huge_memory.c 81.3 KB
Newer Older
1 2 3 4 5 6 7
/*
 *  Copyright (C) 2009  Red Hat, Inc.
 *
 *  This work is licensed under the terms of the GNU GPL, version 2. See
 *  the COPYING file in the top-level directory.
 */

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

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

30 31 32 33
#include <asm/tlb.h>
#include <asm/pgalloc.h>
#include "internal.h"

A
Andrea Arcangeli 已提交
34
/*
35 36 37 38 39 40
 * 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
41
 */
42
unsigned long transparent_hugepage_flags __read_mostly =
43
#ifdef CONFIG_TRANSPARENT_HUGEPAGE_ALWAYS
A
Andrea Arcangeli 已提交
44
	(1<<TRANSPARENT_HUGEPAGE_FLAG)|
45 46 47 48
#endif
#ifdef CONFIG_TRANSPARENT_HUGEPAGE_MADVISE
	(1<<TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG)|
#endif
A
Andrea Arcangeli 已提交
49
	(1<<TRANSPARENT_HUGEPAGE_DEFRAG_FLAG)|
50 51
	(1<<TRANSPARENT_HUGEPAGE_DEFRAG_KHUGEPAGED_FLAG)|
	(1<<TRANSPARENT_HUGEPAGE_USE_ZERO_PAGE_FLAG);
A
Andrea Arcangeli 已提交
52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72

/* 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);
73
static void khugepaged_slab_exit(void);
A
Andrea Arcangeli 已提交
74

75 76 77
#define MM_SLOTS_HASH_BITS 10
static __read_mostly DEFINE_HASHTABLE(mm_slots_hash, MM_SLOTS_HASH_BITS);

A
Andrea Arcangeli 已提交
78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103
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;
104 105
};
static struct khugepaged_scan khugepaged_scan = {
A
Andrea Arcangeli 已提交
106 107 108
	.mm_head = LIST_HEAD_INIT(khugepaged_scan.mm_head),
};

109

110
static void set_recommended_min_free_kbytes(void)
111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135
{
	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);

136 137 138 139 140 141
	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);

142
		min_free_kbytes = recommended_min;
143
	}
144 145 146
	setup_per_zone_wmarks();
}

147
static int start_stop_khugepaged(void)
A
Andrea Arcangeli 已提交
148 149 150 151 152 153 154
{
	int err = 0;
	if (khugepaged_enabled()) {
		if (!khugepaged_thread)
			khugepaged_thread = kthread_run(khugepaged, NULL,
							"khugepaged");
		if (unlikely(IS_ERR(khugepaged_thread))) {
155
			pr_err("khugepaged: kthread_run(khugepaged) failed\n");
A
Andrea Arcangeli 已提交
156 157
			err = PTR_ERR(khugepaged_thread);
			khugepaged_thread = NULL;
158
			goto fail;
A
Andrea Arcangeli 已提交
159
		}
160 161

		if (!list_empty(&khugepaged_scan.mm_head))
A
Andrea Arcangeli 已提交
162
			wake_up_interruptible(&khugepaged_wait);
163 164

		set_recommended_min_free_kbytes();
165 166 167 168
	} else if (khugepaged_thread) {
		kthread_stop(khugepaged_thread);
		khugepaged_thread = NULL;
	}
169
fail:
A
Andrea Arcangeli 已提交
170 171
	return err;
}
172

173
static atomic_t huge_zero_refcount;
174
struct page *huge_zero_page __read_mostly;
175

176
struct page *get_huge_zero_page(void)
177 178 179 180
{
	struct page *zero_page;
retry:
	if (likely(atomic_inc_not_zero(&huge_zero_refcount)))
181
		return READ_ONCE(huge_zero_page);
182 183

	zero_page = alloc_pages((GFP_TRANSHUGE | __GFP_ZERO) & ~__GFP_MOVABLE,
184
			HPAGE_PMD_ORDER);
185 186
	if (!zero_page) {
		count_vm_event(THP_ZERO_PAGE_ALLOC_FAILED);
187
		return NULL;
188 189
	}
	count_vm_event(THP_ZERO_PAGE_ALLOC);
190
	preempt_disable();
191
	if (cmpxchg(&huge_zero_page, NULL, zero_page)) {
192
		preempt_enable();
193
		__free_pages(zero_page, compound_order(zero_page));
194 195 196 197 198 199
		goto retry;
	}

	/* We take additional reference here. It will be put back by shrinker */
	atomic_set(&huge_zero_refcount, 2);
	preempt_enable();
200
	return READ_ONCE(huge_zero_page);
201 202
}

203
static void put_huge_zero_page(void)
204
{
205 206 207 208 209
	/*
	 * Counter should never go to zero here. Only shrinker can put
	 * last reference.
	 */
	BUG_ON(atomic_dec_and_test(&huge_zero_refcount));
210 211
}

212 213
static unsigned long shrink_huge_zero_page_count(struct shrinker *shrink,
					struct shrink_control *sc)
214
{
215 216 217
	/* we can free zero page only if last reference remains */
	return atomic_read(&huge_zero_refcount) == 1 ? HPAGE_PMD_NR : 0;
}
218

219 220 221
static unsigned long shrink_huge_zero_page_scan(struct shrinker *shrink,
				       struct shrink_control *sc)
{
222
	if (atomic_cmpxchg(&huge_zero_refcount, 1, 0) == 1) {
223 224
		struct page *zero_page = xchg(&huge_zero_page, NULL);
		BUG_ON(zero_page == NULL);
225
		__free_pages(zero_page, compound_order(zero_page));
226
		return HPAGE_PMD_NR;
227 228 229
	}

	return 0;
230 231
}

232
static struct shrinker huge_zero_page_shrinker = {
233 234
	.count_objects = shrink_huge_zero_page_count,
	.scan_objects = shrink_huge_zero_page_scan,
235 236 237
	.seeks = DEFAULT_SEEKS,
};

238
#ifdef CONFIG_SYSFS
A
Andrea Arcangeli 已提交
239

240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287
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
288 289 290 291 292 293 294
	ssize_t ret;

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

	if (ret > 0) {
295 296 297
		int err;

		mutex_lock(&khugepaged_mutex);
298
		err = start_stop_khugepaged();
299 300
		mutex_unlock(&khugepaged_mutex);

A
Andrea Arcangeli 已提交
301 302 303 304 305
		if (err)
			ret = err;
	}

	return ret;
306 307 308 309 310 311 312 313
}
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)
{
314 315
	return sprintf(buf, "%d\n",
		       !!test_bit(flag, &transparent_hugepage_flags));
316
}
317

318 319 320 321 322
static ssize_t single_flag_store(struct kobject *kobj,
				 struct kobj_attribute *attr,
				 const char *buf, size_t count,
				 enum transparent_hugepage_flag flag)
{
323 324 325 326 327 328 329 330 331 332
	unsigned long value;
	int ret;

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

	if (value)
333
		set_bit(flag, &transparent_hugepage_flags);
334
	else
335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362
		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);

363 364 365 366 367 368 369 370 371 372 373 374 375 376
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);
377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397
#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,
398
	&use_zero_page_attr.attr,
399 400 401 402 403 404 405 406
#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
407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422
};

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;

423
	err = kstrtoul(buf, 10, &msecs);
A
Andrea Arcangeli 已提交
424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449
	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;

450
	err = kstrtoul(buf, 10, &msecs);
A
Andrea Arcangeli 已提交
451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475
	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;

476
	err = kstrtoul(buf, 10, &pages);
A
Andrea Arcangeli 已提交
477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543
	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;

544
	err = kstrtoul(buf, 10, &max_ptes_none);
A
Andrea Arcangeli 已提交
545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569
	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",
570 571
};

S
Shaohua Li 已提交
572
static int __init hugepage_init_sysfs(struct kobject **hugepage_kobj)
573 574 575
{
	int err;

S
Shaohua Li 已提交
576 577
	*hugepage_kobj = kobject_create_and_add("transparent_hugepage", mm_kobj);
	if (unlikely(!*hugepage_kobj)) {
578
		pr_err("failed to create transparent hugepage kobject\n");
S
Shaohua Li 已提交
579
		return -ENOMEM;
A
Andrea Arcangeli 已提交
580 581
	}

S
Shaohua Li 已提交
582
	err = sysfs_create_group(*hugepage_kobj, &hugepage_attr_group);
A
Andrea Arcangeli 已提交
583
	if (err) {
584
		pr_err("failed to register transparent hugepage group\n");
S
Shaohua Li 已提交
585
		goto delete_obj;
A
Andrea Arcangeli 已提交
586 587
	}

S
Shaohua Li 已提交
588
	err = sysfs_create_group(*hugepage_kobj, &khugepaged_attr_group);
A
Andrea Arcangeli 已提交
589
	if (err) {
590
		pr_err("failed to register transparent hugepage group\n");
S
Shaohua Li 已提交
591
		goto remove_hp_group;
A
Andrea Arcangeli 已提交
592
	}
S
Shaohua Li 已提交
593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631

	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)
632
		goto err_sysfs;
A
Andrea Arcangeli 已提交
633 634 635

	err = khugepaged_slab_init();
	if (err)
636
		goto err_slab;
A
Andrea Arcangeli 已提交
637

638 639 640
	err = register_shrinker(&huge_zero_page_shrinker);
	if (err)
		goto err_hzp_shrinker;
641

642 643 644 645 646
	/*
	 * 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.
	 */
647
	if (totalram_pages < (512 << (20 - PAGE_SHIFT))) {
648
		transparent_hugepage_flags = 0;
649 650
		return 0;
	}
651

652
	err = start_stop_khugepaged();
653 654
	if (err)
		goto err_khugepaged;
A
Andrea Arcangeli 已提交
655

S
Shaohua Li 已提交
656
	return 0;
657 658 659 660 661
err_khugepaged:
	unregister_shrinker(&huge_zero_page_shrinker);
err_hzp_shrinker:
	khugepaged_slab_exit();
err_slab:
S
Shaohua Li 已提交
662
	hugepage_exit_sysfs(hugepage_kobj);
663
err_sysfs:
A
Andrea Arcangeli 已提交
664
	return err;
665
}
666
subsys_initcall(hugepage_init);
667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693

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)
694
		pr_warn("transparent_hugepage= cannot parse, ignored\n");
695 696 697 698
	return ret;
}
__setup("transparent_hugepage=", setup_transparent_hugepage);

699
pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma)
700 701 702 703 704 705
{
	if (likely(vma->vm_flags & VM_WRITE))
		pmd = pmd_mkwrite(pmd);
	return pmd;
}

706
static inline pmd_t mk_huge_pmd(struct page *page, pgprot_t prot)
B
Bob Liu 已提交
707 708
{
	pmd_t entry;
709
	entry = mk_pmd(page, prot);
B
Bob Liu 已提交
710 711 712 713
	entry = pmd_mkhuge(entry);
	return entry;
}

714 715
static int __do_huge_pmd_anonymous_page(struct mm_struct *mm,
					struct vm_area_struct *vma,
716
					unsigned long address, pmd_t *pmd,
717 718
					struct page *page, gfp_t gfp,
					unsigned int flags)
719
{
720
	struct mem_cgroup *memcg;
721
	pgtable_t pgtable;
722
	spinlock_t *ptl;
723
	unsigned long haddr = address & HPAGE_PMD_MASK;
724

725
	VM_BUG_ON_PAGE(!PageCompound(page), page);
726

727 728 729 730 731
	if (mem_cgroup_try_charge(page, mm, gfp, &memcg)) {
		put_page(page);
		count_vm_event(THP_FAULT_FALLBACK);
		return VM_FAULT_FALLBACK;
	}
732

733
	pgtable = pte_alloc_one(mm, haddr);
734 735
	if (unlikely(!pgtable)) {
		mem_cgroup_cancel_charge(page, memcg);
736
		put_page(page);
737
		return VM_FAULT_OOM;
738
	}
739 740

	clear_huge_page(page, haddr, HPAGE_PMD_NR);
741 742 743 744 745
	/*
	 * The memory barrier inside __SetPageUptodate makes sure that
	 * clear_huge_page writes become visible before the set_pmd_at()
	 * write.
	 */
746 747
	__SetPageUptodate(page);

748
	ptl = pmd_lock(mm, pmd);
749
	if (unlikely(!pmd_none(*pmd))) {
750
		spin_unlock(ptl);
751
		mem_cgroup_cancel_charge(page, memcg);
752 753 754 755
		put_page(page);
		pte_free(mm, pgtable);
	} else {
		pmd_t entry;
756 757 758 759 760 761 762 763 764

		/* 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);
765
			ret = handle_userfault(vma, address, flags,
766 767 768 769 770
					       VM_UFFD_MISSING);
			VM_BUG_ON(ret & VM_FAULT_FALLBACK);
			return ret;
		}

771 772
		entry = mk_huge_pmd(page, vma->vm_page_prot);
		entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma);
773
		page_add_new_anon_rmap(page, vma, haddr);
774 775
		mem_cgroup_commit_charge(page, memcg, false);
		lru_cache_add_active_or_unevictable(page, vma);
776
		pgtable_trans_huge_deposit(mm, pmd, pgtable);
777 778
		set_pmd_at(mm, haddr, pmd, entry);
		add_mm_counter(mm, MM_ANONPAGES, HPAGE_PMD_NR);
779
		atomic_long_inc(&mm->nr_ptes);
780
		spin_unlock(ptl);
781
		count_vm_event(THP_FAULT_ALLOC);
782 783
	}

784
	return 0;
785 786
}

787
static inline gfp_t alloc_hugepage_gfpmask(int defrag, gfp_t extra_gfp)
788
{
789
	return (GFP_TRANSHUGE & ~(defrag ? 0 : __GFP_WAIT)) | extra_gfp;
790 791
}

792
/* Caller must hold page table lock. */
793
static bool set_huge_zero_page(pgtable_t pgtable, struct mm_struct *mm,
794
		struct vm_area_struct *vma, unsigned long haddr, pmd_t *pmd,
795
		struct page *zero_page)
796 797
{
	pmd_t entry;
A
Andrew Morton 已提交
798 799
	if (!pmd_none(*pmd))
		return false;
800
	entry = mk_pmd(zero_page, vma->vm_page_prot);
801
	entry = pmd_mkhuge(entry);
802
	pgtable_trans_huge_deposit(mm, pmd, pgtable);
803
	set_pmd_at(mm, haddr, pmd, entry);
804
	atomic_long_inc(&mm->nr_ptes);
A
Andrew Morton 已提交
805
	return true;
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
		ret = 0;
		set = false;
		if (pmd_none(*pmd)) {
			if (userfaultfd_missing(vma)) {
				spin_unlock(ptl);
844
				ret = handle_userfault(vma, address, flags,
845 846 847 848 849 850 851 852 853 854 855
						       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
	}
868 869
	return __do_huge_pmd_anonymous_page(mm, vma, address, pmd, page, gfp,
					    flags);
870 871
}

872
static void insert_pfn_pmd(struct vm_area_struct *vma, unsigned long addr,
M
Matthew Wilcox 已提交
873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910
		pmd_t *pmd, unsigned long pfn, pgprot_t prot, bool write)
{
	struct mm_struct *mm = vma->vm_mm;
	pmd_t entry;
	spinlock_t *ptl;

	ptl = pmd_lock(mm, pmd);
	if (pmd_none(*pmd)) {
		entry = pmd_mkhuge(pfn_pmd(pfn, prot));
		if (write) {
			entry = pmd_mkyoung(pmd_mkdirty(entry));
			entry = maybe_pmd_mkwrite(entry, vma);
		}
		set_pmd_at(mm, addr, pmd, entry);
		update_mmu_cache_pmd(vma, addr, pmd);
	}
	spin_unlock(ptl);
}

int vmf_insert_pfn_pmd(struct vm_area_struct *vma, unsigned long addr,
			pmd_t *pmd, unsigned long pfn, bool write)
{
	pgprot_t pgprot = vma->vm_page_prot;
	/*
	 * If we had pmd_special, we could avoid all these restrictions,
	 * but we need to be consistent with PTEs and architectures that
	 * can't support a 'special' bit.
	 */
	BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
	BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
						(VM_PFNMAP|VM_MIXEDMAP));
	BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
	BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));

	if (addr < vma->vm_start || addr >= vma->vm_end)
		return VM_FAULT_SIGBUS;
	if (track_pfn_insert(vma, &pgprot, pfn))
		return VM_FAULT_SIGBUS;
911 912
	insert_pfn_pmd(vma, addr, pmd, pfn, pgprot, write);
	return VM_FAULT_NOPAGE;
M
Matthew Wilcox 已提交
913 914
}

915 916 917 918
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)
{
919
	spinlock_t *dst_ptl, *src_ptl;
920 921 922 923 924 925 926 927 928 929
	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;

930 931 932
	dst_ptl = pmd_lock(dst_mm, dst_pmd);
	src_ptl = pmd_lockptr(src_mm, src_pmd);
	spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
933 934 935 936 937 938 939

	ret = -EAGAIN;
	pmd = *src_pmd;
	if (unlikely(!pmd_trans_huge(pmd))) {
		pte_free(dst_mm, pgtable);
		goto out_unlock;
	}
940
	/*
941
	 * When page table lock is held, the huge zero pmd should not be
942 943 944 945
	 * under splitting since we don't split the page itself, only pmd to
	 * a page table.
	 */
	if (is_huge_zero_pmd(pmd)) {
946
		struct page *zero_page;
947 948 949 950 951
		/*
		 * 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.
		 */
952
		zero_page = get_huge_zero_page();
953
		set_huge_zero_page(pgtable, dst_mm, vma, addr, dst_pmd,
954
				zero_page);
955 956 957
		ret = 0;
		goto out_unlock;
	}
958

959 960
	if (unlikely(pmd_trans_splitting(pmd))) {
		/* split huge page running from under us */
961 962
		spin_unlock(src_ptl);
		spin_unlock(dst_ptl);
963 964 965 966 967 968
		pte_free(dst_mm, pgtable);

		wait_split_huge_page(vma->anon_vma, src_pmd); /* src_vma */
		goto out;
	}
	src_page = pmd_page(pmd);
969
	VM_BUG_ON_PAGE(!PageHead(src_page), src_page);
970 971 972 973 974 975
	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));
976
	pgtable_trans_huge_deposit(dst_mm, dst_pmd, pgtable);
977
	set_pmd_at(dst_mm, addr, dst_pmd, pmd);
978
	atomic_long_inc(&dst_mm->nr_ptes);
979 980 981

	ret = 0;
out_unlock:
982 983
	spin_unlock(src_ptl);
	spin_unlock(dst_ptl);
984 985 986 987
out:
	return ret;
}

988 989 990 991 992 993
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)
{
994
	spinlock_t *ptl;
995 996 997
	pmd_t entry;
	unsigned long haddr;

998
	ptl = pmd_lock(mm, pmd);
999 1000 1001 1002 1003 1004 1005 1006 1007
	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:
1008
	spin_unlock(ptl);
1009 1010
}

1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041
/*
 * 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);
	}
}

1042 1043 1044 1045 1046 1047 1048
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)
{
1049
	struct mem_cgroup *memcg;
1050
	spinlock_t *ptl;
1051 1052 1053 1054
	pgtable_t pgtable;
	pmd_t _pmd;
	int ret = 0, i;
	struct page **pages;
1055 1056
	unsigned long mmun_start;	/* For mmu_notifiers */
	unsigned long mmun_end;		/* For mmu_notifiers */
1057 1058 1059 1060 1061 1062 1063 1064 1065

	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++) {
1066 1067
		pages[i] = alloc_page_vma_node(GFP_HIGHUSER_MOVABLE |
					       __GFP_OTHER_NODE,
1068
					       vma, address, page_to_nid(page));
A
Andrea Arcangeli 已提交
1069
		if (unlikely(!pages[i] ||
1070 1071
			     mem_cgroup_try_charge(pages[i], mm, GFP_KERNEL,
						   &memcg))) {
A
Andrea Arcangeli 已提交
1072
			if (pages[i])
1073
				put_page(pages[i]);
A
Andrea Arcangeli 已提交
1074
			while (--i >= 0) {
1075 1076 1077
				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
1078 1079
				put_page(pages[i]);
			}
1080 1081 1082 1083
			kfree(pages);
			ret |= VM_FAULT_OOM;
			goto out;
		}
1084
		set_page_private(pages[i], (unsigned long)memcg);
1085 1086 1087 1088
	}

	for (i = 0; i < HPAGE_PMD_NR; i++) {
		copy_user_highpage(pages[i], page + i,
1089
				   haddr + PAGE_SIZE * i, vma);
1090 1091 1092 1093
		__SetPageUptodate(pages[i]);
		cond_resched();
	}

1094 1095 1096 1097
	mmun_start = haddr;
	mmun_end   = haddr + HPAGE_PMD_SIZE;
	mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);

1098
	ptl = pmd_lock(mm, pmd);
1099 1100
	if (unlikely(!pmd_same(*pmd, orig_pmd)))
		goto out_free_pages;
1101
	VM_BUG_ON_PAGE(!PageHead(page), page);
1102

1103
	pmdp_huge_clear_flush_notify(vma, haddr, pmd);
1104 1105
	/* leave pmd empty until pte is filled */

1106
	pgtable = pgtable_trans_huge_withdraw(mm, pmd);
1107 1108 1109 1110 1111 1112
	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);
1113 1114
		memcg = (void *)page_private(pages[i]);
		set_page_private(pages[i], 0);
1115
		page_add_new_anon_rmap(pages[i], vma, haddr);
1116 1117
		mem_cgroup_commit_charge(pages[i], memcg, false);
		lru_cache_add_active_or_unevictable(pages[i], vma);
1118 1119 1120 1121 1122 1123 1124 1125 1126 1127
		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);
1128
	spin_unlock(ptl);
1129

1130 1131
	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);

1132 1133 1134 1135 1136 1137 1138
	ret |= VM_FAULT_WRITE;
	put_page(page);

out:
	return ret;

out_free_pages:
1139
	spin_unlock(ptl);
1140
	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
A
Andrea Arcangeli 已提交
1141
	for (i = 0; i < HPAGE_PMD_NR; i++) {
1142 1143 1144
		memcg = (void *)page_private(pages[i]);
		set_page_private(pages[i], 0);
		mem_cgroup_cancel_charge(pages[i], memcg);
1145
		put_page(pages[i]);
A
Andrea Arcangeli 已提交
1146
	}
1147 1148 1149 1150 1151 1152 1153
	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)
{
1154
	spinlock_t *ptl;
1155
	int ret = 0;
1156
	struct page *page = NULL, *new_page;
1157
	struct mem_cgroup *memcg;
1158
	unsigned long haddr;
1159 1160
	unsigned long mmun_start;	/* For mmu_notifiers */
	unsigned long mmun_end;		/* For mmu_notifiers */
1161
	gfp_t huge_gfp;			/* for allocation and charge */
1162

1163
	ptl = pmd_lockptr(mm, pmd);
1164
	VM_BUG_ON_VMA(!vma->anon_vma, vma);
1165 1166 1167
	haddr = address & HPAGE_PMD_MASK;
	if (is_huge_zero_pmd(orig_pmd))
		goto alloc;
1168
	spin_lock(ptl);
1169 1170 1171 1172
	if (unlikely(!pmd_same(*pmd, orig_pmd)))
		goto out_unlock;

	page = pmd_page(orig_pmd);
1173
	VM_BUG_ON_PAGE(!PageCompound(page) || !PageHead(page), page);
1174 1175 1176 1177 1178
	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))
1179
			update_mmu_cache_pmd(vma, address, pmd);
1180 1181 1182
		ret |= VM_FAULT_WRITE;
		goto out_unlock;
	}
1183
	get_user_huge_page(page);
1184
	spin_unlock(ptl);
1185
alloc:
1186
	if (transparent_hugepage_enabled(vma) &&
1187
	    !transparent_hugepage_debug_cow()) {
1188 1189
		huge_gfp = alloc_hugepage_gfpmask(transparent_hugepage_defrag(vma), 0);
		new_page = alloc_hugepage_vma(huge_gfp, vma, haddr, HPAGE_PMD_ORDER);
1190
	} else
1191 1192 1193
		new_page = NULL;

	if (unlikely(!new_page)) {
1194
		if (!page) {
1195 1196
			split_huge_page_pmd(vma, address, pmd);
			ret |= VM_FAULT_FALLBACK;
1197 1198 1199
		} else {
			ret = do_huge_pmd_wp_page_fallback(mm, vma, address,
					pmd, orig_pmd, page, haddr);
1200
			if (ret & VM_FAULT_OOM) {
1201
				split_huge_page(page);
1202 1203
				ret |= VM_FAULT_FALLBACK;
			}
1204
			put_user_huge_page(page);
1205
		}
1206
		count_vm_event(THP_FAULT_FALLBACK);
1207 1208 1209
		goto out;
	}

1210
	if (unlikely(mem_cgroup_try_charge(new_page, mm, huge_gfp, &memcg))) {
A
Andrea Arcangeli 已提交
1211
		put_page(new_page);
1212 1213
		if (page) {
			split_huge_page(page);
1214
			put_user_huge_page(page);
1215 1216 1217
		} else
			split_huge_page_pmd(vma, address, pmd);
		ret |= VM_FAULT_FALLBACK;
1218
		count_vm_event(THP_FAULT_FALLBACK);
A
Andrea Arcangeli 已提交
1219 1220 1221
		goto out;
	}

1222 1223
	count_vm_event(THP_FAULT_ALLOC);

1224
	if (!page)
1225 1226 1227
		clear_huge_page(new_page, haddr, HPAGE_PMD_NR);
	else
		copy_user_huge_page(new_page, page, haddr, vma, HPAGE_PMD_NR);
1228 1229
	__SetPageUptodate(new_page);

1230 1231 1232 1233
	mmun_start = haddr;
	mmun_end   = haddr + HPAGE_PMD_SIZE;
	mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);

1234
	spin_lock(ptl);
1235
	if (page)
1236
		put_user_huge_page(page);
A
Andrea Arcangeli 已提交
1237
	if (unlikely(!pmd_same(*pmd, orig_pmd))) {
1238
		spin_unlock(ptl);
1239
		mem_cgroup_cancel_charge(new_page, memcg);
1240
		put_page(new_page);
1241
		goto out_mn;
A
Andrea Arcangeli 已提交
1242
	} else {
1243
		pmd_t entry;
1244 1245
		entry = mk_huge_pmd(new_page, vma->vm_page_prot);
		entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma);
1246
		pmdp_huge_clear_flush_notify(vma, haddr, pmd);
1247
		page_add_new_anon_rmap(new_page, vma, haddr);
1248 1249
		mem_cgroup_commit_charge(new_page, memcg, false);
		lru_cache_add_active_or_unevictable(new_page, vma);
1250
		set_pmd_at(mm, haddr, pmd, entry);
1251
		update_mmu_cache_pmd(vma, address, pmd);
1252
		if (!page) {
1253
			add_mm_counter(mm, MM_ANONPAGES, HPAGE_PMD_NR);
1254 1255
			put_huge_zero_page();
		} else {
1256
			VM_BUG_ON_PAGE(!PageHead(page), page);
1257 1258 1259
			page_remove_rmap(page);
			put_page(page);
		}
1260 1261
		ret |= VM_FAULT_WRITE;
	}
1262
	spin_unlock(ptl);
1263 1264
out_mn:
	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
1265 1266
out:
	return ret;
1267
out_unlock:
1268
	spin_unlock(ptl);
1269
	return ret;
1270 1271
}

1272
struct page *follow_trans_huge_pmd(struct vm_area_struct *vma,
1273 1274 1275 1276
				   unsigned long addr,
				   pmd_t *pmd,
				   unsigned int flags)
{
1277
	struct mm_struct *mm = vma->vm_mm;
1278 1279
	struct page *page = NULL;

1280
	assert_spin_locked(pmd_lockptr(mm, pmd));
1281 1282 1283 1284

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

1285 1286 1287 1288
	/* Avoid dumping huge zero page */
	if ((flags & FOLL_DUMP) && is_huge_zero_pmd(*pmd))
		return ERR_PTR(-EFAULT);

1289
	/* Full NUMA hinting faults to serialise migration in fault paths */
1290
	if ((flags & FOLL_NUMA) && pmd_protnone(*pmd))
1291 1292
		goto out;

1293
	page = pmd_page(*pmd);
1294
	VM_BUG_ON_PAGE(!PageHead(page), page);
1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305
	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));
1306 1307 1308
		if (pmdp_set_access_flags(vma, addr & HPAGE_PMD_MASK,
					  pmd, _pmd,  1))
			update_mmu_cache_pmd(vma, addr, pmd);
1309
	}
1310
	if ((flags & FOLL_POPULATE) && (vma->vm_flags & VM_LOCKED)) {
1311 1312 1313 1314 1315 1316 1317
		if (page->mapping && trylock_page(page)) {
			lru_add_drain();
			if (page->mapping)
				mlock_vma_page(page);
			unlock_page(page);
		}
	}
1318
	page += (addr & ~HPAGE_PMD_MASK) >> PAGE_SHIFT;
1319
	VM_BUG_ON_PAGE(!PageCompound(page), page);
1320
	if (flags & FOLL_GET)
1321
		get_page_foll(page);
1322 1323 1324 1325 1326

out:
	return page;
}

1327
/* NUMA hinting page fault entry point for trans huge pmds */
1328 1329
int do_huge_pmd_numa_page(struct mm_struct *mm, struct vm_area_struct *vma,
				unsigned long addr, pmd_t pmd, pmd_t *pmdp)
1330
{
1331
	spinlock_t *ptl;
1332
	struct anon_vma *anon_vma = NULL;
1333
	struct page *page;
1334
	unsigned long haddr = addr & HPAGE_PMD_MASK;
1335
	int page_nid = -1, this_nid = numa_node_id();
1336
	int target_nid, last_cpupid = -1;
1337 1338
	bool page_locked;
	bool migrated = false;
1339
	bool was_writable;
1340
	int flags = 0;
1341

1342 1343 1344
	/* A PROT_NONE fault should not end up here */
	BUG_ON(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)));

1345
	ptl = pmd_lock(mm, pmdp);
1346 1347 1348
	if (unlikely(!pmd_same(pmd, *pmdp)))
		goto out_unlock;

1349 1350 1351 1352 1353 1354
	/*
	 * 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))) {
1355
		page = pmd_page(*pmdp);
1356
		spin_unlock(ptl);
1357
		wait_on_page_locked(page);
1358 1359 1360
		goto out;
	}

1361
	page = pmd_page(pmd);
1362
	BUG_ON(is_huge_zero_page(page));
1363
	page_nid = page_to_nid(page);
1364
	last_cpupid = page_cpupid_last(page);
1365
	count_vm_numa_event(NUMA_HINT_FAULTS);
1366
	if (page_nid == this_nid) {
1367
		count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL);
1368 1369
		flags |= TNF_FAULT_LOCAL;
	}
1370

1371 1372
	/* See similar comment in do_numa_page for explanation */
	if (!(vma->vm_flags & VM_WRITE))
1373 1374
		flags |= TNF_NO_GROUP;

1375 1376 1377 1378
	/*
	 * Acquire the page lock to serialise THP migrations but avoid dropping
	 * page_table_lock if at all possible
	 */
1379 1380 1381 1382
	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 */
1383
		if (page_locked)
1384
			goto clear_pmdnuma;
1385
	}
1386

1387
	/* Migration could have started since the pmd_trans_migrating check */
1388
	if (!page_locked) {
1389
		spin_unlock(ptl);
1390
		wait_on_page_locked(page);
1391
		page_nid = -1;
1392 1393 1394
		goto out;
	}

1395 1396 1397 1398
	/*
	 * Page is misplaced. Page lock serialises migrations. Acquire anon_vma
	 * to serialises splits
	 */
1399
	get_page(page);
1400
	spin_unlock(ptl);
1401
	anon_vma = page_lock_anon_vma_read(page);
1402

P
Peter Zijlstra 已提交
1403
	/* Confirm the PMD did not change while page_table_lock was released */
1404
	spin_lock(ptl);
1405 1406 1407
	if (unlikely(!pmd_same(pmd, *pmdp))) {
		unlock_page(page);
		put_page(page);
1408
		page_nid = -1;
1409
		goto out_unlock;
1410
	}
1411

1412 1413 1414 1415 1416 1417 1418
	/* 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;
	}

1419 1420
	/*
	 * Migrate the THP to the requested node, returns with page unlocked
1421
	 * and access rights restored.
1422
	 */
1423
	spin_unlock(ptl);
1424
	migrated = migrate_misplaced_transhuge_page(mm, vma,
1425
				pmdp, pmd, addr, page, target_nid);
1426 1427
	if (migrated) {
		flags |= TNF_MIGRATED;
1428
		page_nid = target_nid;
1429 1430
	} else
		flags |= TNF_MIGRATE_FAIL;
1431

1432
	goto out;
1433
clear_pmdnuma:
1434
	BUG_ON(!PageLocked(page));
1435
	was_writable = pmd_write(pmd);
1436
	pmd = pmd_modify(pmd, vma->vm_page_prot);
1437
	pmd = pmd_mkyoung(pmd);
1438 1439
	if (was_writable)
		pmd = pmd_mkwrite(pmd);
1440 1441
	set_pmd_at(mm, haddr, pmdp, pmd);
	update_mmu_cache_pmd(vma, addr, pmdp);
1442
	unlock_page(page);
1443
out_unlock:
1444
	spin_unlock(ptl);
1445 1446 1447 1448 1449

out:
	if (anon_vma)
		page_unlock_anon_vma_read(anon_vma);

1450
	if (page_nid != -1)
1451
		task_numa_fault(last_cpupid, page_nid, HPAGE_PMD_NR, flags);
1452

1453 1454 1455
	return 0;
}

1456
int zap_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
S
Shaohua Li 已提交
1457
		 pmd_t *pmd, unsigned long addr)
1458
{
1459
	pmd_t orig_pmd;
1460
	spinlock_t *ptl;
1461

1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475
	if (__pmd_trans_huge_lock(pmd, vma, &ptl) != 1)
		return 0;
	/*
	 * For architectures like ppc64 we look at deposited pgtable
	 * when calling pmdp_huge_get_and_clear. So do the
	 * pgtable_trans_huge_withdraw after finishing pmdp related
	 * operations.
	 */
	orig_pmd = pmdp_huge_get_and_clear_full(tlb->mm, addr, pmd,
			tlb->fullmm);
	tlb_remove_pmd_tlb_entry(tlb, pmd, addr);
	if (vma_is_dax(vma)) {
		spin_unlock(ptl);
		if (is_huge_zero_pmd(orig_pmd))
1476
			put_huge_zero_page();
1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491
	} else if (is_huge_zero_pmd(orig_pmd)) {
		pte_free(tlb->mm, pgtable_trans_huge_withdraw(tlb->mm, pmd));
		atomic_long_dec(&tlb->mm->nr_ptes);
		spin_unlock(ptl);
		put_huge_zero_page();
	} else {
		struct page *page = pmd_page(orig_pmd);
		page_remove_rmap(page);
		VM_BUG_ON_PAGE(page_mapcount(page) < 0, page);
		add_mm_counter(tlb->mm, MM_ANONPAGES, -HPAGE_PMD_NR);
		VM_BUG_ON_PAGE(!PageHead(page), page);
		pte_free(tlb->mm, pgtable_trans_huge_withdraw(tlb->mm, pmd));
		atomic_long_dec(&tlb->mm->nr_ptes);
		spin_unlock(ptl);
		tlb_remove_page(tlb, page);
1492
	}
1493
	return 1;
1494 1495
}

1496 1497 1498 1499 1500
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)
{
1501
	spinlock_t *old_ptl, *new_ptl;
1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521
	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;
	}

1522 1523 1524 1525 1526
	/*
	 * 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);
1527
	if (ret == 1) {
1528 1529 1530
		new_ptl = pmd_lockptr(mm, new_pmd);
		if (new_ptl != old_ptl)
			spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
1531
		pmd = pmdp_huge_get_and_clear(mm, old_addr, old_pmd);
1532
		VM_BUG_ON(!pmd_none(*new_pmd));
1533

1534 1535
		if (pmd_move_must_withdraw(new_ptl, old_ptl)) {
			pgtable_t pgtable;
1536 1537 1538
			pgtable = pgtable_trans_huge_withdraw(mm, old_pmd);
			pgtable_trans_huge_deposit(mm, new_pmd, pgtable);
		}
1539 1540 1541
		set_pmd_at(mm, new_addr, new_pmd, pmd_mksoft_dirty(pmd));
		if (new_ptl != old_ptl)
			spin_unlock(new_ptl);
1542
		spin_unlock(old_ptl);
1543 1544 1545 1546 1547
	}
out:
	return ret;
}

1548 1549 1550 1551 1552 1553
/*
 * 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
 */
1554
int change_huge_pmd(struct vm_area_struct *vma, pmd_t *pmd,
1555
		unsigned long addr, pgprot_t newprot, int prot_numa)
1556 1557
{
	struct mm_struct *mm = vma->vm_mm;
1558
	spinlock_t *ptl;
1559 1560
	int ret = 0;

1561
	if (__pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
1562
		pmd_t entry;
1563
		bool preserve_write = prot_numa && pmd_write(*pmd);
1564
		ret = 1;
1565 1566 1567 1568 1569 1570 1571 1572

		/*
		 * 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);
1573
			return ret;
1574 1575
		}

1576
		if (!prot_numa || !pmd_protnone(*pmd)) {
1577
			entry = pmdp_huge_get_and_clear_notify(mm, addr, pmd);
1578
			entry = pmd_modify(entry, newprot);
1579 1580
			if (preserve_write)
				entry = pmd_mkwrite(entry);
1581 1582
			ret = HPAGE_PMD_NR;
			set_pmd_at(mm, addr, pmd, entry);
1583
			BUG_ON(!preserve_write && pmd_write(entry));
1584
		}
1585
		spin_unlock(ptl);
1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597
	}

	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.
 */
1598 1599
int __pmd_trans_huge_lock(pmd_t *pmd, struct vm_area_struct *vma,
		spinlock_t **ptl)
1600
{
1601
	*ptl = pmd_lock(vma->vm_mm, pmd);
1602 1603
	if (likely(pmd_trans_huge(*pmd))) {
		if (unlikely(pmd_trans_splitting(*pmd))) {
1604
			spin_unlock(*ptl);
1605
			wait_split_huge_page(vma->anon_vma, pmd);
1606
			return -1;
1607
		} else {
1608 1609 1610
			/* Thp mapped by 'pmd' is stable, so we can
			 * handle it as it is. */
			return 1;
1611
		}
1612
	}
1613
	spin_unlock(*ptl);
1614
	return 0;
1615 1616
}

1617 1618 1619 1620 1621 1622 1623 1624
/*
 * 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.
 */
1625 1626 1627
pmd_t *page_check_address_pmd(struct page *page,
			      struct mm_struct *mm,
			      unsigned long address,
1628 1629
			      enum page_check_address_pmd_flag flag,
			      spinlock_t **ptl)
1630
{
1631 1632
	pgd_t *pgd;
	pud_t *pud;
1633
	pmd_t *pmd;
1634 1635

	if (address & ~HPAGE_PMD_MASK)
1636
		return NULL;
1637

1638 1639
	pgd = pgd_offset(mm, address);
	if (!pgd_present(*pgd))
1640
		return NULL;
1641 1642 1643 1644 1645
	pud = pud_offset(pgd, address);
	if (!pud_present(*pud))
		return NULL;
	pmd = pmd_offset(pud, address);

1646
	*ptl = pmd_lock(mm, pmd);
1647
	if (!pmd_present(*pmd))
1648
		goto unlock;
1649
	if (pmd_page(*pmd) != page)
1650
		goto unlock;
1651 1652 1653 1654 1655 1656 1657 1658 1659
	/*
	 * 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))
1660
		goto unlock;
1661 1662 1663
	if (pmd_trans_huge(*pmd)) {
		VM_BUG_ON(flag == PAGE_CHECK_ADDRESS_PMD_SPLITTING_FLAG &&
			  !pmd_trans_splitting(*pmd));
1664
		return pmd;
1665
	}
1666 1667 1668
unlock:
	spin_unlock(*ptl);
	return NULL;
1669 1670 1671 1672 1673 1674 1675
}

static int __split_huge_page_splitting(struct page *page,
				       struct vm_area_struct *vma,
				       unsigned long address)
{
	struct mm_struct *mm = vma->vm_mm;
1676
	spinlock_t *ptl;
1677 1678
	pmd_t *pmd;
	int ret = 0;
1679 1680 1681
	/* For mmu_notifiers */
	const unsigned long mmun_start = address;
	const unsigned long mmun_end   = address + HPAGE_PMD_SIZE;
1682

1683
	mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
1684
	pmd = page_check_address_pmd(page, mm, address,
1685
			PAGE_CHECK_ADDRESS_PMD_NOTSPLITTING_FLAG, &ptl);
1686 1687 1688 1689 1690
	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
1691
		 * and it won't wait on the anon_vma->root->rwsem to
1692 1693
		 * serialize against split_huge_page*.
		 */
1694
		pmdp_splitting_flush(vma, address, pmd);
1695

1696
		ret = 1;
1697
		spin_unlock(ptl);
1698
	}
1699
	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
1700 1701 1702 1703

	return ret;
}

1704 1705
static void __split_huge_page_refcount(struct page *page,
				       struct list_head *list)
1706 1707 1708
{
	int i;
	struct zone *zone = page_zone(page);
1709
	struct lruvec *lruvec;
1710
	int tail_count = 0;
1711 1712 1713

	/* prevent PageLRU to go away from under us, and freeze lru stats */
	spin_lock_irq(&zone->lru_lock);
1714 1715
	lruvec = mem_cgroup_page_lruvec(page, zone);

1716
	compound_lock(page);
1717 1718
	/* complete memcg works before add pages to LRU */
	mem_cgroup_split_huge_fixup(page);
1719

1720
	for (i = HPAGE_PMD_NR - 1; i >= 1; i--) {
1721 1722
		struct page *page_tail = page + i;

1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743
		/* 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);
1744 1745

		/* after clearing PageTail the gup refcount can be released */
1746
		smp_mb__after_atomic();
1747

1748
		page_tail->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1749 1750 1751 1752
		page_tail->flags |= (page->flags &
				     ((1L << PG_referenced) |
				      (1L << PG_swapbacked) |
				      (1L << PG_mlocked) |
1753 1754 1755
				      (1L << PG_uptodate) |
				      (1L << PG_active) |
				      (1L << PG_unevictable)));
1756 1757
		page_tail->flags |= (1L << PG_dirty);

1758
		/* clear PageTail before overwriting first_page */
1759 1760
		smp_wmb();

1761 1762 1763 1764 1765
		if (page_is_young(page))
			set_page_young(page_tail);
		if (page_is_idle(page))
			set_page_idle(page_tail);

1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784
		/*
		 * __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;

1785
		page_tail->index = page->index + i;
1786
		page_cpupid_xchg_last(page_tail, page_cpupid_last(page));
1787 1788 1789 1790 1791 1792

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

1793
		lru_add_page_tail(page, page_tail, lruvec, list);
1794
	}
1795 1796
	atomic_sub(tail_count, &page->_count);
	BUG_ON(atomic_read(&page->_count) <= 0);
1797

1798
	__mod_zone_page_state(zone, NR_ANON_TRANSPARENT_HUGEPAGES, -1);
1799

1800 1801 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
	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;
1829
	spinlock_t *ptl;
1830 1831 1832 1833 1834 1835
	pmd_t *pmd, _pmd;
	int ret = 0, i;
	pgtable_t pgtable;
	unsigned long haddr;

	pmd = page_check_address_pmd(page, mm, address,
1836
			PAGE_CHECK_ADDRESS_PMD_SPLITTING_FLAG, &ptl);
1837
	if (pmd) {
1838
		pgtable = pgtable_trans_huge_withdraw(mm, pmd);
1839
		pmd_populate(mm, &_pmd, pgtable);
1840 1841
		if (pmd_write(*pmd))
			BUG_ON(page_mapcount(page) != 1);
1842

1843 1844
		haddr = address;
		for (i = 0; i < HPAGE_PMD_NR; i++, haddr += PAGE_SIZE) {
1845 1846
			pte_t *pte, entry;
			BUG_ON(PageCompound(page+i));
1847
			/*
1848 1849 1850
			 * Note that NUMA hinting access restrictions are not
			 * transferred to avoid any possibility of altering
			 * permissions across VMAs.
1851
			 */
1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890
			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
1891
		pmdp_invalidate(vma, address, pmd);
1892 1893
		pmd_populate(mm, pmd, pgtable);
		ret = 1;
1894
		spin_unlock(ptl);
1895 1896 1897 1898 1899
	}

	return ret;
}

1900
/* must be called with anon_vma->root->rwsem held */
1901
static void __split_huge_page(struct page *page,
1902 1903
			      struct anon_vma *anon_vma,
			      struct list_head *list)
1904 1905
{
	int mapcount, mapcount2;
1906
	pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1907 1908 1909 1910 1911 1912
	struct anon_vma_chain *avc;

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

	mapcount = 0;
1913
	anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
1914 1915 1916 1917 1918
		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);
	}
1919 1920 1921 1922 1923 1924 1925 1926 1927 1928
	/*
	 * 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.
	 */
1929
	if (mapcount != page_mapcount(page)) {
1930 1931
		pr_err("mapcount %d page_mapcount %d\n",
			mapcount, page_mapcount(page));
1932 1933
		BUG();
	}
1934

1935
	__split_huge_page_refcount(page, list);
1936 1937

	mapcount2 = 0;
1938
	anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
1939 1940 1941 1942 1943
		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);
	}
1944
	if (mapcount != mapcount2) {
1945 1946
		pr_err("mapcount %d mapcount2 %d page_mapcount %d\n",
			mapcount, mapcount2, page_mapcount(page));
1947 1948
		BUG();
	}
1949 1950
}

1951 1952 1953 1954 1955 1956 1957 1958
/*
 * 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)
1959 1960 1961 1962
{
	struct anon_vma *anon_vma;
	int ret = 1;

1963
	BUG_ON(is_huge_zero_page(page));
1964
	BUG_ON(!PageAnon(page));
1965 1966 1967 1968 1969 1970 1971 1972 1973

	/*
	 * 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);
1974 1975
	if (!anon_vma)
		goto out;
1976 1977
	anon_vma_lock_write(anon_vma);

1978 1979 1980 1981 1982
	ret = 0;
	if (!PageCompound(page))
		goto out_unlock;

	BUG_ON(!PageSwapBacked(page));
1983
	__split_huge_page(page, anon_vma, list);
1984
	count_vm_event(THP_SPLIT);
1985 1986 1987

	BUG_ON(PageCompound(page));
out_unlock:
1988
	anon_vma_unlock_write(anon_vma);
1989
	put_anon_vma(anon_vma);
1990 1991 1992 1993
out:
	return ret;
}

1994
#define VM_NO_THP (VM_SPECIAL | VM_HUGETLB | VM_SHARED | VM_MAYSHARE)
1995

1996 1997
int hugepage_madvise(struct vm_area_struct *vma,
		     unsigned long *vm_flags, int advice)
A
Andrea Arcangeli 已提交
1998
{
A
Andrea Arcangeli 已提交
1999 2000
	switch (advice) {
	case MADV_HUGEPAGE:
2001 2002 2003 2004 2005 2006 2007 2008 2009
#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
2010 2011 2012
		/*
		 * Be somewhat over-protective like KSM for now!
		 */
2013
		if (*vm_flags & (VM_HUGEPAGE | VM_NO_THP))
A
Andrea Arcangeli 已提交
2014 2015 2016
			return -EINVAL;
		*vm_flags &= ~VM_NOHUGEPAGE;
		*vm_flags |= VM_HUGEPAGE;
2017 2018 2019 2020 2021
		/*
		 * If the vma become good for khugepaged to scan,
		 * register it here without waiting a page fault that
		 * may not happen any time soon.
		 */
2022
		if (unlikely(khugepaged_enter_vma_merge(vma, *vm_flags)))
2023
			return -ENOMEM;
A
Andrea Arcangeli 已提交
2024 2025 2026 2027 2028
		break;
	case MADV_NOHUGEPAGE:
		/*
		 * Be somewhat over-protective like KSM for now!
		 */
2029
		if (*vm_flags & (VM_NOHUGEPAGE | VM_NO_THP))
A
Andrea Arcangeli 已提交
2030 2031 2032
			return -EINVAL;
		*vm_flags &= ~VM_HUGEPAGE;
		*vm_flags |= VM_NOHUGEPAGE;
2033 2034 2035 2036 2037
		/*
		 * 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
2038 2039
		break;
	}
A
Andrea Arcangeli 已提交
2040 2041 2042 2043

	return 0;
}

A
Andrea Arcangeli 已提交
2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054
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;
}

2055 2056 2057 2058 2059
static void __init khugepaged_slab_exit(void)
{
	kmem_cache_destroy(mm_slot_cache);
}

A
Andrea Arcangeli 已提交
2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075
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;

2076
	hash_for_each_possible(mm_slots_hash, mm_slot, hash, (unsigned long)mm)
A
Andrea Arcangeli 已提交
2077 2078
		if (mm == mm_slot->mm)
			return mm_slot;
2079

A
Andrea Arcangeli 已提交
2080 2081 2082 2083 2084 2085 2086
	return NULL;
}

static void insert_to_mm_slots_hash(struct mm_struct *mm,
				    struct mm_slot *mm_slot)
{
	mm_slot->mm = mm;
2087
	hash_add(mm_slots_hash, &mm_slot->hash, (long)mm);
A
Andrea Arcangeli 已提交
2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104
}

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
2105
	VM_BUG_ON_MM(khugepaged_test_exit(mm), mm);
A
Andrea Arcangeli 已提交
2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127
	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;
}

2128 2129
int khugepaged_enter_vma_merge(struct vm_area_struct *vma,
			       unsigned long vm_flags)
A
Andrea Arcangeli 已提交
2130 2131 2132 2133 2134 2135 2136 2137
{
	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;
2138
	if (vma->vm_ops)
A
Andrea Arcangeli 已提交
2139 2140
		/* khugepaged not yet working on file or special mappings */
		return 0;
2141
	VM_BUG_ON_VMA(vm_flags & VM_NO_THP, vma);
A
Andrea Arcangeli 已提交
2142 2143 2144
	hstart = (vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK;
	hend = vma->vm_end & HPAGE_PMD_MASK;
	if (hstart < hend)
2145
		return khugepaged_enter(vma, vm_flags);
A
Andrea Arcangeli 已提交
2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156
	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) {
2157
		hash_del(&mm_slot->hash);
A
Andrea Arcangeli 已提交
2158 2159 2160
		list_del(&mm_slot->mm_node);
		free = 1;
	}
2161
	spin_unlock(&khugepaged_mm_lock);
A
Andrea Arcangeli 已提交
2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177

	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);
2178
	}
A
Andrea Arcangeli 已提交
2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192
}

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;
2193
		if (!pte_none(pteval) && !is_zero_pfn(pte_pfn(pteval)))
A
Andrea Arcangeli 已提交
2194 2195 2196 2197 2198 2199 2200 2201 2202 2203
			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;
2204
	int none_or_zero = 0;
2205
	bool referenced = false, writable = false;
A
Andrea Arcangeli 已提交
2206 2207 2208
	for (_pte = pte; _pte < pte+HPAGE_PMD_NR;
	     _pte++, address += PAGE_SIZE) {
		pte_t pteval = *_pte;
2209
		if (pte_none(pteval) || is_zero_pfn(pte_pfn(pteval))) {
2210 2211
			if (!userfaultfd_armed(vma) &&
			    ++none_or_zero <= khugepaged_max_ptes_none)
A
Andrea Arcangeli 已提交
2212
				continue;
2213
			else
A
Andrea Arcangeli 已提交
2214 2215
				goto out;
		}
2216
		if (!pte_present(pteval))
A
Andrea Arcangeli 已提交
2217 2218
			goto out;
		page = vm_normal_page(vma, address, pteval);
2219
		if (unlikely(!page))
A
Andrea Arcangeli 已提交
2220
			goto out;
2221

2222 2223 2224
		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
2225 2226 2227 2228 2229 2230 2231

		/*
		 * 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.
		 */
2232
		if (!trylock_page(page))
A
Andrea Arcangeli 已提交
2233
			goto out;
2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256

		/*
		 * 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
2257 2258 2259 2260 2261 2262 2263 2264 2265 2266
		/*
		 * 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);
2267 2268
		VM_BUG_ON_PAGE(!PageLocked(page), page);
		VM_BUG_ON_PAGE(PageLRU(page), page);
A
Andrea Arcangeli 已提交
2269 2270

		/* If there is no mapped pte young don't collapse the page */
2271 2272
		if (pte_young(pteval) ||
		    page_is_young(page) || PageReferenced(page) ||
A
Andrea Arcangeli 已提交
2273
		    mmu_notifier_test_young(vma->vm_mm, address))
2274
			referenced = true;
A
Andrea Arcangeli 已提交
2275
	}
2276
	if (likely(referenced && writable))
2277
		return 1;
A
Andrea Arcangeli 已提交
2278
out:
2279 2280
	release_pte_pages(pte, _pte);
	return 0;
A
Andrea Arcangeli 已提交
2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292
}

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;

2293
		if (pte_none(pteval) || is_zero_pfn(pte_pfn(pteval))) {
A
Andrea Arcangeli 已提交
2294 2295
			clear_user_highpage(page, address);
			add_mm_counter(vma->vm_mm, MM_ANONPAGES, 1);
2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307
			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
2308 2309 2310
		} else {
			src_page = pte_page(pteval);
			copy_user_highpage(page, src_page, address, vma);
2311
			VM_BUG_ON_PAGE(page_mapcount(src_page) != 1, src_page);
A
Andrea Arcangeli 已提交
2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333
			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++;
	}
}

2334
static void khugepaged_alloc_sleep(void)
A
Andrea Arcangeli 已提交
2335
{
2336 2337 2338 2339 2340 2341
	DEFINE_WAIT(wait);

	add_wait_queue(&khugepaged_wait, &wait);
	freezable_schedule_timeout_interruptible(
		msecs_to_jiffies(khugepaged_alloc_sleep_millisecs));
	remove_wait_queue(&khugepaged_wait, &wait);
2342
}
A
Andrea Arcangeli 已提交
2343

2344 2345
static int khugepaged_node_load[MAX_NUMNODES];

2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369
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;
}

2370
#ifdef CONFIG_NUMA
2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395
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;
}

2396 2397 2398 2399 2400 2401 2402
static bool khugepaged_prealloc_page(struct page **hpage, bool *wait)
{
	if (IS_ERR(*hpage)) {
		if (!*wait)
			return false;

		*wait = false;
2403
		*hpage = NULL;
2404 2405 2406 2407 2408 2409 2410 2411 2412
		khugepaged_alloc_sleep();
	} else if (*hpage) {
		put_page(*hpage);
		*hpage = NULL;
	}

	return true;
}

2413 2414
static struct page *
khugepaged_alloc_page(struct page **hpage, gfp_t gfp, struct mm_struct *mm,
2415 2416 2417
		       struct vm_area_struct *vma, unsigned long address,
		       int node)
{
2418
	VM_BUG_ON_PAGE(*hpage, *hpage);
2419

2420
	/*
2421 2422 2423 2424
	 * 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.
2425
	 */
2426 2427
	up_read(&mm->mmap_sem);

2428
	*hpage = __alloc_pages_node(node, gfp, HPAGE_PMD_ORDER);
2429
	if (unlikely(!*hpage)) {
2430
		count_vm_event(THP_COLLAPSE_ALLOC_FAILED);
2431
		*hpage = ERR_PTR(-ENOMEM);
2432
		return NULL;
2433
	}
2434

2435
	count_vm_event(THP_COLLAPSE_ALLOC);
2436 2437 2438
	return *hpage;
}
#else
2439 2440 2441 2442 2443
static int khugepaged_find_target_node(void)
{
	return 0;
}

2444 2445 2446 2447 2448 2449
static inline struct page *alloc_hugepage(int defrag)
{
	return alloc_pages(alloc_hugepage_gfpmask(defrag, 0),
			   HPAGE_PMD_ORDER);
}

2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480
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;
}

2481 2482
static struct page *
khugepaged_alloc_page(struct page **hpage, gfp_t gfp, struct mm_struct *mm,
2483 2484 2485 2486 2487
		       struct vm_area_struct *vma, unsigned long address,
		       int node)
{
	up_read(&mm->mmap_sem);
	VM_BUG_ON(!*hpage);
2488

2489 2490
	return  *hpage;
}
2491 2492
#endif

B
Bob Liu 已提交
2493 2494 2495 2496 2497 2498 2499 2500 2501 2502
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;
2503
	VM_BUG_ON_VMA(vma->vm_flags & VM_NO_THP, vma);
B
Bob Liu 已提交
2504 2505 2506
	return true;
}

2507 2508 2509 2510 2511 2512 2513 2514 2515 2516
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;
2517
	spinlock_t *pmd_ptl, *pte_ptl;
2518 2519
	int isolated;
	unsigned long hstart, hend;
2520
	struct mem_cgroup *memcg;
2521 2522
	unsigned long mmun_start;	/* For mmu_notifiers */
	unsigned long mmun_end;		/* For mmu_notifiers */
2523
	gfp_t gfp;
2524 2525 2526

	VM_BUG_ON(address & ~HPAGE_PMD_MASK);

2527 2528 2529 2530
	/* Only allocate from the target node */
	gfp = alloc_hugepage_gfpmask(khugepaged_defrag(), __GFP_OTHER_NODE) |
		__GFP_THISNODE;

2531
	/* release the mmap_sem read lock. */
2532
	new_page = khugepaged_alloc_page(hpage, gfp, mm, vma, address, node);
2533 2534 2535
	if (!new_page)
		return;

2536
	if (unlikely(mem_cgroup_try_charge(new_page, mm,
2537
					   gfp, &memcg)))
2538
		return;
A
Andrea Arcangeli 已提交
2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549

	/*
	 * 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);
2550 2551
	if (!vma)
		goto out;
A
Andrea Arcangeli 已提交
2552 2553 2554 2555
	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
2556
	if (!hugepage_vma_check(vma))
2557
		goto out;
B
Bob Liu 已提交
2558 2559
	pmd = mm_find_pmd(mm, address);
	if (!pmd)
A
Andrea Arcangeli 已提交
2560 2561
		goto out;

2562
	anon_vma_lock_write(vma->anon_vma);
A
Andrea Arcangeli 已提交
2563 2564

	pte = pte_offset_map(pmd, address);
2565
	pte_ptl = pte_lockptr(mm, pmd);
A
Andrea Arcangeli 已提交
2566

2567 2568 2569
	mmun_start = address;
	mmun_end   = address + HPAGE_PMD_SIZE;
	mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
2570
	pmd_ptl = pmd_lock(mm, pmd); /* probably unnecessary */
A
Andrea Arcangeli 已提交
2571 2572 2573 2574 2575 2576
	/*
	 * 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.
	 */
2577
	_pmd = pmdp_collapse_flush(vma, address, pmd);
2578
	spin_unlock(pmd_ptl);
2579
	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
A
Andrea Arcangeli 已提交
2580

2581
	spin_lock(pte_ptl);
A
Andrea Arcangeli 已提交
2582
	isolated = __collapse_huge_page_isolate(vma, address, pte);
2583
	spin_unlock(pte_ptl);
A
Andrea Arcangeli 已提交
2584 2585

	if (unlikely(!isolated)) {
2586
		pte_unmap(pte);
2587
		spin_lock(pmd_ptl);
A
Andrea Arcangeli 已提交
2588
		BUG_ON(!pmd_none(*pmd));
2589 2590 2591 2592 2593 2594
		/*
		 * 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));
2595
		spin_unlock(pmd_ptl);
2596
		anon_vma_unlock_write(vma->anon_vma);
2597
		goto out;
A
Andrea Arcangeli 已提交
2598 2599 2600 2601 2602 2603
	}

	/*
	 * All pages are isolated and locked so anon_vma rmap
	 * can't run anymore.
	 */
2604
	anon_vma_unlock_write(vma->anon_vma);
A
Andrea Arcangeli 已提交
2605

2606
	__collapse_huge_page_copy(pte, new_page, vma, address, pte_ptl);
2607
	pte_unmap(pte);
A
Andrea Arcangeli 已提交
2608 2609 2610
	__SetPageUptodate(new_page);
	pgtable = pmd_pgtable(_pmd);

2611 2612
	_pmd = mk_huge_pmd(new_page, vma->vm_page_prot);
	_pmd = maybe_pmd_mkwrite(pmd_mkdirty(_pmd), vma);
A
Andrea Arcangeli 已提交
2613 2614 2615 2616 2617 2618 2619 2620

	/*
	 * 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();

2621
	spin_lock(pmd_ptl);
A
Andrea Arcangeli 已提交
2622 2623
	BUG_ON(!pmd_none(*pmd));
	page_add_new_anon_rmap(new_page, vma, address);
2624 2625
	mem_cgroup_commit_charge(new_page, memcg, false);
	lru_cache_add_active_or_unevictable(new_page, vma);
2626
	pgtable_trans_huge_deposit(mm, pmd, pgtable);
A
Andrea Arcangeli 已提交
2627
	set_pmd_at(mm, address, pmd, _pmd);
2628
	update_mmu_cache_pmd(vma, address, pmd);
2629
	spin_unlock(pmd_ptl);
A
Andrea Arcangeli 已提交
2630 2631

	*hpage = NULL;
2632

A
Andrea Arcangeli 已提交
2633
	khugepaged_pages_collapsed++;
2634
out_up_write:
A
Andrea Arcangeli 已提交
2635
	up_write(&mm->mmap_sem);
2636 2637
	return;

2638
out:
2639
	mem_cgroup_cancel_charge(new_page, memcg);
2640
	goto out_up_write;
A
Andrea Arcangeli 已提交
2641 2642 2643 2644 2645 2646 2647 2648 2649
}

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;
2650
	int ret = 0, none_or_zero = 0;
A
Andrea Arcangeli 已提交
2651 2652 2653
	struct page *page;
	unsigned long _address;
	spinlock_t *ptl;
D
David Rientjes 已提交
2654
	int node = NUMA_NO_NODE;
2655
	bool writable = false, referenced = false;
A
Andrea Arcangeli 已提交
2656 2657 2658

	VM_BUG_ON(address & ~HPAGE_PMD_MASK);

B
Bob Liu 已提交
2659 2660
	pmd = mm_find_pmd(mm, address);
	if (!pmd)
A
Andrea Arcangeli 已提交
2661 2662
		goto out;

2663
	memset(khugepaged_node_load, 0, sizeof(khugepaged_node_load));
A
Andrea Arcangeli 已提交
2664 2665 2666 2667
	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;
2668
		if (pte_none(pteval) || is_zero_pfn(pte_pfn(pteval))) {
2669 2670
			if (!userfaultfd_armed(vma) &&
			    ++none_or_zero <= khugepaged_max_ptes_none)
A
Andrea Arcangeli 已提交
2671 2672 2673 2674
				continue;
			else
				goto out_unmap;
		}
2675
		if (!pte_present(pteval))
A
Andrea Arcangeli 已提交
2676
			goto out_unmap;
2677 2678 2679
		if (pte_write(pteval))
			writable = true;

A
Andrea Arcangeli 已提交
2680 2681 2682
		page = vm_normal_page(vma, _address, pteval);
		if (unlikely(!page))
			goto out_unmap;
2683
		/*
2684 2685 2686 2687
		 * 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.
2688
		 */
2689
		node = page_to_nid(page);
2690 2691
		if (khugepaged_scan_abort(node))
			goto out_unmap;
2692
		khugepaged_node_load[node]++;
2693
		VM_BUG_ON_PAGE(PageCompound(page), page);
A
Andrea Arcangeli 已提交
2694 2695
		if (!PageLRU(page) || PageLocked(page) || !PageAnon(page))
			goto out_unmap;
2696 2697 2698 2699 2700 2701
		/*
		 * 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
2702
			goto out_unmap;
2703 2704
		if (pte_young(pteval) ||
		    page_is_young(page) || PageReferenced(page) ||
A
Andrea Arcangeli 已提交
2705
		    mmu_notifier_test_young(vma->vm_mm, address))
2706
			referenced = true;
A
Andrea Arcangeli 已提交
2707
	}
2708
	if (referenced && writable)
A
Andrea Arcangeli 已提交
2709 2710 2711
		ret = 1;
out_unmap:
	pte_unmap_unlock(pte, ptl);
2712 2713
	if (ret) {
		node = khugepaged_find_target_node();
2714
		/* collapse_huge_page will return with the mmap_sem released */
2715
		collapse_huge_page(mm, address, hpage, vma, node);
2716
	}
A
Andrea Arcangeli 已提交
2717 2718 2719 2720 2721 2722 2723 2724
out:
	return ret;
}

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

2725
	VM_BUG_ON(NR_CPUS != 1 && !spin_is_locked(&khugepaged_mm_lock));
A
Andrea Arcangeli 已提交
2726 2727 2728

	if (khugepaged_test_exit(mm)) {
		/* free mm_slot */
2729
		hash_del(&mm_slot->hash);
A
Andrea Arcangeli 已提交
2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745
		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)
2746 2747
	__releases(&khugepaged_mm_lock)
	__acquires(&khugepaged_mm_lock)
A
Andrea Arcangeli 已提交
2748 2749 2750 2751 2752 2753 2754
{
	struct mm_slot *mm_slot;
	struct mm_struct *mm;
	struct vm_area_struct *vma;
	int progress = 0;

	VM_BUG_ON(!pages);
2755
	VM_BUG_ON(NR_CPUS != 1 && !spin_is_locked(&khugepaged_mm_lock));
A
Andrea Arcangeli 已提交
2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782

	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
2783 2784
		if (!hugepage_vma_check(vma)) {
skip:
A
Andrea Arcangeli 已提交
2785 2786 2787 2788 2789
			progress++;
			continue;
		}
		hstart = (vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK;
		hend = vma->vm_end & HPAGE_PMD_MASK;
2790 2791 2792 2793
		if (hstart >= hend)
			goto skip;
		if (khugepaged_scan.address > hend)
			goto skip;
A
Andrea Arcangeli 已提交
2794 2795
		if (khugepaged_scan.address < hstart)
			khugepaged_scan.address = hstart;
2796
		VM_BUG_ON(khugepaged_scan.address & ~HPAGE_PMD_MASK);
A
Andrea Arcangeli 已提交
2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824

		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);
2825
	VM_BUG_ON(khugepaged_scan.mm_slot != mm_slot);
A
Andrea Arcangeli 已提交
2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860
	/*
	 * 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) ||
2861
		kthread_should_stop();
A
Andrea Arcangeli 已提交
2862 2863
}

2864
static void khugepaged_do_scan(void)
A
Andrea Arcangeli 已提交
2865
{
2866
	struct page *hpage = NULL;
A
Andrea Arcangeli 已提交
2867 2868
	unsigned int progress = 0, pass_through_head = 0;
	unsigned int pages = khugepaged_pages_to_scan;
2869
	bool wait = true;
A
Andrea Arcangeli 已提交
2870 2871 2872 2873

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

	while (progress < pages) {
2874
		if (!khugepaged_prealloc_page(&hpage, &wait))
2875
			break;
2876

2877
		cond_resched();
A
Andrea Arcangeli 已提交
2878

2879
		if (unlikely(kthread_should_stop() || try_to_freeze()))
2880 2881
			break;

A
Andrea Arcangeli 已提交
2882 2883 2884 2885 2886 2887
		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,
2888
							    &hpage);
A
Andrea Arcangeli 已提交
2889 2890 2891 2892 2893
		else
			progress = pages;
		spin_unlock(&khugepaged_mm_lock);
	}

2894 2895
	if (!IS_ERR_OR_NULL(hpage))
		put_page(hpage);
2896 2897
}

2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913
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
2914 2915 2916 2917
static int khugepaged(void *none)
{
	struct mm_slot *mm_slot;

2918
	set_freezable();
2919
	set_user_nice(current, MAX_NICE);
A
Andrea Arcangeli 已提交
2920

X
Xiao Guangrong 已提交
2921 2922 2923 2924
	while (!kthread_should_stop()) {
		khugepaged_do_scan();
		khugepaged_wait_work();
	}
A
Andrea Arcangeli 已提交
2925 2926 2927 2928 2929 2930 2931 2932 2933 2934

	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;
}

2935 2936 2937 2938 2939 2940 2941 2942
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;

2943
	pmdp_huge_clear_flush_notify(vma, haddr, pmd);
2944 2945
	/* leave pmd empty until pte is filled */

2946
	pgtable = pgtable_trans_huge_withdraw(mm, pmd);
2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959
	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);
2960
	put_huge_zero_page();
2961 2962
}

2963 2964
void __split_huge_page_pmd(struct vm_area_struct *vma, unsigned long address,
		pmd_t *pmd)
2965
{
2966
	spinlock_t *ptl;
M
Matthew Wilcox 已提交
2967
	struct page *page = NULL;
2968
	struct mm_struct *mm = vma->vm_mm;
2969 2970 2971
	unsigned long haddr = address & HPAGE_PMD_MASK;
	unsigned long mmun_start;	/* For mmu_notifiers */
	unsigned long mmun_end;		/* For mmu_notifiers */
2972 2973

	BUG_ON(vma->vm_start > haddr || vma->vm_end < haddr + HPAGE_PMD_SIZE);
2974

2975 2976
	mmun_start = haddr;
	mmun_end   = haddr + HPAGE_PMD_SIZE;
2977
again:
2978
	mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
2979
	ptl = pmd_lock(mm, pmd);
M
Matthew Wilcox 已提交
2980 2981 2982
	if (unlikely(!pmd_trans_huge(*pmd)))
		goto unlock;
	if (vma_is_dax(vma)) {
2983 2984 2985
		pmd_t _pmd = pmdp_huge_clear_flush_notify(vma, haddr, pmd);
		if (is_huge_zero_pmd(_pmd))
			put_huge_zero_page();
M
Matthew Wilcox 已提交
2986
	} else if (is_huge_zero_pmd(*pmd)) {
2987
		__split_huge_zero_page_pmd(vma, haddr, pmd);
M
Matthew Wilcox 已提交
2988 2989 2990 2991
	} else {
		page = pmd_page(*pmd);
		VM_BUG_ON_PAGE(!page_count(page), page);
		get_page(page);
2992
	}
M
Matthew Wilcox 已提交
2993
 unlock:
2994
	spin_unlock(ptl);
2995
	mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
2996

M
Matthew Wilcox 已提交
2997 2998
	if (!page)
		return;
2999

M
Matthew Wilcox 已提交
3000
	split_huge_page(page);
3001
	put_page(page);
3002 3003 3004 3005 3006 3007 3008 3009

	/*
	 * 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;
3010
}
3011

3012 3013 3014 3015 3016 3017 3018 3019 3020 3021
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);
}

3022 3023 3024
static void split_huge_page_address(struct mm_struct *mm,
				    unsigned long address)
{
3025 3026
	pgd_t *pgd;
	pud_t *pud;
3027 3028 3029 3030
	pmd_t *pmd;

	VM_BUG_ON(!(address & ~HPAGE_PMD_MASK));

3031 3032 3033 3034 3035 3036 3037 3038 3039 3040
	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))
3041 3042 3043 3044 3045
		return;
	/*
	 * Caller holds the mmap_sem write mode, so a huge pmd cannot
	 * materialize from under us.
	 */
3046
	split_huge_page_pmd_mm(mm, address, pmd);
3047 3048
}

3049
void vma_adjust_trans_huge(struct vm_area_struct *vma,
3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088
			     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);
	}
}