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 static 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 void mm_init_cgroup(struct mm_struct
*mm
, struct task_struct
*p
)
247 struct mem_cgroup
*mem
;
249 mem
= mem_cgroup_from_task(p
);
251 mm
->mem_cgroup
= mem
;
254 void mm_free_cgroup(struct mm_struct
*mm
)
256 css_put(&mm
->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 page_cgroup
*pc
)
292 int from
= 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_ACTIVE
) -= 1;
298 MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_INACTIVE
) -= 1;
300 mem_cgroup_charge_statistics(pc
->mem_cgroup
, pc
->flags
, false);
301 list_del_init(&pc
->lru
);
304 static void __mem_cgroup_add_list(struct page_cgroup
*pc
)
306 int to
= pc
->flags
& PAGE_CGROUP_FLAG_ACTIVE
;
307 struct mem_cgroup_per_zone
*mz
= page_cgroup_zoneinfo(pc
);
310 MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_INACTIVE
) += 1;
311 list_add(&pc
->lru
, &mz
->inactive_list
);
313 MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_ACTIVE
) += 1;
314 list_add(&pc
->lru
, &mz
->active_list
);
316 mem_cgroup_charge_statistics(pc
->mem_cgroup
, pc
->flags
, true);
319 static void __mem_cgroup_move_lists(struct page_cgroup
*pc
, bool active
)
321 int from
= pc
->flags
& PAGE_CGROUP_FLAG_ACTIVE
;
322 struct mem_cgroup_per_zone
*mz
= page_cgroup_zoneinfo(pc
);
325 MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_ACTIVE
) -= 1;
327 MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_INACTIVE
) -= 1;
330 MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_ACTIVE
) += 1;
331 pc
->flags
|= PAGE_CGROUP_FLAG_ACTIVE
;
332 list_move(&pc
->lru
, &mz
->active_list
);
334 MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_INACTIVE
) += 1;
335 pc
->flags
&= ~PAGE_CGROUP_FLAG_ACTIVE
;
336 list_move(&pc
->lru
, &mz
->inactive_list
);
340 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*mem
)
345 ret
= task
->mm
&& mm_match_cgroup(task
->mm
, mem
);
351 * This routine assumes that the appropriate zone's lru lock is already held
353 void mem_cgroup_move_lists(struct page
*page
, bool active
)
355 struct page_cgroup
*pc
;
356 struct mem_cgroup
*mem
;
357 struct mem_cgroup_per_zone
*mz
;
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
))
371 * Now page_cgroup is stable, but we cannot acquire mz->lru_lock
372 * while holding it, because mem_cgroup_force_empty_list does the
373 * reverse. Get a hold on the mem_cgroup before unlocking, so that
374 * the zoneinfo remains stable, then take mz->lru_lock; then check
375 * that page still points to pc and pc (even if freed and reassigned
376 * to that same page meanwhile) still points to the same mem_cgroup.
377 * Then we know mz still points to the right spinlock, so it's safe
378 * to move_lists (page->page_cgroup might be reset while we do so, but
379 * that doesn't matter: pc->page is stable till we drop mz->lru_lock).
380 * We're being a little naughty not to try_lock_page_cgroup again
381 * inside there, but we are safe, aren't we? Aren't we? Whistle...
383 pc
= page_get_page_cgroup(page
);
385 mem
= pc
->mem_cgroup
;
386 mz
= page_cgroup_zoneinfo(pc
);
389 unlock_page_cgroup(page
);
391 spin_lock_irqsave(&mz
->lru_lock
, flags
);
392 if (page_get_page_cgroup(page
) == pc
&& pc
->mem_cgroup
== mem
)
393 __mem_cgroup_move_lists(pc
, active
);
394 spin_unlock_irqrestore(&mz
->lru_lock
, flags
);
398 unlock_page_cgroup(page
);
402 * Calculate mapped_ratio under memory controller. This will be used in
403 * vmscan.c for deteremining we have to reclaim mapped pages.
405 int mem_cgroup_calc_mapped_ratio(struct mem_cgroup
*mem
)
410 * usage is recorded in bytes. But, here, we assume the number of
411 * physical pages can be represented by "long" on any arch.
413 total
= (long) (mem
->res
.usage
>> PAGE_SHIFT
) + 1L;
414 rss
= (long)mem_cgroup_read_stat(&mem
->stat
, MEM_CGROUP_STAT_RSS
);
415 return (int)((rss
* 100L) / total
);
419 * This function is called from vmscan.c. In page reclaiming loop. balance
420 * between active and inactive list is calculated. For memory controller
421 * page reclaiming, we should use using mem_cgroup's imbalance rather than
422 * zone's global lru imbalance.
424 long mem_cgroup_reclaim_imbalance(struct mem_cgroup
*mem
)
426 unsigned long active
, inactive
;
427 /* active and inactive are the number of pages. 'long' is ok.*/
428 active
= mem_cgroup_get_all_zonestat(mem
, MEM_CGROUP_ZSTAT_ACTIVE
);
429 inactive
= mem_cgroup_get_all_zonestat(mem
, MEM_CGROUP_ZSTAT_INACTIVE
);
430 return (long) (active
/ (inactive
+ 1));
434 * prev_priority control...this will be used in memory reclaim path.
436 int mem_cgroup_get_reclaim_priority(struct mem_cgroup
*mem
)
438 return mem
->prev_priority
;
441 void mem_cgroup_note_reclaim_priority(struct mem_cgroup
*mem
, int priority
)
443 if (priority
< mem
->prev_priority
)
444 mem
->prev_priority
= priority
;
447 void mem_cgroup_record_reclaim_priority(struct mem_cgroup
*mem
, int priority
)
449 mem
->prev_priority
= priority
;
453 * Calculate # of pages to be scanned in this priority/zone.
456 * priority starts from "DEF_PRIORITY" and decremented in each loop.
457 * (see include/linux/mmzone.h)
460 long mem_cgroup_calc_reclaim_active(struct mem_cgroup
*mem
,
461 struct zone
*zone
, int priority
)
464 int nid
= zone
->zone_pgdat
->node_id
;
465 int zid
= zone_idx(zone
);
466 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(mem
, nid
, zid
);
468 nr_active
= MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_ACTIVE
);
469 return (nr_active
>> priority
);
472 long mem_cgroup_calc_reclaim_inactive(struct mem_cgroup
*mem
,
473 struct zone
*zone
, int priority
)
476 int nid
= zone
->zone_pgdat
->node_id
;
477 int zid
= zone_idx(zone
);
478 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(mem
, nid
, zid
);
480 nr_inactive
= MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_INACTIVE
);
481 return (nr_inactive
>> priority
);
484 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan
,
485 struct list_head
*dst
,
486 unsigned long *scanned
, int order
,
487 int mode
, struct zone
*z
,
488 struct mem_cgroup
*mem_cont
,
491 unsigned long nr_taken
= 0;
495 struct list_head
*src
;
496 struct page_cgroup
*pc
, *tmp
;
497 int nid
= z
->zone_pgdat
->node_id
;
498 int zid
= zone_idx(z
);
499 struct mem_cgroup_per_zone
*mz
;
501 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
503 src
= &mz
->active_list
;
505 src
= &mz
->inactive_list
;
508 spin_lock(&mz
->lru_lock
);
510 list_for_each_entry_safe_reverse(pc
, tmp
, src
, lru
) {
511 if (scan
>= nr_to_scan
)
515 if (unlikely(!PageLRU(page
)))
518 if (PageActive(page
) && !active
) {
519 __mem_cgroup_move_lists(pc
, true);
522 if (!PageActive(page
) && active
) {
523 __mem_cgroup_move_lists(pc
, false);
528 list_move(&pc
->lru
, &pc_list
);
530 if (__isolate_lru_page(page
, mode
) == 0) {
531 list_move(&page
->lru
, dst
);
536 list_splice(&pc_list
, src
);
537 spin_unlock(&mz
->lru_lock
);
544 * Charge the memory controller for page usage.
546 * 0 if the charge was successful
547 * < 0 if the cgroup is over its limit
549 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
550 gfp_t gfp_mask
, enum charge_type ctype
)
552 struct mem_cgroup
*mem
;
553 struct page_cgroup
*pc
;
555 unsigned long nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
556 struct mem_cgroup_per_zone
*mz
;
559 * Should page_cgroup's go to their own slab?
560 * One could optimize the performance of the charging routine
561 * by saving a bit in the page_flags and using it as a lock
562 * to see if the cgroup page already has a page_cgroup associated
566 lock_page_cgroup(page
);
567 pc
= page_get_page_cgroup(page
);
569 * The page_cgroup exists and
570 * the page has already been accounted.
573 VM_BUG_ON(pc
->page
!= page
);
574 VM_BUG_ON(pc
->ref_cnt
<= 0);
577 unlock_page_cgroup(page
);
580 unlock_page_cgroup(page
);
582 pc
= kzalloc(sizeof(struct page_cgroup
), gfp_mask
);
587 * We always charge the cgroup the mm_struct belongs to.
588 * The mm_struct's mem_cgroup changes on task migration if the
589 * thread group leader migrates. It's possible that mm is not
590 * set, if so charge the init_mm (happens for pagecache usage).
596 mem
= rcu_dereference(mm
->mem_cgroup
);
598 * For every charge from the cgroup, increment reference count
603 while (res_counter_charge(&mem
->res
, PAGE_SIZE
)) {
604 if (!(gfp_mask
& __GFP_WAIT
))
607 if (try_to_free_mem_cgroup_pages(mem
, gfp_mask
))
611 * try_to_free_mem_cgroup_pages() might not give us a full
612 * picture of reclaim. Some pages are reclaimed and might be
613 * moved to swap cache or just unmapped from the cgroup.
614 * Check the limit again to see if the reclaim reduced the
615 * current usage of the cgroup before giving up
617 if (res_counter_check_under_limit(&mem
->res
))
621 mem_cgroup_out_of_memory(mem
, gfp_mask
);
624 congestion_wait(WRITE
, HZ
/10);
628 pc
->mem_cgroup
= mem
;
630 pc
->flags
= PAGE_CGROUP_FLAG_ACTIVE
;
631 if (ctype
== MEM_CGROUP_CHARGE_TYPE_CACHE
)
632 pc
->flags
|= PAGE_CGROUP_FLAG_CACHE
;
634 lock_page_cgroup(page
);
635 if (page_get_page_cgroup(page
)) {
636 unlock_page_cgroup(page
);
638 * Another charge has been added to this page already.
639 * We take lock_page_cgroup(page) again and read
640 * page->cgroup, increment refcnt.... just retry is OK.
642 res_counter_uncharge(&mem
->res
, PAGE_SIZE
);
647 page_assign_page_cgroup(page
, pc
);
648 unlock_page_cgroup(page
);
650 mz
= page_cgroup_zoneinfo(pc
);
651 spin_lock_irqsave(&mz
->lru_lock
, flags
);
652 __mem_cgroup_add_list(pc
);
653 spin_unlock_irqrestore(&mz
->lru_lock
, flags
);
664 int mem_cgroup_charge(struct page
*page
, struct mm_struct
*mm
, gfp_t gfp_mask
)
666 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
667 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
670 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
675 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
676 MEM_CGROUP_CHARGE_TYPE_CACHE
);
680 * Uncharging is always a welcome operation, we never complain, simply
683 void mem_cgroup_uncharge_page(struct page
*page
)
685 struct page_cgroup
*pc
;
686 struct mem_cgroup
*mem
;
687 struct mem_cgroup_per_zone
*mz
;
691 * Check if our page_cgroup is valid
693 lock_page_cgroup(page
);
694 pc
= page_get_page_cgroup(page
);
698 VM_BUG_ON(pc
->page
!= page
);
699 VM_BUG_ON(pc
->ref_cnt
<= 0);
701 if (--(pc
->ref_cnt
) == 0) {
702 page_assign_page_cgroup(page
, NULL
);
703 unlock_page_cgroup(page
);
705 mz
= page_cgroup_zoneinfo(pc
);
706 spin_lock_irqsave(&mz
->lru_lock
, flags
);
707 __mem_cgroup_remove_list(pc
);
708 spin_unlock_irqrestore(&mz
->lru_lock
, flags
);
710 mem
= pc
->mem_cgroup
;
711 res_counter_uncharge(&mem
->res
, PAGE_SIZE
);
719 unlock_page_cgroup(page
);
723 * Returns non-zero if a page (under migration) has valid page_cgroup member.
724 * Refcnt of page_cgroup is incremented.
726 int mem_cgroup_prepare_migration(struct page
*page
)
728 struct page_cgroup
*pc
;
730 lock_page_cgroup(page
);
731 pc
= page_get_page_cgroup(page
);
734 unlock_page_cgroup(page
);
738 void mem_cgroup_end_migration(struct page
*page
)
740 mem_cgroup_uncharge_page(page
);
744 * We know both *page* and *newpage* are now not-on-LRU and PG_locked.
745 * And no race with uncharge() routines because page_cgroup for *page*
746 * has extra one reference by mem_cgroup_prepare_migration.
748 void mem_cgroup_page_migration(struct page
*page
, struct page
*newpage
)
750 struct page_cgroup
*pc
;
751 struct mem_cgroup_per_zone
*mz
;
754 lock_page_cgroup(page
);
755 pc
= page_get_page_cgroup(page
);
757 unlock_page_cgroup(page
);
761 page_assign_page_cgroup(page
, NULL
);
762 unlock_page_cgroup(page
);
764 mz
= page_cgroup_zoneinfo(pc
);
765 spin_lock_irqsave(&mz
->lru_lock
, flags
);
766 __mem_cgroup_remove_list(pc
);
767 spin_unlock_irqrestore(&mz
->lru_lock
, flags
);
770 lock_page_cgroup(newpage
);
771 page_assign_page_cgroup(newpage
, pc
);
772 unlock_page_cgroup(newpage
);
774 mz
= page_cgroup_zoneinfo(pc
);
775 spin_lock_irqsave(&mz
->lru_lock
, flags
);
776 __mem_cgroup_add_list(pc
);
777 spin_unlock_irqrestore(&mz
->lru_lock
, flags
);
781 * This routine traverse page_cgroup in given list and drop them all.
782 * This routine ignores page_cgroup->ref_cnt.
783 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
785 #define FORCE_UNCHARGE_BATCH (128)
786 static void mem_cgroup_force_empty_list(struct mem_cgroup
*mem
,
787 struct mem_cgroup_per_zone
*mz
,
790 struct page_cgroup
*pc
;
794 struct list_head
*list
;
797 list
= &mz
->active_list
;
799 list
= &mz
->inactive_list
;
801 if (list_empty(list
))
804 count
= FORCE_UNCHARGE_BATCH
;
805 spin_lock_irqsave(&mz
->lru_lock
, flags
);
807 while (--count
&& !list_empty(list
)) {
808 pc
= list_entry(list
->prev
, struct page_cgroup
, lru
);
810 lock_page_cgroup(page
);
811 if (page_get_page_cgroup(page
) == pc
) {
812 page_assign_page_cgroup(page
, NULL
);
813 unlock_page_cgroup(page
);
814 __mem_cgroup_remove_list(pc
);
815 res_counter_uncharge(&mem
->res
, PAGE_SIZE
);
819 /* racing uncharge: let page go then retry */
820 unlock_page_cgroup(page
);
825 spin_unlock_irqrestore(&mz
->lru_lock
, flags
);
826 if (!list_empty(list
)) {
833 * make mem_cgroup's charge to be 0 if there is no task.
834 * This enables deleting this mem_cgroup.
836 static int mem_cgroup_force_empty(struct mem_cgroup
*mem
)
843 * page reclaim code (kswapd etc..) will move pages between
844 * active_list <-> inactive_list while we don't take a lock.
845 * So, we have to do loop here until all lists are empty.
847 while (mem
->res
.usage
> 0) {
848 if (atomic_read(&mem
->css
.cgroup
->count
) > 0)
850 for_each_node_state(node
, N_POSSIBLE
)
851 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
852 struct mem_cgroup_per_zone
*mz
;
853 mz
= mem_cgroup_zoneinfo(mem
, node
, zid
);
854 /* drop all page_cgroup in active_list */
855 mem_cgroup_force_empty_list(mem
, mz
, 1);
856 /* drop all page_cgroup in inactive_list */
857 mem_cgroup_force_empty_list(mem
, mz
, 0);
866 static int mem_cgroup_write_strategy(char *buf
, unsigned long long *tmp
)
868 *tmp
= memparse(buf
, &buf
);
873 * Round up the value to the closest page size
875 *tmp
= ((*tmp
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
) << PAGE_SHIFT
;
879 static ssize_t
mem_cgroup_read(struct cgroup
*cont
,
880 struct cftype
*cft
, struct file
*file
,
881 char __user
*userbuf
, size_t nbytes
, loff_t
*ppos
)
883 return res_counter_read(&mem_cgroup_from_cont(cont
)->res
,
884 cft
->private, userbuf
, nbytes
, ppos
,
888 static ssize_t
mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
889 struct file
*file
, const char __user
*userbuf
,
890 size_t nbytes
, loff_t
*ppos
)
892 return res_counter_write(&mem_cgroup_from_cont(cont
)->res
,
893 cft
->private, userbuf
, nbytes
, ppos
,
894 mem_cgroup_write_strategy
);
897 static ssize_t
mem_force_empty_write(struct cgroup
*cont
,
898 struct cftype
*cft
, struct file
*file
,
899 const char __user
*userbuf
,
900 size_t nbytes
, loff_t
*ppos
)
902 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
903 int ret
= mem_cgroup_force_empty(mem
);
910 * Note: This should be removed if cgroup supports write-only file.
912 static ssize_t
mem_force_empty_read(struct cgroup
*cont
,
914 struct file
*file
, char __user
*userbuf
,
915 size_t nbytes
, loff_t
*ppos
)
920 static const struct mem_cgroup_stat_desc
{
923 } mem_cgroup_stat_desc
[] = {
924 [MEM_CGROUP_STAT_CACHE
] = { "cache", PAGE_SIZE
, },
925 [MEM_CGROUP_STAT_RSS
] = { "rss", PAGE_SIZE
, },
928 static int mem_control_stat_show(struct seq_file
*m
, void *arg
)
930 struct cgroup
*cont
= m
->private;
931 struct mem_cgroup
*mem_cont
= mem_cgroup_from_cont(cont
);
932 struct mem_cgroup_stat
*stat
= &mem_cont
->stat
;
935 for (i
= 0; i
< ARRAY_SIZE(stat
->cpustat
[0].count
); i
++) {
938 val
= mem_cgroup_read_stat(stat
, i
);
939 val
*= mem_cgroup_stat_desc
[i
].unit
;
940 seq_printf(m
, "%s %lld\n", mem_cgroup_stat_desc
[i
].msg
,
943 /* showing # of active pages */
945 unsigned long active
, inactive
;
947 inactive
= mem_cgroup_get_all_zonestat(mem_cont
,
948 MEM_CGROUP_ZSTAT_INACTIVE
);
949 active
= mem_cgroup_get_all_zonestat(mem_cont
,
950 MEM_CGROUP_ZSTAT_ACTIVE
);
951 seq_printf(m
, "active %ld\n", (active
) * PAGE_SIZE
);
952 seq_printf(m
, "inactive %ld\n", (inactive
) * PAGE_SIZE
);
957 static const struct file_operations mem_control_stat_file_operations
= {
960 .release
= single_release
,
963 static int mem_control_stat_open(struct inode
*unused
, struct file
*file
)
966 struct cgroup
*cont
= file
->f_dentry
->d_parent
->d_fsdata
;
968 file
->f_op
= &mem_control_stat_file_operations
;
969 return single_open(file
, mem_control_stat_show
, cont
);
972 static struct cftype mem_cgroup_files
[] = {
974 .name
= "usage_in_bytes",
975 .private = RES_USAGE
,
976 .read
= mem_cgroup_read
,
979 .name
= "limit_in_bytes",
980 .private = RES_LIMIT
,
981 .write
= mem_cgroup_write
,
982 .read
= mem_cgroup_read
,
986 .private = RES_FAILCNT
,
987 .read
= mem_cgroup_read
,
990 .name
= "force_empty",
991 .write
= mem_force_empty_write
,
992 .read
= mem_force_empty_read
,
996 .open
= mem_control_stat_open
,
1000 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*mem
, int node
)
1002 struct mem_cgroup_per_node
*pn
;
1003 struct mem_cgroup_per_zone
*mz
;
1006 * This routine is called against possible nodes.
1007 * But it's BUG to call kmalloc() against offline node.
1009 * TODO: this routine can waste much memory for nodes which will
1010 * never be onlined. It's better to use memory hotplug callback
1013 if (node_state(node
, N_HIGH_MEMORY
))
1014 pn
= kmalloc_node(sizeof(*pn
), GFP_KERNEL
, node
);
1016 pn
= kmalloc(sizeof(*pn
), GFP_KERNEL
);
1020 mem
->info
.nodeinfo
[node
] = pn
;
1021 memset(pn
, 0, sizeof(*pn
));
1023 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
1024 mz
= &pn
->zoneinfo
[zone
];
1025 INIT_LIST_HEAD(&mz
->active_list
);
1026 INIT_LIST_HEAD(&mz
->inactive_list
);
1027 spin_lock_init(&mz
->lru_lock
);
1032 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*mem
, int node
)
1034 kfree(mem
->info
.nodeinfo
[node
]);
1037 static struct cgroup_subsys_state
*
1038 mem_cgroup_create(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
1040 struct mem_cgroup
*mem
;
1043 if (unlikely((cont
->parent
) == NULL
)) {
1044 mem
= &init_mem_cgroup
;
1045 init_mm
.mem_cgroup
= mem
;
1047 mem
= kzalloc(sizeof(struct mem_cgroup
), GFP_KERNEL
);
1050 return ERR_PTR(-ENOMEM
);
1052 res_counter_init(&mem
->res
);
1054 memset(&mem
->info
, 0, sizeof(mem
->info
));
1056 for_each_node_state(node
, N_POSSIBLE
)
1057 if (alloc_mem_cgroup_per_zone_info(mem
, node
))
1062 for_each_node_state(node
, N_POSSIBLE
)
1063 free_mem_cgroup_per_zone_info(mem
, node
);
1064 if (cont
->parent
!= NULL
)
1066 return ERR_PTR(-ENOMEM
);
1069 static void mem_cgroup_pre_destroy(struct cgroup_subsys
*ss
,
1070 struct cgroup
*cont
)
1072 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
1073 mem_cgroup_force_empty(mem
);
1076 static void mem_cgroup_destroy(struct cgroup_subsys
*ss
,
1077 struct cgroup
*cont
)
1080 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
1082 for_each_node_state(node
, N_POSSIBLE
)
1083 free_mem_cgroup_per_zone_info(mem
, node
);
1085 kfree(mem_cgroup_from_cont(cont
));
1088 static int mem_cgroup_populate(struct cgroup_subsys
*ss
,
1089 struct cgroup
*cont
)
1091 return cgroup_add_files(cont
, ss
, mem_cgroup_files
,
1092 ARRAY_SIZE(mem_cgroup_files
));
1095 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
1096 struct cgroup
*cont
,
1097 struct cgroup
*old_cont
,
1098 struct task_struct
*p
)
1100 struct mm_struct
*mm
;
1101 struct mem_cgroup
*mem
, *old_mem
;
1103 mm
= get_task_mm(p
);
1107 mem
= mem_cgroup_from_cont(cont
);
1108 old_mem
= mem_cgroup_from_cont(old_cont
);
1114 * Only thread group leaders are allowed to migrate, the mm_struct is
1115 * in effect owned by the leader
1117 if (p
->tgid
!= p
->pid
)
1121 rcu_assign_pointer(mm
->mem_cgroup
, mem
);
1122 css_put(&old_mem
->css
);
1128 struct cgroup_subsys mem_cgroup_subsys
= {
1130 .subsys_id
= mem_cgroup_subsys_id
,
1131 .create
= mem_cgroup_create
,
1132 .pre_destroy
= mem_cgroup_pre_destroy
,
1133 .destroy
= mem_cgroup_destroy
,
1134 .populate
= mem_cgroup_populate
,
1135 .attach
= mem_cgroup_move_task
,