1 /* 2 * Copyright 2018 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 "soc15.h" 30 #include "soc15d.h" 31 #include "amdgpu_pm.h" 32 #include "amdgpu_psp.h" 33 #include "mmsch_v2_0.h" 34 #include "vcn_v2_0.h" 35 36 #include "vcn/vcn_2_0_0_offset.h" 37 #include "vcn/vcn_2_0_0_sh_mask.h" 38 #include "ivsrcid/vcn/irqsrcs_vcn_2_0.h" 39 40 #define mmUVD_CONTEXT_ID_INTERNAL_OFFSET 0x1fd 41 #define mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET 0x503 42 #define mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET 0x504 43 #define mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET 0x505 44 #define mmUVD_NO_OP_INTERNAL_OFFSET 0x53f 45 #define mmUVD_GP_SCRATCH8_INTERNAL_OFFSET 0x54a 46 #define mmUVD_SCRATCH9_INTERNAL_OFFSET 0xc01d 47 48 #define mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET 0x1e1 49 #define mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET 0x5a6 50 #define mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET 0x5a7 51 #define mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET 0x1e2 52 53 static void vcn_v2_0_set_dec_ring_funcs(struct amdgpu_device *adev); 54 static void vcn_v2_0_set_enc_ring_funcs(struct amdgpu_device *adev); 55 static void vcn_v2_0_set_irq_funcs(struct amdgpu_device *adev); 56 static int vcn_v2_0_set_powergating_state(void *handle, 57 enum amd_powergating_state state); 58 static int vcn_v2_0_pause_dpg_mode(struct amdgpu_device *adev, 59 int inst_idx, struct dpg_pause_state *new_state); 60 static int vcn_v2_0_start_sriov(struct amdgpu_device *adev); 61 /** 62 * vcn_v2_0_early_init - set function pointers 63 * 64 * @handle: amdgpu_device pointer 65 * 66 * Set ring and irq function pointers 67 */ 68 static int vcn_v2_0_early_init(void *handle) 69 { 70 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 71 72 adev->vcn.num_vcn_inst = 1; 73 if (amdgpu_sriov_vf(adev)) 74 adev->vcn.num_enc_rings = 1; 75 else 76 adev->vcn.num_enc_rings = 2; 77 78 vcn_v2_0_set_dec_ring_funcs(adev); 79 vcn_v2_0_set_enc_ring_funcs(adev); 80 vcn_v2_0_set_irq_funcs(adev); 81 82 return 0; 83 } 84 85 /** 86 * vcn_v2_0_sw_init - sw init for VCN block 87 * 88 * @handle: amdgpu_device pointer 89 * 90 * Load firmware and sw initialization 91 */ 92 static int vcn_v2_0_sw_init(void *handle) 93 { 94 struct amdgpu_ring *ring; 95 int i, r; 96 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 97 volatile struct amdgpu_fw_shared *fw_shared; 98 99 /* VCN DEC TRAP */ 100 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_VCN, 101 VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT, 102 &adev->vcn.inst->irq); 103 if (r) 104 return r; 105 106 /* VCN ENC TRAP */ 107 for (i = 0; i < adev->vcn.num_enc_rings; ++i) { 108 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_VCN, 109 i + VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE, 110 &adev->vcn.inst->irq); 111 if (r) 112 return r; 113 } 114 115 r = amdgpu_vcn_sw_init(adev); 116 if (r) 117 return r; 118 119 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) { 120 const struct common_firmware_header *hdr; 121 hdr = (const struct common_firmware_header *)adev->vcn.fw->data; 122 adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].ucode_id = AMDGPU_UCODE_ID_VCN; 123 adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].fw = adev->vcn.fw; 124 adev->firmware.fw_size += 125 roundup2(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE); 126 dev_info(adev->dev, "Will use PSP to load VCN firmware\n"); 127 } 128 129 r = amdgpu_vcn_resume(adev); 130 if (r) 131 return r; 132 133 ring = &adev->vcn.inst->ring_dec; 134 135 ring->use_doorbell = true; 136 ring->doorbell_index = adev->doorbell_index.vcn.vcn_ring0_1 << 1; 137 138 snprintf(ring->name, sizeof(ring->name), "vcn_dec"); 139 r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst->irq, 0, 140 AMDGPU_RING_PRIO_DEFAULT, NULL); 141 if (r) 142 return r; 143 144 adev->vcn.internal.context_id = mmUVD_CONTEXT_ID_INTERNAL_OFFSET; 145 adev->vcn.internal.ib_vmid = mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET; 146 adev->vcn.internal.ib_bar_low = mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET; 147 adev->vcn.internal.ib_bar_high = mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET; 148 adev->vcn.internal.ib_size = mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET; 149 adev->vcn.internal.gp_scratch8 = mmUVD_GP_SCRATCH8_INTERNAL_OFFSET; 150 151 adev->vcn.internal.scratch9 = mmUVD_SCRATCH9_INTERNAL_OFFSET; 152 adev->vcn.inst->external.scratch9 = SOC15_REG_OFFSET(UVD, 0, mmUVD_SCRATCH9); 153 adev->vcn.internal.data0 = mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET; 154 adev->vcn.inst->external.data0 = SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0); 155 adev->vcn.internal.data1 = mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET; 156 adev->vcn.inst->external.data1 = SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1); 157 adev->vcn.internal.cmd = mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET; 158 adev->vcn.inst->external.cmd = SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD); 159 adev->vcn.internal.nop = mmUVD_NO_OP_INTERNAL_OFFSET; 160 adev->vcn.inst->external.nop = SOC15_REG_OFFSET(UVD, 0, mmUVD_NO_OP); 161 162 for (i = 0; i < adev->vcn.num_enc_rings; ++i) { 163 ring = &adev->vcn.inst->ring_enc[i]; 164 ring->use_doorbell = true; 165 if (!amdgpu_sriov_vf(adev)) 166 ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 2 + i; 167 else 168 ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 1 + i; 169 snprintf(ring->name, sizeof(ring->name), "vcn_enc%d", i); 170 r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst->irq, 0, 171 AMDGPU_RING_PRIO_DEFAULT, NULL); 172 if (r) 173 return r; 174 } 175 176 adev->vcn.pause_dpg_mode = vcn_v2_0_pause_dpg_mode; 177 178 r = amdgpu_virt_alloc_mm_table(adev); 179 if (r) 180 return r; 181 182 fw_shared = adev->vcn.inst->fw_shared_cpu_addr; 183 fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_VCN_MULTI_QUEUE_FLAG); 184 return 0; 185 } 186 187 /** 188 * vcn_v2_0_sw_fini - sw fini for VCN block 189 * 190 * @handle: amdgpu_device pointer 191 * 192 * VCN suspend and free up sw allocation 193 */ 194 static int vcn_v2_0_sw_fini(void *handle) 195 { 196 int r, idx; 197 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 198 volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst->fw_shared_cpu_addr; 199 200 if (drm_dev_enter(&adev->ddev, &idx)) { 201 fw_shared->present_flag_0 = 0; 202 drm_dev_exit(idx); 203 } 204 205 amdgpu_virt_free_mm_table(adev); 206 207 r = amdgpu_vcn_suspend(adev); 208 if (r) 209 return r; 210 211 r = amdgpu_vcn_sw_fini(adev); 212 213 return r; 214 } 215 216 /** 217 * vcn_v2_0_hw_init - start and test VCN block 218 * 219 * @handle: amdgpu_device pointer 220 * 221 * Initialize the hardware, boot up the VCPU and do some testing 222 */ 223 static int vcn_v2_0_hw_init(void *handle) 224 { 225 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 226 struct amdgpu_ring *ring = &adev->vcn.inst->ring_dec; 227 int i, r; 228 229 adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell, 230 ring->doorbell_index, 0); 231 232 if (amdgpu_sriov_vf(adev)) 233 vcn_v2_0_start_sriov(adev); 234 235 r = amdgpu_ring_test_helper(ring); 236 if (r) 237 goto done; 238 239 //Disable vcn decode for sriov 240 if (amdgpu_sriov_vf(adev)) 241 ring->sched.ready = false; 242 243 for (i = 0; i < adev->vcn.num_enc_rings; ++i) { 244 ring = &adev->vcn.inst->ring_enc[i]; 245 r = amdgpu_ring_test_helper(ring); 246 if (r) 247 goto done; 248 } 249 250 done: 251 if (!r) 252 DRM_INFO("VCN decode and encode initialized successfully(under %s).\n", 253 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode"); 254 255 return r; 256 } 257 258 /** 259 * vcn_v2_0_hw_fini - stop the hardware block 260 * 261 * @handle: amdgpu_device pointer 262 * 263 * Stop the VCN block, mark ring as not ready any more 264 */ 265 static int vcn_v2_0_hw_fini(void *handle) 266 { 267 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 268 269 cancel_delayed_work_sync(&adev->vcn.idle_work); 270 271 if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) || 272 (adev->vcn.cur_state != AMD_PG_STATE_GATE && 273 RREG32_SOC15(VCN, 0, mmUVD_STATUS))) 274 vcn_v2_0_set_powergating_state(adev, AMD_PG_STATE_GATE); 275 276 return 0; 277 } 278 279 /** 280 * vcn_v2_0_suspend - suspend VCN block 281 * 282 * @handle: amdgpu_device pointer 283 * 284 * HW fini and suspend VCN block 285 */ 286 static int vcn_v2_0_suspend(void *handle) 287 { 288 int r; 289 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 290 291 r = vcn_v2_0_hw_fini(adev); 292 if (r) 293 return r; 294 295 r = amdgpu_vcn_suspend(adev); 296 297 return r; 298 } 299 300 /** 301 * vcn_v2_0_resume - resume VCN block 302 * 303 * @handle: amdgpu_device pointer 304 * 305 * Resume firmware and hw init VCN block 306 */ 307 static int vcn_v2_0_resume(void *handle) 308 { 309 int r; 310 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 311 312 r = amdgpu_vcn_resume(adev); 313 if (r) 314 return r; 315 316 r = vcn_v2_0_hw_init(adev); 317 318 return r; 319 } 320 321 /** 322 * vcn_v2_0_mc_resume - memory controller programming 323 * 324 * @adev: amdgpu_device pointer 325 * 326 * Let the VCN memory controller know it's offsets 327 */ 328 static void vcn_v2_0_mc_resume(struct amdgpu_device *adev) 329 { 330 uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4); 331 uint32_t offset; 332 333 if (amdgpu_sriov_vf(adev)) 334 return; 335 336 /* cache window 0: fw */ 337 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) { 338 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW, 339 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_lo)); 340 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH, 341 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_hi)); 342 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0, 0); 343 offset = 0; 344 } else { 345 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW, 346 lower_32_bits(adev->vcn.inst->gpu_addr)); 347 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH, 348 upper_32_bits(adev->vcn.inst->gpu_addr)); 349 offset = size; 350 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0, 351 AMDGPU_UVD_FIRMWARE_OFFSET >> 3); 352 } 353 354 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE0, size); 355 356 /* cache window 1: stack */ 357 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW, 358 lower_32_bits(adev->vcn.inst->gpu_addr + offset)); 359 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH, 360 upper_32_bits(adev->vcn.inst->gpu_addr + offset)); 361 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET1, 0); 362 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE); 363 364 /* cache window 2: context */ 365 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW, 366 lower_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE)); 367 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH, 368 upper_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE)); 369 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET2, 0); 370 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE); 371 372 /* non-cache window */ 373 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW, 374 lower_32_bits(adev->vcn.inst->fw_shared_gpu_addr)); 375 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH, 376 upper_32_bits(adev->vcn.inst->fw_shared_gpu_addr)); 377 WREG32_SOC15(UVD, 0, mmUVD_VCPU_NONCACHE_OFFSET0, 0); 378 WREG32_SOC15(UVD, 0, mmUVD_VCPU_NONCACHE_SIZE0, 379 AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared))); 380 381 WREG32_SOC15(UVD, 0, mmUVD_GFX10_ADDR_CONFIG, adev->gfx.config.gb_addr_config); 382 } 383 384 static void vcn_v2_0_mc_resume_dpg_mode(struct amdgpu_device *adev, bool indirect) 385 { 386 uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4); 387 uint32_t offset; 388 389 /* cache window 0: fw */ 390 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) { 391 if (!indirect) { 392 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 393 UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 394 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_lo), 0, indirect); 395 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 396 UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 397 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_hi), 0, indirect); 398 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 399 UVD, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect); 400 } else { 401 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 402 UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect); 403 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 404 UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect); 405 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 406 UVD, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect); 407 } 408 offset = 0; 409 } else { 410 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 411 UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 412 lower_32_bits(adev->vcn.inst->gpu_addr), 0, indirect); 413 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 414 UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 415 upper_32_bits(adev->vcn.inst->gpu_addr), 0, indirect); 416 offset = size; 417 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 418 UVD, 0, mmUVD_VCPU_CACHE_OFFSET0), 419 AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect); 420 } 421 422 if (!indirect) 423 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 424 UVD, 0, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect); 425 else 426 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 427 UVD, 0, mmUVD_VCPU_CACHE_SIZE0), 0, 0, indirect); 428 429 /* cache window 1: stack */ 430 if (!indirect) { 431 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 432 UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 433 lower_32_bits(adev->vcn.inst->gpu_addr + offset), 0, indirect); 434 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 435 UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 436 upper_32_bits(adev->vcn.inst->gpu_addr + offset), 0, indirect); 437 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 438 UVD, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect); 439 } else { 440 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 441 UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect); 442 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 443 UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect); 444 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 445 UVD, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect); 446 } 447 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 448 UVD, 0, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect); 449 450 /* cache window 2: context */ 451 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 452 UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW), 453 lower_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect); 454 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 455 UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH), 456 upper_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect); 457 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 458 UVD, 0, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect); 459 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 460 UVD, 0, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect); 461 462 /* non-cache window */ 463 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 464 UVD, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW), 465 lower_32_bits(adev->vcn.inst->fw_shared_gpu_addr), 0, indirect); 466 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 467 UVD, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH), 468 upper_32_bits(adev->vcn.inst->fw_shared_gpu_addr), 0, indirect); 469 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 470 UVD, 0, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect); 471 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 472 UVD, 0, mmUVD_VCPU_NONCACHE_SIZE0), 473 AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)), 0, indirect); 474 475 /* VCN global tiling registers */ 476 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 477 UVD, 0, mmUVD_GFX10_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect); 478 } 479 480 /** 481 * vcn_v2_0_disable_clock_gating - disable VCN clock gating 482 * 483 * @adev: amdgpu_device pointer 484 * 485 * Disable clock gating for VCN block 486 */ 487 static void vcn_v2_0_disable_clock_gating(struct amdgpu_device *adev) 488 { 489 uint32_t data; 490 491 if (amdgpu_sriov_vf(adev)) 492 return; 493 494 /* UVD disable CGC */ 495 data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL); 496 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG) 497 data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 498 else 499 data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK; 500 data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT; 501 data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT; 502 WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data); 503 504 data = RREG32_SOC15(VCN, 0, mmUVD_CGC_GATE); 505 data &= ~(UVD_CGC_GATE__SYS_MASK 506 | UVD_CGC_GATE__UDEC_MASK 507 | UVD_CGC_GATE__MPEG2_MASK 508 | UVD_CGC_GATE__REGS_MASK 509 | UVD_CGC_GATE__RBC_MASK 510 | UVD_CGC_GATE__LMI_MC_MASK 511 | UVD_CGC_GATE__LMI_UMC_MASK 512 | UVD_CGC_GATE__IDCT_MASK 513 | UVD_CGC_GATE__MPRD_MASK 514 | UVD_CGC_GATE__MPC_MASK 515 | UVD_CGC_GATE__LBSI_MASK 516 | UVD_CGC_GATE__LRBBM_MASK 517 | UVD_CGC_GATE__UDEC_RE_MASK 518 | UVD_CGC_GATE__UDEC_CM_MASK 519 | UVD_CGC_GATE__UDEC_IT_MASK 520 | UVD_CGC_GATE__UDEC_DB_MASK 521 | UVD_CGC_GATE__UDEC_MP_MASK 522 | UVD_CGC_GATE__WCB_MASK 523 | UVD_CGC_GATE__VCPU_MASK 524 | UVD_CGC_GATE__SCPU_MASK); 525 WREG32_SOC15(VCN, 0, mmUVD_CGC_GATE, data); 526 527 data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL); 528 data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK 529 | UVD_CGC_CTRL__UDEC_CM_MODE_MASK 530 | UVD_CGC_CTRL__UDEC_IT_MODE_MASK 531 | UVD_CGC_CTRL__UDEC_DB_MODE_MASK 532 | UVD_CGC_CTRL__UDEC_MP_MODE_MASK 533 | UVD_CGC_CTRL__SYS_MODE_MASK 534 | UVD_CGC_CTRL__UDEC_MODE_MASK 535 | UVD_CGC_CTRL__MPEG2_MODE_MASK 536 | UVD_CGC_CTRL__REGS_MODE_MASK 537 | UVD_CGC_CTRL__RBC_MODE_MASK 538 | UVD_CGC_CTRL__LMI_MC_MODE_MASK 539 | UVD_CGC_CTRL__LMI_UMC_MODE_MASK 540 | UVD_CGC_CTRL__IDCT_MODE_MASK 541 | UVD_CGC_CTRL__MPRD_MODE_MASK 542 | UVD_CGC_CTRL__MPC_MODE_MASK 543 | UVD_CGC_CTRL__LBSI_MODE_MASK 544 | UVD_CGC_CTRL__LRBBM_MODE_MASK 545 | UVD_CGC_CTRL__WCB_MODE_MASK 546 | UVD_CGC_CTRL__VCPU_MODE_MASK 547 | UVD_CGC_CTRL__SCPU_MODE_MASK); 548 WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data); 549 550 /* turn on */ 551 data = RREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_GATE); 552 data |= (UVD_SUVD_CGC_GATE__SRE_MASK 553 | UVD_SUVD_CGC_GATE__SIT_MASK 554 | UVD_SUVD_CGC_GATE__SMP_MASK 555 | UVD_SUVD_CGC_GATE__SCM_MASK 556 | UVD_SUVD_CGC_GATE__SDB_MASK 557 | UVD_SUVD_CGC_GATE__SRE_H264_MASK 558 | UVD_SUVD_CGC_GATE__SRE_HEVC_MASK 559 | UVD_SUVD_CGC_GATE__SIT_H264_MASK 560 | UVD_SUVD_CGC_GATE__SIT_HEVC_MASK 561 | UVD_SUVD_CGC_GATE__SCM_H264_MASK 562 | UVD_SUVD_CGC_GATE__SCM_HEVC_MASK 563 | UVD_SUVD_CGC_GATE__SDB_H264_MASK 564 | UVD_SUVD_CGC_GATE__SDB_HEVC_MASK 565 | UVD_SUVD_CGC_GATE__SCLR_MASK 566 | UVD_SUVD_CGC_GATE__UVD_SC_MASK 567 | UVD_SUVD_CGC_GATE__ENT_MASK 568 | UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK 569 | UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK 570 | UVD_SUVD_CGC_GATE__SITE_MASK 571 | UVD_SUVD_CGC_GATE__SRE_VP9_MASK 572 | UVD_SUVD_CGC_GATE__SCM_VP9_MASK 573 | UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK 574 | UVD_SUVD_CGC_GATE__SDB_VP9_MASK 575 | UVD_SUVD_CGC_GATE__IME_HEVC_MASK); 576 WREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_GATE, data); 577 578 data = RREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL); 579 data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK 580 | UVD_SUVD_CGC_CTRL__SIT_MODE_MASK 581 | UVD_SUVD_CGC_CTRL__SMP_MODE_MASK 582 | UVD_SUVD_CGC_CTRL__SCM_MODE_MASK 583 | UVD_SUVD_CGC_CTRL__SDB_MODE_MASK 584 | UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK 585 | UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK 586 | UVD_SUVD_CGC_CTRL__ENT_MODE_MASK 587 | UVD_SUVD_CGC_CTRL__IME_MODE_MASK 588 | UVD_SUVD_CGC_CTRL__SITE_MODE_MASK); 589 WREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL, data); 590 } 591 592 static void vcn_v2_0_clock_gating_dpg_mode(struct amdgpu_device *adev, 593 uint8_t sram_sel, uint8_t indirect) 594 { 595 uint32_t reg_data = 0; 596 597 /* enable sw clock gating control */ 598 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG) 599 reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 600 else 601 reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 602 reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT; 603 reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT; 604 reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK | 605 UVD_CGC_CTRL__UDEC_CM_MODE_MASK | 606 UVD_CGC_CTRL__UDEC_IT_MODE_MASK | 607 UVD_CGC_CTRL__UDEC_DB_MODE_MASK | 608 UVD_CGC_CTRL__UDEC_MP_MODE_MASK | 609 UVD_CGC_CTRL__SYS_MODE_MASK | 610 UVD_CGC_CTRL__UDEC_MODE_MASK | 611 UVD_CGC_CTRL__MPEG2_MODE_MASK | 612 UVD_CGC_CTRL__REGS_MODE_MASK | 613 UVD_CGC_CTRL__RBC_MODE_MASK | 614 UVD_CGC_CTRL__LMI_MC_MODE_MASK | 615 UVD_CGC_CTRL__LMI_UMC_MODE_MASK | 616 UVD_CGC_CTRL__IDCT_MODE_MASK | 617 UVD_CGC_CTRL__MPRD_MODE_MASK | 618 UVD_CGC_CTRL__MPC_MODE_MASK | 619 UVD_CGC_CTRL__LBSI_MODE_MASK | 620 UVD_CGC_CTRL__LRBBM_MODE_MASK | 621 UVD_CGC_CTRL__WCB_MODE_MASK | 622 UVD_CGC_CTRL__VCPU_MODE_MASK | 623 UVD_CGC_CTRL__SCPU_MODE_MASK); 624 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 625 UVD, 0, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect); 626 627 /* turn off clock gating */ 628 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 629 UVD, 0, mmUVD_CGC_GATE), 0, sram_sel, indirect); 630 631 /* turn on SUVD clock gating */ 632 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 633 UVD, 0, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect); 634 635 /* turn on sw mode in UVD_SUVD_CGC_CTRL */ 636 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 637 UVD, 0, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect); 638 } 639 640 /** 641 * vcn_v2_0_enable_clock_gating - enable VCN clock gating 642 * 643 * @adev: amdgpu_device pointer 644 * 645 * Enable clock gating for VCN block 646 */ 647 static void vcn_v2_0_enable_clock_gating(struct amdgpu_device *adev) 648 { 649 uint32_t data = 0; 650 651 if (amdgpu_sriov_vf(adev)) 652 return; 653 654 /* enable UVD CGC */ 655 data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL); 656 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG) 657 data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 658 else 659 data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 660 data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT; 661 data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT; 662 WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data); 663 664 data = RREG32_SOC15(VCN, 0, 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__SCPU_MODE_MASK); 685 WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data); 686 687 data = RREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL); 688 data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK 689 | UVD_SUVD_CGC_CTRL__SIT_MODE_MASK 690 | UVD_SUVD_CGC_CTRL__SMP_MODE_MASK 691 | UVD_SUVD_CGC_CTRL__SCM_MODE_MASK 692 | UVD_SUVD_CGC_CTRL__SDB_MODE_MASK 693 | UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK 694 | UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK 695 | UVD_SUVD_CGC_CTRL__ENT_MODE_MASK 696 | UVD_SUVD_CGC_CTRL__IME_MODE_MASK 697 | UVD_SUVD_CGC_CTRL__SITE_MODE_MASK); 698 WREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL, data); 699 } 700 701 static void vcn_v2_0_disable_static_power_gating(struct amdgpu_device *adev) 702 { 703 uint32_t data = 0; 704 705 if (amdgpu_sriov_vf(adev)) 706 return; 707 708 if (adev->pg_flags & AMD_PG_SUPPORT_VCN) { 709 data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT 710 | 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT 711 | 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT 712 | 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT 713 | 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT 714 | 2 << UVD_PGFSM_CONFIG__UVDIL_PWR_CONFIG__SHIFT 715 | 2 << UVD_PGFSM_CONFIG__UVDIR_PWR_CONFIG__SHIFT 716 | 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT 717 | 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT 718 | 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT); 719 720 WREG32_SOC15(VCN, 0, mmUVD_PGFSM_CONFIG, data); 721 SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_PGFSM_STATUS, 722 UVD_PGFSM_STATUS__UVDM_UVDU_PWR_ON_2_0, 0xFFFFF); 723 } else { 724 data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT 725 | 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT 726 | 1 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT 727 | 1 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT 728 | 1 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT 729 | 1 << UVD_PGFSM_CONFIG__UVDIL_PWR_CONFIG__SHIFT 730 | 1 << UVD_PGFSM_CONFIG__UVDIR_PWR_CONFIG__SHIFT 731 | 1 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT 732 | 1 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT 733 | 1 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT); 734 WREG32_SOC15(VCN, 0, mmUVD_PGFSM_CONFIG, data); 735 SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_PGFSM_STATUS, 0, 0xFFFFF); 736 } 737 738 /* polling UVD_PGFSM_STATUS to confirm UVDM_PWR_STATUS, 739 * UVDU_PWR_STATUS are 0 (power on) */ 740 741 data = RREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS); 742 data &= ~0x103; 743 if (adev->pg_flags & AMD_PG_SUPPORT_VCN) 744 data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON | 745 UVD_POWER_STATUS__UVD_PG_EN_MASK; 746 747 WREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS, data); 748 } 749 750 static void vcn_v2_0_enable_static_power_gating(struct amdgpu_device *adev) 751 { 752 uint32_t data = 0; 753 754 if (amdgpu_sriov_vf(adev)) 755 return; 756 757 if (adev->pg_flags & AMD_PG_SUPPORT_VCN) { 758 /* Before power off, this indicator has to be turned on */ 759 data = RREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS); 760 data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK; 761 data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF; 762 WREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS, data); 763 764 765 data = (2 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT 766 | 2 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT 767 | 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT 768 | 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT 769 | 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT 770 | 2 << UVD_PGFSM_CONFIG__UVDIL_PWR_CONFIG__SHIFT 771 | 2 << UVD_PGFSM_CONFIG__UVDIR_PWR_CONFIG__SHIFT 772 | 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT 773 | 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT 774 | 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT); 775 776 WREG32_SOC15(VCN, 0, mmUVD_PGFSM_CONFIG, data); 777 778 data = (2 << UVD_PGFSM_STATUS__UVDM_PWR_STATUS__SHIFT 779 | 2 << UVD_PGFSM_STATUS__UVDU_PWR_STATUS__SHIFT 780 | 2 << UVD_PGFSM_STATUS__UVDF_PWR_STATUS__SHIFT 781 | 2 << UVD_PGFSM_STATUS__UVDC_PWR_STATUS__SHIFT 782 | 2 << UVD_PGFSM_STATUS__UVDB_PWR_STATUS__SHIFT 783 | 2 << UVD_PGFSM_STATUS__UVDIL_PWR_STATUS__SHIFT 784 | 2 << UVD_PGFSM_STATUS__UVDIR_PWR_STATUS__SHIFT 785 | 2 << UVD_PGFSM_STATUS__UVDTD_PWR_STATUS__SHIFT 786 | 2 << UVD_PGFSM_STATUS__UVDTE_PWR_STATUS__SHIFT 787 | 2 << UVD_PGFSM_STATUS__UVDE_PWR_STATUS__SHIFT); 788 SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_PGFSM_STATUS, data, 0xFFFFF); 789 } 790 } 791 792 static int vcn_v2_0_start_dpg_mode(struct amdgpu_device *adev, bool indirect) 793 { 794 volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst->fw_shared_cpu_addr; 795 struct amdgpu_ring *ring = &adev->vcn.inst->ring_dec; 796 uint32_t rb_bufsz, tmp; 797 798 vcn_v2_0_enable_static_power_gating(adev); 799 800 /* enable dynamic power gating mode */ 801 tmp = RREG32_SOC15(UVD, 0, mmUVD_POWER_STATUS); 802 tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK; 803 tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK; 804 WREG32_SOC15(UVD, 0, mmUVD_POWER_STATUS, tmp); 805 806 if (indirect) 807 adev->vcn.inst->dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst->dpg_sram_cpu_addr; 808 809 /* enable clock gating */ 810 vcn_v2_0_clock_gating_dpg_mode(adev, 0, indirect); 811 812 /* enable VCPU clock */ 813 tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT); 814 tmp |= UVD_VCPU_CNTL__CLK_EN_MASK; 815 tmp |= UVD_VCPU_CNTL__MIF_WR_LOW_THRESHOLD_BP_MASK; 816 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 817 UVD, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect); 818 819 /* disable master interupt */ 820 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 821 UVD, 0, mmUVD_MASTINT_EN), 0, 0, indirect); 822 823 /* setup mmUVD_LMI_CTRL */ 824 tmp = (UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK | 825 UVD_LMI_CTRL__REQ_MODE_MASK | 826 UVD_LMI_CTRL__CRC_RESET_MASK | 827 UVD_LMI_CTRL__MASK_MC_URGENT_MASK | 828 UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK | 829 UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK | 830 (8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) | 831 0x00100000L); 832 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 833 UVD, 0, mmUVD_LMI_CTRL), tmp, 0, indirect); 834 835 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 836 UVD, 0, mmUVD_MPC_CNTL), 837 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect); 838 839 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 840 UVD, 0, mmUVD_MPC_SET_MUXA0), 841 ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) | 842 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) | 843 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) | 844 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect); 845 846 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 847 UVD, 0, mmUVD_MPC_SET_MUXB0), 848 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) | 849 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) | 850 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) | 851 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect); 852 853 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 854 UVD, 0, mmUVD_MPC_SET_MUX), 855 ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) | 856 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) | 857 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect); 858 859 vcn_v2_0_mc_resume_dpg_mode(adev, indirect); 860 861 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 862 UVD, 0, mmUVD_REG_XX_MASK), 0x10, 0, indirect); 863 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 864 UVD, 0, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect); 865 866 /* release VCPU reset to boot */ 867 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 868 UVD, 0, mmUVD_SOFT_RESET), 0, 0, indirect); 869 870 /* enable LMI MC and UMC channels */ 871 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 872 UVD, 0, mmUVD_LMI_CTRL2), 873 0x1F << UVD_LMI_CTRL2__RE_OFLD_MIF_WR_REQ_NUM__SHIFT, 0, indirect); 874 875 /* enable master interrupt */ 876 WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET( 877 UVD, 0, mmUVD_MASTINT_EN), 878 UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect); 879 880 if (indirect) 881 psp_update_vcn_sram(adev, 0, adev->vcn.inst->dpg_sram_gpu_addr, 882 (uint32_t)((uintptr_t)adev->vcn.inst->dpg_sram_curr_addr - 883 (uintptr_t)adev->vcn.inst->dpg_sram_cpu_addr)); 884 885 /* force RBC into idle state */ 886 rb_bufsz = order_base_2(ring->ring_size); 887 tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz); 888 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1); 889 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1); 890 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1); 891 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1); 892 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_CNTL, tmp); 893 894 /* Stall DPG before WPTR/RPTR reset */ 895 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_POWER_STATUS), 896 UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK, 897 ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK); 898 fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET; 899 900 /* set the write pointer delay */ 901 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR_CNTL, 0); 902 903 /* set the wb address */ 904 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR_ADDR, 905 (upper_32_bits(ring->gpu_addr) >> 2)); 906 907 /* program the RB_BASE for ring buffer */ 908 WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW, 909 lower_32_bits(ring->gpu_addr)); 910 WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH, 911 upper_32_bits(ring->gpu_addr)); 912 913 /* Initialize the ring buffer's read and write pointers */ 914 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR, 0); 915 916 WREG32_SOC15(UVD, 0, mmUVD_SCRATCH2, 0); 917 918 ring->wptr = RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR); 919 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR, 920 lower_32_bits(ring->wptr)); 921 922 fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET; 923 /* Unstall DPG */ 924 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_POWER_STATUS), 925 0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK); 926 return 0; 927 } 928 929 static int vcn_v2_0_start(struct amdgpu_device *adev) 930 { 931 volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst->fw_shared_cpu_addr; 932 struct amdgpu_ring *ring = &adev->vcn.inst->ring_dec; 933 uint32_t rb_bufsz, tmp; 934 uint32_t lmi_swap_cntl; 935 int i, j, r; 936 937 if (adev->pm.dpm_enabled) 938 amdgpu_dpm_enable_uvd(adev, true); 939 940 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) 941 return vcn_v2_0_start_dpg_mode(adev, adev->vcn.indirect_sram); 942 943 vcn_v2_0_disable_static_power_gating(adev); 944 945 /* set uvd status busy */ 946 tmp = RREG32_SOC15(UVD, 0, mmUVD_STATUS) | UVD_STATUS__UVD_BUSY; 947 WREG32_SOC15(UVD, 0, mmUVD_STATUS, tmp); 948 949 /*SW clock gating */ 950 vcn_v2_0_disable_clock_gating(adev); 951 952 /* enable VCPU clock */ 953 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_VCPU_CNTL), 954 UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK); 955 956 /* disable master interrupt */ 957 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_MASTINT_EN), 0, 958 ~UVD_MASTINT_EN__VCPU_EN_MASK); 959 960 /* setup mmUVD_LMI_CTRL */ 961 tmp = RREG32_SOC15(UVD, 0, mmUVD_LMI_CTRL); 962 WREG32_SOC15(UVD, 0, mmUVD_LMI_CTRL, tmp | 963 UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK | 964 UVD_LMI_CTRL__MASK_MC_URGENT_MASK | 965 UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK | 966 UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK); 967 968 /* setup mmUVD_MPC_CNTL */ 969 tmp = RREG32_SOC15(UVD, 0, mmUVD_MPC_CNTL); 970 tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK; 971 tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT; 972 WREG32_SOC15(VCN, 0, mmUVD_MPC_CNTL, tmp); 973 974 /* setup UVD_MPC_SET_MUXA0 */ 975 WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUXA0, 976 ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) | 977 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) | 978 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) | 979 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT))); 980 981 /* setup UVD_MPC_SET_MUXB0 */ 982 WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUXB0, 983 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) | 984 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) | 985 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) | 986 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT))); 987 988 /* setup mmUVD_MPC_SET_MUX */ 989 WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUX, 990 ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) | 991 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) | 992 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT))); 993 994 vcn_v2_0_mc_resume(adev); 995 996 /* release VCPU reset to boot */ 997 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 0, 998 ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK); 999 1000 /* enable LMI MC and UMC channels */ 1001 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_CTRL2), 0, 1002 ~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK); 1003 1004 tmp = RREG32_SOC15(VCN, 0, mmUVD_SOFT_RESET); 1005 tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK; 1006 tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK; 1007 WREG32_SOC15(VCN, 0, mmUVD_SOFT_RESET, tmp); 1008 1009 /* disable byte swapping */ 1010 lmi_swap_cntl = 0; 1011 #ifdef __BIG_ENDIAN 1012 /* swap (8 in 32) RB and IB */ 1013 lmi_swap_cntl = 0xa; 1014 #endif 1015 WREG32_SOC15(UVD, 0, mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl); 1016 1017 for (i = 0; i < 10; ++i) { 1018 uint32_t status; 1019 1020 for (j = 0; j < 100; ++j) { 1021 status = RREG32_SOC15(UVD, 0, mmUVD_STATUS); 1022 if (status & 2) 1023 break; 1024 mdelay(10); 1025 } 1026 r = 0; 1027 if (status & 2) 1028 break; 1029 1030 DRM_ERROR("VCN decode not responding, trying to reset the VCPU!!!\n"); 1031 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 1032 UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK, 1033 ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK); 1034 mdelay(10); 1035 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 0, 1036 ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK); 1037 mdelay(10); 1038 r = -1; 1039 } 1040 1041 if (r) { 1042 DRM_ERROR("VCN decode not responding, giving up!!!\n"); 1043 return r; 1044 } 1045 1046 /* enable master interrupt */ 1047 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_MASTINT_EN), 1048 UVD_MASTINT_EN__VCPU_EN_MASK, 1049 ~UVD_MASTINT_EN__VCPU_EN_MASK); 1050 1051 /* clear the busy bit of VCN_STATUS */ 1052 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_STATUS), 0, 1053 ~(2 << UVD_STATUS__VCPU_REPORT__SHIFT)); 1054 1055 WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_VMID, 0); 1056 1057 /* force RBC into idle state */ 1058 rb_bufsz = order_base_2(ring->ring_size); 1059 tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz); 1060 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1); 1061 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1); 1062 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1); 1063 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1); 1064 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_CNTL, tmp); 1065 1066 fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET; 1067 /* program the RB_BASE for ring buffer */ 1068 WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW, 1069 lower_32_bits(ring->gpu_addr)); 1070 WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH, 1071 upper_32_bits(ring->gpu_addr)); 1072 1073 /* Initialize the ring buffer's read and write pointers */ 1074 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR, 0); 1075 1076 ring->wptr = RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR); 1077 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR, 1078 lower_32_bits(ring->wptr)); 1079 fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET; 1080 1081 fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET; 1082 ring = &adev->vcn.inst->ring_enc[0]; 1083 WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR, lower_32_bits(ring->wptr)); 1084 WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR, lower_32_bits(ring->wptr)); 1085 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO, ring->gpu_addr); 1086 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr)); 1087 WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE, ring->ring_size / 4); 1088 fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET; 1089 1090 fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET; 1091 ring = &adev->vcn.inst->ring_enc[1]; 1092 WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr)); 1093 WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr)); 1094 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO2, ring->gpu_addr); 1095 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr)); 1096 WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE2, ring->ring_size / 4); 1097 fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET; 1098 1099 return 0; 1100 } 1101 1102 static int vcn_v2_0_stop_dpg_mode(struct amdgpu_device *adev) 1103 { 1104 uint32_t tmp; 1105 1106 /* Wait for power status to be 1 */ 1107 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 1, 1108 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); 1109 1110 /* wait for read ptr to be equal to write ptr */ 1111 tmp = RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR); 1112 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF); 1113 1114 tmp = RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2); 1115 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF); 1116 1117 tmp = RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF; 1118 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF); 1119 1120 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 1, 1121 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); 1122 1123 /* disable dynamic power gating mode */ 1124 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_POWER_STATUS), 0, 1125 ~UVD_POWER_STATUS__UVD_PG_MODE_MASK); 1126 1127 return 0; 1128 } 1129 1130 static int vcn_v2_0_stop(struct amdgpu_device *adev) 1131 { 1132 uint32_t tmp; 1133 int r; 1134 1135 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) { 1136 r = vcn_v2_0_stop_dpg_mode(adev); 1137 if (r) 1138 return r; 1139 goto power_off; 1140 } 1141 1142 /* wait for uvd idle */ 1143 r = SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7); 1144 if (r) 1145 return r; 1146 1147 tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK | 1148 UVD_LMI_STATUS__READ_CLEAN_MASK | 1149 UVD_LMI_STATUS__WRITE_CLEAN_MASK | 1150 UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK; 1151 r = SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_LMI_STATUS, tmp, tmp); 1152 if (r) 1153 return r; 1154 1155 /* stall UMC channel */ 1156 tmp = RREG32_SOC15(VCN, 0, mmUVD_LMI_CTRL2); 1157 tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK; 1158 WREG32_SOC15(VCN, 0, mmUVD_LMI_CTRL2, tmp); 1159 1160 tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK| 1161 UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK; 1162 r = SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_LMI_STATUS, tmp, tmp); 1163 if (r) 1164 return r; 1165 1166 /* disable VCPU clock */ 1167 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_VCPU_CNTL), 0, 1168 ~(UVD_VCPU_CNTL__CLK_EN_MASK)); 1169 1170 /* reset LMI UMC */ 1171 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 1172 UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK, 1173 ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK); 1174 1175 /* reset LMI */ 1176 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 1177 UVD_SOFT_RESET__LMI_SOFT_RESET_MASK, 1178 ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK); 1179 1180 /* reset VCPU */ 1181 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 1182 UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK, 1183 ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK); 1184 1185 /* clear status */ 1186 WREG32_SOC15(VCN, 0, mmUVD_STATUS, 0); 1187 1188 vcn_v2_0_enable_clock_gating(adev); 1189 vcn_v2_0_enable_static_power_gating(adev); 1190 1191 power_off: 1192 if (adev->pm.dpm_enabled) 1193 amdgpu_dpm_enable_uvd(adev, false); 1194 1195 return 0; 1196 } 1197 1198 static int vcn_v2_0_pause_dpg_mode(struct amdgpu_device *adev, 1199 int inst_idx, struct dpg_pause_state *new_state) 1200 { 1201 struct amdgpu_ring *ring; 1202 uint32_t reg_data = 0; 1203 int ret_code; 1204 1205 /* pause/unpause if state is changed */ 1206 if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) { 1207 DRM_DEBUG("dpg pause state changed %d -> %d", 1208 adev->vcn.inst[inst_idx].pause_state.fw_based, new_state->fw_based); 1209 reg_data = RREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE) & 1210 (~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK); 1211 1212 if (new_state->fw_based == VCN_DPG_STATE__PAUSE) { 1213 ret_code = SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 0x1, 1214 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); 1215 1216 if (!ret_code) { 1217 volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst->fw_shared_cpu_addr; 1218 /* pause DPG */ 1219 reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK; 1220 WREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE, reg_data); 1221 1222 /* wait for ACK */ 1223 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_DPG_PAUSE, 1224 UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK, 1225 UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK); 1226 1227 /* Stall DPG before WPTR/RPTR reset */ 1228 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_POWER_STATUS), 1229 UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK, 1230 ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK); 1231 /* Restore */ 1232 fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET; 1233 ring = &adev->vcn.inst->ring_enc[0]; 1234 ring->wptr = 0; 1235 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO, ring->gpu_addr); 1236 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr)); 1237 WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE, ring->ring_size / 4); 1238 WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR, lower_32_bits(ring->wptr)); 1239 WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR, lower_32_bits(ring->wptr)); 1240 fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET; 1241 1242 fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET; 1243 ring = &adev->vcn.inst->ring_enc[1]; 1244 ring->wptr = 0; 1245 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO2, ring->gpu_addr); 1246 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr)); 1247 WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE2, ring->ring_size / 4); 1248 WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr)); 1249 WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr)); 1250 fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET; 1251 1252 fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET; 1253 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR, 1254 RREG32_SOC15(UVD, 0, mmUVD_SCRATCH2) & 0x7FFFFFFF); 1255 fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET; 1256 /* Unstall DPG */ 1257 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_POWER_STATUS), 1258 0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK); 1259 1260 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 1261 UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, 1262 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); 1263 } 1264 } else { 1265 /* unpause dpg, no need to wait */ 1266 reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK; 1267 WREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE, reg_data); 1268 } 1269 adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based; 1270 } 1271 1272 return 0; 1273 } 1274 1275 static bool vcn_v2_0_is_idle(void *handle) 1276 { 1277 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1278 1279 return (RREG32_SOC15(VCN, 0, mmUVD_STATUS) == UVD_STATUS__IDLE); 1280 } 1281 1282 static int vcn_v2_0_wait_for_idle(void *handle) 1283 { 1284 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1285 int ret; 1286 1287 ret = SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_STATUS, UVD_STATUS__IDLE, 1288 UVD_STATUS__IDLE); 1289 1290 return ret; 1291 } 1292 1293 static int vcn_v2_0_set_clockgating_state(void *handle, 1294 enum amd_clockgating_state state) 1295 { 1296 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1297 bool enable = (state == AMD_CG_STATE_GATE); 1298 1299 if (amdgpu_sriov_vf(adev)) 1300 return 0; 1301 1302 if (enable) { 1303 /* wait for STATUS to clear */ 1304 if (!vcn_v2_0_is_idle(handle)) 1305 return -EBUSY; 1306 vcn_v2_0_enable_clock_gating(adev); 1307 } else { 1308 /* disable HW gating and enable Sw gating */ 1309 vcn_v2_0_disable_clock_gating(adev); 1310 } 1311 return 0; 1312 } 1313 1314 /** 1315 * vcn_v2_0_dec_ring_get_rptr - get read pointer 1316 * 1317 * @ring: amdgpu_ring pointer 1318 * 1319 * Returns the current hardware read pointer 1320 */ 1321 static uint64_t vcn_v2_0_dec_ring_get_rptr(struct amdgpu_ring *ring) 1322 { 1323 struct amdgpu_device *adev = ring->adev; 1324 1325 return RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR); 1326 } 1327 1328 /** 1329 * vcn_v2_0_dec_ring_get_wptr - get write pointer 1330 * 1331 * @ring: amdgpu_ring pointer 1332 * 1333 * Returns the current hardware write pointer 1334 */ 1335 static uint64_t vcn_v2_0_dec_ring_get_wptr(struct amdgpu_ring *ring) 1336 { 1337 struct amdgpu_device *adev = ring->adev; 1338 1339 if (ring->use_doorbell) 1340 return adev->wb.wb[ring->wptr_offs]; 1341 else 1342 return RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR); 1343 } 1344 1345 /** 1346 * vcn_v2_0_dec_ring_set_wptr - set write pointer 1347 * 1348 * @ring: amdgpu_ring pointer 1349 * 1350 * Commits the write pointer to the hardware 1351 */ 1352 static void vcn_v2_0_dec_ring_set_wptr(struct amdgpu_ring *ring) 1353 { 1354 struct amdgpu_device *adev = ring->adev; 1355 1356 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) 1357 WREG32_SOC15(UVD, 0, mmUVD_SCRATCH2, 1358 lower_32_bits(ring->wptr) | 0x80000000); 1359 1360 if (ring->use_doorbell) { 1361 adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr); 1362 WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr)); 1363 } else { 1364 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr)); 1365 } 1366 } 1367 1368 /** 1369 * vcn_v2_0_dec_ring_insert_start - insert a start command 1370 * 1371 * @ring: amdgpu_ring pointer 1372 * 1373 * Write a start command to the ring. 1374 */ 1375 void vcn_v2_0_dec_ring_insert_start(struct amdgpu_ring *ring) 1376 { 1377 struct amdgpu_device *adev = ring->adev; 1378 1379 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data0, 0)); 1380 amdgpu_ring_write(ring, 0); 1381 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0)); 1382 amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_PACKET_START << 1)); 1383 } 1384 1385 /** 1386 * vcn_v2_0_dec_ring_insert_end - insert a end command 1387 * 1388 * @ring: amdgpu_ring pointer 1389 * 1390 * Write a end command to the ring. 1391 */ 1392 void vcn_v2_0_dec_ring_insert_end(struct amdgpu_ring *ring) 1393 { 1394 struct amdgpu_device *adev = ring->adev; 1395 1396 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0)); 1397 amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_PACKET_END << 1)); 1398 } 1399 1400 /** 1401 * vcn_v2_0_dec_ring_insert_nop - insert a nop command 1402 * 1403 * @ring: amdgpu_ring pointer 1404 * @count: the number of NOP packets to insert 1405 * 1406 * Write a nop command to the ring. 1407 */ 1408 void vcn_v2_0_dec_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) 1409 { 1410 struct amdgpu_device *adev = ring->adev; 1411 int i; 1412 1413 WARN_ON(ring->wptr % 2 || count % 2); 1414 1415 for (i = 0; i < count / 2; i++) { 1416 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.nop, 0)); 1417 amdgpu_ring_write(ring, 0); 1418 } 1419 } 1420 1421 /** 1422 * vcn_v2_0_dec_ring_emit_fence - emit an fence & trap command 1423 * 1424 * @ring: amdgpu_ring pointer 1425 * @addr: address 1426 * @seq: sequence number 1427 * @flags: fence related flags 1428 * 1429 * Write a fence and a trap command to the ring. 1430 */ 1431 void vcn_v2_0_dec_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq, 1432 unsigned flags) 1433 { 1434 struct amdgpu_device *adev = ring->adev; 1435 1436 WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT); 1437 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.context_id, 0)); 1438 amdgpu_ring_write(ring, seq); 1439 1440 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data0, 0)); 1441 amdgpu_ring_write(ring, addr & 0xffffffff); 1442 1443 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data1, 0)); 1444 amdgpu_ring_write(ring, upper_32_bits(addr) & 0xff); 1445 1446 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0)); 1447 amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_FENCE << 1)); 1448 1449 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data0, 0)); 1450 amdgpu_ring_write(ring, 0); 1451 1452 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data1, 0)); 1453 amdgpu_ring_write(ring, 0); 1454 1455 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0)); 1456 1457 amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_TRAP << 1)); 1458 } 1459 1460 /** 1461 * vcn_v2_0_dec_ring_emit_ib - execute indirect buffer 1462 * 1463 * @ring: amdgpu_ring pointer 1464 * @job: job to retrieve vmid from 1465 * @ib: indirect buffer to execute 1466 * @flags: unused 1467 * 1468 * Write ring commands to execute the indirect buffer 1469 */ 1470 void vcn_v2_0_dec_ring_emit_ib(struct amdgpu_ring *ring, 1471 struct amdgpu_job *job, 1472 struct amdgpu_ib *ib, 1473 uint32_t flags) 1474 { 1475 struct amdgpu_device *adev = ring->adev; 1476 unsigned vmid = AMDGPU_JOB_GET_VMID(job); 1477 1478 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.ib_vmid, 0)); 1479 amdgpu_ring_write(ring, vmid); 1480 1481 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.ib_bar_low, 0)); 1482 amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr)); 1483 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.ib_bar_high, 0)); 1484 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr)); 1485 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.ib_size, 0)); 1486 amdgpu_ring_write(ring, ib->length_dw); 1487 } 1488 1489 void vcn_v2_0_dec_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg, 1490 uint32_t val, uint32_t mask) 1491 { 1492 struct amdgpu_device *adev = ring->adev; 1493 1494 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data0, 0)); 1495 amdgpu_ring_write(ring, reg << 2); 1496 1497 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data1, 0)); 1498 amdgpu_ring_write(ring, val); 1499 1500 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.gp_scratch8, 0)); 1501 amdgpu_ring_write(ring, mask); 1502 1503 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0)); 1504 1505 amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_REG_READ_COND_WAIT << 1)); 1506 } 1507 1508 void vcn_v2_0_dec_ring_emit_vm_flush(struct amdgpu_ring *ring, 1509 unsigned vmid, uint64_t pd_addr) 1510 { 1511 struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub]; 1512 uint32_t data0, data1, mask; 1513 1514 pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr); 1515 1516 /* wait for register write */ 1517 data0 = hub->ctx0_ptb_addr_lo32 + vmid * hub->ctx_addr_distance; 1518 data1 = lower_32_bits(pd_addr); 1519 mask = 0xffffffff; 1520 vcn_v2_0_dec_ring_emit_reg_wait(ring, data0, data1, mask); 1521 } 1522 1523 void vcn_v2_0_dec_ring_emit_wreg(struct amdgpu_ring *ring, 1524 uint32_t reg, uint32_t val) 1525 { 1526 struct amdgpu_device *adev = ring->adev; 1527 1528 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data0, 0)); 1529 amdgpu_ring_write(ring, reg << 2); 1530 1531 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data1, 0)); 1532 amdgpu_ring_write(ring, val); 1533 1534 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0)); 1535 1536 amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_WRITE_REG << 1)); 1537 } 1538 1539 /** 1540 * vcn_v2_0_enc_ring_get_rptr - get enc read pointer 1541 * 1542 * @ring: amdgpu_ring pointer 1543 * 1544 * Returns the current hardware enc read pointer 1545 */ 1546 static uint64_t vcn_v2_0_enc_ring_get_rptr(struct amdgpu_ring *ring) 1547 { 1548 struct amdgpu_device *adev = ring->adev; 1549 1550 if (ring == &adev->vcn.inst->ring_enc[0]) 1551 return RREG32_SOC15(UVD, 0, mmUVD_RB_RPTR); 1552 else 1553 return RREG32_SOC15(UVD, 0, mmUVD_RB_RPTR2); 1554 } 1555 1556 /** 1557 * vcn_v2_0_enc_ring_get_wptr - get enc write pointer 1558 * 1559 * @ring: amdgpu_ring pointer 1560 * 1561 * Returns the current hardware enc write pointer 1562 */ 1563 static uint64_t vcn_v2_0_enc_ring_get_wptr(struct amdgpu_ring *ring) 1564 { 1565 struct amdgpu_device *adev = ring->adev; 1566 1567 if (ring == &adev->vcn.inst->ring_enc[0]) { 1568 if (ring->use_doorbell) 1569 return adev->wb.wb[ring->wptr_offs]; 1570 else 1571 return RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR); 1572 } else { 1573 if (ring->use_doorbell) 1574 return adev->wb.wb[ring->wptr_offs]; 1575 else 1576 return RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2); 1577 } 1578 } 1579 1580 /** 1581 * vcn_v2_0_enc_ring_set_wptr - set enc write pointer 1582 * 1583 * @ring: amdgpu_ring pointer 1584 * 1585 * Commits the enc write pointer to the hardware 1586 */ 1587 static void vcn_v2_0_enc_ring_set_wptr(struct amdgpu_ring *ring) 1588 { 1589 struct amdgpu_device *adev = ring->adev; 1590 1591 if (ring == &adev->vcn.inst->ring_enc[0]) { 1592 if (ring->use_doorbell) { 1593 adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr); 1594 WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr)); 1595 } else { 1596 WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR, lower_32_bits(ring->wptr)); 1597 } 1598 } else { 1599 if (ring->use_doorbell) { 1600 adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr); 1601 WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr)); 1602 } else { 1603 WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr)); 1604 } 1605 } 1606 } 1607 1608 /** 1609 * vcn_v2_0_enc_ring_emit_fence - emit an enc fence & trap command 1610 * 1611 * @ring: amdgpu_ring pointer 1612 * @addr: address 1613 * @seq: sequence number 1614 * @flags: fence related flags 1615 * 1616 * Write enc a fence and a trap command to the ring. 1617 */ 1618 void vcn_v2_0_enc_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, 1619 u64 seq, unsigned flags) 1620 { 1621 WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT); 1622 1623 amdgpu_ring_write(ring, VCN_ENC_CMD_FENCE); 1624 amdgpu_ring_write(ring, addr); 1625 amdgpu_ring_write(ring, upper_32_bits(addr)); 1626 amdgpu_ring_write(ring, seq); 1627 amdgpu_ring_write(ring, VCN_ENC_CMD_TRAP); 1628 } 1629 1630 void vcn_v2_0_enc_ring_insert_end(struct amdgpu_ring *ring) 1631 { 1632 amdgpu_ring_write(ring, VCN_ENC_CMD_END); 1633 } 1634 1635 /** 1636 * vcn_v2_0_enc_ring_emit_ib - enc execute indirect buffer 1637 * 1638 * @ring: amdgpu_ring pointer 1639 * @job: job to retrive vmid from 1640 * @ib: indirect buffer to execute 1641 * @flags: unused 1642 * 1643 * Write enc ring commands to execute the indirect buffer 1644 */ 1645 void vcn_v2_0_enc_ring_emit_ib(struct amdgpu_ring *ring, 1646 struct amdgpu_job *job, 1647 struct amdgpu_ib *ib, 1648 uint32_t flags) 1649 { 1650 unsigned vmid = AMDGPU_JOB_GET_VMID(job); 1651 1652 amdgpu_ring_write(ring, VCN_ENC_CMD_IB); 1653 amdgpu_ring_write(ring, vmid); 1654 amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr)); 1655 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr)); 1656 amdgpu_ring_write(ring, ib->length_dw); 1657 } 1658 1659 void vcn_v2_0_enc_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg, 1660 uint32_t val, uint32_t mask) 1661 { 1662 amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WAIT); 1663 amdgpu_ring_write(ring, reg << 2); 1664 amdgpu_ring_write(ring, mask); 1665 amdgpu_ring_write(ring, val); 1666 } 1667 1668 void vcn_v2_0_enc_ring_emit_vm_flush(struct amdgpu_ring *ring, 1669 unsigned int vmid, uint64_t pd_addr) 1670 { 1671 struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub]; 1672 1673 pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr); 1674 1675 /* wait for reg writes */ 1676 vcn_v2_0_enc_ring_emit_reg_wait(ring, hub->ctx0_ptb_addr_lo32 + 1677 vmid * hub->ctx_addr_distance, 1678 lower_32_bits(pd_addr), 0xffffffff); 1679 } 1680 1681 void vcn_v2_0_enc_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg, uint32_t val) 1682 { 1683 amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WRITE); 1684 amdgpu_ring_write(ring, reg << 2); 1685 amdgpu_ring_write(ring, val); 1686 } 1687 1688 static int vcn_v2_0_set_interrupt_state(struct amdgpu_device *adev, 1689 struct amdgpu_irq_src *source, 1690 unsigned type, 1691 enum amdgpu_interrupt_state state) 1692 { 1693 return 0; 1694 } 1695 1696 static int vcn_v2_0_process_interrupt(struct amdgpu_device *adev, 1697 struct amdgpu_irq_src *source, 1698 struct amdgpu_iv_entry *entry) 1699 { 1700 DRM_DEBUG("IH: VCN TRAP\n"); 1701 1702 switch (entry->src_id) { 1703 case VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT: 1704 amdgpu_fence_process(&adev->vcn.inst->ring_dec); 1705 break; 1706 case VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE: 1707 amdgpu_fence_process(&adev->vcn.inst->ring_enc[0]); 1708 break; 1709 case VCN_2_0__SRCID__UVD_ENC_LOW_LATENCY: 1710 amdgpu_fence_process(&adev->vcn.inst->ring_enc[1]); 1711 break; 1712 default: 1713 DRM_ERROR("Unhandled interrupt: %d %d\n", 1714 entry->src_id, entry->src_data[0]); 1715 break; 1716 } 1717 1718 return 0; 1719 } 1720 1721 int vcn_v2_0_dec_ring_test_ring(struct amdgpu_ring *ring) 1722 { 1723 struct amdgpu_device *adev = ring->adev; 1724 uint32_t tmp = 0; 1725 unsigned i; 1726 int r; 1727 1728 if (amdgpu_sriov_vf(adev)) 1729 return 0; 1730 1731 WREG32(adev->vcn.inst[ring->me].external.scratch9, 0xCAFEDEAD); 1732 r = amdgpu_ring_alloc(ring, 4); 1733 if (r) 1734 return r; 1735 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0)); 1736 amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_PACKET_START << 1)); 1737 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.scratch9, 0)); 1738 amdgpu_ring_write(ring, 0xDEADBEEF); 1739 amdgpu_ring_commit(ring); 1740 for (i = 0; i < adev->usec_timeout; i++) { 1741 tmp = RREG32(adev->vcn.inst[ring->me].external.scratch9); 1742 if (tmp == 0xDEADBEEF) 1743 break; 1744 udelay(1); 1745 } 1746 1747 if (i >= adev->usec_timeout) 1748 r = -ETIMEDOUT; 1749 1750 return r; 1751 } 1752 1753 1754 static int vcn_v2_0_set_powergating_state(void *handle, 1755 enum amd_powergating_state state) 1756 { 1757 /* This doesn't actually powergate the VCN block. 1758 * That's done in the dpm code via the SMC. This 1759 * just re-inits the block as necessary. The actual 1760 * gating still happens in the dpm code. We should 1761 * revisit this when there is a cleaner line between 1762 * the smc and the hw blocks 1763 */ 1764 int ret; 1765 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1766 1767 if (amdgpu_sriov_vf(adev)) { 1768 adev->vcn.cur_state = AMD_PG_STATE_UNGATE; 1769 return 0; 1770 } 1771 1772 if (state == adev->vcn.cur_state) 1773 return 0; 1774 1775 if (state == AMD_PG_STATE_GATE) 1776 ret = vcn_v2_0_stop(adev); 1777 else 1778 ret = vcn_v2_0_start(adev); 1779 1780 if (!ret) 1781 adev->vcn.cur_state = state; 1782 return ret; 1783 } 1784 1785 static int vcn_v2_0_start_mmsch(struct amdgpu_device *adev, 1786 struct amdgpu_mm_table *table) 1787 { 1788 uint32_t data = 0, loop; 1789 uint64_t addr = table->gpu_addr; 1790 struct mmsch_v2_0_init_header *header; 1791 uint32_t size; 1792 int i; 1793 1794 header = (struct mmsch_v2_0_init_header *)table->cpu_addr; 1795 size = header->header_size + header->vcn_table_size; 1796 1797 /* 1, write to vce_mmsch_vf_ctx_addr_lo/hi register with GPU mc addr 1798 * of memory descriptor location 1799 */ 1800 WREG32_SOC15(UVD, 0, mmMMSCH_VF_CTX_ADDR_LO, lower_32_bits(addr)); 1801 WREG32_SOC15(UVD, 0, mmMMSCH_VF_CTX_ADDR_HI, upper_32_bits(addr)); 1802 1803 /* 2, update vmid of descriptor */ 1804 data = RREG32_SOC15(UVD, 0, mmMMSCH_VF_VMID); 1805 data &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK; 1806 /* use domain0 for MM scheduler */ 1807 data |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT); 1808 WREG32_SOC15(UVD, 0, mmMMSCH_VF_VMID, data); 1809 1810 /* 3, notify mmsch about the size of this descriptor */ 1811 WREG32_SOC15(UVD, 0, mmMMSCH_VF_CTX_SIZE, size); 1812 1813 /* 4, set resp to zero */ 1814 WREG32_SOC15(UVD, 0, mmMMSCH_VF_MAILBOX_RESP, 0); 1815 1816 adev->vcn.inst->ring_dec.wptr = 0; 1817 adev->vcn.inst->ring_dec.wptr_old = 0; 1818 vcn_v2_0_dec_ring_set_wptr(&adev->vcn.inst->ring_dec); 1819 1820 for (i = 0; i < adev->vcn.num_enc_rings; ++i) { 1821 adev->vcn.inst->ring_enc[i].wptr = 0; 1822 adev->vcn.inst->ring_enc[i].wptr_old = 0; 1823 vcn_v2_0_enc_ring_set_wptr(&adev->vcn.inst->ring_enc[i]); 1824 } 1825 1826 /* 5, kick off the initialization and wait until 1827 * VCE_MMSCH_VF_MAILBOX_RESP becomes non-zero 1828 */ 1829 WREG32_SOC15(UVD, 0, mmMMSCH_VF_MAILBOX_HOST, 0x10000001); 1830 1831 data = RREG32_SOC15(UVD, 0, mmMMSCH_VF_MAILBOX_RESP); 1832 loop = 1000; 1833 while ((data & 0x10000002) != 0x10000002) { 1834 udelay(10); 1835 data = RREG32_SOC15(UVD, 0, mmMMSCH_VF_MAILBOX_RESP); 1836 loop--; 1837 if (!loop) 1838 break; 1839 } 1840 1841 if (!loop) { 1842 DRM_ERROR("failed to init MMSCH, " \ 1843 "mmMMSCH_VF_MAILBOX_RESP = 0x%08x\n", data); 1844 return -EBUSY; 1845 } 1846 1847 return 0; 1848 } 1849 1850 static int vcn_v2_0_start_sriov(struct amdgpu_device *adev) 1851 { 1852 int r; 1853 uint32_t tmp; 1854 struct amdgpu_ring *ring; 1855 uint32_t offset, size; 1856 uint32_t table_size = 0; 1857 struct mmsch_v2_0_cmd_direct_write direct_wt = { {0} }; 1858 struct mmsch_v2_0_cmd_direct_read_modify_write direct_rd_mod_wt = { {0} }; 1859 struct mmsch_v2_0_cmd_end end = { {0} }; 1860 struct mmsch_v2_0_init_header *header; 1861 uint32_t *init_table = adev->virt.mm_table.cpu_addr; 1862 uint8_t i = 0; 1863 1864 header = (struct mmsch_v2_0_init_header *)init_table; 1865 direct_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_WRITE; 1866 direct_rd_mod_wt.cmd_header.command_type = 1867 MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE; 1868 end.cmd_header.command_type = MMSCH_COMMAND__END; 1869 1870 if (header->vcn_table_offset == 0 && header->vcn_table_size == 0) { 1871 header->version = MMSCH_VERSION; 1872 header->header_size = sizeof(struct mmsch_v2_0_init_header) >> 2; 1873 1874 header->vcn_table_offset = header->header_size; 1875 1876 init_table += header->vcn_table_offset; 1877 1878 size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4); 1879 1880 MMSCH_V2_0_INSERT_DIRECT_RD_MOD_WT( 1881 SOC15_REG_OFFSET(UVD, i, mmUVD_STATUS), 1882 0xFFFFFFFF, 0x00000004); 1883 1884 /* mc resume*/ 1885 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) { 1886 tmp = AMDGPU_UCODE_ID_VCN; 1887 MMSCH_V2_0_INSERT_DIRECT_WT( 1888 SOC15_REG_OFFSET(UVD, i, 1889 mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 1890 adev->firmware.ucode[tmp].tmr_mc_addr_lo); 1891 MMSCH_V2_0_INSERT_DIRECT_WT( 1892 SOC15_REG_OFFSET(UVD, i, 1893 mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 1894 adev->firmware.ucode[tmp].tmr_mc_addr_hi); 1895 offset = 0; 1896 } else { 1897 MMSCH_V2_0_INSERT_DIRECT_WT( 1898 SOC15_REG_OFFSET(UVD, i, 1899 mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 1900 lower_32_bits(adev->vcn.inst->gpu_addr)); 1901 MMSCH_V2_0_INSERT_DIRECT_WT( 1902 SOC15_REG_OFFSET(UVD, i, 1903 mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 1904 upper_32_bits(adev->vcn.inst->gpu_addr)); 1905 offset = size; 1906 } 1907 1908 MMSCH_V2_0_INSERT_DIRECT_WT( 1909 SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_OFFSET0), 1910 0); 1911 MMSCH_V2_0_INSERT_DIRECT_WT( 1912 SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_SIZE0), 1913 size); 1914 1915 MMSCH_V2_0_INSERT_DIRECT_WT( 1916 SOC15_REG_OFFSET(UVD, i, 1917 mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 1918 lower_32_bits(adev->vcn.inst->gpu_addr + offset)); 1919 MMSCH_V2_0_INSERT_DIRECT_WT( 1920 SOC15_REG_OFFSET(UVD, i, 1921 mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 1922 upper_32_bits(adev->vcn.inst->gpu_addr + offset)); 1923 MMSCH_V2_0_INSERT_DIRECT_WT( 1924 SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_OFFSET1), 1925 0); 1926 MMSCH_V2_0_INSERT_DIRECT_WT( 1927 SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_SIZE1), 1928 AMDGPU_VCN_STACK_SIZE); 1929 1930 MMSCH_V2_0_INSERT_DIRECT_WT( 1931 SOC15_REG_OFFSET(UVD, i, 1932 mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW), 1933 lower_32_bits(adev->vcn.inst->gpu_addr + offset + 1934 AMDGPU_VCN_STACK_SIZE)); 1935 MMSCH_V2_0_INSERT_DIRECT_WT( 1936 SOC15_REG_OFFSET(UVD, i, 1937 mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH), 1938 upper_32_bits(adev->vcn.inst->gpu_addr + offset + 1939 AMDGPU_VCN_STACK_SIZE)); 1940 MMSCH_V2_0_INSERT_DIRECT_WT( 1941 SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_OFFSET2), 1942 0); 1943 MMSCH_V2_0_INSERT_DIRECT_WT( 1944 SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_SIZE2), 1945 AMDGPU_VCN_CONTEXT_SIZE); 1946 1947 for (r = 0; r < adev->vcn.num_enc_rings; ++r) { 1948 ring = &adev->vcn.inst->ring_enc[r]; 1949 ring->wptr = 0; 1950 MMSCH_V2_0_INSERT_DIRECT_WT( 1951 SOC15_REG_OFFSET(UVD, i, mmUVD_RB_BASE_LO), 1952 lower_32_bits(ring->gpu_addr)); 1953 MMSCH_V2_0_INSERT_DIRECT_WT( 1954 SOC15_REG_OFFSET(UVD, i, mmUVD_RB_BASE_HI), 1955 upper_32_bits(ring->gpu_addr)); 1956 MMSCH_V2_0_INSERT_DIRECT_WT( 1957 SOC15_REG_OFFSET(UVD, i, mmUVD_RB_SIZE), 1958 ring->ring_size / 4); 1959 } 1960 1961 ring = &adev->vcn.inst->ring_dec; 1962 ring->wptr = 0; 1963 MMSCH_V2_0_INSERT_DIRECT_WT( 1964 SOC15_REG_OFFSET(UVD, i, 1965 mmUVD_LMI_RBC_RB_64BIT_BAR_LOW), 1966 lower_32_bits(ring->gpu_addr)); 1967 MMSCH_V2_0_INSERT_DIRECT_WT( 1968 SOC15_REG_OFFSET(UVD, i, 1969 mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH), 1970 upper_32_bits(ring->gpu_addr)); 1971 /* force RBC into idle state */ 1972 tmp = order_base_2(ring->ring_size); 1973 tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, tmp); 1974 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1); 1975 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1); 1976 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1); 1977 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1); 1978 MMSCH_V2_0_INSERT_DIRECT_WT( 1979 SOC15_REG_OFFSET(UVD, i, mmUVD_RBC_RB_CNTL), tmp); 1980 1981 /* add end packet */ 1982 tmp = sizeof(struct mmsch_v2_0_cmd_end); 1983 memcpy((void *)init_table, &end, tmp); 1984 table_size += (tmp / 4); 1985 header->vcn_table_size = table_size; 1986 1987 } 1988 return vcn_v2_0_start_mmsch(adev, &adev->virt.mm_table); 1989 } 1990 1991 static const struct amd_ip_funcs vcn_v2_0_ip_funcs = { 1992 .name = "vcn_v2_0", 1993 .early_init = vcn_v2_0_early_init, 1994 .late_init = NULL, 1995 .sw_init = vcn_v2_0_sw_init, 1996 .sw_fini = vcn_v2_0_sw_fini, 1997 .hw_init = vcn_v2_0_hw_init, 1998 .hw_fini = vcn_v2_0_hw_fini, 1999 .suspend = vcn_v2_0_suspend, 2000 .resume = vcn_v2_0_resume, 2001 .is_idle = vcn_v2_0_is_idle, 2002 .wait_for_idle = vcn_v2_0_wait_for_idle, 2003 .check_soft_reset = NULL, 2004 .pre_soft_reset = NULL, 2005 .soft_reset = NULL, 2006 .post_soft_reset = NULL, 2007 .set_clockgating_state = vcn_v2_0_set_clockgating_state, 2008 .set_powergating_state = vcn_v2_0_set_powergating_state, 2009 }; 2010 2011 static const struct amdgpu_ring_funcs vcn_v2_0_dec_ring_vm_funcs = { 2012 .type = AMDGPU_RING_TYPE_VCN_DEC, 2013 .align_mask = 0xf, 2014 .vmhub = AMDGPU_MMHUB_0, 2015 .get_rptr = vcn_v2_0_dec_ring_get_rptr, 2016 .get_wptr = vcn_v2_0_dec_ring_get_wptr, 2017 .set_wptr = vcn_v2_0_dec_ring_set_wptr, 2018 .emit_frame_size = 2019 SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 + 2020 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 + 2021 8 + /* vcn_v2_0_dec_ring_emit_vm_flush */ 2022 14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */ 2023 6, 2024 .emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */ 2025 .emit_ib = vcn_v2_0_dec_ring_emit_ib, 2026 .emit_fence = vcn_v2_0_dec_ring_emit_fence, 2027 .emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush, 2028 .test_ring = vcn_v2_0_dec_ring_test_ring, 2029 .test_ib = amdgpu_vcn_dec_ring_test_ib, 2030 .insert_nop = vcn_v2_0_dec_ring_insert_nop, 2031 .insert_start = vcn_v2_0_dec_ring_insert_start, 2032 .insert_end = vcn_v2_0_dec_ring_insert_end, 2033 .pad_ib = amdgpu_ring_generic_pad_ib, 2034 .begin_use = amdgpu_vcn_ring_begin_use, 2035 .end_use = amdgpu_vcn_ring_end_use, 2036 .emit_wreg = vcn_v2_0_dec_ring_emit_wreg, 2037 .emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait, 2038 .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper, 2039 }; 2040 2041 static const struct amdgpu_ring_funcs vcn_v2_0_enc_ring_vm_funcs = { 2042 .type = AMDGPU_RING_TYPE_VCN_ENC, 2043 .align_mask = 0x3f, 2044 .nop = VCN_ENC_CMD_NO_OP, 2045 .vmhub = AMDGPU_MMHUB_0, 2046 .get_rptr = vcn_v2_0_enc_ring_get_rptr, 2047 .get_wptr = vcn_v2_0_enc_ring_get_wptr, 2048 .set_wptr = vcn_v2_0_enc_ring_set_wptr, 2049 .emit_frame_size = 2050 SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 + 2051 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 + 2052 4 + /* vcn_v2_0_enc_ring_emit_vm_flush */ 2053 5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */ 2054 1, /* vcn_v2_0_enc_ring_insert_end */ 2055 .emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */ 2056 .emit_ib = vcn_v2_0_enc_ring_emit_ib, 2057 .emit_fence = vcn_v2_0_enc_ring_emit_fence, 2058 .emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush, 2059 .test_ring = amdgpu_vcn_enc_ring_test_ring, 2060 .test_ib = amdgpu_vcn_enc_ring_test_ib, 2061 .insert_nop = amdgpu_ring_insert_nop, 2062 .insert_end = vcn_v2_0_enc_ring_insert_end, 2063 .pad_ib = amdgpu_ring_generic_pad_ib, 2064 .begin_use = amdgpu_vcn_ring_begin_use, 2065 .end_use = amdgpu_vcn_ring_end_use, 2066 .emit_wreg = vcn_v2_0_enc_ring_emit_wreg, 2067 .emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait, 2068 .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper, 2069 }; 2070 2071 static void vcn_v2_0_set_dec_ring_funcs(struct amdgpu_device *adev) 2072 { 2073 adev->vcn.inst->ring_dec.funcs = &vcn_v2_0_dec_ring_vm_funcs; 2074 DRM_INFO("VCN decode is enabled in VM mode\n"); 2075 } 2076 2077 static void vcn_v2_0_set_enc_ring_funcs(struct amdgpu_device *adev) 2078 { 2079 int i; 2080 2081 for (i = 0; i < adev->vcn.num_enc_rings; ++i) 2082 adev->vcn.inst->ring_enc[i].funcs = &vcn_v2_0_enc_ring_vm_funcs; 2083 2084 DRM_INFO("VCN encode is enabled in VM mode\n"); 2085 } 2086 2087 static const struct amdgpu_irq_src_funcs vcn_v2_0_irq_funcs = { 2088 .set = vcn_v2_0_set_interrupt_state, 2089 .process = vcn_v2_0_process_interrupt, 2090 }; 2091 2092 static void vcn_v2_0_set_irq_funcs(struct amdgpu_device *adev) 2093 { 2094 adev->vcn.inst->irq.num_types = adev->vcn.num_enc_rings + 1; 2095 adev->vcn.inst->irq.funcs = &vcn_v2_0_irq_funcs; 2096 } 2097 2098 const struct amdgpu_ip_block_version vcn_v2_0_ip_block = 2099 { 2100 .type = AMD_IP_BLOCK_TYPE_VCN, 2101 .major = 2, 2102 .minor = 0, 2103 .rev = 0, 2104 .funcs = &vcn_v2_0_ip_funcs, 2105 }; 2106