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)
54 static DEFINE_MUTEX(memcg_tasklist); /* can be hold under cgroup_mutex */
57 * Statistics for memory cgroup.
59 enum mem_cgroup_stat_index {
61 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
63 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
64 MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */
65 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
66 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
68 MEM_CGROUP_STAT_NSTATS,
71 struct mem_cgroup_stat_cpu {
72 s64 count[MEM_CGROUP_STAT_NSTATS];
73 } ____cacheline_aligned_in_smp;
75 struct mem_cgroup_stat {
76 struct mem_cgroup_stat_cpu cpustat[0];
80 * For accounting under irq disable, no need for increment preempt count.
82 static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat,
83 enum mem_cgroup_stat_index idx, int val)
85 stat->count[idx] += val;
88 static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat,
89 enum mem_cgroup_stat_index idx)
93 for_each_possible_cpu(cpu)
94 ret += stat->cpustat[cpu].count[idx];
99 * per-zone information in memory controller.
101 struct mem_cgroup_per_zone {
103 * spin_lock to protect the per cgroup LRU
105 struct list_head lists[NR_LRU_LISTS];
106 unsigned long count[NR_LRU_LISTS];
108 struct zone_reclaim_stat reclaim_stat;
110 /* Macro for accessing counter */
111 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
113 struct mem_cgroup_per_node {
114 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
117 struct mem_cgroup_lru_info {
118 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
122 * The memory controller data structure. The memory controller controls both
123 * page cache and RSS per cgroup. We would eventually like to provide
124 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
125 * to help the administrator determine what knobs to tune.
127 * TODO: Add a water mark for the memory controller. Reclaim will begin when
128 * we hit the water mark. May be even add a low water mark, such that
129 * no reclaim occurs from a cgroup at it's low water mark, this is
130 * a feature that will be implemented much later in the future.
133 struct cgroup_subsys_state css;
135 * the counter to account for memory usage
137 struct res_counter res;
139 * the counter to account for mem+swap usage.
141 struct res_counter memsw;
143 * Per cgroup active and inactive list, similar to the
144 * per zone LRU lists.
146 struct mem_cgroup_lru_info info;
149 protect against reclaim related member.
151 spinlock_t reclaim_param_lock;
153 int prev_priority; /* for recording reclaim priority */
156 * While reclaiming in a hiearchy, we cache the last child we
157 * reclaimed from. Protected by cgroup_lock()
159 struct mem_cgroup *last_scanned_child;
161 * Should the accounting and control be hierarchical, per subtree?
164 unsigned long last_oom_jiffies;
167 unsigned int swappiness;
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);
285 static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
287 struct mem_cgroup *mem = NULL;
289 * Because we have no locks, mm->owner's may be being moved to other
290 * cgroup. We use css_tryget() here even if this looks
291 * pessimistic (rather than adding locks here).
295 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
298 } while (!css_tryget(&mem->css));
303 static bool mem_cgroup_is_obsolete(struct mem_cgroup *mem)
307 return css_is_removed(&mem->css);
311 * Following LRU functions are allowed to be used without PCG_LOCK.
312 * Operations are called by routine of global LRU independently from memcg.
313 * What we have to take care of here is validness of pc->mem_cgroup.
315 * Changes to pc->mem_cgroup happens when
318 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
319 * It is added to LRU before charge.
320 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
321 * When moving account, the page is not on LRU. It's isolated.
324 void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
326 struct page_cgroup *pc;
327 struct mem_cgroup *mem;
328 struct mem_cgroup_per_zone *mz;
330 if (mem_cgroup_disabled())
332 pc = lookup_page_cgroup(page);
333 /* can happen while we handle swapcache. */
334 if (list_empty(&pc->lru))
336 mz = page_cgroup_zoneinfo(pc);
337 mem = pc->mem_cgroup;
338 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
339 list_del_init(&pc->lru);
343 void mem_cgroup_del_lru(struct page *page)
345 mem_cgroup_del_lru_list(page, page_lru(page));
348 void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
350 struct mem_cgroup_per_zone *mz;
351 struct page_cgroup *pc;
353 if (mem_cgroup_disabled())
356 pc = lookup_page_cgroup(page);
358 /* unused page is not rotated. */
359 if (!PageCgroupUsed(pc))
361 mz = page_cgroup_zoneinfo(pc);
362 list_move(&pc->lru, &mz->lists[lru]);
365 void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
367 struct page_cgroup *pc;
368 struct mem_cgroup_per_zone *mz;
370 if (mem_cgroup_disabled())
372 pc = lookup_page_cgroup(page);
373 /* barrier to sync with "charge" */
375 if (!PageCgroupUsed(pc))
378 mz = page_cgroup_zoneinfo(pc);
379 MEM_CGROUP_ZSTAT(mz, lru) += 1;
380 list_add(&pc->lru, &mz->lists[lru]);
383 * To add swapcache into LRU. Be careful to all this function.
384 * zone->lru_lock shouldn't be held and irq must not be disabled.
386 static void mem_cgroup_lru_fixup(struct page *page)
388 if (!isolate_lru_page(page))
389 putback_lru_page(page);
392 void mem_cgroup_move_lists(struct page *page,
393 enum lru_list from, enum lru_list to)
395 if (mem_cgroup_disabled())
397 mem_cgroup_del_lru_list(page, from);
398 mem_cgroup_add_lru_list(page, to);
401 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
406 ret = task->mm && mm_match_cgroup(task->mm, mem);
412 * Calculate mapped_ratio under memory controller. This will be used in
413 * vmscan.c for deteremining we have to reclaim mapped pages.
415 int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem)
420 * usage is recorded in bytes. But, here, we assume the number of
421 * physical pages can be represented by "long" on any arch.
423 total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L;
424 rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
425 return (int)((rss * 100L) / total);
429 * prev_priority control...this will be used in memory reclaim path.
431 int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
435 spin_lock(&mem->reclaim_param_lock);
436 prev_priority = mem->prev_priority;
437 spin_unlock(&mem->reclaim_param_lock);
439 return prev_priority;
442 void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
444 spin_lock(&mem->reclaim_param_lock);
445 if (priority < mem->prev_priority)
446 mem->prev_priority = priority;
447 spin_unlock(&mem->reclaim_param_lock);
450 void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
452 spin_lock(&mem->reclaim_param_lock);
453 mem->prev_priority = priority;
454 spin_unlock(&mem->reclaim_param_lock);
457 static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages)
459 unsigned long active;
460 unsigned long inactive;
462 unsigned long inactive_ratio;
464 inactive = mem_cgroup_get_all_zonestat(memcg, LRU_INACTIVE_ANON);
465 active = mem_cgroup_get_all_zonestat(memcg, LRU_ACTIVE_ANON);
467 gb = (inactive + active) >> (30 - PAGE_SHIFT);
469 inactive_ratio = int_sqrt(10 * gb);
474 present_pages[0] = inactive;
475 present_pages[1] = active;
478 return inactive_ratio;
481 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg)
483 unsigned long active;
484 unsigned long inactive;
485 unsigned long present_pages[2];
486 unsigned long inactive_ratio;
488 inactive_ratio = calc_inactive_ratio(memcg, present_pages);
490 inactive = present_pages[0];
491 active = present_pages[1];
493 if (inactive * inactive_ratio < active)
499 unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg,
503 int nid = zone->zone_pgdat->node_id;
504 int zid = zone_idx(zone);
505 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
507 return MEM_CGROUP_ZSTAT(mz, lru);
510 struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
513 int nid = zone->zone_pgdat->node_id;
514 int zid = zone_idx(zone);
515 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
517 return &mz->reclaim_stat;
520 struct zone_reclaim_stat *
521 mem_cgroup_get_reclaim_stat_from_page(struct page *page)
523 struct page_cgroup *pc;
524 struct mem_cgroup_per_zone *mz;
526 if (mem_cgroup_disabled())
529 pc = lookup_page_cgroup(page);
530 mz = page_cgroup_zoneinfo(pc);
534 return &mz->reclaim_stat;
537 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
538 struct list_head *dst,
539 unsigned long *scanned, int order,
540 int mode, struct zone *z,
541 struct mem_cgroup *mem_cont,
542 int active, int file)
544 unsigned long nr_taken = 0;
548 struct list_head *src;
549 struct page_cgroup *pc, *tmp;
550 int nid = z->zone_pgdat->node_id;
551 int zid = zone_idx(z);
552 struct mem_cgroup_per_zone *mz;
553 int lru = LRU_FILE * !!file + !!active;
556 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
557 src = &mz->lists[lru];
560 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
561 if (scan >= nr_to_scan)
565 if (unlikely(!PageCgroupUsed(pc)))
567 if (unlikely(!PageLRU(page)))
571 if (__isolate_lru_page(page, mode, file) == 0) {
572 list_move(&page->lru, dst);
581 #define mem_cgroup_from_res_counter(counter, member) \
582 container_of(counter, struct mem_cgroup, member)
585 * This routine finds the DFS walk successor. This routine should be
586 * called with cgroup_mutex held
588 static struct mem_cgroup *
589 mem_cgroup_get_next_node(struct mem_cgroup *curr, struct mem_cgroup *root_mem)
591 struct cgroup *cgroup, *curr_cgroup, *root_cgroup;
593 curr_cgroup = curr->css.cgroup;
594 root_cgroup = root_mem->css.cgroup;
596 if (!list_empty(&curr_cgroup->children)) {
598 * Walk down to children
600 mem_cgroup_put(curr);
601 cgroup = list_entry(curr_cgroup->children.next,
602 struct cgroup, sibling);
603 curr = mem_cgroup_from_cont(cgroup);
604 mem_cgroup_get(curr);
609 if (curr_cgroup == root_cgroup) {
610 mem_cgroup_put(curr);
612 mem_cgroup_get(curr);
619 if (curr_cgroup->sibling.next != &curr_cgroup->parent->children) {
620 mem_cgroup_put(curr);
621 cgroup = list_entry(curr_cgroup->sibling.next, struct cgroup,
623 curr = mem_cgroup_from_cont(cgroup);
624 mem_cgroup_get(curr);
629 * Go up to next parent and next parent's sibling if need be
631 curr_cgroup = curr_cgroup->parent;
635 root_mem->last_scanned_child = curr;
640 * Visit the first child (need not be the first child as per the ordering
641 * of the cgroup list, since we track last_scanned_child) of @mem and use
642 * that to reclaim free pages from.
644 static struct mem_cgroup *
645 mem_cgroup_get_first_node(struct mem_cgroup *root_mem)
647 struct cgroup *cgroup;
648 struct mem_cgroup *ret;
651 obsolete = mem_cgroup_is_obsolete(root_mem->last_scanned_child);
654 * Scan all children under the mem_cgroup mem
657 if (list_empty(&root_mem->css.cgroup->children)) {
662 if (!root_mem->last_scanned_child || obsolete) {
664 if (obsolete && root_mem->last_scanned_child)
665 mem_cgroup_put(root_mem->last_scanned_child);
667 cgroup = list_first_entry(&root_mem->css.cgroup->children,
668 struct cgroup, sibling);
669 ret = mem_cgroup_from_cont(cgroup);
672 ret = mem_cgroup_get_next_node(root_mem->last_scanned_child,
676 root_mem->last_scanned_child = ret;
681 static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem)
683 if (do_swap_account) {
684 if (res_counter_check_under_limit(&mem->res) &&
685 res_counter_check_under_limit(&mem->memsw))
688 if (res_counter_check_under_limit(&mem->res))
693 static unsigned int get_swappiness(struct mem_cgroup *memcg)
695 struct cgroup *cgrp = memcg->css.cgroup;
696 unsigned int swappiness;
699 if (cgrp->parent == NULL)
700 return vm_swappiness;
702 spin_lock(&memcg->reclaim_param_lock);
703 swappiness = memcg->swappiness;
704 spin_unlock(&memcg->reclaim_param_lock);
710 * Dance down the hierarchy if needed to reclaim memory. We remember the
711 * last child we reclaimed from, so that we don't end up penalizing
712 * one child extensively based on its position in the children list.
714 * root_mem is the original ancestor that we've been reclaim from.
716 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
717 gfp_t gfp_mask, bool noswap)
719 struct mem_cgroup *next_mem;
723 * Reclaim unconditionally and don't check for return value.
724 * We need to reclaim in the current group and down the tree.
725 * One might think about checking for children before reclaiming,
726 * but there might be left over accounting, even after children
729 ret = try_to_free_mem_cgroup_pages(root_mem, gfp_mask, noswap,
730 get_swappiness(root_mem));
731 if (mem_cgroup_check_under_limit(root_mem))
733 if (!root_mem->use_hierarchy)
736 next_mem = mem_cgroup_get_first_node(root_mem);
738 while (next_mem != root_mem) {
739 if (mem_cgroup_is_obsolete(next_mem)) {
740 mem_cgroup_put(next_mem);
742 next_mem = mem_cgroup_get_first_node(root_mem);
746 ret = try_to_free_mem_cgroup_pages(next_mem, gfp_mask, noswap,
747 get_swappiness(next_mem));
748 if (mem_cgroup_check_under_limit(root_mem))
751 next_mem = mem_cgroup_get_next_node(next_mem, root_mem);
757 bool mem_cgroup_oom_called(struct task_struct *task)
760 struct mem_cgroup *mem;
761 struct mm_struct *mm;
767 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
768 if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10))
774 * Unlike exported interface, "oom" parameter is added. if oom==true,
775 * oom-killer can be invoked.
777 static int __mem_cgroup_try_charge(struct mm_struct *mm,
778 gfp_t gfp_mask, struct mem_cgroup **memcg,
781 struct mem_cgroup *mem, *mem_over_limit;
782 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
783 struct res_counter *fail_res;
785 if (unlikely(test_thread_flag(TIF_MEMDIE))) {
786 /* Don't account this! */
792 * We always charge the cgroup the mm_struct belongs to.
793 * The mm_struct's mem_cgroup changes on task migration if the
794 * thread group leader migrates. It's possible that mm is not
795 * set, if so charge the init_mm (happens for pagecache usage).
799 mem = try_get_mem_cgroup_from_mm(mm);
807 VM_BUG_ON(mem_cgroup_is_obsolete(mem));
813 ret = res_counter_charge(&mem->res, PAGE_SIZE, &fail_res);
815 if (!do_swap_account)
817 ret = res_counter_charge(&mem->memsw, PAGE_SIZE,
821 /* mem+swap counter fails */
822 res_counter_uncharge(&mem->res, PAGE_SIZE);
824 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
827 /* mem counter fails */
828 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
831 if (!(gfp_mask & __GFP_WAIT))
834 ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, gfp_mask,
838 * try_to_free_mem_cgroup_pages() might not give us a full
839 * picture of reclaim. Some pages are reclaimed and might be
840 * moved to swap cache or just unmapped from the cgroup.
841 * Check the limit again to see if the reclaim reduced the
842 * current usage of the cgroup before giving up
845 if (mem_cgroup_check_under_limit(mem_over_limit))
850 mutex_lock(&memcg_tasklist);
851 mem_cgroup_out_of_memory(mem_over_limit, gfp_mask);
852 mutex_unlock(&memcg_tasklist);
853 mem_over_limit->last_oom_jiffies = jiffies;
865 * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be
866 * USED state. If already USED, uncharge and return.
869 static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
870 struct page_cgroup *pc,
871 enum charge_type ctype)
873 /* try_charge() can return NULL to *memcg, taking care of it. */
877 lock_page_cgroup(pc);
878 if (unlikely(PageCgroupUsed(pc))) {
879 unlock_page_cgroup(pc);
880 res_counter_uncharge(&mem->res, PAGE_SIZE);
882 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
886 pc->mem_cgroup = mem;
888 pc->flags = pcg_default_flags[ctype];
890 mem_cgroup_charge_statistics(mem, pc, true);
892 unlock_page_cgroup(pc);
896 * mem_cgroup_move_account - move account of the page
897 * @pc: page_cgroup of the page.
898 * @from: mem_cgroup which the page is moved from.
899 * @to: mem_cgroup which the page is moved to. @from != @to.
901 * The caller must confirm following.
902 * - page is not on LRU (isolate_page() is useful.)
904 * returns 0 at success,
905 * returns -EBUSY when lock is busy or "pc" is unstable.
907 * This function does "uncharge" from old cgroup but doesn't do "charge" to
908 * new cgroup. It should be done by a caller.
911 static int mem_cgroup_move_account(struct page_cgroup *pc,
912 struct mem_cgroup *from, struct mem_cgroup *to)
914 struct mem_cgroup_per_zone *from_mz, *to_mz;
918 VM_BUG_ON(from == to);
919 VM_BUG_ON(PageLRU(pc->page));
921 nid = page_cgroup_nid(pc);
922 zid = page_cgroup_zid(pc);
923 from_mz = mem_cgroup_zoneinfo(from, nid, zid);
924 to_mz = mem_cgroup_zoneinfo(to, nid, zid);
926 if (!trylock_page_cgroup(pc))
929 if (!PageCgroupUsed(pc))
932 if (pc->mem_cgroup != from)
936 res_counter_uncharge(&from->res, PAGE_SIZE);
937 mem_cgroup_charge_statistics(from, pc, false);
939 res_counter_uncharge(&from->memsw, PAGE_SIZE);
941 mem_cgroup_charge_statistics(to, pc, true);
945 unlock_page_cgroup(pc);
950 * move charges to its parent.
953 static int mem_cgroup_move_parent(struct page_cgroup *pc,
954 struct mem_cgroup *child,
957 struct page *page = pc->page;
958 struct cgroup *cg = child->css.cgroup;
959 struct cgroup *pcg = cg->parent;
960 struct mem_cgroup *parent;
968 parent = mem_cgroup_from_cont(pcg);
971 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false);
975 if (!get_page_unless_zero(page))
978 ret = isolate_lru_page(page);
983 ret = mem_cgroup_move_account(pc, child, parent);
985 /* drop extra refcnt by try_charge() (move_account increment one) */
986 css_put(&parent->css);
987 putback_lru_page(page);
992 /* uncharge if move fails */
994 res_counter_uncharge(&parent->res, PAGE_SIZE);
996 res_counter_uncharge(&parent->memsw, PAGE_SIZE);
1002 * Charge the memory controller for page usage.
1004 * 0 if the charge was successful
1005 * < 0 if the cgroup is over its limit
1007 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
1008 gfp_t gfp_mask, enum charge_type ctype,
1009 struct mem_cgroup *memcg)
1011 struct mem_cgroup *mem;
1012 struct page_cgroup *pc;
1015 pc = lookup_page_cgroup(page);
1016 /* can happen at boot */
1022 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true);
1026 __mem_cgroup_commit_charge(mem, pc, ctype);
1030 int mem_cgroup_newpage_charge(struct page *page,
1031 struct mm_struct *mm, gfp_t gfp_mask)
1033 if (mem_cgroup_disabled())
1035 if (PageCompound(page))
1038 * If already mapped, we don't have to account.
1039 * If page cache, page->mapping has address_space.
1040 * But page->mapping may have out-of-use anon_vma pointer,
1041 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
1044 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
1048 return mem_cgroup_charge_common(page, mm, gfp_mask,
1049 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
1052 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
1055 if (mem_cgroup_disabled())
1057 if (PageCompound(page))
1060 * Corner case handling. This is called from add_to_page_cache()
1061 * in usual. But some FS (shmem) precharges this page before calling it
1062 * and call add_to_page_cache() with GFP_NOWAIT.
1064 * For GFP_NOWAIT case, the page may be pre-charged before calling
1065 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
1066 * charge twice. (It works but has to pay a bit larger cost.)
1068 if (!(gfp_mask & __GFP_WAIT)) {
1069 struct page_cgroup *pc;
1072 pc = lookup_page_cgroup(page);
1075 lock_page_cgroup(pc);
1076 if (PageCgroupUsed(pc)) {
1077 unlock_page_cgroup(pc);
1080 unlock_page_cgroup(pc);
1086 if (page_is_file_cache(page))
1087 return mem_cgroup_charge_common(page, mm, gfp_mask,
1088 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
1090 return mem_cgroup_charge_common(page, mm, gfp_mask,
1091 MEM_CGROUP_CHARGE_TYPE_SHMEM, NULL);
1095 * While swap-in, try_charge -> commit or cancel, the page is locked.
1096 * And when try_charge() successfully returns, one refcnt to memcg without
1097 * struct page_cgroup is aquired. This refcnt will be cumsumed by
1098 * "commit()" or removed by "cancel()"
1100 int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
1102 gfp_t mask, struct mem_cgroup **ptr)
1104 struct mem_cgroup *mem;
1108 if (mem_cgroup_disabled())
1111 if (!do_swap_account)
1115 * A racing thread's fault, or swapoff, may have already updated
1116 * the pte, and even removed page from swap cache: return success
1117 * to go on to do_swap_page()'s pte_same() test, which should fail.
1119 if (!PageSwapCache(page))
1122 ent.val = page_private(page);
1124 mem = lookup_swap_cgroup(ent);
1127 if (!css_tryget(&mem->css))
1130 ret = __mem_cgroup_try_charge(NULL, mask, ptr, true);
1131 /* drop extra refcnt from tryget */
1137 return __mem_cgroup_try_charge(mm, mask, ptr, true);
1142 int mem_cgroup_cache_charge_swapin(struct page *page,
1143 struct mm_struct *mm, gfp_t mask, bool locked)
1147 if (mem_cgroup_disabled())
1154 * If not locked, the page can be dropped from SwapCache until
1157 if (PageSwapCache(page)) {
1158 struct mem_cgroup *mem = NULL;
1161 ent.val = page_private(page);
1162 if (do_swap_account) {
1163 mem = lookup_swap_cgroup(ent);
1165 if (css_tryget(&mem->css))
1166 mm = NULL; /* charge to recorded */
1168 mem = NULL; /* charge to current */
1171 ret = mem_cgroup_charge_common(page, mm, mask,
1172 MEM_CGROUP_CHARGE_TYPE_SHMEM, mem);
1173 /* drop extra refcnt from tryget */
1177 if (!ret && do_swap_account) {
1178 /* avoid double counting */
1179 mem = swap_cgroup_record(ent, NULL);
1181 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1182 mem_cgroup_put(mem);
1188 /* add this page(page_cgroup) to the LRU we want. */
1189 mem_cgroup_lru_fixup(page);
1195 void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
1197 struct page_cgroup *pc;
1199 if (mem_cgroup_disabled())
1203 pc = lookup_page_cgroup(page);
1204 __mem_cgroup_commit_charge(ptr, pc, MEM_CGROUP_CHARGE_TYPE_MAPPED);
1206 * Now swap is on-memory. This means this page may be
1207 * counted both as mem and swap....double count.
1208 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
1209 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
1210 * may call delete_from_swap_cache() before reach here.
1212 if (do_swap_account && PageSwapCache(page)) {
1213 swp_entry_t ent = {.val = page_private(page)};
1214 struct mem_cgroup *memcg;
1215 memcg = swap_cgroup_record(ent, NULL);
1217 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1218 mem_cgroup_put(memcg);
1222 /* add this page(page_cgroup) to the LRU we want. */
1223 mem_cgroup_lru_fixup(page);
1226 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
1228 if (mem_cgroup_disabled())
1232 res_counter_uncharge(&mem->res, PAGE_SIZE);
1233 if (do_swap_account)
1234 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1240 * uncharge if !page_mapped(page)
1242 static struct mem_cgroup *
1243 __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
1245 struct page_cgroup *pc;
1246 struct mem_cgroup *mem = NULL;
1247 struct mem_cgroup_per_zone *mz;
1249 if (mem_cgroup_disabled())
1252 if (PageSwapCache(page))
1256 * Check if our page_cgroup is valid
1258 pc = lookup_page_cgroup(page);
1259 if (unlikely(!pc || !PageCgroupUsed(pc)))
1262 lock_page_cgroup(pc);
1264 mem = pc->mem_cgroup;
1266 if (!PageCgroupUsed(pc))
1270 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
1271 if (page_mapped(page))
1274 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
1275 if (!PageAnon(page)) { /* Shared memory */
1276 if (page->mapping && !page_is_file_cache(page))
1278 } else if (page_mapped(page)) /* Anon */
1285 res_counter_uncharge(&mem->res, PAGE_SIZE);
1286 if (do_swap_account && (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT))
1287 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1289 mem_cgroup_charge_statistics(mem, pc, false);
1290 ClearPageCgroupUsed(pc);
1292 mz = page_cgroup_zoneinfo(pc);
1293 unlock_page_cgroup(pc);
1295 /* at swapout, this memcg will be accessed to record to swap */
1296 if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
1302 unlock_page_cgroup(pc);
1306 void mem_cgroup_uncharge_page(struct page *page)
1309 if (page_mapped(page))
1311 if (page->mapping && !PageAnon(page))
1313 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
1316 void mem_cgroup_uncharge_cache_page(struct page *page)
1318 VM_BUG_ON(page_mapped(page));
1319 VM_BUG_ON(page->mapping);
1320 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
1324 * called from __delete_from_swap_cache() and drop "page" account.
1325 * memcg information is recorded to swap_cgroup of "ent"
1327 void mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent)
1329 struct mem_cgroup *memcg;
1331 memcg = __mem_cgroup_uncharge_common(page,
1332 MEM_CGROUP_CHARGE_TYPE_SWAPOUT);
1333 /* record memcg information */
1334 if (do_swap_account && memcg) {
1335 swap_cgroup_record(ent, memcg);
1336 mem_cgroup_get(memcg);
1339 css_put(&memcg->css);
1342 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1344 * called from swap_entry_free(). remove record in swap_cgroup and
1345 * uncharge "memsw" account.
1347 void mem_cgroup_uncharge_swap(swp_entry_t ent)
1349 struct mem_cgroup *memcg;
1351 if (!do_swap_account)
1354 memcg = swap_cgroup_record(ent, NULL);
1356 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1357 mem_cgroup_put(memcg);
1363 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
1366 int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr)
1368 struct page_cgroup *pc;
1369 struct mem_cgroup *mem = NULL;
1372 if (mem_cgroup_disabled())
1375 pc = lookup_page_cgroup(page);
1376 lock_page_cgroup(pc);
1377 if (PageCgroupUsed(pc)) {
1378 mem = pc->mem_cgroup;
1381 unlock_page_cgroup(pc);
1384 ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false);
1391 /* remove redundant charge if migration failed*/
1392 void mem_cgroup_end_migration(struct mem_cgroup *mem,
1393 struct page *oldpage, struct page *newpage)
1395 struct page *target, *unused;
1396 struct page_cgroup *pc;
1397 enum charge_type ctype;
1402 /* at migration success, oldpage->mapping is NULL. */
1403 if (oldpage->mapping) {
1411 if (PageAnon(target))
1412 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
1413 else if (page_is_file_cache(target))
1414 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
1416 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
1418 /* unused page is not on radix-tree now. */
1420 __mem_cgroup_uncharge_common(unused, ctype);
1422 pc = lookup_page_cgroup(target);
1424 * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup.
1425 * So, double-counting is effectively avoided.
1427 __mem_cgroup_commit_charge(mem, pc, ctype);
1430 * Both of oldpage and newpage are still under lock_page().
1431 * Then, we don't have to care about race in radix-tree.
1432 * But we have to be careful that this page is unmapped or not.
1434 * There is a case for !page_mapped(). At the start of
1435 * migration, oldpage was mapped. But now, it's zapped.
1436 * But we know *target* page is not freed/reused under us.
1437 * mem_cgroup_uncharge_page() does all necessary checks.
1439 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
1440 mem_cgroup_uncharge_page(target);
1444 * A call to try to shrink memory usage under specified resource controller.
1445 * This is typically used for page reclaiming for shmem for reducing side
1446 * effect of page allocation from shmem, which is used by some mem_cgroup.
1448 int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask)
1450 struct mem_cgroup *mem;
1452 int retry = MEM_CGROUP_RECLAIM_RETRIES;
1454 if (mem_cgroup_disabled())
1459 mem = try_get_mem_cgroup_from_mm(mm);
1464 progress = mem_cgroup_hierarchical_reclaim(mem, gfp_mask, true);
1465 progress += mem_cgroup_check_under_limit(mem);
1466 } while (!progress && --retry);
1474 static DEFINE_MUTEX(set_limit_mutex);
1476 static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
1477 unsigned long long val)
1480 int retry_count = MEM_CGROUP_RECLAIM_RETRIES;
1485 while (retry_count) {
1486 if (signal_pending(current)) {
1491 * Rather than hide all in some function, I do this in
1492 * open coded manner. You see what this really does.
1493 * We have to guarantee mem->res.limit < mem->memsw.limit.
1495 mutex_lock(&set_limit_mutex);
1496 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1497 if (memswlimit < val) {
1499 mutex_unlock(&set_limit_mutex);
1502 ret = res_counter_set_limit(&memcg->res, val);
1503 mutex_unlock(&set_limit_mutex);
1508 progress = mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL,
1510 if (!progress) retry_count--;
1516 int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
1517 unsigned long long val)
1519 int retry_count = MEM_CGROUP_RECLAIM_RETRIES;
1520 u64 memlimit, oldusage, curusage;
1523 if (!do_swap_account)
1526 while (retry_count) {
1527 if (signal_pending(current)) {
1532 * Rather than hide all in some function, I do this in
1533 * open coded manner. You see what this really does.
1534 * We have to guarantee mem->res.limit < mem->memsw.limit.
1536 mutex_lock(&set_limit_mutex);
1537 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1538 if (memlimit > val) {
1540 mutex_unlock(&set_limit_mutex);
1543 ret = res_counter_set_limit(&memcg->memsw, val);
1544 mutex_unlock(&set_limit_mutex);
1549 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
1550 mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL, true);
1551 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
1552 if (curusage >= oldusage)
1559 * This routine traverse page_cgroup in given list and drop them all.
1560 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
1562 static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
1563 int node, int zid, enum lru_list lru)
1566 struct mem_cgroup_per_zone *mz;
1567 struct page_cgroup *pc, *busy;
1568 unsigned long flags, loop;
1569 struct list_head *list;
1572 zone = &NODE_DATA(node)->node_zones[zid];
1573 mz = mem_cgroup_zoneinfo(mem, node, zid);
1574 list = &mz->lists[lru];
1576 loop = MEM_CGROUP_ZSTAT(mz, lru);
1577 /* give some margin against EBUSY etc...*/
1582 spin_lock_irqsave(&zone->lru_lock, flags);
1583 if (list_empty(list)) {
1584 spin_unlock_irqrestore(&zone->lru_lock, flags);
1587 pc = list_entry(list->prev, struct page_cgroup, lru);
1589 list_move(&pc->lru, list);
1591 spin_unlock_irqrestore(&zone->lru_lock, flags);
1594 spin_unlock_irqrestore(&zone->lru_lock, flags);
1596 ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL);
1600 if (ret == -EBUSY || ret == -EINVAL) {
1601 /* found lock contention or "pc" is obsolete. */
1608 if (!ret && !list_empty(list))
1614 * make mem_cgroup's charge to be 0 if there is no task.
1615 * This enables deleting this mem_cgroup.
1617 static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
1620 int node, zid, shrink;
1621 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
1622 struct cgroup *cgrp = mem->css.cgroup;
1627 /* should free all ? */
1631 while (mem->res.usage > 0) {
1633 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
1636 if (signal_pending(current))
1638 /* This is for making all *used* pages to be on LRU. */
1639 lru_add_drain_all();
1641 for_each_node_state(node, N_POSSIBLE) {
1642 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
1645 ret = mem_cgroup_force_empty_list(mem,
1654 /* it seems parent cgroup doesn't have enough mem */
1665 /* returns EBUSY if there is a task or if we come here twice. */
1666 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
1670 /* we call try-to-free pages for make this cgroup empty */
1671 lru_add_drain_all();
1672 /* try to free all pages in this cgroup */
1674 while (nr_retries && mem->res.usage > 0) {
1677 if (signal_pending(current)) {
1681 progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL,
1682 false, get_swappiness(mem));
1685 /* maybe some writeback is necessary */
1686 congestion_wait(WRITE, HZ/10);
1691 /* try move_account...there may be some *locked* pages. */
1698 int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
1700 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
1704 static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
1706 return mem_cgroup_from_cont(cont)->use_hierarchy;
1709 static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
1713 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1714 struct cgroup *parent = cont->parent;
1715 struct mem_cgroup *parent_mem = NULL;
1718 parent_mem = mem_cgroup_from_cont(parent);
1722 * If parent's use_hiearchy is set, we can't make any modifications
1723 * in the child subtrees. If it is unset, then the change can
1724 * occur, provided the current cgroup has no children.
1726 * For the root cgroup, parent_mem is NULL, we allow value to be
1727 * set if there are no children.
1729 if ((!parent_mem || !parent_mem->use_hierarchy) &&
1730 (val == 1 || val == 0)) {
1731 if (list_empty(&cont->children))
1732 mem->use_hierarchy = val;
1742 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
1744 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1748 type = MEMFILE_TYPE(cft->private);
1749 name = MEMFILE_ATTR(cft->private);
1752 val = res_counter_read_u64(&mem->res, name);
1755 if (do_swap_account)
1756 val = res_counter_read_u64(&mem->memsw, name);
1765 * The user of this function is...
1768 static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
1771 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
1773 unsigned long long val;
1776 type = MEMFILE_TYPE(cft->private);
1777 name = MEMFILE_ATTR(cft->private);
1780 /* This function does all necessary parse...reuse it */
1781 ret = res_counter_memparse_write_strategy(buffer, &val);
1785 ret = mem_cgroup_resize_limit(memcg, val);
1787 ret = mem_cgroup_resize_memsw_limit(memcg, val);
1790 ret = -EINVAL; /* should be BUG() ? */
1796 static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
1797 unsigned long long *mem_limit, unsigned long long *memsw_limit)
1799 struct cgroup *cgroup;
1800 unsigned long long min_limit, min_memsw_limit, tmp;
1802 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1803 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1804 cgroup = memcg->css.cgroup;
1805 if (!memcg->use_hierarchy)
1808 while (cgroup->parent) {
1809 cgroup = cgroup->parent;
1810 memcg = mem_cgroup_from_cont(cgroup);
1811 if (!memcg->use_hierarchy)
1813 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
1814 min_limit = min(min_limit, tmp);
1815 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1816 min_memsw_limit = min(min_memsw_limit, tmp);
1819 *mem_limit = min_limit;
1820 *memsw_limit = min_memsw_limit;
1824 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
1826 struct mem_cgroup *mem;
1829 mem = mem_cgroup_from_cont(cont);
1830 type = MEMFILE_TYPE(event);
1831 name = MEMFILE_ATTR(event);
1835 res_counter_reset_max(&mem->res);
1837 res_counter_reset_max(&mem->memsw);
1841 res_counter_reset_failcnt(&mem->res);
1843 res_counter_reset_failcnt(&mem->memsw);
1849 static const struct mem_cgroup_stat_desc {
1852 } mem_cgroup_stat_desc[] = {
1853 [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, },
1854 [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, },
1855 [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, },
1856 [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, },
1859 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
1860 struct cgroup_map_cb *cb)
1862 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
1863 struct mem_cgroup_stat *stat = &mem_cont->stat;
1866 for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) {
1869 val = mem_cgroup_read_stat(stat, i);
1870 val *= mem_cgroup_stat_desc[i].unit;
1871 cb->fill(cb, mem_cgroup_stat_desc[i].msg, val);
1873 /* showing # of active pages */
1875 unsigned long active_anon, inactive_anon;
1876 unsigned long active_file, inactive_file;
1877 unsigned long unevictable;
1879 inactive_anon = mem_cgroup_get_all_zonestat(mem_cont,
1881 active_anon = mem_cgroup_get_all_zonestat(mem_cont,
1883 inactive_file = mem_cgroup_get_all_zonestat(mem_cont,
1885 active_file = mem_cgroup_get_all_zonestat(mem_cont,
1887 unevictable = mem_cgroup_get_all_zonestat(mem_cont,
1890 cb->fill(cb, "active_anon", (active_anon) * PAGE_SIZE);
1891 cb->fill(cb, "inactive_anon", (inactive_anon) * PAGE_SIZE);
1892 cb->fill(cb, "active_file", (active_file) * PAGE_SIZE);
1893 cb->fill(cb, "inactive_file", (inactive_file) * PAGE_SIZE);
1894 cb->fill(cb, "unevictable", unevictable * PAGE_SIZE);
1898 unsigned long long limit, memsw_limit;
1899 memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit);
1900 cb->fill(cb, "hierarchical_memory_limit", limit);
1901 if (do_swap_account)
1902 cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
1905 #ifdef CONFIG_DEBUG_VM
1906 cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL));
1910 struct mem_cgroup_per_zone *mz;
1911 unsigned long recent_rotated[2] = {0, 0};
1912 unsigned long recent_scanned[2] = {0, 0};
1914 for_each_online_node(nid)
1915 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1916 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
1918 recent_rotated[0] +=
1919 mz->reclaim_stat.recent_rotated[0];
1920 recent_rotated[1] +=
1921 mz->reclaim_stat.recent_rotated[1];
1922 recent_scanned[0] +=
1923 mz->reclaim_stat.recent_scanned[0];
1924 recent_scanned[1] +=
1925 mz->reclaim_stat.recent_scanned[1];
1927 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
1928 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
1929 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
1930 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
1937 static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
1939 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
1941 return get_swappiness(memcg);
1944 static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
1947 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
1948 struct mem_cgroup *parent;
1952 if (cgrp->parent == NULL)
1955 parent = mem_cgroup_from_cont(cgrp->parent);
1956 /* If under hierarchy, only empty-root can set this value */
1957 if ((parent->use_hierarchy) ||
1958 (memcg->use_hierarchy && !list_empty(&cgrp->children)))
1961 spin_lock(&memcg->reclaim_param_lock);
1962 memcg->swappiness = val;
1963 spin_unlock(&memcg->reclaim_param_lock);
1969 static struct cftype mem_cgroup_files[] = {
1971 .name = "usage_in_bytes",
1972 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
1973 .read_u64 = mem_cgroup_read,
1976 .name = "max_usage_in_bytes",
1977 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
1978 .trigger = mem_cgroup_reset,
1979 .read_u64 = mem_cgroup_read,
1982 .name = "limit_in_bytes",
1983 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
1984 .write_string = mem_cgroup_write,
1985 .read_u64 = mem_cgroup_read,
1989 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
1990 .trigger = mem_cgroup_reset,
1991 .read_u64 = mem_cgroup_read,
1995 .read_map = mem_control_stat_show,
1998 .name = "force_empty",
1999 .trigger = mem_cgroup_force_empty_write,
2002 .name = "use_hierarchy",
2003 .write_u64 = mem_cgroup_hierarchy_write,
2004 .read_u64 = mem_cgroup_hierarchy_read,
2007 .name = "swappiness",
2008 .read_u64 = mem_cgroup_swappiness_read,
2009 .write_u64 = mem_cgroup_swappiness_write,
2013 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2014 static struct cftype memsw_cgroup_files[] = {
2016 .name = "memsw.usage_in_bytes",
2017 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
2018 .read_u64 = mem_cgroup_read,
2021 .name = "memsw.max_usage_in_bytes",
2022 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
2023 .trigger = mem_cgroup_reset,
2024 .read_u64 = mem_cgroup_read,
2027 .name = "memsw.limit_in_bytes",
2028 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
2029 .write_string = mem_cgroup_write,
2030 .read_u64 = mem_cgroup_read,
2033 .name = "memsw.failcnt",
2034 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
2035 .trigger = mem_cgroup_reset,
2036 .read_u64 = mem_cgroup_read,
2040 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
2042 if (!do_swap_account)
2044 return cgroup_add_files(cont, ss, memsw_cgroup_files,
2045 ARRAY_SIZE(memsw_cgroup_files));
2048 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
2054 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
2056 struct mem_cgroup_per_node *pn;
2057 struct mem_cgroup_per_zone *mz;
2059 int zone, tmp = node;
2061 * This routine is called against possible nodes.
2062 * But it's BUG to call kmalloc() against offline node.
2064 * TODO: this routine can waste much memory for nodes which will
2065 * never be onlined. It's better to use memory hotplug callback
2068 if (!node_state(node, N_NORMAL_MEMORY))
2070 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
2074 mem->info.nodeinfo[node] = pn;
2075 memset(pn, 0, sizeof(*pn));
2077 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
2078 mz = &pn->zoneinfo[zone];
2080 INIT_LIST_HEAD(&mz->lists[l]);
2085 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
2087 kfree(mem->info.nodeinfo[node]);
2090 static int mem_cgroup_size(void)
2092 int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu);
2093 return sizeof(struct mem_cgroup) + cpustat_size;
2096 static struct mem_cgroup *mem_cgroup_alloc(void)
2098 struct mem_cgroup *mem;
2099 int size = mem_cgroup_size();
2101 if (size < PAGE_SIZE)
2102 mem = kmalloc(size, GFP_KERNEL);
2104 mem = vmalloc(size);
2107 memset(mem, 0, size);
2112 * At destroying mem_cgroup, references from swap_cgroup can remain.
2113 * (scanning all at force_empty is too costly...)
2115 * Instead of clearing all references at force_empty, we remember
2116 * the number of reference from swap_cgroup and free mem_cgroup when
2117 * it goes down to 0.
2119 * Removal of cgroup itself succeeds regardless of refs from swap.
2122 static void __mem_cgroup_free(struct mem_cgroup *mem)
2126 for_each_node_state(node, N_POSSIBLE)
2127 free_mem_cgroup_per_zone_info(mem, node);
2129 if (mem_cgroup_size() < PAGE_SIZE)
2135 static void mem_cgroup_get(struct mem_cgroup *mem)
2137 atomic_inc(&mem->refcnt);
2140 static void mem_cgroup_put(struct mem_cgroup *mem)
2142 if (atomic_dec_and_test(&mem->refcnt))
2143 __mem_cgroup_free(mem);
2147 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2148 static void __init enable_swap_cgroup(void)
2150 if (!mem_cgroup_disabled() && really_do_swap_account)
2151 do_swap_account = 1;
2154 static void __init enable_swap_cgroup(void)
2159 static struct cgroup_subsys_state *
2160 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
2162 struct mem_cgroup *mem, *parent;
2165 mem = mem_cgroup_alloc();
2167 return ERR_PTR(-ENOMEM);
2169 for_each_node_state(node, N_POSSIBLE)
2170 if (alloc_mem_cgroup_per_zone_info(mem, node))
2173 if (cont->parent == NULL) {
2174 enable_swap_cgroup();
2177 parent = mem_cgroup_from_cont(cont->parent);
2178 mem->use_hierarchy = parent->use_hierarchy;
2181 if (parent && parent->use_hierarchy) {
2182 res_counter_init(&mem->res, &parent->res);
2183 res_counter_init(&mem->memsw, &parent->memsw);
2185 res_counter_init(&mem->res, NULL);
2186 res_counter_init(&mem->memsw, NULL);
2188 mem->last_scanned_child = NULL;
2189 spin_lock_init(&mem->reclaim_param_lock);
2192 mem->swappiness = get_swappiness(parent);
2193 atomic_set(&mem->refcnt, 1);
2196 __mem_cgroup_free(mem);
2197 return ERR_PTR(-ENOMEM);
2200 static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
2201 struct cgroup *cont)
2203 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2204 mem_cgroup_force_empty(mem, false);
2207 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
2208 struct cgroup *cont)
2210 mem_cgroup_put(mem_cgroup_from_cont(cont));
2213 static int mem_cgroup_populate(struct cgroup_subsys *ss,
2214 struct cgroup *cont)
2218 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
2219 ARRAY_SIZE(mem_cgroup_files));
2222 ret = register_memsw_files(cont, ss);
2226 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
2227 struct cgroup *cont,
2228 struct cgroup *old_cont,
2229 struct task_struct *p)
2231 mutex_lock(&memcg_tasklist);
2233 * FIXME: It's better to move charges of this process from old
2234 * memcg to new memcg. But it's just on TODO-List now.
2236 mutex_unlock(&memcg_tasklist);
2239 struct cgroup_subsys mem_cgroup_subsys = {
2241 .subsys_id = mem_cgroup_subsys_id,
2242 .create = mem_cgroup_create,
2243 .pre_destroy = mem_cgroup_pre_destroy,
2244 .destroy = mem_cgroup_destroy,
2245 .populate = mem_cgroup_populate,
2246 .attach = mem_cgroup_move_task,
2250 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2252 static int __init disable_swap_account(char *s)
2254 really_do_swap_account = 0;
2257 __setup("noswapaccount", disable_swap_account);