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
{
93 * spin_lock to protect the per cgroup LRU
96 struct list_head active_list
;
97 struct list_head inactive_list
;
98 unsigned long count
[NR_MEM_CGROUP_ZSTAT
];
100 /* Macro for accessing counter */
101 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
103 struct mem_cgroup_per_node
{
104 struct mem_cgroup_per_zone zoneinfo
[MAX_NR_ZONES
];
107 struct mem_cgroup_lru_info
{
108 struct mem_cgroup_per_node
*nodeinfo
[MAX_NUMNODES
];
112 * The memory controller data structure. The memory controller controls both
113 * page cache and RSS per cgroup. We would eventually like to provide
114 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
115 * to help the administrator determine what knobs to tune.
117 * TODO: Add a water mark for the memory controller. Reclaim will begin when
118 * we hit the water mark. May be even add a low water mark, such that
119 * no reclaim occurs from a cgroup at it's low water mark, this is
120 * a feature that will be implemented much later in the future.
123 struct cgroup_subsys_state css
;
125 * the counter to account for memory usage
127 struct res_counter res
;
129 * Per cgroup active and inactive list, similar to the
130 * per zone LRU lists.
132 struct mem_cgroup_lru_info info
;
134 int prev_priority
; /* for recording reclaim priority */
138 struct mem_cgroup_stat stat
;
140 static struct mem_cgroup init_mem_cgroup
;
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 at least two
145 * byte aligned (based on comments from Nick Piggin). But since
146 * bit_spin_lock doesn't actually set that lock bit in a non-debug
147 * uniprocessor kernel, we should avoid setting it here too.
149 #define PAGE_CGROUP_LOCK_BIT 0x0
150 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK)
151 #define PAGE_CGROUP_LOCK (1 << PAGE_CGROUP_LOCK_BIT)
153 #define PAGE_CGROUP_LOCK 0x0
157 * A page_cgroup page is associated with every page descriptor. The
158 * page_cgroup helps us identify information about the cgroup
161 struct list_head lru
; /* per cgroup LRU list */
163 struct mem_cgroup
*mem_cgroup
;
164 int ref_cnt
; /* cached, mapped, migrating */
167 #define PAGE_CGROUP_FLAG_CACHE (0x1) /* charged as cache */
168 #define PAGE_CGROUP_FLAG_ACTIVE (0x2) /* page is active in this cgroup */
170 static int page_cgroup_nid(struct page_cgroup
*pc
)
172 return page_to_nid(pc
->page
);
175 static enum zone_type
page_cgroup_zid(struct page_cgroup
*pc
)
177 return page_zonenum(pc
->page
);
181 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
182 MEM_CGROUP_CHARGE_TYPE_MAPPED
,
186 * Always modified under lru lock. Then, not necessary to preempt_disable()
188 static void mem_cgroup_charge_statistics(struct mem_cgroup
*mem
, int flags
,
191 int val
= (charge
)? 1 : -1;
192 struct mem_cgroup_stat
*stat
= &mem
->stat
;
194 VM_BUG_ON(!irqs_disabled());
195 if (flags
& PAGE_CGROUP_FLAG_CACHE
)
196 __mem_cgroup_stat_add_safe(stat
, MEM_CGROUP_STAT_CACHE
, val
);
198 __mem_cgroup_stat_add_safe(stat
, MEM_CGROUP_STAT_RSS
, val
);
201 static struct mem_cgroup_per_zone
*
202 mem_cgroup_zoneinfo(struct mem_cgroup
*mem
, int nid
, int zid
)
204 return &mem
->info
.nodeinfo
[nid
]->zoneinfo
[zid
];
207 static struct mem_cgroup_per_zone
*
208 page_cgroup_zoneinfo(struct page_cgroup
*pc
)
210 struct mem_cgroup
*mem
= pc
->mem_cgroup
;
211 int nid
= page_cgroup_nid(pc
);
212 int zid
= page_cgroup_zid(pc
);
214 return mem_cgroup_zoneinfo(mem
, nid
, zid
);
217 static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup
*mem
,
218 enum mem_cgroup_zstat_index idx
)
221 struct mem_cgroup_per_zone
*mz
;
224 for_each_online_node(nid
)
225 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
226 mz
= mem_cgroup_zoneinfo(mem
, nid
, zid
);
227 total
+= MEM_CGROUP_ZSTAT(mz
, idx
);
232 static struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
234 return container_of(cgroup_subsys_state(cont
,
235 mem_cgroup_subsys_id
), struct mem_cgroup
,
239 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
241 return container_of(task_subsys_state(p
, mem_cgroup_subsys_id
),
242 struct mem_cgroup
, css
);
245 static inline int page_cgroup_locked(struct page
*page
)
247 return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT
, &page
->page_cgroup
);
250 static void page_assign_page_cgroup(struct page
*page
, struct page_cgroup
*pc
)
252 VM_BUG_ON(!page_cgroup_locked(page
));
253 page
->page_cgroup
= ((unsigned long)pc
| PAGE_CGROUP_LOCK
);
256 struct page_cgroup
*page_get_page_cgroup(struct page
*page
)
258 return (struct page_cgroup
*) (page
->page_cgroup
& ~PAGE_CGROUP_LOCK
);
261 static void lock_page_cgroup(struct page
*page
)
263 bit_spin_lock(PAGE_CGROUP_LOCK_BIT
, &page
->page_cgroup
);
266 static int try_lock_page_cgroup(struct page
*page
)
268 return bit_spin_trylock(PAGE_CGROUP_LOCK_BIT
, &page
->page_cgroup
);
271 static void unlock_page_cgroup(struct page
*page
)
273 bit_spin_unlock(PAGE_CGROUP_LOCK_BIT
, &page
->page_cgroup
);
276 static void __mem_cgroup_remove_list(struct page_cgroup
*pc
)
278 int from
= pc
->flags
& PAGE_CGROUP_FLAG_ACTIVE
;
279 struct mem_cgroup_per_zone
*mz
= page_cgroup_zoneinfo(pc
);
282 MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_ACTIVE
) -= 1;
284 MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_INACTIVE
) -= 1;
286 mem_cgroup_charge_statistics(pc
->mem_cgroup
, pc
->flags
, false);
287 list_del_init(&pc
->lru
);
290 static void __mem_cgroup_add_list(struct page_cgroup
*pc
)
292 int to
= pc
->flags
& PAGE_CGROUP_FLAG_ACTIVE
;
293 struct mem_cgroup_per_zone
*mz
= page_cgroup_zoneinfo(pc
);
296 MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_INACTIVE
) += 1;
297 list_add(&pc
->lru
, &mz
->inactive_list
);
299 MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_ACTIVE
) += 1;
300 list_add(&pc
->lru
, &mz
->active_list
);
302 mem_cgroup_charge_statistics(pc
->mem_cgroup
, pc
->flags
, true);
305 static void __mem_cgroup_move_lists(struct page_cgroup
*pc
, bool active
)
307 int from
= pc
->flags
& PAGE_CGROUP_FLAG_ACTIVE
;
308 struct mem_cgroup_per_zone
*mz
= page_cgroup_zoneinfo(pc
);
311 MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_ACTIVE
) -= 1;
313 MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_INACTIVE
) -= 1;
316 MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_ACTIVE
) += 1;
317 pc
->flags
|= PAGE_CGROUP_FLAG_ACTIVE
;
318 list_move(&pc
->lru
, &mz
->active_list
);
320 MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_INACTIVE
) += 1;
321 pc
->flags
&= ~PAGE_CGROUP_FLAG_ACTIVE
;
322 list_move(&pc
->lru
, &mz
->inactive_list
);
326 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*mem
)
331 ret
= task
->mm
&& mm_match_cgroup(task
->mm
, mem
);
337 * This routine assumes that the appropriate zone's lru lock is already held
339 void mem_cgroup_move_lists(struct page
*page
, bool active
)
341 struct page_cgroup
*pc
;
342 struct mem_cgroup_per_zone
*mz
;
346 * We cannot lock_page_cgroup while holding zone's lru_lock,
347 * because other holders of lock_page_cgroup can be interrupted
348 * with an attempt to rotate_reclaimable_page. But we cannot
349 * safely get to page_cgroup without it, so just try_lock it:
350 * mem_cgroup_isolate_pages allows for page left on wrong list.
352 if (!try_lock_page_cgroup(page
))
355 pc
= page_get_page_cgroup(page
);
357 mz
= page_cgroup_zoneinfo(pc
);
358 spin_lock_irqsave(&mz
->lru_lock
, flags
);
359 __mem_cgroup_move_lists(pc
, active
);
360 spin_unlock_irqrestore(&mz
->lru_lock
, flags
);
362 unlock_page_cgroup(page
);
366 * Calculate mapped_ratio under memory controller. This will be used in
367 * vmscan.c for deteremining we have to reclaim mapped pages.
369 int mem_cgroup_calc_mapped_ratio(struct mem_cgroup
*mem
)
374 * usage is recorded in bytes. But, here, we assume the number of
375 * physical pages can be represented by "long" on any arch.
377 total
= (long) (mem
->res
.usage
>> PAGE_SHIFT
) + 1L;
378 rss
= (long)mem_cgroup_read_stat(&mem
->stat
, MEM_CGROUP_STAT_RSS
);
379 return (int)((rss
* 100L) / total
);
383 * This function is called from vmscan.c. In page reclaiming loop. balance
384 * between active and inactive list is calculated. For memory controller
385 * page reclaiming, we should use using mem_cgroup's imbalance rather than
386 * zone's global lru imbalance.
388 long mem_cgroup_reclaim_imbalance(struct mem_cgroup
*mem
)
390 unsigned long active
, inactive
;
391 /* active and inactive are the number of pages. 'long' is ok.*/
392 active
= mem_cgroup_get_all_zonestat(mem
, MEM_CGROUP_ZSTAT_ACTIVE
);
393 inactive
= mem_cgroup_get_all_zonestat(mem
, MEM_CGROUP_ZSTAT_INACTIVE
);
394 return (long) (active
/ (inactive
+ 1));
398 * prev_priority control...this will be used in memory reclaim path.
400 int mem_cgroup_get_reclaim_priority(struct mem_cgroup
*mem
)
402 return mem
->prev_priority
;
405 void mem_cgroup_note_reclaim_priority(struct mem_cgroup
*mem
, int priority
)
407 if (priority
< mem
->prev_priority
)
408 mem
->prev_priority
= priority
;
411 void mem_cgroup_record_reclaim_priority(struct mem_cgroup
*mem
, int priority
)
413 mem
->prev_priority
= priority
;
417 * Calculate # of pages to be scanned in this priority/zone.
420 * priority starts from "DEF_PRIORITY" and decremented in each loop.
421 * (see include/linux/mmzone.h)
424 long mem_cgroup_calc_reclaim_active(struct mem_cgroup
*mem
,
425 struct zone
*zone
, int priority
)
428 int nid
= zone
->zone_pgdat
->node_id
;
429 int zid
= zone_idx(zone
);
430 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(mem
, nid
, zid
);
432 nr_active
= MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_ACTIVE
);
433 return (nr_active
>> priority
);
436 long mem_cgroup_calc_reclaim_inactive(struct mem_cgroup
*mem
,
437 struct zone
*zone
, int priority
)
440 int nid
= zone
->zone_pgdat
->node_id
;
441 int zid
= zone_idx(zone
);
442 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(mem
, nid
, zid
);
444 nr_inactive
= MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_INACTIVE
);
445 return (nr_inactive
>> priority
);
448 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan
,
449 struct list_head
*dst
,
450 unsigned long *scanned
, int order
,
451 int mode
, struct zone
*z
,
452 struct mem_cgroup
*mem_cont
,
455 unsigned long nr_taken
= 0;
459 struct list_head
*src
;
460 struct page_cgroup
*pc
, *tmp
;
461 int nid
= z
->zone_pgdat
->node_id
;
462 int zid
= zone_idx(z
);
463 struct mem_cgroup_per_zone
*mz
;
466 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
468 src
= &mz
->active_list
;
470 src
= &mz
->inactive_list
;
473 spin_lock(&mz
->lru_lock
);
475 list_for_each_entry_safe_reverse(pc
, tmp
, src
, lru
) {
476 if (scan
>= nr_to_scan
)
480 if (unlikely(!PageLRU(page
)))
483 if (PageActive(page
) && !active
) {
484 __mem_cgroup_move_lists(pc
, true);
487 if (!PageActive(page
) && active
) {
488 __mem_cgroup_move_lists(pc
, false);
493 list_move(&pc
->lru
, &pc_list
);
495 if (__isolate_lru_page(page
, mode
) == 0) {
496 list_move(&page
->lru
, dst
);
501 list_splice(&pc_list
, src
);
502 spin_unlock(&mz
->lru_lock
);
509 * Charge the memory controller for page usage.
511 * 0 if the charge was successful
512 * < 0 if the cgroup is over its limit
514 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
515 gfp_t gfp_mask
, enum charge_type ctype
)
517 struct mem_cgroup
*mem
;
518 struct page_cgroup
*pc
;
520 unsigned long nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
521 struct mem_cgroup_per_zone
*mz
;
523 if (mem_cgroup_subsys
.disabled
)
527 * Should page_cgroup's go to their own slab?
528 * One could optimize the performance of the charging routine
529 * by saving a bit in the page_flags and using it as a lock
530 * to see if the cgroup page already has a page_cgroup associated
534 lock_page_cgroup(page
);
535 pc
= page_get_page_cgroup(page
);
537 * The page_cgroup exists and
538 * the page has already been accounted.
541 VM_BUG_ON(pc
->page
!= page
);
542 VM_BUG_ON(pc
->ref_cnt
<= 0);
545 unlock_page_cgroup(page
);
548 unlock_page_cgroup(page
);
550 pc
= kzalloc(sizeof(struct page_cgroup
), gfp_mask
);
555 * We always charge the cgroup the mm_struct belongs to.
556 * The mm_struct's mem_cgroup changes on task migration if the
557 * thread group leader migrates. It's possible that mm is not
558 * set, if so charge the init_mm (happens for pagecache usage).
564 mem
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
566 * For every charge from the cgroup, increment reference count
571 while (res_counter_charge(&mem
->res
, PAGE_SIZE
)) {
572 if (!(gfp_mask
& __GFP_WAIT
))
575 if (try_to_free_mem_cgroup_pages(mem
, gfp_mask
))
579 * try_to_free_mem_cgroup_pages() might not give us a full
580 * picture of reclaim. Some pages are reclaimed and might be
581 * moved to swap cache or just unmapped from the cgroup.
582 * Check the limit again to see if the reclaim reduced the
583 * current usage of the cgroup before giving up
585 if (res_counter_check_under_limit(&mem
->res
))
589 mem_cgroup_out_of_memory(mem
, gfp_mask
);
592 congestion_wait(WRITE
, HZ
/10);
596 pc
->mem_cgroup
= mem
;
598 pc
->flags
= PAGE_CGROUP_FLAG_ACTIVE
;
599 if (ctype
== MEM_CGROUP_CHARGE_TYPE_CACHE
)
600 pc
->flags
|= PAGE_CGROUP_FLAG_CACHE
;
602 lock_page_cgroup(page
);
603 if (page_get_page_cgroup(page
)) {
604 unlock_page_cgroup(page
);
606 * Another charge has been added to this page already.
607 * We take lock_page_cgroup(page) again and read
608 * page->cgroup, increment refcnt.... just retry is OK.
610 res_counter_uncharge(&mem
->res
, PAGE_SIZE
);
615 page_assign_page_cgroup(page
, pc
);
617 mz
= page_cgroup_zoneinfo(pc
);
618 spin_lock_irqsave(&mz
->lru_lock
, flags
);
619 __mem_cgroup_add_list(pc
);
620 spin_unlock_irqrestore(&mz
->lru_lock
, flags
);
622 unlock_page_cgroup(page
);
632 int mem_cgroup_charge(struct page
*page
, struct mm_struct
*mm
, gfp_t gfp_mask
)
634 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
635 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
638 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
643 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
644 MEM_CGROUP_CHARGE_TYPE_CACHE
);
648 * Uncharging is always a welcome operation, we never complain, simply
651 void mem_cgroup_uncharge_page(struct page
*page
)
653 struct page_cgroup
*pc
;
654 struct mem_cgroup
*mem
;
655 struct mem_cgroup_per_zone
*mz
;
658 if (mem_cgroup_subsys
.disabled
)
662 * Check if our page_cgroup is valid
664 lock_page_cgroup(page
);
665 pc
= page_get_page_cgroup(page
);
669 VM_BUG_ON(pc
->page
!= page
);
670 VM_BUG_ON(pc
->ref_cnt
<= 0);
672 if (--(pc
->ref_cnt
) == 0) {
673 mz
= page_cgroup_zoneinfo(pc
);
674 spin_lock_irqsave(&mz
->lru_lock
, flags
);
675 __mem_cgroup_remove_list(pc
);
676 spin_unlock_irqrestore(&mz
->lru_lock
, flags
);
678 page_assign_page_cgroup(page
, NULL
);
679 unlock_page_cgroup(page
);
681 mem
= pc
->mem_cgroup
;
682 res_counter_uncharge(&mem
->res
, PAGE_SIZE
);
690 unlock_page_cgroup(page
);
694 * Returns non-zero if a page (under migration) has valid page_cgroup member.
695 * Refcnt of page_cgroup is incremented.
697 int mem_cgroup_prepare_migration(struct page
*page
)
699 struct page_cgroup
*pc
;
701 if (mem_cgroup_subsys
.disabled
)
704 lock_page_cgroup(page
);
705 pc
= page_get_page_cgroup(page
);
708 unlock_page_cgroup(page
);
712 void mem_cgroup_end_migration(struct page
*page
)
714 mem_cgroup_uncharge_page(page
);
718 * We know both *page* and *newpage* are now not-on-LRU and PG_locked.
719 * And no race with uncharge() routines because page_cgroup for *page*
720 * has extra one reference by mem_cgroup_prepare_migration.
722 void mem_cgroup_page_migration(struct page
*page
, struct page
*newpage
)
724 struct page_cgroup
*pc
;
725 struct mem_cgroup_per_zone
*mz
;
728 lock_page_cgroup(page
);
729 pc
= page_get_page_cgroup(page
);
731 unlock_page_cgroup(page
);
735 mz
= page_cgroup_zoneinfo(pc
);
736 spin_lock_irqsave(&mz
->lru_lock
, flags
);
737 __mem_cgroup_remove_list(pc
);
738 spin_unlock_irqrestore(&mz
->lru_lock
, flags
);
740 page_assign_page_cgroup(page
, NULL
);
741 unlock_page_cgroup(page
);
744 lock_page_cgroup(newpage
);
745 page_assign_page_cgroup(newpage
, pc
);
747 mz
= page_cgroup_zoneinfo(pc
);
748 spin_lock_irqsave(&mz
->lru_lock
, flags
);
749 __mem_cgroup_add_list(pc
);
750 spin_unlock_irqrestore(&mz
->lru_lock
, flags
);
752 unlock_page_cgroup(newpage
);
756 * This routine traverse page_cgroup in given list and drop them all.
757 * This routine ignores page_cgroup->ref_cnt.
758 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
760 #define FORCE_UNCHARGE_BATCH (128)
761 static void mem_cgroup_force_empty_list(struct mem_cgroup
*mem
,
762 struct mem_cgroup_per_zone
*mz
,
765 struct page_cgroup
*pc
;
767 int count
= FORCE_UNCHARGE_BATCH
;
769 struct list_head
*list
;
772 list
= &mz
->active_list
;
774 list
= &mz
->inactive_list
;
776 spin_lock_irqsave(&mz
->lru_lock
, flags
);
777 while (!list_empty(list
)) {
778 pc
= list_entry(list
->prev
, struct page_cgroup
, lru
);
781 spin_unlock_irqrestore(&mz
->lru_lock
, flags
);
782 mem_cgroup_uncharge_page(page
);
785 count
= FORCE_UNCHARGE_BATCH
;
788 spin_lock_irqsave(&mz
->lru_lock
, flags
);
790 spin_unlock_irqrestore(&mz
->lru_lock
, flags
);
794 * make mem_cgroup's charge to be 0 if there is no task.
795 * This enables deleting this mem_cgroup.
797 static int mem_cgroup_force_empty(struct mem_cgroup
*mem
)
802 if (mem_cgroup_subsys
.disabled
)
807 * page reclaim code (kswapd etc..) will move pages between
808 * active_list <-> inactive_list while we don't take a lock.
809 * So, we have to do loop here until all lists are empty.
811 while (mem
->res
.usage
> 0) {
812 if (atomic_read(&mem
->css
.cgroup
->count
) > 0)
814 for_each_node_state(node
, N_POSSIBLE
)
815 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
816 struct mem_cgroup_per_zone
*mz
;
817 mz
= mem_cgroup_zoneinfo(mem
, node
, zid
);
818 /* drop all page_cgroup in active_list */
819 mem_cgroup_force_empty_list(mem
, mz
, 1);
820 /* drop all page_cgroup in inactive_list */
821 mem_cgroup_force_empty_list(mem
, mz
, 0);
830 static int mem_cgroup_write_strategy(char *buf
, unsigned long long *tmp
)
832 *tmp
= memparse(buf
, &buf
);
837 * Round up the value to the closest page size
839 *tmp
= ((*tmp
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
) << PAGE_SHIFT
;
843 static u64
mem_cgroup_read(struct cgroup
*cont
, struct cftype
*cft
)
845 return res_counter_read_u64(&mem_cgroup_from_cont(cont
)->res
,
849 static ssize_t
mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
850 struct file
*file
, const char __user
*userbuf
,
851 size_t nbytes
, loff_t
*ppos
)
853 return res_counter_write(&mem_cgroup_from_cont(cont
)->res
,
854 cft
->private, userbuf
, nbytes
, ppos
,
855 mem_cgroup_write_strategy
);
858 static ssize_t
mem_force_empty_write(struct cgroup
*cont
,
859 struct cftype
*cft
, struct file
*file
,
860 const char __user
*userbuf
,
861 size_t nbytes
, loff_t
*ppos
)
863 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
864 int ret
= mem_cgroup_force_empty(mem
);
870 static const struct mem_cgroup_stat_desc
{
873 } mem_cgroup_stat_desc
[] = {
874 [MEM_CGROUP_STAT_CACHE
] = { "cache", PAGE_SIZE
, },
875 [MEM_CGROUP_STAT_RSS
] = { "rss", PAGE_SIZE
, },
878 static int mem_control_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
879 struct cgroup_map_cb
*cb
)
881 struct mem_cgroup
*mem_cont
= mem_cgroup_from_cont(cont
);
882 struct mem_cgroup_stat
*stat
= &mem_cont
->stat
;
885 for (i
= 0; i
< ARRAY_SIZE(stat
->cpustat
[0].count
); i
++) {
888 val
= mem_cgroup_read_stat(stat
, i
);
889 val
*= mem_cgroup_stat_desc
[i
].unit
;
890 cb
->fill(cb
, mem_cgroup_stat_desc
[i
].msg
, val
);
892 /* showing # of active pages */
894 unsigned long active
, inactive
;
896 inactive
= mem_cgroup_get_all_zonestat(mem_cont
,
897 MEM_CGROUP_ZSTAT_INACTIVE
);
898 active
= mem_cgroup_get_all_zonestat(mem_cont
,
899 MEM_CGROUP_ZSTAT_ACTIVE
);
900 cb
->fill(cb
, "active", (active
) * PAGE_SIZE
);
901 cb
->fill(cb
, "inactive", (inactive
) * PAGE_SIZE
);
906 static struct cftype mem_cgroup_files
[] = {
908 .name
= "usage_in_bytes",
909 .private = RES_USAGE
,
910 .read_u64
= mem_cgroup_read
,
913 .name
= "limit_in_bytes",
914 .private = RES_LIMIT
,
915 .write
= mem_cgroup_write
,
916 .read_u64
= mem_cgroup_read
,
920 .private = RES_FAILCNT
,
921 .read_u64
= mem_cgroup_read
,
924 .name
= "force_empty",
925 .write
= mem_force_empty_write
,
929 .read_map
= mem_control_stat_show
,
933 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*mem
, int node
)
935 struct mem_cgroup_per_node
*pn
;
936 struct mem_cgroup_per_zone
*mz
;
937 int zone
, tmp
= node
;
939 * This routine is called against possible nodes.
940 * But it's BUG to call kmalloc() against offline node.
942 * TODO: this routine can waste much memory for nodes which will
943 * never be onlined. It's better to use memory hotplug callback
946 if (!node_state(node
, N_NORMAL_MEMORY
))
948 pn
= kmalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
952 mem
->info
.nodeinfo
[node
] = pn
;
953 memset(pn
, 0, sizeof(*pn
));
955 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
956 mz
= &pn
->zoneinfo
[zone
];
957 INIT_LIST_HEAD(&mz
->active_list
);
958 INIT_LIST_HEAD(&mz
->inactive_list
);
959 spin_lock_init(&mz
->lru_lock
);
964 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*mem
, int node
)
966 kfree(mem
->info
.nodeinfo
[node
]);
969 static struct cgroup_subsys_state
*
970 mem_cgroup_create(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
972 struct mem_cgroup
*mem
;
975 if (unlikely((cont
->parent
) == NULL
))
976 mem
= &init_mem_cgroup
;
978 mem
= kzalloc(sizeof(struct mem_cgroup
), GFP_KERNEL
);
981 return ERR_PTR(-ENOMEM
);
983 res_counter_init(&mem
->res
);
985 memset(&mem
->info
, 0, sizeof(mem
->info
));
987 for_each_node_state(node
, N_POSSIBLE
)
988 if (alloc_mem_cgroup_per_zone_info(mem
, node
))
993 for_each_node_state(node
, N_POSSIBLE
)
994 free_mem_cgroup_per_zone_info(mem
, node
);
995 if (cont
->parent
!= NULL
)
997 return ERR_PTR(-ENOMEM
);
1000 static void mem_cgroup_pre_destroy(struct cgroup_subsys
*ss
,
1001 struct cgroup
*cont
)
1003 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
1004 mem_cgroup_force_empty(mem
);
1007 static void mem_cgroup_destroy(struct cgroup_subsys
*ss
,
1008 struct cgroup
*cont
)
1011 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
1013 for_each_node_state(node
, N_POSSIBLE
)
1014 free_mem_cgroup_per_zone_info(mem
, node
);
1016 kfree(mem_cgroup_from_cont(cont
));
1019 static int mem_cgroup_populate(struct cgroup_subsys
*ss
,
1020 struct cgroup
*cont
)
1022 if (mem_cgroup_subsys
.disabled
)
1024 return cgroup_add_files(cont
, ss
, mem_cgroup_files
,
1025 ARRAY_SIZE(mem_cgroup_files
));
1028 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
1029 struct cgroup
*cont
,
1030 struct cgroup
*old_cont
,
1031 struct task_struct
*p
)
1033 struct mm_struct
*mm
;
1034 struct mem_cgroup
*mem
, *old_mem
;
1036 if (mem_cgroup_subsys
.disabled
)
1039 mm
= get_task_mm(p
);
1043 mem
= mem_cgroup_from_cont(cont
);
1044 old_mem
= mem_cgroup_from_cont(old_cont
);
1050 * Only thread group leaders are allowed to migrate, the mm_struct is
1051 * in effect owned by the leader
1053 if (!thread_group_leader(p
))
1060 struct cgroup_subsys mem_cgroup_subsys
= {
1062 .subsys_id
= mem_cgroup_subsys_id
,
1063 .create
= mem_cgroup_create
,
1064 .pre_destroy
= mem_cgroup_pre_destroy
,
1065 .destroy
= mem_cgroup_destroy
,
1066 .populate
= mem_cgroup_populate
,
1067 .attach
= mem_cgroup_move_task
,