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/smp.h>
25 #include <linux/page-flags.h>
26 #include <linux/backing-dev.h>
27 #include <linux/bit_spinlock.h>
28 #include <linux/rcupdate.h>
29 #include <linux/swap.h>
30 #include <linux/spinlock.h>
32 #include <linux/seq_file.h>
34 #include <asm/uaccess.h>
36 struct cgroup_subsys mem_cgroup_subsys;
37 static const int MEM_CGROUP_RECLAIM_RETRIES = 5;
40 * Statistics for memory cgroup.
42 enum mem_cgroup_stat_index {
44 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
46 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
47 MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */
49 MEM_CGROUP_STAT_NSTATS,
52 struct mem_cgroup_stat_cpu {
53 s64 count[MEM_CGROUP_STAT_NSTATS];
54 } ____cacheline_aligned_in_smp;
56 struct mem_cgroup_stat {
57 struct mem_cgroup_stat_cpu cpustat[NR_CPUS];
61 * For accounting under irq disable, no need for increment preempt count.
63 static void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat *stat,
64 enum mem_cgroup_stat_index idx, int val)
66 int cpu = smp_processor_id();
67 stat->cpustat[cpu].count[idx] += val;
70 static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat,
71 enum mem_cgroup_stat_index idx)
75 for_each_possible_cpu(cpu)
76 ret += stat->cpustat[cpu].count[idx];
81 * per-zone information in memory controller.
84 enum mem_cgroup_zstat_index {
85 MEM_CGROUP_ZSTAT_ACTIVE,
86 MEM_CGROUP_ZSTAT_INACTIVE,
91 struct mem_cgroup_per_zone {
92 struct list_head active_list;
93 struct list_head inactive_list;
94 unsigned long count[NR_MEM_CGROUP_ZSTAT];
96 /* Macro for accessing counter */
97 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
99 struct mem_cgroup_per_node {
100 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
103 struct mem_cgroup_lru_info {
104 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
108 * The memory controller data structure. The memory controller controls both
109 * page cache and RSS per cgroup. We would eventually like to provide
110 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
111 * to help the administrator determine what knobs to tune.
113 * TODO: Add a water mark for the memory controller. Reclaim will begin when
114 * we hit the water mark. May be even add a low water mark, such that
115 * no reclaim occurs from a cgroup at it's low water mark, this is
116 * a feature that will be implemented much later in the future.
119 struct cgroup_subsys_state css;
121 * the counter to account for memory usage
123 struct res_counter res;
125 * Per cgroup active and inactive list, similar to the
126 * per zone LRU lists.
128 struct mem_cgroup_lru_info info;
130 * spin_lock to protect the per cgroup LRU
133 unsigned long control_type; /* control RSS or RSS+Pagecache */
134 int prev_priority; /* for recording reclaim priority */
138 struct mem_cgroup_stat stat;
142 * We use the lower bit of the page->page_cgroup pointer as a bit spin
143 * lock. We need to ensure that page->page_cgroup is atleast two
144 * byte aligned (based on comments from Nick Piggin)
146 #define PAGE_CGROUP_LOCK_BIT 0x0
147 #define PAGE_CGROUP_LOCK (1 << PAGE_CGROUP_LOCK_BIT)
150 * A page_cgroup page is associated with every page descriptor. The
151 * page_cgroup helps us identify information about the cgroup
154 struct list_head lru; /* per cgroup LRU list */
156 struct mem_cgroup *mem_cgroup;
157 atomic_t ref_cnt; /* Helpful when pages move b/w */
158 /* mapped and cached states */
161 #define PAGE_CGROUP_FLAG_CACHE (0x1) /* charged as cache */
162 #define PAGE_CGROUP_FLAG_ACTIVE (0x2) /* page is active in this cgroup */
164 static inline int page_cgroup_nid(struct page_cgroup *pc)
166 return page_to_nid(pc->page);
169 static inline enum zone_type page_cgroup_zid(struct page_cgroup *pc)
171 return page_zonenum(pc->page);
175 MEM_CGROUP_TYPE_UNSPEC = 0,
176 MEM_CGROUP_TYPE_MAPPED,
177 MEM_CGROUP_TYPE_CACHED,
183 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
184 MEM_CGROUP_CHARGE_TYPE_MAPPED,
189 * Always modified under lru lock. Then, not necessary to preempt_disable()
191 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, int flags,
194 int val = (charge)? 1 : -1;
195 struct mem_cgroup_stat *stat = &mem->stat;
196 VM_BUG_ON(!irqs_disabled());
198 if (flags & PAGE_CGROUP_FLAG_CACHE)
199 __mem_cgroup_stat_add_safe(stat,
200 MEM_CGROUP_STAT_CACHE, val);
202 __mem_cgroup_stat_add_safe(stat, MEM_CGROUP_STAT_RSS, val);
205 static inline struct mem_cgroup_per_zone *
206 mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
208 BUG_ON(!mem->info.nodeinfo[nid]);
209 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
212 static inline struct mem_cgroup_per_zone *
213 page_cgroup_zoneinfo(struct page_cgroup *pc)
215 struct mem_cgroup *mem = pc->mem_cgroup;
216 int nid = page_cgroup_nid(pc);
217 int zid = page_cgroup_zid(pc);
219 return mem_cgroup_zoneinfo(mem, nid, zid);
222 static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem,
223 enum mem_cgroup_zstat_index idx)
226 struct mem_cgroup_per_zone *mz;
229 for_each_online_node(nid)
230 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
231 mz = mem_cgroup_zoneinfo(mem, nid, zid);
232 total += MEM_CGROUP_ZSTAT(mz, idx);
237 static struct mem_cgroup init_mem_cgroup;
240 struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
242 return container_of(cgroup_subsys_state(cont,
243 mem_cgroup_subsys_id), struct mem_cgroup,
248 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
250 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
251 struct mem_cgroup, css);
254 void mm_init_cgroup(struct mm_struct *mm, struct task_struct *p)
256 struct mem_cgroup *mem;
258 mem = mem_cgroup_from_task(p);
260 mm->mem_cgroup = mem;
263 void mm_free_cgroup(struct mm_struct *mm)
265 css_put(&mm->mem_cgroup->css);
268 static inline int page_cgroup_locked(struct page *page)
270 return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT,
274 void page_assign_page_cgroup(struct page *page, struct page_cgroup *pc)
279 * While resetting the page_cgroup we might not hold the
280 * page_cgroup lock. free_hot_cold_page() is an example
284 VM_BUG_ON(!page_cgroup_locked(page));
285 locked = (page->page_cgroup & PAGE_CGROUP_LOCK);
286 page->page_cgroup = ((unsigned long)pc | locked);
289 struct page_cgroup *page_get_page_cgroup(struct page *page)
291 return (struct page_cgroup *)
292 (page->page_cgroup & ~PAGE_CGROUP_LOCK);
295 static void __always_inline lock_page_cgroup(struct page *page)
297 bit_spin_lock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
298 VM_BUG_ON(!page_cgroup_locked(page));
301 static void __always_inline unlock_page_cgroup(struct page *page)
303 bit_spin_unlock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
307 * Tie new page_cgroup to struct page under lock_page_cgroup()
308 * This can fail if the page has been tied to a page_cgroup.
309 * If success, returns 0.
311 static int page_cgroup_assign_new_page_cgroup(struct page *page,
312 struct page_cgroup *pc)
316 lock_page_cgroup(page);
317 if (!page_get_page_cgroup(page))
318 page_assign_page_cgroup(page, pc);
319 else /* A page is tied to other pc. */
321 unlock_page_cgroup(page);
326 * Clear page->page_cgroup member under lock_page_cgroup().
327 * If given "pc" value is different from one page->page_cgroup,
328 * page->cgroup is not cleared.
329 * Returns a value of page->page_cgroup at lock taken.
330 * A can can detect failure of clearing by following
331 * clear_page_cgroup(page, pc) == pc
334 static struct page_cgroup *clear_page_cgroup(struct page *page,
335 struct page_cgroup *pc)
337 struct page_cgroup *ret;
339 lock_page_cgroup(page);
340 ret = page_get_page_cgroup(page);
341 if (likely(ret == pc))
342 page_assign_page_cgroup(page, NULL);
343 unlock_page_cgroup(page);
347 static void __mem_cgroup_remove_list(struct page_cgroup *pc)
349 int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
350 struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc);
353 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1;
355 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1;
357 mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, false);
358 list_del_init(&pc->lru);
361 static void __mem_cgroup_add_list(struct page_cgroup *pc)
363 int to = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
364 struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc);
367 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1;
368 list_add(&pc->lru, &mz->inactive_list);
370 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1;
371 list_add(&pc->lru, &mz->active_list);
373 mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, true);
376 static void __mem_cgroup_move_lists(struct page_cgroup *pc, bool active)
378 int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
379 struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc);
382 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1;
384 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1;
387 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1;
388 pc->flags |= PAGE_CGROUP_FLAG_ACTIVE;
389 list_move(&pc->lru, &mz->active_list);
391 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1;
392 pc->flags &= ~PAGE_CGROUP_FLAG_ACTIVE;
393 list_move(&pc->lru, &mz->inactive_list);
397 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
402 ret = task->mm && mm_cgroup(task->mm) == mem;
408 * This routine assumes that the appropriate zone's lru lock is already held
410 void mem_cgroup_move_lists(struct page_cgroup *pc, bool active)
412 struct mem_cgroup *mem;
416 mem = pc->mem_cgroup;
418 spin_lock(&mem->lru_lock);
419 __mem_cgroup_move_lists(pc, active);
420 spin_unlock(&mem->lru_lock);
424 * Calculate mapped_ratio under memory controller. This will be used in
425 * vmscan.c for deteremining we have to reclaim mapped pages.
427 int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem)
432 * usage is recorded in bytes. But, here, we assume the number of
433 * physical pages can be represented by "long" on any arch.
435 total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L;
436 rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
437 return (int)((rss * 100L) / total);
440 * This function is called from vmscan.c. In page reclaiming loop. balance
441 * between active and inactive list is calculated. For memory controller
442 * page reclaiming, we should use using mem_cgroup's imbalance rather than
443 * zone's global lru imbalance.
445 long mem_cgroup_reclaim_imbalance(struct mem_cgroup *mem)
447 unsigned long active, inactive;
448 /* active and inactive are the number of pages. 'long' is ok.*/
449 active = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_ACTIVE);
450 inactive = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_INACTIVE);
451 return (long) (active / (inactive + 1));
455 * prev_priority control...this will be used in memory reclaim path.
457 int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
459 return mem->prev_priority;
462 void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
464 if (priority < mem->prev_priority)
465 mem->prev_priority = priority;
468 void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
470 mem->prev_priority = priority;
474 * Calculate # of pages to be scanned in this priority/zone.
477 * priority starts from "DEF_PRIORITY" and decremented in each loop.
478 * (see include/linux/mmzone.h)
481 long mem_cgroup_calc_reclaim_active(struct mem_cgroup *mem,
482 struct zone *zone, int priority)
485 int nid = zone->zone_pgdat->node_id;
486 int zid = zone_idx(zone);
487 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid);
489 nr_active = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE);
490 return (nr_active >> priority);
493 long mem_cgroup_calc_reclaim_inactive(struct mem_cgroup *mem,
494 struct zone *zone, int priority)
497 int nid = zone->zone_pgdat->node_id;
498 int zid = zone_idx(zone);
499 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid);
501 nr_inactive = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE);
503 return (nr_inactive >> priority);
506 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
507 struct list_head *dst,
508 unsigned long *scanned, int order,
509 int mode, struct zone *z,
510 struct mem_cgroup *mem_cont,
513 unsigned long nr_taken = 0;
517 struct list_head *src;
518 struct page_cgroup *pc, *tmp;
519 int nid = z->zone_pgdat->node_id;
520 int zid = zone_idx(z);
521 struct mem_cgroup_per_zone *mz;
523 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
525 src = &mz->active_list;
527 src = &mz->inactive_list;
530 spin_lock(&mem_cont->lru_lock);
532 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
533 if (scan >= nr_to_scan)
538 if (unlikely(!PageLRU(page)))
541 if (PageActive(page) && !active) {
542 __mem_cgroup_move_lists(pc, true);
545 if (!PageActive(page) && active) {
546 __mem_cgroup_move_lists(pc, false);
551 list_move(&pc->lru, &pc_list);
553 if (__isolate_lru_page(page, mode) == 0) {
554 list_move(&page->lru, dst);
559 list_splice(&pc_list, src);
560 spin_unlock(&mem_cont->lru_lock);
567 * Charge the memory controller for page usage.
569 * 0 if the charge was successful
570 * < 0 if the cgroup is over its limit
572 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
573 gfp_t gfp_mask, enum charge_type ctype)
575 struct mem_cgroup *mem;
576 struct page_cgroup *pc;
578 unsigned long nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
581 * Should page_cgroup's go to their own slab?
582 * One could optimize the performance of the charging routine
583 * by saving a bit in the page_flags and using it as a lock
584 * to see if the cgroup page already has a page_cgroup associated
589 lock_page_cgroup(page);
590 pc = page_get_page_cgroup(page);
592 * The page_cgroup exists and
593 * the page has already been accounted.
596 if (unlikely(!atomic_inc_not_zero(&pc->ref_cnt))) {
597 /* this page is under being uncharged ? */
598 unlock_page_cgroup(page);
602 unlock_page_cgroup(page);
606 unlock_page_cgroup(page);
609 pc = kzalloc(sizeof(struct page_cgroup), gfp_mask);
614 * We always charge the cgroup the mm_struct belongs to.
615 * The mm_struct's mem_cgroup changes on task migration if the
616 * thread group leader migrates. It's possible that mm is not
617 * set, if so charge the init_mm (happens for pagecache usage).
623 mem = rcu_dereference(mm->mem_cgroup);
625 * For every charge from the cgroup, increment reference
632 * If we created the page_cgroup, we should free it on exceeding
635 while (res_counter_charge(&mem->res, PAGE_SIZE)) {
636 if (!(gfp_mask & __GFP_WAIT))
639 if (try_to_free_mem_cgroup_pages(mem, gfp_mask))
643 * try_to_free_mem_cgroup_pages() might not give us a full
644 * picture of reclaim. Some pages are reclaimed and might be
645 * moved to swap cache or just unmapped from the cgroup.
646 * Check the limit again to see if the reclaim reduced the
647 * current usage of the cgroup before giving up
649 if (res_counter_check_under_limit(&mem->res))
653 mem_cgroup_out_of_memory(mem, gfp_mask);
656 congestion_wait(WRITE, HZ/10);
659 atomic_set(&pc->ref_cnt, 1);
660 pc->mem_cgroup = mem;
662 pc->flags = PAGE_CGROUP_FLAG_ACTIVE;
663 if (ctype == MEM_CGROUP_CHARGE_TYPE_CACHE)
664 pc->flags |= PAGE_CGROUP_FLAG_CACHE;
666 if (!page || page_cgroup_assign_new_page_cgroup(page, pc)) {
668 * Another charge has been added to this page already.
669 * We take lock_page_cgroup(page) again and read
670 * page->cgroup, increment refcnt.... just retry is OK.
672 res_counter_uncharge(&mem->res, PAGE_SIZE);
680 spin_lock_irqsave(&mem->lru_lock, flags);
681 /* Update statistics vector */
682 __mem_cgroup_add_list(pc);
683 spin_unlock_irqrestore(&mem->lru_lock, flags);
694 int mem_cgroup_charge(struct page *page, struct mm_struct *mm,
697 return mem_cgroup_charge_common(page, mm, gfp_mask,
698 MEM_CGROUP_CHARGE_TYPE_MAPPED);
702 * See if the cached pages should be charged at all?
704 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
708 struct mem_cgroup *mem;
713 mem = rcu_dereference(mm->mem_cgroup);
716 if (mem->control_type == MEM_CGROUP_TYPE_ALL)
717 ret = mem_cgroup_charge_common(page, mm, gfp_mask,
718 MEM_CGROUP_CHARGE_TYPE_CACHE);
724 * Uncharging is always a welcome operation, we never complain, simply
727 void mem_cgroup_uncharge(struct page_cgroup *pc)
729 struct mem_cgroup *mem;
734 * This can handle cases when a page is not charged at all and we
735 * are switching between handling the control_type.
740 if (atomic_dec_and_test(&pc->ref_cnt)) {
743 * get page->cgroup and clear it under lock.
744 * force_empty can drop page->cgroup without checking refcnt.
746 if (clear_page_cgroup(page, pc) == pc) {
747 mem = pc->mem_cgroup;
749 res_counter_uncharge(&mem->res, PAGE_SIZE);
750 spin_lock_irqsave(&mem->lru_lock, flags);
751 __mem_cgroup_remove_list(pc);
752 spin_unlock_irqrestore(&mem->lru_lock, flags);
759 * Returns non-zero if a page (under migration) has valid page_cgroup member.
760 * Refcnt of page_cgroup is incremented.
763 int mem_cgroup_prepare_migration(struct page *page)
765 struct page_cgroup *pc;
767 lock_page_cgroup(page);
768 pc = page_get_page_cgroup(page);
769 if (pc && atomic_inc_not_zero(&pc->ref_cnt))
771 unlock_page_cgroup(page);
775 void mem_cgroup_end_migration(struct page *page)
777 struct page_cgroup *pc = page_get_page_cgroup(page);
778 mem_cgroup_uncharge(pc);
781 * We know both *page* and *newpage* are now not-on-LRU and Pg_locked.
782 * And no race with uncharge() routines because page_cgroup for *page*
783 * has extra one reference by mem_cgroup_prepare_migration.
786 void mem_cgroup_page_migration(struct page *page, struct page *newpage)
788 struct page_cgroup *pc;
789 struct mem_cgroup *mem;
792 pc = page_get_page_cgroup(page);
795 mem = pc->mem_cgroup;
796 if (clear_page_cgroup(page, pc) != pc)
799 spin_lock_irqsave(&mem->lru_lock, flags);
801 __mem_cgroup_remove_list(pc);
803 lock_page_cgroup(newpage);
804 page_assign_page_cgroup(newpage, pc);
805 unlock_page_cgroup(newpage);
806 __mem_cgroup_add_list(pc);
808 spin_unlock_irqrestore(&mem->lru_lock, flags);
813 * This routine traverse page_cgroup in given list and drop them all.
814 * This routine ignores page_cgroup->ref_cnt.
815 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
817 #define FORCE_UNCHARGE_BATCH (128)
819 mem_cgroup_force_empty_list(struct mem_cgroup *mem, struct list_head *list)
821 struct page_cgroup *pc;
826 if (list_empty(list))
829 count = FORCE_UNCHARGE_BATCH;
830 spin_lock_irqsave(&mem->lru_lock, flags);
832 while (--count && !list_empty(list)) {
833 pc = list_entry(list->prev, struct page_cgroup, lru);
835 /* Avoid race with charge */
836 atomic_set(&pc->ref_cnt, 0);
837 if (clear_page_cgroup(page, pc) == pc) {
839 res_counter_uncharge(&mem->res, PAGE_SIZE);
840 __mem_cgroup_remove_list(pc);
842 } else /* being uncharged ? ...do relax */
845 spin_unlock_irqrestore(&mem->lru_lock, flags);
846 if (!list_empty(list)) {
854 * make mem_cgroup's charge to be 0 if there is no task.
855 * This enables deleting this mem_cgroup.
858 int mem_cgroup_force_empty(struct mem_cgroup *mem)
864 * page reclaim code (kswapd etc..) will move pages between
865 ` * active_list <-> inactive_list while we don't take a lock.
866 * So, we have to do loop here until all lists are empty.
868 while (mem->res.usage > 0) {
869 if (atomic_read(&mem->css.cgroup->count) > 0)
871 for_each_node_state(node, N_POSSIBLE)
872 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
873 struct mem_cgroup_per_zone *mz;
874 mz = mem_cgroup_zoneinfo(mem, node, zid);
875 /* drop all page_cgroup in active_list */
876 mem_cgroup_force_empty_list(mem,
878 /* drop all page_cgroup in inactive_list */
879 mem_cgroup_force_empty_list(mem,
891 int mem_cgroup_write_strategy(char *buf, unsigned long long *tmp)
893 *tmp = memparse(buf, &buf);
898 * Round up the value to the closest page size
900 *tmp = ((*tmp + PAGE_SIZE - 1) >> PAGE_SHIFT) << PAGE_SHIFT;
904 static ssize_t mem_cgroup_read(struct cgroup *cont,
905 struct cftype *cft, struct file *file,
906 char __user *userbuf, size_t nbytes, loff_t *ppos)
908 return res_counter_read(&mem_cgroup_from_cont(cont)->res,
909 cft->private, userbuf, nbytes, ppos,
913 static ssize_t mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
914 struct file *file, const char __user *userbuf,
915 size_t nbytes, loff_t *ppos)
917 return res_counter_write(&mem_cgroup_from_cont(cont)->res,
918 cft->private, userbuf, nbytes, ppos,
919 mem_cgroup_write_strategy);
922 static ssize_t mem_control_type_write(struct cgroup *cont,
923 struct cftype *cft, struct file *file,
924 const char __user *userbuf,
925 size_t nbytes, loff_t *pos)
930 struct mem_cgroup *mem;
932 mem = mem_cgroup_from_cont(cont);
933 buf = kmalloc(nbytes + 1, GFP_KERNEL);
940 if (copy_from_user(buf, userbuf, nbytes))
944 tmp = simple_strtoul(buf, &end, 10);
948 if (tmp <= MEM_CGROUP_TYPE_UNSPEC || tmp >= MEM_CGROUP_TYPE_MAX)
951 mem->control_type = tmp;
959 static ssize_t mem_control_type_read(struct cgroup *cont,
961 struct file *file, char __user *userbuf,
962 size_t nbytes, loff_t *ppos)
966 struct mem_cgroup *mem;
968 mem = mem_cgroup_from_cont(cont);
970 val = mem->control_type;
971 s += sprintf(s, "%lu\n", val);
972 return simple_read_from_buffer((void __user *)userbuf, nbytes,
977 static ssize_t mem_force_empty_write(struct cgroup *cont,
978 struct cftype *cft, struct file *file,
979 const char __user *userbuf,
980 size_t nbytes, loff_t *ppos)
982 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
984 ret = mem_cgroup_force_empty(mem);
991 * Note: This should be removed if cgroup supports write-only file.
994 static ssize_t mem_force_empty_read(struct cgroup *cont,
996 struct file *file, char __user *userbuf,
997 size_t nbytes, loff_t *ppos)
1003 static const struct mem_cgroup_stat_desc {
1006 } mem_cgroup_stat_desc[] = {
1007 [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, },
1008 [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, },
1011 static int mem_control_stat_show(struct seq_file *m, void *arg)
1013 struct cgroup *cont = m->private;
1014 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
1015 struct mem_cgroup_stat *stat = &mem_cont->stat;
1018 for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) {
1021 val = mem_cgroup_read_stat(stat, i);
1022 val *= mem_cgroup_stat_desc[i].unit;
1023 seq_printf(m, "%s %lld\n", mem_cgroup_stat_desc[i].msg,
1026 /* showing # of active pages */
1028 unsigned long active, inactive;
1030 inactive = mem_cgroup_get_all_zonestat(mem_cont,
1031 MEM_CGROUP_ZSTAT_INACTIVE);
1032 active = mem_cgroup_get_all_zonestat(mem_cont,
1033 MEM_CGROUP_ZSTAT_ACTIVE);
1034 seq_printf(m, "active %ld\n", (active) * PAGE_SIZE);
1035 seq_printf(m, "inactive %ld\n", (inactive) * PAGE_SIZE);
1040 static const struct file_operations mem_control_stat_file_operations = {
1042 .llseek = seq_lseek,
1043 .release = single_release,
1046 static int mem_control_stat_open(struct inode *unused, struct file *file)
1049 struct cgroup *cont = file->f_dentry->d_parent->d_fsdata;
1051 file->f_op = &mem_control_stat_file_operations;
1052 return single_open(file, mem_control_stat_show, cont);
1057 static struct cftype mem_cgroup_files[] = {
1059 .name = "usage_in_bytes",
1060 .private = RES_USAGE,
1061 .read = mem_cgroup_read,
1064 .name = "limit_in_bytes",
1065 .private = RES_LIMIT,
1066 .write = mem_cgroup_write,
1067 .read = mem_cgroup_read,
1071 .private = RES_FAILCNT,
1072 .read = mem_cgroup_read,
1075 .name = "control_type",
1076 .write = mem_control_type_write,
1077 .read = mem_control_type_read,
1080 .name = "force_empty",
1081 .write = mem_force_empty_write,
1082 .read = mem_force_empty_read,
1086 .open = mem_control_stat_open,
1090 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
1092 struct mem_cgroup_per_node *pn;
1093 struct mem_cgroup_per_zone *mz;
1096 * This routine is called against possible nodes.
1097 * But it's BUG to call kmalloc() against offline node.
1099 * TODO: this routine can waste much memory for nodes which will
1100 * never be onlined. It's better to use memory hotplug callback
1103 if (node_state(node, N_HIGH_MEMORY))
1104 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, node);
1106 pn = kmalloc(sizeof(*pn), GFP_KERNEL);
1110 mem->info.nodeinfo[node] = pn;
1111 memset(pn, 0, sizeof(*pn));
1113 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
1114 mz = &pn->zoneinfo[zone];
1115 INIT_LIST_HEAD(&mz->active_list);
1116 INIT_LIST_HEAD(&mz->inactive_list);
1121 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
1123 kfree(mem->info.nodeinfo[node]);
1127 static struct mem_cgroup init_mem_cgroup;
1129 static struct cgroup_subsys_state *
1130 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
1132 struct mem_cgroup *mem;
1135 if (unlikely((cont->parent) == NULL)) {
1136 mem = &init_mem_cgroup;
1137 init_mm.mem_cgroup = mem;
1139 mem = kzalloc(sizeof(struct mem_cgroup), GFP_KERNEL);
1144 res_counter_init(&mem->res);
1146 spin_lock_init(&mem->lru_lock);
1147 mem->control_type = MEM_CGROUP_TYPE_ALL;
1148 memset(&mem->info, 0, sizeof(mem->info));
1150 for_each_node_state(node, N_POSSIBLE)
1151 if (alloc_mem_cgroup_per_zone_info(mem, node))
1156 for_each_node_state(node, N_POSSIBLE)
1157 free_mem_cgroup_per_zone_info(mem, node);
1158 if (cont->parent != NULL)
1163 static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
1164 struct cgroup *cont)
1166 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1167 mem_cgroup_force_empty(mem);
1170 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
1171 struct cgroup *cont)
1174 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1176 for_each_node_state(node, N_POSSIBLE)
1177 free_mem_cgroup_per_zone_info(mem, node);
1179 kfree(mem_cgroup_from_cont(cont));
1182 static int mem_cgroup_populate(struct cgroup_subsys *ss,
1183 struct cgroup *cont)
1185 return cgroup_add_files(cont, ss, mem_cgroup_files,
1186 ARRAY_SIZE(mem_cgroup_files));
1189 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
1190 struct cgroup *cont,
1191 struct cgroup *old_cont,
1192 struct task_struct *p)
1194 struct mm_struct *mm;
1195 struct mem_cgroup *mem, *old_mem;
1197 mm = get_task_mm(p);
1201 mem = mem_cgroup_from_cont(cont);
1202 old_mem = mem_cgroup_from_cont(old_cont);
1208 * Only thread group leaders are allowed to migrate, the mm_struct is
1209 * in effect owned by the leader
1211 if (p->tgid != p->pid)
1215 rcu_assign_pointer(mm->mem_cgroup, mem);
1216 css_put(&old_mem->css);
1223 struct cgroup_subsys mem_cgroup_subsys = {
1225 .subsys_id = mem_cgroup_subsys_id,
1226 .create = mem_cgroup_create,
1227 .pre_destroy = mem_cgroup_pre_destroy,
1228 .destroy = mem_cgroup_destroy,
1229 .populate = mem_cgroup_populate,
1230 .attach = mem_cgroup_move_task,