1 /* memcontrol.c - Memory Controller
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
6 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2 of the License, or
12 * (at your option) any later version.
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
20 #include <linux/res_counter.h>
21 #include <linux/memcontrol.h>
22 #include <linux/cgroup.h>
24 #include <linux/pagemap.h>
25 #include <linux/smp.h>
26 #include <linux/page-flags.h>
27 #include <linux/backing-dev.h>
28 #include <linux/bit_spinlock.h>
29 #include <linux/rcupdate.h>
30 #include <linux/mutex.h>
31 #include <linux/slab.h>
32 #include <linux/swap.h>
33 #include <linux/spinlock.h>
35 #include <linux/seq_file.h>
36 #include <linux/vmalloc.h>
37 #include <linux/mm_inline.h>
38 #include <linux/page_cgroup.h>
41 #include <asm/uaccess.h>
43 struct cgroup_subsys mem_cgroup_subsys __read_mostly;
44 #define MEM_CGROUP_RECLAIM_RETRIES 5
46 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
47 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 0 */
48 int do_swap_account __read_mostly;
49 static int really_do_swap_account __initdata = 1; /* for remember boot option*/
51 #define do_swap_account (0)
56 * Statistics for memory cgroup.
58 enum mem_cgroup_stat_index {
60 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
62 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
63 MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */
64 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
65 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
67 MEM_CGROUP_STAT_NSTATS,
70 struct mem_cgroup_stat_cpu {
71 s64 count[MEM_CGROUP_STAT_NSTATS];
72 } ____cacheline_aligned_in_smp;
74 struct mem_cgroup_stat {
75 struct mem_cgroup_stat_cpu cpustat[0];
79 * For accounting under irq disable, no need for increment preempt count.
81 static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat,
82 enum mem_cgroup_stat_index idx, int val)
84 stat->count[idx] += val;
87 static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat,
88 enum mem_cgroup_stat_index idx)
92 for_each_possible_cpu(cpu)
93 ret += stat->cpustat[cpu].count[idx];
98 * per-zone information in memory controller.
100 struct mem_cgroup_per_zone {
102 * spin_lock to protect the per cgroup LRU
104 struct list_head lists[NR_LRU_LISTS];
105 unsigned long count[NR_LRU_LISTS];
107 struct zone_reclaim_stat reclaim_stat;
109 /* Macro for accessing counter */
110 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
112 struct mem_cgroup_per_node {
113 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
116 struct mem_cgroup_lru_info {
117 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
121 * The memory controller data structure. The memory controller controls both
122 * page cache and RSS per cgroup. We would eventually like to provide
123 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
124 * to help the administrator determine what knobs to tune.
126 * TODO: Add a water mark for the memory controller. Reclaim will begin when
127 * we hit the water mark. May be even add a low water mark, such that
128 * no reclaim occurs from a cgroup at it's low water mark, this is
129 * a feature that will be implemented much later in the future.
132 struct cgroup_subsys_state css;
134 * the counter to account for memory usage
136 struct res_counter res;
138 * the counter to account for mem+swap usage.
140 struct res_counter memsw;
142 * Per cgroup active and inactive list, similar to the
143 * per zone LRU lists.
145 struct mem_cgroup_lru_info info;
148 protect against reclaim related member.
150 spinlock_t reclaim_param_lock;
152 int prev_priority; /* for recording reclaim priority */
155 * While reclaiming in a hiearchy, we cache the last child we
156 * reclaimed from. Protected by cgroup_lock()
158 struct mem_cgroup *last_scanned_child;
160 * Should the accounting and control be hierarchical, per subtree?
163 unsigned long last_oom_jiffies;
167 unsigned int inactive_ratio;
170 * statistics. This must be placed at the end of memcg.
172 struct mem_cgroup_stat stat;
176 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
177 MEM_CGROUP_CHARGE_TYPE_MAPPED,
178 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
179 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
180 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
184 /* only for here (for easy reading.) */
185 #define PCGF_CACHE (1UL << PCG_CACHE)
186 #define PCGF_USED (1UL << PCG_USED)
187 #define PCGF_LOCK (1UL << PCG_LOCK)
188 static const unsigned long
189 pcg_default_flags[NR_CHARGE_TYPE] = {
190 PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* File Cache */
191 PCGF_USED | PCGF_LOCK, /* Anon */
192 PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* Shmem */
196 /* for encoding cft->private value on file */
199 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
200 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
201 #define MEMFILE_ATTR(val) ((val) & 0xffff)
203 static void mem_cgroup_get(struct mem_cgroup *mem);
204 static void mem_cgroup_put(struct mem_cgroup *mem);
206 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
207 struct page_cgroup *pc,
210 int val = (charge)? 1 : -1;
211 struct mem_cgroup_stat *stat = &mem->stat;
212 struct mem_cgroup_stat_cpu *cpustat;
215 cpustat = &stat->cpustat[cpu];
216 if (PageCgroupCache(pc))
217 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val);
219 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val);
222 __mem_cgroup_stat_add_safe(cpustat,
223 MEM_CGROUP_STAT_PGPGIN_COUNT, 1);
225 __mem_cgroup_stat_add_safe(cpustat,
226 MEM_CGROUP_STAT_PGPGOUT_COUNT, 1);
230 static struct mem_cgroup_per_zone *
231 mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
233 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
236 static struct mem_cgroup_per_zone *
237 page_cgroup_zoneinfo(struct page_cgroup *pc)
239 struct mem_cgroup *mem = pc->mem_cgroup;
240 int nid = page_cgroup_nid(pc);
241 int zid = page_cgroup_zid(pc);
246 return mem_cgroup_zoneinfo(mem, nid, zid);
249 static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem,
253 struct mem_cgroup_per_zone *mz;
256 for_each_online_node(nid)
257 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
258 mz = mem_cgroup_zoneinfo(mem, nid, zid);
259 total += MEM_CGROUP_ZSTAT(mz, idx);
264 static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
266 return container_of(cgroup_subsys_state(cont,
267 mem_cgroup_subsys_id), struct mem_cgroup,
271 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
274 * mm_update_next_owner() may clear mm->owner to NULL
275 * if it races with swapoff, page migration, etc.
276 * So this can be called with p == NULL.
281 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
282 struct mem_cgroup, css);
286 * Following LRU functions are allowed to be used without PCG_LOCK.
287 * Operations are called by routine of global LRU independently from memcg.
288 * What we have to take care of here is validness of pc->mem_cgroup.
290 * Changes to pc->mem_cgroup happens when
293 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
294 * It is added to LRU before charge.
295 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
296 * When moving account, the page is not on LRU. It's isolated.
299 void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
301 struct page_cgroup *pc;
302 struct mem_cgroup *mem;
303 struct mem_cgroup_per_zone *mz;
305 if (mem_cgroup_disabled())
307 pc = lookup_page_cgroup(page);
308 /* can happen while we handle swapcache. */
309 if (list_empty(&pc->lru))
311 mz = page_cgroup_zoneinfo(pc);
312 mem = pc->mem_cgroup;
313 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
314 list_del_init(&pc->lru);
318 void mem_cgroup_del_lru(struct page *page)
320 mem_cgroup_del_lru_list(page, page_lru(page));
323 void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
325 struct mem_cgroup_per_zone *mz;
326 struct page_cgroup *pc;
328 if (mem_cgroup_disabled())
331 pc = lookup_page_cgroup(page);
333 /* unused page is not rotated. */
334 if (!PageCgroupUsed(pc))
336 mz = page_cgroup_zoneinfo(pc);
337 list_move(&pc->lru, &mz->lists[lru]);
340 void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
342 struct page_cgroup *pc;
343 struct mem_cgroup_per_zone *mz;
345 if (mem_cgroup_disabled())
347 pc = lookup_page_cgroup(page);
348 /* barrier to sync with "charge" */
350 if (!PageCgroupUsed(pc))
353 mz = page_cgroup_zoneinfo(pc);
354 MEM_CGROUP_ZSTAT(mz, lru) += 1;
355 list_add(&pc->lru, &mz->lists[lru]);
358 * To add swapcache into LRU. Be careful to all this function.
359 * zone->lru_lock shouldn't be held and irq must not be disabled.
361 static void mem_cgroup_lru_fixup(struct page *page)
363 if (!isolate_lru_page(page))
364 putback_lru_page(page);
367 void mem_cgroup_move_lists(struct page *page,
368 enum lru_list from, enum lru_list to)
370 if (mem_cgroup_disabled())
372 mem_cgroup_del_lru_list(page, from);
373 mem_cgroup_add_lru_list(page, to);
376 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
381 ret = task->mm && mm_match_cgroup(task->mm, mem);
387 * Calculate mapped_ratio under memory controller. This will be used in
388 * vmscan.c for deteremining we have to reclaim mapped pages.
390 int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem)
395 * usage is recorded in bytes. But, here, we assume the number of
396 * physical pages can be represented by "long" on any arch.
398 total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L;
399 rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
400 return (int)((rss * 100L) / total);
404 * prev_priority control...this will be used in memory reclaim path.
406 int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
410 spin_lock(&mem->reclaim_param_lock);
411 prev_priority = mem->prev_priority;
412 spin_unlock(&mem->reclaim_param_lock);
414 return prev_priority;
417 void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
419 spin_lock(&mem->reclaim_param_lock);
420 if (priority < mem->prev_priority)
421 mem->prev_priority = priority;
422 spin_unlock(&mem->reclaim_param_lock);
425 void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
427 spin_lock(&mem->reclaim_param_lock);
428 mem->prev_priority = priority;
429 spin_unlock(&mem->reclaim_param_lock);
432 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg, struct zone *zone)
434 unsigned long active;
435 unsigned long inactive;
437 inactive = mem_cgroup_get_all_zonestat(memcg, LRU_INACTIVE_ANON);
438 active = mem_cgroup_get_all_zonestat(memcg, LRU_ACTIVE_ANON);
440 if (inactive * memcg->inactive_ratio < active)
446 unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg,
450 int nid = zone->zone_pgdat->node_id;
451 int zid = zone_idx(zone);
452 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
454 return MEM_CGROUP_ZSTAT(mz, lru);
457 struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
460 int nid = zone->zone_pgdat->node_id;
461 int zid = zone_idx(zone);
462 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
464 return &mz->reclaim_stat;
467 struct zone_reclaim_stat *
468 mem_cgroup_get_reclaim_stat_from_page(struct page *page)
470 struct page_cgroup *pc;
471 struct mem_cgroup_per_zone *mz;
473 if (mem_cgroup_disabled())
476 pc = lookup_page_cgroup(page);
477 mz = page_cgroup_zoneinfo(pc);
481 return &mz->reclaim_stat;
484 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
485 struct list_head *dst,
486 unsigned long *scanned, int order,
487 int mode, struct zone *z,
488 struct mem_cgroup *mem_cont,
489 int active, int file)
491 unsigned long nr_taken = 0;
495 struct list_head *src;
496 struct page_cgroup *pc, *tmp;
497 int nid = z->zone_pgdat->node_id;
498 int zid = zone_idx(z);
499 struct mem_cgroup_per_zone *mz;
500 int lru = LRU_FILE * !!file + !!active;
503 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
504 src = &mz->lists[lru];
507 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
508 if (scan >= nr_to_scan)
512 if (unlikely(!PageCgroupUsed(pc)))
514 if (unlikely(!PageLRU(page)))
518 if (__isolate_lru_page(page, mode, file) == 0) {
519 list_move(&page->lru, dst);
528 #define mem_cgroup_from_res_counter(counter, member) \
529 container_of(counter, struct mem_cgroup, member)
532 * This routine finds the DFS walk successor. This routine should be
533 * called with cgroup_mutex held
535 static struct mem_cgroup *
536 mem_cgroup_get_next_node(struct mem_cgroup *curr, struct mem_cgroup *root_mem)
538 struct cgroup *cgroup, *curr_cgroup, *root_cgroup;
540 curr_cgroup = curr->css.cgroup;
541 root_cgroup = root_mem->css.cgroup;
543 if (!list_empty(&curr_cgroup->children)) {
545 * Walk down to children
547 mem_cgroup_put(curr);
548 cgroup = list_entry(curr_cgroup->children.next,
549 struct cgroup, sibling);
550 curr = mem_cgroup_from_cont(cgroup);
551 mem_cgroup_get(curr);
556 if (curr_cgroup == root_cgroup) {
557 mem_cgroup_put(curr);
559 mem_cgroup_get(curr);
566 if (curr_cgroup->sibling.next != &curr_cgroup->parent->children) {
567 mem_cgroup_put(curr);
568 cgroup = list_entry(curr_cgroup->sibling.next, struct cgroup,
570 curr = mem_cgroup_from_cont(cgroup);
571 mem_cgroup_get(curr);
576 * Go up to next parent and next parent's sibling if need be
578 curr_cgroup = curr_cgroup->parent;
582 root_mem->last_scanned_child = curr;
587 * Visit the first child (need not be the first child as per the ordering
588 * of the cgroup list, since we track last_scanned_child) of @mem and use
589 * that to reclaim free pages from.
591 static struct mem_cgroup *
592 mem_cgroup_get_first_node(struct mem_cgroup *root_mem)
594 struct cgroup *cgroup;
595 struct mem_cgroup *ret;
596 bool obsolete = (root_mem->last_scanned_child &&
597 root_mem->last_scanned_child->obsolete);
600 * Scan all children under the mem_cgroup mem
603 if (list_empty(&root_mem->css.cgroup->children)) {
608 if (!root_mem->last_scanned_child || obsolete) {
611 mem_cgroup_put(root_mem->last_scanned_child);
613 cgroup = list_first_entry(&root_mem->css.cgroup->children,
614 struct cgroup, sibling);
615 ret = mem_cgroup_from_cont(cgroup);
618 ret = mem_cgroup_get_next_node(root_mem->last_scanned_child,
622 root_mem->last_scanned_child = ret;
627 static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem)
629 if (do_swap_account) {
630 if (res_counter_check_under_limit(&mem->res) &&
631 res_counter_check_under_limit(&mem->memsw))
634 if (res_counter_check_under_limit(&mem->res))
640 * Dance down the hierarchy if needed to reclaim memory. We remember the
641 * last child we reclaimed from, so that we don't end up penalizing
642 * one child extensively based on its position in the children list.
644 * root_mem is the original ancestor that we've been reclaim from.
646 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
647 gfp_t gfp_mask, bool noswap)
649 struct mem_cgroup *next_mem;
653 * Reclaim unconditionally and don't check for return value.
654 * We need to reclaim in the current group and down the tree.
655 * One might think about checking for children before reclaiming,
656 * but there might be left over accounting, even after children
659 ret = try_to_free_mem_cgroup_pages(root_mem, gfp_mask, noswap);
660 if (mem_cgroup_check_under_limit(root_mem))
662 if (!root_mem->use_hierarchy)
665 next_mem = mem_cgroup_get_first_node(root_mem);
667 while (next_mem != root_mem) {
668 if (next_mem->obsolete) {
669 mem_cgroup_put(next_mem);
671 next_mem = mem_cgroup_get_first_node(root_mem);
675 ret = try_to_free_mem_cgroup_pages(next_mem, gfp_mask, noswap);
676 if (mem_cgroup_check_under_limit(root_mem))
679 next_mem = mem_cgroup_get_next_node(next_mem, root_mem);
685 bool mem_cgroup_oom_called(struct task_struct *task)
688 struct mem_cgroup *mem;
689 struct mm_struct *mm;
695 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
696 if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10))
702 * Unlike exported interface, "oom" parameter is added. if oom==true,
703 * oom-killer can be invoked.
705 static int __mem_cgroup_try_charge(struct mm_struct *mm,
706 gfp_t gfp_mask, struct mem_cgroup **memcg,
709 struct mem_cgroup *mem, *mem_over_limit;
710 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
711 struct res_counter *fail_res;
713 if (unlikely(test_thread_flag(TIF_MEMDIE))) {
714 /* Don't account this! */
720 * We always charge the cgroup the mm_struct belongs to.
721 * The mm_struct's mem_cgroup changes on task migration if the
722 * thread group leader migrates. It's possible that mm is not
723 * set, if so charge the init_mm (happens for pagecache usage).
725 if (likely(!*memcg)) {
727 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
728 if (unlikely(!mem)) {
733 * For every charge from the cgroup, increment reference count
747 ret = res_counter_charge(&mem->res, PAGE_SIZE, &fail_res);
749 if (!do_swap_account)
751 ret = res_counter_charge(&mem->memsw, PAGE_SIZE,
755 /* mem+swap counter fails */
756 res_counter_uncharge(&mem->res, PAGE_SIZE);
758 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
761 /* mem counter fails */
762 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
765 if (!(gfp_mask & __GFP_WAIT))
768 ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, gfp_mask,
772 * try_to_free_mem_cgroup_pages() might not give us a full
773 * picture of reclaim. Some pages are reclaimed and might be
774 * moved to swap cache or just unmapped from the cgroup.
775 * Check the limit again to see if the reclaim reduced the
776 * current usage of the cgroup before giving up
779 if (mem_cgroup_check_under_limit(mem_over_limit))
784 mem_cgroup_out_of_memory(mem_over_limit, gfp_mask);
785 mem_over_limit->last_oom_jiffies = jiffies;
797 * mem_cgroup_try_charge - get charge of PAGE_SIZE.
798 * @mm: an mm_struct which is charged against. (when *memcg is NULL)
799 * @gfp_mask: gfp_mask for reclaim.
800 * @memcg: a pointer to memory cgroup which is charged against.
802 * charge against memory cgroup pointed by *memcg. if *memcg == NULL, estimated
803 * memory cgroup from @mm is got and stored in *memcg.
805 * Returns 0 if success. -ENOMEM at failure.
806 * This call can invoke OOM-Killer.
809 int mem_cgroup_try_charge(struct mm_struct *mm,
810 gfp_t mask, struct mem_cgroup **memcg)
812 return __mem_cgroup_try_charge(mm, mask, memcg, true);
816 * commit a charge got by mem_cgroup_try_charge() and makes page_cgroup to be
817 * USED state. If already USED, uncharge and return.
820 static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
821 struct page_cgroup *pc,
822 enum charge_type ctype)
824 /* try_charge() can return NULL to *memcg, taking care of it. */
828 lock_page_cgroup(pc);
829 if (unlikely(PageCgroupUsed(pc))) {
830 unlock_page_cgroup(pc);
831 res_counter_uncharge(&mem->res, PAGE_SIZE);
833 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
837 pc->mem_cgroup = mem;
839 pc->flags = pcg_default_flags[ctype];
841 mem_cgroup_charge_statistics(mem, pc, true);
843 unlock_page_cgroup(pc);
847 * mem_cgroup_move_account - move account of the page
848 * @pc: page_cgroup of the page.
849 * @from: mem_cgroup which the page is moved from.
850 * @to: mem_cgroup which the page is moved to. @from != @to.
852 * The caller must confirm following.
853 * - page is not on LRU (isolate_page() is useful.)
855 * returns 0 at success,
856 * returns -EBUSY when lock is busy or "pc" is unstable.
858 * This function does "uncharge" from old cgroup but doesn't do "charge" to
859 * new cgroup. It should be done by a caller.
862 static int mem_cgroup_move_account(struct page_cgroup *pc,
863 struct mem_cgroup *from, struct mem_cgroup *to)
865 struct mem_cgroup_per_zone *from_mz, *to_mz;
869 VM_BUG_ON(from == to);
870 VM_BUG_ON(PageLRU(pc->page));
872 nid = page_cgroup_nid(pc);
873 zid = page_cgroup_zid(pc);
874 from_mz = mem_cgroup_zoneinfo(from, nid, zid);
875 to_mz = mem_cgroup_zoneinfo(to, nid, zid);
877 if (!trylock_page_cgroup(pc))
880 if (!PageCgroupUsed(pc))
883 if (pc->mem_cgroup != from)
887 res_counter_uncharge(&from->res, PAGE_SIZE);
888 mem_cgroup_charge_statistics(from, pc, false);
890 res_counter_uncharge(&from->memsw, PAGE_SIZE);
892 mem_cgroup_charge_statistics(to, pc, true);
896 unlock_page_cgroup(pc);
901 * move charges to its parent.
904 static int mem_cgroup_move_parent(struct page_cgroup *pc,
905 struct mem_cgroup *child,
908 struct page *page = pc->page;
909 struct cgroup *cg = child->css.cgroup;
910 struct cgroup *pcg = cg->parent;
911 struct mem_cgroup *parent;
919 parent = mem_cgroup_from_cont(pcg);
922 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false);
926 if (!get_page_unless_zero(page))
929 ret = isolate_lru_page(page);
934 ret = mem_cgroup_move_account(pc, child, parent);
936 /* drop extra refcnt by try_charge() (move_account increment one) */
937 css_put(&parent->css);
938 putback_lru_page(page);
943 /* uncharge if move fails */
945 res_counter_uncharge(&parent->res, PAGE_SIZE);
947 res_counter_uncharge(&parent->memsw, PAGE_SIZE);
953 * Charge the memory controller for page usage.
955 * 0 if the charge was successful
956 * < 0 if the cgroup is over its limit
958 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
959 gfp_t gfp_mask, enum charge_type ctype,
960 struct mem_cgroup *memcg)
962 struct mem_cgroup *mem;
963 struct page_cgroup *pc;
966 pc = lookup_page_cgroup(page);
967 /* can happen at boot */
973 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true);
977 __mem_cgroup_commit_charge(mem, pc, ctype);
981 int mem_cgroup_newpage_charge(struct page *page,
982 struct mm_struct *mm, gfp_t gfp_mask)
984 if (mem_cgroup_disabled())
986 if (PageCompound(page))
989 * If already mapped, we don't have to account.
990 * If page cache, page->mapping has address_space.
991 * But page->mapping may have out-of-use anon_vma pointer,
992 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
995 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
999 return mem_cgroup_charge_common(page, mm, gfp_mask,
1000 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
1003 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
1006 if (mem_cgroup_disabled())
1008 if (PageCompound(page))
1011 * Corner case handling. This is called from add_to_page_cache()
1012 * in usual. But some FS (shmem) precharges this page before calling it
1013 * and call add_to_page_cache() with GFP_NOWAIT.
1015 * For GFP_NOWAIT case, the page may be pre-charged before calling
1016 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
1017 * charge twice. (It works but has to pay a bit larger cost.)
1019 if (!(gfp_mask & __GFP_WAIT)) {
1020 struct page_cgroup *pc;
1023 pc = lookup_page_cgroup(page);
1026 lock_page_cgroup(pc);
1027 if (PageCgroupUsed(pc)) {
1028 unlock_page_cgroup(pc);
1031 unlock_page_cgroup(pc);
1037 if (page_is_file_cache(page))
1038 return mem_cgroup_charge_common(page, mm, gfp_mask,
1039 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
1041 return mem_cgroup_charge_common(page, mm, gfp_mask,
1042 MEM_CGROUP_CHARGE_TYPE_SHMEM, NULL);
1045 int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
1047 gfp_t mask, struct mem_cgroup **ptr)
1049 struct mem_cgroup *mem;
1052 if (mem_cgroup_disabled())
1055 if (!do_swap_account)
1059 * A racing thread's fault, or swapoff, may have already updated
1060 * the pte, and even removed page from swap cache: return success
1061 * to go on to do_swap_page()'s pte_same() test, which should fail.
1063 if (!PageSwapCache(page))
1066 ent.val = page_private(page);
1068 mem = lookup_swap_cgroup(ent);
1069 if (!mem || mem->obsolete)
1072 return __mem_cgroup_try_charge(NULL, mask, ptr, true);
1076 return __mem_cgroup_try_charge(mm, mask, ptr, true);
1081 int mem_cgroup_cache_charge_swapin(struct page *page,
1082 struct mm_struct *mm, gfp_t mask, bool locked)
1086 if (mem_cgroup_disabled())
1093 * If not locked, the page can be dropped from SwapCache until
1096 if (PageSwapCache(page)) {
1097 struct mem_cgroup *mem = NULL;
1100 ent.val = page_private(page);
1101 if (do_swap_account) {
1102 mem = lookup_swap_cgroup(ent);
1103 if (mem && mem->obsolete)
1108 ret = mem_cgroup_charge_common(page, mm, mask,
1109 MEM_CGROUP_CHARGE_TYPE_SHMEM, mem);
1111 if (!ret && do_swap_account) {
1112 /* avoid double counting */
1113 mem = swap_cgroup_record(ent, NULL);
1115 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1116 mem_cgroup_put(mem);
1122 /* add this page(page_cgroup) to the LRU we want. */
1123 mem_cgroup_lru_fixup(page);
1129 void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
1131 struct page_cgroup *pc;
1133 if (mem_cgroup_disabled())
1137 pc = lookup_page_cgroup(page);
1138 __mem_cgroup_commit_charge(ptr, pc, MEM_CGROUP_CHARGE_TYPE_MAPPED);
1140 * Now swap is on-memory. This means this page may be
1141 * counted both as mem and swap....double count.
1142 * Fix it by uncharging from memsw. This SwapCache is stable
1143 * because we're still under lock_page().
1145 if (do_swap_account) {
1146 swp_entry_t ent = {.val = page_private(page)};
1147 struct mem_cgroup *memcg;
1148 memcg = swap_cgroup_record(ent, NULL);
1150 /* If memcg is obsolete, memcg can be != ptr */
1151 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1152 mem_cgroup_put(memcg);
1156 /* add this page(page_cgroup) to the LRU we want. */
1157 mem_cgroup_lru_fixup(page);
1160 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
1162 if (mem_cgroup_disabled())
1166 res_counter_uncharge(&mem->res, PAGE_SIZE);
1167 if (do_swap_account)
1168 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1174 * uncharge if !page_mapped(page)
1176 static struct mem_cgroup *
1177 __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
1179 struct page_cgroup *pc;
1180 struct mem_cgroup *mem = NULL;
1181 struct mem_cgroup_per_zone *mz;
1183 if (mem_cgroup_disabled())
1186 if (PageSwapCache(page))
1190 * Check if our page_cgroup is valid
1192 pc = lookup_page_cgroup(page);
1193 if (unlikely(!pc || !PageCgroupUsed(pc)))
1196 lock_page_cgroup(pc);
1198 mem = pc->mem_cgroup;
1200 if (!PageCgroupUsed(pc))
1204 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
1205 if (page_mapped(page))
1208 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
1209 if (!PageAnon(page)) { /* Shared memory */
1210 if (page->mapping && !page_is_file_cache(page))
1212 } else if (page_mapped(page)) /* Anon */
1219 res_counter_uncharge(&mem->res, PAGE_SIZE);
1220 if (do_swap_account && (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT))
1221 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1223 mem_cgroup_charge_statistics(mem, pc, false);
1224 ClearPageCgroupUsed(pc);
1226 mz = page_cgroup_zoneinfo(pc);
1227 unlock_page_cgroup(pc);
1229 /* at swapout, this memcg will be accessed to record to swap */
1230 if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
1236 unlock_page_cgroup(pc);
1240 void mem_cgroup_uncharge_page(struct page *page)
1243 if (page_mapped(page))
1245 if (page->mapping && !PageAnon(page))
1247 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
1250 void mem_cgroup_uncharge_cache_page(struct page *page)
1252 VM_BUG_ON(page_mapped(page));
1253 VM_BUG_ON(page->mapping);
1254 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
1258 * called from __delete_from_swap_cache() and drop "page" account.
1259 * memcg information is recorded to swap_cgroup of "ent"
1261 void mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent)
1263 struct mem_cgroup *memcg;
1265 memcg = __mem_cgroup_uncharge_common(page,
1266 MEM_CGROUP_CHARGE_TYPE_SWAPOUT);
1267 /* record memcg information */
1268 if (do_swap_account && memcg) {
1269 swap_cgroup_record(ent, memcg);
1270 mem_cgroup_get(memcg);
1273 css_put(&memcg->css);
1276 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1278 * called from swap_entry_free(). remove record in swap_cgroup and
1279 * uncharge "memsw" account.
1281 void mem_cgroup_uncharge_swap(swp_entry_t ent)
1283 struct mem_cgroup *memcg;
1285 if (!do_swap_account)
1288 memcg = swap_cgroup_record(ent, NULL);
1290 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1291 mem_cgroup_put(memcg);
1297 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
1300 int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr)
1302 struct page_cgroup *pc;
1303 struct mem_cgroup *mem = NULL;
1306 if (mem_cgroup_disabled())
1309 pc = lookup_page_cgroup(page);
1310 lock_page_cgroup(pc);
1311 if (PageCgroupUsed(pc)) {
1312 mem = pc->mem_cgroup;
1315 unlock_page_cgroup(pc);
1318 ret = mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem);
1325 /* remove redundant charge if migration failed*/
1326 void mem_cgroup_end_migration(struct mem_cgroup *mem,
1327 struct page *oldpage, struct page *newpage)
1329 struct page *target, *unused;
1330 struct page_cgroup *pc;
1331 enum charge_type ctype;
1336 /* at migration success, oldpage->mapping is NULL. */
1337 if (oldpage->mapping) {
1345 if (PageAnon(target))
1346 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
1347 else if (page_is_file_cache(target))
1348 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
1350 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
1352 /* unused page is not on radix-tree now. */
1354 __mem_cgroup_uncharge_common(unused, ctype);
1356 pc = lookup_page_cgroup(target);
1358 * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup.
1359 * So, double-counting is effectively avoided.
1361 __mem_cgroup_commit_charge(mem, pc, ctype);
1364 * Both of oldpage and newpage are still under lock_page().
1365 * Then, we don't have to care about race in radix-tree.
1366 * But we have to be careful that this page is unmapped or not.
1368 * There is a case for !page_mapped(). At the start of
1369 * migration, oldpage was mapped. But now, it's zapped.
1370 * But we know *target* page is not freed/reused under us.
1371 * mem_cgroup_uncharge_page() does all necessary checks.
1373 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
1374 mem_cgroup_uncharge_page(target);
1378 * A call to try to shrink memory usage under specified resource controller.
1379 * This is typically used for page reclaiming for shmem for reducing side
1380 * effect of page allocation from shmem, which is used by some mem_cgroup.
1382 int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask)
1384 struct mem_cgroup *mem;
1386 int retry = MEM_CGROUP_RECLAIM_RETRIES;
1388 if (mem_cgroup_disabled())
1394 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
1395 if (unlikely(!mem)) {
1403 progress = try_to_free_mem_cgroup_pages(mem, gfp_mask, true);
1404 progress += mem_cgroup_check_under_limit(mem);
1405 } while (!progress && --retry);
1414 * The inactive anon list should be small enough that the VM never has to
1415 * do too much work, but large enough that each inactive page has a chance
1416 * to be referenced again before it is swapped out.
1418 * this calculation is straightforward porting from
1419 * page_alloc.c::setup_per_zone_inactive_ratio().
1420 * it describe more detail.
1422 static void mem_cgroup_set_inactive_ratio(struct mem_cgroup *memcg)
1424 unsigned int gb, ratio;
1426 gb = res_counter_read_u64(&memcg->res, RES_LIMIT) >> 30;
1428 ratio = int_sqrt(10 * gb);
1432 memcg->inactive_ratio = ratio;
1436 static DEFINE_MUTEX(set_limit_mutex);
1438 static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
1439 unsigned long long val)
1442 int retry_count = MEM_CGROUP_RECLAIM_RETRIES;
1447 while (retry_count) {
1448 if (signal_pending(current)) {
1453 * Rather than hide all in some function, I do this in
1454 * open coded manner. You see what this really does.
1455 * We have to guarantee mem->res.limit < mem->memsw.limit.
1457 mutex_lock(&set_limit_mutex);
1458 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1459 if (memswlimit < val) {
1461 mutex_unlock(&set_limit_mutex);
1464 ret = res_counter_set_limit(&memcg->res, val);
1465 mutex_unlock(&set_limit_mutex);
1470 progress = try_to_free_mem_cgroup_pages(memcg,
1472 if (!progress) retry_count--;
1476 mem_cgroup_set_inactive_ratio(memcg);
1481 int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
1482 unsigned long long val)
1484 int retry_count = MEM_CGROUP_RECLAIM_RETRIES;
1485 u64 memlimit, oldusage, curusage;
1488 if (!do_swap_account)
1491 while (retry_count) {
1492 if (signal_pending(current)) {
1497 * Rather than hide all in some function, I do this in
1498 * open coded manner. You see what this really does.
1499 * We have to guarantee mem->res.limit < mem->memsw.limit.
1501 mutex_lock(&set_limit_mutex);
1502 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1503 if (memlimit > val) {
1505 mutex_unlock(&set_limit_mutex);
1508 ret = res_counter_set_limit(&memcg->memsw, val);
1509 mutex_unlock(&set_limit_mutex);
1514 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
1515 try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, true);
1516 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
1517 if (curusage >= oldusage)
1524 * This routine traverse page_cgroup in given list and drop them all.
1525 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
1527 static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
1528 int node, int zid, enum lru_list lru)
1531 struct mem_cgroup_per_zone *mz;
1532 struct page_cgroup *pc, *busy;
1533 unsigned long flags, loop;
1534 struct list_head *list;
1537 zone = &NODE_DATA(node)->node_zones[zid];
1538 mz = mem_cgroup_zoneinfo(mem, node, zid);
1539 list = &mz->lists[lru];
1541 loop = MEM_CGROUP_ZSTAT(mz, lru);
1542 /* give some margin against EBUSY etc...*/
1547 spin_lock_irqsave(&zone->lru_lock, flags);
1548 if (list_empty(list)) {
1549 spin_unlock_irqrestore(&zone->lru_lock, flags);
1552 pc = list_entry(list->prev, struct page_cgroup, lru);
1554 list_move(&pc->lru, list);
1556 spin_unlock_irqrestore(&zone->lru_lock, flags);
1559 spin_unlock_irqrestore(&zone->lru_lock, flags);
1561 ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL);
1565 if (ret == -EBUSY || ret == -EINVAL) {
1566 /* found lock contention or "pc" is obsolete. */
1573 if (!ret && !list_empty(list))
1579 * make mem_cgroup's charge to be 0 if there is no task.
1580 * This enables deleting this mem_cgroup.
1582 static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
1585 int node, zid, shrink;
1586 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
1587 struct cgroup *cgrp = mem->css.cgroup;
1592 /* should free all ? */
1596 while (mem->res.usage > 0) {
1598 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
1601 if (signal_pending(current))
1603 /* This is for making all *used* pages to be on LRU. */
1604 lru_add_drain_all();
1606 for_each_node_state(node, N_POSSIBLE) {
1607 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
1610 ret = mem_cgroup_force_empty_list(mem,
1619 /* it seems parent cgroup doesn't have enough mem */
1630 /* returns EBUSY if there is a task or if we come here twice. */
1631 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
1635 /* we call try-to-free pages for make this cgroup empty */
1636 lru_add_drain_all();
1637 /* try to free all pages in this cgroup */
1639 while (nr_retries && mem->res.usage > 0) {
1642 if (signal_pending(current)) {
1646 progress = try_to_free_mem_cgroup_pages(mem,
1650 /* maybe some writeback is necessary */
1651 congestion_wait(WRITE, HZ/10);
1656 /* try move_account...there may be some *locked* pages. */
1663 int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
1665 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
1669 static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
1671 return mem_cgroup_from_cont(cont)->use_hierarchy;
1674 static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
1678 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1679 struct cgroup *parent = cont->parent;
1680 struct mem_cgroup *parent_mem = NULL;
1683 parent_mem = mem_cgroup_from_cont(parent);
1687 * If parent's use_hiearchy is set, we can't make any modifications
1688 * in the child subtrees. If it is unset, then the change can
1689 * occur, provided the current cgroup has no children.
1691 * For the root cgroup, parent_mem is NULL, we allow value to be
1692 * set if there are no children.
1694 if ((!parent_mem || !parent_mem->use_hierarchy) &&
1695 (val == 1 || val == 0)) {
1696 if (list_empty(&cont->children))
1697 mem->use_hierarchy = val;
1707 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
1709 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1713 type = MEMFILE_TYPE(cft->private);
1714 name = MEMFILE_ATTR(cft->private);
1717 val = res_counter_read_u64(&mem->res, name);
1720 if (do_swap_account)
1721 val = res_counter_read_u64(&mem->memsw, name);
1730 * The user of this function is...
1733 static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
1736 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
1738 unsigned long long val;
1741 type = MEMFILE_TYPE(cft->private);
1742 name = MEMFILE_ATTR(cft->private);
1745 /* This function does all necessary parse...reuse it */
1746 ret = res_counter_memparse_write_strategy(buffer, &val);
1750 ret = mem_cgroup_resize_limit(memcg, val);
1752 ret = mem_cgroup_resize_memsw_limit(memcg, val);
1755 ret = -EINVAL; /* should be BUG() ? */
1761 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
1763 struct mem_cgroup *mem;
1766 mem = mem_cgroup_from_cont(cont);
1767 type = MEMFILE_TYPE(event);
1768 name = MEMFILE_ATTR(event);
1772 res_counter_reset_max(&mem->res);
1774 res_counter_reset_max(&mem->memsw);
1778 res_counter_reset_failcnt(&mem->res);
1780 res_counter_reset_failcnt(&mem->memsw);
1786 static const struct mem_cgroup_stat_desc {
1789 } mem_cgroup_stat_desc[] = {
1790 [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, },
1791 [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, },
1792 [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, },
1793 [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, },
1796 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
1797 struct cgroup_map_cb *cb)
1799 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
1800 struct mem_cgroup_stat *stat = &mem_cont->stat;
1803 for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) {
1806 val = mem_cgroup_read_stat(stat, i);
1807 val *= mem_cgroup_stat_desc[i].unit;
1808 cb->fill(cb, mem_cgroup_stat_desc[i].msg, val);
1810 /* showing # of active pages */
1812 unsigned long active_anon, inactive_anon;
1813 unsigned long active_file, inactive_file;
1814 unsigned long unevictable;
1816 inactive_anon = mem_cgroup_get_all_zonestat(mem_cont,
1818 active_anon = mem_cgroup_get_all_zonestat(mem_cont,
1820 inactive_file = mem_cgroup_get_all_zonestat(mem_cont,
1822 active_file = mem_cgroup_get_all_zonestat(mem_cont,
1824 unevictable = mem_cgroup_get_all_zonestat(mem_cont,
1827 cb->fill(cb, "active_anon", (active_anon) * PAGE_SIZE);
1828 cb->fill(cb, "inactive_anon", (inactive_anon) * PAGE_SIZE);
1829 cb->fill(cb, "active_file", (active_file) * PAGE_SIZE);
1830 cb->fill(cb, "inactive_file", (inactive_file) * PAGE_SIZE);
1831 cb->fill(cb, "unevictable", unevictable * PAGE_SIZE);
1835 #ifdef CONFIG_DEBUG_VM
1836 cb->fill(cb, "inactive_ratio", mem_cont->inactive_ratio);
1840 struct mem_cgroup_per_zone *mz;
1841 unsigned long recent_rotated[2] = {0, 0};
1842 unsigned long recent_scanned[2] = {0, 0};
1844 for_each_online_node(nid)
1845 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1846 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
1848 recent_rotated[0] +=
1849 mz->reclaim_stat.recent_rotated[0];
1850 recent_rotated[1] +=
1851 mz->reclaim_stat.recent_rotated[1];
1852 recent_scanned[0] +=
1853 mz->reclaim_stat.recent_scanned[0];
1854 recent_scanned[1] +=
1855 mz->reclaim_stat.recent_scanned[1];
1857 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
1858 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
1859 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
1860 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
1868 static struct cftype mem_cgroup_files[] = {
1870 .name = "usage_in_bytes",
1871 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
1872 .read_u64 = mem_cgroup_read,
1875 .name = "max_usage_in_bytes",
1876 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
1877 .trigger = mem_cgroup_reset,
1878 .read_u64 = mem_cgroup_read,
1881 .name = "limit_in_bytes",
1882 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
1883 .write_string = mem_cgroup_write,
1884 .read_u64 = mem_cgroup_read,
1888 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
1889 .trigger = mem_cgroup_reset,
1890 .read_u64 = mem_cgroup_read,
1894 .read_map = mem_control_stat_show,
1897 .name = "force_empty",
1898 .trigger = mem_cgroup_force_empty_write,
1901 .name = "use_hierarchy",
1902 .write_u64 = mem_cgroup_hierarchy_write,
1903 .read_u64 = mem_cgroup_hierarchy_read,
1907 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1908 static struct cftype memsw_cgroup_files[] = {
1910 .name = "memsw.usage_in_bytes",
1911 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
1912 .read_u64 = mem_cgroup_read,
1915 .name = "memsw.max_usage_in_bytes",
1916 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
1917 .trigger = mem_cgroup_reset,
1918 .read_u64 = mem_cgroup_read,
1921 .name = "memsw.limit_in_bytes",
1922 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
1923 .write_string = mem_cgroup_write,
1924 .read_u64 = mem_cgroup_read,
1927 .name = "memsw.failcnt",
1928 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
1929 .trigger = mem_cgroup_reset,
1930 .read_u64 = mem_cgroup_read,
1934 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
1936 if (!do_swap_account)
1938 return cgroup_add_files(cont, ss, memsw_cgroup_files,
1939 ARRAY_SIZE(memsw_cgroup_files));
1942 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
1948 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
1950 struct mem_cgroup_per_node *pn;
1951 struct mem_cgroup_per_zone *mz;
1953 int zone, tmp = node;
1955 * This routine is called against possible nodes.
1956 * But it's BUG to call kmalloc() against offline node.
1958 * TODO: this routine can waste much memory for nodes which will
1959 * never be onlined. It's better to use memory hotplug callback
1962 if (!node_state(node, N_NORMAL_MEMORY))
1964 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
1968 mem->info.nodeinfo[node] = pn;
1969 memset(pn, 0, sizeof(*pn));
1971 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
1972 mz = &pn->zoneinfo[zone];
1974 INIT_LIST_HEAD(&mz->lists[l]);
1979 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
1981 kfree(mem->info.nodeinfo[node]);
1984 static int mem_cgroup_size(void)
1986 int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu);
1987 return sizeof(struct mem_cgroup) + cpustat_size;
1990 static struct mem_cgroup *mem_cgroup_alloc(void)
1992 struct mem_cgroup *mem;
1993 int size = mem_cgroup_size();
1995 if (size < PAGE_SIZE)
1996 mem = kmalloc(size, GFP_KERNEL);
1998 mem = vmalloc(size);
2001 memset(mem, 0, size);
2006 * At destroying mem_cgroup, references from swap_cgroup can remain.
2007 * (scanning all at force_empty is too costly...)
2009 * Instead of clearing all references at force_empty, we remember
2010 * the number of reference from swap_cgroup and free mem_cgroup when
2011 * it goes down to 0.
2013 * When mem_cgroup is destroyed, mem->obsolete will be set to 0 and
2014 * entry which points to this memcg will be ignore at swapin.
2016 * Removal of cgroup itself succeeds regardless of refs from swap.
2019 static void mem_cgroup_free(struct mem_cgroup *mem)
2023 if (atomic_read(&mem->refcnt) > 0)
2027 for_each_node_state(node, N_POSSIBLE)
2028 free_mem_cgroup_per_zone_info(mem, node);
2030 if (mem_cgroup_size() < PAGE_SIZE)
2036 static void mem_cgroup_get(struct mem_cgroup *mem)
2038 atomic_inc(&mem->refcnt);
2041 static void mem_cgroup_put(struct mem_cgroup *mem)
2043 if (atomic_dec_and_test(&mem->refcnt)) {
2046 mem_cgroup_free(mem);
2051 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2052 static void __init enable_swap_cgroup(void)
2054 if (!mem_cgroup_disabled() && really_do_swap_account)
2055 do_swap_account = 1;
2058 static void __init enable_swap_cgroup(void)
2063 static struct cgroup_subsys_state *
2064 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
2066 struct mem_cgroup *mem, *parent;
2069 mem = mem_cgroup_alloc();
2071 return ERR_PTR(-ENOMEM);
2073 for_each_node_state(node, N_POSSIBLE)
2074 if (alloc_mem_cgroup_per_zone_info(mem, node))
2077 if (cont->parent == NULL) {
2078 enable_swap_cgroup();
2081 parent = mem_cgroup_from_cont(cont->parent);
2082 mem->use_hierarchy = parent->use_hierarchy;
2085 if (parent && parent->use_hierarchy) {
2086 res_counter_init(&mem->res, &parent->res);
2087 res_counter_init(&mem->memsw, &parent->memsw);
2089 res_counter_init(&mem->res, NULL);
2090 res_counter_init(&mem->memsw, NULL);
2092 mem_cgroup_set_inactive_ratio(mem);
2093 mem->last_scanned_child = NULL;
2094 spin_lock_init(&mem->reclaim_param_lock);
2098 for_each_node_state(node, N_POSSIBLE)
2099 free_mem_cgroup_per_zone_info(mem, node);
2100 mem_cgroup_free(mem);
2101 return ERR_PTR(-ENOMEM);
2104 static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
2105 struct cgroup *cont)
2107 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2109 mem_cgroup_force_empty(mem, false);
2112 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
2113 struct cgroup *cont)
2115 mem_cgroup_free(mem_cgroup_from_cont(cont));
2118 static int mem_cgroup_populate(struct cgroup_subsys *ss,
2119 struct cgroup *cont)
2123 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
2124 ARRAY_SIZE(mem_cgroup_files));
2127 ret = register_memsw_files(cont, ss);
2131 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
2132 struct cgroup *cont,
2133 struct cgroup *old_cont,
2134 struct task_struct *p)
2137 * FIXME: It's better to move charges of this process from old
2138 * memcg to new memcg. But it's just on TODO-List now.
2142 struct cgroup_subsys mem_cgroup_subsys = {
2144 .subsys_id = mem_cgroup_subsys_id,
2145 .create = mem_cgroup_create,
2146 .pre_destroy = mem_cgroup_pre_destroy,
2147 .destroy = mem_cgroup_destroy,
2148 .populate = mem_cgroup_populate,
2149 .attach = mem_cgroup_move_task,
2153 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2155 static int __init disable_swap_account(char *s)
2157 really_do_swap_account = 0;
2160 __setup("noswapaccount", disable_swap_account);