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 #include "amdgpu_amdkfd.h" 24 #include "amd_shared.h" 25 #include <drm/drmP.h> 26 #include "amdgpu.h" 27 #include <linux/module.h> 28 29 const struct kfd2kgd_calls *kfd2kgd; 30 const struct kgd2kfd_calls *kgd2kfd; 31 bool (*kgd2kfd_init_p)(unsigned, const struct kgd2kfd_calls**); 32 33 bool amdgpu_amdkfd_init(void) 34 { 35 #if defined(CONFIG_HSA_AMD_MODULE) 36 bool (*kgd2kfd_init_p)(unsigned, const struct kgd2kfd_calls**); 37 38 kgd2kfd_init_p = symbol_request(kgd2kfd_init); 39 40 if (kgd2kfd_init_p == NULL) 41 return false; 42 #endif 43 return true; 44 } 45 46 bool amdgpu_amdkfd_load_interface(struct amdgpu_device *rdev) 47 { 48 #if defined(CONFIG_HSA_AMD_MODULE) 49 bool (*kgd2kfd_init_p)(unsigned, const struct kgd2kfd_calls**); 50 #endif 51 52 switch (rdev->asic_type) { 53 #ifdef CONFIG_DRM_AMDGPU_CIK 54 case CHIP_KAVERI: 55 kfd2kgd = amdgpu_amdkfd_gfx_7_get_functions(); 56 break; 57 #endif 58 case CHIP_CARRIZO: 59 kfd2kgd = amdgpu_amdkfd_gfx_8_0_get_functions(); 60 break; 61 default: 62 return false; 63 } 64 65 #if defined(CONFIG_HSA_AMD_MODULE) 66 kgd2kfd_init_p = symbol_request(kgd2kfd_init); 67 68 if (kgd2kfd_init_p == NULL) { 69 kfd2kgd = NULL; 70 return false; 71 } 72 73 if (!kgd2kfd_init_p(KFD_INTERFACE_VERSION, &kgd2kfd)) { 74 symbol_put(kgd2kfd_init); 75 kfd2kgd = NULL; 76 kgd2kfd = NULL; 77 78 return false; 79 } 80 81 return true; 82 #elif defined(CONFIG_HSA_AMD) 83 if (!kgd2kfd_init(KFD_INTERFACE_VERSION, &kgd2kfd)) { 84 kfd2kgd = NULL; 85 kgd2kfd = NULL; 86 return false; 87 } 88 89 return true; 90 #else 91 kfd2kgd = NULL; 92 return false; 93 #endif 94 } 95 96 void amdgpu_amdkfd_fini(void) 97 { 98 if (kgd2kfd) { 99 kgd2kfd->exit(); 100 symbol_put(kgd2kfd_init); 101 } 102 } 103 104 void amdgpu_amdkfd_device_probe(struct amdgpu_device *rdev) 105 { 106 if (kgd2kfd) 107 rdev->kfd = kgd2kfd->probe((struct kgd_dev *)rdev, 108 rdev->pdev, kfd2kgd); 109 } 110 111 void amdgpu_amdkfd_device_init(struct amdgpu_device *rdev) 112 { 113 if (rdev->kfd) { 114 struct kgd2kfd_shared_resources gpu_resources = { 115 .compute_vmid_bitmap = 0xFF00, 116 117 .first_compute_pipe = 1, 118 .compute_pipe_count = 4 - 1, 119 }; 120 121 amdgpu_doorbell_get_kfd_info(rdev, 122 &gpu_resources.doorbell_physical_address, 123 &gpu_resources.doorbell_aperture_size, 124 &gpu_resources.doorbell_start_offset); 125 126 kgd2kfd->device_init(rdev->kfd, &gpu_resources); 127 } 128 } 129 130 void amdgpu_amdkfd_device_fini(struct amdgpu_device *rdev) 131 { 132 if (rdev->kfd) { 133 kgd2kfd->device_exit(rdev->kfd); 134 rdev->kfd = NULL; 135 } 136 } 137 138 void amdgpu_amdkfd_interrupt(struct amdgpu_device *rdev, 139 const void *ih_ring_entry) 140 { 141 if (rdev->kfd) 142 kgd2kfd->interrupt(rdev->kfd, ih_ring_entry); 143 } 144 145 void amdgpu_amdkfd_suspend(struct amdgpu_device *rdev) 146 { 147 if (rdev->kfd) 148 kgd2kfd->suspend(rdev->kfd); 149 } 150 151 int amdgpu_amdkfd_resume(struct amdgpu_device *rdev) 152 { 153 int r = 0; 154 155 if (rdev->kfd) 156 r = kgd2kfd->resume(rdev->kfd); 157 158 return r; 159 } 160 161 u32 pool_to_domain(enum kgd_memory_pool p) 162 { 163 switch (p) { 164 case KGD_POOL_FRAMEBUFFER: return AMDGPU_GEM_DOMAIN_VRAM; 165 default: return AMDGPU_GEM_DOMAIN_GTT; 166 } 167 } 168 169 int alloc_gtt_mem(struct kgd_dev *kgd, size_t size, 170 void **mem_obj, uint64_t *gpu_addr, 171 void **cpu_ptr) 172 { 173 struct amdgpu_device *rdev = (struct amdgpu_device *)kgd; 174 struct kgd_mem **mem = (struct kgd_mem **) mem_obj; 175 int r; 176 177 BUG_ON(kgd == NULL); 178 BUG_ON(gpu_addr == NULL); 179 BUG_ON(cpu_ptr == NULL); 180 181 *mem = kmalloc(sizeof(struct kgd_mem), GFP_KERNEL); 182 if ((*mem) == NULL) 183 return -ENOMEM; 184 185 r = amdgpu_bo_create(rdev, size, PAGE_SIZE, true, AMDGPU_GEM_DOMAIN_GTT, 186 AMDGPU_GEM_CREATE_CPU_GTT_USWC, NULL, NULL, &(*mem)->bo); 187 if (r) { 188 dev_err(rdev->dev, 189 "failed to allocate BO for amdkfd (%d)\n", r); 190 return r; 191 } 192 193 /* map the buffer */ 194 r = amdgpu_bo_reserve((*mem)->bo, true); 195 if (r) { 196 dev_err(rdev->dev, "(%d) failed to reserve bo for amdkfd\n", r); 197 goto allocate_mem_reserve_bo_failed; 198 } 199 200 r = amdgpu_bo_pin((*mem)->bo, AMDGPU_GEM_DOMAIN_GTT, 201 &(*mem)->gpu_addr); 202 if (r) { 203 dev_err(rdev->dev, "(%d) failed to pin bo for amdkfd\n", r); 204 goto allocate_mem_pin_bo_failed; 205 } 206 *gpu_addr = (*mem)->gpu_addr; 207 208 r = amdgpu_bo_kmap((*mem)->bo, &(*mem)->cpu_ptr); 209 if (r) { 210 dev_err(rdev->dev, 211 "(%d) failed to map bo to kernel for amdkfd\n", r); 212 goto allocate_mem_kmap_bo_failed; 213 } 214 *cpu_ptr = (*mem)->cpu_ptr; 215 216 amdgpu_bo_unreserve((*mem)->bo); 217 218 return 0; 219 220 allocate_mem_kmap_bo_failed: 221 amdgpu_bo_unpin((*mem)->bo); 222 allocate_mem_pin_bo_failed: 223 amdgpu_bo_unreserve((*mem)->bo); 224 allocate_mem_reserve_bo_failed: 225 amdgpu_bo_unref(&(*mem)->bo); 226 227 return r; 228 } 229 230 void free_gtt_mem(struct kgd_dev *kgd, void *mem_obj) 231 { 232 struct kgd_mem *mem = (struct kgd_mem *) mem_obj; 233 234 BUG_ON(mem == NULL); 235 236 amdgpu_bo_reserve(mem->bo, true); 237 amdgpu_bo_kunmap(mem->bo); 238 amdgpu_bo_unpin(mem->bo); 239 amdgpu_bo_unreserve(mem->bo); 240 amdgpu_bo_unref(&(mem->bo)); 241 kfree(mem); 242 } 243 244 uint64_t get_vmem_size(struct kgd_dev *kgd) 245 { 246 struct amdgpu_device *rdev = 247 (struct amdgpu_device *)kgd; 248 249 BUG_ON(kgd == NULL); 250 251 return rdev->mc.real_vram_size; 252 } 253 254 uint64_t get_gpu_clock_counter(struct kgd_dev *kgd) 255 { 256 struct amdgpu_device *rdev = (struct amdgpu_device *)kgd; 257 258 if (rdev->asic_funcs->get_gpu_clock_counter) 259 return rdev->asic_funcs->get_gpu_clock_counter(rdev); 260 return 0; 261 } 262 263 uint32_t get_max_engine_clock_in_mhz(struct kgd_dev *kgd) 264 { 265 struct amdgpu_device *rdev = (struct amdgpu_device *)kgd; 266 267 /* The sclk is in quantas of 10kHz */ 268 return rdev->pm.dpm.dyn_state.max_clock_voltage_on_ac.sclk / 100; 269 } 270