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