #define VMWGFX_DRIVER_MAJOR 2 #define VMWGFX_DRIVER_MINOR 5 #define VMWGFX_DRIVER_PATCHLEVEL 0 @@ -75,10 +75,14 @@ #define VMW_RES_FENCE ttm_driver_type3 #define VMW_RES_SHADER ttm_driver_type4 +struct vmw_compat_shader_manager; + struct vmw_fpriv { struct drm_master *locked_master; struct ttm_object_file *tfile; struct list_head fence_events; + bool gb_aware; + struct vmw_compat_shader_manager *shman; }; struct vmw_dma_buffer { @@ -272,6 +276,7 @@ struct vmw_ctx_bindinfo { struct vmw_resource *ctx; struct vmw_resource *res; enum vmw_ctx_binding_type bt; + bool scrubbed; union { SVGA3dShaderType shader_type; SVGA3dRenderTargetType rt_type; @@ -318,7 +323,7 @@ struct vmw_sw_context{ struct drm_open_hash res_ht; bool res_ht_initialized; bool kernel; /**< is the called made from the kernel */ - struct ttm_object_file *tfile; + struct vmw_fpriv *fp; struct list_head validate_nodes; struct vmw_relocation relocs[VMWGFX_MAX_RELOCATIONS]; uint32_t cur_reloc; @@ -336,6 +341,7 @@ struct vmw_sw_context{ bool needs_post_query_barrier; struct vmw_resource *error_resource; struct vmw_ctx_binding_state staged_bindings; + struct list_head staged_shaders; }; struct vmw_legacy_display; @@ -380,6 +386,7 @@ struct vmw_private { uint32_t max_gmr_ids; uint32_t max_gmr_pages; uint32_t max_mob_pages; + uint32_t max_mob_size; uint32_t memory_size; bool has_gmr; bool has_mob; @@ -569,6 +576,8 @@ struct vmw_user_resource_conv; extern void vmw_resource_unreference(struct vmw_resource **p_res); extern struct vmw_resource *vmw_resource_reference(struct vmw_resource *res); +extern struct vmw_resource * +vmw_resource_reference_unless_doomed(struct vmw_resource *res); extern int vmw_resource_validate(struct vmw_resource *res); extern int vmw_resource_reserve(struct vmw_resource *res, bool no_backup); extern bool vmw_resource_needs_backup(const struct vmw_resource *res); @@ -957,6 +966,9 @@ extern void vmw_context_binding_state_transfer(struct vmw_resource *res, struct vmw_ctx_binding_state *cbs); extern void vmw_context_binding_res_list_kill(struct list_head *head); +extern void vmw_context_binding_res_list_scrub(struct list_head *head); +extern int vmw_context_rebind_all(struct vmw_resource *ctx); +extern struct list_head *vmw_context_binding_list(struct vmw_resource *ctx); /* * Surface management - vmwgfx_surface.c @@ -991,6 +1003,28 @@ extern int vmw_shader_define_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv); extern int vmw_shader_destroy_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv); +extern int vmw_compat_shader_lookup(struct vmw_compat_shader_manager *man, + SVGA3dShaderType shader_type, + u32 *user_key); +extern void vmw_compat_shaders_commit(struct vmw_compat_shader_manager *man, + struct list_head *list); +extern void vmw_compat_shaders_revert(struct vmw_compat_shader_manager *man, + struct list_head *list); +extern int vmw_compat_shader_remove(struct vmw_compat_shader_manager *man, + u32 user_key, + SVGA3dShaderType shader_type, + struct list_head *list); +extern int vmw_compat_shader_add(struct vmw_compat_shader_manager *man, + u32 user_key, const void *bytecode, + SVGA3dShaderType shader_type, + size_t size, + struct ttm_object_file *tfile, + struct list_head *list); +extern struct vmw_compat_shader_manager * +vmw_compat_shader_man_create(struct vmw_private *dev_priv); +extern void +vmw_compat_shader_man_destroy(struct vmw_compat_shader_manager *man); + /** * Inline helper functions diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c index 7a5f1eb..efb575a 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c @@ -114,8 +114,10 @@ static void vmw_resource_list_unreserve(struct list_head *list, * persistent context binding tracker. */ if (unlikely(val->staged_bindings)) { - vmw_context_binding_state_transfer - (val->res, val->staged_bindings); + if (!backoff) { + vmw_context_binding_state_transfer + (val->res, val->staged_bindings); + } kfree(val->staged_bindings); val->staged_bindings = NULL; } @@ -178,6 +180,44 @@ static int vmw_resource_val_add(struct vmw_sw_context *sw_context, } /** + * vmw_resource_context_res_add - Put resources previously bound to a context on + * the validation list + * + * @dev_priv: Pointer to a device private structure + * @sw_context: Pointer to a software context used for this command submission + * @ctx: Pointer to the context resource + * + * This function puts all resources that were previously bound to @ctx on + * the resource validation list. This is part of the context state reemission + */ +static int vmw_resource_context_res_add(struct vmw_private *dev_priv, + struct vmw_sw_context *sw_context, + struct vmw_resource *ctx) +{ + struct list_head *binding_list; + struct vmw_ctx_binding *entry; + int ret = 0; + struct vmw_resource *res; + + mutex_lock(&dev_priv->binding_mutex); + binding_list = vmw_context_binding_list(ctx); + + list_for_each_entry(entry, binding_list, ctx_list) { + res = vmw_resource_reference_unless_doomed(entry->bi.res); + if (unlikely(res == NULL)) + continue; + + ret = vmw_resource_val_add(sw_context, entry->bi.res, NULL); + vmw_resource_unreference(&res); + if (unlikely(ret != 0)) + break; + } + + mutex_unlock(&dev_priv->binding_mutex); + return ret; +} + +/** * vmw_resource_relocation_add - Add a relocation to the relocation list * * @list: Pointer to head of relocation list. @@ -233,8 +273,12 @@ static void vmw_resource_relocations_apply(uint32_t *cb, { struct vmw_resource_relocation *rel; - list_for_each_entry(rel, list, head) - cb[rel->offset] = rel->res->id; + list_for_each_entry(rel, list, head) { + if (likely(rel->res != NULL)) + cb[rel->offset] = rel->res->id; + else + cb[rel->offset] = SVGA_3D_CMD_NOP; + } } static int vmw_cmd_invalid(struct vmw_private *dev_priv, @@ -379,22 +423,27 @@ static int vmw_resources_validate(struct vmw_sw_context *sw_context) } /** - * vmw_cmd_res_check - Check that a resource is present and if so, put it + * vmw_cmd_compat_res_check - Check that a resource is present and if so, put it * on the resource validate list unless it's already there. * * @dev_priv: Pointer to a device private structure. * @sw_context: Pointer to the software context. * @res_type: Resource type. * @converter: User-space visisble type specific information. - * @id: Pointer to the location in the command buffer currently being + * @id: user-space resource id handle. + * @id_loc: Pointer to the location in the command buffer currently being * parsed from where the user-space resource id handle is located. + * @p_val: Pointer to pointer to resource validalidation node. Populated + * on exit. */ -static int vmw_cmd_res_check(struct vmw_private *dev_priv, - struct vmw_sw_context *sw_context, - enum vmw_res_type res_type, - const struct vmw_user_resource_conv *converter, - uint32_t *id, - struct vmw_resource_val_node **p_val) +static int +vmw_cmd_compat_res_check(struct vmw_private *dev_priv, + struct vmw_sw_context *sw_context, + enum vmw_res_type res_type, + const struct vmw_user_resource_conv *converter, + uint32_t id, + uint32_t *id_loc, + struct vmw_resource_val_node **p_val) { struct vmw_res_cache_entry *rcache = &sw_context->res_cache[res_type]; @@ -402,7 +451,7 @@ static int vmw_cmd_res_check(struct vmw_private *dev_priv, struct vmw_resource_val_node *node; int ret; - if (*id == SVGA3D_INVALID_ID) { + if (id == SVGA3D_INVALID_ID) { if (p_val) *p_val = NULL; if (res_type == vmw_res_context) { @@ -417,7 +466,7 @@ static int vmw_cmd_res_check(struct vmw_private *dev_priv, * resource */ - if (likely(rcache->valid && *id == rcache->handle)) { + if (likely(rcache->valid && id == rcache->handle)) { const struct vmw_resource *res = rcache->res; rcache->node->first_usage = false; @@ -426,28 +475,28 @@ static int vmw_cmd_res_check(struct vmw_private *dev_priv, return vmw_resource_relocation_add (&sw_context->res_relocations, res, - id - sw_context->buf_start); + id_loc - sw_context->buf_start); } ret = vmw_user_resource_lookup_handle(dev_priv, - sw_context->tfile, - *id, + sw_context->fp->tfile, + id, converter, &res); if (unlikely(ret != 0)) { DRM_ERROR("Could not find or use resource 0x%08x.\n", - (unsigned) *id); + (unsigned) id); dump_stack(); return ret; } rcache->valid = true; rcache->res = res; - rcache->handle = *id; + rcache->handle = id; ret = vmw_resource_relocation_add(&sw_context->res_relocations, res, - id - sw_context->buf_start); + id_loc - sw_context->buf_start); if (unlikely(ret != 0)) goto out_no_reloc; @@ -459,7 +508,11 @@ static int vmw_cmd_res_check(struct vmw_private *dev_priv, if (p_val) *p_val = node; - if (node->first_usage && res_type == vmw_res_context) { + if (dev_priv->has_mob && node->first_usage && + res_type == vmw_res_context) { + ret = vmw_resource_context_res_add(dev_priv, sw_context, res); + if (unlikely(ret != 0)) + goto out_no_reloc; node->staged_bindings = kzalloc(sizeof(*node->staged_bindings), GFP_KERNEL); if (node->staged_bindings == NULL) { @@ -481,6 +534,59 @@ out_no_reloc: } /** + * vmw_cmd_res_check - Check that a resource is present and if so, put it + * on the resource validate list unless it's already there. + * + * @dev_priv: Pointer to a device private structure. + * @sw_context: Pointer to the software context. + * @res_type: Resource type. + * @converter: User-space visisble type specific information. + * @id_loc: Pointer to the location in the command buffer currently being + * parsed from where the user-space resource id handle is located. + * @p_val: Pointer to pointer to resource validalidation node. Populated + * on exit. + */ +static int +vmw_cmd_res_check(struct vmw_private *dev_priv, + struct vmw_sw_context *sw_context, + enum vmw_res_type res_type, + const struct vmw_user_resource_conv *converter, + uint32_t *id_loc, + struct vmw_resource_val_node **p_val) +{ + return vmw_cmd_compat_res_check(dev_priv, sw_context, res_type, + converter, *id_loc, id_loc, p_val); +} + +/** + * vmw_rebind_contexts - Rebind all resources previously bound to + * referenced contexts. + * + * @sw_context: Pointer to the software context. + * + * Rebind context binding points that have been scrubbed because of eviction. + */ +static int vmw_rebind_contexts(struct vmw_sw_context *sw_context) +{ + struct vmw_resource_val_node *val; + int ret; + + list_for_each_entry(val, &sw_context->resource_list, head) { + if (likely(!val->staged_bindings)) + continue; + + ret = vmw_context_rebind_all(val->res); + if (unlikely(ret != 0)) { + if (ret != -ERESTARTSYS) + DRM_ERROR("Failed to rebind context.\n"); + return ret; + } + } + + return 0; +} + +/** * vmw_cmd_cid_check - Check a command header for valid context information. * * @dev_priv: Pointer to a device private structure. @@ -496,7 +602,7 @@ static int vmw_cmd_cid_check(struct vmw_private *dev_priv, { struct vmw_cid_cmd { SVGA3dCmdHeader header; - __le32 cid; + uint32_t cid; } *cmd; cmd = container_of(header, struct vmw_cid_cmd, header); @@ -767,7 +873,7 @@ static int vmw_translate_mob_ptr(struct vmw_private *dev_priv, struct vmw_relocation *reloc; int ret; - ret = vmw_user_dmabuf_lookup(sw_context->tfile, handle, &vmw_bo); + ret = vmw_user_dmabuf_lookup(sw_context->fp->tfile, handle, &vmw_bo); if (unlikely(ret != 0)) { DRM_ERROR("Could not find or use MOB buffer.\n"); return -EINVAL; @@ -828,7 +934,7 @@ static int vmw_translate_guest_ptr(struct vmw_private *dev_priv, struct vmw_relocation *reloc; int ret; - ret = vmw_user_dmabuf_lookup(sw_context->tfile, handle, &vmw_bo); + ret = vmw_user_dmabuf_lookup(sw_context->fp->tfile, handle, &vmw_bo); if (unlikely(ret != 0)) { DRM_ERROR("Could not find or use GMR region.\n"); return -EINVAL; @@ -1127,7 +1233,8 @@ static int vmw_cmd_dma(struct vmw_private *dev_priv, srf = vmw_res_to_srf(sw_context->res_cache[vmw_res_surface].res); - vmw_kms_cursor_snoop(srf, sw_context->tfile, &vmw_bo->base, header); + vmw_kms_cursor_snoop(srf, sw_context->fp->tfile, &vmw_bo->base, + header); out_no_surface: vmw_dmabuf_unreference(&vmw_bo); @@ -1478,6 +1585,98 @@ static int vmw_cmd_invalidate_gb_surface(struct vmw_private *dev_priv, &cmd->body.sid, NULL); } + +/** + * vmw_cmd_shader_define - Validate an SVGA_3D_CMD_SHADER_DEFINE + * command + * + * @dev_priv: Pointer to a device private struct. + * @sw_context: The software context being used for this batch. + * @header: Pointer to the command header in the command stream. + */ +static int vmw_cmd_shader_define(struct vmw_private *dev_priv, + struct vmw_sw_context *sw_context, + SVGA3dCmdHeader *header) +{ + struct vmw_shader_define_cmd { + SVGA3dCmdHeader header; + SVGA3dCmdDefineShader body; + } *cmd; + int ret; + size_t size; + + cmd = container_of(header, struct vmw_shader_define_cmd, + header); + + ret = vmw_cmd_res_check(dev_priv, sw_context, vmw_res_context, + user_context_converter, &cmd->body.cid, + NULL); + if (unlikely(ret != 0)) + return ret; + + if (unlikely(!dev_priv->has_mob)) + return 0; + + size = cmd->header.size - sizeof(cmd->body); + ret = vmw_compat_shader_add(sw_context->fp->shman, + cmd->body.shid, cmd + 1, + cmd->body.type, size, + sw_context->fp->tfile, + &sw_context->staged_shaders); + if (unlikely(ret != 0)) + return ret; + + return vmw_resource_relocation_add(&sw_context->res_relocations, + NULL, &cmd->header.id - + sw_context->buf_start); + + return 0; +} + +/** + * vmw_cmd_shader_destroy - Validate an SVGA_3D_CMD_SHADER_DESTROY + * command + * + * @dev_priv: Pointer to a device private struct. + * @sw_context: The software context being used for this batch. + * @header: Pointer to the command header in the command stream. + */ +static int vmw_cmd_shader_destroy(struct vmw_private *dev_priv, + struct vmw_sw_context *sw_context, + SVGA3dCmdHeader *header) +{ + struct vmw_shader_destroy_cmd { + SVGA3dCmdHeader header; + SVGA3dCmdDestroyShader body; + } *cmd; + int ret; + + cmd = container_of(header, struct vmw_shader_destroy_cmd, + header); + + ret = vmw_cmd_res_check(dev_priv, sw_context, vmw_res_context, + user_context_converter, &cmd->body.cid, + NULL); + if (unlikely(ret != 0)) + return ret; + + if (unlikely(!dev_priv->has_mob)) + return 0; + + ret = vmw_compat_shader_remove(sw_context->fp->shman, + cmd->body.shid, + cmd->body.type, + &sw_context->staged_shaders); + if (unlikely(ret != 0)) + return ret; + + return vmw_resource_relocation_add(&sw_context->res_relocations, + NULL, &cmd->header.id - + sw_context->buf_start); + + return 0; +} + /** * vmw_cmd_set_shader - Validate an SVGA_3D_CMD_SET_SHADER * command @@ -1509,10 +1708,18 @@ static int vmw_cmd_set_shader(struct vmw_private *dev_priv, if (dev_priv->has_mob) { struct vmw_ctx_bindinfo bi; struct vmw_resource_val_node *res_node; - - ret = vmw_cmd_res_check(dev_priv, sw_context, vmw_res_shader, - user_shader_converter, - &cmd->body.shid, &res_node); + u32 shid = cmd->body.shid; + + if (shid != SVGA3D_INVALID_ID) + (void) vmw_compat_shader_lookup(sw_context->fp->shman, + cmd->body.type, + &shid); + + ret = vmw_cmd_compat_res_check(dev_priv, sw_context, + vmw_res_shader, + user_shader_converter, + shid, + &cmd->body.shid, &res_node); if (unlikely(ret != 0)) return ret; @@ -1527,6 +1734,39 @@ static int vmw_cmd_set_shader(struct vmw_private *dev_priv, } /** + * vmw_cmd_set_shader_const - Validate an SVGA_3D_CMD_SET_SHADER_CONST + * command + * + * @dev_priv: Pointer to a device private struct. + * @sw_context: The software context being used for this batch. + * @header: Pointer to the command header in the command stream. + */ +static int vmw_cmd_set_shader_const(struct vmw_private *dev_priv, + struct vmw_sw_context *sw_context, + SVGA3dCmdHeader *header) +{ + struct vmw_set_shader_const_cmd { + SVGA3dCmdHeader header; + SVGA3dCmdSetShaderConst body; + } *cmd; + int ret; + + cmd = container_of(header, struct vmw_set_shader_const_cmd, + header); + + ret = vmw_cmd_res_check(dev_priv, sw_context, vmw_res_context, + user_context_converter, &cmd->body.cid, + NULL); + if (unlikely(ret != 0)) + return ret; + + if (dev_priv->has_mob) + header->id = SVGA_3D_CMD_SET_GB_SHADERCONSTS_INLINE; + + return 0; +} + +/** * vmw_cmd_bind_gb_shader - Validate an SVGA_3D_CMD_BIND_GB_SHADER * command * @@ -1595,7 +1835,7 @@ static int vmw_cmd_check_not_3d(struct vmw_private *dev_priv, return 0; } -static const struct vmw_cmd_entry const vmw_cmd_entries[SVGA_3D_CMD_MAX] = { +static const struct vmw_cmd_entry vmw_cmd_entries[SVGA_3D_CMD_MAX] = { VMW_CMD_DEF(SVGA_3D_CMD_SURFACE_DEFINE, &vmw_cmd_invalid, false, false, false), VMW_CMD_DEF(SVGA_3D_CMD_SURFACE_DESTROY, &vmw_cmd_invalid, @@ -1634,14 +1874,14 @@ static const struct vmw_cmd_entry const vmw_cmd_entries[SVGA_3D_CMD_MAX] = { true, false, false), VMW_CMD_DEF(SVGA_3D_CMD_PRESENT, &vmw_cmd_present_check, false, false, false), - VMW_CMD_DEF(SVGA_3D_CMD_SHADER_DEFINE, &vmw_cmd_cid_check, - true, true, false), - VMW_CMD_DEF(SVGA_3D_CMD_SHADER_DESTROY, &vmw_cmd_cid_check, - true, true, false), + VMW_CMD_DEF(SVGA_3D_CMD_SHADER_DEFINE, &vmw_cmd_shader_define, + true, false, false), + VMW_CMD_DEF(SVGA_3D_CMD_SHADER_DESTROY, &vmw_cmd_shader_destroy, + true, false, false), VMW_CMD_DEF(SVGA_3D_CMD_SET_SHADER, &vmw_cmd_set_shader, true, false, false), - VMW_CMD_DEF(SVGA_3D_CMD_SET_SHADER_CONST, &vmw_cmd_cid_check, - true, true, false), + VMW_CMD_DEF(SVGA_3D_CMD_SET_SHADER_CONST, &vmw_cmd_set_shader_const, + true, false, false), VMW_CMD_DEF(SVGA_3D_CMD_DRAW_PRIMITIVES, &vmw_cmd_draw, true, false, false), VMW_CMD_DEF(SVGA_3D_CMD_SETSCISSORRECT, &vmw_cmd_cid_check, @@ -1792,6 +2032,9 @@ static int vmw_cmd_check(struct vmw_private *dev_priv, goto out_invalid; entry = &vmw_cmd_entries[cmd_id]; + if (unlikely(!entry->func)) + goto out_invalid; + if (unlikely(!entry->user_allow && !sw_context->kernel)) goto out_privileged; @@ -2171,7 +2414,7 @@ int vmw_execbuf_process(struct drm_file *file_priv, } else sw_context->kernel = true; - sw_context->tfile = vmw_fpriv(file_priv)->tfile; + sw_context->fp = vmw_fpriv(file_priv); sw_context->cur_reloc = 0; sw_context->cur_val_buf = 0; sw_context->fence_flags = 0; @@ -2188,16 +2431,17 @@ int vmw_execbuf_process(struct drm_file *file_priv, goto out_unlock; sw_context->res_ht_initialized = true; } + INIT_LIST_HEAD(&sw_context->staged_shaders); INIT_LIST_HEAD(&resource_list); ret = vmw_cmd_check_all(dev_priv, sw_context, kernel_commands, command_size); if (unlikely(ret != 0)) - goto out_err; + goto out_err_nores; ret = vmw_resources_reserve(sw_context); if (unlikely(ret != 0)) - goto out_err; + goto out_err_nores; ret = ttm_eu_reserve_buffers(&ticket, &sw_context->validate_nodes); if (unlikely(ret != 0)) @@ -2225,6 +2469,12 @@ int vmw_execbuf_process(struct drm_file *file_priv, goto out_err; } + if (dev_priv->has_mob) { + ret = vmw_rebind_contexts(sw_context); + if (unlikely(ret != 0)) + goto out_unlock_binding; + } + cmd = vmw_fifo_reserve(dev_priv, command_size); if (unlikely(cmd == NULL)) { DRM_ERROR("Failed reserving fifo space for commands.\n"); @@ -2276,6 +2526,8 @@ int vmw_execbuf_process(struct drm_file *file_priv, } list_splice_init(&sw_context->resource_list, &resource_list); + vmw_compat_shaders_commit(sw_context->fp->shman, + &sw_context->staged_shaders); mutex_unlock(&dev_priv->cmdbuf_mutex); /* @@ -2289,10 +2541,11 @@ int vmw_execbuf_process(struct drm_file *file_priv, out_unlock_binding: mutex_unlock(&dev_priv->binding_mutex); out_err: - vmw_resource_relocations_free(&sw_context->res_relocations); - vmw_free_relocations(sw_context); ttm_eu_backoff_reservation(&ticket, &sw_context->validate_nodes); +out_err_nores: vmw_resource_list_unreserve(&sw_context->resource_list, true); + vmw_resource_relocations_free(&sw_context->res_relocations); + vmw_free_relocations(sw_context); vmw_clear_validations(sw_context); if (unlikely(dev_priv->pinned_bo != NULL && !dev_priv->query_cid_valid)) @@ -2301,6 +2554,8 @@ out_unlock: list_splice_init(&sw_context->resource_list, &resource_list); error_resource = sw_context->error_resource; sw_context->error_resource = NULL; + vmw_compat_shaders_revert(sw_context->fp->shman, + &sw_context->staged_shaders); mutex_unlock(&dev_priv->cmdbuf_mutex); /* diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c b/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c index 116c497..47b7094 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c @@ -29,12 +29,18 @@ #include <drm/vmwgfx_drm.h> #include "vmwgfx_kms.h" +struct svga_3d_compat_cap { + SVGA3dCapsRecordHeader header; + SVGA3dCapPair pairs[SVGA3D_DEVCAP_MAX]; +}; + int vmw_getparam_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv) { struct vmw_private *dev_priv = vmw_priv(dev); struct drm_vmw_getparam_arg *param = (struct drm_vmw_getparam_arg *)data; + struct vmw_fpriv *vmw_fp = vmw_fpriv(file_priv); switch (param->param) { case DRM_VMW_PARAM_NUM_STREAMS: @@ -60,6 +66,11 @@ int vmw_getparam_ioctl(struct drm_device *dev, void *data, __le32 __iomem *fifo_mem = dev_priv->mmio_virt; const struct vmw_fifo_state *fifo = &dev_priv->fifo; + if ((dev_priv->capabilities & SVGA_CAP_GBOBJECTS)) { + param->value = SVGA3D_HWVERSION_WS8_B1; + break; + } + param->value = ioread32(fifo_mem + ((fifo->capabilities & @@ -69,19 +80,31 @@ int vmw_getparam_ioctl(struct drm_device *dev, void *data, break; } case DRM_VMW_PARAM_MAX_SURF_MEMORY: - param->value = dev_priv->memory_size; + if ((dev_priv->capabilities & SVGA_CAP_GBOBJECTS) && + !vmw_fp->gb_aware) + param->value = dev_priv->max_mob_pages * PAGE_SIZE / 2; + else + param->value = dev_priv->memory_size; break; case DRM_VMW_PARAM_3D_CAPS_SIZE: - if (dev_priv->capabilities & SVGA_CAP_GBOBJECTS) - param->value = SVGA3D_DEVCAP_MAX; + if ((dev_priv->capabilities & SVGA_CAP_GBOBJECTS) && + vmw_fp->gb_aware) + param->value = SVGA3D_DEVCAP_MAX * sizeof(uint32_t); + else if (dev_priv->capabilities & SVGA_CAP_GBOBJECTS) + param->value = sizeof(struct svga_3d_compat_cap) + + sizeof(uint32_t); else param->value = (SVGA_FIFO_3D_CAPS_LAST - - SVGA_FIFO_3D_CAPS + 1); - param->value *= sizeof(uint32_t); + SVGA_FIFO_3D_CAPS + 1) * + sizeof(uint32_t); break; case DRM_VMW_PARAM_MAX_MOB_MEMORY: + vmw_fp->gb_aware = true; param->value = dev_priv->max_mob_pages * PAGE_SIZE; break; + case DRM_VMW_PARAM_MAX_MOB_SIZE: + param->value = dev_priv->max_mob_size; + break; default: DRM_ERROR("Illegal vmwgfx get param request: %d\n", param->param); @@ -91,6 +114,38 @@ int vmw_getparam_ioctl(struct drm_device *dev, void *data, return 0; } +static int vmw_fill_compat_cap(struct vmw_private *dev_priv, void *bounce, + size_t size) +{ + struct svga_3d_compat_cap *compat_cap = + (struct svga_3d_compat_cap *) bounce; + unsigned int i; + size_t pair_offset = offsetof(struct svga_3d_compat_cap, pairs); + unsigned int max_size; + + if (size < pair_offset) + return -EINVAL; + + max_size = (size - pair_offset) / sizeof(SVGA3dCapPair); + + if (max_size > SVGA3D_DEVCAP_MAX) + max_size = SVGA3D_DEVCAP_MAX; + + compat_cap->header.length = + (pair_offset + max_size * sizeof(SVGA3dCapPair)) / sizeof(u32); + compat_cap->header.type = SVGA3DCAPS_RECORD_DEVCAPS; + + mutex_lock(&dev_priv->hw_mutex); + for (i = 0; i < max_size; ++i) { + vmw_write(dev_priv, SVGA_REG_DEV_CAP, i); + compat_cap->pairs[i][0] = i; + compat_cap->pairs[i][1] = vmw_read(dev_priv, SVGA_REG_DEV_CAP); + } + mutex_unlock(&dev_priv->hw_mutex); + + return 0; +} + int vmw_get_cap_3d_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv) @@ -104,41 +159,49 @@ int vmw_get_cap_3d_ioctl(struct drm_device *dev, void *data, void *bounce; int ret; bool gb_objects = !!(dev_priv->capabilities & SVGA_CAP_GBOBJECTS); + struct vmw_fpriv *vmw_fp = vmw_fpriv(file_priv); if (unlikely(arg->pad64 != 0)) { DRM_ERROR("Illegal GET_3D_CAP argument.\n"); return -EINVAL; } - if (gb_objects) - size = SVGA3D_DEVCAP_MAX; + if (gb_objects && vmw_fp->gb_aware) + size = SVGA3D_DEVCAP_MAX * sizeof(uint32_t); + else if (gb_objects) + size = sizeof(struct svga_3d_compat_cap) + sizeof(uint32_t); else - size = (SVGA_FIFO_3D_CAPS_LAST - SVGA_FIFO_3D_CAPS + 1); - - size *= sizeof(uint32_t); + size = (SVGA_FIFO_3D_CAPS_LAST - SVGA_FIFO_3D_CAPS + 1) * + sizeof(uint32_t); if (arg->max_size < size) size = arg->max_size; - bounce = vmalloc(size); + bounce = vzalloc(size); if (unlikely(bounce == NULL)) { DRM_ERROR("Failed to allocate bounce buffer for 3D caps.\n"); return -ENOMEM; } - if (gb_objects) { - int i; + if (gb_objects && vmw_fp->gb_aware) { + int i, num; uint32_t *bounce32 = (uint32_t *) bounce; + num = size / sizeof(uint32_t); + if (num > SVGA3D_DEVCAP_MAX) + num = SVGA3D_DEVCAP_MAX; + mutex_lock(&dev_priv->hw_mutex); - for (i = 0; i < SVGA3D_DEVCAP_MAX; ++i) { + for (i = 0; i < num; ++i) { vmw_write(dev_priv, SVGA_REG_DEV_CAP, i); *bounce32++ = vmw_read(dev_priv, SVGA_REG_DEV_CAP); } mutex_unlock(&dev_priv->hw_mutex); - + } else if (gb_objects) { + ret = vmw_fill_compat_cap(dev_priv, bounce, size); + if (unlikely(ret != 0)) + goto out_err; } else { - fifo_mem = dev_priv->mmio_virt; memcpy_fromio(bounce, &fifo_mem[SVGA_FIFO_3D_CAPS], size); } @@ -146,6 +209,7 @@ int vmw_get_cap_3d_ioctl(struct drm_device *dev, void *data, ret = copy_to_user(buffer, bounce, size); if (ret) ret = -EFAULT; +out_err: vfree(bounce); if (unlikely(ret != 0)) diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c b/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c index 4910e7b..04a64b8 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_mob.c @@ -134,6 +134,7 @@ static int vmw_setup_otable_base(struct vmw_private *dev_priv, cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd)); if (unlikely(cmd == NULL)) { DRM_ERROR("Failed reserving FIFO space for OTable setup.\n"); + ret = -ENOMEM; goto out_no_fifo; } @@ -187,18 +188,20 @@ static void vmw_takedown_otable_base(struct vmw_private *dev_priv, bo = otable->page_table->pt_bo; cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd)); - if (unlikely(cmd == NULL)) - DRM_ERROR("Failed reserving FIFO space for OTable setup.\n"); - - memset(cmd, 0, sizeof(*cmd)); - cmd->header.id = SVGA_3D_CMD_SET_OTABLE_BASE; - cmd->header.size = sizeof(cmd->body); - cmd->body.type = type; - cmd->body.baseAddress = 0; - cmd->body.sizeInBytes = 0; - cmd->body.validSizeInBytes = 0; - cmd->body.ptDepth = SVGA3D_MOBFMT_INVALID; - vmw_fifo_commit(dev_priv, sizeof(*cmd)); + if (unlikely(cmd == NULL)) { + DRM_ERROR("Failed reserving FIFO space for OTable " + "takedown.\n"); + } else { + memset(cmd, 0, sizeof(*cmd)); + cmd->header.id = SVGA_3D_CMD_SET_OTABLE_BASE; + cmd->header.size = sizeof(cmd->body); + cmd->body.type = type; + cmd->body.baseAddress = 0; + cmd->body.sizeInBytes = 0; + cmd->body.validSizeInBytes = 0; + cmd->body.ptDepth = SVGA3D_MOBFMT_INVALID; + vmw_fifo_commit(dev_priv, sizeof(*cmd)); + } if (bo) { int ret; @@ -561,11 +564,12 @@ void vmw_mob_unbind(struct vmw_private *dev_priv, if (unlikely(cmd == NULL)) { DRM_ERROR("Failed reserving FIFO space for Memory " "Object unbinding.\n"); + } else { + cmd->header.id = SVGA_3D_CMD_DESTROY_GB_MOB; + cmd->header.size = sizeof(cmd->body); + cmd->body.mobid = mob->id; + vmw_fifo_commit(dev_priv, sizeof(*cmd)); } - cmd->header.id = SVGA_3D_CMD_DESTROY_GB_MOB; - cmd->header.size = sizeof(cmd->body); - cmd->body.mobid = mob->id; - vmw_fifo_commit(dev_priv, sizeof(*cmd)); if (bo) { vmw_fence_single_bo(bo, NULL); ttm_bo_unreserve(bo); diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c index 6fdd82d..9757b57 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c @@ -88,6 +88,11 @@ struct vmw_resource *vmw_resource_reference(struct vmw_resource *res) return res; } +struct vmw_resource * +vmw_resource_reference_unless_doomed(struct vmw_resource *res) +{ + return kref_get_unless_zero(&res->kref) ? res : NULL; +} /** * vmw_resource_release_id - release a resource id to the id manager. @@ -136,8 +141,12 @@ static void vmw_resource_release(struct kref *kref) vmw_dmabuf_unreference(&res->backup); } - if (likely(res->hw_destroy != NULL)) + if (likely(res->hw_destroy != NULL)) { res->hw_destroy(res); + mutex_lock(&dev_priv->binding_mutex); + vmw_context_binding_res_list_kill(&res->binding_head); + mutex_unlock(&dev_priv->binding_mutex); + } id = res->id; if (res->res_free != NULL) @@ -418,8 +427,7 @@ int vmw_dmabuf_init(struct vmw_private *dev_priv, INIT_LIST_HEAD(&vmw_bo->res_list); ret = ttm_bo_init(bdev, &vmw_bo->base, size, - (user) ? ttm_bo_type_device : - ttm_bo_type_kernel, placement, + ttm_bo_type_device, placement, 0, interruptible, NULL, acc_size, NULL, bo_free); return ret; diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c b/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c index 1457ec4b..ee38565 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_shader.c @@ -29,6 +29,8 @@ #include "vmwgfx_resource_priv.h" #include "ttm/ttm_placement.h" +#define VMW_COMPAT_SHADER_HT_ORDER 12 + struct vmw_shader { struct vmw_resource res; SVGA3dShaderType type; @@ -40,6 +42,50 @@ struct vmw_user_shader { struct vmw_shader shader; }; +/** + * enum vmw_compat_shader_state - Staging state for compat shaders + */ +enum vmw_compat_shader_state { + VMW_COMPAT_COMMITED, + VMW_COMPAT_ADD, + VMW_COMPAT_DEL +}; + +/** + * struct vmw_compat_shader - Metadata for compat shaders. + * + * @handle: The TTM handle of the guest backed shader. + * @tfile: The struct ttm_object_file the guest backed shader is registered + * with. + * @hash: Hash item for lookup. + * @head: List head for staging lists or the compat shader manager list. + * @state: Staging state. + * + * The structure is protected by the cmdbuf lock. + */ +struct vmw_compat_shader { + u32 handle; + struct ttm_object_file *tfile; + struct drm_hash_item hash; + struct list_head head; + enum vmw_compat_shader_state state; +}; + +/** + * struct vmw_compat_shader_manager - Compat shader manager. + * + * @shaders: Hash table containing staged and commited compat shaders + * @list: List of commited shaders. + * @dev_priv: Pointer to a device private structure. + * + * @shaders and @list are protected by the cmdbuf mutex for now. + */ +struct vmw_compat_shader_manager { + struct drm_open_hash shaders; + struct list_head list; + struct vmw_private *dev_priv; +}; + static void vmw_user_shader_free(struct vmw_resource *res); static struct vmw_resource * vmw_user_shader_base_to_res(struct ttm_base_object *base); @@ -258,7 +304,7 @@ static int vmw_gb_shader_destroy(struct vmw_resource *res) return 0; mutex_lock(&dev_priv->binding_mutex); - vmw_context_binding_res_list_kill(&res->binding_head); + vmw_context_binding_res_list_scrub(&res->binding_head); cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd)); if (unlikely(cmd == NULL)) { @@ -325,13 +371,81 @@ int vmw_shader_destroy_ioctl(struct drm_device *dev, void *data, TTM_REF_USAGE); } +static int vmw_shader_alloc(struct vmw_private *dev_priv, + struct vmw_dma_buffer *buffer, + size_t shader_size, + size_t offset, + SVGA3dShaderType shader_type, + struct ttm_object_file *tfile, + u32 *handle) +{ + struct vmw_user_shader *ushader; + struct vmw_resource *res, *tmp; + int ret; + + /* + * Approximate idr memory usage with 128 bytes. It will be limited + * by maximum number_of shaders anyway. + */ + if (unlikely(vmw_user_shader_size == 0)) + vmw_user_shader_size = + ttm_round_pot(sizeof(struct vmw_user_shader)) + 128; + + ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), + vmw_user_shader_size, + false, true); + if (unlikely(ret != 0)) { + if (ret != -ERESTARTSYS) + DRM_ERROR("Out of graphics memory for shader " + "creation.\n"); + goto out; + } + + ushader = kzalloc(sizeof(*ushader), GFP_KERNEL); + if (unlikely(ushader == NULL)) { + ttm_mem_global_free(vmw_mem_glob(dev_priv), + vmw_user_shader_size); + ret = -ENOMEM; + goto out; + } + + res = &ushader->shader.res; + ushader->base.shareable = false; + ushader->base.tfile = NULL; + + /* + * From here on, the destructor takes over resource freeing. + */ + + ret = vmw_gb_shader_init(dev_priv, res, shader_size, + offset, shader_type, buffer, + vmw_user_shader_free); + if (unlikely(ret != 0)) + goto out; + + tmp = vmw_resource_reference(res); + ret = ttm_base_object_init(tfile, &ushader->base, false, + VMW_RES_SHADER, + &vmw_user_shader_base_release, NULL); + + if (unlikely(ret != 0)) { + vmw_resource_unreference(&tmp); + goto out_err; + } + + if (handle) + *handle = ushader->base.hash.key; +out_err: + vmw_resource_unreference(&res); +out: + return ret; +} + + int vmw_shader_define_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv) { struct vmw_private *dev_priv = vmw_priv(dev); - struct vmw_user_shader *ushader; - struct vmw_resource *res; - struct vmw_resource *tmp; struct drm_vmw_shader_create_arg *arg = (struct drm_vmw_shader_create_arg *)data; struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile; @@ -373,69 +487,326 @@ int vmw_shader_define_ioctl(struct drm_device *dev, void *data, goto out_bad_arg; } - /* - * Approximate idr memory usage with 128 bytes. It will be limited - * by maximum number_of shaders anyway. - */ + ret = ttm_read_lock(&vmaster->lock, true); + if (unlikely(ret != 0)) + goto out_bad_arg; - if (unlikely(vmw_user_shader_size == 0)) - vmw_user_shader_size = ttm_round_pot(sizeof(*ushader)) - + 128; + ret = vmw_shader_alloc(dev_priv, buffer, arg->size, arg->offset, + shader_type, tfile, &arg->shader_handle); - ret = ttm_read_lock(&vmaster->lock, true); + ttm_read_unlock(&vmaster->lock); +out_bad_arg: + vmw_dmabuf_unreference(&buffer); + return ret; +} + +/** + * vmw_compat_shader_lookup - Look up a compat shader + * + * @man: Pointer to the compat shader manager. + * @shader_type: The shader type, that combined with the user_key identifies + * the shader. + * @user_key: On entry, this should be a pointer to the user_key. + * On successful exit, it will contain the guest-backed shader's TTM handle. + * + * Returns 0 on success. Non-zero on failure, in which case the value pointed + * to by @user_key is unmodified. + */ +int vmw_compat_shader_lookup(struct vmw_compat_shader_manager *man, + SVGA3dShaderType shader_type, + u32 *user_key) +{ + struct drm_hash_item *hash; + int ret; + unsigned long key = *user_key | (shader_type << 24); + + ret = drm_ht_find_item(&man->shaders, key, &hash); if (unlikely(ret != 0)) return ret; - ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), - vmw_user_shader_size, - false, true); - if (unlikely(ret != 0)) { - if (ret != -ERESTARTSYS) - DRM_ERROR("Out of graphics memory for shader" - " creation.\n"); - goto out_unlock; + *user_key = drm_hash_entry(hash, struct vmw_compat_shader, + hash)->handle; + + return 0; +} + +/** + * vmw_compat_shader_free - Free a compat shader. + * + * @man: Pointer to the compat shader manager. + * @entry: Pointer to a struct vmw_compat_shader. + * + * Frees a struct vmw_compat_shder entry and drops its reference to the + * guest backed shader. + */ +static void vmw_compat_shader_free(struct vmw_compat_shader_manager *man, + struct vmw_compat_shader *entry) +{ + list_del(&entry->head); + WARN_ON(drm_ht_remove_item(&man->shaders, &entry->hash)); + WARN_ON(ttm_ref_object_base_unref(entry->tfile, entry->handle, + TTM_REF_USAGE)); + kfree(entry); +} + +/** + * vmw_compat_shaders_commit - Commit a list of compat shader actions. + * + * @man: Pointer to the compat shader manager. + * @list: Caller's list of compat shader actions. + * + * This function commits a list of compat shader additions or removals. + * It is typically called when the execbuf ioctl call triggering these + * actions has commited the fifo contents to the device. + */ +void vmw_compat_shaders_commit(struct vmw_compat_shader_manager *man, + struct list_head *list) +{ + struct vmw_compat_shader *entry, *next; + + list_for_each_entry_safe(entry, next, list, head) { + list_del(&entry->head); + switch (entry->state) { + case VMW_COMPAT_ADD: + entry->state = VMW_COMPAT_COMMITED; + list_add_tail(&entry->head, &man->list); + break; + case VMW_COMPAT_DEL: + ttm_ref_object_base_unref(entry->tfile, entry->handle, + TTM_REF_USAGE); + kfree(entry); + break; + default: + BUG(); + break; + } } +} - ushader = kzalloc(sizeof(*ushader), GFP_KERNEL); - if (unlikely(ushader == NULL)) { - ttm_mem_global_free(vmw_mem_glob(dev_priv), - vmw_user_shader_size); - ret = -ENOMEM; - goto out_unlock; +/** + * vmw_compat_shaders_revert - Revert a list of compat shader actions + * + * @man: Pointer to the compat shader manager. + * @list: Caller's list of compat shader actions. + * + * This function reverts a list of compat shader additions or removals. + * It is typically called when the execbuf ioctl call triggering these + * actions failed for some reason, and the command stream was never + * submitted. + */ +void vmw_compat_shaders_revert(struct vmw_compat_shader_manager *man, + struct list_head *list) +{ + struct vmw_compat_shader *entry, *next; + int ret; + + list_for_each_entry_safe(entry, next, list, head) { + switch (entry->state) { + case VMW_COMPAT_ADD: + vmw_compat_shader_free(man, entry); + break; + case VMW_COMPAT_DEL: + ret = drm_ht_insert_item(&man->shaders, &entry->hash); + list_del(&entry->head); + list_add_tail(&entry->head, &man->list); + entry->state = VMW_COMPAT_COMMITED; + break; + default: + BUG(); + break; + } } +} - res = &ushader->shader.res; - ushader->base.shareable = false; - ushader->base.tfile = NULL; +/** + * vmw_compat_shader_remove - Stage a compat shader for removal. + * + * @man: Pointer to the compat shader manager + * @user_key: The key that is used to identify the shader. The key is + * unique to the shader type. + * @shader_type: Shader type. + * @list: Caller's list of staged shader actions. + * + * This function stages a compat shader for removal and removes the key from + * the shader manager's hash table. If the shader was previously only staged + * for addition it is completely removed (But the execbuf code may keep a + * reference if it was bound to a context between addition and removal). If + * it was previously commited to the manager, it is staged for removal. + */ +int vmw_compat_shader_remove(struct vmw_compat_shader_manager *man, + u32 user_key, SVGA3dShaderType shader_type, + struct list_head *list) +{ + struct vmw_compat_shader *entry; + struct drm_hash_item *hash; + int ret; - /* - * From here on, the destructor takes over resource freeing. - */ + ret = drm_ht_find_item(&man->shaders, user_key | (shader_type << 24), + &hash); + if (likely(ret != 0)) + return -EINVAL; - ret = vmw_gb_shader_init(dev_priv, res, arg->size, - arg->offset, shader_type, buffer, - vmw_user_shader_free); + entry = drm_hash_entry(hash, struct vmw_compat_shader, hash); + + switch (entry->state) { + case VMW_COMPAT_ADD: + vmw_compat_shader_free(man, entry); + break; + case VMW_COMPAT_COMMITED: + (void) drm_ht_remove_item(&man->shaders, &entry->hash); + list_del(&entry->head); + entry->state = VMW_COMPAT_DEL; + list_add_tail(&entry->head, list); + break; + default: + BUG(); + break; + } + + return 0; +} + +/** + * vmw_compat_shader_add - Create a compat shader and add the + * key to the manager + * + * @man: Pointer to the compat shader manager + * @user_key: The key that is used to identify the shader. The key is + * unique to the shader type. + * @bytecode: Pointer to the bytecode of the shader. + * @shader_type: Shader type. + * @tfile: Pointer to a struct ttm_object_file that the guest-backed shader is + * to be created with. + * @list: Caller's list of staged shader actions. + * + * Note that only the key is added to the shader manager's hash table. + * The shader is not yet added to the shader manager's list of shaders. + */ +int vmw_compat_shader_add(struct vmw_compat_shader_manager *man, + u32 user_key, const void *bytecode, + SVGA3dShaderType shader_type, + size_t size, + struct ttm_object_file *tfile, + struct list_head *list) +{ + struct vmw_dma_buffer *buf; + struct ttm_bo_kmap_obj map; + bool is_iomem; + struct vmw_compat_shader *compat; + u32 handle; + int ret; + + if (user_key > ((1 << 24) - 1) || (unsigned) shader_type > 16) + return -EINVAL; + + /* Allocate and pin a DMA buffer */ + buf = kzalloc(sizeof(*buf), GFP_KERNEL); + if (unlikely(buf == NULL)) + return -ENOMEM; + + ret = vmw_dmabuf_init(man->dev_priv, buf, size, &vmw_sys_ne_placement, + true, vmw_dmabuf_bo_free); if (unlikely(ret != 0)) - goto out_unlock; + goto out; - tmp = vmw_resource_reference(res); - ret = ttm_base_object_init(tfile, &ushader->base, false, - VMW_RES_SHADER, - &vmw_user_shader_base_release, NULL); + ret = ttm_bo_reserve(&buf->base, false, true, false, NULL); + if (unlikely(ret != 0)) + goto no_reserve; + /* Map and copy shader bytecode. */ + ret = ttm_bo_kmap(&buf->base, 0, PAGE_ALIGN(size) >> PAGE_SHIFT, + &map); if (unlikely(ret != 0)) { - vmw_resource_unreference(&tmp); - goto out_err; + ttm_bo_unreserve(&buf->base); + goto no_reserve; } - arg->shader_handle = ushader->base.hash.key; -out_err: - vmw_resource_unreference(&res); -out_unlock: - ttm_read_unlock(&vmaster->lock); -out_bad_arg: - vmw_dmabuf_unreference(&buffer); + memcpy(ttm_kmap_obj_virtual(&map, &is_iomem), bytecode, size); + WARN_ON(is_iomem); + + ttm_bo_kunmap(&map); + ret = ttm_bo_validate(&buf->base, &vmw_sys_placement, false, true); + WARN_ON(ret != 0); + ttm_bo_unreserve(&buf->base); + + /* Create a guest-backed shader container backed by the dma buffer */ + ret = vmw_shader_alloc(man->dev_priv, buf, size, 0, shader_type, + tfile, &handle); + vmw_dmabuf_unreference(&buf); + if (unlikely(ret != 0)) + goto no_reserve; + /* + * Create a compat shader structure and stage it for insertion + * in the manager + */ + compat = kzalloc(sizeof(*compat), GFP_KERNEL); + if (compat == NULL) + goto no_compat; + + compat->hash.key = user_key | (shader_type << 24); + ret = drm_ht_insert_item(&man->shaders, &compat->hash); + if (unlikely(ret != 0)) + goto out_invalid_key; + + compat->state = VMW_COMPAT_ADD; + compat->handle = handle; + compat->tfile = tfile; + list_add_tail(&compat->head, list); + return 0; + +out_invalid_key: + kfree(compat); +no_compat: + ttm_ref_object_base_unref(tfile, handle, TTM_REF_USAGE); +no_reserve: +out: return ret; +} + +/** + * vmw_compat_shader_man_create - Create a compat shader manager + * + * @dev_priv: Pointer to a device private structure. + * + * Typically done at file open time. If successful returns a pointer to a + * compat shader manager. Otherwise returns an error pointer. + */ +struct vmw_compat_shader_manager * +vmw_compat_shader_man_create(struct vmw_private *dev_priv) +{ + struct vmw_compat_shader_manager *man; + int ret; + + man = kzalloc(sizeof(*man), GFP_KERNEL); + if (man == NULL) + return ERR_PTR(-ENOMEM); + + man->dev_priv = dev_priv; + INIT_LIST_HEAD(&man->list); + ret = drm_ht_create(&man->shaders, VMW_COMPAT_SHADER_HT_ORDER); + if (ret == 0) + return man; + + kfree(man); + return ERR_PTR(ret); +} + +/** + * vmw_compat_shader_man_destroy - Destroy a compat shader manager + * + * @man: Pointer to the shader manager to destroy. + * + * Typically done at file close time. + */ +void vmw_compat_shader_man_destroy(struct vmw_compat_shader_manager *man) +{ + struct vmw_compat_shader *entry, *next; + + mutex_lock(&man->dev_priv->cmdbuf_mutex); + list_for_each_entry_safe(entry, next, &man->list, head) + vmw_compat_shader_free(man, entry); + mutex_unlock(&man->dev_priv->cmdbuf_mutex); + kfree(man); } diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c b/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c index 979da1c..82468d9 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c @@ -908,8 +908,8 @@ int vmw_surface_reference_ioctl(struct drm_device *dev, void *data, rep->size_addr; if (user_sizes) - ret = copy_to_user(user_sizes, srf->sizes, - srf->num_sizes * sizeof(*srf->sizes)); + ret = copy_to_user(user_sizes, &srf->base_size, + sizeof(srf->base_size)); if (unlikely(ret != 0)) { DRM_ERROR("copy_to_user failed %p %u\n", user_sizes, srf->num_sizes); @@ -1111,7 +1111,7 @@ static int vmw_gb_surface_destroy(struct vmw_resource *res) return 0; mutex_lock(&dev_priv->binding_mutex); - vmw_context_binding_res_list_kill(&res->binding_head); + vmw_context_binding_res_list_scrub(&res->binding_head); cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd)); if (unlikely(cmd == NULL)) { |