1 /* 2 * Copyright 2014-2015 Broadcom 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 7 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8 * and/or sell copies of the Software, and to permit persons to whom the 9 * Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice (including the next 12 * paragraph) shall be included in all copies or substantial portions of the 13 * Software. 14 * 15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS 21 * IN THE SOFTWARE. 22 */ 23 24 #ifndef _VC4_DRM_H_ 25 #define _VC4_DRM_H_ 26 27 #include "drm.h" 28 29 #if defined(__cplusplus) 30 extern "C" { 31 #endif 32 33 #define DRM_VC4_SUBMIT_CL 0x00 34 #define DRM_VC4_WAIT_SEQNO 0x01 35 #define DRM_VC4_WAIT_BO 0x02 36 #define DRM_VC4_CREATE_BO 0x03 37 #define DRM_VC4_MMAP_BO 0x04 38 #define DRM_VC4_CREATE_SHADER_BO 0x05 39 #define DRM_VC4_GET_HANG_STATE 0x06 40 #define DRM_VC4_GET_PARAM 0x07 41 #define DRM_VC4_SET_TILING 0x08 42 #define DRM_VC4_GET_TILING 0x09 43 #define DRM_VC4_LABEL_BO 0x0a 44 #define DRM_VC4_GEM_MADVISE 0x0b 45 #define DRM_VC4_PERFMON_CREATE 0x0c 46 #define DRM_VC4_PERFMON_DESTROY 0x0d 47 #define DRM_VC4_PERFMON_GET_VALUES 0x0e 48 49 #define DRM_IOCTL_VC4_SUBMIT_CL DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_SUBMIT_CL, struct drm_vc4_submit_cl) 50 #define DRM_IOCTL_VC4_WAIT_SEQNO DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_WAIT_SEQNO, struct drm_vc4_wait_seqno) 51 #define DRM_IOCTL_VC4_WAIT_BO DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_WAIT_BO, struct drm_vc4_wait_bo) 52 #define DRM_IOCTL_VC4_CREATE_BO DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_CREATE_BO, struct drm_vc4_create_bo) 53 #define DRM_IOCTL_VC4_MMAP_BO DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_MMAP_BO, struct drm_vc4_mmap_bo) 54 #define DRM_IOCTL_VC4_CREATE_SHADER_BO DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_CREATE_SHADER_BO, struct drm_vc4_create_shader_bo) 55 #define DRM_IOCTL_VC4_GET_HANG_STATE DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_GET_HANG_STATE, struct drm_vc4_get_hang_state) 56 #define DRM_IOCTL_VC4_GET_PARAM DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_GET_PARAM, struct drm_vc4_get_param) 57 #define DRM_IOCTL_VC4_SET_TILING DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_SET_TILING, struct drm_vc4_set_tiling) 58 #define DRM_IOCTL_VC4_GET_TILING DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_GET_TILING, struct drm_vc4_get_tiling) 59 #define DRM_IOCTL_VC4_LABEL_BO DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_LABEL_BO, struct drm_vc4_label_bo) 60 #define DRM_IOCTL_VC4_GEM_MADVISE DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_GEM_MADVISE, struct drm_vc4_gem_madvise) 61 #define DRM_IOCTL_VC4_PERFMON_CREATE DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_PERFMON_CREATE, struct drm_vc4_perfmon_create) 62 #define DRM_IOCTL_VC4_PERFMON_DESTROY DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_PERFMON_DESTROY, struct drm_vc4_perfmon_destroy) 63 #define DRM_IOCTL_VC4_PERFMON_GET_VALUES DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_PERFMON_GET_VALUES, struct drm_vc4_perfmon_get_values) 64 65 struct drm_vc4_submit_rcl_surface { 66 __u32 hindex; /* Handle index, or ~0 if not present. */ 67 __u32 offset; /* Offset to start of buffer. */ 68 /* 69 * Bits for either render config (color_write) or load/store packet. 70 * Bits should all be 0 for MSAA load/stores. 71 */ 72 __u16 bits; 73 74 #define VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES (1 << 0) 75 __u16 flags; 76 }; 77 78 /** 79 * struct drm_vc4_submit_cl - ioctl argument for submitting commands to the 3D 80 * engine. 81 * 82 * Drivers typically use GPU BOs to store batchbuffers / command lists and 83 * their associated state. However, because the VC4 lacks an MMU, we have to 84 * do validation of memory accesses by the GPU commands. If we were to store 85 * our commands in BOs, we'd need to do uncached readback from them to do the 86 * validation process, which is too expensive. Instead, userspace accumulates 87 * commands and associated state in plain memory, then the kernel copies the 88 * data to its own address space, and then validates and stores it in a GPU 89 * BO. 90 */ 91 struct drm_vc4_submit_cl { 92 /* Pointer to the binner command list. 93 * 94 * This is the first set of commands executed, which runs the 95 * coordinate shader to determine where primitives land on the screen, 96 * then writes out the state updates and draw calls necessary per tile 97 * to the tile allocation BO. 98 */ 99 __u64 bin_cl; 100 101 /* Pointer to the shader records. 102 * 103 * Shader records are the structures read by the hardware that contain 104 * pointers to uniforms, shaders, and vertex attributes. The 105 * reference to the shader record has enough information to determine 106 * how many pointers are necessary (fixed number for shaders/uniforms, 107 * and an attribute count), so those BO indices into bo_handles are 108 * just stored as __u32s before each shader record passed in. 109 */ 110 __u64 shader_rec; 111 112 /* Pointer to uniform data and texture handles for the textures 113 * referenced by the shader. 114 * 115 * For each shader state record, there is a set of uniform data in the 116 * order referenced by the record (FS, VS, then CS). Each set of 117 * uniform data has a __u32 index into bo_handles per texture 118 * sample operation, in the order the QPU_W_TMUn_S writes appear in 119 * the program. Following the texture BO handle indices is the actual 120 * uniform data. 121 * 122 * The individual uniform state blocks don't have sizes passed in, 123 * because the kernel has to determine the sizes anyway during shader 124 * code validation. 125 */ 126 __u64 uniforms; 127 __u64 bo_handles; 128 129 /* Size in bytes of the binner command list. */ 130 __u32 bin_cl_size; 131 /* Size in bytes of the set of shader records. */ 132 __u32 shader_rec_size; 133 /* Number of shader records. 134 * 135 * This could just be computed from the contents of shader_records and 136 * the address bits of references to them from the bin CL, but it 137 * keeps the kernel from having to resize some allocations it makes. 138 */ 139 __u32 shader_rec_count; 140 /* Size in bytes of the uniform state. */ 141 __u32 uniforms_size; 142 143 /* Number of BO handles passed in (size is that times 4). */ 144 __u32 bo_handle_count; 145 146 /* RCL setup: */ 147 __u16 width; 148 __u16 height; 149 __u8 min_x_tile; 150 __u8 min_y_tile; 151 __u8 max_x_tile; 152 __u8 max_y_tile; 153 struct drm_vc4_submit_rcl_surface color_read; 154 struct drm_vc4_submit_rcl_surface color_write; 155 struct drm_vc4_submit_rcl_surface zs_read; 156 struct drm_vc4_submit_rcl_surface zs_write; 157 struct drm_vc4_submit_rcl_surface msaa_color_write; 158 struct drm_vc4_submit_rcl_surface msaa_zs_write; 159 __u32 clear_color[2]; 160 __u32 clear_z; 161 __u8 clear_s; 162 163 __u32 pad:24; 164 165 #define VC4_SUBMIT_CL_USE_CLEAR_COLOR (1 << 0) 166 /* By default, the kernel gets to choose the order that the tiles are 167 * rendered in. If this is set, then the tiles will be rendered in a 168 * raster order, with the right-to-left vs left-to-right and 169 * top-to-bottom vs bottom-to-top dictated by 170 * VC4_SUBMIT_CL_RCL_ORDER_INCREASING_*. This allows overlapping 171 * blits to be implemented using the 3D engine. 172 */ 173 #define VC4_SUBMIT_CL_FIXED_RCL_ORDER (1 << 1) 174 #define VC4_SUBMIT_CL_RCL_ORDER_INCREASING_X (1 << 2) 175 #define VC4_SUBMIT_CL_RCL_ORDER_INCREASING_Y (1 << 3) 176 __u32 flags; 177 178 /* Returned value of the seqno of this render job (for the 179 * wait ioctl). 180 */ 181 __u64 seqno; 182 183 /* ID of the perfmon to attach to this job. 0 means no perfmon. */ 184 __u32 perfmonid; 185 186 /* Unused field to align this struct on 64 bits. Must be set to 0. 187 * If one ever needs to add an u32 field to this struct, this field 188 * can be used. 189 */ 190 __u32 pad2; 191 }; 192 193 /** 194 * struct drm_vc4_wait_seqno - ioctl argument for waiting for 195 * DRM_VC4_SUBMIT_CL completion using its returned seqno. 196 * 197 * timeout_ns is the timeout in nanoseconds, where "0" means "don't 198 * block, just return the status." 199 */ 200 struct drm_vc4_wait_seqno { 201 __u64 seqno; 202 __u64 timeout_ns; 203 }; 204 205 /** 206 * struct drm_vc4_wait_bo - ioctl argument for waiting for 207 * completion of the last DRM_VC4_SUBMIT_CL on a BO. 208 * 209 * This is useful for cases where multiple processes might be 210 * rendering to a BO and you want to wait for all rendering to be 211 * completed. 212 */ 213 struct drm_vc4_wait_bo { 214 __u32 handle; 215 __u32 pad; 216 __u64 timeout_ns; 217 }; 218 219 /** 220 * struct drm_vc4_create_bo - ioctl argument for creating VC4 BOs. 221 * 222 * There are currently no values for the flags argument, but it may be 223 * used in a future extension. 224 */ 225 struct drm_vc4_create_bo { 226 __u32 size; 227 __u32 flags; 228 /** Returned GEM handle for the BO. */ 229 __u32 handle; 230 __u32 pad; 231 }; 232 233 /** 234 * struct drm_vc4_mmap_bo - ioctl argument for mapping VC4 BOs. 235 * 236 * This doesn't actually perform an mmap. Instead, it returns the 237 * offset you need to use in an mmap on the DRM device node. This 238 * means that tools like valgrind end up knowing about the mapped 239 * memory. 240 * 241 * There are currently no values for the flags argument, but it may be 242 * used in a future extension. 243 */ 244 struct drm_vc4_mmap_bo { 245 /** Handle for the object being mapped. */ 246 __u32 handle; 247 __u32 flags; 248 /** offset into the drm node to use for subsequent mmap call. */ 249 __u64 offset; 250 }; 251 252 /** 253 * struct drm_vc4_create_shader_bo - ioctl argument for creating VC4 254 * shader BOs. 255 * 256 * Since allowing a shader to be overwritten while it's also being 257 * executed from would allow privlege escalation, shaders must be 258 * created using this ioctl, and they can't be mmapped later. 259 */ 260 struct drm_vc4_create_shader_bo { 261 /* Size of the data argument. */ 262 __u32 size; 263 /* Flags, currently must be 0. */ 264 __u32 flags; 265 266 /* Pointer to the data. */ 267 __u64 data; 268 269 /** Returned GEM handle for the BO. */ 270 __u32 handle; 271 /* Pad, must be 0. */ 272 __u32 pad; 273 }; 274 275 struct drm_vc4_get_hang_state_bo { 276 __u32 handle; 277 __u32 paddr; 278 __u32 size; 279 __u32 pad; 280 }; 281 282 /** 283 * struct drm_vc4_hang_state - ioctl argument for collecting state 284 * from a GPU hang for analysis. 285 */ 286 struct drm_vc4_get_hang_state { 287 /** Pointer to array of struct drm_vc4_get_hang_state_bo. */ 288 __u64 bo; 289 /** 290 * On input, the size of the bo array. Output is the number 291 * of bos to be returned. 292 */ 293 __u32 bo_count; 294 295 __u32 start_bin, start_render; 296 297 __u32 ct0ca, ct0ea; 298 __u32 ct1ca, ct1ea; 299 __u32 ct0cs, ct1cs; 300 __u32 ct0ra0, ct1ra0; 301 302 __u32 bpca, bpcs; 303 __u32 bpoa, bpos; 304 305 __u32 vpmbase; 306 307 __u32 dbge; 308 __u32 fdbgo; 309 __u32 fdbgb; 310 __u32 fdbgr; 311 __u32 fdbgs; 312 __u32 errstat; 313 314 /* Pad that we may save more registers into in the future. */ 315 __u32 pad[16]; 316 }; 317 318 #define DRM_VC4_PARAM_V3D_IDENT0 0 319 #define DRM_VC4_PARAM_V3D_IDENT1 1 320 #define DRM_VC4_PARAM_V3D_IDENT2 2 321 #define DRM_VC4_PARAM_SUPPORTS_BRANCHES 3 322 #define DRM_VC4_PARAM_SUPPORTS_ETC1 4 323 #define DRM_VC4_PARAM_SUPPORTS_THREADED_FS 5 324 #define DRM_VC4_PARAM_SUPPORTS_FIXED_RCL_ORDER 6 325 #define DRM_VC4_PARAM_SUPPORTS_MADVISE 7 326 #define DRM_VC4_PARAM_SUPPORTS_PERFMON 8 327 328 struct drm_vc4_get_param { 329 __u32 param; 330 __u32 pad; 331 __u64 value; 332 }; 333 334 struct drm_vc4_get_tiling { 335 __u32 handle; 336 __u32 flags; 337 __u64 modifier; 338 }; 339 340 struct drm_vc4_set_tiling { 341 __u32 handle; 342 __u32 flags; 343 __u64 modifier; 344 }; 345 346 /** 347 * struct drm_vc4_label_bo - Attach a name to a BO for debug purposes. 348 */ 349 struct drm_vc4_label_bo { 350 __u32 handle; 351 __u32 len; 352 __u64 name; 353 }; 354 355 /* 356 * States prefixed with '__' are internal states and cannot be passed to the 357 * DRM_IOCTL_VC4_GEM_MADVISE ioctl. 358 */ 359 #define VC4_MADV_WILLNEED 0 360 #define VC4_MADV_DONTNEED 1 361 #define __VC4_MADV_PURGED 2 362 #define __VC4_MADV_NOTSUPP 3 363 364 struct drm_vc4_gem_madvise { 365 __u32 handle; 366 __u32 madv; 367 __u32 retained; 368 __u32 pad; 369 }; 370 371 enum { 372 VC4_PERFCNT_FEP_VALID_PRIMS_NO_RENDER, 373 VC4_PERFCNT_FEP_VALID_PRIMS_RENDER, 374 VC4_PERFCNT_FEP_CLIPPED_QUADS, 375 VC4_PERFCNT_FEP_VALID_QUADS, 376 VC4_PERFCNT_TLB_QUADS_NOT_PASSING_STENCIL, 377 VC4_PERFCNT_TLB_QUADS_NOT_PASSING_Z_AND_STENCIL, 378 VC4_PERFCNT_TLB_QUADS_PASSING_Z_AND_STENCIL, 379 VC4_PERFCNT_TLB_QUADS_ZERO_COVERAGE, 380 VC4_PERFCNT_TLB_QUADS_NON_ZERO_COVERAGE, 381 VC4_PERFCNT_TLB_QUADS_WRITTEN_TO_COLOR_BUF, 382 VC4_PERFCNT_PLB_PRIMS_OUTSIDE_VIEWPORT, 383 VC4_PERFCNT_PLB_PRIMS_NEED_CLIPPING, 384 VC4_PERFCNT_PSE_PRIMS_REVERSED, 385 VC4_PERFCNT_QPU_TOTAL_IDLE_CYCLES, 386 VC4_PERFCNT_QPU_TOTAL_CLK_CYCLES_VERTEX_COORD_SHADING, 387 VC4_PERFCNT_QPU_TOTAL_CLK_CYCLES_FRAGMENT_SHADING, 388 VC4_PERFCNT_QPU_TOTAL_CLK_CYCLES_EXEC_VALID_INST, 389 VC4_PERFCNT_QPU_TOTAL_CLK_CYCLES_WAITING_TMUS, 390 VC4_PERFCNT_QPU_TOTAL_CLK_CYCLES_WAITING_SCOREBOARD, 391 VC4_PERFCNT_QPU_TOTAL_CLK_CYCLES_WAITING_VARYINGS, 392 VC4_PERFCNT_QPU_TOTAL_INST_CACHE_HIT, 393 VC4_PERFCNT_QPU_TOTAL_INST_CACHE_MISS, 394 VC4_PERFCNT_QPU_TOTAL_UNIFORM_CACHE_HIT, 395 VC4_PERFCNT_QPU_TOTAL_UNIFORM_CACHE_MISS, 396 VC4_PERFCNT_TMU_TOTAL_TEXT_QUADS_PROCESSED, 397 VC4_PERFCNT_TMU_TOTAL_TEXT_CACHE_MISS, 398 VC4_PERFCNT_VPM_TOTAL_CLK_CYCLES_VDW_STALLED, 399 VC4_PERFCNT_VPM_TOTAL_CLK_CYCLES_VCD_STALLED, 400 VC4_PERFCNT_L2C_TOTAL_L2_CACHE_HIT, 401 VC4_PERFCNT_L2C_TOTAL_L2_CACHE_MISS, 402 VC4_PERFCNT_NUM_EVENTS, 403 }; 404 405 #define DRM_VC4_MAX_PERF_COUNTERS 16 406 407 struct drm_vc4_perfmon_create { 408 __u32 id; 409 __u32 ncounters; 410 __u8 events[DRM_VC4_MAX_PERF_COUNTERS]; 411 }; 412 413 struct drm_vc4_perfmon_destroy { 414 __u32 id; 415 }; 416 417 /* 418 * Returns the values of the performance counters tracked by this 419 * perfmon (as an array of ncounters u64 values). 420 * 421 * No implicit synchronization is performed, so the user has to 422 * guarantee that any jobs using this perfmon have already been 423 * completed (probably by blocking on the seqno returned by the 424 * last exec that used the perfmon). 425 */ 426 struct drm_vc4_perfmon_get_values { 427 __u32 id; 428 __u64 values_ptr; 429 }; 430 431 #if defined(__cplusplus) 432 } 433 #endif 434 435 #endif /* _VC4_DRM_H_ */ 436