1 /* $NetBSD: amdgpu_amdkfd.c,v 1.3 2018/08/27 14:04:50 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 #include <sys/cdefs.h> 26 __KERNEL_RCSID(0, "$NetBSD: amdgpu_amdkfd.c,v 1.3 2018/08/27 14:04:50 riastradh Exp $"); 27 28 #include "amdgpu_amdkfd.h" 29 #include "amd_shared.h" 30 #include <drm/drmP.h> 31 #include "amdgpu.h" 32 #include <linux/module.h> 33 34 const struct kfd2kgd_calls *kfd2kgd; 35 const struct kgd2kfd_calls *kgd2kfd; 36 bool (*kgd2kfd_init_p)(unsigned, const struct kgd2kfd_calls**); 37 38 bool amdgpu_amdkfd_init(void) 39 { 40 #if defined(CONFIG_HSA_AMD_MODULE) 41 bool (*kgd2kfd_init_p)(unsigned, const struct kgd2kfd_calls**); 42 43 kgd2kfd_init_p = symbol_request(kgd2kfd_init); 44 45 if (kgd2kfd_init_p == NULL) 46 return false; 47 #endif 48 return true; 49 } 50 51 bool amdgpu_amdkfd_load_interface(struct amdgpu_device *rdev) 52 { 53 #if defined(CONFIG_HSA_AMD_MODULE) 54 bool (*kgd2kfd_init_p)(unsigned, const struct kgd2kfd_calls**); 55 #endif 56 57 switch (rdev->asic_type) { 58 #ifdef CONFIG_DRM_AMDGPU_CIK 59 case CHIP_KAVERI: 60 kfd2kgd = amdgpu_amdkfd_gfx_7_get_functions(); 61 break; 62 #endif 63 case CHIP_CARRIZO: 64 kfd2kgd = amdgpu_amdkfd_gfx_8_0_get_functions(); 65 break; 66 default: 67 return false; 68 } 69 70 #if defined(CONFIG_HSA_AMD_MODULE) 71 kgd2kfd_init_p = symbol_request(kgd2kfd_init); 72 73 if (kgd2kfd_init_p == NULL) { 74 kfd2kgd = NULL; 75 return false; 76 } 77 78 if (!kgd2kfd_init_p(KFD_INTERFACE_VERSION, &kgd2kfd)) { 79 symbol_put(kgd2kfd_init); 80 kfd2kgd = NULL; 81 kgd2kfd = NULL; 82 83 return false; 84 } 85 86 return true; 87 #elif defined(CONFIG_HSA_AMD) 88 if (!kgd2kfd_init(KFD_INTERFACE_VERSION, &kgd2kfd)) { 89 kfd2kgd = NULL; 90 kgd2kfd = NULL; 91 return false; 92 } 93 94 return true; 95 #else 96 kfd2kgd = NULL; 97 return false; 98 #endif 99 } 100 101 void amdgpu_amdkfd_fini(void) 102 { 103 if (kgd2kfd) { 104 kgd2kfd->exit(); 105 #ifndef __NetBSD__ 106 symbol_put(kgd2kfd_init); 107 #endif 108 } 109 } 110 111 void amdgpu_amdkfd_device_probe(struct amdgpu_device *rdev) 112 { 113 if (kgd2kfd) 114 rdev->kfd = kgd2kfd->probe((struct kgd_dev *)rdev, 115 rdev->pdev, kfd2kgd); 116 } 117 118 void amdgpu_amdkfd_device_init(struct amdgpu_device *rdev) 119 { 120 if (rdev->kfd) { 121 struct kgd2kfd_shared_resources gpu_resources = { 122 .compute_vmid_bitmap = 0xFF00, 123 124 .first_compute_pipe = 1, 125 .compute_pipe_count = 4 - 1, 126 }; 127 128 amdgpu_doorbell_get_kfd_info(rdev, 129 &gpu_resources.doorbell_physical_address, 130 &gpu_resources.doorbell_aperture_size, 131 &gpu_resources.doorbell_start_offset); 132 133 kgd2kfd->device_init(rdev->kfd, &gpu_resources); 134 } 135 } 136 137 void amdgpu_amdkfd_device_fini(struct amdgpu_device *rdev) 138 { 139 if (rdev->kfd) { 140 kgd2kfd->device_exit(rdev->kfd); 141 rdev->kfd = NULL; 142 } 143 } 144 145 void amdgpu_amdkfd_interrupt(struct amdgpu_device *rdev, 146 const void *ih_ring_entry) 147 { 148 if (rdev->kfd) 149 kgd2kfd->interrupt(rdev->kfd, ih_ring_entry); 150 } 151 152 void amdgpu_amdkfd_suspend(struct amdgpu_device *rdev) 153 { 154 if (rdev->kfd) 155 kgd2kfd->suspend(rdev->kfd); 156 } 157 158 int amdgpu_amdkfd_resume(struct amdgpu_device *rdev) 159 { 160 int r = 0; 161 162 if (rdev->kfd) 163 r = kgd2kfd->resume(rdev->kfd); 164 165 return r; 166 } 167 168 #ifndef __NetBSD__ /* XXX unused? */ 169 u32 pool_to_domain(enum kgd_memory_pool p) 170 { 171 switch (p) { 172 case KGD_POOL_FRAMEBUFFER: return AMDGPU_GEM_DOMAIN_VRAM; 173 default: return AMDGPU_GEM_DOMAIN_GTT; 174 } 175 } 176 #endif 177 178 int alloc_gtt_mem(struct kgd_dev *kgd, size_t size, 179 void **mem_obj, uint64_t *gpu_addr, 180 void **cpu_ptr) 181 { 182 struct amdgpu_device *rdev = (struct amdgpu_device *)kgd; 183 struct kgd_mem **mem = (struct kgd_mem **) mem_obj; 184 int r; 185 186 BUG_ON(kgd == NULL); 187 BUG_ON(gpu_addr == NULL); 188 BUG_ON(cpu_ptr == NULL); 189 190 *mem = kmalloc(sizeof(struct kgd_mem), GFP_KERNEL); 191 if ((*mem) == NULL) 192 return -ENOMEM; 193 194 r = amdgpu_bo_create(rdev, size, PAGE_SIZE, true, AMDGPU_GEM_DOMAIN_GTT, 195 AMDGPU_GEM_CREATE_CPU_GTT_USWC, NULL, NULL, &(*mem)->bo); 196 if (r) { 197 dev_err(rdev->dev, 198 "failed to allocate BO for amdkfd (%d)\n", r); 199 return r; 200 } 201 202 /* map the buffer */ 203 r = amdgpu_bo_reserve((*mem)->bo, true); 204 if (r) { 205 dev_err(rdev->dev, "(%d) failed to reserve bo for amdkfd\n", r); 206 goto allocate_mem_reserve_bo_failed; 207 } 208 209 r = amdgpu_bo_pin((*mem)->bo, AMDGPU_GEM_DOMAIN_GTT, 210 &(*mem)->gpu_addr); 211 if (r) { 212 dev_err(rdev->dev, "(%d) failed to pin bo for amdkfd\n", r); 213 goto allocate_mem_pin_bo_failed; 214 } 215 *gpu_addr = (*mem)->gpu_addr; 216 217 r = amdgpu_bo_kmap((*mem)->bo, &(*mem)->cpu_ptr); 218 if (r) { 219 dev_err(rdev->dev, 220 "(%d) failed to map bo to kernel for amdkfd\n", r); 221 goto allocate_mem_kmap_bo_failed; 222 } 223 *cpu_ptr = (*mem)->cpu_ptr; 224 225 amdgpu_bo_unreserve((*mem)->bo); 226 227 return 0; 228 229 allocate_mem_kmap_bo_failed: 230 amdgpu_bo_unpin((*mem)->bo); 231 allocate_mem_pin_bo_failed: 232 amdgpu_bo_unreserve((*mem)->bo); 233 allocate_mem_reserve_bo_failed: 234 amdgpu_bo_unref(&(*mem)->bo); 235 236 return r; 237 } 238 239 void free_gtt_mem(struct kgd_dev *kgd, void *mem_obj) 240 { 241 struct kgd_mem *mem = (struct kgd_mem *) mem_obj; 242 243 BUG_ON(mem == NULL); 244 245 amdgpu_bo_reserve(mem->bo, true); 246 amdgpu_bo_kunmap(mem->bo); 247 amdgpu_bo_unpin(mem->bo); 248 amdgpu_bo_unreserve(mem->bo); 249 amdgpu_bo_unref(&(mem->bo)); 250 kfree(mem); 251 } 252 253 uint64_t get_vmem_size(struct kgd_dev *kgd) 254 { 255 struct amdgpu_device *rdev = 256 (struct amdgpu_device *)kgd; 257 258 BUG_ON(kgd == NULL); 259 260 return rdev->mc.real_vram_size; 261 } 262 263 uint64_t get_gpu_clock_counter(struct kgd_dev *kgd) 264 { 265 struct amdgpu_device *rdev = (struct amdgpu_device *)kgd; 266 267 if (rdev->asic_funcs->get_gpu_clock_counter) 268 return rdev->asic_funcs->get_gpu_clock_counter(rdev); 269 return 0; 270 } 271 272 uint32_t get_max_engine_clock_in_mhz(struct kgd_dev *kgd) 273 { 274 struct amdgpu_device *rdev = (struct amdgpu_device *)kgd; 275 276 /* The sclk is in quantas of 10kHz */ 277 return rdev->pm.dpm.dyn_state.max_clock_voltage_on_ac.sclk / 100; 278 } 279