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
);
118 xas_create_range(&xas
);
121 for (i
= 0; i
< nr
; i
++) {
122 VM_BUG_ON_PAGE(xas
.xa_index
!= idx
+ i
, page
);
123 old
= xas_load(&xas
);
124 if (xa_is_value(old
)) {
128 set_page_private(page
+ i
, entry
.val
+ i
);
129 xas_store(&xas
, page
);
132 address_space
->nrpages
+= nr
;
133 __mod_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, nr
);
134 __mod_lruvec_page_state(page
, NR_SWAPCACHE
, nr
);
135 ADD_CACHE_INFO(add_total
, nr
);
137 xas_unlock_irq(&xas
);
138 } while (xas_nomem(&xas
, gfp
));
140 if (!xas_error(&xas
))
143 ClearPageSwapCache(page
);
144 page_ref_sub(page
, nr
);
145 return xas_error(&xas
);
149 * This must be called only on pages that have
150 * been verified to be in the swap cache.
152 void __delete_from_swap_cache(struct page
*page
,
153 swp_entry_t entry
, void *shadow
)
155 struct address_space
*address_space
= swap_address_space(entry
);
156 int i
, nr
= thp_nr_pages(page
);
157 pgoff_t idx
= swp_offset(entry
);
158 XA_STATE(xas
, &address_space
->i_pages
, idx
);
160 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
161 VM_BUG_ON_PAGE(!PageSwapCache(page
), page
);
162 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
164 for (i
= 0; i
< nr
; i
++) {
165 void *entry
= xas_store(&xas
, shadow
);
166 VM_BUG_ON_PAGE(entry
!= page
, entry
);
167 set_page_private(page
+ i
, 0);
170 ClearPageSwapCache(page
);
171 address_space
->nrpages
-= nr
;
172 __mod_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, -nr
);
173 __mod_lruvec_page_state(page
, NR_SWAPCACHE
, -nr
);
174 ADD_CACHE_INFO(del_total
, nr
);
178 * add_to_swap - allocate swap space for a page
179 * @page: page we want to move to swap
181 * Allocate swap space for the page and add the page to the
182 * swap cache. Caller needs to hold the page lock.
184 int add_to_swap(struct page
*page
)
189 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
190 VM_BUG_ON_PAGE(!PageUptodate(page
), page
);
192 entry
= get_swap_page(page
);
197 * XArray node allocations from PF_MEMALLOC contexts could
198 * completely exhaust the page allocator. __GFP_NOMEMALLOC
199 * stops emergency reserves from being allocated.
201 * TODO: this could cause a theoretical memory reclaim
202 * deadlock in the swap out path.
205 * Add it to the swap cache.
207 err
= add_to_swap_cache(page
, entry
,
208 __GFP_HIGH
|__GFP_NOMEMALLOC
|__GFP_NOWARN
, NULL
);
211 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
212 * clear SWAP_HAS_CACHE flag.
216 * Normally the page will be dirtied in unmap because its pte should be
217 * dirty. A special case is MADV_FREE page. The page's pte could have
218 * dirty bit cleared but the page's SwapBacked bit is still set because
219 * clearing the dirty bit and SwapBacked bit has no lock protected. For
220 * such page, unmap will not set dirty bit for it, so page reclaim will
221 * not write the page out. This can cause data corruption when the page
222 * is swap in later. Always setting the dirty bit for the page solves
225 set_page_dirty(page
);
230 put_swap_page(page
, entry
);
235 * This must be called only on pages that have
236 * been verified to be in the swap cache and locked.
237 * It will never put the page into the free list,
238 * the caller has a reference on the page.
240 void delete_from_swap_cache(struct page
*page
)
242 swp_entry_t entry
= { .val
= page_private(page
) };
243 struct address_space
*address_space
= swap_address_space(entry
);
245 xa_lock_irq(&address_space
->i_pages
);
246 __delete_from_swap_cache(page
, entry
, NULL
);
247 xa_unlock_irq(&address_space
->i_pages
);
249 put_swap_page(page
, entry
);
250 page_ref_sub(page
, thp_nr_pages(page
));
253 void clear_shadow_from_swap_cache(int type
, unsigned long begin
,
256 unsigned long curr
= begin
;
260 swp_entry_t entry
= swp_entry(type
, curr
);
261 struct address_space
*address_space
= swap_address_space(entry
);
262 XA_STATE(xas
, &address_space
->i_pages
, curr
);
264 xa_lock_irq(&address_space
->i_pages
);
265 xas_for_each(&xas
, old
, end
) {
266 if (!xa_is_value(old
))
268 xas_store(&xas
, NULL
);
270 xa_unlock_irq(&address_space
->i_pages
);
272 /* search the next swapcache until we meet end */
273 curr
>>= SWAP_ADDRESS_SPACE_SHIFT
;
275 curr
<<= SWAP_ADDRESS_SPACE_SHIFT
;
282 * If we are the only user, then try to free up the swap cache.
284 * Its ok to check for PageSwapCache without the page lock
285 * here because we are going to recheck again inside
286 * try_to_free_swap() _with_ the lock.
289 void free_swap_cache(struct page
*page
)
291 if (PageSwapCache(page
) && !page_mapped(page
) && trylock_page(page
)) {
292 try_to_free_swap(page
);
298 * Perform a free_page(), also freeing any swap cache associated with
299 * this page if it is the last user of the page.
301 void free_page_and_swap_cache(struct page
*page
)
303 free_swap_cache(page
);
304 if (!is_huge_zero_page(page
))
309 * Passed an array of pages, drop them all from swapcache and then release
310 * them. They are removed from the LRU and freed if this is their last use.
312 void free_pages_and_swap_cache(struct page
**pages
, int nr
)
314 struct page
**pagep
= pages
;
318 for (i
= 0; i
< nr
; i
++)
319 free_swap_cache(pagep
[i
]);
320 release_pages(pagep
, nr
);
323 static inline bool swap_use_vma_readahead(void)
325 return READ_ONCE(enable_vma_readahead
) && !atomic_read(&nr_rotate_swap
);
329 * Lookup a swap entry in the swap cache. A found page will be returned
330 * unlocked and with its refcount incremented - we rely on the kernel
331 * lock getting page table operations atomic even if we drop the page
332 * lock before returning.
334 struct page
*lookup_swap_cache(swp_entry_t entry
, struct vm_area_struct
*vma
,
338 struct swap_info_struct
*si
;
340 si
= get_swap_device(entry
);
343 page
= find_get_page(swap_address_space(entry
), swp_offset(entry
));
346 INC_CACHE_INFO(find_total
);
348 bool vma_ra
= swap_use_vma_readahead();
351 INC_CACHE_INFO(find_success
);
353 * At the moment, we don't support PG_readahead for anon THP
354 * so let's bail out rather than confusing the readahead stat.
356 if (unlikely(PageTransCompound(page
)))
359 readahead
= TestClearPageReadahead(page
);
361 unsigned long ra_val
;
364 ra_val
= GET_SWAP_RA_VAL(vma
);
365 win
= SWAP_RA_WIN(ra_val
);
366 hits
= SWAP_RA_HITS(ra_val
);
368 hits
= min_t(int, hits
+ 1, SWAP_RA_HITS_MAX
);
369 atomic_long_set(&vma
->swap_readahead_info
,
370 SWAP_RA_VAL(addr
, win
, hits
));
374 count_vm_event(SWAP_RA_HIT
);
376 atomic_inc(&swapin_readahead_hits
);
384 * find_get_incore_page - Find and get a page from the page or swap caches.
385 * @mapping: The address_space to search.
386 * @index: The page cache index.
388 * This differs from find_get_page() in that it will also look for the
389 * page in the swap cache.
391 * Return: The found page or %NULL.
393 struct page
*find_get_incore_page(struct address_space
*mapping
, pgoff_t index
)
396 struct swap_info_struct
*si
;
397 struct page
*page
= pagecache_get_page(mapping
, index
,
398 FGP_ENTRY
| FGP_HEAD
, 0);
402 if (!xa_is_value(page
))
403 return find_subpage(page
, index
);
404 if (!shmem_mapping(mapping
))
407 swp
= radix_to_swp_entry(page
);
408 /* Prevent swapoff from happening to us */
409 si
= get_swap_device(swp
);
412 page
= find_get_page(swap_address_space(swp
), swp_offset(swp
));
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.
481 schedule_timeout_uninterruptible(1);
485 * The swap entry is ours to swap in. Prepare the new page.
488 __SetPageLocked(page
);
489 __SetPageSwapBacked(page
);
491 if (mem_cgroup_swapin_charge_page(page
, NULL
, gfp_mask
, entry
))
494 /* May fail (-ENOMEM) if XArray node allocation failed. */
495 if (add_to_swap_cache(page
, entry
, gfp_mask
& GFP_RECLAIM_MASK
, &shadow
))
498 mem_cgroup_swapin_uncharge_swap(entry
);
501 workingset_refault(page
, shadow
);
503 /* Caller will initiate read into locked page */
505 *new_page_allocated
= true;
509 put_swap_page(page
, entry
);
516 * Locate a page of swap in physical memory, reserving swap cache space
517 * and reading the disk if it is not already cached.
518 * A failure return means that either the page allocation failed or that
519 * the swap entry is no longer in use.
521 struct page
*read_swap_cache_async(swp_entry_t entry
, gfp_t gfp_mask
,
522 struct vm_area_struct
*vma
, unsigned long addr
, bool do_poll
)
524 bool page_was_allocated
;
525 struct page
*retpage
= __read_swap_cache_async(entry
, gfp_mask
,
526 vma
, addr
, &page_was_allocated
);
528 if (page_was_allocated
)
529 swap_readpage(retpage
, do_poll
);
534 static unsigned int __swapin_nr_pages(unsigned long prev_offset
,
535 unsigned long offset
,
540 unsigned int pages
, last_ra
;
543 * This heuristic has been found to work well on both sequential and
544 * random loads, swapping to hard disk or to SSD: please don't ask
545 * what the "+ 2" means, it just happens to work well, that's all.
550 * We can have no readahead hits to judge by: but must not get
551 * stuck here forever, so check for an adjacent offset instead
552 * (and don't even bother to check whether swap type is same).
554 if (offset
!= prev_offset
+ 1 && offset
!= prev_offset
- 1)
557 unsigned int roundup
= 4;
558 while (roundup
< pages
)
563 if (pages
> max_pages
)
566 /* Don't shrink readahead too fast */
567 last_ra
= prev_win
/ 2;
574 static unsigned long swapin_nr_pages(unsigned long offset
)
576 static unsigned long prev_offset
;
577 unsigned int hits
, pages
, max_pages
;
578 static atomic_t last_readahead_pages
;
580 max_pages
= 1 << READ_ONCE(page_cluster
);
584 hits
= atomic_xchg(&swapin_readahead_hits
, 0);
585 pages
= __swapin_nr_pages(READ_ONCE(prev_offset
), offset
, hits
,
587 atomic_read(&last_readahead_pages
));
589 WRITE_ONCE(prev_offset
, offset
);
590 atomic_set(&last_readahead_pages
, pages
);
596 * swap_cluster_readahead - swap in pages in hope we need them soon
597 * @entry: swap entry of this memory
598 * @gfp_mask: memory allocation flags
599 * @vmf: fault information
601 * Returns the struct page for entry and addr, after queueing swapin.
603 * Primitive swap readahead code. We simply read an aligned block of
604 * (1 << page_cluster) entries in the swap area. This method is chosen
605 * because it doesn't cost us any seek time. We also make sure to queue
606 * the 'original' request together with the readahead ones...
608 * This has been extended to use the NUMA policies from the mm triggering
611 * Caller must hold read mmap_lock if vmf->vma is not NULL.
613 struct page
*swap_cluster_readahead(swp_entry_t entry
, gfp_t gfp_mask
,
614 struct vm_fault
*vmf
)
617 unsigned long entry_offset
= swp_offset(entry
);
618 unsigned long offset
= entry_offset
;
619 unsigned long start_offset
, end_offset
;
621 struct swap_info_struct
*si
= swp_swap_info(entry
);
622 struct blk_plug plug
;
623 bool do_poll
= true, page_allocated
;
624 struct vm_area_struct
*vma
= vmf
->vma
;
625 unsigned long addr
= vmf
->address
;
627 mask
= swapin_nr_pages(offset
) - 1;
632 /* Read a page_cluster sized and aligned cluster around offset. */
633 start_offset
= offset
& ~mask
;
634 end_offset
= offset
| mask
;
635 if (!start_offset
) /* First page is swap header. */
637 if (end_offset
>= si
->max
)
638 end_offset
= si
->max
- 1;
640 blk_start_plug(&plug
);
641 for (offset
= start_offset
; offset
<= end_offset
; offset
++) {
642 /* Ok, do the async read-ahead now */
643 page
= __read_swap_cache_async(
644 swp_entry(swp_type(entry
), offset
),
645 gfp_mask
, vma
, addr
, &page_allocated
);
648 if (page_allocated
) {
649 swap_readpage(page
, false);
650 if (offset
!= entry_offset
) {
651 SetPageReadahead(page
);
652 count_vm_event(SWAP_RA
);
657 blk_finish_plug(&plug
);
659 lru_add_drain(); /* Push any new pages onto the LRU now */
661 return read_swap_cache_async(entry
, gfp_mask
, vma
, addr
, do_poll
);
664 int init_swap_address_space(unsigned int type
, unsigned long nr_pages
)
666 struct address_space
*spaces
, *space
;
669 nr
= DIV_ROUND_UP(nr_pages
, SWAP_ADDRESS_SPACE_PAGES
);
670 spaces
= kvcalloc(nr
, sizeof(struct address_space
), GFP_KERNEL
);
673 for (i
= 0; i
< nr
; i
++) {
675 xa_init_flags(&space
->i_pages
, XA_FLAGS_LOCK_IRQ
);
676 atomic_set(&space
->i_mmap_writable
, 0);
677 space
->a_ops
= &swap_aops
;
678 /* swap cache doesn't use writeback related tags */
679 mapping_set_no_writeback_tags(space
);
681 nr_swapper_spaces
[type
] = nr
;
682 swapper_spaces
[type
] = spaces
;
687 void exit_swap_address_space(unsigned int type
)
690 struct address_space
*spaces
= swapper_spaces
[type
];
692 for (i
= 0; i
< nr_swapper_spaces
[type
]; i
++)
693 VM_WARN_ON_ONCE(!mapping_empty(&spaces
[i
]));
695 nr_swapper_spaces
[type
] = 0;
696 swapper_spaces
[type
] = NULL
;
699 static inline void swap_ra_clamp_pfn(struct vm_area_struct
*vma
,
703 unsigned long *start
,
706 *start
= max3(lpfn
, PFN_DOWN(vma
->vm_start
),
707 PFN_DOWN(faddr
& PMD_MASK
));
708 *end
= min3(rpfn
, PFN_DOWN(vma
->vm_end
),
709 PFN_DOWN((faddr
& PMD_MASK
) + PMD_SIZE
));
712 static void swap_ra_info(struct vm_fault
*vmf
,
713 struct vma_swap_readahead
*ra_info
)
715 struct vm_area_struct
*vma
= vmf
->vma
;
716 unsigned long ra_val
;
717 unsigned long faddr
, pfn
, fpfn
;
718 unsigned long start
, end
;
719 pte_t
*pte
, *orig_pte
;
720 unsigned int max_win
, hits
, prev_win
, win
, left
;
725 max_win
= 1 << min_t(unsigned int, READ_ONCE(page_cluster
),
726 SWAP_RA_ORDER_CEILING
);
732 faddr
= vmf
->address
;
733 orig_pte
= pte
= pte_offset_map(vmf
->pmd
, faddr
);
735 fpfn
= PFN_DOWN(faddr
);
736 ra_val
= GET_SWAP_RA_VAL(vma
);
737 pfn
= PFN_DOWN(SWAP_RA_ADDR(ra_val
));
738 prev_win
= SWAP_RA_WIN(ra_val
);
739 hits
= SWAP_RA_HITS(ra_val
);
740 ra_info
->win
= win
= __swapin_nr_pages(pfn
, fpfn
, hits
,
742 atomic_long_set(&vma
->swap_readahead_info
,
743 SWAP_RA_VAL(faddr
, win
, 0));
750 /* Copy the PTEs because the page table may be unmapped */
752 swap_ra_clamp_pfn(vma
, faddr
, fpfn
, fpfn
+ win
, &start
, &end
);
753 else if (pfn
== fpfn
+ 1)
754 swap_ra_clamp_pfn(vma
, faddr
, fpfn
- win
+ 1, fpfn
+ 1,
757 left
= (win
- 1) / 2;
758 swap_ra_clamp_pfn(vma
, faddr
, fpfn
- left
, fpfn
+ win
- left
,
761 ra_info
->nr_pte
= end
- start
;
762 ra_info
->offset
= fpfn
- start
;
763 pte
-= ra_info
->offset
;
767 tpte
= ra_info
->ptes
;
768 for (pfn
= start
; pfn
!= end
; pfn
++)
775 * swap_vma_readahead - swap in pages in hope we need them soon
776 * @fentry: swap entry of this memory
777 * @gfp_mask: memory allocation flags
778 * @vmf: fault information
780 * Returns the struct page for entry and addr, after queueing swapin.
782 * Primitive swap readahead code. We simply read in a few pages whose
783 * virtual addresses are around the fault address in the same vma.
785 * Caller must hold read mmap_lock if vmf->vma is not NULL.
788 static struct page
*swap_vma_readahead(swp_entry_t fentry
, gfp_t gfp_mask
,
789 struct vm_fault
*vmf
)
791 struct blk_plug plug
;
792 struct vm_area_struct
*vma
= vmf
->vma
;
798 struct vma_swap_readahead ra_info
= {
802 swap_ra_info(vmf
, &ra_info
);
803 if (ra_info
.win
== 1)
806 blk_start_plug(&plug
);
807 for (i
= 0, pte
= ra_info
.ptes
; i
< ra_info
.nr_pte
;
810 if (pte_none(pentry
))
812 if (pte_present(pentry
))
814 entry
= pte_to_swp_entry(pentry
);
815 if (unlikely(non_swap_entry(entry
)))
817 page
= __read_swap_cache_async(entry
, gfp_mask
, vma
,
818 vmf
->address
, &page_allocated
);
821 if (page_allocated
) {
822 swap_readpage(page
, false);
823 if (i
!= ra_info
.offset
) {
824 SetPageReadahead(page
);
825 count_vm_event(SWAP_RA
);
830 blk_finish_plug(&plug
);
833 return read_swap_cache_async(fentry
, gfp_mask
, vma
, vmf
->address
,
838 * swapin_readahead - swap in pages in hope we need them soon
839 * @entry: swap entry of this memory
840 * @gfp_mask: memory allocation flags
841 * @vmf: fault information
843 * Returns the struct page for entry and addr, after queueing swapin.
845 * It's a main entry function for swap readahead. By the configuration,
846 * it will read ahead blocks by cluster-based(ie, physical disk based)
847 * or vma-based(ie, virtual address based on faulty address) readahead.
849 struct page
*swapin_readahead(swp_entry_t entry
, gfp_t gfp_mask
,
850 struct vm_fault
*vmf
)
852 return swap_use_vma_readahead() ?
853 swap_vma_readahead(entry
, gfp_mask
, vmf
) :
854 swap_cluster_readahead(entry
, gfp_mask
, vmf
);
858 static ssize_t
vma_ra_enabled_show(struct kobject
*kobj
,
859 struct kobj_attribute
*attr
, char *buf
)
861 return sysfs_emit(buf
, "%s\n",
862 enable_vma_readahead
? "true" : "false");
864 static ssize_t
vma_ra_enabled_store(struct kobject
*kobj
,
865 struct kobj_attribute
*attr
,
866 const char *buf
, size_t count
)
868 if (!strncmp(buf
, "true", 4) || !strncmp(buf
, "1", 1))
869 enable_vma_readahead
= true;
870 else if (!strncmp(buf
, "false", 5) || !strncmp(buf
, "0", 1))
871 enable_vma_readahead
= false;
877 static struct kobj_attribute vma_ra_enabled_attr
=
878 __ATTR(vma_ra_enabled
, 0644, vma_ra_enabled_show
,
879 vma_ra_enabled_store
);
881 static struct attribute
*swap_attrs
[] = {
882 &vma_ra_enabled_attr
.attr
,
886 static const struct attribute_group swap_attr_group
= {
890 static int __init
swap_init_sysfs(void)
893 struct kobject
*swap_kobj
;
895 swap_kobj
= kobject_create_and_add("swap", mm_kobj
);
897 pr_err("failed to create swap kobject\n");
900 err
= sysfs_create_group(swap_kobj
, &swap_attr_group
);
902 pr_err("failed to register swap group\n");
908 kobject_put(swap_kobj
);
911 subsys_initcall(swap_init_sysfs
);