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 * This file defines the private interface between the 25 * AMD kernel graphics drivers and the AMD KFD. 26 */ 27 28 #ifndef KGD_KFD_INTERFACE_H_INCLUDED 29 #define KGD_KFD_INTERFACE_H_INCLUDED 30 31 #include <linux/types.h> 32 #include <linux/bitmap.h> 33 #include <linux/dma-fence.h> 34 35 struct pci_dev; 36 37 #define KFD_INTERFACE_VERSION 2 38 #define KGD_MAX_QUEUES 128 39 40 struct kfd_dev; 41 struct kgd_dev; 42 43 struct kgd_mem; 44 45 enum kfd_preempt_type { 46 KFD_PREEMPT_TYPE_WAVEFRONT_DRAIN = 0, 47 KFD_PREEMPT_TYPE_WAVEFRONT_RESET, 48 }; 49 50 struct kfd_cu_info { 51 uint32_t num_shader_engines; 52 uint32_t num_shader_arrays_per_engine; 53 uint32_t num_cu_per_sh; 54 uint32_t cu_active_number; 55 uint32_t cu_ao_mask; 56 uint32_t simd_per_cu; 57 uint32_t max_waves_per_simd; 58 uint32_t wave_front_size; 59 uint32_t max_scratch_slots_per_cu; 60 uint32_t lds_size; 61 uint32_t cu_bitmap[4][4]; 62 }; 63 64 /* For getting GPU local memory information from KGD */ 65 struct kfd_local_mem_info { 66 uint64_t local_mem_size_private; 67 uint64_t local_mem_size_public; 68 uint32_t vram_width; 69 uint32_t mem_clk_max; 70 }; 71 72 enum kgd_memory_pool { 73 KGD_POOL_SYSTEM_CACHEABLE = 1, 74 KGD_POOL_SYSTEM_WRITECOMBINE = 2, 75 KGD_POOL_FRAMEBUFFER = 3, 76 }; 77 78 enum kgd_engine_type { 79 KGD_ENGINE_PFP = 1, 80 KGD_ENGINE_ME, 81 KGD_ENGINE_CE, 82 KGD_ENGINE_MEC1, 83 KGD_ENGINE_MEC2, 84 KGD_ENGINE_RLC, 85 KGD_ENGINE_SDMA1, 86 KGD_ENGINE_SDMA2, 87 KGD_ENGINE_MAX 88 }; 89 90 struct kgd2kfd_shared_resources { 91 /* Bit n == 1 means VMID n is available for KFD. */ 92 unsigned int compute_vmid_bitmap; 93 94 /* number of pipes per mec */ 95 uint32_t num_pipe_per_mec; 96 97 /* number of queues per pipe */ 98 uint32_t num_queue_per_pipe; 99 100 /* Bit n == 1 means Queue n is available for KFD */ 101 DECLARE_BITMAP(queue_bitmap, KGD_MAX_QUEUES); 102 103 /* Base address of doorbell aperture. */ 104 phys_addr_t doorbell_physical_address; 105 106 /* Size in bytes of doorbell aperture. */ 107 size_t doorbell_aperture_size; 108 109 /* Number of bytes at start of aperture reserved for KGD. */ 110 size_t doorbell_start_offset; 111 112 /* GPUVM address space size in bytes */ 113 uint64_t gpuvm_size; 114 115 /* Minor device number of the render node */ 116 int drm_render_minor; 117 }; 118 119 struct tile_config { 120 uint32_t *tile_config_ptr; 121 uint32_t *macro_tile_config_ptr; 122 uint32_t num_tile_configs; 123 uint32_t num_macro_tile_configs; 124 125 uint32_t gb_addr_config; 126 uint32_t num_banks; 127 uint32_t num_ranks; 128 }; 129 130 131 /* 132 * Allocation flag domains 133 * NOTE: This must match the corresponding definitions in kfd_ioctl.h. 134 */ 135 #define ALLOC_MEM_FLAGS_VRAM (1 << 0) 136 #define ALLOC_MEM_FLAGS_GTT (1 << 1) 137 #define ALLOC_MEM_FLAGS_USERPTR (1 << 2) /* TODO */ 138 #define ALLOC_MEM_FLAGS_DOORBELL (1 << 3) /* TODO */ 139 140 /* 141 * Allocation flags attributes/access options. 142 * NOTE: This must match the corresponding definitions in kfd_ioctl.h. 143 */ 144 #define ALLOC_MEM_FLAGS_WRITABLE (1 << 31) 145 #define ALLOC_MEM_FLAGS_EXECUTABLE (1 << 30) 146 #define ALLOC_MEM_FLAGS_PUBLIC (1 << 29) 147 #define ALLOC_MEM_FLAGS_NO_SUBSTITUTE (1 << 28) /* TODO */ 148 #define ALLOC_MEM_FLAGS_AQL_QUEUE_MEM (1 << 27) 149 #define ALLOC_MEM_FLAGS_COHERENT (1 << 26) /* For GFXv9 or later */ 150 151 /** 152 * struct kfd2kgd_calls 153 * 154 * @init_gtt_mem_allocation: Allocate a buffer on the gart aperture. 155 * The buffer can be used for mqds, hpds, kernel queue, fence and runlists 156 * 157 * @free_gtt_mem: Frees a buffer that was allocated on the gart aperture 158 * 159 * @get_local_mem_info: Retrieves information about GPU local memory 160 * 161 * @get_gpu_clock_counter: Retrieves GPU clock counter 162 * 163 * @get_max_engine_clock_in_mhz: Retrieves maximum GPU clock in MHz 164 * 165 * @alloc_pasid: Allocate a PASID 166 * @free_pasid: Free a PASID 167 * 168 * @program_sh_mem_settings: A function that should initiate the memory 169 * properties such as main aperture memory type (cache / non cached) and 170 * secondary aperture base address, size and memory type. 171 * This function is used only for no cp scheduling mode. 172 * 173 * @set_pasid_vmid_mapping: Exposes pasid/vmid pair to the H/W for no cp 174 * scheduling mode. Only used for no cp scheduling mode. 175 * 176 * @init_pipeline: Initialized the compute pipelines. 177 * 178 * @hqd_load: Loads the mqd structure to a H/W hqd slot. used only for no cp 179 * sceduling mode. 180 * 181 * @hqd_sdma_load: Loads the SDMA mqd structure to a H/W SDMA hqd slot. 182 * used only for no HWS mode. 183 * 184 * @hqd_dump: Dumps CPC HQD registers to an array of address-value pairs. 185 * Array is allocated with kmalloc, needs to be freed with kfree by caller. 186 * 187 * @hqd_sdma_dump: Dumps SDMA HQD registers to an array of address-value pairs. 188 * Array is allocated with kmalloc, needs to be freed with kfree by caller. 189 * 190 * @hqd_is_occupies: Checks if a hqd slot is occupied. 191 * 192 * @hqd_destroy: Destructs and preempts the queue assigned to that hqd slot. 193 * 194 * @hqd_sdma_is_occupied: Checks if an SDMA hqd slot is occupied. 195 * 196 * @hqd_sdma_destroy: Destructs and preempts the SDMA queue assigned to that 197 * SDMA hqd slot. 198 * 199 * @get_fw_version: Returns FW versions from the header 200 * 201 * @set_scratch_backing_va: Sets VA for scratch backing memory of a VMID. 202 * Only used for no cp scheduling mode 203 * 204 * @get_tile_config: Returns GPU-specific tiling mode information 205 * 206 * @get_cu_info: Retrieves activated cu info 207 * 208 * @get_vram_usage: Returns current VRAM usage 209 * 210 * @create_process_vm: Create a VM address space for a given process and GPU 211 * 212 * @destroy_process_vm: Destroy a VM 213 * 214 * @get_process_page_dir: Get physical address of a VM page directory 215 * 216 * @set_vm_context_page_table_base: Program page table base for a VMID 217 * 218 * @alloc_memory_of_gpu: Allocate GPUVM memory 219 * 220 * @free_memory_of_gpu: Free GPUVM memory 221 * 222 * @map_memory_to_gpu: Map GPUVM memory into a specific VM address 223 * space. Allocates and updates page tables and page directories as 224 * needed. This function may return before all page table updates have 225 * completed. This allows multiple map operations (on multiple GPUs) 226 * to happen concurrently. Use sync_memory to synchronize with all 227 * pending updates. 228 * 229 * @unmap_memor_to_gpu: Unmap GPUVM memory from a specific VM address space 230 * 231 * @sync_memory: Wait for pending page table updates to complete 232 * 233 * @map_gtt_bo_to_kernel: Map a GTT BO for kernel access 234 * Pins the BO, maps it to kernel address space. Such BOs are never evicted. 235 * The kernel virtual address remains valid until the BO is freed. 236 * 237 * @restore_process_bos: Restore all BOs that belong to the 238 * process. This is intended for restoring memory mappings after a TTM 239 * eviction. 240 * 241 * @invalidate_tlbs: Invalidate TLBs for a specific PASID 242 * 243 * @invalidate_tlbs_vmid: Invalidate TLBs for a specific VMID 244 * 245 * @submit_ib: Submits an IB to the engine specified by inserting the 246 * IB to the corresponding ring (ring type). The IB is executed with the 247 * specified VMID in a user mode context. 248 * 249 * This structure contains function pointers to services that the kgd driver 250 * provides to amdkfd driver. 251 * 252 */ 253 struct kfd2kgd_calls { 254 int (*init_gtt_mem_allocation)(struct kgd_dev *kgd, size_t size, 255 void **mem_obj, uint64_t *gpu_addr, 256 void **cpu_ptr); 257 258 void (*free_gtt_mem)(struct kgd_dev *kgd, void *mem_obj); 259 260 void (*get_local_mem_info)(struct kgd_dev *kgd, 261 struct kfd_local_mem_info *mem_info); 262 uint64_t (*get_gpu_clock_counter)(struct kgd_dev *kgd); 263 264 uint32_t (*get_max_engine_clock_in_mhz)(struct kgd_dev *kgd); 265 266 int (*alloc_pasid)(unsigned int bits); 267 void (*free_pasid)(unsigned int pasid); 268 269 /* Register access functions */ 270 void (*program_sh_mem_settings)(struct kgd_dev *kgd, uint32_t vmid, 271 uint32_t sh_mem_config, uint32_t sh_mem_ape1_base, 272 uint32_t sh_mem_ape1_limit, uint32_t sh_mem_bases); 273 274 int (*set_pasid_vmid_mapping)(struct kgd_dev *kgd, unsigned int pasid, 275 unsigned int vmid); 276 277 int (*init_pipeline)(struct kgd_dev *kgd, uint32_t pipe_id, 278 uint32_t hpd_size, uint64_t hpd_gpu_addr); 279 280 int (*init_interrupts)(struct kgd_dev *kgd, uint32_t pipe_id); 281 282 int (*hqd_load)(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id, 283 uint32_t queue_id, uint32_t __user *wptr, 284 uint32_t wptr_shift, uint32_t wptr_mask, 285 struct mm_struct *mm); 286 287 int (*hqd_sdma_load)(struct kgd_dev *kgd, void *mqd, 288 uint32_t __user *wptr, struct mm_struct *mm); 289 290 int (*hqd_dump)(struct kgd_dev *kgd, 291 uint32_t pipe_id, uint32_t queue_id, 292 uint32_t (**dump)[2], uint32_t *n_regs); 293 294 int (*hqd_sdma_dump)(struct kgd_dev *kgd, 295 uint32_t engine_id, uint32_t queue_id, 296 uint32_t (**dump)[2], uint32_t *n_regs); 297 298 bool (*hqd_is_occupied)(struct kgd_dev *kgd, uint64_t queue_address, 299 uint32_t pipe_id, uint32_t queue_id); 300 301 int (*hqd_destroy)(struct kgd_dev *kgd, void *mqd, uint32_t reset_type, 302 unsigned int timeout, uint32_t pipe_id, 303 uint32_t queue_id); 304 305 bool (*hqd_sdma_is_occupied)(struct kgd_dev *kgd, void *mqd); 306 307 int (*hqd_sdma_destroy)(struct kgd_dev *kgd, void *mqd, 308 unsigned int timeout); 309 310 int (*address_watch_disable)(struct kgd_dev *kgd); 311 int (*address_watch_execute)(struct kgd_dev *kgd, 312 unsigned int watch_point_id, 313 uint32_t cntl_val, 314 uint32_t addr_hi, 315 uint32_t addr_lo); 316 int (*wave_control_execute)(struct kgd_dev *kgd, 317 uint32_t gfx_index_val, 318 uint32_t sq_cmd); 319 uint32_t (*address_watch_get_offset)(struct kgd_dev *kgd, 320 unsigned int watch_point_id, 321 unsigned int reg_offset); 322 bool (*get_atc_vmid_pasid_mapping_valid)( 323 struct kgd_dev *kgd, 324 uint8_t vmid); 325 uint16_t (*get_atc_vmid_pasid_mapping_pasid)( 326 struct kgd_dev *kgd, 327 uint8_t vmid); 328 329 uint16_t (*get_fw_version)(struct kgd_dev *kgd, 330 enum kgd_engine_type type); 331 void (*set_scratch_backing_va)(struct kgd_dev *kgd, 332 uint64_t va, uint32_t vmid); 333 int (*get_tile_config)(struct kgd_dev *kgd, struct tile_config *config); 334 335 void (*get_cu_info)(struct kgd_dev *kgd, 336 struct kfd_cu_info *cu_info); 337 uint64_t (*get_vram_usage)(struct kgd_dev *kgd); 338 339 int (*create_process_vm)(struct kgd_dev *kgd, void **vm, 340 void **process_info, struct dma_fence **ef); 341 int (*acquire_process_vm)(struct kgd_dev *kgd, struct file *filp, 342 void **vm, void **process_info, struct dma_fence **ef); 343 void (*destroy_process_vm)(struct kgd_dev *kgd, void *vm); 344 uint32_t (*get_process_page_dir)(void *vm); 345 void (*set_vm_context_page_table_base)(struct kgd_dev *kgd, 346 uint32_t vmid, uint32_t page_table_base); 347 int (*alloc_memory_of_gpu)(struct kgd_dev *kgd, uint64_t va, 348 uint64_t size, void *vm, 349 struct kgd_mem **mem, uint64_t *offset, 350 uint32_t flags); 351 int (*free_memory_of_gpu)(struct kgd_dev *kgd, struct kgd_mem *mem); 352 int (*map_memory_to_gpu)(struct kgd_dev *kgd, struct kgd_mem *mem, 353 void *vm); 354 int (*unmap_memory_to_gpu)(struct kgd_dev *kgd, struct kgd_mem *mem, 355 void *vm); 356 int (*sync_memory)(struct kgd_dev *kgd, struct kgd_mem *mem, bool intr); 357 int (*map_gtt_bo_to_kernel)(struct kgd_dev *kgd, struct kgd_mem *mem, 358 void **kptr, uint64_t *size); 359 int (*restore_process_bos)(void *process_info, struct dma_fence **ef); 360 361 int (*invalidate_tlbs)(struct kgd_dev *kgd, uint16_t pasid); 362 int (*invalidate_tlbs_vmid)(struct kgd_dev *kgd, uint16_t vmid); 363 364 int (*submit_ib)(struct kgd_dev *kgd, enum kgd_engine_type engine, 365 uint32_t vmid, uint64_t gpu_addr, 366 uint32_t *ib_cmd, uint32_t ib_len); 367 }; 368 369 /** 370 * struct kgd2kfd_calls 371 * 372 * @exit: Notifies amdkfd that kgd module is unloaded 373 * 374 * @probe: Notifies amdkfd about a probe done on a device in the kgd driver. 375 * 376 * @device_init: Initialize the newly probed device (if it is a device that 377 * amdkfd supports) 378 * 379 * @device_exit: Notifies amdkfd about a removal of a kgd device 380 * 381 * @suspend: Notifies amdkfd about a suspend action done to a kgd device 382 * 383 * @resume: Notifies amdkfd about a resume action done to a kgd device 384 * 385 * @schedule_evict_and_restore_process: Schedules work queue that will prepare 386 * for safe eviction of KFD BOs that belong to the specified process. 387 * 388 * This structure contains function callback pointers so the kgd driver 389 * will notify to the amdkfd about certain status changes. 390 * 391 */ 392 struct kgd2kfd_calls { 393 void (*exit)(void); 394 struct kfd_dev* (*probe)(struct kgd_dev *kgd, struct pci_dev *pdev, 395 const struct kfd2kgd_calls *f2g); 396 bool (*device_init)(struct kfd_dev *kfd, 397 const struct kgd2kfd_shared_resources *gpu_resources); 398 void (*device_exit)(struct kfd_dev *kfd); 399 void (*interrupt)(struct kfd_dev *kfd, const void *ih_ring_entry); 400 void (*suspend)(struct kfd_dev *kfd); 401 int (*resume)(struct kfd_dev *kfd); 402 int (*schedule_evict_and_restore_process)(struct mm_struct *mm, 403 struct dma_fence *fence); 404 }; 405 406 int kgd2kfd_init(unsigned interface_version, 407 const struct kgd2kfd_calls **g2f); 408 409 #endif /* KGD_KFD_INTERFACE_H_INCLUDED */ 410