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);
705 static int mem_cgroup_count_children_cb(struct mem_cgroup *mem, void *data)
712 * This function returns the number of memcg under hierarchy tree. Returns
713 * 1(self count) if no children.
715 static int mem_cgroup_count_children(struct mem_cgroup *mem)
718 mem_cgroup_walk_tree(mem, &num, mem_cgroup_count_children_cb);
723 * Visit the first child (need not be the first child as per the ordering
724 * of the cgroup list, since we track last_scanned_child) of @mem and use
725 * that to reclaim free pages from.
727 static struct mem_cgroup *
728 mem_cgroup_select_victim(struct mem_cgroup *root_mem)
730 struct mem_cgroup *ret = NULL;
731 struct cgroup_subsys_state *css;
734 if (!root_mem->use_hierarchy) {
735 css_get(&root_mem->css);
741 nextid = root_mem->last_scanned_child + 1;
742 css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css,
744 if (css && css_tryget(css))
745 ret = container_of(css, struct mem_cgroup, css);
748 /* Updates scanning parameter */
749 spin_lock(&root_mem->reclaim_param_lock);
751 /* this means start scan from ID:1 */
752 root_mem->last_scanned_child = 0;
754 root_mem->last_scanned_child = found;
755 spin_unlock(&root_mem->reclaim_param_lock);
762 * Scan the hierarchy if needed to reclaim memory. We remember the last child
763 * we reclaimed from, so that we don't end up penalizing one child extensively
764 * based on its position in the children list.
766 * root_mem is the original ancestor that we've been reclaim from.
768 * We give up and return to the caller when we visit root_mem twice.
769 * (other groups can be removed while we're walking....)
771 * If shrink==true, for avoiding to free too much, this returns immedieately.
773 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
774 gfp_t gfp_mask, bool noswap, bool shrink)
776 struct mem_cgroup *victim;
781 victim = mem_cgroup_select_victim(root_mem);
782 if (victim == root_mem)
784 if (!mem_cgroup_local_usage(&victim->stat)) {
785 /* this cgroup's local usage == 0 */
786 css_put(&victim->css);
789 /* we use swappiness of local cgroup */
790 ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, noswap,
791 get_swappiness(victim));
792 css_put(&victim->css);
794 * At shrinking usage, we can't check we should stop here or
795 * reclaim more. It's depends on callers. last_scanned_child
796 * will work enough for keeping fairness under tree.
801 if (mem_cgroup_check_under_limit(root_mem))
807 bool mem_cgroup_oom_called(struct task_struct *task)
810 struct mem_cgroup *mem;
811 struct mm_struct *mm;
817 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
818 if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10))
824 * Unlike exported interface, "oom" parameter is added. if oom==true,
825 * oom-killer can be invoked.
827 static int __mem_cgroup_try_charge(struct mm_struct *mm,
828 gfp_t gfp_mask, struct mem_cgroup **memcg,
831 struct mem_cgroup *mem, *mem_over_limit;
832 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
833 struct res_counter *fail_res;
835 if (unlikely(test_thread_flag(TIF_MEMDIE))) {
836 /* Don't account this! */
842 * We always charge the cgroup the mm_struct belongs to.
843 * The mm_struct's mem_cgroup changes on task migration if the
844 * thread group leader migrates. It's possible that mm is not
845 * set, if so charge the init_mm (happens for pagecache usage).
849 mem = try_get_mem_cgroup_from_mm(mm);
857 VM_BUG_ON(mem_cgroup_is_obsolete(mem));
863 ret = res_counter_charge(&mem->res, PAGE_SIZE, &fail_res);
865 if (!do_swap_account)
867 ret = res_counter_charge(&mem->memsw, PAGE_SIZE,
871 /* mem+swap counter fails */
872 res_counter_uncharge(&mem->res, PAGE_SIZE);
874 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
877 /* mem counter fails */
878 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
881 if (!(gfp_mask & __GFP_WAIT))
884 ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, gfp_mask,
890 * try_to_free_mem_cgroup_pages() might not give us a full
891 * picture of reclaim. Some pages are reclaimed and might be
892 * moved to swap cache or just unmapped from the cgroup.
893 * Check the limit again to see if the reclaim reduced the
894 * current usage of the cgroup before giving up
897 if (mem_cgroup_check_under_limit(mem_over_limit))
902 mutex_lock(&memcg_tasklist);
903 mem_cgroup_out_of_memory(mem_over_limit, gfp_mask);
904 mutex_unlock(&memcg_tasklist);
905 mem_over_limit->last_oom_jiffies = jiffies;
916 static struct mem_cgroup *try_get_mem_cgroup_from_swapcache(struct page *page)
918 struct mem_cgroup *mem;
921 if (!PageSwapCache(page))
924 ent.val = page_private(page);
925 mem = lookup_swap_cgroup(ent);
928 if (!css_tryget(&mem->css))
934 * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be
935 * USED state. If already USED, uncharge and return.
938 static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
939 struct page_cgroup *pc,
940 enum charge_type ctype)
942 /* try_charge() can return NULL to *memcg, taking care of it. */
946 lock_page_cgroup(pc);
947 if (unlikely(PageCgroupUsed(pc))) {
948 unlock_page_cgroup(pc);
949 res_counter_uncharge(&mem->res, PAGE_SIZE);
951 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
955 pc->mem_cgroup = mem;
957 pc->flags = pcg_default_flags[ctype];
959 mem_cgroup_charge_statistics(mem, pc, true);
961 unlock_page_cgroup(pc);
965 * mem_cgroup_move_account - move account of the page
966 * @pc: page_cgroup of the page.
967 * @from: mem_cgroup which the page is moved from.
968 * @to: mem_cgroup which the page is moved to. @from != @to.
970 * The caller must confirm following.
971 * - page is not on LRU (isolate_page() is useful.)
973 * returns 0 at success,
974 * returns -EBUSY when lock is busy or "pc" is unstable.
976 * This function does "uncharge" from old cgroup but doesn't do "charge" to
977 * new cgroup. It should be done by a caller.
980 static int mem_cgroup_move_account(struct page_cgroup *pc,
981 struct mem_cgroup *from, struct mem_cgroup *to)
983 struct mem_cgroup_per_zone *from_mz, *to_mz;
987 VM_BUG_ON(from == to);
988 VM_BUG_ON(PageLRU(pc->page));
990 nid = page_cgroup_nid(pc);
991 zid = page_cgroup_zid(pc);
992 from_mz = mem_cgroup_zoneinfo(from, nid, zid);
993 to_mz = mem_cgroup_zoneinfo(to, nid, zid);
995 if (!trylock_page_cgroup(pc))
998 if (!PageCgroupUsed(pc))
1001 if (pc->mem_cgroup != from)
1004 res_counter_uncharge(&from->res, PAGE_SIZE);
1005 mem_cgroup_charge_statistics(from, pc, false);
1006 if (do_swap_account)
1007 res_counter_uncharge(&from->memsw, PAGE_SIZE);
1008 css_put(&from->css);
1011 pc->mem_cgroup = to;
1012 mem_cgroup_charge_statistics(to, pc, true);
1015 unlock_page_cgroup(pc);
1020 * move charges to its parent.
1023 static int mem_cgroup_move_parent(struct page_cgroup *pc,
1024 struct mem_cgroup *child,
1027 struct page *page = pc->page;
1028 struct cgroup *cg = child->css.cgroup;
1029 struct cgroup *pcg = cg->parent;
1030 struct mem_cgroup *parent;
1038 parent = mem_cgroup_from_cont(pcg);
1041 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false);
1045 if (!get_page_unless_zero(page)) {
1050 ret = isolate_lru_page(page);
1055 ret = mem_cgroup_move_account(pc, child, parent);
1057 putback_lru_page(page);
1060 /* drop extra refcnt by try_charge() */
1061 css_put(&parent->css);
1068 /* drop extra refcnt by try_charge() */
1069 css_put(&parent->css);
1070 /* uncharge if move fails */
1071 res_counter_uncharge(&parent->res, PAGE_SIZE);
1072 if (do_swap_account)
1073 res_counter_uncharge(&parent->memsw, PAGE_SIZE);
1078 * Charge the memory controller for page usage.
1080 * 0 if the charge was successful
1081 * < 0 if the cgroup is over its limit
1083 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
1084 gfp_t gfp_mask, enum charge_type ctype,
1085 struct mem_cgroup *memcg)
1087 struct mem_cgroup *mem;
1088 struct page_cgroup *pc;
1091 pc = lookup_page_cgroup(page);
1092 /* can happen at boot */
1098 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true);
1102 __mem_cgroup_commit_charge(mem, pc, ctype);
1106 int mem_cgroup_newpage_charge(struct page *page,
1107 struct mm_struct *mm, gfp_t gfp_mask)
1109 if (mem_cgroup_disabled())
1111 if (PageCompound(page))
1114 * If already mapped, we don't have to account.
1115 * If page cache, page->mapping has address_space.
1116 * But page->mapping may have out-of-use anon_vma pointer,
1117 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
1120 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
1124 return mem_cgroup_charge_common(page, mm, gfp_mask,
1125 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
1128 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
1131 struct mem_cgroup *mem = NULL;
1134 if (mem_cgroup_disabled())
1136 if (PageCompound(page))
1139 * Corner case handling. This is called from add_to_page_cache()
1140 * in usual. But some FS (shmem) precharges this page before calling it
1141 * and call add_to_page_cache() with GFP_NOWAIT.
1143 * For GFP_NOWAIT case, the page may be pre-charged before calling
1144 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
1145 * charge twice. (It works but has to pay a bit larger cost.)
1146 * And when the page is SwapCache, it should take swap information
1147 * into account. This is under lock_page() now.
1149 if (!(gfp_mask & __GFP_WAIT)) {
1150 struct page_cgroup *pc;
1153 pc = lookup_page_cgroup(page);
1156 lock_page_cgroup(pc);
1157 if (PageCgroupUsed(pc)) {
1158 unlock_page_cgroup(pc);
1161 unlock_page_cgroup(pc);
1164 if (do_swap_account && PageSwapCache(page)) {
1165 mem = try_get_mem_cgroup_from_swapcache(page);
1170 /* SwapCache may be still linked to LRU now. */
1171 mem_cgroup_lru_del_before_commit_swapcache(page);
1174 if (unlikely(!mm && !mem))
1177 if (page_is_file_cache(page))
1178 return mem_cgroup_charge_common(page, mm, gfp_mask,
1179 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
1181 ret = mem_cgroup_charge_common(page, mm, gfp_mask,
1182 MEM_CGROUP_CHARGE_TYPE_SHMEM, mem);
1185 if (PageSwapCache(page))
1186 mem_cgroup_lru_add_after_commit_swapcache(page);
1188 if (do_swap_account && !ret && PageSwapCache(page)) {
1189 swp_entry_t ent = {.val = page_private(page)};
1190 /* avoid double counting */
1191 mem = swap_cgroup_record(ent, NULL);
1193 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1194 mem_cgroup_put(mem);
1201 * While swap-in, try_charge -> commit or cancel, the page is locked.
1202 * And when try_charge() successfully returns, one refcnt to memcg without
1203 * struct page_cgroup is aquired. This refcnt will be cumsumed by
1204 * "commit()" or removed by "cancel()"
1206 int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
1208 gfp_t mask, struct mem_cgroup **ptr)
1210 struct mem_cgroup *mem;
1213 if (mem_cgroup_disabled())
1216 if (!do_swap_account)
1219 * A racing thread's fault, or swapoff, may have already updated
1220 * the pte, and even removed page from swap cache: return success
1221 * to go on to do_swap_page()'s pte_same() test, which should fail.
1223 if (!PageSwapCache(page))
1225 mem = try_get_mem_cgroup_from_swapcache(page);
1229 ret = __mem_cgroup_try_charge(NULL, mask, ptr, true);
1230 /* drop extra refcnt from tryget */
1236 return __mem_cgroup_try_charge(mm, mask, ptr, true);
1239 void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
1241 struct page_cgroup *pc;
1243 if (mem_cgroup_disabled())
1247 pc = lookup_page_cgroup(page);
1248 mem_cgroup_lru_del_before_commit_swapcache(page);
1249 __mem_cgroup_commit_charge(ptr, pc, MEM_CGROUP_CHARGE_TYPE_MAPPED);
1250 mem_cgroup_lru_add_after_commit_swapcache(page);
1252 * Now swap is on-memory. This means this page may be
1253 * counted both as mem and swap....double count.
1254 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
1255 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
1256 * may call delete_from_swap_cache() before reach here.
1258 if (do_swap_account && PageSwapCache(page)) {
1259 swp_entry_t ent = {.val = page_private(page)};
1260 struct mem_cgroup *memcg;
1261 memcg = swap_cgroup_record(ent, NULL);
1263 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1264 mem_cgroup_put(memcg);
1268 /* add this page(page_cgroup) to the LRU we want. */
1272 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
1274 if (mem_cgroup_disabled())
1278 res_counter_uncharge(&mem->res, PAGE_SIZE);
1279 if (do_swap_account)
1280 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1286 * uncharge if !page_mapped(page)
1288 static struct mem_cgroup *
1289 __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
1291 struct page_cgroup *pc;
1292 struct mem_cgroup *mem = NULL;
1293 struct mem_cgroup_per_zone *mz;
1295 if (mem_cgroup_disabled())
1298 if (PageSwapCache(page))
1302 * Check if our page_cgroup is valid
1304 pc = lookup_page_cgroup(page);
1305 if (unlikely(!pc || !PageCgroupUsed(pc)))
1308 lock_page_cgroup(pc);
1310 mem = pc->mem_cgroup;
1312 if (!PageCgroupUsed(pc))
1316 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
1317 if (page_mapped(page))
1320 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
1321 if (!PageAnon(page)) { /* Shared memory */
1322 if (page->mapping && !page_is_file_cache(page))
1324 } else if (page_mapped(page)) /* Anon */
1331 res_counter_uncharge(&mem->res, PAGE_SIZE);
1332 if (do_swap_account && (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT))
1333 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1334 mem_cgroup_charge_statistics(mem, pc, false);
1336 ClearPageCgroupUsed(pc);
1338 * pc->mem_cgroup is not cleared here. It will be accessed when it's
1339 * freed from LRU. This is safe because uncharged page is expected not
1340 * to be reused (freed soon). Exception is SwapCache, it's handled by
1341 * special functions.
1344 mz = page_cgroup_zoneinfo(pc);
1345 unlock_page_cgroup(pc);
1347 /* at swapout, this memcg will be accessed to record to swap */
1348 if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
1354 unlock_page_cgroup(pc);
1358 void mem_cgroup_uncharge_page(struct page *page)
1361 if (page_mapped(page))
1363 if (page->mapping && !PageAnon(page))
1365 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
1368 void mem_cgroup_uncharge_cache_page(struct page *page)
1370 VM_BUG_ON(page_mapped(page));
1371 VM_BUG_ON(page->mapping);
1372 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
1376 * called from __delete_from_swap_cache() and drop "page" account.
1377 * memcg information is recorded to swap_cgroup of "ent"
1379 void mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent)
1381 struct mem_cgroup *memcg;
1383 memcg = __mem_cgroup_uncharge_common(page,
1384 MEM_CGROUP_CHARGE_TYPE_SWAPOUT);
1385 /* record memcg information */
1386 if (do_swap_account && memcg) {
1387 swap_cgroup_record(ent, memcg);
1388 mem_cgroup_get(memcg);
1391 css_put(&memcg->css);
1394 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1396 * called from swap_entry_free(). remove record in swap_cgroup and
1397 * uncharge "memsw" account.
1399 void mem_cgroup_uncharge_swap(swp_entry_t ent)
1401 struct mem_cgroup *memcg;
1403 if (!do_swap_account)
1406 memcg = swap_cgroup_record(ent, NULL);
1408 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1409 mem_cgroup_put(memcg);
1415 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
1418 int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr)
1420 struct page_cgroup *pc;
1421 struct mem_cgroup *mem = NULL;
1424 if (mem_cgroup_disabled())
1427 pc = lookup_page_cgroup(page);
1428 lock_page_cgroup(pc);
1429 if (PageCgroupUsed(pc)) {
1430 mem = pc->mem_cgroup;
1433 unlock_page_cgroup(pc);
1436 ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false);
1443 /* remove redundant charge if migration failed*/
1444 void mem_cgroup_end_migration(struct mem_cgroup *mem,
1445 struct page *oldpage, struct page *newpage)
1447 struct page *target, *unused;
1448 struct page_cgroup *pc;
1449 enum charge_type ctype;
1454 /* at migration success, oldpage->mapping is NULL. */
1455 if (oldpage->mapping) {
1463 if (PageAnon(target))
1464 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
1465 else if (page_is_file_cache(target))
1466 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
1468 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
1470 /* unused page is not on radix-tree now. */
1472 __mem_cgroup_uncharge_common(unused, ctype);
1474 pc = lookup_page_cgroup(target);
1476 * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup.
1477 * So, double-counting is effectively avoided.
1479 __mem_cgroup_commit_charge(mem, pc, ctype);
1482 * Both of oldpage and newpage are still under lock_page().
1483 * Then, we don't have to care about race in radix-tree.
1484 * But we have to be careful that this page is unmapped or not.
1486 * There is a case for !page_mapped(). At the start of
1487 * migration, oldpage was mapped. But now, it's zapped.
1488 * But we know *target* page is not freed/reused under us.
1489 * mem_cgroup_uncharge_page() does all necessary checks.
1491 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
1492 mem_cgroup_uncharge_page(target);
1496 * A call to try to shrink memory usage under specified resource controller.
1497 * This is typically used for page reclaiming for shmem for reducing side
1498 * effect of page allocation from shmem, which is used by some mem_cgroup.
1500 int mem_cgroup_shrink_usage(struct page *page,
1501 struct mm_struct *mm,
1504 struct mem_cgroup *mem = NULL;
1506 int retry = MEM_CGROUP_RECLAIM_RETRIES;
1508 if (mem_cgroup_disabled())
1511 mem = try_get_mem_cgroup_from_swapcache(page);
1513 mem = try_get_mem_cgroup_from_mm(mm);
1518 progress = mem_cgroup_hierarchical_reclaim(mem,
1519 gfp_mask, true, false);
1520 progress += mem_cgroup_check_under_limit(mem);
1521 } while (!progress && --retry);
1529 static DEFINE_MUTEX(set_limit_mutex);
1531 static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
1532 unsigned long long val)
1538 int children = mem_cgroup_count_children(memcg);
1539 u64 curusage, oldusage;
1542 * For keeping hierarchical_reclaim simple, how long we should retry
1543 * is depends on callers. We set our retry-count to be function
1544 * of # of children which we should visit in this loop.
1546 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
1548 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
1550 while (retry_count) {
1551 if (signal_pending(current)) {
1556 * Rather than hide all in some function, I do this in
1557 * open coded manner. You see what this really does.
1558 * We have to guarantee mem->res.limit < mem->memsw.limit.
1560 mutex_lock(&set_limit_mutex);
1561 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1562 if (memswlimit < val) {
1564 mutex_unlock(&set_limit_mutex);
1567 ret = res_counter_set_limit(&memcg->res, val);
1568 mutex_unlock(&set_limit_mutex);
1573 progress = mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL,
1575 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
1576 /* Usage is reduced ? */
1577 if (curusage >= oldusage)
1580 oldusage = curusage;
1586 int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
1587 unsigned long long val)
1590 u64 memlimit, oldusage, curusage;
1591 int children = mem_cgroup_count_children(memcg);
1594 if (!do_swap_account)
1596 /* see mem_cgroup_resize_res_limit */
1597 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
1598 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
1599 while (retry_count) {
1600 if (signal_pending(current)) {
1605 * Rather than hide all in some function, I do this in
1606 * open coded manner. You see what this really does.
1607 * We have to guarantee mem->res.limit < mem->memsw.limit.
1609 mutex_lock(&set_limit_mutex);
1610 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1611 if (memlimit > val) {
1613 mutex_unlock(&set_limit_mutex);
1616 ret = res_counter_set_limit(&memcg->memsw, val);
1617 mutex_unlock(&set_limit_mutex);
1622 mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL, true, true);
1623 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
1624 /* Usage is reduced ? */
1625 if (curusage >= oldusage)
1628 oldusage = curusage;
1634 * This routine traverse page_cgroup in given list and drop them all.
1635 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
1637 static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
1638 int node, int zid, enum lru_list lru)
1641 struct mem_cgroup_per_zone *mz;
1642 struct page_cgroup *pc, *busy;
1643 unsigned long flags, loop;
1644 struct list_head *list;
1647 zone = &NODE_DATA(node)->node_zones[zid];
1648 mz = mem_cgroup_zoneinfo(mem, node, zid);
1649 list = &mz->lists[lru];
1651 loop = MEM_CGROUP_ZSTAT(mz, lru);
1652 /* give some margin against EBUSY etc...*/
1657 spin_lock_irqsave(&zone->lru_lock, flags);
1658 if (list_empty(list)) {
1659 spin_unlock_irqrestore(&zone->lru_lock, flags);
1662 pc = list_entry(list->prev, struct page_cgroup, lru);
1664 list_move(&pc->lru, list);
1666 spin_unlock_irqrestore(&zone->lru_lock, flags);
1669 spin_unlock_irqrestore(&zone->lru_lock, flags);
1671 ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL);
1675 if (ret == -EBUSY || ret == -EINVAL) {
1676 /* found lock contention or "pc" is obsolete. */
1683 if (!ret && !list_empty(list))
1689 * make mem_cgroup's charge to be 0 if there is no task.
1690 * This enables deleting this mem_cgroup.
1692 static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
1695 int node, zid, shrink;
1696 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
1697 struct cgroup *cgrp = mem->css.cgroup;
1702 /* should free all ? */
1706 while (mem->res.usage > 0) {
1708 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
1711 if (signal_pending(current))
1713 /* This is for making all *used* pages to be on LRU. */
1714 lru_add_drain_all();
1716 for_each_node_state(node, N_HIGH_MEMORY) {
1717 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
1720 ret = mem_cgroup_force_empty_list(mem,
1729 /* it seems parent cgroup doesn't have enough mem */
1740 /* returns EBUSY if there is a task or if we come here twice. */
1741 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
1745 /* we call try-to-free pages for make this cgroup empty */
1746 lru_add_drain_all();
1747 /* try to free all pages in this cgroup */
1749 while (nr_retries && mem->res.usage > 0) {
1752 if (signal_pending(current)) {
1756 progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL,
1757 false, get_swappiness(mem));
1760 /* maybe some writeback is necessary */
1761 congestion_wait(WRITE, HZ/10);
1766 /* try move_account...there may be some *locked* pages. */
1773 int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
1775 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
1779 static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
1781 return mem_cgroup_from_cont(cont)->use_hierarchy;
1784 static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
1788 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1789 struct cgroup *parent = cont->parent;
1790 struct mem_cgroup *parent_mem = NULL;
1793 parent_mem = mem_cgroup_from_cont(parent);
1797 * If parent's use_hiearchy is set, we can't make any modifications
1798 * in the child subtrees. If it is unset, then the change can
1799 * occur, provided the current cgroup has no children.
1801 * For the root cgroup, parent_mem is NULL, we allow value to be
1802 * set if there are no children.
1804 if ((!parent_mem || !parent_mem->use_hierarchy) &&
1805 (val == 1 || val == 0)) {
1806 if (list_empty(&cont->children))
1807 mem->use_hierarchy = val;
1817 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
1819 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1823 type = MEMFILE_TYPE(cft->private);
1824 name = MEMFILE_ATTR(cft->private);
1827 val = res_counter_read_u64(&mem->res, name);
1830 if (do_swap_account)
1831 val = res_counter_read_u64(&mem->memsw, name);
1840 * The user of this function is...
1843 static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
1846 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
1848 unsigned long long val;
1851 type = MEMFILE_TYPE(cft->private);
1852 name = MEMFILE_ATTR(cft->private);
1855 /* This function does all necessary parse...reuse it */
1856 ret = res_counter_memparse_write_strategy(buffer, &val);
1860 ret = mem_cgroup_resize_limit(memcg, val);
1862 ret = mem_cgroup_resize_memsw_limit(memcg, val);
1865 ret = -EINVAL; /* should be BUG() ? */
1871 static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
1872 unsigned long long *mem_limit, unsigned long long *memsw_limit)
1874 struct cgroup *cgroup;
1875 unsigned long long min_limit, min_memsw_limit, tmp;
1877 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1878 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1879 cgroup = memcg->css.cgroup;
1880 if (!memcg->use_hierarchy)
1883 while (cgroup->parent) {
1884 cgroup = cgroup->parent;
1885 memcg = mem_cgroup_from_cont(cgroup);
1886 if (!memcg->use_hierarchy)
1888 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
1889 min_limit = min(min_limit, tmp);
1890 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1891 min_memsw_limit = min(min_memsw_limit, tmp);
1894 *mem_limit = min_limit;
1895 *memsw_limit = min_memsw_limit;
1899 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
1901 struct mem_cgroup *mem;
1904 mem = mem_cgroup_from_cont(cont);
1905 type = MEMFILE_TYPE(event);
1906 name = MEMFILE_ATTR(event);
1910 res_counter_reset_max(&mem->res);
1912 res_counter_reset_max(&mem->memsw);
1916 res_counter_reset_failcnt(&mem->res);
1918 res_counter_reset_failcnt(&mem->memsw);
1925 /* For read statistics */
1939 struct mcs_total_stat {
1940 s64 stat[NR_MCS_STAT];
1946 } memcg_stat_strings[NR_MCS_STAT] = {
1947 {"cache", "total_cache"},
1948 {"rss", "total_rss"},
1949 {"pgpgin", "total_pgpgin"},
1950 {"pgpgout", "total_pgpgout"},
1951 {"inactive_anon", "total_inactive_anon"},
1952 {"active_anon", "total_active_anon"},
1953 {"inactive_file", "total_inactive_file"},
1954 {"active_file", "total_active_file"},
1955 {"unevictable", "total_unevictable"}
1959 static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data)
1961 struct mcs_total_stat *s = data;
1965 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_CACHE);
1966 s->stat[MCS_CACHE] += val * PAGE_SIZE;
1967 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
1968 s->stat[MCS_RSS] += val * PAGE_SIZE;
1969 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGIN_COUNT);
1970 s->stat[MCS_PGPGIN] += val;
1971 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGOUT_COUNT);
1972 s->stat[MCS_PGPGOUT] += val;
1975 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON);
1976 s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE;
1977 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON);
1978 s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE;
1979 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE);
1980 s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE;
1981 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE);
1982 s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE;
1983 val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE);
1984 s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE;
1989 mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s)
1991 mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat);
1994 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
1995 struct cgroup_map_cb *cb)
1997 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
1998 struct mcs_total_stat mystat;
2001 memset(&mystat, 0, sizeof(mystat));
2002 mem_cgroup_get_local_stat(mem_cont, &mystat);
2004 for (i = 0; i < NR_MCS_STAT; i++)
2005 cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
2007 /* Hierarchical information */
2009 unsigned long long limit, memsw_limit;
2010 memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit);
2011 cb->fill(cb, "hierarchical_memory_limit", limit);
2012 if (do_swap_account)
2013 cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
2016 memset(&mystat, 0, sizeof(mystat));
2017 mem_cgroup_get_total_stat(mem_cont, &mystat);
2018 for (i = 0; i < NR_MCS_STAT; i++)
2019 cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
2022 #ifdef CONFIG_DEBUG_VM
2023 cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL));
2027 struct mem_cgroup_per_zone *mz;
2028 unsigned long recent_rotated[2] = {0, 0};
2029 unsigned long recent_scanned[2] = {0, 0};
2031 for_each_online_node(nid)
2032 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
2033 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
2035 recent_rotated[0] +=
2036 mz->reclaim_stat.recent_rotated[0];
2037 recent_rotated[1] +=
2038 mz->reclaim_stat.recent_rotated[1];
2039 recent_scanned[0] +=
2040 mz->reclaim_stat.recent_scanned[0];
2041 recent_scanned[1] +=
2042 mz->reclaim_stat.recent_scanned[1];
2044 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
2045 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
2046 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
2047 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
2054 static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
2056 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
2058 return get_swappiness(memcg);
2061 static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
2064 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
2065 struct mem_cgroup *parent;
2070 if (cgrp->parent == NULL)
2073 parent = mem_cgroup_from_cont(cgrp->parent);
2077 /* If under hierarchy, only empty-root can set this value */
2078 if ((parent->use_hierarchy) ||
2079 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
2084 spin_lock(&memcg->reclaim_param_lock);
2085 memcg->swappiness = val;
2086 spin_unlock(&memcg->reclaim_param_lock);
2094 static struct cftype mem_cgroup_files[] = {
2096 .name = "usage_in_bytes",
2097 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
2098 .read_u64 = mem_cgroup_read,
2101 .name = "max_usage_in_bytes",
2102 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
2103 .trigger = mem_cgroup_reset,
2104 .read_u64 = mem_cgroup_read,
2107 .name = "limit_in_bytes",
2108 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
2109 .write_string = mem_cgroup_write,
2110 .read_u64 = mem_cgroup_read,
2114 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
2115 .trigger = mem_cgroup_reset,
2116 .read_u64 = mem_cgroup_read,
2120 .read_map = mem_control_stat_show,
2123 .name = "force_empty",
2124 .trigger = mem_cgroup_force_empty_write,
2127 .name = "use_hierarchy",
2128 .write_u64 = mem_cgroup_hierarchy_write,
2129 .read_u64 = mem_cgroup_hierarchy_read,
2132 .name = "swappiness",
2133 .read_u64 = mem_cgroup_swappiness_read,
2134 .write_u64 = mem_cgroup_swappiness_write,
2138 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2139 static struct cftype memsw_cgroup_files[] = {
2141 .name = "memsw.usage_in_bytes",
2142 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
2143 .read_u64 = mem_cgroup_read,
2146 .name = "memsw.max_usage_in_bytes",
2147 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
2148 .trigger = mem_cgroup_reset,
2149 .read_u64 = mem_cgroup_read,
2152 .name = "memsw.limit_in_bytes",
2153 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
2154 .write_string = mem_cgroup_write,
2155 .read_u64 = mem_cgroup_read,
2158 .name = "memsw.failcnt",
2159 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
2160 .trigger = mem_cgroup_reset,
2161 .read_u64 = mem_cgroup_read,
2165 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
2167 if (!do_swap_account)
2169 return cgroup_add_files(cont, ss, memsw_cgroup_files,
2170 ARRAY_SIZE(memsw_cgroup_files));
2173 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
2179 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
2181 struct mem_cgroup_per_node *pn;
2182 struct mem_cgroup_per_zone *mz;
2184 int zone, tmp = node;
2186 * This routine is called against possible nodes.
2187 * But it's BUG to call kmalloc() against offline node.
2189 * TODO: this routine can waste much memory for nodes which will
2190 * never be onlined. It's better to use memory hotplug callback
2193 if (!node_state(node, N_NORMAL_MEMORY))
2195 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
2199 mem->info.nodeinfo[node] = pn;
2200 memset(pn, 0, sizeof(*pn));
2202 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
2203 mz = &pn->zoneinfo[zone];
2205 INIT_LIST_HEAD(&mz->lists[l]);
2210 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
2212 kfree(mem->info.nodeinfo[node]);
2215 static int mem_cgroup_size(void)
2217 int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu);
2218 return sizeof(struct mem_cgroup) + cpustat_size;
2221 static struct mem_cgroup *mem_cgroup_alloc(void)
2223 struct mem_cgroup *mem;
2224 int size = mem_cgroup_size();
2226 if (size < PAGE_SIZE)
2227 mem = kmalloc(size, GFP_KERNEL);
2229 mem = vmalloc(size);
2232 memset(mem, 0, size);
2237 * At destroying mem_cgroup, references from swap_cgroup can remain.
2238 * (scanning all at force_empty is too costly...)
2240 * Instead of clearing all references at force_empty, we remember
2241 * the number of reference from swap_cgroup and free mem_cgroup when
2242 * it goes down to 0.
2244 * Removal of cgroup itself succeeds regardless of refs from swap.
2247 static void __mem_cgroup_free(struct mem_cgroup *mem)
2251 free_css_id(&mem_cgroup_subsys, &mem->css);
2253 for_each_node_state(node, N_POSSIBLE)
2254 free_mem_cgroup_per_zone_info(mem, node);
2256 if (mem_cgroup_size() < PAGE_SIZE)
2262 static void mem_cgroup_get(struct mem_cgroup *mem)
2264 atomic_inc(&mem->refcnt);
2267 static void mem_cgroup_put(struct mem_cgroup *mem)
2269 if (atomic_dec_and_test(&mem->refcnt)) {
2270 struct mem_cgroup *parent = parent_mem_cgroup(mem);
2271 __mem_cgroup_free(mem);
2273 mem_cgroup_put(parent);
2278 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
2280 static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem)
2282 if (!mem->res.parent)
2284 return mem_cgroup_from_res_counter(mem->res.parent, res);
2287 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2288 static void __init enable_swap_cgroup(void)
2290 if (!mem_cgroup_disabled() && really_do_swap_account)
2291 do_swap_account = 1;
2294 static void __init enable_swap_cgroup(void)
2299 static struct cgroup_subsys_state * __ref
2300 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
2302 struct mem_cgroup *mem, *parent;
2303 long error = -ENOMEM;
2306 mem = mem_cgroup_alloc();
2308 return ERR_PTR(error);
2310 for_each_node_state(node, N_POSSIBLE)
2311 if (alloc_mem_cgroup_per_zone_info(mem, node))
2314 if (cont->parent == NULL) {
2315 enable_swap_cgroup();
2318 parent = mem_cgroup_from_cont(cont->parent);
2319 mem->use_hierarchy = parent->use_hierarchy;
2322 if (parent && parent->use_hierarchy) {
2323 res_counter_init(&mem->res, &parent->res);
2324 res_counter_init(&mem->memsw, &parent->memsw);
2326 * We increment refcnt of the parent to ensure that we can
2327 * safely access it on res_counter_charge/uncharge.
2328 * This refcnt will be decremented when freeing this
2329 * mem_cgroup(see mem_cgroup_put).
2331 mem_cgroup_get(parent);
2333 res_counter_init(&mem->res, NULL);
2334 res_counter_init(&mem->memsw, NULL);
2336 mem->last_scanned_child = 0;
2337 spin_lock_init(&mem->reclaim_param_lock);
2340 mem->swappiness = get_swappiness(parent);
2341 atomic_set(&mem->refcnt, 1);
2344 __mem_cgroup_free(mem);
2345 return ERR_PTR(error);
2348 static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
2349 struct cgroup *cont)
2351 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2353 return mem_cgroup_force_empty(mem, false);
2356 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
2357 struct cgroup *cont)
2359 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2361 mem_cgroup_put(mem);
2364 static int mem_cgroup_populate(struct cgroup_subsys *ss,
2365 struct cgroup *cont)
2369 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
2370 ARRAY_SIZE(mem_cgroup_files));
2373 ret = register_memsw_files(cont, ss);
2377 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
2378 struct cgroup *cont,
2379 struct cgroup *old_cont,
2380 struct task_struct *p)
2382 mutex_lock(&memcg_tasklist);
2384 * FIXME: It's better to move charges of this process from old
2385 * memcg to new memcg. But it's just on TODO-List now.
2387 mutex_unlock(&memcg_tasklist);
2390 struct cgroup_subsys mem_cgroup_subsys = {
2392 .subsys_id = mem_cgroup_subsys_id,
2393 .create = mem_cgroup_create,
2394 .pre_destroy = mem_cgroup_pre_destroy,
2395 .destroy = mem_cgroup_destroy,
2396 .populate = mem_cgroup_populate,
2397 .attach = mem_cgroup_move_task,
2402 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2404 static int __init disable_swap_account(char *s)
2406 really_do_swap_account = 0;
2409 __setup("noswapaccount", disable_swap_account);