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>
35 #include <linux/mm_inline.h>
37 #include <asm/uaccess.h>
39 struct cgroup_subsys mem_cgroup_subsys __read_mostly
;
40 static struct kmem_cache
*page_cgroup_cache __read_mostly
;
41 #define MEM_CGROUP_RECLAIM_RETRIES 5
44 * Statistics for memory cgroup.
46 enum mem_cgroup_stat_index
{
48 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
50 MEM_CGROUP_STAT_CACHE
, /* # of pages charged as cache */
51 MEM_CGROUP_STAT_RSS
, /* # of pages charged as rss */
52 MEM_CGROUP_STAT_PGPGIN_COUNT
, /* # of pages paged in */
53 MEM_CGROUP_STAT_PGPGOUT_COUNT
, /* # of pages paged out */
55 MEM_CGROUP_STAT_NSTATS
,
58 struct mem_cgroup_stat_cpu
{
59 s64 count
[MEM_CGROUP_STAT_NSTATS
];
60 } ____cacheline_aligned_in_smp
;
62 struct mem_cgroup_stat
{
63 struct mem_cgroup_stat_cpu cpustat
[NR_CPUS
];
67 * For accounting under irq disable, no need for increment preempt count.
69 static void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat
*stat
,
70 enum mem_cgroup_stat_index idx
, int val
)
72 int cpu
= smp_processor_id();
73 stat
->cpustat
[cpu
].count
[idx
] += val
;
76 static s64
mem_cgroup_read_stat(struct mem_cgroup_stat
*stat
,
77 enum mem_cgroup_stat_index idx
)
81 for_each_possible_cpu(cpu
)
82 ret
+= stat
->cpustat
[cpu
].count
[idx
];
87 * per-zone information in memory controller.
89 struct mem_cgroup_per_zone
{
91 * spin_lock to protect the per cgroup LRU
94 struct list_head lists
[NR_LRU_LISTS
];
95 unsigned long count
[NR_LRU_LISTS
];
97 /* Macro for accessing counter */
98 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
100 struct mem_cgroup_per_node
{
101 struct mem_cgroup_per_zone zoneinfo
[MAX_NR_ZONES
];
104 struct mem_cgroup_lru_info
{
105 struct mem_cgroup_per_node
*nodeinfo
[MAX_NUMNODES
];
109 * The memory controller data structure. The memory controller controls both
110 * page cache and RSS per cgroup. We would eventually like to provide
111 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
112 * to help the administrator determine what knobs to tune.
114 * TODO: Add a water mark for the memory controller. Reclaim will begin when
115 * we hit the water mark. May be even add a low water mark, such that
116 * no reclaim occurs from a cgroup at it's low water mark, this is
117 * a feature that will be implemented much later in the future.
120 struct cgroup_subsys_state css
;
122 * the counter to account for memory usage
124 struct res_counter res
;
126 * Per cgroup active and inactive list, similar to the
127 * per zone LRU lists.
129 struct mem_cgroup_lru_info info
;
131 int prev_priority
; /* for recording reclaim priority */
135 struct mem_cgroup_stat stat
;
137 static struct mem_cgroup init_mem_cgroup
;
140 * We use the lower bit of the page->page_cgroup pointer as a bit spin
141 * lock. We need to ensure that page->page_cgroup is at least two
142 * byte aligned (based on comments from Nick Piggin). But since
143 * bit_spin_lock doesn't actually set that lock bit in a non-debug
144 * uniprocessor kernel, we should avoid setting it here too.
146 #define PAGE_CGROUP_LOCK_BIT 0x0
147 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK)
148 #define PAGE_CGROUP_LOCK (1 << PAGE_CGROUP_LOCK_BIT)
150 #define PAGE_CGROUP_LOCK 0x0
154 * A page_cgroup page is associated with every page descriptor. The
155 * page_cgroup helps us identify information about the cgroup
158 struct list_head lru
; /* per cgroup LRU list */
160 struct mem_cgroup
*mem_cgroup
;
163 #define PAGE_CGROUP_FLAG_CACHE (0x1) /* charged as cache */
164 #define PAGE_CGROUP_FLAG_ACTIVE (0x2) /* page is active in this cgroup */
165 #define PAGE_CGROUP_FLAG_FILE (0x4) /* page is file system backed */
167 static int page_cgroup_nid(struct page_cgroup
*pc
)
169 return page_to_nid(pc
->page
);
172 static enum zone_type
page_cgroup_zid(struct page_cgroup
*pc
)
174 return page_zonenum(pc
->page
);
178 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
179 MEM_CGROUP_CHARGE_TYPE_MAPPED
,
180 MEM_CGROUP_CHARGE_TYPE_FORCE
, /* used by force_empty */
181 MEM_CGROUP_CHARGE_TYPE_SHMEM
, /* used by page migration of shmem */
185 * Always modified under lru lock. Then, not necessary to preempt_disable()
187 static void mem_cgroup_charge_statistics(struct mem_cgroup
*mem
, int flags
,
190 int val
= (charge
)? 1 : -1;
191 struct mem_cgroup_stat
*stat
= &mem
->stat
;
193 VM_BUG_ON(!irqs_disabled());
194 if (flags
& PAGE_CGROUP_FLAG_CACHE
)
195 __mem_cgroup_stat_add_safe(stat
, MEM_CGROUP_STAT_CACHE
, val
);
197 __mem_cgroup_stat_add_safe(stat
, MEM_CGROUP_STAT_RSS
, val
);
200 __mem_cgroup_stat_add_safe(stat
,
201 MEM_CGROUP_STAT_PGPGIN_COUNT
, 1);
203 __mem_cgroup_stat_add_safe(stat
,
204 MEM_CGROUP_STAT_PGPGOUT_COUNT
, 1);
207 static struct mem_cgroup_per_zone
*
208 mem_cgroup_zoneinfo(struct mem_cgroup
*mem
, int nid
, int zid
)
210 return &mem
->info
.nodeinfo
[nid
]->zoneinfo
[zid
];
213 static 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
,
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
*mem_cgroup_from_cont(struct cgroup
*cont
)
240 return container_of(cgroup_subsys_state(cont
,
241 mem_cgroup_subsys_id
), struct mem_cgroup
,
245 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
248 * mm_update_next_owner() may clear mm->owner to NULL
249 * if it races with swapoff, page migration, etc.
250 * So this can be called with p == NULL.
255 return container_of(task_subsys_state(p
, mem_cgroup_subsys_id
),
256 struct mem_cgroup
, css
);
259 static inline int page_cgroup_locked(struct page
*page
)
261 return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT
, &page
->page_cgroup
);
264 static void page_assign_page_cgroup(struct page
*page
, struct page_cgroup
*pc
)
266 VM_BUG_ON(!page_cgroup_locked(page
));
267 page
->page_cgroup
= ((unsigned long)pc
| PAGE_CGROUP_LOCK
);
270 struct page_cgroup
*page_get_page_cgroup(struct page
*page
)
272 return (struct page_cgroup
*) (page
->page_cgroup
& ~PAGE_CGROUP_LOCK
);
275 static void lock_page_cgroup(struct page
*page
)
277 bit_spin_lock(PAGE_CGROUP_LOCK_BIT
, &page
->page_cgroup
);
280 static int try_lock_page_cgroup(struct page
*page
)
282 return bit_spin_trylock(PAGE_CGROUP_LOCK_BIT
, &page
->page_cgroup
);
285 static void unlock_page_cgroup(struct page
*page
)
287 bit_spin_unlock(PAGE_CGROUP_LOCK_BIT
, &page
->page_cgroup
);
290 static void __mem_cgroup_remove_list(struct mem_cgroup_per_zone
*mz
,
291 struct page_cgroup
*pc
)
295 if (pc
->flags
& PAGE_CGROUP_FLAG_ACTIVE
)
297 if (pc
->flags
& PAGE_CGROUP_FLAG_FILE
)
300 MEM_CGROUP_ZSTAT(mz
, lru
) -= 1;
302 mem_cgroup_charge_statistics(pc
->mem_cgroup
, pc
->flags
, false);
306 static void __mem_cgroup_add_list(struct mem_cgroup_per_zone
*mz
,
307 struct page_cgroup
*pc
)
311 if (pc
->flags
& PAGE_CGROUP_FLAG_ACTIVE
)
313 if (pc
->flags
& PAGE_CGROUP_FLAG_FILE
)
316 MEM_CGROUP_ZSTAT(mz
, lru
) += 1;
317 list_add(&pc
->lru
, &mz
->lists
[lru
]);
319 mem_cgroup_charge_statistics(pc
->mem_cgroup
, pc
->flags
, true);
322 static void __mem_cgroup_move_lists(struct page_cgroup
*pc
, bool active
)
324 struct mem_cgroup_per_zone
*mz
= page_cgroup_zoneinfo(pc
);
325 int from
= pc
->flags
& PAGE_CGROUP_FLAG_ACTIVE
;
326 int file
= pc
->flags
& PAGE_CGROUP_FLAG_FILE
;
327 int lru
= LRU_FILE
* !!file
+ !!from
;
329 MEM_CGROUP_ZSTAT(mz
, lru
) -= 1;
332 pc
->flags
|= PAGE_CGROUP_FLAG_ACTIVE
;
334 pc
->flags
&= ~PAGE_CGROUP_FLAG_ACTIVE
;
336 lru
= LRU_FILE
* !!file
+ !!active
;
337 MEM_CGROUP_ZSTAT(mz
, lru
) += 1;
338 list_move(&pc
->lru
, &mz
->lists
[lru
]);
341 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*mem
)
346 ret
= task
->mm
&& mm_match_cgroup(task
->mm
, mem
);
352 * This routine assumes that the appropriate zone's lru lock is already held
354 void mem_cgroup_move_lists(struct page
*page
, bool active
)
356 struct page_cgroup
*pc
;
357 struct mem_cgroup_per_zone
*mz
;
360 if (mem_cgroup_subsys
.disabled
)
364 * We cannot lock_page_cgroup while holding zone's lru_lock,
365 * because other holders of lock_page_cgroup can be interrupted
366 * with an attempt to rotate_reclaimable_page. But we cannot
367 * safely get to page_cgroup without it, so just try_lock it:
368 * mem_cgroup_isolate_pages allows for page left on wrong list.
370 if (!try_lock_page_cgroup(page
))
373 pc
= page_get_page_cgroup(page
);
375 mz
= page_cgroup_zoneinfo(pc
);
376 spin_lock_irqsave(&mz
->lru_lock
, flags
);
377 __mem_cgroup_move_lists(pc
, active
);
378 spin_unlock_irqrestore(&mz
->lru_lock
, flags
);
380 unlock_page_cgroup(page
);
384 * Calculate mapped_ratio under memory controller. This will be used in
385 * vmscan.c for deteremining we have to reclaim mapped pages.
387 int mem_cgroup_calc_mapped_ratio(struct mem_cgroup
*mem
)
392 * usage is recorded in bytes. But, here, we assume the number of
393 * physical pages can be represented by "long" on any arch.
395 total
= (long) (mem
->res
.usage
>> PAGE_SHIFT
) + 1L;
396 rss
= (long)mem_cgroup_read_stat(&mem
->stat
, MEM_CGROUP_STAT_RSS
);
397 return (int)((rss
* 100L) / total
);
401 * prev_priority control...this will be used in memory reclaim path.
403 int mem_cgroup_get_reclaim_priority(struct mem_cgroup
*mem
)
405 return mem
->prev_priority
;
408 void mem_cgroup_note_reclaim_priority(struct mem_cgroup
*mem
, int priority
)
410 if (priority
< mem
->prev_priority
)
411 mem
->prev_priority
= priority
;
414 void mem_cgroup_record_reclaim_priority(struct mem_cgroup
*mem
, int priority
)
416 mem
->prev_priority
= priority
;
420 * Calculate # of pages to be scanned in this priority/zone.
423 * priority starts from "DEF_PRIORITY" and decremented in each loop.
424 * (see include/linux/mmzone.h)
427 long mem_cgroup_calc_reclaim(struct mem_cgroup
*mem
, struct zone
*zone
,
428 int priority
, enum lru_list lru
)
431 int nid
= zone
->zone_pgdat
->node_id
;
432 int zid
= zone_idx(zone
);
433 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(mem
, nid
, zid
);
435 nr_pages
= MEM_CGROUP_ZSTAT(mz
, lru
);
437 return (nr_pages
>> priority
);
440 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan
,
441 struct list_head
*dst
,
442 unsigned long *scanned
, int order
,
443 int mode
, struct zone
*z
,
444 struct mem_cgroup
*mem_cont
,
445 int active
, int file
)
447 unsigned long nr_taken
= 0;
451 struct list_head
*src
;
452 struct page_cgroup
*pc
, *tmp
;
453 int nid
= z
->zone_pgdat
->node_id
;
454 int zid
= zone_idx(z
);
455 struct mem_cgroup_per_zone
*mz
;
456 int lru
= LRU_FILE
* !!file
+ !!active
;
459 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
460 src
= &mz
->lists
[lru
];
462 spin_lock(&mz
->lru_lock
);
464 list_for_each_entry_safe_reverse(pc
, tmp
, src
, lru
) {
465 if (scan
>= nr_to_scan
)
469 if (unlikely(!PageLRU(page
)))
473 * TODO: play better with lumpy reclaim, grabbing anything.
475 if (PageActive(page
) && !active
) {
476 __mem_cgroup_move_lists(pc
, true);
479 if (!PageActive(page
) && active
) {
480 __mem_cgroup_move_lists(pc
, false);
485 list_move(&pc
->lru
, &pc_list
);
487 if (__isolate_lru_page(page
, mode
, file
) == 0) {
488 list_move(&page
->lru
, dst
);
493 list_splice(&pc_list
, src
);
494 spin_unlock(&mz
->lru_lock
);
501 * Charge the memory controller for page usage.
503 * 0 if the charge was successful
504 * < 0 if the cgroup is over its limit
506 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
507 gfp_t gfp_mask
, enum charge_type ctype
,
508 struct mem_cgroup
*memcg
)
510 struct mem_cgroup
*mem
;
511 struct page_cgroup
*pc
;
513 unsigned long nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
514 struct mem_cgroup_per_zone
*mz
;
516 pc
= kmem_cache_alloc(page_cgroup_cache
, gfp_mask
);
517 if (unlikely(pc
== NULL
))
521 * We always charge the cgroup the mm_struct belongs to.
522 * The mm_struct's mem_cgroup changes on task migration if the
523 * thread group leader migrates. It's possible that mm is not
524 * set, if so charge the init_mm (happens for pagecache usage).
526 if (likely(!memcg
)) {
528 mem
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
529 if (unlikely(!mem
)) {
531 kmem_cache_free(page_cgroup_cache
, pc
);
535 * For every charge from the cgroup, increment reference count
541 css_get(&memcg
->css
);
544 while (res_counter_charge(&mem
->res
, PAGE_SIZE
)) {
545 if (!(gfp_mask
& __GFP_WAIT
))
548 if (try_to_free_mem_cgroup_pages(mem
, gfp_mask
))
552 * try_to_free_mem_cgroup_pages() might not give us a full
553 * picture of reclaim. Some pages are reclaimed and might be
554 * moved to swap cache or just unmapped from the cgroup.
555 * Check the limit again to see if the reclaim reduced the
556 * current usage of the cgroup before giving up
558 if (res_counter_check_under_limit(&mem
->res
))
562 mem_cgroup_out_of_memory(mem
, gfp_mask
);
567 pc
->mem_cgroup
= mem
;
570 * If a page is accounted as a page cache, insert to inactive list.
571 * If anon, insert to active list.
573 if (ctype
== MEM_CGROUP_CHARGE_TYPE_CACHE
) {
574 pc
->flags
= PAGE_CGROUP_FLAG_CACHE
;
575 if (page_is_file_cache(page
))
576 pc
->flags
|= PAGE_CGROUP_FLAG_FILE
;
578 pc
->flags
|= PAGE_CGROUP_FLAG_ACTIVE
;
579 } else if (ctype
== MEM_CGROUP_CHARGE_TYPE_MAPPED
)
580 pc
->flags
= PAGE_CGROUP_FLAG_ACTIVE
;
581 else /* MEM_CGROUP_CHARGE_TYPE_SHMEM */
582 pc
->flags
= PAGE_CGROUP_FLAG_CACHE
| PAGE_CGROUP_FLAG_ACTIVE
;
584 lock_page_cgroup(page
);
585 if (unlikely(page_get_page_cgroup(page
))) {
586 unlock_page_cgroup(page
);
587 res_counter_uncharge(&mem
->res
, PAGE_SIZE
);
589 kmem_cache_free(page_cgroup_cache
, pc
);
592 page_assign_page_cgroup(page
, pc
);
594 mz
= page_cgroup_zoneinfo(pc
);
595 spin_lock_irqsave(&mz
->lru_lock
, flags
);
596 __mem_cgroup_add_list(mz
, pc
);
597 spin_unlock_irqrestore(&mz
->lru_lock
, flags
);
599 unlock_page_cgroup(page
);
604 kmem_cache_free(page_cgroup_cache
, pc
);
609 int mem_cgroup_charge(struct page
*page
, struct mm_struct
*mm
, gfp_t gfp_mask
)
611 if (mem_cgroup_subsys
.disabled
)
615 * If already mapped, we don't have to account.
616 * If page cache, page->mapping has address_space.
617 * But page->mapping may have out-of-use anon_vma pointer,
618 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
621 if (page_mapped(page
) || (page
->mapping
&& !PageAnon(page
)))
625 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
626 MEM_CGROUP_CHARGE_TYPE_MAPPED
, NULL
);
629 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
632 if (mem_cgroup_subsys
.disabled
)
636 * Corner case handling. This is called from add_to_page_cache()
637 * in usual. But some FS (shmem) precharges this page before calling it
638 * and call add_to_page_cache() with GFP_NOWAIT.
640 * For GFP_NOWAIT case, the page may be pre-charged before calling
641 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
642 * charge twice. (It works but has to pay a bit larger cost.)
644 if (!(gfp_mask
& __GFP_WAIT
)) {
645 struct page_cgroup
*pc
;
647 lock_page_cgroup(page
);
648 pc
= page_get_page_cgroup(page
);
650 VM_BUG_ON(pc
->page
!= page
);
651 VM_BUG_ON(!pc
->mem_cgroup
);
652 unlock_page_cgroup(page
);
655 unlock_page_cgroup(page
);
661 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
662 MEM_CGROUP_CHARGE_TYPE_CACHE
, NULL
);
666 * uncharge if !page_mapped(page)
669 __mem_cgroup_uncharge_common(struct page
*page
, enum charge_type ctype
)
671 struct page_cgroup
*pc
;
672 struct mem_cgroup
*mem
;
673 struct mem_cgroup_per_zone
*mz
;
676 if (mem_cgroup_subsys
.disabled
)
680 * Check if our page_cgroup is valid
682 lock_page_cgroup(page
);
683 pc
= page_get_page_cgroup(page
);
687 VM_BUG_ON(pc
->page
!= page
);
689 if ((ctype
== MEM_CGROUP_CHARGE_TYPE_MAPPED
)
690 && ((pc
->flags
& PAGE_CGROUP_FLAG_CACHE
)
691 || page_mapped(page
)))
694 mz
= page_cgroup_zoneinfo(pc
);
695 spin_lock_irqsave(&mz
->lru_lock
, flags
);
696 __mem_cgroup_remove_list(mz
, pc
);
697 spin_unlock_irqrestore(&mz
->lru_lock
, flags
);
699 page_assign_page_cgroup(page
, NULL
);
700 unlock_page_cgroup(page
);
702 mem
= pc
->mem_cgroup
;
703 res_counter_uncharge(&mem
->res
, PAGE_SIZE
);
706 kmem_cache_free(page_cgroup_cache
, pc
);
709 unlock_page_cgroup(page
);
712 void mem_cgroup_uncharge_page(struct page
*page
)
714 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_MAPPED
);
717 void mem_cgroup_uncharge_cache_page(struct page
*page
)
719 VM_BUG_ON(page_mapped(page
));
720 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_CACHE
);
724 * Before starting migration, account against new page.
726 int mem_cgroup_prepare_migration(struct page
*page
, struct page
*newpage
)
728 struct page_cgroup
*pc
;
729 struct mem_cgroup
*mem
= NULL
;
730 enum charge_type ctype
= MEM_CGROUP_CHARGE_TYPE_MAPPED
;
733 if (mem_cgroup_subsys
.disabled
)
736 lock_page_cgroup(page
);
737 pc
= page_get_page_cgroup(page
);
739 mem
= pc
->mem_cgroup
;
741 if (pc
->flags
& PAGE_CGROUP_FLAG_CACHE
) {
742 if (page_is_file_cache(page
))
743 ctype
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
745 ctype
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
748 unlock_page_cgroup(page
);
750 ret
= mem_cgroup_charge_common(newpage
, NULL
, GFP_KERNEL
,
757 /* remove redundant charge if migration failed*/
758 void mem_cgroup_end_migration(struct page
*newpage
)
761 * At success, page->mapping is not NULL.
762 * special rollback care is necessary when
763 * 1. at migration failure. (newpage->mapping is cleared in this case)
764 * 2. the newpage was moved but not remapped again because the task
765 * exits and the newpage is obsolete. In this case, the new page
766 * may be a swapcache. So, we just call mem_cgroup_uncharge_page()
767 * always for avoiding mess. The page_cgroup will be removed if
768 * unnecessary. File cache pages is still on radix-tree. Don't
771 if (!newpage
->mapping
)
772 __mem_cgroup_uncharge_common(newpage
,
773 MEM_CGROUP_CHARGE_TYPE_FORCE
);
774 else if (PageAnon(newpage
))
775 mem_cgroup_uncharge_page(newpage
);
779 * A call to try to shrink memory usage under specified resource controller.
780 * This is typically used for page reclaiming for shmem for reducing side
781 * effect of page allocation from shmem, which is used by some mem_cgroup.
783 int mem_cgroup_shrink_usage(struct mm_struct
*mm
, gfp_t gfp_mask
)
785 struct mem_cgroup
*mem
;
787 int retry
= MEM_CGROUP_RECLAIM_RETRIES
;
789 if (mem_cgroup_subsys
.disabled
)
795 mem
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
796 if (unlikely(!mem
)) {
804 progress
= try_to_free_mem_cgroup_pages(mem
, gfp_mask
);
805 progress
+= res_counter_check_under_limit(&mem
->res
);
806 } while (!progress
&& --retry
);
814 int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
, unsigned long long val
)
817 int retry_count
= MEM_CGROUP_RECLAIM_RETRIES
;
821 while (res_counter_set_limit(&memcg
->res
, val
)) {
822 if (signal_pending(current
)) {
830 progress
= try_to_free_mem_cgroup_pages(memcg
, GFP_KERNEL
);
839 * This routine traverse page_cgroup in given list and drop them all.
840 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
842 #define FORCE_UNCHARGE_BATCH (128)
843 static void mem_cgroup_force_empty_list(struct mem_cgroup
*mem
,
844 struct mem_cgroup_per_zone
*mz
,
847 struct page_cgroup
*pc
;
849 int count
= FORCE_UNCHARGE_BATCH
;
851 struct list_head
*list
;
853 list
= &mz
->lists
[lru
];
855 spin_lock_irqsave(&mz
->lru_lock
, flags
);
856 while (!list_empty(list
)) {
857 pc
= list_entry(list
->prev
, struct page_cgroup
, lru
);
860 spin_unlock_irqrestore(&mz
->lru_lock
, flags
);
862 * Check if this page is on LRU. !LRU page can be found
863 * if it's under page migration.
866 __mem_cgroup_uncharge_common(page
,
867 MEM_CGROUP_CHARGE_TYPE_FORCE
);
870 count
= FORCE_UNCHARGE_BATCH
;
875 spin_lock_irqsave(&mz
->lru_lock
, flags
);
877 spin_unlock_irqrestore(&mz
->lru_lock
, flags
);
881 * make mem_cgroup's charge to be 0 if there is no task.
882 * This enables deleting this mem_cgroup.
884 static int mem_cgroup_force_empty(struct mem_cgroup
*mem
)
891 * page reclaim code (kswapd etc..) will move pages between
892 * active_list <-> inactive_list while we don't take a lock.
893 * So, we have to do loop here until all lists are empty.
895 while (mem
->res
.usage
> 0) {
896 if (atomic_read(&mem
->css
.cgroup
->count
) > 0)
898 for_each_node_state(node
, N_POSSIBLE
)
899 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
900 struct mem_cgroup_per_zone
*mz
;
902 mz
= mem_cgroup_zoneinfo(mem
, node
, zid
);
904 mem_cgroup_force_empty_list(mem
, mz
, l
);
913 static u64
mem_cgroup_read(struct cgroup
*cont
, struct cftype
*cft
)
915 return res_counter_read_u64(&mem_cgroup_from_cont(cont
)->res
,
919 * The user of this function is...
922 static int mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
925 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
926 unsigned long long val
;
929 switch (cft
->private) {
931 /* This function does all necessary parse...reuse it */
932 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
934 ret
= mem_cgroup_resize_limit(memcg
, val
);
937 ret
= -EINVAL
; /* should be BUG() ? */
943 static int mem_cgroup_reset(struct cgroup
*cont
, unsigned int event
)
945 struct mem_cgroup
*mem
;
947 mem
= mem_cgroup_from_cont(cont
);
950 res_counter_reset_max(&mem
->res
);
953 res_counter_reset_failcnt(&mem
->res
);
959 static int mem_force_empty_write(struct cgroup
*cont
, unsigned int event
)
961 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont
));
964 static const struct mem_cgroup_stat_desc
{
967 } mem_cgroup_stat_desc
[] = {
968 [MEM_CGROUP_STAT_CACHE
] = { "cache", PAGE_SIZE
, },
969 [MEM_CGROUP_STAT_RSS
] = { "rss", PAGE_SIZE
, },
970 [MEM_CGROUP_STAT_PGPGIN_COUNT
] = {"pgpgin", 1, },
971 [MEM_CGROUP_STAT_PGPGOUT_COUNT
] = {"pgpgout", 1, },
974 static int mem_control_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
975 struct cgroup_map_cb
*cb
)
977 struct mem_cgroup
*mem_cont
= mem_cgroup_from_cont(cont
);
978 struct mem_cgroup_stat
*stat
= &mem_cont
->stat
;
981 for (i
= 0; i
< ARRAY_SIZE(stat
->cpustat
[0].count
); i
++) {
984 val
= mem_cgroup_read_stat(stat
, i
);
985 val
*= mem_cgroup_stat_desc
[i
].unit
;
986 cb
->fill(cb
, mem_cgroup_stat_desc
[i
].msg
, val
);
988 /* showing # of active pages */
990 unsigned long active_anon
, inactive_anon
;
991 unsigned long active_file
, inactive_file
;
993 inactive_anon
= mem_cgroup_get_all_zonestat(mem_cont
,
995 active_anon
= mem_cgroup_get_all_zonestat(mem_cont
,
997 inactive_file
= mem_cgroup_get_all_zonestat(mem_cont
,
999 active_file
= mem_cgroup_get_all_zonestat(mem_cont
,
1001 cb
->fill(cb
, "active_anon", (active_anon
) * PAGE_SIZE
);
1002 cb
->fill(cb
, "inactive_anon", (inactive_anon
) * PAGE_SIZE
);
1003 cb
->fill(cb
, "active_file", (active_file
) * PAGE_SIZE
);
1004 cb
->fill(cb
, "inactive_file", (inactive_file
) * 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
;
1048 int zone
, tmp
= node
;
1050 * This routine is called against possible nodes.
1051 * But it's BUG to call kmalloc() against offline node.
1053 * TODO: this routine can waste much memory for nodes which will
1054 * never be onlined. It's better to use memory hotplug callback
1057 if (!node_state(node
, N_NORMAL_MEMORY
))
1059 pn
= kmalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
1063 mem
->info
.nodeinfo
[node
] = pn
;
1064 memset(pn
, 0, sizeof(*pn
));
1066 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
1067 mz
= &pn
->zoneinfo
[zone
];
1068 spin_lock_init(&mz
->lru_lock
);
1070 INIT_LIST_HEAD(&mz
->lists
[l
]);
1075 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*mem
, int node
)
1077 kfree(mem
->info
.nodeinfo
[node
]);
1080 static struct mem_cgroup
*mem_cgroup_alloc(void)
1082 struct mem_cgroup
*mem
;
1084 if (sizeof(*mem
) < PAGE_SIZE
)
1085 mem
= kmalloc(sizeof(*mem
), GFP_KERNEL
);
1087 mem
= vmalloc(sizeof(*mem
));
1090 memset(mem
, 0, sizeof(*mem
));
1094 static void mem_cgroup_free(struct mem_cgroup
*mem
)
1096 if (sizeof(*mem
) < PAGE_SIZE
)
1103 static struct cgroup_subsys_state
*
1104 mem_cgroup_create(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
1106 struct mem_cgroup
*mem
;
1109 if (unlikely((cont
->parent
) == NULL
)) {
1110 mem
= &init_mem_cgroup
;
1111 page_cgroup_cache
= KMEM_CACHE(page_cgroup
, SLAB_PANIC
);
1113 mem
= mem_cgroup_alloc();
1115 return ERR_PTR(-ENOMEM
);
1118 res_counter_init(&mem
->res
);
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 free_mem_cgroup_per_zone_info(mem
, node
);
1128 if (cont
->parent
!= NULL
)
1129 mem_cgroup_free(mem
);
1130 return ERR_PTR(-ENOMEM
);
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 free_mem_cgroup_per_zone_info(mem
, node
);
1149 mem_cgroup_free(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
);
1175 * Only thread group leaders are allowed to migrate, the mm_struct is
1176 * in effect owned by the leader
1178 if (!thread_group_leader(p
))
1185 struct cgroup_subsys mem_cgroup_subsys
= {
1187 .subsys_id
= mem_cgroup_subsys_id
,
1188 .create
= mem_cgroup_create
,
1189 .pre_destroy
= mem_cgroup_pre_destroy
,
1190 .destroy
= mem_cgroup_destroy
,
1191 .populate
= mem_cgroup_populate
,
1192 .attach
= mem_cgroup_move_task
,