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 /* Macro for accessing counter */
108 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
110 struct mem_cgroup_per_node {
111 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
114 struct mem_cgroup_lru_info {
115 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
119 * The memory controller data structure. The memory controller controls both
120 * page cache and RSS per cgroup. We would eventually like to provide
121 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
122 * to help the administrator determine what knobs to tune.
124 * TODO: Add a water mark for the memory controller. Reclaim will begin when
125 * we hit the water mark. May be even add a low water mark, such that
126 * no reclaim occurs from a cgroup at it's low water mark, this is
127 * a feature that will be implemented much later in the future.
130 struct cgroup_subsys_state css;
132 * the counter to account for memory usage
134 struct res_counter res;
136 * the counter to account for mem+swap usage.
138 struct res_counter memsw;
140 * Per cgroup active and inactive list, similar to the
141 * per zone LRU lists.
143 struct mem_cgroup_lru_info info;
145 int prev_priority; /* for recording reclaim priority */
148 * While reclaiming in a hiearchy, we cache the last child we
149 * reclaimed from. Protected by cgroup_lock()
151 struct mem_cgroup *last_scanned_child;
153 * Should the accounting and control be hierarchical, per subtree?
156 unsigned long last_oom_jiffies;
160 * statistics. This must be placed at the end of memcg.
162 struct mem_cgroup_stat stat;
166 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
167 MEM_CGROUP_CHARGE_TYPE_MAPPED,
168 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
169 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
170 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
174 /* only for here (for easy reading.) */
175 #define PCGF_CACHE (1UL << PCG_CACHE)
176 #define PCGF_USED (1UL << PCG_USED)
177 #define PCGF_LOCK (1UL << PCG_LOCK)
178 static const unsigned long
179 pcg_default_flags[NR_CHARGE_TYPE] = {
180 PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* File Cache */
181 PCGF_USED | PCGF_LOCK, /* Anon */
182 PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* Shmem */
187 /* for encoding cft->private value on file */
190 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
191 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
192 #define MEMFILE_ATTR(val) ((val) & 0xffff)
194 static void mem_cgroup_get(struct mem_cgroup *mem);
195 static void mem_cgroup_put(struct mem_cgroup *mem);
197 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
198 struct page_cgroup *pc,
201 int val = (charge)? 1 : -1;
202 struct mem_cgroup_stat *stat = &mem->stat;
203 struct mem_cgroup_stat_cpu *cpustat;
206 cpustat = &stat->cpustat[cpu];
207 if (PageCgroupCache(pc))
208 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val);
210 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val);
213 __mem_cgroup_stat_add_safe(cpustat,
214 MEM_CGROUP_STAT_PGPGIN_COUNT, 1);
216 __mem_cgroup_stat_add_safe(cpustat,
217 MEM_CGROUP_STAT_PGPGOUT_COUNT, 1);
221 static struct mem_cgroup_per_zone *
222 mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
224 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
227 static struct mem_cgroup_per_zone *
228 page_cgroup_zoneinfo(struct page_cgroup *pc)
230 struct mem_cgroup *mem = pc->mem_cgroup;
231 int nid = page_cgroup_nid(pc);
232 int zid = page_cgroup_zid(pc);
234 return mem_cgroup_zoneinfo(mem, nid, zid);
237 static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem,
241 struct mem_cgroup_per_zone *mz;
244 for_each_online_node(nid)
245 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
246 mz = mem_cgroup_zoneinfo(mem, nid, zid);
247 total += MEM_CGROUP_ZSTAT(mz, idx);
252 static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
254 return container_of(cgroup_subsys_state(cont,
255 mem_cgroup_subsys_id), struct mem_cgroup,
259 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
262 * mm_update_next_owner() may clear mm->owner to NULL
263 * if it races with swapoff, page migration, etc.
264 * So this can be called with p == NULL.
269 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
270 struct mem_cgroup, css);
274 * Following LRU functions are allowed to be used without PCG_LOCK.
275 * Operations are called by routine of global LRU independently from memcg.
276 * What we have to take care of here is validness of pc->mem_cgroup.
278 * Changes to pc->mem_cgroup happens when
281 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
282 * It is added to LRU before charge.
283 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
284 * When moving account, the page is not on LRU. It's isolated.
287 void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
289 struct page_cgroup *pc;
290 struct mem_cgroup *mem;
291 struct mem_cgroup_per_zone *mz;
293 if (mem_cgroup_disabled())
295 pc = lookup_page_cgroup(page);
296 /* can happen while we handle swapcache. */
297 if (list_empty(&pc->lru))
299 mz = page_cgroup_zoneinfo(pc);
300 mem = pc->mem_cgroup;
301 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
302 list_del_init(&pc->lru);
306 void mem_cgroup_del_lru(struct page *page)
308 mem_cgroup_del_lru_list(page, page_lru(page));
311 void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
313 struct mem_cgroup_per_zone *mz;
314 struct page_cgroup *pc;
316 if (mem_cgroup_disabled())
319 pc = lookup_page_cgroup(page);
321 /* unused page is not rotated. */
322 if (!PageCgroupUsed(pc))
324 mz = page_cgroup_zoneinfo(pc);
325 list_move(&pc->lru, &mz->lists[lru]);
328 void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
330 struct page_cgroup *pc;
331 struct mem_cgroup_per_zone *mz;
333 if (mem_cgroup_disabled())
335 pc = lookup_page_cgroup(page);
336 /* barrier to sync with "charge" */
338 if (!PageCgroupUsed(pc))
341 mz = page_cgroup_zoneinfo(pc);
342 MEM_CGROUP_ZSTAT(mz, lru) += 1;
343 list_add(&pc->lru, &mz->lists[lru]);
346 * To add swapcache into LRU. Be careful to all this function.
347 * zone->lru_lock shouldn't be held and irq must not be disabled.
349 static void mem_cgroup_lru_fixup(struct page *page)
351 if (!isolate_lru_page(page))
352 putback_lru_page(page);
355 void mem_cgroup_move_lists(struct page *page,
356 enum lru_list from, enum lru_list to)
358 if (mem_cgroup_disabled())
360 mem_cgroup_del_lru_list(page, from);
361 mem_cgroup_add_lru_list(page, to);
364 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
369 ret = task->mm && mm_match_cgroup(task->mm, mem);
375 * Calculate mapped_ratio under memory controller. This will be used in
376 * vmscan.c for deteremining we have to reclaim mapped pages.
378 int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem)
383 * usage is recorded in bytes. But, here, we assume the number of
384 * physical pages can be represented by "long" on any arch.
386 total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L;
387 rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
388 return (int)((rss * 100L) / total);
392 * prev_priority control...this will be used in memory reclaim path.
394 int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
396 return mem->prev_priority;
399 void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
401 if (priority < mem->prev_priority)
402 mem->prev_priority = priority;
405 void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
407 mem->prev_priority = priority;
411 * Calculate # of pages to be scanned in this priority/zone.
414 * priority starts from "DEF_PRIORITY" and decremented in each loop.
415 * (see include/linux/mmzone.h)
418 long mem_cgroup_calc_reclaim(struct mem_cgroup *mem, struct zone *zone,
419 int priority, enum lru_list lru)
422 int nid = zone->zone_pgdat->node_id;
423 int zid = zone_idx(zone);
424 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid);
426 nr_pages = MEM_CGROUP_ZSTAT(mz, lru);
428 return (nr_pages >> priority);
431 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
432 struct list_head *dst,
433 unsigned long *scanned, int order,
434 int mode, struct zone *z,
435 struct mem_cgroup *mem_cont,
436 int active, int file)
438 unsigned long nr_taken = 0;
442 struct list_head *src;
443 struct page_cgroup *pc, *tmp;
444 int nid = z->zone_pgdat->node_id;
445 int zid = zone_idx(z);
446 struct mem_cgroup_per_zone *mz;
447 int lru = LRU_FILE * !!file + !!active;
450 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
451 src = &mz->lists[lru];
454 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
455 if (scan >= nr_to_scan)
459 if (unlikely(!PageCgroupUsed(pc)))
461 if (unlikely(!PageLRU(page)))
465 if (__isolate_lru_page(page, mode, file) == 0) {
466 list_move(&page->lru, dst);
475 #define mem_cgroup_from_res_counter(counter, member) \
476 container_of(counter, struct mem_cgroup, member)
479 * This routine finds the DFS walk successor. This routine should be
480 * called with cgroup_mutex held
482 static struct mem_cgroup *
483 mem_cgroup_get_next_node(struct mem_cgroup *curr, struct mem_cgroup *root_mem)
485 struct cgroup *cgroup, *curr_cgroup, *root_cgroup;
487 curr_cgroup = curr->css.cgroup;
488 root_cgroup = root_mem->css.cgroup;
490 if (!list_empty(&curr_cgroup->children)) {
492 * Walk down to children
494 mem_cgroup_put(curr);
495 cgroup = list_entry(curr_cgroup->children.next,
496 struct cgroup, sibling);
497 curr = mem_cgroup_from_cont(cgroup);
498 mem_cgroup_get(curr);
503 if (curr_cgroup == root_cgroup) {
504 mem_cgroup_put(curr);
506 mem_cgroup_get(curr);
513 if (curr_cgroup->sibling.next != &curr_cgroup->parent->children) {
514 mem_cgroup_put(curr);
515 cgroup = list_entry(curr_cgroup->sibling.next, struct cgroup,
517 curr = mem_cgroup_from_cont(cgroup);
518 mem_cgroup_get(curr);
523 * Go up to next parent and next parent's sibling if need be
525 curr_cgroup = curr_cgroup->parent;
529 root_mem->last_scanned_child = curr;
534 * Visit the first child (need not be the first child as per the ordering
535 * of the cgroup list, since we track last_scanned_child) of @mem and use
536 * that to reclaim free pages from.
538 static struct mem_cgroup *
539 mem_cgroup_get_first_node(struct mem_cgroup *root_mem)
541 struct cgroup *cgroup;
542 struct mem_cgroup *ret;
543 bool obsolete = (root_mem->last_scanned_child &&
544 root_mem->last_scanned_child->obsolete);
547 * Scan all children under the mem_cgroup mem
550 if (list_empty(&root_mem->css.cgroup->children)) {
555 if (!root_mem->last_scanned_child || obsolete) {
558 mem_cgroup_put(root_mem->last_scanned_child);
560 cgroup = list_first_entry(&root_mem->css.cgroup->children,
561 struct cgroup, sibling);
562 ret = mem_cgroup_from_cont(cgroup);
565 ret = mem_cgroup_get_next_node(root_mem->last_scanned_child,
569 root_mem->last_scanned_child = ret;
574 static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem)
576 if (do_swap_account) {
577 if (res_counter_check_under_limit(&mem->res) &&
578 res_counter_check_under_limit(&mem->memsw))
581 if (res_counter_check_under_limit(&mem->res))
587 * Dance down the hierarchy if needed to reclaim memory. We remember the
588 * last child we reclaimed from, so that we don't end up penalizing
589 * one child extensively based on its position in the children list.
591 * root_mem is the original ancestor that we've been reclaim from.
593 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
594 gfp_t gfp_mask, bool noswap)
596 struct mem_cgroup *next_mem;
600 * Reclaim unconditionally and don't check for return value.
601 * We need to reclaim in the current group and down the tree.
602 * One might think about checking for children before reclaiming,
603 * but there might be left over accounting, even after children
606 ret = try_to_free_mem_cgroup_pages(root_mem, gfp_mask, noswap);
607 if (mem_cgroup_check_under_limit(root_mem))
610 next_mem = mem_cgroup_get_first_node(root_mem);
612 while (next_mem != root_mem) {
613 if (next_mem->obsolete) {
614 mem_cgroup_put(next_mem);
616 next_mem = mem_cgroup_get_first_node(root_mem);
620 ret = try_to_free_mem_cgroup_pages(next_mem, gfp_mask, noswap);
621 if (mem_cgroup_check_under_limit(root_mem))
624 next_mem = mem_cgroup_get_next_node(next_mem, root_mem);
630 bool mem_cgroup_oom_called(struct task_struct *task)
633 struct mem_cgroup *mem;
634 struct mm_struct *mm;
640 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
641 if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10))
647 * Unlike exported interface, "oom" parameter is added. if oom==true,
648 * oom-killer can be invoked.
650 static int __mem_cgroup_try_charge(struct mm_struct *mm,
651 gfp_t gfp_mask, struct mem_cgroup **memcg,
654 struct mem_cgroup *mem, *mem_over_limit;
655 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
656 struct res_counter *fail_res;
658 if (unlikely(test_thread_flag(TIF_MEMDIE))) {
659 /* Don't account this! */
665 * We always charge the cgroup the mm_struct belongs to.
666 * The mm_struct's mem_cgroup changes on task migration if the
667 * thread group leader migrates. It's possible that mm is not
668 * set, if so charge the init_mm (happens for pagecache usage).
670 if (likely(!*memcg)) {
672 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
673 if (unlikely(!mem)) {
678 * For every charge from the cgroup, increment reference count
692 ret = res_counter_charge(&mem->res, PAGE_SIZE, &fail_res);
694 if (!do_swap_account)
696 ret = res_counter_charge(&mem->memsw, PAGE_SIZE,
700 /* mem+swap counter fails */
701 res_counter_uncharge(&mem->res, PAGE_SIZE);
703 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
706 /* mem counter fails */
707 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
710 if (!(gfp_mask & __GFP_WAIT))
713 ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, gfp_mask,
717 * try_to_free_mem_cgroup_pages() might not give us a full
718 * picture of reclaim. Some pages are reclaimed and might be
719 * moved to swap cache or just unmapped from the cgroup.
720 * Check the limit again to see if the reclaim reduced the
721 * current usage of the cgroup before giving up
724 if (mem_cgroup_check_under_limit(mem_over_limit))
729 mem_cgroup_out_of_memory(mem_over_limit, gfp_mask);
730 mem_over_limit->last_oom_jiffies = jiffies;
742 * mem_cgroup_try_charge - get charge of PAGE_SIZE.
743 * @mm: an mm_struct which is charged against. (when *memcg is NULL)
744 * @gfp_mask: gfp_mask for reclaim.
745 * @memcg: a pointer to memory cgroup which is charged against.
747 * charge against memory cgroup pointed by *memcg. if *memcg == NULL, estimated
748 * memory cgroup from @mm is got and stored in *memcg.
750 * Returns 0 if success. -ENOMEM at failure.
751 * This call can invoke OOM-Killer.
754 int mem_cgroup_try_charge(struct mm_struct *mm,
755 gfp_t mask, struct mem_cgroup **memcg)
757 return __mem_cgroup_try_charge(mm, mask, memcg, true);
761 * commit a charge got by mem_cgroup_try_charge() and makes page_cgroup to be
762 * USED state. If already USED, uncharge and return.
765 static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
766 struct page_cgroup *pc,
767 enum charge_type ctype)
769 /* try_charge() can return NULL to *memcg, taking care of it. */
773 lock_page_cgroup(pc);
774 if (unlikely(PageCgroupUsed(pc))) {
775 unlock_page_cgroup(pc);
776 res_counter_uncharge(&mem->res, PAGE_SIZE);
778 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
782 pc->mem_cgroup = mem;
784 pc->flags = pcg_default_flags[ctype];
786 mem_cgroup_charge_statistics(mem, pc, true);
788 unlock_page_cgroup(pc);
792 * mem_cgroup_move_account - move account of the page
793 * @pc: page_cgroup of the page.
794 * @from: mem_cgroup which the page is moved from.
795 * @to: mem_cgroup which the page is moved to. @from != @to.
797 * The caller must confirm following.
798 * - page is not on LRU (isolate_page() is useful.)
800 * returns 0 at success,
801 * returns -EBUSY when lock is busy or "pc" is unstable.
803 * This function does "uncharge" from old cgroup but doesn't do "charge" to
804 * new cgroup. It should be done by a caller.
807 static int mem_cgroup_move_account(struct page_cgroup *pc,
808 struct mem_cgroup *from, struct mem_cgroup *to)
810 struct mem_cgroup_per_zone *from_mz, *to_mz;
814 VM_BUG_ON(from == to);
815 VM_BUG_ON(PageLRU(pc->page));
817 nid = page_cgroup_nid(pc);
818 zid = page_cgroup_zid(pc);
819 from_mz = mem_cgroup_zoneinfo(from, nid, zid);
820 to_mz = mem_cgroup_zoneinfo(to, nid, zid);
822 if (!trylock_page_cgroup(pc))
825 if (!PageCgroupUsed(pc))
828 if (pc->mem_cgroup != from)
832 res_counter_uncharge(&from->res, PAGE_SIZE);
833 mem_cgroup_charge_statistics(from, pc, false);
835 res_counter_uncharge(&from->memsw, PAGE_SIZE);
837 mem_cgroup_charge_statistics(to, pc, true);
841 unlock_page_cgroup(pc);
846 * move charges to its parent.
849 static int mem_cgroup_move_parent(struct page_cgroup *pc,
850 struct mem_cgroup *child,
853 struct page *page = pc->page;
854 struct cgroup *cg = child->css.cgroup;
855 struct cgroup *pcg = cg->parent;
856 struct mem_cgroup *parent;
864 parent = mem_cgroup_from_cont(pcg);
867 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false);
871 if (!get_page_unless_zero(page))
874 ret = isolate_lru_page(page);
879 ret = mem_cgroup_move_account(pc, child, parent);
881 /* drop extra refcnt by try_charge() (move_account increment one) */
882 css_put(&parent->css);
883 putback_lru_page(page);
888 /* uncharge if move fails */
890 res_counter_uncharge(&parent->res, PAGE_SIZE);
892 res_counter_uncharge(&parent->memsw, PAGE_SIZE);
898 * Charge the memory controller for page usage.
900 * 0 if the charge was successful
901 * < 0 if the cgroup is over its limit
903 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
904 gfp_t gfp_mask, enum charge_type ctype,
905 struct mem_cgroup *memcg)
907 struct mem_cgroup *mem;
908 struct page_cgroup *pc;
911 pc = lookup_page_cgroup(page);
912 /* can happen at boot */
918 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true);
922 __mem_cgroup_commit_charge(mem, pc, ctype);
926 int mem_cgroup_newpage_charge(struct page *page,
927 struct mm_struct *mm, gfp_t gfp_mask)
929 if (mem_cgroup_disabled())
931 if (PageCompound(page))
934 * If already mapped, we don't have to account.
935 * If page cache, page->mapping has address_space.
936 * But page->mapping may have out-of-use anon_vma pointer,
937 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
940 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
944 return mem_cgroup_charge_common(page, mm, gfp_mask,
945 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
948 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
951 if (mem_cgroup_disabled())
953 if (PageCompound(page))
956 * Corner case handling. This is called from add_to_page_cache()
957 * in usual. But some FS (shmem) precharges this page before calling it
958 * and call add_to_page_cache() with GFP_NOWAIT.
960 * For GFP_NOWAIT case, the page may be pre-charged before calling
961 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
962 * charge twice. (It works but has to pay a bit larger cost.)
964 if (!(gfp_mask & __GFP_WAIT)) {
965 struct page_cgroup *pc;
968 pc = lookup_page_cgroup(page);
971 lock_page_cgroup(pc);
972 if (PageCgroupUsed(pc)) {
973 unlock_page_cgroup(pc);
976 unlock_page_cgroup(pc);
982 if (page_is_file_cache(page))
983 return mem_cgroup_charge_common(page, mm, gfp_mask,
984 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
986 return mem_cgroup_charge_common(page, mm, gfp_mask,
987 MEM_CGROUP_CHARGE_TYPE_SHMEM, NULL);
990 int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
992 gfp_t mask, struct mem_cgroup **ptr)
994 struct mem_cgroup *mem;
997 if (mem_cgroup_disabled())
1000 if (!do_swap_account)
1004 * A racing thread's fault, or swapoff, may have already updated
1005 * the pte, and even removed page from swap cache: return success
1006 * to go on to do_swap_page()'s pte_same() test, which should fail.
1008 if (!PageSwapCache(page))
1011 ent.val = page_private(page);
1013 mem = lookup_swap_cgroup(ent);
1014 if (!mem || mem->obsolete)
1017 return __mem_cgroup_try_charge(NULL, mask, ptr, true);
1021 return __mem_cgroup_try_charge(mm, mask, ptr, true);
1026 int mem_cgroup_cache_charge_swapin(struct page *page,
1027 struct mm_struct *mm, gfp_t mask, bool locked)
1031 if (mem_cgroup_disabled())
1038 * If not locked, the page can be dropped from SwapCache until
1041 if (PageSwapCache(page)) {
1042 struct mem_cgroup *mem = NULL;
1045 ent.val = page_private(page);
1046 if (do_swap_account) {
1047 mem = lookup_swap_cgroup(ent);
1048 if (mem && mem->obsolete)
1053 ret = mem_cgroup_charge_common(page, mm, mask,
1054 MEM_CGROUP_CHARGE_TYPE_SHMEM, mem);
1056 if (!ret && do_swap_account) {
1057 /* avoid double counting */
1058 mem = swap_cgroup_record(ent, NULL);
1060 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1061 mem_cgroup_put(mem);
1067 /* add this page(page_cgroup) to the LRU we want. */
1068 mem_cgroup_lru_fixup(page);
1074 void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
1076 struct page_cgroup *pc;
1078 if (mem_cgroup_disabled())
1082 pc = lookup_page_cgroup(page);
1083 __mem_cgroup_commit_charge(ptr, pc, MEM_CGROUP_CHARGE_TYPE_MAPPED);
1085 * Now swap is on-memory. This means this page may be
1086 * counted both as mem and swap....double count.
1087 * Fix it by uncharging from memsw. This SwapCache is stable
1088 * because we're still under lock_page().
1090 if (do_swap_account) {
1091 swp_entry_t ent = {.val = page_private(page)};
1092 struct mem_cgroup *memcg;
1093 memcg = swap_cgroup_record(ent, NULL);
1095 /* If memcg is obsolete, memcg can be != ptr */
1096 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1097 mem_cgroup_put(memcg);
1101 /* add this page(page_cgroup) to the LRU we want. */
1102 mem_cgroup_lru_fixup(page);
1105 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
1107 if (mem_cgroup_disabled())
1111 res_counter_uncharge(&mem->res, PAGE_SIZE);
1112 if (do_swap_account)
1113 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1119 * uncharge if !page_mapped(page)
1121 static struct mem_cgroup *
1122 __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
1124 struct page_cgroup *pc;
1125 struct mem_cgroup *mem = NULL;
1126 struct mem_cgroup_per_zone *mz;
1128 if (mem_cgroup_disabled())
1131 if (PageSwapCache(page))
1135 * Check if our page_cgroup is valid
1137 pc = lookup_page_cgroup(page);
1138 if (unlikely(!pc || !PageCgroupUsed(pc)))
1141 lock_page_cgroup(pc);
1143 mem = pc->mem_cgroup;
1145 if (!PageCgroupUsed(pc))
1149 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
1150 if (page_mapped(page))
1153 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
1154 if (!PageAnon(page)) { /* Shared memory */
1155 if (page->mapping && !page_is_file_cache(page))
1157 } else if (page_mapped(page)) /* Anon */
1164 res_counter_uncharge(&mem->res, PAGE_SIZE);
1165 if (do_swap_account && (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT))
1166 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1168 mem_cgroup_charge_statistics(mem, pc, false);
1169 ClearPageCgroupUsed(pc);
1171 mz = page_cgroup_zoneinfo(pc);
1172 unlock_page_cgroup(pc);
1174 /* at swapout, this memcg will be accessed to record to swap */
1175 if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
1181 unlock_page_cgroup(pc);
1185 void mem_cgroup_uncharge_page(struct page *page)
1188 if (page_mapped(page))
1190 if (page->mapping && !PageAnon(page))
1192 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
1195 void mem_cgroup_uncharge_cache_page(struct page *page)
1197 VM_BUG_ON(page_mapped(page));
1198 VM_BUG_ON(page->mapping);
1199 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
1203 * called from __delete_from_swap_cache() and drop "page" account.
1204 * memcg information is recorded to swap_cgroup of "ent"
1206 void mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent)
1208 struct mem_cgroup *memcg;
1210 memcg = __mem_cgroup_uncharge_common(page,
1211 MEM_CGROUP_CHARGE_TYPE_SWAPOUT);
1212 /* record memcg information */
1213 if (do_swap_account && memcg) {
1214 swap_cgroup_record(ent, memcg);
1215 mem_cgroup_get(memcg);
1218 css_put(&memcg->css);
1221 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1223 * called from swap_entry_free(). remove record in swap_cgroup and
1224 * uncharge "memsw" account.
1226 void mem_cgroup_uncharge_swap(swp_entry_t ent)
1228 struct mem_cgroup *memcg;
1230 if (!do_swap_account)
1233 memcg = swap_cgroup_record(ent, NULL);
1235 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1236 mem_cgroup_put(memcg);
1242 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
1245 int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr)
1247 struct page_cgroup *pc;
1248 struct mem_cgroup *mem = NULL;
1251 if (mem_cgroup_disabled())
1254 pc = lookup_page_cgroup(page);
1255 lock_page_cgroup(pc);
1256 if (PageCgroupUsed(pc)) {
1257 mem = pc->mem_cgroup;
1260 unlock_page_cgroup(pc);
1263 ret = mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem);
1270 /* remove redundant charge if migration failed*/
1271 void mem_cgroup_end_migration(struct mem_cgroup *mem,
1272 struct page *oldpage, struct page *newpage)
1274 struct page *target, *unused;
1275 struct page_cgroup *pc;
1276 enum charge_type ctype;
1281 /* at migration success, oldpage->mapping is NULL. */
1282 if (oldpage->mapping) {
1290 if (PageAnon(target))
1291 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
1292 else if (page_is_file_cache(target))
1293 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
1295 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
1297 /* unused page is not on radix-tree now. */
1299 __mem_cgroup_uncharge_common(unused, ctype);
1301 pc = lookup_page_cgroup(target);
1303 * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup.
1304 * So, double-counting is effectively avoided.
1306 __mem_cgroup_commit_charge(mem, pc, ctype);
1309 * Both of oldpage and newpage are still under lock_page().
1310 * Then, we don't have to care about race in radix-tree.
1311 * But we have to be careful that this page is unmapped or not.
1313 * There is a case for !page_mapped(). At the start of
1314 * migration, oldpage was mapped. But now, it's zapped.
1315 * But we know *target* page is not freed/reused under us.
1316 * mem_cgroup_uncharge_page() does all necessary checks.
1318 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
1319 mem_cgroup_uncharge_page(target);
1323 * A call to try to shrink memory usage under specified resource controller.
1324 * This is typically used for page reclaiming for shmem for reducing side
1325 * effect of page allocation from shmem, which is used by some mem_cgroup.
1327 int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask)
1329 struct mem_cgroup *mem;
1331 int retry = MEM_CGROUP_RECLAIM_RETRIES;
1333 if (mem_cgroup_disabled())
1339 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
1340 if (unlikely(!mem)) {
1348 progress = try_to_free_mem_cgroup_pages(mem, gfp_mask, true);
1349 progress += mem_cgroup_check_under_limit(mem);
1350 } while (!progress && --retry);
1358 static DEFINE_MUTEX(set_limit_mutex);
1360 static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
1361 unsigned long long val)
1364 int retry_count = MEM_CGROUP_RECLAIM_RETRIES;
1369 while (retry_count) {
1370 if (signal_pending(current)) {
1375 * Rather than hide all in some function, I do this in
1376 * open coded manner. You see what this really does.
1377 * We have to guarantee mem->res.limit < mem->memsw.limit.
1379 mutex_lock(&set_limit_mutex);
1380 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1381 if (memswlimit < val) {
1383 mutex_unlock(&set_limit_mutex);
1386 ret = res_counter_set_limit(&memcg->res, val);
1387 mutex_unlock(&set_limit_mutex);
1392 progress = try_to_free_mem_cgroup_pages(memcg,
1394 if (!progress) retry_count--;
1399 int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
1400 unsigned long long val)
1402 int retry_count = MEM_CGROUP_RECLAIM_RETRIES;
1403 u64 memlimit, oldusage, curusage;
1406 if (!do_swap_account)
1409 while (retry_count) {
1410 if (signal_pending(current)) {
1415 * Rather than hide all in some function, I do this in
1416 * open coded manner. You see what this really does.
1417 * We have to guarantee mem->res.limit < mem->memsw.limit.
1419 mutex_lock(&set_limit_mutex);
1420 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1421 if (memlimit > val) {
1423 mutex_unlock(&set_limit_mutex);
1426 ret = res_counter_set_limit(&memcg->memsw, val);
1427 mutex_unlock(&set_limit_mutex);
1432 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
1433 try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, true);
1434 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
1435 if (curusage >= oldusage)
1442 * This routine traverse page_cgroup in given list and drop them all.
1443 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
1445 static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
1446 int node, int zid, enum lru_list lru)
1449 struct mem_cgroup_per_zone *mz;
1450 struct page_cgroup *pc, *busy;
1451 unsigned long flags, loop;
1452 struct list_head *list;
1455 zone = &NODE_DATA(node)->node_zones[zid];
1456 mz = mem_cgroup_zoneinfo(mem, node, zid);
1457 list = &mz->lists[lru];
1459 loop = MEM_CGROUP_ZSTAT(mz, lru);
1460 /* give some margin against EBUSY etc...*/
1465 spin_lock_irqsave(&zone->lru_lock, flags);
1466 if (list_empty(list)) {
1467 spin_unlock_irqrestore(&zone->lru_lock, flags);
1470 pc = list_entry(list->prev, struct page_cgroup, lru);
1472 list_move(&pc->lru, list);
1474 spin_unlock_irqrestore(&zone->lru_lock, flags);
1477 spin_unlock_irqrestore(&zone->lru_lock, flags);
1479 ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL);
1483 if (ret == -EBUSY || ret == -EINVAL) {
1484 /* found lock contention or "pc" is obsolete. */
1491 if (!ret && !list_empty(list))
1497 * make mem_cgroup's charge to be 0 if there is no task.
1498 * This enables deleting this mem_cgroup.
1500 static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
1503 int node, zid, shrink;
1504 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
1505 struct cgroup *cgrp = mem->css.cgroup;
1510 /* should free all ? */
1514 while (mem->res.usage > 0) {
1516 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
1519 if (signal_pending(current))
1521 /* This is for making all *used* pages to be on LRU. */
1522 lru_add_drain_all();
1524 for_each_node_state(node, N_POSSIBLE) {
1525 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
1528 ret = mem_cgroup_force_empty_list(mem,
1537 /* it seems parent cgroup doesn't have enough mem */
1548 /* returns EBUSY if there is a task or if we come here twice. */
1549 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
1553 /* we call try-to-free pages for make this cgroup empty */
1554 lru_add_drain_all();
1555 /* try to free all pages in this cgroup */
1557 while (nr_retries && mem->res.usage > 0) {
1560 if (signal_pending(current)) {
1564 progress = try_to_free_mem_cgroup_pages(mem,
1568 /* maybe some writeback is necessary */
1569 congestion_wait(WRITE, HZ/10);
1574 /* try move_account...there may be some *locked* pages. */
1581 int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
1583 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
1587 static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
1589 return mem_cgroup_from_cont(cont)->use_hierarchy;
1592 static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
1596 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1597 struct cgroup *parent = cont->parent;
1598 struct mem_cgroup *parent_mem = NULL;
1601 parent_mem = mem_cgroup_from_cont(parent);
1605 * If parent's use_hiearchy is set, we can't make any modifications
1606 * in the child subtrees. If it is unset, then the change can
1607 * occur, provided the current cgroup has no children.
1609 * For the root cgroup, parent_mem is NULL, we allow value to be
1610 * set if there are no children.
1612 if ((!parent_mem || !parent_mem->use_hierarchy) &&
1613 (val == 1 || val == 0)) {
1614 if (list_empty(&cont->children))
1615 mem->use_hierarchy = val;
1625 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
1627 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1631 type = MEMFILE_TYPE(cft->private);
1632 name = MEMFILE_ATTR(cft->private);
1635 val = res_counter_read_u64(&mem->res, name);
1638 if (do_swap_account)
1639 val = res_counter_read_u64(&mem->memsw, name);
1648 * The user of this function is...
1651 static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
1654 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
1656 unsigned long long val;
1659 type = MEMFILE_TYPE(cft->private);
1660 name = MEMFILE_ATTR(cft->private);
1663 /* This function does all necessary parse...reuse it */
1664 ret = res_counter_memparse_write_strategy(buffer, &val);
1668 ret = mem_cgroup_resize_limit(memcg, val);
1670 ret = mem_cgroup_resize_memsw_limit(memcg, val);
1673 ret = -EINVAL; /* should be BUG() ? */
1679 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
1681 struct mem_cgroup *mem;
1684 mem = mem_cgroup_from_cont(cont);
1685 type = MEMFILE_TYPE(event);
1686 name = MEMFILE_ATTR(event);
1690 res_counter_reset_max(&mem->res);
1692 res_counter_reset_max(&mem->memsw);
1696 res_counter_reset_failcnt(&mem->res);
1698 res_counter_reset_failcnt(&mem->memsw);
1704 static const struct mem_cgroup_stat_desc {
1707 } mem_cgroup_stat_desc[] = {
1708 [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, },
1709 [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, },
1710 [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, },
1711 [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, },
1714 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
1715 struct cgroup_map_cb *cb)
1717 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
1718 struct mem_cgroup_stat *stat = &mem_cont->stat;
1721 for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) {
1724 val = mem_cgroup_read_stat(stat, i);
1725 val *= mem_cgroup_stat_desc[i].unit;
1726 cb->fill(cb, mem_cgroup_stat_desc[i].msg, val);
1728 /* showing # of active pages */
1730 unsigned long active_anon, inactive_anon;
1731 unsigned long active_file, inactive_file;
1732 unsigned long unevictable;
1734 inactive_anon = mem_cgroup_get_all_zonestat(mem_cont,
1736 active_anon = mem_cgroup_get_all_zonestat(mem_cont,
1738 inactive_file = mem_cgroup_get_all_zonestat(mem_cont,
1740 active_file = mem_cgroup_get_all_zonestat(mem_cont,
1742 unevictable = mem_cgroup_get_all_zonestat(mem_cont,
1745 cb->fill(cb, "active_anon", (active_anon) * PAGE_SIZE);
1746 cb->fill(cb, "inactive_anon", (inactive_anon) * PAGE_SIZE);
1747 cb->fill(cb, "active_file", (active_file) * PAGE_SIZE);
1748 cb->fill(cb, "inactive_file", (inactive_file) * PAGE_SIZE);
1749 cb->fill(cb, "unevictable", unevictable * PAGE_SIZE);
1756 static struct cftype mem_cgroup_files[] = {
1758 .name = "usage_in_bytes",
1759 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
1760 .read_u64 = mem_cgroup_read,
1763 .name = "max_usage_in_bytes",
1764 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
1765 .trigger = mem_cgroup_reset,
1766 .read_u64 = mem_cgroup_read,
1769 .name = "limit_in_bytes",
1770 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
1771 .write_string = mem_cgroup_write,
1772 .read_u64 = mem_cgroup_read,
1776 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
1777 .trigger = mem_cgroup_reset,
1778 .read_u64 = mem_cgroup_read,
1782 .read_map = mem_control_stat_show,
1785 .name = "force_empty",
1786 .trigger = mem_cgroup_force_empty_write,
1789 .name = "use_hierarchy",
1790 .write_u64 = mem_cgroup_hierarchy_write,
1791 .read_u64 = mem_cgroup_hierarchy_read,
1795 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1796 static struct cftype memsw_cgroup_files[] = {
1798 .name = "memsw.usage_in_bytes",
1799 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
1800 .read_u64 = mem_cgroup_read,
1803 .name = "memsw.max_usage_in_bytes",
1804 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
1805 .trigger = mem_cgroup_reset,
1806 .read_u64 = mem_cgroup_read,
1809 .name = "memsw.limit_in_bytes",
1810 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
1811 .write_string = mem_cgroup_write,
1812 .read_u64 = mem_cgroup_read,
1815 .name = "memsw.failcnt",
1816 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
1817 .trigger = mem_cgroup_reset,
1818 .read_u64 = mem_cgroup_read,
1822 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
1824 if (!do_swap_account)
1826 return cgroup_add_files(cont, ss, memsw_cgroup_files,
1827 ARRAY_SIZE(memsw_cgroup_files));
1830 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
1836 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
1838 struct mem_cgroup_per_node *pn;
1839 struct mem_cgroup_per_zone *mz;
1841 int zone, tmp = node;
1843 * This routine is called against possible nodes.
1844 * But it's BUG to call kmalloc() against offline node.
1846 * TODO: this routine can waste much memory for nodes which will
1847 * never be onlined. It's better to use memory hotplug callback
1850 if (!node_state(node, N_NORMAL_MEMORY))
1852 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
1856 mem->info.nodeinfo[node] = pn;
1857 memset(pn, 0, sizeof(*pn));
1859 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
1860 mz = &pn->zoneinfo[zone];
1862 INIT_LIST_HEAD(&mz->lists[l]);
1867 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
1869 kfree(mem->info.nodeinfo[node]);
1872 static int mem_cgroup_size(void)
1874 int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu);
1875 return sizeof(struct mem_cgroup) + cpustat_size;
1878 static struct mem_cgroup *mem_cgroup_alloc(void)
1880 struct mem_cgroup *mem;
1881 int size = mem_cgroup_size();
1883 if (size < PAGE_SIZE)
1884 mem = kmalloc(size, GFP_KERNEL);
1886 mem = vmalloc(size);
1889 memset(mem, 0, size);
1894 * At destroying mem_cgroup, references from swap_cgroup can remain.
1895 * (scanning all at force_empty is too costly...)
1897 * Instead of clearing all references at force_empty, we remember
1898 * the number of reference from swap_cgroup and free mem_cgroup when
1899 * it goes down to 0.
1901 * When mem_cgroup is destroyed, mem->obsolete will be set to 0 and
1902 * entry which points to this memcg will be ignore at swapin.
1904 * Removal of cgroup itself succeeds regardless of refs from swap.
1907 static void mem_cgroup_free(struct mem_cgroup *mem)
1911 if (atomic_read(&mem->refcnt) > 0)
1915 for_each_node_state(node, N_POSSIBLE)
1916 free_mem_cgroup_per_zone_info(mem, node);
1918 if (mem_cgroup_size() < PAGE_SIZE)
1924 static void mem_cgroup_get(struct mem_cgroup *mem)
1926 atomic_inc(&mem->refcnt);
1929 static void mem_cgroup_put(struct mem_cgroup *mem)
1931 if (atomic_dec_and_test(&mem->refcnt)) {
1934 mem_cgroup_free(mem);
1939 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1940 static void __init enable_swap_cgroup(void)
1942 if (!mem_cgroup_disabled() && really_do_swap_account)
1943 do_swap_account = 1;
1946 static void __init enable_swap_cgroup(void)
1951 static struct cgroup_subsys_state *
1952 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
1954 struct mem_cgroup *mem, *parent;
1957 mem = mem_cgroup_alloc();
1959 return ERR_PTR(-ENOMEM);
1961 for_each_node_state(node, N_POSSIBLE)
1962 if (alloc_mem_cgroup_per_zone_info(mem, node))
1965 if (cont->parent == NULL) {
1966 enable_swap_cgroup();
1969 parent = mem_cgroup_from_cont(cont->parent);
1970 mem->use_hierarchy = parent->use_hierarchy;
1973 if (parent && parent->use_hierarchy) {
1974 res_counter_init(&mem->res, &parent->res);
1975 res_counter_init(&mem->memsw, &parent->memsw);
1977 res_counter_init(&mem->res, NULL);
1978 res_counter_init(&mem->memsw, NULL);
1981 mem->last_scanned_child = NULL;
1985 for_each_node_state(node, N_POSSIBLE)
1986 free_mem_cgroup_per_zone_info(mem, node);
1987 mem_cgroup_free(mem);
1988 return ERR_PTR(-ENOMEM);
1991 static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
1992 struct cgroup *cont)
1994 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1996 mem_cgroup_force_empty(mem, false);
1999 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
2000 struct cgroup *cont)
2002 mem_cgroup_free(mem_cgroup_from_cont(cont));
2005 static int mem_cgroup_populate(struct cgroup_subsys *ss,
2006 struct cgroup *cont)
2010 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
2011 ARRAY_SIZE(mem_cgroup_files));
2014 ret = register_memsw_files(cont, ss);
2018 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
2019 struct cgroup *cont,
2020 struct cgroup *old_cont,
2021 struct task_struct *p)
2024 * FIXME: It's better to move charges of this process from old
2025 * memcg to new memcg. But it's just on TODO-List now.
2029 struct cgroup_subsys mem_cgroup_subsys = {
2031 .subsys_id = mem_cgroup_subsys_id,
2032 .create = mem_cgroup_create,
2033 .pre_destroy = mem_cgroup_pre_destroy,
2034 .destroy = mem_cgroup_destroy,
2035 .populate = mem_cgroup_populate,
2036 .attach = mem_cgroup_move_task,
2040 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2042 static int __init disable_swap_account(char *s)
2044 really_do_swap_account = 0;
2047 __setup("noswapaccount", disable_swap_account);