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>
29 * swapper_space is a fiction, retained to simplify the path through
30 * vmscan's shrink_page_list.
32 static const struct address_space_operations swap_aops
= {
33 .writepage
= swap_writepage
,
34 .dirty_folio
= noop_dirty_folio
,
35 #ifdef CONFIG_MIGRATION
36 .migrate_folio
= migrate_folio
,
40 struct address_space
*swapper_spaces
[MAX_SWAPFILES
] __read_mostly
;
41 static unsigned int nr_swapper_spaces
[MAX_SWAPFILES
] __read_mostly
;
42 static bool enable_vma_readahead __read_mostly
= true;
44 #define SWAP_RA_WIN_SHIFT (PAGE_SHIFT / 2)
45 #define SWAP_RA_HITS_MASK ((1UL << SWAP_RA_WIN_SHIFT) - 1)
46 #define SWAP_RA_HITS_MAX SWAP_RA_HITS_MASK
47 #define SWAP_RA_WIN_MASK (~PAGE_MASK & ~SWAP_RA_HITS_MASK)
49 #define SWAP_RA_HITS(v) ((v) & SWAP_RA_HITS_MASK)
50 #define SWAP_RA_WIN(v) (((v) & SWAP_RA_WIN_MASK) >> SWAP_RA_WIN_SHIFT)
51 #define SWAP_RA_ADDR(v) ((v) & PAGE_MASK)
53 #define SWAP_RA_VAL(addr, win, hits) \
54 (((addr) & PAGE_MASK) | \
55 (((win) << SWAP_RA_WIN_SHIFT) & SWAP_RA_WIN_MASK) | \
56 ((hits) & SWAP_RA_HITS_MASK))
58 /* Initial readahead hits is 4 to start up with a small window */
59 #define GET_SWAP_RA_VAL(vma) \
60 (atomic_long_read(&(vma)->swap_readahead_info) ? : 4)
62 static atomic_t swapin_readahead_hits
= ATOMIC_INIT(4);
64 void show_swap_cache_info(void)
66 printk("%lu pages in swap cache\n", total_swapcache_pages());
67 printk("Free swap = %ldkB\n",
68 get_nr_swap_pages() << (PAGE_SHIFT
- 10));
69 printk("Total swap = %lukB\n", total_swap_pages
<< (PAGE_SHIFT
- 10));
72 void *get_shadow_from_swap_cache(swp_entry_t entry
)
74 struct address_space
*address_space
= swap_address_space(entry
);
75 pgoff_t idx
= swp_offset(entry
);
78 page
= xa_load(&address_space
->i_pages
, idx
);
79 if (xa_is_value(page
))
85 * add_to_swap_cache resembles filemap_add_folio on swapper_space,
86 * but sets SwapCache flag and private instead of mapping and index.
88 int add_to_swap_cache(struct page
*page
, swp_entry_t entry
,
89 gfp_t gfp
, void **shadowp
)
91 struct address_space
*address_space
= swap_address_space(entry
);
92 pgoff_t idx
= swp_offset(entry
);
93 XA_STATE_ORDER(xas
, &address_space
->i_pages
, idx
, compound_order(page
));
94 unsigned long i
, nr
= thp_nr_pages(page
);
97 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
98 VM_BUG_ON_PAGE(PageSwapCache(page
), page
);
99 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
101 page_ref_add(page
, nr
);
102 SetPageSwapCache(page
);
106 xas_create_range(&xas
);
109 for (i
= 0; i
< nr
; i
++) {
110 VM_BUG_ON_PAGE(xas
.xa_index
!= idx
+ i
, page
);
111 old
= xas_load(&xas
);
112 if (xa_is_value(old
)) {
116 set_page_private(page
+ i
, entry
.val
+ i
);
117 xas_store(&xas
, page
);
120 address_space
->nrpages
+= nr
;
121 __mod_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, nr
);
122 __mod_lruvec_page_state(page
, NR_SWAPCACHE
, nr
);
124 xas_unlock_irq(&xas
);
125 } while (xas_nomem(&xas
, gfp
));
127 if (!xas_error(&xas
))
130 ClearPageSwapCache(page
);
131 page_ref_sub(page
, nr
);
132 return xas_error(&xas
);
136 * This must be called only on folios that have
137 * been verified to be in the swap cache.
139 void __delete_from_swap_cache(struct folio
*folio
,
140 swp_entry_t entry
, void *shadow
)
142 struct address_space
*address_space
= swap_address_space(entry
);
144 long nr
= folio_nr_pages(folio
);
145 pgoff_t idx
= swp_offset(entry
);
146 XA_STATE(xas
, &address_space
->i_pages
, idx
);
148 VM_BUG_ON_FOLIO(!folio_test_locked(folio
), folio
);
149 VM_BUG_ON_FOLIO(!folio_test_swapcache(folio
), folio
);
150 VM_BUG_ON_FOLIO(folio_test_writeback(folio
), folio
);
152 for (i
= 0; i
< nr
; i
++) {
153 void *entry
= xas_store(&xas
, shadow
);
154 VM_BUG_ON_FOLIO(entry
!= folio
, folio
);
155 set_page_private(folio_page(folio
, i
), 0);
158 folio_clear_swapcache(folio
);
159 address_space
->nrpages
-= nr
;
160 __node_stat_mod_folio(folio
, NR_FILE_PAGES
, -nr
);
161 __lruvec_stat_mod_folio(folio
, NR_SWAPCACHE
, -nr
);
165 * add_to_swap - allocate swap space for a folio
166 * @folio: folio we want to move to swap
168 * Allocate swap space for the folio and add the folio to the
171 * Context: Caller needs to hold the folio lock.
172 * Return: Whether the folio was added to the swap cache.
174 bool add_to_swap(struct folio
*folio
)
179 VM_BUG_ON_FOLIO(!folio_test_locked(folio
), folio
);
180 VM_BUG_ON_FOLIO(!folio_test_uptodate(folio
), folio
);
182 entry
= folio_alloc_swap(folio
);
187 * XArray node allocations from PF_MEMALLOC contexts could
188 * completely exhaust the page allocator. __GFP_NOMEMALLOC
189 * stops emergency reserves from being allocated.
191 * TODO: this could cause a theoretical memory reclaim
192 * deadlock in the swap out path.
195 * Add it to the swap cache.
197 err
= add_to_swap_cache(&folio
->page
, entry
,
198 __GFP_HIGH
|__GFP_NOMEMALLOC
|__GFP_NOWARN
, NULL
);
201 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
202 * clear SWAP_HAS_CACHE flag.
206 * Normally the folio will be dirtied in unmap because its
207 * pte should be dirty. A special case is MADV_FREE page. The
208 * page's pte could have dirty bit cleared but the folio's
209 * SwapBacked flag is still set because clearing the dirty bit
210 * and SwapBacked flag has no lock protected. For such folio,
211 * unmap will not set dirty bit for it, so folio reclaim will
212 * not write the folio out. This can cause data corruption when
213 * the folio is swapped in later. Always setting the dirty flag
214 * for the folio solves the problem.
216 folio_mark_dirty(folio
);
221 put_swap_page(&folio
->page
, entry
);
226 * This must be called only on folios that have
227 * been verified to be in the swap cache and locked.
228 * It will never put the folio into the free list,
229 * the caller has a reference on the folio.
231 void delete_from_swap_cache(struct folio
*folio
)
233 swp_entry_t entry
= folio_swap_entry(folio
);
234 struct address_space
*address_space
= swap_address_space(entry
);
236 xa_lock_irq(&address_space
->i_pages
);
237 __delete_from_swap_cache(folio
, entry
, NULL
);
238 xa_unlock_irq(&address_space
->i_pages
);
240 put_swap_page(&folio
->page
, entry
);
241 folio_ref_sub(folio
, folio_nr_pages(folio
));
244 void clear_shadow_from_swap_cache(int type
, unsigned long begin
,
247 unsigned long curr
= begin
;
251 swp_entry_t entry
= swp_entry(type
, curr
);
252 struct address_space
*address_space
= swap_address_space(entry
);
253 XA_STATE(xas
, &address_space
->i_pages
, curr
);
255 xa_lock_irq(&address_space
->i_pages
);
256 xas_for_each(&xas
, old
, end
) {
257 if (!xa_is_value(old
))
259 xas_store(&xas
, NULL
);
261 xa_unlock_irq(&address_space
->i_pages
);
263 /* search the next swapcache until we meet end */
264 curr
>>= SWAP_ADDRESS_SPACE_SHIFT
;
266 curr
<<= SWAP_ADDRESS_SPACE_SHIFT
;
273 * If we are the only user, then try to free up the swap cache.
275 * Its ok to check for PageSwapCache without the page lock
276 * here because we are going to recheck again inside
277 * try_to_free_swap() _with_ the lock.
280 void free_swap_cache(struct page
*page
)
282 if (PageSwapCache(page
) && !page_mapped(page
) && trylock_page(page
)) {
283 try_to_free_swap(page
);
289 * Perform a free_page(), also freeing any swap cache associated with
290 * this page if it is the last user of the page.
292 void free_page_and_swap_cache(struct page
*page
)
294 free_swap_cache(page
);
295 if (!is_huge_zero_page(page
))
300 * Passed an array of pages, drop them all from swapcache and then release
301 * them. They are removed from the LRU and freed if this is their last use.
303 void free_pages_and_swap_cache(struct page
**pages
, int nr
)
305 struct page
**pagep
= pages
;
309 for (i
= 0; i
< nr
; i
++)
310 free_swap_cache(pagep
[i
]);
311 release_pages(pagep
, nr
);
314 static inline bool swap_use_vma_readahead(void)
316 return READ_ONCE(enable_vma_readahead
) && !atomic_read(&nr_rotate_swap
);
320 * Lookup a swap entry in the swap cache. A found page will be returned
321 * unlocked and with its refcount incremented - we rely on the kernel
322 * lock getting page table operations atomic even if we drop the page
323 * lock before returning.
325 struct page
*lookup_swap_cache(swp_entry_t entry
, struct vm_area_struct
*vma
,
329 struct swap_info_struct
*si
;
331 si
= get_swap_device(entry
);
334 page
= find_get_page(swap_address_space(entry
), swp_offset(entry
));
338 bool vma_ra
= swap_use_vma_readahead();
342 * At the moment, we don't support PG_readahead for anon THP
343 * so let's bail out rather than confusing the readahead stat.
345 if (unlikely(PageTransCompound(page
)))
348 readahead
= TestClearPageReadahead(page
);
350 unsigned long ra_val
;
353 ra_val
= GET_SWAP_RA_VAL(vma
);
354 win
= SWAP_RA_WIN(ra_val
);
355 hits
= SWAP_RA_HITS(ra_val
);
357 hits
= min_t(int, hits
+ 1, SWAP_RA_HITS_MAX
);
358 atomic_long_set(&vma
->swap_readahead_info
,
359 SWAP_RA_VAL(addr
, win
, hits
));
363 count_vm_event(SWAP_RA_HIT
);
365 atomic_inc(&swapin_readahead_hits
);
373 * find_get_incore_page - Find and get a page from the page or swap caches.
374 * @mapping: The address_space to search.
375 * @index: The page cache index.
377 * This differs from find_get_page() in that it will also look for the
378 * page in the swap cache.
380 * Return: The found page or %NULL.
382 struct page
*find_get_incore_page(struct address_space
*mapping
, pgoff_t index
)
385 struct swap_info_struct
*si
;
386 struct page
*page
= pagecache_get_page(mapping
, index
,
387 FGP_ENTRY
| FGP_HEAD
, 0);
391 if (!xa_is_value(page
))
392 return find_subpage(page
, index
);
393 if (!shmem_mapping(mapping
))
396 swp
= radix_to_swp_entry(page
);
397 /* There might be swapin error entries in shmem mapping. */
398 if (non_swap_entry(swp
))
400 /* Prevent swapoff from happening to us */
401 si
= get_swap_device(swp
);
404 page
= find_get_page(swap_address_space(swp
), swp_offset(swp
));
409 struct page
*__read_swap_cache_async(swp_entry_t entry
, gfp_t gfp_mask
,
410 struct vm_area_struct
*vma
, unsigned long addr
,
411 bool *new_page_allocated
)
413 struct swap_info_struct
*si
;
417 *new_page_allocated
= false;
422 * First check the swap cache. Since this is normally
423 * called after lookup_swap_cache() failed, re-calling
424 * that would confuse statistics.
426 si
= get_swap_device(entry
);
429 page
= find_get_page(swap_address_space(entry
),
436 * Just skip read ahead for unused swap slot.
437 * During swap_off when swap_slot_cache is disabled,
438 * we have to handle the race between putting
439 * swap entry in swap cache and marking swap slot
440 * as SWAP_HAS_CACHE. That's done in later part of code or
441 * else swap_off will be aborted if we return NULL.
443 if (!__swp_swapcount(entry
) && swap_slot_cache_enabled
)
447 * Get a new page to read into from swap. Allocate it now,
448 * before marking swap_map SWAP_HAS_CACHE, when -EEXIST will
449 * cause any racers to loop around until we add it to cache.
451 page
= alloc_page_vma(gfp_mask
, vma
, addr
);
456 * Swap entry may have been freed since our caller observed it.
458 err
= swapcache_prepare(entry
);
467 * We might race against __delete_from_swap_cache(), and
468 * stumble across a swap_map entry whose SWAP_HAS_CACHE
469 * has not yet been cleared. Or race against another
470 * __read_swap_cache_async(), which has set SWAP_HAS_CACHE
471 * in swap_map, but not yet added its page to swap cache.
473 schedule_timeout_uninterruptible(1);
477 * The swap entry is ours to swap in. Prepare the new page.
480 __SetPageLocked(page
);
481 __SetPageSwapBacked(page
);
483 if (mem_cgroup_swapin_charge_page(page
, NULL
, gfp_mask
, entry
))
486 /* May fail (-ENOMEM) if XArray node allocation failed. */
487 if (add_to_swap_cache(page
, entry
, gfp_mask
& GFP_RECLAIM_MASK
, &shadow
))
490 mem_cgroup_swapin_uncharge_swap(entry
);
493 workingset_refault(page_folio(page
), shadow
);
495 /* Caller will initiate read into locked page */
497 *new_page_allocated
= true;
501 put_swap_page(page
, entry
);
508 * Locate a page of swap in physical memory, reserving swap cache space
509 * and reading the disk if it is not already cached.
510 * A failure return means that either the page allocation failed or that
511 * the swap entry is no longer in use.
513 struct page
*read_swap_cache_async(swp_entry_t entry
, gfp_t gfp_mask
,
514 struct vm_area_struct
*vma
,
515 unsigned long addr
, bool do_poll
,
516 struct swap_iocb
**plug
)
518 bool page_was_allocated
;
519 struct page
*retpage
= __read_swap_cache_async(entry
, gfp_mask
,
520 vma
, addr
, &page_was_allocated
);
522 if (page_was_allocated
)
523 swap_readpage(retpage
, do_poll
, plug
);
528 static unsigned int __swapin_nr_pages(unsigned long prev_offset
,
529 unsigned long offset
,
534 unsigned int pages
, last_ra
;
537 * This heuristic has been found to work well on both sequential and
538 * random loads, swapping to hard disk or to SSD: please don't ask
539 * what the "+ 2" means, it just happens to work well, that's all.
544 * We can have no readahead hits to judge by: but must not get
545 * stuck here forever, so check for an adjacent offset instead
546 * (and don't even bother to check whether swap type is same).
548 if (offset
!= prev_offset
+ 1 && offset
!= prev_offset
- 1)
551 unsigned int roundup
= 4;
552 while (roundup
< pages
)
557 if (pages
> max_pages
)
560 /* Don't shrink readahead too fast */
561 last_ra
= prev_win
/ 2;
568 static unsigned long swapin_nr_pages(unsigned long offset
)
570 static unsigned long prev_offset
;
571 unsigned int hits
, pages
, max_pages
;
572 static atomic_t last_readahead_pages
;
574 max_pages
= 1 << READ_ONCE(page_cluster
);
578 hits
= atomic_xchg(&swapin_readahead_hits
, 0);
579 pages
= __swapin_nr_pages(READ_ONCE(prev_offset
), offset
, hits
,
581 atomic_read(&last_readahead_pages
));
583 WRITE_ONCE(prev_offset
, offset
);
584 atomic_set(&last_readahead_pages
, pages
);
590 * swap_cluster_readahead - swap in pages in hope we need them soon
591 * @entry: swap entry of this memory
592 * @gfp_mask: memory allocation flags
593 * @vmf: fault information
595 * Returns the struct page for entry and addr, after queueing swapin.
597 * Primitive swap readahead code. We simply read an aligned block of
598 * (1 << page_cluster) entries in the swap area. This method is chosen
599 * because it doesn't cost us any seek time. We also make sure to queue
600 * the 'original' request together with the readahead ones...
602 * This has been extended to use the NUMA policies from the mm triggering
605 * Caller must hold read mmap_lock if vmf->vma is not NULL.
607 struct page
*swap_cluster_readahead(swp_entry_t entry
, gfp_t gfp_mask
,
608 struct vm_fault
*vmf
)
611 unsigned long entry_offset
= swp_offset(entry
);
612 unsigned long offset
= entry_offset
;
613 unsigned long start_offset
, end_offset
;
615 struct swap_info_struct
*si
= swp_swap_info(entry
);
616 struct blk_plug plug
;
617 struct swap_iocb
*splug
= NULL
;
618 bool do_poll
= true, page_allocated
;
619 struct vm_area_struct
*vma
= vmf
->vma
;
620 unsigned long addr
= vmf
->address
;
622 mask
= swapin_nr_pages(offset
) - 1;
627 /* Read a page_cluster sized and aligned cluster around offset. */
628 start_offset
= offset
& ~mask
;
629 end_offset
= offset
| mask
;
630 if (!start_offset
) /* First page is swap header. */
632 if (end_offset
>= si
->max
)
633 end_offset
= si
->max
- 1;
635 blk_start_plug(&plug
);
636 for (offset
= start_offset
; offset
<= end_offset
; offset
++) {
637 /* Ok, do the async read-ahead now */
638 page
= __read_swap_cache_async(
639 swp_entry(swp_type(entry
), offset
),
640 gfp_mask
, vma
, addr
, &page_allocated
);
643 if (page_allocated
) {
644 swap_readpage(page
, false, &splug
);
645 if (offset
!= entry_offset
) {
646 SetPageReadahead(page
);
647 count_vm_event(SWAP_RA
);
652 blk_finish_plug(&plug
);
653 swap_read_unplug(splug
);
655 lru_add_drain(); /* Push any new pages onto the LRU now */
657 /* The page was likely read above, so no need for plugging here */
658 return read_swap_cache_async(entry
, gfp_mask
, vma
, addr
, do_poll
, NULL
);
661 int init_swap_address_space(unsigned int type
, unsigned long nr_pages
)
663 struct address_space
*spaces
, *space
;
666 nr
= DIV_ROUND_UP(nr_pages
, SWAP_ADDRESS_SPACE_PAGES
);
667 spaces
= kvcalloc(nr
, sizeof(struct address_space
), GFP_KERNEL
);
670 for (i
= 0; i
< nr
; i
++) {
672 xa_init_flags(&space
->i_pages
, XA_FLAGS_LOCK_IRQ
);
673 atomic_set(&space
->i_mmap_writable
, 0);
674 space
->a_ops
= &swap_aops
;
675 /* swap cache doesn't use writeback related tags */
676 mapping_set_no_writeback_tags(space
);
678 nr_swapper_spaces
[type
] = nr
;
679 swapper_spaces
[type
] = spaces
;
684 void exit_swap_address_space(unsigned int type
)
687 struct address_space
*spaces
= swapper_spaces
[type
];
689 for (i
= 0; i
< nr_swapper_spaces
[type
]; i
++)
690 VM_WARN_ON_ONCE(!mapping_empty(&spaces
[i
]));
692 nr_swapper_spaces
[type
] = 0;
693 swapper_spaces
[type
] = NULL
;
696 static inline void swap_ra_clamp_pfn(struct vm_area_struct
*vma
,
700 unsigned long *start
,
703 *start
= max3(lpfn
, PFN_DOWN(vma
->vm_start
),
704 PFN_DOWN(faddr
& PMD_MASK
));
705 *end
= min3(rpfn
, PFN_DOWN(vma
->vm_end
),
706 PFN_DOWN((faddr
& PMD_MASK
) + PMD_SIZE
));
709 static void swap_ra_info(struct vm_fault
*vmf
,
710 struct vma_swap_readahead
*ra_info
)
712 struct vm_area_struct
*vma
= vmf
->vma
;
713 unsigned long ra_val
;
714 unsigned long faddr
, pfn
, fpfn
;
715 unsigned long start
, end
;
716 pte_t
*pte
, *orig_pte
;
717 unsigned int max_win
, hits
, prev_win
, win
, left
;
722 max_win
= 1 << min_t(unsigned int, READ_ONCE(page_cluster
),
723 SWAP_RA_ORDER_CEILING
);
729 faddr
= vmf
->address
;
730 orig_pte
= pte
= pte_offset_map(vmf
->pmd
, faddr
);
732 fpfn
= PFN_DOWN(faddr
);
733 ra_val
= GET_SWAP_RA_VAL(vma
);
734 pfn
= PFN_DOWN(SWAP_RA_ADDR(ra_val
));
735 prev_win
= SWAP_RA_WIN(ra_val
);
736 hits
= SWAP_RA_HITS(ra_val
);
737 ra_info
->win
= win
= __swapin_nr_pages(pfn
, fpfn
, hits
,
739 atomic_long_set(&vma
->swap_readahead_info
,
740 SWAP_RA_VAL(faddr
, win
, 0));
747 /* Copy the PTEs because the page table may be unmapped */
749 swap_ra_clamp_pfn(vma
, faddr
, fpfn
, fpfn
+ win
, &start
, &end
);
750 else if (pfn
== fpfn
+ 1)
751 swap_ra_clamp_pfn(vma
, faddr
, fpfn
- win
+ 1, fpfn
+ 1,
754 left
= (win
- 1) / 2;
755 swap_ra_clamp_pfn(vma
, faddr
, fpfn
- left
, fpfn
+ win
- left
,
758 ra_info
->nr_pte
= end
- start
;
759 ra_info
->offset
= fpfn
- start
;
760 pte
-= ra_info
->offset
;
764 tpte
= ra_info
->ptes
;
765 for (pfn
= start
; pfn
!= end
; pfn
++)
772 * swap_vma_readahead - swap in pages in hope we need them soon
773 * @fentry: swap entry of this memory
774 * @gfp_mask: memory allocation flags
775 * @vmf: fault information
777 * Returns the struct page for entry and addr, after queueing swapin.
779 * Primitive swap readahead code. We simply read in a few pages whose
780 * virtual addresses are around the fault address in the same vma.
782 * Caller must hold read mmap_lock if vmf->vma is not NULL.
785 static struct page
*swap_vma_readahead(swp_entry_t fentry
, gfp_t gfp_mask
,
786 struct vm_fault
*vmf
)
788 struct blk_plug plug
;
789 struct swap_iocb
*splug
= NULL
;
790 struct vm_area_struct
*vma
= vmf
->vma
;
796 struct vma_swap_readahead ra_info
= {
800 swap_ra_info(vmf
, &ra_info
);
801 if (ra_info
.win
== 1)
804 blk_start_plug(&plug
);
805 for (i
= 0, pte
= ra_info
.ptes
; i
< ra_info
.nr_pte
;
808 if (!is_swap_pte(pentry
))
810 entry
= pte_to_swp_entry(pentry
);
811 if (unlikely(non_swap_entry(entry
)))
813 page
= __read_swap_cache_async(entry
, gfp_mask
, vma
,
814 vmf
->address
, &page_allocated
);
817 if (page_allocated
) {
818 swap_readpage(page
, false, &splug
);
819 if (i
!= ra_info
.offset
) {
820 SetPageReadahead(page
);
821 count_vm_event(SWAP_RA
);
826 blk_finish_plug(&plug
);
827 swap_read_unplug(splug
);
830 /* The page was likely read above, so no need for plugging here */
831 return read_swap_cache_async(fentry
, gfp_mask
, vma
, vmf
->address
,
832 ra_info
.win
== 1, NULL
);
836 * swapin_readahead - swap in pages in hope we need them soon
837 * @entry: swap entry of this memory
838 * @gfp_mask: memory allocation flags
839 * @vmf: fault information
841 * Returns the struct page for entry and addr, after queueing swapin.
843 * It's a main entry function for swap readahead. By the configuration,
844 * it will read ahead blocks by cluster-based(ie, physical disk based)
845 * or vma-based(ie, virtual address based on faulty address) readahead.
847 struct page
*swapin_readahead(swp_entry_t entry
, gfp_t gfp_mask
,
848 struct vm_fault
*vmf
)
850 return swap_use_vma_readahead() ?
851 swap_vma_readahead(entry
, gfp_mask
, vmf
) :
852 swap_cluster_readahead(entry
, gfp_mask
, vmf
);
856 static ssize_t
vma_ra_enabled_show(struct kobject
*kobj
,
857 struct kobj_attribute
*attr
, char *buf
)
859 return sysfs_emit(buf
, "%s\n",
860 enable_vma_readahead
? "true" : "false");
862 static ssize_t
vma_ra_enabled_store(struct kobject
*kobj
,
863 struct kobj_attribute
*attr
,
864 const char *buf
, size_t count
)
868 ret
= kstrtobool(buf
, &enable_vma_readahead
);
874 static struct kobj_attribute vma_ra_enabled_attr
= __ATTR_RW(vma_ra_enabled
);
876 static struct attribute
*swap_attrs
[] = {
877 &vma_ra_enabled_attr
.attr
,
881 static const struct attribute_group swap_attr_group
= {
885 static int __init
swap_init_sysfs(void)
888 struct kobject
*swap_kobj
;
890 swap_kobj
= kobject_create_and_add("swap", mm_kobj
);
892 pr_err("failed to create swap kobject\n");
895 err
= sysfs_create_group(swap_kobj
, &swap_attr_group
);
897 pr_err("failed to register swap group\n");
903 kobject_put(swap_kobj
);
906 subsys_initcall(swap_init_sysfs
);