1 /* 2 * Copyright 2016 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 26 #include "amdgpu.h" 27 #include "amdgpu_vcn.h" 28 #include "amdgpu_pm.h" 29 #include "soc15.h" 30 #include "soc15d.h" 31 #include "soc15_common.h" 32 33 #include "vcn/vcn_1_0_offset.h" 34 #include "vcn/vcn_1_0_sh_mask.h" 35 #include "mmhub/mmhub_9_1_offset.h" 36 #include "mmhub/mmhub_9_1_sh_mask.h" 37 38 #include "ivsrcid/vcn/irqsrcs_vcn_1_0.h" 39 #include "jpeg_v1_0.h" 40 #include "vcn_v1_0.h" 41 42 #define mmUVD_RBC_XX_IB_REG_CHECK_1_0 0x05ab 43 #define mmUVD_RBC_XX_IB_REG_CHECK_1_0_BASE_IDX 1 44 #define mmUVD_REG_XX_MASK_1_0 0x05ac 45 #define mmUVD_REG_XX_MASK_1_0_BASE_IDX 1 46 47 static int vcn_v1_0_stop(struct amdgpu_device *adev); 48 static void vcn_v1_0_set_dec_ring_funcs(struct amdgpu_device *adev); 49 static void vcn_v1_0_set_enc_ring_funcs(struct amdgpu_device *adev); 50 static void vcn_v1_0_set_irq_funcs(struct amdgpu_device *adev); 51 static int vcn_v1_0_set_powergating_state(void *handle, enum amd_powergating_state state); 52 static int vcn_v1_0_pause_dpg_mode(struct amdgpu_device *adev, 53 int inst_idx, struct dpg_pause_state *new_state); 54 55 static void vcn_v1_0_idle_work_handler(struct work_struct *work); 56 static void vcn_v1_0_ring_begin_use(struct amdgpu_ring *ring); 57 58 /** 59 * vcn_v1_0_early_init - set function pointers 60 * 61 * @handle: amdgpu_device pointer 62 * 63 * Set ring and irq function pointers 64 */ 65 static int vcn_v1_0_early_init(void *handle) 66 { 67 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 68 69 adev->vcn.num_vcn_inst = 1; 70 adev->vcn.num_enc_rings = 2; 71 72 vcn_v1_0_set_dec_ring_funcs(adev); 73 vcn_v1_0_set_enc_ring_funcs(adev); 74 vcn_v1_0_set_irq_funcs(adev); 75 76 jpeg_v1_0_early_init(handle); 77 78 return 0; 79 } 80 81 /** 82 * vcn_v1_0_sw_init - sw init for VCN block 83 * 84 * @handle: amdgpu_device pointer 85 * 86 * Load firmware and sw initialization 87 */ 88 static int vcn_v1_0_sw_init(void *handle) 89 { 90 struct amdgpu_ring *ring; 91 int i, r; 92 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 93 94 /* VCN DEC TRAP */ 95 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_VCN, 96 VCN_1_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT, &adev->vcn.inst->irq); 97 if (r) 98 return r; 99 100 /* VCN ENC TRAP */ 101 for (i = 0; i < adev->vcn.num_enc_rings; ++i) { 102 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_VCN, i + VCN_1_0__SRCID__UVD_ENC_GENERAL_PURPOSE, 103 &adev->vcn.inst->irq); 104 if (r) 105 return r; 106 } 107 108 r = amdgpu_vcn_sw_init(adev); 109 if (r) 110 return r; 111 112 /* Override the work func */ 113 adev->vcn.idle_work.work.func = vcn_v1_0_idle_work_handler; 114 115 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) { 116 const struct common_firmware_header *hdr; 117 hdr = (const struct common_firmware_header *)adev->vcn.fw->data; 118 adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].ucode_id = AMDGPU_UCODE_ID_VCN; 119 adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].fw = adev->vcn.fw; 120 adev->firmware.fw_size += 121 ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE); 122 dev_info(adev->dev, "Will use PSP to load VCN firmware\n"); 123 } 124 125 r = amdgpu_vcn_resume(adev); 126 if (r) 127 return r; 128 129 ring = &adev->vcn.inst->ring_dec; 130 sprintf(ring->name, "vcn_dec"); 131 r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst->irq, 0, 132 AMDGPU_RING_PRIO_DEFAULT, NULL); 133 if (r) 134 return r; 135 136 adev->vcn.internal.scratch9 = adev->vcn.inst->external.scratch9 = 137 SOC15_REG_OFFSET(UVD, 0, mmUVD_SCRATCH9); 138 adev->vcn.internal.data0 = adev->vcn.inst->external.data0 = 139 SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0); 140 adev->vcn.internal.data1 = adev->vcn.inst->external.data1 = 141 SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1); 142 adev->vcn.internal.cmd = adev->vcn.inst->external.cmd = 143 SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD); 144 adev->vcn.internal.nop = adev->vcn.inst->external.nop = 145 SOC15_REG_OFFSET(UVD, 0, mmUVD_NO_OP); 146 147 for (i = 0; i < adev->vcn.num_enc_rings; ++i) { 148 ring = &adev->vcn.inst->ring_enc[i]; 149 sprintf(ring->name, "vcn_enc%d", i); 150 r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst->irq, 0, 151 AMDGPU_RING_PRIO_DEFAULT, NULL); 152 if (r) 153 return r; 154 } 155 156 adev->vcn.pause_dpg_mode = vcn_v1_0_pause_dpg_mode; 157 158 r = jpeg_v1_0_sw_init(handle); 159 160 return r; 161 } 162 163 /** 164 * vcn_v1_0_sw_fini - sw fini for VCN block 165 * 166 * @handle: amdgpu_device pointer 167 * 168 * VCN suspend and free up sw allocation 169 */ 170 static int vcn_v1_0_sw_fini(void *handle) 171 { 172 int r; 173 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 174 175 r = amdgpu_vcn_suspend(adev); 176 if (r) 177 return r; 178 179 jpeg_v1_0_sw_fini(handle); 180 181 r = amdgpu_vcn_sw_fini(adev); 182 183 return r; 184 } 185 186 /** 187 * vcn_v1_0_hw_init - start and test VCN block 188 * 189 * @handle: amdgpu_device pointer 190 * 191 * Initialize the hardware, boot up the VCPU and do some testing 192 */ 193 static int vcn_v1_0_hw_init(void *handle) 194 { 195 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 196 struct amdgpu_ring *ring = &adev->vcn.inst->ring_dec; 197 int i, r; 198 199 r = amdgpu_ring_test_helper(ring); 200 if (r) 201 goto done; 202 203 for (i = 0; i < adev->vcn.num_enc_rings; ++i) { 204 ring = &adev->vcn.inst->ring_enc[i]; 205 r = amdgpu_ring_test_helper(ring); 206 if (r) 207 goto done; 208 } 209 210 ring = &adev->jpeg.inst->ring_dec; 211 r = amdgpu_ring_test_helper(ring); 212 if (r) 213 goto done; 214 215 done: 216 if (!r) 217 DRM_INFO("VCN decode and encode initialized successfully(under %s).\n", 218 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode"); 219 220 return r; 221 } 222 223 /** 224 * vcn_v1_0_hw_fini - stop the hardware block 225 * 226 * @handle: amdgpu_device pointer 227 * 228 * Stop the VCN block, mark ring as not ready any more 229 */ 230 static int vcn_v1_0_hw_fini(void *handle) 231 { 232 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 233 234 cancel_delayed_work_sync(&adev->vcn.idle_work); 235 236 if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) || 237 (adev->vcn.cur_state != AMD_PG_STATE_GATE && 238 RREG32_SOC15(VCN, 0, mmUVD_STATUS))) { 239 vcn_v1_0_set_powergating_state(adev, AMD_PG_STATE_GATE); 240 } 241 242 return 0; 243 } 244 245 /** 246 * vcn_v1_0_suspend - suspend VCN block 247 * 248 * @handle: amdgpu_device pointer 249 * 250 * HW fini and suspend VCN block 251 */ 252 static int vcn_v1_0_suspend(void *handle) 253 { 254 int r; 255 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 256 257 r = vcn_v1_0_hw_fini(adev); 258 if (r) 259 return r; 260 261 r = amdgpu_vcn_suspend(adev); 262 263 return r; 264 } 265 266 /** 267 * vcn_v1_0_resume - resume VCN block 268 * 269 * @handle: amdgpu_device pointer 270 * 271 * Resume firmware and hw init VCN block 272 */ 273 static int vcn_v1_0_resume(void *handle) 274 { 275 int r; 276 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 277 278 r = amdgpu_vcn_resume(adev); 279 if (r) 280 return r; 281 282 r = vcn_v1_0_hw_init(adev); 283 284 return r; 285 } 286 287 /** 288 * vcn_v1_0_mc_resume_spg_mode - memory controller programming 289 * 290 * @adev: amdgpu_device pointer 291 * 292 * Let the VCN memory controller know it's offsets 293 */ 294 static void vcn_v1_0_mc_resume_spg_mode(struct amdgpu_device *adev) 295 { 296 uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4); 297 uint32_t offset; 298 299 /* cache window 0: fw */ 300 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) { 301 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW, 302 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_lo)); 303 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH, 304 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_hi)); 305 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0, 0); 306 offset = 0; 307 } else { 308 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW, 309 lower_32_bits(adev->vcn.inst->gpu_addr)); 310 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH, 311 upper_32_bits(adev->vcn.inst->gpu_addr)); 312 offset = size; 313 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0, 314 AMDGPU_UVD_FIRMWARE_OFFSET >> 3); 315 } 316 317 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE0, size); 318 319 /* cache window 1: stack */ 320 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW, 321 lower_32_bits(adev->vcn.inst->gpu_addr + offset)); 322 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH, 323 upper_32_bits(adev->vcn.inst->gpu_addr + offset)); 324 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET1, 0); 325 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE); 326 327 /* cache window 2: context */ 328 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW, 329 lower_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE)); 330 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH, 331 upper_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE)); 332 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET2, 0); 333 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE); 334 335 WREG32_SOC15(UVD, 0, mmUVD_UDEC_ADDR_CONFIG, 336 adev->gfx.config.gb_addr_config); 337 WREG32_SOC15(UVD, 0, mmUVD_UDEC_DB_ADDR_CONFIG, 338 adev->gfx.config.gb_addr_config); 339 WREG32_SOC15(UVD, 0, mmUVD_UDEC_DBW_ADDR_CONFIG, 340 adev->gfx.config.gb_addr_config); 341 WREG32_SOC15(UVD, 0, mmUVD_UDEC_DBW_UV_ADDR_CONFIG, 342 adev->gfx.config.gb_addr_config); 343 WREG32_SOC15(UVD, 0, mmUVD_MIF_CURR_ADDR_CONFIG, 344 adev->gfx.config.gb_addr_config); 345 WREG32_SOC15(UVD, 0, mmUVD_MIF_CURR_UV_ADDR_CONFIG, 346 adev->gfx.config.gb_addr_config); 347 WREG32_SOC15(UVD, 0, mmUVD_MIF_RECON1_ADDR_CONFIG, 348 adev->gfx.config.gb_addr_config); 349 WREG32_SOC15(UVD, 0, mmUVD_MIF_RECON1_UV_ADDR_CONFIG, 350 adev->gfx.config.gb_addr_config); 351 WREG32_SOC15(UVD, 0, mmUVD_MIF_REF_ADDR_CONFIG, 352 adev->gfx.config.gb_addr_config); 353 WREG32_SOC15(UVD, 0, mmUVD_MIF_REF_UV_ADDR_CONFIG, 354 adev->gfx.config.gb_addr_config); 355 WREG32_SOC15(UVD, 0, mmUVD_JPEG_ADDR_CONFIG, 356 adev->gfx.config.gb_addr_config); 357 WREG32_SOC15(UVD, 0, mmUVD_JPEG_UV_ADDR_CONFIG, 358 adev->gfx.config.gb_addr_config); 359 } 360 361 static void vcn_v1_0_mc_resume_dpg_mode(struct amdgpu_device *adev) 362 { 363 uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4); 364 uint32_t offset; 365 366 /* cache window 0: fw */ 367 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) { 368 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW, 369 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_lo), 370 0xFFFFFFFF, 0); 371 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH, 372 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_hi), 373 0xFFFFFFFF, 0); 374 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0, 0, 375 0xFFFFFFFF, 0); 376 offset = 0; 377 } else { 378 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW, 379 lower_32_bits(adev->vcn.inst->gpu_addr), 0xFFFFFFFF, 0); 380 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH, 381 upper_32_bits(adev->vcn.inst->gpu_addr), 0xFFFFFFFF, 0); 382 offset = size; 383 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0, 384 AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0xFFFFFFFF, 0); 385 } 386 387 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_VCPU_CACHE_SIZE0, size, 0xFFFFFFFF, 0); 388 389 /* cache window 1: stack */ 390 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW, 391 lower_32_bits(adev->vcn.inst->gpu_addr + offset), 0xFFFFFFFF, 0); 392 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH, 393 upper_32_bits(adev->vcn.inst->gpu_addr + offset), 0xFFFFFFFF, 0); 394 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_VCPU_CACHE_OFFSET1, 0, 395 0xFFFFFFFF, 0); 396 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE, 397 0xFFFFFFFF, 0); 398 399 /* cache window 2: context */ 400 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW, 401 lower_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 402 0xFFFFFFFF, 0); 403 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH, 404 upper_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 405 0xFFFFFFFF, 0); 406 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_VCPU_CACHE_OFFSET2, 0, 0xFFFFFFFF, 0); 407 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE, 408 0xFFFFFFFF, 0); 409 410 /* VCN global tiling registers */ 411 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_UDEC_ADDR_CONFIG, 412 adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); 413 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_UDEC_DB_ADDR_CONFIG, 414 adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); 415 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_UDEC_DBW_ADDR_CONFIG, 416 adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); 417 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_UDEC_DBW_UV_ADDR_CONFIG, 418 adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); 419 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MIF_CURR_ADDR_CONFIG, 420 adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); 421 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MIF_CURR_UV_ADDR_CONFIG, 422 adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); 423 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MIF_RECON1_ADDR_CONFIG, 424 adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); 425 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MIF_RECON1_UV_ADDR_CONFIG, 426 adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); 427 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MIF_REF_ADDR_CONFIG, 428 adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); 429 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MIF_REF_UV_ADDR_CONFIG, 430 adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); 431 } 432 433 /** 434 * vcn_v1_0_disable_clock_gating - disable VCN clock gating 435 * 436 * @adev: amdgpu_device pointer 437 * 438 * Disable clock gating for VCN block 439 */ 440 static void vcn_v1_0_disable_clock_gating(struct amdgpu_device *adev) 441 { 442 uint32_t data; 443 444 /* JPEG disable CGC */ 445 data = RREG32_SOC15(VCN, 0, mmJPEG_CGC_CTRL); 446 447 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG) 448 data |= 1 << JPEG_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 449 else 450 data &= ~JPEG_CGC_CTRL__DYN_CLOCK_MODE_MASK; 451 452 data |= 1 << JPEG_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT; 453 data |= 4 << JPEG_CGC_CTRL__CLK_OFF_DELAY__SHIFT; 454 WREG32_SOC15(VCN, 0, mmJPEG_CGC_CTRL, data); 455 456 data = RREG32_SOC15(VCN, 0, mmJPEG_CGC_GATE); 457 data &= ~(JPEG_CGC_GATE__JPEG_MASK | JPEG_CGC_GATE__JPEG2_MASK); 458 WREG32_SOC15(VCN, 0, mmJPEG_CGC_GATE, data); 459 460 /* UVD disable CGC */ 461 data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL); 462 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG) 463 data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 464 else 465 data &= ~ UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK; 466 467 data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT; 468 data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT; 469 WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data); 470 471 data = RREG32_SOC15(VCN, 0, mmUVD_CGC_GATE); 472 data &= ~(UVD_CGC_GATE__SYS_MASK 473 | UVD_CGC_GATE__UDEC_MASK 474 | UVD_CGC_GATE__MPEG2_MASK 475 | UVD_CGC_GATE__REGS_MASK 476 | UVD_CGC_GATE__RBC_MASK 477 | UVD_CGC_GATE__LMI_MC_MASK 478 | UVD_CGC_GATE__LMI_UMC_MASK 479 | UVD_CGC_GATE__IDCT_MASK 480 | UVD_CGC_GATE__MPRD_MASK 481 | UVD_CGC_GATE__MPC_MASK 482 | UVD_CGC_GATE__LBSI_MASK 483 | UVD_CGC_GATE__LRBBM_MASK 484 | UVD_CGC_GATE__UDEC_RE_MASK 485 | UVD_CGC_GATE__UDEC_CM_MASK 486 | UVD_CGC_GATE__UDEC_IT_MASK 487 | UVD_CGC_GATE__UDEC_DB_MASK 488 | UVD_CGC_GATE__UDEC_MP_MASK 489 | UVD_CGC_GATE__WCB_MASK 490 | UVD_CGC_GATE__VCPU_MASK 491 | UVD_CGC_GATE__SCPU_MASK); 492 WREG32_SOC15(VCN, 0, mmUVD_CGC_GATE, data); 493 494 data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL); 495 data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK 496 | UVD_CGC_CTRL__UDEC_CM_MODE_MASK 497 | UVD_CGC_CTRL__UDEC_IT_MODE_MASK 498 | UVD_CGC_CTRL__UDEC_DB_MODE_MASK 499 | UVD_CGC_CTRL__UDEC_MP_MODE_MASK 500 | UVD_CGC_CTRL__SYS_MODE_MASK 501 | UVD_CGC_CTRL__UDEC_MODE_MASK 502 | UVD_CGC_CTRL__MPEG2_MODE_MASK 503 | UVD_CGC_CTRL__REGS_MODE_MASK 504 | UVD_CGC_CTRL__RBC_MODE_MASK 505 | UVD_CGC_CTRL__LMI_MC_MODE_MASK 506 | UVD_CGC_CTRL__LMI_UMC_MODE_MASK 507 | UVD_CGC_CTRL__IDCT_MODE_MASK 508 | UVD_CGC_CTRL__MPRD_MODE_MASK 509 | UVD_CGC_CTRL__MPC_MODE_MASK 510 | UVD_CGC_CTRL__LBSI_MODE_MASK 511 | UVD_CGC_CTRL__LRBBM_MODE_MASK 512 | UVD_CGC_CTRL__WCB_MODE_MASK 513 | UVD_CGC_CTRL__VCPU_MODE_MASK 514 | UVD_CGC_CTRL__SCPU_MODE_MASK); 515 WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data); 516 517 /* turn on */ 518 data = RREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_GATE); 519 data |= (UVD_SUVD_CGC_GATE__SRE_MASK 520 | UVD_SUVD_CGC_GATE__SIT_MASK 521 | UVD_SUVD_CGC_GATE__SMP_MASK 522 | UVD_SUVD_CGC_GATE__SCM_MASK 523 | UVD_SUVD_CGC_GATE__SDB_MASK 524 | UVD_SUVD_CGC_GATE__SRE_H264_MASK 525 | UVD_SUVD_CGC_GATE__SRE_HEVC_MASK 526 | UVD_SUVD_CGC_GATE__SIT_H264_MASK 527 | UVD_SUVD_CGC_GATE__SIT_HEVC_MASK 528 | UVD_SUVD_CGC_GATE__SCM_H264_MASK 529 | UVD_SUVD_CGC_GATE__SCM_HEVC_MASK 530 | UVD_SUVD_CGC_GATE__SDB_H264_MASK 531 | UVD_SUVD_CGC_GATE__SDB_HEVC_MASK 532 | UVD_SUVD_CGC_GATE__SCLR_MASK 533 | UVD_SUVD_CGC_GATE__UVD_SC_MASK 534 | UVD_SUVD_CGC_GATE__ENT_MASK 535 | UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK 536 | UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK 537 | UVD_SUVD_CGC_GATE__SITE_MASK 538 | UVD_SUVD_CGC_GATE__SRE_VP9_MASK 539 | UVD_SUVD_CGC_GATE__SCM_VP9_MASK 540 | UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK 541 | UVD_SUVD_CGC_GATE__SDB_VP9_MASK 542 | UVD_SUVD_CGC_GATE__IME_HEVC_MASK); 543 WREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_GATE, data); 544 545 data = RREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL); 546 data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK 547 | UVD_SUVD_CGC_CTRL__SIT_MODE_MASK 548 | UVD_SUVD_CGC_CTRL__SMP_MODE_MASK 549 | UVD_SUVD_CGC_CTRL__SCM_MODE_MASK 550 | UVD_SUVD_CGC_CTRL__SDB_MODE_MASK 551 | UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK 552 | UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK 553 | UVD_SUVD_CGC_CTRL__ENT_MODE_MASK 554 | UVD_SUVD_CGC_CTRL__IME_MODE_MASK 555 | UVD_SUVD_CGC_CTRL__SITE_MODE_MASK); 556 WREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL, data); 557 } 558 559 /** 560 * vcn_v1_0_enable_clock_gating - enable VCN clock gating 561 * 562 * @adev: amdgpu_device pointer 563 * 564 * Enable clock gating for VCN block 565 */ 566 static void vcn_v1_0_enable_clock_gating(struct amdgpu_device *adev) 567 { 568 uint32_t data = 0; 569 570 /* enable JPEG CGC */ 571 data = RREG32_SOC15(VCN, 0, mmJPEG_CGC_CTRL); 572 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG) 573 data |= 1 << JPEG_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 574 else 575 data |= 0 << JPEG_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 576 data |= 1 << JPEG_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT; 577 data |= 4 << JPEG_CGC_CTRL__CLK_OFF_DELAY__SHIFT; 578 WREG32_SOC15(VCN, 0, mmJPEG_CGC_CTRL, data); 579 580 data = RREG32_SOC15(VCN, 0, mmJPEG_CGC_GATE); 581 data |= (JPEG_CGC_GATE__JPEG_MASK | JPEG_CGC_GATE__JPEG2_MASK); 582 WREG32_SOC15(VCN, 0, mmJPEG_CGC_GATE, data); 583 584 /* enable UVD CGC */ 585 data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL); 586 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG) 587 data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 588 else 589 data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 590 data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT; 591 data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT; 592 WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data); 593 594 data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL); 595 data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK 596 | UVD_CGC_CTRL__UDEC_CM_MODE_MASK 597 | UVD_CGC_CTRL__UDEC_IT_MODE_MASK 598 | UVD_CGC_CTRL__UDEC_DB_MODE_MASK 599 | UVD_CGC_CTRL__UDEC_MP_MODE_MASK 600 | UVD_CGC_CTRL__SYS_MODE_MASK 601 | UVD_CGC_CTRL__UDEC_MODE_MASK 602 | UVD_CGC_CTRL__MPEG2_MODE_MASK 603 | UVD_CGC_CTRL__REGS_MODE_MASK 604 | UVD_CGC_CTRL__RBC_MODE_MASK 605 | UVD_CGC_CTRL__LMI_MC_MODE_MASK 606 | UVD_CGC_CTRL__LMI_UMC_MODE_MASK 607 | UVD_CGC_CTRL__IDCT_MODE_MASK 608 | UVD_CGC_CTRL__MPRD_MODE_MASK 609 | UVD_CGC_CTRL__MPC_MODE_MASK 610 | UVD_CGC_CTRL__LBSI_MODE_MASK 611 | UVD_CGC_CTRL__LRBBM_MODE_MASK 612 | UVD_CGC_CTRL__WCB_MODE_MASK 613 | UVD_CGC_CTRL__VCPU_MODE_MASK 614 | UVD_CGC_CTRL__SCPU_MODE_MASK); 615 WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data); 616 617 data = RREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL); 618 data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK 619 | UVD_SUVD_CGC_CTRL__SIT_MODE_MASK 620 | UVD_SUVD_CGC_CTRL__SMP_MODE_MASK 621 | UVD_SUVD_CGC_CTRL__SCM_MODE_MASK 622 | UVD_SUVD_CGC_CTRL__SDB_MODE_MASK 623 | UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK 624 | UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK 625 | UVD_SUVD_CGC_CTRL__ENT_MODE_MASK 626 | UVD_SUVD_CGC_CTRL__IME_MODE_MASK 627 | UVD_SUVD_CGC_CTRL__SITE_MODE_MASK); 628 WREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL, data); 629 } 630 631 static void vcn_v1_0_clock_gating_dpg_mode(struct amdgpu_device *adev, uint8_t sram_sel) 632 { 633 uint32_t reg_data = 0; 634 635 /* disable JPEG CGC */ 636 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG) 637 reg_data = 1 << JPEG_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 638 else 639 reg_data = 0 << JPEG_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 640 reg_data |= 1 << JPEG_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT; 641 reg_data |= 4 << JPEG_CGC_CTRL__CLK_OFF_DELAY__SHIFT; 642 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmJPEG_CGC_CTRL, reg_data, 0xFFFFFFFF, sram_sel); 643 644 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmJPEG_CGC_GATE, 0, 0xFFFFFFFF, sram_sel); 645 646 /* enable sw clock gating control */ 647 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG) 648 reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 649 else 650 reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 651 reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT; 652 reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT; 653 reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK | 654 UVD_CGC_CTRL__UDEC_CM_MODE_MASK | 655 UVD_CGC_CTRL__UDEC_IT_MODE_MASK | 656 UVD_CGC_CTRL__UDEC_DB_MODE_MASK | 657 UVD_CGC_CTRL__UDEC_MP_MODE_MASK | 658 UVD_CGC_CTRL__SYS_MODE_MASK | 659 UVD_CGC_CTRL__UDEC_MODE_MASK | 660 UVD_CGC_CTRL__MPEG2_MODE_MASK | 661 UVD_CGC_CTRL__REGS_MODE_MASK | 662 UVD_CGC_CTRL__RBC_MODE_MASK | 663 UVD_CGC_CTRL__LMI_MC_MODE_MASK | 664 UVD_CGC_CTRL__LMI_UMC_MODE_MASK | 665 UVD_CGC_CTRL__IDCT_MODE_MASK | 666 UVD_CGC_CTRL__MPRD_MODE_MASK | 667 UVD_CGC_CTRL__MPC_MODE_MASK | 668 UVD_CGC_CTRL__LBSI_MODE_MASK | 669 UVD_CGC_CTRL__LRBBM_MODE_MASK | 670 UVD_CGC_CTRL__WCB_MODE_MASK | 671 UVD_CGC_CTRL__VCPU_MODE_MASK | 672 UVD_CGC_CTRL__SCPU_MODE_MASK); 673 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_CGC_CTRL, reg_data, 0xFFFFFFFF, sram_sel); 674 675 /* turn off clock gating */ 676 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_CGC_GATE, 0, 0xFFFFFFFF, sram_sel); 677 678 /* turn on SUVD clock gating */ 679 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_SUVD_CGC_GATE, 1, 0xFFFFFFFF, sram_sel); 680 681 /* turn on sw mode in UVD_SUVD_CGC_CTRL */ 682 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_SUVD_CGC_CTRL, 0, 0xFFFFFFFF, sram_sel); 683 } 684 685 static void vcn_1_0_disable_static_power_gating(struct amdgpu_device *adev) 686 { 687 uint32_t data = 0; 688 689 if (adev->pg_flags & AMD_PG_SUPPORT_VCN) { 690 data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT 691 | 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT 692 | 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT 693 | 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT 694 | 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT 695 | 2 << UVD_PGFSM_CONFIG__UVDIL_PWR_CONFIG__SHIFT 696 | 2 << UVD_PGFSM_CONFIG__UVDIR_PWR_CONFIG__SHIFT 697 | 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT 698 | 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT 699 | 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT 700 | 2 << UVD_PGFSM_CONFIG__UVDW_PWR_CONFIG__SHIFT); 701 702 WREG32_SOC15(VCN, 0, mmUVD_PGFSM_CONFIG, data); 703 SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_PGFSM_STATUS, UVD_PGFSM_STATUS__UVDM_UVDU_PWR_ON, 0xFFFFFF); 704 } else { 705 data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT 706 | 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT 707 | 1 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT 708 | 1 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT 709 | 1 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT 710 | 1 << UVD_PGFSM_CONFIG__UVDIL_PWR_CONFIG__SHIFT 711 | 1 << UVD_PGFSM_CONFIG__UVDIR_PWR_CONFIG__SHIFT 712 | 1 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT 713 | 1 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT 714 | 1 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT 715 | 1 << UVD_PGFSM_CONFIG__UVDW_PWR_CONFIG__SHIFT); 716 WREG32_SOC15(VCN, 0, mmUVD_PGFSM_CONFIG, data); 717 SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_PGFSM_STATUS, 0, 0xFFFFFFFF); 718 } 719 720 /* polling UVD_PGFSM_STATUS to confirm UVDM_PWR_STATUS , UVDU_PWR_STATUS are 0 (power on) */ 721 722 data = RREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS); 723 data &= ~0x103; 724 if (adev->pg_flags & AMD_PG_SUPPORT_VCN) 725 data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON | UVD_POWER_STATUS__UVD_PG_EN_MASK; 726 727 WREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS, data); 728 } 729 730 static void vcn_1_0_enable_static_power_gating(struct amdgpu_device *adev) 731 { 732 uint32_t data = 0; 733 734 if (adev->pg_flags & AMD_PG_SUPPORT_VCN) { 735 /* Before power off, this indicator has to be turned on */ 736 data = RREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS); 737 data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK; 738 data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF; 739 WREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS, data); 740 741 742 data = (2 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT 743 | 2 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT 744 | 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT 745 | 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT 746 | 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT 747 | 2 << UVD_PGFSM_CONFIG__UVDIL_PWR_CONFIG__SHIFT 748 | 2 << UVD_PGFSM_CONFIG__UVDIR_PWR_CONFIG__SHIFT 749 | 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT 750 | 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT 751 | 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT 752 | 2 << UVD_PGFSM_CONFIG__UVDW_PWR_CONFIG__SHIFT); 753 754 WREG32_SOC15(VCN, 0, mmUVD_PGFSM_CONFIG, data); 755 756 data = (2 << UVD_PGFSM_STATUS__UVDM_PWR_STATUS__SHIFT 757 | 2 << UVD_PGFSM_STATUS__UVDU_PWR_STATUS__SHIFT 758 | 2 << UVD_PGFSM_STATUS__UVDF_PWR_STATUS__SHIFT 759 | 2 << UVD_PGFSM_STATUS__UVDC_PWR_STATUS__SHIFT 760 | 2 << UVD_PGFSM_STATUS__UVDB_PWR_STATUS__SHIFT 761 | 2 << UVD_PGFSM_STATUS__UVDIL_PWR_STATUS__SHIFT 762 | 2 << UVD_PGFSM_STATUS__UVDIR_PWR_STATUS__SHIFT 763 | 2 << UVD_PGFSM_STATUS__UVDTD_PWR_STATUS__SHIFT 764 | 2 << UVD_PGFSM_STATUS__UVDTE_PWR_STATUS__SHIFT 765 | 2 << UVD_PGFSM_STATUS__UVDE_PWR_STATUS__SHIFT 766 | 2 << UVD_PGFSM_STATUS__UVDW_PWR_STATUS__SHIFT); 767 SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_PGFSM_STATUS, data, 0xFFFFFFFF); 768 } 769 } 770 771 /** 772 * vcn_v1_0_start_spg_mode - start VCN block 773 * 774 * @adev: amdgpu_device pointer 775 * 776 * Setup and start the VCN block 777 */ 778 static int vcn_v1_0_start_spg_mode(struct amdgpu_device *adev) 779 { 780 struct amdgpu_ring *ring = &adev->vcn.inst->ring_dec; 781 uint32_t rb_bufsz, tmp; 782 uint32_t lmi_swap_cntl; 783 int i, j, r; 784 785 /* disable byte swapping */ 786 lmi_swap_cntl = 0; 787 788 vcn_1_0_disable_static_power_gating(adev); 789 790 tmp = RREG32_SOC15(UVD, 0, mmUVD_STATUS) | UVD_STATUS__UVD_BUSY; 791 WREG32_SOC15(UVD, 0, mmUVD_STATUS, tmp); 792 793 /* disable clock gating */ 794 vcn_v1_0_disable_clock_gating(adev); 795 796 /* disable interupt */ 797 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_MASTINT_EN), 0, 798 ~UVD_MASTINT_EN__VCPU_EN_MASK); 799 800 /* initialize VCN memory controller */ 801 tmp = RREG32_SOC15(UVD, 0, mmUVD_LMI_CTRL); 802 WREG32_SOC15(UVD, 0, mmUVD_LMI_CTRL, tmp | 803 UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK | 804 UVD_LMI_CTRL__MASK_MC_URGENT_MASK | 805 UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK | 806 UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK); 807 808 #ifdef __BIG_ENDIAN 809 /* swap (8 in 32) RB and IB */ 810 lmi_swap_cntl = 0xa; 811 #endif 812 WREG32_SOC15(UVD, 0, mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl); 813 814 tmp = RREG32_SOC15(UVD, 0, mmUVD_MPC_CNTL); 815 tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK; 816 tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT; 817 WREG32_SOC15(UVD, 0, mmUVD_MPC_CNTL, tmp); 818 819 WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUXA0, 820 ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) | 821 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) | 822 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) | 823 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT))); 824 825 WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUXB0, 826 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) | 827 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) | 828 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) | 829 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT))); 830 831 WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUX, 832 ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) | 833 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) | 834 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT))); 835 836 vcn_v1_0_mc_resume_spg_mode(adev); 837 838 WREG32_SOC15(UVD, 0, mmUVD_REG_XX_MASK_1_0, 0x10); 839 WREG32_SOC15(UVD, 0, mmUVD_RBC_XX_IB_REG_CHECK_1_0, 840 RREG32_SOC15(UVD, 0, mmUVD_RBC_XX_IB_REG_CHECK_1_0) | 0x3); 841 842 /* enable VCPU clock */ 843 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CNTL, UVD_VCPU_CNTL__CLK_EN_MASK); 844 845 /* boot up the VCPU */ 846 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 0, 847 ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK); 848 849 /* enable UMC */ 850 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_CTRL2), 0, 851 ~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK); 852 853 tmp = RREG32_SOC15(UVD, 0, mmUVD_SOFT_RESET); 854 tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK; 855 tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK; 856 WREG32_SOC15(UVD, 0, mmUVD_SOFT_RESET, tmp); 857 858 for (i = 0; i < 10; ++i) { 859 uint32_t status; 860 861 for (j = 0; j < 100; ++j) { 862 status = RREG32_SOC15(UVD, 0, mmUVD_STATUS); 863 if (status & UVD_STATUS__IDLE) 864 break; 865 mdelay(10); 866 } 867 r = 0; 868 if (status & UVD_STATUS__IDLE) 869 break; 870 871 DRM_ERROR("VCN decode not responding, trying to reset the VCPU!!!\n"); 872 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 873 UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK, 874 ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK); 875 mdelay(10); 876 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 0, 877 ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK); 878 mdelay(10); 879 r = -1; 880 } 881 882 if (r) { 883 DRM_ERROR("VCN decode not responding, giving up!!!\n"); 884 return r; 885 } 886 /* enable master interrupt */ 887 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_MASTINT_EN), 888 UVD_MASTINT_EN__VCPU_EN_MASK, ~UVD_MASTINT_EN__VCPU_EN_MASK); 889 890 /* enable system interrupt for JRBC, TODO: move to set interrupt*/ 891 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SYS_INT_EN), 892 UVD_SYS_INT_EN__UVD_JRBC_EN_MASK, 893 ~UVD_SYS_INT_EN__UVD_JRBC_EN_MASK); 894 895 /* clear the busy bit of UVD_STATUS */ 896 tmp = RREG32_SOC15(UVD, 0, mmUVD_STATUS) & ~UVD_STATUS__UVD_BUSY; 897 WREG32_SOC15(UVD, 0, mmUVD_STATUS, tmp); 898 899 /* force RBC into idle state */ 900 rb_bufsz = order_base_2(ring->ring_size); 901 tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz); 902 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1); 903 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1); 904 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1); 905 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1); 906 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_CNTL, tmp); 907 908 /* set the write pointer delay */ 909 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR_CNTL, 0); 910 911 /* set the wb address */ 912 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR_ADDR, 913 (upper_32_bits(ring->gpu_addr) >> 2)); 914 915 /* program the RB_BASE for ring buffer */ 916 WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW, 917 lower_32_bits(ring->gpu_addr)); 918 WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH, 919 upper_32_bits(ring->gpu_addr)); 920 921 /* Initialize the ring buffer's read and write pointers */ 922 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR, 0); 923 924 WREG32_SOC15(UVD, 0, mmUVD_SCRATCH2, 0); 925 926 ring->wptr = RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR); 927 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR, 928 lower_32_bits(ring->wptr)); 929 930 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_RBC_RB_CNTL), 0, 931 ~UVD_RBC_RB_CNTL__RB_NO_FETCH_MASK); 932 933 ring = &adev->vcn.inst->ring_enc[0]; 934 WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR, lower_32_bits(ring->wptr)); 935 WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR, lower_32_bits(ring->wptr)); 936 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO, ring->gpu_addr); 937 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr)); 938 WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE, ring->ring_size / 4); 939 940 ring = &adev->vcn.inst->ring_enc[1]; 941 WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr)); 942 WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr)); 943 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO2, ring->gpu_addr); 944 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr)); 945 WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE2, ring->ring_size / 4); 946 947 jpeg_v1_0_start(adev, 0); 948 949 return 0; 950 } 951 952 static int vcn_v1_0_start_dpg_mode(struct amdgpu_device *adev) 953 { 954 struct amdgpu_ring *ring = &adev->vcn.inst->ring_dec; 955 uint32_t rb_bufsz, tmp; 956 uint32_t lmi_swap_cntl; 957 958 /* disable byte swapping */ 959 lmi_swap_cntl = 0; 960 961 vcn_1_0_enable_static_power_gating(adev); 962 963 /* enable dynamic power gating mode */ 964 tmp = RREG32_SOC15(UVD, 0, mmUVD_POWER_STATUS); 965 tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK; 966 tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK; 967 WREG32_SOC15(UVD, 0, mmUVD_POWER_STATUS, tmp); 968 969 /* enable clock gating */ 970 vcn_v1_0_clock_gating_dpg_mode(adev, 0); 971 972 /* enable VCPU clock */ 973 tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT); 974 tmp |= UVD_VCPU_CNTL__CLK_EN_MASK; 975 tmp |= UVD_VCPU_CNTL__MIF_WR_LOW_THRESHOLD_BP_MASK; 976 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_VCPU_CNTL, tmp, 0xFFFFFFFF, 0); 977 978 /* disable interupt */ 979 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MASTINT_EN, 980 0, UVD_MASTINT_EN__VCPU_EN_MASK, 0); 981 982 /* initialize VCN memory controller */ 983 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_CTRL, 984 (8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) | 985 UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK | 986 UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK | 987 UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK | 988 UVD_LMI_CTRL__REQ_MODE_MASK | 989 UVD_LMI_CTRL__CRC_RESET_MASK | 990 UVD_LMI_CTRL__MASK_MC_URGENT_MASK | 991 0x00100000L, 0xFFFFFFFF, 0); 992 993 #ifdef __BIG_ENDIAN 994 /* swap (8 in 32) RB and IB */ 995 lmi_swap_cntl = 0xa; 996 #endif 997 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl, 0xFFFFFFFF, 0); 998 999 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MPC_CNTL, 1000 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0xFFFFFFFF, 0); 1001 1002 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MPC_SET_MUXA0, 1003 ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) | 1004 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) | 1005 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) | 1006 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0xFFFFFFFF, 0); 1007 1008 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MPC_SET_MUXB0, 1009 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) | 1010 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) | 1011 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) | 1012 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0xFFFFFFFF, 0); 1013 1014 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MPC_SET_MUX, 1015 ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) | 1016 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) | 1017 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0xFFFFFFFF, 0); 1018 1019 vcn_v1_0_mc_resume_dpg_mode(adev); 1020 1021 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_REG_XX_MASK, 0x10, 0xFFFFFFFF, 0); 1022 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_RBC_XX_IB_REG_CHECK, 0x3, 0xFFFFFFFF, 0); 1023 1024 /* boot up the VCPU */ 1025 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_SOFT_RESET, 0, 0xFFFFFFFF, 0); 1026 1027 /* enable UMC */ 1028 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_CTRL2, 1029 0x1F << UVD_LMI_CTRL2__RE_OFLD_MIF_WR_REQ_NUM__SHIFT, 1030 0xFFFFFFFF, 0); 1031 1032 /* enable master interrupt */ 1033 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MASTINT_EN, 1034 UVD_MASTINT_EN__VCPU_EN_MASK, UVD_MASTINT_EN__VCPU_EN_MASK, 0); 1035 1036 vcn_v1_0_clock_gating_dpg_mode(adev, 1); 1037 /* setup mmUVD_LMI_CTRL */ 1038 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_CTRL, 1039 (8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) | 1040 UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK | 1041 UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK | 1042 UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK | 1043 UVD_LMI_CTRL__REQ_MODE_MASK | 1044 UVD_LMI_CTRL__CRC_RESET_MASK | 1045 UVD_LMI_CTRL__MASK_MC_URGENT_MASK | 1046 0x00100000L, 0xFFFFFFFF, 1); 1047 1048 tmp = adev->gfx.config.gb_addr_config; 1049 /* setup VCN global tiling registers */ 1050 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_JPEG_ADDR_CONFIG, tmp, 0xFFFFFFFF, 1); 1051 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_JPEG_UV_ADDR_CONFIG, tmp, 0xFFFFFFFF, 1); 1052 1053 /* enable System Interrupt for JRBC */ 1054 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_SYS_INT_EN, 1055 UVD_SYS_INT_EN__UVD_JRBC_EN_MASK, 0xFFFFFFFF, 1); 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 /* set the write pointer delay */ 1067 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR_CNTL, 0); 1068 1069 /* set the wb address */ 1070 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR_ADDR, 1071 (upper_32_bits(ring->gpu_addr) >> 2)); 1072 1073 /* program the RB_BASE for ring buffer */ 1074 WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW, 1075 lower_32_bits(ring->gpu_addr)); 1076 WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH, 1077 upper_32_bits(ring->gpu_addr)); 1078 1079 /* Initialize the ring buffer's read and write pointers */ 1080 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR, 0); 1081 1082 WREG32_SOC15(UVD, 0, mmUVD_SCRATCH2, 0); 1083 1084 ring->wptr = RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR); 1085 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR, 1086 lower_32_bits(ring->wptr)); 1087 1088 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_RBC_RB_CNTL), 0, 1089 ~UVD_RBC_RB_CNTL__RB_NO_FETCH_MASK); 1090 1091 jpeg_v1_0_start(adev, 1); 1092 1093 return 0; 1094 } 1095 1096 static int vcn_v1_0_start(struct amdgpu_device *adev) 1097 { 1098 int r; 1099 1100 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) 1101 r = vcn_v1_0_start_dpg_mode(adev); 1102 else 1103 r = vcn_v1_0_start_spg_mode(adev); 1104 return r; 1105 } 1106 1107 /** 1108 * vcn_v1_0_stop_spg_mode - stop VCN block 1109 * 1110 * @adev: amdgpu_device pointer 1111 * 1112 * stop the VCN block 1113 */ 1114 static int vcn_v1_0_stop_spg_mode(struct amdgpu_device *adev) 1115 { 1116 int tmp; 1117 1118 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7); 1119 1120 tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK | 1121 UVD_LMI_STATUS__READ_CLEAN_MASK | 1122 UVD_LMI_STATUS__WRITE_CLEAN_MASK | 1123 UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK; 1124 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_LMI_STATUS, tmp, tmp); 1125 1126 /* stall UMC channel */ 1127 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_CTRL2), 1128 UVD_LMI_CTRL2__STALL_ARB_UMC_MASK, 1129 ~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK); 1130 1131 tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK | 1132 UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK; 1133 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_LMI_STATUS, tmp, tmp); 1134 1135 /* disable VCPU clock */ 1136 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_VCPU_CNTL), 0, 1137 ~UVD_VCPU_CNTL__CLK_EN_MASK); 1138 1139 /* reset LMI UMC/LMI */ 1140 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 1141 UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK, 1142 ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK); 1143 1144 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 1145 UVD_SOFT_RESET__LMI_SOFT_RESET_MASK, 1146 ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK); 1147 1148 /* put VCPU into reset */ 1149 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 1150 UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK, 1151 ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK); 1152 1153 WREG32_SOC15(UVD, 0, mmUVD_STATUS, 0); 1154 1155 vcn_v1_0_enable_clock_gating(adev); 1156 vcn_1_0_enable_static_power_gating(adev); 1157 return 0; 1158 } 1159 1160 static int vcn_v1_0_stop_dpg_mode(struct amdgpu_device *adev) 1161 { 1162 uint32_t tmp; 1163 1164 /* Wait for power status to be UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF */ 1165 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 1166 UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF, 1167 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); 1168 1169 /* wait for read ptr to be equal to write ptr */ 1170 tmp = RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR); 1171 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF); 1172 1173 tmp = RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2); 1174 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF); 1175 1176 tmp = RREG32_SOC15(UVD, 0, mmUVD_JRBC_RB_WPTR); 1177 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_JRBC_RB_RPTR, tmp, 0xFFFFFFFF); 1178 1179 tmp = RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF; 1180 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF); 1181 1182 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 1183 UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF, 1184 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); 1185 1186 /* disable dynamic power gating mode */ 1187 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_POWER_STATUS), 0, 1188 ~UVD_POWER_STATUS__UVD_PG_MODE_MASK); 1189 1190 return 0; 1191 } 1192 1193 static int vcn_v1_0_stop(struct amdgpu_device *adev) 1194 { 1195 int r; 1196 1197 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) 1198 r = vcn_v1_0_stop_dpg_mode(adev); 1199 else 1200 r = vcn_v1_0_stop_spg_mode(adev); 1201 1202 return r; 1203 } 1204 1205 static int vcn_v1_0_pause_dpg_mode(struct amdgpu_device *adev, 1206 int inst_idx, struct dpg_pause_state *new_state) 1207 { 1208 int ret_code; 1209 uint32_t reg_data = 0; 1210 uint32_t reg_data2 = 0; 1211 struct amdgpu_ring *ring; 1212 1213 /* pause/unpause if state is changed */ 1214 if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) { 1215 DRM_DEBUG("dpg pause state changed %d:%d -> %d:%d", 1216 adev->vcn.inst[inst_idx].pause_state.fw_based, 1217 adev->vcn.inst[inst_idx].pause_state.jpeg, 1218 new_state->fw_based, new_state->jpeg); 1219 1220 reg_data = RREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE) & 1221 (~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK); 1222 1223 if (new_state->fw_based == VCN_DPG_STATE__PAUSE) { 1224 ret_code = 0; 1225 1226 if (!(reg_data & UVD_DPG_PAUSE__JPEG_PAUSE_DPG_ACK_MASK)) 1227 ret_code = SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 1228 UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF, 1229 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); 1230 1231 if (!ret_code) { 1232 /* pause DPG non-jpeg */ 1233 reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK; 1234 WREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE, reg_data); 1235 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_DPG_PAUSE, 1236 UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK, 1237 UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK); 1238 1239 /* Restore */ 1240 ring = &adev->vcn.inst->ring_enc[0]; 1241 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO, ring->gpu_addr); 1242 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr)); 1243 WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE, ring->ring_size / 4); 1244 WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR, lower_32_bits(ring->wptr)); 1245 WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR, lower_32_bits(ring->wptr)); 1246 1247 ring = &adev->vcn.inst->ring_enc[1]; 1248 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO2, ring->gpu_addr); 1249 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr)); 1250 WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE2, ring->ring_size / 4); 1251 WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr)); 1252 WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr)); 1253 1254 ring = &adev->vcn.inst->ring_dec; 1255 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR, 1256 RREG32_SOC15(UVD, 0, mmUVD_SCRATCH2) & 0x7FFFFFFF); 1257 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 1258 UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, 1259 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); 1260 } 1261 } else { 1262 /* unpause dpg non-jpeg, no need to wait */ 1263 reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK; 1264 WREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE, reg_data); 1265 } 1266 adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based; 1267 } 1268 1269 /* pause/unpause if state is changed */ 1270 if (adev->vcn.inst[inst_idx].pause_state.jpeg != new_state->jpeg) { 1271 DRM_DEBUG("dpg pause state changed %d:%d -> %d:%d", 1272 adev->vcn.inst[inst_idx].pause_state.fw_based, 1273 adev->vcn.inst[inst_idx].pause_state.jpeg, 1274 new_state->fw_based, new_state->jpeg); 1275 1276 reg_data = RREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE) & 1277 (~UVD_DPG_PAUSE__JPEG_PAUSE_DPG_ACK_MASK); 1278 1279 if (new_state->jpeg == VCN_DPG_STATE__PAUSE) { 1280 ret_code = 0; 1281 1282 if (!(reg_data & UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK)) 1283 ret_code = SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 1284 UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF, 1285 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); 1286 1287 if (!ret_code) { 1288 /* Make sure JPRG Snoop is disabled before sending the pause */ 1289 reg_data2 = RREG32_SOC15(UVD, 0, mmUVD_POWER_STATUS); 1290 reg_data2 |= UVD_POWER_STATUS__JRBC_SNOOP_DIS_MASK; 1291 WREG32_SOC15(UVD, 0, mmUVD_POWER_STATUS, reg_data2); 1292 1293 /* pause DPG jpeg */ 1294 reg_data |= UVD_DPG_PAUSE__JPEG_PAUSE_DPG_REQ_MASK; 1295 WREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE, reg_data); 1296 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_DPG_PAUSE, 1297 UVD_DPG_PAUSE__JPEG_PAUSE_DPG_ACK_MASK, 1298 UVD_DPG_PAUSE__JPEG_PAUSE_DPG_ACK_MASK); 1299 1300 /* Restore */ 1301 ring = &adev->jpeg.inst->ring_dec; 1302 WREG32_SOC15(UVD, 0, mmUVD_LMI_JRBC_RB_VMID, 0); 1303 WREG32_SOC15(UVD, 0, mmUVD_JRBC_RB_CNTL, 1304 UVD_JRBC_RB_CNTL__RB_NO_FETCH_MASK | 1305 UVD_JRBC_RB_CNTL__RB_RPTR_WR_EN_MASK); 1306 WREG32_SOC15(UVD, 0, mmUVD_LMI_JRBC_RB_64BIT_BAR_LOW, 1307 lower_32_bits(ring->gpu_addr)); 1308 WREG32_SOC15(UVD, 0, mmUVD_LMI_JRBC_RB_64BIT_BAR_HIGH, 1309 upper_32_bits(ring->gpu_addr)); 1310 WREG32_SOC15(UVD, 0, mmUVD_JRBC_RB_RPTR, ring->wptr); 1311 WREG32_SOC15(UVD, 0, mmUVD_JRBC_RB_WPTR, ring->wptr); 1312 WREG32_SOC15(UVD, 0, mmUVD_JRBC_RB_CNTL, 1313 UVD_JRBC_RB_CNTL__RB_RPTR_WR_EN_MASK); 1314 1315 ring = &adev->vcn.inst->ring_dec; 1316 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR, 1317 RREG32_SOC15(UVD, 0, mmUVD_SCRATCH2) & 0x7FFFFFFF); 1318 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 1319 UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, 1320 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); 1321 } 1322 } else { 1323 /* unpause dpg jpeg, no need to wait */ 1324 reg_data &= ~UVD_DPG_PAUSE__JPEG_PAUSE_DPG_REQ_MASK; 1325 WREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE, reg_data); 1326 } 1327 adev->vcn.inst[inst_idx].pause_state.jpeg = new_state->jpeg; 1328 } 1329 1330 return 0; 1331 } 1332 1333 static bool vcn_v1_0_is_idle(void *handle) 1334 { 1335 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1336 1337 return (RREG32_SOC15(VCN, 0, mmUVD_STATUS) == UVD_STATUS__IDLE); 1338 } 1339 1340 static int vcn_v1_0_wait_for_idle(void *handle) 1341 { 1342 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1343 int ret; 1344 1345 ret = SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_STATUS, UVD_STATUS__IDLE, 1346 UVD_STATUS__IDLE); 1347 1348 return ret; 1349 } 1350 1351 static int vcn_v1_0_set_clockgating_state(void *handle, 1352 enum amd_clockgating_state state) 1353 { 1354 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1355 bool enable = (state == AMD_CG_STATE_GATE); 1356 1357 if (enable) { 1358 /* wait for STATUS to clear */ 1359 if (!vcn_v1_0_is_idle(handle)) 1360 return -EBUSY; 1361 vcn_v1_0_enable_clock_gating(adev); 1362 } else { 1363 /* disable HW gating and enable Sw gating */ 1364 vcn_v1_0_disable_clock_gating(adev); 1365 } 1366 return 0; 1367 } 1368 1369 /** 1370 * vcn_v1_0_dec_ring_get_rptr - get read pointer 1371 * 1372 * @ring: amdgpu_ring pointer 1373 * 1374 * Returns the current hardware read pointer 1375 */ 1376 static uint64_t vcn_v1_0_dec_ring_get_rptr(struct amdgpu_ring *ring) 1377 { 1378 struct amdgpu_device *adev = ring->adev; 1379 1380 return RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR); 1381 } 1382 1383 /** 1384 * vcn_v1_0_dec_ring_get_wptr - get write pointer 1385 * 1386 * @ring: amdgpu_ring pointer 1387 * 1388 * Returns the current hardware write pointer 1389 */ 1390 static uint64_t vcn_v1_0_dec_ring_get_wptr(struct amdgpu_ring *ring) 1391 { 1392 struct amdgpu_device *adev = ring->adev; 1393 1394 return RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR); 1395 } 1396 1397 /** 1398 * vcn_v1_0_dec_ring_set_wptr - set write pointer 1399 * 1400 * @ring: amdgpu_ring pointer 1401 * 1402 * Commits the write pointer to the hardware 1403 */ 1404 static void vcn_v1_0_dec_ring_set_wptr(struct amdgpu_ring *ring) 1405 { 1406 struct amdgpu_device *adev = ring->adev; 1407 1408 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) 1409 WREG32_SOC15(UVD, 0, mmUVD_SCRATCH2, 1410 lower_32_bits(ring->wptr) | 0x80000000); 1411 1412 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr)); 1413 } 1414 1415 /** 1416 * vcn_v1_0_dec_ring_insert_start - insert a start command 1417 * 1418 * @ring: amdgpu_ring pointer 1419 * 1420 * Write a start command to the ring. 1421 */ 1422 static void vcn_v1_0_dec_ring_insert_start(struct amdgpu_ring *ring) 1423 { 1424 struct amdgpu_device *adev = ring->adev; 1425 1426 amdgpu_ring_write(ring, 1427 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0), 0)); 1428 amdgpu_ring_write(ring, 0); 1429 amdgpu_ring_write(ring, 1430 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD), 0)); 1431 amdgpu_ring_write(ring, VCN_DEC_CMD_PACKET_START << 1); 1432 } 1433 1434 /** 1435 * vcn_v1_0_dec_ring_insert_end - insert a end command 1436 * 1437 * @ring: amdgpu_ring pointer 1438 * 1439 * Write a end command to the ring. 1440 */ 1441 static void vcn_v1_0_dec_ring_insert_end(struct amdgpu_ring *ring) 1442 { 1443 struct amdgpu_device *adev = ring->adev; 1444 1445 amdgpu_ring_write(ring, 1446 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD), 0)); 1447 amdgpu_ring_write(ring, VCN_DEC_CMD_PACKET_END << 1); 1448 } 1449 1450 /** 1451 * vcn_v1_0_dec_ring_emit_fence - emit an fence & trap command 1452 * 1453 * @ring: amdgpu_ring pointer 1454 * @addr: address 1455 * @seq: sequence number 1456 * @flags: fence related flags 1457 * 1458 * Write a fence and a trap command to the ring. 1459 */ 1460 static void vcn_v1_0_dec_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq, 1461 unsigned flags) 1462 { 1463 struct amdgpu_device *adev = ring->adev; 1464 1465 WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT); 1466 1467 amdgpu_ring_write(ring, 1468 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_CONTEXT_ID), 0)); 1469 amdgpu_ring_write(ring, seq); 1470 amdgpu_ring_write(ring, 1471 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0), 0)); 1472 amdgpu_ring_write(ring, addr & 0xffffffff); 1473 amdgpu_ring_write(ring, 1474 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1), 0)); 1475 amdgpu_ring_write(ring, upper_32_bits(addr) & 0xff); 1476 amdgpu_ring_write(ring, 1477 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD), 0)); 1478 amdgpu_ring_write(ring, VCN_DEC_CMD_FENCE << 1); 1479 1480 amdgpu_ring_write(ring, 1481 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0), 0)); 1482 amdgpu_ring_write(ring, 0); 1483 amdgpu_ring_write(ring, 1484 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1), 0)); 1485 amdgpu_ring_write(ring, 0); 1486 amdgpu_ring_write(ring, 1487 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD), 0)); 1488 amdgpu_ring_write(ring, VCN_DEC_CMD_TRAP << 1); 1489 } 1490 1491 /** 1492 * vcn_v1_0_dec_ring_emit_ib - execute indirect buffer 1493 * 1494 * @ring: amdgpu_ring pointer 1495 * @job: job to retrieve vmid from 1496 * @ib: indirect buffer to execute 1497 * @flags: unused 1498 * 1499 * Write ring commands to execute the indirect buffer 1500 */ 1501 static void vcn_v1_0_dec_ring_emit_ib(struct amdgpu_ring *ring, 1502 struct amdgpu_job *job, 1503 struct amdgpu_ib *ib, 1504 uint32_t flags) 1505 { 1506 struct amdgpu_device *adev = ring->adev; 1507 unsigned vmid = AMDGPU_JOB_GET_VMID(job); 1508 1509 amdgpu_ring_write(ring, 1510 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_RBC_IB_VMID), 0)); 1511 amdgpu_ring_write(ring, vmid); 1512 1513 amdgpu_ring_write(ring, 1514 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_RBC_IB_64BIT_BAR_LOW), 0)); 1515 amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr)); 1516 amdgpu_ring_write(ring, 1517 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH), 0)); 1518 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr)); 1519 amdgpu_ring_write(ring, 1520 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_RBC_IB_SIZE), 0)); 1521 amdgpu_ring_write(ring, ib->length_dw); 1522 } 1523 1524 static void vcn_v1_0_dec_ring_emit_reg_wait(struct amdgpu_ring *ring, 1525 uint32_t reg, uint32_t val, 1526 uint32_t mask) 1527 { 1528 struct amdgpu_device *adev = ring->adev; 1529 1530 amdgpu_ring_write(ring, 1531 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0), 0)); 1532 amdgpu_ring_write(ring, reg << 2); 1533 amdgpu_ring_write(ring, 1534 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1), 0)); 1535 amdgpu_ring_write(ring, val); 1536 amdgpu_ring_write(ring, 1537 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GP_SCRATCH8), 0)); 1538 amdgpu_ring_write(ring, mask); 1539 amdgpu_ring_write(ring, 1540 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD), 0)); 1541 amdgpu_ring_write(ring, VCN_DEC_CMD_REG_READ_COND_WAIT << 1); 1542 } 1543 1544 static void vcn_v1_0_dec_ring_emit_vm_flush(struct amdgpu_ring *ring, 1545 unsigned vmid, uint64_t pd_addr) 1546 { 1547 struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub]; 1548 uint32_t data0, data1, mask; 1549 1550 pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr); 1551 1552 /* wait for register write */ 1553 data0 = hub->ctx0_ptb_addr_lo32 + vmid * hub->ctx_addr_distance; 1554 data1 = lower_32_bits(pd_addr); 1555 mask = 0xffffffff; 1556 vcn_v1_0_dec_ring_emit_reg_wait(ring, data0, data1, mask); 1557 } 1558 1559 static void vcn_v1_0_dec_ring_emit_wreg(struct amdgpu_ring *ring, 1560 uint32_t reg, uint32_t val) 1561 { 1562 struct amdgpu_device *adev = ring->adev; 1563 1564 amdgpu_ring_write(ring, 1565 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0), 0)); 1566 amdgpu_ring_write(ring, reg << 2); 1567 amdgpu_ring_write(ring, 1568 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1), 0)); 1569 amdgpu_ring_write(ring, val); 1570 amdgpu_ring_write(ring, 1571 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD), 0)); 1572 amdgpu_ring_write(ring, VCN_DEC_CMD_WRITE_REG << 1); 1573 } 1574 1575 /** 1576 * vcn_v1_0_enc_ring_get_rptr - get enc read pointer 1577 * 1578 * @ring: amdgpu_ring pointer 1579 * 1580 * Returns the current hardware enc read pointer 1581 */ 1582 static uint64_t vcn_v1_0_enc_ring_get_rptr(struct amdgpu_ring *ring) 1583 { 1584 struct amdgpu_device *adev = ring->adev; 1585 1586 if (ring == &adev->vcn.inst->ring_enc[0]) 1587 return RREG32_SOC15(UVD, 0, mmUVD_RB_RPTR); 1588 else 1589 return RREG32_SOC15(UVD, 0, mmUVD_RB_RPTR2); 1590 } 1591 1592 /** 1593 * vcn_v1_0_enc_ring_get_wptr - get enc write pointer 1594 * 1595 * @ring: amdgpu_ring pointer 1596 * 1597 * Returns the current hardware enc write pointer 1598 */ 1599 static uint64_t vcn_v1_0_enc_ring_get_wptr(struct amdgpu_ring *ring) 1600 { 1601 struct amdgpu_device *adev = ring->adev; 1602 1603 if (ring == &adev->vcn.inst->ring_enc[0]) 1604 return RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR); 1605 else 1606 return RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2); 1607 } 1608 1609 /** 1610 * vcn_v1_0_enc_ring_set_wptr - set enc write pointer 1611 * 1612 * @ring: amdgpu_ring pointer 1613 * 1614 * Commits the enc write pointer to the hardware 1615 */ 1616 static void vcn_v1_0_enc_ring_set_wptr(struct amdgpu_ring *ring) 1617 { 1618 struct amdgpu_device *adev = ring->adev; 1619 1620 if (ring == &adev->vcn.inst->ring_enc[0]) 1621 WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR, 1622 lower_32_bits(ring->wptr)); 1623 else 1624 WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2, 1625 lower_32_bits(ring->wptr)); 1626 } 1627 1628 /** 1629 * vcn_v1_0_enc_ring_emit_fence - emit an enc fence & trap command 1630 * 1631 * @ring: amdgpu_ring pointer 1632 * @addr: address 1633 * @seq: sequence number 1634 * @flags: fence related flags 1635 * 1636 * Write enc a fence and a trap command to the ring. 1637 */ 1638 static void vcn_v1_0_enc_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, 1639 u64 seq, unsigned flags) 1640 { 1641 WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT); 1642 1643 amdgpu_ring_write(ring, VCN_ENC_CMD_FENCE); 1644 amdgpu_ring_write(ring, addr); 1645 amdgpu_ring_write(ring, upper_32_bits(addr)); 1646 amdgpu_ring_write(ring, seq); 1647 amdgpu_ring_write(ring, VCN_ENC_CMD_TRAP); 1648 } 1649 1650 static void vcn_v1_0_enc_ring_insert_end(struct amdgpu_ring *ring) 1651 { 1652 amdgpu_ring_write(ring, VCN_ENC_CMD_END); 1653 } 1654 1655 /** 1656 * vcn_v1_0_enc_ring_emit_ib - enc execute indirect buffer 1657 * 1658 * @ring: amdgpu_ring pointer 1659 * @job: job to retrive vmid from 1660 * @ib: indirect buffer to execute 1661 * @flags: unused 1662 * 1663 * Write enc ring commands to execute the indirect buffer 1664 */ 1665 static void vcn_v1_0_enc_ring_emit_ib(struct amdgpu_ring *ring, 1666 struct amdgpu_job *job, 1667 struct amdgpu_ib *ib, 1668 uint32_t flags) 1669 { 1670 unsigned vmid = AMDGPU_JOB_GET_VMID(job); 1671 1672 amdgpu_ring_write(ring, VCN_ENC_CMD_IB); 1673 amdgpu_ring_write(ring, vmid); 1674 amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr)); 1675 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr)); 1676 amdgpu_ring_write(ring, ib->length_dw); 1677 } 1678 1679 static void vcn_v1_0_enc_ring_emit_reg_wait(struct amdgpu_ring *ring, 1680 uint32_t reg, uint32_t val, 1681 uint32_t mask) 1682 { 1683 amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WAIT); 1684 amdgpu_ring_write(ring, reg << 2); 1685 amdgpu_ring_write(ring, mask); 1686 amdgpu_ring_write(ring, val); 1687 } 1688 1689 static void vcn_v1_0_enc_ring_emit_vm_flush(struct amdgpu_ring *ring, 1690 unsigned int vmid, uint64_t pd_addr) 1691 { 1692 struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub]; 1693 1694 pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr); 1695 1696 /* wait for reg writes */ 1697 vcn_v1_0_enc_ring_emit_reg_wait(ring, hub->ctx0_ptb_addr_lo32 + 1698 vmid * hub->ctx_addr_distance, 1699 lower_32_bits(pd_addr), 0xffffffff); 1700 } 1701 1702 static void vcn_v1_0_enc_ring_emit_wreg(struct amdgpu_ring *ring, 1703 uint32_t reg, uint32_t val) 1704 { 1705 amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WRITE); 1706 amdgpu_ring_write(ring, reg << 2); 1707 amdgpu_ring_write(ring, val); 1708 } 1709 1710 static int vcn_v1_0_set_interrupt_state(struct amdgpu_device *adev, 1711 struct amdgpu_irq_src *source, 1712 unsigned type, 1713 enum amdgpu_interrupt_state state) 1714 { 1715 return 0; 1716 } 1717 1718 static int vcn_v1_0_process_interrupt(struct amdgpu_device *adev, 1719 struct amdgpu_irq_src *source, 1720 struct amdgpu_iv_entry *entry) 1721 { 1722 DRM_DEBUG("IH: VCN TRAP\n"); 1723 1724 switch (entry->src_id) { 1725 case 124: 1726 amdgpu_fence_process(&adev->vcn.inst->ring_dec); 1727 break; 1728 case 119: 1729 amdgpu_fence_process(&adev->vcn.inst->ring_enc[0]); 1730 break; 1731 case 120: 1732 amdgpu_fence_process(&adev->vcn.inst->ring_enc[1]); 1733 break; 1734 default: 1735 DRM_ERROR("Unhandled interrupt: %d %d\n", 1736 entry->src_id, entry->src_data[0]); 1737 break; 1738 } 1739 1740 return 0; 1741 } 1742 1743 static void vcn_v1_0_dec_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) 1744 { 1745 struct amdgpu_device *adev = ring->adev; 1746 int i; 1747 1748 WARN_ON(ring->wptr % 2 || count % 2); 1749 1750 for (i = 0; i < count / 2; i++) { 1751 amdgpu_ring_write(ring, PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_NO_OP), 0)); 1752 amdgpu_ring_write(ring, 0); 1753 } 1754 } 1755 1756 static int vcn_v1_0_set_powergating_state(void *handle, 1757 enum amd_powergating_state state) 1758 { 1759 /* This doesn't actually powergate the VCN block. 1760 * That's done in the dpm code via the SMC. This 1761 * just re-inits the block as necessary. The actual 1762 * gating still happens in the dpm code. We should 1763 * revisit this when there is a cleaner line between 1764 * the smc and the hw blocks 1765 */ 1766 int ret; 1767 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1768 1769 if(state == adev->vcn.cur_state) 1770 return 0; 1771 1772 if (state == AMD_PG_STATE_GATE) 1773 ret = vcn_v1_0_stop(adev); 1774 else 1775 ret = vcn_v1_0_start(adev); 1776 1777 if(!ret) 1778 adev->vcn.cur_state = state; 1779 return ret; 1780 } 1781 1782 static void vcn_v1_0_idle_work_handler(struct work_struct *work) 1783 { 1784 struct amdgpu_device *adev = 1785 container_of(work, struct amdgpu_device, vcn.idle_work.work); 1786 unsigned int fences = 0, i; 1787 1788 for (i = 0; i < adev->vcn.num_enc_rings; ++i) 1789 fences += amdgpu_fence_count_emitted(&adev->vcn.inst->ring_enc[i]); 1790 1791 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) { 1792 struct dpg_pause_state new_state; 1793 1794 if (fences) 1795 new_state.fw_based = VCN_DPG_STATE__PAUSE; 1796 else 1797 new_state.fw_based = VCN_DPG_STATE__UNPAUSE; 1798 1799 if (amdgpu_fence_count_emitted(&adev->jpeg.inst->ring_dec)) 1800 new_state.jpeg = VCN_DPG_STATE__PAUSE; 1801 else 1802 new_state.jpeg = VCN_DPG_STATE__UNPAUSE; 1803 1804 adev->vcn.pause_dpg_mode(adev, 0, &new_state); 1805 } 1806 1807 fences += amdgpu_fence_count_emitted(&adev->jpeg.inst->ring_dec); 1808 fences += amdgpu_fence_count_emitted(&adev->vcn.inst->ring_dec); 1809 1810 if (fences == 0) { 1811 amdgpu_gfx_off_ctrl(adev, true); 1812 if (adev->pm.dpm_enabled) 1813 amdgpu_dpm_enable_uvd(adev, false); 1814 else 1815 amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN, 1816 AMD_PG_STATE_GATE); 1817 } else { 1818 schedule_delayed_work(&adev->vcn.idle_work, VCN_IDLE_TIMEOUT); 1819 } 1820 } 1821 1822 static void vcn_v1_0_ring_begin_use(struct amdgpu_ring *ring) 1823 { 1824 struct amdgpu_device *adev = ring->adev; 1825 bool set_clocks = !cancel_delayed_work_sync(&adev->vcn.idle_work); 1826 1827 mutex_lock(&adev->vcn.vcn1_jpeg1_workaround); 1828 1829 if (amdgpu_fence_wait_empty(&ring->adev->jpeg.inst->ring_dec)) 1830 DRM_ERROR("VCN dec: jpeg dec ring may not be empty\n"); 1831 1832 vcn_v1_0_set_pg_for_begin_use(ring, set_clocks); 1833 1834 } 1835 1836 void vcn_v1_0_set_pg_for_begin_use(struct amdgpu_ring *ring, bool set_clocks) 1837 { 1838 struct amdgpu_device *adev = ring->adev; 1839 1840 if (set_clocks) { 1841 amdgpu_gfx_off_ctrl(adev, false); 1842 if (adev->pm.dpm_enabled) 1843 amdgpu_dpm_enable_uvd(adev, true); 1844 else 1845 amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN, 1846 AMD_PG_STATE_UNGATE); 1847 } 1848 1849 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) { 1850 struct dpg_pause_state new_state; 1851 unsigned int fences = 0, i; 1852 1853 for (i = 0; i < adev->vcn.num_enc_rings; ++i) 1854 fences += amdgpu_fence_count_emitted(&adev->vcn.inst->ring_enc[i]); 1855 1856 if (fences) 1857 new_state.fw_based = VCN_DPG_STATE__PAUSE; 1858 else 1859 new_state.fw_based = VCN_DPG_STATE__UNPAUSE; 1860 1861 if (amdgpu_fence_count_emitted(&adev->jpeg.inst->ring_dec)) 1862 new_state.jpeg = VCN_DPG_STATE__PAUSE; 1863 else 1864 new_state.jpeg = VCN_DPG_STATE__UNPAUSE; 1865 1866 if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC) 1867 new_state.fw_based = VCN_DPG_STATE__PAUSE; 1868 else if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_JPEG) 1869 new_state.jpeg = VCN_DPG_STATE__PAUSE; 1870 1871 adev->vcn.pause_dpg_mode(adev, 0, &new_state); 1872 } 1873 } 1874 1875 void vcn_v1_0_ring_end_use(struct amdgpu_ring *ring) 1876 { 1877 schedule_delayed_work(&ring->adev->vcn.idle_work, VCN_IDLE_TIMEOUT); 1878 mutex_unlock(&ring->adev->vcn.vcn1_jpeg1_workaround); 1879 } 1880 1881 static const struct amd_ip_funcs vcn_v1_0_ip_funcs = { 1882 .name = "vcn_v1_0", 1883 .early_init = vcn_v1_0_early_init, 1884 .late_init = NULL, 1885 .sw_init = vcn_v1_0_sw_init, 1886 .sw_fini = vcn_v1_0_sw_fini, 1887 .hw_init = vcn_v1_0_hw_init, 1888 .hw_fini = vcn_v1_0_hw_fini, 1889 .suspend = vcn_v1_0_suspend, 1890 .resume = vcn_v1_0_resume, 1891 .is_idle = vcn_v1_0_is_idle, 1892 .wait_for_idle = vcn_v1_0_wait_for_idle, 1893 .check_soft_reset = NULL /* vcn_v1_0_check_soft_reset */, 1894 .pre_soft_reset = NULL /* vcn_v1_0_pre_soft_reset */, 1895 .soft_reset = NULL /* vcn_v1_0_soft_reset */, 1896 .post_soft_reset = NULL /* vcn_v1_0_post_soft_reset */, 1897 .set_clockgating_state = vcn_v1_0_set_clockgating_state, 1898 .set_powergating_state = vcn_v1_0_set_powergating_state, 1899 }; 1900 1901 static const struct amdgpu_ring_funcs vcn_v1_0_dec_ring_vm_funcs = { 1902 .type = AMDGPU_RING_TYPE_VCN_DEC, 1903 .align_mask = 0xf, 1904 .support_64bit_ptrs = false, 1905 .no_user_fence = true, 1906 .vmhub = AMDGPU_MMHUB_0, 1907 .get_rptr = vcn_v1_0_dec_ring_get_rptr, 1908 .get_wptr = vcn_v1_0_dec_ring_get_wptr, 1909 .set_wptr = vcn_v1_0_dec_ring_set_wptr, 1910 .emit_frame_size = 1911 6 + 6 + /* hdp invalidate / flush */ 1912 SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 + 1913 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 + 1914 8 + /* vcn_v1_0_dec_ring_emit_vm_flush */ 1915 14 + 14 + /* vcn_v1_0_dec_ring_emit_fence x2 vm fence */ 1916 6, 1917 .emit_ib_size = 8, /* vcn_v1_0_dec_ring_emit_ib */ 1918 .emit_ib = vcn_v1_0_dec_ring_emit_ib, 1919 .emit_fence = vcn_v1_0_dec_ring_emit_fence, 1920 .emit_vm_flush = vcn_v1_0_dec_ring_emit_vm_flush, 1921 .test_ring = amdgpu_vcn_dec_ring_test_ring, 1922 .test_ib = amdgpu_vcn_dec_ring_test_ib, 1923 .insert_nop = vcn_v1_0_dec_ring_insert_nop, 1924 .insert_start = vcn_v1_0_dec_ring_insert_start, 1925 .insert_end = vcn_v1_0_dec_ring_insert_end, 1926 .pad_ib = amdgpu_ring_generic_pad_ib, 1927 .begin_use = vcn_v1_0_ring_begin_use, 1928 .end_use = vcn_v1_0_ring_end_use, 1929 .emit_wreg = vcn_v1_0_dec_ring_emit_wreg, 1930 .emit_reg_wait = vcn_v1_0_dec_ring_emit_reg_wait, 1931 .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper, 1932 }; 1933 1934 static const struct amdgpu_ring_funcs vcn_v1_0_enc_ring_vm_funcs = { 1935 .type = AMDGPU_RING_TYPE_VCN_ENC, 1936 .align_mask = 0x3f, 1937 .nop = VCN_ENC_CMD_NO_OP, 1938 .support_64bit_ptrs = false, 1939 .no_user_fence = true, 1940 .vmhub = AMDGPU_MMHUB_0, 1941 .get_rptr = vcn_v1_0_enc_ring_get_rptr, 1942 .get_wptr = vcn_v1_0_enc_ring_get_wptr, 1943 .set_wptr = vcn_v1_0_enc_ring_set_wptr, 1944 .emit_frame_size = 1945 SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 + 1946 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 + 1947 4 + /* vcn_v1_0_enc_ring_emit_vm_flush */ 1948 5 + 5 + /* vcn_v1_0_enc_ring_emit_fence x2 vm fence */ 1949 1, /* vcn_v1_0_enc_ring_insert_end */ 1950 .emit_ib_size = 5, /* vcn_v1_0_enc_ring_emit_ib */ 1951 .emit_ib = vcn_v1_0_enc_ring_emit_ib, 1952 .emit_fence = vcn_v1_0_enc_ring_emit_fence, 1953 .emit_vm_flush = vcn_v1_0_enc_ring_emit_vm_flush, 1954 .test_ring = amdgpu_vcn_enc_ring_test_ring, 1955 .test_ib = amdgpu_vcn_enc_ring_test_ib, 1956 .insert_nop = amdgpu_ring_insert_nop, 1957 .insert_end = vcn_v1_0_enc_ring_insert_end, 1958 .pad_ib = amdgpu_ring_generic_pad_ib, 1959 .begin_use = vcn_v1_0_ring_begin_use, 1960 .end_use = vcn_v1_0_ring_end_use, 1961 .emit_wreg = vcn_v1_0_enc_ring_emit_wreg, 1962 .emit_reg_wait = vcn_v1_0_enc_ring_emit_reg_wait, 1963 .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper, 1964 }; 1965 1966 static void vcn_v1_0_set_dec_ring_funcs(struct amdgpu_device *adev) 1967 { 1968 adev->vcn.inst->ring_dec.funcs = &vcn_v1_0_dec_ring_vm_funcs; 1969 DRM_INFO("VCN decode is enabled in VM mode\n"); 1970 } 1971 1972 static void vcn_v1_0_set_enc_ring_funcs(struct amdgpu_device *adev) 1973 { 1974 int i; 1975 1976 for (i = 0; i < adev->vcn.num_enc_rings; ++i) 1977 adev->vcn.inst->ring_enc[i].funcs = &vcn_v1_0_enc_ring_vm_funcs; 1978 1979 DRM_INFO("VCN encode is enabled in VM mode\n"); 1980 } 1981 1982 static const struct amdgpu_irq_src_funcs vcn_v1_0_irq_funcs = { 1983 .set = vcn_v1_0_set_interrupt_state, 1984 .process = vcn_v1_0_process_interrupt, 1985 }; 1986 1987 static void vcn_v1_0_set_irq_funcs(struct amdgpu_device *adev) 1988 { 1989 adev->vcn.inst->irq.num_types = adev->vcn.num_enc_rings + 2; 1990 adev->vcn.inst->irq.funcs = &vcn_v1_0_irq_funcs; 1991 } 1992 1993 const struct amdgpu_ip_block_version vcn_v1_0_ip_block = 1994 { 1995 .type = AMD_IP_BLOCK_TYPE_VCN, 1996 .major = 1, 1997 .minor = 0, 1998 .rev = 0, 1999 .funcs = &vcn_v1_0_ip_funcs, 2000 }; 2001