The patch titled Subject: memcg: add mlock statistic in memory.stat has been removed from the -mm tree. Its filename was memcg-add-mlock-statistic-in-memorystat.patch This patch was dropped because it was withdrawn The current -mm tree may be found at http://userweb.kernel.org/~akpm/mmotm/ ------------------------------------------------------ From: Ying Han <yinghan@xxxxxxxxxx> Subject: memcg: add mlock statistic in memory.stat We have the nr_mlock stat both in meminfo as well as vmstat system wide, this patch adds the mlock field into per-memcg memory stat. The stat itself enhances the metrics exported by memcg since the unevictable lru includes more than mlock()'d page like SHM_LOCK'd. Why we need to count mlock'd pages while they are unevictable and we can not do much on them anyway? This is true. The mlock stat I am proposing is more helpful for system admin and kernel developer to understand the system workload. The same information should be helpful to add into OOM log as well. Many times in the past that we need to read the mlock stat from the per-container meminfo for different reason. Afterall, we do have the ability to read the mlock from meminfo and this patch fills the info in memcg. Note: Here are the places where I didn't add the hook: 1. in the mlock_migrate_page() since the owner of oldpage and newpage is the same. 2. in the freeing path since page shouldn't get to there at the first place. Testing: 1 ) $ cat /dev/cgroup/memory/memory.use_hierarchy 1 $ mkdir /dev/cgroup/memory/A $ mkdir /dev/cgroup/memory/A/B $ echo 1g >/dev/cgroup/memory/A/memory.limit_in_bytes $ echo 1g >/dev/cgroup/memory/B/memory.limit_in_bytes 1. Run memtoy in B and mlock 512m file pages: memtoy>file /export/hda3/file_512m private memtoy>map file_512m 0 512m memtoy>lock file_512m memtoy: mlock of file_512m [131072 pages] took 5.296secs. $ cat /dev/cgroup/memory/A/B/memory.stat mlock 536870912 unevictable 536870912 .. total_mlock 536870912 total_unevictable 536870912 $ cat /dev/cgroup/memory/A/memory.stat mlock 0 unevictable 0 .. total_mlock 536870912 total_unevictable 536870912 2) Create 20g memcg and run single thread page fault test (pft) w/ 10g mlock memory, here it measures faults/cpu/second: x before.txt + after.txt +--------------------------------------------------------------------------+ N Min Max Median Avg Stddev x 10 346345.92 349113.01 347470.52 347651.93 819.71411 + 10 345934.67 348973.58 347677.9 347495.33 833.58657 No difference proven at 95.0% confidence [akpm@xxxxxxxxxxxxxxxxxxxx: tweak comments] Signed-off-by: Ying Han <yinghan@xxxxxxxxxx> Cc: Michal Hocko <mhocko@xxxxxxx> Cc: Johannes Weiner <hannes@xxxxxxxxxxx> Cc: Mel Gorman <mel@xxxxxxxxx> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx> Cc: Rik van Riel <riel@xxxxxxxxxx> Cc: Hillf Danton <dhillf@xxxxxxxxx> Cc: Hugh Dickins <hughd@xxxxxxxxxx> Cc: Dan Magenheimer <dan.magenheimer@xxxxxxxxxx> Acked-by: KOSAKI Motohiro <kosaki.motohiro@xxxxxxxxxxxxxx> Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> --- Documentation/cgroups/memory.txt | 1 + include/linux/memcontrol.h | 1 + mm/internal.h | 18 ++++++++++++++++++ mm/memcontrol.c | 16 ++++++++++++++++ mm/mlock.c | 17 ++++++++++++++++- mm/page_alloc.c | 10 ++++++++++ 6 files changed, 62 insertions(+), 1 deletion(-) diff -puN Documentation/cgroups/memory.txt~memcg-add-mlock-statistic-in-memorystat Documentation/cgroups/memory.txt --- a/Documentation/cgroups/memory.txt~memcg-add-mlock-statistic-in-memorystat +++ a/Documentation/cgroups/memory.txt @@ -428,6 +428,7 @@ memory.stat file includes following stat cache - # of bytes of page cache memory. rss - # of bytes of anonymous and swap cache memory. mapped_file - # of bytes of mapped file (includes tmpfs/shmem) +mlock - # of bytes of mlocked memory. pgpgin - # of charging events to the memory cgroup. The charging event happens each time a page is accounted as either mapped anon page(RSS) or cache page(Page Cache) to the cgroup. diff -puN include/linux/memcontrol.h~memcg-add-mlock-statistic-in-memorystat include/linux/memcontrol.h --- a/include/linux/memcontrol.h~memcg-add-mlock-statistic-in-memorystat +++ a/include/linux/memcontrol.h @@ -30,6 +30,7 @@ struct mm_struct; /* Stats that can be updated by kernel. */ enum mem_cgroup_page_stat_item { MEMCG_NR_FILE_MAPPED, /* # of pages charged as file rss */ + MEMCG_NR_MLOCK, /* # of pages charged as mlock */ }; struct mem_cgroup_reclaim_cookie { diff -puN mm/internal.h~memcg-add-mlock-statistic-in-memorystat mm/internal.h --- a/mm/internal.h~memcg-add-mlock-statistic-in-memorystat +++ a/mm/internal.h @@ -12,6 +12,7 @@ #define __MM_INTERNAL_H #include <linux/mm.h> +#include <linux/memcontrol.h> void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma, unsigned long floor, unsigned long ceiling); @@ -174,15 +175,22 @@ static inline void munlock_vma_pages_all static inline int mlocked_vma_newpage(struct vm_area_struct *vma, struct page *page) { + bool locked; + unsigned long flags; + VM_BUG_ON(PageLRU(page)); if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED)) return 0; + mem_cgroup_begin_update_page_stat(page, &locked, &flags); if (!TestSetPageMlocked(page)) { inc_zone_page_state(page, NR_MLOCK); + mem_cgroup_inc_page_stat(page, MEMCG_NR_MLOCK); count_vm_event(UNEVICTABLE_PGMLOCKED); } + mem_cgroup_end_update_page_stat(page, &locked, &flags); + return 1; } @@ -204,8 +212,13 @@ extern void munlock_vma_page(struct page extern void __clear_page_mlock(struct page *page); static inline void clear_page_mlock(struct page *page) { + bool locked; + unsigned long flags; + + mem_cgroup_begin_update_page_stat(page, &locked, &flags); if (unlikely(TestClearPageMlocked(page))) __clear_page_mlock(page); + mem_cgroup_end_update_page_stat(page, &locked, &flags); } /* @@ -214,6 +227,11 @@ static inline void clear_page_mlock(stru */ static inline void mlock_migrate_page(struct page *newpage, struct page *page) { + /* + * Here we are supposed to update the page memcg's mlock stat and the + * newpage memcgs' mlock. Since the two pages are always being charged + * to the same memcg there is no need for this. + */ if (TestClearPageMlocked(page)) { unsigned long flags; diff -puN mm/memcontrol.c~memcg-add-mlock-statistic-in-memorystat mm/memcontrol.c --- a/mm/memcontrol.c~memcg-add-mlock-statistic-in-memorystat +++ a/mm/memcontrol.c @@ -97,6 +97,7 @@ enum mem_cgroup_stat_index { MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */ MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */ + MEM_CGROUP_STAT_MLOCK, /* # of pages charged as mlock()ed */ MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */ MEM_CGROUP_STAT_DATA, /* end of data requires synchronization */ MEM_CGROUP_STAT_NSTATS, @@ -1989,6 +1990,9 @@ void mem_cgroup_update_page_stat(struct case MEMCG_NR_FILE_MAPPED: idx = MEM_CGROUP_STAT_FILE_MAPPED; break; + case MEMCG_NR_MLOCK: + idx = MEM_CGROUP_STAT_MLOCK; + break; default: BUG(); } @@ -2641,6 +2645,14 @@ static int mem_cgroup_move_account(struc __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); preempt_enable(); } + + if (PageMlocked(page)) { + /* Update mlocked data for mem_cgroup */ + preempt_disable(); + __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_MLOCK]); + __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_MLOCK]); + preempt_enable(); + } mem_cgroup_charge_statistics(from, anon, -nr_pages); if (uncharge) /* This is not "cancel", but cancel_charge does all we need. */ @@ -4253,6 +4265,7 @@ enum { MCS_CACHE, MCS_RSS, MCS_FILE_MAPPED, + MCS_MLOCK, MCS_PGPGIN, MCS_PGPGOUT, MCS_SWAP, @@ -4277,6 +4290,7 @@ static struct { {"cache", "total_cache"}, {"rss", "total_rss"}, {"mapped_file", "total_mapped_file"}, + {"mlock", "total_mlock"}, {"pgpgin", "total_pgpgin"}, {"pgpgout", "total_pgpgout"}, {"swap", "total_swap"}, @@ -4302,6 +4316,8 @@ mem_cgroup_get_local_stat(struct mem_cgr s->stat[MCS_RSS] += val * PAGE_SIZE; val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED); s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE; + val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_MLOCK); + s->stat[MCS_MLOCK] += val * PAGE_SIZE; val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGIN); s->stat[MCS_PGPGIN] += val; val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGOUT); diff -puN mm/mlock.c~memcg-add-mlock-statistic-in-memorystat mm/mlock.c --- a/mm/mlock.c~memcg-add-mlock-statistic-in-memorystat +++ a/mm/mlock.c @@ -49,7 +49,9 @@ EXPORT_SYMBOL(can_do_mlock); */ /* - * LRU accounting for clear_page_mlock() + * LRU accounting for clear_page_mlock() + * The caller must use mem_cgroup_begin[end]_update_page_stat to prevent a race + * between "move" and "page stat accounting". */ void __clear_page_mlock(struct page *page) { @@ -60,6 +62,7 @@ void __clear_page_mlock(struct page *pag } dec_zone_page_state(page, NR_MLOCK); + mem_cgroup_dec_page_stat(page, MEMCG_NR_MLOCK); count_vm_event(UNEVICTABLE_PGCLEARED); if (!isolate_lru_page(page)) { putback_lru_page(page); @@ -78,14 +81,20 @@ void __clear_page_mlock(struct page *pag */ void mlock_vma_page(struct page *page) { + bool locked; + unsigned long flags; + BUG_ON(!PageLocked(page)); + mem_cgroup_begin_update_page_stat(page, &locked, &flags); if (!TestSetPageMlocked(page)) { inc_zone_page_state(page, NR_MLOCK); + mem_cgroup_inc_page_stat(page, MEMCG_NR_MLOCK); count_vm_event(UNEVICTABLE_PGMLOCKED); if (!isolate_lru_page(page)) putback_lru_page(page); } + mem_cgroup_end_update_page_stat(page, &locked, &flags); } /** @@ -105,10 +114,15 @@ void mlock_vma_page(struct page *page) */ void munlock_vma_page(struct page *page) { + bool locked; + unsigned long flags; + BUG_ON(!PageLocked(page)); + mem_cgroup_begin_update_page_stat(page, &locked, &flags); if (TestClearPageMlocked(page)) { dec_zone_page_state(page, NR_MLOCK); + mem_cgroup_dec_page_stat(page, MEMCG_NR_MLOCK); if (!isolate_lru_page(page)) { int ret = SWAP_AGAIN; @@ -141,6 +155,7 @@ void munlock_vma_page(struct page *page) count_vm_event(UNEVICTABLE_PGMUNLOCKED); } } + mem_cgroup_end_update_page_stat(page, &locked, &flags); } /** diff -puN mm/page_alloc.c~memcg-add-mlock-statistic-in-memorystat mm/page_alloc.c --- a/mm/page_alloc.c~memcg-add-mlock-statistic-in-memorystat +++ a/mm/page_alloc.c @@ -722,6 +722,11 @@ static void __free_pages_ok(struct page return; local_irq_save(flags); + /* + * Note: we didn't update the page memcg's mlock stat since we believe + * the mlocked page shouldn't get to here. However, we could be wrong + * and a warn_once would tell us. + */ if (unlikely(wasMlocked)) free_page_mlock(page); __count_vm_events(PGFREE, 1 << order); @@ -1297,6 +1302,11 @@ void free_hot_cold_page(struct page *pag migratetype = get_pageblock_migratetype(page); set_page_private(page, migratetype); local_irq_save(flags); + /* + * Note: we didn't update the page memcg's mlock stat since we believe + * the mlocked page shouldn't get to here. However, we could be wrong + * and a warn_once would tell us. + */ if (unlikely(wasMlocked)) free_page_mlock(page); __count_vm_event(PGFREE); _ Patches currently in -mm which might be from yinghan@xxxxxxxxxx are mm-vmscan-remove-lumpy-reclaim.patch mm-vmscan-do-not-stall-on-writeback-during-memory-compaction.patch mm-vmscan-remove-reclaim_mode_t.patch mm-rename-is_mlocked_vma-to-mlocked_vma_newpage.patch mm-rename-is_mlocked_vma-to-mlocked_vma_newpage-fix.patch documentation-memcg-future-proof-hierarchical-statistics-documentation.patch memcg-fix-error-code-in-hugetlb_force_memcg_empty.patch rescounters-add-res_counter_uncharge_until.patch memcg-use-res_counter_uncharge_until-in-move_parent.patch memcg-move-charges-to-root-cgroup-if-use_hierarchy=0.patch memcg-dont-uncharge-in-mem_cgroup_move_account.patch remove-__must_check-for-res_counter_charge_nofail.patch mm-memcg-remove-obsolete-statistics-array-boundary-enum-item.patch mm-memcg-convert-numa-stat-to-read_seq_string-interface.patch mm-memcg-print-statistics-directly-to-seq_file.patch mm-memcg-keep-ratelimit-counter-separate-from-event-counters.patch mm-memcg-group-swapped-out-statistics-counter-logically.patch mm-memcg-print-statistics-from-live-counters.patch mm-memcg-print-statistics-from-live-counters-checkpatch-fixes.patch -- 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