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/limits.h>
31 #include <linux/mutex.h>
32 #include <linux/rbtree.h>
33 #include <linux/slab.h>
34 #include <linux/swap.h>
35 #include <linux/spinlock.h>
37 #include <linux/seq_file.h>
38 #include <linux/vmalloc.h>
39 #include <linux/mm_inline.h>
40 #include <linux/page_cgroup.h>
43 #include <asm/uaccess.h>
45 struct cgroup_subsys mem_cgroup_subsys __read_mostly;
46 #define MEM_CGROUP_RECLAIM_RETRIES 5
47 struct mem_cgroup *root_mem_cgroup __read_mostly;
49 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
50 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
51 int do_swap_account __read_mostly;
52 static int really_do_swap_account __initdata = 1; /* for remember boot option*/
54 #define do_swap_account (0)
57 static DEFINE_MUTEX(memcg_tasklist); /* can be hold under cgroup_mutex */
58 #define SOFTLIMIT_EVENTS_THRESH (1000)
61 * Statistics for memory cgroup.
63 enum mem_cgroup_stat_index {
65 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
67 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
68 MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
69 MEM_CGROUP_STAT_MAPPED_FILE, /* # of pages charged as file rss */
70 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
71 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
72 MEM_CGROUP_STAT_EVENTS, /* sum of pagein + pageout for internal use */
74 MEM_CGROUP_STAT_NSTATS,
77 struct mem_cgroup_stat_cpu {
78 s64 count[MEM_CGROUP_STAT_NSTATS];
79 } ____cacheline_aligned_in_smp;
81 struct mem_cgroup_stat {
82 struct mem_cgroup_stat_cpu cpustat[0];
86 __mem_cgroup_stat_reset_safe(struct mem_cgroup_stat_cpu *stat,
87 enum mem_cgroup_stat_index idx)
93 __mem_cgroup_stat_read_local(struct mem_cgroup_stat_cpu *stat,
94 enum mem_cgroup_stat_index idx)
96 return stat->count[idx];
100 * For accounting under irq disable, no need for increment preempt count.
102 static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat,
103 enum mem_cgroup_stat_index idx, int val)
105 stat->count[idx] += val;
108 static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat,
109 enum mem_cgroup_stat_index idx)
113 for_each_possible_cpu(cpu)
114 ret += stat->cpustat[cpu].count[idx];
118 static s64 mem_cgroup_local_usage(struct mem_cgroup_stat *stat)
122 ret = mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_CACHE);
123 ret += mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_RSS);
128 * per-zone information in memory controller.
130 struct mem_cgroup_per_zone {
132 * spin_lock to protect the per cgroup LRU
134 struct list_head lists[NR_LRU_LISTS];
135 unsigned long count[NR_LRU_LISTS];
137 struct zone_reclaim_stat reclaim_stat;
138 struct rb_node tree_node; /* RB tree node */
139 unsigned long long usage_in_excess;/* Set to the value by which */
140 /* the soft limit is exceeded*/
143 /* Macro for accessing counter */
144 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
146 struct mem_cgroup_per_node {
147 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
150 struct mem_cgroup_lru_info {
151 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
155 * Cgroups above their limits are maintained in a RB-Tree, independent of
156 * their hierarchy representation
159 struct mem_cgroup_tree_per_zone {
160 struct rb_root rb_root;
164 struct mem_cgroup_tree_per_node {
165 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
168 struct mem_cgroup_tree {
169 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
172 static struct mem_cgroup_tree soft_limit_tree __read_mostly;
175 * The memory controller data structure. The memory controller controls both
176 * page cache and RSS per cgroup. We would eventually like to provide
177 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
178 * to help the administrator determine what knobs to tune.
180 * TODO: Add a water mark for the memory controller. Reclaim will begin when
181 * we hit the water mark. May be even add a low water mark, such that
182 * no reclaim occurs from a cgroup at it's low water mark, this is
183 * a feature that will be implemented much later in the future.
186 struct cgroup_subsys_state css;
188 * the counter to account for memory usage
190 struct res_counter res;
192 * the counter to account for mem+swap usage.
194 struct res_counter memsw;
196 * Per cgroup active and inactive list, similar to the
197 * per zone LRU lists.
199 struct mem_cgroup_lru_info info;
202 protect against reclaim related member.
204 spinlock_t reclaim_param_lock;
206 int prev_priority; /* for recording reclaim priority */
209 * While reclaiming in a hiearchy, we cache the last child we
212 int last_scanned_child;
214 * Should the accounting and control be hierarchical, per subtree?
217 unsigned long last_oom_jiffies;
220 unsigned int swappiness;
222 /* set when res.limit == memsw.limit */
223 bool memsw_is_minimum;
226 * statistics. This must be placed at the end of memcg.
228 struct mem_cgroup_stat stat;
232 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
233 MEM_CGROUP_CHARGE_TYPE_MAPPED,
234 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
235 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
236 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
237 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
241 /* only for here (for easy reading.) */
242 #define PCGF_CACHE (1UL << PCG_CACHE)
243 #define PCGF_USED (1UL << PCG_USED)
244 #define PCGF_LOCK (1UL << PCG_LOCK)
245 /* Not used, but added here for completeness */
246 #define PCGF_ACCT (1UL << PCG_ACCT)
248 /* for encoding cft->private value on file */
251 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
252 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
253 #define MEMFILE_ATTR(val) ((val) & 0xffff)
256 * Reclaim flags for mem_cgroup_hierarchical_reclaim
258 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
259 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
260 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
261 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
263 static void mem_cgroup_get(struct mem_cgroup *mem);
264 static void mem_cgroup_put(struct mem_cgroup *mem);
265 static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem);
267 static struct mem_cgroup_per_zone *
268 mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
270 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
273 static struct mem_cgroup_per_zone *
274 page_cgroup_zoneinfo(struct page_cgroup *pc)
276 struct mem_cgroup *mem = pc->mem_cgroup;
277 int nid = page_cgroup_nid(pc);
278 int zid = page_cgroup_zid(pc);
283 return mem_cgroup_zoneinfo(mem, nid, zid);
286 static struct mem_cgroup_tree_per_zone *
287 soft_limit_tree_node_zone(int nid, int zid)
289 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
292 static struct mem_cgroup_tree_per_zone *
293 soft_limit_tree_from_page(struct page *page)
295 int nid = page_to_nid(page);
296 int zid = page_zonenum(page);
298 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
302 mem_cgroup_insert_exceeded(struct mem_cgroup *mem,
303 struct mem_cgroup_per_zone *mz,
304 struct mem_cgroup_tree_per_zone *mctz)
306 struct rb_node **p = &mctz->rb_root.rb_node;
307 struct rb_node *parent = NULL;
308 struct mem_cgroup_per_zone *mz_node;
313 mz->usage_in_excess = res_counter_soft_limit_excess(&mem->res);
314 spin_lock(&mctz->lock);
317 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
319 if (mz->usage_in_excess < mz_node->usage_in_excess)
322 * We can't avoid mem cgroups that are over their soft
323 * limit by the same amount
325 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
328 rb_link_node(&mz->tree_node, parent, p);
329 rb_insert_color(&mz->tree_node, &mctz->rb_root);
331 spin_unlock(&mctz->lock);
335 mem_cgroup_remove_exceeded(struct mem_cgroup *mem,
336 struct mem_cgroup_per_zone *mz,
337 struct mem_cgroup_tree_per_zone *mctz)
339 spin_lock(&mctz->lock);
340 rb_erase(&mz->tree_node, &mctz->rb_root);
342 spin_unlock(&mctz->lock);
345 static bool mem_cgroup_soft_limit_check(struct mem_cgroup *mem)
350 struct mem_cgroup_stat_cpu *cpustat;
353 cpustat = &mem->stat.cpustat[cpu];
354 val = __mem_cgroup_stat_read_local(cpustat, MEM_CGROUP_STAT_EVENTS);
355 if (unlikely(val > SOFTLIMIT_EVENTS_THRESH)) {
356 __mem_cgroup_stat_reset_safe(cpustat, MEM_CGROUP_STAT_EVENTS);
363 static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page)
365 unsigned long long prev_usage_in_excess, new_usage_in_excess;
366 bool updated_tree = false;
367 struct mem_cgroup_per_zone *mz;
368 struct mem_cgroup_tree_per_zone *mctz;
370 mz = mem_cgroup_zoneinfo(mem, page_to_nid(page), page_zonenum(page));
371 mctz = soft_limit_tree_from_page(page);
374 * We do updates in lazy mode, mem's are removed
375 * lazily from the per-zone, per-node rb tree
377 prev_usage_in_excess = mz->usage_in_excess;
379 new_usage_in_excess = res_counter_soft_limit_excess(&mem->res);
380 if (prev_usage_in_excess) {
381 mem_cgroup_remove_exceeded(mem, mz, mctz);
384 if (!new_usage_in_excess)
386 mem_cgroup_insert_exceeded(mem, mz, mctz);
390 spin_lock(&mctz->lock);
391 mz->usage_in_excess = new_usage_in_excess;
392 spin_unlock(&mctz->lock);
396 static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem)
399 struct mem_cgroup_per_zone *mz;
400 struct mem_cgroup_tree_per_zone *mctz;
402 for_each_node_state(node, N_POSSIBLE) {
403 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
404 mz = mem_cgroup_zoneinfo(mem, node, zone);
405 mctz = soft_limit_tree_node_zone(node, zone);
406 mem_cgroup_remove_exceeded(mem, mz, mctz);
411 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
412 struct page_cgroup *pc,
415 int val = (charge)? 1 : -1;
416 struct mem_cgroup_stat *stat = &mem->stat;
417 struct mem_cgroup_stat_cpu *cpustat;
420 cpustat = &stat->cpustat[cpu];
421 if (PageCgroupCache(pc))
422 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val);
424 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val);
427 __mem_cgroup_stat_add_safe(cpustat,
428 MEM_CGROUP_STAT_PGPGIN_COUNT, 1);
430 __mem_cgroup_stat_add_safe(cpustat,
431 MEM_CGROUP_STAT_PGPGOUT_COUNT, 1);
432 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_EVENTS, 1);
436 static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem,
440 struct mem_cgroup_per_zone *mz;
443 for_each_online_node(nid)
444 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
445 mz = mem_cgroup_zoneinfo(mem, nid, zid);
446 total += MEM_CGROUP_ZSTAT(mz, idx);
451 static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
453 return container_of(cgroup_subsys_state(cont,
454 mem_cgroup_subsys_id), struct mem_cgroup,
458 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
461 * mm_update_next_owner() may clear mm->owner to NULL
462 * if it races with swapoff, page migration, etc.
463 * So this can be called with p == NULL.
468 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
469 struct mem_cgroup, css);
472 static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
474 struct mem_cgroup *mem = NULL;
479 * Because we have no locks, mm->owner's may be being moved to other
480 * cgroup. We use css_tryget() here even if this looks
481 * pessimistic (rather than adding locks here).
485 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
488 } while (!css_tryget(&mem->css));
494 * Call callback function against all cgroup under hierarchy tree.
496 static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data,
497 int (*func)(struct mem_cgroup *, void *))
499 int found, ret, nextid;
500 struct cgroup_subsys_state *css;
501 struct mem_cgroup *mem;
503 if (!root->use_hierarchy)
504 return (*func)(root, data);
512 css = css_get_next(&mem_cgroup_subsys, nextid, &root->css,
514 if (css && css_tryget(css))
515 mem = container_of(css, struct mem_cgroup, css);
519 ret = (*func)(mem, data);
523 } while (!ret && css);
528 static inline bool mem_cgroup_is_root(struct mem_cgroup *mem)
530 return (mem == root_mem_cgroup);
534 * Following LRU functions are allowed to be used without PCG_LOCK.
535 * Operations are called by routine of global LRU independently from memcg.
536 * What we have to take care of here is validness of pc->mem_cgroup.
538 * Changes to pc->mem_cgroup happens when
541 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
542 * It is added to LRU before charge.
543 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
544 * When moving account, the page is not on LRU. It's isolated.
547 void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
549 struct page_cgroup *pc;
550 struct mem_cgroup_per_zone *mz;
552 if (mem_cgroup_disabled())
554 pc = lookup_page_cgroup(page);
555 /* can happen while we handle swapcache. */
556 if (!TestClearPageCgroupAcctLRU(pc))
558 VM_BUG_ON(!pc->mem_cgroup);
560 * We don't check PCG_USED bit. It's cleared when the "page" is finally
561 * removed from global LRU.
563 mz = page_cgroup_zoneinfo(pc);
564 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
565 if (mem_cgroup_is_root(pc->mem_cgroup))
567 VM_BUG_ON(list_empty(&pc->lru));
568 list_del_init(&pc->lru);
572 void mem_cgroup_del_lru(struct page *page)
574 mem_cgroup_del_lru_list(page, page_lru(page));
577 void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
579 struct mem_cgroup_per_zone *mz;
580 struct page_cgroup *pc;
582 if (mem_cgroup_disabled())
585 pc = lookup_page_cgroup(page);
587 * Used bit is set without atomic ops but after smp_wmb().
588 * For making pc->mem_cgroup visible, insert smp_rmb() here.
591 /* unused or root page is not rotated. */
592 if (!PageCgroupUsed(pc) || mem_cgroup_is_root(pc->mem_cgroup))
594 mz = page_cgroup_zoneinfo(pc);
595 list_move(&pc->lru, &mz->lists[lru]);
598 void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
600 struct page_cgroup *pc;
601 struct mem_cgroup_per_zone *mz;
603 if (mem_cgroup_disabled())
605 pc = lookup_page_cgroup(page);
606 VM_BUG_ON(PageCgroupAcctLRU(pc));
608 * Used bit is set without atomic ops but after smp_wmb().
609 * For making pc->mem_cgroup visible, insert smp_rmb() here.
612 if (!PageCgroupUsed(pc))
615 mz = page_cgroup_zoneinfo(pc);
616 MEM_CGROUP_ZSTAT(mz, lru) += 1;
617 SetPageCgroupAcctLRU(pc);
618 if (mem_cgroup_is_root(pc->mem_cgroup))
620 list_add(&pc->lru, &mz->lists[lru]);
624 * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to
625 * lru because the page may.be reused after it's fully uncharged (because of
626 * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge
627 * it again. This function is only used to charge SwapCache. It's done under
628 * lock_page and expected that zone->lru_lock is never held.
630 static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page)
633 struct zone *zone = page_zone(page);
634 struct page_cgroup *pc = lookup_page_cgroup(page);
636 spin_lock_irqsave(&zone->lru_lock, flags);
638 * Forget old LRU when this page_cgroup is *not* used. This Used bit
639 * is guarded by lock_page() because the page is SwapCache.
641 if (!PageCgroupUsed(pc))
642 mem_cgroup_del_lru_list(page, page_lru(page));
643 spin_unlock_irqrestore(&zone->lru_lock, flags);
646 static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page)
649 struct zone *zone = page_zone(page);
650 struct page_cgroup *pc = lookup_page_cgroup(page);
652 spin_lock_irqsave(&zone->lru_lock, flags);
653 /* link when the page is linked to LRU but page_cgroup isn't */
654 if (PageLRU(page) && !PageCgroupAcctLRU(pc))
655 mem_cgroup_add_lru_list(page, page_lru(page));
656 spin_unlock_irqrestore(&zone->lru_lock, flags);
660 void mem_cgroup_move_lists(struct page *page,
661 enum lru_list from, enum lru_list to)
663 if (mem_cgroup_disabled())
665 mem_cgroup_del_lru_list(page, from);
666 mem_cgroup_add_lru_list(page, to);
669 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
672 struct mem_cgroup *curr = NULL;
676 curr = try_get_mem_cgroup_from_mm(task->mm);
681 if (curr->use_hierarchy)
682 ret = css_is_ancestor(&curr->css, &mem->css);
690 * prev_priority control...this will be used in memory reclaim path.
692 int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
696 spin_lock(&mem->reclaim_param_lock);
697 prev_priority = mem->prev_priority;
698 spin_unlock(&mem->reclaim_param_lock);
700 return prev_priority;
703 void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
705 spin_lock(&mem->reclaim_param_lock);
706 if (priority < mem->prev_priority)
707 mem->prev_priority = priority;
708 spin_unlock(&mem->reclaim_param_lock);
711 void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
713 spin_lock(&mem->reclaim_param_lock);
714 mem->prev_priority = priority;
715 spin_unlock(&mem->reclaim_param_lock);
718 static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages)
720 unsigned long active;
721 unsigned long inactive;
723 unsigned long inactive_ratio;
725 inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON);
726 active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON);
728 gb = (inactive + active) >> (30 - PAGE_SHIFT);
730 inactive_ratio = int_sqrt(10 * gb);
735 present_pages[0] = inactive;
736 present_pages[1] = active;
739 return inactive_ratio;
742 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg)
744 unsigned long active;
745 unsigned long inactive;
746 unsigned long present_pages[2];
747 unsigned long inactive_ratio;
749 inactive_ratio = calc_inactive_ratio(memcg, present_pages);
751 inactive = present_pages[0];
752 active = present_pages[1];
754 if (inactive * inactive_ratio < active)
760 int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg)
762 unsigned long active;
763 unsigned long inactive;
765 inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE);
766 active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE);
768 return (active > inactive);
771 unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg,
775 int nid = zone->zone_pgdat->node_id;
776 int zid = zone_idx(zone);
777 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
779 return MEM_CGROUP_ZSTAT(mz, lru);
782 struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
785 int nid = zone->zone_pgdat->node_id;
786 int zid = zone_idx(zone);
787 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
789 return &mz->reclaim_stat;
792 struct zone_reclaim_stat *
793 mem_cgroup_get_reclaim_stat_from_page(struct page *page)
795 struct page_cgroup *pc;
796 struct mem_cgroup_per_zone *mz;
798 if (mem_cgroup_disabled())
801 pc = lookup_page_cgroup(page);
803 * Used bit is set without atomic ops but after smp_wmb().
804 * For making pc->mem_cgroup visible, insert smp_rmb() here.
807 if (!PageCgroupUsed(pc))
810 mz = page_cgroup_zoneinfo(pc);
814 return &mz->reclaim_stat;
817 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
818 struct list_head *dst,
819 unsigned long *scanned, int order,
820 int mode, struct zone *z,
821 struct mem_cgroup *mem_cont,
822 int active, int file)
824 unsigned long nr_taken = 0;
828 struct list_head *src;
829 struct page_cgroup *pc, *tmp;
830 int nid = z->zone_pgdat->node_id;
831 int zid = zone_idx(z);
832 struct mem_cgroup_per_zone *mz;
833 int lru = LRU_FILE * file + active;
837 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
838 src = &mz->lists[lru];
841 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
842 if (scan >= nr_to_scan)
846 if (unlikely(!PageCgroupUsed(pc)))
848 if (unlikely(!PageLRU(page)))
852 ret = __isolate_lru_page(page, mode, file);
855 list_move(&page->lru, dst);
856 mem_cgroup_del_lru(page);
860 /* we don't affect global LRU but rotate in our LRU */
861 mem_cgroup_rotate_lru_list(page, page_lru(page));
872 #define mem_cgroup_from_res_counter(counter, member) \
873 container_of(counter, struct mem_cgroup, member)
875 static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem)
877 if (do_swap_account) {
878 if (res_counter_check_under_limit(&mem->res) &&
879 res_counter_check_under_limit(&mem->memsw))
882 if (res_counter_check_under_limit(&mem->res))
887 static unsigned int get_swappiness(struct mem_cgroup *memcg)
889 struct cgroup *cgrp = memcg->css.cgroup;
890 unsigned int swappiness;
893 if (cgrp->parent == NULL)
894 return vm_swappiness;
896 spin_lock(&memcg->reclaim_param_lock);
897 swappiness = memcg->swappiness;
898 spin_unlock(&memcg->reclaim_param_lock);
903 static int mem_cgroup_count_children_cb(struct mem_cgroup *mem, void *data)
911 * mem_cgroup_print_mem_info: Called from OOM with tasklist_lock held in read mode.
912 * @memcg: The memory cgroup that went over limit
913 * @p: Task that is going to be killed
915 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
918 void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
920 struct cgroup *task_cgrp;
921 struct cgroup *mem_cgrp;
923 * Need a buffer in BSS, can't rely on allocations. The code relies
924 * on the assumption that OOM is serialized for memory controller.
925 * If this assumption is broken, revisit this code.
927 static char memcg_name[PATH_MAX];
936 mem_cgrp = memcg->css.cgroup;
937 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
939 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
942 * Unfortunately, we are unable to convert to a useful name
943 * But we'll still print out the usage information
950 printk(KERN_INFO "Task in %s killed", memcg_name);
953 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
961 * Continues from above, so we don't need an KERN_ level
963 printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
966 printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
967 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
968 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
969 res_counter_read_u64(&memcg->res, RES_FAILCNT));
970 printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
972 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
973 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
974 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
978 * This function returns the number of memcg under hierarchy tree. Returns
979 * 1(self count) if no children.
981 static int mem_cgroup_count_children(struct mem_cgroup *mem)
984 mem_cgroup_walk_tree(mem, &num, mem_cgroup_count_children_cb);
989 * Visit the first child (need not be the first child as per the ordering
990 * of the cgroup list, since we track last_scanned_child) of @mem and use
991 * that to reclaim free pages from.
993 static struct mem_cgroup *
994 mem_cgroup_select_victim(struct mem_cgroup *root_mem)
996 struct mem_cgroup *ret = NULL;
997 struct cgroup_subsys_state *css;
1000 if (!root_mem->use_hierarchy) {
1001 css_get(&root_mem->css);
1007 nextid = root_mem->last_scanned_child + 1;
1008 css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css,
1010 if (css && css_tryget(css))
1011 ret = container_of(css, struct mem_cgroup, css);
1014 /* Updates scanning parameter */
1015 spin_lock(&root_mem->reclaim_param_lock);
1017 /* this means start scan from ID:1 */
1018 root_mem->last_scanned_child = 0;
1020 root_mem->last_scanned_child = found;
1021 spin_unlock(&root_mem->reclaim_param_lock);
1028 * Scan the hierarchy if needed to reclaim memory. We remember the last child
1029 * we reclaimed from, so that we don't end up penalizing one child extensively
1030 * based on its position in the children list.
1032 * root_mem is the original ancestor that we've been reclaim from.
1034 * We give up and return to the caller when we visit root_mem twice.
1035 * (other groups can be removed while we're walking....)
1037 * If shrink==true, for avoiding to free too much, this returns immedieately.
1039 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
1041 unsigned long reclaim_options)
1043 struct mem_cgroup *victim;
1046 bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP;
1047 bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK;
1049 /* If memsw_is_minimum==1, swap-out is of-no-use. */
1050 if (root_mem->memsw_is_minimum)
1054 victim = mem_cgroup_select_victim(root_mem);
1055 if (victim == root_mem)
1057 if (!mem_cgroup_local_usage(&victim->stat)) {
1058 /* this cgroup's local usage == 0 */
1059 css_put(&victim->css);
1062 /* we use swappiness of local cgroup */
1063 ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, noswap,
1064 get_swappiness(victim));
1065 css_put(&victim->css);
1067 * At shrinking usage, we can't check we should stop here or
1068 * reclaim more. It's depends on callers. last_scanned_child
1069 * will work enough for keeping fairness under tree.
1074 if (mem_cgroup_check_under_limit(root_mem))
1080 bool mem_cgroup_oom_called(struct task_struct *task)
1083 struct mem_cgroup *mem;
1084 struct mm_struct *mm;
1090 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
1091 if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10))
1097 static int record_last_oom_cb(struct mem_cgroup *mem, void *data)
1099 mem->last_oom_jiffies = jiffies;
1103 static void record_last_oom(struct mem_cgroup *mem)
1105 mem_cgroup_walk_tree(mem, NULL, record_last_oom_cb);
1109 * Currently used to update mapped file statistics, but the routine can be
1110 * generalized to update other statistics as well.
1112 void mem_cgroup_update_mapped_file_stat(struct page *page, int val)
1114 struct mem_cgroup *mem;
1115 struct mem_cgroup_stat *stat;
1116 struct mem_cgroup_stat_cpu *cpustat;
1118 struct page_cgroup *pc;
1120 if (!page_is_file_cache(page))
1123 pc = lookup_page_cgroup(page);
1127 lock_page_cgroup(pc);
1128 mem = pc->mem_cgroup;
1132 if (!PageCgroupUsed(pc))
1136 * Preemption is already disabled, we don't need get_cpu()
1138 cpu = smp_processor_id();
1140 cpustat = &stat->cpustat[cpu];
1142 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_MAPPED_FILE, val);
1144 unlock_page_cgroup(pc);
1148 * Unlike exported interface, "oom" parameter is added. if oom==true,
1149 * oom-killer can be invoked.
1151 static int __mem_cgroup_try_charge(struct mm_struct *mm,
1152 gfp_t gfp_mask, struct mem_cgroup **memcg,
1153 bool oom, struct page *page)
1155 struct mem_cgroup *mem, *mem_over_limit, *mem_over_soft_limit;
1156 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
1157 struct res_counter *fail_res, *soft_fail_res = NULL;
1159 if (unlikely(test_thread_flag(TIF_MEMDIE))) {
1160 /* Don't account this! */
1166 * We always charge the cgroup the mm_struct belongs to.
1167 * The mm_struct's mem_cgroup changes on task migration if the
1168 * thread group leader migrates. It's possible that mm is not
1169 * set, if so charge the init_mm (happens for pagecache usage).
1173 mem = try_get_mem_cgroup_from_mm(mm);
1181 VM_BUG_ON(css_is_removed(&mem->css));
1185 unsigned long flags = 0;
1187 ret = res_counter_charge(&mem->res, PAGE_SIZE, &fail_res,
1190 if (!do_swap_account)
1192 ret = res_counter_charge(&mem->memsw, PAGE_SIZE,
1196 /* mem+swap counter fails */
1197 res_counter_uncharge(&mem->res, PAGE_SIZE, NULL);
1198 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
1199 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
1202 /* mem counter fails */
1203 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
1206 if (!(gfp_mask & __GFP_WAIT))
1209 ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, gfp_mask,
1215 * try_to_free_mem_cgroup_pages() might not give us a full
1216 * picture of reclaim. Some pages are reclaimed and might be
1217 * moved to swap cache or just unmapped from the cgroup.
1218 * Check the limit again to see if the reclaim reduced the
1219 * current usage of the cgroup before giving up
1222 if (mem_cgroup_check_under_limit(mem_over_limit))
1225 if (!nr_retries--) {
1227 mutex_lock(&memcg_tasklist);
1228 mem_cgroup_out_of_memory(mem_over_limit, gfp_mask);
1229 mutex_unlock(&memcg_tasklist);
1230 record_last_oom(mem_over_limit);
1236 * Insert just the ancestor, we should trickle down to the correct
1237 * cgroup for reclaim, since the other nodes will be below their
1240 if (soft_fail_res) {
1241 mem_over_soft_limit =
1242 mem_cgroup_from_res_counter(soft_fail_res, res);
1243 if (mem_cgroup_soft_limit_check(mem_over_soft_limit))
1244 mem_cgroup_update_tree(mem_over_soft_limit, page);
1253 * A helper function to get mem_cgroup from ID. must be called under
1254 * rcu_read_lock(). The caller must check css_is_removed() or some if
1255 * it's concern. (dropping refcnt from swap can be called against removed
1258 static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
1260 struct cgroup_subsys_state *css;
1262 /* ID 0 is unused ID */
1265 css = css_lookup(&mem_cgroup_subsys, id);
1268 return container_of(css, struct mem_cgroup, css);
1271 static struct mem_cgroup *try_get_mem_cgroup_from_swapcache(struct page *page)
1273 struct mem_cgroup *mem;
1274 struct page_cgroup *pc;
1278 VM_BUG_ON(!PageLocked(page));
1280 if (!PageSwapCache(page))
1283 pc = lookup_page_cgroup(page);
1284 lock_page_cgroup(pc);
1285 if (PageCgroupUsed(pc)) {
1286 mem = pc->mem_cgroup;
1287 if (mem && !css_tryget(&mem->css))
1290 ent.val = page_private(page);
1291 id = lookup_swap_cgroup(ent);
1293 mem = mem_cgroup_lookup(id);
1294 if (mem && !css_tryget(&mem->css))
1298 unlock_page_cgroup(pc);
1303 * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be
1304 * USED state. If already USED, uncharge and return.
1307 static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
1308 struct page_cgroup *pc,
1309 enum charge_type ctype)
1311 /* try_charge() can return NULL to *memcg, taking care of it. */
1315 lock_page_cgroup(pc);
1316 if (unlikely(PageCgroupUsed(pc))) {
1317 unlock_page_cgroup(pc);
1318 res_counter_uncharge(&mem->res, PAGE_SIZE, NULL);
1319 if (do_swap_account)
1320 res_counter_uncharge(&mem->memsw, PAGE_SIZE, NULL);
1325 pc->mem_cgroup = mem;
1327 * We access a page_cgroup asynchronously without lock_page_cgroup().
1328 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
1329 * is accessed after testing USED bit. To make pc->mem_cgroup visible
1330 * before USED bit, we need memory barrier here.
1331 * See mem_cgroup_add_lru_list(), etc.
1335 case MEM_CGROUP_CHARGE_TYPE_CACHE:
1336 case MEM_CGROUP_CHARGE_TYPE_SHMEM:
1337 SetPageCgroupCache(pc);
1338 SetPageCgroupUsed(pc);
1340 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
1341 ClearPageCgroupCache(pc);
1342 SetPageCgroupUsed(pc);
1348 mem_cgroup_charge_statistics(mem, pc, true);
1350 unlock_page_cgroup(pc);
1354 * mem_cgroup_move_account - move account of the page
1355 * @pc: page_cgroup of the page.
1356 * @from: mem_cgroup which the page is moved from.
1357 * @to: mem_cgroup which the page is moved to. @from != @to.
1359 * The caller must confirm following.
1360 * - page is not on LRU (isolate_page() is useful.)
1362 * returns 0 at success,
1363 * returns -EBUSY when lock is busy or "pc" is unstable.
1365 * This function does "uncharge" from old cgroup but doesn't do "charge" to
1366 * new cgroup. It should be done by a caller.
1369 static int mem_cgroup_move_account(struct page_cgroup *pc,
1370 struct mem_cgroup *from, struct mem_cgroup *to)
1372 struct mem_cgroup_per_zone *from_mz, *to_mz;
1377 struct mem_cgroup_stat *stat;
1378 struct mem_cgroup_stat_cpu *cpustat;
1380 VM_BUG_ON(from == to);
1381 VM_BUG_ON(PageLRU(pc->page));
1383 nid = page_cgroup_nid(pc);
1384 zid = page_cgroup_zid(pc);
1385 from_mz = mem_cgroup_zoneinfo(from, nid, zid);
1386 to_mz = mem_cgroup_zoneinfo(to, nid, zid);
1388 if (!trylock_page_cgroup(pc))
1391 if (!PageCgroupUsed(pc))
1394 if (pc->mem_cgroup != from)
1397 res_counter_uncharge(&from->res, PAGE_SIZE, NULL);
1398 mem_cgroup_charge_statistics(from, pc, false);
1401 if (page_is_file_cache(page) && page_mapped(page)) {
1402 cpu = smp_processor_id();
1403 /* Update mapped_file data for mem_cgroup "from" */
1405 cpustat = &stat->cpustat[cpu];
1406 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_MAPPED_FILE,
1409 /* Update mapped_file data for mem_cgroup "to" */
1411 cpustat = &stat->cpustat[cpu];
1412 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_MAPPED_FILE,
1416 if (do_swap_account)
1417 res_counter_uncharge(&from->memsw, PAGE_SIZE, NULL);
1418 css_put(&from->css);
1421 pc->mem_cgroup = to;
1422 mem_cgroup_charge_statistics(to, pc, true);
1425 unlock_page_cgroup(pc);
1427 * We charges against "to" which may not have any tasks. Then, "to"
1428 * can be under rmdir(). But in current implementation, caller of
1429 * this function is just force_empty() and it's garanteed that
1430 * "to" is never removed. So, we don't check rmdir status here.
1436 * move charges to its parent.
1439 static int mem_cgroup_move_parent(struct page_cgroup *pc,
1440 struct mem_cgroup *child,
1443 struct page *page = pc->page;
1444 struct cgroup *cg = child->css.cgroup;
1445 struct cgroup *pcg = cg->parent;
1446 struct mem_cgroup *parent;
1454 parent = mem_cgroup_from_cont(pcg);
1457 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false, page);
1461 if (!get_page_unless_zero(page)) {
1466 ret = isolate_lru_page(page);
1471 ret = mem_cgroup_move_account(pc, child, parent);
1473 putback_lru_page(page);
1476 /* drop extra refcnt by try_charge() */
1477 css_put(&parent->css);
1484 /* drop extra refcnt by try_charge() */
1485 css_put(&parent->css);
1486 /* uncharge if move fails */
1487 res_counter_uncharge(&parent->res, PAGE_SIZE, NULL);
1488 if (do_swap_account)
1489 res_counter_uncharge(&parent->memsw, PAGE_SIZE, NULL);
1494 * Charge the memory controller for page usage.
1496 * 0 if the charge was successful
1497 * < 0 if the cgroup is over its limit
1499 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
1500 gfp_t gfp_mask, enum charge_type ctype,
1501 struct mem_cgroup *memcg)
1503 struct mem_cgroup *mem;
1504 struct page_cgroup *pc;
1507 pc = lookup_page_cgroup(page);
1508 /* can happen at boot */
1514 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true, page);
1518 __mem_cgroup_commit_charge(mem, pc, ctype);
1522 int mem_cgroup_newpage_charge(struct page *page,
1523 struct mm_struct *mm, gfp_t gfp_mask)
1525 if (mem_cgroup_disabled())
1527 if (PageCompound(page))
1530 * If already mapped, we don't have to account.
1531 * If page cache, page->mapping has address_space.
1532 * But page->mapping may have out-of-use anon_vma pointer,
1533 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
1536 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
1540 return mem_cgroup_charge_common(page, mm, gfp_mask,
1541 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
1545 __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
1546 enum charge_type ctype);
1548 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
1551 struct mem_cgroup *mem = NULL;
1554 if (mem_cgroup_disabled())
1556 if (PageCompound(page))
1559 * Corner case handling. This is called from add_to_page_cache()
1560 * in usual. But some FS (shmem) precharges this page before calling it
1561 * and call add_to_page_cache() with GFP_NOWAIT.
1563 * For GFP_NOWAIT case, the page may be pre-charged before calling
1564 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
1565 * charge twice. (It works but has to pay a bit larger cost.)
1566 * And when the page is SwapCache, it should take swap information
1567 * into account. This is under lock_page() now.
1569 if (!(gfp_mask & __GFP_WAIT)) {
1570 struct page_cgroup *pc;
1573 pc = lookup_page_cgroup(page);
1576 lock_page_cgroup(pc);
1577 if (PageCgroupUsed(pc)) {
1578 unlock_page_cgroup(pc);
1581 unlock_page_cgroup(pc);
1584 if (unlikely(!mm && !mem))
1587 if (page_is_file_cache(page))
1588 return mem_cgroup_charge_common(page, mm, gfp_mask,
1589 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
1592 if (PageSwapCache(page)) {
1593 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
1595 __mem_cgroup_commit_charge_swapin(page, mem,
1596 MEM_CGROUP_CHARGE_TYPE_SHMEM);
1598 ret = mem_cgroup_charge_common(page, mm, gfp_mask,
1599 MEM_CGROUP_CHARGE_TYPE_SHMEM, mem);
1605 * While swap-in, try_charge -> commit or cancel, the page is locked.
1606 * And when try_charge() successfully returns, one refcnt to memcg without
1607 * struct page_cgroup is aquired. This refcnt will be cumsumed by
1608 * "commit()" or removed by "cancel()"
1610 int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
1612 gfp_t mask, struct mem_cgroup **ptr)
1614 struct mem_cgroup *mem;
1617 if (mem_cgroup_disabled())
1620 if (!do_swap_account)
1623 * A racing thread's fault, or swapoff, may have already updated
1624 * the pte, and even removed page from swap cache: return success
1625 * to go on to do_swap_page()'s pte_same() test, which should fail.
1627 if (!PageSwapCache(page))
1629 mem = try_get_mem_cgroup_from_swapcache(page);
1633 ret = __mem_cgroup_try_charge(NULL, mask, ptr, true, page);
1634 /* drop extra refcnt from tryget */
1640 return __mem_cgroup_try_charge(mm, mask, ptr, true, page);
1644 __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
1645 enum charge_type ctype)
1647 struct page_cgroup *pc;
1649 if (mem_cgroup_disabled())
1653 cgroup_exclude_rmdir(&ptr->css);
1654 pc = lookup_page_cgroup(page);
1655 mem_cgroup_lru_del_before_commit_swapcache(page);
1656 __mem_cgroup_commit_charge(ptr, pc, ctype);
1657 mem_cgroup_lru_add_after_commit_swapcache(page);
1659 * Now swap is on-memory. This means this page may be
1660 * counted both as mem and swap....double count.
1661 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
1662 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
1663 * may call delete_from_swap_cache() before reach here.
1665 if (do_swap_account && PageSwapCache(page)) {
1666 swp_entry_t ent = {.val = page_private(page)};
1668 struct mem_cgroup *memcg;
1670 id = swap_cgroup_record(ent, 0);
1672 memcg = mem_cgroup_lookup(id);
1675 * This recorded memcg can be obsolete one. So, avoid
1676 * calling css_tryget
1678 res_counter_uncharge(&memcg->memsw, PAGE_SIZE, NULL);
1679 mem_cgroup_put(memcg);
1684 * At swapin, we may charge account against cgroup which has no tasks.
1685 * So, rmdir()->pre_destroy() can be called while we do this charge.
1686 * In that case, we need to call pre_destroy() again. check it here.
1688 cgroup_release_and_wakeup_rmdir(&ptr->css);
1691 void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
1693 __mem_cgroup_commit_charge_swapin(page, ptr,
1694 MEM_CGROUP_CHARGE_TYPE_MAPPED);
1697 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
1699 if (mem_cgroup_disabled())
1703 res_counter_uncharge(&mem->res, PAGE_SIZE, NULL);
1704 if (do_swap_account)
1705 res_counter_uncharge(&mem->memsw, PAGE_SIZE, NULL);
1711 * uncharge if !page_mapped(page)
1713 static struct mem_cgroup *
1714 __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
1716 struct page_cgroup *pc;
1717 struct mem_cgroup *mem = NULL;
1718 struct mem_cgroup_per_zone *mz;
1719 bool soft_limit_excess = false;
1721 if (mem_cgroup_disabled())
1724 if (PageSwapCache(page))
1728 * Check if our page_cgroup is valid
1730 pc = lookup_page_cgroup(page);
1731 if (unlikely(!pc || !PageCgroupUsed(pc)))
1734 lock_page_cgroup(pc);
1736 mem = pc->mem_cgroup;
1738 if (!PageCgroupUsed(pc))
1742 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
1743 case MEM_CGROUP_CHARGE_TYPE_DROP:
1744 if (page_mapped(page))
1747 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
1748 if (!PageAnon(page)) { /* Shared memory */
1749 if (page->mapping && !page_is_file_cache(page))
1751 } else if (page_mapped(page)) /* Anon */
1758 res_counter_uncharge(&mem->res, PAGE_SIZE, &soft_limit_excess);
1759 if (do_swap_account && (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT))
1760 res_counter_uncharge(&mem->memsw, PAGE_SIZE, NULL);
1761 mem_cgroup_charge_statistics(mem, pc, false);
1763 ClearPageCgroupUsed(pc);
1765 * pc->mem_cgroup is not cleared here. It will be accessed when it's
1766 * freed from LRU. This is safe because uncharged page is expected not
1767 * to be reused (freed soon). Exception is SwapCache, it's handled by
1768 * special functions.
1771 mz = page_cgroup_zoneinfo(pc);
1772 unlock_page_cgroup(pc);
1774 if (soft_limit_excess && mem_cgroup_soft_limit_check(mem))
1775 mem_cgroup_update_tree(mem, page);
1776 /* at swapout, this memcg will be accessed to record to swap */
1777 if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
1783 unlock_page_cgroup(pc);
1787 void mem_cgroup_uncharge_page(struct page *page)
1790 if (page_mapped(page))
1792 if (page->mapping && !PageAnon(page))
1794 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
1797 void mem_cgroup_uncharge_cache_page(struct page *page)
1799 VM_BUG_ON(page_mapped(page));
1800 VM_BUG_ON(page->mapping);
1801 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
1806 * called after __delete_from_swap_cache() and drop "page" account.
1807 * memcg information is recorded to swap_cgroup of "ent"
1810 mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
1812 struct mem_cgroup *memcg;
1813 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
1815 if (!swapout) /* this was a swap cache but the swap is unused ! */
1816 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
1818 memcg = __mem_cgroup_uncharge_common(page, ctype);
1820 /* record memcg information */
1821 if (do_swap_account && swapout && memcg) {
1822 swap_cgroup_record(ent, css_id(&memcg->css));
1823 mem_cgroup_get(memcg);
1825 if (swapout && memcg)
1826 css_put(&memcg->css);
1830 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1832 * called from swap_entry_free(). remove record in swap_cgroup and
1833 * uncharge "memsw" account.
1835 void mem_cgroup_uncharge_swap(swp_entry_t ent)
1837 struct mem_cgroup *memcg;
1840 if (!do_swap_account)
1843 id = swap_cgroup_record(ent, 0);
1845 memcg = mem_cgroup_lookup(id);
1848 * We uncharge this because swap is freed.
1849 * This memcg can be obsolete one. We avoid calling css_tryget
1851 res_counter_uncharge(&memcg->memsw, PAGE_SIZE, NULL);
1852 mem_cgroup_put(memcg);
1859 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
1862 int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr)
1864 struct page_cgroup *pc;
1865 struct mem_cgroup *mem = NULL;
1868 if (mem_cgroup_disabled())
1871 pc = lookup_page_cgroup(page);
1872 lock_page_cgroup(pc);
1873 if (PageCgroupUsed(pc)) {
1874 mem = pc->mem_cgroup;
1877 unlock_page_cgroup(pc);
1880 ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false,
1888 /* remove redundant charge if migration failed*/
1889 void mem_cgroup_end_migration(struct mem_cgroup *mem,
1890 struct page *oldpage, struct page *newpage)
1892 struct page *target, *unused;
1893 struct page_cgroup *pc;
1894 enum charge_type ctype;
1898 cgroup_exclude_rmdir(&mem->css);
1899 /* at migration success, oldpage->mapping is NULL. */
1900 if (oldpage->mapping) {
1908 if (PageAnon(target))
1909 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
1910 else if (page_is_file_cache(target))
1911 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
1913 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
1915 /* unused page is not on radix-tree now. */
1917 __mem_cgroup_uncharge_common(unused, ctype);
1919 pc = lookup_page_cgroup(target);
1921 * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup.
1922 * So, double-counting is effectively avoided.
1924 __mem_cgroup_commit_charge(mem, pc, ctype);
1927 * Both of oldpage and newpage are still under lock_page().
1928 * Then, we don't have to care about race in radix-tree.
1929 * But we have to be careful that this page is unmapped or not.
1931 * There is a case for !page_mapped(). At the start of
1932 * migration, oldpage was mapped. But now, it's zapped.
1933 * But we know *target* page is not freed/reused under us.
1934 * mem_cgroup_uncharge_page() does all necessary checks.
1936 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
1937 mem_cgroup_uncharge_page(target);
1939 * At migration, we may charge account against cgroup which has no tasks
1940 * So, rmdir()->pre_destroy() can be called while we do this charge.
1941 * In that case, we need to call pre_destroy() again. check it here.
1943 cgroup_release_and_wakeup_rmdir(&mem->css);
1947 * A call to try to shrink memory usage on charge failure at shmem's swapin.
1948 * Calling hierarchical_reclaim is not enough because we should update
1949 * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM.
1950 * Moreover considering hierarchy, we should reclaim from the mem_over_limit,
1951 * not from the memcg which this page would be charged to.
1952 * try_charge_swapin does all of these works properly.
1954 int mem_cgroup_shmem_charge_fallback(struct page *page,
1955 struct mm_struct *mm,
1958 struct mem_cgroup *mem = NULL;
1961 if (mem_cgroup_disabled())
1964 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
1966 mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */
1971 static DEFINE_MUTEX(set_limit_mutex);
1973 static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
1974 unsigned long long val)
1980 int children = mem_cgroup_count_children(memcg);
1981 u64 curusage, oldusage;
1984 * For keeping hierarchical_reclaim simple, how long we should retry
1985 * is depends on callers. We set our retry-count to be function
1986 * of # of children which we should visit in this loop.
1988 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
1990 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
1992 while (retry_count) {
1993 if (signal_pending(current)) {
1998 * Rather than hide all in some function, I do this in
1999 * open coded manner. You see what this really does.
2000 * We have to guarantee mem->res.limit < mem->memsw.limit.
2002 mutex_lock(&set_limit_mutex);
2003 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2004 if (memswlimit < val) {
2006 mutex_unlock(&set_limit_mutex);
2009 ret = res_counter_set_limit(&memcg->res, val);
2011 if (memswlimit == val)
2012 memcg->memsw_is_minimum = true;
2014 memcg->memsw_is_minimum = false;
2016 mutex_unlock(&set_limit_mutex);
2021 progress = mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL,
2022 MEM_CGROUP_RECLAIM_SHRINK);
2023 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
2024 /* Usage is reduced ? */
2025 if (curusage >= oldusage)
2028 oldusage = curusage;
2034 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
2035 unsigned long long val)
2038 u64 memlimit, oldusage, curusage;
2039 int children = mem_cgroup_count_children(memcg);
2042 /* see mem_cgroup_resize_res_limit */
2043 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
2044 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
2045 while (retry_count) {
2046 if (signal_pending(current)) {
2051 * Rather than hide all in some function, I do this in
2052 * open coded manner. You see what this really does.
2053 * We have to guarantee mem->res.limit < mem->memsw.limit.
2055 mutex_lock(&set_limit_mutex);
2056 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
2057 if (memlimit > val) {
2059 mutex_unlock(&set_limit_mutex);
2062 ret = res_counter_set_limit(&memcg->memsw, val);
2064 if (memlimit == val)
2065 memcg->memsw_is_minimum = true;
2067 memcg->memsw_is_minimum = false;
2069 mutex_unlock(&set_limit_mutex);
2074 mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL,
2075 MEM_CGROUP_RECLAIM_NOSWAP |
2076 MEM_CGROUP_RECLAIM_SHRINK);
2077 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
2078 /* Usage is reduced ? */
2079 if (curusage >= oldusage)
2082 oldusage = curusage;
2088 * This routine traverse page_cgroup in given list and drop them all.
2089 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
2091 static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
2092 int node, int zid, enum lru_list lru)
2095 struct mem_cgroup_per_zone *mz;
2096 struct page_cgroup *pc, *busy;
2097 unsigned long flags, loop;
2098 struct list_head *list;
2101 zone = &NODE_DATA(node)->node_zones[zid];
2102 mz = mem_cgroup_zoneinfo(mem, node, zid);
2103 list = &mz->lists[lru];
2105 loop = MEM_CGROUP_ZSTAT(mz, lru);
2106 /* give some margin against EBUSY etc...*/
2111 spin_lock_irqsave(&zone->lru_lock, flags);
2112 if (list_empty(list)) {
2113 spin_unlock_irqrestore(&zone->lru_lock, flags);
2116 pc = list_entry(list->prev, struct page_cgroup, lru);
2118 list_move(&pc->lru, list);
2120 spin_unlock_irqrestore(&zone->lru_lock, flags);
2123 spin_unlock_irqrestore(&zone->lru_lock, flags);
2125 ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL);
2129 if (ret == -EBUSY || ret == -EINVAL) {
2130 /* found lock contention or "pc" is obsolete. */
2137 if (!ret && !list_empty(list))
2143 * make mem_cgroup's charge to be 0 if there is no task.
2144 * This enables deleting this mem_cgroup.
2146 static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
2149 int node, zid, shrink;
2150 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
2151 struct cgroup *cgrp = mem->css.cgroup;
2156 /* should free all ? */
2160 while (mem->res.usage > 0) {
2162 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
2165 if (signal_pending(current))
2167 /* This is for making all *used* pages to be on LRU. */
2168 lru_add_drain_all();
2170 for_each_node_state(node, N_HIGH_MEMORY) {
2171 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
2174 ret = mem_cgroup_force_empty_list(mem,
2183 /* it seems parent cgroup doesn't have enough mem */
2194 /* returns EBUSY if there is a task or if we come here twice. */
2195 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
2199 /* we call try-to-free pages for make this cgroup empty */
2200 lru_add_drain_all();
2201 /* try to free all pages in this cgroup */
2203 while (nr_retries && mem->res.usage > 0) {
2206 if (signal_pending(current)) {
2210 progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL,
2211 false, get_swappiness(mem));
2214 /* maybe some writeback is necessary */
2215 congestion_wait(BLK_RW_ASYNC, HZ/10);
2220 /* try move_account...there may be some *locked* pages. */
2227 int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
2229 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
2233 static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
2235 return mem_cgroup_from_cont(cont)->use_hierarchy;
2238 static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
2242 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2243 struct cgroup *parent = cont->parent;
2244 struct mem_cgroup *parent_mem = NULL;
2247 parent_mem = mem_cgroup_from_cont(parent);
2251 * If parent's use_hiearchy is set, we can't make any modifications
2252 * in the child subtrees. If it is unset, then the change can
2253 * occur, provided the current cgroup has no children.
2255 * For the root cgroup, parent_mem is NULL, we allow value to be
2256 * set if there are no children.
2258 if ((!parent_mem || !parent_mem->use_hierarchy) &&
2259 (val == 1 || val == 0)) {
2260 if (list_empty(&cont->children))
2261 mem->use_hierarchy = val;
2271 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
2273 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2277 type = MEMFILE_TYPE(cft->private);
2278 name = MEMFILE_ATTR(cft->private);
2281 val = res_counter_read_u64(&mem->res, name);
2284 val = res_counter_read_u64(&mem->memsw, name);
2293 * The user of this function is...
2296 static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
2299 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
2301 unsigned long long val;
2304 type = MEMFILE_TYPE(cft->private);
2305 name = MEMFILE_ATTR(cft->private);
2308 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
2312 /* This function does all necessary parse...reuse it */
2313 ret = res_counter_memparse_write_strategy(buffer, &val);
2317 ret = mem_cgroup_resize_limit(memcg, val);
2319 ret = mem_cgroup_resize_memsw_limit(memcg, val);
2321 case RES_SOFT_LIMIT:
2322 ret = res_counter_memparse_write_strategy(buffer, &val);
2326 * For memsw, soft limits are hard to implement in terms
2327 * of semantics, for now, we support soft limits for
2328 * control without swap
2331 ret = res_counter_set_soft_limit(&memcg->res, val);
2336 ret = -EINVAL; /* should be BUG() ? */
2342 static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
2343 unsigned long long *mem_limit, unsigned long long *memsw_limit)
2345 struct cgroup *cgroup;
2346 unsigned long long min_limit, min_memsw_limit, tmp;
2348 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
2349 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2350 cgroup = memcg->css.cgroup;
2351 if (!memcg->use_hierarchy)
2354 while (cgroup->parent) {
2355 cgroup = cgroup->parent;
2356 memcg = mem_cgroup_from_cont(cgroup);
2357 if (!memcg->use_hierarchy)
2359 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
2360 min_limit = min(min_limit, tmp);
2361 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2362 min_memsw_limit = min(min_memsw_limit, tmp);
2365 *mem_limit = min_limit;
2366 *memsw_limit = min_memsw_limit;
2370 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
2372 struct mem_cgroup *mem;
2375 mem = mem_cgroup_from_cont(cont);
2376 type = MEMFILE_TYPE(event);
2377 name = MEMFILE_ATTR(event);
2381 res_counter_reset_max(&mem->res);
2383 res_counter_reset_max(&mem->memsw);
2387 res_counter_reset_failcnt(&mem->res);
2389 res_counter_reset_failcnt(&mem->memsw);
2397 /* For read statistics */
2412 struct mcs_total_stat {
2413 s64 stat[NR_MCS_STAT];
2419 } memcg_stat_strings[NR_MCS_STAT] = {
2420 {"cache", "total_cache"},
2421 {"rss", "total_rss"},
2422 {"mapped_file", "total_mapped_file"},
2423 {"pgpgin", "total_pgpgin"},
2424 {"pgpgout", "total_pgpgout"},
2425 {"inactive_anon", "total_inactive_anon"},
2426 {"active_anon", "total_active_anon"},
2427 {"inactive_file", "total_inactive_file"},
2428 {"active_file", "total_active_file"},
2429 {"unevictable", "total_unevictable"}
2433 static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data)
2435 struct mcs_total_stat *s = data;
2439 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_CACHE);
2440 s->stat[MCS_CACHE] += val * PAGE_SIZE;
2441 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
2442 s->stat[MCS_RSS] += val * PAGE_SIZE;
2443 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_MAPPED_FILE);
2444 s->stat[MCS_MAPPED_FILE] += val * PAGE_SIZE;
2445 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGIN_COUNT);
2446 s->stat[MCS_PGPGIN] += val;
2447 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGOUT_COUNT);
2448 s->stat[MCS_PGPGOUT] += val;
2451 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON);
2452 s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE;
2453 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON);
2454 s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE;
2455 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE);
2456 s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE;
2457 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE);
2458 s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE;
2459 val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE);
2460 s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE;
2465 mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s)
2467 mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat);
2470 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
2471 struct cgroup_map_cb *cb)
2473 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
2474 struct mcs_total_stat mystat;
2477 memset(&mystat, 0, sizeof(mystat));
2478 mem_cgroup_get_local_stat(mem_cont, &mystat);
2480 for (i = 0; i < NR_MCS_STAT; i++)
2481 cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
2483 /* Hierarchical information */
2485 unsigned long long limit, memsw_limit;
2486 memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit);
2487 cb->fill(cb, "hierarchical_memory_limit", limit);
2488 if (do_swap_account)
2489 cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
2492 memset(&mystat, 0, sizeof(mystat));
2493 mem_cgroup_get_total_stat(mem_cont, &mystat);
2494 for (i = 0; i < NR_MCS_STAT; i++)
2495 cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
2498 #ifdef CONFIG_DEBUG_VM
2499 cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL));
2503 struct mem_cgroup_per_zone *mz;
2504 unsigned long recent_rotated[2] = {0, 0};
2505 unsigned long recent_scanned[2] = {0, 0};
2507 for_each_online_node(nid)
2508 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
2509 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
2511 recent_rotated[0] +=
2512 mz->reclaim_stat.recent_rotated[0];
2513 recent_rotated[1] +=
2514 mz->reclaim_stat.recent_rotated[1];
2515 recent_scanned[0] +=
2516 mz->reclaim_stat.recent_scanned[0];
2517 recent_scanned[1] +=
2518 mz->reclaim_stat.recent_scanned[1];
2520 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
2521 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
2522 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
2523 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
2530 static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
2532 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
2534 return get_swappiness(memcg);
2537 static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
2540 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
2541 struct mem_cgroup *parent;
2546 if (cgrp->parent == NULL)
2549 parent = mem_cgroup_from_cont(cgrp->parent);
2553 /* If under hierarchy, only empty-root can set this value */
2554 if ((parent->use_hierarchy) ||
2555 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
2560 spin_lock(&memcg->reclaim_param_lock);
2561 memcg->swappiness = val;
2562 spin_unlock(&memcg->reclaim_param_lock);
2570 static struct cftype mem_cgroup_files[] = {
2572 .name = "usage_in_bytes",
2573 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
2574 .read_u64 = mem_cgroup_read,
2577 .name = "max_usage_in_bytes",
2578 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
2579 .trigger = mem_cgroup_reset,
2580 .read_u64 = mem_cgroup_read,
2583 .name = "limit_in_bytes",
2584 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
2585 .write_string = mem_cgroup_write,
2586 .read_u64 = mem_cgroup_read,
2589 .name = "soft_limit_in_bytes",
2590 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
2591 .write_string = mem_cgroup_write,
2592 .read_u64 = mem_cgroup_read,
2596 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
2597 .trigger = mem_cgroup_reset,
2598 .read_u64 = mem_cgroup_read,
2602 .read_map = mem_control_stat_show,
2605 .name = "force_empty",
2606 .trigger = mem_cgroup_force_empty_write,
2609 .name = "use_hierarchy",
2610 .write_u64 = mem_cgroup_hierarchy_write,
2611 .read_u64 = mem_cgroup_hierarchy_read,
2614 .name = "swappiness",
2615 .read_u64 = mem_cgroup_swappiness_read,
2616 .write_u64 = mem_cgroup_swappiness_write,
2620 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2621 static struct cftype memsw_cgroup_files[] = {
2623 .name = "memsw.usage_in_bytes",
2624 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
2625 .read_u64 = mem_cgroup_read,
2628 .name = "memsw.max_usage_in_bytes",
2629 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
2630 .trigger = mem_cgroup_reset,
2631 .read_u64 = mem_cgroup_read,
2634 .name = "memsw.limit_in_bytes",
2635 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
2636 .write_string = mem_cgroup_write,
2637 .read_u64 = mem_cgroup_read,
2640 .name = "memsw.failcnt",
2641 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
2642 .trigger = mem_cgroup_reset,
2643 .read_u64 = mem_cgroup_read,
2647 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
2649 if (!do_swap_account)
2651 return cgroup_add_files(cont, ss, memsw_cgroup_files,
2652 ARRAY_SIZE(memsw_cgroup_files));
2655 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
2661 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
2663 struct mem_cgroup_per_node *pn;
2664 struct mem_cgroup_per_zone *mz;
2666 int zone, tmp = node;
2668 * This routine is called against possible nodes.
2669 * But it's BUG to call kmalloc() against offline node.
2671 * TODO: this routine can waste much memory for nodes which will
2672 * never be onlined. It's better to use memory hotplug callback
2675 if (!node_state(node, N_NORMAL_MEMORY))
2677 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
2681 mem->info.nodeinfo[node] = pn;
2682 memset(pn, 0, sizeof(*pn));
2684 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
2685 mz = &pn->zoneinfo[zone];
2687 INIT_LIST_HEAD(&mz->lists[l]);
2688 mz->usage_in_excess = 0;
2693 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
2695 kfree(mem->info.nodeinfo[node]);
2698 static int mem_cgroup_size(void)
2700 int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu);
2701 return sizeof(struct mem_cgroup) + cpustat_size;
2704 static struct mem_cgroup *mem_cgroup_alloc(void)
2706 struct mem_cgroup *mem;
2707 int size = mem_cgroup_size();
2709 if (size < PAGE_SIZE)
2710 mem = kmalloc(size, GFP_KERNEL);
2712 mem = vmalloc(size);
2715 memset(mem, 0, size);
2720 * At destroying mem_cgroup, references from swap_cgroup can remain.
2721 * (scanning all at force_empty is too costly...)
2723 * Instead of clearing all references at force_empty, we remember
2724 * the number of reference from swap_cgroup and free mem_cgroup when
2725 * it goes down to 0.
2727 * Removal of cgroup itself succeeds regardless of refs from swap.
2730 static void __mem_cgroup_free(struct mem_cgroup *mem)
2734 mem_cgroup_remove_from_trees(mem);
2735 free_css_id(&mem_cgroup_subsys, &mem->css);
2737 for_each_node_state(node, N_POSSIBLE)
2738 free_mem_cgroup_per_zone_info(mem, node);
2740 if (mem_cgroup_size() < PAGE_SIZE)
2746 static void mem_cgroup_get(struct mem_cgroup *mem)
2748 atomic_inc(&mem->refcnt);
2751 static void mem_cgroup_put(struct mem_cgroup *mem)
2753 if (atomic_dec_and_test(&mem->refcnt)) {
2754 struct mem_cgroup *parent = parent_mem_cgroup(mem);
2755 __mem_cgroup_free(mem);
2757 mem_cgroup_put(parent);
2762 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
2764 static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem)
2766 if (!mem->res.parent)
2768 return mem_cgroup_from_res_counter(mem->res.parent, res);
2771 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2772 static void __init enable_swap_cgroup(void)
2774 if (!mem_cgroup_disabled() && really_do_swap_account)
2775 do_swap_account = 1;
2778 static void __init enable_swap_cgroup(void)
2783 static int mem_cgroup_soft_limit_tree_init(void)
2785 struct mem_cgroup_tree_per_node *rtpn;
2786 struct mem_cgroup_tree_per_zone *rtpz;
2787 int tmp, node, zone;
2789 for_each_node_state(node, N_POSSIBLE) {
2791 if (!node_state(node, N_NORMAL_MEMORY))
2793 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
2797 soft_limit_tree.rb_tree_per_node[node] = rtpn;
2799 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
2800 rtpz = &rtpn->rb_tree_per_zone[zone];
2801 rtpz->rb_root = RB_ROOT;
2802 spin_lock_init(&rtpz->lock);
2808 static struct cgroup_subsys_state * __ref
2809 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
2811 struct mem_cgroup *mem, *parent;
2812 long error = -ENOMEM;
2815 mem = mem_cgroup_alloc();
2817 return ERR_PTR(error);
2819 for_each_node_state(node, N_POSSIBLE)
2820 if (alloc_mem_cgroup_per_zone_info(mem, node))
2824 if (cont->parent == NULL) {
2825 enable_swap_cgroup();
2827 root_mem_cgroup = mem;
2828 if (mem_cgroup_soft_limit_tree_init())
2832 parent = mem_cgroup_from_cont(cont->parent);
2833 mem->use_hierarchy = parent->use_hierarchy;
2836 if (parent && parent->use_hierarchy) {
2837 res_counter_init(&mem->res, &parent->res);
2838 res_counter_init(&mem->memsw, &parent->memsw);
2840 * We increment refcnt of the parent to ensure that we can
2841 * safely access it on res_counter_charge/uncharge.
2842 * This refcnt will be decremented when freeing this
2843 * mem_cgroup(see mem_cgroup_put).
2845 mem_cgroup_get(parent);
2847 res_counter_init(&mem->res, NULL);
2848 res_counter_init(&mem->memsw, NULL);
2850 mem->last_scanned_child = 0;
2851 spin_lock_init(&mem->reclaim_param_lock);
2854 mem->swappiness = get_swappiness(parent);
2855 atomic_set(&mem->refcnt, 1);
2858 __mem_cgroup_free(mem);
2859 root_mem_cgroup = NULL;
2860 return ERR_PTR(error);
2863 static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
2864 struct cgroup *cont)
2866 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2868 return mem_cgroup_force_empty(mem, false);
2871 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
2872 struct cgroup *cont)
2874 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2876 mem_cgroup_put(mem);
2879 static int mem_cgroup_populate(struct cgroup_subsys *ss,
2880 struct cgroup *cont)
2884 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
2885 ARRAY_SIZE(mem_cgroup_files));
2888 ret = register_memsw_files(cont, ss);
2892 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
2893 struct cgroup *cont,
2894 struct cgroup *old_cont,
2895 struct task_struct *p,
2898 mutex_lock(&memcg_tasklist);
2900 * FIXME: It's better to move charges of this process from old
2901 * memcg to new memcg. But it's just on TODO-List now.
2903 mutex_unlock(&memcg_tasklist);
2906 struct cgroup_subsys mem_cgroup_subsys = {
2908 .subsys_id = mem_cgroup_subsys_id,
2909 .create = mem_cgroup_create,
2910 .pre_destroy = mem_cgroup_pre_destroy,
2911 .destroy = mem_cgroup_destroy,
2912 .populate = mem_cgroup_populate,
2913 .attach = mem_cgroup_move_task,
2918 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2920 static int __init disable_swap_account(char *s)
2922 really_do_swap_account = 0;
2925 __setup("noswapaccount", disable_swap_account);