2 * Copyright 2009 VMware, Inc.
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:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
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.
22 * Authors: Michel Dänzer
25 #include <drm/radeon_drm.h>
26 #include "radeon_reg.h"
29 #define RADEON_TEST_COPY_BLIT 1
30 #define RADEON_TEST_COPY_DMA 0
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)
36 struct radeon_bo *vram_obj = NULL;
37 struct radeon_bo **gtt_obj = NULL;
38 uint64_t gtt_addr, vram_addr;
43 case RADEON_TEST_COPY_DMA:
44 ring = radeon_copy_dma_ring_index(rdev);
46 case RADEON_TEST_COPY_BLIT:
47 ring = radeon_copy_blit_ring_index(rdev);
50 DRM_ERROR("Unknown copy method\n");
57 * (Total GTT - IB pool - writeback page - ring buffers) / test size
59 n = rdev->mc.gtt_size - rdev->gart_pin_size;
62 gtt_obj = kzalloc(n * sizeof(*gtt_obj), GFP_KERNEL);
64 DRM_ERROR("Failed to allocate %d pointers\n", n);
69 r = radeon_bo_create(rdev, size, PAGE_SIZE, true, RADEON_GEM_DOMAIN_VRAM,
70 0, NULL, NULL, &vram_obj);
72 DRM_ERROR("Failed to create VRAM object\n");
75 r = radeon_bo_reserve(vram_obj, false);
78 r = radeon_bo_pin(vram_obj, RADEON_GEM_DOMAIN_VRAM, &vram_addr);
80 DRM_ERROR("Failed to pin VRAM object\n");
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;
89 r = radeon_bo_create(rdev, size, PAGE_SIZE, true,
90 RADEON_GEM_DOMAIN_GTT, 0, NULL, NULL,
93 DRM_ERROR("Failed to create GTT object %d\n", i);
97 r = radeon_bo_reserve(gtt_obj[i], false);
99 goto out_lclean_unref;
100 r = radeon_bo_pin(gtt_obj[i], RADEON_GEM_DOMAIN_GTT, >t_addr);
102 DRM_ERROR("Failed to pin GTT object %d\n", i);
103 goto out_lclean_unres;
106 r = radeon_bo_kmap(gtt_obj[i], >t_map);
108 DRM_ERROR("Failed to map GTT object %d\n", i);
109 goto out_lclean_unpin;
112 for (gtt_start = gtt_map, gtt_end = gtt_map + size;
115 *gtt_start = gtt_start;
117 radeon_bo_kunmap(gtt_obj[i]);
119 if (ring == R600_RING_TYPE_DMA_INDEX)
120 fence = radeon_copy_dma(rdev, gtt_addr, vram_addr,
121 size / RADEON_GPU_PAGE_SIZE,
124 fence = radeon_copy_blit(rdev, gtt_addr, vram_addr,
125 size / RADEON_GPU_PAGE_SIZE,
128 DRM_ERROR("Failed GTT->VRAM copy %d\n", i);
130 goto out_lclean_unpin;
133 r = radeon_fence_wait(fence, false);
135 DRM_ERROR("Failed to wait for GTT->VRAM fence %d\n", i);
136 goto out_lclean_unpin;
139 radeon_fence_unref(&fence);
141 r = radeon_bo_kmap(vram_obj, &vram_map);
143 DRM_ERROR("Failed to map VRAM object after copy %d\n", i);
144 goto out_lclean_unpin;
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,
157 (gtt_addr - rdev->mc.gtt_start +
158 (void*)gtt_start - gtt_map),
160 (vram_addr - rdev->mc.vram_start +
161 (void*)gtt_start - gtt_map));
162 radeon_bo_kunmap(vram_obj);
163 goto out_lclean_unpin;
165 *vram_start = vram_start;
168 radeon_bo_kunmap(vram_obj);
170 if (ring == R600_RING_TYPE_DMA_INDEX)
171 fence = radeon_copy_dma(rdev, vram_addr, gtt_addr,
172 size / RADEON_GPU_PAGE_SIZE,
175 fence = radeon_copy_blit(rdev, vram_addr, gtt_addr,
176 size / RADEON_GPU_PAGE_SIZE,
179 DRM_ERROR("Failed VRAM->GTT copy %d\n", i);
181 goto out_lclean_unpin;
184 r = radeon_fence_wait(fence, false);
186 DRM_ERROR("Failed to wait for VRAM->GTT fence %d\n", i);
187 goto out_lclean_unpin;
190 radeon_fence_unref(&fence);
192 r = radeon_bo_kmap(gtt_obj[i], >t_map);
194 DRM_ERROR("Failed to map GTT object after copy %d\n", i);
195 goto out_lclean_unpin;
198 for (gtt_start = gtt_map, gtt_end = gtt_map + size,
199 vram_start = vram_map, vram_end = vram_map + size;
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,
208 (vram_addr - rdev->mc.vram_start +
209 (void*)vram_start - vram_map),
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;
218 radeon_bo_kunmap(gtt_obj[i]);
220 DRM_INFO("Tested GTT->VRAM and VRAM->GTT copy for GTT offset 0x%llx\n",
221 gtt_addr - rdev->mc.gtt_start);
225 radeon_bo_unpin(gtt_obj[i]);
227 radeon_bo_unreserve(gtt_obj[i]);
229 radeon_bo_unref(>t_obj[i]);
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]);
236 if (fence && !IS_ERR(fence))
237 radeon_fence_unref(&fence);
241 radeon_bo_unpin(vram_obj);
243 radeon_bo_unreserve(vram_obj);
245 radeon_bo_unref(&vram_obj);
249 printk(KERN_WARNING "Error while testing BO move.\n");
253 void radeon_test_moves(struct radeon_device *rdev)
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);
261 static int radeon_test_create_and_emit_fence(struct radeon_device *rdev,
262 struct radeon_ring *ring,
263 struct radeon_fence **fence)
265 uint32_t handle = ring->idx ^ 0xdeafbeef;
268 if (ring->idx == R600_RING_TYPE_UVD_INDEX) {
269 r = radeon_uvd_get_create_msg(rdev, ring->idx, handle, NULL);
271 DRM_ERROR("Failed to get dummy create msg\n");
275 r = radeon_uvd_get_destroy_msg(rdev, ring->idx, handle, fence);
277 DRM_ERROR("Failed to get dummy destroy msg\n");
281 } else if (ring->idx == TN_RING_TYPE_VCE1_INDEX ||
282 ring->idx == TN_RING_TYPE_VCE2_INDEX) {
283 r = radeon_vce_get_create_msg(rdev, ring->idx, handle, NULL);
285 DRM_ERROR("Failed to get dummy create msg\n");
289 r = radeon_vce_get_destroy_msg(rdev, ring->idx, handle, fence);
291 DRM_ERROR("Failed to get dummy destroy msg\n");
296 r = radeon_ring_lock(rdev, ring, 64);
298 DRM_ERROR("Failed to lock ring A %d\n", ring->idx);
301 radeon_fence_emit(rdev, fence, ring->idx);
302 radeon_ring_unlock_commit(rdev, ring, false);
307 void radeon_test_ring_sync(struct radeon_device *rdev,
308 struct radeon_ring *ringA,
309 struct radeon_ring *ringB)
311 struct radeon_fence *fence1 = NULL, *fence2 = NULL;
312 struct radeon_semaphore *semaphore = NULL;
315 r = radeon_semaphore_create(rdev, &semaphore);
317 DRM_ERROR("Failed to create semaphore\n");
321 r = radeon_ring_lock(rdev, ringA, 64);
323 DRM_ERROR("Failed to lock ring A %d\n", ringA->idx);
326 radeon_semaphore_emit_wait(rdev, ringA->idx, semaphore);
327 radeon_ring_unlock_commit(rdev, ringA, false);
329 r = radeon_test_create_and_emit_fence(rdev, ringA, &fence1);
333 r = radeon_ring_lock(rdev, ringA, 64);
335 DRM_ERROR("Failed to lock ring A %d\n", ringA->idx);
338 radeon_semaphore_emit_wait(rdev, ringA->idx, semaphore);
339 radeon_ring_unlock_commit(rdev, ringA, false);
341 r = radeon_test_create_and_emit_fence(rdev, ringA, &fence2);
347 if (radeon_fence_signaled(fence1)) {
348 DRM_ERROR("Fence 1 signaled without waiting for semaphore.\n");
352 r = radeon_ring_lock(rdev, ringB, 64);
354 DRM_ERROR("Failed to lock ring B %p\n", ringB);
357 radeon_semaphore_emit_signal(rdev, ringB->idx, semaphore);
358 radeon_ring_unlock_commit(rdev, ringB, false);
360 r = radeon_fence_wait(fence1, false);
362 DRM_ERROR("Failed to wait for sync fence 1\n");
368 if (radeon_fence_signaled(fence2)) {
369 DRM_ERROR("Fence 2 signaled without waiting for semaphore.\n");
373 r = radeon_ring_lock(rdev, ringB, 64);
375 DRM_ERROR("Failed to lock ring B %p\n", ringB);
378 radeon_semaphore_emit_signal(rdev, ringB->idx, semaphore);
379 radeon_ring_unlock_commit(rdev, ringB, false);
381 r = radeon_fence_wait(fence2, false);
383 DRM_ERROR("Failed to wait for sync fence 1\n");
388 radeon_semaphore_free(rdev, &semaphore, NULL);
391 radeon_fence_unref(&fence1);
394 radeon_fence_unref(&fence2);
397 printk(KERN_WARNING "Error while testing ring sync (%d).\n", r);
400 static void radeon_test_ring_sync2(struct radeon_device *rdev,
401 struct radeon_ring *ringA,
402 struct radeon_ring *ringB,
403 struct radeon_ring *ringC)
405 struct radeon_fence *fenceA = NULL, *fenceB = NULL;
406 struct radeon_semaphore *semaphore = NULL;
410 r = radeon_semaphore_create(rdev, &semaphore);
412 DRM_ERROR("Failed to create semaphore\n");
416 r = radeon_ring_lock(rdev, ringA, 64);
418 DRM_ERROR("Failed to lock ring A %d\n", ringA->idx);
421 radeon_semaphore_emit_wait(rdev, ringA->idx, semaphore);
422 radeon_ring_unlock_commit(rdev, ringA, false);
424 r = radeon_test_create_and_emit_fence(rdev, ringA, &fenceA);
428 r = radeon_ring_lock(rdev, ringB, 64);
430 DRM_ERROR("Failed to lock ring B %d\n", ringB->idx);
433 radeon_semaphore_emit_wait(rdev, ringB->idx, semaphore);
434 radeon_ring_unlock_commit(rdev, ringB, false);
435 r = radeon_test_create_and_emit_fence(rdev, ringB, &fenceB);
441 if (radeon_fence_signaled(fenceA)) {
442 DRM_ERROR("Fence A signaled without waiting for semaphore.\n");
445 if (radeon_fence_signaled(fenceB)) {
446 DRM_ERROR("Fence B signaled without waiting for semaphore.\n");
450 r = radeon_ring_lock(rdev, ringC, 64);
452 DRM_ERROR("Failed to lock ring B %p\n", ringC);
455 radeon_semaphore_emit_signal(rdev, ringC->idx, semaphore);
456 radeon_ring_unlock_commit(rdev, ringC, false);
458 for (i = 0; i < 30; ++i) {
460 sigA = radeon_fence_signaled(fenceA);
461 sigB = radeon_fence_signaled(fenceB);
466 if (!sigA && !sigB) {
467 DRM_ERROR("Neither fence A nor B has been signaled\n");
469 } else if (sigA && sigB) {
470 DRM_ERROR("Both fence A and B has been signaled\n");
474 DRM_INFO("Fence %c was first signaled\n", sigA ? 'A' : 'B');
476 r = radeon_ring_lock(rdev, ringC, 64);
478 DRM_ERROR("Failed to lock ring B %p\n", ringC);
481 radeon_semaphore_emit_signal(rdev, ringC->idx, semaphore);
482 radeon_ring_unlock_commit(rdev, ringC, false);
486 r = radeon_fence_wait(fenceA, false);
488 DRM_ERROR("Failed to wait for sync fence A\n");
491 r = radeon_fence_wait(fenceB, false);
493 DRM_ERROR("Failed to wait for sync fence B\n");
498 radeon_semaphore_free(rdev, &semaphore, NULL);
501 radeon_fence_unref(&fenceA);
504 radeon_fence_unref(&fenceB);
507 printk(KERN_WARNING "Error while testing ring sync (%d).\n", r);
510 static bool radeon_test_sync_possible(struct radeon_ring *ringA,
511 struct radeon_ring *ringB)
513 if (ringA->idx == TN_RING_TYPE_VCE2_INDEX &&
514 ringB->idx == TN_RING_TYPE_VCE1_INDEX)
520 void radeon_test_syncing(struct radeon_device *rdev)
524 for (i = 1; i < RADEON_NUM_RINGS; ++i) {
525 struct radeon_ring *ringA = &rdev->ring[i];
529 for (j = 0; j < i; ++j) {
530 struct radeon_ring *ringB = &rdev->ring[j];
534 if (!radeon_test_sync_possible(ringA, ringB))
537 DRM_INFO("Testing syncing between rings %d and %d...\n", i, j);
538 radeon_test_ring_sync(rdev, ringA, ringB);
540 DRM_INFO("Testing syncing between rings %d and %d...\n", j, i);
541 radeon_test_ring_sync(rdev, ringB, ringA);
543 for (k = 0; k < j; ++k) {
544 struct radeon_ring *ringC = &rdev->ring[k];
548 if (!radeon_test_sync_possible(ringA, ringC))
551 if (!radeon_test_sync_possible(ringB, ringC))
554 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", i, j, k);
555 radeon_test_ring_sync2(rdev, ringA, ringB, ringC);
557 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", i, k, j);
558 radeon_test_ring_sync2(rdev, ringA, ringC, ringB);
560 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", j, i, k);
561 radeon_test_ring_sync2(rdev, ringB, ringA, ringC);
563 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", j, k, i);
564 radeon_test_ring_sync2(rdev, ringB, ringC, ringA);
566 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", k, i, j);
567 radeon_test_ring_sync2(rdev, ringC, ringA, ringB);
569 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", k, j, i);
570 radeon_test_ring_sync2(rdev, ringC, ringB, ringA);