1 /* $NetBSD: amdgpu_amdkfd.h,v 1.3 2021/12/18 23:44:58 riastradh Exp $ */ 2 3 /* 4 * Copyright 2014 Advanced Micro Devices, Inc. 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a 7 * copy of this software and associated documentation files (the "Software"), 8 * to deal in the Software without restriction, including without limitation 9 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 10 * and/or sell copies of the Software, and to permit persons to whom the 11 * Software is furnished to do so, subject to the following conditions: 12 * 13 * The above copyright notice and this permission notice shall be included in 14 * all copies or substantial portions of the Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 19 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR 20 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 21 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 22 * OTHER DEALINGS IN THE SOFTWARE. 23 */ 24 25 /* amdgpu_amdkfd.h defines the private interface between amdgpu and amdkfd. */ 26 27 #ifndef AMDGPU_AMDKFD_H_INCLUDED 28 #define AMDGPU_AMDKFD_H_INCLUDED 29 30 #include <linux/types.h> 31 #include <linux/mm.h> 32 #include <linux/workqueue.h> 33 #include <kgd_kfd_interface.h> 34 #include <drm/ttm/ttm_execbuf_util.h> 35 #include "amdgpu_sync.h" 36 #include "amdgpu_vm.h" 37 38 extern uint64_t amdgpu_amdkfd_total_mem_size; 39 40 struct amdgpu_device; 41 42 struct kfd_bo_va_list { 43 struct list_head bo_list; 44 struct amdgpu_bo_va *bo_va; 45 void *kgd_dev; 46 bool is_mapped; 47 uint64_t va; 48 uint64_t pte_flags; 49 }; 50 51 struct kgd_mem { 52 struct mutex lock; 53 struct amdgpu_bo *bo; 54 struct list_head bo_va_list; 55 /* protected by amdkfd_process_info.lock */ 56 struct ttm_validate_buffer validate_list; 57 struct ttm_validate_buffer resv_list; 58 uint32_t domain; 59 unsigned int mapped_to_gpu_memory; 60 uint64_t va; 61 62 uint32_t alloc_flags; 63 64 atomic_t invalid; 65 struct amdkfd_process_info *process_info; 66 67 struct amdgpu_sync sync; 68 69 bool aql_queue; 70 }; 71 72 /* KFD Memory Eviction */ 73 struct amdgpu_amdkfd_fence { 74 struct dma_fence base; 75 struct mm_struct *mm; 76 spinlock_t lock; 77 char timeline_name[TASK_COMM_LEN]; 78 }; 79 80 struct amdgpu_kfd_dev { 81 struct kfd_dev *dev; 82 uint64_t vram_used; 83 }; 84 85 enum kgd_engine_type { 86 KGD_ENGINE_PFP = 1, 87 KGD_ENGINE_ME, 88 KGD_ENGINE_CE, 89 KGD_ENGINE_MEC1, 90 KGD_ENGINE_MEC2, 91 KGD_ENGINE_RLC, 92 KGD_ENGINE_SDMA1, 93 KGD_ENGINE_SDMA2, 94 KGD_ENGINE_MAX 95 }; 96 97 struct amdgpu_amdkfd_fence *amdgpu_amdkfd_fence_create(u64 context, 98 struct mm_struct *mm); 99 bool amdkfd_fence_check_mm(struct dma_fence *f, struct mm_struct *mm); 100 struct amdgpu_amdkfd_fence *to_amdgpu_amdkfd_fence(struct dma_fence *f); 101 102 struct amdkfd_process_info { 103 /* List head of all VMs that belong to a KFD process */ 104 struct list_head vm_list_head; 105 /* List head for all KFD BOs that belong to a KFD process. */ 106 struct list_head kfd_bo_list; 107 /* List of userptr BOs that are valid or invalid */ 108 struct list_head userptr_valid_list; 109 struct list_head userptr_inval_list; 110 /* Lock to protect kfd_bo_list */ 111 struct mutex lock; 112 113 /* Number of VMs */ 114 unsigned int n_vms; 115 /* Eviction Fence */ 116 struct amdgpu_amdkfd_fence *eviction_fence; 117 118 /* MMU-notifier related fields */ 119 atomic_t evicted_bos; 120 struct delayed_work restore_userptr_work; 121 struct pid *pid; 122 }; 123 124 int amdgpu_amdkfd_init(void); 125 void amdgpu_amdkfd_fini(void); 126 127 void amdgpu_amdkfd_suspend(struct amdgpu_device *adev); 128 int amdgpu_amdkfd_resume(struct amdgpu_device *adev); 129 void amdgpu_amdkfd_interrupt(struct amdgpu_device *adev, 130 const void *ih_ring_entry); 131 void amdgpu_amdkfd_device_probe(struct amdgpu_device *adev); 132 void amdgpu_amdkfd_device_init(struct amdgpu_device *adev); 133 void amdgpu_amdkfd_device_fini(struct amdgpu_device *adev); 134 135 int amdgpu_amdkfd_evict_userptr(struct kgd_mem *mem, struct mm_struct *mm); 136 int amdgpu_amdkfd_submit_ib(struct kgd_dev *kgd, enum kgd_engine_type engine, 137 uint32_t vmid, uint64_t gpu_addr, 138 uint32_t *ib_cmd, uint32_t ib_len); 139 void amdgpu_amdkfd_set_compute_idle(struct kgd_dev *kgd, bool idle); 140 bool amdgpu_amdkfd_have_atomics_support(struct kgd_dev *kgd); 141 int amdgpu_amdkfd_flush_gpu_tlb_vmid(struct kgd_dev *kgd, uint16_t vmid); 142 int amdgpu_amdkfd_flush_gpu_tlb_pasid(struct kgd_dev *kgd, uint16_t pasid); 143 144 bool amdgpu_amdkfd_is_kfd_vmid(struct amdgpu_device *adev, u32 vmid); 145 146 int amdgpu_amdkfd_pre_reset(struct amdgpu_device *adev); 147 148 int amdgpu_amdkfd_post_reset(struct amdgpu_device *adev); 149 150 void amdgpu_amdkfd_gpu_reset(struct kgd_dev *kgd); 151 152 /* Shared API */ 153 int amdgpu_amdkfd_alloc_gtt_mem(struct kgd_dev *kgd, size_t size, 154 void **mem_obj, uint64_t *gpu_addr, 155 void **cpu_ptr, bool mqd_gfx9); 156 void amdgpu_amdkfd_free_gtt_mem(struct kgd_dev *kgd, void *mem_obj); 157 int amdgpu_amdkfd_alloc_gws(struct kgd_dev *kgd, size_t size, void **mem_obj); 158 void amdgpu_amdkfd_free_gws(struct kgd_dev *kgd, void *mem_obj); 159 int amdgpu_amdkfd_add_gws_to_process(void *info, void *gws, struct kgd_mem **mem); 160 int amdgpu_amdkfd_remove_gws_from_process(void *info, void *mem); 161 uint32_t amdgpu_amdkfd_get_fw_version(struct kgd_dev *kgd, 162 enum kgd_engine_type type); 163 void amdgpu_amdkfd_get_local_mem_info(struct kgd_dev *kgd, 164 struct kfd_local_mem_info *mem_info); 165 uint64_t amdgpu_amdkfd_get_gpu_clock_counter(struct kgd_dev *kgd); 166 167 uint32_t amdgpu_amdkfd_get_max_engine_clock_in_mhz(struct kgd_dev *kgd); 168 void amdgpu_amdkfd_get_cu_info(struct kgd_dev *kgd, struct kfd_cu_info *cu_info); 169 int amdgpu_amdkfd_get_dmabuf_info(struct kgd_dev *kgd, int dma_buf_fd, 170 struct kgd_dev **dmabuf_kgd, 171 uint64_t *bo_size, void *metadata_buffer, 172 size_t buffer_size, uint32_t *metadata_size, 173 uint32_t *flags); 174 uint64_t amdgpu_amdkfd_get_vram_usage(struct kgd_dev *kgd); 175 uint64_t amdgpu_amdkfd_get_hive_id(struct kgd_dev *kgd); 176 uint64_t amdgpu_amdkfd_get_mmio_remap_phys_addr(struct kgd_dev *kgd); 177 uint32_t amdgpu_amdkfd_get_num_gws(struct kgd_dev *kgd); 178 uint8_t amdgpu_amdkfd_get_xgmi_hops_count(struct kgd_dev *dst, struct kgd_dev *src); 179 180 /* Read user wptr from a specified user address space with page fault 181 * disabled. The memory must be pinned and mapped to the hardware when 182 * this is called in hqd_load functions, so it should never fault in 183 * the first place. This resolves a circular lock dependency involving 184 * four locks, including the DQM lock and mmap_sem. 185 */ 186 #define read_user_wptr(mmptr, wptr, dst) \ 187 ({ \ 188 bool valid = false; \ 189 if ((mmptr) && (wptr)) { \ 190 pagefault_disable(); \ 191 if ((mmptr) == current->mm) { \ 192 valid = !get_user((dst), (wptr)); \ 193 } else if (current->mm == NULL) { \ 194 use_mm(mmptr); \ 195 valid = !get_user((dst), (wptr)); \ 196 unuse_mm(mmptr); \ 197 } \ 198 pagefault_enable(); \ 199 } \ 200 valid; \ 201 }) 202 203 /* GPUVM API */ 204 int amdgpu_amdkfd_gpuvm_create_process_vm(struct kgd_dev *kgd, unsigned int pasid, 205 void **vm, void **process_info, 206 struct dma_fence **ef); 207 int amdgpu_amdkfd_gpuvm_acquire_process_vm(struct kgd_dev *kgd, 208 struct file *filp, unsigned int pasid, 209 void **vm, void **process_info, 210 struct dma_fence **ef); 211 void amdgpu_amdkfd_gpuvm_destroy_cb(struct amdgpu_device *adev, 212 struct amdgpu_vm *vm); 213 void amdgpu_amdkfd_gpuvm_destroy_process_vm(struct kgd_dev *kgd, void *vm); 214 void amdgpu_amdkfd_gpuvm_release_process_vm(struct kgd_dev *kgd, void *vm); 215 uint64_t amdgpu_amdkfd_gpuvm_get_process_page_dir(void *vm); 216 int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu( 217 struct kgd_dev *kgd, uint64_t va, uint64_t size, 218 void *vm, struct kgd_mem **mem, 219 uint64_t *offset, uint32_t flags); 220 int amdgpu_amdkfd_gpuvm_free_memory_of_gpu( 221 struct kgd_dev *kgd, struct kgd_mem *mem); 222 int amdgpu_amdkfd_gpuvm_map_memory_to_gpu( 223 struct kgd_dev *kgd, struct kgd_mem *mem, void *vm); 224 int amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu( 225 struct kgd_dev *kgd, struct kgd_mem *mem, void *vm); 226 int amdgpu_amdkfd_gpuvm_sync_memory( 227 struct kgd_dev *kgd, struct kgd_mem *mem, bool intr); 228 int amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel(struct kgd_dev *kgd, 229 struct kgd_mem *mem, void **kptr, uint64_t *size); 230 int amdgpu_amdkfd_gpuvm_restore_process_bos(void *process_info, 231 struct dma_fence **ef); 232 233 int amdgpu_amdkfd_gpuvm_get_vm_fault_info(struct kgd_dev *kgd, 234 struct kfd_vm_fault_info *info); 235 236 int amdgpu_amdkfd_gpuvm_import_dmabuf(struct kgd_dev *kgd, 237 struct dma_buf *dmabuf, 238 uint64_t va, void *vm, 239 struct kgd_mem **mem, uint64_t *size, 240 uint64_t *mmap_offset); 241 242 void amdgpu_amdkfd_gpuvm_init_mem_limits(void); 243 void amdgpu_amdkfd_unreserve_memory_limit(struct amdgpu_bo *bo); 244 245 /* KGD2KFD callbacks */ 246 int kgd2kfd_init(void); 247 void kgd2kfd_exit(void); 248 struct kfd_dev *kgd2kfd_probe(struct kgd_dev *kgd, struct pci_dev *pdev, 249 unsigned int asic_type, bool vf); 250 bool kgd2kfd_device_init(struct kfd_dev *kfd, 251 struct drm_device *ddev, 252 const struct kgd2kfd_shared_resources *gpu_resources); 253 void kgd2kfd_device_exit(struct kfd_dev *kfd); 254 void kgd2kfd_suspend(struct kfd_dev *kfd); 255 int kgd2kfd_resume(struct kfd_dev *kfd); 256 int kgd2kfd_pre_reset(struct kfd_dev *kfd); 257 int kgd2kfd_post_reset(struct kfd_dev *kfd); 258 void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry); 259 int kgd2kfd_quiesce_mm(struct mm_struct *mm); 260 int kgd2kfd_resume_mm(struct mm_struct *mm); 261 int kgd2kfd_schedule_evict_and_restore_process(struct mm_struct *mm, 262 struct dma_fence *fence); 263 void kgd2kfd_set_sram_ecc_flag(struct kfd_dev *kfd); 264 265 #endif /* AMDGPU_AMDKFD_H_INCLUDED */ 266