huge_memory.c 92.8 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
A
Andrea Arcangeli 已提交
81
	(1<<TRANSPARENT_HUGEPAGE_DEFRAG_FLAG)|
82 83
	(1<<TRANSPARENT_HUGEPAGE_DEFRAG_KHUGEPAGED_FLAG)|
	(1<<TRANSPARENT_HUGEPAGE_USE_ZERO_PAGE_FLAG);
A
Andrea Arcangeli 已提交
84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104

/* default scan 8*512 pte (or vmas) every 30 second */
static unsigned int khugepaged_pages_to_scan __read_mostly = HPAGE_PMD_NR*8;
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;
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.
 */
static unsigned int khugepaged_max_ptes_none __read_mostly = HPAGE_PMD_NR-1;

static int khugepaged(void *none);
static int khugepaged_slab_init(void);
105
static void khugepaged_slab_exit(void);
A
Andrea Arcangeli 已提交
106

107 108 109
#define MM_SLOTS_HASH_BITS 10
static __read_mostly DEFINE_HASHTABLE(mm_slots_hash, MM_SLOTS_HASH_BITS);

A
Andrea Arcangeli 已提交
110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135
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;
136 137
};
static struct khugepaged_scan khugepaged_scan = {
A
Andrea Arcangeli 已提交
138 139 140
	.mm_head = LIST_HEAD_INIT(khugepaged_scan.mm_head),
};

141
static struct shrinker deferred_split_shrinker;
142

143
static void set_recommended_min_free_kbytes(void)
144 145 146 147 148 149 150 151
{
	struct zone *zone;
	int nr_zones = 0;
	unsigned long recommended_min;

	for_each_populated_zone(zone)
		nr_zones++;

152
	/* Ensure 2 pageblocks are free to assist fragmentation avoidance */
153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168
	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);

169 170 171 172 173 174
	if (recommended_min > min_free_kbytes) {
		if (user_min_free_kbytes >= 0)
			pr_info("raising min_free_kbytes from %d to %lu "
				"to help transparent hugepage allocations\n",
				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
static 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

273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320
static ssize_t double_flag_show(struct kobject *kobj,
				struct kobj_attribute *attr, char *buf,
				enum transparent_hugepage_flag enabled,
				enum transparent_hugepage_flag req_madv)
{
	if (test_bit(enabled, &transparent_hugepage_flags)) {
		VM_BUG_ON(test_bit(req_madv, &transparent_hugepage_flags));
		return sprintf(buf, "[always] madvise never\n");
	} else if (test_bit(req_madv, &transparent_hugepage_flags))
		return sprintf(buf, "always [madvise] never\n");
	else
		return sprintf(buf, "always madvise [never]\n");
}
static ssize_t double_flag_store(struct kobject *kobj,
				 struct kobj_attribute *attr,
				 const char *buf, size_t count,
				 enum transparent_hugepage_flag enabled,
				 enum transparent_hugepage_flag req_madv)
{
	if (!memcmp("always", buf,
		    min(sizeof("always")-1, count))) {
		set_bit(enabled, &transparent_hugepage_flags);
		clear_bit(req_madv, &transparent_hugepage_flags);
	} else if (!memcmp("madvise", buf,
			   min(sizeof("madvise")-1, count))) {
		clear_bit(enabled, &transparent_hugepage_flags);
		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);
	} else
		return -EINVAL;

	return count;
}

static ssize_t enabled_show(struct kobject *kobj,
			    struct kobj_attribute *attr, char *buf)
{
	return double_flag_show(kobj, attr, buf,
				TRANSPARENT_HUGEPAGE_FLAG,
				TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG);
}
static ssize_t enabled_store(struct kobject *kobj,
			     struct kobj_attribute *attr,
			     const char *buf, size_t count)
{
A
Andrea Arcangeli 已提交
321 322 323 324 325 326 327
	ssize_t ret;

	ret = double_flag_store(kobj, attr, buf, count,
				TRANSPARENT_HUGEPAGE_FLAG,
				TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG);

	if (ret > 0) {
328 329 330
		int err;

		mutex_lock(&khugepaged_mutex);
331
		err = start_stop_khugepaged();
332 333
		mutex_unlock(&khugepaged_mutex);

A
Andrea Arcangeli 已提交
334 335 336 337 338
		if (err)
			ret = err;
	}

	return ret;
339 340 341 342 343 344 345 346
}
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)
{
347 348
	return sprintf(buf, "%d\n",
		       !!test_bit(flag, &transparent_hugepage_flags));
349
}
350

351 352 353 354 355
static ssize_t single_flag_store(struct kobject *kobj,
				 struct kobj_attribute *attr,
				 const char *buf, size_t count,
				 enum transparent_hugepage_flag flag)
{
356 357 358 359 360 361 362 363 364 365
	unsigned long value;
	int ret;

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

	if (value)
366
		set_bit(flag, &transparent_hugepage_flags);
367
	else
368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395
		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)
{
	return double_flag_show(kobj, attr, buf,
				TRANSPARENT_HUGEPAGE_DEFRAG_FLAG,
				TRANSPARENT_HUGEPAGE_DEFRAG_REQ_MADV_FLAG);
}
static ssize_t defrag_store(struct kobject *kobj,
			    struct kobj_attribute *attr,
			    const char *buf, size_t count)
{
	return double_flag_store(kobj, attr, buf, count,
				 TRANSPARENT_HUGEPAGE_DEFRAG_FLAG,
				 TRANSPARENT_HUGEPAGE_DEFRAG_REQ_MADV_FLAG);
}
static struct kobj_attribute defrag_attr =
	__ATTR(defrag, 0644, defrag_show, defrag_store);

396 397 398 399 400 401 402 403 404 405 406 407 408 409
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);
410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430
#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,
431
	&use_zero_page_attr.attr,
432 433 434 435 436 437 438 439
#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
440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455
};

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;

456
	err = kstrtoul(buf, 10, &msecs);
A
Andrea Arcangeli 已提交
457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482
	if (err || msecs > UINT_MAX)
		return -EINVAL;

	khugepaged_scan_sleep_millisecs = msecs;
	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;

483
	err = kstrtoul(buf, 10, &msecs);
A
Andrea Arcangeli 已提交
484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508
	if (err || msecs > UINT_MAX)
		return -EINVAL;

	khugepaged_alloc_sleep_millisecs = msecs;
	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;

509
	err = kstrtoul(buf, 10, &pages);
A
Andrea Arcangeli 已提交
510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576
	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;

577
	err = kstrtoul(buf, 10, &max_ptes_none);
A
Andrea Arcangeli 已提交
578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602
	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",
603 604
};

S
Shaohua Li 已提交
605
static int __init hugepage_init_sysfs(struct kobject **hugepage_kobj)
606 607 608
{
	int err;

S
Shaohua Li 已提交
609 610
	*hugepage_kobj = kobject_create_and_add("transparent_hugepage", mm_kobj);
	if (unlikely(!*hugepage_kobj)) {
611
		pr_err("failed to create transparent hugepage kobject\n");
S
Shaohua Li 已提交
612
		return -ENOMEM;
A
Andrea Arcangeli 已提交
613 614
	}

S
Shaohua Li 已提交
615
	err = sysfs_create_group(*hugepage_kobj, &hugepage_attr_group);
A
Andrea Arcangeli 已提交
616
	if (err) {
617
		pr_err("failed to register transparent hugepage group\n");
S
Shaohua Li 已提交
618
		goto delete_obj;
A
Andrea Arcangeli 已提交
619 620
	}

S
Shaohua Li 已提交
621
	err = sysfs_create_group(*hugepage_kobj, &khugepaged_attr_group);
A
Andrea Arcangeli 已提交
622
	if (err) {
623
		pr_err("failed to register transparent hugepage group\n");
S
Shaohua Li 已提交
624
		goto remove_hp_group;
A
Andrea Arcangeli 已提交
625
	}
S
Shaohua Li 已提交
626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664

	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;
	}

	err = hugepage_init_sysfs(&hugepage_kobj);
	if (err)
665
		goto err_sysfs;
A
Andrea Arcangeli 已提交
666 667 668

	err = khugepaged_slab_init();
	if (err)
669
		goto err_slab;
A
Andrea Arcangeli 已提交
670

671 672 673
	err = register_shrinker(&huge_zero_page_shrinker);
	if (err)
		goto err_hzp_shrinker;
674 675 676
	err = register_shrinker(&deferred_split_shrinker);
	if (err)
		goto err_split_shrinker;
677

678 679 680 681 682
	/*
	 * 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.
	 */
683
	if (totalram_pages < (512 << (20 - PAGE_SHIFT))) {
684
		transparent_hugepage_flags = 0;
685 686
		return 0;
	}
687

688
	err = start_stop_khugepaged();
689 690
	if (err)
		goto err_khugepaged;
A
Andrea Arcangeli 已提交
691

S
Shaohua Li 已提交
692
	return 0;
693
err_khugepaged:
694 695
	unregister_shrinker(&deferred_split_shrinker);
err_split_shrinker:
696 697 698 699
	unregister_shrinker(&huge_zero_page_shrinker);
err_hzp_shrinker:
	khugepaged_slab_exit();
err_slab:
S
Shaohua Li 已提交
700
	hugepage_exit_sysfs(hugepage_kobj);
701
err_sysfs:
A
Andrea Arcangeli 已提交
702
	return err;
703
}
704
subsys_initcall(hugepage_init);
705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731

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)
732
		pr_warn("transparent_hugepage= cannot parse, ignored\n");
733 734 735 736
	return ret;
}
__setup("transparent_hugepage=", setup_transparent_hugepage);

737
pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma)
738 739 740 741 742 743
{
	if (likely(vma->vm_flags & VM_WRITE))
		pmd = pmd_mkwrite(pmd);
	return pmd;
}

744
static inline pmd_t mk_huge_pmd(struct page *page, pgprot_t prot)
B
Bob Liu 已提交
745 746
{
	pmd_t entry;
747
	entry = mk_pmd(page, prot);
B
Bob Liu 已提交
748 749 750 751
	entry = pmd_mkhuge(entry);
	return entry;
}

752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772
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
	 */
	BUILD_BUG_ON(HPAGE_PMD_ORDER < 2);

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

773 774
static int __do_huge_pmd_anonymous_page(struct mm_struct *mm,
					struct vm_area_struct *vma,
775
					unsigned long address, pmd_t *pmd,
776 777
					struct page *page, gfp_t gfp,
					unsigned int flags)
778
{
779
	struct mem_cgroup *memcg;
780
	pgtable_t pgtable;
781
	spinlock_t *ptl;
782
	unsigned long haddr = address & HPAGE_PMD_MASK;
783

784
	VM_BUG_ON_PAGE(!PageCompound(page), page);
785

786
	if (mem_cgroup_try_charge(page, mm, gfp, &memcg, true)) {
787 788 789 790
		put_page(page);
		count_vm_event(THP_FAULT_FALLBACK);
		return VM_FAULT_FALLBACK;
	}
791

792
	pgtable = pte_alloc_one(mm, haddr);
793
	if (unlikely(!pgtable)) {
794
		mem_cgroup_cancel_charge(page, memcg, true);
795
		put_page(page);
796
		return VM_FAULT_OOM;
797
	}
798 799

	clear_huge_page(page, haddr, HPAGE_PMD_NR);
800 801 802 803 804
	/*
	 * The memory barrier inside __SetPageUptodate makes sure that
	 * clear_huge_page writes become visible before the set_pmd_at()
	 * write.
	 */
805 806
	__SetPageUptodate(page);

807
	ptl = pmd_lock(mm, pmd);
808
	if (unlikely(!pmd_none(*pmd))) {
809
		spin_unlock(ptl);
810
		mem_cgroup_cancel_charge(page, memcg, true);
811 812 813 814
		put_page(page);
		pte_free(mm, pgtable);
	} else {
		pmd_t entry;
815 816 817 818 819 820

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

			spin_unlock(ptl);
821
			mem_cgroup_cancel_charge(page, memcg, true);
822 823
			put_page(page);
			pte_free(mm, pgtable);
824
			ret = handle_userfault(vma, address, flags,
825 826 827 828 829
					       VM_UFFD_MISSING);
			VM_BUG_ON(ret & VM_FAULT_FALLBACK);
			return ret;
		}

830 831
		entry = mk_huge_pmd(page, vma->vm_page_prot);
		entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma);
832
		page_add_new_anon_rmap(page, vma, haddr, true);
833
		mem_cgroup_commit_charge(page, memcg, false, true);
834
		lru_cache_add_active_or_unevictable(page, vma);
835
		pgtable_trans_huge_deposit(mm, pmd, pgtable);
836 837
		set_pmd_at(mm, haddr, pmd, entry);
		add_mm_counter(mm, MM_ANONPAGES, HPAGE_PMD_NR);
838
		atomic_long_inc(&mm->nr_ptes);
839
		spin_unlock(ptl);
840
		count_vm_event(THP_FAULT_ALLOC);
841 842
	}

843
	return 0;
844 845
}

846
static inline gfp_t alloc_hugepage_gfpmask(int defrag, gfp_t extra_gfp)
847
{
848
	return (GFP_TRANSHUGE & ~(defrag ? 0 : __GFP_RECLAIM)) | extra_gfp;
849 850
}

851
/* Caller must hold page table lock. */
852
static bool set_huge_zero_page(pgtable_t pgtable, struct mm_struct *mm,
853
		struct vm_area_struct *vma, unsigned long haddr, pmd_t *pmd,
854
		struct page *zero_page)
855 856
{
	pmd_t entry;
A
Andrew Morton 已提交
857 858
	if (!pmd_none(*pmd))
		return false;
859
	entry = mk_pmd(zero_page, vma->vm_page_prot);
860
	entry = pmd_mkhuge(entry);
861 862
	if (pgtable)
		pgtable_trans_huge_deposit(mm, pmd, pgtable);
863
	set_pmd_at(mm, haddr, pmd, entry);
864
	atomic_long_inc(&mm->nr_ptes);
A
Andrew Morton 已提交
865
	return true;
866 867
}

868 869 870 871
int do_huge_pmd_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
			       unsigned long address, pmd_t *pmd,
			       unsigned int flags)
{
872
	gfp_t gfp;
873 874 875
	struct page *page;
	unsigned long haddr = address & HPAGE_PMD_MASK;

876
	if (haddr < vma->vm_start || haddr + HPAGE_PMD_SIZE > vma->vm_end)
877
		return VM_FAULT_FALLBACK;
878 879
	if (unlikely(anon_vma_prepare(vma)))
		return VM_FAULT_OOM;
880
	if (unlikely(khugepaged_enter(vma, vma->vm_flags)))
881
		return VM_FAULT_OOM;
882
	if (!(flags & FAULT_FLAG_WRITE) && !mm_forbids_zeropage(mm) &&
883
			transparent_hugepage_use_zero_page()) {
884
		spinlock_t *ptl;
885 886 887
		pgtable_t pgtable;
		struct page *zero_page;
		bool set;
888
		int ret;
889 890
		pgtable = pte_alloc_one(mm, haddr);
		if (unlikely(!pgtable))
A
Andrea Arcangeli 已提交
891
			return VM_FAULT_OOM;
892 893 894
		zero_page = get_huge_zero_page();
		if (unlikely(!zero_page)) {
			pte_free(mm, pgtable);
895
			count_vm_event(THP_FAULT_FALLBACK);
896
			return VM_FAULT_FALLBACK;
A
Andrea Arcangeli 已提交
897
		}
898
		ptl = pmd_lock(mm, pmd);
899 900 901 902 903
		ret = 0;
		set = false;
		if (pmd_none(*pmd)) {
			if (userfaultfd_missing(vma)) {
				spin_unlock(ptl);
904
				ret = handle_userfault(vma, address, flags,
905 906 907 908 909 910 911 912 913 914 915
						       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);
916 917 918
		if (!set) {
			pte_free(mm, pgtable);
			put_huge_zero_page();
919
		}
920
		return ret;
921
	}
922 923
	gfp = alloc_hugepage_gfpmask(transparent_hugepage_defrag(vma), 0);
	page = alloc_hugepage_vma(gfp, vma, haddr, HPAGE_PMD_ORDER);
924 925
	if (unlikely(!page)) {
		count_vm_event(THP_FAULT_FALLBACK);
926
		return VM_FAULT_FALLBACK;
927
	}
928
	prep_transhuge_page(page);
929 930
	return __do_huge_pmd_anonymous_page(mm, vma, address, pmd, page, gfp,
					    flags);
931 932
}

933
static void insert_pfn_pmd(struct vm_area_struct *vma, unsigned long addr,
934
		pmd_t *pmd, pfn_t pfn, pgprot_t prot, bool write)
M
Matthew Wilcox 已提交
935 936 937 938 939 940
{
	struct mm_struct *mm = vma->vm_mm;
	pmd_t entry;
	spinlock_t *ptl;

	ptl = pmd_lock(mm, pmd);
941 942 943
	entry = pmd_mkhuge(pfn_t_pmd(pfn, prot));
	if (pfn_t_devmap(pfn))
		entry = pmd_mkdevmap(entry);
944 945 946
	if (write) {
		entry = pmd_mkyoung(pmd_mkdirty(entry));
		entry = maybe_pmd_mkwrite(entry, vma);
M
Matthew Wilcox 已提交
947
	}
948 949
	set_pmd_at(mm, addr, pmd, entry);
	update_mmu_cache_pmd(vma, addr, pmd);
M
Matthew Wilcox 已提交
950 951 952 953
	spin_unlock(ptl);
}

int vmf_insert_pfn_pmd(struct vm_area_struct *vma, unsigned long addr,
954
			pmd_t *pmd, pfn_t pfn, bool write)
M
Matthew Wilcox 已提交
955 956 957 958 959 960 961 962 963 964 965
{
	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));
966
	BUG_ON(!pfn_t_devmap(pfn));
M
Matthew Wilcox 已提交
967 968 969 970 971

	if (addr < vma->vm_start || addr >= vma->vm_end)
		return VM_FAULT_SIGBUS;
	if (track_pfn_insert(vma, &pgprot, pfn))
		return VM_FAULT_SIGBUS;
972 973
	insert_pfn_pmd(vma, addr, pmd, pfn, pgprot, write);
	return VM_FAULT_NOPAGE;
M
Matthew Wilcox 已提交
974 975
}

976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032
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;
}

1033 1034 1035 1036
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)
{
1037
	spinlock_t *dst_ptl, *src_ptl;
1038 1039
	struct page *src_page;
	pmd_t pmd;
1040
	pgtable_t pgtable = NULL;
1041 1042
	int ret;

1043 1044 1045 1046 1047 1048
	if (!vma_is_dax(vma)) {
		ret = -ENOMEM;
		pgtable = pte_alloc_one(dst_mm, addr);
		if (unlikely(!pgtable))
			goto out;
	}
1049

1050 1051 1052
	dst_ptl = pmd_lock(dst_mm, dst_pmd);
	src_ptl = pmd_lockptr(src_mm, src_pmd);
	spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
1053 1054 1055

	ret = -EAGAIN;
	pmd = *src_pmd;
1056
	if (unlikely(!pmd_trans_huge(pmd) && !pmd_devmap(pmd))) {
1057 1058 1059
		pte_free(dst_mm, pgtable);
		goto out_unlock;
	}
1060
	/*
1061
	 * When page table lock is held, the huge zero pmd should not be
1062 1063 1064 1065
	 * under splitting since we don't split the page itself, only pmd to
	 * a page table.
	 */
	if (is_huge_zero_pmd(pmd)) {
1066
		struct page *zero_page;
1067 1068 1069 1070 1071
		/*
		 * 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.
		 */
1072
		zero_page = get_huge_zero_page();
1073
		set_huge_zero_page(pgtable, dst_mm, vma, addr, dst_pmd,
1074
				zero_page);
1075 1076 1077
		ret = 0;
		goto out_unlock;
	}
1078

1079
	if (!vma_is_dax(vma)) {
1080 1081 1082 1083 1084 1085 1086 1087 1088
		/* 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);
	}
1089 1090 1091 1092 1093 1094 1095

	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:
1096 1097
	spin_unlock(src_ptl);
	spin_unlock(dst_ptl);
1098 1099 1100 1101
out:
	return ret;
}

1102 1103 1104 1105 1106 1107
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)
{
1108
	spinlock_t *ptl;
1109 1110 1111
	pmd_t entry;
	unsigned long haddr;

1112
	ptl = pmd_lock(mm, pmd);
1113 1114 1115 1116 1117 1118 1119 1120 1121
	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:
1122
	spin_unlock(ptl);
1123 1124
}

1125 1126 1127 1128 1129 1130 1131
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)
{
1132
	struct mem_cgroup *memcg;
1133
	spinlock_t *ptl;
1134 1135 1136 1137
	pgtable_t pgtable;
	pmd_t _pmd;
	int ret = 0, i;
	struct page **pages;
1138 1139
	unsigned long mmun_start;	/* For mmu_notifiers */
	unsigned long mmun_end;		/* For mmu_notifiers */
1140 1141 1142 1143 1144 1145 1146 1147 1148

	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++) {
1149 1150
		pages[i] = alloc_page_vma_node(GFP_HIGHUSER_MOVABLE |
					       __GFP_OTHER_NODE,
1151
					       vma, address, page_to_nid(page));
A
Andrea Arcangeli 已提交
1152
		if (unlikely(!pages[i] ||
1153
			     mem_cgroup_try_charge(pages[i], mm, GFP_KERNEL,
1154
						   &memcg, false))) {
A
Andrea Arcangeli 已提交
1155
			if (pages[i])
1156
				put_page(pages[i]);
A
Andrea Arcangeli 已提交
1157
			while (--i >= 0) {
1158 1159
				memcg = (void *)page_private(pages[i]);
				set_page_private(pages[i], 0);
1160 1161
				mem_cgroup_cancel_charge(pages[i], memcg,
						false);
A
Andrea Arcangeli 已提交
1162 1163
				put_page(pages[i]);
			}
1164 1165 1166 1167
			kfree(pages);
			ret |= VM_FAULT_OOM;
			goto out;
		}
1168
		set_page_private(pages[i], (unsigned long)memcg);
1169 1170 1171 1172
	}

	for (i = 0; i < HPAGE_PMD_NR; i++) {
		copy_user_highpage(pages[i], page + i,
1173
				   haddr + PAGE_SIZE * i, vma);
1174 1175 1176 1177
		__SetPageUptodate(pages[i]);
		cond_resched();
	}

1178 1179 1180 1181
	mmun_start = haddr;
	mmun_end   = haddr + HPAGE_PMD_SIZE;
	mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);

1182
	ptl = pmd_lock(mm, pmd);
1183 1184
	if (unlikely(!pmd_same(*pmd, orig_pmd)))
		goto out_free_pages;
1185
	VM_BUG_ON_PAGE(!PageHead(page), page);
1186

1187
	pmdp_huge_clear_flush_notify(vma, haddr, pmd);
1188 1189
	/* leave pmd empty until pte is filled */

1190
	pgtable = pgtable_trans_huge_withdraw(mm, pmd);
1191 1192 1193 1194 1195 1196
	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);
1197 1198
		memcg = (void *)page_private(pages[i]);
		set_page_private(pages[i], 0);
1199
		page_add_new_anon_rmap(pages[i], vma, haddr, false);
1200
		mem_cgroup_commit_charge(pages[i], memcg, false, false);
1201
		lru_cache_add_active_or_unevictable(pages[i], vma);
1202 1203 1204 1205 1206 1207 1208 1209 1210
		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);
1211
	page_remove_rmap(page, true);
1212
	spin_unlock(ptl);
1213

1214 1215
	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);

1216 1217 1218 1219 1220 1221 1222
	ret |= VM_FAULT_WRITE;
	put_page(page);

out:
	return ret;

out_free_pages:
1223
	spin_unlock(ptl);
1224
	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
A
Andrea Arcangeli 已提交
1225
	for (i = 0; i < HPAGE_PMD_NR; i++) {
1226 1227
		memcg = (void *)page_private(pages[i]);
		set_page_private(pages[i], 0);
1228
		mem_cgroup_cancel_charge(pages[i], memcg, false);
1229
		put_page(pages[i]);
A
Andrea Arcangeli 已提交
1230
	}
1231 1232 1233 1234 1235 1236 1237
	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)
{
1238
	spinlock_t *ptl;
1239
	int ret = 0;
1240
	struct page *page = NULL, *new_page;
1241
	struct mem_cgroup *memcg;
1242
	unsigned long haddr;
1243 1244
	unsigned long mmun_start;	/* For mmu_notifiers */
	unsigned long mmun_end;		/* For mmu_notifiers */
1245
	gfp_t huge_gfp;			/* for allocation and charge */
1246

1247
	ptl = pmd_lockptr(mm, pmd);
1248
	VM_BUG_ON_VMA(!vma->anon_vma, vma);
1249 1250 1251
	haddr = address & HPAGE_PMD_MASK;
	if (is_huge_zero_pmd(orig_pmd))
		goto alloc;
1252
	spin_lock(ptl);
1253 1254 1255 1256
	if (unlikely(!pmd_same(*pmd, orig_pmd)))
		goto out_unlock;

	page = pmd_page(orig_pmd);
1257
	VM_BUG_ON_PAGE(!PageCompound(page) || !PageHead(page), page);
1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268
	/*
	 * We can only reuse the page if nobody else maps the huge page or it's
	 * part. We can do it by checking page_mapcount() on each sub-page, but
	 * it's expensive.
	 * The cheaper way is to check page_count() to be equal 1: every
	 * mapcount takes page reference reference, so this way we can
	 * guarantee, that the PMD is the only mapping.
	 * This can give false negative if somebody pinned the page, but that's
	 * fine.
	 */
	if (page_mapcount(page) == 1 && page_count(page) == 1) {
1269 1270 1271 1272
		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))
1273
			update_mmu_cache_pmd(vma, address, pmd);
1274 1275 1276
		ret |= VM_FAULT_WRITE;
		goto out_unlock;
	}
1277
	get_page(page);
1278
	spin_unlock(ptl);
1279
alloc:
1280
	if (transparent_hugepage_enabled(vma) &&
1281
	    !transparent_hugepage_debug_cow()) {
1282 1283
		huge_gfp = alloc_hugepage_gfpmask(transparent_hugepage_defrag(vma), 0);
		new_page = alloc_hugepage_vma(huge_gfp, vma, haddr, HPAGE_PMD_ORDER);
1284
	} else
1285 1286
		new_page = NULL;

1287 1288 1289
	if (likely(new_page)) {
		prep_transhuge_page(new_page);
	} else {
1290
		if (!page) {
1291
			split_huge_pmd(vma, pmd, address);
1292
			ret |= VM_FAULT_FALLBACK;
1293 1294 1295
		} else {
			ret = do_huge_pmd_wp_page_fallback(mm, vma, address,
					pmd, orig_pmd, page, haddr);
1296
			if (ret & VM_FAULT_OOM) {
1297
				split_huge_pmd(vma, pmd, address);
1298 1299
				ret |= VM_FAULT_FALLBACK;
			}
1300
			put_page(page);
1301
		}
1302
		count_vm_event(THP_FAULT_FALLBACK);
1303 1304 1305
		goto out;
	}

1306 1307
	if (unlikely(mem_cgroup_try_charge(new_page, mm, huge_gfp, &memcg,
					   true))) {
A
Andrea Arcangeli 已提交
1308
		put_page(new_page);
1309
		if (page) {
1310
			split_huge_pmd(vma, pmd, address);
1311
			put_page(page);
1312
		} else
1313
			split_huge_pmd(vma, pmd, address);
1314
		ret |= VM_FAULT_FALLBACK;
1315
		count_vm_event(THP_FAULT_FALLBACK);
A
Andrea Arcangeli 已提交
1316 1317 1318
		goto out;
	}

1319 1320
	count_vm_event(THP_FAULT_ALLOC);

1321
	if (!page)
1322 1323 1324
		clear_huge_page(new_page, haddr, HPAGE_PMD_NR);
	else
		copy_user_huge_page(new_page, page, haddr, vma, HPAGE_PMD_NR);
1325 1326
	__SetPageUptodate(new_page);

1327 1328 1329 1330
	mmun_start = haddr;
	mmun_end   = haddr + HPAGE_PMD_SIZE;
	mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);

1331
	spin_lock(ptl);
1332
	if (page)
1333
		put_page(page);
A
Andrea Arcangeli 已提交
1334
	if (unlikely(!pmd_same(*pmd, orig_pmd))) {
1335
		spin_unlock(ptl);
1336
		mem_cgroup_cancel_charge(new_page, memcg, true);
1337
		put_page(new_page);
1338
		goto out_mn;
A
Andrea Arcangeli 已提交
1339
	} else {
1340
		pmd_t entry;
1341 1342
		entry = mk_huge_pmd(new_page, vma->vm_page_prot);
		entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma);
1343
		pmdp_huge_clear_flush_notify(vma, haddr, pmd);
1344
		page_add_new_anon_rmap(new_page, vma, haddr, true);
1345
		mem_cgroup_commit_charge(new_page, memcg, false, true);
1346
		lru_cache_add_active_or_unevictable(new_page, vma);
1347
		set_pmd_at(mm, haddr, pmd, entry);
1348
		update_mmu_cache_pmd(vma, address, pmd);
1349
		if (!page) {
1350
			add_mm_counter(mm, MM_ANONPAGES, HPAGE_PMD_NR);
1351 1352
			put_huge_zero_page();
		} else {
1353
			VM_BUG_ON_PAGE(!PageHead(page), page);
1354
			page_remove_rmap(page, true);
1355 1356
			put_page(page);
		}
1357 1358
		ret |= VM_FAULT_WRITE;
	}
1359
	spin_unlock(ptl);
1360 1361
out_mn:
	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
1362 1363
out:
	return ret;
1364
out_unlock:
1365
	spin_unlock(ptl);
1366
	return ret;
1367 1368
}

1369
struct page *follow_trans_huge_pmd(struct vm_area_struct *vma,
1370 1371 1372 1373
				   unsigned long addr,
				   pmd_t *pmd,
				   unsigned int flags)
{
1374
	struct mm_struct *mm = vma->vm_mm;
1375 1376
	struct page *page = NULL;

1377
	assert_spin_locked(pmd_lockptr(mm, pmd));
1378 1379 1380 1381

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

1382 1383 1384 1385
	/* Avoid dumping huge zero page */
	if ((flags & FOLL_DUMP) && is_huge_zero_pmd(*pmd))
		return ERR_PTR(-EFAULT);

1386
	/* Full NUMA hinting faults to serialise migration in fault paths */
1387
	if ((flags & FOLL_NUMA) && pmd_protnone(*pmd))
1388 1389
		goto out;

1390
	page = pmd_page(*pmd);
1391
	VM_BUG_ON_PAGE(!PageHead(page), page);
1392 1393
	if (flags & FOLL_TOUCH)
		touch_pmd(vma, addr, pmd);
E
Eric B Munson 已提交
1394
	if ((flags & FOLL_MLOCK) && (vma->vm_flags & VM_LOCKED)) {
1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408
		/*
		 * 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)) {
1409 1410 1411 1412 1413 1414
			lru_add_drain();
			if (page->mapping)
				mlock_vma_page(page);
			unlock_page(page);
		}
	}
1415
	page += (addr & ~HPAGE_PMD_MASK) >> PAGE_SHIFT;
1416
	VM_BUG_ON_PAGE(!PageCompound(page), page);
1417
	if (flags & FOLL_GET)
1418
		get_page(page);
1419 1420 1421 1422 1423

out:
	return page;
}

1424
/* NUMA hinting page fault entry point for trans huge pmds */
1425 1426
int do_huge_pmd_numa_page(struct mm_struct *mm, struct vm_area_struct *vma,
				unsigned long addr, pmd_t pmd, pmd_t *pmdp)
1427
{
1428
	spinlock_t *ptl;
1429
	struct anon_vma *anon_vma = NULL;
1430
	struct page *page;
1431
	unsigned long haddr = addr & HPAGE_PMD_MASK;
1432
	int page_nid = -1, this_nid = numa_node_id();
1433
	int target_nid, last_cpupid = -1;
1434 1435
	bool page_locked;
	bool migrated = false;
1436
	bool was_writable;
1437
	int flags = 0;
1438

1439 1440 1441
	/* A PROT_NONE fault should not end up here */
	BUG_ON(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)));

1442
	ptl = pmd_lock(mm, pmdp);
1443 1444 1445
	if (unlikely(!pmd_same(pmd, *pmdp)))
		goto out_unlock;

1446 1447 1448 1449 1450 1451
	/*
	 * 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))) {
1452
		page = pmd_page(*pmdp);
1453
		spin_unlock(ptl);
1454
		wait_on_page_locked(page);
1455 1456 1457
		goto out;
	}

1458
	page = pmd_page(pmd);
1459
	BUG_ON(is_huge_zero_page(page));
1460
	page_nid = page_to_nid(page);
1461
	last_cpupid = page_cpupid_last(page);
1462
	count_vm_numa_event(NUMA_HINT_FAULTS);
1463
	if (page_nid == this_nid) {
1464
		count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL);
1465 1466
		flags |= TNF_FAULT_LOCAL;
	}
1467

1468 1469
	/* See similar comment in do_numa_page for explanation */
	if (!(vma->vm_flags & VM_WRITE))
1470 1471
		flags |= TNF_NO_GROUP;

1472 1473 1474 1475
	/*
	 * Acquire the page lock to serialise THP migrations but avoid dropping
	 * page_table_lock if at all possible
	 */
1476 1477 1478 1479
	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 */
1480
		if (page_locked)
1481
			goto clear_pmdnuma;
1482
	}
1483

1484
	/* Migration could have started since the pmd_trans_migrating check */
1485
	if (!page_locked) {
1486
		spin_unlock(ptl);
1487
		wait_on_page_locked(page);
1488
		page_nid = -1;
1489 1490 1491
		goto out;
	}

1492 1493 1494 1495
	/*
	 * Page is misplaced. Page lock serialises migrations. Acquire anon_vma
	 * to serialises splits
	 */
1496
	get_page(page);
1497
	spin_unlock(ptl);
1498
	anon_vma = page_lock_anon_vma_read(page);
1499

P
Peter Zijlstra 已提交
1500
	/* Confirm the PMD did not change while page_table_lock was released */
1501
	spin_lock(ptl);
1502 1503 1504
	if (unlikely(!pmd_same(pmd, *pmdp))) {
		unlock_page(page);
		put_page(page);
1505
		page_nid = -1;
1506
		goto out_unlock;
1507
	}
1508

1509 1510 1511 1512 1513 1514 1515
	/* 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;
	}

1516 1517
	/*
	 * Migrate the THP to the requested node, returns with page unlocked
1518
	 * and access rights restored.
1519
	 */
1520
	spin_unlock(ptl);
1521
	migrated = migrate_misplaced_transhuge_page(mm, vma,
1522
				pmdp, pmd, addr, page, target_nid);
1523 1524
	if (migrated) {
		flags |= TNF_MIGRATED;
1525
		page_nid = target_nid;
1526 1527
	} else
		flags |= TNF_MIGRATE_FAIL;
1528

1529
	goto out;
1530
clear_pmdnuma:
1531
	BUG_ON(!PageLocked(page));
1532
	was_writable = pmd_write(pmd);
1533
	pmd = pmd_modify(pmd, vma->vm_page_prot);
1534
	pmd = pmd_mkyoung(pmd);
1535 1536
	if (was_writable)
		pmd = pmd_mkwrite(pmd);
1537 1538
	set_pmd_at(mm, haddr, pmdp, pmd);
	update_mmu_cache_pmd(vma, addr, pmdp);
1539
	unlock_page(page);
1540
out_unlock:
1541
	spin_unlock(ptl);
1542 1543 1544 1545 1546

out:
	if (anon_vma)
		page_unlock_anon_vma_read(anon_vma);

1547
	if (page_nid != -1)
1548
		task_numa_fault(last_cpupid, page_nid, HPAGE_PMD_NR, flags);
1549

1550 1551 1552
	return 0;
}

1553 1554 1555 1556 1557 1558 1559 1560 1561 1562
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;

1563 1564
	ptl = pmd_trans_huge_lock(pmd, vma);
	if (!ptl)
1565
		goto out_unlocked;
1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624

	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);
		if (split_huge_page(page)) {
			put_page(page);
			unlock_page(page);
			goto out_unlocked;
		}
		put_page(page);
		unlock_page(page);
		ret = 1;
		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;
}

1625
int zap_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
S
Shaohua Li 已提交
1626
		 pmd_t *pmd, unsigned long addr)
1627
{
1628
	pmd_t orig_pmd;
1629
	spinlock_t *ptl;
1630

1631 1632
	ptl = __pmd_trans_huge_lock(pmd, vma);
	if (!ptl)
1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645
		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))
1646
			put_huge_zero_page();
1647 1648 1649 1650 1651 1652 1653
	} 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);
		put_huge_zero_page();
	} else {
		struct page *page = pmd_page(orig_pmd);
1654
		page_remove_rmap(page, true);
1655 1656 1657 1658 1659 1660 1661
		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);
1662
	}
1663
	return 1;
1664 1665
}

1666
bool move_huge_pmd(struct vm_area_struct *vma, struct vm_area_struct *new_vma,
1667 1668 1669 1670
		  unsigned long old_addr,
		  unsigned long new_addr, unsigned long old_end,
		  pmd_t *old_pmd, pmd_t *new_pmd)
{
1671
	spinlock_t *old_ptl, *new_ptl;
1672 1673 1674 1675 1676 1677 1678 1679
	pmd_t pmd;

	struct mm_struct *mm = vma->vm_mm;

	if ((old_addr & ~HPAGE_PMD_MASK) ||
	    (new_addr & ~HPAGE_PMD_MASK) ||
	    old_end - old_addr < HPAGE_PMD_SIZE ||
	    (new_vma->vm_flags & VM_NOHUGEPAGE))
1680
		return false;
1681 1682 1683 1684 1685 1686 1687

	/*
	 * 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));
1688
		return false;
1689 1690
	}

1691 1692 1693 1694
	/*
	 * We don't have to worry about the ordering of src and dst
	 * ptlocks because exclusive mmap_sem prevents deadlock.
	 */
1695 1696
	old_ptl = __pmd_trans_huge_lock(old_pmd, vma);
	if (old_ptl) {
1697 1698 1699
		new_ptl = pmd_lockptr(mm, new_pmd);
		if (new_ptl != old_ptl)
			spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
1700
		pmd = pmdp_huge_get_and_clear(mm, old_addr, old_pmd);
1701
		VM_BUG_ON(!pmd_none(*new_pmd));
1702

1703 1704
		if (pmd_move_must_withdraw(new_ptl, old_ptl)) {
			pgtable_t pgtable;
1705 1706 1707
			pgtable = pgtable_trans_huge_withdraw(mm, old_pmd);
			pgtable_trans_huge_deposit(mm, new_pmd, pgtable);
		}
1708 1709 1710
		set_pmd_at(mm, new_addr, new_pmd, pmd_mksoft_dirty(pmd));
		if (new_ptl != old_ptl)
			spin_unlock(new_ptl);
1711
		spin_unlock(old_ptl);
1712
		return true;
1713
	}
1714
	return false;
1715 1716
}

1717 1718 1719 1720 1721 1722
/*
 * 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
 */
1723
int change_huge_pmd(struct vm_area_struct *vma, pmd_t *pmd,
1724
		unsigned long addr, pgprot_t newprot, int prot_numa)
1725 1726
{
	struct mm_struct *mm = vma->vm_mm;
1727
	spinlock_t *ptl;
1728 1729
	int ret = 0;

1730 1731
	ptl = __pmd_trans_huge_lock(pmd, vma);
	if (ptl) {
1732
		pmd_t entry;
1733
		bool preserve_write = prot_numa && pmd_write(*pmd);
1734
		ret = 1;
1735 1736 1737 1738 1739 1740 1741 1742

		/*
		 * 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);
1743
			return ret;
1744 1745
		}

1746
		if (!prot_numa || !pmd_protnone(*pmd)) {
1747
			entry = pmdp_huge_get_and_clear_notify(mm, addr, pmd);
1748
			entry = pmd_modify(entry, newprot);
1749 1750
			if (preserve_write)
				entry = pmd_mkwrite(entry);
1751 1752
			ret = HPAGE_PMD_NR;
			set_pmd_at(mm, addr, pmd, entry);
1753
			BUG_ON(!preserve_write && pmd_write(entry));
1754
		}
1755
		spin_unlock(ptl);
1756 1757 1758 1759 1760 1761
	}

	return ret;
}

/*
1762
 * Returns true if a given pmd maps a thp, false otherwise.
1763
 *
1764 1765
 * Note that if it returns true, this routine returns without unlocking page
 * table lock. So callers must unlock it.
1766
 */
1767
spinlock_t *__pmd_trans_huge_lock(pmd_t *pmd, struct vm_area_struct *vma)
1768
{
1769 1770
	spinlock_t *ptl;
	ptl = pmd_lock(vma->vm_mm, pmd);
1771
	if (likely(pmd_trans_huge(*pmd) || pmd_devmap(*pmd)))
1772 1773 1774
		return ptl;
	spin_unlock(ptl);
	return NULL;
1775 1776
}

1777
#define VM_NO_THP (VM_SPECIAL | VM_HUGETLB | VM_SHARED | VM_MAYSHARE)
1778

1779 1780
int hugepage_madvise(struct vm_area_struct *vma,
		     unsigned long *vm_flags, int advice)
A
Andrea Arcangeli 已提交
1781
{
A
Andrea Arcangeli 已提交
1782 1783
	switch (advice) {
	case MADV_HUGEPAGE:
1784 1785 1786 1787 1788 1789 1790 1791 1792
#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
1793 1794 1795
		/*
		 * Be somewhat over-protective like KSM for now!
		 */
1796
		if (*vm_flags & VM_NO_THP)
A
Andrea Arcangeli 已提交
1797 1798 1799
			return -EINVAL;
		*vm_flags &= ~VM_NOHUGEPAGE;
		*vm_flags |= VM_HUGEPAGE;
1800 1801 1802 1803 1804
		/*
		 * If the vma become good for khugepaged to scan,
		 * register it here without waiting a page fault that
		 * may not happen any time soon.
		 */
1805
		if (unlikely(khugepaged_enter_vma_merge(vma, *vm_flags)))
1806
			return -ENOMEM;
A
Andrea Arcangeli 已提交
1807 1808 1809 1810 1811
		break;
	case MADV_NOHUGEPAGE:
		/*
		 * Be somewhat over-protective like KSM for now!
		 */
1812
		if (*vm_flags & VM_NO_THP)
A
Andrea Arcangeli 已提交
1813 1814 1815
			return -EINVAL;
		*vm_flags &= ~VM_HUGEPAGE;
		*vm_flags |= VM_NOHUGEPAGE;
1816 1817 1818 1819 1820
		/*
		 * 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
1821 1822
		break;
	}
A
Andrea Arcangeli 已提交
1823 1824 1825 1826

	return 0;
}

A
Andrea Arcangeli 已提交
1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837
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;
}

1838 1839 1840 1841 1842
static void __init khugepaged_slab_exit(void)
{
	kmem_cache_destroy(mm_slot_cache);
}

A
Andrea Arcangeli 已提交
1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858
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;

1859
	hash_for_each_possible(mm_slots_hash, mm_slot, hash, (unsigned long)mm)
A
Andrea Arcangeli 已提交
1860 1861
		if (mm == mm_slot->mm)
			return mm_slot;
1862

A
Andrea Arcangeli 已提交
1863 1864 1865 1866 1867 1868 1869
	return NULL;
}

static void insert_to_mm_slots_hash(struct mm_struct *mm,
				    struct mm_slot *mm_slot)
{
	mm_slot->mm = mm;
1870
	hash_add(mm_slots_hash, &mm_slot->hash, (long)mm);
A
Andrea Arcangeli 已提交
1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887
}

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
1888
	VM_BUG_ON_MM(khugepaged_test_exit(mm), mm);
A
Andrea Arcangeli 已提交
1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910
	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;
}

1911 1912
int khugepaged_enter_vma_merge(struct vm_area_struct *vma,
			       unsigned long vm_flags)
A
Andrea Arcangeli 已提交
1913 1914 1915 1916 1917 1918 1919 1920
{
	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;
1921
	if (vma->vm_ops)
A
Andrea Arcangeli 已提交
1922 1923
		/* khugepaged not yet working on file or special mappings */
		return 0;
1924
	VM_BUG_ON_VMA(vm_flags & VM_NO_THP, vma);
A
Andrea Arcangeli 已提交
1925 1926 1927
	hstart = (vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK;
	hend = vma->vm_end & HPAGE_PMD_MASK;
	if (hstart < hend)
1928
		return khugepaged_enter(vma, vm_flags);
A
Andrea Arcangeli 已提交
1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939
	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) {
1940
		hash_del(&mm_slot->hash);
A
Andrea Arcangeli 已提交
1941 1942 1943
		list_del(&mm_slot->mm_node);
		free = 1;
	}
1944
	spin_unlock(&khugepaged_mm_lock);
A
Andrea Arcangeli 已提交
1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960

	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);
1961
	}
A
Andrea Arcangeli 已提交
1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975
}

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;
1976
		if (!pte_none(pteval) && !is_zero_pfn(pte_pfn(pteval)))
A
Andrea Arcangeli 已提交
1977 1978 1979 1980 1981 1982 1983 1984
			release_pte_page(pte_page(pteval));
	}
}

static int __collapse_huge_page_isolate(struct vm_area_struct *vma,
					unsigned long address,
					pte_t *pte)
{
1985
	struct page *page = NULL;
A
Andrea Arcangeli 已提交
1986
	pte_t *_pte;
1987
	int none_or_zero = 0, result = 0;
1988
	bool referenced = false, writable = false;
1989

A
Andrea Arcangeli 已提交
1990 1991 1992
	for (_pte = pte; _pte < pte+HPAGE_PMD_NR;
	     _pte++, address += PAGE_SIZE) {
		pte_t pteval = *_pte;
1993 1994
		if (pte_none(pteval) || (pte_present(pteval) &&
				is_zero_pfn(pte_pfn(pteval)))) {
1995
			if (!userfaultfd_armed(vma) &&
1996
			    ++none_or_zero <= khugepaged_max_ptes_none) {
A
Andrea Arcangeli 已提交
1997
				continue;
1998 1999
			} else {
				result = SCAN_EXCEED_NONE_PTE;
A
Andrea Arcangeli 已提交
2000
				goto out;
2001
			}
A
Andrea Arcangeli 已提交
2002
		}
2003 2004
		if (!pte_present(pteval)) {
			result = SCAN_PTE_NON_PRESENT;
A
Andrea Arcangeli 已提交
2005
			goto out;
2006
		}
A
Andrea Arcangeli 已提交
2007
		page = vm_normal_page(vma, address, pteval);
2008 2009
		if (unlikely(!page)) {
			result = SCAN_PAGE_NULL;
A
Andrea Arcangeli 已提交
2010
			goto out;
2011
		}
2012

2013 2014 2015
		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
2016 2017 2018 2019 2020 2021 2022

		/*
		 * 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.
		 */
2023 2024
		if (!trylock_page(page)) {
			result = SCAN_PAGE_LOCK;
A
Andrea Arcangeli 已提交
2025
			goto out;
2026
		}
2027 2028 2029 2030 2031 2032 2033 2034

		/*
		 * 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);
2035
			result = SCAN_PAGE_COUNT;
2036 2037 2038 2039 2040 2041 2042
			goto out;
		}
		if (pte_write(pteval)) {
			writable = true;
		} else {
			if (PageSwapCache(page) && !reuse_swap_page(page)) {
				unlock_page(page);
2043
				result = SCAN_SWAP_CACHE_PAGE;
2044 2045 2046 2047 2048 2049 2050 2051
				goto out;
			}
			/*
			 * Page is not in the swap cache. It can be collapsed
			 * into a THP.
			 */
		}

A
Andrea Arcangeli 已提交
2052 2053 2054 2055 2056 2057
		/*
		 * Isolate the page to avoid collapsing an hugepage
		 * currently in use by the VM.
		 */
		if (isolate_lru_page(page)) {
			unlock_page(page);
2058
			result = SCAN_DEL_PAGE_LRU;
A
Andrea Arcangeli 已提交
2059 2060 2061 2062
			goto out;
		}
		/* 0 stands for page_is_file_cache(page) == false */
		inc_zone_page_state(page, NR_ISOLATED_ANON + 0);
2063 2064
		VM_BUG_ON_PAGE(!PageLocked(page), page);
		VM_BUG_ON_PAGE(PageLRU(page), page);
A
Andrea Arcangeli 已提交
2065 2066

		/* If there is no mapped pte young don't collapse the page */
2067 2068
		if (pte_young(pteval) ||
		    page_is_young(page) || PageReferenced(page) ||
A
Andrea Arcangeli 已提交
2069
		    mmu_notifier_test_young(vma->vm_mm, address))
2070
			referenced = true;
A
Andrea Arcangeli 已提交
2071
	}
2072 2073 2074
	if (likely(writable)) {
		if (likely(referenced)) {
			result = SCAN_SUCCEED;
2075
			trace_mm_collapse_huge_page_isolate(page, none_or_zero,
2076 2077 2078 2079 2080 2081 2082
							    referenced, writable, result);
			return 1;
		}
	} else {
		result = SCAN_PAGE_RO;
	}

A
Andrea Arcangeli 已提交
2083
out:
2084
	release_pte_pages(pte, _pte);
2085
	trace_mm_collapse_huge_page_isolate(page, none_or_zero,
2086
					    referenced, writable, result);
2087
	return 0;
A
Andrea Arcangeli 已提交
2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099
}

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;

2100
		if (pte_none(pteval) || is_zero_pfn(pte_pfn(pteval))) {
A
Andrea Arcangeli 已提交
2101 2102
			clear_user_highpage(page, address);
			add_mm_counter(vma->vm_mm, MM_ANONPAGES, 1);
2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114
			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
2115 2116 2117
		} else {
			src_page = pte_page(pteval);
			copy_user_highpage(page, src_page, address, vma);
2118
			VM_BUG_ON_PAGE(page_mapcount(src_page) != 1, src_page);
A
Andrea Arcangeli 已提交
2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130
			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);
2131
			page_remove_rmap(src_page, false);
A
Andrea Arcangeli 已提交
2132 2133 2134 2135 2136 2137 2138 2139 2140
			spin_unlock(ptl);
			free_page_and_swap_cache(src_page);
		}

		address += PAGE_SIZE;
		page++;
	}
}

2141
static void khugepaged_alloc_sleep(void)
A
Andrea Arcangeli 已提交
2142
{
2143 2144 2145 2146 2147 2148
	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);
2149
}
A
Andrea Arcangeli 已提交
2150

2151 2152
static int khugepaged_node_load[MAX_NUMNODES];

2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176
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;
}

2177
#ifdef CONFIG_NUMA
2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202
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;
}

2203 2204 2205 2206 2207 2208 2209
static bool khugepaged_prealloc_page(struct page **hpage, bool *wait)
{
	if (IS_ERR(*hpage)) {
		if (!*wait)
			return false;

		*wait = false;
2210
		*hpage = NULL;
2211 2212 2213 2214 2215 2216 2217 2218 2219
		khugepaged_alloc_sleep();
	} else if (*hpage) {
		put_page(*hpage);
		*hpage = NULL;
	}

	return true;
}

2220 2221
static struct page *
khugepaged_alloc_page(struct page **hpage, gfp_t gfp, struct mm_struct *mm,
2222
		       unsigned long address, int node)
2223
{
2224
	VM_BUG_ON_PAGE(*hpage, *hpage);
2225

2226
	/*
2227 2228 2229 2230
	 * 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.
2231
	 */
2232 2233
	up_read(&mm->mmap_sem);

2234
	*hpage = __alloc_pages_node(node, gfp, HPAGE_PMD_ORDER);
2235
	if (unlikely(!*hpage)) {
2236
		count_vm_event(THP_COLLAPSE_ALLOC_FAILED);
2237
		*hpage = ERR_PTR(-ENOMEM);
2238
		return NULL;
2239
	}
2240

2241
	prep_transhuge_page(*hpage);
2242
	count_vm_event(THP_COLLAPSE_ALLOC);
2243 2244 2245
	return *hpage;
}
#else
2246 2247 2248 2249 2250
static int khugepaged_find_target_node(void)
{
	return 0;
}

2251 2252
static inline struct page *alloc_hugepage(int defrag)
{
2253 2254 2255 2256 2257 2258
	struct page *page;

	page = alloc_pages(alloc_hugepage_gfpmask(defrag, 0), HPAGE_PMD_ORDER);
	if (page)
		prep_transhuge_page(page);
	return page;
2259 2260
}

2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291
static struct page *khugepaged_alloc_hugepage(bool *wait)
{
	struct page *hpage;

	do {
		hpage = alloc_hugepage(khugepaged_defrag());
		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;
}

2292 2293
static struct page *
khugepaged_alloc_page(struct page **hpage, gfp_t gfp, struct mm_struct *mm,
2294
		       unsigned long address, int node)
2295 2296 2297
{
	up_read(&mm->mmap_sem);
	VM_BUG_ON(!*hpage);
2298

2299 2300
	return  *hpage;
}
2301 2302
#endif

B
Bob Liu 已提交
2303 2304 2305 2306 2307 2308 2309 2310 2311
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;
2312
	VM_BUG_ON_VMA(vma->vm_flags & VM_NO_THP, vma);
B
Bob Liu 已提交
2313 2314 2315
	return true;
}

2316 2317 2318 2319 2320 2321 2322 2323 2324 2325
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;
2326
	spinlock_t *pmd_ptl, *pte_ptl;
2327
	int isolated = 0, result = 0;
2328
	unsigned long hstart, hend;
2329
	struct mem_cgroup *memcg;
2330 2331
	unsigned long mmun_start;	/* For mmu_notifiers */
	unsigned long mmun_end;		/* For mmu_notifiers */
2332
	gfp_t gfp;
2333 2334 2335

	VM_BUG_ON(address & ~HPAGE_PMD_MASK);

2336 2337 2338 2339
	/* Only allocate from the target node */
	gfp = alloc_hugepage_gfpmask(khugepaged_defrag(), __GFP_OTHER_NODE) |
		__GFP_THISNODE;

2340
	/* release the mmap_sem read lock. */
2341
	new_page = khugepaged_alloc_page(hpage, gfp, mm, address, node);
2342 2343 2344 2345
	if (!new_page) {
		result = SCAN_ALLOC_HUGE_PAGE_FAIL;
		goto out_nolock;
	}
2346

2347
	if (unlikely(mem_cgroup_try_charge(new_page, mm, gfp, &memcg, true))) {
2348 2349 2350
		result = SCAN_CGROUP_CHARGE_FAIL;
		goto out_nolock;
	}
A
Andrea Arcangeli 已提交
2351 2352 2353 2354 2355 2356 2357

	/*
	 * 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);
2358 2359
	if (unlikely(khugepaged_test_exit(mm))) {
		result = SCAN_ANY_PROCESS;
A
Andrea Arcangeli 已提交
2360
		goto out;
2361
	}
A
Andrea Arcangeli 已提交
2362 2363

	vma = find_vma(mm, address);
2364 2365
	if (!vma) {
		result = SCAN_VMA_NULL;
2366
		goto out;
2367
	}
A
Andrea Arcangeli 已提交
2368 2369
	hstart = (vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK;
	hend = vma->vm_end & HPAGE_PMD_MASK;
2370 2371
	if (address < hstart || address + HPAGE_PMD_SIZE > hend) {
		result = SCAN_ADDRESS_RANGE;
A
Andrea Arcangeli 已提交
2372
		goto out;
2373 2374 2375
	}
	if (!hugepage_vma_check(vma)) {
		result = SCAN_VMA_CHECK;
2376
		goto out;
2377
	}
B
Bob Liu 已提交
2378
	pmd = mm_find_pmd(mm, address);
2379 2380
	if (!pmd) {
		result = SCAN_PMD_NULL;
A
Andrea Arcangeli 已提交
2381
		goto out;
2382
	}
A
Andrea Arcangeli 已提交
2383

2384
	anon_vma_lock_write(vma->anon_vma);
A
Andrea Arcangeli 已提交
2385 2386

	pte = pte_offset_map(pmd, address);
2387
	pte_ptl = pte_lockptr(mm, pmd);
A
Andrea Arcangeli 已提交
2388

2389 2390 2391
	mmun_start = address;
	mmun_end   = address + HPAGE_PMD_SIZE;
	mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
2392
	pmd_ptl = pmd_lock(mm, pmd); /* probably unnecessary */
A
Andrea Arcangeli 已提交
2393 2394 2395 2396 2397 2398
	/*
	 * 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.
	 */
2399
	_pmd = pmdp_collapse_flush(vma, address, pmd);
2400
	spin_unlock(pmd_ptl);
2401
	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
A
Andrea Arcangeli 已提交
2402

2403
	spin_lock(pte_ptl);
A
Andrea Arcangeli 已提交
2404
	isolated = __collapse_huge_page_isolate(vma, address, pte);
2405
	spin_unlock(pte_ptl);
A
Andrea Arcangeli 已提交
2406 2407

	if (unlikely(!isolated)) {
2408
		pte_unmap(pte);
2409
		spin_lock(pmd_ptl);
A
Andrea Arcangeli 已提交
2410
		BUG_ON(!pmd_none(*pmd));
2411 2412 2413 2414 2415 2416
		/*
		 * 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));
2417
		spin_unlock(pmd_ptl);
2418
		anon_vma_unlock_write(vma->anon_vma);
2419
		result = SCAN_FAIL;
2420
		goto out;
A
Andrea Arcangeli 已提交
2421 2422 2423 2424 2425 2426
	}

	/*
	 * All pages are isolated and locked so anon_vma rmap
	 * can't run anymore.
	 */
2427
	anon_vma_unlock_write(vma->anon_vma);
A
Andrea Arcangeli 已提交
2428

2429
	__collapse_huge_page_copy(pte, new_page, vma, address, pte_ptl);
2430
	pte_unmap(pte);
A
Andrea Arcangeli 已提交
2431 2432 2433
	__SetPageUptodate(new_page);
	pgtable = pmd_pgtable(_pmd);

2434 2435
	_pmd = mk_huge_pmd(new_page, vma->vm_page_prot);
	_pmd = maybe_pmd_mkwrite(pmd_mkdirty(_pmd), vma);
A
Andrea Arcangeli 已提交
2436 2437 2438 2439 2440 2441 2442 2443

	/*
	 * 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();

2444
	spin_lock(pmd_ptl);
A
Andrea Arcangeli 已提交
2445
	BUG_ON(!pmd_none(*pmd));
2446
	page_add_new_anon_rmap(new_page, vma, address, true);
2447
	mem_cgroup_commit_charge(new_page, memcg, false, true);
2448
	lru_cache_add_active_or_unevictable(new_page, vma);
2449
	pgtable_trans_huge_deposit(mm, pmd, pgtable);
A
Andrea Arcangeli 已提交
2450
	set_pmd_at(mm, address, pmd, _pmd);
2451
	update_mmu_cache_pmd(vma, address, pmd);
2452
	spin_unlock(pmd_ptl);
A
Andrea Arcangeli 已提交
2453 2454

	*hpage = NULL;
2455

A
Andrea Arcangeli 已提交
2456
	khugepaged_pages_collapsed++;
2457
	result = SCAN_SUCCEED;
2458
out_up_write:
A
Andrea Arcangeli 已提交
2459
	up_write(&mm->mmap_sem);
2460
	trace_mm_collapse_huge_page(mm, isolated, result);
2461 2462
	return;

2463 2464 2465
out_nolock:
	trace_mm_collapse_huge_page(mm, isolated, result);
	return;
2466
out:
2467
	mem_cgroup_cancel_charge(new_page, memcg, true);
2468
	goto out_up_write;
A
Andrea Arcangeli 已提交
2469 2470 2471 2472 2473 2474 2475 2476 2477
}

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;
2478 2479
	int ret = 0, none_or_zero = 0, result = 0;
	struct page *page = NULL;
A
Andrea Arcangeli 已提交
2480 2481
	unsigned long _address;
	spinlock_t *ptl;
D
David Rientjes 已提交
2482
	int node = NUMA_NO_NODE;
2483
	bool writable = false, referenced = false;
A
Andrea Arcangeli 已提交
2484 2485 2486

	VM_BUG_ON(address & ~HPAGE_PMD_MASK);

B
Bob Liu 已提交
2487
	pmd = mm_find_pmd(mm, address);
2488 2489
	if (!pmd) {
		result = SCAN_PMD_NULL;
A
Andrea Arcangeli 已提交
2490
		goto out;
2491
	}
A
Andrea Arcangeli 已提交
2492

2493
	memset(khugepaged_node_load, 0, sizeof(khugepaged_node_load));
A
Andrea Arcangeli 已提交
2494 2495 2496 2497
	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;
2498
		if (pte_none(pteval) || is_zero_pfn(pte_pfn(pteval))) {
2499
			if (!userfaultfd_armed(vma) &&
2500
			    ++none_or_zero <= khugepaged_max_ptes_none) {
A
Andrea Arcangeli 已提交
2501
				continue;
2502 2503
			} else {
				result = SCAN_EXCEED_NONE_PTE;
A
Andrea Arcangeli 已提交
2504
				goto out_unmap;
2505
			}
A
Andrea Arcangeli 已提交
2506
		}
2507 2508
		if (!pte_present(pteval)) {
			result = SCAN_PTE_NON_PRESENT;
A
Andrea Arcangeli 已提交
2509
			goto out_unmap;
2510
		}
2511 2512 2513
		if (pte_write(pteval))
			writable = true;

A
Andrea Arcangeli 已提交
2514
		page = vm_normal_page(vma, _address, pteval);
2515 2516
		if (unlikely(!page)) {
			result = SCAN_PAGE_NULL;
A
Andrea Arcangeli 已提交
2517
			goto out_unmap;
2518
		}
2519 2520 2521 2522 2523 2524 2525

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

2526
		/*
2527 2528 2529 2530
		 * 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.
2531
		 */
2532
		node = page_to_nid(page);
2533 2534
		if (khugepaged_scan_abort(node)) {
			result = SCAN_SCAN_ABORT;
2535
			goto out_unmap;
2536
		}
2537
		khugepaged_node_load[node]++;
2538 2539 2540 2541 2542 2543
		if (!PageLRU(page)) {
			result = SCAN_SCAN_ABORT;
			goto out_unmap;
		}
		if (PageLocked(page)) {
			result = SCAN_PAGE_LOCK;
A
Andrea Arcangeli 已提交
2544
			goto out_unmap;
2545 2546 2547 2548 2549 2550
		}
		if (!PageAnon(page)) {
			result = SCAN_PAGE_ANON;
			goto out_unmap;
		}

2551 2552 2553 2554 2555
		/*
		 * 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.
		 */
2556 2557
		if (page_count(page) != 1 + !!PageSwapCache(page)) {
			result = SCAN_PAGE_COUNT;
A
Andrea Arcangeli 已提交
2558
			goto out_unmap;
2559
		}
2560 2561
		if (pte_young(pteval) ||
		    page_is_young(page) || PageReferenced(page) ||
A
Andrea Arcangeli 已提交
2562
		    mmu_notifier_test_young(vma->vm_mm, address))
2563
			referenced = true;
A
Andrea Arcangeli 已提交
2564
	}
2565 2566 2567 2568 2569 2570 2571 2572 2573 2574
	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
2575 2576
out_unmap:
	pte_unmap_unlock(pte, ptl);
2577 2578
	if (ret) {
		node = khugepaged_find_target_node();
2579
		/* collapse_huge_page will return with the mmap_sem released */
2580
		collapse_huge_page(mm, address, hpage, vma, node);
2581
	}
A
Andrea Arcangeli 已提交
2582
out:
2583
	trace_mm_khugepaged_scan_pmd(mm, page, writable, referenced,
2584
				     none_or_zero, result);
A
Andrea Arcangeli 已提交
2585 2586 2587 2588 2589 2590 2591
	return ret;
}

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

2592
	VM_BUG_ON(NR_CPUS != 1 && !spin_is_locked(&khugepaged_mm_lock));
A
Andrea Arcangeli 已提交
2593 2594 2595

	if (khugepaged_test_exit(mm)) {
		/* free mm_slot */
2596
		hash_del(&mm_slot->hash);
A
Andrea Arcangeli 已提交
2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612
		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)
2613 2614
	__releases(&khugepaged_mm_lock)
	__acquires(&khugepaged_mm_lock)
A
Andrea Arcangeli 已提交
2615 2616 2617 2618 2619 2620 2621
{
	struct mm_slot *mm_slot;
	struct mm_struct *mm;
	struct vm_area_struct *vma;
	int progress = 0;

	VM_BUG_ON(!pages);
2622
	VM_BUG_ON(NR_CPUS != 1 && !spin_is_locked(&khugepaged_mm_lock));
A
Andrea Arcangeli 已提交
2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649

	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
2650 2651
		if (!hugepage_vma_check(vma)) {
skip:
A
Andrea Arcangeli 已提交
2652 2653 2654 2655 2656
			progress++;
			continue;
		}
		hstart = (vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK;
		hend = vma->vm_end & HPAGE_PMD_MASK;
2657 2658 2659 2660
		if (hstart >= hend)
			goto skip;
		if (khugepaged_scan.address > hend)
			goto skip;
A
Andrea Arcangeli 已提交
2661 2662
		if (khugepaged_scan.address < hstart)
			khugepaged_scan.address = hstart;
2663
		VM_BUG_ON(khugepaged_scan.address & ~HPAGE_PMD_MASK);
A
Andrea Arcangeli 已提交
2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691

		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);
2692
	VM_BUG_ON(khugepaged_scan.mm_slot != mm_slot);
A
Andrea Arcangeli 已提交
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 2720 2721 2722 2723 2724 2725 2726 2727
	/*
	 * 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) ||
2728
		kthread_should_stop();
A
Andrea Arcangeli 已提交
2729 2730
}

2731
static void khugepaged_do_scan(void)
A
Andrea Arcangeli 已提交
2732
{
2733
	struct page *hpage = NULL;
A
Andrea Arcangeli 已提交
2734 2735
	unsigned int progress = 0, pass_through_head = 0;
	unsigned int pages = khugepaged_pages_to_scan;
2736
	bool wait = true;
A
Andrea Arcangeli 已提交
2737 2738 2739 2740

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

	while (progress < pages) {
2741
		if (!khugepaged_prealloc_page(&hpage, &wait))
2742
			break;
2743

2744
		cond_resched();
A
Andrea Arcangeli 已提交
2745

2746
		if (unlikely(kthread_should_stop() || try_to_freeze()))
2747 2748
			break;

A
Andrea Arcangeli 已提交
2749 2750 2751 2752 2753 2754
		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,
2755
							    &hpage);
A
Andrea Arcangeli 已提交
2756 2757 2758 2759 2760
		else
			progress = pages;
		spin_unlock(&khugepaged_mm_lock);
	}

2761 2762
	if (!IS_ERR_OR_NULL(hpage))
		put_page(hpage);
2763 2764
}

2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780
static void khugepaged_wait_work(void)
{
	if (khugepaged_has_work()) {
		if (!khugepaged_scan_sleep_millisecs)
			return;

		wait_event_freezable_timeout(khugepaged_wait,
					     kthread_should_stop(),
			msecs_to_jiffies(khugepaged_scan_sleep_millisecs));
		return;
	}

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

A
Andrea Arcangeli 已提交
2781 2782 2783 2784
static int khugepaged(void *none)
{
	struct mm_slot *mm_slot;

2785
	set_freezable();
2786
	set_user_nice(current, MAX_NICE);
A
Andrea Arcangeli 已提交
2787

X
Xiao Guangrong 已提交
2788 2789 2790 2791
	while (!kthread_should_stop()) {
		khugepaged_do_scan();
		khugepaged_wait_work();
	}
A
Andrea Arcangeli 已提交
2792 2793 2794 2795 2796 2797 2798 2799 2800 2801

	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;
}

2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830
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,
2831
		unsigned long haddr, bool freeze)
2832 2833 2834 2835 2836
{
	struct mm_struct *mm = vma->vm_mm;
	struct page *page;
	pgtable_t pgtable;
	pmd_t _pmd;
2837
	bool young, write, dirty;
2838 2839 2840 2841 2842
	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);
2843
	VM_BUG_ON(!pmd_trans_huge(*pmd) && !pmd_devmap(*pmd));
2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860

	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);
	atomic_add(HPAGE_PMD_NR - 1, &page->_count);
	write = pmd_write(*pmd);
	young = pmd_young(*pmd);
2861
	dirty = pmd_dirty(*pmd);
2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872

	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 entry, *pte;
		/*
		 * Note that NUMA hinting access restrictions are not
		 * transferred to avoid any possibility of altering
		 * permissions across VMAs.
		 */
2873 2874 2875 2876 2877 2878
		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);
2879
			entry = maybe_mkwrite(entry, vma);
2880 2881 2882 2883 2884
			if (!write)
				entry = pte_wrprotect(entry);
			if (!young)
				entry = pte_mkold(entry);
		}
2885 2886
		if (dirty)
			SetPageDirty(page + i);
2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913
		pte = pte_offset_map(&_pmd, haddr);
		BUG_ON(!pte_none(*pte));
		set_pte_at(mm, haddr, pte, entry);
		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 */
2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935
	/*
	 * 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);
2936
	pmd_populate(mm, pmd, pgtable);
2937 2938 2939 2940 2941 2942 2943

	if (freeze) {
		for (i = 0; i < HPAGE_PMD_NR; i++, haddr += PAGE_SIZE) {
			page_remove_rmap(page + i, false);
			put_page(page + i);
		}
	}
2944 2945 2946 2947 2948 2949 2950
}

void __split_huge_pmd(struct vm_area_struct *vma, pmd_t *pmd,
		unsigned long address)
{
	spinlock_t *ptl;
	struct mm_struct *mm = vma->vm_mm;
2951
	struct page *page = NULL;
2952 2953 2954 2955
	unsigned long haddr = address & HPAGE_PMD_MASK;

	mmu_notifier_invalidate_range_start(mm, haddr, haddr + HPAGE_PMD_SIZE);
	ptl = pmd_lock(mm, pmd);
2956 2957 2958 2959 2960 2961 2962
	if (pmd_trans_huge(*pmd)) {
		page = pmd_page(*pmd);
		if (PageMlocked(page))
			get_page(page);
		else
			page = NULL;
	} else if (!pmd_devmap(*pmd))
2963 2964 2965
		goto out;
	__split_huge_pmd_locked(vma, pmd, haddr, false);
out:
2966 2967
	spin_unlock(ptl);
	mmu_notifier_invalidate_range_end(mm, haddr, haddr + HPAGE_PMD_SIZE);
2968 2969 2970 2971 2972 2973
	if (page) {
		lock_page(page);
		munlock_vma_page(page);
		unlock_page(page);
		put_page(page);
	}
2974 2975
}

2976
static void split_huge_pmd_address(struct vm_area_struct *vma,
2977 2978
				    unsigned long address)
{
2979 2980
	pgd_t *pgd;
	pud_t *pud;
2981 2982 2983 2984
	pmd_t *pmd;

	VM_BUG_ON(!(address & ~HPAGE_PMD_MASK));

2985
	pgd = pgd_offset(vma->vm_mm, address);
2986 2987 2988 2989 2990 2991 2992 2993
	if (!pgd_present(*pgd))
		return;

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

	pmd = pmd_offset(pud, address);
2994
	if (!pmd_present(*pmd) || (!pmd_trans_huge(*pmd) && !pmd_devmap(*pmd)))
2995 2996 2997 2998 2999
		return;
	/*
	 * Caller holds the mmap_sem write mode, so a huge pmd cannot
	 * materialize from under us.
	 */
3000
	split_huge_pmd(vma, pmd, address);
3001 3002
}

3003
void vma_adjust_trans_huge(struct vm_area_struct *vma,
3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015
			     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)
3016
		split_huge_pmd_address(vma, start);
3017 3018 3019 3020 3021 3022 3023 3024 3025

	/*
	 * 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)
3026
		split_huge_pmd_address(vma, end);
3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039

	/*
	 * 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)
3040
			split_huge_pmd_address(next, nstart);
3041 3042
	}
}
3043 3044 3045 3046

static void freeze_page_vma(struct vm_area_struct *vma, struct page *page,
		unsigned long address)
{
3047
	unsigned long haddr = address & HPAGE_PMD_MASK;
3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076
	spinlock_t *ptl;
	pgd_t *pgd;
	pud_t *pud;
	pmd_t *pmd;
	pte_t *pte;
	int i, nr = HPAGE_PMD_NR;

	/* Skip pages which doesn't belong to the VMA */
	if (address < vma->vm_start) {
		int off = (vma->vm_start - address) >> PAGE_SHIFT;
		page += off;
		nr -= off;
		address = vma->vm_start;
	}

	pgd = pgd_offset(vma->vm_mm, address);
	if (!pgd_present(*pgd))
		return;
	pud = pud_offset(pgd, address);
	if (!pud_present(*pud))
		return;
	pmd = pmd_offset(pud, address);
	ptl = pmd_lock(vma->vm_mm, pmd);
	if (!pmd_present(*pmd)) {
		spin_unlock(ptl);
		return;
	}
	if (pmd_trans_huge(*pmd)) {
		if (page == pmd_page(*pmd))
3077
			__split_huge_pmd_locked(vma, pmd, haddr, true);
3078 3079 3080 3081 3082 3083
		spin_unlock(ptl);
		return;
	}
	spin_unlock(ptl);

	pte = pte_offset_map_lock(vma->vm_mm, pmd, address, &ptl);
3084
	for (i = 0; i < nr; i++, address += PAGE_SIZE, page++, pte++) {
3085 3086 3087
		pte_t entry, swp_pte;
		swp_entry_t swp_entry;

3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101
		/*
		 * We've just crossed page table boundary: need to map next one.
		 * It can happen if THP was mremaped to non PMD-aligned address.
		 */
		if (unlikely(address == haddr + HPAGE_PMD_SIZE)) {
			pte_unmap_unlock(pte - 1, ptl);
			pmd = mm_find_pmd(vma->vm_mm, address);
			if (!pmd)
				return;
			pte = pte_offset_map_lock(vma->vm_mm, pmd,
					address, &ptl);
		}

		if (!pte_present(*pte))
3102
			continue;
3103
		if (page_to_pfn(page) != pte_pfn(*pte))
3104 3105
			continue;
		flush_cache_page(vma, address, page_to_pfn(page));
3106
		entry = ptep_clear_flush(vma, address, pte);
3107 3108
		if (pte_dirty(entry))
			SetPageDirty(page);
3109 3110 3111 3112
		swp_entry = make_migration_entry(page, pte_write(entry));
		swp_pte = swp_entry_to_pte(swp_entry);
		if (pte_soft_dirty(entry))
			swp_pte = pte_swp_mksoft_dirty(swp_pte);
3113
		set_pte_at(vma->vm_mm, address, pte, swp_pte);
3114 3115 3116
		page_remove_rmap(page, false);
		put_page(page);
	}
3117
	pte_unmap_unlock(pte - 1, ptl);
3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128
}

static void freeze_page(struct anon_vma *anon_vma, struct page *page)
{
	struct anon_vma_chain *avc;
	pgoff_t pgoff = page_to_pgoff(page);

	VM_BUG_ON_PAGE(!PageHead(page), page);

	anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff,
			pgoff + HPAGE_PMD_NR - 1) {
3129
		unsigned long address = __vma_address(page, avc->vma);
3130 3131

		mmu_notifier_invalidate_range_start(avc->vma->vm_mm,
3132 3133
				address, address + HPAGE_PMD_SIZE);
		freeze_page_vma(avc->vma, page, address);
3134
		mmu_notifier_invalidate_range_end(avc->vma->vm_mm,
3135
				address, address + HPAGE_PMD_SIZE);
3136 3137 3138 3139 3140 3141 3142 3143 3144 3145
	}
}

static void unfreeze_page_vma(struct vm_area_struct *vma, struct page *page,
		unsigned long address)
{
	spinlock_t *ptl;
	pmd_t *pmd;
	pte_t *pte, entry;
	swp_entry_t swp_entry;
3146
	unsigned long haddr = address & HPAGE_PMD_MASK;
3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159
	int i, nr = HPAGE_PMD_NR;

	/* Skip pages which doesn't belong to the VMA */
	if (address < vma->vm_start) {
		int off = (vma->vm_start - address) >> PAGE_SHIFT;
		page += off;
		nr -= off;
		address = vma->vm_start;
	}

	pmd = mm_find_pmd(vma->vm_mm, address);
	if (!pmd)
		return;
3160

3161
	pte = pte_offset_map_lock(vma->vm_mm, pmd, address, &ptl);
3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176
	for (i = 0; i < nr; i++, address += PAGE_SIZE, page++, pte++) {
		/*
		 * We've just crossed page table boundary: need to map next one.
		 * It can happen if THP was mremaped to non-PMD aligned address.
		 */
		if (unlikely(address == haddr + HPAGE_PMD_SIZE)) {
			pte_unmap_unlock(pte - 1, ptl);
			pmd = mm_find_pmd(vma->vm_mm, address);
			if (!pmd)
				return;
			pte = pte_offset_map_lock(vma->vm_mm, pmd,
					address, &ptl);
		}

		if (!is_swap_pte(*pte))
3177 3178
			continue;

3179
		swp_entry = pte_to_swp_entry(*pte);
3180 3181 3182 3183 3184 3185 3186 3187 3188
		if (!is_migration_entry(swp_entry))
			continue;
		if (migration_entry_to_page(swp_entry) != page)
			continue;

		get_page(page);
		page_add_anon_rmap(page, vma, address, false);

		entry = pte_mkold(mk_pte(page, vma->vm_page_prot));
3189 3190
		if (PageDirty(page))
			entry = pte_mkdirty(entry);
3191 3192 3193 3194
		if (is_write_migration_entry(swp_entry))
			entry = maybe_mkwrite(entry, vma);

		flush_dcache_page(page);
3195
		set_pte_at(vma->vm_mm, address, pte, entry);
3196 3197

		/* No need to invalidate - it was non-present before */
3198
		update_mmu_cache(vma, address, pte);
3199
	}
3200
	pte_unmap_unlock(pte - 1, ptl);
3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 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
}

static void unfreeze_page(struct anon_vma *anon_vma, struct page *page)
{
	struct anon_vma_chain *avc;
	pgoff_t pgoff = page_to_pgoff(page);

	anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root,
			pgoff, pgoff + HPAGE_PMD_NR - 1) {
		unsigned long address = __vma_address(page, avc->vma);

		mmu_notifier_invalidate_range_start(avc->vma->vm_mm,
				address, address + HPAGE_PMD_SIZE);
		unfreeze_page_vma(avc->vma, page, address);
		mmu_notifier_invalidate_range_end(avc->vma->vm_mm,
				address, address + HPAGE_PMD_SIZE);
	}
}

static int __split_huge_page_tail(struct page *head, int tail,
		struct lruvec *lruvec, struct list_head *list)
{
	int mapcount;
	struct page *page_tail = head + tail;

	mapcount = atomic_read(&page_tail->_mapcount) + 1;
	VM_BUG_ON_PAGE(atomic_read(&page_tail->_count) != 0, page_tail);

	/*
	 * tail_page->_count is zero and not changing from under us. But
	 * get_page_unless_zero() may be running from under us on the
	 * tail_page. If we used atomic_set() below instead of atomic_add(), we
	 * 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),
	 * it's safer to use atomic_add().
	 */
	atomic_add(mapcount + 1, &page_tail->_count);


	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) |
3251 3252
			 (1L << PG_unevictable) |
			 (1L << PG_dirty)));
3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267

	/*
	 * 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 */
3268
	VM_BUG_ON_PAGE(tail > 2 && page_tail->mapping != TAIL_MAPPING,
3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314 3315 3316 3317 3318 3319
			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);

	return mapcount;
}

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;
	int i, tail_mapcount;

	/* 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);

	tail_mapcount = 0;
	for (i = HPAGE_PMD_NR - 1; i >= 1; i--)
		tail_mapcount += __split_huge_page_tail(head, i, lruvec, list);
	atomic_sub(tail_mapcount, &head->_count);

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

	unfreeze_page(page_anon_vma(head), head);

	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);
	}
}

3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338
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;
}

3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360
/*
 * 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);
3361
	struct pglist_data *pgdata = NODE_DATA(page_to_nid(head));
3362 3363
	struct anon_vma *anon_vma;
	int count, mapcount, ret;
3364
	bool mlocked;
3365
	unsigned long flags;
3366 3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395

	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;
	}

3396
	mlocked = PageMlocked(page);
3397 3398 3399
	freeze_page(anon_vma, head);
	VM_BUG_ON_PAGE(compound_mapcount(head), head);

3400 3401 3402 3403
	/* Make sure the page is not on per-CPU pagevec as it takes pin */
	if (mlocked)
		lru_add_drain();

3404
	/* Prevent deferred_split_scan() touching ->_count */
3405
	spin_lock_irqsave(&pgdata->split_queue_lock, flags);
3406 3407
	count = page_count(head);
	mapcount = total_mapcount(head);
3408
	if (!mapcount && count == 1) {
3409
		if (!list_empty(page_deferred_list(head))) {
3410
			pgdata->split_queue_len--;
3411 3412
			list_del(page_deferred_list(head));
		}
3413
		spin_unlock_irqrestore(&pgdata->split_queue_lock, flags);
3414 3415
		__split_huge_page(page, list);
		ret = 0;
3416
	} else if (IS_ENABLED(CONFIG_DEBUG_VM) && mapcount) {
3417
		spin_unlock_irqrestore(&pgdata->split_queue_lock, flags);
3418 3419 3420 3421
		pr_alert("total_mapcount: %u, page_count(): %u\n",
				mapcount, count);
		if (PageTail(page))
			dump_page(head, NULL);
3422
		dump_page(page, "total_mapcount(head) > 0");
3423 3424
		BUG();
	} else {
3425
		spin_unlock_irqrestore(&pgdata->split_queue_lock, flags);
3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436
		unfreeze_page(anon_vma, head);
		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;
}
3437 3438 3439

void free_transhuge_page(struct page *page)
{
3440
	struct pglist_data *pgdata = NODE_DATA(page_to_nid(page));
3441 3442
	unsigned long flags;

3443
	spin_lock_irqsave(&pgdata->split_queue_lock, flags);
3444
	if (!list_empty(page_deferred_list(page))) {
3445
		pgdata->split_queue_len--;
3446 3447
		list_del(page_deferred_list(page));
	}
3448
	spin_unlock_irqrestore(&pgdata->split_queue_lock, flags);
3449 3450 3451 3452 3453
	free_compound_page(page);
}

void deferred_split_huge_page(struct page *page)
{
3454
	struct pglist_data *pgdata = NODE_DATA(page_to_nid(page));
3455 3456 3457 3458
	unsigned long flags;

	VM_BUG_ON_PAGE(!PageTransHuge(page), page);

3459
	spin_lock_irqsave(&pgdata->split_queue_lock, flags);
3460
	if (list_empty(page_deferred_list(page))) {
3461 3462
		list_add_tail(page_deferred_list(page), &pgdata->split_queue);
		pgdata->split_queue_len++;
3463
	}
3464
	spin_unlock_irqrestore(&pgdata->split_queue_lock, flags);
3465 3466 3467 3468 3469
}

static unsigned long deferred_split_count(struct shrinker *shrink,
		struct shrink_control *sc)
{
3470
	struct pglist_data *pgdata = NODE_DATA(sc->nid);
3471
	return ACCESS_ONCE(pgdata->split_queue_len);
3472 3473 3474 3475 3476
}

static unsigned long deferred_split_scan(struct shrinker *shrink,
		struct shrink_control *sc)
{
3477
	struct pglist_data *pgdata = NODE_DATA(sc->nid);
3478 3479 3480 3481 3482
	unsigned long flags;
	LIST_HEAD(list), *pos, *next;
	struct page *page;
	int split = 0;

3483
	spin_lock_irqsave(&pgdata->split_queue_lock, flags);
3484
	/* Take pin on all head pages to avoid freeing them under us */
3485
	list_for_each_safe(pos, next, &pgdata->split_queue) {
3486 3487
		page = list_entry((void *)pos, struct page, mapping);
		page = compound_head(page);
3488 3489 3490 3491
		if (get_page_unless_zero(page)) {
			list_move(page_deferred_list(page), &list);
		} else {
			/* We lost race with put_compound_page() */
3492
			list_del_init(page_deferred_list(page));
3493
			pgdata->split_queue_len--;
3494
		}
3495 3496
		if (!--sc->nr_to_scan)
			break;
3497
	}
3498
	spin_unlock_irqrestore(&pgdata->split_queue_lock, flags);
3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509

	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);
	}

3510 3511 3512
	spin_lock_irqsave(&pgdata->split_queue_lock, flags);
	list_splice_tail(&list, &pgdata->split_queue);
	spin_unlock_irqrestore(&pgdata->split_queue_lock, flags);
3513

3514 3515 3516 3517 3518 3519 3520
	/*
	 * 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;
3521 3522 3523 3524 3525 3526
}

static struct shrinker deferred_split_shrinker = {
	.count_objects = deferred_split_count,
	.scan_objects = deferred_split_scan,
	.seeks = DEFAULT_SEEKS,
3527
	.flags = SHRINKER_NUMA_AWARE,
3528
};
3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585 3586

#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);
		}
	}

	pr_info("%lu of %lu THP split", split, total);

	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;

	ret = debugfs_create_file("split_huge_pages", 0644, NULL, NULL,
			&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