defconfig: exynos9610: Re-add dropped Wi-Fi AP options lost
[GitHub/LineageOS/android_kernel_motorola_exynos9610.git] / mm / swap_state.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/mm/swap_state.c
4 *
5 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
6 * Swap reorganised 29.12.95, Stephen Tweedie
7 *
8 * Rewritten to use page cache, (C) 1998 Stephen Tweedie
9 */
10 #include <linux/mm.h>
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
25 #include <asm/pgtable.h>
26
27 /*
28 * swapper_space is a fiction, retained to simplify the path through
29 * vmscan's shrink_page_list.
30 */
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,
36 #endif
37 };
38
39 struct address_space *swapper_spaces[MAX_SWAPFILES];
40 static unsigned int nr_swapper_spaces[MAX_SWAPFILES];
41 bool swap_vma_readahead = true;
42
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)
47
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)
51
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))
56
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)
60
61 #define INC_CACHE_INFO(x) do { swap_cache_info.x++; } while (0)
62 #define ADD_CACHE_INFO(x, nr) do { swap_cache_info.x += (nr); } while (0)
63
64 static struct {
65 unsigned long add_total;
66 unsigned long del_total;
67 unsigned long find_success;
68 unsigned long find_total;
69 } swap_cache_info;
70
71 unsigned long total_swapcache_pages(void)
72 {
73 unsigned int i, j, nr;
74 unsigned long ret = 0;
75 struct address_space *spaces;
76
77 rcu_read_lock();
78 for (i = 0; i < MAX_SWAPFILES; i++) {
79 /*
80 * The corresponding entries in nr_swapper_spaces and
81 * swapper_spaces will be reused only after at least
82 * one grace period. So it is impossible for them
83 * belongs to different usage.
84 */
85 nr = nr_swapper_spaces[i];
86 spaces = rcu_dereference(swapper_spaces[i]);
87 if (!nr || !spaces)
88 continue;
89 for (j = 0; j < nr; j++)
90 ret += spaces[j].nrpages;
91 }
92 rcu_read_unlock();
93 return ret;
94 }
95
96 static atomic_t swapin_readahead_hits = ATOMIC_INIT(4);
97
98 void show_swap_cache_info(void)
99 {
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));
107 }
108
109 /*
110 * __add_to_swap_cache resembles add_to_page_cache_locked on swapper_space,
111 * but sets SwapCache flag and private instead of mapping and index.
112 */
113 int __add_to_swap_cache(struct page *page, swp_entry_t entry)
114 {
115 int error, i, nr = hpage_nr_pages(page);
116 struct address_space *address_space;
117 pgoff_t idx = swp_offset(entry);
118
119 VM_BUG_ON_PAGE(!PageLocked(page), page);
120 VM_BUG_ON_PAGE(PageSwapCache(page), page);
121 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
122
123 page_ref_add(page, nr);
124 SetPageSwapCache(page);
125
126 address_space = swap_address_space(entry);
127 spin_lock_irq(&address_space->tree_lock);
128 for (i = 0; i < nr; i++) {
129 set_page_private(page + i, entry.val + i);
130 error = radix_tree_insert(&address_space->page_tree,
131 idx + i, page + i);
132 if (unlikely(error))
133 break;
134 }
135 if (likely(!error)) {
136 address_space->nrpages += nr;
137 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
138 ADD_CACHE_INFO(add_total, nr);
139 } else {
140 /*
141 * Only the context which have set SWAP_HAS_CACHE flag
142 * would call add_to_swap_cache().
143 * So add_to_swap_cache() doesn't returns -EEXIST.
144 */
145 VM_BUG_ON(error == -EEXIST);
146 set_page_private(page + i, 0UL);
147 while (i--) {
148 radix_tree_delete(&address_space->page_tree, idx + i);
149 set_page_private(page + i, 0UL);
150 }
151 ClearPageSwapCache(page);
152 page_ref_sub(page, nr);
153 }
154 spin_unlock_irq(&address_space->tree_lock);
155
156 return error;
157 }
158
159
160 int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp_mask)
161 {
162 int error;
163
164 error = radix_tree_maybe_preload_order(gfp_mask, compound_order(page));
165 if (!error) {
166 error = __add_to_swap_cache(page, entry);
167 radix_tree_preload_end();
168 }
169 return error;
170 }
171
172 /*
173 * This must be called only on pages that have
174 * been verified to be in the swap cache.
175 */
176 void __delete_from_swap_cache(struct page *page)
177 {
178 struct address_space *address_space;
179 int i, nr = hpage_nr_pages(page);
180 swp_entry_t entry;
181 pgoff_t idx;
182
183 VM_BUG_ON_PAGE(!PageLocked(page), page);
184 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
185 VM_BUG_ON_PAGE(PageWriteback(page), page);
186
187 entry.val = page_private(page);
188 address_space = swap_address_space(entry);
189 idx = swp_offset(entry);
190 for (i = 0; i < nr; i++) {
191 radix_tree_delete(&address_space->page_tree, idx + i);
192 set_page_private(page + i, 0);
193 }
194 ClearPageSwapCache(page);
195 address_space->nrpages -= nr;
196 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
197 ADD_CACHE_INFO(del_total, nr);
198 }
199
200 /**
201 * add_to_swap - allocate swap space for a page
202 * @page: page we want to move to swap
203 *
204 * Allocate swap space for the page and add the page to the
205 * swap cache. Caller needs to hold the page lock.
206 */
207 int add_to_swap(struct page *page)
208 {
209 swp_entry_t entry;
210 int err;
211
212 VM_BUG_ON_PAGE(!PageLocked(page), page);
213 VM_BUG_ON_PAGE(!PageUptodate(page), page);
214
215 entry = get_swap_page(page);
216 if (!entry.val)
217 return 0;
218
219 if (mem_cgroup_try_charge_swap(page, entry))
220 goto fail;
221
222 /*
223 * Radix-tree node allocations from PF_MEMALLOC contexts could
224 * completely exhaust the page allocator. __GFP_NOMEMALLOC
225 * stops emergency reserves from being allocated.
226 *
227 * TODO: this could cause a theoretical memory reclaim
228 * deadlock in the swap out path.
229 */
230 /*
231 * Add it to the swap cache.
232 */
233 err = add_to_swap_cache(page, entry,
234 __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN);
235 /* -ENOMEM radix-tree allocation failure */
236 if (err)
237 /*
238 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
239 * clear SWAP_HAS_CACHE flag.
240 */
241 goto fail;
242 /*
243 * Normally the page will be dirtied in unmap because its pte should be
244 * dirty. A special case is MADV_FREE page. The page'e pte could have
245 * dirty bit cleared but the page's SwapBacked bit is still set because
246 * clearing the dirty bit and SwapBacked bit has no lock protected. For
247 * such page, unmap will not set dirty bit for it, so page reclaim will
248 * not write the page out. This can cause data corruption when the page
249 * is swap in later. Always setting the dirty bit for the page solves
250 * the problem.
251 */
252 set_page_dirty(page);
253
254 return 1;
255
256 fail:
257 put_swap_page(page, entry);
258 return 0;
259 }
260
261 /*
262 * This must be called only on pages that have
263 * been verified to be in the swap cache and locked.
264 * It will never put the page into the free list,
265 * the caller has a reference on the page.
266 */
267 void delete_from_swap_cache(struct page *page)
268 {
269 swp_entry_t entry;
270 struct address_space *address_space;
271
272 entry.val = page_private(page);
273
274 address_space = swap_address_space(entry);
275 spin_lock_irq(&address_space->tree_lock);
276 __delete_from_swap_cache(page);
277 spin_unlock_irq(&address_space->tree_lock);
278
279 put_swap_page(page, entry);
280 page_ref_sub(page, hpage_nr_pages(page));
281 }
282
283 /*
284 * If we are the only user, then try to free up the swap cache.
285 *
286 * Its ok to check for PageSwapCache without the page lock
287 * here because we are going to recheck again inside
288 * try_to_free_swap() _with_ the lock.
289 * - Marcelo
290 */
291 static inline void free_swap_cache(struct page *page)
292 {
293 if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) {
294 try_to_free_swap(page);
295 unlock_page(page);
296 }
297 }
298
299 /*
300 * Perform a free_page(), also freeing any swap cache associated with
301 * this page if it is the last user of the page.
302 */
303 void free_page_and_swap_cache(struct page *page)
304 {
305 free_swap_cache(page);
306 if (!is_huge_zero_page(page))
307 put_page(page);
308 }
309
310 /*
311 * Passed an array of pages, drop them all from swapcache and then release
312 * them. They are removed from the LRU and freed if this is their last use.
313 */
314 void free_pages_and_swap_cache(struct page **pages, int nr)
315 {
316 struct page **pagep = pages;
317 int i;
318
319 lru_add_drain();
320 for (i = 0; i < nr; i++)
321 free_swap_cache(pagep[i]);
322 release_pages(pagep, nr, false);
323 }
324
325 /*
326 * Lookup a swap entry in the swap cache. A found page will be returned
327 * unlocked and with its refcount incremented - we rely on the kernel
328 * lock getting page table operations atomic even if we drop the page
329 * lock before returning.
330 */
331 struct page *lookup_swap_cache(swp_entry_t entry, struct vm_area_struct *vma,
332 unsigned long addr)
333 {
334 struct page *page;
335 unsigned long ra_info;
336 int win, hits, readahead;
337
338 page = find_get_page(swap_address_space(entry), swp_offset(entry));
339
340 INC_CACHE_INFO(find_total);
341 if (page) {
342 INC_CACHE_INFO(find_success);
343 if (unlikely(PageTransCompound(page)))
344 return page;
345 readahead = TestClearPageReadahead(page);
346 if (vma) {
347 ra_info = GET_SWAP_RA_VAL(vma);
348 win = SWAP_RA_WIN(ra_info);
349 hits = SWAP_RA_HITS(ra_info);
350 if (readahead)
351 hits = min_t(int, hits + 1, SWAP_RA_HITS_MAX);
352 atomic_long_set(&vma->swap_readahead_info,
353 SWAP_RA_VAL(addr, win, hits));
354 }
355 if (readahead) {
356 count_vm_event(SWAP_RA_HIT);
357 if (!vma)
358 atomic_inc(&swapin_readahead_hits);
359 }
360 }
361 return page;
362 }
363
364 struct page *__read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
365 struct vm_area_struct *vma, unsigned long addr,
366 bool *new_page_allocated)
367 {
368 struct page *found_page, *new_page = NULL;
369 struct address_space *swapper_space = swap_address_space(entry);
370 int err;
371 *new_page_allocated = false;
372
373 do {
374 /*
375 * First check the swap cache. Since this is normally
376 * called after lookup_swap_cache() failed, re-calling
377 * that would confuse statistics.
378 */
379 found_page = find_get_page(swapper_space, swp_offset(entry));
380 if (found_page)
381 break;
382
383 /*
384 * Just skip read ahead for unused swap slot.
385 * During swap_off when swap_slot_cache is disabled,
386 * we have to handle the race between putting
387 * swap entry in swap cache and marking swap slot
388 * as SWAP_HAS_CACHE. That's done in later part of code or
389 * else swap_off will be aborted if we return NULL.
390 */
391 if (!__swp_swapcount(entry) && swap_slot_cache_enabled)
392 break;
393
394 /*
395 * Get a new page to read into from swap.
396 */
397 if (!new_page) {
398 new_page = alloc_page_vma(gfp_mask, vma, addr);
399 if (!new_page)
400 break; /* Out of memory */
401 }
402
403 /*
404 * call radix_tree_preload() while we can wait.
405 */
406 err = radix_tree_maybe_preload(gfp_mask & GFP_KERNEL);
407 if (err)
408 break;
409
410 /*
411 * Swap entry may have been freed since our caller observed it.
412 */
413 err = swapcache_prepare(entry);
414 if (err == -EEXIST) {
415 radix_tree_preload_end();
416 /*
417 * We might race against get_swap_page() and stumble
418 * across a SWAP_HAS_CACHE swap_map entry whose page
419 * has not been brought into the swapcache yet.
420 */
421 cond_resched();
422 continue;
423 }
424 if (err) { /* swp entry is obsolete ? */
425 radix_tree_preload_end();
426 break;
427 }
428
429 /* May fail (-ENOMEM) if radix-tree node allocation failed. */
430 __SetPageLocked(new_page);
431 __SetPageSwapBacked(new_page);
432 err = __add_to_swap_cache(new_page, entry);
433 if (likely(!err)) {
434 radix_tree_preload_end();
435 /*
436 * Initiate read into locked page and return.
437 */
438 SetPageWorkingset(new_page);
439 lru_cache_add_anon(new_page);
440 *new_page_allocated = true;
441 return new_page;
442 }
443 radix_tree_preload_end();
444 __ClearPageLocked(new_page);
445 /*
446 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
447 * clear SWAP_HAS_CACHE flag.
448 */
449 put_swap_page(new_page, entry);
450 } while (err != -ENOMEM);
451
452 if (new_page)
453 put_page(new_page);
454 return found_page;
455 }
456
457 /*
458 * Locate a page of swap in physical memory, reserving swap cache space
459 * and reading the disk if it is not already cached.
460 * A failure return means that either the page allocation failed or that
461 * the swap entry is no longer in use.
462 */
463 struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
464 struct vm_area_struct *vma, unsigned long addr, bool do_poll)
465 {
466 bool page_was_allocated;
467 struct page *retpage = __read_swap_cache_async(entry, gfp_mask,
468 vma, addr, &page_was_allocated);
469
470 if (page_was_allocated)
471 swap_readpage(retpage, do_poll);
472
473 return retpage;
474 }
475
476 static unsigned int __swapin_nr_pages(unsigned long prev_offset,
477 unsigned long offset,
478 int hits,
479 int max_pages,
480 int prev_win)
481 {
482 unsigned int pages, last_ra;
483
484 /*
485 * This heuristic has been found to work well on both sequential and
486 * random loads, swapping to hard disk or to SSD: please don't ask
487 * what the "+ 2" means, it just happens to work well, that's all.
488 */
489 pages = hits + 2;
490 if (pages == 2) {
491 /*
492 * We can have no readahead hits to judge by: but must not get
493 * stuck here forever, so check for an adjacent offset instead
494 * (and don't even bother to check whether swap type is same).
495 */
496 if (offset != prev_offset + 1 && offset != prev_offset - 1)
497 pages = 1;
498 } else {
499 unsigned int roundup = 4;
500 while (roundup < pages)
501 roundup <<= 1;
502 pages = roundup;
503 }
504
505 if (pages > max_pages)
506 pages = max_pages;
507
508 /* Don't shrink readahead too fast */
509 last_ra = prev_win / 2;
510 if (pages < last_ra)
511 pages = last_ra;
512
513 return pages;
514 }
515
516 static unsigned long swapin_nr_pages(unsigned long offset)
517 {
518 static unsigned long prev_offset;
519 unsigned int hits, pages, max_pages;
520 static atomic_t last_readahead_pages;
521
522 max_pages = 1 << READ_ONCE(page_cluster);
523 if (max_pages <= 1)
524 return 1;
525
526 hits = atomic_xchg(&swapin_readahead_hits, 0);
527 pages = __swapin_nr_pages(prev_offset, offset, hits, max_pages,
528 atomic_read(&last_readahead_pages));
529 if (!hits)
530 prev_offset = offset;
531 atomic_set(&last_readahead_pages, pages);
532
533 return pages;
534 }
535
536 /**
537 * swapin_readahead - swap in pages in hope we need them soon
538 * @entry: swap entry of this memory
539 * @gfp_mask: memory allocation flags
540 * @vma: user vma this address belongs to
541 * @addr: target address for mempolicy
542 *
543 * Returns the struct page for entry and addr, after queueing swapin.
544 *
545 * Primitive swap readahead code. We simply read an aligned block of
546 * (1 << page_cluster) entries in the swap area. This method is chosen
547 * because it doesn't cost us any seek time. We also make sure to queue
548 * the 'original' request together with the readahead ones...
549 *
550 * This has been extended to use the NUMA policies from the mm triggering
551 * the readahead.
552 *
553 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
554 */
555 struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask,
556 struct vm_area_struct *vma, unsigned long addr)
557 {
558 struct page *page;
559 unsigned long entry_offset = swp_offset(entry);
560 unsigned long offset = entry_offset;
561 unsigned long start_offset, end_offset;
562 unsigned long mask;
563 struct blk_plug plug;
564 bool do_poll = true, page_allocated;
565
566 mask = swapin_nr_pages(offset) - 1;
567 if (!mask)
568 goto skip;
569
570 do_poll = false;
571 /* Read a page_cluster sized and aligned cluster around offset. */
572 start_offset = offset & ~mask;
573 end_offset = offset | mask;
574 if (!start_offset) /* First page is swap header. */
575 start_offset++;
576
577 blk_start_plug(&plug);
578 for (offset = start_offset; offset <= end_offset ; offset++) {
579 /* Ok, do the async read-ahead now */
580 page = __read_swap_cache_async(
581 swp_entry(swp_type(entry), offset),
582 gfp_mask, vma, addr, &page_allocated);
583 if (!page)
584 continue;
585 if (page_allocated) {
586 swap_readpage(page, false);
587 if (offset != entry_offset &&
588 likely(!PageTransCompound(page))) {
589 SetPageReadahead(page);
590 count_vm_event(SWAP_RA);
591 }
592 }
593 put_page(page);
594 }
595 blk_finish_plug(&plug);
596
597 lru_add_drain(); /* Push any new pages onto the LRU now */
598 skip:
599 return read_swap_cache_async(entry, gfp_mask, vma, addr, do_poll);
600 }
601
602 int init_swap_address_space(unsigned int type, unsigned long nr_pages)
603 {
604 struct address_space *spaces, *space;
605 unsigned int i, nr;
606
607 nr = DIV_ROUND_UP(nr_pages, SWAP_ADDRESS_SPACE_PAGES);
608 spaces = kvzalloc(sizeof(struct address_space) * nr, GFP_KERNEL);
609 if (!spaces)
610 return -ENOMEM;
611 for (i = 0; i < nr; i++) {
612 space = spaces + i;
613 INIT_RADIX_TREE(&space->page_tree, GFP_ATOMIC|__GFP_NOWARN);
614 atomic_set(&space->i_mmap_writable, 0);
615 space->a_ops = &swap_aops;
616 /* swap cache doesn't use writeback related tags */
617 mapping_set_no_writeback_tags(space);
618 spin_lock_init(&space->tree_lock);
619 }
620 nr_swapper_spaces[type] = nr;
621 rcu_assign_pointer(swapper_spaces[type], spaces);
622
623 return 0;
624 }
625
626 void exit_swap_address_space(unsigned int type)
627 {
628 struct address_space *spaces;
629
630 spaces = swapper_spaces[type];
631 nr_swapper_spaces[type] = 0;
632 rcu_assign_pointer(swapper_spaces[type], NULL);
633 synchronize_rcu();
634 kvfree(spaces);
635 }
636
637 static inline void swap_ra_clamp_pfn(struct vm_area_struct *vma,
638 unsigned long faddr,
639 unsigned long lpfn,
640 unsigned long rpfn,
641 unsigned long *start,
642 unsigned long *end)
643 {
644 *start = max3(lpfn, PFN_DOWN(vma->vm_start),
645 PFN_DOWN(faddr & PMD_MASK));
646 *end = min3(rpfn, PFN_DOWN(vma->vm_end),
647 PFN_DOWN((faddr & PMD_MASK) + PMD_SIZE));
648 }
649
650 struct page *swap_readahead_detect(struct vm_fault *vmf,
651 struct vma_swap_readahead *swap_ra)
652 {
653 struct vm_area_struct *vma = vmf->vma;
654 unsigned long swap_ra_info;
655 struct page *page;
656 swp_entry_t entry;
657 unsigned long faddr, pfn, fpfn;
658 unsigned long start, end;
659 pte_t *pte;
660 unsigned int max_win, hits, prev_win, win, left;
661 #ifndef CONFIG_64BIT
662 pte_t *tpte;
663 #endif
664
665 max_win = 1 << min_t(unsigned int, READ_ONCE(page_cluster),
666 SWAP_RA_ORDER_CEILING);
667 if (max_win == 1) {
668 swap_ra->win = 1;
669 return NULL;
670 }
671
672 faddr = vmf->address;
673 entry = pte_to_swp_entry(vmf->orig_pte);
674 if ((unlikely(non_swap_entry(entry))))
675 return NULL;
676 page = lookup_swap_cache(entry, vma, faddr);
677 if (page)
678 return page;
679
680 fpfn = PFN_DOWN(faddr);
681 swap_ra_info = GET_SWAP_RA_VAL(vma);
682 pfn = PFN_DOWN(SWAP_RA_ADDR(swap_ra_info));
683 prev_win = SWAP_RA_WIN(swap_ra_info);
684 hits = SWAP_RA_HITS(swap_ra_info);
685 swap_ra->win = win = __swapin_nr_pages(pfn, fpfn, hits,
686 max_win, prev_win);
687 atomic_long_set(&vma->swap_readahead_info,
688 SWAP_RA_VAL(faddr, win, 0));
689
690 if (win == 1)
691 return NULL;
692
693 /* Copy the PTEs because the page table may be unmapped */
694 if (fpfn == pfn + 1)
695 swap_ra_clamp_pfn(vma, faddr, fpfn, fpfn + win, &start, &end);
696 else if (pfn == fpfn + 1)
697 swap_ra_clamp_pfn(vma, faddr, fpfn - win + 1, fpfn + 1,
698 &start, &end);
699 else {
700 left = (win - 1) / 2;
701 swap_ra_clamp_pfn(vma, faddr, fpfn - left, fpfn + win - left,
702 &start, &end);
703 }
704 swap_ra->nr_pte = end - start;
705 swap_ra->offset = fpfn - start;
706 pte = vmf->pte - swap_ra->offset;
707 #ifdef CONFIG_64BIT
708 swap_ra->ptes = pte;
709 #else
710 tpte = swap_ra->ptes;
711 for (pfn = start; pfn != end; pfn++)
712 *tpte++ = *pte++;
713 #endif
714
715 return NULL;
716 }
717
718 struct page *do_swap_page_readahead(swp_entry_t fentry, gfp_t gfp_mask,
719 struct vm_fault *vmf,
720 struct vma_swap_readahead *swap_ra)
721 {
722 struct blk_plug plug;
723 struct vm_area_struct *vma = vmf->vma;
724 struct page *page;
725 pte_t *pte, pentry;
726 swp_entry_t entry;
727 unsigned int i;
728 bool page_allocated;
729
730 if (swap_ra->win == 1)
731 goto skip;
732
733 blk_start_plug(&plug);
734 for (i = 0, pte = swap_ra->ptes; i < swap_ra->nr_pte;
735 i++, pte++) {
736 pentry = *pte;
737 if (pte_none(pentry))
738 continue;
739 if (pte_present(pentry))
740 continue;
741 entry = pte_to_swp_entry(pentry);
742 if (unlikely(non_swap_entry(entry)))
743 continue;
744 page = __read_swap_cache_async(entry, gfp_mask, vma,
745 vmf->address, &page_allocated);
746 if (!page)
747 continue;
748 if (page_allocated) {
749 swap_readpage(page, false);
750 if (i != swap_ra->offset &&
751 likely(!PageTransCompound(page))) {
752 SetPageReadahead(page);
753 count_vm_event(SWAP_RA);
754 }
755 }
756 put_page(page);
757 }
758 blk_finish_plug(&plug);
759 lru_add_drain();
760 skip:
761 return read_swap_cache_async(fentry, gfp_mask, vma, vmf->address,
762 swap_ra->win == 1);
763 }
764
765 #ifdef CONFIG_SYSFS
766 static ssize_t vma_ra_enabled_show(struct kobject *kobj,
767 struct kobj_attribute *attr, char *buf)
768 {
769 return sprintf(buf, "%s\n", swap_vma_readahead ? "true" : "false");
770 }
771 static ssize_t vma_ra_enabled_store(struct kobject *kobj,
772 struct kobj_attribute *attr,
773 const char *buf, size_t count)
774 {
775 if (!strncmp(buf, "true", 4) || !strncmp(buf, "1", 1))
776 swap_vma_readahead = true;
777 else if (!strncmp(buf, "false", 5) || !strncmp(buf, "0", 1))
778 swap_vma_readahead = false;
779 else
780 return -EINVAL;
781
782 return count;
783 }
784 static struct kobj_attribute vma_ra_enabled_attr =
785 __ATTR(vma_ra_enabled, 0644, vma_ra_enabled_show,
786 vma_ra_enabled_store);
787
788 static struct attribute *swap_attrs[] = {
789 &vma_ra_enabled_attr.attr,
790 NULL,
791 };
792
793 static struct attribute_group swap_attr_group = {
794 .attrs = swap_attrs,
795 };
796
797 static int __init swap_init_sysfs(void)
798 {
799 int err;
800 struct kobject *swap_kobj;
801
802 swap_kobj = kobject_create_and_add("swap", mm_kobj);
803 if (!swap_kobj) {
804 pr_err("failed to create swap kobject\n");
805 return -ENOMEM;
806 }
807 err = sysfs_create_group(swap_kobj, &swap_attr_group);
808 if (err) {
809 pr_err("failed to register swap group\n");
810 goto delete_obj;
811 }
812 return 0;
813
814 delete_obj:
815 kobject_put(swap_kobj);
816 return err;
817 }
818 subsys_initcall(swap_init_sysfs);
819 #endif