1 /* 2 * Copyright 2018 Advanced Micro Devices, Inc. 3 * All Rights Reserved. 4 * 5 * Permission is hereby granted, free of charge, to any person obtaining a 6 * copy of this software and associated documentation files (the 7 * "Software"), to deal in the Software without restriction, including 8 * without limitation the rights to use, copy, modify, merge, publish, 9 * distribute, sub license, and/or sell copies of the Software, and to 10 * permit persons to whom the Software is furnished to do so, subject to 11 * the following conditions: 12 * 13 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 14 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 15 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL 16 * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, 17 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 18 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 19 * USE OR OTHER DEALINGS IN THE SOFTWARE. 20 * 21 * The above copyright notice and this permission notice (including the 22 * next paragraph) shall be included in all copies or substantial portions 23 * of the Software. 24 * 25 */ 26 #ifndef __AMDGPU_GMC_H__ 27 #define __AMDGPU_GMC_H__ 28 29 #include <linux/types.h> 30 31 #include "amdgpu_irq.h" 32 #include "amdgpu_xgmi.h" 33 #include "amdgpu_ras.h" 34 35 /* VA hole for 48bit addresses on Vega10 */ 36 #define AMDGPU_GMC_HOLE_START 0x0000800000000000ULL 37 #define AMDGPU_GMC_HOLE_END 0xffff800000000000ULL 38 39 /* 40 * Hardware is programmed as if the hole doesn't exists with start and end 41 * address values. 42 * 43 * This mask is used to remove the upper 16bits of the VA and so come up with 44 * the linear addr value. 45 */ 46 #define AMDGPU_GMC_HOLE_MASK 0x0000ffffffffffffULL 47 48 /* 49 * Ring size as power of two for the log of recent faults. 50 */ 51 #define AMDGPU_GMC_FAULT_RING_ORDER 8 52 #define AMDGPU_GMC_FAULT_RING_SIZE (1 << AMDGPU_GMC_FAULT_RING_ORDER) 53 54 /* 55 * Hash size as power of two for the log of recent faults 56 */ 57 #define AMDGPU_GMC_FAULT_HASH_ORDER 8 58 #define AMDGPU_GMC_FAULT_HASH_SIZE (1 << AMDGPU_GMC_FAULT_HASH_ORDER) 59 60 /* 61 * Number of IH timestamp ticks until a fault is considered handled 62 */ 63 #define AMDGPU_GMC_FAULT_TIMEOUT 5000ULL 64 65 /* XNACK flags */ 66 #define AMDGPU_GMC_XNACK_FLAG_CHAIN BIT(0) 67 68 struct firmware; 69 70 enum amdgpu_memory_partition { 71 UNKNOWN_MEMORY_PARTITION_MODE = 0, 72 AMDGPU_NPS1_PARTITION_MODE = 1, 73 AMDGPU_NPS2_PARTITION_MODE = 2, 74 AMDGPU_NPS3_PARTITION_MODE = 3, 75 AMDGPU_NPS4_PARTITION_MODE = 4, 76 AMDGPU_NPS6_PARTITION_MODE = 6, 77 AMDGPU_NPS8_PARTITION_MODE = 8, 78 }; 79 80 #define AMDGPU_ALL_NPS_MASK \ 81 (BIT(AMDGPU_NPS1_PARTITION_MODE) | BIT(AMDGPU_NPS2_PARTITION_MODE) | \ 82 BIT(AMDGPU_NPS3_PARTITION_MODE) | BIT(AMDGPU_NPS4_PARTITION_MODE) | \ 83 BIT(AMDGPU_NPS6_PARTITION_MODE) | BIT(AMDGPU_NPS8_PARTITION_MODE)) 84 85 #define AMDGPU_GMC_INIT_RESET_NPS BIT(0) 86 87 #define AMDGPU_MAX_MEM_RANGES 8 88 89 #define AMDGPU_GMC9_FAULT_SOURCE_DATA_RETRY 0x80 90 #define AMDGPU_GMC9_FAULT_SOURCE_DATA_READ 0x40 91 #define AMDGPU_GMC9_FAULT_SOURCE_DATA_WRITE 0x20 92 #define AMDGPU_GMC9_FAULT_SOURCE_DATA_EXE 0x10 93 94 /* 95 * GMC page fault information 96 */ 97 struct amdgpu_gmc_fault { 98 uint64_t timestamp:48; 99 uint64_t next:AMDGPU_GMC_FAULT_RING_ORDER; 100 atomic64_t key; 101 uint64_t timestamp_expiry:48; 102 }; 103 104 /* 105 * VMHUB structures, functions & helpers 106 */ 107 struct amdgpu_vmhub_funcs { 108 void (*print_l2_protection_fault_status)(struct amdgpu_device *adev, 109 uint32_t status); 110 uint32_t (*get_invalidate_req)(unsigned int vmid, uint32_t flush_type); 111 }; 112 113 struct amdgpu_vmhub { 114 uint32_t ctx0_ptb_addr_lo32; 115 uint32_t ctx0_ptb_addr_hi32; 116 uint32_t vm_inv_eng0_sem; 117 uint32_t vm_inv_eng0_req; 118 uint32_t vm_inv_eng0_ack; 119 uint32_t vm_context0_cntl; 120 uint32_t vm_l2_pro_fault_status; 121 uint32_t vm_l2_pro_fault_cntl; 122 123 /* 124 * store the register distances between two continuous context domain 125 * and invalidation engine. 126 */ 127 uint32_t ctx_distance; 128 uint32_t ctx_addr_distance; /* include LO32/HI32 */ 129 uint32_t eng_distance; 130 uint32_t eng_addr_distance; /* include LO32/HI32 */ 131 132 uint32_t vm_cntx_cntl; 133 uint32_t vm_cntx_cntl_vm_fault; 134 uint32_t vm_l2_bank_select_reserved_cid2; 135 136 uint32_t vm_contexts_disable; 137 138 bool sdma_invalidation_workaround; 139 140 const struct amdgpu_vmhub_funcs *vmhub_funcs; 141 }; 142 143 /* 144 * GPU MC structures, functions & helpers 145 */ 146 struct amdgpu_gmc_funcs { 147 /* flush the vm tlb via mmio */ 148 void (*flush_gpu_tlb)(struct amdgpu_device *adev, uint32_t vmid, 149 uint32_t vmhub, uint32_t flush_type); 150 /* flush the vm tlb via pasid */ 151 void (*flush_gpu_tlb_pasid)(struct amdgpu_device *adev, uint16_t pasid, 152 uint32_t flush_type, bool all_hub, 153 uint32_t inst); 154 /* flush the vm tlb via ring */ 155 uint64_t (*emit_flush_gpu_tlb)(struct amdgpu_ring *ring, unsigned vmid, 156 uint64_t pd_addr); 157 /* Change the VMID -> PASID mapping */ 158 void (*emit_pasid_mapping)(struct amdgpu_ring *ring, unsigned vmid, 159 unsigned pasid); 160 /* enable/disable PRT support */ 161 void (*set_prt)(struct amdgpu_device *adev, bool enable); 162 /* get the pde for a given mc addr */ 163 void (*get_vm_pde)(struct amdgpu_device *adev, int level, 164 u64 *dst, u64 *flags); 165 /* get the pte flags to use for PTEs */ 166 void (*get_vm_pte)(struct amdgpu_device *adev, 167 struct amdgpu_vm *vm, 168 struct amdgpu_bo *bo, 169 uint32_t vm_flags, 170 uint64_t *pte_flags); 171 /* override per-page pte flags */ 172 void (*override_vm_pte_flags)(struct amdgpu_device *dev, 173 struct amdgpu_vm *vm, 174 uint64_t addr, uint64_t *flags); 175 /* get the amount of memory used by the vbios for pre-OS console */ 176 unsigned int (*get_vbios_fb_size)(struct amdgpu_device *adev); 177 /* get the DCC buffer alignment */ 178 unsigned int (*get_dcc_alignment)(struct amdgpu_device *adev); 179 180 enum amdgpu_memory_partition (*query_mem_partition_mode)( 181 struct amdgpu_device *adev); 182 /* Request NPS mode */ 183 int (*request_mem_partition_mode)(struct amdgpu_device *adev, 184 int nps_mode); 185 bool (*need_reset_on_init)(struct amdgpu_device *adev); 186 }; 187 188 struct amdgpu_mem_partition_info { 189 union { 190 struct { 191 uint32_t fpfn; 192 uint32_t lpfn; 193 } range; 194 struct { 195 int node; 196 } numa; 197 }; 198 uint64_t size; 199 }; 200 201 #define INVALID_PFN -1 202 203 struct amdgpu_gmc_memrange { 204 uint64_t base_address; 205 uint64_t limit_address; 206 uint32_t flags; 207 int nid_mask; 208 }; 209 210 enum amdgpu_gart_placement { 211 AMDGPU_GART_PLACEMENT_BEST_FIT = 0, 212 AMDGPU_GART_PLACEMENT_HIGH, 213 AMDGPU_GART_PLACEMENT_LOW, 214 }; 215 216 struct amdgpu_gmc { 217 /* FB's physical address in MMIO space (for CPU to 218 * map FB). This is different compared to the agp/ 219 * gart/vram_start/end field as the later is from 220 * GPU's view and aper_base is from CPU's view. 221 */ 222 resource_size_t aper_size; 223 resource_size_t aper_base; 224 /* for some chips with <= 32MB we need to lie 225 * about vram size near mc fb location */ 226 u64 mc_vram_size; 227 u64 visible_vram_size; 228 /* AGP aperture start and end in MC address space 229 * Driver find a hole in the MC address space 230 * to place AGP by setting MC_VM_AGP_BOT/TOP registers 231 * Under VMID0, logical address == MC address. AGP 232 * aperture maps to physical bus or IOVA addressed. 233 * AGP aperture is used to simulate FB in ZFB case. 234 * AGP aperture is also used for page table in system 235 * memory (mainly for APU). 236 * 237 */ 238 u64 agp_size; 239 u64 agp_start; 240 u64 agp_end; 241 /* GART aperture start and end in MC address space 242 * Driver find a hole in the MC address space 243 * to place GART by setting VM_CONTEXT0_PAGE_TABLE_START/END_ADDR 244 * registers 245 * Under VMID0, logical address inside GART aperture will 246 * be translated through gpuvm gart page table to access 247 * paged system memory 248 */ 249 u64 gart_size; 250 u64 gart_start; 251 u64 gart_end; 252 /* Frame buffer aperture of this GPU device. Different from 253 * fb_start (see below), this only covers the local GPU device. 254 * If driver uses FB aperture to access FB, driver get fb_start from 255 * MC_VM_FB_LOCATION_BASE (set by vbios) and calculate vram_start 256 * of this local device by adding an offset inside the XGMI hive. 257 * If driver uses GART table for VMID0 FB access, driver finds a hole in 258 * VMID0's virtual address space to place the SYSVM aperture inside 259 * which the first part is vram and the second part is gart (covering 260 * system ram). 261 */ 262 u64 vram_start; 263 u64 vram_end; 264 /* FB region , it's same as local vram region in single GPU, in XGMI 265 * configuration, this region covers all GPUs in the same hive , 266 * each GPU in the hive has the same view of this FB region . 267 * GPU0's vram starts at offset (0 * segment size) , 268 * GPU1 starts at offset (1 * segment size), etc. 269 */ 270 u64 fb_start; 271 u64 fb_end; 272 unsigned vram_width; 273 u64 real_vram_size; 274 int vram_mtrr; 275 u64 mc_mask; 276 const struct firmware *fw; /* MC firmware */ 277 uint32_t fw_version; 278 struct amdgpu_irq_src vm_fault; 279 uint32_t vram_type; 280 uint8_t vram_vendor; 281 uint32_t srbm_soft_reset; 282 bool prt_warning; 283 uint32_t sdpif_register; 284 /* apertures */ 285 u64 shared_aperture_start; 286 u64 shared_aperture_end; 287 u64 private_aperture_start; 288 u64 private_aperture_end; 289 /* protects concurrent invalidation */ 290 spinlock_t invalidate_lock; 291 bool translate_further; 292 struct kfd_vm_fault_info *vm_fault_info; 293 atomic_t vm_fault_info_updated; 294 295 struct amdgpu_gmc_fault fault_ring[AMDGPU_GMC_FAULT_RING_SIZE]; 296 struct { 297 uint64_t idx:AMDGPU_GMC_FAULT_RING_ORDER; 298 } fault_hash[AMDGPU_GMC_FAULT_HASH_SIZE]; 299 uint64_t last_fault:AMDGPU_GMC_FAULT_RING_ORDER; 300 301 bool tmz_enabled; 302 bool is_app_apu; 303 304 struct amdgpu_mem_partition_info *mem_partitions; 305 uint8_t num_mem_partitions; 306 const struct amdgpu_gmc_funcs *gmc_funcs; 307 enum amdgpu_memory_partition requested_nps_mode; 308 uint32_t supported_nps_modes; 309 uint32_t reset_flags; 310 311 struct amdgpu_xgmi xgmi; 312 struct amdgpu_irq_src ecc_irq; 313 int noretry; 314 uint32_t xnack_flags; 315 316 uint32_t vmid0_page_table_block_size; 317 uint32_t vmid0_page_table_depth; 318 struct amdgpu_bo *pdb0_bo; 319 /* CPU kmapped address of pdb0*/ 320 void *ptr_pdb0; 321 322 /* MALL size */ 323 u64 mall_size; 324 uint32_t m_half_use; 325 326 /* number of UMC instances */ 327 int num_umc; 328 /* mode2 save restore */ 329 u64 VM_L2_CNTL; 330 u64 VM_L2_CNTL2; 331 u64 VM_DUMMY_PAGE_FAULT_CNTL; 332 u64 VM_DUMMY_PAGE_FAULT_ADDR_LO32; 333 u64 VM_DUMMY_PAGE_FAULT_ADDR_HI32; 334 u64 VM_L2_PROTECTION_FAULT_CNTL; 335 u64 VM_L2_PROTECTION_FAULT_CNTL2; 336 u64 VM_L2_PROTECTION_FAULT_MM_CNTL3; 337 u64 VM_L2_PROTECTION_FAULT_MM_CNTL4; 338 u64 VM_L2_PROTECTION_FAULT_ADDR_LO32; 339 u64 VM_L2_PROTECTION_FAULT_ADDR_HI32; 340 u64 VM_DEBUG; 341 u64 VM_L2_MM_GROUP_RT_CLASSES; 342 u64 VM_L2_BANK_SELECT_RESERVED_CID; 343 u64 VM_L2_BANK_SELECT_RESERVED_CID2; 344 u64 VM_L2_CACHE_PARITY_CNTL; 345 u64 VM_L2_IH_LOG_CNTL; 346 u64 VM_CONTEXT_CNTL[16]; 347 u64 VM_CONTEXT_PAGE_TABLE_BASE_ADDR_LO32[16]; 348 u64 VM_CONTEXT_PAGE_TABLE_BASE_ADDR_HI32[16]; 349 u64 VM_CONTEXT_PAGE_TABLE_START_ADDR_LO32[16]; 350 u64 VM_CONTEXT_PAGE_TABLE_START_ADDR_HI32[16]; 351 u64 VM_CONTEXT_PAGE_TABLE_END_ADDR_LO32[16]; 352 u64 VM_CONTEXT_PAGE_TABLE_END_ADDR_HI32[16]; 353 u64 MC_VM_MX_L1_TLB_CNTL; 354 355 u64 noretry_flags; 356 357 bool flush_tlb_needs_extra_type_0; 358 bool flush_tlb_needs_extra_type_2; 359 bool flush_pasid_uses_kiq; 360 }; 361 362 #define amdgpu_gmc_emit_flush_gpu_tlb(r, vmid, addr) (r)->adev->gmc.gmc_funcs->emit_flush_gpu_tlb((r), (vmid), (addr)) 363 #define amdgpu_gmc_emit_pasid_mapping(r, vmid, pasid) (r)->adev->gmc.gmc_funcs->emit_pasid_mapping((r), (vmid), (pasid)) 364 #define amdgpu_gmc_get_vm_pde(adev, level, dst, flags) (adev)->gmc.gmc_funcs->get_vm_pde((adev), (level), (dst), (flags)) 365 #define amdgpu_gmc_get_vm_pte(adev, vm, bo, vm_flags, pte_flags) \ 366 ((adev)->gmc.gmc_funcs->get_vm_pte((adev), (vm), (bo), (vm_flags), \ 367 (pte_flags))) 368 #define amdgpu_gmc_override_vm_pte_flags(adev, vm, addr, pte_flags) \ 369 (adev)->gmc.gmc_funcs->override_vm_pte_flags \ 370 ((adev), (vm), (addr), (pte_flags)) 371 #define amdgpu_gmc_get_vbios_fb_size(adev) (adev)->gmc.gmc_funcs->get_vbios_fb_size((adev)) 372 #define amdgpu_gmc_get_dcc_alignment(adev) ({ \ 373 typeof(adev) _adev = (adev); \ 374 _adev->gmc.gmc_funcs->get_dcc_alignment(_adev); \ 375 }) 376 377 /** 378 * amdgpu_gmc_vram_full_visible - Check if full VRAM is visible through the BAR 379 * 380 * @adev: amdgpu_device pointer 381 * 382 * Returns: 383 * True if full VRAM is visible through the BAR 384 */ 385 static inline bool amdgpu_gmc_vram_full_visible(struct amdgpu_gmc *gmc) 386 { 387 WARN_ON(gmc->real_vram_size < gmc->visible_vram_size); 388 389 return (gmc->real_vram_size == gmc->visible_vram_size); 390 } 391 392 /** 393 * amdgpu_gmc_sign_extend - sign extend the given gmc address 394 * 395 * @addr: address to extend 396 */ 397 static inline uint64_t amdgpu_gmc_sign_extend(uint64_t addr) 398 { 399 if (addr >= AMDGPU_GMC_HOLE_START) 400 addr |= AMDGPU_GMC_HOLE_END; 401 402 return addr; 403 } 404 405 bool amdgpu_gmc_is_pdb0_enabled(struct amdgpu_device *adev); 406 int amdgpu_gmc_pdb0_alloc(struct amdgpu_device *adev); 407 void amdgpu_gmc_get_pde_for_bo(struct amdgpu_bo *bo, int level, 408 uint64_t *addr, uint64_t *flags); 409 int amdgpu_gmc_set_pte_pde(struct amdgpu_device *adev, void *cpu_pt_addr, 410 uint32_t gpu_page_idx, uint64_t addr, 411 uint64_t flags); 412 uint64_t amdgpu_gmc_pd_addr(struct amdgpu_bo *bo); 413 uint64_t amdgpu_gmc_agp_addr(struct ttm_buffer_object *bo); 414 void amdgpu_gmc_sysvm_location(struct amdgpu_device *adev, struct amdgpu_gmc *mc); 415 void amdgpu_gmc_vram_location(struct amdgpu_device *adev, struct amdgpu_gmc *mc, 416 u64 base); 417 void amdgpu_gmc_gart_location(struct amdgpu_device *adev, 418 struct amdgpu_gmc *mc, 419 enum amdgpu_gart_placement gart_placement); 420 void amdgpu_gmc_agp_location(struct amdgpu_device *adev, 421 struct amdgpu_gmc *mc); 422 void amdgpu_gmc_set_agp_default(struct amdgpu_device *adev, 423 struct amdgpu_gmc *mc); 424 bool amdgpu_gmc_filter_faults(struct amdgpu_device *adev, 425 struct amdgpu_ih_ring *ih, uint64_t addr, 426 uint16_t pasid, uint64_t timestamp); 427 void amdgpu_gmc_filter_faults_remove(struct amdgpu_device *adev, uint64_t addr, 428 uint16_t pasid); 429 int amdgpu_gmc_ras_sw_init(struct amdgpu_device *adev); 430 int amdgpu_gmc_ras_late_init(struct amdgpu_device *adev); 431 void amdgpu_gmc_ras_fini(struct amdgpu_device *adev); 432 int amdgpu_gmc_allocate_vm_inv_eng(struct amdgpu_device *adev); 433 void amdgpu_gmc_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid, 434 uint32_t vmhub, uint32_t flush_type); 435 int amdgpu_gmc_flush_gpu_tlb_pasid(struct amdgpu_device *adev, uint16_t pasid, 436 uint32_t flush_type, bool all_hub, 437 uint32_t inst); 438 void amdgpu_gmc_fw_reg_write_reg_wait(struct amdgpu_device *adev, 439 uint32_t reg0, uint32_t reg1, 440 uint32_t ref, uint32_t mask, 441 uint32_t xcc_inst); 442 443 extern void amdgpu_gmc_tmz_set(struct amdgpu_device *adev); 444 extern void amdgpu_gmc_noretry_set(struct amdgpu_device *adev); 445 446 extern void 447 amdgpu_gmc_set_vm_fault_masks(struct amdgpu_device *adev, int hub_type, 448 bool enable); 449 450 void amdgpu_gmc_get_vbios_allocations(struct amdgpu_device *adev); 451 452 void amdgpu_gmc_init_pdb0(struct amdgpu_device *adev); 453 uint64_t amdgpu_gmc_vram_mc2pa(struct amdgpu_device *adev, uint64_t mc_addr); 454 uint64_t amdgpu_gmc_vram_pa(struct amdgpu_device *adev, struct amdgpu_bo *bo); 455 int amdgpu_gmc_vram_checking(struct amdgpu_device *adev); 456 int amdgpu_gmc_sysfs_init(struct amdgpu_device *adev); 457 void amdgpu_gmc_sysfs_fini(struct amdgpu_device *adev); 458 459 int amdgpu_gmc_get_nps_memranges(struct amdgpu_device *adev, 460 struct amdgpu_mem_partition_info *mem_ranges, 461 uint8_t *exp_ranges); 462 463 int amdgpu_gmc_request_memory_partition(struct amdgpu_device *adev, 464 int nps_mode); 465 void amdgpu_gmc_prepare_nps_mode_change(struct amdgpu_device *adev); 466 bool amdgpu_gmc_need_reset_on_init(struct amdgpu_device *adev); 467 enum amdgpu_memory_partition 468 amdgpu_gmc_get_vf_memory_partition(struct amdgpu_device *adev); 469 enum amdgpu_memory_partition 470 amdgpu_gmc_get_memory_partition(struct amdgpu_device *adev, u32 *supp_modes); 471 enum amdgpu_memory_partition 472 amdgpu_gmc_query_memory_partition(struct amdgpu_device *adev); 473 int amdgpu_gmc_init_mem_ranges(struct amdgpu_device *adev); 474 void amdgpu_gmc_init_sw_mem_ranges(struct amdgpu_device *adev, 475 struct amdgpu_mem_partition_info *mem_ranges); 476 #endif 477