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 #ifndef KFD_PRIV_H_INCLUDED 24 #define KFD_PRIV_H_INCLUDED 25 26 #include <linux/hashtable.h> 27 #include <linux/mmu_notifier.h> 28 #include <linux/mutex.h> 29 #include <linux/types.h> 30 #include <linux/atomic.h> 31 #include <linux/workqueue.h> 32 #include <linux/spinlock.h> 33 #include <linux/kfd_ioctl.h> 34 #include <linux/idr.h> 35 #include <linux/kfifo.h> 36 #include <linux/seq_file.h> 37 #include <linux/kref.h> 38 #include <linux/sysfs.h> 39 #include <linux/device_cgroup.h> 40 #include <drm/drm_file.h> 41 #include <drm/drm_drv.h> 42 #include <drm/drm_device.h> 43 #include <kgd_kfd_interface.h> 44 #include <linux/swap.h> 45 46 #include "amd_shared.h" 47 48 #define KFD_MAX_RING_ENTRY_SIZE 8 49 50 #define KFD_SYSFS_FILE_MODE 0444 51 52 /* GPU ID hash width in bits */ 53 #define KFD_GPU_ID_HASH_WIDTH 16 54 55 /* Use upper bits of mmap offset to store KFD driver specific information. 56 * BITS[63:62] - Encode MMAP type 57 * BITS[61:46] - Encode gpu_id. To identify to which GPU the offset belongs to 58 * BITS[45:0] - MMAP offset value 59 * 60 * NOTE: struct vm_area_struct.vm_pgoff uses offset in pages. Hence, these 61 * defines are w.r.t to PAGE_SIZE 62 */ 63 #define KFD_MMAP_TYPE_SHIFT 62 64 #define KFD_MMAP_TYPE_MASK (0x3ULL << KFD_MMAP_TYPE_SHIFT) 65 #define KFD_MMAP_TYPE_DOORBELL (0x3ULL << KFD_MMAP_TYPE_SHIFT) 66 #define KFD_MMAP_TYPE_EVENTS (0x2ULL << KFD_MMAP_TYPE_SHIFT) 67 #define KFD_MMAP_TYPE_RESERVED_MEM (0x1ULL << KFD_MMAP_TYPE_SHIFT) 68 #define KFD_MMAP_TYPE_MMIO (0x0ULL << KFD_MMAP_TYPE_SHIFT) 69 70 #define KFD_MMAP_GPU_ID_SHIFT 46 71 #define KFD_MMAP_GPU_ID_MASK (((1ULL << KFD_GPU_ID_HASH_WIDTH) - 1) \ 72 << KFD_MMAP_GPU_ID_SHIFT) 73 #define KFD_MMAP_GPU_ID(gpu_id) ((((uint64_t)gpu_id) << KFD_MMAP_GPU_ID_SHIFT)\ 74 & KFD_MMAP_GPU_ID_MASK) 75 #define KFD_MMAP_GET_GPU_ID(offset) ((offset & KFD_MMAP_GPU_ID_MASK) \ 76 >> KFD_MMAP_GPU_ID_SHIFT) 77 78 /* 79 * When working with cp scheduler we should assign the HIQ manually or via 80 * the amdgpu driver to a fixed hqd slot, here are the fixed HIQ hqd slot 81 * definitions for Kaveri. In Kaveri only the first ME queues participates 82 * in the cp scheduling taking that in mind we set the HIQ slot in the 83 * second ME. 84 */ 85 #define KFD_CIK_HIQ_PIPE 4 86 #define KFD_CIK_HIQ_QUEUE 0 87 88 /* Macro for allocating structures */ 89 #define kfd_alloc_struct(ptr_to_struct) \ 90 ((typeof(ptr_to_struct)) kzalloc(sizeof(*ptr_to_struct), GFP_KERNEL)) 91 92 #define KFD_MAX_NUM_OF_PROCESSES 512 93 #define KFD_MAX_NUM_OF_QUEUES_PER_PROCESS 1024 94 95 /* 96 * Size of the per-process TBA+TMA buffer: 2 pages 97 * 98 * The first page is the TBA used for the CWSR ISA code. The second 99 * page is used as TMA for daisy changing a user-mode trap handler. 100 */ 101 #define KFD_CWSR_TBA_TMA_SIZE (PAGE_SIZE * 2) 102 #define KFD_CWSR_TMA_OFFSET PAGE_SIZE 103 104 #define KFD_MAX_NUM_OF_QUEUES_PER_DEVICE \ 105 (KFD_MAX_NUM_OF_PROCESSES * \ 106 KFD_MAX_NUM_OF_QUEUES_PER_PROCESS) 107 108 #define KFD_KERNEL_QUEUE_SIZE 2048 109 110 #define KFD_UNMAP_LATENCY_MS (4000) 111 112 /* 113 * 512 = 0x200 114 * The doorbell index distance between SDMA RLC (2*i) and (2*i+1) in the 115 * same SDMA engine on SOC15, which has 8-byte doorbells for SDMA. 116 * 512 8-byte doorbell distance (i.e. one page away) ensures that SDMA RLC 117 * (2*i+1) doorbells (in terms of the lower 12 bit address) lie exactly in 118 * the OFFSET and SIZE set in registers like BIF_SDMA0_DOORBELL_RANGE. 119 */ 120 #define KFD_QUEUE_DOORBELL_MIRROR_OFFSET 512 121 122 123 /* 124 * Kernel module parameter to specify maximum number of supported queues per 125 * device 126 */ 127 extern int max_num_of_queues_per_device; 128 129 130 /* Kernel module parameter to specify the scheduling policy */ 131 extern int sched_policy; 132 133 /* 134 * Kernel module parameter to specify the maximum process 135 * number per HW scheduler 136 */ 137 extern int hws_max_conc_proc; 138 139 extern int cwsr_enable; 140 141 /* 142 * Kernel module parameter to specify whether to send sigterm to HSA process on 143 * unhandled exception 144 */ 145 extern int send_sigterm; 146 147 /* 148 * This kernel module is used to simulate large bar machine on non-large bar 149 * enabled machines. 150 */ 151 extern int debug_largebar; 152 153 /* 154 * Ignore CRAT table during KFD initialization, can be used to work around 155 * broken CRAT tables on some AMD systems 156 */ 157 extern int ignore_crat; 158 159 /* 160 * Set sh_mem_config.retry_disable on Vega10 161 */ 162 extern int amdgpu_noretry; 163 164 /* 165 * Halt if HWS hang is detected 166 */ 167 extern int halt_if_hws_hang; 168 169 /* 170 * Whether MEC FW support GWS barriers 171 */ 172 extern bool hws_gws_support; 173 174 /* 175 * Queue preemption timeout in ms 176 */ 177 extern int queue_preemption_timeout_ms; 178 179 enum cache_policy { 180 cache_policy_coherent, 181 cache_policy_noncoherent 182 }; 183 184 #define KFD_IS_SOC15(chip) ((chip) >= CHIP_VEGA10) 185 186 struct kfd_event_interrupt_class { 187 bool (*interrupt_isr)(struct kfd_dev *dev, 188 const uint32_t *ih_ring_entry, uint32_t *patched_ihre, 189 bool *patched_flag); 190 void (*interrupt_wq)(struct kfd_dev *dev, 191 const uint32_t *ih_ring_entry); 192 }; 193 194 struct kfd_device_info { 195 enum amd_asic_type asic_family; 196 const char *asic_name; 197 const struct kfd_event_interrupt_class *event_interrupt_class; 198 unsigned int max_pasid_bits; 199 unsigned int max_no_of_hqd; 200 unsigned int doorbell_size; 201 size_t ih_ring_entry_size; 202 uint8_t num_of_watch_points; 203 uint16_t mqd_size_aligned; 204 bool supports_cwsr; 205 bool needs_iommu_device; 206 bool needs_pci_atomics; 207 unsigned int num_sdma_engines; 208 unsigned int num_xgmi_sdma_engines; 209 unsigned int num_sdma_queues_per_engine; 210 }; 211 212 struct kfd_mem_obj { 213 uint32_t range_start; 214 uint32_t range_end; 215 uint64_t gpu_addr; 216 uint32_t *cpu_ptr; 217 void *gtt_mem; 218 }; 219 220 struct kfd_vmid_info { 221 uint32_t first_vmid_kfd; 222 uint32_t last_vmid_kfd; 223 uint32_t vmid_num_kfd; 224 }; 225 226 struct kfd_dev { 227 struct kgd_dev *kgd; 228 229 const struct kfd_device_info *device_info; 230 struct pci_dev *pdev; 231 struct drm_device *ddev; 232 233 unsigned int id; /* topology stub index */ 234 235 phys_addr_t doorbell_base; /* Start of actual doorbells used by 236 * KFD. It is aligned for mapping 237 * into user mode 238 */ 239 size_t doorbell_base_dw_offset; /* Offset from the start of the PCI 240 * doorbell BAR to the first KFD 241 * doorbell in dwords. GFX reserves 242 * the segment before this offset. 243 */ 244 u32 __iomem *doorbell_kernel_ptr; /* This is a pointer for a doorbells 245 * page used by kernel queue 246 */ 247 248 struct kgd2kfd_shared_resources shared_resources; 249 struct kfd_vmid_info vm_info; 250 251 const struct kfd2kgd_calls *kfd2kgd; 252 struct mutex doorbell_mutex; 253 DECLARE_BITMAP(doorbell_available_index, 254 KFD_MAX_NUM_OF_QUEUES_PER_PROCESS); 255 256 void *gtt_mem; 257 uint64_t gtt_start_gpu_addr; 258 void *gtt_start_cpu_ptr; 259 void *gtt_sa_bitmap; 260 struct mutex gtt_sa_lock; 261 unsigned int gtt_sa_chunk_size; 262 unsigned int gtt_sa_num_of_chunks; 263 264 /* Interrupts */ 265 struct kfifo ih_fifo; 266 struct workqueue_struct *ih_wq; 267 struct work_struct interrupt_work; 268 spinlock_t interrupt_lock; 269 270 /* QCM Device instance */ 271 struct device_queue_manager *dqm; 272 273 bool init_complete; 274 /* 275 * Interrupts of interest to KFD are copied 276 * from the HW ring into a SW ring. 277 */ 278 bool interrupts_active; 279 280 /* Debug manager */ 281 struct kfd_dbgmgr *dbgmgr; 282 283 /* Firmware versions */ 284 uint16_t mec_fw_version; 285 uint16_t mec2_fw_version; 286 uint16_t sdma_fw_version; 287 288 /* Maximum process number mapped to HW scheduler */ 289 unsigned int max_proc_per_quantum; 290 291 /* CWSR */ 292 bool cwsr_enabled; 293 const void *cwsr_isa; 294 unsigned int cwsr_isa_size; 295 296 /* xGMI */ 297 uint64_t hive_id; 298 299 /* UUID */ 300 uint64_t unique_id; 301 302 bool pci_atomic_requested; 303 304 /* SRAM ECC flag */ 305 atomic_t sram_ecc_flag; 306 307 /* Compute Profile ref. count */ 308 atomic_t compute_profile; 309 310 /* Global GWS resource shared b/t processes*/ 311 void *gws; 312 }; 313 314 enum kfd_mempool { 315 KFD_MEMPOOL_SYSTEM_CACHEABLE = 1, 316 KFD_MEMPOOL_SYSTEM_WRITECOMBINE = 2, 317 KFD_MEMPOOL_FRAMEBUFFER = 3, 318 }; 319 320 /* Character device interface */ 321 int kfd_chardev_init(void); 322 void kfd_chardev_exit(void); 323 struct device *kfd_chardev(void); 324 325 /** 326 * enum kfd_unmap_queues_filter 327 * 328 * @KFD_UNMAP_QUEUES_FILTER_SINGLE_QUEUE: Preempts single queue. 329 * 330 * @KFD_UNMAP_QUEUES_FILTER_ALL_QUEUES: Preempts all queues in the 331 * running queues list. 332 * 333 * @KFD_UNMAP_QUEUES_FILTER_BY_PASID: Preempts queues that belongs to 334 * specific process. 335 * 336 */ 337 enum kfd_unmap_queues_filter { 338 KFD_UNMAP_QUEUES_FILTER_SINGLE_QUEUE, 339 KFD_UNMAP_QUEUES_FILTER_ALL_QUEUES, 340 KFD_UNMAP_QUEUES_FILTER_DYNAMIC_QUEUES, 341 KFD_UNMAP_QUEUES_FILTER_BY_PASID 342 }; 343 344 /** 345 * enum kfd_queue_type 346 * 347 * @KFD_QUEUE_TYPE_COMPUTE: Regular user mode queue type. 348 * 349 * @KFD_QUEUE_TYPE_SDMA: Sdma user mode queue type. 350 * 351 * @KFD_QUEUE_TYPE_HIQ: HIQ queue type. 352 * 353 * @KFD_QUEUE_TYPE_DIQ: DIQ queue type. 354 */ 355 enum kfd_queue_type { 356 KFD_QUEUE_TYPE_COMPUTE, 357 KFD_QUEUE_TYPE_SDMA, 358 KFD_QUEUE_TYPE_HIQ, 359 KFD_QUEUE_TYPE_DIQ, 360 KFD_QUEUE_TYPE_SDMA_XGMI 361 }; 362 363 enum kfd_queue_format { 364 KFD_QUEUE_FORMAT_PM4, 365 KFD_QUEUE_FORMAT_AQL 366 }; 367 368 enum KFD_QUEUE_PRIORITY { 369 KFD_QUEUE_PRIORITY_MINIMUM = 0, 370 KFD_QUEUE_PRIORITY_MAXIMUM = 15 371 }; 372 373 /** 374 * struct queue_properties 375 * 376 * @type: The queue type. 377 * 378 * @queue_id: Queue identifier. 379 * 380 * @queue_address: Queue ring buffer address. 381 * 382 * @queue_size: Queue ring buffer size. 383 * 384 * @priority: Defines the queue priority relative to other queues in the 385 * process. 386 * This is just an indication and HW scheduling may override the priority as 387 * necessary while keeping the relative prioritization. 388 * the priority granularity is from 0 to f which f is the highest priority. 389 * currently all queues are initialized with the highest priority. 390 * 391 * @queue_percent: This field is partially implemented and currently a zero in 392 * this field defines that the queue is non active. 393 * 394 * @read_ptr: User space address which points to the number of dwords the 395 * cp read from the ring buffer. This field updates automatically by the H/W. 396 * 397 * @write_ptr: Defines the number of dwords written to the ring buffer. 398 * 399 * @doorbell_ptr: This field aim is to notify the H/W of new packet written to 400 * the queue ring buffer. This field should be similar to write_ptr and the 401 * user should update this field after he updated the write_ptr. 402 * 403 * @doorbell_off: The doorbell offset in the doorbell pci-bar. 404 * 405 * @is_interop: Defines if this is a interop queue. Interop queue means that 406 * the queue can access both graphics and compute resources. 407 * 408 * @is_evicted: Defines if the queue is evicted. Only active queues 409 * are evicted, rendering them inactive. 410 * 411 * @is_active: Defines if the queue is active or not. @is_active and 412 * @is_evicted are protected by the DQM lock. 413 * 414 * @is_gws: Defines if the queue has been updated to be GWS-capable or not. 415 * @is_gws should be protected by the DQM lock, since changing it can yield the 416 * possibility of updating DQM state on number of GWS queues. 417 * 418 * @vmid: If the scheduling mode is no cp scheduling the field defines the vmid 419 * of the queue. 420 * 421 * This structure represents the queue properties for each queue no matter if 422 * it's user mode or kernel mode queue. 423 * 424 */ 425 struct queue_properties { 426 enum kfd_queue_type type; 427 enum kfd_queue_format format; 428 unsigned int queue_id; 429 uint64_t queue_address; 430 uint64_t queue_size; 431 uint32_t priority; 432 uint32_t queue_percent; 433 uint32_t *read_ptr; 434 uint32_t *write_ptr; 435 void __iomem *doorbell_ptr; 436 uint32_t doorbell_off; 437 bool is_interop; 438 bool is_evicted; 439 bool is_active; 440 bool is_gws; 441 /* Not relevant for user mode queues in cp scheduling */ 442 unsigned int vmid; 443 /* Relevant only for sdma queues*/ 444 uint32_t sdma_engine_id; 445 uint32_t sdma_queue_id; 446 uint32_t sdma_vm_addr; 447 /* Relevant only for VI */ 448 uint64_t eop_ring_buffer_address; 449 uint32_t eop_ring_buffer_size; 450 uint64_t ctx_save_restore_area_address; 451 uint32_t ctx_save_restore_area_size; 452 uint32_t ctl_stack_size; 453 uint64_t tba_addr; 454 uint64_t tma_addr; 455 /* Relevant for CU */ 456 uint32_t cu_mask_count; /* Must be a multiple of 32 */ 457 uint32_t *cu_mask; 458 }; 459 460 #define QUEUE_IS_ACTIVE(q) ((q).queue_size > 0 && \ 461 (q).queue_address != 0 && \ 462 (q).queue_percent > 0 && \ 463 !(q).is_evicted) 464 465 /** 466 * struct queue 467 * 468 * @list: Queue linked list. 469 * 470 * @mqd: The queue MQD. 471 * 472 * @mqd_mem_obj: The MQD local gpu memory object. 473 * 474 * @gart_mqd_addr: The MQD gart mc address. 475 * 476 * @properties: The queue properties. 477 * 478 * @mec: Used only in no cp scheduling mode and identifies to micro engine id 479 * that the queue should be execute on. 480 * 481 * @pipe: Used only in no cp scheduling mode and identifies the queue's pipe 482 * id. 483 * 484 * @queue: Used only in no cp scheduliong mode and identifies the queue's slot. 485 * 486 * @process: The kfd process that created this queue. 487 * 488 * @device: The kfd device that created this queue. 489 * 490 * @gws: Pointing to gws kgd_mem if this is a gws control queue; NULL 491 * otherwise. 492 * 493 * This structure represents user mode compute queues. 494 * It contains all the necessary data to handle such queues. 495 * 496 */ 497 498 struct queue { 499 struct list_head list; 500 void *mqd; 501 struct kfd_mem_obj *mqd_mem_obj; 502 uint64_t gart_mqd_addr; 503 struct queue_properties properties; 504 505 uint32_t mec; 506 uint32_t pipe; 507 uint32_t queue; 508 509 unsigned int sdma_id; 510 unsigned int doorbell_id; 511 512 struct kfd_process *process; 513 struct kfd_dev *device; 514 void *gws; 515 516 /* procfs */ 517 struct kobject kobj; 518 }; 519 520 /* 521 * Please read the kfd_mqd_manager.h description. 522 */ 523 enum KFD_MQD_TYPE { 524 KFD_MQD_TYPE_HIQ = 0, /* for hiq */ 525 KFD_MQD_TYPE_CP, /* for cp queues and diq */ 526 KFD_MQD_TYPE_SDMA, /* for sdma queues */ 527 KFD_MQD_TYPE_DIQ, /* for diq */ 528 KFD_MQD_TYPE_MAX 529 }; 530 531 enum KFD_PIPE_PRIORITY { 532 KFD_PIPE_PRIORITY_CS_LOW = 0, 533 KFD_PIPE_PRIORITY_CS_MEDIUM, 534 KFD_PIPE_PRIORITY_CS_HIGH 535 }; 536 537 struct scheduling_resources { 538 unsigned int vmid_mask; 539 enum kfd_queue_type type; 540 uint64_t queue_mask; 541 uint64_t gws_mask; 542 uint32_t oac_mask; 543 uint32_t gds_heap_base; 544 uint32_t gds_heap_size; 545 }; 546 547 struct process_queue_manager { 548 /* data */ 549 struct kfd_process *process; 550 struct list_head queues; 551 unsigned long *queue_slot_bitmap; 552 }; 553 554 struct qcm_process_device { 555 /* The Device Queue Manager that owns this data */ 556 struct device_queue_manager *dqm; 557 struct process_queue_manager *pqm; 558 /* Queues list */ 559 struct list_head queues_list; 560 struct list_head priv_queue_list; 561 562 unsigned int queue_count; 563 unsigned int vmid; 564 bool is_debug; 565 unsigned int evicted; /* eviction counter, 0=active */ 566 567 /* This flag tells if we should reset all wavefronts on 568 * process termination 569 */ 570 bool reset_wavefronts; 571 572 /* This flag tells us if this process has a GWS-capable 573 * queue that will be mapped into the runlist. It's 574 * possible to request a GWS BO, but not have the queue 575 * currently mapped, and this changes how the MAP_PROCESS 576 * PM4 packet is configured. 577 */ 578 bool mapped_gws_queue; 579 580 /* 581 * All the memory management data should be here too 582 */ 583 uint64_t gds_context_area; 584 /* Contains page table flags such as AMDGPU_PTE_VALID since gfx9 */ 585 uint64_t page_table_base; 586 uint32_t sh_mem_config; 587 uint32_t sh_mem_bases; 588 uint32_t sh_mem_ape1_base; 589 uint32_t sh_mem_ape1_limit; 590 uint32_t gds_size; 591 uint32_t num_gws; 592 uint32_t num_oac; 593 uint32_t sh_hidden_private_base; 594 595 /* CWSR memory */ 596 void *cwsr_kaddr; 597 uint64_t cwsr_base; 598 uint64_t tba_addr; 599 uint64_t tma_addr; 600 601 /* IB memory */ 602 uint64_t ib_base; 603 void *ib_kaddr; 604 605 /* doorbell resources per process per device */ 606 unsigned long *doorbell_bitmap; 607 }; 608 609 /* KFD Memory Eviction */ 610 611 /* Approx. wait time before attempting to restore evicted BOs */ 612 #define PROCESS_RESTORE_TIME_MS 100 613 /* Approx. back off time if restore fails due to lack of memory */ 614 #define PROCESS_BACK_OFF_TIME_MS 100 615 /* Approx. time before evicting the process again */ 616 #define PROCESS_ACTIVE_TIME_MS 10 617 618 /* 8 byte handle containing GPU ID in the most significant 4 bytes and 619 * idr_handle in the least significant 4 bytes 620 */ 621 #define MAKE_HANDLE(gpu_id, idr_handle) \ 622 (((uint64_t)(gpu_id) << 32) + idr_handle) 623 #define GET_GPU_ID(handle) (handle >> 32) 624 #define GET_IDR_HANDLE(handle) (handle & 0xFFFFFFFF) 625 626 enum kfd_pdd_bound { 627 PDD_UNBOUND = 0, 628 PDD_BOUND, 629 PDD_BOUND_SUSPENDED, 630 }; 631 632 #define MAX_SYSFS_FILENAME_LEN 11 633 634 /* 635 * SDMA counter runs at 100MHz frequency. 636 * We display SDMA activity in microsecond granularity in sysfs. 637 * As a result, the divisor is 100. 638 */ 639 #define SDMA_ACTIVITY_DIVISOR 100 640 641 /* Data that is per-process-per device. */ 642 struct kfd_process_device { 643 /* 644 * List of all per-device data for a process. 645 * Starts from kfd_process.per_device_data. 646 */ 647 struct list_head per_device_list; 648 649 /* The device that owns this data. */ 650 struct kfd_dev *dev; 651 652 /* The process that owns this kfd_process_device. */ 653 struct kfd_process *process; 654 655 /* per-process-per device QCM data structure */ 656 struct qcm_process_device qpd; 657 658 /*Apertures*/ 659 uint64_t lds_base; 660 uint64_t lds_limit; 661 uint64_t gpuvm_base; 662 uint64_t gpuvm_limit; 663 uint64_t scratch_base; 664 uint64_t scratch_limit; 665 666 /* VM context for GPUVM allocations */ 667 struct file *drm_file; 668 void *vm; 669 670 /* GPUVM allocations storage */ 671 struct idr alloc_idr; 672 673 /* Flag used to tell the pdd has dequeued from the dqm. 674 * This is used to prevent dev->dqm->ops.process_termination() from 675 * being called twice when it is already called in IOMMU callback 676 * function. 677 */ 678 bool already_dequeued; 679 bool runtime_inuse; 680 681 /* Is this process/pasid bound to this device? (amd_iommu_bind_pasid) */ 682 enum kfd_pdd_bound bound; 683 684 /* VRAM usage */ 685 uint64_t vram_usage; 686 struct attribute attr_vram; 687 char vram_filename[MAX_SYSFS_FILENAME_LEN]; 688 689 /* SDMA activity tracking */ 690 uint64_t sdma_past_activity_counter; 691 struct attribute attr_sdma; 692 char sdma_filename[MAX_SYSFS_FILENAME_LEN]; 693 }; 694 695 #define qpd_to_pdd(x) container_of(x, struct kfd_process_device, qpd) 696 697 /* Process data */ 698 struct kfd_process { 699 /* 700 * kfd_process are stored in an mm_struct*->kfd_process* 701 * hash table (kfd_processes in kfd_process.c) 702 */ 703 struct hlist_node kfd_processes; 704 705 /* 706 * Opaque pointer to mm_struct. We don't hold a reference to 707 * it so it should never be dereferenced from here. This is 708 * only used for looking up processes by their mm. 709 */ 710 void *mm; 711 712 struct kref ref; 713 struct work_struct release_work; 714 715 struct mutex mutex; 716 717 /* 718 * In any process, the thread that started main() is the lead 719 * thread and outlives the rest. 720 * It is here because amd_iommu_bind_pasid wants a task_struct. 721 * It can also be used for safely getting a reference to the 722 * mm_struct of the process. 723 */ 724 struct task_struct *lead_thread; 725 726 /* We want to receive a notification when the mm_struct is destroyed */ 727 struct mmu_notifier mmu_notifier; 728 729 uint16_t pasid; 730 unsigned int doorbell_index; 731 732 /* 733 * List of kfd_process_device structures, 734 * one for each device the process is using. 735 */ 736 struct list_head per_device_data; 737 738 struct process_queue_manager pqm; 739 740 /*Is the user space process 32 bit?*/ 741 bool is_32bit_user_mode; 742 743 /* Event-related data */ 744 struct mutex event_mutex; 745 /* Event ID allocator and lookup */ 746 struct idr event_idr; 747 /* Event page */ 748 struct kfd_signal_page *signal_page; 749 size_t signal_mapped_size; 750 size_t signal_event_count; 751 bool signal_event_limit_reached; 752 753 /* Information used for memory eviction */ 754 void *kgd_process_info; 755 /* Eviction fence that is attached to all the BOs of this process. The 756 * fence will be triggered during eviction and new one will be created 757 * during restore 758 */ 759 struct dma_fence *ef; 760 761 /* Work items for evicting and restoring BOs */ 762 struct delayed_work eviction_work; 763 struct delayed_work restore_work; 764 /* seqno of the last scheduled eviction */ 765 unsigned int last_eviction_seqno; 766 /* Approx. the last timestamp (in jiffies) when the process was 767 * restored after an eviction 768 */ 769 unsigned long last_restore_timestamp; 770 771 /* Kobj for our procfs */ 772 struct kobject *kobj; 773 struct kobject *kobj_queues; 774 struct attribute attr_pasid; 775 }; 776 777 #define KFD_PROCESS_TABLE_SIZE 5 /* bits: 32 entries */ 778 extern DECLARE_HASHTABLE(kfd_processes_table, KFD_PROCESS_TABLE_SIZE); 779 extern struct srcu_struct kfd_processes_srcu; 780 781 /** 782 * Ioctl function type. 783 * 784 * \param filep pointer to file structure. 785 * \param p amdkfd process pointer. 786 * \param data pointer to arg that was copied from user. 787 */ 788 typedef int amdkfd_ioctl_t(struct file *filep, struct kfd_process *p, 789 void *data); 790 791 struct amdkfd_ioctl_desc { 792 unsigned int cmd; 793 int flags; 794 amdkfd_ioctl_t *func; 795 unsigned int cmd_drv; 796 const char *name; 797 }; 798 bool kfd_dev_is_large_bar(struct kfd_dev *dev); 799 800 int kfd_process_create_wq(void); 801 void kfd_process_destroy_wq(void); 802 struct kfd_process *kfd_create_process(struct file *filep); 803 struct kfd_process *kfd_get_process(const struct task_struct *); 804 struct kfd_process *kfd_lookup_process_by_pasid(unsigned int pasid); 805 struct kfd_process *kfd_lookup_process_by_mm(const struct mm_struct *mm); 806 void kfd_unref_process(struct kfd_process *p); 807 int kfd_process_evict_queues(struct kfd_process *p); 808 int kfd_process_restore_queues(struct kfd_process *p); 809 void kfd_suspend_all_processes(void); 810 int kfd_resume_all_processes(void); 811 812 int kfd_process_device_init_vm(struct kfd_process_device *pdd, 813 struct file *drm_file); 814 struct kfd_process_device *kfd_bind_process_to_device(struct kfd_dev *dev, 815 struct kfd_process *p); 816 struct kfd_process_device *kfd_get_process_device_data(struct kfd_dev *dev, 817 struct kfd_process *p); 818 struct kfd_process_device *kfd_create_process_device_data(struct kfd_dev *dev, 819 struct kfd_process *p); 820 821 int kfd_reserved_mem_mmap(struct kfd_dev *dev, struct kfd_process *process, 822 struct vm_area_struct *vma); 823 824 /* KFD process API for creating and translating handles */ 825 int kfd_process_device_create_obj_handle(struct kfd_process_device *pdd, 826 void *mem); 827 void *kfd_process_device_translate_handle(struct kfd_process_device *p, 828 int handle); 829 void kfd_process_device_remove_obj_handle(struct kfd_process_device *pdd, 830 int handle); 831 832 /* Process device data iterator */ 833 struct kfd_process_device *kfd_get_first_process_device_data( 834 struct kfd_process *p); 835 struct kfd_process_device *kfd_get_next_process_device_data( 836 struct kfd_process *p, 837 struct kfd_process_device *pdd); 838 bool kfd_has_process_device_data(struct kfd_process *p); 839 840 /* PASIDs */ 841 int kfd_pasid_init(void); 842 void kfd_pasid_exit(void); 843 bool kfd_set_pasid_limit(unsigned int new_limit); 844 unsigned int kfd_get_pasid_limit(void); 845 unsigned int kfd_pasid_alloc(void); 846 void kfd_pasid_free(unsigned int pasid); 847 848 /* Doorbells */ 849 size_t kfd_doorbell_process_slice(struct kfd_dev *kfd); 850 int kfd_doorbell_init(struct kfd_dev *kfd); 851 void kfd_doorbell_fini(struct kfd_dev *kfd); 852 int kfd_doorbell_mmap(struct kfd_dev *dev, struct kfd_process *process, 853 struct vm_area_struct *vma); 854 void __iomem *kfd_get_kernel_doorbell(struct kfd_dev *kfd, 855 unsigned int *doorbell_off); 856 void kfd_release_kernel_doorbell(struct kfd_dev *kfd, u32 __iomem *db_addr); 857 u32 read_kernel_doorbell(u32 __iomem *db); 858 void write_kernel_doorbell(void __iomem *db, u32 value); 859 void write_kernel_doorbell64(void __iomem *db, u64 value); 860 unsigned int kfd_get_doorbell_dw_offset_in_bar(struct kfd_dev *kfd, 861 struct kfd_process *process, 862 unsigned int doorbell_id); 863 phys_addr_t kfd_get_process_doorbells(struct kfd_dev *dev, 864 struct kfd_process *process); 865 int kfd_alloc_process_doorbells(struct kfd_process *process); 866 void kfd_free_process_doorbells(struct kfd_process *process); 867 868 /* GTT Sub-Allocator */ 869 870 int kfd_gtt_sa_allocate(struct kfd_dev *kfd, unsigned int size, 871 struct kfd_mem_obj **mem_obj); 872 873 int kfd_gtt_sa_free(struct kfd_dev *kfd, struct kfd_mem_obj *mem_obj); 874 875 extern struct device *kfd_device; 876 877 /* KFD's procfs */ 878 void kfd_procfs_init(void); 879 void kfd_procfs_shutdown(void); 880 int kfd_procfs_add_queue(struct queue *q); 881 void kfd_procfs_del_queue(struct queue *q); 882 883 /* Topology */ 884 int kfd_topology_init(void); 885 void kfd_topology_shutdown(void); 886 int kfd_topology_add_device(struct kfd_dev *gpu); 887 int kfd_topology_remove_device(struct kfd_dev *gpu); 888 struct kfd_topology_device *kfd_topology_device_by_proximity_domain( 889 uint32_t proximity_domain); 890 struct kfd_topology_device *kfd_topology_device_by_id(uint32_t gpu_id); 891 struct kfd_dev *kfd_device_by_id(uint32_t gpu_id); 892 struct kfd_dev *kfd_device_by_pci_dev(const struct pci_dev *pdev); 893 struct kfd_dev *kfd_device_by_kgd(const struct kgd_dev *kgd); 894 int kfd_topology_enum_kfd_devices(uint8_t idx, struct kfd_dev **kdev); 895 int kfd_numa_node_to_apic_id(int numa_node_id); 896 897 /* Interrupts */ 898 int kfd_interrupt_init(struct kfd_dev *dev); 899 void kfd_interrupt_exit(struct kfd_dev *dev); 900 bool enqueue_ih_ring_entry(struct kfd_dev *kfd, const void *ih_ring_entry); 901 bool interrupt_is_wanted(struct kfd_dev *dev, 902 const uint32_t *ih_ring_entry, 903 uint32_t *patched_ihre, bool *flag); 904 905 /* amdkfd Apertures */ 906 int kfd_init_apertures(struct kfd_process *process); 907 908 /* Queue Context Management */ 909 int init_queue(struct queue **q, const struct queue_properties *properties); 910 void uninit_queue(struct queue *q); 911 void print_queue_properties(struct queue_properties *q); 912 void print_queue(struct queue *q); 913 914 struct mqd_manager *mqd_manager_init_cik(enum KFD_MQD_TYPE type, 915 struct kfd_dev *dev); 916 struct mqd_manager *mqd_manager_init_cik_hawaii(enum KFD_MQD_TYPE type, 917 struct kfd_dev *dev); 918 struct mqd_manager *mqd_manager_init_vi(enum KFD_MQD_TYPE type, 919 struct kfd_dev *dev); 920 struct mqd_manager *mqd_manager_init_vi_tonga(enum KFD_MQD_TYPE type, 921 struct kfd_dev *dev); 922 struct mqd_manager *mqd_manager_init_v9(enum KFD_MQD_TYPE type, 923 struct kfd_dev *dev); 924 struct mqd_manager *mqd_manager_init_v10(enum KFD_MQD_TYPE type, 925 struct kfd_dev *dev); 926 struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev); 927 void device_queue_manager_uninit(struct device_queue_manager *dqm); 928 struct kernel_queue *kernel_queue_init(struct kfd_dev *dev, 929 enum kfd_queue_type type); 930 void kernel_queue_uninit(struct kernel_queue *kq, bool hanging); 931 int kfd_process_vm_fault(struct device_queue_manager *dqm, unsigned int pasid); 932 933 /* Process Queue Manager */ 934 struct process_queue_node { 935 struct queue *q; 936 struct kernel_queue *kq; 937 struct list_head process_queue_list; 938 }; 939 940 void kfd_process_dequeue_from_device(struct kfd_process_device *pdd); 941 void kfd_process_dequeue_from_all_devices(struct kfd_process *p); 942 int pqm_init(struct process_queue_manager *pqm, struct kfd_process *p); 943 void pqm_uninit(struct process_queue_manager *pqm); 944 int pqm_create_queue(struct process_queue_manager *pqm, 945 struct kfd_dev *dev, 946 struct file *f, 947 struct queue_properties *properties, 948 unsigned int *qid, 949 uint32_t *p_doorbell_offset_in_process); 950 int pqm_destroy_queue(struct process_queue_manager *pqm, unsigned int qid); 951 int pqm_update_queue(struct process_queue_manager *pqm, unsigned int qid, 952 struct queue_properties *p); 953 int pqm_set_cu_mask(struct process_queue_manager *pqm, unsigned int qid, 954 struct queue_properties *p); 955 int pqm_set_gws(struct process_queue_manager *pqm, unsigned int qid, 956 void *gws); 957 struct kernel_queue *pqm_get_kernel_queue(struct process_queue_manager *pqm, 958 unsigned int qid); 959 struct queue *pqm_get_user_queue(struct process_queue_manager *pqm, 960 unsigned int qid); 961 int pqm_get_wave_state(struct process_queue_manager *pqm, 962 unsigned int qid, 963 void __user *ctl_stack, 964 u32 *ctl_stack_used_size, 965 u32 *save_area_used_size); 966 967 int amdkfd_fence_wait_timeout(unsigned int *fence_addr, 968 unsigned int fence_value, 969 unsigned int timeout_ms); 970 971 /* Packet Manager */ 972 973 #define KFD_FENCE_COMPLETED (100) 974 #define KFD_FENCE_INIT (10) 975 976 struct packet_manager { 977 struct device_queue_manager *dqm; 978 struct kernel_queue *priv_queue; 979 struct mutex lock; 980 bool allocated; 981 struct kfd_mem_obj *ib_buffer_obj; 982 unsigned int ib_size_bytes; 983 bool is_over_subscription; 984 985 const struct packet_manager_funcs *pmf; 986 }; 987 988 struct packet_manager_funcs { 989 /* Support ASIC-specific packet formats for PM4 packets */ 990 int (*map_process)(struct packet_manager *pm, uint32_t *buffer, 991 struct qcm_process_device *qpd); 992 int (*runlist)(struct packet_manager *pm, uint32_t *buffer, 993 uint64_t ib, size_t ib_size_in_dwords, bool chain); 994 int (*set_resources)(struct packet_manager *pm, uint32_t *buffer, 995 struct scheduling_resources *res); 996 int (*map_queues)(struct packet_manager *pm, uint32_t *buffer, 997 struct queue *q, bool is_static); 998 int (*unmap_queues)(struct packet_manager *pm, uint32_t *buffer, 999 enum kfd_queue_type type, 1000 enum kfd_unmap_queues_filter mode, 1001 uint32_t filter_param, bool reset, 1002 unsigned int sdma_engine); 1003 int (*query_status)(struct packet_manager *pm, uint32_t *buffer, 1004 uint64_t fence_address, uint32_t fence_value); 1005 int (*release_mem)(uint64_t gpu_addr, uint32_t *buffer); 1006 1007 /* Packet sizes */ 1008 int map_process_size; 1009 int runlist_size; 1010 int set_resources_size; 1011 int map_queues_size; 1012 int unmap_queues_size; 1013 int query_status_size; 1014 int release_mem_size; 1015 }; 1016 1017 extern const struct packet_manager_funcs kfd_vi_pm_funcs; 1018 extern const struct packet_manager_funcs kfd_v9_pm_funcs; 1019 1020 int pm_init(struct packet_manager *pm, struct device_queue_manager *dqm); 1021 void pm_uninit(struct packet_manager *pm, bool hanging); 1022 int pm_send_set_resources(struct packet_manager *pm, 1023 struct scheduling_resources *res); 1024 int pm_send_runlist(struct packet_manager *pm, struct list_head *dqm_queues); 1025 int pm_send_query_status(struct packet_manager *pm, uint64_t fence_address, 1026 uint32_t fence_value); 1027 1028 int pm_send_unmap_queue(struct packet_manager *pm, enum kfd_queue_type type, 1029 enum kfd_unmap_queues_filter mode, 1030 uint32_t filter_param, bool reset, 1031 unsigned int sdma_engine); 1032 1033 void pm_release_ib(struct packet_manager *pm); 1034 1035 /* Following PM funcs can be shared among VI and AI */ 1036 unsigned int pm_build_pm4_header(unsigned int opcode, size_t packet_size); 1037 1038 uint64_t kfd_get_number_elems(struct kfd_dev *kfd); 1039 1040 /* Events */ 1041 extern const struct kfd_event_interrupt_class event_interrupt_class_cik; 1042 extern const struct kfd_event_interrupt_class event_interrupt_class_v9; 1043 1044 extern const struct kfd_device_global_init_class device_global_init_class_cik; 1045 1046 void kfd_event_init_process(struct kfd_process *p); 1047 void kfd_event_free_process(struct kfd_process *p); 1048 int kfd_event_mmap(struct kfd_process *process, struct vm_area_struct *vma); 1049 int kfd_wait_on_events(struct kfd_process *p, 1050 uint32_t num_events, void __user *data, 1051 bool all, uint32_t user_timeout_ms, 1052 uint32_t *wait_result); 1053 void kfd_signal_event_interrupt(unsigned int pasid, uint32_t partial_id, 1054 uint32_t valid_id_bits); 1055 void kfd_signal_iommu_event(struct kfd_dev *dev, 1056 unsigned int pasid, unsigned long address, 1057 bool is_write_requested, bool is_execute_requested); 1058 void kfd_signal_hw_exception_event(unsigned int pasid); 1059 int kfd_set_event(struct kfd_process *p, uint32_t event_id); 1060 int kfd_reset_event(struct kfd_process *p, uint32_t event_id); 1061 int kfd_event_page_set(struct kfd_process *p, void *kernel_address, 1062 uint64_t size); 1063 int kfd_event_create(struct file *devkfd, struct kfd_process *p, 1064 uint32_t event_type, bool auto_reset, uint32_t node_id, 1065 uint32_t *event_id, uint32_t *event_trigger_data, 1066 uint64_t *event_page_offset, uint32_t *event_slot_index); 1067 int kfd_event_destroy(struct kfd_process *p, uint32_t event_id); 1068 1069 void kfd_signal_vm_fault_event(struct kfd_dev *dev, unsigned int pasid, 1070 struct kfd_vm_fault_info *info); 1071 1072 void kfd_signal_reset_event(struct kfd_dev *dev); 1073 1074 void kfd_flush_tlb(struct kfd_process_device *pdd); 1075 1076 int dbgdev_wave_reset_wavefronts(struct kfd_dev *dev, struct kfd_process *p); 1077 1078 bool kfd_is_locked(void); 1079 1080 /* Compute profile */ 1081 void kfd_inc_compute_active(struct kfd_dev *dev); 1082 void kfd_dec_compute_active(struct kfd_dev *dev); 1083 1084 /* Cgroup Support */ 1085 /* Check with device cgroup if @kfd device is accessible */ 1086 static inline int kfd_devcgroup_check_permission(struct kfd_dev *kfd) 1087 { 1088 #if defined(CONFIG_CGROUP_DEVICE) 1089 struct drm_device *ddev = kfd->ddev; 1090 1091 return devcgroup_check_permission(DEVCG_DEV_CHAR, ddev->driver->major, 1092 ddev->render->index, 1093 DEVCG_ACC_WRITE | DEVCG_ACC_READ); 1094 #else 1095 return 0; 1096 #endif 1097 } 1098 1099 /* Debugfs */ 1100 #if defined(CONFIG_DEBUG_FS) 1101 1102 void kfd_debugfs_init(void); 1103 void kfd_debugfs_fini(void); 1104 int kfd_debugfs_mqds_by_process(struct seq_file *m, void *data); 1105 int pqm_debugfs_mqds(struct seq_file *m, void *data); 1106 int kfd_debugfs_hqds_by_device(struct seq_file *m, void *data); 1107 int dqm_debugfs_hqds(struct seq_file *m, void *data); 1108 int kfd_debugfs_rls_by_device(struct seq_file *m, void *data); 1109 int pm_debugfs_runlist(struct seq_file *m, void *data); 1110 1111 int kfd_debugfs_hang_hws(struct kfd_dev *dev); 1112 int pm_debugfs_hang_hws(struct packet_manager *pm); 1113 int dqm_debugfs_execute_queues(struct device_queue_manager *dqm); 1114 1115 #else 1116 1117 static inline void kfd_debugfs_init(void) {} 1118 static inline void kfd_debugfs_fini(void) {} 1119 1120 #endif 1121 1122 #endif 1123