huge_memory.c 92.2 KB
Newer Older
1 2 3 4 5 6 7
/*
 *  Copyright (C) 2009  Red Hat, Inc.
 *
 *  This work is licensed under the terms of the GNU GPL, version 2. See
 *  the COPYING file in the top-level directory.
 */

8 9
#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt

10 11 12 13 14 15 16
#include <linux/mm.h>
#include <linux/sched.h>
#include <linux/highmem.h>
#include <linux/hugetlb.h>
#include <linux/mmu_notifier.h>
#include <linux/rmap.h>
#include <linux/swap.h>
17
#include <linux/shrinker.h>
A
Andrea Arcangeli 已提交
18
#include <linux/mm_inline.h>
19
#include <linux/swapops.h>
M
Matthew Wilcox 已提交
20
#include <linux/dax.h>
A
Andrea Arcangeli 已提交
21 22
#include <linux/kthread.h>
#include <linux/khugepaged.h>
23
#include <linux/freezer.h>
24
#include <linux/pfn_t.h>
A
Andrea Arcangeli 已提交
25
#include <linux/mman.h>
26
#include <linux/memremap.h>
R
Ralf Baechle 已提交
27
#include <linux/pagemap.h>
28
#include <linux/debugfs.h>
29
#include <linux/migrate.h>
30
#include <linux/hashtable.h>
31
#include <linux/userfaultfd_k.h>
32
#include <linux/page_idle.h>
33

34 35 36 37
#include <asm/tlb.h>
#include <asm/pgalloc.h>
#include "internal.h"

38 39 40 41 42 43 44 45 46 47 48 49 50 51
enum scan_result {
	SCAN_FAIL,
	SCAN_SUCCEED,
	SCAN_PMD_NULL,
	SCAN_EXCEED_NONE_PTE,
	SCAN_PTE_NON_PRESENT,
	SCAN_PAGE_RO,
	SCAN_NO_REFERENCED_PAGE,
	SCAN_PAGE_NULL,
	SCAN_SCAN_ABORT,
	SCAN_PAGE_COUNT,
	SCAN_PAGE_LRU,
	SCAN_PAGE_LOCK,
	SCAN_PAGE_ANON,
52
	SCAN_PAGE_COMPOUND,
53 54 55 56 57 58 59 60 61 62 63 64 65
	SCAN_ANY_PROCESS,
	SCAN_VMA_NULL,
	SCAN_VMA_CHECK,
	SCAN_ADDRESS_RANGE,
	SCAN_SWAP_CACHE_PAGE,
	SCAN_DEL_PAGE_LRU,
	SCAN_ALLOC_HUGE_PAGE_FAIL,
	SCAN_CGROUP_CHARGE_FAIL
};

#define CREATE_TRACE_POINTS
#include <trace/events/huge_memory.h>

A
Andrea Arcangeli 已提交
66
/*
67 68 69 70 71 72
 * By default transparent hugepage support is disabled in order that avoid
 * to risk increase the memory footprint of applications without a guaranteed
 * benefit. When transparent hugepage support is enabled, is for all mappings,
 * and khugepaged scans all mappings.
 * Defrag is invoked by khugepaged hugepage allocations and by page faults
 * for all hugepage allocations.
A
Andrea Arcangeli 已提交
73
 */
74
unsigned long transparent_hugepage_flags __read_mostly =
75
#ifdef CONFIG_TRANSPARENT_HUGEPAGE_ALWAYS
A
Andrea Arcangeli 已提交
76
	(1<<TRANSPARENT_HUGEPAGE_FLAG)|
77 78 79 80
#endif
#ifdef CONFIG_TRANSPARENT_HUGEPAGE_MADVISE
	(1<<TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG)|
#endif
81
	(1<<TRANSPARENT_HUGEPAGE_DEFRAG_REQ_MADV_FLAG)|
82 83
	(1<<TRANSPARENT_HUGEPAGE_DEFRAG_KHUGEPAGED_FLAG)|
	(1<<TRANSPARENT_HUGEPAGE_USE_ZERO_PAGE_FLAG);
A
Andrea Arcangeli 已提交
84 85

/* default scan 8*512 pte (or vmas) every 30 second */
86
static unsigned int khugepaged_pages_to_scan __read_mostly;
A
Andrea Arcangeli 已提交
87 88 89 90 91
static unsigned int khugepaged_pages_collapsed;
static unsigned int khugepaged_full_scans;
static unsigned int khugepaged_scan_sleep_millisecs __read_mostly = 10000;
/* during fragmentation poll the hugepage allocator once every minute */
static unsigned int khugepaged_alloc_sleep_millisecs __read_mostly = 60000;
92
static unsigned long khugepaged_sleep_expire;
A
Andrea Arcangeli 已提交
93 94 95 96 97 98 99 100 101
static struct task_struct *khugepaged_thread __read_mostly;
static DEFINE_MUTEX(khugepaged_mutex);
static DEFINE_SPINLOCK(khugepaged_mm_lock);
static DECLARE_WAIT_QUEUE_HEAD(khugepaged_wait);
/*
 * default collapse hugepages if there is at least one pte mapped like
 * it would have happened if the vma was large enough during page
 * fault.
 */
102
static unsigned int khugepaged_max_ptes_none __read_mostly;
A
Andrea Arcangeli 已提交
103 104 105

static int khugepaged(void *none);
static int khugepaged_slab_init(void);
106
static void khugepaged_slab_exit(void);
A
Andrea Arcangeli 已提交
107

108 109 110
#define MM_SLOTS_HASH_BITS 10
static __read_mostly DEFINE_HASHTABLE(mm_slots_hash, MM_SLOTS_HASH_BITS);

A
Andrea Arcangeli 已提交
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
static struct kmem_cache *mm_slot_cache __read_mostly;

/**
 * struct mm_slot - hash lookup from mm to mm_slot
 * @hash: hash collision list
 * @mm_node: khugepaged scan list headed in khugepaged_scan.mm_head
 * @mm: the mm that this information is valid for
 */
struct mm_slot {
	struct hlist_node hash;
	struct list_head mm_node;
	struct mm_struct *mm;
};

/**
 * struct khugepaged_scan - cursor for scanning
 * @mm_head: the head of the mm list to scan
 * @mm_slot: the current mm_slot we are scanning
 * @address: the next address inside that to be scanned
 *
 * There is only the one khugepaged_scan instance of this cursor structure.
 */
struct khugepaged_scan {
	struct list_head mm_head;
	struct mm_slot *mm_slot;
	unsigned long address;
137 138
};
static struct khugepaged_scan khugepaged_scan = {
A
Andrea Arcangeli 已提交
139 140 141
	.mm_head = LIST_HEAD_INIT(khugepaged_scan.mm_head),
};

142
static struct shrinker deferred_split_shrinker;
143

144
static void set_recommended_min_free_kbytes(void)
145 146 147 148 149 150 151 152
{
	struct zone *zone;
	int nr_zones = 0;
	unsigned long recommended_min;

	for_each_populated_zone(zone)
		nr_zones++;

153
	/* Ensure 2 pageblocks are free to assist fragmentation avoidance */
154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169
	recommended_min = pageblock_nr_pages * nr_zones * 2;

	/*
	 * Make sure that on average at least two pageblocks are almost free
	 * of another type, one for a migratetype to fall back to and a
	 * second to avoid subsequent fallbacks of other types There are 3
	 * MIGRATE_TYPES we care about.
	 */
	recommended_min += pageblock_nr_pages * nr_zones *
			   MIGRATE_PCPTYPES * MIGRATE_PCPTYPES;

	/* don't ever allow to reserve more than 5% of the lowmem */
	recommended_min = min(recommended_min,
			      (unsigned long) nr_free_buffer_pages() / 20);
	recommended_min <<= (PAGE_SHIFT-10);

170 171
	if (recommended_min > min_free_kbytes) {
		if (user_min_free_kbytes >= 0)
J
Joe Perches 已提交
172
			pr_info("raising min_free_kbytes from %d to %lu to help transparent hugepage allocations\n",
173 174
				min_free_kbytes, recommended_min);

175
		min_free_kbytes = recommended_min;
176
	}
177 178 179
	setup_per_zone_wmarks();
}

180
static int start_stop_khugepaged(void)
A
Andrea Arcangeli 已提交
181 182 183 184 185 186
{
	int err = 0;
	if (khugepaged_enabled()) {
		if (!khugepaged_thread)
			khugepaged_thread = kthread_run(khugepaged, NULL,
							"khugepaged");
187
		if (IS_ERR(khugepaged_thread)) {
188
			pr_err("khugepaged: kthread_run(khugepaged) failed\n");
A
Andrea Arcangeli 已提交
189 190
			err = PTR_ERR(khugepaged_thread);
			khugepaged_thread = NULL;
191
			goto fail;
A
Andrea Arcangeli 已提交
192
		}
193 194

		if (!list_empty(&khugepaged_scan.mm_head))
A
Andrea Arcangeli 已提交
195
			wake_up_interruptible(&khugepaged_wait);
196 197

		set_recommended_min_free_kbytes();
198 199 200 201
	} else if (khugepaged_thread) {
		kthread_stop(khugepaged_thread);
		khugepaged_thread = NULL;
	}
202
fail:
A
Andrea Arcangeli 已提交
203 204
	return err;
}
205

206
static atomic_t huge_zero_refcount;
207
struct page *huge_zero_page __read_mostly;
208

209
struct page *get_huge_zero_page(void)
210 211 212 213
{
	struct page *zero_page;
retry:
	if (likely(atomic_inc_not_zero(&huge_zero_refcount)))
214
		return READ_ONCE(huge_zero_page);
215 216

	zero_page = alloc_pages((GFP_TRANSHUGE | __GFP_ZERO) & ~__GFP_MOVABLE,
217
			HPAGE_PMD_ORDER);
218 219
	if (!zero_page) {
		count_vm_event(THP_ZERO_PAGE_ALLOC_FAILED);
220
		return NULL;
221 222
	}
	count_vm_event(THP_ZERO_PAGE_ALLOC);
223
	preempt_disable();
224
	if (cmpxchg(&huge_zero_page, NULL, zero_page)) {
225
		preempt_enable();
226
		__free_pages(zero_page, compound_order(zero_page));
227 228 229 230 231 232
		goto retry;
	}

	/* We take additional reference here. It will be put back by shrinker */
	atomic_set(&huge_zero_refcount, 2);
	preempt_enable();
233
	return READ_ONCE(huge_zero_page);
234 235
}

236
void put_huge_zero_page(void)
237
{
238 239 240 241 242
	/*
	 * Counter should never go to zero here. Only shrinker can put
	 * last reference.
	 */
	BUG_ON(atomic_dec_and_test(&huge_zero_refcount));
243 244
}

245 246
static unsigned long shrink_huge_zero_page_count(struct shrinker *shrink,
					struct shrink_control *sc)
247
{
248 249 250
	/* we can free zero page only if last reference remains */
	return atomic_read(&huge_zero_refcount) == 1 ? HPAGE_PMD_NR : 0;
}
251

252 253 254
static unsigned long shrink_huge_zero_page_scan(struct shrinker *shrink,
				       struct shrink_control *sc)
{
255
	if (atomic_cmpxchg(&huge_zero_refcount, 1, 0) == 1) {
256 257
		struct page *zero_page = xchg(&huge_zero_page, NULL);
		BUG_ON(zero_page == NULL);
258
		__free_pages(zero_page, compound_order(zero_page));
259
		return HPAGE_PMD_NR;
260 261 262
	}

	return 0;
263 264
}

265
static struct shrinker huge_zero_page_shrinker = {
266 267
	.count_objects = shrink_huge_zero_page_count,
	.scan_objects = shrink_huge_zero_page_scan,
268 269 270
	.seeks = DEFAULT_SEEKS,
};

271
#ifdef CONFIG_SYSFS
A
Andrea Arcangeli 已提交
272

273
static ssize_t triple_flag_store(struct kobject *kobj,
274 275 276
				 struct kobj_attribute *attr,
				 const char *buf, size_t count,
				 enum transparent_hugepage_flag enabled,
277
				 enum transparent_hugepage_flag deferred,
278 279
				 enum transparent_hugepage_flag req_madv)
{
280 281 282 283 284 285 286 287
	if (!memcmp("defer", buf,
		    min(sizeof("defer")-1, count))) {
		if (enabled == deferred)
			return -EINVAL;
		clear_bit(enabled, &transparent_hugepage_flags);
		clear_bit(req_madv, &transparent_hugepage_flags);
		set_bit(deferred, &transparent_hugepage_flags);
	} else if (!memcmp("always", buf,
288
		    min(sizeof("always")-1, count))) {
289
		clear_bit(deferred, &transparent_hugepage_flags);
290
		clear_bit(req_madv, &transparent_hugepage_flags);
291
		set_bit(enabled, &transparent_hugepage_flags);
292 293 294
	} else if (!memcmp("madvise", buf,
			   min(sizeof("madvise")-1, count))) {
		clear_bit(enabled, &transparent_hugepage_flags);
295
		clear_bit(deferred, &transparent_hugepage_flags);
296 297 298 299 300
		set_bit(req_madv, &transparent_hugepage_flags);
	} else if (!memcmp("never", buf,
			   min(sizeof("never")-1, count))) {
		clear_bit(enabled, &transparent_hugepage_flags);
		clear_bit(req_madv, &transparent_hugepage_flags);
301
		clear_bit(deferred, &transparent_hugepage_flags);
302 303 304 305 306 307 308 309 310
	} else
		return -EINVAL;

	return count;
}

static ssize_t enabled_show(struct kobject *kobj,
			    struct kobj_attribute *attr, char *buf)
{
311 312 313 314 315 316
	if (test_bit(TRANSPARENT_HUGEPAGE_FLAG, &transparent_hugepage_flags))
		return sprintf(buf, "[always] madvise never\n");
	else if (test_bit(TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG, &transparent_hugepage_flags))
		return sprintf(buf, "always [madvise] never\n");
	else
		return sprintf(buf, "always madvise [never]\n");
317
}
318

319 320 321 322
static ssize_t enabled_store(struct kobject *kobj,
			     struct kobj_attribute *attr,
			     const char *buf, size_t count)
{
A
Andrea Arcangeli 已提交
323 324
	ssize_t ret;

325 326
	ret = triple_flag_store(kobj, attr, buf, count,
				TRANSPARENT_HUGEPAGE_FLAG,
A
Andrea Arcangeli 已提交
327 328 329 330
				TRANSPARENT_HUGEPAGE_FLAG,
				TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG);

	if (ret > 0) {
331 332 333
		int err;

		mutex_lock(&khugepaged_mutex);
334
		err = start_stop_khugepaged();
335 336
		mutex_unlock(&khugepaged_mutex);

A
Andrea Arcangeli 已提交
337 338 339 340 341
		if (err)
			ret = err;
	}

	return ret;
342 343 344 345 346 347 348 349
}
static struct kobj_attribute enabled_attr =
	__ATTR(enabled, 0644, enabled_show, enabled_store);

static ssize_t single_flag_show(struct kobject *kobj,
				struct kobj_attribute *attr, char *buf,
				enum transparent_hugepage_flag flag)
{
350 351
	return sprintf(buf, "%d\n",
		       !!test_bit(flag, &transparent_hugepage_flags));
352
}
353

354 355 356 357 358
static ssize_t single_flag_store(struct kobject *kobj,
				 struct kobj_attribute *attr,
				 const char *buf, size_t count,
				 enum transparent_hugepage_flag flag)
{
359 360 361 362 363 364 365 366 367 368
	unsigned long value;
	int ret;

	ret = kstrtoul(buf, 10, &value);
	if (ret < 0)
		return ret;
	if (value > 1)
		return -EINVAL;

	if (value)
369
		set_bit(flag, &transparent_hugepage_flags);
370
	else
371 372 373 374 375 376 377 378 379 380 381 382 383
		clear_bit(flag, &transparent_hugepage_flags);

	return count;
}

/*
 * Currently defrag only disables __GFP_NOWAIT for allocation. A blind
 * __GFP_REPEAT is too aggressive, it's never worth swapping tons of
 * memory just to allocate one more hugepage.
 */
static ssize_t defrag_show(struct kobject *kobj,
			   struct kobj_attribute *attr, char *buf)
{
384 385 386 387 388 389 390 391 392
	if (test_bit(TRANSPARENT_HUGEPAGE_DEFRAG_DIRECT_FLAG, &transparent_hugepage_flags))
		return sprintf(buf, "[always] defer madvise never\n");
	if (test_bit(TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_FLAG, &transparent_hugepage_flags))
		return sprintf(buf, "always [defer] madvise never\n");
	else if (test_bit(TRANSPARENT_HUGEPAGE_DEFRAG_REQ_MADV_FLAG, &transparent_hugepage_flags))
		return sprintf(buf, "always defer [madvise] never\n");
	else
		return sprintf(buf, "always defer madvise [never]\n");

393 394 395 396 397
}
static ssize_t defrag_store(struct kobject *kobj,
			    struct kobj_attribute *attr,
			    const char *buf, size_t count)
{
398 399 400
	return triple_flag_store(kobj, attr, buf, count,
				 TRANSPARENT_HUGEPAGE_DEFRAG_DIRECT_FLAG,
				 TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_FLAG,
401 402 403 404 405
				 TRANSPARENT_HUGEPAGE_DEFRAG_REQ_MADV_FLAG);
}
static struct kobj_attribute defrag_attr =
	__ATTR(defrag, 0644, defrag_show, defrag_store);

406 407 408 409 410 411 412 413 414 415 416 417 418 419
static ssize_t use_zero_page_show(struct kobject *kobj,
		struct kobj_attribute *attr, char *buf)
{
	return single_flag_show(kobj, attr, buf,
				TRANSPARENT_HUGEPAGE_USE_ZERO_PAGE_FLAG);
}
static ssize_t use_zero_page_store(struct kobject *kobj,
		struct kobj_attribute *attr, const char *buf, size_t count)
{
	return single_flag_store(kobj, attr, buf, count,
				 TRANSPARENT_HUGEPAGE_USE_ZERO_PAGE_FLAG);
}
static struct kobj_attribute use_zero_page_attr =
	__ATTR(use_zero_page, 0644, use_zero_page_show, use_zero_page_store);
420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440
#ifdef CONFIG_DEBUG_VM
static ssize_t debug_cow_show(struct kobject *kobj,
				struct kobj_attribute *attr, char *buf)
{
	return single_flag_show(kobj, attr, buf,
				TRANSPARENT_HUGEPAGE_DEBUG_COW_FLAG);
}
static ssize_t debug_cow_store(struct kobject *kobj,
			       struct kobj_attribute *attr,
			       const char *buf, size_t count)
{
	return single_flag_store(kobj, attr, buf, count,
				 TRANSPARENT_HUGEPAGE_DEBUG_COW_FLAG);
}
static struct kobj_attribute debug_cow_attr =
	__ATTR(debug_cow, 0644, debug_cow_show, debug_cow_store);
#endif /* CONFIG_DEBUG_VM */

static struct attribute *hugepage_attr[] = {
	&enabled_attr.attr,
	&defrag_attr.attr,
441
	&use_zero_page_attr.attr,
442 443 444 445 446 447 448 449
#ifdef CONFIG_DEBUG_VM
	&debug_cow_attr.attr,
#endif
	NULL,
};

static struct attribute_group hugepage_attr_group = {
	.attrs = hugepage_attr,
A
Andrea Arcangeli 已提交
450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465
};

static ssize_t scan_sleep_millisecs_show(struct kobject *kobj,
					 struct kobj_attribute *attr,
					 char *buf)
{
	return sprintf(buf, "%u\n", khugepaged_scan_sleep_millisecs);
}

static ssize_t scan_sleep_millisecs_store(struct kobject *kobj,
					  struct kobj_attribute *attr,
					  const char *buf, size_t count)
{
	unsigned long msecs;
	int err;

466
	err = kstrtoul(buf, 10, &msecs);
A
Andrea Arcangeli 已提交
467 468 469 470
	if (err || msecs > UINT_MAX)
		return -EINVAL;

	khugepaged_scan_sleep_millisecs = msecs;
471
	khugepaged_sleep_expire = 0;
A
Andrea Arcangeli 已提交
472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493
	wake_up_interruptible(&khugepaged_wait);

	return count;
}
static struct kobj_attribute scan_sleep_millisecs_attr =
	__ATTR(scan_sleep_millisecs, 0644, scan_sleep_millisecs_show,
	       scan_sleep_millisecs_store);

static ssize_t alloc_sleep_millisecs_show(struct kobject *kobj,
					  struct kobj_attribute *attr,
					  char *buf)
{
	return sprintf(buf, "%u\n", khugepaged_alloc_sleep_millisecs);
}

static ssize_t alloc_sleep_millisecs_store(struct kobject *kobj,
					   struct kobj_attribute *attr,
					   const char *buf, size_t count)
{
	unsigned long msecs;
	int err;

494
	err = kstrtoul(buf, 10, &msecs);
A
Andrea Arcangeli 已提交
495 496 497 498
	if (err || msecs > UINT_MAX)
		return -EINVAL;

	khugepaged_alloc_sleep_millisecs = msecs;
499
	khugepaged_sleep_expire = 0;
A
Andrea Arcangeli 已提交
500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520
	wake_up_interruptible(&khugepaged_wait);

	return count;
}
static struct kobj_attribute alloc_sleep_millisecs_attr =
	__ATTR(alloc_sleep_millisecs, 0644, alloc_sleep_millisecs_show,
	       alloc_sleep_millisecs_store);

static ssize_t pages_to_scan_show(struct kobject *kobj,
				  struct kobj_attribute *attr,
				  char *buf)
{
	return sprintf(buf, "%u\n", khugepaged_pages_to_scan);
}
static ssize_t pages_to_scan_store(struct kobject *kobj,
				   struct kobj_attribute *attr,
				   const char *buf, size_t count)
{
	int err;
	unsigned long pages;

521
	err = kstrtoul(buf, 10, &pages);
A
Andrea Arcangeli 已提交
522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588
	if (err || !pages || pages > UINT_MAX)
		return -EINVAL;

	khugepaged_pages_to_scan = pages;

	return count;
}
static struct kobj_attribute pages_to_scan_attr =
	__ATTR(pages_to_scan, 0644, pages_to_scan_show,
	       pages_to_scan_store);

static ssize_t pages_collapsed_show(struct kobject *kobj,
				    struct kobj_attribute *attr,
				    char *buf)
{
	return sprintf(buf, "%u\n", khugepaged_pages_collapsed);
}
static struct kobj_attribute pages_collapsed_attr =
	__ATTR_RO(pages_collapsed);

static ssize_t full_scans_show(struct kobject *kobj,
			       struct kobj_attribute *attr,
			       char *buf)
{
	return sprintf(buf, "%u\n", khugepaged_full_scans);
}
static struct kobj_attribute full_scans_attr =
	__ATTR_RO(full_scans);

static ssize_t khugepaged_defrag_show(struct kobject *kobj,
				      struct kobj_attribute *attr, char *buf)
{
	return single_flag_show(kobj, attr, buf,
				TRANSPARENT_HUGEPAGE_DEFRAG_KHUGEPAGED_FLAG);
}
static ssize_t khugepaged_defrag_store(struct kobject *kobj,
				       struct kobj_attribute *attr,
				       const char *buf, size_t count)
{
	return single_flag_store(kobj, attr, buf, count,
				 TRANSPARENT_HUGEPAGE_DEFRAG_KHUGEPAGED_FLAG);
}
static struct kobj_attribute khugepaged_defrag_attr =
	__ATTR(defrag, 0644, khugepaged_defrag_show,
	       khugepaged_defrag_store);

/*
 * max_ptes_none controls if khugepaged should collapse hugepages over
 * any unmapped ptes in turn potentially increasing the memory
 * footprint of the vmas. When max_ptes_none is 0 khugepaged will not
 * reduce the available free memory in the system as it
 * runs. Increasing max_ptes_none will instead potentially reduce the
 * free memory in the system during the khugepaged scan.
 */
static ssize_t khugepaged_max_ptes_none_show(struct kobject *kobj,
					     struct kobj_attribute *attr,
					     char *buf)
{
	return sprintf(buf, "%u\n", khugepaged_max_ptes_none);
}
static ssize_t khugepaged_max_ptes_none_store(struct kobject *kobj,
					      struct kobj_attribute *attr,
					      const char *buf, size_t count)
{
	int err;
	unsigned long max_ptes_none;

589
	err = kstrtoul(buf, 10, &max_ptes_none);
A
Andrea Arcangeli 已提交
590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614
	if (err || max_ptes_none > HPAGE_PMD_NR-1)
		return -EINVAL;

	khugepaged_max_ptes_none = max_ptes_none;

	return count;
}
static struct kobj_attribute khugepaged_max_ptes_none_attr =
	__ATTR(max_ptes_none, 0644, khugepaged_max_ptes_none_show,
	       khugepaged_max_ptes_none_store);

static struct attribute *khugepaged_attr[] = {
	&khugepaged_defrag_attr.attr,
	&khugepaged_max_ptes_none_attr.attr,
	&pages_to_scan_attr.attr,
	&pages_collapsed_attr.attr,
	&full_scans_attr.attr,
	&scan_sleep_millisecs_attr.attr,
	&alloc_sleep_millisecs_attr.attr,
	NULL,
};

static struct attribute_group khugepaged_attr_group = {
	.attrs = khugepaged_attr,
	.name = "khugepaged",
615 616
};

S
Shaohua Li 已提交
617
static int __init hugepage_init_sysfs(struct kobject **hugepage_kobj)
618 619 620
{
	int err;

S
Shaohua Li 已提交
621 622
	*hugepage_kobj = kobject_create_and_add("transparent_hugepage", mm_kobj);
	if (unlikely(!*hugepage_kobj)) {
623
		pr_err("failed to create transparent hugepage kobject\n");
S
Shaohua Li 已提交
624
		return -ENOMEM;
A
Andrea Arcangeli 已提交
625 626
	}

S
Shaohua Li 已提交
627
	err = sysfs_create_group(*hugepage_kobj, &hugepage_attr_group);
A
Andrea Arcangeli 已提交
628
	if (err) {
629
		pr_err("failed to register transparent hugepage group\n");
S
Shaohua Li 已提交
630
		goto delete_obj;
A
Andrea Arcangeli 已提交
631 632
	}

S
Shaohua Li 已提交
633
	err = sysfs_create_group(*hugepage_kobj, &khugepaged_attr_group);
A
Andrea Arcangeli 已提交
634
	if (err) {
635
		pr_err("failed to register transparent hugepage group\n");
S
Shaohua Li 已提交
636
		goto remove_hp_group;
A
Andrea Arcangeli 已提交
637
	}
S
Shaohua Li 已提交
638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674

	return 0;

remove_hp_group:
	sysfs_remove_group(*hugepage_kobj, &hugepage_attr_group);
delete_obj:
	kobject_put(*hugepage_kobj);
	return err;
}

static void __init hugepage_exit_sysfs(struct kobject *hugepage_kobj)
{
	sysfs_remove_group(hugepage_kobj, &khugepaged_attr_group);
	sysfs_remove_group(hugepage_kobj, &hugepage_attr_group);
	kobject_put(hugepage_kobj);
}
#else
static inline int hugepage_init_sysfs(struct kobject **hugepage_kobj)
{
	return 0;
}

static inline void hugepage_exit_sysfs(struct kobject *hugepage_kobj)
{
}
#endif /* CONFIG_SYSFS */

static int __init hugepage_init(void)
{
	int err;
	struct kobject *hugepage_kobj;

	if (!has_transparent_hugepage()) {
		transparent_hugepage_flags = 0;
		return -EINVAL;
	}

675 676 677 678 679 680 681 682 683 684 685 686
	khugepaged_pages_to_scan = HPAGE_PMD_NR * 8;
	khugepaged_max_ptes_none = HPAGE_PMD_NR - 1;
	/*
	 * hugepages can't be allocated by the buddy allocator
	 */
	MAYBE_BUILD_BUG_ON(HPAGE_PMD_ORDER >= MAX_ORDER);
	/*
	 * we use page->mapping and page->index in second tail page
	 * as list_head: assuming THP order >= 2
	 */
	MAYBE_BUILD_BUG_ON(HPAGE_PMD_ORDER < 2);

S
Shaohua Li 已提交
687 688
	err = hugepage_init_sysfs(&hugepage_kobj);
	if (err)
689
		goto err_sysfs;
A
Andrea Arcangeli 已提交
690 691 692

	err = khugepaged_slab_init();
	if (err)
693
		goto err_slab;
A
Andrea Arcangeli 已提交
694

695 696 697
	err = register_shrinker(&huge_zero_page_shrinker);
	if (err)
		goto err_hzp_shrinker;
698 699 700
	err = register_shrinker(&deferred_split_shrinker);
	if (err)
		goto err_split_shrinker;
701

702 703 704 705 706
	/*
	 * By default disable transparent hugepages on smaller systems,
	 * where the extra memory used could hurt more than TLB overhead
	 * is likely to save.  The admin can still enable it through /sys.
	 */
707
	if (totalram_pages < (512 << (20 - PAGE_SHIFT))) {
708
		transparent_hugepage_flags = 0;
709 710
		return 0;
	}
711

712
	err = start_stop_khugepaged();
713 714
	if (err)
		goto err_khugepaged;
A
Andrea Arcangeli 已提交
715

S
Shaohua Li 已提交
716
	return 0;
717
err_khugepaged:
718 719
	unregister_shrinker(&deferred_split_shrinker);
err_split_shrinker:
720 721 722 723
	unregister_shrinker(&huge_zero_page_shrinker);
err_hzp_shrinker:
	khugepaged_slab_exit();
err_slab:
S
Shaohua Li 已提交
724
	hugepage_exit_sysfs(hugepage_kobj);
725
err_sysfs:
A
Andrea Arcangeli 已提交
726
	return err;
727
}
728
subsys_initcall(hugepage_init);
729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755

static int __init setup_transparent_hugepage(char *str)
{
	int ret = 0;
	if (!str)
		goto out;
	if (!strcmp(str, "always")) {
		set_bit(TRANSPARENT_HUGEPAGE_FLAG,
			&transparent_hugepage_flags);
		clear_bit(TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG,
			  &transparent_hugepage_flags);
		ret = 1;
	} else if (!strcmp(str, "madvise")) {
		clear_bit(TRANSPARENT_HUGEPAGE_FLAG,
			  &transparent_hugepage_flags);
		set_bit(TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG,
			&transparent_hugepage_flags);
		ret = 1;
	} else if (!strcmp(str, "never")) {
		clear_bit(TRANSPARENT_HUGEPAGE_FLAG,
			  &transparent_hugepage_flags);
		clear_bit(TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG,
			  &transparent_hugepage_flags);
		ret = 1;
	}
out:
	if (!ret)
756
		pr_warn("transparent_hugepage= cannot parse, ignored\n");
757 758 759 760
	return ret;
}
__setup("transparent_hugepage=", setup_transparent_hugepage);

761
pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma)
762 763 764 765 766 767
{
	if (likely(vma->vm_flags & VM_WRITE))
		pmd = pmd_mkwrite(pmd);
	return pmd;
}

768
static inline pmd_t mk_huge_pmd(struct page *page, pgprot_t prot)
B
Bob Liu 已提交
769
{
770
	return pmd_mkhuge(mk_pmd(page, prot));
B
Bob Liu 已提交
771 772
}

773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792
static inline struct list_head *page_deferred_list(struct page *page)
{
	/*
	 * ->lru in the tail pages is occupied by compound_head.
	 * Let's use ->mapping + ->index in the second tail page as list_head.
	 */
	return (struct list_head *)&page[2].mapping;
}

void prep_transhuge_page(struct page *page)
{
	/*
	 * we use page->mapping and page->indexlru in second tail page
	 * as list_head: assuming THP order >= 2
	 */

	INIT_LIST_HEAD(page_deferred_list(page));
	set_compound_page_dtor(page, TRANSHUGE_PAGE_DTOR);
}

793 794
static int __do_huge_pmd_anonymous_page(struct mm_struct *mm,
					struct vm_area_struct *vma,
795
					unsigned long address, pmd_t *pmd,
796 797
					struct page *page, gfp_t gfp,
					unsigned int flags)
798
{
799
	struct mem_cgroup *memcg;
800
	pgtable_t pgtable;
801
	spinlock_t *ptl;
802
	unsigned long haddr = address & HPAGE_PMD_MASK;
803

804
	VM_BUG_ON_PAGE(!PageCompound(page), page);
805

806
	if (mem_cgroup_try_charge(page, mm, gfp, &memcg, true)) {
807 808 809 810
		put_page(page);
		count_vm_event(THP_FAULT_FALLBACK);
		return VM_FAULT_FALLBACK;
	}
811

812
	pgtable = pte_alloc_one(mm, haddr);
813
	if (unlikely(!pgtable)) {
814
		mem_cgroup_cancel_charge(page, memcg, true);
815
		put_page(page);
816
		return VM_FAULT_OOM;
817
	}
818 819

	clear_huge_page(page, haddr, HPAGE_PMD_NR);
820 821 822 823 824
	/*
	 * The memory barrier inside __SetPageUptodate makes sure that
	 * clear_huge_page writes become visible before the set_pmd_at()
	 * write.
	 */
825 826
	__SetPageUptodate(page);

827
	ptl = pmd_lock(mm, pmd);
828
	if (unlikely(!pmd_none(*pmd))) {
829
		spin_unlock(ptl);
830
		mem_cgroup_cancel_charge(page, memcg, true);
831 832 833 834
		put_page(page);
		pte_free(mm, pgtable);
	} else {
		pmd_t entry;
835 836 837 838 839 840

		/* Deliver the page fault to userland */
		if (userfaultfd_missing(vma)) {
			int ret;

			spin_unlock(ptl);
841
			mem_cgroup_cancel_charge(page, memcg, true);
842 843
			put_page(page);
			pte_free(mm, pgtable);
844
			ret = handle_userfault(vma, address, flags,
845 846 847 848 849
					       VM_UFFD_MISSING);
			VM_BUG_ON(ret & VM_FAULT_FALLBACK);
			return ret;
		}

850 851
		entry = mk_huge_pmd(page, vma->vm_page_prot);
		entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma);
852
		page_add_new_anon_rmap(page, vma, haddr, true);
853
		mem_cgroup_commit_charge(page, memcg, false, true);
854
		lru_cache_add_active_or_unevictable(page, vma);
855
		pgtable_trans_huge_deposit(mm, pmd, pgtable);
856 857
		set_pmd_at(mm, haddr, pmd, entry);
		add_mm_counter(mm, MM_ANONPAGES, HPAGE_PMD_NR);
858
		atomic_long_inc(&mm->nr_ptes);
859
		spin_unlock(ptl);
860
		count_vm_event(THP_FAULT_ALLOC);
861 862
	}

863
	return 0;
864 865
}

866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887
/*
 * If THP is set to always then directly reclaim/compact as necessary
 * If set to defer then do no reclaim and defer to khugepaged
 * If set to madvise and the VMA is flagged then directly reclaim/compact
 */
static inline gfp_t alloc_hugepage_direct_gfpmask(struct vm_area_struct *vma)
{
	gfp_t reclaim_flags = 0;

	if (test_bit(TRANSPARENT_HUGEPAGE_DEFRAG_REQ_MADV_FLAG, &transparent_hugepage_flags) &&
	    (vma->vm_flags & VM_HUGEPAGE))
		reclaim_flags = __GFP_DIRECT_RECLAIM;
	else if (test_bit(TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_FLAG, &transparent_hugepage_flags))
		reclaim_flags = __GFP_KSWAPD_RECLAIM;
	else if (test_bit(TRANSPARENT_HUGEPAGE_DEFRAG_DIRECT_FLAG, &transparent_hugepage_flags))
		reclaim_flags = __GFP_DIRECT_RECLAIM;

	return GFP_TRANSHUGE | reclaim_flags;
}

/* Defrag for khugepaged will enter direct reclaim/compaction if necessary */
static inline gfp_t alloc_hugepage_khugepaged_gfpmask(void)
888
{
889
	return GFP_TRANSHUGE | (khugepaged_defrag() ? __GFP_DIRECT_RECLAIM : 0);
890 891
}

892
/* Caller must hold page table lock. */
893
static bool set_huge_zero_page(pgtable_t pgtable, struct mm_struct *mm,
894
		struct vm_area_struct *vma, unsigned long haddr, pmd_t *pmd,
895
		struct page *zero_page)
896 897
{
	pmd_t entry;
A
Andrew Morton 已提交
898 899
	if (!pmd_none(*pmd))
		return false;
900
	entry = mk_pmd(zero_page, vma->vm_page_prot);
901
	entry = pmd_mkhuge(entry);
902 903
	if (pgtable)
		pgtable_trans_huge_deposit(mm, pmd, pgtable);
904
	set_pmd_at(mm, haddr, pmd, entry);
905
	atomic_long_inc(&mm->nr_ptes);
A
Andrew Morton 已提交
906
	return true;
907 908
}

909 910 911 912
int do_huge_pmd_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
			       unsigned long address, pmd_t *pmd,
			       unsigned int flags)
{
913
	gfp_t gfp;
914 915 916
	struct page *page;
	unsigned long haddr = address & HPAGE_PMD_MASK;

917
	if (haddr < vma->vm_start || haddr + HPAGE_PMD_SIZE > vma->vm_end)
918
		return VM_FAULT_FALLBACK;
919 920
	if (unlikely(anon_vma_prepare(vma)))
		return VM_FAULT_OOM;
921
	if (unlikely(khugepaged_enter(vma, vma->vm_flags)))
922
		return VM_FAULT_OOM;
923
	if (!(flags & FAULT_FLAG_WRITE) && !mm_forbids_zeropage(mm) &&
924
			transparent_hugepage_use_zero_page()) {
925
		spinlock_t *ptl;
926 927 928
		pgtable_t pgtable;
		struct page *zero_page;
		bool set;
929
		int ret;
930 931
		pgtable = pte_alloc_one(mm, haddr);
		if (unlikely(!pgtable))
A
Andrea Arcangeli 已提交
932
			return VM_FAULT_OOM;
933 934 935
		zero_page = get_huge_zero_page();
		if (unlikely(!zero_page)) {
			pte_free(mm, pgtable);
936
			count_vm_event(THP_FAULT_FALLBACK);
937
			return VM_FAULT_FALLBACK;
A
Andrea Arcangeli 已提交
938
		}
939
		ptl = pmd_lock(mm, pmd);
940 941 942 943 944
		ret = 0;
		set = false;
		if (pmd_none(*pmd)) {
			if (userfaultfd_missing(vma)) {
				spin_unlock(ptl);
945
				ret = handle_userfault(vma, address, flags,
946 947 948 949 950 951 952 953 954 955 956
						       VM_UFFD_MISSING);
				VM_BUG_ON(ret & VM_FAULT_FALLBACK);
			} else {
				set_huge_zero_page(pgtable, mm, vma,
						   haddr, pmd,
						   zero_page);
				spin_unlock(ptl);
				set = true;
			}
		} else
			spin_unlock(ptl);
957 958 959
		if (!set) {
			pte_free(mm, pgtable);
			put_huge_zero_page();
960
		}
961
		return ret;
962
	}
963
	gfp = alloc_hugepage_direct_gfpmask(vma);
964
	page = alloc_hugepage_vma(gfp, vma, haddr, HPAGE_PMD_ORDER);
965 966
	if (unlikely(!page)) {
		count_vm_event(THP_FAULT_FALLBACK);
967
		return VM_FAULT_FALLBACK;
968
	}
969
	prep_transhuge_page(page);
970 971
	return __do_huge_pmd_anonymous_page(mm, vma, address, pmd, page, gfp,
					    flags);
972 973
}

974
static void insert_pfn_pmd(struct vm_area_struct *vma, unsigned long addr,
975
		pmd_t *pmd, pfn_t pfn, pgprot_t prot, bool write)
M
Matthew Wilcox 已提交
976 977 978 979 980 981
{
	struct mm_struct *mm = vma->vm_mm;
	pmd_t entry;
	spinlock_t *ptl;

	ptl = pmd_lock(mm, pmd);
982 983 984
	entry = pmd_mkhuge(pfn_t_pmd(pfn, prot));
	if (pfn_t_devmap(pfn))
		entry = pmd_mkdevmap(entry);
985 986 987
	if (write) {
		entry = pmd_mkyoung(pmd_mkdirty(entry));
		entry = maybe_pmd_mkwrite(entry, vma);
M
Matthew Wilcox 已提交
988
	}
989 990
	set_pmd_at(mm, addr, pmd, entry);
	update_mmu_cache_pmd(vma, addr, pmd);
M
Matthew Wilcox 已提交
991 992 993 994
	spin_unlock(ptl);
}

int vmf_insert_pfn_pmd(struct vm_area_struct *vma, unsigned long addr,
995
			pmd_t *pmd, pfn_t pfn, bool write)
M
Matthew Wilcox 已提交
996 997 998 999 1000 1001 1002 1003 1004 1005 1006
{
	pgprot_t pgprot = vma->vm_page_prot;
	/*
	 * If we had pmd_special, we could avoid all these restrictions,
	 * but we need to be consistent with PTEs and architectures that
	 * can't support a 'special' bit.
	 */
	BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
	BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
						(VM_PFNMAP|VM_MIXEDMAP));
	BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
1007
	BUG_ON(!pfn_t_devmap(pfn));
M
Matthew Wilcox 已提交
1008 1009 1010 1011 1012

	if (addr < vma->vm_start || addr >= vma->vm_end)
		return VM_FAULT_SIGBUS;
	if (track_pfn_insert(vma, &pgprot, pfn))
		return VM_FAULT_SIGBUS;
1013 1014
	insert_pfn_pmd(vma, addr, pmd, pfn, pgprot, write);
	return VM_FAULT_NOPAGE;
M
Matthew Wilcox 已提交
1015
}
1016
EXPORT_SYMBOL_GPL(vmf_insert_pfn_pmd);
M
Matthew Wilcox 已提交
1017

1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074
static void touch_pmd(struct vm_area_struct *vma, unsigned long addr,
		pmd_t *pmd)
{
	pmd_t _pmd;

	/*
	 * We should set the dirty bit only for FOLL_WRITE but for now
	 * the dirty bit in the pmd is meaningless.  And if the dirty
	 * bit will become meaningful and we'll only set it with
	 * FOLL_WRITE, an atomic set_bit will be required on the pmd to
	 * set the young bit, instead of the current set_pmd_at.
	 */
	_pmd = pmd_mkyoung(pmd_mkdirty(*pmd));
	if (pmdp_set_access_flags(vma, addr & HPAGE_PMD_MASK,
				pmd, _pmd,  1))
		update_mmu_cache_pmd(vma, addr, pmd);
}

struct page *follow_devmap_pmd(struct vm_area_struct *vma, unsigned long addr,
		pmd_t *pmd, int flags)
{
	unsigned long pfn = pmd_pfn(*pmd);
	struct mm_struct *mm = vma->vm_mm;
	struct dev_pagemap *pgmap;
	struct page *page;

	assert_spin_locked(pmd_lockptr(mm, pmd));

	if (flags & FOLL_WRITE && !pmd_write(*pmd))
		return NULL;

	if (pmd_present(*pmd) && pmd_devmap(*pmd))
		/* pass */;
	else
		return NULL;

	if (flags & FOLL_TOUCH)
		touch_pmd(vma, addr, pmd);

	/*
	 * device mapped pages can only be returned if the
	 * caller will manage the page reference count.
	 */
	if (!(flags & FOLL_GET))
		return ERR_PTR(-EEXIST);

	pfn += (addr & ~PMD_MASK) >> PAGE_SHIFT;
	pgmap = get_dev_pagemap(pfn, NULL);
	if (!pgmap)
		return ERR_PTR(-EFAULT);
	page = pfn_to_page(pfn);
	get_page(page);
	put_dev_pagemap(pgmap);

	return page;
}

1075 1076 1077 1078
int copy_huge_pmd(struct mm_struct *dst_mm, struct mm_struct *src_mm,
		  pmd_t *dst_pmd, pmd_t *src_pmd, unsigned long addr,
		  struct vm_area_struct *vma)
{
1079
	spinlock_t *dst_ptl, *src_ptl;
1080 1081
	struct page *src_page;
	pmd_t pmd;
1082
	pgtable_t pgtable = NULL;
1083 1084
	int ret;

1085 1086 1087 1088 1089 1090
	if (!vma_is_dax(vma)) {
		ret = -ENOMEM;
		pgtable = pte_alloc_one(dst_mm, addr);
		if (unlikely(!pgtable))
			goto out;
	}
1091

1092 1093 1094
	dst_ptl = pmd_lock(dst_mm, dst_pmd);
	src_ptl = pmd_lockptr(src_mm, src_pmd);
	spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
1095 1096 1097

	ret = -EAGAIN;
	pmd = *src_pmd;
1098
	if (unlikely(!pmd_trans_huge(pmd) && !pmd_devmap(pmd))) {
1099 1100 1101
		pte_free(dst_mm, pgtable);
		goto out_unlock;
	}
1102
	/*
1103
	 * When page table lock is held, the huge zero pmd should not be
1104 1105 1106 1107
	 * under splitting since we don't split the page itself, only pmd to
	 * a page table.
	 */
	if (is_huge_zero_pmd(pmd)) {
1108
		struct page *zero_page;
1109 1110 1111 1112 1113
		/*
		 * get_huge_zero_page() will never allocate a new page here,
		 * since we already have a zero page to copy. It just takes a
		 * reference.
		 */
1114
		zero_page = get_huge_zero_page();
1115
		set_huge_zero_page(pgtable, dst_mm, vma, addr, dst_pmd,
1116
				zero_page);
1117 1118 1119
		ret = 0;
		goto out_unlock;
	}
1120

1121
	if (!vma_is_dax(vma)) {
1122 1123 1124 1125 1126 1127 1128 1129 1130
		/* thp accounting separate from pmd_devmap accounting */
		src_page = pmd_page(pmd);
		VM_BUG_ON_PAGE(!PageHead(src_page), src_page);
		get_page(src_page);
		page_dup_rmap(src_page, true);
		add_mm_counter(dst_mm, MM_ANONPAGES, HPAGE_PMD_NR);
		atomic_long_inc(&dst_mm->nr_ptes);
		pgtable_trans_huge_deposit(dst_mm, dst_pmd, pgtable);
	}
1131 1132 1133 1134 1135 1136 1137

	pmdp_set_wrprotect(src_mm, addr, src_pmd);
	pmd = pmd_mkold(pmd_wrprotect(pmd));
	set_pmd_at(dst_mm, addr, dst_pmd, pmd);

	ret = 0;
out_unlock:
1138 1139
	spin_unlock(src_ptl);
	spin_unlock(dst_ptl);
1140 1141 1142 1143
out:
	return ret;
}

1144 1145 1146 1147 1148 1149
void huge_pmd_set_accessed(struct mm_struct *mm,
			   struct vm_area_struct *vma,
			   unsigned long address,
			   pmd_t *pmd, pmd_t orig_pmd,
			   int dirty)
{
1150
	spinlock_t *ptl;
1151 1152 1153
	pmd_t entry;
	unsigned long haddr;

1154
	ptl = pmd_lock(mm, pmd);
1155 1156 1157 1158 1159 1160 1161 1162 1163
	if (unlikely(!pmd_same(*pmd, orig_pmd)))
		goto unlock;

	entry = pmd_mkyoung(orig_pmd);
	haddr = address & HPAGE_PMD_MASK;
	if (pmdp_set_access_flags(vma, haddr, pmd, entry, dirty))
		update_mmu_cache_pmd(vma, address, pmd);

unlock:
1164
	spin_unlock(ptl);
1165 1166
}

1167 1168 1169 1170 1171 1172 1173
static int do_huge_pmd_wp_page_fallback(struct mm_struct *mm,
					struct vm_area_struct *vma,
					unsigned long address,
					pmd_t *pmd, pmd_t orig_pmd,
					struct page *page,
					unsigned long haddr)
{
1174
	struct mem_cgroup *memcg;
1175
	spinlock_t *ptl;
1176 1177 1178 1179
	pgtable_t pgtable;
	pmd_t _pmd;
	int ret = 0, i;
	struct page **pages;
1180 1181
	unsigned long mmun_start;	/* For mmu_notifiers */
	unsigned long mmun_end;		/* For mmu_notifiers */
1182 1183 1184 1185 1186 1187 1188 1189 1190

	pages = kmalloc(sizeof(struct page *) * HPAGE_PMD_NR,
			GFP_KERNEL);
	if (unlikely(!pages)) {
		ret |= VM_FAULT_OOM;
		goto out;
	}

	for (i = 0; i < HPAGE_PMD_NR; i++) {
1191 1192
		pages[i] = alloc_page_vma_node(GFP_HIGHUSER_MOVABLE |
					       __GFP_OTHER_NODE,
1193
					       vma, address, page_to_nid(page));
A
Andrea Arcangeli 已提交
1194
		if (unlikely(!pages[i] ||
1195
			     mem_cgroup_try_charge(pages[i], mm, GFP_KERNEL,
1196
						   &memcg, false))) {
A
Andrea Arcangeli 已提交
1197
			if (pages[i])
1198
				put_page(pages[i]);
A
Andrea Arcangeli 已提交
1199
			while (--i >= 0) {
1200 1201
				memcg = (void *)page_private(pages[i]);
				set_page_private(pages[i], 0);
1202 1203
				mem_cgroup_cancel_charge(pages[i], memcg,
						false);
A
Andrea Arcangeli 已提交
1204 1205
				put_page(pages[i]);
			}
1206 1207 1208 1209
			kfree(pages);
			ret |= VM_FAULT_OOM;
			goto out;
		}
1210
		set_page_private(pages[i], (unsigned long)memcg);
1211 1212 1213 1214
	}

	for (i = 0; i < HPAGE_PMD_NR; i++) {
		copy_user_highpage(pages[i], page + i,
1215
				   haddr + PAGE_SIZE * i, vma);
1216 1217 1218 1219
		__SetPageUptodate(pages[i]);
		cond_resched();
	}

1220 1221 1222 1223
	mmun_start = haddr;
	mmun_end   = haddr + HPAGE_PMD_SIZE;
	mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);

1224
	ptl = pmd_lock(mm, pmd);
1225 1226
	if (unlikely(!pmd_same(*pmd, orig_pmd)))
		goto out_free_pages;
1227
	VM_BUG_ON_PAGE(!PageHead(page), page);
1228

1229
	pmdp_huge_clear_flush_notify(vma, haddr, pmd);
1230 1231
	/* leave pmd empty until pte is filled */

1232
	pgtable = pgtable_trans_huge_withdraw(mm, pmd);
1233 1234 1235 1236 1237 1238
	pmd_populate(mm, &_pmd, pgtable);

	for (i = 0; i < HPAGE_PMD_NR; i++, haddr += PAGE_SIZE) {
		pte_t *pte, entry;
		entry = mk_pte(pages[i], vma->vm_page_prot);
		entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1239 1240
		memcg = (void *)page_private(pages[i]);
		set_page_private(pages[i], 0);
1241
		page_add_new_anon_rmap(pages[i], vma, haddr, false);
1242
		mem_cgroup_commit_charge(pages[i], memcg, false, false);
1243
		lru_cache_add_active_or_unevictable(pages[i], vma);
1244 1245 1246 1247 1248 1249 1250 1251 1252
		pte = pte_offset_map(&_pmd, haddr);
		VM_BUG_ON(!pte_none(*pte));
		set_pte_at(mm, haddr, pte, entry);
		pte_unmap(pte);
	}
	kfree(pages);

	smp_wmb(); /* make pte visible before pmd */
	pmd_populate(mm, pmd, pgtable);
1253
	page_remove_rmap(page, true);
1254
	spin_unlock(ptl);
1255

1256 1257
	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);

1258 1259 1260 1261 1262 1263 1264
	ret |= VM_FAULT_WRITE;
	put_page(page);

out:
	return ret;

out_free_pages:
1265
	spin_unlock(ptl);
1266
	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
A
Andrea Arcangeli 已提交
1267
	for (i = 0; i < HPAGE_PMD_NR; i++) {
1268 1269
		memcg = (void *)page_private(pages[i]);
		set_page_private(pages[i], 0);
1270
		mem_cgroup_cancel_charge(pages[i], memcg, false);
1271
		put_page(pages[i]);
A
Andrea Arcangeli 已提交
1272
	}
1273 1274 1275 1276 1277 1278 1279
	kfree(pages);
	goto out;
}

int do_huge_pmd_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
			unsigned long address, pmd_t *pmd, pmd_t orig_pmd)
{
1280
	spinlock_t *ptl;
1281
	int ret = 0;
1282
	struct page *page = NULL, *new_page;
1283
	struct mem_cgroup *memcg;
1284
	unsigned long haddr;
1285 1286
	unsigned long mmun_start;	/* For mmu_notifiers */
	unsigned long mmun_end;		/* For mmu_notifiers */
1287
	gfp_t huge_gfp;			/* for allocation and charge */
1288

1289
	ptl = pmd_lockptr(mm, pmd);
1290
	VM_BUG_ON_VMA(!vma->anon_vma, vma);
1291 1292 1293
	haddr = address & HPAGE_PMD_MASK;
	if (is_huge_zero_pmd(orig_pmd))
		goto alloc;
1294
	spin_lock(ptl);
1295 1296 1297 1298
	if (unlikely(!pmd_same(*pmd, orig_pmd)))
		goto out_unlock;

	page = pmd_page(orig_pmd);
1299
	VM_BUG_ON_PAGE(!PageCompound(page) || !PageHead(page), page);
1300 1301
	/*
	 * We can only reuse the page if nobody else maps the huge page or it's
1302
	 * part.
1303
	 */
1304
	if (page_trans_huge_mapcount(page, NULL) == 1) {
1305 1306 1307 1308
		pmd_t entry;
		entry = pmd_mkyoung(orig_pmd);
		entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma);
		if (pmdp_set_access_flags(vma, haddr, pmd, entry,  1))
1309
			update_mmu_cache_pmd(vma, address, pmd);
1310 1311 1312
		ret |= VM_FAULT_WRITE;
		goto out_unlock;
	}
1313
	get_page(page);
1314
	spin_unlock(ptl);
1315
alloc:
1316
	if (transparent_hugepage_enabled(vma) &&
1317
	    !transparent_hugepage_debug_cow()) {
1318
		huge_gfp = alloc_hugepage_direct_gfpmask(vma);
1319
		new_page = alloc_hugepage_vma(huge_gfp, vma, haddr, HPAGE_PMD_ORDER);
1320
	} else
1321 1322
		new_page = NULL;

1323 1324 1325
	if (likely(new_page)) {
		prep_transhuge_page(new_page);
	} else {
1326
		if (!page) {
1327
			split_huge_pmd(vma, pmd, address);
1328
			ret |= VM_FAULT_FALLBACK;
1329 1330 1331
		} else {
			ret = do_huge_pmd_wp_page_fallback(mm, vma, address,
					pmd, orig_pmd, page, haddr);
1332
			if (ret & VM_FAULT_OOM) {
1333
				split_huge_pmd(vma, pmd, address);
1334 1335
				ret |= VM_FAULT_FALLBACK;
			}
1336
			put_page(page);
1337
		}
1338
		count_vm_event(THP_FAULT_FALLBACK);
1339 1340 1341
		goto out;
	}

1342 1343
	if (unlikely(mem_cgroup_try_charge(new_page, mm, huge_gfp, &memcg,
					   true))) {
A
Andrea Arcangeli 已提交
1344
		put_page(new_page);
1345
		if (page) {
1346
			split_huge_pmd(vma, pmd, address);
1347
			put_page(page);
1348
		} else
1349
			split_huge_pmd(vma, pmd, address);
1350
		ret |= VM_FAULT_FALLBACK;
1351
		count_vm_event(THP_FAULT_FALLBACK);
A
Andrea Arcangeli 已提交
1352 1353 1354
		goto out;
	}

1355 1356
	count_vm_event(THP_FAULT_ALLOC);

1357
	if (!page)
1358 1359 1360
		clear_huge_page(new_page, haddr, HPAGE_PMD_NR);
	else
		copy_user_huge_page(new_page, page, haddr, vma, HPAGE_PMD_NR);
1361 1362
	__SetPageUptodate(new_page);

1363 1364 1365 1366
	mmun_start = haddr;
	mmun_end   = haddr + HPAGE_PMD_SIZE;
	mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);

1367
	spin_lock(ptl);
1368
	if (page)
1369
		put_page(page);
A
Andrea Arcangeli 已提交
1370
	if (unlikely(!pmd_same(*pmd, orig_pmd))) {
1371
		spin_unlock(ptl);
1372
		mem_cgroup_cancel_charge(new_page, memcg, true);
1373
		put_page(new_page);
1374
		goto out_mn;
A
Andrea Arcangeli 已提交
1375
	} else {
1376
		pmd_t entry;
1377 1378
		entry = mk_huge_pmd(new_page, vma->vm_page_prot);
		entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma);
1379
		pmdp_huge_clear_flush_notify(vma, haddr, pmd);
1380
		page_add_new_anon_rmap(new_page, vma, haddr, true);
1381
		mem_cgroup_commit_charge(new_page, memcg, false, true);
1382
		lru_cache_add_active_or_unevictable(new_page, vma);
1383
		set_pmd_at(mm, haddr, pmd, entry);
1384
		update_mmu_cache_pmd(vma, address, pmd);
1385
		if (!page) {
1386
			add_mm_counter(mm, MM_ANONPAGES, HPAGE_PMD_NR);
1387 1388
			put_huge_zero_page();
		} else {
1389
			VM_BUG_ON_PAGE(!PageHead(page), page);
1390
			page_remove_rmap(page, true);
1391 1392
			put_page(page);
		}
1393 1394
		ret |= VM_FAULT_WRITE;
	}
1395
	spin_unlock(ptl);
1396 1397
out_mn:
	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
1398 1399
out:
	return ret;
1400
out_unlock:
1401
	spin_unlock(ptl);
1402
	return ret;
1403 1404
}

1405
struct page *follow_trans_huge_pmd(struct vm_area_struct *vma,
1406 1407 1408 1409
				   unsigned long addr,
				   pmd_t *pmd,
				   unsigned int flags)
{
1410
	struct mm_struct *mm = vma->vm_mm;
1411 1412
	struct page *page = NULL;

1413
	assert_spin_locked(pmd_lockptr(mm, pmd));
1414 1415 1416 1417

	if (flags & FOLL_WRITE && !pmd_write(*pmd))
		goto out;

1418 1419 1420 1421
	/* Avoid dumping huge zero page */
	if ((flags & FOLL_DUMP) && is_huge_zero_pmd(*pmd))
		return ERR_PTR(-EFAULT);

1422
	/* Full NUMA hinting faults to serialise migration in fault paths */
1423
	if ((flags & FOLL_NUMA) && pmd_protnone(*pmd))
1424 1425
		goto out;

1426
	page = pmd_page(*pmd);
1427
	VM_BUG_ON_PAGE(!PageHead(page), page);
1428 1429
	if (flags & FOLL_TOUCH)
		touch_pmd(vma, addr, pmd);
E
Eric B Munson 已提交
1430
	if ((flags & FOLL_MLOCK) && (vma->vm_flags & VM_LOCKED)) {
1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444
		/*
		 * We don't mlock() pte-mapped THPs. This way we can avoid
		 * leaking mlocked pages into non-VM_LOCKED VMAs.
		 *
		 * In most cases the pmd is the only mapping of the page as we
		 * break COW for the mlock() -- see gup_flags |= FOLL_WRITE for
		 * writable private mappings in populate_vma_page_range().
		 *
		 * The only scenario when we have the page shared here is if we
		 * mlocking read-only mapping shared over fork(). We skip
		 * mlocking such pages.
		 */
		if (compound_mapcount(page) == 1 && !PageDoubleMap(page) &&
				page->mapping && trylock_page(page)) {
1445 1446 1447 1448 1449 1450
			lru_add_drain();
			if (page->mapping)
				mlock_vma_page(page);
			unlock_page(page);
		}
	}
1451
	page += (addr & ~HPAGE_PMD_MASK) >> PAGE_SHIFT;
1452
	VM_BUG_ON_PAGE(!PageCompound(page), page);
1453
	if (flags & FOLL_GET)
1454
		get_page(page);
1455 1456 1457 1458 1459

out:
	return page;
}

1460
/* NUMA hinting page fault entry point for trans huge pmds */
1461 1462
int do_huge_pmd_numa_page(struct mm_struct *mm, struct vm_area_struct *vma,
				unsigned long addr, pmd_t pmd, pmd_t *pmdp)
1463
{
1464
	spinlock_t *ptl;
1465
	struct anon_vma *anon_vma = NULL;
1466
	struct page *page;
1467
	unsigned long haddr = addr & HPAGE_PMD_MASK;
1468
	int page_nid = -1, this_nid = numa_node_id();
1469
	int target_nid, last_cpupid = -1;
1470 1471
	bool page_locked;
	bool migrated = false;
1472
	bool was_writable;
1473
	int flags = 0;
1474

1475 1476 1477
	/* A PROT_NONE fault should not end up here */
	BUG_ON(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)));

1478
	ptl = pmd_lock(mm, pmdp);
1479 1480 1481
	if (unlikely(!pmd_same(pmd, *pmdp)))
		goto out_unlock;

1482 1483 1484 1485 1486 1487
	/*
	 * If there are potential migrations, wait for completion and retry
	 * without disrupting NUMA hinting information. Do not relock and
	 * check_same as the page may no longer be mapped.
	 */
	if (unlikely(pmd_trans_migrating(*pmdp))) {
1488
		page = pmd_page(*pmdp);
1489
		spin_unlock(ptl);
1490
		wait_on_page_locked(page);
1491 1492 1493
		goto out;
	}

1494
	page = pmd_page(pmd);
1495
	BUG_ON(is_huge_zero_page(page));
1496
	page_nid = page_to_nid(page);
1497
	last_cpupid = page_cpupid_last(page);
1498
	count_vm_numa_event(NUMA_HINT_FAULTS);
1499
	if (page_nid == this_nid) {
1500
		count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL);
1501 1502
		flags |= TNF_FAULT_LOCAL;
	}
1503

1504 1505
	/* See similar comment in do_numa_page for explanation */
	if (!(vma->vm_flags & VM_WRITE))
1506 1507
		flags |= TNF_NO_GROUP;

1508 1509 1510 1511
	/*
	 * Acquire the page lock to serialise THP migrations but avoid dropping
	 * page_table_lock if at all possible
	 */
1512 1513 1514 1515
	page_locked = trylock_page(page);
	target_nid = mpol_misplaced(page, vma, haddr);
	if (target_nid == -1) {
		/* If the page was locked, there are no parallel migrations */
1516
		if (page_locked)
1517
			goto clear_pmdnuma;
1518
	}
1519

1520
	/* Migration could have started since the pmd_trans_migrating check */
1521
	if (!page_locked) {
1522
		spin_unlock(ptl);
1523
		wait_on_page_locked(page);
1524
		page_nid = -1;
1525 1526 1527
		goto out;
	}

1528 1529 1530 1531
	/*
	 * Page is misplaced. Page lock serialises migrations. Acquire anon_vma
	 * to serialises splits
	 */
1532
	get_page(page);
1533
	spin_unlock(ptl);
1534
	anon_vma = page_lock_anon_vma_read(page);
1535

P
Peter Zijlstra 已提交
1536
	/* Confirm the PMD did not change while page_table_lock was released */
1537
	spin_lock(ptl);
1538 1539 1540
	if (unlikely(!pmd_same(pmd, *pmdp))) {
		unlock_page(page);
		put_page(page);
1541
		page_nid = -1;
1542
		goto out_unlock;
1543
	}
1544

1545 1546 1547 1548 1549 1550 1551
	/* Bail if we fail to protect against THP splits for any reason */
	if (unlikely(!anon_vma)) {
		put_page(page);
		page_nid = -1;
		goto clear_pmdnuma;
	}

1552 1553
	/*
	 * Migrate the THP to the requested node, returns with page unlocked
1554
	 * and access rights restored.
1555
	 */
1556
	spin_unlock(ptl);
1557
	migrated = migrate_misplaced_transhuge_page(mm, vma,
1558
				pmdp, pmd, addr, page, target_nid);
1559 1560
	if (migrated) {
		flags |= TNF_MIGRATED;
1561
		page_nid = target_nid;
1562 1563
	} else
		flags |= TNF_MIGRATE_FAIL;
1564

1565
	goto out;
1566
clear_pmdnuma:
1567
	BUG_ON(!PageLocked(page));
1568
	was_writable = pmd_write(pmd);
1569
	pmd = pmd_modify(pmd, vma->vm_page_prot);
1570
	pmd = pmd_mkyoung(pmd);
1571 1572
	if (was_writable)
		pmd = pmd_mkwrite(pmd);
1573 1574
	set_pmd_at(mm, haddr, pmdp, pmd);
	update_mmu_cache_pmd(vma, addr, pmdp);
1575
	unlock_page(page);
1576
out_unlock:
1577
	spin_unlock(ptl);
1578 1579 1580 1581 1582

out:
	if (anon_vma)
		page_unlock_anon_vma_read(anon_vma);

1583
	if (page_nid != -1)
1584
		task_numa_fault(last_cpupid, page_nid, HPAGE_PMD_NR, flags);
1585

1586 1587 1588
	return 0;
}

1589 1590 1591 1592 1593 1594 1595 1596 1597 1598
int madvise_free_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
		pmd_t *pmd, unsigned long addr, unsigned long next)

{
	spinlock_t *ptl;
	pmd_t orig_pmd;
	struct page *page;
	struct mm_struct *mm = tlb->mm;
	int ret = 0;

1599 1600
	ptl = pmd_trans_huge_lock(pmd, vma);
	if (!ptl)
1601
		goto out_unlocked;
1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626

	orig_pmd = *pmd;
	if (is_huge_zero_pmd(orig_pmd)) {
		ret = 1;
		goto out;
	}

	page = pmd_page(orig_pmd);
	/*
	 * If other processes are mapping this page, we couldn't discard
	 * the page unless they all do MADV_FREE so let's skip the page.
	 */
	if (page_mapcount(page) != 1)
		goto out;

	if (!trylock_page(page))
		goto out;

	/*
	 * If user want to discard part-pages of THP, split it so MADV_FREE
	 * will deactivate only them.
	 */
	if (next - addr != HPAGE_PMD_SIZE) {
		get_page(page);
		spin_unlock(ptl);
1627
		split_huge_page(page);
1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655
		put_page(page);
		unlock_page(page);
		goto out_unlocked;
	}

	if (PageDirty(page))
		ClearPageDirty(page);
	unlock_page(page);

	if (PageActive(page))
		deactivate_page(page);

	if (pmd_young(orig_pmd) || pmd_dirty(orig_pmd)) {
		orig_pmd = pmdp_huge_get_and_clear_full(tlb->mm, addr, pmd,
			tlb->fullmm);
		orig_pmd = pmd_mkold(orig_pmd);
		orig_pmd = pmd_mkclean(orig_pmd);

		set_pmd_at(mm, addr, pmd, orig_pmd);
		tlb_remove_pmd_tlb_entry(tlb, pmd, addr);
	}
	ret = 1;
out:
	spin_unlock(ptl);
out_unlocked:
	return ret;
}

1656
int zap_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
S
Shaohua Li 已提交
1657
		 pmd_t *pmd, unsigned long addr)
1658
{
1659
	pmd_t orig_pmd;
1660
	spinlock_t *ptl;
1661

1662 1663
	ptl = __pmd_trans_huge_lock(pmd, vma);
	if (!ptl)
1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676
		return 0;
	/*
	 * For architectures like ppc64 we look at deposited pgtable
	 * when calling pmdp_huge_get_and_clear. So do the
	 * pgtable_trans_huge_withdraw after finishing pmdp related
	 * operations.
	 */
	orig_pmd = pmdp_huge_get_and_clear_full(tlb->mm, addr, pmd,
			tlb->fullmm);
	tlb_remove_pmd_tlb_entry(tlb, pmd, addr);
	if (vma_is_dax(vma)) {
		spin_unlock(ptl);
		if (is_huge_zero_pmd(orig_pmd))
1677
			tlb_remove_page(tlb, pmd_page(orig_pmd));
1678 1679 1680 1681
	} else if (is_huge_zero_pmd(orig_pmd)) {
		pte_free(tlb->mm, pgtable_trans_huge_withdraw(tlb->mm, pmd));
		atomic_long_dec(&tlb->mm->nr_ptes);
		spin_unlock(ptl);
1682
		tlb_remove_page(tlb, pmd_page(orig_pmd));
1683 1684
	} else {
		struct page *page = pmd_page(orig_pmd);
1685
		page_remove_rmap(page, true);
1686 1687 1688 1689 1690 1691 1692
		VM_BUG_ON_PAGE(page_mapcount(page) < 0, page);
		add_mm_counter(tlb->mm, MM_ANONPAGES, -HPAGE_PMD_NR);
		VM_BUG_ON_PAGE(!PageHead(page), page);
		pte_free(tlb->mm, pgtable_trans_huge_withdraw(tlb->mm, pmd));
		atomic_long_dec(&tlb->mm->nr_ptes);
		spin_unlock(ptl);
		tlb_remove_page(tlb, page);
1693
	}
1694
	return 1;
1695 1696
}

1697
bool move_huge_pmd(struct vm_area_struct *vma, unsigned long old_addr,
1698 1699 1700
		  unsigned long new_addr, unsigned long old_end,
		  pmd_t *old_pmd, pmd_t *new_pmd)
{
1701
	spinlock_t *old_ptl, *new_ptl;
1702 1703 1704 1705 1706
	pmd_t pmd;
	struct mm_struct *mm = vma->vm_mm;

	if ((old_addr & ~HPAGE_PMD_MASK) ||
	    (new_addr & ~HPAGE_PMD_MASK) ||
1707
	    old_end - old_addr < HPAGE_PMD_SIZE)
1708
		return false;
1709 1710 1711 1712 1713 1714 1715

	/*
	 * The destination pmd shouldn't be established, free_pgtables()
	 * should have release it.
	 */
	if (WARN_ON(!pmd_none(*new_pmd))) {
		VM_BUG_ON(pmd_trans_huge(*new_pmd));
1716
		return false;
1717 1718
	}

1719 1720 1721 1722
	/*
	 * We don't have to worry about the ordering of src and dst
	 * ptlocks because exclusive mmap_sem prevents deadlock.
	 */
1723 1724
	old_ptl = __pmd_trans_huge_lock(old_pmd, vma);
	if (old_ptl) {
1725 1726 1727
		new_ptl = pmd_lockptr(mm, new_pmd);
		if (new_ptl != old_ptl)
			spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
1728
		pmd = pmdp_huge_get_and_clear(mm, old_addr, old_pmd);
1729
		VM_BUG_ON(!pmd_none(*new_pmd));
1730

1731 1732
		if (pmd_move_must_withdraw(new_ptl, old_ptl) &&
				vma_is_anonymous(vma)) {
1733
			pgtable_t pgtable;
1734 1735 1736
			pgtable = pgtable_trans_huge_withdraw(mm, old_pmd);
			pgtable_trans_huge_deposit(mm, new_pmd, pgtable);
		}
1737 1738 1739
		set_pmd_at(mm, new_addr, new_pmd, pmd_mksoft_dirty(pmd));
		if (new_ptl != old_ptl)
			spin_unlock(new_ptl);
1740
		spin_unlock(old_ptl);
1741
		return true;
1742
	}
1743
	return false;
1744 1745
}

1746 1747 1748 1749 1750 1751
/*
 * Returns
 *  - 0 if PMD could not be locked
 *  - 1 if PMD was locked but protections unchange and TLB flush unnecessary
 *  - HPAGE_PMD_NR is protections changed and TLB flush necessary
 */
1752
int change_huge_pmd(struct vm_area_struct *vma, pmd_t *pmd,
1753
		unsigned long addr, pgprot_t newprot, int prot_numa)
1754 1755
{
	struct mm_struct *mm = vma->vm_mm;
1756
	spinlock_t *ptl;
1757 1758
	int ret = 0;

1759 1760
	ptl = __pmd_trans_huge_lock(pmd, vma);
	if (ptl) {
1761
		pmd_t entry;
1762
		bool preserve_write = prot_numa && pmd_write(*pmd);
1763
		ret = 1;
1764 1765 1766 1767 1768 1769 1770 1771

		/*
		 * Avoid trapping faults against the zero page. The read-only
		 * data is likely to be read-cached on the local CPU and
		 * local/remote hits to the zero page are not interesting.
		 */
		if (prot_numa && is_huge_zero_pmd(*pmd)) {
			spin_unlock(ptl);
1772
			return ret;
1773 1774
		}

1775
		if (!prot_numa || !pmd_protnone(*pmd)) {
1776
			entry = pmdp_huge_get_and_clear_notify(mm, addr, pmd);
1777
			entry = pmd_modify(entry, newprot);
1778 1779
			if (preserve_write)
				entry = pmd_mkwrite(entry);
1780 1781
			ret = HPAGE_PMD_NR;
			set_pmd_at(mm, addr, pmd, entry);
1782
			BUG_ON(!preserve_write && pmd_write(entry));
1783
		}
1784
		spin_unlock(ptl);
1785 1786 1787 1788 1789 1790
	}

	return ret;
}

/*
1791
 * Returns true if a given pmd maps a thp, false otherwise.
1792
 *
1793 1794
 * Note that if it returns true, this routine returns without unlocking page
 * table lock. So callers must unlock it.
1795
 */
1796
spinlock_t *__pmd_trans_huge_lock(pmd_t *pmd, struct vm_area_struct *vma)
1797
{
1798 1799
	spinlock_t *ptl;
	ptl = pmd_lock(vma->vm_mm, pmd);
1800
	if (likely(pmd_trans_huge(*pmd) || pmd_devmap(*pmd)))
1801 1802 1803
		return ptl;
	spin_unlock(ptl);
	return NULL;
1804 1805
}

1806
#define VM_NO_THP (VM_SPECIAL | VM_HUGETLB | VM_SHARED | VM_MAYSHARE)
1807

1808 1809
int hugepage_madvise(struct vm_area_struct *vma,
		     unsigned long *vm_flags, int advice)
A
Andrea Arcangeli 已提交
1810
{
A
Andrea Arcangeli 已提交
1811 1812
	switch (advice) {
	case MADV_HUGEPAGE:
1813 1814 1815 1816 1817 1818 1819 1820 1821
#ifdef CONFIG_S390
		/*
		 * qemu blindly sets MADV_HUGEPAGE on all allocations, but s390
		 * can't handle this properly after s390_enable_sie, so we simply
		 * ignore the madvise to prevent qemu from causing a SIGSEGV.
		 */
		if (mm_has_pgste(vma->vm_mm))
			return 0;
#endif
A
Andrea Arcangeli 已提交
1822 1823 1824
		/*
		 * Be somewhat over-protective like KSM for now!
		 */
1825
		if (*vm_flags & VM_NO_THP)
A
Andrea Arcangeli 已提交
1826 1827 1828
			return -EINVAL;
		*vm_flags &= ~VM_NOHUGEPAGE;
		*vm_flags |= VM_HUGEPAGE;
1829 1830 1831 1832 1833
		/*
		 * If the vma become good for khugepaged to scan,
		 * register it here without waiting a page fault that
		 * may not happen any time soon.
		 */
1834
		if (unlikely(khugepaged_enter_vma_merge(vma, *vm_flags)))
1835
			return -ENOMEM;
A
Andrea Arcangeli 已提交
1836 1837 1838 1839 1840
		break;
	case MADV_NOHUGEPAGE:
		/*
		 * Be somewhat over-protective like KSM for now!
		 */
1841
		if (*vm_flags & VM_NO_THP)
A
Andrea Arcangeli 已提交
1842 1843 1844
			return -EINVAL;
		*vm_flags &= ~VM_HUGEPAGE;
		*vm_flags |= VM_NOHUGEPAGE;
1845 1846 1847 1848 1849
		/*
		 * Setting VM_NOHUGEPAGE will prevent khugepaged from scanning
		 * this vma even if we leave the mm registered in khugepaged if
		 * it got registered before VM_NOHUGEPAGE was set.
		 */
A
Andrea Arcangeli 已提交
1850 1851
		break;
	}
A
Andrea Arcangeli 已提交
1852 1853 1854 1855

	return 0;
}

A
Andrea Arcangeli 已提交
1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866
static int __init khugepaged_slab_init(void)
{
	mm_slot_cache = kmem_cache_create("khugepaged_mm_slot",
					  sizeof(struct mm_slot),
					  __alignof__(struct mm_slot), 0, NULL);
	if (!mm_slot_cache)
		return -ENOMEM;

	return 0;
}

1867 1868 1869 1870 1871
static void __init khugepaged_slab_exit(void)
{
	kmem_cache_destroy(mm_slot_cache);
}

A
Andrea Arcangeli 已提交
1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887
static inline struct mm_slot *alloc_mm_slot(void)
{
	if (!mm_slot_cache)	/* initialization failed */
		return NULL;
	return kmem_cache_zalloc(mm_slot_cache, GFP_KERNEL);
}

static inline void free_mm_slot(struct mm_slot *mm_slot)
{
	kmem_cache_free(mm_slot_cache, mm_slot);
}

static struct mm_slot *get_mm_slot(struct mm_struct *mm)
{
	struct mm_slot *mm_slot;

1888
	hash_for_each_possible(mm_slots_hash, mm_slot, hash, (unsigned long)mm)
A
Andrea Arcangeli 已提交
1889 1890
		if (mm == mm_slot->mm)
			return mm_slot;
1891

A
Andrea Arcangeli 已提交
1892 1893 1894 1895 1896 1897 1898
	return NULL;
}

static void insert_to_mm_slots_hash(struct mm_struct *mm,
				    struct mm_slot *mm_slot)
{
	mm_slot->mm = mm;
1899
	hash_add(mm_slots_hash, &mm_slot->hash, (long)mm);
A
Andrea Arcangeli 已提交
1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916
}

static inline int khugepaged_test_exit(struct mm_struct *mm)
{
	return atomic_read(&mm->mm_users) == 0;
}

int __khugepaged_enter(struct mm_struct *mm)
{
	struct mm_slot *mm_slot;
	int wakeup;

	mm_slot = alloc_mm_slot();
	if (!mm_slot)
		return -ENOMEM;

	/* __khugepaged_exit() must not run from under us */
S
Sasha Levin 已提交
1917
	VM_BUG_ON_MM(khugepaged_test_exit(mm), mm);
A
Andrea Arcangeli 已提交
1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939
	if (unlikely(test_and_set_bit(MMF_VM_HUGEPAGE, &mm->flags))) {
		free_mm_slot(mm_slot);
		return 0;
	}

	spin_lock(&khugepaged_mm_lock);
	insert_to_mm_slots_hash(mm, mm_slot);
	/*
	 * Insert just behind the scanning cursor, to let the area settle
	 * down a little.
	 */
	wakeup = list_empty(&khugepaged_scan.mm_head);
	list_add_tail(&mm_slot->mm_node, &khugepaged_scan.mm_head);
	spin_unlock(&khugepaged_mm_lock);

	atomic_inc(&mm->mm_count);
	if (wakeup)
		wake_up_interruptible(&khugepaged_wait);

	return 0;
}

1940 1941
int khugepaged_enter_vma_merge(struct vm_area_struct *vma,
			       unsigned long vm_flags)
A
Andrea Arcangeli 已提交
1942 1943 1944 1945 1946 1947 1948 1949
{
	unsigned long hstart, hend;
	if (!vma->anon_vma)
		/*
		 * Not yet faulted in so we will register later in the
		 * page fault if needed.
		 */
		return 0;
1950
	if (vma->vm_ops || (vm_flags & VM_NO_THP))
A
Andrea Arcangeli 已提交
1951 1952 1953 1954 1955
		/* khugepaged not yet working on file or special mappings */
		return 0;
	hstart = (vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK;
	hend = vma->vm_end & HPAGE_PMD_MASK;
	if (hstart < hend)
1956
		return khugepaged_enter(vma, vm_flags);
A
Andrea Arcangeli 已提交
1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967
	return 0;
}

void __khugepaged_exit(struct mm_struct *mm)
{
	struct mm_slot *mm_slot;
	int free = 0;

	spin_lock(&khugepaged_mm_lock);
	mm_slot = get_mm_slot(mm);
	if (mm_slot && khugepaged_scan.mm_slot != mm_slot) {
1968
		hash_del(&mm_slot->hash);
A
Andrea Arcangeli 已提交
1969 1970 1971
		list_del(&mm_slot->mm_node);
		free = 1;
	}
1972
	spin_unlock(&khugepaged_mm_lock);
A
Andrea Arcangeli 已提交
1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988

	if (free) {
		clear_bit(MMF_VM_HUGEPAGE, &mm->flags);
		free_mm_slot(mm_slot);
		mmdrop(mm);
	} else if (mm_slot) {
		/*
		 * This is required to serialize against
		 * khugepaged_test_exit() (which is guaranteed to run
		 * under mmap sem read mode). Stop here (after we
		 * return all pagetables will be destroyed) until
		 * khugepaged has finished working on the pagetables
		 * under the mmap_sem.
		 */
		down_write(&mm->mmap_sem);
		up_write(&mm->mmap_sem);
1989
	}
A
Andrea Arcangeli 已提交
1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
}

static void release_pte_page(struct page *page)
{
	/* 0 stands for page_is_file_cache(page) == false */
	dec_zone_page_state(page, NR_ISOLATED_ANON + 0);
	unlock_page(page);
	putback_lru_page(page);
}

static void release_pte_pages(pte_t *pte, pte_t *_pte)
{
	while (--_pte >= pte) {
		pte_t pteval = *_pte;
2004
		if (!pte_none(pteval) && !is_zero_pfn(pte_pfn(pteval)))
A
Andrea Arcangeli 已提交
2005 2006 2007 2008 2009 2010 2011 2012
			release_pte_page(pte_page(pteval));
	}
}

static int __collapse_huge_page_isolate(struct vm_area_struct *vma,
					unsigned long address,
					pte_t *pte)
{
2013
	struct page *page = NULL;
A
Andrea Arcangeli 已提交
2014
	pte_t *_pte;
2015
	int none_or_zero = 0, result = 0;
2016
	bool referenced = false, writable = false;
2017

A
Andrea Arcangeli 已提交
2018 2019 2020
	for (_pte = pte; _pte < pte+HPAGE_PMD_NR;
	     _pte++, address += PAGE_SIZE) {
		pte_t pteval = *_pte;
2021 2022
		if (pte_none(pteval) || (pte_present(pteval) &&
				is_zero_pfn(pte_pfn(pteval)))) {
2023
			if (!userfaultfd_armed(vma) &&
2024
			    ++none_or_zero <= khugepaged_max_ptes_none) {
A
Andrea Arcangeli 已提交
2025
				continue;
2026 2027
			} else {
				result = SCAN_EXCEED_NONE_PTE;
A
Andrea Arcangeli 已提交
2028
				goto out;
2029
			}
A
Andrea Arcangeli 已提交
2030
		}
2031 2032
		if (!pte_present(pteval)) {
			result = SCAN_PTE_NON_PRESENT;
A
Andrea Arcangeli 已提交
2033
			goto out;
2034
		}
A
Andrea Arcangeli 已提交
2035
		page = vm_normal_page(vma, address, pteval);
2036 2037
		if (unlikely(!page)) {
			result = SCAN_PAGE_NULL;
A
Andrea Arcangeli 已提交
2038
			goto out;
2039
		}
2040

2041 2042 2043
		VM_BUG_ON_PAGE(PageCompound(page), page);
		VM_BUG_ON_PAGE(!PageAnon(page), page);
		VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
A
Andrea Arcangeli 已提交
2044 2045 2046 2047 2048 2049 2050

		/*
		 * We can do it before isolate_lru_page because the
		 * page can't be freed from under us. NOTE: PG_lock
		 * is needed to serialize against split_huge_page
		 * when invoked from the VM.
		 */
2051 2052
		if (!trylock_page(page)) {
			result = SCAN_PAGE_LOCK;
A
Andrea Arcangeli 已提交
2053
			goto out;
2054
		}
2055 2056 2057 2058 2059 2060 2061 2062

		/*
		 * cannot use mapcount: can't collapse if there's a gup pin.
		 * The page must only be referenced by the scanned process
		 * and page swap cache.
		 */
		if (page_count(page) != 1 + !!PageSwapCache(page)) {
			unlock_page(page);
2063
			result = SCAN_PAGE_COUNT;
2064 2065 2066 2067 2068
			goto out;
		}
		if (pte_write(pteval)) {
			writable = true;
		} else {
2069 2070
			if (PageSwapCache(page) &&
			    !reuse_swap_page(page, NULL)) {
2071
				unlock_page(page);
2072
				result = SCAN_SWAP_CACHE_PAGE;
2073 2074 2075 2076 2077 2078 2079 2080
				goto out;
			}
			/*
			 * Page is not in the swap cache. It can be collapsed
			 * into a THP.
			 */
		}

A
Andrea Arcangeli 已提交
2081 2082 2083 2084 2085 2086
		/*
		 * Isolate the page to avoid collapsing an hugepage
		 * currently in use by the VM.
		 */
		if (isolate_lru_page(page)) {
			unlock_page(page);
2087
			result = SCAN_DEL_PAGE_LRU;
A
Andrea Arcangeli 已提交
2088 2089 2090 2091
			goto out;
		}
		/* 0 stands for page_is_file_cache(page) == false */
		inc_zone_page_state(page, NR_ISOLATED_ANON + 0);
2092 2093
		VM_BUG_ON_PAGE(!PageLocked(page), page);
		VM_BUG_ON_PAGE(PageLRU(page), page);
A
Andrea Arcangeli 已提交
2094 2095

		/* If there is no mapped pte young don't collapse the page */
2096 2097
		if (pte_young(pteval) ||
		    page_is_young(page) || PageReferenced(page) ||
A
Andrea Arcangeli 已提交
2098
		    mmu_notifier_test_young(vma->vm_mm, address))
2099
			referenced = true;
A
Andrea Arcangeli 已提交
2100
	}
2101 2102 2103
	if (likely(writable)) {
		if (likely(referenced)) {
			result = SCAN_SUCCEED;
2104
			trace_mm_collapse_huge_page_isolate(page, none_or_zero,
2105 2106 2107 2108 2109 2110 2111
							    referenced, writable, result);
			return 1;
		}
	} else {
		result = SCAN_PAGE_RO;
	}

A
Andrea Arcangeli 已提交
2112
out:
2113
	release_pte_pages(pte, _pte);
2114
	trace_mm_collapse_huge_page_isolate(page, none_or_zero,
2115
					    referenced, writable, result);
2116
	return 0;
A
Andrea Arcangeli 已提交
2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128
}

static void __collapse_huge_page_copy(pte_t *pte, struct page *page,
				      struct vm_area_struct *vma,
				      unsigned long address,
				      spinlock_t *ptl)
{
	pte_t *_pte;
	for (_pte = pte; _pte < pte+HPAGE_PMD_NR; _pte++) {
		pte_t pteval = *_pte;
		struct page *src_page;

2129
		if (pte_none(pteval) || is_zero_pfn(pte_pfn(pteval))) {
A
Andrea Arcangeli 已提交
2130 2131
			clear_user_highpage(page, address);
			add_mm_counter(vma->vm_mm, MM_ANONPAGES, 1);
2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143
			if (is_zero_pfn(pte_pfn(pteval))) {
				/*
				 * ptl mostly unnecessary.
				 */
				spin_lock(ptl);
				/*
				 * paravirt calls inside pte_clear here are
				 * superfluous.
				 */
				pte_clear(vma->vm_mm, address, _pte);
				spin_unlock(ptl);
			}
A
Andrea Arcangeli 已提交
2144 2145 2146
		} else {
			src_page = pte_page(pteval);
			copy_user_highpage(page, src_page, address, vma);
2147
			VM_BUG_ON_PAGE(page_mapcount(src_page) != 1, src_page);
A
Andrea Arcangeli 已提交
2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159
			release_pte_page(src_page);
			/*
			 * ptl mostly unnecessary, but preempt has to
			 * be disabled to update the per-cpu stats
			 * inside page_remove_rmap().
			 */
			spin_lock(ptl);
			/*
			 * paravirt calls inside pte_clear here are
			 * superfluous.
			 */
			pte_clear(vma->vm_mm, address, _pte);
2160
			page_remove_rmap(src_page, false);
A
Andrea Arcangeli 已提交
2161 2162 2163 2164 2165 2166 2167 2168 2169
			spin_unlock(ptl);
			free_page_and_swap_cache(src_page);
		}

		address += PAGE_SIZE;
		page++;
	}
}

2170
static void khugepaged_alloc_sleep(void)
A
Andrea Arcangeli 已提交
2171
{
2172 2173 2174 2175 2176 2177
	DEFINE_WAIT(wait);

	add_wait_queue(&khugepaged_wait, &wait);
	freezable_schedule_timeout_interruptible(
		msecs_to_jiffies(khugepaged_alloc_sleep_millisecs));
	remove_wait_queue(&khugepaged_wait, &wait);
2178
}
A
Andrea Arcangeli 已提交
2179

2180 2181
static int khugepaged_node_load[MAX_NUMNODES];

2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205
static bool khugepaged_scan_abort(int nid)
{
	int i;

	/*
	 * If zone_reclaim_mode is disabled, then no extra effort is made to
	 * allocate memory locally.
	 */
	if (!zone_reclaim_mode)
		return false;

	/* If there is a count for this node already, it must be acceptable */
	if (khugepaged_node_load[nid])
		return false;

	for (i = 0; i < MAX_NUMNODES; i++) {
		if (!khugepaged_node_load[i])
			continue;
		if (node_distance(nid, i) > RECLAIM_DISTANCE)
			return true;
	}
	return false;
}

2206
#ifdef CONFIG_NUMA
2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231
static int khugepaged_find_target_node(void)
{
	static int last_khugepaged_target_node = NUMA_NO_NODE;
	int nid, target_node = 0, max_value = 0;

	/* find first node with max normal pages hit */
	for (nid = 0; nid < MAX_NUMNODES; nid++)
		if (khugepaged_node_load[nid] > max_value) {
			max_value = khugepaged_node_load[nid];
			target_node = nid;
		}

	/* do some balance if several nodes have the same hit record */
	if (target_node <= last_khugepaged_target_node)
		for (nid = last_khugepaged_target_node + 1; nid < MAX_NUMNODES;
				nid++)
			if (max_value == khugepaged_node_load[nid]) {
				target_node = nid;
				break;
			}

	last_khugepaged_target_node = target_node;
	return target_node;
}

2232 2233 2234 2235 2236 2237 2238
static bool khugepaged_prealloc_page(struct page **hpage, bool *wait)
{
	if (IS_ERR(*hpage)) {
		if (!*wait)
			return false;

		*wait = false;
2239
		*hpage = NULL;
2240 2241 2242 2243 2244 2245 2246 2247 2248
		khugepaged_alloc_sleep();
	} else if (*hpage) {
		put_page(*hpage);
		*hpage = NULL;
	}

	return true;
}

2249 2250
static struct page *
khugepaged_alloc_page(struct page **hpage, gfp_t gfp, struct mm_struct *mm,
2251
		       unsigned long address, int node)
2252
{
2253
	VM_BUG_ON_PAGE(*hpage, *hpage);
2254

2255
	/*
2256 2257 2258 2259
	 * Before allocating the hugepage, release the mmap_sem read lock.
	 * The allocation can take potentially a long time if it involves
	 * sync compaction, and we do not need to hold the mmap_sem during
	 * that. We will recheck the vma after taking it again in write mode.
2260
	 */
2261 2262
	up_read(&mm->mmap_sem);

2263
	*hpage = __alloc_pages_node(node, gfp, HPAGE_PMD_ORDER);
2264
	if (unlikely(!*hpage)) {
2265
		count_vm_event(THP_COLLAPSE_ALLOC_FAILED);
2266
		*hpage = ERR_PTR(-ENOMEM);
2267
		return NULL;
2268
	}
2269

2270
	prep_transhuge_page(*hpage);
2271
	count_vm_event(THP_COLLAPSE_ALLOC);
2272 2273 2274
	return *hpage;
}
#else
2275 2276 2277 2278 2279
static int khugepaged_find_target_node(void)
{
	return 0;
}

2280
static inline struct page *alloc_khugepaged_hugepage(void)
2281
{
2282 2283
	struct page *page;

2284 2285
	page = alloc_pages(alloc_hugepage_khugepaged_gfpmask(),
			   HPAGE_PMD_ORDER);
2286 2287 2288
	if (page)
		prep_transhuge_page(page);
	return page;
2289 2290
}

2291 2292 2293 2294 2295
static struct page *khugepaged_alloc_hugepage(bool *wait)
{
	struct page *hpage;

	do {
2296
		hpage = alloc_khugepaged_hugepage();
2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321
		if (!hpage) {
			count_vm_event(THP_COLLAPSE_ALLOC_FAILED);
			if (!*wait)
				return NULL;

			*wait = false;
			khugepaged_alloc_sleep();
		} else
			count_vm_event(THP_COLLAPSE_ALLOC);
	} while (unlikely(!hpage) && likely(khugepaged_enabled()));

	return hpage;
}

static bool khugepaged_prealloc_page(struct page **hpage, bool *wait)
{
	if (!*hpage)
		*hpage = khugepaged_alloc_hugepage(wait);

	if (unlikely(!*hpage))
		return false;

	return true;
}

2322 2323
static struct page *
khugepaged_alloc_page(struct page **hpage, gfp_t gfp, struct mm_struct *mm,
2324
		       unsigned long address, int node)
2325 2326 2327
{
	up_read(&mm->mmap_sem);
	VM_BUG_ON(!*hpage);
2328

2329 2330
	return  *hpage;
}
2331 2332
#endif

B
Bob Liu 已提交
2333 2334 2335 2336 2337 2338 2339 2340 2341
static bool hugepage_vma_check(struct vm_area_struct *vma)
{
	if ((!(vma->vm_flags & VM_HUGEPAGE) && !khugepaged_always()) ||
	    (vma->vm_flags & VM_NOHUGEPAGE))
		return false;
	if (!vma->anon_vma || vma->vm_ops)
		return false;
	if (is_vma_temporary_stack(vma))
		return false;
2342
	return !(vma->vm_flags & VM_NO_THP);
B
Bob Liu 已提交
2343 2344
}

2345 2346 2347 2348 2349 2350 2351 2352 2353 2354
static void collapse_huge_page(struct mm_struct *mm,
				   unsigned long address,
				   struct page **hpage,
				   struct vm_area_struct *vma,
				   int node)
{
	pmd_t *pmd, _pmd;
	pte_t *pte;
	pgtable_t pgtable;
	struct page *new_page;
2355
	spinlock_t *pmd_ptl, *pte_ptl;
2356
	int isolated = 0, result = 0;
2357
	unsigned long hstart, hend;
2358
	struct mem_cgroup *memcg;
2359 2360
	unsigned long mmun_start;	/* For mmu_notifiers */
	unsigned long mmun_end;		/* For mmu_notifiers */
2361
	gfp_t gfp;
2362 2363 2364

	VM_BUG_ON(address & ~HPAGE_PMD_MASK);

2365
	/* Only allocate from the target node */
2366
	gfp = alloc_hugepage_khugepaged_gfpmask() | __GFP_OTHER_NODE | __GFP_THISNODE;
2367

2368
	/* release the mmap_sem read lock. */
2369
	new_page = khugepaged_alloc_page(hpage, gfp, mm, address, node);
2370 2371 2372 2373
	if (!new_page) {
		result = SCAN_ALLOC_HUGE_PAGE_FAIL;
		goto out_nolock;
	}
2374

2375
	if (unlikely(mem_cgroup_try_charge(new_page, mm, gfp, &memcg, true))) {
2376 2377 2378
		result = SCAN_CGROUP_CHARGE_FAIL;
		goto out_nolock;
	}
A
Andrea Arcangeli 已提交
2379 2380 2381 2382 2383 2384 2385

	/*
	 * Prevent all access to pagetables with the exception of
	 * gup_fast later hanlded by the ptep_clear_flush and the VM
	 * handled by the anon_vma lock + PG_lock.
	 */
	down_write(&mm->mmap_sem);
2386 2387
	if (unlikely(khugepaged_test_exit(mm))) {
		result = SCAN_ANY_PROCESS;
A
Andrea Arcangeli 已提交
2388
		goto out;
2389
	}
A
Andrea Arcangeli 已提交
2390 2391

	vma = find_vma(mm, address);
2392 2393
	if (!vma) {
		result = SCAN_VMA_NULL;
2394
		goto out;
2395
	}
A
Andrea Arcangeli 已提交
2396 2397
	hstart = (vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK;
	hend = vma->vm_end & HPAGE_PMD_MASK;
2398 2399
	if (address < hstart || address + HPAGE_PMD_SIZE > hend) {
		result = SCAN_ADDRESS_RANGE;
A
Andrea Arcangeli 已提交
2400
		goto out;
2401 2402 2403
	}
	if (!hugepage_vma_check(vma)) {
		result = SCAN_VMA_CHECK;
2404
		goto out;
2405
	}
B
Bob Liu 已提交
2406
	pmd = mm_find_pmd(mm, address);
2407 2408
	if (!pmd) {
		result = SCAN_PMD_NULL;
A
Andrea Arcangeli 已提交
2409
		goto out;
2410
	}
A
Andrea Arcangeli 已提交
2411

2412
	anon_vma_lock_write(vma->anon_vma);
A
Andrea Arcangeli 已提交
2413 2414

	pte = pte_offset_map(pmd, address);
2415
	pte_ptl = pte_lockptr(mm, pmd);
A
Andrea Arcangeli 已提交
2416

2417 2418 2419
	mmun_start = address;
	mmun_end   = address + HPAGE_PMD_SIZE;
	mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
2420
	pmd_ptl = pmd_lock(mm, pmd); /* probably unnecessary */
A
Andrea Arcangeli 已提交
2421 2422 2423 2424 2425 2426
	/*
	 * After this gup_fast can't run anymore. This also removes
	 * any huge TLB entry from the CPU so we won't allow
	 * huge and small TLB entries for the same virtual address
	 * to avoid the risk of CPU bugs in that area.
	 */
2427
	_pmd = pmdp_collapse_flush(vma, address, pmd);
2428
	spin_unlock(pmd_ptl);
2429
	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
A
Andrea Arcangeli 已提交
2430

2431
	spin_lock(pte_ptl);
A
Andrea Arcangeli 已提交
2432
	isolated = __collapse_huge_page_isolate(vma, address, pte);
2433
	spin_unlock(pte_ptl);
A
Andrea Arcangeli 已提交
2434 2435

	if (unlikely(!isolated)) {
2436
		pte_unmap(pte);
2437
		spin_lock(pmd_ptl);
A
Andrea Arcangeli 已提交
2438
		BUG_ON(!pmd_none(*pmd));
2439 2440 2441 2442 2443 2444
		/*
		 * We can only use set_pmd_at when establishing
		 * hugepmds and never for establishing regular pmds that
		 * points to regular pagetables. Use pmd_populate for that
		 */
		pmd_populate(mm, pmd, pmd_pgtable(_pmd));
2445
		spin_unlock(pmd_ptl);
2446
		anon_vma_unlock_write(vma->anon_vma);
2447
		result = SCAN_FAIL;
2448
		goto out;
A
Andrea Arcangeli 已提交
2449 2450 2451 2452 2453 2454
	}

	/*
	 * All pages are isolated and locked so anon_vma rmap
	 * can't run anymore.
	 */
2455
	anon_vma_unlock_write(vma->anon_vma);
A
Andrea Arcangeli 已提交
2456

2457
	__collapse_huge_page_copy(pte, new_page, vma, address, pte_ptl);
2458
	pte_unmap(pte);
A
Andrea Arcangeli 已提交
2459 2460 2461
	__SetPageUptodate(new_page);
	pgtable = pmd_pgtable(_pmd);

2462 2463
	_pmd = mk_huge_pmd(new_page, vma->vm_page_prot);
	_pmd = maybe_pmd_mkwrite(pmd_mkdirty(_pmd), vma);
A
Andrea Arcangeli 已提交
2464 2465 2466 2467 2468 2469 2470 2471

	/*
	 * spin_lock() below is not the equivalent of smp_wmb(), so
	 * this is needed to avoid the copy_huge_page writes to become
	 * visible after the set_pmd_at() write.
	 */
	smp_wmb();

2472
	spin_lock(pmd_ptl);
A
Andrea Arcangeli 已提交
2473
	BUG_ON(!pmd_none(*pmd));
2474
	page_add_new_anon_rmap(new_page, vma, address, true);
2475
	mem_cgroup_commit_charge(new_page, memcg, false, true);
2476
	lru_cache_add_active_or_unevictable(new_page, vma);
2477
	pgtable_trans_huge_deposit(mm, pmd, pgtable);
A
Andrea Arcangeli 已提交
2478
	set_pmd_at(mm, address, pmd, _pmd);
2479
	update_mmu_cache_pmd(vma, address, pmd);
2480
	spin_unlock(pmd_ptl);
A
Andrea Arcangeli 已提交
2481 2482

	*hpage = NULL;
2483

A
Andrea Arcangeli 已提交
2484
	khugepaged_pages_collapsed++;
2485
	result = SCAN_SUCCEED;
2486
out_up_write:
A
Andrea Arcangeli 已提交
2487
	up_write(&mm->mmap_sem);
2488
	trace_mm_collapse_huge_page(mm, isolated, result);
2489 2490
	return;

2491 2492 2493
out_nolock:
	trace_mm_collapse_huge_page(mm, isolated, result);
	return;
2494
out:
2495
	mem_cgroup_cancel_charge(new_page, memcg, true);
2496
	goto out_up_write;
A
Andrea Arcangeli 已提交
2497 2498 2499 2500 2501 2502 2503 2504 2505
}

static int khugepaged_scan_pmd(struct mm_struct *mm,
			       struct vm_area_struct *vma,
			       unsigned long address,
			       struct page **hpage)
{
	pmd_t *pmd;
	pte_t *pte, *_pte;
2506 2507
	int ret = 0, none_or_zero = 0, result = 0;
	struct page *page = NULL;
A
Andrea Arcangeli 已提交
2508 2509
	unsigned long _address;
	spinlock_t *ptl;
D
David Rientjes 已提交
2510
	int node = NUMA_NO_NODE;
2511
	bool writable = false, referenced = false;
A
Andrea Arcangeli 已提交
2512 2513 2514

	VM_BUG_ON(address & ~HPAGE_PMD_MASK);

B
Bob Liu 已提交
2515
	pmd = mm_find_pmd(mm, address);
2516 2517
	if (!pmd) {
		result = SCAN_PMD_NULL;
A
Andrea Arcangeli 已提交
2518
		goto out;
2519
	}
A
Andrea Arcangeli 已提交
2520

2521
	memset(khugepaged_node_load, 0, sizeof(khugepaged_node_load));
A
Andrea Arcangeli 已提交
2522 2523 2524 2525
	pte = pte_offset_map_lock(mm, pmd, address, &ptl);
	for (_address = address, _pte = pte; _pte < pte+HPAGE_PMD_NR;
	     _pte++, _address += PAGE_SIZE) {
		pte_t pteval = *_pte;
2526
		if (pte_none(pteval) || is_zero_pfn(pte_pfn(pteval))) {
2527
			if (!userfaultfd_armed(vma) &&
2528
			    ++none_or_zero <= khugepaged_max_ptes_none) {
A
Andrea Arcangeli 已提交
2529
				continue;
2530 2531
			} else {
				result = SCAN_EXCEED_NONE_PTE;
A
Andrea Arcangeli 已提交
2532
				goto out_unmap;
2533
			}
A
Andrea Arcangeli 已提交
2534
		}
2535 2536
		if (!pte_present(pteval)) {
			result = SCAN_PTE_NON_PRESENT;
A
Andrea Arcangeli 已提交
2537
			goto out_unmap;
2538
		}
2539 2540 2541
		if (pte_write(pteval))
			writable = true;

A
Andrea Arcangeli 已提交
2542
		page = vm_normal_page(vma, _address, pteval);
2543 2544
		if (unlikely(!page)) {
			result = SCAN_PAGE_NULL;
A
Andrea Arcangeli 已提交
2545
			goto out_unmap;
2546
		}
2547 2548 2549 2550 2551 2552 2553

		/* TODO: teach khugepaged to collapse THP mapped with pte */
		if (PageCompound(page)) {
			result = SCAN_PAGE_COMPOUND;
			goto out_unmap;
		}

2554
		/*
2555 2556 2557 2558
		 * Record which node the original page is from and save this
		 * information to khugepaged_node_load[].
		 * Khupaged will allocate hugepage from the node has the max
		 * hit record.
2559
		 */
2560
		node = page_to_nid(page);
2561 2562
		if (khugepaged_scan_abort(node)) {
			result = SCAN_SCAN_ABORT;
2563
			goto out_unmap;
2564
		}
2565
		khugepaged_node_load[node]++;
2566
		if (!PageLRU(page)) {
2567
			result = SCAN_PAGE_LRU;
2568 2569 2570 2571
			goto out_unmap;
		}
		if (PageLocked(page)) {
			result = SCAN_PAGE_LOCK;
A
Andrea Arcangeli 已提交
2572
			goto out_unmap;
2573 2574 2575 2576 2577 2578
		}
		if (!PageAnon(page)) {
			result = SCAN_PAGE_ANON;
			goto out_unmap;
		}

2579 2580 2581 2582 2583
		/*
		 * cannot use mapcount: can't collapse if there's a gup pin.
		 * The page must only be referenced by the scanned process
		 * and page swap cache.
		 */
2584 2585
		if (page_count(page) != 1 + !!PageSwapCache(page)) {
			result = SCAN_PAGE_COUNT;
A
Andrea Arcangeli 已提交
2586
			goto out_unmap;
2587
		}
2588 2589
		if (pte_young(pteval) ||
		    page_is_young(page) || PageReferenced(page) ||
A
Andrea Arcangeli 已提交
2590
		    mmu_notifier_test_young(vma->vm_mm, address))
2591
			referenced = true;
A
Andrea Arcangeli 已提交
2592
	}
2593 2594 2595 2596 2597 2598 2599 2600 2601 2602
	if (writable) {
		if (referenced) {
			result = SCAN_SUCCEED;
			ret = 1;
		} else {
			result = SCAN_NO_REFERENCED_PAGE;
		}
	} else {
		result = SCAN_PAGE_RO;
	}
A
Andrea Arcangeli 已提交
2603 2604
out_unmap:
	pte_unmap_unlock(pte, ptl);
2605 2606
	if (ret) {
		node = khugepaged_find_target_node();
2607
		/* collapse_huge_page will return with the mmap_sem released */
2608
		collapse_huge_page(mm, address, hpage, vma, node);
2609
	}
A
Andrea Arcangeli 已提交
2610
out:
2611
	trace_mm_khugepaged_scan_pmd(mm, page, writable, referenced,
2612
				     none_or_zero, result);
A
Andrea Arcangeli 已提交
2613 2614 2615 2616 2617 2618 2619
	return ret;
}

static void collect_mm_slot(struct mm_slot *mm_slot)
{
	struct mm_struct *mm = mm_slot->mm;

2620
	VM_BUG_ON(NR_CPUS != 1 && !spin_is_locked(&khugepaged_mm_lock));
A
Andrea Arcangeli 已提交
2621 2622 2623

	if (khugepaged_test_exit(mm)) {
		/* free mm_slot */
2624
		hash_del(&mm_slot->hash);
A
Andrea Arcangeli 已提交
2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640
		list_del(&mm_slot->mm_node);

		/*
		 * Not strictly needed because the mm exited already.
		 *
		 * clear_bit(MMF_VM_HUGEPAGE, &mm->flags);
		 */

		/* khugepaged_mm_lock actually not necessary for the below */
		free_mm_slot(mm_slot);
		mmdrop(mm);
	}
}

static unsigned int khugepaged_scan_mm_slot(unsigned int pages,
					    struct page **hpage)
2641 2642
	__releases(&khugepaged_mm_lock)
	__acquires(&khugepaged_mm_lock)
A
Andrea Arcangeli 已提交
2643 2644 2645 2646 2647 2648 2649
{
	struct mm_slot *mm_slot;
	struct mm_struct *mm;
	struct vm_area_struct *vma;
	int progress = 0;

	VM_BUG_ON(!pages);
2650
	VM_BUG_ON(NR_CPUS != 1 && !spin_is_locked(&khugepaged_mm_lock));
A
Andrea Arcangeli 已提交
2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677

	if (khugepaged_scan.mm_slot)
		mm_slot = khugepaged_scan.mm_slot;
	else {
		mm_slot = list_entry(khugepaged_scan.mm_head.next,
				     struct mm_slot, mm_node);
		khugepaged_scan.address = 0;
		khugepaged_scan.mm_slot = mm_slot;
	}
	spin_unlock(&khugepaged_mm_lock);

	mm = mm_slot->mm;
	down_read(&mm->mmap_sem);
	if (unlikely(khugepaged_test_exit(mm)))
		vma = NULL;
	else
		vma = find_vma(mm, khugepaged_scan.address);

	progress++;
	for (; vma; vma = vma->vm_next) {
		unsigned long hstart, hend;

		cond_resched();
		if (unlikely(khugepaged_test_exit(mm))) {
			progress++;
			break;
		}
B
Bob Liu 已提交
2678 2679
		if (!hugepage_vma_check(vma)) {
skip:
A
Andrea Arcangeli 已提交
2680 2681 2682 2683 2684
			progress++;
			continue;
		}
		hstart = (vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK;
		hend = vma->vm_end & HPAGE_PMD_MASK;
2685 2686 2687 2688
		if (hstart >= hend)
			goto skip;
		if (khugepaged_scan.address > hend)
			goto skip;
A
Andrea Arcangeli 已提交
2689 2690
		if (khugepaged_scan.address < hstart)
			khugepaged_scan.address = hstart;
2691
		VM_BUG_ON(khugepaged_scan.address & ~HPAGE_PMD_MASK);
A
Andrea Arcangeli 已提交
2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719

		while (khugepaged_scan.address < hend) {
			int ret;
			cond_resched();
			if (unlikely(khugepaged_test_exit(mm)))
				goto breakouterloop;

			VM_BUG_ON(khugepaged_scan.address < hstart ||
				  khugepaged_scan.address + HPAGE_PMD_SIZE >
				  hend);
			ret = khugepaged_scan_pmd(mm, vma,
						  khugepaged_scan.address,
						  hpage);
			/* move to next address */
			khugepaged_scan.address += HPAGE_PMD_SIZE;
			progress += HPAGE_PMD_NR;
			if (ret)
				/* we released mmap_sem so break loop */
				goto breakouterloop_mmap_sem;
			if (progress >= pages)
				goto breakouterloop;
		}
	}
breakouterloop:
	up_read(&mm->mmap_sem); /* exit_mmap will destroy ptes after this */
breakouterloop_mmap_sem:

	spin_lock(&khugepaged_mm_lock);
2720
	VM_BUG_ON(khugepaged_scan.mm_slot != mm_slot);
A
Andrea Arcangeli 已提交
2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755
	/*
	 * Release the current mm_slot if this mm is about to die, or
	 * if we scanned all vmas of this mm.
	 */
	if (khugepaged_test_exit(mm) || !vma) {
		/*
		 * Make sure that if mm_users is reaching zero while
		 * khugepaged runs here, khugepaged_exit will find
		 * mm_slot not pointing to the exiting mm.
		 */
		if (mm_slot->mm_node.next != &khugepaged_scan.mm_head) {
			khugepaged_scan.mm_slot = list_entry(
				mm_slot->mm_node.next,
				struct mm_slot, mm_node);
			khugepaged_scan.address = 0;
		} else {
			khugepaged_scan.mm_slot = NULL;
			khugepaged_full_scans++;
		}

		collect_mm_slot(mm_slot);
	}

	return progress;
}

static int khugepaged_has_work(void)
{
	return !list_empty(&khugepaged_scan.mm_head) &&
		khugepaged_enabled();
}

static int khugepaged_wait_event(void)
{
	return !list_empty(&khugepaged_scan.mm_head) ||
2756
		kthread_should_stop();
A
Andrea Arcangeli 已提交
2757 2758
}

2759
static void khugepaged_do_scan(void)
A
Andrea Arcangeli 已提交
2760
{
2761
	struct page *hpage = NULL;
A
Andrea Arcangeli 已提交
2762 2763
	unsigned int progress = 0, pass_through_head = 0;
	unsigned int pages = khugepaged_pages_to_scan;
2764
	bool wait = true;
A
Andrea Arcangeli 已提交
2765 2766 2767 2768

	barrier(); /* write khugepaged_pages_to_scan to local stack */

	while (progress < pages) {
2769
		if (!khugepaged_prealloc_page(&hpage, &wait))
2770
			break;
2771

2772
		cond_resched();
A
Andrea Arcangeli 已提交
2773

2774
		if (unlikely(kthread_should_stop() || try_to_freeze()))
2775 2776
			break;

A
Andrea Arcangeli 已提交
2777 2778 2779 2780 2781 2782
		spin_lock(&khugepaged_mm_lock);
		if (!khugepaged_scan.mm_slot)
			pass_through_head++;
		if (khugepaged_has_work() &&
		    pass_through_head < 2)
			progress += khugepaged_scan_mm_slot(pages - progress,
2783
							    &hpage);
A
Andrea Arcangeli 已提交
2784 2785 2786 2787 2788
		else
			progress = pages;
		spin_unlock(&khugepaged_mm_lock);
	}

2789 2790
	if (!IS_ERR_OR_NULL(hpage))
		put_page(hpage);
2791 2792
}

2793 2794 2795 2796 2797 2798
static bool khugepaged_should_wakeup(void)
{
	return kthread_should_stop() ||
	       time_after_eq(jiffies, khugepaged_sleep_expire);
}

2799 2800 2801
static void khugepaged_wait_work(void)
{
	if (khugepaged_has_work()) {
2802 2803 2804 2805
		const unsigned long scan_sleep_jiffies =
			msecs_to_jiffies(khugepaged_scan_sleep_millisecs);

		if (!scan_sleep_jiffies)
2806 2807
			return;

2808
		khugepaged_sleep_expire = jiffies + scan_sleep_jiffies;
2809
		wait_event_freezable_timeout(khugepaged_wait,
2810 2811
					     khugepaged_should_wakeup(),
					     scan_sleep_jiffies);
2812 2813 2814 2815 2816 2817 2818
		return;
	}

	if (khugepaged_enabled())
		wait_event_freezable(khugepaged_wait, khugepaged_wait_event());
}

A
Andrea Arcangeli 已提交
2819 2820 2821 2822
static int khugepaged(void *none)
{
	struct mm_slot *mm_slot;

2823
	set_freezable();
2824
	set_user_nice(current, MAX_NICE);
A
Andrea Arcangeli 已提交
2825

X
Xiao Guangrong 已提交
2826 2827 2828 2829
	while (!kthread_should_stop()) {
		khugepaged_do_scan();
		khugepaged_wait_work();
	}
A
Andrea Arcangeli 已提交
2830 2831 2832 2833 2834 2835 2836 2837 2838 2839

	spin_lock(&khugepaged_mm_lock);
	mm_slot = khugepaged_scan.mm_slot;
	khugepaged_scan.mm_slot = NULL;
	if (mm_slot)
		collect_mm_slot(mm_slot);
	spin_unlock(&khugepaged_mm_lock);
	return 0;
}

2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868
static void __split_huge_zero_page_pmd(struct vm_area_struct *vma,
		unsigned long haddr, pmd_t *pmd)
{
	struct mm_struct *mm = vma->vm_mm;
	pgtable_t pgtable;
	pmd_t _pmd;
	int i;

	/* leave pmd empty until pte is filled */
	pmdp_huge_clear_flush_notify(vma, haddr, pmd);

	pgtable = pgtable_trans_huge_withdraw(mm, pmd);
	pmd_populate(mm, &_pmd, pgtable);

	for (i = 0; i < HPAGE_PMD_NR; i++, haddr += PAGE_SIZE) {
		pte_t *pte, entry;
		entry = pfn_pte(my_zero_pfn(haddr), vma->vm_page_prot);
		entry = pte_mkspecial(entry);
		pte = pte_offset_map(&_pmd, haddr);
		VM_BUG_ON(!pte_none(*pte));
		set_pte_at(mm, haddr, pte, entry);
		pte_unmap(pte);
	}
	smp_wmb(); /* make pte visible before pmd */
	pmd_populate(mm, pmd, pgtable);
	put_huge_zero_page();
}

static void __split_huge_pmd_locked(struct vm_area_struct *vma, pmd_t *pmd,
2869
		unsigned long haddr, bool freeze)
2870 2871 2872 2873 2874
{
	struct mm_struct *mm = vma->vm_mm;
	struct page *page;
	pgtable_t pgtable;
	pmd_t _pmd;
2875
	bool young, write, dirty;
2876
	unsigned long addr;
2877 2878 2879 2880 2881
	int i;

	VM_BUG_ON(haddr & ~HPAGE_PMD_MASK);
	VM_BUG_ON_VMA(vma->vm_start > haddr, vma);
	VM_BUG_ON_VMA(vma->vm_end < haddr + HPAGE_PMD_SIZE, vma);
2882
	VM_BUG_ON(!pmd_trans_huge(*pmd) && !pmd_devmap(*pmd));
2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896

	count_vm_event(THP_SPLIT_PMD);

	if (vma_is_dax(vma)) {
		pmd_t _pmd = pmdp_huge_clear_flush_notify(vma, haddr, pmd);
		if (is_huge_zero_pmd(_pmd))
			put_huge_zero_page();
		return;
	} else if (is_huge_zero_pmd(*pmd)) {
		return __split_huge_zero_page_pmd(vma, haddr, pmd);
	}

	page = pmd_page(*pmd);
	VM_BUG_ON_PAGE(!page_count(page), page);
2897
	page_ref_add(page, HPAGE_PMD_NR - 1);
2898 2899
	write = pmd_write(*pmd);
	young = pmd_young(*pmd);
2900
	dirty = pmd_dirty(*pmd);
2901

2902
	pmdp_huge_split_prepare(vma, haddr, pmd);
2903 2904 2905
	pgtable = pgtable_trans_huge_withdraw(mm, pmd);
	pmd_populate(mm, &_pmd, pgtable);

2906
	for (i = 0, addr = haddr; i < HPAGE_PMD_NR; i++, addr += PAGE_SIZE) {
2907 2908 2909 2910 2911 2912
		pte_t entry, *pte;
		/*
		 * Note that NUMA hinting access restrictions are not
		 * transferred to avoid any possibility of altering
		 * permissions across VMAs.
		 */
2913 2914 2915 2916 2917 2918
		if (freeze) {
			swp_entry_t swp_entry;
			swp_entry = make_migration_entry(page + i, write);
			entry = swp_entry_to_pte(swp_entry);
		} else {
			entry = mk_pte(page + i, vma->vm_page_prot);
2919
			entry = maybe_mkwrite(entry, vma);
2920 2921 2922 2923 2924
			if (!write)
				entry = pte_wrprotect(entry);
			if (!young)
				entry = pte_mkold(entry);
		}
2925 2926
		if (dirty)
			SetPageDirty(page + i);
2927
		pte = pte_offset_map(&_pmd, addr);
2928
		BUG_ON(!pte_none(*pte));
2929
		set_pte_at(mm, addr, pte, entry);
2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953
		atomic_inc(&page[i]._mapcount);
		pte_unmap(pte);
	}

	/*
	 * Set PG_double_map before dropping compound_mapcount to avoid
	 * false-negative page_mapped().
	 */
	if (compound_mapcount(page) > 1 && !TestSetPageDoubleMap(page)) {
		for (i = 0; i < HPAGE_PMD_NR; i++)
			atomic_inc(&page[i]._mapcount);
	}

	if (atomic_add_negative(-1, compound_mapcount_ptr(page))) {
		/* Last compound_mapcount is gone. */
		__dec_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES);
		if (TestClearPageDoubleMap(page)) {
			/* No need in mapcount reference anymore */
			for (i = 0; i < HPAGE_PMD_NR; i++)
				atomic_dec(&page[i]._mapcount);
		}
	}

	smp_wmb(); /* make pte visible before pmd */
2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975
	/*
	 * Up to this point the pmd is present and huge and userland has the
	 * whole access to the hugepage during the split (which happens in
	 * place). If we overwrite the pmd with the not-huge version pointing
	 * to the pte here (which of course we could if all CPUs were bug
	 * free), userland could trigger a small page size TLB miss on the
	 * small sized TLB while the hugepage TLB entry is still established in
	 * the huge TLB. Some CPU doesn't like that.
	 * See http://support.amd.com/us/Processor_TechDocs/41322.pdf, Erratum
	 * 383 on page 93. Intel should be safe but is also warns that it's
	 * only safe if the permission and cache attributes of the two entries
	 * loaded in the two TLB is identical (which should be the case here).
	 * But it is generally safer to never allow small and huge TLB entries
	 * for the same virtual address to be loaded simultaneously. So instead
	 * of doing "pmd_populate(); flush_pmd_tlb_range();" we first mark the
	 * current pmd notpresent (atomically because here the pmd_trans_huge
	 * and pmd_trans_splitting must remain set at all times on the pmd
	 * until the split is complete for this pmd), then we flush the SMP TLB
	 * and finally we write the non-huge version of the pmd entry with
	 * pmd_populate.
	 */
	pmdp_invalidate(vma, haddr, pmd);
2976
	pmd_populate(mm, pmd, pgtable);
2977 2978

	if (freeze) {
2979
		for (i = 0; i < HPAGE_PMD_NR; i++) {
2980 2981 2982 2983
			page_remove_rmap(page + i, false);
			put_page(page + i);
		}
	}
2984 2985 2986
}

void __split_huge_pmd(struct vm_area_struct *vma, pmd_t *pmd,
2987
		unsigned long address, bool freeze, struct page *page)
2988 2989 2990 2991 2992 2993 2994
{
	spinlock_t *ptl;
	struct mm_struct *mm = vma->vm_mm;
	unsigned long haddr = address & HPAGE_PMD_MASK;

	mmu_notifier_invalidate_range_start(mm, haddr, haddr + HPAGE_PMD_SIZE);
	ptl = pmd_lock(mm, pmd);
2995 2996 2997 2998 2999 3000 3001 3002 3003

	/*
	 * If caller asks to setup a migration entries, we need a page to check
	 * pmd against. Otherwise we can end up replacing wrong page.
	 */
	VM_BUG_ON(freeze && !page);
	if (page && page != pmd_page(*pmd))
	        goto out;

3004
	if (pmd_trans_huge(*pmd)) {
3005
		page = pmd_page(*pmd);
3006
		if (PageMlocked(page))
3007
			clear_page_mlock(page);
3008
	} else if (!pmd_devmap(*pmd))
3009
		goto out;
3010
	__split_huge_pmd_locked(vma, pmd, haddr, freeze);
3011
out:
3012 3013 3014 3015
	spin_unlock(ptl);
	mmu_notifier_invalidate_range_end(mm, haddr, haddr + HPAGE_PMD_SIZE);
}

3016 3017
void split_huge_pmd_address(struct vm_area_struct *vma, unsigned long address,
		bool freeze, struct page *page)
3018
{
3019 3020
	pgd_t *pgd;
	pud_t *pud;
3021 3022
	pmd_t *pmd;

3023
	pgd = pgd_offset(vma->vm_mm, address);
3024 3025 3026 3027 3028 3029 3030 3031
	if (!pgd_present(*pgd))
		return;

	pud = pud_offset(pgd, address);
	if (!pud_present(*pud))
		return;

	pmd = pmd_offset(pud, address);
3032

3033
	__split_huge_pmd(vma, pmd, address, freeze, page);
3034 3035
}

3036
void vma_adjust_trans_huge(struct vm_area_struct *vma,
3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048
			     unsigned long start,
			     unsigned long end,
			     long adjust_next)
{
	/*
	 * If the new start address isn't hpage aligned and it could
	 * previously contain an hugepage: check if we need to split
	 * an huge pmd.
	 */
	if (start & ~HPAGE_PMD_MASK &&
	    (start & HPAGE_PMD_MASK) >= vma->vm_start &&
	    (start & HPAGE_PMD_MASK) + HPAGE_PMD_SIZE <= vma->vm_end)
3049
		split_huge_pmd_address(vma, start, false, NULL);
3050 3051 3052 3053 3054 3055 3056 3057 3058

	/*
	 * If the new end address isn't hpage aligned and it could
	 * previously contain an hugepage: check if we need to split
	 * an huge pmd.
	 */
	if (end & ~HPAGE_PMD_MASK &&
	    (end & HPAGE_PMD_MASK) >= vma->vm_start &&
	    (end & HPAGE_PMD_MASK) + HPAGE_PMD_SIZE <= vma->vm_end)
3059
		split_huge_pmd_address(vma, end, false, NULL);
3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072

	/*
	 * If we're also updating the vma->vm_next->vm_start, if the new
	 * vm_next->vm_start isn't page aligned and it could previously
	 * contain an hugepage: check if we need to split an huge pmd.
	 */
	if (adjust_next > 0) {
		struct vm_area_struct *next = vma->vm_next;
		unsigned long nstart = next->vm_start;
		nstart += adjust_next << PAGE_SHIFT;
		if (nstart & ~HPAGE_PMD_MASK &&
		    (nstart & HPAGE_PMD_MASK) >= next->vm_start &&
		    (nstart & HPAGE_PMD_MASK) + HPAGE_PMD_SIZE <= next->vm_end)
3073
			split_huge_pmd_address(next, nstart, false, NULL);
3074 3075
	}
}
3076

3077
static void freeze_page(struct page *page)
3078
{
3079 3080 3081
	enum ttu_flags ttu_flags = TTU_MIGRATION | TTU_IGNORE_MLOCK |
		TTU_IGNORE_ACCESS | TTU_RMAP_LOCKED;
	int i, ret;
3082 3083 3084

	VM_BUG_ON_PAGE(!PageHead(page), page);

3085 3086 3087 3088 3089 3090
	/* We only need TTU_SPLIT_HUGE_PMD once */
	ret = try_to_unmap(page, ttu_flags | TTU_SPLIT_HUGE_PMD);
	for (i = 1; !ret && i < HPAGE_PMD_NR; i++) {
		/* Cut short if the page is unmapped */
		if (page_count(page) == 1)
			return;
3091

3092
		ret = try_to_unmap(page + i, ttu_flags);
3093
	}
3094
	VM_BUG_ON(ret);
3095 3096
}

3097
static void unfreeze_page(struct page *page)
3098
{
3099
	int i;
3100

3101 3102
	for (i = 0; i < HPAGE_PMD_NR; i++)
		remove_migration_ptes(page + i, page + i, true);
3103 3104
}

3105
static void __split_huge_page_tail(struct page *head, int tail,
3106 3107 3108 3109
		struct lruvec *lruvec, struct list_head *list)
{
	struct page *page_tail = head + tail;

3110
	VM_BUG_ON_PAGE(atomic_read(&page_tail->_mapcount) != -1, page_tail);
3111
	VM_BUG_ON_PAGE(page_ref_count(page_tail) != 0, page_tail);
3112 3113

	/*
3114
	 * tail_page->_refcount is zero and not changing from under us. But
3115
	 * get_page_unless_zero() may be running from under us on the
3116
	 * tail_page. If we used atomic_set() below instead of atomic_inc(), we
3117 3118 3119 3120 3121
	 * would then run atomic_set() concurrently with
	 * get_page_unless_zero(), and atomic_set() is implemented in C not
	 * using locked ops. spin_unlock on x86 sometime uses locked ops
	 * because of PPro errata 66, 92, so unless somebody can guarantee
	 * atomic_set() here would be safe on all archs (and not only on x86),
3122
	 * it's safer to use atomic_inc().
3123
	 */
3124
	page_ref_inc(page_tail);
3125 3126 3127 3128 3129 3130 3131 3132 3133

	page_tail->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
	page_tail->flags |= (head->flags &
			((1L << PG_referenced) |
			 (1L << PG_swapbacked) |
			 (1L << PG_mlocked) |
			 (1L << PG_uptodate) |
			 (1L << PG_active) |
			 (1L << PG_locked) |
3134 3135
			 (1L << PG_unevictable) |
			 (1L << PG_dirty)));
3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150

	/*
	 * After clearing PageTail the gup refcount can be released.
	 * Page flags also must be visible before we make the page non-compound.
	 */
	smp_wmb();

	clear_compound_head(page_tail);

	if (page_is_young(head))
		set_page_young(page_tail);
	if (page_is_idle(head))
		set_page_idle(page_tail);

	/* ->mapping in first tail page is compound_mapcount */
3151
	VM_BUG_ON_PAGE(tail > 2 && page_tail->mapping != TAIL_MAPPING,
3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164
			page_tail);
	page_tail->mapping = head->mapping;

	page_tail->index = head->index + tail;
	page_cpupid_xchg_last(page_tail, page_cpupid_last(head));
	lru_add_page_tail(head, page_tail, lruvec, list);
}

static void __split_huge_page(struct page *page, struct list_head *list)
{
	struct page *head = compound_head(page);
	struct zone *zone = page_zone(head);
	struct lruvec *lruvec;
3165
	int i;
3166 3167 3168 3169 3170 3171 3172 3173 3174

	/* prevent PageLRU to go away from under us, and freeze lru stats */
	spin_lock_irq(&zone->lru_lock);
	lruvec = mem_cgroup_page_lruvec(head, zone);

	/* complete memcg works before add pages to LRU */
	mem_cgroup_split_huge_fixup(head);

	for (i = HPAGE_PMD_NR - 1; i >= 1; i--)
3175
		__split_huge_page_tail(head, i, lruvec, list);
3176 3177 3178 3179

	ClearPageCompound(head);
	spin_unlock_irq(&zone->lru_lock);

3180
	unfreeze_page(head);
3181 3182 3183 3184 3185 3186 3187 3188 3189 3190 3191 3192 3193 3194 3195 3196 3197 3198

	for (i = 0; i < HPAGE_PMD_NR; i++) {
		struct page *subpage = head + i;
		if (subpage == page)
			continue;
		unlock_page(subpage);

		/*
		 * Subpages may be freed if there wasn't any mapping
		 * like if add_to_swap() is running on a lru page that
		 * had its mapping zapped. And freeing these pages
		 * requires taking the lru_lock so we do the put_page
		 * of the tail pages after the split is complete.
		 */
		put_page(subpage);
	}
}

3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217
int total_mapcount(struct page *page)
{
	int i, ret;

	VM_BUG_ON_PAGE(PageTail(page), page);

	if (likely(!PageCompound(page)))
		return atomic_read(&page->_mapcount) + 1;

	ret = compound_mapcount(page);
	if (PageHuge(page))
		return ret;
	for (i = 0; i < HPAGE_PMD_NR; i++)
		ret += atomic_read(&page[i]._mapcount) + 1;
	if (PageDoubleMap(page))
		ret -= HPAGE_PMD_NR;
	return ret;
}

3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275
/*
 * This calculates accurately how many mappings a transparent hugepage
 * has (unlike page_mapcount() which isn't fully accurate). This full
 * accuracy is primarily needed to know if copy-on-write faults can
 * reuse the page and change the mapping to read-write instead of
 * copying them. At the same time this returns the total_mapcount too.
 *
 * The function returns the highest mapcount any one of the subpages
 * has. If the return value is one, even if different processes are
 * mapping different subpages of the transparent hugepage, they can
 * all reuse it, because each process is reusing a different subpage.
 *
 * The total_mapcount is instead counting all virtual mappings of the
 * subpages. If the total_mapcount is equal to "one", it tells the
 * caller all mappings belong to the same "mm" and in turn the
 * anon_vma of the transparent hugepage can become the vma->anon_vma
 * local one as no other process may be mapping any of the subpages.
 *
 * It would be more accurate to replace page_mapcount() with
 * page_trans_huge_mapcount(), however we only use
 * page_trans_huge_mapcount() in the copy-on-write faults where we
 * need full accuracy to avoid breaking page pinning, because
 * page_trans_huge_mapcount() is slower than page_mapcount().
 */
int page_trans_huge_mapcount(struct page *page, int *total_mapcount)
{
	int i, ret, _total_mapcount, mapcount;

	/* hugetlbfs shouldn't call it */
	VM_BUG_ON_PAGE(PageHuge(page), page);

	if (likely(!PageTransCompound(page))) {
		mapcount = atomic_read(&page->_mapcount) + 1;
		if (total_mapcount)
			*total_mapcount = mapcount;
		return mapcount;
	}

	page = compound_head(page);

	_total_mapcount = ret = 0;
	for (i = 0; i < HPAGE_PMD_NR; i++) {
		mapcount = atomic_read(&page[i]._mapcount) + 1;
		ret = max(ret, mapcount);
		_total_mapcount += mapcount;
	}
	if (PageDoubleMap(page)) {
		ret -= 1;
		_total_mapcount -= HPAGE_PMD_NR;
	}
	mapcount = compound_mapcount(page);
	ret += mapcount;
	_total_mapcount += mapcount;
	if (total_mapcount)
		*total_mapcount = _total_mapcount;
	return ret;
}

3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297
/*
 * This function splits huge page into normal pages. @page can point to any
 * subpage of huge page to split. Split doesn't change the position of @page.
 *
 * Only caller must hold pin on the @page, otherwise split fails with -EBUSY.
 * The huge page must be locked.
 *
 * If @list is null, tail pages will be added to LRU list, otherwise, to @list.
 *
 * Both head page and tail pages will inherit mapping, flags, and so on from
 * the hugepage.
 *
 * GUP pin and PG_locked transferred to @page. Rest subpages can be freed if
 * they are not mapped.
 *
 * Returns 0 if the hugepage is split successfully.
 * Returns -EBUSY if the page is pinned or if anon_vma disappeared from under
 * us.
 */
int split_huge_page_to_list(struct page *page, struct list_head *list)
{
	struct page *head = compound_head(page);
3298
	struct pglist_data *pgdata = NODE_DATA(page_to_nid(head));
3299 3300
	struct anon_vma *anon_vma;
	int count, mapcount, ret;
3301
	bool mlocked;
3302
	unsigned long flags;
3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332

	VM_BUG_ON_PAGE(is_huge_zero_page(page), page);
	VM_BUG_ON_PAGE(!PageAnon(page), page);
	VM_BUG_ON_PAGE(!PageLocked(page), page);
	VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
	VM_BUG_ON_PAGE(!PageCompound(page), page);

	/*
	 * The caller does not necessarily hold an mmap_sem that would prevent
	 * the anon_vma disappearing so we first we take a reference to it
	 * and then lock the anon_vma for write. This is similar to
	 * page_lock_anon_vma_read except the write lock is taken to serialise
	 * against parallel split or collapse operations.
	 */
	anon_vma = page_get_anon_vma(head);
	if (!anon_vma) {
		ret = -EBUSY;
		goto out;
	}
	anon_vma_lock_write(anon_vma);

	/*
	 * Racy check if we can split the page, before freeze_page() will
	 * split PMDs
	 */
	if (total_mapcount(head) != page_count(head) - 1) {
		ret = -EBUSY;
		goto out_unlock;
	}

3333
	mlocked = PageMlocked(page);
3334
	freeze_page(head);
3335 3336
	VM_BUG_ON_PAGE(compound_mapcount(head), head);

3337 3338 3339 3340
	/* Make sure the page is not on per-CPU pagevec as it takes pin */
	if (mlocked)
		lru_add_drain();

3341
	/* Prevent deferred_split_scan() touching ->_refcount */
3342
	spin_lock_irqsave(&pgdata->split_queue_lock, flags);
3343 3344
	count = page_count(head);
	mapcount = total_mapcount(head);
3345
	if (!mapcount && count == 1) {
3346
		if (!list_empty(page_deferred_list(head))) {
3347
			pgdata->split_queue_len--;
3348 3349
			list_del(page_deferred_list(head));
		}
3350
		spin_unlock_irqrestore(&pgdata->split_queue_lock, flags);
3351 3352
		__split_huge_page(page, list);
		ret = 0;
3353
	} else if (IS_ENABLED(CONFIG_DEBUG_VM) && mapcount) {
3354
		spin_unlock_irqrestore(&pgdata->split_queue_lock, flags);
3355 3356 3357 3358
		pr_alert("total_mapcount: %u, page_count(): %u\n",
				mapcount, count);
		if (PageTail(page))
			dump_page(head, NULL);
3359
		dump_page(page, "total_mapcount(head) > 0");
3360 3361
		BUG();
	} else {
3362
		spin_unlock_irqrestore(&pgdata->split_queue_lock, flags);
3363
		unfreeze_page(head);
3364 3365 3366 3367 3368 3369 3370 3371 3372 3373
		ret = -EBUSY;
	}

out_unlock:
	anon_vma_unlock_write(anon_vma);
	put_anon_vma(anon_vma);
out:
	count_vm_event(!ret ? THP_SPLIT_PAGE : THP_SPLIT_PAGE_FAILED);
	return ret;
}
3374 3375 3376

void free_transhuge_page(struct page *page)
{
3377
	struct pglist_data *pgdata = NODE_DATA(page_to_nid(page));
3378 3379
	unsigned long flags;

3380
	spin_lock_irqsave(&pgdata->split_queue_lock, flags);
3381
	if (!list_empty(page_deferred_list(page))) {
3382
		pgdata->split_queue_len--;
3383 3384
		list_del(page_deferred_list(page));
	}
3385
	spin_unlock_irqrestore(&pgdata->split_queue_lock, flags);
3386 3387 3388 3389 3390
	free_compound_page(page);
}

void deferred_split_huge_page(struct page *page)
{
3391
	struct pglist_data *pgdata = NODE_DATA(page_to_nid(page));
3392 3393 3394 3395
	unsigned long flags;

	VM_BUG_ON_PAGE(!PageTransHuge(page), page);

3396
	spin_lock_irqsave(&pgdata->split_queue_lock, flags);
3397
	if (list_empty(page_deferred_list(page))) {
3398
		count_vm_event(THP_DEFERRED_SPLIT_PAGE);
3399 3400
		list_add_tail(page_deferred_list(page), &pgdata->split_queue);
		pgdata->split_queue_len++;
3401
	}
3402
	spin_unlock_irqrestore(&pgdata->split_queue_lock, flags);
3403 3404 3405 3406 3407
}

static unsigned long deferred_split_count(struct shrinker *shrink,
		struct shrink_control *sc)
{
3408
	struct pglist_data *pgdata = NODE_DATA(sc->nid);
3409
	return ACCESS_ONCE(pgdata->split_queue_len);
3410 3411 3412 3413 3414
}

static unsigned long deferred_split_scan(struct shrinker *shrink,
		struct shrink_control *sc)
{
3415
	struct pglist_data *pgdata = NODE_DATA(sc->nid);
3416 3417 3418 3419 3420
	unsigned long flags;
	LIST_HEAD(list), *pos, *next;
	struct page *page;
	int split = 0;

3421
	spin_lock_irqsave(&pgdata->split_queue_lock, flags);
3422
	/* Take pin on all head pages to avoid freeing them under us */
3423
	list_for_each_safe(pos, next, &pgdata->split_queue) {
3424 3425
		page = list_entry((void *)pos, struct page, mapping);
		page = compound_head(page);
3426 3427 3428 3429
		if (get_page_unless_zero(page)) {
			list_move(page_deferred_list(page), &list);
		} else {
			/* We lost race with put_compound_page() */
3430
			list_del_init(page_deferred_list(page));
3431
			pgdata->split_queue_len--;
3432
		}
3433 3434
		if (!--sc->nr_to_scan)
			break;
3435
	}
3436
	spin_unlock_irqrestore(&pgdata->split_queue_lock, flags);
3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447

	list_for_each_safe(pos, next, &list) {
		page = list_entry((void *)pos, struct page, mapping);
		lock_page(page);
		/* split_huge_page() removes page from list on success */
		if (!split_huge_page(page))
			split++;
		unlock_page(page);
		put_page(page);
	}

3448 3449 3450
	spin_lock_irqsave(&pgdata->split_queue_lock, flags);
	list_splice_tail(&list, &pgdata->split_queue);
	spin_unlock_irqrestore(&pgdata->split_queue_lock, flags);
3451

3452 3453 3454 3455 3456 3457 3458
	/*
	 * Stop shrinker if we didn't split any page, but the queue is empty.
	 * This can happen if pages were freed under us.
	 */
	if (!split && list_empty(&pgdata->split_queue))
		return SHRINK_STOP;
	return split;
3459 3460 3461 3462 3463 3464
}

static struct shrinker deferred_split_shrinker = {
	.count_objects = deferred_split_count,
	.scan_objects = deferred_split_scan,
	.seeks = DEFAULT_SEEKS,
3465
	.flags = SHRINKER_NUMA_AWARE,
3466
};
3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505

#ifdef CONFIG_DEBUG_FS
static int split_huge_pages_set(void *data, u64 val)
{
	struct zone *zone;
	struct page *page;
	unsigned long pfn, max_zone_pfn;
	unsigned long total = 0, split = 0;

	if (val != 1)
		return -EINVAL;

	for_each_populated_zone(zone) {
		max_zone_pfn = zone_end_pfn(zone);
		for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) {
			if (!pfn_valid(pfn))
				continue;

			page = pfn_to_page(pfn);
			if (!get_page_unless_zero(page))
				continue;

			if (zone != page_zone(page))
				goto next;

			if (!PageHead(page) || !PageAnon(page) ||
					PageHuge(page))
				goto next;

			total++;
			lock_page(page);
			if (!split_huge_page(page))
				split++;
			unlock_page(page);
next:
			put_page(page);
		}
	}

3506
	pr_info("%lu of %lu THP split\n", split, total);
3507 3508 3509 3510 3511 3512 3513 3514 3515 3516

	return 0;
}
DEFINE_SIMPLE_ATTRIBUTE(split_huge_pages_fops, NULL, split_huge_pages_set,
		"%llu\n");

static int __init split_huge_pages_debugfs(void)
{
	void *ret;

3517
	ret = debugfs_create_file("split_huge_pages", 0200, NULL, NULL,
3518 3519 3520 3521 3522 3523 3524
			&split_huge_pages_fops);
	if (!ret)
		pr_warn("Failed to create split_huge_pages in debugfs");
	return 0;
}
late_initcall(split_huge_pages_debugfs);
#endif