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 = 1 */
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 anon rss */
66 MEM_CGROUP_STAT_MAPPED_FILE, /* # of pages charged as file rss */
67 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
68 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
70 MEM_CGROUP_STAT_NSTATS,
73 struct mem_cgroup_stat_cpu {
74 s64 count[MEM_CGROUP_STAT_NSTATS];
75 } ____cacheline_aligned_in_smp;
77 struct mem_cgroup_stat {
78 struct mem_cgroup_stat_cpu cpustat[0];
82 * For accounting under irq disable, no need for increment preempt count.
84 static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat,
85 enum mem_cgroup_stat_index idx, int val)
87 stat->count[idx] += val;
90 static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat,
91 enum mem_cgroup_stat_index idx)
95 for_each_possible_cpu(cpu)
96 ret += stat->cpustat[cpu].count[idx];
100 static s64 mem_cgroup_local_usage(struct mem_cgroup_stat *stat)
104 ret = mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_CACHE);
105 ret += mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_RSS);
110 * per-zone information in memory controller.
112 struct mem_cgroup_per_zone {
114 * spin_lock to protect the per cgroup LRU
116 struct list_head lists[NR_LRU_LISTS];
117 unsigned long count[NR_LRU_LISTS];
119 struct zone_reclaim_stat reclaim_stat;
121 /* Macro for accessing counter */
122 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
124 struct mem_cgroup_per_node {
125 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
128 struct mem_cgroup_lru_info {
129 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
133 * The memory controller data structure. The memory controller controls both
134 * page cache and RSS per cgroup. We would eventually like to provide
135 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
136 * to help the administrator determine what knobs to tune.
138 * TODO: Add a water mark for the memory controller. Reclaim will begin when
139 * we hit the water mark. May be even add a low water mark, such that
140 * no reclaim occurs from a cgroup at it's low water mark, this is
141 * a feature that will be implemented much later in the future.
144 struct cgroup_subsys_state css;
146 * the counter to account for memory usage
148 struct res_counter res;
150 * the counter to account for mem+swap usage.
152 struct res_counter memsw;
154 * Per cgroup active and inactive list, similar to the
155 * per zone LRU lists.
157 struct mem_cgroup_lru_info info;
160 protect against reclaim related member.
162 spinlock_t reclaim_param_lock;
164 int prev_priority; /* for recording reclaim priority */
167 * While reclaiming in a hiearchy, we cache the last child we
170 int last_scanned_child;
172 * Should the accounting and control be hierarchical, per subtree?
175 unsigned long last_oom_jiffies;
178 unsigned int swappiness;
181 * statistics. This must be placed at the end of memcg.
183 struct mem_cgroup_stat stat;
187 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
188 MEM_CGROUP_CHARGE_TYPE_MAPPED,
189 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
190 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
191 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
195 /* only for here (for easy reading.) */
196 #define PCGF_CACHE (1UL << PCG_CACHE)
197 #define PCGF_USED (1UL << PCG_USED)
198 #define PCGF_LOCK (1UL << PCG_LOCK)
199 static const unsigned long
200 pcg_default_flags[NR_CHARGE_TYPE] = {
201 PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* File Cache */
202 PCGF_USED | PCGF_LOCK, /* Anon */
203 PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* Shmem */
207 /* for encoding cft->private value on file */
210 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
211 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
212 #define MEMFILE_ATTR(val) ((val) & 0xffff)
214 static void mem_cgroup_get(struct mem_cgroup *mem);
215 static void mem_cgroup_put(struct mem_cgroup *mem);
216 static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem);
218 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
219 struct page_cgroup *pc,
222 int val = (charge)? 1 : -1;
223 struct mem_cgroup_stat *stat = &mem->stat;
224 struct mem_cgroup_stat_cpu *cpustat;
227 cpustat = &stat->cpustat[cpu];
228 if (PageCgroupCache(pc))
229 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val);
231 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val);
234 __mem_cgroup_stat_add_safe(cpustat,
235 MEM_CGROUP_STAT_PGPGIN_COUNT, 1);
237 __mem_cgroup_stat_add_safe(cpustat,
238 MEM_CGROUP_STAT_PGPGOUT_COUNT, 1);
242 static struct mem_cgroup_per_zone *
243 mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
245 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
248 static struct mem_cgroup_per_zone *
249 page_cgroup_zoneinfo(struct page_cgroup *pc)
251 struct mem_cgroup *mem = pc->mem_cgroup;
252 int nid = page_cgroup_nid(pc);
253 int zid = page_cgroup_zid(pc);
258 return mem_cgroup_zoneinfo(mem, nid, zid);
261 static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem,
265 struct mem_cgroup_per_zone *mz;
268 for_each_online_node(nid)
269 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
270 mz = mem_cgroup_zoneinfo(mem, nid, zid);
271 total += MEM_CGROUP_ZSTAT(mz, idx);
276 static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
278 return container_of(cgroup_subsys_state(cont,
279 mem_cgroup_subsys_id), struct mem_cgroup,
283 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
286 * mm_update_next_owner() may clear mm->owner to NULL
287 * if it races with swapoff, page migration, etc.
288 * So this can be called with p == NULL.
293 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
294 struct mem_cgroup, css);
297 static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
299 struct mem_cgroup *mem = NULL;
304 * Because we have no locks, mm->owner's may be being moved to other
305 * cgroup. We use css_tryget() here even if this looks
306 * pessimistic (rather than adding locks here).
310 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
313 } while (!css_tryget(&mem->css));
319 * Call callback function against all cgroup under hierarchy tree.
321 static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data,
322 int (*func)(struct mem_cgroup *, void *))
324 int found, ret, nextid;
325 struct cgroup_subsys_state *css;
326 struct mem_cgroup *mem;
328 if (!root->use_hierarchy)
329 return (*func)(root, data);
337 css = css_get_next(&mem_cgroup_subsys, nextid, &root->css,
339 if (css && css_tryget(css))
340 mem = container_of(css, struct mem_cgroup, css);
344 ret = (*func)(mem, data);
348 } while (!ret && css);
354 * Following LRU functions are allowed to be used without PCG_LOCK.
355 * Operations are called by routine of global LRU independently from memcg.
356 * What we have to take care of here is validness of pc->mem_cgroup.
358 * Changes to pc->mem_cgroup happens when
361 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
362 * It is added to LRU before charge.
363 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
364 * When moving account, the page is not on LRU. It's isolated.
367 void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
369 struct page_cgroup *pc;
370 struct mem_cgroup *mem;
371 struct mem_cgroup_per_zone *mz;
373 if (mem_cgroup_disabled())
375 pc = lookup_page_cgroup(page);
376 /* can happen while we handle swapcache. */
377 if (list_empty(&pc->lru) || !pc->mem_cgroup)
380 * We don't check PCG_USED bit. It's cleared when the "page" is finally
381 * removed from global LRU.
383 mz = page_cgroup_zoneinfo(pc);
384 mem = pc->mem_cgroup;
385 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
386 list_del_init(&pc->lru);
390 void mem_cgroup_del_lru(struct page *page)
392 mem_cgroup_del_lru_list(page, page_lru(page));
395 void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
397 struct mem_cgroup_per_zone *mz;
398 struct page_cgroup *pc;
400 if (mem_cgroup_disabled())
403 pc = lookup_page_cgroup(page);
405 * Used bit is set without atomic ops but after smp_wmb().
406 * For making pc->mem_cgroup visible, insert smp_rmb() here.
409 /* unused page is not rotated. */
410 if (!PageCgroupUsed(pc))
412 mz = page_cgroup_zoneinfo(pc);
413 list_move(&pc->lru, &mz->lists[lru]);
416 void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
418 struct page_cgroup *pc;
419 struct mem_cgroup_per_zone *mz;
421 if (mem_cgroup_disabled())
423 pc = lookup_page_cgroup(page);
425 * Used bit is set without atomic ops but after smp_wmb().
426 * For making pc->mem_cgroup visible, insert smp_rmb() here.
429 if (!PageCgroupUsed(pc))
432 mz = page_cgroup_zoneinfo(pc);
433 MEM_CGROUP_ZSTAT(mz, lru) += 1;
434 list_add(&pc->lru, &mz->lists[lru]);
438 * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to
439 * lru because the page may.be reused after it's fully uncharged (because of
440 * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge
441 * it again. This function is only used to charge SwapCache. It's done under
442 * lock_page and expected that zone->lru_lock is never held.
444 static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page)
447 struct zone *zone = page_zone(page);
448 struct page_cgroup *pc = lookup_page_cgroup(page);
450 spin_lock_irqsave(&zone->lru_lock, flags);
452 * Forget old LRU when this page_cgroup is *not* used. This Used bit
453 * is guarded by lock_page() because the page is SwapCache.
455 if (!PageCgroupUsed(pc))
456 mem_cgroup_del_lru_list(page, page_lru(page));
457 spin_unlock_irqrestore(&zone->lru_lock, flags);
460 static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page)
463 struct zone *zone = page_zone(page);
464 struct page_cgroup *pc = lookup_page_cgroup(page);
466 spin_lock_irqsave(&zone->lru_lock, flags);
467 /* link when the page is linked to LRU but page_cgroup isn't */
468 if (PageLRU(page) && list_empty(&pc->lru))
469 mem_cgroup_add_lru_list(page, page_lru(page));
470 spin_unlock_irqrestore(&zone->lru_lock, flags);
474 void mem_cgroup_move_lists(struct page *page,
475 enum lru_list from, enum lru_list to)
477 if (mem_cgroup_disabled())
479 mem_cgroup_del_lru_list(page, from);
480 mem_cgroup_add_lru_list(page, to);
483 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
486 struct mem_cgroup *curr = NULL;
490 curr = try_get_mem_cgroup_from_mm(task->mm);
495 if (curr->use_hierarchy)
496 ret = css_is_ancestor(&curr->css, &mem->css);
504 * prev_priority control...this will be used in memory reclaim path.
506 int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
510 spin_lock(&mem->reclaim_param_lock);
511 prev_priority = mem->prev_priority;
512 spin_unlock(&mem->reclaim_param_lock);
514 return prev_priority;
517 void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
519 spin_lock(&mem->reclaim_param_lock);
520 if (priority < mem->prev_priority)
521 mem->prev_priority = priority;
522 spin_unlock(&mem->reclaim_param_lock);
525 void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
527 spin_lock(&mem->reclaim_param_lock);
528 mem->prev_priority = priority;
529 spin_unlock(&mem->reclaim_param_lock);
532 static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages)
534 unsigned long active;
535 unsigned long inactive;
537 unsigned long inactive_ratio;
539 inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON);
540 active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON);
542 gb = (inactive + active) >> (30 - PAGE_SHIFT);
544 inactive_ratio = int_sqrt(10 * gb);
549 present_pages[0] = inactive;
550 present_pages[1] = active;
553 return inactive_ratio;
556 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg)
558 unsigned long active;
559 unsigned long inactive;
560 unsigned long present_pages[2];
561 unsigned long inactive_ratio;
563 inactive_ratio = calc_inactive_ratio(memcg, present_pages);
565 inactive = present_pages[0];
566 active = present_pages[1];
568 if (inactive * inactive_ratio < active)
574 int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg)
576 unsigned long active;
577 unsigned long inactive;
579 inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE);
580 active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE);
582 return (active > inactive);
585 unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg,
589 int nid = zone->zone_pgdat->node_id;
590 int zid = zone_idx(zone);
591 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
593 return MEM_CGROUP_ZSTAT(mz, lru);
596 struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
599 int nid = zone->zone_pgdat->node_id;
600 int zid = zone_idx(zone);
601 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
603 return &mz->reclaim_stat;
606 struct zone_reclaim_stat *
607 mem_cgroup_get_reclaim_stat_from_page(struct page *page)
609 struct page_cgroup *pc;
610 struct mem_cgroup_per_zone *mz;
612 if (mem_cgroup_disabled())
615 pc = lookup_page_cgroup(page);
617 * Used bit is set without atomic ops but after smp_wmb().
618 * For making pc->mem_cgroup visible, insert smp_rmb() here.
621 if (!PageCgroupUsed(pc))
624 mz = page_cgroup_zoneinfo(pc);
628 return &mz->reclaim_stat;
631 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
632 struct list_head *dst,
633 unsigned long *scanned, int order,
634 int mode, struct zone *z,
635 struct mem_cgroup *mem_cont,
636 int active, int file)
638 unsigned long nr_taken = 0;
642 struct list_head *src;
643 struct page_cgroup *pc, *tmp;
644 int nid = z->zone_pgdat->node_id;
645 int zid = zone_idx(z);
646 struct mem_cgroup_per_zone *mz;
647 int lru = LRU_FILE * !!file + !!active;
650 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
651 src = &mz->lists[lru];
654 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
655 if (scan >= nr_to_scan)
659 if (unlikely(!PageCgroupUsed(pc)))
661 if (unlikely(!PageLRU(page)))
665 if (__isolate_lru_page(page, mode, file) == 0) {
666 list_move(&page->lru, dst);
675 #define mem_cgroup_from_res_counter(counter, member) \
676 container_of(counter, struct mem_cgroup, member)
678 static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem)
680 if (do_swap_account) {
681 if (res_counter_check_under_limit(&mem->res) &&
682 res_counter_check_under_limit(&mem->memsw))
685 if (res_counter_check_under_limit(&mem->res))
690 static unsigned int get_swappiness(struct mem_cgroup *memcg)
692 struct cgroup *cgrp = memcg->css.cgroup;
693 unsigned int swappiness;
696 if (cgrp->parent == NULL)
697 return vm_swappiness;
699 spin_lock(&memcg->reclaim_param_lock);
700 swappiness = memcg->swappiness;
701 spin_unlock(&memcg->reclaim_param_lock);
706 static int mem_cgroup_count_children_cb(struct mem_cgroup *mem, void *data)
714 * mem_cgroup_print_mem_info: Called from OOM with tasklist_lock held in read mode.
715 * @memcg: The memory cgroup that went over limit
716 * @p: Task that is going to be killed
718 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
721 void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
723 struct cgroup *task_cgrp;
724 struct cgroup *mem_cgrp;
726 * Need a buffer in BSS, can't rely on allocations. The code relies
727 * on the assumption that OOM is serialized for memory controller.
728 * If this assumption is broken, revisit this code.
730 static char memcg_name[PATH_MAX];
739 mem_cgrp = memcg->css.cgroup;
740 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
742 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
745 * Unfortunately, we are unable to convert to a useful name
746 * But we'll still print out the usage information
753 printk(KERN_INFO "Task in %s killed", memcg_name);
756 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
764 * Continues from above, so we don't need an KERN_ level
766 printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
769 printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
770 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
771 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
772 res_counter_read_u64(&memcg->res, RES_FAILCNT));
773 printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
775 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
776 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
777 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
781 * This function returns the number of memcg under hierarchy tree. Returns
782 * 1(self count) if no children.
784 static int mem_cgroup_count_children(struct mem_cgroup *mem)
787 mem_cgroup_walk_tree(mem, &num, mem_cgroup_count_children_cb);
792 * Visit the first child (need not be the first child as per the ordering
793 * of the cgroup list, since we track last_scanned_child) of @mem and use
794 * that to reclaim free pages from.
796 static struct mem_cgroup *
797 mem_cgroup_select_victim(struct mem_cgroup *root_mem)
799 struct mem_cgroup *ret = NULL;
800 struct cgroup_subsys_state *css;
803 if (!root_mem->use_hierarchy) {
804 css_get(&root_mem->css);
810 nextid = root_mem->last_scanned_child + 1;
811 css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css,
813 if (css && css_tryget(css))
814 ret = container_of(css, struct mem_cgroup, css);
817 /* Updates scanning parameter */
818 spin_lock(&root_mem->reclaim_param_lock);
820 /* this means start scan from ID:1 */
821 root_mem->last_scanned_child = 0;
823 root_mem->last_scanned_child = found;
824 spin_unlock(&root_mem->reclaim_param_lock);
831 * Scan the hierarchy if needed to reclaim memory. We remember the last child
832 * we reclaimed from, so that we don't end up penalizing one child extensively
833 * based on its position in the children list.
835 * root_mem is the original ancestor that we've been reclaim from.
837 * We give up and return to the caller when we visit root_mem twice.
838 * (other groups can be removed while we're walking....)
840 * If shrink==true, for avoiding to free too much, this returns immedieately.
842 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
843 gfp_t gfp_mask, bool noswap, bool shrink)
845 struct mem_cgroup *victim;
850 victim = mem_cgroup_select_victim(root_mem);
851 if (victim == root_mem)
853 if (!mem_cgroup_local_usage(&victim->stat)) {
854 /* this cgroup's local usage == 0 */
855 css_put(&victim->css);
858 /* we use swappiness of local cgroup */
859 ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, noswap,
860 get_swappiness(victim));
861 css_put(&victim->css);
863 * At shrinking usage, we can't check we should stop here or
864 * reclaim more. It's depends on callers. last_scanned_child
865 * will work enough for keeping fairness under tree.
870 if (mem_cgroup_check_under_limit(root_mem))
876 bool mem_cgroup_oom_called(struct task_struct *task)
879 struct mem_cgroup *mem;
880 struct mm_struct *mm;
886 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
887 if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10))
893 static int record_last_oom_cb(struct mem_cgroup *mem, void *data)
895 mem->last_oom_jiffies = jiffies;
899 static void record_last_oom(struct mem_cgroup *mem)
901 mem_cgroup_walk_tree(mem, NULL, record_last_oom_cb);
905 * Currently used to update mapped file statistics, but the routine can be
906 * generalized to update other statistics as well.
908 void mem_cgroup_update_mapped_file_stat(struct page *page, int val)
910 struct mem_cgroup *mem;
911 struct mem_cgroup_stat *stat;
912 struct mem_cgroup_stat_cpu *cpustat;
914 struct page_cgroup *pc;
916 if (!page_is_file_cache(page))
919 pc = lookup_page_cgroup(page);
923 lock_page_cgroup(pc);
924 mem = pc->mem_cgroup;
928 if (!PageCgroupUsed(pc))
932 * Preemption is already disabled, we don't need get_cpu()
934 cpu = smp_processor_id();
936 cpustat = &stat->cpustat[cpu];
938 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_MAPPED_FILE, val);
940 unlock_page_cgroup(pc);
944 * Unlike exported interface, "oom" parameter is added. if oom==true,
945 * oom-killer can be invoked.
947 static int __mem_cgroup_try_charge(struct mm_struct *mm,
948 gfp_t gfp_mask, struct mem_cgroup **memcg,
951 struct mem_cgroup *mem, *mem_over_limit;
952 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
953 struct res_counter *fail_res;
955 if (unlikely(test_thread_flag(TIF_MEMDIE))) {
956 /* Don't account this! */
962 * We always charge the cgroup the mm_struct belongs to.
963 * The mm_struct's mem_cgroup changes on task migration if the
964 * thread group leader migrates. It's possible that mm is not
965 * set, if so charge the init_mm (happens for pagecache usage).
969 mem = try_get_mem_cgroup_from_mm(mm);
977 VM_BUG_ON(css_is_removed(&mem->css));
983 ret = res_counter_charge(&mem->res, PAGE_SIZE, &fail_res);
985 if (!do_swap_account)
987 ret = res_counter_charge(&mem->memsw, PAGE_SIZE,
991 /* mem+swap counter fails */
992 res_counter_uncharge(&mem->res, PAGE_SIZE);
994 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
997 /* mem counter fails */
998 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
1001 if (!(gfp_mask & __GFP_WAIT))
1004 ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, gfp_mask,
1010 * try_to_free_mem_cgroup_pages() might not give us a full
1011 * picture of reclaim. Some pages are reclaimed and might be
1012 * moved to swap cache or just unmapped from the cgroup.
1013 * Check the limit again to see if the reclaim reduced the
1014 * current usage of the cgroup before giving up
1017 if (mem_cgroup_check_under_limit(mem_over_limit))
1020 if (!nr_retries--) {
1022 mutex_lock(&memcg_tasklist);
1023 mem_cgroup_out_of_memory(mem_over_limit, gfp_mask);
1024 mutex_unlock(&memcg_tasklist);
1025 record_last_oom(mem_over_limit);
1038 * A helper function to get mem_cgroup from ID. must be called under
1039 * rcu_read_lock(). The caller must check css_is_removed() or some if
1040 * it's concern. (dropping refcnt from swap can be called against removed
1043 static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
1045 struct cgroup_subsys_state *css;
1047 /* ID 0 is unused ID */
1050 css = css_lookup(&mem_cgroup_subsys, id);
1053 return container_of(css, struct mem_cgroup, css);
1056 static struct mem_cgroup *try_get_mem_cgroup_from_swapcache(struct page *page)
1058 struct mem_cgroup *mem;
1059 struct page_cgroup *pc;
1063 VM_BUG_ON(!PageLocked(page));
1065 if (!PageSwapCache(page))
1068 pc = lookup_page_cgroup(page);
1069 lock_page_cgroup(pc);
1070 if (PageCgroupUsed(pc)) {
1071 mem = pc->mem_cgroup;
1072 if (mem && !css_tryget(&mem->css))
1075 ent.val = page_private(page);
1076 id = lookup_swap_cgroup(ent);
1078 mem = mem_cgroup_lookup(id);
1079 if (mem && !css_tryget(&mem->css))
1083 unlock_page_cgroup(pc);
1088 * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be
1089 * USED state. If already USED, uncharge and return.
1092 static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
1093 struct page_cgroup *pc,
1094 enum charge_type ctype)
1096 /* try_charge() can return NULL to *memcg, taking care of it. */
1100 lock_page_cgroup(pc);
1101 if (unlikely(PageCgroupUsed(pc))) {
1102 unlock_page_cgroup(pc);
1103 res_counter_uncharge(&mem->res, PAGE_SIZE);
1104 if (do_swap_account)
1105 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1109 pc->mem_cgroup = mem;
1111 pc->flags = pcg_default_flags[ctype];
1113 mem_cgroup_charge_statistics(mem, pc, true);
1115 unlock_page_cgroup(pc);
1119 * mem_cgroup_move_account - move account of the page
1120 * @pc: page_cgroup of the page.
1121 * @from: mem_cgroup which the page is moved from.
1122 * @to: mem_cgroup which the page is moved to. @from != @to.
1124 * The caller must confirm following.
1125 * - page is not on LRU (isolate_page() is useful.)
1127 * returns 0 at success,
1128 * returns -EBUSY when lock is busy or "pc" is unstable.
1130 * This function does "uncharge" from old cgroup but doesn't do "charge" to
1131 * new cgroup. It should be done by a caller.
1134 static int mem_cgroup_move_account(struct page_cgroup *pc,
1135 struct mem_cgroup *from, struct mem_cgroup *to)
1137 struct mem_cgroup_per_zone *from_mz, *to_mz;
1142 struct mem_cgroup_stat *stat;
1143 struct mem_cgroup_stat_cpu *cpustat;
1145 VM_BUG_ON(from == to);
1146 VM_BUG_ON(PageLRU(pc->page));
1148 nid = page_cgroup_nid(pc);
1149 zid = page_cgroup_zid(pc);
1150 from_mz = mem_cgroup_zoneinfo(from, nid, zid);
1151 to_mz = mem_cgroup_zoneinfo(to, nid, zid);
1153 if (!trylock_page_cgroup(pc))
1156 if (!PageCgroupUsed(pc))
1159 if (pc->mem_cgroup != from)
1162 res_counter_uncharge(&from->res, PAGE_SIZE);
1163 mem_cgroup_charge_statistics(from, pc, false);
1166 if (page_is_file_cache(page) && page_mapped(page)) {
1167 cpu = smp_processor_id();
1168 /* Update mapped_file data for mem_cgroup "from" */
1170 cpustat = &stat->cpustat[cpu];
1171 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_MAPPED_FILE,
1174 /* Update mapped_file data for mem_cgroup "to" */
1176 cpustat = &stat->cpustat[cpu];
1177 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_MAPPED_FILE,
1181 if (do_swap_account)
1182 res_counter_uncharge(&from->memsw, PAGE_SIZE);
1183 css_put(&from->css);
1186 pc->mem_cgroup = to;
1187 mem_cgroup_charge_statistics(to, pc, true);
1190 unlock_page_cgroup(pc);
1195 * move charges to its parent.
1198 static int mem_cgroup_move_parent(struct page_cgroup *pc,
1199 struct mem_cgroup *child,
1202 struct page *page = pc->page;
1203 struct cgroup *cg = child->css.cgroup;
1204 struct cgroup *pcg = cg->parent;
1205 struct mem_cgroup *parent;
1213 parent = mem_cgroup_from_cont(pcg);
1216 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false);
1220 if (!get_page_unless_zero(page)) {
1225 ret = isolate_lru_page(page);
1230 ret = mem_cgroup_move_account(pc, child, parent);
1232 putback_lru_page(page);
1235 /* drop extra refcnt by try_charge() */
1236 css_put(&parent->css);
1243 /* drop extra refcnt by try_charge() */
1244 css_put(&parent->css);
1245 /* uncharge if move fails */
1246 res_counter_uncharge(&parent->res, PAGE_SIZE);
1247 if (do_swap_account)
1248 res_counter_uncharge(&parent->memsw, PAGE_SIZE);
1253 * Charge the memory controller for page usage.
1255 * 0 if the charge was successful
1256 * < 0 if the cgroup is over its limit
1258 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
1259 gfp_t gfp_mask, enum charge_type ctype,
1260 struct mem_cgroup *memcg)
1262 struct mem_cgroup *mem;
1263 struct page_cgroup *pc;
1266 pc = lookup_page_cgroup(page);
1267 /* can happen at boot */
1273 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true);
1277 __mem_cgroup_commit_charge(mem, pc, ctype);
1281 int mem_cgroup_newpage_charge(struct page *page,
1282 struct mm_struct *mm, gfp_t gfp_mask)
1284 if (mem_cgroup_disabled())
1286 if (PageCompound(page))
1289 * If already mapped, we don't have to account.
1290 * If page cache, page->mapping has address_space.
1291 * But page->mapping may have out-of-use anon_vma pointer,
1292 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
1295 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
1299 return mem_cgroup_charge_common(page, mm, gfp_mask,
1300 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
1304 __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
1305 enum charge_type ctype);
1307 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
1310 struct mem_cgroup *mem = NULL;
1313 if (mem_cgroup_disabled())
1315 if (PageCompound(page))
1318 * Corner case handling. This is called from add_to_page_cache()
1319 * in usual. But some FS (shmem) precharges this page before calling it
1320 * and call add_to_page_cache() with GFP_NOWAIT.
1322 * For GFP_NOWAIT case, the page may be pre-charged before calling
1323 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
1324 * charge twice. (It works but has to pay a bit larger cost.)
1325 * And when the page is SwapCache, it should take swap information
1326 * into account. This is under lock_page() now.
1328 if (!(gfp_mask & __GFP_WAIT)) {
1329 struct page_cgroup *pc;
1332 pc = lookup_page_cgroup(page);
1335 lock_page_cgroup(pc);
1336 if (PageCgroupUsed(pc)) {
1337 unlock_page_cgroup(pc);
1340 unlock_page_cgroup(pc);
1343 if (unlikely(!mm && !mem))
1346 if (page_is_file_cache(page))
1347 return mem_cgroup_charge_common(page, mm, gfp_mask,
1348 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
1351 if (PageSwapCache(page)) {
1352 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
1354 __mem_cgroup_commit_charge_swapin(page, mem,
1355 MEM_CGROUP_CHARGE_TYPE_SHMEM);
1357 ret = mem_cgroup_charge_common(page, mm, gfp_mask,
1358 MEM_CGROUP_CHARGE_TYPE_SHMEM, mem);
1364 * While swap-in, try_charge -> commit or cancel, the page is locked.
1365 * And when try_charge() successfully returns, one refcnt to memcg without
1366 * struct page_cgroup is aquired. This refcnt will be cumsumed by
1367 * "commit()" or removed by "cancel()"
1369 int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
1371 gfp_t mask, struct mem_cgroup **ptr)
1373 struct mem_cgroup *mem;
1376 if (mem_cgroup_disabled())
1379 if (!do_swap_account)
1382 * A racing thread's fault, or swapoff, may have already updated
1383 * the pte, and even removed page from swap cache: return success
1384 * to go on to do_swap_page()'s pte_same() test, which should fail.
1386 if (!PageSwapCache(page))
1388 mem = try_get_mem_cgroup_from_swapcache(page);
1392 ret = __mem_cgroup_try_charge(NULL, mask, ptr, true);
1393 /* drop extra refcnt from tryget */
1399 return __mem_cgroup_try_charge(mm, mask, ptr, true);
1403 __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
1404 enum charge_type ctype)
1406 struct page_cgroup *pc;
1408 if (mem_cgroup_disabled())
1412 pc = lookup_page_cgroup(page);
1413 mem_cgroup_lru_del_before_commit_swapcache(page);
1414 __mem_cgroup_commit_charge(ptr, pc, ctype);
1415 mem_cgroup_lru_add_after_commit_swapcache(page);
1417 * Now swap is on-memory. This means this page may be
1418 * counted both as mem and swap....double count.
1419 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
1420 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
1421 * may call delete_from_swap_cache() before reach here.
1423 if (do_swap_account && PageSwapCache(page)) {
1424 swp_entry_t ent = {.val = page_private(page)};
1426 struct mem_cgroup *memcg;
1428 id = swap_cgroup_record(ent, 0);
1430 memcg = mem_cgroup_lookup(id);
1433 * This recorded memcg can be obsolete one. So, avoid
1434 * calling css_tryget
1436 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1437 mem_cgroup_put(memcg);
1441 /* add this page(page_cgroup) to the LRU we want. */
1445 void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
1447 __mem_cgroup_commit_charge_swapin(page, ptr,
1448 MEM_CGROUP_CHARGE_TYPE_MAPPED);
1451 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
1453 if (mem_cgroup_disabled())
1457 res_counter_uncharge(&mem->res, PAGE_SIZE);
1458 if (do_swap_account)
1459 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1465 * uncharge if !page_mapped(page)
1467 static struct mem_cgroup *
1468 __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
1470 struct page_cgroup *pc;
1471 struct mem_cgroup *mem = NULL;
1472 struct mem_cgroup_per_zone *mz;
1474 if (mem_cgroup_disabled())
1477 if (PageSwapCache(page))
1481 * Check if our page_cgroup is valid
1483 pc = lookup_page_cgroup(page);
1484 if (unlikely(!pc || !PageCgroupUsed(pc)))
1487 lock_page_cgroup(pc);
1489 mem = pc->mem_cgroup;
1491 if (!PageCgroupUsed(pc))
1495 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
1496 if (page_mapped(page))
1499 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
1500 if (!PageAnon(page)) { /* Shared memory */
1501 if (page->mapping && !page_is_file_cache(page))
1503 } else if (page_mapped(page)) /* Anon */
1510 res_counter_uncharge(&mem->res, PAGE_SIZE);
1511 if (do_swap_account && (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT))
1512 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1513 mem_cgroup_charge_statistics(mem, pc, false);
1515 ClearPageCgroupUsed(pc);
1517 * pc->mem_cgroup is not cleared here. It will be accessed when it's
1518 * freed from LRU. This is safe because uncharged page is expected not
1519 * to be reused (freed soon). Exception is SwapCache, it's handled by
1520 * special functions.
1523 mz = page_cgroup_zoneinfo(pc);
1524 unlock_page_cgroup(pc);
1526 /* at swapout, this memcg will be accessed to record to swap */
1527 if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
1533 unlock_page_cgroup(pc);
1537 void mem_cgroup_uncharge_page(struct page *page)
1540 if (page_mapped(page))
1542 if (page->mapping && !PageAnon(page))
1544 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
1547 void mem_cgroup_uncharge_cache_page(struct page *page)
1549 VM_BUG_ON(page_mapped(page));
1550 VM_BUG_ON(page->mapping);
1551 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
1556 * called after __delete_from_swap_cache() and drop "page" account.
1557 * memcg information is recorded to swap_cgroup of "ent"
1559 void mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent)
1561 struct mem_cgroup *memcg;
1563 memcg = __mem_cgroup_uncharge_common(page,
1564 MEM_CGROUP_CHARGE_TYPE_SWAPOUT);
1565 /* record memcg information */
1566 if (do_swap_account && memcg) {
1567 swap_cgroup_record(ent, css_id(&memcg->css));
1568 mem_cgroup_get(memcg);
1571 css_put(&memcg->css);
1575 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1577 * called from swap_entry_free(). remove record in swap_cgroup and
1578 * uncharge "memsw" account.
1580 void mem_cgroup_uncharge_swap(swp_entry_t ent)
1582 struct mem_cgroup *memcg;
1585 if (!do_swap_account)
1588 id = swap_cgroup_record(ent, 0);
1590 memcg = mem_cgroup_lookup(id);
1593 * We uncharge this because swap is freed.
1594 * This memcg can be obsolete one. We avoid calling css_tryget
1596 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1597 mem_cgroup_put(memcg);
1604 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
1607 int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr)
1609 struct page_cgroup *pc;
1610 struct mem_cgroup *mem = NULL;
1613 if (mem_cgroup_disabled())
1616 pc = lookup_page_cgroup(page);
1617 lock_page_cgroup(pc);
1618 if (PageCgroupUsed(pc)) {
1619 mem = pc->mem_cgroup;
1622 unlock_page_cgroup(pc);
1625 ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false);
1632 /* remove redundant charge if migration failed*/
1633 void mem_cgroup_end_migration(struct mem_cgroup *mem,
1634 struct page *oldpage, struct page *newpage)
1636 struct page *target, *unused;
1637 struct page_cgroup *pc;
1638 enum charge_type ctype;
1643 /* at migration success, oldpage->mapping is NULL. */
1644 if (oldpage->mapping) {
1652 if (PageAnon(target))
1653 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
1654 else if (page_is_file_cache(target))
1655 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
1657 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
1659 /* unused page is not on radix-tree now. */
1661 __mem_cgroup_uncharge_common(unused, ctype);
1663 pc = lookup_page_cgroup(target);
1665 * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup.
1666 * So, double-counting is effectively avoided.
1668 __mem_cgroup_commit_charge(mem, pc, ctype);
1671 * Both of oldpage and newpage are still under lock_page().
1672 * Then, we don't have to care about race in radix-tree.
1673 * But we have to be careful that this page is unmapped or not.
1675 * There is a case for !page_mapped(). At the start of
1676 * migration, oldpage was mapped. But now, it's zapped.
1677 * But we know *target* page is not freed/reused under us.
1678 * mem_cgroup_uncharge_page() does all necessary checks.
1680 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
1681 mem_cgroup_uncharge_page(target);
1685 * A call to try to shrink memory usage on charge failure at shmem's swapin.
1686 * Calling hierarchical_reclaim is not enough because we should update
1687 * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM.
1688 * Moreover considering hierarchy, we should reclaim from the mem_over_limit,
1689 * not from the memcg which this page would be charged to.
1690 * try_charge_swapin does all of these works properly.
1692 int mem_cgroup_shmem_charge_fallback(struct page *page,
1693 struct mm_struct *mm,
1696 struct mem_cgroup *mem = NULL;
1699 if (mem_cgroup_disabled())
1702 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
1704 mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */
1709 static DEFINE_MUTEX(set_limit_mutex);
1711 static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
1712 unsigned long long val)
1718 int children = mem_cgroup_count_children(memcg);
1719 u64 curusage, oldusage;
1722 * For keeping hierarchical_reclaim simple, how long we should retry
1723 * is depends on callers. We set our retry-count to be function
1724 * of # of children which we should visit in this loop.
1726 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
1728 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
1730 while (retry_count) {
1731 if (signal_pending(current)) {
1736 * Rather than hide all in some function, I do this in
1737 * open coded manner. You see what this really does.
1738 * We have to guarantee mem->res.limit < mem->memsw.limit.
1740 mutex_lock(&set_limit_mutex);
1741 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1742 if (memswlimit < val) {
1744 mutex_unlock(&set_limit_mutex);
1747 ret = res_counter_set_limit(&memcg->res, val);
1748 mutex_unlock(&set_limit_mutex);
1753 progress = mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL,
1755 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
1756 /* Usage is reduced ? */
1757 if (curusage >= oldusage)
1760 oldusage = curusage;
1766 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
1767 unsigned long long val)
1770 u64 memlimit, oldusage, curusage;
1771 int children = mem_cgroup_count_children(memcg);
1774 /* see mem_cgroup_resize_res_limit */
1775 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
1776 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
1777 while (retry_count) {
1778 if (signal_pending(current)) {
1783 * Rather than hide all in some function, I do this in
1784 * open coded manner. You see what this really does.
1785 * We have to guarantee mem->res.limit < mem->memsw.limit.
1787 mutex_lock(&set_limit_mutex);
1788 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1789 if (memlimit > val) {
1791 mutex_unlock(&set_limit_mutex);
1794 ret = res_counter_set_limit(&memcg->memsw, val);
1795 mutex_unlock(&set_limit_mutex);
1800 mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL, true, true);
1801 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
1802 /* Usage is reduced ? */
1803 if (curusage >= oldusage)
1806 oldusage = curusage;
1812 * This routine traverse page_cgroup in given list and drop them all.
1813 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
1815 static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
1816 int node, int zid, enum lru_list lru)
1819 struct mem_cgroup_per_zone *mz;
1820 struct page_cgroup *pc, *busy;
1821 unsigned long flags, loop;
1822 struct list_head *list;
1825 zone = &NODE_DATA(node)->node_zones[zid];
1826 mz = mem_cgroup_zoneinfo(mem, node, zid);
1827 list = &mz->lists[lru];
1829 loop = MEM_CGROUP_ZSTAT(mz, lru);
1830 /* give some margin against EBUSY etc...*/
1835 spin_lock_irqsave(&zone->lru_lock, flags);
1836 if (list_empty(list)) {
1837 spin_unlock_irqrestore(&zone->lru_lock, flags);
1840 pc = list_entry(list->prev, struct page_cgroup, lru);
1842 list_move(&pc->lru, list);
1844 spin_unlock_irqrestore(&zone->lru_lock, flags);
1847 spin_unlock_irqrestore(&zone->lru_lock, flags);
1849 ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL);
1853 if (ret == -EBUSY || ret == -EINVAL) {
1854 /* found lock contention or "pc" is obsolete. */
1861 if (!ret && !list_empty(list))
1867 * make mem_cgroup's charge to be 0 if there is no task.
1868 * This enables deleting this mem_cgroup.
1870 static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
1873 int node, zid, shrink;
1874 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
1875 struct cgroup *cgrp = mem->css.cgroup;
1880 /* should free all ? */
1884 while (mem->res.usage > 0) {
1886 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
1889 if (signal_pending(current))
1891 /* This is for making all *used* pages to be on LRU. */
1892 lru_add_drain_all();
1894 for_each_node_state(node, N_HIGH_MEMORY) {
1895 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
1898 ret = mem_cgroup_force_empty_list(mem,
1907 /* it seems parent cgroup doesn't have enough mem */
1918 /* returns EBUSY if there is a task or if we come here twice. */
1919 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
1923 /* we call try-to-free pages for make this cgroup empty */
1924 lru_add_drain_all();
1925 /* try to free all pages in this cgroup */
1927 while (nr_retries && mem->res.usage > 0) {
1930 if (signal_pending(current)) {
1934 progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL,
1935 false, get_swappiness(mem));
1938 /* maybe some writeback is necessary */
1939 congestion_wait(WRITE, HZ/10);
1944 /* try move_account...there may be some *locked* pages. */
1951 int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
1953 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
1957 static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
1959 return mem_cgroup_from_cont(cont)->use_hierarchy;
1962 static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
1966 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1967 struct cgroup *parent = cont->parent;
1968 struct mem_cgroup *parent_mem = NULL;
1971 parent_mem = mem_cgroup_from_cont(parent);
1975 * If parent's use_hiearchy is set, we can't make any modifications
1976 * in the child subtrees. If it is unset, then the change can
1977 * occur, provided the current cgroup has no children.
1979 * For the root cgroup, parent_mem is NULL, we allow value to be
1980 * set if there are no children.
1982 if ((!parent_mem || !parent_mem->use_hierarchy) &&
1983 (val == 1 || val == 0)) {
1984 if (list_empty(&cont->children))
1985 mem->use_hierarchy = val;
1995 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
1997 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2001 type = MEMFILE_TYPE(cft->private);
2002 name = MEMFILE_ATTR(cft->private);
2005 val = res_counter_read_u64(&mem->res, name);
2008 val = res_counter_read_u64(&mem->memsw, name);
2017 * The user of this function is...
2020 static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
2023 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
2025 unsigned long long val;
2028 type = MEMFILE_TYPE(cft->private);
2029 name = MEMFILE_ATTR(cft->private);
2032 /* This function does all necessary parse...reuse it */
2033 ret = res_counter_memparse_write_strategy(buffer, &val);
2037 ret = mem_cgroup_resize_limit(memcg, val);
2039 ret = mem_cgroup_resize_memsw_limit(memcg, val);
2042 ret = -EINVAL; /* should be BUG() ? */
2048 static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
2049 unsigned long long *mem_limit, unsigned long long *memsw_limit)
2051 struct cgroup *cgroup;
2052 unsigned long long min_limit, min_memsw_limit, tmp;
2054 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
2055 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2056 cgroup = memcg->css.cgroup;
2057 if (!memcg->use_hierarchy)
2060 while (cgroup->parent) {
2061 cgroup = cgroup->parent;
2062 memcg = mem_cgroup_from_cont(cgroup);
2063 if (!memcg->use_hierarchy)
2065 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
2066 min_limit = min(min_limit, tmp);
2067 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2068 min_memsw_limit = min(min_memsw_limit, tmp);
2071 *mem_limit = min_limit;
2072 *memsw_limit = min_memsw_limit;
2076 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
2078 struct mem_cgroup *mem;
2081 mem = mem_cgroup_from_cont(cont);
2082 type = MEMFILE_TYPE(event);
2083 name = MEMFILE_ATTR(event);
2087 res_counter_reset_max(&mem->res);
2089 res_counter_reset_max(&mem->memsw);
2093 res_counter_reset_failcnt(&mem->res);
2095 res_counter_reset_failcnt(&mem->memsw);
2102 /* For read statistics */
2117 struct mcs_total_stat {
2118 s64 stat[NR_MCS_STAT];
2124 } memcg_stat_strings[NR_MCS_STAT] = {
2125 {"cache", "total_cache"},
2126 {"rss", "total_rss"},
2127 {"mapped_file", "total_mapped_file"},
2128 {"pgpgin", "total_pgpgin"},
2129 {"pgpgout", "total_pgpgout"},
2130 {"inactive_anon", "total_inactive_anon"},
2131 {"active_anon", "total_active_anon"},
2132 {"inactive_file", "total_inactive_file"},
2133 {"active_file", "total_active_file"},
2134 {"unevictable", "total_unevictable"}
2138 static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data)
2140 struct mcs_total_stat *s = data;
2144 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_CACHE);
2145 s->stat[MCS_CACHE] += val * PAGE_SIZE;
2146 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
2147 s->stat[MCS_RSS] += val * PAGE_SIZE;
2148 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_MAPPED_FILE);
2149 s->stat[MCS_MAPPED_FILE] += val * PAGE_SIZE;
2150 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGIN_COUNT);
2151 s->stat[MCS_PGPGIN] += val;
2152 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGOUT_COUNT);
2153 s->stat[MCS_PGPGOUT] += val;
2156 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON);
2157 s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE;
2158 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON);
2159 s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE;
2160 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE);
2161 s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE;
2162 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE);
2163 s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE;
2164 val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE);
2165 s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE;
2170 mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s)
2172 mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat);
2175 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
2176 struct cgroup_map_cb *cb)
2178 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
2179 struct mcs_total_stat mystat;
2182 memset(&mystat, 0, sizeof(mystat));
2183 mem_cgroup_get_local_stat(mem_cont, &mystat);
2185 for (i = 0; i < NR_MCS_STAT; i++)
2186 cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
2188 /* Hierarchical information */
2190 unsigned long long limit, memsw_limit;
2191 memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit);
2192 cb->fill(cb, "hierarchical_memory_limit", limit);
2193 if (do_swap_account)
2194 cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
2197 memset(&mystat, 0, sizeof(mystat));
2198 mem_cgroup_get_total_stat(mem_cont, &mystat);
2199 for (i = 0; i < NR_MCS_STAT; i++)
2200 cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
2203 #ifdef CONFIG_DEBUG_VM
2204 cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL));
2208 struct mem_cgroup_per_zone *mz;
2209 unsigned long recent_rotated[2] = {0, 0};
2210 unsigned long recent_scanned[2] = {0, 0};
2212 for_each_online_node(nid)
2213 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
2214 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
2216 recent_rotated[0] +=
2217 mz->reclaim_stat.recent_rotated[0];
2218 recent_rotated[1] +=
2219 mz->reclaim_stat.recent_rotated[1];
2220 recent_scanned[0] +=
2221 mz->reclaim_stat.recent_scanned[0];
2222 recent_scanned[1] +=
2223 mz->reclaim_stat.recent_scanned[1];
2225 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
2226 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
2227 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
2228 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
2235 static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
2237 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
2239 return get_swappiness(memcg);
2242 static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
2245 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
2246 struct mem_cgroup *parent;
2251 if (cgrp->parent == NULL)
2254 parent = mem_cgroup_from_cont(cgrp->parent);
2258 /* If under hierarchy, only empty-root can set this value */
2259 if ((parent->use_hierarchy) ||
2260 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
2265 spin_lock(&memcg->reclaim_param_lock);
2266 memcg->swappiness = val;
2267 spin_unlock(&memcg->reclaim_param_lock);
2275 static struct cftype mem_cgroup_files[] = {
2277 .name = "usage_in_bytes",
2278 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
2279 .read_u64 = mem_cgroup_read,
2282 .name = "max_usage_in_bytes",
2283 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
2284 .trigger = mem_cgroup_reset,
2285 .read_u64 = mem_cgroup_read,
2288 .name = "limit_in_bytes",
2289 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
2290 .write_string = mem_cgroup_write,
2291 .read_u64 = mem_cgroup_read,
2295 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
2296 .trigger = mem_cgroup_reset,
2297 .read_u64 = mem_cgroup_read,
2301 .read_map = mem_control_stat_show,
2304 .name = "force_empty",
2305 .trigger = mem_cgroup_force_empty_write,
2308 .name = "use_hierarchy",
2309 .write_u64 = mem_cgroup_hierarchy_write,
2310 .read_u64 = mem_cgroup_hierarchy_read,
2313 .name = "swappiness",
2314 .read_u64 = mem_cgroup_swappiness_read,
2315 .write_u64 = mem_cgroup_swappiness_write,
2319 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2320 static struct cftype memsw_cgroup_files[] = {
2322 .name = "memsw.usage_in_bytes",
2323 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
2324 .read_u64 = mem_cgroup_read,
2327 .name = "memsw.max_usage_in_bytes",
2328 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
2329 .trigger = mem_cgroup_reset,
2330 .read_u64 = mem_cgroup_read,
2333 .name = "memsw.limit_in_bytes",
2334 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
2335 .write_string = mem_cgroup_write,
2336 .read_u64 = mem_cgroup_read,
2339 .name = "memsw.failcnt",
2340 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
2341 .trigger = mem_cgroup_reset,
2342 .read_u64 = mem_cgroup_read,
2346 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
2348 if (!do_swap_account)
2350 return cgroup_add_files(cont, ss, memsw_cgroup_files,
2351 ARRAY_SIZE(memsw_cgroup_files));
2354 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
2360 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
2362 struct mem_cgroup_per_node *pn;
2363 struct mem_cgroup_per_zone *mz;
2365 int zone, tmp = node;
2367 * This routine is called against possible nodes.
2368 * But it's BUG to call kmalloc() against offline node.
2370 * TODO: this routine can waste much memory for nodes which will
2371 * never be onlined. It's better to use memory hotplug callback
2374 if (!node_state(node, N_NORMAL_MEMORY))
2376 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
2380 mem->info.nodeinfo[node] = pn;
2381 memset(pn, 0, sizeof(*pn));
2383 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
2384 mz = &pn->zoneinfo[zone];
2386 INIT_LIST_HEAD(&mz->lists[l]);
2391 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
2393 kfree(mem->info.nodeinfo[node]);
2396 static int mem_cgroup_size(void)
2398 int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu);
2399 return sizeof(struct mem_cgroup) + cpustat_size;
2402 static struct mem_cgroup *mem_cgroup_alloc(void)
2404 struct mem_cgroup *mem;
2405 int size = mem_cgroup_size();
2407 if (size < PAGE_SIZE)
2408 mem = kmalloc(size, GFP_KERNEL);
2410 mem = vmalloc(size);
2413 memset(mem, 0, size);
2418 * At destroying mem_cgroup, references from swap_cgroup can remain.
2419 * (scanning all at force_empty is too costly...)
2421 * Instead of clearing all references at force_empty, we remember
2422 * the number of reference from swap_cgroup and free mem_cgroup when
2423 * it goes down to 0.
2425 * Removal of cgroup itself succeeds regardless of refs from swap.
2428 static void __mem_cgroup_free(struct mem_cgroup *mem)
2432 free_css_id(&mem_cgroup_subsys, &mem->css);
2434 for_each_node_state(node, N_POSSIBLE)
2435 free_mem_cgroup_per_zone_info(mem, node);
2437 if (mem_cgroup_size() < PAGE_SIZE)
2443 static void mem_cgroup_get(struct mem_cgroup *mem)
2445 atomic_inc(&mem->refcnt);
2448 static void mem_cgroup_put(struct mem_cgroup *mem)
2450 if (atomic_dec_and_test(&mem->refcnt)) {
2451 struct mem_cgroup *parent = parent_mem_cgroup(mem);
2452 __mem_cgroup_free(mem);
2454 mem_cgroup_put(parent);
2459 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
2461 static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem)
2463 if (!mem->res.parent)
2465 return mem_cgroup_from_res_counter(mem->res.parent, res);
2468 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2469 static void __init enable_swap_cgroup(void)
2471 if (!mem_cgroup_disabled() && really_do_swap_account)
2472 do_swap_account = 1;
2475 static void __init enable_swap_cgroup(void)
2480 static struct cgroup_subsys_state * __ref
2481 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
2483 struct mem_cgroup *mem, *parent;
2484 long error = -ENOMEM;
2487 mem = mem_cgroup_alloc();
2489 return ERR_PTR(error);
2491 for_each_node_state(node, N_POSSIBLE)
2492 if (alloc_mem_cgroup_per_zone_info(mem, node))
2495 if (cont->parent == NULL) {
2496 enable_swap_cgroup();
2499 parent = mem_cgroup_from_cont(cont->parent);
2500 mem->use_hierarchy = parent->use_hierarchy;
2503 if (parent && parent->use_hierarchy) {
2504 res_counter_init(&mem->res, &parent->res);
2505 res_counter_init(&mem->memsw, &parent->memsw);
2507 * We increment refcnt of the parent to ensure that we can
2508 * safely access it on res_counter_charge/uncharge.
2509 * This refcnt will be decremented when freeing this
2510 * mem_cgroup(see mem_cgroup_put).
2512 mem_cgroup_get(parent);
2514 res_counter_init(&mem->res, NULL);
2515 res_counter_init(&mem->memsw, NULL);
2517 mem->last_scanned_child = 0;
2518 spin_lock_init(&mem->reclaim_param_lock);
2521 mem->swappiness = get_swappiness(parent);
2522 atomic_set(&mem->refcnt, 1);
2525 __mem_cgroup_free(mem);
2526 return ERR_PTR(error);
2529 static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
2530 struct cgroup *cont)
2532 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2534 return mem_cgroup_force_empty(mem, false);
2537 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
2538 struct cgroup *cont)
2540 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2542 mem_cgroup_put(mem);
2545 static int mem_cgroup_populate(struct cgroup_subsys *ss,
2546 struct cgroup *cont)
2550 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
2551 ARRAY_SIZE(mem_cgroup_files));
2554 ret = register_memsw_files(cont, ss);
2558 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
2559 struct cgroup *cont,
2560 struct cgroup *old_cont,
2561 struct task_struct *p)
2563 mutex_lock(&memcg_tasklist);
2565 * FIXME: It's better to move charges of this process from old
2566 * memcg to new memcg. But it's just on TODO-List now.
2568 mutex_unlock(&memcg_tasklist);
2571 struct cgroup_subsys mem_cgroup_subsys = {
2573 .subsys_id = mem_cgroup_subsys_id,
2574 .create = mem_cgroup_create,
2575 .pre_destroy = mem_cgroup_pre_destroy,
2576 .destroy = mem_cgroup_destroy,
2577 .populate = mem_cgroup_populate,
2578 .attach = mem_cgroup_move_task,
2583 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2585 static int __init disable_swap_account(char *s)
2587 really_do_swap_account = 0;
2590 __setup("noswapaccount", disable_swap_account);