1 /* memcontrol.c - Memory Controller
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
6 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2 of the License, or
12 * (at your option) any later version.
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
20 #include <linux/res_counter.h>
21 #include <linux/memcontrol.h>
22 #include <linux/cgroup.h>
24 #include <linux/pagemap.h>
25 #include <linux/smp.h>
26 #include <linux/page-flags.h>
27 #include <linux/backing-dev.h>
28 #include <linux/bit_spinlock.h>
29 #include <linux/rcupdate.h>
30 #include <linux/mutex.h>
31 #include <linux/slab.h>
32 #include <linux/swap.h>
33 #include <linux/spinlock.h>
35 #include <linux/seq_file.h>
36 #include <linux/vmalloc.h>
37 #include <linux/mm_inline.h>
38 #include <linux/page_cgroup.h>
41 #include <asm/uaccess.h>
43 struct cgroup_subsys mem_cgroup_subsys __read_mostly;
44 #define MEM_CGROUP_RECLAIM_RETRIES 5
46 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
47 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 0 */
48 int do_swap_account __read_mostly;
49 static int really_do_swap_account __initdata = 1; /* for remember boot option*/
51 #define do_swap_account (0)
56 * Statistics for memory cgroup.
58 enum mem_cgroup_stat_index {
60 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
62 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
63 MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */
64 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
65 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
67 MEM_CGROUP_STAT_NSTATS,
70 struct mem_cgroup_stat_cpu {
71 s64 count[MEM_CGROUP_STAT_NSTATS];
72 } ____cacheline_aligned_in_smp;
74 struct mem_cgroup_stat {
75 struct mem_cgroup_stat_cpu cpustat[0];
79 * For accounting under irq disable, no need for increment preempt count.
81 static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat,
82 enum mem_cgroup_stat_index idx, int val)
84 stat->count[idx] += val;
87 static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat,
88 enum mem_cgroup_stat_index idx)
92 for_each_possible_cpu(cpu)
93 ret += stat->cpustat[cpu].count[idx];
98 * per-zone information in memory controller.
100 struct mem_cgroup_per_zone {
102 * spin_lock to protect the per cgroup LRU
104 struct list_head lists[NR_LRU_LISTS];
105 unsigned long count[NR_LRU_LISTS];
107 /* Macro for accessing counter */
108 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
110 struct mem_cgroup_per_node {
111 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
114 struct mem_cgroup_lru_info {
115 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
119 * The memory controller data structure. The memory controller controls both
120 * page cache and RSS per cgroup. We would eventually like to provide
121 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
122 * to help the administrator determine what knobs to tune.
124 * TODO: Add a water mark for the memory controller. Reclaim will begin when
125 * we hit the water mark. May be even add a low water mark, such that
126 * no reclaim occurs from a cgroup at it's low water mark, this is
127 * a feature that will be implemented much later in the future.
130 struct cgroup_subsys_state css;
132 * the counter to account for memory usage
134 struct res_counter res;
136 * the counter to account for mem+swap usage.
138 struct res_counter memsw;
140 * Per cgroup active and inactive list, similar to the
141 * per zone LRU lists.
143 struct mem_cgroup_lru_info info;
145 int prev_priority; /* for recording reclaim priority */
149 * statistics. This must be placed at the end of memcg.
151 struct mem_cgroup_stat stat;
155 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
156 MEM_CGROUP_CHARGE_TYPE_MAPPED,
157 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
158 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
159 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
163 /* only for here (for easy reading.) */
164 #define PCGF_CACHE (1UL << PCG_CACHE)
165 #define PCGF_USED (1UL << PCG_USED)
166 #define PCGF_LOCK (1UL << PCG_LOCK)
167 static const unsigned long
168 pcg_default_flags[NR_CHARGE_TYPE] = {
169 PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* File Cache */
170 PCGF_USED | PCGF_LOCK, /* Anon */
171 PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* Shmem */
176 /* for encoding cft->private value on file */
179 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
180 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
181 #define MEMFILE_ATTR(val) ((val) & 0xffff)
183 static void mem_cgroup_get(struct mem_cgroup *mem);
184 static void mem_cgroup_put(struct mem_cgroup *mem);
186 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
187 struct page_cgroup *pc,
190 int val = (charge)? 1 : -1;
191 struct mem_cgroup_stat *stat = &mem->stat;
192 struct mem_cgroup_stat_cpu *cpustat;
195 cpustat = &stat->cpustat[cpu];
196 if (PageCgroupCache(pc))
197 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val);
199 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val);
202 __mem_cgroup_stat_add_safe(cpustat,
203 MEM_CGROUP_STAT_PGPGIN_COUNT, 1);
205 __mem_cgroup_stat_add_safe(cpustat,
206 MEM_CGROUP_STAT_PGPGOUT_COUNT, 1);
210 static struct mem_cgroup_per_zone *
211 mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
213 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
216 static struct mem_cgroup_per_zone *
217 page_cgroup_zoneinfo(struct page_cgroup *pc)
219 struct mem_cgroup *mem = pc->mem_cgroup;
220 int nid = page_cgroup_nid(pc);
221 int zid = page_cgroup_zid(pc);
223 return mem_cgroup_zoneinfo(mem, nid, zid);
226 static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem,
230 struct mem_cgroup_per_zone *mz;
233 for_each_online_node(nid)
234 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
235 mz = mem_cgroup_zoneinfo(mem, nid, zid);
236 total += MEM_CGROUP_ZSTAT(mz, idx);
241 static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
243 return container_of(cgroup_subsys_state(cont,
244 mem_cgroup_subsys_id), struct mem_cgroup,
248 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
251 * mm_update_next_owner() may clear mm->owner to NULL
252 * if it races with swapoff, page migration, etc.
253 * So this can be called with p == NULL.
258 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
259 struct mem_cgroup, css);
263 * Following LRU functions are allowed to be used without PCG_LOCK.
264 * Operations are called by routine of global LRU independently from memcg.
265 * What we have to take care of here is validness of pc->mem_cgroup.
267 * Changes to pc->mem_cgroup happens when
270 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
271 * It is added to LRU before charge.
272 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
273 * When moving account, the page is not on LRU. It's isolated.
276 void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
278 struct page_cgroup *pc;
279 struct mem_cgroup *mem;
280 struct mem_cgroup_per_zone *mz;
282 if (mem_cgroup_subsys.disabled)
284 pc = lookup_page_cgroup(page);
285 /* can happen while we handle swapcache. */
286 if (list_empty(&pc->lru))
288 mz = page_cgroup_zoneinfo(pc);
289 mem = pc->mem_cgroup;
290 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
291 list_del_init(&pc->lru);
295 void mem_cgroup_del_lru(struct page *page)
297 mem_cgroup_del_lru_list(page, page_lru(page));
300 void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
302 struct mem_cgroup_per_zone *mz;
303 struct page_cgroup *pc;
305 if (mem_cgroup_subsys.disabled)
308 pc = lookup_page_cgroup(page);
310 /* unused page is not rotated. */
311 if (!PageCgroupUsed(pc))
313 mz = page_cgroup_zoneinfo(pc);
314 list_move(&pc->lru, &mz->lists[lru]);
317 void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
319 struct page_cgroup *pc;
320 struct mem_cgroup_per_zone *mz;
322 if (mem_cgroup_subsys.disabled)
324 pc = lookup_page_cgroup(page);
325 /* barrier to sync with "charge" */
327 if (!PageCgroupUsed(pc))
330 mz = page_cgroup_zoneinfo(pc);
331 MEM_CGROUP_ZSTAT(mz, lru) += 1;
332 list_add(&pc->lru, &mz->lists[lru]);
335 * To add swapcache into LRU. Be careful to all this function.
336 * zone->lru_lock shouldn't be held and irq must not be disabled.
338 static void mem_cgroup_lru_fixup(struct page *page)
340 if (!isolate_lru_page(page))
341 putback_lru_page(page);
344 void mem_cgroup_move_lists(struct page *page,
345 enum lru_list from, enum lru_list to)
347 if (mem_cgroup_subsys.disabled)
349 mem_cgroup_del_lru_list(page, from);
350 mem_cgroup_add_lru_list(page, to);
353 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
358 ret = task->mm && mm_match_cgroup(task->mm, mem);
364 * Calculate mapped_ratio under memory controller. This will be used in
365 * vmscan.c for deteremining we have to reclaim mapped pages.
367 int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem)
372 * usage is recorded in bytes. But, here, we assume the number of
373 * physical pages can be represented by "long" on any arch.
375 total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L;
376 rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
377 return (int)((rss * 100L) / total);
381 * prev_priority control...this will be used in memory reclaim path.
383 int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
385 return mem->prev_priority;
388 void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
390 if (priority < mem->prev_priority)
391 mem->prev_priority = priority;
394 void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
396 mem->prev_priority = priority;
400 * Calculate # of pages to be scanned in this priority/zone.
403 * priority starts from "DEF_PRIORITY" and decremented in each loop.
404 * (see include/linux/mmzone.h)
407 long mem_cgroup_calc_reclaim(struct mem_cgroup *mem, struct zone *zone,
408 int priority, enum lru_list lru)
411 int nid = zone->zone_pgdat->node_id;
412 int zid = zone_idx(zone);
413 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid);
415 nr_pages = MEM_CGROUP_ZSTAT(mz, lru);
417 return (nr_pages >> priority);
420 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
421 struct list_head *dst,
422 unsigned long *scanned, int order,
423 int mode, struct zone *z,
424 struct mem_cgroup *mem_cont,
425 int active, int file)
427 unsigned long nr_taken = 0;
431 struct list_head *src;
432 struct page_cgroup *pc, *tmp;
433 int nid = z->zone_pgdat->node_id;
434 int zid = zone_idx(z);
435 struct mem_cgroup_per_zone *mz;
436 int lru = LRU_FILE * !!file + !!active;
439 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
440 src = &mz->lists[lru];
443 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
444 if (scan >= nr_to_scan)
448 if (unlikely(!PageCgroupUsed(pc)))
450 if (unlikely(!PageLRU(page)))
454 if (__isolate_lru_page(page, mode, file) == 0) {
455 list_move(&page->lru, dst);
465 * Unlike exported interface, "oom" parameter is added. if oom==true,
466 * oom-killer can be invoked.
468 static int __mem_cgroup_try_charge(struct mm_struct *mm,
469 gfp_t gfp_mask, struct mem_cgroup **memcg,
472 struct mem_cgroup *mem;
473 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
475 * We always charge the cgroup the mm_struct belongs to.
476 * The mm_struct's mem_cgroup changes on task migration if the
477 * thread group leader migrates. It's possible that mm is not
478 * set, if so charge the init_mm (happens for pagecache usage).
480 if (likely(!*memcg)) {
482 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
483 if (unlikely(!mem)) {
488 * For every charge from the cgroup, increment reference count
502 ret = res_counter_charge(&mem->res, PAGE_SIZE);
504 if (!do_swap_account)
506 ret = res_counter_charge(&mem->memsw, PAGE_SIZE);
509 /* mem+swap counter fails */
510 res_counter_uncharge(&mem->res, PAGE_SIZE);
513 if (!(gfp_mask & __GFP_WAIT))
516 if (try_to_free_mem_cgroup_pages(mem, gfp_mask, noswap))
520 * try_to_free_mem_cgroup_pages() might not give us a full
521 * picture of reclaim. Some pages are reclaimed and might be
522 * moved to swap cache or just unmapped from the cgroup.
523 * Check the limit again to see if the reclaim reduced the
524 * current usage of the cgroup before giving up
527 if (!do_swap_account &&
528 res_counter_check_under_limit(&mem->res))
530 if (do_swap_account &&
531 res_counter_check_under_limit(&mem->memsw))
536 mem_cgroup_out_of_memory(mem, gfp_mask);
547 * mem_cgroup_try_charge - get charge of PAGE_SIZE.
548 * @mm: an mm_struct which is charged against. (when *memcg is NULL)
549 * @gfp_mask: gfp_mask for reclaim.
550 * @memcg: a pointer to memory cgroup which is charged against.
552 * charge against memory cgroup pointed by *memcg. if *memcg == NULL, estimated
553 * memory cgroup from @mm is got and stored in *memcg.
555 * Returns 0 if success. -ENOMEM at failure.
556 * This call can invoke OOM-Killer.
559 int mem_cgroup_try_charge(struct mm_struct *mm,
560 gfp_t mask, struct mem_cgroup **memcg)
562 return __mem_cgroup_try_charge(mm, mask, memcg, true);
566 * commit a charge got by mem_cgroup_try_charge() and makes page_cgroup to be
567 * USED state. If already USED, uncharge and return.
570 static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
571 struct page_cgroup *pc,
572 enum charge_type ctype)
574 /* try_charge() can return NULL to *memcg, taking care of it. */
578 lock_page_cgroup(pc);
579 if (unlikely(PageCgroupUsed(pc))) {
580 unlock_page_cgroup(pc);
581 res_counter_uncharge(&mem->res, PAGE_SIZE);
583 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
587 pc->mem_cgroup = mem;
589 pc->flags = pcg_default_flags[ctype];
591 mem_cgroup_charge_statistics(mem, pc, true);
593 unlock_page_cgroup(pc);
597 * mem_cgroup_move_account - move account of the page
598 * @pc: page_cgroup of the page.
599 * @from: mem_cgroup which the page is moved from.
600 * @to: mem_cgroup which the page is moved to. @from != @to.
602 * The caller must confirm following.
603 * - page is not on LRU (isolate_page() is useful.)
605 * returns 0 at success,
606 * returns -EBUSY when lock is busy or "pc" is unstable.
608 * This function does "uncharge" from old cgroup but doesn't do "charge" to
609 * new cgroup. It should be done by a caller.
612 static int mem_cgroup_move_account(struct page_cgroup *pc,
613 struct mem_cgroup *from, struct mem_cgroup *to)
615 struct mem_cgroup_per_zone *from_mz, *to_mz;
619 VM_BUG_ON(from == to);
620 VM_BUG_ON(PageLRU(pc->page));
622 nid = page_cgroup_nid(pc);
623 zid = page_cgroup_zid(pc);
624 from_mz = mem_cgroup_zoneinfo(from, nid, zid);
625 to_mz = mem_cgroup_zoneinfo(to, nid, zid);
627 if (!trylock_page_cgroup(pc))
630 if (!PageCgroupUsed(pc))
633 if (pc->mem_cgroup != from)
637 res_counter_uncharge(&from->res, PAGE_SIZE);
638 mem_cgroup_charge_statistics(from, pc, false);
640 res_counter_uncharge(&from->memsw, PAGE_SIZE);
642 mem_cgroup_charge_statistics(to, pc, true);
646 unlock_page_cgroup(pc);
651 * move charges to its parent.
654 static int mem_cgroup_move_parent(struct page_cgroup *pc,
655 struct mem_cgroup *child,
658 struct page *page = pc->page;
659 struct cgroup *cg = child->css.cgroup;
660 struct cgroup *pcg = cg->parent;
661 struct mem_cgroup *parent;
669 parent = mem_cgroup_from_cont(pcg);
672 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false);
676 if (!get_page_unless_zero(page))
679 ret = isolate_lru_page(page);
684 ret = mem_cgroup_move_account(pc, child, parent);
686 /* drop extra refcnt by try_charge() (move_account increment one) */
687 css_put(&parent->css);
688 putback_lru_page(page);
693 /* uncharge if move fails */
695 res_counter_uncharge(&parent->res, PAGE_SIZE);
697 res_counter_uncharge(&parent->memsw, PAGE_SIZE);
703 * Charge the memory controller for page usage.
705 * 0 if the charge was successful
706 * < 0 if the cgroup is over its limit
708 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
709 gfp_t gfp_mask, enum charge_type ctype,
710 struct mem_cgroup *memcg)
712 struct mem_cgroup *mem;
713 struct page_cgroup *pc;
716 pc = lookup_page_cgroup(page);
717 /* can happen at boot */
723 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true);
727 __mem_cgroup_commit_charge(mem, pc, ctype);
731 int mem_cgroup_newpage_charge(struct page *page,
732 struct mm_struct *mm, gfp_t gfp_mask)
734 if (mem_cgroup_subsys.disabled)
736 if (PageCompound(page))
739 * If already mapped, we don't have to account.
740 * If page cache, page->mapping has address_space.
741 * But page->mapping may have out-of-use anon_vma pointer,
742 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
745 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
749 return mem_cgroup_charge_common(page, mm, gfp_mask,
750 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
753 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
756 if (mem_cgroup_subsys.disabled)
758 if (PageCompound(page))
761 * Corner case handling. This is called from add_to_page_cache()
762 * in usual. But some FS (shmem) precharges this page before calling it
763 * and call add_to_page_cache() with GFP_NOWAIT.
765 * For GFP_NOWAIT case, the page may be pre-charged before calling
766 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
767 * charge twice. (It works but has to pay a bit larger cost.)
769 if (!(gfp_mask & __GFP_WAIT)) {
770 struct page_cgroup *pc;
773 pc = lookup_page_cgroup(page);
776 lock_page_cgroup(pc);
777 if (PageCgroupUsed(pc)) {
778 unlock_page_cgroup(pc);
781 unlock_page_cgroup(pc);
787 if (page_is_file_cache(page))
788 return mem_cgroup_charge_common(page, mm, gfp_mask,
789 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
791 return mem_cgroup_charge_common(page, mm, gfp_mask,
792 MEM_CGROUP_CHARGE_TYPE_SHMEM, NULL);
795 int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
797 gfp_t mask, struct mem_cgroup **ptr)
799 struct mem_cgroup *mem;
802 if (mem_cgroup_subsys.disabled)
805 if (!do_swap_account)
809 * A racing thread's fault, or swapoff, may have already updated
810 * the pte, and even removed page from swap cache: return success
811 * to go on to do_swap_page()'s pte_same() test, which should fail.
813 if (!PageSwapCache(page))
816 ent.val = page_private(page);
818 mem = lookup_swap_cgroup(ent);
819 if (!mem || mem->obsolete)
822 return __mem_cgroup_try_charge(NULL, mask, ptr, true);
826 return __mem_cgroup_try_charge(mm, mask, ptr, true);
831 int mem_cgroup_cache_charge_swapin(struct page *page,
832 struct mm_struct *mm, gfp_t mask, bool locked)
836 if (mem_cgroup_subsys.disabled)
843 * If not locked, the page can be dropped from SwapCache until
846 if (PageSwapCache(page)) {
847 struct mem_cgroup *mem = NULL;
850 ent.val = page_private(page);
851 if (do_swap_account) {
852 mem = lookup_swap_cgroup(ent);
853 if (mem && mem->obsolete)
858 ret = mem_cgroup_charge_common(page, mm, mask,
859 MEM_CGROUP_CHARGE_TYPE_SHMEM, mem);
861 if (!ret && do_swap_account) {
862 /* avoid double counting */
863 mem = swap_cgroup_record(ent, NULL);
865 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
872 /* add this page(page_cgroup) to the LRU we want. */
873 mem_cgroup_lru_fixup(page);
879 void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
881 struct page_cgroup *pc;
883 if (mem_cgroup_subsys.disabled)
887 pc = lookup_page_cgroup(page);
888 __mem_cgroup_commit_charge(ptr, pc, MEM_CGROUP_CHARGE_TYPE_MAPPED);
890 * Now swap is on-memory. This means this page may be
891 * counted both as mem and swap....double count.
892 * Fix it by uncharging from memsw. This SwapCache is stable
893 * because we're still under lock_page().
895 if (do_swap_account) {
896 swp_entry_t ent = {.val = page_private(page)};
897 struct mem_cgroup *memcg;
898 memcg = swap_cgroup_record(ent, NULL);
900 /* If memcg is obsolete, memcg can be != ptr */
901 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
902 mem_cgroup_put(memcg);
906 /* add this page(page_cgroup) to the LRU we want. */
907 mem_cgroup_lru_fixup(page);
910 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
912 if (mem_cgroup_subsys.disabled)
916 res_counter_uncharge(&mem->res, PAGE_SIZE);
918 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
924 * uncharge if !page_mapped(page)
926 static struct mem_cgroup *
927 __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
929 struct page_cgroup *pc;
930 struct mem_cgroup *mem = NULL;
931 struct mem_cgroup_per_zone *mz;
933 if (mem_cgroup_subsys.disabled)
936 if (PageSwapCache(page))
940 * Check if our page_cgroup is valid
942 pc = lookup_page_cgroup(page);
943 if (unlikely(!pc || !PageCgroupUsed(pc)))
946 lock_page_cgroup(pc);
948 mem = pc->mem_cgroup;
950 if (!PageCgroupUsed(pc))
954 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
955 if (page_mapped(page))
958 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
959 if (!PageAnon(page)) { /* Shared memory */
960 if (page->mapping && !page_is_file_cache(page))
962 } else if (page_mapped(page)) /* Anon */
969 res_counter_uncharge(&mem->res, PAGE_SIZE);
970 if (do_swap_account && (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT))
971 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
973 mem_cgroup_charge_statistics(mem, pc, false);
974 ClearPageCgroupUsed(pc);
976 mz = page_cgroup_zoneinfo(pc);
977 unlock_page_cgroup(pc);
984 unlock_page_cgroup(pc);
988 void mem_cgroup_uncharge_page(struct page *page)
991 if (page_mapped(page))
993 if (page->mapping && !PageAnon(page))
995 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
998 void mem_cgroup_uncharge_cache_page(struct page *page)
1000 VM_BUG_ON(page_mapped(page));
1001 VM_BUG_ON(page->mapping);
1002 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
1006 * called from __delete_from_swap_cache() and drop "page" account.
1007 * memcg information is recorded to swap_cgroup of "ent"
1009 void mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent)
1011 struct mem_cgroup *memcg;
1013 memcg = __mem_cgroup_uncharge_common(page,
1014 MEM_CGROUP_CHARGE_TYPE_SWAPOUT);
1015 /* record memcg information */
1016 if (do_swap_account && memcg) {
1017 swap_cgroup_record(ent, memcg);
1018 mem_cgroup_get(memcg);
1022 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1024 * called from swap_entry_free(). remove record in swap_cgroup and
1025 * uncharge "memsw" account.
1027 void mem_cgroup_uncharge_swap(swp_entry_t ent)
1029 struct mem_cgroup *memcg;
1031 if (!do_swap_account)
1034 memcg = swap_cgroup_record(ent, NULL);
1036 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1037 mem_cgroup_put(memcg);
1043 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
1046 int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr)
1048 struct page_cgroup *pc;
1049 struct mem_cgroup *mem = NULL;
1052 if (mem_cgroup_subsys.disabled)
1055 pc = lookup_page_cgroup(page);
1056 lock_page_cgroup(pc);
1057 if (PageCgroupUsed(pc)) {
1058 mem = pc->mem_cgroup;
1061 unlock_page_cgroup(pc);
1064 ret = mem_cgroup_try_charge(NULL, GFP_HIGHUSER_MOVABLE, &mem);
1071 /* remove redundant charge if migration failed*/
1072 void mem_cgroup_end_migration(struct mem_cgroup *mem,
1073 struct page *oldpage, struct page *newpage)
1075 struct page *target, *unused;
1076 struct page_cgroup *pc;
1077 enum charge_type ctype;
1082 /* at migration success, oldpage->mapping is NULL. */
1083 if (oldpage->mapping) {
1091 if (PageAnon(target))
1092 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
1093 else if (page_is_file_cache(target))
1094 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
1096 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
1098 /* unused page is not on radix-tree now. */
1100 __mem_cgroup_uncharge_common(unused, ctype);
1102 pc = lookup_page_cgroup(target);
1104 * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup.
1105 * So, double-counting is effectively avoided.
1107 __mem_cgroup_commit_charge(mem, pc, ctype);
1110 * Both of oldpage and newpage are still under lock_page().
1111 * Then, we don't have to care about race in radix-tree.
1112 * But we have to be careful that this page is unmapped or not.
1114 * There is a case for !page_mapped(). At the start of
1115 * migration, oldpage was mapped. But now, it's zapped.
1116 * But we know *target* page is not freed/reused under us.
1117 * mem_cgroup_uncharge_page() does all necessary checks.
1119 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
1120 mem_cgroup_uncharge_page(target);
1124 * A call to try to shrink memory usage under specified resource controller.
1125 * This is typically used for page reclaiming for shmem for reducing side
1126 * effect of page allocation from shmem, which is used by some mem_cgroup.
1128 int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask)
1130 struct mem_cgroup *mem;
1132 int retry = MEM_CGROUP_RECLAIM_RETRIES;
1134 if (mem_cgroup_subsys.disabled)
1140 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
1141 if (unlikely(!mem)) {
1149 progress = try_to_free_mem_cgroup_pages(mem, gfp_mask, true);
1150 progress += res_counter_check_under_limit(&mem->res);
1151 } while (!progress && --retry);
1159 static DEFINE_MUTEX(set_limit_mutex);
1161 static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
1162 unsigned long long val)
1165 int retry_count = MEM_CGROUP_RECLAIM_RETRIES;
1170 while (retry_count) {
1171 if (signal_pending(current)) {
1176 * Rather than hide all in some function, I do this in
1177 * open coded manner. You see what this really does.
1178 * We have to guarantee mem->res.limit < mem->memsw.limit.
1180 mutex_lock(&set_limit_mutex);
1181 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1182 if (memswlimit < val) {
1184 mutex_unlock(&set_limit_mutex);
1187 ret = res_counter_set_limit(&memcg->res, val);
1188 mutex_unlock(&set_limit_mutex);
1193 progress = try_to_free_mem_cgroup_pages(memcg,
1194 GFP_HIGHUSER_MOVABLE, false);
1195 if (!progress) retry_count--;
1200 int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
1201 unsigned long long val)
1203 int retry_count = MEM_CGROUP_RECLAIM_RETRIES;
1204 u64 memlimit, oldusage, curusage;
1207 if (!do_swap_account)
1210 while (retry_count) {
1211 if (signal_pending(current)) {
1216 * Rather than hide all in some function, I do this in
1217 * open coded manner. You see what this really does.
1218 * We have to guarantee mem->res.limit < mem->memsw.limit.
1220 mutex_lock(&set_limit_mutex);
1221 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1222 if (memlimit > val) {
1224 mutex_unlock(&set_limit_mutex);
1227 ret = res_counter_set_limit(&memcg->memsw, val);
1228 mutex_unlock(&set_limit_mutex);
1233 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
1234 try_to_free_mem_cgroup_pages(memcg, GFP_HIGHUSER_MOVABLE, true);
1235 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
1236 if (curusage >= oldusage)
1243 * This routine traverse page_cgroup in given list and drop them all.
1244 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
1246 static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
1247 int node, int zid, enum lru_list lru)
1250 struct mem_cgroup_per_zone *mz;
1251 struct page_cgroup *pc, *busy;
1252 unsigned long flags, loop;
1253 struct list_head *list;
1256 zone = &NODE_DATA(node)->node_zones[zid];
1257 mz = mem_cgroup_zoneinfo(mem, node, zid);
1258 list = &mz->lists[lru];
1260 loop = MEM_CGROUP_ZSTAT(mz, lru);
1261 /* give some margin against EBUSY etc...*/
1266 spin_lock_irqsave(&zone->lru_lock, flags);
1267 if (list_empty(list)) {
1268 spin_unlock_irqrestore(&zone->lru_lock, flags);
1271 pc = list_entry(list->prev, struct page_cgroup, lru);
1273 list_move(&pc->lru, list);
1275 spin_unlock_irqrestore(&zone->lru_lock, flags);
1278 spin_unlock_irqrestore(&zone->lru_lock, flags);
1280 ret = mem_cgroup_move_parent(pc, mem, GFP_HIGHUSER_MOVABLE);
1284 if (ret == -EBUSY || ret == -EINVAL) {
1285 /* found lock contention or "pc" is obsolete. */
1292 if (!ret && !list_empty(list))
1298 * make mem_cgroup's charge to be 0 if there is no task.
1299 * This enables deleting this mem_cgroup.
1301 static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
1304 int node, zid, shrink;
1305 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
1306 struct cgroup *cgrp = mem->css.cgroup;
1311 /* should free all ? */
1315 while (mem->res.usage > 0) {
1317 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
1320 if (signal_pending(current))
1322 /* This is for making all *used* pages to be on LRU. */
1323 lru_add_drain_all();
1325 for_each_node_state(node, N_POSSIBLE) {
1326 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
1329 ret = mem_cgroup_force_empty_list(mem,
1338 /* it seems parent cgroup doesn't have enough mem */
1349 /* returns EBUSY if there is a task or if we come here twice. */
1350 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
1354 /* we call try-to-free pages for make this cgroup empty */
1355 lru_add_drain_all();
1356 /* try to free all pages in this cgroup */
1358 while (nr_retries && mem->res.usage > 0) {
1361 if (signal_pending(current)) {
1365 progress = try_to_free_mem_cgroup_pages(mem,
1366 GFP_HIGHUSER_MOVABLE, false);
1369 /* maybe some writeback is necessary */
1370 congestion_wait(WRITE, HZ/10);
1375 /* try move_account...there may be some *locked* pages. */
1382 int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
1384 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
1388 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
1390 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1394 type = MEMFILE_TYPE(cft->private);
1395 name = MEMFILE_ATTR(cft->private);
1398 val = res_counter_read_u64(&mem->res, name);
1401 if (do_swap_account)
1402 val = res_counter_read_u64(&mem->memsw, name);
1411 * The user of this function is...
1414 static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
1417 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
1419 unsigned long long val;
1422 type = MEMFILE_TYPE(cft->private);
1423 name = MEMFILE_ATTR(cft->private);
1426 /* This function does all necessary parse...reuse it */
1427 ret = res_counter_memparse_write_strategy(buffer, &val);
1431 ret = mem_cgroup_resize_limit(memcg, val);
1433 ret = mem_cgroup_resize_memsw_limit(memcg, val);
1436 ret = -EINVAL; /* should be BUG() ? */
1442 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
1444 struct mem_cgroup *mem;
1447 mem = mem_cgroup_from_cont(cont);
1448 type = MEMFILE_TYPE(event);
1449 name = MEMFILE_ATTR(event);
1453 res_counter_reset_max(&mem->res);
1455 res_counter_reset_max(&mem->memsw);
1459 res_counter_reset_failcnt(&mem->res);
1461 res_counter_reset_failcnt(&mem->memsw);
1467 static const struct mem_cgroup_stat_desc {
1470 } mem_cgroup_stat_desc[] = {
1471 [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, },
1472 [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, },
1473 [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, },
1474 [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, },
1477 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
1478 struct cgroup_map_cb *cb)
1480 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
1481 struct mem_cgroup_stat *stat = &mem_cont->stat;
1484 for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) {
1487 val = mem_cgroup_read_stat(stat, i);
1488 val *= mem_cgroup_stat_desc[i].unit;
1489 cb->fill(cb, mem_cgroup_stat_desc[i].msg, val);
1491 /* showing # of active pages */
1493 unsigned long active_anon, inactive_anon;
1494 unsigned long active_file, inactive_file;
1495 unsigned long unevictable;
1497 inactive_anon = mem_cgroup_get_all_zonestat(mem_cont,
1499 active_anon = mem_cgroup_get_all_zonestat(mem_cont,
1501 inactive_file = mem_cgroup_get_all_zonestat(mem_cont,
1503 active_file = mem_cgroup_get_all_zonestat(mem_cont,
1505 unevictable = mem_cgroup_get_all_zonestat(mem_cont,
1508 cb->fill(cb, "active_anon", (active_anon) * PAGE_SIZE);
1509 cb->fill(cb, "inactive_anon", (inactive_anon) * PAGE_SIZE);
1510 cb->fill(cb, "active_file", (active_file) * PAGE_SIZE);
1511 cb->fill(cb, "inactive_file", (inactive_file) * PAGE_SIZE);
1512 cb->fill(cb, "unevictable", unevictable * PAGE_SIZE);
1519 static struct cftype mem_cgroup_files[] = {
1521 .name = "usage_in_bytes",
1522 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
1523 .read_u64 = mem_cgroup_read,
1526 .name = "max_usage_in_bytes",
1527 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
1528 .trigger = mem_cgroup_reset,
1529 .read_u64 = mem_cgroup_read,
1532 .name = "limit_in_bytes",
1533 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
1534 .write_string = mem_cgroup_write,
1535 .read_u64 = mem_cgroup_read,
1539 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
1540 .trigger = mem_cgroup_reset,
1541 .read_u64 = mem_cgroup_read,
1545 .read_map = mem_control_stat_show,
1548 .name = "force_empty",
1549 .trigger = mem_cgroup_force_empty_write,
1553 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1554 static struct cftype memsw_cgroup_files[] = {
1556 .name = "memsw.usage_in_bytes",
1557 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
1558 .read_u64 = mem_cgroup_read,
1561 .name = "memsw.max_usage_in_bytes",
1562 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
1563 .trigger = mem_cgroup_reset,
1564 .read_u64 = mem_cgroup_read,
1567 .name = "memsw.limit_in_bytes",
1568 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
1569 .write_string = mem_cgroup_write,
1570 .read_u64 = mem_cgroup_read,
1573 .name = "memsw.failcnt",
1574 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
1575 .trigger = mem_cgroup_reset,
1576 .read_u64 = mem_cgroup_read,
1580 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
1582 if (!do_swap_account)
1584 return cgroup_add_files(cont, ss, memsw_cgroup_files,
1585 ARRAY_SIZE(memsw_cgroup_files));
1588 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
1594 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
1596 struct mem_cgroup_per_node *pn;
1597 struct mem_cgroup_per_zone *mz;
1599 int zone, tmp = node;
1601 * This routine is called against possible nodes.
1602 * But it's BUG to call kmalloc() against offline node.
1604 * TODO: this routine can waste much memory for nodes which will
1605 * never be onlined. It's better to use memory hotplug callback
1608 if (!node_state(node, N_NORMAL_MEMORY))
1610 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
1614 mem->info.nodeinfo[node] = pn;
1615 memset(pn, 0, sizeof(*pn));
1617 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
1618 mz = &pn->zoneinfo[zone];
1620 INIT_LIST_HEAD(&mz->lists[l]);
1625 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
1627 kfree(mem->info.nodeinfo[node]);
1630 static int mem_cgroup_size(void)
1632 int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu);
1633 return sizeof(struct mem_cgroup) + cpustat_size;
1636 static struct mem_cgroup *mem_cgroup_alloc(void)
1638 struct mem_cgroup *mem;
1639 int size = mem_cgroup_size();
1641 if (size < PAGE_SIZE)
1642 mem = kmalloc(size, GFP_KERNEL);
1644 mem = vmalloc(size);
1647 memset(mem, 0, size);
1652 * At destroying mem_cgroup, references from swap_cgroup can remain.
1653 * (scanning all at force_empty is too costly...)
1655 * Instead of clearing all references at force_empty, we remember
1656 * the number of reference from swap_cgroup and free mem_cgroup when
1657 * it goes down to 0.
1659 * When mem_cgroup is destroyed, mem->obsolete will be set to 0 and
1660 * entry which points to this memcg will be ignore at swapin.
1662 * Removal of cgroup itself succeeds regardless of refs from swap.
1665 static void mem_cgroup_free(struct mem_cgroup *mem)
1669 if (atomic_read(&mem->refcnt) > 0)
1673 for_each_node_state(node, N_POSSIBLE)
1674 free_mem_cgroup_per_zone_info(mem, node);
1676 if (mem_cgroup_size() < PAGE_SIZE)
1682 static void mem_cgroup_get(struct mem_cgroup *mem)
1684 atomic_inc(&mem->refcnt);
1687 static void mem_cgroup_put(struct mem_cgroup *mem)
1689 if (atomic_dec_and_test(&mem->refcnt)) {
1692 mem_cgroup_free(mem);
1697 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1698 static void __init enable_swap_cgroup(void)
1700 if (!mem_cgroup_subsys.disabled && really_do_swap_account)
1701 do_swap_account = 1;
1704 static void __init enable_swap_cgroup(void)
1709 static struct cgroup_subsys_state *
1710 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
1712 struct mem_cgroup *mem;
1715 mem = mem_cgroup_alloc();
1717 return ERR_PTR(-ENOMEM);
1719 res_counter_init(&mem->res);
1720 res_counter_init(&mem->memsw);
1722 for_each_node_state(node, N_POSSIBLE)
1723 if (alloc_mem_cgroup_per_zone_info(mem, node))
1726 if (cont->parent == NULL)
1727 enable_swap_cgroup();
1731 for_each_node_state(node, N_POSSIBLE)
1732 free_mem_cgroup_per_zone_info(mem, node);
1733 mem_cgroup_free(mem);
1734 return ERR_PTR(-ENOMEM);
1737 static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
1738 struct cgroup *cont)
1740 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1742 mem_cgroup_force_empty(mem, false);
1745 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
1746 struct cgroup *cont)
1748 mem_cgroup_free(mem_cgroup_from_cont(cont));
1751 static int mem_cgroup_populate(struct cgroup_subsys *ss,
1752 struct cgroup *cont)
1756 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
1757 ARRAY_SIZE(mem_cgroup_files));
1760 ret = register_memsw_files(cont, ss);
1764 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
1765 struct cgroup *cont,
1766 struct cgroup *old_cont,
1767 struct task_struct *p)
1769 struct mm_struct *mm;
1770 struct mem_cgroup *mem, *old_mem;
1772 mm = get_task_mm(p);
1776 mem = mem_cgroup_from_cont(cont);
1777 old_mem = mem_cgroup_from_cont(old_cont);
1780 * Only thread group leaders are allowed to migrate, the mm_struct is
1781 * in effect owned by the leader
1783 if (!thread_group_leader(p))
1790 struct cgroup_subsys mem_cgroup_subsys = {
1792 .subsys_id = mem_cgroup_subsys_id,
1793 .create = mem_cgroup_create,
1794 .pre_destroy = mem_cgroup_pre_destroy,
1795 .destroy = mem_cgroup_destroy,
1796 .populate = mem_cgroup_populate,
1797 .attach = mem_cgroup_move_task,
1801 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1803 static int __init disable_swap_account(char *s)
1805 really_do_swap_account = 0;
1808 __setup("noswapaccount", disable_swap_account);