The patch titled memcg: use generic percpu instead of private implementation has been added to the -mm tree. Its filename is memcg-use-generic-percpu-instead-of-private-implementation.patch Before you just go and hit "reply", please: a) Consider who else should be cc'ed b) Prefer to cc a suitable mailing list as well c) Ideally: find the original patch on the mailing list and do a reply-to-all to that, adding suitable additional cc's *** Remember to use Documentation/SubmitChecklist when testing your code *** See http://userweb.kernel.org/~akpm/stuff/added-to-mm.txt to find out what to do about this The current -mm tree may be found at http://userweb.kernel.org/~akpm/mmotm/ ------------------------------------------------------ Subject: memcg: use generic percpu instead of private implementation From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx> When per-cpu counter for memcg was implemneted, dynamic percpu allocator was not very good. But now, we have good one and useful macros. This patch replaces memcg's private percpu counter implementation with generic dynamic percpu allocator. The benefits are - We can remove private implementation. - The counters will be NUMA-aware. (Current one is not...) - This patch makes sizeof struct mem_cgroup smaller. Then, struct mem_cgroup may be fit in page size on small config. - About basic performance aspects, see below. [Before] # size mm/memcontrol.o text data bss dec hex filename 24373 2528 4132 31033 7939 mm/memcontrol.o [page-fault-throuput test on 8cpu/SMP in root cgroup] # /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8 Performance counter stats for './multi-fault-fork 8' (5 runs): 45878618 page-faults ( +- 0.110% ) 602635826 cache-misses ( +- 0.105% ) 61.005373262 seconds time elapsed ( +- 0.004% ) Then cache-miss/page fault = 13.14 [After] #size mm/memcontrol.o text data bss dec hex filename 23913 2528 4132 30573 776d mm/memcontrol.o # /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8 Performance counter stats for './multi-fault-fork 8' (5 runs): 48179400 page-faults ( +- 0.271% ) 588628407 cache-misses ( +- 0.136% ) 61.004615021 seconds time elapsed ( +- 0.004% ) Then cache-miss/page fault = 12.22 Text size is reduced. This performance improvement is not big and will be invisible in real world applications. But this result shows this patch has some good effect even on (small) SMP. Here is a test program I used. 1. fork() processes on each cpus. 2. do page fault repeatedly on each process. 3. after 60secs, kill all childredn and exit. (3 is necessary for getting stable data, this is improvement from previous one.) #define _GNU_SOURCE #include <stdio.h> #include <sched.h> #include <sys/mman.h> #include <sys/types.h> #include <sys/stat.h> #include <fcntl.h> #include <signal.h> #include <stdlib.h> /* * For avoiding contention in page table lock, FAULT area is * sparse. If FAULT_LENGTH is too large for your cpus, decrease it. */ #define FAULT_LENGTH (2 * 1024 * 1024) #define PAGE_SIZE 4096 #define MAXNUM (128) void alarm_handler(int sig) { } void *worker(int cpu, int ppid) { void *start, *end; char *c; cpu_set_t set; int i; CPU_ZERO(&set); CPU_SET(cpu, &set); sched_setaffinity(0, sizeof(set), &set); start = mmap(NULL, FAULT_LENGTH, PROT_READ|PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0); if (start == MAP_FAILED) { perror("mmap"); exit(1); } end = start + FAULT_LENGTH; pause(); //fprintf(stderr, "run%d", cpu); while (1) { for (c = (char*)start; (void *)c < end; c += PAGE_SIZE) *c = 0; madvise(start, FAULT_LENGTH, MADV_DONTNEED); } return NULL; } int main(int argc, char *argv[]) { int num, i, ret, pid, status; int pids[MAXNUM]; if (argc < 2) return 0; setpgid(0, 0); signal(SIGALRM, alarm_handler); num = atoi(argv[1]); pid = getpid(); for (i = 0; i < num; ++i) { ret = fork(); if (!ret) { worker(i, pid); exit(0); } pids[i] = ret; } sleep(1); kill(-pid, SIGALRM); sleep(60); for (i = 0; i < num; i++) kill(pids[i], SIGKILL); for (i = 0; i < num; i++) waitpid(pids[i], &status, 0); return 0; } Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx> Cc: Daisuke Nishimura <nishimura@xxxxxxxxxxxxxxxxx> Cc: Balbir Singh <balbir@xxxxxxxxxx> Cc: Pavel Emelyanov <xemul@xxxxxxxxxx> Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> --- mm/memcontrol.c | 184 +++++++++++++++------------------------------- 1 file changed, 63 insertions(+), 121 deletions(-) diff -puN mm/memcontrol.c~memcg-use-generic-percpu-instead-of-private-implementation mm/memcontrol.c --- a/mm/memcontrol.c~memcg-use-generic-percpu-instead-of-private-implementation +++ a/mm/memcontrol.c @@ -89,54 +89,8 @@ enum mem_cgroup_stat_index { struct mem_cgroup_stat_cpu { s64 count[MEM_CGROUP_STAT_NSTATS]; -} ____cacheline_aligned_in_smp; - -struct mem_cgroup_stat { - struct mem_cgroup_stat_cpu cpustat[0]; }; -static inline void -__mem_cgroup_stat_set_safe(struct mem_cgroup_stat_cpu *stat, - enum mem_cgroup_stat_index idx, s64 val) -{ - stat->count[idx] = val; -} - -static inline s64 -__mem_cgroup_stat_read_local(struct mem_cgroup_stat_cpu *stat, - enum mem_cgroup_stat_index idx) -{ - return stat->count[idx]; -} - -/* - * For accounting under irq disable, no need for increment preempt count. - */ -static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat, - enum mem_cgroup_stat_index idx, int val) -{ - stat->count[idx] += val; -} - -static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat, - enum mem_cgroup_stat_index idx) -{ - int cpu; - s64 ret = 0; - for_each_possible_cpu(cpu) - ret += stat->cpustat[cpu].count[idx]; - return ret; -} - -static s64 mem_cgroup_local_usage(struct mem_cgroup_stat *stat) -{ - s64 ret; - - ret = mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_CACHE); - ret += mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_RSS); - return ret; -} - /* * per-zone information in memory controller. */ @@ -270,9 +224,9 @@ struct mem_cgroup { unsigned long move_charge_at_immigrate; /* - * statistics. This must be placed at the end of memcg. + * percpu counter. */ - struct mem_cgroup_stat stat; + struct mem_cgroup_stat_cpu *stat; }; /* Stuffs for move charges at task migration. */ @@ -441,19 +395,14 @@ mem_cgroup_remove_exceeded(struct mem_cg static bool mem_cgroup_soft_limit_check(struct mem_cgroup *mem) { bool ret = false; - int cpu; s64 val; - struct mem_cgroup_stat_cpu *cpustat; - cpu = get_cpu(); - cpustat = &mem->stat.cpustat[cpu]; - val = __mem_cgroup_stat_read_local(cpustat, MEM_CGROUP_STAT_SOFTLIMIT); + val = this_cpu_read(mem->stat->count[MEM_CGROUP_STAT_SOFTLIMIT]); if (unlikely(val < 0)) { - __mem_cgroup_stat_set_safe(cpustat, MEM_CGROUP_STAT_SOFTLIMIT, + this_cpu_write(mem->stat->count[MEM_CGROUP_STAT_SOFTLIMIT], SOFTLIMIT_EVENTS_THRESH); ret = true; } - put_cpu(); return ret; } @@ -549,17 +498,31 @@ mem_cgroup_largest_soft_limit_node(struc return mz; } +static s64 mem_cgroup_read_stat(struct mem_cgroup *mem, + enum mem_cgroup_stat_index idx) +{ + int cpu; + s64 val = 0; + + for_each_possible_cpu(cpu) + val += per_cpu(mem->stat->count[idx], cpu); + return val; +} + +static s64 mem_cgroup_local_usage(struct mem_cgroup *mem) +{ + s64 ret; + + ret = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS); + ret += mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE); + return ret; +} + static void mem_cgroup_swap_statistics(struct mem_cgroup *mem, bool charge) { int val = (charge) ? 1 : -1; - struct mem_cgroup_stat *stat = &mem->stat; - struct mem_cgroup_stat_cpu *cpustat; - int cpu = get_cpu(); - - cpustat = &stat->cpustat[cpu]; - __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_SWAPOUT, val); - put_cpu(); + this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_SWAPOUT], val); } static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, @@ -567,26 +530,22 @@ static void mem_cgroup_charge_statistics bool charge) { int val = (charge) ? 1 : -1; - struct mem_cgroup_stat *stat = &mem->stat; - struct mem_cgroup_stat_cpu *cpustat; - int cpu = get_cpu(); - cpustat = &stat->cpustat[cpu]; + preempt_disable(); + if (PageCgroupCache(pc)) - __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val); + __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_CACHE], val); else - __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val); + __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_RSS], val); if (charge) - __mem_cgroup_stat_add_safe(cpustat, - MEM_CGROUP_STAT_PGPGIN_COUNT, 1); + __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGIN_COUNT]); else - __mem_cgroup_stat_add_safe(cpustat, - MEM_CGROUP_STAT_PGPGOUT_COUNT, 1); - __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_SOFTLIMIT, -1); - __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_THRESHOLDS, -1); + __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGOUT_COUNT]); + __this_cpu_dec(mem->stat->count[MEM_CGROUP_STAT_SOFTLIMIT]); + __this_cpu_dec(mem->stat->count[MEM_CGROUP_STAT_THRESHOLDS]); - put_cpu(); + preempt_enable(); } static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem, @@ -1244,7 +1203,7 @@ static int mem_cgroup_hierarchical_recla } } } - if (!mem_cgroup_local_usage(&victim->stat)) { + if (!mem_cgroup_local_usage(victim)) { /* this cgroup's local usage == 0 */ css_put(&victim->css); continue; @@ -1310,9 +1269,6 @@ static void record_last_oom(struct mem_c void mem_cgroup_update_file_mapped(struct page *page, int val) { struct mem_cgroup *mem; - struct mem_cgroup_stat *stat; - struct mem_cgroup_stat_cpu *cpustat; - int cpu; struct page_cgroup *pc; pc = lookup_page_cgroup(page); @@ -1328,13 +1284,10 @@ void mem_cgroup_update_file_mapped(struc goto done; /* - * Preemption is already disabled, we don't need get_cpu() + * Preemption is already disabled. We can use __this_cpu_xxx */ - cpu = smp_processor_id(); - stat = &mem->stat; - cpustat = &stat->cpustat[cpu]; + __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], val); - __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_FILE_MAPPED, val); done: unlock_page_cgroup(pc); } @@ -1761,9 +1714,6 @@ static void __mem_cgroup_move_account(st struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge) { struct page *page; - int cpu; - struct mem_cgroup_stat *stat; - struct mem_cgroup_stat_cpu *cpustat; VM_BUG_ON(from == to); VM_BUG_ON(PageLRU(pc->page)); @@ -1773,18 +1723,11 @@ static void __mem_cgroup_move_account(st page = pc->page; if (page_mapped(page) && !PageAnon(page)) { - cpu = smp_processor_id(); - /* Update mapped_file data for mem_cgroup "from" */ - stat = &from->stat; - cpustat = &stat->cpustat[cpu]; - __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_FILE_MAPPED, - -1); - - /* Update mapped_file data for mem_cgroup "to" */ - stat = &to->stat; - cpustat = &stat->cpustat[cpu]; - __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_FILE_MAPPED, - 1); + /* Update mapped_file data for mem_cgroup */ + preempt_disable(); + __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); + __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); + preempt_enable(); } mem_cgroup_charge_statistics(from, pc, false); if (uncharge) @@ -2885,7 +2828,7 @@ static int mem_cgroup_get_idx_stat(struct mem_cgroup *mem, void *data) { struct mem_cgroup_idx_data *d = data; - d->val += mem_cgroup_read_stat(&mem->stat, d->idx); + d->val += mem_cgroup_read_stat(mem, d->idx); return 0; } @@ -3126,18 +3069,18 @@ static int mem_cgroup_get_local_stat(str s64 val; /* per cpu stat */ - val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_CACHE); + val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE); s->stat[MCS_CACHE] += val * PAGE_SIZE; - val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); + val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS); s->stat[MCS_RSS] += val * PAGE_SIZE; - val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_FILE_MAPPED); + val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_FILE_MAPPED); s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE; - val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGIN_COUNT); + val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGIN_COUNT); s->stat[MCS_PGPGIN] += val; - val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGOUT_COUNT); + val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGOUT_COUNT); s->stat[MCS_PGPGOUT] += val; if (do_swap_account) { - val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_SWAPOUT); + val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_SWAPOUT); s->stat[MCS_SWAP] += val * PAGE_SIZE; } @@ -3268,19 +3211,14 @@ static int mem_cgroup_swappiness_write(s static bool mem_cgroup_threshold_check(struct mem_cgroup *mem) { bool ret = false; - int cpu; s64 val; - struct mem_cgroup_stat_cpu *cpustat; - cpu = get_cpu(); - cpustat = &mem->stat.cpustat[cpu]; - val = __mem_cgroup_stat_read_local(cpustat, MEM_CGROUP_STAT_THRESHOLDS); + val = this_cpu_read(mem->stat->count[MEM_CGROUP_STAT_THRESHOLDS]); if (unlikely(val < 0)) { - __mem_cgroup_stat_set_safe(cpustat, MEM_CGROUP_STAT_THRESHOLDS, + this_cpu_write(mem->stat->count[MEM_CGROUP_STAT_THRESHOLDS], THRESHOLDS_EVENTS_THRESH); ret = true; } - put_cpu(); return ret; } @@ -3668,17 +3606,12 @@ static void free_mem_cgroup_per_zone_inf kfree(mem->info.nodeinfo[node]); } -static int mem_cgroup_size(void) -{ - int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu); - return sizeof(struct mem_cgroup) + cpustat_size; -} - static struct mem_cgroup *mem_cgroup_alloc(void) { struct mem_cgroup *mem; - int size = mem_cgroup_size(); + int size = sizeof(struct mem_cgroup); + /* Can be very big if MAX_NUMNODES is very big */ if (size < PAGE_SIZE) mem = kmalloc(size, GFP_KERNEL); else @@ -3686,6 +3619,14 @@ static struct mem_cgroup *mem_cgroup_all if (mem) memset(mem, 0, size); + mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu); + if (!mem->stat) { + if (size < PAGE_SIZE) + kfree(mem); + else + vfree(mem); + mem = NULL; + } return mem; } @@ -3710,7 +3651,8 @@ static void __mem_cgroup_free(struct mem for_each_node_state(node, N_POSSIBLE) free_mem_cgroup_per_zone_info(mem, node); - if (mem_cgroup_size() < PAGE_SIZE) + free_percpu(mem->stat); + if (sizeof(struct mem_cgroup) < PAGE_SIZE) kfree(mem); else vfree(mem); _ Patches currently in -mm which might be from kamezawa.hiroyu@xxxxxxxxxxxxxx are origin.patch mm-clean-up-mm_counter.patch mm-avoid-false-sharing-of-mm_counter.patch mm-avoid-false-sharing-of-mm_counter-checkpatch-fixes.patch mm-count-swap-usage.patch mm-count-swap-usage-checkpatch-fixes.patch mm-add-lowmem-detection-logic.patch mm-add-lowmem-detection-logic-fix.patch vmscan-get_scan_ratio-cleanup.patch mm-restore-zone-all_unreclaimable-to-independence-word.patch mm-restore-zone-all_unreclaimable-to-independence-word-fix.patch mm-restore-zone-all_unreclaimable-to-independence-word-fix-2.patch mm-migratec-kill-anon-local-variable-from-migrate_page_copy.patch nodemaskh-remove-macro-any_online_node.patch resources-introduce-generic-page_is_ram.patch x86-remove-bios-data-range-from-e820.patch x86-use-the-generic-page_is_ram.patch cgroup-introduce-cancel_attach.patch cgroup-introduce-coalesce-css_get-and-css_put.patch cgroups-revamp-subsys-array.patch cgroups-subsystem-module-loading-interface.patch cgroups-subsystem-module-loading-interface-fix.patch cgroups-subsystem-module-unloading.patch cgroups-net_cls-as-module.patch cgroups-blkio-subsystem-as-module.patch cgroups-clean-up-cgroup_pidlist_find-a-bit.patch memcg-add-interface-to-move-charge-at-task-migration.patch memcg-move-charges-of-anonymous-page.patch memcg-move-charges-of-anonymous-page-cleanup.patch memcg-improve-performance-in-moving-charge.patch memcg-avoid-oom-during-moving-charge.patch memcg-move-charges-of-anonymous-swap.patch memcg-move-charges-of-anonymous-swap-fix.patch memcg-improve-performance-in-moving-swap-charge.patch memcg-improve-performance-in-moving-swap-charge-fix.patch cgroup-implement-eventfd-based-generic-api-for-notifications.patch cgroup-implement-eventfd-based-generic-api-for-notifications-kconfig-fix.patch cgroup-implement-eventfd-based-generic-api-for-notifications-fixes.patch memcg-extract-mem_group_usage-from-mem_cgroup_read.patch memcg-rework-usage-of-stats-by-soft-limit.patch memcg-implement-memory-thresholds.patch memcg-implement-memory-thresholds-checkpatch-fixes.patch memcg-implement-memory-thresholds-checkpatch-fixes-fix.patch memcg-typo-in-comment-to-mem_cgroup_print_oom_info.patch memcg-use-generic-percpu-instead-of-private-implementation.patch sysctl-clean-up-vm-related-variable-declarations.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