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_cs.h" 28 #include "amdgpu_vcn.h" 29 #include "amdgpu_pm.h" 30 #include "soc15.h" 31 #include "soc15d.h" 32 #include "soc15_common.h" 33 34 #include "vcn/vcn_1_0_offset.h" 35 #include "vcn/vcn_1_0_sh_mask.h" 36 #include "mmhub/mmhub_9_1_offset.h" 37 #include "mmhub/mmhub_9_1_sh_mask.h" 38 39 #include "ivsrcid/vcn/irqsrcs_vcn_1_0.h" 40 #include "jpeg_v1_0.h" 41 #include "vcn_v1_0.h" 42 43 #define mmUVD_RBC_XX_IB_REG_CHECK_1_0 0x05ab 44 #define mmUVD_RBC_XX_IB_REG_CHECK_1_0_BASE_IDX 1 45 #define mmUVD_REG_XX_MASK_1_0 0x05ac 46 #define mmUVD_REG_XX_MASK_1_0_BASE_IDX 1 47 48 static const struct amdgpu_hwip_reg_entry vcn_reg_list_1_0[] = { 49 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_POWER_STATUS), 50 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_STATUS), 51 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_CONTEXT_ID), 52 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_CONTEXT_ID2), 53 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_GPCOM_VCPU_DATA0), 54 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_GPCOM_VCPU_DATA1), 55 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_GPCOM_VCPU_CMD), 56 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_HI), 57 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_LO), 58 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_HI2), 59 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_LO2), 60 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_HI3), 61 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_LO3), 62 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_HI4), 63 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_LO4), 64 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_RPTR), 65 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_WPTR), 66 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_RPTR2), 67 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_WPTR2), 68 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_RPTR3), 69 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_WPTR3), 70 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_RPTR4), 71 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_WPTR4), 72 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_SIZE), 73 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_SIZE2), 74 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_SIZE3), 75 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_SIZE4), 76 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_PGFSM_CONFIG), 77 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_PGFSM_STATUS), 78 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_DPG_LMA_CTL), 79 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_DPG_LMA_DATA), 80 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_DPG_LMA_MASK), 81 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_DPG_PAUSE) 82 }; 83 84 static int vcn_v1_0_stop(struct amdgpu_vcn_inst *vinst); 85 static void vcn_v1_0_set_dec_ring_funcs(struct amdgpu_device *adev); 86 static void vcn_v1_0_set_enc_ring_funcs(struct amdgpu_device *adev); 87 static void vcn_v1_0_set_irq_funcs(struct amdgpu_device *adev); 88 static int vcn_v1_0_set_pg_state(struct amdgpu_vcn_inst *vinst, 89 enum amd_powergating_state state); 90 static int vcn_v1_0_pause_dpg_mode(struct amdgpu_vcn_inst *vinst, 91 struct dpg_pause_state *new_state); 92 93 static void vcn_v1_0_idle_work_handler(struct work_struct *work); 94 static void vcn_v1_0_ring_begin_use(struct amdgpu_ring *ring); 95 96 /** 97 * vcn_v1_0_early_init - set function pointers and load microcode 98 * 99 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance. 100 * 101 * Set ring and irq function pointers 102 * Load microcode from filesystem 103 */ 104 static int vcn_v1_0_early_init(struct amdgpu_ip_block *ip_block) 105 { 106 struct amdgpu_device *adev = ip_block->adev; 107 108 adev->vcn.inst[0].num_enc_rings = 2; 109 adev->vcn.inst[0].set_pg_state = vcn_v1_0_set_pg_state; 110 111 vcn_v1_0_set_dec_ring_funcs(adev); 112 vcn_v1_0_set_enc_ring_funcs(adev); 113 vcn_v1_0_set_irq_funcs(adev); 114 115 jpeg_v1_0_early_init(ip_block); 116 117 return amdgpu_vcn_early_init(adev, 0); 118 } 119 120 /** 121 * vcn_v1_0_sw_init - sw init for VCN block 122 * 123 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance. 124 * 125 * Load firmware and sw initialization 126 */ 127 static int vcn_v1_0_sw_init(struct amdgpu_ip_block *ip_block) 128 { 129 struct amdgpu_ring *ring; 130 int i, r; 131 uint32_t reg_count = ARRAY_SIZE(vcn_reg_list_1_0); 132 uint32_t *ptr; 133 struct amdgpu_device *adev = ip_block->adev; 134 135 /* VCN DEC TRAP */ 136 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_VCN, 137 VCN_1_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT, &adev->vcn.inst->irq); 138 if (r) 139 return r; 140 141 /* VCN ENC TRAP */ 142 for (i = 0; i < adev->vcn.inst[0].num_enc_rings; ++i) { 143 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_VCN, i + VCN_1_0__SRCID__UVD_ENC_GENERAL_PURPOSE, 144 &adev->vcn.inst->irq); 145 if (r) 146 return r; 147 } 148 149 r = amdgpu_vcn_sw_init(adev, 0); 150 if (r) 151 return r; 152 153 /* Override the work func */ 154 adev->vcn.inst[0].idle_work.work.func = vcn_v1_0_idle_work_handler; 155 156 amdgpu_vcn_setup_ucode(adev, 0); 157 158 r = amdgpu_vcn_resume(adev, 0); 159 if (r) 160 return r; 161 162 ring = &adev->vcn.inst->ring_dec; 163 ring->vm_hub = AMDGPU_MMHUB0(0); 164 sprintf(ring->name, "vcn_dec"); 165 r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst->irq, 0, 166 AMDGPU_RING_PRIO_DEFAULT, NULL); 167 if (r) 168 return r; 169 170 adev->vcn.inst[0].internal.scratch9 = adev->vcn.inst->external.scratch9 = 171 SOC15_REG_OFFSET(UVD, 0, mmUVD_SCRATCH9); 172 adev->vcn.inst[0].internal.data0 = adev->vcn.inst->external.data0 = 173 SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0); 174 adev->vcn.inst[0].internal.data1 = adev->vcn.inst->external.data1 = 175 SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1); 176 adev->vcn.inst[0].internal.cmd = adev->vcn.inst->external.cmd = 177 SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD); 178 adev->vcn.inst[0].internal.nop = adev->vcn.inst->external.nop = 179 SOC15_REG_OFFSET(UVD, 0, mmUVD_NO_OP); 180 181 for (i = 0; i < adev->vcn.inst[0].num_enc_rings; ++i) { 182 enum amdgpu_ring_priority_level hw_prio = amdgpu_vcn_get_enc_ring_prio(i); 183 184 ring = &adev->vcn.inst->ring_enc[i]; 185 ring->vm_hub = AMDGPU_MMHUB0(0); 186 sprintf(ring->name, "vcn_enc%d", i); 187 r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst->irq, 0, 188 hw_prio, NULL); 189 if (r) 190 return r; 191 } 192 193 adev->vcn.inst[0].pause_dpg_mode = vcn_v1_0_pause_dpg_mode; 194 195 if (amdgpu_vcnfw_log) { 196 struct amdgpu_fw_shared *fw_shared = adev->vcn.inst->fw_shared.cpu_addr; 197 198 fw_shared->present_flag_0 = 0; 199 amdgpu_vcn_fwlog_init(adev->vcn.inst); 200 } 201 202 r = jpeg_v1_0_sw_init(ip_block); 203 204 /* Allocate memory for VCN IP Dump buffer */ 205 ptr = kcalloc(adev->vcn.num_vcn_inst * reg_count, sizeof(uint32_t), GFP_KERNEL); 206 if (!ptr) { 207 DRM_ERROR("Failed to allocate memory for VCN IP Dump\n"); 208 adev->vcn.ip_dump = NULL; 209 } else { 210 adev->vcn.ip_dump = ptr; 211 } 212 return r; 213 } 214 215 /** 216 * vcn_v1_0_sw_fini - sw fini for VCN block 217 * 218 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance. 219 * 220 * VCN suspend and free up sw allocation 221 */ 222 static int vcn_v1_0_sw_fini(struct amdgpu_ip_block *ip_block) 223 { 224 int r; 225 struct amdgpu_device *adev = ip_block->adev; 226 227 r = amdgpu_vcn_suspend(adev, 0); 228 if (r) 229 return r; 230 231 jpeg_v1_0_sw_fini(ip_block); 232 233 amdgpu_vcn_sw_fini(adev, 0); 234 235 kfree(adev->vcn.ip_dump); 236 237 return 0; 238 } 239 240 /** 241 * vcn_v1_0_hw_init - start and test VCN block 242 * 243 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance. 244 * 245 * Initialize the hardware, boot up the VCPU and do some testing 246 */ 247 static int vcn_v1_0_hw_init(struct amdgpu_ip_block *ip_block) 248 { 249 struct amdgpu_device *adev = ip_block->adev; 250 struct amdgpu_ring *ring = &adev->vcn.inst->ring_dec; 251 int i, r; 252 253 r = amdgpu_ring_test_helper(ring); 254 if (r) 255 return r; 256 257 for (i = 0; i < adev->vcn.inst[0].num_enc_rings; ++i) { 258 ring = &adev->vcn.inst->ring_enc[i]; 259 r = amdgpu_ring_test_helper(ring); 260 if (r) 261 return r; 262 } 263 264 ring = adev->jpeg.inst->ring_dec; 265 r = amdgpu_ring_test_helper(ring); 266 267 return r; 268 } 269 270 /** 271 * vcn_v1_0_hw_fini - stop the hardware block 272 * 273 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance. 274 * 275 * Stop the VCN block, mark ring as not ready any more 276 */ 277 static int vcn_v1_0_hw_fini(struct amdgpu_ip_block *ip_block) 278 { 279 struct amdgpu_device *adev = ip_block->adev; 280 struct amdgpu_vcn_inst *vinst = adev->vcn.inst; 281 282 cancel_delayed_work_sync(&vinst->idle_work); 283 284 if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) || 285 (vinst->cur_state != AMD_PG_STATE_GATE && 286 RREG32_SOC15(VCN, 0, mmUVD_STATUS))) { 287 vinst->set_pg_state(vinst, AMD_PG_STATE_GATE); 288 } 289 290 return 0; 291 } 292 293 /** 294 * vcn_v1_0_suspend - suspend VCN block 295 * 296 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance. 297 * 298 * HW fini and suspend VCN block 299 */ 300 static int vcn_v1_0_suspend(struct amdgpu_ip_block *ip_block) 301 { 302 int r; 303 struct amdgpu_device *adev = ip_block->adev; 304 bool idle_work_unexecuted; 305 306 idle_work_unexecuted = cancel_delayed_work_sync(&adev->vcn.inst[0].idle_work); 307 if (idle_work_unexecuted) { 308 if (adev->pm.dpm_enabled) 309 amdgpu_dpm_enable_vcn(adev, false, 0); 310 } 311 312 r = vcn_v1_0_hw_fini(ip_block); 313 if (r) 314 return r; 315 316 r = amdgpu_vcn_suspend(adev, 0); 317 318 return r; 319 } 320 321 /** 322 * vcn_v1_0_resume - resume VCN block 323 * 324 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance. 325 * 326 * Resume firmware and hw init VCN block 327 */ 328 static int vcn_v1_0_resume(struct amdgpu_ip_block *ip_block) 329 { 330 int r; 331 332 r = amdgpu_vcn_resume(ip_block->adev, 0); 333 if (r) 334 return r; 335 336 r = vcn_v1_0_hw_init(ip_block); 337 338 return r; 339 } 340 341 /** 342 * vcn_v1_0_mc_resume_spg_mode - memory controller programming 343 * 344 * @vinst: VCN instance 345 * 346 * Let the VCN memory controller know it's offsets 347 */ 348 static void vcn_v1_0_mc_resume_spg_mode(struct amdgpu_vcn_inst *vinst) 349 { 350 struct amdgpu_device *adev = vinst->adev; 351 uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.inst[0].fw->size + 4); 352 uint32_t offset; 353 354 /* cache window 0: fw */ 355 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) { 356 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW, 357 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_lo)); 358 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH, 359 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_hi)); 360 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0, 0); 361 offset = 0; 362 } else { 363 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW, 364 lower_32_bits(adev->vcn.inst->gpu_addr)); 365 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH, 366 upper_32_bits(adev->vcn.inst->gpu_addr)); 367 offset = size; 368 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0, 369 AMDGPU_UVD_FIRMWARE_OFFSET >> 3); 370 } 371 372 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE0, size); 373 374 /* cache window 1: stack */ 375 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW, 376 lower_32_bits(adev->vcn.inst->gpu_addr + offset)); 377 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH, 378 upper_32_bits(adev->vcn.inst->gpu_addr + offset)); 379 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET1, 0); 380 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE); 381 382 /* cache window 2: context */ 383 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW, 384 lower_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE)); 385 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH, 386 upper_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE)); 387 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET2, 0); 388 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE); 389 390 WREG32_SOC15(UVD, 0, mmUVD_UDEC_ADDR_CONFIG, 391 adev->gfx.config.gb_addr_config); 392 WREG32_SOC15(UVD, 0, mmUVD_UDEC_DB_ADDR_CONFIG, 393 adev->gfx.config.gb_addr_config); 394 WREG32_SOC15(UVD, 0, mmUVD_UDEC_DBW_ADDR_CONFIG, 395 adev->gfx.config.gb_addr_config); 396 WREG32_SOC15(UVD, 0, mmUVD_UDEC_DBW_UV_ADDR_CONFIG, 397 adev->gfx.config.gb_addr_config); 398 WREG32_SOC15(UVD, 0, mmUVD_MIF_CURR_ADDR_CONFIG, 399 adev->gfx.config.gb_addr_config); 400 WREG32_SOC15(UVD, 0, mmUVD_MIF_CURR_UV_ADDR_CONFIG, 401 adev->gfx.config.gb_addr_config); 402 WREG32_SOC15(UVD, 0, mmUVD_MIF_RECON1_ADDR_CONFIG, 403 adev->gfx.config.gb_addr_config); 404 WREG32_SOC15(UVD, 0, mmUVD_MIF_RECON1_UV_ADDR_CONFIG, 405 adev->gfx.config.gb_addr_config); 406 WREG32_SOC15(UVD, 0, mmUVD_MIF_REF_ADDR_CONFIG, 407 adev->gfx.config.gb_addr_config); 408 WREG32_SOC15(UVD, 0, mmUVD_MIF_REF_UV_ADDR_CONFIG, 409 adev->gfx.config.gb_addr_config); 410 WREG32_SOC15(UVD, 0, mmUVD_JPEG_ADDR_CONFIG, 411 adev->gfx.config.gb_addr_config); 412 WREG32_SOC15(UVD, 0, mmUVD_JPEG_UV_ADDR_CONFIG, 413 adev->gfx.config.gb_addr_config); 414 } 415 416 static void vcn_v1_0_mc_resume_dpg_mode(struct amdgpu_vcn_inst *vinst) 417 { 418 struct amdgpu_device *adev = vinst->adev; 419 uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.inst[0].fw->size + 4); 420 uint32_t offset; 421 422 /* cache window 0: fw */ 423 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) { 424 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW, 425 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_lo), 426 0xFFFFFFFF, 0); 427 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH, 428 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_hi), 429 0xFFFFFFFF, 0); 430 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0, 0, 431 0xFFFFFFFF, 0); 432 offset = 0; 433 } else { 434 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW, 435 lower_32_bits(adev->vcn.inst->gpu_addr), 0xFFFFFFFF, 0); 436 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH, 437 upper_32_bits(adev->vcn.inst->gpu_addr), 0xFFFFFFFF, 0); 438 offset = size; 439 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0, 440 AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0xFFFFFFFF, 0); 441 } 442 443 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_VCPU_CACHE_SIZE0, size, 0xFFFFFFFF, 0); 444 445 /* cache window 1: stack */ 446 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW, 447 lower_32_bits(adev->vcn.inst->gpu_addr + offset), 0xFFFFFFFF, 0); 448 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH, 449 upper_32_bits(adev->vcn.inst->gpu_addr + offset), 0xFFFFFFFF, 0); 450 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_VCPU_CACHE_OFFSET1, 0, 451 0xFFFFFFFF, 0); 452 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE, 453 0xFFFFFFFF, 0); 454 455 /* cache window 2: context */ 456 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW, 457 lower_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 458 0xFFFFFFFF, 0); 459 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH, 460 upper_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 461 0xFFFFFFFF, 0); 462 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_VCPU_CACHE_OFFSET2, 0, 0xFFFFFFFF, 0); 463 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE, 464 0xFFFFFFFF, 0); 465 466 /* VCN global tiling registers */ 467 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_UDEC_ADDR_CONFIG, 468 adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); 469 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_UDEC_DB_ADDR_CONFIG, 470 adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); 471 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_UDEC_DBW_ADDR_CONFIG, 472 adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); 473 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_UDEC_DBW_UV_ADDR_CONFIG, 474 adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); 475 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MIF_CURR_ADDR_CONFIG, 476 adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); 477 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MIF_CURR_UV_ADDR_CONFIG, 478 adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); 479 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MIF_RECON1_ADDR_CONFIG, 480 adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); 481 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MIF_RECON1_UV_ADDR_CONFIG, 482 adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); 483 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MIF_REF_ADDR_CONFIG, 484 adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); 485 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MIF_REF_UV_ADDR_CONFIG, 486 adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); 487 } 488 489 /** 490 * vcn_v1_0_disable_clock_gating - disable VCN clock gating 491 * 492 * @vinst: VCN instance 493 * 494 * Disable clock gating for VCN block 495 */ 496 static void vcn_v1_0_disable_clock_gating(struct amdgpu_vcn_inst *vinst) 497 { 498 struct amdgpu_device *adev = vinst->adev; 499 uint32_t data; 500 501 /* JPEG disable CGC */ 502 data = RREG32_SOC15(VCN, 0, mmJPEG_CGC_CTRL); 503 504 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG) 505 data |= 1 << JPEG_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 506 else 507 data &= ~JPEG_CGC_CTRL__DYN_CLOCK_MODE_MASK; 508 509 data |= 1 << JPEG_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT; 510 data |= 4 << JPEG_CGC_CTRL__CLK_OFF_DELAY__SHIFT; 511 WREG32_SOC15(VCN, 0, mmJPEG_CGC_CTRL, data); 512 513 data = RREG32_SOC15(VCN, 0, mmJPEG_CGC_GATE); 514 data &= ~(JPEG_CGC_GATE__JPEG_MASK | JPEG_CGC_GATE__JPEG2_MASK); 515 WREG32_SOC15(VCN, 0, mmJPEG_CGC_GATE, data); 516 517 /* UVD disable CGC */ 518 data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL); 519 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG) 520 data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 521 else 522 data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK; 523 524 data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT; 525 data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT; 526 WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data); 527 528 data = RREG32_SOC15(VCN, 0, mmUVD_CGC_GATE); 529 data &= ~(UVD_CGC_GATE__SYS_MASK 530 | UVD_CGC_GATE__UDEC_MASK 531 | UVD_CGC_GATE__MPEG2_MASK 532 | UVD_CGC_GATE__REGS_MASK 533 | UVD_CGC_GATE__RBC_MASK 534 | UVD_CGC_GATE__LMI_MC_MASK 535 | UVD_CGC_GATE__LMI_UMC_MASK 536 | UVD_CGC_GATE__IDCT_MASK 537 | UVD_CGC_GATE__MPRD_MASK 538 | UVD_CGC_GATE__MPC_MASK 539 | UVD_CGC_GATE__LBSI_MASK 540 | UVD_CGC_GATE__LRBBM_MASK 541 | UVD_CGC_GATE__UDEC_RE_MASK 542 | UVD_CGC_GATE__UDEC_CM_MASK 543 | UVD_CGC_GATE__UDEC_IT_MASK 544 | UVD_CGC_GATE__UDEC_DB_MASK 545 | UVD_CGC_GATE__UDEC_MP_MASK 546 | UVD_CGC_GATE__WCB_MASK 547 | UVD_CGC_GATE__VCPU_MASK 548 | UVD_CGC_GATE__SCPU_MASK); 549 WREG32_SOC15(VCN, 0, mmUVD_CGC_GATE, data); 550 551 data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL); 552 data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK 553 | UVD_CGC_CTRL__UDEC_CM_MODE_MASK 554 | UVD_CGC_CTRL__UDEC_IT_MODE_MASK 555 | UVD_CGC_CTRL__UDEC_DB_MODE_MASK 556 | UVD_CGC_CTRL__UDEC_MP_MODE_MASK 557 | UVD_CGC_CTRL__SYS_MODE_MASK 558 | UVD_CGC_CTRL__UDEC_MODE_MASK 559 | UVD_CGC_CTRL__MPEG2_MODE_MASK 560 | UVD_CGC_CTRL__REGS_MODE_MASK 561 | UVD_CGC_CTRL__RBC_MODE_MASK 562 | UVD_CGC_CTRL__LMI_MC_MODE_MASK 563 | UVD_CGC_CTRL__LMI_UMC_MODE_MASK 564 | UVD_CGC_CTRL__IDCT_MODE_MASK 565 | UVD_CGC_CTRL__MPRD_MODE_MASK 566 | UVD_CGC_CTRL__MPC_MODE_MASK 567 | UVD_CGC_CTRL__LBSI_MODE_MASK 568 | UVD_CGC_CTRL__LRBBM_MODE_MASK 569 | UVD_CGC_CTRL__WCB_MODE_MASK 570 | UVD_CGC_CTRL__VCPU_MODE_MASK 571 | UVD_CGC_CTRL__SCPU_MODE_MASK); 572 WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data); 573 574 /* turn on */ 575 data = RREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_GATE); 576 data |= (UVD_SUVD_CGC_GATE__SRE_MASK 577 | UVD_SUVD_CGC_GATE__SIT_MASK 578 | UVD_SUVD_CGC_GATE__SMP_MASK 579 | UVD_SUVD_CGC_GATE__SCM_MASK 580 | UVD_SUVD_CGC_GATE__SDB_MASK 581 | UVD_SUVD_CGC_GATE__SRE_H264_MASK 582 | UVD_SUVD_CGC_GATE__SRE_HEVC_MASK 583 | UVD_SUVD_CGC_GATE__SIT_H264_MASK 584 | UVD_SUVD_CGC_GATE__SIT_HEVC_MASK 585 | UVD_SUVD_CGC_GATE__SCM_H264_MASK 586 | UVD_SUVD_CGC_GATE__SCM_HEVC_MASK 587 | UVD_SUVD_CGC_GATE__SDB_H264_MASK 588 | UVD_SUVD_CGC_GATE__SDB_HEVC_MASK 589 | UVD_SUVD_CGC_GATE__SCLR_MASK 590 | UVD_SUVD_CGC_GATE__UVD_SC_MASK 591 | UVD_SUVD_CGC_GATE__ENT_MASK 592 | UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK 593 | UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK 594 | UVD_SUVD_CGC_GATE__SITE_MASK 595 | UVD_SUVD_CGC_GATE__SRE_VP9_MASK 596 | UVD_SUVD_CGC_GATE__SCM_VP9_MASK 597 | UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK 598 | UVD_SUVD_CGC_GATE__SDB_VP9_MASK 599 | UVD_SUVD_CGC_GATE__IME_HEVC_MASK); 600 WREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_GATE, data); 601 602 data = RREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL); 603 data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK 604 | UVD_SUVD_CGC_CTRL__SIT_MODE_MASK 605 | UVD_SUVD_CGC_CTRL__SMP_MODE_MASK 606 | UVD_SUVD_CGC_CTRL__SCM_MODE_MASK 607 | UVD_SUVD_CGC_CTRL__SDB_MODE_MASK 608 | UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK 609 | UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK 610 | UVD_SUVD_CGC_CTRL__ENT_MODE_MASK 611 | UVD_SUVD_CGC_CTRL__IME_MODE_MASK 612 | UVD_SUVD_CGC_CTRL__SITE_MODE_MASK); 613 WREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL, data); 614 } 615 616 /** 617 * vcn_v1_0_enable_clock_gating - enable VCN clock gating 618 * 619 * @vinst: Pointer to the VCN instance structure 620 * 621 * Enable clock gating for VCN block 622 */ 623 static void vcn_v1_0_enable_clock_gating(struct amdgpu_vcn_inst *vinst) 624 { 625 struct amdgpu_device *adev = vinst->adev; 626 uint32_t data = 0; 627 628 /* enable JPEG CGC */ 629 data = RREG32_SOC15(VCN, 0, mmJPEG_CGC_CTRL); 630 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG) 631 data |= 1 << JPEG_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 632 else 633 data |= 0 << JPEG_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 634 data |= 1 << JPEG_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT; 635 data |= 4 << JPEG_CGC_CTRL__CLK_OFF_DELAY__SHIFT; 636 WREG32_SOC15(VCN, 0, mmJPEG_CGC_CTRL, data); 637 638 data = RREG32_SOC15(VCN, 0, mmJPEG_CGC_GATE); 639 data |= (JPEG_CGC_GATE__JPEG_MASK | JPEG_CGC_GATE__JPEG2_MASK); 640 WREG32_SOC15(VCN, 0, mmJPEG_CGC_GATE, data); 641 642 /* enable UVD CGC */ 643 data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL); 644 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG) 645 data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 646 else 647 data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 648 data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT; 649 data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT; 650 WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data); 651 652 data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL); 653 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(VCN, 0, mmUVD_CGC_CTRL, data); 674 675 data = RREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL); 676 data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK 677 | UVD_SUVD_CGC_CTRL__SIT_MODE_MASK 678 | UVD_SUVD_CGC_CTRL__SMP_MODE_MASK 679 | UVD_SUVD_CGC_CTRL__SCM_MODE_MASK 680 | UVD_SUVD_CGC_CTRL__SDB_MODE_MASK 681 | UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK 682 | UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK 683 | UVD_SUVD_CGC_CTRL__ENT_MODE_MASK 684 | UVD_SUVD_CGC_CTRL__IME_MODE_MASK 685 | UVD_SUVD_CGC_CTRL__SITE_MODE_MASK); 686 WREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL, data); 687 } 688 689 static void vcn_v1_0_clock_gating_dpg_mode(struct amdgpu_vcn_inst *vinst, 690 uint8_t sram_sel) 691 { 692 struct amdgpu_device *adev = vinst->adev; 693 uint32_t reg_data = 0; 694 695 /* disable JPEG CGC */ 696 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG) 697 reg_data = 1 << JPEG_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 698 else 699 reg_data = 0 << JPEG_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 700 reg_data |= 1 << JPEG_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT; 701 reg_data |= 4 << JPEG_CGC_CTRL__CLK_OFF_DELAY__SHIFT; 702 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmJPEG_CGC_CTRL, reg_data, 0xFFFFFFFF, sram_sel); 703 704 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmJPEG_CGC_GATE, 0, 0xFFFFFFFF, sram_sel); 705 706 /* enable sw clock gating control */ 707 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG) 708 reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 709 else 710 reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; 711 reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT; 712 reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT; 713 reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK | 714 UVD_CGC_CTRL__UDEC_CM_MODE_MASK | 715 UVD_CGC_CTRL__UDEC_IT_MODE_MASK | 716 UVD_CGC_CTRL__UDEC_DB_MODE_MASK | 717 UVD_CGC_CTRL__UDEC_MP_MODE_MASK | 718 UVD_CGC_CTRL__SYS_MODE_MASK | 719 UVD_CGC_CTRL__UDEC_MODE_MASK | 720 UVD_CGC_CTRL__MPEG2_MODE_MASK | 721 UVD_CGC_CTRL__REGS_MODE_MASK | 722 UVD_CGC_CTRL__RBC_MODE_MASK | 723 UVD_CGC_CTRL__LMI_MC_MODE_MASK | 724 UVD_CGC_CTRL__LMI_UMC_MODE_MASK | 725 UVD_CGC_CTRL__IDCT_MODE_MASK | 726 UVD_CGC_CTRL__MPRD_MODE_MASK | 727 UVD_CGC_CTRL__MPC_MODE_MASK | 728 UVD_CGC_CTRL__LBSI_MODE_MASK | 729 UVD_CGC_CTRL__LRBBM_MODE_MASK | 730 UVD_CGC_CTRL__WCB_MODE_MASK | 731 UVD_CGC_CTRL__VCPU_MODE_MASK | 732 UVD_CGC_CTRL__SCPU_MODE_MASK); 733 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_CGC_CTRL, reg_data, 0xFFFFFFFF, sram_sel); 734 735 /* turn off clock gating */ 736 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_CGC_GATE, 0, 0xFFFFFFFF, sram_sel); 737 738 /* turn on SUVD clock gating */ 739 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_SUVD_CGC_GATE, 1, 0xFFFFFFFF, sram_sel); 740 741 /* turn on sw mode in UVD_SUVD_CGC_CTRL */ 742 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_SUVD_CGC_CTRL, 0, 0xFFFFFFFF, sram_sel); 743 } 744 745 static void vcn_1_0_disable_static_power_gating(struct amdgpu_vcn_inst *vinst) 746 { 747 struct amdgpu_device *adev = vinst->adev; 748 uint32_t data = 0; 749 750 if (adev->pg_flags & AMD_PG_SUPPORT_VCN) { 751 data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT 752 | 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT 753 | 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT 754 | 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT 755 | 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT 756 | 2 << UVD_PGFSM_CONFIG__UVDIL_PWR_CONFIG__SHIFT 757 | 2 << UVD_PGFSM_CONFIG__UVDIR_PWR_CONFIG__SHIFT 758 | 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT 759 | 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT 760 | 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT 761 | 2 << UVD_PGFSM_CONFIG__UVDW_PWR_CONFIG__SHIFT); 762 763 WREG32_SOC15(VCN, 0, mmUVD_PGFSM_CONFIG, data); 764 SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_PGFSM_STATUS, UVD_PGFSM_STATUS__UVDM_UVDU_PWR_ON, 0xFFFFFF); 765 } else { 766 data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT 767 | 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT 768 | 1 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT 769 | 1 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT 770 | 1 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT 771 | 1 << UVD_PGFSM_CONFIG__UVDIL_PWR_CONFIG__SHIFT 772 | 1 << UVD_PGFSM_CONFIG__UVDIR_PWR_CONFIG__SHIFT 773 | 1 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT 774 | 1 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT 775 | 1 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT 776 | 1 << UVD_PGFSM_CONFIG__UVDW_PWR_CONFIG__SHIFT); 777 WREG32_SOC15(VCN, 0, mmUVD_PGFSM_CONFIG, data); 778 SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_PGFSM_STATUS, 0, 0xFFFFFFFF); 779 } 780 781 /* polling UVD_PGFSM_STATUS to confirm UVDM_PWR_STATUS , UVDU_PWR_STATUS are 0 (power on) */ 782 783 data = RREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS); 784 data &= ~0x103; 785 if (adev->pg_flags & AMD_PG_SUPPORT_VCN) 786 data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON | UVD_POWER_STATUS__UVD_PG_EN_MASK; 787 788 WREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS, data); 789 } 790 791 static void vcn_1_0_enable_static_power_gating(struct amdgpu_vcn_inst *vinst) 792 { 793 struct amdgpu_device *adev = vinst->adev; 794 uint32_t data = 0; 795 796 if (adev->pg_flags & AMD_PG_SUPPORT_VCN) { 797 /* Before power off, this indicator has to be turned on */ 798 data = RREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS); 799 data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK; 800 data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF; 801 WREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS, data); 802 803 804 data = (2 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT 805 | 2 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT 806 | 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT 807 | 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT 808 | 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT 809 | 2 << UVD_PGFSM_CONFIG__UVDIL_PWR_CONFIG__SHIFT 810 | 2 << UVD_PGFSM_CONFIG__UVDIR_PWR_CONFIG__SHIFT 811 | 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT 812 | 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT 813 | 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT 814 | 2 << UVD_PGFSM_CONFIG__UVDW_PWR_CONFIG__SHIFT); 815 816 WREG32_SOC15(VCN, 0, mmUVD_PGFSM_CONFIG, data); 817 818 data = (2 << UVD_PGFSM_STATUS__UVDM_PWR_STATUS__SHIFT 819 | 2 << UVD_PGFSM_STATUS__UVDU_PWR_STATUS__SHIFT 820 | 2 << UVD_PGFSM_STATUS__UVDF_PWR_STATUS__SHIFT 821 | 2 << UVD_PGFSM_STATUS__UVDC_PWR_STATUS__SHIFT 822 | 2 << UVD_PGFSM_STATUS__UVDB_PWR_STATUS__SHIFT 823 | 2 << UVD_PGFSM_STATUS__UVDIL_PWR_STATUS__SHIFT 824 | 2 << UVD_PGFSM_STATUS__UVDIR_PWR_STATUS__SHIFT 825 | 2 << UVD_PGFSM_STATUS__UVDTD_PWR_STATUS__SHIFT 826 | 2 << UVD_PGFSM_STATUS__UVDTE_PWR_STATUS__SHIFT 827 | 2 << UVD_PGFSM_STATUS__UVDE_PWR_STATUS__SHIFT 828 | 2 << UVD_PGFSM_STATUS__UVDW_PWR_STATUS__SHIFT); 829 SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_PGFSM_STATUS, data, 0xFFFFFFFF); 830 } 831 } 832 833 /** 834 * vcn_v1_0_start_spg_mode - start VCN block 835 * 836 * @vinst: VCN instance 837 * 838 * Setup and start the VCN block 839 */ 840 static int vcn_v1_0_start_spg_mode(struct amdgpu_vcn_inst *vinst) 841 { 842 struct amdgpu_device *adev = vinst->adev; 843 struct amdgpu_ring *ring = &adev->vcn.inst->ring_dec; 844 uint32_t rb_bufsz, tmp; 845 uint32_t lmi_swap_cntl; 846 int i, j, r; 847 848 /* disable byte swapping */ 849 lmi_swap_cntl = 0; 850 851 vcn_1_0_disable_static_power_gating(vinst); 852 853 tmp = RREG32_SOC15(UVD, 0, mmUVD_STATUS) | UVD_STATUS__UVD_BUSY; 854 WREG32_SOC15(UVD, 0, mmUVD_STATUS, tmp); 855 856 /* disable clock gating */ 857 vcn_v1_0_disable_clock_gating(vinst); 858 859 /* disable interupt */ 860 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_MASTINT_EN), 0, 861 ~UVD_MASTINT_EN__VCPU_EN_MASK); 862 863 /* initialize VCN memory controller */ 864 tmp = RREG32_SOC15(UVD, 0, mmUVD_LMI_CTRL); 865 WREG32_SOC15(UVD, 0, mmUVD_LMI_CTRL, tmp | 866 UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK | 867 UVD_LMI_CTRL__MASK_MC_URGENT_MASK | 868 UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK | 869 UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK); 870 871 #ifdef __BIG_ENDIAN 872 /* swap (8 in 32) RB and IB */ 873 lmi_swap_cntl = 0xa; 874 #endif 875 WREG32_SOC15(UVD, 0, mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl); 876 877 tmp = RREG32_SOC15(UVD, 0, mmUVD_MPC_CNTL); 878 tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK; 879 tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT; 880 WREG32_SOC15(UVD, 0, mmUVD_MPC_CNTL, tmp); 881 882 WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUXA0, 883 ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) | 884 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) | 885 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) | 886 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT))); 887 888 WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUXB0, 889 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) | 890 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) | 891 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) | 892 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT))); 893 894 WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUX, 895 ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) | 896 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) | 897 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT))); 898 899 vcn_v1_0_mc_resume_spg_mode(vinst); 900 901 WREG32_SOC15(UVD, 0, mmUVD_REG_XX_MASK_1_0, 0x10); 902 WREG32_SOC15(UVD, 0, mmUVD_RBC_XX_IB_REG_CHECK_1_0, 903 RREG32_SOC15(UVD, 0, mmUVD_RBC_XX_IB_REG_CHECK_1_0) | 0x3); 904 905 /* enable VCPU clock */ 906 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CNTL, UVD_VCPU_CNTL__CLK_EN_MASK); 907 908 /* boot up the VCPU */ 909 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 0, 910 ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK); 911 912 /* enable UMC */ 913 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_CTRL2), 0, 914 ~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK); 915 916 tmp = RREG32_SOC15(UVD, 0, mmUVD_SOFT_RESET); 917 tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK; 918 tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK; 919 WREG32_SOC15(UVD, 0, mmUVD_SOFT_RESET, tmp); 920 921 for (i = 0; i < 10; ++i) { 922 uint32_t status; 923 924 for (j = 0; j < 100; ++j) { 925 status = RREG32_SOC15(UVD, 0, mmUVD_STATUS); 926 if (status & UVD_STATUS__IDLE) 927 break; 928 mdelay(10); 929 } 930 r = 0; 931 if (status & UVD_STATUS__IDLE) 932 break; 933 934 DRM_ERROR("VCN decode not responding, trying to reset the VCPU!!!\n"); 935 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 936 UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK, 937 ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK); 938 mdelay(10); 939 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 0, 940 ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK); 941 mdelay(10); 942 r = -1; 943 } 944 945 if (r) { 946 DRM_ERROR("VCN decode not responding, giving up!!!\n"); 947 return r; 948 } 949 /* enable master interrupt */ 950 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_MASTINT_EN), 951 UVD_MASTINT_EN__VCPU_EN_MASK, ~UVD_MASTINT_EN__VCPU_EN_MASK); 952 953 /* enable system interrupt for JRBC, TODO: move to set interrupt*/ 954 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SYS_INT_EN), 955 UVD_SYS_INT_EN__UVD_JRBC_EN_MASK, 956 ~UVD_SYS_INT_EN__UVD_JRBC_EN_MASK); 957 958 /* clear the busy bit of UVD_STATUS */ 959 tmp = RREG32_SOC15(UVD, 0, mmUVD_STATUS) & ~UVD_STATUS__UVD_BUSY; 960 WREG32_SOC15(UVD, 0, mmUVD_STATUS, tmp); 961 962 /* force RBC into idle state */ 963 rb_bufsz = order_base_2(ring->ring_size); 964 tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz); 965 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1); 966 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1); 967 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1); 968 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1); 969 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_CNTL, tmp); 970 971 /* set the write pointer delay */ 972 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR_CNTL, 0); 973 974 /* set the wb address */ 975 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR_ADDR, 976 (upper_32_bits(ring->gpu_addr) >> 2)); 977 978 /* program the RB_BASE for ring buffer */ 979 WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW, 980 lower_32_bits(ring->gpu_addr)); 981 WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH, 982 upper_32_bits(ring->gpu_addr)); 983 984 /* Initialize the ring buffer's read and write pointers */ 985 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR, 0); 986 987 WREG32_SOC15(UVD, 0, mmUVD_SCRATCH2, 0); 988 989 ring->wptr = RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR); 990 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR, 991 lower_32_bits(ring->wptr)); 992 993 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_RBC_RB_CNTL), 0, 994 ~UVD_RBC_RB_CNTL__RB_NO_FETCH_MASK); 995 996 ring = &adev->vcn.inst->ring_enc[0]; 997 WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR, lower_32_bits(ring->wptr)); 998 WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR, lower_32_bits(ring->wptr)); 999 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO, ring->gpu_addr); 1000 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr)); 1001 WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE, ring->ring_size / 4); 1002 1003 ring = &adev->vcn.inst->ring_enc[1]; 1004 WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr)); 1005 WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr)); 1006 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO2, ring->gpu_addr); 1007 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr)); 1008 WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE2, ring->ring_size / 4); 1009 1010 jpeg_v1_0_start(adev, 0); 1011 1012 /* Keeping one read-back to ensure all register writes are done, 1013 * otherwise it may introduce race conditions. 1014 */ 1015 RREG32_SOC15(UVD, 0, mmUVD_STATUS); 1016 1017 return 0; 1018 } 1019 1020 static int vcn_v1_0_start_dpg_mode(struct amdgpu_vcn_inst *vinst) 1021 { 1022 struct amdgpu_device *adev = vinst->adev; 1023 struct amdgpu_ring *ring = &adev->vcn.inst->ring_dec; 1024 uint32_t rb_bufsz, tmp; 1025 uint32_t lmi_swap_cntl; 1026 1027 /* disable byte swapping */ 1028 lmi_swap_cntl = 0; 1029 1030 vcn_1_0_enable_static_power_gating(vinst); 1031 1032 /* enable dynamic power gating mode */ 1033 tmp = RREG32_SOC15(UVD, 0, mmUVD_POWER_STATUS); 1034 tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK; 1035 tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK; 1036 WREG32_SOC15(UVD, 0, mmUVD_POWER_STATUS, tmp); 1037 1038 /* enable clock gating */ 1039 vcn_v1_0_clock_gating_dpg_mode(vinst, 0); 1040 1041 /* enable VCPU clock */ 1042 tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT); 1043 tmp |= UVD_VCPU_CNTL__CLK_EN_MASK; 1044 tmp |= UVD_VCPU_CNTL__MIF_WR_LOW_THRESHOLD_BP_MASK; 1045 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_VCPU_CNTL, tmp, 0xFFFFFFFF, 0); 1046 1047 /* disable interupt */ 1048 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MASTINT_EN, 1049 0, UVD_MASTINT_EN__VCPU_EN_MASK, 0); 1050 1051 /* initialize VCN memory controller */ 1052 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_CTRL, 1053 (8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) | 1054 UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK | 1055 UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK | 1056 UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK | 1057 UVD_LMI_CTRL__REQ_MODE_MASK | 1058 UVD_LMI_CTRL__CRC_RESET_MASK | 1059 UVD_LMI_CTRL__MASK_MC_URGENT_MASK | 1060 0x00100000L, 0xFFFFFFFF, 0); 1061 1062 #ifdef __BIG_ENDIAN 1063 /* swap (8 in 32) RB and IB */ 1064 lmi_swap_cntl = 0xa; 1065 #endif 1066 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl, 0xFFFFFFFF, 0); 1067 1068 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MPC_CNTL, 1069 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0xFFFFFFFF, 0); 1070 1071 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MPC_SET_MUXA0, 1072 ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) | 1073 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) | 1074 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) | 1075 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0xFFFFFFFF, 0); 1076 1077 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MPC_SET_MUXB0, 1078 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) | 1079 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) | 1080 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) | 1081 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0xFFFFFFFF, 0); 1082 1083 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MPC_SET_MUX, 1084 ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) | 1085 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) | 1086 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0xFFFFFFFF, 0); 1087 1088 vcn_v1_0_mc_resume_dpg_mode(vinst); 1089 1090 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_REG_XX_MASK, 0x10, 0xFFFFFFFF, 0); 1091 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_RBC_XX_IB_REG_CHECK, 0x3, 0xFFFFFFFF, 0); 1092 1093 /* boot up the VCPU */ 1094 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_SOFT_RESET, 0, 0xFFFFFFFF, 0); 1095 1096 /* enable UMC */ 1097 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_CTRL2, 1098 0x1F << UVD_LMI_CTRL2__RE_OFLD_MIF_WR_REQ_NUM__SHIFT, 1099 0xFFFFFFFF, 0); 1100 1101 /* enable master interrupt */ 1102 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MASTINT_EN, 1103 UVD_MASTINT_EN__VCPU_EN_MASK, UVD_MASTINT_EN__VCPU_EN_MASK, 0); 1104 1105 vcn_v1_0_clock_gating_dpg_mode(vinst, 1); 1106 /* setup mmUVD_LMI_CTRL */ 1107 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_CTRL, 1108 (8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) | 1109 UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK | 1110 UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK | 1111 UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK | 1112 UVD_LMI_CTRL__REQ_MODE_MASK | 1113 UVD_LMI_CTRL__CRC_RESET_MASK | 1114 UVD_LMI_CTRL__MASK_MC_URGENT_MASK | 1115 0x00100000L, 0xFFFFFFFF, 1); 1116 1117 tmp = adev->gfx.config.gb_addr_config; 1118 /* setup VCN global tiling registers */ 1119 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_JPEG_ADDR_CONFIG, tmp, 0xFFFFFFFF, 1); 1120 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_JPEG_UV_ADDR_CONFIG, tmp, 0xFFFFFFFF, 1); 1121 1122 /* enable System Interrupt for JRBC */ 1123 WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_SYS_INT_EN, 1124 UVD_SYS_INT_EN__UVD_JRBC_EN_MASK, 0xFFFFFFFF, 1); 1125 1126 /* force RBC into idle state */ 1127 rb_bufsz = order_base_2(ring->ring_size); 1128 tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz); 1129 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1); 1130 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1); 1131 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1); 1132 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1); 1133 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_CNTL, tmp); 1134 1135 /* set the write pointer delay */ 1136 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR_CNTL, 0); 1137 1138 /* set the wb address */ 1139 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR_ADDR, 1140 (upper_32_bits(ring->gpu_addr) >> 2)); 1141 1142 /* program the RB_BASE for ring buffer */ 1143 WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW, 1144 lower_32_bits(ring->gpu_addr)); 1145 WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH, 1146 upper_32_bits(ring->gpu_addr)); 1147 1148 /* Initialize the ring buffer's read and write pointers */ 1149 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR, 0); 1150 1151 WREG32_SOC15(UVD, 0, mmUVD_SCRATCH2, 0); 1152 1153 ring->wptr = RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR); 1154 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR, 1155 lower_32_bits(ring->wptr)); 1156 1157 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_RBC_RB_CNTL), 0, 1158 ~UVD_RBC_RB_CNTL__RB_NO_FETCH_MASK); 1159 1160 jpeg_v1_0_start(adev, 1); 1161 1162 /* Keeping one read-back to ensure all register writes are done, 1163 * otherwise it may introduce race conditions. 1164 */ 1165 RREG32_SOC15(UVD, 0, mmUVD_STATUS); 1166 1167 return 0; 1168 } 1169 1170 static int vcn_v1_0_start(struct amdgpu_vcn_inst *vinst) 1171 { 1172 struct amdgpu_device *adev = vinst->adev; 1173 1174 return (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ? 1175 vcn_v1_0_start_dpg_mode(vinst) : vcn_v1_0_start_spg_mode(vinst); 1176 } 1177 1178 /** 1179 * vcn_v1_0_stop_spg_mode - stop VCN block 1180 * 1181 * @vinst: VCN instance 1182 * 1183 * stop the VCN block 1184 */ 1185 static int vcn_v1_0_stop_spg_mode(struct amdgpu_vcn_inst *vinst) 1186 { 1187 struct amdgpu_device *adev = vinst->adev; 1188 int tmp; 1189 1190 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7); 1191 1192 tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK | 1193 UVD_LMI_STATUS__READ_CLEAN_MASK | 1194 UVD_LMI_STATUS__WRITE_CLEAN_MASK | 1195 UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK; 1196 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_LMI_STATUS, tmp, tmp); 1197 1198 /* stall UMC channel */ 1199 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_CTRL2), 1200 UVD_LMI_CTRL2__STALL_ARB_UMC_MASK, 1201 ~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK); 1202 1203 tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK | 1204 UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK; 1205 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_LMI_STATUS, tmp, tmp); 1206 1207 /* disable VCPU clock */ 1208 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_VCPU_CNTL), 0, 1209 ~UVD_VCPU_CNTL__CLK_EN_MASK); 1210 1211 /* reset LMI UMC/LMI */ 1212 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 1213 UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK, 1214 ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK); 1215 1216 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 1217 UVD_SOFT_RESET__LMI_SOFT_RESET_MASK, 1218 ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK); 1219 1220 /* put VCPU into reset */ 1221 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 1222 UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK, 1223 ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK); 1224 1225 WREG32_SOC15(UVD, 0, mmUVD_STATUS, 0); 1226 1227 vcn_v1_0_enable_clock_gating(vinst); 1228 vcn_1_0_enable_static_power_gating(vinst); 1229 1230 /* Keeping one read-back to ensure all register writes are done, 1231 * otherwise it may introduce race conditions. 1232 */ 1233 RREG32_SOC15(UVD, 0, mmUVD_STATUS); 1234 1235 return 0; 1236 } 1237 1238 static int vcn_v1_0_stop_dpg_mode(struct amdgpu_vcn_inst *vinst) 1239 { 1240 struct amdgpu_device *adev = vinst->adev; 1241 uint32_t tmp; 1242 1243 /* Wait for power status to be UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF */ 1244 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 1245 UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF, 1246 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); 1247 1248 /* wait for read ptr to be equal to write ptr */ 1249 tmp = RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR); 1250 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF); 1251 1252 tmp = RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2); 1253 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF); 1254 1255 tmp = RREG32_SOC15(UVD, 0, mmUVD_JRBC_RB_WPTR); 1256 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_JRBC_RB_RPTR, tmp, 0xFFFFFFFF); 1257 1258 tmp = RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF; 1259 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF); 1260 1261 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 1262 UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF, 1263 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); 1264 1265 /* disable dynamic power gating mode */ 1266 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_POWER_STATUS), 0, 1267 ~UVD_POWER_STATUS__UVD_PG_MODE_MASK); 1268 1269 /* Keeping one read-back to ensure all register writes are done, 1270 * otherwise it may introduce race conditions. 1271 */ 1272 RREG32_SOC15(UVD, 0, mmUVD_STATUS); 1273 1274 return 0; 1275 } 1276 1277 static int vcn_v1_0_stop(struct amdgpu_vcn_inst *vinst) 1278 { 1279 struct amdgpu_device *adev = vinst->adev; 1280 int r; 1281 1282 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) 1283 r = vcn_v1_0_stop_dpg_mode(vinst); 1284 else 1285 r = vcn_v1_0_stop_spg_mode(vinst); 1286 1287 return r; 1288 } 1289 1290 static int vcn_v1_0_pause_dpg_mode(struct amdgpu_vcn_inst *vinst, 1291 struct dpg_pause_state *new_state) 1292 { 1293 struct amdgpu_device *adev = vinst->adev; 1294 int inst_idx = vinst->inst; 1295 int ret_code; 1296 uint32_t reg_data = 0; 1297 uint32_t reg_data2 = 0; 1298 struct amdgpu_ring *ring; 1299 1300 /* pause/unpause if state is changed */ 1301 if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) { 1302 DRM_DEBUG("dpg pause state changed %d:%d -> %d:%d", 1303 adev->vcn.inst[inst_idx].pause_state.fw_based, 1304 adev->vcn.inst[inst_idx].pause_state.jpeg, 1305 new_state->fw_based, new_state->jpeg); 1306 1307 reg_data = RREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE) & 1308 (~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK); 1309 1310 if (new_state->fw_based == VCN_DPG_STATE__PAUSE) { 1311 ret_code = 0; 1312 1313 if (!(reg_data & UVD_DPG_PAUSE__JPEG_PAUSE_DPG_ACK_MASK)) 1314 ret_code = SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 1315 UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF, 1316 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); 1317 1318 if (!ret_code) { 1319 /* pause DPG non-jpeg */ 1320 reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK; 1321 WREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE, reg_data); 1322 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_DPG_PAUSE, 1323 UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK, 1324 UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK); 1325 1326 /* Restore */ 1327 ring = &adev->vcn.inst->ring_enc[0]; 1328 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO, ring->gpu_addr); 1329 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr)); 1330 WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE, ring->ring_size / 4); 1331 WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR, lower_32_bits(ring->wptr)); 1332 WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR, lower_32_bits(ring->wptr)); 1333 1334 ring = &adev->vcn.inst->ring_enc[1]; 1335 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO2, ring->gpu_addr); 1336 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr)); 1337 WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE2, ring->ring_size / 4); 1338 WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr)); 1339 WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr)); 1340 1341 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR, 1342 RREG32_SOC15(UVD, 0, mmUVD_SCRATCH2) & 0x7FFFFFFF); 1343 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 1344 UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, 1345 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); 1346 } 1347 } else { 1348 /* unpause dpg non-jpeg, no need to wait */ 1349 reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK; 1350 WREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE, reg_data); 1351 } 1352 adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based; 1353 } 1354 1355 /* pause/unpause if state is changed */ 1356 if (adev->vcn.inst[inst_idx].pause_state.jpeg != new_state->jpeg) { 1357 DRM_DEBUG("dpg pause state changed %d:%d -> %d:%d", 1358 adev->vcn.inst[inst_idx].pause_state.fw_based, 1359 adev->vcn.inst[inst_idx].pause_state.jpeg, 1360 new_state->fw_based, new_state->jpeg); 1361 1362 reg_data = RREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE) & 1363 (~UVD_DPG_PAUSE__JPEG_PAUSE_DPG_ACK_MASK); 1364 1365 if (new_state->jpeg == VCN_DPG_STATE__PAUSE) { 1366 ret_code = 0; 1367 1368 if (!(reg_data & UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK)) 1369 ret_code = SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 1370 UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF, 1371 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); 1372 1373 if (!ret_code) { 1374 /* Make sure JPRG Snoop is disabled before sending the pause */ 1375 reg_data2 = RREG32_SOC15(UVD, 0, mmUVD_POWER_STATUS); 1376 reg_data2 |= UVD_POWER_STATUS__JRBC_SNOOP_DIS_MASK; 1377 WREG32_SOC15(UVD, 0, mmUVD_POWER_STATUS, reg_data2); 1378 1379 /* pause DPG jpeg */ 1380 reg_data |= UVD_DPG_PAUSE__JPEG_PAUSE_DPG_REQ_MASK; 1381 WREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE, reg_data); 1382 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_DPG_PAUSE, 1383 UVD_DPG_PAUSE__JPEG_PAUSE_DPG_ACK_MASK, 1384 UVD_DPG_PAUSE__JPEG_PAUSE_DPG_ACK_MASK); 1385 1386 /* Restore */ 1387 ring = adev->jpeg.inst->ring_dec; 1388 WREG32_SOC15(UVD, 0, mmUVD_LMI_JRBC_RB_VMID, 0); 1389 WREG32_SOC15(UVD, 0, mmUVD_JRBC_RB_CNTL, 1390 UVD_JRBC_RB_CNTL__RB_NO_FETCH_MASK | 1391 UVD_JRBC_RB_CNTL__RB_RPTR_WR_EN_MASK); 1392 WREG32_SOC15(UVD, 0, mmUVD_LMI_JRBC_RB_64BIT_BAR_LOW, 1393 lower_32_bits(ring->gpu_addr)); 1394 WREG32_SOC15(UVD, 0, mmUVD_LMI_JRBC_RB_64BIT_BAR_HIGH, 1395 upper_32_bits(ring->gpu_addr)); 1396 WREG32_SOC15(UVD, 0, mmUVD_JRBC_RB_RPTR, ring->wptr); 1397 WREG32_SOC15(UVD, 0, mmUVD_JRBC_RB_WPTR, ring->wptr); 1398 WREG32_SOC15(UVD, 0, mmUVD_JRBC_RB_CNTL, 1399 UVD_JRBC_RB_CNTL__RB_RPTR_WR_EN_MASK); 1400 1401 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR, 1402 RREG32_SOC15(UVD, 0, mmUVD_SCRATCH2) & 0x7FFFFFFF); 1403 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 1404 UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, 1405 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); 1406 } 1407 } else { 1408 /* unpause dpg jpeg, no need to wait */ 1409 reg_data &= ~UVD_DPG_PAUSE__JPEG_PAUSE_DPG_REQ_MASK; 1410 WREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE, reg_data); 1411 } 1412 adev->vcn.inst[inst_idx].pause_state.jpeg = new_state->jpeg; 1413 } 1414 1415 return 0; 1416 } 1417 1418 static bool vcn_v1_0_is_idle(struct amdgpu_ip_block *ip_block) 1419 { 1420 struct amdgpu_device *adev = ip_block->adev; 1421 1422 return (RREG32_SOC15(VCN, 0, mmUVD_STATUS) == UVD_STATUS__IDLE); 1423 } 1424 1425 static int vcn_v1_0_wait_for_idle(struct amdgpu_ip_block *ip_block) 1426 { 1427 struct amdgpu_device *adev = ip_block->adev; 1428 int ret; 1429 1430 ret = SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_STATUS, UVD_STATUS__IDLE, 1431 UVD_STATUS__IDLE); 1432 1433 return ret; 1434 } 1435 1436 static int vcn_v1_0_set_clockgating_state(struct amdgpu_ip_block *ip_block, 1437 enum amd_clockgating_state state) 1438 { 1439 struct amdgpu_device *adev = ip_block->adev; 1440 struct amdgpu_vcn_inst *vinst = adev->vcn.inst; 1441 bool enable = (state == AMD_CG_STATE_GATE); 1442 1443 if (enable) { 1444 /* wait for STATUS to clear */ 1445 if (!vcn_v1_0_is_idle(ip_block)) 1446 return -EBUSY; 1447 vcn_v1_0_enable_clock_gating(vinst); 1448 } else { 1449 /* disable HW gating and enable Sw gating */ 1450 vcn_v1_0_disable_clock_gating(vinst); 1451 } 1452 return 0; 1453 } 1454 1455 /** 1456 * vcn_v1_0_dec_ring_get_rptr - get read pointer 1457 * 1458 * @ring: amdgpu_ring pointer 1459 * 1460 * Returns the current hardware read pointer 1461 */ 1462 static uint64_t vcn_v1_0_dec_ring_get_rptr(struct amdgpu_ring *ring) 1463 { 1464 struct amdgpu_device *adev = ring->adev; 1465 1466 return RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR); 1467 } 1468 1469 /** 1470 * vcn_v1_0_dec_ring_get_wptr - get write pointer 1471 * 1472 * @ring: amdgpu_ring pointer 1473 * 1474 * Returns the current hardware write pointer 1475 */ 1476 static uint64_t vcn_v1_0_dec_ring_get_wptr(struct amdgpu_ring *ring) 1477 { 1478 struct amdgpu_device *adev = ring->adev; 1479 1480 return RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR); 1481 } 1482 1483 /** 1484 * vcn_v1_0_dec_ring_set_wptr - set write pointer 1485 * 1486 * @ring: amdgpu_ring pointer 1487 * 1488 * Commits the write pointer to the hardware 1489 */ 1490 static void vcn_v1_0_dec_ring_set_wptr(struct amdgpu_ring *ring) 1491 { 1492 struct amdgpu_device *adev = ring->adev; 1493 1494 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) 1495 WREG32_SOC15(UVD, 0, mmUVD_SCRATCH2, 1496 lower_32_bits(ring->wptr) | 0x80000000); 1497 1498 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr)); 1499 } 1500 1501 /** 1502 * vcn_v1_0_dec_ring_insert_start - insert a start command 1503 * 1504 * @ring: amdgpu_ring pointer 1505 * 1506 * Write a start command to the ring. 1507 */ 1508 static void vcn_v1_0_dec_ring_insert_start(struct amdgpu_ring *ring) 1509 { 1510 struct amdgpu_device *adev = ring->adev; 1511 1512 amdgpu_ring_write(ring, 1513 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0), 0)); 1514 amdgpu_ring_write(ring, 0); 1515 amdgpu_ring_write(ring, 1516 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD), 0)); 1517 amdgpu_ring_write(ring, VCN_DEC_CMD_PACKET_START << 1); 1518 } 1519 1520 /** 1521 * vcn_v1_0_dec_ring_insert_end - insert a end command 1522 * 1523 * @ring: amdgpu_ring pointer 1524 * 1525 * Write a end command to the ring. 1526 */ 1527 static void vcn_v1_0_dec_ring_insert_end(struct amdgpu_ring *ring) 1528 { 1529 struct amdgpu_device *adev = ring->adev; 1530 1531 amdgpu_ring_write(ring, 1532 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD), 0)); 1533 amdgpu_ring_write(ring, VCN_DEC_CMD_PACKET_END << 1); 1534 } 1535 1536 /** 1537 * vcn_v1_0_dec_ring_emit_fence - emit an fence & trap command 1538 * 1539 * @ring: amdgpu_ring pointer 1540 * @addr: address 1541 * @seq: sequence number 1542 * @flags: fence related flags 1543 * 1544 * Write a fence and a trap command to the ring. 1545 */ 1546 static void vcn_v1_0_dec_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq, 1547 unsigned flags) 1548 { 1549 struct amdgpu_device *adev = ring->adev; 1550 1551 WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT); 1552 1553 amdgpu_ring_write(ring, 1554 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_CONTEXT_ID), 0)); 1555 amdgpu_ring_write(ring, seq); 1556 amdgpu_ring_write(ring, 1557 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0), 0)); 1558 amdgpu_ring_write(ring, addr & 0xffffffff); 1559 amdgpu_ring_write(ring, 1560 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1), 0)); 1561 amdgpu_ring_write(ring, upper_32_bits(addr) & 0xff); 1562 amdgpu_ring_write(ring, 1563 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD), 0)); 1564 amdgpu_ring_write(ring, VCN_DEC_CMD_FENCE << 1); 1565 1566 amdgpu_ring_write(ring, 1567 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0), 0)); 1568 amdgpu_ring_write(ring, 0); 1569 amdgpu_ring_write(ring, 1570 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1), 0)); 1571 amdgpu_ring_write(ring, 0); 1572 amdgpu_ring_write(ring, 1573 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD), 0)); 1574 amdgpu_ring_write(ring, VCN_DEC_CMD_TRAP << 1); 1575 } 1576 1577 /** 1578 * vcn_v1_0_dec_ring_emit_ib - execute indirect buffer 1579 * 1580 * @ring: amdgpu_ring pointer 1581 * @job: job to retrieve vmid from 1582 * @ib: indirect buffer to execute 1583 * @flags: unused 1584 * 1585 * Write ring commands to execute the indirect buffer 1586 */ 1587 static void vcn_v1_0_dec_ring_emit_ib(struct amdgpu_ring *ring, 1588 struct amdgpu_job *job, 1589 struct amdgpu_ib *ib, 1590 uint32_t flags) 1591 { 1592 struct amdgpu_device *adev = ring->adev; 1593 unsigned vmid = AMDGPU_JOB_GET_VMID(job); 1594 1595 amdgpu_ring_write(ring, 1596 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_RBC_IB_VMID), 0)); 1597 amdgpu_ring_write(ring, vmid); 1598 1599 amdgpu_ring_write(ring, 1600 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_RBC_IB_64BIT_BAR_LOW), 0)); 1601 amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr)); 1602 amdgpu_ring_write(ring, 1603 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH), 0)); 1604 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr)); 1605 amdgpu_ring_write(ring, 1606 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_RBC_IB_SIZE), 0)); 1607 amdgpu_ring_write(ring, ib->length_dw); 1608 } 1609 1610 static void vcn_v1_0_dec_ring_emit_reg_wait(struct amdgpu_ring *ring, 1611 uint32_t reg, uint32_t val, 1612 uint32_t mask) 1613 { 1614 struct amdgpu_device *adev = ring->adev; 1615 1616 amdgpu_ring_write(ring, 1617 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0), 0)); 1618 amdgpu_ring_write(ring, reg << 2); 1619 amdgpu_ring_write(ring, 1620 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1), 0)); 1621 amdgpu_ring_write(ring, val); 1622 amdgpu_ring_write(ring, 1623 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GP_SCRATCH8), 0)); 1624 amdgpu_ring_write(ring, mask); 1625 amdgpu_ring_write(ring, 1626 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD), 0)); 1627 amdgpu_ring_write(ring, VCN_DEC_CMD_REG_READ_COND_WAIT << 1); 1628 } 1629 1630 static void vcn_v1_0_dec_ring_emit_vm_flush(struct amdgpu_ring *ring, 1631 unsigned vmid, uint64_t pd_addr) 1632 { 1633 struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->vm_hub]; 1634 uint32_t data0, data1, mask; 1635 1636 pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr); 1637 1638 /* wait for register write */ 1639 data0 = hub->ctx0_ptb_addr_lo32 + vmid * hub->ctx_addr_distance; 1640 data1 = lower_32_bits(pd_addr); 1641 mask = 0xffffffff; 1642 vcn_v1_0_dec_ring_emit_reg_wait(ring, data0, data1, mask); 1643 } 1644 1645 static void vcn_v1_0_dec_ring_emit_wreg(struct amdgpu_ring *ring, 1646 uint32_t reg, uint32_t val) 1647 { 1648 struct amdgpu_device *adev = ring->adev; 1649 1650 amdgpu_ring_write(ring, 1651 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0), 0)); 1652 amdgpu_ring_write(ring, reg << 2); 1653 amdgpu_ring_write(ring, 1654 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1), 0)); 1655 amdgpu_ring_write(ring, val); 1656 amdgpu_ring_write(ring, 1657 PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD), 0)); 1658 amdgpu_ring_write(ring, VCN_DEC_CMD_WRITE_REG << 1); 1659 } 1660 1661 /** 1662 * vcn_v1_0_enc_ring_get_rptr - get enc read pointer 1663 * 1664 * @ring: amdgpu_ring pointer 1665 * 1666 * Returns the current hardware enc read pointer 1667 */ 1668 static uint64_t vcn_v1_0_enc_ring_get_rptr(struct amdgpu_ring *ring) 1669 { 1670 struct amdgpu_device *adev = ring->adev; 1671 1672 if (ring == &adev->vcn.inst->ring_enc[0]) 1673 return RREG32_SOC15(UVD, 0, mmUVD_RB_RPTR); 1674 else 1675 return RREG32_SOC15(UVD, 0, mmUVD_RB_RPTR2); 1676 } 1677 1678 /** 1679 * vcn_v1_0_enc_ring_get_wptr - get enc write pointer 1680 * 1681 * @ring: amdgpu_ring pointer 1682 * 1683 * Returns the current hardware enc write pointer 1684 */ 1685 static uint64_t vcn_v1_0_enc_ring_get_wptr(struct amdgpu_ring *ring) 1686 { 1687 struct amdgpu_device *adev = ring->adev; 1688 1689 if (ring == &adev->vcn.inst->ring_enc[0]) 1690 return RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR); 1691 else 1692 return RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2); 1693 } 1694 1695 /** 1696 * vcn_v1_0_enc_ring_set_wptr - set enc write pointer 1697 * 1698 * @ring: amdgpu_ring pointer 1699 * 1700 * Commits the enc write pointer to the hardware 1701 */ 1702 static void vcn_v1_0_enc_ring_set_wptr(struct amdgpu_ring *ring) 1703 { 1704 struct amdgpu_device *adev = ring->adev; 1705 1706 if (ring == &adev->vcn.inst->ring_enc[0]) 1707 WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR, 1708 lower_32_bits(ring->wptr)); 1709 else 1710 WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2, 1711 lower_32_bits(ring->wptr)); 1712 } 1713 1714 /** 1715 * vcn_v1_0_enc_ring_emit_fence - emit an enc fence & trap command 1716 * 1717 * @ring: amdgpu_ring pointer 1718 * @addr: address 1719 * @seq: sequence number 1720 * @flags: fence related flags 1721 * 1722 * Write enc a fence and a trap command to the ring. 1723 */ 1724 static void vcn_v1_0_enc_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, 1725 u64 seq, unsigned flags) 1726 { 1727 WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT); 1728 1729 amdgpu_ring_write(ring, VCN_ENC_CMD_FENCE); 1730 amdgpu_ring_write(ring, addr); 1731 amdgpu_ring_write(ring, upper_32_bits(addr)); 1732 amdgpu_ring_write(ring, seq); 1733 amdgpu_ring_write(ring, VCN_ENC_CMD_TRAP); 1734 } 1735 1736 static void vcn_v1_0_enc_ring_insert_end(struct amdgpu_ring *ring) 1737 { 1738 amdgpu_ring_write(ring, VCN_ENC_CMD_END); 1739 } 1740 1741 /** 1742 * vcn_v1_0_enc_ring_emit_ib - enc execute indirect buffer 1743 * 1744 * @ring: amdgpu_ring pointer 1745 * @job: job to retrive vmid from 1746 * @ib: indirect buffer to execute 1747 * @flags: unused 1748 * 1749 * Write enc ring commands to execute the indirect buffer 1750 */ 1751 static void vcn_v1_0_enc_ring_emit_ib(struct amdgpu_ring *ring, 1752 struct amdgpu_job *job, 1753 struct amdgpu_ib *ib, 1754 uint32_t flags) 1755 { 1756 unsigned vmid = AMDGPU_JOB_GET_VMID(job); 1757 1758 amdgpu_ring_write(ring, VCN_ENC_CMD_IB); 1759 amdgpu_ring_write(ring, vmid); 1760 amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr)); 1761 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr)); 1762 amdgpu_ring_write(ring, ib->length_dw); 1763 } 1764 1765 static void vcn_v1_0_enc_ring_emit_reg_wait(struct amdgpu_ring *ring, 1766 uint32_t reg, uint32_t val, 1767 uint32_t mask) 1768 { 1769 amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WAIT); 1770 amdgpu_ring_write(ring, reg << 2); 1771 amdgpu_ring_write(ring, mask); 1772 amdgpu_ring_write(ring, val); 1773 } 1774 1775 static void vcn_v1_0_enc_ring_emit_vm_flush(struct amdgpu_ring *ring, 1776 unsigned int vmid, uint64_t pd_addr) 1777 { 1778 struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->vm_hub]; 1779 1780 pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr); 1781 1782 /* wait for reg writes */ 1783 vcn_v1_0_enc_ring_emit_reg_wait(ring, hub->ctx0_ptb_addr_lo32 + 1784 vmid * hub->ctx_addr_distance, 1785 lower_32_bits(pd_addr), 0xffffffff); 1786 } 1787 1788 static void vcn_v1_0_enc_ring_emit_wreg(struct amdgpu_ring *ring, 1789 uint32_t reg, uint32_t val) 1790 { 1791 amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WRITE); 1792 amdgpu_ring_write(ring, reg << 2); 1793 amdgpu_ring_write(ring, val); 1794 } 1795 1796 static int vcn_v1_0_set_interrupt_state(struct amdgpu_device *adev, 1797 struct amdgpu_irq_src *source, 1798 unsigned type, 1799 enum amdgpu_interrupt_state state) 1800 { 1801 return 0; 1802 } 1803 1804 static int vcn_v1_0_process_interrupt(struct amdgpu_device *adev, 1805 struct amdgpu_irq_src *source, 1806 struct amdgpu_iv_entry *entry) 1807 { 1808 DRM_DEBUG("IH: VCN TRAP\n"); 1809 1810 switch (entry->src_id) { 1811 case 124: 1812 amdgpu_fence_process(&adev->vcn.inst->ring_dec); 1813 break; 1814 case 119: 1815 amdgpu_fence_process(&adev->vcn.inst->ring_enc[0]); 1816 break; 1817 case 120: 1818 amdgpu_fence_process(&adev->vcn.inst->ring_enc[1]); 1819 break; 1820 default: 1821 DRM_ERROR("Unhandled interrupt: %d %d\n", 1822 entry->src_id, entry->src_data[0]); 1823 break; 1824 } 1825 1826 return 0; 1827 } 1828 1829 static void vcn_v1_0_dec_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) 1830 { 1831 struct amdgpu_device *adev = ring->adev; 1832 int i; 1833 1834 WARN_ON(ring->wptr % 2 || count % 2); 1835 1836 for (i = 0; i < count / 2; i++) { 1837 amdgpu_ring_write(ring, PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_NO_OP), 0)); 1838 amdgpu_ring_write(ring, 0); 1839 } 1840 } 1841 1842 static int vcn_v1_0_set_pg_state(struct amdgpu_vcn_inst *vinst, 1843 enum amd_powergating_state state) 1844 { 1845 /* This doesn't actually powergate the VCN block. 1846 * That's done in the dpm code via the SMC. This 1847 * just re-inits the block as necessary. The actual 1848 * gating still happens in the dpm code. We should 1849 * revisit this when there is a cleaner line between 1850 * the smc and the hw blocks 1851 */ 1852 int ret; 1853 1854 if (state == vinst->cur_state) 1855 return 0; 1856 1857 if (state == AMD_PG_STATE_GATE) 1858 ret = vcn_v1_0_stop(vinst); 1859 else 1860 ret = vcn_v1_0_start(vinst); 1861 1862 if (!ret) 1863 vinst->cur_state = state; 1864 1865 return ret; 1866 } 1867 1868 static void vcn_v1_0_idle_work_handler(struct work_struct *work) 1869 { 1870 struct amdgpu_vcn_inst *vcn_inst = 1871 container_of(work, struct amdgpu_vcn_inst, idle_work.work); 1872 struct amdgpu_device *adev = vcn_inst->adev; 1873 unsigned int fences = 0, i; 1874 1875 for (i = 0; i < adev->vcn.inst[0].num_enc_rings; ++i) 1876 fences += amdgpu_fence_count_emitted(&adev->vcn.inst->ring_enc[i]); 1877 1878 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) { 1879 struct dpg_pause_state new_state; 1880 1881 if (fences) 1882 new_state.fw_based = VCN_DPG_STATE__PAUSE; 1883 else 1884 new_state.fw_based = VCN_DPG_STATE__UNPAUSE; 1885 1886 if (amdgpu_fence_count_emitted(adev->jpeg.inst->ring_dec)) 1887 new_state.jpeg = VCN_DPG_STATE__PAUSE; 1888 else 1889 new_state.jpeg = VCN_DPG_STATE__UNPAUSE; 1890 1891 adev->vcn.inst->pause_dpg_mode(vcn_inst, &new_state); 1892 } 1893 1894 fences += amdgpu_fence_count_emitted(adev->jpeg.inst->ring_dec); 1895 fences += amdgpu_fence_count_emitted(&adev->vcn.inst->ring_dec); 1896 1897 if (fences == 0) { 1898 amdgpu_gfx_off_ctrl(adev, true); 1899 if (adev->pm.dpm_enabled) 1900 amdgpu_dpm_enable_vcn(adev, false, 0); 1901 else 1902 amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN, 1903 AMD_PG_STATE_GATE); 1904 } else { 1905 schedule_delayed_work(&adev->vcn.inst[0].idle_work, VCN_IDLE_TIMEOUT); 1906 } 1907 } 1908 1909 static void vcn_v1_0_ring_begin_use(struct amdgpu_ring *ring) 1910 { 1911 struct amdgpu_device *adev = ring->adev; 1912 bool set_clocks = !cancel_delayed_work_sync(&adev->vcn.inst[0].idle_work); 1913 1914 mutex_lock(&adev->vcn.inst[0].vcn1_jpeg1_workaround); 1915 1916 if (amdgpu_fence_wait_empty(ring->adev->jpeg.inst->ring_dec)) 1917 DRM_ERROR("VCN dec: jpeg dec ring may not be empty\n"); 1918 1919 vcn_v1_0_set_pg_for_begin_use(ring, set_clocks); 1920 1921 } 1922 1923 void vcn_v1_0_set_pg_for_begin_use(struct amdgpu_ring *ring, bool set_clocks) 1924 { 1925 struct amdgpu_device *adev = ring->adev; 1926 1927 if (set_clocks) { 1928 amdgpu_gfx_off_ctrl(adev, false); 1929 if (adev->pm.dpm_enabled) 1930 amdgpu_dpm_enable_vcn(adev, true, 0); 1931 else 1932 amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN, 1933 AMD_PG_STATE_UNGATE); 1934 } 1935 1936 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) { 1937 struct dpg_pause_state new_state; 1938 unsigned int fences = 0, i; 1939 1940 for (i = 0; i < adev->vcn.inst[0].num_enc_rings; ++i) 1941 fences += amdgpu_fence_count_emitted(&adev->vcn.inst->ring_enc[i]); 1942 1943 if (fences) 1944 new_state.fw_based = VCN_DPG_STATE__PAUSE; 1945 else 1946 new_state.fw_based = VCN_DPG_STATE__UNPAUSE; 1947 1948 if (amdgpu_fence_count_emitted(adev->jpeg.inst->ring_dec)) 1949 new_state.jpeg = VCN_DPG_STATE__PAUSE; 1950 else 1951 new_state.jpeg = VCN_DPG_STATE__UNPAUSE; 1952 1953 if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC) 1954 new_state.fw_based = VCN_DPG_STATE__PAUSE; 1955 else if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_JPEG) 1956 new_state.jpeg = VCN_DPG_STATE__PAUSE; 1957 1958 adev->vcn.inst->pause_dpg_mode(adev->vcn.inst, &new_state); 1959 } 1960 } 1961 1962 void vcn_v1_0_ring_end_use(struct amdgpu_ring *ring) 1963 { 1964 schedule_delayed_work(&ring->adev->vcn.inst[0].idle_work, VCN_IDLE_TIMEOUT); 1965 mutex_unlock(&ring->adev->vcn.inst[0].vcn1_jpeg1_workaround); 1966 } 1967 1968 static void vcn_v1_0_print_ip_state(struct amdgpu_ip_block *ip_block, struct drm_printer *p) 1969 { 1970 struct amdgpu_device *adev = ip_block->adev; 1971 int i, j; 1972 uint32_t reg_count = ARRAY_SIZE(vcn_reg_list_1_0); 1973 uint32_t inst_off, is_powered; 1974 1975 if (!adev->vcn.ip_dump) 1976 return; 1977 1978 drm_printf(p, "num_instances:%d\n", adev->vcn.num_vcn_inst); 1979 for (i = 0; i < adev->vcn.num_vcn_inst; i++) { 1980 if (adev->vcn.harvest_config & (1 << i)) { 1981 drm_printf(p, "\nHarvested Instance:VCN%d Skipping dump\n", i); 1982 continue; 1983 } 1984 1985 inst_off = i * reg_count; 1986 is_powered = (adev->vcn.ip_dump[inst_off] & 1987 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK) != 1; 1988 1989 if (is_powered) { 1990 drm_printf(p, "\nActive Instance:VCN%d\n", i); 1991 for (j = 0; j < reg_count; j++) 1992 drm_printf(p, "%-50s \t 0x%08x\n", vcn_reg_list_1_0[j].reg_name, 1993 adev->vcn.ip_dump[inst_off + j]); 1994 } else { 1995 drm_printf(p, "\nInactive Instance:VCN%d\n", i); 1996 } 1997 } 1998 } 1999 2000 static void vcn_v1_0_dump_ip_state(struct amdgpu_ip_block *ip_block) 2001 { 2002 struct amdgpu_device *adev = ip_block->adev; 2003 int i, j; 2004 bool is_powered; 2005 uint32_t inst_off; 2006 uint32_t reg_count = ARRAY_SIZE(vcn_reg_list_1_0); 2007 2008 if (!adev->vcn.ip_dump) 2009 return; 2010 2011 for (i = 0; i < adev->vcn.num_vcn_inst; i++) { 2012 if (adev->vcn.harvest_config & (1 << i)) 2013 continue; 2014 2015 inst_off = i * reg_count; 2016 /* mmUVD_POWER_STATUS is always readable and is first element of the array */ 2017 adev->vcn.ip_dump[inst_off] = RREG32_SOC15(VCN, i, mmUVD_POWER_STATUS); 2018 is_powered = (adev->vcn.ip_dump[inst_off] & 2019 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK) != 1; 2020 2021 if (is_powered) 2022 for (j = 1; j < reg_count; j++) 2023 adev->vcn.ip_dump[inst_off + j] = 2024 RREG32(SOC15_REG_ENTRY_OFFSET_INST(vcn_reg_list_1_0[j], i)); 2025 } 2026 } 2027 2028 static const struct amd_ip_funcs vcn_v1_0_ip_funcs = { 2029 .name = "vcn_v1_0", 2030 .early_init = vcn_v1_0_early_init, 2031 .sw_init = vcn_v1_0_sw_init, 2032 .sw_fini = vcn_v1_0_sw_fini, 2033 .hw_init = vcn_v1_0_hw_init, 2034 .hw_fini = vcn_v1_0_hw_fini, 2035 .suspend = vcn_v1_0_suspend, 2036 .resume = vcn_v1_0_resume, 2037 .is_idle = vcn_v1_0_is_idle, 2038 .wait_for_idle = vcn_v1_0_wait_for_idle, 2039 .set_clockgating_state = vcn_v1_0_set_clockgating_state, 2040 .set_powergating_state = vcn_set_powergating_state, 2041 .dump_ip_state = vcn_v1_0_dump_ip_state, 2042 .print_ip_state = vcn_v1_0_print_ip_state, 2043 }; 2044 2045 /* 2046 * It is a hardware issue that VCN can't handle a GTT TMZ buffer on 2047 * CHIP_RAVEN series ASIC. Move such a GTT TMZ buffer to VRAM domain 2048 * before command submission as a workaround. 2049 */ 2050 static int vcn_v1_0_validate_bo(struct amdgpu_cs_parser *parser, 2051 struct amdgpu_job *job, 2052 uint64_t addr) 2053 { 2054 struct ttm_operation_ctx ctx = { false, false }; 2055 struct amdgpu_fpriv *fpriv = parser->filp->driver_priv; 2056 struct amdgpu_vm *vm = &fpriv->vm; 2057 struct amdgpu_bo_va_mapping *mapping; 2058 struct amdgpu_bo *bo; 2059 int r; 2060 2061 addr &= AMDGPU_GMC_HOLE_MASK; 2062 if (addr & 0x7) { 2063 DRM_ERROR("VCN messages must be 8 byte aligned!\n"); 2064 return -EINVAL; 2065 } 2066 2067 mapping = amdgpu_vm_bo_lookup_mapping(vm, addr/AMDGPU_GPU_PAGE_SIZE); 2068 if (!mapping || !mapping->bo_va || !mapping->bo_va->base.bo) 2069 return -EINVAL; 2070 2071 bo = mapping->bo_va->base.bo; 2072 if (!(bo->flags & AMDGPU_GEM_CREATE_ENCRYPTED)) 2073 return 0; 2074 2075 amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_VRAM); 2076 r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx); 2077 if (r) { 2078 DRM_ERROR("Failed to validate the VCN message BO (%d)!\n", r); 2079 return r; 2080 } 2081 2082 return r; 2083 } 2084 2085 static int vcn_v1_0_ring_patch_cs_in_place(struct amdgpu_cs_parser *p, 2086 struct amdgpu_job *job, 2087 struct amdgpu_ib *ib) 2088 { 2089 uint32_t msg_lo = 0, msg_hi = 0; 2090 int i, r; 2091 2092 if (!(ib->flags & AMDGPU_IB_FLAGS_SECURE)) 2093 return 0; 2094 2095 for (i = 0; i < ib->length_dw; i += 2) { 2096 uint32_t reg = amdgpu_ib_get_value(ib, i); 2097 uint32_t val = amdgpu_ib_get_value(ib, i + 1); 2098 2099 if (reg == PACKET0(p->adev->vcn.inst[0].internal.data0, 0)) { 2100 msg_lo = val; 2101 } else if (reg == PACKET0(p->adev->vcn.inst[0].internal.data1, 0)) { 2102 msg_hi = val; 2103 } else if (reg == PACKET0(p->adev->vcn.inst[0].internal.cmd, 0)) { 2104 r = vcn_v1_0_validate_bo(p, job, 2105 ((u64)msg_hi) << 32 | msg_lo); 2106 if (r) 2107 return r; 2108 } 2109 } 2110 2111 return 0; 2112 } 2113 2114 static const struct amdgpu_ring_funcs vcn_v1_0_dec_ring_vm_funcs = { 2115 .type = AMDGPU_RING_TYPE_VCN_DEC, 2116 .align_mask = 0xf, 2117 .support_64bit_ptrs = false, 2118 .no_user_fence = true, 2119 .secure_submission_supported = true, 2120 .get_rptr = vcn_v1_0_dec_ring_get_rptr, 2121 .get_wptr = vcn_v1_0_dec_ring_get_wptr, 2122 .set_wptr = vcn_v1_0_dec_ring_set_wptr, 2123 .patch_cs_in_place = vcn_v1_0_ring_patch_cs_in_place, 2124 .emit_frame_size = 2125 6 + 6 + /* hdp invalidate / flush */ 2126 SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 + 2127 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 + 2128 8 + /* vcn_v1_0_dec_ring_emit_vm_flush */ 2129 14 + 14 + /* vcn_v1_0_dec_ring_emit_fence x2 vm fence */ 2130 6, 2131 .emit_ib_size = 8, /* vcn_v1_0_dec_ring_emit_ib */ 2132 .emit_ib = vcn_v1_0_dec_ring_emit_ib, 2133 .emit_fence = vcn_v1_0_dec_ring_emit_fence, 2134 .emit_vm_flush = vcn_v1_0_dec_ring_emit_vm_flush, 2135 .test_ring = amdgpu_vcn_dec_ring_test_ring, 2136 .test_ib = amdgpu_vcn_dec_ring_test_ib, 2137 .insert_nop = vcn_v1_0_dec_ring_insert_nop, 2138 .insert_start = vcn_v1_0_dec_ring_insert_start, 2139 .insert_end = vcn_v1_0_dec_ring_insert_end, 2140 .pad_ib = amdgpu_ring_generic_pad_ib, 2141 .begin_use = vcn_v1_0_ring_begin_use, 2142 .end_use = vcn_v1_0_ring_end_use, 2143 .emit_wreg = vcn_v1_0_dec_ring_emit_wreg, 2144 .emit_reg_wait = vcn_v1_0_dec_ring_emit_reg_wait, 2145 .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper, 2146 }; 2147 2148 static const struct amdgpu_ring_funcs vcn_v1_0_enc_ring_vm_funcs = { 2149 .type = AMDGPU_RING_TYPE_VCN_ENC, 2150 .align_mask = 0x3f, 2151 .nop = VCN_ENC_CMD_NO_OP, 2152 .support_64bit_ptrs = false, 2153 .no_user_fence = true, 2154 .get_rptr = vcn_v1_0_enc_ring_get_rptr, 2155 .get_wptr = vcn_v1_0_enc_ring_get_wptr, 2156 .set_wptr = vcn_v1_0_enc_ring_set_wptr, 2157 .emit_frame_size = 2158 SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 + 2159 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 + 2160 4 + /* vcn_v1_0_enc_ring_emit_vm_flush */ 2161 5 + 5 + /* vcn_v1_0_enc_ring_emit_fence x2 vm fence */ 2162 1, /* vcn_v1_0_enc_ring_insert_end */ 2163 .emit_ib_size = 5, /* vcn_v1_0_enc_ring_emit_ib */ 2164 .emit_ib = vcn_v1_0_enc_ring_emit_ib, 2165 .emit_fence = vcn_v1_0_enc_ring_emit_fence, 2166 .emit_vm_flush = vcn_v1_0_enc_ring_emit_vm_flush, 2167 .test_ring = amdgpu_vcn_enc_ring_test_ring, 2168 .test_ib = amdgpu_vcn_enc_ring_test_ib, 2169 .insert_nop = amdgpu_ring_insert_nop, 2170 .insert_end = vcn_v1_0_enc_ring_insert_end, 2171 .pad_ib = amdgpu_ring_generic_pad_ib, 2172 .begin_use = vcn_v1_0_ring_begin_use, 2173 .end_use = vcn_v1_0_ring_end_use, 2174 .emit_wreg = vcn_v1_0_enc_ring_emit_wreg, 2175 .emit_reg_wait = vcn_v1_0_enc_ring_emit_reg_wait, 2176 .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper, 2177 }; 2178 2179 static void vcn_v1_0_set_dec_ring_funcs(struct amdgpu_device *adev) 2180 { 2181 adev->vcn.inst->ring_dec.funcs = &vcn_v1_0_dec_ring_vm_funcs; 2182 } 2183 2184 static void vcn_v1_0_set_enc_ring_funcs(struct amdgpu_device *adev) 2185 { 2186 int i; 2187 2188 for (i = 0; i < adev->vcn.inst[0].num_enc_rings; ++i) 2189 adev->vcn.inst->ring_enc[i].funcs = &vcn_v1_0_enc_ring_vm_funcs; 2190 } 2191 2192 static const struct amdgpu_irq_src_funcs vcn_v1_0_irq_funcs = { 2193 .set = vcn_v1_0_set_interrupt_state, 2194 .process = vcn_v1_0_process_interrupt, 2195 }; 2196 2197 static void vcn_v1_0_set_irq_funcs(struct amdgpu_device *adev) 2198 { 2199 adev->vcn.inst->irq.num_types = adev->vcn.inst[0].num_enc_rings + 2; 2200 adev->vcn.inst->irq.funcs = &vcn_v1_0_irq_funcs; 2201 } 2202 2203 const struct amdgpu_ip_block_version vcn_v1_0_ip_block = { 2204 .type = AMD_IP_BLOCK_TYPE_VCN, 2205 .major = 1, 2206 .minor = 0, 2207 .rev = 0, 2208 .funcs = &vcn_v1_0_ip_funcs, 2209 }; 2210