1 /*
2  * Copyright © 2009,2012,2013 Intel Corporation
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 (including the next
12  * paragraph) shall be included in all copies or substantial portions of the
13  * Software.
14  *
15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21  * IN THE SOFTWARE.
22  *
23  * Authors:
24  *    Eric Anholt <eric@anholt.net>
25  *    Chris Wilson <chris@chris-wilson.co.uk>
26  *    Daniel Vetter <daniel.vetter@ffwll.ch>
27  *
28  */
29 
30 /** @file gem_concurrent.c
31  *
32  * This is a test of pread/pwrite/mmap behavior when writing to active
33  * buffers.
34  *
35  * Based on gem_gtt_concurrent_blt.
36  */
37 
38 #include "igt.h"
39 #include "igt_vgem.h"
40 #include <stdlib.h>
41 #include <stdio.h>
42 #include <string.h>
43 #include <fcntl.h>
44 #include <inttypes.h>
45 #include <errno.h>
46 #include <sys/resource.h>
47 #include <sys/stat.h>
48 #include <sys/time.h>
49 #include <sys/wait.h>
50 
51 #include <drm.h>
52 
53 #include "intel_bufmgr.h"
54 
55 IGT_TEST_DESCRIPTION("Test of pread/pwrite/mmap behavior when writing to active"
56 		     " buffers.");
57 
58 int fd, devid, gen;
59 int vgem_drv = -1;
60 int all;
61 int pass;
62 
63 struct create {
64 	const char *name;
65 	void (*require)(const struct create *, unsigned);
66 	drm_intel_bo *(*create)(drm_intel_bufmgr *, uint64_t size);
67 };
68 
69 struct size {
70 	const char *name;
71 	int width, height;
72 };
73 
74 struct buffers {
75 	const char *name;
76 	const struct create *create;
77 	const struct access_mode *mode;
78 	const struct size *size;
79 	drm_intel_bufmgr *bufmgr;
80 	struct intel_batchbuffer *batch;
81 	drm_intel_bo **src, **dst;
82 	drm_intel_bo *snoop, *spare;
83 	uint32_t *tmp;
84 	int width, height, npixels, page_size;
85 	int count, num_buffers;
86 };
87 
88 #define MIN_BUFFERS 3
89 
90 static void blt_copy_bo(struct buffers *b, drm_intel_bo *dst, drm_intel_bo *src);
91 
92 static void
nop_release_bo(drm_intel_bo * bo)93 nop_release_bo(drm_intel_bo *bo)
94 {
95 	drm_intel_bo_unreference(bo);
96 }
97 
98 static void
prw_set_bo(struct buffers * b,drm_intel_bo * bo,uint32_t val)99 prw_set_bo(struct buffers *b, drm_intel_bo *bo, uint32_t val)
100 {
101 	for (int i = 0; i < b->npixels; i++)
102 		b->tmp[i] = val;
103 	drm_intel_bo_subdata(bo, 0, 4*b->npixels, b->tmp);
104 }
105 
106 static void
prw_cmp_bo(struct buffers * b,drm_intel_bo * bo,uint32_t val)107 prw_cmp_bo(struct buffers *b, drm_intel_bo *bo, uint32_t val)
108 {
109 	uint32_t *vaddr;
110 
111 	vaddr = b->tmp;
112 	do_or_die(drm_intel_bo_get_subdata(bo, 0, 4*b->npixels, vaddr));
113 	for (int i = 0; i < b->npixels; i++)
114 		igt_assert_eq_u32(vaddr[i], val);
115 }
116 
117 #define pixel(y, width) ((y)*(width) + (((y) + pass)%(width)))
118 
119 static void
partial_set_bo(struct buffers * b,drm_intel_bo * bo,uint32_t val)120 partial_set_bo(struct buffers *b, drm_intel_bo *bo, uint32_t val)
121 {
122 	for (int y = 0; y < b->height; y++)
123 		do_or_die(drm_intel_bo_subdata(bo, 4*pixel(y, b->width), 4, &val));
124 }
125 
126 static void
partial_cmp_bo(struct buffers * b,drm_intel_bo * bo,uint32_t val)127 partial_cmp_bo(struct buffers *b, drm_intel_bo *bo, uint32_t val)
128 {
129 	for (int y = 0; y < b->height; y++) {
130 		uint32_t buf;
131 		do_or_die(drm_intel_bo_get_subdata(bo, 4*pixel(y, b->width), 4, &buf));
132 		igt_assert_eq_u32(buf, val);
133 	}
134 }
135 
136 static drm_intel_bo *
create_normal_bo(drm_intel_bufmgr * bufmgr,uint64_t size)137 create_normal_bo(drm_intel_bufmgr *bufmgr, uint64_t size)
138 {
139 	drm_intel_bo *bo;
140 
141 	bo = drm_intel_bo_alloc(bufmgr, "bo", size, 0);
142 	igt_assert(bo);
143 
144 	return bo;
145 }
146 
can_create_normal(const struct create * create,unsigned count)147 static void can_create_normal(const struct create *create, unsigned count)
148 {
149 }
150 
151 #if HAVE_CREATE_PRIVATE
152 static drm_intel_bo *
create_private_bo(drm_intel_bufmgr * bufmgr,uint64_t size)153 create_private_bo(drm_intel_bufmgr *bufmgr, uint64_t size)
154 {
155 	drm_intel_bo *bo;
156 	uint32_t handle;
157 
158 	/* XXX gem_create_with_flags(fd, size, I915_CREATE_PRIVATE); */
159 
160 	handle = gem_create(fd, size);
161 	bo = gem_handle_to_libdrm_bo(bufmgr, fd, "stolen", handle);
162 	gem_close(fd, handle);
163 
164 	return bo;
165 }
166 
can_create_private(const struct create * create,unsigned count)167 static void can_create_private(const struct create *create, unsigned count)
168 {
169 	igt_require(0);
170 }
171 #endif
172 
173 #if HAVE_CREATE_STOLEN
174 static drm_intel_bo *
create_stolen_bo(drm_intel_bufmgr * bufmgr,uint64_t size)175 create_stolen_bo(drm_intel_bufmgr *bufmgr, uint64_t size)
176 {
177 	drm_intel_bo *bo;
178 	uint32_t handle;
179 
180 	/* XXX gem_create_with_flags(fd, size, I915_CREATE_STOLEN); */
181 
182 	handle = gem_create(fd, size);
183 	bo = gem_handle_to_libdrm_bo(bufmgr, fd, "stolen", handle);
184 	gem_close(fd, handle);
185 
186 	return bo;
187 }
188 
can_create_stolen(const struct create * create,unsigned count)189 static void can_create_stolen(const struct create *create, unsigned count)
190 {
191 	/* XXX check num_buffers against available stolen */
192 	igt_require(0);
193 }
194 #endif
195 
create_cpu_require(const struct create * create,unsigned count)196 static void create_cpu_require(const struct create *create, unsigned count)
197 {
198 #if HAVE_CREATE_STOLEN
199 	igt_require(create->create != create_stolen_bo);
200 #endif
201 }
202 
203 static drm_intel_bo *
unmapped_create_bo(const struct buffers * b)204 unmapped_create_bo(const struct buffers *b)
205 {
206 	return b->create->create(b->bufmgr, 4*b->npixels);
207 }
208 
create_snoop_require(const struct create * create,unsigned count)209 static void create_snoop_require(const struct create *create, unsigned count)
210 {
211 	create_cpu_require(create, count);
212 	igt_require(!gem_has_llc(fd));
213 }
214 
215 static drm_intel_bo *
snoop_create_bo(const struct buffers * b)216 snoop_create_bo(const struct buffers *b)
217 {
218 	drm_intel_bo *bo;
219 
220 	bo = unmapped_create_bo(b);
221 	gem_set_caching(fd, bo->handle, I915_CACHING_CACHED);
222 	drm_intel_bo_disable_reuse(bo);
223 
224 	return bo;
225 }
226 
create_userptr_require(const struct create * create,unsigned count)227 static void create_userptr_require(const struct create *create, unsigned count)
228 {
229 	static int has_userptr = -1;
230 	if (has_userptr < 0) {
231 		struct drm_i915_gem_userptr arg;
232 
233 		has_userptr = 0;
234 
235 		memset(&arg, 0, sizeof(arg));
236 		arg.user_ptr = -4096ULL;
237 		arg.user_size = 8192;
238 		errno = 0;
239 		drmIoctl(fd, LOCAL_IOCTL_I915_GEM_USERPTR, &arg);
240 		if (errno == EFAULT) {
241 			igt_assert(posix_memalign((void **)&arg.user_ptr,
242 						  4096, arg.user_size) == 0);
243 			has_userptr = drmIoctl(fd,
244 					 LOCAL_IOCTL_I915_GEM_USERPTR,
245 					 &arg) == 0;
246 			free(from_user_pointer(arg.user_ptr));
247 		}
248 
249 	}
250 	igt_require(has_userptr);
251 }
252 
253 static drm_intel_bo *
userptr_create_bo(const struct buffers * b)254 userptr_create_bo(const struct buffers *b)
255 {
256 	struct local_i915_gem_userptr userptr;
257 	drm_intel_bo *bo;
258 	void *ptr;
259 
260 	memset(&userptr, 0, sizeof(userptr));
261 	userptr.user_size = b->page_size;
262 
263 	ptr = mmap(NULL, userptr.user_size,
264 		   PROT_READ | PROT_WRITE, MAP_ANON | MAP_SHARED, -1, 0);
265 	igt_assert(ptr != (void *)-1);
266 	userptr.user_ptr = to_user_pointer(ptr);
267 
268 #if 0
269 	do_or_die(drmIoctl(fd, LOCAL_IOCTL_I915_GEM_USERPTR, &userptr));
270 	bo = gem_handle_to_libdrm_bo(b->bufmgr, fd, "userptr", userptr.handle);
271 	gem_close(fd, userptr.handle);
272 #else
273 	bo = drm_intel_bo_alloc_userptr(b->bufmgr, "name",
274 					ptr, I915_TILING_NONE, 0,
275 					userptr.user_size, 0);
276 	igt_assert(bo);
277 #endif
278 	bo->virtual = from_user_pointer(userptr.user_ptr);
279 
280 	return bo;
281 }
282 
283 static void
userptr_set_bo(struct buffers * b,drm_intel_bo * bo,uint32_t val)284 userptr_set_bo(struct buffers *b, drm_intel_bo *bo, uint32_t val)
285 {
286 	int size = b->npixels;
287 	uint32_t *vaddr = bo->virtual;
288 
289 	gem_set_domain(fd, bo->handle,
290 		       I915_GEM_DOMAIN_CPU, I915_GEM_DOMAIN_CPU);
291 	while (size--)
292 		*vaddr++ = val;
293 }
294 
295 static void
userptr_cmp_bo(struct buffers * b,drm_intel_bo * bo,uint32_t val)296 userptr_cmp_bo(struct buffers *b, drm_intel_bo *bo, uint32_t val)
297 {
298 	int size =  b->npixels;
299 	uint32_t *vaddr = bo->virtual;
300 
301 	gem_set_domain(fd, bo->handle,
302 		       I915_GEM_DOMAIN_CPU, 0);
303 	while (size--)
304 		igt_assert_eq_u32(*vaddr++, val);
305 }
306 
307 static void
userptr_release_bo(drm_intel_bo * bo)308 userptr_release_bo(drm_intel_bo *bo)
309 {
310 	igt_assert(bo->virtual);
311 
312 	munmap(bo->virtual, bo->size);
313 	bo->virtual = NULL;
314 
315 	drm_intel_bo_unreference(bo);
316 }
317 
create_dmabuf_require(const struct create * create,unsigned count)318 static void create_dmabuf_require(const struct create *create, unsigned count)
319 {
320 	static int has_dmabuf = -1;
321 	if (has_dmabuf < 0) {
322 		struct drm_prime_handle args;
323 		void *ptr;
324 
325 		memset(&args, 0, sizeof(args));
326 		args.handle = gem_create(fd, 4096);
327 		args.flags = DRM_RDWR;
328 		args.fd = -1;
329 
330 		drmIoctl(fd, DRM_IOCTL_PRIME_HANDLE_TO_FD, &args);
331 		gem_close(fd, args.handle);
332 
333 		has_dmabuf = 0;
334 		ptr = mmap(NULL, 4096, PROT_READ, MAP_SHARED, args.fd, 0);
335 		if (ptr != MAP_FAILED) {
336 			has_dmabuf = 1;
337 			munmap(ptr, 4096);
338 		}
339 
340 		close(args.fd);
341 	}
342 	igt_require(has_dmabuf);
343 	intel_require_files(2*count);
344 }
345 
346 struct dmabuf {
347 	int fd;
348 	void *map;
349 };
350 
351 static drm_intel_bo *
dmabuf_create_bo(const struct buffers * b)352 dmabuf_create_bo(const struct buffers *b)
353 {
354 	struct drm_prime_handle args;
355 	drm_intel_bo *bo;
356 	struct dmabuf *dmabuf;
357 	int size;
358 
359 	size = b->page_size;
360 
361 	memset(&args, 0, sizeof(args));
362 	args.handle = gem_create(fd, size);
363 	args.flags = DRM_RDWR;
364 	args.fd = -1;
365 
366 	do_ioctl(fd, DRM_IOCTL_PRIME_HANDLE_TO_FD, &args);
367 	gem_close(fd, args.handle);
368 
369 	bo = drm_intel_bo_gem_create_from_prime(b->bufmgr, args.fd, size);
370 	igt_assert(bo);
371 
372 	dmabuf = malloc(sizeof(*dmabuf));
373 	igt_assert(dmabuf);
374 
375 	dmabuf->fd = args.fd;
376 	dmabuf->map = mmap(NULL, size,
377 			   PROT_READ | PROT_WRITE, MAP_SHARED,
378 			   dmabuf->fd, 0);
379 	igt_assert(dmabuf->map != (void *)-1);
380 
381 	bo->virtual = dmabuf;
382 
383 	return bo;
384 }
385 
386 static void
dmabuf_set_bo(struct buffers * b,drm_intel_bo * bo,uint32_t val)387 dmabuf_set_bo(struct buffers *b, drm_intel_bo *bo, uint32_t val)
388 {
389 	struct dmabuf *dmabuf = bo->virtual;
390 	uint32_t *v = dmabuf->map;
391 	int y;
392 
393 	prime_sync_start(dmabuf->fd, true);
394 	for (y = 0; y < b->height; y++)
395 		v[pixel(y, b->width)] = val;
396 	prime_sync_end(dmabuf->fd, true);
397 }
398 
399 static void
dmabuf_cmp_bo(struct buffers * b,drm_intel_bo * bo,uint32_t val)400 dmabuf_cmp_bo(struct buffers *b, drm_intel_bo *bo, uint32_t val)
401 {
402 	struct dmabuf *dmabuf = bo->virtual;
403 	uint32_t *v = dmabuf->map;
404 	int y;
405 
406 	prime_sync_start(dmabuf->fd, false);
407 	for (y = 0; y < b->height; y++)
408 		igt_assert_eq_u32(v[pixel(y, b->width)], val);
409 	prime_sync_end(dmabuf->fd, false);
410 }
411 
412 static void
dmabuf_release_bo(drm_intel_bo * bo)413 dmabuf_release_bo(drm_intel_bo *bo)
414 {
415 	struct dmabuf *dmabuf = bo->virtual;
416 	igt_assert(dmabuf);
417 
418 	munmap(dmabuf->map, bo->size);
419 	close(dmabuf->fd);
420 	free(dmabuf);
421 
422 	bo->virtual = NULL;
423 	drm_intel_bo_unreference(bo);
424 }
425 
has_prime_export(int _fd)426 static bool has_prime_export(int _fd)
427 {
428 	uint64_t value;
429 
430 	if (drmGetCap(_fd, DRM_CAP_PRIME, &value))
431 		return false;
432 
433 	return value & DRM_PRIME_CAP_EXPORT;
434 }
435 
create_vgem_require(const struct create * create,unsigned count)436 static void create_vgem_require(const struct create *create, unsigned count)
437 {
438 	igt_require(vgem_drv != -1);
439 	igt_require(has_prime_export(vgem_drv));
440 	create_dmabuf_require(create, count);
441 }
442 
443 static drm_intel_bo *
vgem_create_bo(const struct buffers * b)444 vgem_create_bo(const struct buffers *b)
445 {
446 	struct drm_prime_handle args;
447 	drm_intel_bo *bo;
448 	struct vgem_bo vgem;
449 	struct dmabuf *dmabuf;
450 
451 	igt_assert(vgem_drv != -1);
452 
453 	vgem.width = b->width;
454 	vgem.height = b->height;
455 	vgem.bpp = 32;
456 	vgem_create(vgem_drv, &vgem);
457 
458 	memset(&args, 0, sizeof(args));
459 	args.handle = vgem.handle;
460 	args.flags = DRM_RDWR;
461 	args.fd = -1;
462 
463 	do_ioctl(vgem_drv, DRM_IOCTL_PRIME_HANDLE_TO_FD, &args);
464 	gem_close(vgem_drv, args.handle);
465 	igt_assert(args.fd != -1);
466 
467 	bo = drm_intel_bo_gem_create_from_prime(b->bufmgr, args.fd, vgem.size);
468 	igt_assert(bo);
469 
470 	dmabuf = malloc(sizeof(*dmabuf));
471 	igt_assert(dmabuf);
472 
473 	dmabuf->fd = args.fd;
474 	dmabuf->map = mmap(NULL, vgem.size,
475 			   PROT_READ | PROT_WRITE, MAP_SHARED,
476 			   dmabuf->fd, 0);
477 	igt_assert(dmabuf->map != (void *)-1);
478 
479 	bo->virtual = dmabuf;
480 
481 	return bo;
482 }
483 
484 static void
gtt_set_bo(struct buffers * b,drm_intel_bo * bo,uint32_t val)485 gtt_set_bo(struct buffers *b, drm_intel_bo *bo, uint32_t val)
486 {
487 	uint32_t *vaddr = bo->virtual;
488 
489 	drm_intel_gem_bo_start_gtt_access(bo, true);
490 	for (int y = 0; y < b->height; y++)
491 		vaddr[pixel(y, b->width)] = val;
492 }
493 
494 static void
gtt_cmp_bo(struct buffers * b,drm_intel_bo * bo,uint32_t val)495 gtt_cmp_bo(struct buffers *b, drm_intel_bo *bo, uint32_t val)
496 {
497 	uint32_t *vaddr = bo->virtual;
498 
499 	/* GTT access is slow. So we just compare a few points */
500 	drm_intel_gem_bo_start_gtt_access(bo, false);
501 	for (int y = 0; y < b->height; y++)
502 		igt_assert_eq_u32(vaddr[pixel(y, b->width)], val);
503 }
504 
505 static drm_intel_bo *
map_bo(drm_intel_bo * bo)506 map_bo(drm_intel_bo *bo)
507 {
508 	/* gtt map doesn't have a write parameter, so just keep the mapping
509 	 * around (to avoid the set_domain with the gtt write domain set) and
510 	 * manually tell the kernel when we start access the gtt. */
511 	do_or_die(drm_intel_gem_bo_map_gtt(bo));
512 
513 	return bo;
514 }
515 
516 static drm_intel_bo *
tile_bo(drm_intel_bo * bo,int width)517 tile_bo(drm_intel_bo *bo, int width)
518 {
519 	uint32_t tiling = I915_TILING_X;
520 	uint32_t stride = width * 4;
521 
522 	do_or_die(drm_intel_bo_set_tiling(bo, &tiling, stride));
523 
524 	return bo;
525 }
526 
527 static drm_intel_bo *
gtt_create_bo(const struct buffers * b)528 gtt_create_bo(const struct buffers *b)
529 {
530 	return map_bo(unmapped_create_bo(b));
531 }
532 
533 static drm_intel_bo *
gttX_create_bo(const struct buffers * b)534 gttX_create_bo(const struct buffers *b)
535 {
536 	return tile_bo(gtt_create_bo(b), b->width);
537 }
538 
bit17_require(void)539 static void bit17_require(void)
540 {
541 	static struct drm_i915_gem_get_tiling2 {
542 		uint32_t handle;
543 		uint32_t tiling_mode;
544 		uint32_t swizzle_mode;
545 		uint32_t phys_swizzle_mode;
546 	} arg;
547 #define DRM_IOCTL_I915_GEM_GET_TILING2	DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_GET_TILING, struct drm_i915_gem_get_tiling2)
548 
549 	if (arg.handle == 0) {
550 		arg.handle = gem_create(fd, 4096);
551 		gem_set_tiling(fd, arg.handle, I915_TILING_X, 512);
552 
553 		do_ioctl(fd, DRM_IOCTL_I915_GEM_GET_TILING2, &arg);
554 		gem_close(fd, arg.handle);
555 	}
556 	igt_require(arg.phys_swizzle_mode == arg.swizzle_mode);
557 }
558 
wc_require(void)559 static void wc_require(void)
560 {
561 	bit17_require();
562 	gem_require_mmap_wc(fd);
563 }
564 
565 static void
wc_create_require(const struct create * create,unsigned count)566 wc_create_require(const struct create *create, unsigned count)
567 {
568 	wc_require();
569 }
570 
571 static drm_intel_bo *
wc_create_bo(const struct buffers * b)572 wc_create_bo(const struct buffers *b)
573 {
574 	drm_intel_bo *bo;
575 
576 	bo = unmapped_create_bo(b);
577 	bo->virtual = gem_mmap__wc(fd, bo->handle, 0, bo->size, PROT_READ | PROT_WRITE);
578 	return bo;
579 }
580 
581 static void
wc_release_bo(drm_intel_bo * bo)582 wc_release_bo(drm_intel_bo *bo)
583 {
584 	igt_assert(bo->virtual);
585 
586 	munmap(bo->virtual, bo->size);
587 	bo->virtual = NULL;
588 
589 	nop_release_bo(bo);
590 }
591 
592 static drm_intel_bo *
gpu_create_bo(const struct buffers * b)593 gpu_create_bo(const struct buffers *b)
594 {
595 	return unmapped_create_bo(b);
596 }
597 
598 static drm_intel_bo *
gpuX_create_bo(const struct buffers * b)599 gpuX_create_bo(const struct buffers *b)
600 {
601 	return tile_bo(gpu_create_bo(b), b->width);
602 }
603 
604 static void
cpu_set_bo(struct buffers * b,drm_intel_bo * bo,uint32_t val)605 cpu_set_bo(struct buffers *b, drm_intel_bo *bo, uint32_t val)
606 {
607 	int size = b->npixels;
608 	uint32_t *vaddr;
609 
610 	do_or_die(drm_intel_bo_map(bo, true));
611 	vaddr = bo->virtual;
612 	while (size--)
613 		*vaddr++ = val;
614 	drm_intel_bo_unmap(bo);
615 }
616 
617 static void
cpu_cmp_bo(struct buffers * b,drm_intel_bo * bo,uint32_t val)618 cpu_cmp_bo(struct buffers *b, drm_intel_bo *bo, uint32_t val)
619 {
620 	int size = b->npixels;
621 	uint32_t *vaddr;
622 
623 	do_or_die(drm_intel_bo_map(bo, false));
624 	vaddr = bo->virtual;
625 	while (size--)
626 		igt_assert_eq_u32(*vaddr++, val);
627 	drm_intel_bo_unmap(bo);
628 }
629 
630 static void
gpu_set_bo(struct buffers * buffers,drm_intel_bo * bo,uint32_t val)631 gpu_set_bo(struct buffers *buffers, drm_intel_bo *bo, uint32_t val)
632 {
633 	struct drm_i915_gem_relocation_entry reloc[1];
634 	struct drm_i915_gem_exec_object2 gem_exec[2];
635 	struct drm_i915_gem_execbuffer2 execbuf;
636 	uint32_t buf[10], *b;
637 	uint32_t tiling, swizzle;
638 
639 	drm_intel_bo_get_tiling(bo, &tiling, &swizzle);
640 
641 	memset(reloc, 0, sizeof(reloc));
642 	memset(gem_exec, 0, sizeof(gem_exec));
643 	memset(&execbuf, 0, sizeof(execbuf));
644 
645 	b = buf;
646 	*b++ = XY_COLOR_BLT_CMD_NOLEN |
647 		((gen >= 8) ? 5 : 4) |
648 		COLOR_BLT_WRITE_ALPHA | XY_COLOR_BLT_WRITE_RGB;
649 	if (gen >= 4 && tiling) {
650 		b[-1] |= XY_COLOR_BLT_TILED;
651 		*b = buffers->width;
652 	} else
653 		*b = buffers->width << 2;
654 	*b++ |= 0xf0 << 16 | 1 << 25 | 1 << 24;
655 	*b++ = 0;
656 	*b++ = buffers->height << 16 | buffers->width;
657 	reloc[0].offset = (b - buf) * sizeof(uint32_t);
658 	reloc[0].target_handle = bo->handle;
659 	reloc[0].read_domains = I915_GEM_DOMAIN_RENDER;
660 	reloc[0].write_domain = I915_GEM_DOMAIN_RENDER;
661 	*b++ = 0;
662 	if (gen >= 8)
663 		*b++ = 0;
664 	*b++ = val;
665 	*b++ = MI_BATCH_BUFFER_END;
666 	if ((b - buf) & 1)
667 		*b++ = 0;
668 
669 	gem_exec[0].handle = bo->handle;
670 	gem_exec[0].flags = EXEC_OBJECT_NEEDS_FENCE;
671 
672 	gem_exec[1].handle = gem_create(fd, 4096);
673 	gem_exec[1].relocation_count = 1;
674 	gem_exec[1].relocs_ptr = to_user_pointer(reloc);
675 
676 	execbuf.buffers_ptr = to_user_pointer(gem_exec);
677 	execbuf.buffer_count = 2;
678 	execbuf.batch_len = (b - buf) * sizeof(buf[0]);
679 	if (gen >= 6)
680 		execbuf.flags = I915_EXEC_BLT;
681 
682 	gem_write(fd, gem_exec[1].handle, 0, buf, execbuf.batch_len);
683 	gem_execbuf(fd, &execbuf);
684 
685 	gem_close(fd, gem_exec[1].handle);
686 }
687 
688 static void
gpu_cmp_bo(struct buffers * b,drm_intel_bo * bo,uint32_t val)689 gpu_cmp_bo(struct buffers *b, drm_intel_bo *bo, uint32_t val)
690 {
691 	blt_copy_bo(b, b->snoop, bo);
692 	cpu_cmp_bo(b, b->snoop, val);
693 }
694 
695 struct access_mode {
696 	const char *name;
697 	void (*require)(const struct create *, unsigned);
698 	drm_intel_bo *(*create_bo)(const struct buffers *b);
699 	void (*set_bo)(struct buffers *b, drm_intel_bo *bo, uint32_t val);
700 	void (*cmp_bo)(struct buffers *b, drm_intel_bo *bo, uint32_t val);
701 	void (*release_bo)(drm_intel_bo *bo);
702 };
703 igt_render_copyfunc_t rendercopy;
704 
read_sysctl(const char * path)705 static int read_sysctl(const char *path)
706 {
707 	FILE *file = fopen(path, "r");
708 	int max = 0;
709 	if (file) {
710 		if (fscanf(file, "%d", &max) != 1)
711 			max = 0; /* silence! */
712 		fclose(file);
713 	}
714 	return max;
715 }
716 
write_sysctl(const char * path,int value)717 static int write_sysctl(const char *path, int value)
718 {
719 	FILE *file = fopen(path, "w");
720 	if (file) {
721 		fprintf(file, "%d", value);
722 		fclose(file);
723 	}
724 	return read_sysctl(path);
725 }
726 
set_max_map_count(int num_buffers)727 static bool set_max_map_count(int num_buffers)
728 {
729 	int max = read_sysctl("/proc/sys/vm/max_map_count");
730 	if (max < num_buffers + 1024)
731 		max = write_sysctl("/proc/sys/vm/max_map_count",
732 				   num_buffers + 1024);
733 	return max > num_buffers;
734 }
735 
buffers_init(struct buffers * b,const char * name,const struct create * create,const struct access_mode * mode,const struct size * size,int num_buffers,int _fd,int enable_reuse)736 static void buffers_init(struct buffers *b,
737 			 const char *name,
738 			 const struct create *create,
739 			 const struct access_mode *mode,
740 			 const struct size *size,
741 			 int num_buffers,
742 			 int _fd, int enable_reuse)
743 {
744 	memset(b, 0, sizeof(*b));
745 	b->name = name;
746 	b->create = create;
747 	b->mode = mode;
748 	b->size = size;
749 	b->num_buffers = num_buffers;
750 	b->count = 0;
751 
752 	b->width = size->width;
753 	b->height = size->height;
754 	b->npixels = size->width * size->height;
755 	b->page_size = 4*b->npixels;
756 	b->page_size = (b->page_size + 4095) & -4096;
757 	b->tmp = malloc(b->page_size);
758 	igt_assert(b->tmp);
759 
760 	b->bufmgr = drm_intel_bufmgr_gem_init(_fd, 4096);
761 	igt_assert(b->bufmgr);
762 
763 	b->src = malloc(2*sizeof(drm_intel_bo *)*num_buffers);
764 	igt_assert(b->src);
765 	b->dst = b->src + num_buffers;
766 
767 	if (enable_reuse)
768 		drm_intel_bufmgr_gem_enable_reuse(b->bufmgr);
769 	b->batch = intel_batchbuffer_alloc(b->bufmgr, devid);
770 	igt_assert(b->batch);
771 }
772 
buffers_destroy(struct buffers * b)773 static void buffers_destroy(struct buffers *b)
774 {
775 	int count = b->count;
776 	if (count == 0)
777 		return;
778 
779 	/* Be safe so that we can clean up a partial creation */
780 	b->count = 0;
781 	for (int i = 0; i < count; i++) {
782 		if (b->src[i]) {
783 			b->mode->release_bo(b->src[i]);
784 			b->src[i] = NULL;
785 		} else
786 			break;
787 
788 		if (b->dst[i]) {
789 			b->mode->release_bo(b->dst[i]);
790 			b->dst[i] = NULL;
791 		}
792 	}
793 	if (b->snoop) {
794 		nop_release_bo(b->snoop);
795 		b->snoop = NULL;
796 	}
797 	if (b->spare) {
798 		b->mode->release_bo(b->spare);
799 		b->spare = NULL;
800 	}
801 }
802 
buffers_create(struct buffers * b)803 static void buffers_create(struct buffers *b)
804 {
805 	int count = b->num_buffers;
806 	igt_assert(b->bufmgr);
807 
808 	buffers_destroy(b);
809 	igt_assert(b->count == 0);
810 	b->count = count;
811 
812 	for (int i = 0; i < count; i++) {
813 		b->src[i] = b->mode->create_bo(b);
814 		b->dst[i] = b->mode->create_bo(b);
815 	}
816 	b->spare = b->mode->create_bo(b);
817 	b->snoop = snoop_create_bo(b);
818 }
819 
buffers_reset(struct buffers * b,bool enable_reuse)820 static void buffers_reset(struct buffers *b, bool enable_reuse)
821 {
822 	b->bufmgr = drm_intel_bufmgr_gem_init(fd, 4096);
823 	igt_assert(b->bufmgr);
824 
825 	if (enable_reuse)
826 		drm_intel_bufmgr_gem_enable_reuse(b->bufmgr);
827 	b->batch = intel_batchbuffer_alloc(b->bufmgr, devid);
828 	igt_assert(b->batch);
829 }
830 
buffers_fini(struct buffers * b)831 static void buffers_fini(struct buffers *b)
832 {
833 	if (b->bufmgr == NULL)
834 		return;
835 
836 	buffers_destroy(b);
837 
838 	free(b->tmp);
839 	free(b->src);
840 
841 	intel_batchbuffer_free(b->batch);
842 	drm_intel_bufmgr_destroy(b->bufmgr);
843 
844 	memset(b, 0, sizeof(*b));
845 }
846 
847 typedef void (*do_copy)(struct buffers *b, drm_intel_bo *dst, drm_intel_bo *src);
848 typedef igt_hang_t (*do_hang)(void);
849 
render_copy_bo(struct buffers * b,drm_intel_bo * dst,drm_intel_bo * src)850 static void render_copy_bo(struct buffers *b, drm_intel_bo *dst, drm_intel_bo *src)
851 {
852 	struct igt_buf d = {
853 		.bo = dst,
854 		.size = b->npixels * 4,
855 		.num_tiles = b->npixels * 4,
856 		.stride = b->width * 4,
857 		.bpp = 32,
858 	}, s = {
859 		.bo = src,
860 		.size = b->npixels * 4,
861 		.num_tiles = b->npixels * 4,
862 		.stride = b->width * 4,
863 		.bpp = 32,
864 	};
865 	uint32_t swizzle;
866 
867 	drm_intel_bo_get_tiling(dst, &d.tiling, &swizzle);
868 	drm_intel_bo_get_tiling(src, &s.tiling, &swizzle);
869 
870 	rendercopy(b->batch, NULL,
871 		   &s, 0, 0,
872 		   b->width, b->height,
873 		   &d, 0, 0);
874 }
875 
blt_copy_bo(struct buffers * b,drm_intel_bo * dst,drm_intel_bo * src)876 static void blt_copy_bo(struct buffers *b, drm_intel_bo *dst, drm_intel_bo *src)
877 {
878 	intel_blt_copy(b->batch,
879 		       src, 0, 0, 4*b->width,
880 		       dst, 0, 0, 4*b->width,
881 		       b->width, b->height, 32);
882 }
883 
cpu_copy_bo(struct buffers * b,drm_intel_bo * dst,drm_intel_bo * src)884 static void cpu_copy_bo(struct buffers *b, drm_intel_bo *dst, drm_intel_bo *src)
885 {
886 	const int size = b->page_size;
887 	void *d, *s;
888 
889 	gem_set_domain(fd, src->handle, I915_GEM_DOMAIN_CPU, 0);
890 	gem_set_domain(fd, dst->handle, I915_GEM_DOMAIN_CPU, I915_GEM_DOMAIN_CPU);
891 	s = gem_mmap__cpu(fd, src->handle, 0, size, PROT_READ);
892 	d = gem_mmap__cpu(fd, dst->handle, 0, size, PROT_WRITE);
893 
894 	memcpy(d, s, size);
895 
896 	munmap(d, size);
897 	munmap(s, size);
898 }
899 
gtt_copy_bo(struct buffers * b,drm_intel_bo * dst,drm_intel_bo * src)900 static void gtt_copy_bo(struct buffers *b, drm_intel_bo *dst, drm_intel_bo *src)
901 {
902 	const int size = b->page_size;
903 	void *d, *s;
904 
905 	gem_set_domain(fd, src->handle, I915_GEM_DOMAIN_GTT, 0);
906 	gem_set_domain(fd, dst->handle, I915_GEM_DOMAIN_GTT, I915_GEM_DOMAIN_GTT);
907 
908 	s = gem_mmap__gtt(fd, src->handle, size, PROT_READ);
909 	d = gem_mmap__gtt(fd, dst->handle, size, PROT_WRITE);
910 
911 	memcpy(d, s, size);
912 
913 	munmap(d, size);
914 	munmap(s, size);
915 }
916 
wc_copy_bo(struct buffers * b,drm_intel_bo * dst,drm_intel_bo * src)917 static void wc_copy_bo(struct buffers *b, drm_intel_bo *dst, drm_intel_bo *src)
918 {
919 	const int size = b->page_size;
920 	void *d, *s;
921 
922 	gem_set_domain(fd, src->handle, I915_GEM_DOMAIN_WC, 0);
923 	gem_set_domain(fd, dst->handle, I915_GEM_DOMAIN_WC, I915_GEM_DOMAIN_WC);
924 
925 	s = gem_mmap__wc(fd, src->handle, 0, size, PROT_READ);
926 	d = gem_mmap__wc(fd, dst->handle, 0, size, PROT_WRITE);
927 
928 	memcpy(d, s, size);
929 
930 	munmap(d, size);
931 	munmap(s, size);
932 }
933 
no_hang(void)934 static igt_hang_t no_hang(void)
935 {
936 	return (igt_hang_t){0, 0};
937 }
938 
bcs_hang(void)939 static igt_hang_t bcs_hang(void)
940 {
941 	return igt_hang_ring(fd, I915_EXEC_BLT);
942 }
943 
rcs_hang(void)944 static igt_hang_t rcs_hang(void)
945 {
946 	return igt_hang_ring(fd, I915_EXEC_RENDER);
947 }
948 
all_hang(void)949 static igt_hang_t all_hang(void)
950 {
951 	igt_hang_t hang = igt_hang_ring(fd, I915_EXEC_RENDER);
952 	unsigned engine;
953 
954 	for_each_physical_engine(fd, engine) {
955 		struct drm_i915_gem_execbuffer2 eb = hang.spin->execbuf;
956 
957 		if (engine == I915_EXEC_RENDER)
958 			continue;
959 
960 		eb.flags = engine;
961 		__gem_execbuf(fd, &eb);
962 	}
963 
964 	return hang;
965 }
966 
do_basic0(struct buffers * buffers,do_copy do_copy_func,do_hang do_hang_func)967 static void do_basic0(struct buffers *buffers,
968 		      do_copy do_copy_func,
969 		      do_hang do_hang_func)
970 {
971 	buffers->mode->set_bo(buffers, buffers->src[0], 0xdeadbeef);
972 	for (int i = 0; i < buffers->count; i++) {
973 		igt_hang_t hang = do_hang_func();
974 
975 		do_copy_func(buffers, buffers->dst[i], buffers->src[0]);
976 		buffers->mode->cmp_bo(buffers, buffers->dst[i], 0xdeadbeef);
977 
978 		igt_post_hang_ring(fd, hang);
979 	}
980 }
981 
do_basic1(struct buffers * buffers,do_copy do_copy_func,do_hang do_hang_func)982 static void do_basic1(struct buffers *buffers,
983 		      do_copy do_copy_func,
984 		      do_hang do_hang_func)
985 {
986 	for (int i = 0; i < buffers->count; i++) {
987 		igt_hang_t hang = do_hang_func();
988 
989 		buffers->mode->set_bo(buffers, buffers->src[i], i);
990 		buffers->mode->set_bo(buffers, buffers->dst[i], ~i);
991 
992 		do_copy_func(buffers, buffers->dst[i], buffers->src[i]);
993 		usleep(0); /* let someone else claim the mutex */
994 		buffers->mode->cmp_bo(buffers, buffers->dst[i], i);
995 
996 		igt_post_hang_ring(fd, hang);
997 	}
998 }
999 
do_basicN(struct buffers * buffers,do_copy do_copy_func,do_hang do_hang_func)1000 static void do_basicN(struct buffers *buffers,
1001 		      do_copy do_copy_func,
1002 		      do_hang do_hang_func)
1003 {
1004 	igt_hang_t hang;
1005 
1006 	for (int i = 0; i < buffers->count; i++) {
1007 		buffers->mode->set_bo(buffers, buffers->src[i], i);
1008 		buffers->mode->set_bo(buffers, buffers->dst[i], ~i);
1009 	}
1010 
1011 	hang = do_hang_func();
1012 
1013 	for (int i = 0; i < buffers->count; i++) {
1014 		do_copy_func(buffers, buffers->dst[i], buffers->src[i]);
1015 		usleep(0); /* let someone else claim the mutex */
1016 	}
1017 
1018 	for (int i = 0; i < buffers->count; i++)
1019 		buffers->mode->cmp_bo(buffers, buffers->dst[i], i);
1020 
1021 	igt_post_hang_ring(fd, hang);
1022 }
1023 
do_overwrite_source(struct buffers * buffers,do_copy do_copy_func,do_hang do_hang_func)1024 static void do_overwrite_source(struct buffers *buffers,
1025 				do_copy do_copy_func,
1026 				do_hang do_hang_func)
1027 {
1028 	igt_hang_t hang;
1029 	int i;
1030 
1031 	for (i = 0; i < buffers->count; i++) {
1032 		buffers->mode->set_bo(buffers, buffers->src[i], i);
1033 		buffers->mode->set_bo(buffers, buffers->dst[i], ~i);
1034 	}
1035 	for (i = 0; i < buffers->count; i++)
1036 		do_copy_func(buffers, buffers->dst[i], buffers->src[i]);
1037 	hang = do_hang_func();
1038 	for (i = buffers->count; i--; )
1039 		buffers->mode->set_bo(buffers, buffers->src[i], 0xdeadbeef);
1040 	for (i = 0; i < buffers->count; i++)
1041 		buffers->mode->cmp_bo(buffers, buffers->dst[i], i);
1042 	igt_post_hang_ring(fd, hang);
1043 }
1044 
do_overwrite_source_read(struct buffers * buffers,do_copy do_copy_func,do_hang do_hang_func,int do_rcs)1045 static void do_overwrite_source_read(struct buffers *buffers,
1046 				     do_copy do_copy_func,
1047 				     do_hang do_hang_func,
1048 				     int do_rcs)
1049 {
1050 	const int half = buffers->count/2;
1051 	igt_hang_t hang;
1052 	int i;
1053 
1054 	for (i = 0; i < half; i++) {
1055 		buffers->mode->set_bo(buffers, buffers->src[i], i);
1056 		buffers->mode->set_bo(buffers, buffers->dst[i], ~i);
1057 		buffers->mode->set_bo(buffers, buffers->dst[i+half], ~i);
1058 	}
1059 	for (i = 0; i < half; i++) {
1060 		do_copy_func(buffers, buffers->dst[i], buffers->src[i]);
1061 		if (do_rcs)
1062 			render_copy_bo(buffers, buffers->dst[i+half], buffers->src[i]);
1063 		else
1064 			blt_copy_bo(buffers, buffers->dst[i+half], buffers->src[i]);
1065 	}
1066 	hang = do_hang_func();
1067 	for (i = half; i--; )
1068 		buffers->mode->set_bo(buffers, buffers->src[i], 0xdeadbeef);
1069 	for (i = 0; i < half; i++) {
1070 		buffers->mode->cmp_bo(buffers, buffers->dst[i], i);
1071 		buffers->mode->cmp_bo(buffers, buffers->dst[i+half], i);
1072 	}
1073 	igt_post_hang_ring(fd, hang);
1074 }
1075 
do_overwrite_source_read_bcs(struct buffers * buffers,do_copy do_copy_func,do_hang do_hang_func)1076 static void do_overwrite_source_read_bcs(struct buffers *buffers,
1077 					 do_copy do_copy_func,
1078 					 do_hang do_hang_func)
1079 {
1080 	do_overwrite_source_read(buffers, do_copy_func, do_hang_func, 0);
1081 }
1082 
do_overwrite_source_read_rcs(struct buffers * buffers,do_copy do_copy_func,do_hang do_hang_func)1083 static void do_overwrite_source_read_rcs(struct buffers *buffers,
1084 					 do_copy do_copy_func,
1085 					 do_hang do_hang_func)
1086 {
1087 	do_overwrite_source_read(buffers, do_copy_func, do_hang_func, 1);
1088 }
1089 
do_overwrite_source__rev(struct buffers * buffers,do_copy do_copy_func,do_hang do_hang_func)1090 static void do_overwrite_source__rev(struct buffers *buffers,
1091 				     do_copy do_copy_func,
1092 				     do_hang do_hang_func)
1093 {
1094 	igt_hang_t hang;
1095 	int i;
1096 
1097 	for (i = 0; i < buffers->count; i++) {
1098 		buffers->mode->set_bo(buffers, buffers->src[i], i);
1099 		buffers->mode->set_bo(buffers, buffers->dst[i], ~i);
1100 	}
1101 	for (i = 0; i < buffers->count; i++)
1102 		do_copy_func(buffers, buffers->dst[i], buffers->src[i]);
1103 	hang = do_hang_func();
1104 	for (i = 0; i < buffers->count; i++)
1105 		buffers->mode->set_bo(buffers, buffers->src[i], 0xdeadbeef);
1106 	for (i = buffers->count; i--; )
1107 		buffers->mode->cmp_bo(buffers, buffers->dst[i], i);
1108 	igt_post_hang_ring(fd, hang);
1109 }
1110 
do_overwrite_source__one(struct buffers * buffers,do_copy do_copy_func,do_hang do_hang_func)1111 static void do_overwrite_source__one(struct buffers *buffers,
1112 				     do_copy do_copy_func,
1113 				     do_hang do_hang_func)
1114 {
1115 	igt_hang_t hang;
1116 
1117 	buffers->mode->set_bo(buffers, buffers->src[0], 0);
1118 	buffers->mode->set_bo(buffers, buffers->dst[0], ~0);
1119 	do_copy_func(buffers, buffers->dst[0], buffers->src[0]);
1120 	hang = do_hang_func();
1121 	buffers->mode->set_bo(buffers, buffers->src[0], 0xdeadbeef);
1122 	buffers->mode->cmp_bo(buffers, buffers->dst[0], 0);
1123 	igt_post_hang_ring(fd, hang);
1124 }
1125 
do_intermix(struct buffers * buffers,do_copy do_copy_func,do_hang do_hang_func,int do_rcs)1126 static void do_intermix(struct buffers *buffers,
1127 			do_copy do_copy_func,
1128 			do_hang do_hang_func,
1129 			int do_rcs)
1130 {
1131 	const int half = buffers->count/2;
1132 	igt_hang_t hang;
1133 	int i;
1134 
1135 	for (i = 0; i < buffers->count; i++) {
1136 		buffers->mode->set_bo(buffers, buffers->src[i], 0xdeadbeef^~i);
1137 		buffers->mode->set_bo(buffers, buffers->dst[i], i);
1138 	}
1139 	for (i = 0; i < half; i++) {
1140 		if (do_rcs == 1 || (do_rcs == -1 && i & 1))
1141 			render_copy_bo(buffers, buffers->dst[i], buffers->src[i]);
1142 		else
1143 			blt_copy_bo(buffers, buffers->dst[i], buffers->src[i]);
1144 
1145 		do_copy_func(buffers, buffers->dst[i+half], buffers->src[i]);
1146 
1147 		if (do_rcs == 1 || (do_rcs == -1 && (i & 1) == 0))
1148 			render_copy_bo(buffers, buffers->dst[i], buffers->dst[i+half]);
1149 		else
1150 			blt_copy_bo(buffers, buffers->dst[i], buffers->dst[i+half]);
1151 
1152 		do_copy_func(buffers, buffers->dst[i+half], buffers->src[i+half]);
1153 	}
1154 	hang = do_hang_func();
1155 	for (i = 0; i < 2*half; i++)
1156 		buffers->mode->cmp_bo(buffers, buffers->dst[i], 0xdeadbeef^~i);
1157 	igt_post_hang_ring(fd, hang);
1158 }
1159 
do_intermix_rcs(struct buffers * buffers,do_copy do_copy_func,do_hang do_hang_func)1160 static void do_intermix_rcs(struct buffers *buffers,
1161 			    do_copy do_copy_func,
1162 			    do_hang do_hang_func)
1163 {
1164 	do_intermix(buffers, do_copy_func, do_hang_func, 1);
1165 }
1166 
do_intermix_bcs(struct buffers * buffers,do_copy do_copy_func,do_hang do_hang_func)1167 static void do_intermix_bcs(struct buffers *buffers,
1168 			    do_copy do_copy_func,
1169 			    do_hang do_hang_func)
1170 {
1171 	do_intermix(buffers, do_copy_func, do_hang_func, 0);
1172 }
1173 
do_intermix_both(struct buffers * buffers,do_copy do_copy_func,do_hang do_hang_func)1174 static void do_intermix_both(struct buffers *buffers,
1175 			     do_copy do_copy_func,
1176 			     do_hang do_hang_func)
1177 {
1178 	do_intermix(buffers, do_copy_func, do_hang_func, -1);
1179 }
1180 
do_early_read(struct buffers * buffers,do_copy do_copy_func,do_hang do_hang_func)1181 static void do_early_read(struct buffers *buffers,
1182 			  do_copy do_copy_func,
1183 			  do_hang do_hang_func)
1184 {
1185 	igt_hang_t hang;
1186 	int i;
1187 
1188 	for (i = buffers->count; i--; )
1189 		buffers->mode->set_bo(buffers, buffers->src[i], 0xdeadbeef);
1190 	for (i = 0; i < buffers->count; i++)
1191 		do_copy_func(buffers, buffers->dst[i], buffers->src[i]);
1192 	hang = do_hang_func();
1193 	for (i = buffers->count; i--; )
1194 		buffers->mode->cmp_bo(buffers, buffers->dst[i], 0xdeadbeef);
1195 	igt_post_hang_ring(fd, hang);
1196 }
1197 
do_read_read_bcs(struct buffers * buffers,do_copy do_copy_func,do_hang do_hang_func)1198 static void do_read_read_bcs(struct buffers *buffers,
1199 			     do_copy do_copy_func,
1200 			     do_hang do_hang_func)
1201 {
1202 	igt_hang_t hang;
1203 	int i;
1204 
1205 	for (i = buffers->count; i--; )
1206 		buffers->mode->set_bo(buffers, buffers->src[i], 0xdeadbeef ^ i);
1207 	for (i = 0; i < buffers->count; i++) {
1208 		do_copy_func(buffers, buffers->dst[i], buffers->src[i]);
1209 		blt_copy_bo(buffers, buffers->spare, buffers->src[i]);
1210 	}
1211 	buffers->mode->cmp_bo(buffers, buffers->spare, 0xdeadbeef^(buffers->count-1));
1212 	hang = do_hang_func();
1213 	for (i = buffers->count; i--; )
1214 		buffers->mode->cmp_bo(buffers, buffers->dst[i], 0xdeadbeef ^ i);
1215 	igt_post_hang_ring(fd, hang);
1216 }
1217 
do_write_read_bcs(struct buffers * buffers,do_copy do_copy_func,do_hang do_hang_func)1218 static void do_write_read_bcs(struct buffers *buffers,
1219 			      do_copy do_copy_func,
1220 			      do_hang do_hang_func)
1221 {
1222 	igt_hang_t hang;
1223 	int i;
1224 
1225 	for (i = buffers->count; i--; )
1226 		buffers->mode->set_bo(buffers, buffers->src[i], 0xdeadbeef ^ i);
1227 	for (i = 0; i < buffers->count; i++) {
1228 		blt_copy_bo(buffers, buffers->spare, buffers->src[i]);
1229 		do_copy_func(buffers, buffers->dst[i], buffers->spare);
1230 	}
1231 	hang = do_hang_func();
1232 	for (i = buffers->count; i--; )
1233 		buffers->mode->cmp_bo(buffers, buffers->dst[i], 0xdeadbeef ^ i);
1234 	igt_post_hang_ring(fd, hang);
1235 }
1236 
do_read_read_rcs(struct buffers * buffers,do_copy do_copy_func,do_hang do_hang_func)1237 static void do_read_read_rcs(struct buffers *buffers,
1238 			     do_copy do_copy_func,
1239 			     do_hang do_hang_func)
1240 {
1241 	igt_hang_t hang;
1242 	int i;
1243 
1244 	for (i = buffers->count; i--; )
1245 		buffers->mode->set_bo(buffers, buffers->src[i], 0xdeadbeef ^ i);
1246 	for (i = 0; i < buffers->count; i++) {
1247 		do_copy_func(buffers, buffers->dst[i], buffers->src[i]);
1248 		render_copy_bo(buffers, buffers->spare, buffers->src[i]);
1249 	}
1250 	buffers->mode->cmp_bo(buffers, buffers->spare, 0xdeadbeef^(buffers->count-1));
1251 	hang = do_hang_func();
1252 	for (i = buffers->count; i--; )
1253 		buffers->mode->cmp_bo(buffers, buffers->dst[i], 0xdeadbeef ^ i);
1254 	igt_post_hang_ring(fd, hang);
1255 }
1256 
do_write_read_rcs(struct buffers * buffers,do_copy do_copy_func,do_hang do_hang_func)1257 static void do_write_read_rcs(struct buffers *buffers,
1258 			      do_copy do_copy_func,
1259 			      do_hang do_hang_func)
1260 {
1261 	igt_hang_t hang;
1262 	int i;
1263 
1264 	for (i = buffers->count; i--; )
1265 		buffers->mode->set_bo(buffers, buffers->src[i], 0xdeadbeef ^ i);
1266 	for (i = 0; i < buffers->count; i++) {
1267 		render_copy_bo(buffers, buffers->spare, buffers->src[i]);
1268 		do_copy_func(buffers, buffers->dst[i], buffers->spare);
1269 	}
1270 	hang = do_hang_func();
1271 	for (i = buffers->count; i--; )
1272 		buffers->mode->cmp_bo(buffers, buffers->dst[i], 0xdeadbeef ^ i);
1273 	igt_post_hang_ring(fd, hang);
1274 }
1275 
do_gpu_read_after_write(struct buffers * buffers,do_copy do_copy_func,do_hang do_hang_func)1276 static void do_gpu_read_after_write(struct buffers *buffers,
1277 				    do_copy do_copy_func,
1278 				    do_hang do_hang_func)
1279 {
1280 	igt_hang_t hang;
1281 	int i;
1282 
1283 	for (i = buffers->count; i--; )
1284 		buffers->mode->set_bo(buffers, buffers->src[i], 0xabcdabcd);
1285 	for (i = 0; i < buffers->count; i++)
1286 		do_copy_func(buffers, buffers->dst[i], buffers->src[i]);
1287 	for (i = buffers->count; i--; )
1288 		do_copy_func(buffers, buffers->spare, buffers->dst[i]);
1289 	hang = do_hang_func();
1290 	for (i = buffers->count; i--; )
1291 		buffers->mode->cmp_bo(buffers, buffers->dst[i], 0xabcdabcd);
1292 	igt_post_hang_ring(fd, hang);
1293 }
1294 
1295 typedef void (*do_test)(struct buffers *buffers,
1296 			do_copy do_copy_func,
1297 			do_hang do_hang_func);
1298 
1299 typedef void (*run_wrap)(struct buffers *buffers,
1300 			 do_test do_test_func,
1301 			 do_copy do_copy_func,
1302 			 do_hang do_hang_func);
1303 
run_single(struct buffers * buffers,do_test do_test_func,do_copy do_copy_func,do_hang do_hang_func)1304 static void run_single(struct buffers *buffers,
1305 		       do_test do_test_func,
1306 		       do_copy do_copy_func,
1307 		       do_hang do_hang_func)
1308 {
1309 	pass = 0;
1310 	do_test_func(buffers, do_copy_func, do_hang_func);
1311 	igt_assert_eq(intel_detect_and_clear_missed_interrupts(fd), 0);
1312 }
1313 
run_interruptible(struct buffers * buffers,do_test do_test_func,do_copy do_copy_func,do_hang do_hang_func)1314 static void run_interruptible(struct buffers *buffers,
1315 			      do_test do_test_func,
1316 			      do_copy do_copy_func,
1317 			      do_hang do_hang_func)
1318 {
1319 	pass = 0;
1320 	igt_while_interruptible(true)
1321 		do_test_func(buffers, do_copy_func, do_hang_func);
1322 	igt_assert_eq(intel_detect_and_clear_missed_interrupts(fd), 0);
1323 }
1324 
run_child(struct buffers * buffers,do_test do_test_func,do_copy do_copy_func,do_hang do_hang_func)1325 static void run_child(struct buffers *buffers,
1326 		      do_test do_test_func,
1327 		      do_copy do_copy_func,
1328 		      do_hang do_hang_func)
1329 
1330 {
1331 	/* We inherit the buffers from the parent, but the bufmgr/batch
1332 	 * needs to be local as the cache of reusable itself will be COWed,
1333 	 * leading to the child closing an object without the parent knowing.
1334 	 */
1335 	pass = 0;
1336 	igt_fork(child, 1)
1337 		do_test_func(buffers, do_copy_func, do_hang_func);
1338 	igt_waitchildren();
1339 	igt_assert_eq(intel_detect_and_clear_missed_interrupts(fd), 0);
1340 }
1341 
__run_forked(struct buffers * buffers,int num_children,int loops,bool interrupt,do_test do_test_func,do_copy do_copy_func,do_hang do_hang_func)1342 static void __run_forked(struct buffers *buffers,
1343 			 int num_children, int loops, bool interrupt,
1344 			 do_test do_test_func,
1345 			 do_copy do_copy_func,
1346 			 do_hang do_hang_func)
1347 
1348 {
1349 	/* purge the libdrm caches before cloing the process */
1350 	buffers_destroy(buffers);
1351 	intel_batchbuffer_free(buffers->batch);
1352 	drm_intel_bufmgr_destroy(buffers->bufmgr);
1353 
1354 	igt_fork(child, num_children) {
1355 		int num_buffers;
1356 
1357 		/* recreate process local variables */
1358 		fd = drm_open_driver(DRIVER_INTEL);
1359 
1360 		num_buffers = buffers->num_buffers / num_children;
1361 		num_buffers += MIN_BUFFERS;
1362 		if (num_buffers < buffers->num_buffers)
1363 			buffers->num_buffers = num_buffers;
1364 
1365 		buffers_reset(buffers, true);
1366 		buffers_create(buffers);
1367 
1368 		igt_while_interruptible(interrupt) {
1369 			for (pass = 0; pass < loops; pass++)
1370 				do_test_func(buffers,
1371 					     do_copy_func,
1372 					     do_hang_func);
1373 		}
1374 	}
1375 	igt_waitchildren();
1376 	igt_assert_eq(intel_detect_and_clear_missed_interrupts(fd), 0);
1377 
1378 	buffers_reset(buffers, true);
1379 }
1380 
run_forked(struct buffers * buffers,do_test do_test_func,do_copy do_copy_func,do_hang do_hang_func)1381 static void run_forked(struct buffers *buffers,
1382 		       do_test do_test_func,
1383 		       do_copy do_copy_func,
1384 		       do_hang do_hang_func)
1385 {
1386 	const int ncpus = sysconf(_SC_NPROCESSORS_ONLN);
1387 	__run_forked(buffers, ncpus, ncpus, false,
1388 		     do_test_func, do_copy_func, do_hang_func);
1389 }
1390 
run_bomb(struct buffers * buffers,do_test do_test_func,do_copy do_copy_func,do_hang do_hang_func)1391 static void run_bomb(struct buffers *buffers,
1392 		     do_test do_test_func,
1393 		     do_copy do_copy_func,
1394 		     do_hang do_hang_func)
1395 {
1396 	const int ncpus = sysconf(_SC_NPROCESSORS_ONLN);
1397 	__run_forked(buffers, 8*ncpus, 2, true,
1398 		     do_test_func, do_copy_func, do_hang_func);
1399 }
1400 
cpu_require(void)1401 static void cpu_require(void)
1402 {
1403 	bit17_require();
1404 }
1405 
gtt_require(void)1406 static void gtt_require(void)
1407 {
1408 }
1409 
bcs_require(void)1410 static void bcs_require(void)
1411 {
1412 }
1413 
rcs_require(void)1414 static void rcs_require(void)
1415 {
1416 	igt_require(rendercopy);
1417 }
1418 
1419 static void
run_mode(const char * prefix,const struct create * create,const struct access_mode * mode,const struct size * size,const int num_buffers,const char * suffix,run_wrap run_wrap_func)1420 run_mode(const char *prefix,
1421 	 const struct create *create,
1422 	 const struct access_mode *mode,
1423 	 const struct size *size,
1424 	 const int num_buffers,
1425 	 const char *suffix,
1426 	 run_wrap run_wrap_func)
1427 {
1428 	const struct {
1429 		const char *prefix;
1430 		do_copy copy;
1431 		void (*require)(void);
1432 	} pipelines[] = {
1433 		{ "cpu", cpu_copy_bo, cpu_require },
1434 		{ "gtt", gtt_copy_bo, gtt_require },
1435 		{ "wc", wc_copy_bo, wc_require },
1436 		{ "blt", blt_copy_bo, bcs_require },
1437 		{ "render", render_copy_bo, rcs_require },
1438 		{ NULL, NULL }
1439 	}, *pskip = pipelines + 3, *p;
1440 	const struct {
1441 		const char *suffix;
1442 		do_hang hang;
1443 	} hangs[] = {
1444 		{ "", no_hang },
1445 		{ "-hang-blt", bcs_hang },
1446 		{ "-hang-render", rcs_hang },
1447 		{ "-hang-all", all_hang },
1448 		{ NULL, NULL },
1449 	}, *h;
1450 	struct buffers buffers;
1451 
1452 	igt_fixture
1453 		buffers_init(&buffers, prefix, create, mode,
1454 			     size, num_buffers,
1455 			     fd, run_wrap_func != run_child);
1456 
1457 	for (h = hangs; h->suffix; h++) {
1458 		if (!all && *h->suffix)
1459 			continue;
1460 
1461 		if (!*h->suffix)
1462 			igt_fixture
1463 				igt_fork_hang_detector(fd);
1464 
1465 		for (p = all ? pipelines : pskip; p->prefix; p++) {
1466 			igt_subtest_group  {
1467 				igt_fixture p->require();
1468 
1469 				igt_subtest_f("%s-%s-%s-sanitycheck0%s%s", prefix, mode->name, p->prefix, suffix, h->suffix) {
1470 					buffers_create(&buffers);
1471 					run_wrap_func(&buffers, do_basic0,
1472 							p->copy, h->hang);
1473 				}
1474 
1475 				igt_subtest_f("%s-%s-%s-sanitycheck1%s%s", prefix, mode->name, p->prefix, suffix, h->suffix) {
1476 					buffers_create(&buffers);
1477 					run_wrap_func(&buffers, do_basic1,
1478 							p->copy, h->hang);
1479 				}
1480 
1481 				igt_subtest_f("%s-%s-%s-sanitycheckN%s%s", prefix, mode->name, p->prefix, suffix, h->suffix) {
1482 					buffers_create(&buffers);
1483 					run_wrap_func(&buffers, do_basicN,
1484 							p->copy, h->hang);
1485 				}
1486 
1487 				/* try to overwrite the source values */
1488 				igt_subtest_f("%s-%s-%s-overwrite-source-one%s%s", prefix, mode->name, p->prefix, suffix, h->suffix) {
1489 					buffers_create(&buffers);
1490 					run_wrap_func(&buffers,
1491 							do_overwrite_source__one,
1492 							p->copy, h->hang);
1493 				}
1494 
1495 				igt_subtest_f("%s-%s-%s-overwrite-source%s%s", prefix, mode->name, p->prefix, suffix, h->suffix) {
1496 					buffers_create(&buffers);
1497 					run_wrap_func(&buffers,
1498 							do_overwrite_source,
1499 							p->copy, h->hang);
1500 				}
1501 
1502 				igt_subtest_f("%s-%s-%s-overwrite-source-read-bcs%s%s", prefix, mode->name, p->prefix, suffix, h->suffix) {
1503 					buffers_create(&buffers);
1504 					run_wrap_func(&buffers,
1505 							do_overwrite_source_read_bcs,
1506 							p->copy, h->hang);
1507 				}
1508 
1509 				igt_subtest_f("%s-%s-%s-overwrite-source-read-rcs%s%s", prefix, mode->name, p->prefix, suffix, h->suffix) {
1510 					igt_require(rendercopy);
1511 					buffers_create(&buffers);
1512 					run_wrap_func(&buffers,
1513 							do_overwrite_source_read_rcs,
1514 							p->copy, h->hang);
1515 				}
1516 
1517 				igt_subtest_f("%s-%s-%s-overwrite-source-rev%s%s", prefix, mode->name, p->prefix, suffix, h->suffix) {
1518 					buffers_create(&buffers);
1519 					run_wrap_func(&buffers,
1520 							do_overwrite_source__rev,
1521 							p->copy, h->hang);
1522 				}
1523 
1524 				/* try to intermix copies with GPU copies*/
1525 				igt_subtest_f("%s-%s-%s-intermix-rcs%s%s", prefix, mode->name, p->prefix, suffix, h->suffix) {
1526 					igt_require(rendercopy);
1527 					buffers_create(&buffers);
1528 					run_wrap_func(&buffers,
1529 							do_intermix_rcs,
1530 							p->copy, h->hang);
1531 				}
1532 				igt_subtest_f("%s-%s-%s-intermix-bcs%s%s", prefix, mode->name, p->prefix, suffix, h->suffix) {
1533 					igt_require(rendercopy);
1534 					buffers_create(&buffers);
1535 					run_wrap_func(&buffers,
1536 							do_intermix_bcs,
1537 							p->copy, h->hang);
1538 				}
1539 				igt_subtest_f("%s-%s-%s-intermix-both%s%s", prefix, mode->name, p->prefix, suffix, h->suffix) {
1540 					igt_require(rendercopy);
1541 					buffers_create(&buffers);
1542 					run_wrap_func(&buffers,
1543 							do_intermix_both,
1544 							p->copy, h->hang);
1545 				}
1546 
1547 				/* try to read the results before the copy completes */
1548 				igt_subtest_f("%s-%s-%s-early-read%s%s", prefix, mode->name, p->prefix, suffix, h->suffix) {
1549 					buffers_create(&buffers);
1550 					run_wrap_func(&buffers,
1551 							do_early_read,
1552 							p->copy, h->hang);
1553 				}
1554 
1555 				/* concurrent reads */
1556 				igt_subtest_f("%s-%s-%s-read-read-bcs%s%s", prefix, mode->name, p->prefix, suffix, h->suffix) {
1557 					buffers_create(&buffers);
1558 					run_wrap_func(&buffers,
1559 							do_read_read_bcs,
1560 							p->copy, h->hang);
1561 				}
1562 				igt_subtest_f("%s-%s-%s-read-read-rcs%s%s", prefix, mode->name, p->prefix, suffix, h->suffix) {
1563 					igt_require(rendercopy);
1564 					buffers_create(&buffers);
1565 					run_wrap_func(&buffers,
1566 							do_read_read_rcs,
1567 							p->copy, h->hang);
1568 				}
1569 
1570 				/* split copying between rings */
1571 				igt_subtest_f("%s-%s-%s-write-read-bcs%s%s", prefix, mode->name, p->prefix, suffix, h->suffix) {
1572 					buffers_create(&buffers);
1573 					run_wrap_func(&buffers,
1574 							do_write_read_bcs,
1575 							p->copy, h->hang);
1576 				}
1577 				igt_subtest_f("%s-%s-%s-write-read-rcs%s%s", prefix, mode->name, p->prefix, suffix, h->suffix) {
1578 					igt_require(rendercopy);
1579 					buffers_create(&buffers);
1580 					run_wrap_func(&buffers,
1581 							do_write_read_rcs,
1582 							p->copy, h->hang);
1583 				}
1584 
1585 				/* and finally try to trick the kernel into loosing the pending write */
1586 				igt_subtest_f("%s-%s-%s-gpu-read-after-write%s%s", prefix, mode->name, p->prefix, suffix, h->suffix) {
1587 					buffers_create(&buffers);
1588 					run_wrap_func(&buffers,
1589 							do_gpu_read_after_write,
1590 							p->copy, h->hang);
1591 				}
1592 			}
1593 		}
1594 
1595 		if (!*h->suffix)
1596 			igt_fixture
1597 				igt_stop_hang_detector();
1598 	}
1599 
1600 	igt_fixture
1601 		buffers_fini(&buffers);
1602 }
1603 
1604 static void
run_modes(const char * style,const struct create * create,const struct access_mode * mode,const struct size * size,const int num)1605 run_modes(const char *style,
1606 	  const struct create *create,
1607 	  const struct access_mode *mode,
1608 	  const struct size *size,
1609 	  const int num)
1610 {
1611 	const struct wrap {
1612 		const char *suffix;
1613 		run_wrap func;
1614 	} wrappers[] = {
1615 		{ "", run_single },
1616 		{ "-child", run_child },
1617 		{ "-forked", run_forked },
1618 		{ "-interruptible", run_interruptible },
1619 		{ "-bomb", run_bomb },
1620 		{ NULL },
1621 	};
1622 
1623 	while (mode->name) {
1624 		igt_subtest_group {
1625 			igt_fixture {
1626 				if (mode->require)
1627 					mode->require(create, num);
1628 			}
1629 
1630 			for (const struct wrap *w = wrappers; w->suffix; w++) {
1631 				run_mode(style, create, mode, size, num,
1632 					 w->suffix, w->func);
1633 			}
1634 		}
1635 
1636 		mode++;
1637 	}
1638 }
1639 
1640 static unsigned
num_buffers(uint64_t max,const struct size * s,const struct create * c,unsigned allow_mem)1641 num_buffers(uint64_t max,
1642 	    const struct size *s,
1643 	    const struct create *c,
1644 	    unsigned allow_mem)
1645 {
1646 	unsigned size = 4*s->width*s->height;
1647 	uint64_t n;
1648 
1649 	igt_assert(size);
1650 	n = max / (2*size);
1651 	n += MIN_BUFFERS;
1652 
1653 	igt_require(n < INT32_MAX);
1654 	igt_require(set_max_map_count(2*n));
1655 
1656 	if (c->require)
1657 		c->require(c, n);
1658 
1659 	intel_require_memory(2*n, size, allow_mem);
1660 
1661 	return n;
1662 }
1663 
1664 igt_main
1665 {
1666 	const struct access_mode modes[] = {
1667 		{
1668 			.name = "prw",
1669 			.create_bo = unmapped_create_bo,
1670 			.set_bo = prw_set_bo,
1671 			.cmp_bo = prw_cmp_bo,
1672 			.release_bo = nop_release_bo,
1673 		},
1674 		{
1675 			.name = "partial",
1676 			.create_bo = unmapped_create_bo,
1677 			.set_bo = partial_set_bo,
1678 			.cmp_bo = partial_cmp_bo,
1679 			.release_bo = nop_release_bo,
1680 		},
1681 		{
1682 			.name = "cpu",
1683 			.create_bo = unmapped_create_bo,
1684 			.require = create_cpu_require,
1685 			.set_bo = cpu_set_bo,
1686 			.cmp_bo = cpu_cmp_bo,
1687 			.release_bo = nop_release_bo,
1688 		},
1689 		{
1690 			.name = "snoop",
1691 			.create_bo = snoop_create_bo,
1692 			.require = create_snoop_require,
1693 			.set_bo = cpu_set_bo,
1694 			.cmp_bo = cpu_cmp_bo,
1695 			.release_bo = nop_release_bo,
1696 		},
1697 		{
1698 			.name = "userptr",
1699 			.create_bo = userptr_create_bo,
1700 			.require = create_userptr_require,
1701 			.set_bo = userptr_set_bo,
1702 			.cmp_bo = userptr_cmp_bo,
1703 			.release_bo = userptr_release_bo,
1704 		},
1705 		{
1706 			.name = "dmabuf",
1707 			.create_bo = dmabuf_create_bo,
1708 			.require = create_dmabuf_require,
1709 			.set_bo = dmabuf_set_bo,
1710 			.cmp_bo = dmabuf_cmp_bo,
1711 			.release_bo = dmabuf_release_bo,
1712 		},
1713 		{
1714 			.name = "vgem",
1715 			.create_bo = vgem_create_bo,
1716 			.require = create_vgem_require,
1717 			.set_bo = dmabuf_set_bo,
1718 			.cmp_bo = dmabuf_cmp_bo,
1719 			.release_bo = dmabuf_release_bo,
1720 		},
1721 		{
1722 			.name = "gtt",
1723 			.create_bo = gtt_create_bo,
1724 			.set_bo = gtt_set_bo,
1725 			.cmp_bo = gtt_cmp_bo,
1726 			.release_bo = nop_release_bo,
1727 		},
1728 		{
1729 			.name = "gttX",
1730 			.create_bo = gttX_create_bo,
1731 			.set_bo = gtt_set_bo,
1732 			.cmp_bo = gtt_cmp_bo,
1733 			.release_bo = nop_release_bo,
1734 		},
1735 		{
1736 			.name = "wc",
1737 			.require = wc_create_require,
1738 			.create_bo = wc_create_bo,
1739 			.set_bo = gtt_set_bo,
1740 			.cmp_bo = gtt_cmp_bo,
1741 			.release_bo = wc_release_bo,
1742 		},
1743 		{
1744 			.name = "gpu",
1745 			.create_bo = gpu_create_bo,
1746 			.set_bo = gpu_set_bo,
1747 			.cmp_bo = gpu_cmp_bo,
1748 			.release_bo = nop_release_bo,
1749 		},
1750 		{
1751 			.name = "gpuX",
1752 			.create_bo = gpuX_create_bo,
1753 			.set_bo = gpu_set_bo,
1754 			.cmp_bo = gpu_cmp_bo,
1755 			.release_bo = nop_release_bo,
1756 		},
1757 		{ NULL },
1758 	};
1759 	const struct create create[] = {
1760 		{ "", can_create_normal, create_normal_bo},
1761 #if HAVE_CREATE_PRIVATE
1762 		{ "private-", can_create_private, create_private_bo},
1763 #endif
1764 #if HAVE_CREATE_STOLEN
1765 		{ "stolen-", can_create_stolen, create_stolen_bo},
1766 #endif
1767 		{ NULL, NULL }
1768 	};
1769 	const struct size sizes[] = {
1770 		{ "4KiB", 128, 8 },
1771 		{ "256KiB", 128, 128 },
1772 		{ "1MiB", 512, 512 },
1773 		{ "16MiB", 2048, 2048 },
1774 		{ NULL}
1775 	};
1776 	uint64_t pin_sz = 0;
1777 	void *pinned = NULL;
1778 	char name[80];
1779 	int count = 0;
1780 
1781 	igt_skip_on_simulation();
1782 
1783 	if (strstr(igt_test_name(), "all"))
1784 		all = true;
1785 
1786 	igt_fixture {
1787 		igt_allow_unlimited_files();
1788 
1789 		fd = drm_open_driver(DRIVER_INTEL);
1790 		igt_require_gem(fd);
1791 		intel_detect_and_clear_missed_interrupts(fd);
1792 		devid = intel_get_drm_devid(fd);
1793 		gen = intel_gen(devid);
1794 		rendercopy = igt_get_render_copyfunc(devid);
1795 
1796 		vgem_drv = __drm_open_driver(DRIVER_VGEM);
1797 	}
1798 
1799 	for (const struct create *c = create; c->name; c++) {
1800 		for (const struct size *s = sizes; s->name; s++) {
1801 			/* Minimum test set */
1802 			snprintf(name, sizeof(name), "%s%s-%s",
1803 				 c->name, s->name, "tiny");
1804 			igt_subtest_group {
1805 				igt_fixture {
1806 					count = num_buffers(0, s, c, CHECK_RAM);
1807 				}
1808 				run_modes(name, c, modes, s, count);
1809 			}
1810 
1811 			/* "Average" test set */
1812 			snprintf(name, sizeof(name), "%s%s-%s",
1813 				 c->name, s->name, "small");
1814 			igt_subtest_group {
1815 				igt_fixture {
1816 					count = num_buffers(gem_mappable_aperture_size()/4,
1817 							    s, c, CHECK_RAM);
1818 				}
1819 				run_modes(name, c, modes, s, count);
1820 			}
1821 
1822 			/* Use the entire mappable aperture */
1823 			snprintf(name, sizeof(name), "%s%s-%s",
1824 				 c->name, s->name, "thrash");
1825 			igt_subtest_group {
1826 				igt_fixture {
1827 					count = num_buffers(gem_mappable_aperture_size(),
1828 							    s, c, CHECK_RAM);
1829 				}
1830 				run_modes(name, c, modes, s, count);
1831 			}
1832 
1833 			/* Use the entire global GTT */
1834 			snprintf(name, sizeof(name), "%s%s-%s",
1835 				 c->name, s->name, "global");
1836 			igt_subtest_group {
1837 				igt_fixture {
1838 					count = num_buffers(gem_global_aperture_size(fd),
1839 							    s, c, CHECK_RAM);
1840 				}
1841 				run_modes(name, c, modes, s, count);
1842 			}
1843 
1844 			/* Use the entire per-process GTT */
1845 			snprintf(name, sizeof(name), "%s%s-%s",
1846 				 c->name, s->name, "full");
1847 			igt_subtest_group {
1848 				igt_fixture {
1849 					count = num_buffers(gem_aperture_size(fd),
1850 							    s, c, CHECK_RAM);
1851 				}
1852 				run_modes(name, c, modes, s, count);
1853 			}
1854 
1855 			snprintf(name, sizeof(name), "%s%s-%s",
1856 				 c->name, s->name, "shrink");
1857 			igt_subtest_group {
1858 				igt_fixture {
1859 					count = num_buffers(gem_mappable_aperture_size(),
1860 							    s, c, CHECK_RAM);
1861 
1862 					igt_fork_shrink_helper(fd);
1863 				}
1864 				run_modes(name, c, modes, s, count);
1865 
1866 				igt_fixture
1867 					igt_stop_shrink_helper();
1868 			}
1869 
1870 			/* Use the entire mappable aperture, force swapping */
1871 			snprintf(name, sizeof(name), "%s%s-%s",
1872 				 c->name, s->name, "swap");
1873 			igt_subtest_group {
1874 				igt_fixture {
1875 					if (intel_get_avail_ram_mb() > gem_mappable_aperture_size()/(1024*1024)) {
1876 						pin_sz = intel_get_avail_ram_mb() - gem_mappable_aperture_size()/(1024*1024);
1877 
1878 						igt_debug("Pinning %lld MiB\n", (long long)pin_sz);
1879 						pin_sz *= 1024 * 1024;
1880 
1881 						if (posix_memalign(&pinned, 4096, pin_sz) ||
1882 						    mlock(pinned, pin_sz) ||
1883 						    madvise(pinned, pin_sz, MADV_DONTFORK)) {
1884 							free(pinned);
1885 							pinned = NULL;
1886 						}
1887 						igt_require(pinned);
1888 					}
1889 
1890 					count = num_buffers(gem_mappable_aperture_size(),
1891 							    s, c, CHECK_RAM | CHECK_SWAP);
1892 				}
1893 				run_modes(name, c, modes, s, count);
1894 
1895 				igt_fixture {
1896 					if (pinned) {
1897 						munlock(pinned, pin_sz);
1898 						free(pinned);
1899 						pinned = NULL;
1900 					}
1901 				}
1902 			}
1903 		}
1904 	}
1905 }
1906