1 // SPDX-License-Identifier: GPL-2.0
3 * linux/mm/swap_state.c
5 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
6 * Swap reorganised 29.12.95, Stephen Tweedie
8 * Rewritten to use page cache, (C) 1998 Stephen Tweedie
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 #include <linux/shmem_fs.h>
28 * swapper_space is a fiction, retained to simplify the path through
29 * vmscan's shrink_page_list.
31 static 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
,
39 struct address_space
*swapper_spaces
[MAX_SWAPFILES
] __read_mostly
;
40 static unsigned int nr_swapper_spaces
[MAX_SWAPFILES
] __read_mostly
;
41 static bool enable_vma_readahead __read_mostly
= true;
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)
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)
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))
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)
61 #define INC_CACHE_INFO(x) data_race(swap_cache_info.x++)
62 #define ADD_CACHE_INFO(x, nr) data_race(swap_cache_info.x += (nr))
65 unsigned long add_total
;
66 unsigned long del_total
;
67 unsigned long find_success
;
68 unsigned long find_total
;
71 static atomic_t swapin_readahead_hits
= ATOMIC_INIT(4);
73 void show_swap_cache_info(void)
75 printk("%lu pages in swap cache\n", total_swapcache_pages());
76 printk("Swap cache stats: add %lu, delete %lu, find %lu/%lu\n",
77 swap_cache_info
.add_total
, swap_cache_info
.del_total
,
78 swap_cache_info
.find_success
, swap_cache_info
.find_total
);
79 printk("Free swap = %ldkB\n",
80 get_nr_swap_pages() << (PAGE_SHIFT
- 10));
81 printk("Total swap = %lukB\n", total_swap_pages
<< (PAGE_SHIFT
- 10));
84 void *get_shadow_from_swap_cache(swp_entry_t entry
)
86 struct address_space
*address_space
= swap_address_space(entry
);
87 pgoff_t idx
= swp_offset(entry
);
90 page
= xa_load(&address_space
->i_pages
, idx
);
91 if (xa_is_value(page
))
97 * add_to_swap_cache resembles add_to_page_cache_locked on swapper_space,
98 * but sets SwapCache flag and private instead of mapping and index.
100 int add_to_swap_cache(struct page
*page
, swp_entry_t entry
,
101 gfp_t gfp
, void **shadowp
)
103 struct address_space
*address_space
= swap_address_space(entry
);
104 pgoff_t idx
= swp_offset(entry
);
105 XA_STATE_ORDER(xas
, &address_space
->i_pages
, idx
, compound_order(page
));
106 unsigned long i
, nr
= thp_nr_pages(page
);
109 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
110 VM_BUG_ON_PAGE(PageSwapCache(page
), page
);
111 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
113 page_ref_add(page
, nr
);
114 SetPageSwapCache(page
);
117 unsigned long nr_shadows
= 0;
120 xas_create_range(&xas
);
123 for (i
= 0; i
< nr
; i
++) {
124 VM_BUG_ON_PAGE(xas
.xa_index
!= idx
+ i
, page
);
125 old
= xas_load(&xas
);
126 if (xa_is_value(old
)) {
131 set_page_private(page
+ i
, entry
.val
+ i
);
132 xas_store(&xas
, page
);
135 address_space
->nrexceptional
-= nr_shadows
;
136 address_space
->nrpages
+= nr
;
137 __mod_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, nr
);
138 __mod_lruvec_page_state(page
, NR_SWAPCACHE
, nr
);
139 ADD_CACHE_INFO(add_total
, nr
);
141 xas_unlock_irq(&xas
);
142 } while (xas_nomem(&xas
, gfp
));
144 if (!xas_error(&xas
))
147 ClearPageSwapCache(page
);
148 page_ref_sub(page
, nr
);
149 return xas_error(&xas
);
153 * This must be called only on pages that have
154 * been verified to be in the swap cache.
156 void __delete_from_swap_cache(struct page
*page
,
157 swp_entry_t entry
, void *shadow
)
159 struct address_space
*address_space
= swap_address_space(entry
);
160 int i
, nr
= thp_nr_pages(page
);
161 pgoff_t idx
= swp_offset(entry
);
162 XA_STATE(xas
, &address_space
->i_pages
, idx
);
164 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
165 VM_BUG_ON_PAGE(!PageSwapCache(page
), page
);
166 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
168 for (i
= 0; i
< nr
; i
++) {
169 void *entry
= xas_store(&xas
, shadow
);
170 VM_BUG_ON_PAGE(entry
!= page
, entry
);
171 set_page_private(page
+ i
, 0);
174 ClearPageSwapCache(page
);
176 address_space
->nrexceptional
+= nr
;
177 address_space
->nrpages
-= nr
;
178 __mod_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, -nr
);
179 __mod_lruvec_page_state(page
, NR_SWAPCACHE
, -nr
);
180 ADD_CACHE_INFO(del_total
, nr
);
184 * add_to_swap - allocate swap space for a page
185 * @page: page we want to move to swap
187 * Allocate swap space for the page and add the page to the
188 * swap cache. Caller needs to hold the page lock.
190 int add_to_swap(struct page
*page
)
195 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
196 VM_BUG_ON_PAGE(!PageUptodate(page
), page
);
198 entry
= get_swap_page(page
);
203 * XArray node allocations from PF_MEMALLOC contexts could
204 * completely exhaust the page allocator. __GFP_NOMEMALLOC
205 * stops emergency reserves from being allocated.
207 * TODO: this could cause a theoretical memory reclaim
208 * deadlock in the swap out path.
211 * Add it to the swap cache.
213 err
= add_to_swap_cache(page
, entry
,
214 __GFP_HIGH
|__GFP_NOMEMALLOC
|__GFP_NOWARN
, NULL
);
217 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
218 * clear SWAP_HAS_CACHE flag.
222 * Normally the page will be dirtied in unmap because its pte should be
223 * dirty. A special case is MADV_FREE page. The page's pte could have
224 * dirty bit cleared but the page's SwapBacked bit is still set because
225 * clearing the dirty bit and SwapBacked bit has no lock protected. For
226 * such page, unmap will not set dirty bit for it, so page reclaim will
227 * not write the page out. This can cause data corruption when the page
228 * is swap in later. Always setting the dirty bit for the page solves
231 set_page_dirty(page
);
236 put_swap_page(page
, entry
);
241 * This must be called only on pages that have
242 * been verified to be in the swap cache and locked.
243 * It will never put the page into the free list,
244 * the caller has a reference on the page.
246 void delete_from_swap_cache(struct page
*page
)
248 swp_entry_t entry
= { .val
= page_private(page
) };
249 struct address_space
*address_space
= swap_address_space(entry
);
251 xa_lock_irq(&address_space
->i_pages
);
252 __delete_from_swap_cache(page
, entry
, NULL
);
253 xa_unlock_irq(&address_space
->i_pages
);
255 put_swap_page(page
, entry
);
256 page_ref_sub(page
, thp_nr_pages(page
));
259 void clear_shadow_from_swap_cache(int type
, unsigned long begin
,
262 unsigned long curr
= begin
;
266 unsigned long nr_shadows
= 0;
267 swp_entry_t entry
= swp_entry(type
, curr
);
268 struct address_space
*address_space
= swap_address_space(entry
);
269 XA_STATE(xas
, &address_space
->i_pages
, curr
);
271 xa_lock_irq(&address_space
->i_pages
);
272 xas_for_each(&xas
, old
, end
) {
273 if (!xa_is_value(old
))
275 xas_store(&xas
, NULL
);
278 address_space
->nrexceptional
-= nr_shadows
;
279 xa_unlock_irq(&address_space
->i_pages
);
281 /* search the next swapcache until we meet end */
282 curr
>>= SWAP_ADDRESS_SPACE_SHIFT
;
284 curr
<<= SWAP_ADDRESS_SPACE_SHIFT
;
291 * If we are the only user, then try to free up the swap cache.
293 * Its ok to check for PageSwapCache without the page lock
294 * here because we are going to recheck again inside
295 * try_to_free_swap() _with_ the lock.
298 static inline void free_swap_cache(struct page
*page
)
300 if (PageSwapCache(page
) && !page_mapped(page
) && trylock_page(page
)) {
301 try_to_free_swap(page
);
307 * Perform a free_page(), also freeing any swap cache associated with
308 * this page if it is the last user of the page.
310 void free_page_and_swap_cache(struct page
*page
)
312 free_swap_cache(page
);
313 if (!is_huge_zero_page(page
))
318 * Passed an array of pages, drop them all from swapcache and then release
319 * them. They are removed from the LRU and freed if this is their last use.
321 void free_pages_and_swap_cache(struct page
**pages
, int nr
)
323 struct page
**pagep
= pages
;
327 for (i
= 0; i
< nr
; i
++)
328 free_swap_cache(pagep
[i
]);
329 release_pages(pagep
, nr
);
332 static inline bool swap_use_vma_readahead(void)
334 return READ_ONCE(enable_vma_readahead
) && !atomic_read(&nr_rotate_swap
);
338 * Lookup a swap entry in the swap cache. A found page will be returned
339 * unlocked and with its refcount incremented - we rely on the kernel
340 * lock getting page table operations atomic even if we drop the page
341 * lock before returning.
343 struct page
*lookup_swap_cache(swp_entry_t entry
, struct vm_area_struct
*vma
,
347 struct swap_info_struct
*si
;
349 si
= get_swap_device(entry
);
352 page
= find_get_page(swap_address_space(entry
), swp_offset(entry
));
355 INC_CACHE_INFO(find_total
);
357 bool vma_ra
= swap_use_vma_readahead();
360 INC_CACHE_INFO(find_success
);
362 * At the moment, we don't support PG_readahead for anon THP
363 * so let's bail out rather than confusing the readahead stat.
365 if (unlikely(PageTransCompound(page
)))
368 readahead
= TestClearPageReadahead(page
);
370 unsigned long ra_val
;
373 ra_val
= GET_SWAP_RA_VAL(vma
);
374 win
= SWAP_RA_WIN(ra_val
);
375 hits
= SWAP_RA_HITS(ra_val
);
377 hits
= min_t(int, hits
+ 1, SWAP_RA_HITS_MAX
);
378 atomic_long_set(&vma
->swap_readahead_info
,
379 SWAP_RA_VAL(addr
, win
, hits
));
383 count_vm_event(SWAP_RA_HIT
);
385 atomic_inc(&swapin_readahead_hits
);
393 * find_get_incore_page - Find and get a page from the page or swap caches.
394 * @mapping: The address_space to search.
395 * @index: The page cache index.
397 * This differs from find_get_page() in that it will also look for the
398 * page in the swap cache.
400 * Return: The found page or %NULL.
402 struct page
*find_get_incore_page(struct address_space
*mapping
, pgoff_t index
)
405 struct swap_info_struct
*si
;
406 struct page
*page
= pagecache_get_page(mapping
, index
,
407 FGP_ENTRY
| FGP_HEAD
, 0);
411 if (!xa_is_value(page
))
412 return find_subpage(page
, index
);
413 if (!shmem_mapping(mapping
))
416 swp
= radix_to_swp_entry(page
);
417 /* Prevent swapoff from happening to us */
418 si
= get_swap_device(swp
);
421 page
= find_get_page(swap_address_space(swp
), swp_offset(swp
));
426 struct page
*__read_swap_cache_async(swp_entry_t entry
, gfp_t gfp_mask
,
427 struct vm_area_struct
*vma
, unsigned long addr
,
428 bool *new_page_allocated
)
430 struct swap_info_struct
*si
;
434 *new_page_allocated
= false;
439 * First check the swap cache. Since this is normally
440 * called after lookup_swap_cache() failed, re-calling
441 * that would confuse statistics.
443 si
= get_swap_device(entry
);
446 page
= find_get_page(swap_address_space(entry
),
453 * Just skip read ahead for unused swap slot.
454 * During swap_off when swap_slot_cache is disabled,
455 * we have to handle the race between putting
456 * swap entry in swap cache and marking swap slot
457 * as SWAP_HAS_CACHE. That's done in later part of code or
458 * else swap_off will be aborted if we return NULL.
460 if (!__swp_swapcount(entry
) && swap_slot_cache_enabled
)
464 * Get a new page to read into from swap. Allocate it now,
465 * before marking swap_map SWAP_HAS_CACHE, when -EEXIST will
466 * cause any racers to loop around until we add it to cache.
468 page
= alloc_page_vma(gfp_mask
, vma
, addr
);
473 * Swap entry may have been freed since our caller observed it.
475 err
= swapcache_prepare(entry
);
484 * We might race against __delete_from_swap_cache(), and
485 * stumble across a swap_map entry whose SWAP_HAS_CACHE
486 * has not yet been cleared. Or race against another
487 * __read_swap_cache_async(), which has set SWAP_HAS_CACHE
488 * in swap_map, but not yet added its page to swap cache.
494 * The swap entry is ours to swap in. Prepare the new page.
497 __SetPageLocked(page
);
498 __SetPageSwapBacked(page
);
500 /* May fail (-ENOMEM) if XArray node allocation failed. */
501 if (add_to_swap_cache(page
, entry
, gfp_mask
& GFP_RECLAIM_MASK
, &shadow
)) {
502 put_swap_page(page
, entry
);
506 if (mem_cgroup_charge(page
, NULL
, gfp_mask
)) {
507 delete_from_swap_cache(page
);
512 workingset_refault(page
, shadow
);
514 /* Caller will initiate read into locked page */
516 *new_page_allocated
= true;
526 * Locate a page of swap in physical memory, reserving swap cache space
527 * and reading the disk if it is not already cached.
528 * A failure return means that either the page allocation failed or that
529 * the swap entry is no longer in use.
531 struct page
*read_swap_cache_async(swp_entry_t entry
, gfp_t gfp_mask
,
532 struct vm_area_struct
*vma
, unsigned long addr
, bool do_poll
)
534 bool page_was_allocated
;
535 struct page
*retpage
= __read_swap_cache_async(entry
, gfp_mask
,
536 vma
, addr
, &page_was_allocated
);
538 if (page_was_allocated
)
539 swap_readpage(retpage
, do_poll
);
544 static unsigned int __swapin_nr_pages(unsigned long prev_offset
,
545 unsigned long offset
,
550 unsigned int pages
, last_ra
;
553 * This heuristic has been found to work well on both sequential and
554 * random loads, swapping to hard disk or to SSD: please don't ask
555 * what the "+ 2" means, it just happens to work well, that's all.
560 * We can have no readahead hits to judge by: but must not get
561 * stuck here forever, so check for an adjacent offset instead
562 * (and don't even bother to check whether swap type is same).
564 if (offset
!= prev_offset
+ 1 && offset
!= prev_offset
- 1)
567 unsigned int roundup
= 4;
568 while (roundup
< pages
)
573 if (pages
> max_pages
)
576 /* Don't shrink readahead too fast */
577 last_ra
= prev_win
/ 2;
584 static unsigned long swapin_nr_pages(unsigned long offset
)
586 static unsigned long prev_offset
;
587 unsigned int hits
, pages
, max_pages
;
588 static atomic_t last_readahead_pages
;
590 max_pages
= 1 << READ_ONCE(page_cluster
);
594 hits
= atomic_xchg(&swapin_readahead_hits
, 0);
595 pages
= __swapin_nr_pages(READ_ONCE(prev_offset
), offset
, hits
,
597 atomic_read(&last_readahead_pages
));
599 WRITE_ONCE(prev_offset
, offset
);
600 atomic_set(&last_readahead_pages
, pages
);
606 * swap_cluster_readahead - swap in pages in hope we need them soon
607 * @entry: swap entry of this memory
608 * @gfp_mask: memory allocation flags
609 * @vmf: fault information
611 * Returns the struct page for entry and addr, after queueing swapin.
613 * Primitive swap readahead code. We simply read an aligned block of
614 * (1 << page_cluster) entries in the swap area. This method is chosen
615 * because it doesn't cost us any seek time. We also make sure to queue
616 * the 'original' request together with the readahead ones...
618 * This has been extended to use the NUMA policies from the mm triggering
621 * Caller must hold read mmap_lock if vmf->vma is not NULL.
623 struct page
*swap_cluster_readahead(swp_entry_t entry
, gfp_t gfp_mask
,
624 struct vm_fault
*vmf
)
627 unsigned long entry_offset
= swp_offset(entry
);
628 unsigned long offset
= entry_offset
;
629 unsigned long start_offset
, end_offset
;
631 struct swap_info_struct
*si
= swp_swap_info(entry
);
632 struct blk_plug plug
;
633 bool do_poll
= true, page_allocated
;
634 struct vm_area_struct
*vma
= vmf
->vma
;
635 unsigned long addr
= vmf
->address
;
637 mask
= swapin_nr_pages(offset
) - 1;
641 /* Test swap type to make sure the dereference is safe */
642 if (likely(si
->flags
& (SWP_BLKDEV
| SWP_FS_OPS
))) {
643 struct inode
*inode
= si
->swap_file
->f_mapping
->host
;
644 if (inode_read_congested(inode
))
649 /* Read a page_cluster sized and aligned cluster around offset. */
650 start_offset
= offset
& ~mask
;
651 end_offset
= offset
| mask
;
652 if (!start_offset
) /* First page is swap header. */
654 if (end_offset
>= si
->max
)
655 end_offset
= si
->max
- 1;
657 blk_start_plug(&plug
);
658 for (offset
= start_offset
; offset
<= end_offset
; offset
++) {
659 /* Ok, do the async read-ahead now */
660 page
= __read_swap_cache_async(
661 swp_entry(swp_type(entry
), offset
),
662 gfp_mask
, vma
, addr
, &page_allocated
);
665 if (page_allocated
) {
666 swap_readpage(page
, false);
667 if (offset
!= entry_offset
) {
668 SetPageReadahead(page
);
669 count_vm_event(SWAP_RA
);
674 blk_finish_plug(&plug
);
676 lru_add_drain(); /* Push any new pages onto the LRU now */
678 return read_swap_cache_async(entry
, gfp_mask
, vma
, addr
, do_poll
);
681 int init_swap_address_space(unsigned int type
, unsigned long nr_pages
)
683 struct address_space
*spaces
, *space
;
686 nr
= DIV_ROUND_UP(nr_pages
, SWAP_ADDRESS_SPACE_PAGES
);
687 spaces
= kvcalloc(nr
, sizeof(struct address_space
), GFP_KERNEL
);
690 for (i
= 0; i
< nr
; i
++) {
692 xa_init_flags(&space
->i_pages
, XA_FLAGS_LOCK_IRQ
);
693 atomic_set(&space
->i_mmap_writable
, 0);
694 space
->a_ops
= &swap_aops
;
695 /* swap cache doesn't use writeback related tags */
696 mapping_set_no_writeback_tags(space
);
698 nr_swapper_spaces
[type
] = nr
;
699 swapper_spaces
[type
] = spaces
;
704 void exit_swap_address_space(unsigned int type
)
706 kvfree(swapper_spaces
[type
]);
707 nr_swapper_spaces
[type
] = 0;
708 swapper_spaces
[type
] = NULL
;
711 static inline void swap_ra_clamp_pfn(struct vm_area_struct
*vma
,
715 unsigned long *start
,
718 *start
= max3(lpfn
, PFN_DOWN(vma
->vm_start
),
719 PFN_DOWN(faddr
& PMD_MASK
));
720 *end
= min3(rpfn
, PFN_DOWN(vma
->vm_end
),
721 PFN_DOWN((faddr
& PMD_MASK
) + PMD_SIZE
));
724 static void swap_ra_info(struct vm_fault
*vmf
,
725 struct vma_swap_readahead
*ra_info
)
727 struct vm_area_struct
*vma
= vmf
->vma
;
728 unsigned long ra_val
;
730 unsigned long faddr
, pfn
, fpfn
;
731 unsigned long start
, end
;
732 pte_t
*pte
, *orig_pte
;
733 unsigned int max_win
, hits
, prev_win
, win
, left
;
738 max_win
= 1 << min_t(unsigned int, READ_ONCE(page_cluster
),
739 SWAP_RA_ORDER_CEILING
);
745 faddr
= vmf
->address
;
746 orig_pte
= pte
= pte_offset_map(vmf
->pmd
, faddr
);
747 entry
= pte_to_swp_entry(*pte
);
748 if ((unlikely(non_swap_entry(entry
)))) {
753 fpfn
= PFN_DOWN(faddr
);
754 ra_val
= GET_SWAP_RA_VAL(vma
);
755 pfn
= PFN_DOWN(SWAP_RA_ADDR(ra_val
));
756 prev_win
= SWAP_RA_WIN(ra_val
);
757 hits
= SWAP_RA_HITS(ra_val
);
758 ra_info
->win
= win
= __swapin_nr_pages(pfn
, fpfn
, hits
,
760 atomic_long_set(&vma
->swap_readahead_info
,
761 SWAP_RA_VAL(faddr
, win
, 0));
768 /* Copy the PTEs because the page table may be unmapped */
770 swap_ra_clamp_pfn(vma
, faddr
, fpfn
, fpfn
+ win
, &start
, &end
);
771 else if (pfn
== fpfn
+ 1)
772 swap_ra_clamp_pfn(vma
, faddr
, fpfn
- win
+ 1, fpfn
+ 1,
775 left
= (win
- 1) / 2;
776 swap_ra_clamp_pfn(vma
, faddr
, fpfn
- left
, fpfn
+ win
- left
,
779 ra_info
->nr_pte
= end
- start
;
780 ra_info
->offset
= fpfn
- start
;
781 pte
-= ra_info
->offset
;
785 tpte
= ra_info
->ptes
;
786 for (pfn
= start
; pfn
!= end
; pfn
++)
793 * swap_vma_readahead - swap in pages in hope we need them soon
794 * @fentry: swap entry of this memory
795 * @gfp_mask: memory allocation flags
796 * @vmf: fault information
798 * Returns the struct page for entry and addr, after queueing swapin.
800 * Primitive swap readahead code. We simply read in a few pages whoes
801 * virtual addresses are around the fault address in the same vma.
803 * Caller must hold read mmap_lock if vmf->vma is not NULL.
806 static struct page
*swap_vma_readahead(swp_entry_t fentry
, gfp_t gfp_mask
,
807 struct vm_fault
*vmf
)
809 struct blk_plug plug
;
810 struct vm_area_struct
*vma
= vmf
->vma
;
816 struct vma_swap_readahead ra_info
= {
820 swap_ra_info(vmf
, &ra_info
);
821 if (ra_info
.win
== 1)
824 blk_start_plug(&plug
);
825 for (i
= 0, pte
= ra_info
.ptes
; i
< ra_info
.nr_pte
;
828 if (pte_none(pentry
))
830 if (pte_present(pentry
))
832 entry
= pte_to_swp_entry(pentry
);
833 if (unlikely(non_swap_entry(entry
)))
835 page
= __read_swap_cache_async(entry
, gfp_mask
, vma
,
836 vmf
->address
, &page_allocated
);
839 if (page_allocated
) {
840 swap_readpage(page
, false);
841 if (i
!= ra_info
.offset
) {
842 SetPageReadahead(page
);
843 count_vm_event(SWAP_RA
);
848 blk_finish_plug(&plug
);
851 return read_swap_cache_async(fentry
, gfp_mask
, vma
, vmf
->address
,
856 * swapin_readahead - swap in pages in hope we need them soon
857 * @entry: swap entry of this memory
858 * @gfp_mask: memory allocation flags
859 * @vmf: fault information
861 * Returns the struct page for entry and addr, after queueing swapin.
863 * It's a main entry function for swap readahead. By the configuration,
864 * it will read ahead blocks by cluster-based(ie, physical disk based)
865 * or vma-based(ie, virtual address based on faulty address) readahead.
867 struct page
*swapin_readahead(swp_entry_t entry
, gfp_t gfp_mask
,
868 struct vm_fault
*vmf
)
870 return swap_use_vma_readahead() ?
871 swap_vma_readahead(entry
, gfp_mask
, vmf
) :
872 swap_cluster_readahead(entry
, gfp_mask
, vmf
);
876 static ssize_t
vma_ra_enabled_show(struct kobject
*kobj
,
877 struct kobj_attribute
*attr
, char *buf
)
879 return sysfs_emit(buf
, "%s\n",
880 enable_vma_readahead
? "true" : "false");
882 static ssize_t
vma_ra_enabled_store(struct kobject
*kobj
,
883 struct kobj_attribute
*attr
,
884 const char *buf
, size_t count
)
886 if (!strncmp(buf
, "true", 4) || !strncmp(buf
, "1", 1))
887 enable_vma_readahead
= true;
888 else if (!strncmp(buf
, "false", 5) || !strncmp(buf
, "0", 1))
889 enable_vma_readahead
= false;
895 static struct kobj_attribute vma_ra_enabled_attr
=
896 __ATTR(vma_ra_enabled
, 0644, vma_ra_enabled_show
,
897 vma_ra_enabled_store
);
899 static struct attribute
*swap_attrs
[] = {
900 &vma_ra_enabled_attr
.attr
,
904 static const struct attribute_group swap_attr_group
= {
908 static int __init
swap_init_sysfs(void)
911 struct kobject
*swap_kobj
;
913 swap_kobj
= kobject_create_and_add("swap", mm_kobj
);
915 pr_err("failed to create swap kobject\n");
918 err
= sysfs_create_group(swap_kobj
, &swap_attr_group
);
920 pr_err("failed to register swap group\n");
926 kobject_put(swap_kobj
);
929 subsys_initcall(swap_init_sysfs
);