1 /* 2 * Copyright 2019 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 #include <linux/firmware.h> 25 #include <drm/drm_drv.h> 26 27 #include "amdgpu.h" 28 #include "amdgpu_vcn.h" 29 #include "amdgpu_pm.h" 30 #include "soc15.h" 31 #include "soc15d.h" 32 #include "vcn_v2_0.h" 33 #include "mmsch_v1_0.h" 34 #include "vcn_v2_5.h" 35 36 #include "vcn/vcn_2_5_offset.h" 37 #include "vcn/vcn_2_5_sh_mask.h" 38 #include "ivsrcid/vcn/irqsrcs_vcn_2_0.h" 39 40 #define VCN_VID_SOC_ADDRESS_2_0 0x1fa00 41 #define VCN1_VID_SOC_ADDRESS_3_0 0x48200 42 43 #define mmUVD_CONTEXT_ID_INTERNAL_OFFSET 0x27 44 #define mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET 0x0f 45 #define mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET 0x10 46 #define mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET 0x11 47 #define mmUVD_NO_OP_INTERNAL_OFFSET 0x29 48 #define mmUVD_GP_SCRATCH8_INTERNAL_OFFSET 0x66 49 #define mmUVD_SCRATCH9_INTERNAL_OFFSET 0xc01d 50 51 #define mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET 0x431 52 #define mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET 0x3b4 53 #define mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET 0x3b5 54 #define mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET 0x25c 55 56 #define VCN25_MAX_HW_INSTANCES_ARCTURUS 2 57 58 static const struct amdgpu_hwip_reg_entry vcn_reg_list_2_5[] = { 59 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_POWER_STATUS), 60 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_POWER_STATUS), 61 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_STATUS), 62 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_CONTEXT_ID), 63 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_CONTEXT_ID2), 64 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_GPCOM_VCPU_DATA0), 65 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_GPCOM_VCPU_DATA1), 66 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_GPCOM_VCPU_CMD), 67 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_HI), 68 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_LO), 69 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_HI2), 70 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_LO2), 71 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_HI3), 72 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_LO3), 73 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_HI4), 74 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_LO4), 75 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_RPTR), 76 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_WPTR), 77 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_RPTR2), 78 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_WPTR2), 79 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_RPTR3), 80 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_WPTR3), 81 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_RPTR4), 82 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_WPTR4), 83 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_SIZE), 84 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_SIZE2), 85 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_SIZE3), 86 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_SIZE4), 87 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_PGFSM_CONFIG), 88 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_PGFSM_STATUS), 89 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_DPG_LMA_CTL), 90 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_DPG_LMA_DATA), 91 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_DPG_LMA_MASK), 92 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_DPG_PAUSE) 93 }; 94 95 static void vcn_v2_5_set_dec_ring_funcs(struct amdgpu_device *adev); 96 static void vcn_v2_5_set_enc_ring_funcs(struct amdgpu_device *adev); 97 static void vcn_v2_5_set_irq_funcs(struct amdgpu_device *adev); 98 static int vcn_v2_5_set_powergating_state(void *handle, 99 enum amd_powergating_state state); 100 static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev, 101 int inst_idx, struct dpg_pause_state *new_state); 102 static int vcn_v2_5_sriov_start(struct amdgpu_device *adev); 103 static void vcn_v2_5_set_ras_funcs(struct amdgpu_device *adev); 104 105 static int amdgpu_ih_clientid_vcns[] = { 106 SOC15_IH_CLIENTID_VCN, 107 SOC15_IH_CLIENTID_VCN1 108 }; 109 110 /** 111 * vcn_v2_5_early_init - set function pointers and load microcode 112 * 113 * @handle: amdgpu_device pointer 114 * 115 * Set ring and irq function pointers 116 * Load microcode from filesystem 117 */ 118 static int vcn_v2_5_early_init(void *handle) 119 { 120 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 121 122 if (amdgpu_sriov_vf(adev)) { 123 adev->vcn.num_vcn_inst = 2; 124 adev->vcn.harvest_config = 0; 125 adev->vcn.num_enc_rings = 1; 126 } else { 127 u32 harvest; 128 int i; 129 130 for (i = 0; i < adev->vcn.num_vcn_inst; i++) { 131 harvest = RREG32_SOC15(VCN, i, mmCC_UVD_HARVESTING); 132 if (harvest & CC_UVD_HARVESTING__UVD_DISABLE_MASK) 133 adev->vcn.harvest_config |= 1 << i; 134 } 135 if (adev->vcn.harvest_config == (AMDGPU_VCN_HARVEST_VCN0 | 136 AMDGPU_VCN_HARVEST_VCN1)) 137 /* both instances are harvested, disable the block */ 138 return -ENOENT; 139 140 adev->vcn.num_enc_rings = 2; 141 } 142 143 vcn_v2_5_set_dec_ring_funcs(adev); 144 vcn_v2_5_set_enc_ring_funcs(adev); 145 vcn_v2_5_set_irq_funcs(adev); 146 vcn_v2_5_set_ras_funcs(adev); 147 148 return amdgpu_vcn_early_init(adev); 149 } 150 151 /** 152 * vcn_v2_5_sw_init - sw init for VCN block 153 * 154 * @handle: amdgpu_device pointer 155 * 156 * Load firmware and sw initialization 157 */ 158 static int vcn_v2_5_sw_init(void *handle) 159 { 160 struct amdgpu_ring *ring; 161 int i, j, r; 162 uint32_t reg_count = ARRAY_SIZE(vcn_reg_list_2_5); 163 uint32_t *ptr; 164 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 165 166 for (j = 0; j < adev->vcn.num_vcn_inst; j++) { 167 if (adev->vcn.harvest_config & (1 << j)) 168 continue; 169 /* VCN DEC TRAP */ 170 r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[j], 171 VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT, &adev->vcn.inst[j].irq); 172 if (r) 173 return r; 174 175 /* VCN ENC TRAP */ 176 for (i = 0; i < adev->vcn.num_enc_rings; ++i) { 177 r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[j], 178 i + VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[j].irq); 179 if (r) 180 return r; 181 } 182 183 /* VCN POISON TRAP */ 184 r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[j], 185 VCN_2_6__SRCID_UVD_POISON, &adev->vcn.inst[j].ras_poison_irq); 186 if (r) 187 return r; 188 } 189 190 r = amdgpu_vcn_sw_init(adev); 191 if (r) 192 return r; 193 194 amdgpu_vcn_setup_ucode(adev); 195 196 r = amdgpu_vcn_resume(adev); 197 if (r) 198 return r; 199 200 for (j = 0; j < adev->vcn.num_vcn_inst; j++) { 201 volatile struct amdgpu_fw_shared *fw_shared; 202 203 if (adev->vcn.harvest_config & (1 << j)) 204 continue; 205 adev->vcn.internal.context_id = mmUVD_CONTEXT_ID_INTERNAL_OFFSET; 206 adev->vcn.internal.ib_vmid = mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET; 207 adev->vcn.internal.ib_bar_low = mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET; 208 adev->vcn.internal.ib_bar_high = mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET; 209 adev->vcn.internal.ib_size = mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET; 210 adev->vcn.internal.gp_scratch8 = mmUVD_GP_SCRATCH8_INTERNAL_OFFSET; 211 212 adev->vcn.internal.scratch9 = mmUVD_SCRATCH9_INTERNAL_OFFSET; 213 adev->vcn.inst[j].external.scratch9 = SOC15_REG_OFFSET(VCN, j, mmUVD_SCRATCH9); 214 adev->vcn.internal.data0 = mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET; 215 adev->vcn.inst[j].external.data0 = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_DATA0); 216 adev->vcn.internal.data1 = mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET; 217 adev->vcn.inst[j].external.data1 = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_DATA1); 218 adev->vcn.internal.cmd = mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET; 219 adev->vcn.inst[j].external.cmd = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_CMD); 220 adev->vcn.internal.nop = mmUVD_NO_OP_INTERNAL_OFFSET; 221 adev->vcn.inst[j].external.nop = SOC15_REG_OFFSET(VCN, j, mmUVD_NO_OP); 222 223 ring = &adev->vcn.inst[j].ring_dec; 224 ring->use_doorbell = true; 225 226 ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 227 (amdgpu_sriov_vf(adev) ? 2*j : 8*j); 228 229 if (amdgpu_ip_version(adev, UVD_HWIP, 0) == IP_VERSION(2, 5, 0)) 230 ring->vm_hub = AMDGPU_MMHUB1(0); 231 else 232 ring->vm_hub = AMDGPU_MMHUB0(0); 233 234 sprintf(ring->name, "vcn_dec_%d", j); 235 r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[j].irq, 236 0, AMDGPU_RING_PRIO_DEFAULT, NULL); 237 if (r) 238 return r; 239 240 for (i = 0; i < adev->vcn.num_enc_rings; ++i) { 241 enum amdgpu_ring_priority_level hw_prio = amdgpu_vcn_get_enc_ring_prio(i); 242 243 ring = &adev->vcn.inst[j].ring_enc[i]; 244 ring->use_doorbell = true; 245 246 ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 247 (amdgpu_sriov_vf(adev) ? (1 + i + 2*j) : (2 + i + 8*j)); 248 249 if (amdgpu_ip_version(adev, UVD_HWIP, 0) == 250 IP_VERSION(2, 5, 0)) 251 ring->vm_hub = AMDGPU_MMHUB1(0); 252 else 253 ring->vm_hub = AMDGPU_MMHUB0(0); 254 255 sprintf(ring->name, "vcn_enc_%d.%d", j, i); 256 r = amdgpu_ring_init(adev, ring, 512, 257 &adev->vcn.inst[j].irq, 0, 258 hw_prio, NULL); 259 if (r) 260 return r; 261 } 262 263 fw_shared = adev->vcn.inst[j].fw_shared.cpu_addr; 264 fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_VCN_MULTI_QUEUE_FLAG); 265 266 if (amdgpu_vcnfw_log) 267 amdgpu_vcn_fwlog_init(&adev->vcn.inst[i]); 268 } 269 270 if (amdgpu_sriov_vf(adev)) { 271 r = amdgpu_virt_alloc_mm_table(adev); 272 if (r) 273 return r; 274 } 275 276 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) 277 adev->vcn.pause_dpg_mode = vcn_v2_5_pause_dpg_mode; 278 279 r = amdgpu_vcn_ras_sw_init(adev); 280 if (r) 281 return r; 282 283 /* Allocate memory for VCN IP Dump buffer */ 284 ptr = kcalloc(adev->vcn.num_vcn_inst * reg_count, sizeof(uint32_t), GFP_KERNEL); 285 if (!ptr) { 286 DRM_ERROR("Failed to allocate memory for VCN IP Dump\n"); 287 adev->vcn.ip_dump = NULL; 288 } else { 289 adev->vcn.ip_dump = ptr; 290 } 291 292 return 0; 293 } 294 295 /** 296 * vcn_v2_5_sw_fini - sw fini for VCN block 297 * 298 * @handle: amdgpu_device pointer 299 * 300 * VCN suspend and free up sw allocation 301 */ 302 static int vcn_v2_5_sw_fini(void *handle) 303 { 304 int i, r, idx; 305 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 306 volatile struct amdgpu_fw_shared *fw_shared; 307 308 if (drm_dev_enter(adev_to_drm(adev), &idx)) { 309 for (i = 0; i < adev->vcn.num_vcn_inst; i++) { 310 if (adev->vcn.harvest_config & (1 << i)) 311 continue; 312 fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr; 313 fw_shared->present_flag_0 = 0; 314 } 315 drm_dev_exit(idx); 316 } 317 318 319 if (amdgpu_sriov_vf(adev)) 320 amdgpu_virt_free_mm_table(adev); 321 322 r = amdgpu_vcn_suspend(adev); 323 if (r) 324 return r; 325 326 r = amdgpu_vcn_sw_fini(adev); 327 328 kfree(adev->vcn.ip_dump); 329 330 return r; 331 } 332 333 /** 334 * vcn_v2_5_hw_init - start and test VCN block 335 * 336 * @handle: amdgpu_device pointer 337 * 338 * Initialize the hardware, boot up the VCPU and do some testing 339 */ 340 static int vcn_v2_5_hw_init(void *handle) 341 { 342 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 343 struct amdgpu_ring *ring; 344 int i, j, r = 0; 345 346 if (amdgpu_sriov_vf(adev)) 347 r = vcn_v2_5_sriov_start(adev); 348 349 for (j = 0; j < adev->vcn.num_vcn_inst; ++j) { 350 if (adev->vcn.harvest_config & (1 << j)) 351 continue; 352 353 if (amdgpu_sriov_vf(adev)) { 354 adev->vcn.inst[j].ring_enc[0].sched.ready = true; 355 adev->vcn.inst[j].ring_enc[1].sched.ready = false; 356 adev->vcn.inst[j].ring_enc[2].sched.ready = false; 357 adev->vcn.inst[j].ring_dec.sched.ready = true; 358 } else { 359 360 ring = &adev->vcn.inst[j].ring_dec; 361 362 adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell, 363 ring->doorbell_index, j); 364 365 r = amdgpu_ring_test_helper(ring); 366 if (r) 367 return r; 368 369 for (i = 0; i < adev->vcn.num_enc_rings; ++i) { 370 ring = &adev->vcn.inst[j].ring_enc[i]; 371 r = amdgpu_ring_test_helper(ring); 372 if (r) 373 return r; 374 } 375 } 376 } 377 378 return r; 379 } 380 381 /** 382 * vcn_v2_5_hw_fini - stop the hardware block 383 * 384 * @handle: amdgpu_device pointer 385 * 386 * Stop the VCN block, mark ring as not ready any more 387 */ 388 static int vcn_v2_5_hw_fini(void *handle) 389 { 390 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 391 int i; 392 393 cancel_delayed_work_sync(&adev->vcn.idle_work); 394 395 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { 396 if (adev->vcn.harvest_config & (1 << i)) 397 continue; 398 399 if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) || 400 (adev->vcn.cur_state != AMD_PG_STATE_GATE && 401 RREG32_SOC15(VCN, i, mmUVD_STATUS))) 402 vcn_v2_5_set_powergating_state(adev, AMD_PG_STATE_GATE); 403 404 if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__VCN)) 405 amdgpu_irq_put(adev, &adev->vcn.inst[i].ras_poison_irq, 0); 406 } 407 408 return 0; 409 } 410 411 /** 412 * vcn_v2_5_suspend - suspend VCN block 413 * 414 * @handle: amdgpu_device pointer 415 * 416 * HW fini and suspend VCN block 417 */ 418 static int vcn_v2_5_suspend(void *handle) 419 { 420 int r; 421 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 422 423 r = vcn_v2_5_hw_fini(adev); 424 if (r) 425 return r; 426 427 r = amdgpu_vcn_suspend(adev); 428 429 return r; 430 } 431 432 /** 433 * vcn_v2_5_resume - resume VCN block 434 * 435 * @handle: amdgpu_device pointer 436 * 437 * Resume firmware and hw init VCN block 438 */ 439 static int vcn_v2_5_resume(void *handle) 440 { 441 int r; 442 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 443 444 r = amdgpu_vcn_resume(adev); 445 if (r) 446 return r; 447 448 r = vcn_v2_5_hw_init(adev); 449 450 return r; 451 } 452 453 /** 454 * vcn_v2_5_mc_resume - memory controller programming 455 * 456 * @adev: amdgpu_device pointer 457 * 458 * Let the VCN memory controller know it's offsets 459 */ 460 static void vcn_v2_5_mc_resume(struct amdgpu_device *adev) 461 { 462 uint32_t size; 463 uint32_t offset; 464 int i; 465 466 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { 467 if (adev->vcn.harvest_config & (1 << i)) 468 continue; 469 470 size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw[i]->size + 4); 471 /* cache window 0: fw */ 472 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) { 473 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW, 474 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo)); 475 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH, 476 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi)); 477 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET0, 0); 478 offset = 0; 479 } else { 480 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW, 481 lower_32_bits(adev->vcn.inst[i].gpu_addr)); 482 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH, 483 upper_32_bits(adev->vcn.inst[i].gpu_addr)); 484 offset = size; 485 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET0, 486 AMDGPU_UVD_FIRMWARE_OFFSET >> 3); 487 } 488 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE0, size); 489 490 /* cache window 1: stack */ 491 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW, 492 lower_32_bits(adev->vcn.inst[i].gpu_addr + offset)); 493 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH, 494 upper_32_bits(adev->vcn.inst[i].gpu_addr + offset)); 495 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET1, 0); 496 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE); 497 498 /* cache window 2: context */ 499 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW, 500 lower_32_bits(adev->vcn.inst[i].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE)); 501 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH, 502 upper_32_bits(adev->vcn.inst[i].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE)); 503 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET2, 0); 504 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE); 505 506 /* non-cache window */ 507 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW, 508 lower_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr)); 509 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH, 510 upper_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr)); 511 WREG32_SOC15(VCN, i, mmUVD_VCPU_NONCACHE_OFFSET0, 0); 512 WREG32_SOC15(VCN, i, mmUVD_VCPU_NONCACHE_SIZE0, 513 AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared))); 514 } 515 } 516 517 static void vcn_v2_5_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect) 518 { 519 uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw[inst_idx]->size + 4); 520 uint32_t offset; 521 522 /* cache window 0: fw */ 523 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) { 524 if (!indirect) { 525 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 526 VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 527 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect); 528 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 529 VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 530 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect); 531 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 532 VCN, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect); 533 } else { 534 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 535 VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect); 536 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 537 VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect); 538 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 539 VCN, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect); 540 } 541 offset = 0; 542 } else { 543 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 544 VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 545 lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect); 546 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 547 VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 548 upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect); 549 offset = size; 550 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 551 VCN, 0, mmUVD_VCPU_CACHE_OFFSET0), 552 AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect); 553 } 554 555 if (!indirect) 556 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 557 VCN, 0, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect); 558 else 559 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 560 VCN, 0, mmUVD_VCPU_CACHE_SIZE0), 0, 0, indirect); 561 562 /* cache window 1: stack */ 563 if (!indirect) { 564 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 565 VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 566 lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect); 567 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 568 VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 569 upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect); 570 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 571 VCN, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect); 572 } else { 573 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 574 VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect); 575 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 576 VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect); 577 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 578 VCN, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect); 579 } 580 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 581 VCN, 0, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect); 582 583 /* cache window 2: context */ 584 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 585 VCN, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW), 586 lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect); 587 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 588 VCN, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH), 589 upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect); 590 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 591 VCN, 0, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect); 592 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 593 VCN, 0, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect); 594 595 /* non-cache window */ 596 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 597 VCN, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW), 598 lower_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect); 599 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 600 VCN, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH), 601 upper_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect); 602 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 603 VCN, 0, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect); 604 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 605 VCN, 0, mmUVD_VCPU_NONCACHE_SIZE0), 606 AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)), 0, indirect); 607 608 /* VCN global tiling registers */ 609 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 610 VCN, 0, mmUVD_GFX8_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect); 611 } 612 613 /** 614 * vcn_v2_5_disable_clock_gating - disable VCN clock gating 615 * 616 * @adev: amdgpu_device pointer 617 * 618 * Disable clock gating for VCN block 619 */ 620 static void vcn_v2_5_disable_clock_gating(struct amdgpu_device *adev) 621 { 622 uint32_t data; 623 int i; 624 625 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { 626 if (adev->vcn.harvest_config & (1 << i)) 627 continue; 628 /* UVD disable CGC */ 629 data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL); 630 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG) 631 data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 632 else 633 data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK; 634 data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT; 635 data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT; 636 WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data); 637 638 data = RREG32_SOC15(VCN, i, mmUVD_CGC_GATE); 639 data &= ~(UVD_CGC_GATE__SYS_MASK 640 | UVD_CGC_GATE__UDEC_MASK 641 | UVD_CGC_GATE__MPEG2_MASK 642 | UVD_CGC_GATE__REGS_MASK 643 | UVD_CGC_GATE__RBC_MASK 644 | UVD_CGC_GATE__LMI_MC_MASK 645 | UVD_CGC_GATE__LMI_UMC_MASK 646 | UVD_CGC_GATE__IDCT_MASK 647 | UVD_CGC_GATE__MPRD_MASK 648 | UVD_CGC_GATE__MPC_MASK 649 | UVD_CGC_GATE__LBSI_MASK 650 | UVD_CGC_GATE__LRBBM_MASK 651 | UVD_CGC_GATE__UDEC_RE_MASK 652 | UVD_CGC_GATE__UDEC_CM_MASK 653 | UVD_CGC_GATE__UDEC_IT_MASK 654 | UVD_CGC_GATE__UDEC_DB_MASK 655 | UVD_CGC_GATE__UDEC_MP_MASK 656 | UVD_CGC_GATE__WCB_MASK 657 | UVD_CGC_GATE__VCPU_MASK 658 | UVD_CGC_GATE__MMSCH_MASK); 659 660 WREG32_SOC15(VCN, i, mmUVD_CGC_GATE, data); 661 662 SOC15_WAIT_ON_RREG(VCN, i, mmUVD_CGC_GATE, 0, 0xFFFFFFFF); 663 664 data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL); 665 data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK 666 | UVD_CGC_CTRL__UDEC_CM_MODE_MASK 667 | UVD_CGC_CTRL__UDEC_IT_MODE_MASK 668 | UVD_CGC_CTRL__UDEC_DB_MODE_MASK 669 | UVD_CGC_CTRL__UDEC_MP_MODE_MASK 670 | UVD_CGC_CTRL__SYS_MODE_MASK 671 | UVD_CGC_CTRL__UDEC_MODE_MASK 672 | UVD_CGC_CTRL__MPEG2_MODE_MASK 673 | UVD_CGC_CTRL__REGS_MODE_MASK 674 | UVD_CGC_CTRL__RBC_MODE_MASK 675 | UVD_CGC_CTRL__LMI_MC_MODE_MASK 676 | UVD_CGC_CTRL__LMI_UMC_MODE_MASK 677 | UVD_CGC_CTRL__IDCT_MODE_MASK 678 | UVD_CGC_CTRL__MPRD_MODE_MASK 679 | UVD_CGC_CTRL__MPC_MODE_MASK 680 | UVD_CGC_CTRL__LBSI_MODE_MASK 681 | UVD_CGC_CTRL__LRBBM_MODE_MASK 682 | UVD_CGC_CTRL__WCB_MODE_MASK 683 | UVD_CGC_CTRL__VCPU_MODE_MASK 684 | UVD_CGC_CTRL__MMSCH_MODE_MASK); 685 WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data); 686 687 /* turn on */ 688 data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_GATE); 689 data |= (UVD_SUVD_CGC_GATE__SRE_MASK 690 | UVD_SUVD_CGC_GATE__SIT_MASK 691 | UVD_SUVD_CGC_GATE__SMP_MASK 692 | UVD_SUVD_CGC_GATE__SCM_MASK 693 | UVD_SUVD_CGC_GATE__SDB_MASK 694 | UVD_SUVD_CGC_GATE__SRE_H264_MASK 695 | UVD_SUVD_CGC_GATE__SRE_HEVC_MASK 696 | UVD_SUVD_CGC_GATE__SIT_H264_MASK 697 | UVD_SUVD_CGC_GATE__SIT_HEVC_MASK 698 | UVD_SUVD_CGC_GATE__SCM_H264_MASK 699 | UVD_SUVD_CGC_GATE__SCM_HEVC_MASK 700 | UVD_SUVD_CGC_GATE__SDB_H264_MASK 701 | UVD_SUVD_CGC_GATE__SDB_HEVC_MASK 702 | UVD_SUVD_CGC_GATE__SCLR_MASK 703 | UVD_SUVD_CGC_GATE__UVD_SC_MASK 704 | UVD_SUVD_CGC_GATE__ENT_MASK 705 | UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK 706 | UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK 707 | UVD_SUVD_CGC_GATE__SITE_MASK 708 | UVD_SUVD_CGC_GATE__SRE_VP9_MASK 709 | UVD_SUVD_CGC_GATE__SCM_VP9_MASK 710 | UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK 711 | UVD_SUVD_CGC_GATE__SDB_VP9_MASK 712 | UVD_SUVD_CGC_GATE__IME_HEVC_MASK); 713 WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_GATE, data); 714 715 data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL); 716 data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK 717 | UVD_SUVD_CGC_CTRL__SIT_MODE_MASK 718 | UVD_SUVD_CGC_CTRL__SMP_MODE_MASK 719 | UVD_SUVD_CGC_CTRL__SCM_MODE_MASK 720 | UVD_SUVD_CGC_CTRL__SDB_MODE_MASK 721 | UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK 722 | UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK 723 | UVD_SUVD_CGC_CTRL__ENT_MODE_MASK 724 | UVD_SUVD_CGC_CTRL__IME_MODE_MASK 725 | UVD_SUVD_CGC_CTRL__SITE_MODE_MASK); 726 WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL, data); 727 } 728 } 729 730 static void vcn_v2_5_clock_gating_dpg_mode(struct amdgpu_device *adev, 731 uint8_t sram_sel, int inst_idx, uint8_t indirect) 732 { 733 uint32_t reg_data = 0; 734 735 /* enable sw clock gating control */ 736 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG) 737 reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 738 else 739 reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 740 reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT; 741 reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT; 742 reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK | 743 UVD_CGC_CTRL__UDEC_CM_MODE_MASK | 744 UVD_CGC_CTRL__UDEC_IT_MODE_MASK | 745 UVD_CGC_CTRL__UDEC_DB_MODE_MASK | 746 UVD_CGC_CTRL__UDEC_MP_MODE_MASK | 747 UVD_CGC_CTRL__SYS_MODE_MASK | 748 UVD_CGC_CTRL__UDEC_MODE_MASK | 749 UVD_CGC_CTRL__MPEG2_MODE_MASK | 750 UVD_CGC_CTRL__REGS_MODE_MASK | 751 UVD_CGC_CTRL__RBC_MODE_MASK | 752 UVD_CGC_CTRL__LMI_MC_MODE_MASK | 753 UVD_CGC_CTRL__LMI_UMC_MODE_MASK | 754 UVD_CGC_CTRL__IDCT_MODE_MASK | 755 UVD_CGC_CTRL__MPRD_MODE_MASK | 756 UVD_CGC_CTRL__MPC_MODE_MASK | 757 UVD_CGC_CTRL__LBSI_MODE_MASK | 758 UVD_CGC_CTRL__LRBBM_MODE_MASK | 759 UVD_CGC_CTRL__WCB_MODE_MASK | 760 UVD_CGC_CTRL__VCPU_MODE_MASK | 761 UVD_CGC_CTRL__MMSCH_MODE_MASK); 762 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 763 VCN, 0, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect); 764 765 /* turn off clock gating */ 766 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 767 VCN, 0, mmUVD_CGC_GATE), 0, sram_sel, indirect); 768 769 /* turn on SUVD clock gating */ 770 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 771 VCN, 0, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect); 772 773 /* turn on sw mode in UVD_SUVD_CGC_CTRL */ 774 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 775 VCN, 0, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect); 776 } 777 778 /** 779 * vcn_v2_5_enable_clock_gating - enable VCN clock gating 780 * 781 * @adev: amdgpu_device pointer 782 * 783 * Enable clock gating for VCN block 784 */ 785 static void vcn_v2_5_enable_clock_gating(struct amdgpu_device *adev) 786 { 787 uint32_t data = 0; 788 int i; 789 790 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { 791 if (adev->vcn.harvest_config & (1 << i)) 792 continue; 793 /* enable UVD CGC */ 794 data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL); 795 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG) 796 data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 797 else 798 data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 799 data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT; 800 data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT; 801 WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data); 802 803 data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL); 804 data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK 805 | UVD_CGC_CTRL__UDEC_CM_MODE_MASK 806 | UVD_CGC_CTRL__UDEC_IT_MODE_MASK 807 | UVD_CGC_CTRL__UDEC_DB_MODE_MASK 808 | UVD_CGC_CTRL__UDEC_MP_MODE_MASK 809 | UVD_CGC_CTRL__SYS_MODE_MASK 810 | UVD_CGC_CTRL__UDEC_MODE_MASK 811 | UVD_CGC_CTRL__MPEG2_MODE_MASK 812 | UVD_CGC_CTRL__REGS_MODE_MASK 813 | UVD_CGC_CTRL__RBC_MODE_MASK 814 | UVD_CGC_CTRL__LMI_MC_MODE_MASK 815 | UVD_CGC_CTRL__LMI_UMC_MODE_MASK 816 | UVD_CGC_CTRL__IDCT_MODE_MASK 817 | UVD_CGC_CTRL__MPRD_MODE_MASK 818 | UVD_CGC_CTRL__MPC_MODE_MASK 819 | UVD_CGC_CTRL__LBSI_MODE_MASK 820 | UVD_CGC_CTRL__LRBBM_MODE_MASK 821 | UVD_CGC_CTRL__WCB_MODE_MASK 822 | UVD_CGC_CTRL__VCPU_MODE_MASK); 823 WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data); 824 825 data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL); 826 data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK 827 | UVD_SUVD_CGC_CTRL__SIT_MODE_MASK 828 | UVD_SUVD_CGC_CTRL__SMP_MODE_MASK 829 | UVD_SUVD_CGC_CTRL__SCM_MODE_MASK 830 | UVD_SUVD_CGC_CTRL__SDB_MODE_MASK 831 | UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK 832 | UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK 833 | UVD_SUVD_CGC_CTRL__ENT_MODE_MASK 834 | UVD_SUVD_CGC_CTRL__IME_MODE_MASK 835 | UVD_SUVD_CGC_CTRL__SITE_MODE_MASK); 836 WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL, data); 837 } 838 } 839 840 static void vcn_v2_6_enable_ras(struct amdgpu_device *adev, int inst_idx, 841 bool indirect) 842 { 843 uint32_t tmp; 844 845 if (amdgpu_ip_version(adev, UVD_HWIP, 0) != IP_VERSION(2, 6, 0)) 846 return; 847 848 tmp = VCN_RAS_CNTL__VCPU_VCODEC_REARM_MASK | 849 VCN_RAS_CNTL__VCPU_VCODEC_IH_EN_MASK | 850 VCN_RAS_CNTL__VCPU_VCODEC_PMI_EN_MASK | 851 VCN_RAS_CNTL__VCPU_VCODEC_STALL_EN_MASK; 852 WREG32_SOC15_DPG_MODE(inst_idx, 853 SOC15_DPG_MODE_OFFSET(VCN, 0, mmVCN_RAS_CNTL), 854 tmp, 0, indirect); 855 856 tmp = UVD_VCPU_INT_EN__RASCNTL_VCPU_VCODEC_EN_MASK; 857 WREG32_SOC15_DPG_MODE(inst_idx, 858 SOC15_DPG_MODE_OFFSET(VCN, 0, mmUVD_VCPU_INT_EN), 859 tmp, 0, indirect); 860 861 tmp = UVD_SYS_INT_EN__RASCNTL_VCPU_VCODEC_EN_MASK; 862 WREG32_SOC15_DPG_MODE(inst_idx, 863 SOC15_DPG_MODE_OFFSET(VCN, 0, mmUVD_SYS_INT_EN), 864 tmp, 0, indirect); 865 } 866 867 static int vcn_v2_5_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect) 868 { 869 volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr; 870 struct amdgpu_ring *ring; 871 uint32_t rb_bufsz, tmp; 872 873 /* disable register anti-hang mechanism */ 874 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 1, 875 ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); 876 /* enable dynamic power gating mode */ 877 tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS); 878 tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK; 879 tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK; 880 WREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS, tmp); 881 882 if (indirect) 883 adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr; 884 885 /* enable clock gating */ 886 vcn_v2_5_clock_gating_dpg_mode(adev, 0, inst_idx, indirect); 887 888 /* enable VCPU clock */ 889 tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT); 890 tmp |= UVD_VCPU_CNTL__CLK_EN_MASK; 891 tmp |= UVD_VCPU_CNTL__BLK_RST_MASK; 892 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 893 VCN, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect); 894 895 /* disable master interupt */ 896 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 897 VCN, 0, mmUVD_MASTINT_EN), 0, 0, indirect); 898 899 /* setup mmUVD_LMI_CTRL */ 900 tmp = (0x8 | UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK | 901 UVD_LMI_CTRL__REQ_MODE_MASK | 902 UVD_LMI_CTRL__CRC_RESET_MASK | 903 UVD_LMI_CTRL__MASK_MC_URGENT_MASK | 904 UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK | 905 UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK | 906 (8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) | 907 0x00100000L); 908 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 909 VCN, 0, mmUVD_LMI_CTRL), tmp, 0, indirect); 910 911 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 912 VCN, 0, mmUVD_MPC_CNTL), 913 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect); 914 915 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 916 VCN, 0, mmUVD_MPC_SET_MUXA0), 917 ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) | 918 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) | 919 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) | 920 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect); 921 922 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 923 VCN, 0, mmUVD_MPC_SET_MUXB0), 924 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) | 925 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) | 926 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) | 927 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect); 928 929 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 930 VCN, 0, mmUVD_MPC_SET_MUX), 931 ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) | 932 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) | 933 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect); 934 935 vcn_v2_5_mc_resume_dpg_mode(adev, inst_idx, indirect); 936 937 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 938 VCN, 0, mmUVD_REG_XX_MASK), 0x10, 0, indirect); 939 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 940 VCN, 0, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect); 941 942 /* enable LMI MC and UMC channels */ 943 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 944 VCN, 0, mmUVD_LMI_CTRL2), 0, 0, indirect); 945 946 vcn_v2_6_enable_ras(adev, inst_idx, indirect); 947 948 /* unblock VCPU register access */ 949 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 950 VCN, 0, mmUVD_RB_ARB_CTRL), 0, 0, indirect); 951 952 tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT); 953 tmp |= UVD_VCPU_CNTL__CLK_EN_MASK; 954 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 955 VCN, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect); 956 957 /* enable master interrupt */ 958 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( 959 VCN, 0, mmUVD_MASTINT_EN), 960 UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect); 961 962 if (indirect) 963 amdgpu_vcn_psp_update_sram(adev, inst_idx, 0); 964 965 ring = &adev->vcn.inst[inst_idx].ring_dec; 966 /* force RBC into idle state */ 967 rb_bufsz = order_base_2(ring->ring_size); 968 tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz); 969 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1); 970 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1); 971 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1); 972 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1); 973 WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_CNTL, tmp); 974 975 /* Stall DPG before WPTR/RPTR reset */ 976 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 977 UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK, 978 ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK); 979 fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET; 980 981 /* set the write pointer delay */ 982 WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR_CNTL, 0); 983 984 /* set the wb address */ 985 WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR_ADDR, 986 (upper_32_bits(ring->gpu_addr) >> 2)); 987 988 /* program the RB_BASE for ring buffer */ 989 WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW, 990 lower_32_bits(ring->gpu_addr)); 991 WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH, 992 upper_32_bits(ring->gpu_addr)); 993 994 /* Initialize the ring buffer's read and write pointers */ 995 WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR, 0); 996 997 WREG32_SOC15(VCN, inst_idx, mmUVD_SCRATCH2, 0); 998 999 ring->wptr = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR); 1000 WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR, 1001 lower_32_bits(ring->wptr)); 1002 1003 fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET; 1004 /* Unstall DPG */ 1005 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 1006 0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK); 1007 1008 return 0; 1009 } 1010 1011 static int vcn_v2_5_start(struct amdgpu_device *adev) 1012 { 1013 struct amdgpu_ring *ring; 1014 uint32_t rb_bufsz, tmp; 1015 int i, j, k, r; 1016 1017 if (adev->pm.dpm_enabled) 1018 amdgpu_dpm_enable_uvd(adev, true); 1019 1020 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { 1021 if (adev->vcn.harvest_config & (1 << i)) 1022 continue; 1023 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) { 1024 r = vcn_v2_5_start_dpg_mode(adev, i, adev->vcn.indirect_sram); 1025 continue; 1026 } 1027 1028 /* disable register anti-hang mechanism */ 1029 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_POWER_STATUS), 0, 1030 ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); 1031 1032 /* set uvd status busy */ 1033 tmp = RREG32_SOC15(VCN, i, mmUVD_STATUS) | UVD_STATUS__UVD_BUSY; 1034 WREG32_SOC15(VCN, i, mmUVD_STATUS, tmp); 1035 } 1036 1037 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) 1038 return 0; 1039 1040 /*SW clock gating */ 1041 vcn_v2_5_disable_clock_gating(adev); 1042 1043 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { 1044 if (adev->vcn.harvest_config & (1 << i)) 1045 continue; 1046 /* enable VCPU clock */ 1047 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 1048 UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK); 1049 1050 /* disable master interrupt */ 1051 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN), 0, 1052 ~UVD_MASTINT_EN__VCPU_EN_MASK); 1053 1054 /* setup mmUVD_LMI_CTRL */ 1055 tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL); 1056 tmp &= ~0xff; 1057 WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL, tmp | 0x8| 1058 UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK | 1059 UVD_LMI_CTRL__MASK_MC_URGENT_MASK | 1060 UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK | 1061 UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK); 1062 1063 /* setup mmUVD_MPC_CNTL */ 1064 tmp = RREG32_SOC15(VCN, i, mmUVD_MPC_CNTL); 1065 tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK; 1066 tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT; 1067 WREG32_SOC15(VCN, i, mmUVD_MPC_CNTL, tmp); 1068 1069 /* setup UVD_MPC_SET_MUXA0 */ 1070 WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXA0, 1071 ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) | 1072 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) | 1073 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) | 1074 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT))); 1075 1076 /* setup UVD_MPC_SET_MUXB0 */ 1077 WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXB0, 1078 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) | 1079 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) | 1080 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) | 1081 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT))); 1082 1083 /* setup mmUVD_MPC_SET_MUX */ 1084 WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUX, 1085 ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) | 1086 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) | 1087 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT))); 1088 } 1089 1090 vcn_v2_5_mc_resume(adev); 1091 1092 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { 1093 volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr; 1094 if (adev->vcn.harvest_config & (1 << i)) 1095 continue; 1096 /* VCN global tiling registers */ 1097 WREG32_SOC15(VCN, i, mmUVD_GFX8_ADDR_CONFIG, 1098 adev->gfx.config.gb_addr_config); 1099 WREG32_SOC15(VCN, i, mmUVD_GFX8_ADDR_CONFIG, 1100 adev->gfx.config.gb_addr_config); 1101 1102 /* enable LMI MC and UMC channels */ 1103 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_LMI_CTRL2), 0, 1104 ~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK); 1105 1106 /* unblock VCPU register access */ 1107 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL), 0, 1108 ~UVD_RB_ARB_CTRL__VCPU_DIS_MASK); 1109 1110 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0, 1111 ~UVD_VCPU_CNTL__BLK_RST_MASK); 1112 1113 for (k = 0; k < 10; ++k) { 1114 uint32_t status; 1115 1116 for (j = 0; j < 100; ++j) { 1117 status = RREG32_SOC15(VCN, i, mmUVD_STATUS); 1118 if (status & 2) 1119 break; 1120 if (amdgpu_emu_mode == 1) 1121 msleep(500); 1122 else 1123 mdelay(10); 1124 } 1125 r = 0; 1126 if (status & 2) 1127 break; 1128 1129 DRM_ERROR("VCN decode not responding, trying to reset the VCPU!!!\n"); 1130 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 1131 UVD_VCPU_CNTL__BLK_RST_MASK, 1132 ~UVD_VCPU_CNTL__BLK_RST_MASK); 1133 mdelay(10); 1134 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0, 1135 ~UVD_VCPU_CNTL__BLK_RST_MASK); 1136 1137 mdelay(10); 1138 r = -1; 1139 } 1140 1141 if (r) { 1142 DRM_ERROR("VCN decode not responding, giving up!!!\n"); 1143 return r; 1144 } 1145 1146 /* enable master interrupt */ 1147 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN), 1148 UVD_MASTINT_EN__VCPU_EN_MASK, 1149 ~UVD_MASTINT_EN__VCPU_EN_MASK); 1150 1151 /* clear the busy bit of VCN_STATUS */ 1152 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS), 0, 1153 ~(2 << UVD_STATUS__VCPU_REPORT__SHIFT)); 1154 1155 WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_VMID, 0); 1156 1157 ring = &adev->vcn.inst[i].ring_dec; 1158 /* force RBC into idle state */ 1159 rb_bufsz = order_base_2(ring->ring_size); 1160 tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz); 1161 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1); 1162 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1); 1163 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1); 1164 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1); 1165 WREG32_SOC15(VCN, i, mmUVD_RBC_RB_CNTL, tmp); 1166 1167 fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET; 1168 /* program the RB_BASE for ring buffer */ 1169 WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW, 1170 lower_32_bits(ring->gpu_addr)); 1171 WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH, 1172 upper_32_bits(ring->gpu_addr)); 1173 1174 /* Initialize the ring buffer's read and write pointers */ 1175 WREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR, 0); 1176 1177 ring->wptr = RREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR); 1178 WREG32_SOC15(VCN, i, mmUVD_RBC_RB_WPTR, 1179 lower_32_bits(ring->wptr)); 1180 fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET; 1181 1182 fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET; 1183 ring = &adev->vcn.inst[i].ring_enc[0]; 1184 WREG32_SOC15(VCN, i, mmUVD_RB_RPTR, lower_32_bits(ring->wptr)); 1185 WREG32_SOC15(VCN, i, mmUVD_RB_WPTR, lower_32_bits(ring->wptr)); 1186 WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO, ring->gpu_addr); 1187 WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr)); 1188 WREG32_SOC15(VCN, i, mmUVD_RB_SIZE, ring->ring_size / 4); 1189 fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET; 1190 1191 fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET; 1192 ring = &adev->vcn.inst[i].ring_enc[1]; 1193 WREG32_SOC15(VCN, i, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr)); 1194 WREG32_SOC15(VCN, i, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr)); 1195 WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO2, ring->gpu_addr); 1196 WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr)); 1197 WREG32_SOC15(VCN, i, mmUVD_RB_SIZE2, ring->ring_size / 4); 1198 fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET; 1199 } 1200 1201 return 0; 1202 } 1203 1204 static int vcn_v2_5_mmsch_start(struct amdgpu_device *adev, 1205 struct amdgpu_mm_table *table) 1206 { 1207 uint32_t data = 0, loop = 0, size = 0; 1208 uint64_t addr = table->gpu_addr; 1209 struct mmsch_v1_1_init_header *header = NULL; 1210 1211 header = (struct mmsch_v1_1_init_header *)table->cpu_addr; 1212 size = header->total_size; 1213 1214 /* 1215 * 1, write to vce_mmsch_vf_ctx_addr_lo/hi register with GPU mc addr of 1216 * memory descriptor location 1217 */ 1218 WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_LO, lower_32_bits(addr)); 1219 WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_HI, upper_32_bits(addr)); 1220 1221 /* 2, update vmid of descriptor */ 1222 data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID); 1223 data &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK; 1224 /* use domain0 for MM scheduler */ 1225 data |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT); 1226 WREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID, data); 1227 1228 /* 3, notify mmsch about the size of this descriptor */ 1229 WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_SIZE, size); 1230 1231 /* 4, set resp to zero */ 1232 WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP, 0); 1233 1234 /* 1235 * 5, kick off the initialization and wait until 1236 * VCE_MMSCH_VF_MAILBOX_RESP becomes non-zero 1237 */ 1238 WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_HOST, 0x10000001); 1239 1240 data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP); 1241 loop = 10; 1242 while ((data & 0x10000002) != 0x10000002) { 1243 udelay(100); 1244 data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP); 1245 loop--; 1246 if (!loop) 1247 break; 1248 } 1249 1250 if (!loop) { 1251 dev_err(adev->dev, 1252 "failed to init MMSCH, mmMMSCH_VF_MAILBOX_RESP = %x\n", 1253 data); 1254 return -EBUSY; 1255 } 1256 1257 return 0; 1258 } 1259 1260 static int vcn_v2_5_sriov_start(struct amdgpu_device *adev) 1261 { 1262 struct amdgpu_ring *ring; 1263 uint32_t offset, size, tmp, i, rb_bufsz; 1264 uint32_t table_size = 0; 1265 struct mmsch_v1_0_cmd_direct_write direct_wt = { { 0 } }; 1266 struct mmsch_v1_0_cmd_direct_read_modify_write direct_rd_mod_wt = { { 0 } }; 1267 struct mmsch_v1_0_cmd_end end = { { 0 } }; 1268 uint32_t *init_table = adev->virt.mm_table.cpu_addr; 1269 struct mmsch_v1_1_init_header *header = (struct mmsch_v1_1_init_header *)init_table; 1270 1271 direct_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_WRITE; 1272 direct_rd_mod_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE; 1273 end.cmd_header.command_type = MMSCH_COMMAND__END; 1274 1275 header->version = MMSCH_VERSION; 1276 header->total_size = sizeof(struct mmsch_v1_1_init_header) >> 2; 1277 init_table += header->total_size; 1278 1279 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { 1280 header->eng[i].table_offset = header->total_size; 1281 header->eng[i].init_status = 0; 1282 header->eng[i].table_size = 0; 1283 1284 table_size = 0; 1285 1286 MMSCH_V1_0_INSERT_DIRECT_RD_MOD_WT( 1287 SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS), 1288 ~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY); 1289 1290 size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw[i]->size + 4); 1291 /* mc resume*/ 1292 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) { 1293 MMSCH_V1_0_INSERT_DIRECT_WT( 1294 SOC15_REG_OFFSET(VCN, i, 1295 mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 1296 adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo); 1297 MMSCH_V1_0_INSERT_DIRECT_WT( 1298 SOC15_REG_OFFSET(VCN, i, 1299 mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 1300 adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi); 1301 offset = 0; 1302 MMSCH_V1_0_INSERT_DIRECT_WT( 1303 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET0), 0); 1304 } else { 1305 MMSCH_V1_0_INSERT_DIRECT_WT( 1306 SOC15_REG_OFFSET(VCN, i, 1307 mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 1308 lower_32_bits(adev->vcn.inst[i].gpu_addr)); 1309 MMSCH_V1_0_INSERT_DIRECT_WT( 1310 SOC15_REG_OFFSET(VCN, i, 1311 mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 1312 upper_32_bits(adev->vcn.inst[i].gpu_addr)); 1313 offset = size; 1314 MMSCH_V1_0_INSERT_DIRECT_WT( 1315 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET0), 1316 AMDGPU_UVD_FIRMWARE_OFFSET >> 3); 1317 } 1318 1319 MMSCH_V1_0_INSERT_DIRECT_WT( 1320 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE0), 1321 size); 1322 MMSCH_V1_0_INSERT_DIRECT_WT( 1323 SOC15_REG_OFFSET(VCN, i, 1324 mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 1325 lower_32_bits(adev->vcn.inst[i].gpu_addr + offset)); 1326 MMSCH_V1_0_INSERT_DIRECT_WT( 1327 SOC15_REG_OFFSET(VCN, i, 1328 mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 1329 upper_32_bits(adev->vcn.inst[i].gpu_addr + offset)); 1330 MMSCH_V1_0_INSERT_DIRECT_WT( 1331 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET1), 1332 0); 1333 MMSCH_V1_0_INSERT_DIRECT_WT( 1334 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE1), 1335 AMDGPU_VCN_STACK_SIZE); 1336 MMSCH_V1_0_INSERT_DIRECT_WT( 1337 SOC15_REG_OFFSET(VCN, i, 1338 mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW), 1339 lower_32_bits(adev->vcn.inst[i].gpu_addr + offset + 1340 AMDGPU_VCN_STACK_SIZE)); 1341 MMSCH_V1_0_INSERT_DIRECT_WT( 1342 SOC15_REG_OFFSET(VCN, i, 1343 mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH), 1344 upper_32_bits(adev->vcn.inst[i].gpu_addr + offset + 1345 AMDGPU_VCN_STACK_SIZE)); 1346 MMSCH_V1_0_INSERT_DIRECT_WT( 1347 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET2), 1348 0); 1349 MMSCH_V1_0_INSERT_DIRECT_WT( 1350 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE2), 1351 AMDGPU_VCN_CONTEXT_SIZE); 1352 1353 ring = &adev->vcn.inst[i].ring_enc[0]; 1354 ring->wptr = 0; 1355 1356 MMSCH_V1_0_INSERT_DIRECT_WT( 1357 SOC15_REG_OFFSET(VCN, i, mmUVD_RB_BASE_LO), 1358 lower_32_bits(ring->gpu_addr)); 1359 MMSCH_V1_0_INSERT_DIRECT_WT( 1360 SOC15_REG_OFFSET(VCN, i, mmUVD_RB_BASE_HI), 1361 upper_32_bits(ring->gpu_addr)); 1362 MMSCH_V1_0_INSERT_DIRECT_WT( 1363 SOC15_REG_OFFSET(VCN, i, mmUVD_RB_SIZE), 1364 ring->ring_size / 4); 1365 1366 ring = &adev->vcn.inst[i].ring_dec; 1367 ring->wptr = 0; 1368 MMSCH_V1_0_INSERT_DIRECT_WT( 1369 SOC15_REG_OFFSET(VCN, i, 1370 mmUVD_LMI_RBC_RB_64BIT_BAR_LOW), 1371 lower_32_bits(ring->gpu_addr)); 1372 MMSCH_V1_0_INSERT_DIRECT_WT( 1373 SOC15_REG_OFFSET(VCN, i, 1374 mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH), 1375 upper_32_bits(ring->gpu_addr)); 1376 1377 /* force RBC into idle state */ 1378 rb_bufsz = order_base_2(ring->ring_size); 1379 tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz); 1380 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1); 1381 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1); 1382 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1); 1383 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1); 1384 MMSCH_V1_0_INSERT_DIRECT_WT( 1385 SOC15_REG_OFFSET(VCN, i, mmUVD_RBC_RB_CNTL), tmp); 1386 1387 /* add end packet */ 1388 memcpy((void *)init_table, &end, sizeof(struct mmsch_v1_0_cmd_end)); 1389 table_size += sizeof(struct mmsch_v1_0_cmd_end) / 4; 1390 init_table += sizeof(struct mmsch_v1_0_cmd_end) / 4; 1391 1392 /* refine header */ 1393 header->eng[i].table_size = table_size; 1394 header->total_size += table_size; 1395 } 1396 1397 return vcn_v2_5_mmsch_start(adev, &adev->virt.mm_table); 1398 } 1399 1400 static int vcn_v2_5_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx) 1401 { 1402 uint32_t tmp; 1403 1404 /* Wait for power status to be 1 */ 1405 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1, 1406 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); 1407 1408 /* wait for read ptr to be equal to write ptr */ 1409 tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR); 1410 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF); 1411 1412 tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2); 1413 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF); 1414 1415 tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF; 1416 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF); 1417 1418 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1, 1419 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); 1420 1421 /* disable dynamic power gating mode */ 1422 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 0, 1423 ~UVD_POWER_STATUS__UVD_PG_MODE_MASK); 1424 1425 return 0; 1426 } 1427 1428 static int vcn_v2_5_stop(struct amdgpu_device *adev) 1429 { 1430 uint32_t tmp; 1431 int i, r = 0; 1432 1433 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { 1434 if (adev->vcn.harvest_config & (1 << i)) 1435 continue; 1436 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) { 1437 r = vcn_v2_5_stop_dpg_mode(adev, i); 1438 continue; 1439 } 1440 1441 /* wait for vcn idle */ 1442 r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7); 1443 if (r) 1444 return r; 1445 1446 tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK | 1447 UVD_LMI_STATUS__READ_CLEAN_MASK | 1448 UVD_LMI_STATUS__WRITE_CLEAN_MASK | 1449 UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK; 1450 r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp); 1451 if (r) 1452 return r; 1453 1454 /* block LMI UMC channel */ 1455 tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2); 1456 tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK; 1457 WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2, tmp); 1458 1459 tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK| 1460 UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK; 1461 r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp); 1462 if (r) 1463 return r; 1464 1465 /* block VCPU register access */ 1466 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL), 1467 UVD_RB_ARB_CTRL__VCPU_DIS_MASK, 1468 ~UVD_RB_ARB_CTRL__VCPU_DIS_MASK); 1469 1470 /* reset VCPU */ 1471 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 1472 UVD_VCPU_CNTL__BLK_RST_MASK, 1473 ~UVD_VCPU_CNTL__BLK_RST_MASK); 1474 1475 /* disable VCPU clock */ 1476 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0, 1477 ~(UVD_VCPU_CNTL__CLK_EN_MASK)); 1478 1479 /* clear status */ 1480 WREG32_SOC15(VCN, i, mmUVD_STATUS, 0); 1481 1482 vcn_v2_5_enable_clock_gating(adev); 1483 1484 /* enable register anti-hang mechanism */ 1485 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_POWER_STATUS), 1486 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, 1487 ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); 1488 } 1489 1490 if (adev->pm.dpm_enabled) 1491 amdgpu_dpm_enable_uvd(adev, false); 1492 1493 return 0; 1494 } 1495 1496 static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev, 1497 int inst_idx, struct dpg_pause_state *new_state) 1498 { 1499 struct amdgpu_ring *ring; 1500 uint32_t reg_data = 0; 1501 int ret_code = 0; 1502 1503 /* pause/unpause if state is changed */ 1504 if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) { 1505 DRM_DEBUG("dpg pause state changed %d -> %d", 1506 adev->vcn.inst[inst_idx].pause_state.fw_based, new_state->fw_based); 1507 reg_data = RREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE) & 1508 (~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK); 1509 1510 if (new_state->fw_based == VCN_DPG_STATE__PAUSE) { 1511 ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1, 1512 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); 1513 1514 if (!ret_code) { 1515 volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr; 1516 1517 /* pause DPG */ 1518 reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK; 1519 WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data); 1520 1521 /* wait for ACK */ 1522 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_DPG_PAUSE, 1523 UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK, 1524 UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK); 1525 1526 /* Stall DPG before WPTR/RPTR reset */ 1527 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 1528 UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK, 1529 ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK); 1530 1531 /* Restore */ 1532 fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET; 1533 ring = &adev->vcn.inst[inst_idx].ring_enc[0]; 1534 ring->wptr = 0; 1535 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO, ring->gpu_addr); 1536 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr)); 1537 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE, ring->ring_size / 4); 1538 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR, lower_32_bits(ring->wptr)); 1539 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR, lower_32_bits(ring->wptr)); 1540 fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET; 1541 1542 fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET; 1543 ring = &adev->vcn.inst[inst_idx].ring_enc[1]; 1544 ring->wptr = 0; 1545 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO2, ring->gpu_addr); 1546 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr)); 1547 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE2, ring->ring_size / 4); 1548 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr)); 1549 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr)); 1550 fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET; 1551 1552 /* Unstall DPG */ 1553 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 1554 0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK); 1555 1556 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1557 UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); 1558 } 1559 } else { 1560 reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK; 1561 WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data); 1562 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1, 1563 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); 1564 } 1565 adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based; 1566 } 1567 1568 return 0; 1569 } 1570 1571 /** 1572 * vcn_v2_5_dec_ring_get_rptr - get read pointer 1573 * 1574 * @ring: amdgpu_ring pointer 1575 * 1576 * Returns the current hardware read pointer 1577 */ 1578 static uint64_t vcn_v2_5_dec_ring_get_rptr(struct amdgpu_ring *ring) 1579 { 1580 struct amdgpu_device *adev = ring->adev; 1581 1582 return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_RPTR); 1583 } 1584 1585 /** 1586 * vcn_v2_5_dec_ring_get_wptr - get write pointer 1587 * 1588 * @ring: amdgpu_ring pointer 1589 * 1590 * Returns the current hardware write pointer 1591 */ 1592 static uint64_t vcn_v2_5_dec_ring_get_wptr(struct amdgpu_ring *ring) 1593 { 1594 struct amdgpu_device *adev = ring->adev; 1595 1596 if (ring->use_doorbell) 1597 return *ring->wptr_cpu_addr; 1598 else 1599 return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR); 1600 } 1601 1602 /** 1603 * vcn_v2_5_dec_ring_set_wptr - set write pointer 1604 * 1605 * @ring: amdgpu_ring pointer 1606 * 1607 * Commits the write pointer to the hardware 1608 */ 1609 static void vcn_v2_5_dec_ring_set_wptr(struct amdgpu_ring *ring) 1610 { 1611 struct amdgpu_device *adev = ring->adev; 1612 1613 if (ring->use_doorbell) { 1614 *ring->wptr_cpu_addr = lower_32_bits(ring->wptr); 1615 WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr)); 1616 } else { 1617 WREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr)); 1618 } 1619 } 1620 1621 static const struct amdgpu_ring_funcs vcn_v2_5_dec_ring_vm_funcs = { 1622 .type = AMDGPU_RING_TYPE_VCN_DEC, 1623 .align_mask = 0xf, 1624 .secure_submission_supported = true, 1625 .get_rptr = vcn_v2_5_dec_ring_get_rptr, 1626 .get_wptr = vcn_v2_5_dec_ring_get_wptr, 1627 .set_wptr = vcn_v2_5_dec_ring_set_wptr, 1628 .emit_frame_size = 1629 SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 + 1630 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 + 1631 8 + /* vcn_v2_0_dec_ring_emit_vm_flush */ 1632 14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */ 1633 6, 1634 .emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */ 1635 .emit_ib = vcn_v2_0_dec_ring_emit_ib, 1636 .emit_fence = vcn_v2_0_dec_ring_emit_fence, 1637 .emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush, 1638 .test_ring = vcn_v2_0_dec_ring_test_ring, 1639 .test_ib = amdgpu_vcn_dec_ring_test_ib, 1640 .insert_nop = vcn_v2_0_dec_ring_insert_nop, 1641 .insert_start = vcn_v2_0_dec_ring_insert_start, 1642 .insert_end = vcn_v2_0_dec_ring_insert_end, 1643 .pad_ib = amdgpu_ring_generic_pad_ib, 1644 .begin_use = amdgpu_vcn_ring_begin_use, 1645 .end_use = amdgpu_vcn_ring_end_use, 1646 .emit_wreg = vcn_v2_0_dec_ring_emit_wreg, 1647 .emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait, 1648 .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper, 1649 }; 1650 1651 /** 1652 * vcn_v2_5_enc_ring_get_rptr - get enc read pointer 1653 * 1654 * @ring: amdgpu_ring pointer 1655 * 1656 * Returns the current hardware enc read pointer 1657 */ 1658 static uint64_t vcn_v2_5_enc_ring_get_rptr(struct amdgpu_ring *ring) 1659 { 1660 struct amdgpu_device *adev = ring->adev; 1661 1662 if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) 1663 return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR); 1664 else 1665 return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR2); 1666 } 1667 1668 /** 1669 * vcn_v2_5_enc_ring_get_wptr - get enc write pointer 1670 * 1671 * @ring: amdgpu_ring pointer 1672 * 1673 * Returns the current hardware enc write pointer 1674 */ 1675 static uint64_t vcn_v2_5_enc_ring_get_wptr(struct amdgpu_ring *ring) 1676 { 1677 struct amdgpu_device *adev = ring->adev; 1678 1679 if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) { 1680 if (ring->use_doorbell) 1681 return *ring->wptr_cpu_addr; 1682 else 1683 return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR); 1684 } else { 1685 if (ring->use_doorbell) 1686 return *ring->wptr_cpu_addr; 1687 else 1688 return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2); 1689 } 1690 } 1691 1692 /** 1693 * vcn_v2_5_enc_ring_set_wptr - set enc write pointer 1694 * 1695 * @ring: amdgpu_ring pointer 1696 * 1697 * Commits the enc write pointer to the hardware 1698 */ 1699 static void vcn_v2_5_enc_ring_set_wptr(struct amdgpu_ring *ring) 1700 { 1701 struct amdgpu_device *adev = ring->adev; 1702 1703 if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) { 1704 if (ring->use_doorbell) { 1705 *ring->wptr_cpu_addr = lower_32_bits(ring->wptr); 1706 WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr)); 1707 } else { 1708 WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR, lower_32_bits(ring->wptr)); 1709 } 1710 } else { 1711 if (ring->use_doorbell) { 1712 *ring->wptr_cpu_addr = lower_32_bits(ring->wptr); 1713 WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr)); 1714 } else { 1715 WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr)); 1716 } 1717 } 1718 } 1719 1720 static const struct amdgpu_ring_funcs vcn_v2_5_enc_ring_vm_funcs = { 1721 .type = AMDGPU_RING_TYPE_VCN_ENC, 1722 .align_mask = 0x3f, 1723 .nop = VCN_ENC_CMD_NO_OP, 1724 .get_rptr = vcn_v2_5_enc_ring_get_rptr, 1725 .get_wptr = vcn_v2_5_enc_ring_get_wptr, 1726 .set_wptr = vcn_v2_5_enc_ring_set_wptr, 1727 .emit_frame_size = 1728 SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 + 1729 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 + 1730 4 + /* vcn_v2_0_enc_ring_emit_vm_flush */ 1731 5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */ 1732 1, /* vcn_v2_0_enc_ring_insert_end */ 1733 .emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */ 1734 .emit_ib = vcn_v2_0_enc_ring_emit_ib, 1735 .emit_fence = vcn_v2_0_enc_ring_emit_fence, 1736 .emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush, 1737 .test_ring = amdgpu_vcn_enc_ring_test_ring, 1738 .test_ib = amdgpu_vcn_enc_ring_test_ib, 1739 .insert_nop = amdgpu_ring_insert_nop, 1740 .insert_end = vcn_v2_0_enc_ring_insert_end, 1741 .pad_ib = amdgpu_ring_generic_pad_ib, 1742 .begin_use = amdgpu_vcn_ring_begin_use, 1743 .end_use = amdgpu_vcn_ring_end_use, 1744 .emit_wreg = vcn_v2_0_enc_ring_emit_wreg, 1745 .emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait, 1746 .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper, 1747 }; 1748 1749 static void vcn_v2_5_set_dec_ring_funcs(struct amdgpu_device *adev) 1750 { 1751 int i; 1752 1753 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { 1754 if (adev->vcn.harvest_config & (1 << i)) 1755 continue; 1756 adev->vcn.inst[i].ring_dec.funcs = &vcn_v2_5_dec_ring_vm_funcs; 1757 adev->vcn.inst[i].ring_dec.me = i; 1758 } 1759 } 1760 1761 static void vcn_v2_5_set_enc_ring_funcs(struct amdgpu_device *adev) 1762 { 1763 int i, j; 1764 1765 for (j = 0; j < adev->vcn.num_vcn_inst; ++j) { 1766 if (adev->vcn.harvest_config & (1 << j)) 1767 continue; 1768 for (i = 0; i < adev->vcn.num_enc_rings; ++i) { 1769 adev->vcn.inst[j].ring_enc[i].funcs = &vcn_v2_5_enc_ring_vm_funcs; 1770 adev->vcn.inst[j].ring_enc[i].me = j; 1771 } 1772 } 1773 } 1774 1775 static bool vcn_v2_5_is_idle(void *handle) 1776 { 1777 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1778 int i, ret = 1; 1779 1780 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { 1781 if (adev->vcn.harvest_config & (1 << i)) 1782 continue; 1783 ret &= (RREG32_SOC15(VCN, i, mmUVD_STATUS) == UVD_STATUS__IDLE); 1784 } 1785 1786 return ret; 1787 } 1788 1789 static int vcn_v2_5_wait_for_idle(void *handle) 1790 { 1791 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1792 int i, ret = 0; 1793 1794 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { 1795 if (adev->vcn.harvest_config & (1 << i)) 1796 continue; 1797 ret = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE, 1798 UVD_STATUS__IDLE); 1799 if (ret) 1800 return ret; 1801 } 1802 1803 return ret; 1804 } 1805 1806 static int vcn_v2_5_set_clockgating_state(void *handle, 1807 enum amd_clockgating_state state) 1808 { 1809 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1810 bool enable = (state == AMD_CG_STATE_GATE); 1811 1812 if (amdgpu_sriov_vf(adev)) 1813 return 0; 1814 1815 if (enable) { 1816 if (!vcn_v2_5_is_idle(handle)) 1817 return -EBUSY; 1818 vcn_v2_5_enable_clock_gating(adev); 1819 } else { 1820 vcn_v2_5_disable_clock_gating(adev); 1821 } 1822 1823 return 0; 1824 } 1825 1826 static int vcn_v2_5_set_powergating_state(void *handle, 1827 enum amd_powergating_state state) 1828 { 1829 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1830 int ret; 1831 1832 if (amdgpu_sriov_vf(adev)) 1833 return 0; 1834 1835 if(state == adev->vcn.cur_state) 1836 return 0; 1837 1838 if (state == AMD_PG_STATE_GATE) 1839 ret = vcn_v2_5_stop(adev); 1840 else 1841 ret = vcn_v2_5_start(adev); 1842 1843 if(!ret) 1844 adev->vcn.cur_state = state; 1845 1846 return ret; 1847 } 1848 1849 static int vcn_v2_5_set_interrupt_state(struct amdgpu_device *adev, 1850 struct amdgpu_irq_src *source, 1851 unsigned type, 1852 enum amdgpu_interrupt_state state) 1853 { 1854 return 0; 1855 } 1856 1857 static int vcn_v2_6_set_ras_interrupt_state(struct amdgpu_device *adev, 1858 struct amdgpu_irq_src *source, 1859 unsigned int type, 1860 enum amdgpu_interrupt_state state) 1861 { 1862 return 0; 1863 } 1864 1865 static int vcn_v2_5_process_interrupt(struct amdgpu_device *adev, 1866 struct amdgpu_irq_src *source, 1867 struct amdgpu_iv_entry *entry) 1868 { 1869 uint32_t ip_instance; 1870 1871 switch (entry->client_id) { 1872 case SOC15_IH_CLIENTID_VCN: 1873 ip_instance = 0; 1874 break; 1875 case SOC15_IH_CLIENTID_VCN1: 1876 ip_instance = 1; 1877 break; 1878 default: 1879 DRM_ERROR("Unhandled client id: %d\n", entry->client_id); 1880 return 0; 1881 } 1882 1883 DRM_DEBUG("IH: VCN TRAP\n"); 1884 1885 switch (entry->src_id) { 1886 case VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT: 1887 amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_dec); 1888 break; 1889 case VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE: 1890 amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]); 1891 break; 1892 case VCN_2_0__SRCID__UVD_ENC_LOW_LATENCY: 1893 amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[1]); 1894 break; 1895 default: 1896 DRM_ERROR("Unhandled interrupt: %d %d\n", 1897 entry->src_id, entry->src_data[0]); 1898 break; 1899 } 1900 1901 return 0; 1902 } 1903 1904 static const struct amdgpu_irq_src_funcs vcn_v2_5_irq_funcs = { 1905 .set = vcn_v2_5_set_interrupt_state, 1906 .process = vcn_v2_5_process_interrupt, 1907 }; 1908 1909 static const struct amdgpu_irq_src_funcs vcn_v2_6_ras_irq_funcs = { 1910 .set = vcn_v2_6_set_ras_interrupt_state, 1911 .process = amdgpu_vcn_process_poison_irq, 1912 }; 1913 1914 static void vcn_v2_5_set_irq_funcs(struct amdgpu_device *adev) 1915 { 1916 int i; 1917 1918 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { 1919 if (adev->vcn.harvest_config & (1 << i)) 1920 continue; 1921 adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1; 1922 adev->vcn.inst[i].irq.funcs = &vcn_v2_5_irq_funcs; 1923 1924 adev->vcn.inst[i].ras_poison_irq.num_types = adev->vcn.num_enc_rings + 1; 1925 adev->vcn.inst[i].ras_poison_irq.funcs = &vcn_v2_6_ras_irq_funcs; 1926 } 1927 } 1928 1929 static void vcn_v2_5_print_ip_state(void *handle, struct drm_printer *p) 1930 { 1931 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1932 int i, j; 1933 uint32_t reg_count = ARRAY_SIZE(vcn_reg_list_2_5); 1934 uint32_t inst_off, is_powered; 1935 1936 if (!adev->vcn.ip_dump) 1937 return; 1938 1939 drm_printf(p, "num_instances:%d\n", adev->vcn.num_vcn_inst); 1940 for (i = 0; i < adev->vcn.num_vcn_inst; i++) { 1941 if (adev->vcn.harvest_config & (1 << i)) { 1942 drm_printf(p, "\nHarvested Instance:VCN%d Skipping dump\n", i); 1943 continue; 1944 } 1945 1946 inst_off = i * reg_count; 1947 is_powered = (adev->vcn.ip_dump[inst_off] & 1948 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK) != 1; 1949 1950 if (is_powered) { 1951 drm_printf(p, "\nActive Instance:VCN%d\n", i); 1952 for (j = 0; j < reg_count; j++) 1953 drm_printf(p, "%-50s \t 0x%08x\n", vcn_reg_list_2_5[j].reg_name, 1954 adev->vcn.ip_dump[inst_off + j]); 1955 } else { 1956 drm_printf(p, "\nInactive Instance:VCN%d\n", i); 1957 } 1958 } 1959 } 1960 1961 static void vcn_v2_5_dump_ip_state(void *handle) 1962 { 1963 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1964 int i, j; 1965 bool is_powered; 1966 uint32_t inst_off; 1967 uint32_t reg_count = ARRAY_SIZE(vcn_reg_list_2_5); 1968 1969 if (!adev->vcn.ip_dump) 1970 return; 1971 1972 for (i = 0; i < adev->vcn.num_vcn_inst; i++) { 1973 if (adev->vcn.harvest_config & (1 << i)) 1974 continue; 1975 1976 inst_off = i * reg_count; 1977 /* mmUVD_POWER_STATUS is always readable and is first element of the array */ 1978 adev->vcn.ip_dump[inst_off] = RREG32_SOC15(VCN, i, mmUVD_POWER_STATUS); 1979 is_powered = (adev->vcn.ip_dump[inst_off] & 1980 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK) != 1; 1981 1982 if (is_powered) 1983 for (j = 1; j < reg_count; j++) 1984 adev->vcn.ip_dump[inst_off + j] = 1985 RREG32(SOC15_REG_ENTRY_OFFSET_INST(vcn_reg_list_2_5[j], i)); 1986 } 1987 } 1988 1989 static const struct amd_ip_funcs vcn_v2_5_ip_funcs = { 1990 .name = "vcn_v2_5", 1991 .early_init = vcn_v2_5_early_init, 1992 .late_init = NULL, 1993 .sw_init = vcn_v2_5_sw_init, 1994 .sw_fini = vcn_v2_5_sw_fini, 1995 .hw_init = vcn_v2_5_hw_init, 1996 .hw_fini = vcn_v2_5_hw_fini, 1997 .suspend = vcn_v2_5_suspend, 1998 .resume = vcn_v2_5_resume, 1999 .is_idle = vcn_v2_5_is_idle, 2000 .wait_for_idle = vcn_v2_5_wait_for_idle, 2001 .check_soft_reset = NULL, 2002 .pre_soft_reset = NULL, 2003 .soft_reset = NULL, 2004 .post_soft_reset = NULL, 2005 .set_clockgating_state = vcn_v2_5_set_clockgating_state, 2006 .set_powergating_state = vcn_v2_5_set_powergating_state, 2007 .dump_ip_state = vcn_v2_5_dump_ip_state, 2008 .print_ip_state = vcn_v2_5_print_ip_state, 2009 }; 2010 2011 static const struct amd_ip_funcs vcn_v2_6_ip_funcs = { 2012 .name = "vcn_v2_6", 2013 .early_init = vcn_v2_5_early_init, 2014 .late_init = NULL, 2015 .sw_init = vcn_v2_5_sw_init, 2016 .sw_fini = vcn_v2_5_sw_fini, 2017 .hw_init = vcn_v2_5_hw_init, 2018 .hw_fini = vcn_v2_5_hw_fini, 2019 .suspend = vcn_v2_5_suspend, 2020 .resume = vcn_v2_5_resume, 2021 .is_idle = vcn_v2_5_is_idle, 2022 .wait_for_idle = vcn_v2_5_wait_for_idle, 2023 .check_soft_reset = NULL, 2024 .pre_soft_reset = NULL, 2025 .soft_reset = NULL, 2026 .post_soft_reset = NULL, 2027 .set_clockgating_state = vcn_v2_5_set_clockgating_state, 2028 .set_powergating_state = vcn_v2_5_set_powergating_state, 2029 .dump_ip_state = vcn_v2_5_dump_ip_state, 2030 .print_ip_state = vcn_v2_5_print_ip_state, 2031 }; 2032 2033 const struct amdgpu_ip_block_version vcn_v2_5_ip_block = 2034 { 2035 .type = AMD_IP_BLOCK_TYPE_VCN, 2036 .major = 2, 2037 .minor = 5, 2038 .rev = 0, 2039 .funcs = &vcn_v2_5_ip_funcs, 2040 }; 2041 2042 const struct amdgpu_ip_block_version vcn_v2_6_ip_block = 2043 { 2044 .type = AMD_IP_BLOCK_TYPE_VCN, 2045 .major = 2, 2046 .minor = 6, 2047 .rev = 0, 2048 .funcs = &vcn_v2_6_ip_funcs, 2049 }; 2050 2051 static uint32_t vcn_v2_6_query_poison_by_instance(struct amdgpu_device *adev, 2052 uint32_t instance, uint32_t sub_block) 2053 { 2054 uint32_t poison_stat = 0, reg_value = 0; 2055 2056 switch (sub_block) { 2057 case AMDGPU_VCN_V2_6_VCPU_VCODEC: 2058 reg_value = RREG32_SOC15(VCN, instance, mmUVD_RAS_VCPU_VCODEC_STATUS); 2059 poison_stat = REG_GET_FIELD(reg_value, UVD_RAS_VCPU_VCODEC_STATUS, POISONED_PF); 2060 break; 2061 default: 2062 break; 2063 } 2064 2065 if (poison_stat) 2066 dev_info(adev->dev, "Poison detected in VCN%d, sub_block%d\n", 2067 instance, sub_block); 2068 2069 return poison_stat; 2070 } 2071 2072 static bool vcn_v2_6_query_poison_status(struct amdgpu_device *adev) 2073 { 2074 uint32_t inst, sub; 2075 uint32_t poison_stat = 0; 2076 2077 for (inst = 0; inst < adev->vcn.num_vcn_inst; inst++) 2078 for (sub = 0; sub < AMDGPU_VCN_V2_6_MAX_SUB_BLOCK; sub++) 2079 poison_stat += 2080 vcn_v2_6_query_poison_by_instance(adev, inst, sub); 2081 2082 return !!poison_stat; 2083 } 2084 2085 const struct amdgpu_ras_block_hw_ops vcn_v2_6_ras_hw_ops = { 2086 .query_poison_status = vcn_v2_6_query_poison_status, 2087 }; 2088 2089 static struct amdgpu_vcn_ras vcn_v2_6_ras = { 2090 .ras_block = { 2091 .hw_ops = &vcn_v2_6_ras_hw_ops, 2092 .ras_late_init = amdgpu_vcn_ras_late_init, 2093 }, 2094 }; 2095 2096 static void vcn_v2_5_set_ras_funcs(struct amdgpu_device *adev) 2097 { 2098 switch (amdgpu_ip_version(adev, VCN_HWIP, 0)) { 2099 case IP_VERSION(2, 6, 0): 2100 adev->vcn.ras = &vcn_v2_6_ras; 2101 break; 2102 default: 2103 break; 2104 } 2105 } 2106