| 1 | /* |
| 2 | * Copyright 2014 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 __AMDGPU_GFX_H__ |
| 25 | #define __AMDGPU_GFX_H__ |
| 26 | |
| 27 | /* |
| 28 | * GFX stuff |
| 29 | */ |
| 30 | #include "clearstate_defs.h" |
| 31 | #include "amdgpu_ring.h" |
| 32 | #include "amdgpu_rlc.h" |
| 33 | #include "amdgpu_imu.h" |
| 34 | #include "soc15.h" |
| 35 | #include "amdgpu_ras.h" |
| 36 | #include "amdgpu_ring_mux.h" |
| 37 | #include "amdgpu_xcp.h" |
| 38 | |
| 39 | /* GFX current status */ |
| 40 | #define AMDGPU_GFX_NORMAL_MODE 0x00000000L |
| 41 | #define AMDGPU_GFX_SAFE_MODE 0x00000001L |
| 42 | #define AMDGPU_GFX_PG_DISABLED_MODE 0x00000002L |
| 43 | #define AMDGPU_GFX_CG_DISABLED_MODE 0x00000004L |
| 44 | #define AMDGPU_GFX_LBPW_DISABLED_MODE 0x00000008L |
| 45 | |
| 46 | #define AMDGPU_MAX_GC_INSTANCES 8 |
| 47 | #define AMDGPU_MAX_QUEUES 128 |
| 48 | |
| 49 | #define AMDGPU_MAX_GFX_QUEUES AMDGPU_MAX_QUEUES |
| 50 | #define AMDGPU_MAX_COMPUTE_QUEUES AMDGPU_MAX_QUEUES |
| 51 | |
| 52 | enum amdgpu_gfx_pipe_priority { |
| 53 | AMDGPU_GFX_PIPE_PRIO_NORMAL = AMDGPU_RING_PRIO_1, |
| 54 | AMDGPU_GFX_PIPE_PRIO_HIGH = AMDGPU_RING_PRIO_2 |
| 55 | }; |
| 56 | |
| 57 | #define AMDGPU_GFX_QUEUE_PRIORITY_MINIMUM 0 |
| 58 | #define AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM 15 |
| 59 | |
| 60 | /* 1 second timeout */ |
| 61 | #define GFX_PROFILE_IDLE_TIMEOUT msecs_to_jiffies(1000) |
| 62 | |
| 63 | enum amdgpu_gfx_partition { |
| 64 | AMDGPU_SPX_PARTITION_MODE = 0, |
| 65 | AMDGPU_DPX_PARTITION_MODE = 1, |
| 66 | AMDGPU_TPX_PARTITION_MODE = 2, |
| 67 | AMDGPU_QPX_PARTITION_MODE = 3, |
| 68 | AMDGPU_CPX_PARTITION_MODE = 4, |
| 69 | AMDGPU_UNKNOWN_COMPUTE_PARTITION_MODE = -1, |
| 70 | /* Automatically choose the right mode */ |
| 71 | AMDGPU_AUTO_COMPUTE_PARTITION_MODE = -2, |
| 72 | }; |
| 73 | |
| 74 | #define NUM_XCC(x) hweight16(x) |
| 75 | |
| 76 | enum amdgpu_gfx_ras_mem_id_type { |
| 77 | AMDGPU_GFX_CP_MEM = 0, |
| 78 | AMDGPU_GFX_GCEA_MEM, |
| 79 | AMDGPU_GFX_GC_CANE_MEM, |
| 80 | AMDGPU_GFX_GCUTCL2_MEM, |
| 81 | AMDGPU_GFX_GDS_MEM, |
| 82 | AMDGPU_GFX_LDS_MEM, |
| 83 | AMDGPU_GFX_RLC_MEM, |
| 84 | AMDGPU_GFX_SP_MEM, |
| 85 | AMDGPU_GFX_SPI_MEM, |
| 86 | AMDGPU_GFX_SQC_MEM, |
| 87 | AMDGPU_GFX_SQ_MEM, |
| 88 | AMDGPU_GFX_TA_MEM, |
| 89 | AMDGPU_GFX_TCC_MEM, |
| 90 | AMDGPU_GFX_TCA_MEM, |
| 91 | AMDGPU_GFX_TCI_MEM, |
| 92 | AMDGPU_GFX_TCP_MEM, |
| 93 | AMDGPU_GFX_TD_MEM, |
| 94 | AMDGPU_GFX_TCX_MEM, |
| 95 | AMDGPU_GFX_ATC_L2_MEM, |
| 96 | AMDGPU_GFX_UTCL2_MEM, |
| 97 | AMDGPU_GFX_VML2_MEM, |
| 98 | AMDGPU_GFX_VML2_WALKER_MEM, |
| 99 | AMDGPU_GFX_MEM_TYPE_NUM |
| 100 | }; |
| 101 | |
| 102 | struct amdgpu_mec { |
| 103 | struct amdgpu_bo *hpd_eop_obj; |
| 104 | u64 hpd_eop_gpu_addr; |
| 105 | struct amdgpu_bo *mec_fw_obj; |
| 106 | u64 mec_fw_gpu_addr; |
| 107 | struct amdgpu_bo *mec_fw_data_obj; |
| 108 | u64 mec_fw_data_gpu_addr; |
| 109 | |
| 110 | u32 num_mec; |
| 111 | u32 num_pipe_per_mec; |
| 112 | u32 num_queue_per_pipe; |
| 113 | void *mqd_backup[AMDGPU_MAX_COMPUTE_RINGS * AMDGPU_MAX_GC_INSTANCES]; |
| 114 | }; |
| 115 | |
| 116 | struct amdgpu_mec_bitmap { |
| 117 | /* These are the resources for which amdgpu takes ownership */ |
| 118 | DECLARE_BITMAP(queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES); |
| 119 | }; |
| 120 | |
| 121 | enum amdgpu_unmap_queues_action { |
| 122 | PREEMPT_QUEUES = 0, |
| 123 | RESET_QUEUES, |
| 124 | DISABLE_PROCESS_QUEUES, |
| 125 | PREEMPT_QUEUES_NO_UNMAP, |
| 126 | }; |
| 127 | |
| 128 | struct kiq_pm4_funcs { |
| 129 | /* Support ASIC-specific kiq pm4 packets*/ |
| 130 | void (*kiq_set_resources)(struct amdgpu_ring *kiq_ring, |
| 131 | uint64_t queue_mask); |
| 132 | void (*kiq_map_queues)(struct amdgpu_ring *kiq_ring, |
| 133 | struct amdgpu_ring *ring); |
| 134 | void (*kiq_unmap_queues)(struct amdgpu_ring *kiq_ring, |
| 135 | struct amdgpu_ring *ring, |
| 136 | enum amdgpu_unmap_queues_action action, |
| 137 | u64 gpu_addr, u64 seq); |
| 138 | void (*kiq_query_status)(struct amdgpu_ring *kiq_ring, |
| 139 | struct amdgpu_ring *ring, |
| 140 | u64 addr, |
| 141 | u64 seq); |
| 142 | void (*kiq_invalidate_tlbs)(struct amdgpu_ring *kiq_ring, |
| 143 | uint16_t pasid, uint32_t flush_type, |
| 144 | bool all_hub); |
| 145 | void (*kiq_reset_hw_queue)(struct amdgpu_ring *kiq_ring, |
| 146 | uint32_t queue_type, uint32_t me_id, |
| 147 | uint32_t pipe_id, uint32_t queue_id, |
| 148 | uint32_t xcc_id, uint32_t vmid); |
| 149 | /* Packet sizes */ |
| 150 | int set_resources_size; |
| 151 | int map_queues_size; |
| 152 | int unmap_queues_size; |
| 153 | int query_status_size; |
| 154 | int invalidate_tlbs_size; |
| 155 | }; |
| 156 | |
| 157 | struct amdgpu_kiq { |
| 158 | u64 eop_gpu_addr; |
| 159 | struct amdgpu_bo *eop_obj; |
| 160 | spinlock_t ring_lock; |
| 161 | struct amdgpu_ring ring; |
| 162 | struct amdgpu_irq_src irq; |
| 163 | const struct kiq_pm4_funcs *pmf; |
| 164 | void *mqd_backup; |
| 165 | }; |
| 166 | |
| 167 | /* |
| 168 | * GFX configurations |
| 169 | */ |
| 170 | #define AMDGPU_GFX_MAX_SE 4 |
| 171 | #define AMDGPU_GFX_MAX_SH_PER_SE 2 |
| 172 | |
| 173 | /** |
| 174 | * amdgpu_rb_config - Configure a single Render Backend (RB) |
| 175 | * |
| 176 | * Bad RBs are fused off and there is a harvest register the driver reads to |
| 177 | * determine which RB(s) are fused off so that the driver can configure the |
| 178 | * hardware state so that nothing gets sent to them. There are also user |
| 179 | * harvest registers that the driver can program to disable additional RBs, |
| 180 | * etc., for testing purposes. |
| 181 | */ |
| 182 | struct amdgpu_rb_config { |
| 183 | /** |
| 184 | * @rb_backend_disable: |
| 185 | * |
| 186 | * The value captured from register RB_BACKEND_DISABLE indicates if the |
| 187 | * RB backend is disabled or not. |
| 188 | */ |
| 189 | uint32_t rb_backend_disable; |
| 190 | |
| 191 | /** |
| 192 | * @user_rb_backend_disable: |
| 193 | * |
| 194 | * The value captured from register USER_RB_BACKEND_DISABLE indicates |
| 195 | * if the User RB backend is disabled or not. |
| 196 | */ |
| 197 | uint32_t user_rb_backend_disable; |
| 198 | |
| 199 | /** |
| 200 | * @raster_config: |
| 201 | * |
| 202 | * To set up all of the states, it is necessary to have two registers |
| 203 | * to keep all of the states. This field holds the first register. |
| 204 | */ |
| 205 | uint32_t raster_config; |
| 206 | |
| 207 | /** |
| 208 | * @raster_config_1: |
| 209 | * |
| 210 | * To set up all of the states, it is necessary to have two registers |
| 211 | * to keep all of the states. This field holds the second register. |
| 212 | */ |
| 213 | uint32_t raster_config_1; |
| 214 | }; |
| 215 | |
| 216 | struct gb_addr_config { |
| 217 | uint16_t pipe_interleave_size; |
| 218 | uint8_t num_pipes; |
| 219 | uint8_t max_compress_frags; |
| 220 | uint8_t num_banks; |
| 221 | uint8_t num_se; |
| 222 | uint8_t num_rb_per_se; |
| 223 | uint8_t num_pkrs; |
| 224 | }; |
| 225 | |
| 226 | struct amdgpu_gfx_config { |
| 227 | unsigned max_shader_engines; |
| 228 | unsigned max_tile_pipes; |
| 229 | unsigned max_cu_per_sh; |
| 230 | unsigned max_sh_per_se; |
| 231 | unsigned max_backends_per_se; |
| 232 | unsigned max_texture_channel_caches; |
| 233 | unsigned max_gprs; |
| 234 | unsigned max_gs_threads; |
| 235 | unsigned max_hw_contexts; |
| 236 | unsigned sc_prim_fifo_size_frontend; |
| 237 | unsigned sc_prim_fifo_size_backend; |
| 238 | unsigned sc_hiz_tile_fifo_size; |
| 239 | unsigned sc_earlyz_tile_fifo_size; |
| 240 | |
| 241 | unsigned num_tile_pipes; |
| 242 | unsigned backend_enable_mask; |
| 243 | unsigned mem_max_burst_length_bytes; |
| 244 | unsigned mem_row_size_in_kb; |
| 245 | unsigned shader_engine_tile_size; |
| 246 | unsigned num_gpus; |
| 247 | unsigned multi_gpu_tile_size; |
| 248 | unsigned mc_arb_ramcfg; |
| 249 | unsigned num_banks; |
| 250 | unsigned num_ranks; |
| 251 | unsigned gb_addr_config; |
| 252 | unsigned num_rbs; |
| 253 | unsigned gs_vgt_table_depth; |
| 254 | unsigned gs_prim_buffer_depth; |
| 255 | |
| 256 | uint32_t tile_mode_array[32]; |
| 257 | uint32_t macrotile_mode_array[16]; |
| 258 | |
| 259 | struct gb_addr_config gb_addr_config_fields; |
| 260 | |
| 261 | /** |
| 262 | * @rb_config: |
| 263 | * |
| 264 | * Matrix that keeps all the Render Backend (color and depth buffer |
| 265 | * handling) configuration on the 3D engine. |
| 266 | */ |
| 267 | struct amdgpu_rb_config rb_config[AMDGPU_GFX_MAX_SE][AMDGPU_GFX_MAX_SH_PER_SE]; |
| 268 | |
| 269 | /* gfx configure feature */ |
| 270 | uint32_t double_offchip_lds_buf; |
| 271 | /* cached value of DB_DEBUG2 */ |
| 272 | uint32_t db_debug2; |
| 273 | /* gfx10 specific config */ |
| 274 | uint32_t num_sc_per_sh; |
| 275 | uint32_t num_packer_per_sc; |
| 276 | uint32_t pa_sc_tile_steering_override; |
| 277 | /* Whether texture coordinate truncation is conformant. */ |
| 278 | bool ta_cntl2_truncate_coord_mode; |
| 279 | uint64_t tcc_disabled_mask; |
| 280 | uint32_t gc_num_tcp_per_sa; |
| 281 | uint32_t gc_num_sdp_interface; |
| 282 | uint32_t gc_num_tcps; |
| 283 | uint32_t gc_num_tcp_per_wpg; |
| 284 | uint32_t gc_tcp_l1_size; |
| 285 | uint32_t gc_num_sqc_per_wgp; |
| 286 | uint32_t gc_l1_instruction_cache_size_per_sqc; |
| 287 | uint32_t gc_l1_data_cache_size_per_sqc; |
| 288 | uint32_t gc_gl1c_per_sa; |
| 289 | uint32_t gc_gl1c_size_per_instance; |
| 290 | uint32_t gc_gl2c_per_gpu; |
| 291 | uint32_t gc_tcp_size_per_cu; |
| 292 | uint32_t gc_num_cu_per_sqc; |
| 293 | uint32_t gc_tcc_size; |
| 294 | uint32_t gc_tcp_cache_line_size; |
| 295 | uint32_t gc_instruction_cache_size_per_sqc; |
| 296 | uint32_t gc_instruction_cache_line_size; |
| 297 | uint32_t gc_scalar_data_cache_size_per_sqc; |
| 298 | uint32_t gc_scalar_data_cache_line_size; |
| 299 | uint32_t gc_tcc_cache_line_size; |
| 300 | }; |
| 301 | |
| 302 | struct amdgpu_cu_info { |
| 303 | uint32_t simd_per_cu; |
| 304 | uint32_t max_waves_per_simd; |
| 305 | uint32_t wave_front_size; |
| 306 | uint32_t max_scratch_slots_per_cu; |
| 307 | uint32_t lds_size; |
| 308 | |
| 309 | /* total active CU number */ |
| 310 | uint32_t number; |
| 311 | uint32_t ao_cu_mask; |
| 312 | uint32_t ao_cu_bitmap[4][4]; |
| 313 | uint32_t bitmap[AMDGPU_MAX_GC_INSTANCES][4][4]; |
| 314 | }; |
| 315 | |
| 316 | struct amdgpu_gfx_ras { |
| 317 | struct amdgpu_ras_block_object ras_block; |
| 318 | void (*enable_watchdog_timer)(struct amdgpu_device *adev); |
| 319 | int (*rlc_gc_fed_irq)(struct amdgpu_device *adev, |
| 320 | struct amdgpu_irq_src *source, |
| 321 | struct amdgpu_iv_entry *entry); |
| 322 | int (*poison_consumption_handler)(struct amdgpu_device *adev, |
| 323 | struct amdgpu_iv_entry *entry); |
| 324 | }; |
| 325 | |
| 326 | struct amdgpu_gfx_shadow_info { |
| 327 | u32 shadow_size; |
| 328 | u32 shadow_alignment; |
| 329 | u32 csa_size; |
| 330 | u32 csa_alignment; |
| 331 | }; |
| 332 | |
| 333 | struct amdgpu_gfx_funcs { |
| 334 | /* get the gpu clock counter */ |
| 335 | uint64_t (*get_gpu_clock_counter)(struct amdgpu_device *adev); |
| 336 | void (*select_se_sh)(struct amdgpu_device *adev, u32 se_num, |
| 337 | u32 sh_num, u32 instance, int xcc_id); |
| 338 | void (*read_wave_data)(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd, |
| 339 | uint32_t wave, uint32_t *dst, int *no_fields); |
| 340 | void (*read_wave_vgprs)(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd, |
| 341 | uint32_t wave, uint32_t thread, uint32_t start, |
| 342 | uint32_t size, uint32_t *dst); |
| 343 | void (*read_wave_sgprs)(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd, |
| 344 | uint32_t wave, uint32_t start, uint32_t size, |
| 345 | uint32_t *dst); |
| 346 | void (*select_me_pipe_q)(struct amdgpu_device *adev, u32 me, u32 pipe, |
| 347 | u32 queue, u32 vmid, u32 xcc_id); |
| 348 | void (*init_spm_golden)(struct amdgpu_device *adev); |
| 349 | void (*update_perfmon_mgcg)(struct amdgpu_device *adev, bool enable); |
| 350 | int (*get_gfx_shadow_info)(struct amdgpu_device *adev, |
| 351 | struct amdgpu_gfx_shadow_info *shadow_info, |
| 352 | bool skip_check); |
| 353 | enum amdgpu_gfx_partition |
| 354 | (*query_partition_mode)(struct amdgpu_device *adev); |
| 355 | int (*switch_partition_mode)(struct amdgpu_device *adev, |
| 356 | int num_xccs_per_xcp); |
| 357 | int (*ih_node_to_logical_xcc)(struct amdgpu_device *adev, int ih_node); |
| 358 | int (*get_xccs_per_xcp)(struct amdgpu_device *adev); |
| 359 | }; |
| 360 | |
| 361 | struct sq_work { |
| 362 | struct work_struct work; |
| 363 | unsigned ih_data; |
| 364 | }; |
| 365 | |
| 366 | struct amdgpu_pfp { |
| 367 | struct amdgpu_bo *pfp_fw_obj; |
| 368 | uint64_t pfp_fw_gpu_addr; |
| 369 | uint32_t *pfp_fw_ptr; |
| 370 | |
| 371 | struct amdgpu_bo *pfp_fw_data_obj; |
| 372 | uint64_t pfp_fw_data_gpu_addr; |
| 373 | uint32_t *pfp_fw_data_ptr; |
| 374 | }; |
| 375 | |
| 376 | struct amdgpu_ce { |
| 377 | struct amdgpu_bo *ce_fw_obj; |
| 378 | uint64_t ce_fw_gpu_addr; |
| 379 | uint32_t *ce_fw_ptr; |
| 380 | }; |
| 381 | |
| 382 | struct amdgpu_me { |
| 383 | struct amdgpu_bo *me_fw_obj; |
| 384 | uint64_t me_fw_gpu_addr; |
| 385 | uint32_t *me_fw_ptr; |
| 386 | |
| 387 | struct amdgpu_bo *me_fw_data_obj; |
| 388 | uint64_t me_fw_data_gpu_addr; |
| 389 | uint32_t *me_fw_data_ptr; |
| 390 | |
| 391 | uint32_t num_me; |
| 392 | uint32_t num_pipe_per_me; |
| 393 | uint32_t num_queue_per_pipe; |
| 394 | void *mqd_backup[AMDGPU_MAX_GFX_RINGS]; |
| 395 | |
| 396 | /* These are the resources for which amdgpu takes ownership */ |
| 397 | DECLARE_BITMAP(queue_bitmap, AMDGPU_MAX_GFX_QUEUES); |
| 398 | }; |
| 399 | |
| 400 | struct amdgpu_isolation_work { |
| 401 | struct amdgpu_device *adev; |
| 402 | u32 xcp_id; |
| 403 | struct delayed_work work; |
| 404 | }; |
| 405 | |
| 406 | struct amdgpu_gfx { |
| 407 | struct mutex gpu_clock_mutex; |
| 408 | struct amdgpu_gfx_config config; |
| 409 | struct amdgpu_rlc rlc; |
| 410 | struct amdgpu_pfp pfp; |
| 411 | struct amdgpu_ce ce; |
| 412 | struct amdgpu_me me; |
| 413 | struct amdgpu_mec mec; |
| 414 | struct amdgpu_mec_bitmap mec_bitmap[AMDGPU_MAX_GC_INSTANCES]; |
| 415 | struct amdgpu_kiq kiq[AMDGPU_MAX_GC_INSTANCES]; |
| 416 | struct amdgpu_imu imu; |
| 417 | bool rs64_enable; /* firmware format */ |
| 418 | const struct firmware *me_fw; /* ME firmware */ |
| 419 | uint32_t me_fw_version; |
| 420 | const struct firmware *pfp_fw; /* PFP firmware */ |
| 421 | uint32_t pfp_fw_version; |
| 422 | const struct firmware *ce_fw; /* CE firmware */ |
| 423 | uint32_t ce_fw_version; |
| 424 | const struct firmware *rlc_fw; /* RLC firmware */ |
| 425 | uint32_t rlc_fw_version; |
| 426 | const struct firmware *mec_fw; /* MEC firmware */ |
| 427 | uint32_t mec_fw_version; |
| 428 | const struct firmware *mec2_fw; /* MEC2 firmware */ |
| 429 | uint32_t mec2_fw_version; |
| 430 | const struct firmware *imu_fw; /* IMU firmware */ |
| 431 | uint32_t imu_fw_version; |
| 432 | uint32_t me_feature_version; |
| 433 | uint32_t ce_feature_version; |
| 434 | uint32_t pfp_feature_version; |
| 435 | uint32_t rlc_feature_version; |
| 436 | uint32_t rlc_srlc_fw_version; |
| 437 | uint32_t rlc_srlc_feature_version; |
| 438 | uint32_t rlc_srlg_fw_version; |
| 439 | uint32_t rlc_srlg_feature_version; |
| 440 | uint32_t rlc_srls_fw_version; |
| 441 | uint32_t rlc_srls_feature_version; |
| 442 | uint32_t rlcp_ucode_version; |
| 443 | uint32_t rlcp_ucode_feature_version; |
| 444 | uint32_t rlcv_ucode_version; |
| 445 | uint32_t rlcv_ucode_feature_version; |
| 446 | uint32_t mec_feature_version; |
| 447 | uint32_t mec2_feature_version; |
| 448 | bool mec_fw_write_wait; |
| 449 | bool me_fw_write_wait; |
| 450 | bool cp_fw_write_wait; |
| 451 | struct amdgpu_ring gfx_ring[AMDGPU_MAX_GFX_RINGS]; |
| 452 | unsigned num_gfx_rings; |
| 453 | struct amdgpu_ring compute_ring[AMDGPU_MAX_COMPUTE_RINGS * AMDGPU_MAX_GC_INSTANCES]; |
| 454 | unsigned num_compute_rings; |
| 455 | struct amdgpu_irq_src eop_irq; |
| 456 | struct amdgpu_irq_src priv_reg_irq; |
| 457 | struct amdgpu_irq_src priv_inst_irq; |
| 458 | struct amdgpu_irq_src bad_op_irq; |
| 459 | struct amdgpu_irq_src cp_ecc_error_irq; |
| 460 | struct amdgpu_irq_src sq_irq; |
| 461 | struct amdgpu_irq_src rlc_gc_fed_irq; |
| 462 | struct sq_work sq_work; |
| 463 | |
| 464 | /* gfx status */ |
| 465 | uint32_t gfx_current_status; |
| 466 | /* ce ram size*/ |
| 467 | unsigned ce_ram_size; |
| 468 | struct amdgpu_cu_info cu_info; |
| 469 | const struct amdgpu_gfx_funcs *funcs; |
| 470 | |
| 471 | /* reset mask */ |
| 472 | uint32_t grbm_soft_reset; |
| 473 | uint32_t srbm_soft_reset; |
| 474 | uint32_t gfx_supported_reset; |
| 475 | uint32_t compute_supported_reset; |
| 476 | |
| 477 | /* gfx off */ |
| 478 | bool gfx_off_state; /* true: enabled, false: disabled */ |
| 479 | struct mutex gfx_off_mutex; /* mutex to change gfxoff state */ |
| 480 | uint32_t gfx_off_req_count; /* default 1, enable gfx off: dec 1, disable gfx off: add 1 */ |
| 481 | struct delayed_work gfx_off_delay_work; /* async work to set gfx block off */ |
| 482 | uint32_t gfx_off_residency; /* last logged residency */ |
| 483 | uint64_t gfx_off_entrycount; /* count of times GPU has get into GFXOFF state */ |
| 484 | |
| 485 | /* pipe reservation */ |
| 486 | struct mutex pipe_reserve_mutex; |
| 487 | DECLARE_BITMAP (pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES); |
| 488 | |
| 489 | /*ras */ |
| 490 | struct ras_common_if *ras_if; |
| 491 | struct amdgpu_gfx_ras *ras; |
| 492 | |
| 493 | bool is_poweron; |
| 494 | |
| 495 | struct amdgpu_ring sw_gfx_ring[AMDGPU_MAX_SW_GFX_RINGS]; |
| 496 | struct amdgpu_ring_mux muxer; |
| 497 | |
| 498 | bool cp_gfx_shadow; /* for gfx11 */ |
| 499 | |
| 500 | uint16_t xcc_mask; |
| 501 | uint32_t num_xcc_per_xcp; |
| 502 | struct mutex partition_mutex; |
| 503 | bool mcbp; /* mid command buffer preemption */ |
| 504 | |
| 505 | /* IP reg dump */ |
| 506 | uint32_t *ip_dump_core; |
| 507 | uint32_t *ip_dump_compute_queues; |
| 508 | uint32_t *ip_dump_gfx_queues; |
| 509 | |
| 510 | struct mutex reset_sem_mutex; |
| 511 | |
| 512 | /* cleaner shader */ |
| 513 | struct amdgpu_bo *cleaner_shader_obj; |
| 514 | unsigned int cleaner_shader_size; |
| 515 | u64 cleaner_shader_gpu_addr; |
| 516 | void *cleaner_shader_cpu_ptr; |
| 517 | const void *cleaner_shader_ptr; |
| 518 | bool enable_cleaner_shader; |
| 519 | struct amdgpu_isolation_work enforce_isolation[MAX_XCP]; |
| 520 | /* Mutex for synchronizing KFD scheduler operations */ |
| 521 | struct mutex userq_sch_mutex; |
| 522 | u64 userq_sch_req_count[MAX_XCP]; |
| 523 | bool userq_sch_inactive[MAX_XCP]; |
| 524 | unsigned long enforce_isolation_jiffies[MAX_XCP]; |
| 525 | unsigned long enforce_isolation_time[MAX_XCP]; |
| 526 | |
| 527 | atomic_t total_submission_cnt; |
| 528 | struct delayed_work idle_work; |
| 529 | bool workload_profile_active; |
| 530 | struct mutex workload_profile_mutex; |
| 531 | |
| 532 | bool disable_kq; |
| 533 | bool disable_uq; |
| 534 | }; |
| 535 | |
| 536 | struct amdgpu_gfx_ras_reg_entry { |
| 537 | struct amdgpu_ras_err_status_reg_entry reg_entry; |
| 538 | enum amdgpu_gfx_ras_mem_id_type mem_id_type; |
| 539 | uint32_t se_num; |
| 540 | }; |
| 541 | |
| 542 | struct amdgpu_gfx_ras_mem_id_entry { |
| 543 | const struct amdgpu_ras_memory_id_entry *mem_id_ent; |
| 544 | uint32_t size; |
| 545 | }; |
| 546 | |
| 547 | #define AMDGPU_GFX_MEMID_ENT(x) {(x), ARRAY_SIZE(x)}, |
| 548 | |
| 549 | #define amdgpu_gfx_get_gpu_clock_counter(adev) (adev)->gfx.funcs->get_gpu_clock_counter((adev)) |
| 550 | #define amdgpu_gfx_select_se_sh(adev, se, sh, instance, xcc_id) ((adev)->gfx.funcs->select_se_sh((adev), (se), (sh), (instance), (xcc_id))) |
| 551 | #define amdgpu_gfx_select_me_pipe_q(adev, me, pipe, q, vmid, xcc_id) ((adev)->gfx.funcs->select_me_pipe_q((adev), (me), (pipe), (q), (vmid), (xcc_id))) |
| 552 | #define amdgpu_gfx_init_spm_golden(adev) (adev)->gfx.funcs->init_spm_golden((adev)) |
| 553 | #define amdgpu_gfx_get_gfx_shadow_info(adev, si) ((adev)->gfx.funcs->get_gfx_shadow_info((adev), (si), false)) |
| 554 | |
| 555 | /** |
| 556 | * amdgpu_gfx_create_bitmask - create a bitmask |
| 557 | * |
| 558 | * @bit_width: length of the mask |
| 559 | * |
| 560 | * create a variable length bit mask. |
| 561 | * Returns the bitmask. |
| 562 | */ |
| 563 | static inline u32 amdgpu_gfx_create_bitmask(u32 bit_width) |
| 564 | { |
| 565 | return (u32)((1ULL << bit_width) - 1); |
| 566 | } |
| 567 | |
| 568 | void amdgpu_gfx_parse_disable_cu(unsigned *mask, unsigned max_se, |
| 569 | unsigned max_sh); |
| 570 | |
| 571 | int amdgpu_gfx_kiq_init_ring(struct amdgpu_device *adev, int xcc_id); |
| 572 | |
| 573 | void amdgpu_gfx_kiq_free_ring(struct amdgpu_ring *ring); |
| 574 | |
| 575 | void amdgpu_gfx_kiq_fini(struct amdgpu_device *adev, int xcc_id); |
| 576 | int amdgpu_gfx_kiq_init(struct amdgpu_device *adev, |
| 577 | unsigned hpd_size, int xcc_id); |
| 578 | |
| 579 | int amdgpu_gfx_mqd_sw_init(struct amdgpu_device *adev, |
| 580 | unsigned mqd_size, int xcc_id); |
| 581 | void amdgpu_gfx_mqd_sw_fini(struct amdgpu_device *adev, int xcc_id); |
| 582 | int amdgpu_gfx_disable_kcq(struct amdgpu_device *adev, int xcc_id); |
| 583 | int amdgpu_gfx_enable_kcq(struct amdgpu_device *adev, int xcc_id); |
| 584 | int amdgpu_gfx_disable_kgq(struct amdgpu_device *adev, int xcc_id); |
| 585 | int amdgpu_gfx_enable_kgq(struct amdgpu_device *adev, int xcc_id); |
| 586 | |
| 587 | void amdgpu_gfx_compute_queue_acquire(struct amdgpu_device *adev); |
| 588 | void amdgpu_gfx_graphics_queue_acquire(struct amdgpu_device *adev); |
| 589 | |
| 590 | int amdgpu_gfx_mec_queue_to_bit(struct amdgpu_device *adev, int mec, |
| 591 | int pipe, int queue); |
| 592 | void amdgpu_queue_mask_bit_to_mec_queue(struct amdgpu_device *adev, int bit, |
| 593 | int *mec, int *pipe, int *queue); |
| 594 | bool amdgpu_gfx_is_mec_queue_enabled(struct amdgpu_device *adev, int xcc_id, |
| 595 | int mec, int pipe, int queue); |
| 596 | bool amdgpu_gfx_is_high_priority_compute_queue(struct amdgpu_device *adev, |
| 597 | struct amdgpu_ring *ring); |
| 598 | bool amdgpu_gfx_is_high_priority_graphics_queue(struct amdgpu_device *adev, |
| 599 | struct amdgpu_ring *ring); |
| 600 | bool amdgpu_gfx_is_me_queue_enabled(struct amdgpu_device *adev, int me, |
| 601 | int pipe, int queue); |
| 602 | void amdgpu_gfx_off_ctrl(struct amdgpu_device *adev, bool enable); |
| 603 | void amdgpu_gfx_off_ctrl_immediate(struct amdgpu_device *adev, bool enable); |
| 604 | int amdgpu_get_gfx_off_status(struct amdgpu_device *adev, uint32_t *value); |
| 605 | int amdgpu_gfx_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block); |
| 606 | void amdgpu_gfx_ras_fini(struct amdgpu_device *adev); |
| 607 | int amdgpu_get_gfx_off_entrycount(struct amdgpu_device *adev, u64 *value); |
| 608 | int amdgpu_get_gfx_off_residency(struct amdgpu_device *adev, u32 *residency); |
| 609 | int amdgpu_set_gfx_off_residency(struct amdgpu_device *adev, bool value); |
| 610 | int amdgpu_gfx_process_ras_data_cb(struct amdgpu_device *adev, |
| 611 | void *err_data, |
| 612 | struct amdgpu_iv_entry *entry); |
| 613 | int amdgpu_gfx_cp_ecc_error_irq(struct amdgpu_device *adev, |
| 614 | struct amdgpu_irq_src *source, |
| 615 | struct amdgpu_iv_entry *entry); |
| 616 | uint32_t amdgpu_kiq_rreg(struct amdgpu_device *adev, uint32_t reg, uint32_t xcc_id); |
| 617 | void amdgpu_kiq_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v, uint32_t xcc_id); |
| 618 | int amdgpu_gfx_get_num_kcq(struct amdgpu_device *adev); |
| 619 | void amdgpu_gfx_cp_init_microcode(struct amdgpu_device *adev, uint32_t ucode_id); |
| 620 | |
| 621 | int amdgpu_gfx_ras_sw_init(struct amdgpu_device *adev); |
| 622 | int amdgpu_gfx_poison_consumption_handler(struct amdgpu_device *adev, |
| 623 | struct amdgpu_iv_entry *entry); |
| 624 | |
| 625 | bool amdgpu_gfx_is_master_xcc(struct amdgpu_device *adev, int xcc_id); |
| 626 | int amdgpu_gfx_sysfs_init(struct amdgpu_device *adev); |
| 627 | void amdgpu_gfx_sysfs_fini(struct amdgpu_device *adev); |
| 628 | void amdgpu_gfx_ras_error_func(struct amdgpu_device *adev, |
| 629 | void *ras_error_status, |
| 630 | void (*func)(struct amdgpu_device *adev, void *ras_error_status, |
| 631 | int xcc_id)); |
| 632 | int amdgpu_gfx_cleaner_shader_sw_init(struct amdgpu_device *adev, |
| 633 | unsigned int cleaner_shader_size); |
| 634 | void amdgpu_gfx_cleaner_shader_sw_fini(struct amdgpu_device *adev); |
| 635 | void amdgpu_gfx_cleaner_shader_init(struct amdgpu_device *adev, |
| 636 | unsigned int cleaner_shader_size, |
| 637 | const void *cleaner_shader_ptr); |
| 638 | void amdgpu_gfx_enforce_isolation_handler(struct work_struct *work); |
| 639 | void amdgpu_gfx_enforce_isolation_ring_begin_use(struct amdgpu_ring *ring); |
| 640 | void amdgpu_gfx_enforce_isolation_ring_end_use(struct amdgpu_ring *ring); |
| 641 | |
| 642 | void amdgpu_gfx_profile_idle_work_handler(struct work_struct *work); |
| 643 | void amdgpu_gfx_profile_ring_begin_use(struct amdgpu_ring *ring); |
| 644 | void amdgpu_gfx_profile_ring_end_use(struct amdgpu_ring *ring); |
| 645 | u32 amdgpu_gfx_csb_preamble_start(volatile u32 *buffer); |
| 646 | u32 amdgpu_gfx_csb_data_parser(struct amdgpu_device *adev, volatile u32 *buffer, u32 count); |
| 647 | void amdgpu_gfx_csb_preamble_end(volatile u32 *buffer, u32 count); |
| 648 | |
| 649 | void amdgpu_debugfs_gfx_sched_mask_init(struct amdgpu_device *adev); |
| 650 | void amdgpu_debugfs_compute_sched_mask_init(struct amdgpu_device *adev); |
| 651 | |
| 652 | static inline const char *amdgpu_gfx_compute_mode_desc(int mode) |
| 653 | { |
| 654 | switch (mode) { |
| 655 | case AMDGPU_SPX_PARTITION_MODE: |
| 656 | return "SPX" ; |
| 657 | case AMDGPU_DPX_PARTITION_MODE: |
| 658 | return "DPX" ; |
| 659 | case AMDGPU_TPX_PARTITION_MODE: |
| 660 | return "TPX" ; |
| 661 | case AMDGPU_QPX_PARTITION_MODE: |
| 662 | return "QPX" ; |
| 663 | case AMDGPU_CPX_PARTITION_MODE: |
| 664 | return "CPX" ; |
| 665 | default: |
| 666 | return "UNKNOWN" ; |
| 667 | } |
| 668 | } |
| 669 | |
| 670 | #endif |
| 671 | |