1 /* 2 * Copyright 2016 Advanced Micro Devices, Inc. 3 * All Rights Reserved. 4 * 5 * Permission is hereby granted, free of charge, to any person obtaining a 6 * copy of this software and associated documentation files (the 7 * "Software"), to deal in the Software without restriction, including 8 * without limitation the rights to use, copy, modify, merge, publish, 9 * distribute, sub license, and/or sell copies of the Software, and to 10 * permit persons to whom the Software is furnished to do so, subject to 11 * the following conditions: 12 * 13 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 14 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 15 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL 16 * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, 17 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 18 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 19 * USE OR OTHER DEALINGS IN THE SOFTWARE. 20 * 21 * The above copyright notice and this permission notice (including the 22 * next paragraph) shall be included in all copies or substantial portions 23 * of the Software. 24 * 25 */ 26 27 #include <linux/firmware.h> 28 #include <linux/module.h> 29 #include <linux/dmi.h> 30 #include <linux/pci.h> 31 #include <linux/debugfs.h> 32 #include <drm/drm_drv.h> 33 34 #include "amdgpu.h" 35 #include "amdgpu_pm.h" 36 #include "amdgpu_vcn.h" 37 #include "soc15d.h" 38 39 /* Firmware Names */ 40 #define FIRMWARE_RAVEN "amdgpu/raven_vcn.bin" 41 #define FIRMWARE_PICASSO "amdgpu/picasso_vcn.bin" 42 #define FIRMWARE_RAVEN2 "amdgpu/raven2_vcn.bin" 43 #define FIRMWARE_ARCTURUS "amdgpu/arcturus_vcn.bin" 44 #define FIRMWARE_RENOIR "amdgpu/renoir_vcn.bin" 45 #define FIRMWARE_GREEN_SARDINE "amdgpu/green_sardine_vcn.bin" 46 #define FIRMWARE_NAVI10 "amdgpu/navi10_vcn.bin" 47 #define FIRMWARE_NAVI14 "amdgpu/navi14_vcn.bin" 48 #define FIRMWARE_NAVI12 "amdgpu/navi12_vcn.bin" 49 #define FIRMWARE_SIENNA_CICHLID "amdgpu/sienna_cichlid_vcn.bin" 50 #define FIRMWARE_NAVY_FLOUNDER "amdgpu/navy_flounder_vcn.bin" 51 #define FIRMWARE_VANGOGH "amdgpu/vangogh_vcn.bin" 52 #define FIRMWARE_DIMGREY_CAVEFISH "amdgpu/dimgrey_cavefish_vcn.bin" 53 #define FIRMWARE_ALDEBARAN "amdgpu/aldebaran_vcn.bin" 54 #define FIRMWARE_BEIGE_GOBY "amdgpu/beige_goby_vcn.bin" 55 #define FIRMWARE_YELLOW_CARP "amdgpu/yellow_carp_vcn.bin" 56 #define FIRMWARE_VCN_3_1_2 "amdgpu/vcn_3_1_2.bin" 57 #define FIRMWARE_VCN4_0_0 "amdgpu/vcn_4_0_0.bin" 58 #define FIRMWARE_VCN4_0_2 "amdgpu/vcn_4_0_2.bin" 59 #define FIRMWARE_VCN4_0_3 "amdgpu/vcn_4_0_3.bin" 60 #define FIRMWARE_VCN4_0_4 "amdgpu/vcn_4_0_4.bin" 61 62 MODULE_FIRMWARE(FIRMWARE_RAVEN); 63 MODULE_FIRMWARE(FIRMWARE_PICASSO); 64 MODULE_FIRMWARE(FIRMWARE_RAVEN2); 65 MODULE_FIRMWARE(FIRMWARE_ARCTURUS); 66 MODULE_FIRMWARE(FIRMWARE_RENOIR); 67 MODULE_FIRMWARE(FIRMWARE_GREEN_SARDINE); 68 MODULE_FIRMWARE(FIRMWARE_ALDEBARAN); 69 MODULE_FIRMWARE(FIRMWARE_NAVI10); 70 MODULE_FIRMWARE(FIRMWARE_NAVI14); 71 MODULE_FIRMWARE(FIRMWARE_NAVI12); 72 MODULE_FIRMWARE(FIRMWARE_SIENNA_CICHLID); 73 MODULE_FIRMWARE(FIRMWARE_NAVY_FLOUNDER); 74 MODULE_FIRMWARE(FIRMWARE_VANGOGH); 75 MODULE_FIRMWARE(FIRMWARE_DIMGREY_CAVEFISH); 76 MODULE_FIRMWARE(FIRMWARE_BEIGE_GOBY); 77 MODULE_FIRMWARE(FIRMWARE_YELLOW_CARP); 78 MODULE_FIRMWARE(FIRMWARE_VCN_3_1_2); 79 MODULE_FIRMWARE(FIRMWARE_VCN4_0_0); 80 MODULE_FIRMWARE(FIRMWARE_VCN4_0_2); 81 MODULE_FIRMWARE(FIRMWARE_VCN4_0_3); 82 MODULE_FIRMWARE(FIRMWARE_VCN4_0_4); 83 84 static void amdgpu_vcn_idle_work_handler(struct work_struct *work); 85 86 int amdgpu_vcn_early_init(struct amdgpu_device *adev) 87 { 88 char ucode_prefix[30]; 89 char fw_name[40]; 90 int r; 91 92 amdgpu_ucode_ip_version_decode(adev, UVD_HWIP, ucode_prefix, sizeof(ucode_prefix)); 93 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s.bin", ucode_prefix); 94 r = amdgpu_ucode_request(adev, &adev->vcn.fw, fw_name); 95 if (r) 96 amdgpu_ucode_release(&adev->vcn.fw); 97 98 return r; 99 } 100 101 int amdgpu_vcn_sw_init(struct amdgpu_device *adev) 102 { 103 unsigned long bo_size; 104 const struct common_firmware_header *hdr; 105 unsigned char fw_check; 106 unsigned int fw_shared_size, log_offset; 107 int i, r; 108 109 INIT_DELAYED_WORK(&adev->vcn.idle_work, amdgpu_vcn_idle_work_handler); 110 mutex_init(&adev->vcn.vcn_pg_lock); 111 mutex_init(&adev->vcn.vcn1_jpeg1_workaround); 112 atomic_set(&adev->vcn.total_submission_cnt, 0); 113 for (i = 0; i < adev->vcn.num_vcn_inst; i++) 114 atomic_set(&adev->vcn.inst[i].dpg_enc_submission_cnt, 0); 115 116 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) && 117 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)) 118 adev->vcn.indirect_sram = true; 119 120 /* 121 * Some Steam Deck's BIOS versions are incompatible with the 122 * indirect SRAM mode, leading to amdgpu being unable to get 123 * properly probed (and even potentially crashing the kernel). 124 * Hence, check for these versions here - notice this is 125 * restricted to Vangogh (Deck's APU). 126 */ 127 if (adev->ip_versions[UVD_HWIP][0] == IP_VERSION(3, 0, 2)) { 128 const char *bios_ver = dmi_get_system_info(DMI_BIOS_VERSION); 129 130 if (bios_ver && (!strncmp("F7A0113", bios_ver, 7) || 131 !strncmp("F7A0114", bios_ver, 7))) { 132 adev->vcn.indirect_sram = false; 133 dev_info(adev->dev, 134 "Steam Deck quirk: indirect SRAM disabled on BIOS %s\n", bios_ver); 135 } 136 } 137 138 hdr = (const struct common_firmware_header *)adev->vcn.fw->data; 139 adev->vcn.fw_version = le32_to_cpu(hdr->ucode_version); 140 141 /* Bit 20-23, it is encode major and non-zero for new naming convention. 142 * This field is part of version minor and DRM_DISABLED_FLAG in old naming 143 * convention. Since the l:wq!atest version minor is 0x5B and DRM_DISABLED_FLAG 144 * is zero in old naming convention, this field is always zero so far. 145 * These four bits are used to tell which naming convention is present. 146 */ 147 fw_check = (le32_to_cpu(hdr->ucode_version) >> 20) & 0xf; 148 if (fw_check) { 149 unsigned int dec_ver, enc_major, enc_minor, vep, fw_rev; 150 151 fw_rev = le32_to_cpu(hdr->ucode_version) & 0xfff; 152 enc_minor = (le32_to_cpu(hdr->ucode_version) >> 12) & 0xff; 153 enc_major = fw_check; 154 dec_ver = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xf; 155 vep = (le32_to_cpu(hdr->ucode_version) >> 28) & 0xf; 156 DRM_INFO("Found VCN firmware Version ENC: %u.%u DEC: %u VEP: %u Revision: %u\n", 157 enc_major, enc_minor, dec_ver, vep, fw_rev); 158 } else { 159 unsigned int version_major, version_minor, family_id; 160 161 family_id = le32_to_cpu(hdr->ucode_version) & 0xff; 162 version_major = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xff; 163 version_minor = (le32_to_cpu(hdr->ucode_version) >> 8) & 0xff; 164 DRM_INFO("Found VCN firmware Version: %u.%u Family ID: %u\n", 165 version_major, version_minor, family_id); 166 } 167 168 bo_size = AMDGPU_VCN_STACK_SIZE + AMDGPU_VCN_CONTEXT_SIZE; 169 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) 170 bo_size += AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8); 171 172 if (adev->ip_versions[UVD_HWIP][0] >= IP_VERSION(4, 0, 0)) { 173 fw_shared_size = AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)); 174 log_offset = offsetof(struct amdgpu_vcn4_fw_shared, fw_log); 175 } else { 176 fw_shared_size = AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)); 177 log_offset = offsetof(struct amdgpu_fw_shared, fw_log); 178 } 179 180 bo_size += fw_shared_size; 181 182 if (amdgpu_vcnfw_log) 183 bo_size += AMDGPU_VCNFW_LOG_SIZE; 184 185 for (i = 0; i < adev->vcn.num_vcn_inst; i++) { 186 if (adev->vcn.harvest_config & (1 << i)) 187 continue; 188 189 r = amdgpu_bo_create_kernel(adev, bo_size, PAGE_SIZE, 190 AMDGPU_GEM_DOMAIN_VRAM | 191 AMDGPU_GEM_DOMAIN_GTT, 192 &adev->vcn.inst[i].vcpu_bo, 193 &adev->vcn.inst[i].gpu_addr, 194 &adev->vcn.inst[i].cpu_addr); 195 if (r) { 196 dev_err(adev->dev, "(%d) failed to allocate vcn bo\n", r); 197 return r; 198 } 199 200 adev->vcn.inst[i].fw_shared.cpu_addr = adev->vcn.inst[i].cpu_addr + 201 bo_size - fw_shared_size; 202 adev->vcn.inst[i].fw_shared.gpu_addr = adev->vcn.inst[i].gpu_addr + 203 bo_size - fw_shared_size; 204 205 adev->vcn.inst[i].fw_shared.mem_size = fw_shared_size; 206 207 if (amdgpu_vcnfw_log) { 208 adev->vcn.inst[i].fw_shared.cpu_addr -= AMDGPU_VCNFW_LOG_SIZE; 209 adev->vcn.inst[i].fw_shared.gpu_addr -= AMDGPU_VCNFW_LOG_SIZE; 210 adev->vcn.inst[i].fw_shared.log_offset = log_offset; 211 } 212 213 if (adev->vcn.indirect_sram) { 214 r = amdgpu_bo_create_kernel(adev, 64 * 2 * 4, PAGE_SIZE, 215 AMDGPU_GEM_DOMAIN_VRAM | 216 AMDGPU_GEM_DOMAIN_GTT, 217 &adev->vcn.inst[i].dpg_sram_bo, 218 &adev->vcn.inst[i].dpg_sram_gpu_addr, 219 &adev->vcn.inst[i].dpg_sram_cpu_addr); 220 if (r) { 221 dev_err(adev->dev, "VCN %d (%d) failed to allocate DPG bo\n", i, r); 222 return r; 223 } 224 } 225 } 226 227 return 0; 228 } 229 230 int amdgpu_vcn_sw_fini(struct amdgpu_device *adev) 231 { 232 int i, j; 233 234 for (j = 0; j < adev->vcn.num_vcn_inst; ++j) { 235 if (adev->vcn.harvest_config & (1 << j)) 236 continue; 237 238 amdgpu_bo_free_kernel( 239 &adev->vcn.inst[j].dpg_sram_bo, 240 &adev->vcn.inst[j].dpg_sram_gpu_addr, 241 (void **)&adev->vcn.inst[j].dpg_sram_cpu_addr); 242 243 kvfree(adev->vcn.inst[j].saved_bo); 244 245 amdgpu_bo_free_kernel(&adev->vcn.inst[j].vcpu_bo, 246 &adev->vcn.inst[j].gpu_addr, 247 (void **)&adev->vcn.inst[j].cpu_addr); 248 249 amdgpu_ring_fini(&adev->vcn.inst[j].ring_dec); 250 251 for (i = 0; i < adev->vcn.num_enc_rings; ++i) 252 amdgpu_ring_fini(&adev->vcn.inst[j].ring_enc[i]); 253 } 254 255 amdgpu_ucode_release(&adev->vcn.fw); 256 mutex_destroy(&adev->vcn.vcn1_jpeg1_workaround); 257 mutex_destroy(&adev->vcn.vcn_pg_lock); 258 259 return 0; 260 } 261 262 /* from vcn4 and above, only unified queue is used */ 263 static bool amdgpu_vcn_using_unified_queue(struct amdgpu_ring *ring) 264 { 265 struct amdgpu_device *adev = ring->adev; 266 bool ret = false; 267 268 if (adev->ip_versions[UVD_HWIP][0] >= IP_VERSION(4, 0, 0)) 269 ret = true; 270 271 return ret; 272 } 273 274 bool amdgpu_vcn_is_disabled_vcn(struct amdgpu_device *adev, enum vcn_ring_type type, uint32_t vcn_instance) 275 { 276 bool ret = false; 277 int vcn_config = adev->vcn.vcn_config[vcn_instance]; 278 279 if ((type == VCN_ENCODE_RING) && (vcn_config & VCN_BLOCK_ENCODE_DISABLE_MASK)) 280 ret = true; 281 else if ((type == VCN_DECODE_RING) && (vcn_config & VCN_BLOCK_DECODE_DISABLE_MASK)) 282 ret = true; 283 else if ((type == VCN_UNIFIED_RING) && (vcn_config & VCN_BLOCK_QUEUE_DISABLE_MASK)) 284 ret = true; 285 286 return ret; 287 } 288 289 int amdgpu_vcn_suspend(struct amdgpu_device *adev) 290 { 291 unsigned int size; 292 void *ptr; 293 int i, idx; 294 295 bool in_ras_intr = amdgpu_ras_intr_triggered(); 296 297 cancel_delayed_work_sync(&adev->vcn.idle_work); 298 299 /* err_event_athub will corrupt VCPU buffer, so we need to 300 * restore fw data and clear buffer in amdgpu_vcn_resume() */ 301 if (in_ras_intr) 302 return 0; 303 304 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { 305 if (adev->vcn.harvest_config & (1 << i)) 306 continue; 307 if (adev->vcn.inst[i].vcpu_bo == NULL) 308 return 0; 309 310 size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo); 311 ptr = adev->vcn.inst[i].cpu_addr; 312 313 adev->vcn.inst[i].saved_bo = kvmalloc(size, GFP_KERNEL); 314 if (!adev->vcn.inst[i].saved_bo) 315 return -ENOMEM; 316 317 if (drm_dev_enter(adev_to_drm(adev), &idx)) { 318 memcpy_fromio(adev->vcn.inst[i].saved_bo, ptr, size); 319 drm_dev_exit(idx); 320 } 321 } 322 return 0; 323 } 324 325 int amdgpu_vcn_resume(struct amdgpu_device *adev) 326 { 327 unsigned int size; 328 void *ptr; 329 int i, idx; 330 331 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { 332 if (adev->vcn.harvest_config & (1 << i)) 333 continue; 334 if (adev->vcn.inst[i].vcpu_bo == NULL) 335 return -EINVAL; 336 337 size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo); 338 ptr = adev->vcn.inst[i].cpu_addr; 339 340 if (adev->vcn.inst[i].saved_bo != NULL) { 341 if (drm_dev_enter(adev_to_drm(adev), &idx)) { 342 memcpy_toio(ptr, adev->vcn.inst[i].saved_bo, size); 343 drm_dev_exit(idx); 344 } 345 kvfree(adev->vcn.inst[i].saved_bo); 346 adev->vcn.inst[i].saved_bo = NULL; 347 } else { 348 const struct common_firmware_header *hdr; 349 unsigned int offset; 350 351 hdr = (const struct common_firmware_header *)adev->vcn.fw->data; 352 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) { 353 offset = le32_to_cpu(hdr->ucode_array_offset_bytes); 354 if (drm_dev_enter(adev_to_drm(adev), &idx)) { 355 memcpy_toio(adev->vcn.inst[i].cpu_addr, adev->vcn.fw->data + offset, 356 le32_to_cpu(hdr->ucode_size_bytes)); 357 drm_dev_exit(idx); 358 } 359 size -= le32_to_cpu(hdr->ucode_size_bytes); 360 ptr += le32_to_cpu(hdr->ucode_size_bytes); 361 } 362 memset_io(ptr, 0, size); 363 } 364 } 365 return 0; 366 } 367 368 static void amdgpu_vcn_idle_work_handler(struct work_struct *work) 369 { 370 struct amdgpu_device *adev = 371 container_of(work, struct amdgpu_device, vcn.idle_work.work); 372 unsigned int fences = 0, fence[AMDGPU_MAX_VCN_INSTANCES] = {0}; 373 unsigned int i, j; 374 int r = 0; 375 376 for (j = 0; j < adev->vcn.num_vcn_inst; ++j) { 377 if (adev->vcn.harvest_config & (1 << j)) 378 continue; 379 380 for (i = 0; i < adev->vcn.num_enc_rings; ++i) 381 fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_enc[i]); 382 383 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) { 384 struct dpg_pause_state new_state; 385 386 if (fence[j] || 387 unlikely(atomic_read(&adev->vcn.inst[j].dpg_enc_submission_cnt))) 388 new_state.fw_based = VCN_DPG_STATE__PAUSE; 389 else 390 new_state.fw_based = VCN_DPG_STATE__UNPAUSE; 391 392 adev->vcn.pause_dpg_mode(adev, j, &new_state); 393 } 394 395 fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_dec); 396 fences += fence[j]; 397 } 398 399 if (!fences && !atomic_read(&adev->vcn.total_submission_cnt)) { 400 amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN, 401 AMD_PG_STATE_GATE); 402 r = amdgpu_dpm_switch_power_profile(adev, PP_SMC_POWER_PROFILE_VIDEO, 403 false); 404 if (r) 405 dev_warn(adev->dev, "(%d) failed to disable video power profile mode\n", r); 406 } else { 407 schedule_delayed_work(&adev->vcn.idle_work, VCN_IDLE_TIMEOUT); 408 } 409 } 410 411 void amdgpu_vcn_ring_begin_use(struct amdgpu_ring *ring) 412 { 413 struct amdgpu_device *adev = ring->adev; 414 int r = 0; 415 416 atomic_inc(&adev->vcn.total_submission_cnt); 417 418 if (!cancel_delayed_work_sync(&adev->vcn.idle_work)) { 419 r = amdgpu_dpm_switch_power_profile(adev, PP_SMC_POWER_PROFILE_VIDEO, 420 true); 421 if (r) 422 dev_warn(adev->dev, "(%d) failed to switch to video power profile mode\n", r); 423 } 424 425 mutex_lock(&adev->vcn.vcn_pg_lock); 426 amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN, 427 AMD_PG_STATE_UNGATE); 428 429 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) { 430 struct dpg_pause_state new_state; 431 432 if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC) { 433 atomic_inc(&adev->vcn.inst[ring->me].dpg_enc_submission_cnt); 434 new_state.fw_based = VCN_DPG_STATE__PAUSE; 435 } else { 436 unsigned int fences = 0; 437 unsigned int i; 438 439 for (i = 0; i < adev->vcn.num_enc_rings; ++i) 440 fences += amdgpu_fence_count_emitted(&adev->vcn.inst[ring->me].ring_enc[i]); 441 442 if (fences || atomic_read(&adev->vcn.inst[ring->me].dpg_enc_submission_cnt)) 443 new_state.fw_based = VCN_DPG_STATE__PAUSE; 444 else 445 new_state.fw_based = VCN_DPG_STATE__UNPAUSE; 446 } 447 448 adev->vcn.pause_dpg_mode(adev, ring->me, &new_state); 449 } 450 mutex_unlock(&adev->vcn.vcn_pg_lock); 451 } 452 453 void amdgpu_vcn_ring_end_use(struct amdgpu_ring *ring) 454 { 455 if (ring->adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG && 456 ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC) 457 atomic_dec(&ring->adev->vcn.inst[ring->me].dpg_enc_submission_cnt); 458 459 atomic_dec(&ring->adev->vcn.total_submission_cnt); 460 461 schedule_delayed_work(&ring->adev->vcn.idle_work, VCN_IDLE_TIMEOUT); 462 } 463 464 int amdgpu_vcn_dec_ring_test_ring(struct amdgpu_ring *ring) 465 { 466 struct amdgpu_device *adev = ring->adev; 467 uint32_t tmp = 0; 468 unsigned int i; 469 int r; 470 471 /* VCN in SRIOV does not support direct register read/write */ 472 if (amdgpu_sriov_vf(adev)) 473 return 0; 474 475 WREG32(adev->vcn.inst[ring->me].external.scratch9, 0xCAFEDEAD); 476 r = amdgpu_ring_alloc(ring, 3); 477 if (r) 478 return r; 479 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.scratch9, 0)); 480 amdgpu_ring_write(ring, 0xDEADBEEF); 481 amdgpu_ring_commit(ring); 482 for (i = 0; i < adev->usec_timeout; i++) { 483 tmp = RREG32(adev->vcn.inst[ring->me].external.scratch9); 484 if (tmp == 0xDEADBEEF) 485 break; 486 udelay(1); 487 } 488 489 if (i >= adev->usec_timeout) 490 r = -ETIMEDOUT; 491 492 return r; 493 } 494 495 int amdgpu_vcn_dec_sw_ring_test_ring(struct amdgpu_ring *ring) 496 { 497 struct amdgpu_device *adev = ring->adev; 498 uint32_t rptr; 499 unsigned int i; 500 int r; 501 502 if (amdgpu_sriov_vf(adev)) 503 return 0; 504 505 r = amdgpu_ring_alloc(ring, 16); 506 if (r) 507 return r; 508 509 rptr = amdgpu_ring_get_rptr(ring); 510 511 amdgpu_ring_write(ring, VCN_DEC_SW_CMD_END); 512 amdgpu_ring_commit(ring); 513 514 for (i = 0; i < adev->usec_timeout; i++) { 515 if (amdgpu_ring_get_rptr(ring) != rptr) 516 break; 517 udelay(1); 518 } 519 520 if (i >= adev->usec_timeout) 521 r = -ETIMEDOUT; 522 523 return r; 524 } 525 526 static int amdgpu_vcn_dec_send_msg(struct amdgpu_ring *ring, 527 struct amdgpu_ib *ib_msg, 528 struct dma_fence **fence) 529 { 530 u64 addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr); 531 struct amdgpu_device *adev = ring->adev; 532 struct dma_fence *f = NULL; 533 struct amdgpu_job *job; 534 struct amdgpu_ib *ib; 535 int i, r; 536 537 r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL, 538 64, AMDGPU_IB_POOL_DIRECT, 539 &job); 540 if (r) 541 goto err; 542 543 ib = &job->ibs[0]; 544 ib->ptr[0] = PACKET0(adev->vcn.internal.data0, 0); 545 ib->ptr[1] = addr; 546 ib->ptr[2] = PACKET0(adev->vcn.internal.data1, 0); 547 ib->ptr[3] = addr >> 32; 548 ib->ptr[4] = PACKET0(adev->vcn.internal.cmd, 0); 549 ib->ptr[5] = 0; 550 for (i = 6; i < 16; i += 2) { 551 ib->ptr[i] = PACKET0(adev->vcn.internal.nop, 0); 552 ib->ptr[i+1] = 0; 553 } 554 ib->length_dw = 16; 555 556 r = amdgpu_job_submit_direct(job, ring, &f); 557 if (r) 558 goto err_free; 559 560 amdgpu_ib_free(adev, ib_msg, f); 561 562 if (fence) 563 *fence = dma_fence_get(f); 564 dma_fence_put(f); 565 566 return 0; 567 568 err_free: 569 amdgpu_job_free(job); 570 err: 571 amdgpu_ib_free(adev, ib_msg, f); 572 return r; 573 } 574 575 static int amdgpu_vcn_dec_get_create_msg(struct amdgpu_ring *ring, uint32_t handle, 576 struct amdgpu_ib *ib) 577 { 578 struct amdgpu_device *adev = ring->adev; 579 uint32_t *msg; 580 int r, i; 581 582 memset(ib, 0, sizeof(*ib)); 583 r = amdgpu_ib_get(adev, NULL, AMDGPU_GPU_PAGE_SIZE * 2, 584 AMDGPU_IB_POOL_DIRECT, 585 ib); 586 if (r) 587 return r; 588 589 msg = (uint32_t *)AMDGPU_GPU_PAGE_ALIGN((unsigned long)ib->ptr); 590 msg[0] = cpu_to_le32(0x00000028); 591 msg[1] = cpu_to_le32(0x00000038); 592 msg[2] = cpu_to_le32(0x00000001); 593 msg[3] = cpu_to_le32(0x00000000); 594 msg[4] = cpu_to_le32(handle); 595 msg[5] = cpu_to_le32(0x00000000); 596 msg[6] = cpu_to_le32(0x00000001); 597 msg[7] = cpu_to_le32(0x00000028); 598 msg[8] = cpu_to_le32(0x00000010); 599 msg[9] = cpu_to_le32(0x00000000); 600 msg[10] = cpu_to_le32(0x00000007); 601 msg[11] = cpu_to_le32(0x00000000); 602 msg[12] = cpu_to_le32(0x00000780); 603 msg[13] = cpu_to_le32(0x00000440); 604 for (i = 14; i < 1024; ++i) 605 msg[i] = cpu_to_le32(0x0); 606 607 return 0; 608 } 609 610 static int amdgpu_vcn_dec_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle, 611 struct amdgpu_ib *ib) 612 { 613 struct amdgpu_device *adev = ring->adev; 614 uint32_t *msg; 615 int r, i; 616 617 memset(ib, 0, sizeof(*ib)); 618 r = amdgpu_ib_get(adev, NULL, AMDGPU_GPU_PAGE_SIZE * 2, 619 AMDGPU_IB_POOL_DIRECT, 620 ib); 621 if (r) 622 return r; 623 624 msg = (uint32_t *)AMDGPU_GPU_PAGE_ALIGN((unsigned long)ib->ptr); 625 msg[0] = cpu_to_le32(0x00000028); 626 msg[1] = cpu_to_le32(0x00000018); 627 msg[2] = cpu_to_le32(0x00000000); 628 msg[3] = cpu_to_le32(0x00000002); 629 msg[4] = cpu_to_le32(handle); 630 msg[5] = cpu_to_le32(0x00000000); 631 for (i = 6; i < 1024; ++i) 632 msg[i] = cpu_to_le32(0x0); 633 634 return 0; 635 } 636 637 int amdgpu_vcn_dec_ring_test_ib(struct amdgpu_ring *ring, long timeout) 638 { 639 struct dma_fence *fence = NULL; 640 struct amdgpu_ib ib; 641 long r; 642 643 r = amdgpu_vcn_dec_get_create_msg(ring, 1, &ib); 644 if (r) 645 goto error; 646 647 r = amdgpu_vcn_dec_send_msg(ring, &ib, NULL); 648 if (r) 649 goto error; 650 r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &ib); 651 if (r) 652 goto error; 653 654 r = amdgpu_vcn_dec_send_msg(ring, &ib, &fence); 655 if (r) 656 goto error; 657 658 r = dma_fence_wait_timeout(fence, false, timeout); 659 if (r == 0) 660 r = -ETIMEDOUT; 661 else if (r > 0) 662 r = 0; 663 664 dma_fence_put(fence); 665 error: 666 return r; 667 } 668 669 static uint32_t *amdgpu_vcn_unified_ring_ib_header(struct amdgpu_ib *ib, 670 uint32_t ib_pack_in_dw, bool enc) 671 { 672 uint32_t *ib_checksum; 673 674 ib->ptr[ib->length_dw++] = 0x00000010; /* single queue checksum */ 675 ib->ptr[ib->length_dw++] = 0x30000002; 676 ib_checksum = &ib->ptr[ib->length_dw++]; 677 ib->ptr[ib->length_dw++] = ib_pack_in_dw; 678 679 ib->ptr[ib->length_dw++] = 0x00000010; /* engine info */ 680 ib->ptr[ib->length_dw++] = 0x30000001; 681 ib->ptr[ib->length_dw++] = enc ? 0x2 : 0x3; 682 ib->ptr[ib->length_dw++] = ib_pack_in_dw * sizeof(uint32_t); 683 684 return ib_checksum; 685 } 686 687 static void amdgpu_vcn_unified_ring_ib_checksum(uint32_t **ib_checksum, 688 uint32_t ib_pack_in_dw) 689 { 690 uint32_t i; 691 uint32_t checksum = 0; 692 693 for (i = 0; i < ib_pack_in_dw; i++) 694 checksum += *(*ib_checksum + 2 + i); 695 696 **ib_checksum = checksum; 697 } 698 699 static int amdgpu_vcn_dec_sw_send_msg(struct amdgpu_ring *ring, 700 struct amdgpu_ib *ib_msg, 701 struct dma_fence **fence) 702 { 703 struct amdgpu_vcn_decode_buffer *decode_buffer = NULL; 704 unsigned int ib_size_dw = 64; 705 struct amdgpu_device *adev = ring->adev; 706 struct dma_fence *f = NULL; 707 struct amdgpu_job *job; 708 struct amdgpu_ib *ib; 709 uint64_t addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr); 710 bool sq = amdgpu_vcn_using_unified_queue(ring); 711 uint32_t *ib_checksum; 712 uint32_t ib_pack_in_dw; 713 int i, r; 714 715 if (sq) 716 ib_size_dw += 8; 717 718 r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL, 719 ib_size_dw * 4, AMDGPU_IB_POOL_DIRECT, 720 &job); 721 if (r) 722 goto err; 723 724 ib = &job->ibs[0]; 725 ib->length_dw = 0; 726 727 /* single queue headers */ 728 if (sq) { 729 ib_pack_in_dw = sizeof(struct amdgpu_vcn_decode_buffer) / sizeof(uint32_t) 730 + 4 + 2; /* engine info + decoding ib in dw */ 731 ib_checksum = amdgpu_vcn_unified_ring_ib_header(ib, ib_pack_in_dw, false); 732 } 733 734 ib->ptr[ib->length_dw++] = sizeof(struct amdgpu_vcn_decode_buffer) + 8; 735 ib->ptr[ib->length_dw++] = cpu_to_le32(AMDGPU_VCN_IB_FLAG_DECODE_BUFFER); 736 decode_buffer = (struct amdgpu_vcn_decode_buffer *)&(ib->ptr[ib->length_dw]); 737 ib->length_dw += sizeof(struct amdgpu_vcn_decode_buffer) / 4; 738 memset(decode_buffer, 0, sizeof(struct amdgpu_vcn_decode_buffer)); 739 740 decode_buffer->valid_buf_flag |= cpu_to_le32(AMDGPU_VCN_CMD_FLAG_MSG_BUFFER); 741 decode_buffer->msg_buffer_address_hi = cpu_to_le32(addr >> 32); 742 decode_buffer->msg_buffer_address_lo = cpu_to_le32(addr); 743 744 for (i = ib->length_dw; i < ib_size_dw; ++i) 745 ib->ptr[i] = 0x0; 746 747 if (sq) 748 amdgpu_vcn_unified_ring_ib_checksum(&ib_checksum, ib_pack_in_dw); 749 750 r = amdgpu_job_submit_direct(job, ring, &f); 751 if (r) 752 goto err_free; 753 754 amdgpu_ib_free(adev, ib_msg, f); 755 756 if (fence) 757 *fence = dma_fence_get(f); 758 dma_fence_put(f); 759 760 return 0; 761 762 err_free: 763 amdgpu_job_free(job); 764 err: 765 amdgpu_ib_free(adev, ib_msg, f); 766 return r; 767 } 768 769 int amdgpu_vcn_dec_sw_ring_test_ib(struct amdgpu_ring *ring, long timeout) 770 { 771 struct dma_fence *fence = NULL; 772 struct amdgpu_ib ib; 773 long r; 774 775 r = amdgpu_vcn_dec_get_create_msg(ring, 1, &ib); 776 if (r) 777 goto error; 778 779 r = amdgpu_vcn_dec_sw_send_msg(ring, &ib, NULL); 780 if (r) 781 goto error; 782 r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &ib); 783 if (r) 784 goto error; 785 786 r = amdgpu_vcn_dec_sw_send_msg(ring, &ib, &fence); 787 if (r) 788 goto error; 789 790 r = dma_fence_wait_timeout(fence, false, timeout); 791 if (r == 0) 792 r = -ETIMEDOUT; 793 else if (r > 0) 794 r = 0; 795 796 dma_fence_put(fence); 797 error: 798 return r; 799 } 800 801 int amdgpu_vcn_enc_ring_test_ring(struct amdgpu_ring *ring) 802 { 803 struct amdgpu_device *adev = ring->adev; 804 uint32_t rptr; 805 unsigned int i; 806 int r; 807 808 if (amdgpu_sriov_vf(adev)) 809 return 0; 810 811 r = amdgpu_ring_alloc(ring, 16); 812 if (r) 813 return r; 814 815 rptr = amdgpu_ring_get_rptr(ring); 816 817 amdgpu_ring_write(ring, VCN_ENC_CMD_END); 818 amdgpu_ring_commit(ring); 819 820 for (i = 0; i < adev->usec_timeout; i++) { 821 if (amdgpu_ring_get_rptr(ring) != rptr) 822 break; 823 udelay(1); 824 } 825 826 if (i >= adev->usec_timeout) 827 r = -ETIMEDOUT; 828 829 return r; 830 } 831 832 static int amdgpu_vcn_enc_get_create_msg(struct amdgpu_ring *ring, uint32_t handle, 833 struct amdgpu_ib *ib_msg, 834 struct dma_fence **fence) 835 { 836 unsigned int ib_size_dw = 16; 837 struct amdgpu_job *job; 838 struct amdgpu_ib *ib; 839 struct dma_fence *f = NULL; 840 uint32_t *ib_checksum = NULL; 841 uint64_t addr; 842 bool sq = amdgpu_vcn_using_unified_queue(ring); 843 int i, r; 844 845 if (sq) 846 ib_size_dw += 8; 847 848 r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL, 849 ib_size_dw * 4, AMDGPU_IB_POOL_DIRECT, 850 &job); 851 if (r) 852 return r; 853 854 ib = &job->ibs[0]; 855 addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr); 856 857 ib->length_dw = 0; 858 859 if (sq) 860 ib_checksum = amdgpu_vcn_unified_ring_ib_header(ib, 0x11, true); 861 862 ib->ptr[ib->length_dw++] = 0x00000018; 863 ib->ptr[ib->length_dw++] = 0x00000001; /* session info */ 864 ib->ptr[ib->length_dw++] = handle; 865 ib->ptr[ib->length_dw++] = upper_32_bits(addr); 866 ib->ptr[ib->length_dw++] = addr; 867 ib->ptr[ib->length_dw++] = 0x0000000b; 868 869 ib->ptr[ib->length_dw++] = 0x00000014; 870 ib->ptr[ib->length_dw++] = 0x00000002; /* task info */ 871 ib->ptr[ib->length_dw++] = 0x0000001c; 872 ib->ptr[ib->length_dw++] = 0x00000000; 873 ib->ptr[ib->length_dw++] = 0x00000000; 874 875 ib->ptr[ib->length_dw++] = 0x00000008; 876 ib->ptr[ib->length_dw++] = 0x08000001; /* op initialize */ 877 878 for (i = ib->length_dw; i < ib_size_dw; ++i) 879 ib->ptr[i] = 0x0; 880 881 if (sq) 882 amdgpu_vcn_unified_ring_ib_checksum(&ib_checksum, 0x11); 883 884 r = amdgpu_job_submit_direct(job, ring, &f); 885 if (r) 886 goto err; 887 888 if (fence) 889 *fence = dma_fence_get(f); 890 dma_fence_put(f); 891 892 return 0; 893 894 err: 895 amdgpu_job_free(job); 896 return r; 897 } 898 899 static int amdgpu_vcn_enc_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle, 900 struct amdgpu_ib *ib_msg, 901 struct dma_fence **fence) 902 { 903 unsigned int ib_size_dw = 16; 904 struct amdgpu_job *job; 905 struct amdgpu_ib *ib; 906 struct dma_fence *f = NULL; 907 uint32_t *ib_checksum = NULL; 908 uint64_t addr; 909 bool sq = amdgpu_vcn_using_unified_queue(ring); 910 int i, r; 911 912 if (sq) 913 ib_size_dw += 8; 914 915 r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL, 916 ib_size_dw * 4, AMDGPU_IB_POOL_DIRECT, 917 &job); 918 if (r) 919 return r; 920 921 ib = &job->ibs[0]; 922 addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr); 923 924 ib->length_dw = 0; 925 926 if (sq) 927 ib_checksum = amdgpu_vcn_unified_ring_ib_header(ib, 0x11, true); 928 929 ib->ptr[ib->length_dw++] = 0x00000018; 930 ib->ptr[ib->length_dw++] = 0x00000001; 931 ib->ptr[ib->length_dw++] = handle; 932 ib->ptr[ib->length_dw++] = upper_32_bits(addr); 933 ib->ptr[ib->length_dw++] = addr; 934 ib->ptr[ib->length_dw++] = 0x0000000b; 935 936 ib->ptr[ib->length_dw++] = 0x00000014; 937 ib->ptr[ib->length_dw++] = 0x00000002; 938 ib->ptr[ib->length_dw++] = 0x0000001c; 939 ib->ptr[ib->length_dw++] = 0x00000000; 940 ib->ptr[ib->length_dw++] = 0x00000000; 941 942 ib->ptr[ib->length_dw++] = 0x00000008; 943 ib->ptr[ib->length_dw++] = 0x08000002; /* op close session */ 944 945 for (i = ib->length_dw; i < ib_size_dw; ++i) 946 ib->ptr[i] = 0x0; 947 948 if (sq) 949 amdgpu_vcn_unified_ring_ib_checksum(&ib_checksum, 0x11); 950 951 r = amdgpu_job_submit_direct(job, ring, &f); 952 if (r) 953 goto err; 954 955 if (fence) 956 *fence = dma_fence_get(f); 957 dma_fence_put(f); 958 959 return 0; 960 961 err: 962 amdgpu_job_free(job); 963 return r; 964 } 965 966 int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout) 967 { 968 struct amdgpu_device *adev = ring->adev; 969 struct dma_fence *fence = NULL; 970 struct amdgpu_ib ib; 971 long r; 972 973 memset(&ib, 0, sizeof(ib)); 974 r = amdgpu_ib_get(adev, NULL, (128 << 10) + AMDGPU_GPU_PAGE_SIZE, 975 AMDGPU_IB_POOL_DIRECT, 976 &ib); 977 if (r) 978 return r; 979 980 r = amdgpu_vcn_enc_get_create_msg(ring, 1, &ib, NULL); 981 if (r) 982 goto error; 983 984 r = amdgpu_vcn_enc_get_destroy_msg(ring, 1, &ib, &fence); 985 if (r) 986 goto error; 987 988 r = dma_fence_wait_timeout(fence, false, timeout); 989 if (r == 0) 990 r = -ETIMEDOUT; 991 else if (r > 0) 992 r = 0; 993 994 error: 995 amdgpu_ib_free(adev, &ib, fence); 996 dma_fence_put(fence); 997 998 return r; 999 } 1000 1001 int amdgpu_vcn_unified_ring_test_ib(struct amdgpu_ring *ring, long timeout) 1002 { 1003 struct amdgpu_device *adev = ring->adev; 1004 long r; 1005 1006 if (adev->ip_versions[UVD_HWIP][0] != IP_VERSION(4, 0, 3)) { 1007 r = amdgpu_vcn_enc_ring_test_ib(ring, timeout); 1008 if (r) 1009 goto error; 1010 } 1011 1012 r = amdgpu_vcn_dec_sw_ring_test_ib(ring, timeout); 1013 1014 error: 1015 return r; 1016 } 1017 1018 enum amdgpu_ring_priority_level amdgpu_vcn_get_enc_ring_prio(int ring) 1019 { 1020 switch (ring) { 1021 case 0: 1022 return AMDGPU_RING_PRIO_0; 1023 case 1: 1024 return AMDGPU_RING_PRIO_1; 1025 case 2: 1026 return AMDGPU_RING_PRIO_2; 1027 default: 1028 return AMDGPU_RING_PRIO_0; 1029 } 1030 } 1031 1032 void amdgpu_vcn_setup_ucode(struct amdgpu_device *adev) 1033 { 1034 int i; 1035 unsigned int idx; 1036 1037 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) { 1038 const struct common_firmware_header *hdr; 1039 1040 hdr = (const struct common_firmware_header *)adev->vcn.fw->data; 1041 1042 for (i = 0; i < adev->vcn.num_vcn_inst; i++) { 1043 if (adev->vcn.harvest_config & (1 << i)) 1044 continue; 1045 /* currently only support 2 FW instances */ 1046 if (i >= 2) { 1047 dev_info(adev->dev, "More then 2 VCN FW instances!\n"); 1048 break; 1049 } 1050 idx = AMDGPU_UCODE_ID_VCN + i; 1051 adev->firmware.ucode[idx].ucode_id = idx; 1052 adev->firmware.ucode[idx].fw = adev->vcn.fw; 1053 adev->firmware.fw_size += 1054 ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE); 1055 1056 if (adev->ip_versions[UVD_HWIP][0] == IP_VERSION(4, 0, 3)) 1057 break; 1058 } 1059 dev_info(adev->dev, "Will use PSP to load VCN firmware\n"); 1060 } 1061 } 1062 1063 /* 1064 * debugfs for mapping vcn firmware log buffer. 1065 */ 1066 #if defined(CONFIG_DEBUG_FS) 1067 static ssize_t amdgpu_debugfs_vcn_fwlog_read(struct file *f, char __user *buf, 1068 size_t size, loff_t *pos) 1069 { 1070 struct amdgpu_vcn_inst *vcn; 1071 void *log_buf; 1072 volatile struct amdgpu_vcn_fwlog *plog; 1073 unsigned int read_pos, write_pos, available, i, read_bytes = 0; 1074 unsigned int read_num[2] = {0}; 1075 1076 vcn = file_inode(f)->i_private; 1077 if (!vcn) 1078 return -ENODEV; 1079 1080 if (!vcn->fw_shared.cpu_addr || !amdgpu_vcnfw_log) 1081 return -EFAULT; 1082 1083 log_buf = vcn->fw_shared.cpu_addr + vcn->fw_shared.mem_size; 1084 1085 plog = (volatile struct amdgpu_vcn_fwlog *)log_buf; 1086 read_pos = plog->rptr; 1087 write_pos = plog->wptr; 1088 1089 if (read_pos > AMDGPU_VCNFW_LOG_SIZE || write_pos > AMDGPU_VCNFW_LOG_SIZE) 1090 return -EFAULT; 1091 1092 if (!size || (read_pos == write_pos)) 1093 return 0; 1094 1095 if (write_pos > read_pos) { 1096 available = write_pos - read_pos; 1097 read_num[0] = min(size, (size_t)available); 1098 } else { 1099 read_num[0] = AMDGPU_VCNFW_LOG_SIZE - read_pos; 1100 available = read_num[0] + write_pos - plog->header_size; 1101 if (size > available) 1102 read_num[1] = write_pos - plog->header_size; 1103 else if (size > read_num[0]) 1104 read_num[1] = size - read_num[0]; 1105 else 1106 read_num[0] = size; 1107 } 1108 1109 for (i = 0; i < 2; i++) { 1110 if (read_num[i]) { 1111 if (read_pos == AMDGPU_VCNFW_LOG_SIZE) 1112 read_pos = plog->header_size; 1113 if (read_num[i] == copy_to_user((buf + read_bytes), 1114 (log_buf + read_pos), read_num[i])) 1115 return -EFAULT; 1116 1117 read_bytes += read_num[i]; 1118 read_pos += read_num[i]; 1119 } 1120 } 1121 1122 plog->rptr = read_pos; 1123 *pos += read_bytes; 1124 return read_bytes; 1125 } 1126 1127 static const struct file_operations amdgpu_debugfs_vcnfwlog_fops = { 1128 .owner = THIS_MODULE, 1129 .read = amdgpu_debugfs_vcn_fwlog_read, 1130 .llseek = default_llseek 1131 }; 1132 #endif 1133 1134 void amdgpu_debugfs_vcn_fwlog_init(struct amdgpu_device *adev, uint8_t i, 1135 struct amdgpu_vcn_inst *vcn) 1136 { 1137 #if defined(CONFIG_DEBUG_FS) 1138 struct drm_minor *minor = adev_to_drm(adev)->primary; 1139 struct dentry *root = minor->debugfs_root; 1140 char name[32]; 1141 1142 sprintf(name, "amdgpu_vcn_%d_fwlog", i); 1143 debugfs_create_file_size(name, S_IFREG | 0444, root, vcn, 1144 &amdgpu_debugfs_vcnfwlog_fops, 1145 AMDGPU_VCNFW_LOG_SIZE); 1146 #endif 1147 } 1148 1149 void amdgpu_vcn_fwlog_init(struct amdgpu_vcn_inst *vcn) 1150 { 1151 #if defined(CONFIG_DEBUG_FS) 1152 volatile uint32_t *flag = vcn->fw_shared.cpu_addr; 1153 void *fw_log_cpu_addr = vcn->fw_shared.cpu_addr + vcn->fw_shared.mem_size; 1154 uint64_t fw_log_gpu_addr = vcn->fw_shared.gpu_addr + vcn->fw_shared.mem_size; 1155 volatile struct amdgpu_vcn_fwlog *log_buf = fw_log_cpu_addr; 1156 volatile struct amdgpu_fw_shared_fw_logging *fw_log = vcn->fw_shared.cpu_addr 1157 + vcn->fw_shared.log_offset; 1158 *flag |= cpu_to_le32(AMDGPU_VCN_FW_LOGGING_FLAG); 1159 fw_log->is_enabled = 1; 1160 fw_log->addr_lo = cpu_to_le32(fw_log_gpu_addr & 0xFFFFFFFF); 1161 fw_log->addr_hi = cpu_to_le32(fw_log_gpu_addr >> 32); 1162 fw_log->size = cpu_to_le32(AMDGPU_VCNFW_LOG_SIZE); 1163 1164 log_buf->header_size = sizeof(struct amdgpu_vcn_fwlog); 1165 log_buf->buffer_size = AMDGPU_VCNFW_LOG_SIZE; 1166 log_buf->rptr = log_buf->header_size; 1167 log_buf->wptr = log_buf->header_size; 1168 log_buf->wrapped = 0; 1169 #endif 1170 } 1171 1172 int amdgpu_vcn_process_poison_irq(struct amdgpu_device *adev, 1173 struct amdgpu_irq_src *source, 1174 struct amdgpu_iv_entry *entry) 1175 { 1176 struct ras_common_if *ras_if = adev->vcn.ras_if; 1177 struct ras_dispatch_if ih_data = { 1178 .entry = entry, 1179 }; 1180 1181 if (!ras_if) 1182 return 0; 1183 1184 if (!amdgpu_sriov_vf(adev)) { 1185 ih_data.head = *ras_if; 1186 amdgpu_ras_interrupt_dispatch(adev, &ih_data); 1187 } else { 1188 if (adev->virt.ops && adev->virt.ops->ras_poison_handler) 1189 adev->virt.ops->ras_poison_handler(adev); 1190 else 1191 dev_warn(adev->dev, 1192 "No ras_poison_handler interface in SRIOV for VCN!\n"); 1193 } 1194 1195 return 0; 1196 } 1197 1198 int amdgpu_vcn_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block) 1199 { 1200 int r, i; 1201 1202 r = amdgpu_ras_block_late_init(adev, ras_block); 1203 if (r) 1204 return r; 1205 1206 if (amdgpu_ras_is_supported(adev, ras_block->block)) { 1207 for (i = 0; i < adev->vcn.num_vcn_inst; i++) { 1208 if (adev->vcn.harvest_config & (1 << i) || 1209 !adev->vcn.inst[i].ras_poison_irq.funcs) 1210 continue; 1211 1212 r = amdgpu_irq_get(adev, &adev->vcn.inst[i].ras_poison_irq, 0); 1213 if (r) 1214 goto late_fini; 1215 } 1216 } 1217 return 0; 1218 1219 late_fini: 1220 amdgpu_ras_block_late_fini(adev, ras_block); 1221 return r; 1222 } 1223 1224 int amdgpu_vcn_ras_sw_init(struct amdgpu_device *adev) 1225 { 1226 int err; 1227 struct amdgpu_vcn_ras *ras; 1228 1229 if (!adev->vcn.ras) 1230 return 0; 1231 1232 ras = adev->vcn.ras; 1233 err = amdgpu_ras_register_ras_block(adev, &ras->ras_block); 1234 if (err) { 1235 dev_err(adev->dev, "Failed to register vcn ras block!\n"); 1236 return err; 1237 } 1238 1239 strcpy(ras->ras_block.ras_comm.name, "vcn"); 1240 ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__VCN; 1241 ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__POISON; 1242 adev->vcn.ras_if = &ras->ras_block.ras_comm; 1243 1244 if (!ras->ras_block.ras_late_init) 1245 ras->ras_block.ras_late_init = amdgpu_vcn_ras_late_init; 1246 1247 return 0; 1248 } 1249 1250 int amdgpu_vcn_psp_update_sram(struct amdgpu_device *adev, int inst_idx, 1251 enum AMDGPU_UCODE_ID ucode_id) 1252 { 1253 struct amdgpu_firmware_info ucode = { 1254 .ucode_id = (ucode_id ? ucode_id : 1255 (inst_idx ? AMDGPU_UCODE_ID_VCN1_RAM : 1256 AMDGPU_UCODE_ID_VCN0_RAM)), 1257 .mc_addr = adev->vcn.inst[inst_idx].dpg_sram_gpu_addr, 1258 .ucode_size = ((uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_curr_addr - 1259 (uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr), 1260 }; 1261 1262 return psp_execute_ip_fw_load(&adev->psp, &ucode); 1263 } 1264