drivers/gpu/drm/radeon/radeon_test.c

   1 /*
   2  * Copyright 2009 VMware, Inc.
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice shall be included in
  12  * all copies or substantial portions of the Software.
  13  *
  14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
  18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  20  * OTHER DEALINGS IN THE SOFTWARE.
  21  *
  22  * Authors: Michel Dänzer
  23  */
  24 #include <drm/drmP.h>
  25 #include <drm/radeon_drm.h>
  26 #include "radeon_reg.h"
  27 #include "radeon.h"
  28
  29
  30 /* Test BO GTT->VRAM and VRAM->GTT GPU copies across the whole GTT aperture */
  31 void radeon_test_moves(struct radeon_device *rdev)
  32 {
  33         struct radeon_bo *vram_obj = NULL;
  34         struct radeon_bo **gtt_obj = NULL;
  35         struct radeon_fence *fence = NULL;
  36         uint64_t gtt_addr, vram_addr;
  37         unsigned i, n, size;
  38         int r;
  39
  40         size = 1024 * 1024;
  41
  42         /* Number of tests =
  43          * (Total GTT - IB pool - writeback page - ring buffers) / test size
  44          */
  45         n = rdev->mc.gtt_size - RADEON_IB_POOL_SIZE*64*1024;
  46         for (i = 0; i < RADEON_NUM_RINGS; ++i)
  47                 n -= rdev->ring[i].ring_size;
  48         if (rdev->wb.wb_obj)
  49                 n -= RADEON_GPU_PAGE_SIZE;
  50         if (rdev->ih.ring_obj)
  51                 n -= rdev->ih.ring_size;
  52         n /= size;
  53
  54         gtt_obj = kzalloc(n * sizeof(*gtt_obj), GFP_KERNEL);
  55         if (!gtt_obj) {
  56                 DRM_ERROR("Failed to allocate %d pointers\n", n);
  57                 r = 1;
  58                 goto out_cleanup;
  59         }
  60
  61         r = radeon_bo_create(rdev, size, PAGE_SIZE, true, RADEON_GEM_DOMAIN_VRAM,
  62                              NULL, &vram_obj);
  63         if (r) {
  64                 DRM_ERROR("Failed to create VRAM object\n");
  65                 goto out_cleanup;
  66         }
  67         r = radeon_bo_reserve(vram_obj, false);
  68         if (unlikely(r != 0))
  69                 goto out_cleanup;
  70         r = radeon_bo_pin(vram_obj, RADEON_GEM_DOMAIN_VRAM, &vram_addr);
  71         if (r) {
  72                 DRM_ERROR("Failed to pin VRAM object\n");
  73                 goto out_cleanup;
  74         }
  75         for (i = 0; i < n; i++) {
  76                 void *gtt_map, *vram_map;
  77                 void **gtt_start, **gtt_end;
  78                 void **vram_start, **vram_end;
  79
  80                 r = radeon_bo_create(rdev, size, PAGE_SIZE, true,
  81                                      RADEON_GEM_DOMAIN_GTT, NULL, gtt_obj + i);
  82                 if (r) {
  83                         DRM_ERROR("Failed to create GTT object %d\n", i);
  84                         goto out_cleanup;
  85                 }
  86
  87                 r = radeon_bo_reserve(gtt_obj[i], false);
  88                 if (unlikely(r != 0))
  89                         goto out_cleanup;
  90                 r = radeon_bo_pin(gtt_obj[i], RADEON_GEM_DOMAIN_GTT, &gtt_addr);
  91                 if (r) {
  92                         DRM_ERROR("Failed to pin GTT object %d\n", i);
  93                         goto out_cleanup;
  94                 }
  95
  96                 r = radeon_bo_kmap(gtt_obj[i], &gtt_map);
  97                 if (r) {
  98                         DRM_ERROR("Failed to map GTT object %d\n", i);
  99                         goto out_cleanup;
 100                 }
 101
 102                 for (gtt_start = gtt_map, gtt_end = gtt_map + size;
 103                      gtt_start < gtt_end;
 104                      gtt_start++)
 105                         *gtt_start = gtt_start;
 106
 107                 radeon_bo_kunmap(gtt_obj[i]);
 108
 109                 r = radeon_copy(rdev, gtt_addr, vram_addr, size / RADEON_GPU_PAGE_SIZE, &fence);
 110                 if (r) {
 111                         DRM_ERROR("Failed GTT->VRAM copy %d\n", i);
 112                         goto out_cleanup;
 113                 }
 114
 115                 r = radeon_fence_wait(fence, false);
 116                 if (r) {
 117                         DRM_ERROR("Failed to wait for GTT->VRAM fence %d\n", i);
 118                         goto out_cleanup;
 119                 }
 120
 121                 radeon_fence_unref(&fence);
 122
 123                 r = radeon_bo_kmap(vram_obj, &vram_map);
 124                 if (r) {
 125                         DRM_ERROR("Failed to map VRAM object after copy %d\n", i);
 126                         goto out_cleanup;
 127                 }
 128
 129                 for (gtt_start = gtt_map, gtt_end = gtt_map + size,
 130                      vram_start = vram_map, vram_end = vram_map + size;
 131                      vram_start < vram_end;
 132                      gtt_start++, vram_start++) {
 133                         if (*vram_start != gtt_start) {
 134                                 DRM_ERROR("Incorrect GTT->VRAM copy %d: Got 0x%p, "
 135                                           "expected 0x%p (GTT/VRAM offset "
 136                                           "0x%16llx/0x%16llx)\n",
 137                                           i, *vram_start, gtt_start,
 138                                           (unsigned long long)
 139                                           (gtt_addr - rdev->mc.gtt_start +
 140                                            (void*)gtt_start - gtt_map),
 141                                           (unsigned long long)
 142                                           (vram_addr - rdev->mc.vram_start +
 143                                            (void*)gtt_start - gtt_map));
 144                                 radeon_bo_kunmap(vram_obj);
 145                                 goto out_cleanup;
 146                         }
 147                         *vram_start = vram_start;
 148                 }
 149
 150                 radeon_bo_kunmap(vram_obj);
 151
 152                 r = radeon_copy(rdev, vram_addr, gtt_addr, size / RADEON_GPU_PAGE_SIZE, &fence);
 153                 if (r) {
 154                         DRM_ERROR("Failed VRAM->GTT copy %d\n", i);
 155                         goto out_cleanup;
 156                 }
 157
 158                 r = radeon_fence_wait(fence, false);
 159                 if (r) {
 160                         DRM_ERROR("Failed to wait for VRAM->GTT fence %d\n", i);
 161                         goto out_cleanup;
 162                 }
 163
 164                 radeon_fence_unref(&fence);
 165
 166                 r = radeon_bo_kmap(gtt_obj[i], &gtt_map);
 167                 if (r) {
 168                         DRM_ERROR("Failed to map GTT object after copy %d\n", i);
 169                         goto out_cleanup;
 170                 }
 171
 172                 for (gtt_start = gtt_map, gtt_end = gtt_map + size,
 173                      vram_start = vram_map, vram_end = vram_map + size;
 174                      gtt_start < gtt_end;
 175                      gtt_start++, vram_start++) {
 176                         if (*gtt_start != vram_start) {
 177                                 DRM_ERROR("Incorrect VRAM->GTT copy %d: Got 0x%p, "
 178                                           "expected 0x%p (VRAM/GTT offset "
 179                                           "0x%16llx/0x%16llx)\n",
 180                                           i, *gtt_start, vram_start,
 181                                           (unsigned long long)
 182                                           (vram_addr - rdev->mc.vram_start +
 183                                            (void*)vram_start - vram_map),
 184                                           (unsigned long long)
 185                                           (gtt_addr - rdev->mc.gtt_start +
 186                                            (void*)vram_start - vram_map));
 187                                 radeon_bo_kunmap(gtt_obj[i]);
 188                                 goto out_cleanup;
 189                         }
 190                 }
 191
 192                 radeon_bo_kunmap(gtt_obj[i]);
 193
 194                 DRM_INFO("Tested GTT->VRAM and VRAM->GTT copy for GTT offset 0x%llx\n",
 195                          gtt_addr - rdev->mc.gtt_start);
 196         }
 197
 198 out_cleanup:
 199         if (vram_obj) {
 200                 if (radeon_bo_is_reserved(vram_obj)) {
 201                         radeon_bo_unpin(vram_obj);
 202                         radeon_bo_unreserve(vram_obj);
 203                 }
 204                 radeon_bo_unref(&vram_obj);
 205         }
 206         if (gtt_obj) {
 207                 for (i = 0; i < n; i++) {
 208                         if (gtt_obj[i]) {
 209                                 if (radeon_bo_is_reserved(gtt_obj[i])) {
 210                                         radeon_bo_unpin(gtt_obj[i]);
 211                                         radeon_bo_unreserve(gtt_obj[i]);
 212                                 }
 213                                 radeon_bo_unref(&gtt_obj[i]);
 214                         }
 215                 }
 216                 kfree(gtt_obj);
 217         }
 218         if (fence) {
 219                 radeon_fence_unref(&fence);
 220         }
 221         if (r) {
 222                 printk(KERN_WARNING "Error while testing BO move.\n");
 223         }
 224 }
 225
 226 void radeon_test_ring_sync(struct radeon_device *rdev,
 227                            struct radeon_ring *ringA,
 228                            struct radeon_ring *ringB)
 229 {
 230         struct radeon_fence *fence1 = NULL, *fence2 = NULL;
 231         struct radeon_semaphore *semaphore = NULL;
 232         int ridxA = radeon_ring_index(rdev, ringA);
 233         int ridxB = radeon_ring_index(rdev, ringB);
 234         int r;
 235
 236         r = radeon_semaphore_create(rdev, &semaphore);
 237         if (r) {
 238                 DRM_ERROR("Failed to create semaphore\n");
 239                 goto out_cleanup;
 240         }
 241
 242         r = radeon_ring_lock(rdev, ringA, 64);
 243         if (r) {
 244                 DRM_ERROR("Failed to lock ring A %d\n", ridxA);
 245                 goto out_cleanup;
 246         }
 247         radeon_semaphore_emit_wait(rdev, ridxA, semaphore);
 248         r = radeon_fence_emit(rdev, &fence1, ridxA);
 249         if (r) {
 250                 DRM_ERROR("Failed to emit fence 1\n");
 251                 radeon_ring_unlock_undo(rdev, ringA);
 252                 goto out_cleanup;
 253         }
 254         radeon_semaphore_emit_wait(rdev, ridxA, semaphore);
 255         r = radeon_fence_emit(rdev, &fence2, ridxA);
 256         if (r) {
 257                 DRM_ERROR("Failed to emit fence 2\n");
 258                 radeon_ring_unlock_undo(rdev, ringA);
 259                 goto out_cleanup;
 260         }
 261         radeon_ring_unlock_commit(rdev, ringA);
 262
 263         mdelay(1000);
 264
 265         if (radeon_fence_signaled(fence1)) {
 266                 DRM_ERROR("Fence 1 signaled without waiting for semaphore.\n");
 267                 goto out_cleanup;
 268         }
 269
 270         r = radeon_ring_lock(rdev, ringB, 64);
 271         if (r) {
 272                 DRM_ERROR("Failed to lock ring B %p\n", ringB);
 273                 goto out_cleanup;
 274         }
 275         radeon_semaphore_emit_signal(rdev, ridxB, semaphore);
 276         radeon_ring_unlock_commit(rdev, ringB);
 277
 278         r = radeon_fence_wait(fence1, false);
 279         if (r) {
 280                 DRM_ERROR("Failed to wait for sync fence 1\n");
 281                 goto out_cleanup;
 282         }
 283
 284         mdelay(1000);
 285
 286         if (radeon_fence_signaled(fence2)) {
 287                 DRM_ERROR("Fence 2 signaled without waiting for semaphore.\n");
 288                 goto out_cleanup;
 289         }
 290
 291         r = radeon_ring_lock(rdev, ringB, 64);
 292         if (r) {
 293                 DRM_ERROR("Failed to lock ring B %p\n", ringB);
 294                 goto out_cleanup;
 295         }
 296         radeon_semaphore_emit_signal(rdev, ridxB, semaphore);
 297         radeon_ring_unlock_commit(rdev, ringB);
 298
 299         r = radeon_fence_wait(fence2, false);
 300         if (r) {
 301                 DRM_ERROR("Failed to wait for sync fence 1\n");
 302                 goto out_cleanup;
 303         }
 304
 305 out_cleanup:
 306         radeon_semaphore_free(rdev, &semaphore, NULL);
 307
 308         if (fence1)
 309                 radeon_fence_unref(&fence1);
 310
 311         if (fence2)
 312                 radeon_fence_unref(&fence2);
 313
 314         if (r)
 315                 printk(KERN_WARNING "Error while testing ring sync (%d).\n", r);
 316 }
 317
 318 void radeon_test_ring_sync2(struct radeon_device *rdev,
 319                             struct radeon_ring *ringA,
 320                             struct radeon_ring *ringB,
 321                             struct radeon_ring *ringC)
 322 {
 323         struct radeon_fence *fenceA = NULL, *fenceB = NULL;
 324         struct radeon_semaphore *semaphore = NULL;
 325         int ridxA = radeon_ring_index(rdev, ringA);
 326         int ridxB = radeon_ring_index(rdev, ringB);
 327         int ridxC = radeon_ring_index(rdev, ringC);
 328         bool sigA, sigB;
 329         int i, r;
 330
 331         r = radeon_semaphore_create(rdev, &semaphore);
 332         if (r) {
 333                 DRM_ERROR("Failed to create semaphore\n");
 334                 goto out_cleanup;
 335         }
 336
 337         r = radeon_ring_lock(rdev, ringA, 64);
 338         if (r) {
 339                 DRM_ERROR("Failed to lock ring A %d\n", ridxA);
 340                 goto out_cleanup;
 341         }
 342         radeon_semaphore_emit_wait(rdev, ridxA, semaphore);
 343         r = radeon_fence_emit(rdev, &fenceA, ridxA);
 344         if (r) {
 345                 DRM_ERROR("Failed to emit sync fence 1\n");
 346                 radeon_ring_unlock_undo(rdev, ringA);
 347                 goto out_cleanup;
 348         }
 349         radeon_ring_unlock_commit(rdev, ringA);
 350
 351         r = radeon_ring_lock(rdev, ringB, 64);
 352         if (r) {
 353                 DRM_ERROR("Failed to lock ring B %d\n", ridxB);
 354                 goto out_cleanup;
 355         }
 356         radeon_semaphore_emit_wait(rdev, ridxB, semaphore);
 357         r = radeon_fence_emit(rdev, &fenceB, ridxB);
 358         if (r) {
 359                 DRM_ERROR("Failed to create sync fence 2\n");
 360                 radeon_ring_unlock_undo(rdev, ringB);
 361                 goto out_cleanup;
 362         }
 363         radeon_ring_unlock_commit(rdev, ringB);
 364
 365         mdelay(1000);
 366
 367         if (radeon_fence_signaled(fenceA)) {
 368                 DRM_ERROR("Fence A signaled without waiting for semaphore.\n");
 369                 goto out_cleanup;
 370         }
 371         if (radeon_fence_signaled(fenceB)) {
 372                 DRM_ERROR("Fence A signaled without waiting for semaphore.\n");
 373                 goto out_cleanup;
 374         }
 375
 376         r = radeon_ring_lock(rdev, ringC, 64);
 377         if (r) {
 378                 DRM_ERROR("Failed to lock ring B %p\n", ringC);
 379                 goto out_cleanup;
 380         }
 381         radeon_semaphore_emit_signal(rdev, ridxC, semaphore);
 382         radeon_ring_unlock_commit(rdev, ringC);
 383
 384         for (i = 0; i < 30; ++i) {
 385                 mdelay(100);
 386                 sigA = radeon_fence_signaled(fenceA);
 387                 sigB = radeon_fence_signaled(fenceB);
 388                 if (sigA || sigB)
 389                         break;
 390         }
 391
 392         if (!sigA && !sigB) {
 393                 DRM_ERROR("Neither fence A nor B has been signaled\n");
 394                 goto out_cleanup;
 395         } else if (sigA && sigB) {
 396                 DRM_ERROR("Both fence A and B has been signaled\n");
 397                 goto out_cleanup;
 398         }
 399
 400         DRM_INFO("Fence %c was first signaled\n", sigA ? 'A' : 'B');
 401
 402         r = radeon_ring_lock(rdev, ringC, 64);
 403         if (r) {
 404                 DRM_ERROR("Failed to lock ring B %p\n", ringC);
 405                 goto out_cleanup;
 406         }
 407         radeon_semaphore_emit_signal(rdev, ridxC, semaphore);
 408         radeon_ring_unlock_commit(rdev, ringC);
 409
 410         mdelay(1000);
 411
 412         r = radeon_fence_wait(fenceA, false);
 413         if (r) {
 414                 DRM_ERROR("Failed to wait for sync fence A\n");
 415                 goto out_cleanup;
 416         }
 417         r = radeon_fence_wait(fenceB, false);
 418         if (r) {
 419                 DRM_ERROR("Failed to wait for sync fence B\n");
 420                 goto out_cleanup;
 421         }
 422
 423 out_cleanup:
 424         radeon_semaphore_free(rdev, &semaphore, NULL);
 425
 426         if (fenceA)
 427                 radeon_fence_unref(&fenceA);
 428
 429         if (fenceB)
 430                 radeon_fence_unref(&fenceB);
 431
 432         if (r)
 433                 printk(KERN_WARNING "Error while testing ring sync (%d).\n", r);
 434 }
 435
 436 void radeon_test_syncing(struct radeon_device *rdev)
 437 {
 438         int i, j, k;
 439
 440         for (i = 1; i < RADEON_NUM_RINGS; ++i) {
 441                 struct radeon_ring *ringA = &rdev->ring[i];
 442                 if (!ringA->ready)
 443                         continue;
 444
 445                 for (j = 0; j < i; ++j) {
 446                         struct radeon_ring *ringB = &rdev->ring[j];
 447                         if (!ringB->ready)
 448                                 continue;
 449
 450                         DRM_INFO("Testing syncing between rings %d and %d...\n", i, j);
 451                         radeon_test_ring_sync(rdev, ringA, ringB);
 452
 453                         DRM_INFO("Testing syncing between rings %d and %d...\n", j, i);
 454                         radeon_test_ring_sync(rdev, ringB, ringA);
 455
 456                         for (k = 0; k < j; ++k) {
 457                                 struct radeon_ring *ringC = &rdev->ring[k];
 458                                 if (!ringC->ready)
 459                                         continue;
 460
 461                                 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", i, j, k);
 462                                 radeon_test_ring_sync2(rdev, ringA, ringB, ringC);
 463
 464                                 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", i, k, j);
 465                                 radeon_test_ring_sync2(rdev, ringA, ringC, ringB);
 466
 467                                 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", j, i, k);
 468                                 radeon_test_ring_sync2(rdev, ringB, ringA, ringC);
 469
 470                                 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", j, k, i);
 471                                 radeon_test_ring_sync2(rdev, ringB, ringC, ringA);
 472
 473                                 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", k, i, j);
 474                                 radeon_test_ring_sync2(rdev, ringC, ringA, ringB);
 475
 476                                 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", k, j, i);
 477                                 radeon_test_ring_sync2(rdev, ringC, ringB, ringA);
 478                         }
 479                 }
 480         }
 481 }