linux/mm/vmscan.c
<<
>>
Prefs
   1/*
   2 *  linux/mm/vmscan.c
   3 *
   4 *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
   5 *
   6 *  Swap reorganised 29.12.95, Stephen Tweedie.
   7 *  kswapd added: 7.1.96  sct
   8 *  Removed kswapd_ctl limits, and swap out as many pages as needed
   9 *  to bring the system back to freepages.high: 2.4.97, Rik van Riel.
  10 *  Zone aware kswapd started 02/00, Kanoj Sarcar (kanoj@sgi.com).
  11 *  Multiqueue VM started 5.8.00, Rik van Riel.
  12 */
  13
  14#include <linux/mm.h>
  15#include <linux/module.h>
  16#include <linux/slab.h>
  17#include <linux/kernel_stat.h>
  18#include <linux/swap.h>
  19#include <linux/pagemap.h>
  20#include <linux/init.h>
  21#include <linux/highmem.h>
  22#include <linux/vmstat.h>
  23#include <linux/file.h>
  24#include <linux/writeback.h>
  25#include <linux/blkdev.h>
  26#include <linux/buffer_head.h>  /* for try_to_release_page(),
  27                                        buffer_heads_over_limit */
  28#include <linux/mm_inline.h>
  29#include <linux/pagevec.h>
  30#include <linux/backing-dev.h>
  31#include <linux/rmap.h>
  32#include <linux/topology.h>
  33#include <linux/cpu.h>
  34#include <linux/cpuset.h>
  35#include <linux/notifier.h>
  36#include <linux/rwsem.h>
  37#include <linux/delay.h>
  38#include <linux/kthread.h>
  39#include <linux/freezer.h>
  40#include <linux/memcontrol.h>
  41#include <linux/delayacct.h>
  42#include <linux/sysctl.h>
  43
  44#include <asm/tlbflush.h>
  45#include <asm/div64.h>
  46
  47#include <linux/swapops.h>
  48
  49#include "internal.h"
  50
  51struct scan_control {
  52        /* Incremented by the number of inactive pages that were scanned */
  53        unsigned long nr_scanned;
  54
  55        /* Number of pages freed so far during a call to shrink_zones() */
  56        unsigned long nr_reclaimed;
  57
  58        /* This context's GFP mask */
  59        gfp_t gfp_mask;
  60
  61        int may_writepage;
  62
  63        /* Can mapped pages be reclaimed? */
  64        int may_unmap;
  65
  66        /* Can pages be swapped as part of reclaim? */
  67        int may_swap;
  68
  69        /* This context's SWAP_CLUSTER_MAX. If freeing memory for
  70         * suspend, we effectively ignore SWAP_CLUSTER_MAX.
  71         * In this context, it doesn't matter that we scan the
  72         * whole list at once. */
  73        int swap_cluster_max;
  74
  75        int swappiness;
  76
  77        int all_unreclaimable;
  78
  79        int order;
  80
  81        /* Which cgroup do we reclaim from */
  82        struct mem_cgroup *mem_cgroup;
  83
  84        /*
  85         * Nodemask of nodes allowed by the caller. If NULL, all nodes
  86         * are scanned.
  87         */
  88        nodemask_t      *nodemask;
  89
  90        /* Pluggable isolate pages callback */
  91        unsigned long (*isolate_pages)(unsigned long nr, struct list_head *dst,
  92                        unsigned long *scanned, int order, int mode,
  93                        struct zone *z, struct mem_cgroup *mem_cont,
  94                        int active, int file);
  95};
  96
  97#define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru))
  98
  99#ifdef ARCH_HAS_PREFETCH
 100#define prefetch_prev_lru_page(_page, _base, _field)                    \
 101        do {                                                            \
 102                if ((_page)->lru.prev != _base) {                       \
 103                        struct page *prev;                              \
 104                                                                        \
 105                        prev = lru_to_page(&(_page->lru));              \
 106                        prefetch(&prev->_field);                        \
 107                }                                                       \
 108        } while (0)
 109#else
 110#define prefetch_prev_lru_page(_page, _base, _field) do { } while (0)
 111#endif
 112
 113#ifdef ARCH_HAS_PREFETCHW
 114#define prefetchw_prev_lru_page(_page, _base, _field)                   \
 115        do {                                                            \
 116                if ((_page)->lru.prev != _base) {                       \
 117                        struct page *prev;                              \
 118                                                                        \
 119                        prev = lru_to_page(&(_page->lru));              \
 120                        prefetchw(&prev->_field);                       \
 121                }                                                       \
 122        } while (0)
 123#else
 124#define prefetchw_prev_lru_page(_page, _base, _field) do { } while (0)
 125#endif
 126
 127/*
 128 * From 0 .. 100.  Higher means more swappy.
 129 */
 130int vm_swappiness = 60;
 131long vm_total_pages;    /* The total number of pages which the VM controls */
 132
 133static LIST_HEAD(shrinker_list);
 134static DECLARE_RWSEM(shrinker_rwsem);
 135
 136#ifdef CONFIG_CGROUP_MEM_RES_CTLR
 137#define scanning_global_lru(sc) (!(sc)->mem_cgroup)
 138#else
 139#define scanning_global_lru(sc) (1)
 140#endif
 141
 142static struct zone_reclaim_stat *get_reclaim_stat(struct zone *zone,
 143                                                  struct scan_control *sc)
 144{
 145        if (!scanning_global_lru(sc))
 146                return mem_cgroup_get_reclaim_stat(sc->mem_cgroup, zone);
 147
 148        return &zone->reclaim_stat;
 149}
 150
 151static unsigned long zone_nr_pages(struct zone *zone, struct scan_control *sc,
 152                                   enum lru_list lru)
 153{
 154        if (!scanning_global_lru(sc))
 155                return mem_cgroup_zone_nr_pages(sc->mem_cgroup, zone, lru);
 156
 157        return zone_page_state(zone, NR_LRU_BASE + lru);
 158}
 159
 160
 161/*
 162 * Add a shrinker callback to be called from the vm
 163 */
 164void register_shrinker(struct shrinker *shrinker)
 165{
 166        shrinker->nr = 0;
 167        down_write(&shrinker_rwsem);
 168        list_add_tail(&shrinker->list, &shrinker_list);
 169        up_write(&shrinker_rwsem);
 170}
 171EXPORT_SYMBOL(register_shrinker);
 172
 173/*
 174 * Remove one
 175 */
 176void unregister_shrinker(struct shrinker *shrinker)
 177{
 178        down_write(&shrinker_rwsem);
 179        list_del(&shrinker->list);
 180        up_write(&shrinker_rwsem);
 181}
 182EXPORT_SYMBOL(unregister_shrinker);
 183
 184#define SHRINK_BATCH 128
 185/*
 186 * Call the shrink functions to age shrinkable caches
 187 *
 188 * Here we assume it costs one seek to replace a lru page and that it also
 189 * takes a seek to recreate a cache object.  With this in mind we age equal
 190 * percentages of the lru and ageable caches.  This should balance the seeks
 191 * generated by these structures.
 192 *
 193 * If the vm encountered mapped pages on the LRU it increase the pressure on
 194 * slab to avoid swapping.
 195 *
 196 * We do weird things to avoid (scanned*seeks*entries) overflowing 32 bits.
 197 *
 198 * `lru_pages' represents the number of on-LRU pages in all the zones which
 199 * are eligible for the caller's allocation attempt.  It is used for balancing
 200 * slab reclaim versus page reclaim.
 201 *
 202 * Returns the number of slab objects which we shrunk.
 203 */
 204unsigned long shrink_slab(unsigned long scanned, gfp_t gfp_mask,
 205                        unsigned long lru_pages)
 206{
 207        struct shrinker *shrinker;
 208        unsigned long ret = 0;
 209
 210        if (scanned == 0)
 211                scanned = SWAP_CLUSTER_MAX;
 212
 213        if (!down_read_trylock(&shrinker_rwsem))
 214                return 1;       /* Assume we'll be able to shrink next time */
 215
 216        list_for_each_entry(shrinker, &shrinker_list, list) {
 217                unsigned long long delta;
 218                unsigned long total_scan;
 219                unsigned long max_pass = (*shrinker->shrink)(0, gfp_mask);
 220
 221                delta = (4 * scanned) / shrinker->seeks;
 222                delta *= max_pass;
 223                do_div(delta, lru_pages + 1);
 224                shrinker->nr += delta;
 225                if (shrinker->nr < 0) {
 226                        printk(KERN_ERR "shrink_slab: %pF negative objects to "
 227                               "delete nr=%ld\n",
 228                               shrinker->shrink, shrinker->nr);
 229                        shrinker->nr = max_pass;
 230                }
 231
 232                /*
 233                 * Avoid risking looping forever due to too large nr value:
 234                 * never try to free more than twice the estimate number of
 235                 * freeable entries.
 236                 */
 237                if (shrinker->nr > max_pass * 2)
 238                        shrinker->nr = max_pass * 2;
 239
 240                total_scan = shrinker->nr;
 241                shrinker->nr = 0;
 242
 243                while (total_scan >= SHRINK_BATCH) {
 244                        long this_scan = SHRINK_BATCH;
 245                        int shrink_ret;
 246                        int nr_before;
 247
 248                        nr_before = (*shrinker->shrink)(0, gfp_mask);
 249                        shrink_ret = (*shrinker->shrink)(this_scan, gfp_mask);
 250                        if (shrink_ret == -1)
 251                                break;
 252                        if (shrink_ret < nr_before)
 253                                ret += nr_before - shrink_ret;
 254                        count_vm_events(SLABS_SCANNED, this_scan);
 255                        total_scan -= this_scan;
 256
 257                        cond_resched();
 258                }
 259
 260                shrinker->nr += total_scan;
 261        }
 262        up_read(&shrinker_rwsem);
 263        return ret;
 264}
 265
 266/* Called without lock on whether page is mapped, so answer is unstable */
 267static inline int page_mapping_inuse(struct page *page)
 268{
 269        struct address_space *mapping;
 270
 271        /* Page is in somebody's page tables. */
 272        if (page_mapped(page))
 273                return 1;
 274
 275        /* Be more reluctant to reclaim swapcache than pagecache */
 276        if (PageSwapCache(page))
 277                return 1;
 278
 279        mapping = page_mapping(page);
 280        if (!mapping)
 281                return 0;
 282
 283        /* File is mmap'd by somebody? */
 284        return mapping_mapped(mapping);
 285}
 286
 287static inline int is_page_cache_freeable(struct page *page)
 288{
 289        return page_count(page) - !!page_has_private(page) == 2;
 290}
 291
 292static int may_write_to_queue(struct backing_dev_info *bdi)
 293{
 294        if (current->flags & PF_SWAPWRITE)
 295                return 1;
 296        if (!bdi_write_congested(bdi))
 297                return 1;
 298        if (bdi == current->backing_dev_info)
 299                return 1;
 300        return 0;
 301}
 302
 303/*
 304 * We detected a synchronous write error writing a page out.  Probably
 305 * -ENOSPC.  We need to propagate that into the address_space for a subsequent
 306 * fsync(), msync() or close().
 307 *
 308 * The tricky part is that after writepage we cannot touch the mapping: nothing
 309 * prevents it from being freed up.  But we have a ref on the page and once
 310 * that page is locked, the mapping is pinned.
 311 *
 312 * We're allowed to run sleeping lock_page() here because we know the caller has
 313 * __GFP_FS.
 314 */
 315static void handle_write_error(struct address_space *mapping,
 316                                struct page *page, int error)
 317{
 318        lock_page(page);
 319        if (page_mapping(page) == mapping)
 320                mapping_set_error(mapping, error);
 321        unlock_page(page);
 322}
 323
 324/* Request for sync pageout. */
 325enum pageout_io {
 326        PAGEOUT_IO_ASYNC,
 327        PAGEOUT_IO_SYNC,
 328};
 329
 330/* possible outcome of pageout() */
 331typedef enum {
 332        /* failed to write page out, page is locked */
 333        PAGE_KEEP,
 334        /* move page to the active list, page is locked */
 335        PAGE_ACTIVATE,
 336        /* page has been sent to the disk successfully, page is unlocked */
 337        PAGE_SUCCESS,
 338        /* page is clean and locked */
 339        PAGE_CLEAN,
 340} pageout_t;
 341
 342/*
 343 * pageout is called by shrink_page_list() for each dirty page.
 344 * Calls ->writepage().
 345 */
 346static pageout_t pageout(struct page *page, struct address_space *mapping,
 347                                                enum pageout_io sync_writeback)
 348{
 349        /*
 350         * If the page is dirty, only perform writeback if that write
 351         * will be non-blocking.  To prevent this allocation from being
 352         * stalled by pagecache activity.  But note that there may be
 353         * stalls if we need to run get_block().  We could test
 354         * PagePrivate for that.
 355         *
 356         * If this process is currently in generic_file_write() against
 357         * this page's queue, we can perform writeback even if that
 358         * will block.
 359         *
 360         * If the page is swapcache, write it back even if that would
 361         * block, for some throttling. This happens by accident, because
 362         * swap_backing_dev_info is bust: it doesn't reflect the
 363         * congestion state of the swapdevs.  Easy to fix, if needed.
 364         * See swapfile.c:page_queue_congested().
 365         */
 366        if (!is_page_cache_freeable(page))
 367                return PAGE_KEEP;
 368        if (!mapping) {
 369                /*
 370                 * Some data journaling orphaned pages can have
 371                 * page->mapping == NULL while being dirty with clean buffers.
 372                 */
 373                if (page_has_private(page)) {
 374                        if (try_to_free_buffers(page)) {
 375                                ClearPageDirty(page);
 376                                printk("%s: orphaned page\n", __func__);
 377                                return PAGE_CLEAN;
 378                        }
 379                }
 380                return PAGE_KEEP;
 381        }
 382        if (mapping->a_ops->writepage == NULL)
 383                return PAGE_ACTIVATE;
 384        if (!may_write_to_queue(mapping->backing_dev_info))
 385                return PAGE_KEEP;
 386
 387        if (clear_page_dirty_for_io(page)) {
 388                int res;
 389                struct writeback_control wbc = {
 390                        .sync_mode = WB_SYNC_NONE,
 391                        .nr_to_write = SWAP_CLUSTER_MAX,
 392                        .range_start = 0,
 393                        .range_end = LLONG_MAX,
 394                        .nonblocking = 1,
 395                        .for_reclaim = 1,
 396                };
 397
 398                SetPageReclaim(page);
 399                res = mapping->a_ops->writepage(page, &wbc);
 400                if (res < 0)
 401                        handle_write_error(mapping, page, res);
 402                if (res == AOP_WRITEPAGE_ACTIVATE) {
 403                        ClearPageReclaim(page);
 404                        return PAGE_ACTIVATE;
 405                }
 406
 407                /*
 408                 * Wait on writeback if requested to. This happens when
 409                 * direct reclaiming a large contiguous area and the
 410                 * first attempt to free a range of pages fails.
 411                 */
 412                if (PageWriteback(page) && sync_writeback == PAGEOUT_IO_SYNC)
 413                        wait_on_page_writeback(page);
 414
 415                if (!PageWriteback(page)) {
 416                        /* synchronous write or broken a_ops? */
 417                        ClearPageReclaim(page);
 418                }
 419                inc_zone_page_state(page, NR_VMSCAN_WRITE);
 420                return PAGE_SUCCESS;
 421        }
 422
 423        return PAGE_CLEAN;
 424}
 425
 426/*
 427 * Same as remove_mapping, but if the page is removed from the mapping, it
 428 * gets returned with a refcount of 0.
 429 */
 430static int __remove_mapping(struct address_space *mapping, struct page *page)
 431{
 432        BUG_ON(!PageLocked(page));
 433        BUG_ON(mapping != page_mapping(page));
 434
 435        spin_lock_irq(&mapping->tree_lock);
 436        /*
 437         * The non racy check for a busy page.
 438         *
 439         * Must be careful with the order of the tests. When someone has
 440         * a ref to the page, it may be possible that they dirty it then
 441         * drop the reference. So if PageDirty is tested before page_count
 442         * here, then the following race may occur:
 443         *
 444         * get_user_pages(&page);
 445         * [user mapping goes away]
 446         * write_to(page);
 447         *                              !PageDirty(page)    [good]
 448         * SetPageDirty(page);
 449         * put_page(page);
 450         *                              !page_count(page)   [good, discard it]
 451         *
 452         * [oops, our write_to data is lost]
 453         *
 454         * Reversing the order of the tests ensures such a situation cannot
 455         * escape unnoticed. The smp_rmb is needed to ensure the page->flags
 456         * load is not satisfied before that of page->_count.
 457         *
 458         * Note that if SetPageDirty is always performed via set_page_dirty,
 459         * and thus under tree_lock, then this ordering is not required.
 460         */
 461        if (!page_freeze_refs(page, 2))
 462                goto cannot_free;
 463        /* note: atomic_cmpxchg in page_freeze_refs provides the smp_rmb */
 464        if (unlikely(PageDirty(page))) {
 465                page_unfreeze_refs(page, 2);
 466                goto cannot_free;
 467        }
 468
 469        if (PageSwapCache(page)) {
 470                swp_entry_t swap = { .val = page_private(page) };
 471                __delete_from_swap_cache(page);
 472                spin_unlock_irq(&mapping->tree_lock);
 473                mem_cgroup_uncharge_swapcache(page, swap);
 474                swap_free(swap);
 475        } else {
 476                __remove_from_page_cache(page);
 477                spin_unlock_irq(&mapping->tree_lock);
 478                mem_cgroup_uncharge_cache_page(page);
 479        }
 480
 481        return 1;
 482
 483cannot_free:
 484        spin_unlock_irq(&mapping->tree_lock);
 485        return 0;
 486}
 487
 488/*
 489 * Attempt to detach a locked page from its ->mapping.  If it is dirty or if
 490 * someone else has a ref on the page, abort and return 0.  If it was
 491 * successfully detached, return 1.  Assumes the caller has a single ref on
 492 * this page.
 493 */
 494int remove_mapping(struct address_space *mapping, struct page *page)
 495{
 496        if (__remove_mapping(mapping, page)) {
 497                /*
 498                 * Unfreezing the refcount with 1 rather than 2 effectively
 499                 * drops the pagecache ref for us without requiring another
 500                 * atomic operation.
 501                 */
 502                page_unfreeze_refs(page, 1);
 503                return 1;
 504        }
 505        return 0;
 506}
 507
 508/**
 509 * putback_lru_page - put previously isolated page onto appropriate LRU list
 510 * @page: page to be put back to appropriate lru list
 511 *
 512 * Add previously isolated @page to appropriate LRU list.
 513 * Page may still be unevictable for other reasons.
 514 *
 515 * lru_lock must not be held, interrupts must be enabled.
 516 */
 517#ifdef CONFIG_UNEVICTABLE_LRU
 518void putback_lru_page(struct page *page)
 519{
 520        int lru;
 521        int active = !!TestClearPageActive(page);
 522        int was_unevictable = PageUnevictable(page);
 523
 524        VM_BUG_ON(PageLRU(page));
 525
 526redo:
 527        ClearPageUnevictable(page);
 528
 529        if (page_evictable(page, NULL)) {
 530                /*
 531                 * For evictable pages, we can use the cache.
 532                 * In event of a race, worst case is we end up with an
 533                 * unevictable page on [in]active list.
 534                 * We know how to handle that.
 535                 */
 536                lru = active + page_is_file_cache(page);
 537                lru_cache_add_lru(page, lru);
 538        } else {
 539                /*
 540                 * Put unevictable pages directly on zone's unevictable
 541                 * list.
 542                 */
 543                lru = LRU_UNEVICTABLE;
 544                add_page_to_unevictable_list(page);
 545        }
 546
 547        /*
 548         * page's status can change while we move it among lru. If an evictable
 549         * page is on unevictable list, it never be freed. To avoid that,
 550         * check after we added it to the list, again.
 551         */
 552        if (lru == LRU_UNEVICTABLE && page_evictable(page, NULL)) {
 553                if (!isolate_lru_page(page)) {
 554                        put_page(page);
 555                        goto redo;
 556                }
 557                /* This means someone else dropped this page from LRU
 558                 * So, it will be freed or putback to LRU again. There is
 559                 * nothing to do here.
 560                 */
 561        }
 562
 563        if (was_unevictable && lru != LRU_UNEVICTABLE)
 564                count_vm_event(UNEVICTABLE_PGRESCUED);
 565        else if (!was_unevictable && lru == LRU_UNEVICTABLE)
 566                count_vm_event(UNEVICTABLE_PGCULLED);
 567
 568        put_page(page);         /* drop ref from isolate */
 569}
 570
 571#else /* CONFIG_UNEVICTABLE_LRU */
 572
 573void putback_lru_page(struct page *page)
 574{
 575        int lru;
 576        VM_BUG_ON(PageLRU(page));
 577
 578        lru = !!TestClearPageActive(page) + page_is_file_cache(page);
 579        lru_cache_add_lru(page, lru);
 580        put_page(page);
 581}
 582#endif /* CONFIG_UNEVICTABLE_LRU */
 583
 584
 585/*
 586 * shrink_page_list() returns the number of reclaimed pages
 587 */
 588static unsigned long shrink_page_list(struct list_head *page_list,
 589                                        struct scan_control *sc,
 590                                        enum pageout_io sync_writeback)
 591{
 592        LIST_HEAD(ret_pages);
 593        struct pagevec freed_pvec;
 594        int pgactivate = 0;
 595        unsigned long nr_reclaimed = 0;
 596
 597        cond_resched();
 598
 599        pagevec_init(&freed_pvec, 1);
 600        while (!list_empty(page_list)) {
 601                struct address_space *mapping;
 602                struct page *page;
 603                int may_enter_fs;
 604                int referenced;
 605
 606                cond_resched();
 607
 608                page = lru_to_page(page_list);
 609                list_del(&page->lru);
 610
 611                if (!trylock_page(page))
 612                        goto keep;
 613
 614                VM_BUG_ON(PageActive(page));
 615
 616                sc->nr_scanned++;
 617
 618                if (unlikely(!page_evictable(page, NULL)))
 619                        goto cull_mlocked;
 620
 621                if (!sc->may_unmap && page_mapped(page))
 622                        goto keep_locked;
 623
 624                /* Double the slab pressure for mapped and swapcache pages */
 625                if (page_mapped(page) || PageSwapCache(page))
 626                        sc->nr_scanned++;
 627
 628                may_enter_fs = (sc->gfp_mask & __GFP_FS) ||
 629                        (PageSwapCache(page) && (sc->gfp_mask & __GFP_IO));
 630
 631                if (PageWriteback(page)) {
 632                        /*
 633                         * Synchronous reclaim is performed in two passes,
 634                         * first an asynchronous pass over the list to
 635                         * start parallel writeback, and a second synchronous
 636                         * pass to wait for the IO to complete.  Wait here
 637                         * for any page for which writeback has already
 638                         * started.
 639                         */
 640                        if (sync_writeback == PAGEOUT_IO_SYNC && may_enter_fs)
 641                                wait_on_page_writeback(page);
 642                        else
 643                                goto keep_locked;
 644                }
 645
 646                referenced = page_referenced(page, 1, sc->mem_cgroup);
 647                /* In active use or really unfreeable?  Activate it. */
 648                if (sc->order <= PAGE_ALLOC_COSTLY_ORDER &&
 649                                        referenced && page_mapping_inuse(page))
 650                        goto activate_locked;
 651
 652                /*
 653                 * Anonymous process memory has backing store?
 654                 * Try to allocate it some swap space here.
 655                 */
 656                if (PageAnon(page) && !PageSwapCache(page)) {
 657                        if (!(sc->gfp_mask & __GFP_IO))
 658                                goto keep_locked;
 659                        if (!add_to_swap(page))
 660                                goto activate_locked;
 661                        may_enter_fs = 1;
 662                }
 663
 664                mapping = page_mapping(page);
 665
 666                /*
 667                 * The page is mapped into the page tables of one or more
 668                 * processes. Try to unmap it here.
 669                 */
 670                if (page_mapped(page) && mapping) {
 671                        switch (try_to_unmap(page, 0)) {
 672                        case SWAP_FAIL:
 673                                goto activate_locked;
 674                        case SWAP_AGAIN:
 675                                goto keep_locked;
 676                        case SWAP_MLOCK:
 677                                goto cull_mlocked;
 678                        case SWAP_SUCCESS:
 679                                ; /* try to free the page below */
 680                        }
 681                }
 682
 683                if (PageDirty(page)) {
 684                        if (sc->order <= PAGE_ALLOC_COSTLY_ORDER && referenced)
 685                                goto keep_locked;
 686                        if (!may_enter_fs)
 687                                goto keep_locked;
 688                        if (!sc->may_writepage)
 689                                goto keep_locked;
 690
 691                        /* Page is dirty, try to write it out here */
 692                        switch (pageout(page, mapping, sync_writeback)) {
 693                        case PAGE_KEEP:
 694                                goto keep_locked;
 695                        case PAGE_ACTIVATE:
 696                                goto activate_locked;
 697                        case PAGE_SUCCESS:
 698                                if (PageWriteback(page) || PageDirty(page))
 699                                        goto keep;
 700                                /*
 701                                 * A synchronous write - probably a ramdisk.  Go
 702                                 * ahead and try to reclaim the page.
 703                                 */
 704                                if (!trylock_page(page))
 705                                        goto keep;
 706                                if (PageDirty(page) || PageWriteback(page))
 707                                        goto keep_locked;
 708                                mapping = page_mapping(page);
 709                        case PAGE_CLEAN:
 710                                ; /* try to free the page below */
 711                        }
 712                }
 713
 714                /*
 715                 * If the page has buffers, try to free the buffer mappings
 716                 * associated with this page. If we succeed we try to free
 717                 * the page as well.
 718                 *
 719                 * We do this even if the page is PageDirty().
 720                 * try_to_release_page() does not perform I/O, but it is
 721                 * possible for a page to have PageDirty set, but it is actually
 722                 * clean (all its buffers are clean).  This happens if the
 723                 * buffers were written out directly, with submit_bh(). ext3
 724                 * will do this, as well as the blockdev mapping.
 725                 * try_to_release_page() will discover that cleanness and will
 726                 * drop the buffers and mark the page clean - it can be freed.
 727                 *
 728                 * Rarely, pages can have buffers and no ->mapping.  These are
 729                 * the pages which were not successfully invalidated in
 730                 * truncate_complete_page().  We try to drop those buffers here
 731                 * and if that worked, and the page is no longer mapped into
 732                 * process address space (page_count == 1) it can be freed.
 733                 * Otherwise, leave the page on the LRU so it is swappable.
 734                 */
 735                if (page_has_private(page)) {
 736                        if (!try_to_release_page(page, sc->gfp_mask))
 737                                goto activate_locked;
 738                        if (!mapping && page_count(page) == 1) {
 739                                unlock_page(page);
 740                                if (put_page_testzero(page))
 741                                        goto free_it;
 742                                else {
 743                                        /*
 744                                         * rare race with speculative reference.
 745                                         * the speculative reference will free
 746                                         * this page shortly, so we may
 747                                         * increment nr_reclaimed here (and
 748                                         * leave it off the LRU).
 749                                         */
 750                                        nr_reclaimed++;
 751                                        continue;
 752                                }
 753                        }
 754                }
 755
 756                if (!mapping || !__remove_mapping(mapping, page))
 757                        goto keep_locked;
 758
 759                /*
 760                 * At this point, we have no other references and there is
 761                 * no way to pick any more up (removed from LRU, removed
 762                 * from pagecache). Can use non-atomic bitops now (and
 763                 * we obviously don't have to worry about waking up a process
 764                 * waiting on the page lock, because there are no references.
 765                 */
 766                __clear_page_locked(page);
 767free_it:
 768                nr_reclaimed++;
 769                if (!pagevec_add(&freed_pvec, page)) {
 770                        __pagevec_free(&freed_pvec);
 771                        pagevec_reinit(&freed_pvec);
 772                }
 773                continue;
 774
 775cull_mlocked:
 776                if (PageSwapCache(page))
 777                        try_to_free_swap(page);
 778                unlock_page(page);
 779                putback_lru_page(page);
 780                continue;
 781
 782activate_locked:
 783                /* Not a candidate for swapping, so reclaim swap space. */
 784                if (PageSwapCache(page) && vm_swap_full())
 785                        try_to_free_swap(page);
 786                VM_BUG_ON(PageActive(page));
 787                SetPageActive(page);
 788                pgactivate++;
 789keep_locked:
 790                unlock_page(page);
 791keep:
 792                list_add(&page->lru, &ret_pages);
 793                VM_BUG_ON(PageLRU(page) || PageUnevictable(page));
 794        }
 795        list_splice(&ret_pages, page_list);
 796        if (pagevec_count(&freed_pvec))
 797                __pagevec_free(&freed_pvec);
 798        count_vm_events(PGACTIVATE, pgactivate);
 799        return nr_reclaimed;
 800}
 801
 802/* LRU Isolation modes. */
 803#define ISOLATE_INACTIVE 0      /* Isolate inactive pages. */
 804#define ISOLATE_ACTIVE 1        /* Isolate active pages. */
 805#define ISOLATE_BOTH 2          /* Isolate both active and inactive pages. */
 806
 807/*
 808 * Attempt to remove the specified page from its LRU.  Only take this page
 809 * if it is of the appropriate PageActive status.  Pages which are being
 810 * freed elsewhere are also ignored.
 811 *
 812 * page:        page to consider
 813 * mode:        one of the LRU isolation modes defined above
 814 *
 815 * returns 0 on success, -ve errno on failure.
 816 */
 817int __isolate_lru_page(struct page *page, int mode, int file)
 818{
 819        int ret = -EINVAL;
 820
 821        /* Only take pages on the LRU. */
 822        if (!PageLRU(page))
 823                return ret;
 824
 825        /*
 826         * When checking the active state, we need to be sure we are
 827         * dealing with comparible boolean values.  Take the logical not
 828         * of each.
 829         */
 830        if (mode != ISOLATE_BOTH && (!PageActive(page) != !mode))
 831                return ret;
 832
 833        if (mode != ISOLATE_BOTH && (!page_is_file_cache(page) != !file))
 834                return ret;
 835
 836        /*
 837         * When this function is being called for lumpy reclaim, we
 838         * initially look into all LRU pages, active, inactive and
 839         * unevictable; only give shrink_page_list evictable pages.
 840         */
 841        if (PageUnevictable(page))
 842                return ret;
 843
 844        ret = -EBUSY;
 845
 846        if (likely(get_page_unless_zero(page))) {
 847                /*
 848                 * Be careful not to clear PageLRU until after we're
 849                 * sure the page is not being freed elsewhere -- the
 850                 * page release code relies on it.
 851                 */
 852                ClearPageLRU(page);
 853                ret = 0;
 854                mem_cgroup_del_lru(page);
 855        }
 856
 857        return ret;
 858}
 859
 860/*
 861 * zone->lru_lock is heavily contended.  Some of the functions that
 862 * shrink the lists perform better by taking out a batch of pages
 863 * and working on them outside the LRU lock.
 864 *
 865 * For pagecache intensive workloads, this function is the hottest
 866 * spot in the kernel (apart from copy_*_user functions).
 867 *
 868 * Appropriate locks must be held before calling this function.
 869 *
 870 * @nr_to_scan: The number of pages to look through on the list.
 871 * @src:        The LRU list to pull pages off.
 872 * @dst:        The temp list to put pages on to.
 873 * @scanned:    The number of pages that were scanned.
 874 * @order:      The caller's attempted allocation order
 875 * @mode:       One of the LRU isolation modes
 876 * @file:       True [1] if isolating file [!anon] pages
 877 *
 878 * returns how many pages were moved onto *@dst.
 879 */
 880static unsigned long isolate_lru_pages(unsigned long nr_to_scan,
 881                struct list_head *src, struct list_head *dst,
 882                unsigned long *scanned, int order, int mode, int file)
 883{
 884        unsigned long nr_taken = 0;
 885        unsigned long scan;
 886
 887        for (scan = 0; scan < nr_to_scan && !list_empty(src); scan++) {
 888                struct page *page;
 889                unsigned long pfn;
 890                unsigned long end_pfn;
 891                unsigned long page_pfn;
 892                int zone_id;
 893
 894                page = lru_to_page(src);
 895                prefetchw_prev_lru_page(page, src, flags);
 896
 897                VM_BUG_ON(!PageLRU(page));
 898
 899                switch (__isolate_lru_page(page, mode, file)) {
 900                case 0:
 901                        list_move(&page->lru, dst);
 902                        nr_taken++;
 903                        break;
 904
 905                case -EBUSY:
 906                        /* else it is being freed elsewhere */
 907                        list_move(&page->lru, src);
 908                        continue;
 909
 910                default:
 911                        BUG();
 912                }
 913
 914                if (!order)
 915                        continue;
 916
 917                /*
 918                 * Attempt to take all pages in the order aligned region
 919                 * surrounding the tag page.  Only take those pages of
 920                 * the same active state as that tag page.  We may safely
 921                 * round the target page pfn down to the requested order
 922                 * as the mem_map is guarenteed valid out to MAX_ORDER,
 923                 * where that page is in a different zone we will detect
 924                 * it from its zone id and abort this block scan.
 925                 */
 926                zone_id = page_zone_id(page);
 927                page_pfn = page_to_pfn(page);
 928                pfn = page_pfn & ~((1 << order) - 1);
 929                end_pfn = pfn + (1 << order);
 930                for (; pfn < end_pfn; pfn++) {
 931                        struct page *cursor_page;
 932
 933                        /* The target page is in the block, ignore it. */
 934                        if (unlikely(pfn == page_pfn))
 935                                continue;
 936
 937                        /* Avoid holes within the zone. */
 938                        if (unlikely(!pfn_valid_within(pfn)))
 939                                break;
 940
 941                        cursor_page = pfn_to_page(pfn);
 942
 943                        /* Check that we have not crossed a zone boundary. */
 944                        if (unlikely(page_zone_id(cursor_page) != zone_id))
 945                                continue;
 946                        switch (__isolate_lru_page(cursor_page, mode, file)) {
 947                        case 0:
 948                                list_move(&cursor_page->lru, dst);
 949                                nr_taken++;
 950                                scan++;
 951                                break;
 952
 953                        case -EBUSY:
 954                                /* else it is being freed elsewhere */
 955                                list_move(&cursor_page->lru, src);
 956                        default:
 957                                break;  /* ! on LRU or wrong list */
 958                        }
 959                }
 960        }
 961
 962        *scanned = scan;
 963        return nr_taken;
 964}
 965
 966static unsigned long isolate_pages_global(unsigned long nr,
 967                                        struct list_head *dst,
 968                                        unsigned long *scanned, int order,
 969                                        int mode, struct zone *z,
 970                                        struct mem_cgroup *mem_cont,
 971                                        int active, int file)
 972{
 973        int lru = LRU_BASE;
 974        if (active)
 975                lru += LRU_ACTIVE;
 976        if (file)
 977                lru += LRU_FILE;
 978        return isolate_lru_pages(nr, &z->lru[lru].list, dst, scanned, order,
 979                                                                mode, !!file);
 980}
 981
 982/*
 983 * clear_active_flags() is a helper for shrink_active_list(), clearing
 984 * any active bits from the pages in the list.
 985 */
 986static unsigned long clear_active_flags(struct list_head *page_list,
 987                                        unsigned int *count)
 988{
 989        int nr_active = 0;
 990        int lru;
 991        struct page *page;
 992
 993        list_for_each_entry(page, page_list, lru) {
 994                lru = page_is_file_cache(page);
 995                if (PageActive(page)) {
 996                        lru += LRU_ACTIVE;
 997                        ClearPageActive(page);
 998                        nr_active++;
 999                }
1000                count[lru]++;
1001        }
1002
1003        return nr_active;
1004}
1005
1006/**
1007 * isolate_lru_page - tries to isolate a page from its LRU list
1008 * @page: page to isolate from its LRU list
1009 *
1010 * Isolates a @page from an LRU list, clears PageLRU and adjusts the
1011 * vmstat statistic corresponding to whatever LRU list the page was on.
1012 *
1013 * Returns 0 if the page was removed from an LRU list.
1014 * Returns -EBUSY if the page was not on an LRU list.
1015 *
1016 * The returned page will have PageLRU() cleared.  If it was found on
1017 * the active list, it will have PageActive set.  If it was found on
1018 * the unevictable list, it will have the PageUnevictable bit set. That flag
1019 * may need to be cleared by the caller before letting the page go.
1020 *
1021 * The vmstat statistic corresponding to the list on which the page was
1022 * found will be decremented.
1023 *
1024 * Restrictions:
1025 * (1) Must be called with an elevated refcount on the page. This is a
1026 *     fundamentnal difference from isolate_lru_pages (which is called
1027 *     without a stable reference).
1028 * (2) the lru_lock must not be held.
1029 * (3) interrupts must be enabled.
1030 */
1031int isolate_lru_page(struct page *page)
1032{
1033        int ret = -EBUSY;
1034
1035        if (PageLRU(page)) {
1036                struct zone *zone = page_zone(page);
1037
1038                spin_lock_irq(&zone->lru_lock);
1039                if (PageLRU(page) && get_page_unless_zero(page)) {
1040                        int lru = page_lru(page);
1041                        ret = 0;
1042                        ClearPageLRU(page);
1043
1044                        del_page_from_lru_list(zone, page, lru);
1045                }
1046                spin_unlock_irq(&zone->lru_lock);
1047        }
1048        return ret;
1049}
1050
1051/*
1052 * shrink_inactive_list() is a helper for shrink_zone().  It returns the number
1053 * of reclaimed pages
1054 */
1055static unsigned long shrink_inactive_list(unsigned long max_scan,
1056                        struct zone *zone, struct scan_control *sc,
1057                        int priority, int file)
1058{
1059        LIST_HEAD(page_list);
1060        struct pagevec pvec;
1061        unsigned long nr_scanned = 0;
1062        unsigned long nr_reclaimed = 0;
1063        struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc);
1064
1065        pagevec_init(&pvec, 1);
1066
1067        lru_add_drain();
1068        spin_lock_irq(&zone->lru_lock);
1069        do {
1070                struct page *page;
1071                unsigned long nr_taken;
1072                unsigned long nr_scan;
1073                unsigned long nr_freed;
1074                unsigned long nr_active;
1075                unsigned int count[NR_LRU_LISTS] = { 0, };
1076                int mode = ISOLATE_INACTIVE;
1077
1078                /*
1079                 * If we need a large contiguous chunk of memory, or have
1080                 * trouble getting a small set of contiguous pages, we
1081                 * will reclaim both active and inactive pages.
1082                 *
1083                 * We use the same threshold as pageout congestion_wait below.
1084                 */
1085                if (sc->order > PAGE_ALLOC_COSTLY_ORDER)
1086                        mode = ISOLATE_BOTH;
1087                else if (sc->order && priority < DEF_PRIORITY - 2)
1088                        mode = ISOLATE_BOTH;
1089
1090                nr_taken = sc->isolate_pages(sc->swap_cluster_max,
1091                             &page_list, &nr_scan, sc->order, mode,
1092                                zone, sc->mem_cgroup, 0, file);
1093                nr_active = clear_active_flags(&page_list, count);
1094                __count_vm_events(PGDEACTIVATE, nr_active);
1095
1096                __mod_zone_page_state(zone, NR_ACTIVE_FILE,
1097                                                -count[LRU_ACTIVE_FILE]);
1098                __mod_zone_page_state(zone, NR_INACTIVE_FILE,
1099                                                -count[LRU_INACTIVE_FILE]);
1100                __mod_zone_page_state(zone, NR_ACTIVE_ANON,
1101                                                -count[LRU_ACTIVE_ANON]);
1102                __mod_zone_page_state(zone, NR_INACTIVE_ANON,
1103                                                -count[LRU_INACTIVE_ANON]);
1104
1105                if (scanning_global_lru(sc))
1106                        zone->pages_scanned += nr_scan;
1107
1108                reclaim_stat->recent_scanned[0] += count[LRU_INACTIVE_ANON];
1109                reclaim_stat->recent_scanned[0] += count[LRU_ACTIVE_ANON];
1110                reclaim_stat->recent_scanned[1] += count[LRU_INACTIVE_FILE];
1111                reclaim_stat->recent_scanned[1] += count[LRU_ACTIVE_FILE];
1112
1113                spin_unlock_irq(&zone->lru_lock);
1114
1115                nr_scanned += nr_scan;
1116                nr_freed = shrink_page_list(&page_list, sc, PAGEOUT_IO_ASYNC);
1117
1118                /*
1119                 * If we are direct reclaiming for contiguous pages and we do
1120                 * not reclaim everything in the list, try again and wait
1121                 * for IO to complete. This will stall high-order allocations
1122                 * but that should be acceptable to the caller
1123                 */
1124                if (nr_freed < nr_taken && !current_is_kswapd() &&
1125                                        sc->order > PAGE_ALLOC_COSTLY_ORDER) {
1126                        congestion_wait(WRITE, HZ/10);
1127
1128                        /*
1129                         * The attempt at page out may have made some
1130                         * of the pages active, mark them inactive again.
1131                         */
1132                        nr_active = clear_active_flags(&page_list, count);
1133                        count_vm_events(PGDEACTIVATE, nr_active);
1134
1135                        nr_freed += shrink_page_list(&page_list, sc,
1136                                                        PAGEOUT_IO_SYNC);
1137                }
1138
1139                nr_reclaimed += nr_freed;
1140                local_irq_disable();
1141                if (current_is_kswapd()) {
1142                        __count_zone_vm_events(PGSCAN_KSWAPD, zone, nr_scan);
1143                        __count_vm_events(KSWAPD_STEAL, nr_freed);
1144                } else if (scanning_global_lru(sc))
1145                        __count_zone_vm_events(PGSCAN_DIRECT, zone, nr_scan);
1146
1147                __count_zone_vm_events(PGSTEAL, zone, nr_freed);
1148
1149                if (nr_taken == 0)
1150                        goto done;
1151
1152                spin_lock(&zone->lru_lock);
1153                /*
1154                 * Put back any unfreeable pages.
1155                 */
1156                while (!list_empty(&page_list)) {
1157                        int lru;
1158                        page = lru_to_page(&page_list);
1159                        VM_BUG_ON(PageLRU(page));
1160                        list_del(&page->lru);
1161                        if (unlikely(!page_evictable(page, NULL))) {
1162                                spin_unlock_irq(&zone->lru_lock);
1163                                putback_lru_page(page);
1164                                spin_lock_irq(&zone->lru_lock);
1165                                continue;
1166                        }
1167                        SetPageLRU(page);
1168                        lru = page_lru(page);
1169                        add_page_to_lru_list(zone, page, lru);
1170                        if (PageActive(page)) {
1171                                int file = !!page_is_file_cache(page);
1172                                reclaim_stat->recent_rotated[file]++;
1173                        }
1174                        if (!pagevec_add(&pvec, page)) {
1175                                spin_unlock_irq(&zone->lru_lock);
1176                                __pagevec_release(&pvec);
1177                                spin_lock_irq(&zone->lru_lock);
1178                        }
1179                }
1180        } while (nr_scanned < max_scan);
1181        spin_unlock(&zone->lru_lock);
1182done:
1183        local_irq_enable();
1184        pagevec_release(&pvec);
1185        return nr_reclaimed;
1186}
1187
1188/*
1189 * We are about to scan this zone at a certain priority level.  If that priority
1190 * level is smaller (ie: more urgent) than the previous priority, then note
1191 * that priority level within the zone.  This is done so that when the next
1192 * process comes in to scan this zone, it will immediately start out at this
1193 * priority level rather than having to build up its own scanning priority.
1194 * Here, this priority affects only the reclaim-mapped threshold.
1195 */
1196static inline void note_zone_scanning_priority(struct zone *zone, int priority)
1197{
1198        if (priority < zone->prev_priority)
1199                zone->prev_priority = priority;
1200}
1201
1202/*
1203 * This moves pages from the active list to the inactive list.
1204 *
1205 * We move them the other way if the page is referenced by one or more
1206 * processes, from rmap.
1207 *
1208 * If the pages are mostly unmapped, the processing is fast and it is
1209 * appropriate to hold zone->lru_lock across the whole operation.  But if
1210 * the pages are mapped, the processing is slow (page_referenced()) so we
1211 * should drop zone->lru_lock around each page.  It's impossible to balance
1212 * this, so instead we remove the pages from the LRU while processing them.
1213 * It is safe to rely on PG_active against the non-LRU pages in here because
1214 * nobody will play with that bit on a non-LRU page.
1215 *
1216 * The downside is that we have to touch page->_count against each page.
1217 * But we had to alter page->flags anyway.
1218 */
1219
1220
1221static void shrink_active_list(unsigned long nr_pages, struct zone *zone,
1222                        struct scan_control *sc, int priority, int file)
1223{
1224        unsigned long pgmoved;
1225        int pgdeactivate = 0;
1226        unsigned long pgscanned;
1227        LIST_HEAD(l_hold);      /* The pages which were snipped off */
1228        LIST_HEAD(l_inactive);
1229        struct page *page;
1230        struct pagevec pvec;
1231        enum lru_list lru;
1232        struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc);
1233
1234        lru_add_drain();
1235        spin_lock_irq(&zone->lru_lock);
1236        pgmoved = sc->isolate_pages(nr_pages, &l_hold, &pgscanned, sc->order,
1237                                        ISOLATE_ACTIVE, zone,
1238                                        sc->mem_cgroup, 1, file);
1239        /*
1240         * zone->pages_scanned is used for detect zone's oom
1241         * mem_cgroup remembers nr_scan by itself.
1242         */
1243        if (scanning_global_lru(sc)) {
1244                zone->pages_scanned += pgscanned;
1245        }
1246        reclaim_stat->recent_scanned[!!file] += pgmoved;
1247
1248        if (file)
1249                __mod_zone_page_state(zone, NR_ACTIVE_FILE, -pgmoved);
1250        else
1251                __mod_zone_page_state(zone, NR_ACTIVE_ANON, -pgmoved);
1252        spin_unlock_irq(&zone->lru_lock);
1253
1254        pgmoved = 0;
1255        while (!list_empty(&l_hold)) {
1256                cond_resched();
1257                page = lru_to_page(&l_hold);
1258                list_del(&page->lru);
1259
1260                if (unlikely(!page_evictable(page, NULL))) {
1261                        putback_lru_page(page);
1262                        continue;
1263                }
1264
1265                /* page_referenced clears PageReferenced */
1266                if (page_mapping_inuse(page) &&
1267                    page_referenced(page, 0, sc->mem_cgroup))
1268                        pgmoved++;
1269
1270                list_add(&page->lru, &l_inactive);
1271        }
1272
1273        /*
1274         * Move the pages to the [file or anon] inactive list.
1275         */
1276        pagevec_init(&pvec, 1);
1277        lru = LRU_BASE + file * LRU_FILE;
1278
1279        spin_lock_irq(&zone->lru_lock);
1280        /*
1281         * Count referenced pages from currently used mappings as
1282         * rotated, even though they are moved to the inactive list.
1283         * This helps balance scan pressure between file and anonymous
1284         * pages in get_scan_ratio.
1285         */
1286        reclaim_stat->recent_rotated[!!file] += pgmoved;
1287
1288        pgmoved = 0;
1289        while (!list_empty(&l_inactive)) {
1290                page = lru_to_page(&l_inactive);
1291                prefetchw_prev_lru_page(page, &l_inactive, flags);
1292                VM_BUG_ON(PageLRU(page));
1293                SetPageLRU(page);
1294                VM_BUG_ON(!PageActive(page));
1295                ClearPageActive(page);
1296
1297                list_move(&page->lru, &zone->lru[lru].list);
1298                mem_cgroup_add_lru_list(page, lru);
1299                pgmoved++;
1300                if (!pagevec_add(&pvec, page)) {
1301                        __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved);
1302                        spin_unlock_irq(&zone->lru_lock);
1303                        pgdeactivate += pgmoved;
1304                        pgmoved = 0;
1305                        if (buffer_heads_over_limit)
1306                                pagevec_strip(&pvec);
1307                        __pagevec_release(&pvec);
1308                        spin_lock_irq(&zone->lru_lock);
1309                }
1310        }
1311        __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved);
1312        pgdeactivate += pgmoved;
1313        __count_zone_vm_events(PGREFILL, zone, pgscanned);
1314        __count_vm_events(PGDEACTIVATE, pgdeactivate);
1315        spin_unlock_irq(&zone->lru_lock);
1316        if (buffer_heads_over_limit)
1317                pagevec_strip(&pvec);
1318        pagevec_release(&pvec);
1319}
1320
1321static int inactive_anon_is_low_global(struct zone *zone)
1322{
1323        unsigned long active, inactive;
1324
1325        active = zone_page_state(zone, NR_ACTIVE_ANON);
1326        inactive = zone_page_state(zone, NR_INACTIVE_ANON);
1327
1328        if (inactive * zone->inactive_ratio < active)
1329                return 1;
1330
1331        return 0;
1332}
1333
1334/**
1335 * inactive_anon_is_low - check if anonymous pages need to be deactivated
1336 * @zone: zone to check
1337 * @sc:   scan control of this context
1338 *
1339 * Returns true if the zone does not have enough inactive anon pages,
1340 * meaning some active anon pages need to be deactivated.
1341 */
1342static int inactive_anon_is_low(struct zone *zone, struct scan_control *sc)
1343{
1344        int low;
1345
1346        if (scanning_global_lru(sc))
1347                low = inactive_anon_is_low_global(zone);
1348        else
1349                low = mem_cgroup_inactive_anon_is_low(sc->mem_cgroup);
1350        return low;
1351}
1352
1353static unsigned long shrink_list(enum lru_list lru, unsigned long nr_to_scan,
1354        struct zone *zone, struct scan_control *sc, int priority)
1355{
1356        int file = is_file_lru(lru);
1357
1358        if (lru == LRU_ACTIVE_FILE) {
1359                shrink_active_list(nr_to_scan, zone, sc, priority, file);
1360                return 0;
1361        }
1362
1363        if (lru == LRU_ACTIVE_ANON && inactive_anon_is_low(zone, sc)) {
1364                shrink_active_list(nr_to_scan, zone, sc, priority, file);
1365                return 0;
1366        }
1367        return shrink_inactive_list(nr_to_scan, zone, sc, priority, file);
1368}
1369
1370/*
1371 * Determine how aggressively the anon and file LRU lists should be
1372 * scanned.  The relative value of each set of LRU lists is determined
1373 * by looking at the fraction of the pages scanned we did rotate back
1374 * onto the active list instead of evict.
1375 *
1376 * percent[0] specifies how much pressure to put on ram/swap backed
1377 * memory, while percent[1] determines pressure on the file LRUs.
1378 */
1379static void get_scan_ratio(struct zone *zone, struct scan_control *sc,
1380                                        unsigned long *percent)
1381{
1382        unsigned long anon, file, free;
1383        unsigned long anon_prio, file_prio;
1384        unsigned long ap, fp;
1385        struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc);
1386
1387        /* If we have no swap space, do not bother scanning anon pages. */
1388        if (!sc->may_swap || (nr_swap_pages <= 0)) {
1389                percent[0] = 0;
1390                percent[1] = 100;
1391                return;
1392        }
1393
1394        anon  = zone_nr_pages(zone, sc, LRU_ACTIVE_ANON) +
1395                zone_nr_pages(zone, sc, LRU_INACTIVE_ANON);
1396        file  = zone_nr_pages(zone, sc, LRU_ACTIVE_FILE) +
1397                zone_nr_pages(zone, sc, LRU_INACTIVE_FILE);
1398
1399        if (scanning_global_lru(sc)) {
1400                free  = zone_page_state(zone, NR_FREE_PAGES);
1401                /* If we have very few page cache pages,
1402                   force-scan anon pages. */
1403                if (unlikely(file + free <= zone->pages_high)) {
1404                        percent[0] = 100;
1405                        percent[1] = 0;
1406                        return;
1407                }
1408        }
1409
1410        /*
1411         * OK, so we have swap space and a fair amount of page cache
1412         * pages.  We use the recently rotated / recently scanned
1413         * ratios to determine how valuable each cache is.
1414         *
1415         * Because workloads change over time (and to avoid overflow)
1416         * we keep these statistics as a floating average, which ends
1417         * up weighing recent references more than old ones.
1418         *
1419         * anon in [0], file in [1]
1420         */
1421        if (unlikely(reclaim_stat->recent_scanned[0] > anon / 4)) {
1422                spin_lock_irq(&zone->lru_lock);
1423                reclaim_stat->recent_scanned[0] /= 2;
1424                reclaim_stat->recent_rotated[0] /= 2;
1425                spin_unlock_irq(&zone->lru_lock);
1426        }
1427
1428        if (unlikely(reclaim_stat->recent_scanned[1] > file / 4)) {
1429                spin_lock_irq(&zone->lru_lock);
1430                reclaim_stat->recent_scanned[1] /= 2;
1431                reclaim_stat->recent_rotated[1] /= 2;
1432                spin_unlock_irq(&zone->lru_lock);
1433        }
1434
1435        /*
1436         * With swappiness at 100, anonymous and file have the same priority.
1437         * This scanning priority is essentially the inverse of IO cost.
1438         */
1439        anon_prio = sc->swappiness;
1440        file_prio = 200 - sc->swappiness;
1441
1442        /*
1443         * The amount of pressure on anon vs file pages is inversely
1444         * proportional to the fraction of recently scanned pages on
1445         * each list that were recently referenced and in active use.
1446         */
1447        ap = (anon_prio + 1) * (reclaim_stat->recent_scanned[0] + 1);
1448        ap /= reclaim_stat->recent_rotated[0] + 1;
1449
1450        fp = (file_prio + 1) * (reclaim_stat->recent_scanned[1] + 1);
1451        fp /= reclaim_stat->recent_rotated[1] + 1;
1452
1453        /* Normalize to percentages */
1454        percent[0] = 100 * ap / (ap + fp + 1);
1455        percent[1] = 100 - percent[0];
1456}
1457
1458
1459/*
1460 * This is a basic per-zone page freer.  Used by both kswapd and direct reclaim.
1461 */
1462static void shrink_zone(int priority, struct zone *zone,
1463                                struct scan_control *sc)
1464{
1465        unsigned long nr[NR_LRU_LISTS];
1466        unsigned long nr_to_scan;
1467        unsigned long percent[2];       /* anon @ 0; file @ 1 */
1468        enum lru_list l;
1469        unsigned long nr_reclaimed = sc->nr_reclaimed;
1470        unsigned long swap_cluster_max = sc->swap_cluster_max;
1471
1472        get_scan_ratio(zone, sc, percent);
1473
1474        for_each_evictable_lru(l) {
1475                int file = is_file_lru(l);
1476                unsigned long scan;
1477
1478                scan = zone_nr_pages(zone, sc, l);
1479                if (priority) {
1480                        scan >>= priority;
1481                        scan = (scan * percent[file]) / 100;
1482                }
1483                if (scanning_global_lru(sc)) {
1484                        zone->lru[l].nr_scan += scan;
1485                        nr[l] = zone->lru[l].nr_scan;
1486                        if (nr[l] >= swap_cluster_max)
1487                                zone->lru[l].nr_scan = 0;
1488                        else
1489                                nr[l] = 0;
1490                } else
1491                        nr[l] = scan;
1492        }
1493
1494        while (nr[LRU_INACTIVE_ANON] || nr[LRU_ACTIVE_FILE] ||
1495                                        nr[LRU_INACTIVE_FILE]) {
1496                for_each_evictable_lru(l) {
1497                        if (nr[l]) {
1498                                nr_to_scan = min(nr[l], swap_cluster_max);
1499                                nr[l] -= nr_to_scan;
1500
1501                                nr_reclaimed += shrink_list(l, nr_to_scan,
1502                                                            zone, sc, priority);
1503                        }
1504                }
1505                /*
1506                 * On large memory systems, scan >> priority can become
1507                 * really large. This is fine for the starting priority;
1508                 * we want to put equal scanning pressure on each zone.
1509                 * However, if the VM has a harder time of freeing pages,
1510                 * with multiple processes reclaiming pages, the total
1511                 * freeing target can get unreasonably large.
1512                 */
1513                if (nr_reclaimed > swap_cluster_max &&
1514                        priority < DEF_PRIORITY && !current_is_kswapd())
1515                        break;
1516        }
1517
1518        sc->nr_reclaimed = nr_reclaimed;
1519
1520        /*
1521         * Even if we did not try to evict anon pages at all, we want to
1522         * rebalance the anon lru active/inactive ratio.
1523         */
1524        if (inactive_anon_is_low(zone, sc))
1525                shrink_active_list(SWAP_CLUSTER_MAX, zone, sc, priority, 0);
1526
1527        throttle_vm_writeout(sc->gfp_mask);
1528}
1529
1530/*
1531 * This is the direct reclaim path, for page-allocating processes.  We only
1532 * try to reclaim pages from zones which will satisfy the caller's allocation
1533 * request.
1534 *
1535 * We reclaim from a zone even if that zone is over pages_high.  Because:
1536 * a) The caller may be trying to free *extra* pages to satisfy a higher-order
1537 *    allocation or
1538 * b) The zones may be over pages_high but they must go *over* pages_high to
1539 *    satisfy the `incremental min' zone defense algorithm.
1540 *
1541 * If a zone is deemed to be full of pinned pages then just give it a light
1542 * scan then give up on it.
1543 */
1544static void shrink_zones(int priority, struct zonelist *zonelist,
1545                                        struct scan_control *sc)
1546{
1547        enum zone_type high_zoneidx = gfp_zone(sc->gfp_mask);
1548        struct zoneref *z;
1549        struct zone *zone;
1550
1551        sc->all_unreclaimable = 1;
1552        for_each_zone_zonelist_nodemask(zone, z, zonelist, high_zoneidx,
1553                                        sc->nodemask) {
1554                if (!populated_zone(zone))
1555                        continue;
1556                /*
1557                 * Take care memory controller reclaiming has small influence
1558                 * to global LRU.
1559                 */
1560                if (scanning_global_lru(sc)) {
1561                        if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
1562                                continue;
1563                        note_zone_scanning_priority(zone, priority);
1564
1565                        if (zone_is_all_unreclaimable(zone) &&
1566                                                priority != DEF_PRIORITY)
1567                                continue;       /* Let kswapd poll it */
1568                        sc->all_unreclaimable = 0;
1569                } else {
1570                        /*
1571                         * Ignore cpuset limitation here. We just want to reduce
1572                         * # of used pages by us regardless of memory shortage.
1573                         */
1574                        sc->all_unreclaimable = 0;
1575                        mem_cgroup_note_reclaim_priority(sc->mem_cgroup,
1576                                                        priority);
1577                }
1578
1579                shrink_zone(priority, zone, sc);
1580        }
1581}
1582
1583/*
1584 * This is the main entry point to direct page reclaim.
1585 *
1586 * If a full scan of the inactive list fails to free enough memory then we
1587 * are "out of memory" and something needs to be killed.
1588 *
1589 * If the caller is !__GFP_FS then the probability of a failure is reasonably
1590 * high - the zone may be full of dirty or under-writeback pages, which this
1591 * caller can't do much about.  We kick pdflush and take explicit naps in the
1592 * hope that some of these pages can be written.  But if the allocating task
1593 * holds filesystem locks which prevent writeout this might not work, and the
1594 * allocation attempt will fail.
1595 *
1596 * returns:     0, if no pages reclaimed
1597 *              else, the number of pages reclaimed
1598 */
1599static unsigned long do_try_to_free_pages(struct zonelist *zonelist,
1600                                        struct scan_control *sc)
1601{
1602        int priority;
1603        unsigned long ret = 0;
1604        unsigned long total_scanned = 0;
1605        struct reclaim_state *reclaim_state = current->reclaim_state;
1606        unsigned long lru_pages = 0;
1607        struct zoneref *z;
1608        struct zone *zone;
1609        enum zone_type high_zoneidx = gfp_zone(sc->gfp_mask);
1610
1611        delayacct_freepages_start();
1612
1613        if (scanning_global_lru(sc))
1614                count_vm_event(ALLOCSTALL);
1615        /*
1616         * mem_cgroup will not do shrink_slab.
1617         */
1618        if (scanning_global_lru(sc)) {
1619                for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
1620
1621                        if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
1622                                continue;
1623
1624                        lru_pages += zone_lru_pages(zone);
1625                }
1626        }
1627
1628        for (priority = DEF_PRIORITY; priority >= 0; priority--) {
1629                sc->nr_scanned = 0;
1630                if (!priority)
1631                        disable_swap_token();
1632                shrink_zones(priority, zonelist, sc);
1633                /*
1634                 * Don't shrink slabs when reclaiming memory from
1635                 * over limit cgroups
1636                 */
1637                if (scanning_global_lru(sc)) {
1638                        shrink_slab(sc->nr_scanned, sc->gfp_mask, lru_pages);
1639                        if (reclaim_state) {
1640                                sc->nr_reclaimed += reclaim_state->reclaimed_slab;
1641                                reclaim_state->reclaimed_slab = 0;
1642                        }
1643                }
1644                total_scanned += sc->nr_scanned;
1645                if (sc->nr_reclaimed >= sc->swap_cluster_max) {
1646                        ret = sc->nr_reclaimed;
1647                        goto out;
1648                }
1649
1650                /*
1651                 * Try to write back as many pages as we just scanned.  This
1652                 * tends to cause slow streaming writers to write data to the
1653                 * disk smoothly, at the dirtying rate, which is nice.   But
1654                 * that's undesirable in laptop mode, where we *want* lumpy
1655                 * writeout.  So in laptop mode, write out the whole world.
1656                 */
1657                if (total_scanned > sc->swap_cluster_max +
1658                                        sc->swap_cluster_max / 2) {
1659                        wakeup_pdflush(laptop_mode ? 0 : total_scanned);
1660                        sc->may_writepage = 1;
1661                }
1662
1663                /* Take a nap, wait for some writeback to complete */
1664                if (sc->nr_scanned && priority < DEF_PRIORITY - 2)
1665                        congestion_wait(WRITE, HZ/10);
1666        }
1667        /* top priority shrink_zones still had more to do? don't OOM, then */
1668        if (!sc->all_unreclaimable && scanning_global_lru(sc))
1669                ret = sc->nr_reclaimed;
1670out:
1671        /*
1672         * Now that we've scanned all the zones at this priority level, note
1673         * that level within the zone so that the next thread which performs
1674         * scanning of this zone will immediately start out at this priority
1675         * level.  This affects only the decision whether or not to bring
1676         * mapped pages onto the inactive list.
1677         */
1678        if (priority < 0)
1679                priority = 0;
1680
1681        if (scanning_global_lru(sc)) {
1682                for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
1683
1684                        if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
1685                                continue;
1686
1687                        zone->prev_priority = priority;
1688                }
1689        } else
1690                mem_cgroup_record_reclaim_priority(sc->mem_cgroup, priority);
1691
1692        delayacct_freepages_end();
1693
1694        return ret;
1695}
1696
1697unsigned long try_to_free_pages(struct zonelist *zonelist, int order,
1698                                gfp_t gfp_mask, nodemask_t *nodemask)
1699{
1700        struct scan_control sc = {
1701                .gfp_mask = gfp_mask,
1702                .may_writepage = !laptop_mode,
1703                .swap_cluster_max = SWAP_CLUSTER_MAX,
1704                .may_unmap = 1,
1705                .may_swap = 1,
1706                .swappiness = vm_swappiness,
1707                .order = order,
1708                .mem_cgroup = NULL,
1709                .isolate_pages = isolate_pages_global,
1710                .nodemask = nodemask,
1711        };
1712
1713        return do_try_to_free_pages(zonelist, &sc);
1714}
1715
1716#ifdef CONFIG_CGROUP_MEM_RES_CTLR
1717
1718unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *mem_cont,
1719                                           gfp_t gfp_mask,
1720                                           bool noswap,
1721                                           unsigned int swappiness)
1722{
1723        struct scan_control sc = {
1724                .may_writepage = !laptop_mode,
1725                .may_unmap = 1,
1726                .may_swap = !noswap,
1727                .swap_cluster_max = SWAP_CLUSTER_MAX,
1728                .swappiness = swappiness,
1729                .order = 0,
1730                .mem_cgroup = mem_cont,
1731                .isolate_pages = mem_cgroup_isolate_pages,
1732                .nodemask = NULL, /* we don't care the placement */
1733        };
1734        struct zonelist *zonelist;
1735
1736        sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) |
1737                        (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK);
1738        zonelist = NODE_DATA(numa_node_id())->node_zonelists;
1739        return do_try_to_free_pages(zonelist, &sc);
1740}
1741#endif
1742
1743/*
1744 * For kswapd, balance_pgdat() will work across all this node's zones until
1745 * they are all at pages_high.
1746 *
1747 * Returns the number of pages which were actually freed.
1748 *
1749 * There is special handling here for zones which are full of pinned pages.
1750 * This can happen if the pages are all mlocked, or if they are all used by
1751 * device drivers (say, ZONE_DMA).  Or if they are all in use by hugetlb.
1752 * What we do is to detect the case where all pages in the zone have been
1753 * scanned twice and there has been zero successful reclaim.  Mark the zone as
1754 * dead and from now on, only perform a short scan.  Basically we're polling
1755 * the zone for when the problem goes away.
1756 *
1757 * kswapd scans the zones in the highmem->normal->dma direction.  It skips
1758 * zones which have free_pages > pages_high, but once a zone is found to have
1759 * free_pages <= pages_high, we scan that zone and the lower zones regardless
1760 * of the number of free pages in the lower zones.  This interoperates with
1761 * the page allocator fallback scheme to ensure that aging of pages is balanced
1762 * across the zones.
1763 */
1764static unsigned long balance_pgdat(pg_data_t *pgdat, int order)
1765{
1766        int all_zones_ok;
1767        int priority;
1768        int i;
1769        unsigned long total_scanned;
1770        struct reclaim_state *reclaim_state = current->reclaim_state;
1771        struct scan_control sc = {
1772                .gfp_mask = GFP_KERNEL,
1773                .may_unmap = 1,
1774                .may_swap = 1,
1775                .swap_cluster_max = SWAP_CLUSTER_MAX,
1776                .swappiness = vm_swappiness,
1777                .order = order,
1778                .mem_cgroup = NULL,
1779                .isolate_pages = isolate_pages_global,
1780        };
1781        /*
1782         * temp_priority is used to remember the scanning priority at which
1783         * this zone was successfully refilled to free_pages == pages_high.
1784         */
1785        int temp_priority[MAX_NR_ZONES];
1786
1787loop_again:
1788        total_scanned = 0;
1789        sc.nr_reclaimed = 0;
1790        sc.may_writepage = !laptop_mode;
1791        count_vm_event(PAGEOUTRUN);
1792
1793        for (i = 0; i < pgdat->nr_zones; i++)
1794                temp_priority[i] = DEF_PRIORITY;
1795
1796        for (priority = DEF_PRIORITY; priority >= 0; priority--) {
1797                int end_zone = 0;       /* Inclusive.  0 = ZONE_DMA */
1798                unsigned long lru_pages = 0;
1799
1800                /* The swap token gets in the way of swapout... */
1801                if (!priority)
1802                        disable_swap_token();
1803
1804                all_zones_ok = 1;
1805
1806                /*
1807                 * Scan in the highmem->dma direction for the highest
1808                 * zone which needs scanning
1809                 */
1810                for (i = pgdat->nr_zones - 1; i >= 0; i--) {
1811                        struct zone *zone = pgdat->node_zones + i;
1812
1813                        if (!populated_zone(zone))
1814                                continue;
1815
1816                        if (zone_is_all_unreclaimable(zone) &&
1817                            priority != DEF_PRIORITY)
1818                                continue;
1819
1820                        /*
1821                         * Do some background aging of the anon list, to give
1822                         * pages a chance to be referenced before reclaiming.
1823                         */
1824                        if (inactive_anon_is_low(zone, &sc))
1825                                shrink_active_list(SWAP_CLUSTER_MAX, zone,
1826                                                        &sc, priority, 0);
1827
1828                        if (!zone_watermark_ok(zone, order, zone->pages_high,
1829                                               0, 0)) {
1830                                end_zone = i;
1831                                break;
1832                        }
1833                }
1834                if (i < 0)
1835                        goto out;
1836
1837                for (i = 0; i <= end_zone; i++) {
1838                        struct zone *zone = pgdat->node_zones + i;
1839
1840                        lru_pages += zone_lru_pages(zone);
1841                }
1842
1843                /*
1844                 * Now scan the zone in the dma->highmem direction, stopping
1845                 * at the last zone which needs scanning.
1846                 *
1847                 * We do this because the page allocator works in the opposite
1848                 * direction.  This prevents the page allocator from allocating
1849                 * pages behind kswapd's direction of progress, which would
1850                 * cause too much scanning of the lower zones.
1851                 */
1852                for (i = 0; i <= end_zone; i++) {
1853                        struct zone *zone = pgdat->node_zones + i;
1854                        int nr_slab;
1855
1856                        if (!populated_zone(zone))
1857                                continue;
1858
1859                        if (zone_is_all_unreclaimable(zone) &&
1860                                        priority != DEF_PRIORITY)
1861                                continue;
1862
1863                        if (!zone_watermark_ok(zone, order, zone->pages_high,
1864                                               end_zone, 0))
1865                                all_zones_ok = 0;
1866                        temp_priority[i] = priority;
1867                        sc.nr_scanned = 0;
1868                        note_zone_scanning_priority(zone, priority);
1869                        /*
1870                         * We put equal pressure on every zone, unless one
1871                         * zone has way too many pages free already.
1872                         */
1873                        if (!zone_watermark_ok(zone, order, 8*zone->pages_high,
1874                                                end_zone, 0))
1875                                shrink_zone(priority, zone, &sc);
1876                        reclaim_state->reclaimed_slab = 0;
1877                        nr_slab = shrink_slab(sc.nr_scanned, GFP_KERNEL,
1878                                                lru_pages);
1879                        sc.nr_reclaimed += reclaim_state->reclaimed_slab;
1880                        total_scanned += sc.nr_scanned;
1881                        if (zone_is_all_unreclaimable(zone))
1882                                continue;
1883                        if (nr_slab == 0 && zone->pages_scanned >=
1884                                                (zone_lru_pages(zone) * 6))
1885                                        zone_set_flag(zone,
1886                                                      ZONE_ALL_UNRECLAIMABLE);
1887                        /*
1888                         * If we've done a decent amount of scanning and
1889                         * the reclaim ratio is low, start doing writepage
1890                         * even in laptop mode
1891                         */
1892                        if (total_scanned > SWAP_CLUSTER_MAX * 2 &&
1893                            total_scanned > sc.nr_reclaimed + sc.nr_reclaimed / 2)
1894                                sc.may_writepage = 1;
1895                }
1896                if (all_zones_ok)
1897                        break;          /* kswapd: all done */
1898                /*
1899                 * OK, kswapd is getting into trouble.  Take a nap, then take
1900                 * another pass across the zones.
1901                 */
1902                if (total_scanned && priority < DEF_PRIORITY - 2)
1903                        congestion_wait(WRITE, HZ/10);
1904
1905                /*
1906                 * We do this so kswapd doesn't build up large priorities for
1907                 * example when it is freeing in parallel with allocators. It
1908                 * matches the direct reclaim path behaviour in terms of impact
1909                 * on zone->*_priority.
1910                 */
1911                if (sc.nr_reclaimed >= SWAP_CLUSTER_MAX)
1912                        break;
1913        }
1914out:
1915        /*
1916         * Note within each zone the priority level at which this zone was
1917         * brought into a happy state.  So that the next thread which scans this
1918         * zone will start out at that priority level.
1919         */
1920        for (i = 0; i < pgdat->nr_zones; i++) {
1921                struct zone *zone = pgdat->node_zones + i;
1922
1923                zone->prev_priority = temp_priority[i];
1924        }
1925        if (!all_zones_ok) {
1926                cond_resched();
1927
1928                try_to_freeze();
1929
1930                /*
1931                 * Fragmentation may mean that the system cannot be
1932                 * rebalanced for high-order allocations in all zones.
1933                 * At this point, if nr_reclaimed < SWAP_CLUSTER_MAX,
1934                 * it means the zones have been fully scanned and are still
1935                 * not balanced. For high-order allocations, there is
1936                 * little point trying all over again as kswapd may
1937                 * infinite loop.
1938                 *
1939                 * Instead, recheck all watermarks at order-0 as they
1940                 * are the most important. If watermarks are ok, kswapd will go
1941                 * back to sleep. High-order users can still perform direct
1942                 * reclaim if they wish.
1943                 */
1944                if (sc.nr_reclaimed < SWAP_CLUSTER_MAX)
1945                        order = sc.order = 0;
1946
1947                goto loop_again;
1948        }
1949
1950        return sc.nr_reclaimed;
1951}
1952
1953/*
1954 * The background pageout daemon, started as a kernel thread
1955 * from the init process.
1956 *
1957 * This basically trickles out pages so that we have _some_
1958 * free memory available even if there is no other activity
1959 * that frees anything up. This is needed for things like routing
1960 * etc, where we otherwise might have all activity going on in
1961 * asynchronous contexts that cannot page things out.
1962 *
1963 * If there are applications that are active memory-allocators
1964 * (most normal use), this basically shouldn't matter.
1965 */
1966static int kswapd(void *p)
1967{
1968        unsigned long order;
1969        pg_data_t *pgdat = (pg_data_t*)p;
1970        struct task_struct *tsk = current;
1971        DEFINE_WAIT(wait);
1972        struct reclaim_state reclaim_state = {
1973                .reclaimed_slab = 0,
1974        };
1975        const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
1976
1977        lockdep_set_current_reclaim_state(GFP_KERNEL);
1978
1979        if (!cpumask_empty(cpumask))
1980                set_cpus_allowed_ptr(tsk, cpumask);
1981        current->reclaim_state = &reclaim_state;
1982
1983        /*
1984         * Tell the memory management that we're a "memory allocator",
1985         * and that if we need more memory we should get access to it
1986         * regardless (see "__alloc_pages()"). "kswapd" should
1987         * never get caught in the normal page freeing logic.
1988         *
1989         * (Kswapd normally doesn't need memory anyway, but sometimes
1990         * you need a small amount of memory in order to be able to
1991         * page out something else, and this flag essentially protects
1992         * us from recursively trying to free more memory as we're
1993         * trying to free the first piece of memory in the first place).
1994         */
1995        tsk->flags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD;
1996        set_freezable();
1997
1998        order = 0;
1999        for ( ; ; ) {
2000                unsigned long new_order;
2001
2002                prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE);
2003                new_order = pgdat->kswapd_max_order;
2004                pgdat->kswapd_max_order = 0;
2005                if (order < new_order) {
2006                        /*
2007                         * Don't sleep if someone wants a larger 'order'
2008                         * allocation
2009                         */
2010                        order = new_order;
2011                } else {
2012                        if (!freezing(current))
2013                                schedule();
2014
2015                        order = pgdat->kswapd_max_order;
2016                }
2017                finish_wait(&pgdat->kswapd_wait, &wait);
2018
2019                if (!try_to_freeze()) {
2020                        /* We can speed up thawing tasks if we don't call
2021                         * balance_pgdat after returning from the refrigerator
2022                         */
2023                        balance_pgdat(pgdat, order);
2024                }
2025        }
2026        return 0;
2027}
2028
2029/*
2030 * A zone is low on free memory, so wake its kswapd task to service it.
2031 */
2032void wakeup_kswapd(struct zone *zone, int order)
2033{
2034        pg_data_t *pgdat;
2035
2036        if (!populated_zone(zone))
2037                return;
2038
2039        pgdat = zone->zone_pgdat;
2040        if (zone_watermark_ok(zone, order, zone->pages_low, 0, 0))
2041                return;
2042        if (pgdat->kswapd_max_order < order)
2043                pgdat->kswapd_max_order = order;
2044        if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
2045                return;
2046        if (!waitqueue_active(&pgdat->kswapd_wait))
2047                return;
2048        wake_up_interruptible(&pgdat->kswapd_wait);
2049}
2050
2051unsigned long global_lru_pages(void)
2052{
2053        return global_page_state(NR_ACTIVE_ANON)
2054                + global_page_state(NR_ACTIVE_FILE)
2055                + global_page_state(NR_INACTIVE_ANON)
2056                + global_page_state(NR_INACTIVE_FILE);
2057}
2058
2059#ifdef CONFIG_PM
2060/*
2061 * Helper function for shrink_all_memory().  Tries to reclaim 'nr_pages' pages
2062 * from LRU lists system-wide, for given pass and priority.
2063 *
2064 * For pass > 3 we also try to shrink the LRU lists that contain a few pages
2065 */
2066static void shrink_all_zones(unsigned long nr_pages, int prio,
2067                                      int pass, struct scan_control *sc)
2068{
2069        struct zone *zone;
2070        unsigned long nr_reclaimed = 0;
2071
2072        for_each_populated_zone(zone) {
2073                enum lru_list l;
2074
2075                if (zone_is_all_unreclaimable(zone) && prio != DEF_PRIORITY)
2076                        continue;
2077
2078                for_each_evictable_lru(l) {
2079                        enum zone_stat_item ls = NR_LRU_BASE + l;
2080                        unsigned long lru_pages = zone_page_state(zone, ls);
2081
2082                        /* For pass = 0, we don't shrink the active list */
2083                        if (pass == 0 && (l == LRU_ACTIVE_ANON ||
2084                                                l == LRU_ACTIVE_FILE))
2085                                continue;
2086
2087                        zone->lru[l].nr_scan += (lru_pages >> prio) + 1;
2088                        if (zone->lru[l].nr_scan >= nr_pages || pass > 3) {
2089                                unsigned long nr_to_scan;
2090
2091                                zone->lru[l].nr_scan = 0;
2092                                nr_to_scan = min(nr_pages, lru_pages);
2093                                nr_reclaimed += shrink_list(l, nr_to_scan, zone,
2094                                                                sc, prio);
2095                                if (nr_reclaimed >= nr_pages) {
2096                                        sc->nr_reclaimed += nr_reclaimed;
2097                                        return;
2098                                }
2099                        }
2100                }
2101        }
2102        sc->nr_reclaimed += nr_reclaimed;
2103}
2104
2105/*
2106 * Try to free `nr_pages' of memory, system-wide, and return the number of
2107 * freed pages.
2108 *
2109 * Rather than trying to age LRUs the aim is to preserve the overall
2110 * LRU order by reclaiming preferentially
2111 * inactive > active > active referenced > active mapped
2112 */
2113unsigned long shrink_all_memory(unsigned long nr_pages)
2114{
2115        unsigned long lru_pages, nr_slab;
2116        int pass;
2117        struct reclaim_state reclaim_state;
2118        struct scan_control sc = {
2119                .gfp_mask = GFP_KERNEL,
2120                .may_unmap = 0,
2121                .may_writepage = 1,
2122                .isolate_pages = isolate_pages_global,
2123                .nr_reclaimed = 0,
2124        };
2125
2126        current->reclaim_state = &reclaim_state;
2127
2128        lru_pages = global_lru_pages();
2129        nr_slab = global_page_state(NR_SLAB_RECLAIMABLE);
2130        /* If slab caches are huge, it's better to hit them first */
2131        while (nr_slab >= lru_pages) {
2132                reclaim_state.reclaimed_slab = 0;
2133                shrink_slab(nr_pages, sc.gfp_mask, lru_pages);
2134                if (!reclaim_state.reclaimed_slab)
2135                        break;
2136
2137                sc.nr_reclaimed += reclaim_state.reclaimed_slab;
2138                if (sc.nr_reclaimed >= nr_pages)
2139                        goto out;
2140
2141                nr_slab -= reclaim_state.reclaimed_slab;
2142        }
2143
2144        /*
2145         * We try to shrink LRUs in 5 passes:
2146         * 0 = Reclaim from inactive_list only
2147         * 1 = Reclaim from active list but don't reclaim mapped
2148         * 2 = 2nd pass of type 1
2149         * 3 = Reclaim mapped (normal reclaim)
2150         * 4 = 2nd pass of type 3
2151         */
2152        for (pass = 0; pass < 5; pass++) {
2153                int prio;
2154
2155                /* Force reclaiming mapped pages in the passes #3 and #4 */
2156                if (pass > 2)
2157                        sc.may_unmap = 1;
2158
2159                for (prio = DEF_PRIORITY; prio >= 0; prio--) {
2160                        unsigned long nr_to_scan = nr_pages - sc.nr_reclaimed;
2161
2162                        sc.nr_scanned = 0;
2163                        sc.swap_cluster_max = nr_to_scan;
2164                        shrink_all_zones(nr_to_scan, prio, pass, &sc);
2165                        if (sc.nr_reclaimed >= nr_pages)
2166                                goto out;
2167
2168                        reclaim_state.reclaimed_slab = 0;
2169                        shrink_slab(sc.nr_scanned, sc.gfp_mask,
2170                                        global_lru_pages());
2171                        sc.nr_reclaimed += reclaim_state.reclaimed_slab;
2172                        if (sc.nr_reclaimed >= nr_pages)
2173                                goto out;
2174
2175                        if (sc.nr_scanned && prio < DEF_PRIORITY - 2)
2176                                congestion_wait(WRITE, HZ / 10);
2177                }
2178        }
2179
2180        /*
2181         * If sc.nr_reclaimed = 0, we could not shrink LRUs, but there may be
2182         * something in slab caches
2183         */
2184        if (!sc.nr_reclaimed) {
2185                do {
2186                        reclaim_state.reclaimed_slab = 0;
2187                        shrink_slab(nr_pages, sc.gfp_mask, global_lru_pages());
2188                        sc.nr_reclaimed += reclaim_state.reclaimed_slab;
2189                } while (sc.nr_reclaimed < nr_pages &&
2190                                reclaim_state.reclaimed_slab > 0);
2191        }
2192
2193
2194out:
2195        current->reclaim_state = NULL;
2196
2197        return sc.nr_reclaimed;
2198}
2199#endif
2200
2201/* It's optimal to keep kswapds on the same CPUs as their memory, but
2202   not required for correctness.  So if the last cpu in a node goes
2203   away, we get changed to run anywhere: as the first one comes back,
2204   restore their cpu bindings. */
2205static int __devinit cpu_callback(struct notifier_block *nfb,
2206                                  unsigned long action, void *hcpu)
2207{
2208        int nid;
2209
2210        if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN) {
2211                for_each_node_state(nid, N_HIGH_MEMORY) {
2212                        pg_data_t *pgdat = NODE_DATA(nid);
2213                        const struct cpumask *mask;
2214
2215                        mask = cpumask_of_node(pgdat->node_id);
2216
2217                        if (cpumask_any_and(cpu_online_mask, mask) < nr_cpu_ids)
2218                                /* One of our CPUs online: restore mask */
2219                                set_cpus_allowed_ptr(pgdat->kswapd, mask);
2220                }
2221        }
2222        return NOTIFY_OK;
2223}
2224
2225/*
2226 * This kswapd start function will be called by init and node-hot-add.
2227 * On node-hot-add, kswapd will moved to proper cpus if cpus are hot-added.
2228 */
2229int kswapd_run(int nid)
2230{
2231        pg_data_t *pgdat = NODE_DATA(nid);
2232        int ret = 0;
2233
2234        if (pgdat->kswapd)
2235                return 0;
2236
2237        pgdat->kswapd = kthread_run(kswapd, pgdat, "kswapd%d", nid);
2238        if (IS_ERR(pgdat->kswapd)) {
2239                /* failure at boot is fatal */
2240                BUG_ON(system_state == SYSTEM_BOOTING);
2241                printk("Failed to start kswapd on node %d\n",nid);
2242                ret = -1;
2243        }
2244        return ret;
2245}
2246
2247static int __init kswapd_init(void)
2248{
2249        int nid;
2250
2251        swap_setup();
2252        for_each_node_state(nid, N_HIGH_MEMORY)
2253                kswapd_run(nid);
2254        hotcpu_notifier(cpu_callback, 0);
2255        return 0;
2256}
2257
2258module_init(kswapd_init)
2259
2260#ifdef CONFIG_NUMA
2261/*
2262 * Zone reclaim mode
2263 *
2264 * If non-zero call zone_reclaim when the number of free pages falls below
2265 * the watermarks.
2266 */
2267int zone_reclaim_mode __read_mostly;
2268
2269#define RECLAIM_OFF 0
2270#define RECLAIM_ZONE (1<<0)     /* Run shrink_inactive_list on the zone */
2271#define RECLAIM_WRITE (1<<1)    /* Writeout pages during reclaim */
2272#define RECLAIM_SWAP (1<<2)     /* Swap pages out during reclaim */
2273
2274/*
2275 * Priority for ZONE_RECLAIM. This determines the fraction of pages
2276 * of a node considered for each zone_reclaim. 4 scans 1/16th of
2277 * a zone.
2278 */
2279#define ZONE_RECLAIM_PRIORITY 4
2280
2281/*
2282 * Percentage of pages in a zone that must be unmapped for zone_reclaim to
2283 * occur.
2284 */
2285int sysctl_min_unmapped_ratio = 1;
2286
2287/*
2288 * If the number of slab pages in a zone grows beyond this percentage then
2289 * slab reclaim needs to occur.
2290 */
2291int sysctl_min_slab_ratio = 5;
2292
2293static inline unsigned long zone_unmapped_file_pages(struct zone *zone)
2294{
2295        unsigned long file_mapped = zone_page_state(zone, NR_FILE_MAPPED);
2296        unsigned long file_lru = zone_page_state(zone, NR_INACTIVE_FILE) +
2297                zone_page_state(zone, NR_ACTIVE_FILE);
2298
2299        /*
2300         * It's possible for there to be more file mapped pages than
2301         * accounted for by the pages on the file LRU lists because
2302         * tmpfs pages accounted for as ANON can also be FILE_MAPPED
2303         */
2304        return (file_lru > file_mapped) ? (file_lru - file_mapped) : 0;
2305}
2306
2307/* Work out how many page cache pages we can reclaim in this reclaim_mode */
2308static long zone_pagecache_reclaimable(struct zone *zone)
2309{
2310        long nr_pagecache_reclaimable;
2311        long delta = 0;
2312
2313        /*
2314         * If RECLAIM_SWAP is set, then all file pages are considered
2315         * potentially reclaimable. Otherwise, we have to worry about
2316         * pages like swapcache and zone_unmapped_file_pages() provides
2317         * a better estimate
2318         */
2319        if (zone_reclaim_mode & RECLAIM_SWAP)
2320                nr_pagecache_reclaimable = zone_page_state(zone, NR_FILE_PAGES);
2321        else
2322                nr_pagecache_reclaimable = zone_unmapped_file_pages(zone);
2323
2324        /* If we can't clean pages, remove dirty pages from consideration */
2325        if (!(zone_reclaim_mode & RECLAIM_WRITE))
2326                delta += zone_page_state(zone, NR_FILE_DIRTY);
2327
2328        /* Watch for any possible underflows due to delta */
2329        if (unlikely(delta > nr_pagecache_reclaimable))
2330                delta = nr_pagecache_reclaimable;
2331
2332        return nr_pagecache_reclaimable - delta;
2333}
2334
2335/*
2336 * Try to free up some pages from this zone through reclaim.
2337 */
2338static int __zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order)
2339{
2340        /* Minimum pages needed in order to stay on node */
2341        const unsigned long nr_pages = 1 << order;
2342        struct task_struct *p = current;
2343        struct reclaim_state reclaim_state;
2344        int priority;
2345        struct scan_control sc = {
2346                .may_writepage = !!(zone_reclaim_mode & RECLAIM_WRITE),
2347                .may_unmap = !!(zone_reclaim_mode & RECLAIM_SWAP),
2348                .may_swap = 1,
2349                .swap_cluster_max = max_t(unsigned long, nr_pages,
2350                                        SWAP_CLUSTER_MAX),
2351                .gfp_mask = gfp_mask,
2352                .swappiness = vm_swappiness,
2353                .order = order,
2354                .isolate_pages = isolate_pages_global,
2355        };
2356        unsigned long slab_reclaimable;
2357
2358        disable_swap_token();
2359        cond_resched();
2360        /*
2361         * We need to be able to allocate from the reserves for RECLAIM_SWAP
2362         * and we also need to be able to write out pages for RECLAIM_WRITE
2363         * and RECLAIM_SWAP.
2364         */
2365        p->flags |= PF_MEMALLOC | PF_SWAPWRITE;
2366        reclaim_state.reclaimed_slab = 0;
2367        p->reclaim_state = &reclaim_state;
2368
2369        if (zone_pagecache_reclaimable(zone) > zone->min_unmapped_pages) {
2370                /*
2371                 * Free memory by calling shrink zone with increasing
2372                 * priorities until we have enough memory freed.
2373                 */
2374                priority = ZONE_RECLAIM_PRIORITY;
2375                do {
2376                        note_zone_scanning_priority(zone, priority);
2377                        shrink_zone(priority, zone, &sc);
2378                        priority--;
2379                } while (priority >= 0 && sc.nr_reclaimed < nr_pages);
2380        }
2381
2382        slab_reclaimable = zone_page_state(zone, NR_SLAB_RECLAIMABLE);
2383        if (slab_reclaimable > zone->min_slab_pages) {
2384                /*
2385                 * shrink_slab() does not currently allow us to determine how
2386                 * many pages were freed in this zone. So we take the current
2387                 * number of slab pages and shake the slab until it is reduced
2388                 * by the same nr_pages that we used for reclaiming unmapped
2389                 * pages.
2390                 *
2391                 * Note that shrink_slab will free memory on all zones and may
2392                 * take a long time.
2393                 */
2394                while (shrink_slab(sc.nr_scanned, gfp_mask, order) &&
2395                        zone_page_state(zone, NR_SLAB_RECLAIMABLE) >
2396                                slab_reclaimable - nr_pages)
2397                        ;
2398
2399                /*
2400                 * Update nr_reclaimed by the number of slab pages we
2401                 * reclaimed from this zone.
2402                 */
2403                sc.nr_reclaimed += slab_reclaimable -
2404                        zone_page_state(zone, NR_SLAB_RECLAIMABLE);
2405        }
2406
2407        p->reclaim_state = NULL;
2408        current->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE);
2409        return sc.nr_reclaimed >= nr_pages;
2410}
2411
2412int zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order)
2413{
2414        int node_id;
2415        int ret;
2416
2417        /*
2418         * Zone reclaim reclaims unmapped file backed pages and
2419         * slab pages if we are over the defined limits.
2420         *
2421         * A small portion of unmapped file backed pages is needed for
2422         * file I/O otherwise pages read by file I/O will be immediately
2423         * thrown out if the zone is overallocated. So we do not reclaim
2424         * if less than a specified percentage of the zone is used by
2425         * unmapped file backed pages.
2426         */
2427        if (zone_pagecache_reclaimable(zone) <= zone->min_unmapped_pages &&
2428            zone_page_state(zone, NR_SLAB_RECLAIMABLE) <= zone->min_slab_pages)
2429                return ZONE_RECLAIM_FULL;
2430
2431        if (zone_is_all_unreclaimable(zone))
2432                return ZONE_RECLAIM_FULL;
2433
2434        /*
2435         * Do not scan if the allocation should not be delayed.
2436         */
2437        if (!(gfp_mask & __GFP_WAIT) || (current->flags & PF_MEMALLOC))
2438                return ZONE_RECLAIM_NOSCAN;
2439
2440        /*
2441         * Only run zone reclaim on the local zone or on zones that do not
2442         * have associated processors. This will favor the local processor
2443         * over remote processors and spread off node memory allocations
2444         * as wide as possible.
2445         */
2446        node_id = zone_to_nid(zone);
2447        if (node_state(node_id, N_CPU) && node_id != numa_node_id())
2448                return ZONE_RECLAIM_NOSCAN;
2449
2450        if (zone_test_and_set_flag(zone, ZONE_RECLAIM_LOCKED))
2451                return ZONE_RECLAIM_NOSCAN;
2452
2453        ret = __zone_reclaim(zone, gfp_mask, order);
2454        zone_clear_flag(zone, ZONE_RECLAIM_LOCKED);
2455
2456        if (!ret)
2457                count_vm_event(PGSCAN_ZONE_RECLAIM_FAILED);
2458
2459        return ret;
2460}
2461#endif
2462
2463#ifdef CONFIG_UNEVICTABLE_LRU
2464/*
2465 * page_evictable - test whether a page is evictable
2466 * @page: the page to test
2467 * @vma: the VMA in which the page is or will be mapped, may be NULL
2468 *
2469 * Test whether page is evictable--i.e., should be placed on active/inactive
2470 * lists vs unevictable list.  The vma argument is !NULL when called from the
2471 * fault path to determine how to instantate a new page.
2472 *
2473 * Reasons page might not be evictable:
2474 * (1) page's mapping marked unevictable
2475 * (2) page is part of an mlocked VMA
2476 *
2477 */
2478int page_evictable(struct page *page, struct vm_area_struct *vma)
2479{
2480
2481        if (mapping_unevictable(page_mapping(page)))
2482                return 0;
2483
2484        if (PageMlocked(page) || (vma && is_mlocked_vma(vma, page)))
2485                return 0;
2486
2487        return 1;
2488}
2489
2490/**
2491 * check_move_unevictable_page - check page for evictability and move to appropriate zone lru list
2492 * @page: page to check evictability and move to appropriate lru list
2493 * @zone: zone page is in
2494 *
2495 * Checks a page for evictability and moves the page to the appropriate
2496 * zone lru list.
2497 *
2498 * Restrictions: zone->lru_lock must be held, page must be on LRU and must
2499 * have PageUnevictable set.
2500 */
2501static void check_move_unevictable_page(struct page *page, struct zone *zone)
2502{
2503        VM_BUG_ON(PageActive(page));
2504
2505retry:
2506        ClearPageUnevictable(page);
2507        if (page_evictable(page, NULL)) {
2508                enum lru_list l = LRU_INACTIVE_ANON + page_is_file_cache(page);
2509
2510                __dec_zone_state(zone, NR_UNEVICTABLE);
2511                list_move(&page->lru, &zone->lru[l].list);
2512                mem_cgroup_move_lists(page, LRU_UNEVICTABLE, l);
2513                __inc_zone_state(zone, NR_INACTIVE_ANON + l);
2514                __count_vm_event(UNEVICTABLE_PGRESCUED);
2515        } else {
2516                /*
2517                 * rotate unevictable list
2518                 */
2519                SetPageUnevictable(page);
2520                list_move(&page->lru, &zone->lru[LRU_UNEVICTABLE].list);
2521                mem_cgroup_rotate_lru_list(page, LRU_UNEVICTABLE);
2522                if (page_evictable(page, NULL))
2523                        goto retry;
2524        }
2525}
2526
2527/**
2528 * scan_mapping_unevictable_pages - scan an address space for evictable pages
2529 * @mapping: struct address_space to scan for evictable pages
2530 *
2531 * Scan all pages in mapping.  Check unevictable pages for
2532 * evictability and move them to the appropriate zone lru list.
2533 */
2534void scan_mapping_unevictable_pages(struct address_space *mapping)
2535{
2536        pgoff_t next = 0;
2537        pgoff_t end   = (i_size_read(mapping->host) + PAGE_CACHE_SIZE - 1) >>
2538                         PAGE_CACHE_SHIFT;
2539        struct zone *zone;
2540        struct pagevec pvec;
2541
2542        if (mapping->nrpages == 0)
2543                return;
2544
2545        pagevec_init(&pvec, 0);
2546        while (next < end &&
2547                pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
2548                int i;
2549                int pg_scanned = 0;
2550
2551                zone = NULL;
2552
2553                for (i = 0; i < pagevec_count(&pvec); i++) {
2554                        struct page *page = pvec.pages[i];
2555                        pgoff_t page_index = page->index;
2556                        struct zone *pagezone = page_zone(page);
2557
2558                        pg_scanned++;
2559                        if (page_index > next)
2560                                next = page_index;
2561                        next++;
2562
2563                        if (pagezone != zone) {
2564                                if (zone)
2565                                        spin_unlock_irq(&zone->lru_lock);
2566                                zone = pagezone;
2567                                spin_lock_irq(&zone->lru_lock);
2568                        }
2569
2570                        if (PageLRU(page) && PageUnevictable(page))
2571                                check_move_unevictable_page(page, zone);
2572                }
2573                if (zone)
2574                        spin_unlock_irq(&zone->lru_lock);
2575                pagevec_release(&pvec);
2576
2577                count_vm_events(UNEVICTABLE_PGSCANNED, pg_scanned);
2578        }
2579
2580}
2581
2582/**
2583 * scan_zone_unevictable_pages - check unevictable list for evictable pages
2584 * @zone - zone of which to scan the unevictable list
2585 *
2586 * Scan @zone's unevictable LRU lists to check for pages that have become
2587 * evictable.  Move those that have to @zone's inactive list where they
2588 * become candidates for reclaim, unless shrink_inactive_zone() decides
2589 * to reactivate them.  Pages that are still unevictable are rotated
2590 * back onto @zone's unevictable list.
2591 */
2592#define SCAN_UNEVICTABLE_BATCH_SIZE 16UL /* arbitrary lock hold batch size */
2593static void scan_zone_unevictable_pages(struct zone *zone)
2594{
2595        struct list_head *l_unevictable = &zone->lru[LRU_UNEVICTABLE].list;
2596        unsigned long scan;
2597        unsigned long nr_to_scan = zone_page_state(zone, NR_UNEVICTABLE);
2598
2599        while (nr_to_scan > 0) {
2600                unsigned long batch_size = min(nr_to_scan,
2601                                                SCAN_UNEVICTABLE_BATCH_SIZE);
2602
2603                spin_lock_irq(&zone->lru_lock);
2604                for (scan = 0;  scan < batch_size; scan++) {
2605                        struct page *page = lru_to_page(l_unevictable);
2606
2607                        if (!trylock_page(page))
2608                                continue;
2609
2610                        prefetchw_prev_lru_page(page, l_unevictable, flags);
2611
2612                        if (likely(PageLRU(page) && PageUnevictable(page)))
2613                                check_move_unevictable_page(page, zone);
2614
2615                        unlock_page(page);
2616                }
2617                spin_unlock_irq(&zone->lru_lock);
2618
2619                nr_to_scan -= batch_size;
2620        }
2621}
2622
2623
2624/**
2625 * scan_all_zones_unevictable_pages - scan all unevictable lists for evictable pages
2626 *
2627 * A really big hammer:  scan all zones' unevictable LRU lists to check for
2628 * pages that have become evictable.  Move those back to the zones'
2629 * inactive list where they become candidates for reclaim.
2630 * This occurs when, e.g., we have unswappable pages on the unevictable lists,
2631 * and we add swap to the system.  As such, it runs in the context of a task
2632 * that has possibly/probably made some previously unevictable pages
2633 * evictable.
2634 */
2635static void scan_all_zones_unevictable_pages(void)
2636{
2637        struct zone *zone;
2638
2639        for_each_zone(zone) {
2640                scan_zone_unevictable_pages(zone);
2641        }
2642}
2643
2644/*
2645 * scan_unevictable_pages [vm] sysctl handler.  On demand re-scan of
2646 * all nodes' unevictable lists for evictable pages
2647 */
2648unsigned long scan_unevictable_pages;
2649
2650int scan_unevictable_handler(struct ctl_table *table, int write,
2651                           struct file *file, void __user *buffer,
2652                           size_t *length, loff_t *ppos)
2653{
2654        proc_doulongvec_minmax(table, write, file, buffer, length, ppos);
2655
2656        if (write && *(unsigned long *)table->data)
2657                scan_all_zones_unevictable_pages();
2658
2659        scan_unevictable_pages = 0;
2660        return 0;
2661}
2662
2663/*
2664 * per node 'scan_unevictable_pages' attribute.  On demand re-scan of
2665 * a specified node's per zone unevictable lists for evictable pages.
2666 */
2667
2668static ssize_t read_scan_unevictable_node(struct sys_device *dev,
2669                                          struct sysdev_attribute *attr,
2670                                          char *buf)
2671{
2672        return sprintf(buf, "0\n");     /* always zero; should fit... */
2673}
2674
2675static ssize_t write_scan_unevictable_node(struct sys_device *dev,
2676                                           struct sysdev_attribute *attr,
2677                                        const char *buf, size_t count)
2678{
2679        struct zone *node_zones = NODE_DATA(dev->id)->node_zones;
2680        struct zone *zone;
2681        unsigned long res;
2682        unsigned long req = strict_strtoul(buf, 10, &res);
2683
2684        if (!req)
2685                return 1;       /* zero is no-op */
2686
2687        for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
2688                if (!populated_zone(zone))
2689                        continue;
2690                scan_zone_unevictable_pages(zone);
2691        }
2692        return 1;
2693}
2694
2695
2696static SYSDEV_ATTR(scan_unevictable_pages, S_IRUGO | S_IWUSR,
2697                        read_scan_unevictable_node,
2698                        write_scan_unevictable_node);
2699
2700int scan_unevictable_register_node(struct node *node)
2701{
2702        return sysdev_create_file(&node->sysdev, &attr_scan_unevictable_pages);
2703}
2704
2705void scan_unevictable_unregister_node(struct node *node)
2706{
2707        sysdev_remove_file(&node->sysdev, &attr_scan_unevictable_pages);
2708}
2709
2710#endif
2711