]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - mm/swap_state.c
dax: Convert page fault handlers to XArray
[mirror_ubuntu-jammy-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
783cb68e
CD
39struct address_space *swapper_spaces[MAX_SWAPFILES] __read_mostly;
40static unsigned int nr_swapper_spaces[MAX_SWAPFILES] __read_mostly;
f5c754d6 41static bool enable_vma_readahead __read_mostly = true;
ec560175 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/*
8d93b41c 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 */
8d93b41c 113int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp)
1da177e4 114{
8d93b41c 115 struct address_space *address_space = swap_address_space(entry);
38d8b4e6 116 pgoff_t idx = swp_offset(entry);
8d93b41c
MW
117 XA_STATE_ORDER(xas, &address_space->i_pages, idx, compound_order(page));
118 unsigned long i, nr = 1UL << compound_order(page);
1da177e4 119
309381fe
SL
120 VM_BUG_ON_PAGE(!PageLocked(page), page);
121 VM_BUG_ON_PAGE(PageSwapCache(page), page);
122 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
51726b12 123
38d8b4e6 124 page_ref_add(page, nr);
31a56396 125 SetPageSwapCache(page);
31a56396 126
8d93b41c
MW
127 do {
128 xas_lock_irq(&xas);
129 xas_create_range(&xas);
130 if (xas_error(&xas))
131 goto unlock;
132 for (i = 0; i < nr; i++) {
133 VM_BUG_ON_PAGE(xas.xa_index != idx + i, page);
134 set_page_private(page + i, entry.val + i);
135 xas_store(&xas, page + i);
136 xas_next(&xas);
137 }
38d8b4e6
HY
138 address_space->nrpages += nr;
139 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
140 ADD_CACHE_INFO(add_total, nr);
8d93b41c
MW
141unlock:
142 xas_unlock_irq(&xas);
143 } while (xas_nomem(&xas, gfp));
31a56396 144
8d93b41c
MW
145 if (!xas_error(&xas))
146 return 0;
31a56396 147
8d93b41c
MW
148 ClearPageSwapCache(page);
149 page_ref_sub(page, nr);
150 return xas_error(&xas);
1da177e4
LT
151}
152
1da177e4
LT
153/*
154 * This must be called only on pages that have
155 * been verified to be in the swap cache.
156 */
4e17ec25 157void __delete_from_swap_cache(struct page *page, swp_entry_t entry)
1da177e4 158{
4e17ec25 159 struct address_space *address_space = swap_address_space(entry);
38d8b4e6 160 int i, nr = hpage_nr_pages(page);
4e17ec25
MW
161 pgoff_t idx = swp_offset(entry);
162 XA_STATE(xas, &address_space->i_pages, idx);
33806f06 163
309381fe
SL
164 VM_BUG_ON_PAGE(!PageLocked(page), page);
165 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
166 VM_BUG_ON_PAGE(PageWriteback(page), page);
1da177e4 167
38d8b4e6 168 for (i = 0; i < nr; i++) {
4e17ec25
MW
169 void *entry = xas_store(&xas, NULL);
170 VM_BUG_ON_PAGE(entry != page + i, entry);
38d8b4e6 171 set_page_private(page + i, 0);
4e17ec25 172 xas_next(&xas);
38d8b4e6 173 }
1da177e4 174 ClearPageSwapCache(page);
38d8b4e6
HY
175 address_space->nrpages -= nr;
176 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
177 ADD_CACHE_INFO(del_total, nr);
1da177e4
LT
178}
179
180/**
181 * add_to_swap - allocate swap space for a page
182 * @page: page we want to move to swap
183 *
184 * Allocate swap space for the page and add the page to the
185 * swap cache. Caller needs to hold the page lock.
186 */
0f074658 187int add_to_swap(struct page *page)
1da177e4
LT
188{
189 swp_entry_t entry;
1da177e4
LT
190 int err;
191
309381fe
SL
192 VM_BUG_ON_PAGE(!PageLocked(page), page);
193 VM_BUG_ON_PAGE(!PageUptodate(page), page);
1da177e4 194
38d8b4e6 195 entry = get_swap_page(page);
2ca4532a 196 if (!entry.val)
0f074658
MK
197 return 0;
198
2ca4532a 199 /*
8d93b41c 200 * XArray node allocations from PF_MEMALLOC contexts could
2ca4532a
DN
201 * completely exhaust the page allocator. __GFP_NOMEMALLOC
202 * stops emergency reserves from being allocated.
203 *
204 * TODO: this could cause a theoretical memory reclaim
205 * deadlock in the swap out path.
206 */
207 /*
854e9ed0 208 * Add it to the swap cache.
2ca4532a
DN
209 */
210 err = add_to_swap_cache(page, entry,
211 __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN);
38d8b4e6 212 if (err)
bd53b714 213 /*
2ca4532a
DN
214 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
215 * clear SWAP_HAS_CACHE flag.
1da177e4 216 */
0f074658 217 goto fail;
9625456c
SL
218 /*
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'e 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
226 * the problem.
227 */
228 set_page_dirty(page);
38d8b4e6
HY
229
230 return 1;
231
38d8b4e6 232fail:
0f074658 233 put_swap_page(page, entry);
38d8b4e6 234 return 0;
1da177e4
LT
235}
236
237/*
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.
242 */
243void delete_from_swap_cache(struct page *page)
244{
4e17ec25
MW
245 swp_entry_t entry = { .val = page_private(page) };
246 struct address_space *address_space = swap_address_space(entry);
1da177e4 247
b93b0163 248 xa_lock_irq(&address_space->i_pages);
4e17ec25 249 __delete_from_swap_cache(page, entry);
b93b0163 250 xa_unlock_irq(&address_space->i_pages);
1da177e4 251
75f6d6d2 252 put_swap_page(page, entry);
38d8b4e6 253 page_ref_sub(page, hpage_nr_pages(page));
1da177e4
LT
254}
255
1da177e4
LT
256/*
257 * If we are the only user, then try to free up the swap cache.
258 *
259 * Its ok to check for PageSwapCache without the page lock
a2c43eed
HD
260 * here because we are going to recheck again inside
261 * try_to_free_swap() _with_ the lock.
1da177e4
LT
262 * - Marcelo
263 */
264static inline void free_swap_cache(struct page *page)
265{
a2c43eed
HD
266 if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) {
267 try_to_free_swap(page);
1da177e4
LT
268 unlock_page(page);
269 }
270}
271
272/*
273 * Perform a free_page(), also freeing any swap cache associated with
b8072f09 274 * this page if it is the last user of the page.
1da177e4
LT
275 */
276void free_page_and_swap_cache(struct page *page)
277{
278 free_swap_cache(page);
6fcb52a5 279 if (!is_huge_zero_page(page))
770a5370 280 put_page(page);
1da177e4
LT
281}
282
283/*
284 * Passed an array of pages, drop them all from swapcache and then release
285 * them. They are removed from the LRU and freed if this is their last use.
286 */
287void free_pages_and_swap_cache(struct page **pages, int nr)
288{
1da177e4 289 struct page **pagep = pages;
aabfb572 290 int i;
1da177e4
LT
291
292 lru_add_drain();
aabfb572
MH
293 for (i = 0; i < nr; i++)
294 free_swap_cache(pagep[i]);
c6f92f9f 295 release_pages(pagep, nr);
1da177e4
LT
296}
297
e9e9b7ec
MK
298static inline bool swap_use_vma_readahead(void)
299{
300 return READ_ONCE(enable_vma_readahead) && !atomic_read(&nr_rotate_swap);
301}
302
1da177e4
LT
303/*
304 * Lookup a swap entry in the swap cache. A found page will be returned
305 * unlocked and with its refcount incremented - we rely on the kernel
306 * lock getting page table operations atomic even if we drop the page
307 * lock before returning.
308 */
ec560175
HY
309struct page *lookup_swap_cache(swp_entry_t entry, struct vm_area_struct *vma,
310 unsigned long addr)
1da177e4
LT
311{
312 struct page *page;
313
f6ab1f7f 314 page = find_get_page(swap_address_space(entry), swp_offset(entry));
1da177e4 315
ec560175
HY
316 INC_CACHE_INFO(find_total);
317 if (page) {
eaf649eb
MK
318 bool vma_ra = swap_use_vma_readahead();
319 bool readahead;
320
1da177e4 321 INC_CACHE_INFO(find_success);
eaf649eb
MK
322 /*
323 * At the moment, we don't support PG_readahead for anon THP
324 * so let's bail out rather than confusing the readahead stat.
325 */
ec560175
HY
326 if (unlikely(PageTransCompound(page)))
327 return page;
eaf649eb 328
ec560175 329 readahead = TestClearPageReadahead(page);
eaf649eb
MK
330 if (vma && vma_ra) {
331 unsigned long ra_val;
332 int win, hits;
333
334 ra_val = GET_SWAP_RA_VAL(vma);
335 win = SWAP_RA_WIN(ra_val);
336 hits = SWAP_RA_HITS(ra_val);
ec560175
HY
337 if (readahead)
338 hits = min_t(int, hits + 1, SWAP_RA_HITS_MAX);
339 atomic_long_set(&vma->swap_readahead_info,
340 SWAP_RA_VAL(addr, win, hits));
341 }
eaf649eb 342
ec560175 343 if (readahead) {
cbc65df2 344 count_vm_event(SWAP_RA_HIT);
eaf649eb 345 if (!vma || !vma_ra)
ec560175 346 atomic_inc(&swapin_readahead_hits);
cbc65df2 347 }
579f8290 348 }
eaf649eb 349
1da177e4
LT
350 return page;
351}
352
5b999aad
DS
353struct page *__read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
354 struct vm_area_struct *vma, unsigned long addr,
355 bool *new_page_allocated)
1da177e4
LT
356{
357 struct page *found_page, *new_page = NULL;
5b999aad 358 struct address_space *swapper_space = swap_address_space(entry);
1da177e4 359 int err;
5b999aad 360 *new_page_allocated = false;
1da177e4
LT
361
362 do {
363 /*
364 * First check the swap cache. Since this is normally
365 * called after lookup_swap_cache() failed, re-calling
366 * that would confuse statistics.
367 */
f6ab1f7f 368 found_page = find_get_page(swapper_space, swp_offset(entry));
1da177e4
LT
369 if (found_page)
370 break;
371
ba81f838
HY
372 /*
373 * Just skip read ahead for unused swap slot.
374 * During swap_off when swap_slot_cache is disabled,
375 * we have to handle the race between putting
376 * swap entry in swap cache and marking swap slot
377 * as SWAP_HAS_CACHE. That's done in later part of code or
378 * else swap_off will be aborted if we return NULL.
379 */
380 if (!__swp_swapcount(entry) && swap_slot_cache_enabled)
381 break;
e8c26ab6 382
1da177e4
LT
383 /*
384 * Get a new page to read into from swap.
385 */
386 if (!new_page) {
02098fea 387 new_page = alloc_page_vma(gfp_mask, vma, addr);
1da177e4
LT
388 if (!new_page)
389 break; /* Out of memory */
390 }
391
f000944d
HD
392 /*
393 * Swap entry may have been freed since our caller observed it.
394 */
355cfa73 395 err = swapcache_prepare(entry);
cbab0e4e 396 if (err == -EEXIST) {
cbab0e4e
RA
397 /*
398 * We might race against get_swap_page() and stumble
399 * across a SWAP_HAS_CACHE swap_map entry whose page
9c1cc2e4 400 * has not been brought into the swapcache yet.
cbab0e4e
RA
401 */
402 cond_resched();
355cfa73 403 continue;
8d93b41c 404 } else if (err) /* swp entry is obsolete ? */
f000944d
HD
405 break;
406
8d93b41c 407 /* May fail (-ENOMEM) if XArray node allocation failed. */
48c935ad 408 __SetPageLocked(new_page);
fa9949da 409 __SetPageSwapBacked(new_page);
8d93b41c 410 err = add_to_swap_cache(new_page, entry, gfp_mask & GFP_KERNEL);
529ae9aa 411 if (likely(!err)) {
8d93b41c 412 /* Initiate read into locked page */
c5fdae46 413 lru_cache_add_anon(new_page);
5b999aad 414 *new_page_allocated = true;
1da177e4
LT
415 return new_page;
416 }
48c935ad 417 __ClearPageLocked(new_page);
2ca4532a
DN
418 /*
419 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
420 * clear SWAP_HAS_CACHE flag.
421 */
75f6d6d2 422 put_swap_page(new_page, entry);
f000944d 423 } while (err != -ENOMEM);
1da177e4
LT
424
425 if (new_page)
09cbfeaf 426 put_page(new_page);
1da177e4
LT
427 return found_page;
428}
46017e95 429
5b999aad
DS
430/*
431 * Locate a page of swap in physical memory, reserving swap cache space
432 * and reading the disk if it is not already cached.
433 * A failure return means that either the page allocation failed or that
434 * the swap entry is no longer in use.
435 */
436struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
23955622 437 struct vm_area_struct *vma, unsigned long addr, bool do_poll)
5b999aad
DS
438{
439 bool page_was_allocated;
440 struct page *retpage = __read_swap_cache_async(entry, gfp_mask,
441 vma, addr, &page_was_allocated);
442
443 if (page_was_allocated)
23955622 444 swap_readpage(retpage, do_poll);
5b999aad
DS
445
446 return retpage;
447}
448
ec560175
HY
449static unsigned int __swapin_nr_pages(unsigned long prev_offset,
450 unsigned long offset,
451 int hits,
452 int max_pages,
453 int prev_win)
579f8290 454{
ec560175 455 unsigned int pages, last_ra;
579f8290
SL
456
457 /*
458 * This heuristic has been found to work well on both sequential and
459 * random loads, swapping to hard disk or to SSD: please don't ask
460 * what the "+ 2" means, it just happens to work well, that's all.
461 */
ec560175 462 pages = hits + 2;
579f8290
SL
463 if (pages == 2) {
464 /*
465 * We can have no readahead hits to judge by: but must not get
466 * stuck here forever, so check for an adjacent offset instead
467 * (and don't even bother to check whether swap type is same).
468 */
469 if (offset != prev_offset + 1 && offset != prev_offset - 1)
470 pages = 1;
579f8290
SL
471 } else {
472 unsigned int roundup = 4;
473 while (roundup < pages)
474 roundup <<= 1;
475 pages = roundup;
476 }
477
478 if (pages > max_pages)
479 pages = max_pages;
480
481 /* Don't shrink readahead too fast */
ec560175 482 last_ra = prev_win / 2;
579f8290
SL
483 if (pages < last_ra)
484 pages = last_ra;
ec560175
HY
485
486 return pages;
487}
488
489static unsigned long swapin_nr_pages(unsigned long offset)
490{
491 static unsigned long prev_offset;
492 unsigned int hits, pages, max_pages;
493 static atomic_t last_readahead_pages;
494
495 max_pages = 1 << READ_ONCE(page_cluster);
496 if (max_pages <= 1)
497 return 1;
498
499 hits = atomic_xchg(&swapin_readahead_hits, 0);
500 pages = __swapin_nr_pages(prev_offset, offset, hits, max_pages,
501 atomic_read(&last_readahead_pages));
502 if (!hits)
503 prev_offset = offset;
579f8290
SL
504 atomic_set(&last_readahead_pages, pages);
505
506 return pages;
507}
508
46017e95 509/**
e9e9b7ec 510 * swap_cluster_readahead - swap in pages in hope we need them soon
46017e95 511 * @entry: swap entry of this memory
7682486b 512 * @gfp_mask: memory allocation flags
e9e9b7ec 513 * @vmf: fault information
46017e95
HD
514 *
515 * Returns the struct page for entry and addr, after queueing swapin.
516 *
517 * Primitive swap readahead code. We simply read an aligned block of
518 * (1 << page_cluster) entries in the swap area. This method is chosen
519 * because it doesn't cost us any seek time. We also make sure to queue
520 * the 'original' request together with the readahead ones...
521 *
522 * This has been extended to use the NUMA policies from the mm triggering
523 * the readahead.
524 *
e9e9b7ec 525 * Caller must hold down_read on the vma->vm_mm if vmf->vma is not NULL.
46017e95 526 */
e9e9b7ec
MK
527struct page *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask,
528 struct vm_fault *vmf)
46017e95 529{
46017e95 530 struct page *page;
579f8290
SL
531 unsigned long entry_offset = swp_offset(entry);
532 unsigned long offset = entry_offset;
67f96aa2 533 unsigned long start_offset, end_offset;
579f8290 534 unsigned long mask;
e9a6effa 535 struct swap_info_struct *si = swp_swap_info(entry);
3fb5c298 536 struct blk_plug plug;
c4fa6309 537 bool do_poll = true, page_allocated;
e9e9b7ec
MK
538 struct vm_area_struct *vma = vmf->vma;
539 unsigned long addr = vmf->address;
46017e95 540
579f8290
SL
541 mask = swapin_nr_pages(offset) - 1;
542 if (!mask)
543 goto skip;
544
23955622 545 do_poll = false;
67f96aa2
RR
546 /* Read a page_cluster sized and aligned cluster around offset. */
547 start_offset = offset & ~mask;
548 end_offset = offset | mask;
549 if (!start_offset) /* First page is swap header. */
550 start_offset++;
e9a6effa
HY
551 if (end_offset >= si->max)
552 end_offset = si->max - 1;
67f96aa2 553
3fb5c298 554 blk_start_plug(&plug);
67f96aa2 555 for (offset = start_offset; offset <= end_offset ; offset++) {
46017e95 556 /* Ok, do the async read-ahead now */
c4fa6309
HY
557 page = __read_swap_cache_async(
558 swp_entry(swp_type(entry), offset),
559 gfp_mask, vma, addr, &page_allocated);
46017e95 560 if (!page)
67f96aa2 561 continue;
c4fa6309
HY
562 if (page_allocated) {
563 swap_readpage(page, false);
eaf649eb 564 if (offset != entry_offset) {
c4fa6309
HY
565 SetPageReadahead(page);
566 count_vm_event(SWAP_RA);
567 }
cbc65df2 568 }
09cbfeaf 569 put_page(page);
46017e95 570 }
3fb5c298
CE
571 blk_finish_plug(&plug);
572
46017e95 573 lru_add_drain(); /* Push any new pages onto the LRU now */
579f8290 574skip:
23955622 575 return read_swap_cache_async(entry, gfp_mask, vma, addr, do_poll);
46017e95 576}
4b3ef9da
HY
577
578int init_swap_address_space(unsigned int type, unsigned long nr_pages)
579{
580 struct address_space *spaces, *space;
581 unsigned int i, nr;
582
583 nr = DIV_ROUND_UP(nr_pages, SWAP_ADDRESS_SPACE_PAGES);
778e1cdd 584 spaces = kvcalloc(nr, sizeof(struct address_space), GFP_KERNEL);
4b3ef9da
HY
585 if (!spaces)
586 return -ENOMEM;
587 for (i = 0; i < nr; i++) {
588 space = spaces + i;
b93b0163 589 INIT_RADIX_TREE(&space->i_pages, GFP_ATOMIC|__GFP_NOWARN);
4b3ef9da
HY
590 atomic_set(&space->i_mmap_writable, 0);
591 space->a_ops = &swap_aops;
592 /* swap cache doesn't use writeback related tags */
593 mapping_set_no_writeback_tags(space);
4b3ef9da
HY
594 }
595 nr_swapper_spaces[type] = nr;
596 rcu_assign_pointer(swapper_spaces[type], spaces);
597
598 return 0;
599}
600
601void exit_swap_address_space(unsigned int type)
602{
603 struct address_space *spaces;
604
605 spaces = swapper_spaces[type];
606 nr_swapper_spaces[type] = 0;
607 rcu_assign_pointer(swapper_spaces[type], NULL);
608 synchronize_rcu();
609 kvfree(spaces);
610}
ec560175
HY
611
612static inline void swap_ra_clamp_pfn(struct vm_area_struct *vma,
613 unsigned long faddr,
614 unsigned long lpfn,
615 unsigned long rpfn,
616 unsigned long *start,
617 unsigned long *end)
618{
619 *start = max3(lpfn, PFN_DOWN(vma->vm_start),
620 PFN_DOWN(faddr & PMD_MASK));
621 *end = min3(rpfn, PFN_DOWN(vma->vm_end),
622 PFN_DOWN((faddr & PMD_MASK) + PMD_SIZE));
623}
624
eaf649eb
MK
625static void swap_ra_info(struct vm_fault *vmf,
626 struct vma_swap_readahead *ra_info)
ec560175
HY
627{
628 struct vm_area_struct *vma = vmf->vma;
eaf649eb 629 unsigned long ra_val;
ec560175
HY
630 swp_entry_t entry;
631 unsigned long faddr, pfn, fpfn;
632 unsigned long start, end;
eaf649eb 633 pte_t *pte, *orig_pte;
ec560175
HY
634 unsigned int max_win, hits, prev_win, win, left;
635#ifndef CONFIG_64BIT
636 pte_t *tpte;
637#endif
638
61b63972
HY
639 max_win = 1 << min_t(unsigned int, READ_ONCE(page_cluster),
640 SWAP_RA_ORDER_CEILING);
641 if (max_win == 1) {
eaf649eb
MK
642 ra_info->win = 1;
643 return;
61b63972
HY
644 }
645
ec560175 646 faddr = vmf->address;
eaf649eb
MK
647 orig_pte = pte = pte_offset_map(vmf->pmd, faddr);
648 entry = pte_to_swp_entry(*pte);
649 if ((unlikely(non_swap_entry(entry)))) {
650 pte_unmap(orig_pte);
651 return;
652 }
ec560175 653
ec560175 654 fpfn = PFN_DOWN(faddr);
eaf649eb
MK
655 ra_val = GET_SWAP_RA_VAL(vma);
656 pfn = PFN_DOWN(SWAP_RA_ADDR(ra_val));
657 prev_win = SWAP_RA_WIN(ra_val);
658 hits = SWAP_RA_HITS(ra_val);
659 ra_info->win = win = __swapin_nr_pages(pfn, fpfn, hits,
ec560175
HY
660 max_win, prev_win);
661 atomic_long_set(&vma->swap_readahead_info,
662 SWAP_RA_VAL(faddr, win, 0));
663
eaf649eb
MK
664 if (win == 1) {
665 pte_unmap(orig_pte);
666 return;
667 }
ec560175
HY
668
669 /* Copy the PTEs because the page table may be unmapped */
670 if (fpfn == pfn + 1)
671 swap_ra_clamp_pfn(vma, faddr, fpfn, fpfn + win, &start, &end);
672 else if (pfn == fpfn + 1)
673 swap_ra_clamp_pfn(vma, faddr, fpfn - win + 1, fpfn + 1,
674 &start, &end);
675 else {
676 left = (win - 1) / 2;
677 swap_ra_clamp_pfn(vma, faddr, fpfn - left, fpfn + win - left,
678 &start, &end);
679 }
eaf649eb
MK
680 ra_info->nr_pte = end - start;
681 ra_info->offset = fpfn - start;
682 pte -= ra_info->offset;
ec560175 683#ifdef CONFIG_64BIT
eaf649eb 684 ra_info->ptes = pte;
ec560175 685#else
eaf649eb 686 tpte = ra_info->ptes;
ec560175
HY
687 for (pfn = start; pfn != end; pfn++)
688 *tpte++ = *pte++;
689#endif
eaf649eb 690 pte_unmap(orig_pte);
ec560175
HY
691}
692
f5c754d6
CIK
693static struct page *swap_vma_readahead(swp_entry_t fentry, gfp_t gfp_mask,
694 struct vm_fault *vmf)
ec560175
HY
695{
696 struct blk_plug plug;
697 struct vm_area_struct *vma = vmf->vma;
698 struct page *page;
699 pte_t *pte, pentry;
700 swp_entry_t entry;
701 unsigned int i;
702 bool page_allocated;
eaf649eb 703 struct vma_swap_readahead ra_info = {0,};
ec560175 704
eaf649eb
MK
705 swap_ra_info(vmf, &ra_info);
706 if (ra_info.win == 1)
ec560175
HY
707 goto skip;
708
709 blk_start_plug(&plug);
eaf649eb 710 for (i = 0, pte = ra_info.ptes; i < ra_info.nr_pte;
ec560175
HY
711 i++, pte++) {
712 pentry = *pte;
713 if (pte_none(pentry))
714 continue;
715 if (pte_present(pentry))
716 continue;
717 entry = pte_to_swp_entry(pentry);
718 if (unlikely(non_swap_entry(entry)))
719 continue;
720 page = __read_swap_cache_async(entry, gfp_mask, vma,
721 vmf->address, &page_allocated);
722 if (!page)
723 continue;
724 if (page_allocated) {
725 swap_readpage(page, false);
eaf649eb 726 if (i != ra_info.offset) {
ec560175
HY
727 SetPageReadahead(page);
728 count_vm_event(SWAP_RA);
729 }
730 }
731 put_page(page);
732 }
733 blk_finish_plug(&plug);
734 lru_add_drain();
735skip:
736 return read_swap_cache_async(fentry, gfp_mask, vma, vmf->address,
eaf649eb 737 ra_info.win == 1);
ec560175 738}
d9bfcfdc 739
e9e9b7ec
MK
740/**
741 * swapin_readahead - swap in pages in hope we need them soon
742 * @entry: swap entry of this memory
743 * @gfp_mask: memory allocation flags
744 * @vmf: fault information
745 *
746 * Returns the struct page for entry and addr, after queueing swapin.
747 *
748 * It's a main entry function for swap readahead. By the configuration,
749 * it will read ahead blocks by cluster-based(ie, physical disk based)
750 * or vma-based(ie, virtual address based on faulty address) readahead.
751 */
752struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask,
753 struct vm_fault *vmf)
754{
755 return swap_use_vma_readahead() ?
756 swap_vma_readahead(entry, gfp_mask, vmf) :
757 swap_cluster_readahead(entry, gfp_mask, vmf);
758}
759
d9bfcfdc
HY
760#ifdef CONFIG_SYSFS
761static ssize_t vma_ra_enabled_show(struct kobject *kobj,
762 struct kobj_attribute *attr, char *buf)
763{
e9e9b7ec 764 return sprintf(buf, "%s\n", enable_vma_readahead ? "true" : "false");
d9bfcfdc
HY
765}
766static ssize_t vma_ra_enabled_store(struct kobject *kobj,
767 struct kobj_attribute *attr,
768 const char *buf, size_t count)
769{
770 if (!strncmp(buf, "true", 4) || !strncmp(buf, "1", 1))
e9e9b7ec 771 enable_vma_readahead = true;
d9bfcfdc 772 else if (!strncmp(buf, "false", 5) || !strncmp(buf, "0", 1))
e9e9b7ec 773 enable_vma_readahead = false;
d9bfcfdc
HY
774 else
775 return -EINVAL;
776
777 return count;
778}
779static struct kobj_attribute vma_ra_enabled_attr =
780 __ATTR(vma_ra_enabled, 0644, vma_ra_enabled_show,
781 vma_ra_enabled_store);
782
d9bfcfdc
HY
783static struct attribute *swap_attrs[] = {
784 &vma_ra_enabled_attr.attr,
d9bfcfdc
HY
785 NULL,
786};
787
788static struct attribute_group swap_attr_group = {
789 .attrs = swap_attrs,
790};
791
792static int __init swap_init_sysfs(void)
793{
794 int err;
795 struct kobject *swap_kobj;
796
797 swap_kobj = kobject_create_and_add("swap", mm_kobj);
798 if (!swap_kobj) {
799 pr_err("failed to create swap kobject\n");
800 return -ENOMEM;
801 }
802 err = sysfs_create_group(swap_kobj, &swap_attr_group);
803 if (err) {
804 pr_err("failed to register swap group\n");
805 goto delete_obj;
806 }
807 return 0;
808
809delete_obj:
810 kobject_put(swap_kobj);
811 return err;
812}
813subsys_initcall(swap_init_sysfs);
814#endif