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

   1 /*
   2  * Copyright 2013 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  * Authors: Christian König <christian.koenig@amd.com>
  26  */
  27
  28 #include <linux/firmware.h>
  29 #include <linux/module.h>
  30 #include <drm/drmP.h>
  31 #include <drm/drm.h>
  32
  33 #include "amdgpu.h"
  34 #include "amdgpu_pm.h"
  35 #include "amdgpu_vce.h"
  36 #include "cikd.h"
  37
  38 /* 1 second timeout */
  39 #define VCE_IDLE_TIMEOUT_MS     1000
  40
  41 /* Firmware Names */
  42 #ifdef CONFIG_DRM_AMDGPU_CIK
  43 #define FIRMWARE_BONAIRE        "radeon/bonaire_vce.bin"
  44 #define FIRMWARE_KABINI         "radeon/kabini_vce.bin"
  45 #define FIRMWARE_KAVERI         "radeon/kaveri_vce.bin"
  46 #define FIRMWARE_HAWAII         "radeon/hawaii_vce.bin"
  47 #define FIRMWARE_MULLINS        "radeon/mullins_vce.bin"
  48 #endif
  49 #define FIRMWARE_TONGA          "amdgpu/tonga_vce.bin"
  50 #define FIRMWARE_CARRIZO        "amdgpu/carrizo_vce.bin"
  51 #define FIRMWARE_FIJI           "amdgpu/fiji_vce.bin"
  52 #define FIRMWARE_STONEY         "amdgpu/stoney_vce.bin"
  53
  54 #ifdef CONFIG_DRM_AMDGPU_CIK
  55 MODULE_FIRMWARE(FIRMWARE_BONAIRE);
  56 MODULE_FIRMWARE(FIRMWARE_KABINI);
  57 MODULE_FIRMWARE(FIRMWARE_KAVERI);
  58 MODULE_FIRMWARE(FIRMWARE_HAWAII);
  59 MODULE_FIRMWARE(FIRMWARE_MULLINS);
  60 #endif
  61 MODULE_FIRMWARE(FIRMWARE_TONGA);
  62 MODULE_FIRMWARE(FIRMWARE_CARRIZO);
  63 MODULE_FIRMWARE(FIRMWARE_FIJI);
  64 MODULE_FIRMWARE(FIRMWARE_STONEY);
  65
  66 static void amdgpu_vce_idle_work_handler(struct work_struct *work);
  67
  68 /**
  69  * amdgpu_vce_init - allocate memory, load vce firmware
  70  *
  71  * @adev: amdgpu_device pointer
  72  *
  73  * First step to get VCE online, allocate memory and load the firmware
  74  */
  75 int amdgpu_vce_sw_init(struct amdgpu_device *adev, unsigned long size)
  76 {
  77         const char *fw_name;
  78         const struct common_firmware_header *hdr;
  79         unsigned ucode_version, version_major, version_minor, binary_id;
  80         int i, r;
  81
  82         INIT_DELAYED_WORK(&adev->vce.idle_work, amdgpu_vce_idle_work_handler);
  83
  84         switch (adev->asic_type) {
  85 #ifdef CONFIG_DRM_AMDGPU_CIK
  86         case CHIP_BONAIRE:
  87                 fw_name = FIRMWARE_BONAIRE;
  88                 break;
  89         case CHIP_KAVERI:
  90                 fw_name = FIRMWARE_KAVERI;
  91                 break;
  92         case CHIP_KABINI:
  93                 fw_name = FIRMWARE_KABINI;
  94                 break;
  95         case CHIP_HAWAII:
  96                 fw_name = FIRMWARE_HAWAII;
  97                 break;
  98         case CHIP_MULLINS:
  99                 fw_name = FIRMWARE_MULLINS;
 100                 break;
 101 #endif
 102         case CHIP_TONGA:
 103                 fw_name = FIRMWARE_TONGA;
 104                 break;
 105         case CHIP_CARRIZO:
 106                 fw_name = FIRMWARE_CARRIZO;
 107                 break;
 108         case CHIP_FIJI:
 109                 fw_name = FIRMWARE_FIJI;
 110                 break;
 111         case CHIP_STONEY:
 112                 fw_name = FIRMWARE_STONEY;
 113                 break;
 114
 115         default:
 116                 return -EINVAL;
 117         }
 118
 119         r = request_firmware(&adev->vce.fw, fw_name, adev->dev);
 120         if (r) {
 121                 dev_err(adev->dev, "amdgpu_vce: Can't load firmware \"%s\"\n",
 122                         fw_name);
 123                 return r;
 124         }
 125
 126         r = amdgpu_ucode_validate(adev->vce.fw);
 127         if (r) {
 128                 dev_err(adev->dev, "amdgpu_vce: Can't validate firmware \"%s\"\n",
 129                         fw_name);
 130                 release_firmware(adev->vce.fw);
 131                 adev->vce.fw = NULL;
 132                 return r;
 133         }
 134
 135         hdr = (const struct common_firmware_header *)adev->vce.fw->data;
 136
 137         ucode_version = le32_to_cpu(hdr->ucode_version);
 138         version_major = (ucode_version >> 20) & 0xfff;
 139         version_minor = (ucode_version >> 8) & 0xfff;
 140         binary_id = ucode_version & 0xff;
 141         DRM_INFO("Found VCE firmware Version: %hhd.%hhd Binary ID: %hhd\n",
 142                 version_major, version_minor, binary_id);
 143         adev->vce.fw_version = ((version_major << 24) | (version_minor << 16) |
 144                                 (binary_id << 8));
 145
 146         /* allocate firmware, stack and heap BO */
 147
 148         r = amdgpu_bo_create(adev, size, PAGE_SIZE, true,
 149                              AMDGPU_GEM_DOMAIN_VRAM,
 150                              AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
 151                              NULL, NULL, &adev->vce.vcpu_bo);
 152         if (r) {
 153                 dev_err(adev->dev, "(%d) failed to allocate VCE bo\n", r);
 154                 return r;
 155         }
 156
 157         r = amdgpu_bo_reserve(adev->vce.vcpu_bo, false);
 158         if (r) {
 159                 amdgpu_bo_unref(&adev->vce.vcpu_bo);
 160                 dev_err(adev->dev, "(%d) failed to reserve VCE bo\n", r);
 161                 return r;
 162         }
 163
 164         r = amdgpu_bo_pin(adev->vce.vcpu_bo, AMDGPU_GEM_DOMAIN_VRAM,
 165                           &adev->vce.gpu_addr);
 166         amdgpu_bo_unreserve(adev->vce.vcpu_bo);
 167         if (r) {
 168                 amdgpu_bo_unref(&adev->vce.vcpu_bo);
 169                 dev_err(adev->dev, "(%d) VCE bo pin failed\n", r);
 170                 return r;
 171         }
 172
 173         for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) {
 174                 atomic_set(&adev->vce.handles[i], 0);
 175                 adev->vce.filp[i] = NULL;
 176         }
 177
 178         return 0;
 179 }
 180
 181 /**
 182  * amdgpu_vce_fini - free memory
 183  *
 184  * @adev: amdgpu_device pointer
 185  *
 186  * Last step on VCE teardown, free firmware memory
 187  */
 188 int amdgpu_vce_sw_fini(struct amdgpu_device *adev)
 189 {
 190         if (adev->vce.vcpu_bo == NULL)
 191                 return 0;
 192
 193         amdgpu_bo_unref(&adev->vce.vcpu_bo);
 194
 195         amdgpu_ring_fini(&adev->vce.ring[0]);
 196         amdgpu_ring_fini(&adev->vce.ring[1]);
 197
 198         release_firmware(adev->vce.fw);
 199
 200         return 0;
 201 }
 202
 203 /**
 204  * amdgpu_vce_suspend - unpin VCE fw memory
 205  *
 206  * @adev: amdgpu_device pointer
 207  *
 208  */
 209 int amdgpu_vce_suspend(struct amdgpu_device *adev)
 210 {
 211         int i;
 212
 213         if (adev->vce.vcpu_bo == NULL)
 214                 return 0;
 215
 216         for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i)
 217                 if (atomic_read(&adev->vce.handles[i]))
 218                         break;
 219
 220         if (i == AMDGPU_MAX_VCE_HANDLES)
 221                 return 0;
 222
 223         /* TODO: suspending running encoding sessions isn't supported */
 224         return -EINVAL;
 225 }
 226
 227 /**
 228  * amdgpu_vce_resume - pin VCE fw memory
 229  *
 230  * @adev: amdgpu_device pointer
 231  *
 232  */
 233 int amdgpu_vce_resume(struct amdgpu_device *adev)
 234 {
 235         void *cpu_addr;
 236         const struct common_firmware_header *hdr;
 237         unsigned offset;
 238         int r;
 239
 240         if (adev->vce.vcpu_bo == NULL)
 241                 return -EINVAL;
 242
 243         r = amdgpu_bo_reserve(adev->vce.vcpu_bo, false);
 244         if (r) {
 245                 dev_err(adev->dev, "(%d) failed to reserve VCE bo\n", r);
 246                 return r;
 247         }
 248
 249         r = amdgpu_bo_kmap(adev->vce.vcpu_bo, &cpu_addr);
 250         if (r) {
 251                 amdgpu_bo_unreserve(adev->vce.vcpu_bo);
 252                 dev_err(adev->dev, "(%d) VCE map failed\n", r);
 253                 return r;
 254         }
 255
 256         hdr = (const struct common_firmware_header *)adev->vce.fw->data;
 257         offset = le32_to_cpu(hdr->ucode_array_offset_bytes);
 258         memcpy(cpu_addr, (adev->vce.fw->data) + offset,
 259                 (adev->vce.fw->size) - offset);
 260
 261         amdgpu_bo_kunmap(adev->vce.vcpu_bo);
 262
 263         amdgpu_bo_unreserve(adev->vce.vcpu_bo);
 264
 265         return 0;
 266 }
 267
 268 /**
 269  * amdgpu_vce_idle_work_handler - power off VCE
 270  *
 271  * @work: pointer to work structure
 272  *
 273  * power of VCE when it's not used any more
 274  */
 275 static void amdgpu_vce_idle_work_handler(struct work_struct *work)
 276 {
 277         struct amdgpu_device *adev =
 278                 container_of(work, struct amdgpu_device, vce.idle_work.work);
 279
 280         if ((amdgpu_fence_count_emitted(&adev->vce.ring[0]) == 0) &&
 281             (amdgpu_fence_count_emitted(&adev->vce.ring[1]) == 0)) {
 282                 if (adev->pm.dpm_enabled) {
 283                         amdgpu_dpm_enable_vce(adev, false);
 284                 } else {
 285                         amdgpu_asic_set_vce_clocks(adev, 0, 0);
 286                 }
 287         } else {
 288                 schedule_delayed_work(&adev->vce.idle_work,
 289                                       msecs_to_jiffies(VCE_IDLE_TIMEOUT_MS));
 290         }
 291 }
 292
 293 /**
 294  * amdgpu_vce_note_usage - power up VCE
 295  *
 296  * @adev: amdgpu_device pointer
 297  *
 298  * Make sure VCE is powerd up when we want to use it
 299  */
 300 static void amdgpu_vce_note_usage(struct amdgpu_device *adev)
 301 {
 302         bool streams_changed = false;
 303         bool set_clocks = !cancel_delayed_work_sync(&adev->vce.idle_work);
 304         set_clocks &= schedule_delayed_work(&adev->vce.idle_work,
 305                                             msecs_to_jiffies(VCE_IDLE_TIMEOUT_MS));
 306
 307         if (adev->pm.dpm_enabled) {
 308                 /* XXX figure out if the streams changed */
 309                 streams_changed = false;
 310         }
 311
 312         if (set_clocks || streams_changed) {
 313                 if (adev->pm.dpm_enabled) {
 314                         amdgpu_dpm_enable_vce(adev, true);
 315                 } else {
 316                         amdgpu_asic_set_vce_clocks(adev, 53300, 40000);
 317                 }
 318         }
 319 }
 320
 321 /**
 322  * amdgpu_vce_free_handles - free still open VCE handles
 323  *
 324  * @adev: amdgpu_device pointer
 325  * @filp: drm file pointer
 326  *
 327  * Close all VCE handles still open by this file pointer
 328  */
 329 void amdgpu_vce_free_handles(struct amdgpu_device *adev, struct drm_file *filp)
 330 {
 331         struct amdgpu_ring *ring = &adev->vce.ring[0];
 332         int i, r;
 333         for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) {
 334                 uint32_t handle = atomic_read(&adev->vce.handles[i]);
 335                 if (!handle || adev->vce.filp[i] != filp)
 336                         continue;
 337
 338                 amdgpu_vce_note_usage(adev);
 339
 340                 r = amdgpu_vce_get_destroy_msg(ring, handle, NULL);
 341                 if (r)
 342                         DRM_ERROR("Error destroying VCE handle (%d)!\n", r);
 343
 344                 adev->vce.filp[i] = NULL;
 345                 atomic_set(&adev->vce.handles[i], 0);
 346         }
 347 }
 348
 349 static int amdgpu_vce_free_job(
 350         struct amdgpu_job *job)
 351 {
 352         amdgpu_ib_free(job->adev, job->ibs);
 353         kfree(job->ibs);
 354         return 0;
 355 }
 356
 357 /**
 358  * amdgpu_vce_get_create_msg - generate a VCE create msg
 359  *
 360  * @adev: amdgpu_device pointer
 361  * @ring: ring we should submit the msg to
 362  * @handle: VCE session handle to use
 363  * @fence: optional fence to return
 364  *
 365  * Open up a stream for HW test
 366  */
 367 int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
 368                               struct fence **fence)
 369 {
 370         const unsigned ib_size_dw = 1024;
 371         struct amdgpu_ib *ib = NULL;
 372         struct fence *f = NULL;
 373         struct amdgpu_device *adev = ring->adev;
 374         uint64_t dummy;
 375         int i, r;
 376
 377         ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);
 378         if (!ib)
 379                 return -ENOMEM;
 380         r = amdgpu_ib_get(ring, NULL, ib_size_dw * 4, ib);
 381         if (r) {
 382                 DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
 383                 kfree(ib);
 384                 return r;
 385         }
 386
 387         dummy = ib->gpu_addr + 1024;
 388
 389         /* stitch together an VCE create msg */
 390         ib->length_dw = 0;
 391         ib->ptr[ib->length_dw++] = 0x0000000c; /* len */
 392         ib->ptr[ib->length_dw++] = 0x00000001; /* session cmd */
 393         ib->ptr[ib->length_dw++] = handle;
 394
 395         if ((ring->adev->vce.fw_version >> 24) >= 52)
 396                 ib->ptr[ib->length_dw++] = 0x00000040; /* len */
 397         else
 398                 ib->ptr[ib->length_dw++] = 0x00000030; /* len */
 399         ib->ptr[ib->length_dw++] = 0x01000001; /* create cmd */
 400         ib->ptr[ib->length_dw++] = 0x00000000;
 401         ib->ptr[ib->length_dw++] = 0x00000042;
 402         ib->ptr[ib->length_dw++] = 0x0000000a;
 403         ib->ptr[ib->length_dw++] = 0x00000001;
 404         ib->ptr[ib->length_dw++] = 0x00000080;
 405         ib->ptr[ib->length_dw++] = 0x00000060;
 406         ib->ptr[ib->length_dw++] = 0x00000100;
 407         ib->ptr[ib->length_dw++] = 0x00000100;
 408         ib->ptr[ib->length_dw++] = 0x0000000c;
 409         ib->ptr[ib->length_dw++] = 0x00000000;
 410         if ((ring->adev->vce.fw_version >> 24) >= 52) {
 411                 ib->ptr[ib->length_dw++] = 0x00000000;
 412                 ib->ptr[ib->length_dw++] = 0x00000000;
 413                 ib->ptr[ib->length_dw++] = 0x00000000;
 414                 ib->ptr[ib->length_dw++] = 0x00000000;
 415         }
 416
 417         ib->ptr[ib->length_dw++] = 0x00000014; /* len */
 418         ib->ptr[ib->length_dw++] = 0x05000005; /* feedback buffer */
 419         ib->ptr[ib->length_dw++] = upper_32_bits(dummy);
 420         ib->ptr[ib->length_dw++] = dummy;
 421         ib->ptr[ib->length_dw++] = 0x00000001;
 422
 423         for (i = ib->length_dw; i < ib_size_dw; ++i)
 424                 ib->ptr[i] = 0x0;
 425
 426         r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1,
 427                                                  &amdgpu_vce_free_job,
 428                                                  AMDGPU_FENCE_OWNER_UNDEFINED,
 429                                                  &f);
 430         if (r)
 431                 goto err;
 432         if (fence)
 433                 *fence = fence_get(f);
 434         fence_put(f);
 435         if (amdgpu_enable_scheduler)
 436                 return 0;
 437 err:
 438         amdgpu_ib_free(adev, ib);
 439         kfree(ib);
 440         return r;
 441 }
 442
 443 /**
 444  * amdgpu_vce_get_destroy_msg - generate a VCE destroy msg
 445  *
 446  * @adev: amdgpu_device pointer
 447  * @ring: ring we should submit the msg to
 448  * @handle: VCE session handle to use
 449  * @fence: optional fence to return
 450  *
 451  * Close up a stream for HW test or if userspace failed to do so
 452  */
 453 int amdgpu_vce_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
 454                                struct fence **fence)
 455 {
 456         const unsigned ib_size_dw = 1024;
 457         struct amdgpu_ib *ib = NULL;
 458         struct fence *f = NULL;
 459         struct amdgpu_device *adev = ring->adev;
 460         uint64_t dummy;
 461         int i, r;
 462
 463         ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);
 464         if (!ib)
 465                 return -ENOMEM;
 466
 467         r = amdgpu_ib_get(ring, NULL, ib_size_dw * 4, ib);
 468         if (r) {
 469                 kfree(ib);
 470                 DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
 471                 return r;
 472         }
 473
 474         dummy = ib->gpu_addr + 1024;
 475
 476         /* stitch together an VCE destroy msg */
 477         ib->length_dw = 0;
 478         ib->ptr[ib->length_dw++] = 0x0000000c; /* len */
 479         ib->ptr[ib->length_dw++] = 0x00000001; /* session cmd */
 480         ib->ptr[ib->length_dw++] = handle;
 481
 482         ib->ptr[ib->length_dw++] = 0x00000014; /* len */
 483         ib->ptr[ib->length_dw++] = 0x05000005; /* feedback buffer */
 484         ib->ptr[ib->length_dw++] = upper_32_bits(dummy);
 485         ib->ptr[ib->length_dw++] = dummy;
 486         ib->ptr[ib->length_dw++] = 0x00000001;
 487
 488         ib->ptr[ib->length_dw++] = 0x00000008; /* len */
 489         ib->ptr[ib->length_dw++] = 0x02000001; /* destroy cmd */
 490
 491         for (i = ib->length_dw; i < ib_size_dw; ++i)
 492                 ib->ptr[i] = 0x0;
 493         r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1,
 494                                                  &amdgpu_vce_free_job,
 495                                                  AMDGPU_FENCE_OWNER_UNDEFINED,
 496                                                  &f);
 497         if (r)
 498                 goto err;
 499         if (fence)
 500                 *fence = fence_get(f);
 501         fence_put(f);
 502         if (amdgpu_enable_scheduler)
 503                 return 0;
 504 err:
 505         amdgpu_ib_free(adev, ib);
 506         kfree(ib);
 507         return r;
 508 }
 509
 510 /**
 511  * amdgpu_vce_cs_reloc - command submission relocation
 512  *
 513  * @p: parser context
 514  * @lo: address of lower dword
 515  * @hi: address of higher dword
 516  * @size: minimum size
 517  *
 518  * Patch relocation inside command stream with real buffer address
 519  */
 520 static int amdgpu_vce_cs_reloc(struct amdgpu_cs_parser *p, uint32_t ib_idx,
 521                                int lo, int hi, unsigned size, uint32_t index)
 522 {
 523         struct amdgpu_bo_va_mapping *mapping;
 524         struct amdgpu_ib *ib = &p->ibs[ib_idx];
 525         struct amdgpu_bo *bo;
 526         uint64_t addr;
 527
 528         if (index == 0xffffffff)
 529                 index = 0;
 530
 531         addr = ((uint64_t)amdgpu_get_ib_value(p, ib_idx, lo)) |
 532                ((uint64_t)amdgpu_get_ib_value(p, ib_idx, hi)) << 32;
 533         addr += ((uint64_t)size) * ((uint64_t)index);
 534
 535         mapping = amdgpu_cs_find_mapping(p, addr, &bo);
 536         if (mapping == NULL) {
 537                 DRM_ERROR("Can't find BO for addr 0x%010Lx %d %d %d %d\n",
 538                           addr, lo, hi, size, index);
 539                 return -EINVAL;
 540         }
 541
 542         if ((addr + (uint64_t)size) >
 543             ((uint64_t)mapping->it.last + 1) * AMDGPU_GPU_PAGE_SIZE) {
 544                 DRM_ERROR("BO to small for addr 0x%010Lx %d %d\n",
 545                           addr, lo, hi);
 546                 return -EINVAL;
 547         }
 548
 549         addr -= ((uint64_t)mapping->it.start) * AMDGPU_GPU_PAGE_SIZE;
 550         addr += amdgpu_bo_gpu_offset(bo);
 551         addr -= ((uint64_t)size) * ((uint64_t)index);
 552
 553         ib->ptr[lo] = addr & 0xFFFFFFFF;
 554         ib->ptr[hi] = addr >> 32;
 555
 556         return 0;
 557 }
 558
 559 /**
 560  * amdgpu_vce_validate_handle - validate stream handle
 561  *
 562  * @p: parser context
 563  * @handle: handle to validate
 564  * @allocated: allocated a new handle?
 565  *
 566  * Validates the handle and return the found session index or -EINVAL
 567  * we we don't have another free session index.
 568  */
 569 static int amdgpu_vce_validate_handle(struct amdgpu_cs_parser *p,
 570                                       uint32_t handle, bool *allocated)
 571 {
 572         unsigned i;
 573
 574         *allocated = false;
 575
 576         /* validate the handle */
 577         for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) {
 578                 if (atomic_read(&p->adev->vce.handles[i]) == handle) {
 579                         if (p->adev->vce.filp[i] != p->filp) {
 580                                 DRM_ERROR("VCE handle collision detected!\n");
 581                                 return -EINVAL;
 582                         }
 583                         return i;
 584                 }
 585         }
 586
 587         /* handle not found try to alloc a new one */
 588         for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) {
 589                 if (!atomic_cmpxchg(&p->adev->vce.handles[i], 0, handle)) {
 590                         p->adev->vce.filp[i] = p->filp;
 591                         p->adev->vce.img_size[i] = 0;
 592                         *allocated = true;
 593                         return i;
 594                 }
 595         }
 596
 597         DRM_ERROR("No more free VCE handles!\n");
 598         return -EINVAL;
 599 }
 600
 601 /**
 602  * amdgpu_vce_cs_parse - parse and validate the command stream
 603  *
 604  * @p: parser context
 605  *
 606  */
 607 int amdgpu_vce_ring_parse_cs(struct amdgpu_cs_parser *p, uint32_t ib_idx)
 608 {
 609         struct amdgpu_ib *ib = &p->ibs[ib_idx];
 610         unsigned fb_idx = 0, bs_idx = 0;
 611         int session_idx = -1;
 612         bool destroyed = false;
 613         bool created = false;
 614         bool allocated = false;
 615         uint32_t tmp, handle = 0;
 616         uint32_t *size = &tmp;
 617         int i, r = 0, idx = 0;
 618
 619         amdgpu_vce_note_usage(p->adev);
 620
 621         while (idx < ib->length_dw) {
 622                 uint32_t len = amdgpu_get_ib_value(p, ib_idx, idx);
 623                 uint32_t cmd = amdgpu_get_ib_value(p, ib_idx, idx + 1);
 624
 625                 if ((len < 8) || (len & 3)) {
 626                         DRM_ERROR("invalid VCE command length (%d)!\n", len);
 627                         r = -EINVAL;
 628                         goto out;
 629                 }
 630
 631                 if (destroyed) {
 632                         DRM_ERROR("No other command allowed after destroy!\n");
 633                         r = -EINVAL;
 634                         goto out;
 635                 }
 636
 637                 switch (cmd) {
 638                 case 0x00000001: // session
 639                         handle = amdgpu_get_ib_value(p, ib_idx, idx + 2);
 640                         session_idx = amdgpu_vce_validate_handle(p, handle,
 641                                                                  &allocated);
 642                         if (session_idx < 0)
 643                                 return session_idx;
 644                         size = &p->adev->vce.img_size[session_idx];
 645                         break;
 646
 647                 case 0x00000002: // task info
 648                         fb_idx = amdgpu_get_ib_value(p, ib_idx, idx + 6);
 649                         bs_idx = amdgpu_get_ib_value(p, ib_idx, idx + 7);
 650                         break;
 651
 652                 case 0x01000001: // create
 653                         created = true;
 654                         if (!allocated) {
 655                                 DRM_ERROR("Handle already in use!\n");
 656                                 r = -EINVAL;
 657                                 goto out;
 658                         }
 659
 660                         *size = amdgpu_get_ib_value(p, ib_idx, idx + 8) *
 661                                 amdgpu_get_ib_value(p, ib_idx, idx + 10) *
 662                                 8 * 3 / 2;
 663                         break;
 664
 665                 case 0x04000001: // config extension
 666                 case 0x04000002: // pic control
 667                 case 0x04000005: // rate control
 668                 case 0x04000007: // motion estimation
 669                 case 0x04000008: // rdo
 670                 case 0x04000009: // vui
 671                 case 0x05000002: // auxiliary buffer
 672                         break;
 673
 674                 case 0x03000001: // encode
 675                         r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 10, idx + 9,
 676                                                 *size, 0);
 677                         if (r)
 678                                 goto out;
 679
 680                         r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 12, idx + 11,
 681                                                 *size / 3, 0);
 682                         if (r)
 683                                 goto out;
 684                         break;
 685
 686                 case 0x02000001: // destroy
 687                         destroyed = true;
 688                         break;
 689
 690                 case 0x05000001: // context buffer
 691                         r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 3, idx + 2,
 692                                                 *size * 2, 0);
 693                         if (r)
 694                                 goto out;
 695                         break;
 696
 697                 case 0x05000004: // video bitstream buffer
 698                         tmp = amdgpu_get_ib_value(p, ib_idx, idx + 4);
 699                         r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 3, idx + 2,
 700                                                 tmp, bs_idx);
 701                         if (r)
 702                                 goto out;
 703                         break;
 704
 705                 case 0x05000005: // feedback buffer
 706                         r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 3, idx + 2,
 707                                                 4096, fb_idx);
 708                         if (r)
 709                                 goto out;
 710                         break;
 711
 712                 default:
 713                         DRM_ERROR("invalid VCE command (0x%x)!\n", cmd);
 714                         r = -EINVAL;
 715                         goto out;
 716                 }
 717
 718                 if (session_idx == -1) {
 719                         DRM_ERROR("no session command at start of IB\n");
 720                         r = -EINVAL;
 721                         goto out;
 722                 }
 723
 724                 idx += len / 4;
 725         }
 726
 727         if (allocated && !created) {
 728                 DRM_ERROR("New session without create command!\n");
 729                 r = -ENOENT;
 730         }
 731
 732 out:
 733         if ((!r && destroyed) || (r && allocated)) {
 734                 /*
 735                  * IB contains a destroy msg or we have allocated an
 736                  * handle and got an error, anyway free the handle
 737                  */
 738                 for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i)
 739                         atomic_cmpxchg(&p->adev->vce.handles[i], handle, 0);
 740         }
 741
 742         return r;
 743 }
 744
 745 /**
 746  * amdgpu_vce_ring_emit_semaphore - emit a semaphore command
 747  *
 748  * @ring: engine to use
 749  * @semaphore: address of semaphore
 750  * @emit_wait: true=emit wait, false=emit signal
 751  *
 752  */
 753 bool amdgpu_vce_ring_emit_semaphore(struct amdgpu_ring *ring,
 754                                     struct amdgpu_semaphore *semaphore,
 755                                     bool emit_wait)
 756 {
 757         uint64_t addr = semaphore->gpu_addr;
 758
 759         amdgpu_ring_write(ring, VCE_CMD_SEMAPHORE);
 760         amdgpu_ring_write(ring, (addr >> 3) & 0x000FFFFF);
 761         amdgpu_ring_write(ring, (addr >> 23) & 0x000FFFFF);
 762         amdgpu_ring_write(ring, 0x01003000 | (emit_wait ? 1 : 0));
 763         if (!emit_wait)
 764                 amdgpu_ring_write(ring, VCE_CMD_END);
 765
 766         return true;
 767 }
 768
 769 /**
 770  * amdgpu_vce_ring_emit_ib - execute indirect buffer
 771  *
 772  * @ring: engine to use
 773  * @ib: the IB to execute
 774  *
 775  */
 776 void amdgpu_vce_ring_emit_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib)
 777 {
 778         amdgpu_ring_write(ring, VCE_CMD_IB);
 779         amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
 780         amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
 781         amdgpu_ring_write(ring, ib->length_dw);
 782 }
 783
 784 /**
 785  * amdgpu_vce_ring_emit_fence - add a fence command to the ring
 786  *
 787  * @ring: engine to use
 788  * @fence: the fence
 789  *
 790  */
 791 void amdgpu_vce_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
 792                                 unsigned flags)
 793 {
 794         WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
 795
 796         amdgpu_ring_write(ring, VCE_CMD_FENCE);
 797         amdgpu_ring_write(ring, addr);
 798         amdgpu_ring_write(ring, upper_32_bits(addr));
 799         amdgpu_ring_write(ring, seq);
 800         amdgpu_ring_write(ring, VCE_CMD_TRAP);
 801         amdgpu_ring_write(ring, VCE_CMD_END);
 802 }
 803
 804 /**
 805  * amdgpu_vce_ring_test_ring - test if VCE ring is working
 806  *
 807  * @ring: the engine to test on
 808  *
 809  */
 810 int amdgpu_vce_ring_test_ring(struct amdgpu_ring *ring)
 811 {
 812         struct amdgpu_device *adev = ring->adev;
 813         uint32_t rptr = amdgpu_ring_get_rptr(ring);
 814         unsigned i;
 815         int r;
 816
 817         r = amdgpu_ring_lock(ring, 16);
 818         if (r) {
 819                 DRM_ERROR("amdgpu: vce failed to lock ring %d (%d).\n",
 820                           ring->idx, r);
 821                 return r;
 822         }
 823         amdgpu_ring_write(ring, VCE_CMD_END);
 824         amdgpu_ring_unlock_commit(ring);
 825
 826         for (i = 0; i < adev->usec_timeout; i++) {
 827                 if (amdgpu_ring_get_rptr(ring) != rptr)
 828                         break;
 829                 DRM_UDELAY(1);
 830         }
 831
 832         if (i < adev->usec_timeout) {
 833                 DRM_INFO("ring test on %d succeeded in %d usecs\n",
 834                          ring->idx, i);
 835         } else {
 836                 DRM_ERROR("amdgpu: ring %d test failed\n",
 837                           ring->idx);
 838                 r = -ETIMEDOUT;
 839         }
 840
 841         return r;
 842 }
 843
 844 /**
 845  * amdgpu_vce_ring_test_ib - test if VCE IBs are working
 846  *
 847  * @ring: the engine to test on
 848  *
 849  */
 850 int amdgpu_vce_ring_test_ib(struct amdgpu_ring *ring)
 851 {
 852         struct fence *fence = NULL;
 853         int r;
 854
 855         /* skip vce ring1 ib test for now, since it's not reliable */
 856         if (ring == &ring->adev->vce.ring[1])
 857                 return 0;
 858
 859         r = amdgpu_vce_get_create_msg(ring, 1, NULL);
 860         if (r) {
 861                 DRM_ERROR("amdgpu: failed to get create msg (%d).\n", r);
 862                 goto error;
 863         }
 864
 865         r = amdgpu_vce_get_destroy_msg(ring, 1, &fence);
 866         if (r) {
 867                 DRM_ERROR("amdgpu: failed to get destroy ib (%d).\n", r);
 868                 goto error;
 869         }
 870
 871         r = fence_wait(fence, false);
 872         if (r) {
 873                 DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
 874         } else {
 875                 DRM_INFO("ib test on ring %d succeeded\n", ring->idx);
 876         }
 877 error:
 878         fence_put(fence);
 879         return r;
 880 }