drivers/gpu/drm/radeon/radeon_uvd.c

   1 /*
   2  * Copyright 2011 Advanced Micro Devices, Inc.
   3  * All Rights Reserved.
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the
   7  * "Software"), to deal in the Software without restriction, including
   8  * without limitation the rights to use, copy, modify, merge, publish,
   9  * distribute, sub license, and/or sell copies of the Software, and to
  10  * permit persons to whom the Software is furnished to do so, subject to
  11  * the following conditions:
  12  *
  13  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  14  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  15  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  16  * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
  17  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  18  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  19  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  20  *
  21  * The above copyright notice and this permission notice (including the
  22  * next paragraph) shall be included in all copies or substantial portions
  23  * of the Software.
  24  *
  25  */
  26 /*
  27  * Authors:
  28  *    Christian König <deathsimple@vodafone.de>
  29  */
  30
  31 #include <linux/firmware.h>
  32 #include <linux/module.h>
  33 #include <drm/drmP.h>
  34 #include <drm/drm.h>
  35
  36 #include "radeon.h"
  37 #include "r600d.h"
  38
  39 /* 1 second timeout */
  40 #define UVD_IDLE_TIMEOUT_MS     1000
  41
  42 /* Firmware Names */
  43 #define FIRMWARE_R600           "radeon/R600_uvd.bin"
  44 #define FIRMWARE_RS780          "radeon/RS780_uvd.bin"
  45 #define FIRMWARE_RV770          "radeon/RV770_uvd.bin"
  46 #define FIRMWARE_RV710          "radeon/RV710_uvd.bin"
  47 #define FIRMWARE_CYPRESS        "radeon/CYPRESS_uvd.bin"
  48 #define FIRMWARE_SUMO           "radeon/SUMO_uvd.bin"
  49 #define FIRMWARE_TAHITI         "radeon/TAHITI_uvd.bin"
  50 #define FIRMWARE_BONAIRE        "radeon/BONAIRE_uvd.bin"
  51
  52 MODULE_FIRMWARE(FIRMWARE_R600);
  53 MODULE_FIRMWARE(FIRMWARE_RS780);
  54 MODULE_FIRMWARE(FIRMWARE_RV770);
  55 MODULE_FIRMWARE(FIRMWARE_RV710);
  56 MODULE_FIRMWARE(FIRMWARE_CYPRESS);
  57 MODULE_FIRMWARE(FIRMWARE_SUMO);
  58 MODULE_FIRMWARE(FIRMWARE_TAHITI);
  59 MODULE_FIRMWARE(FIRMWARE_BONAIRE);
  60
  61 static void radeon_uvd_idle_work_handler(struct work_struct *work);
  62
  63 int radeon_uvd_init(struct radeon_device *rdev)
  64 {
  65         unsigned long bo_size;
  66         const char *fw_name;
  67         int i, r;
  68
  69         INIT_DELAYED_WORK(&rdev->uvd.idle_work, radeon_uvd_idle_work_handler);
  70
  71         switch (rdev->family) {
  72         case CHIP_RV610:
  73         case CHIP_RV630:
  74         case CHIP_RV670:
  75         case CHIP_RV620:
  76         case CHIP_RV635:
  77                 fw_name = FIRMWARE_R600;
  78                 break;
  79
  80         case CHIP_RS780:
  81         case CHIP_RS880:
  82                 fw_name = FIRMWARE_RS780;
  83                 break;
  84
  85         case CHIP_RV770:
  86                 fw_name = FIRMWARE_RV770;
  87                 break;
  88
  89         case CHIP_RV710:
  90         case CHIP_RV730:
  91         case CHIP_RV740:
  92                 fw_name = FIRMWARE_RV710;
  93                 break;
  94
  95         case CHIP_CYPRESS:
  96         case CHIP_HEMLOCK:
  97         case CHIP_JUNIPER:
  98         case CHIP_REDWOOD:
  99         case CHIP_CEDAR:
 100                 fw_name = FIRMWARE_CYPRESS;
 101                 break;
 102
 103         case CHIP_SUMO:
 104         case CHIP_SUMO2:
 105         case CHIP_PALM:
 106         case CHIP_CAYMAN:
 107         case CHIP_BARTS:
 108         case CHIP_TURKS:
 109         case CHIP_CAICOS:
 110                 fw_name = FIRMWARE_SUMO;
 111                 break;
 112
 113         case CHIP_TAHITI:
 114         case CHIP_VERDE:
 115         case CHIP_PITCAIRN:
 116         case CHIP_ARUBA:
 117         case CHIP_OLAND:
 118                 fw_name = FIRMWARE_TAHITI;
 119                 break;
 120
 121         case CHIP_BONAIRE:
 122         case CHIP_KABINI:
 123         case CHIP_KAVERI:
 124         case CHIP_HAWAII:
 125         case CHIP_MULLINS:
 126                 fw_name = FIRMWARE_BONAIRE;
 127                 break;
 128
 129         default:
 130                 return -EINVAL;
 131         }
 132
 133         r = request_firmware(&rdev->uvd_fw, fw_name, rdev->dev);
 134         if (r) {
 135                 dev_err(rdev->dev, "radeon_uvd: Can't load firmware \"%s\"\n",
 136                         fw_name);
 137                 return r;
 138         }
 139
 140         bo_size = RADEON_GPU_PAGE_ALIGN(rdev->uvd_fw->size + 8) +
 141                   RADEON_UVD_STACK_SIZE + RADEON_UVD_HEAP_SIZE +
 142                   RADEON_GPU_PAGE_SIZE;
 143         r = radeon_bo_create(rdev, bo_size, PAGE_SIZE, true,
 144                              RADEON_GEM_DOMAIN_VRAM, 0, NULL,
 145                              NULL, &rdev->uvd.vcpu_bo);
 146         if (r) {
 147                 dev_err(rdev->dev, "(%d) failed to allocate UVD bo\n", r);
 148                 return r;
 149         }
 150
 151         r = radeon_bo_reserve(rdev->uvd.vcpu_bo, false);
 152         if (r) {
 153                 radeon_bo_unref(&rdev->uvd.vcpu_bo);
 154                 dev_err(rdev->dev, "(%d) failed to reserve UVD bo\n", r);
 155                 return r;
 156         }
 157
 158         r = radeon_bo_pin(rdev->uvd.vcpu_bo, RADEON_GEM_DOMAIN_VRAM,
 159                           &rdev->uvd.gpu_addr);
 160         if (r) {
 161                 radeon_bo_unreserve(rdev->uvd.vcpu_bo);
 162                 radeon_bo_unref(&rdev->uvd.vcpu_bo);
 163                 dev_err(rdev->dev, "(%d) UVD bo pin failed\n", r);
 164                 return r;
 165         }
 166
 167         r = radeon_bo_kmap(rdev->uvd.vcpu_bo, &rdev->uvd.cpu_addr);
 168         if (r) {
 169                 dev_err(rdev->dev, "(%d) UVD map failed\n", r);
 170                 return r;
 171         }
 172
 173         radeon_bo_unreserve(rdev->uvd.vcpu_bo);
 174
 175         for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) {
 176                 atomic_set(&rdev->uvd.handles[i], 0);
 177                 rdev->uvd.filp[i] = NULL;
 178                 rdev->uvd.img_size[i] = 0;
 179         }
 180
 181         return 0;
 182 }
 183
 184 void radeon_uvd_fini(struct radeon_device *rdev)
 185 {
 186         int r;
 187
 188         if (rdev->uvd.vcpu_bo == NULL)
 189                 return;
 190
 191         r = radeon_bo_reserve(rdev->uvd.vcpu_bo, false);
 192         if (!r) {
 193                 radeon_bo_kunmap(rdev->uvd.vcpu_bo);
 194                 radeon_bo_unpin(rdev->uvd.vcpu_bo);
 195                 radeon_bo_unreserve(rdev->uvd.vcpu_bo);
 196         }
 197
 198         radeon_bo_unref(&rdev->uvd.vcpu_bo);
 199
 200         radeon_ring_fini(rdev, &rdev->ring[R600_RING_TYPE_UVD_INDEX]);
 201
 202         release_firmware(rdev->uvd_fw);
 203 }
 204
 205 int radeon_uvd_suspend(struct radeon_device *rdev)
 206 {
 207         unsigned size;
 208         void *ptr;
 209         int i;
 210
 211         if (rdev->uvd.vcpu_bo == NULL)
 212                 return 0;
 213
 214         for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i)
 215                 if (atomic_read(&rdev->uvd.handles[i]))
 216                         break;
 217
 218         if (i == RADEON_MAX_UVD_HANDLES)
 219                 return 0;
 220
 221         size = radeon_bo_size(rdev->uvd.vcpu_bo);
 222         size -= rdev->uvd_fw->size;
 223
 224         ptr = rdev->uvd.cpu_addr;
 225         ptr += rdev->uvd_fw->size;
 226
 227         rdev->uvd.saved_bo = kmalloc(size, GFP_KERNEL);
 228         memcpy(rdev->uvd.saved_bo, ptr, size);
 229
 230         return 0;
 231 }
 232
 233 int radeon_uvd_resume(struct radeon_device *rdev)
 234 {
 235         unsigned size;
 236         void *ptr;
 237
 238         if (rdev->uvd.vcpu_bo == NULL)
 239                 return -EINVAL;
 240
 241         memcpy(rdev->uvd.cpu_addr, rdev->uvd_fw->data, rdev->uvd_fw->size);
 242
 243         size = radeon_bo_size(rdev->uvd.vcpu_bo);
 244         size -= rdev->uvd_fw->size;
 245
 246         ptr = rdev->uvd.cpu_addr;
 247         ptr += rdev->uvd_fw->size;
 248
 249         if (rdev->uvd.saved_bo != NULL) {
 250                 memcpy(ptr, rdev->uvd.saved_bo, size);
 251                 kfree(rdev->uvd.saved_bo);
 252                 rdev->uvd.saved_bo = NULL;
 253         } else
 254                 memset(ptr, 0, size);
 255
 256         return 0;
 257 }
 258
 259 void radeon_uvd_force_into_uvd_segment(struct radeon_bo *rbo,
 260                                        uint32_t allowed_domains)
 261 {
 262         int i;
 263
 264         for (i = 0; i < rbo->placement.num_placement; ++i) {
 265                 rbo->placements[i].fpfn = 0 >> PAGE_SHIFT;
 266                 rbo->placements[i].lpfn = (256 * 1024 * 1024) >> PAGE_SHIFT;
 267         }
 268
 269         /* If it must be in VRAM it must be in the first segment as well */
 270         if (allowed_domains == RADEON_GEM_DOMAIN_VRAM)
 271                 return;
 272
 273         /* abort if we already have more than one placement */
 274         if (rbo->placement.num_placement > 1)
 275                 return;
 276
 277         /* add another 256MB segment */
 278         rbo->placements[1] = rbo->placements[0];
 279         rbo->placements[1].fpfn += (256 * 1024 * 1024) >> PAGE_SHIFT;
 280         rbo->placements[1].lpfn += (256 * 1024 * 1024) >> PAGE_SHIFT;
 281         rbo->placement.num_placement++;
 282         rbo->placement.num_busy_placement++;
 283 }
 284
 285 void radeon_uvd_free_handles(struct radeon_device *rdev, struct drm_file *filp)
 286 {
 287         int i, r;
 288         for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) {
 289                 uint32_t handle = atomic_read(&rdev->uvd.handles[i]);
 290                 if (handle != 0 && rdev->uvd.filp[i] == filp) {
 291                         struct radeon_fence *fence;
 292
 293                         radeon_uvd_note_usage(rdev);
 294
 295                         r = radeon_uvd_get_destroy_msg(rdev,
 296                                 R600_RING_TYPE_UVD_INDEX, handle, &fence);
 297                         if (r) {
 298                                 DRM_ERROR("Error destroying UVD (%d)!\n", r);
 299                                 continue;
 300                         }
 301
 302                         radeon_fence_wait(fence, false);
 303                         radeon_fence_unref(&fence);
 304
 305                         rdev->uvd.filp[i] = NULL;
 306                         atomic_set(&rdev->uvd.handles[i], 0);
 307                 }
 308         }
 309 }
 310
 311 static int radeon_uvd_cs_msg_decode(uint32_t *msg, unsigned buf_sizes[])
 312 {
 313         unsigned stream_type = msg[4];
 314         unsigned width = msg[6];
 315         unsigned height = msg[7];
 316         unsigned dpb_size = msg[9];
 317         unsigned pitch = msg[28];
 318
 319         unsigned width_in_mb = width / 16;
 320         unsigned height_in_mb = ALIGN(height / 16, 2);
 321
 322         unsigned image_size, tmp, min_dpb_size;
 323
 324         image_size = width * height;
 325         image_size += image_size / 2;
 326         image_size = ALIGN(image_size, 1024);
 327
 328         switch (stream_type) {
 329         case 0: /* H264 */
 330
 331                 /* reference picture buffer */
 332                 min_dpb_size = image_size * 17;
 333
 334                 /* macroblock context buffer */
 335                 min_dpb_size += width_in_mb * height_in_mb * 17 * 192;
 336
 337                 /* IT surface buffer */
 338                 min_dpb_size += width_in_mb * height_in_mb * 32;
 339                 break;
 340
 341         case 1: /* VC1 */
 342
 343                 /* reference picture buffer */
 344                 min_dpb_size = image_size * 3;
 345
 346                 /* CONTEXT_BUFFER */
 347                 min_dpb_size += width_in_mb * height_in_mb * 128;
 348
 349                 /* IT surface buffer */
 350                 min_dpb_size += width_in_mb * 64;
 351
 352                 /* DB surface buffer */
 353                 min_dpb_size += width_in_mb * 128;
 354
 355                 /* BP */
 356                 tmp = max(width_in_mb, height_in_mb);
 357                 min_dpb_size += ALIGN(tmp * 7 * 16, 64);
 358                 break;
 359
 360         case 3: /* MPEG2 */
 361
 362                 /* reference picture buffer */
 363                 min_dpb_size = image_size * 3;
 364                 break;
 365
 366         case 4: /* MPEG4 */
 367
 368                 /* reference picture buffer */
 369                 min_dpb_size = image_size * 3;
 370
 371                 /* CM */
 372                 min_dpb_size += width_in_mb * height_in_mb * 64;
 373
 374                 /* IT surface buffer */
 375                 min_dpb_size += ALIGN(width_in_mb * height_in_mb * 32, 64);
 376                 break;
 377
 378         default:
 379                 DRM_ERROR("UVD codec not handled %d!\n", stream_type);
 380                 return -EINVAL;
 381         }
 382
 383         if (width > pitch) {
 384                 DRM_ERROR("Invalid UVD decoding target pitch!\n");
 385                 return -EINVAL;
 386         }
 387
 388         if (dpb_size < min_dpb_size) {
 389                 DRM_ERROR("Invalid dpb_size in UVD message (%d / %d)!\n",
 390                           dpb_size, min_dpb_size);
 391                 return -EINVAL;
 392         }
 393
 394         buf_sizes[0x1] = dpb_size;
 395         buf_sizes[0x2] = image_size;
 396         return 0;
 397 }
 398
 399 static int radeon_uvd_cs_msg(struct radeon_cs_parser *p, struct radeon_bo *bo,
 400                              unsigned offset, unsigned buf_sizes[])
 401 {
 402         int32_t *msg, msg_type, handle;
 403         unsigned img_size = 0;
 404         struct fence *f;
 405         void *ptr;
 406
 407         int i, r;
 408
 409         if (offset & 0x3F) {
 410                 DRM_ERROR("UVD messages must be 64 byte aligned!\n");
 411                 return -EINVAL;
 412         }
 413
 414         f = reservation_object_get_excl(bo->tbo.resv);
 415         if (f) {
 416                 r = radeon_fence_wait((struct radeon_fence *)f, false);
 417                 if (r) {
 418                         DRM_ERROR("Failed waiting for UVD message (%d)!\n", r);
 419                         return r;
 420                 }
 421         }
 422
 423         r = radeon_bo_kmap(bo, &ptr);
 424         if (r) {
 425                 DRM_ERROR("Failed mapping the UVD message (%d)!\n", r);
 426                 return r;
 427         }
 428
 429         msg = ptr + offset;
 430
 431         msg_type = msg[1];
 432         handle = msg[2];
 433
 434         if (handle == 0) {
 435                 DRM_ERROR("Invalid UVD handle!\n");
 436                 return -EINVAL;
 437         }
 438
 439         if (msg_type == 1) {
 440                 /* it's a decode msg, calc buffer sizes */
 441                 r = radeon_uvd_cs_msg_decode(msg, buf_sizes);
 442                 /* calc image size (width * height) */
 443                 img_size = msg[6] * msg[7];
 444                 radeon_bo_kunmap(bo);
 445                 if (r)
 446                         return r;
 447
 448         } else if (msg_type == 2) {
 449                 /* it's a destroy msg, free the handle */
 450                 for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i)
 451                         atomic_cmpxchg(&p->rdev->uvd.handles[i], handle, 0);
 452                 radeon_bo_kunmap(bo);
 453                 return 0;
 454         } else {
 455                 /* it's a create msg, calc image size (width * height) */
 456                 img_size = msg[7] * msg[8];
 457                 radeon_bo_kunmap(bo);
 458
 459                 if (msg_type != 0) {
 460                         DRM_ERROR("Illegal UVD message type (%d)!\n", msg_type);
 461                         return -EINVAL;
 462                 }
 463
 464                 /* it's a create msg, no special handling needed */
 465         }
 466
 467         /* create or decode, validate the handle */
 468         for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) {
 469                 if (atomic_read(&p->rdev->uvd.handles[i]) == handle)
 470                         return 0;
 471         }
 472
 473         /* handle not found try to alloc a new one */
 474         for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) {
 475                 if (!atomic_cmpxchg(&p->rdev->uvd.handles[i], 0, handle)) {
 476                         p->rdev->uvd.filp[i] = p->filp;
 477                         p->rdev->uvd.img_size[i] = img_size;
 478                         return 0;
 479                 }
 480         }
 481
 482         DRM_ERROR("No more free UVD handles!\n");
 483         return -EINVAL;
 484 }
 485
 486 static int radeon_uvd_cs_reloc(struct radeon_cs_parser *p,
 487                                int data0, int data1,
 488                                unsigned buf_sizes[], bool *has_msg_cmd)
 489 {
 490         struct radeon_cs_chunk *relocs_chunk;
 491         struct radeon_bo_list *reloc;
 492         unsigned idx, cmd, offset;
 493         uint64_t start, end;
 494         int r;
 495
 496         relocs_chunk = p->chunk_relocs;
 497         offset = radeon_get_ib_value(p, data0);
 498         idx = radeon_get_ib_value(p, data1);
 499         if (idx >= relocs_chunk->length_dw) {
 500                 DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
 501                           idx, relocs_chunk->length_dw);
 502                 return -EINVAL;
 503         }
 504
 505         reloc = &p->relocs[(idx / 4)];
 506         start = reloc->gpu_offset;
 507         end = start + radeon_bo_size(reloc->robj);
 508         start += offset;
 509
 510         p->ib.ptr[data0] = start & 0xFFFFFFFF;
 511         p->ib.ptr[data1] = start >> 32;
 512
 513         cmd = radeon_get_ib_value(p, p->idx) >> 1;
 514
 515         if (cmd < 0x4) {
 516                 if (end <= start) {
 517                         DRM_ERROR("invalid reloc offset %X!\n", offset);
 518                         return -EINVAL;
 519                 }
 520                 if ((end - start) < buf_sizes[cmd]) {
 521                         DRM_ERROR("buffer (%d) to small (%d / %d)!\n", cmd,
 522                                   (unsigned)(end - start), buf_sizes[cmd]);
 523                         return -EINVAL;
 524                 }
 525
 526         } else if (cmd != 0x100) {
 527                 DRM_ERROR("invalid UVD command %X!\n", cmd);
 528                 return -EINVAL;
 529         }
 530
 531         if ((start >> 28) != ((end - 1) >> 28)) {
 532                 DRM_ERROR("reloc %LX-%LX crossing 256MB boundary!\n",
 533                           start, end);
 534                 return -EINVAL;
 535         }
 536
 537         /* TODO: is this still necessary on NI+ ? */
 538         if ((cmd == 0 || cmd == 0x3) &&
 539             (start >> 28) != (p->rdev->uvd.gpu_addr >> 28)) {
 540                 DRM_ERROR("msg/fb buffer %LX-%LX out of 256MB segment!\n",
 541                           start, end);
 542                 return -EINVAL;
 543         }
 544
 545         if (cmd == 0) {
 546                 if (*has_msg_cmd) {
 547                         DRM_ERROR("More than one message in a UVD-IB!\n");
 548                         return -EINVAL;
 549                 }
 550                 *has_msg_cmd = true;
 551                 r = radeon_uvd_cs_msg(p, reloc->robj, offset, buf_sizes);
 552                 if (r)
 553                         return r;
 554         } else if (!*has_msg_cmd) {
 555                 DRM_ERROR("Message needed before other commands are send!\n");
 556                 return -EINVAL;
 557         }
 558
 559         return 0;
 560 }
 561
 562 static int radeon_uvd_cs_reg(struct radeon_cs_parser *p,
 563                              struct radeon_cs_packet *pkt,
 564                              int *data0, int *data1,
 565                              unsigned buf_sizes[],
 566                              bool *has_msg_cmd)
 567 {
 568         int i, r;
 569
 570         p->idx++;
 571         for (i = 0; i <= pkt->count; ++i) {
 572                 switch (pkt->reg + i*4) {
 573                 case UVD_GPCOM_VCPU_DATA0:
 574                         *data0 = p->idx;
 575                         break;
 576                 case UVD_GPCOM_VCPU_DATA1:
 577                         *data1 = p->idx;
 578                         break;
 579                 case UVD_GPCOM_VCPU_CMD:
 580                         r = radeon_uvd_cs_reloc(p, *data0, *data1,
 581                                                 buf_sizes, has_msg_cmd);
 582                         if (r)
 583                                 return r;
 584                         break;
 585                 case UVD_ENGINE_CNTL:
 586                         break;
 587                 default:
 588                         DRM_ERROR("Invalid reg 0x%X!\n",
 589                                   pkt->reg + i*4);
 590                         return -EINVAL;
 591                 }
 592                 p->idx++;
 593         }
 594         return 0;
 595 }
 596
 597 int radeon_uvd_cs_parse(struct radeon_cs_parser *p)
 598 {
 599         struct radeon_cs_packet pkt;
 600         int r, data0 = 0, data1 = 0;
 601
 602         /* does the IB has a msg command */
 603         bool has_msg_cmd = false;
 604
 605         /* minimum buffer sizes */
 606         unsigned buf_sizes[] = {
 607                 [0x00000000]    =       2048,
 608                 [0x00000001]    =       32 * 1024 * 1024,
 609                 [0x00000002]    =       2048 * 1152 * 3,
 610                 [0x00000003]    =       2048,
 611         };
 612
 613         if (p->chunk_ib->length_dw % 16) {
 614                 DRM_ERROR("UVD IB length (%d) not 16 dwords aligned!\n",
 615                           p->chunk_ib->length_dw);
 616                 return -EINVAL;
 617         }
 618
 619         if (p->chunk_relocs == NULL) {
 620                 DRM_ERROR("No relocation chunk !\n");
 621                 return -EINVAL;
 622         }
 623
 624
 625         do {
 626                 r = radeon_cs_packet_parse(p, &pkt, p->idx);
 627                 if (r)
 628                         return r;
 629                 switch (pkt.type) {
 630                 case RADEON_PACKET_TYPE0:
 631                         r = radeon_uvd_cs_reg(p, &pkt, &data0, &data1,
 632                                               buf_sizes, &has_msg_cmd);
 633                         if (r)
 634                                 return r;
 635                         break;
 636                 case RADEON_PACKET_TYPE2:
 637                         p->idx += pkt.count + 2;
 638                         break;
 639                 default:
 640                         DRM_ERROR("Unknown packet type %d !\n", pkt.type);
 641                         return -EINVAL;
 642                 }
 643         } while (p->idx < p->chunk_ib->length_dw);
 644
 645         if (!has_msg_cmd) {
 646                 DRM_ERROR("UVD-IBs need a msg command!\n");
 647                 return -EINVAL;
 648         }
 649
 650         return 0;
 651 }
 652
 653 static int radeon_uvd_send_msg(struct radeon_device *rdev,
 654                                int ring, uint64_t addr,
 655                                struct radeon_fence **fence)
 656 {
 657         struct radeon_ib ib;
 658         int i, r;
 659
 660         r = radeon_ib_get(rdev, ring, &ib, NULL, 64);
 661         if (r)
 662                 return r;
 663
 664         ib.ptr[0] = PACKET0(UVD_GPCOM_VCPU_DATA0, 0);
 665         ib.ptr[1] = addr;
 666         ib.ptr[2] = PACKET0(UVD_GPCOM_VCPU_DATA1, 0);
 667         ib.ptr[3] = addr >> 32;
 668         ib.ptr[4] = PACKET0(UVD_GPCOM_VCPU_CMD, 0);
 669         ib.ptr[5] = 0;
 670         for (i = 6; i < 16; ++i)
 671                 ib.ptr[i] = PACKET2(0);
 672         ib.length_dw = 16;
 673
 674         r = radeon_ib_schedule(rdev, &ib, NULL, false);
 675
 676         if (fence)
 677                 *fence = radeon_fence_ref(ib.fence);
 678
 679         radeon_ib_free(rdev, &ib);
 680         return r;
 681 }
 682
 683 /* multiple fence commands without any stream commands in between can
 684    crash the vcpu so just try to emmit a dummy create/destroy msg to
 685    avoid this */
 686 int radeon_uvd_get_create_msg(struct radeon_device *rdev, int ring,
 687                               uint32_t handle, struct radeon_fence **fence)
 688 {
 689         /* we use the last page of the vcpu bo for the UVD message */
 690         uint64_t offs = radeon_bo_size(rdev->uvd.vcpu_bo) -
 691                 RADEON_GPU_PAGE_SIZE;
 692
 693         uint32_t *msg = rdev->uvd.cpu_addr + offs;
 694         uint64_t addr = rdev->uvd.gpu_addr + offs;
 695
 696         int r, i;
 697
 698         r = radeon_bo_reserve(rdev->uvd.vcpu_bo, true);
 699         if (r)
 700                 return r;
 701
 702         /* stitch together an UVD create msg */
 703         msg[0] = cpu_to_le32(0x00000de4);
 704         msg[1] = cpu_to_le32(0x00000000);
 705         msg[2] = cpu_to_le32(handle);
 706         msg[3] = cpu_to_le32(0x00000000);
 707         msg[4] = cpu_to_le32(0x00000000);
 708         msg[5] = cpu_to_le32(0x00000000);
 709         msg[6] = cpu_to_le32(0x00000000);
 710         msg[7] = cpu_to_le32(0x00000780);
 711         msg[8] = cpu_to_le32(0x00000440);
 712         msg[9] = cpu_to_le32(0x00000000);
 713         msg[10] = cpu_to_le32(0x01b37000);
 714         for (i = 11; i < 1024; ++i)
 715                 msg[i] = cpu_to_le32(0x0);
 716
 717         r = radeon_uvd_send_msg(rdev, ring, addr, fence);
 718         radeon_bo_unreserve(rdev->uvd.vcpu_bo);
 719         return r;
 720 }
 721
 722 int radeon_uvd_get_destroy_msg(struct radeon_device *rdev, int ring,
 723                                uint32_t handle, struct radeon_fence **fence)
 724 {
 725         /* we use the last page of the vcpu bo for the UVD message */
 726         uint64_t offs = radeon_bo_size(rdev->uvd.vcpu_bo) -
 727                 RADEON_GPU_PAGE_SIZE;
 728
 729         uint32_t *msg = rdev->uvd.cpu_addr + offs;
 730         uint64_t addr = rdev->uvd.gpu_addr + offs;
 731
 732         int r, i;
 733
 734         r = radeon_bo_reserve(rdev->uvd.vcpu_bo, true);
 735         if (r)
 736                 return r;
 737
 738         /* stitch together an UVD destroy msg */
 739         msg[0] = cpu_to_le32(0x00000de4);
 740         msg[1] = cpu_to_le32(0x00000002);
 741         msg[2] = cpu_to_le32(handle);
 742         msg[3] = cpu_to_le32(0x00000000);
 743         for (i = 4; i < 1024; ++i)
 744                 msg[i] = cpu_to_le32(0x0);
 745
 746         r = radeon_uvd_send_msg(rdev, ring, addr, fence);
 747         radeon_bo_unreserve(rdev->uvd.vcpu_bo);
 748         return r;
 749 }
 750
 751 /**
 752  * radeon_uvd_count_handles - count number of open streams
 753  *
 754  * @rdev: radeon_device pointer
 755  * @sd: number of SD streams
 756  * @hd: number of HD streams
 757  *
 758  * Count the number of open SD/HD streams as a hint for power mangement
 759  */
 760 static void radeon_uvd_count_handles(struct radeon_device *rdev,
 761                                      unsigned *sd, unsigned *hd)
 762 {
 763         unsigned i;
 764
 765         *sd = 0;
 766         *hd = 0;
 767
 768         for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) {
 769                 if (!atomic_read(&rdev->uvd.handles[i]))
 770                         continue;
 771
 772                 if (rdev->uvd.img_size[i] >= 720*576)
 773                         ++(*hd);
 774                 else
 775                         ++(*sd);
 776         }
 777 }
 778
 779 static void radeon_uvd_idle_work_handler(struct work_struct *work)
 780 {
 781         struct radeon_device *rdev =
 782                 container_of(work, struct radeon_device, uvd.idle_work.work);
 783
 784         if (radeon_fence_count_emitted(rdev, R600_RING_TYPE_UVD_INDEX) == 0) {
 785                 if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) {
 786                         radeon_uvd_count_handles(rdev, &rdev->pm.dpm.sd,
 787                                                  &rdev->pm.dpm.hd);
 788                         radeon_dpm_enable_uvd(rdev, false);
 789                 } else {
 790                         radeon_set_uvd_clocks(rdev, 0, 0);
 791                 }
 792         } else {
 793                 schedule_delayed_work(&rdev->uvd.idle_work,
 794                                       msecs_to_jiffies(UVD_IDLE_TIMEOUT_MS));
 795         }
 796 }
 797
 798 void radeon_uvd_note_usage(struct radeon_device *rdev)
 799 {
 800         bool streams_changed = false;
 801         bool set_clocks = !cancel_delayed_work_sync(&rdev->uvd.idle_work);
 802         set_clocks &= schedule_delayed_work(&rdev->uvd.idle_work,
 803                                             msecs_to_jiffies(UVD_IDLE_TIMEOUT_MS));
 804
 805         if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) {
 806                 unsigned hd = 0, sd = 0;
 807                 radeon_uvd_count_handles(rdev, &sd, &hd);
 808                 if ((rdev->pm.dpm.sd != sd) ||
 809                     (rdev->pm.dpm.hd != hd)) {
 810                         rdev->pm.dpm.sd = sd;
 811                         rdev->pm.dpm.hd = hd;
 812                         /* disable this for now */
 813                         /*streams_changed = true;*/
 814                 }
 815         }
 816
 817         if (set_clocks || streams_changed) {
 818                 if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) {
 819                         radeon_dpm_enable_uvd(rdev, true);
 820                 } else {
 821                         radeon_set_uvd_clocks(rdev, 53300, 40000);
 822                 }
 823         }
 824 }
 825
 826 static unsigned radeon_uvd_calc_upll_post_div(unsigned vco_freq,
 827                                               unsigned target_freq,
 828                                               unsigned pd_min,
 829                                               unsigned pd_even)
 830 {
 831         unsigned post_div = vco_freq / target_freq;
 832
 833         /* adjust to post divider minimum value */
 834         if (post_div < pd_min)
 835                 post_div = pd_min;
 836
 837         /* we alway need a frequency less than or equal the target */
 838         if ((vco_freq / post_div) > target_freq)
 839                 post_div += 1;
 840
 841         /* post dividers above a certain value must be even */
 842         if (post_div > pd_even && post_div % 2)
 843                 post_div += 1;
 844
 845         return post_div;
 846 }
 847
 848 /**
 849  * radeon_uvd_calc_upll_dividers - calc UPLL clock dividers
 850  *
 851  * @rdev: radeon_device pointer
 852  * @vclk: wanted VCLK
 853  * @dclk: wanted DCLK
 854  * @vco_min: minimum VCO frequency
 855  * @vco_max: maximum VCO frequency
 856  * @fb_factor: factor to multiply vco freq with
 857  * @fb_mask: limit and bitmask for feedback divider
 858  * @pd_min: post divider minimum
 859  * @pd_max: post divider maximum
 860  * @pd_even: post divider must be even above this value
 861  * @optimal_fb_div: resulting feedback divider
 862  * @optimal_vclk_div: resulting vclk post divider
 863  * @optimal_dclk_div: resulting dclk post divider
 864  *
 865  * Calculate dividers for UVDs UPLL (R6xx-SI, except APUs).
 866  * Returns zero on success -EINVAL on error.
 867  */
 868 int radeon_uvd_calc_upll_dividers(struct radeon_device *rdev,
 869                                   unsigned vclk, unsigned dclk,
 870                                   unsigned vco_min, unsigned vco_max,
 871                                   unsigned fb_factor, unsigned fb_mask,
 872                                   unsigned pd_min, unsigned pd_max,
 873                                   unsigned pd_even,
 874                                   unsigned *optimal_fb_div,
 875                                   unsigned *optimal_vclk_div,
 876                                   unsigned *optimal_dclk_div)
 877 {
 878         unsigned vco_freq, ref_freq = rdev->clock.spll.reference_freq;
 879
 880         /* start off with something large */
 881         unsigned optimal_score = ~0;
 882
 883         /* loop through vco from low to high */
 884         vco_min = max(max(vco_min, vclk), dclk);
 885         for (vco_freq = vco_min; vco_freq <= vco_max; vco_freq += 100) {
 886
 887                 uint64_t fb_div = (uint64_t)vco_freq * fb_factor;
 888                 unsigned vclk_div, dclk_div, score;
 889
 890                 do_div(fb_div, ref_freq);
 891
 892                 /* fb div out of range ? */
 893                 if (fb_div > fb_mask)
 894                         break; /* it can oly get worse */
 895
 896                 fb_div &= fb_mask;
 897
 898                 /* calc vclk divider with current vco freq */
 899                 vclk_div = radeon_uvd_calc_upll_post_div(vco_freq, vclk,
 900                                                          pd_min, pd_even);
 901                 if (vclk_div > pd_max)
 902                         break; /* vco is too big, it has to stop */
 903
 904                 /* calc dclk divider with current vco freq */
 905                 dclk_div = radeon_uvd_calc_upll_post_div(vco_freq, dclk,
 906                                                          pd_min, pd_even);
 907                 if (vclk_div > pd_max)
 908                         break; /* vco is too big, it has to stop */
 909
 910                 /* calc score with current vco freq */
 911                 score = vclk - (vco_freq / vclk_div) + dclk - (vco_freq / dclk_div);
 912
 913                 /* determine if this vco setting is better than current optimal settings */
 914                 if (score < optimal_score) {
 915                         *optimal_fb_div = fb_div;
 916                         *optimal_vclk_div = vclk_div;
 917                         *optimal_dclk_div = dclk_div;
 918                         optimal_score = score;
 919                         if (optimal_score == 0)
 920                                 break; /* it can't get better than this */
 921                 }
 922         }
 923
 924         /* did we found a valid setup ? */
 925         if (optimal_score == ~0)
 926                 return -EINVAL;
 927
 928         return 0;
 929 }
 930
 931 int radeon_uvd_send_upll_ctlreq(struct radeon_device *rdev,
 932                                 unsigned cg_upll_func_cntl)
 933 {
 934         unsigned i;
 935
 936         /* make sure UPLL_CTLREQ is deasserted */
 937         WREG32_P(cg_upll_func_cntl, 0, ~UPLL_CTLREQ_MASK);
 938
 939         mdelay(10);
 940
 941         /* assert UPLL_CTLREQ */
 942         WREG32_P(cg_upll_func_cntl, UPLL_CTLREQ_MASK, ~UPLL_CTLREQ_MASK);
 943
 944         /* wait for CTLACK and CTLACK2 to get asserted */
 945         for (i = 0; i < 100; ++i) {
 946                 uint32_t mask = UPLL_CTLACK_MASK | UPLL_CTLACK2_MASK;
 947                 if ((RREG32(cg_upll_func_cntl) & mask) == mask)
 948                         break;
 949                 mdelay(10);
 950         }
 951
 952         /* deassert UPLL_CTLREQ */
 953         WREG32_P(cg_upll_func_cntl, 0, ~UPLL_CTLREQ_MASK);
 954
 955         if (i == 100) {
 956                 DRM_ERROR("Timeout setting UVD clocks!\n");
 957                 return -ETIMEDOUT;
 958         }
 959
 960         return 0;
 961 }