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/limits.h>
31 #include <linux/mutex.h>
32 #include <linux/slab.h>
33 #include <linux/swap.h>
34 #include <linux/spinlock.h>
36 #include <linux/seq_file.h>
37 #include <linux/vmalloc.h>
38 #include <linux/mm_inline.h>
39 #include <linux/page_cgroup.h>
42 #include <asm/uaccess.h>
44 struct cgroup_subsys mem_cgroup_subsys __read_mostly;
45 #define MEM_CGROUP_RECLAIM_RETRIES 5
47 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
48 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 0 */
49 int do_swap_account __read_mostly;
50 static int really_do_swap_account __initdata = 1; /* for remember boot option*/
52 #define do_swap_account (0)
55 static DEFINE_MUTEX(memcg_tasklist); /* can be hold under cgroup_mutex */
58 * Statistics for memory cgroup.
60 enum mem_cgroup_stat_index {
62 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
64 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
65 MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */
66 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
67 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
69 MEM_CGROUP_STAT_NSTATS,
72 struct mem_cgroup_stat_cpu {
73 s64 count[MEM_CGROUP_STAT_NSTATS];
74 } ____cacheline_aligned_in_smp;
76 struct mem_cgroup_stat {
77 struct mem_cgroup_stat_cpu cpustat[0];
81 * For accounting under irq disable, no need for increment preempt count.
83 static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat,
84 enum mem_cgroup_stat_index idx, int val)
86 stat->count[idx] += val;
89 static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat,
90 enum mem_cgroup_stat_index idx)
94 for_each_possible_cpu(cpu)
95 ret += stat->cpustat[cpu].count[idx];
99 static s64 mem_cgroup_local_usage(struct mem_cgroup_stat *stat)
103 ret = mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_CACHE);
104 ret += mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_RSS);
109 * per-zone information in memory controller.
111 struct mem_cgroup_per_zone {
113 * spin_lock to protect the per cgroup LRU
115 struct list_head lists[NR_LRU_LISTS];
116 unsigned long count[NR_LRU_LISTS];
118 struct zone_reclaim_stat reclaim_stat;
120 /* Macro for accessing counter */
121 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
123 struct mem_cgroup_per_node {
124 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
127 struct mem_cgroup_lru_info {
128 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
132 * The memory controller data structure. The memory controller controls both
133 * page cache and RSS per cgroup. We would eventually like to provide
134 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
135 * to help the administrator determine what knobs to tune.
137 * TODO: Add a water mark for the memory controller. Reclaim will begin when
138 * we hit the water mark. May be even add a low water mark, such that
139 * no reclaim occurs from a cgroup at it's low water mark, this is
140 * a feature that will be implemented much later in the future.
143 struct cgroup_subsys_state css;
145 * the counter to account for memory usage
147 struct res_counter res;
149 * the counter to account for mem+swap usage.
151 struct res_counter memsw;
153 * Per cgroup active and inactive list, similar to the
154 * per zone LRU lists.
156 struct mem_cgroup_lru_info info;
159 protect against reclaim related member.
161 spinlock_t reclaim_param_lock;
163 int prev_priority; /* for recording reclaim priority */
166 * While reclaiming in a hiearchy, we cache the last child we
169 int last_scanned_child;
171 * Should the accounting and control be hierarchical, per subtree?
174 unsigned long last_oom_jiffies;
177 unsigned int swappiness;
180 * statistics. This must be placed at the end of memcg.
182 struct mem_cgroup_stat stat;
186 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
187 MEM_CGROUP_CHARGE_TYPE_MAPPED,
188 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
189 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
190 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
194 /* only for here (for easy reading.) */
195 #define PCGF_CACHE (1UL << PCG_CACHE)
196 #define PCGF_USED (1UL << PCG_USED)
197 #define PCGF_LOCK (1UL << PCG_LOCK)
198 static const unsigned long
199 pcg_default_flags[NR_CHARGE_TYPE] = {
200 PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* File Cache */
201 PCGF_USED | PCGF_LOCK, /* Anon */
202 PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* Shmem */
206 /* for encoding cft->private value on file */
209 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
210 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
211 #define MEMFILE_ATTR(val) ((val) & 0xffff)
213 static void mem_cgroup_get(struct mem_cgroup *mem);
214 static void mem_cgroup_put(struct mem_cgroup *mem);
215 static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem);
217 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
218 struct page_cgroup *pc,
221 int val = (charge)? 1 : -1;
222 struct mem_cgroup_stat *stat = &mem->stat;
223 struct mem_cgroup_stat_cpu *cpustat;
226 cpustat = &stat->cpustat[cpu];
227 if (PageCgroupCache(pc))
228 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val);
230 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val);
233 __mem_cgroup_stat_add_safe(cpustat,
234 MEM_CGROUP_STAT_PGPGIN_COUNT, 1);
236 __mem_cgroup_stat_add_safe(cpustat,
237 MEM_CGROUP_STAT_PGPGOUT_COUNT, 1);
241 static struct mem_cgroup_per_zone *
242 mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
244 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
247 static struct mem_cgroup_per_zone *
248 page_cgroup_zoneinfo(struct page_cgroup *pc)
250 struct mem_cgroup *mem = pc->mem_cgroup;
251 int nid = page_cgroup_nid(pc);
252 int zid = page_cgroup_zid(pc);
257 return mem_cgroup_zoneinfo(mem, nid, zid);
260 static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem,
264 struct mem_cgroup_per_zone *mz;
267 for_each_online_node(nid)
268 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
269 mz = mem_cgroup_zoneinfo(mem, nid, zid);
270 total += MEM_CGROUP_ZSTAT(mz, idx);
275 static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
277 return container_of(cgroup_subsys_state(cont,
278 mem_cgroup_subsys_id), struct mem_cgroup,
282 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
285 * mm_update_next_owner() may clear mm->owner to NULL
286 * if it races with swapoff, page migration, etc.
287 * So this can be called with p == NULL.
292 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
293 struct mem_cgroup, css);
296 static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
298 struct mem_cgroup *mem = NULL;
303 * Because we have no locks, mm->owner's may be being moved to other
304 * cgroup. We use css_tryget() here even if this looks
305 * pessimistic (rather than adding locks here).
309 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
312 } while (!css_tryget(&mem->css));
317 static bool mem_cgroup_is_obsolete(struct mem_cgroup *mem)
321 return css_is_removed(&mem->css);
326 * Call callback function against all cgroup under hierarchy tree.
328 static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data,
329 int (*func)(struct mem_cgroup *, void *))
331 int found, ret, nextid;
332 struct cgroup_subsys_state *css;
333 struct mem_cgroup *mem;
335 if (!root->use_hierarchy)
336 return (*func)(root, data);
344 css = css_get_next(&mem_cgroup_subsys, nextid, &root->css,
346 if (css && css_tryget(css))
347 mem = container_of(css, struct mem_cgroup, css);
351 ret = (*func)(mem, data);
355 } while (!ret && css);
361 * Following LRU functions are allowed to be used without PCG_LOCK.
362 * Operations are called by routine of global LRU independently from memcg.
363 * What we have to take care of here is validness of pc->mem_cgroup.
365 * Changes to pc->mem_cgroup happens when
368 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
369 * It is added to LRU before charge.
370 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
371 * When moving account, the page is not on LRU. It's isolated.
374 void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
376 struct page_cgroup *pc;
377 struct mem_cgroup *mem;
378 struct mem_cgroup_per_zone *mz;
380 if (mem_cgroup_disabled())
382 pc = lookup_page_cgroup(page);
383 /* can happen while we handle swapcache. */
384 if (list_empty(&pc->lru) || !pc->mem_cgroup)
387 * We don't check PCG_USED bit. It's cleared when the "page" is finally
388 * removed from global LRU.
390 mz = page_cgroup_zoneinfo(pc);
391 mem = pc->mem_cgroup;
392 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
393 list_del_init(&pc->lru);
397 void mem_cgroup_del_lru(struct page *page)
399 mem_cgroup_del_lru_list(page, page_lru(page));
402 void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
404 struct mem_cgroup_per_zone *mz;
405 struct page_cgroup *pc;
407 if (mem_cgroup_disabled())
410 pc = lookup_page_cgroup(page);
412 * Used bit is set without atomic ops but after smp_wmb().
413 * For making pc->mem_cgroup visible, insert smp_rmb() here.
416 /* unused page is not rotated. */
417 if (!PageCgroupUsed(pc))
419 mz = page_cgroup_zoneinfo(pc);
420 list_move(&pc->lru, &mz->lists[lru]);
423 void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
425 struct page_cgroup *pc;
426 struct mem_cgroup_per_zone *mz;
428 if (mem_cgroup_disabled())
430 pc = lookup_page_cgroup(page);
432 * Used bit is set without atomic ops but after smp_wmb().
433 * For making pc->mem_cgroup visible, insert smp_rmb() here.
436 if (!PageCgroupUsed(pc))
439 mz = page_cgroup_zoneinfo(pc);
440 MEM_CGROUP_ZSTAT(mz, lru) += 1;
441 list_add(&pc->lru, &mz->lists[lru]);
445 * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to
446 * lru because the page may.be reused after it's fully uncharged (because of
447 * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge
448 * it again. This function is only used to charge SwapCache. It's done under
449 * lock_page and expected that zone->lru_lock is never held.
451 static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page)
454 struct zone *zone = page_zone(page);
455 struct page_cgroup *pc = lookup_page_cgroup(page);
457 spin_lock_irqsave(&zone->lru_lock, flags);
459 * Forget old LRU when this page_cgroup is *not* used. This Used bit
460 * is guarded by lock_page() because the page is SwapCache.
462 if (!PageCgroupUsed(pc))
463 mem_cgroup_del_lru_list(page, page_lru(page));
464 spin_unlock_irqrestore(&zone->lru_lock, flags);
467 static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page)
470 struct zone *zone = page_zone(page);
471 struct page_cgroup *pc = lookup_page_cgroup(page);
473 spin_lock_irqsave(&zone->lru_lock, flags);
474 /* link when the page is linked to LRU but page_cgroup isn't */
475 if (PageLRU(page) && list_empty(&pc->lru))
476 mem_cgroup_add_lru_list(page, page_lru(page));
477 spin_unlock_irqrestore(&zone->lru_lock, flags);
481 void mem_cgroup_move_lists(struct page *page,
482 enum lru_list from, enum lru_list to)
484 if (mem_cgroup_disabled())
486 mem_cgroup_del_lru_list(page, from);
487 mem_cgroup_add_lru_list(page, to);
490 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
493 struct mem_cgroup *curr = NULL;
497 curr = try_get_mem_cgroup_from_mm(task->mm);
502 if (curr->use_hierarchy)
503 ret = css_is_ancestor(&curr->css, &mem->css);
511 * Calculate mapped_ratio under memory controller. This will be used in
512 * vmscan.c for deteremining we have to reclaim mapped pages.
514 int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem)
519 * usage is recorded in bytes. But, here, we assume the number of
520 * physical pages can be represented by "long" on any arch.
522 total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L;
523 rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
524 return (int)((rss * 100L) / total);
528 * prev_priority control...this will be used in memory reclaim path.
530 int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
534 spin_lock(&mem->reclaim_param_lock);
535 prev_priority = mem->prev_priority;
536 spin_unlock(&mem->reclaim_param_lock);
538 return prev_priority;
541 void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
543 spin_lock(&mem->reclaim_param_lock);
544 if (priority < mem->prev_priority)
545 mem->prev_priority = priority;
546 spin_unlock(&mem->reclaim_param_lock);
549 void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
551 spin_lock(&mem->reclaim_param_lock);
552 mem->prev_priority = priority;
553 spin_unlock(&mem->reclaim_param_lock);
556 static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages)
558 unsigned long active;
559 unsigned long inactive;
561 unsigned long inactive_ratio;
563 inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON);
564 active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON);
566 gb = (inactive + active) >> (30 - PAGE_SHIFT);
568 inactive_ratio = int_sqrt(10 * gb);
573 present_pages[0] = inactive;
574 present_pages[1] = active;
577 return inactive_ratio;
580 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg)
582 unsigned long active;
583 unsigned long inactive;
584 unsigned long present_pages[2];
585 unsigned long inactive_ratio;
587 inactive_ratio = calc_inactive_ratio(memcg, present_pages);
589 inactive = present_pages[0];
590 active = present_pages[1];
592 if (inactive * inactive_ratio < active)
598 unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg,
602 int nid = zone->zone_pgdat->node_id;
603 int zid = zone_idx(zone);
604 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
606 return MEM_CGROUP_ZSTAT(mz, lru);
609 struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
612 int nid = zone->zone_pgdat->node_id;
613 int zid = zone_idx(zone);
614 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
616 return &mz->reclaim_stat;
619 struct zone_reclaim_stat *
620 mem_cgroup_get_reclaim_stat_from_page(struct page *page)
622 struct page_cgroup *pc;
623 struct mem_cgroup_per_zone *mz;
625 if (mem_cgroup_disabled())
628 pc = lookup_page_cgroup(page);
630 * Used bit is set without atomic ops but after smp_wmb().
631 * For making pc->mem_cgroup visible, insert smp_rmb() here.
634 if (!PageCgroupUsed(pc))
637 mz = page_cgroup_zoneinfo(pc);
641 return &mz->reclaim_stat;
644 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
645 struct list_head *dst,
646 unsigned long *scanned, int order,
647 int mode, struct zone *z,
648 struct mem_cgroup *mem_cont,
649 int active, int file)
651 unsigned long nr_taken = 0;
655 struct list_head *src;
656 struct page_cgroup *pc, *tmp;
657 int nid = z->zone_pgdat->node_id;
658 int zid = zone_idx(z);
659 struct mem_cgroup_per_zone *mz;
660 int lru = LRU_FILE * !!file + !!active;
663 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
664 src = &mz->lists[lru];
667 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
668 if (scan >= nr_to_scan)
672 if (unlikely(!PageCgroupUsed(pc)))
674 if (unlikely(!PageLRU(page)))
678 if (__isolate_lru_page(page, mode, file) == 0) {
679 list_move(&page->lru, dst);
688 #define mem_cgroup_from_res_counter(counter, member) \
689 container_of(counter, struct mem_cgroup, member)
691 static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem)
693 if (do_swap_account) {
694 if (res_counter_check_under_limit(&mem->res) &&
695 res_counter_check_under_limit(&mem->memsw))
698 if (res_counter_check_under_limit(&mem->res))
703 static unsigned int get_swappiness(struct mem_cgroup *memcg)
705 struct cgroup *cgrp = memcg->css.cgroup;
706 unsigned int swappiness;
709 if (cgrp->parent == NULL)
710 return vm_swappiness;
712 spin_lock(&memcg->reclaim_param_lock);
713 swappiness = memcg->swappiness;
714 spin_unlock(&memcg->reclaim_param_lock);
719 static int mem_cgroup_count_children_cb(struct mem_cgroup *mem, void *data)
727 * mem_cgroup_print_mem_info: Called from OOM with tasklist_lock held in read mode.
728 * @memcg: The memory cgroup that went over limit
729 * @p: Task that is going to be killed
731 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
734 void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
736 struct cgroup *task_cgrp;
737 struct cgroup *mem_cgrp;
739 * Need a buffer in BSS, can't rely on allocations. The code relies
740 * on the assumption that OOM is serialized for memory controller.
741 * If this assumption is broken, revisit this code.
743 static char memcg_name[PATH_MAX];
752 mem_cgrp = memcg->css.cgroup;
753 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
755 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
758 * Unfortunately, we are unable to convert to a useful name
759 * But we'll still print out the usage information
766 printk(KERN_INFO "Task in %s killed", memcg_name);
769 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
777 * Continues from above, so we don't need an KERN_ level
779 printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
782 printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
783 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
784 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
785 res_counter_read_u64(&memcg->res, RES_FAILCNT));
786 printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
788 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
789 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
790 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
794 * This function returns the number of memcg under hierarchy tree. Returns
795 * 1(self count) if no children.
797 static int mem_cgroup_count_children(struct mem_cgroup *mem)
800 mem_cgroup_walk_tree(mem, &num, mem_cgroup_count_children_cb);
805 * Visit the first child (need not be the first child as per the ordering
806 * of the cgroup list, since we track last_scanned_child) of @mem and use
807 * that to reclaim free pages from.
809 static struct mem_cgroup *
810 mem_cgroup_select_victim(struct mem_cgroup *root_mem)
812 struct mem_cgroup *ret = NULL;
813 struct cgroup_subsys_state *css;
816 if (!root_mem->use_hierarchy) {
817 css_get(&root_mem->css);
823 nextid = root_mem->last_scanned_child + 1;
824 css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css,
826 if (css && css_tryget(css))
827 ret = container_of(css, struct mem_cgroup, css);
830 /* Updates scanning parameter */
831 spin_lock(&root_mem->reclaim_param_lock);
833 /* this means start scan from ID:1 */
834 root_mem->last_scanned_child = 0;
836 root_mem->last_scanned_child = found;
837 spin_unlock(&root_mem->reclaim_param_lock);
844 * Scan the hierarchy if needed to reclaim memory. We remember the last child
845 * we reclaimed from, so that we don't end up penalizing one child extensively
846 * based on its position in the children list.
848 * root_mem is the original ancestor that we've been reclaim from.
850 * We give up and return to the caller when we visit root_mem twice.
851 * (other groups can be removed while we're walking....)
853 * If shrink==true, for avoiding to free too much, this returns immedieately.
855 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
856 gfp_t gfp_mask, bool noswap, bool shrink)
858 struct mem_cgroup *victim;
863 victim = mem_cgroup_select_victim(root_mem);
864 if (victim == root_mem)
866 if (!mem_cgroup_local_usage(&victim->stat)) {
867 /* this cgroup's local usage == 0 */
868 css_put(&victim->css);
871 /* we use swappiness of local cgroup */
872 ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, noswap,
873 get_swappiness(victim));
874 css_put(&victim->css);
876 * At shrinking usage, we can't check we should stop here or
877 * reclaim more. It's depends on callers. last_scanned_child
878 * will work enough for keeping fairness under tree.
883 if (mem_cgroup_check_under_limit(root_mem))
889 bool mem_cgroup_oom_called(struct task_struct *task)
892 struct mem_cgroup *mem;
893 struct mm_struct *mm;
899 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
900 if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10))
906 static int record_last_oom_cb(struct mem_cgroup *mem, void *data)
908 mem->last_oom_jiffies = jiffies;
912 static void record_last_oom(struct mem_cgroup *mem)
914 mem_cgroup_walk_tree(mem, NULL, record_last_oom_cb);
919 * Unlike exported interface, "oom" parameter is added. if oom==true,
920 * oom-killer can be invoked.
922 static int __mem_cgroup_try_charge(struct mm_struct *mm,
923 gfp_t gfp_mask, struct mem_cgroup **memcg,
926 struct mem_cgroup *mem, *mem_over_limit;
927 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
928 struct res_counter *fail_res;
930 if (unlikely(test_thread_flag(TIF_MEMDIE))) {
931 /* Don't account this! */
937 * We always charge the cgroup the mm_struct belongs to.
938 * The mm_struct's mem_cgroup changes on task migration if the
939 * thread group leader migrates. It's possible that mm is not
940 * set, if so charge the init_mm (happens for pagecache usage).
944 mem = try_get_mem_cgroup_from_mm(mm);
952 VM_BUG_ON(mem_cgroup_is_obsolete(mem));
958 ret = res_counter_charge(&mem->res, PAGE_SIZE, &fail_res);
960 if (!do_swap_account)
962 ret = res_counter_charge(&mem->memsw, PAGE_SIZE,
966 /* mem+swap counter fails */
967 res_counter_uncharge(&mem->res, PAGE_SIZE);
969 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
972 /* mem counter fails */
973 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
976 if (!(gfp_mask & __GFP_WAIT))
979 ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, gfp_mask,
985 * try_to_free_mem_cgroup_pages() might not give us a full
986 * picture of reclaim. Some pages are reclaimed and might be
987 * moved to swap cache or just unmapped from the cgroup.
988 * Check the limit again to see if the reclaim reduced the
989 * current usage of the cgroup before giving up
992 if (mem_cgroup_check_under_limit(mem_over_limit))
997 mutex_lock(&memcg_tasklist);
998 mem_cgroup_out_of_memory(mem_over_limit, gfp_mask);
999 mutex_unlock(&memcg_tasklist);
1000 record_last_oom(mem_over_limit);
1011 static struct mem_cgroup *try_get_mem_cgroup_from_swapcache(struct page *page)
1013 struct mem_cgroup *mem;
1016 if (!PageSwapCache(page))
1019 ent.val = page_private(page);
1020 mem = lookup_swap_cgroup(ent);
1023 if (!css_tryget(&mem->css))
1029 * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be
1030 * USED state. If already USED, uncharge and return.
1033 static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
1034 struct page_cgroup *pc,
1035 enum charge_type ctype)
1037 /* try_charge() can return NULL to *memcg, taking care of it. */
1041 lock_page_cgroup(pc);
1042 if (unlikely(PageCgroupUsed(pc))) {
1043 unlock_page_cgroup(pc);
1044 res_counter_uncharge(&mem->res, PAGE_SIZE);
1045 if (do_swap_account)
1046 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1050 pc->mem_cgroup = mem;
1052 pc->flags = pcg_default_flags[ctype];
1054 mem_cgroup_charge_statistics(mem, pc, true);
1056 unlock_page_cgroup(pc);
1060 * mem_cgroup_move_account - move account of the page
1061 * @pc: page_cgroup of the page.
1062 * @from: mem_cgroup which the page is moved from.
1063 * @to: mem_cgroup which the page is moved to. @from != @to.
1065 * The caller must confirm following.
1066 * - page is not on LRU (isolate_page() is useful.)
1068 * returns 0 at success,
1069 * returns -EBUSY when lock is busy or "pc" is unstable.
1071 * This function does "uncharge" from old cgroup but doesn't do "charge" to
1072 * new cgroup. It should be done by a caller.
1075 static int mem_cgroup_move_account(struct page_cgroup *pc,
1076 struct mem_cgroup *from, struct mem_cgroup *to)
1078 struct mem_cgroup_per_zone *from_mz, *to_mz;
1082 VM_BUG_ON(from == to);
1083 VM_BUG_ON(PageLRU(pc->page));
1085 nid = page_cgroup_nid(pc);
1086 zid = page_cgroup_zid(pc);
1087 from_mz = mem_cgroup_zoneinfo(from, nid, zid);
1088 to_mz = mem_cgroup_zoneinfo(to, nid, zid);
1090 if (!trylock_page_cgroup(pc))
1093 if (!PageCgroupUsed(pc))
1096 if (pc->mem_cgroup != from)
1099 res_counter_uncharge(&from->res, PAGE_SIZE);
1100 mem_cgroup_charge_statistics(from, pc, false);
1101 if (do_swap_account)
1102 res_counter_uncharge(&from->memsw, PAGE_SIZE);
1103 css_put(&from->css);
1106 pc->mem_cgroup = to;
1107 mem_cgroup_charge_statistics(to, pc, true);
1110 unlock_page_cgroup(pc);
1115 * move charges to its parent.
1118 static int mem_cgroup_move_parent(struct page_cgroup *pc,
1119 struct mem_cgroup *child,
1122 struct page *page = pc->page;
1123 struct cgroup *cg = child->css.cgroup;
1124 struct cgroup *pcg = cg->parent;
1125 struct mem_cgroup *parent;
1133 parent = mem_cgroup_from_cont(pcg);
1136 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false);
1140 if (!get_page_unless_zero(page)) {
1145 ret = isolate_lru_page(page);
1150 ret = mem_cgroup_move_account(pc, child, parent);
1152 putback_lru_page(page);
1155 /* drop extra refcnt by try_charge() */
1156 css_put(&parent->css);
1163 /* drop extra refcnt by try_charge() */
1164 css_put(&parent->css);
1165 /* uncharge if move fails */
1166 res_counter_uncharge(&parent->res, PAGE_SIZE);
1167 if (do_swap_account)
1168 res_counter_uncharge(&parent->memsw, PAGE_SIZE);
1173 * Charge the memory controller for page usage.
1175 * 0 if the charge was successful
1176 * < 0 if the cgroup is over its limit
1178 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
1179 gfp_t gfp_mask, enum charge_type ctype,
1180 struct mem_cgroup *memcg)
1182 struct mem_cgroup *mem;
1183 struct page_cgroup *pc;
1186 pc = lookup_page_cgroup(page);
1187 /* can happen at boot */
1193 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true);
1197 __mem_cgroup_commit_charge(mem, pc, ctype);
1201 int mem_cgroup_newpage_charge(struct page *page,
1202 struct mm_struct *mm, gfp_t gfp_mask)
1204 if (mem_cgroup_disabled())
1206 if (PageCompound(page))
1209 * If already mapped, we don't have to account.
1210 * If page cache, page->mapping has address_space.
1211 * But page->mapping may have out-of-use anon_vma pointer,
1212 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
1215 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
1219 return mem_cgroup_charge_common(page, mm, gfp_mask,
1220 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
1223 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
1226 struct mem_cgroup *mem = NULL;
1229 if (mem_cgroup_disabled())
1231 if (PageCompound(page))
1234 * Corner case handling. This is called from add_to_page_cache()
1235 * in usual. But some FS (shmem) precharges this page before calling it
1236 * and call add_to_page_cache() with GFP_NOWAIT.
1238 * For GFP_NOWAIT case, the page may be pre-charged before calling
1239 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
1240 * charge twice. (It works but has to pay a bit larger cost.)
1241 * And when the page is SwapCache, it should take swap information
1242 * into account. This is under lock_page() now.
1244 if (!(gfp_mask & __GFP_WAIT)) {
1245 struct page_cgroup *pc;
1248 pc = lookup_page_cgroup(page);
1251 lock_page_cgroup(pc);
1252 if (PageCgroupUsed(pc)) {
1253 unlock_page_cgroup(pc);
1256 unlock_page_cgroup(pc);
1259 if (do_swap_account && PageSwapCache(page)) {
1260 mem = try_get_mem_cgroup_from_swapcache(page);
1265 /* SwapCache may be still linked to LRU now. */
1266 mem_cgroup_lru_del_before_commit_swapcache(page);
1269 if (unlikely(!mm && !mem))
1272 if (page_is_file_cache(page))
1273 return mem_cgroup_charge_common(page, mm, gfp_mask,
1274 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
1276 ret = mem_cgroup_charge_common(page, mm, gfp_mask,
1277 MEM_CGROUP_CHARGE_TYPE_SHMEM, mem);
1280 if (PageSwapCache(page))
1281 mem_cgroup_lru_add_after_commit_swapcache(page);
1283 if (do_swap_account && !ret && PageSwapCache(page)) {
1284 swp_entry_t ent = {.val = page_private(page)};
1285 /* avoid double counting */
1286 mem = swap_cgroup_record(ent, NULL);
1288 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1289 mem_cgroup_put(mem);
1296 * While swap-in, try_charge -> commit or cancel, the page is locked.
1297 * And when try_charge() successfully returns, one refcnt to memcg without
1298 * struct page_cgroup is aquired. This refcnt will be cumsumed by
1299 * "commit()" or removed by "cancel()"
1301 int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
1303 gfp_t mask, struct mem_cgroup **ptr)
1305 struct mem_cgroup *mem;
1308 if (mem_cgroup_disabled())
1311 if (!do_swap_account)
1314 * A racing thread's fault, or swapoff, may have already updated
1315 * the pte, and even removed page from swap cache: return success
1316 * to go on to do_swap_page()'s pte_same() test, which should fail.
1318 if (!PageSwapCache(page))
1320 mem = try_get_mem_cgroup_from_swapcache(page);
1324 ret = __mem_cgroup_try_charge(NULL, mask, ptr, true);
1325 /* drop extra refcnt from tryget */
1331 return __mem_cgroup_try_charge(mm, mask, ptr, true);
1334 void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
1336 struct page_cgroup *pc;
1338 if (mem_cgroup_disabled())
1342 pc = lookup_page_cgroup(page);
1343 mem_cgroup_lru_del_before_commit_swapcache(page);
1344 __mem_cgroup_commit_charge(ptr, pc, MEM_CGROUP_CHARGE_TYPE_MAPPED);
1345 mem_cgroup_lru_add_after_commit_swapcache(page);
1347 * Now swap is on-memory. This means this page may be
1348 * counted both as mem and swap....double count.
1349 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
1350 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
1351 * may call delete_from_swap_cache() before reach here.
1353 if (do_swap_account && PageSwapCache(page)) {
1354 swp_entry_t ent = {.val = page_private(page)};
1355 struct mem_cgroup *memcg;
1356 memcg = swap_cgroup_record(ent, NULL);
1358 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1359 mem_cgroup_put(memcg);
1363 /* add this page(page_cgroup) to the LRU we want. */
1367 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
1369 if (mem_cgroup_disabled())
1373 res_counter_uncharge(&mem->res, PAGE_SIZE);
1374 if (do_swap_account)
1375 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1381 * uncharge if !page_mapped(page)
1383 static struct mem_cgroup *
1384 __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
1386 struct page_cgroup *pc;
1387 struct mem_cgroup *mem = NULL;
1388 struct mem_cgroup_per_zone *mz;
1390 if (mem_cgroup_disabled())
1393 if (PageSwapCache(page))
1397 * Check if our page_cgroup is valid
1399 pc = lookup_page_cgroup(page);
1400 if (unlikely(!pc || !PageCgroupUsed(pc)))
1403 lock_page_cgroup(pc);
1405 mem = pc->mem_cgroup;
1407 if (!PageCgroupUsed(pc))
1411 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
1412 if (page_mapped(page))
1415 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
1416 if (!PageAnon(page)) { /* Shared memory */
1417 if (page->mapping && !page_is_file_cache(page))
1419 } else if (page_mapped(page)) /* Anon */
1426 res_counter_uncharge(&mem->res, PAGE_SIZE);
1427 if (do_swap_account && (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT))
1428 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1429 mem_cgroup_charge_statistics(mem, pc, false);
1431 ClearPageCgroupUsed(pc);
1433 * pc->mem_cgroup is not cleared here. It will be accessed when it's
1434 * freed from LRU. This is safe because uncharged page is expected not
1435 * to be reused (freed soon). Exception is SwapCache, it's handled by
1436 * special functions.
1439 mz = page_cgroup_zoneinfo(pc);
1440 unlock_page_cgroup(pc);
1442 /* at swapout, this memcg will be accessed to record to swap */
1443 if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
1449 unlock_page_cgroup(pc);
1453 void mem_cgroup_uncharge_page(struct page *page)
1456 if (page_mapped(page))
1458 if (page->mapping && !PageAnon(page))
1460 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
1463 void mem_cgroup_uncharge_cache_page(struct page *page)
1465 VM_BUG_ON(page_mapped(page));
1466 VM_BUG_ON(page->mapping);
1467 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
1471 * called from __delete_from_swap_cache() and drop "page" account.
1472 * memcg information is recorded to swap_cgroup of "ent"
1474 void mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent)
1476 struct mem_cgroup *memcg;
1478 memcg = __mem_cgroup_uncharge_common(page,
1479 MEM_CGROUP_CHARGE_TYPE_SWAPOUT);
1480 /* record memcg information */
1481 if (do_swap_account && memcg) {
1482 swap_cgroup_record(ent, memcg);
1483 mem_cgroup_get(memcg);
1486 css_put(&memcg->css);
1489 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1491 * called from swap_entry_free(). remove record in swap_cgroup and
1492 * uncharge "memsw" account.
1494 void mem_cgroup_uncharge_swap(swp_entry_t ent)
1496 struct mem_cgroup *memcg;
1498 if (!do_swap_account)
1501 memcg = swap_cgroup_record(ent, NULL);
1503 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1504 mem_cgroup_put(memcg);
1510 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
1513 int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr)
1515 struct page_cgroup *pc;
1516 struct mem_cgroup *mem = NULL;
1519 if (mem_cgroup_disabled())
1522 pc = lookup_page_cgroup(page);
1523 lock_page_cgroup(pc);
1524 if (PageCgroupUsed(pc)) {
1525 mem = pc->mem_cgroup;
1528 unlock_page_cgroup(pc);
1531 ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false);
1538 /* remove redundant charge if migration failed*/
1539 void mem_cgroup_end_migration(struct mem_cgroup *mem,
1540 struct page *oldpage, struct page *newpage)
1542 struct page *target, *unused;
1543 struct page_cgroup *pc;
1544 enum charge_type ctype;
1549 /* at migration success, oldpage->mapping is NULL. */
1550 if (oldpage->mapping) {
1558 if (PageAnon(target))
1559 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
1560 else if (page_is_file_cache(target))
1561 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
1563 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
1565 /* unused page is not on radix-tree now. */
1567 __mem_cgroup_uncharge_common(unused, ctype);
1569 pc = lookup_page_cgroup(target);
1571 * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup.
1572 * So, double-counting is effectively avoided.
1574 __mem_cgroup_commit_charge(mem, pc, ctype);
1577 * Both of oldpage and newpage are still under lock_page().
1578 * Then, we don't have to care about race in radix-tree.
1579 * But we have to be careful that this page is unmapped or not.
1581 * There is a case for !page_mapped(). At the start of
1582 * migration, oldpage was mapped. But now, it's zapped.
1583 * But we know *target* page is not freed/reused under us.
1584 * mem_cgroup_uncharge_page() does all necessary checks.
1586 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
1587 mem_cgroup_uncharge_page(target);
1591 * A call to try to shrink memory usage under specified resource controller.
1592 * This is typically used for page reclaiming for shmem for reducing side
1593 * effect of page allocation from shmem, which is used by some mem_cgroup.
1595 int mem_cgroup_shrink_usage(struct page *page,
1596 struct mm_struct *mm,
1599 struct mem_cgroup *mem = NULL;
1601 int retry = MEM_CGROUP_RECLAIM_RETRIES;
1603 if (mem_cgroup_disabled())
1606 mem = try_get_mem_cgroup_from_swapcache(page);
1608 mem = try_get_mem_cgroup_from_mm(mm);
1613 progress = mem_cgroup_hierarchical_reclaim(mem,
1614 gfp_mask, true, false);
1615 progress += mem_cgroup_check_under_limit(mem);
1616 } while (!progress && --retry);
1624 static DEFINE_MUTEX(set_limit_mutex);
1626 static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
1627 unsigned long long val)
1633 int children = mem_cgroup_count_children(memcg);
1634 u64 curusage, oldusage;
1637 * For keeping hierarchical_reclaim simple, how long we should retry
1638 * is depends on callers. We set our retry-count to be function
1639 * of # of children which we should visit in this loop.
1641 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
1643 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
1645 while (retry_count) {
1646 if (signal_pending(current)) {
1651 * Rather than hide all in some function, I do this in
1652 * open coded manner. You see what this really does.
1653 * We have to guarantee mem->res.limit < mem->memsw.limit.
1655 mutex_lock(&set_limit_mutex);
1656 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1657 if (memswlimit < val) {
1659 mutex_unlock(&set_limit_mutex);
1662 ret = res_counter_set_limit(&memcg->res, val);
1663 mutex_unlock(&set_limit_mutex);
1668 progress = mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL,
1670 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
1671 /* Usage is reduced ? */
1672 if (curusage >= oldusage)
1675 oldusage = curusage;
1681 int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
1682 unsigned long long val)
1685 u64 memlimit, oldusage, curusage;
1686 int children = mem_cgroup_count_children(memcg);
1689 if (!do_swap_account)
1691 /* see mem_cgroup_resize_res_limit */
1692 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
1693 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
1694 while (retry_count) {
1695 if (signal_pending(current)) {
1700 * Rather than hide all in some function, I do this in
1701 * open coded manner. You see what this really does.
1702 * We have to guarantee mem->res.limit < mem->memsw.limit.
1704 mutex_lock(&set_limit_mutex);
1705 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1706 if (memlimit > val) {
1708 mutex_unlock(&set_limit_mutex);
1711 ret = res_counter_set_limit(&memcg->memsw, val);
1712 mutex_unlock(&set_limit_mutex);
1717 mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL, true, true);
1718 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
1719 /* Usage is reduced ? */
1720 if (curusage >= oldusage)
1723 oldusage = curusage;
1729 * This routine traverse page_cgroup in given list and drop them all.
1730 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
1732 static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
1733 int node, int zid, enum lru_list lru)
1736 struct mem_cgroup_per_zone *mz;
1737 struct page_cgroup *pc, *busy;
1738 unsigned long flags, loop;
1739 struct list_head *list;
1742 zone = &NODE_DATA(node)->node_zones[zid];
1743 mz = mem_cgroup_zoneinfo(mem, node, zid);
1744 list = &mz->lists[lru];
1746 loop = MEM_CGROUP_ZSTAT(mz, lru);
1747 /* give some margin against EBUSY etc...*/
1752 spin_lock_irqsave(&zone->lru_lock, flags);
1753 if (list_empty(list)) {
1754 spin_unlock_irqrestore(&zone->lru_lock, flags);
1757 pc = list_entry(list->prev, struct page_cgroup, lru);
1759 list_move(&pc->lru, list);
1761 spin_unlock_irqrestore(&zone->lru_lock, flags);
1764 spin_unlock_irqrestore(&zone->lru_lock, flags);
1766 ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL);
1770 if (ret == -EBUSY || ret == -EINVAL) {
1771 /* found lock contention or "pc" is obsolete. */
1778 if (!ret && !list_empty(list))
1784 * make mem_cgroup's charge to be 0 if there is no task.
1785 * This enables deleting this mem_cgroup.
1787 static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
1790 int node, zid, shrink;
1791 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
1792 struct cgroup *cgrp = mem->css.cgroup;
1797 /* should free all ? */
1801 while (mem->res.usage > 0) {
1803 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
1806 if (signal_pending(current))
1808 /* This is for making all *used* pages to be on LRU. */
1809 lru_add_drain_all();
1811 for_each_node_state(node, N_HIGH_MEMORY) {
1812 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
1815 ret = mem_cgroup_force_empty_list(mem,
1824 /* it seems parent cgroup doesn't have enough mem */
1835 /* returns EBUSY if there is a task or if we come here twice. */
1836 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
1840 /* we call try-to-free pages for make this cgroup empty */
1841 lru_add_drain_all();
1842 /* try to free all pages in this cgroup */
1844 while (nr_retries && mem->res.usage > 0) {
1847 if (signal_pending(current)) {
1851 progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL,
1852 false, get_swappiness(mem));
1855 /* maybe some writeback is necessary */
1856 congestion_wait(WRITE, HZ/10);
1861 /* try move_account...there may be some *locked* pages. */
1868 int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
1870 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
1874 static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
1876 return mem_cgroup_from_cont(cont)->use_hierarchy;
1879 static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
1883 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1884 struct cgroup *parent = cont->parent;
1885 struct mem_cgroup *parent_mem = NULL;
1888 parent_mem = mem_cgroup_from_cont(parent);
1892 * If parent's use_hiearchy is set, we can't make any modifications
1893 * in the child subtrees. If it is unset, then the change can
1894 * occur, provided the current cgroup has no children.
1896 * For the root cgroup, parent_mem is NULL, we allow value to be
1897 * set if there are no children.
1899 if ((!parent_mem || !parent_mem->use_hierarchy) &&
1900 (val == 1 || val == 0)) {
1901 if (list_empty(&cont->children))
1902 mem->use_hierarchy = val;
1912 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
1914 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1918 type = MEMFILE_TYPE(cft->private);
1919 name = MEMFILE_ATTR(cft->private);
1922 val = res_counter_read_u64(&mem->res, name);
1925 if (do_swap_account)
1926 val = res_counter_read_u64(&mem->memsw, name);
1935 * The user of this function is...
1938 static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
1941 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
1943 unsigned long long val;
1946 type = MEMFILE_TYPE(cft->private);
1947 name = MEMFILE_ATTR(cft->private);
1950 /* This function does all necessary parse...reuse it */
1951 ret = res_counter_memparse_write_strategy(buffer, &val);
1955 ret = mem_cgroup_resize_limit(memcg, val);
1957 ret = mem_cgroup_resize_memsw_limit(memcg, val);
1960 ret = -EINVAL; /* should be BUG() ? */
1966 static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
1967 unsigned long long *mem_limit, unsigned long long *memsw_limit)
1969 struct cgroup *cgroup;
1970 unsigned long long min_limit, min_memsw_limit, tmp;
1972 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1973 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1974 cgroup = memcg->css.cgroup;
1975 if (!memcg->use_hierarchy)
1978 while (cgroup->parent) {
1979 cgroup = cgroup->parent;
1980 memcg = mem_cgroup_from_cont(cgroup);
1981 if (!memcg->use_hierarchy)
1983 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
1984 min_limit = min(min_limit, tmp);
1985 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1986 min_memsw_limit = min(min_memsw_limit, tmp);
1989 *mem_limit = min_limit;
1990 *memsw_limit = min_memsw_limit;
1994 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
1996 struct mem_cgroup *mem;
1999 mem = mem_cgroup_from_cont(cont);
2000 type = MEMFILE_TYPE(event);
2001 name = MEMFILE_ATTR(event);
2005 res_counter_reset_max(&mem->res);
2007 res_counter_reset_max(&mem->memsw);
2011 res_counter_reset_failcnt(&mem->res);
2013 res_counter_reset_failcnt(&mem->memsw);
2020 /* For read statistics */
2034 struct mcs_total_stat {
2035 s64 stat[NR_MCS_STAT];
2041 } memcg_stat_strings[NR_MCS_STAT] = {
2042 {"cache", "total_cache"},
2043 {"rss", "total_rss"},
2044 {"pgpgin", "total_pgpgin"},
2045 {"pgpgout", "total_pgpgout"},
2046 {"inactive_anon", "total_inactive_anon"},
2047 {"active_anon", "total_active_anon"},
2048 {"inactive_file", "total_inactive_file"},
2049 {"active_file", "total_active_file"},
2050 {"unevictable", "total_unevictable"}
2054 static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data)
2056 struct mcs_total_stat *s = data;
2060 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_CACHE);
2061 s->stat[MCS_CACHE] += val * PAGE_SIZE;
2062 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
2063 s->stat[MCS_RSS] += val * PAGE_SIZE;
2064 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGIN_COUNT);
2065 s->stat[MCS_PGPGIN] += val;
2066 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGOUT_COUNT);
2067 s->stat[MCS_PGPGOUT] += val;
2070 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON);
2071 s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE;
2072 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON);
2073 s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE;
2074 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE);
2075 s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE;
2076 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE);
2077 s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE;
2078 val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE);
2079 s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE;
2084 mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s)
2086 mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat);
2089 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
2090 struct cgroup_map_cb *cb)
2092 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
2093 struct mcs_total_stat mystat;
2096 memset(&mystat, 0, sizeof(mystat));
2097 mem_cgroup_get_local_stat(mem_cont, &mystat);
2099 for (i = 0; i < NR_MCS_STAT; i++)
2100 cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
2102 /* Hierarchical information */
2104 unsigned long long limit, memsw_limit;
2105 memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit);
2106 cb->fill(cb, "hierarchical_memory_limit", limit);
2107 if (do_swap_account)
2108 cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
2111 memset(&mystat, 0, sizeof(mystat));
2112 mem_cgroup_get_total_stat(mem_cont, &mystat);
2113 for (i = 0; i < NR_MCS_STAT; i++)
2114 cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
2117 #ifdef CONFIG_DEBUG_VM
2118 cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL));
2122 struct mem_cgroup_per_zone *mz;
2123 unsigned long recent_rotated[2] = {0, 0};
2124 unsigned long recent_scanned[2] = {0, 0};
2126 for_each_online_node(nid)
2127 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
2128 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
2130 recent_rotated[0] +=
2131 mz->reclaim_stat.recent_rotated[0];
2132 recent_rotated[1] +=
2133 mz->reclaim_stat.recent_rotated[1];
2134 recent_scanned[0] +=
2135 mz->reclaim_stat.recent_scanned[0];
2136 recent_scanned[1] +=
2137 mz->reclaim_stat.recent_scanned[1];
2139 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
2140 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
2141 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
2142 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
2149 static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
2151 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
2153 return get_swappiness(memcg);
2156 static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
2159 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
2160 struct mem_cgroup *parent;
2165 if (cgrp->parent == NULL)
2168 parent = mem_cgroup_from_cont(cgrp->parent);
2172 /* If under hierarchy, only empty-root can set this value */
2173 if ((parent->use_hierarchy) ||
2174 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
2179 spin_lock(&memcg->reclaim_param_lock);
2180 memcg->swappiness = val;
2181 spin_unlock(&memcg->reclaim_param_lock);
2189 static struct cftype mem_cgroup_files[] = {
2191 .name = "usage_in_bytes",
2192 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
2193 .read_u64 = mem_cgroup_read,
2196 .name = "max_usage_in_bytes",
2197 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
2198 .trigger = mem_cgroup_reset,
2199 .read_u64 = mem_cgroup_read,
2202 .name = "limit_in_bytes",
2203 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
2204 .write_string = mem_cgroup_write,
2205 .read_u64 = mem_cgroup_read,
2209 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
2210 .trigger = mem_cgroup_reset,
2211 .read_u64 = mem_cgroup_read,
2215 .read_map = mem_control_stat_show,
2218 .name = "force_empty",
2219 .trigger = mem_cgroup_force_empty_write,
2222 .name = "use_hierarchy",
2223 .write_u64 = mem_cgroup_hierarchy_write,
2224 .read_u64 = mem_cgroup_hierarchy_read,
2227 .name = "swappiness",
2228 .read_u64 = mem_cgroup_swappiness_read,
2229 .write_u64 = mem_cgroup_swappiness_write,
2233 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2234 static struct cftype memsw_cgroup_files[] = {
2236 .name = "memsw.usage_in_bytes",
2237 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
2238 .read_u64 = mem_cgroup_read,
2241 .name = "memsw.max_usage_in_bytes",
2242 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
2243 .trigger = mem_cgroup_reset,
2244 .read_u64 = mem_cgroup_read,
2247 .name = "memsw.limit_in_bytes",
2248 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
2249 .write_string = mem_cgroup_write,
2250 .read_u64 = mem_cgroup_read,
2253 .name = "memsw.failcnt",
2254 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
2255 .trigger = mem_cgroup_reset,
2256 .read_u64 = mem_cgroup_read,
2260 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
2262 if (!do_swap_account)
2264 return cgroup_add_files(cont, ss, memsw_cgroup_files,
2265 ARRAY_SIZE(memsw_cgroup_files));
2268 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
2274 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
2276 struct mem_cgroup_per_node *pn;
2277 struct mem_cgroup_per_zone *mz;
2279 int zone, tmp = node;
2281 * This routine is called against possible nodes.
2282 * But it's BUG to call kmalloc() against offline node.
2284 * TODO: this routine can waste much memory for nodes which will
2285 * never be onlined. It's better to use memory hotplug callback
2288 if (!node_state(node, N_NORMAL_MEMORY))
2290 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
2294 mem->info.nodeinfo[node] = pn;
2295 memset(pn, 0, sizeof(*pn));
2297 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
2298 mz = &pn->zoneinfo[zone];
2300 INIT_LIST_HEAD(&mz->lists[l]);
2305 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
2307 kfree(mem->info.nodeinfo[node]);
2310 static int mem_cgroup_size(void)
2312 int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu);
2313 return sizeof(struct mem_cgroup) + cpustat_size;
2316 static struct mem_cgroup *mem_cgroup_alloc(void)
2318 struct mem_cgroup *mem;
2319 int size = mem_cgroup_size();
2321 if (size < PAGE_SIZE)
2322 mem = kmalloc(size, GFP_KERNEL);
2324 mem = vmalloc(size);
2327 memset(mem, 0, size);
2332 * At destroying mem_cgroup, references from swap_cgroup can remain.
2333 * (scanning all at force_empty is too costly...)
2335 * Instead of clearing all references at force_empty, we remember
2336 * the number of reference from swap_cgroup and free mem_cgroup when
2337 * it goes down to 0.
2339 * Removal of cgroup itself succeeds regardless of refs from swap.
2342 static void __mem_cgroup_free(struct mem_cgroup *mem)
2346 free_css_id(&mem_cgroup_subsys, &mem->css);
2348 for_each_node_state(node, N_POSSIBLE)
2349 free_mem_cgroup_per_zone_info(mem, node);
2351 if (mem_cgroup_size() < PAGE_SIZE)
2357 static void mem_cgroup_get(struct mem_cgroup *mem)
2359 atomic_inc(&mem->refcnt);
2362 static void mem_cgroup_put(struct mem_cgroup *mem)
2364 if (atomic_dec_and_test(&mem->refcnt)) {
2365 struct mem_cgroup *parent = parent_mem_cgroup(mem);
2366 __mem_cgroup_free(mem);
2368 mem_cgroup_put(parent);
2373 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
2375 static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem)
2377 if (!mem->res.parent)
2379 return mem_cgroup_from_res_counter(mem->res.parent, res);
2382 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2383 static void __init enable_swap_cgroup(void)
2385 if (!mem_cgroup_disabled() && really_do_swap_account)
2386 do_swap_account = 1;
2389 static void __init enable_swap_cgroup(void)
2394 static struct cgroup_subsys_state * __ref
2395 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
2397 struct mem_cgroup *mem, *parent;
2398 long error = -ENOMEM;
2401 mem = mem_cgroup_alloc();
2403 return ERR_PTR(error);
2405 for_each_node_state(node, N_POSSIBLE)
2406 if (alloc_mem_cgroup_per_zone_info(mem, node))
2409 if (cont->parent == NULL) {
2410 enable_swap_cgroup();
2413 parent = mem_cgroup_from_cont(cont->parent);
2414 mem->use_hierarchy = parent->use_hierarchy;
2417 if (parent && parent->use_hierarchy) {
2418 res_counter_init(&mem->res, &parent->res);
2419 res_counter_init(&mem->memsw, &parent->memsw);
2421 * We increment refcnt of the parent to ensure that we can
2422 * safely access it on res_counter_charge/uncharge.
2423 * This refcnt will be decremented when freeing this
2424 * mem_cgroup(see mem_cgroup_put).
2426 mem_cgroup_get(parent);
2428 res_counter_init(&mem->res, NULL);
2429 res_counter_init(&mem->memsw, NULL);
2431 mem->last_scanned_child = 0;
2432 spin_lock_init(&mem->reclaim_param_lock);
2435 mem->swappiness = get_swappiness(parent);
2436 atomic_set(&mem->refcnt, 1);
2439 __mem_cgroup_free(mem);
2440 return ERR_PTR(error);
2443 static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
2444 struct cgroup *cont)
2446 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2448 return mem_cgroup_force_empty(mem, false);
2451 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
2452 struct cgroup *cont)
2454 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2456 mem_cgroup_put(mem);
2459 static int mem_cgroup_populate(struct cgroup_subsys *ss,
2460 struct cgroup *cont)
2464 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
2465 ARRAY_SIZE(mem_cgroup_files));
2468 ret = register_memsw_files(cont, ss);
2472 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
2473 struct cgroup *cont,
2474 struct cgroup *old_cont,
2475 struct task_struct *p)
2477 mutex_lock(&memcg_tasklist);
2479 * FIXME: It's better to move charges of this process from old
2480 * memcg to new memcg. But it's just on TODO-List now.
2482 mutex_unlock(&memcg_tasklist);
2485 struct cgroup_subsys mem_cgroup_subsys = {
2487 .subsys_id = mem_cgroup_subsys_id,
2488 .create = mem_cgroup_create,
2489 .pre_destroy = mem_cgroup_pre_destroy,
2490 .destroy = mem_cgroup_destroy,
2491 .populate = mem_cgroup_populate,
2492 .attach = mem_cgroup_move_task,
2497 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2499 static int __init disable_swap_account(char *s)
2501 really_do_swap_account = 0;
2504 __setup("noswapaccount", disable_swap_account);