diff options
| -rw-r--r-- | include/linux/mmzone.h | 18 | ||||
| -rw-r--r-- | mm/vmscan.c | 19 |
2 files changed, 27 insertions, 10 deletions
diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h index 3e822335f214..d863698a84e0 100644 --- a/include/linux/mmzone.h +++ b/include/linux/mmzone.h @@ -293,9 +293,21 @@ static inline bool is_active_lru(enum lru_list lru) #define ANON_AND_FILE 2 enum lruvec_flags { - LRUVEC_CONGESTED, /* lruvec has many dirty pages - * backed by a congested BDI - */ + /* + * An lruvec has many dirty pages backed by a congested BDI: + * 1. LRUVEC_CGROUP_CONGESTED is set by cgroup-level reclaim. + * It can be cleared by cgroup reclaim or kswapd. + * 2. LRUVEC_NODE_CONGESTED is set by kswapd node-level reclaim. + * It can only be cleared by kswapd. + * + * Essentially, kswapd can unthrottle an lruvec throttled by cgroup + * reclaim, but not vice versa. This only applies to the root cgroup. + * The goal is to prevent cgroup reclaim on the root cgroup (e.g. + * memory.reclaim) to unthrottle an unbalanced node (that was throttled + * by kswapd). + */ + LRUVEC_CGROUP_CONGESTED, + LRUVEC_NODE_CONGESTED, }; #endif /* !__GENERATING_BOUNDS_H */ diff --git a/mm/vmscan.c b/mm/vmscan.c index b7068be8a034..1080209a568b 100644 --- a/mm/vmscan.c +++ b/mm/vmscan.c @@ -6578,10 +6578,13 @@ again: * Legacy memcg will stall in page writeback so avoid forcibly * stalling in reclaim_throttle(). */ - if ((current_is_kswapd() || - (cgroup_reclaim(sc) && writeback_throttling_sane(sc))) && - sc->nr.dirty && sc->nr.dirty == sc->nr.congested) - set_bit(LRUVEC_CONGESTED, &target_lruvec->flags); + if (sc->nr.dirty && sc->nr.dirty == sc->nr.congested) { + if (cgroup_reclaim(sc) && writeback_throttling_sane(sc)) + set_bit(LRUVEC_CGROUP_CONGESTED, &target_lruvec->flags); + + if (current_is_kswapd()) + set_bit(LRUVEC_NODE_CONGESTED, &target_lruvec->flags); + } /* * Stall direct reclaim for IO completions if the lruvec is @@ -6591,7 +6594,8 @@ again: */ if (!current_is_kswapd() && current_may_throttle() && !sc->hibernation_mode && - test_bit(LRUVEC_CONGESTED, &target_lruvec->flags)) + (test_bit(LRUVEC_CGROUP_CONGESTED, &target_lruvec->flags) || + test_bit(LRUVEC_NODE_CONGESTED, &target_lruvec->flags))) reclaim_throttle(pgdat, VMSCAN_THROTTLE_CONGESTED); if (should_continue_reclaim(pgdat, nr_node_reclaimed, sc)) @@ -6848,7 +6852,7 @@ retry: lruvec = mem_cgroup_lruvec(sc->target_mem_cgroup, zone->zone_pgdat); - clear_bit(LRUVEC_CONGESTED, &lruvec->flags); + clear_bit(LRUVEC_CGROUP_CONGESTED, &lruvec->flags); } } @@ -7237,7 +7241,8 @@ static void clear_pgdat_congested(pg_data_t *pgdat) { struct lruvec *lruvec = mem_cgroup_lruvec(NULL, pgdat); - clear_bit(LRUVEC_CONGESTED, &lruvec->flags); + clear_bit(LRUVEC_NODE_CONGESTED, &lruvec->flags); + clear_bit(LRUVEC_CGROUP_CONGESTED, &lruvec->flags); clear_bit(PGDAT_DIRTY, &pgdat->flags); clear_bit(PGDAT_WRITEBACK, &pgdat->flags); } |