Rev 5271 | Details | Compare with Previous | Last modification | View Log | RSS feed
| Rev | Author | Line No. | Line |
|---|---|---|---|
| 5078 | serge | 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 |
||
| 25 | #include |
||
| 26 | #include "radeon_reg.h" |
||
| 27 | #include "radeon.h" |
||
| 28 | |||
| 29 | #define RADEON_TEST_COPY_BLIT 1 |
||
| 30 | #define RADEON_TEST_COPY_DMA 0 |
||
| 31 | |||
| 32 | |||
| 33 | /* Test BO GTT->VRAM and VRAM->GTT GPU copies across the whole GTT aperture */ |
||
| 34 | static void radeon_do_test_moves(struct radeon_device *rdev, int flag) |
||
| 35 | { |
||
| 36 | struct radeon_bo *vram_obj = NULL; |
||
| 37 | struct radeon_bo **gtt_obj = NULL; |
||
| 38 | uint64_t gtt_addr, vram_addr; |
||
| 39 | unsigned n, size; |
||
| 40 | int i, r, ring; |
||
| 41 | |||
| 42 | switch (flag) { |
||
| 43 | case RADEON_TEST_COPY_DMA: |
||
| 44 | ring = radeon_copy_dma_ring_index(rdev); |
||
| 45 | break; |
||
| 46 | case RADEON_TEST_COPY_BLIT: |
||
| 47 | ring = radeon_copy_blit_ring_index(rdev); |
||
| 48 | break; |
||
| 49 | default: |
||
| 50 | DRM_ERROR("Unknown copy method\n"); |
||
| 51 | return; |
||
| 52 | } |
||
| 53 | |||
| 54 | size = 1024 * 1024; |
||
| 55 | |||
| 56 | /* Number of tests = |
||
| 57 | * (Total GTT - IB pool - writeback page - ring buffers) / test size |
||
| 58 | */ |
||
| 59 | n = rdev->mc.gtt_size - rdev->gart_pin_size; |
||
| 60 | n /= size; |
||
| 61 | |||
| 62 | gtt_obj = kzalloc(n * sizeof(*gtt_obj), GFP_KERNEL); |
||
| 63 | if (!gtt_obj) { |
||
| 64 | DRM_ERROR("Failed to allocate %d pointers\n", n); |
||
| 65 | r = 1; |
||
| 66 | goto out_cleanup; |
||
| 67 | } |
||
| 68 | |||
| 69 | r = radeon_bo_create(rdev, size, PAGE_SIZE, true, RADEON_GEM_DOMAIN_VRAM, |
||
| 5271 | serge | 70 | 0, NULL, NULL, &vram_obj); |
| 5078 | serge | 71 | if (r) { |
| 72 | DRM_ERROR("Failed to create VRAM object\n"); |
||
| 73 | goto out_cleanup; |
||
| 74 | } |
||
| 75 | r = radeon_bo_reserve(vram_obj, false); |
||
| 76 | if (unlikely(r != 0)) |
||
| 77 | goto out_unref; |
||
| 78 | r = radeon_bo_pin(vram_obj, RADEON_GEM_DOMAIN_VRAM, &vram_addr); |
||
| 79 | if (r) { |
||
| 80 | DRM_ERROR("Failed to pin VRAM object\n"); |
||
| 81 | goto out_unres; |
||
| 82 | } |
||
| 83 | for (i = 0; i < n; i++) { |
||
| 84 | void *gtt_map, *vram_map; |
||
| 85 | void **gtt_start, **gtt_end; |
||
| 86 | void **vram_start, **vram_end; |
||
| 87 | struct radeon_fence *fence = NULL; |
||
| 88 | |||
| 89 | r = radeon_bo_create(rdev, size, PAGE_SIZE, true, |
||
| 5271 | serge | 90 | RADEON_GEM_DOMAIN_GTT, 0, NULL, NULL, |
| 91 | gtt_obj + i); |
||
| 5078 | serge | 92 | if (r) { |
| 93 | DRM_ERROR("Failed to create GTT object %d\n", i); |
||
| 94 | goto out_lclean; |
||
| 95 | } |
||
| 96 | |||
| 97 | r = radeon_bo_reserve(gtt_obj[i], false); |
||
| 98 | if (unlikely(r != 0)) |
||
| 99 | goto out_lclean_unref; |
||
| 100 | r = radeon_bo_pin(gtt_obj[i], RADEON_GEM_DOMAIN_GTT, >t_addr); |
||
| 101 | if (r) { |
||
| 102 | DRM_ERROR("Failed to pin GTT object %d\n", i); |
||
| 103 | goto out_lclean_unres; |
||
| 104 | } |
||
| 105 | |||
| 106 | r = radeon_bo_kmap(gtt_obj[i], >t_map); |
||
| 107 | if (r) { |
||
| 108 | DRM_ERROR("Failed to map GTT object %d\n", i); |
||
| 109 | goto out_lclean_unpin; |
||
| 110 | } |
||
| 111 | |||
| 112 | for (gtt_start = gtt_map, gtt_end = gtt_map + size; |
||
| 113 | gtt_start < gtt_end; |
||
| 114 | gtt_start++) |
||
| 115 | *gtt_start = gtt_start; |
||
| 116 | |||
| 117 | radeon_bo_kunmap(gtt_obj[i]); |
||
| 118 | |||
| 119 | if (ring == R600_RING_TYPE_DMA_INDEX) |
||
| 5271 | serge | 120 | fence = radeon_copy_dma(rdev, gtt_addr, vram_addr, |
| 121 | size / RADEON_GPU_PAGE_SIZE, |
||
| 6104 | serge | 122 | vram_obj->tbo.resv); |
| 5078 | serge | 123 | else |
| 5271 | serge | 124 | fence = radeon_copy_blit(rdev, gtt_addr, vram_addr, |
| 125 | size / RADEON_GPU_PAGE_SIZE, |
||
| 6104 | serge | 126 | vram_obj->tbo.resv); |
| 5271 | serge | 127 | if (IS_ERR(fence)) { |
| 5078 | serge | 128 | DRM_ERROR("Failed GTT->VRAM copy %d\n", i); |
| 5271 | serge | 129 | r = PTR_ERR(fence); |
| 5078 | serge | 130 | goto out_lclean_unpin; |
| 131 | } |
||
| 132 | |||
| 133 | r = radeon_fence_wait(fence, false); |
||
| 134 | if (r) { |
||
| 135 | DRM_ERROR("Failed to wait for GTT->VRAM fence %d\n", i); |
||
| 136 | goto out_lclean_unpin; |
||
| 137 | } |
||
| 138 | |||
| 139 | radeon_fence_unref(&fence); |
||
| 140 | |||
| 141 | r = radeon_bo_kmap(vram_obj, &vram_map); |
||
| 142 | if (r) { |
||
| 143 | DRM_ERROR("Failed to map VRAM object after copy %d\n", i); |
||
| 144 | goto out_lclean_unpin; |
||
| 145 | } |
||
| 146 | |||
| 147 | for (gtt_start = gtt_map, gtt_end = gtt_map + size, |
||
| 148 | vram_start = vram_map, vram_end = vram_map + size; |
||
| 149 | vram_start < vram_end; |
||
| 150 | gtt_start++, vram_start++) { |
||
| 151 | if (*vram_start != gtt_start) { |
||
| 152 | DRM_ERROR("Incorrect GTT->VRAM copy %d: Got 0x%p, " |
||
| 153 | "expected 0x%p (GTT/VRAM offset " |
||
| 154 | "0x%16llx/0x%16llx)\n", |
||
| 155 | i, *vram_start, gtt_start, |
||
| 156 | (unsigned long long) |
||
| 157 | (gtt_addr - rdev->mc.gtt_start + |
||
| 158 | (void*)gtt_start - gtt_map), |
||
| 159 | (unsigned long long) |
||
| 160 | (vram_addr - rdev->mc.vram_start + |
||
| 161 | (void*)gtt_start - gtt_map)); |
||
| 162 | radeon_bo_kunmap(vram_obj); |
||
| 163 | goto out_lclean_unpin; |
||
| 164 | } |
||
| 165 | *vram_start = vram_start; |
||
| 166 | } |
||
| 167 | |||
| 168 | radeon_bo_kunmap(vram_obj); |
||
| 169 | |||
| 170 | if (ring == R600_RING_TYPE_DMA_INDEX) |
||
| 5271 | serge | 171 | fence = radeon_copy_dma(rdev, vram_addr, gtt_addr, |
| 172 | size / RADEON_GPU_PAGE_SIZE, |
||
| 6104 | serge | 173 | vram_obj->tbo.resv); |
| 5078 | serge | 174 | else |
| 5271 | serge | 175 | fence = radeon_copy_blit(rdev, vram_addr, gtt_addr, |
| 176 | size / RADEON_GPU_PAGE_SIZE, |
||
| 6104 | serge | 177 | vram_obj->tbo.resv); |
| 5271 | serge | 178 | if (IS_ERR(fence)) { |
| 5078 | serge | 179 | DRM_ERROR("Failed VRAM->GTT copy %d\n", i); |
| 5271 | serge | 180 | r = PTR_ERR(fence); |
| 5078 | serge | 181 | goto out_lclean_unpin; |
| 182 | } |
||
| 183 | |||
| 184 | r = radeon_fence_wait(fence, false); |
||
| 185 | if (r) { |
||
| 186 | DRM_ERROR("Failed to wait for VRAM->GTT fence %d\n", i); |
||
| 187 | goto out_lclean_unpin; |
||
| 188 | } |
||
| 189 | |||
| 190 | radeon_fence_unref(&fence); |
||
| 191 | |||
| 192 | r = radeon_bo_kmap(gtt_obj[i], >t_map); |
||
| 193 | if (r) { |
||
| 194 | DRM_ERROR("Failed to map GTT object after copy %d\n", i); |
||
| 195 | goto out_lclean_unpin; |
||
| 196 | } |
||
| 197 | |||
| 198 | for (gtt_start = gtt_map, gtt_end = gtt_map + size, |
||
| 199 | vram_start = vram_map, vram_end = vram_map + size; |
||
| 200 | gtt_start < gtt_end; |
||
| 201 | gtt_start++, vram_start++) { |
||
| 202 | if (*gtt_start != vram_start) { |
||
| 203 | DRM_ERROR("Incorrect VRAM->GTT copy %d: Got 0x%p, " |
||
| 204 | "expected 0x%p (VRAM/GTT offset " |
||
| 205 | "0x%16llx/0x%16llx)\n", |
||
| 206 | i, *gtt_start, vram_start, |
||
| 207 | (unsigned long long) |
||
| 208 | (vram_addr - rdev->mc.vram_start + |
||
| 209 | (void*)vram_start - vram_map), |
||
| 210 | (unsigned long long) |
||
| 211 | (gtt_addr - rdev->mc.gtt_start + |
||
| 212 | (void*)vram_start - vram_map)); |
||
| 213 | radeon_bo_kunmap(gtt_obj[i]); |
||
| 214 | goto out_lclean_unpin; |
||
| 215 | } |
||
| 216 | } |
||
| 217 | |||
| 218 | radeon_bo_kunmap(gtt_obj[i]); |
||
| 219 | |||
| 220 | DRM_INFO("Tested GTT->VRAM and VRAM->GTT copy for GTT offset 0x%llx\n", |
||
| 221 | gtt_addr - rdev->mc.gtt_start); |
||
| 222 | continue; |
||
| 223 | |||
| 224 | out_lclean_unpin: |
||
| 225 | radeon_bo_unpin(gtt_obj[i]); |
||
| 226 | out_lclean_unres: |
||
| 227 | radeon_bo_unreserve(gtt_obj[i]); |
||
| 228 | out_lclean_unref: |
||
| 229 | radeon_bo_unref(>t_obj[i]); |
||
| 230 | out_lclean: |
||
| 231 | for (--i; i >= 0; --i) { |
||
| 232 | radeon_bo_unpin(gtt_obj[i]); |
||
| 233 | radeon_bo_unreserve(gtt_obj[i]); |
||
| 234 | radeon_bo_unref(>t_obj[i]); |
||
| 235 | } |
||
| 5271 | serge | 236 | if (fence && !IS_ERR(fence)) |
| 5078 | serge | 237 | radeon_fence_unref(&fence); |
| 238 | break; |
||
| 239 | } |
||
| 240 | |||
| 241 | radeon_bo_unpin(vram_obj); |
||
| 242 | out_unres: |
||
| 243 | radeon_bo_unreserve(vram_obj); |
||
| 244 | out_unref: |
||
| 245 | radeon_bo_unref(&vram_obj); |
||
| 246 | out_cleanup: |
||
| 247 | kfree(gtt_obj); |
||
| 248 | if (r) { |
||
| 249 | printk(KERN_WARNING "Error while testing BO move.\n"); |
||
| 250 | } |
||
| 251 | } |
||
| 252 | |||
| 253 | void radeon_test_moves(struct radeon_device *rdev) |
||
| 254 | { |
||
| 255 | if (rdev->asic->copy.dma) |
||
| 256 | radeon_do_test_moves(rdev, RADEON_TEST_COPY_DMA); |
||
| 257 | if (rdev->asic->copy.blit) |
||
| 258 | radeon_do_test_moves(rdev, RADEON_TEST_COPY_BLIT); |
||
| 259 | } |
||
| 260 | |||
| 261 | static int radeon_test_create_and_emit_fence(struct radeon_device *rdev, |
||
| 262 | struct radeon_ring *ring, |
||
| 263 | struct radeon_fence **fence) |
||
| 264 | { |
||
| 265 | uint32_t handle = ring->idx ^ 0xdeafbeef; |
||
| 266 | int r; |
||
| 267 | |||
| 268 | if (ring->idx == R600_RING_TYPE_UVD_INDEX) { |
||
| 269 | #if 0 |
||
| 270 | r = radeon_uvd_get_create_msg(rdev, ring->idx, handle, NULL); |
||
| 271 | if (r) { |
||
| 272 | DRM_ERROR("Failed to get dummy create msg\n"); |
||
| 273 | return r; |
||
| 274 | } |
||
| 275 | |||
| 276 | r = radeon_uvd_get_destroy_msg(rdev, ring->idx, handle, fence); |
||
| 277 | if (r) { |
||
| 278 | DRM_ERROR("Failed to get dummy destroy msg\n"); |
||
| 279 | return r; |
||
| 280 | } |
||
| 281 | #endif |
||
| 282 | } else if (ring->idx == TN_RING_TYPE_VCE1_INDEX || |
||
| 283 | ring->idx == TN_RING_TYPE_VCE2_INDEX) { |
||
| 284 | #if 0 |
||
| 285 | r = radeon_vce_get_create_msg(rdev, ring->idx, handle, NULL); |
||
| 286 | if (r) { |
||
| 287 | DRM_ERROR("Failed to get dummy create msg\n"); |
||
| 288 | return r; |
||
| 289 | } |
||
| 290 | |||
| 291 | r = radeon_vce_get_destroy_msg(rdev, ring->idx, handle, fence); |
||
| 292 | if (r) { |
||
| 293 | DRM_ERROR("Failed to get dummy destroy msg\n"); |
||
| 294 | return r; |
||
| 295 | } |
||
| 296 | #endif |
||
| 297 | } else { |
||
| 298 | r = radeon_ring_lock(rdev, ring, 64); |
||
| 299 | if (r) { |
||
| 300 | DRM_ERROR("Failed to lock ring A %d\n", ring->idx); |
||
| 301 | return r; |
||
| 302 | } |
||
| 303 | radeon_fence_emit(rdev, fence, ring->idx); |
||
| 304 | radeon_ring_unlock_commit(rdev, ring, false); |
||
| 305 | } |
||
| 306 | return 0; |
||
| 307 | } |
||
| 308 | |||
| 309 | void radeon_test_ring_sync(struct radeon_device *rdev, |
||
| 310 | struct radeon_ring *ringA, |
||
| 311 | struct radeon_ring *ringB) |
||
| 312 | { |
||
| 313 | struct radeon_fence *fence1 = NULL, *fence2 = NULL; |
||
| 314 | struct radeon_semaphore *semaphore = NULL; |
||
| 315 | int r; |
||
| 316 | |||
| 317 | r = radeon_semaphore_create(rdev, &semaphore); |
||
| 318 | if (r) { |
||
| 319 | DRM_ERROR("Failed to create semaphore\n"); |
||
| 320 | goto out_cleanup; |
||
| 321 | } |
||
| 322 | |||
| 323 | r = radeon_ring_lock(rdev, ringA, 64); |
||
| 324 | if (r) { |
||
| 325 | DRM_ERROR("Failed to lock ring A %d\n", ringA->idx); |
||
| 326 | goto out_cleanup; |
||
| 327 | } |
||
| 328 | radeon_semaphore_emit_wait(rdev, ringA->idx, semaphore); |
||
| 329 | radeon_ring_unlock_commit(rdev, ringA, false); |
||
| 330 | |||
| 331 | r = radeon_test_create_and_emit_fence(rdev, ringA, &fence1); |
||
| 332 | if (r) |
||
| 333 | goto out_cleanup; |
||
| 334 | |||
| 335 | r = radeon_ring_lock(rdev, ringA, 64); |
||
| 336 | if (r) { |
||
| 337 | DRM_ERROR("Failed to lock ring A %d\n", ringA->idx); |
||
| 338 | goto out_cleanup; |
||
| 339 | } |
||
| 340 | radeon_semaphore_emit_wait(rdev, ringA->idx, semaphore); |
||
| 341 | radeon_ring_unlock_commit(rdev, ringA, false); |
||
| 342 | |||
| 343 | r = radeon_test_create_and_emit_fence(rdev, ringA, &fence2); |
||
| 344 | if (r) |
||
| 345 | goto out_cleanup; |
||
| 346 | |||
| 347 | mdelay(1000); |
||
| 348 | |||
| 349 | if (radeon_fence_signaled(fence1)) { |
||
| 350 | DRM_ERROR("Fence 1 signaled without waiting for semaphore.\n"); |
||
| 351 | goto out_cleanup; |
||
| 352 | } |
||
| 353 | |||
| 354 | r = radeon_ring_lock(rdev, ringB, 64); |
||
| 355 | if (r) { |
||
| 356 | DRM_ERROR("Failed to lock ring B %p\n", ringB); |
||
| 357 | goto out_cleanup; |
||
| 358 | } |
||
| 359 | radeon_semaphore_emit_signal(rdev, ringB->idx, semaphore); |
||
| 360 | radeon_ring_unlock_commit(rdev, ringB, false); |
||
| 361 | |||
| 362 | r = radeon_fence_wait(fence1, false); |
||
| 363 | if (r) { |
||
| 364 | DRM_ERROR("Failed to wait for sync fence 1\n"); |
||
| 365 | goto out_cleanup; |
||
| 366 | } |
||
| 367 | |||
| 368 | mdelay(1000); |
||
| 369 | |||
| 370 | if (radeon_fence_signaled(fence2)) { |
||
| 371 | DRM_ERROR("Fence 2 signaled without waiting for semaphore.\n"); |
||
| 372 | goto out_cleanup; |
||
| 373 | } |
||
| 374 | |||
| 375 | r = radeon_ring_lock(rdev, ringB, 64); |
||
| 376 | if (r) { |
||
| 377 | DRM_ERROR("Failed to lock ring B %p\n", ringB); |
||
| 378 | goto out_cleanup; |
||
| 379 | } |
||
| 380 | radeon_semaphore_emit_signal(rdev, ringB->idx, semaphore); |
||
| 381 | radeon_ring_unlock_commit(rdev, ringB, false); |
||
| 382 | |||
| 383 | r = radeon_fence_wait(fence2, false); |
||
| 384 | if (r) { |
||
| 385 | DRM_ERROR("Failed to wait for sync fence 1\n"); |
||
| 386 | goto out_cleanup; |
||
| 387 | } |
||
| 388 | |||
| 389 | out_cleanup: |
||
| 390 | radeon_semaphore_free(rdev, &semaphore, NULL); |
||
| 391 | |||
| 392 | if (fence1) |
||
| 393 | radeon_fence_unref(&fence1); |
||
| 394 | |||
| 395 | if (fence2) |
||
| 396 | radeon_fence_unref(&fence2); |
||
| 397 | |||
| 398 | if (r) |
||
| 399 | printk(KERN_WARNING "Error while testing ring sync (%d).\n", r); |
||
| 400 | } |
||
| 401 | |||
| 402 | static void radeon_test_ring_sync2(struct radeon_device *rdev, |
||
| 403 | struct radeon_ring *ringA, |
||
| 404 | struct radeon_ring *ringB, |
||
| 405 | struct radeon_ring *ringC) |
||
| 406 | { |
||
| 407 | struct radeon_fence *fenceA = NULL, *fenceB = NULL; |
||
| 408 | struct radeon_semaphore *semaphore = NULL; |
||
| 409 | bool sigA, sigB; |
||
| 410 | int i, r; |
||
| 411 | |||
| 412 | r = radeon_semaphore_create(rdev, &semaphore); |
||
| 413 | if (r) { |
||
| 414 | DRM_ERROR("Failed to create semaphore\n"); |
||
| 415 | goto out_cleanup; |
||
| 416 | } |
||
| 417 | |||
| 418 | r = radeon_ring_lock(rdev, ringA, 64); |
||
| 419 | if (r) { |
||
| 420 | DRM_ERROR("Failed to lock ring A %d\n", ringA->idx); |
||
| 421 | goto out_cleanup; |
||
| 422 | } |
||
| 423 | radeon_semaphore_emit_wait(rdev, ringA->idx, semaphore); |
||
| 424 | radeon_ring_unlock_commit(rdev, ringA, false); |
||
| 425 | |||
| 426 | r = radeon_test_create_and_emit_fence(rdev, ringA, &fenceA); |
||
| 427 | if (r) |
||
| 428 | goto out_cleanup; |
||
| 429 | |||
| 430 | r = radeon_ring_lock(rdev, ringB, 64); |
||
| 431 | if (r) { |
||
| 432 | DRM_ERROR("Failed to lock ring B %d\n", ringB->idx); |
||
| 433 | goto out_cleanup; |
||
| 434 | } |
||
| 435 | radeon_semaphore_emit_wait(rdev, ringB->idx, semaphore); |
||
| 436 | radeon_ring_unlock_commit(rdev, ringB, false); |
||
| 437 | r = radeon_test_create_and_emit_fence(rdev, ringB, &fenceB); |
||
| 438 | if (r) |
||
| 439 | goto out_cleanup; |
||
| 440 | |||
| 441 | mdelay(1000); |
||
| 442 | |||
| 443 | if (radeon_fence_signaled(fenceA)) { |
||
| 444 | DRM_ERROR("Fence A signaled without waiting for semaphore.\n"); |
||
| 445 | goto out_cleanup; |
||
| 446 | } |
||
| 447 | if (radeon_fence_signaled(fenceB)) { |
||
| 448 | DRM_ERROR("Fence B signaled without waiting for semaphore.\n"); |
||
| 449 | goto out_cleanup; |
||
| 450 | } |
||
| 451 | |||
| 452 | r = radeon_ring_lock(rdev, ringC, 64); |
||
| 453 | if (r) { |
||
| 454 | DRM_ERROR("Failed to lock ring B %p\n", ringC); |
||
| 455 | goto out_cleanup; |
||
| 456 | } |
||
| 457 | radeon_semaphore_emit_signal(rdev, ringC->idx, semaphore); |
||
| 458 | radeon_ring_unlock_commit(rdev, ringC, false); |
||
| 459 | |||
| 460 | for (i = 0; i < 30; ++i) { |
||
| 461 | mdelay(100); |
||
| 462 | sigA = radeon_fence_signaled(fenceA); |
||
| 463 | sigB = radeon_fence_signaled(fenceB); |
||
| 464 | if (sigA || sigB) |
||
| 465 | break; |
||
| 466 | } |
||
| 467 | |||
| 468 | if (!sigA && !sigB) { |
||
| 469 | DRM_ERROR("Neither fence A nor B has been signaled\n"); |
||
| 470 | goto out_cleanup; |
||
| 471 | } else if (sigA && sigB) { |
||
| 472 | DRM_ERROR("Both fence A and B has been signaled\n"); |
||
| 473 | goto out_cleanup; |
||
| 474 | } |
||
| 475 | |||
| 476 | DRM_INFO("Fence %c was first signaled\n", sigA ? 'A' : 'B'); |
||
| 477 | |||
| 478 | r = radeon_ring_lock(rdev, ringC, 64); |
||
| 479 | if (r) { |
||
| 480 | DRM_ERROR("Failed to lock ring B %p\n", ringC); |
||
| 481 | goto out_cleanup; |
||
| 482 | } |
||
| 483 | radeon_semaphore_emit_signal(rdev, ringC->idx, semaphore); |
||
| 484 | radeon_ring_unlock_commit(rdev, ringC, false); |
||
| 485 | |||
| 486 | mdelay(1000); |
||
| 487 | |||
| 488 | r = radeon_fence_wait(fenceA, false); |
||
| 489 | if (r) { |
||
| 490 | DRM_ERROR("Failed to wait for sync fence A\n"); |
||
| 491 | goto out_cleanup; |
||
| 492 | } |
||
| 493 | r = radeon_fence_wait(fenceB, false); |
||
| 494 | if (r) { |
||
| 495 | DRM_ERROR("Failed to wait for sync fence B\n"); |
||
| 496 | goto out_cleanup; |
||
| 497 | } |
||
| 498 | |||
| 499 | out_cleanup: |
||
| 500 | radeon_semaphore_free(rdev, &semaphore, NULL); |
||
| 501 | |||
| 502 | if (fenceA) |
||
| 503 | radeon_fence_unref(&fenceA); |
||
| 504 | |||
| 505 | if (fenceB) |
||
| 506 | radeon_fence_unref(&fenceB); |
||
| 507 | |||
| 508 | if (r) |
||
| 509 | printk(KERN_WARNING "Error while testing ring sync (%d).\n", r); |
||
| 510 | } |
||
| 511 | |||
| 512 | static bool radeon_test_sync_possible(struct radeon_ring *ringA, |
||
| 513 | struct radeon_ring *ringB) |
||
| 514 | { |
||
| 515 | if (ringA->idx == TN_RING_TYPE_VCE2_INDEX && |
||
| 516 | ringB->idx == TN_RING_TYPE_VCE1_INDEX) |
||
| 517 | return false; |
||
| 518 | |||
| 519 | return true; |
||
| 520 | } |
||
| 521 | |||
| 522 | void radeon_test_syncing(struct radeon_device *rdev) |
||
| 523 | { |
||
| 524 | int i, j, k; |
||
| 525 | |||
| 526 | for (i = 1; i < RADEON_NUM_RINGS; ++i) { |
||
| 527 | struct radeon_ring *ringA = &rdev->ring[i]; |
||
| 528 | if (!ringA->ready) |
||
| 529 | continue; |
||
| 530 | |||
| 531 | for (j = 0; j < i; ++j) { |
||
| 532 | struct radeon_ring *ringB = &rdev->ring[j]; |
||
| 533 | if (!ringB->ready) |
||
| 534 | continue; |
||
| 535 | |||
| 536 | if (!radeon_test_sync_possible(ringA, ringB)) |
||
| 537 | continue; |
||
| 538 | |||
| 539 | DRM_INFO("Testing syncing between rings %d and %d...\n", i, j); |
||
| 540 | radeon_test_ring_sync(rdev, ringA, ringB); |
||
| 541 | |||
| 542 | DRM_INFO("Testing syncing between rings %d and %d...\n", j, i); |
||
| 543 | radeon_test_ring_sync(rdev, ringB, ringA); |
||
| 544 | |||
| 545 | for (k = 0; k < j; ++k) { |
||
| 546 | struct radeon_ring *ringC = &rdev->ring[k]; |
||
| 547 | if (!ringC->ready) |
||
| 548 | continue; |
||
| 549 | |||
| 550 | if (!radeon_test_sync_possible(ringA, ringC)) |
||
| 551 | continue; |
||
| 552 | |||
| 553 | if (!radeon_test_sync_possible(ringB, ringC)) |
||
| 554 | continue; |
||
| 555 | |||
| 556 | DRM_INFO("Testing syncing between rings %d, %d and %d...\n", i, j, k); |
||
| 557 | radeon_test_ring_sync2(rdev, ringA, ringB, ringC); |
||
| 558 | |||
| 559 | DRM_INFO("Testing syncing between rings %d, %d and %d...\n", i, k, j); |
||
| 560 | radeon_test_ring_sync2(rdev, ringA, ringC, ringB); |
||
| 561 | |||
| 562 | DRM_INFO("Testing syncing between rings %d, %d and %d...\n", j, i, k); |
||
| 563 | radeon_test_ring_sync2(rdev, ringB, ringA, ringC); |
||
| 564 | |||
| 565 | DRM_INFO("Testing syncing between rings %d, %d and %d...\n", j, k, i); |
||
| 566 | radeon_test_ring_sync2(rdev, ringB, ringC, ringA); |
||
| 567 | |||
| 568 | DRM_INFO("Testing syncing between rings %d, %d and %d...\n", k, i, j); |
||
| 569 | radeon_test_ring_sync2(rdev, ringC, ringA, ringB); |
||
| 570 | |||
| 571 | DRM_INFO("Testing syncing between rings %d, %d and %d...\n", k, j, i); |
||
| 572 | radeon_test_ring_sync2(rdev, ringC, ringB, ringA); |
||
| 573 | } |
||
| 574 | } |
||
| 575 | } |
||
| 576 | } |