This reverts commit 96e51ccf1af33e82f429a0d6baebba29c6448d0f. Recently we started running the kernel with rstat infrastructure on production traffic and begin to see negative memcg stats values. Particularly the 'sock' stat is the one which we observed having negative value. $ grep "sock " /mnt/memory/job/memory.stat sock 253952 total_sock 18446744073708724224 Re-run after couple of seconds $ grep "sock " /mnt/memory/job/memory.stat sock 253952 total_sock 53248 For now we are only seeing this issue on large machines (256 CPUs) and only with 'sock' stat. I think the networking stack increase the stat on one cpu and decrease it on another cpu much more often. So, this negative sock is due to rstat flusher flushing the stats on the CPU that has seen the decrement of sock but missed the CPU that has increments. A typical race condition. For easy stable backport, revert is the most simple solution. For long term solution, I am thinking of two directions. First is just reduce the race window by optimizing the rstat flusher. Second is if the reader sees a negative stat value, force flush and restart the stat collection. Basically retry but limited.
Showing the changes to include/linux (1 of 1 file). Show all
Unified · Split
| @@ -987,19 +987,30 @@ static inline void mod_memcg_page_state(struct page *page, · 986 unmodified lines | |||
| 987 | 987 | ||
| 988 | static inline unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx) | 988 | static inline unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx) |
| 989 | { | 989 | { |
| 990 | return READ_ONCE(memcg->vmstats.state[idx]); | 990 | long x = READ_ONCE(memcg->vmstats.state[idx]); |
| 991 | #ifdef CONFIG_SMP | ||
| 992 | if (x < 0) | ||
| 993 | x = 0; | ||
| 994 | #endif | ||
| 995 | return x; | ||
| 991 | } | 996 | } |
| 992 | 997 | ||
| 993 | static inline unsigned long lruvec_page_state(struct lruvec *lruvec, | 998 | static inline unsigned long lruvec_page_state(struct lruvec *lruvec, |
| 994 | enum node_stat_item idx) | 999 | enum node_stat_item idx) |
| 995 | { | 1000 | { |
| 996 | struct mem_cgroup_per_node *pn; | 1001 | struct mem_cgroup_per_node *pn; |
| 1002 | long x; | ||
| 997 | 1003 | ||
| 998 | if (mem_cgroup_disabled()) | 1004 | if (mem_cgroup_disabled()) |
| 999 | return node_page_state(lruvec_pgdat(lruvec), idx); | 1005 | return node_page_state(lruvec_pgdat(lruvec), idx); |
| 1000 | 1006 | ||
| 1001 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); | 1007 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
| 1002 | return READ_ONCE(pn->lruvec_stats.state[idx]); | 1008 | x = READ_ONCE(pn->lruvec_stats.state[idx]); |
| 1009 | #ifdef CONFIG_SMP | ||
| 1010 | if (x < 0) | ||
| 1011 | x = 0; | ||
| 1012 | #endif | ||
| 1013 | return x; | ||
| 1003 | } | 1014 | } |
| 1004 | 1015 | ||
| 1005 | static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec, | 1016 | static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec, |