In order to infer whether compaction is likely to enable an allocation that maps/unmaps a pageblock, we need to know how many fully-free pageblocks of each type there are. Signed-off-by: Brendan Jackman --- include/linux/mmzone.h | 5 ++++- include/linux/vmstat.h | 10 ++++++++++ mm/compaction.c | 5 ++--- mm/page_alloc.c | 13 ++++++++++--- mm/vmscan.c | 49 ++++++++++++++++++++++++++++--------------------- mm/vmstat.c | 3 ++- 6 files changed, 56 insertions(+), 29 deletions(-) diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h index 2533a228f71fd..e063a59bbc883 100644 --- a/include/linux/mmzone.h +++ b/include/linux/mmzone.h @@ -181,7 +181,10 @@ enum numa_stat_item { enum zone_stat_item { NR_FREE_PAGES, - NR_FREE_PAGES_BLOCKS, + /* Number of free pages in entirely free pageblocks, with direct map */ + NR_FREE_PAGES_BLOCKS_MAPPED, + /* Ditto, without direct map */ + NR_FREE_PAGES_BLOCKS_UNMAPPED, NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */ NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE, NR_ZONE_ACTIVE_ANON, diff --git a/include/linux/vmstat.h b/include/linux/vmstat.h index 5b31d8e7ae405..debd593756149 100644 --- a/include/linux/vmstat.h +++ b/include/linux/vmstat.h @@ -224,6 +224,16 @@ static inline unsigned long zone_page_state(struct zone *zone, return x; } +/* + * Approx number of pages in entirely free pageblocks. Due to races this could + * actually return a value more than the number of pages in the zone. + */ +static inline unsigned long zone_free_pages_blocks(struct zone *zone) +{ + return zone_page_state(zone, NR_FREE_PAGES_BLOCKS_MAPPED) + + zone_page_state(zone, NR_FREE_PAGES_BLOCKS_UNMAPPED); +} + /* * More accurate version that also considers the currently pending * deltas. For that we need to loop over all cpus to find the current diff --git a/mm/compaction.c b/mm/compaction.c index c9eb3947ffc79..ed12d2fc6fad3 100644 --- a/mm/compaction.c +++ b/mm/compaction.c @@ -2353,8 +2353,7 @@ static enum compact_result __compact_finished(struct compact_control *cc) if (__zone_watermark_ok(cc->zone, cc->order, high_wmark_pages(cc->zone), cc->highest_zoneidx, cc->alloc_flags, - zone_page_state(cc->zone, - NR_FREE_PAGES_BLOCKS))) + zone_free_pages_blocks(cc->zone))) return COMPACT_SUCCESS; return COMPACT_CONTINUE; @@ -2538,7 +2537,7 @@ compaction_suit_allocation_order(struct zone *zone, unsigned int order, unsigned long watermark; if (kcompactd && defrag_mode) - free_pages = zone_page_state(zone, NR_FREE_PAGES_BLOCKS); + free_pages = zone_free_pages_blocks(zone); else free_pages = zone_page_state(zone, NR_FREE_PAGES); diff --git a/mm/page_alloc.c b/mm/page_alloc.c index ac2f6190117ae..d12ce84662ab7 100644 --- a/mm/page_alloc.c +++ b/mm/page_alloc.c @@ -866,6 +866,13 @@ static inline void account_freepages(struct zone *zone, int nr_pages, zone->nr_free_highatomic + nr_pages); } +static inline enum zone_stat_item free_pages_blocks_stat(freetype_t ft) +{ + if (freetype_flags(ft) & FREETYPE_UNMAPPED) + return NR_FREE_PAGES_BLOCKS_UNMAPPED; + return NR_FREE_PAGES_BLOCKS_MAPPED; +} + /* Used for pages not on another list */ static inline void __add_to_free_list(struct page *page, struct zone *zone, unsigned int order, freetype_t freetype, @@ -890,7 +897,7 @@ static inline void __add_to_free_list(struct page *page, struct zone *zone, area->nr_free++; if (order >= pageblock_order && !is_migrate_isolate(free_to_migratetype(freetype))) - __mod_zone_page_state(zone, NR_FREE_PAGES_BLOCKS, nr_pages); + __mod_zone_page_state(zone, free_pages_blocks_stat(freetype), nr_pages); } /* @@ -926,7 +933,7 @@ static inline void move_to_free_list(struct page *page, struct zone *zone, is_migrate_isolate(old_mt) != is_migrate_isolate(new_mt)) { if (!is_migrate_isolate(old_mt)) nr_pages = -nr_pages; - __mod_zone_page_state(zone, NR_FREE_PAGES_BLOCKS, nr_pages); + __mod_zone_page_state(zone, free_pages_blocks_stat(new_ft), nr_pages); } } @@ -954,7 +961,7 @@ static inline void __del_page_from_free_list(struct page *page, struct zone *zon zone->free_area[order].nr_free--; if (order >= pageblock_order && !is_migrate_isolate(free_to_migratetype(freetype))) - __mod_zone_page_state(zone, NR_FREE_PAGES_BLOCKS, -nr_pages); + __mod_zone_page_state(zone, free_pages_blocks_stat(freetype), -nr_pages); } static inline void del_page_from_free_list(struct page *page, struct zone *zone, diff --git a/mm/vmscan.c b/mm/vmscan.c index 566c4e837c7d5..5789c39a0a729 100644 --- a/mm/vmscan.c +++ b/mm/vmscan.c @@ -6935,6 +6935,28 @@ static bool pgdat_watermark_boosted(pg_data_t *pgdat, int highest_zoneidx) return false; } +/* + * Helper to get *FREE_PAGES* zone stats with accuracy heuristics. + * + * When there is a high number of CPUs in the system, the cumulative error from + * the vmstat per-cpu cache can blur the line between the watermarks. In that + * case, be safe and get an accurate snapshot. + * + * TODO: NR_FREE_PAGES_BLOCKS_* move in steps of pageblock_nr_pages, while the + * vmstat pcp threshold is limited to 125. On many configurations that counter + * won't actually be per-cpu cached. But keep things simple for now; revisit + * when somebody cares. + */ +static inline unsigned long get_free_pages_stat(struct zone *zone, + enum zone_stat_item item) +{ + unsigned long free_pages = zone_page_state(zone, item); + + if (zone->percpu_drift_mark && free_pages < zone->percpu_drift_mark) + return zone_page_state_snapshot(zone, item); + return free_pages; +} + /* * Returns true if there is an eligible zone balanced for the request order * and highest_zoneidx @@ -6950,7 +6972,6 @@ static bool pgdat_balanced(pg_data_t *pgdat, int order, int highest_zoneidx) * meet watermarks. */ for_each_managed_zone_pgdat(zone, pgdat, i, highest_zoneidx) { - enum zone_stat_item item; unsigned long free_pages; if (sysctl_numa_balancing_mode & NUMA_BALANCING_MEMORY_TIERING) @@ -6968,26 +6989,12 @@ static bool pgdat_balanced(pg_data_t *pgdat, int order, int highest_zoneidx) * has dropped order, simply ensure there are enough * base pages for compaction, wake kcompactd & sleep. */ - if (defrag_mode && order) - item = NR_FREE_PAGES_BLOCKS; - else - item = NR_FREE_PAGES; - - /* - * When there is a high number of CPUs in the system, - * the cumulative error from the vmstat per-cpu cache - * can blur the line between the watermarks. In that - * case, be safe and get an accurate snapshot. - * - * TODO: NR_FREE_PAGES_BLOCKS moves in steps of - * pageblock_nr_pages, while the vmstat pcp threshold - * is limited to 125. On many configurations that - * counter won't actually be per-cpu cached. But keep - * things simple for now; revisit when somebody cares. - */ - free_pages = zone_page_state(zone, item); - if (zone->percpu_drift_mark && free_pages < zone->percpu_drift_mark) - free_pages = zone_page_state_snapshot(zone, item); + if (defrag_mode && order) { + free_pages = get_free_pages_stat(zone, NR_FREE_PAGES_BLOCKS_UNMAPPED) + + get_free_pages_stat(zone, NR_FREE_PAGES_BLOCKS_MAPPED); + } else { + free_pages = get_free_pages_stat(zone, NR_FREE_PAGES); + } if (__zone_watermark_ok(zone, order, mark, highest_zoneidx, 0, free_pages)) diff --git a/mm/vmstat.c b/mm/vmstat.c index cb57714539fb5..f5ab6ab641c6d 100644 --- a/mm/vmstat.c +++ b/mm/vmstat.c @@ -1200,7 +1200,8 @@ const char * const vmstat_text[] = { /* enum zone_stat_item counters */ #define I(x) (x) [I(NR_FREE_PAGES)] = "nr_free_pages", - [I(NR_FREE_PAGES_BLOCKS)] = "nr_free_pages_blocks", + [I(NR_FREE_PAGES_BLOCKS_MAPPED)] = "nr_free_pages_blocks_mapped", + [I(NR_FREE_PAGES_BLOCKS_UNMAPPED)] = "nr_free_pages_blocks_unmapped", [I(NR_ZONE_INACTIVE_ANON)] = "nr_zone_inactive_anon", [I(NR_ZONE_ACTIVE_ANON)] = "nr_zone_active_anon", [I(NR_ZONE_INACTIVE_FILE)] = "nr_zone_inactive_file", -- 2.54.0