hugetlb.c 79.7 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6 7 8
/*
 * Generic hugetlb support.
 * (C) William Irwin, April 2004
 */
#include <linux/list.h>
#include <linux/init.h>
#include <linux/module.h>
#include <linux/mm.h>
9
#include <linux/seq_file.h>
L
Linus Torvalds 已提交
10 11
#include <linux/sysctl.h>
#include <linux/highmem.h>
A
Andrea Arcangeli 已提交
12
#include <linux/mmu_notifier.h>
L
Linus Torvalds 已提交
13
#include <linux/nodemask.h>
D
David Gibson 已提交
14
#include <linux/pagemap.h>
15
#include <linux/mempolicy.h>
16
#include <linux/cpuset.h>
17
#include <linux/mutex.h>
18
#include <linux/bootmem.h>
19
#include <linux/sysfs.h>
20
#include <linux/slab.h>
21
#include <linux/rmap.h>
22 23
#include <linux/swap.h>
#include <linux/swapops.h>
24

D
David Gibson 已提交
25 26
#include <asm/page.h>
#include <asm/pgtable.h>
27
#include <asm/tlb.h>
D
David Gibson 已提交
28

29
#include <linux/io.h>
D
David Gibson 已提交
30
#include <linux/hugetlb.h>
31
#include <linux/hugetlb_cgroup.h>
32
#include <linux/node.h>
33
#include "internal.h"
L
Linus Torvalds 已提交
34 35

const unsigned long hugetlb_zero = 0, hugetlb_infinity = ~0UL;
36 37
static gfp_t htlb_alloc_mask = GFP_HIGHUSER;
unsigned long hugepages_treat_as_movable;
38

39
int hugetlb_max_hstate __read_mostly;
40 41 42
unsigned int default_hstate_idx;
struct hstate hstates[HUGE_MAX_HSTATE];

43 44
__initdata LIST_HEAD(huge_boot_pages);

45 46 47
/* for command line parsing */
static struct hstate * __initdata parsed_hstate;
static unsigned long __initdata default_hstate_max_huge_pages;
48
static unsigned long __initdata default_hstate_size;
49

50 51 52
/*
 * Protects updates to hugepage_freelists, nr_huge_pages, and free_huge_pages
 */
53
DEFINE_SPINLOCK(hugetlb_lock);
54

55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132
static inline void unlock_or_release_subpool(struct hugepage_subpool *spool)
{
	bool free = (spool->count == 0) && (spool->used_hpages == 0);

	spin_unlock(&spool->lock);

	/* If no pages are used, and no other handles to the subpool
	 * remain, free the subpool the subpool remain */
	if (free)
		kfree(spool);
}

struct hugepage_subpool *hugepage_new_subpool(long nr_blocks)
{
	struct hugepage_subpool *spool;

	spool = kmalloc(sizeof(*spool), GFP_KERNEL);
	if (!spool)
		return NULL;

	spin_lock_init(&spool->lock);
	spool->count = 1;
	spool->max_hpages = nr_blocks;
	spool->used_hpages = 0;

	return spool;
}

void hugepage_put_subpool(struct hugepage_subpool *spool)
{
	spin_lock(&spool->lock);
	BUG_ON(!spool->count);
	spool->count--;
	unlock_or_release_subpool(spool);
}

static int hugepage_subpool_get_pages(struct hugepage_subpool *spool,
				      long delta)
{
	int ret = 0;

	if (!spool)
		return 0;

	spin_lock(&spool->lock);
	if ((spool->used_hpages + delta) <= spool->max_hpages) {
		spool->used_hpages += delta;
	} else {
		ret = -ENOMEM;
	}
	spin_unlock(&spool->lock);

	return ret;
}

static void hugepage_subpool_put_pages(struct hugepage_subpool *spool,
				       long delta)
{
	if (!spool)
		return;

	spin_lock(&spool->lock);
	spool->used_hpages -= delta;
	/* If hugetlbfs_put_super couldn't free spool due to
	* an outstanding quota reference, free it now. */
	unlock_or_release_subpool(spool);
}

static inline struct hugepage_subpool *subpool_inode(struct inode *inode)
{
	return HUGETLBFS_SB(inode->i_sb)->spool;
}

static inline struct hugepage_subpool *subpool_vma(struct vm_area_struct *vma)
{
	return subpool_inode(vma->vm_file->f_dentry->d_inode);
}

133 134 135
/*
 * Region tracking -- allows tracking of reservations and instantiated pages
 *                    across the pages in a mapping.
136 137 138 139 140 141
 *
 * The region data structures are protected by a combination of the mmap_sem
 * and the hugetlb_instantion_mutex.  To access or modify a region the caller
 * must either hold the mmap_sem for write, or the mmap_sem for read and
 * the hugetlb_instantiation mutex:
 *
142
 *	down_write(&mm->mmap_sem);
143
 * or
144 145
 *	down_read(&mm->mmap_sem);
 *	mutex_lock(&hugetlb_instantiation_mutex);
146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225
 */
struct file_region {
	struct list_head link;
	long from;
	long to;
};

static long region_add(struct list_head *head, long f, long t)
{
	struct file_region *rg, *nrg, *trg;

	/* Locate the region we are either in or before. */
	list_for_each_entry(rg, head, link)
		if (f <= rg->to)
			break;

	/* Round our left edge to the current segment if it encloses us. */
	if (f > rg->from)
		f = rg->from;

	/* Check for and consume any regions we now overlap with. */
	nrg = rg;
	list_for_each_entry_safe(rg, trg, rg->link.prev, link) {
		if (&rg->link == head)
			break;
		if (rg->from > t)
			break;

		/* If this area reaches higher then extend our area to
		 * include it completely.  If this is not the first area
		 * which we intend to reuse, free it. */
		if (rg->to > t)
			t = rg->to;
		if (rg != nrg) {
			list_del(&rg->link);
			kfree(rg);
		}
	}
	nrg->from = f;
	nrg->to = t;
	return 0;
}

static long region_chg(struct list_head *head, long f, long t)
{
	struct file_region *rg, *nrg;
	long chg = 0;

	/* Locate the region we are before or in. */
	list_for_each_entry(rg, head, link)
		if (f <= rg->to)
			break;

	/* If we are below the current region then a new region is required.
	 * Subtle, allocate a new region at the position but make it zero
	 * size such that we can guarantee to record the reservation. */
	if (&rg->link == head || t < rg->from) {
		nrg = kmalloc(sizeof(*nrg), GFP_KERNEL);
		if (!nrg)
			return -ENOMEM;
		nrg->from = f;
		nrg->to   = f;
		INIT_LIST_HEAD(&nrg->link);
		list_add(&nrg->link, rg->link.prev);

		return t - f;
	}

	/* Round our left edge to the current segment if it encloses us. */
	if (f > rg->from)
		f = rg->from;
	chg = t - f;

	/* Check for and consume any regions we now overlap with. */
	list_for_each_entry(rg, rg->link.prev, link) {
		if (&rg->link == head)
			break;
		if (rg->from > t)
			return chg;

L
Lucas De Marchi 已提交
226
		/* We overlap with this area, if it extends further than
227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267
		 * us then we must extend ourselves.  Account for its
		 * existing reservation. */
		if (rg->to > t) {
			chg += rg->to - t;
			t = rg->to;
		}
		chg -= rg->to - rg->from;
	}
	return chg;
}

static long region_truncate(struct list_head *head, long end)
{
	struct file_region *rg, *trg;
	long chg = 0;

	/* Locate the region we are either in or before. */
	list_for_each_entry(rg, head, link)
		if (end <= rg->to)
			break;
	if (&rg->link == head)
		return 0;

	/* If we are in the middle of a region then adjust it. */
	if (end > rg->from) {
		chg = rg->to - end;
		rg->to = end;
		rg = list_entry(rg->link.next, typeof(*rg), link);
	}

	/* Drop any remaining regions. */
	list_for_each_entry_safe(rg, trg, rg->link.prev, link) {
		if (&rg->link == head)
			break;
		chg += rg->to - rg->from;
		list_del(&rg->link);
		kfree(rg);
	}
	return chg;
}

268 269 270 271 272 273 274
static long region_count(struct list_head *head, long f, long t)
{
	struct file_region *rg;
	long chg = 0;

	/* Locate each segment we overlap with, and count that overlap. */
	list_for_each_entry(rg, head, link) {
275 276
		long seg_from;
		long seg_to;
277 278 279 280 281 282 283 284 285 286 287 288 289 290 291

		if (rg->to <= f)
			continue;
		if (rg->from >= t)
			break;

		seg_from = max(rg->from, f);
		seg_to = min(rg->to, t);

		chg += seg_to - seg_from;
	}

	return chg;
}

292 293 294 295
/*
 * Convert the address within this vma to the page offset within
 * the mapping, in pagecache page units; huge pages here.
 */
296 297
static pgoff_t vma_hugecache_offset(struct hstate *h,
			struct vm_area_struct *vma, unsigned long address)
298
{
299 300
	return ((address - vma->vm_start) >> huge_page_shift(h)) +
			(vma->vm_pgoff >> huge_page_order(h));
301 302
}

303 304 305 306 307 308
pgoff_t linear_hugepage_index(struct vm_area_struct *vma,
				     unsigned long address)
{
	return vma_hugecache_offset(hstate_vma(vma), vma, address);
}

309 310 311 312 313 314 315 316 317 318 319 320 321 322 323
/*
 * Return the size of the pages allocated when backing a VMA. In the majority
 * cases this will be same size as used by the page table entries.
 */
unsigned long vma_kernel_pagesize(struct vm_area_struct *vma)
{
	struct hstate *hstate;

	if (!is_vm_hugetlb_page(vma))
		return PAGE_SIZE;

	hstate = hstate_vma(vma);

	return 1UL << (hstate->order + PAGE_SHIFT);
}
324
EXPORT_SYMBOL_GPL(vma_kernel_pagesize);
325

326 327 328 329 330 331 332 333 334 335 336 337 338
/*
 * Return the page size being used by the MMU to back a VMA. In the majority
 * of cases, the page size used by the kernel matches the MMU size. On
 * architectures where it differs, an architecture-specific version of this
 * function is required.
 */
#ifndef vma_mmu_pagesize
unsigned long vma_mmu_pagesize(struct vm_area_struct *vma)
{
	return vma_kernel_pagesize(vma);
}
#endif

339 340 341 342 343 344 345
/*
 * Flags for MAP_PRIVATE reservations.  These are stored in the bottom
 * bits of the reservation map pointer, which are always clear due to
 * alignment.
 */
#define HPAGE_RESV_OWNER    (1UL << 0)
#define HPAGE_RESV_UNMAPPED (1UL << 1)
346
#define HPAGE_RESV_MASK (HPAGE_RESV_OWNER | HPAGE_RESV_UNMAPPED)
347

348 349 350 351 352 353 354 355 356
/*
 * These helpers are used to track how many pages are reserved for
 * faults in a MAP_PRIVATE mapping. Only the process that called mmap()
 * is guaranteed to have their future faults succeed.
 *
 * With the exception of reset_vma_resv_huge_pages() which is called at fork(),
 * the reserve counters are updated with the hugetlb_lock held. It is safe
 * to reset the VMA at fork() time as it is not in use yet and there is no
 * chance of the global counters getting corrupted as a result of the values.
357 358 359 360 361 362 363 364 365
 *
 * The private mapping reservation is represented in a subtly different
 * manner to a shared mapping.  A shared mapping has a region map associated
 * with the underlying file, this region map represents the backing file
 * pages which have ever had a reservation assigned which this persists even
 * after the page is instantiated.  A private mapping has a region map
 * associated with the original mmap which is attached to all VMAs which
 * reference it, this region map represents those offsets which have consumed
 * reservation ie. where pages have been instantiated.
366
 */
367 368 369 370 371 372 373 374 375 376 377
static unsigned long get_vma_private_data(struct vm_area_struct *vma)
{
	return (unsigned long)vma->vm_private_data;
}

static void set_vma_private_data(struct vm_area_struct *vma,
							unsigned long value)
{
	vma->vm_private_data = (void *)value;
}

378 379 380 381 382
struct resv_map {
	struct kref refs;
	struct list_head regions;
};

383
static struct resv_map *resv_map_alloc(void)
384 385 386 387 388 389 390 391 392 393 394
{
	struct resv_map *resv_map = kmalloc(sizeof(*resv_map), GFP_KERNEL);
	if (!resv_map)
		return NULL;

	kref_init(&resv_map->refs);
	INIT_LIST_HEAD(&resv_map->regions);

	return resv_map;
}

395
static void resv_map_release(struct kref *ref)
396 397 398 399 400 401 402 403 404
{
	struct resv_map *resv_map = container_of(ref, struct resv_map, refs);

	/* Clear out any active regions before we release the map. */
	region_truncate(&resv_map->regions, 0);
	kfree(resv_map);
}

static struct resv_map *vma_resv_map(struct vm_area_struct *vma)
405 406
{
	VM_BUG_ON(!is_vm_hugetlb_page(vma));
407
	if (!(vma->vm_flags & VM_MAYSHARE))
408 409
		return (struct resv_map *)(get_vma_private_data(vma) &
							~HPAGE_RESV_MASK);
410
	return NULL;
411 412
}

413
static void set_vma_resv_map(struct vm_area_struct *vma, struct resv_map *map)
414 415
{
	VM_BUG_ON(!is_vm_hugetlb_page(vma));
416
	VM_BUG_ON(vma->vm_flags & VM_MAYSHARE);
417

418 419
	set_vma_private_data(vma, (get_vma_private_data(vma) &
				HPAGE_RESV_MASK) | (unsigned long)map);
420 421 422 423 424
}

static void set_vma_resv_flags(struct vm_area_struct *vma, unsigned long flags)
{
	VM_BUG_ON(!is_vm_hugetlb_page(vma));
425
	VM_BUG_ON(vma->vm_flags & VM_MAYSHARE);
426 427

	set_vma_private_data(vma, get_vma_private_data(vma) | flags);
428 429 430 431 432
}

static int is_vma_resv_set(struct vm_area_struct *vma, unsigned long flag)
{
	VM_BUG_ON(!is_vm_hugetlb_page(vma));
433 434

	return (get_vma_private_data(vma) & flag) != 0;
435 436 437
}

/* Decrement the reserved pages in the hugepage pool by one */
438 439
static void decrement_hugepage_resv_vma(struct hstate *h,
			struct vm_area_struct *vma)
440
{
441 442 443
	if (vma->vm_flags & VM_NORESERVE)
		return;

444
	if (vma->vm_flags & VM_MAYSHARE) {
445
		/* Shared mappings always use reserves */
446
		h->resv_huge_pages--;
447
	} else if (is_vma_resv_set(vma, HPAGE_RESV_OWNER)) {
448 449 450 451
		/*
		 * Only the process that called mmap() has reserves for
		 * private mappings.
		 */
452
		h->resv_huge_pages--;
453 454 455
	}
}

456
/* Reset counters to 0 and clear all HPAGE_RESV_* flags */
457 458 459
void reset_vma_resv_huge_pages(struct vm_area_struct *vma)
{
	VM_BUG_ON(!is_vm_hugetlb_page(vma));
460
	if (!(vma->vm_flags & VM_MAYSHARE))
461 462 463 464
		vma->vm_private_data = (void *)0;
}

/* Returns true if the VMA has associated reserve pages */
465
static int vma_has_reserves(struct vm_area_struct *vma)
466
{
467
	if (vma->vm_flags & VM_MAYSHARE)
468 469 470 471
		return 1;
	if (is_vma_resv_set(vma, HPAGE_RESV_OWNER))
		return 1;
	return 0;
472 473
}

474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507
static void copy_gigantic_page(struct page *dst, struct page *src)
{
	int i;
	struct hstate *h = page_hstate(src);
	struct page *dst_base = dst;
	struct page *src_base = src;

	for (i = 0; i < pages_per_huge_page(h); ) {
		cond_resched();
		copy_highpage(dst, src);

		i++;
		dst = mem_map_next(dst, dst_base, i);
		src = mem_map_next(src, src_base, i);
	}
}

void copy_huge_page(struct page *dst, struct page *src)
{
	int i;
	struct hstate *h = page_hstate(src);

	if (unlikely(pages_per_huge_page(h) > MAX_ORDER_NR_PAGES)) {
		copy_gigantic_page(dst, src);
		return;
	}

	might_sleep();
	for (i = 0; i < pages_per_huge_page(h); i++) {
		cond_resched();
		copy_highpage(dst + i, src + i);
	}
}

508
static void enqueue_huge_page(struct hstate *h, struct page *page)
L
Linus Torvalds 已提交
509 510
{
	int nid = page_to_nid(page);
511
	list_move(&page->lru, &h->hugepage_freelists[nid]);
512 513
	h->free_huge_pages++;
	h->free_huge_pages_node[nid]++;
L
Linus Torvalds 已提交
514 515
}

516 517 518 519 520 521 522
static struct page *dequeue_huge_page_node(struct hstate *h, int nid)
{
	struct page *page;

	if (list_empty(&h->hugepage_freelists[nid]))
		return NULL;
	page = list_entry(h->hugepage_freelists[nid].next, struct page, lru);
523
	list_move(&page->lru, &h->hugepage_activelist);
524
	set_page_refcounted(page);
525 526 527 528 529
	h->free_huge_pages--;
	h->free_huge_pages_node[nid]--;
	return page;
}

530 531
static struct page *dequeue_huge_page_vma(struct hstate *h,
				struct vm_area_struct *vma,
532
				unsigned long address, int avoid_reserve)
L
Linus Torvalds 已提交
533
{
534
	struct page *page = NULL;
535
	struct mempolicy *mpol;
536
	nodemask_t *nodemask;
537
	struct zonelist *zonelist;
538 539
	struct zone *zone;
	struct zoneref *z;
540
	unsigned int cpuset_mems_cookie;
L
Linus Torvalds 已提交
541

542 543
retry_cpuset:
	cpuset_mems_cookie = get_mems_allowed();
544 545
	zonelist = huge_zonelist(vma, address,
					htlb_alloc_mask, &mpol, &nodemask);
546 547 548 549 550
	/*
	 * A child process with MAP_PRIVATE mappings created by their parent
	 * have no page reserves. This check ensures that reservations are
	 * not "stolen". The child may still get SIGKILLed
	 */
551
	if (!vma_has_reserves(vma) &&
552
			h->free_huge_pages - h->resv_huge_pages == 0)
553
		goto err;
554

555
	/* If reserves cannot be used, ensure enough pages are in the pool */
556
	if (avoid_reserve && h->free_huge_pages - h->resv_huge_pages == 0)
557
		goto err;
558

559 560
	for_each_zone_zonelist_nodemask(zone, z, zonelist,
						MAX_NR_ZONES - 1, nodemask) {
561 562 563 564 565 566 567
		if (cpuset_zone_allowed_softwall(zone, htlb_alloc_mask)) {
			page = dequeue_huge_page_node(h, zone_to_nid(zone));
			if (page) {
				if (!avoid_reserve)
					decrement_hugepage_resv_vma(h, vma);
				break;
			}
A
Andrew Morton 已提交
568
		}
L
Linus Torvalds 已提交
569
	}
570

571
	mpol_cond_put(mpol);
572 573
	if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page))
		goto retry_cpuset;
L
Linus Torvalds 已提交
574
	return page;
575 576 577 578

err:
	mpol_cond_put(mpol);
	return NULL;
L
Linus Torvalds 已提交
579 580
}

581
static void update_and_free_page(struct hstate *h, struct page *page)
A
Adam Litke 已提交
582 583
{
	int i;
584

585 586
	VM_BUG_ON(h->order >= MAX_ORDER);

587 588 589
	h->nr_huge_pages--;
	h->nr_huge_pages_node[page_to_nid(page)]--;
	for (i = 0; i < pages_per_huge_page(h); i++) {
590 591 592 593
		page[i].flags &= ~(1 << PG_locked | 1 << PG_error |
				1 << PG_referenced | 1 << PG_dirty |
				1 << PG_active | 1 << PG_reserved |
				1 << PG_private | 1 << PG_writeback);
A
Adam Litke 已提交
594
	}
595
	VM_BUG_ON(hugetlb_cgroup_from_page(page));
A
Adam Litke 已提交
596 597
	set_compound_page_dtor(page, NULL);
	set_page_refcounted(page);
598
	arch_release_hugepage(page);
599
	__free_pages(page, huge_page_order(h));
A
Adam Litke 已提交
600 601
}

602 603 604 605 606 607 608 609 610 611 612
struct hstate *size_to_hstate(unsigned long size)
{
	struct hstate *h;

	for_each_hstate(h) {
		if (huge_page_size(h) == size)
			return h;
	}
	return NULL;
}

613 614
static void free_huge_page(struct page *page)
{
615 616 617 618
	/*
	 * Can't pass hstate in here because it is called from the
	 * compound page destructor.
	 */
619
	struct hstate *h = page_hstate(page);
620
	int nid = page_to_nid(page);
621 622
	struct hugepage_subpool *spool =
		(struct hugepage_subpool *)page_private(page);
623

624
	set_page_private(page, 0);
625
	page->mapping = NULL;
626
	BUG_ON(page_count(page));
627
	BUG_ON(page_mapcount(page));
628 629

	spin_lock(&hugetlb_lock);
630
	if (h->surplus_huge_pages_node[nid] && huge_page_order(h) < MAX_ORDER) {
631 632
		/* remove the page from active list */
		list_del(&page->lru);
633 634 635
		update_and_free_page(h, page);
		h->surplus_huge_pages--;
		h->surplus_huge_pages_node[nid]--;
636
	} else {
637
		enqueue_huge_page(h, page);
638
	}
639
	spin_unlock(&hugetlb_lock);
640
	hugepage_subpool_put_pages(spool, 1);
641 642
}

643
static void prep_new_huge_page(struct hstate *h, struct page *page, int nid)
644
{
645
	INIT_LIST_HEAD(&page->lru);
646 647
	set_compound_page_dtor(page, free_huge_page);
	spin_lock(&hugetlb_lock);
648
	set_hugetlb_cgroup(page, NULL);
649 650
	h->nr_huge_pages++;
	h->nr_huge_pages_node[nid]++;
651 652 653 654
	spin_unlock(&hugetlb_lock);
	put_page(page); /* free it into the hugepage allocator */
}

655 656 657 658 659 660 661 662 663 664 665
static void prep_compound_gigantic_page(struct page *page, unsigned long order)
{
	int i;
	int nr_pages = 1 << order;
	struct page *p = page + 1;

	/* we rely on prep_new_huge_page to set the destructor */
	set_compound_order(page, order);
	__SetPageHead(page);
	for (i = 1; i < nr_pages; i++, p = mem_map_next(p, page, i)) {
		__SetPageTail(p);
666
		set_page_count(p, 0);
667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682
		p->first_page = page;
	}
}

int PageHuge(struct page *page)
{
	compound_page_dtor *dtor;

	if (!PageCompound(page))
		return 0;

	page = compound_head(page);
	dtor = get_compound_page_dtor(page);

	return dtor == free_huge_page;
}
683 684
EXPORT_SYMBOL_GPL(PageHuge);

685
static struct page *alloc_fresh_huge_page_node(struct hstate *h, int nid)
L
Linus Torvalds 已提交
686 687
{
	struct page *page;
688

689 690 691
	if (h->order >= MAX_ORDER)
		return NULL;

692
	page = alloc_pages_exact_node(nid,
693 694
		htlb_alloc_mask|__GFP_COMP|__GFP_THISNODE|
						__GFP_REPEAT|__GFP_NOWARN,
695
		huge_page_order(h));
L
Linus Torvalds 已提交
696
	if (page) {
697
		if (arch_prepare_hugepage(page)) {
698
			__free_pages(page, huge_page_order(h));
699
			return NULL;
700
		}
701
		prep_new_huge_page(h, page, nid);
L
Linus Torvalds 已提交
702
	}
703 704 705 706

	return page;
}

707
/*
708 709 710 711 712
 * common helper functions for hstate_next_node_to_{alloc|free}.
 * We may have allocated or freed a huge page based on a different
 * nodes_allowed previously, so h->next_node_to_{alloc|free} might
 * be outside of *nodes_allowed.  Ensure that we use an allowed
 * node for alloc or free.
713
 */
714
static int next_node_allowed(int nid, nodemask_t *nodes_allowed)
715
{
716
	nid = next_node(nid, *nodes_allowed);
717
	if (nid == MAX_NUMNODES)
718
		nid = first_node(*nodes_allowed);
719 720 721 722 723
	VM_BUG_ON(nid >= MAX_NUMNODES);

	return nid;
}

724 725 726 727 728 729 730
static int get_valid_node_allowed(int nid, nodemask_t *nodes_allowed)
{
	if (!node_isset(nid, *nodes_allowed))
		nid = next_node_allowed(nid, nodes_allowed);
	return nid;
}

731
/*
732 733 734 735
 * returns the previously saved node ["this node"] from which to
 * allocate a persistent huge page for the pool and advance the
 * next node from which to allocate, handling wrap at end of node
 * mask.
736
 */
737 738
static int hstate_next_node_to_alloc(struct hstate *h,
					nodemask_t *nodes_allowed)
739
{
740 741 742 743 744 745
	int nid;

	VM_BUG_ON(!nodes_allowed);

	nid = get_valid_node_allowed(h->next_nid_to_alloc, nodes_allowed);
	h->next_nid_to_alloc = next_node_allowed(nid, nodes_allowed);
746 747

	return nid;
748 749
}

750
static int alloc_fresh_huge_page(struct hstate *h, nodemask_t *nodes_allowed)
751 752 753 754 755 756
{
	struct page *page;
	int start_nid;
	int next_nid;
	int ret = 0;

757
	start_nid = hstate_next_node_to_alloc(h, nodes_allowed);
758
	next_nid = start_nid;
759 760

	do {
761
		page = alloc_fresh_huge_page_node(h, next_nid);
762
		if (page) {
763
			ret = 1;
764 765
			break;
		}
766
		next_nid = hstate_next_node_to_alloc(h, nodes_allowed);
767
	} while (next_nid != start_nid);
768

769 770 771 772 773
	if (ret)
		count_vm_event(HTLB_BUDDY_PGALLOC);
	else
		count_vm_event(HTLB_BUDDY_PGALLOC_FAIL);

774
	return ret;
L
Linus Torvalds 已提交
775 776
}

777
/*
778 779 780 781
 * helper for free_pool_huge_page() - return the previously saved
 * node ["this node"] from which to free a huge page.  Advance the
 * next node id whether or not we find a free huge page to free so
 * that the next attempt to free addresses the next node.
782
 */
783
static int hstate_next_node_to_free(struct hstate *h, nodemask_t *nodes_allowed)
784
{
785 786 787 788 789 790
	int nid;

	VM_BUG_ON(!nodes_allowed);

	nid = get_valid_node_allowed(h->next_nid_to_free, nodes_allowed);
	h->next_nid_to_free = next_node_allowed(nid, nodes_allowed);
791 792

	return nid;
793 794 795 796 797 798 799 800
}

/*
 * Free huge page from pool from next node to free.
 * Attempt to keep persistent huge pages more or less
 * balanced over allowed nodes.
 * Called with hugetlb_lock locked.
 */
801 802
static int free_pool_huge_page(struct hstate *h, nodemask_t *nodes_allowed,
							 bool acct_surplus)
803 804 805 806 807
{
	int start_nid;
	int next_nid;
	int ret = 0;

808
	start_nid = hstate_next_node_to_free(h, nodes_allowed);
809 810 811
	next_nid = start_nid;

	do {
812 813 814 815 816 817
		/*
		 * If we're returning unused surplus pages, only examine
		 * nodes with surplus pages.
		 */
		if ((!acct_surplus || h->surplus_huge_pages_node[next_nid]) &&
		    !list_empty(&h->hugepage_freelists[next_nid])) {
818 819 820 821 822 823
			struct page *page =
				list_entry(h->hugepage_freelists[next_nid].next,
					  struct page, lru);
			list_del(&page->lru);
			h->free_huge_pages--;
			h->free_huge_pages_node[next_nid]--;
824 825 826 827
			if (acct_surplus) {
				h->surplus_huge_pages--;
				h->surplus_huge_pages_node[next_nid]--;
			}
828 829
			update_and_free_page(h, page);
			ret = 1;
830
			break;
831
		}
832
		next_nid = hstate_next_node_to_free(h, nodes_allowed);
833
	} while (next_nid != start_nid);
834 835 836 837

	return ret;
}

838
static struct page *alloc_buddy_huge_page(struct hstate *h, int nid)
839 840
{
	struct page *page;
841
	unsigned int r_nid;
842

843 844 845
	if (h->order >= MAX_ORDER)
		return NULL;

846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869
	/*
	 * Assume we will successfully allocate the surplus page to
	 * prevent racing processes from causing the surplus to exceed
	 * overcommit
	 *
	 * This however introduces a different race, where a process B
	 * tries to grow the static hugepage pool while alloc_pages() is
	 * called by process A. B will only examine the per-node
	 * counters in determining if surplus huge pages can be
	 * converted to normal huge pages in adjust_pool_surplus(). A
	 * won't be able to increment the per-node counter, until the
	 * lock is dropped by B, but B doesn't drop hugetlb_lock until
	 * no more huge pages can be converted from surplus to normal
	 * state (and doesn't try to convert again). Thus, we have a
	 * case where a surplus huge page exists, the pool is grown, and
	 * the surplus huge page still exists after, even though it
	 * should just have been converted to a normal huge page. This
	 * does not leak memory, though, as the hugepage will be freed
	 * once it is out of use. It also does not allow the counters to
	 * go out of whack in adjust_pool_surplus() as we don't modify
	 * the node values until we've gotten the hugepage and only the
	 * per-node value is checked there.
	 */
	spin_lock(&hugetlb_lock);
870
	if (h->surplus_huge_pages >= h->nr_overcommit_huge_pages) {
871 872 873
		spin_unlock(&hugetlb_lock);
		return NULL;
	} else {
874 875
		h->nr_huge_pages++;
		h->surplus_huge_pages++;
876 877 878
	}
	spin_unlock(&hugetlb_lock);

879 880 881 882 883 884 885 886
	if (nid == NUMA_NO_NODE)
		page = alloc_pages(htlb_alloc_mask|__GFP_COMP|
				   __GFP_REPEAT|__GFP_NOWARN,
				   huge_page_order(h));
	else
		page = alloc_pages_exact_node(nid,
			htlb_alloc_mask|__GFP_COMP|__GFP_THISNODE|
			__GFP_REPEAT|__GFP_NOWARN, huge_page_order(h));
887

888 889
	if (page && arch_prepare_hugepage(page)) {
		__free_pages(page, huge_page_order(h));
890
		page = NULL;
891 892
	}

893
	spin_lock(&hugetlb_lock);
894
	if (page) {
895
		INIT_LIST_HEAD(&page->lru);
896
		r_nid = page_to_nid(page);
897
		set_compound_page_dtor(page, free_huge_page);
898
		set_hugetlb_cgroup(page, NULL);
899 900 901
		/*
		 * We incremented the global counters already
		 */
902 903
		h->nr_huge_pages_node[r_nid]++;
		h->surplus_huge_pages_node[r_nid]++;
904
		__count_vm_event(HTLB_BUDDY_PGALLOC);
905
	} else {
906 907
		h->nr_huge_pages--;
		h->surplus_huge_pages--;
908
		__count_vm_event(HTLB_BUDDY_PGALLOC_FAIL);
909
	}
910
	spin_unlock(&hugetlb_lock);
911 912 913 914

	return page;
}

915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933
/*
 * This allocation function is useful in the context where vma is irrelevant.
 * E.g. soft-offlining uses this function because it only cares physical
 * address of error page.
 */
struct page *alloc_huge_page_node(struct hstate *h, int nid)
{
	struct page *page;

	spin_lock(&hugetlb_lock);
	page = dequeue_huge_page_node(h, nid);
	spin_unlock(&hugetlb_lock);

	if (!page)
		page = alloc_buddy_huge_page(h, nid);

	return page;
}

934
/*
L
Lucas De Marchi 已提交
935
 * Increase the hugetlb pool such that it can accommodate a reservation
936 937
 * of size 'delta'.
 */
938
static int gather_surplus_pages(struct hstate *h, int delta)
939 940 941 942 943
{
	struct list_head surplus_list;
	struct page *page, *tmp;
	int ret, i;
	int needed, allocated;
944
	bool alloc_ok = true;
945

946
	needed = (h->resv_huge_pages + delta) - h->free_huge_pages;
947
	if (needed <= 0) {
948
		h->resv_huge_pages += delta;
949
		return 0;
950
	}
951 952 953 954 955 956 957 958

	allocated = 0;
	INIT_LIST_HEAD(&surplus_list);

	ret = -ENOMEM;
retry:
	spin_unlock(&hugetlb_lock);
	for (i = 0; i < needed; i++) {
959
		page = alloc_buddy_huge_page(h, NUMA_NO_NODE);
960 961 962 963
		if (!page) {
			alloc_ok = false;
			break;
		}
964 965
		list_add(&page->lru, &surplus_list);
	}
966
	allocated += i;
967 968 969 970 971 972

	/*
	 * After retaking hugetlb_lock, we need to recalculate 'needed'
	 * because either resv_huge_pages or free_huge_pages may have changed.
	 */
	spin_lock(&hugetlb_lock);
973 974
	needed = (h->resv_huge_pages + delta) -
			(h->free_huge_pages + allocated);
975 976 977 978 979 980 981 982 983 984
	if (needed > 0) {
		if (alloc_ok)
			goto retry;
		/*
		 * We were not able to allocate enough pages to
		 * satisfy the entire reservation so we free what
		 * we've allocated so far.
		 */
		goto free;
	}
985 986
	/*
	 * The surplus_list now contains _at_least_ the number of extra pages
L
Lucas De Marchi 已提交
987
	 * needed to accommodate the reservation.  Add the appropriate number
988
	 * of pages to the hugetlb pool and free the extras back to the buddy
989 990 991
	 * allocator.  Commit the entire reservation here to prevent another
	 * process from stealing the pages as they are added to the pool but
	 * before they are reserved.
992 993
	 */
	needed += allocated;
994
	h->resv_huge_pages += delta;
995
	ret = 0;
996

997
	/* Free the needed pages to the hugetlb pool */
998
	list_for_each_entry_safe(page, tmp, &surplus_list, lru) {
999 1000
		if ((--needed) < 0)
			break;
1001 1002 1003 1004 1005 1006
		/*
		 * This page is now managed by the hugetlb allocator and has
		 * no users -- drop the buddy allocator's reference.
		 */
		put_page_testzero(page);
		VM_BUG_ON(page_count(page));
1007
		enqueue_huge_page(h, page);
1008
	}
1009
free:
1010
	spin_unlock(&hugetlb_lock);
1011 1012 1013 1014

	/* Free unnecessary surplus pages to the buddy allocator */
	if (!list_empty(&surplus_list)) {
		list_for_each_entry_safe(page, tmp, &surplus_list, lru) {
1015
			put_page(page);
1016
		}
1017
	}
1018
	spin_lock(&hugetlb_lock);
1019 1020 1021 1022 1023 1024 1025 1026

	return ret;
}

/*
 * When releasing a hugetlb pool reservation, any surplus pages that were
 * allocated to satisfy the reservation must be explicitly freed if they were
 * never used.
1027
 * Called with hugetlb_lock held.
1028
 */
1029 1030
static void return_unused_surplus_pages(struct hstate *h,
					unsigned long unused_resv_pages)
1031 1032 1033
{
	unsigned long nr_pages;

1034
	/* Uncommit the reservation */
1035
	h->resv_huge_pages -= unused_resv_pages;
1036

1037 1038 1039 1040
	/* Cannot return gigantic pages currently */
	if (h->order >= MAX_ORDER)
		return;

1041
	nr_pages = min(unused_resv_pages, h->surplus_huge_pages);
1042

1043 1044
	/*
	 * We want to release as many surplus pages as possible, spread
1045 1046 1047 1048 1049
	 * evenly across all nodes with memory. Iterate across these nodes
	 * until we can no longer free unreserved surplus pages. This occurs
	 * when the nodes with surplus pages have no free pages.
	 * free_pool_huge_page() will balance the the freed pages across the
	 * on-line nodes with memory and will handle the hstate accounting.
1050 1051
	 */
	while (nr_pages--) {
1052
		if (!free_pool_huge_page(h, &node_states[N_HIGH_MEMORY], 1))
1053
			break;
1054 1055 1056
	}
}

1057 1058 1059
/*
 * Determine if the huge page at addr within the vma has an associated
 * reservation.  Where it does not we will need to logically increase
1060 1061 1062 1063 1064 1065
 * reservation and actually increase subpool usage before an allocation
 * can occur.  Where any new reservation would be required the
 * reservation change is prepared, but not committed.  Once the page
 * has been allocated from the subpool and instantiated the change should
 * be committed via vma_commit_reservation.  No action is required on
 * failure.
1066
 */
1067
static long vma_needs_reservation(struct hstate *h,
1068
			struct vm_area_struct *vma, unsigned long addr)
1069 1070 1071 1072
{
	struct address_space *mapping = vma->vm_file->f_mapping;
	struct inode *inode = mapping->host;

1073
	if (vma->vm_flags & VM_MAYSHARE) {
1074
		pgoff_t idx = vma_hugecache_offset(h, vma, addr);
1075 1076 1077
		return region_chg(&inode->i_mapping->private_list,
							idx, idx + 1);

1078 1079
	} else if (!is_vma_resv_set(vma, HPAGE_RESV_OWNER)) {
		return 1;
1080

1081
	} else  {
1082
		long err;
1083
		pgoff_t idx = vma_hugecache_offset(h, vma, addr);
1084 1085 1086 1087 1088 1089 1090
		struct resv_map *reservations = vma_resv_map(vma);

		err = region_chg(&reservations->regions, idx, idx + 1);
		if (err < 0)
			return err;
		return 0;
	}
1091
}
1092 1093
static void vma_commit_reservation(struct hstate *h,
			struct vm_area_struct *vma, unsigned long addr)
1094 1095 1096 1097
{
	struct address_space *mapping = vma->vm_file->f_mapping;
	struct inode *inode = mapping->host;

1098
	if (vma->vm_flags & VM_MAYSHARE) {
1099
		pgoff_t idx = vma_hugecache_offset(h, vma, addr);
1100
		region_add(&inode->i_mapping->private_list, idx, idx + 1);
1101 1102

	} else if (is_vma_resv_set(vma, HPAGE_RESV_OWNER)) {
1103
		pgoff_t idx = vma_hugecache_offset(h, vma, addr);
1104 1105 1106 1107
		struct resv_map *reservations = vma_resv_map(vma);

		/* Mark this page used in the map. */
		region_add(&reservations->regions, idx, idx + 1);
1108 1109 1110
	}
}

1111
static struct page *alloc_huge_page(struct vm_area_struct *vma,
1112
				    unsigned long addr, int avoid_reserve)
L
Linus Torvalds 已提交
1113
{
1114
	struct hugepage_subpool *spool = subpool_vma(vma);
1115
	struct hstate *h = hstate_vma(vma);
1116
	struct page *page;
1117
	long chg;
1118 1119

	/*
1120 1121 1122 1123 1124 1125
	 * Processes that did not create the mapping will have no
	 * reserves and will not have accounted against subpool
	 * limit. Check that the subpool limit can be made before
	 * satisfying the allocation MAP_NORESERVE mappings may also
	 * need pages and subpool limit allocated allocated if no reserve
	 * mapping overlaps.
1126
	 */
1127
	chg = vma_needs_reservation(h, vma, addr);
1128
	if (chg < 0)
1129
		return ERR_PTR(-ENOMEM);
1130
	if (chg)
1131
		if (hugepage_subpool_get_pages(spool, chg))
1132
			return ERR_PTR(-ENOSPC);
L
Linus Torvalds 已提交
1133 1134

	spin_lock(&hugetlb_lock);
1135
	page = dequeue_huge_page_vma(h, vma, addr, avoid_reserve);
L
Linus Torvalds 已提交
1136
	spin_unlock(&hugetlb_lock);
1137

K
Ken Chen 已提交
1138
	if (!page) {
1139
		page = alloc_buddy_huge_page(h, NUMA_NO_NODE);
K
Ken Chen 已提交
1140
		if (!page) {
1141
			hugepage_subpool_put_pages(spool, chg);
1142
			return ERR_PTR(-ENOSPC);
K
Ken Chen 已提交
1143 1144
		}
	}
1145

1146
	set_page_private(page, (unsigned long)spool);
1147

1148
	vma_commit_reservation(h, vma, addr);
1149

1150
	return page;
1151 1152
}

1153
int __weak alloc_bootmem_huge_page(struct hstate *h)
1154 1155
{
	struct huge_bootmem_page *m;
1156
	int nr_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
1157 1158 1159 1160 1161

	while (nr_nodes) {
		void *addr;

		addr = __alloc_bootmem_node_nopanic(
1162
				NODE_DATA(hstate_next_node_to_alloc(h,
1163
						&node_states[N_HIGH_MEMORY])),
1164 1165 1166 1167 1168 1169 1170 1171 1172
				huge_page_size(h), huge_page_size(h), 0);

		if (addr) {
			/*
			 * Use the beginning of the huge page to store the
			 * huge_bootmem_page struct (until gather_bootmem
			 * puts them into the mem_map).
			 */
			m = addr;
1173
			goto found;
1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186
		}
		nr_nodes--;
	}
	return 0;

found:
	BUG_ON((unsigned long)virt_to_phys(m) & (huge_page_size(h) - 1));
	/* Put them into a private list first because mem_map is not up yet */
	list_add(&m->list, &huge_boot_pages);
	m->hstate = h;
	return 1;
}

1187 1188 1189 1190 1191 1192 1193 1194
static void prep_compound_huge_page(struct page *page, int order)
{
	if (unlikely(order > (MAX_ORDER - 1)))
		prep_compound_gigantic_page(page, order);
	else
		prep_compound_page(page, order);
}

1195 1196 1197 1198 1199 1200 1201
/* Put bootmem huge pages into the standard lists after mem_map is up */
static void __init gather_bootmem_prealloc(void)
{
	struct huge_bootmem_page *m;

	list_for_each_entry(m, &huge_boot_pages, list) {
		struct hstate *h = m->hstate;
1202 1203 1204 1205 1206 1207 1208 1209 1210
		struct page *page;

#ifdef CONFIG_HIGHMEM
		page = pfn_to_page(m->phys >> PAGE_SHIFT);
		free_bootmem_late((unsigned long)m,
				  sizeof(struct huge_bootmem_page));
#else
		page = virt_to_page(m);
#endif
1211 1212
		__ClearPageReserved(page);
		WARN_ON(page_count(page) != 1);
1213
		prep_compound_huge_page(page, h->order);
1214
		prep_new_huge_page(h, page, page_to_nid(page));
1215 1216 1217 1218 1219 1220 1221 1222
		/*
		 * If we had gigantic hugepages allocated at boot time, we need
		 * to restore the 'stolen' pages to totalram_pages in order to
		 * fix confusing memory reports from free(1) and another
		 * side-effects, like CommitLimit going negative.
		 */
		if (h->order > (MAX_ORDER - 1))
			totalram_pages += 1 << h->order;
1223 1224 1225
	}
}

1226
static void __init hugetlb_hstate_alloc_pages(struct hstate *h)
L
Linus Torvalds 已提交
1227 1228
{
	unsigned long i;
1229

1230
	for (i = 0; i < h->max_huge_pages; ++i) {
1231 1232 1233
		if (h->order >= MAX_ORDER) {
			if (!alloc_bootmem_huge_page(h))
				break;
1234 1235
		} else if (!alloc_fresh_huge_page(h,
					 &node_states[N_HIGH_MEMORY]))
L
Linus Torvalds 已提交
1236 1237
			break;
	}
1238
	h->max_huge_pages = i;
1239 1240 1241 1242 1243 1244 1245
}

static void __init hugetlb_init_hstates(void)
{
	struct hstate *h;

	for_each_hstate(h) {
1246 1247 1248
		/* oversize hugepages were init'ed in early boot */
		if (h->order < MAX_ORDER)
			hugetlb_hstate_alloc_pages(h);
1249 1250 1251
	}
}

A
Andi Kleen 已提交
1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262
static char * __init memfmt(char *buf, unsigned long n)
{
	if (n >= (1UL << 30))
		sprintf(buf, "%lu GB", n >> 30);
	else if (n >= (1UL << 20))
		sprintf(buf, "%lu MB", n >> 20);
	else
		sprintf(buf, "%lu KB", n >> 10);
	return buf;
}

1263 1264 1265 1266 1267
static void __init report_hugepages(void)
{
	struct hstate *h;

	for_each_hstate(h) {
A
Andi Kleen 已提交
1268 1269 1270 1271 1272
		char buf[32];
		printk(KERN_INFO "HugeTLB registered %s page size, "
				 "pre-allocated %ld pages\n",
			memfmt(buf, huge_page_size(h)),
			h->free_huge_pages);
1273 1274 1275
	}
}

L
Linus Torvalds 已提交
1276
#ifdef CONFIG_HIGHMEM
1277 1278
static void try_to_free_low(struct hstate *h, unsigned long count,
						nodemask_t *nodes_allowed)
L
Linus Torvalds 已提交
1279
{
1280 1281
	int i;

1282 1283 1284
	if (h->order >= MAX_ORDER)
		return;

1285
	for_each_node_mask(i, *nodes_allowed) {
L
Linus Torvalds 已提交
1286
		struct page *page, *next;
1287 1288 1289
		struct list_head *freel = &h->hugepage_freelists[i];
		list_for_each_entry_safe(page, next, freel, lru) {
			if (count >= h->nr_huge_pages)
1290
				return;
L
Linus Torvalds 已提交
1291 1292 1293
			if (PageHighMem(page))
				continue;
			list_del(&page->lru);
1294
			update_and_free_page(h, page);
1295 1296
			h->free_huge_pages--;
			h->free_huge_pages_node[page_to_nid(page)]--;
L
Linus Torvalds 已提交
1297 1298 1299 1300
		}
	}
}
#else
1301 1302
static inline void try_to_free_low(struct hstate *h, unsigned long count,
						nodemask_t *nodes_allowed)
L
Linus Torvalds 已提交
1303 1304 1305 1306
{
}
#endif

1307 1308 1309 1310 1311
/*
 * Increment or decrement surplus_huge_pages.  Keep node-specific counters
 * balanced by operating on them in a round-robin fashion.
 * Returns 1 if an adjustment was made.
 */
1312 1313
static int adjust_pool_surplus(struct hstate *h, nodemask_t *nodes_allowed,
				int delta)
1314
{
1315
	int start_nid, next_nid;
1316 1317 1318 1319
	int ret = 0;

	VM_BUG_ON(delta != -1 && delta != 1);

1320
	if (delta < 0)
1321
		start_nid = hstate_next_node_to_alloc(h, nodes_allowed);
1322
	else
1323
		start_nid = hstate_next_node_to_free(h, nodes_allowed);
1324 1325 1326 1327 1328 1329 1330 1331
	next_nid = start_nid;

	do {
		int nid = next_nid;
		if (delta < 0)  {
			/*
			 * To shrink on this node, there must be a surplus page
			 */
1332
			if (!h->surplus_huge_pages_node[nid]) {
1333 1334
				next_nid = hstate_next_node_to_alloc(h,
								nodes_allowed);
1335
				continue;
1336
			}
1337 1338 1339 1340 1341 1342
		}
		if (delta > 0) {
			/*
			 * Surplus cannot exceed the total number of pages
			 */
			if (h->surplus_huge_pages_node[nid] >=
1343
						h->nr_huge_pages_node[nid]) {
1344 1345
				next_nid = hstate_next_node_to_free(h,
								nodes_allowed);
1346
				continue;
1347
			}
1348
		}
1349 1350 1351 1352 1353

		h->surplus_huge_pages += delta;
		h->surplus_huge_pages_node[nid] += delta;
		ret = 1;
		break;
1354
	} while (next_nid != start_nid);
1355 1356 1357 1358

	return ret;
}

1359
#define persistent_huge_pages(h) (h->nr_huge_pages - h->surplus_huge_pages)
1360 1361
static unsigned long set_max_huge_pages(struct hstate *h, unsigned long count,
						nodemask_t *nodes_allowed)
L
Linus Torvalds 已提交
1362
{
1363
	unsigned long min_count, ret;
L
Linus Torvalds 已提交
1364

1365 1366 1367
	if (h->order >= MAX_ORDER)
		return h->max_huge_pages;

1368 1369 1370 1371
	/*
	 * Increase the pool size
	 * First take pages out of surplus state.  Then make up the
	 * remaining difference by allocating fresh huge pages.
1372 1373 1374 1375 1376 1377
	 *
	 * We might race with alloc_buddy_huge_page() here and be unable
	 * to convert a surplus huge page to a normal huge page. That is
	 * not critical, though, it just means the overall size of the
	 * pool might be one hugepage larger than it needs to be, but
	 * within all the constraints specified by the sysctls.
1378
	 */
L
Linus Torvalds 已提交
1379
	spin_lock(&hugetlb_lock);
1380
	while (h->surplus_huge_pages && count > persistent_huge_pages(h)) {
1381
		if (!adjust_pool_surplus(h, nodes_allowed, -1))
1382 1383 1384
			break;
	}

1385
	while (count > persistent_huge_pages(h)) {
1386 1387 1388 1389 1390 1391
		/*
		 * If this allocation races such that we no longer need the
		 * page, free_huge_page will handle it by freeing the page
		 * and reducing the surplus.
		 */
		spin_unlock(&hugetlb_lock);
1392
		ret = alloc_fresh_huge_page(h, nodes_allowed);
1393 1394 1395 1396
		spin_lock(&hugetlb_lock);
		if (!ret)
			goto out;

1397 1398 1399
		/* Bail for signals. Probably ctrl-c from user */
		if (signal_pending(current))
			goto out;
1400 1401 1402 1403 1404 1405 1406 1407
	}

	/*
	 * Decrease the pool size
	 * First return free pages to the buddy allocator (being careful
	 * to keep enough around to satisfy reservations).  Then place
	 * pages into surplus state as needed so the pool will shrink
	 * to the desired size as pages become free.
1408 1409 1410 1411 1412 1413 1414 1415
	 *
	 * By placing pages into the surplus state independent of the
	 * overcommit value, we are allowing the surplus pool size to
	 * exceed overcommit. There are few sane options here. Since
	 * alloc_buddy_huge_page() is checking the global counter,
	 * though, we'll note that we're not allowed to exceed surplus
	 * and won't grow the pool anywhere else. Not until one of the
	 * sysctls are changed, or the surplus pages go out of use.
1416
	 */
1417
	min_count = h->resv_huge_pages + h->nr_huge_pages - h->free_huge_pages;
1418
	min_count = max(count, min_count);
1419
	try_to_free_low(h, min_count, nodes_allowed);
1420
	while (min_count < persistent_huge_pages(h)) {
1421
		if (!free_pool_huge_page(h, nodes_allowed, 0))
L
Linus Torvalds 已提交
1422 1423
			break;
	}
1424
	while (count < persistent_huge_pages(h)) {
1425
		if (!adjust_pool_surplus(h, nodes_allowed, 1))
1426 1427 1428
			break;
	}
out:
1429
	ret = persistent_huge_pages(h);
L
Linus Torvalds 已提交
1430
	spin_unlock(&hugetlb_lock);
1431
	return ret;
L
Linus Torvalds 已提交
1432 1433
}

1434 1435 1436 1437 1438 1439 1440 1441 1442 1443
#define HSTATE_ATTR_RO(_name) \
	static struct kobj_attribute _name##_attr = __ATTR_RO(_name)

#define HSTATE_ATTR(_name) \
	static struct kobj_attribute _name##_attr = \
		__ATTR(_name, 0644, _name##_show, _name##_store)

static struct kobject *hugepages_kobj;
static struct kobject *hstate_kobjs[HUGE_MAX_HSTATE];

1444 1445 1446
static struct hstate *kobj_to_node_hstate(struct kobject *kobj, int *nidp);

static struct hstate *kobj_to_hstate(struct kobject *kobj, int *nidp)
1447 1448
{
	int i;
1449

1450
	for (i = 0; i < HUGE_MAX_HSTATE; i++)
1451 1452 1453
		if (hstate_kobjs[i] == kobj) {
			if (nidp)
				*nidp = NUMA_NO_NODE;
1454
			return &hstates[i];
1455 1456 1457
		}

	return kobj_to_node_hstate(kobj, nidp);
1458 1459
}

1460
static ssize_t nr_hugepages_show_common(struct kobject *kobj,
1461 1462
					struct kobj_attribute *attr, char *buf)
{
1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473
	struct hstate *h;
	unsigned long nr_huge_pages;
	int nid;

	h = kobj_to_hstate(kobj, &nid);
	if (nid == NUMA_NO_NODE)
		nr_huge_pages = h->nr_huge_pages;
	else
		nr_huge_pages = h->nr_huge_pages_node[nid];

	return sprintf(buf, "%lu\n", nr_huge_pages);
1474
}
1475

1476 1477 1478
static ssize_t nr_hugepages_store_common(bool obey_mempolicy,
			struct kobject *kobj, struct kobj_attribute *attr,
			const char *buf, size_t len)
1479 1480
{
	int err;
1481
	int nid;
1482
	unsigned long count;
1483
	struct hstate *h;
1484
	NODEMASK_ALLOC(nodemask_t, nodes_allowed, GFP_KERNEL | __GFP_NORETRY);
1485

1486
	err = strict_strtoul(buf, 10, &count);
1487
	if (err)
1488
		goto out;
1489

1490
	h = kobj_to_hstate(kobj, &nid);
1491 1492 1493 1494 1495
	if (h->order >= MAX_ORDER) {
		err = -EINVAL;
		goto out;
	}

1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514
	if (nid == NUMA_NO_NODE) {
		/*
		 * global hstate attribute
		 */
		if (!(obey_mempolicy &&
				init_nodemask_of_mempolicy(nodes_allowed))) {
			NODEMASK_FREE(nodes_allowed);
			nodes_allowed = &node_states[N_HIGH_MEMORY];
		}
	} else if (nodes_allowed) {
		/*
		 * per node hstate attribute: adjust count to global,
		 * but restrict alloc/free to the specified node.
		 */
		count += h->nr_huge_pages - h->nr_huge_pages_node[nid];
		init_nodemask_of_node(nodes_allowed, nid);
	} else
		nodes_allowed = &node_states[N_HIGH_MEMORY];

1515
	h->max_huge_pages = set_max_huge_pages(h, count, nodes_allowed);
1516

1517
	if (nodes_allowed != &node_states[N_HIGH_MEMORY])
1518 1519 1520
		NODEMASK_FREE(nodes_allowed);

	return len;
1521 1522 1523
out:
	NODEMASK_FREE(nodes_allowed);
	return err;
1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535
}

static ssize_t nr_hugepages_show(struct kobject *kobj,
				       struct kobj_attribute *attr, char *buf)
{
	return nr_hugepages_show_common(kobj, attr, buf);
}

static ssize_t nr_hugepages_store(struct kobject *kobj,
	       struct kobj_attribute *attr, const char *buf, size_t len)
{
	return nr_hugepages_store_common(false, kobj, attr, buf, len);
1536 1537 1538
}
HSTATE_ATTR(nr_hugepages);

1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559
#ifdef CONFIG_NUMA

/*
 * hstate attribute for optionally mempolicy-based constraint on persistent
 * huge page alloc/free.
 */
static ssize_t nr_hugepages_mempolicy_show(struct kobject *kobj,
				       struct kobj_attribute *attr, char *buf)
{
	return nr_hugepages_show_common(kobj, attr, buf);
}

static ssize_t nr_hugepages_mempolicy_store(struct kobject *kobj,
	       struct kobj_attribute *attr, const char *buf, size_t len)
{
	return nr_hugepages_store_common(true, kobj, attr, buf, len);
}
HSTATE_ATTR(nr_hugepages_mempolicy);
#endif


1560 1561 1562
static ssize_t nr_overcommit_hugepages_show(struct kobject *kobj,
					struct kobj_attribute *attr, char *buf)
{
1563
	struct hstate *h = kobj_to_hstate(kobj, NULL);
1564 1565
	return sprintf(buf, "%lu\n", h->nr_overcommit_huge_pages);
}
1566

1567 1568 1569 1570 1571
static ssize_t nr_overcommit_hugepages_store(struct kobject *kobj,
		struct kobj_attribute *attr, const char *buf, size_t count)
{
	int err;
	unsigned long input;
1572
	struct hstate *h = kobj_to_hstate(kobj, NULL);
1573

1574 1575 1576
	if (h->order >= MAX_ORDER)
		return -EINVAL;

1577 1578
	err = strict_strtoul(buf, 10, &input);
	if (err)
1579
		return err;
1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591

	spin_lock(&hugetlb_lock);
	h->nr_overcommit_huge_pages = input;
	spin_unlock(&hugetlb_lock);

	return count;
}
HSTATE_ATTR(nr_overcommit_hugepages);

static ssize_t free_hugepages_show(struct kobject *kobj,
					struct kobj_attribute *attr, char *buf)
{
1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602
	struct hstate *h;
	unsigned long free_huge_pages;
	int nid;

	h = kobj_to_hstate(kobj, &nid);
	if (nid == NUMA_NO_NODE)
		free_huge_pages = h->free_huge_pages;
	else
		free_huge_pages = h->free_huge_pages_node[nid];

	return sprintf(buf, "%lu\n", free_huge_pages);
1603 1604 1605 1606 1607 1608
}
HSTATE_ATTR_RO(free_hugepages);

static ssize_t resv_hugepages_show(struct kobject *kobj,
					struct kobj_attribute *attr, char *buf)
{
1609
	struct hstate *h = kobj_to_hstate(kobj, NULL);
1610 1611 1612 1613 1614 1615 1616
	return sprintf(buf, "%lu\n", h->resv_huge_pages);
}
HSTATE_ATTR_RO(resv_hugepages);

static ssize_t surplus_hugepages_show(struct kobject *kobj,
					struct kobj_attribute *attr, char *buf)
{
1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627
	struct hstate *h;
	unsigned long surplus_huge_pages;
	int nid;

	h = kobj_to_hstate(kobj, &nid);
	if (nid == NUMA_NO_NODE)
		surplus_huge_pages = h->surplus_huge_pages;
	else
		surplus_huge_pages = h->surplus_huge_pages_node[nid];

	return sprintf(buf, "%lu\n", surplus_huge_pages);
1628 1629 1630 1631 1632 1633 1634 1635 1636
}
HSTATE_ATTR_RO(surplus_hugepages);

static struct attribute *hstate_attrs[] = {
	&nr_hugepages_attr.attr,
	&nr_overcommit_hugepages_attr.attr,
	&free_hugepages_attr.attr,
	&resv_hugepages_attr.attr,
	&surplus_hugepages_attr.attr,
1637 1638 1639
#ifdef CONFIG_NUMA
	&nr_hugepages_mempolicy_attr.attr,
#endif
1640 1641 1642 1643 1644 1645 1646
	NULL,
};

static struct attribute_group hstate_attr_group = {
	.attrs = hstate_attrs,
};

J
Jeff Mahoney 已提交
1647 1648 1649
static int hugetlb_sysfs_add_hstate(struct hstate *h, struct kobject *parent,
				    struct kobject **hstate_kobjs,
				    struct attribute_group *hstate_attr_group)
1650 1651
{
	int retval;
1652
	int hi = hstate_index(h);
1653

1654 1655
	hstate_kobjs[hi] = kobject_create_and_add(h->name, parent);
	if (!hstate_kobjs[hi])
1656 1657
		return -ENOMEM;

1658
	retval = sysfs_create_group(hstate_kobjs[hi], hstate_attr_group);
1659
	if (retval)
1660
		kobject_put(hstate_kobjs[hi]);
1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674

	return retval;
}

static void __init hugetlb_sysfs_init(void)
{
	struct hstate *h;
	int err;

	hugepages_kobj = kobject_create_and_add("hugepages", mm_kobj);
	if (!hugepages_kobj)
		return;

	for_each_hstate(h) {
1675 1676
		err = hugetlb_sysfs_add_hstate(h, hugepages_kobj,
					 hstate_kobjs, &hstate_attr_group);
1677 1678 1679 1680 1681 1682
		if (err)
			printk(KERN_ERR "Hugetlb: Unable to add hstate %s",
								h->name);
	}
}

1683 1684 1685 1686
#ifdef CONFIG_NUMA

/*
 * node_hstate/s - associate per node hstate attributes, via their kobjects,
1687 1688 1689
 * with node devices in node_devices[] using a parallel array.  The array
 * index of a node device or _hstate == node id.
 * This is here to avoid any static dependency of the node device driver, in
1690 1691 1692 1693 1694 1695 1696 1697 1698
 * the base kernel, on the hugetlb module.
 */
struct node_hstate {
	struct kobject		*hugepages_kobj;
	struct kobject		*hstate_kobjs[HUGE_MAX_HSTATE];
};
struct node_hstate node_hstates[MAX_NUMNODES];

/*
1699
 * A subset of global hstate attributes for node devices
1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712
 */
static struct attribute *per_node_hstate_attrs[] = {
	&nr_hugepages_attr.attr,
	&free_hugepages_attr.attr,
	&surplus_hugepages_attr.attr,
	NULL,
};

static struct attribute_group per_node_hstate_attr_group = {
	.attrs = per_node_hstate_attrs,
};

/*
1713
 * kobj_to_node_hstate - lookup global hstate for node device hstate attr kobj.
1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735
 * Returns node id via non-NULL nidp.
 */
static struct hstate *kobj_to_node_hstate(struct kobject *kobj, int *nidp)
{
	int nid;

	for (nid = 0; nid < nr_node_ids; nid++) {
		struct node_hstate *nhs = &node_hstates[nid];
		int i;
		for (i = 0; i < HUGE_MAX_HSTATE; i++)
			if (nhs->hstate_kobjs[i] == kobj) {
				if (nidp)
					*nidp = nid;
				return &hstates[i];
			}
	}

	BUG();
	return NULL;
}

/*
1736
 * Unregister hstate attributes from a single node device.
1737 1738 1739 1740 1741
 * No-op if no hstate attributes attached.
 */
void hugetlb_unregister_node(struct node *node)
{
	struct hstate *h;
1742
	struct node_hstate *nhs = &node_hstates[node->dev.id];
1743 1744

	if (!nhs->hugepages_kobj)
1745
		return;		/* no hstate attributes */
1746

1747 1748 1749 1750 1751
	for_each_hstate(h) {
		int idx = hstate_index(h);
		if (nhs->hstate_kobjs[idx]) {
			kobject_put(nhs->hstate_kobjs[idx]);
			nhs->hstate_kobjs[idx] = NULL;
1752
		}
1753
	}
1754 1755 1756 1757 1758 1759

	kobject_put(nhs->hugepages_kobj);
	nhs->hugepages_kobj = NULL;
}

/*
1760
 * hugetlb module exit:  unregister hstate attributes from node devices
1761 1762 1763 1764 1765 1766 1767
 * that have them.
 */
static void hugetlb_unregister_all_nodes(void)
{
	int nid;

	/*
1768
	 * disable node device registrations.
1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779
	 */
	register_hugetlbfs_with_node(NULL, NULL);

	/*
	 * remove hstate attributes from any nodes that have them.
	 */
	for (nid = 0; nid < nr_node_ids; nid++)
		hugetlb_unregister_node(&node_devices[nid]);
}

/*
1780
 * Register hstate attributes for a single node device.
1781 1782 1783 1784 1785
 * No-op if attributes already registered.
 */
void hugetlb_register_node(struct node *node)
{
	struct hstate *h;
1786
	struct node_hstate *nhs = &node_hstates[node->dev.id];
1787 1788 1789 1790 1791 1792
	int err;

	if (nhs->hugepages_kobj)
		return;		/* already allocated */

	nhs->hugepages_kobj = kobject_create_and_add("hugepages",
1793
							&node->dev.kobj);
1794 1795 1796 1797 1798 1799 1800 1801 1802 1803
	if (!nhs->hugepages_kobj)
		return;

	for_each_hstate(h) {
		err = hugetlb_sysfs_add_hstate(h, nhs->hugepages_kobj,
						nhs->hstate_kobjs,
						&per_node_hstate_attr_group);
		if (err) {
			printk(KERN_ERR "Hugetlb: Unable to add hstate %s"
					" for node %d\n",
1804
						h->name, node->dev.id);
1805 1806 1807 1808 1809 1810 1811
			hugetlb_unregister_node(node);
			break;
		}
	}
}

/*
1812
 * hugetlb init time:  register hstate attributes for all registered node
1813 1814
 * devices of nodes that have memory.  All on-line nodes should have
 * registered their associated device by this time.
1815 1816 1817 1818 1819
 */
static void hugetlb_register_all_nodes(void)
{
	int nid;

1820
	for_each_node_state(nid, N_HIGH_MEMORY) {
1821
		struct node *node = &node_devices[nid];
1822
		if (node->dev.id == nid)
1823 1824 1825 1826
			hugetlb_register_node(node);
	}

	/*
1827
	 * Let the node device driver know we're here so it can
1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848
	 * [un]register hstate attributes on node hotplug.
	 */
	register_hugetlbfs_with_node(hugetlb_register_node,
				     hugetlb_unregister_node);
}
#else	/* !CONFIG_NUMA */

static struct hstate *kobj_to_node_hstate(struct kobject *kobj, int *nidp)
{
	BUG();
	if (nidp)
		*nidp = -1;
	return NULL;
}

static void hugetlb_unregister_all_nodes(void) { }

static void hugetlb_register_all_nodes(void) { }

#endif

1849 1850 1851 1852
static void __exit hugetlb_exit(void)
{
	struct hstate *h;

1853 1854
	hugetlb_unregister_all_nodes();

1855
	for_each_hstate(h) {
1856
		kobject_put(hstate_kobjs[hstate_index(h)]);
1857 1858 1859 1860 1861 1862 1863 1864
	}

	kobject_put(hugepages_kobj);
}
module_exit(hugetlb_exit);

static int __init hugetlb_init(void)
{
1865 1866 1867 1868 1869 1870
	/* Some platform decide whether they support huge pages at boot
	 * time. On these, such as powerpc, HPAGE_SHIFT is set to 0 when
	 * there is no such support
	 */
	if (HPAGE_SHIFT == 0)
		return 0;
1871

1872 1873 1874 1875
	if (!size_to_hstate(default_hstate_size)) {
		default_hstate_size = HPAGE_SIZE;
		if (!size_to_hstate(default_hstate_size))
			hugetlb_add_hstate(HUGETLB_PAGE_ORDER);
1876
	}
1877
	default_hstate_idx = hstate_index(size_to_hstate(default_hstate_size));
1878 1879
	if (default_hstate_max_huge_pages)
		default_hstate.max_huge_pages = default_hstate_max_huge_pages;
1880 1881 1882

	hugetlb_init_hstates();

1883 1884
	gather_bootmem_prealloc();

1885 1886 1887 1888
	report_hugepages();

	hugetlb_sysfs_init();

1889 1890
	hugetlb_register_all_nodes();

1891 1892 1893 1894 1895 1896 1897 1898
	return 0;
}
module_init(hugetlb_init);

/* Should be called on processing a hugepagesz=... option */
void __init hugetlb_add_hstate(unsigned order)
{
	struct hstate *h;
1899 1900
	unsigned long i;

1901 1902 1903 1904
	if (size_to_hstate(PAGE_SIZE << order)) {
		printk(KERN_WARNING "hugepagesz= specified twice, ignoring\n");
		return;
	}
1905
	BUG_ON(hugetlb_max_hstate >= HUGE_MAX_HSTATE);
1906
	BUG_ON(order == 0);
1907
	h = &hstates[hugetlb_max_hstate++];
1908 1909
	h->order = order;
	h->mask = ~((1ULL << (order + PAGE_SHIFT)) - 1);
1910 1911 1912 1913
	h->nr_huge_pages = 0;
	h->free_huge_pages = 0;
	for (i = 0; i < MAX_NUMNODES; ++i)
		INIT_LIST_HEAD(&h->hugepage_freelists[i]);
1914
	INIT_LIST_HEAD(&h->hugepage_activelist);
1915 1916
	h->next_nid_to_alloc = first_node(node_states[N_HIGH_MEMORY]);
	h->next_nid_to_free = first_node(node_states[N_HIGH_MEMORY]);
1917 1918
	snprintf(h->name, HSTATE_NAME_LEN, "hugepages-%lukB",
					huge_page_size(h)/1024);
1919

1920 1921 1922
	parsed_hstate = h;
}

1923
static int __init hugetlb_nrpages_setup(char *s)
1924 1925
{
	unsigned long *mhp;
1926
	static unsigned long *last_mhp;
1927 1928

	/*
1929
	 * !hugetlb_max_hstate means we haven't parsed a hugepagesz= parameter yet,
1930 1931
	 * so this hugepages= parameter goes to the "default hstate".
	 */
1932
	if (!hugetlb_max_hstate)
1933 1934 1935 1936
		mhp = &default_hstate_max_huge_pages;
	else
		mhp = &parsed_hstate->max_huge_pages;

1937 1938 1939 1940 1941 1942
	if (mhp == last_mhp) {
		printk(KERN_WARNING "hugepages= specified twice without "
			"interleaving hugepagesz=, ignoring\n");
		return 1;
	}

1943 1944 1945
	if (sscanf(s, "%lu", mhp) <= 0)
		*mhp = 0;

1946 1947 1948 1949 1950
	/*
	 * Global state is always initialized later in hugetlb_init.
	 * But we need to allocate >= MAX_ORDER hstates here early to still
	 * use the bootmem allocator.
	 */
1951
	if (hugetlb_max_hstate && parsed_hstate->order >= MAX_ORDER)
1952 1953 1954 1955
		hugetlb_hstate_alloc_pages(parsed_hstate);

	last_mhp = mhp;

1956 1957
	return 1;
}
1958 1959 1960 1961 1962 1963 1964 1965
__setup("hugepages=", hugetlb_nrpages_setup);

static int __init hugetlb_default_setup(char *s)
{
	default_hstate_size = memparse(s, &s);
	return 1;
}
__setup("default_hugepagesz=", hugetlb_default_setup);
1966

1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978
static unsigned int cpuset_mems_nr(unsigned int *array)
{
	int node;
	unsigned int nr = 0;

	for_each_node_mask(node, cpuset_current_mems_allowed)
		nr += array[node];

	return nr;
}

#ifdef CONFIG_SYSCTL
1979 1980 1981
static int hugetlb_sysctl_handler_common(bool obey_mempolicy,
			 struct ctl_table *table, int write,
			 void __user *buffer, size_t *length, loff_t *ppos)
L
Linus Torvalds 已提交
1982
{
1983 1984
	struct hstate *h = &default_hstate;
	unsigned long tmp;
1985
	int ret;
1986

1987
	tmp = h->max_huge_pages;
1988

1989 1990 1991
	if (write && h->order >= MAX_ORDER)
		return -EINVAL;

1992 1993
	table->data = &tmp;
	table->maxlen = sizeof(unsigned long);
1994 1995 1996
	ret = proc_doulongvec_minmax(table, write, buffer, length, ppos);
	if (ret)
		goto out;
1997

1998
	if (write) {
1999 2000
		NODEMASK_ALLOC(nodemask_t, nodes_allowed,
						GFP_KERNEL | __GFP_NORETRY);
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
		if (!(obey_mempolicy &&
			       init_nodemask_of_mempolicy(nodes_allowed))) {
			NODEMASK_FREE(nodes_allowed);
			nodes_allowed = &node_states[N_HIGH_MEMORY];
		}
		h->max_huge_pages = set_max_huge_pages(h, tmp, nodes_allowed);

		if (nodes_allowed != &node_states[N_HIGH_MEMORY])
			NODEMASK_FREE(nodes_allowed);
	}
2011 2012
out:
	return ret;
L
Linus Torvalds 已提交
2013
}
2014

2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031
int hugetlb_sysctl_handler(struct ctl_table *table, int write,
			  void __user *buffer, size_t *length, loff_t *ppos)
{

	return hugetlb_sysctl_handler_common(false, table, write,
							buffer, length, ppos);
}

#ifdef CONFIG_NUMA
int hugetlb_mempolicy_sysctl_handler(struct ctl_table *table, int write,
			  void __user *buffer, size_t *length, loff_t *ppos)
{
	return hugetlb_sysctl_handler_common(true, table, write,
							buffer, length, ppos);
}
#endif /* CONFIG_NUMA */

2032
int hugetlb_treat_movable_handler(struct ctl_table *table, int write,
2033
			void __user *buffer,
2034 2035
			size_t *length, loff_t *ppos)
{
2036
	proc_dointvec(table, write, buffer, length, ppos);
2037 2038 2039 2040 2041 2042 2043
	if (hugepages_treat_as_movable)
		htlb_alloc_mask = GFP_HIGHUSER_MOVABLE;
	else
		htlb_alloc_mask = GFP_HIGHUSER;
	return 0;
}

2044
int hugetlb_overcommit_handler(struct ctl_table *table, int write,
2045
			void __user *buffer,
2046 2047
			size_t *length, loff_t *ppos)
{
2048
	struct hstate *h = &default_hstate;
2049
	unsigned long tmp;
2050
	int ret;
2051

2052
	tmp = h->nr_overcommit_huge_pages;
2053

2054 2055 2056
	if (write && h->order >= MAX_ORDER)
		return -EINVAL;

2057 2058
	table->data = &tmp;
	table->maxlen = sizeof(unsigned long);
2059 2060 2061
	ret = proc_doulongvec_minmax(table, write, buffer, length, ppos);
	if (ret)
		goto out;
2062 2063 2064 2065 2066 2067

	if (write) {
		spin_lock(&hugetlb_lock);
		h->nr_overcommit_huge_pages = tmp;
		spin_unlock(&hugetlb_lock);
	}
2068 2069
out:
	return ret;
2070 2071
}

L
Linus Torvalds 已提交
2072 2073
#endif /* CONFIG_SYSCTL */

2074
void hugetlb_report_meminfo(struct seq_file *m)
L
Linus Torvalds 已提交
2075
{
2076
	struct hstate *h = &default_hstate;
2077
	seq_printf(m,
2078 2079 2080 2081 2082
			"HugePages_Total:   %5lu\n"
			"HugePages_Free:    %5lu\n"
			"HugePages_Rsvd:    %5lu\n"
			"HugePages_Surp:    %5lu\n"
			"Hugepagesize:   %8lu kB\n",
2083 2084 2085 2086 2087
			h->nr_huge_pages,
			h->free_huge_pages,
			h->resv_huge_pages,
			h->surplus_huge_pages,
			1UL << (huge_page_order(h) + PAGE_SHIFT - 10));
L
Linus Torvalds 已提交
2088 2089 2090 2091
}

int hugetlb_report_node_meminfo(int nid, char *buf)
{
2092
	struct hstate *h = &default_hstate;
L
Linus Torvalds 已提交
2093 2094
	return sprintf(buf,
		"Node %d HugePages_Total: %5u\n"
2095 2096
		"Node %d HugePages_Free:  %5u\n"
		"Node %d HugePages_Surp:  %5u\n",
2097 2098 2099
		nid, h->nr_huge_pages_node[nid],
		nid, h->free_huge_pages_node[nid],
		nid, h->surplus_huge_pages_node[nid]);
L
Linus Torvalds 已提交
2100 2101 2102 2103 2104
}

/* Return the number pages of memory we physically have, in PAGE_SIZE units. */
unsigned long hugetlb_total_pages(void)
{
2105 2106
	struct hstate *h = &default_hstate;
	return h->nr_huge_pages * pages_per_huge_page(h);
L
Linus Torvalds 已提交
2107 2108
}

2109
static int hugetlb_acct_memory(struct hstate *h, long delta)
M
Mel Gorman 已提交
2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131
{
	int ret = -ENOMEM;

	spin_lock(&hugetlb_lock);
	/*
	 * When cpuset is configured, it breaks the strict hugetlb page
	 * reservation as the accounting is done on a global variable. Such
	 * reservation is completely rubbish in the presence of cpuset because
	 * the reservation is not checked against page availability for the
	 * current cpuset. Application can still potentially OOM'ed by kernel
	 * with lack of free htlb page in cpuset that the task is in.
	 * Attempt to enforce strict accounting with cpuset is almost
	 * impossible (or too ugly) because cpuset is too fluid that
	 * task or memory node can be dynamically moved between cpusets.
	 *
	 * The change of semantics for shared hugetlb mapping with cpuset is
	 * undesirable. However, in order to preserve some of the semantics,
	 * we fall back to check against current free page availability as
	 * a best attempt and hopefully to minimize the impact of changing
	 * semantics that cpuset has.
	 */
	if (delta > 0) {
2132
		if (gather_surplus_pages(h, delta) < 0)
M
Mel Gorman 已提交
2133 2134
			goto out;

2135 2136
		if (delta > cpuset_mems_nr(h->free_huge_pages_node)) {
			return_unused_surplus_pages(h, delta);
M
Mel Gorman 已提交
2137 2138 2139 2140 2141 2142
			goto out;
		}
	}

	ret = 0;
	if (delta < 0)
2143
		return_unused_surplus_pages(h, (unsigned long) -delta);
M
Mel Gorman 已提交
2144 2145 2146 2147 2148 2149

out:
	spin_unlock(&hugetlb_lock);
	return ret;
}

2150 2151 2152 2153 2154 2155 2156 2157
static void hugetlb_vm_op_open(struct vm_area_struct *vma)
{
	struct resv_map *reservations = vma_resv_map(vma);

	/*
	 * This new VMA should share its siblings reservation map if present.
	 * The VMA will only ever have a valid reservation map pointer where
	 * it is being copied for another still existing VMA.  As that VMA
L
Lucas De Marchi 已提交
2158
	 * has a reference to the reservation map it cannot disappear until
2159 2160 2161 2162 2163 2164 2165
	 * after this open call completes.  It is therefore safe to take a
	 * new reference here without additional locking.
	 */
	if (reservations)
		kref_get(&reservations->refs);
}

2166 2167 2168 2169 2170 2171 2172 2173 2174
static void resv_map_put(struct vm_area_struct *vma)
{
	struct resv_map *reservations = vma_resv_map(vma);

	if (!reservations)
		return;
	kref_put(&reservations->refs, resv_map_release);
}

2175 2176
static void hugetlb_vm_op_close(struct vm_area_struct *vma)
{
2177
	struct hstate *h = hstate_vma(vma);
2178
	struct resv_map *reservations = vma_resv_map(vma);
2179
	struct hugepage_subpool *spool = subpool_vma(vma);
2180 2181 2182 2183 2184
	unsigned long reserve;
	unsigned long start;
	unsigned long end;

	if (reservations) {
2185 2186
		start = vma_hugecache_offset(h, vma, vma->vm_start);
		end = vma_hugecache_offset(h, vma, vma->vm_end);
2187 2188 2189 2190

		reserve = (end - start) -
			region_count(&reservations->regions, start, end);

2191
		resv_map_put(vma);
2192

2193
		if (reserve) {
2194
			hugetlb_acct_memory(h, -reserve);
2195
			hugepage_subpool_put_pages(spool, reserve);
2196
		}
2197
	}
2198 2199
}

L
Linus Torvalds 已提交
2200 2201 2202 2203 2204 2205
/*
 * We cannot handle pagefaults against hugetlb pages at all.  They cause
 * handle_mm_fault() to try to instantiate regular-sized pages in the
 * hugegpage VMA.  do_page_fault() is supposed to trap this, so BUG is we get
 * this far.
 */
N
Nick Piggin 已提交
2206
static int hugetlb_vm_op_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
L
Linus Torvalds 已提交
2207 2208
{
	BUG();
N
Nick Piggin 已提交
2209
	return 0;
L
Linus Torvalds 已提交
2210 2211
}

2212
const struct vm_operations_struct hugetlb_vm_ops = {
N
Nick Piggin 已提交
2213
	.fault = hugetlb_vm_op_fault,
2214
	.open = hugetlb_vm_op_open,
2215
	.close = hugetlb_vm_op_close,
L
Linus Torvalds 已提交
2216 2217
};

2218 2219
static pte_t make_huge_pte(struct vm_area_struct *vma, struct page *page,
				int writable)
D
David Gibson 已提交
2220 2221 2222
{
	pte_t entry;

2223
	if (writable) {
D
David Gibson 已提交
2224 2225 2226
		entry =
		    pte_mkwrite(pte_mkdirty(mk_pte(page, vma->vm_page_prot)));
	} else {
2227
		entry = huge_pte_wrprotect(mk_pte(page, vma->vm_page_prot));
D
David Gibson 已提交
2228 2229 2230
	}
	entry = pte_mkyoung(entry);
	entry = pte_mkhuge(entry);
2231
	entry = arch_make_huge_pte(entry, vma, page, writable);
D
David Gibson 已提交
2232 2233 2234 2235

	return entry;
}

2236 2237 2238 2239 2240
static void set_huge_ptep_writable(struct vm_area_struct *vma,
				   unsigned long address, pte_t *ptep)
{
	pte_t entry;

2241
	entry = pte_mkwrite(pte_mkdirty(huge_ptep_get(ptep)));
2242
	if (huge_ptep_set_access_flags(vma, address, ptep, entry, 1))
2243
		update_mmu_cache(vma, address, ptep);
2244 2245 2246
}


D
David Gibson 已提交
2247 2248 2249 2250 2251
int copy_hugetlb_page_range(struct mm_struct *dst, struct mm_struct *src,
			    struct vm_area_struct *vma)
{
	pte_t *src_pte, *dst_pte, entry;
	struct page *ptepage;
2252
	unsigned long addr;
2253
	int cow;
2254 2255
	struct hstate *h = hstate_vma(vma);
	unsigned long sz = huge_page_size(h);
2256 2257

	cow = (vma->vm_flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
D
David Gibson 已提交
2258

2259
	for (addr = vma->vm_start; addr < vma->vm_end; addr += sz) {
H
Hugh Dickins 已提交
2260 2261 2262
		src_pte = huge_pte_offset(src, addr);
		if (!src_pte)
			continue;
2263
		dst_pte = huge_pte_alloc(dst, addr, sz);
D
David Gibson 已提交
2264 2265
		if (!dst_pte)
			goto nomem;
2266 2267 2268 2269 2270

		/* If the pagetables are shared don't copy or take references */
		if (dst_pte == src_pte)
			continue;

H
Hugh Dickins 已提交
2271
		spin_lock(&dst->page_table_lock);
N
Nick Piggin 已提交
2272
		spin_lock_nested(&src->page_table_lock, SINGLE_DEPTH_NESTING);
2273
		if (!huge_pte_none(huge_ptep_get(src_pte))) {
2274
			if (cow)
2275 2276
				huge_ptep_set_wrprotect(src, addr, src_pte);
			entry = huge_ptep_get(src_pte);
2277 2278
			ptepage = pte_page(entry);
			get_page(ptepage);
2279
			page_dup_rmap(ptepage);
2280 2281 2282
			set_huge_pte_at(dst, addr, dst_pte, entry);
		}
		spin_unlock(&src->page_table_lock);
H
Hugh Dickins 已提交
2283
		spin_unlock(&dst->page_table_lock);
D
David Gibson 已提交
2284 2285 2286 2287 2288 2289 2290
	}
	return 0;

nomem:
	return -ENOMEM;
}

N
Naoya Horiguchi 已提交
2291 2292 2293 2294 2295 2296 2297
static int is_hugetlb_entry_migration(pte_t pte)
{
	swp_entry_t swp;

	if (huge_pte_none(pte) || pte_present(pte))
		return 0;
	swp = pte_to_swp_entry(pte);
2298
	if (non_swap_entry(swp) && is_migration_entry(swp))
N
Naoya Horiguchi 已提交
2299
		return 1;
2300
	else
N
Naoya Horiguchi 已提交
2301 2302 2303
		return 0;
}

2304 2305 2306 2307 2308 2309 2310
static int is_hugetlb_entry_hwpoisoned(pte_t pte)
{
	swp_entry_t swp;

	if (huge_pte_none(pte) || pte_present(pte))
		return 0;
	swp = pte_to_swp_entry(pte);
2311
	if (non_swap_entry(swp) && is_hwpoison_entry(swp))
2312
		return 1;
2313
	else
2314 2315 2316
		return 0;
}

2317 2318 2319
void __unmap_hugepage_range(struct mmu_gather *tlb, struct vm_area_struct *vma,
			    unsigned long start, unsigned long end,
			    struct page *ref_page)
D
David Gibson 已提交
2320
{
2321
	int force_flush = 0;
D
David Gibson 已提交
2322 2323
	struct mm_struct *mm = vma->vm_mm;
	unsigned long address;
2324
	pte_t *ptep;
D
David Gibson 已提交
2325 2326
	pte_t pte;
	struct page *page;
2327 2328 2329
	struct hstate *h = hstate_vma(vma);
	unsigned long sz = huge_page_size(h);

D
David Gibson 已提交
2330
	WARN_ON(!is_vm_hugetlb_page(vma));
2331 2332
	BUG_ON(start & ~huge_page_mask(h));
	BUG_ON(end & ~huge_page_mask(h));
D
David Gibson 已提交
2333

2334
	tlb_start_vma(tlb, vma);
A
Andrea Arcangeli 已提交
2335
	mmu_notifier_invalidate_range_start(mm, start, end);
2336
again:
2337
	spin_lock(&mm->page_table_lock);
2338
	for (address = start; address < end; address += sz) {
2339
		ptep = huge_pte_offset(mm, address);
A
Adam Litke 已提交
2340
		if (!ptep)
2341 2342
			continue;

2343 2344 2345
		if (huge_pmd_unshare(mm, &address, ptep))
			continue;

2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356
		pte = huge_ptep_get(ptep);
		if (huge_pte_none(pte))
			continue;

		/*
		 * HWPoisoned hugepage is already unmapped and dropped reference
		 */
		if (unlikely(is_hugetlb_entry_hwpoisoned(pte)))
			continue;

		page = pte_page(pte);
2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373
		/*
		 * If a reference page is supplied, it is because a specific
		 * page is being unmapped, not a range. Ensure the page we
		 * are about to unmap is the actual page of interest.
		 */
		if (ref_page) {
			if (page != ref_page)
				continue;

			/*
			 * Mark the VMA as having unmapped its page so that
			 * future faults in this VMA will fail rather than
			 * looking like data was lost
			 */
			set_vma_resv_flags(vma, HPAGE_RESV_UNMAPPED);
		}

2374
		pte = huge_ptep_get_and_clear(mm, address, ptep);
2375
		tlb_remove_tlb_entry(tlb, ptep, address);
2376 2377
		if (pte_dirty(pte))
			set_page_dirty(page);
2378

2379 2380 2381 2382
		page_remove_rmap(page);
		force_flush = !__tlb_remove_page(tlb, page);
		if (force_flush)
			break;
2383 2384 2385
		/* Bail out after unmapping reference page if supplied */
		if (ref_page)
			break;
D
David Gibson 已提交
2386
	}
2387
	spin_unlock(&mm->page_table_lock);
2388 2389 2390 2391 2392 2393 2394 2395 2396 2397
	/*
	 * mmu_gather ran out of room to batch pages, we break out of
	 * the PTE lock to avoid doing the potential expensive TLB invalidate
	 * and page-free while holding it.
	 */
	if (force_flush) {
		force_flush = 0;
		tlb_flush_mmu(tlb);
		if (address < end && !ref_page)
			goto again;
2398
	}
2399 2400
	mmu_notifier_invalidate_range_end(mm, start, end);
	tlb_end_vma(tlb, vma);
L
Linus Torvalds 已提交
2401
}
D
David Gibson 已提交
2402

2403
void unmap_hugepage_range(struct vm_area_struct *vma, unsigned long start,
2404
			  unsigned long end, struct page *ref_page)
2405
{
2406 2407 2408 2409 2410 2411 2412 2413
	struct mm_struct *mm;
	struct mmu_gather tlb;

	mm = vma->vm_mm;

	tlb_gather_mmu(&tlb, mm, 0);
	__unmap_hugepage_range(&tlb, vma, start, end, ref_page);
	tlb_finish_mmu(&tlb, start, end);
2414 2415
}

2416 2417 2418 2419 2420 2421
/*
 * This is called when the original mapper is failing to COW a MAP_PRIVATE
 * mappping it owns the reserve page for. The intention is to unmap the page
 * from other VMAs and let the children be SIGKILLed if they are faulting the
 * same region.
 */
2422 2423
static int unmap_ref_private(struct mm_struct *mm, struct vm_area_struct *vma,
				struct page *page, unsigned long address)
2424
{
2425
	struct hstate *h = hstate_vma(vma);
2426 2427 2428 2429 2430 2431 2432 2433 2434
	struct vm_area_struct *iter_vma;
	struct address_space *mapping;
	struct prio_tree_iter iter;
	pgoff_t pgoff;

	/*
	 * vm_pgoff is in PAGE_SIZE units, hence the different calculation
	 * from page cache lookup which is in HPAGE_SIZE units.
	 */
2435
	address = address & huge_page_mask(h);
2436
	pgoff = vma_hugecache_offset(h, vma, address);
2437
	mapping = vma->vm_file->f_dentry->d_inode->i_mapping;
2438

2439 2440 2441 2442 2443
	/*
	 * Take the mapping lock for the duration of the table walk. As
	 * this mapping should be shared between all the VMAs,
	 * __unmap_hugepage_range() is called as the lock is already held
	 */
2444
	mutex_lock(&mapping->i_mmap_mutex);
2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457
	vma_prio_tree_foreach(iter_vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
		/* Do not unmap the current VMA */
		if (iter_vma == vma)
			continue;

		/*
		 * Unmap the page from other VMAs without their own reserves.
		 * They get marked to be SIGKILLed if they fault in these
		 * areas. This is because a future no-page fault on this VMA
		 * could insert a zeroed page instead of the data existing
		 * from the time of fork. This would look like data corruption
		 */
		if (!is_vma_resv_set(iter_vma, HPAGE_RESV_OWNER))
2458 2459
			unmap_hugepage_range(iter_vma, address,
					     address + huge_page_size(h), page);
2460
	}
2461
	mutex_unlock(&mapping->i_mmap_mutex);
2462 2463 2464 2465

	return 1;
}

2466 2467
/*
 * Hugetlb_cow() should be called with page lock of the original hugepage held.
2468 2469 2470
 * Called with hugetlb_instantiation_mutex held and pte_page locked so we
 * cannot race with other handlers or page migration.
 * Keep the pte_same checks anyway to make transition from the mutex easier.
2471
 */
2472
static int hugetlb_cow(struct mm_struct *mm, struct vm_area_struct *vma,
2473 2474
			unsigned long address, pte_t *ptep, pte_t pte,
			struct page *pagecache_page)
2475
{
2476
	struct hstate *h = hstate_vma(vma);
2477
	struct page *old_page, *new_page;
2478
	int avoidcopy;
2479
	int outside_reserve = 0;
2480 2481 2482

	old_page = pte_page(pte);

2483
retry_avoidcopy:
2484 2485
	/* If no-one else is actually using this page, avoid the copy
	 * and just make the page writable */
2486
	avoidcopy = (page_mapcount(old_page) == 1);
2487
	if (avoidcopy) {
2488 2489
		if (PageAnon(old_page))
			page_move_anon_rmap(old_page, vma, address);
2490
		set_huge_ptep_writable(vma, address, ptep);
N
Nick Piggin 已提交
2491
		return 0;
2492 2493
	}

2494 2495 2496 2497 2498 2499 2500 2501 2502
	/*
	 * If the process that created a MAP_PRIVATE mapping is about to
	 * perform a COW due to a shared page count, attempt to satisfy
	 * the allocation without using the existing reserves. The pagecache
	 * page is used to determine if the reserve at this address was
	 * consumed or not. If reserves were used, a partial faulted mapping
	 * at the time of fork() could consume its reserves on COW instead
	 * of the full address range.
	 */
2503
	if (!(vma->vm_flags & VM_MAYSHARE) &&
2504 2505 2506 2507
			is_vma_resv_set(vma, HPAGE_RESV_OWNER) &&
			old_page != pagecache_page)
		outside_reserve = 1;

2508
	page_cache_get(old_page);
2509 2510 2511

	/* Drop page_table_lock as buddy allocator may be called */
	spin_unlock(&mm->page_table_lock);
2512
	new_page = alloc_huge_page(vma, address, outside_reserve);
2513

2514
	if (IS_ERR(new_page)) {
2515
		long err = PTR_ERR(new_page);
2516
		page_cache_release(old_page);
2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528

		/*
		 * If a process owning a MAP_PRIVATE mapping fails to COW,
		 * it is due to references held by a child and an insufficient
		 * huge page pool. To guarantee the original mappers
		 * reliability, unmap the page from child processes. The child
		 * may get SIGKILLed if it later faults.
		 */
		if (outside_reserve) {
			BUG_ON(huge_pte_none(pte));
			if (unmap_ref_private(mm, vma, old_page, address)) {
				BUG_ON(huge_pte_none(pte));
2529
				spin_lock(&mm->page_table_lock);
2530 2531 2532 2533 2534 2535 2536 2537
				ptep = huge_pte_offset(mm, address & huge_page_mask(h));
				if (likely(pte_same(huge_ptep_get(ptep), pte)))
					goto retry_avoidcopy;
				/*
				 * race occurs while re-acquiring page_table_lock, and
				 * our job is done.
				 */
				return 0;
2538 2539 2540 2541
			}
			WARN_ON_ONCE(1);
		}

2542 2543
		/* Caller expects lock to be held */
		spin_lock(&mm->page_table_lock);
2544 2545 2546 2547
		if (err == -ENOMEM)
			return VM_FAULT_OOM;
		else
			return VM_FAULT_SIGBUS;
2548 2549
	}

2550 2551 2552 2553
	/*
	 * When the original hugepage is shared one, it does not have
	 * anon_vma prepared.
	 */
2554
	if (unlikely(anon_vma_prepare(vma))) {
2555 2556
		page_cache_release(new_page);
		page_cache_release(old_page);
2557 2558
		/* Caller expects lock to be held */
		spin_lock(&mm->page_table_lock);
2559
		return VM_FAULT_OOM;
2560
	}
2561

A
Andrea Arcangeli 已提交
2562 2563
	copy_user_huge_page(new_page, old_page, address, vma,
			    pages_per_huge_page(h));
N
Nick Piggin 已提交
2564
	__SetPageUptodate(new_page);
2565

2566 2567 2568 2569 2570
	/*
	 * Retake the page_table_lock to check for racing updates
	 * before the page tables are altered
	 */
	spin_lock(&mm->page_table_lock);
2571
	ptep = huge_pte_offset(mm, address & huge_page_mask(h));
2572
	if (likely(pte_same(huge_ptep_get(ptep), pte))) {
2573
		/* Break COW */
2574 2575 2576
		mmu_notifier_invalidate_range_start(mm,
			address & huge_page_mask(h),
			(address & huge_page_mask(h)) + huge_page_size(h));
2577
		huge_ptep_clear_flush(vma, address, ptep);
2578 2579
		set_huge_pte_at(mm, address, ptep,
				make_huge_pte(vma, new_page, 1));
2580
		page_remove_rmap(old_page);
2581
		hugepage_add_new_anon_rmap(new_page, vma, address);
2582 2583
		/* Make the old page be freed below */
		new_page = old_page;
2584 2585 2586
		mmu_notifier_invalidate_range_end(mm,
			address & huge_page_mask(h),
			(address & huge_page_mask(h)) + huge_page_size(h));
2587 2588 2589
	}
	page_cache_release(new_page);
	page_cache_release(old_page);
N
Nick Piggin 已提交
2590
	return 0;
2591 2592
}

2593
/* Return the pagecache page at a given address within a VMA */
2594 2595
static struct page *hugetlbfs_pagecache_page(struct hstate *h,
			struct vm_area_struct *vma, unsigned long address)
2596 2597
{
	struct address_space *mapping;
2598
	pgoff_t idx;
2599 2600

	mapping = vma->vm_file->f_mapping;
2601
	idx = vma_hugecache_offset(h, vma, address);
2602 2603 2604 2605

	return find_lock_page(mapping, idx);
}

H
Hugh Dickins 已提交
2606 2607 2608 2609 2610
/*
 * Return whether there is a pagecache page to back given address within VMA.
 * Caller follow_hugetlb_page() holds page_table_lock so we cannot lock_page.
 */
static bool hugetlbfs_pagecache_present(struct hstate *h,
H
Hugh Dickins 已提交
2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625
			struct vm_area_struct *vma, unsigned long address)
{
	struct address_space *mapping;
	pgoff_t idx;
	struct page *page;

	mapping = vma->vm_file->f_mapping;
	idx = vma_hugecache_offset(h, vma, address);

	page = find_get_page(mapping, idx);
	if (page)
		put_page(page);
	return page != NULL;
}

2626
static int hugetlb_no_page(struct mm_struct *mm, struct vm_area_struct *vma,
2627
			unsigned long address, pte_t *ptep, unsigned int flags)
2628
{
2629
	struct hstate *h = hstate_vma(vma);
2630
	int ret = VM_FAULT_SIGBUS;
2631
	int anon_rmap = 0;
2632
	pgoff_t idx;
A
Adam Litke 已提交
2633 2634 2635
	unsigned long size;
	struct page *page;
	struct address_space *mapping;
2636
	pte_t new_pte;
A
Adam Litke 已提交
2637

2638 2639 2640
	/*
	 * Currently, we are forced to kill the process in the event the
	 * original mapper has unmapped pages from the child due to a failed
L
Lucas De Marchi 已提交
2641
	 * COW. Warn that such a situation has occurred as it may not be obvious
2642 2643 2644 2645 2646 2647 2648 2649
	 */
	if (is_vma_resv_set(vma, HPAGE_RESV_UNMAPPED)) {
		printk(KERN_WARNING
			"PID %d killed due to inadequate hugepage pool\n",
			current->pid);
		return ret;
	}

A
Adam Litke 已提交
2650
	mapping = vma->vm_file->f_mapping;
2651
	idx = vma_hugecache_offset(h, vma, address);
A
Adam Litke 已提交
2652 2653 2654 2655 2656

	/*
	 * Use page lock to guard against racing truncation
	 * before we get page_table_lock.
	 */
2657 2658 2659
retry:
	page = find_lock_page(mapping, idx);
	if (!page) {
2660
		size = i_size_read(mapping->host) >> huge_page_shift(h);
2661 2662
		if (idx >= size)
			goto out;
2663
		page = alloc_huge_page(vma, address, 0);
2664
		if (IS_ERR(page)) {
2665 2666 2667 2668 2669
			ret = PTR_ERR(page);
			if (ret == -ENOMEM)
				ret = VM_FAULT_OOM;
			else
				ret = VM_FAULT_SIGBUS;
2670 2671
			goto out;
		}
A
Andrea Arcangeli 已提交
2672
		clear_huge_page(page, address, pages_per_huge_page(h));
N
Nick Piggin 已提交
2673
		__SetPageUptodate(page);
2674

2675
		if (vma->vm_flags & VM_MAYSHARE) {
2676
			int err;
K
Ken Chen 已提交
2677
			struct inode *inode = mapping->host;
2678 2679 2680 2681 2682 2683 2684 2685

			err = add_to_page_cache(page, mapping, idx, GFP_KERNEL);
			if (err) {
				put_page(page);
				if (err == -EEXIST)
					goto retry;
				goto out;
			}
K
Ken Chen 已提交
2686 2687

			spin_lock(&inode->i_lock);
2688
			inode->i_blocks += blocks_per_huge_page(h);
K
Ken Chen 已提交
2689
			spin_unlock(&inode->i_lock);
2690
		} else {
2691
			lock_page(page);
2692 2693 2694 2695
			if (unlikely(anon_vma_prepare(vma))) {
				ret = VM_FAULT_OOM;
				goto backout_unlocked;
			}
2696
			anon_rmap = 1;
2697
		}
2698
	} else {
2699 2700 2701 2702 2703 2704
		/*
		 * If memory error occurs between mmap() and fault, some process
		 * don't have hwpoisoned swap entry for errored virtual address.
		 * So we need to block hugepage fault by PG_hwpoison bit check.
		 */
		if (unlikely(PageHWPoison(page))) {
2705
			ret = VM_FAULT_HWPOISON |
2706
				VM_FAULT_SET_HINDEX(hstate_index(h));
2707 2708
			goto backout_unlocked;
		}
2709
	}
2710

2711 2712 2713 2714 2715 2716
	/*
	 * If we are going to COW a private mapping later, we examine the
	 * pending reservations for this page now. This will ensure that
	 * any allocations necessary to record that reservation occur outside
	 * the spinlock.
	 */
2717
	if ((flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED))
2718 2719 2720 2721
		if (vma_needs_reservation(h, vma, address) < 0) {
			ret = VM_FAULT_OOM;
			goto backout_unlocked;
		}
2722

2723
	spin_lock(&mm->page_table_lock);
2724
	size = i_size_read(mapping->host) >> huge_page_shift(h);
A
Adam Litke 已提交
2725 2726 2727
	if (idx >= size)
		goto backout;

N
Nick Piggin 已提交
2728
	ret = 0;
2729
	if (!huge_pte_none(huge_ptep_get(ptep)))
A
Adam Litke 已提交
2730 2731
		goto backout;

2732 2733 2734 2735
	if (anon_rmap)
		hugepage_add_new_anon_rmap(page, vma, address);
	else
		page_dup_rmap(page);
2736 2737 2738 2739
	new_pte = make_huge_pte(vma, page, ((vma->vm_flags & VM_WRITE)
				&& (vma->vm_flags & VM_SHARED)));
	set_huge_pte_at(mm, address, ptep, new_pte);

2740
	if ((flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) {
2741
		/* Optimization, do the COW without a second fault */
2742
		ret = hugetlb_cow(mm, vma, address, ptep, new_pte, page);
2743 2744
	}

2745
	spin_unlock(&mm->page_table_lock);
A
Adam Litke 已提交
2746 2747
	unlock_page(page);
out:
2748
	return ret;
A
Adam Litke 已提交
2749 2750 2751

backout:
	spin_unlock(&mm->page_table_lock);
2752
backout_unlocked:
A
Adam Litke 已提交
2753 2754 2755
	unlock_page(page);
	put_page(page);
	goto out;
2756 2757
}

2758
int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2759
			unsigned long address, unsigned int flags)
2760 2761 2762
{
	pte_t *ptep;
	pte_t entry;
2763
	int ret;
2764
	struct page *page = NULL;
2765
	struct page *pagecache_page = NULL;
2766
	static DEFINE_MUTEX(hugetlb_instantiation_mutex);
2767
	struct hstate *h = hstate_vma(vma);
2768

2769 2770
	address &= huge_page_mask(h);

2771 2772 2773
	ptep = huge_pte_offset(mm, address);
	if (ptep) {
		entry = huge_ptep_get(ptep);
N
Naoya Horiguchi 已提交
2774 2775 2776 2777
		if (unlikely(is_hugetlb_entry_migration(entry))) {
			migration_entry_wait(mm, (pmd_t *)ptep, address);
			return 0;
		} else if (unlikely(is_hugetlb_entry_hwpoisoned(entry)))
2778
			return VM_FAULT_HWPOISON_LARGE |
2779
				VM_FAULT_SET_HINDEX(hstate_index(h));
2780 2781
	}

2782
	ptep = huge_pte_alloc(mm, address, huge_page_size(h));
2783 2784 2785
	if (!ptep)
		return VM_FAULT_OOM;

2786 2787 2788 2789 2790 2791
	/*
	 * Serialize hugepage allocation and instantiation, so that we don't
	 * get spurious allocation failures if two CPUs race to instantiate
	 * the same page in the page cache.
	 */
	mutex_lock(&hugetlb_instantiation_mutex);
2792 2793
	entry = huge_ptep_get(ptep);
	if (huge_pte_none(entry)) {
2794
		ret = hugetlb_no_page(mm, vma, address, ptep, flags);
2795
		goto out_mutex;
2796
	}
2797

N
Nick Piggin 已提交
2798
	ret = 0;
2799

2800 2801 2802 2803 2804 2805 2806 2807
	/*
	 * If we are going to COW the mapping later, we examine the pending
	 * reservations for this page now. This will ensure that any
	 * allocations necessary to record that reservation occur outside the
	 * spinlock. For private mappings, we also lookup the pagecache
	 * page now as it is used to determine if a reservation has been
	 * consumed.
	 */
2808
	if ((flags & FAULT_FLAG_WRITE) && !pte_write(entry)) {
2809 2810
		if (vma_needs_reservation(h, vma, address) < 0) {
			ret = VM_FAULT_OOM;
2811
			goto out_mutex;
2812
		}
2813

2814
		if (!(vma->vm_flags & VM_MAYSHARE))
2815 2816 2817 2818
			pagecache_page = hugetlbfs_pagecache_page(h,
								vma, address);
	}

2819 2820 2821 2822 2823 2824 2825 2826
	/*
	 * hugetlb_cow() requires page locks of pte_page(entry) and
	 * pagecache_page, so here we need take the former one
	 * when page != pagecache_page or !pagecache_page.
	 * Note that locking order is always pagecache_page -> page,
	 * so no worry about deadlock.
	 */
	page = pte_page(entry);
2827
	get_page(page);
2828
	if (page != pagecache_page)
2829 2830
		lock_page(page);

2831 2832
	spin_lock(&mm->page_table_lock);
	/* Check for a racing update before calling hugetlb_cow */
2833 2834 2835 2836
	if (unlikely(!pte_same(entry, huge_ptep_get(ptep))))
		goto out_page_table_lock;


2837
	if (flags & FAULT_FLAG_WRITE) {
2838
		if (!pte_write(entry)) {
2839 2840
			ret = hugetlb_cow(mm, vma, address, ptep, entry,
							pagecache_page);
2841 2842 2843 2844 2845
			goto out_page_table_lock;
		}
		entry = pte_mkdirty(entry);
	}
	entry = pte_mkyoung(entry);
2846 2847
	if (huge_ptep_set_access_flags(vma, address, ptep, entry,
						flags & FAULT_FLAG_WRITE))
2848
		update_mmu_cache(vma, address, ptep);
2849 2850

out_page_table_lock:
2851
	spin_unlock(&mm->page_table_lock);
2852 2853 2854 2855 2856

	if (pagecache_page) {
		unlock_page(pagecache_page);
		put_page(pagecache_page);
	}
2857 2858
	if (page != pagecache_page)
		unlock_page(page);
2859
	put_page(page);
2860

2861
out_mutex:
2862
	mutex_unlock(&hugetlb_instantiation_mutex);
2863 2864

	return ret;
2865 2866
}

A
Andi Kleen 已提交
2867 2868 2869 2870 2871 2872 2873 2874 2875
/* Can be overriden by architectures */
__attribute__((weak)) struct page *
follow_huge_pud(struct mm_struct *mm, unsigned long address,
	       pud_t *pud, int write)
{
	BUG();
	return NULL;
}

D
David Gibson 已提交
2876 2877
int follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma,
			struct page **pages, struct vm_area_struct **vmas,
2878
			unsigned long *position, int *length, int i,
H
Hugh Dickins 已提交
2879
			unsigned int flags)
D
David Gibson 已提交
2880
{
2881 2882
	unsigned long pfn_offset;
	unsigned long vaddr = *position;
D
David Gibson 已提交
2883
	int remainder = *length;
2884
	struct hstate *h = hstate_vma(vma);
D
David Gibson 已提交
2885

2886
	spin_lock(&mm->page_table_lock);
D
David Gibson 已提交
2887
	while (vaddr < vma->vm_end && remainder) {
A
Adam Litke 已提交
2888
		pte_t *pte;
H
Hugh Dickins 已提交
2889
		int absent;
A
Adam Litke 已提交
2890
		struct page *page;
D
David Gibson 已提交
2891

A
Adam Litke 已提交
2892 2893
		/*
		 * Some archs (sparc64, sh*) have multiple pte_ts to
H
Hugh Dickins 已提交
2894
		 * each hugepage.  We have to make sure we get the
A
Adam Litke 已提交
2895 2896
		 * first, for the page indexing below to work.
		 */
2897
		pte = huge_pte_offset(mm, vaddr & huge_page_mask(h));
H
Hugh Dickins 已提交
2898 2899 2900 2901
		absent = !pte || huge_pte_none(huge_ptep_get(pte));

		/*
		 * When coredumping, it suits get_dump_page if we just return
H
Hugh Dickins 已提交
2902 2903 2904 2905
		 * an error where there's an empty slot with no huge pagecache
		 * to back it.  This way, we avoid allocating a hugepage, and
		 * the sparse dumpfile avoids allocating disk blocks, but its
		 * huge holes still show up with zeroes where they need to be.
H
Hugh Dickins 已提交
2906
		 */
H
Hugh Dickins 已提交
2907 2908
		if (absent && (flags & FOLL_DUMP) &&
		    !hugetlbfs_pagecache_present(h, vma, vaddr)) {
H
Hugh Dickins 已提交
2909 2910 2911
			remainder = 0;
			break;
		}
D
David Gibson 已提交
2912

H
Hugh Dickins 已提交
2913 2914
		if (absent ||
		    ((flags & FOLL_WRITE) && !pte_write(huge_ptep_get(pte)))) {
A
Adam Litke 已提交
2915
			int ret;
D
David Gibson 已提交
2916

A
Adam Litke 已提交
2917
			spin_unlock(&mm->page_table_lock);
H
Hugh Dickins 已提交
2918 2919
			ret = hugetlb_fault(mm, vma, vaddr,
				(flags & FOLL_WRITE) ? FAULT_FLAG_WRITE : 0);
A
Adam Litke 已提交
2920
			spin_lock(&mm->page_table_lock);
2921
			if (!(ret & VM_FAULT_ERROR))
A
Adam Litke 已提交
2922
				continue;
D
David Gibson 已提交
2923

A
Adam Litke 已提交
2924 2925 2926 2927
			remainder = 0;
			break;
		}

2928
		pfn_offset = (vaddr & ~huge_page_mask(h)) >> PAGE_SHIFT;
2929
		page = pte_page(huge_ptep_get(pte));
2930
same_page:
2931
		if (pages) {
H
Hugh Dickins 已提交
2932
			pages[i] = mem_map_offset(page, pfn_offset);
K
KOSAKI Motohiro 已提交
2933
			get_page(pages[i]);
2934
		}
D
David Gibson 已提交
2935 2936 2937 2938 2939

		if (vmas)
			vmas[i] = vma;

		vaddr += PAGE_SIZE;
2940
		++pfn_offset;
D
David Gibson 已提交
2941 2942
		--remainder;
		++i;
2943
		if (vaddr < vma->vm_end && remainder &&
2944
				pfn_offset < pages_per_huge_page(h)) {
2945 2946 2947 2948 2949 2950
			/*
			 * We use pfn_offset to avoid touching the pageframes
			 * of this compound page.
			 */
			goto same_page;
		}
D
David Gibson 已提交
2951
	}
2952
	spin_unlock(&mm->page_table_lock);
D
David Gibson 已提交
2953 2954 2955
	*length = remainder;
	*position = vaddr;

H
Hugh Dickins 已提交
2956
	return i ? i : -EFAULT;
D
David Gibson 已提交
2957
}
2958 2959 2960 2961 2962 2963 2964 2965

void hugetlb_change_protection(struct vm_area_struct *vma,
		unsigned long address, unsigned long end, pgprot_t newprot)
{
	struct mm_struct *mm = vma->vm_mm;
	unsigned long start = address;
	pte_t *ptep;
	pte_t pte;
2966
	struct hstate *h = hstate_vma(vma);
2967 2968 2969 2970

	BUG_ON(address >= end);
	flush_cache_range(vma, address, end);

2971
	mutex_lock(&vma->vm_file->f_mapping->i_mmap_mutex);
2972
	spin_lock(&mm->page_table_lock);
2973
	for (; address < end; address += huge_page_size(h)) {
2974 2975 2976
		ptep = huge_pte_offset(mm, address);
		if (!ptep)
			continue;
2977 2978
		if (huge_pmd_unshare(mm, &address, ptep))
			continue;
2979
		if (!huge_pte_none(huge_ptep_get(ptep))) {
2980 2981 2982 2983 2984 2985
			pte = huge_ptep_get_and_clear(mm, address, ptep);
			pte = pte_mkhuge(pte_modify(pte, newprot));
			set_huge_pte_at(mm, address, ptep, pte);
		}
	}
	spin_unlock(&mm->page_table_lock);
2986
	mutex_unlock(&vma->vm_file->f_mapping->i_mmap_mutex);
2987 2988 2989 2990

	flush_tlb_range(vma, start, end);
}

2991 2992
int hugetlb_reserve_pages(struct inode *inode,
					long from, long to,
2993
					struct vm_area_struct *vma,
2994
					vm_flags_t vm_flags)
2995
{
2996
	long ret, chg;
2997
	struct hstate *h = hstate_inode(inode);
2998
	struct hugepage_subpool *spool = subpool_inode(inode);
2999

3000 3001 3002
	/*
	 * Only apply hugepage reservation if asked. At fault time, an
	 * attempt will be made for VM_NORESERVE to allocate a page
3003
	 * without using reserves
3004
	 */
3005
	if (vm_flags & VM_NORESERVE)
3006 3007
		return 0;

3008 3009 3010 3011 3012 3013
	/*
	 * Shared mappings base their reservation on the number of pages that
	 * are already allocated on behalf of the file. Private mappings need
	 * to reserve the full area even if read-only as mprotect() may be
	 * called to make the mapping read-write. Assume !vma is a shm mapping
	 */
3014
	if (!vma || vma->vm_flags & VM_MAYSHARE)
3015
		chg = region_chg(&inode->i_mapping->private_list, from, to);
3016 3017 3018 3019 3020
	else {
		struct resv_map *resv_map = resv_map_alloc();
		if (!resv_map)
			return -ENOMEM;

3021
		chg = to - from;
3022

3023 3024 3025 3026
		set_vma_resv_map(vma, resv_map);
		set_vma_resv_flags(vma, HPAGE_RESV_OWNER);
	}

3027 3028 3029 3030
	if (chg < 0) {
		ret = chg;
		goto out_err;
	}
3031

3032
	/* There must be enough pages in the subpool for the mapping */
3033 3034 3035 3036
	if (hugepage_subpool_get_pages(spool, chg)) {
		ret = -ENOSPC;
		goto out_err;
	}
3037 3038

	/*
3039
	 * Check enough hugepages are available for the reservation.
3040
	 * Hand the pages back to the subpool if there are not
3041
	 */
3042
	ret = hugetlb_acct_memory(h, chg);
K
Ken Chen 已提交
3043
	if (ret < 0) {
3044
		hugepage_subpool_put_pages(spool, chg);
3045
		goto out_err;
K
Ken Chen 已提交
3046
	}
3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058

	/*
	 * Account for the reservations made. Shared mappings record regions
	 * that have reservations as they are shared by multiple VMAs.
	 * When the last VMA disappears, the region map says how much
	 * the reservation was and the page cache tells how much of
	 * the reservation was consumed. Private mappings are per-VMA and
	 * only the consumed reservations are tracked. When the VMA
	 * disappears, the original reservation is the VMA size and the
	 * consumed reservations are stored in the map. Hence, nothing
	 * else has to be done for private mappings here
	 */
3059
	if (!vma || vma->vm_flags & VM_MAYSHARE)
3060
		region_add(&inode->i_mapping->private_list, from, to);
3061
	return 0;
3062
out_err:
3063 3064
	if (vma)
		resv_map_put(vma);
3065
	return ret;
3066 3067 3068 3069
}

void hugetlb_unreserve_pages(struct inode *inode, long offset, long freed)
{
3070
	struct hstate *h = hstate_inode(inode);
3071
	long chg = region_truncate(&inode->i_mapping->private_list, offset);
3072
	struct hugepage_subpool *spool = subpool_inode(inode);
K
Ken Chen 已提交
3073 3074

	spin_lock(&inode->i_lock);
3075
	inode->i_blocks -= (blocks_per_huge_page(h) * freed);
K
Ken Chen 已提交
3076 3077
	spin_unlock(&inode->i_lock);

3078
	hugepage_subpool_put_pages(spool, (chg - freed));
3079
	hugetlb_acct_memory(h, -(chg - freed));
3080
}
3081

3082 3083
#ifdef CONFIG_MEMORY_FAILURE

3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097
/* Should be called in hugetlb_lock */
static int is_hugepage_on_freelist(struct page *hpage)
{
	struct page *page;
	struct page *tmp;
	struct hstate *h = page_hstate(hpage);
	int nid = page_to_nid(hpage);

	list_for_each_entry_safe(page, tmp, &h->hugepage_freelists[nid], lru)
		if (page == hpage)
			return 1;
	return 0;
}

3098 3099 3100 3101
/*
 * This function is called from memory failure code.
 * Assume the caller holds page lock of the head page.
 */
3102
int dequeue_hwpoisoned_huge_page(struct page *hpage)
3103 3104 3105
{
	struct hstate *h = page_hstate(hpage);
	int nid = page_to_nid(hpage);
3106
	int ret = -EBUSY;
3107 3108

	spin_lock(&hugetlb_lock);
3109 3110
	if (is_hugepage_on_freelist(hpage)) {
		list_del(&hpage->lru);
3111
		set_page_refcounted(hpage);
3112 3113 3114 3115
		h->free_huge_pages--;
		h->free_huge_pages_node[nid]--;
		ret = 0;
	}
3116
	spin_unlock(&hugetlb_lock);
3117
	return ret;
3118
}
3119
#endif