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 hierarchy_mutex
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) || !pc->mem_cgroup)
337 * We don't check PCG_USED bit. It's cleared when the "page" is finally
338 * removed from global LRU.
340 mz = page_cgroup_zoneinfo(pc);
341 mem = pc->mem_cgroup;
342 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
343 list_del_init(&pc->lru);
347 void mem_cgroup_del_lru(struct page *page)
349 mem_cgroup_del_lru_list(page, page_lru(page));
352 void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
354 struct mem_cgroup_per_zone *mz;
355 struct page_cgroup *pc;
357 if (mem_cgroup_disabled())
360 pc = lookup_page_cgroup(page);
362 * Used bit is set without atomic ops but after smp_wmb().
363 * For making pc->mem_cgroup visible, insert smp_rmb() here.
366 /* unused page is not rotated. */
367 if (!PageCgroupUsed(pc))
369 mz = page_cgroup_zoneinfo(pc);
370 list_move(&pc->lru, &mz->lists[lru]);
373 void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
375 struct page_cgroup *pc;
376 struct mem_cgroup_per_zone *mz;
378 if (mem_cgroup_disabled())
380 pc = lookup_page_cgroup(page);
382 * Used bit is set without atomic ops but after smp_wmb().
383 * For making pc->mem_cgroup visible, insert smp_rmb() here.
386 if (!PageCgroupUsed(pc))
389 mz = page_cgroup_zoneinfo(pc);
390 MEM_CGROUP_ZSTAT(mz, lru) += 1;
391 list_add(&pc->lru, &mz->lists[lru]);
395 * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to
396 * lru because the page may.be reused after it's fully uncharged (because of
397 * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge
398 * it again. This function is only used to charge SwapCache. It's done under
399 * lock_page and expected that zone->lru_lock is never held.
401 static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page)
404 struct zone *zone = page_zone(page);
405 struct page_cgroup *pc = lookup_page_cgroup(page);
407 spin_lock_irqsave(&zone->lru_lock, flags);
409 * Forget old LRU when this page_cgroup is *not* used. This Used bit
410 * is guarded by lock_page() because the page is SwapCache.
412 if (!PageCgroupUsed(pc))
413 mem_cgroup_del_lru_list(page, page_lru(page));
414 spin_unlock_irqrestore(&zone->lru_lock, flags);
417 static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page)
420 struct zone *zone = page_zone(page);
421 struct page_cgroup *pc = lookup_page_cgroup(page);
423 spin_lock_irqsave(&zone->lru_lock, flags);
424 /* link when the page is linked to LRU but page_cgroup isn't */
425 if (PageLRU(page) && list_empty(&pc->lru))
426 mem_cgroup_add_lru_list(page, page_lru(page));
427 spin_unlock_irqrestore(&zone->lru_lock, flags);
431 void mem_cgroup_move_lists(struct page *page,
432 enum lru_list from, enum lru_list to)
434 if (mem_cgroup_disabled())
436 mem_cgroup_del_lru_list(page, from);
437 mem_cgroup_add_lru_list(page, to);
440 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
445 ret = task->mm && mm_match_cgroup(task->mm, mem);
451 * Calculate mapped_ratio under memory controller. This will be used in
452 * vmscan.c for deteremining we have to reclaim mapped pages.
454 int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem)
459 * usage is recorded in bytes. But, here, we assume the number of
460 * physical pages can be represented by "long" on any arch.
462 total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L;
463 rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
464 return (int)((rss * 100L) / total);
468 * prev_priority control...this will be used in memory reclaim path.
470 int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
474 spin_lock(&mem->reclaim_param_lock);
475 prev_priority = mem->prev_priority;
476 spin_unlock(&mem->reclaim_param_lock);
478 return prev_priority;
481 void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
483 spin_lock(&mem->reclaim_param_lock);
484 if (priority < mem->prev_priority)
485 mem->prev_priority = priority;
486 spin_unlock(&mem->reclaim_param_lock);
489 void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
491 spin_lock(&mem->reclaim_param_lock);
492 mem->prev_priority = priority;
493 spin_unlock(&mem->reclaim_param_lock);
496 static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages)
498 unsigned long active;
499 unsigned long inactive;
501 unsigned long inactive_ratio;
503 inactive = mem_cgroup_get_all_zonestat(memcg, LRU_INACTIVE_ANON);
504 active = mem_cgroup_get_all_zonestat(memcg, LRU_ACTIVE_ANON);
506 gb = (inactive + active) >> (30 - PAGE_SHIFT);
508 inactive_ratio = int_sqrt(10 * gb);
513 present_pages[0] = inactive;
514 present_pages[1] = active;
517 return inactive_ratio;
520 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg)
522 unsigned long active;
523 unsigned long inactive;
524 unsigned long present_pages[2];
525 unsigned long inactive_ratio;
527 inactive_ratio = calc_inactive_ratio(memcg, present_pages);
529 inactive = present_pages[0];
530 active = present_pages[1];
532 if (inactive * inactive_ratio < active)
538 unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg,
542 int nid = zone->zone_pgdat->node_id;
543 int zid = zone_idx(zone);
544 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
546 return MEM_CGROUP_ZSTAT(mz, lru);
549 struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
552 int nid = zone->zone_pgdat->node_id;
553 int zid = zone_idx(zone);
554 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
556 return &mz->reclaim_stat;
559 struct zone_reclaim_stat *
560 mem_cgroup_get_reclaim_stat_from_page(struct page *page)
562 struct page_cgroup *pc;
563 struct mem_cgroup_per_zone *mz;
565 if (mem_cgroup_disabled())
568 pc = lookup_page_cgroup(page);
570 * Used bit is set without atomic ops but after smp_wmb().
571 * For making pc->mem_cgroup visible, insert smp_rmb() here.
574 if (!PageCgroupUsed(pc))
577 mz = page_cgroup_zoneinfo(pc);
581 return &mz->reclaim_stat;
584 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
585 struct list_head *dst,
586 unsigned long *scanned, int order,
587 int mode, struct zone *z,
588 struct mem_cgroup *mem_cont,
589 int active, int file)
591 unsigned long nr_taken = 0;
595 struct list_head *src;
596 struct page_cgroup *pc, *tmp;
597 int nid = z->zone_pgdat->node_id;
598 int zid = zone_idx(z);
599 struct mem_cgroup_per_zone *mz;
600 int lru = LRU_FILE * !!file + !!active;
603 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
604 src = &mz->lists[lru];
607 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
608 if (scan >= nr_to_scan)
612 if (unlikely(!PageCgroupUsed(pc)))
614 if (unlikely(!PageLRU(page)))
618 if (__isolate_lru_page(page, mode, file) == 0) {
619 list_move(&page->lru, dst);
628 #define mem_cgroup_from_res_counter(counter, member) \
629 container_of(counter, struct mem_cgroup, member)
632 * This routine finds the DFS walk successor. This routine should be
633 * called with hierarchy_mutex held
635 static struct mem_cgroup *
636 __mem_cgroup_get_next_node(struct mem_cgroup *curr, struct mem_cgroup *root_mem)
638 struct cgroup *cgroup, *curr_cgroup, *root_cgroup;
640 curr_cgroup = curr->css.cgroup;
641 root_cgroup = root_mem->css.cgroup;
643 if (!list_empty(&curr_cgroup->children)) {
645 * Walk down to children
647 cgroup = list_entry(curr_cgroup->children.next,
648 struct cgroup, sibling);
649 curr = mem_cgroup_from_cont(cgroup);
654 if (curr_cgroup == root_cgroup) {
655 /* caller handles NULL case */
663 if (curr_cgroup->sibling.next != &curr_cgroup->parent->children) {
664 cgroup = list_entry(curr_cgroup->sibling.next, struct cgroup,
666 curr = mem_cgroup_from_cont(cgroup);
671 * Go up to next parent and next parent's sibling if need be
673 curr_cgroup = curr_cgroup->parent;
681 * Visit the first child (need not be the first child as per the ordering
682 * of the cgroup list, since we track last_scanned_child) of @mem and use
683 * that to reclaim free pages from.
685 static struct mem_cgroup *
686 mem_cgroup_get_next_node(struct mem_cgroup *root_mem)
688 struct cgroup *cgroup;
689 struct mem_cgroup *orig, *next;
693 * Scan all children under the mem_cgroup mem
695 mutex_lock(&mem_cgroup_subsys.hierarchy_mutex);
697 orig = root_mem->last_scanned_child;
698 obsolete = mem_cgroup_is_obsolete(orig);
700 if (list_empty(&root_mem->css.cgroup->children)) {
702 * root_mem might have children before and last_scanned_child
703 * may point to one of them. We put it later.
706 VM_BUG_ON(!obsolete);
711 if (!orig || obsolete) {
712 cgroup = list_first_entry(&root_mem->css.cgroup->children,
713 struct cgroup, sibling);
714 next = mem_cgroup_from_cont(cgroup);
716 next = __mem_cgroup_get_next_node(orig, root_mem);
720 mem_cgroup_get(next);
721 root_mem->last_scanned_child = next;
723 mem_cgroup_put(orig);
724 mutex_unlock(&mem_cgroup_subsys.hierarchy_mutex);
725 return (next) ? next : root_mem;
728 static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem)
730 if (do_swap_account) {
731 if (res_counter_check_under_limit(&mem->res) &&
732 res_counter_check_under_limit(&mem->memsw))
735 if (res_counter_check_under_limit(&mem->res))
740 static unsigned int get_swappiness(struct mem_cgroup *memcg)
742 struct cgroup *cgrp = memcg->css.cgroup;
743 unsigned int swappiness;
746 if (cgrp->parent == NULL)
747 return vm_swappiness;
749 spin_lock(&memcg->reclaim_param_lock);
750 swappiness = memcg->swappiness;
751 spin_unlock(&memcg->reclaim_param_lock);
757 * Dance down the hierarchy if needed to reclaim memory. We remember the
758 * last child we reclaimed from, so that we don't end up penalizing
759 * one child extensively based on its position in the children list.
761 * root_mem is the original ancestor that we've been reclaim from.
763 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
764 gfp_t gfp_mask, bool noswap)
766 struct mem_cgroup *next_mem;
770 * Reclaim unconditionally and don't check for return value.
771 * We need to reclaim in the current group and down the tree.
772 * One might think about checking for children before reclaiming,
773 * but there might be left over accounting, even after children
776 ret = try_to_free_mem_cgroup_pages(root_mem, gfp_mask, noswap,
777 get_swappiness(root_mem));
778 if (mem_cgroup_check_under_limit(root_mem))
780 if (!root_mem->use_hierarchy)
783 next_mem = mem_cgroup_get_next_node(root_mem);
785 while (next_mem != root_mem) {
786 if (mem_cgroup_is_obsolete(next_mem)) {
787 next_mem = mem_cgroup_get_next_node(root_mem);
790 ret = try_to_free_mem_cgroup_pages(next_mem, gfp_mask, noswap,
791 get_swappiness(next_mem));
792 if (mem_cgroup_check_under_limit(root_mem))
794 next_mem = mem_cgroup_get_next_node(root_mem);
799 bool mem_cgroup_oom_called(struct task_struct *task)
802 struct mem_cgroup *mem;
803 struct mm_struct *mm;
809 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
810 if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10))
816 * Unlike exported interface, "oom" parameter is added. if oom==true,
817 * oom-killer can be invoked.
819 static int __mem_cgroup_try_charge(struct mm_struct *mm,
820 gfp_t gfp_mask, struct mem_cgroup **memcg,
823 struct mem_cgroup *mem, *mem_over_limit;
824 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
825 struct res_counter *fail_res;
827 if (unlikely(test_thread_flag(TIF_MEMDIE))) {
828 /* Don't account this! */
834 * We always charge the cgroup the mm_struct belongs to.
835 * The mm_struct's mem_cgroup changes on task migration if the
836 * thread group leader migrates. It's possible that mm is not
837 * set, if so charge the init_mm (happens for pagecache usage).
841 mem = try_get_mem_cgroup_from_mm(mm);
849 VM_BUG_ON(mem_cgroup_is_obsolete(mem));
855 ret = res_counter_charge(&mem->res, PAGE_SIZE, &fail_res);
857 if (!do_swap_account)
859 ret = res_counter_charge(&mem->memsw, PAGE_SIZE,
863 /* mem+swap counter fails */
864 res_counter_uncharge(&mem->res, PAGE_SIZE);
866 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
869 /* mem counter fails */
870 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
873 if (!(gfp_mask & __GFP_WAIT))
876 ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, gfp_mask,
880 * try_to_free_mem_cgroup_pages() might not give us a full
881 * picture of reclaim. Some pages are reclaimed and might be
882 * moved to swap cache or just unmapped from the cgroup.
883 * Check the limit again to see if the reclaim reduced the
884 * current usage of the cgroup before giving up
887 if (mem_cgroup_check_under_limit(mem_over_limit))
892 mutex_lock(&memcg_tasklist);
893 mem_cgroup_out_of_memory(mem_over_limit, gfp_mask);
894 mutex_unlock(&memcg_tasklist);
895 mem_over_limit->last_oom_jiffies = jiffies;
906 static struct mem_cgroup *try_get_mem_cgroup_from_swapcache(struct page *page)
908 struct mem_cgroup *mem;
911 if (!PageSwapCache(page))
914 ent.val = page_private(page);
915 mem = lookup_swap_cgroup(ent);
918 if (!css_tryget(&mem->css))
924 * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be
925 * USED state. If already USED, uncharge and return.
928 static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
929 struct page_cgroup *pc,
930 enum charge_type ctype)
932 /* try_charge() can return NULL to *memcg, taking care of it. */
936 lock_page_cgroup(pc);
937 if (unlikely(PageCgroupUsed(pc))) {
938 unlock_page_cgroup(pc);
939 res_counter_uncharge(&mem->res, PAGE_SIZE);
941 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
945 pc->mem_cgroup = mem;
947 pc->flags = pcg_default_flags[ctype];
949 mem_cgroup_charge_statistics(mem, pc, true);
951 unlock_page_cgroup(pc);
955 * mem_cgroup_move_account - move account of the page
956 * @pc: page_cgroup of the page.
957 * @from: mem_cgroup which the page is moved from.
958 * @to: mem_cgroup which the page is moved to. @from != @to.
960 * The caller must confirm following.
961 * - page is not on LRU (isolate_page() is useful.)
963 * returns 0 at success,
964 * returns -EBUSY when lock is busy or "pc" is unstable.
966 * This function does "uncharge" from old cgroup but doesn't do "charge" to
967 * new cgroup. It should be done by a caller.
970 static int mem_cgroup_move_account(struct page_cgroup *pc,
971 struct mem_cgroup *from, struct mem_cgroup *to)
973 struct mem_cgroup_per_zone *from_mz, *to_mz;
977 VM_BUG_ON(from == to);
978 VM_BUG_ON(PageLRU(pc->page));
980 nid = page_cgroup_nid(pc);
981 zid = page_cgroup_zid(pc);
982 from_mz = mem_cgroup_zoneinfo(from, nid, zid);
983 to_mz = mem_cgroup_zoneinfo(to, nid, zid);
985 if (!trylock_page_cgroup(pc))
988 if (!PageCgroupUsed(pc))
991 if (pc->mem_cgroup != from)
994 res_counter_uncharge(&from->res, PAGE_SIZE);
995 mem_cgroup_charge_statistics(from, pc, false);
997 res_counter_uncharge(&from->memsw, PAGE_SIZE);
1001 pc->mem_cgroup = to;
1002 mem_cgroup_charge_statistics(to, pc, true);
1005 unlock_page_cgroup(pc);
1010 * move charges to its parent.
1013 static int mem_cgroup_move_parent(struct page_cgroup *pc,
1014 struct mem_cgroup *child,
1017 struct page *page = pc->page;
1018 struct cgroup *cg = child->css.cgroup;
1019 struct cgroup *pcg = cg->parent;
1020 struct mem_cgroup *parent;
1028 parent = mem_cgroup_from_cont(pcg);
1031 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false);
1035 if (!get_page_unless_zero(page)) {
1040 ret = isolate_lru_page(page);
1045 ret = mem_cgroup_move_account(pc, child, parent);
1047 putback_lru_page(page);
1050 /* drop extra refcnt by try_charge() */
1051 css_put(&parent->css);
1058 /* drop extra refcnt by try_charge() */
1059 css_put(&parent->css);
1060 /* uncharge if move fails */
1061 res_counter_uncharge(&parent->res, PAGE_SIZE);
1062 if (do_swap_account)
1063 res_counter_uncharge(&parent->memsw, PAGE_SIZE);
1068 * Charge the memory controller for page usage.
1070 * 0 if the charge was successful
1071 * < 0 if the cgroup is over its limit
1073 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
1074 gfp_t gfp_mask, enum charge_type ctype,
1075 struct mem_cgroup *memcg)
1077 struct mem_cgroup *mem;
1078 struct page_cgroup *pc;
1081 pc = lookup_page_cgroup(page);
1082 /* can happen at boot */
1088 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true);
1092 __mem_cgroup_commit_charge(mem, pc, ctype);
1096 int mem_cgroup_newpage_charge(struct page *page,
1097 struct mm_struct *mm, gfp_t gfp_mask)
1099 if (mem_cgroup_disabled())
1101 if (PageCompound(page))
1104 * If already mapped, we don't have to account.
1105 * If page cache, page->mapping has address_space.
1106 * But page->mapping may have out-of-use anon_vma pointer,
1107 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
1110 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
1114 return mem_cgroup_charge_common(page, mm, gfp_mask,
1115 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
1118 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
1121 struct mem_cgroup *mem = NULL;
1124 if (mem_cgroup_disabled())
1126 if (PageCompound(page))
1129 * Corner case handling. This is called from add_to_page_cache()
1130 * in usual. But some FS (shmem) precharges this page before calling it
1131 * and call add_to_page_cache() with GFP_NOWAIT.
1133 * For GFP_NOWAIT case, the page may be pre-charged before calling
1134 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
1135 * charge twice. (It works but has to pay a bit larger cost.)
1136 * And when the page is SwapCache, it should take swap information
1137 * into account. This is under lock_page() now.
1139 if (!(gfp_mask & __GFP_WAIT)) {
1140 struct page_cgroup *pc;
1143 pc = lookup_page_cgroup(page);
1146 lock_page_cgroup(pc);
1147 if (PageCgroupUsed(pc)) {
1148 unlock_page_cgroup(pc);
1151 unlock_page_cgroup(pc);
1154 if (do_swap_account && PageSwapCache(page)) {
1155 mem = try_get_mem_cgroup_from_swapcache(page);
1160 /* SwapCache may be still linked to LRU now. */
1161 mem_cgroup_lru_del_before_commit_swapcache(page);
1164 if (unlikely(!mm && !mem))
1167 if (page_is_file_cache(page))
1168 return mem_cgroup_charge_common(page, mm, gfp_mask,
1169 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
1171 ret = mem_cgroup_charge_common(page, mm, gfp_mask,
1172 MEM_CGROUP_CHARGE_TYPE_SHMEM, mem);
1175 if (PageSwapCache(page))
1176 mem_cgroup_lru_add_after_commit_swapcache(page);
1178 if (do_swap_account && !ret && PageSwapCache(page)) {
1179 swp_entry_t ent = {.val = page_private(page)};
1180 /* avoid double counting */
1181 mem = swap_cgroup_record(ent, NULL);
1183 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1184 mem_cgroup_put(mem);
1191 * While swap-in, try_charge -> commit or cancel, the page is locked.
1192 * And when try_charge() successfully returns, one refcnt to memcg without
1193 * struct page_cgroup is aquired. This refcnt will be cumsumed by
1194 * "commit()" or removed by "cancel()"
1196 int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
1198 gfp_t mask, struct mem_cgroup **ptr)
1200 struct mem_cgroup *mem;
1203 if (mem_cgroup_disabled())
1206 if (!do_swap_account)
1209 * A racing thread's fault, or swapoff, may have already updated
1210 * the pte, and even removed page from swap cache: return success
1211 * to go on to do_swap_page()'s pte_same() test, which should fail.
1213 if (!PageSwapCache(page))
1215 mem = try_get_mem_cgroup_from_swapcache(page);
1219 ret = __mem_cgroup_try_charge(NULL, mask, ptr, true);
1220 /* drop extra refcnt from tryget */
1226 return __mem_cgroup_try_charge(mm, mask, ptr, true);
1229 void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
1231 struct page_cgroup *pc;
1233 if (mem_cgroup_disabled())
1237 pc = lookup_page_cgroup(page);
1238 mem_cgroup_lru_del_before_commit_swapcache(page);
1239 __mem_cgroup_commit_charge(ptr, pc, MEM_CGROUP_CHARGE_TYPE_MAPPED);
1240 mem_cgroup_lru_add_after_commit_swapcache(page);
1242 * Now swap is on-memory. This means this page may be
1243 * counted both as mem and swap....double count.
1244 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
1245 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
1246 * may call delete_from_swap_cache() before reach here.
1248 if (do_swap_account && PageSwapCache(page)) {
1249 swp_entry_t ent = {.val = page_private(page)};
1250 struct mem_cgroup *memcg;
1251 memcg = swap_cgroup_record(ent, NULL);
1253 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1254 mem_cgroup_put(memcg);
1258 /* add this page(page_cgroup) to the LRU we want. */
1262 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
1264 if (mem_cgroup_disabled())
1268 res_counter_uncharge(&mem->res, PAGE_SIZE);
1269 if (do_swap_account)
1270 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1276 * uncharge if !page_mapped(page)
1278 static struct mem_cgroup *
1279 __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
1281 struct page_cgroup *pc;
1282 struct mem_cgroup *mem = NULL;
1283 struct mem_cgroup_per_zone *mz;
1285 if (mem_cgroup_disabled())
1288 if (PageSwapCache(page))
1292 * Check if our page_cgroup is valid
1294 pc = lookup_page_cgroup(page);
1295 if (unlikely(!pc || !PageCgroupUsed(pc)))
1298 lock_page_cgroup(pc);
1300 mem = pc->mem_cgroup;
1302 if (!PageCgroupUsed(pc))
1306 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
1307 if (page_mapped(page))
1310 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
1311 if (!PageAnon(page)) { /* Shared memory */
1312 if (page->mapping && !page_is_file_cache(page))
1314 } else if (page_mapped(page)) /* Anon */
1321 res_counter_uncharge(&mem->res, PAGE_SIZE);
1322 if (do_swap_account && (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT))
1323 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1325 mem_cgroup_charge_statistics(mem, pc, false);
1326 ClearPageCgroupUsed(pc);
1328 * pc->mem_cgroup is not cleared here. It will be accessed when it's
1329 * freed from LRU. This is safe because uncharged page is expected not
1330 * to be reused (freed soon). Exception is SwapCache, it's handled by
1331 * special functions.
1334 mz = page_cgroup_zoneinfo(pc);
1335 unlock_page_cgroup(pc);
1337 /* at swapout, this memcg will be accessed to record to swap */
1338 if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
1344 unlock_page_cgroup(pc);
1348 void mem_cgroup_uncharge_page(struct page *page)
1351 if (page_mapped(page))
1353 if (page->mapping && !PageAnon(page))
1355 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
1358 void mem_cgroup_uncharge_cache_page(struct page *page)
1360 VM_BUG_ON(page_mapped(page));
1361 VM_BUG_ON(page->mapping);
1362 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
1366 * called from __delete_from_swap_cache() and drop "page" account.
1367 * memcg information is recorded to swap_cgroup of "ent"
1369 void mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent)
1371 struct mem_cgroup *memcg;
1373 memcg = __mem_cgroup_uncharge_common(page,
1374 MEM_CGROUP_CHARGE_TYPE_SWAPOUT);
1375 /* record memcg information */
1376 if (do_swap_account && memcg) {
1377 swap_cgroup_record(ent, memcg);
1378 mem_cgroup_get(memcg);
1381 css_put(&memcg->css);
1384 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1386 * called from swap_entry_free(). remove record in swap_cgroup and
1387 * uncharge "memsw" account.
1389 void mem_cgroup_uncharge_swap(swp_entry_t ent)
1391 struct mem_cgroup *memcg;
1393 if (!do_swap_account)
1396 memcg = swap_cgroup_record(ent, NULL);
1398 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1399 mem_cgroup_put(memcg);
1405 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
1408 int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr)
1410 struct page_cgroup *pc;
1411 struct mem_cgroup *mem = NULL;
1414 if (mem_cgroup_disabled())
1417 pc = lookup_page_cgroup(page);
1418 lock_page_cgroup(pc);
1419 if (PageCgroupUsed(pc)) {
1420 mem = pc->mem_cgroup;
1423 unlock_page_cgroup(pc);
1426 ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false);
1433 /* remove redundant charge if migration failed*/
1434 void mem_cgroup_end_migration(struct mem_cgroup *mem,
1435 struct page *oldpage, struct page *newpage)
1437 struct page *target, *unused;
1438 struct page_cgroup *pc;
1439 enum charge_type ctype;
1444 /* at migration success, oldpage->mapping is NULL. */
1445 if (oldpage->mapping) {
1453 if (PageAnon(target))
1454 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
1455 else if (page_is_file_cache(target))
1456 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
1458 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
1460 /* unused page is not on radix-tree now. */
1462 __mem_cgroup_uncharge_common(unused, ctype);
1464 pc = lookup_page_cgroup(target);
1466 * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup.
1467 * So, double-counting is effectively avoided.
1469 __mem_cgroup_commit_charge(mem, pc, ctype);
1472 * Both of oldpage and newpage are still under lock_page().
1473 * Then, we don't have to care about race in radix-tree.
1474 * But we have to be careful that this page is unmapped or not.
1476 * There is a case for !page_mapped(). At the start of
1477 * migration, oldpage was mapped. But now, it's zapped.
1478 * But we know *target* page is not freed/reused under us.
1479 * mem_cgroup_uncharge_page() does all necessary checks.
1481 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
1482 mem_cgroup_uncharge_page(target);
1486 * A call to try to shrink memory usage under specified resource controller.
1487 * This is typically used for page reclaiming for shmem for reducing side
1488 * effect of page allocation from shmem, which is used by some mem_cgroup.
1490 int mem_cgroup_shrink_usage(struct page *page,
1491 struct mm_struct *mm,
1494 struct mem_cgroup *mem = NULL;
1496 int retry = MEM_CGROUP_RECLAIM_RETRIES;
1498 if (mem_cgroup_disabled())
1501 mem = try_get_mem_cgroup_from_swapcache(page);
1503 mem = try_get_mem_cgroup_from_mm(mm);
1508 progress = mem_cgroup_hierarchical_reclaim(mem, gfp_mask, true);
1509 progress += mem_cgroup_check_under_limit(mem);
1510 } while (!progress && --retry);
1518 static DEFINE_MUTEX(set_limit_mutex);
1520 static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
1521 unsigned long long val)
1524 int retry_count = MEM_CGROUP_RECLAIM_RETRIES;
1529 while (retry_count) {
1530 if (signal_pending(current)) {
1535 * Rather than hide all in some function, I do this in
1536 * open coded manner. You see what this really does.
1537 * We have to guarantee mem->res.limit < mem->memsw.limit.
1539 mutex_lock(&set_limit_mutex);
1540 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1541 if (memswlimit < val) {
1543 mutex_unlock(&set_limit_mutex);
1546 ret = res_counter_set_limit(&memcg->res, val);
1547 mutex_unlock(&set_limit_mutex);
1552 progress = mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL,
1554 if (!progress) retry_count--;
1560 int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
1561 unsigned long long val)
1563 int retry_count = MEM_CGROUP_RECLAIM_RETRIES;
1564 u64 memlimit, oldusage, curusage;
1567 if (!do_swap_account)
1570 while (retry_count) {
1571 if (signal_pending(current)) {
1576 * Rather than hide all in some function, I do this in
1577 * open coded manner. You see what this really does.
1578 * We have to guarantee mem->res.limit < mem->memsw.limit.
1580 mutex_lock(&set_limit_mutex);
1581 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1582 if (memlimit > val) {
1584 mutex_unlock(&set_limit_mutex);
1587 ret = res_counter_set_limit(&memcg->memsw, val);
1588 mutex_unlock(&set_limit_mutex);
1593 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
1594 mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL, true);
1595 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
1596 if (curusage >= oldusage)
1603 * This routine traverse page_cgroup in given list and drop them all.
1604 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
1606 static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
1607 int node, int zid, enum lru_list lru)
1610 struct mem_cgroup_per_zone *mz;
1611 struct page_cgroup *pc, *busy;
1612 unsigned long flags, loop;
1613 struct list_head *list;
1616 zone = &NODE_DATA(node)->node_zones[zid];
1617 mz = mem_cgroup_zoneinfo(mem, node, zid);
1618 list = &mz->lists[lru];
1620 loop = MEM_CGROUP_ZSTAT(mz, lru);
1621 /* give some margin against EBUSY etc...*/
1626 spin_lock_irqsave(&zone->lru_lock, flags);
1627 if (list_empty(list)) {
1628 spin_unlock_irqrestore(&zone->lru_lock, flags);
1631 pc = list_entry(list->prev, struct page_cgroup, lru);
1633 list_move(&pc->lru, list);
1635 spin_unlock_irqrestore(&zone->lru_lock, flags);
1638 spin_unlock_irqrestore(&zone->lru_lock, flags);
1640 ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL);
1644 if (ret == -EBUSY || ret == -EINVAL) {
1645 /* found lock contention or "pc" is obsolete. */
1652 if (!ret && !list_empty(list))
1658 * make mem_cgroup's charge to be 0 if there is no task.
1659 * This enables deleting this mem_cgroup.
1661 static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
1664 int node, zid, shrink;
1665 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
1666 struct cgroup *cgrp = mem->css.cgroup;
1671 /* should free all ? */
1675 while (mem->res.usage > 0) {
1677 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
1680 if (signal_pending(current))
1682 /* This is for making all *used* pages to be on LRU. */
1683 lru_add_drain_all();
1685 for_each_node_state(node, N_POSSIBLE) {
1686 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
1689 ret = mem_cgroup_force_empty_list(mem,
1698 /* it seems parent cgroup doesn't have enough mem */
1709 /* returns EBUSY if there is a task or if we come here twice. */
1710 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
1714 /* we call try-to-free pages for make this cgroup empty */
1715 lru_add_drain_all();
1716 /* try to free all pages in this cgroup */
1718 while (nr_retries && mem->res.usage > 0) {
1721 if (signal_pending(current)) {
1725 progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL,
1726 false, get_swappiness(mem));
1729 /* maybe some writeback is necessary */
1730 congestion_wait(WRITE, HZ/10);
1735 /* try move_account...there may be some *locked* pages. */
1742 int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
1744 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
1748 static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
1750 return mem_cgroup_from_cont(cont)->use_hierarchy;
1753 static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
1757 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1758 struct cgroup *parent = cont->parent;
1759 struct mem_cgroup *parent_mem = NULL;
1762 parent_mem = mem_cgroup_from_cont(parent);
1766 * If parent's use_hiearchy is set, we can't make any modifications
1767 * in the child subtrees. If it is unset, then the change can
1768 * occur, provided the current cgroup has no children.
1770 * For the root cgroup, parent_mem is NULL, we allow value to be
1771 * set if there are no children.
1773 if ((!parent_mem || !parent_mem->use_hierarchy) &&
1774 (val == 1 || val == 0)) {
1775 if (list_empty(&cont->children))
1776 mem->use_hierarchy = val;
1786 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
1788 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1792 type = MEMFILE_TYPE(cft->private);
1793 name = MEMFILE_ATTR(cft->private);
1796 val = res_counter_read_u64(&mem->res, name);
1799 if (do_swap_account)
1800 val = res_counter_read_u64(&mem->memsw, name);
1809 * The user of this function is...
1812 static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
1815 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
1817 unsigned long long val;
1820 type = MEMFILE_TYPE(cft->private);
1821 name = MEMFILE_ATTR(cft->private);
1824 /* This function does all necessary parse...reuse it */
1825 ret = res_counter_memparse_write_strategy(buffer, &val);
1829 ret = mem_cgroup_resize_limit(memcg, val);
1831 ret = mem_cgroup_resize_memsw_limit(memcg, val);
1834 ret = -EINVAL; /* should be BUG() ? */
1840 static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
1841 unsigned long long *mem_limit, unsigned long long *memsw_limit)
1843 struct cgroup *cgroup;
1844 unsigned long long min_limit, min_memsw_limit, tmp;
1846 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1847 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1848 cgroup = memcg->css.cgroup;
1849 if (!memcg->use_hierarchy)
1852 while (cgroup->parent) {
1853 cgroup = cgroup->parent;
1854 memcg = mem_cgroup_from_cont(cgroup);
1855 if (!memcg->use_hierarchy)
1857 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
1858 min_limit = min(min_limit, tmp);
1859 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1860 min_memsw_limit = min(min_memsw_limit, tmp);
1863 *mem_limit = min_limit;
1864 *memsw_limit = min_memsw_limit;
1868 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
1870 struct mem_cgroup *mem;
1873 mem = mem_cgroup_from_cont(cont);
1874 type = MEMFILE_TYPE(event);
1875 name = MEMFILE_ATTR(event);
1879 res_counter_reset_max(&mem->res);
1881 res_counter_reset_max(&mem->memsw);
1885 res_counter_reset_failcnt(&mem->res);
1887 res_counter_reset_failcnt(&mem->memsw);
1893 static const struct mem_cgroup_stat_desc {
1896 } mem_cgroup_stat_desc[] = {
1897 [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, },
1898 [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, },
1899 [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, },
1900 [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, },
1903 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
1904 struct cgroup_map_cb *cb)
1906 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
1907 struct mem_cgroup_stat *stat = &mem_cont->stat;
1910 for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) {
1913 val = mem_cgroup_read_stat(stat, i);
1914 val *= mem_cgroup_stat_desc[i].unit;
1915 cb->fill(cb, mem_cgroup_stat_desc[i].msg, val);
1917 /* showing # of active pages */
1919 unsigned long active_anon, inactive_anon;
1920 unsigned long active_file, inactive_file;
1921 unsigned long unevictable;
1923 inactive_anon = mem_cgroup_get_all_zonestat(mem_cont,
1925 active_anon = mem_cgroup_get_all_zonestat(mem_cont,
1927 inactive_file = mem_cgroup_get_all_zonestat(mem_cont,
1929 active_file = mem_cgroup_get_all_zonestat(mem_cont,
1931 unevictable = mem_cgroup_get_all_zonestat(mem_cont,
1934 cb->fill(cb, "active_anon", (active_anon) * PAGE_SIZE);
1935 cb->fill(cb, "inactive_anon", (inactive_anon) * PAGE_SIZE);
1936 cb->fill(cb, "active_file", (active_file) * PAGE_SIZE);
1937 cb->fill(cb, "inactive_file", (inactive_file) * PAGE_SIZE);
1938 cb->fill(cb, "unevictable", unevictable * PAGE_SIZE);
1942 unsigned long long limit, memsw_limit;
1943 memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit);
1944 cb->fill(cb, "hierarchical_memory_limit", limit);
1945 if (do_swap_account)
1946 cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
1949 #ifdef CONFIG_DEBUG_VM
1950 cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL));
1954 struct mem_cgroup_per_zone *mz;
1955 unsigned long recent_rotated[2] = {0, 0};
1956 unsigned long recent_scanned[2] = {0, 0};
1958 for_each_online_node(nid)
1959 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1960 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
1962 recent_rotated[0] +=
1963 mz->reclaim_stat.recent_rotated[0];
1964 recent_rotated[1] +=
1965 mz->reclaim_stat.recent_rotated[1];
1966 recent_scanned[0] +=
1967 mz->reclaim_stat.recent_scanned[0];
1968 recent_scanned[1] +=
1969 mz->reclaim_stat.recent_scanned[1];
1971 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
1972 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
1973 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
1974 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
1981 static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
1983 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
1985 return get_swappiness(memcg);
1988 static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
1991 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
1992 struct mem_cgroup *parent;
1996 if (cgrp->parent == NULL)
1999 parent = mem_cgroup_from_cont(cgrp->parent);
2000 /* If under hierarchy, only empty-root can set this value */
2001 if ((parent->use_hierarchy) ||
2002 (memcg->use_hierarchy && !list_empty(&cgrp->children)))
2005 spin_lock(&memcg->reclaim_param_lock);
2006 memcg->swappiness = val;
2007 spin_unlock(&memcg->reclaim_param_lock);
2013 static struct cftype mem_cgroup_files[] = {
2015 .name = "usage_in_bytes",
2016 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
2017 .read_u64 = mem_cgroup_read,
2020 .name = "max_usage_in_bytes",
2021 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
2022 .trigger = mem_cgroup_reset,
2023 .read_u64 = mem_cgroup_read,
2026 .name = "limit_in_bytes",
2027 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
2028 .write_string = mem_cgroup_write,
2029 .read_u64 = mem_cgroup_read,
2033 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
2034 .trigger = mem_cgroup_reset,
2035 .read_u64 = mem_cgroup_read,
2039 .read_map = mem_control_stat_show,
2042 .name = "force_empty",
2043 .trigger = mem_cgroup_force_empty_write,
2046 .name = "use_hierarchy",
2047 .write_u64 = mem_cgroup_hierarchy_write,
2048 .read_u64 = mem_cgroup_hierarchy_read,
2051 .name = "swappiness",
2052 .read_u64 = mem_cgroup_swappiness_read,
2053 .write_u64 = mem_cgroup_swappiness_write,
2057 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2058 static struct cftype memsw_cgroup_files[] = {
2060 .name = "memsw.usage_in_bytes",
2061 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
2062 .read_u64 = mem_cgroup_read,
2065 .name = "memsw.max_usage_in_bytes",
2066 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
2067 .trigger = mem_cgroup_reset,
2068 .read_u64 = mem_cgroup_read,
2071 .name = "memsw.limit_in_bytes",
2072 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
2073 .write_string = mem_cgroup_write,
2074 .read_u64 = mem_cgroup_read,
2077 .name = "memsw.failcnt",
2078 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
2079 .trigger = mem_cgroup_reset,
2080 .read_u64 = mem_cgroup_read,
2084 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
2086 if (!do_swap_account)
2088 return cgroup_add_files(cont, ss, memsw_cgroup_files,
2089 ARRAY_SIZE(memsw_cgroup_files));
2092 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
2098 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
2100 struct mem_cgroup_per_node *pn;
2101 struct mem_cgroup_per_zone *mz;
2103 int zone, tmp = node;
2105 * This routine is called against possible nodes.
2106 * But it's BUG to call kmalloc() against offline node.
2108 * TODO: this routine can waste much memory for nodes which will
2109 * never be onlined. It's better to use memory hotplug callback
2112 if (!node_state(node, N_NORMAL_MEMORY))
2114 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
2118 mem->info.nodeinfo[node] = pn;
2119 memset(pn, 0, sizeof(*pn));
2121 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
2122 mz = &pn->zoneinfo[zone];
2124 INIT_LIST_HEAD(&mz->lists[l]);
2129 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
2131 kfree(mem->info.nodeinfo[node]);
2134 static int mem_cgroup_size(void)
2136 int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu);
2137 return sizeof(struct mem_cgroup) + cpustat_size;
2140 static struct mem_cgroup *mem_cgroup_alloc(void)
2142 struct mem_cgroup *mem;
2143 int size = mem_cgroup_size();
2145 if (size < PAGE_SIZE)
2146 mem = kmalloc(size, GFP_KERNEL);
2148 mem = vmalloc(size);
2151 memset(mem, 0, size);
2156 * At destroying mem_cgroup, references from swap_cgroup can remain.
2157 * (scanning all at force_empty is too costly...)
2159 * Instead of clearing all references at force_empty, we remember
2160 * the number of reference from swap_cgroup and free mem_cgroup when
2161 * it goes down to 0.
2163 * Removal of cgroup itself succeeds regardless of refs from swap.
2166 static void __mem_cgroup_free(struct mem_cgroup *mem)
2170 for_each_node_state(node, N_POSSIBLE)
2171 free_mem_cgroup_per_zone_info(mem, node);
2173 if (mem_cgroup_size() < PAGE_SIZE)
2179 static void mem_cgroup_get(struct mem_cgroup *mem)
2181 atomic_inc(&mem->refcnt);
2184 static void mem_cgroup_put(struct mem_cgroup *mem)
2186 if (atomic_dec_and_test(&mem->refcnt))
2187 __mem_cgroup_free(mem);
2191 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2192 static void __init enable_swap_cgroup(void)
2194 if (!mem_cgroup_disabled() && really_do_swap_account)
2195 do_swap_account = 1;
2198 static void __init enable_swap_cgroup(void)
2203 static struct cgroup_subsys_state *
2204 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
2206 struct mem_cgroup *mem, *parent;
2209 mem = mem_cgroup_alloc();
2211 return ERR_PTR(-ENOMEM);
2213 for_each_node_state(node, N_POSSIBLE)
2214 if (alloc_mem_cgroup_per_zone_info(mem, node))
2217 if (cont->parent == NULL) {
2218 enable_swap_cgroup();
2221 parent = mem_cgroup_from_cont(cont->parent);
2222 mem->use_hierarchy = parent->use_hierarchy;
2225 if (parent && parent->use_hierarchy) {
2226 res_counter_init(&mem->res, &parent->res);
2227 res_counter_init(&mem->memsw, &parent->memsw);
2229 res_counter_init(&mem->res, NULL);
2230 res_counter_init(&mem->memsw, NULL);
2232 mem->last_scanned_child = NULL;
2233 spin_lock_init(&mem->reclaim_param_lock);
2236 mem->swappiness = get_swappiness(parent);
2237 atomic_set(&mem->refcnt, 1);
2240 __mem_cgroup_free(mem);
2241 return ERR_PTR(-ENOMEM);
2244 static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
2245 struct cgroup *cont)
2247 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2248 mem_cgroup_force_empty(mem, false);
2251 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
2252 struct cgroup *cont)
2254 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2255 struct mem_cgroup *last_scanned_child = mem->last_scanned_child;
2257 if (last_scanned_child) {
2258 VM_BUG_ON(!mem_cgroup_is_obsolete(last_scanned_child));
2259 mem_cgroup_put(last_scanned_child);
2261 mem_cgroup_put(mem);
2264 static int mem_cgroup_populate(struct cgroup_subsys *ss,
2265 struct cgroup *cont)
2269 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
2270 ARRAY_SIZE(mem_cgroup_files));
2273 ret = register_memsw_files(cont, ss);
2277 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
2278 struct cgroup *cont,
2279 struct cgroup *old_cont,
2280 struct task_struct *p)
2282 mutex_lock(&memcg_tasklist);
2284 * FIXME: It's better to move charges of this process from old
2285 * memcg to new memcg. But it's just on TODO-List now.
2287 mutex_unlock(&memcg_tasklist);
2290 struct cgroup_subsys mem_cgroup_subsys = {
2292 .subsys_id = mem_cgroup_subsys_id,
2293 .create = mem_cgroup_create,
2294 .pre_destroy = mem_cgroup_pre_destroy,
2295 .destroy = mem_cgroup_destroy,
2296 .populate = mem_cgroup_populate,
2297 .attach = mem_cgroup_move_task,
2301 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2303 static int __init disable_swap_account(char *s)
2305 really_do_swap_account = 0;
2308 __setup("noswapaccount", disable_swap_account);