1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 *  linux/mm/swap_state.c
  4 *
  5 *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
  6 *  Swap reorganised 29.12.95, Stephen Tweedie
  7 *
  8 *  Rewritten to use page cache, (C) 1998 Stephen Tweedie
  9 */
 10#include <linux/mm.h>
 11#include <linux/gfp.h>
 12#include <linux/kernel_stat.h>
 13#include <linux/swap.h>
 14#include <linux/swapops.h>
 15#include <linux/init.h>
 16#include <linux/pagemap.h>
 17#include <linux/backing-dev.h>
 18#include <linux/blkdev.h>
 19#include <linux/pagevec.h>
 20#include <linux/migrate.h>
 21#include <linux/vmalloc.h>
 22#include <linux/swap_slots.h>
 23#include <linux/huge_mm.h>
 24
 25#include <asm/pgtable.h>
 26
 27/*
 28 * swapper_space is a fiction, retained to simplify the path through
 29 * vmscan's shrink_page_list.
 30 */
 31static const struct address_space_operations swap_aops = {
 32	.writepage	= swap_writepage,
 33	.set_page_dirty	= swap_set_page_dirty,
 34#ifdef CONFIG_MIGRATION
 35	.migratepage	= migrate_page,
 36#endif
 37};
 38
 39struct address_space *swapper_spaces[MAX_SWAPFILES] __read_mostly;
 40static unsigned int nr_swapper_spaces[MAX_SWAPFILES] __read_mostly;
 41static bool enable_vma_readahead __read_mostly = true;
 42
 43#define SWAP_RA_WIN_SHIFT	(PAGE_SHIFT / 2)
 44#define SWAP_RA_HITS_MASK	((1UL << SWAP_RA_WIN_SHIFT) - 1)
 45#define SWAP_RA_HITS_MAX	SWAP_RA_HITS_MASK
 46#define SWAP_RA_WIN_MASK	(~PAGE_MASK & ~SWAP_RA_HITS_MASK)
 47
 48#define SWAP_RA_HITS(v)		((v) & SWAP_RA_HITS_MASK)
 49#define SWAP_RA_WIN(v)		(((v) & SWAP_RA_WIN_MASK) >> SWAP_RA_WIN_SHIFT)
 50#define SWAP_RA_ADDR(v)		((v) & PAGE_MASK)
 51
 52#define SWAP_RA_VAL(addr, win, hits)				\
 53	(((addr) & PAGE_MASK) |					\
 54	 (((win) << SWAP_RA_WIN_SHIFT) & SWAP_RA_WIN_MASK) |	\
 55	 ((hits) & SWAP_RA_HITS_MASK))
 56
 57/* Initial readahead hits is 4 to start up with a small window */
 58#define GET_SWAP_RA_VAL(vma)					\
 59	(atomic_long_read(&(vma)->swap_readahead_info) ? : 4)
 60
 61#define INC_CACHE_INFO(x)	do { swap_cache_info.x++; } while (0)
 62#define ADD_CACHE_INFO(x, nr)	do { swap_cache_info.x += (nr); } while (0)
 63
 64static struct {
 65	unsigned long add_total;
 66	unsigned long del_total;
 67	unsigned long find_success;
 68	unsigned long find_total;
 69} swap_cache_info;
 70
 71unsigned long total_swapcache_pages(void)
 72{
 73	unsigned int i, j, nr;
 74	unsigned long ret = 0;
 75	struct address_space *spaces;
 76	struct swap_info_struct *si;
 77
 78	for (i = 0; i < MAX_SWAPFILES; i++) {
 79		swp_entry_t entry = swp_entry(i, 1);
 80
 81		/* Avoid get_swap_device() to warn for bad swap entry */
 82		if (!swp_swap_info(entry))
 83			continue;
 84		/* Prevent swapoff to free swapper_spaces */
 85		si = get_swap_device(entry);
 86		if (!si)
 87			continue;
 88		nr = nr_swapper_spaces[i];
 89		spaces = swapper_spaces[i];
 90		for (j = 0; j < nr; j++)
 91			ret += spaces[j].nrpages;
 92		put_swap_device(si);
 93	}
 94	return ret;
 95}
 96
 97static atomic_t swapin_readahead_hits = ATOMIC_INIT(4);
 98
 99void show_swap_cache_info(void)
100{
101	printk("%lu pages in swap cache\n", total_swapcache_pages());
102	printk("Swap cache stats: add %lu, delete %lu, find %lu/%lu\n",
103		swap_cache_info.add_total, swap_cache_info.del_total,
104		swap_cache_info.find_success, swap_cache_info.find_total);
105	printk("Free swap  = %ldkB\n",
106		get_nr_swap_pages() << (PAGE_SHIFT - 10));
107	printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10));
108}
109
110/*
111 * add_to_swap_cache resembles add_to_page_cache_locked on swapper_space,
112 * but sets SwapCache flag and private instead of mapping and index.
113 */
114int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp)
115{
116	struct address_space *address_space = swap_address_space(entry);
117	pgoff_t idx = swp_offset(entry);
118	XA_STATE_ORDER(xas, &address_space->i_pages, idx, compound_order(page));
119	unsigned long i, nr = compound_nr(page);
120
121	VM_BUG_ON_PAGE(!PageLocked(page), page);
122	VM_BUG_ON_PAGE(PageSwapCache(page), page);
123	VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
124
125	page_ref_add(page, nr);
126	SetPageSwapCache(page);
 
127
128	do {
129		xas_lock_irq(&xas);
130		xas_create_range(&xas);
131		if (xas_error(&xas))
132			goto unlock;
133		for (i = 0; i < nr; i++) {
134			VM_BUG_ON_PAGE(xas.xa_index != idx + i, page);
135			set_page_private(page + i, entry.val + i);
136			xas_store(&xas, page);
137			xas_next(&xas);
138		}
139		address_space->nrpages += nr;
140		__mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
141		ADD_CACHE_INFO(add_total, nr);
142unlock:
143		xas_unlock_irq(&xas);
144	} while (xas_nomem(&xas, gfp));
145
146	if (!xas_error(&xas))
147		return 0;
 
 
 
 
 
 
 
 
 
148
149	ClearPageSwapCache(page);
150	page_ref_sub(page, nr);
151	return xas_error(&xas);
 
 
 
 
 
 
 
 
 
 
 
152}
153
154/*
155 * This must be called only on pages that have
156 * been verified to be in the swap cache.
157 */
158void __delete_from_swap_cache(struct page *page, swp_entry_t entry)
159{
160	struct address_space *address_space = swap_address_space(entry);
161	int i, nr = hpage_nr_pages(page);
162	pgoff_t idx = swp_offset(entry);
163	XA_STATE(xas, &address_space->i_pages, idx);
164
165	VM_BUG_ON_PAGE(!PageLocked(page), page);
166	VM_BUG_ON_PAGE(!PageSwapCache(page), page);
167	VM_BUG_ON_PAGE(PageWriteback(page), page);
168
169	for (i = 0; i < nr; i++) {
170		void *entry = xas_store(&xas, NULL);
171		VM_BUG_ON_PAGE(entry != page, entry);
172		set_page_private(page + i, 0);
173		xas_next(&xas);
174	}
175	ClearPageSwapCache(page);
176	address_space->nrpages -= nr;
177	__mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
178	ADD_CACHE_INFO(del_total, nr);
179}
180
181/**
182 * add_to_swap - allocate swap space for a page
183 * @page: page we want to move to swap
184 *
185 * Allocate swap space for the page and add the page to the
186 * swap cache.  Caller needs to hold the page lock. 
187 */
188int add_to_swap(struct page *page)
189{
190	swp_entry_t entry;
191	int err;
192
193	VM_BUG_ON_PAGE(!PageLocked(page), page);
194	VM_BUG_ON_PAGE(!PageUptodate(page), page);
195
196	entry = get_swap_page(page);
197	if (!entry.val)
198		return 0;
199
 
 
 
 
 
 
 
 
 
 
 
200	/*
201	 * XArray node allocations from PF_MEMALLOC contexts could
202	 * completely exhaust the page allocator. __GFP_NOMEMALLOC
203	 * stops emergency reserves from being allocated.
204	 *
205	 * TODO: this could cause a theoretical memory reclaim
206	 * deadlock in the swap out path.
207	 */
208	/*
209	 * Add it to the swap cache.
210	 */
211	err = add_to_swap_cache(page, entry,
212			__GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN);
213	if (err)
 
 
 
214		/*
215		 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
216		 * clear SWAP_HAS_CACHE flag.
217		 */
218		goto fail;
219	/*
220	 * Normally the page will be dirtied in unmap because its pte should be
221	 * dirty. A special case is MADV_FREE page. The page'e pte could have
222	 * dirty bit cleared but the page's SwapBacked bit is still set because
223	 * clearing the dirty bit and SwapBacked bit has no lock protected. For
224	 * such page, unmap will not set dirty bit for it, so page reclaim will
225	 * not write the page out. This can cause data corruption when the page
226	 * is swap in later. Always setting the dirty bit for the page solves
227	 * the problem.
228	 */
229	set_page_dirty(page);
230
231	return 1;
232
233fail:
234	put_swap_page(page, entry);
235	return 0;
236}
237
238/*
239 * This must be called only on pages that have
240 * been verified to be in the swap cache and locked.
241 * It will never put the page into the free list,
242 * the caller has a reference on the page.
243 */
244void delete_from_swap_cache(struct page *page)
245{
246	swp_entry_t entry = { .val = page_private(page) };
247	struct address_space *address_space = swap_address_space(entry);
 
 
248
249	xa_lock_irq(&address_space->i_pages);
250	__delete_from_swap_cache(page, entry);
251	xa_unlock_irq(&address_space->i_pages);
 
252
253	put_swap_page(page, entry);
254	page_ref_sub(page, hpage_nr_pages(page));
255}
256
257/* 
258 * If we are the only user, then try to free up the swap cache. 
259 * 
260 * Its ok to check for PageSwapCache without the page lock
261 * here because we are going to recheck again inside
262 * try_to_free_swap() _with_ the lock.
263 * 					- Marcelo
264 */
265static inline void free_swap_cache(struct page *page)
266{
267	if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) {
268		try_to_free_swap(page);
269		unlock_page(page);
270	}
271}
272
273/* 
274 * Perform a free_page(), also freeing any swap cache associated with
275 * this page if it is the last user of the page.
276 */
277void free_page_and_swap_cache(struct page *page)
278{
279	free_swap_cache(page);
280	if (!is_huge_zero_page(page))
281		put_page(page);
282}
283
284/*
285 * Passed an array of pages, drop them all from swapcache and then release
286 * them.  They are removed from the LRU and freed if this is their last use.
287 */
288void free_pages_and_swap_cache(struct page **pages, int nr)
289{
290	struct page **pagep = pages;
291	int i;
292
293	lru_add_drain();
294	for (i = 0; i < nr; i++)
295		free_swap_cache(pagep[i]);
296	release_pages(pagep, nr);
297}
298
299static inline bool swap_use_vma_readahead(void)
300{
301	return READ_ONCE(enable_vma_readahead) && !atomic_read(&nr_rotate_swap);
302}
303
304/*
305 * Lookup a swap entry in the swap cache. A found page will be returned
306 * unlocked and with its refcount incremented - we rely on the kernel
307 * lock getting page table operations atomic even if we drop the page
308 * lock before returning.
309 */
310struct page *lookup_swap_cache(swp_entry_t entry, struct vm_area_struct *vma,
311			       unsigned long addr)
312{
313	struct page *page;
314	struct swap_info_struct *si;
315
316	si = get_swap_device(entry);
317	if (!si)
318		return NULL;
319	page = find_get_page(swap_address_space(entry), swp_offset(entry));
320	put_swap_device(si);
321
322	INC_CACHE_INFO(find_total);
323	if (page) {
324		bool vma_ra = swap_use_vma_readahead();
325		bool readahead;
326
327		INC_CACHE_INFO(find_success);
328		/*
329		 * At the moment, we don't support PG_readahead for anon THP
330		 * so let's bail out rather than confusing the readahead stat.
331		 */
332		if (unlikely(PageTransCompound(page)))
333			return page;
334
335		readahead = TestClearPageReadahead(page);
336		if (vma && vma_ra) {
337			unsigned long ra_val;
338			int win, hits;
339
340			ra_val = GET_SWAP_RA_VAL(vma);
341			win = SWAP_RA_WIN(ra_val);
342			hits = SWAP_RA_HITS(ra_val);
343			if (readahead)
344				hits = min_t(int, hits + 1, SWAP_RA_HITS_MAX);
345			atomic_long_set(&vma->swap_readahead_info,
346					SWAP_RA_VAL(addr, win, hits));
347		}
348
349		if (readahead) {
350			count_vm_event(SWAP_RA_HIT);
351			if (!vma || !vma_ra)
352				atomic_inc(&swapin_readahead_hits);
353		}
354	}
355
 
356	return page;
357}
358
359struct page *__read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
360			struct vm_area_struct *vma, unsigned long addr,
361			bool *new_page_allocated)
362{
363	struct page *found_page = NULL, *new_page = NULL;
364	struct swap_info_struct *si;
365	int err;
366	*new_page_allocated = false;
367
368	do {
369		/*
370		 * First check the swap cache.  Since this is normally
371		 * called after lookup_swap_cache() failed, re-calling
372		 * that would confuse statistics.
373		 */
374		si = get_swap_device(entry);
375		if (!si)
376			break;
377		found_page = find_get_page(swap_address_space(entry),
378					   swp_offset(entry));
379		put_swap_device(si);
380		if (found_page)
381			break;
382
383		/*
384		 * Just skip read ahead for unused swap slot.
385		 * During swap_off when swap_slot_cache is disabled,
386		 * we have to handle the race between putting
387		 * swap entry in swap cache and marking swap slot
388		 * as SWAP_HAS_CACHE.  That's done in later part of code or
389		 * else swap_off will be aborted if we return NULL.
390		 */
391		if (!__swp_swapcount(entry) && swap_slot_cache_enabled)
392			break;
393
394		/*
395		 * Get a new page to read into from swap.
396		 */
397		if (!new_page) {
398			new_page = alloc_page_vma(gfp_mask, vma, addr);
399			if (!new_page)
400				break;		/* Out of memory */
401		}
402
403		/*
 
 
 
 
 
 
 
404		 * Swap entry may have been freed since our caller observed it.
405		 */
406		err = swapcache_prepare(entry);
407		if (err == -EEXIST) {
 
408			/*
409			 * We might race against get_swap_page() and stumble
410			 * across a SWAP_HAS_CACHE swap_map entry whose page
411			 * has not been brought into the swapcache yet.
 
 
 
 
 
 
 
 
 
 
412			 */
413			cond_resched();
414			continue;
415		} else if (err)		/* swp entry is obsolete ? */
 
 
416			break;
 
417
418		/* May fail (-ENOMEM) if XArray node allocation failed. */
419		__SetPageLocked(new_page);
420		__SetPageSwapBacked(new_page);
421		err = add_to_swap_cache(new_page, entry, gfp_mask & GFP_KERNEL);
422		if (likely(!err)) {
423			/* Initiate read into locked page */
424			SetPageWorkingset(new_page);
 
 
425			lru_cache_add_anon(new_page);
426			*new_page_allocated = true;
427			return new_page;
428		}
 
429		__ClearPageLocked(new_page);
430		/*
431		 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
432		 * clear SWAP_HAS_CACHE flag.
433		 */
434		put_swap_page(new_page, entry);
435	} while (err != -ENOMEM);
436
437	if (new_page)
438		put_page(new_page);
439	return found_page;
440}
441
442/*
443 * Locate a page of swap in physical memory, reserving swap cache space
444 * and reading the disk if it is not already cached.
445 * A failure return means that either the page allocation failed or that
446 * the swap entry is no longer in use.
447 */
448struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
449		struct vm_area_struct *vma, unsigned long addr, bool do_poll)
450{
451	bool page_was_allocated;
452	struct page *retpage = __read_swap_cache_async(entry, gfp_mask,
453			vma, addr, &page_was_allocated);
454
455	if (page_was_allocated)
456		swap_readpage(retpage, do_poll);
457
458	return retpage;
459}
460
461static unsigned int __swapin_nr_pages(unsigned long prev_offset,
462				      unsigned long offset,
463				      int hits,
464				      int max_pages,
465				      int prev_win)
466{
467	unsigned int pages, last_ra;
 
 
 
 
 
 
468
469	/*
470	 * This heuristic has been found to work well on both sequential and
471	 * random loads, swapping to hard disk or to SSD: please don't ask
472	 * what the "+ 2" means, it just happens to work well, that's all.
473	 */
474	pages = hits + 2;
475	if (pages == 2) {
476		/*
477		 * We can have no readahead hits to judge by: but must not get
478		 * stuck here forever, so check for an adjacent offset instead
479		 * (and don't even bother to check whether swap type is same).
480		 */
481		if (offset != prev_offset + 1 && offset != prev_offset - 1)
482			pages = 1;
 
483	} else {
484		unsigned int roundup = 4;
485		while (roundup < pages)
486			roundup <<= 1;
487		pages = roundup;
488	}
489
490	if (pages > max_pages)
491		pages = max_pages;
492
493	/* Don't shrink readahead too fast */
494	last_ra = prev_win / 2;
495	if (pages < last_ra)
496		pages = last_ra;
497
498	return pages;
499}
500
501static unsigned long swapin_nr_pages(unsigned long offset)
502{
503	static unsigned long prev_offset;
504	unsigned int hits, pages, max_pages;
505	static atomic_t last_readahead_pages;
506
507	max_pages = 1 << READ_ONCE(page_cluster);
508	if (max_pages <= 1)
509		return 1;
510
511	hits = atomic_xchg(&swapin_readahead_hits, 0);
512	pages = __swapin_nr_pages(prev_offset, offset, hits, max_pages,
513				  atomic_read(&last_readahead_pages));
514	if (!hits)
515		prev_offset = offset;
516	atomic_set(&last_readahead_pages, pages);
517
518	return pages;
519}
520
521/**
522 * swap_cluster_readahead - swap in pages in hope we need them soon
523 * @entry: swap entry of this memory
524 * @gfp_mask: memory allocation flags
525 * @vmf: fault information
 
526 *
527 * Returns the struct page for entry and addr, after queueing swapin.
528 *
529 * Primitive swap readahead code. We simply read an aligned block of
530 * (1 << page_cluster) entries in the swap area. This method is chosen
531 * because it doesn't cost us any seek time.  We also make sure to queue
532 * the 'original' request together with the readahead ones...
533 *
534 * This has been extended to use the NUMA policies from the mm triggering
535 * the readahead.
536 *
537 * Caller must hold read mmap_sem if vmf->vma is not NULL.
538 */
539struct page *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask,
540				struct vm_fault *vmf)
541{
542	struct page *page;
543	unsigned long entry_offset = swp_offset(entry);
544	unsigned long offset = entry_offset;
545	unsigned long start_offset, end_offset;
546	unsigned long mask;
547	struct swap_info_struct *si = swp_swap_info(entry);
548	struct blk_plug plug;
549	bool do_poll = true, page_allocated;
550	struct vm_area_struct *vma = vmf->vma;
551	unsigned long addr = vmf->address;
552
553	mask = swapin_nr_pages(offset) - 1;
554	if (!mask)
555		goto skip;
556
557	/* Test swap type to make sure the dereference is safe */
558	if (likely(si->flags & (SWP_BLKDEV | SWP_FS))) {
559		struct inode *inode = si->swap_file->f_mapping->host;
560		if (inode_read_congested(inode))
561			goto skip;
562	}
563
564	do_poll = false;
565	/* Read a page_cluster sized and aligned cluster around offset. */
566	start_offset = offset & ~mask;
567	end_offset = offset | mask;
568	if (!start_offset)	/* First page is swap header. */
569		start_offset++;
570	if (end_offset >= si->max)
571		end_offset = si->max - 1;
572
573	blk_start_plug(&plug);
574	for (offset = start_offset; offset <= end_offset ; offset++) {
575		/* Ok, do the async read-ahead now */
576		page = __read_swap_cache_async(
577			swp_entry(swp_type(entry), offset),
578			gfp_mask, vma, addr, &page_allocated);
579		if (!page)
580			continue;
581		if (page_allocated) {
582			swap_readpage(page, false);
583			if (offset != entry_offset) {
584				SetPageReadahead(page);
585				count_vm_event(SWAP_RA);
586			}
587		}
588		put_page(page);
589	}
590	blk_finish_plug(&plug);
591
592	lru_add_drain();	/* Push any new pages onto the LRU now */
593skip:
594	return read_swap_cache_async(entry, gfp_mask, vma, addr, do_poll);
595}
596
597int init_swap_address_space(unsigned int type, unsigned long nr_pages)
598{
599	struct address_space *spaces, *space;
600	unsigned int i, nr;
601
602	nr = DIV_ROUND_UP(nr_pages, SWAP_ADDRESS_SPACE_PAGES);
603	spaces = kvcalloc(nr, sizeof(struct address_space), GFP_KERNEL);
604	if (!spaces)
605		return -ENOMEM;
606	for (i = 0; i < nr; i++) {
607		space = spaces + i;
608		xa_init_flags(&space->i_pages, XA_FLAGS_LOCK_IRQ);
609		atomic_set(&space->i_mmap_writable, 0);
610		space->a_ops = &swap_aops;
611		/* swap cache doesn't use writeback related tags */
612		mapping_set_no_writeback_tags(space);
613	}
614	nr_swapper_spaces[type] = nr;
615	swapper_spaces[type] = spaces;
616
617	return 0;
618}
619
620void exit_swap_address_space(unsigned int type)
621{
622	kvfree(swapper_spaces[type]);
623	nr_swapper_spaces[type] = 0;
624	swapper_spaces[type] = NULL;
625}
626
627static inline void swap_ra_clamp_pfn(struct vm_area_struct *vma,
628				     unsigned long faddr,
629				     unsigned long lpfn,
630				     unsigned long rpfn,
631				     unsigned long *start,
632				     unsigned long *end)
633{
634	*start = max3(lpfn, PFN_DOWN(vma->vm_start),
635		      PFN_DOWN(faddr & PMD_MASK));
636	*end = min3(rpfn, PFN_DOWN(vma->vm_end),
637		    PFN_DOWN((faddr & PMD_MASK) + PMD_SIZE));
638}
639
640static void swap_ra_info(struct vm_fault *vmf,
641			struct vma_swap_readahead *ra_info)
642{
643	struct vm_area_struct *vma = vmf->vma;
644	unsigned long ra_val;
645	swp_entry_t entry;
646	unsigned long faddr, pfn, fpfn;
647	unsigned long start, end;
648	pte_t *pte, *orig_pte;
649	unsigned int max_win, hits, prev_win, win, left;
650#ifndef CONFIG_64BIT
651	pte_t *tpte;
652#endif
653
654	max_win = 1 << min_t(unsigned int, READ_ONCE(page_cluster),
655			     SWAP_RA_ORDER_CEILING);
656	if (max_win == 1) {
657		ra_info->win = 1;
658		return;
659	}
660
661	faddr = vmf->address;
662	orig_pte = pte = pte_offset_map(vmf->pmd, faddr);
663	entry = pte_to_swp_entry(*pte);
664	if ((unlikely(non_swap_entry(entry)))) {
665		pte_unmap(orig_pte);
666		return;
667	}
668
669	fpfn = PFN_DOWN(faddr);
670	ra_val = GET_SWAP_RA_VAL(vma);
671	pfn = PFN_DOWN(SWAP_RA_ADDR(ra_val));
672	prev_win = SWAP_RA_WIN(ra_val);
673	hits = SWAP_RA_HITS(ra_val);
674	ra_info->win = win = __swapin_nr_pages(pfn, fpfn, hits,
675					       max_win, prev_win);
676	atomic_long_set(&vma->swap_readahead_info,
677			SWAP_RA_VAL(faddr, win, 0));
678
679	if (win == 1) {
680		pte_unmap(orig_pte);
681		return;
682	}
683
684	/* Copy the PTEs because the page table may be unmapped */
685	if (fpfn == pfn + 1)
686		swap_ra_clamp_pfn(vma, faddr, fpfn, fpfn + win, &start, &end);
687	else if (pfn == fpfn + 1)
688		swap_ra_clamp_pfn(vma, faddr, fpfn - win + 1, fpfn + 1,
689				  &start, &end);
690	else {
691		left = (win - 1) / 2;
692		swap_ra_clamp_pfn(vma, faddr, fpfn - left, fpfn + win - left,
693				  &start, &end);
694	}
695	ra_info->nr_pte = end - start;
696	ra_info->offset = fpfn - start;
697	pte -= ra_info->offset;
698#ifdef CONFIG_64BIT
699	ra_info->ptes = pte;
700#else
701	tpte = ra_info->ptes;
702	for (pfn = start; pfn != end; pfn++)
703		*tpte++ = *pte++;
704#endif
705	pte_unmap(orig_pte);
706}
707
708/**
709 * swap_vma_readahead - swap in pages in hope we need them soon
710 * @entry: swap entry of this memory
711 * @gfp_mask: memory allocation flags
712 * @vmf: fault information
713 *
714 * Returns the struct page for entry and addr, after queueing swapin.
715 *
716 * Primitive swap readahead code. We simply read in a few pages whoes
717 * virtual addresses are around the fault address in the same vma.
718 *
719 * Caller must hold read mmap_sem if vmf->vma is not NULL.
720 *
721 */
722static struct page *swap_vma_readahead(swp_entry_t fentry, gfp_t gfp_mask,
723				       struct vm_fault *vmf)
724{
725	struct blk_plug plug;
726	struct vm_area_struct *vma = vmf->vma;
727	struct page *page;
728	pte_t *pte, pentry;
729	swp_entry_t entry;
730	unsigned int i;
731	bool page_allocated;
732	struct vma_swap_readahead ra_info = {0,};
733
734	swap_ra_info(vmf, &ra_info);
735	if (ra_info.win == 1)
736		goto skip;
737
738	blk_start_plug(&plug);
739	for (i = 0, pte = ra_info.ptes; i < ra_info.nr_pte;
740	     i++, pte++) {
741		pentry = *pte;
742		if (pte_none(pentry))
743			continue;
744		if (pte_present(pentry))
745			continue;
746		entry = pte_to_swp_entry(pentry);
747		if (unlikely(non_swap_entry(entry)))
748			continue;
749		page = __read_swap_cache_async(entry, gfp_mask, vma,
750					       vmf->address, &page_allocated);
751		if (!page)
752			continue;
753		if (page_allocated) {
754			swap_readpage(page, false);
755			if (i != ra_info.offset) {
756				SetPageReadahead(page);
757				count_vm_event(SWAP_RA);
758			}
759		}
760		put_page(page);
761	}
762	blk_finish_plug(&plug);
763	lru_add_drain();
764skip:
765	return read_swap_cache_async(fentry, gfp_mask, vma, vmf->address,
766				     ra_info.win == 1);
767}
768
769/**
770 * swapin_readahead - swap in pages in hope we need them soon
771 * @entry: swap entry of this memory
772 * @gfp_mask: memory allocation flags
773 * @vmf: fault information
774 *
775 * Returns the struct page for entry and addr, after queueing swapin.
776 *
777 * It's a main entry function for swap readahead. By the configuration,
778 * it will read ahead blocks by cluster-based(ie, physical disk based)
779 * or vma-based(ie, virtual address based on faulty address) readahead.
780 */
781struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask,
782				struct vm_fault *vmf)
783{
784	return swap_use_vma_readahead() ?
785			swap_vma_readahead(entry, gfp_mask, vmf) :
786			swap_cluster_readahead(entry, gfp_mask, vmf);
787}
788
789#ifdef CONFIG_SYSFS
790static ssize_t vma_ra_enabled_show(struct kobject *kobj,
791				     struct kobj_attribute *attr, char *buf)
792{
793	return sprintf(buf, "%s\n", enable_vma_readahead ? "true" : "false");
794}
795static ssize_t vma_ra_enabled_store(struct kobject *kobj,
796				      struct kobj_attribute *attr,
797				      const char *buf, size_t count)
798{
799	if (!strncmp(buf, "true", 4) || !strncmp(buf, "1", 1))
800		enable_vma_readahead = true;
801	else if (!strncmp(buf, "false", 5) || !strncmp(buf, "0", 1))
802		enable_vma_readahead = false;
803	else
804		return -EINVAL;
805
806	return count;
807}
808static struct kobj_attribute vma_ra_enabled_attr =
809	__ATTR(vma_ra_enabled, 0644, vma_ra_enabled_show,
810	       vma_ra_enabled_store);
811
812static struct attribute *swap_attrs[] = {
813	&vma_ra_enabled_attr.attr,
814	NULL,
815};
816
817static struct attribute_group swap_attr_group = {
818	.attrs = swap_attrs,
819};
820
821static int __init swap_init_sysfs(void)
822{
823	int err;
824	struct kobject *swap_kobj;
825
826	swap_kobj = kobject_create_and_add("swap", mm_kobj);
827	if (!swap_kobj) {
828		pr_err("failed to create swap kobject\n");
829		return -ENOMEM;
830	}
831	err = sysfs_create_group(swap_kobj, &swap_attr_group);
832	if (err) {
833		pr_err("failed to register swap group\n");
834		goto delete_obj;
835	}
836	return 0;
837
838delete_obj:
839	kobject_put(swap_kobj);
840	return err;
841}
842subsys_initcall(swap_init_sysfs);
843#endif