mm: __isolate_lru_page_prepare() in isolate_migratepages_block()
authorHugh Dickins <hughd@google.com>
Tue, 22 Mar 2022 21:45:41 +0000 (14:45 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 22 Mar 2022 22:57:08 +0000 (15:57 -0700)
__isolate_lru_page_prepare() conflates two unrelated functions, with the
flags to one disjoint from the flags to the other; and hides some of the
important checks outside of isolate_migratepages_block(), where the
sequence is better to be visible.  It comes from the days of lumpy
reclaim, before compaction, when the combination made more sense.

Move what's needed by mm/compaction.c isolate_migratepages_block() inline
there, and what's needed by mm/vmscan.c isolate_lru_pages() inline there.

Shorten "isolate_mode" to "mode", so the sequence of conditions is easier
to read.  Declare a "mapping" variable, to save one call to page_mapping()
(but not another: calling again after page is locked is necessary).
Simplify isolate_lru_pages() with a "move_to" list pointer.

Link: https://lkml.kernel.org/r/879d62a8-91cc-d3c6-fb3b-69768236df68@google.com
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: David Rientjes <rientjes@google.com>
Reviewed-by: Alex Shi <alexs@kernel.org>
Cc: Alexander Duyck <alexander.duyck@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
include/linux/swap.h
mm/compaction.c
mm/vmscan.c

index 3db4312..a246c13 100644 (file)
@@ -387,7 +387,6 @@ extern void lru_cache_add_inactive_or_unevictable(struct page *page,
 extern unsigned long zone_reclaimable_pages(struct zone *zone);
 extern unsigned long try_to_free_pages(struct zonelist *zonelist, int order,
                                        gfp_t gfp_mask, nodemask_t *mask);
-extern bool __isolate_lru_page_prepare(struct page *page, isolate_mode_t mode);
 extern unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *memcg,
                                                  unsigned long nr_pages,
                                                  gfp_t gfp_mask,
index b4e94cd..b72fb9d 100644 (file)
@@ -785,7 +785,7 @@ static bool too_many_isolated(pg_data_t *pgdat)
  * @cc:                Compaction control structure.
  * @low_pfn:   The first PFN to isolate
  * @end_pfn:   The one-past-the-last PFN to isolate, within same pageblock
- * @isolate_mode: Isolation mode to be used.
+ * @mode:      Isolation mode to be used.
  *
  * Isolate all pages that can be migrated from the range specified by
  * [low_pfn, end_pfn). The range is expected to be within same pageblock.
@@ -798,7 +798,7 @@ static bool too_many_isolated(pg_data_t *pgdat)
  */
 static int
 isolate_migratepages_block(struct compact_control *cc, unsigned long low_pfn,
-                       unsigned long end_pfn, isolate_mode_t isolate_mode)
+                       unsigned long end_pfn, isolate_mode_t mode)
 {
        pg_data_t *pgdat = cc->zone->zone_pgdat;
        unsigned long nr_scanned = 0, nr_isolated = 0;
@@ -806,6 +806,7 @@ isolate_migratepages_block(struct compact_control *cc, unsigned long low_pfn,
        unsigned long flags = 0;
        struct lruvec *locked = NULL;
        struct page *page = NULL, *valid_page = NULL;
+       struct address_space *mapping;
        unsigned long start_pfn = low_pfn;
        bool skip_on_failure = false;
        unsigned long next_skip_pfn = 0;
@@ -990,7 +991,7 @@ isolate_migratepages_block(struct compact_control *cc, unsigned long low_pfn,
                                        locked = NULL;
                                }
 
-                               if (!isolate_movable_page(page, isolate_mode))
+                               if (!isolate_movable_page(page, mode))
                                        goto isolate_success;
                        }
 
@@ -1002,15 +1003,15 @@ isolate_migratepages_block(struct compact_control *cc, unsigned long low_pfn,
                 * so avoid taking lru_lock and isolating it unnecessarily in an
                 * admittedly racy check.
                 */
-               if (!page_mapping(page) &&
-                   page_count(page) > page_mapcount(page))
+               mapping = page_mapping(page);
+               if (!mapping && page_count(page) > page_mapcount(page))
                        goto isolate_fail;
 
                /*
                 * Only allow to migrate anonymous pages in GFP_NOFS context
                 * because those do not depend on fs locks.
                 */
-               if (!(cc->gfp_mask & __GFP_FS) && page_mapping(page))
+               if (!(cc->gfp_mask & __GFP_FS) && mapping)
                        goto isolate_fail;
 
                /*
@@ -1021,9 +1022,45 @@ isolate_migratepages_block(struct compact_control *cc, unsigned long low_pfn,
                if (unlikely(!get_page_unless_zero(page)))
                        goto isolate_fail;
 
-               if (!__isolate_lru_page_prepare(page, isolate_mode))
+               /* Only take pages on LRU: a check now makes later tests safe */
+               if (!PageLRU(page))
+                       goto isolate_fail_put;
+
+               /* Compaction might skip unevictable pages but CMA takes them */
+               if (!(mode & ISOLATE_UNEVICTABLE) && PageUnevictable(page))
+                       goto isolate_fail_put;
+
+               /*
+                * To minimise LRU disruption, the caller can indicate with
+                * ISOLATE_ASYNC_MIGRATE that it only wants to isolate pages
+                * it will be able to migrate without blocking - clean pages
+                * for the most part.  PageWriteback would require blocking.
+                */
+               if ((mode & ISOLATE_ASYNC_MIGRATE) && PageWriteback(page))
                        goto isolate_fail_put;
 
+               if ((mode & ISOLATE_ASYNC_MIGRATE) && PageDirty(page)) {
+                       bool migrate_dirty;
+
+                       /*
+                        * Only pages without mappings or that have a
+                        * ->migratepage callback are possible to migrate
+                        * without blocking. However, we can be racing with
+                        * truncation so it's necessary to lock the page
+                        * to stabilise the mapping as truncation holds
+                        * the page lock until after the page is removed
+                        * from the page cache.
+                        */
+                       if (!trylock_page(page))
+                               goto isolate_fail_put;
+
+                       mapping = page_mapping(page);
+                       migrate_dirty = !mapping || mapping->a_ops->migratepage;
+                       unlock_page(page);
+                       if (!migrate_dirty)
+                               goto isolate_fail_put;
+               }
+
                /* Try isolate the page */
                if (!TestClearPageLRU(page))
                        goto isolate_fail_put;
index d416896..24601f3 100644 (file)
@@ -1998,69 +1998,6 @@ unsigned int reclaim_clean_pages_from_list(struct zone *zone,
        return nr_reclaimed;
 }
 
-/*
- * Attempt to remove the specified page from its LRU.  Only take this page
- * if it is of the appropriate PageActive status.  Pages which are being
- * freed elsewhere are also ignored.
- *
- * page:       page to consider
- * mode:       one of the LRU isolation modes defined above
- *
- * returns true on success, false on failure.
- */
-bool __isolate_lru_page_prepare(struct page *page, isolate_mode_t mode)
-{
-       /* Only take pages on the LRU. */
-       if (!PageLRU(page))
-               return false;
-
-       /* Compaction should not handle unevictable pages but CMA can do so */
-       if (PageUnevictable(page) && !(mode & ISOLATE_UNEVICTABLE))
-               return false;
-
-       /*
-        * To minimise LRU disruption, the caller can indicate that it only
-        * wants to isolate pages it will be able to operate on without
-        * blocking - clean pages for the most part.
-        *
-        * ISOLATE_ASYNC_MIGRATE is used to indicate that it only wants to pages
-        * that it is possible to migrate without blocking
-        */
-       if (mode & ISOLATE_ASYNC_MIGRATE) {
-               /* All the caller can do on PageWriteback is block */
-               if (PageWriteback(page))
-                       return false;
-
-               if (PageDirty(page)) {
-                       struct address_space *mapping;
-                       bool migrate_dirty;
-
-                       /*
-                        * Only pages without mappings or that have a
-                        * ->migratepage callback are possible to migrate
-                        * without blocking. However, we can be racing with
-                        * truncation so it's necessary to lock the page
-                        * to stabilise the mapping as truncation holds
-                        * the page lock until after the page is removed
-                        * from the page cache.
-                        */
-                       if (!trylock_page(page))
-                               return false;
-
-                       mapping = page_mapping(page);
-                       migrate_dirty = !mapping || mapping->a_ops->migratepage;
-                       unlock_page(page);
-                       if (!migrate_dirty)
-                               return false;
-               }
-       }
-
-       if ((mode & ISOLATE_UNMAPPED) && page_mapped(page))
-               return false;
-
-       return true;
-}
-
 /*
  * Update LRU sizes after isolating pages. The LRU size updates must
  * be complete before mem_cgroup_update_lru_size due to a sanity check.
@@ -2112,11 +2049,11 @@ static unsigned long isolate_lru_pages(unsigned long nr_to_scan,
        unsigned long skipped = 0;
        unsigned long scan, total_scan, nr_pages;
        LIST_HEAD(pages_skipped);
-       isolate_mode_t mode = (sc->may_unmap ? 0 : ISOLATE_UNMAPPED);
 
        total_scan = 0;
        scan = 0;
        while (scan < nr_to_scan && !list_empty(src)) {
+               struct list_head *move_to = src;
                struct page *page;
 
                page = lru_to_page(src);
@@ -2126,9 +2063,9 @@ static unsigned long isolate_lru_pages(unsigned long nr_to_scan,
                total_scan += nr_pages;
 
                if (page_zonenum(page) > sc->reclaim_idx) {
-                       list_move(&page->lru, &pages_skipped);
                        nr_skipped[page_zonenum(page)] += nr_pages;
-                       continue;
+                       move_to = &pages_skipped;
+                       goto move;
                }
 
                /*
@@ -2136,37 +2073,34 @@ static unsigned long isolate_lru_pages(unsigned long nr_to_scan,
                 * return with no isolated pages if the LRU mostly contains
                 * ineligible pages.  This causes the VM to not reclaim any
                 * pages, triggering a premature OOM.
-                *
-                * Account all tail pages of THP.  This would not cause
-                * premature OOM since __isolate_lru_page() returns -EBUSY
-                * only when the page is being freed somewhere else.
+                * Account all tail pages of THP.
                 */
                scan += nr_pages;
-               if (!__isolate_lru_page_prepare(page, mode)) {
-                       /* It is being freed elsewhere */
-                       list_move(&page->lru, src);
-                       continue;
-               }
+
+               if (!PageLRU(page))
+                       goto move;
+               if (!sc->may_unmap && page_mapped(page))
+                       goto move;
+
                /*
                 * Be careful not to clear PageLRU until after we're
                 * sure the page is not being freed elsewhere -- the
                 * page release code relies on it.
                 */
-               if (unlikely(!get_page_unless_zero(page))) {
-                       list_move(&page->lru, src);
-                       continue;
-               }
+               if (unlikely(!get_page_unless_zero(page)))
+                       goto move;
 
                if (!TestClearPageLRU(page)) {
                        /* Another thread is already isolating this page */
                        put_page(page);
-                       list_move(&page->lru, src);
-                       continue;
+                       goto move;
                }
 
                nr_taken += nr_pages;
                nr_zone_taken[page_zonenum(page)] += nr_pages;
-               list_move(&page->lru, dst);
+               move_to = dst;
+move:
+               list_move(&page->lru, move_to);
        }
 
        /*
@@ -2190,7 +2124,8 @@ static unsigned long isolate_lru_pages(unsigned long nr_to_scan,
        }
        *nr_scanned = total_scan;
        trace_mm_vmscan_lru_isolate(sc->reclaim_idx, sc->order, nr_to_scan,
-                                   total_scan, skipped, nr_taken, mode, lru);
+                                   total_scan, skipped, nr_taken,
+                                   sc->may_unmap ? 0 : ISOLATE_UNMAPPED, lru);
        update_lru_sizes(lruvec, lru, nr_zone_taken);
        return nr_taken;
 }