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;
147 int prev_priority; /* for recording reclaim priority */
150 * While reclaiming in a hiearchy, we cache the last child we
151 * reclaimed from. Protected by cgroup_lock()
153 struct mem_cgroup *last_scanned_child;
155 * Should the accounting and control be hierarchical, per subtree?
158 unsigned long last_oom_jiffies;
162 unsigned int inactive_ratio;
165 * statistics. This must be placed at the end of memcg.
167 struct mem_cgroup_stat stat;
171 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
172 MEM_CGROUP_CHARGE_TYPE_MAPPED,
173 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
174 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
175 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
179 /* only for here (for easy reading.) */
180 #define PCGF_CACHE (1UL << PCG_CACHE)
181 #define PCGF_USED (1UL << PCG_USED)
182 #define PCGF_LOCK (1UL << PCG_LOCK)
183 static const unsigned long
184 pcg_default_flags[NR_CHARGE_TYPE] = {
185 PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* File Cache */
186 PCGF_USED | PCGF_LOCK, /* Anon */
187 PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* Shmem */
191 /* for encoding cft->private value on file */
194 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
195 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
196 #define MEMFILE_ATTR(val) ((val) & 0xffff)
198 static void mem_cgroup_get(struct mem_cgroup *mem);
199 static void mem_cgroup_put(struct mem_cgroup *mem);
201 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
202 struct page_cgroup *pc,
205 int val = (charge)? 1 : -1;
206 struct mem_cgroup_stat *stat = &mem->stat;
207 struct mem_cgroup_stat_cpu *cpustat;
210 cpustat = &stat->cpustat[cpu];
211 if (PageCgroupCache(pc))
212 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val);
214 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val);
217 __mem_cgroup_stat_add_safe(cpustat,
218 MEM_CGROUP_STAT_PGPGIN_COUNT, 1);
220 __mem_cgroup_stat_add_safe(cpustat,
221 MEM_CGROUP_STAT_PGPGOUT_COUNT, 1);
225 static struct mem_cgroup_per_zone *
226 mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
228 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
231 static struct mem_cgroup_per_zone *
232 page_cgroup_zoneinfo(struct page_cgroup *pc)
234 struct mem_cgroup *mem = pc->mem_cgroup;
235 int nid = page_cgroup_nid(pc);
236 int zid = page_cgroup_zid(pc);
241 return mem_cgroup_zoneinfo(mem, nid, zid);
244 static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem,
248 struct mem_cgroup_per_zone *mz;
251 for_each_online_node(nid)
252 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
253 mz = mem_cgroup_zoneinfo(mem, nid, zid);
254 total += MEM_CGROUP_ZSTAT(mz, idx);
259 static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
261 return container_of(cgroup_subsys_state(cont,
262 mem_cgroup_subsys_id), struct mem_cgroup,
266 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
269 * mm_update_next_owner() may clear mm->owner to NULL
270 * if it races with swapoff, page migration, etc.
271 * So this can be called with p == NULL.
276 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
277 struct mem_cgroup, css);
281 * Following LRU functions are allowed to be used without PCG_LOCK.
282 * Operations are called by routine of global LRU independently from memcg.
283 * What we have to take care of here is validness of pc->mem_cgroup.
285 * Changes to pc->mem_cgroup happens when
288 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
289 * It is added to LRU before charge.
290 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
291 * When moving account, the page is not on LRU. It's isolated.
294 void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
296 struct page_cgroup *pc;
297 struct mem_cgroup *mem;
298 struct mem_cgroup_per_zone *mz;
300 if (mem_cgroup_disabled())
302 pc = lookup_page_cgroup(page);
303 /* can happen while we handle swapcache. */
304 if (list_empty(&pc->lru))
306 mz = page_cgroup_zoneinfo(pc);
307 mem = pc->mem_cgroup;
308 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
309 list_del_init(&pc->lru);
313 void mem_cgroup_del_lru(struct page *page)
315 mem_cgroup_del_lru_list(page, page_lru(page));
318 void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
320 struct mem_cgroup_per_zone *mz;
321 struct page_cgroup *pc;
323 if (mem_cgroup_disabled())
326 pc = lookup_page_cgroup(page);
328 /* unused page is not rotated. */
329 if (!PageCgroupUsed(pc))
331 mz = page_cgroup_zoneinfo(pc);
332 list_move(&pc->lru, &mz->lists[lru]);
335 void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
337 struct page_cgroup *pc;
338 struct mem_cgroup_per_zone *mz;
340 if (mem_cgroup_disabled())
342 pc = lookup_page_cgroup(page);
343 /* barrier to sync with "charge" */
345 if (!PageCgroupUsed(pc))
348 mz = page_cgroup_zoneinfo(pc);
349 MEM_CGROUP_ZSTAT(mz, lru) += 1;
350 list_add(&pc->lru, &mz->lists[lru]);
353 * To add swapcache into LRU. Be careful to all this function.
354 * zone->lru_lock shouldn't be held and irq must not be disabled.
356 static void mem_cgroup_lru_fixup(struct page *page)
358 if (!isolate_lru_page(page))
359 putback_lru_page(page);
362 void mem_cgroup_move_lists(struct page *page,
363 enum lru_list from, enum lru_list to)
365 if (mem_cgroup_disabled())
367 mem_cgroup_del_lru_list(page, from);
368 mem_cgroup_add_lru_list(page, to);
371 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
376 ret = task->mm && mm_match_cgroup(task->mm, mem);
382 * Calculate mapped_ratio under memory controller. This will be used in
383 * vmscan.c for deteremining we have to reclaim mapped pages.
385 int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem)
390 * usage is recorded in bytes. But, here, we assume the number of
391 * physical pages can be represented by "long" on any arch.
393 total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L;
394 rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
395 return (int)((rss * 100L) / total);
399 * prev_priority control...this will be used in memory reclaim path.
401 int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
403 return mem->prev_priority;
406 void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
408 if (priority < mem->prev_priority)
409 mem->prev_priority = priority;
412 void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
414 mem->prev_priority = priority;
417 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg, struct zone *zone)
419 unsigned long active;
420 unsigned long inactive;
422 inactive = mem_cgroup_get_all_zonestat(memcg, LRU_INACTIVE_ANON);
423 active = mem_cgroup_get_all_zonestat(memcg, LRU_ACTIVE_ANON);
425 if (inactive * memcg->inactive_ratio < active)
431 unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg,
435 int nid = zone->zone_pgdat->node_id;
436 int zid = zone_idx(zone);
437 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
439 return MEM_CGROUP_ZSTAT(mz, lru);
442 struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
445 int nid = zone->zone_pgdat->node_id;
446 int zid = zone_idx(zone);
447 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
449 return &mz->reclaim_stat;
452 struct zone_reclaim_stat *
453 mem_cgroup_get_reclaim_stat_from_page(struct page *page)
455 struct page_cgroup *pc;
456 struct mem_cgroup_per_zone *mz;
458 if (mem_cgroup_disabled())
461 pc = lookup_page_cgroup(page);
462 mz = page_cgroup_zoneinfo(pc);
466 return &mz->reclaim_stat;
469 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
470 struct list_head *dst,
471 unsigned long *scanned, int order,
472 int mode, struct zone *z,
473 struct mem_cgroup *mem_cont,
474 int active, int file)
476 unsigned long nr_taken = 0;
480 struct list_head *src;
481 struct page_cgroup *pc, *tmp;
482 int nid = z->zone_pgdat->node_id;
483 int zid = zone_idx(z);
484 struct mem_cgroup_per_zone *mz;
485 int lru = LRU_FILE * !!file + !!active;
488 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
489 src = &mz->lists[lru];
492 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
493 if (scan >= nr_to_scan)
497 if (unlikely(!PageCgroupUsed(pc)))
499 if (unlikely(!PageLRU(page)))
503 if (__isolate_lru_page(page, mode, file) == 0) {
504 list_move(&page->lru, dst);
513 #define mem_cgroup_from_res_counter(counter, member) \
514 container_of(counter, struct mem_cgroup, member)
517 * This routine finds the DFS walk successor. This routine should be
518 * called with cgroup_mutex held
520 static struct mem_cgroup *
521 mem_cgroup_get_next_node(struct mem_cgroup *curr, struct mem_cgroup *root_mem)
523 struct cgroup *cgroup, *curr_cgroup, *root_cgroup;
525 curr_cgroup = curr->css.cgroup;
526 root_cgroup = root_mem->css.cgroup;
528 if (!list_empty(&curr_cgroup->children)) {
530 * Walk down to children
532 mem_cgroup_put(curr);
533 cgroup = list_entry(curr_cgroup->children.next,
534 struct cgroup, sibling);
535 curr = mem_cgroup_from_cont(cgroup);
536 mem_cgroup_get(curr);
541 if (curr_cgroup == root_cgroup) {
542 mem_cgroup_put(curr);
544 mem_cgroup_get(curr);
551 if (curr_cgroup->sibling.next != &curr_cgroup->parent->children) {
552 mem_cgroup_put(curr);
553 cgroup = list_entry(curr_cgroup->sibling.next, struct cgroup,
555 curr = mem_cgroup_from_cont(cgroup);
556 mem_cgroup_get(curr);
561 * Go up to next parent and next parent's sibling if need be
563 curr_cgroup = curr_cgroup->parent;
567 root_mem->last_scanned_child = curr;
572 * Visit the first child (need not be the first child as per the ordering
573 * of the cgroup list, since we track last_scanned_child) of @mem and use
574 * that to reclaim free pages from.
576 static struct mem_cgroup *
577 mem_cgroup_get_first_node(struct mem_cgroup *root_mem)
579 struct cgroup *cgroup;
580 struct mem_cgroup *ret;
581 bool obsolete = (root_mem->last_scanned_child &&
582 root_mem->last_scanned_child->obsolete);
585 * Scan all children under the mem_cgroup mem
588 if (list_empty(&root_mem->css.cgroup->children)) {
593 if (!root_mem->last_scanned_child || obsolete) {
596 mem_cgroup_put(root_mem->last_scanned_child);
598 cgroup = list_first_entry(&root_mem->css.cgroup->children,
599 struct cgroup, sibling);
600 ret = mem_cgroup_from_cont(cgroup);
603 ret = mem_cgroup_get_next_node(root_mem->last_scanned_child,
607 root_mem->last_scanned_child = ret;
612 static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem)
614 if (do_swap_account) {
615 if (res_counter_check_under_limit(&mem->res) &&
616 res_counter_check_under_limit(&mem->memsw))
619 if (res_counter_check_under_limit(&mem->res))
625 * Dance down the hierarchy if needed to reclaim memory. We remember the
626 * last child we reclaimed from, so that we don't end up penalizing
627 * one child extensively based on its position in the children list.
629 * root_mem is the original ancestor that we've been reclaim from.
631 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
632 gfp_t gfp_mask, bool noswap)
634 struct mem_cgroup *next_mem;
638 * Reclaim unconditionally and don't check for return value.
639 * We need to reclaim in the current group and down the tree.
640 * One might think about checking for children before reclaiming,
641 * but there might be left over accounting, even after children
644 ret = try_to_free_mem_cgroup_pages(root_mem, gfp_mask, noswap);
645 if (mem_cgroup_check_under_limit(root_mem))
647 if (!root_mem->use_hierarchy)
650 next_mem = mem_cgroup_get_first_node(root_mem);
652 while (next_mem != root_mem) {
653 if (next_mem->obsolete) {
654 mem_cgroup_put(next_mem);
656 next_mem = mem_cgroup_get_first_node(root_mem);
660 ret = try_to_free_mem_cgroup_pages(next_mem, gfp_mask, noswap);
661 if (mem_cgroup_check_under_limit(root_mem))
664 next_mem = mem_cgroup_get_next_node(next_mem, root_mem);
670 bool mem_cgroup_oom_called(struct task_struct *task)
673 struct mem_cgroup *mem;
674 struct mm_struct *mm;
680 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
681 if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10))
687 * Unlike exported interface, "oom" parameter is added. if oom==true,
688 * oom-killer can be invoked.
690 static int __mem_cgroup_try_charge(struct mm_struct *mm,
691 gfp_t gfp_mask, struct mem_cgroup **memcg,
694 struct mem_cgroup *mem, *mem_over_limit;
695 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
696 struct res_counter *fail_res;
698 if (unlikely(test_thread_flag(TIF_MEMDIE))) {
699 /* Don't account this! */
705 * We always charge the cgroup the mm_struct belongs to.
706 * The mm_struct's mem_cgroup changes on task migration if the
707 * thread group leader migrates. It's possible that mm is not
708 * set, if so charge the init_mm (happens for pagecache usage).
710 if (likely(!*memcg)) {
712 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
713 if (unlikely(!mem)) {
718 * For every charge from the cgroup, increment reference count
732 ret = res_counter_charge(&mem->res, PAGE_SIZE, &fail_res);
734 if (!do_swap_account)
736 ret = res_counter_charge(&mem->memsw, PAGE_SIZE,
740 /* mem+swap counter fails */
741 res_counter_uncharge(&mem->res, PAGE_SIZE);
743 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
746 /* mem counter fails */
747 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
750 if (!(gfp_mask & __GFP_WAIT))
753 ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, gfp_mask,
757 * try_to_free_mem_cgroup_pages() might not give us a full
758 * picture of reclaim. Some pages are reclaimed and might be
759 * moved to swap cache or just unmapped from the cgroup.
760 * Check the limit again to see if the reclaim reduced the
761 * current usage of the cgroup before giving up
764 if (mem_cgroup_check_under_limit(mem_over_limit))
769 mem_cgroup_out_of_memory(mem_over_limit, gfp_mask);
770 mem_over_limit->last_oom_jiffies = jiffies;
782 * mem_cgroup_try_charge - get charge of PAGE_SIZE.
783 * @mm: an mm_struct which is charged against. (when *memcg is NULL)
784 * @gfp_mask: gfp_mask for reclaim.
785 * @memcg: a pointer to memory cgroup which is charged against.
787 * charge against memory cgroup pointed by *memcg. if *memcg == NULL, estimated
788 * memory cgroup from @mm is got and stored in *memcg.
790 * Returns 0 if success. -ENOMEM at failure.
791 * This call can invoke OOM-Killer.
794 int mem_cgroup_try_charge(struct mm_struct *mm,
795 gfp_t mask, struct mem_cgroup **memcg)
797 return __mem_cgroup_try_charge(mm, mask, memcg, true);
801 * commit a charge got by mem_cgroup_try_charge() and makes page_cgroup to be
802 * USED state. If already USED, uncharge and return.
805 static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
806 struct page_cgroup *pc,
807 enum charge_type ctype)
809 /* try_charge() can return NULL to *memcg, taking care of it. */
813 lock_page_cgroup(pc);
814 if (unlikely(PageCgroupUsed(pc))) {
815 unlock_page_cgroup(pc);
816 res_counter_uncharge(&mem->res, PAGE_SIZE);
818 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
822 pc->mem_cgroup = mem;
824 pc->flags = pcg_default_flags[ctype];
826 mem_cgroup_charge_statistics(mem, pc, true);
828 unlock_page_cgroup(pc);
832 * mem_cgroup_move_account - move account of the page
833 * @pc: page_cgroup of the page.
834 * @from: mem_cgroup which the page is moved from.
835 * @to: mem_cgroup which the page is moved to. @from != @to.
837 * The caller must confirm following.
838 * - page is not on LRU (isolate_page() is useful.)
840 * returns 0 at success,
841 * returns -EBUSY when lock is busy or "pc" is unstable.
843 * This function does "uncharge" from old cgroup but doesn't do "charge" to
844 * new cgroup. It should be done by a caller.
847 static int mem_cgroup_move_account(struct page_cgroup *pc,
848 struct mem_cgroup *from, struct mem_cgroup *to)
850 struct mem_cgroup_per_zone *from_mz, *to_mz;
854 VM_BUG_ON(from == to);
855 VM_BUG_ON(PageLRU(pc->page));
857 nid = page_cgroup_nid(pc);
858 zid = page_cgroup_zid(pc);
859 from_mz = mem_cgroup_zoneinfo(from, nid, zid);
860 to_mz = mem_cgroup_zoneinfo(to, nid, zid);
862 if (!trylock_page_cgroup(pc))
865 if (!PageCgroupUsed(pc))
868 if (pc->mem_cgroup != from)
872 res_counter_uncharge(&from->res, PAGE_SIZE);
873 mem_cgroup_charge_statistics(from, pc, false);
875 res_counter_uncharge(&from->memsw, PAGE_SIZE);
877 mem_cgroup_charge_statistics(to, pc, true);
881 unlock_page_cgroup(pc);
886 * move charges to its parent.
889 static int mem_cgroup_move_parent(struct page_cgroup *pc,
890 struct mem_cgroup *child,
893 struct page *page = pc->page;
894 struct cgroup *cg = child->css.cgroup;
895 struct cgroup *pcg = cg->parent;
896 struct mem_cgroup *parent;
904 parent = mem_cgroup_from_cont(pcg);
907 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false);
911 if (!get_page_unless_zero(page))
914 ret = isolate_lru_page(page);
919 ret = mem_cgroup_move_account(pc, child, parent);
921 /* drop extra refcnt by try_charge() (move_account increment one) */
922 css_put(&parent->css);
923 putback_lru_page(page);
928 /* uncharge if move fails */
930 res_counter_uncharge(&parent->res, PAGE_SIZE);
932 res_counter_uncharge(&parent->memsw, PAGE_SIZE);
938 * Charge the memory controller for page usage.
940 * 0 if the charge was successful
941 * < 0 if the cgroup is over its limit
943 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
944 gfp_t gfp_mask, enum charge_type ctype,
945 struct mem_cgroup *memcg)
947 struct mem_cgroup *mem;
948 struct page_cgroup *pc;
951 pc = lookup_page_cgroup(page);
952 /* can happen at boot */
958 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true);
962 __mem_cgroup_commit_charge(mem, pc, ctype);
966 int mem_cgroup_newpage_charge(struct page *page,
967 struct mm_struct *mm, gfp_t gfp_mask)
969 if (mem_cgroup_disabled())
971 if (PageCompound(page))
974 * If already mapped, we don't have to account.
975 * If page cache, page->mapping has address_space.
976 * But page->mapping may have out-of-use anon_vma pointer,
977 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
980 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
984 return mem_cgroup_charge_common(page, mm, gfp_mask,
985 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
988 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
991 if (mem_cgroup_disabled())
993 if (PageCompound(page))
996 * Corner case handling. This is called from add_to_page_cache()
997 * in usual. But some FS (shmem) precharges this page before calling it
998 * and call add_to_page_cache() with GFP_NOWAIT.
1000 * For GFP_NOWAIT case, the page may be pre-charged before calling
1001 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
1002 * charge twice. (It works but has to pay a bit larger cost.)
1004 if (!(gfp_mask & __GFP_WAIT)) {
1005 struct page_cgroup *pc;
1008 pc = lookup_page_cgroup(page);
1011 lock_page_cgroup(pc);
1012 if (PageCgroupUsed(pc)) {
1013 unlock_page_cgroup(pc);
1016 unlock_page_cgroup(pc);
1022 if (page_is_file_cache(page))
1023 return mem_cgroup_charge_common(page, mm, gfp_mask,
1024 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
1026 return mem_cgroup_charge_common(page, mm, gfp_mask,
1027 MEM_CGROUP_CHARGE_TYPE_SHMEM, NULL);
1030 int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
1032 gfp_t mask, struct mem_cgroup **ptr)
1034 struct mem_cgroup *mem;
1037 if (mem_cgroup_disabled())
1040 if (!do_swap_account)
1044 * A racing thread's fault, or swapoff, may have already updated
1045 * the pte, and even removed page from swap cache: return success
1046 * to go on to do_swap_page()'s pte_same() test, which should fail.
1048 if (!PageSwapCache(page))
1051 ent.val = page_private(page);
1053 mem = lookup_swap_cgroup(ent);
1054 if (!mem || mem->obsolete)
1057 return __mem_cgroup_try_charge(NULL, mask, ptr, true);
1061 return __mem_cgroup_try_charge(mm, mask, ptr, true);
1066 int mem_cgroup_cache_charge_swapin(struct page *page,
1067 struct mm_struct *mm, gfp_t mask, bool locked)
1071 if (mem_cgroup_disabled())
1078 * If not locked, the page can be dropped from SwapCache until
1081 if (PageSwapCache(page)) {
1082 struct mem_cgroup *mem = NULL;
1085 ent.val = page_private(page);
1086 if (do_swap_account) {
1087 mem = lookup_swap_cgroup(ent);
1088 if (mem && mem->obsolete)
1093 ret = mem_cgroup_charge_common(page, mm, mask,
1094 MEM_CGROUP_CHARGE_TYPE_SHMEM, mem);
1096 if (!ret && do_swap_account) {
1097 /* avoid double counting */
1098 mem = swap_cgroup_record(ent, NULL);
1100 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1101 mem_cgroup_put(mem);
1107 /* add this page(page_cgroup) to the LRU we want. */
1108 mem_cgroup_lru_fixup(page);
1114 void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
1116 struct page_cgroup *pc;
1118 if (mem_cgroup_disabled())
1122 pc = lookup_page_cgroup(page);
1123 __mem_cgroup_commit_charge(ptr, pc, MEM_CGROUP_CHARGE_TYPE_MAPPED);
1125 * Now swap is on-memory. This means this page may be
1126 * counted both as mem and swap....double count.
1127 * Fix it by uncharging from memsw. This SwapCache is stable
1128 * because we're still under lock_page().
1130 if (do_swap_account) {
1131 swp_entry_t ent = {.val = page_private(page)};
1132 struct mem_cgroup *memcg;
1133 memcg = swap_cgroup_record(ent, NULL);
1135 /* If memcg is obsolete, memcg can be != ptr */
1136 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1137 mem_cgroup_put(memcg);
1141 /* add this page(page_cgroup) to the LRU we want. */
1142 mem_cgroup_lru_fixup(page);
1145 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
1147 if (mem_cgroup_disabled())
1151 res_counter_uncharge(&mem->res, PAGE_SIZE);
1152 if (do_swap_account)
1153 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1159 * uncharge if !page_mapped(page)
1161 static struct mem_cgroup *
1162 __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
1164 struct page_cgroup *pc;
1165 struct mem_cgroup *mem = NULL;
1166 struct mem_cgroup_per_zone *mz;
1168 if (mem_cgroup_disabled())
1171 if (PageSwapCache(page))
1175 * Check if our page_cgroup is valid
1177 pc = lookup_page_cgroup(page);
1178 if (unlikely(!pc || !PageCgroupUsed(pc)))
1181 lock_page_cgroup(pc);
1183 mem = pc->mem_cgroup;
1185 if (!PageCgroupUsed(pc))
1189 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
1190 if (page_mapped(page))
1193 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
1194 if (!PageAnon(page)) { /* Shared memory */
1195 if (page->mapping && !page_is_file_cache(page))
1197 } else if (page_mapped(page)) /* Anon */
1204 res_counter_uncharge(&mem->res, PAGE_SIZE);
1205 if (do_swap_account && (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT))
1206 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1208 mem_cgroup_charge_statistics(mem, pc, false);
1209 ClearPageCgroupUsed(pc);
1211 mz = page_cgroup_zoneinfo(pc);
1212 unlock_page_cgroup(pc);
1214 /* at swapout, this memcg will be accessed to record to swap */
1215 if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
1221 unlock_page_cgroup(pc);
1225 void mem_cgroup_uncharge_page(struct page *page)
1228 if (page_mapped(page))
1230 if (page->mapping && !PageAnon(page))
1232 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
1235 void mem_cgroup_uncharge_cache_page(struct page *page)
1237 VM_BUG_ON(page_mapped(page));
1238 VM_BUG_ON(page->mapping);
1239 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
1243 * called from __delete_from_swap_cache() and drop "page" account.
1244 * memcg information is recorded to swap_cgroup of "ent"
1246 void mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent)
1248 struct mem_cgroup *memcg;
1250 memcg = __mem_cgroup_uncharge_common(page,
1251 MEM_CGROUP_CHARGE_TYPE_SWAPOUT);
1252 /* record memcg information */
1253 if (do_swap_account && memcg) {
1254 swap_cgroup_record(ent, memcg);
1255 mem_cgroup_get(memcg);
1258 css_put(&memcg->css);
1261 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1263 * called from swap_entry_free(). remove record in swap_cgroup and
1264 * uncharge "memsw" account.
1266 void mem_cgroup_uncharge_swap(swp_entry_t ent)
1268 struct mem_cgroup *memcg;
1270 if (!do_swap_account)
1273 memcg = swap_cgroup_record(ent, NULL);
1275 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1276 mem_cgroup_put(memcg);
1282 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
1285 int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr)
1287 struct page_cgroup *pc;
1288 struct mem_cgroup *mem = NULL;
1291 if (mem_cgroup_disabled())
1294 pc = lookup_page_cgroup(page);
1295 lock_page_cgroup(pc);
1296 if (PageCgroupUsed(pc)) {
1297 mem = pc->mem_cgroup;
1300 unlock_page_cgroup(pc);
1303 ret = mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem);
1310 /* remove redundant charge if migration failed*/
1311 void mem_cgroup_end_migration(struct mem_cgroup *mem,
1312 struct page *oldpage, struct page *newpage)
1314 struct page *target, *unused;
1315 struct page_cgroup *pc;
1316 enum charge_type ctype;
1321 /* at migration success, oldpage->mapping is NULL. */
1322 if (oldpage->mapping) {
1330 if (PageAnon(target))
1331 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
1332 else if (page_is_file_cache(target))
1333 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
1335 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
1337 /* unused page is not on radix-tree now. */
1339 __mem_cgroup_uncharge_common(unused, ctype);
1341 pc = lookup_page_cgroup(target);
1343 * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup.
1344 * So, double-counting is effectively avoided.
1346 __mem_cgroup_commit_charge(mem, pc, ctype);
1349 * Both of oldpage and newpage are still under lock_page().
1350 * Then, we don't have to care about race in radix-tree.
1351 * But we have to be careful that this page is unmapped or not.
1353 * There is a case for !page_mapped(). At the start of
1354 * migration, oldpage was mapped. But now, it's zapped.
1355 * But we know *target* page is not freed/reused under us.
1356 * mem_cgroup_uncharge_page() does all necessary checks.
1358 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
1359 mem_cgroup_uncharge_page(target);
1363 * A call to try to shrink memory usage under specified resource controller.
1364 * This is typically used for page reclaiming for shmem for reducing side
1365 * effect of page allocation from shmem, which is used by some mem_cgroup.
1367 int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask)
1369 struct mem_cgroup *mem;
1371 int retry = MEM_CGROUP_RECLAIM_RETRIES;
1373 if (mem_cgroup_disabled())
1379 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
1380 if (unlikely(!mem)) {
1388 progress = try_to_free_mem_cgroup_pages(mem, gfp_mask, true);
1389 progress += mem_cgroup_check_under_limit(mem);
1390 } while (!progress && --retry);
1399 * The inactive anon list should be small enough that the VM never has to
1400 * do too much work, but large enough that each inactive page has a chance
1401 * to be referenced again before it is swapped out.
1403 * this calculation is straightforward porting from
1404 * page_alloc.c::setup_per_zone_inactive_ratio().
1405 * it describe more detail.
1407 static void mem_cgroup_set_inactive_ratio(struct mem_cgroup *memcg)
1409 unsigned int gb, ratio;
1411 gb = res_counter_read_u64(&memcg->res, RES_LIMIT) >> 30;
1413 ratio = int_sqrt(10 * gb);
1417 memcg->inactive_ratio = ratio;
1421 static DEFINE_MUTEX(set_limit_mutex);
1423 static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
1424 unsigned long long val)
1427 int retry_count = MEM_CGROUP_RECLAIM_RETRIES;
1432 while (retry_count) {
1433 if (signal_pending(current)) {
1438 * Rather than hide all in some function, I do this in
1439 * open coded manner. You see what this really does.
1440 * We have to guarantee mem->res.limit < mem->memsw.limit.
1442 mutex_lock(&set_limit_mutex);
1443 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1444 if (memswlimit < val) {
1446 mutex_unlock(&set_limit_mutex);
1449 ret = res_counter_set_limit(&memcg->res, val);
1450 mutex_unlock(&set_limit_mutex);
1455 progress = try_to_free_mem_cgroup_pages(memcg,
1457 if (!progress) retry_count--;
1461 mem_cgroup_set_inactive_ratio(memcg);
1466 int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
1467 unsigned long long val)
1469 int retry_count = MEM_CGROUP_RECLAIM_RETRIES;
1470 u64 memlimit, oldusage, curusage;
1473 if (!do_swap_account)
1476 while (retry_count) {
1477 if (signal_pending(current)) {
1482 * Rather than hide all in some function, I do this in
1483 * open coded manner. You see what this really does.
1484 * We have to guarantee mem->res.limit < mem->memsw.limit.
1486 mutex_lock(&set_limit_mutex);
1487 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1488 if (memlimit > val) {
1490 mutex_unlock(&set_limit_mutex);
1493 ret = res_counter_set_limit(&memcg->memsw, val);
1494 mutex_unlock(&set_limit_mutex);
1499 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
1500 try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, true);
1501 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
1502 if (curusage >= oldusage)
1509 * This routine traverse page_cgroup in given list and drop them all.
1510 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
1512 static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
1513 int node, int zid, enum lru_list lru)
1516 struct mem_cgroup_per_zone *mz;
1517 struct page_cgroup *pc, *busy;
1518 unsigned long flags, loop;
1519 struct list_head *list;
1522 zone = &NODE_DATA(node)->node_zones[zid];
1523 mz = mem_cgroup_zoneinfo(mem, node, zid);
1524 list = &mz->lists[lru];
1526 loop = MEM_CGROUP_ZSTAT(mz, lru);
1527 /* give some margin against EBUSY etc...*/
1532 spin_lock_irqsave(&zone->lru_lock, flags);
1533 if (list_empty(list)) {
1534 spin_unlock_irqrestore(&zone->lru_lock, flags);
1537 pc = list_entry(list->prev, struct page_cgroup, lru);
1539 list_move(&pc->lru, list);
1541 spin_unlock_irqrestore(&zone->lru_lock, flags);
1544 spin_unlock_irqrestore(&zone->lru_lock, flags);
1546 ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL);
1550 if (ret == -EBUSY || ret == -EINVAL) {
1551 /* found lock contention or "pc" is obsolete. */
1558 if (!ret && !list_empty(list))
1564 * make mem_cgroup's charge to be 0 if there is no task.
1565 * This enables deleting this mem_cgroup.
1567 static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
1570 int node, zid, shrink;
1571 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
1572 struct cgroup *cgrp = mem->css.cgroup;
1577 /* should free all ? */
1581 while (mem->res.usage > 0) {
1583 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
1586 if (signal_pending(current))
1588 /* This is for making all *used* pages to be on LRU. */
1589 lru_add_drain_all();
1591 for_each_node_state(node, N_POSSIBLE) {
1592 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
1595 ret = mem_cgroup_force_empty_list(mem,
1604 /* it seems parent cgroup doesn't have enough mem */
1615 /* returns EBUSY if there is a task or if we come here twice. */
1616 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
1620 /* we call try-to-free pages for make this cgroup empty */
1621 lru_add_drain_all();
1622 /* try to free all pages in this cgroup */
1624 while (nr_retries && mem->res.usage > 0) {
1627 if (signal_pending(current)) {
1631 progress = try_to_free_mem_cgroup_pages(mem,
1635 /* maybe some writeback is necessary */
1636 congestion_wait(WRITE, HZ/10);
1641 /* try move_account...there may be some *locked* pages. */
1648 int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
1650 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
1654 static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
1656 return mem_cgroup_from_cont(cont)->use_hierarchy;
1659 static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
1663 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1664 struct cgroup *parent = cont->parent;
1665 struct mem_cgroup *parent_mem = NULL;
1668 parent_mem = mem_cgroup_from_cont(parent);
1672 * If parent's use_hiearchy is set, we can't make any modifications
1673 * in the child subtrees. If it is unset, then the change can
1674 * occur, provided the current cgroup has no children.
1676 * For the root cgroup, parent_mem is NULL, we allow value to be
1677 * set if there are no children.
1679 if ((!parent_mem || !parent_mem->use_hierarchy) &&
1680 (val == 1 || val == 0)) {
1681 if (list_empty(&cont->children))
1682 mem->use_hierarchy = val;
1692 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
1694 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1698 type = MEMFILE_TYPE(cft->private);
1699 name = MEMFILE_ATTR(cft->private);
1702 val = res_counter_read_u64(&mem->res, name);
1705 if (do_swap_account)
1706 val = res_counter_read_u64(&mem->memsw, name);
1715 * The user of this function is...
1718 static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
1721 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
1723 unsigned long long val;
1726 type = MEMFILE_TYPE(cft->private);
1727 name = MEMFILE_ATTR(cft->private);
1730 /* This function does all necessary parse...reuse it */
1731 ret = res_counter_memparse_write_strategy(buffer, &val);
1735 ret = mem_cgroup_resize_limit(memcg, val);
1737 ret = mem_cgroup_resize_memsw_limit(memcg, val);
1740 ret = -EINVAL; /* should be BUG() ? */
1746 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
1748 struct mem_cgroup *mem;
1751 mem = mem_cgroup_from_cont(cont);
1752 type = MEMFILE_TYPE(event);
1753 name = MEMFILE_ATTR(event);
1757 res_counter_reset_max(&mem->res);
1759 res_counter_reset_max(&mem->memsw);
1763 res_counter_reset_failcnt(&mem->res);
1765 res_counter_reset_failcnt(&mem->memsw);
1771 static const struct mem_cgroup_stat_desc {
1774 } mem_cgroup_stat_desc[] = {
1775 [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, },
1776 [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, },
1777 [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, },
1778 [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, },
1781 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
1782 struct cgroup_map_cb *cb)
1784 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
1785 struct mem_cgroup_stat *stat = &mem_cont->stat;
1788 for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) {
1791 val = mem_cgroup_read_stat(stat, i);
1792 val *= mem_cgroup_stat_desc[i].unit;
1793 cb->fill(cb, mem_cgroup_stat_desc[i].msg, val);
1795 /* showing # of active pages */
1797 unsigned long active_anon, inactive_anon;
1798 unsigned long active_file, inactive_file;
1799 unsigned long unevictable;
1801 inactive_anon = mem_cgroup_get_all_zonestat(mem_cont,
1803 active_anon = mem_cgroup_get_all_zonestat(mem_cont,
1805 inactive_file = mem_cgroup_get_all_zonestat(mem_cont,
1807 active_file = mem_cgroup_get_all_zonestat(mem_cont,
1809 unevictable = mem_cgroup_get_all_zonestat(mem_cont,
1812 cb->fill(cb, "active_anon", (active_anon) * PAGE_SIZE);
1813 cb->fill(cb, "inactive_anon", (inactive_anon) * PAGE_SIZE);
1814 cb->fill(cb, "active_file", (active_file) * PAGE_SIZE);
1815 cb->fill(cb, "inactive_file", (inactive_file) * PAGE_SIZE);
1816 cb->fill(cb, "unevictable", unevictable * PAGE_SIZE);
1820 #ifdef CONFIG_DEBUG_VM
1821 cb->fill(cb, "inactive_ratio", mem_cont->inactive_ratio);
1825 struct mem_cgroup_per_zone *mz;
1826 unsigned long recent_rotated[2] = {0, 0};
1827 unsigned long recent_scanned[2] = {0, 0};
1829 for_each_online_node(nid)
1830 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1831 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
1833 recent_rotated[0] +=
1834 mz->reclaim_stat.recent_rotated[0];
1835 recent_rotated[1] +=
1836 mz->reclaim_stat.recent_rotated[1];
1837 recent_scanned[0] +=
1838 mz->reclaim_stat.recent_scanned[0];
1839 recent_scanned[1] +=
1840 mz->reclaim_stat.recent_scanned[1];
1842 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
1843 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
1844 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
1845 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
1853 static struct cftype mem_cgroup_files[] = {
1855 .name = "usage_in_bytes",
1856 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
1857 .read_u64 = mem_cgroup_read,
1860 .name = "max_usage_in_bytes",
1861 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
1862 .trigger = mem_cgroup_reset,
1863 .read_u64 = mem_cgroup_read,
1866 .name = "limit_in_bytes",
1867 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
1868 .write_string = mem_cgroup_write,
1869 .read_u64 = mem_cgroup_read,
1873 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
1874 .trigger = mem_cgroup_reset,
1875 .read_u64 = mem_cgroup_read,
1879 .read_map = mem_control_stat_show,
1882 .name = "force_empty",
1883 .trigger = mem_cgroup_force_empty_write,
1886 .name = "use_hierarchy",
1887 .write_u64 = mem_cgroup_hierarchy_write,
1888 .read_u64 = mem_cgroup_hierarchy_read,
1892 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1893 static struct cftype memsw_cgroup_files[] = {
1895 .name = "memsw.usage_in_bytes",
1896 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
1897 .read_u64 = mem_cgroup_read,
1900 .name = "memsw.max_usage_in_bytes",
1901 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
1902 .trigger = mem_cgroup_reset,
1903 .read_u64 = mem_cgroup_read,
1906 .name = "memsw.limit_in_bytes",
1907 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
1908 .write_string = mem_cgroup_write,
1909 .read_u64 = mem_cgroup_read,
1912 .name = "memsw.failcnt",
1913 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
1914 .trigger = mem_cgroup_reset,
1915 .read_u64 = mem_cgroup_read,
1919 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
1921 if (!do_swap_account)
1923 return cgroup_add_files(cont, ss, memsw_cgroup_files,
1924 ARRAY_SIZE(memsw_cgroup_files));
1927 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
1933 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
1935 struct mem_cgroup_per_node *pn;
1936 struct mem_cgroup_per_zone *mz;
1938 int zone, tmp = node;
1940 * This routine is called against possible nodes.
1941 * But it's BUG to call kmalloc() against offline node.
1943 * TODO: this routine can waste much memory for nodes which will
1944 * never be onlined. It's better to use memory hotplug callback
1947 if (!node_state(node, N_NORMAL_MEMORY))
1949 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
1953 mem->info.nodeinfo[node] = pn;
1954 memset(pn, 0, sizeof(*pn));
1956 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
1957 mz = &pn->zoneinfo[zone];
1959 INIT_LIST_HEAD(&mz->lists[l]);
1964 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
1966 kfree(mem->info.nodeinfo[node]);
1969 static int mem_cgroup_size(void)
1971 int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu);
1972 return sizeof(struct mem_cgroup) + cpustat_size;
1975 static struct mem_cgroup *mem_cgroup_alloc(void)
1977 struct mem_cgroup *mem;
1978 int size = mem_cgroup_size();
1980 if (size < PAGE_SIZE)
1981 mem = kmalloc(size, GFP_KERNEL);
1983 mem = vmalloc(size);
1986 memset(mem, 0, size);
1991 * At destroying mem_cgroup, references from swap_cgroup can remain.
1992 * (scanning all at force_empty is too costly...)
1994 * Instead of clearing all references at force_empty, we remember
1995 * the number of reference from swap_cgroup and free mem_cgroup when
1996 * it goes down to 0.
1998 * When mem_cgroup is destroyed, mem->obsolete will be set to 0 and
1999 * entry which points to this memcg will be ignore at swapin.
2001 * Removal of cgroup itself succeeds regardless of refs from swap.
2004 static void mem_cgroup_free(struct mem_cgroup *mem)
2008 if (atomic_read(&mem->refcnt) > 0)
2012 for_each_node_state(node, N_POSSIBLE)
2013 free_mem_cgroup_per_zone_info(mem, node);
2015 if (mem_cgroup_size() < PAGE_SIZE)
2021 static void mem_cgroup_get(struct mem_cgroup *mem)
2023 atomic_inc(&mem->refcnt);
2026 static void mem_cgroup_put(struct mem_cgroup *mem)
2028 if (atomic_dec_and_test(&mem->refcnt)) {
2031 mem_cgroup_free(mem);
2036 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2037 static void __init enable_swap_cgroup(void)
2039 if (!mem_cgroup_disabled() && really_do_swap_account)
2040 do_swap_account = 1;
2043 static void __init enable_swap_cgroup(void)
2048 static struct cgroup_subsys_state *
2049 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
2051 struct mem_cgroup *mem, *parent;
2054 mem = mem_cgroup_alloc();
2056 return ERR_PTR(-ENOMEM);
2058 for_each_node_state(node, N_POSSIBLE)
2059 if (alloc_mem_cgroup_per_zone_info(mem, node))
2062 if (cont->parent == NULL) {
2063 enable_swap_cgroup();
2066 parent = mem_cgroup_from_cont(cont->parent);
2067 mem->use_hierarchy = parent->use_hierarchy;
2070 if (parent && parent->use_hierarchy) {
2071 res_counter_init(&mem->res, &parent->res);
2072 res_counter_init(&mem->memsw, &parent->memsw);
2074 res_counter_init(&mem->res, NULL);
2075 res_counter_init(&mem->memsw, NULL);
2077 mem_cgroup_set_inactive_ratio(mem);
2078 mem->last_scanned_child = NULL;
2082 for_each_node_state(node, N_POSSIBLE)
2083 free_mem_cgroup_per_zone_info(mem, node);
2084 mem_cgroup_free(mem);
2085 return ERR_PTR(-ENOMEM);
2088 static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
2089 struct cgroup *cont)
2091 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2093 mem_cgroup_force_empty(mem, false);
2096 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
2097 struct cgroup *cont)
2099 mem_cgroup_free(mem_cgroup_from_cont(cont));
2102 static int mem_cgroup_populate(struct cgroup_subsys *ss,
2103 struct cgroup *cont)
2107 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
2108 ARRAY_SIZE(mem_cgroup_files));
2111 ret = register_memsw_files(cont, ss);
2115 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
2116 struct cgroup *cont,
2117 struct cgroup *old_cont,
2118 struct task_struct *p)
2121 * FIXME: It's better to move charges of this process from old
2122 * memcg to new memcg. But it's just on TODO-List now.
2126 struct cgroup_subsys mem_cgroup_subsys = {
2128 .subsys_id = mem_cgroup_subsys_id,
2129 .create = mem_cgroup_create,
2130 .pre_destroy = mem_cgroup_pre_destroy,
2131 .destroy = mem_cgroup_destroy,
2132 .populate = mem_cgroup_populate,
2133 .attach = mem_cgroup_move_task,
2137 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2139 static int __init disable_swap_account(char *s)
2141 really_do_swap_account = 0;
2144 __setup("noswapaccount", disable_swap_account);