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

   1 /*
   2  * Copyright 2016 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 #include <linux/firmware.h>
  28 #include <linux/module.h>
  29 #include <drm/drmP.h>
  30 #include <drm/drm.h>
  31
  32 #include "amdgpu.h"
  33 #include "amdgpu_pm.h"
  34 #include "amdgpu_vcn.h"
  35 #include "soc15d.h"
  36 #include "soc15_common.h"
  37
  38 #include "vcn/vcn_1_0_offset.h"
  39 #include "vcn/vcn_1_0_sh_mask.h"
  40
  41 /* 1 second timeout */
  42 #define VCN_IDLE_TIMEOUT        msecs_to_jiffies(1000)
  43
  44 /* Firmware Names */
  45 #define FIRMWARE_RAVEN          "amdgpu/raven_vcn.bin"
  46 #define FIRMWARE_PICASSO        "amdgpu/picasso_vcn.bin"
  47 #define FIRMWARE_RAVEN2         "amdgpu/raven2_vcn.bin"
  48 #define FIRMWARE_NAVI10         "amdgpu/navi10_vcn.bin"
  49
  50 MODULE_FIRMWARE(FIRMWARE_RAVEN);
  51 MODULE_FIRMWARE(FIRMWARE_PICASSO);
  52 MODULE_FIRMWARE(FIRMWARE_RAVEN2);
  53 MODULE_FIRMWARE(FIRMWARE_NAVI10);
  54
  55 static void amdgpu_vcn_idle_work_handler(struct work_struct *work);
  56
  57 int amdgpu_vcn_sw_init(struct amdgpu_device *adev)
  58 {
  59         unsigned long bo_size;
  60         const char *fw_name;
  61         const struct common_firmware_header *hdr;
  62         unsigned char fw_check;
  63         int r;
  64
  65         INIT_DELAYED_WORK(&adev->vcn.idle_work, amdgpu_vcn_idle_work_handler);
  66
  67         switch (adev->asic_type) {
  68         case CHIP_RAVEN:
  69                 if (adev->rev_id >= 8)
  70                         fw_name = FIRMWARE_RAVEN2;
  71                 else if (adev->pdev->device == 0x15d8)
  72                         fw_name = FIRMWARE_PICASSO;
  73                 else
  74                         fw_name = FIRMWARE_RAVEN;
  75                 break;
  76         case CHIP_NAVI10:
  77                 fw_name = FIRMWARE_NAVI10;
  78                 break;
  79         default:
  80                 return -EINVAL;
  81         }
  82
  83         r = request_firmware(&adev->vcn.fw, fw_name, adev->dev);
  84         if (r) {
  85                 dev_err(adev->dev, "amdgpu_vcn: Can't load firmware \"%s\"\n",
  86                         fw_name);
  87                 return r;
  88         }
  89
  90         r = amdgpu_ucode_validate(adev->vcn.fw);
  91         if (r) {
  92                 dev_err(adev->dev, "amdgpu_vcn: Can't validate firmware \"%s\"\n",
  93                         fw_name);
  94                 release_firmware(adev->vcn.fw);
  95                 adev->vcn.fw = NULL;
  96                 return r;
  97         }
  98
  99         hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
 100         adev->vcn.fw_version = le32_to_cpu(hdr->ucode_version);
 101
 102         /* Bit 20-23, it is encode major and non-zero for new naming convention.
 103          * This field is part of version minor and DRM_DISABLED_FLAG in old naming
 104          * convention. Since the l:wq!atest version minor is 0x5B and DRM_DISABLED_FLAG
 105          * is zero in old naming convention, this field is always zero so far.
 106          * These four bits are used to tell which naming convention is present.
 107          */
 108         fw_check = (le32_to_cpu(hdr->ucode_version) >> 20) & 0xf;
 109         if (fw_check) {
 110                 unsigned int dec_ver, enc_major, enc_minor, vep, fw_rev;
 111
 112                 fw_rev = le32_to_cpu(hdr->ucode_version) & 0xfff;
 113                 enc_minor = (le32_to_cpu(hdr->ucode_version) >> 12) & 0xff;
 114                 enc_major = fw_check;
 115                 dec_ver = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xf;
 116                 vep = (le32_to_cpu(hdr->ucode_version) >> 28) & 0xf;
 117                 DRM_INFO("Found VCN firmware Version ENC: %hu.%hu DEC: %hu VEP: %hu Revision: %hu\n",
 118                         enc_major, enc_minor, dec_ver, vep, fw_rev);
 119         } else {
 120                 unsigned int version_major, version_minor, family_id;
 121
 122                 family_id = le32_to_cpu(hdr->ucode_version) & 0xff;
 123                 version_major = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xff;
 124                 version_minor = (le32_to_cpu(hdr->ucode_version) >> 8) & 0xff;
 125                 DRM_INFO("Found VCN firmware Version: %hu.%hu Family ID: %hu\n",
 126                         version_major, version_minor, family_id);
 127         }
 128
 129         bo_size = AMDGPU_VCN_STACK_SIZE + AMDGPU_VCN_CONTEXT_SIZE;
 130         if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP)
 131                 bo_size += AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
 132         r = amdgpu_bo_create_kernel(adev, bo_size, PAGE_SIZE,
 133                                     AMDGPU_GEM_DOMAIN_VRAM, &adev->vcn.vcpu_bo,
 134                                     &adev->vcn.gpu_addr, &adev->vcn.cpu_addr);
 135         if (r) {
 136                 dev_err(adev->dev, "(%d) failed to allocate vcn bo\n", r);
 137                 return r;
 138         }
 139
 140         return 0;
 141 }
 142
 143 int amdgpu_vcn_sw_fini(struct amdgpu_device *adev)
 144 {
 145         int i;
 146
 147         kvfree(adev->vcn.saved_bo);
 148
 149         amdgpu_bo_free_kernel(&adev->vcn.vcpu_bo,
 150                               &adev->vcn.gpu_addr,
 151                               (void **)&adev->vcn.cpu_addr);
 152
 153         amdgpu_ring_fini(&adev->vcn.ring_dec);
 154
 155         for (i = 0; i < adev->vcn.num_enc_rings; ++i)
 156                 amdgpu_ring_fini(&adev->vcn.ring_enc[i]);
 157
 158         amdgpu_ring_fini(&adev->vcn.ring_jpeg);
 159
 160         release_firmware(adev->vcn.fw);
 161
 162         return 0;
 163 }
 164
 165 int amdgpu_vcn_suspend(struct amdgpu_device *adev)
 166 {
 167         unsigned size;
 168         void *ptr;
 169
 170         cancel_delayed_work_sync(&adev->vcn.idle_work);
 171
 172         if (adev->vcn.vcpu_bo == NULL)
 173                 return 0;
 174
 175         size = amdgpu_bo_size(adev->vcn.vcpu_bo);
 176         ptr = adev->vcn.cpu_addr;
 177
 178         adev->vcn.saved_bo = kvmalloc(size, GFP_KERNEL);
 179         if (!adev->vcn.saved_bo)
 180                 return -ENOMEM;
 181
 182         memcpy_fromio(adev->vcn.saved_bo, ptr, size);
 183
 184         return 0;
 185 }
 186
 187 int amdgpu_vcn_resume(struct amdgpu_device *adev)
 188 {
 189         unsigned size;
 190         void *ptr;
 191
 192         if (adev->vcn.vcpu_bo == NULL)
 193                 return -EINVAL;
 194
 195         size = amdgpu_bo_size(adev->vcn.vcpu_bo);
 196         ptr = adev->vcn.cpu_addr;
 197
 198         if (adev->vcn.saved_bo != NULL) {
 199                 memcpy_toio(ptr, adev->vcn.saved_bo, size);
 200                 kvfree(adev->vcn.saved_bo);
 201                 adev->vcn.saved_bo = NULL;
 202         } else {
 203                 const struct common_firmware_header *hdr;
 204                 unsigned offset;
 205
 206                 hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
 207                 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
 208                         offset = le32_to_cpu(hdr->ucode_array_offset_bytes);
 209                         memcpy_toio(adev->vcn.cpu_addr, adev->vcn.fw->data + offset,
 210                                     le32_to_cpu(hdr->ucode_size_bytes));
 211                         size -= le32_to_cpu(hdr->ucode_size_bytes);
 212                         ptr += le32_to_cpu(hdr->ucode_size_bytes);
 213                 }
 214                 memset_io(ptr, 0, size);
 215         }
 216
 217         return 0;
 218 }
 219
 220 static void amdgpu_vcn_idle_work_handler(struct work_struct *work)
 221 {
 222         struct amdgpu_device *adev =
 223                 container_of(work, struct amdgpu_device, vcn.idle_work.work);
 224         unsigned int fences = 0;
 225         unsigned int i;
 226
 227         for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
 228                 fences += amdgpu_fence_count_emitted(&adev->vcn.ring_enc[i]);
 229         }
 230
 231         if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)    {
 232                 struct dpg_pause_state new_state;
 233
 234                 if (fences)
 235                         new_state.fw_based = VCN_DPG_STATE__PAUSE;
 236                 else
 237                         new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
 238
 239                 if (amdgpu_fence_count_emitted(&adev->vcn.ring_jpeg))
 240                         new_state.jpeg = VCN_DPG_STATE__PAUSE;
 241                 else
 242                         new_state.jpeg = VCN_DPG_STATE__UNPAUSE;
 243
 244                 adev->vcn.pause_dpg_mode(adev, &new_state);
 245         }
 246
 247         fences += amdgpu_fence_count_emitted(&adev->vcn.ring_jpeg);
 248         fences += amdgpu_fence_count_emitted(&adev->vcn.ring_dec);
 249
 250         if (fences == 0) {
 251                 amdgpu_gfx_off_ctrl(adev, true);
 252                 if (adev->pm.dpm_enabled)
 253                         amdgpu_dpm_enable_uvd(adev, false);
 254                 else
 255                         amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
 256                                                                AMD_PG_STATE_GATE);
 257         } else {
 258                 schedule_delayed_work(&adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
 259         }
 260 }
 261
 262 void amdgpu_vcn_ring_begin_use(struct amdgpu_ring *ring)
 263 {
 264         struct amdgpu_device *adev = ring->adev;
 265         bool set_clocks = !cancel_delayed_work_sync(&adev->vcn.idle_work);
 266
 267         if (set_clocks) {
 268                 amdgpu_gfx_off_ctrl(adev, false);
 269                 if (adev->pm.dpm_enabled)
 270                         amdgpu_dpm_enable_uvd(adev, true);
 271                 else
 272                         amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
 273                                                                AMD_PG_STATE_UNGATE);
 274         }
 275
 276         if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)    {
 277                 struct dpg_pause_state new_state;
 278                 unsigned int fences = 0;
 279                 unsigned int i;
 280
 281                 for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
 282                         fences += amdgpu_fence_count_emitted(&adev->vcn.ring_enc[i]);
 283                 }
 284                 if (fences)
 285                         new_state.fw_based = VCN_DPG_STATE__PAUSE;
 286                 else
 287                         new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
 288
 289                 if (amdgpu_fence_count_emitted(&adev->vcn.ring_jpeg))
 290                         new_state.jpeg = VCN_DPG_STATE__PAUSE;
 291                 else
 292                         new_state.jpeg = VCN_DPG_STATE__UNPAUSE;
 293
 294                 if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC)
 295                         new_state.fw_based = VCN_DPG_STATE__PAUSE;
 296                 else if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_JPEG)
 297                         new_state.jpeg = VCN_DPG_STATE__PAUSE;
 298
 299                 adev->vcn.pause_dpg_mode(adev, &new_state);
 300         }
 301 }
 302
 303 void amdgpu_vcn_ring_end_use(struct amdgpu_ring *ring)
 304 {
 305         schedule_delayed_work(&ring->adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
 306 }
 307
 308 int amdgpu_vcn_dec_ring_test_ring(struct amdgpu_ring *ring)
 309 {
 310         struct amdgpu_device *adev = ring->adev;
 311         uint32_t tmp = 0;
 312         unsigned i;
 313         int r;
 314
 315         WREG32(adev->vcn.external.scratch9, 0xCAFEDEAD);
 316         r = amdgpu_ring_alloc(ring, 3);
 317         if (r)
 318                 return r;
 319         amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.scratch9, 0));
 320         amdgpu_ring_write(ring, 0xDEADBEEF);
 321         amdgpu_ring_commit(ring);
 322         for (i = 0; i < adev->usec_timeout; i++) {
 323                 tmp = RREG32(adev->vcn.external.scratch9);
 324                 if (tmp == 0xDEADBEEF)
 325                         break;
 326                 DRM_UDELAY(1);
 327         }
 328
 329         if (i >= adev->usec_timeout)
 330                 r = -ETIMEDOUT;
 331
 332         return r;
 333 }
 334
 335 static int amdgpu_vcn_dec_send_msg(struct amdgpu_ring *ring,
 336                                    struct amdgpu_bo *bo,
 337                                    struct dma_fence **fence)
 338 {
 339         struct amdgpu_device *adev = ring->adev;
 340         struct dma_fence *f = NULL;
 341         struct amdgpu_job *job;
 342         struct amdgpu_ib *ib;
 343         uint64_t addr;
 344         int i, r;
 345
 346         r = amdgpu_job_alloc_with_ib(adev, 64, &job);
 347         if (r)
 348                 goto err;
 349
 350         ib = &job->ibs[0];
 351         addr = amdgpu_bo_gpu_offset(bo);
 352         ib->ptr[0] = PACKET0(adev->vcn.internal.data0, 0);
 353         ib->ptr[1] = addr;
 354         ib->ptr[2] = PACKET0(adev->vcn.internal.data1, 0);
 355         ib->ptr[3] = addr >> 32;
 356         ib->ptr[4] = PACKET0(adev->vcn.internal.cmd, 0);
 357         ib->ptr[5] = 0;
 358         for (i = 6; i < 16; i += 2) {
 359                 ib->ptr[i] = PACKET0(adev->vcn.internal.nop, 0);
 360                 ib->ptr[i+1] = 0;
 361         }
 362         ib->length_dw = 16;
 363
 364         r = amdgpu_job_submit_direct(job, ring, &f);
 365         if (r)
 366                 goto err_free;
 367
 368         amdgpu_bo_fence(bo, f, false);
 369         amdgpu_bo_unreserve(bo);
 370         amdgpu_bo_unref(&bo);
 371
 372         if (fence)
 373                 *fence = dma_fence_get(f);
 374         dma_fence_put(f);
 375
 376         return 0;
 377
 378 err_free:
 379         amdgpu_job_free(job);
 380
 381 err:
 382         amdgpu_bo_unreserve(bo);
 383         amdgpu_bo_unref(&bo);
 384         return r;
 385 }
 386
 387 static int amdgpu_vcn_dec_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
 388                               struct dma_fence **fence)
 389 {
 390         struct amdgpu_device *adev = ring->adev;
 391         struct amdgpu_bo *bo = NULL;
 392         uint32_t *msg;
 393         int r, i;
 394
 395         r = amdgpu_bo_create_reserved(adev, 1024, PAGE_SIZE,
 396                                       AMDGPU_GEM_DOMAIN_VRAM,
 397                                       &bo, NULL, (void **)&msg);
 398         if (r)
 399                 return r;
 400
 401         msg[0] = cpu_to_le32(0x00000028);
 402         msg[1] = cpu_to_le32(0x00000038);
 403         msg[2] = cpu_to_le32(0x00000001);
 404         msg[3] = cpu_to_le32(0x00000000);
 405         msg[4] = cpu_to_le32(handle);
 406         msg[5] = cpu_to_le32(0x00000000);
 407         msg[6] = cpu_to_le32(0x00000001);
 408         msg[7] = cpu_to_le32(0x00000028);
 409         msg[8] = cpu_to_le32(0x00000010);
 410         msg[9] = cpu_to_le32(0x00000000);
 411         msg[10] = cpu_to_le32(0x00000007);
 412         msg[11] = cpu_to_le32(0x00000000);
 413         msg[12] = cpu_to_le32(0x00000780);
 414         msg[13] = cpu_to_le32(0x00000440);
 415         for (i = 14; i < 1024; ++i)
 416                 msg[i] = cpu_to_le32(0x0);
 417
 418         return amdgpu_vcn_dec_send_msg(ring, bo, fence);
 419 }
 420
 421 static int amdgpu_vcn_dec_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
 422                                struct dma_fence **fence)
 423 {
 424         struct amdgpu_device *adev = ring->adev;
 425         struct amdgpu_bo *bo = NULL;
 426         uint32_t *msg;
 427         int r, i;
 428
 429         r = amdgpu_bo_create_reserved(adev, 1024, PAGE_SIZE,
 430                                       AMDGPU_GEM_DOMAIN_VRAM,
 431                                       &bo, NULL, (void **)&msg);
 432         if (r)
 433                 return r;
 434
 435         msg[0] = cpu_to_le32(0x00000028);
 436         msg[1] = cpu_to_le32(0x00000018);
 437         msg[2] = cpu_to_le32(0x00000000);
 438         msg[3] = cpu_to_le32(0x00000002);
 439         msg[4] = cpu_to_le32(handle);
 440         msg[5] = cpu_to_le32(0x00000000);
 441         for (i = 6; i < 1024; ++i)
 442                 msg[i] = cpu_to_le32(0x0);
 443
 444         return amdgpu_vcn_dec_send_msg(ring, bo, fence);
 445 }
 446
 447 int amdgpu_vcn_dec_ring_test_ib(struct amdgpu_ring *ring, long timeout)
 448 {
 449         struct dma_fence *fence;
 450         long r;
 451
 452         r = amdgpu_vcn_dec_get_create_msg(ring, 1, NULL);
 453         if (r)
 454                 goto error;
 455
 456         r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &fence);
 457         if (r)
 458                 goto error;
 459
 460         r = dma_fence_wait_timeout(fence, false, timeout);
 461         if (r == 0)
 462                 r = -ETIMEDOUT;
 463         else if (r > 0)
 464                 r = 0;
 465
 466         dma_fence_put(fence);
 467 error:
 468         return r;
 469 }
 470
 471 int amdgpu_vcn_enc_ring_test_ring(struct amdgpu_ring *ring)
 472 {
 473         struct amdgpu_device *adev = ring->adev;
 474         uint32_t rptr;
 475         unsigned i;
 476         int r;
 477
 478         r = amdgpu_ring_alloc(ring, 16);
 479         if (r)
 480                 return r;
 481
 482         rptr = amdgpu_ring_get_rptr(ring);
 483
 484         amdgpu_ring_write(ring, VCN_ENC_CMD_END);
 485         amdgpu_ring_commit(ring);
 486
 487         for (i = 0; i < adev->usec_timeout; i++) {
 488                 if (amdgpu_ring_get_rptr(ring) != rptr)
 489                         break;
 490                 DRM_UDELAY(1);
 491         }
 492
 493         if (i >= adev->usec_timeout)
 494                 r = -ETIMEDOUT;
 495
 496         return r;
 497 }
 498
 499 static int amdgpu_vcn_enc_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
 500                               struct dma_fence **fence)
 501 {
 502         const unsigned ib_size_dw = 16;
 503         struct amdgpu_job *job;
 504         struct amdgpu_ib *ib;
 505         struct dma_fence *f = NULL;
 506         uint64_t dummy;
 507         int i, r;
 508
 509         r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4, &job);
 510         if (r)
 511                 return r;
 512
 513         ib = &job->ibs[0];
 514         dummy = ib->gpu_addr + 1024;
 515
 516         ib->length_dw = 0;
 517         ib->ptr[ib->length_dw++] = 0x00000018;
 518         ib->ptr[ib->length_dw++] = 0x00000001; /* session info */
 519         ib->ptr[ib->length_dw++] = handle;
 520         ib->ptr[ib->length_dw++] = upper_32_bits(dummy);
 521         ib->ptr[ib->length_dw++] = dummy;
 522         ib->ptr[ib->length_dw++] = 0x0000000b;
 523
 524         ib->ptr[ib->length_dw++] = 0x00000014;
 525         ib->ptr[ib->length_dw++] = 0x00000002; /* task info */
 526         ib->ptr[ib->length_dw++] = 0x0000001c;
 527         ib->ptr[ib->length_dw++] = 0x00000000;
 528         ib->ptr[ib->length_dw++] = 0x00000000;
 529
 530         ib->ptr[ib->length_dw++] = 0x00000008;
 531         ib->ptr[ib->length_dw++] = 0x08000001; /* op initialize */
 532
 533         for (i = ib->length_dw; i < ib_size_dw; ++i)
 534                 ib->ptr[i] = 0x0;
 535
 536         r = amdgpu_job_submit_direct(job, ring, &f);
 537         if (r)
 538                 goto err;
 539
 540         if (fence)
 541                 *fence = dma_fence_get(f);
 542         dma_fence_put(f);
 543
 544         return 0;
 545
 546 err:
 547         amdgpu_job_free(job);
 548         return r;
 549 }
 550
 551 static int amdgpu_vcn_enc_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
 552                                 struct dma_fence **fence)
 553 {
 554         const unsigned ib_size_dw = 16;
 555         struct amdgpu_job *job;
 556         struct amdgpu_ib *ib;
 557         struct dma_fence *f = NULL;
 558         uint64_t dummy;
 559         int i, r;
 560
 561         r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4, &job);
 562         if (r)
 563                 return r;
 564
 565         ib = &job->ibs[0];
 566         dummy = ib->gpu_addr + 1024;
 567
 568         ib->length_dw = 0;
 569         ib->ptr[ib->length_dw++] = 0x00000018;
 570         ib->ptr[ib->length_dw++] = 0x00000001;
 571         ib->ptr[ib->length_dw++] = handle;
 572         ib->ptr[ib->length_dw++] = upper_32_bits(dummy);
 573         ib->ptr[ib->length_dw++] = dummy;
 574         ib->ptr[ib->length_dw++] = 0x0000000b;
 575
 576         ib->ptr[ib->length_dw++] = 0x00000014;
 577         ib->ptr[ib->length_dw++] = 0x00000002;
 578         ib->ptr[ib->length_dw++] = 0x0000001c;
 579         ib->ptr[ib->length_dw++] = 0x00000000;
 580         ib->ptr[ib->length_dw++] = 0x00000000;
 581
 582         ib->ptr[ib->length_dw++] = 0x00000008;
 583         ib->ptr[ib->length_dw++] = 0x08000002; /* op close session */
 584
 585         for (i = ib->length_dw; i < ib_size_dw; ++i)
 586                 ib->ptr[i] = 0x0;
 587
 588         r = amdgpu_job_submit_direct(job, ring, &f);
 589         if (r)
 590                 goto err;
 591
 592         if (fence)
 593                 *fence = dma_fence_get(f);
 594         dma_fence_put(f);
 595
 596         return 0;
 597
 598 err:
 599         amdgpu_job_free(job);
 600         return r;
 601 }
 602
 603 int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout)
 604 {
 605         struct dma_fence *fence = NULL;
 606         long r;
 607
 608         r = amdgpu_vcn_enc_get_create_msg(ring, 1, NULL);
 609         if (r)
 610                 goto error;
 611
 612         r = amdgpu_vcn_enc_get_destroy_msg(ring, 1, &fence);
 613         if (r)
 614                 goto error;
 615
 616         r = dma_fence_wait_timeout(fence, false, timeout);
 617         if (r == 0)
 618                 r = -ETIMEDOUT;
 619         else if (r > 0)
 620                 r = 0;
 621
 622 error:
 623         dma_fence_put(fence);
 624         return r;
 625 }
 626
 627 int amdgpu_vcn_jpeg_ring_test_ring(struct amdgpu_ring *ring)
 628 {
 629         struct amdgpu_device *adev = ring->adev;
 630         uint32_t tmp = 0;
 631         unsigned i;
 632         int r;
 633
 634         WREG32(SOC15_REG_OFFSET(UVD, 0, mmUVD_SCRATCH9), 0xCAFEDEAD);
 635         r = amdgpu_ring_alloc(ring, 3);
 636
 637         if (r)
 638                 return r;
 639
 640         amdgpu_ring_write(ring,
 641                 PACKETJ(SOC15_REG_OFFSET(UVD, 0, mmUVD_SCRATCH9), 0, 0, 0));
 642         amdgpu_ring_write(ring, 0xDEADBEEF);
 643         amdgpu_ring_commit(ring);
 644
 645         for (i = 0; i < adev->usec_timeout; i++) {
 646                 tmp = RREG32(SOC15_REG_OFFSET(UVD, 0, mmUVD_SCRATCH9));
 647                 if (tmp == 0xDEADBEEF)
 648                         break;
 649                 DRM_UDELAY(1);
 650         }
 651
 652         if (i >= adev->usec_timeout)
 653                 r = -ETIMEDOUT;
 654
 655         return r;
 656 }
 657
 658 static int amdgpu_vcn_jpeg_set_reg(struct amdgpu_ring *ring, uint32_t handle,
 659                 struct dma_fence **fence)
 660 {
 661         struct amdgpu_device *adev = ring->adev;
 662         struct amdgpu_job *job;
 663         struct amdgpu_ib *ib;
 664         struct dma_fence *f = NULL;
 665         const unsigned ib_size_dw = 16;
 666         int i, r;
 667
 668         r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4, &job);
 669         if (r)
 670                 return r;
 671
 672         ib = &job->ibs[0];
 673
 674         ib->ptr[0] = PACKETJ(SOC15_REG_OFFSET(UVD, 0, mmUVD_SCRATCH9), 0, 0, PACKETJ_TYPE0);
 675         ib->ptr[1] = 0xDEADBEEF;
 676         for (i = 2; i < 16; i += 2) {
 677                 ib->ptr[i] = PACKETJ(0, 0, 0, PACKETJ_TYPE6);
 678                 ib->ptr[i+1] = 0;
 679         }
 680         ib->length_dw = 16;
 681
 682         r = amdgpu_job_submit_direct(job, ring, &f);
 683         if (r)
 684                 goto err;
 685
 686         if (fence)
 687                 *fence = dma_fence_get(f);
 688         dma_fence_put(f);
 689
 690         return 0;
 691
 692 err:
 693         amdgpu_job_free(job);
 694         return r;
 695 }
 696
 697 int amdgpu_vcn_jpeg_ring_test_ib(struct amdgpu_ring *ring, long timeout)
 698 {
 699         struct amdgpu_device *adev = ring->adev;
 700         uint32_t tmp = 0;
 701         unsigned i;
 702         struct dma_fence *fence = NULL;
 703         long r = 0;
 704
 705         r = amdgpu_vcn_jpeg_set_reg(ring, 1, &fence);
 706         if (r)
 707                 goto error;
 708
 709         r = dma_fence_wait_timeout(fence, false, timeout);
 710         if (r == 0) {
 711                 r = -ETIMEDOUT;
 712                 goto error;
 713         } else if (r < 0) {
 714                 goto error;
 715         } else {
 716                 r = 0;
 717         }
 718
 719         for (i = 0; i < adev->usec_timeout; i++) {
 720                 tmp = RREG32(SOC15_REG_OFFSET(UVD, 0, mmUVD_SCRATCH9));
 721                 if (tmp == 0xDEADBEEF)
 722                         break;
 723                 DRM_UDELAY(1);
 724         }
 725
 726         if (i >= adev->usec_timeout)
 727                 r = -ETIMEDOUT;
 728
 729         dma_fence_put(fence);
 730 error:
 731         return r;
 732 }