drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c

   1 /*
   2  * Copyright 2013 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 <[email protected]>
  23  */
  24
  25 #include <linux/firmware.h>
  26 #include <drm/drmP.h>
  27 #include "amdgpu.h"
  28 #include "amdgpu_uvd.h"
  29 #include "cikd.h"
  30
  31 #include "uvd/uvd_4_2_d.h"
  32 #include "uvd/uvd_4_2_sh_mask.h"
  33
  34 #include "oss/oss_2_0_d.h"
  35 #include "oss/oss_2_0_sh_mask.h"
  36
  37 #include "bif/bif_4_1_d.h"
  38
  39 #include "smu/smu_7_0_1_d.h"
  40 #include "smu/smu_7_0_1_sh_mask.h"
  41
  42 static void uvd_v4_2_mc_resume(struct amdgpu_device *adev);
  43 static void uvd_v4_2_set_ring_funcs(struct amdgpu_device *adev);
  44 static void uvd_v4_2_set_irq_funcs(struct amdgpu_device *adev);
  45 static int uvd_v4_2_start(struct amdgpu_device *adev);
  46 static void uvd_v4_2_stop(struct amdgpu_device *adev);
  47 static int uvd_v4_2_set_clockgating_state(void *handle,
  48                                 enum amd_clockgating_state state);
  49 static void uvd_v4_2_set_dcm(struct amdgpu_device *adev,
  50                              bool sw_mode);
  51 /**
  52  * uvd_v4_2_ring_get_rptr - get read pointer
  53  *
  54  * @ring: amdgpu_ring pointer
  55  *
  56  * Returns the current hardware read pointer
  57  */
  58 static uint32_t uvd_v4_2_ring_get_rptr(struct amdgpu_ring *ring)
  59 {
  60         struct amdgpu_device *adev = ring->adev;
  61
  62         return RREG32(mmUVD_RBC_RB_RPTR);
  63 }
  64
  65 /**
  66  * uvd_v4_2_ring_get_wptr - get write pointer
  67  *
  68  * @ring: amdgpu_ring pointer
  69  *
  70  * Returns the current hardware write pointer
  71  */
  72 static uint32_t uvd_v4_2_ring_get_wptr(struct amdgpu_ring *ring)
  73 {
  74         struct amdgpu_device *adev = ring->adev;
  75
  76         return RREG32(mmUVD_RBC_RB_WPTR);
  77 }
  78
  79 /**
  80  * uvd_v4_2_ring_set_wptr - set write pointer
  81  *
  82  * @ring: amdgpu_ring pointer
  83  *
  84  * Commits the write pointer to the hardware
  85  */
  86 static void uvd_v4_2_ring_set_wptr(struct amdgpu_ring *ring)
  87 {
  88         struct amdgpu_device *adev = ring->adev;
  89
  90         WREG32(mmUVD_RBC_RB_WPTR, ring->wptr);
  91 }
  92
  93 static int uvd_v4_2_early_init(void *handle)
  94 {
  95         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
  96
  97         uvd_v4_2_set_ring_funcs(adev);
  98         uvd_v4_2_set_irq_funcs(adev);
  99
 100         return 0;
 101 }
 102
 103 static int uvd_v4_2_sw_init(void *handle)
 104 {
 105         struct amdgpu_ring *ring;
 106         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 107         int r;
 108
 109         /* UVD TRAP */
 110         r = amdgpu_irq_add_id(adev, 124, &adev->uvd.irq);
 111         if (r)
 112                 return r;
 113
 114         r = amdgpu_uvd_sw_init(adev);
 115         if (r)
 116                 return r;
 117
 118         r = amdgpu_uvd_resume(adev);
 119         if (r)
 120                 return r;
 121
 122         ring = &adev->uvd.ring;
 123         sprintf(ring->name, "uvd");
 124         r = amdgpu_ring_init(adev, ring, 512, &adev->uvd.irq, 0);
 125
 126         return r;
 127 }
 128
 129 static int uvd_v4_2_sw_fini(void *handle)
 130 {
 131         int r;
 132         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 133
 134         r = amdgpu_uvd_suspend(adev);
 135         if (r)
 136                 return r;
 137
 138         r = amdgpu_uvd_sw_fini(adev);
 139         if (r)
 140                 return r;
 141
 142         return r;
 143 }
 144 static void uvd_v4_2_enable_mgcg(struct amdgpu_device *adev,
 145                                  bool enable);
 146 /**
 147  * uvd_v4_2_hw_init - start and test UVD block
 148  *
 149  * @adev: amdgpu_device pointer
 150  *
 151  * Initialize the hardware, boot up the VCPU and do some testing
 152  */
 153 static int uvd_v4_2_hw_init(void *handle)
 154 {
 155         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 156         struct amdgpu_ring *ring = &adev->uvd.ring;
 157         uint32_t tmp;
 158         int r;
 159
 160         uvd_v4_2_enable_mgcg(adev, true);
 161         amdgpu_asic_set_uvd_clocks(adev, 10000, 10000);
 162
 163         ring->ready = true;
 164         r = amdgpu_ring_test_ring(ring);
 165         if (r) {
 166                 ring->ready = false;
 167                 goto done;
 168         }
 169
 170         r = amdgpu_ring_alloc(ring, 10);
 171         if (r) {
 172                 DRM_ERROR("amdgpu: ring failed to lock UVD ring (%d).\n", r);
 173                 goto done;
 174         }
 175
 176         tmp = PACKET0(mmUVD_SEMA_WAIT_FAULT_TIMEOUT_CNTL, 0);
 177         amdgpu_ring_write(ring, tmp);
 178         amdgpu_ring_write(ring, 0xFFFFF);
 179
 180         tmp = PACKET0(mmUVD_SEMA_WAIT_INCOMPLETE_TIMEOUT_CNTL, 0);
 181         amdgpu_ring_write(ring, tmp);
 182         amdgpu_ring_write(ring, 0xFFFFF);
 183
 184         tmp = PACKET0(mmUVD_SEMA_SIGNAL_INCOMPLETE_TIMEOUT_CNTL, 0);
 185         amdgpu_ring_write(ring, tmp);
 186         amdgpu_ring_write(ring, 0xFFFFF);
 187
 188         /* Clear timeout status bits */
 189         amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_TIMEOUT_STATUS, 0));
 190         amdgpu_ring_write(ring, 0x8);
 191
 192         amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_CNTL, 0));
 193         amdgpu_ring_write(ring, 3);
 194
 195         amdgpu_ring_commit(ring);
 196
 197 done:
 198         if (!r)
 199                 DRM_INFO("UVD initialized successfully.\n");
 200
 201         return r;
 202 }
 203
 204 /**
 205  * uvd_v4_2_hw_fini - stop the hardware block
 206  *
 207  * @adev: amdgpu_device pointer
 208  *
 209  * Stop the UVD block, mark ring as not ready any more
 210  */
 211 static int uvd_v4_2_hw_fini(void *handle)
 212 {
 213         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 214         struct amdgpu_ring *ring = &adev->uvd.ring;
 215
 216         if (RREG32(mmUVD_STATUS) != 0)
 217                 uvd_v4_2_stop(adev);
 218
 219         ring->ready = false;
 220
 221         return 0;
 222 }
 223
 224 static int uvd_v4_2_suspend(void *handle)
 225 {
 226         int r;
 227         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 228
 229         r = uvd_v4_2_hw_fini(adev);
 230         if (r)
 231                 return r;
 232
 233         r = amdgpu_uvd_suspend(adev);
 234         if (r)
 235                 return r;
 236
 237         return r;
 238 }
 239
 240 static int uvd_v4_2_resume(void *handle)
 241 {
 242         int r;
 243         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 244
 245         r = amdgpu_uvd_resume(adev);
 246         if (r)
 247                 return r;
 248
 249         r = uvd_v4_2_hw_init(adev);
 250         if (r)
 251                 return r;
 252
 253         return r;
 254 }
 255
 256 /**
 257  * uvd_v4_2_start - start UVD block
 258  *
 259  * @adev: amdgpu_device pointer
 260  *
 261  * Setup and start the UVD block
 262  */
 263 static int uvd_v4_2_start(struct amdgpu_device *adev)
 264 {
 265         struct amdgpu_ring *ring = &adev->uvd.ring;
 266         uint32_t rb_bufsz;
 267         int i, j, r;
 268         u32 tmp;
 269         /* disable byte swapping */
 270         u32 lmi_swap_cntl = 0;
 271         u32 mp_swap_cntl = 0;
 272
 273         /* set uvd busy */
 274         WREG32_P(mmUVD_STATUS, 1<<2, ~(1<<2));
 275
 276         uvd_v4_2_set_dcm(adev, true);
 277         WREG32(mmUVD_CGC_GATE, 0);
 278
 279         /* take UVD block out of reset */
 280         WREG32_P(mmSRBM_SOFT_RESET, 0, ~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
 281         mdelay(5);
 282
 283         /* enable VCPU clock */
 284         WREG32(mmUVD_VCPU_CNTL,  1 << 9);
 285
 286         /* disable interupt */
 287         WREG32_P(mmUVD_MASTINT_EN, 0, ~(1 << 1));
 288
 289 #ifdef __BIG_ENDIAN
 290         /* swap (8 in 32) RB and IB */
 291         lmi_swap_cntl = 0xa;
 292         mp_swap_cntl = 0;
 293 #endif
 294         WREG32(mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl);
 295         WREG32(mmUVD_MP_SWAP_CNTL, mp_swap_cntl);
 296         /* initialize UVD memory controller */
 297         WREG32(mmUVD_LMI_CTRL, 0x203108);
 298
 299         tmp = RREG32(mmUVD_MPC_CNTL);
 300         WREG32(mmUVD_MPC_CNTL, tmp | 0x10);
 301
 302         WREG32(mmUVD_MPC_SET_MUXA0, 0x40c2040);
 303         WREG32(mmUVD_MPC_SET_MUXA1, 0x0);
 304         WREG32(mmUVD_MPC_SET_MUXB0, 0x40c2040);
 305         WREG32(mmUVD_MPC_SET_MUXB1, 0x0);
 306         WREG32(mmUVD_MPC_SET_ALU, 0);
 307         WREG32(mmUVD_MPC_SET_MUX, 0x88);
 308
 309         uvd_v4_2_mc_resume(adev);
 310
 311         tmp = RREG32_UVD_CTX(ixUVD_LMI_CACHE_CTRL);
 312         WREG32_UVD_CTX(ixUVD_LMI_CACHE_CTRL, tmp & (~0x10));
 313
 314         /* enable UMC */
 315         WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8));
 316
 317         WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK);
 318
 319         WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
 320
 321         WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
 322
 323         mdelay(10);
 324
 325         for (i = 0; i < 10; ++i) {
 326                 uint32_t status;
 327                 for (j = 0; j < 100; ++j) {
 328                         status = RREG32(mmUVD_STATUS);
 329                         if (status & 2)
 330                                 break;
 331                         mdelay(10);
 332                 }
 333                 r = 0;
 334                 if (status & 2)
 335                         break;
 336
 337                 DRM_ERROR("UVD not responding, trying to reset the VCPU!!!\n");
 338                 WREG32_P(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK,
 339                                 ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
 340                 mdelay(10);
 341                 WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
 342                 mdelay(10);
 343                 r = -1;
 344         }
 345
 346         if (r) {
 347                 DRM_ERROR("UVD not responding, giving up!!!\n");
 348                 return r;
 349         }
 350
 351         /* enable interupt */
 352         WREG32_P(mmUVD_MASTINT_EN, 3<<1, ~(3 << 1));
 353
 354         WREG32_P(mmUVD_STATUS, 0, ~(1<<2));
 355
 356         /* force RBC into idle state */
 357         WREG32(mmUVD_RBC_RB_CNTL, 0x11010101);
 358
 359         /* Set the write pointer delay */
 360         WREG32(mmUVD_RBC_RB_WPTR_CNTL, 0);
 361
 362         /* programm the 4GB memory segment for rptr and ring buffer */
 363         WREG32(mmUVD_LMI_EXT40_ADDR, upper_32_bits(ring->gpu_addr) |
 364                                    (0x7 << 16) | (0x1 << 31));
 365
 366         /* Initialize the ring buffer's read and write pointers */
 367         WREG32(mmUVD_RBC_RB_RPTR, 0x0);
 368
 369         ring->wptr = RREG32(mmUVD_RBC_RB_RPTR);
 370         WREG32(mmUVD_RBC_RB_WPTR, ring->wptr);
 371
 372         /* set the ring address */
 373         WREG32(mmUVD_RBC_RB_BASE, ring->gpu_addr);
 374
 375         /* Set ring buffer size */
 376         rb_bufsz = order_base_2(ring->ring_size);
 377         rb_bufsz = (0x1 << 8) | rb_bufsz;
 378         WREG32_P(mmUVD_RBC_RB_CNTL, rb_bufsz, ~0x11f1f);
 379
 380         return 0;
 381 }
 382
 383 /**
 384  * uvd_v4_2_stop - stop UVD block
 385  *
 386  * @adev: amdgpu_device pointer
 387  *
 388  * stop the UVD block
 389  */
 390 static void uvd_v4_2_stop(struct amdgpu_device *adev)
 391 {
 392         uint32_t i, j;
 393         uint32_t status;
 394
 395         WREG32(mmUVD_RBC_RB_CNTL, 0x11010101);
 396
 397         for (i = 0; i < 10; ++i) {
 398                 for (j = 0; j < 100; ++j) {
 399                         status = RREG32(mmUVD_STATUS);
 400                         if (status & 2)
 401                                 break;
 402                         mdelay(1);
 403                 }
 404                 if (status & 2)
 405                         break;
 406         }
 407
 408         for (i = 0; i < 10; ++i) {
 409                 for (j = 0; j < 100; ++j) {
 410                         status = RREG32(mmUVD_LMI_STATUS);
 411                         if (status & 0xf)
 412                                 break;
 413                         mdelay(1);
 414                 }
 415                 if (status & 0xf)
 416                         break;
 417         }
 418
 419         /* Stall UMC and register bus before resetting VCPU */
 420         WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8));
 421
 422         for (i = 0; i < 10; ++i) {
 423                 for (j = 0; j < 100; ++j) {
 424                         status = RREG32(mmUVD_LMI_STATUS);
 425                         if (status & 0x240)
 426                                 break;
 427                         mdelay(1);
 428                 }
 429                 if (status & 0x240)
 430                         break;
 431         }
 432
 433         WREG32_P(0x3D49, 0, ~(1 << 2));
 434
 435         WREG32_P(mmUVD_VCPU_CNTL, 0, ~(1 << 9));
 436
 437         /* put LMI, VCPU, RBC etc... into reset */
 438         WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__LMI_SOFT_RESET_MASK |
 439                 UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK |
 440                 UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
 441
 442         WREG32(mmUVD_STATUS, 0);
 443
 444         uvd_v4_2_set_dcm(adev, false);
 445 }
 446
 447 /**
 448  * uvd_v4_2_ring_emit_fence - emit an fence & trap command
 449  *
 450  * @ring: amdgpu_ring pointer
 451  * @fence: fence to emit
 452  *
 453  * Write a fence and a trap command to the ring.
 454  */
 455 static void uvd_v4_2_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
 456                                      unsigned flags)
 457 {
 458         WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
 459
 460         amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
 461         amdgpu_ring_write(ring, seq);
 462         amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
 463         amdgpu_ring_write(ring, addr & 0xffffffff);
 464         amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
 465         amdgpu_ring_write(ring, upper_32_bits(addr) & 0xff);
 466         amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
 467         amdgpu_ring_write(ring, 0);
 468
 469         amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
 470         amdgpu_ring_write(ring, 0);
 471         amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
 472         amdgpu_ring_write(ring, 0);
 473         amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
 474         amdgpu_ring_write(ring, 2);
 475 }
 476
 477 /**
 478  * uvd_v4_2_ring_emit_hdp_flush - emit an hdp flush
 479  *
 480  * @ring: amdgpu_ring pointer
 481  *
 482  * Emits an hdp flush.
 483  */
 484 static void uvd_v4_2_ring_emit_hdp_flush(struct amdgpu_ring *ring)
 485 {
 486         amdgpu_ring_write(ring, PACKET0(mmHDP_MEM_COHERENCY_FLUSH_CNTL, 0));
 487         amdgpu_ring_write(ring, 0);
 488 }
 489
 490 /**
 491  * uvd_v4_2_ring_hdp_invalidate - emit an hdp invalidate
 492  *
 493  * @ring: amdgpu_ring pointer
 494  *
 495  * Emits an hdp invalidate.
 496  */
 497 static void uvd_v4_2_ring_emit_hdp_invalidate(struct amdgpu_ring *ring)
 498 {
 499         amdgpu_ring_write(ring, PACKET0(mmHDP_DEBUG0, 0));
 500         amdgpu_ring_write(ring, 1);
 501 }
 502
 503 /**
 504  * uvd_v4_2_ring_test_ring - register write test
 505  *
 506  * @ring: amdgpu_ring pointer
 507  *
 508  * Test if we can successfully write to the context register
 509  */
 510 static int uvd_v4_2_ring_test_ring(struct amdgpu_ring *ring)
 511 {
 512         struct amdgpu_device *adev = ring->adev;
 513         uint32_t tmp = 0;
 514         unsigned i;
 515         int r;
 516
 517         WREG32(mmUVD_CONTEXT_ID, 0xCAFEDEAD);
 518         r = amdgpu_ring_alloc(ring, 3);
 519         if (r) {
 520                 DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n",
 521                           ring->idx, r);
 522                 return r;
 523         }
 524         amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
 525         amdgpu_ring_write(ring, 0xDEADBEEF);
 526         amdgpu_ring_commit(ring);
 527         for (i = 0; i < adev->usec_timeout; i++) {
 528                 tmp = RREG32(mmUVD_CONTEXT_ID);
 529                 if (tmp == 0xDEADBEEF)
 530                         break;
 531                 DRM_UDELAY(1);
 532         }
 533
 534         if (i < adev->usec_timeout) {
 535                 DRM_INFO("ring test on %d succeeded in %d usecs\n",
 536                          ring->idx, i);
 537         } else {
 538                 DRM_ERROR("amdgpu: ring %d test failed (0x%08X)\n",
 539                           ring->idx, tmp);
 540                 r = -EINVAL;
 541         }
 542         return r;
 543 }
 544
 545 /**
 546  * uvd_v4_2_ring_emit_ib - execute indirect buffer
 547  *
 548  * @ring: amdgpu_ring pointer
 549  * @ib: indirect buffer to execute
 550  *
 551  * Write ring commands to execute the indirect buffer
 552  */
 553 static void uvd_v4_2_ring_emit_ib(struct amdgpu_ring *ring,
 554                                   struct amdgpu_ib *ib,
 555                                   unsigned vm_id, bool ctx_switch)
 556 {
 557         amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_BASE, 0));
 558         amdgpu_ring_write(ring, ib->gpu_addr);
 559         amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_SIZE, 0));
 560         amdgpu_ring_write(ring, ib->length_dw);
 561 }
 562
 563 /**
 564  * uvd_v4_2_mc_resume - memory controller programming
 565  *
 566  * @adev: amdgpu_device pointer
 567  *
 568  * Let the UVD memory controller know it's offsets
 569  */
 570 static void uvd_v4_2_mc_resume(struct amdgpu_device *adev)
 571 {
 572         uint64_t addr;
 573         uint32_t size;
 574
 575         /* programm the VCPU memory controller bits 0-27 */
 576         addr = (adev->uvd.gpu_addr + AMDGPU_UVD_FIRMWARE_OFFSET) >> 3;
 577         size = AMDGPU_GPU_PAGE_ALIGN(adev->uvd.fw->size + 4) >> 3;
 578         WREG32(mmUVD_VCPU_CACHE_OFFSET0, addr);
 579         WREG32(mmUVD_VCPU_CACHE_SIZE0, size);
 580
 581         addr += size;
 582         size = AMDGPU_UVD_HEAP_SIZE >> 3;
 583         WREG32(mmUVD_VCPU_CACHE_OFFSET1, addr);
 584         WREG32(mmUVD_VCPU_CACHE_SIZE1, size);
 585
 586         addr += size;
 587         size = (AMDGPU_UVD_STACK_SIZE +
 588                (AMDGPU_UVD_SESSION_SIZE * adev->uvd.max_handles)) >> 3;
 589         WREG32(mmUVD_VCPU_CACHE_OFFSET2, addr);
 590         WREG32(mmUVD_VCPU_CACHE_SIZE2, size);
 591
 592         /* bits 28-31 */
 593         addr = (adev->uvd.gpu_addr >> 28) & 0xF;
 594         WREG32(mmUVD_LMI_ADDR_EXT, (addr << 12) | (addr << 0));
 595
 596         /* bits 32-39 */
 597         addr = (adev->uvd.gpu_addr >> 32) & 0xFF;
 598         WREG32(mmUVD_LMI_EXT40_ADDR, addr | (0x9 << 16) | (0x1 << 31));
 599
 600         WREG32(mmUVD_UDEC_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
 601         WREG32(mmUVD_UDEC_DB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
 602         WREG32(mmUVD_UDEC_DBW_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
 603 }
 604
 605 static void uvd_v4_2_enable_mgcg(struct amdgpu_device *adev,
 606                                  bool enable)
 607 {
 608         u32 orig, data;
 609
 610         if (enable && (adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG)) {
 611                 data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
 612                 data |= 0xfff;
 613                 WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
 614
 615                 orig = data = RREG32(mmUVD_CGC_CTRL);
 616                 data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
 617                 if (orig != data)
 618                         WREG32(mmUVD_CGC_CTRL, data);
 619         } else {
 620                 data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
 621                 data &= ~0xfff;
 622                 WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
 623
 624                 orig = data = RREG32(mmUVD_CGC_CTRL);
 625                 data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
 626                 if (orig != data)
 627                         WREG32(mmUVD_CGC_CTRL, data);
 628         }
 629 }
 630
 631 static void uvd_v4_2_set_dcm(struct amdgpu_device *adev,
 632                              bool sw_mode)
 633 {
 634         u32 tmp, tmp2;
 635
 636         WREG32_FIELD(UVD_CGC_GATE, REGS, 0);
 637
 638         tmp = RREG32(mmUVD_CGC_CTRL);
 639         tmp &= ~(UVD_CGC_CTRL__CLK_OFF_DELAY_MASK | UVD_CGC_CTRL__CLK_GATE_DLY_TIMER_MASK);
 640         tmp |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK |
 641                 (1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT) |
 642                 (4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT);
 643
 644         if (sw_mode) {
 645                 tmp &= ~0x7ffff800;
 646                 tmp2 = UVD_CGC_CTRL2__DYN_OCLK_RAMP_EN_MASK |
 647                         UVD_CGC_CTRL2__DYN_RCLK_RAMP_EN_MASK |
 648                         (7 << UVD_CGC_CTRL2__GATER_DIV_ID__SHIFT);
 649         } else {
 650                 tmp |= 0x7ffff800;
 651                 tmp2 = 0;
 652         }
 653
 654         WREG32(mmUVD_CGC_CTRL, tmp);
 655         WREG32_UVD_CTX(ixUVD_CGC_CTRL2, tmp2);
 656 }
 657
 658 static bool uvd_v4_2_is_idle(void *handle)
 659 {
 660         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 661
 662         return !(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK);
 663 }
 664
 665 static int uvd_v4_2_wait_for_idle(void *handle)
 666 {
 667         unsigned i;
 668         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 669
 670         for (i = 0; i < adev->usec_timeout; i++) {
 671                 if (!(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK))
 672                         return 0;
 673         }
 674         return -ETIMEDOUT;
 675 }
 676
 677 static int uvd_v4_2_soft_reset(void *handle)
 678 {
 679         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 680
 681         uvd_v4_2_stop(adev);
 682
 683         WREG32_P(mmSRBM_SOFT_RESET, SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK,
 684                         ~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
 685         mdelay(5);
 686
 687         return uvd_v4_2_start(adev);
 688 }
 689
 690 static int uvd_v4_2_set_interrupt_state(struct amdgpu_device *adev,
 691                                         struct amdgpu_irq_src *source,
 692                                         unsigned type,
 693                                         enum amdgpu_interrupt_state state)
 694 {
 695         // TODO
 696         return 0;
 697 }
 698
 699 static int uvd_v4_2_process_interrupt(struct amdgpu_device *adev,
 700                                       struct amdgpu_irq_src *source,
 701                                       struct amdgpu_iv_entry *entry)
 702 {
 703         DRM_DEBUG("IH: UVD TRAP\n");
 704         amdgpu_fence_process(&adev->uvd.ring);
 705         return 0;
 706 }
 707
 708 static int uvd_v4_2_set_clockgating_state(void *handle,
 709                                           enum amd_clockgating_state state)
 710 {
 711         return 0;
 712 }
 713
 714 static int uvd_v4_2_set_powergating_state(void *handle,
 715                                           enum amd_powergating_state state)
 716 {
 717         /* This doesn't actually powergate the UVD block.
 718          * That's done in the dpm code via the SMC.  This
 719          * just re-inits the block as necessary.  The actual
 720          * gating still happens in the dpm code.  We should
 721          * revisit this when there is a cleaner line between
 722          * the smc and the hw blocks
 723          */
 724         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 725
 726         if (state == AMD_PG_STATE_GATE) {
 727                 uvd_v4_2_stop(adev);
 728                 if (adev->pg_flags & AMD_PG_SUPPORT_UVD && amdgpu_dpm == 0) {
 729                         if (!(RREG32_SMC(ixCURRENT_PG_STATUS) &
 730                                 CURRENT_PG_STATUS__UVD_PG_STATUS_MASK)) {
 731                                 WREG32(mmUVD_PGFSM_CONFIG, (UVD_PGFSM_CONFIG__UVD_PGFSM_FSM_ADDR_MASK   |
 732                                                         UVD_PGFSM_CONFIG__UVD_PGFSM_POWER_DOWN_MASK |
 733                                                         UVD_PGFSM_CONFIG__UVD_PGFSM_P1_SELECT_MASK));
 734                                 mdelay(20);
 735                         }
 736                 }
 737                 return 0;
 738         } else {
 739                 if (adev->pg_flags & AMD_PG_SUPPORT_UVD && amdgpu_dpm == 0) {
 740                         if (RREG32_SMC(ixCURRENT_PG_STATUS) &
 741                                 CURRENT_PG_STATUS__UVD_PG_STATUS_MASK) {
 742                                 WREG32(mmUVD_PGFSM_CONFIG, (UVD_PGFSM_CONFIG__UVD_PGFSM_FSM_ADDR_MASK   |
 743                                                 UVD_PGFSM_CONFIG__UVD_PGFSM_POWER_UP_MASK |
 744                                                 UVD_PGFSM_CONFIG__UVD_PGFSM_P1_SELECT_MASK));
 745                                 mdelay(30);
 746                         }
 747                 }
 748                 return uvd_v4_2_start(adev);
 749         }
 750 }
 751
 752 static const struct amd_ip_funcs uvd_v4_2_ip_funcs = {
 753         .name = "uvd_v4_2",
 754         .early_init = uvd_v4_2_early_init,
 755         .late_init = NULL,
 756         .sw_init = uvd_v4_2_sw_init,
 757         .sw_fini = uvd_v4_2_sw_fini,
 758         .hw_init = uvd_v4_2_hw_init,
 759         .hw_fini = uvd_v4_2_hw_fini,
 760         .suspend = uvd_v4_2_suspend,
 761         .resume = uvd_v4_2_resume,
 762         .is_idle = uvd_v4_2_is_idle,
 763         .wait_for_idle = uvd_v4_2_wait_for_idle,
 764         .soft_reset = uvd_v4_2_soft_reset,
 765         .set_clockgating_state = uvd_v4_2_set_clockgating_state,
 766         .set_powergating_state = uvd_v4_2_set_powergating_state,
 767 };
 768
 769 static const struct amdgpu_ring_funcs uvd_v4_2_ring_funcs = {
 770         .type = AMDGPU_RING_TYPE_UVD,
 771         .align_mask = 0xf,
 772         .nop = PACKET0(mmUVD_NO_OP, 0),
 773         .get_rptr = uvd_v4_2_ring_get_rptr,
 774         .get_wptr = uvd_v4_2_ring_get_wptr,
 775         .set_wptr = uvd_v4_2_ring_set_wptr,
 776         .parse_cs = amdgpu_uvd_ring_parse_cs,
 777         .emit_frame_size =
 778                 2 + /* uvd_v4_2_ring_emit_hdp_flush */
 779                 2 + /* uvd_v4_2_ring_emit_hdp_invalidate */
 780                 14, /* uvd_v4_2_ring_emit_fence  x1 no user fence */
 781         .emit_ib_size = 4, /* uvd_v4_2_ring_emit_ib */
 782         .emit_ib = uvd_v4_2_ring_emit_ib,
 783         .emit_fence = uvd_v4_2_ring_emit_fence,
 784         .emit_hdp_flush = uvd_v4_2_ring_emit_hdp_flush,
 785         .emit_hdp_invalidate = uvd_v4_2_ring_emit_hdp_invalidate,
 786         .test_ring = uvd_v4_2_ring_test_ring,
 787         .test_ib = amdgpu_uvd_ring_test_ib,
 788         .insert_nop = amdgpu_ring_insert_nop,
 789         .pad_ib = amdgpu_ring_generic_pad_ib,
 790         .begin_use = amdgpu_uvd_ring_begin_use,
 791         .end_use = amdgpu_uvd_ring_end_use,
 792 };
 793
 794 static void uvd_v4_2_set_ring_funcs(struct amdgpu_device *adev)
 795 {
 796         adev->uvd.ring.funcs = &uvd_v4_2_ring_funcs;
 797 }
 798
 799 static const struct amdgpu_irq_src_funcs uvd_v4_2_irq_funcs = {
 800         .set = uvd_v4_2_set_interrupt_state,
 801         .process = uvd_v4_2_process_interrupt,
 802 };
 803
 804 static void uvd_v4_2_set_irq_funcs(struct amdgpu_device *adev)
 805 {
 806         adev->uvd.irq.num_types = 1;
 807         adev->uvd.irq.funcs = &uvd_v4_2_irq_funcs;
 808 }
 809
 810 const struct amdgpu_ip_block_version uvd_v4_2_ip_block =
 811 {
 812                 .type = AMD_IP_BLOCK_TYPE_UVD,
 813                 .major = 4,
 814                 .minor = 2,
 815                 .rev = 0,
 816                 .funcs = &uvd_v4_2_ip_funcs,
 817 };