1 /* 2 * Copyright 2014 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 * Authors: Christian König <christian.koenig@amd.com> 23 */ 24 25 #include <linux/firmware.h> 26 #include <drm/drmP.h> 27 #include "amdgpu.h" 28 #include "amdgpu_uvd.h" 29 #include "vid.h" 30 #include "uvd/uvd_5_0_d.h" 31 #include "uvd/uvd_5_0_sh_mask.h" 32 #include "oss/oss_2_0_d.h" 33 #include "oss/oss_2_0_sh_mask.h" 34 #include "bif/bif_5_0_d.h" 35 #include "vi.h" 36 #include "smu/smu_7_1_2_d.h" 37 #include "smu/smu_7_1_2_sh_mask.h" 38 39 static void uvd_v5_0_set_ring_funcs(struct amdgpu_device *adev); 40 static void uvd_v5_0_set_irq_funcs(struct amdgpu_device *adev); 41 static int uvd_v5_0_start(struct amdgpu_device *adev); 42 static void uvd_v5_0_stop(struct amdgpu_device *adev); 43 static int uvd_v5_0_set_clockgating_state(void *handle, 44 enum amd_clockgating_state state); 45 static void uvd_v5_0_enable_mgcg(struct amdgpu_device *adev, 46 bool enable); 47 /** 48 * uvd_v5_0_ring_get_rptr - get read pointer 49 * 50 * @ring: amdgpu_ring pointer 51 * 52 * Returns the current hardware read pointer 53 */ 54 static uint32_t uvd_v5_0_ring_get_rptr(struct amdgpu_ring *ring) 55 { 56 struct amdgpu_device *adev = ring->adev; 57 58 return RREG32(mmUVD_RBC_RB_RPTR); 59 } 60 61 /** 62 * uvd_v5_0_ring_get_wptr - get write pointer 63 * 64 * @ring: amdgpu_ring pointer 65 * 66 * Returns the current hardware write pointer 67 */ 68 static uint32_t uvd_v5_0_ring_get_wptr(struct amdgpu_ring *ring) 69 { 70 struct amdgpu_device *adev = ring->adev; 71 72 return RREG32(mmUVD_RBC_RB_WPTR); 73 } 74 75 /** 76 * uvd_v5_0_ring_set_wptr - set write pointer 77 * 78 * @ring: amdgpu_ring pointer 79 * 80 * Commits the write pointer to the hardware 81 */ 82 static void uvd_v5_0_ring_set_wptr(struct amdgpu_ring *ring) 83 { 84 struct amdgpu_device *adev = ring->adev; 85 86 WREG32(mmUVD_RBC_RB_WPTR, ring->wptr); 87 } 88 89 static int uvd_v5_0_early_init(void *handle) 90 { 91 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 92 93 uvd_v5_0_set_ring_funcs(adev); 94 uvd_v5_0_set_irq_funcs(adev); 95 96 return 0; 97 } 98 99 static int uvd_v5_0_sw_init(void *handle) 100 { 101 struct amdgpu_ring *ring; 102 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 103 int r; 104 105 /* UVD TRAP */ 106 r = amdgpu_irq_add_id(adev, 124, &adev->uvd.irq); 107 if (r) 108 return r; 109 110 r = amdgpu_uvd_sw_init(adev); 111 if (r) 112 return r; 113 114 r = amdgpu_uvd_resume(adev); 115 if (r) 116 return r; 117 118 ring = &adev->uvd.ring; 119 sprintf(ring->name, "uvd"); 120 r = amdgpu_ring_init(adev, ring, 512, &adev->uvd.irq, 0); 121 122 return r; 123 } 124 125 static int uvd_v5_0_sw_fini(void *handle) 126 { 127 int r; 128 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 129 130 r = amdgpu_uvd_suspend(adev); 131 if (r) 132 return r; 133 134 r = amdgpu_uvd_sw_fini(adev); 135 if (r) 136 return r; 137 138 return r; 139 } 140 141 /** 142 * uvd_v5_0_hw_init - start and test UVD block 143 * 144 * @adev: amdgpu_device pointer 145 * 146 * Initialize the hardware, boot up the VCPU and do some testing 147 */ 148 static int uvd_v5_0_hw_init(void *handle) 149 { 150 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 151 struct amdgpu_ring *ring = &adev->uvd.ring; 152 uint32_t tmp; 153 int r; 154 155 r = uvd_v5_0_start(adev); 156 if (r) 157 goto done; 158 159 ring->ready = true; 160 r = amdgpu_ring_test_ring(ring); 161 if (r) { 162 ring->ready = false; 163 goto done; 164 } 165 166 r = amdgpu_ring_alloc(ring, 10); 167 if (r) { 168 DRM_ERROR("amdgpu: ring failed to lock UVD ring (%d).\n", r); 169 goto done; 170 } 171 172 tmp = PACKET0(mmUVD_SEMA_WAIT_FAULT_TIMEOUT_CNTL, 0); 173 amdgpu_ring_write(ring, tmp); 174 amdgpu_ring_write(ring, 0xFFFFF); 175 176 tmp = PACKET0(mmUVD_SEMA_WAIT_INCOMPLETE_TIMEOUT_CNTL, 0); 177 amdgpu_ring_write(ring, tmp); 178 amdgpu_ring_write(ring, 0xFFFFF); 179 180 tmp = PACKET0(mmUVD_SEMA_SIGNAL_INCOMPLETE_TIMEOUT_CNTL, 0); 181 amdgpu_ring_write(ring, tmp); 182 amdgpu_ring_write(ring, 0xFFFFF); 183 184 /* Clear timeout status bits */ 185 amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_TIMEOUT_STATUS, 0)); 186 amdgpu_ring_write(ring, 0x8); 187 188 amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_CNTL, 0)); 189 amdgpu_ring_write(ring, 3); 190 191 amdgpu_ring_commit(ring); 192 done: 193 if (!r) 194 DRM_INFO("UVD initialized successfully.\n"); 195 196 return r; 197 } 198 199 /** 200 * uvd_v5_0_hw_fini - stop the hardware block 201 * 202 * @adev: amdgpu_device pointer 203 * 204 * Stop the UVD block, mark ring as not ready any more 205 */ 206 static int uvd_v5_0_hw_fini(void *handle) 207 { 208 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 209 struct amdgpu_ring *ring = &adev->uvd.ring; 210 211 uvd_v5_0_stop(adev); 212 ring->ready = false; 213 214 return 0; 215 } 216 217 static int uvd_v5_0_suspend(void *handle) 218 { 219 int r; 220 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 221 222 r = uvd_v5_0_hw_fini(adev); 223 if (r) 224 return r; 225 uvd_v5_0_set_clockgating_state(adev, AMD_CG_STATE_GATE); 226 227 r = amdgpu_uvd_suspend(adev); 228 if (r) 229 return r; 230 231 return r; 232 } 233 234 static int uvd_v5_0_resume(void *handle) 235 { 236 int r; 237 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 238 239 r = amdgpu_uvd_resume(adev); 240 if (r) 241 return r; 242 243 r = uvd_v5_0_hw_init(adev); 244 if (r) 245 return r; 246 247 return r; 248 } 249 250 /** 251 * uvd_v5_0_mc_resume - memory controller programming 252 * 253 * @adev: amdgpu_device pointer 254 * 255 * Let the UVD memory controller know it's offsets 256 */ 257 static void uvd_v5_0_mc_resume(struct amdgpu_device *adev) 258 { 259 uint64_t offset; 260 uint32_t size; 261 262 /* programm memory controller bits 0-27 */ 263 WREG32(mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW, 264 lower_32_bits(adev->uvd.gpu_addr)); 265 WREG32(mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH, 266 upper_32_bits(adev->uvd.gpu_addr)); 267 268 offset = AMDGPU_UVD_FIRMWARE_OFFSET; 269 size = AMDGPU_GPU_PAGE_ALIGN(adev->uvd.fw->size + 4); 270 WREG32(mmUVD_VCPU_CACHE_OFFSET0, offset >> 3); 271 WREG32(mmUVD_VCPU_CACHE_SIZE0, size); 272 273 offset += size; 274 size = AMDGPU_UVD_HEAP_SIZE; 275 WREG32(mmUVD_VCPU_CACHE_OFFSET1, offset >> 3); 276 WREG32(mmUVD_VCPU_CACHE_SIZE1, size); 277 278 offset += size; 279 size = AMDGPU_UVD_STACK_SIZE + 280 (AMDGPU_UVD_SESSION_SIZE * adev->uvd.max_handles); 281 WREG32(mmUVD_VCPU_CACHE_OFFSET2, offset >> 3); 282 WREG32(mmUVD_VCPU_CACHE_SIZE2, size); 283 284 WREG32(mmUVD_UDEC_ADDR_CONFIG, adev->gfx.config.gb_addr_config); 285 WREG32(mmUVD_UDEC_DB_ADDR_CONFIG, adev->gfx.config.gb_addr_config); 286 WREG32(mmUVD_UDEC_DBW_ADDR_CONFIG, adev->gfx.config.gb_addr_config); 287 } 288 289 /** 290 * uvd_v5_0_start - start UVD block 291 * 292 * @adev: amdgpu_device pointer 293 * 294 * Setup and start the UVD block 295 */ 296 static int uvd_v5_0_start(struct amdgpu_device *adev) 297 { 298 struct amdgpu_ring *ring = &adev->uvd.ring; 299 uint32_t rb_bufsz, tmp; 300 uint32_t lmi_swap_cntl; 301 uint32_t mp_swap_cntl; 302 int i, j, r; 303 304 /*disable DPG */ 305 WREG32_P(mmUVD_POWER_STATUS, 0, ~(1 << 2)); 306 307 /* disable byte swapping */ 308 lmi_swap_cntl = 0; 309 mp_swap_cntl = 0; 310 311 uvd_v5_0_mc_resume(adev); 312 313 amdgpu_asic_set_uvd_clocks(adev, 10000, 10000); 314 uvd_v5_0_set_clockgating_state(adev, AMD_CG_STATE_UNGATE); 315 uvd_v5_0_enable_mgcg(adev, true); 316 317 /* disable interupt */ 318 WREG32_P(mmUVD_MASTINT_EN, 0, ~(1 << 1)); 319 320 /* stall UMC and register bus before resetting VCPU */ 321 WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8)); 322 mdelay(1); 323 324 /* put LMI, VCPU, RBC etc... into reset */ 325 WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__LMI_SOFT_RESET_MASK | 326 UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK | UVD_SOFT_RESET__LBSI_SOFT_RESET_MASK | 327 UVD_SOFT_RESET__RBC_SOFT_RESET_MASK | UVD_SOFT_RESET__CSM_SOFT_RESET_MASK | 328 UVD_SOFT_RESET__CXW_SOFT_RESET_MASK | UVD_SOFT_RESET__TAP_SOFT_RESET_MASK | 329 UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK); 330 mdelay(5); 331 332 /* take UVD block out of reset */ 333 WREG32_P(mmSRBM_SOFT_RESET, 0, ~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK); 334 mdelay(5); 335 336 /* initialize UVD memory controller */ 337 WREG32(mmUVD_LMI_CTRL, 0x40 | (1 << 8) | (1 << 13) | 338 (1 << 21) | (1 << 9) | (1 << 20)); 339 340 #ifdef __BIG_ENDIAN 341 /* swap (8 in 32) RB and IB */ 342 lmi_swap_cntl = 0xa; 343 mp_swap_cntl = 0; 344 #endif 345 WREG32(mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl); 346 WREG32(mmUVD_MP_SWAP_CNTL, mp_swap_cntl); 347 348 WREG32(mmUVD_MPC_SET_MUXA0, 0x40c2040); 349 WREG32(mmUVD_MPC_SET_MUXA1, 0x0); 350 WREG32(mmUVD_MPC_SET_MUXB0, 0x40c2040); 351 WREG32(mmUVD_MPC_SET_MUXB1, 0x0); 352 WREG32(mmUVD_MPC_SET_ALU, 0); 353 WREG32(mmUVD_MPC_SET_MUX, 0x88); 354 355 /* take all subblocks out of reset, except VCPU */ 356 WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK); 357 mdelay(5); 358 359 /* enable VCPU clock */ 360 WREG32(mmUVD_VCPU_CNTL, 1 << 9); 361 362 /* enable UMC */ 363 WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8)); 364 365 /* boot up the VCPU */ 366 WREG32(mmUVD_SOFT_RESET, 0); 367 mdelay(10); 368 369 for (i = 0; i < 10; ++i) { 370 uint32_t status; 371 for (j = 0; j < 100; ++j) { 372 status = RREG32(mmUVD_STATUS); 373 if (status & 2) 374 break; 375 mdelay(10); 376 } 377 r = 0; 378 if (status & 2) 379 break; 380 381 DRM_ERROR("UVD not responding, trying to reset the VCPU!!!\n"); 382 WREG32_P(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK, 383 ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK); 384 mdelay(10); 385 WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK); 386 mdelay(10); 387 r = -1; 388 } 389 390 if (r) { 391 DRM_ERROR("UVD not responding, giving up!!!\n"); 392 return r; 393 } 394 /* enable master interrupt */ 395 WREG32_P(mmUVD_MASTINT_EN, 3 << 1, ~(3 << 1)); 396 397 /* clear the bit 4 of UVD_STATUS */ 398 WREG32_P(mmUVD_STATUS, 0, ~(2 << 1)); 399 400 rb_bufsz = order_base_2(ring->ring_size); 401 tmp = 0; 402 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz); 403 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1); 404 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1); 405 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_WPTR_POLL_EN, 0); 406 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1); 407 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1); 408 /* force RBC into idle state */ 409 WREG32(mmUVD_RBC_RB_CNTL, tmp); 410 411 /* set the write pointer delay */ 412 WREG32(mmUVD_RBC_RB_WPTR_CNTL, 0); 413 414 /* set the wb address */ 415 WREG32(mmUVD_RBC_RB_RPTR_ADDR, (upper_32_bits(ring->gpu_addr) >> 2)); 416 417 /* programm the RB_BASE for ring buffer */ 418 WREG32(mmUVD_LMI_RBC_RB_64BIT_BAR_LOW, 419 lower_32_bits(ring->gpu_addr)); 420 WREG32(mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH, 421 upper_32_bits(ring->gpu_addr)); 422 423 /* Initialize the ring buffer's read and write pointers */ 424 WREG32(mmUVD_RBC_RB_RPTR, 0); 425 426 ring->wptr = RREG32(mmUVD_RBC_RB_RPTR); 427 WREG32(mmUVD_RBC_RB_WPTR, ring->wptr); 428 429 WREG32_P(mmUVD_RBC_RB_CNTL, 0, ~UVD_RBC_RB_CNTL__RB_NO_FETCH_MASK); 430 431 return 0; 432 } 433 434 /** 435 * uvd_v5_0_stop - stop UVD block 436 * 437 * @adev: amdgpu_device pointer 438 * 439 * stop the UVD block 440 */ 441 static void uvd_v5_0_stop(struct amdgpu_device *adev) 442 { 443 /* force RBC into idle state */ 444 WREG32(mmUVD_RBC_RB_CNTL, 0x11010101); 445 446 /* Stall UMC and register bus before resetting VCPU */ 447 WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8)); 448 mdelay(1); 449 450 /* put VCPU into reset */ 451 WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK); 452 mdelay(5); 453 454 /* disable VCPU clock */ 455 WREG32(mmUVD_VCPU_CNTL, 0x0); 456 457 /* Unstall UMC and register bus */ 458 WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8)); 459 } 460 461 /** 462 * uvd_v5_0_ring_emit_fence - emit an fence & trap command 463 * 464 * @ring: amdgpu_ring pointer 465 * @fence: fence to emit 466 * 467 * Write a fence and a trap command to the ring. 468 */ 469 static void uvd_v5_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq, 470 unsigned flags) 471 { 472 WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT); 473 474 amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0)); 475 amdgpu_ring_write(ring, seq); 476 amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0)); 477 amdgpu_ring_write(ring, addr & 0xffffffff); 478 amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0)); 479 amdgpu_ring_write(ring, upper_32_bits(addr) & 0xff); 480 amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0)); 481 amdgpu_ring_write(ring, 0); 482 483 amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0)); 484 amdgpu_ring_write(ring, 0); 485 amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0)); 486 amdgpu_ring_write(ring, 0); 487 amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0)); 488 amdgpu_ring_write(ring, 2); 489 } 490 491 /** 492 * uvd_v5_0_ring_emit_hdp_flush - emit an hdp flush 493 * 494 * @ring: amdgpu_ring pointer 495 * 496 * Emits an hdp flush. 497 */ 498 static void uvd_v5_0_ring_emit_hdp_flush(struct amdgpu_ring *ring) 499 { 500 amdgpu_ring_write(ring, PACKET0(mmHDP_MEM_COHERENCY_FLUSH_CNTL, 0)); 501 amdgpu_ring_write(ring, 0); 502 } 503 504 /** 505 * uvd_v5_0_ring_hdp_invalidate - emit an hdp invalidate 506 * 507 * @ring: amdgpu_ring pointer 508 * 509 * Emits an hdp invalidate. 510 */ 511 static void uvd_v5_0_ring_emit_hdp_invalidate(struct amdgpu_ring *ring) 512 { 513 amdgpu_ring_write(ring, PACKET0(mmHDP_DEBUG0, 0)); 514 amdgpu_ring_write(ring, 1); 515 } 516 517 /** 518 * uvd_v5_0_ring_test_ring - register write test 519 * 520 * @ring: amdgpu_ring pointer 521 * 522 * Test if we can successfully write to the context register 523 */ 524 static int uvd_v5_0_ring_test_ring(struct amdgpu_ring *ring) 525 { 526 struct amdgpu_device *adev = ring->adev; 527 uint32_t tmp = 0; 528 unsigned i; 529 int r; 530 531 WREG32(mmUVD_CONTEXT_ID, 0xCAFEDEAD); 532 r = amdgpu_ring_alloc(ring, 3); 533 if (r) { 534 DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n", 535 ring->idx, r); 536 return r; 537 } 538 amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0)); 539 amdgpu_ring_write(ring, 0xDEADBEEF); 540 amdgpu_ring_commit(ring); 541 for (i = 0; i < adev->usec_timeout; i++) { 542 tmp = RREG32(mmUVD_CONTEXT_ID); 543 if (tmp == 0xDEADBEEF) 544 break; 545 DRM_UDELAY(1); 546 } 547 548 if (i < adev->usec_timeout) { 549 DRM_INFO("ring test on %d succeeded in %d usecs\n", 550 ring->idx, i); 551 } else { 552 DRM_ERROR("amdgpu: ring %d test failed (0x%08X)\n", 553 ring->idx, tmp); 554 r = -EINVAL; 555 } 556 return r; 557 } 558 559 /** 560 * uvd_v5_0_ring_emit_ib - execute indirect buffer 561 * 562 * @ring: amdgpu_ring pointer 563 * @ib: indirect buffer to execute 564 * 565 * Write ring commands to execute the indirect buffer 566 */ 567 static void uvd_v5_0_ring_emit_ib(struct amdgpu_ring *ring, 568 struct amdgpu_ib *ib, 569 unsigned vm_id, bool ctx_switch) 570 { 571 amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_LOW, 0)); 572 amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr)); 573 amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH, 0)); 574 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr)); 575 amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_SIZE, 0)); 576 amdgpu_ring_write(ring, ib->length_dw); 577 } 578 579 static bool uvd_v5_0_is_idle(void *handle) 580 { 581 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 582 583 return !(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK); 584 } 585 586 static int uvd_v5_0_wait_for_idle(void *handle) 587 { 588 unsigned i; 589 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 590 591 for (i = 0; i < adev->usec_timeout; i++) { 592 if (!(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK)) 593 return 0; 594 } 595 return -ETIMEDOUT; 596 } 597 598 static int uvd_v5_0_soft_reset(void *handle) 599 { 600 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 601 602 uvd_v5_0_stop(adev); 603 604 WREG32_P(mmSRBM_SOFT_RESET, SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK, 605 ~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK); 606 mdelay(5); 607 608 return uvd_v5_0_start(adev); 609 } 610 611 static int uvd_v5_0_set_interrupt_state(struct amdgpu_device *adev, 612 struct amdgpu_irq_src *source, 613 unsigned type, 614 enum amdgpu_interrupt_state state) 615 { 616 // TODO 617 return 0; 618 } 619 620 static int uvd_v5_0_process_interrupt(struct amdgpu_device *adev, 621 struct amdgpu_irq_src *source, 622 struct amdgpu_iv_entry *entry) 623 { 624 DRM_DEBUG("IH: UVD TRAP\n"); 625 amdgpu_fence_process(&adev->uvd.ring); 626 return 0; 627 } 628 629 static void uvd_v5_0_enable_clock_gating(struct amdgpu_device *adev, bool enable) 630 { 631 uint32_t data1, data3, suvd_flags; 632 633 data1 = RREG32(mmUVD_SUVD_CGC_GATE); 634 data3 = RREG32(mmUVD_CGC_GATE); 635 636 suvd_flags = UVD_SUVD_CGC_GATE__SRE_MASK | 637 UVD_SUVD_CGC_GATE__SIT_MASK | 638 UVD_SUVD_CGC_GATE__SMP_MASK | 639 UVD_SUVD_CGC_GATE__SCM_MASK | 640 UVD_SUVD_CGC_GATE__SDB_MASK; 641 642 if (enable) { 643 data3 |= (UVD_CGC_GATE__SYS_MASK | 644 UVD_CGC_GATE__UDEC_MASK | 645 UVD_CGC_GATE__MPEG2_MASK | 646 UVD_CGC_GATE__RBC_MASK | 647 UVD_CGC_GATE__LMI_MC_MASK | 648 UVD_CGC_GATE__IDCT_MASK | 649 UVD_CGC_GATE__MPRD_MASK | 650 UVD_CGC_GATE__MPC_MASK | 651 UVD_CGC_GATE__LBSI_MASK | 652 UVD_CGC_GATE__LRBBM_MASK | 653 UVD_CGC_GATE__UDEC_RE_MASK | 654 UVD_CGC_GATE__UDEC_CM_MASK | 655 UVD_CGC_GATE__UDEC_IT_MASK | 656 UVD_CGC_GATE__UDEC_DB_MASK | 657 UVD_CGC_GATE__UDEC_MP_MASK | 658 UVD_CGC_GATE__WCB_MASK | 659 UVD_CGC_GATE__JPEG_MASK | 660 UVD_CGC_GATE__SCPU_MASK); 661 /* only in pg enabled, we can gate clock to vcpu*/ 662 if (adev->pg_flags & AMD_PG_SUPPORT_UVD) 663 data3 |= UVD_CGC_GATE__VCPU_MASK; 664 data3 &= ~UVD_CGC_GATE__REGS_MASK; 665 data1 |= suvd_flags; 666 } else { 667 data3 = 0; 668 data1 = 0; 669 } 670 671 WREG32(mmUVD_SUVD_CGC_GATE, data1); 672 WREG32(mmUVD_CGC_GATE, data3); 673 } 674 675 static void uvd_v5_0_set_sw_clock_gating(struct amdgpu_device *adev) 676 { 677 uint32_t data, data2; 678 679 data = RREG32(mmUVD_CGC_CTRL); 680 data2 = RREG32(mmUVD_SUVD_CGC_CTRL); 681 682 683 data &= ~(UVD_CGC_CTRL__CLK_OFF_DELAY_MASK | 684 UVD_CGC_CTRL__CLK_GATE_DLY_TIMER_MASK); 685 686 687 data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK | 688 (1 << REG_FIELD_SHIFT(UVD_CGC_CTRL, CLK_GATE_DLY_TIMER)) | 689 (4 << REG_FIELD_SHIFT(UVD_CGC_CTRL, CLK_OFF_DELAY)); 690 691 data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK | 692 UVD_CGC_CTRL__UDEC_CM_MODE_MASK | 693 UVD_CGC_CTRL__UDEC_IT_MODE_MASK | 694 UVD_CGC_CTRL__UDEC_DB_MODE_MASK | 695 UVD_CGC_CTRL__UDEC_MP_MODE_MASK | 696 UVD_CGC_CTRL__SYS_MODE_MASK | 697 UVD_CGC_CTRL__UDEC_MODE_MASK | 698 UVD_CGC_CTRL__MPEG2_MODE_MASK | 699 UVD_CGC_CTRL__REGS_MODE_MASK | 700 UVD_CGC_CTRL__RBC_MODE_MASK | 701 UVD_CGC_CTRL__LMI_MC_MODE_MASK | 702 UVD_CGC_CTRL__LMI_UMC_MODE_MASK | 703 UVD_CGC_CTRL__IDCT_MODE_MASK | 704 UVD_CGC_CTRL__MPRD_MODE_MASK | 705 UVD_CGC_CTRL__MPC_MODE_MASK | 706 UVD_CGC_CTRL__LBSI_MODE_MASK | 707 UVD_CGC_CTRL__LRBBM_MODE_MASK | 708 UVD_CGC_CTRL__WCB_MODE_MASK | 709 UVD_CGC_CTRL__VCPU_MODE_MASK | 710 UVD_CGC_CTRL__JPEG_MODE_MASK | 711 UVD_CGC_CTRL__SCPU_MODE_MASK); 712 data2 &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK | 713 UVD_SUVD_CGC_CTRL__SIT_MODE_MASK | 714 UVD_SUVD_CGC_CTRL__SMP_MODE_MASK | 715 UVD_SUVD_CGC_CTRL__SCM_MODE_MASK | 716 UVD_SUVD_CGC_CTRL__SDB_MODE_MASK); 717 718 WREG32(mmUVD_CGC_CTRL, data); 719 WREG32(mmUVD_SUVD_CGC_CTRL, data2); 720 } 721 722 #if 0 723 static void uvd_v5_0_set_hw_clock_gating(struct amdgpu_device *adev) 724 { 725 uint32_t data, data1, cgc_flags, suvd_flags; 726 727 data = RREG32(mmUVD_CGC_GATE); 728 data1 = RREG32(mmUVD_SUVD_CGC_GATE); 729 730 cgc_flags = UVD_CGC_GATE__SYS_MASK | 731 UVD_CGC_GATE__UDEC_MASK | 732 UVD_CGC_GATE__MPEG2_MASK | 733 UVD_CGC_GATE__RBC_MASK | 734 UVD_CGC_GATE__LMI_MC_MASK | 735 UVD_CGC_GATE__IDCT_MASK | 736 UVD_CGC_GATE__MPRD_MASK | 737 UVD_CGC_GATE__MPC_MASK | 738 UVD_CGC_GATE__LBSI_MASK | 739 UVD_CGC_GATE__LRBBM_MASK | 740 UVD_CGC_GATE__UDEC_RE_MASK | 741 UVD_CGC_GATE__UDEC_CM_MASK | 742 UVD_CGC_GATE__UDEC_IT_MASK | 743 UVD_CGC_GATE__UDEC_DB_MASK | 744 UVD_CGC_GATE__UDEC_MP_MASK | 745 UVD_CGC_GATE__WCB_MASK | 746 UVD_CGC_GATE__VCPU_MASK | 747 UVD_CGC_GATE__SCPU_MASK; 748 749 suvd_flags = UVD_SUVD_CGC_GATE__SRE_MASK | 750 UVD_SUVD_CGC_GATE__SIT_MASK | 751 UVD_SUVD_CGC_GATE__SMP_MASK | 752 UVD_SUVD_CGC_GATE__SCM_MASK | 753 UVD_SUVD_CGC_GATE__SDB_MASK; 754 755 data |= cgc_flags; 756 data1 |= suvd_flags; 757 758 WREG32(mmUVD_CGC_GATE, data); 759 WREG32(mmUVD_SUVD_CGC_GATE, data1); 760 } 761 #endif 762 763 static void uvd_v5_0_enable_mgcg(struct amdgpu_device *adev, 764 bool enable) 765 { 766 u32 orig, data; 767 768 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG)) { 769 data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL); 770 data |= 0xfff; 771 WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data); 772 773 orig = data = RREG32(mmUVD_CGC_CTRL); 774 data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK; 775 if (orig != data) 776 WREG32(mmUVD_CGC_CTRL, data); 777 } else { 778 data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL); 779 data &= ~0xfff; 780 WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data); 781 782 orig = data = RREG32(mmUVD_CGC_CTRL); 783 data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK; 784 if (orig != data) 785 WREG32(mmUVD_CGC_CTRL, data); 786 } 787 } 788 789 static int uvd_v5_0_set_clockgating_state(void *handle, 790 enum amd_clockgating_state state) 791 { 792 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 793 bool enable = (state == AMD_CG_STATE_GATE) ? true : false; 794 static int curstate = -1; 795 796 if (!(adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG)) 797 return 0; 798 799 if (curstate == state) 800 return 0; 801 802 curstate = state; 803 if (enable) { 804 /* wait for STATUS to clear */ 805 if (uvd_v5_0_wait_for_idle(handle)) 806 return -EBUSY; 807 uvd_v5_0_enable_clock_gating(adev, true); 808 809 /* enable HW gates because UVD is idle */ 810 /* uvd_v5_0_set_hw_clock_gating(adev); */ 811 } else { 812 uvd_v5_0_enable_clock_gating(adev, false); 813 } 814 815 uvd_v5_0_set_sw_clock_gating(adev); 816 return 0; 817 } 818 819 static int uvd_v5_0_set_powergating_state(void *handle, 820 enum amd_powergating_state state) 821 { 822 /* This doesn't actually powergate the UVD block. 823 * That's done in the dpm code via the SMC. This 824 * just re-inits the block as necessary. The actual 825 * gating still happens in the dpm code. We should 826 * revisit this when there is a cleaner line between 827 * the smc and the hw blocks 828 */ 829 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 830 831 if (!(adev->pg_flags & AMD_PG_SUPPORT_UVD)) 832 return 0; 833 834 if (state == AMD_PG_STATE_GATE) { 835 uvd_v5_0_stop(adev); 836 return 0; 837 } else { 838 return uvd_v5_0_start(adev); 839 } 840 } 841 842 static const struct amd_ip_funcs uvd_v5_0_ip_funcs = { 843 .name = "uvd_v5_0", 844 .early_init = uvd_v5_0_early_init, 845 .late_init = NULL, 846 .sw_init = uvd_v5_0_sw_init, 847 .sw_fini = uvd_v5_0_sw_fini, 848 .hw_init = uvd_v5_0_hw_init, 849 .hw_fini = uvd_v5_0_hw_fini, 850 .suspend = uvd_v5_0_suspend, 851 .resume = uvd_v5_0_resume, 852 .is_idle = uvd_v5_0_is_idle, 853 .wait_for_idle = uvd_v5_0_wait_for_idle, 854 .soft_reset = uvd_v5_0_soft_reset, 855 .set_clockgating_state = uvd_v5_0_set_clockgating_state, 856 .set_powergating_state = uvd_v5_0_set_powergating_state, 857 }; 858 859 static const struct amdgpu_ring_funcs uvd_v5_0_ring_funcs = { 860 .type = AMDGPU_RING_TYPE_UVD, 861 .align_mask = 0xf, 862 .nop = PACKET0(mmUVD_NO_OP, 0), 863 .get_rptr = uvd_v5_0_ring_get_rptr, 864 .get_wptr = uvd_v5_0_ring_get_wptr, 865 .set_wptr = uvd_v5_0_ring_set_wptr, 866 .parse_cs = amdgpu_uvd_ring_parse_cs, 867 .emit_frame_size = 868 2 + /* uvd_v5_0_ring_emit_hdp_flush */ 869 2 + /* uvd_v5_0_ring_emit_hdp_invalidate */ 870 14, /* uvd_v5_0_ring_emit_fence x1 no user fence */ 871 .emit_ib_size = 6, /* uvd_v5_0_ring_emit_ib */ 872 .emit_ib = uvd_v5_0_ring_emit_ib, 873 .emit_fence = uvd_v5_0_ring_emit_fence, 874 .emit_hdp_flush = uvd_v5_0_ring_emit_hdp_flush, 875 .emit_hdp_invalidate = uvd_v5_0_ring_emit_hdp_invalidate, 876 .test_ring = uvd_v5_0_ring_test_ring, 877 .test_ib = amdgpu_uvd_ring_test_ib, 878 .insert_nop = amdgpu_ring_insert_nop, 879 .pad_ib = amdgpu_ring_generic_pad_ib, 880 .begin_use = amdgpu_uvd_ring_begin_use, 881 .end_use = amdgpu_uvd_ring_end_use, 882 }; 883 884 static void uvd_v5_0_set_ring_funcs(struct amdgpu_device *adev) 885 { 886 adev->uvd.ring.funcs = &uvd_v5_0_ring_funcs; 887 } 888 889 static const struct amdgpu_irq_src_funcs uvd_v5_0_irq_funcs = { 890 .set = uvd_v5_0_set_interrupt_state, 891 .process = uvd_v5_0_process_interrupt, 892 }; 893 894 static void uvd_v5_0_set_irq_funcs(struct amdgpu_device *adev) 895 { 896 adev->uvd.irq.num_types = 1; 897 adev->uvd.irq.funcs = &uvd_v5_0_irq_funcs; 898 } 899 900 const struct amdgpu_ip_block_version uvd_v5_0_ip_block = 901 { 902 .type = AMD_IP_BLOCK_TYPE_UVD, 903 .major = 5, 904 .minor = 0, 905 .rev = 0, 906 .funcs = &uvd_v5_0_ip_funcs, 907 }; 908