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/slab.h>
30 #include <linux/swap.h>
31 #include <linux/spinlock.h>
33 #include <linux/seq_file.h>
35 #include <asm/uaccess.h>
37 struct cgroup_subsys mem_cgroup_subsys;
38 static const int MEM_CGROUP_RECLAIM_RETRIES = 5;
39 static struct kmem_cache *page_cgroup_cache;
42 * Statistics for memory cgroup.
44 enum mem_cgroup_stat_index {
46 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
48 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
49 MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */
51 MEM_CGROUP_STAT_NSTATS,
54 struct mem_cgroup_stat_cpu {
55 s64 count[MEM_CGROUP_STAT_NSTATS];
56 } ____cacheline_aligned_in_smp;
58 struct mem_cgroup_stat {
59 struct mem_cgroup_stat_cpu cpustat[NR_CPUS];
63 * For accounting under irq disable, no need for increment preempt count.
65 static void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat *stat,
66 enum mem_cgroup_stat_index idx, int val)
68 int cpu = smp_processor_id();
69 stat->cpustat[cpu].count[idx] += val;
72 static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat,
73 enum mem_cgroup_stat_index idx)
77 for_each_possible_cpu(cpu)
78 ret += stat->cpustat[cpu].count[idx];
83 * per-zone information in memory controller.
86 enum mem_cgroup_zstat_index {
87 MEM_CGROUP_ZSTAT_ACTIVE,
88 MEM_CGROUP_ZSTAT_INACTIVE,
93 struct mem_cgroup_per_zone {
95 * spin_lock to protect the per cgroup LRU
98 struct list_head active_list;
99 struct list_head inactive_list;
100 unsigned long count[NR_MEM_CGROUP_ZSTAT];
102 /* Macro for accessing counter */
103 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
105 struct mem_cgroup_per_node {
106 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
109 struct mem_cgroup_lru_info {
110 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
114 * The memory controller data structure. The memory controller controls both
115 * page cache and RSS per cgroup. We would eventually like to provide
116 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
117 * to help the administrator determine what knobs to tune.
119 * TODO: Add a water mark for the memory controller. Reclaim will begin when
120 * we hit the water mark. May be even add a low water mark, such that
121 * no reclaim occurs from a cgroup at it's low water mark, this is
122 * a feature that will be implemented much later in the future.
125 struct cgroup_subsys_state css;
127 * the counter to account for memory usage
129 struct res_counter res;
131 * Per cgroup active and inactive list, similar to the
132 * per zone LRU lists.
134 struct mem_cgroup_lru_info info;
136 int prev_priority; /* for recording reclaim priority */
140 struct mem_cgroup_stat stat;
142 static struct mem_cgroup init_mem_cgroup;
145 * We use the lower bit of the page->page_cgroup pointer as a bit spin
146 * lock. We need to ensure that page->page_cgroup is at least two
147 * byte aligned (based on comments from Nick Piggin). But since
148 * bit_spin_lock doesn't actually set that lock bit in a non-debug
149 * uniprocessor kernel, we should avoid setting it here too.
151 #define PAGE_CGROUP_LOCK_BIT 0x0
152 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK)
153 #define PAGE_CGROUP_LOCK (1 << PAGE_CGROUP_LOCK_BIT)
155 #define PAGE_CGROUP_LOCK 0x0
159 * A page_cgroup page is associated with every page descriptor. The
160 * page_cgroup helps us identify information about the cgroup
163 struct list_head lru; /* per cgroup LRU list */
165 struct mem_cgroup *mem_cgroup;
166 int ref_cnt; /* cached, mapped, migrating */
169 #define PAGE_CGROUP_FLAG_CACHE (0x1) /* charged as cache */
170 #define PAGE_CGROUP_FLAG_ACTIVE (0x2) /* page is active in this cgroup */
172 static int page_cgroup_nid(struct page_cgroup *pc)
174 return page_to_nid(pc->page);
177 static enum zone_type page_cgroup_zid(struct page_cgroup *pc)
179 return page_zonenum(pc->page);
183 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
184 MEM_CGROUP_CHARGE_TYPE_MAPPED,
188 * Always modified under lru lock. Then, not necessary to preempt_disable()
190 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, int flags,
193 int val = (charge)? 1 : -1;
194 struct mem_cgroup_stat *stat = &mem->stat;
196 VM_BUG_ON(!irqs_disabled());
197 if (flags & PAGE_CGROUP_FLAG_CACHE)
198 __mem_cgroup_stat_add_safe(stat, MEM_CGROUP_STAT_CACHE, val);
200 __mem_cgroup_stat_add_safe(stat, MEM_CGROUP_STAT_RSS, val);
203 static struct mem_cgroup_per_zone *
204 mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
206 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
209 static struct mem_cgroup_per_zone *
210 page_cgroup_zoneinfo(struct page_cgroup *pc)
212 struct mem_cgroup *mem = pc->mem_cgroup;
213 int nid = page_cgroup_nid(pc);
214 int zid = page_cgroup_zid(pc);
216 return mem_cgroup_zoneinfo(mem, nid, zid);
219 static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem,
220 enum mem_cgroup_zstat_index idx)
223 struct mem_cgroup_per_zone *mz;
226 for_each_online_node(nid)
227 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
228 mz = mem_cgroup_zoneinfo(mem, nid, zid);
229 total += MEM_CGROUP_ZSTAT(mz, idx);
234 static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
236 return container_of(cgroup_subsys_state(cont,
237 mem_cgroup_subsys_id), struct mem_cgroup,
241 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
243 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
244 struct mem_cgroup, css);
247 static inline int page_cgroup_locked(struct page *page)
249 return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
252 static void page_assign_page_cgroup(struct page *page, struct page_cgroup *pc)
254 VM_BUG_ON(!page_cgroup_locked(page));
255 page->page_cgroup = ((unsigned long)pc | PAGE_CGROUP_LOCK);
258 struct page_cgroup *page_get_page_cgroup(struct page *page)
260 return (struct page_cgroup *) (page->page_cgroup & ~PAGE_CGROUP_LOCK);
263 static void lock_page_cgroup(struct page *page)
265 bit_spin_lock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
268 static int try_lock_page_cgroup(struct page *page)
270 return bit_spin_trylock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
273 static void unlock_page_cgroup(struct page *page)
275 bit_spin_unlock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
278 static void __mem_cgroup_remove_list(struct mem_cgroup_per_zone *mz,
279 struct page_cgroup *pc)
281 int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
284 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1;
286 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1;
288 mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, false);
289 list_del_init(&pc->lru);
292 static void __mem_cgroup_add_list(struct mem_cgroup_per_zone *mz,
293 struct page_cgroup *pc)
295 int to = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
298 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1;
299 list_add(&pc->lru, &mz->inactive_list);
301 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1;
302 list_add(&pc->lru, &mz->active_list);
304 mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, true);
307 static void __mem_cgroup_move_lists(struct page_cgroup *pc, bool active)
309 int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
310 struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc);
313 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1;
315 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1;
318 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1;
319 pc->flags |= PAGE_CGROUP_FLAG_ACTIVE;
320 list_move(&pc->lru, &mz->active_list);
322 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1;
323 pc->flags &= ~PAGE_CGROUP_FLAG_ACTIVE;
324 list_move(&pc->lru, &mz->inactive_list);
328 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
333 ret = task->mm && mm_match_cgroup(task->mm, mem);
339 * This routine assumes that the appropriate zone's lru lock is already held
341 void mem_cgroup_move_lists(struct page *page, bool active)
343 struct page_cgroup *pc;
344 struct mem_cgroup_per_zone *mz;
348 * We cannot lock_page_cgroup while holding zone's lru_lock,
349 * because other holders of lock_page_cgroup can be interrupted
350 * with an attempt to rotate_reclaimable_page. But we cannot
351 * safely get to page_cgroup without it, so just try_lock it:
352 * mem_cgroup_isolate_pages allows for page left on wrong list.
354 if (!try_lock_page_cgroup(page))
357 pc = page_get_page_cgroup(page);
359 mz = page_cgroup_zoneinfo(pc);
360 spin_lock_irqsave(&mz->lru_lock, flags);
361 __mem_cgroup_move_lists(pc, active);
362 spin_unlock_irqrestore(&mz->lru_lock, flags);
364 unlock_page_cgroup(page);
368 * Calculate mapped_ratio under memory controller. This will be used in
369 * vmscan.c for deteremining we have to reclaim mapped pages.
371 int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem)
376 * usage is recorded in bytes. But, here, we assume the number of
377 * physical pages can be represented by "long" on any arch.
379 total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L;
380 rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
381 return (int)((rss * 100L) / total);
385 * This function is called from vmscan.c. In page reclaiming loop. balance
386 * between active and inactive list is calculated. For memory controller
387 * page reclaiming, we should use using mem_cgroup's imbalance rather than
388 * zone's global lru imbalance.
390 long mem_cgroup_reclaim_imbalance(struct mem_cgroup *mem)
392 unsigned long active, inactive;
393 /* active and inactive are the number of pages. 'long' is ok.*/
394 active = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_ACTIVE);
395 inactive = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_INACTIVE);
396 return (long) (active / (inactive + 1));
400 * prev_priority control...this will be used in memory reclaim path.
402 int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
404 return mem->prev_priority;
407 void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
409 if (priority < mem->prev_priority)
410 mem->prev_priority = priority;
413 void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
415 mem->prev_priority = priority;
419 * Calculate # of pages to be scanned in this priority/zone.
422 * priority starts from "DEF_PRIORITY" and decremented in each loop.
423 * (see include/linux/mmzone.h)
426 long mem_cgroup_calc_reclaim_active(struct mem_cgroup *mem,
427 struct zone *zone, int priority)
430 int nid = zone->zone_pgdat->node_id;
431 int zid = zone_idx(zone);
432 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid);
434 nr_active = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE);
435 return (nr_active >> priority);
438 long mem_cgroup_calc_reclaim_inactive(struct mem_cgroup *mem,
439 struct zone *zone, int priority)
442 int nid = zone->zone_pgdat->node_id;
443 int zid = zone_idx(zone);
444 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid);
446 nr_inactive = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE);
447 return (nr_inactive >> priority);
450 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
451 struct list_head *dst,
452 unsigned long *scanned, int order,
453 int mode, struct zone *z,
454 struct mem_cgroup *mem_cont,
457 unsigned long nr_taken = 0;
461 struct list_head *src;
462 struct page_cgroup *pc, *tmp;
463 int nid = z->zone_pgdat->node_id;
464 int zid = zone_idx(z);
465 struct mem_cgroup_per_zone *mz;
468 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
470 src = &mz->active_list;
472 src = &mz->inactive_list;
475 spin_lock(&mz->lru_lock);
477 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
478 if (scan >= nr_to_scan)
482 if (unlikely(!PageLRU(page)))
485 if (PageActive(page) && !active) {
486 __mem_cgroup_move_lists(pc, true);
489 if (!PageActive(page) && active) {
490 __mem_cgroup_move_lists(pc, false);
495 list_move(&pc->lru, &pc_list);
497 if (__isolate_lru_page(page, mode) == 0) {
498 list_move(&page->lru, dst);
503 list_splice(&pc_list, src);
504 spin_unlock(&mz->lru_lock);
511 * Charge the memory controller for page usage.
513 * 0 if the charge was successful
514 * < 0 if the cgroup is over its limit
516 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
517 gfp_t gfp_mask, enum charge_type ctype)
519 struct mem_cgroup *mem;
520 struct page_cgroup *pc;
522 unsigned long nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
523 struct mem_cgroup_per_zone *mz;
525 if (mem_cgroup_subsys.disabled)
529 * Should page_cgroup's go to their own slab?
530 * One could optimize the performance of the charging routine
531 * by saving a bit in the page_flags and using it as a lock
532 * to see if the cgroup page already has a page_cgroup associated
536 lock_page_cgroup(page);
537 pc = page_get_page_cgroup(page);
539 * The page_cgroup exists and
540 * the page has already been accounted.
543 VM_BUG_ON(pc->page != page);
544 VM_BUG_ON(pc->ref_cnt <= 0);
547 unlock_page_cgroup(page);
550 unlock_page_cgroup(page);
552 pc = kmem_cache_zalloc(page_cgroup_cache, gfp_mask);
557 * We always charge the cgroup the mm_struct belongs to.
558 * The mm_struct's mem_cgroup changes on task migration if the
559 * thread group leader migrates. It's possible that mm is not
560 * set, if so charge the init_mm (happens for pagecache usage).
566 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
568 * For every charge from the cgroup, increment reference count
573 while (res_counter_charge(&mem->res, PAGE_SIZE)) {
574 if (!(gfp_mask & __GFP_WAIT))
577 if (try_to_free_mem_cgroup_pages(mem, gfp_mask))
581 * try_to_free_mem_cgroup_pages() might not give us a full
582 * picture of reclaim. Some pages are reclaimed and might be
583 * moved to swap cache or just unmapped from the cgroup.
584 * Check the limit again to see if the reclaim reduced the
585 * current usage of the cgroup before giving up
587 if (res_counter_check_under_limit(&mem->res))
591 mem_cgroup_out_of_memory(mem, gfp_mask);
597 pc->mem_cgroup = mem;
599 pc->flags = PAGE_CGROUP_FLAG_ACTIVE;
600 if (ctype == MEM_CGROUP_CHARGE_TYPE_CACHE)
601 pc->flags = PAGE_CGROUP_FLAG_CACHE;
603 lock_page_cgroup(page);
604 if (page_get_page_cgroup(page)) {
605 unlock_page_cgroup(page);
607 * Another charge has been added to this page already.
608 * We take lock_page_cgroup(page) again and read
609 * page->cgroup, increment refcnt.... just retry is OK.
611 res_counter_uncharge(&mem->res, PAGE_SIZE);
613 kmem_cache_free(page_cgroup_cache, pc);
616 page_assign_page_cgroup(page, pc);
618 mz = page_cgroup_zoneinfo(pc);
619 spin_lock_irqsave(&mz->lru_lock, flags);
620 __mem_cgroup_add_list(mz, pc);
621 spin_unlock_irqrestore(&mz->lru_lock, flags);
623 unlock_page_cgroup(page);
628 kmem_cache_free(page_cgroup_cache, pc);
633 int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask)
635 return mem_cgroup_charge_common(page, mm, gfp_mask,
636 MEM_CGROUP_CHARGE_TYPE_MAPPED);
639 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
644 return mem_cgroup_charge_common(page, mm, gfp_mask,
645 MEM_CGROUP_CHARGE_TYPE_CACHE);
649 * Uncharging is always a welcome operation, we never complain, simply
652 void mem_cgroup_uncharge_page(struct page *page)
654 struct page_cgroup *pc;
655 struct mem_cgroup *mem;
656 struct mem_cgroup_per_zone *mz;
659 if (mem_cgroup_subsys.disabled)
663 * Check if our page_cgroup is valid
665 lock_page_cgroup(page);
666 pc = page_get_page_cgroup(page);
670 VM_BUG_ON(pc->page != page);
671 VM_BUG_ON(pc->ref_cnt <= 0);
673 if (--(pc->ref_cnt) == 0) {
674 mz = page_cgroup_zoneinfo(pc);
675 spin_lock_irqsave(&mz->lru_lock, flags);
676 __mem_cgroup_remove_list(mz, pc);
677 spin_unlock_irqrestore(&mz->lru_lock, flags);
679 page_assign_page_cgroup(page, NULL);
680 unlock_page_cgroup(page);
682 mem = pc->mem_cgroup;
683 res_counter_uncharge(&mem->res, PAGE_SIZE);
686 kmem_cache_free(page_cgroup_cache, pc);
691 unlock_page_cgroup(page);
695 * Returns non-zero if a page (under migration) has valid page_cgroup member.
696 * Refcnt of page_cgroup is incremented.
698 int mem_cgroup_prepare_migration(struct page *page)
700 struct page_cgroup *pc;
702 if (mem_cgroup_subsys.disabled)
705 lock_page_cgroup(page);
706 pc = page_get_page_cgroup(page);
709 unlock_page_cgroup(page);
713 void mem_cgroup_end_migration(struct page *page)
715 mem_cgroup_uncharge_page(page);
719 * We know both *page* and *newpage* are now not-on-LRU and PG_locked.
720 * And no race with uncharge() routines because page_cgroup for *page*
721 * has extra one reference by mem_cgroup_prepare_migration.
723 void mem_cgroup_page_migration(struct page *page, struct page *newpage)
725 struct page_cgroup *pc;
726 struct mem_cgroup_per_zone *mz;
729 lock_page_cgroup(page);
730 pc = page_get_page_cgroup(page);
732 unlock_page_cgroup(page);
736 mz = page_cgroup_zoneinfo(pc);
737 spin_lock_irqsave(&mz->lru_lock, flags);
738 __mem_cgroup_remove_list(mz, pc);
739 spin_unlock_irqrestore(&mz->lru_lock, flags);
741 page_assign_page_cgroup(page, NULL);
742 unlock_page_cgroup(page);
745 lock_page_cgroup(newpage);
746 page_assign_page_cgroup(newpage, pc);
748 mz = page_cgroup_zoneinfo(pc);
749 spin_lock_irqsave(&mz->lru_lock, flags);
750 __mem_cgroup_add_list(mz, pc);
751 spin_unlock_irqrestore(&mz->lru_lock, flags);
753 unlock_page_cgroup(newpage);
757 * This routine traverse page_cgroup in given list and drop them all.
758 * This routine ignores page_cgroup->ref_cnt.
759 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
761 #define FORCE_UNCHARGE_BATCH (128)
762 static void mem_cgroup_force_empty_list(struct mem_cgroup *mem,
763 struct mem_cgroup_per_zone *mz,
766 struct page_cgroup *pc;
768 int count = FORCE_UNCHARGE_BATCH;
770 struct list_head *list;
773 list = &mz->active_list;
775 list = &mz->inactive_list;
777 spin_lock_irqsave(&mz->lru_lock, flags);
778 while (!list_empty(list)) {
779 pc = list_entry(list->prev, struct page_cgroup, lru);
782 spin_unlock_irqrestore(&mz->lru_lock, flags);
783 mem_cgroup_uncharge_page(page);
786 count = FORCE_UNCHARGE_BATCH;
789 spin_lock_irqsave(&mz->lru_lock, flags);
791 spin_unlock_irqrestore(&mz->lru_lock, flags);
795 * make mem_cgroup's charge to be 0 if there is no task.
796 * This enables deleting this mem_cgroup.
798 static int mem_cgroup_force_empty(struct mem_cgroup *mem)
803 if (mem_cgroup_subsys.disabled)
808 * page reclaim code (kswapd etc..) will move pages between
809 * active_list <-> inactive_list while we don't take a lock.
810 * So, we have to do loop here until all lists are empty.
812 while (mem->res.usage > 0) {
813 if (atomic_read(&mem->css.cgroup->count) > 0)
815 for_each_node_state(node, N_POSSIBLE)
816 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
817 struct mem_cgroup_per_zone *mz;
818 mz = mem_cgroup_zoneinfo(mem, node, zid);
819 /* drop all page_cgroup in active_list */
820 mem_cgroup_force_empty_list(mem, mz, 1);
821 /* drop all page_cgroup in inactive_list */
822 mem_cgroup_force_empty_list(mem, mz, 0);
831 static int mem_cgroup_write_strategy(char *buf, unsigned long long *tmp)
833 *tmp = memparse(buf, &buf);
838 * Round up the value to the closest page size
840 *tmp = ((*tmp + PAGE_SIZE - 1) >> PAGE_SHIFT) << PAGE_SHIFT;
844 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
846 return res_counter_read_u64(&mem_cgroup_from_cont(cont)->res,
850 static ssize_t mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
851 struct file *file, const char __user *userbuf,
852 size_t nbytes, loff_t *ppos)
854 return res_counter_write(&mem_cgroup_from_cont(cont)->res,
855 cft->private, userbuf, nbytes, ppos,
856 mem_cgroup_write_strategy);
859 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
861 struct mem_cgroup *mem;
863 mem = mem_cgroup_from_cont(cont);
866 res_counter_reset_max(&mem->res);
869 res_counter_reset_failcnt(&mem->res);
875 static int mem_force_empty_write(struct cgroup *cont, unsigned int event)
877 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont));
880 static const struct mem_cgroup_stat_desc {
883 } mem_cgroup_stat_desc[] = {
884 [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, },
885 [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, },
888 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
889 struct cgroup_map_cb *cb)
891 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
892 struct mem_cgroup_stat *stat = &mem_cont->stat;
895 for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) {
898 val = mem_cgroup_read_stat(stat, i);
899 val *= mem_cgroup_stat_desc[i].unit;
900 cb->fill(cb, mem_cgroup_stat_desc[i].msg, val);
902 /* showing # of active pages */
904 unsigned long active, inactive;
906 inactive = mem_cgroup_get_all_zonestat(mem_cont,
907 MEM_CGROUP_ZSTAT_INACTIVE);
908 active = mem_cgroup_get_all_zonestat(mem_cont,
909 MEM_CGROUP_ZSTAT_ACTIVE);
910 cb->fill(cb, "active", (active) * PAGE_SIZE);
911 cb->fill(cb, "inactive", (inactive) * PAGE_SIZE);
916 static struct cftype mem_cgroup_files[] = {
918 .name = "usage_in_bytes",
919 .private = RES_USAGE,
920 .read_u64 = mem_cgroup_read,
923 .name = "max_usage_in_bytes",
924 .private = RES_MAX_USAGE,
925 .trigger = mem_cgroup_reset,
926 .read_u64 = mem_cgroup_read,
929 .name = "limit_in_bytes",
930 .private = RES_LIMIT,
931 .write = mem_cgroup_write,
932 .read_u64 = mem_cgroup_read,
936 .private = RES_FAILCNT,
937 .trigger = mem_cgroup_reset,
938 .read_u64 = mem_cgroup_read,
941 .name = "force_empty",
942 .trigger = mem_force_empty_write,
946 .read_map = mem_control_stat_show,
950 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
952 struct mem_cgroup_per_node *pn;
953 struct mem_cgroup_per_zone *mz;
954 int zone, tmp = node;
956 * This routine is called against possible nodes.
957 * But it's BUG to call kmalloc() against offline node.
959 * TODO: this routine can waste much memory for nodes which will
960 * never be onlined. It's better to use memory hotplug callback
963 if (!node_state(node, N_NORMAL_MEMORY))
965 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
969 mem->info.nodeinfo[node] = pn;
970 memset(pn, 0, sizeof(*pn));
972 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
973 mz = &pn->zoneinfo[zone];
974 INIT_LIST_HEAD(&mz->active_list);
975 INIT_LIST_HEAD(&mz->inactive_list);
976 spin_lock_init(&mz->lru_lock);
981 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
983 kfree(mem->info.nodeinfo[node]);
986 static struct cgroup_subsys_state *
987 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
989 struct mem_cgroup *mem;
992 if (unlikely((cont->parent) == NULL)) {
993 mem = &init_mem_cgroup;
994 page_cgroup_cache = KMEM_CACHE(page_cgroup, SLAB_PANIC);
996 mem = kzalloc(sizeof(struct mem_cgroup), GFP_KERNEL);
1000 return ERR_PTR(-ENOMEM);
1002 res_counter_init(&mem->res);
1004 memset(&mem->info, 0, sizeof(mem->info));
1006 for_each_node_state(node, N_POSSIBLE)
1007 if (alloc_mem_cgroup_per_zone_info(mem, node))
1012 for_each_node_state(node, N_POSSIBLE)
1013 free_mem_cgroup_per_zone_info(mem, node);
1014 if (cont->parent != NULL)
1016 return ERR_PTR(-ENOMEM);
1019 static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
1020 struct cgroup *cont)
1022 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1023 mem_cgroup_force_empty(mem);
1026 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
1027 struct cgroup *cont)
1030 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1032 for_each_node_state(node, N_POSSIBLE)
1033 free_mem_cgroup_per_zone_info(mem, node);
1035 kfree(mem_cgroup_from_cont(cont));
1038 static int mem_cgroup_populate(struct cgroup_subsys *ss,
1039 struct cgroup *cont)
1041 if (mem_cgroup_subsys.disabled)
1043 return cgroup_add_files(cont, ss, mem_cgroup_files,
1044 ARRAY_SIZE(mem_cgroup_files));
1047 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
1048 struct cgroup *cont,
1049 struct cgroup *old_cont,
1050 struct task_struct *p)
1052 struct mm_struct *mm;
1053 struct mem_cgroup *mem, *old_mem;
1055 if (mem_cgroup_subsys.disabled)
1058 mm = get_task_mm(p);
1062 mem = mem_cgroup_from_cont(cont);
1063 old_mem = mem_cgroup_from_cont(old_cont);
1069 * Only thread group leaders are allowed to migrate, the mm_struct is
1070 * in effect owned by the leader
1072 if (!thread_group_leader(p))
1079 struct cgroup_subsys mem_cgroup_subsys = {
1081 .subsys_id = mem_cgroup_subsys_id,
1082 .create = mem_cgroup_create,
1083 .pre_destroy = mem_cgroup_pre_destroy,
1084 .destroy = mem_cgroup_destroy,
1085 .populate = mem_cgroup_populate,
1086 .attach = mem_cgroup_move_task,