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 - RADEON_IB_POOL_SIZE*64*1024;
60 for (i = 0; i < RADEON_NUM_RINGS; ++i)
61 n -= rdev->ring[i].ring_size;
63 n -= RADEON_GPU_PAGE_SIZE;
64 if (rdev->ih.ring_obj)
65 n -= rdev->ih.ring_size;
68 gtt_obj = kzalloc(n * sizeof(*gtt_obj), GFP_KERNEL);
70 DRM_ERROR("Failed to allocate %d pointers\n", n);
75 r = radeon_bo_create(rdev, size, PAGE_SIZE, true, RADEON_GEM_DOMAIN_VRAM,
78 DRM_ERROR("Failed to create VRAM object\n");
81 r = radeon_bo_reserve(vram_obj, false);
84 r = radeon_bo_pin(vram_obj, RADEON_GEM_DOMAIN_VRAM, &vram_addr);
86 DRM_ERROR("Failed to pin VRAM object\n");
89 for (i = 0; i < n; i++) {
90 void *gtt_map, *vram_map;
91 void **gtt_start, **gtt_end;
92 void **vram_start, **vram_end;
93 struct radeon_fence *fence = NULL;
95 r = radeon_bo_create(rdev, size, PAGE_SIZE, true,
96 RADEON_GEM_DOMAIN_GTT, NULL, gtt_obj + i);
98 DRM_ERROR("Failed to create GTT object %d\n", i);
102 r = radeon_bo_reserve(gtt_obj[i], false);
103 if (unlikely(r != 0))
104 goto out_lclean_unref;
105 r = radeon_bo_pin(gtt_obj[i], RADEON_GEM_DOMAIN_GTT, >t_addr);
107 DRM_ERROR("Failed to pin GTT object %d\n", i);
108 goto out_lclean_unres;
111 r = radeon_bo_kmap(gtt_obj[i], >t_map);
113 DRM_ERROR("Failed to map GTT object %d\n", i);
114 goto out_lclean_unpin;
117 for (gtt_start = gtt_map, gtt_end = gtt_map + size;
120 *gtt_start = gtt_start;
122 radeon_bo_kunmap(gtt_obj[i]);
124 if (ring == R600_RING_TYPE_DMA_INDEX)
125 r = radeon_copy_dma(rdev, gtt_addr, vram_addr, size / RADEON_GPU_PAGE_SIZE, &fence);
127 r = radeon_copy_blit(rdev, gtt_addr, vram_addr, size / RADEON_GPU_PAGE_SIZE, &fence);
129 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 r = radeon_copy_dma(rdev, vram_addr, gtt_addr, size / RADEON_GPU_PAGE_SIZE, &fence);
173 r = radeon_copy_blit(rdev, vram_addr, gtt_addr, size / RADEON_GPU_PAGE_SIZE, &fence);
175 DRM_ERROR("Failed VRAM->GTT copy %d\n", i);
176 goto out_lclean_unpin;
179 r = radeon_fence_wait(fence, false);
181 DRM_ERROR("Failed to wait for VRAM->GTT fence %d\n", i);
182 goto out_lclean_unpin;
185 radeon_fence_unref(&fence);
187 r = radeon_bo_kmap(gtt_obj[i], >t_map);
189 DRM_ERROR("Failed to map GTT object after copy %d\n", i);
190 goto out_lclean_unpin;
193 for (gtt_start = gtt_map, gtt_end = gtt_map + size,
194 vram_start = vram_map, vram_end = vram_map + size;
196 gtt_start++, vram_start++) {
197 if (*gtt_start != vram_start) {
198 DRM_ERROR("Incorrect VRAM->GTT copy %d: Got 0x%p, "
199 "expected 0x%p (VRAM/GTT offset "
200 "0x%16llx/0x%16llx)\n",
201 i, *gtt_start, vram_start,
203 (vram_addr - rdev->mc.vram_start +
204 (void*)vram_start - vram_map),
206 (gtt_addr - rdev->mc.gtt_start +
207 (void*)vram_start - vram_map));
208 radeon_bo_kunmap(gtt_obj[i]);
209 goto out_lclean_unpin;
213 radeon_bo_kunmap(gtt_obj[i]);
215 DRM_INFO("Tested GTT->VRAM and VRAM->GTT copy for GTT offset 0x%llx\n",
216 gtt_addr - rdev->mc.gtt_start);
220 radeon_bo_unpin(gtt_obj[i]);
222 radeon_bo_unreserve(gtt_obj[i]);
224 radeon_bo_unref(>t_obj[i]);
226 for (--i; i >= 0; --i) {
227 radeon_bo_unpin(gtt_obj[i]);
228 radeon_bo_unreserve(gtt_obj[i]);
229 radeon_bo_unref(>t_obj[i]);
232 radeon_fence_unref(&fence);
236 radeon_bo_unpin(vram_obj);
238 radeon_bo_unreserve(vram_obj);
240 radeon_bo_unref(&vram_obj);
244 printk(KERN_WARNING "Error while testing BO move.\n");
248 void radeon_test_moves(struct radeon_device *rdev)
250 if (rdev->asic->copy.dma)
251 radeon_do_test_moves(rdev, RADEON_TEST_COPY_DMA);
252 if (rdev->asic->copy.blit)
253 radeon_do_test_moves(rdev, RADEON_TEST_COPY_BLIT);
256 static int radeon_test_create_and_emit_fence(struct radeon_device *rdev,
257 struct radeon_ring *ring,
258 struct radeon_fence **fence)
262 if (ring->idx == R600_RING_TYPE_UVD_INDEX) {
263 r = radeon_uvd_get_create_msg(rdev, ring->idx, 1, NULL);
265 DRM_ERROR("Failed to get dummy create msg\n");
269 r = radeon_uvd_get_destroy_msg(rdev, ring->idx, 1, fence);
271 DRM_ERROR("Failed to get dummy destroy msg\n");
275 r = radeon_ring_lock(rdev, ring, 64);
277 DRM_ERROR("Failed to lock ring A %d\n", ring->idx);
280 radeon_fence_emit(rdev, fence, ring->idx);
281 radeon_ring_unlock_commit(rdev, ring);
286 void radeon_test_ring_sync(struct radeon_device *rdev,
287 struct radeon_ring *ringA,
288 struct radeon_ring *ringB)
290 struct radeon_fence *fence1 = NULL, *fence2 = NULL;
291 struct radeon_semaphore *semaphore = NULL;
294 r = radeon_semaphore_create(rdev, &semaphore);
296 DRM_ERROR("Failed to create semaphore\n");
300 r = radeon_ring_lock(rdev, ringA, 64);
302 DRM_ERROR("Failed to lock ring A %d\n", ringA->idx);
305 radeon_semaphore_emit_wait(rdev, ringA->idx, semaphore);
306 radeon_ring_unlock_commit(rdev, ringA);
308 r = radeon_test_create_and_emit_fence(rdev, ringA, &fence1);
312 r = radeon_ring_lock(rdev, ringA, 64);
314 DRM_ERROR("Failed to lock ring A %d\n", ringA->idx);
317 radeon_semaphore_emit_wait(rdev, ringA->idx, semaphore);
318 radeon_ring_unlock_commit(rdev, ringA);
320 r = radeon_test_create_and_emit_fence(rdev, ringA, &fence2);
326 if (radeon_fence_signaled(fence1)) {
327 DRM_ERROR("Fence 1 signaled without waiting for semaphore.\n");
331 r = radeon_ring_lock(rdev, ringB, 64);
333 DRM_ERROR("Failed to lock ring B %p\n", ringB);
336 radeon_semaphore_emit_signal(rdev, ringB->idx, semaphore);
337 radeon_ring_unlock_commit(rdev, ringB);
339 r = radeon_fence_wait(fence1, false);
341 DRM_ERROR("Failed to wait for sync fence 1\n");
347 if (radeon_fence_signaled(fence2)) {
348 DRM_ERROR("Fence 2 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);
360 r = radeon_fence_wait(fence2, false);
362 DRM_ERROR("Failed to wait for sync fence 1\n");
367 radeon_semaphore_free(rdev, &semaphore, NULL);
370 radeon_fence_unref(&fence1);
373 radeon_fence_unref(&fence2);
376 printk(KERN_WARNING "Error while testing ring sync (%d).\n", r);
379 static void radeon_test_ring_sync2(struct radeon_device *rdev,
380 struct radeon_ring *ringA,
381 struct radeon_ring *ringB,
382 struct radeon_ring *ringC)
384 struct radeon_fence *fenceA = NULL, *fenceB = NULL;
385 struct radeon_semaphore *semaphore = NULL;
389 r = radeon_semaphore_create(rdev, &semaphore);
391 DRM_ERROR("Failed to create semaphore\n");
395 r = radeon_ring_lock(rdev, ringA, 64);
397 DRM_ERROR("Failed to lock ring A %d\n", ringA->idx);
400 radeon_semaphore_emit_wait(rdev, ringA->idx, semaphore);
401 radeon_ring_unlock_commit(rdev, ringA);
403 r = radeon_test_create_and_emit_fence(rdev, ringA, &fenceA);
407 r = radeon_ring_lock(rdev, ringB, 64);
409 DRM_ERROR("Failed to lock ring B %d\n", ringB->idx);
412 radeon_semaphore_emit_wait(rdev, ringB->idx, semaphore);
413 radeon_ring_unlock_commit(rdev, ringB);
414 r = radeon_test_create_and_emit_fence(rdev, ringB, &fenceB);
420 if (radeon_fence_signaled(fenceA)) {
421 DRM_ERROR("Fence A signaled without waiting for semaphore.\n");
424 if (radeon_fence_signaled(fenceB)) {
425 DRM_ERROR("Fence B signaled without waiting for semaphore.\n");
429 r = radeon_ring_lock(rdev, ringC, 64);
431 DRM_ERROR("Failed to lock ring B %p\n", ringC);
434 radeon_semaphore_emit_signal(rdev, ringC->idx, semaphore);
435 radeon_ring_unlock_commit(rdev, ringC);
437 for (i = 0; i < 30; ++i) {
439 sigA = radeon_fence_signaled(fenceA);
440 sigB = radeon_fence_signaled(fenceB);
445 if (!sigA && !sigB) {
446 DRM_ERROR("Neither fence A nor B has been signaled\n");
448 } else if (sigA && sigB) {
449 DRM_ERROR("Both fence A and B has been signaled\n");
453 DRM_INFO("Fence %c was first signaled\n", sigA ? 'A' : 'B');
455 r = radeon_ring_lock(rdev, ringC, 64);
457 DRM_ERROR("Failed to lock ring B %p\n", ringC);
460 radeon_semaphore_emit_signal(rdev, ringC->idx, semaphore);
461 radeon_ring_unlock_commit(rdev, ringC);
465 r = radeon_fence_wait(fenceA, false);
467 DRM_ERROR("Failed to wait for sync fence A\n");
470 r = radeon_fence_wait(fenceB, false);
472 DRM_ERROR("Failed to wait for sync fence B\n");
477 radeon_semaphore_free(rdev, &semaphore, NULL);
480 radeon_fence_unref(&fenceA);
483 radeon_fence_unref(&fenceB);
486 printk(KERN_WARNING "Error while testing ring sync (%d).\n", r);
489 void radeon_test_syncing(struct radeon_device *rdev)
493 for (i = 1; i < RADEON_NUM_RINGS; ++i) {
494 struct radeon_ring *ringA = &rdev->ring[i];
498 for (j = 0; j < i; ++j) {
499 struct radeon_ring *ringB = &rdev->ring[j];
503 DRM_INFO("Testing syncing between rings %d and %d...\n", i, j);
504 radeon_test_ring_sync(rdev, ringA, ringB);
506 DRM_INFO("Testing syncing between rings %d and %d...\n", j, i);
507 radeon_test_ring_sync(rdev, ringB, ringA);
509 for (k = 0; k < j; ++k) {
510 struct radeon_ring *ringC = &rdev->ring[k];
514 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", i, j, k);
515 radeon_test_ring_sync2(rdev, ringA, ringB, ringC);
517 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", i, k, j);
518 radeon_test_ring_sync2(rdev, ringA, ringC, ringB);
520 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", j, i, k);
521 radeon_test_ring_sync2(rdev, ringB, ringA, ringC);
523 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", j, k, i);
524 radeon_test_ring_sync2(rdev, ringB, ringC, ringA);
526 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", k, i, j);
527 radeon_test_ring_sync2(rdev, ringC, ringA, ringB);
529 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", k, j, i);
530 radeon_test_ring_sync2(rdev, ringC, ringB, ringA);