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 25#include <sys/cdefs.h> 26__FBSDID("$FreeBSD$"); 27 28#include <dev/drm2/drmP.h> 29#include <dev/drm2/radeon/radeon_drm.h> 30#include "radeon_reg.h" 31#include "radeon.h" 32 33#define RADEON_TEST_COPY_BLIT 1 34#define RADEON_TEST_COPY_DMA 0 35 36 37/* Test BO GTT->VRAM and VRAM->GTT GPU copies across the whole GTT aperture */ 38static void radeon_do_test_moves(struct radeon_device *rdev, int flag) 39{ 40 struct radeon_bo *vram_obj = NULL; 41 struct radeon_bo **gtt_obj = NULL; 42 struct radeon_fence *fence = NULL; 43 uint64_t gtt_addr, vram_addr; 44 unsigned i, n, size; 45 int r, ring; 46 47 switch (flag) { 48 case RADEON_TEST_COPY_DMA: 49 ring = radeon_copy_dma_ring_index(rdev); 50 break; 51 case RADEON_TEST_COPY_BLIT: 52 ring = radeon_copy_blit_ring_index(rdev); 53 break; 54 default: 55 DRM_ERROR("Unknown copy method\n"); 56 return; 57 } 58 59 size = 1024 * 1024; 60 61 /* Number of tests = 62 * (Total GTT - IB pool - writeback page - ring buffers) / test size 63 */ 64 n = rdev->mc.gtt_size - RADEON_IB_POOL_SIZE*64*1024; 65 for (i = 0; i < RADEON_NUM_RINGS; ++i) 66 n -= rdev->ring[i].ring_size; 67 if (rdev->wb.wb_obj) 68 n -= RADEON_GPU_PAGE_SIZE; 69 if (rdev->ih.ring_obj) 70 n -= rdev->ih.ring_size; 71 n /= size; 72 73 gtt_obj = malloc(n * sizeof(*gtt_obj), DRM_MEM_DRIVER, M_NOWAIT | M_ZERO); 74 if (!gtt_obj) { 75 DRM_ERROR("Failed to allocate %d pointers\n", n); 76 r = 1; 77 goto out_cleanup; 78 } 79 80 r = radeon_bo_create(rdev, size, PAGE_SIZE, true, RADEON_GEM_DOMAIN_VRAM, 81 NULL, &vram_obj); 82 if (r) { 83 DRM_ERROR("Failed to create VRAM object\n"); 84 goto out_cleanup; 85 } 86 r = radeon_bo_reserve(vram_obj, false); 87 if (unlikely(r != 0)) 88 goto out_cleanup; 89 r = radeon_bo_pin(vram_obj, RADEON_GEM_DOMAIN_VRAM, &vram_addr); 90 if (r) { 91 DRM_ERROR("Failed to pin VRAM object\n"); 92 goto out_cleanup; 93 } 94 for (i = 0; i < n; i++) { 95 void *gtt_map, *vram_map; 96 void **gtt_start, **gtt_end; 97 void **vram_start, **vram_end; 98 99 r = radeon_bo_create(rdev, size, PAGE_SIZE, true, 100 RADEON_GEM_DOMAIN_GTT, NULL, gtt_obj + i); 101 if (r) { 102 DRM_ERROR("Failed to create GTT object %d\n", i); 103 goto out_cleanup; 104 } 105 106 r = radeon_bo_reserve(gtt_obj[i], false); 107 if (unlikely(r != 0)) 108 goto out_cleanup; 109 r = radeon_bo_pin(gtt_obj[i], RADEON_GEM_DOMAIN_GTT, >t_addr); 110 if (r) { 111 DRM_ERROR("Failed to pin GTT object %d\n", i); 112 goto out_cleanup; 113 } 114 115 r = radeon_bo_kmap(gtt_obj[i], >t_map); 116 if (r) { 117 DRM_ERROR("Failed to map GTT object %d\n", i); 118 goto out_cleanup; 119 } 120 121 for (gtt_start = gtt_map, gtt_end = (void *)((uintptr_t)gtt_map + size); 122 gtt_start < gtt_end; 123 gtt_start++) 124 *gtt_start = gtt_start; 125 126 radeon_bo_kunmap(gtt_obj[i]); 127 128 if (ring == R600_RING_TYPE_DMA_INDEX) 129 r = radeon_copy_dma(rdev, gtt_addr, vram_addr, size / RADEON_GPU_PAGE_SIZE, &fence); 130 else 131 r = radeon_copy_blit(rdev, gtt_addr, vram_addr, size / RADEON_GPU_PAGE_SIZE, &fence); 132 if (r) { 133 DRM_ERROR("Failed GTT->VRAM copy %d\n", i); 134 goto out_cleanup; 135 } 136 137 r = radeon_fence_wait(fence, false); 138 if (r) { 139 DRM_ERROR("Failed to wait for GTT->VRAM fence %d\n", i); 140 goto out_cleanup; 141 } 142 143 radeon_fence_unref(&fence); 144 145 r = radeon_bo_kmap(vram_obj, &vram_map); 146 if (r) { 147 DRM_ERROR("Failed to map VRAM object after copy %d\n", i); 148 goto out_cleanup; 149 } 150 151 for (gtt_start = gtt_map, gtt_end = (void *)((uintptr_t)gtt_map + size), 152 vram_start = vram_map, vram_end = (void *)((uintptr_t)vram_map + size); 153 vram_start < vram_end; 154 gtt_start++, vram_start++) { 155 if (*vram_start != gtt_start) { 156 DRM_ERROR("Incorrect GTT->VRAM copy %d: Got 0x%p, " 157 "expected 0x%p (GTT/VRAM offset " 158 "0x%16llx/0x%16llx)\n", 159 i, *vram_start, gtt_start, 160 (unsigned long long) 161 ((uintptr_t)gtt_addr - (uintptr_t)rdev->mc.gtt_start + 162 (uintptr_t)gtt_start - (uintptr_t)gtt_map), 163 (unsigned long long) 164 ((uintptr_t)vram_addr - (uintptr_t)rdev->mc.vram_start + 165 (uintptr_t)gtt_start - (uintptr_t)gtt_map)); 166 radeon_bo_kunmap(vram_obj); 167 goto out_cleanup; 168 } 169 *vram_start = vram_start; 170 } 171 172 radeon_bo_kunmap(vram_obj); 173 174 if (ring == R600_RING_TYPE_DMA_INDEX) 175 r = radeon_copy_dma(rdev, vram_addr, gtt_addr, size / RADEON_GPU_PAGE_SIZE, &fence); 176 else 177 r = radeon_copy_blit(rdev, vram_addr, gtt_addr, size / RADEON_GPU_PAGE_SIZE, &fence); 178 if (r) { 179 DRM_ERROR("Failed VRAM->GTT copy %d\n", i); 180 goto out_cleanup; 181 } 182 183 r = radeon_fence_wait(fence, false); 184 if (r) { 185 DRM_ERROR("Failed to wait for VRAM->GTT fence %d\n", i); 186 goto out_cleanup; 187 } 188 189 radeon_fence_unref(&fence); 190 191 r = radeon_bo_kmap(gtt_obj[i], >t_map); 192 if (r) { 193 DRM_ERROR("Failed to map GTT object after copy %d\n", i); 194 goto out_cleanup; 195 } 196 197 for (gtt_start = gtt_map, gtt_end = (void *)((uintptr_t)gtt_map + size), 198 vram_start = vram_map, vram_end = (void *)((uintptr_t)vram_map + size); 199 gtt_start < gtt_end; 200 gtt_start++, vram_start++) { 201 if (*gtt_start != vram_start) { 202 DRM_ERROR("Incorrect VRAM->GTT copy %d: Got 0x%p, " 203 "expected 0x%p (VRAM/GTT offset " 204 "0x%16llx/0x%16llx)\n", 205 i, *gtt_start, vram_start, 206 (unsigned long long) 207 ((uintptr_t)vram_addr - (uintptr_t)rdev->mc.vram_start + 208 (uintptr_t)vram_start - (uintptr_t)vram_map), 209 (unsigned long long) 210 ((uintptr_t)gtt_addr - (uintptr_t)rdev->mc.gtt_start + 211 (uintptr_t)vram_start - (uintptr_t)vram_map)); 212 radeon_bo_kunmap(gtt_obj[i]); 213 goto out_cleanup; 214 } 215 } 216 217 radeon_bo_kunmap(gtt_obj[i]); 218 219 DRM_INFO("Tested GTT->VRAM and VRAM->GTT copy for GTT offset 0x%jx\n", 220 (uintmax_t)gtt_addr - rdev->mc.gtt_start); 221 } 222 223out_cleanup: 224 if (vram_obj) { 225 if (radeon_bo_is_reserved(vram_obj)) { 226 radeon_bo_unpin(vram_obj); 227 radeon_bo_unreserve(vram_obj); 228 } 229 radeon_bo_unref(&vram_obj); 230 } 231 if (gtt_obj) { 232 for (i = 0; i < n; i++) { 233 if (gtt_obj[i]) { 234 if (radeon_bo_is_reserved(gtt_obj[i])) { 235 radeon_bo_unpin(gtt_obj[i]); 236 radeon_bo_unreserve(gtt_obj[i]); 237 } 238 radeon_bo_unref(>t_obj[i]); 239 } 240 } 241 free(gtt_obj, DRM_MEM_DRIVER); 242 } 243 if (fence) { 244 radeon_fence_unref(&fence); 245 } 246 if (r) { 247 DRM_ERROR("Error while testing BO move.\n"); 248 } 249} 250 251void radeon_test_moves(struct radeon_device *rdev) 252{ 253 if (rdev->asic->copy.dma) 254 radeon_do_test_moves(rdev, RADEON_TEST_COPY_DMA); 255 if (rdev->asic->copy.blit) 256 radeon_do_test_moves(rdev, RADEON_TEST_COPY_BLIT); 257} 258 259void radeon_test_ring_sync(struct radeon_device *rdev, 260 struct radeon_ring *ringA, 261 struct radeon_ring *ringB) 262{ 263 struct radeon_fence *fence1 = NULL, *fence2 = NULL; 264 struct radeon_semaphore *semaphore = NULL; 265 int r; 266 267 r = radeon_semaphore_create(rdev, &semaphore); 268 if (r) { 269 DRM_ERROR("Failed to create semaphore\n"); 270 goto out_cleanup; 271 } 272 273 r = radeon_ring_lock(rdev, ringA, 64); 274 if (r) { 275 DRM_ERROR("Failed to lock ring A %d\n", ringA->idx); 276 goto out_cleanup; 277 } 278 radeon_semaphore_emit_wait(rdev, ringA->idx, semaphore); 279 r = radeon_fence_emit(rdev, &fence1, ringA->idx); 280 if (r) { 281 DRM_ERROR("Failed to emit fence 1\n"); 282 radeon_ring_unlock_undo(rdev, ringA); 283 goto out_cleanup; 284 } 285 radeon_semaphore_emit_wait(rdev, ringA->idx, semaphore); 286 r = radeon_fence_emit(rdev, &fence2, ringA->idx); 287 if (r) { 288 DRM_ERROR("Failed to emit fence 2\n"); 289 radeon_ring_unlock_undo(rdev, ringA); 290 goto out_cleanup; 291 } 292 radeon_ring_unlock_commit(rdev, ringA); 293 294 mdelay(1000); 295 296 if (radeon_fence_signaled(fence1)) { 297 DRM_ERROR("Fence 1 signaled without waiting for semaphore.\n"); 298 goto out_cleanup; 299 } 300 301 r = radeon_ring_lock(rdev, ringB, 64); 302 if (r) { 303 DRM_ERROR("Failed to lock ring B %p\n", ringB); 304 goto out_cleanup; 305 } 306 radeon_semaphore_emit_signal(rdev, ringB->idx, semaphore); 307 radeon_ring_unlock_commit(rdev, ringB); 308 309 r = radeon_fence_wait(fence1, false); 310 if (r) { 311 DRM_ERROR("Failed to wait for sync fence 1\n"); 312 goto out_cleanup; 313 } 314 315 mdelay(1000); 316 317 if (radeon_fence_signaled(fence2)) { 318 DRM_ERROR("Fence 2 signaled without waiting for semaphore.\n"); 319 goto out_cleanup; 320 } 321 322 r = radeon_ring_lock(rdev, ringB, 64); 323 if (r) { 324 DRM_ERROR("Failed to lock ring B %p\n", ringB); 325 goto out_cleanup; 326 } 327 radeon_semaphore_emit_signal(rdev, ringB->idx, semaphore); 328 radeon_ring_unlock_commit(rdev, ringB); 329 330 r = radeon_fence_wait(fence2, false); 331 if (r) { 332 DRM_ERROR("Failed to wait for sync fence 1\n"); 333 goto out_cleanup; 334 } 335 336out_cleanup: 337 radeon_semaphore_free(rdev, &semaphore, NULL); 338 339 if (fence1) 340 radeon_fence_unref(&fence1); 341 342 if (fence2) 343 radeon_fence_unref(&fence2); 344 345 if (r) 346 DRM_ERROR("Error while testing ring sync (%d).\n", r); 347} 348 349static void radeon_test_ring_sync2(struct radeon_device *rdev, 350 struct radeon_ring *ringA, 351 struct radeon_ring *ringB, 352 struct radeon_ring *ringC) 353{ 354 struct radeon_fence *fenceA = NULL, *fenceB = NULL; 355 struct radeon_semaphore *semaphore = NULL; 356 bool sigA, sigB; 357 int i, r; 358 359 r = radeon_semaphore_create(rdev, &semaphore); 360 if (r) { 361 DRM_ERROR("Failed to create semaphore\n"); 362 goto out_cleanup; 363 } 364 365 r = radeon_ring_lock(rdev, ringA, 64); 366 if (r) { 367 DRM_ERROR("Failed to lock ring A %d\n", ringA->idx); 368 goto out_cleanup; 369 } 370 radeon_semaphore_emit_wait(rdev, ringA->idx, semaphore); 371 r = radeon_fence_emit(rdev, &fenceA, ringA->idx); 372 if (r) { 373 DRM_ERROR("Failed to emit sync fence 1\n"); 374 radeon_ring_unlock_undo(rdev, ringA); 375 goto out_cleanup; 376 } 377 radeon_ring_unlock_commit(rdev, ringA); 378 379 r = radeon_ring_lock(rdev, ringB, 64); 380 if (r) { 381 DRM_ERROR("Failed to lock ring B %d\n", ringB->idx); 382 goto out_cleanup; 383 } 384 radeon_semaphore_emit_wait(rdev, ringB->idx, semaphore); 385 r = radeon_fence_emit(rdev, &fenceB, ringB->idx); 386 if (r) { 387 DRM_ERROR("Failed to create sync fence 2\n"); 388 radeon_ring_unlock_undo(rdev, ringB); 389 goto out_cleanup; 390 } 391 radeon_ring_unlock_commit(rdev, ringB); 392 393 mdelay(1000); 394 395 if (radeon_fence_signaled(fenceA)) { 396 DRM_ERROR("Fence A signaled without waiting for semaphore.\n"); 397 goto out_cleanup; 398 } 399 if (radeon_fence_signaled(fenceB)) { 400 DRM_ERROR("Fence A signaled without waiting for semaphore.\n"); 401 goto out_cleanup; 402 } 403 404 r = radeon_ring_lock(rdev, ringC, 64); 405 if (r) { 406 DRM_ERROR("Failed to lock ring B %p\n", ringC); 407 goto out_cleanup; 408 } 409 radeon_semaphore_emit_signal(rdev, ringC->idx, semaphore); 410 radeon_ring_unlock_commit(rdev, ringC); 411 412 for (i = 0; i < 30; ++i) { 413 mdelay(100); 414 sigA = radeon_fence_signaled(fenceA); 415 sigB = radeon_fence_signaled(fenceB); 416 if (sigA || sigB) 417 break; 418 } 419 420 if (!sigA && !sigB) { 421 DRM_ERROR("Neither fence A nor B has been signaled\n"); 422 goto out_cleanup; 423 } else if (sigA && sigB) { 424 DRM_ERROR("Both fence A and B has been signaled\n"); 425 goto out_cleanup; 426 } 427 428 DRM_INFO("Fence %c was first signaled\n", sigA ? 'A' : 'B'); 429 430 r = radeon_ring_lock(rdev, ringC, 64); 431 if (r) { 432 DRM_ERROR("Failed to lock ring B %p\n", ringC); 433 goto out_cleanup; 434 } 435 radeon_semaphore_emit_signal(rdev, ringC->idx, semaphore); 436 radeon_ring_unlock_commit(rdev, ringC); 437 438 mdelay(1000); 439 440 r = radeon_fence_wait(fenceA, false); 441 if (r) { 442 DRM_ERROR("Failed to wait for sync fence A\n"); 443 goto out_cleanup; 444 } 445 r = radeon_fence_wait(fenceB, false); 446 if (r) { 447 DRM_ERROR("Failed to wait for sync fence B\n"); 448 goto out_cleanup; 449 } 450 451out_cleanup: 452 radeon_semaphore_free(rdev, &semaphore, NULL); 453 454 if (fenceA) 455 radeon_fence_unref(&fenceA); 456 457 if (fenceB) 458 radeon_fence_unref(&fenceB); 459 460 if (r) 461 DRM_ERROR("Error while testing ring sync (%d).\n", r); 462} 463 464void radeon_test_syncing(struct radeon_device *rdev) 465{ 466 int i, j, k; 467 468 for (i = 1; i < RADEON_NUM_RINGS; ++i) { 469 struct radeon_ring *ringA = &rdev->ring[i]; 470 if (!ringA->ready) 471 continue; 472 473 for (j = 0; j < i; ++j) { 474 struct radeon_ring *ringB = &rdev->ring[j]; 475 if (!ringB->ready) 476 continue; 477 478 DRM_INFO("Testing syncing between rings %d and %d...\n", i, j); 479 radeon_test_ring_sync(rdev, ringA, ringB); 480 481 DRM_INFO("Testing syncing between rings %d and %d...\n", j, i); 482 radeon_test_ring_sync(rdev, ringB, ringA); 483 484 for (k = 0; k < j; ++k) { 485 struct radeon_ring *ringC = &rdev->ring[k]; 486 if (!ringC->ready) 487 continue; 488 489 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", i, j, k); 490 radeon_test_ring_sync2(rdev, ringA, ringB, ringC); 491 492 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", i, k, j); 493 radeon_test_ring_sync2(rdev, ringA, ringC, ringB); 494 495 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", j, i, k); 496 radeon_test_ring_sync2(rdev, ringB, ringA, ringC); 497 498 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", j, k, i); 499 radeon_test_ring_sync2(rdev, ringB, ringC, ringA); 500 501 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", k, i, j); 502 radeon_test_ring_sync2(rdev, ringC, ringA, ringB); 503 504 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", k, j, i); 505 radeon_test_ring_sync2(rdev, ringC, ringB, ringA); 506 } 507 } 508 } 509} 510