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 unsigned long count[NR_MEM_CGROUP_ZSTAT];
94 /* Macro for accessing counter */
95 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
97 struct mem_cgroup_per_node {
98 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
101 struct mem_cgroup_lru_info {
102 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
106 * The memory controller data structure. The memory controller controls both
107 * page cache and RSS per cgroup. We would eventually like to provide
108 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
109 * to help the administrator determine what knobs to tune.
111 * TODO: Add a water mark for the memory controller. Reclaim will begin when
112 * we hit the water mark. May be even add a low water mark, such that
113 * no reclaim occurs from a cgroup at it's low water mark, this is
114 * a feature that will be implemented much later in the future.
117 struct cgroup_subsys_state css;
119 * the counter to account for memory usage
121 struct res_counter res;
123 * Per cgroup active and inactive list, similar to the
124 * per zone LRU lists.
125 * TODO: Consider making these lists per zone
127 struct list_head active_list;
128 struct list_head inactive_list;
129 struct mem_cgroup_lru_info info;
131 * spin_lock to protect the per cgroup LRU
134 unsigned long control_type; /* control RSS or RSS+Pagecache */
135 int prev_priority; /* for recording reclaim priority */
139 struct mem_cgroup_stat stat;
143 * We use the lower bit of the page->page_cgroup pointer as a bit spin
144 * lock. We need to ensure that page->page_cgroup is atleast two
145 * byte aligned (based on comments from Nick Piggin)
147 #define PAGE_CGROUP_LOCK_BIT 0x0
148 #define PAGE_CGROUP_LOCK (1 << PAGE_CGROUP_LOCK_BIT)
151 * A page_cgroup page is associated with every page descriptor. The
152 * page_cgroup helps us identify information about the cgroup
155 struct list_head lru; /* per cgroup LRU list */
157 struct mem_cgroup *mem_cgroup;
158 atomic_t ref_cnt; /* Helpful when pages move b/w */
159 /* mapped and cached states */
162 #define PAGE_CGROUP_FLAG_CACHE (0x1) /* charged as cache */
163 #define PAGE_CGROUP_FLAG_ACTIVE (0x2) /* page is active in this cgroup */
165 static inline int page_cgroup_nid(struct page_cgroup *pc)
167 return page_to_nid(pc->page);
170 static inline enum zone_type page_cgroup_zid(struct page_cgroup *pc)
172 return page_zonenum(pc->page);
176 MEM_CGROUP_TYPE_UNSPEC = 0,
177 MEM_CGROUP_TYPE_MAPPED,
178 MEM_CGROUP_TYPE_CACHED,
184 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
185 MEM_CGROUP_CHARGE_TYPE_MAPPED,
190 * Always modified under lru lock. Then, not necessary to preempt_disable()
192 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, int flags,
195 int val = (charge)? 1 : -1;
196 struct mem_cgroup_stat *stat = &mem->stat;
197 VM_BUG_ON(!irqs_disabled());
199 if (flags & PAGE_CGROUP_FLAG_CACHE)
200 __mem_cgroup_stat_add_safe(stat,
201 MEM_CGROUP_STAT_CACHE, val);
203 __mem_cgroup_stat_add_safe(stat, MEM_CGROUP_STAT_RSS, val);
206 static inline struct mem_cgroup_per_zone *
207 mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
209 BUG_ON(!mem->info.nodeinfo[nid]);
210 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
213 static inline struct mem_cgroup_per_zone *
214 page_cgroup_zoneinfo(struct page_cgroup *pc)
216 struct mem_cgroup *mem = pc->mem_cgroup;
217 int nid = page_cgroup_nid(pc);
218 int zid = page_cgroup_zid(pc);
220 return mem_cgroup_zoneinfo(mem, nid, zid);
223 static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem,
224 enum mem_cgroup_zstat_index idx)
227 struct mem_cgroup_per_zone *mz;
230 for_each_online_node(nid)
231 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
232 mz = mem_cgroup_zoneinfo(mem, nid, zid);
233 total += MEM_CGROUP_ZSTAT(mz, idx);
238 static struct mem_cgroup init_mem_cgroup;
241 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,
249 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
251 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
252 struct mem_cgroup, css);
255 void mm_init_cgroup(struct mm_struct *mm, struct task_struct *p)
257 struct mem_cgroup *mem;
259 mem = mem_cgroup_from_task(p);
261 mm->mem_cgroup = mem;
264 void mm_free_cgroup(struct mm_struct *mm)
266 css_put(&mm->mem_cgroup->css);
269 static inline int page_cgroup_locked(struct page *page)
271 return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT,
275 void page_assign_page_cgroup(struct page *page, struct page_cgroup *pc)
280 * While resetting the page_cgroup we might not hold the
281 * page_cgroup lock. free_hot_cold_page() is an example
285 VM_BUG_ON(!page_cgroup_locked(page));
286 locked = (page->page_cgroup & PAGE_CGROUP_LOCK);
287 page->page_cgroup = ((unsigned long)pc | locked);
290 struct page_cgroup *page_get_page_cgroup(struct page *page)
292 return (struct page_cgroup *)
293 (page->page_cgroup & ~PAGE_CGROUP_LOCK);
296 static void __always_inline lock_page_cgroup(struct page *page)
298 bit_spin_lock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
299 VM_BUG_ON(!page_cgroup_locked(page));
302 static void __always_inline unlock_page_cgroup(struct page *page)
304 bit_spin_unlock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
308 * Tie new page_cgroup to struct page under lock_page_cgroup()
309 * This can fail if the page has been tied to a page_cgroup.
310 * If success, returns 0.
312 static int page_cgroup_assign_new_page_cgroup(struct page *page,
313 struct page_cgroup *pc)
317 lock_page_cgroup(page);
318 if (!page_get_page_cgroup(page))
319 page_assign_page_cgroup(page, pc);
320 else /* A page is tied to other pc. */
322 unlock_page_cgroup(page);
327 * Clear page->page_cgroup member under lock_page_cgroup().
328 * If given "pc" value is different from one page->page_cgroup,
329 * page->cgroup is not cleared.
330 * Returns a value of page->page_cgroup at lock taken.
331 * A can can detect failure of clearing by following
332 * clear_page_cgroup(page, pc) == pc
335 static struct page_cgroup *clear_page_cgroup(struct page *page,
336 struct page_cgroup *pc)
338 struct page_cgroup *ret;
340 lock_page_cgroup(page);
341 ret = page_get_page_cgroup(page);
342 if (likely(ret == pc))
343 page_assign_page_cgroup(page, NULL);
344 unlock_page_cgroup(page);
348 static void __mem_cgroup_remove_list(struct page_cgroup *pc)
350 int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
351 struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc);
354 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1;
356 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1;
358 mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, false);
359 list_del_init(&pc->lru);
362 static void __mem_cgroup_add_list(struct page_cgroup *pc)
364 int to = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
365 struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc);
368 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1;
369 list_add(&pc->lru, &pc->mem_cgroup->inactive_list);
371 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1;
372 list_add(&pc->lru, &pc->mem_cgroup->active_list);
374 mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, true);
377 static void __mem_cgroup_move_lists(struct page_cgroup *pc, bool active)
379 int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
380 struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc);
383 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1;
385 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1;
388 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1;
389 pc->flags |= PAGE_CGROUP_FLAG_ACTIVE;
390 list_move(&pc->lru, &pc->mem_cgroup->active_list);
392 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1;
393 pc->flags &= ~PAGE_CGROUP_FLAG_ACTIVE;
394 list_move(&pc->lru, &pc->mem_cgroup->inactive_list);
398 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
403 ret = task->mm && mm_cgroup(task->mm) == mem;
409 * This routine assumes that the appropriate zone's lru lock is already held
411 void mem_cgroup_move_lists(struct page_cgroup *pc, bool active)
413 struct mem_cgroup *mem;
417 mem = pc->mem_cgroup;
419 spin_lock(&mem->lru_lock);
420 __mem_cgroup_move_lists(pc, active);
421 spin_unlock(&mem->lru_lock);
425 * Calculate mapped_ratio under memory controller. This will be used in
426 * vmscan.c for deteremining we have to reclaim mapped pages.
428 int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem)
433 * usage is recorded in bytes. But, here, we assume the number of
434 * physical pages can be represented by "long" on any arch.
436 total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L;
437 rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
438 return (int)((rss * 100L) / total);
441 * This function is called from vmscan.c. In page reclaiming loop. balance
442 * between active and inactive list is calculated. For memory controller
443 * page reclaiming, we should use using mem_cgroup's imbalance rather than
444 * zone's global lru imbalance.
446 long mem_cgroup_reclaim_imbalance(struct mem_cgroup *mem)
448 unsigned long active, inactive;
449 /* active and inactive are the number of pages. 'long' is ok.*/
450 active = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_ACTIVE);
451 inactive = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_INACTIVE);
452 return (long) (active / (inactive + 1));
456 * prev_priority control...this will be used in memory reclaim path.
458 int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
460 return mem->prev_priority;
463 void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
465 if (priority < mem->prev_priority)
466 mem->prev_priority = priority;
469 void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
471 mem->prev_priority = priority;
475 * Calculate # of pages to be scanned in this priority/zone.
478 * priority starts from "DEF_PRIORITY" and decremented in each loop.
479 * (see include/linux/mmzone.h)
482 long mem_cgroup_calc_reclaim_active(struct mem_cgroup *mem,
483 struct zone *zone, int priority)
486 int nid = zone->zone_pgdat->node_id;
487 int zid = zone_idx(zone);
488 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid);
490 nr_active = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE);
491 return (nr_active >> priority);
494 long mem_cgroup_calc_reclaim_inactive(struct mem_cgroup *mem,
495 struct zone *zone, int priority)
498 int nid = zone->zone_pgdat->node_id;
499 int zid = zone_idx(zone);
500 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid);
502 nr_inactive = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE);
504 return (nr_inactive >> priority);
507 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
508 struct list_head *dst,
509 unsigned long *scanned, int order,
510 int mode, struct zone *z,
511 struct mem_cgroup *mem_cont,
514 unsigned long nr_taken = 0;
518 struct list_head *src;
519 struct page_cgroup *pc, *tmp;
522 src = &mem_cont->active_list;
524 src = &mem_cont->inactive_list;
526 spin_lock(&mem_cont->lru_lock);
528 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
529 if (scan >= nr_to_scan)
534 if (unlikely(!PageLRU(page)))
537 if (PageActive(page) && !active) {
538 __mem_cgroup_move_lists(pc, true);
541 if (!PageActive(page) && active) {
542 __mem_cgroup_move_lists(pc, false);
548 * TODO: make the active/inactive lists per zone
550 if (page_zone(page) != z)
554 list_move(&pc->lru, &pc_list);
556 if (__isolate_lru_page(page, mode) == 0) {
557 list_move(&page->lru, dst);
562 list_splice(&pc_list, src);
563 spin_unlock(&mem_cont->lru_lock);
570 * Charge the memory controller for page usage.
572 * 0 if the charge was successful
573 * < 0 if the cgroup is over its limit
575 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
576 gfp_t gfp_mask, enum charge_type ctype)
578 struct mem_cgroup *mem;
579 struct page_cgroup *pc;
581 unsigned long nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
584 * Should page_cgroup's go to their own slab?
585 * One could optimize the performance of the charging routine
586 * by saving a bit in the page_flags and using it as a lock
587 * to see if the cgroup page already has a page_cgroup associated
592 lock_page_cgroup(page);
593 pc = page_get_page_cgroup(page);
595 * The page_cgroup exists and
596 * the page has already been accounted.
599 if (unlikely(!atomic_inc_not_zero(&pc->ref_cnt))) {
600 /* this page is under being uncharged ? */
601 unlock_page_cgroup(page);
605 unlock_page_cgroup(page);
609 unlock_page_cgroup(page);
612 pc = kzalloc(sizeof(struct page_cgroup), gfp_mask);
617 * We always charge the cgroup the mm_struct belongs to.
618 * The mm_struct's mem_cgroup changes on task migration if the
619 * thread group leader migrates. It's possible that mm is not
620 * set, if so charge the init_mm (happens for pagecache usage).
626 mem = rcu_dereference(mm->mem_cgroup);
628 * For every charge from the cgroup, increment reference
635 * If we created the page_cgroup, we should free it on exceeding
638 while (res_counter_charge(&mem->res, PAGE_SIZE)) {
639 if (!(gfp_mask & __GFP_WAIT))
642 if (try_to_free_mem_cgroup_pages(mem, gfp_mask))
646 * try_to_free_mem_cgroup_pages() might not give us a full
647 * picture of reclaim. Some pages are reclaimed and might be
648 * moved to swap cache or just unmapped from the cgroup.
649 * Check the limit again to see if the reclaim reduced the
650 * current usage of the cgroup before giving up
652 if (res_counter_check_under_limit(&mem->res))
656 mem_cgroup_out_of_memory(mem, gfp_mask);
659 congestion_wait(WRITE, HZ/10);
662 atomic_set(&pc->ref_cnt, 1);
663 pc->mem_cgroup = mem;
665 pc->flags = PAGE_CGROUP_FLAG_ACTIVE;
666 if (ctype == MEM_CGROUP_CHARGE_TYPE_CACHE)
667 pc->flags |= PAGE_CGROUP_FLAG_CACHE;
669 if (!page || page_cgroup_assign_new_page_cgroup(page, pc)) {
671 * Another charge has been added to this page already.
672 * We take lock_page_cgroup(page) again and read
673 * page->cgroup, increment refcnt.... just retry is OK.
675 res_counter_uncharge(&mem->res, PAGE_SIZE);
683 spin_lock_irqsave(&mem->lru_lock, flags);
684 /* Update statistics vector */
685 __mem_cgroup_add_list(pc);
686 spin_unlock_irqrestore(&mem->lru_lock, flags);
697 int mem_cgroup_charge(struct page *page, struct mm_struct *mm,
700 return mem_cgroup_charge_common(page, mm, gfp_mask,
701 MEM_CGROUP_CHARGE_TYPE_MAPPED);
705 * See if the cached pages should be charged at all?
707 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
711 struct mem_cgroup *mem;
716 mem = rcu_dereference(mm->mem_cgroup);
719 if (mem->control_type == MEM_CGROUP_TYPE_ALL)
720 ret = mem_cgroup_charge_common(page, mm, gfp_mask,
721 MEM_CGROUP_CHARGE_TYPE_CACHE);
727 * Uncharging is always a welcome operation, we never complain, simply
730 void mem_cgroup_uncharge(struct page_cgroup *pc)
732 struct mem_cgroup *mem;
737 * This can handle cases when a page is not charged at all and we
738 * are switching between handling the control_type.
743 if (atomic_dec_and_test(&pc->ref_cnt)) {
746 * get page->cgroup and clear it under lock.
747 * force_empty can drop page->cgroup without checking refcnt.
749 if (clear_page_cgroup(page, pc) == pc) {
750 mem = pc->mem_cgroup;
752 res_counter_uncharge(&mem->res, PAGE_SIZE);
753 spin_lock_irqsave(&mem->lru_lock, flags);
754 __mem_cgroup_remove_list(pc);
755 spin_unlock_irqrestore(&mem->lru_lock, flags);
762 * Returns non-zero if a page (under migration) has valid page_cgroup member.
763 * Refcnt of page_cgroup is incremented.
766 int mem_cgroup_prepare_migration(struct page *page)
768 struct page_cgroup *pc;
770 lock_page_cgroup(page);
771 pc = page_get_page_cgroup(page);
772 if (pc && atomic_inc_not_zero(&pc->ref_cnt))
774 unlock_page_cgroup(page);
778 void mem_cgroup_end_migration(struct page *page)
780 struct page_cgroup *pc = page_get_page_cgroup(page);
781 mem_cgroup_uncharge(pc);
784 * We know both *page* and *newpage* are now not-on-LRU and Pg_locked.
785 * And no race with uncharge() routines because page_cgroup for *page*
786 * has extra one reference by mem_cgroup_prepare_migration.
789 void mem_cgroup_page_migration(struct page *page, struct page *newpage)
791 struct page_cgroup *pc;
792 struct mem_cgroup *mem;
795 pc = page_get_page_cgroup(page);
798 mem = pc->mem_cgroup;
799 if (clear_page_cgroup(page, pc) != pc)
802 spin_lock_irqsave(&mem->lru_lock, flags);
804 __mem_cgroup_remove_list(pc);
806 lock_page_cgroup(newpage);
807 page_assign_page_cgroup(newpage, pc);
808 unlock_page_cgroup(newpage);
809 __mem_cgroup_add_list(pc);
811 spin_unlock_irqrestore(&mem->lru_lock, flags);
816 * This routine traverse page_cgroup in given list and drop them all.
817 * This routine ignores page_cgroup->ref_cnt.
818 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
820 #define FORCE_UNCHARGE_BATCH (128)
822 mem_cgroup_force_empty_list(struct mem_cgroup *mem, struct list_head *list)
824 struct page_cgroup *pc;
830 count = FORCE_UNCHARGE_BATCH;
831 spin_lock_irqsave(&mem->lru_lock, flags);
833 while (--count && !list_empty(list)) {
834 pc = list_entry(list->prev, struct page_cgroup, lru);
836 /* Avoid race with charge */
837 atomic_set(&pc->ref_cnt, 0);
838 if (clear_page_cgroup(page, pc) == pc) {
840 res_counter_uncharge(&mem->res, PAGE_SIZE);
841 __mem_cgroup_remove_list(pc);
843 } else /* being uncharged ? ...do relax */
846 spin_unlock_irqrestore(&mem->lru_lock, flags);
847 if (!list_empty(list)) {
855 * make mem_cgroup's charge to be 0 if there is no task.
856 * This enables deleting this mem_cgroup.
859 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 (!(list_empty(&mem->active_list) &&
869 list_empty(&mem->inactive_list))) {
870 if (atomic_read(&mem->css.cgroup->count) > 0)
872 /* drop all page_cgroup in active_list */
873 mem_cgroup_force_empty_list(mem, &mem->active_list);
874 /* drop all page_cgroup in inactive_list */
875 mem_cgroup_force_empty_list(mem, &mem->inactive_list);
885 int mem_cgroup_write_strategy(char *buf, unsigned long long *tmp)
887 *tmp = memparse(buf, &buf);
892 * Round up the value to the closest page size
894 *tmp = ((*tmp + PAGE_SIZE - 1) >> PAGE_SHIFT) << PAGE_SHIFT;
898 static ssize_t mem_cgroup_read(struct cgroup *cont,
899 struct cftype *cft, struct file *file,
900 char __user *userbuf, size_t nbytes, loff_t *ppos)
902 return res_counter_read(&mem_cgroup_from_cont(cont)->res,
903 cft->private, userbuf, nbytes, ppos,
907 static ssize_t mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
908 struct file *file, const char __user *userbuf,
909 size_t nbytes, loff_t *ppos)
911 return res_counter_write(&mem_cgroup_from_cont(cont)->res,
912 cft->private, userbuf, nbytes, ppos,
913 mem_cgroup_write_strategy);
916 static ssize_t mem_control_type_write(struct cgroup *cont,
917 struct cftype *cft, struct file *file,
918 const char __user *userbuf,
919 size_t nbytes, loff_t *pos)
924 struct mem_cgroup *mem;
926 mem = mem_cgroup_from_cont(cont);
927 buf = kmalloc(nbytes + 1, GFP_KERNEL);
934 if (copy_from_user(buf, userbuf, nbytes))
938 tmp = simple_strtoul(buf, &end, 10);
942 if (tmp <= MEM_CGROUP_TYPE_UNSPEC || tmp >= MEM_CGROUP_TYPE_MAX)
945 mem->control_type = tmp;
953 static ssize_t mem_control_type_read(struct cgroup *cont,
955 struct file *file, char __user *userbuf,
956 size_t nbytes, loff_t *ppos)
960 struct mem_cgroup *mem;
962 mem = mem_cgroup_from_cont(cont);
964 val = mem->control_type;
965 s += sprintf(s, "%lu\n", val);
966 return simple_read_from_buffer((void __user *)userbuf, nbytes,
971 static ssize_t mem_force_empty_write(struct cgroup *cont,
972 struct cftype *cft, struct file *file,
973 const char __user *userbuf,
974 size_t nbytes, loff_t *ppos)
976 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
978 ret = mem_cgroup_force_empty(mem);
985 * Note: This should be removed if cgroup supports write-only file.
988 static ssize_t mem_force_empty_read(struct cgroup *cont,
990 struct file *file, char __user *userbuf,
991 size_t nbytes, loff_t *ppos)
997 static const struct mem_cgroup_stat_desc {
1000 } mem_cgroup_stat_desc[] = {
1001 [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, },
1002 [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, },
1005 static int mem_control_stat_show(struct seq_file *m, void *arg)
1007 struct cgroup *cont = m->private;
1008 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
1009 struct mem_cgroup_stat *stat = &mem_cont->stat;
1012 for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) {
1015 val = mem_cgroup_read_stat(stat, i);
1016 val *= mem_cgroup_stat_desc[i].unit;
1017 seq_printf(m, "%s %lld\n", mem_cgroup_stat_desc[i].msg,
1020 /* showing # of active pages */
1022 unsigned long active, inactive;
1024 inactive = mem_cgroup_get_all_zonestat(mem_cont,
1025 MEM_CGROUP_ZSTAT_INACTIVE);
1026 active = mem_cgroup_get_all_zonestat(mem_cont,
1027 MEM_CGROUP_ZSTAT_ACTIVE);
1028 seq_printf(m, "active %ld\n", (active) * PAGE_SIZE);
1029 seq_printf(m, "inactive %ld\n", (inactive) * PAGE_SIZE);
1034 static const struct file_operations mem_control_stat_file_operations = {
1036 .llseek = seq_lseek,
1037 .release = single_release,
1040 static int mem_control_stat_open(struct inode *unused, struct file *file)
1043 struct cgroup *cont = file->f_dentry->d_parent->d_fsdata;
1045 file->f_op = &mem_control_stat_file_operations;
1046 return single_open(file, mem_control_stat_show, cont);
1051 static struct cftype mem_cgroup_files[] = {
1053 .name = "usage_in_bytes",
1054 .private = RES_USAGE,
1055 .read = mem_cgroup_read,
1058 .name = "limit_in_bytes",
1059 .private = RES_LIMIT,
1060 .write = mem_cgroup_write,
1061 .read = mem_cgroup_read,
1065 .private = RES_FAILCNT,
1066 .read = mem_cgroup_read,
1069 .name = "control_type",
1070 .write = mem_control_type_write,
1071 .read = mem_control_type_read,
1074 .name = "force_empty",
1075 .write = mem_force_empty_write,
1076 .read = mem_force_empty_read,
1080 .open = mem_control_stat_open,
1084 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
1086 struct mem_cgroup_per_node *pn;
1088 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, node);
1091 mem->info.nodeinfo[node] = pn;
1092 memset(pn, 0, sizeof(*pn));
1096 static struct mem_cgroup init_mem_cgroup;
1098 static struct cgroup_subsys_state *
1099 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
1101 struct mem_cgroup *mem;
1104 if (unlikely((cont->parent) == NULL)) {
1105 mem = &init_mem_cgroup;
1106 init_mm.mem_cgroup = mem;
1108 mem = kzalloc(sizeof(struct mem_cgroup), GFP_KERNEL);
1113 res_counter_init(&mem->res);
1114 INIT_LIST_HEAD(&mem->active_list);
1115 INIT_LIST_HEAD(&mem->inactive_list);
1116 spin_lock_init(&mem->lru_lock);
1117 mem->control_type = MEM_CGROUP_TYPE_ALL;
1118 memset(&mem->info, 0, sizeof(mem->info));
1120 for_each_node_state(node, N_POSSIBLE)
1121 if (alloc_mem_cgroup_per_zone_info(mem, node))
1126 for_each_node_state(node, N_POSSIBLE)
1127 kfree(mem->info.nodeinfo[node]);
1128 if (cont->parent != NULL)
1133 static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
1134 struct cgroup *cont)
1136 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1137 mem_cgroup_force_empty(mem);
1140 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
1141 struct cgroup *cont)
1144 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1146 for_each_node_state(node, N_POSSIBLE)
1147 kfree(mem->info.nodeinfo[node]);
1149 kfree(mem_cgroup_from_cont(cont));
1152 static int mem_cgroup_populate(struct cgroup_subsys *ss,
1153 struct cgroup *cont)
1155 return cgroup_add_files(cont, ss, mem_cgroup_files,
1156 ARRAY_SIZE(mem_cgroup_files));
1159 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
1160 struct cgroup *cont,
1161 struct cgroup *old_cont,
1162 struct task_struct *p)
1164 struct mm_struct *mm;
1165 struct mem_cgroup *mem, *old_mem;
1167 mm = get_task_mm(p);
1171 mem = mem_cgroup_from_cont(cont);
1172 old_mem = mem_cgroup_from_cont(old_cont);
1178 * Only thread group leaders are allowed to migrate, the mm_struct is
1179 * in effect owned by the leader
1181 if (p->tgid != p->pid)
1185 rcu_assign_pointer(mm->mem_cgroup, mem);
1186 css_put(&old_mem->css);
1193 struct cgroup_subsys mem_cgroup_subsys = {
1195 .subsys_id = mem_cgroup_subsys_id,
1196 .create = mem_cgroup_create,
1197 .pre_destroy = mem_cgroup_pre_destroy,
1198 .destroy = mem_cgroup_destroy,
1199 .populate = mem_cgroup_populate,
1200 .attach = mem_cgroup_move_task,