summaryrefslogtreecommitdiffstats
path: root/drivers/gpu/drm/i915/i915_gem_shrinker.c
blob: 6da795c7e62e47a12207cca861efafb1ea0f45f3 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
/*
 * Copyright © 2008-2015 Intel Corporation
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 *
 */

#include <linux/oom.h>
#include <linux/sched/mm.h>
#include <linux/shmem_fs.h>
#include <linux/slab.h>
#include <linux/swap.h>
#include <linux/pci.h>
#include <linux/dma-buf.h>
#include <linux/vmalloc.h>
#include <drm/i915_drm.h>

#include "i915_drv.h"
#include "i915_trace.h"

static bool shrinker_lock(struct drm_i915_private *i915,
			  unsigned int flags,
			  bool *unlock)
{
	struct mutex *m = &i915->drm.struct_mutex;

	switch (mutex_trylock_recursive(m)) {
	case MUTEX_TRYLOCK_RECURSIVE:
		*unlock = false;
		return true;

	case MUTEX_TRYLOCK_FAILED:
		*unlock = false;
		if (flags & I915_SHRINK_ACTIVE &&
		    mutex_lock_killable_nested(m, I915_MM_SHRINKER) == 0)
			*unlock = true;
		return *unlock;

	case MUTEX_TRYLOCK_SUCCESS:
		*unlock = true;
		return true;
	}

	BUG();
}

static void shrinker_unlock(struct drm_i915_private *i915, bool unlock)
{
	if (!unlock)
		return;

	mutex_unlock(&i915->drm.struct_mutex);
}

static bool swap_available(void)
{
	return get_nr_swap_pages() > 0;
}

static bool can_release_pages(struct drm_i915_gem_object *obj)
{
	/* Consider only shrinkable ojects. */
	if (!i915_gem_object_is_shrinkable(obj))
		return false;

	/* Only report true if by unbinding the object and putting its pages
	 * we can actually make forward progress towards freeing physical
	 * pages.
	 *
	 * If the pages are pinned for any other reason than being bound
	 * to the GPU, simply unbinding from the GPU is not going to succeed
	 * in releasing our pin count on the pages themselves.
	 */
	if (atomic_read(&obj->mm.pages_pin_count) > obj->bind_count)
		return false;

	/* If any vma are "permanently" pinned, it will prevent us from
	 * reclaiming the obj->mm.pages. We only allow scanout objects to claim
	 * a permanent pin, along with a few others like the context objects.
	 * To simplify the scan, and to avoid walking the list of vma under the
	 * object, we just check the count of its permanently pinned.
	 */
	if (READ_ONCE(obj->pin_global))
		return false;

	/* We can only return physical pages to the system if we can either
	 * discard the contents (because the user has marked them as being
	 * purgeable) or if we can move their contents out to swap.
	 */
	return swap_available() || obj->mm.madv == I915_MADV_DONTNEED;
}

static bool unsafe_drop_pages(struct drm_i915_gem_object *obj)
{
	if (i915_gem_object_unbind(obj) == 0)
		__i915_gem_object_put_pages(obj, I915_MM_SHRINKER);
	return !i915_gem_object_has_pages(obj);
}

/**
 * i915_gem_shrink - Shrink buffer object caches
 * @i915: i915 device
 * @target: amount of memory to make available, in pages
 * @nr_scanned: optional output for number of pages scanned (incremental)
 * @flags: control flags for selecting cache types
 *
 * This function is the main interface to the shrinker. It will try to release
 * up to @target pages of main memory backing storage from buffer objects.
 * Selection of the specific caches can be done with @flags. This is e.g. useful
 * when purgeable objects should be removed from caches preferentially.
 *
 * Note that it's not guaranteed that released amount is actually available as
 * free system memory - the pages might still be in-used to due to other reasons
 * (like cpu mmaps) or the mm core has reused them before we could grab them.
 * Therefore code that needs to explicitly shrink buffer objects caches (e.g. to
 * avoid deadlocks in memory reclaim) must fall back to i915_gem_shrink_all().
 *
 * Also note that any kind of pinning (both per-vma address space pins and
 * backing storage pins at the buffer object level) result in the shrinker code
 * having to skip the object.
 *
 * Returns:
 * The number of pages of backing storage actually released.
 */
unsigned long
i915_gem_shrink(struct drm_i915_private *i915,
		unsigned long target,
		unsigned long *nr_scanned,
		unsigned flags)
{
	const struct {
		struct list_head *list;
		unsigned int bit;
	} phases[] = {
		{ &i915->mm.unbound_list, I915_SHRINK_UNBOUND },
		{ &i915->mm.bound_list, I915_SHRINK_BOUND },
		{ NULL, 0 },
	}, *phase;
	intel_wakeref_t wakeref = 0;
	unsigned long count = 0;
	unsigned long scanned = 0;
	bool unlock;

	if (!shrinker_lock(i915, flags, &unlock))
		return 0;

	/*
	 * When shrinking the active list, also consider active contexts.
	 * Active contexts are pinned until they are retired, and so can
	 * not be simply unbound to retire and unpin their pages. To shrink
	 * the contexts, we must wait until the gpu is idle.
	 *
	 * We don't care about errors here; if we cannot wait upon the GPU,
	 * we will free as much as we can and hope to get a second chance.
	 */
	if (flags & I915_SHRINK_ACTIVE)
		i915_gem_wait_for_idle(i915,
				       I915_WAIT_LOCKED,
				       MAX_SCHEDULE_TIMEOUT);

	trace_i915_gem_shrink(i915, target, flags);
	i915_retire_requests(i915);

	/*
	 * Unbinding of objects will require HW access; Let us not wake the
	 * device just to recover a little memory. If absolutely necessary,
	 * we will force the wake during oom-notifier.
	 */
	if (flags & I915_SHRINK_BOUND) {
		wakeref = intel_runtime_pm_get_if_in_use(i915);
		if (!wakeref)
			flags &= ~I915_SHRINK_BOUND;
	}

	/*
	 * As we may completely rewrite the (un)bound list whilst unbinding
	 * (due to retiring requests) we have to strictly process only
	 * one element of the list at the time, and recheck the list
	 * on every iteration.
	 *
	 * In particular, we must hold a reference whilst removing the
	 * object as we may end up waiting for and/or retiring the objects.
	 * This might release the final reference (held by the active list)
	 * and result in the object being freed from under us. This is
	 * similar to the precautions the eviction code must take whilst
	 * removing objects.
	 *
	 * Also note that although these lists do not hold a reference to
	 * the object we can safely grab one here: The final object
	 * unreferencing and the bound_list are both protected by the
	 * dev->struct_mutex and so we won't ever be able to observe an
	 * object on the bound_list with a reference count equals 0.
	 */
	for (phase = phases; phase->list; phase++) {
		struct list_head still_in_list;
		struct drm_i915_gem_object *obj;

		if ((flags & phase->bit) == 0)
			continue;

		INIT_LIST_HEAD(&still_in_list);

		/*
		 * We serialize our access to unreferenced objects through
		 * the use of the struct_mutex. While the objects are not
		 * yet freed (due to RCU then a workqueue) we still want
		 * to be able to shrink their pages, so they remain on
		 * the unbound/bound list until actually freed.
		 */
		spin_lock(&i915->mm.obj_lock);
		while (count < target &&
		       (obj = list_first_entry_or_null(phase->list,
						       typeof(*obj),
						       mm.link))) {
			list_move_tail(&obj->mm.link, &still_in_list);

			if (flags & I915_SHRINK_PURGEABLE &&
			    obj->mm.madv != I915_MADV_DONTNEED)
				continue;

			if (flags & I915_SHRINK_VMAPS &&
			    !is_vmalloc_addr(obj->mm.mapping))
				continue;

			if (!(flags & I915_SHRINK_ACTIVE) &&
			    (i915_gem_object_is_active(obj) ||
			     i915_gem_object_is_framebuffer(obj)))
				continue;

			if (!can_release_pages(obj))
				continue;

			spin_unlock(&i915->mm.obj_lock);

			if (unsafe_drop_pages(obj)) {
				/* May arrive from get_pages on another bo */
				mutex_lock_nested(&obj->mm.lock,
						  I915_MM_SHRINKER);
				if (!i915_gem_object_has_pages(obj)) {
					__i915_gem_object_invalidate(obj);
					count += obj->base.size >> PAGE_SHIFT;
				}
				mutex_unlock(&obj->mm.lock);
			}
			scanned += obj->base.size >> PAGE_SHIFT;

			spin_lock(&i915->mm.obj_lock);
		}
		list_splice_tail(&still_in_list, phase->list);
		spin_unlock(&i915->mm.obj_lock);
	}

	if (flags & I915_SHRINK_BOUND)
		intel_runtime_pm_put(i915, wakeref);

	i915_retire_requests(i915);

	shrinker_unlock(i915, unlock);

	if (nr_scanned)
		*nr_scanned += scanned;
	return count;
}

/**
 * i915_gem_shrink_all - Shrink buffer object caches completely
 * @i915: i915 device
 *
 * This is a simple wraper around i915_gem_shrink() to aggressively shrink all
 * caches completely. It also first waits for and retires all outstanding
 * requests to also be able to release backing storage for active objects.
 *
 * This should only be used in code to intentionally quiescent the gpu or as a
 * last-ditch effort when memory seems to have run out.
 *
 * Returns:
 * The number of pages of backing storage actually released.
 */
unsigned long i915_gem_shrink_all(struct drm_i915_private *i915)
{
	intel_wakeref_t wakeref;
	unsigned long freed = 0;

	with_intel_runtime_pm(i915, wakeref) {
		freed = i915_gem_shrink(i915, -1UL, NULL,
					I915_SHRINK_BOUND |
					I915_SHRINK_UNBOUND |
					I915_SHRINK_ACTIVE);
	}

	return freed;
}

static unsigned long
i915_gem_shrinker_count(struct shrinker *shrinker, struct shrink_control *sc)
{
	struct drm_i915_private *i915 =
		container_of(shrinker, struct drm_i915_private, mm.shrinker);
	struct drm_i915_gem_object *obj;
	unsigned long num_objects = 0;
	unsigned long count = 0;

	spin_lock(&i915->mm.obj_lock);
	list_for_each_entry(obj, &i915->mm.unbound_list, mm.link)
		if (can_release_pages(obj)) {
			count += obj->base.size >> PAGE_SHIFT;
			num_objects++;
		}

	list_for_each_entry(obj, &i915->mm.bound_list, mm.link)
		if (!i915_gem_object_is_active(obj) && can_release_pages(obj)) {
			count += obj->base.size >> PAGE_SHIFT;
			num_objects++;
		}
	spin_unlock(&i915->mm.obj_lock);

	/* Update our preferred vmscan batch size for the next pass.
	 * Our rough guess for an effective batch size is roughly 2
	 * available GEM objects worth of pages. That is we don't want
	 * the shrinker to fire, until it is worth the cost of freeing an
	 * entire GEM object.
	 */
	if (num_objects) {
		unsigned long avg = 2 * count / num_objects;

		i915->mm.shrinker.batch =
			max((i915->mm.shrinker.batch + avg) >> 1,
			    128ul /* default SHRINK_BATCH */);
	}

	return count;
}

static unsigned long
i915_gem_shrinker_scan(struct shrinker *shrinker, struct shrink_control *sc)
{
	struct drm_i915_private *i915 =
		container_of(shrinker, struct drm_i915_private, mm.shrinker);
	unsigned long freed;
	bool unlock;

	sc->nr_scanned = 0;

	if (!shrinker_lock(i915, 0, &unlock))
		return SHRINK_STOP;

	freed = i915_gem_shrink(i915,
				sc->nr_to_scan,
				&sc->nr_scanned,
				I915_SHRINK_BOUND |
				I915_SHRINK_UNBOUND |
				I915_SHRINK_PURGEABLE);
	if (sc->nr_scanned < sc->nr_to_scan)
		freed += i915_gem_shrink(i915,
					 sc->nr_to_scan - sc->nr_scanned,
					 &sc->nr_scanned,
					 I915_SHRINK_BOUND |
					 I915_SHRINK_UNBOUND);
	if (sc->nr_scanned < sc->nr_to_scan && current_is_kswapd()) {
		intel_wakeref_t wakeref;

		with_intel_runtime_pm(i915, wakeref) {
			freed += i915_gem_shrink(i915,
						 sc->nr_to_scan - sc->nr_scanned,
						 &sc->nr_scanned,
						 I915_SHRINK_ACTIVE |
						 I915_SHRINK_BOUND |
						 I915_SHRINK_UNBOUND);
		}
	}

	shrinker_unlock(i915, unlock);

	return sc->nr_scanned ? freed : SHRINK_STOP;
}

static int
i915_gem_shrinker_oom(struct notifier_block *nb, unsigned long event, void *ptr)
{
	struct drm_i915_private *i915 =
		container_of(nb, struct drm_i915_private, mm.oom_notifier);
	struct drm_i915_gem_object *obj;
	unsigned long unevictable, bound, unbound, freed_pages;
	intel_wakeref_t wakeref;

	freed_pages = 0;
	with_intel_runtime_pm(i915, wakeref)
		freed_pages += i915_gem_shrink(i915, -1UL, NULL,
					       I915_SHRINK_BOUND |
					       I915_SHRINK_UNBOUND);

	/* Because we may be allocating inside our own driver, we cannot
	 * assert that there are no objects with pinned pages that are not
	 * being pointed to by hardware.
	 */
	unbound = bound = unevictable = 0;
	spin_lock(&i915->mm.obj_lock);
	list_for_each_entry(obj, &i915->mm.unbound_list, mm.link) {
		if (!can_release_pages(obj))
			unevictable += obj->base.size >> PAGE_SHIFT;
		else
			unbound += obj->base.size >> PAGE_SHIFT;
	}
	list_for_each_entry(obj, &i915->mm.bound_list, mm.link) {
		if (!can_release_pages(obj))
			unevictable += obj->base.size >> PAGE_SHIFT;
		else
			bound += obj->base.size >> PAGE_SHIFT;
	}
	spin_unlock(&i915->mm.obj_lock);

	if (freed_pages || unbound || bound)
		pr_info("Purging GPU memory, %lu pages freed, "
			"%lu pages still pinned.\n",
			freed_pages, unevictable);

	*(unsigned long *)ptr += freed_pages;
	return NOTIFY_DONE;
}

static int
i915_gem_shrinker_vmap(struct notifier_block *nb, unsigned long event, void *ptr)
{
	struct drm_i915_private *i915 =
		container_of(nb, struct drm_i915_private, mm.vmap_notifier);
	struct i915_vma *vma, *next;
	unsigned long freed_pages = 0;
	intel_wakeref_t wakeref;
	bool unlock;

	if (!shrinker_lock(i915, 0, &unlock))
		return NOTIFY_DONE;

	/* Force everything onto the inactive lists */
	if (i915_gem_wait_for_idle(i915,
				   I915_WAIT_LOCKED,
				   MAX_SCHEDULE_TIMEOUT))
		goto out;

	with_intel_runtime_pm(i915, wakeref)
		freed_pages += i915_gem_shrink(i915, -1UL, NULL,
					       I915_SHRINK_BOUND |
					       I915_SHRINK_UNBOUND |
					       I915_SHRINK_VMAPS);

	/* We also want to clear any cached iomaps as they wrap vmap */
	mutex_lock(&i915->ggtt.vm.mutex);
	list_for_each_entry_safe(vma, next,
				 &i915->ggtt.vm.bound_list, vm_link) {
		unsigned long count = vma->node.size >> PAGE_SHIFT;

		if (!vma->iomap || i915_vma_is_active(vma))
			continue;

		mutex_unlock(&i915->ggtt.vm.mutex);
		if (i915_vma_unbind(vma) == 0)
			freed_pages += count;
		mutex_lock(&i915->ggtt.vm.mutex);
	}
	mutex_unlock(&i915->ggtt.vm.mutex);

out:
	shrinker_unlock(i915, unlock);

	*(unsigned long *)ptr += freed_pages;
	return NOTIFY_DONE;
}

/**
 * i915_gem_shrinker_register - Register the i915 shrinker
 * @i915: i915 device
 *
 * This function registers and sets up the i915 shrinker and OOM handler.
 */
void i915_gem_shrinker_register(struct drm_i915_private *i915)
{
	i915->mm.shrinker.scan_objects = i915_gem_shrinker_scan;
	i915->mm.shrinker.count_objects = i915_gem_shrinker_count;
	i915->mm.shrinker.seeks = DEFAULT_SEEKS;
	i915->mm.shrinker.batch = 4096;
	WARN_ON(register_shrinker(&i915->mm.shrinker));

	i915->mm.oom_notifier.notifier_call = i915_gem_shrinker_oom;
	WARN_ON(register_oom_notifier(&i915->mm.oom_notifier));

	i915->mm.vmap_notifier.notifier_call = i915_gem_shrinker_vmap;
	WARN_ON(register_vmap_purge_notifier(&i915->mm.vmap_notifier));
}

/**
 * i915_gem_shrinker_unregister - Unregisters the i915 shrinker
 * @i915: i915 device
 *
 * This function unregisters the i915 shrinker and OOM handler.
 */
void i915_gem_shrinker_unregister(struct drm_i915_private *i915)
{
	WARN_ON(unregister_vmap_purge_notifier(&i915->mm.vmap_notifier));
	WARN_ON(unregister_oom_notifier(&i915->mm.oom_notifier));
	unregister_shrinker(&i915->mm.shrinker);
}

void i915_gem_shrinker_taints_mutex(struct drm_i915_private *i915,
				    struct mutex *mutex)
{
	bool unlock = false;

	if (!IS_ENABLED(CONFIG_LOCKDEP))
		return;

	if (!lockdep_is_held_type(&i915->drm.struct_mutex, -1)) {
		mutex_acquire(&i915->drm.struct_mutex.dep_map,
			      I915_MM_NORMAL, 0, _RET_IP_);
		unlock = true;
	}

	fs_reclaim_acquire(GFP_KERNEL);

	/*
	 * As we invariably rely on the struct_mutex within the shrinker,
	 * but have a complicated recursion dance, taint all the mutexes used
	 * within the shrinker with the struct_mutex. For completeness, we
	 * taint with all subclass of struct_mutex, even though we should
	 * only need tainting by I915_MM_NORMAL to catch possible ABBA
	 * deadlocks from using struct_mutex inside @mutex.
	 */
	mutex_acquire(&i915->drm.struct_mutex.dep_map,
		      I915_MM_SHRINKER, 0, _RET_IP_);

	mutex_acquire(&mutex->dep_map, 0, 0, _RET_IP_);
	mutex_release(&mutex->dep_map, 0, _RET_IP_);

	mutex_release(&i915->drm.struct_mutex.dep_map, 0, _RET_IP_);

	fs_reclaim_release(GFP_KERNEL);

	if (unlock)
		mutex_release(&i915->drm.struct_mutex.dep_map, 0, _RET_IP_);
}