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 unsigned long control_type
; /* control RSS or RSS+Pagecache */
135 int prev_priority
; /* for recording reclaim priority */
139 struct mem_cgroup_stat stat
;
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 atleast two
145 * byte aligned (based on comments from Nick Piggin)
147 #define PAGE_CGROUP_LOCK_BIT 0x0
148 #define PAGE_CGROUP_LOCK (1 << PAGE_CGROUP_LOCK_BIT)
151 * A page_cgroup page is associated with every page descriptor. The
152 * page_cgroup helps us identify information about the cgroup
155 struct list_head lru
; /* per cgroup LRU list */
157 struct mem_cgroup
*mem_cgroup
;
158 atomic_t ref_cnt
; /* Helpful when pages move b/w */
159 /* mapped and cached states */
162 #define PAGE_CGROUP_FLAG_CACHE (0x1) /* charged as cache */
163 #define PAGE_CGROUP_FLAG_ACTIVE (0x2) /* page is active in this cgroup */
165 static inline int page_cgroup_nid(struct page_cgroup
*pc
)
167 return page_to_nid(pc
->page
);
170 static inline enum zone_type
page_cgroup_zid(struct page_cgroup
*pc
)
172 return page_zonenum(pc
->page
);
176 MEM_CGROUP_TYPE_UNSPEC
= 0,
177 MEM_CGROUP_TYPE_MAPPED
,
178 MEM_CGROUP_TYPE_CACHED
,
184 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
185 MEM_CGROUP_CHARGE_TYPE_MAPPED
,
190 * Always modified under lru lock. Then, not necessary to preempt_disable()
192 static void mem_cgroup_charge_statistics(struct mem_cgroup
*mem
, int flags
,
195 int val
= (charge
)? 1 : -1;
196 struct mem_cgroup_stat
*stat
= &mem
->stat
;
197 VM_BUG_ON(!irqs_disabled());
199 if (flags
& PAGE_CGROUP_FLAG_CACHE
)
200 __mem_cgroup_stat_add_safe(stat
,
201 MEM_CGROUP_STAT_CACHE
, val
);
203 __mem_cgroup_stat_add_safe(stat
, MEM_CGROUP_STAT_RSS
, val
);
206 static inline struct mem_cgroup_per_zone
*
207 mem_cgroup_zoneinfo(struct mem_cgroup
*mem
, int nid
, int zid
)
209 BUG_ON(!mem
->info
.nodeinfo
[nid
]);
210 return &mem
->info
.nodeinfo
[nid
]->zoneinfo
[zid
];
213 static inline 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
,
224 enum mem_cgroup_zstat_index idx
)
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 init_mem_cgroup
;
241 struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
243 return container_of(cgroup_subsys_state(cont
,
244 mem_cgroup_subsys_id
), struct mem_cgroup
,
249 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
251 return container_of(task_subsys_state(p
, mem_cgroup_subsys_id
),
252 struct mem_cgroup
, css
);
255 void mm_init_cgroup(struct mm_struct
*mm
, struct task_struct
*p
)
257 struct mem_cgroup
*mem
;
259 mem
= mem_cgroup_from_task(p
);
261 mm
->mem_cgroup
= mem
;
264 void mm_free_cgroup(struct mm_struct
*mm
)
266 css_put(&mm
->mem_cgroup
->css
);
269 static inline int page_cgroup_locked(struct page
*page
)
271 return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT
,
275 void page_assign_page_cgroup(struct page
*page
, struct page_cgroup
*pc
)
280 * While resetting the page_cgroup we might not hold the
281 * page_cgroup lock. free_hot_cold_page() is an example
285 VM_BUG_ON(!page_cgroup_locked(page
));
286 locked
= (page
->page_cgroup
& PAGE_CGROUP_LOCK
);
287 page
->page_cgroup
= ((unsigned long)pc
| locked
);
290 struct page_cgroup
*page_get_page_cgroup(struct page
*page
)
292 return (struct page_cgroup
*)
293 (page
->page_cgroup
& ~PAGE_CGROUP_LOCK
);
296 static void __always_inline
lock_page_cgroup(struct page
*page
)
298 bit_spin_lock(PAGE_CGROUP_LOCK_BIT
, &page
->page_cgroup
);
299 VM_BUG_ON(!page_cgroup_locked(page
));
302 static void __always_inline
unlock_page_cgroup(struct page
*page
)
304 bit_spin_unlock(PAGE_CGROUP_LOCK_BIT
, &page
->page_cgroup
);
308 * Tie new page_cgroup to struct page under lock_page_cgroup()
309 * This can fail if the page has been tied to a page_cgroup.
310 * If success, returns 0.
312 static int page_cgroup_assign_new_page_cgroup(struct page
*page
,
313 struct page_cgroup
*pc
)
317 lock_page_cgroup(page
);
318 if (!page_get_page_cgroup(page
))
319 page_assign_page_cgroup(page
, pc
);
320 else /* A page is tied to other pc. */
322 unlock_page_cgroup(page
);
327 * Clear page->page_cgroup member under lock_page_cgroup().
328 * If given "pc" value is different from one page->page_cgroup,
329 * page->cgroup is not cleared.
330 * Returns a value of page->page_cgroup at lock taken.
331 * A can can detect failure of clearing by following
332 * clear_page_cgroup(page, pc) == pc
335 static struct page_cgroup
*clear_page_cgroup(struct page
*page
,
336 struct page_cgroup
*pc
)
338 struct page_cgroup
*ret
;
340 lock_page_cgroup(page
);
341 ret
= page_get_page_cgroup(page
);
342 if (likely(ret
== pc
))
343 page_assign_page_cgroup(page
, NULL
);
344 unlock_page_cgroup(page
);
348 static void __mem_cgroup_remove_list(struct page_cgroup
*pc
)
350 int from
= pc
->flags
& PAGE_CGROUP_FLAG_ACTIVE
;
351 struct mem_cgroup_per_zone
*mz
= page_cgroup_zoneinfo(pc
);
354 MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_ACTIVE
) -= 1;
356 MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_INACTIVE
) -= 1;
358 mem_cgroup_charge_statistics(pc
->mem_cgroup
, pc
->flags
, false);
359 list_del_init(&pc
->lru
);
362 static void __mem_cgroup_add_list(struct page_cgroup
*pc
)
364 int to
= pc
->flags
& PAGE_CGROUP_FLAG_ACTIVE
;
365 struct mem_cgroup_per_zone
*mz
= page_cgroup_zoneinfo(pc
);
368 MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_INACTIVE
) += 1;
369 list_add(&pc
->lru
, &mz
->inactive_list
);
371 MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_ACTIVE
) += 1;
372 list_add(&pc
->lru
, &mz
->active_list
);
374 mem_cgroup_charge_statistics(pc
->mem_cgroup
, pc
->flags
, true);
377 static void __mem_cgroup_move_lists(struct page_cgroup
*pc
, bool active
)
379 int from
= pc
->flags
& PAGE_CGROUP_FLAG_ACTIVE
;
380 struct mem_cgroup_per_zone
*mz
= page_cgroup_zoneinfo(pc
);
383 MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_ACTIVE
) -= 1;
385 MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_INACTIVE
) -= 1;
388 MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_ACTIVE
) += 1;
389 pc
->flags
|= PAGE_CGROUP_FLAG_ACTIVE
;
390 list_move(&pc
->lru
, &mz
->active_list
);
392 MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_INACTIVE
) += 1;
393 pc
->flags
&= ~PAGE_CGROUP_FLAG_ACTIVE
;
394 list_move(&pc
->lru
, &mz
->inactive_list
);
398 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*mem
)
403 ret
= task
->mm
&& mm_cgroup(task
->mm
) == mem
;
409 * This routine assumes that the appropriate zone's lru lock is already held
411 void mem_cgroup_move_lists(struct page_cgroup
*pc
, bool active
)
413 struct mem_cgroup_per_zone
*mz
;
419 mz
= page_cgroup_zoneinfo(pc
);
420 spin_lock_irqsave(&mz
->lru_lock
, flags
);
421 __mem_cgroup_move_lists(pc
, active
);
422 spin_unlock_irqrestore(&mz
->lru_lock
, flags
);
426 * Calculate mapped_ratio under memory controller. This will be used in
427 * vmscan.c for deteremining we have to reclaim mapped pages.
429 int mem_cgroup_calc_mapped_ratio(struct mem_cgroup
*mem
)
434 * usage is recorded in bytes. But, here, we assume the number of
435 * physical pages can be represented by "long" on any arch.
437 total
= (long) (mem
->res
.usage
>> PAGE_SHIFT
) + 1L;
438 rss
= (long)mem_cgroup_read_stat(&mem
->stat
, MEM_CGROUP_STAT_RSS
);
439 return (int)((rss
* 100L) / total
);
442 * This function is called from vmscan.c. In page reclaiming loop. balance
443 * between active and inactive list is calculated. For memory controller
444 * page reclaiming, we should use using mem_cgroup's imbalance rather than
445 * zone's global lru imbalance.
447 long mem_cgroup_reclaim_imbalance(struct mem_cgroup
*mem
)
449 unsigned long active
, inactive
;
450 /* active and inactive are the number of pages. 'long' is ok.*/
451 active
= mem_cgroup_get_all_zonestat(mem
, MEM_CGROUP_ZSTAT_ACTIVE
);
452 inactive
= mem_cgroup_get_all_zonestat(mem
, MEM_CGROUP_ZSTAT_INACTIVE
);
453 return (long) (active
/ (inactive
+ 1));
457 * prev_priority control...this will be used in memory reclaim path.
459 int mem_cgroup_get_reclaim_priority(struct mem_cgroup
*mem
)
461 return mem
->prev_priority
;
464 void mem_cgroup_note_reclaim_priority(struct mem_cgroup
*mem
, int priority
)
466 if (priority
< mem
->prev_priority
)
467 mem
->prev_priority
= priority
;
470 void mem_cgroup_record_reclaim_priority(struct mem_cgroup
*mem
, int priority
)
472 mem
->prev_priority
= priority
;
476 * Calculate # of pages to be scanned in this priority/zone.
479 * priority starts from "DEF_PRIORITY" and decremented in each loop.
480 * (see include/linux/mmzone.h)
483 long mem_cgroup_calc_reclaim_active(struct mem_cgroup
*mem
,
484 struct zone
*zone
, int priority
)
487 int nid
= zone
->zone_pgdat
->node_id
;
488 int zid
= zone_idx(zone
);
489 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(mem
, nid
, zid
);
491 nr_active
= MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_ACTIVE
);
492 return (nr_active
>> priority
);
495 long mem_cgroup_calc_reclaim_inactive(struct mem_cgroup
*mem
,
496 struct zone
*zone
, int priority
)
499 int nid
= zone
->zone_pgdat
->node_id
;
500 int zid
= zone_idx(zone
);
501 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(mem
, nid
, zid
);
503 nr_inactive
= MEM_CGROUP_ZSTAT(mz
, MEM_CGROUP_ZSTAT_INACTIVE
);
505 return (nr_inactive
>> priority
);
508 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan
,
509 struct list_head
*dst
,
510 unsigned long *scanned
, int order
,
511 int mode
, struct zone
*z
,
512 struct mem_cgroup
*mem_cont
,
515 unsigned long nr_taken
= 0;
519 struct list_head
*src
;
520 struct page_cgroup
*pc
, *tmp
;
521 int nid
= z
->zone_pgdat
->node_id
;
522 int zid
= zone_idx(z
);
523 struct mem_cgroup_per_zone
*mz
;
525 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
527 src
= &mz
->active_list
;
529 src
= &mz
->inactive_list
;
532 spin_lock(&mz
->lru_lock
);
534 list_for_each_entry_safe_reverse(pc
, tmp
, src
, lru
) {
535 if (scan
>= nr_to_scan
)
540 if (unlikely(!PageLRU(page
)))
543 if (PageActive(page
) && !active
) {
544 __mem_cgroup_move_lists(pc
, true);
547 if (!PageActive(page
) && active
) {
548 __mem_cgroup_move_lists(pc
, false);
553 list_move(&pc
->lru
, &pc_list
);
555 if (__isolate_lru_page(page
, mode
) == 0) {
556 list_move(&page
->lru
, dst
);
561 list_splice(&pc_list
, src
);
562 spin_unlock(&mz
->lru_lock
);
569 * Charge the memory controller for page usage.
571 * 0 if the charge was successful
572 * < 0 if the cgroup is over its limit
574 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
575 gfp_t gfp_mask
, enum charge_type ctype
)
577 struct mem_cgroup
*mem
;
578 struct page_cgroup
*pc
;
580 unsigned long nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
581 struct mem_cgroup_per_zone
*mz
;
584 * Should page_cgroup's go to their own slab?
585 * One could optimize the performance of the charging routine
586 * by saving a bit in the page_flags and using it as a lock
587 * to see if the cgroup page already has a page_cgroup associated
592 lock_page_cgroup(page
);
593 pc
= page_get_page_cgroup(page
);
595 * The page_cgroup exists and
596 * the page has already been accounted.
599 if (unlikely(!atomic_inc_not_zero(&pc
->ref_cnt
))) {
600 /* this page is under being uncharged ? */
601 unlock_page_cgroup(page
);
605 unlock_page_cgroup(page
);
609 unlock_page_cgroup(page
);
612 pc
= kzalloc(sizeof(struct page_cgroup
), gfp_mask
);
617 * We always charge the cgroup the mm_struct belongs to.
618 * The mm_struct's mem_cgroup changes on task migration if the
619 * thread group leader migrates. It's possible that mm is not
620 * set, if so charge the init_mm (happens for pagecache usage).
626 mem
= rcu_dereference(mm
->mem_cgroup
);
628 * For every charge from the cgroup, increment reference
635 * If we created the page_cgroup, we should free it on exceeding
638 while (res_counter_charge(&mem
->res
, PAGE_SIZE
)) {
639 if (!(gfp_mask
& __GFP_WAIT
))
642 if (try_to_free_mem_cgroup_pages(mem
, gfp_mask
))
646 * try_to_free_mem_cgroup_pages() might not give us a full
647 * picture of reclaim. Some pages are reclaimed and might be
648 * moved to swap cache or just unmapped from the cgroup.
649 * Check the limit again to see if the reclaim reduced the
650 * current usage of the cgroup before giving up
652 if (res_counter_check_under_limit(&mem
->res
))
656 mem_cgroup_out_of_memory(mem
, gfp_mask
);
659 congestion_wait(WRITE
, HZ
/10);
662 atomic_set(&pc
->ref_cnt
, 1);
663 pc
->mem_cgroup
= mem
;
665 pc
->flags
= PAGE_CGROUP_FLAG_ACTIVE
;
666 if (ctype
== MEM_CGROUP_CHARGE_TYPE_CACHE
)
667 pc
->flags
|= PAGE_CGROUP_FLAG_CACHE
;
669 if (!page
|| page_cgroup_assign_new_page_cgroup(page
, pc
)) {
671 * Another charge has been added to this page already.
672 * We take lock_page_cgroup(page) again and read
673 * page->cgroup, increment refcnt.... just retry is OK.
675 res_counter_uncharge(&mem
->res
, PAGE_SIZE
);
683 mz
= page_cgroup_zoneinfo(pc
);
684 spin_lock_irqsave(&mz
->lru_lock
, flags
);
685 /* Update statistics vector */
686 __mem_cgroup_add_list(pc
);
687 spin_unlock_irqrestore(&mz
->lru_lock
, flags
);
698 int mem_cgroup_charge(struct page
*page
, struct mm_struct
*mm
,
701 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
702 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
706 * See if the cached pages should be charged at all?
708 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
712 struct mem_cgroup
*mem
;
717 mem
= rcu_dereference(mm
->mem_cgroup
);
720 if (mem
->control_type
== MEM_CGROUP_TYPE_ALL
)
721 ret
= mem_cgroup_charge_common(page
, mm
, gfp_mask
,
722 MEM_CGROUP_CHARGE_TYPE_CACHE
);
728 * Uncharging is always a welcome operation, we never complain, simply
731 void mem_cgroup_uncharge(struct page_cgroup
*pc
)
733 struct mem_cgroup
*mem
;
734 struct mem_cgroup_per_zone
*mz
;
739 * This can handle cases when a page is not charged at all and we
740 * are switching between handling the control_type.
745 if (atomic_dec_and_test(&pc
->ref_cnt
)) {
747 mz
= page_cgroup_zoneinfo(pc
);
749 * get page->cgroup and clear it under lock.
750 * force_empty can drop page->cgroup without checking refcnt.
752 if (clear_page_cgroup(page
, pc
) == pc
) {
753 mem
= pc
->mem_cgroup
;
755 res_counter_uncharge(&mem
->res
, PAGE_SIZE
);
756 spin_lock_irqsave(&mz
->lru_lock
, flags
);
757 __mem_cgroup_remove_list(pc
);
758 spin_unlock_irqrestore(&mz
->lru_lock
, flags
);
765 * Returns non-zero if a page (under migration) has valid page_cgroup member.
766 * Refcnt of page_cgroup is incremented.
769 int mem_cgroup_prepare_migration(struct page
*page
)
771 struct page_cgroup
*pc
;
773 lock_page_cgroup(page
);
774 pc
= page_get_page_cgroup(page
);
775 if (pc
&& atomic_inc_not_zero(&pc
->ref_cnt
))
777 unlock_page_cgroup(page
);
781 void mem_cgroup_end_migration(struct page
*page
)
783 struct page_cgroup
*pc
= page_get_page_cgroup(page
);
784 mem_cgroup_uncharge(pc
);
787 * We know both *page* and *newpage* are now not-on-LRU and Pg_locked.
788 * And no race with uncharge() routines because page_cgroup for *page*
789 * has extra one reference by mem_cgroup_prepare_migration.
792 void mem_cgroup_page_migration(struct page
*page
, struct page
*newpage
)
794 struct page_cgroup
*pc
;
795 struct mem_cgroup
*mem
;
797 struct mem_cgroup_per_zone
*mz
;
799 pc
= page_get_page_cgroup(page
);
802 mem
= pc
->mem_cgroup
;
803 mz
= page_cgroup_zoneinfo(pc
);
804 if (clear_page_cgroup(page
, pc
) != pc
)
806 spin_lock_irqsave(&mz
->lru_lock
, flags
);
808 __mem_cgroup_remove_list(pc
);
809 spin_unlock_irqrestore(&mz
->lru_lock
, flags
);
812 lock_page_cgroup(newpage
);
813 page_assign_page_cgroup(newpage
, pc
);
814 unlock_page_cgroup(newpage
);
816 mz
= page_cgroup_zoneinfo(pc
);
817 spin_lock_irqsave(&mz
->lru_lock
, flags
);
818 __mem_cgroup_add_list(pc
);
819 spin_unlock_irqrestore(&mz
->lru_lock
, flags
);
824 * This routine traverse page_cgroup in given list and drop them all.
825 * This routine ignores page_cgroup->ref_cnt.
826 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
828 #define FORCE_UNCHARGE_BATCH (128)
830 mem_cgroup_force_empty_list(struct mem_cgroup
*mem
,
831 struct mem_cgroup_per_zone
*mz
,
834 struct page_cgroup
*pc
;
838 struct list_head
*list
;
841 list
= &mz
->active_list
;
843 list
= &mz
->inactive_list
;
845 if (list_empty(list
))
848 count
= FORCE_UNCHARGE_BATCH
;
849 spin_lock_irqsave(&mz
->lru_lock
, flags
);
851 while (--count
&& !list_empty(list
)) {
852 pc
= list_entry(list
->prev
, struct page_cgroup
, lru
);
854 /* Avoid race with charge */
855 atomic_set(&pc
->ref_cnt
, 0);
856 if (clear_page_cgroup(page
, pc
) == pc
) {
858 res_counter_uncharge(&mem
->res
, PAGE_SIZE
);
859 __mem_cgroup_remove_list(pc
);
861 } else /* being uncharged ? ...do relax */
864 spin_unlock_irqrestore(&mz
->lru_lock
, flags
);
865 if (!list_empty(list
)) {
873 * make mem_cgroup's charge to be 0 if there is no task.
874 * This enables deleting this mem_cgroup.
877 int mem_cgroup_force_empty(struct mem_cgroup
*mem
)
883 * page reclaim code (kswapd etc..) will move pages between
884 ` * active_list <-> inactive_list while we don't take a lock.
885 * So, we have to do loop here until all lists are empty.
887 while (mem
->res
.usage
> 0) {
888 if (atomic_read(&mem
->css
.cgroup
->count
) > 0)
890 for_each_node_state(node
, N_POSSIBLE
)
891 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
892 struct mem_cgroup_per_zone
*mz
;
893 mz
= mem_cgroup_zoneinfo(mem
, node
, zid
);
894 /* drop all page_cgroup in active_list */
895 mem_cgroup_force_empty_list(mem
, mz
, 1);
896 /* drop all page_cgroup in inactive_list */
897 mem_cgroup_force_empty_list(mem
, mz
, 0);
908 int mem_cgroup_write_strategy(char *buf
, unsigned long long *tmp
)
910 *tmp
= memparse(buf
, &buf
);
915 * Round up the value to the closest page size
917 *tmp
= ((*tmp
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
) << PAGE_SHIFT
;
921 static ssize_t
mem_cgroup_read(struct cgroup
*cont
,
922 struct cftype
*cft
, struct file
*file
,
923 char __user
*userbuf
, size_t nbytes
, loff_t
*ppos
)
925 return res_counter_read(&mem_cgroup_from_cont(cont
)->res
,
926 cft
->private, userbuf
, nbytes
, ppos
,
930 static ssize_t
mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
931 struct file
*file
, const char __user
*userbuf
,
932 size_t nbytes
, loff_t
*ppos
)
934 return res_counter_write(&mem_cgroup_from_cont(cont
)->res
,
935 cft
->private, userbuf
, nbytes
, ppos
,
936 mem_cgroup_write_strategy
);
939 static ssize_t
mem_control_type_write(struct cgroup
*cont
,
940 struct cftype
*cft
, struct file
*file
,
941 const char __user
*userbuf
,
942 size_t nbytes
, loff_t
*pos
)
947 struct mem_cgroup
*mem
;
949 mem
= mem_cgroup_from_cont(cont
);
950 buf
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
957 if (copy_from_user(buf
, userbuf
, nbytes
))
961 tmp
= simple_strtoul(buf
, &end
, 10);
965 if (tmp
<= MEM_CGROUP_TYPE_UNSPEC
|| tmp
>= MEM_CGROUP_TYPE_MAX
)
968 mem
->control_type
= tmp
;
976 static ssize_t
mem_control_type_read(struct cgroup
*cont
,
978 struct file
*file
, char __user
*userbuf
,
979 size_t nbytes
, loff_t
*ppos
)
983 struct mem_cgroup
*mem
;
985 mem
= mem_cgroup_from_cont(cont
);
987 val
= mem
->control_type
;
988 s
+= sprintf(s
, "%lu\n", val
);
989 return simple_read_from_buffer((void __user
*)userbuf
, nbytes
,
994 static ssize_t
mem_force_empty_write(struct cgroup
*cont
,
995 struct cftype
*cft
, struct file
*file
,
996 const char __user
*userbuf
,
997 size_t nbytes
, loff_t
*ppos
)
999 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
1001 ret
= mem_cgroup_force_empty(mem
);
1008 * Note: This should be removed if cgroup supports write-only file.
1011 static ssize_t
mem_force_empty_read(struct cgroup
*cont
,
1013 struct file
*file
, char __user
*userbuf
,
1014 size_t nbytes
, loff_t
*ppos
)
1020 static const struct mem_cgroup_stat_desc
{
1023 } mem_cgroup_stat_desc
[] = {
1024 [MEM_CGROUP_STAT_CACHE
] = { "cache", PAGE_SIZE
, },
1025 [MEM_CGROUP_STAT_RSS
] = { "rss", PAGE_SIZE
, },
1028 static int mem_control_stat_show(struct seq_file
*m
, void *arg
)
1030 struct cgroup
*cont
= m
->private;
1031 struct mem_cgroup
*mem_cont
= mem_cgroup_from_cont(cont
);
1032 struct mem_cgroup_stat
*stat
= &mem_cont
->stat
;
1035 for (i
= 0; i
< ARRAY_SIZE(stat
->cpustat
[0].count
); i
++) {
1038 val
= mem_cgroup_read_stat(stat
, i
);
1039 val
*= mem_cgroup_stat_desc
[i
].unit
;
1040 seq_printf(m
, "%s %lld\n", mem_cgroup_stat_desc
[i
].msg
,
1043 /* showing # of active pages */
1045 unsigned long active
, inactive
;
1047 inactive
= mem_cgroup_get_all_zonestat(mem_cont
,
1048 MEM_CGROUP_ZSTAT_INACTIVE
);
1049 active
= mem_cgroup_get_all_zonestat(mem_cont
,
1050 MEM_CGROUP_ZSTAT_ACTIVE
);
1051 seq_printf(m
, "active %ld\n", (active
) * PAGE_SIZE
);
1052 seq_printf(m
, "inactive %ld\n", (inactive
) * PAGE_SIZE
);
1057 static const struct file_operations mem_control_stat_file_operations
= {
1059 .llseek
= seq_lseek
,
1060 .release
= single_release
,
1063 static int mem_control_stat_open(struct inode
*unused
, struct file
*file
)
1066 struct cgroup
*cont
= file
->f_dentry
->d_parent
->d_fsdata
;
1068 file
->f_op
= &mem_control_stat_file_operations
;
1069 return single_open(file
, mem_control_stat_show
, cont
);
1074 static struct cftype mem_cgroup_files
[] = {
1076 .name
= "usage_in_bytes",
1077 .private = RES_USAGE
,
1078 .read
= mem_cgroup_read
,
1081 .name
= "limit_in_bytes",
1082 .private = RES_LIMIT
,
1083 .write
= mem_cgroup_write
,
1084 .read
= mem_cgroup_read
,
1088 .private = RES_FAILCNT
,
1089 .read
= mem_cgroup_read
,
1092 .name
= "control_type",
1093 .write
= mem_control_type_write
,
1094 .read
= mem_control_type_read
,
1097 .name
= "force_empty",
1098 .write
= mem_force_empty_write
,
1099 .read
= mem_force_empty_read
,
1103 .open
= mem_control_stat_open
,
1107 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*mem
, int node
)
1109 struct mem_cgroup_per_node
*pn
;
1110 struct mem_cgroup_per_zone
*mz
;
1113 * This routine is called against possible nodes.
1114 * But it's BUG to call kmalloc() against offline node.
1116 * TODO: this routine can waste much memory for nodes which will
1117 * never be onlined. It's better to use memory hotplug callback
1120 if (node_state(node
, N_HIGH_MEMORY
))
1121 pn
= kmalloc_node(sizeof(*pn
), GFP_KERNEL
, node
);
1123 pn
= kmalloc(sizeof(*pn
), GFP_KERNEL
);
1127 mem
->info
.nodeinfo
[node
] = pn
;
1128 memset(pn
, 0, sizeof(*pn
));
1130 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
1131 mz
= &pn
->zoneinfo
[zone
];
1132 INIT_LIST_HEAD(&mz
->active_list
);
1133 INIT_LIST_HEAD(&mz
->inactive_list
);
1134 spin_lock_init(&mz
->lru_lock
);
1139 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*mem
, int node
)
1141 kfree(mem
->info
.nodeinfo
[node
]);
1145 static struct mem_cgroup init_mem_cgroup
;
1147 static struct cgroup_subsys_state
*
1148 mem_cgroup_create(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
1150 struct mem_cgroup
*mem
;
1153 if (unlikely((cont
->parent
) == NULL
)) {
1154 mem
= &init_mem_cgroup
;
1155 init_mm
.mem_cgroup
= mem
;
1157 mem
= kzalloc(sizeof(struct mem_cgroup
), GFP_KERNEL
);
1162 res_counter_init(&mem
->res
);
1164 mem
->control_type
= MEM_CGROUP_TYPE_ALL
;
1165 memset(&mem
->info
, 0, sizeof(mem
->info
));
1167 for_each_node_state(node
, N_POSSIBLE
)
1168 if (alloc_mem_cgroup_per_zone_info(mem
, node
))
1173 for_each_node_state(node
, N_POSSIBLE
)
1174 free_mem_cgroup_per_zone_info(mem
, node
);
1175 if (cont
->parent
!= NULL
)
1180 static void mem_cgroup_pre_destroy(struct cgroup_subsys
*ss
,
1181 struct cgroup
*cont
)
1183 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
1184 mem_cgroup_force_empty(mem
);
1187 static void mem_cgroup_destroy(struct cgroup_subsys
*ss
,
1188 struct cgroup
*cont
)
1191 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
1193 for_each_node_state(node
, N_POSSIBLE
)
1194 free_mem_cgroup_per_zone_info(mem
, node
);
1196 kfree(mem_cgroup_from_cont(cont
));
1199 static int mem_cgroup_populate(struct cgroup_subsys
*ss
,
1200 struct cgroup
*cont
)
1202 return cgroup_add_files(cont
, ss
, mem_cgroup_files
,
1203 ARRAY_SIZE(mem_cgroup_files
));
1206 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
1207 struct cgroup
*cont
,
1208 struct cgroup
*old_cont
,
1209 struct task_struct
*p
)
1211 struct mm_struct
*mm
;
1212 struct mem_cgroup
*mem
, *old_mem
;
1214 mm
= get_task_mm(p
);
1218 mem
= mem_cgroup_from_cont(cont
);
1219 old_mem
= mem_cgroup_from_cont(old_cont
);
1225 * Only thread group leaders are allowed to migrate, the mm_struct is
1226 * in effect owned by the leader
1228 if (p
->tgid
!= p
->pid
)
1232 rcu_assign_pointer(mm
->mem_cgroup
, mem
);
1233 css_put(&old_mem
->css
);
1240 struct cgroup_subsys mem_cgroup_subsys
= {
1242 .subsys_id
= mem_cgroup_subsys_id
,
1243 .create
= mem_cgroup_create
,
1244 .pre_destroy
= mem_cgroup_pre_destroy
,
1245 .destroy
= mem_cgroup_destroy
,
1246 .populate
= mem_cgroup_populate
,
1247 .attach
= mem_cgroup_move_task
,