huge_memory.c 79.5 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 19 20
#include <linux/mm_inline.h>
#include <linux/kthread.h>
#include <linux/khugepaged.h>
21
#include <linux/freezer.h>
A
Andrea Arcangeli 已提交
22
#include <linux/mman.h>
R
Ralf Baechle 已提交
23
#include <linux/pagemap.h>
24
#include <linux/migrate.h>
25
#include <linux/hashtable.h>
26
#include <linux/userfaultfd_k.h>
27

28 29 30 31
#include <asm/tlb.h>
#include <asm/pgalloc.h>
#include "internal.h"

A
Andrea Arcangeli 已提交
32
/*
33 34 35 36 37 38
 * 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
39
 */
40
unsigned long transparent_hugepage_flags __read_mostly =
41
#ifdef CONFIG_TRANSPARENT_HUGEPAGE_ALWAYS
A
Andrea Arcangeli 已提交
42
	(1<<TRANSPARENT_HUGEPAGE_FLAG)|
43 44 45 46
#endif
#ifdef CONFIG_TRANSPARENT_HUGEPAGE_MADVISE
	(1<<TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG)|
#endif
A
Andrea Arcangeli 已提交
47
	(1<<TRANSPARENT_HUGEPAGE_DEFRAG_FLAG)|
48 49
	(1<<TRANSPARENT_HUGEPAGE_DEFRAG_KHUGEPAGED_FLAG)|
	(1<<TRANSPARENT_HUGEPAGE_USE_ZERO_PAGE_FLAG);
A
Andrea Arcangeli 已提交
50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70

/* 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);
71
static void khugepaged_slab_exit(void);
A
Andrea Arcangeli 已提交
72

73 74 75
#define MM_SLOTS_HASH_BITS 10
static __read_mostly DEFINE_HASHTABLE(mm_slots_hash, MM_SLOTS_HASH_BITS);

A
Andrea Arcangeli 已提交
76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101
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;
102 103
};
static struct khugepaged_scan khugepaged_scan = {
A
Andrea Arcangeli 已提交
104 105 106
	.mm_head = LIST_HEAD_INIT(khugepaged_scan.mm_head),
};

107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133

static int set_recommended_min_free_kbytes(void)
{
	struct zone *zone;
	int nr_zones = 0;
	unsigned long recommended_min;

	for_each_populated_zone(zone)
		nr_zones++;

	/* Make sure at least 2 hugepages are free for MIGRATE_RESERVE */
	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);

134 135 136 137 138 139
	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);

140
		min_free_kbytes = recommended_min;
141
	}
142 143 144 145
	setup_per_zone_wmarks();
	return 0;
}

146
static int start_stop_khugepaged(void)
A
Andrea Arcangeli 已提交
147 148 149 150 151 152 153
{
	int err = 0;
	if (khugepaged_enabled()) {
		if (!khugepaged_thread)
			khugepaged_thread = kthread_run(khugepaged, NULL,
							"khugepaged");
		if (unlikely(IS_ERR(khugepaged_thread))) {
154
			pr_err("khugepaged: kthread_run(khugepaged) failed\n");
A
Andrea Arcangeli 已提交
155 156
			err = PTR_ERR(khugepaged_thread);
			khugepaged_thread = NULL;
157
			goto fail;
A
Andrea Arcangeli 已提交
158
		}
159 160

		if (!list_empty(&khugepaged_scan.mm_head))
A
Andrea Arcangeli 已提交
161
			wake_up_interruptible(&khugepaged_wait);
162 163

		set_recommended_min_free_kbytes();
164 165 166 167
	} else if (khugepaged_thread) {
		kthread_stop(khugepaged_thread);
		khugepaged_thread = NULL;
	}
168
fail:
A
Andrea Arcangeli 已提交
169 170
	return err;
}
171

172
static atomic_t huge_zero_refcount;
173
struct page *huge_zero_page __read_mostly;
174

175 176
static inline bool is_huge_zero_pmd(pmd_t pmd)
{
177
	return is_huge_zero_page(pmd_page(pmd));
178 179
}

180
static struct page *get_huge_zero_page(void)
181 182 183 184
{
	struct page *zero_page;
retry:
	if (likely(atomic_inc_not_zero(&huge_zero_refcount)))
185
		return READ_ONCE(huge_zero_page);
186 187

	zero_page = alloc_pages((GFP_TRANSHUGE | __GFP_ZERO) & ~__GFP_MOVABLE,
188
			HPAGE_PMD_ORDER);
189 190
	if (!zero_page) {
		count_vm_event(THP_ZERO_PAGE_ALLOC_FAILED);
191
		return NULL;
192 193
	}
	count_vm_event(THP_ZERO_PAGE_ALLOC);
194
	preempt_disable();
195
	if (cmpxchg(&huge_zero_page, NULL, zero_page)) {
196
		preempt_enable();
197
		__free_pages(zero_page, compound_order(zero_page));
198 199 200 201 202 203
		goto retry;
	}

	/* We take additional reference here. It will be put back by shrinker */
	atomic_set(&huge_zero_refcount, 2);
	preempt_enable();
204
	return READ_ONCE(huge_zero_page);
205 206
}

207
static void put_huge_zero_page(void)
208
{
209 210 211 212 213
	/*
	 * Counter should never go to zero here. Only shrinker can put
	 * last reference.
	 */
	BUG_ON(atomic_dec_and_test(&huge_zero_refcount));
214 215
}

216 217
static unsigned long shrink_huge_zero_page_count(struct shrinker *shrink,
					struct shrink_control *sc)
218
{
219 220 221
	/* we can free zero page only if last reference remains */
	return atomic_read(&huge_zero_refcount) == 1 ? HPAGE_PMD_NR : 0;
}
222

223 224 225
static unsigned long shrink_huge_zero_page_scan(struct shrinker *shrink,
				       struct shrink_control *sc)
{
226
	if (atomic_cmpxchg(&huge_zero_refcount, 1, 0) == 1) {
227 228
		struct page *zero_page = xchg(&huge_zero_page, NULL);
		BUG_ON(zero_page == NULL);
229
		__free_pages(zero_page, compound_order(zero_page));
230
		return HPAGE_PMD_NR;
231 232 233
	}

	return 0;
234 235
}

236
static struct shrinker huge_zero_page_shrinker = {
237 238
	.count_objects = shrink_huge_zero_page_count,
	.scan_objects = shrink_huge_zero_page_scan,
239 240 241
	.seeks = DEFAULT_SEEKS,
};

242
#ifdef CONFIG_SYSFS
A
Andrea Arcangeli 已提交
243

244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291
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
292 293 294 295 296 297 298
	ssize_t ret;

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

	if (ret > 0) {
299 300 301
		int err;

		mutex_lock(&khugepaged_mutex);
302
		err = start_stop_khugepaged();
303 304
		mutex_unlock(&khugepaged_mutex);

A
Andrea Arcangeli 已提交
305 306 307 308 309
		if (err)
			ret = err;
	}

	return ret;
310 311 312 313 314 315 316 317
}
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)
{
318 319
	return sprintf(buf, "%d\n",
		       !!test_bit(flag, &transparent_hugepage_flags));
320
}
321

322 323 324 325 326
static ssize_t single_flag_store(struct kobject *kobj,
				 struct kobj_attribute *attr,
				 const char *buf, size_t count,
				 enum transparent_hugepage_flag flag)
{
327 328 329 330 331 332 333 334 335 336
	unsigned long value;
	int ret;

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

	if (value)
337
		set_bit(flag, &transparent_hugepage_flags);
338
	else
339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366
		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);

367 368 369 370 371 372 373 374 375 376 377 378 379 380
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);
381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401
#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,
402
	&use_zero_page_attr.attr,
403 404 405 406 407 408 409 410
#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
411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426
};

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;

427
	err = kstrtoul(buf, 10, &msecs);
A
Andrea Arcangeli 已提交
428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453
	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;

454
	err = kstrtoul(buf, 10, &msecs);
A
Andrea Arcangeli 已提交
455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479
	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;

480
	err = kstrtoul(buf, 10, &pages);
A
Andrea Arcangeli 已提交
481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547
	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;

548
	err = kstrtoul(buf, 10, &max_ptes_none);
A
Andrea Arcangeli 已提交
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
	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",
574 575
};

S
Shaohua Li 已提交
576
static int __init hugepage_init_sysfs(struct kobject **hugepage_kobj)
577 578 579
{
	int err;

S
Shaohua Li 已提交
580 581
	*hugepage_kobj = kobject_create_and_add("transparent_hugepage", mm_kobj);
	if (unlikely(!*hugepage_kobj)) {
582
		pr_err("failed to create transparent hugepage kobject\n");
S
Shaohua Li 已提交
583
		return -ENOMEM;
A
Andrea Arcangeli 已提交
584 585
	}

S
Shaohua Li 已提交
586
	err = sysfs_create_group(*hugepage_kobj, &hugepage_attr_group);
A
Andrea Arcangeli 已提交
587
	if (err) {
588
		pr_err("failed to register transparent hugepage group\n");
S
Shaohua Li 已提交
589
		goto delete_obj;
A
Andrea Arcangeli 已提交
590 591
	}

S
Shaohua Li 已提交
592
	err = sysfs_create_group(*hugepage_kobj, &khugepaged_attr_group);
A
Andrea Arcangeli 已提交
593
	if (err) {
594
		pr_err("failed to register transparent hugepage group\n");
S
Shaohua Li 已提交
595
		goto remove_hp_group;
A
Andrea Arcangeli 已提交
596
	}
S
Shaohua Li 已提交
597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635

	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)
636
		goto err_sysfs;
A
Andrea Arcangeli 已提交
637 638 639

	err = khugepaged_slab_init();
	if (err)
640
		goto err_slab;
A
Andrea Arcangeli 已提交
641

642 643 644
	err = register_shrinker(&huge_zero_page_shrinker);
	if (err)
		goto err_hzp_shrinker;
645

646 647 648 649 650
	/*
	 * 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.
	 */
651
	if (totalram_pages < (512 << (20 - PAGE_SHIFT))) {
652
		transparent_hugepage_flags = 0;
653 654
		return 0;
	}
655

656
	err = start_stop_khugepaged();
657 658
	if (err)
		goto err_khugepaged;
A
Andrea Arcangeli 已提交
659

S
Shaohua Li 已提交
660
	return 0;
661 662 663 664 665
err_khugepaged:
	unregister_shrinker(&huge_zero_page_shrinker);
err_hzp_shrinker:
	khugepaged_slab_exit();
err_slab:
S
Shaohua Li 已提交
666
	hugepage_exit_sysfs(hugepage_kobj);
667
err_sysfs:
A
Andrea Arcangeli 已提交
668
	return err;
669
}
670
subsys_initcall(hugepage_init);
671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697

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)
698
		pr_warn("transparent_hugepage= cannot parse, ignored\n");
699 700 701 702
	return ret;
}
__setup("transparent_hugepage=", setup_transparent_hugepage);

703
pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma)
704 705 706 707 708 709
{
	if (likely(vma->vm_flags & VM_WRITE))
		pmd = pmd_mkwrite(pmd);
	return pmd;
}

710
static inline pmd_t mk_huge_pmd(struct page *page, pgprot_t prot)
B
Bob Liu 已提交
711 712
{
	pmd_t entry;
713
	entry = mk_pmd(page, prot);
B
Bob Liu 已提交
714 715 716 717
	entry = pmd_mkhuge(entry);
	return entry;
}

718 719 720
static int __do_huge_pmd_anonymous_page(struct mm_struct *mm,
					struct vm_area_struct *vma,
					unsigned long haddr, pmd_t *pmd,
721 722
					struct page *page, gfp_t gfp,
					unsigned int flags)
723
{
724
	struct mem_cgroup *memcg;
725
	pgtable_t pgtable;
726
	spinlock_t *ptl;
727

728
	VM_BUG_ON_PAGE(!PageCompound(page), page);
729

730 731 732 733 734
	if (mem_cgroup_try_charge(page, mm, gfp, &memcg)) {
		put_page(page);
		count_vm_event(THP_FAULT_FALLBACK);
		return VM_FAULT_FALLBACK;
	}
735

736
	pgtable = pte_alloc_one(mm, haddr);
737 738
	if (unlikely(!pgtable)) {
		mem_cgroup_cancel_charge(page, memcg);
739
		put_page(page);
740
		return VM_FAULT_OOM;
741
	}
742 743

	clear_huge_page(page, haddr, HPAGE_PMD_NR);
744 745 746 747 748
	/*
	 * The memory barrier inside __SetPageUptodate makes sure that
	 * clear_huge_page writes become visible before the set_pmd_at()
	 * write.
	 */
749 750
	__SetPageUptodate(page);

751
	ptl = pmd_lock(mm, pmd);
752
	if (unlikely(!pmd_none(*pmd))) {
753
		spin_unlock(ptl);
754
		mem_cgroup_cancel_charge(page, memcg);
755 756 757 758
		put_page(page);
		pte_free(mm, pgtable);
	} else {
		pmd_t entry;
759 760 761 762 763 764 765 766 767 768 769 770 771 772 773

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

			spin_unlock(ptl);
			mem_cgroup_cancel_charge(page, memcg);
			put_page(page);
			pte_free(mm, pgtable);
			ret = handle_userfault(vma, haddr, flags,
					       VM_UFFD_MISSING);
			VM_BUG_ON(ret & VM_FAULT_FALLBACK);
			return ret;
		}

774 775
		entry = mk_huge_pmd(page, vma->vm_page_prot);
		entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma);
776
		page_add_new_anon_rmap(page, vma, haddr);
777 778
		mem_cgroup_commit_charge(page, memcg, false);
		lru_cache_add_active_or_unevictable(page, vma);
779
		pgtable_trans_huge_deposit(mm, pmd, pgtable);
780 781
		set_pmd_at(mm, haddr, pmd, entry);
		add_mm_counter(mm, MM_ANONPAGES, HPAGE_PMD_NR);
782
		atomic_long_inc(&mm->nr_ptes);
783
		spin_unlock(ptl);
784
		count_vm_event(THP_FAULT_ALLOC);
785 786
	}

787
	return 0;
788 789
}

790
static inline gfp_t alloc_hugepage_gfpmask(int defrag, gfp_t extra_gfp)
791
{
792
	return (GFP_TRANSHUGE & ~(defrag ? 0 : __GFP_WAIT)) | extra_gfp;
793 794
}

795
/* Caller must hold page table lock. */
796
static void set_huge_zero_page(pgtable_t pgtable, struct mm_struct *mm,
797
		struct vm_area_struct *vma, unsigned long haddr, pmd_t *pmd,
798
		struct page *zero_page)
799 800
{
	pmd_t entry;
801
	entry = mk_pmd(zero_page, vma->vm_page_prot);
802
	entry = pmd_mkhuge(entry);
803
	pgtable_trans_huge_deposit(mm, pmd, pgtable);
804
	set_pmd_at(mm, haddr, pmd, entry);
805
	atomic_long_inc(&mm->nr_ptes);
806 807
}

808 809 810 811
int do_huge_pmd_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
			       unsigned long address, pmd_t *pmd,
			       unsigned int flags)
{
812
	gfp_t gfp;
813 814 815
	struct page *page;
	unsigned long haddr = address & HPAGE_PMD_MASK;

816
	if (haddr < vma->vm_start || haddr + HPAGE_PMD_SIZE > vma->vm_end)
817
		return VM_FAULT_FALLBACK;
818 819
	if (unlikely(anon_vma_prepare(vma)))
		return VM_FAULT_OOM;
820
	if (unlikely(khugepaged_enter(vma, vma->vm_flags)))
821
		return VM_FAULT_OOM;
822
	if (!(flags & FAULT_FLAG_WRITE) && !mm_forbids_zeropage(mm) &&
823
			transparent_hugepage_use_zero_page()) {
824
		spinlock_t *ptl;
825 826 827
		pgtable_t pgtable;
		struct page *zero_page;
		bool set;
828
		int ret;
829 830
		pgtable = pte_alloc_one(mm, haddr);
		if (unlikely(!pgtable))
A
Andrea Arcangeli 已提交
831
			return VM_FAULT_OOM;
832 833 834
		zero_page = get_huge_zero_page();
		if (unlikely(!zero_page)) {
			pte_free(mm, pgtable);
835
			count_vm_event(THP_FAULT_FALLBACK);
836
			return VM_FAULT_FALLBACK;
A
Andrea Arcangeli 已提交
837
		}
838
		ptl = pmd_lock(mm, pmd);
839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855
		ret = 0;
		set = false;
		if (pmd_none(*pmd)) {
			if (userfaultfd_missing(vma)) {
				spin_unlock(ptl);
				ret = handle_userfault(vma, haddr, flags,
						       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);
856 857 858
		if (!set) {
			pte_free(mm, pgtable);
			put_huge_zero_page();
859
		}
860
		return ret;
861
	}
862 863
	gfp = alloc_hugepage_gfpmask(transparent_hugepage_defrag(vma), 0);
	page = alloc_hugepage_vma(gfp, vma, haddr, HPAGE_PMD_ORDER);
864 865
	if (unlikely(!page)) {
		count_vm_event(THP_FAULT_FALLBACK);
866
		return VM_FAULT_FALLBACK;
867
	}
868
	return __do_huge_pmd_anonymous_page(mm, vma, haddr, pmd, page, gfp, flags);
869 870 871 872 873 874
}

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)
{
875
	spinlock_t *dst_ptl, *src_ptl;
876 877 878 879 880 881 882 883 884 885
	struct page *src_page;
	pmd_t pmd;
	pgtable_t pgtable;
	int ret;

	ret = -ENOMEM;
	pgtable = pte_alloc_one(dst_mm, addr);
	if (unlikely(!pgtable))
		goto out;

886 887 888
	dst_ptl = pmd_lock(dst_mm, dst_pmd);
	src_ptl = pmd_lockptr(src_mm, src_pmd);
	spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
889 890 891 892 893 894 895

	ret = -EAGAIN;
	pmd = *src_pmd;
	if (unlikely(!pmd_trans_huge(pmd))) {
		pte_free(dst_mm, pgtable);
		goto out_unlock;
	}
896
	/*
897
	 * When page table lock is held, the huge zero pmd should not be
898 899 900 901
	 * under splitting since we don't split the page itself, only pmd to
	 * a page table.
	 */
	if (is_huge_zero_pmd(pmd)) {
902
		struct page *zero_page;
903 904 905 906 907
		/*
		 * 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.
		 */
908
		zero_page = get_huge_zero_page();
909
		set_huge_zero_page(pgtable, dst_mm, vma, addr, dst_pmd,
910
				zero_page);
911 912 913
		ret = 0;
		goto out_unlock;
	}
914

915 916
	if (unlikely(pmd_trans_splitting(pmd))) {
		/* split huge page running from under us */
917 918
		spin_unlock(src_ptl);
		spin_unlock(dst_ptl);
919 920 921 922 923 924
		pte_free(dst_mm, pgtable);

		wait_split_huge_page(vma->anon_vma, src_pmd); /* src_vma */
		goto out;
	}
	src_page = pmd_page(pmd);
925
	VM_BUG_ON_PAGE(!PageHead(src_page), src_page);
926 927 928 929 930 931
	get_page(src_page);
	page_dup_rmap(src_page);
	add_mm_counter(dst_mm, MM_ANONPAGES, HPAGE_PMD_NR);

	pmdp_set_wrprotect(src_mm, addr, src_pmd);
	pmd = pmd_mkold(pmd_wrprotect(pmd));
932
	pgtable_trans_huge_deposit(dst_mm, dst_pmd, pgtable);
933
	set_pmd_at(dst_mm, addr, dst_pmd, pmd);
934
	atomic_long_inc(&dst_mm->nr_ptes);
935 936 937

	ret = 0;
out_unlock:
938 939
	spin_unlock(src_ptl);
	spin_unlock(dst_ptl);
940 941 942 943
out:
	return ret;
}

944 945 946 947 948 949
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)
{
950
	spinlock_t *ptl;
951 952 953
	pmd_t entry;
	unsigned long haddr;

954
	ptl = pmd_lock(mm, pmd);
955 956 957 958 959 960 961 962 963
	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:
964
	spin_unlock(ptl);
965 966
}

967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997
/*
 * Save CONFIG_DEBUG_PAGEALLOC from faulting falsely on tail pages
 * during copy_user_huge_page()'s copy_page_rep(): in the case when
 * the source page gets split and a tail freed before copy completes.
 * Called under pmd_lock of checked pmd, so safe from splitting itself.
 */
static void get_user_huge_page(struct page *page)
{
	if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC)) {
		struct page *endpage = page + HPAGE_PMD_NR;

		atomic_add(HPAGE_PMD_NR, &page->_count);
		while (++page < endpage)
			get_huge_page_tail(page);
	} else {
		get_page(page);
	}
}

static void put_user_huge_page(struct page *page)
{
	if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC)) {
		struct page *endpage = page + HPAGE_PMD_NR;

		while (page < endpage)
			put_page(page++);
	} else {
		put_page(page);
	}
}

998 999 1000 1001 1002 1003 1004
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)
{
1005
	struct mem_cgroup *memcg;
1006
	spinlock_t *ptl;
1007 1008 1009 1010
	pgtable_t pgtable;
	pmd_t _pmd;
	int ret = 0, i;
	struct page **pages;
1011 1012
	unsigned long mmun_start;	/* For mmu_notifiers */
	unsigned long mmun_end;		/* For mmu_notifiers */
1013 1014 1015 1016 1017 1018 1019 1020 1021

	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++) {
1022 1023
		pages[i] = alloc_page_vma_node(GFP_HIGHUSER_MOVABLE |
					       __GFP_OTHER_NODE,
1024
					       vma, address, page_to_nid(page));
A
Andrea Arcangeli 已提交
1025
		if (unlikely(!pages[i] ||
1026 1027
			     mem_cgroup_try_charge(pages[i], mm, GFP_KERNEL,
						   &memcg))) {
A
Andrea Arcangeli 已提交
1028
			if (pages[i])
1029
				put_page(pages[i]);
A
Andrea Arcangeli 已提交
1030
			while (--i >= 0) {
1031 1032 1033
				memcg = (void *)page_private(pages[i]);
				set_page_private(pages[i], 0);
				mem_cgroup_cancel_charge(pages[i], memcg);
A
Andrea Arcangeli 已提交
1034 1035
				put_page(pages[i]);
			}
1036 1037 1038 1039
			kfree(pages);
			ret |= VM_FAULT_OOM;
			goto out;
		}
1040
		set_page_private(pages[i], (unsigned long)memcg);
1041 1042 1043 1044
	}

	for (i = 0; i < HPAGE_PMD_NR; i++) {
		copy_user_highpage(pages[i], page + i,
1045
				   haddr + PAGE_SIZE * i, vma);
1046 1047 1048 1049
		__SetPageUptodate(pages[i]);
		cond_resched();
	}

1050 1051 1052 1053
	mmun_start = haddr;
	mmun_end   = haddr + HPAGE_PMD_SIZE;
	mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);

1054
	ptl = pmd_lock(mm, pmd);
1055 1056
	if (unlikely(!pmd_same(*pmd, orig_pmd)))
		goto out_free_pages;
1057
	VM_BUG_ON_PAGE(!PageHead(page), page);
1058

1059
	pmdp_huge_clear_flush_notify(vma, haddr, pmd);
1060 1061
	/* leave pmd empty until pte is filled */

1062
	pgtable = pgtable_trans_huge_withdraw(mm, pmd);
1063 1064 1065 1066 1067 1068
	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);
1069 1070
		memcg = (void *)page_private(pages[i]);
		set_page_private(pages[i], 0);
1071
		page_add_new_anon_rmap(pages[i], vma, haddr);
1072 1073
		mem_cgroup_commit_charge(pages[i], memcg, false);
		lru_cache_add_active_or_unevictable(pages[i], vma);
1074 1075 1076 1077 1078 1079 1080 1081 1082 1083
		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);
	page_remove_rmap(page);
1084
	spin_unlock(ptl);
1085

1086 1087
	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);

1088 1089 1090 1091 1092 1093 1094
	ret |= VM_FAULT_WRITE;
	put_page(page);

out:
	return ret;

out_free_pages:
1095
	spin_unlock(ptl);
1096
	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
A
Andrea Arcangeli 已提交
1097
	for (i = 0; i < HPAGE_PMD_NR; i++) {
1098 1099 1100
		memcg = (void *)page_private(pages[i]);
		set_page_private(pages[i], 0);
		mem_cgroup_cancel_charge(pages[i], memcg);
1101
		put_page(pages[i]);
A
Andrea Arcangeli 已提交
1102
	}
1103 1104 1105 1106 1107 1108 1109
	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)
{
1110
	spinlock_t *ptl;
1111
	int ret = 0;
1112
	struct page *page = NULL, *new_page;
1113
	struct mem_cgroup *memcg;
1114
	unsigned long haddr;
1115 1116
	unsigned long mmun_start;	/* For mmu_notifiers */
	unsigned long mmun_end;		/* For mmu_notifiers */
1117
	gfp_t huge_gfp;			/* for allocation and charge */
1118

1119
	ptl = pmd_lockptr(mm, pmd);
1120
	VM_BUG_ON_VMA(!vma->anon_vma, vma);
1121 1122 1123
	haddr = address & HPAGE_PMD_MASK;
	if (is_huge_zero_pmd(orig_pmd))
		goto alloc;
1124
	spin_lock(ptl);
1125 1126 1127 1128
	if (unlikely(!pmd_same(*pmd, orig_pmd)))
		goto out_unlock;

	page = pmd_page(orig_pmd);
1129
	VM_BUG_ON_PAGE(!PageCompound(page) || !PageHead(page), page);
1130 1131 1132 1133 1134
	if (page_mapcount(page) == 1) {
		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))
1135
			update_mmu_cache_pmd(vma, address, pmd);
1136 1137 1138
		ret |= VM_FAULT_WRITE;
		goto out_unlock;
	}
1139
	get_user_huge_page(page);
1140
	spin_unlock(ptl);
1141
alloc:
1142
	if (transparent_hugepage_enabled(vma) &&
1143
	    !transparent_hugepage_debug_cow()) {
1144 1145
		huge_gfp = alloc_hugepage_gfpmask(transparent_hugepage_defrag(vma), 0);
		new_page = alloc_hugepage_vma(huge_gfp, vma, haddr, HPAGE_PMD_ORDER);
1146
	} else
1147 1148 1149
		new_page = NULL;

	if (unlikely(!new_page)) {
1150
		if (!page) {
1151 1152
			split_huge_page_pmd(vma, address, pmd);
			ret |= VM_FAULT_FALLBACK;
1153 1154 1155
		} else {
			ret = do_huge_pmd_wp_page_fallback(mm, vma, address,
					pmd, orig_pmd, page, haddr);
1156
			if (ret & VM_FAULT_OOM) {
1157
				split_huge_page(page);
1158 1159
				ret |= VM_FAULT_FALLBACK;
			}
1160
			put_user_huge_page(page);
1161
		}
1162
		count_vm_event(THP_FAULT_FALLBACK);
1163 1164 1165
		goto out;
	}

1166
	if (unlikely(mem_cgroup_try_charge(new_page, mm, huge_gfp, &memcg))) {
A
Andrea Arcangeli 已提交
1167
		put_page(new_page);
1168 1169
		if (page) {
			split_huge_page(page);
1170
			put_user_huge_page(page);
1171 1172 1173
		} else
			split_huge_page_pmd(vma, address, pmd);
		ret |= VM_FAULT_FALLBACK;
1174
		count_vm_event(THP_FAULT_FALLBACK);
A
Andrea Arcangeli 已提交
1175 1176 1177
		goto out;
	}

1178 1179
	count_vm_event(THP_FAULT_ALLOC);

1180
	if (!page)
1181 1182 1183
		clear_huge_page(new_page, haddr, HPAGE_PMD_NR);
	else
		copy_user_huge_page(new_page, page, haddr, vma, HPAGE_PMD_NR);
1184 1185
	__SetPageUptodate(new_page);

1186 1187 1188 1189
	mmun_start = haddr;
	mmun_end   = haddr + HPAGE_PMD_SIZE;
	mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);

1190
	spin_lock(ptl);
1191
	if (page)
1192
		put_user_huge_page(page);
A
Andrea Arcangeli 已提交
1193
	if (unlikely(!pmd_same(*pmd, orig_pmd))) {
1194
		spin_unlock(ptl);
1195
		mem_cgroup_cancel_charge(new_page, memcg);
1196
		put_page(new_page);
1197
		goto out_mn;
A
Andrea Arcangeli 已提交
1198
	} else {
1199
		pmd_t entry;
1200 1201
		entry = mk_huge_pmd(new_page, vma->vm_page_prot);
		entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma);
1202
		pmdp_huge_clear_flush_notify(vma, haddr, pmd);
1203
		page_add_new_anon_rmap(new_page, vma, haddr);
1204 1205
		mem_cgroup_commit_charge(new_page, memcg, false);
		lru_cache_add_active_or_unevictable(new_page, vma);
1206
		set_pmd_at(mm, haddr, pmd, entry);
1207
		update_mmu_cache_pmd(vma, address, pmd);
1208
		if (!page) {
1209
			add_mm_counter(mm, MM_ANONPAGES, HPAGE_PMD_NR);
1210 1211
			put_huge_zero_page();
		} else {
1212
			VM_BUG_ON_PAGE(!PageHead(page), page);
1213 1214 1215
			page_remove_rmap(page);
			put_page(page);
		}
1216 1217
		ret |= VM_FAULT_WRITE;
	}
1218
	spin_unlock(ptl);
1219 1220
out_mn:
	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
1221 1222
out:
	return ret;
1223
out_unlock:
1224
	spin_unlock(ptl);
1225
	return ret;
1226 1227
}

1228
struct page *follow_trans_huge_pmd(struct vm_area_struct *vma,
1229 1230 1231 1232
				   unsigned long addr,
				   pmd_t *pmd,
				   unsigned int flags)
{
1233
	struct mm_struct *mm = vma->vm_mm;
1234 1235
	struct page *page = NULL;

1236
	assert_spin_locked(pmd_lockptr(mm, pmd));
1237 1238 1239 1240

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

1241 1242 1243 1244
	/* Avoid dumping huge zero page */
	if ((flags & FOLL_DUMP) && is_huge_zero_pmd(*pmd))
		return ERR_PTR(-EFAULT);

1245
	/* Full NUMA hinting faults to serialise migration in fault paths */
1246
	if ((flags & FOLL_NUMA) && pmd_protnone(*pmd))
1247 1248
		goto out;

1249
	page = pmd_page(*pmd);
1250
	VM_BUG_ON_PAGE(!PageHead(page), page);
1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261
	if (flags & FOLL_TOUCH) {
		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));
1262 1263 1264
		if (pmdp_set_access_flags(vma, addr & HPAGE_PMD_MASK,
					  pmd, _pmd,  1))
			update_mmu_cache_pmd(vma, addr, pmd);
1265
	}
1266
	if ((flags & FOLL_POPULATE) && (vma->vm_flags & VM_LOCKED)) {
1267 1268 1269 1270 1271 1272 1273
		if (page->mapping && trylock_page(page)) {
			lru_add_drain();
			if (page->mapping)
				mlock_vma_page(page);
			unlock_page(page);
		}
	}
1274
	page += (addr & ~HPAGE_PMD_MASK) >> PAGE_SHIFT;
1275
	VM_BUG_ON_PAGE(!PageCompound(page), page);
1276
	if (flags & FOLL_GET)
1277
		get_page_foll(page);
1278 1279 1280 1281 1282

out:
	return page;
}

1283
/* NUMA hinting page fault entry point for trans huge pmds */
1284 1285
int do_huge_pmd_numa_page(struct mm_struct *mm, struct vm_area_struct *vma,
				unsigned long addr, pmd_t pmd, pmd_t *pmdp)
1286
{
1287
	spinlock_t *ptl;
1288
	struct anon_vma *anon_vma = NULL;
1289
	struct page *page;
1290
	unsigned long haddr = addr & HPAGE_PMD_MASK;
1291
	int page_nid = -1, this_nid = numa_node_id();
1292
	int target_nid, last_cpupid = -1;
1293 1294
	bool page_locked;
	bool migrated = false;
1295
	bool was_writable;
1296
	int flags = 0;
1297

1298 1299 1300
	/* A PROT_NONE fault should not end up here */
	BUG_ON(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)));

1301
	ptl = pmd_lock(mm, pmdp);
1302 1303 1304
	if (unlikely(!pmd_same(pmd, *pmdp)))
		goto out_unlock;

1305 1306 1307 1308 1309 1310
	/*
	 * 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))) {
1311
		page = pmd_page(*pmdp);
1312
		spin_unlock(ptl);
1313
		wait_on_page_locked(page);
1314 1315 1316
		goto out;
	}

1317
	page = pmd_page(pmd);
1318
	BUG_ON(is_huge_zero_page(page));
1319
	page_nid = page_to_nid(page);
1320
	last_cpupid = page_cpupid_last(page);
1321
	count_vm_numa_event(NUMA_HINT_FAULTS);
1322
	if (page_nid == this_nid) {
1323
		count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL);
1324 1325
		flags |= TNF_FAULT_LOCAL;
	}
1326

1327 1328
	/* See similar comment in do_numa_page for explanation */
	if (!(vma->vm_flags & VM_WRITE))
1329 1330
		flags |= TNF_NO_GROUP;

1331 1332 1333 1334
	/*
	 * Acquire the page lock to serialise THP migrations but avoid dropping
	 * page_table_lock if at all possible
	 */
1335 1336 1337 1338
	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 */
1339
		if (page_locked)
1340
			goto clear_pmdnuma;
1341
	}
1342

1343
	/* Migration could have started since the pmd_trans_migrating check */
1344
	if (!page_locked) {
1345
		spin_unlock(ptl);
1346
		wait_on_page_locked(page);
1347
		page_nid = -1;
1348 1349 1350
		goto out;
	}

1351 1352 1353 1354
	/*
	 * Page is misplaced. Page lock serialises migrations. Acquire anon_vma
	 * to serialises splits
	 */
1355
	get_page(page);
1356
	spin_unlock(ptl);
1357
	anon_vma = page_lock_anon_vma_read(page);
1358

P
Peter Zijlstra 已提交
1359
	/* Confirm the PMD did not change while page_table_lock was released */
1360
	spin_lock(ptl);
1361 1362 1363
	if (unlikely(!pmd_same(pmd, *pmdp))) {
		unlock_page(page);
		put_page(page);
1364
		page_nid = -1;
1365
		goto out_unlock;
1366
	}
1367

1368 1369 1370 1371 1372 1373 1374
	/* 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;
	}

1375 1376
	/*
	 * Migrate the THP to the requested node, returns with page unlocked
1377
	 * and access rights restored.
1378
	 */
1379
	spin_unlock(ptl);
1380
	migrated = migrate_misplaced_transhuge_page(mm, vma,
1381
				pmdp, pmd, addr, page, target_nid);
1382 1383
	if (migrated) {
		flags |= TNF_MIGRATED;
1384
		page_nid = target_nid;
1385 1386
	} else
		flags |= TNF_MIGRATE_FAIL;
1387

1388
	goto out;
1389
clear_pmdnuma:
1390
	BUG_ON(!PageLocked(page));
1391
	was_writable = pmd_write(pmd);
1392
	pmd = pmd_modify(pmd, vma->vm_page_prot);
1393
	pmd = pmd_mkyoung(pmd);
1394 1395
	if (was_writable)
		pmd = pmd_mkwrite(pmd);
1396 1397
	set_pmd_at(mm, haddr, pmdp, pmd);
	update_mmu_cache_pmd(vma, addr, pmdp);
1398
	unlock_page(page);
1399
out_unlock:
1400
	spin_unlock(ptl);
1401 1402 1403 1404 1405

out:
	if (anon_vma)
		page_unlock_anon_vma_read(anon_vma);

1406
	if (page_nid != -1)
1407
		task_numa_fault(last_cpupid, page_nid, HPAGE_PMD_NR, flags);
1408

1409 1410 1411
	return 0;
}

1412
int zap_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
S
Shaohua Li 已提交
1413
		 pmd_t *pmd, unsigned long addr)
1414
{
1415
	spinlock_t *ptl;
1416 1417
	int ret = 0;

1418
	if (__pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
1419 1420
		struct page *page;
		pgtable_t pgtable;
1421
		pmd_t orig_pmd;
1422 1423
		/*
		 * For architectures like ppc64 we look at deposited pgtable
1424
		 * when calling pmdp_huge_get_and_clear. So do the
1425 1426 1427
		 * pgtable_trans_huge_withdraw after finishing pmdp related
		 * operations.
		 */
1428 1429
		orig_pmd = pmdp_huge_get_and_clear_full(tlb->mm, addr, pmd,
							tlb->fullmm);
1430
		tlb_remove_pmd_tlb_entry(tlb, pmd, addr);
1431
		pgtable = pgtable_trans_huge_withdraw(tlb->mm, pmd);
1432
		if (is_huge_zero_pmd(orig_pmd)) {
1433
			atomic_long_dec(&tlb->mm->nr_ptes);
1434
			spin_unlock(ptl);
1435
			put_huge_zero_page();
1436 1437 1438
		} else {
			page = pmd_page(orig_pmd);
			page_remove_rmap(page);
1439
			VM_BUG_ON_PAGE(page_mapcount(page) < 0, page);
1440
			add_mm_counter(tlb->mm, MM_ANONPAGES, -HPAGE_PMD_NR);
1441
			VM_BUG_ON_PAGE(!PageHead(page), page);
1442
			atomic_long_dec(&tlb->mm->nr_ptes);
1443
			spin_unlock(ptl);
1444 1445
			tlb_remove_page(tlb, page);
		}
1446 1447 1448
		pte_free(tlb->mm, pgtable);
		ret = 1;
	}
1449 1450 1451
	return ret;
}

1452 1453 1454 1455 1456
int move_huge_pmd(struct vm_area_struct *vma, struct vm_area_struct *new_vma,
		  unsigned long old_addr,
		  unsigned long new_addr, unsigned long old_end,
		  pmd_t *old_pmd, pmd_t *new_pmd)
{
1457
	spinlock_t *old_ptl, *new_ptl;
1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477
	int ret = 0;
	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))
		goto out;

	/*
	 * 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));
		goto out;
	}

1478 1479 1480 1481 1482
	/*
	 * We don't have to worry about the ordering of src and dst
	 * ptlocks because exclusive mmap_sem prevents deadlock.
	 */
	ret = __pmd_trans_huge_lock(old_pmd, vma, &old_ptl);
1483
	if (ret == 1) {
1484 1485 1486
		new_ptl = pmd_lockptr(mm, new_pmd);
		if (new_ptl != old_ptl)
			spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
1487
		pmd = pmdp_huge_get_and_clear(mm, old_addr, old_pmd);
1488
		VM_BUG_ON(!pmd_none(*new_pmd));
1489

1490 1491
		if (pmd_move_must_withdraw(new_ptl, old_ptl)) {
			pgtable_t pgtable;
1492 1493 1494
			pgtable = pgtable_trans_huge_withdraw(mm, old_pmd);
			pgtable_trans_huge_deposit(mm, new_pmd, pgtable);
		}
1495 1496 1497
		set_pmd_at(mm, new_addr, new_pmd, pmd_mksoft_dirty(pmd));
		if (new_ptl != old_ptl)
			spin_unlock(new_ptl);
1498
		spin_unlock(old_ptl);
1499 1500 1501 1502 1503
	}
out:
	return ret;
}

1504 1505 1506 1507 1508 1509
/*
 * 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
 */
1510
int change_huge_pmd(struct vm_area_struct *vma, pmd_t *pmd,
1511
		unsigned long addr, pgprot_t newprot, int prot_numa)
1512 1513
{
	struct mm_struct *mm = vma->vm_mm;
1514
	spinlock_t *ptl;
1515 1516
	int ret = 0;

1517
	if (__pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
1518
		pmd_t entry;
1519
		bool preserve_write = prot_numa && pmd_write(*pmd);
1520
		ret = 1;
1521 1522 1523 1524 1525 1526 1527 1528

		/*
		 * 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);
1529
			return ret;
1530 1531
		}

1532
		if (!prot_numa || !pmd_protnone(*pmd)) {
1533
			entry = pmdp_huge_get_and_clear_notify(mm, addr, pmd);
1534
			entry = pmd_modify(entry, newprot);
1535 1536
			if (preserve_write)
				entry = pmd_mkwrite(entry);
1537 1538
			ret = HPAGE_PMD_NR;
			set_pmd_at(mm, addr, pmd, entry);
1539
			BUG_ON(!preserve_write && pmd_write(entry));
1540
		}
1541
		spin_unlock(ptl);
1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553
	}

	return ret;
}

/*
 * Returns 1 if a given pmd maps a stable (not under splitting) thp.
 * Returns -1 if it maps a thp under splitting. Returns 0 otherwise.
 *
 * Note that if it returns 1, this routine returns without unlocking page
 * table locks. So callers must unlock them.
 */
1554 1555
int __pmd_trans_huge_lock(pmd_t *pmd, struct vm_area_struct *vma,
		spinlock_t **ptl)
1556
{
1557
	*ptl = pmd_lock(vma->vm_mm, pmd);
1558 1559
	if (likely(pmd_trans_huge(*pmd))) {
		if (unlikely(pmd_trans_splitting(*pmd))) {
1560
			spin_unlock(*ptl);
1561
			wait_split_huge_page(vma->anon_vma, pmd);
1562
			return -1;
1563
		} else {
1564 1565 1566
			/* Thp mapped by 'pmd' is stable, so we can
			 * handle it as it is. */
			return 1;
1567
		}
1568
	}
1569
	spin_unlock(*ptl);
1570
	return 0;
1571 1572
}

1573 1574 1575 1576 1577 1578 1579 1580
/*
 * This function returns whether a given @page is mapped onto the @address
 * in the virtual space of @mm.
 *
 * When it's true, this function returns *pmd with holding the page table lock
 * and passing it back to the caller via @ptl.
 * If it's false, returns NULL without holding the page table lock.
 */
1581 1582 1583
pmd_t *page_check_address_pmd(struct page *page,
			      struct mm_struct *mm,
			      unsigned long address,
1584 1585
			      enum page_check_address_pmd_flag flag,
			      spinlock_t **ptl)
1586
{
1587 1588
	pgd_t *pgd;
	pud_t *pud;
1589
	pmd_t *pmd;
1590 1591

	if (address & ~HPAGE_PMD_MASK)
1592
		return NULL;
1593

1594 1595
	pgd = pgd_offset(mm, address);
	if (!pgd_present(*pgd))
1596
		return NULL;
1597 1598 1599 1600 1601
	pud = pud_offset(pgd, address);
	if (!pud_present(*pud))
		return NULL;
	pmd = pmd_offset(pud, address);

1602
	*ptl = pmd_lock(mm, pmd);
1603
	if (!pmd_present(*pmd))
1604
		goto unlock;
1605
	if (pmd_page(*pmd) != page)
1606
		goto unlock;
1607 1608 1609 1610 1611 1612 1613 1614 1615
	/*
	 * split_vma() may create temporary aliased mappings. There is
	 * no risk as long as all huge pmd are found and have their
	 * splitting bit set before __split_huge_page_refcount
	 * runs. Finding the same huge pmd more than once during the
	 * same rmap walk is not a problem.
	 */
	if (flag == PAGE_CHECK_ADDRESS_PMD_NOTSPLITTING_FLAG &&
	    pmd_trans_splitting(*pmd))
1616
		goto unlock;
1617 1618 1619
	if (pmd_trans_huge(*pmd)) {
		VM_BUG_ON(flag == PAGE_CHECK_ADDRESS_PMD_SPLITTING_FLAG &&
			  !pmd_trans_splitting(*pmd));
1620
		return pmd;
1621
	}
1622 1623 1624
unlock:
	spin_unlock(*ptl);
	return NULL;
1625 1626 1627 1628 1629 1630 1631
}

static int __split_huge_page_splitting(struct page *page,
				       struct vm_area_struct *vma,
				       unsigned long address)
{
	struct mm_struct *mm = vma->vm_mm;
1632
	spinlock_t *ptl;
1633 1634
	pmd_t *pmd;
	int ret = 0;
1635 1636 1637
	/* For mmu_notifiers */
	const unsigned long mmun_start = address;
	const unsigned long mmun_end   = address + HPAGE_PMD_SIZE;
1638

1639
	mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
1640
	pmd = page_check_address_pmd(page, mm, address,
1641
			PAGE_CHECK_ADDRESS_PMD_NOTSPLITTING_FLAG, &ptl);
1642 1643 1644 1645 1646
	if (pmd) {
		/*
		 * We can't temporarily set the pmd to null in order
		 * to split it, the pmd must remain marked huge at all
		 * times or the VM won't take the pmd_trans_huge paths
1647
		 * and it won't wait on the anon_vma->root->rwsem to
1648 1649
		 * serialize against split_huge_page*.
		 */
1650
		pmdp_splitting_flush(vma, address, pmd);
1651

1652
		ret = 1;
1653
		spin_unlock(ptl);
1654
	}
1655
	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
1656 1657 1658 1659

	return ret;
}

1660 1661
static void __split_huge_page_refcount(struct page *page,
				       struct list_head *list)
1662 1663 1664
{
	int i;
	struct zone *zone = page_zone(page);
1665
	struct lruvec *lruvec;
1666
	int tail_count = 0;
1667 1668 1669

	/* prevent PageLRU to go away from under us, and freeze lru stats */
	spin_lock_irq(&zone->lru_lock);
1670 1671
	lruvec = mem_cgroup_page_lruvec(page, zone);

1672
	compound_lock(page);
1673 1674
	/* complete memcg works before add pages to LRU */
	mem_cgroup_split_huge_fixup(page);
1675

1676
	for (i = HPAGE_PMD_NR - 1; i >= 1; i--) {
1677 1678
		struct page *page_tail = page + i;

1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699
		/* tail_page->_mapcount cannot change */
		BUG_ON(page_mapcount(page_tail) < 0);
		tail_count += page_mapcount(page_tail);
		/* check for overflow */
		BUG_ON(tail_count < 0);
		BUG_ON(atomic_read(&page_tail->_count) != 0);
		/*
		 * 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(page_mapcount(page) + page_mapcount(page_tail) + 1,
			   &page_tail->_count);
1700 1701

		/* after clearing PageTail the gup refcount can be released */
1702
		smp_mb__after_atomic();
1703

1704
		page_tail->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1705 1706 1707 1708
		page_tail->flags |= (page->flags &
				     ((1L << PG_referenced) |
				      (1L << PG_swapbacked) |
				      (1L << PG_mlocked) |
1709 1710 1711
				      (1L << PG_uptodate) |
				      (1L << PG_active) |
				      (1L << PG_unevictable)));
1712 1713
		page_tail->flags |= (1L << PG_dirty);

1714
		/* clear PageTail before overwriting first_page */
1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735
		smp_wmb();

		/*
		 * __split_huge_page_splitting() already set the
		 * splitting bit in all pmd that could map this
		 * hugepage, that will ensure no CPU can alter the
		 * mapcount on the head page. The mapcount is only
		 * accounted in the head page and it has to be
		 * transferred to all tail pages in the below code. So
		 * for this code to be safe, the split the mapcount
		 * can't change. But that doesn't mean userland can't
		 * keep changing and reading the page contents while
		 * we transfer the mapcount, so the pmd splitting
		 * status is achieved setting a reserved bit in the
		 * pmd, not by clearing the present bit.
		*/
		page_tail->_mapcount = page->_mapcount;

		BUG_ON(page_tail->mapping);
		page_tail->mapping = page->mapping;

1736
		page_tail->index = page->index + i;
1737
		page_cpupid_xchg_last(page_tail, page_cpupid_last(page));
1738 1739 1740 1741 1742 1743

		BUG_ON(!PageAnon(page_tail));
		BUG_ON(!PageUptodate(page_tail));
		BUG_ON(!PageDirty(page_tail));
		BUG_ON(!PageSwapBacked(page_tail));

1744
		lru_add_page_tail(page, page_tail, lruvec, list);
1745
	}
1746 1747
	atomic_sub(tail_count, &page->_count);
	BUG_ON(atomic_read(&page->_count) <= 0);
1748

1749
	__mod_zone_page_state(zone, NR_ANON_TRANSPARENT_HUGEPAGES, -1);
1750

1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779
	ClearPageCompound(page);
	compound_unlock(page);
	spin_unlock_irq(&zone->lru_lock);

	for (i = 1; i < HPAGE_PMD_NR; i++) {
		struct page *page_tail = page + i;
		BUG_ON(page_count(page_tail) <= 0);
		/*
		 * Tail pages 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(page_tail);
	}

	/*
	 * Only the head page (now become a regular page) is required
	 * to be pinned by the caller.
	 */
	BUG_ON(page_count(page) <= 0);
}

static int __split_huge_page_map(struct page *page,
				 struct vm_area_struct *vma,
				 unsigned long address)
{
	struct mm_struct *mm = vma->vm_mm;
1780
	spinlock_t *ptl;
1781 1782 1783 1784 1785 1786
	pmd_t *pmd, _pmd;
	int ret = 0, i;
	pgtable_t pgtable;
	unsigned long haddr;

	pmd = page_check_address_pmd(page, mm, address,
1787
			PAGE_CHECK_ADDRESS_PMD_SPLITTING_FLAG, &ptl);
1788
	if (pmd) {
1789
		pgtable = pgtable_trans_huge_withdraw(mm, pmd);
1790
		pmd_populate(mm, &_pmd, pgtable);
1791 1792
		if (pmd_write(*pmd))
			BUG_ON(page_mapcount(page) != 1);
1793

1794 1795
		haddr = address;
		for (i = 0; i < HPAGE_PMD_NR; i++, haddr += PAGE_SIZE) {
1796 1797
			pte_t *pte, entry;
			BUG_ON(PageCompound(page+i));
1798
			/*
1799 1800 1801
			 * Note that NUMA hinting access restrictions are not
			 * transferred to avoid any possibility of altering
			 * permissions across VMAs.
1802
			 */
1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841
			entry = mk_pte(page + i, vma->vm_page_prot);
			entry = maybe_mkwrite(pte_mkdirty(entry), vma);
			if (!pmd_write(*pmd))
				entry = pte_wrprotect(entry);
			if (!pmd_young(*pmd))
				entry = pte_mkold(entry);
			pte = pte_offset_map(&_pmd, haddr);
			BUG_ON(!pte_none(*pte));
			set_pte_at(mm, haddr, pte, entry);
			pte_unmap(pte);
		}

		smp_wmb(); /* make pte visible before pmd */
		/*
		 * 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_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.
		 */
G
Gerald Schaefer 已提交
1842
		pmdp_invalidate(vma, address, pmd);
1843 1844
		pmd_populate(mm, pmd, pgtable);
		ret = 1;
1845
		spin_unlock(ptl);
1846 1847 1848 1849 1850
	}

	return ret;
}

1851
/* must be called with anon_vma->root->rwsem held */
1852
static void __split_huge_page(struct page *page,
1853 1854
			      struct anon_vma *anon_vma,
			      struct list_head *list)
1855 1856
{
	int mapcount, mapcount2;
1857
	pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1858 1859 1860 1861 1862 1863
	struct anon_vma_chain *avc;

	BUG_ON(!PageHead(page));
	BUG_ON(PageTail(page));

	mapcount = 0;
1864
	anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
1865 1866 1867 1868 1869
		struct vm_area_struct *vma = avc->vma;
		unsigned long addr = vma_address(page, vma);
		BUG_ON(is_vma_temporary_stack(vma));
		mapcount += __split_huge_page_splitting(page, vma, addr);
	}
1870 1871 1872 1873 1874 1875 1876 1877 1878 1879
	/*
	 * It is critical that new vmas are added to the tail of the
	 * anon_vma list. This guarantes that if copy_huge_pmd() runs
	 * and establishes a child pmd before
	 * __split_huge_page_splitting() freezes the parent pmd (so if
	 * we fail to prevent copy_huge_pmd() from running until the
	 * whole __split_huge_page() is complete), we will still see
	 * the newly established pmd of the child later during the
	 * walk, to be able to set it as pmd_trans_splitting too.
	 */
1880
	if (mapcount != page_mapcount(page)) {
1881 1882
		pr_err("mapcount %d page_mapcount %d\n",
			mapcount, page_mapcount(page));
1883 1884
		BUG();
	}
1885

1886
	__split_huge_page_refcount(page, list);
1887 1888

	mapcount2 = 0;
1889
	anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
1890 1891 1892 1893 1894
		struct vm_area_struct *vma = avc->vma;
		unsigned long addr = vma_address(page, vma);
		BUG_ON(is_vma_temporary_stack(vma));
		mapcount2 += __split_huge_page_map(page, vma, addr);
	}
1895
	if (mapcount != mapcount2) {
1896 1897
		pr_err("mapcount %d mapcount2 %d page_mapcount %d\n",
			mapcount, mapcount2, page_mapcount(page));
1898 1899
		BUG();
	}
1900 1901
}

1902 1903 1904 1905 1906 1907 1908 1909
/*
 * Split a hugepage into normal pages. This doesn't change the position of head
 * page. 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.
 * Return 0 if the hugepage is split successfully otherwise return 1.
 */
int split_huge_page_to_list(struct page *page, struct list_head *list)
1910 1911 1912 1913
{
	struct anon_vma *anon_vma;
	int ret = 1;

1914
	BUG_ON(is_huge_zero_page(page));
1915
	BUG_ON(!PageAnon(page));
1916 1917 1918 1919 1920 1921 1922 1923 1924

	/*
	 * 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(page);
1925 1926
	if (!anon_vma)
		goto out;
1927 1928
	anon_vma_lock_write(anon_vma);

1929 1930 1931 1932 1933
	ret = 0;
	if (!PageCompound(page))
		goto out_unlock;

	BUG_ON(!PageSwapBacked(page));
1934
	__split_huge_page(page, anon_vma, list);
1935
	count_vm_event(THP_SPLIT);
1936 1937 1938

	BUG_ON(PageCompound(page));
out_unlock:
1939
	anon_vma_unlock_write(anon_vma);
1940
	put_anon_vma(anon_vma);
1941 1942 1943 1944
out:
	return ret;
}

1945
#define VM_NO_THP (VM_SPECIAL | VM_HUGETLB | VM_SHARED | VM_MAYSHARE)
1946

1947 1948
int hugepage_madvise(struct vm_area_struct *vma,
		     unsigned long *vm_flags, int advice)
A
Andrea Arcangeli 已提交
1949
{
A
Andrea Arcangeli 已提交
1950 1951
	switch (advice) {
	case MADV_HUGEPAGE:
1952 1953 1954 1955 1956 1957 1958 1959 1960
#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
1961 1962 1963
		/*
		 * Be somewhat over-protective like KSM for now!
		 */
1964
		if (*vm_flags & (VM_HUGEPAGE | VM_NO_THP))
A
Andrea Arcangeli 已提交
1965 1966 1967
			return -EINVAL;
		*vm_flags &= ~VM_NOHUGEPAGE;
		*vm_flags |= VM_HUGEPAGE;
1968 1969 1970 1971 1972
		/*
		 * If the vma become good for khugepaged to scan,
		 * register it here without waiting a page fault that
		 * may not happen any time soon.
		 */
1973
		if (unlikely(khugepaged_enter_vma_merge(vma, *vm_flags)))
1974
			return -ENOMEM;
A
Andrea Arcangeli 已提交
1975 1976 1977 1978 1979
		break;
	case MADV_NOHUGEPAGE:
		/*
		 * Be somewhat over-protective like KSM for now!
		 */
1980
		if (*vm_flags & (VM_NOHUGEPAGE | VM_NO_THP))
A
Andrea Arcangeli 已提交
1981 1982 1983
			return -EINVAL;
		*vm_flags &= ~VM_HUGEPAGE;
		*vm_flags |= VM_NOHUGEPAGE;
1984 1985 1986 1987 1988
		/*
		 * 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
1989 1990
		break;
	}
A
Andrea Arcangeli 已提交
1991 1992 1993 1994

	return 0;
}

A
Andrea Arcangeli 已提交
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
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;
}

2006 2007 2008 2009 2010
static void __init khugepaged_slab_exit(void)
{
	kmem_cache_destroy(mm_slot_cache);
}

A
Andrea Arcangeli 已提交
2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026
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;

2027
	hash_for_each_possible(mm_slots_hash, mm_slot, hash, (unsigned long)mm)
A
Andrea Arcangeli 已提交
2028 2029
		if (mm == mm_slot->mm)
			return mm_slot;
2030

A
Andrea Arcangeli 已提交
2031 2032 2033 2034 2035 2036 2037
	return NULL;
}

static void insert_to_mm_slots_hash(struct mm_struct *mm,
				    struct mm_slot *mm_slot)
{
	mm_slot->mm = mm;
2038
	hash_add(mm_slots_hash, &mm_slot->hash, (long)mm);
A
Andrea Arcangeli 已提交
2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055
}

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
2056
	VM_BUG_ON_MM(khugepaged_test_exit(mm), mm);
A
Andrea Arcangeli 已提交
2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078
	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;
}

2079 2080
int khugepaged_enter_vma_merge(struct vm_area_struct *vma,
			       unsigned long vm_flags)
A
Andrea Arcangeli 已提交
2081 2082 2083 2084 2085 2086 2087 2088
{
	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;
2089
	if (vma->vm_ops)
A
Andrea Arcangeli 已提交
2090 2091
		/* khugepaged not yet working on file or special mappings */
		return 0;
2092
	VM_BUG_ON_VMA(vm_flags & VM_NO_THP, vma);
A
Andrea Arcangeli 已提交
2093 2094 2095
	hstart = (vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK;
	hend = vma->vm_end & HPAGE_PMD_MASK;
	if (hstart < hend)
2096
		return khugepaged_enter(vma, vm_flags);
A
Andrea Arcangeli 已提交
2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107
	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) {
2108
		hash_del(&mm_slot->hash);
A
Andrea Arcangeli 已提交
2109 2110 2111
		list_del(&mm_slot->mm_node);
		free = 1;
	}
2112
	spin_unlock(&khugepaged_mm_lock);
A
Andrea Arcangeli 已提交
2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128

	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);
2129
	}
A
Andrea Arcangeli 已提交
2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143
}

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;
2144
		if (!pte_none(pteval) && !is_zero_pfn(pte_pfn(pteval)))
A
Andrea Arcangeli 已提交
2145 2146 2147 2148 2149 2150 2151 2152 2153 2154
			release_pte_page(pte_page(pteval));
	}
}

static int __collapse_huge_page_isolate(struct vm_area_struct *vma,
					unsigned long address,
					pte_t *pte)
{
	struct page *page;
	pte_t *_pte;
2155
	int none_or_zero = 0;
2156
	bool referenced = false, writable = false;
A
Andrea Arcangeli 已提交
2157 2158 2159
	for (_pte = pte; _pte < pte+HPAGE_PMD_NR;
	     _pte++, address += PAGE_SIZE) {
		pte_t pteval = *_pte;
2160 2161
		if (pte_none(pteval) || is_zero_pfn(pte_pfn(pteval))) {
			if (++none_or_zero <= khugepaged_max_ptes_none)
A
Andrea Arcangeli 已提交
2162
				continue;
2163
			else
A
Andrea Arcangeli 已提交
2164 2165
				goto out;
		}
2166
		if (!pte_present(pteval))
A
Andrea Arcangeli 已提交
2167 2168
			goto out;
		page = vm_normal_page(vma, address, pteval);
2169
		if (unlikely(!page))
A
Andrea Arcangeli 已提交
2170
			goto out;
2171

2172 2173 2174
		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
2175 2176 2177 2178 2179 2180 2181

		/*
		 * 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.
		 */
2182
		if (!trylock_page(page))
A
Andrea Arcangeli 已提交
2183
			goto out;
2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206

		/*
		 * 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);
			goto out;
		}
		if (pte_write(pteval)) {
			writable = true;
		} else {
			if (PageSwapCache(page) && !reuse_swap_page(page)) {
				unlock_page(page);
				goto out;
			}
			/*
			 * Page is not in the swap cache. It can be collapsed
			 * into a THP.
			 */
		}

A
Andrea Arcangeli 已提交
2207 2208 2209 2210 2211 2212 2213 2214 2215 2216
		/*
		 * Isolate the page to avoid collapsing an hugepage
		 * currently in use by the VM.
		 */
		if (isolate_lru_page(page)) {
			unlock_page(page);
			goto out;
		}
		/* 0 stands for page_is_file_cache(page) == false */
		inc_zone_page_state(page, NR_ISOLATED_ANON + 0);
2217 2218
		VM_BUG_ON_PAGE(!PageLocked(page), page);
		VM_BUG_ON_PAGE(PageLRU(page), page);
A
Andrea Arcangeli 已提交
2219 2220

		/* If there is no mapped pte young don't collapse the page */
A
Andrea Arcangeli 已提交
2221 2222
		if (pte_young(pteval) || PageReferenced(page) ||
		    mmu_notifier_test_young(vma->vm_mm, address))
2223
			referenced = true;
A
Andrea Arcangeli 已提交
2224
	}
2225
	if (likely(referenced && writable))
2226
		return 1;
A
Andrea Arcangeli 已提交
2227
out:
2228 2229
	release_pte_pages(pte, _pte);
	return 0;
A
Andrea Arcangeli 已提交
2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241
}

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;

2242
		if (pte_none(pteval) || is_zero_pfn(pte_pfn(pteval))) {
A
Andrea Arcangeli 已提交
2243 2244
			clear_user_highpage(page, address);
			add_mm_counter(vma->vm_mm, MM_ANONPAGES, 1);
2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256
			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
2257 2258 2259
		} else {
			src_page = pte_page(pteval);
			copy_user_highpage(page, src_page, address, vma);
2260
			VM_BUG_ON_PAGE(page_mapcount(src_page) != 1, src_page);
A
Andrea Arcangeli 已提交
2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282
			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);
			page_remove_rmap(src_page);
			spin_unlock(ptl);
			free_page_and_swap_cache(src_page);
		}

		address += PAGE_SIZE;
		page++;
	}
}

2283
static void khugepaged_alloc_sleep(void)
A
Andrea Arcangeli 已提交
2284
{
2285 2286 2287
	wait_event_freezable_timeout(khugepaged_wait, false,
			msecs_to_jiffies(khugepaged_alloc_sleep_millisecs));
}
A
Andrea Arcangeli 已提交
2288

2289 2290
static int khugepaged_node_load[MAX_NUMNODES];

2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314
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;
}

2315
#ifdef CONFIG_NUMA
2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340
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;
}

2341 2342 2343 2344 2345 2346 2347
static bool khugepaged_prealloc_page(struct page **hpage, bool *wait)
{
	if (IS_ERR(*hpage)) {
		if (!*wait)
			return false;

		*wait = false;
2348
		*hpage = NULL;
2349 2350 2351 2352 2353 2354 2355 2356 2357
		khugepaged_alloc_sleep();
	} else if (*hpage) {
		put_page(*hpage);
		*hpage = NULL;
	}

	return true;
}

2358 2359
static struct page *
khugepaged_alloc_page(struct page **hpage, gfp_t gfp, struct mm_struct *mm,
2360 2361 2362
		       struct vm_area_struct *vma, unsigned long address,
		       int node)
{
2363
	VM_BUG_ON_PAGE(*hpage, *hpage);
2364

2365
	/*
2366 2367 2368 2369
	 * 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.
2370
	 */
2371 2372
	up_read(&mm->mmap_sem);

2373
	*hpage = alloc_pages_exact_node(node, gfp, HPAGE_PMD_ORDER);
2374
	if (unlikely(!*hpage)) {
2375
		count_vm_event(THP_COLLAPSE_ALLOC_FAILED);
2376
		*hpage = ERR_PTR(-ENOMEM);
2377
		return NULL;
2378
	}
2379

2380
	count_vm_event(THP_COLLAPSE_ALLOC);
2381 2382 2383
	return *hpage;
}
#else
2384 2385 2386 2387 2388
static int khugepaged_find_target_node(void)
{
	return 0;
}

2389 2390 2391 2392 2393 2394
static inline struct page *alloc_hugepage(int defrag)
{
	return alloc_pages(alloc_hugepage_gfpmask(defrag, 0),
			   HPAGE_PMD_ORDER);
}

2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425
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;
}

2426 2427
static struct page *
khugepaged_alloc_page(struct page **hpage, gfp_t gfp, struct mm_struct *mm,
2428 2429 2430 2431 2432
		       struct vm_area_struct *vma, unsigned long address,
		       int node)
{
	up_read(&mm->mmap_sem);
	VM_BUG_ON(!*hpage);
2433

2434 2435
	return  *hpage;
}
2436 2437
#endif

B
Bob Liu 已提交
2438 2439 2440 2441 2442 2443 2444 2445 2446 2447
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;
2448
	VM_BUG_ON_VMA(vma->vm_flags & VM_NO_THP, vma);
B
Bob Liu 已提交
2449 2450 2451
	return true;
}

2452 2453 2454 2455 2456 2457 2458 2459 2460 2461
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;
2462
	spinlock_t *pmd_ptl, *pte_ptl;
2463 2464
	int isolated;
	unsigned long hstart, hend;
2465
	struct mem_cgroup *memcg;
2466 2467
	unsigned long mmun_start;	/* For mmu_notifiers */
	unsigned long mmun_end;		/* For mmu_notifiers */
2468
	gfp_t gfp;
2469 2470 2471

	VM_BUG_ON(address & ~HPAGE_PMD_MASK);

2472 2473 2474 2475
	/* Only allocate from the target node */
	gfp = alloc_hugepage_gfpmask(khugepaged_defrag(), __GFP_OTHER_NODE) |
		__GFP_THISNODE;

2476
	/* release the mmap_sem read lock. */
2477
	new_page = khugepaged_alloc_page(hpage, gfp, mm, vma, address, node);
2478 2479 2480
	if (!new_page)
		return;

2481
	if (unlikely(mem_cgroup_try_charge(new_page, mm,
2482
					   gfp, &memcg)))
2483
		return;
A
Andrea Arcangeli 已提交
2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494

	/*
	 * 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);
	if (unlikely(khugepaged_test_exit(mm)))
		goto out;

	vma = find_vma(mm, address);
2495 2496
	if (!vma)
		goto out;
A
Andrea Arcangeli 已提交
2497 2498 2499 2500
	hstart = (vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK;
	hend = vma->vm_end & HPAGE_PMD_MASK;
	if (address < hstart || address + HPAGE_PMD_SIZE > hend)
		goto out;
B
Bob Liu 已提交
2501
	if (!hugepage_vma_check(vma))
2502
		goto out;
B
Bob Liu 已提交
2503 2504
	pmd = mm_find_pmd(mm, address);
	if (!pmd)
A
Andrea Arcangeli 已提交
2505 2506
		goto out;

2507
	anon_vma_lock_write(vma->anon_vma);
A
Andrea Arcangeli 已提交
2508 2509

	pte = pte_offset_map(pmd, address);
2510
	pte_ptl = pte_lockptr(mm, pmd);
A
Andrea Arcangeli 已提交
2511

2512 2513 2514
	mmun_start = address;
	mmun_end   = address + HPAGE_PMD_SIZE;
	mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
2515
	pmd_ptl = pmd_lock(mm, pmd); /* probably unnecessary */
A
Andrea Arcangeli 已提交
2516 2517 2518 2519 2520 2521
	/*
	 * 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.
	 */
2522
	_pmd = pmdp_collapse_flush(vma, address, pmd);
2523
	spin_unlock(pmd_ptl);
2524
	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
A
Andrea Arcangeli 已提交
2525

2526
	spin_lock(pte_ptl);
A
Andrea Arcangeli 已提交
2527
	isolated = __collapse_huge_page_isolate(vma, address, pte);
2528
	spin_unlock(pte_ptl);
A
Andrea Arcangeli 已提交
2529 2530

	if (unlikely(!isolated)) {
2531
		pte_unmap(pte);
2532
		spin_lock(pmd_ptl);
A
Andrea Arcangeli 已提交
2533
		BUG_ON(!pmd_none(*pmd));
2534 2535 2536 2537 2538 2539
		/*
		 * 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));
2540
		spin_unlock(pmd_ptl);
2541
		anon_vma_unlock_write(vma->anon_vma);
2542
		goto out;
A
Andrea Arcangeli 已提交
2543 2544 2545 2546 2547 2548
	}

	/*
	 * All pages are isolated and locked so anon_vma rmap
	 * can't run anymore.
	 */
2549
	anon_vma_unlock_write(vma->anon_vma);
A
Andrea Arcangeli 已提交
2550

2551
	__collapse_huge_page_copy(pte, new_page, vma, address, pte_ptl);
2552
	pte_unmap(pte);
A
Andrea Arcangeli 已提交
2553 2554 2555
	__SetPageUptodate(new_page);
	pgtable = pmd_pgtable(_pmd);

2556 2557
	_pmd = mk_huge_pmd(new_page, vma->vm_page_prot);
	_pmd = maybe_pmd_mkwrite(pmd_mkdirty(_pmd), vma);
A
Andrea Arcangeli 已提交
2558 2559 2560 2561 2562 2563 2564 2565

	/*
	 * 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();

2566
	spin_lock(pmd_ptl);
A
Andrea Arcangeli 已提交
2567 2568
	BUG_ON(!pmd_none(*pmd));
	page_add_new_anon_rmap(new_page, vma, address);
2569 2570
	mem_cgroup_commit_charge(new_page, memcg, false);
	lru_cache_add_active_or_unevictable(new_page, vma);
2571
	pgtable_trans_huge_deposit(mm, pmd, pgtable);
A
Andrea Arcangeli 已提交
2572
	set_pmd_at(mm, address, pmd, _pmd);
2573
	update_mmu_cache_pmd(vma, address, pmd);
2574
	spin_unlock(pmd_ptl);
A
Andrea Arcangeli 已提交
2575 2576

	*hpage = NULL;
2577

A
Andrea Arcangeli 已提交
2578
	khugepaged_pages_collapsed++;
2579
out_up_write:
A
Andrea Arcangeli 已提交
2580
	up_write(&mm->mmap_sem);
2581 2582
	return;

2583
out:
2584
	mem_cgroup_cancel_charge(new_page, memcg);
2585
	goto out_up_write;
A
Andrea Arcangeli 已提交
2586 2587 2588 2589 2590 2591 2592 2593 2594
}

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;
2595
	int ret = 0, none_or_zero = 0;
A
Andrea Arcangeli 已提交
2596 2597 2598
	struct page *page;
	unsigned long _address;
	spinlock_t *ptl;
D
David Rientjes 已提交
2599
	int node = NUMA_NO_NODE;
2600
	bool writable = false, referenced = false;
A
Andrea Arcangeli 已提交
2601 2602 2603

	VM_BUG_ON(address & ~HPAGE_PMD_MASK);

B
Bob Liu 已提交
2604 2605
	pmd = mm_find_pmd(mm, address);
	if (!pmd)
A
Andrea Arcangeli 已提交
2606 2607
		goto out;

2608
	memset(khugepaged_node_load, 0, sizeof(khugepaged_node_load));
A
Andrea Arcangeli 已提交
2609 2610 2611 2612
	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;
2613 2614
		if (pte_none(pteval) || is_zero_pfn(pte_pfn(pteval))) {
			if (++none_or_zero <= khugepaged_max_ptes_none)
A
Andrea Arcangeli 已提交
2615 2616 2617 2618
				continue;
			else
				goto out_unmap;
		}
2619
		if (!pte_present(pteval))
A
Andrea Arcangeli 已提交
2620
			goto out_unmap;
2621 2622 2623
		if (pte_write(pteval))
			writable = true;

A
Andrea Arcangeli 已提交
2624 2625 2626
		page = vm_normal_page(vma, _address, pteval);
		if (unlikely(!page))
			goto out_unmap;
2627
		/*
2628 2629 2630 2631
		 * 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.
2632
		 */
2633
		node = page_to_nid(page);
2634 2635
		if (khugepaged_scan_abort(node))
			goto out_unmap;
2636
		khugepaged_node_load[node]++;
2637
		VM_BUG_ON_PAGE(PageCompound(page), page);
A
Andrea Arcangeli 已提交
2638 2639
		if (!PageLRU(page) || PageLocked(page) || !PageAnon(page))
			goto out_unmap;
2640 2641 2642 2643 2644 2645
		/*
		 * 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))
A
Andrea Arcangeli 已提交
2646
			goto out_unmap;
A
Andrea Arcangeli 已提交
2647 2648
		if (pte_young(pteval) || PageReferenced(page) ||
		    mmu_notifier_test_young(vma->vm_mm, address))
2649
			referenced = true;
A
Andrea Arcangeli 已提交
2650
	}
2651
	if (referenced && writable)
A
Andrea Arcangeli 已提交
2652 2653 2654
		ret = 1;
out_unmap:
	pte_unmap_unlock(pte, ptl);
2655 2656
	if (ret) {
		node = khugepaged_find_target_node();
2657
		/* collapse_huge_page will return with the mmap_sem released */
2658
		collapse_huge_page(mm, address, hpage, vma, node);
2659
	}
A
Andrea Arcangeli 已提交
2660 2661 2662 2663 2664 2665 2666 2667
out:
	return ret;
}

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

2668
	VM_BUG_ON(NR_CPUS != 1 && !spin_is_locked(&khugepaged_mm_lock));
A
Andrea Arcangeli 已提交
2669 2670 2671

	if (khugepaged_test_exit(mm)) {
		/* free mm_slot */
2672
		hash_del(&mm_slot->hash);
A
Andrea Arcangeli 已提交
2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688
		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)
2689 2690
	__releases(&khugepaged_mm_lock)
	__acquires(&khugepaged_mm_lock)
A
Andrea Arcangeli 已提交
2691 2692 2693 2694 2695 2696 2697
{
	struct mm_slot *mm_slot;
	struct mm_struct *mm;
	struct vm_area_struct *vma;
	int progress = 0;

	VM_BUG_ON(!pages);
2698
	VM_BUG_ON(NR_CPUS != 1 && !spin_is_locked(&khugepaged_mm_lock));
A
Andrea Arcangeli 已提交
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

	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
2726 2727
		if (!hugepage_vma_check(vma)) {
skip:
A
Andrea Arcangeli 已提交
2728 2729 2730 2731 2732
			progress++;
			continue;
		}
		hstart = (vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK;
		hend = vma->vm_end & HPAGE_PMD_MASK;
2733 2734 2735 2736
		if (hstart >= hend)
			goto skip;
		if (khugepaged_scan.address > hend)
			goto skip;
A
Andrea Arcangeli 已提交
2737 2738
		if (khugepaged_scan.address < hstart)
			khugepaged_scan.address = hstart;
2739
		VM_BUG_ON(khugepaged_scan.address & ~HPAGE_PMD_MASK);
A
Andrea Arcangeli 已提交
2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767

		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);
2768
	VM_BUG_ON(khugepaged_scan.mm_slot != mm_slot);
A
Andrea Arcangeli 已提交
2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803
	/*
	 * 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) ||
2804
		kthread_should_stop();
A
Andrea Arcangeli 已提交
2805 2806
}

2807
static void khugepaged_do_scan(void)
A
Andrea Arcangeli 已提交
2808
{
2809
	struct page *hpage = NULL;
A
Andrea Arcangeli 已提交
2810 2811
	unsigned int progress = 0, pass_through_head = 0;
	unsigned int pages = khugepaged_pages_to_scan;
2812
	bool wait = true;
A
Andrea Arcangeli 已提交
2813 2814 2815 2816

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

	while (progress < pages) {
2817
		if (!khugepaged_prealloc_page(&hpage, &wait))
2818
			break;
2819

2820
		cond_resched();
A
Andrea Arcangeli 已提交
2821

2822
		if (unlikely(kthread_should_stop() || try_to_freeze()))
2823 2824
			break;

A
Andrea Arcangeli 已提交
2825 2826 2827 2828 2829 2830
		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,
2831
							    &hpage);
A
Andrea Arcangeli 已提交
2832 2833 2834 2835 2836
		else
			progress = pages;
		spin_unlock(&khugepaged_mm_lock);
	}

2837 2838
	if (!IS_ERR_OR_NULL(hpage))
		put_page(hpage);
2839 2840
}

2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856
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
2857 2858 2859 2860
static int khugepaged(void *none)
{
	struct mm_slot *mm_slot;

2861
	set_freezable();
2862
	set_user_nice(current, MAX_NICE);
A
Andrea Arcangeli 已提交
2863

X
Xiao Guangrong 已提交
2864 2865 2866 2867
	while (!kthread_should_stop()) {
		khugepaged_do_scan();
		khugepaged_wait_work();
	}
A
Andrea Arcangeli 已提交
2868 2869 2870 2871 2872 2873 2874 2875 2876 2877

	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;
}

2878 2879 2880 2881 2882 2883 2884 2885
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;

2886
	pmdp_huge_clear_flush_notify(vma, haddr, pmd);
2887 2888
	/* leave pmd empty until pte is filled */

2889
	pgtable = pgtable_trans_huge_withdraw(mm, pmd);
2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902
	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);
2903
	put_huge_zero_page();
2904 2905
}

2906 2907
void __split_huge_page_pmd(struct vm_area_struct *vma, unsigned long address,
		pmd_t *pmd)
2908
{
2909
	spinlock_t *ptl;
2910
	struct page *page;
2911
	struct mm_struct *mm = vma->vm_mm;
2912 2913 2914
	unsigned long haddr = address & HPAGE_PMD_MASK;
	unsigned long mmun_start;	/* For mmu_notifiers */
	unsigned long mmun_end;		/* For mmu_notifiers */
2915 2916

	BUG_ON(vma->vm_start > haddr || vma->vm_end < haddr + HPAGE_PMD_SIZE);
2917

2918 2919
	mmun_start = haddr;
	mmun_end   = haddr + HPAGE_PMD_SIZE;
2920
again:
2921
	mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
2922
	ptl = pmd_lock(mm, pmd);
2923
	if (unlikely(!pmd_trans_huge(*pmd))) {
2924
		spin_unlock(ptl);
2925 2926 2927 2928 2929
		mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
		return;
	}
	if (is_huge_zero_pmd(*pmd)) {
		__split_huge_zero_page_pmd(vma, haddr, pmd);
2930
		spin_unlock(ptl);
2931
		mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
2932 2933 2934
		return;
	}
	page = pmd_page(*pmd);
2935
	VM_BUG_ON_PAGE(!page_count(page), page);
2936
	get_page(page);
2937
	spin_unlock(ptl);
2938
	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
2939 2940 2941 2942

	split_huge_page(page);

	put_page(page);
2943 2944 2945 2946 2947 2948 2949 2950

	/*
	 * We don't always have down_write of mmap_sem here: a racing
	 * do_huge_pmd_wp_page() might have copied-on-write to another
	 * huge page before our split_huge_page() got the anon_vma lock.
	 */
	if (unlikely(pmd_trans_huge(*pmd)))
		goto again;
2951
}
2952

2953 2954 2955 2956 2957 2958 2959 2960 2961 2962
void split_huge_page_pmd_mm(struct mm_struct *mm, unsigned long address,
		pmd_t *pmd)
{
	struct vm_area_struct *vma;

	vma = find_vma(mm, address);
	BUG_ON(vma == NULL);
	split_huge_page_pmd(vma, address, pmd);
}

2963 2964 2965
static void split_huge_page_address(struct mm_struct *mm,
				    unsigned long address)
{
2966 2967
	pgd_t *pgd;
	pud_t *pud;
2968 2969 2970 2971
	pmd_t *pmd;

	VM_BUG_ON(!(address & ~HPAGE_PMD_MASK));

2972 2973 2974 2975 2976 2977 2978 2979 2980 2981
	pgd = pgd_offset(mm, address);
	if (!pgd_present(*pgd))
		return;

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

	pmd = pmd_offset(pud, address);
	if (!pmd_present(*pmd))
2982 2983 2984 2985 2986
		return;
	/*
	 * Caller holds the mmap_sem write mode, so a huge pmd cannot
	 * materialize from under us.
	 */
2987
	split_huge_page_pmd_mm(mm, address, pmd);
2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029
}

void __vma_adjust_trans_huge(struct vm_area_struct *vma,
			     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)
		split_huge_page_address(vma->vm_mm, start);

	/*
	 * 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)
		split_huge_page_address(vma->vm_mm, end);

	/*
	 * 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)
			split_huge_page_address(next->vm_mm, nstart);
	}
}