1 /* memcontrol.c - Memory Controller
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
6 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2 of the License, or
12 * (at your option) any later version.
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
20 #include <linux/res_counter.h>
21 #include <linux/memcontrol.h>
22 #include <linux/cgroup.h>
24 #include <linux/pagemap.h>
25 #include <linux/smp.h>
26 #include <linux/page-flags.h>
27 #include <linux/backing-dev.h>
28 #include <linux/bit_spinlock.h>
29 #include <linux/rcupdate.h>
30 #include <linux/mutex.h>
31 #include <linux/slab.h>
32 #include <linux/swap.h>
33 #include <linux/spinlock.h>
35 #include <linux/seq_file.h>
36 #include <linux/vmalloc.h>
37 #include <linux/mm_inline.h>
38 #include <linux/page_cgroup.h>
41 #include <asm/uaccess.h>
43 struct cgroup_subsys mem_cgroup_subsys __read_mostly;
44 #define MEM_CGROUP_RECLAIM_RETRIES 5
46 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
47 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 0 */
48 int do_swap_account __read_mostly;
49 static int really_do_swap_account __initdata = 1; /* for remember boot option*/
51 #define do_swap_account (0)
54 static DEFINE_MUTEX(memcg_tasklist); /* can be hold under cgroup_mutex */
57 * Statistics for memory cgroup.
59 enum mem_cgroup_stat_index {
61 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
63 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
64 MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */
65 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
66 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
68 MEM_CGROUP_STAT_NSTATS,
71 struct mem_cgroup_stat_cpu {
72 s64 count[MEM_CGROUP_STAT_NSTATS];
73 } ____cacheline_aligned_in_smp;
75 struct mem_cgroup_stat {
76 struct mem_cgroup_stat_cpu cpustat[0];
80 * For accounting under irq disable, no need for increment preempt count.
82 static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat,
83 enum mem_cgroup_stat_index idx, int val)
85 stat->count[idx] += val;
88 static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat,
89 enum mem_cgroup_stat_index idx)
93 for_each_possible_cpu(cpu)
94 ret += stat->cpustat[cpu].count[idx];
98 static s64 mem_cgroup_local_usage(struct mem_cgroup_stat *stat)
102 ret = mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_CACHE);
103 ret += mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_RSS);
108 * per-zone information in memory controller.
110 struct mem_cgroup_per_zone {
112 * spin_lock to protect the per cgroup LRU
114 struct list_head lists[NR_LRU_LISTS];
115 unsigned long count[NR_LRU_LISTS];
117 struct zone_reclaim_stat reclaim_stat;
119 /* Macro for accessing counter */
120 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
122 struct mem_cgroup_per_node {
123 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
126 struct mem_cgroup_lru_info {
127 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
131 * The memory controller data structure. The memory controller controls both
132 * page cache and RSS per cgroup. We would eventually like to provide
133 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
134 * to help the administrator determine what knobs to tune.
136 * TODO: Add a water mark for the memory controller. Reclaim will begin when
137 * we hit the water mark. May be even add a low water mark, such that
138 * no reclaim occurs from a cgroup at it's low water mark, this is
139 * a feature that will be implemented much later in the future.
142 struct cgroup_subsys_state css;
144 * the counter to account for memory usage
146 struct res_counter res;
148 * the counter to account for mem+swap usage.
150 struct res_counter memsw;
152 * Per cgroup active and inactive list, similar to the
153 * per zone LRU lists.
155 struct mem_cgroup_lru_info info;
158 protect against reclaim related member.
160 spinlock_t reclaim_param_lock;
162 int prev_priority; /* for recording reclaim priority */
165 * While reclaiming in a hiearchy, we cache the last child we
168 int last_scanned_child;
170 * Should the accounting and control be hierarchical, per subtree?
173 unsigned long last_oom_jiffies;
176 unsigned int swappiness;
179 * statistics. This must be placed at the end of memcg.
181 struct mem_cgroup_stat stat;
185 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
186 MEM_CGROUP_CHARGE_TYPE_MAPPED,
187 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
188 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
189 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
193 /* only for here (for easy reading.) */
194 #define PCGF_CACHE (1UL << PCG_CACHE)
195 #define PCGF_USED (1UL << PCG_USED)
196 #define PCGF_LOCK (1UL << PCG_LOCK)
197 static const unsigned long
198 pcg_default_flags[NR_CHARGE_TYPE] = {
199 PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* File Cache */
200 PCGF_USED | PCGF_LOCK, /* Anon */
201 PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* Shmem */
205 /* for encoding cft->private value on file */
208 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
209 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
210 #define MEMFILE_ATTR(val) ((val) & 0xffff)
212 static void mem_cgroup_get(struct mem_cgroup *mem);
213 static void mem_cgroup_put(struct mem_cgroup *mem);
214 static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem);
216 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
217 struct page_cgroup *pc,
220 int val = (charge)? 1 : -1;
221 struct mem_cgroup_stat *stat = &mem->stat;
222 struct mem_cgroup_stat_cpu *cpustat;
225 cpustat = &stat->cpustat[cpu];
226 if (PageCgroupCache(pc))
227 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val);
229 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val);
232 __mem_cgroup_stat_add_safe(cpustat,
233 MEM_CGROUP_STAT_PGPGIN_COUNT, 1);
235 __mem_cgroup_stat_add_safe(cpustat,
236 MEM_CGROUP_STAT_PGPGOUT_COUNT, 1);
240 static struct mem_cgroup_per_zone *
241 mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
243 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
246 static struct mem_cgroup_per_zone *
247 page_cgroup_zoneinfo(struct page_cgroup *pc)
249 struct mem_cgroup *mem = pc->mem_cgroup;
250 int nid = page_cgroup_nid(pc);
251 int zid = page_cgroup_zid(pc);
256 return mem_cgroup_zoneinfo(mem, nid, zid);
259 static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem,
263 struct mem_cgroup_per_zone *mz;
266 for_each_online_node(nid)
267 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
268 mz = mem_cgroup_zoneinfo(mem, nid, zid);
269 total += MEM_CGROUP_ZSTAT(mz, idx);
274 static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
276 return container_of(cgroup_subsys_state(cont,
277 mem_cgroup_subsys_id), struct mem_cgroup,
281 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
284 * mm_update_next_owner() may clear mm->owner to NULL
285 * if it races with swapoff, page migration, etc.
286 * So this can be called with p == NULL.
291 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
292 struct mem_cgroup, css);
295 static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
297 struct mem_cgroup *mem = NULL;
299 * Because we have no locks, mm->owner's may be being moved to other
300 * cgroup. We use css_tryget() here even if this looks
301 * pessimistic (rather than adding locks here).
305 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
308 } while (!css_tryget(&mem->css));
313 static bool mem_cgroup_is_obsolete(struct mem_cgroup *mem)
317 return css_is_removed(&mem->css);
322 * Call callback function against all cgroup under hierarchy tree.
324 static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data,
325 int (*func)(struct mem_cgroup *, void *))
327 int found, ret, nextid;
328 struct cgroup_subsys_state *css;
329 struct mem_cgroup *mem;
331 if (!root->use_hierarchy)
332 return (*func)(root, data);
340 css = css_get_next(&mem_cgroup_subsys, nextid, &root->css,
342 if (css && css_tryget(css))
343 mem = container_of(css, struct mem_cgroup, css);
347 ret = (*func)(mem, data);
351 } while (!ret && css);
357 * Following LRU functions are allowed to be used without PCG_LOCK.
358 * Operations are called by routine of global LRU independently from memcg.
359 * What we have to take care of here is validness of pc->mem_cgroup.
361 * Changes to pc->mem_cgroup happens when
364 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
365 * It is added to LRU before charge.
366 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
367 * When moving account, the page is not on LRU. It's isolated.
370 void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
372 struct page_cgroup *pc;
373 struct mem_cgroup *mem;
374 struct mem_cgroup_per_zone *mz;
376 if (mem_cgroup_disabled())
378 pc = lookup_page_cgroup(page);
379 /* can happen while we handle swapcache. */
380 if (list_empty(&pc->lru) || !pc->mem_cgroup)
383 * We don't check PCG_USED bit. It's cleared when the "page" is finally
384 * removed from global LRU.
386 mz = page_cgroup_zoneinfo(pc);
387 mem = pc->mem_cgroup;
388 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
389 list_del_init(&pc->lru);
393 void mem_cgroup_del_lru(struct page *page)
395 mem_cgroup_del_lru_list(page, page_lru(page));
398 void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
400 struct mem_cgroup_per_zone *mz;
401 struct page_cgroup *pc;
403 if (mem_cgroup_disabled())
406 pc = lookup_page_cgroup(page);
408 * Used bit is set without atomic ops but after smp_wmb().
409 * For making pc->mem_cgroup visible, insert smp_rmb() here.
412 /* unused page is not rotated. */
413 if (!PageCgroupUsed(pc))
415 mz = page_cgroup_zoneinfo(pc);
416 list_move(&pc->lru, &mz->lists[lru]);
419 void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
421 struct page_cgroup *pc;
422 struct mem_cgroup_per_zone *mz;
424 if (mem_cgroup_disabled())
426 pc = lookup_page_cgroup(page);
428 * Used bit is set without atomic ops but after smp_wmb().
429 * For making pc->mem_cgroup visible, insert smp_rmb() here.
432 if (!PageCgroupUsed(pc))
435 mz = page_cgroup_zoneinfo(pc);
436 MEM_CGROUP_ZSTAT(mz, lru) += 1;
437 list_add(&pc->lru, &mz->lists[lru]);
441 * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to
442 * lru because the page may.be reused after it's fully uncharged (because of
443 * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge
444 * it again. This function is only used to charge SwapCache. It's done under
445 * lock_page and expected that zone->lru_lock is never held.
447 static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page)
450 struct zone *zone = page_zone(page);
451 struct page_cgroup *pc = lookup_page_cgroup(page);
453 spin_lock_irqsave(&zone->lru_lock, flags);
455 * Forget old LRU when this page_cgroup is *not* used. This Used bit
456 * is guarded by lock_page() because the page is SwapCache.
458 if (!PageCgroupUsed(pc))
459 mem_cgroup_del_lru_list(page, page_lru(page));
460 spin_unlock_irqrestore(&zone->lru_lock, flags);
463 static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page)
466 struct zone *zone = page_zone(page);
467 struct page_cgroup *pc = lookup_page_cgroup(page);
469 spin_lock_irqsave(&zone->lru_lock, flags);
470 /* link when the page is linked to LRU but page_cgroup isn't */
471 if (PageLRU(page) && list_empty(&pc->lru))
472 mem_cgroup_add_lru_list(page, page_lru(page));
473 spin_unlock_irqrestore(&zone->lru_lock, flags);
477 void mem_cgroup_move_lists(struct page *page,
478 enum lru_list from, enum lru_list to)
480 if (mem_cgroup_disabled())
482 mem_cgroup_del_lru_list(page, from);
483 mem_cgroup_add_lru_list(page, to);
486 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
491 ret = task->mm && mm_match_cgroup(task->mm, mem);
497 * Calculate mapped_ratio under memory controller. This will be used in
498 * vmscan.c for deteremining we have to reclaim mapped pages.
500 int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem)
505 * usage is recorded in bytes. But, here, we assume the number of
506 * physical pages can be represented by "long" on any arch.
508 total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L;
509 rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
510 return (int)((rss * 100L) / total);
514 * prev_priority control...this will be used in memory reclaim path.
516 int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
520 spin_lock(&mem->reclaim_param_lock);
521 prev_priority = mem->prev_priority;
522 spin_unlock(&mem->reclaim_param_lock);
524 return prev_priority;
527 void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
529 spin_lock(&mem->reclaim_param_lock);
530 if (priority < mem->prev_priority)
531 mem->prev_priority = priority;
532 spin_unlock(&mem->reclaim_param_lock);
535 void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
537 spin_lock(&mem->reclaim_param_lock);
538 mem->prev_priority = priority;
539 spin_unlock(&mem->reclaim_param_lock);
542 static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages)
544 unsigned long active;
545 unsigned long inactive;
547 unsigned long inactive_ratio;
549 inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON);
550 active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON);
552 gb = (inactive + active) >> (30 - PAGE_SHIFT);
554 inactive_ratio = int_sqrt(10 * gb);
559 present_pages[0] = inactive;
560 present_pages[1] = active;
563 return inactive_ratio;
566 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg)
568 unsigned long active;
569 unsigned long inactive;
570 unsigned long present_pages[2];
571 unsigned long inactive_ratio;
573 inactive_ratio = calc_inactive_ratio(memcg, present_pages);
575 inactive = present_pages[0];
576 active = present_pages[1];
578 if (inactive * inactive_ratio < active)
584 unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg,
588 int nid = zone->zone_pgdat->node_id;
589 int zid = zone_idx(zone);
590 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
592 return MEM_CGROUP_ZSTAT(mz, lru);
595 struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
598 int nid = zone->zone_pgdat->node_id;
599 int zid = zone_idx(zone);
600 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
602 return &mz->reclaim_stat;
605 struct zone_reclaim_stat *
606 mem_cgroup_get_reclaim_stat_from_page(struct page *page)
608 struct page_cgroup *pc;
609 struct mem_cgroup_per_zone *mz;
611 if (mem_cgroup_disabled())
614 pc = lookup_page_cgroup(page);
616 * Used bit is set without atomic ops but after smp_wmb().
617 * For making pc->mem_cgroup visible, insert smp_rmb() here.
620 if (!PageCgroupUsed(pc))
623 mz = page_cgroup_zoneinfo(pc);
627 return &mz->reclaim_stat;
630 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
631 struct list_head *dst,
632 unsigned long *scanned, int order,
633 int mode, struct zone *z,
634 struct mem_cgroup *mem_cont,
635 int active, int file)
637 unsigned long nr_taken = 0;
641 struct list_head *src;
642 struct page_cgroup *pc, *tmp;
643 int nid = z->zone_pgdat->node_id;
644 int zid = zone_idx(z);
645 struct mem_cgroup_per_zone *mz;
646 int lru = LRU_FILE * !!file + !!active;
649 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
650 src = &mz->lists[lru];
653 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
654 if (scan >= nr_to_scan)
658 if (unlikely(!PageCgroupUsed(pc)))
660 if (unlikely(!PageLRU(page)))
664 if (__isolate_lru_page(page, mode, file) == 0) {
665 list_move(&page->lru, dst);
674 #define mem_cgroup_from_res_counter(counter, member) \
675 container_of(counter, struct mem_cgroup, member)
677 static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem)
679 if (do_swap_account) {
680 if (res_counter_check_under_limit(&mem->res) &&
681 res_counter_check_under_limit(&mem->memsw))
684 if (res_counter_check_under_limit(&mem->res))
689 static unsigned int get_swappiness(struct mem_cgroup *memcg)
691 struct cgroup *cgrp = memcg->css.cgroup;
692 unsigned int swappiness;
695 if (cgrp->parent == NULL)
696 return vm_swappiness;
698 spin_lock(&memcg->reclaim_param_lock);
699 swappiness = memcg->swappiness;
700 spin_unlock(&memcg->reclaim_param_lock);
706 * Visit the first child (need not be the first child as per the ordering
707 * of the cgroup list, since we track last_scanned_child) of @mem and use
708 * that to reclaim free pages from.
710 static struct mem_cgroup *
711 mem_cgroup_select_victim(struct mem_cgroup *root_mem)
713 struct mem_cgroup *ret = NULL;
714 struct cgroup_subsys_state *css;
717 if (!root_mem->use_hierarchy) {
718 css_get(&root_mem->css);
724 nextid = root_mem->last_scanned_child + 1;
725 css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css,
727 if (css && css_tryget(css))
728 ret = container_of(css, struct mem_cgroup, css);
731 /* Updates scanning parameter */
732 spin_lock(&root_mem->reclaim_param_lock);
734 /* this means start scan from ID:1 */
735 root_mem->last_scanned_child = 0;
737 root_mem->last_scanned_child = found;
738 spin_unlock(&root_mem->reclaim_param_lock);
745 * Scan the hierarchy if needed to reclaim memory. We remember the last child
746 * we reclaimed from, so that we don't end up penalizing one child extensively
747 * based on its position in the children list.
749 * root_mem is the original ancestor that we've been reclaim from.
751 * We give up and return to the caller when we visit root_mem twice.
752 * (other groups can be removed while we're walking....)
754 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
755 gfp_t gfp_mask, bool noswap)
757 struct mem_cgroup *victim;
762 victim = mem_cgroup_select_victim(root_mem);
763 if (victim == root_mem)
765 if (!mem_cgroup_local_usage(&victim->stat)) {
766 /* this cgroup's local usage == 0 */
767 css_put(&victim->css);
770 /* we use swappiness of local cgroup */
771 ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, noswap,
772 get_swappiness(victim));
773 css_put(&victim->css);
775 if (mem_cgroup_check_under_limit(root_mem))
781 bool mem_cgroup_oom_called(struct task_struct *task)
784 struct mem_cgroup *mem;
785 struct mm_struct *mm;
791 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
792 if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10))
798 * Unlike exported interface, "oom" parameter is added. if oom==true,
799 * oom-killer can be invoked.
801 static int __mem_cgroup_try_charge(struct mm_struct *mm,
802 gfp_t gfp_mask, struct mem_cgroup **memcg,
805 struct mem_cgroup *mem, *mem_over_limit;
806 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
807 struct res_counter *fail_res;
809 if (unlikely(test_thread_flag(TIF_MEMDIE))) {
810 /* Don't account this! */
816 * We always charge the cgroup the mm_struct belongs to.
817 * The mm_struct's mem_cgroup changes on task migration if the
818 * thread group leader migrates. It's possible that mm is not
819 * set, if so charge the init_mm (happens for pagecache usage).
823 mem = try_get_mem_cgroup_from_mm(mm);
831 VM_BUG_ON(mem_cgroup_is_obsolete(mem));
837 ret = res_counter_charge(&mem->res, PAGE_SIZE, &fail_res);
839 if (!do_swap_account)
841 ret = res_counter_charge(&mem->memsw, PAGE_SIZE,
845 /* mem+swap counter fails */
846 res_counter_uncharge(&mem->res, PAGE_SIZE);
848 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
851 /* mem counter fails */
852 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
855 if (!(gfp_mask & __GFP_WAIT))
858 ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, gfp_mask,
864 * try_to_free_mem_cgroup_pages() might not give us a full
865 * picture of reclaim. Some pages are reclaimed and might be
866 * moved to swap cache or just unmapped from the cgroup.
867 * Check the limit again to see if the reclaim reduced the
868 * current usage of the cgroup before giving up
871 if (mem_cgroup_check_under_limit(mem_over_limit))
876 mutex_lock(&memcg_tasklist);
877 mem_cgroup_out_of_memory(mem_over_limit, gfp_mask);
878 mutex_unlock(&memcg_tasklist);
879 mem_over_limit->last_oom_jiffies = jiffies;
890 static struct mem_cgroup *try_get_mem_cgroup_from_swapcache(struct page *page)
892 struct mem_cgroup *mem;
895 if (!PageSwapCache(page))
898 ent.val = page_private(page);
899 mem = lookup_swap_cgroup(ent);
902 if (!css_tryget(&mem->css))
908 * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be
909 * USED state. If already USED, uncharge and return.
912 static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
913 struct page_cgroup *pc,
914 enum charge_type ctype)
916 /* try_charge() can return NULL to *memcg, taking care of it. */
920 lock_page_cgroup(pc);
921 if (unlikely(PageCgroupUsed(pc))) {
922 unlock_page_cgroup(pc);
923 res_counter_uncharge(&mem->res, PAGE_SIZE);
925 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
929 pc->mem_cgroup = mem;
931 pc->flags = pcg_default_flags[ctype];
933 mem_cgroup_charge_statistics(mem, pc, true);
935 unlock_page_cgroup(pc);
939 * mem_cgroup_move_account - move account of the page
940 * @pc: page_cgroup of the page.
941 * @from: mem_cgroup which the page is moved from.
942 * @to: mem_cgroup which the page is moved to. @from != @to.
944 * The caller must confirm following.
945 * - page is not on LRU (isolate_page() is useful.)
947 * returns 0 at success,
948 * returns -EBUSY when lock is busy or "pc" is unstable.
950 * This function does "uncharge" from old cgroup but doesn't do "charge" to
951 * new cgroup. It should be done by a caller.
954 static int mem_cgroup_move_account(struct page_cgroup *pc,
955 struct mem_cgroup *from, struct mem_cgroup *to)
957 struct mem_cgroup_per_zone *from_mz, *to_mz;
961 VM_BUG_ON(from == to);
962 VM_BUG_ON(PageLRU(pc->page));
964 nid = page_cgroup_nid(pc);
965 zid = page_cgroup_zid(pc);
966 from_mz = mem_cgroup_zoneinfo(from, nid, zid);
967 to_mz = mem_cgroup_zoneinfo(to, nid, zid);
969 if (!trylock_page_cgroup(pc))
972 if (!PageCgroupUsed(pc))
975 if (pc->mem_cgroup != from)
978 res_counter_uncharge(&from->res, PAGE_SIZE);
979 mem_cgroup_charge_statistics(from, pc, false);
981 res_counter_uncharge(&from->memsw, PAGE_SIZE);
986 mem_cgroup_charge_statistics(to, pc, true);
989 unlock_page_cgroup(pc);
994 * move charges to its parent.
997 static int mem_cgroup_move_parent(struct page_cgroup *pc,
998 struct mem_cgroup *child,
1001 struct page *page = pc->page;
1002 struct cgroup *cg = child->css.cgroup;
1003 struct cgroup *pcg = cg->parent;
1004 struct mem_cgroup *parent;
1012 parent = mem_cgroup_from_cont(pcg);
1015 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false);
1019 if (!get_page_unless_zero(page)) {
1024 ret = isolate_lru_page(page);
1029 ret = mem_cgroup_move_account(pc, child, parent);
1031 putback_lru_page(page);
1034 /* drop extra refcnt by try_charge() */
1035 css_put(&parent->css);
1042 /* drop extra refcnt by try_charge() */
1043 css_put(&parent->css);
1044 /* uncharge if move fails */
1045 res_counter_uncharge(&parent->res, PAGE_SIZE);
1046 if (do_swap_account)
1047 res_counter_uncharge(&parent->memsw, PAGE_SIZE);
1052 * Charge the memory controller for page usage.
1054 * 0 if the charge was successful
1055 * < 0 if the cgroup is over its limit
1057 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
1058 gfp_t gfp_mask, enum charge_type ctype,
1059 struct mem_cgroup *memcg)
1061 struct mem_cgroup *mem;
1062 struct page_cgroup *pc;
1065 pc = lookup_page_cgroup(page);
1066 /* can happen at boot */
1072 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true);
1076 __mem_cgroup_commit_charge(mem, pc, ctype);
1080 int mem_cgroup_newpage_charge(struct page *page,
1081 struct mm_struct *mm, gfp_t gfp_mask)
1083 if (mem_cgroup_disabled())
1085 if (PageCompound(page))
1088 * If already mapped, we don't have to account.
1089 * If page cache, page->mapping has address_space.
1090 * But page->mapping may have out-of-use anon_vma pointer,
1091 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
1094 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
1098 return mem_cgroup_charge_common(page, mm, gfp_mask,
1099 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
1102 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
1105 struct mem_cgroup *mem = NULL;
1108 if (mem_cgroup_disabled())
1110 if (PageCompound(page))
1113 * Corner case handling. This is called from add_to_page_cache()
1114 * in usual. But some FS (shmem) precharges this page before calling it
1115 * and call add_to_page_cache() with GFP_NOWAIT.
1117 * For GFP_NOWAIT case, the page may be pre-charged before calling
1118 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
1119 * charge twice. (It works but has to pay a bit larger cost.)
1120 * And when the page is SwapCache, it should take swap information
1121 * into account. This is under lock_page() now.
1123 if (!(gfp_mask & __GFP_WAIT)) {
1124 struct page_cgroup *pc;
1127 pc = lookup_page_cgroup(page);
1130 lock_page_cgroup(pc);
1131 if (PageCgroupUsed(pc)) {
1132 unlock_page_cgroup(pc);
1135 unlock_page_cgroup(pc);
1138 if (do_swap_account && PageSwapCache(page)) {
1139 mem = try_get_mem_cgroup_from_swapcache(page);
1144 /* SwapCache may be still linked to LRU now. */
1145 mem_cgroup_lru_del_before_commit_swapcache(page);
1148 if (unlikely(!mm && !mem))
1151 if (page_is_file_cache(page))
1152 return mem_cgroup_charge_common(page, mm, gfp_mask,
1153 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
1155 ret = mem_cgroup_charge_common(page, mm, gfp_mask,
1156 MEM_CGROUP_CHARGE_TYPE_SHMEM, mem);
1159 if (PageSwapCache(page))
1160 mem_cgroup_lru_add_after_commit_swapcache(page);
1162 if (do_swap_account && !ret && PageSwapCache(page)) {
1163 swp_entry_t ent = {.val = page_private(page)};
1164 /* avoid double counting */
1165 mem = swap_cgroup_record(ent, NULL);
1167 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1168 mem_cgroup_put(mem);
1175 * While swap-in, try_charge -> commit or cancel, the page is locked.
1176 * And when try_charge() successfully returns, one refcnt to memcg without
1177 * struct page_cgroup is aquired. This refcnt will be cumsumed by
1178 * "commit()" or removed by "cancel()"
1180 int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
1182 gfp_t mask, struct mem_cgroup **ptr)
1184 struct mem_cgroup *mem;
1187 if (mem_cgroup_disabled())
1190 if (!do_swap_account)
1193 * A racing thread's fault, or swapoff, may have already updated
1194 * the pte, and even removed page from swap cache: return success
1195 * to go on to do_swap_page()'s pte_same() test, which should fail.
1197 if (!PageSwapCache(page))
1199 mem = try_get_mem_cgroup_from_swapcache(page);
1203 ret = __mem_cgroup_try_charge(NULL, mask, ptr, true);
1204 /* drop extra refcnt from tryget */
1210 return __mem_cgroup_try_charge(mm, mask, ptr, true);
1213 void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
1215 struct page_cgroup *pc;
1217 if (mem_cgroup_disabled())
1221 pc = lookup_page_cgroup(page);
1222 mem_cgroup_lru_del_before_commit_swapcache(page);
1223 __mem_cgroup_commit_charge(ptr, pc, MEM_CGROUP_CHARGE_TYPE_MAPPED);
1224 mem_cgroup_lru_add_after_commit_swapcache(page);
1226 * Now swap is on-memory. This means this page may be
1227 * counted both as mem and swap....double count.
1228 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
1229 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
1230 * may call delete_from_swap_cache() before reach here.
1232 if (do_swap_account && PageSwapCache(page)) {
1233 swp_entry_t ent = {.val = page_private(page)};
1234 struct mem_cgroup *memcg;
1235 memcg = swap_cgroup_record(ent, NULL);
1237 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1238 mem_cgroup_put(memcg);
1242 /* add this page(page_cgroup) to the LRU we want. */
1246 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
1248 if (mem_cgroup_disabled())
1252 res_counter_uncharge(&mem->res, PAGE_SIZE);
1253 if (do_swap_account)
1254 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1260 * uncharge if !page_mapped(page)
1262 static struct mem_cgroup *
1263 __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
1265 struct page_cgroup *pc;
1266 struct mem_cgroup *mem = NULL;
1267 struct mem_cgroup_per_zone *mz;
1269 if (mem_cgroup_disabled())
1272 if (PageSwapCache(page))
1276 * Check if our page_cgroup is valid
1278 pc = lookup_page_cgroup(page);
1279 if (unlikely(!pc || !PageCgroupUsed(pc)))
1282 lock_page_cgroup(pc);
1284 mem = pc->mem_cgroup;
1286 if (!PageCgroupUsed(pc))
1290 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
1291 if (page_mapped(page))
1294 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
1295 if (!PageAnon(page)) { /* Shared memory */
1296 if (page->mapping && !page_is_file_cache(page))
1298 } else if (page_mapped(page)) /* Anon */
1305 res_counter_uncharge(&mem->res, PAGE_SIZE);
1306 if (do_swap_account && (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT))
1307 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1308 mem_cgroup_charge_statistics(mem, pc, false);
1310 ClearPageCgroupUsed(pc);
1312 * pc->mem_cgroup is not cleared here. It will be accessed when it's
1313 * freed from LRU. This is safe because uncharged page is expected not
1314 * to be reused (freed soon). Exception is SwapCache, it's handled by
1315 * special functions.
1318 mz = page_cgroup_zoneinfo(pc);
1319 unlock_page_cgroup(pc);
1321 /* at swapout, this memcg will be accessed to record to swap */
1322 if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
1328 unlock_page_cgroup(pc);
1332 void mem_cgroup_uncharge_page(struct page *page)
1335 if (page_mapped(page))
1337 if (page->mapping && !PageAnon(page))
1339 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
1342 void mem_cgroup_uncharge_cache_page(struct page *page)
1344 VM_BUG_ON(page_mapped(page));
1345 VM_BUG_ON(page->mapping);
1346 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
1350 * called from __delete_from_swap_cache() and drop "page" account.
1351 * memcg information is recorded to swap_cgroup of "ent"
1353 void mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent)
1355 struct mem_cgroup *memcg;
1357 memcg = __mem_cgroup_uncharge_common(page,
1358 MEM_CGROUP_CHARGE_TYPE_SWAPOUT);
1359 /* record memcg information */
1360 if (do_swap_account && memcg) {
1361 swap_cgroup_record(ent, memcg);
1362 mem_cgroup_get(memcg);
1365 css_put(&memcg->css);
1368 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1370 * called from swap_entry_free(). remove record in swap_cgroup and
1371 * uncharge "memsw" account.
1373 void mem_cgroup_uncharge_swap(swp_entry_t ent)
1375 struct mem_cgroup *memcg;
1377 if (!do_swap_account)
1380 memcg = swap_cgroup_record(ent, NULL);
1382 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1383 mem_cgroup_put(memcg);
1389 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
1392 int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr)
1394 struct page_cgroup *pc;
1395 struct mem_cgroup *mem = NULL;
1398 if (mem_cgroup_disabled())
1401 pc = lookup_page_cgroup(page);
1402 lock_page_cgroup(pc);
1403 if (PageCgroupUsed(pc)) {
1404 mem = pc->mem_cgroup;
1407 unlock_page_cgroup(pc);
1410 ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false);
1417 /* remove redundant charge if migration failed*/
1418 void mem_cgroup_end_migration(struct mem_cgroup *mem,
1419 struct page *oldpage, struct page *newpage)
1421 struct page *target, *unused;
1422 struct page_cgroup *pc;
1423 enum charge_type ctype;
1428 /* at migration success, oldpage->mapping is NULL. */
1429 if (oldpage->mapping) {
1437 if (PageAnon(target))
1438 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
1439 else if (page_is_file_cache(target))
1440 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
1442 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
1444 /* unused page is not on radix-tree now. */
1446 __mem_cgroup_uncharge_common(unused, ctype);
1448 pc = lookup_page_cgroup(target);
1450 * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup.
1451 * So, double-counting is effectively avoided.
1453 __mem_cgroup_commit_charge(mem, pc, ctype);
1456 * Both of oldpage and newpage are still under lock_page().
1457 * Then, we don't have to care about race in radix-tree.
1458 * But we have to be careful that this page is unmapped or not.
1460 * There is a case for !page_mapped(). At the start of
1461 * migration, oldpage was mapped. But now, it's zapped.
1462 * But we know *target* page is not freed/reused under us.
1463 * mem_cgroup_uncharge_page() does all necessary checks.
1465 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
1466 mem_cgroup_uncharge_page(target);
1470 * A call to try to shrink memory usage under specified resource controller.
1471 * This is typically used for page reclaiming for shmem for reducing side
1472 * effect of page allocation from shmem, which is used by some mem_cgroup.
1474 int mem_cgroup_shrink_usage(struct page *page,
1475 struct mm_struct *mm,
1478 struct mem_cgroup *mem = NULL;
1480 int retry = MEM_CGROUP_RECLAIM_RETRIES;
1482 if (mem_cgroup_disabled())
1485 mem = try_get_mem_cgroup_from_swapcache(page);
1487 mem = try_get_mem_cgroup_from_mm(mm);
1492 progress = mem_cgroup_hierarchical_reclaim(mem, gfp_mask, true);
1493 progress += mem_cgroup_check_under_limit(mem);
1494 } while (!progress && --retry);
1502 static DEFINE_MUTEX(set_limit_mutex);
1504 static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
1505 unsigned long long val)
1508 int retry_count = MEM_CGROUP_RECLAIM_RETRIES;
1513 while (retry_count) {
1514 if (signal_pending(current)) {
1519 * Rather than hide all in some function, I do this in
1520 * open coded manner. You see what this really does.
1521 * We have to guarantee mem->res.limit < mem->memsw.limit.
1523 mutex_lock(&set_limit_mutex);
1524 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1525 if (memswlimit < val) {
1527 mutex_unlock(&set_limit_mutex);
1530 ret = res_counter_set_limit(&memcg->res, val);
1531 mutex_unlock(&set_limit_mutex);
1536 progress = mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL,
1538 if (!progress) retry_count--;
1544 int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
1545 unsigned long long val)
1547 int retry_count = MEM_CGROUP_RECLAIM_RETRIES;
1548 u64 memlimit, oldusage, curusage;
1551 if (!do_swap_account)
1554 while (retry_count) {
1555 if (signal_pending(current)) {
1560 * Rather than hide all in some function, I do this in
1561 * open coded manner. You see what this really does.
1562 * We have to guarantee mem->res.limit < mem->memsw.limit.
1564 mutex_lock(&set_limit_mutex);
1565 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1566 if (memlimit > val) {
1568 mutex_unlock(&set_limit_mutex);
1571 ret = res_counter_set_limit(&memcg->memsw, val);
1572 mutex_unlock(&set_limit_mutex);
1577 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
1578 mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL, true);
1579 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
1580 if (curusage >= oldusage)
1587 * This routine traverse page_cgroup in given list and drop them all.
1588 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
1590 static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
1591 int node, int zid, enum lru_list lru)
1594 struct mem_cgroup_per_zone *mz;
1595 struct page_cgroup *pc, *busy;
1596 unsigned long flags, loop;
1597 struct list_head *list;
1600 zone = &NODE_DATA(node)->node_zones[zid];
1601 mz = mem_cgroup_zoneinfo(mem, node, zid);
1602 list = &mz->lists[lru];
1604 loop = MEM_CGROUP_ZSTAT(mz, lru);
1605 /* give some margin against EBUSY etc...*/
1610 spin_lock_irqsave(&zone->lru_lock, flags);
1611 if (list_empty(list)) {
1612 spin_unlock_irqrestore(&zone->lru_lock, flags);
1615 pc = list_entry(list->prev, struct page_cgroup, lru);
1617 list_move(&pc->lru, list);
1619 spin_unlock_irqrestore(&zone->lru_lock, flags);
1622 spin_unlock_irqrestore(&zone->lru_lock, flags);
1624 ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL);
1628 if (ret == -EBUSY || ret == -EINVAL) {
1629 /* found lock contention or "pc" is obsolete. */
1636 if (!ret && !list_empty(list))
1642 * make mem_cgroup's charge to be 0 if there is no task.
1643 * This enables deleting this mem_cgroup.
1645 static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
1648 int node, zid, shrink;
1649 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
1650 struct cgroup *cgrp = mem->css.cgroup;
1655 /* should free all ? */
1659 while (mem->res.usage > 0) {
1661 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
1664 if (signal_pending(current))
1666 /* This is for making all *used* pages to be on LRU. */
1667 lru_add_drain_all();
1669 for_each_node_state(node, N_HIGH_MEMORY) {
1670 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
1673 ret = mem_cgroup_force_empty_list(mem,
1682 /* it seems parent cgroup doesn't have enough mem */
1693 /* returns EBUSY if there is a task or if we come here twice. */
1694 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
1698 /* we call try-to-free pages for make this cgroup empty */
1699 lru_add_drain_all();
1700 /* try to free all pages in this cgroup */
1702 while (nr_retries && mem->res.usage > 0) {
1705 if (signal_pending(current)) {
1709 progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL,
1710 false, get_swappiness(mem));
1713 /* maybe some writeback is necessary */
1714 congestion_wait(WRITE, HZ/10);
1719 /* try move_account...there may be some *locked* pages. */
1726 int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
1728 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
1732 static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
1734 return mem_cgroup_from_cont(cont)->use_hierarchy;
1737 static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
1741 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1742 struct cgroup *parent = cont->parent;
1743 struct mem_cgroup *parent_mem = NULL;
1746 parent_mem = mem_cgroup_from_cont(parent);
1750 * If parent's use_hiearchy is set, we can't make any modifications
1751 * in the child subtrees. If it is unset, then the change can
1752 * occur, provided the current cgroup has no children.
1754 * For the root cgroup, parent_mem is NULL, we allow value to be
1755 * set if there are no children.
1757 if ((!parent_mem || !parent_mem->use_hierarchy) &&
1758 (val == 1 || val == 0)) {
1759 if (list_empty(&cont->children))
1760 mem->use_hierarchy = val;
1770 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
1772 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1776 type = MEMFILE_TYPE(cft->private);
1777 name = MEMFILE_ATTR(cft->private);
1780 val = res_counter_read_u64(&mem->res, name);
1783 if (do_swap_account)
1784 val = res_counter_read_u64(&mem->memsw, name);
1793 * The user of this function is...
1796 static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
1799 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
1801 unsigned long long val;
1804 type = MEMFILE_TYPE(cft->private);
1805 name = MEMFILE_ATTR(cft->private);
1808 /* This function does all necessary parse...reuse it */
1809 ret = res_counter_memparse_write_strategy(buffer, &val);
1813 ret = mem_cgroup_resize_limit(memcg, val);
1815 ret = mem_cgroup_resize_memsw_limit(memcg, val);
1818 ret = -EINVAL; /* should be BUG() ? */
1824 static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
1825 unsigned long long *mem_limit, unsigned long long *memsw_limit)
1827 struct cgroup *cgroup;
1828 unsigned long long min_limit, min_memsw_limit, tmp;
1830 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1831 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1832 cgroup = memcg->css.cgroup;
1833 if (!memcg->use_hierarchy)
1836 while (cgroup->parent) {
1837 cgroup = cgroup->parent;
1838 memcg = mem_cgroup_from_cont(cgroup);
1839 if (!memcg->use_hierarchy)
1841 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
1842 min_limit = min(min_limit, tmp);
1843 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1844 min_memsw_limit = min(min_memsw_limit, tmp);
1847 *mem_limit = min_limit;
1848 *memsw_limit = min_memsw_limit;
1852 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
1854 struct mem_cgroup *mem;
1857 mem = mem_cgroup_from_cont(cont);
1858 type = MEMFILE_TYPE(event);
1859 name = MEMFILE_ATTR(event);
1863 res_counter_reset_max(&mem->res);
1865 res_counter_reset_max(&mem->memsw);
1869 res_counter_reset_failcnt(&mem->res);
1871 res_counter_reset_failcnt(&mem->memsw);
1878 /* For read statistics */
1892 struct mcs_total_stat {
1893 s64 stat[NR_MCS_STAT];
1899 } memcg_stat_strings[NR_MCS_STAT] = {
1900 {"cache", "total_cache"},
1901 {"rss", "total_rss"},
1902 {"pgpgin", "total_pgpgin"},
1903 {"pgpgout", "total_pgpgout"},
1904 {"inactive_anon", "total_inactive_anon"},
1905 {"active_anon", "total_active_anon"},
1906 {"inactive_file", "total_inactive_file"},
1907 {"active_file", "total_active_file"},
1908 {"unevictable", "total_unevictable"}
1912 static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data)
1914 struct mcs_total_stat *s = data;
1918 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_CACHE);
1919 s->stat[MCS_CACHE] += val * PAGE_SIZE;
1920 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
1921 s->stat[MCS_RSS] += val * PAGE_SIZE;
1922 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGIN_COUNT);
1923 s->stat[MCS_PGPGIN] += val;
1924 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGOUT_COUNT);
1925 s->stat[MCS_PGPGOUT] += val;
1928 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON);
1929 s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE;
1930 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON);
1931 s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE;
1932 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE);
1933 s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE;
1934 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE);
1935 s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE;
1936 val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE);
1937 s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE;
1942 mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s)
1944 mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat);
1947 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
1948 struct cgroup_map_cb *cb)
1950 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
1951 struct mcs_total_stat mystat;
1954 memset(&mystat, 0, sizeof(mystat));
1955 mem_cgroup_get_local_stat(mem_cont, &mystat);
1957 for (i = 0; i < NR_MCS_STAT; i++)
1958 cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
1960 /* Hierarchical information */
1962 unsigned long long limit, memsw_limit;
1963 memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit);
1964 cb->fill(cb, "hierarchical_memory_limit", limit);
1965 if (do_swap_account)
1966 cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
1969 memset(&mystat, 0, sizeof(mystat));
1970 mem_cgroup_get_total_stat(mem_cont, &mystat);
1971 for (i = 0; i < NR_MCS_STAT; i++)
1972 cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
1975 #ifdef CONFIG_DEBUG_VM
1976 cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL));
1980 struct mem_cgroup_per_zone *mz;
1981 unsigned long recent_rotated[2] = {0, 0};
1982 unsigned long recent_scanned[2] = {0, 0};
1984 for_each_online_node(nid)
1985 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1986 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
1988 recent_rotated[0] +=
1989 mz->reclaim_stat.recent_rotated[0];
1990 recent_rotated[1] +=
1991 mz->reclaim_stat.recent_rotated[1];
1992 recent_scanned[0] +=
1993 mz->reclaim_stat.recent_scanned[0];
1994 recent_scanned[1] +=
1995 mz->reclaim_stat.recent_scanned[1];
1997 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
1998 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
1999 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
2000 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
2007 static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
2009 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
2011 return get_swappiness(memcg);
2014 static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
2017 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
2018 struct mem_cgroup *parent;
2023 if (cgrp->parent == NULL)
2026 parent = mem_cgroup_from_cont(cgrp->parent);
2030 /* If under hierarchy, only empty-root can set this value */
2031 if ((parent->use_hierarchy) ||
2032 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
2037 spin_lock(&memcg->reclaim_param_lock);
2038 memcg->swappiness = val;
2039 spin_unlock(&memcg->reclaim_param_lock);
2047 static struct cftype mem_cgroup_files[] = {
2049 .name = "usage_in_bytes",
2050 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
2051 .read_u64 = mem_cgroup_read,
2054 .name = "max_usage_in_bytes",
2055 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
2056 .trigger = mem_cgroup_reset,
2057 .read_u64 = mem_cgroup_read,
2060 .name = "limit_in_bytes",
2061 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
2062 .write_string = mem_cgroup_write,
2063 .read_u64 = mem_cgroup_read,
2067 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
2068 .trigger = mem_cgroup_reset,
2069 .read_u64 = mem_cgroup_read,
2073 .read_map = mem_control_stat_show,
2076 .name = "force_empty",
2077 .trigger = mem_cgroup_force_empty_write,
2080 .name = "use_hierarchy",
2081 .write_u64 = mem_cgroup_hierarchy_write,
2082 .read_u64 = mem_cgroup_hierarchy_read,
2085 .name = "swappiness",
2086 .read_u64 = mem_cgroup_swappiness_read,
2087 .write_u64 = mem_cgroup_swappiness_write,
2091 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2092 static struct cftype memsw_cgroup_files[] = {
2094 .name = "memsw.usage_in_bytes",
2095 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
2096 .read_u64 = mem_cgroup_read,
2099 .name = "memsw.max_usage_in_bytes",
2100 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
2101 .trigger = mem_cgroup_reset,
2102 .read_u64 = mem_cgroup_read,
2105 .name = "memsw.limit_in_bytes",
2106 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
2107 .write_string = mem_cgroup_write,
2108 .read_u64 = mem_cgroup_read,
2111 .name = "memsw.failcnt",
2112 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
2113 .trigger = mem_cgroup_reset,
2114 .read_u64 = mem_cgroup_read,
2118 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
2120 if (!do_swap_account)
2122 return cgroup_add_files(cont, ss, memsw_cgroup_files,
2123 ARRAY_SIZE(memsw_cgroup_files));
2126 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
2132 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
2134 struct mem_cgroup_per_node *pn;
2135 struct mem_cgroup_per_zone *mz;
2137 int zone, tmp = node;
2139 * This routine is called against possible nodes.
2140 * But it's BUG to call kmalloc() against offline node.
2142 * TODO: this routine can waste much memory for nodes which will
2143 * never be onlined. It's better to use memory hotplug callback
2146 if (!node_state(node, N_NORMAL_MEMORY))
2148 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
2152 mem->info.nodeinfo[node] = pn;
2153 memset(pn, 0, sizeof(*pn));
2155 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
2156 mz = &pn->zoneinfo[zone];
2158 INIT_LIST_HEAD(&mz->lists[l]);
2163 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
2165 kfree(mem->info.nodeinfo[node]);
2168 static int mem_cgroup_size(void)
2170 int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu);
2171 return sizeof(struct mem_cgroup) + cpustat_size;
2174 static struct mem_cgroup *mem_cgroup_alloc(void)
2176 struct mem_cgroup *mem;
2177 int size = mem_cgroup_size();
2179 if (size < PAGE_SIZE)
2180 mem = kmalloc(size, GFP_KERNEL);
2182 mem = vmalloc(size);
2185 memset(mem, 0, size);
2190 * At destroying mem_cgroup, references from swap_cgroup can remain.
2191 * (scanning all at force_empty is too costly...)
2193 * Instead of clearing all references at force_empty, we remember
2194 * the number of reference from swap_cgroup and free mem_cgroup when
2195 * it goes down to 0.
2197 * Removal of cgroup itself succeeds regardless of refs from swap.
2200 static void __mem_cgroup_free(struct mem_cgroup *mem)
2204 free_css_id(&mem_cgroup_subsys, &mem->css);
2206 for_each_node_state(node, N_POSSIBLE)
2207 free_mem_cgroup_per_zone_info(mem, node);
2209 if (mem_cgroup_size() < PAGE_SIZE)
2215 static void mem_cgroup_get(struct mem_cgroup *mem)
2217 atomic_inc(&mem->refcnt);
2220 static void mem_cgroup_put(struct mem_cgroup *mem)
2222 if (atomic_dec_and_test(&mem->refcnt)) {
2223 struct mem_cgroup *parent = parent_mem_cgroup(mem);
2224 __mem_cgroup_free(mem);
2226 mem_cgroup_put(parent);
2231 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
2233 static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem)
2235 if (!mem->res.parent)
2237 return mem_cgroup_from_res_counter(mem->res.parent, res);
2240 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2241 static void __init enable_swap_cgroup(void)
2243 if (!mem_cgroup_disabled() && really_do_swap_account)
2244 do_swap_account = 1;
2247 static void __init enable_swap_cgroup(void)
2252 static struct cgroup_subsys_state * __ref
2253 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
2255 struct mem_cgroup *mem, *parent;
2256 long error = -ENOMEM;
2259 mem = mem_cgroup_alloc();
2261 return ERR_PTR(error);
2263 for_each_node_state(node, N_POSSIBLE)
2264 if (alloc_mem_cgroup_per_zone_info(mem, node))
2267 if (cont->parent == NULL) {
2268 enable_swap_cgroup();
2271 parent = mem_cgroup_from_cont(cont->parent);
2272 mem->use_hierarchy = parent->use_hierarchy;
2275 if (parent && parent->use_hierarchy) {
2276 res_counter_init(&mem->res, &parent->res);
2277 res_counter_init(&mem->memsw, &parent->memsw);
2279 * We increment refcnt of the parent to ensure that we can
2280 * safely access it on res_counter_charge/uncharge.
2281 * This refcnt will be decremented when freeing this
2282 * mem_cgroup(see mem_cgroup_put).
2284 mem_cgroup_get(parent);
2286 res_counter_init(&mem->res, NULL);
2287 res_counter_init(&mem->memsw, NULL);
2289 mem->last_scanned_child = 0;
2290 spin_lock_init(&mem->reclaim_param_lock);
2293 mem->swappiness = get_swappiness(parent);
2294 atomic_set(&mem->refcnt, 1);
2297 __mem_cgroup_free(mem);
2298 return ERR_PTR(error);
2301 static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
2302 struct cgroup *cont)
2304 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2306 return mem_cgroup_force_empty(mem, false);
2309 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
2310 struct cgroup *cont)
2312 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2314 mem_cgroup_put(mem);
2317 static int mem_cgroup_populate(struct cgroup_subsys *ss,
2318 struct cgroup *cont)
2322 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
2323 ARRAY_SIZE(mem_cgroup_files));
2326 ret = register_memsw_files(cont, ss);
2330 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
2331 struct cgroup *cont,
2332 struct cgroup *old_cont,
2333 struct task_struct *p)
2335 mutex_lock(&memcg_tasklist);
2337 * FIXME: It's better to move charges of this process from old
2338 * memcg to new memcg. But it's just on TODO-List now.
2340 mutex_unlock(&memcg_tasklist);
2343 struct cgroup_subsys mem_cgroup_subsys = {
2345 .subsys_id = mem_cgroup_subsys_id,
2346 .create = mem_cgroup_create,
2347 .pre_destroy = mem_cgroup_pre_destroy,
2348 .destroy = mem_cgroup_destroy,
2349 .populate = mem_cgroup_populate,
2350 .attach = mem_cgroup_move_task,
2355 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2357 static int __init disable_swap_account(char *s)
2359 really_do_swap_account = 0;
2362 __setup("noswapaccount", disable_swap_account);