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 38 /* GFX current status */ 39 #define AMDGPU_GFX_NORMAL_MODE 0x00000000L 40 #define AMDGPU_GFX_SAFE_MODE 0x00000001L 41 #define AMDGPU_GFX_PG_DISABLED_MODE 0x00000002L 42 #define AMDGPU_GFX_CG_DISABLED_MODE 0x00000004L 43 #define AMDGPU_GFX_LBPW_DISABLED_MODE 0x00000008L 44 45 #define AMDGPU_MAX_GFX_QUEUES KGD_MAX_QUEUES 46 #define AMDGPU_MAX_COMPUTE_QUEUES KGD_MAX_QUEUES 47 48 enum amdgpu_gfx_pipe_priority { 49 AMDGPU_GFX_PIPE_PRIO_NORMAL = AMDGPU_RING_PRIO_1, 50 AMDGPU_GFX_PIPE_PRIO_HIGH = AMDGPU_RING_PRIO_2 51 }; 52 53 #define AMDGPU_GFX_QUEUE_PRIORITY_MINIMUM 0 54 #define AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM 15 55 56 struct amdgpu_mec { 57 struct amdgpu_bo *hpd_eop_obj; 58 u64 hpd_eop_gpu_addr; 59 struct amdgpu_bo *mec_fw_obj; 60 u64 mec_fw_gpu_addr; 61 struct amdgpu_bo *mec_fw_data_obj; 62 u64 mec_fw_data_gpu_addr; 63 64 u32 num_mec; 65 u32 num_pipe_per_mec; 66 u32 num_queue_per_pipe; 67 void *mqd_backup[AMDGPU_MAX_COMPUTE_RINGS + 1]; 68 69 /* These are the resources for which amdgpu takes ownership */ 70 DECLARE_BITMAP(queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES); 71 }; 72 73 enum amdgpu_unmap_queues_action { 74 PREEMPT_QUEUES = 0, 75 RESET_QUEUES, 76 DISABLE_PROCESS_QUEUES, 77 PREEMPT_QUEUES_NO_UNMAP, 78 }; 79 80 struct kiq_pm4_funcs { 81 /* Support ASIC-specific kiq pm4 packets*/ 82 void (*kiq_set_resources)(struct amdgpu_ring *kiq_ring, 83 uint64_t queue_mask); 84 void (*kiq_map_queues)(struct amdgpu_ring *kiq_ring, 85 struct amdgpu_ring *ring); 86 void (*kiq_unmap_queues)(struct amdgpu_ring *kiq_ring, 87 struct amdgpu_ring *ring, 88 enum amdgpu_unmap_queues_action action, 89 u64 gpu_addr, u64 seq); 90 void (*kiq_query_status)(struct amdgpu_ring *kiq_ring, 91 struct amdgpu_ring *ring, 92 u64 addr, 93 u64 seq); 94 void (*kiq_invalidate_tlbs)(struct amdgpu_ring *kiq_ring, 95 uint16_t pasid, uint32_t flush_type, 96 bool all_hub); 97 /* Packet sizes */ 98 int set_resources_size; 99 int map_queues_size; 100 int unmap_queues_size; 101 int query_status_size; 102 int invalidate_tlbs_size; 103 }; 104 105 struct amdgpu_kiq { 106 u64 eop_gpu_addr; 107 struct amdgpu_bo *eop_obj; 108 spinlock_t ring_lock; 109 struct amdgpu_ring ring; 110 struct amdgpu_irq_src irq; 111 const struct kiq_pm4_funcs *pmf; 112 }; 113 114 /* 115 * GFX configurations 116 */ 117 #define AMDGPU_GFX_MAX_SE 4 118 #define AMDGPU_GFX_MAX_SH_PER_SE 2 119 120 struct amdgpu_rb_config { 121 uint32_t rb_backend_disable; 122 uint32_t user_rb_backend_disable; 123 uint32_t raster_config; 124 uint32_t raster_config_1; 125 }; 126 127 struct gb_addr_config { 128 uint16_t pipe_interleave_size; 129 uint8_t num_pipes; 130 uint8_t max_compress_frags; 131 uint8_t num_banks; 132 uint8_t num_se; 133 uint8_t num_rb_per_se; 134 uint8_t num_pkrs; 135 }; 136 137 struct amdgpu_gfx_config { 138 unsigned max_shader_engines; 139 unsigned max_tile_pipes; 140 unsigned max_cu_per_sh; 141 unsigned max_sh_per_se; 142 unsigned max_backends_per_se; 143 unsigned max_texture_channel_caches; 144 unsigned max_gprs; 145 unsigned max_gs_threads; 146 unsigned max_hw_contexts; 147 unsigned sc_prim_fifo_size_frontend; 148 unsigned sc_prim_fifo_size_backend; 149 unsigned sc_hiz_tile_fifo_size; 150 unsigned sc_earlyz_tile_fifo_size; 151 152 unsigned num_tile_pipes; 153 unsigned backend_enable_mask; 154 unsigned mem_max_burst_length_bytes; 155 unsigned mem_row_size_in_kb; 156 unsigned shader_engine_tile_size; 157 unsigned num_gpus; 158 unsigned multi_gpu_tile_size; 159 unsigned mc_arb_ramcfg; 160 unsigned num_banks; 161 unsigned num_ranks; 162 unsigned gb_addr_config; 163 unsigned num_rbs; 164 unsigned gs_vgt_table_depth; 165 unsigned gs_prim_buffer_depth; 166 167 uint32_t tile_mode_array[32]; 168 uint32_t macrotile_mode_array[16]; 169 170 struct gb_addr_config gb_addr_config_fields; 171 struct amdgpu_rb_config rb_config[AMDGPU_GFX_MAX_SE][AMDGPU_GFX_MAX_SH_PER_SE]; 172 173 /* gfx configure feature */ 174 uint32_t double_offchip_lds_buf; 175 /* cached value of DB_DEBUG2 */ 176 uint32_t db_debug2; 177 /* gfx10 specific config */ 178 uint32_t num_sc_per_sh; 179 uint32_t num_packer_per_sc; 180 uint32_t pa_sc_tile_steering_override; 181 /* Whether texture coordinate truncation is conformant. */ 182 bool ta_cntl2_truncate_coord_mode; 183 uint64_t tcc_disabled_mask; 184 uint32_t gc_num_tcp_per_sa; 185 uint32_t gc_num_sdp_interface; 186 uint32_t gc_num_tcps; 187 uint32_t gc_num_tcp_per_wpg; 188 uint32_t gc_tcp_l1_size; 189 uint32_t gc_num_sqc_per_wgp; 190 uint32_t gc_l1_instruction_cache_size_per_sqc; 191 uint32_t gc_l1_data_cache_size_per_sqc; 192 uint32_t gc_gl1c_per_sa; 193 uint32_t gc_gl1c_size_per_instance; 194 uint32_t gc_gl2c_per_gpu; 195 }; 196 197 struct amdgpu_cu_info { 198 uint32_t simd_per_cu; 199 uint32_t max_waves_per_simd; 200 uint32_t wave_front_size; 201 uint32_t max_scratch_slots_per_cu; 202 uint32_t lds_size; 203 204 /* total active CU number */ 205 uint32_t number; 206 uint32_t ao_cu_mask; 207 uint32_t ao_cu_bitmap[4][4]; 208 uint32_t bitmap[4][4]; 209 }; 210 211 struct amdgpu_gfx_ras { 212 struct amdgpu_ras_block_object ras_block; 213 void (*enable_watchdog_timer)(struct amdgpu_device *adev); 214 bool (*query_utcl2_poison_status)(struct amdgpu_device *adev); 215 int (*rlc_gc_fed_irq)(struct amdgpu_device *adev, 216 struct amdgpu_irq_src *source, 217 struct amdgpu_iv_entry *entry); 218 int (*poison_consumption_handler)(struct amdgpu_device *adev, 219 struct amdgpu_iv_entry *entry); 220 }; 221 222 struct amdgpu_gfx_funcs { 223 /* get the gpu clock counter */ 224 uint64_t (*get_gpu_clock_counter)(struct amdgpu_device *adev); 225 void (*select_se_sh)(struct amdgpu_device *adev, u32 se_num, 226 u32 sh_num, u32 instance); 227 void (*read_wave_data)(struct amdgpu_device *adev, uint32_t simd, 228 uint32_t wave, uint32_t *dst, int *no_fields); 229 void (*read_wave_vgprs)(struct amdgpu_device *adev, uint32_t simd, 230 uint32_t wave, uint32_t thread, uint32_t start, 231 uint32_t size, uint32_t *dst); 232 void (*read_wave_sgprs)(struct amdgpu_device *adev, uint32_t simd, 233 uint32_t wave, uint32_t start, uint32_t size, 234 uint32_t *dst); 235 void (*select_me_pipe_q)(struct amdgpu_device *adev, u32 me, u32 pipe, 236 u32 queue, u32 vmid); 237 void (*init_spm_golden)(struct amdgpu_device *adev); 238 void (*update_perfmon_mgcg)(struct amdgpu_device *adev, bool enable); 239 }; 240 241 struct sq_work { 242 struct work_struct work; 243 unsigned ih_data; 244 }; 245 246 struct amdgpu_pfp { 247 struct amdgpu_bo *pfp_fw_obj; 248 uint64_t pfp_fw_gpu_addr; 249 uint32_t *pfp_fw_ptr; 250 251 struct amdgpu_bo *pfp_fw_data_obj; 252 uint64_t pfp_fw_data_gpu_addr; 253 uint32_t *pfp_fw_data_ptr; 254 }; 255 256 struct amdgpu_ce { 257 struct amdgpu_bo *ce_fw_obj; 258 uint64_t ce_fw_gpu_addr; 259 uint32_t *ce_fw_ptr; 260 }; 261 262 struct amdgpu_me { 263 struct amdgpu_bo *me_fw_obj; 264 uint64_t me_fw_gpu_addr; 265 uint32_t *me_fw_ptr; 266 267 struct amdgpu_bo *me_fw_data_obj; 268 uint64_t me_fw_data_gpu_addr; 269 uint32_t *me_fw_data_ptr; 270 271 uint32_t num_me; 272 uint32_t num_pipe_per_me; 273 uint32_t num_queue_per_pipe; 274 void *mqd_backup[AMDGPU_MAX_GFX_RINGS]; 275 276 /* These are the resources for which amdgpu takes ownership */ 277 DECLARE_BITMAP(queue_bitmap, AMDGPU_MAX_GFX_QUEUES); 278 }; 279 280 struct amdgpu_gfx { 281 struct mutex gpu_clock_mutex; 282 struct amdgpu_gfx_config config; 283 struct amdgpu_rlc rlc; 284 struct amdgpu_pfp pfp; 285 struct amdgpu_ce ce; 286 struct amdgpu_me me; 287 struct amdgpu_mec mec; 288 struct amdgpu_kiq kiq; 289 struct amdgpu_imu imu; 290 bool rs64_enable; /* firmware format */ 291 const struct firmware *me_fw; /* ME firmware */ 292 uint32_t me_fw_version; 293 const struct firmware *pfp_fw; /* PFP firmware */ 294 uint32_t pfp_fw_version; 295 const struct firmware *ce_fw; /* CE firmware */ 296 uint32_t ce_fw_version; 297 const struct firmware *rlc_fw; /* RLC firmware */ 298 uint32_t rlc_fw_version; 299 const struct firmware *mec_fw; /* MEC firmware */ 300 uint32_t mec_fw_version; 301 const struct firmware *mec2_fw; /* MEC2 firmware */ 302 uint32_t mec2_fw_version; 303 const struct firmware *imu_fw; /* IMU firmware */ 304 uint32_t imu_fw_version; 305 uint32_t me_feature_version; 306 uint32_t ce_feature_version; 307 uint32_t pfp_feature_version; 308 uint32_t rlc_feature_version; 309 uint32_t rlc_srlc_fw_version; 310 uint32_t rlc_srlc_feature_version; 311 uint32_t rlc_srlg_fw_version; 312 uint32_t rlc_srlg_feature_version; 313 uint32_t rlc_srls_fw_version; 314 uint32_t rlc_srls_feature_version; 315 uint32_t rlcp_ucode_version; 316 uint32_t rlcp_ucode_feature_version; 317 uint32_t rlcv_ucode_version; 318 uint32_t rlcv_ucode_feature_version; 319 uint32_t mec_feature_version; 320 uint32_t mec2_feature_version; 321 bool mec_fw_write_wait; 322 bool me_fw_write_wait; 323 bool cp_fw_write_wait; 324 struct amdgpu_ring gfx_ring[AMDGPU_MAX_GFX_RINGS]; 325 unsigned num_gfx_rings; 326 struct amdgpu_ring compute_ring[AMDGPU_MAX_COMPUTE_RINGS]; 327 unsigned num_compute_rings; 328 struct amdgpu_irq_src eop_irq; 329 struct amdgpu_irq_src priv_reg_irq; 330 struct amdgpu_irq_src priv_inst_irq; 331 struct amdgpu_irq_src cp_ecc_error_irq; 332 struct amdgpu_irq_src sq_irq; 333 struct amdgpu_irq_src rlc_gc_fed_irq; 334 struct sq_work sq_work; 335 336 /* gfx status */ 337 uint32_t gfx_current_status; 338 /* ce ram size*/ 339 unsigned ce_ram_size; 340 struct amdgpu_cu_info cu_info; 341 const struct amdgpu_gfx_funcs *funcs; 342 343 /* reset mask */ 344 uint32_t grbm_soft_reset; 345 uint32_t srbm_soft_reset; 346 347 /* gfx off */ 348 bool gfx_off_state; /* true: enabled, false: disabled */ 349 struct mutex gfx_off_mutex; /* mutex to change gfxoff state */ 350 uint32_t gfx_off_req_count; /* default 1, enable gfx off: dec 1, disable gfx off: add 1 */ 351 struct delayed_work gfx_off_delay_work; /* async work to set gfx block off */ 352 uint32_t gfx_off_residency; /* last logged residency */ 353 uint64_t gfx_off_entrycount; /* count of times GPU has get into GFXOFF state */ 354 355 /* pipe reservation */ 356 struct mutex pipe_reserve_mutex; 357 DECLARE_BITMAP (pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES); 358 359 /*ras */ 360 struct ras_common_if *ras_if; 361 struct amdgpu_gfx_ras *ras; 362 363 bool is_poweron; 364 365 struct amdgpu_ring sw_gfx_ring[AMDGPU_MAX_SW_GFX_RINGS]; 366 struct amdgpu_ring_mux muxer; 367 }; 368 369 #define amdgpu_gfx_get_gpu_clock_counter(adev) (adev)->gfx.funcs->get_gpu_clock_counter((adev)) 370 #define amdgpu_gfx_select_se_sh(adev, se, sh, instance) (adev)->gfx.funcs->select_se_sh((adev), (se), (sh), (instance)) 371 #define amdgpu_gfx_select_me_pipe_q(adev, me, pipe, q, vmid) (adev)->gfx.funcs->select_me_pipe_q((adev), (me), (pipe), (q), (vmid)) 372 #define amdgpu_gfx_init_spm_golden(adev) (adev)->gfx.funcs->init_spm_golden((adev)) 373 374 /** 375 * amdgpu_gfx_create_bitmask - create a bitmask 376 * 377 * @bit_width: length of the mask 378 * 379 * create a variable length bit mask. 380 * Returns the bitmask. 381 */ 382 static inline u32 amdgpu_gfx_create_bitmask(u32 bit_width) 383 { 384 return (u32)((1ULL << bit_width) - 1); 385 } 386 387 void amdgpu_gfx_parse_disable_cu(unsigned *mask, unsigned max_se, 388 unsigned max_sh); 389 390 int amdgpu_gfx_kiq_init_ring(struct amdgpu_device *adev, 391 struct amdgpu_ring *ring, 392 struct amdgpu_irq_src *irq); 393 394 void amdgpu_gfx_kiq_free_ring(struct amdgpu_ring *ring); 395 396 void amdgpu_gfx_kiq_fini(struct amdgpu_device *adev); 397 int amdgpu_gfx_kiq_init(struct amdgpu_device *adev, 398 unsigned hpd_size); 399 400 int amdgpu_gfx_mqd_sw_init(struct amdgpu_device *adev, 401 unsigned mqd_size); 402 void amdgpu_gfx_mqd_sw_fini(struct amdgpu_device *adev); 403 int amdgpu_gfx_disable_kcq(struct amdgpu_device *adev); 404 int amdgpu_gfx_enable_kcq(struct amdgpu_device *adev); 405 406 void amdgpu_gfx_compute_queue_acquire(struct amdgpu_device *adev); 407 void amdgpu_gfx_graphics_queue_acquire(struct amdgpu_device *adev); 408 409 int amdgpu_gfx_mec_queue_to_bit(struct amdgpu_device *adev, int mec, 410 int pipe, int queue); 411 void amdgpu_queue_mask_bit_to_mec_queue(struct amdgpu_device *adev, int bit, 412 int *mec, int *pipe, int *queue); 413 bool amdgpu_gfx_is_mec_queue_enabled(struct amdgpu_device *adev, int mec, 414 int pipe, int queue); 415 bool amdgpu_gfx_is_high_priority_compute_queue(struct amdgpu_device *adev, 416 struct amdgpu_ring *ring); 417 bool amdgpu_gfx_is_high_priority_graphics_queue(struct amdgpu_device *adev, 418 struct amdgpu_ring *ring); 419 int amdgpu_gfx_me_queue_to_bit(struct amdgpu_device *adev, int me, 420 int pipe, int queue); 421 void amdgpu_gfx_bit_to_me_queue(struct amdgpu_device *adev, int bit, 422 int *me, int *pipe, int *queue); 423 bool amdgpu_gfx_is_me_queue_enabled(struct amdgpu_device *adev, int me, 424 int pipe, int queue); 425 void amdgpu_gfx_off_ctrl(struct amdgpu_device *adev, bool enable); 426 int amdgpu_get_gfx_off_status(struct amdgpu_device *adev, uint32_t *value); 427 int amdgpu_gfx_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block); 428 void amdgpu_gfx_ras_fini(struct amdgpu_device *adev); 429 int amdgpu_get_gfx_off_entrycount(struct amdgpu_device *adev, u64 *value); 430 int amdgpu_get_gfx_off_residency(struct amdgpu_device *adev, u32 *residency); 431 int amdgpu_set_gfx_off_residency(struct amdgpu_device *adev, bool value); 432 int amdgpu_gfx_process_ras_data_cb(struct amdgpu_device *adev, 433 void *err_data, 434 struct amdgpu_iv_entry *entry); 435 int amdgpu_gfx_cp_ecc_error_irq(struct amdgpu_device *adev, 436 struct amdgpu_irq_src *source, 437 struct amdgpu_iv_entry *entry); 438 uint32_t amdgpu_kiq_rreg(struct amdgpu_device *adev, uint32_t reg); 439 void amdgpu_kiq_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v); 440 int amdgpu_gfx_get_num_kcq(struct amdgpu_device *adev); 441 void amdgpu_gfx_cp_init_microcode(struct amdgpu_device *adev, uint32_t ucode_id); 442 443 int amdgpu_gfx_ras_sw_init(struct amdgpu_device *adev); 444 int amdgpu_gfx_poison_consumption_handler(struct amdgpu_device *adev, 445 struct amdgpu_iv_entry *entry); 446 #endif 447