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 * The memory controller data structure. The memory controller controls both
82 * page cache and RSS per cgroup. We would eventually like to provide
83 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
84 * to help the administrator determine what knobs to tune.
86 * TODO: Add a water mark for the memory controller. Reclaim will begin when
87 * we hit the water mark. May be even add a low water mark, such that
88 * no reclaim occurs from a cgroup at it's low water mark, this is
89 * a feature that will be implemented much later in the future.
92 struct cgroup_subsys_state css
;
94 * the counter to account for memory usage
96 struct res_counter res
;
98 * Per cgroup active and inactive list, similar to the
100 * TODO: Consider making these lists per zone
102 struct list_head active_list
;
103 struct list_head inactive_list
;
105 * spin_lock to protect the per cgroup LRU
108 unsigned long control_type
; /* control RSS or RSS+Pagecache */
112 struct mem_cgroup_stat stat
;
116 * We use the lower bit of the page->page_cgroup pointer as a bit spin
117 * lock. We need to ensure that page->page_cgroup is atleast two
118 * byte aligned (based on comments from Nick Piggin)
120 #define PAGE_CGROUP_LOCK_BIT 0x0
121 #define PAGE_CGROUP_LOCK (1 << PAGE_CGROUP_LOCK_BIT)
124 * A page_cgroup page is associated with every page descriptor. The
125 * page_cgroup helps us identify information about the cgroup
128 struct list_head lru
; /* per cgroup LRU list */
130 struct mem_cgroup
*mem_cgroup
;
131 atomic_t ref_cnt
; /* Helpful when pages move b/w */
132 /* mapped and cached states */
135 #define PAGE_CGROUP_FLAG_CACHE (0x1) /* charged as cache */
136 #define PAGE_CGROUP_FLAG_ACTIVE (0x2) /* page is active in this cgroup */
138 static inline int page_cgroup_nid(struct page_cgroup
*pc
)
140 return page_to_nid(pc
->page
);
143 static inline enum zone_type
page_cgroup_zid(struct page_cgroup
*pc
)
145 return page_zonenum(pc
->page
);
149 MEM_CGROUP_TYPE_UNSPEC
= 0,
150 MEM_CGROUP_TYPE_MAPPED
,
151 MEM_CGROUP_TYPE_CACHED
,
157 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
158 MEM_CGROUP_CHARGE_TYPE_MAPPED
,
162 * Always modified under lru lock. Then, not necessary to preempt_disable()
164 static void mem_cgroup_charge_statistics(struct mem_cgroup
*mem
, int flags
,
167 int val
= (charge
)? 1 : -1;
168 struct mem_cgroup_stat
*stat
= &mem
->stat
;
169 VM_BUG_ON(!irqs_disabled());
171 if (flags
& PAGE_CGROUP_FLAG_CACHE
)
172 __mem_cgroup_stat_add_safe(stat
,
173 MEM_CGROUP_STAT_CACHE
, val
);
175 __mem_cgroup_stat_add_safe(stat
, MEM_CGROUP_STAT_RSS
, val
);
179 static struct mem_cgroup init_mem_cgroup
;
182 struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
184 return container_of(cgroup_subsys_state(cont
,
185 mem_cgroup_subsys_id
), struct mem_cgroup
,
190 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
192 return container_of(task_subsys_state(p
, mem_cgroup_subsys_id
),
193 struct mem_cgroup
, css
);
196 void mm_init_cgroup(struct mm_struct
*mm
, struct task_struct
*p
)
198 struct mem_cgroup
*mem
;
200 mem
= mem_cgroup_from_task(p
);
202 mm
->mem_cgroup
= mem
;
205 void mm_free_cgroup(struct mm_struct
*mm
)
207 css_put(&mm
->mem_cgroup
->css
);
210 static inline int page_cgroup_locked(struct page
*page
)
212 return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT
,
216 void page_assign_page_cgroup(struct page
*page
, struct page_cgroup
*pc
)
221 * While resetting the page_cgroup we might not hold the
222 * page_cgroup lock. free_hot_cold_page() is an example
226 VM_BUG_ON(!page_cgroup_locked(page
));
227 locked
= (page
->page_cgroup
& PAGE_CGROUP_LOCK
);
228 page
->page_cgroup
= ((unsigned long)pc
| locked
);
231 struct page_cgroup
*page_get_page_cgroup(struct page
*page
)
233 return (struct page_cgroup
*)
234 (page
->page_cgroup
& ~PAGE_CGROUP_LOCK
);
237 static void __always_inline
lock_page_cgroup(struct page
*page
)
239 bit_spin_lock(PAGE_CGROUP_LOCK_BIT
, &page
->page_cgroup
);
240 VM_BUG_ON(!page_cgroup_locked(page
));
243 static void __always_inline
unlock_page_cgroup(struct page
*page
)
245 bit_spin_unlock(PAGE_CGROUP_LOCK_BIT
, &page
->page_cgroup
);
249 * Tie new page_cgroup to struct page under lock_page_cgroup()
250 * This can fail if the page has been tied to a page_cgroup.
251 * If success, returns 0.
253 static int page_cgroup_assign_new_page_cgroup(struct page
*page
,
254 struct page_cgroup
*pc
)
258 lock_page_cgroup(page
);
259 if (!page_get_page_cgroup(page
))
260 page_assign_page_cgroup(page
, pc
);
261 else /* A page is tied to other pc. */
263 unlock_page_cgroup(page
);
268 * Clear page->page_cgroup member under lock_page_cgroup().
269 * If given "pc" value is different from one page->page_cgroup,
270 * page->cgroup is not cleared.
271 * Returns a value of page->page_cgroup at lock taken.
272 * A can can detect failure of clearing by following
273 * clear_page_cgroup(page, pc) == pc
276 static struct page_cgroup
*clear_page_cgroup(struct page
*page
,
277 struct page_cgroup
*pc
)
279 struct page_cgroup
*ret
;
281 lock_page_cgroup(page
);
282 ret
= page_get_page_cgroup(page
);
283 if (likely(ret
== pc
))
284 page_assign_page_cgroup(page
, NULL
);
285 unlock_page_cgroup(page
);
289 static void __mem_cgroup_move_lists(struct page_cgroup
*pc
, bool active
)
292 pc
->flags
|= PAGE_CGROUP_FLAG_ACTIVE
;
293 list_move(&pc
->lru
, &pc
->mem_cgroup
->active_list
);
295 pc
->flags
&= ~PAGE_CGROUP_FLAG_ACTIVE
;
296 list_move(&pc
->lru
, &pc
->mem_cgroup
->inactive_list
);
300 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*mem
)
305 ret
= task
->mm
&& mm_cgroup(task
->mm
) == mem
;
311 * This routine assumes that the appropriate zone's lru lock is already held
313 void mem_cgroup_move_lists(struct page_cgroup
*pc
, bool active
)
315 struct mem_cgroup
*mem
;
319 mem
= pc
->mem_cgroup
;
321 spin_lock(&mem
->lru_lock
);
322 __mem_cgroup_move_lists(pc
, active
);
323 spin_unlock(&mem
->lru_lock
);
326 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan
,
327 struct list_head
*dst
,
328 unsigned long *scanned
, int order
,
329 int mode
, struct zone
*z
,
330 struct mem_cgroup
*mem_cont
,
333 unsigned long nr_taken
= 0;
337 struct list_head
*src
;
338 struct page_cgroup
*pc
, *tmp
;
341 src
= &mem_cont
->active_list
;
343 src
= &mem_cont
->inactive_list
;
345 spin_lock(&mem_cont
->lru_lock
);
347 list_for_each_entry_safe_reverse(pc
, tmp
, src
, lru
) {
348 if (scan
>= nr_to_scan
)
353 if (unlikely(!PageLRU(page
)))
356 if (PageActive(page
) && !active
) {
357 __mem_cgroup_move_lists(pc
, true);
360 if (!PageActive(page
) && active
) {
361 __mem_cgroup_move_lists(pc
, false);
367 * TODO: make the active/inactive lists per zone
369 if (page_zone(page
) != z
)
373 list_move(&pc
->lru
, &pc_list
);
375 if (__isolate_lru_page(page
, mode
) == 0) {
376 list_move(&page
->lru
, dst
);
381 list_splice(&pc_list
, src
);
382 spin_unlock(&mem_cont
->lru_lock
);
389 * Charge the memory controller for page usage.
391 * 0 if the charge was successful
392 * < 0 if the cgroup is over its limit
394 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
395 gfp_t gfp_mask
, enum charge_type ctype
)
397 struct mem_cgroup
*mem
;
398 struct page_cgroup
*pc
;
400 unsigned long nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
403 * Should page_cgroup's go to their own slab?
404 * One could optimize the performance of the charging routine
405 * by saving a bit in the page_flags and using it as a lock
406 * to see if the cgroup page already has a page_cgroup associated
411 lock_page_cgroup(page
);
412 pc
= page_get_page_cgroup(page
);
414 * The page_cgroup exists and
415 * the page has already been accounted.
418 if (unlikely(!atomic_inc_not_zero(&pc
->ref_cnt
))) {
419 /* this page is under being uncharged ? */
420 unlock_page_cgroup(page
);
424 unlock_page_cgroup(page
);
428 unlock_page_cgroup(page
);
431 pc
= kzalloc(sizeof(struct page_cgroup
), gfp_mask
);
436 * We always charge the cgroup the mm_struct belongs to.
437 * The mm_struct's mem_cgroup changes on task migration if the
438 * thread group leader migrates. It's possible that mm is not
439 * set, if so charge the init_mm (happens for pagecache usage).
445 mem
= rcu_dereference(mm
->mem_cgroup
);
447 * For every charge from the cgroup, increment reference
454 * If we created the page_cgroup, we should free it on exceeding
457 while (res_counter_charge(&mem
->res
, PAGE_SIZE
)) {
458 if (!(gfp_mask
& __GFP_WAIT
))
461 if (try_to_free_mem_cgroup_pages(mem
, gfp_mask
))
465 * try_to_free_mem_cgroup_pages() might not give us a full
466 * picture of reclaim. Some pages are reclaimed and might be
467 * moved to swap cache or just unmapped from the cgroup.
468 * Check the limit again to see if the reclaim reduced the
469 * current usage of the cgroup before giving up
471 if (res_counter_check_under_limit(&mem
->res
))
475 mem_cgroup_out_of_memory(mem
, gfp_mask
);
478 congestion_wait(WRITE
, HZ
/10);
481 atomic_set(&pc
->ref_cnt
, 1);
482 pc
->mem_cgroup
= mem
;
484 pc
->flags
= PAGE_CGROUP_FLAG_ACTIVE
;
485 if (ctype
== MEM_CGROUP_CHARGE_TYPE_CACHE
)
486 pc
->flags
|= PAGE_CGROUP_FLAG_CACHE
;
488 if (!page
|| page_cgroup_assign_new_page_cgroup(page
, pc
)) {
490 * Another charge has been added to this page already.
491 * We take lock_page_cgroup(page) again and read
492 * page->cgroup, increment refcnt.... just retry is OK.
494 res_counter_uncharge(&mem
->res
, PAGE_SIZE
);
502 spin_lock_irqsave(&mem
->lru_lock
, flags
);
503 /* Update statistics vector */
504 mem_cgroup_charge_statistics(mem
, pc
->flags
, true);
505 list_add(&pc
->lru
, &mem
->active_list
);
506 spin_unlock_irqrestore(&mem
->lru_lock
, flags
);
517 int mem_cgroup_charge(struct page
*page
, struct mm_struct
*mm
,
520 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
521 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
525 * See if the cached pages should be charged at all?
527 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
531 struct mem_cgroup
*mem
;
536 mem
= rcu_dereference(mm
->mem_cgroup
);
539 if (mem
->control_type
== MEM_CGROUP_TYPE_ALL
)
540 ret
= mem_cgroup_charge_common(page
, mm
, gfp_mask
,
541 MEM_CGROUP_CHARGE_TYPE_CACHE
);
547 * Uncharging is always a welcome operation, we never complain, simply
550 void mem_cgroup_uncharge(struct page_cgroup
*pc
)
552 struct mem_cgroup
*mem
;
557 * This can handle cases when a page is not charged at all and we
558 * are switching between handling the control_type.
563 if (atomic_dec_and_test(&pc
->ref_cnt
)) {
566 * get page->cgroup and clear it under lock.
567 * force_empty can drop page->cgroup without checking refcnt.
569 if (clear_page_cgroup(page
, pc
) == pc
) {
570 mem
= pc
->mem_cgroup
;
572 res_counter_uncharge(&mem
->res
, PAGE_SIZE
);
573 spin_lock_irqsave(&mem
->lru_lock
, flags
);
574 list_del_init(&pc
->lru
);
575 mem_cgroup_charge_statistics(mem
, pc
->flags
, false);
576 spin_unlock_irqrestore(&mem
->lru_lock
, flags
);
582 * Returns non-zero if a page (under migration) has valid page_cgroup member.
583 * Refcnt of page_cgroup is incremented.
586 int mem_cgroup_prepare_migration(struct page
*page
)
588 struct page_cgroup
*pc
;
590 lock_page_cgroup(page
);
591 pc
= page_get_page_cgroup(page
);
592 if (pc
&& atomic_inc_not_zero(&pc
->ref_cnt
))
594 unlock_page_cgroup(page
);
598 void mem_cgroup_end_migration(struct page
*page
)
600 struct page_cgroup
*pc
= page_get_page_cgroup(page
);
601 mem_cgroup_uncharge(pc
);
604 * We know both *page* and *newpage* are now not-on-LRU and Pg_locked.
605 * And no race with uncharge() routines because page_cgroup for *page*
606 * has extra one reference by mem_cgroup_prepare_migration.
609 void mem_cgroup_page_migration(struct page
*page
, struct page
*newpage
)
611 struct page_cgroup
*pc
;
613 pc
= page_get_page_cgroup(page
);
616 if (clear_page_cgroup(page
, pc
) != pc
)
619 lock_page_cgroup(newpage
);
620 page_assign_page_cgroup(newpage
, pc
);
621 unlock_page_cgroup(newpage
);
626 * This routine traverse page_cgroup in given list and drop them all.
627 * This routine ignores page_cgroup->ref_cnt.
628 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
630 #define FORCE_UNCHARGE_BATCH (128)
632 mem_cgroup_force_empty_list(struct mem_cgroup
*mem
, struct list_head
*list
)
634 struct page_cgroup
*pc
;
640 count
= FORCE_UNCHARGE_BATCH
;
641 spin_lock_irqsave(&mem
->lru_lock
, flags
);
643 while (--count
&& !list_empty(list
)) {
644 pc
= list_entry(list
->prev
, struct page_cgroup
, lru
);
646 /* Avoid race with charge */
647 atomic_set(&pc
->ref_cnt
, 0);
648 if (clear_page_cgroup(page
, pc
) == pc
) {
650 res_counter_uncharge(&mem
->res
, PAGE_SIZE
);
651 list_del_init(&pc
->lru
);
652 mem_cgroup_charge_statistics(mem
, pc
->flags
, false);
654 } else /* being uncharged ? ...do relax */
657 spin_unlock_irqrestore(&mem
->lru_lock
, flags
);
658 if (!list_empty(list
)) {
666 * make mem_cgroup's charge to be 0 if there is no task.
667 * This enables deleting this mem_cgroup.
670 int mem_cgroup_force_empty(struct mem_cgroup
*mem
)
675 * page reclaim code (kswapd etc..) will move pages between
676 ` * active_list <-> inactive_list while we don't take a lock.
677 * So, we have to do loop here until all lists are empty.
679 while (!(list_empty(&mem
->active_list
) &&
680 list_empty(&mem
->inactive_list
))) {
681 if (atomic_read(&mem
->css
.cgroup
->count
) > 0)
683 /* drop all page_cgroup in active_list */
684 mem_cgroup_force_empty_list(mem
, &mem
->active_list
);
685 /* drop all page_cgroup in inactive_list */
686 mem_cgroup_force_empty_list(mem
, &mem
->inactive_list
);
696 int mem_cgroup_write_strategy(char *buf
, unsigned long long *tmp
)
698 *tmp
= memparse(buf
, &buf
);
703 * Round up the value to the closest page size
705 *tmp
= ((*tmp
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
) << PAGE_SHIFT
;
709 static ssize_t
mem_cgroup_read(struct cgroup
*cont
,
710 struct cftype
*cft
, struct file
*file
,
711 char __user
*userbuf
, size_t nbytes
, loff_t
*ppos
)
713 return res_counter_read(&mem_cgroup_from_cont(cont
)->res
,
714 cft
->private, userbuf
, nbytes
, ppos
,
718 static ssize_t
mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
719 struct file
*file
, const char __user
*userbuf
,
720 size_t nbytes
, loff_t
*ppos
)
722 return res_counter_write(&mem_cgroup_from_cont(cont
)->res
,
723 cft
->private, userbuf
, nbytes
, ppos
,
724 mem_cgroup_write_strategy
);
727 static ssize_t
mem_control_type_write(struct cgroup
*cont
,
728 struct cftype
*cft
, struct file
*file
,
729 const char __user
*userbuf
,
730 size_t nbytes
, loff_t
*pos
)
735 struct mem_cgroup
*mem
;
737 mem
= mem_cgroup_from_cont(cont
);
738 buf
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
745 if (copy_from_user(buf
, userbuf
, nbytes
))
749 tmp
= simple_strtoul(buf
, &end
, 10);
753 if (tmp
<= MEM_CGROUP_TYPE_UNSPEC
|| tmp
>= MEM_CGROUP_TYPE_MAX
)
756 mem
->control_type
= tmp
;
764 static ssize_t
mem_control_type_read(struct cgroup
*cont
,
766 struct file
*file
, char __user
*userbuf
,
767 size_t nbytes
, loff_t
*ppos
)
771 struct mem_cgroup
*mem
;
773 mem
= mem_cgroup_from_cont(cont
);
775 val
= mem
->control_type
;
776 s
+= sprintf(s
, "%lu\n", val
);
777 return simple_read_from_buffer((void __user
*)userbuf
, nbytes
,
782 static ssize_t
mem_force_empty_write(struct cgroup
*cont
,
783 struct cftype
*cft
, struct file
*file
,
784 const char __user
*userbuf
,
785 size_t nbytes
, loff_t
*ppos
)
787 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
789 ret
= mem_cgroup_force_empty(mem
);
796 * Note: This should be removed if cgroup supports write-only file.
799 static ssize_t
mem_force_empty_read(struct cgroup
*cont
,
801 struct file
*file
, char __user
*userbuf
,
802 size_t nbytes
, loff_t
*ppos
)
808 static const struct mem_cgroup_stat_desc
{
811 } mem_cgroup_stat_desc
[] = {
812 [MEM_CGROUP_STAT_CACHE
] = { "cache", PAGE_SIZE
, },
813 [MEM_CGROUP_STAT_RSS
] = { "rss", PAGE_SIZE
, },
816 static int mem_control_stat_show(struct seq_file
*m
, void *arg
)
818 struct cgroup
*cont
= m
->private;
819 struct mem_cgroup
*mem_cont
= mem_cgroup_from_cont(cont
);
820 struct mem_cgroup_stat
*stat
= &mem_cont
->stat
;
823 for (i
= 0; i
< ARRAY_SIZE(stat
->cpustat
[0].count
); i
++) {
826 val
= mem_cgroup_read_stat(stat
, i
);
827 val
*= mem_cgroup_stat_desc
[i
].unit
;
828 seq_printf(m
, "%s %lld\n", mem_cgroup_stat_desc
[i
].msg
,
834 static const struct file_operations mem_control_stat_file_operations
= {
837 .release
= single_release
,
840 static int mem_control_stat_open(struct inode
*unused
, struct file
*file
)
843 struct cgroup
*cont
= file
->f_dentry
->d_parent
->d_fsdata
;
845 file
->f_op
= &mem_control_stat_file_operations
;
846 return single_open(file
, mem_control_stat_show
, cont
);
851 static struct cftype mem_cgroup_files
[] = {
853 .name
= "usage_in_bytes",
854 .private = RES_USAGE
,
855 .read
= mem_cgroup_read
,
858 .name
= "limit_in_bytes",
859 .private = RES_LIMIT
,
860 .write
= mem_cgroup_write
,
861 .read
= mem_cgroup_read
,
865 .private = RES_FAILCNT
,
866 .read
= mem_cgroup_read
,
869 .name
= "control_type",
870 .write
= mem_control_type_write
,
871 .read
= mem_control_type_read
,
874 .name
= "force_empty",
875 .write
= mem_force_empty_write
,
876 .read
= mem_force_empty_read
,
880 .open
= mem_control_stat_open
,
884 static struct mem_cgroup init_mem_cgroup
;
886 static struct cgroup_subsys_state
*
887 mem_cgroup_create(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
889 struct mem_cgroup
*mem
;
891 if (unlikely((cont
->parent
) == NULL
)) {
892 mem
= &init_mem_cgroup
;
893 init_mm
.mem_cgroup
= mem
;
895 mem
= kzalloc(sizeof(struct mem_cgroup
), GFP_KERNEL
);
900 res_counter_init(&mem
->res
);
901 INIT_LIST_HEAD(&mem
->active_list
);
902 INIT_LIST_HEAD(&mem
->inactive_list
);
903 spin_lock_init(&mem
->lru_lock
);
904 mem
->control_type
= MEM_CGROUP_TYPE_ALL
;
908 static void mem_cgroup_pre_destroy(struct cgroup_subsys
*ss
,
911 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
912 mem_cgroup_force_empty(mem
);
915 static void mem_cgroup_destroy(struct cgroup_subsys
*ss
,
918 kfree(mem_cgroup_from_cont(cont
));
921 static int mem_cgroup_populate(struct cgroup_subsys
*ss
,
924 return cgroup_add_files(cont
, ss
, mem_cgroup_files
,
925 ARRAY_SIZE(mem_cgroup_files
));
928 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
930 struct cgroup
*old_cont
,
931 struct task_struct
*p
)
933 struct mm_struct
*mm
;
934 struct mem_cgroup
*mem
, *old_mem
;
940 mem
= mem_cgroup_from_cont(cont
);
941 old_mem
= mem_cgroup_from_cont(old_cont
);
947 * Only thread group leaders are allowed to migrate, the mm_struct is
948 * in effect owned by the leader
950 if (p
->tgid
!= p
->pid
)
954 rcu_assign_pointer(mm
->mem_cgroup
, mem
);
955 css_put(&old_mem
->css
);
962 struct cgroup_subsys mem_cgroup_subsys
= {
964 .subsys_id
= mem_cgroup_subsys_id
,
965 .create
= mem_cgroup_create
,
966 .pre_destroy
= mem_cgroup_pre_destroy
,
967 .destroy
= mem_cgroup_destroy
,
968 .populate
= mem_cgroup_populate
,
969 .attach
= mem_cgroup_move_task
,