The patch titled Subject: mm: compaction: handle incorrect MIGRATE_UNMOVABLE type pageblocks has been removed from the -mm tree. Its filename was mm-compaction-handle-incorrect-migrate_unmovable-type-pageblocks.patch This patch was dropped because an updated version will be merged The current -mm tree may be found at http://userweb.kernel.org/~akpm/mmotm/ ------------------------------------------------------ From: Bartlomiej Zolnierkiewicz <b.zolnierkie@xxxxxxxxxxx> Subject: mm: compaction: handle incorrect MIGRATE_UNMOVABLE type pageblocks When MIGRATE_UNMOVABLE pages are freed from MIGRATE_UNMOVABLE type pageblock (and some MIGRATE_MOVABLE pages are left in it) waiting until an allocation takes ownership of the block may take too long. The type of the pageblock remains unchanged so the pageblock cannot be used as a migration target during compaction. Fix it by: * Adding enum compact_mode (COMPACT_ASYNC_[MOVABLE,UNMOVABLE], and COMPACT_SYNC) and then converting sync field in struct compact_control to use it. * Adding nr_pageblocks_skipped field to struct compact_control and tracking how many destination pageblocks were of MIGRATE_UNMOVABLE type. If COMPACT_ASYNC_MOVABLE mode compaction ran fully in try_to_compact_pages() (COMPACT_COMPLETE) it implies that there is not a suitable page for allocation. In this case then check how if there were enough MIGRATE_UNMOVABLE pageblocks to try a second pass in COMPACT_ASYNC_UNMOVABLE mode. * Scanning the MIGRATE_UNMOVABLE pageblocks (during COMPACT_SYNC and COMPACT_ASYNC_UNMOVABLE compaction modes) and building a count based on finding PageBuddy pages, page_count(page) == 0 or PageLRU pages. If all pages within the MIGRATE_UNMOVABLE pageblock are in one of those three sets change the whole pageblock type to MIGRATE_MOVABLE. My particular test case (on a ARM EXYNOS4 device with 512 MiB, which means 131072 standard 4KiB pages in 'Normal' zone) is to: - allocate 95000 pages for kernel's usage - free every second page (47500 pages) of memory just allocated - allocate and use 60000 pages from user space - free remaining 60000 pages of kernel memory (now we have fragmented memory occupied mostly by user space pages) - try to allocate 100 order-9 (2048 KiB) pages for kernel's usage The results: - with compaction disabled I get 10 successful allocations - with compaction enabled - 11 successful allocations - with this patch I'm able to get 25 successful allocations NOTE: If we can make kswapd aware of order-0 requests during compaction, we can enhance kswapd with changing mode to COMPACT_ASYNC_FULL (COMPACT_ASYNC_MOVABLE + COMPACT_ASYNC_UNMOVABLE). Please see the following thread: http://marc.info/?l=linux-mm&m=133552069417068&w=2 [minchan@xxxxxxxxxx: minor cleanups] [akpm@xxxxxxxxxxxxxxxxxxxx: make __count_immobile_pages() return bool, fix some comments] [b.zolnierkie@xxxxxxxxxxx: fix can_rescue_unmovable_pageblock() comment and __count_immobile_pages() documentation] [akpm@xxxxxxxxxxxxxxxxxxxx: remove unused local `end_pfn'] Cc: Hugh Dickins <hughd@xxxxxxxxxx> Cc: KOSAKI Motohiro <kosaki.motohiro@xxxxxxxxx> Cc: Dave Jones <davej@xxxxxxxxxx> Cc: Cong Wang <amwang@xxxxxxxxxx> Cc: Markus Trippelsdorf <markus@xxxxxxxxxxxxxxx> Cc: Mel Gorman <mgorman@xxxxxxx> Cc: Minchan Kim <minchan@xxxxxxxxxx> Cc: Rik van Riel <riel@xxxxxxxxxx> Cc: Marek Szyprowski <m.szyprowski@xxxxxxxxxxx> Signed-off-by: Bartlomiej Zolnierkiewicz <b.zolnierkie@xxxxxxxxxxx> Signed-off-by: Kyungmin Park <kyungmin.park@xxxxxxxxxxx> Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> --- include/linux/compaction.h | 19 ++++ include/linux/vm_event_item.h | 1 mm/compaction.c | 141 ++++++++++++++++++++++++++------ mm/internal.h | 11 ++ mm/page_alloc.c | 33 ++++--- mm/vmstat.c | 1 6 files changed, 170 insertions(+), 36 deletions(-) diff -puN include/linux/compaction.h~mm-compaction-handle-incorrect-migrate_unmovable-type-pageblocks include/linux/compaction.h --- a/include/linux/compaction.h~mm-compaction-handle-incorrect-migrate_unmovable-type-pageblocks +++ a/include/linux/compaction.h @@ -1,6 +1,8 @@ #ifndef _LINUX_COMPACTION_H #define _LINUX_COMPACTION_H +#include <linux/node.h> + /* Return values for compact_zone() and try_to_compact_pages() */ /* compaction didn't start as it was not possible or direct reclaim was more suitable */ #define COMPACT_SKIPPED 0 @@ -11,6 +13,23 @@ /* The full zone was compacted */ #define COMPACT_COMPLETE 3 +/* + * compaction supports three modes + * + * COMPACT_ASYNC_MOVABLE uses asynchronous migration and only scans + * MIGRATE_MOVABLE pageblocks as migration sources and targets. + * COMPACT_ASYNC_UNMOVABLE uses asynchronous migration and only scans + * MIGRATE_MOVABLE pageblocks as migration sources. + * MIGRATE_UNMOVABLE pageblocks are scanned as potential migration + * targets and convers them to MIGRATE_MOVABLE if possible + * COMPACT_SYNC uses synchronous migration and scans all pageblocks + */ +enum compact_mode { + COMPACT_ASYNC_MOVABLE, + COMPACT_ASYNC_UNMOVABLE, + COMPACT_SYNC, +}; + #ifdef CONFIG_COMPACTION extern int sysctl_compact_memory; extern int sysctl_compaction_handler(struct ctl_table *table, int write, diff -puN include/linux/vm_event_item.h~mm-compaction-handle-incorrect-migrate_unmovable-type-pageblocks include/linux/vm_event_item.h --- a/include/linux/vm_event_item.h~mm-compaction-handle-incorrect-migrate_unmovable-type-pageblocks +++ a/include/linux/vm_event_item.h @@ -40,6 +40,7 @@ enum vm_event_item { PGPGIN, PGPGOUT, PS #ifdef CONFIG_COMPACTION COMPACTBLOCKS, COMPACTPAGES, COMPACTPAGEFAILED, COMPACTSTALL, COMPACTFAIL, COMPACTSUCCESS, + COMPACT_RESCUED_UNMOVABLE_BLOCKS, #endif #ifdef CONFIG_HUGETLB_PAGE HTLB_BUDDY_PGALLOC, HTLB_BUDDY_PGALLOC_FAIL, diff -puN mm/compaction.c~mm-compaction-handle-incorrect-migrate_unmovable-type-pageblocks mm/compaction.c --- a/mm/compaction.c~mm-compaction-handle-incorrect-migrate_unmovable-type-pageblocks +++ a/mm/compaction.c @@ -236,7 +236,7 @@ isolate_migratepages_range(struct zone * */ while (unlikely(too_many_isolated(zone))) { /* async migration should just abort */ - if (!cc->sync) + if (cc->mode != COMPACT_SYNC) return 0; congestion_wait(BLK_RW_ASYNC, HZ/10); @@ -304,7 +304,8 @@ isolate_migratepages_range(struct zone * * satisfies the allocation */ pageblock_nr = low_pfn >> pageblock_order; - if (!cc->sync && last_pageblock_nr != pageblock_nr && + if (cc->mode != COMPACT_SYNC && + last_pageblock_nr != pageblock_nr && !migrate_async_suitable(get_pageblock_migratetype(page))) { low_pfn += pageblock_nr_pages; low_pfn = ALIGN(low_pfn, pageblock_nr_pages) - 1; @@ -325,7 +326,7 @@ isolate_migratepages_range(struct zone * continue; } - if (!cc->sync) + if (cc->mode != COMPACT_SYNC) mode |= ISOLATE_ASYNC_MIGRATE; lruvec = mem_cgroup_page_lruvec(page, zone); @@ -360,27 +361,87 @@ isolate_migratepages_range(struct zone * #endif /* CONFIG_COMPACTION || CONFIG_CMA */ #ifdef CONFIG_COMPACTION +/* + * Returns true if MIGRATE_UNMOVABLE pageblock can be successfully + * converted to MIGRATE_MOVABLE type, false otherwise. + */ +static bool can_rescue_unmovable_pageblock(struct page *page) +{ + struct zone *zone; + unsigned long pfn, start_pfn; + struct page *start_page; + + zone = page_zone(page); + pfn = page_to_pfn(page); + start_pfn = pfn & ~(pageblock_nr_pages - 1); + start_page = pfn_to_page(start_pfn); + + /* + * Race with page allocator/reclaimer can happen so it is possible + * that MIGRATE_UNMOVABLE type page will end up in MIGRATE_MOVABLE + * type pageblock. However such situation is rare and not critical + * (because page allocator fallback mechanism can also allocate + * MIGRATE_UNMOVABLE type pages in MIGRATE_MOVABLE type pageblock). + */ + return __count_immobile_pages(zone, start_page, 0); +} + +static void rescue_unmovable_pageblock(struct page *page) +{ + set_pageblock_migratetype(page, MIGRATE_MOVABLE); + move_freepages_block(page_zone(page), page, MIGRATE_MOVABLE); + + count_vm_event(COMPACT_RESCUED_UNMOVABLE_BLOCKS); +} + +/* + * MIGRATE_TARGET: good for migration target + * RESCUE_UNMOVABLE_TARGET: good only if we can rescue the unmovable pageblock. + * UNMOVABLE_TARGET: can't migrate because it's a page in unmovable pageblock. + * SKIP_TARGET: can't migrate by another reasons. + */ +enum smt_result { + MIGRATE_TARGET, + RESCUE_UNMOVABLE_TARGET, + UNMOVABLE_TARGET, + SKIP_TARGET, +}; -/* Returns true if the page is within a block suitable for migration to */ -static bool suitable_migration_target(struct page *page) +/* + * Returns MIGRATE_TARGET if the page is within a block suitable for migration + * to, UNMOVABLE_TARGET if the page is within a MIGRATE_UNMOVABLE block, + * SKIP_TARGET otherwise. + */ +static enum smt_result suitable_migration_target(struct page *page, + struct compact_control *cc) { int migratetype = get_pageblock_migratetype(page); /* Don't interfere with memory hot-remove or the min_free_kbytes blocks */ if (migratetype == MIGRATE_ISOLATE || migratetype == MIGRATE_RESERVE) - return false; + return SKIP_TARGET; /* If the page is a large free page, then allow migration */ if (PageBuddy(page) && page_order(page) >= pageblock_order) - return true; + return MIGRATE_TARGET; /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */ - if (migrate_async_suitable(migratetype)) - return true; + if (cc->mode != COMPACT_ASYNC_UNMOVABLE && + migrate_async_suitable(migratetype)) + return MIGRATE_TARGET; + + if (cc->mode == COMPACT_ASYNC_MOVABLE && + migratetype == MIGRATE_UNMOVABLE) + return UNMOVABLE_TARGET; + + if (cc->mode != COMPACT_ASYNC_MOVABLE && + migratetype == MIGRATE_UNMOVABLE && + can_rescue_unmovable_pageblock(page)) + return RESCUE_UNMOVABLE_TARGET; /* Otherwise skip the block */ - return false; + return SKIP_TARGET; } /* @@ -414,6 +475,13 @@ static void isolate_freepages(struct zon zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages; /* + * isolate_freepages() may be called more than once during + * compact_zone_order() run and we want only the most recent + * count. + */ + cc->nr_unmovable_pageblock = 0; + + /* * Isolate free pages until enough are available to migrate the * pages on cc->migratepages. We stop searching if the migrate * and free page scanners meet or enough free pages are isolated. @@ -421,6 +489,7 @@ static void isolate_freepages(struct zon for (; pfn > low_pfn && cc->nr_migratepages > nr_freepages; pfn -= pageblock_nr_pages) { unsigned long isolated; + enum smt_result ret; if (!pfn_valid(pfn)) continue; @@ -437,9 +506,12 @@ static void isolate_freepages(struct zon continue; /* Check the block is suitable for migration */ - if (!suitable_migration_target(page)) + ret = suitable_migration_target(page, cc); + if (ret != MIGRATE_TARGET && ret != RESCUE_UNMOVABLE_TARGET) { + if (ret == UNMOVABLE_TARGET) + cc->nr_unmovable_pageblock++; continue; - + } /* * Found a block suitable for isolating free pages from. Now * we disabled interrupts, double check things are ok and @@ -448,12 +520,16 @@ static void isolate_freepages(struct zon */ isolated = 0; spin_lock_irqsave(&zone->lock, flags); - if (suitable_migration_target(page)) { + ret = suitable_migration_target(page, cc); + if (ret == MIGRATE_TARGET || ret == RESCUE_UNMOVABLE_TARGET) { + if (ret == RESCUE_UNMOVABLE_TARGET) + rescue_unmovable_pageblock(page); end_pfn = min(pfn + pageblock_nr_pages, zone_end_pfn); isolated = isolate_freepages_block(pfn, end_pfn, freelist, false); nr_freepages += isolated; - } + } else if (ret == UNMOVABLE_TARGET) + cc->nr_unmovable_pageblock++; spin_unlock_irqrestore(&zone->lock, flags); /* @@ -685,8 +761,9 @@ static int compact_zone(struct zone *zon nr_migrate = cc->nr_migratepages; err = migrate_pages(&cc->migratepages, compaction_alloc, - (unsigned long)cc, false, - cc->sync ? MIGRATE_SYNC_LIGHT : MIGRATE_ASYNC); + (unsigned long)&cc->freepages, false, + (cc->mode == COMPACT_SYNC) ? MIGRATE_SYNC_LIGHT + : MIGRATE_ASYNC); update_nr_listpages(cc); nr_remaining = cc->nr_migratepages; @@ -718,7 +795,8 @@ out: static unsigned long compact_zone_order(struct zone *zone, int order, gfp_t gfp_mask, - bool sync) + enum compact_mode mode, + unsigned long *nr_pageblocks_skipped) { struct compact_control cc = { .nr_freepages = 0, @@ -726,12 +804,17 @@ static unsigned long compact_zone_order( .order = order, .migratetype = allocflags_to_migratetype(gfp_mask), .zone = zone, - .sync = sync, + .mode = mode, }; + unsigned long rc; + INIT_LIST_HEAD(&cc.freepages); INIT_LIST_HEAD(&cc.migratepages); - return compact_zone(zone, &cc); + rc = compact_zone(zone, &cc); + *nr_pageblocks_skipped = cc.nr_unmovable_pageblock; + + return rc; } int sysctl_extfrag_threshold = 500; @@ -756,6 +839,8 @@ unsigned long try_to_compact_pages(struc struct zoneref *z; struct zone *zone; int rc = COMPACT_SKIPPED; + unsigned long nr_pageblocks_skipped; + enum compact_mode mode; /* * Check whether it is worth even starting compaction. The order check is @@ -772,12 +857,22 @@ unsigned long try_to_compact_pages(struc nodemask) { int status; - status = compact_zone_order(zone, order, gfp_mask, sync); + mode = sync ? COMPACT_SYNC : COMPACT_ASYNC_MOVABLE; +retry: + status = compact_zone_order(zone, order, gfp_mask, mode, + &nr_pageblocks_skipped); rc = max(status, rc); /* If a normal allocation would succeed, stop compacting */ if (zone_watermark_ok(zone, order, low_wmark_pages(zone), 0, 0)) break; + + if (rc == COMPACT_COMPLETE && mode == COMPACT_ASYNC_MOVABLE) { + if (nr_pageblocks_skipped) { + mode = COMPACT_ASYNC_UNMOVABLE; + goto retry; + } + } } return rc; @@ -811,7 +906,7 @@ static int __compact_pgdat(pg_data_t *pg if (ok && cc->order > zone->compact_order_failed) zone->compact_order_failed = cc->order + 1; /* Currently async compaction is never deferred. */ - else if (!ok && cc->sync) + else if (!ok && cc->mode == COMPACT_SYNC) defer_compaction(zone, cc->order); } @@ -826,7 +921,7 @@ int compact_pgdat(pg_data_t *pgdat, int { struct compact_control cc = { .order = order, - .sync = false, + .mode = COMPACT_ASYNC_MOVABLE, }; return __compact_pgdat(pgdat, &cc); @@ -836,7 +931,7 @@ static int compact_node(int nid) { struct compact_control cc = { .order = -1, - .sync = true, + .mode = COMPACT_SYNC, }; return __compact_pgdat(NODE_DATA(nid), &cc); diff -puN mm/internal.h~mm-compaction-handle-incorrect-migrate_unmovable-type-pageblocks mm/internal.h --- a/mm/internal.h~mm-compaction-handle-incorrect-migrate_unmovable-type-pageblocks +++ a/mm/internal.h @@ -94,6 +94,11 @@ extern void putback_lru_page(struct page /* * in mm/page_alloc.c */ +extern void set_pageblock_migratetype(struct page *page, int migratetype); +extern int move_freepages_block(struct zone *zone, struct page *page, + int migratetype); +extern bool __count_immobile_pages(struct zone *zone, struct page *page, + int count); extern void __free_pages_bootmem(struct page *page, unsigned int order); extern void prep_compound_page(struct page *page, unsigned long order); #ifdef CONFIG_MEMORY_FAILURE @@ -101,6 +106,7 @@ extern bool is_free_buddy_page(struct pa #endif #if defined CONFIG_COMPACTION || defined CONFIG_CMA +#include <linux/compaction.h> /* * in mm/compaction.c @@ -119,11 +125,14 @@ struct compact_control { unsigned long nr_migratepages; /* Number of pages to migrate */ unsigned long free_pfn; /* isolate_freepages search base */ unsigned long migrate_pfn; /* isolate_migratepages search base */ - bool sync; /* Synchronous migration */ + enum compact_mode mode; /* Compaction mode */ int order; /* order a direct compactor needs */ int migratetype; /* MOVABLE, RECLAIMABLE etc */ struct zone *zone; + + /* Number of UNMOVABLE destination pageblocks skipped during scan */ + unsigned long nr_unmovable_pageblock; }; unsigned long diff -puN mm/page_alloc.c~mm-compaction-handle-incorrect-migrate_unmovable-type-pageblocks mm/page_alloc.c --- a/mm/page_alloc.c~mm-compaction-handle-incorrect-migrate_unmovable-type-pageblocks +++ a/mm/page_alloc.c @@ -219,9 +219,8 @@ EXPORT_SYMBOL(nr_online_nodes); int page_group_by_mobility_disabled __read_mostly; -static void set_pageblock_migratetype(struct page *page, int migratetype) +void set_pageblock_migratetype(struct page *page, int migratetype) { - if (unlikely(page_group_by_mobility_disabled)) migratetype = MIGRATE_UNMOVABLE; @@ -953,8 +952,8 @@ static int move_freepages(struct zone *z return pages_moved; } -static int move_freepages_block(struct zone *zone, struct page *page, - int migratetype) +int move_freepages_block(struct zone *zone, struct page *page, + int migratetype) { unsigned long start_pfn, end_pfn; struct page *start_page, *end_page; @@ -5466,14 +5465,19 @@ void set_pageblock_flags_group(struct pa __clear_bit(bitidx + start_bitidx, bitmap); } -/* - * This is designed as sub function...plz see page_isolation.c also. - * set/clear page block's type to be ISOLATE. - * page allocater never alloc memory from ISOLATE block. +/** + * __count_immobile_pages - Check pageblock for MIGRATE_UNMOVABLE type pages. + * @zone: Zone pages are in. + * @page: The first page in the pageblock. + * @count: The count of allowed MIGRATE_UNMOVABLE type pages. + * + * Count the number of MIGRATE_UNMOVABLE type pages in the pageblock + * starting with @page. Returns true If the @zone is of ZONE_MOVABLE + * type or the pageblock is of MIGRATE_MOVABLE or MIGRATE_CMA type. + * If the number of MIGRATE_UNMOVABLE type pages inside the pageblock + * is higher than given by @count returns false, true otherwise. */ - -static int -__count_immobile_pages(struct zone *zone, struct page *page, int count) +bool __count_immobile_pages(struct zone *zone, struct page *page, int count) { unsigned long pfn, iter, found; int mt; @@ -5496,6 +5500,11 @@ __count_immobile_pages(struct zone *zone continue; page = pfn_to_page(check); + + /* Do not deal with pageblocks that overlap zones */ + if (page_zone(page) != zone) + return false; + if (!page_count(page)) { if (PageBuddy(page)) iter += (1 << page_order(page)) - 1; @@ -5650,7 +5659,7 @@ static int __alloc_contig_migrate_range( .nr_migratepages = 0, .order = -1, .zone = page_zone(pfn_to_page(start)), - .sync = true, + .mode = COMPACT_SYNC, }; INIT_LIST_HEAD(&cc.migratepages); diff -puN mm/vmstat.c~mm-compaction-handle-incorrect-migrate_unmovable-type-pageblocks mm/vmstat.c --- a/mm/vmstat.c~mm-compaction-handle-incorrect-migrate_unmovable-type-pageblocks +++ a/mm/vmstat.c @@ -767,6 +767,7 @@ const char * const vmstat_text[] = { "compact_stall", "compact_fail", "compact_success", + "compact_rescued_unmovable_blocks", #endif #ifdef CONFIG_HUGETLB_PAGE _ Patches currently in -mm which might be from b.zolnierkie@xxxxxxxxxxx are -- To unsubscribe from this list: send the line "unsubscribe mm-commits" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html