]> git.ipfire.org Git - people/arne_f/kernel.git/blame - mm/swap_state.c
net: qmi_wwan: fix divide by 0 on bad descriptors
[people/arne_f/kernel.git] / mm / swap_state.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
1da177e4
LT
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 */
1da177e4 10#include <linux/mm.h>
5a0e3ad6 11#include <linux/gfp.h>
1da177e4
LT
12#include <linux/kernel_stat.h>
13#include <linux/swap.h>
46017e95 14#include <linux/swapops.h>
1da177e4
LT
15#include <linux/init.h>
16#include <linux/pagemap.h>
1da177e4 17#include <linux/backing-dev.h>
3fb5c298 18#include <linux/blkdev.h>
c484d410 19#include <linux/pagevec.h>
b20a3503 20#include <linux/migrate.h>
4b3ef9da 21#include <linux/vmalloc.h>
67afa38e 22#include <linux/swap_slots.h>
38d8b4e6 23#include <linux/huge_mm.h>
1da177e4
LT
24
25#include <asm/pgtable.h>
26
27/*
28 * swapper_space is a fiction, retained to simplify the path through
7eaceacc 29 * vmscan's shrink_page_list.
1da177e4 30 */
f5e54d6e 31static const struct address_space_operations swap_aops = {
1da177e4 32 .writepage = swap_writepage,
62c230bc 33 .set_page_dirty = swap_set_page_dirty,
1c93923c 34#ifdef CONFIG_MIGRATION
e965f963 35 .migratepage = migrate_page,
1c93923c 36#endif
1da177e4
LT
37};
38
4b3ef9da
HY
39struct address_space *swapper_spaces[MAX_SWAPFILES];
40static unsigned int nr_swapper_spaces[MAX_SWAPFILES];
ec560175
HY
41bool swap_vma_readahead = true;
42
ec560175
HY
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)
1da177e4
LT
60
61#define INC_CACHE_INFO(x) do { swap_cache_info.x++; } while (0)
38d8b4e6 62#define ADD_CACHE_INFO(x, nr) do { swap_cache_info.x += (nr); } while (0)
1da177e4
LT
63
64static struct {
65 unsigned long add_total;
66 unsigned long del_total;
67 unsigned long find_success;
68 unsigned long find_total;
1da177e4
LT
69} swap_cache_info;
70
33806f06
SL
71unsigned long total_swapcache_pages(void)
72{
4b3ef9da 73 unsigned int i, j, nr;
33806f06 74 unsigned long ret = 0;
4b3ef9da 75 struct address_space *spaces;
33806f06 76
4b3ef9da
HY
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();
33806f06
SL
93 return ret;
94}
95
579f8290
SL
96static atomic_t swapin_readahead_hits = ATOMIC_INIT(4);
97
1da177e4
LT
98void show_swap_cache_info(void)
99{
33806f06 100 printk("%lu pages in swap cache\n", total_swapcache_pages());
2c97b7fc 101 printk("Swap cache stats: add %lu, delete %lu, find %lu/%lu\n",
1da177e4 102 swap_cache_info.add_total, swap_cache_info.del_total,
bb63be0a 103 swap_cache_info.find_success, swap_cache_info.find_total);
ec8acf20
SL
104 printk("Free swap = %ldkB\n",
105 get_nr_swap_pages() << (PAGE_SHIFT - 10));
1da177e4
LT
106 printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10));
107}
108
109/*
31a56396 110 * __add_to_swap_cache resembles add_to_page_cache_locked on swapper_space,
1da177e4
LT
111 * but sets SwapCache flag and private instead of mapping and index.
112 */
2f772e6c 113int __add_to_swap_cache(struct page *page, swp_entry_t entry)
1da177e4 114{
38d8b4e6 115 int error, i, nr = hpage_nr_pages(page);
33806f06 116 struct address_space *address_space;
38d8b4e6 117 pgoff_t idx = swp_offset(entry);
1da177e4 118
309381fe
SL
119 VM_BUG_ON_PAGE(!PageLocked(page), page);
120 VM_BUG_ON_PAGE(PageSwapCache(page), page);
121 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
51726b12 122
38d8b4e6 123 page_ref_add(page, nr);
31a56396 124 SetPageSwapCache(page);
31a56396 125
33806f06
SL
126 address_space = swap_address_space(entry);
127 spin_lock_irq(&address_space->tree_lock);
38d8b4e6
HY
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;
31a56396 134 }
38d8b4e6
HY
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 {
2ca4532a
DN
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);
38d8b4e6
HY
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 }
31a56396 151 ClearPageSwapCache(page);
38d8b4e6 152 page_ref_sub(page, nr);
31a56396 153 }
38d8b4e6 154 spin_unlock_irq(&address_space->tree_lock);
31a56396
DN
155
156 return error;
157}
158
159
160int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp_mask)
161{
162 int error;
163
38d8b4e6 164 error = radix_tree_maybe_preload_order(gfp_mask, compound_order(page));
35c754d7 165 if (!error) {
31a56396 166 error = __add_to_swap_cache(page, entry);
1da177e4 167 radix_tree_preload_end();
fa1de900 168 }
1da177e4
LT
169 return error;
170}
171
1da177e4
LT
172/*
173 * This must be called only on pages that have
174 * been verified to be in the swap cache.
175 */
176void __delete_from_swap_cache(struct page *page)
177{
33806f06 178 struct address_space *address_space;
38d8b4e6
HY
179 int i, nr = hpage_nr_pages(page);
180 swp_entry_t entry;
181 pgoff_t idx;
33806f06 182
309381fe
SL
183 VM_BUG_ON_PAGE(!PageLocked(page), page);
184 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
185 VM_BUG_ON_PAGE(PageWriteback(page), page);
1da177e4 186
33806f06
SL
187 entry.val = page_private(page);
188 address_space = swap_address_space(entry);
38d8b4e6
HY
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 }
1da177e4 194 ClearPageSwapCache(page);
38d8b4e6
HY
195 address_space->nrpages -= nr;
196 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
197 ADD_CACHE_INFO(del_total, nr);
1da177e4
LT
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 */
0f074658 207int add_to_swap(struct page *page)
1da177e4
LT
208{
209 swp_entry_t entry;
1da177e4
LT
210 int err;
211
309381fe
SL
212 VM_BUG_ON_PAGE(!PageLocked(page), page);
213 VM_BUG_ON_PAGE(!PageUptodate(page), page);
1da177e4 214
38d8b4e6 215 entry = get_swap_page(page);
2ca4532a 216 if (!entry.val)
0f074658
MK
217 return 0;
218
38d8b4e6 219 if (mem_cgroup_try_charge_swap(page, entry))
0f074658 220 goto fail;
3f04f62f 221
2ca4532a
DN
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 /*
854e9ed0 231 * Add it to the swap cache.
2ca4532a
DN
232 */
233 err = add_to_swap_cache(page, entry,
234 __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN);
38d8b4e6
HY
235 /* -ENOMEM radix-tree allocation failure */
236 if (err)
bd53b714 237 /*
2ca4532a
DN
238 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
239 * clear SWAP_HAS_CACHE flag.
1da177e4 240 */
0f074658 241 goto fail;
9625456c
SL
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);
38d8b4e6
HY
253
254 return 1;
255
38d8b4e6 256fail:
0f074658 257 put_swap_page(page, entry);
38d8b4e6 258 return 0;
1da177e4
LT
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 */
267void delete_from_swap_cache(struct page *page)
268{
269 swp_entry_t entry;
33806f06 270 struct address_space *address_space;
1da177e4 271
4c21e2f2 272 entry.val = page_private(page);
1da177e4 273
33806f06
SL
274 address_space = swap_address_space(entry);
275 spin_lock_irq(&address_space->tree_lock);
1da177e4 276 __delete_from_swap_cache(page);
33806f06 277 spin_unlock_irq(&address_space->tree_lock);
1da177e4 278
75f6d6d2 279 put_swap_page(page, entry);
38d8b4e6 280 page_ref_sub(page, hpage_nr_pages(page));
1da177e4
LT
281}
282
1da177e4
LT
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
a2c43eed
HD
287 * here because we are going to recheck again inside
288 * try_to_free_swap() _with_ the lock.
1da177e4
LT
289 * - Marcelo
290 */
291static inline void free_swap_cache(struct page *page)
292{
a2c43eed
HD
293 if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) {
294 try_to_free_swap(page);
1da177e4
LT
295 unlock_page(page);
296 }
297}
298
299/*
300 * Perform a free_page(), also freeing any swap cache associated with
b8072f09 301 * this page if it is the last user of the page.
1da177e4
LT
302 */
303void free_page_and_swap_cache(struct page *page)
304{
305 free_swap_cache(page);
6fcb52a5 306 if (!is_huge_zero_page(page))
770a5370 307 put_page(page);
1da177e4
LT
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 */
314void free_pages_and_swap_cache(struct page **pages, int nr)
315{
1da177e4 316 struct page **pagep = pages;
aabfb572 317 int i;
1da177e4
LT
318
319 lru_add_drain();
aabfb572
MH
320 for (i = 0; i < nr; i++)
321 free_swap_cache(pagep[i]);
322 release_pages(pagep, nr, false);
1da177e4
LT
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 */
ec560175
HY
331struct page *lookup_swap_cache(swp_entry_t entry, struct vm_area_struct *vma,
332 unsigned long addr)
1da177e4
LT
333{
334 struct page *page;
ec560175
HY
335 unsigned long ra_info;
336 int win, hits, readahead;
1da177e4 337
f6ab1f7f 338 page = find_get_page(swap_address_space(entry), swp_offset(entry));
1da177e4 339
ec560175
HY
340 INC_CACHE_INFO(find_total);
341 if (page) {
1da177e4 342 INC_CACHE_INFO(find_success);
ec560175
HY
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) {
cbc65df2 356 count_vm_event(SWAP_RA_HIT);
ec560175
HY
357 if (!vma)
358 atomic_inc(&swapin_readahead_hits);
cbc65df2 359 }
579f8290 360 }
1da177e4
LT
361 return page;
362}
363
5b999aad
DS
364struct 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)
1da177e4
LT
367{
368 struct page *found_page, *new_page = NULL;
5b999aad 369 struct address_space *swapper_space = swap_address_space(entry);
1da177e4 370 int err;
5b999aad 371 *new_page_allocated = false;
1da177e4
LT
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 */
f6ab1f7f 379 found_page = find_get_page(swapper_space, swp_offset(entry));
1da177e4
LT
380 if (found_page)
381 break;
382
ba81f838
HY
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;
e8c26ab6 393
1da177e4
LT
394 /*
395 * Get a new page to read into from swap.
396 */
397 if (!new_page) {
02098fea 398 new_page = alloc_page_vma(gfp_mask, vma, addr);
1da177e4
LT
399 if (!new_page)
400 break; /* Out of memory */
401 }
402
31a56396
DN
403 /*
404 * call radix_tree_preload() while we can wait.
405 */
5e4c0d97 406 err = radix_tree_maybe_preload(gfp_mask & GFP_KERNEL);
31a56396
DN
407 if (err)
408 break;
409
f000944d
HD
410 /*
411 * Swap entry may have been freed since our caller observed it.
412 */
355cfa73 413 err = swapcache_prepare(entry);
cbab0e4e 414 if (err == -EEXIST) {
31a56396 415 radix_tree_preload_end();
cbab0e4e
RA
416 /*
417 * We might race against get_swap_page() and stumble
418 * across a SWAP_HAS_CACHE swap_map entry whose page
9c1cc2e4 419 * has not been brought into the swapcache yet.
cbab0e4e
RA
420 */
421 cond_resched();
355cfa73 422 continue;
31a56396
DN
423 }
424 if (err) { /* swp entry is obsolete ? */
425 radix_tree_preload_end();
f000944d 426 break;
31a56396 427 }
f000944d 428
2ca4532a 429 /* May fail (-ENOMEM) if radix-tree node allocation failed. */
48c935ad 430 __SetPageLocked(new_page);
fa9949da 431 __SetPageSwapBacked(new_page);
31a56396 432 err = __add_to_swap_cache(new_page, entry);
529ae9aa 433 if (likely(!err)) {
31a56396 434 radix_tree_preload_end();
1da177e4
LT
435 /*
436 * Initiate read into locked page and return.
437 */
c5fdae46 438 lru_cache_add_anon(new_page);
5b999aad 439 *new_page_allocated = true;
1da177e4
LT
440 return new_page;
441 }
31a56396 442 radix_tree_preload_end();
48c935ad 443 __ClearPageLocked(new_page);
2ca4532a
DN
444 /*
445 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
446 * clear SWAP_HAS_CACHE flag.
447 */
75f6d6d2 448 put_swap_page(new_page, entry);
f000944d 449 } while (err != -ENOMEM);
1da177e4
LT
450
451 if (new_page)
09cbfeaf 452 put_page(new_page);
1da177e4
LT
453 return found_page;
454}
46017e95 455
5b999aad
DS
456/*
457 * Locate a page of swap in physical memory, reserving swap cache space
458 * and reading the disk if it is not already cached.
459 * A failure return means that either the page allocation failed or that
460 * the swap entry is no longer in use.
461 */
462struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
23955622 463 struct vm_area_struct *vma, unsigned long addr, bool do_poll)
5b999aad
DS
464{
465 bool page_was_allocated;
466 struct page *retpage = __read_swap_cache_async(entry, gfp_mask,
467 vma, addr, &page_was_allocated);
468
469 if (page_was_allocated)
23955622 470 swap_readpage(retpage, do_poll);
5b999aad
DS
471
472 return retpage;
473}
474
ec560175
HY
475static unsigned int __swapin_nr_pages(unsigned long prev_offset,
476 unsigned long offset,
477 int hits,
478 int max_pages,
479 int prev_win)
579f8290 480{
ec560175 481 unsigned int pages, last_ra;
579f8290
SL
482
483 /*
484 * This heuristic has been found to work well on both sequential and
485 * random loads, swapping to hard disk or to SSD: please don't ask
486 * what the "+ 2" means, it just happens to work well, that's all.
487 */
ec560175 488 pages = hits + 2;
579f8290
SL
489 if (pages == 2) {
490 /*
491 * We can have no readahead hits to judge by: but must not get
492 * stuck here forever, so check for an adjacent offset instead
493 * (and don't even bother to check whether swap type is same).
494 */
495 if (offset != prev_offset + 1 && offset != prev_offset - 1)
496 pages = 1;
579f8290
SL
497 } else {
498 unsigned int roundup = 4;
499 while (roundup < pages)
500 roundup <<= 1;
501 pages = roundup;
502 }
503
504 if (pages > max_pages)
505 pages = max_pages;
506
507 /* Don't shrink readahead too fast */
ec560175 508 last_ra = prev_win / 2;
579f8290
SL
509 if (pages < last_ra)
510 pages = last_ra;
ec560175
HY
511
512 return pages;
513}
514
515static unsigned long swapin_nr_pages(unsigned long offset)
516{
517 static unsigned long prev_offset;
518 unsigned int hits, pages, max_pages;
519 static atomic_t last_readahead_pages;
520
521 max_pages = 1 << READ_ONCE(page_cluster);
522 if (max_pages <= 1)
523 return 1;
524
525 hits = atomic_xchg(&swapin_readahead_hits, 0);
526 pages = __swapin_nr_pages(prev_offset, offset, hits, max_pages,
527 atomic_read(&last_readahead_pages));
528 if (!hits)
529 prev_offset = offset;
579f8290
SL
530 atomic_set(&last_readahead_pages, pages);
531
532 return pages;
533}
534
46017e95
HD
535/**
536 * swapin_readahead - swap in pages in hope we need them soon
537 * @entry: swap entry of this memory
7682486b 538 * @gfp_mask: memory allocation flags
46017e95
HD
539 * @vma: user vma this address belongs to
540 * @addr: target address for mempolicy
541 *
542 * Returns the struct page for entry and addr, after queueing swapin.
543 *
544 * Primitive swap readahead code. We simply read an aligned block of
545 * (1 << page_cluster) entries in the swap area. This method is chosen
546 * because it doesn't cost us any seek time. We also make sure to queue
547 * the 'original' request together with the readahead ones...
548 *
549 * This has been extended to use the NUMA policies from the mm triggering
550 * the readahead.
551 *
552 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
553 */
02098fea 554struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask,
46017e95
HD
555 struct vm_area_struct *vma, unsigned long addr)
556{
46017e95 557 struct page *page;
579f8290
SL
558 unsigned long entry_offset = swp_offset(entry);
559 unsigned long offset = entry_offset;
67f96aa2 560 unsigned long start_offset, end_offset;
579f8290 561 unsigned long mask;
3fb5c298 562 struct blk_plug plug;
c4fa6309 563 bool do_poll = true, page_allocated;
46017e95 564
579f8290
SL
565 mask = swapin_nr_pages(offset) - 1;
566 if (!mask)
567 goto skip;
568
23955622 569 do_poll = false;
67f96aa2
RR
570 /* Read a page_cluster sized and aligned cluster around offset. */
571 start_offset = offset & ~mask;
572 end_offset = offset | mask;
573 if (!start_offset) /* First page is swap header. */
574 start_offset++;
575
3fb5c298 576 blk_start_plug(&plug);
67f96aa2 577 for (offset = start_offset; offset <= end_offset ; offset++) {
46017e95 578 /* Ok, do the async read-ahead now */
c4fa6309
HY
579 page = __read_swap_cache_async(
580 swp_entry(swp_type(entry), offset),
581 gfp_mask, vma, addr, &page_allocated);
46017e95 582 if (!page)
67f96aa2 583 continue;
c4fa6309
HY
584 if (page_allocated) {
585 swap_readpage(page, false);
586 if (offset != entry_offset &&
587 likely(!PageTransCompound(page))) {
588 SetPageReadahead(page);
589 count_vm_event(SWAP_RA);
590 }
cbc65df2 591 }
09cbfeaf 592 put_page(page);
46017e95 593 }
3fb5c298
CE
594 blk_finish_plug(&plug);
595
46017e95 596 lru_add_drain(); /* Push any new pages onto the LRU now */
579f8290 597skip:
23955622 598 return read_swap_cache_async(entry, gfp_mask, vma, addr, do_poll);
46017e95 599}
4b3ef9da
HY
600
601int init_swap_address_space(unsigned int type, unsigned long nr_pages)
602{
603 struct address_space *spaces, *space;
604 unsigned int i, nr;
605
606 nr = DIV_ROUND_UP(nr_pages, SWAP_ADDRESS_SPACE_PAGES);
54f180d3 607 spaces = kvzalloc(sizeof(struct address_space) * nr, GFP_KERNEL);
4b3ef9da
HY
608 if (!spaces)
609 return -ENOMEM;
610 for (i = 0; i < nr; i++) {
611 space = spaces + i;
612 INIT_RADIX_TREE(&space->page_tree, GFP_ATOMIC|__GFP_NOWARN);
613 atomic_set(&space->i_mmap_writable, 0);
614 space->a_ops = &swap_aops;
615 /* swap cache doesn't use writeback related tags */
616 mapping_set_no_writeback_tags(space);
617 spin_lock_init(&space->tree_lock);
618 }
619 nr_swapper_spaces[type] = nr;
620 rcu_assign_pointer(swapper_spaces[type], spaces);
621
622 return 0;
623}
624
625void exit_swap_address_space(unsigned int type)
626{
627 struct address_space *spaces;
628
629 spaces = swapper_spaces[type];
630 nr_swapper_spaces[type] = 0;
631 rcu_assign_pointer(swapper_spaces[type], NULL);
632 synchronize_rcu();
633 kvfree(spaces);
634}
ec560175
HY
635
636static inline void swap_ra_clamp_pfn(struct vm_area_struct *vma,
637 unsigned long faddr,
638 unsigned long lpfn,
639 unsigned long rpfn,
640 unsigned long *start,
641 unsigned long *end)
642{
643 *start = max3(lpfn, PFN_DOWN(vma->vm_start),
644 PFN_DOWN(faddr & PMD_MASK));
645 *end = min3(rpfn, PFN_DOWN(vma->vm_end),
646 PFN_DOWN((faddr & PMD_MASK) + PMD_SIZE));
647}
648
649struct page *swap_readahead_detect(struct vm_fault *vmf,
650 struct vma_swap_readahead *swap_ra)
651{
652 struct vm_area_struct *vma = vmf->vma;
653 unsigned long swap_ra_info;
654 struct page *page;
655 swp_entry_t entry;
656 unsigned long faddr, pfn, fpfn;
657 unsigned long start, end;
658 pte_t *pte;
659 unsigned int max_win, hits, prev_win, win, left;
660#ifndef CONFIG_64BIT
661 pte_t *tpte;
662#endif
663
61b63972
HY
664 max_win = 1 << min_t(unsigned int, READ_ONCE(page_cluster),
665 SWAP_RA_ORDER_CEILING);
666 if (max_win == 1) {
667 swap_ra->win = 1;
668 return NULL;
669 }
670
ec560175
HY
671 faddr = vmf->address;
672 entry = pte_to_swp_entry(vmf->orig_pte);
673 if ((unlikely(non_swap_entry(entry))))
674 return NULL;
675 page = lookup_swap_cache(entry, vma, faddr);
676 if (page)
677 return page;
678
ec560175
HY
679 fpfn = PFN_DOWN(faddr);
680 swap_ra_info = GET_SWAP_RA_VAL(vma);
681 pfn = PFN_DOWN(SWAP_RA_ADDR(swap_ra_info));
682 prev_win = SWAP_RA_WIN(swap_ra_info);
683 hits = SWAP_RA_HITS(swap_ra_info);
684 swap_ra->win = win = __swapin_nr_pages(pfn, fpfn, hits,
685 max_win, prev_win);
686 atomic_long_set(&vma->swap_readahead_info,
687 SWAP_RA_VAL(faddr, win, 0));
688
689 if (win == 1)
690 return NULL;
691
692 /* Copy the PTEs because the page table may be unmapped */
693 if (fpfn == pfn + 1)
694 swap_ra_clamp_pfn(vma, faddr, fpfn, fpfn + win, &start, &end);
695 else if (pfn == fpfn + 1)
696 swap_ra_clamp_pfn(vma, faddr, fpfn - win + 1, fpfn + 1,
697 &start, &end);
698 else {
699 left = (win - 1) / 2;
700 swap_ra_clamp_pfn(vma, faddr, fpfn - left, fpfn + win - left,
701 &start, &end);
702 }
703 swap_ra->nr_pte = end - start;
704 swap_ra->offset = fpfn - start;
705 pte = vmf->pte - swap_ra->offset;
706#ifdef CONFIG_64BIT
707 swap_ra->ptes = pte;
708#else
709 tpte = swap_ra->ptes;
710 for (pfn = start; pfn != end; pfn++)
711 *tpte++ = *pte++;
712#endif
713
714 return NULL;
715}
716
717struct page *do_swap_page_readahead(swp_entry_t fentry, gfp_t gfp_mask,
718 struct vm_fault *vmf,
719 struct vma_swap_readahead *swap_ra)
720{
721 struct blk_plug plug;
722 struct vm_area_struct *vma = vmf->vma;
723 struct page *page;
724 pte_t *pte, pentry;
725 swp_entry_t entry;
726 unsigned int i;
727 bool page_allocated;
728
729 if (swap_ra->win == 1)
730 goto skip;
731
732 blk_start_plug(&plug);
733 for (i = 0, pte = swap_ra->ptes; i < swap_ra->nr_pte;
734 i++, pte++) {
735 pentry = *pte;
736 if (pte_none(pentry))
737 continue;
738 if (pte_present(pentry))
739 continue;
740 entry = pte_to_swp_entry(pentry);
741 if (unlikely(non_swap_entry(entry)))
742 continue;
743 page = __read_swap_cache_async(entry, gfp_mask, vma,
744 vmf->address, &page_allocated);
745 if (!page)
746 continue;
747 if (page_allocated) {
748 swap_readpage(page, false);
749 if (i != swap_ra->offset &&
750 likely(!PageTransCompound(page))) {
751 SetPageReadahead(page);
752 count_vm_event(SWAP_RA);
753 }
754 }
755 put_page(page);
756 }
757 blk_finish_plug(&plug);
758 lru_add_drain();
759skip:
760 return read_swap_cache_async(fentry, gfp_mask, vma, vmf->address,
761 swap_ra->win == 1);
762}
d9bfcfdc
HY
763
764#ifdef CONFIG_SYSFS
765static ssize_t vma_ra_enabled_show(struct kobject *kobj,
766 struct kobj_attribute *attr, char *buf)
767{
768 return sprintf(buf, "%s\n", swap_vma_readahead ? "true" : "false");
769}
770static ssize_t vma_ra_enabled_store(struct kobject *kobj,
771 struct kobj_attribute *attr,
772 const char *buf, size_t count)
773{
774 if (!strncmp(buf, "true", 4) || !strncmp(buf, "1", 1))
775 swap_vma_readahead = true;
776 else if (!strncmp(buf, "false", 5) || !strncmp(buf, "0", 1))
777 swap_vma_readahead = false;
778 else
779 return -EINVAL;
780
781 return count;
782}
783static struct kobj_attribute vma_ra_enabled_attr =
784 __ATTR(vma_ra_enabled, 0644, vma_ra_enabled_show,
785 vma_ra_enabled_store);
786
d9bfcfdc
HY
787static struct attribute *swap_attrs[] = {
788 &vma_ra_enabled_attr.attr,
d9bfcfdc
HY
789 NULL,
790};
791
792static struct attribute_group swap_attr_group = {
793 .attrs = swap_attrs,
794};
795
796static int __init swap_init_sysfs(void)
797{
798 int err;
799 struct kobject *swap_kobj;
800
801 swap_kobj = kobject_create_and_add("swap", mm_kobj);
802 if (!swap_kobj) {
803 pr_err("failed to create swap kobject\n");
804 return -ENOMEM;
805 }
806 err = sysfs_create_group(swap_kobj, &swap_attr_group);
807 if (err) {
808 pr_err("failed to register swap group\n");
809 goto delete_obj;
810 }
811 return 0;
812
813delete_obj:
814 kobject_put(swap_kobj);
815 return err;
816}
817subsys_initcall(swap_init_sysfs);
818#endif