| 1 | |
| 2 | * Copyright 2018 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 _DISCOVERY_H_
|
| 25 | #define _DISCOVERY_H_
|
| 26 |
|
| 27 | #define PSP_HEADER_SIZE 256
|
| 28 | #define BINARY_SIGNATURE 0x28211407
|
| 29 | #define DISCOVERY_TABLE_SIGNATURE 0x53445049
|
| 30 | #define GC_TABLE_ID 0x4347
|
| 31 | #define HARVEST_TABLE_SIGNATURE 0x56524148
|
| 32 | #define VCN_INFO_TABLE_ID 0x004E4356
|
| 33 | #define MALL_INFO_TABLE_ID 0x4C4C414D
|
| 34 | #define NPS_INFO_TABLE_ID 0x0053504E
|
| 35 |
|
| 36 | typedef enum {
|
| 37 | IP_DISCOVERY = 0,
|
| 38 | GC,
|
| 39 | HARVEST_INFO,
|
| 40 | VCN_INFO,
|
| 41 | MALL_INFO,
|
| 42 | NPS_INFO,
|
| 43 | TOTAL_TABLES = 6
|
| 44 | } table;
|
| 45 |
|
| 46 | #pragma pack(1)
|
| 47 |
|
| 48 | typedef struct table_info
|
| 49 | {
|
| 50 | uint16_t offset;
|
| 51 | uint16_t checksum;
|
| 52 | uint16_t size;
|
| 53 | uint16_t padding;
|
| 54 | } table_info;
|
| 55 |
|
| 56 | typedef struct binary_header
|
| 57 | {
|
| 58 |
|
| 59 | uint32_t binary_signature;
|
| 60 | uint16_t version_major;
|
| 61 | uint16_t version_minor;
|
| 62 | uint16_t binary_checksum;
|
| 63 | uint16_t binary_size;
|
| 64 | table_info table_list[TOTAL_TABLES];
|
| 65 | } binary_header;
|
| 66 |
|
| 67 | typedef struct binary_header_v2
|
| 68 | {
|
| 69 |
|
| 70 | uint32_t binary_signature;
|
| 71 | uint16_t version_major;
|
| 72 | uint16_t version_minor;
|
| 73 | uint16_t binary_checksum;
|
| 74 | uint16_t binary_size;
|
| 75 | uint16_t num_tables;
|
| 76 | uint16_t padding;
|
| 77 | table_info table_list[] __counted_by(num_tables);
|
| 78 | } binary_header_v2;
|
| 79 |
|
| 80 | typedef struct die_info
|
| 81 | {
|
| 82 | uint16_t die_id;
|
| 83 | uint16_t die_offset;
|
| 84 | } die_info;
|
| 85 |
|
| 86 |
|
| 87 | typedef struct ip_discovery_header
|
| 88 | {
|
| 89 | uint32_t signature;
|
| 90 | uint16_t version;
|
| 91 | uint16_t size;
|
| 92 | uint32_t id;
|
| 93 | uint16_t num_dies;
|
| 94 | die_info die_info[16];
|
| 95 | union {
|
| 96 | uint16_t padding[1];
|
| 97 | struct {
|
| 98 | uint8_t base_addr_64_bit : 1;
|
| 99 | uint8_t reserved : 7;
|
| 100 | uint8_t reserved2;
|
| 101 | };
|
| 102 | };
|
| 103 | } ip_discovery_header;
|
| 104 |
|
| 105 | typedef struct ip
|
| 106 | {
|
| 107 | uint16_t hw_id;
|
| 108 | uint8_t number_instance;
|
| 109 | uint8_t num_base_address;
|
| 110 | uint8_t major;
|
| 111 | uint8_t minor;
|
| 112 | uint8_t revision;
|
| 113 | #if defined(__BIG_ENDIAN)
|
| 114 | uint8_t reserved : 4;
|
| 115 | uint8_t harvest : 4;
|
| 116 | #else
|
| 117 | uint8_t harvest : 4;
|
| 118 | uint8_t reserved : 4;
|
| 119 | #endif
|
| 120 | uint32_t base_address[];
|
| 121 | } ip;
|
| 122 |
|
| 123 | typedef struct ip_v3
|
| 124 | {
|
| 125 | uint16_t hw_id;
|
| 126 | uint8_t instance_number;
|
| 127 | uint8_t num_base_address;
|
| 128 | uint8_t major;
|
| 129 | uint8_t minor;
|
| 130 | uint8_t revision;
|
| 131 | #if defined(__BIG_ENDIAN)
|
| 132 | uint8_t variant : 4;
|
| 133 | uint8_t sub_revision : 4;
|
| 134 | #else
|
| 135 | uint8_t sub_revision : 4;
|
| 136 | uint8_t variant : 4;
|
| 137 | #endif
|
| 138 | uint32_t base_address[];
|
| 139 | } ip_v3;
|
| 140 |
|
| 141 | typedef struct ip_v4 {
|
| 142 | uint16_t hw_id;
|
| 143 | uint8_t instance_number;
|
| 144 | uint8_t num_base_address;
|
| 145 | uint8_t major;
|
| 146 | uint8_t minor;
|
| 147 | uint8_t revision;
|
| 148 | #if defined(LITTLEENDIAN_CPU)
|
| 149 | uint8_t sub_revision : 4;
|
| 150 | uint8_t variant : 4;
|
| 151 | #elif defined(BIGENDIAN_CPU)
|
| 152 | uint8_t variant : 4;
|
| 153 | uint8_t sub_revision : 4;
|
| 154 | #endif
|
| 155 | union {
|
| 156 | DECLARE_FLEX_ARRAY(uint32_t, base_address);
|
| 157 | DECLARE_FLEX_ARRAY(uint64_t, base_address_64);
|
| 158 | } __packed;
|
| 159 | } ip_v4;
|
| 160 |
|
| 161 | typedef struct die_header
|
| 162 | {
|
| 163 | uint16_t die_id;
|
| 164 | uint16_t num_ips;
|
| 165 | } die_header;
|
| 166 |
|
| 167 | typedef struct ip_structure
|
| 168 | {
|
| 169 | ip_discovery_header* header;
|
| 170 | struct die
|
| 171 | {
|
| 172 | die_header *die_header;
|
| 173 | union
|
| 174 | {
|
| 175 | ip *ip_list;
|
| 176 | ip_v3 *ip_v3_list;
|
| 177 | ip_v4 *ip_v4_list;
|
| 178 | };
|
| 179 | } die;
|
| 180 | } ip_structure;
|
| 181 |
|
| 182 | struct gpu_info_header {
|
| 183 | uint32_t table_id;
|
| 184 | uint16_t version_major;
|
| 185 | uint16_t version_minor;
|
| 186 | uint32_t size;
|
| 187 | };
|
| 188 |
|
| 189 | struct gc_info_v1_0 {
|
| 190 | struct gpu_info_header header;
|
| 191 |
|
| 192 | uint32_t gc_num_se;
|
| 193 | uint32_t gc_num_wgp0_per_sa;
|
| 194 | uint32_t gc_num_wgp1_per_sa;
|
| 195 | uint32_t gc_num_rb_per_se;
|
| 196 | uint32_t gc_num_gl2c;
|
| 197 | uint32_t gc_num_gprs;
|
| 198 | uint32_t gc_num_max_gs_thds;
|
| 199 | uint32_t gc_gs_table_depth;
|
| 200 | uint32_t gc_gsprim_buff_depth;
|
| 201 | uint32_t gc_parameter_cache_depth;
|
| 202 | uint32_t gc_double_offchip_lds_buffer;
|
| 203 | uint32_t gc_wave_size;
|
| 204 | uint32_t gc_max_waves_per_simd;
|
| 205 | uint32_t gc_max_scratch_slots_per_cu;
|
| 206 | uint32_t gc_lds_size;
|
| 207 | uint32_t gc_num_sc_per_se;
|
| 208 | uint32_t gc_num_sa_per_se;
|
| 209 | uint32_t gc_num_packer_per_sc;
|
| 210 | uint32_t gc_num_gl2a;
|
| 211 | };
|
| 212 |
|
| 213 | struct gc_info_v1_1 {
|
| 214 | struct gpu_info_header header;
|
| 215 |
|
| 216 | uint32_t gc_num_se;
|
| 217 | uint32_t gc_num_wgp0_per_sa;
|
| 218 | uint32_t gc_num_wgp1_per_sa;
|
| 219 | uint32_t gc_num_rb_per_se;
|
| 220 | uint32_t gc_num_gl2c;
|
| 221 | uint32_t gc_num_gprs;
|
| 222 | uint32_t gc_num_max_gs_thds;
|
| 223 | uint32_t gc_gs_table_depth;
|
| 224 | uint32_t gc_gsprim_buff_depth;
|
| 225 | uint32_t gc_parameter_cache_depth;
|
| 226 | uint32_t gc_double_offchip_lds_buffer;
|
| 227 | uint32_t gc_wave_size;
|
| 228 | uint32_t gc_max_waves_per_simd;
|
| 229 | uint32_t gc_max_scratch_slots_per_cu;
|
| 230 | uint32_t gc_lds_size;
|
| 231 | uint32_t gc_num_sc_per_se;
|
| 232 | uint32_t gc_num_sa_per_se;
|
| 233 | uint32_t gc_num_packer_per_sc;
|
| 234 | uint32_t gc_num_gl2a;
|
| 235 | uint32_t gc_num_tcp_per_sa;
|
| 236 | uint32_t gc_num_sdp_interface;
|
| 237 | uint32_t gc_num_tcps;
|
| 238 | };
|
| 239 |
|
| 240 | struct gc_info_v1_2 {
|
| 241 | struct gpu_info_header header;
|
| 242 | uint32_t gc_num_se;
|
| 243 | uint32_t gc_num_wgp0_per_sa;
|
| 244 | uint32_t gc_num_wgp1_per_sa;
|
| 245 | uint32_t gc_num_rb_per_se;
|
| 246 | uint32_t gc_num_gl2c;
|
| 247 | uint32_t gc_num_gprs;
|
| 248 | uint32_t gc_num_max_gs_thds;
|
| 249 | uint32_t gc_gs_table_depth;
|
| 250 | uint32_t gc_gsprim_buff_depth;
|
| 251 | uint32_t gc_parameter_cache_depth;
|
| 252 | uint32_t gc_double_offchip_lds_buffer;
|
| 253 | uint32_t gc_wave_size;
|
| 254 | uint32_t gc_max_waves_per_simd;
|
| 255 | uint32_t gc_max_scratch_slots_per_cu;
|
| 256 | uint32_t gc_lds_size;
|
| 257 | uint32_t gc_num_sc_per_se;
|
| 258 | uint32_t gc_num_sa_per_se;
|
| 259 | uint32_t gc_num_packer_per_sc;
|
| 260 | uint32_t gc_num_gl2a;
|
| 261 | uint32_t gc_num_tcp_per_sa;
|
| 262 | uint32_t gc_num_sdp_interface;
|
| 263 | uint32_t gc_num_tcps;
|
| 264 | uint32_t gc_num_tcp_per_wpg;
|
| 265 | uint32_t gc_tcp_l1_size;
|
| 266 | uint32_t gc_num_sqc_per_wgp;
|
| 267 | uint32_t gc_l1_instruction_cache_size_per_sqc;
|
| 268 | uint32_t gc_l1_data_cache_size_per_sqc;
|
| 269 | uint32_t gc_gl1c_per_sa;
|
| 270 | uint32_t gc_gl1c_size_per_instance;
|
| 271 | uint32_t gc_gl2c_per_gpu;
|
| 272 | };
|
| 273 |
|
| 274 | struct gc_info_v1_3 {
|
| 275 | struct gpu_info_header header;
|
| 276 | uint32_t gc_num_se;
|
| 277 | uint32_t gc_num_wgp0_per_sa;
|
| 278 | uint32_t gc_num_wgp1_per_sa;
|
| 279 | uint32_t gc_num_rb_per_se;
|
| 280 | uint32_t gc_num_gl2c;
|
| 281 | uint32_t gc_num_gprs;
|
| 282 | uint32_t gc_num_max_gs_thds;
|
| 283 | uint32_t gc_gs_table_depth;
|
| 284 | uint32_t gc_gsprim_buff_depth;
|
| 285 | uint32_t gc_parameter_cache_depth;
|
| 286 | uint32_t gc_double_offchip_lds_buffer;
|
| 287 | uint32_t gc_wave_size;
|
| 288 | uint32_t gc_max_waves_per_simd;
|
| 289 | uint32_t gc_max_scratch_slots_per_cu;
|
| 290 | uint32_t gc_lds_size;
|
| 291 | uint32_t gc_num_sc_per_se;
|
| 292 | uint32_t gc_num_sa_per_se;
|
| 293 | uint32_t gc_num_packer_per_sc;
|
| 294 | uint32_t gc_num_gl2a;
|
| 295 | uint32_t gc_num_tcp_per_sa;
|
| 296 | uint32_t gc_num_sdp_interface;
|
| 297 | uint32_t gc_num_tcps;
|
| 298 | uint32_t gc_num_tcp_per_wpg;
|
| 299 | uint32_t gc_tcp_l1_size;
|
| 300 | uint32_t gc_num_sqc_per_wgp;
|
| 301 | uint32_t gc_l1_instruction_cache_size_per_sqc;
|
| 302 | uint32_t gc_l1_data_cache_size_per_sqc;
|
| 303 | uint32_t gc_gl1c_per_sa;
|
| 304 | uint32_t gc_gl1c_size_per_instance;
|
| 305 | uint32_t gc_gl2c_per_gpu;
|
| 306 | uint32_t gc_tcp_size_per_cu;
|
| 307 | uint32_t gc_tcp_cache_line_size;
|
| 308 | uint32_t gc_instruction_cache_size_per_sqc;
|
| 309 | uint32_t gc_instruction_cache_line_size;
|
| 310 | uint32_t gc_scalar_data_cache_size_per_sqc;
|
| 311 | uint32_t gc_scalar_data_cache_line_size;
|
| 312 | uint32_t gc_tcc_size;
|
| 313 | uint32_t gc_tcc_cache_line_size;
|
| 314 | };
|
| 315 |
|
| 316 | struct gc_info_v2_0 {
|
| 317 | struct gpu_info_header header;
|
| 318 |
|
| 319 | uint32_t gc_num_se;
|
| 320 | uint32_t gc_num_cu_per_sh;
|
| 321 | uint32_t gc_num_sh_per_se;
|
| 322 | uint32_t gc_num_rb_per_se;
|
| 323 | uint32_t gc_num_tccs;
|
| 324 | uint32_t gc_num_gprs;
|
| 325 | uint32_t gc_num_max_gs_thds;
|
| 326 | uint32_t gc_gs_table_depth;
|
| 327 | uint32_t gc_gsprim_buff_depth;
|
| 328 | uint32_t gc_parameter_cache_depth;
|
| 329 | uint32_t gc_double_offchip_lds_buffer;
|
| 330 | uint32_t gc_wave_size;
|
| 331 | uint32_t gc_max_waves_per_simd;
|
| 332 | uint32_t gc_max_scratch_slots_per_cu;
|
| 333 | uint32_t gc_lds_size;
|
| 334 | uint32_t gc_num_sc_per_se;
|
| 335 | uint32_t gc_num_packer_per_sc;
|
| 336 | };
|
| 337 |
|
| 338 | struct gc_info_v2_1 {
|
| 339 | struct gpu_info_header header;
|
| 340 |
|
| 341 | uint32_t gc_num_se;
|
| 342 | uint32_t gc_num_cu_per_sh;
|
| 343 | uint32_t gc_num_sh_per_se;
|
| 344 | uint32_t gc_num_rb_per_se;
|
| 345 | uint32_t gc_num_tccs;
|
| 346 | uint32_t gc_num_gprs;
|
| 347 | uint32_t gc_num_max_gs_thds;
|
| 348 | uint32_t gc_gs_table_depth;
|
| 349 | uint32_t gc_gsprim_buff_depth;
|
| 350 | uint32_t gc_parameter_cache_depth;
|
| 351 | uint32_t gc_double_offchip_lds_buffer;
|
| 352 | uint32_t gc_wave_size;
|
| 353 | uint32_t gc_max_waves_per_simd;
|
| 354 | uint32_t gc_max_scratch_slots_per_cu;
|
| 355 | uint32_t gc_lds_size;
|
| 356 | uint32_t gc_num_sc_per_se;
|
| 357 | uint32_t gc_num_packer_per_sc;
|
| 358 |
|
| 359 | uint32_t gc_num_tcp_per_sh;
|
| 360 | uint32_t gc_tcp_size_per_cu;
|
| 361 | uint32_t gc_num_sdp_interface;
|
| 362 | uint32_t gc_num_cu_per_sqc;
|
| 363 | uint32_t gc_instruction_cache_size_per_sqc;
|
| 364 | uint32_t gc_scalar_data_cache_size_per_sqc;
|
| 365 | uint32_t gc_tcc_size;
|
| 366 | };
|
| 367 |
|
| 368 | typedef struct harvest_info_header {
|
| 369 | uint32_t signature;
|
| 370 | uint32_t version;
|
| 371 | } harvest_info_header;
|
| 372 |
|
| 373 | typedef struct harvest_info {
|
| 374 | uint16_t hw_id;
|
| 375 | uint8_t number_instance;
|
| 376 | uint8_t reserved;
|
| 377 | } harvest_info;
|
| 378 |
|
| 379 | typedef struct harvest_table {
|
| 380 | harvest_info_header header;
|
| 381 | harvest_info list[32];
|
| 382 | } harvest_table;
|
| 383 |
|
| 384 | struct mall_info_header {
|
| 385 | uint32_t table_id;
|
| 386 | uint16_t version_major;
|
| 387 | uint16_t version_minor;
|
| 388 | uint32_t size_bytes;
|
| 389 | };
|
| 390 |
|
| 391 | struct mall_info_v1_0 {
|
| 392 | struct mall_info_header header;
|
| 393 | uint32_t mall_size_per_m;
|
| 394 | uint32_t m_s_present;
|
| 395 | uint32_t m_half_use;
|
| 396 | uint32_t m_mall_config;
|
| 397 | uint32_t reserved[5];
|
| 398 | };
|
| 399 |
|
| 400 | struct mall_info_v2_0 {
|
| 401 | struct mall_info_header header;
|
| 402 | uint32_t mall_size_per_umc;
|
| 403 | uint32_t reserved[8];
|
| 404 | };
|
| 405 |
|
| 406 | #define VCN_INFO_TABLE_MAX_NUM_INSTANCES 4
|
| 407 |
|
| 408 | struct vcn_info_header {
|
| 409 | uint32_t table_id;
|
| 410 | uint16_t version_major;
|
| 411 | uint16_t version_minor;
|
| 412 | uint32_t size_bytes;
|
| 413 | };
|
| 414 |
|
| 415 | struct vcn_instance_info_v1_0
|
| 416 | {
|
| 417 | uint32_t instance_num;
|
| 418 | union _fuse_data {
|
| 419 | struct {
|
| 420 | uint32_t av1_disabled : 1;
|
| 421 | uint32_t vp9_disabled : 1;
|
| 422 | uint32_t hevc_disabled : 1;
|
| 423 | uint32_t h264_disabled : 1;
|
| 424 | uint32_t reserved : 28;
|
| 425 | } bits;
|
| 426 | uint32_t all_bits;
|
| 427 | } fuse_data;
|
| 428 | uint32_t reserved[2];
|
| 429 | };
|
| 430 |
|
| 431 | struct vcn_info_v1_0 {
|
| 432 | struct vcn_info_header header;
|
| 433 | uint32_t num_of_instances;
|
| 434 | struct vcn_instance_info_v1_0 instance_info[VCN_INFO_TABLE_MAX_NUM_INSTANCES];
|
| 435 | uint32_t reserved[4];
|
| 436 | };
|
| 437 |
|
| 438 | #define NPS_INFO_TABLE_MAX_NUM_INSTANCES 12
|
| 439 |
|
| 440 | struct nps_info_header {
|
| 441 | uint32_t table_id;
|
| 442 | uint16_t version_major;
|
| 443 | uint16_t version_minor;
|
| 444 | uint32_t size_bytes;
|
| 445 | };
|
| 446 |
|
| 447 | struct nps_instance_info_v1_0 {
|
| 448 | uint64_t base_address;
|
| 449 | uint64_t limit_address;
|
| 450 | };
|
| 451 |
|
| 452 | struct nps_info_v1_0 {
|
| 453 | struct nps_info_header header;
|
| 454 | uint32_t nps_type;
|
| 455 | uint32_t count;
|
| 456 | struct nps_instance_info_v1_0
|
| 457 | instance_info[NPS_INFO_TABLE_MAX_NUM_INSTANCES];
|
| 458 | };
|
| 459 |
|
| 460 | #pragma pack()
|
| 461 |
|
| 462 | #endif
|