1/* 2 * Copyright 2022 Advanced Micro Devices, Inc. 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 shall be included in 12 * all copies or substantial portions of the Software. 13 * 14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR 18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 20 * OTHER DEALINGS IN THE SOFTWARE. 21 * 22 */ 23 24#ifndef __MES_API_DEF_H__ 25#define __MES_API_DEF_H__ 26 27#pragma pack(push, 4) 28 29#define MES_API_VERSION 1 30 31/* Driver submits one API(cmd) as a single Frame and this command size is same 32 * for all API to ease the debugging and parsing of ring buffer. 33 */ 34enum { API_FRAME_SIZE_IN_DWORDS = 64 }; 35 36/* To avoid command in scheduler context to be overwritten whenenver mutilple 37 * interrupts come in, this creates another queue. 38 */ 39enum { API_NUMBER_OF_COMMAND_MAX = 32 }; 40 41enum MES_API_TYPE { 42 MES_API_TYPE_SCHEDULER = 1, 43 MES_API_TYPE_MAX 44}; 45 46enum MES_SCH_API_OPCODE { 47 MES_SCH_API_SET_HW_RSRC = 0, 48 MES_SCH_API_SET_SCHEDULING_CONFIG = 1, /* agreegated db, quantums, etc */ 49 MES_SCH_API_ADD_QUEUE = 2, 50 MES_SCH_API_REMOVE_QUEUE = 3, 51 MES_SCH_API_PERFORM_YIELD = 4, 52 MES_SCH_API_SET_GANG_PRIORITY_LEVEL = 5, 53 MES_SCH_API_SUSPEND = 6, 54 MES_SCH_API_RESUME = 7, 55 MES_SCH_API_RESET = 8, 56 MES_SCH_API_SET_LOG_BUFFER = 9, 57 MES_SCH_API_CHANGE_GANG_PRORITY = 10, 58 MES_SCH_API_QUERY_SCHEDULER_STATUS = 11, 59 MES_SCH_API_PROGRAM_GDS = 12, 60 MES_SCH_API_SET_DEBUG_VMID = 13, 61 MES_SCH_API_MISC = 14, 62 MES_SCH_API_UPDATE_ROOT_PAGE_TABLE = 15, 63 MES_SCH_API_AMD_LOG = 16, 64 MES_SCH_API_MAX = 0xFF 65}; 66 67union MES_API_HEADER { 68 struct { 69 uint32_t type : 4; /* 0 - Invalid; 1 - Scheduling; 2 - TBD */ 70 uint32_t opcode : 8; 71 uint32_t dwsize : 8; /* including header */ 72 uint32_t reserved : 12; 73 }; 74 75 uint32_t u32All; 76}; 77 78enum MES_AMD_PRIORITY_LEVEL { 79 AMD_PRIORITY_LEVEL_LOW = 0, 80 AMD_PRIORITY_LEVEL_NORMAL = 1, 81 AMD_PRIORITY_LEVEL_MEDIUM = 2, 82 AMD_PRIORITY_LEVEL_HIGH = 3, 83 AMD_PRIORITY_LEVEL_REALTIME = 4, 84 AMD_PRIORITY_NUM_LEVELS 85}; 86 87enum MES_QUEUE_TYPE { 88 MES_QUEUE_TYPE_GFX, 89 MES_QUEUE_TYPE_COMPUTE, 90 MES_QUEUE_TYPE_SDMA, 91 MES_QUEUE_TYPE_MAX, 92}; 93 94struct MES_API_STATUS { 95 uint64_t api_completion_fence_addr; 96 uint64_t api_completion_fence_value; 97}; 98 99enum { MAX_COMPUTE_PIPES = 8 }; 100enum { MAX_GFX_PIPES = 2 }; 101enum { MAX_SDMA_PIPES = 2 }; 102 103enum { MAX_COMPUTE_HQD_PER_PIPE = 8 }; 104enum { MAX_GFX_HQD_PER_PIPE = 8 }; 105enum { MAX_SDMA_HQD_PER_PIPE = 10 }; 106enum { MAX_SDMA_HQD_PER_PIPE_11_0 = 8 }; 107 108enum { MAX_QUEUES_IN_A_GANG = 8 }; 109 110enum VM_HUB_TYPE { 111 VM_HUB_TYPE_GC = 0, 112 VM_HUB_TYPE_MM = 1, 113 VM_HUB_TYPE_MAX, 114}; 115 116enum { VMID_INVALID = 0xffff }; 117 118enum { MAX_VMID_GCHUB = 16 }; 119enum { MAX_VMID_MMHUB = 16 }; 120 121enum SET_DEBUG_VMID_OPERATIONS { 122 DEBUG_VMID_OP_PROGRAM = 0, 123 DEBUG_VMID_OP_ALLOCATE = 1, 124 DEBUG_VMID_OP_RELEASE = 2 125}; 126 127enum MES_LOG_OPERATION { 128 MES_LOG_OPERATION_CONTEXT_STATE_CHANGE = 0, 129 MES_LOG_OPERATION_QUEUE_NEW_WORK = 1, 130 MES_LOG_OPERATION_QUEUE_UNWAIT_SYNC_OBJECT = 2, 131 MES_LOG_OPERATION_QUEUE_NO_MORE_WORK = 3, 132 MES_LOG_OPERATION_QUEUE_WAIT_SYNC_OBJECT = 4, 133 MES_LOG_OPERATION_QUEUE_INVALID = 0xF, 134}; 135 136enum MES_LOG_CONTEXT_STATE { 137 MES_LOG_CONTEXT_STATE_IDLE = 0, 138 MES_LOG_CONTEXT_STATE_RUNNING = 1, 139 MES_LOG_CONTEXT_STATE_READY = 2, 140 MES_LOG_CONTEXT_STATE_READY_STANDBY = 3, 141 MES_LOG_CONTEXT_STATE_INVALID = 0xF, 142}; 143 144struct MES_LOG_CONTEXT_STATE_CHANGE { 145 void *h_context; 146 enum MES_LOG_CONTEXT_STATE new_context_state; 147}; 148 149struct MES_LOG_QUEUE_NEW_WORK { 150 uint64_t h_queue; 151 uint64_t reserved; 152}; 153 154struct MES_LOG_QUEUE_UNWAIT_SYNC_OBJECT { 155 uint64_t h_queue; 156 uint64_t h_sync_object; 157}; 158 159struct MES_LOG_QUEUE_NO_MORE_WORK { 160 uint64_t h_queue; 161 uint64_t reserved; 162}; 163 164struct MES_LOG_QUEUE_WAIT_SYNC_OBJECT { 165 uint64_t h_queue; 166 uint64_t h_sync_object; 167}; 168 169struct MES_LOG_ENTRY_HEADER { 170 uint32_t first_free_entry_index; 171 uint32_t wraparound_count; 172 uint64_t number_of_entries; 173 uint64_t reserved[2]; 174}; 175 176struct MES_LOG_ENTRY_DATA { 177 uint64_t gpu_time_stamp; 178 uint32_t operation_type; /* operation_type is of MES_LOG_OPERATION type */ 179 uint32_t reserved_operation_type_bits; 180 union { 181 struct MES_LOG_CONTEXT_STATE_CHANGE context_state_change; 182 struct MES_LOG_QUEUE_NEW_WORK queue_new_work; 183 struct MES_LOG_QUEUE_UNWAIT_SYNC_OBJECT queue_unwait_sync_object; 184 struct MES_LOG_QUEUE_NO_MORE_WORK queue_no_more_work; 185 struct MES_LOG_QUEUE_WAIT_SYNC_OBJECT queue_wait_sync_object; 186 uint64_t all[2]; 187 }; 188}; 189 190struct MES_LOG_BUFFER { 191 struct MES_LOG_ENTRY_HEADER header; 192 struct MES_LOG_ENTRY_DATA entries[1]; 193}; 194 195enum MES_SWIP_TO_HWIP_DEF { 196 MES_MAX_HWIP_SEGMENT = 8, 197}; 198 199union MESAPI_SET_HW_RESOURCES { 200 struct { 201 union MES_API_HEADER header; 202 uint32_t vmid_mask_mmhub; 203 uint32_t vmid_mask_gfxhub; 204 uint32_t gds_size; 205 uint32_t paging_vmid; 206 uint32_t compute_hqd_mask[MAX_COMPUTE_PIPES]; 207 uint32_t gfx_hqd_mask[MAX_GFX_PIPES]; 208 uint32_t sdma_hqd_mask[MAX_SDMA_PIPES]; 209 uint32_t aggregated_doorbells[AMD_PRIORITY_NUM_LEVELS]; 210 uint64_t g_sch_ctx_gpu_mc_ptr; 211 uint64_t query_status_fence_gpu_mc_ptr; 212 uint32_t gc_base[MES_MAX_HWIP_SEGMENT]; 213 uint32_t mmhub_base[MES_MAX_HWIP_SEGMENT]; 214 uint32_t osssys_base[MES_MAX_HWIP_SEGMENT]; 215 struct MES_API_STATUS api_status; 216 union { 217 struct { 218 uint32_t disable_reset : 1; 219 uint32_t use_different_vmid_compute : 1; 220 uint32_t disable_mes_log : 1; 221 uint32_t apply_mmhub_pgvm_invalidate_ack_loss_wa : 1; 222 uint32_t apply_grbm_remote_register_dummy_read_wa : 1; 223 uint32_t second_gfx_pipe_enabled : 1; 224 uint32_t enable_level_process_quantum_check : 1; 225 uint32_t legacy_sch_mode : 1; 226 uint32_t disable_add_queue_wptr_mc_addr : 1; 227 uint32_t enable_mes_event_int_logging : 1; 228 uint32_t enable_reg_active_poll : 1; 229 uint32_t reserved : 21; 230 }; 231 uint32_t uint32_t_all; 232 }; 233 uint32_t oversubscription_timer; 234 uint64_t doorbell_info; 235 uint64_t event_intr_history_gpu_mc_ptr; 236 }; 237 238 uint32_t max_dwords_in_api[API_FRAME_SIZE_IN_DWORDS]; 239}; 240 241union MESAPI__ADD_QUEUE { 242 struct { 243 union MES_API_HEADER header; 244 uint32_t process_id; 245 uint64_t page_table_base_addr; 246 uint64_t process_va_start; 247 uint64_t process_va_end; 248 uint64_t process_quantum; 249 uint64_t process_context_addr; 250 uint64_t gang_quantum; 251 uint64_t gang_context_addr; 252 uint32_t inprocess_gang_priority; 253 enum MES_AMD_PRIORITY_LEVEL gang_global_priority_level; 254 uint32_t doorbell_offset; 255 uint64_t mqd_addr; 256 uint64_t wptr_addr; 257 uint64_t h_context; 258 uint64_t h_queue; 259 enum MES_QUEUE_TYPE queue_type; 260 uint32_t gds_base; 261 uint32_t gds_size; 262 uint32_t gws_base; 263 uint32_t gws_size; 264 uint32_t oa_mask; 265 uint64_t trap_handler_addr; 266 uint32_t vm_context_cntl; 267 268 struct { 269 uint32_t paging : 1; 270 uint32_t debug_vmid : 4; 271 uint32_t program_gds : 1; 272 uint32_t is_gang_suspended : 1; 273 uint32_t is_tmz_queue : 1; 274 uint32_t map_kiq_utility_queue : 1; 275 uint32_t is_kfd_process : 1; 276 uint32_t trap_en : 1; 277 uint32_t is_aql_queue : 1; 278 uint32_t skip_process_ctx_clear : 1; 279 uint32_t map_legacy_kq : 1; 280 uint32_t exclusively_scheduled : 1; 281 uint32_t reserved : 17; 282 }; 283 struct MES_API_STATUS api_status; 284 uint64_t tma_addr; 285 }; 286 287 uint32_t max_dwords_in_api[API_FRAME_SIZE_IN_DWORDS]; 288}; 289 290union MESAPI__REMOVE_QUEUE { 291 struct { 292 union MES_API_HEADER header; 293 uint32_t doorbell_offset; 294 uint64_t gang_context_addr; 295 296 struct { 297 uint32_t unmap_legacy_gfx_queue : 1; 298 uint32_t unmap_kiq_utility_queue : 1; 299 uint32_t preempt_legacy_gfx_queue : 1; 300 uint32_t unmap_legacy_queue : 1; 301 uint32_t reserved : 28; 302 }; 303 struct MES_API_STATUS api_status; 304 305 uint32_t pipe_id; 306 uint32_t queue_id; 307 308 uint64_t tf_addr; 309 uint32_t tf_data; 310 311 enum MES_QUEUE_TYPE queue_type; 312 }; 313 314 uint32_t max_dwords_in_api[API_FRAME_SIZE_IN_DWORDS]; 315}; 316 317union MESAPI__SET_SCHEDULING_CONFIG { 318 struct { 319 union MES_API_HEADER header; 320 /* Grace period when preempting another priority band for this 321 * priority band. The value for idle priority band is ignored, 322 * as it never preempts other bands. 323 */ 324 uint64_t grace_period_other_levels[AMD_PRIORITY_NUM_LEVELS]; 325 /* Default quantum for scheduling across processes within 326 * a priority band. 327 */ 328 uint64_t process_quantum_for_level[AMD_PRIORITY_NUM_LEVELS]; 329 /* Default grace period for processes that preempt each other 330 * within a priority band. 331 */ 332 uint64_t process_grace_period_same_level[AMD_PRIORITY_NUM_LEVELS]; 333 /* For normal level this field specifies the target GPU 334 * percentage in situations when it's starved by the high level. 335 * Valid values are between 0 and 50, with the default being 10. 336 */ 337 uint32_t normal_yield_percent; 338 struct MES_API_STATUS api_status; 339 }; 340 341 uint32_t max_dwords_in_api[API_FRAME_SIZE_IN_DWORDS]; 342}; 343 344union MESAPI__PERFORM_YIELD { 345 struct { 346 union MES_API_HEADER header; 347 uint32_t dummy; 348 struct MES_API_STATUS api_status; 349 }; 350 351 uint32_t max_dwords_in_api[API_FRAME_SIZE_IN_DWORDS]; 352}; 353 354union MESAPI__CHANGE_GANG_PRIORITY_LEVEL { 355 struct { 356 union MES_API_HEADER header; 357 uint32_t inprocess_gang_priority; 358 enum MES_AMD_PRIORITY_LEVEL gang_global_priority_level; 359 uint64_t gang_quantum; 360 uint64_t gang_context_addr; 361 struct MES_API_STATUS api_status; 362 }; 363 364 uint32_t max_dwords_in_api[API_FRAME_SIZE_IN_DWORDS]; 365}; 366 367union MESAPI__SUSPEND { 368 struct { 369 union MES_API_HEADER header; 370 /* false - suspend all gangs; true - specific gang */ 371 struct { 372 uint32_t suspend_all_gangs : 1; 373 uint32_t reserved : 31; 374 }; 375 /* gang_context_addr is valid only if suspend_all = false */ 376 uint64_t gang_context_addr; 377 378 uint64_t suspend_fence_addr; 379 uint32_t suspend_fence_value; 380 381 struct MES_API_STATUS api_status; 382 }; 383 384 uint32_t max_dwords_in_api[API_FRAME_SIZE_IN_DWORDS]; 385}; 386 387union MESAPI__RESUME { 388 struct { 389 union MES_API_HEADER header; 390 /* false - resume all gangs; true - specified gang */ 391 struct { 392 uint32_t resume_all_gangs : 1; 393 uint32_t reserved : 31; 394 }; 395 /* valid only if resume_all_gangs = false */ 396 uint64_t gang_context_addr; 397 398 struct MES_API_STATUS api_status; 399 }; 400 401 uint32_t max_dwords_in_api[API_FRAME_SIZE_IN_DWORDS]; 402}; 403 404union MESAPI__RESET { 405 struct { 406 union MES_API_HEADER header; 407 408 struct { 409 /* Only reset the queue given by doorbell_offset (not entire gang) */ 410 uint32_t reset_queue_only : 1; 411 /* Hang detection first then reset any queues that are hung */ 412 uint32_t hang_detect_then_reset : 1; 413 /* Only do hang detection (no reset) */ 414 uint32_t hang_detect_only : 1; 415 /* Rest HP and LP kernel queues not managed by MES */ 416 uint32_t reset_legacy_gfx : 1; 417 uint32_t reserved : 28; 418 }; 419 420 uint64_t gang_context_addr; 421 422 /* valid only if reset_queue_only = true */ 423 uint32_t doorbell_offset; 424 425 /* valid only if hang_detect_then_reset = true */ 426 uint64_t doorbell_offset_addr; 427 enum MES_QUEUE_TYPE queue_type; 428 429 /* valid only if reset_legacy_gfx = true */ 430 uint32_t pipe_id_lp; 431 uint32_t queue_id_lp; 432 uint32_t vmid_id_lp; 433 uint64_t mqd_mc_addr_lp; 434 uint32_t doorbell_offset_lp; 435 uint64_t wptr_addr_lp; 436 437 uint32_t pipe_id_hp; 438 uint32_t queue_id_hp; 439 uint32_t vmid_id_hp; 440 uint64_t mqd_mc_addr_hp; 441 uint32_t doorbell_offset_hp; 442 uint64_t wptr_addr_hp; 443 444 struct MES_API_STATUS api_status; 445 }; 446 447 uint32_t max_dwords_in_api[API_FRAME_SIZE_IN_DWORDS]; 448}; 449 450union MESAPI__SET_LOGGING_BUFFER { 451 struct { 452 union MES_API_HEADER header; 453 /* There are separate log buffers for each queue type */ 454 enum MES_QUEUE_TYPE log_type; 455 /* Log buffer GPU Address */ 456 uint64_t logging_buffer_addr; 457 /* number of entries in the log buffer */ 458 uint32_t number_of_entries; 459 /* Entry index at which CPU interrupt needs to be signalled */ 460 uint32_t interrupt_entry; 461 462 struct MES_API_STATUS api_status; 463 }; 464 465 uint32_t max_dwords_in_api[API_FRAME_SIZE_IN_DWORDS]; 466}; 467 468union MESAPI__QUERY_MES_STATUS { 469 struct { 470 union MES_API_HEADER header; 471 bool mes_healthy; /* 0 - not healthy, 1 - healthy */ 472 struct MES_API_STATUS api_status; 473 }; 474 475 uint32_t max_dwords_in_api[API_FRAME_SIZE_IN_DWORDS]; 476}; 477 478union MESAPI__PROGRAM_GDS { 479 struct { 480 union MES_API_HEADER header; 481 uint64_t process_context_addr; 482 uint32_t gds_base; 483 uint32_t gds_size; 484 uint32_t gws_base; 485 uint32_t gws_size; 486 uint32_t oa_mask; 487 struct MES_API_STATUS api_status; 488 }; 489 490 uint32_t max_dwords_in_api[API_FRAME_SIZE_IN_DWORDS]; 491}; 492 493union MESAPI__SET_DEBUG_VMID { 494 struct { 495 union MES_API_HEADER header; 496 struct MES_API_STATUS api_status; 497 union { 498 struct { 499 uint32_t use_gds : 1; 500 uint32_t operation : 2; 501 uint32_t reserved : 29; 502 } flags; 503 uint32_t u32All; 504 }; 505 uint32_t reserved; 506 uint32_t debug_vmid; 507 uint64_t process_context_addr; 508 uint64_t page_table_base_addr; 509 uint64_t process_va_start; 510 uint64_t process_va_end; 511 uint32_t gds_base; 512 uint32_t gds_size; 513 uint32_t gws_base; 514 uint32_t gws_size; 515 uint32_t oa_mask; 516 517 /* output addr of the acquired vmid value */ 518 uint64_t output_addr; 519 }; 520 521 uint32_t max_dwords_in_api[API_FRAME_SIZE_IN_DWORDS]; 522}; 523 524enum MESAPI_MISC_OPCODE { 525 MESAPI_MISC__WRITE_REG, 526 MESAPI_MISC__INV_GART, 527 MESAPI_MISC__QUERY_STATUS, 528 MESAPI_MISC__READ_REG, 529 MESAPI_MISC__WAIT_REG_MEM, 530 MESAPI_MISC__SET_SHADER_DEBUGGER, 531 MESAPI_MISC__MAX, 532}; 533 534enum { MISC_DATA_MAX_SIZE_IN_DWORDS = 20 }; 535 536struct WRITE_REG { 537 uint32_t reg_offset; 538 uint32_t reg_value; 539}; 540 541struct READ_REG { 542 uint32_t reg_offset; 543 uint64_t buffer_addr; 544}; 545 546enum WRM_OPERATION { 547 WRM_OPERATION__WAIT_REG_MEM, 548 WRM_OPERATION__WR_WAIT_WR_REG, 549 WRM_OPERATION__MAX, 550}; 551 552struct WAIT_REG_MEM { 553 enum WRM_OPERATION op; 554 uint32_t reference; 555 uint32_t mask; 556 uint32_t reg_offset1; 557 uint32_t reg_offset2; 558}; 559 560struct INV_GART { 561 uint64_t inv_range_va_start; 562 uint64_t inv_range_size; 563}; 564 565struct QUERY_STATUS { 566 uint32_t context_id; 567}; 568 569struct SET_SHADER_DEBUGGER { 570 uint64_t process_context_addr; 571 union { 572 struct { 573 uint32_t single_memop : 1; /* SQ_DEBUG.single_memop */ 574 uint32_t single_alu_op : 1; /* SQ_DEBUG.single_alu_op */ 575 uint32_t reserved : 29; 576 uint32_t process_ctx_flush : 1; 577 }; 578 uint32_t u32all; 579 } flags; 580 uint32_t spi_gdbg_per_vmid_cntl; 581 uint32_t tcp_watch_cntl[4]; /* TCP_WATCHx_CNTL */ 582 uint32_t trap_en; 583}; 584 585union MESAPI__MISC { 586 struct { 587 union MES_API_HEADER header; 588 enum MESAPI_MISC_OPCODE opcode; 589 struct MES_API_STATUS api_status; 590 591 union { 592 struct WRITE_REG write_reg; 593 struct INV_GART inv_gart; 594 struct QUERY_STATUS query_status; 595 struct READ_REG read_reg; 596 struct WAIT_REG_MEM wait_reg_mem; 597 struct SET_SHADER_DEBUGGER set_shader_debugger; 598 enum MES_AMD_PRIORITY_LEVEL queue_sch_level; 599 600 uint32_t data[MISC_DATA_MAX_SIZE_IN_DWORDS]; 601 }; 602 }; 603 604 uint32_t max_dwords_in_api[API_FRAME_SIZE_IN_DWORDS]; 605}; 606 607union MESAPI__UPDATE_ROOT_PAGE_TABLE { 608 struct { 609 union MES_API_HEADER header; 610 uint64_t page_table_base_addr; 611 uint64_t process_context_addr; 612 struct MES_API_STATUS api_status; 613 }; 614 615 uint32_t max_dwords_in_api[API_FRAME_SIZE_IN_DWORDS]; 616}; 617 618union MESAPI_AMD_LOG { 619 struct { 620 union MES_API_HEADER header; 621 uint64_t p_buffer_memory; 622 uint64_t p_buffer_size_used; 623 struct MES_API_STATUS api_status; 624 }; 625 626 uint32_t max_dwords_in_api[API_FRAME_SIZE_IN_DWORDS]; 627}; 628 629#pragma pack(pop) 630#endif 631