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>
27 * swapper_space is a fiction, retained to simplify the path through
28 * vmscan's shrink_page_list.
30 static const struct address_space_operations swap_aops
= {
31 .writepage
= swap_writepage
,
32 .set_page_dirty
= swap_set_page_dirty
,
33 #ifdef CONFIG_MIGRATION
34 .migratepage
= migrate_page
,
38 struct address_space
*swapper_spaces
[MAX_SWAPFILES
] __read_mostly
;
39 static unsigned int nr_swapper_spaces
[MAX_SWAPFILES
] __read_mostly
;
40 static bool enable_vma_readahead __read_mostly
= true;
42 #define SWAP_RA_WIN_SHIFT (PAGE_SHIFT / 2)
43 #define SWAP_RA_HITS_MASK ((1UL << SWAP_RA_WIN_SHIFT) - 1)
44 #define SWAP_RA_HITS_MAX SWAP_RA_HITS_MASK
45 #define SWAP_RA_WIN_MASK (~PAGE_MASK & ~SWAP_RA_HITS_MASK)
47 #define SWAP_RA_HITS(v) ((v) & SWAP_RA_HITS_MASK)
48 #define SWAP_RA_WIN(v) (((v) & SWAP_RA_WIN_MASK) >> SWAP_RA_WIN_SHIFT)
49 #define SWAP_RA_ADDR(v) ((v) & PAGE_MASK)
51 #define SWAP_RA_VAL(addr, win, hits) \
52 (((addr) & PAGE_MASK) | \
53 (((win) << SWAP_RA_WIN_SHIFT) & SWAP_RA_WIN_MASK) | \
54 ((hits) & SWAP_RA_HITS_MASK))
56 /* Initial readahead hits is 4 to start up with a small window */
57 #define GET_SWAP_RA_VAL(vma) \
58 (atomic_long_read(&(vma)->swap_readahead_info) ? : 4)
60 #define INC_CACHE_INFO(x) do { swap_cache_info.x++; } while (0)
61 #define ADD_CACHE_INFO(x, nr) do { swap_cache_info.x += (nr); } while (0)
64 unsigned long add_total
;
65 unsigned long del_total
;
66 unsigned long find_success
;
67 unsigned long find_total
;
70 unsigned long total_swapcache_pages(void)
72 unsigned int i
, j
, nr
;
73 unsigned long ret
= 0;
74 struct address_space
*spaces
;
75 struct swap_info_struct
*si
;
77 for (i
= 0; i
< MAX_SWAPFILES
; i
++) {
78 swp_entry_t entry
= swp_entry(i
, 1);
80 /* Avoid get_swap_device() to warn for bad swap entry */
81 if (!swp_swap_info(entry
))
83 /* Prevent swapoff to free swapper_spaces */
84 si
= get_swap_device(entry
);
87 nr
= nr_swapper_spaces
[i
];
88 spaces
= swapper_spaces
[i
];
89 for (j
= 0; j
< nr
; j
++)
90 ret
+= spaces
[j
].nrpages
;
96 static atomic_t swapin_readahead_hits
= ATOMIC_INIT(4);
98 void show_swap_cache_info(void)
100 printk("%lu pages in swap cache\n", total_swapcache_pages());
101 printk("Swap cache stats: add %lu, delete %lu, find %lu/%lu\n",
102 swap_cache_info
.add_total
, swap_cache_info
.del_total
,
103 swap_cache_info
.find_success
, swap_cache_info
.find_total
);
104 printk("Free swap = %ldkB\n",
105 get_nr_swap_pages() << (PAGE_SHIFT
- 10));
106 printk("Total swap = %lukB\n", total_swap_pages
<< (PAGE_SHIFT
- 10));
109 void *get_shadow_from_swap_cache(swp_entry_t entry
)
111 struct address_space
*address_space
= swap_address_space(entry
);
112 pgoff_t idx
= swp_offset(entry
);
115 page
= find_get_entry(address_space
, idx
);
116 if (xa_is_value(page
))
124 * add_to_swap_cache resembles add_to_page_cache_locked on swapper_space,
125 * but sets SwapCache flag and private instead of mapping and index.
127 int add_to_swap_cache(struct page
*page
, swp_entry_t entry
,
128 gfp_t gfp
, void **shadowp
)
130 struct address_space
*address_space
= swap_address_space(entry
);
131 pgoff_t idx
= swp_offset(entry
);
132 XA_STATE_ORDER(xas
, &address_space
->i_pages
, idx
, compound_order(page
));
133 unsigned long i
, nr
= hpage_nr_pages(page
);
136 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
137 VM_BUG_ON_PAGE(PageSwapCache(page
), page
);
138 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
140 page_ref_add(page
, nr
);
141 SetPageSwapCache(page
);
144 unsigned long nr_shadows
= 0;
147 xas_create_range(&xas
);
150 for (i
= 0; i
< nr
; i
++) {
151 VM_BUG_ON_PAGE(xas
.xa_index
!= idx
+ i
, page
);
152 old
= xas_load(&xas
);
153 if (xa_is_value(old
)) {
158 set_page_private(page
+ i
, entry
.val
+ i
);
159 xas_store(&xas
, page
);
162 address_space
->nrexceptional
-= nr_shadows
;
163 address_space
->nrpages
+= nr
;
164 __mod_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, nr
);
165 ADD_CACHE_INFO(add_total
, nr
);
167 xas_unlock_irq(&xas
);
168 } while (xas_nomem(&xas
, gfp
));
170 if (!xas_error(&xas
))
173 ClearPageSwapCache(page
);
174 page_ref_sub(page
, nr
);
175 return xas_error(&xas
);
179 * This must be called only on pages that have
180 * been verified to be in the swap cache.
182 void __delete_from_swap_cache(struct page
*page
,
183 swp_entry_t entry
, void *shadow
)
185 struct address_space
*address_space
= swap_address_space(entry
);
186 int i
, nr
= hpage_nr_pages(page
);
187 pgoff_t idx
= swp_offset(entry
);
188 XA_STATE(xas
, &address_space
->i_pages
, idx
);
190 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
191 VM_BUG_ON_PAGE(!PageSwapCache(page
), page
);
192 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
194 for (i
= 0; i
< nr
; i
++) {
195 void *entry
= xas_store(&xas
, shadow
);
196 VM_BUG_ON_PAGE(entry
!= page
, entry
);
197 set_page_private(page
+ i
, 0);
200 ClearPageSwapCache(page
);
202 address_space
->nrexceptional
+= nr
;
203 address_space
->nrpages
-= nr
;
204 __mod_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, -nr
);
205 ADD_CACHE_INFO(del_total
, nr
);
209 * add_to_swap - allocate swap space for a page
210 * @page: page we want to move to swap
212 * Allocate swap space for the page and add the page to the
213 * swap cache. Caller needs to hold the page lock.
215 int add_to_swap(struct page
*page
)
220 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
221 VM_BUG_ON_PAGE(!PageUptodate(page
), page
);
223 entry
= get_swap_page(page
);
228 * XArray node allocations from PF_MEMALLOC contexts could
229 * completely exhaust the page allocator. __GFP_NOMEMALLOC
230 * stops emergency reserves from being allocated.
232 * TODO: this could cause a theoretical memory reclaim
233 * deadlock in the swap out path.
236 * Add it to the swap cache.
238 err
= add_to_swap_cache(page
, entry
,
239 __GFP_HIGH
|__GFP_NOMEMALLOC
|__GFP_NOWARN
, NULL
);
242 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
243 * clear SWAP_HAS_CACHE flag.
247 * Normally the page will be dirtied in unmap because its pte should be
248 * dirty. A special case is MADV_FREE page. The page'e pte could have
249 * dirty bit cleared but the page's SwapBacked bit is still set because
250 * clearing the dirty bit and SwapBacked bit has no lock protected. For
251 * such page, unmap will not set dirty bit for it, so page reclaim will
252 * not write the page out. This can cause data corruption when the page
253 * is swap in later. Always setting the dirty bit for the page solves
256 set_page_dirty(page
);
261 put_swap_page(page
, entry
);
266 * This must be called only on pages that have
267 * been verified to be in the swap cache and locked.
268 * It will never put the page into the free list,
269 * the caller has a reference on the page.
271 void delete_from_swap_cache(struct page
*page
)
273 swp_entry_t entry
= { .val
= page_private(page
) };
274 struct address_space
*address_space
= swap_address_space(entry
);
276 xa_lock_irq(&address_space
->i_pages
);
277 __delete_from_swap_cache(page
, entry
, NULL
);
278 xa_unlock_irq(&address_space
->i_pages
);
280 put_swap_page(page
, entry
);
281 page_ref_sub(page
, hpage_nr_pages(page
));
284 void clear_shadow_from_swap_cache(int type
, unsigned long begin
,
287 unsigned long curr
= begin
;
291 unsigned long nr_shadows
= 0;
292 swp_entry_t entry
= swp_entry(type
, curr
);
293 struct address_space
*address_space
= swap_address_space(entry
);
294 XA_STATE(xas
, &address_space
->i_pages
, curr
);
296 xa_lock_irq(&address_space
->i_pages
);
297 xas_for_each(&xas
, old
, end
) {
298 if (!xa_is_value(old
))
300 xas_store(&xas
, NULL
);
303 address_space
->nrexceptional
-= nr_shadows
;
304 xa_unlock_irq(&address_space
->i_pages
);
306 /* search the next swapcache until we meet end */
307 curr
>>= SWAP_ADDRESS_SPACE_SHIFT
;
309 curr
<<= SWAP_ADDRESS_SPACE_SHIFT
;
316 * If we are the only user, then try to free up the swap cache.
318 * Its ok to check for PageSwapCache without the page lock
319 * here because we are going to recheck again inside
320 * try_to_free_swap() _with_ the lock.
323 static inline void free_swap_cache(struct page
*page
)
325 if (PageSwapCache(page
) && !page_mapped(page
) && trylock_page(page
)) {
326 try_to_free_swap(page
);
332 * Perform a free_page(), also freeing any swap cache associated with
333 * this page if it is the last user of the page.
335 void free_page_and_swap_cache(struct page
*page
)
337 free_swap_cache(page
);
338 if (!is_huge_zero_page(page
))
343 * Passed an array of pages, drop them all from swapcache and then release
344 * them. They are removed from the LRU and freed if this is their last use.
346 void free_pages_and_swap_cache(struct page
**pages
, int nr
)
348 struct page
**pagep
= pages
;
352 for (i
= 0; i
< nr
; i
++)
353 free_swap_cache(pagep
[i
]);
354 release_pages(pagep
, nr
);
357 static inline bool swap_use_vma_readahead(void)
359 return READ_ONCE(enable_vma_readahead
) && !atomic_read(&nr_rotate_swap
);
363 * Lookup a swap entry in the swap cache. A found page will be returned
364 * unlocked and with its refcount incremented - we rely on the kernel
365 * lock getting page table operations atomic even if we drop the page
366 * lock before returning.
368 struct page
*lookup_swap_cache(swp_entry_t entry
, struct vm_area_struct
*vma
,
372 struct swap_info_struct
*si
;
374 si
= get_swap_device(entry
);
377 page
= find_get_page(swap_address_space(entry
), swp_offset(entry
));
380 INC_CACHE_INFO(find_total
);
382 bool vma_ra
= swap_use_vma_readahead();
385 INC_CACHE_INFO(find_success
);
387 * At the moment, we don't support PG_readahead for anon THP
388 * so let's bail out rather than confusing the readahead stat.
390 if (unlikely(PageTransCompound(page
)))
393 readahead
= TestClearPageReadahead(page
);
395 unsigned long ra_val
;
398 ra_val
= GET_SWAP_RA_VAL(vma
);
399 win
= SWAP_RA_WIN(ra_val
);
400 hits
= SWAP_RA_HITS(ra_val
);
402 hits
= min_t(int, hits
+ 1, SWAP_RA_HITS_MAX
);
403 atomic_long_set(&vma
->swap_readahead_info
,
404 SWAP_RA_VAL(addr
, win
, hits
));
408 count_vm_event(SWAP_RA_HIT
);
410 atomic_inc(&swapin_readahead_hits
);
417 struct page
*__read_swap_cache_async(swp_entry_t entry
, gfp_t gfp_mask
,
418 struct vm_area_struct
*vma
, unsigned long addr
,
419 bool *new_page_allocated
)
421 struct swap_info_struct
*si
;
425 *new_page_allocated
= false;
430 * First check the swap cache. Since this is normally
431 * called after lookup_swap_cache() failed, re-calling
432 * that would confuse statistics.
434 si
= get_swap_device(entry
);
437 page
= find_get_page(swap_address_space(entry
),
444 * Just skip read ahead for unused swap slot.
445 * During swap_off when swap_slot_cache is disabled,
446 * we have to handle the race between putting
447 * swap entry in swap cache and marking swap slot
448 * as SWAP_HAS_CACHE. That's done in later part of code or
449 * else swap_off will be aborted if we return NULL.
451 if (!__swp_swapcount(entry
) && swap_slot_cache_enabled
)
455 * Get a new page to read into from swap. Allocate it now,
456 * before marking swap_map SWAP_HAS_CACHE, when -EEXIST will
457 * cause any racers to loop around until we add it to cache.
459 page
= alloc_page_vma(gfp_mask
, vma
, addr
);
464 * Swap entry may have been freed since our caller observed it.
466 err
= swapcache_prepare(entry
);
475 * We might race against __delete_from_swap_cache(), and
476 * stumble across a swap_map entry whose SWAP_HAS_CACHE
477 * has not yet been cleared. Or race against another
478 * __read_swap_cache_async(), which has set SWAP_HAS_CACHE
479 * in swap_map, but not yet added its page to swap cache.
485 * The swap entry is ours to swap in. Prepare the new page.
488 __SetPageLocked(page
);
489 __SetPageSwapBacked(page
);
491 /* May fail (-ENOMEM) if XArray node allocation failed. */
492 if (add_to_swap_cache(page
, entry
, gfp_mask
& GFP_RECLAIM_MASK
, &shadow
)) {
493 put_swap_page(page
, entry
);
497 if (mem_cgroup_charge(page
, NULL
, gfp_mask
)) {
498 delete_from_swap_cache(page
);
503 workingset_refault(page
, shadow
);
505 /* Caller will initiate read into locked page */
506 SetPageWorkingset(page
);
508 *new_page_allocated
= true;
518 * Locate a page of swap in physical memory, reserving swap cache space
519 * and reading the disk if it is not already cached.
520 * A failure return means that either the page allocation failed or that
521 * the swap entry is no longer in use.
523 struct page
*read_swap_cache_async(swp_entry_t entry
, gfp_t gfp_mask
,
524 struct vm_area_struct
*vma
, unsigned long addr
, bool do_poll
)
526 bool page_was_allocated
;
527 struct page
*retpage
= __read_swap_cache_async(entry
, gfp_mask
,
528 vma
, addr
, &page_was_allocated
);
530 if (page_was_allocated
)
531 swap_readpage(retpage
, do_poll
);
536 static unsigned int __swapin_nr_pages(unsigned long prev_offset
,
537 unsigned long offset
,
542 unsigned int pages
, last_ra
;
545 * This heuristic has been found to work well on both sequential and
546 * random loads, swapping to hard disk or to SSD: please don't ask
547 * what the "+ 2" means, it just happens to work well, that's all.
552 * We can have no readahead hits to judge by: but must not get
553 * stuck here forever, so check for an adjacent offset instead
554 * (and don't even bother to check whether swap type is same).
556 if (offset
!= prev_offset
+ 1 && offset
!= prev_offset
- 1)
559 unsigned int roundup
= 4;
560 while (roundup
< pages
)
565 if (pages
> max_pages
)
568 /* Don't shrink readahead too fast */
569 last_ra
= prev_win
/ 2;
576 static unsigned long swapin_nr_pages(unsigned long offset
)
578 static unsigned long prev_offset
;
579 unsigned int hits
, pages
, max_pages
;
580 static atomic_t last_readahead_pages
;
582 max_pages
= 1 << READ_ONCE(page_cluster
);
586 hits
= atomic_xchg(&swapin_readahead_hits
, 0);
587 pages
= __swapin_nr_pages(READ_ONCE(prev_offset
), offset
, hits
,
589 atomic_read(&last_readahead_pages
));
591 WRITE_ONCE(prev_offset
, offset
);
592 atomic_set(&last_readahead_pages
, pages
);
598 * swap_cluster_readahead - swap in pages in hope we need them soon
599 * @entry: swap entry of this memory
600 * @gfp_mask: memory allocation flags
601 * @vmf: fault information
603 * Returns the struct page for entry and addr, after queueing swapin.
605 * Primitive swap readahead code. We simply read an aligned block of
606 * (1 << page_cluster) entries in the swap area. This method is chosen
607 * because it doesn't cost us any seek time. We also make sure to queue
608 * the 'original' request together with the readahead ones...
610 * This has been extended to use the NUMA policies from the mm triggering
613 * Caller must hold read mmap_lock if vmf->vma is not NULL.
615 struct page
*swap_cluster_readahead(swp_entry_t entry
, gfp_t gfp_mask
,
616 struct vm_fault
*vmf
)
619 unsigned long entry_offset
= swp_offset(entry
);
620 unsigned long offset
= entry_offset
;
621 unsigned long start_offset
, end_offset
;
623 struct swap_info_struct
*si
= swp_swap_info(entry
);
624 struct blk_plug plug
;
625 bool do_poll
= true, page_allocated
;
626 struct vm_area_struct
*vma
= vmf
->vma
;
627 unsigned long addr
= vmf
->address
;
629 mask
= swapin_nr_pages(offset
) - 1;
633 /* Test swap type to make sure the dereference is safe */
634 if (likely(si
->flags
& (SWP_BLKDEV
| SWP_FS
))) {
635 struct inode
*inode
= si
->swap_file
->f_mapping
->host
;
636 if (inode_read_congested(inode
))
641 /* Read a page_cluster sized and aligned cluster around offset. */
642 start_offset
= offset
& ~mask
;
643 end_offset
= offset
| mask
;
644 if (!start_offset
) /* First page is swap header. */
646 if (end_offset
>= si
->max
)
647 end_offset
= si
->max
- 1;
649 blk_start_plug(&plug
);
650 for (offset
= start_offset
; offset
<= end_offset
; offset
++) {
651 /* Ok, do the async read-ahead now */
652 page
= __read_swap_cache_async(
653 swp_entry(swp_type(entry
), offset
),
654 gfp_mask
, vma
, addr
, &page_allocated
);
657 if (page_allocated
) {
658 swap_readpage(page
, false);
659 if (offset
!= entry_offset
) {
660 SetPageReadahead(page
);
661 count_vm_event(SWAP_RA
);
666 blk_finish_plug(&plug
);
668 lru_add_drain(); /* Push any new pages onto the LRU now */
670 return read_swap_cache_async(entry
, gfp_mask
, vma
, addr
, do_poll
);
673 int init_swap_address_space(unsigned int type
, unsigned long nr_pages
)
675 struct address_space
*spaces
, *space
;
678 nr
= DIV_ROUND_UP(nr_pages
, SWAP_ADDRESS_SPACE_PAGES
);
679 spaces
= kvcalloc(nr
, sizeof(struct address_space
), GFP_KERNEL
);
682 for (i
= 0; i
< nr
; i
++) {
684 xa_init_flags(&space
->i_pages
, XA_FLAGS_LOCK_IRQ
);
685 atomic_set(&space
->i_mmap_writable
, 0);
686 space
->a_ops
= &swap_aops
;
687 /* swap cache doesn't use writeback related tags */
688 mapping_set_no_writeback_tags(space
);
690 nr_swapper_spaces
[type
] = nr
;
691 swapper_spaces
[type
] = spaces
;
696 void exit_swap_address_space(unsigned int type
)
698 kvfree(swapper_spaces
[type
]);
699 nr_swapper_spaces
[type
] = 0;
700 swapper_spaces
[type
] = NULL
;
703 static inline void swap_ra_clamp_pfn(struct vm_area_struct
*vma
,
707 unsigned long *start
,
710 *start
= max3(lpfn
, PFN_DOWN(vma
->vm_start
),
711 PFN_DOWN(faddr
& PMD_MASK
));
712 *end
= min3(rpfn
, PFN_DOWN(vma
->vm_end
),
713 PFN_DOWN((faddr
& PMD_MASK
) + PMD_SIZE
));
716 static void swap_ra_info(struct vm_fault
*vmf
,
717 struct vma_swap_readahead
*ra_info
)
719 struct vm_area_struct
*vma
= vmf
->vma
;
720 unsigned long ra_val
;
722 unsigned long faddr
, pfn
, fpfn
;
723 unsigned long start
, end
;
724 pte_t
*pte
, *orig_pte
;
725 unsigned int max_win
, hits
, prev_win
, win
, left
;
730 max_win
= 1 << min_t(unsigned int, READ_ONCE(page_cluster
),
731 SWAP_RA_ORDER_CEILING
);
737 faddr
= vmf
->address
;
738 orig_pte
= pte
= pte_offset_map(vmf
->pmd
, faddr
);
739 entry
= pte_to_swp_entry(*pte
);
740 if ((unlikely(non_swap_entry(entry
)))) {
745 fpfn
= PFN_DOWN(faddr
);
746 ra_val
= GET_SWAP_RA_VAL(vma
);
747 pfn
= PFN_DOWN(SWAP_RA_ADDR(ra_val
));
748 prev_win
= SWAP_RA_WIN(ra_val
);
749 hits
= SWAP_RA_HITS(ra_val
);
750 ra_info
->win
= win
= __swapin_nr_pages(pfn
, fpfn
, hits
,
752 atomic_long_set(&vma
->swap_readahead_info
,
753 SWAP_RA_VAL(faddr
, win
, 0));
760 /* Copy the PTEs because the page table may be unmapped */
762 swap_ra_clamp_pfn(vma
, faddr
, fpfn
, fpfn
+ win
, &start
, &end
);
763 else if (pfn
== fpfn
+ 1)
764 swap_ra_clamp_pfn(vma
, faddr
, fpfn
- win
+ 1, fpfn
+ 1,
767 left
= (win
- 1) / 2;
768 swap_ra_clamp_pfn(vma
, faddr
, fpfn
- left
, fpfn
+ win
- left
,
771 ra_info
->nr_pte
= end
- start
;
772 ra_info
->offset
= fpfn
- start
;
773 pte
-= ra_info
->offset
;
777 tpte
= ra_info
->ptes
;
778 for (pfn
= start
; pfn
!= end
; pfn
++)
785 * swap_vma_readahead - swap in pages in hope we need them soon
786 * @fentry: swap entry of this memory
787 * @gfp_mask: memory allocation flags
788 * @vmf: fault information
790 * Returns the struct page for entry and addr, after queueing swapin.
792 * Primitive swap readahead code. We simply read in a few pages whoes
793 * virtual addresses are around the fault address in the same vma.
795 * Caller must hold read mmap_lock if vmf->vma is not NULL.
798 static struct page
*swap_vma_readahead(swp_entry_t fentry
, gfp_t gfp_mask
,
799 struct vm_fault
*vmf
)
801 struct blk_plug plug
;
802 struct vm_area_struct
*vma
= vmf
->vma
;
808 struct vma_swap_readahead ra_info
= {0,};
810 swap_ra_info(vmf
, &ra_info
);
811 if (ra_info
.win
== 1)
814 blk_start_plug(&plug
);
815 for (i
= 0, pte
= ra_info
.ptes
; i
< ra_info
.nr_pte
;
818 if (pte_none(pentry
))
820 if (pte_present(pentry
))
822 entry
= pte_to_swp_entry(pentry
);
823 if (unlikely(non_swap_entry(entry
)))
825 page
= __read_swap_cache_async(entry
, gfp_mask
, vma
,
826 vmf
->address
, &page_allocated
);
829 if (page_allocated
) {
830 swap_readpage(page
, false);
831 if (i
!= ra_info
.offset
) {
832 SetPageReadahead(page
);
833 count_vm_event(SWAP_RA
);
838 blk_finish_plug(&plug
);
841 return read_swap_cache_async(fentry
, gfp_mask
, vma
, vmf
->address
,
846 * swapin_readahead - swap in pages in hope we need them soon
847 * @entry: swap entry of this memory
848 * @gfp_mask: memory allocation flags
849 * @vmf: fault information
851 * Returns the struct page for entry and addr, after queueing swapin.
853 * It's a main entry function for swap readahead. By the configuration,
854 * it will read ahead blocks by cluster-based(ie, physical disk based)
855 * or vma-based(ie, virtual address based on faulty address) readahead.
857 struct page
*swapin_readahead(swp_entry_t entry
, gfp_t gfp_mask
,
858 struct vm_fault
*vmf
)
860 return swap_use_vma_readahead() ?
861 swap_vma_readahead(entry
, gfp_mask
, vmf
) :
862 swap_cluster_readahead(entry
, gfp_mask
, vmf
);
866 static ssize_t
vma_ra_enabled_show(struct kobject
*kobj
,
867 struct kobj_attribute
*attr
, char *buf
)
869 return sprintf(buf
, "%s\n", enable_vma_readahead
? "true" : "false");
871 static ssize_t
vma_ra_enabled_store(struct kobject
*kobj
,
872 struct kobj_attribute
*attr
,
873 const char *buf
, size_t count
)
875 if (!strncmp(buf
, "true", 4) || !strncmp(buf
, "1", 1))
876 enable_vma_readahead
= true;
877 else if (!strncmp(buf
, "false", 5) || !strncmp(buf
, "0", 1))
878 enable_vma_readahead
= false;
884 static struct kobj_attribute vma_ra_enabled_attr
=
885 __ATTR(vma_ra_enabled
, 0644, vma_ra_enabled_show
,
886 vma_ra_enabled_store
);
888 static struct attribute
*swap_attrs
[] = {
889 &vma_ra_enabled_attr
.attr
,
893 static struct attribute_group swap_attr_group
= {
897 static int __init
swap_init_sysfs(void)
900 struct kobject
*swap_kobj
;
902 swap_kobj
= kobject_create_and_add("swap", mm_kobj
);
904 pr_err("failed to create swap kobject\n");
907 err
= sysfs_create_group(swap_kobj
, &swap_attr_group
);
909 pr_err("failed to register swap group\n");
915 kobject_put(swap_kobj
);
918 subsys_initcall(swap_init_sysfs
);