drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c

   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 uint64_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 uint64_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, lower_32_bits(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, AMDGPU_IH_CLIENTID_LEGACY, 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         return amdgpu_uvd_sw_fini(adev);
 135 }
 136
 137 /**
 138  * uvd_v5_0_hw_init - start and test UVD block
 139  *
 140  * @adev: amdgpu_device pointer
 141  *
 142  * Initialize the hardware, boot up the VCPU and do some testing
 143  */
 144 static int uvd_v5_0_hw_init(void *handle)
 145 {
 146         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 147         struct amdgpu_ring *ring = &adev->uvd.ring;
 148         uint32_t tmp;
 149         int r;
 150
 151         amdgpu_asic_set_uvd_clocks(adev, 10000, 10000);
 152         uvd_v5_0_set_clockgating_state(adev, AMD_CG_STATE_UNGATE);
 153         uvd_v5_0_enable_mgcg(adev, true);
 154
 155         ring->ready = true;
 156         r = amdgpu_ring_test_ring(ring);
 157         if (r) {
 158                 ring->ready = false;
 159                 goto done;
 160         }
 161
 162         r = amdgpu_ring_alloc(ring, 10);
 163         if (r) {
 164                 DRM_ERROR("amdgpu: ring failed to lock UVD ring (%d).\n", r);
 165                 goto done;
 166         }
 167
 168         tmp = PACKET0(mmUVD_SEMA_WAIT_FAULT_TIMEOUT_CNTL, 0);
 169         amdgpu_ring_write(ring, tmp);
 170         amdgpu_ring_write(ring, 0xFFFFF);
 171
 172         tmp = PACKET0(mmUVD_SEMA_WAIT_INCOMPLETE_TIMEOUT_CNTL, 0);
 173         amdgpu_ring_write(ring, tmp);
 174         amdgpu_ring_write(ring, 0xFFFFF);
 175
 176         tmp = PACKET0(mmUVD_SEMA_SIGNAL_INCOMPLETE_TIMEOUT_CNTL, 0);
 177         amdgpu_ring_write(ring, tmp);
 178         amdgpu_ring_write(ring, 0xFFFFF);
 179
 180         /* Clear timeout status bits */
 181         amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_TIMEOUT_STATUS, 0));
 182         amdgpu_ring_write(ring, 0x8);
 183
 184         amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_CNTL, 0));
 185         amdgpu_ring_write(ring, 3);
 186
 187         amdgpu_ring_commit(ring);
 188
 189 done:
 190         if (!r)
 191                 DRM_INFO("UVD initialized successfully.\n");
 192
 193         return r;
 194
 195 }
 196
 197 /**
 198  * uvd_v5_0_hw_fini - stop the hardware block
 199  *
 200  * @adev: amdgpu_device pointer
 201  *
 202  * Stop the UVD block, mark ring as not ready any more
 203  */
 204 static int uvd_v5_0_hw_fini(void *handle)
 205 {
 206         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 207         struct amdgpu_ring *ring = &adev->uvd.ring;
 208
 209         if (RREG32(mmUVD_STATUS) != 0)
 210                 uvd_v5_0_stop(adev);
 211
 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         return amdgpu_uvd_suspend(adev);
 228 }
 229
 230 static int uvd_v5_0_resume(void *handle)
 231 {
 232         int r;
 233         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 234
 235         r = amdgpu_uvd_resume(adev);
 236         if (r)
 237                 return r;
 238
 239         return uvd_v5_0_hw_init(adev);
 240 }
 241
 242 /**
 243  * uvd_v5_0_mc_resume - memory controller programming
 244  *
 245  * @adev: amdgpu_device pointer
 246  *
 247  * Let the UVD memory controller know it's offsets
 248  */
 249 static void uvd_v5_0_mc_resume(struct amdgpu_device *adev)
 250 {
 251         uint64_t offset;
 252         uint32_t size;
 253
 254         /* programm memory controller bits 0-27 */
 255         WREG32(mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
 256                         lower_32_bits(adev->uvd.gpu_addr));
 257         WREG32(mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
 258                         upper_32_bits(adev->uvd.gpu_addr));
 259
 260         offset = AMDGPU_UVD_FIRMWARE_OFFSET;
 261         size = AMDGPU_GPU_PAGE_ALIGN(adev->uvd.fw->size + 4);
 262         WREG32(mmUVD_VCPU_CACHE_OFFSET0, offset >> 3);
 263         WREG32(mmUVD_VCPU_CACHE_SIZE0, size);
 264
 265         offset += size;
 266         size = AMDGPU_UVD_HEAP_SIZE;
 267         WREG32(mmUVD_VCPU_CACHE_OFFSET1, offset >> 3);
 268         WREG32(mmUVD_VCPU_CACHE_SIZE1, size);
 269
 270         offset += size;
 271         size = AMDGPU_UVD_STACK_SIZE +
 272                (AMDGPU_UVD_SESSION_SIZE * adev->uvd.max_handles);
 273         WREG32(mmUVD_VCPU_CACHE_OFFSET2, offset >> 3);
 274         WREG32(mmUVD_VCPU_CACHE_SIZE2, size);
 275
 276         WREG32(mmUVD_UDEC_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
 277         WREG32(mmUVD_UDEC_DB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
 278         WREG32(mmUVD_UDEC_DBW_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
 279 }
 280
 281 /**
 282  * uvd_v5_0_start - start UVD block
 283  *
 284  * @adev: amdgpu_device pointer
 285  *
 286  * Setup and start the UVD block
 287  */
 288 static int uvd_v5_0_start(struct amdgpu_device *adev)
 289 {
 290         struct amdgpu_ring *ring = &adev->uvd.ring;
 291         uint32_t rb_bufsz, tmp;
 292         uint32_t lmi_swap_cntl;
 293         uint32_t mp_swap_cntl;
 294         int i, j, r;
 295
 296         /*disable DPG */
 297         WREG32_P(mmUVD_POWER_STATUS, 0, ~(1 << 2));
 298
 299         /* disable byte swapping */
 300         lmi_swap_cntl = 0;
 301         mp_swap_cntl = 0;
 302
 303         uvd_v5_0_mc_resume(adev);
 304
 305         /* disable interupt */
 306         WREG32_P(mmUVD_MASTINT_EN, 0, ~(1 << 1));
 307
 308         /* stall UMC and register bus before resetting VCPU */
 309         WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8));
 310         mdelay(1);
 311
 312         /* put LMI, VCPU, RBC etc... into reset */
 313         WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__LMI_SOFT_RESET_MASK |
 314                 UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK | UVD_SOFT_RESET__LBSI_SOFT_RESET_MASK |
 315                 UVD_SOFT_RESET__RBC_SOFT_RESET_MASK | UVD_SOFT_RESET__CSM_SOFT_RESET_MASK |
 316                 UVD_SOFT_RESET__CXW_SOFT_RESET_MASK | UVD_SOFT_RESET__TAP_SOFT_RESET_MASK |
 317                 UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
 318         mdelay(5);
 319
 320         /* take UVD block out of reset */
 321         WREG32_P(mmSRBM_SOFT_RESET, 0, ~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
 322         mdelay(5);
 323
 324         /* initialize UVD memory controller */
 325         WREG32(mmUVD_LMI_CTRL, 0x40 | (1 << 8) | (1 << 13) |
 326                              (1 << 21) | (1 << 9) | (1 << 20));
 327
 328 #ifdef __BIG_ENDIAN
 329         /* swap (8 in 32) RB and IB */
 330         lmi_swap_cntl = 0xa;
 331         mp_swap_cntl = 0;
 332 #endif
 333         WREG32(mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl);
 334         WREG32(mmUVD_MP_SWAP_CNTL, mp_swap_cntl);
 335
 336         WREG32(mmUVD_MPC_SET_MUXA0, 0x40c2040);
 337         WREG32(mmUVD_MPC_SET_MUXA1, 0x0);
 338         WREG32(mmUVD_MPC_SET_MUXB0, 0x40c2040);
 339         WREG32(mmUVD_MPC_SET_MUXB1, 0x0);
 340         WREG32(mmUVD_MPC_SET_ALU, 0);
 341         WREG32(mmUVD_MPC_SET_MUX, 0x88);
 342
 343         /* take all subblocks out of reset, except VCPU */
 344         WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
 345         mdelay(5);
 346
 347         /* enable VCPU clock */
 348         WREG32(mmUVD_VCPU_CNTL,  1 << 9);
 349
 350         /* enable UMC */
 351         WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8));
 352
 353         /* boot up the VCPU */
 354         WREG32(mmUVD_SOFT_RESET, 0);
 355         mdelay(10);
 356
 357         for (i = 0; i < 10; ++i) {
 358                 uint32_t status;
 359                 for (j = 0; j < 100; ++j) {
 360                         status = RREG32(mmUVD_STATUS);
 361                         if (status & 2)
 362                                 break;
 363                         mdelay(10);
 364                 }
 365                 r = 0;
 366                 if (status & 2)
 367                         break;
 368
 369                 DRM_ERROR("UVD not responding, trying to reset the VCPU!!!\n");
 370                 WREG32_P(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK,
 371                                 ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
 372                 mdelay(10);
 373                 WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
 374                 mdelay(10);
 375                 r = -1;
 376         }
 377
 378         if (r) {
 379                 DRM_ERROR("UVD not responding, giving up!!!\n");
 380                 return r;
 381         }
 382         /* enable master interrupt */
 383         WREG32_P(mmUVD_MASTINT_EN, 3 << 1, ~(3 << 1));
 384
 385         /* clear the bit 4 of UVD_STATUS */
 386         WREG32_P(mmUVD_STATUS, 0, ~(2 << 1));
 387
 388         rb_bufsz = order_base_2(ring->ring_size);
 389         tmp = 0;
 390         tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
 391         tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
 392         tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
 393         tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_WPTR_POLL_EN, 0);
 394         tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
 395         tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
 396         /* force RBC into idle state */
 397         WREG32(mmUVD_RBC_RB_CNTL, tmp);
 398
 399         /* set the write pointer delay */
 400         WREG32(mmUVD_RBC_RB_WPTR_CNTL, 0);
 401
 402         /* set the wb address */
 403         WREG32(mmUVD_RBC_RB_RPTR_ADDR, (upper_32_bits(ring->gpu_addr) >> 2));
 404
 405         /* programm the RB_BASE for ring buffer */
 406         WREG32(mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
 407                         lower_32_bits(ring->gpu_addr));
 408         WREG32(mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
 409                         upper_32_bits(ring->gpu_addr));
 410
 411         /* Initialize the ring buffer's read and write pointers */
 412         WREG32(mmUVD_RBC_RB_RPTR, 0);
 413
 414         ring->wptr = RREG32(mmUVD_RBC_RB_RPTR);
 415         WREG32(mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
 416
 417         WREG32_P(mmUVD_RBC_RB_CNTL, 0, ~UVD_RBC_RB_CNTL__RB_NO_FETCH_MASK);
 418
 419         return 0;
 420 }
 421
 422 /**
 423  * uvd_v5_0_stop - stop UVD block
 424  *
 425  * @adev: amdgpu_device pointer
 426  *
 427  * stop the UVD block
 428  */
 429 static void uvd_v5_0_stop(struct amdgpu_device *adev)
 430 {
 431         /* force RBC into idle state */
 432         WREG32(mmUVD_RBC_RB_CNTL, 0x11010101);
 433
 434         /* Stall UMC and register bus before resetting VCPU */
 435         WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8));
 436         mdelay(1);
 437
 438         /* put VCPU into reset */
 439         WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
 440         mdelay(5);
 441
 442         /* disable VCPU clock */
 443         WREG32(mmUVD_VCPU_CNTL, 0x0);
 444
 445         /* Unstall UMC and register bus */
 446         WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8));
 447
 448         WREG32(mmUVD_STATUS, 0);
 449 }
 450
 451 /**
 452  * uvd_v5_0_ring_emit_fence - emit an fence & trap command
 453  *
 454  * @ring: amdgpu_ring pointer
 455  * @fence: fence to emit
 456  *
 457  * Write a fence and a trap command to the ring.
 458  */
 459 static void uvd_v5_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
 460                                      unsigned flags)
 461 {
 462         WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
 463
 464         amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
 465         amdgpu_ring_write(ring, seq);
 466         amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
 467         amdgpu_ring_write(ring, addr & 0xffffffff);
 468         amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
 469         amdgpu_ring_write(ring, upper_32_bits(addr) & 0xff);
 470         amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
 471         amdgpu_ring_write(ring, 0);
 472
 473         amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
 474         amdgpu_ring_write(ring, 0);
 475         amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
 476         amdgpu_ring_write(ring, 0);
 477         amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
 478         amdgpu_ring_write(ring, 2);
 479 }
 480
 481 /**
 482  * uvd_v5_0_ring_emit_hdp_flush - emit an hdp flush
 483  *
 484  * @ring: amdgpu_ring pointer
 485  *
 486  * Emits an hdp flush.
 487  */
 488 static void uvd_v5_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
 489 {
 490         amdgpu_ring_write(ring, PACKET0(mmHDP_MEM_COHERENCY_FLUSH_CNTL, 0));
 491         amdgpu_ring_write(ring, 0);
 492 }
 493
 494 /**
 495  * uvd_v5_0_ring_hdp_invalidate - emit an hdp invalidate
 496  *
 497  * @ring: amdgpu_ring pointer
 498  *
 499  * Emits an hdp invalidate.
 500  */
 501 static void uvd_v5_0_ring_emit_hdp_invalidate(struct amdgpu_ring *ring)
 502 {
 503         amdgpu_ring_write(ring, PACKET0(mmHDP_DEBUG0, 0));
 504         amdgpu_ring_write(ring, 1);
 505 }
 506
 507 /**
 508  * uvd_v5_0_ring_test_ring - register write test
 509  *
 510  * @ring: amdgpu_ring pointer
 511  *
 512  * Test if we can successfully write to the context register
 513  */
 514 static int uvd_v5_0_ring_test_ring(struct amdgpu_ring *ring)
 515 {
 516         struct amdgpu_device *adev = ring->adev;
 517         uint32_t tmp = 0;
 518         unsigned i;
 519         int r;
 520
 521         WREG32(mmUVD_CONTEXT_ID, 0xCAFEDEAD);
 522         r = amdgpu_ring_alloc(ring, 3);
 523         if (r) {
 524                 DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n",
 525                           ring->idx, r);
 526                 return r;
 527         }
 528         amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
 529         amdgpu_ring_write(ring, 0xDEADBEEF);
 530         amdgpu_ring_commit(ring);
 531         for (i = 0; i < adev->usec_timeout; i++) {
 532                 tmp = RREG32(mmUVD_CONTEXT_ID);
 533                 if (tmp == 0xDEADBEEF)
 534                         break;
 535                 DRM_UDELAY(1);
 536         }
 537
 538         if (i < adev->usec_timeout) {
 539                 DRM_INFO("ring test on %d succeeded in %d usecs\n",
 540                          ring->idx, i);
 541         } else {
 542                 DRM_ERROR("amdgpu: ring %d test failed (0x%08X)\n",
 543                           ring->idx, tmp);
 544                 r = -EINVAL;
 545         }
 546         return r;
 547 }
 548
 549 /**
 550  * uvd_v5_0_ring_emit_ib - execute indirect buffer
 551  *
 552  * @ring: amdgpu_ring pointer
 553  * @ib: indirect buffer to execute
 554  *
 555  * Write ring commands to execute the indirect buffer
 556  */
 557 static void uvd_v5_0_ring_emit_ib(struct amdgpu_ring *ring,
 558                                   struct amdgpu_ib *ib,
 559                                   unsigned vm_id, bool ctx_switch)
 560 {
 561         amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_LOW, 0));
 562         amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
 563         amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH, 0));
 564         amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
 565         amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_SIZE, 0));
 566         amdgpu_ring_write(ring, ib->length_dw);
 567 }
 568
 569 static bool uvd_v5_0_is_idle(void *handle)
 570 {
 571         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 572
 573         return !(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK);
 574 }
 575
 576 static int uvd_v5_0_wait_for_idle(void *handle)
 577 {
 578         unsigned i;
 579         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 580
 581         for (i = 0; i < adev->usec_timeout; i++) {
 582                 if (!(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK))
 583                         return 0;
 584         }
 585         return -ETIMEDOUT;
 586 }
 587
 588 static int uvd_v5_0_soft_reset(void *handle)
 589 {
 590         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 591
 592         uvd_v5_0_stop(adev);
 593
 594         WREG32_P(mmSRBM_SOFT_RESET, SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK,
 595                         ~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
 596         mdelay(5);
 597
 598         return uvd_v5_0_start(adev);
 599 }
 600
 601 static int uvd_v5_0_set_interrupt_state(struct amdgpu_device *adev,
 602                                         struct amdgpu_irq_src *source,
 603                                         unsigned type,
 604                                         enum amdgpu_interrupt_state state)
 605 {
 606         // TODO
 607         return 0;
 608 }
 609
 610 static int uvd_v5_0_process_interrupt(struct amdgpu_device *adev,
 611                                       struct amdgpu_irq_src *source,
 612                                       struct amdgpu_iv_entry *entry)
 613 {
 614         DRM_DEBUG("IH: UVD TRAP\n");
 615         amdgpu_fence_process(&adev->uvd.ring);
 616         return 0;
 617 }
 618
 619 static void uvd_v5_0_enable_clock_gating(struct amdgpu_device *adev, bool enable)
 620 {
 621         uint32_t data1, data3, suvd_flags;
 622
 623         data1 = RREG32(mmUVD_SUVD_CGC_GATE);
 624         data3 = RREG32(mmUVD_CGC_GATE);
 625
 626         suvd_flags = UVD_SUVD_CGC_GATE__SRE_MASK |
 627                      UVD_SUVD_CGC_GATE__SIT_MASK |
 628                      UVD_SUVD_CGC_GATE__SMP_MASK |
 629                      UVD_SUVD_CGC_GATE__SCM_MASK |
 630                      UVD_SUVD_CGC_GATE__SDB_MASK;
 631
 632         if (enable) {
 633                 data3 |= (UVD_CGC_GATE__SYS_MASK     |
 634                         UVD_CGC_GATE__UDEC_MASK      |
 635                         UVD_CGC_GATE__MPEG2_MASK     |
 636                         UVD_CGC_GATE__RBC_MASK       |
 637                         UVD_CGC_GATE__LMI_MC_MASK    |
 638                         UVD_CGC_GATE__IDCT_MASK      |
 639                         UVD_CGC_GATE__MPRD_MASK      |
 640                         UVD_CGC_GATE__MPC_MASK       |
 641                         UVD_CGC_GATE__LBSI_MASK      |
 642                         UVD_CGC_GATE__LRBBM_MASK     |
 643                         UVD_CGC_GATE__UDEC_RE_MASK   |
 644                         UVD_CGC_GATE__UDEC_CM_MASK   |
 645                         UVD_CGC_GATE__UDEC_IT_MASK   |
 646                         UVD_CGC_GATE__UDEC_DB_MASK   |
 647                         UVD_CGC_GATE__UDEC_MP_MASK   |
 648                         UVD_CGC_GATE__WCB_MASK       |
 649                         UVD_CGC_GATE__JPEG_MASK      |
 650                         UVD_CGC_GATE__SCPU_MASK);
 651                 /* only in pg enabled, we can gate clock to vcpu*/
 652                 if (adev->pg_flags & AMD_PG_SUPPORT_UVD)
 653                         data3 |= UVD_CGC_GATE__VCPU_MASK;
 654                 data3 &= ~UVD_CGC_GATE__REGS_MASK;
 655                 data1 |= suvd_flags;
 656         } else {
 657                 data3 = 0;
 658                 data1 = 0;
 659         }
 660
 661         WREG32(mmUVD_SUVD_CGC_GATE, data1);
 662         WREG32(mmUVD_CGC_GATE, data3);
 663 }
 664
 665 static void uvd_v5_0_set_sw_clock_gating(struct amdgpu_device *adev)
 666 {
 667         uint32_t data, data2;
 668
 669         data = RREG32(mmUVD_CGC_CTRL);
 670         data2 = RREG32(mmUVD_SUVD_CGC_CTRL);
 671
 672
 673         data &= ~(UVD_CGC_CTRL__CLK_OFF_DELAY_MASK |
 674                   UVD_CGC_CTRL__CLK_GATE_DLY_TIMER_MASK);
 675
 676
 677         data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK |
 678                 (1 << REG_FIELD_SHIFT(UVD_CGC_CTRL, CLK_GATE_DLY_TIMER)) |
 679                 (4 << REG_FIELD_SHIFT(UVD_CGC_CTRL, CLK_OFF_DELAY));
 680
 681         data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
 682                         UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
 683                         UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
 684                         UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
 685                         UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
 686                         UVD_CGC_CTRL__SYS_MODE_MASK |
 687                         UVD_CGC_CTRL__UDEC_MODE_MASK |
 688                         UVD_CGC_CTRL__MPEG2_MODE_MASK |
 689                         UVD_CGC_CTRL__REGS_MODE_MASK |
 690                         UVD_CGC_CTRL__RBC_MODE_MASK |
 691                         UVD_CGC_CTRL__LMI_MC_MODE_MASK |
 692                         UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
 693                         UVD_CGC_CTRL__IDCT_MODE_MASK |
 694                         UVD_CGC_CTRL__MPRD_MODE_MASK |
 695                         UVD_CGC_CTRL__MPC_MODE_MASK |
 696                         UVD_CGC_CTRL__LBSI_MODE_MASK |
 697                         UVD_CGC_CTRL__LRBBM_MODE_MASK |
 698                         UVD_CGC_CTRL__WCB_MODE_MASK |
 699                         UVD_CGC_CTRL__VCPU_MODE_MASK |
 700                         UVD_CGC_CTRL__JPEG_MODE_MASK |
 701                         UVD_CGC_CTRL__SCPU_MODE_MASK);
 702         data2 &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK |
 703                         UVD_SUVD_CGC_CTRL__SIT_MODE_MASK |
 704                         UVD_SUVD_CGC_CTRL__SMP_MODE_MASK |
 705                         UVD_SUVD_CGC_CTRL__SCM_MODE_MASK |
 706                         UVD_SUVD_CGC_CTRL__SDB_MODE_MASK);
 707
 708         WREG32(mmUVD_CGC_CTRL, data);
 709         WREG32(mmUVD_SUVD_CGC_CTRL, data2);
 710 }
 711
 712 #if 0
 713 static void uvd_v5_0_set_hw_clock_gating(struct amdgpu_device *adev)
 714 {
 715         uint32_t data, data1, cgc_flags, suvd_flags;
 716
 717         data = RREG32(mmUVD_CGC_GATE);
 718         data1 = RREG32(mmUVD_SUVD_CGC_GATE);
 719
 720         cgc_flags = UVD_CGC_GATE__SYS_MASK |
 721                                 UVD_CGC_GATE__UDEC_MASK |
 722                                 UVD_CGC_GATE__MPEG2_MASK |
 723                                 UVD_CGC_GATE__RBC_MASK |
 724                                 UVD_CGC_GATE__LMI_MC_MASK |
 725                                 UVD_CGC_GATE__IDCT_MASK |
 726                                 UVD_CGC_GATE__MPRD_MASK |
 727                                 UVD_CGC_GATE__MPC_MASK |
 728                                 UVD_CGC_GATE__LBSI_MASK |
 729                                 UVD_CGC_GATE__LRBBM_MASK |
 730                                 UVD_CGC_GATE__UDEC_RE_MASK |
 731                                 UVD_CGC_GATE__UDEC_CM_MASK |
 732                                 UVD_CGC_GATE__UDEC_IT_MASK |
 733                                 UVD_CGC_GATE__UDEC_DB_MASK |
 734                                 UVD_CGC_GATE__UDEC_MP_MASK |
 735                                 UVD_CGC_GATE__WCB_MASK |
 736                                 UVD_CGC_GATE__VCPU_MASK |
 737                                 UVD_CGC_GATE__SCPU_MASK;
 738
 739         suvd_flags = UVD_SUVD_CGC_GATE__SRE_MASK |
 740                                 UVD_SUVD_CGC_GATE__SIT_MASK |
 741                                 UVD_SUVD_CGC_GATE__SMP_MASK |
 742                                 UVD_SUVD_CGC_GATE__SCM_MASK |
 743                                 UVD_SUVD_CGC_GATE__SDB_MASK;
 744
 745         data |= cgc_flags;
 746         data1 |= suvd_flags;
 747
 748         WREG32(mmUVD_CGC_GATE, data);
 749         WREG32(mmUVD_SUVD_CGC_GATE, data1);
 750 }
 751 #endif
 752
 753 static void uvd_v5_0_enable_mgcg(struct amdgpu_device *adev,
 754                                  bool enable)
 755 {
 756         u32 orig, data;
 757
 758         if (enable && (adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG)) {
 759                 data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
 760                 data |= 0xfff;
 761                 WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
 762
 763                 orig = data = RREG32(mmUVD_CGC_CTRL);
 764                 data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
 765                 if (orig != data)
 766                         WREG32(mmUVD_CGC_CTRL, data);
 767         } else {
 768                 data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
 769                 data &= ~0xfff;
 770                 WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
 771
 772                 orig = data = RREG32(mmUVD_CGC_CTRL);
 773                 data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
 774                 if (orig != data)
 775                         WREG32(mmUVD_CGC_CTRL, data);
 776         }
 777 }
 778
 779 static int uvd_v5_0_set_clockgating_state(void *handle,
 780                                           enum amd_clockgating_state state)
 781 {
 782         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 783         bool enable = (state == AMD_CG_STATE_GATE) ? true : false;
 784
 785         if (enable) {
 786                 /* wait for STATUS to clear */
 787                 if (uvd_v5_0_wait_for_idle(handle))
 788                         return -EBUSY;
 789                 uvd_v5_0_enable_clock_gating(adev, true);
 790
 791                 /* enable HW gates because UVD is idle */
 792 /*              uvd_v5_0_set_hw_clock_gating(adev); */
 793         } else {
 794                 uvd_v5_0_enable_clock_gating(adev, false);
 795         }
 796
 797         uvd_v5_0_set_sw_clock_gating(adev);
 798         return 0;
 799 }
 800
 801 static int uvd_v5_0_set_powergating_state(void *handle,
 802                                           enum amd_powergating_state state)
 803 {
 804         /* This doesn't actually powergate the UVD block.
 805          * That's done in the dpm code via the SMC.  This
 806          * just re-inits the block as necessary.  The actual
 807          * gating still happens in the dpm code.  We should
 808          * revisit this when there is a cleaner line between
 809          * the smc and the hw blocks
 810          */
 811         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 812         int ret = 0;
 813
 814         if (state == AMD_PG_STATE_GATE) {
 815                 uvd_v5_0_stop(adev);
 816         } else {
 817                 ret = uvd_v5_0_start(adev);
 818                 if (ret)
 819                         goto out;
 820         }
 821
 822 out:
 823         return ret;
 824 }
 825
 826 static void uvd_v5_0_get_clockgating_state(void *handle, u32 *flags)
 827 {
 828         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 829         int data;
 830
 831         mutex_lock(&adev->pm.mutex);
 832
 833         if (RREG32_SMC(ixCURRENT_PG_STATUS) &
 834                                 CURRENT_PG_STATUS__UVD_PG_STATUS_MASK) {
 835                 DRM_INFO("Cannot get clockgating state when UVD is powergated.\n");
 836                 goto out;
 837         }
 838
 839         /* AMD_CG_SUPPORT_UVD_MGCG */
 840         data = RREG32(mmUVD_CGC_CTRL);
 841         if (data & UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK)
 842                 *flags |= AMD_CG_SUPPORT_UVD_MGCG;
 843
 844 out:
 845         mutex_unlock(&adev->pm.mutex);
 846 }
 847
 848 static const struct amd_ip_funcs uvd_v5_0_ip_funcs = {
 849         .name = "uvd_v5_0",
 850         .early_init = uvd_v5_0_early_init,
 851         .late_init = NULL,
 852         .sw_init = uvd_v5_0_sw_init,
 853         .sw_fini = uvd_v5_0_sw_fini,
 854         .hw_init = uvd_v5_0_hw_init,
 855         .hw_fini = uvd_v5_0_hw_fini,
 856         .suspend = uvd_v5_0_suspend,
 857         .resume = uvd_v5_0_resume,
 858         .is_idle = uvd_v5_0_is_idle,
 859         .wait_for_idle = uvd_v5_0_wait_for_idle,
 860         .soft_reset = uvd_v5_0_soft_reset,
 861         .set_clockgating_state = uvd_v5_0_set_clockgating_state,
 862         .set_powergating_state = uvd_v5_0_set_powergating_state,
 863         .get_clockgating_state = uvd_v5_0_get_clockgating_state,
 864 };
 865
 866 static const struct amdgpu_ring_funcs uvd_v5_0_ring_funcs = {
 867         .type = AMDGPU_RING_TYPE_UVD,
 868         .align_mask = 0xf,
 869         .nop = PACKET0(mmUVD_NO_OP, 0),
 870         .support_64bit_ptrs = false,
 871         .get_rptr = uvd_v5_0_ring_get_rptr,
 872         .get_wptr = uvd_v5_0_ring_get_wptr,
 873         .set_wptr = uvd_v5_0_ring_set_wptr,
 874         .parse_cs = amdgpu_uvd_ring_parse_cs,
 875         .emit_frame_size =
 876                 2 + /* uvd_v5_0_ring_emit_hdp_flush */
 877                 2 + /* uvd_v5_0_ring_emit_hdp_invalidate */
 878                 14, /* uvd_v5_0_ring_emit_fence  x1 no user fence */
 879         .emit_ib_size = 6, /* uvd_v5_0_ring_emit_ib */
 880         .emit_ib = uvd_v5_0_ring_emit_ib,
 881         .emit_fence = uvd_v5_0_ring_emit_fence,
 882         .emit_hdp_flush = uvd_v5_0_ring_emit_hdp_flush,
 883         .emit_hdp_invalidate = uvd_v5_0_ring_emit_hdp_invalidate,
 884         .test_ring = uvd_v5_0_ring_test_ring,
 885         .test_ib = amdgpu_uvd_ring_test_ib,
 886         .insert_nop = amdgpu_ring_insert_nop,
 887         .pad_ib = amdgpu_ring_generic_pad_ib,
 888         .begin_use = amdgpu_uvd_ring_begin_use,
 889         .end_use = amdgpu_uvd_ring_end_use,
 890 };
 891
 892 static void uvd_v5_0_set_ring_funcs(struct amdgpu_device *adev)
 893 {
 894         adev->uvd.ring.funcs = &uvd_v5_0_ring_funcs;
 895 }
 896
 897 static const struct amdgpu_irq_src_funcs uvd_v5_0_irq_funcs = {
 898         .set = uvd_v5_0_set_interrupt_state,
 899         .process = uvd_v5_0_process_interrupt,
 900 };
 901
 902 static void uvd_v5_0_set_irq_funcs(struct amdgpu_device *adev)
 903 {
 904         adev->uvd.irq.num_types = 1;
 905         adev->uvd.irq.funcs = &uvd_v5_0_irq_funcs;
 906 }
 907
 908 const struct amdgpu_ip_block_version uvd_v5_0_ip_block =
 909 {
 910                 .type = AMD_IP_BLOCK_TYPE_UVD,
 911                 .major = 5,
 912                 .minor = 0,
 913                 .rev = 0,
 914                 .funcs = &uvd_v5_0_ip_funcs,
 915 };