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
= false;
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
->nrpages
+= nr
;
136 __mod_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, nr
);
137 __mod_lruvec_page_state(page
, NR_SWAPCACHE
, nr
);
138 ADD_CACHE_INFO(add_total
, nr
);
140 xas_unlock_irq(&xas
);
141 } while (xas_nomem(&xas
, gfp
));
143 if (!xas_error(&xas
))
146 ClearPageSwapCache(page
);
147 page_ref_sub(page
, nr
);
148 return xas_error(&xas
);
152 * This must be called only on pages that have
153 * been verified to be in the swap cache.
155 void __delete_from_swap_cache(struct page
*page
,
156 swp_entry_t entry
, void *shadow
)
158 struct address_space
*address_space
= swap_address_space(entry
);
159 int i
, nr
= thp_nr_pages(page
);
160 pgoff_t idx
= swp_offset(entry
);
161 XA_STATE(xas
, &address_space
->i_pages
, idx
);
163 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
164 VM_BUG_ON_PAGE(!PageSwapCache(page
), page
);
165 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
167 for (i
= 0; i
< nr
; i
++) {
168 void *entry
= xas_store(&xas
, shadow
);
169 VM_BUG_ON_PAGE(entry
!= page
, entry
);
170 set_page_private(page
+ i
, 0);
173 ClearPageSwapCache(page
);
174 address_space
->nrpages
-= nr
;
175 __mod_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, -nr
);
176 __mod_lruvec_page_state(page
, NR_SWAPCACHE
, -nr
);
177 ADD_CACHE_INFO(del_total
, nr
);
181 * add_to_swap - allocate swap space for a page
182 * @page: page we want to move to swap
184 * Allocate swap space for the page and add the page to the
185 * swap cache. Caller needs to hold the page lock.
187 int add_to_swap(struct page
*page
)
192 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
193 VM_BUG_ON_PAGE(!PageUptodate(page
), page
);
195 entry
= get_swap_page(page
);
200 * XArray node allocations from PF_MEMALLOC contexts could
201 * completely exhaust the page allocator. __GFP_NOMEMALLOC
202 * stops emergency reserves from being allocated.
204 * TODO: this could cause a theoretical memory reclaim
205 * deadlock in the swap out path.
208 * Add it to the swap cache.
210 err
= add_to_swap_cache(page
, entry
,
211 __GFP_HIGH
|__GFP_NOMEMALLOC
|__GFP_NOWARN
, NULL
);
214 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
215 * clear SWAP_HAS_CACHE flag.
219 * Normally the page will be dirtied in unmap because its pte should be
220 * dirty. A special case is MADV_FREE page. The page's pte could have
221 * dirty bit cleared but the page's SwapBacked bit is still set because
222 * clearing the dirty bit and SwapBacked bit has no lock protected. For
223 * such page, unmap will not set dirty bit for it, so page reclaim will
224 * not write the page out. This can cause data corruption when the page
225 * is swap in later. Always setting the dirty bit for the page solves
228 set_page_dirty(page
);
233 put_swap_page(page
, entry
);
238 * This must be called only on pages that have
239 * been verified to be in the swap cache and locked.
240 * It will never put the page into the free list,
241 * the caller has a reference on the page.
243 void delete_from_swap_cache(struct page
*page
)
245 swp_entry_t entry
= { .val
= page_private(page
) };
246 struct address_space
*address_space
= swap_address_space(entry
);
248 xa_lock_irq(&address_space
->i_pages
);
249 __delete_from_swap_cache(page
, entry
, NULL
);
250 xa_unlock_irq(&address_space
->i_pages
);
252 put_swap_page(page
, entry
);
253 page_ref_sub(page
, thp_nr_pages(page
));
256 void clear_shadow_from_swap_cache(int type
, unsigned long begin
,
259 unsigned long curr
= begin
;
263 unsigned long nr_shadows
= 0;
264 swp_entry_t entry
= swp_entry(type
, curr
);
265 struct address_space
*address_space
= swap_address_space(entry
);
266 XA_STATE(xas
, &address_space
->i_pages
, curr
);
268 xa_lock_irq(&address_space
->i_pages
);
269 xas_for_each(&xas
, old
, end
) {
270 if (!xa_is_value(old
))
272 xas_store(&xas
, NULL
);
275 xa_unlock_irq(&address_space
->i_pages
);
277 /* search the next swapcache until we meet end */
278 curr
>>= SWAP_ADDRESS_SPACE_SHIFT
;
280 curr
<<= SWAP_ADDRESS_SPACE_SHIFT
;
287 * If we are the only user, then try to free up the swap cache.
289 * Its ok to check for PageSwapCache without the page lock
290 * here because we are going to recheck again inside
291 * try_to_free_swap() _with_ the lock.
294 static inline void free_swap_cache(struct page
*page
)
296 if (PageSwapCache(page
) && !page_mapped(page
) && trylock_page(page
)) {
297 try_to_free_swap(page
);
303 * Perform a free_page(), also freeing any swap cache associated with
304 * this page if it is the last user of the page.
306 void free_page_and_swap_cache(struct page
*page
)
308 free_swap_cache(page
);
309 if (!is_huge_zero_page(page
))
314 * Passed an array of pages, drop them all from swapcache and then release
315 * them. They are removed from the LRU and freed if this is their last use.
317 void free_pages_and_swap_cache(struct page
**pages
, int nr
)
319 struct page
**pagep
= pages
;
323 for (i
= 0; i
< nr
; i
++)
324 free_swap_cache(pagep
[i
]);
325 release_pages(pagep
, nr
);
328 static inline bool swap_use_vma_readahead(void)
330 return READ_ONCE(enable_vma_readahead
) && !atomic_read(&nr_rotate_swap
);
334 * Lookup a swap entry in the swap cache. A found page will be returned
335 * unlocked and with its refcount incremented - we rely on the kernel
336 * lock getting page table operations atomic even if we drop the page
337 * lock before returning.
339 struct page
*lookup_swap_cache(swp_entry_t entry
, struct vm_area_struct
*vma
,
343 struct swap_info_struct
*si
;
345 si
= get_swap_device(entry
);
348 page
= find_get_page(swap_address_space(entry
), swp_offset(entry
));
351 INC_CACHE_INFO(find_total
);
353 bool vma_ra
= swap_use_vma_readahead();
356 INC_CACHE_INFO(find_success
);
358 * At the moment, we don't support PG_readahead for anon THP
359 * so let's bail out rather than confusing the readahead stat.
361 if (unlikely(PageTransCompound(page
)))
364 readahead
= TestClearPageReadahead(page
);
366 unsigned long ra_val
;
369 ra_val
= GET_SWAP_RA_VAL(vma
);
370 win
= SWAP_RA_WIN(ra_val
);
371 hits
= SWAP_RA_HITS(ra_val
);
373 hits
= min_t(int, hits
+ 1, SWAP_RA_HITS_MAX
);
374 atomic_long_set(&vma
->swap_readahead_info
,
375 SWAP_RA_VAL(addr
, win
, hits
));
379 count_vm_event(SWAP_RA_HIT
);
381 atomic_inc(&swapin_readahead_hits
);
389 * find_get_incore_page - Find and get a page from the page or swap caches.
390 * @mapping: The address_space to search.
391 * @index: The page cache index.
393 * This differs from find_get_page() in that it will also look for the
394 * page in the swap cache.
396 * Return: The found page or %NULL.
398 struct page
*find_get_incore_page(struct address_space
*mapping
, pgoff_t index
)
401 struct swap_info_struct
*si
;
402 struct page
*page
= pagecache_get_page(mapping
, index
,
403 FGP_ENTRY
| FGP_HEAD
, 0);
407 if (!xa_is_value(page
))
408 return find_subpage(page
, index
);
409 if (!shmem_mapping(mapping
))
412 swp
= radix_to_swp_entry(page
);
413 /* Prevent swapoff from happening to us */
414 si
= get_swap_device(swp
);
417 page
= find_get_page(swap_address_space(swp
), swp_offset(swp
));
422 struct page
*__read_swap_cache_async(swp_entry_t entry
, gfp_t gfp_mask
,
423 struct vm_area_struct
*vma
, unsigned long addr
,
424 bool *new_page_allocated
)
426 struct swap_info_struct
*si
;
430 *new_page_allocated
= false;
435 * First check the swap cache. Since this is normally
436 * called after lookup_swap_cache() failed, re-calling
437 * that would confuse statistics.
439 si
= get_swap_device(entry
);
442 page
= find_get_page(swap_address_space(entry
),
449 * Just skip read ahead for unused swap slot.
450 * During swap_off when swap_slot_cache is disabled,
451 * we have to handle the race between putting
452 * swap entry in swap cache and marking swap slot
453 * as SWAP_HAS_CACHE. That's done in later part of code or
454 * else swap_off will be aborted if we return NULL.
456 if (!__swp_swapcount(entry
) && swap_slot_cache_enabled
)
460 * Get a new page to read into from swap. Allocate it now,
461 * before marking swap_map SWAP_HAS_CACHE, when -EEXIST will
462 * cause any racers to loop around until we add it to cache.
464 page
= alloc_page_vma(gfp_mask
, vma
, addr
);
469 * Swap entry may have been freed since our caller observed it.
471 err
= swapcache_prepare(entry
);
480 * We might race against __delete_from_swap_cache(), and
481 * stumble across a swap_map entry whose SWAP_HAS_CACHE
482 * has not yet been cleared. Or race against another
483 * __read_swap_cache_async(), which has set SWAP_HAS_CACHE
484 * in swap_map, but not yet added its page to swap cache.
490 * The swap entry is ours to swap in. Prepare the new page.
493 __SetPageLocked(page
);
494 __SetPageSwapBacked(page
);
496 if (mem_cgroup_swapin_charge_page(page
, NULL
, gfp_mask
, entry
))
499 /* May fail (-ENOMEM) if XArray node allocation failed. */
500 if (add_to_swap_cache(page
, entry
, gfp_mask
& GFP_RECLAIM_MASK
, &shadow
))
503 mem_cgroup_swapin_uncharge_swap(entry
);
506 workingset_refault(page
, shadow
);
508 /* Caller will initiate read into locked page */
510 *new_page_allocated
= true;
514 put_swap_page(page
, entry
);
521 * Locate a page of swap in physical memory, reserving swap cache space
522 * and reading the disk if it is not already cached.
523 * A failure return means that either the page allocation failed or that
524 * the swap entry is no longer in use.
526 struct page
*read_swap_cache_async(swp_entry_t entry
, gfp_t gfp_mask
,
527 struct vm_area_struct
*vma
, unsigned long addr
, bool do_poll
)
529 bool page_was_allocated
;
530 struct page
*retpage
= __read_swap_cache_async(entry
, gfp_mask
,
531 vma
, addr
, &page_was_allocated
);
533 if (page_was_allocated
)
534 swap_readpage(retpage
, do_poll
);
539 static unsigned int __swapin_nr_pages(unsigned long prev_offset
,
540 unsigned long offset
,
545 unsigned int pages
, last_ra
;
548 * This heuristic has been found to work well on both sequential and
549 * random loads, swapping to hard disk or to SSD: please don't ask
550 * what the "+ 2" means, it just happens to work well, that's all.
555 * We can have no readahead hits to judge by: but must not get
556 * stuck here forever, so check for an adjacent offset instead
557 * (and don't even bother to check whether swap type is same).
559 if (offset
!= prev_offset
+ 1 && offset
!= prev_offset
- 1)
562 unsigned int roundup
= 4;
563 while (roundup
< pages
)
568 if (pages
> max_pages
)
571 /* Don't shrink readahead too fast */
572 last_ra
= prev_win
/ 2;
579 static unsigned long swapin_nr_pages(unsigned long offset
)
581 static unsigned long prev_offset
;
582 unsigned int hits
, pages
, max_pages
;
583 static atomic_t last_readahead_pages
;
585 max_pages
= 1 << READ_ONCE(page_cluster
);
589 hits
= atomic_xchg(&swapin_readahead_hits
, 0);
590 pages
= __swapin_nr_pages(READ_ONCE(prev_offset
), offset
, hits
,
592 atomic_read(&last_readahead_pages
));
594 WRITE_ONCE(prev_offset
, offset
);
595 atomic_set(&last_readahead_pages
, pages
);
601 * swap_cluster_readahead - swap in pages in hope we need them soon
602 * @entry: swap entry of this memory
603 * @gfp_mask: memory allocation flags
604 * @vmf: fault information
606 * Returns the struct page for entry and addr, after queueing swapin.
608 * Primitive swap readahead code. We simply read an aligned block of
609 * (1 << page_cluster) entries in the swap area. This method is chosen
610 * because it doesn't cost us any seek time. We also make sure to queue
611 * the 'original' request together with the readahead ones...
613 * This has been extended to use the NUMA policies from the mm triggering
616 * Caller must hold read mmap_lock if vmf->vma is not NULL.
618 struct page
*swap_cluster_readahead(swp_entry_t entry
, gfp_t gfp_mask
,
619 struct vm_fault
*vmf
)
622 unsigned long entry_offset
= swp_offset(entry
);
623 unsigned long offset
= entry_offset
;
624 unsigned long start_offset
, end_offset
;
626 struct swap_info_struct
*si
= swp_swap_info(entry
);
627 struct blk_plug plug
;
628 bool do_poll
= true, page_allocated
;
629 struct vm_area_struct
*vma
= vmf
->vma
;
630 unsigned long addr
= vmf
->address
;
632 mask
= swapin_nr_pages(offset
) - 1;
636 /* Test swap type to make sure the dereference is safe */
637 if (likely(si
->flags
& (SWP_BLKDEV
| SWP_FS_OPS
))) {
638 struct inode
*inode
= si
->swap_file
->f_mapping
->host
;
639 if (inode_read_congested(inode
))
644 /* Read a page_cluster sized and aligned cluster around offset. */
645 start_offset
= offset
& ~mask
;
646 end_offset
= offset
| mask
;
647 if (!start_offset
) /* First page is swap header. */
649 if (end_offset
>= si
->max
)
650 end_offset
= si
->max
- 1;
652 blk_start_plug(&plug
);
653 for (offset
= start_offset
; offset
<= end_offset
; offset
++) {
654 /* Ok, do the async read-ahead now */
655 page
= __read_swap_cache_async(
656 swp_entry(swp_type(entry
), offset
),
657 gfp_mask
, vma
, addr
, &page_allocated
);
660 if (page_allocated
) {
661 swap_readpage(page
, false);
662 if (offset
!= entry_offset
) {
663 SetPageReadahead(page
);
664 count_vm_event(SWAP_RA
);
669 blk_finish_plug(&plug
);
671 lru_add_drain(); /* Push any new pages onto the LRU now */
673 return read_swap_cache_async(entry
, gfp_mask
, vma
, addr
, do_poll
);
676 int init_swap_address_space(unsigned int type
, unsigned long nr_pages
)
678 struct address_space
*spaces
, *space
;
681 nr
= DIV_ROUND_UP(nr_pages
, SWAP_ADDRESS_SPACE_PAGES
);
682 spaces
= kvcalloc(nr
, sizeof(struct address_space
), GFP_KERNEL
);
685 for (i
= 0; i
< nr
; i
++) {
687 xa_init_flags(&space
->i_pages
, XA_FLAGS_LOCK_IRQ
);
688 atomic_set(&space
->i_mmap_writable
, 0);
689 space
->a_ops
= &swap_aops
;
690 /* swap cache doesn't use writeback related tags */
691 mapping_set_no_writeback_tags(space
);
693 nr_swapper_spaces
[type
] = nr
;
694 swapper_spaces
[type
] = spaces
;
699 void exit_swap_address_space(unsigned int type
)
701 kvfree(swapper_spaces
[type
]);
702 nr_swapper_spaces
[type
] = 0;
703 swapper_spaces
[type
] = NULL
;
706 static inline void swap_ra_clamp_pfn(struct vm_area_struct
*vma
,
710 unsigned long *start
,
713 *start
= max3(lpfn
, PFN_DOWN(vma
->vm_start
),
714 PFN_DOWN(faddr
& PMD_MASK
));
715 *end
= min3(rpfn
, PFN_DOWN(vma
->vm_end
),
716 PFN_DOWN((faddr
& PMD_MASK
) + PMD_SIZE
));
719 static void swap_ra_info(struct vm_fault
*vmf
,
720 struct vma_swap_readahead
*ra_info
)
722 struct vm_area_struct
*vma
= vmf
->vma
;
723 unsigned long ra_val
;
725 unsigned long faddr
, pfn
, fpfn
;
726 unsigned long start
, end
;
727 pte_t
*pte
, *orig_pte
;
728 unsigned int max_win
, hits
, prev_win
, win
, left
;
733 max_win
= 1 << min_t(unsigned int, READ_ONCE(page_cluster
),
734 SWAP_RA_ORDER_CEILING
);
740 faddr
= vmf
->address
;
741 orig_pte
= pte
= pte_offset_map(vmf
->pmd
, faddr
);
742 entry
= pte_to_swp_entry(*pte
);
743 if ((unlikely(non_swap_entry(entry
)))) {
748 fpfn
= PFN_DOWN(faddr
);
749 ra_val
= GET_SWAP_RA_VAL(vma
);
750 pfn
= PFN_DOWN(SWAP_RA_ADDR(ra_val
));
751 prev_win
= SWAP_RA_WIN(ra_val
);
752 hits
= SWAP_RA_HITS(ra_val
);
753 ra_info
->win
= win
= __swapin_nr_pages(pfn
, fpfn
, hits
,
755 atomic_long_set(&vma
->swap_readahead_info
,
756 SWAP_RA_VAL(faddr
, win
, 0));
763 /* Copy the PTEs because the page table may be unmapped */
765 swap_ra_clamp_pfn(vma
, faddr
, fpfn
, fpfn
+ win
, &start
, &end
);
766 else if (pfn
== fpfn
+ 1)
767 swap_ra_clamp_pfn(vma
, faddr
, fpfn
- win
+ 1, fpfn
+ 1,
770 left
= (win
- 1) / 2;
771 swap_ra_clamp_pfn(vma
, faddr
, fpfn
- left
, fpfn
+ win
- left
,
774 ra_info
->nr_pte
= end
- start
;
775 ra_info
->offset
= fpfn
- start
;
776 pte
-= ra_info
->offset
;
780 tpte
= ra_info
->ptes
;
781 for (pfn
= start
; pfn
!= end
; pfn
++)
788 * swap_vma_readahead - swap in pages in hope we need them soon
789 * @fentry: swap entry of this memory
790 * @gfp_mask: memory allocation flags
791 * @vmf: fault information
793 * Returns the struct page for entry and addr, after queueing swapin.
795 * Primitive swap readahead code. We simply read in a few pages whose
796 * virtual addresses are around the fault address in the same vma.
798 * Caller must hold read mmap_lock if vmf->vma is not NULL.
801 static struct page
*swap_vma_readahead(swp_entry_t fentry
, gfp_t gfp_mask
,
802 struct vm_fault
*vmf
)
804 struct blk_plug plug
;
805 struct vm_area_struct
*vma
= vmf
->vma
;
811 struct vma_swap_readahead ra_info
= {
815 swap_ra_info(vmf
, &ra_info
);
816 if (ra_info
.win
== 1)
819 blk_start_plug(&plug
);
820 for (i
= 0, pte
= ra_info
.ptes
; i
< ra_info
.nr_pte
;
823 if (pte_none(pentry
))
825 if (pte_present(pentry
))
827 entry
= pte_to_swp_entry(pentry
);
828 if (unlikely(non_swap_entry(entry
)))
830 page
= __read_swap_cache_async(entry
, gfp_mask
, vma
,
831 vmf
->address
, &page_allocated
);
834 if (page_allocated
) {
835 swap_readpage(page
, false);
836 if (i
!= ra_info
.offset
) {
837 SetPageReadahead(page
);
838 count_vm_event(SWAP_RA
);
843 blk_finish_plug(&plug
);
846 return read_swap_cache_async(fentry
, gfp_mask
, vma
, vmf
->address
,
851 * swapin_readahead - swap in pages in hope we need them soon
852 * @entry: swap entry of this memory
853 * @gfp_mask: memory allocation flags
854 * @vmf: fault information
856 * Returns the struct page for entry and addr, after queueing swapin.
858 * It's a main entry function for swap readahead. By the configuration,
859 * it will read ahead blocks by cluster-based(ie, physical disk based)
860 * or vma-based(ie, virtual address based on faulty address) readahead.
862 struct page
*swapin_readahead(swp_entry_t entry
, gfp_t gfp_mask
,
863 struct vm_fault
*vmf
)
865 return swap_use_vma_readahead() ?
866 swap_vma_readahead(entry
, gfp_mask
, vmf
) :
867 swap_cluster_readahead(entry
, gfp_mask
, vmf
);
871 static ssize_t
vma_ra_enabled_show(struct kobject
*kobj
,
872 struct kobj_attribute
*attr
, char *buf
)
874 return sysfs_emit(buf
, "%s\n",
875 enable_vma_readahead
? "true" : "false");
877 static ssize_t
vma_ra_enabled_store(struct kobject
*kobj
,
878 struct kobj_attribute
*attr
,
879 const char *buf
, size_t count
)
881 if (!strncmp(buf
, "true", 4) || !strncmp(buf
, "1", 1))
882 enable_vma_readahead
= true;
883 else if (!strncmp(buf
, "false", 5) || !strncmp(buf
, "0", 1))
884 enable_vma_readahead
= false;
890 static struct kobj_attribute vma_ra_enabled_attr
=
891 __ATTR(vma_ra_enabled
, 0644, vma_ra_enabled_show
,
892 vma_ra_enabled_store
);
894 static struct attribute
*swap_attrs
[] = {
895 &vma_ra_enabled_attr
.attr
,
899 static const struct attribute_group swap_attr_group
= {
903 static int __init
swap_init_sysfs(void)
906 struct kobject
*swap_kobj
;
908 swap_kobj
= kobject_create_and_add("swap", mm_kobj
);
910 pr_err("failed to create swap kobject\n");
913 err
= sysfs_create_group(swap_kobj
, &swap_attr_group
);
915 pr_err("failed to register swap group\n");
921 kobject_put(swap_kobj
);
924 subsys_initcall(swap_init_sysfs
);