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>
34 #include <linux/vmalloc.h>
36 #include <asm/uaccess.h>
38 struct cgroup_subsys mem_cgroup_subsys __read_mostly;
39 static struct kmem_cache *page_cgroup_cache __read_mostly;
40 #define MEM_CGROUP_RECLAIM_RETRIES 5
43 * Statistics for memory cgroup.
45 enum mem_cgroup_stat_index {
47 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
49 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
50 MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */
51 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
52 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
54 MEM_CGROUP_STAT_NSTATS,
57 struct mem_cgroup_stat_cpu {
58 s64 count[MEM_CGROUP_STAT_NSTATS];
59 } ____cacheline_aligned_in_smp;
61 struct mem_cgroup_stat {
62 struct mem_cgroup_stat_cpu cpustat[NR_CPUS];
66 * For accounting under irq disable, no need for increment preempt count.
68 static void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat *stat,
69 enum mem_cgroup_stat_index idx, int val)
71 int cpu = smp_processor_id();
72 stat->cpustat[cpu].count[idx] += val;
75 static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat,
76 enum mem_cgroup_stat_index idx)
80 for_each_possible_cpu(cpu)
81 ret += stat->cpustat[cpu].count[idx];
86 * per-zone information in memory controller.
89 enum mem_cgroup_zstat_index {
90 MEM_CGROUP_ZSTAT_ACTIVE,
91 MEM_CGROUP_ZSTAT_INACTIVE,
96 struct mem_cgroup_per_zone {
98 * spin_lock to protect the per cgroup LRU
101 struct list_head active_list;
102 struct list_head inactive_list;
103 unsigned long count[NR_MEM_CGROUP_ZSTAT];
105 /* Macro for accessing counter */
106 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
108 struct mem_cgroup_per_node {
109 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
112 struct mem_cgroup_lru_info {
113 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
117 * The memory controller data structure. The memory controller controls both
118 * page cache and RSS per cgroup. We would eventually like to provide
119 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
120 * to help the administrator determine what knobs to tune.
122 * TODO: Add a water mark for the memory controller. Reclaim will begin when
123 * we hit the water mark. May be even add a low water mark, such that
124 * no reclaim occurs from a cgroup at it's low water mark, this is
125 * a feature that will be implemented much later in the future.
128 struct cgroup_subsys_state css;
130 * the counter to account for memory usage
132 struct res_counter res;
134 * Per cgroup active and inactive list, similar to the
135 * per zone LRU lists.
137 struct mem_cgroup_lru_info info;
139 int prev_priority; /* for recording reclaim priority */
143 struct mem_cgroup_stat stat;
145 static struct mem_cgroup init_mem_cgroup;
148 * We use the lower bit of the page->page_cgroup pointer as a bit spin
149 * lock. We need to ensure that page->page_cgroup is at least two
150 * byte aligned (based on comments from Nick Piggin). But since
151 * bit_spin_lock doesn't actually set that lock bit in a non-debug
152 * uniprocessor kernel, we should avoid setting it here too.
154 #define PAGE_CGROUP_LOCK_BIT 0x0
155 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK)
156 #define PAGE_CGROUP_LOCK (1 << PAGE_CGROUP_LOCK_BIT)
158 #define PAGE_CGROUP_LOCK 0x0
162 * A page_cgroup page is associated with every page descriptor. The
163 * page_cgroup helps us identify information about the cgroup
166 struct list_head lru; /* per cgroup LRU list */
168 struct mem_cgroup *mem_cgroup;
171 #define PAGE_CGROUP_FLAG_CACHE (0x1) /* charged as cache */
172 #define PAGE_CGROUP_FLAG_ACTIVE (0x2) /* page is active in this cgroup */
174 static int page_cgroup_nid(struct page_cgroup *pc)
176 return page_to_nid(pc->page);
179 static enum zone_type page_cgroup_zid(struct page_cgroup *pc)
181 return page_zonenum(pc->page);
185 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
186 MEM_CGROUP_CHARGE_TYPE_MAPPED,
187 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
191 * Always modified under lru lock. Then, not necessary to preempt_disable()
193 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, int flags,
196 int val = (charge)? 1 : -1;
197 struct mem_cgroup_stat *stat = &mem->stat;
199 VM_BUG_ON(!irqs_disabled());
200 if (flags & PAGE_CGROUP_FLAG_CACHE)
201 __mem_cgroup_stat_add_safe(stat, MEM_CGROUP_STAT_CACHE, val);
203 __mem_cgroup_stat_add_safe(stat, MEM_CGROUP_STAT_RSS, val);
206 __mem_cgroup_stat_add_safe(stat,
207 MEM_CGROUP_STAT_PGPGIN_COUNT, 1);
209 __mem_cgroup_stat_add_safe(stat,
210 MEM_CGROUP_STAT_PGPGOUT_COUNT, 1);
213 static struct mem_cgroup_per_zone *
214 mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
216 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
219 static struct mem_cgroup_per_zone *
220 page_cgroup_zoneinfo(struct page_cgroup *pc)
222 struct mem_cgroup *mem = pc->mem_cgroup;
223 int nid = page_cgroup_nid(pc);
224 int zid = page_cgroup_zid(pc);
226 return mem_cgroup_zoneinfo(mem, nid, zid);
229 static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem,
230 enum mem_cgroup_zstat_index idx)
233 struct mem_cgroup_per_zone *mz;
236 for_each_online_node(nid)
237 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
238 mz = mem_cgroup_zoneinfo(mem, nid, zid);
239 total += MEM_CGROUP_ZSTAT(mz, idx);
244 static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
246 return container_of(cgroup_subsys_state(cont,
247 mem_cgroup_subsys_id), struct mem_cgroup,
251 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
253 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
254 struct mem_cgroup, css);
257 static inline int page_cgroup_locked(struct page *page)
259 return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
262 static void page_assign_page_cgroup(struct page *page, struct page_cgroup *pc)
264 VM_BUG_ON(!page_cgroup_locked(page));
265 page->page_cgroup = ((unsigned long)pc | PAGE_CGROUP_LOCK);
268 struct page_cgroup *page_get_page_cgroup(struct page *page)
270 return (struct page_cgroup *) (page->page_cgroup & ~PAGE_CGROUP_LOCK);
273 static void lock_page_cgroup(struct page *page)
275 bit_spin_lock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
278 static int try_lock_page_cgroup(struct page *page)
280 return bit_spin_trylock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
283 static void unlock_page_cgroup(struct page *page)
285 bit_spin_unlock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
288 static void __mem_cgroup_remove_list(struct mem_cgroup_per_zone *mz,
289 struct page_cgroup *pc)
291 int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
294 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1;
296 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1;
298 mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, false);
302 static void __mem_cgroup_add_list(struct mem_cgroup_per_zone *mz,
303 struct page_cgroup *pc)
305 int to = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
308 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1;
309 list_add(&pc->lru, &mz->inactive_list);
311 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1;
312 list_add(&pc->lru, &mz->active_list);
314 mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, true);
317 static void __mem_cgroup_move_lists(struct page_cgroup *pc, bool active)
319 int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
320 struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc);
323 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1;
325 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1;
328 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1;
329 pc->flags |= PAGE_CGROUP_FLAG_ACTIVE;
330 list_move(&pc->lru, &mz->active_list);
332 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1;
333 pc->flags &= ~PAGE_CGROUP_FLAG_ACTIVE;
334 list_move(&pc->lru, &mz->inactive_list);
338 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
343 ret = task->mm && mm_match_cgroup(task->mm, mem);
349 * This routine assumes that the appropriate zone's lru lock is already held
351 void mem_cgroup_move_lists(struct page *page, bool active)
353 struct page_cgroup *pc;
354 struct mem_cgroup_per_zone *mz;
357 if (mem_cgroup_subsys.disabled)
361 * We cannot lock_page_cgroup while holding zone's lru_lock,
362 * because other holders of lock_page_cgroup can be interrupted
363 * with an attempt to rotate_reclaimable_page. But we cannot
364 * safely get to page_cgroup without it, so just try_lock it:
365 * mem_cgroup_isolate_pages allows for page left on wrong list.
367 if (!try_lock_page_cgroup(page))
370 pc = page_get_page_cgroup(page);
372 mz = page_cgroup_zoneinfo(pc);
373 spin_lock_irqsave(&mz->lru_lock, flags);
374 __mem_cgroup_move_lists(pc, active);
375 spin_unlock_irqrestore(&mz->lru_lock, flags);
377 unlock_page_cgroup(page);
381 * Calculate mapped_ratio under memory controller. This will be used in
382 * vmscan.c for deteremining we have to reclaim mapped pages.
384 int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem)
389 * usage is recorded in bytes. But, here, we assume the number of
390 * physical pages can be represented by "long" on any arch.
392 total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L;
393 rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
394 return (int)((rss * 100L) / total);
398 * This function is called from vmscan.c. In page reclaiming loop. balance
399 * between active and inactive list is calculated. For memory controller
400 * page reclaiming, we should use using mem_cgroup's imbalance rather than
401 * zone's global lru imbalance.
403 long mem_cgroup_reclaim_imbalance(struct mem_cgroup *mem)
405 unsigned long active, inactive;
406 /* active and inactive are the number of pages. 'long' is ok.*/
407 active = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_ACTIVE);
408 inactive = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_INACTIVE);
409 return (long) (active / (inactive + 1));
413 * prev_priority control...this will be used in memory reclaim path.
415 int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
417 return mem->prev_priority;
420 void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
422 if (priority < mem->prev_priority)
423 mem->prev_priority = priority;
426 void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
428 mem->prev_priority = priority;
432 * Calculate # of pages to be scanned in this priority/zone.
435 * priority starts from "DEF_PRIORITY" and decremented in each loop.
436 * (see include/linux/mmzone.h)
439 long mem_cgroup_calc_reclaim_active(struct mem_cgroup *mem,
440 struct zone *zone, int priority)
443 int nid = zone->zone_pgdat->node_id;
444 int zid = zone_idx(zone);
445 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid);
447 nr_active = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE);
448 return (nr_active >> priority);
451 long mem_cgroup_calc_reclaim_inactive(struct mem_cgroup *mem,
452 struct zone *zone, int priority)
455 int nid = zone->zone_pgdat->node_id;
456 int zid = zone_idx(zone);
457 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid);
459 nr_inactive = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE);
460 return (nr_inactive >> priority);
463 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
464 struct list_head *dst,
465 unsigned long *scanned, int order,
466 int mode, struct zone *z,
467 struct mem_cgroup *mem_cont,
470 unsigned long nr_taken = 0;
474 struct list_head *src;
475 struct page_cgroup *pc, *tmp;
476 int nid = z->zone_pgdat->node_id;
477 int zid = zone_idx(z);
478 struct mem_cgroup_per_zone *mz;
481 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
483 src = &mz->active_list;
485 src = &mz->inactive_list;
488 spin_lock(&mz->lru_lock);
490 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
491 if (scan >= nr_to_scan)
495 if (unlikely(!PageLRU(page)))
498 if (PageActive(page) && !active) {
499 __mem_cgroup_move_lists(pc, true);
502 if (!PageActive(page) && active) {
503 __mem_cgroup_move_lists(pc, false);
508 list_move(&pc->lru, &pc_list);
510 if (__isolate_lru_page(page, mode) == 0) {
511 list_move(&page->lru, dst);
516 list_splice(&pc_list, src);
517 spin_unlock(&mz->lru_lock);
524 * Charge the memory controller for page usage.
526 * 0 if the charge was successful
527 * < 0 if the cgroup is over its limit
529 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
530 gfp_t gfp_mask, enum charge_type ctype,
531 struct mem_cgroup *memcg)
533 struct mem_cgroup *mem;
534 struct page_cgroup *pc;
536 unsigned long nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
537 struct mem_cgroup_per_zone *mz;
539 pc = kmem_cache_alloc(page_cgroup_cache, gfp_mask);
540 if (unlikely(pc == NULL))
544 * We always charge the cgroup the mm_struct belongs to.
545 * The mm_struct's mem_cgroup changes on task migration if the
546 * thread group leader migrates. It's possible that mm is not
547 * set, if so charge the init_mm (happens for pagecache usage).
549 if (likely(!memcg)) {
551 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
553 * For every charge from the cgroup, increment reference count
559 css_get(&memcg->css);
562 while (res_counter_charge(&mem->res, PAGE_SIZE)) {
563 if (!(gfp_mask & __GFP_WAIT))
566 if (try_to_free_mem_cgroup_pages(mem, gfp_mask))
570 * try_to_free_mem_cgroup_pages() might not give us a full
571 * picture of reclaim. Some pages are reclaimed and might be
572 * moved to swap cache or just unmapped from the cgroup.
573 * Check the limit again to see if the reclaim reduced the
574 * current usage of the cgroup before giving up
576 if (res_counter_check_under_limit(&mem->res))
580 mem_cgroup_out_of_memory(mem, gfp_mask);
585 pc->mem_cgroup = mem;
588 * If a page is accounted as a page cache, insert to inactive list.
589 * If anon, insert to active list.
591 if (ctype == MEM_CGROUP_CHARGE_TYPE_CACHE)
592 pc->flags = PAGE_CGROUP_FLAG_CACHE;
594 pc->flags = PAGE_CGROUP_FLAG_ACTIVE;
596 lock_page_cgroup(page);
597 if (unlikely(page_get_page_cgroup(page))) {
598 unlock_page_cgroup(page);
599 res_counter_uncharge(&mem->res, PAGE_SIZE);
601 kmem_cache_free(page_cgroup_cache, pc);
604 page_assign_page_cgroup(page, pc);
606 mz = page_cgroup_zoneinfo(pc);
607 spin_lock_irqsave(&mz->lru_lock, flags);
608 __mem_cgroup_add_list(mz, pc);
609 spin_unlock_irqrestore(&mz->lru_lock, flags);
611 unlock_page_cgroup(page);
616 kmem_cache_free(page_cgroup_cache, pc);
621 int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask)
623 if (mem_cgroup_subsys.disabled)
627 * If already mapped, we don't have to account.
628 * If page cache, page->mapping has address_space.
629 * But page->mapping may have out-of-use anon_vma pointer,
630 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
633 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
637 return mem_cgroup_charge_common(page, mm, gfp_mask,
638 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
641 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
644 if (mem_cgroup_subsys.disabled)
648 * Corner case handling. This is called from add_to_page_cache()
649 * in usual. But some FS (shmem) precharges this page before calling it
650 * and call add_to_page_cache() with GFP_NOWAIT.
652 * For GFP_NOWAIT case, the page may be pre-charged before calling
653 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
654 * charge twice. (It works but has to pay a bit larger cost.)
656 if (!(gfp_mask & __GFP_WAIT)) {
657 struct page_cgroup *pc;
659 lock_page_cgroup(page);
660 pc = page_get_page_cgroup(page);
662 VM_BUG_ON(pc->page != page);
663 VM_BUG_ON(!pc->mem_cgroup);
664 unlock_page_cgroup(page);
667 unlock_page_cgroup(page);
673 return mem_cgroup_charge_common(page, mm, gfp_mask,
674 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
678 * uncharge if !page_mapped(page)
681 __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
683 struct page_cgroup *pc;
684 struct mem_cgroup *mem;
685 struct mem_cgroup_per_zone *mz;
688 if (mem_cgroup_subsys.disabled)
692 * Check if our page_cgroup is valid
694 lock_page_cgroup(page);
695 pc = page_get_page_cgroup(page);
699 VM_BUG_ON(pc->page != page);
701 if ((ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
702 && ((pc->flags & PAGE_CGROUP_FLAG_CACHE)
703 || page_mapped(page)))
706 mz = page_cgroup_zoneinfo(pc);
707 spin_lock_irqsave(&mz->lru_lock, flags);
708 __mem_cgroup_remove_list(mz, pc);
709 spin_unlock_irqrestore(&mz->lru_lock, flags);
711 page_assign_page_cgroup(page, NULL);
712 unlock_page_cgroup(page);
714 mem = pc->mem_cgroup;
715 res_counter_uncharge(&mem->res, PAGE_SIZE);
718 kmem_cache_free(page_cgroup_cache, pc);
721 unlock_page_cgroup(page);
724 void mem_cgroup_uncharge_page(struct page *page)
726 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
729 void mem_cgroup_uncharge_cache_page(struct page *page)
731 VM_BUG_ON(page_mapped(page));
732 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
736 * Before starting migration, account against new page.
738 int mem_cgroup_prepare_migration(struct page *page, struct page *newpage)
740 struct page_cgroup *pc;
741 struct mem_cgroup *mem = NULL;
742 enum charge_type ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
745 if (mem_cgroup_subsys.disabled)
748 lock_page_cgroup(page);
749 pc = page_get_page_cgroup(page);
751 mem = pc->mem_cgroup;
753 if (pc->flags & PAGE_CGROUP_FLAG_CACHE)
754 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
756 unlock_page_cgroup(page);
758 ret = mem_cgroup_charge_common(newpage, NULL, GFP_KERNEL,
765 /* remove redundant charge if migration failed*/
766 void mem_cgroup_end_migration(struct page *newpage)
769 * At success, page->mapping is not NULL.
770 * special rollback care is necessary when
771 * 1. at migration failure. (newpage->mapping is cleared in this case)
772 * 2. the newpage was moved but not remapped again because the task
773 * exits and the newpage is obsolete. In this case, the new page
774 * may be a swapcache. So, we just call mem_cgroup_uncharge_page()
775 * always for avoiding mess. The page_cgroup will be removed if
776 * unnecessary. File cache pages is still on radix-tree. Don't
779 if (!newpage->mapping)
780 __mem_cgroup_uncharge_common(newpage,
781 MEM_CGROUP_CHARGE_TYPE_FORCE);
782 else if (PageAnon(newpage))
783 mem_cgroup_uncharge_page(newpage);
787 * A call to try to shrink memory usage under specified resource controller.
788 * This is typically used for page reclaiming for shmem for reducing side
789 * effect of page allocation from shmem, which is used by some mem_cgroup.
791 int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask)
793 struct mem_cgroup *mem;
795 int retry = MEM_CGROUP_RECLAIM_RETRIES;
797 if (mem_cgroup_subsys.disabled)
803 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
808 progress = try_to_free_mem_cgroup_pages(mem, gfp_mask);
809 } while (!progress && --retry);
817 int mem_cgroup_resize_limit(struct mem_cgroup *memcg, unsigned long long val)
820 int retry_count = MEM_CGROUP_RECLAIM_RETRIES;
824 while (res_counter_set_limit(&memcg->res, val)) {
825 if (signal_pending(current)) {
833 progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL);
842 * This routine traverse page_cgroup in given list and drop them all.
843 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
845 #define FORCE_UNCHARGE_BATCH (128)
846 static void mem_cgroup_force_empty_list(struct mem_cgroup *mem,
847 struct mem_cgroup_per_zone *mz,
850 struct page_cgroup *pc;
852 int count = FORCE_UNCHARGE_BATCH;
854 struct list_head *list;
857 list = &mz->active_list;
859 list = &mz->inactive_list;
861 spin_lock_irqsave(&mz->lru_lock, flags);
862 while (!list_empty(list)) {
863 pc = list_entry(list->prev, struct page_cgroup, lru);
866 spin_unlock_irqrestore(&mz->lru_lock, flags);
868 * Check if this page is on LRU. !LRU page can be found
869 * if it's under page migration.
872 __mem_cgroup_uncharge_common(page,
873 MEM_CGROUP_CHARGE_TYPE_FORCE);
876 count = FORCE_UNCHARGE_BATCH;
881 spin_lock_irqsave(&mz->lru_lock, flags);
883 spin_unlock_irqrestore(&mz->lru_lock, flags);
887 * make mem_cgroup's charge to be 0 if there is no task.
888 * This enables deleting this mem_cgroup.
890 static int mem_cgroup_force_empty(struct mem_cgroup *mem)
897 * page reclaim code (kswapd etc..) will move pages between
898 * active_list <-> inactive_list while we don't take a lock.
899 * So, we have to do loop here until all lists are empty.
901 while (mem->res.usage > 0) {
902 if (atomic_read(&mem->css.cgroup->count) > 0)
904 for_each_node_state(node, N_POSSIBLE)
905 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
906 struct mem_cgroup_per_zone *mz;
907 mz = mem_cgroup_zoneinfo(mem, node, zid);
908 /* drop all page_cgroup in active_list */
909 mem_cgroup_force_empty_list(mem, mz, 1);
910 /* drop all page_cgroup in inactive_list */
911 mem_cgroup_force_empty_list(mem, mz, 0);
920 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
922 return res_counter_read_u64(&mem_cgroup_from_cont(cont)->res,
926 * The user of this function is...
929 static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
932 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
933 unsigned long long val;
936 switch (cft->private) {
938 /* This function does all necessary parse...reuse it */
939 ret = res_counter_memparse_write_strategy(buffer, &val);
941 ret = mem_cgroup_resize_limit(memcg, val);
944 ret = -EINVAL; /* should be BUG() ? */
950 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
952 struct mem_cgroup *mem;
954 mem = mem_cgroup_from_cont(cont);
957 res_counter_reset_max(&mem->res);
960 res_counter_reset_failcnt(&mem->res);
966 static int mem_force_empty_write(struct cgroup *cont, unsigned int event)
968 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont));
971 static const struct mem_cgroup_stat_desc {
974 } mem_cgroup_stat_desc[] = {
975 [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, },
976 [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, },
977 [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, },
978 [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, },
981 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
982 struct cgroup_map_cb *cb)
984 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
985 struct mem_cgroup_stat *stat = &mem_cont->stat;
988 for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) {
991 val = mem_cgroup_read_stat(stat, i);
992 val *= mem_cgroup_stat_desc[i].unit;
993 cb->fill(cb, mem_cgroup_stat_desc[i].msg, val);
995 /* showing # of active pages */
997 unsigned long active, inactive;
999 inactive = mem_cgroup_get_all_zonestat(mem_cont,
1000 MEM_CGROUP_ZSTAT_INACTIVE);
1001 active = mem_cgroup_get_all_zonestat(mem_cont,
1002 MEM_CGROUP_ZSTAT_ACTIVE);
1003 cb->fill(cb, "active", (active) * PAGE_SIZE);
1004 cb->fill(cb, "inactive", (inactive) * PAGE_SIZE);
1009 static struct cftype mem_cgroup_files[] = {
1011 .name = "usage_in_bytes",
1012 .private = RES_USAGE,
1013 .read_u64 = mem_cgroup_read,
1016 .name = "max_usage_in_bytes",
1017 .private = RES_MAX_USAGE,
1018 .trigger = mem_cgroup_reset,
1019 .read_u64 = mem_cgroup_read,
1022 .name = "limit_in_bytes",
1023 .private = RES_LIMIT,
1024 .write_string = mem_cgroup_write,
1025 .read_u64 = mem_cgroup_read,
1029 .private = RES_FAILCNT,
1030 .trigger = mem_cgroup_reset,
1031 .read_u64 = mem_cgroup_read,
1034 .name = "force_empty",
1035 .trigger = mem_force_empty_write,
1039 .read_map = mem_control_stat_show,
1043 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
1045 struct mem_cgroup_per_node *pn;
1046 struct mem_cgroup_per_zone *mz;
1047 int zone, tmp = node;
1049 * This routine is called against possible nodes.
1050 * But it's BUG to call kmalloc() against offline node.
1052 * TODO: this routine can waste much memory for nodes which will
1053 * never be onlined. It's better to use memory hotplug callback
1056 if (!node_state(node, N_NORMAL_MEMORY))
1058 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
1062 mem->info.nodeinfo[node] = pn;
1063 memset(pn, 0, sizeof(*pn));
1065 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
1066 mz = &pn->zoneinfo[zone];
1067 INIT_LIST_HEAD(&mz->active_list);
1068 INIT_LIST_HEAD(&mz->inactive_list);
1069 spin_lock_init(&mz->lru_lock);
1074 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
1076 kfree(mem->info.nodeinfo[node]);
1079 static struct mem_cgroup *mem_cgroup_alloc(void)
1081 struct mem_cgroup *mem;
1083 if (sizeof(*mem) < PAGE_SIZE)
1084 mem = kmalloc(sizeof(*mem), GFP_KERNEL);
1086 mem = vmalloc(sizeof(*mem));
1089 memset(mem, 0, sizeof(*mem));
1093 static void mem_cgroup_free(struct mem_cgroup *mem)
1095 if (sizeof(*mem) < PAGE_SIZE)
1102 static struct cgroup_subsys_state *
1103 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
1105 struct mem_cgroup *mem;
1108 if (unlikely((cont->parent) == NULL)) {
1109 mem = &init_mem_cgroup;
1110 page_cgroup_cache = KMEM_CACHE(page_cgroup, SLAB_PANIC);
1112 mem = mem_cgroup_alloc();
1114 return ERR_PTR(-ENOMEM);
1117 res_counter_init(&mem->res);
1119 for_each_node_state(node, N_POSSIBLE)
1120 if (alloc_mem_cgroup_per_zone_info(mem, node))
1125 for_each_node_state(node, N_POSSIBLE)
1126 free_mem_cgroup_per_zone_info(mem, node);
1127 if (cont->parent != NULL)
1128 mem_cgroup_free(mem);
1129 return ERR_PTR(-ENOMEM);
1132 static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
1133 struct cgroup *cont)
1135 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1136 mem_cgroup_force_empty(mem);
1139 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
1140 struct cgroup *cont)
1143 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1145 for_each_node_state(node, N_POSSIBLE)
1146 free_mem_cgroup_per_zone_info(mem, node);
1148 mem_cgroup_free(mem_cgroup_from_cont(cont));
1151 static int mem_cgroup_populate(struct cgroup_subsys *ss,
1152 struct cgroup *cont)
1154 return cgroup_add_files(cont, ss, mem_cgroup_files,
1155 ARRAY_SIZE(mem_cgroup_files));
1158 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
1159 struct cgroup *cont,
1160 struct cgroup *old_cont,
1161 struct task_struct *p)
1163 struct mm_struct *mm;
1164 struct mem_cgroup *mem, *old_mem;
1166 mm = get_task_mm(p);
1170 mem = mem_cgroup_from_cont(cont);
1171 old_mem = mem_cgroup_from_cont(old_cont);
1174 * Only thread group leaders are allowed to migrate, the mm_struct is
1175 * in effect owned by the leader
1177 if (!thread_group_leader(p))
1184 struct cgroup_subsys mem_cgroup_subsys = {
1186 .subsys_id = mem_cgroup_subsys_id,
1187 .create = mem_cgroup_create,
1188 .pre_destroy = mem_cgroup_pre_destroy,
1189 .destroy = mem_cgroup_destroy,
1190 .populate = mem_cgroup_populate,
1191 .attach = mem_cgroup_move_task,