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>
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
13 * This program is free software; you can redistribute it and/or modify
14 * it under the terms of the GNU General Public License as published by
15 * the Free Software Foundation; either version 2 of the License, or
16 * (at your option) any later version.
18 * This program is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 * GNU General Public License for more details.
24 #include <linux/res_counter.h>
25 #include <linux/memcontrol.h>
26 #include <linux/cgroup.h>
28 #include <linux/hugetlb.h>
29 #include <linux/pagemap.h>
30 #include <linux/smp.h>
31 #include <linux/page-flags.h>
32 #include <linux/backing-dev.h>
33 #include <linux/bit_spinlock.h>
34 #include <linux/rcupdate.h>
35 #include <linux/limits.h>
36 #include <linux/mutex.h>
37 #include <linux/rbtree.h>
38 #include <linux/slab.h>
39 #include <linux/swap.h>
40 #include <linux/swapops.h>
41 #include <linux/spinlock.h>
42 #include <linux/eventfd.h>
43 #include <linux/sort.h>
45 #include <linux/seq_file.h>
46 #include <linux/vmalloc.h>
47 #include <linux/mm_inline.h>
48 #include <linux/page_cgroup.h>
49 #include <linux/cpu.h>
50 #include <linux/oom.h>
53 #include <asm/uaccess.h>
55 #include <trace/events/vmscan.h>
57 struct cgroup_subsys mem_cgroup_subsys __read_mostly
;
58 #define MEM_CGROUP_RECLAIM_RETRIES 5
59 struct mem_cgroup
*root_mem_cgroup __read_mostly
;
61 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
62 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
63 int do_swap_account __read_mostly
;
65 /* for remember boot option*/
66 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED
67 static int really_do_swap_account __initdata
= 1;
69 static int really_do_swap_account __initdata
= 0;
73 #define do_swap_account (0)
78 * Statistics for memory cgroup.
80 enum mem_cgroup_stat_index
{
82 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
84 MEM_CGROUP_STAT_CACHE
, /* # of pages charged as cache */
85 MEM_CGROUP_STAT_RSS
, /* # of pages charged as anon rss */
86 MEM_CGROUP_STAT_FILE_MAPPED
, /* # of pages charged as file rss */
87 MEM_CGROUP_STAT_SWAPOUT
, /* # of pages, swapped out */
88 MEM_CGROUP_STAT_DATA
, /* end of data requires synchronization */
89 MEM_CGROUP_ON_MOVE
, /* someone is moving account between groups */
90 MEM_CGROUP_STAT_NSTATS
,
93 enum mem_cgroup_events_index
{
94 MEM_CGROUP_EVENTS_PGPGIN
, /* # of pages paged in */
95 MEM_CGROUP_EVENTS_PGPGOUT
, /* # of pages paged out */
96 MEM_CGROUP_EVENTS_COUNT
, /* # of pages paged in/out */
97 MEM_CGROUP_EVENTS_NSTATS
,
100 * Per memcg event counter is incremented at every pagein/pageout. With THP,
101 * it will be incremated by the number of pages. This counter is used for
102 * for trigger some periodic events. This is straightforward and better
103 * than using jiffies etc. to handle periodic memcg event.
105 enum mem_cgroup_events_target
{
106 MEM_CGROUP_TARGET_THRESH
,
107 MEM_CGROUP_TARGET_SOFTLIMIT
,
110 #define THRESHOLDS_EVENTS_TARGET (128)
111 #define SOFTLIMIT_EVENTS_TARGET (1024)
113 struct mem_cgroup_stat_cpu
{
114 long count
[MEM_CGROUP_STAT_NSTATS
];
115 unsigned long events
[MEM_CGROUP_EVENTS_NSTATS
];
116 unsigned long targets
[MEM_CGROUP_NTARGETS
];
120 * per-zone information in memory controller.
122 struct mem_cgroup_per_zone
{
124 * spin_lock to protect the per cgroup LRU
126 struct list_head lists
[NR_LRU_LISTS
];
127 unsigned long count
[NR_LRU_LISTS
];
129 struct zone_reclaim_stat reclaim_stat
;
130 struct rb_node tree_node
; /* RB tree node */
131 unsigned long long usage_in_excess
;/* Set to the value by which */
132 /* the soft limit is exceeded*/
134 struct mem_cgroup
*mem
; /* Back pointer, we cannot */
135 /* use container_of */
137 /* Macro for accessing counter */
138 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
140 struct mem_cgroup_per_node
{
141 struct mem_cgroup_per_zone zoneinfo
[MAX_NR_ZONES
];
144 struct mem_cgroup_lru_info
{
145 struct mem_cgroup_per_node
*nodeinfo
[MAX_NUMNODES
];
149 * Cgroups above their limits are maintained in a RB-Tree, independent of
150 * their hierarchy representation
153 struct mem_cgroup_tree_per_zone
{
154 struct rb_root rb_root
;
158 struct mem_cgroup_tree_per_node
{
159 struct mem_cgroup_tree_per_zone rb_tree_per_zone
[MAX_NR_ZONES
];
162 struct mem_cgroup_tree
{
163 struct mem_cgroup_tree_per_node
*rb_tree_per_node
[MAX_NUMNODES
];
166 static struct mem_cgroup_tree soft_limit_tree __read_mostly
;
168 struct mem_cgroup_threshold
{
169 struct eventfd_ctx
*eventfd
;
174 struct mem_cgroup_threshold_ary
{
175 /* An array index points to threshold just below usage. */
176 int current_threshold
;
177 /* Size of entries[] */
179 /* Array of thresholds */
180 struct mem_cgroup_threshold entries
[0];
183 struct mem_cgroup_thresholds
{
184 /* Primary thresholds array */
185 struct mem_cgroup_threshold_ary
*primary
;
187 * Spare threshold array.
188 * This is needed to make mem_cgroup_unregister_event() "never fail".
189 * It must be able to store at least primary->size - 1 entries.
191 struct mem_cgroup_threshold_ary
*spare
;
195 struct mem_cgroup_eventfd_list
{
196 struct list_head list
;
197 struct eventfd_ctx
*eventfd
;
200 static void mem_cgroup_threshold(struct mem_cgroup
*mem
);
201 static void mem_cgroup_oom_notify(struct mem_cgroup
*mem
);
204 * The memory controller data structure. The memory controller controls both
205 * page cache and RSS per cgroup. We would eventually like to provide
206 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
207 * to help the administrator determine what knobs to tune.
209 * TODO: Add a water mark for the memory controller. Reclaim will begin when
210 * we hit the water mark. May be even add a low water mark, such that
211 * no reclaim occurs from a cgroup at it's low water mark, this is
212 * a feature that will be implemented much later in the future.
215 struct cgroup_subsys_state css
;
217 * the counter to account for memory usage
219 struct res_counter res
;
221 * the counter to account for mem+swap usage.
223 struct res_counter memsw
;
225 * Per cgroup active and inactive list, similar to the
226 * per zone LRU lists.
228 struct mem_cgroup_lru_info info
;
230 * While reclaiming in a hierarchy, we cache the last child we
233 int last_scanned_child
;
235 * Should the accounting and control be hierarchical, per subtree?
241 unsigned int swappiness
;
242 /* OOM-Killer disable */
243 int oom_kill_disable
;
245 /* set when res.limit == memsw.limit */
246 bool memsw_is_minimum
;
248 /* protect arrays of thresholds */
249 struct mutex thresholds_lock
;
251 /* thresholds for memory usage. RCU-protected */
252 struct mem_cgroup_thresholds thresholds
;
254 /* thresholds for mem+swap usage. RCU-protected */
255 struct mem_cgroup_thresholds memsw_thresholds
;
257 /* For oom notifier event fd */
258 struct list_head oom_notify
;
261 * Should we move charges of a task when a task is moved into this
262 * mem_cgroup ? And what type of charges should we move ?
264 unsigned long move_charge_at_immigrate
;
268 struct mem_cgroup_stat_cpu
*stat
;
270 * used when a cpu is offlined or other synchronizations
271 * See mem_cgroup_read_stat().
273 struct mem_cgroup_stat_cpu nocpu_base
;
274 spinlock_t pcp_counter_lock
;
277 /* Stuffs for move charges at task migration. */
279 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
280 * left-shifted bitmap of these types.
283 MOVE_CHARGE_TYPE_ANON
, /* private anonymous page and swap of it */
284 MOVE_CHARGE_TYPE_FILE
, /* file page(including tmpfs) and swap of it */
288 /* "mc" and its members are protected by cgroup_mutex */
289 static struct move_charge_struct
{
290 spinlock_t lock
; /* for from, to */
291 struct mem_cgroup
*from
;
292 struct mem_cgroup
*to
;
293 unsigned long precharge
;
294 unsigned long moved_charge
;
295 unsigned long moved_swap
;
296 struct task_struct
*moving_task
; /* a task moving charges */
297 wait_queue_head_t waitq
; /* a waitq for other context */
299 .lock
= __SPIN_LOCK_UNLOCKED(mc
.lock
),
300 .waitq
= __WAIT_QUEUE_HEAD_INITIALIZER(mc
.waitq
),
303 static bool move_anon(void)
305 return test_bit(MOVE_CHARGE_TYPE_ANON
,
306 &mc
.to
->move_charge_at_immigrate
);
309 static bool move_file(void)
311 return test_bit(MOVE_CHARGE_TYPE_FILE
,
312 &mc
.to
->move_charge_at_immigrate
);
316 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
317 * limit reclaim to prevent infinite loops, if they ever occur.
319 #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
320 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
323 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
324 MEM_CGROUP_CHARGE_TYPE_MAPPED
,
325 MEM_CGROUP_CHARGE_TYPE_SHMEM
, /* used by page migration of shmem */
326 MEM_CGROUP_CHARGE_TYPE_FORCE
, /* used by force_empty */
327 MEM_CGROUP_CHARGE_TYPE_SWAPOUT
, /* for accounting swapcache */
328 MEM_CGROUP_CHARGE_TYPE_DROP
, /* a page was unused swap cache */
332 /* for encoding cft->private value on file */
335 #define _OOM_TYPE (2)
336 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
337 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
338 #define MEMFILE_ATTR(val) ((val) & 0xffff)
339 /* Used for OOM nofiier */
340 #define OOM_CONTROL (0)
343 * Reclaim flags for mem_cgroup_hierarchical_reclaim
345 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
346 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
347 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
348 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
349 #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2
350 #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT)
352 static void mem_cgroup_get(struct mem_cgroup
*mem
);
353 static void mem_cgroup_put(struct mem_cgroup
*mem
);
354 static struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*mem
);
355 static void drain_all_stock_async(void);
357 static struct mem_cgroup_per_zone
*
358 mem_cgroup_zoneinfo(struct mem_cgroup
*mem
, int nid
, int zid
)
360 return &mem
->info
.nodeinfo
[nid
]->zoneinfo
[zid
];
363 struct cgroup_subsys_state
*mem_cgroup_css(struct mem_cgroup
*mem
)
368 static struct mem_cgroup_per_zone
*
369 page_cgroup_zoneinfo(struct mem_cgroup
*mem
, struct page
*page
)
371 int nid
= page_to_nid(page
);
372 int zid
= page_zonenum(page
);
374 return mem_cgroup_zoneinfo(mem
, nid
, zid
);
377 static struct mem_cgroup_tree_per_zone
*
378 soft_limit_tree_node_zone(int nid
, int zid
)
380 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
383 static struct mem_cgroup_tree_per_zone
*
384 soft_limit_tree_from_page(struct page
*page
)
386 int nid
= page_to_nid(page
);
387 int zid
= page_zonenum(page
);
389 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
393 __mem_cgroup_insert_exceeded(struct mem_cgroup
*mem
,
394 struct mem_cgroup_per_zone
*mz
,
395 struct mem_cgroup_tree_per_zone
*mctz
,
396 unsigned long long new_usage_in_excess
)
398 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
399 struct rb_node
*parent
= NULL
;
400 struct mem_cgroup_per_zone
*mz_node
;
405 mz
->usage_in_excess
= new_usage_in_excess
;
406 if (!mz
->usage_in_excess
)
410 mz_node
= rb_entry(parent
, struct mem_cgroup_per_zone
,
412 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
)
415 * We can't avoid mem cgroups that are over their soft
416 * limit by the same amount
418 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
421 rb_link_node(&mz
->tree_node
, parent
, p
);
422 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
427 __mem_cgroup_remove_exceeded(struct mem_cgroup
*mem
,
428 struct mem_cgroup_per_zone
*mz
,
429 struct mem_cgroup_tree_per_zone
*mctz
)
433 rb_erase(&mz
->tree_node
, &mctz
->rb_root
);
438 mem_cgroup_remove_exceeded(struct mem_cgroup
*mem
,
439 struct mem_cgroup_per_zone
*mz
,
440 struct mem_cgroup_tree_per_zone
*mctz
)
442 spin_lock(&mctz
->lock
);
443 __mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
444 spin_unlock(&mctz
->lock
);
448 static void mem_cgroup_update_tree(struct mem_cgroup
*mem
, struct page
*page
)
450 unsigned long long excess
;
451 struct mem_cgroup_per_zone
*mz
;
452 struct mem_cgroup_tree_per_zone
*mctz
;
453 int nid
= page_to_nid(page
);
454 int zid
= page_zonenum(page
);
455 mctz
= soft_limit_tree_from_page(page
);
458 * Necessary to update all ancestors when hierarchy is used.
459 * because their event counter is not touched.
461 for (; mem
; mem
= parent_mem_cgroup(mem
)) {
462 mz
= mem_cgroup_zoneinfo(mem
, nid
, zid
);
463 excess
= res_counter_soft_limit_excess(&mem
->res
);
465 * We have to update the tree if mz is on RB-tree or
466 * mem is over its softlimit.
468 if (excess
|| mz
->on_tree
) {
469 spin_lock(&mctz
->lock
);
470 /* if on-tree, remove it */
472 __mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
474 * Insert again. mz->usage_in_excess will be updated.
475 * If excess is 0, no tree ops.
477 __mem_cgroup_insert_exceeded(mem
, mz
, mctz
, excess
);
478 spin_unlock(&mctz
->lock
);
483 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*mem
)
486 struct mem_cgroup_per_zone
*mz
;
487 struct mem_cgroup_tree_per_zone
*mctz
;
489 for_each_node_state(node
, N_POSSIBLE
) {
490 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
491 mz
= mem_cgroup_zoneinfo(mem
, node
, zone
);
492 mctz
= soft_limit_tree_node_zone(node
, zone
);
493 mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
498 static struct mem_cgroup_per_zone
*
499 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
501 struct rb_node
*rightmost
= NULL
;
502 struct mem_cgroup_per_zone
*mz
;
506 rightmost
= rb_last(&mctz
->rb_root
);
508 goto done
; /* Nothing to reclaim from */
510 mz
= rb_entry(rightmost
, struct mem_cgroup_per_zone
, tree_node
);
512 * Remove the node now but someone else can add it back,
513 * we will to add it back at the end of reclaim to its correct
514 * position in the tree.
516 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
517 if (!res_counter_soft_limit_excess(&mz
->mem
->res
) ||
518 !css_tryget(&mz
->mem
->css
))
524 static struct mem_cgroup_per_zone
*
525 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
527 struct mem_cgroup_per_zone
*mz
;
529 spin_lock(&mctz
->lock
);
530 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
531 spin_unlock(&mctz
->lock
);
536 * Implementation Note: reading percpu statistics for memcg.
538 * Both of vmstat[] and percpu_counter has threshold and do periodic
539 * synchronization to implement "quick" read. There are trade-off between
540 * reading cost and precision of value. Then, we may have a chance to implement
541 * a periodic synchronizion of counter in memcg's counter.
543 * But this _read() function is used for user interface now. The user accounts
544 * memory usage by memory cgroup and he _always_ requires exact value because
545 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
546 * have to visit all online cpus and make sum. So, for now, unnecessary
547 * synchronization is not implemented. (just implemented for cpu hotplug)
549 * If there are kernel internal actions which can make use of some not-exact
550 * value, and reading all cpu value can be performance bottleneck in some
551 * common workload, threashold and synchonization as vmstat[] should be
554 static long mem_cgroup_read_stat(struct mem_cgroup
*mem
,
555 enum mem_cgroup_stat_index idx
)
561 for_each_online_cpu(cpu
)
562 val
+= per_cpu(mem
->stat
->count
[idx
], cpu
);
563 #ifdef CONFIG_HOTPLUG_CPU
564 spin_lock(&mem
->pcp_counter_lock
);
565 val
+= mem
->nocpu_base
.count
[idx
];
566 spin_unlock(&mem
->pcp_counter_lock
);
572 static long mem_cgroup_local_usage(struct mem_cgroup
*mem
)
576 ret
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_RSS
);
577 ret
+= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_CACHE
);
581 static void mem_cgroup_swap_statistics(struct mem_cgroup
*mem
,
584 int val
= (charge
) ? 1 : -1;
585 this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_SWAPOUT
], val
);
588 static unsigned long mem_cgroup_read_events(struct mem_cgroup
*mem
,
589 enum mem_cgroup_events_index idx
)
591 unsigned long val
= 0;
594 for_each_online_cpu(cpu
)
595 val
+= per_cpu(mem
->stat
->events
[idx
], cpu
);
596 #ifdef CONFIG_HOTPLUG_CPU
597 spin_lock(&mem
->pcp_counter_lock
);
598 val
+= mem
->nocpu_base
.events
[idx
];
599 spin_unlock(&mem
->pcp_counter_lock
);
604 static void mem_cgroup_charge_statistics(struct mem_cgroup
*mem
,
605 bool file
, int nr_pages
)
610 __this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_CACHE
], nr_pages
);
612 __this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_RSS
], nr_pages
);
614 /* pagein of a big page is an event. So, ignore page size */
616 __this_cpu_inc(mem
->stat
->events
[MEM_CGROUP_EVENTS_PGPGIN
]);
618 __this_cpu_inc(mem
->stat
->events
[MEM_CGROUP_EVENTS_PGPGOUT
]);
619 nr_pages
= -nr_pages
; /* for event */
622 __this_cpu_add(mem
->stat
->events
[MEM_CGROUP_EVENTS_COUNT
], nr_pages
);
627 static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup
*mem
,
631 struct mem_cgroup_per_zone
*mz
;
634 for_each_online_node(nid
)
635 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
636 mz
= mem_cgroup_zoneinfo(mem
, nid
, zid
);
637 total
+= MEM_CGROUP_ZSTAT(mz
, idx
);
642 static bool __memcg_event_check(struct mem_cgroup
*mem
, int target
)
644 unsigned long val
, next
;
646 val
= this_cpu_read(mem
->stat
->events
[MEM_CGROUP_EVENTS_COUNT
]);
647 next
= this_cpu_read(mem
->stat
->targets
[target
]);
648 /* from time_after() in jiffies.h */
649 return ((long)next
- (long)val
< 0);
652 static void __mem_cgroup_target_update(struct mem_cgroup
*mem
, int target
)
654 unsigned long val
, next
;
656 val
= this_cpu_read(mem
->stat
->events
[MEM_CGROUP_EVENTS_COUNT
]);
659 case MEM_CGROUP_TARGET_THRESH
:
660 next
= val
+ THRESHOLDS_EVENTS_TARGET
;
662 case MEM_CGROUP_TARGET_SOFTLIMIT
:
663 next
= val
+ SOFTLIMIT_EVENTS_TARGET
;
669 this_cpu_write(mem
->stat
->targets
[target
], next
);
673 * Check events in order.
676 static void memcg_check_events(struct mem_cgroup
*mem
, struct page
*page
)
678 /* threshold event is triggered in finer grain than soft limit */
679 if (unlikely(__memcg_event_check(mem
, MEM_CGROUP_TARGET_THRESH
))) {
680 mem_cgroup_threshold(mem
);
681 __mem_cgroup_target_update(mem
, MEM_CGROUP_TARGET_THRESH
);
682 if (unlikely(__memcg_event_check(mem
,
683 MEM_CGROUP_TARGET_SOFTLIMIT
))){
684 mem_cgroup_update_tree(mem
, page
);
685 __mem_cgroup_target_update(mem
,
686 MEM_CGROUP_TARGET_SOFTLIMIT
);
691 static struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
693 return container_of(cgroup_subsys_state(cont
,
694 mem_cgroup_subsys_id
), struct mem_cgroup
,
698 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
701 * mm_update_next_owner() may clear mm->owner to NULL
702 * if it races with swapoff, page migration, etc.
703 * So this can be called with p == NULL.
708 return container_of(task_subsys_state(p
, mem_cgroup_subsys_id
),
709 struct mem_cgroup
, css
);
712 static struct mem_cgroup
*try_get_mem_cgroup_from_mm(struct mm_struct
*mm
)
714 struct mem_cgroup
*mem
= NULL
;
719 * Because we have no locks, mm->owner's may be being moved to other
720 * cgroup. We use css_tryget() here even if this looks
721 * pessimistic (rather than adding locks here).
725 mem
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
728 } while (!css_tryget(&mem
->css
));
733 /* The caller has to guarantee "mem" exists before calling this */
734 static struct mem_cgroup
*mem_cgroup_start_loop(struct mem_cgroup
*mem
)
736 struct cgroup_subsys_state
*css
;
739 if (!mem
) /* ROOT cgroup has the smallest ID */
740 return root_mem_cgroup
; /*css_put/get against root is ignored*/
741 if (!mem
->use_hierarchy
) {
742 if (css_tryget(&mem
->css
))
748 * searching a memory cgroup which has the smallest ID under given
749 * ROOT cgroup. (ID >= 1)
751 css
= css_get_next(&mem_cgroup_subsys
, 1, &mem
->css
, &found
);
752 if (css
&& css_tryget(css
))
753 mem
= container_of(css
, struct mem_cgroup
, css
);
760 static struct mem_cgroup
*mem_cgroup_get_next(struct mem_cgroup
*iter
,
761 struct mem_cgroup
*root
,
764 int nextid
= css_id(&iter
->css
) + 1;
767 struct cgroup_subsys_state
*css
;
769 hierarchy_used
= iter
->use_hierarchy
;
772 /* If no ROOT, walk all, ignore hierarchy */
773 if (!cond
|| (root
&& !hierarchy_used
))
777 root
= root_mem_cgroup
;
783 css
= css_get_next(&mem_cgroup_subsys
, nextid
,
785 if (css
&& css_tryget(css
))
786 iter
= container_of(css
, struct mem_cgroup
, css
);
788 /* If css is NULL, no more cgroups will be found */
790 } while (css
&& !iter
);
795 * for_eacn_mem_cgroup_tree() for visiting all cgroup under tree. Please
796 * be careful that "break" loop is not allowed. We have reference count.
797 * Instead of that modify "cond" to be false and "continue" to exit the loop.
799 #define for_each_mem_cgroup_tree_cond(iter, root, cond) \
800 for (iter = mem_cgroup_start_loop(root);\
802 iter = mem_cgroup_get_next(iter, root, cond))
804 #define for_each_mem_cgroup_tree(iter, root) \
805 for_each_mem_cgroup_tree_cond(iter, root, true)
807 #define for_each_mem_cgroup_all(iter) \
808 for_each_mem_cgroup_tree_cond(iter, NULL, true)
811 static inline bool mem_cgroup_is_root(struct mem_cgroup
*mem
)
813 return (mem
== root_mem_cgroup
);
817 * Following LRU functions are allowed to be used without PCG_LOCK.
818 * Operations are called by routine of global LRU independently from memcg.
819 * What we have to take care of here is validness of pc->mem_cgroup.
821 * Changes to pc->mem_cgroup happens when
824 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
825 * It is added to LRU before charge.
826 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
827 * When moving account, the page is not on LRU. It's isolated.
830 void mem_cgroup_del_lru_list(struct page
*page
, enum lru_list lru
)
832 struct page_cgroup
*pc
;
833 struct mem_cgroup_per_zone
*mz
;
835 if (mem_cgroup_disabled())
837 pc
= lookup_page_cgroup(page
);
838 /* can happen while we handle swapcache. */
839 if (!TestClearPageCgroupAcctLRU(pc
))
841 VM_BUG_ON(!pc
->mem_cgroup
);
843 * We don't check PCG_USED bit. It's cleared when the "page" is finally
844 * removed from global LRU.
846 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
847 /* huge page split is done under lru_lock. so, we have no races. */
848 MEM_CGROUP_ZSTAT(mz
, lru
) -= 1 << compound_order(page
);
849 if (mem_cgroup_is_root(pc
->mem_cgroup
))
851 VM_BUG_ON(list_empty(&pc
->lru
));
852 list_del_init(&pc
->lru
);
855 void mem_cgroup_del_lru(struct page
*page
)
857 mem_cgroup_del_lru_list(page
, page_lru(page
));
861 * Writeback is about to end against a page which has been marked for immediate
862 * reclaim. If it still appears to be reclaimable, move it to the tail of the
865 void mem_cgroup_rotate_reclaimable_page(struct page
*page
)
867 struct mem_cgroup_per_zone
*mz
;
868 struct page_cgroup
*pc
;
869 enum lru_list lru
= page_lru(page
);
871 if (mem_cgroup_disabled())
874 pc
= lookup_page_cgroup(page
);
875 /* unused or root page is not rotated. */
876 if (!PageCgroupUsed(pc
))
878 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
880 if (mem_cgroup_is_root(pc
->mem_cgroup
))
882 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
883 list_move_tail(&pc
->lru
, &mz
->lists
[lru
]);
886 void mem_cgroup_rotate_lru_list(struct page
*page
, enum lru_list lru
)
888 struct mem_cgroup_per_zone
*mz
;
889 struct page_cgroup
*pc
;
891 if (mem_cgroup_disabled())
894 pc
= lookup_page_cgroup(page
);
895 /* unused or root page is not rotated. */
896 if (!PageCgroupUsed(pc
))
898 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
900 if (mem_cgroup_is_root(pc
->mem_cgroup
))
902 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
903 list_move(&pc
->lru
, &mz
->lists
[lru
]);
906 void mem_cgroup_add_lru_list(struct page
*page
, enum lru_list lru
)
908 struct page_cgroup
*pc
;
909 struct mem_cgroup_per_zone
*mz
;
911 if (mem_cgroup_disabled())
913 pc
= lookup_page_cgroup(page
);
914 VM_BUG_ON(PageCgroupAcctLRU(pc
));
915 if (!PageCgroupUsed(pc
))
917 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
919 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
920 /* huge page split is done under lru_lock. so, we have no races. */
921 MEM_CGROUP_ZSTAT(mz
, lru
) += 1 << compound_order(page
);
922 SetPageCgroupAcctLRU(pc
);
923 if (mem_cgroup_is_root(pc
->mem_cgroup
))
925 list_add(&pc
->lru
, &mz
->lists
[lru
]);
929 * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to
930 * lru because the page may.be reused after it's fully uncharged (because of
931 * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge
932 * it again. This function is only used to charge SwapCache. It's done under
933 * lock_page and expected that zone->lru_lock is never held.
935 static void mem_cgroup_lru_del_before_commit_swapcache(struct page
*page
)
938 struct zone
*zone
= page_zone(page
);
939 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
941 spin_lock_irqsave(&zone
->lru_lock
, flags
);
943 * Forget old LRU when this page_cgroup is *not* used. This Used bit
944 * is guarded by lock_page() because the page is SwapCache.
946 if (!PageCgroupUsed(pc
))
947 mem_cgroup_del_lru_list(page
, page_lru(page
));
948 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
951 static void mem_cgroup_lru_add_after_commit_swapcache(struct page
*page
)
954 struct zone
*zone
= page_zone(page
);
955 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
957 spin_lock_irqsave(&zone
->lru_lock
, flags
);
958 /* link when the page is linked to LRU but page_cgroup isn't */
959 if (PageLRU(page
) && !PageCgroupAcctLRU(pc
))
960 mem_cgroup_add_lru_list(page
, page_lru(page
));
961 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
965 void mem_cgroup_move_lists(struct page
*page
,
966 enum lru_list from
, enum lru_list to
)
968 if (mem_cgroup_disabled())
970 mem_cgroup_del_lru_list(page
, from
);
971 mem_cgroup_add_lru_list(page
, to
);
974 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*mem
)
977 struct mem_cgroup
*curr
= NULL
;
978 struct task_struct
*p
;
980 p
= find_lock_task_mm(task
);
983 curr
= try_get_mem_cgroup_from_mm(p
->mm
);
988 * We should check use_hierarchy of "mem" not "curr". Because checking
989 * use_hierarchy of "curr" here make this function true if hierarchy is
990 * enabled in "curr" and "curr" is a child of "mem" in *cgroup*
991 * hierarchy(even if use_hierarchy is disabled in "mem").
993 if (mem
->use_hierarchy
)
994 ret
= css_is_ancestor(&curr
->css
, &mem
->css
);
1001 static int calc_inactive_ratio(struct mem_cgroup
*memcg
, unsigned long *present_pages
)
1003 unsigned long active
;
1004 unsigned long inactive
;
1006 unsigned long inactive_ratio
;
1008 inactive
= mem_cgroup_get_local_zonestat(memcg
, LRU_INACTIVE_ANON
);
1009 active
= mem_cgroup_get_local_zonestat(memcg
, LRU_ACTIVE_ANON
);
1011 gb
= (inactive
+ active
) >> (30 - PAGE_SHIFT
);
1013 inactive_ratio
= int_sqrt(10 * gb
);
1017 if (present_pages
) {
1018 present_pages
[0] = inactive
;
1019 present_pages
[1] = active
;
1022 return inactive_ratio
;
1025 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup
*memcg
)
1027 unsigned long active
;
1028 unsigned long inactive
;
1029 unsigned long present_pages
[2];
1030 unsigned long inactive_ratio
;
1032 inactive_ratio
= calc_inactive_ratio(memcg
, present_pages
);
1034 inactive
= present_pages
[0];
1035 active
= present_pages
[1];
1037 if (inactive
* inactive_ratio
< active
)
1043 int mem_cgroup_inactive_file_is_low(struct mem_cgroup
*memcg
)
1045 unsigned long active
;
1046 unsigned long inactive
;
1048 inactive
= mem_cgroup_get_local_zonestat(memcg
, LRU_INACTIVE_FILE
);
1049 active
= mem_cgroup_get_local_zonestat(memcg
, LRU_ACTIVE_FILE
);
1051 return (active
> inactive
);
1054 unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup
*memcg
,
1058 int nid
= zone_to_nid(zone
);
1059 int zid
= zone_idx(zone
);
1060 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
1062 return MEM_CGROUP_ZSTAT(mz
, lru
);
1065 struct zone_reclaim_stat
*mem_cgroup_get_reclaim_stat(struct mem_cgroup
*memcg
,
1068 int nid
= zone_to_nid(zone
);
1069 int zid
= zone_idx(zone
);
1070 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
1072 return &mz
->reclaim_stat
;
1075 struct zone_reclaim_stat
*
1076 mem_cgroup_get_reclaim_stat_from_page(struct page
*page
)
1078 struct page_cgroup
*pc
;
1079 struct mem_cgroup_per_zone
*mz
;
1081 if (mem_cgroup_disabled())
1084 pc
= lookup_page_cgroup(page
);
1085 if (!PageCgroupUsed(pc
))
1087 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1089 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
1090 return &mz
->reclaim_stat
;
1093 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan
,
1094 struct list_head
*dst
,
1095 unsigned long *scanned
, int order
,
1096 int mode
, struct zone
*z
,
1097 struct mem_cgroup
*mem_cont
,
1098 int active
, int file
)
1100 unsigned long nr_taken
= 0;
1104 struct list_head
*src
;
1105 struct page_cgroup
*pc
, *tmp
;
1106 int nid
= zone_to_nid(z
);
1107 int zid
= zone_idx(z
);
1108 struct mem_cgroup_per_zone
*mz
;
1109 int lru
= LRU_FILE
* file
+ active
;
1113 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
1114 src
= &mz
->lists
[lru
];
1117 list_for_each_entry_safe_reverse(pc
, tmp
, src
, lru
) {
1118 if (scan
>= nr_to_scan
)
1121 if (unlikely(!PageCgroupUsed(pc
)))
1124 page
= lookup_cgroup_page(pc
);
1126 if (unlikely(!PageLRU(page
)))
1130 ret
= __isolate_lru_page(page
, mode
, file
);
1133 list_move(&page
->lru
, dst
);
1134 mem_cgroup_del_lru(page
);
1135 nr_taken
+= hpage_nr_pages(page
);
1138 /* we don't affect global LRU but rotate in our LRU */
1139 mem_cgroup_rotate_lru_list(page
, page_lru(page
));
1148 trace_mm_vmscan_memcg_isolate(0, nr_to_scan
, scan
, nr_taken
,
1154 #define mem_cgroup_from_res_counter(counter, member) \
1155 container_of(counter, struct mem_cgroup, member)
1158 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1159 * @mem: the memory cgroup
1161 * Returns the maximum amount of memory @mem can be charged with, in
1164 static unsigned long mem_cgroup_margin(struct mem_cgroup
*mem
)
1166 unsigned long long margin
;
1168 margin
= res_counter_margin(&mem
->res
);
1169 if (do_swap_account
)
1170 margin
= min(margin
, res_counter_margin(&mem
->memsw
));
1171 return margin
>> PAGE_SHIFT
;
1174 static unsigned int get_swappiness(struct mem_cgroup
*memcg
)
1176 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
1179 if (cgrp
->parent
== NULL
)
1180 return vm_swappiness
;
1182 return memcg
->swappiness
;
1185 static void mem_cgroup_start_move(struct mem_cgroup
*mem
)
1190 spin_lock(&mem
->pcp_counter_lock
);
1191 for_each_online_cpu(cpu
)
1192 per_cpu(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) += 1;
1193 mem
->nocpu_base
.count
[MEM_CGROUP_ON_MOVE
] += 1;
1194 spin_unlock(&mem
->pcp_counter_lock
);
1200 static void mem_cgroup_end_move(struct mem_cgroup
*mem
)
1207 spin_lock(&mem
->pcp_counter_lock
);
1208 for_each_online_cpu(cpu
)
1209 per_cpu(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) -= 1;
1210 mem
->nocpu_base
.count
[MEM_CGROUP_ON_MOVE
] -= 1;
1211 spin_unlock(&mem
->pcp_counter_lock
);
1215 * 2 routines for checking "mem" is under move_account() or not.
1217 * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used
1218 * for avoiding race in accounting. If true,
1219 * pc->mem_cgroup may be overwritten.
1221 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1222 * under hierarchy of moving cgroups. This is for
1223 * waiting at hith-memory prressure caused by "move".
1226 static bool mem_cgroup_stealed(struct mem_cgroup
*mem
)
1228 VM_BUG_ON(!rcu_read_lock_held());
1229 return this_cpu_read(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
]) > 0;
1232 static bool mem_cgroup_under_move(struct mem_cgroup
*mem
)
1234 struct mem_cgroup
*from
;
1235 struct mem_cgroup
*to
;
1238 * Unlike task_move routines, we access mc.to, mc.from not under
1239 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1241 spin_lock(&mc
.lock
);
1246 if (from
== mem
|| to
== mem
1247 || (mem
->use_hierarchy
&& css_is_ancestor(&from
->css
, &mem
->css
))
1248 || (mem
->use_hierarchy
&& css_is_ancestor(&to
->css
, &mem
->css
)))
1251 spin_unlock(&mc
.lock
);
1255 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*mem
)
1257 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1258 if (mem_cgroup_under_move(mem
)) {
1260 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1261 /* moving charge context might have finished. */
1264 finish_wait(&mc
.waitq
, &wait
);
1272 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1273 * @memcg: The memory cgroup that went over limit
1274 * @p: Task that is going to be killed
1276 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1279 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1281 struct cgroup
*task_cgrp
;
1282 struct cgroup
*mem_cgrp
;
1284 * Need a buffer in BSS, can't rely on allocations. The code relies
1285 * on the assumption that OOM is serialized for memory controller.
1286 * If this assumption is broken, revisit this code.
1288 static char memcg_name
[PATH_MAX
];
1297 mem_cgrp
= memcg
->css
.cgroup
;
1298 task_cgrp
= task_cgroup(p
, mem_cgroup_subsys_id
);
1300 ret
= cgroup_path(task_cgrp
, memcg_name
, PATH_MAX
);
1303 * Unfortunately, we are unable to convert to a useful name
1304 * But we'll still print out the usage information
1311 printk(KERN_INFO
"Task in %s killed", memcg_name
);
1314 ret
= cgroup_path(mem_cgrp
, memcg_name
, PATH_MAX
);
1322 * Continues from above, so we don't need an KERN_ level
1324 printk(KERN_CONT
" as a result of limit of %s\n", memcg_name
);
1327 printk(KERN_INFO
"memory: usage %llukB, limit %llukB, failcnt %llu\n",
1328 res_counter_read_u64(&memcg
->res
, RES_USAGE
) >> 10,
1329 res_counter_read_u64(&memcg
->res
, RES_LIMIT
) >> 10,
1330 res_counter_read_u64(&memcg
->res
, RES_FAILCNT
));
1331 printk(KERN_INFO
"memory+swap: usage %llukB, limit %llukB, "
1333 res_counter_read_u64(&memcg
->memsw
, RES_USAGE
) >> 10,
1334 res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
) >> 10,
1335 res_counter_read_u64(&memcg
->memsw
, RES_FAILCNT
));
1339 * This function returns the number of memcg under hierarchy tree. Returns
1340 * 1(self count) if no children.
1342 static int mem_cgroup_count_children(struct mem_cgroup
*mem
)
1345 struct mem_cgroup
*iter
;
1347 for_each_mem_cgroup_tree(iter
, mem
)
1353 * Return the memory (and swap, if configured) limit for a memcg.
1355 u64
mem_cgroup_get_limit(struct mem_cgroup
*memcg
)
1360 limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
1361 limit
+= total_swap_pages
<< PAGE_SHIFT
;
1363 memsw
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
1365 * If memsw is finite and limits the amount of swap space available
1366 * to this memcg, return that limit.
1368 return min(limit
, memsw
);
1372 * Visit the first child (need not be the first child as per the ordering
1373 * of the cgroup list, since we track last_scanned_child) of @mem and use
1374 * that to reclaim free pages from.
1376 static struct mem_cgroup
*
1377 mem_cgroup_select_victim(struct mem_cgroup
*root_mem
)
1379 struct mem_cgroup
*ret
= NULL
;
1380 struct cgroup_subsys_state
*css
;
1383 if (!root_mem
->use_hierarchy
) {
1384 css_get(&root_mem
->css
);
1390 nextid
= root_mem
->last_scanned_child
+ 1;
1391 css
= css_get_next(&mem_cgroup_subsys
, nextid
, &root_mem
->css
,
1393 if (css
&& css_tryget(css
))
1394 ret
= container_of(css
, struct mem_cgroup
, css
);
1397 /* Updates scanning parameter */
1399 /* this means start scan from ID:1 */
1400 root_mem
->last_scanned_child
= 0;
1402 root_mem
->last_scanned_child
= found
;
1409 * Scan the hierarchy if needed to reclaim memory. We remember the last child
1410 * we reclaimed from, so that we don't end up penalizing one child extensively
1411 * based on its position in the children list.
1413 * root_mem is the original ancestor that we've been reclaim from.
1415 * We give up and return to the caller when we visit root_mem twice.
1416 * (other groups can be removed while we're walking....)
1418 * If shrink==true, for avoiding to free too much, this returns immedieately.
1420 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup
*root_mem
,
1423 unsigned long reclaim_options
)
1425 struct mem_cgroup
*victim
;
1428 bool noswap
= reclaim_options
& MEM_CGROUP_RECLAIM_NOSWAP
;
1429 bool shrink
= reclaim_options
& MEM_CGROUP_RECLAIM_SHRINK
;
1430 bool check_soft
= reclaim_options
& MEM_CGROUP_RECLAIM_SOFT
;
1431 unsigned long excess
;
1433 excess
= res_counter_soft_limit_excess(&root_mem
->res
) >> PAGE_SHIFT
;
1435 /* If memsw_is_minimum==1, swap-out is of-no-use. */
1436 if (root_mem
->memsw_is_minimum
)
1440 victim
= mem_cgroup_select_victim(root_mem
);
1441 if (victim
== root_mem
) {
1444 drain_all_stock_async();
1447 * If we have not been able to reclaim
1448 * anything, it might because there are
1449 * no reclaimable pages under this hierarchy
1451 if (!check_soft
|| !total
) {
1452 css_put(&victim
->css
);
1456 * We want to do more targetted reclaim.
1457 * excess >> 2 is not to excessive so as to
1458 * reclaim too much, nor too less that we keep
1459 * coming back to reclaim from this cgroup
1461 if (total
>= (excess
>> 2) ||
1462 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
)) {
1463 css_put(&victim
->css
);
1468 if (!mem_cgroup_local_usage(victim
)) {
1469 /* this cgroup's local usage == 0 */
1470 css_put(&victim
->css
);
1473 /* we use swappiness of local cgroup */
1475 ret
= mem_cgroup_shrink_node_zone(victim
, gfp_mask
,
1476 noswap
, get_swappiness(victim
), zone
);
1478 ret
= try_to_free_mem_cgroup_pages(victim
, gfp_mask
,
1479 noswap
, get_swappiness(victim
));
1480 css_put(&victim
->css
);
1482 * At shrinking usage, we can't check we should stop here or
1483 * reclaim more. It's depends on callers. last_scanned_child
1484 * will work enough for keeping fairness under tree.
1490 if (!res_counter_soft_limit_excess(&root_mem
->res
))
1492 } else if (mem_cgroup_margin(root_mem
))
1499 * Check OOM-Killer is already running under our hierarchy.
1500 * If someone is running, return false.
1502 static bool mem_cgroup_oom_lock(struct mem_cgroup
*mem
)
1504 int x
, lock_count
= 0;
1505 struct mem_cgroup
*iter
;
1507 for_each_mem_cgroup_tree(iter
, mem
) {
1508 x
= atomic_inc_return(&iter
->oom_lock
);
1509 lock_count
= max(x
, lock_count
);
1512 if (lock_count
== 1)
1517 static int mem_cgroup_oom_unlock(struct mem_cgroup
*mem
)
1519 struct mem_cgroup
*iter
;
1522 * When a new child is created while the hierarchy is under oom,
1523 * mem_cgroup_oom_lock() may not be called. We have to use
1524 * atomic_add_unless() here.
1526 for_each_mem_cgroup_tree(iter
, mem
)
1527 atomic_add_unless(&iter
->oom_lock
, -1, 0);
1532 static DEFINE_MUTEX(memcg_oom_mutex
);
1533 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1535 struct oom_wait_info
{
1536 struct mem_cgroup
*mem
;
1540 static int memcg_oom_wake_function(wait_queue_t
*wait
,
1541 unsigned mode
, int sync
, void *arg
)
1543 struct mem_cgroup
*wake_mem
= (struct mem_cgroup
*)arg
;
1544 struct oom_wait_info
*oom_wait_info
;
1546 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1548 if (oom_wait_info
->mem
== wake_mem
)
1550 /* if no hierarchy, no match */
1551 if (!oom_wait_info
->mem
->use_hierarchy
|| !wake_mem
->use_hierarchy
)
1554 * Both of oom_wait_info->mem and wake_mem are stable under us.
1555 * Then we can use css_is_ancestor without taking care of RCU.
1557 if (!css_is_ancestor(&oom_wait_info
->mem
->css
, &wake_mem
->css
) &&
1558 !css_is_ancestor(&wake_mem
->css
, &oom_wait_info
->mem
->css
))
1562 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1565 static void memcg_wakeup_oom(struct mem_cgroup
*mem
)
1567 /* for filtering, pass "mem" as argument. */
1568 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, mem
);
1571 static void memcg_oom_recover(struct mem_cgroup
*mem
)
1573 if (mem
&& atomic_read(&mem
->oom_lock
))
1574 memcg_wakeup_oom(mem
);
1578 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1580 bool mem_cgroup_handle_oom(struct mem_cgroup
*mem
, gfp_t mask
)
1582 struct oom_wait_info owait
;
1583 bool locked
, need_to_kill
;
1586 owait
.wait
.flags
= 0;
1587 owait
.wait
.func
= memcg_oom_wake_function
;
1588 owait
.wait
.private = current
;
1589 INIT_LIST_HEAD(&owait
.wait
.task_list
);
1590 need_to_kill
= true;
1591 /* At first, try to OOM lock hierarchy under mem.*/
1592 mutex_lock(&memcg_oom_mutex
);
1593 locked
= mem_cgroup_oom_lock(mem
);
1595 * Even if signal_pending(), we can't quit charge() loop without
1596 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1597 * under OOM is always welcomed, use TASK_KILLABLE here.
1599 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1600 if (!locked
|| mem
->oom_kill_disable
)
1601 need_to_kill
= false;
1603 mem_cgroup_oom_notify(mem
);
1604 mutex_unlock(&memcg_oom_mutex
);
1607 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1608 mem_cgroup_out_of_memory(mem
, mask
);
1611 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1613 mutex_lock(&memcg_oom_mutex
);
1614 mem_cgroup_oom_unlock(mem
);
1615 memcg_wakeup_oom(mem
);
1616 mutex_unlock(&memcg_oom_mutex
);
1618 if (test_thread_flag(TIF_MEMDIE
) || fatal_signal_pending(current
))
1620 /* Give chance to dying process */
1621 schedule_timeout(1);
1626 * Currently used to update mapped file statistics, but the routine can be
1627 * generalized to update other statistics as well.
1629 * Notes: Race condition
1631 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1632 * it tends to be costly. But considering some conditions, we doesn't need
1633 * to do so _always_.
1635 * Considering "charge", lock_page_cgroup() is not required because all
1636 * file-stat operations happen after a page is attached to radix-tree. There
1637 * are no race with "charge".
1639 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1640 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1641 * if there are race with "uncharge". Statistics itself is properly handled
1644 * Considering "move", this is an only case we see a race. To make the race
1645 * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are
1646 * possibility of race condition. If there is, we take a lock.
1649 void mem_cgroup_update_page_stat(struct page
*page
,
1650 enum mem_cgroup_page_stat_item idx
, int val
)
1652 struct mem_cgroup
*mem
;
1653 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1654 bool need_unlock
= false;
1655 unsigned long uninitialized_var(flags
);
1661 mem
= pc
->mem_cgroup
;
1662 if (unlikely(!mem
|| !PageCgroupUsed(pc
)))
1664 /* pc->mem_cgroup is unstable ? */
1665 if (unlikely(mem_cgroup_stealed(mem
)) || PageTransHuge(page
)) {
1666 /* take a lock against to access pc->mem_cgroup */
1667 move_lock_page_cgroup(pc
, &flags
);
1669 mem
= pc
->mem_cgroup
;
1670 if (!mem
|| !PageCgroupUsed(pc
))
1675 case MEMCG_NR_FILE_MAPPED
:
1677 SetPageCgroupFileMapped(pc
);
1678 else if (!page_mapped(page
))
1679 ClearPageCgroupFileMapped(pc
);
1680 idx
= MEM_CGROUP_STAT_FILE_MAPPED
;
1686 this_cpu_add(mem
->stat
->count
[idx
], val
);
1689 if (unlikely(need_unlock
))
1690 move_unlock_page_cgroup(pc
, &flags
);
1694 EXPORT_SYMBOL(mem_cgroup_update_page_stat
);
1697 * size of first charge trial. "32" comes from vmscan.c's magic value.
1698 * TODO: maybe necessary to use big numbers in big irons.
1700 #define CHARGE_BATCH 32U
1701 struct memcg_stock_pcp
{
1702 struct mem_cgroup
*cached
; /* this never be root cgroup */
1703 unsigned int nr_pages
;
1704 struct work_struct work
;
1706 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
1707 static atomic_t memcg_drain_count
;
1710 * Try to consume stocked charge on this cpu. If success, one page is consumed
1711 * from local stock and true is returned. If the stock is 0 or charges from a
1712 * cgroup which is not current target, returns false. This stock will be
1715 static bool consume_stock(struct mem_cgroup
*mem
)
1717 struct memcg_stock_pcp
*stock
;
1720 stock
= &get_cpu_var(memcg_stock
);
1721 if (mem
== stock
->cached
&& stock
->nr_pages
)
1723 else /* need to call res_counter_charge */
1725 put_cpu_var(memcg_stock
);
1730 * Returns stocks cached in percpu to res_counter and reset cached information.
1732 static void drain_stock(struct memcg_stock_pcp
*stock
)
1734 struct mem_cgroup
*old
= stock
->cached
;
1736 if (stock
->nr_pages
) {
1737 unsigned long bytes
= stock
->nr_pages
* PAGE_SIZE
;
1739 res_counter_uncharge(&old
->res
, bytes
);
1740 if (do_swap_account
)
1741 res_counter_uncharge(&old
->memsw
, bytes
);
1742 stock
->nr_pages
= 0;
1744 stock
->cached
= NULL
;
1748 * This must be called under preempt disabled or must be called by
1749 * a thread which is pinned to local cpu.
1751 static void drain_local_stock(struct work_struct
*dummy
)
1753 struct memcg_stock_pcp
*stock
= &__get_cpu_var(memcg_stock
);
1758 * Cache charges(val) which is from res_counter, to local per_cpu area.
1759 * This will be consumed by consume_stock() function, later.
1761 static void refill_stock(struct mem_cgroup
*mem
, unsigned int nr_pages
)
1763 struct memcg_stock_pcp
*stock
= &get_cpu_var(memcg_stock
);
1765 if (stock
->cached
!= mem
) { /* reset if necessary */
1767 stock
->cached
= mem
;
1769 stock
->nr_pages
+= nr_pages
;
1770 put_cpu_var(memcg_stock
);
1774 * Tries to drain stocked charges in other cpus. This function is asynchronous
1775 * and just put a work per cpu for draining localy on each cpu. Caller can
1776 * expects some charges will be back to res_counter later but cannot wait for
1779 static void drain_all_stock_async(void)
1782 /* This function is for scheduling "drain" in asynchronous way.
1783 * The result of "drain" is not directly handled by callers. Then,
1784 * if someone is calling drain, we don't have to call drain more.
1785 * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if
1786 * there is a race. We just do loose check here.
1788 if (atomic_read(&memcg_drain_count
))
1790 /* Notify other cpus that system-wide "drain" is running */
1791 atomic_inc(&memcg_drain_count
);
1793 for_each_online_cpu(cpu
) {
1794 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
1795 schedule_work_on(cpu
, &stock
->work
);
1798 atomic_dec(&memcg_drain_count
);
1799 /* We don't wait for flush_work */
1802 /* This is a synchronous drain interface. */
1803 static void drain_all_stock_sync(void)
1805 /* called when force_empty is called */
1806 atomic_inc(&memcg_drain_count
);
1807 schedule_on_each_cpu(drain_local_stock
);
1808 atomic_dec(&memcg_drain_count
);
1812 * This function drains percpu counter value from DEAD cpu and
1813 * move it to local cpu. Note that this function can be preempted.
1815 static void mem_cgroup_drain_pcp_counter(struct mem_cgroup
*mem
, int cpu
)
1819 spin_lock(&mem
->pcp_counter_lock
);
1820 for (i
= 0; i
< MEM_CGROUP_STAT_DATA
; i
++) {
1821 long x
= per_cpu(mem
->stat
->count
[i
], cpu
);
1823 per_cpu(mem
->stat
->count
[i
], cpu
) = 0;
1824 mem
->nocpu_base
.count
[i
] += x
;
1826 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
1827 unsigned long x
= per_cpu(mem
->stat
->events
[i
], cpu
);
1829 per_cpu(mem
->stat
->events
[i
], cpu
) = 0;
1830 mem
->nocpu_base
.events
[i
] += x
;
1832 /* need to clear ON_MOVE value, works as a kind of lock. */
1833 per_cpu(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) = 0;
1834 spin_unlock(&mem
->pcp_counter_lock
);
1837 static void synchronize_mem_cgroup_on_move(struct mem_cgroup
*mem
, int cpu
)
1839 int idx
= MEM_CGROUP_ON_MOVE
;
1841 spin_lock(&mem
->pcp_counter_lock
);
1842 per_cpu(mem
->stat
->count
[idx
], cpu
) = mem
->nocpu_base
.count
[idx
];
1843 spin_unlock(&mem
->pcp_counter_lock
);
1846 static int __cpuinit
memcg_cpu_hotplug_callback(struct notifier_block
*nb
,
1847 unsigned long action
,
1850 int cpu
= (unsigned long)hcpu
;
1851 struct memcg_stock_pcp
*stock
;
1852 struct mem_cgroup
*iter
;
1854 if ((action
== CPU_ONLINE
)) {
1855 for_each_mem_cgroup_all(iter
)
1856 synchronize_mem_cgroup_on_move(iter
, cpu
);
1860 if ((action
!= CPU_DEAD
) || action
!= CPU_DEAD_FROZEN
)
1863 for_each_mem_cgroup_all(iter
)
1864 mem_cgroup_drain_pcp_counter(iter
, cpu
);
1866 stock
= &per_cpu(memcg_stock
, cpu
);
1872 /* See __mem_cgroup_try_charge() for details */
1874 CHARGE_OK
, /* success */
1875 CHARGE_RETRY
, /* need to retry but retry is not bad */
1876 CHARGE_NOMEM
, /* we can't do more. return -ENOMEM */
1877 CHARGE_WOULDBLOCK
, /* GFP_WAIT wasn't set and no enough res. */
1878 CHARGE_OOM_DIE
, /* the current is killed because of OOM */
1881 static int mem_cgroup_do_charge(struct mem_cgroup
*mem
, gfp_t gfp_mask
,
1882 unsigned int nr_pages
, bool oom_check
)
1884 unsigned long csize
= nr_pages
* PAGE_SIZE
;
1885 struct mem_cgroup
*mem_over_limit
;
1886 struct res_counter
*fail_res
;
1887 unsigned long flags
= 0;
1890 ret
= res_counter_charge(&mem
->res
, csize
, &fail_res
);
1893 if (!do_swap_account
)
1895 ret
= res_counter_charge(&mem
->memsw
, csize
, &fail_res
);
1899 res_counter_uncharge(&mem
->res
, csize
);
1900 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, memsw
);
1901 flags
|= MEM_CGROUP_RECLAIM_NOSWAP
;
1903 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, res
);
1905 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
1906 * of regular pages (CHARGE_BATCH), or a single regular page (1).
1908 * Never reclaim on behalf of optional batching, retry with a
1909 * single page instead.
1911 if (nr_pages
== CHARGE_BATCH
)
1912 return CHARGE_RETRY
;
1914 if (!(gfp_mask
& __GFP_WAIT
))
1915 return CHARGE_WOULDBLOCK
;
1917 ret
= mem_cgroup_hierarchical_reclaim(mem_over_limit
, NULL
,
1919 if (mem_cgroup_margin(mem_over_limit
) >= nr_pages
)
1920 return CHARGE_RETRY
;
1922 * Even though the limit is exceeded at this point, reclaim
1923 * may have been able to free some pages. Retry the charge
1924 * before killing the task.
1926 * Only for regular pages, though: huge pages are rather
1927 * unlikely to succeed so close to the limit, and we fall back
1928 * to regular pages anyway in case of failure.
1930 if (nr_pages
== 1 && ret
)
1931 return CHARGE_RETRY
;
1934 * At task move, charge accounts can be doubly counted. So, it's
1935 * better to wait until the end of task_move if something is going on.
1937 if (mem_cgroup_wait_acct_move(mem_over_limit
))
1938 return CHARGE_RETRY
;
1940 /* If we don't need to call oom-killer at el, return immediately */
1942 return CHARGE_NOMEM
;
1944 if (!mem_cgroup_handle_oom(mem_over_limit
, gfp_mask
))
1945 return CHARGE_OOM_DIE
;
1947 return CHARGE_RETRY
;
1951 * Unlike exported interface, "oom" parameter is added. if oom==true,
1952 * oom-killer can be invoked.
1954 static int __mem_cgroup_try_charge(struct mm_struct
*mm
,
1956 unsigned int nr_pages
,
1957 struct mem_cgroup
**memcg
,
1960 unsigned int batch
= max(CHARGE_BATCH
, nr_pages
);
1961 int nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
1962 struct mem_cgroup
*mem
= NULL
;
1966 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
1967 * in system level. So, allow to go ahead dying process in addition to
1970 if (unlikely(test_thread_flag(TIF_MEMDIE
)
1971 || fatal_signal_pending(current
)))
1975 * We always charge the cgroup the mm_struct belongs to.
1976 * The mm_struct's mem_cgroup changes on task migration if the
1977 * thread group leader migrates. It's possible that mm is not
1978 * set, if so charge the init_mm (happens for pagecache usage).
1983 if (*memcg
) { /* css should be a valid one */
1985 VM_BUG_ON(css_is_removed(&mem
->css
));
1986 if (mem_cgroup_is_root(mem
))
1988 if (nr_pages
== 1 && consume_stock(mem
))
1992 struct task_struct
*p
;
1995 p
= rcu_dereference(mm
->owner
);
1997 * Because we don't have task_lock(), "p" can exit.
1998 * In that case, "mem" can point to root or p can be NULL with
1999 * race with swapoff. Then, we have small risk of mis-accouning.
2000 * But such kind of mis-account by race always happens because
2001 * we don't have cgroup_mutex(). It's overkill and we allo that
2003 * (*) swapoff at el will charge against mm-struct not against
2004 * task-struct. So, mm->owner can be NULL.
2006 mem
= mem_cgroup_from_task(p
);
2007 if (!mem
|| mem_cgroup_is_root(mem
)) {
2011 if (nr_pages
== 1 && consume_stock(mem
)) {
2013 * It seems dagerous to access memcg without css_get().
2014 * But considering how consume_stok works, it's not
2015 * necessary. If consume_stock success, some charges
2016 * from this memcg are cached on this cpu. So, we
2017 * don't need to call css_get()/css_tryget() before
2018 * calling consume_stock().
2023 /* after here, we may be blocked. we need to get refcnt */
2024 if (!css_tryget(&mem
->css
)) {
2034 /* If killed, bypass charge */
2035 if (fatal_signal_pending(current
)) {
2041 if (oom
&& !nr_oom_retries
) {
2043 nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2046 ret
= mem_cgroup_do_charge(mem
, gfp_mask
, batch
, oom_check
);
2050 case CHARGE_RETRY
: /* not in OOM situation but retry */
2055 case CHARGE_WOULDBLOCK
: /* !__GFP_WAIT */
2058 case CHARGE_NOMEM
: /* OOM routine works */
2063 /* If oom, we never return -ENOMEM */
2066 case CHARGE_OOM_DIE
: /* Killed by OOM Killer */
2070 } while (ret
!= CHARGE_OK
);
2072 if (batch
> nr_pages
)
2073 refill_stock(mem
, batch
- nr_pages
);
2087 * Somemtimes we have to undo a charge we got by try_charge().
2088 * This function is for that and do uncharge, put css's refcnt.
2089 * gotten by try_charge().
2091 static void __mem_cgroup_cancel_charge(struct mem_cgroup
*mem
,
2092 unsigned int nr_pages
)
2094 if (!mem_cgroup_is_root(mem
)) {
2095 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2097 res_counter_uncharge(&mem
->res
, bytes
);
2098 if (do_swap_account
)
2099 res_counter_uncharge(&mem
->memsw
, bytes
);
2104 * A helper function to get mem_cgroup from ID. must be called under
2105 * rcu_read_lock(). The caller must check css_is_removed() or some if
2106 * it's concern. (dropping refcnt from swap can be called against removed
2109 static struct mem_cgroup
*mem_cgroup_lookup(unsigned short id
)
2111 struct cgroup_subsys_state
*css
;
2113 /* ID 0 is unused ID */
2116 css
= css_lookup(&mem_cgroup_subsys
, id
);
2119 return container_of(css
, struct mem_cgroup
, css
);
2122 struct mem_cgroup
*try_get_mem_cgroup_from_page(struct page
*page
)
2124 struct mem_cgroup
*mem
= NULL
;
2125 struct page_cgroup
*pc
;
2129 VM_BUG_ON(!PageLocked(page
));
2131 pc
= lookup_page_cgroup(page
);
2132 lock_page_cgroup(pc
);
2133 if (PageCgroupUsed(pc
)) {
2134 mem
= pc
->mem_cgroup
;
2135 if (mem
&& !css_tryget(&mem
->css
))
2137 } else if (PageSwapCache(page
)) {
2138 ent
.val
= page_private(page
);
2139 id
= lookup_swap_cgroup(ent
);
2141 mem
= mem_cgroup_lookup(id
);
2142 if (mem
&& !css_tryget(&mem
->css
))
2146 unlock_page_cgroup(pc
);
2150 static void __mem_cgroup_commit_charge(struct mem_cgroup
*mem
,
2152 unsigned int nr_pages
,
2153 struct page_cgroup
*pc
,
2154 enum charge_type ctype
)
2156 lock_page_cgroup(pc
);
2157 if (unlikely(PageCgroupUsed(pc
))) {
2158 unlock_page_cgroup(pc
);
2159 __mem_cgroup_cancel_charge(mem
, nr_pages
);
2163 * we don't need page_cgroup_lock about tail pages, becase they are not
2164 * accessed by any other context at this point.
2166 pc
->mem_cgroup
= mem
;
2168 * We access a page_cgroup asynchronously without lock_page_cgroup().
2169 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2170 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2171 * before USED bit, we need memory barrier here.
2172 * See mem_cgroup_add_lru_list(), etc.
2176 case MEM_CGROUP_CHARGE_TYPE_CACHE
:
2177 case MEM_CGROUP_CHARGE_TYPE_SHMEM
:
2178 SetPageCgroupCache(pc
);
2179 SetPageCgroupUsed(pc
);
2181 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2182 ClearPageCgroupCache(pc
);
2183 SetPageCgroupUsed(pc
);
2189 mem_cgroup_charge_statistics(mem
, PageCgroupCache(pc
), nr_pages
);
2190 unlock_page_cgroup(pc
);
2192 * "charge_statistics" updated event counter. Then, check it.
2193 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2194 * if they exceeds softlimit.
2196 memcg_check_events(mem
, page
);
2199 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2201 #define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\
2202 (1 << PCG_ACCT_LRU) | (1 << PCG_MIGRATION))
2204 * Because tail pages are not marked as "used", set it. We're under
2205 * zone->lru_lock, 'splitting on pmd' and compund_lock.
2207 void mem_cgroup_split_huge_fixup(struct page
*head
, struct page
*tail
)
2209 struct page_cgroup
*head_pc
= lookup_page_cgroup(head
);
2210 struct page_cgroup
*tail_pc
= lookup_page_cgroup(tail
);
2211 unsigned long flags
;
2213 if (mem_cgroup_disabled())
2216 * We have no races with charge/uncharge but will have races with
2217 * page state accounting.
2219 move_lock_page_cgroup(head_pc
, &flags
);
2221 tail_pc
->mem_cgroup
= head_pc
->mem_cgroup
;
2222 smp_wmb(); /* see __commit_charge() */
2223 if (PageCgroupAcctLRU(head_pc
)) {
2225 struct mem_cgroup_per_zone
*mz
;
2228 * LRU flags cannot be copied because we need to add tail
2229 *.page to LRU by generic call and our hook will be called.
2230 * We hold lru_lock, then, reduce counter directly.
2232 lru
= page_lru(head
);
2233 mz
= page_cgroup_zoneinfo(head_pc
->mem_cgroup
, head
);
2234 MEM_CGROUP_ZSTAT(mz
, lru
) -= 1;
2236 tail_pc
->flags
= head_pc
->flags
& ~PCGF_NOCOPY_AT_SPLIT
;
2237 move_unlock_page_cgroup(head_pc
, &flags
);
2242 * mem_cgroup_move_account - move account of the page
2244 * @nr_pages: number of regular pages (>1 for huge pages)
2245 * @pc: page_cgroup of the page.
2246 * @from: mem_cgroup which the page is moved from.
2247 * @to: mem_cgroup which the page is moved to. @from != @to.
2248 * @uncharge: whether we should call uncharge and css_put against @from.
2250 * The caller must confirm following.
2251 * - page is not on LRU (isolate_page() is useful.)
2252 * - compound_lock is held when nr_pages > 1
2254 * This function doesn't do "charge" nor css_get to new cgroup. It should be
2255 * done by a caller(__mem_cgroup_try_charge would be usefull). If @uncharge is
2256 * true, this function does "uncharge" from old cgroup, but it doesn't if
2257 * @uncharge is false, so a caller should do "uncharge".
2259 static int mem_cgroup_move_account(struct page
*page
,
2260 unsigned int nr_pages
,
2261 struct page_cgroup
*pc
,
2262 struct mem_cgroup
*from
,
2263 struct mem_cgroup
*to
,
2266 unsigned long flags
;
2269 VM_BUG_ON(from
== to
);
2270 VM_BUG_ON(PageLRU(page
));
2272 * The page is isolated from LRU. So, collapse function
2273 * will not handle this page. But page splitting can happen.
2274 * Do this check under compound_page_lock(). The caller should
2278 if (nr_pages
> 1 && !PageTransHuge(page
))
2281 lock_page_cgroup(pc
);
2284 if (!PageCgroupUsed(pc
) || pc
->mem_cgroup
!= from
)
2287 move_lock_page_cgroup(pc
, &flags
);
2289 if (PageCgroupFileMapped(pc
)) {
2290 /* Update mapped_file data for mem_cgroup */
2292 __this_cpu_dec(from
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2293 __this_cpu_inc(to
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2296 mem_cgroup_charge_statistics(from
, PageCgroupCache(pc
), -nr_pages
);
2298 /* This is not "cancel", but cancel_charge does all we need. */
2299 __mem_cgroup_cancel_charge(from
, nr_pages
);
2301 /* caller should have done css_get */
2302 pc
->mem_cgroup
= to
;
2303 mem_cgroup_charge_statistics(to
, PageCgroupCache(pc
), nr_pages
);
2305 * We charges against "to" which may not have any tasks. Then, "to"
2306 * can be under rmdir(). But in current implementation, caller of
2307 * this function is just force_empty() and move charge, so it's
2308 * garanteed that "to" is never removed. So, we don't check rmdir
2311 move_unlock_page_cgroup(pc
, &flags
);
2314 unlock_page_cgroup(pc
);
2318 memcg_check_events(to
, page
);
2319 memcg_check_events(from
, page
);
2325 * move charges to its parent.
2328 static int mem_cgroup_move_parent(struct page
*page
,
2329 struct page_cgroup
*pc
,
2330 struct mem_cgroup
*child
,
2333 struct cgroup
*cg
= child
->css
.cgroup
;
2334 struct cgroup
*pcg
= cg
->parent
;
2335 struct mem_cgroup
*parent
;
2336 unsigned int nr_pages
;
2337 unsigned long uninitialized_var(flags
);
2345 if (!get_page_unless_zero(page
))
2347 if (isolate_lru_page(page
))
2350 nr_pages
= hpage_nr_pages(page
);
2352 parent
= mem_cgroup_from_cont(pcg
);
2353 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, nr_pages
, &parent
, false);
2358 flags
= compound_lock_irqsave(page
);
2360 ret
= mem_cgroup_move_account(page
, nr_pages
, pc
, child
, parent
, true);
2362 __mem_cgroup_cancel_charge(parent
, nr_pages
);
2365 compound_unlock_irqrestore(page
, flags
);
2367 putback_lru_page(page
);
2375 * Charge the memory controller for page usage.
2377 * 0 if the charge was successful
2378 * < 0 if the cgroup is over its limit
2380 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
2381 gfp_t gfp_mask
, enum charge_type ctype
)
2383 struct mem_cgroup
*mem
= NULL
;
2384 unsigned int nr_pages
= 1;
2385 struct page_cgroup
*pc
;
2389 if (PageTransHuge(page
)) {
2390 nr_pages
<<= compound_order(page
);
2391 VM_BUG_ON(!PageTransHuge(page
));
2393 * Never OOM-kill a process for a huge page. The
2394 * fault handler will fall back to regular pages.
2399 pc
= lookup_page_cgroup(page
);
2400 BUG_ON(!pc
); /* XXX: remove this and move pc lookup into commit */
2402 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, nr_pages
, &mem
, oom
);
2406 __mem_cgroup_commit_charge(mem
, page
, nr_pages
, pc
, ctype
);
2410 int mem_cgroup_newpage_charge(struct page
*page
,
2411 struct mm_struct
*mm
, gfp_t gfp_mask
)
2413 if (mem_cgroup_disabled())
2416 * If already mapped, we don't have to account.
2417 * If page cache, page->mapping has address_space.
2418 * But page->mapping may have out-of-use anon_vma pointer,
2419 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
2422 if (page_mapped(page
) || (page
->mapping
&& !PageAnon(page
)))
2426 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2427 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2431 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2432 enum charge_type ctype
);
2434 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
2439 if (mem_cgroup_disabled())
2441 if (PageCompound(page
))
2444 * Corner case handling. This is called from add_to_page_cache()
2445 * in usual. But some FS (shmem) precharges this page before calling it
2446 * and call add_to_page_cache() with GFP_NOWAIT.
2448 * For GFP_NOWAIT case, the page may be pre-charged before calling
2449 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
2450 * charge twice. (It works but has to pay a bit larger cost.)
2451 * And when the page is SwapCache, it should take swap information
2452 * into account. This is under lock_page() now.
2454 if (!(gfp_mask
& __GFP_WAIT
)) {
2455 struct page_cgroup
*pc
;
2457 pc
= lookup_page_cgroup(page
);
2460 lock_page_cgroup(pc
);
2461 if (PageCgroupUsed(pc
)) {
2462 unlock_page_cgroup(pc
);
2465 unlock_page_cgroup(pc
);
2471 if (page_is_file_cache(page
))
2472 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2473 MEM_CGROUP_CHARGE_TYPE_CACHE
);
2476 if (PageSwapCache(page
)) {
2477 struct mem_cgroup
*mem
;
2479 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &mem
);
2481 __mem_cgroup_commit_charge_swapin(page
, mem
,
2482 MEM_CGROUP_CHARGE_TYPE_SHMEM
);
2484 ret
= mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2485 MEM_CGROUP_CHARGE_TYPE_SHMEM
);
2491 * While swap-in, try_charge -> commit or cancel, the page is locked.
2492 * And when try_charge() successfully returns, one refcnt to memcg without
2493 * struct page_cgroup is acquired. This refcnt will be consumed by
2494 * "commit()" or removed by "cancel()"
2496 int mem_cgroup_try_charge_swapin(struct mm_struct
*mm
,
2498 gfp_t mask
, struct mem_cgroup
**ptr
)
2500 struct mem_cgroup
*mem
;
2505 if (mem_cgroup_disabled())
2508 if (!do_swap_account
)
2511 * A racing thread's fault, or swapoff, may have already updated
2512 * the pte, and even removed page from swap cache: in those cases
2513 * do_swap_page()'s pte_same() test will fail; but there's also a
2514 * KSM case which does need to charge the page.
2516 if (!PageSwapCache(page
))
2518 mem
= try_get_mem_cgroup_from_page(page
);
2522 ret
= __mem_cgroup_try_charge(NULL
, mask
, 1, ptr
, true);
2528 return __mem_cgroup_try_charge(mm
, mask
, 1, ptr
, true);
2532 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2533 enum charge_type ctype
)
2535 struct page_cgroup
*pc
;
2537 if (mem_cgroup_disabled())
2541 cgroup_exclude_rmdir(&ptr
->css
);
2542 pc
= lookup_page_cgroup(page
);
2543 mem_cgroup_lru_del_before_commit_swapcache(page
);
2544 __mem_cgroup_commit_charge(ptr
, page
, 1, pc
, ctype
);
2545 mem_cgroup_lru_add_after_commit_swapcache(page
);
2547 * Now swap is on-memory. This means this page may be
2548 * counted both as mem and swap....double count.
2549 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2550 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2551 * may call delete_from_swap_cache() before reach here.
2553 if (do_swap_account
&& PageSwapCache(page
)) {
2554 swp_entry_t ent
= {.val
= page_private(page
)};
2556 struct mem_cgroup
*memcg
;
2558 id
= swap_cgroup_record(ent
, 0);
2560 memcg
= mem_cgroup_lookup(id
);
2563 * This recorded memcg can be obsolete one. So, avoid
2564 * calling css_tryget
2566 if (!mem_cgroup_is_root(memcg
))
2567 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
2568 mem_cgroup_swap_statistics(memcg
, false);
2569 mem_cgroup_put(memcg
);
2574 * At swapin, we may charge account against cgroup which has no tasks.
2575 * So, rmdir()->pre_destroy() can be called while we do this charge.
2576 * In that case, we need to call pre_destroy() again. check it here.
2578 cgroup_release_and_wakeup_rmdir(&ptr
->css
);
2581 void mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
)
2583 __mem_cgroup_commit_charge_swapin(page
, ptr
,
2584 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2587 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup
*mem
)
2589 if (mem_cgroup_disabled())
2593 __mem_cgroup_cancel_charge(mem
, 1);
2596 static void mem_cgroup_do_uncharge(struct mem_cgroup
*mem
,
2597 unsigned int nr_pages
,
2598 const enum charge_type ctype
)
2600 struct memcg_batch_info
*batch
= NULL
;
2601 bool uncharge_memsw
= true;
2603 /* If swapout, usage of swap doesn't decrease */
2604 if (!do_swap_account
|| ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
)
2605 uncharge_memsw
= false;
2607 batch
= ¤t
->memcg_batch
;
2609 * In usual, we do css_get() when we remember memcg pointer.
2610 * But in this case, we keep res->usage until end of a series of
2611 * uncharges. Then, it's ok to ignore memcg's refcnt.
2616 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2617 * In those cases, all pages freed continously can be expected to be in
2618 * the same cgroup and we have chance to coalesce uncharges.
2619 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2620 * because we want to do uncharge as soon as possible.
2623 if (!batch
->do_batch
|| test_thread_flag(TIF_MEMDIE
))
2624 goto direct_uncharge
;
2627 goto direct_uncharge
;
2630 * In typical case, batch->memcg == mem. This means we can
2631 * merge a series of uncharges to an uncharge of res_counter.
2632 * If not, we uncharge res_counter ony by one.
2634 if (batch
->memcg
!= mem
)
2635 goto direct_uncharge
;
2636 /* remember freed charge and uncharge it later */
2639 batch
->memsw_nr_pages
++;
2642 res_counter_uncharge(&mem
->res
, nr_pages
* PAGE_SIZE
);
2644 res_counter_uncharge(&mem
->memsw
, nr_pages
* PAGE_SIZE
);
2645 if (unlikely(batch
->memcg
!= mem
))
2646 memcg_oom_recover(mem
);
2651 * uncharge if !page_mapped(page)
2653 static struct mem_cgroup
*
2654 __mem_cgroup_uncharge_common(struct page
*page
, enum charge_type ctype
)
2656 struct mem_cgroup
*mem
= NULL
;
2657 unsigned int nr_pages
= 1;
2658 struct page_cgroup
*pc
;
2660 if (mem_cgroup_disabled())
2663 if (PageSwapCache(page
))
2666 if (PageTransHuge(page
)) {
2667 nr_pages
<<= compound_order(page
);
2668 VM_BUG_ON(!PageTransHuge(page
));
2671 * Check if our page_cgroup is valid
2673 pc
= lookup_page_cgroup(page
);
2674 if (unlikely(!pc
|| !PageCgroupUsed(pc
)))
2677 lock_page_cgroup(pc
);
2679 mem
= pc
->mem_cgroup
;
2681 if (!PageCgroupUsed(pc
))
2685 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2686 case MEM_CGROUP_CHARGE_TYPE_DROP
:
2687 /* See mem_cgroup_prepare_migration() */
2688 if (page_mapped(page
) || PageCgroupMigration(pc
))
2691 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT
:
2692 if (!PageAnon(page
)) { /* Shared memory */
2693 if (page
->mapping
&& !page_is_file_cache(page
))
2695 } else if (page_mapped(page
)) /* Anon */
2702 mem_cgroup_charge_statistics(mem
, PageCgroupCache(pc
), -nr_pages
);
2704 ClearPageCgroupUsed(pc
);
2706 * pc->mem_cgroup is not cleared here. It will be accessed when it's
2707 * freed from LRU. This is safe because uncharged page is expected not
2708 * to be reused (freed soon). Exception is SwapCache, it's handled by
2709 * special functions.
2712 unlock_page_cgroup(pc
);
2714 * even after unlock, we have mem->res.usage here and this memcg
2715 * will never be freed.
2717 memcg_check_events(mem
, page
);
2718 if (do_swap_account
&& ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
) {
2719 mem_cgroup_swap_statistics(mem
, true);
2720 mem_cgroup_get(mem
);
2722 if (!mem_cgroup_is_root(mem
))
2723 mem_cgroup_do_uncharge(mem
, nr_pages
, ctype
);
2728 unlock_page_cgroup(pc
);
2732 void mem_cgroup_uncharge_page(struct page
*page
)
2735 if (page_mapped(page
))
2737 if (page
->mapping
&& !PageAnon(page
))
2739 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2742 void mem_cgroup_uncharge_cache_page(struct page
*page
)
2744 VM_BUG_ON(page_mapped(page
));
2745 VM_BUG_ON(page
->mapping
);
2746 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_CACHE
);
2750 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
2751 * In that cases, pages are freed continuously and we can expect pages
2752 * are in the same memcg. All these calls itself limits the number of
2753 * pages freed at once, then uncharge_start/end() is called properly.
2754 * This may be called prural(2) times in a context,
2757 void mem_cgroup_uncharge_start(void)
2759 current
->memcg_batch
.do_batch
++;
2760 /* We can do nest. */
2761 if (current
->memcg_batch
.do_batch
== 1) {
2762 current
->memcg_batch
.memcg
= NULL
;
2763 current
->memcg_batch
.nr_pages
= 0;
2764 current
->memcg_batch
.memsw_nr_pages
= 0;
2768 void mem_cgroup_uncharge_end(void)
2770 struct memcg_batch_info
*batch
= ¤t
->memcg_batch
;
2772 if (!batch
->do_batch
)
2776 if (batch
->do_batch
) /* If stacked, do nothing. */
2782 * This "batch->memcg" is valid without any css_get/put etc...
2783 * bacause we hide charges behind us.
2785 if (batch
->nr_pages
)
2786 res_counter_uncharge(&batch
->memcg
->res
,
2787 batch
->nr_pages
* PAGE_SIZE
);
2788 if (batch
->memsw_nr_pages
)
2789 res_counter_uncharge(&batch
->memcg
->memsw
,
2790 batch
->memsw_nr_pages
* PAGE_SIZE
);
2791 memcg_oom_recover(batch
->memcg
);
2792 /* forget this pointer (for sanity check) */
2793 batch
->memcg
= NULL
;
2798 * called after __delete_from_swap_cache() and drop "page" account.
2799 * memcg information is recorded to swap_cgroup of "ent"
2802 mem_cgroup_uncharge_swapcache(struct page
*page
, swp_entry_t ent
, bool swapout
)
2804 struct mem_cgroup
*memcg
;
2805 int ctype
= MEM_CGROUP_CHARGE_TYPE_SWAPOUT
;
2807 if (!swapout
) /* this was a swap cache but the swap is unused ! */
2808 ctype
= MEM_CGROUP_CHARGE_TYPE_DROP
;
2810 memcg
= __mem_cgroup_uncharge_common(page
, ctype
);
2813 * record memcg information, if swapout && memcg != NULL,
2814 * mem_cgroup_get() was called in uncharge().
2816 if (do_swap_account
&& swapout
&& memcg
)
2817 swap_cgroup_record(ent
, css_id(&memcg
->css
));
2821 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2823 * called from swap_entry_free(). remove record in swap_cgroup and
2824 * uncharge "memsw" account.
2826 void mem_cgroup_uncharge_swap(swp_entry_t ent
)
2828 struct mem_cgroup
*memcg
;
2831 if (!do_swap_account
)
2834 id
= swap_cgroup_record(ent
, 0);
2836 memcg
= mem_cgroup_lookup(id
);
2839 * We uncharge this because swap is freed.
2840 * This memcg can be obsolete one. We avoid calling css_tryget
2842 if (!mem_cgroup_is_root(memcg
))
2843 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
2844 mem_cgroup_swap_statistics(memcg
, false);
2845 mem_cgroup_put(memcg
);
2851 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2852 * @entry: swap entry to be moved
2853 * @from: mem_cgroup which the entry is moved from
2854 * @to: mem_cgroup which the entry is moved to
2855 * @need_fixup: whether we should fixup res_counters and refcounts.
2857 * It succeeds only when the swap_cgroup's record for this entry is the same
2858 * as the mem_cgroup's id of @from.
2860 * Returns 0 on success, -EINVAL on failure.
2862 * The caller must have charged to @to, IOW, called res_counter_charge() about
2863 * both res and memsw, and called css_get().
2865 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
2866 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
2868 unsigned short old_id
, new_id
;
2870 old_id
= css_id(&from
->css
);
2871 new_id
= css_id(&to
->css
);
2873 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
2874 mem_cgroup_swap_statistics(from
, false);
2875 mem_cgroup_swap_statistics(to
, true);
2877 * This function is only called from task migration context now.
2878 * It postpones res_counter and refcount handling till the end
2879 * of task migration(mem_cgroup_clear_mc()) for performance
2880 * improvement. But we cannot postpone mem_cgroup_get(to)
2881 * because if the process that has been moved to @to does
2882 * swap-in, the refcount of @to might be decreased to 0.
2886 if (!mem_cgroup_is_root(from
))
2887 res_counter_uncharge(&from
->memsw
, PAGE_SIZE
);
2888 mem_cgroup_put(from
);
2890 * we charged both to->res and to->memsw, so we should
2893 if (!mem_cgroup_is_root(to
))
2894 res_counter_uncharge(&to
->res
, PAGE_SIZE
);
2901 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
2902 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
2909 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
2912 int mem_cgroup_prepare_migration(struct page
*page
,
2913 struct page
*newpage
, struct mem_cgroup
**ptr
, gfp_t gfp_mask
)
2915 struct mem_cgroup
*mem
= NULL
;
2916 struct page_cgroup
*pc
;
2917 enum charge_type ctype
;
2922 VM_BUG_ON(PageTransHuge(page
));
2923 if (mem_cgroup_disabled())
2926 pc
= lookup_page_cgroup(page
);
2927 lock_page_cgroup(pc
);
2928 if (PageCgroupUsed(pc
)) {
2929 mem
= pc
->mem_cgroup
;
2932 * At migrating an anonymous page, its mapcount goes down
2933 * to 0 and uncharge() will be called. But, even if it's fully
2934 * unmapped, migration may fail and this page has to be
2935 * charged again. We set MIGRATION flag here and delay uncharge
2936 * until end_migration() is called
2938 * Corner Case Thinking
2940 * When the old page was mapped as Anon and it's unmap-and-freed
2941 * while migration was ongoing.
2942 * If unmap finds the old page, uncharge() of it will be delayed
2943 * until end_migration(). If unmap finds a new page, it's
2944 * uncharged when it make mapcount to be 1->0. If unmap code
2945 * finds swap_migration_entry, the new page will not be mapped
2946 * and end_migration() will find it(mapcount==0).
2949 * When the old page was mapped but migraion fails, the kernel
2950 * remaps it. A charge for it is kept by MIGRATION flag even
2951 * if mapcount goes down to 0. We can do remap successfully
2952 * without charging it again.
2955 * The "old" page is under lock_page() until the end of
2956 * migration, so, the old page itself will not be swapped-out.
2957 * If the new page is swapped out before end_migraton, our
2958 * hook to usual swap-out path will catch the event.
2961 SetPageCgroupMigration(pc
);
2963 unlock_page_cgroup(pc
);
2965 * If the page is not charged at this point,
2972 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, 1, ptr
, false);
2973 css_put(&mem
->css
);/* drop extra refcnt */
2974 if (ret
|| *ptr
== NULL
) {
2975 if (PageAnon(page
)) {
2976 lock_page_cgroup(pc
);
2977 ClearPageCgroupMigration(pc
);
2978 unlock_page_cgroup(pc
);
2980 * The old page may be fully unmapped while we kept it.
2982 mem_cgroup_uncharge_page(page
);
2987 * We charge new page before it's used/mapped. So, even if unlock_page()
2988 * is called before end_migration, we can catch all events on this new
2989 * page. In the case new page is migrated but not remapped, new page's
2990 * mapcount will be finally 0 and we call uncharge in end_migration().
2992 pc
= lookup_page_cgroup(newpage
);
2994 ctype
= MEM_CGROUP_CHARGE_TYPE_MAPPED
;
2995 else if (page_is_file_cache(page
))
2996 ctype
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
2998 ctype
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
2999 __mem_cgroup_commit_charge(mem
, page
, 1, pc
, ctype
);
3003 /* remove redundant charge if migration failed*/
3004 void mem_cgroup_end_migration(struct mem_cgroup
*mem
,
3005 struct page
*oldpage
, struct page
*newpage
, bool migration_ok
)
3007 struct page
*used
, *unused
;
3008 struct page_cgroup
*pc
;
3012 /* blocks rmdir() */
3013 cgroup_exclude_rmdir(&mem
->css
);
3014 if (!migration_ok
) {
3022 * We disallowed uncharge of pages under migration because mapcount
3023 * of the page goes down to zero, temporarly.
3024 * Clear the flag and check the page should be charged.
3026 pc
= lookup_page_cgroup(oldpage
);
3027 lock_page_cgroup(pc
);
3028 ClearPageCgroupMigration(pc
);
3029 unlock_page_cgroup(pc
);
3031 __mem_cgroup_uncharge_common(unused
, MEM_CGROUP_CHARGE_TYPE_FORCE
);
3034 * If a page is a file cache, radix-tree replacement is very atomic
3035 * and we can skip this check. When it was an Anon page, its mapcount
3036 * goes down to 0. But because we added MIGRATION flage, it's not
3037 * uncharged yet. There are several case but page->mapcount check
3038 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3039 * check. (see prepare_charge() also)
3042 mem_cgroup_uncharge_page(used
);
3044 * At migration, we may charge account against cgroup which has no
3046 * So, rmdir()->pre_destroy() can be called while we do this charge.
3047 * In that case, we need to call pre_destroy() again. check it here.
3049 cgroup_release_and_wakeup_rmdir(&mem
->css
);
3053 * A call to try to shrink memory usage on charge failure at shmem's swapin.
3054 * Calling hierarchical_reclaim is not enough because we should update
3055 * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM.
3056 * Moreover considering hierarchy, we should reclaim from the mem_over_limit,
3057 * not from the memcg which this page would be charged to.
3058 * try_charge_swapin does all of these works properly.
3060 int mem_cgroup_shmem_charge_fallback(struct page
*page
,
3061 struct mm_struct
*mm
,
3064 struct mem_cgroup
*mem
;
3067 if (mem_cgroup_disabled())
3070 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &mem
);
3072 mem_cgroup_cancel_charge_swapin(mem
); /* it does !mem check */
3077 #ifdef CONFIG_DEBUG_VM
3078 static struct page_cgroup
*lookup_page_cgroup_used(struct page
*page
)
3080 struct page_cgroup
*pc
;
3082 pc
= lookup_page_cgroup(page
);
3083 if (likely(pc
) && PageCgroupUsed(pc
))
3088 bool mem_cgroup_bad_page_check(struct page
*page
)
3090 if (mem_cgroup_disabled())
3093 return lookup_page_cgroup_used(page
) != NULL
;
3096 void mem_cgroup_print_bad_page(struct page
*page
)
3098 struct page_cgroup
*pc
;
3100 pc
= lookup_page_cgroup_used(page
);
3105 printk(KERN_ALERT
"pc:%p pc->flags:%lx pc->mem_cgroup:%p",
3106 pc
, pc
->flags
, pc
->mem_cgroup
);
3108 path
= kmalloc(PATH_MAX
, GFP_KERNEL
);
3111 ret
= cgroup_path(pc
->mem_cgroup
->css
.cgroup
,
3116 printk(KERN_CONT
"(%s)\n",
3117 (ret
< 0) ? "cannot get the path" : path
);
3123 static DEFINE_MUTEX(set_limit_mutex
);
3125 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
3126 unsigned long long val
)
3129 u64 memswlimit
, memlimit
;
3131 int children
= mem_cgroup_count_children(memcg
);
3132 u64 curusage
, oldusage
;
3136 * For keeping hierarchical_reclaim simple, how long we should retry
3137 * is depends on callers. We set our retry-count to be function
3138 * of # of children which we should visit in this loop.
3140 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
* children
;
3142 oldusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3145 while (retry_count
) {
3146 if (signal_pending(current
)) {
3151 * Rather than hide all in some function, I do this in
3152 * open coded manner. You see what this really does.
3153 * We have to guarantee mem->res.limit < mem->memsw.limit.
3155 mutex_lock(&set_limit_mutex
);
3156 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3157 if (memswlimit
< val
) {
3159 mutex_unlock(&set_limit_mutex
);
3163 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3167 ret
= res_counter_set_limit(&memcg
->res
, val
);
3169 if (memswlimit
== val
)
3170 memcg
->memsw_is_minimum
= true;
3172 memcg
->memsw_is_minimum
= false;
3174 mutex_unlock(&set_limit_mutex
);
3179 mem_cgroup_hierarchical_reclaim(memcg
, NULL
, GFP_KERNEL
,
3180 MEM_CGROUP_RECLAIM_SHRINK
);
3181 curusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3182 /* Usage is reduced ? */
3183 if (curusage
>= oldusage
)
3186 oldusage
= curusage
;
3188 if (!ret
&& enlarge
)
3189 memcg_oom_recover(memcg
);
3194 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
3195 unsigned long long val
)
3198 u64 memlimit
, memswlimit
, oldusage
, curusage
;
3199 int children
= mem_cgroup_count_children(memcg
);
3203 /* see mem_cgroup_resize_res_limit */
3204 retry_count
= children
* MEM_CGROUP_RECLAIM_RETRIES
;
3205 oldusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3206 while (retry_count
) {
3207 if (signal_pending(current
)) {
3212 * Rather than hide all in some function, I do this in
3213 * open coded manner. You see what this really does.
3214 * We have to guarantee mem->res.limit < mem->memsw.limit.
3216 mutex_lock(&set_limit_mutex
);
3217 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3218 if (memlimit
> val
) {
3220 mutex_unlock(&set_limit_mutex
);
3223 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3224 if (memswlimit
< val
)
3226 ret
= res_counter_set_limit(&memcg
->memsw
, val
);
3228 if (memlimit
== val
)
3229 memcg
->memsw_is_minimum
= true;
3231 memcg
->memsw_is_minimum
= false;
3233 mutex_unlock(&set_limit_mutex
);
3238 mem_cgroup_hierarchical_reclaim(memcg
, NULL
, GFP_KERNEL
,
3239 MEM_CGROUP_RECLAIM_NOSWAP
|
3240 MEM_CGROUP_RECLAIM_SHRINK
);
3241 curusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3242 /* Usage is reduced ? */
3243 if (curusage
>= oldusage
)
3246 oldusage
= curusage
;
3248 if (!ret
&& enlarge
)
3249 memcg_oom_recover(memcg
);
3253 unsigned long mem_cgroup_soft_limit_reclaim(struct zone
*zone
, int order
,
3256 unsigned long nr_reclaimed
= 0;
3257 struct mem_cgroup_per_zone
*mz
, *next_mz
= NULL
;
3258 unsigned long reclaimed
;
3260 struct mem_cgroup_tree_per_zone
*mctz
;
3261 unsigned long long excess
;
3266 mctz
= soft_limit_tree_node_zone(zone_to_nid(zone
), zone_idx(zone
));
3268 * This loop can run a while, specially if mem_cgroup's continuously
3269 * keep exceeding their soft limit and putting the system under
3276 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
3280 reclaimed
= mem_cgroup_hierarchical_reclaim(mz
->mem
, zone
,
3282 MEM_CGROUP_RECLAIM_SOFT
);
3283 nr_reclaimed
+= reclaimed
;
3284 spin_lock(&mctz
->lock
);
3287 * If we failed to reclaim anything from this memory cgroup
3288 * it is time to move on to the next cgroup
3294 * Loop until we find yet another one.
3296 * By the time we get the soft_limit lock
3297 * again, someone might have aded the
3298 * group back on the RB tree. Iterate to
3299 * make sure we get a different mem.
3300 * mem_cgroup_largest_soft_limit_node returns
3301 * NULL if no other cgroup is present on
3305 __mem_cgroup_largest_soft_limit_node(mctz
);
3306 if (next_mz
== mz
) {
3307 css_put(&next_mz
->mem
->css
);
3309 } else /* next_mz == NULL or other memcg */
3313 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
3314 excess
= res_counter_soft_limit_excess(&mz
->mem
->res
);
3316 * One school of thought says that we should not add
3317 * back the node to the tree if reclaim returns 0.
3318 * But our reclaim could return 0, simply because due
3319 * to priority we are exposing a smaller subset of
3320 * memory to reclaim from. Consider this as a longer
3323 /* If excess == 0, no tree ops */
3324 __mem_cgroup_insert_exceeded(mz
->mem
, mz
, mctz
, excess
);
3325 spin_unlock(&mctz
->lock
);
3326 css_put(&mz
->mem
->css
);
3329 * Could not reclaim anything and there are no more
3330 * mem cgroups to try or we seem to be looping without
3331 * reclaiming anything.
3333 if (!nr_reclaimed
&&
3335 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
3337 } while (!nr_reclaimed
);
3339 css_put(&next_mz
->mem
->css
);
3340 return nr_reclaimed
;
3344 * This routine traverse page_cgroup in given list and drop them all.
3345 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3347 static int mem_cgroup_force_empty_list(struct mem_cgroup
*mem
,
3348 int node
, int zid
, enum lru_list lru
)
3351 struct mem_cgroup_per_zone
*mz
;
3352 struct page_cgroup
*pc
, *busy
;
3353 unsigned long flags
, loop
;
3354 struct list_head
*list
;
3357 zone
= &NODE_DATA(node
)->node_zones
[zid
];
3358 mz
= mem_cgroup_zoneinfo(mem
, node
, zid
);
3359 list
= &mz
->lists
[lru
];
3361 loop
= MEM_CGROUP_ZSTAT(mz
, lru
);
3362 /* give some margin against EBUSY etc...*/
3369 spin_lock_irqsave(&zone
->lru_lock
, flags
);
3370 if (list_empty(list
)) {
3371 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3374 pc
= list_entry(list
->prev
, struct page_cgroup
, lru
);
3376 list_move(&pc
->lru
, list
);
3378 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3381 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3383 page
= lookup_cgroup_page(pc
);
3385 ret
= mem_cgroup_move_parent(page
, pc
, mem
, GFP_KERNEL
);
3389 if (ret
== -EBUSY
|| ret
== -EINVAL
) {
3390 /* found lock contention or "pc" is obsolete. */
3397 if (!ret
&& !list_empty(list
))
3403 * make mem_cgroup's charge to be 0 if there is no task.
3404 * This enables deleting this mem_cgroup.
3406 static int mem_cgroup_force_empty(struct mem_cgroup
*mem
, bool free_all
)
3409 int node
, zid
, shrink
;
3410 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
3411 struct cgroup
*cgrp
= mem
->css
.cgroup
;
3416 /* should free all ? */
3422 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
))
3425 if (signal_pending(current
))
3427 /* This is for making all *used* pages to be on LRU. */
3428 lru_add_drain_all();
3429 drain_all_stock_sync();
3431 mem_cgroup_start_move(mem
);
3432 for_each_node_state(node
, N_HIGH_MEMORY
) {
3433 for (zid
= 0; !ret
&& zid
< MAX_NR_ZONES
; zid
++) {
3436 ret
= mem_cgroup_force_empty_list(mem
,
3445 mem_cgroup_end_move(mem
);
3446 memcg_oom_recover(mem
);
3447 /* it seems parent cgroup doesn't have enough mem */
3451 /* "ret" should also be checked to ensure all lists are empty. */
3452 } while (mem
->res
.usage
> 0 || ret
);
3458 /* returns EBUSY if there is a task or if we come here twice. */
3459 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
) || shrink
) {
3463 /* we call try-to-free pages for make this cgroup empty */
3464 lru_add_drain_all();
3465 /* try to free all pages in this cgroup */
3467 while (nr_retries
&& mem
->res
.usage
> 0) {
3470 if (signal_pending(current
)) {
3474 progress
= try_to_free_mem_cgroup_pages(mem
, GFP_KERNEL
,
3475 false, get_swappiness(mem
));
3478 /* maybe some writeback is necessary */
3479 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3484 /* try move_account...there may be some *locked* pages. */
3488 int mem_cgroup_force_empty_write(struct cgroup
*cont
, unsigned int event
)
3490 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont
), true);
3494 static u64
mem_cgroup_hierarchy_read(struct cgroup
*cont
, struct cftype
*cft
)
3496 return mem_cgroup_from_cont(cont
)->use_hierarchy
;
3499 static int mem_cgroup_hierarchy_write(struct cgroup
*cont
, struct cftype
*cft
,
3503 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
3504 struct cgroup
*parent
= cont
->parent
;
3505 struct mem_cgroup
*parent_mem
= NULL
;
3508 parent_mem
= mem_cgroup_from_cont(parent
);
3512 * If parent's use_hierarchy is set, we can't make any modifications
3513 * in the child subtrees. If it is unset, then the change can
3514 * occur, provided the current cgroup has no children.
3516 * For the root cgroup, parent_mem is NULL, we allow value to be
3517 * set if there are no children.
3519 if ((!parent_mem
|| !parent_mem
->use_hierarchy
) &&
3520 (val
== 1 || val
== 0)) {
3521 if (list_empty(&cont
->children
))
3522 mem
->use_hierarchy
= val
;
3533 static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup
*mem
,
3534 enum mem_cgroup_stat_index idx
)
3536 struct mem_cgroup
*iter
;
3539 /* Per-cpu values can be negative, use a signed accumulator */
3540 for_each_mem_cgroup_tree(iter
, mem
)
3541 val
+= mem_cgroup_read_stat(iter
, idx
);
3543 if (val
< 0) /* race ? */
3548 static inline u64
mem_cgroup_usage(struct mem_cgroup
*mem
, bool swap
)
3552 if (!mem_cgroup_is_root(mem
)) {
3554 return res_counter_read_u64(&mem
->res
, RES_USAGE
);
3556 return res_counter_read_u64(&mem
->memsw
, RES_USAGE
);
3559 val
= mem_cgroup_recursive_stat(mem
, MEM_CGROUP_STAT_CACHE
);
3560 val
+= mem_cgroup_recursive_stat(mem
, MEM_CGROUP_STAT_RSS
);
3563 val
+= mem_cgroup_recursive_stat(mem
, MEM_CGROUP_STAT_SWAPOUT
);
3565 return val
<< PAGE_SHIFT
;
3568 static u64
mem_cgroup_read(struct cgroup
*cont
, struct cftype
*cft
)
3570 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
3574 type
= MEMFILE_TYPE(cft
->private);
3575 name
= MEMFILE_ATTR(cft
->private);
3578 if (name
== RES_USAGE
)
3579 val
= mem_cgroup_usage(mem
, false);
3581 val
= res_counter_read_u64(&mem
->res
, name
);
3584 if (name
== RES_USAGE
)
3585 val
= mem_cgroup_usage(mem
, true);
3587 val
= res_counter_read_u64(&mem
->memsw
, name
);
3596 * The user of this function is...
3599 static int mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
3602 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3604 unsigned long long val
;
3607 type
= MEMFILE_TYPE(cft
->private);
3608 name
= MEMFILE_ATTR(cft
->private);
3611 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
3615 /* This function does all necessary parse...reuse it */
3616 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3620 ret
= mem_cgroup_resize_limit(memcg
, val
);
3622 ret
= mem_cgroup_resize_memsw_limit(memcg
, val
);
3624 case RES_SOFT_LIMIT
:
3625 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3629 * For memsw, soft limits are hard to implement in terms
3630 * of semantics, for now, we support soft limits for
3631 * control without swap
3634 ret
= res_counter_set_soft_limit(&memcg
->res
, val
);
3639 ret
= -EINVAL
; /* should be BUG() ? */
3645 static void memcg_get_hierarchical_limit(struct mem_cgroup
*memcg
,
3646 unsigned long long *mem_limit
, unsigned long long *memsw_limit
)
3648 struct cgroup
*cgroup
;
3649 unsigned long long min_limit
, min_memsw_limit
, tmp
;
3651 min_limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3652 min_memsw_limit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3653 cgroup
= memcg
->css
.cgroup
;
3654 if (!memcg
->use_hierarchy
)
3657 while (cgroup
->parent
) {
3658 cgroup
= cgroup
->parent
;
3659 memcg
= mem_cgroup_from_cont(cgroup
);
3660 if (!memcg
->use_hierarchy
)
3662 tmp
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3663 min_limit
= min(min_limit
, tmp
);
3664 tmp
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3665 min_memsw_limit
= min(min_memsw_limit
, tmp
);
3668 *mem_limit
= min_limit
;
3669 *memsw_limit
= min_memsw_limit
;
3673 static int mem_cgroup_reset(struct cgroup
*cont
, unsigned int event
)
3675 struct mem_cgroup
*mem
;
3678 mem
= mem_cgroup_from_cont(cont
);
3679 type
= MEMFILE_TYPE(event
);
3680 name
= MEMFILE_ATTR(event
);
3684 res_counter_reset_max(&mem
->res
);
3686 res_counter_reset_max(&mem
->memsw
);
3690 res_counter_reset_failcnt(&mem
->res
);
3692 res_counter_reset_failcnt(&mem
->memsw
);
3699 static u64
mem_cgroup_move_charge_read(struct cgroup
*cgrp
,
3702 return mem_cgroup_from_cont(cgrp
)->move_charge_at_immigrate
;
3706 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
3707 struct cftype
*cft
, u64 val
)
3709 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
3711 if (val
>= (1 << NR_MOVE_TYPE
))
3714 * We check this value several times in both in can_attach() and
3715 * attach(), so we need cgroup lock to prevent this value from being
3719 mem
->move_charge_at_immigrate
= val
;
3725 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
3726 struct cftype
*cft
, u64 val
)
3733 /* For read statistics */
3749 struct mcs_total_stat
{
3750 s64 stat
[NR_MCS_STAT
];
3756 } memcg_stat_strings
[NR_MCS_STAT
] = {
3757 {"cache", "total_cache"},
3758 {"rss", "total_rss"},
3759 {"mapped_file", "total_mapped_file"},
3760 {"pgpgin", "total_pgpgin"},
3761 {"pgpgout", "total_pgpgout"},
3762 {"swap", "total_swap"},
3763 {"inactive_anon", "total_inactive_anon"},
3764 {"active_anon", "total_active_anon"},
3765 {"inactive_file", "total_inactive_file"},
3766 {"active_file", "total_active_file"},
3767 {"unevictable", "total_unevictable"}
3772 mem_cgroup_get_local_stat(struct mem_cgroup
*mem
, struct mcs_total_stat
*s
)
3777 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_CACHE
);
3778 s
->stat
[MCS_CACHE
] += val
* PAGE_SIZE
;
3779 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_RSS
);
3780 s
->stat
[MCS_RSS
] += val
* PAGE_SIZE
;
3781 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_FILE_MAPPED
);
3782 s
->stat
[MCS_FILE_MAPPED
] += val
* PAGE_SIZE
;
3783 val
= mem_cgroup_read_events(mem
, MEM_CGROUP_EVENTS_PGPGIN
);
3784 s
->stat
[MCS_PGPGIN
] += val
;
3785 val
= mem_cgroup_read_events(mem
, MEM_CGROUP_EVENTS_PGPGOUT
);
3786 s
->stat
[MCS_PGPGOUT
] += val
;
3787 if (do_swap_account
) {
3788 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_SWAPOUT
);
3789 s
->stat
[MCS_SWAP
] += val
* PAGE_SIZE
;
3793 val
= mem_cgroup_get_local_zonestat(mem
, LRU_INACTIVE_ANON
);
3794 s
->stat
[MCS_INACTIVE_ANON
] += val
* PAGE_SIZE
;
3795 val
= mem_cgroup_get_local_zonestat(mem
, LRU_ACTIVE_ANON
);
3796 s
->stat
[MCS_ACTIVE_ANON
] += val
* PAGE_SIZE
;
3797 val
= mem_cgroup_get_local_zonestat(mem
, LRU_INACTIVE_FILE
);
3798 s
->stat
[MCS_INACTIVE_FILE
] += val
* PAGE_SIZE
;
3799 val
= mem_cgroup_get_local_zonestat(mem
, LRU_ACTIVE_FILE
);
3800 s
->stat
[MCS_ACTIVE_FILE
] += val
* PAGE_SIZE
;
3801 val
= mem_cgroup_get_local_zonestat(mem
, LRU_UNEVICTABLE
);
3802 s
->stat
[MCS_UNEVICTABLE
] += val
* PAGE_SIZE
;
3806 mem_cgroup_get_total_stat(struct mem_cgroup
*mem
, struct mcs_total_stat
*s
)
3808 struct mem_cgroup
*iter
;
3810 for_each_mem_cgroup_tree(iter
, mem
)
3811 mem_cgroup_get_local_stat(iter
, s
);
3814 static int mem_control_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
3815 struct cgroup_map_cb
*cb
)
3817 struct mem_cgroup
*mem_cont
= mem_cgroup_from_cont(cont
);
3818 struct mcs_total_stat mystat
;
3821 memset(&mystat
, 0, sizeof(mystat
));
3822 mem_cgroup_get_local_stat(mem_cont
, &mystat
);
3824 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
3825 if (i
== MCS_SWAP
&& !do_swap_account
)
3827 cb
->fill(cb
, memcg_stat_strings
[i
].local_name
, mystat
.stat
[i
]);
3830 /* Hierarchical information */
3832 unsigned long long limit
, memsw_limit
;
3833 memcg_get_hierarchical_limit(mem_cont
, &limit
, &memsw_limit
);
3834 cb
->fill(cb
, "hierarchical_memory_limit", limit
);
3835 if (do_swap_account
)
3836 cb
->fill(cb
, "hierarchical_memsw_limit", memsw_limit
);
3839 memset(&mystat
, 0, sizeof(mystat
));
3840 mem_cgroup_get_total_stat(mem_cont
, &mystat
);
3841 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
3842 if (i
== MCS_SWAP
&& !do_swap_account
)
3844 cb
->fill(cb
, memcg_stat_strings
[i
].total_name
, mystat
.stat
[i
]);
3847 #ifdef CONFIG_DEBUG_VM
3848 cb
->fill(cb
, "inactive_ratio", calc_inactive_ratio(mem_cont
, NULL
));
3852 struct mem_cgroup_per_zone
*mz
;
3853 unsigned long recent_rotated
[2] = {0, 0};
3854 unsigned long recent_scanned
[2] = {0, 0};
3856 for_each_online_node(nid
)
3857 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
3858 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
3860 recent_rotated
[0] +=
3861 mz
->reclaim_stat
.recent_rotated
[0];
3862 recent_rotated
[1] +=
3863 mz
->reclaim_stat
.recent_rotated
[1];
3864 recent_scanned
[0] +=
3865 mz
->reclaim_stat
.recent_scanned
[0];
3866 recent_scanned
[1] +=
3867 mz
->reclaim_stat
.recent_scanned
[1];
3869 cb
->fill(cb
, "recent_rotated_anon", recent_rotated
[0]);
3870 cb
->fill(cb
, "recent_rotated_file", recent_rotated
[1]);
3871 cb
->fill(cb
, "recent_scanned_anon", recent_scanned
[0]);
3872 cb
->fill(cb
, "recent_scanned_file", recent_scanned
[1]);
3879 static u64
mem_cgroup_swappiness_read(struct cgroup
*cgrp
, struct cftype
*cft
)
3881 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3883 return get_swappiness(memcg
);
3886 static int mem_cgroup_swappiness_write(struct cgroup
*cgrp
, struct cftype
*cft
,
3889 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3890 struct mem_cgroup
*parent
;
3895 if (cgrp
->parent
== NULL
)
3898 parent
= mem_cgroup_from_cont(cgrp
->parent
);
3902 /* If under hierarchy, only empty-root can set this value */
3903 if ((parent
->use_hierarchy
) ||
3904 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
3909 memcg
->swappiness
= val
;
3916 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
3918 struct mem_cgroup_threshold_ary
*t
;
3924 t
= rcu_dereference(memcg
->thresholds
.primary
);
3926 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
3931 usage
= mem_cgroup_usage(memcg
, swap
);
3934 * current_threshold points to threshold just below usage.
3935 * If it's not true, a threshold was crossed after last
3936 * call of __mem_cgroup_threshold().
3938 i
= t
->current_threshold
;
3941 * Iterate backward over array of thresholds starting from
3942 * current_threshold and check if a threshold is crossed.
3943 * If none of thresholds below usage is crossed, we read
3944 * only one element of the array here.
3946 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
3947 eventfd_signal(t
->entries
[i
].eventfd
, 1);
3949 /* i = current_threshold + 1 */
3953 * Iterate forward over array of thresholds starting from
3954 * current_threshold+1 and check if a threshold is crossed.
3955 * If none of thresholds above usage is crossed, we read
3956 * only one element of the array here.
3958 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
3959 eventfd_signal(t
->entries
[i
].eventfd
, 1);
3961 /* Update current_threshold */
3962 t
->current_threshold
= i
- 1;
3967 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
3970 __mem_cgroup_threshold(memcg
, false);
3971 if (do_swap_account
)
3972 __mem_cgroup_threshold(memcg
, true);
3974 memcg
= parent_mem_cgroup(memcg
);
3978 static int compare_thresholds(const void *a
, const void *b
)
3980 const struct mem_cgroup_threshold
*_a
= a
;
3981 const struct mem_cgroup_threshold
*_b
= b
;
3983 return _a
->threshold
- _b
->threshold
;
3986 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*mem
)
3988 struct mem_cgroup_eventfd_list
*ev
;
3990 list_for_each_entry(ev
, &mem
->oom_notify
, list
)
3991 eventfd_signal(ev
->eventfd
, 1);
3995 static void mem_cgroup_oom_notify(struct mem_cgroup
*mem
)
3997 struct mem_cgroup
*iter
;
3999 for_each_mem_cgroup_tree(iter
, mem
)
4000 mem_cgroup_oom_notify_cb(iter
);
4003 static int mem_cgroup_usage_register_event(struct cgroup
*cgrp
,
4004 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4006 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4007 struct mem_cgroup_thresholds
*thresholds
;
4008 struct mem_cgroup_threshold_ary
*new;
4009 int type
= MEMFILE_TYPE(cft
->private);
4010 u64 threshold
, usage
;
4013 ret
= res_counter_memparse_write_strategy(args
, &threshold
);
4017 mutex_lock(&memcg
->thresholds_lock
);
4020 thresholds
= &memcg
->thresholds
;
4021 else if (type
== _MEMSWAP
)
4022 thresholds
= &memcg
->memsw_thresholds
;
4026 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4028 /* Check if a threshold crossed before adding a new one */
4029 if (thresholds
->primary
)
4030 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4032 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
4034 /* Allocate memory for new array of thresholds */
4035 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
4043 /* Copy thresholds (if any) to new array */
4044 if (thresholds
->primary
) {
4045 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
4046 sizeof(struct mem_cgroup_threshold
));
4049 /* Add new threshold */
4050 new->entries
[size
- 1].eventfd
= eventfd
;
4051 new->entries
[size
- 1].threshold
= threshold
;
4053 /* Sort thresholds. Registering of new threshold isn't time-critical */
4054 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
4055 compare_thresholds
, NULL
);
4057 /* Find current threshold */
4058 new->current_threshold
= -1;
4059 for (i
= 0; i
< size
; i
++) {
4060 if (new->entries
[i
].threshold
< usage
) {
4062 * new->current_threshold will not be used until
4063 * rcu_assign_pointer(), so it's safe to increment
4066 ++new->current_threshold
;
4070 /* Free old spare buffer and save old primary buffer as spare */
4071 kfree(thresholds
->spare
);
4072 thresholds
->spare
= thresholds
->primary
;
4074 rcu_assign_pointer(thresholds
->primary
, new);
4076 /* To be sure that nobody uses thresholds */
4080 mutex_unlock(&memcg
->thresholds_lock
);
4085 static void mem_cgroup_usage_unregister_event(struct cgroup
*cgrp
,
4086 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4088 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4089 struct mem_cgroup_thresholds
*thresholds
;
4090 struct mem_cgroup_threshold_ary
*new;
4091 int type
= MEMFILE_TYPE(cft
->private);
4095 mutex_lock(&memcg
->thresholds_lock
);
4097 thresholds
= &memcg
->thresholds
;
4098 else if (type
== _MEMSWAP
)
4099 thresholds
= &memcg
->memsw_thresholds
;
4104 * Something went wrong if we trying to unregister a threshold
4105 * if we don't have thresholds
4107 BUG_ON(!thresholds
);
4109 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4111 /* Check if a threshold crossed before removing */
4112 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4114 /* Calculate new number of threshold */
4116 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
4117 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
4121 new = thresholds
->spare
;
4123 /* Set thresholds array to NULL if we don't have thresholds */
4132 /* Copy thresholds and find current threshold */
4133 new->current_threshold
= -1;
4134 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
4135 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
4138 new->entries
[j
] = thresholds
->primary
->entries
[i
];
4139 if (new->entries
[j
].threshold
< usage
) {
4141 * new->current_threshold will not be used
4142 * until rcu_assign_pointer(), so it's safe to increment
4145 ++new->current_threshold
;
4151 /* Swap primary and spare array */
4152 thresholds
->spare
= thresholds
->primary
;
4153 rcu_assign_pointer(thresholds
->primary
, new);
4155 /* To be sure that nobody uses thresholds */
4158 mutex_unlock(&memcg
->thresholds_lock
);
4161 static int mem_cgroup_oom_register_event(struct cgroup
*cgrp
,
4162 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4164 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4165 struct mem_cgroup_eventfd_list
*event
;
4166 int type
= MEMFILE_TYPE(cft
->private);
4168 BUG_ON(type
!= _OOM_TYPE
);
4169 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
4173 mutex_lock(&memcg_oom_mutex
);
4175 event
->eventfd
= eventfd
;
4176 list_add(&event
->list
, &memcg
->oom_notify
);
4178 /* already in OOM ? */
4179 if (atomic_read(&memcg
->oom_lock
))
4180 eventfd_signal(eventfd
, 1);
4181 mutex_unlock(&memcg_oom_mutex
);
4186 static void mem_cgroup_oom_unregister_event(struct cgroup
*cgrp
,
4187 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4189 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
4190 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
4191 int type
= MEMFILE_TYPE(cft
->private);
4193 BUG_ON(type
!= _OOM_TYPE
);
4195 mutex_lock(&memcg_oom_mutex
);
4197 list_for_each_entry_safe(ev
, tmp
, &mem
->oom_notify
, list
) {
4198 if (ev
->eventfd
== eventfd
) {
4199 list_del(&ev
->list
);
4204 mutex_unlock(&memcg_oom_mutex
);
4207 static int mem_cgroup_oom_control_read(struct cgroup
*cgrp
,
4208 struct cftype
*cft
, struct cgroup_map_cb
*cb
)
4210 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
4212 cb
->fill(cb
, "oom_kill_disable", mem
->oom_kill_disable
);
4214 if (atomic_read(&mem
->oom_lock
))
4215 cb
->fill(cb
, "under_oom", 1);
4217 cb
->fill(cb
, "under_oom", 0);
4221 static int mem_cgroup_oom_control_write(struct cgroup
*cgrp
,
4222 struct cftype
*cft
, u64 val
)
4224 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
4225 struct mem_cgroup
*parent
;
4227 /* cannot set to root cgroup and only 0 and 1 are allowed */
4228 if (!cgrp
->parent
|| !((val
== 0) || (val
== 1)))
4231 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4234 /* oom-kill-disable is a flag for subhierarchy. */
4235 if ((parent
->use_hierarchy
) ||
4236 (mem
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4240 mem
->oom_kill_disable
= val
;
4242 memcg_oom_recover(mem
);
4247 static struct cftype mem_cgroup_files
[] = {
4249 .name
= "usage_in_bytes",
4250 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
4251 .read_u64
= mem_cgroup_read
,
4252 .register_event
= mem_cgroup_usage_register_event
,
4253 .unregister_event
= mem_cgroup_usage_unregister_event
,
4256 .name
= "max_usage_in_bytes",
4257 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
4258 .trigger
= mem_cgroup_reset
,
4259 .read_u64
= mem_cgroup_read
,
4262 .name
= "limit_in_bytes",
4263 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
4264 .write_string
= mem_cgroup_write
,
4265 .read_u64
= mem_cgroup_read
,
4268 .name
= "soft_limit_in_bytes",
4269 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
4270 .write_string
= mem_cgroup_write
,
4271 .read_u64
= mem_cgroup_read
,
4275 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
4276 .trigger
= mem_cgroup_reset
,
4277 .read_u64
= mem_cgroup_read
,
4281 .read_map
= mem_control_stat_show
,
4284 .name
= "force_empty",
4285 .trigger
= mem_cgroup_force_empty_write
,
4288 .name
= "use_hierarchy",
4289 .write_u64
= mem_cgroup_hierarchy_write
,
4290 .read_u64
= mem_cgroup_hierarchy_read
,
4293 .name
= "swappiness",
4294 .read_u64
= mem_cgroup_swappiness_read
,
4295 .write_u64
= mem_cgroup_swappiness_write
,
4298 .name
= "move_charge_at_immigrate",
4299 .read_u64
= mem_cgroup_move_charge_read
,
4300 .write_u64
= mem_cgroup_move_charge_write
,
4303 .name
= "oom_control",
4304 .read_map
= mem_cgroup_oom_control_read
,
4305 .write_u64
= mem_cgroup_oom_control_write
,
4306 .register_event
= mem_cgroup_oom_register_event
,
4307 .unregister_event
= mem_cgroup_oom_unregister_event
,
4308 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
4312 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4313 static struct cftype memsw_cgroup_files
[] = {
4315 .name
= "memsw.usage_in_bytes",
4316 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
4317 .read_u64
= mem_cgroup_read
,
4318 .register_event
= mem_cgroup_usage_register_event
,
4319 .unregister_event
= mem_cgroup_usage_unregister_event
,
4322 .name
= "memsw.max_usage_in_bytes",
4323 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
4324 .trigger
= mem_cgroup_reset
,
4325 .read_u64
= mem_cgroup_read
,
4328 .name
= "memsw.limit_in_bytes",
4329 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
4330 .write_string
= mem_cgroup_write
,
4331 .read_u64
= mem_cgroup_read
,
4334 .name
= "memsw.failcnt",
4335 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
4336 .trigger
= mem_cgroup_reset
,
4337 .read_u64
= mem_cgroup_read
,
4341 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4343 if (!do_swap_account
)
4345 return cgroup_add_files(cont
, ss
, memsw_cgroup_files
,
4346 ARRAY_SIZE(memsw_cgroup_files
));
4349 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4355 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*mem
, int node
)
4357 struct mem_cgroup_per_node
*pn
;
4358 struct mem_cgroup_per_zone
*mz
;
4360 int zone
, tmp
= node
;
4362 * This routine is called against possible nodes.
4363 * But it's BUG to call kmalloc() against offline node.
4365 * TODO: this routine can waste much memory for nodes which will
4366 * never be onlined. It's better to use memory hotplug callback
4369 if (!node_state(node
, N_NORMAL_MEMORY
))
4371 pn
= kzalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
4375 mem
->info
.nodeinfo
[node
] = pn
;
4376 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4377 mz
= &pn
->zoneinfo
[zone
];
4379 INIT_LIST_HEAD(&mz
->lists
[l
]);
4380 mz
->usage_in_excess
= 0;
4381 mz
->on_tree
= false;
4387 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*mem
, int node
)
4389 kfree(mem
->info
.nodeinfo
[node
]);
4392 static struct mem_cgroup
*mem_cgroup_alloc(void)
4394 struct mem_cgroup
*mem
;
4395 int size
= sizeof(struct mem_cgroup
);
4397 /* Can be very big if MAX_NUMNODES is very big */
4398 if (size
< PAGE_SIZE
)
4399 mem
= kzalloc(size
, GFP_KERNEL
);
4401 mem
= vzalloc(size
);
4406 mem
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4409 spin_lock_init(&mem
->pcp_counter_lock
);
4413 if (size
< PAGE_SIZE
)
4421 * At destroying mem_cgroup, references from swap_cgroup can remain.
4422 * (scanning all at force_empty is too costly...)
4424 * Instead of clearing all references at force_empty, we remember
4425 * the number of reference from swap_cgroup and free mem_cgroup when
4426 * it goes down to 0.
4428 * Removal of cgroup itself succeeds regardless of refs from swap.
4431 static void __mem_cgroup_free(struct mem_cgroup
*mem
)
4435 mem_cgroup_remove_from_trees(mem
);
4436 free_css_id(&mem_cgroup_subsys
, &mem
->css
);
4438 for_each_node_state(node
, N_POSSIBLE
)
4439 free_mem_cgroup_per_zone_info(mem
, node
);
4441 free_percpu(mem
->stat
);
4442 if (sizeof(struct mem_cgroup
) < PAGE_SIZE
)
4448 static void mem_cgroup_get(struct mem_cgroup
*mem
)
4450 atomic_inc(&mem
->refcnt
);
4453 static void __mem_cgroup_put(struct mem_cgroup
*mem
, int count
)
4455 if (atomic_sub_and_test(count
, &mem
->refcnt
)) {
4456 struct mem_cgroup
*parent
= parent_mem_cgroup(mem
);
4457 __mem_cgroup_free(mem
);
4459 mem_cgroup_put(parent
);
4463 static void mem_cgroup_put(struct mem_cgroup
*mem
)
4465 __mem_cgroup_put(mem
, 1);
4469 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4471 static struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*mem
)
4473 if (!mem
->res
.parent
)
4475 return mem_cgroup_from_res_counter(mem
->res
.parent
, res
);
4478 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4479 static void __init
enable_swap_cgroup(void)
4481 if (!mem_cgroup_disabled() && really_do_swap_account
)
4482 do_swap_account
= 1;
4485 static void __init
enable_swap_cgroup(void)
4490 static int mem_cgroup_soft_limit_tree_init(void)
4492 struct mem_cgroup_tree_per_node
*rtpn
;
4493 struct mem_cgroup_tree_per_zone
*rtpz
;
4494 int tmp
, node
, zone
;
4496 for_each_node_state(node
, N_POSSIBLE
) {
4498 if (!node_state(node
, N_NORMAL_MEMORY
))
4500 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
, tmp
);
4504 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
4506 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4507 rtpz
= &rtpn
->rb_tree_per_zone
[zone
];
4508 rtpz
->rb_root
= RB_ROOT
;
4509 spin_lock_init(&rtpz
->lock
);
4515 static struct cgroup_subsys_state
* __ref
4516 mem_cgroup_create(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
4518 struct mem_cgroup
*mem
, *parent
;
4519 long error
= -ENOMEM
;
4522 mem
= mem_cgroup_alloc();
4524 return ERR_PTR(error
);
4526 for_each_node_state(node
, N_POSSIBLE
)
4527 if (alloc_mem_cgroup_per_zone_info(mem
, node
))
4531 if (cont
->parent
== NULL
) {
4533 enable_swap_cgroup();
4535 root_mem_cgroup
= mem
;
4536 if (mem_cgroup_soft_limit_tree_init())
4538 for_each_possible_cpu(cpu
) {
4539 struct memcg_stock_pcp
*stock
=
4540 &per_cpu(memcg_stock
, cpu
);
4541 INIT_WORK(&stock
->work
, drain_local_stock
);
4543 hotcpu_notifier(memcg_cpu_hotplug_callback
, 0);
4545 parent
= mem_cgroup_from_cont(cont
->parent
);
4546 mem
->use_hierarchy
= parent
->use_hierarchy
;
4547 mem
->oom_kill_disable
= parent
->oom_kill_disable
;
4550 if (parent
&& parent
->use_hierarchy
) {
4551 res_counter_init(&mem
->res
, &parent
->res
);
4552 res_counter_init(&mem
->memsw
, &parent
->memsw
);
4554 * We increment refcnt of the parent to ensure that we can
4555 * safely access it on res_counter_charge/uncharge.
4556 * This refcnt will be decremented when freeing this
4557 * mem_cgroup(see mem_cgroup_put).
4559 mem_cgroup_get(parent
);
4561 res_counter_init(&mem
->res
, NULL
);
4562 res_counter_init(&mem
->memsw
, NULL
);
4564 mem
->last_scanned_child
= 0;
4565 INIT_LIST_HEAD(&mem
->oom_notify
);
4568 mem
->swappiness
= get_swappiness(parent
);
4569 atomic_set(&mem
->refcnt
, 1);
4570 mem
->move_charge_at_immigrate
= 0;
4571 mutex_init(&mem
->thresholds_lock
);
4574 __mem_cgroup_free(mem
);
4575 root_mem_cgroup
= NULL
;
4576 return ERR_PTR(error
);
4579 static int mem_cgroup_pre_destroy(struct cgroup_subsys
*ss
,
4580 struct cgroup
*cont
)
4582 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
4584 return mem_cgroup_force_empty(mem
, false);
4587 static void mem_cgroup_destroy(struct cgroup_subsys
*ss
,
4588 struct cgroup
*cont
)
4590 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
4592 mem_cgroup_put(mem
);
4595 static int mem_cgroup_populate(struct cgroup_subsys
*ss
,
4596 struct cgroup
*cont
)
4600 ret
= cgroup_add_files(cont
, ss
, mem_cgroup_files
,
4601 ARRAY_SIZE(mem_cgroup_files
));
4604 ret
= register_memsw_files(cont
, ss
);
4609 /* Handlers for move charge at task migration. */
4610 #define PRECHARGE_COUNT_AT_ONCE 256
4611 static int mem_cgroup_do_precharge(unsigned long count
)
4614 int batch_count
= PRECHARGE_COUNT_AT_ONCE
;
4615 struct mem_cgroup
*mem
= mc
.to
;
4617 if (mem_cgroup_is_root(mem
)) {
4618 mc
.precharge
+= count
;
4619 /* we don't need css_get for root */
4622 /* try to charge at once */
4624 struct res_counter
*dummy
;
4626 * "mem" cannot be under rmdir() because we've already checked
4627 * by cgroup_lock_live_cgroup() that it is not removed and we
4628 * are still under the same cgroup_mutex. So we can postpone
4631 if (res_counter_charge(&mem
->res
, PAGE_SIZE
* count
, &dummy
))
4633 if (do_swap_account
&& res_counter_charge(&mem
->memsw
,
4634 PAGE_SIZE
* count
, &dummy
)) {
4635 res_counter_uncharge(&mem
->res
, PAGE_SIZE
* count
);
4638 mc
.precharge
+= count
;
4642 /* fall back to one by one charge */
4644 if (signal_pending(current
)) {
4648 if (!batch_count
--) {
4649 batch_count
= PRECHARGE_COUNT_AT_ONCE
;
4652 ret
= __mem_cgroup_try_charge(NULL
, GFP_KERNEL
, 1, &mem
, false);
4654 /* mem_cgroup_clear_mc() will do uncharge later */
4662 * is_target_pte_for_mc - check a pte whether it is valid for move charge
4663 * @vma: the vma the pte to be checked belongs
4664 * @addr: the address corresponding to the pte to be checked
4665 * @ptent: the pte to be checked
4666 * @target: the pointer the target page or swap ent will be stored(can be NULL)
4669 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4670 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4671 * move charge. if @target is not NULL, the page is stored in target->page
4672 * with extra refcnt got(Callers should handle it).
4673 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4674 * target for charge migration. if @target is not NULL, the entry is stored
4677 * Called with pte lock held.
4684 enum mc_target_type
{
4685 MC_TARGET_NONE
, /* not used */
4690 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
4691 unsigned long addr
, pte_t ptent
)
4693 struct page
*page
= vm_normal_page(vma
, addr
, ptent
);
4695 if (!page
|| !page_mapped(page
))
4697 if (PageAnon(page
)) {
4698 /* we don't move shared anon */
4699 if (!move_anon() || page_mapcount(page
) > 2)
4701 } else if (!move_file())
4702 /* we ignore mapcount for file pages */
4704 if (!get_page_unless_zero(page
))
4710 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
4711 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
4714 struct page
*page
= NULL
;
4715 swp_entry_t ent
= pte_to_swp_entry(ptent
);
4717 if (!move_anon() || non_swap_entry(ent
))
4719 usage_count
= mem_cgroup_count_swap_user(ent
, &page
);
4720 if (usage_count
> 1) { /* we don't move shared anon */
4725 if (do_swap_account
)
4726 entry
->val
= ent
.val
;
4731 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
4732 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
4734 struct page
*page
= NULL
;
4735 struct inode
*inode
;
4736 struct address_space
*mapping
;
4739 if (!vma
->vm_file
) /* anonymous vma */
4744 inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
4745 mapping
= vma
->vm_file
->f_mapping
;
4746 if (pte_none(ptent
))
4747 pgoff
= linear_page_index(vma
, addr
);
4748 else /* pte_file(ptent) is true */
4749 pgoff
= pte_to_pgoff(ptent
);
4751 /* page is moved even if it's not RSS of this task(page-faulted). */
4752 if (!mapping_cap_swap_backed(mapping
)) { /* normal file */
4753 page
= find_get_page(mapping
, pgoff
);
4754 } else { /* shmem/tmpfs file. we should take account of swap too. */
4756 mem_cgroup_get_shmem_target(inode
, pgoff
, &page
, &ent
);
4757 if (do_swap_account
)
4758 entry
->val
= ent
.val
;
4764 static int is_target_pte_for_mc(struct vm_area_struct
*vma
,
4765 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
4767 struct page
*page
= NULL
;
4768 struct page_cgroup
*pc
;
4770 swp_entry_t ent
= { .val
= 0 };
4772 if (pte_present(ptent
))
4773 page
= mc_handle_present_pte(vma
, addr
, ptent
);
4774 else if (is_swap_pte(ptent
))
4775 page
= mc_handle_swap_pte(vma
, addr
, ptent
, &ent
);
4776 else if (pte_none(ptent
) || pte_file(ptent
))
4777 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
4779 if (!page
&& !ent
.val
)
4782 pc
= lookup_page_cgroup(page
);
4784 * Do only loose check w/o page_cgroup lock.
4785 * mem_cgroup_move_account() checks the pc is valid or not under
4788 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
4789 ret
= MC_TARGET_PAGE
;
4791 target
->page
= page
;
4793 if (!ret
|| !target
)
4796 /* There is a swap entry and a page doesn't exist or isn't charged */
4797 if (ent
.val
&& !ret
&&
4798 css_id(&mc
.from
->css
) == lookup_swap_cgroup(ent
)) {
4799 ret
= MC_TARGET_SWAP
;
4806 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
4807 unsigned long addr
, unsigned long end
,
4808 struct mm_walk
*walk
)
4810 struct vm_area_struct
*vma
= walk
->private;
4814 split_huge_page_pmd(walk
->mm
, pmd
);
4816 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
4817 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
4818 if (is_target_pte_for_mc(vma
, addr
, *pte
, NULL
))
4819 mc
.precharge
++; /* increment precharge temporarily */
4820 pte_unmap_unlock(pte
- 1, ptl
);
4826 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
4828 unsigned long precharge
;
4829 struct vm_area_struct
*vma
;
4831 down_read(&mm
->mmap_sem
);
4832 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
4833 struct mm_walk mem_cgroup_count_precharge_walk
= {
4834 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
4838 if (is_vm_hugetlb_page(vma
))
4840 walk_page_range(vma
->vm_start
, vma
->vm_end
,
4841 &mem_cgroup_count_precharge_walk
);
4843 up_read(&mm
->mmap_sem
);
4845 precharge
= mc
.precharge
;
4851 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
4853 unsigned long precharge
= mem_cgroup_count_precharge(mm
);
4855 VM_BUG_ON(mc
.moving_task
);
4856 mc
.moving_task
= current
;
4857 return mem_cgroup_do_precharge(precharge
);
4860 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4861 static void __mem_cgroup_clear_mc(void)
4863 struct mem_cgroup
*from
= mc
.from
;
4864 struct mem_cgroup
*to
= mc
.to
;
4866 /* we must uncharge all the leftover precharges from mc.to */
4868 __mem_cgroup_cancel_charge(mc
.to
, mc
.precharge
);
4872 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4873 * we must uncharge here.
4875 if (mc
.moved_charge
) {
4876 __mem_cgroup_cancel_charge(mc
.from
, mc
.moved_charge
);
4877 mc
.moved_charge
= 0;
4879 /* we must fixup refcnts and charges */
4880 if (mc
.moved_swap
) {
4881 /* uncharge swap account from the old cgroup */
4882 if (!mem_cgroup_is_root(mc
.from
))
4883 res_counter_uncharge(&mc
.from
->memsw
,
4884 PAGE_SIZE
* mc
.moved_swap
);
4885 __mem_cgroup_put(mc
.from
, mc
.moved_swap
);
4887 if (!mem_cgroup_is_root(mc
.to
)) {
4889 * we charged both to->res and to->memsw, so we should
4892 res_counter_uncharge(&mc
.to
->res
,
4893 PAGE_SIZE
* mc
.moved_swap
);
4895 /* we've already done mem_cgroup_get(mc.to) */
4898 memcg_oom_recover(from
);
4899 memcg_oom_recover(to
);
4900 wake_up_all(&mc
.waitq
);
4903 static void mem_cgroup_clear_mc(void)
4905 struct mem_cgroup
*from
= mc
.from
;
4908 * we must clear moving_task before waking up waiters at the end of
4911 mc
.moving_task
= NULL
;
4912 __mem_cgroup_clear_mc();
4913 spin_lock(&mc
.lock
);
4916 spin_unlock(&mc
.lock
);
4917 mem_cgroup_end_move(from
);
4920 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
4921 struct cgroup
*cgroup
,
4922 struct task_struct
*p
,
4926 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgroup
);
4928 if (mem
->move_charge_at_immigrate
) {
4929 struct mm_struct
*mm
;
4930 struct mem_cgroup
*from
= mem_cgroup_from_task(p
);
4932 VM_BUG_ON(from
== mem
);
4934 mm
= get_task_mm(p
);
4937 /* We move charges only when we move a owner of the mm */
4938 if (mm
->owner
== p
) {
4941 VM_BUG_ON(mc
.precharge
);
4942 VM_BUG_ON(mc
.moved_charge
);
4943 VM_BUG_ON(mc
.moved_swap
);
4944 mem_cgroup_start_move(from
);
4945 spin_lock(&mc
.lock
);
4948 spin_unlock(&mc
.lock
);
4949 /* We set mc.moving_task later */
4951 ret
= mem_cgroup_precharge_mc(mm
);
4953 mem_cgroup_clear_mc();
4960 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
4961 struct cgroup
*cgroup
,
4962 struct task_struct
*p
,
4965 mem_cgroup_clear_mc();
4968 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
4969 unsigned long addr
, unsigned long end
,
4970 struct mm_walk
*walk
)
4973 struct vm_area_struct
*vma
= walk
->private;
4977 split_huge_page_pmd(walk
->mm
, pmd
);
4979 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
4980 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
4981 pte_t ptent
= *(pte
++);
4982 union mc_target target
;
4985 struct page_cgroup
*pc
;
4991 type
= is_target_pte_for_mc(vma
, addr
, ptent
, &target
);
4993 case MC_TARGET_PAGE
:
4995 if (isolate_lru_page(page
))
4997 pc
= lookup_page_cgroup(page
);
4998 if (!mem_cgroup_move_account(page
, 1, pc
,
4999 mc
.from
, mc
.to
, false)) {
5001 /* we uncharge from mc.from later. */
5004 putback_lru_page(page
);
5005 put
: /* is_target_pte_for_mc() gets the page */
5008 case MC_TARGET_SWAP
:
5010 if (!mem_cgroup_move_swap_account(ent
,
5011 mc
.from
, mc
.to
, false)) {
5013 /* we fixup refcnts and charges later. */
5021 pte_unmap_unlock(pte
- 1, ptl
);
5026 * We have consumed all precharges we got in can_attach().
5027 * We try charge one by one, but don't do any additional
5028 * charges to mc.to if we have failed in charge once in attach()
5031 ret
= mem_cgroup_do_precharge(1);
5039 static void mem_cgroup_move_charge(struct mm_struct
*mm
)
5041 struct vm_area_struct
*vma
;
5043 lru_add_drain_all();
5045 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
5047 * Someone who are holding the mmap_sem might be waiting in
5048 * waitq. So we cancel all extra charges, wake up all waiters,
5049 * and retry. Because we cancel precharges, we might not be able
5050 * to move enough charges, but moving charge is a best-effort
5051 * feature anyway, so it wouldn't be a big problem.
5053 __mem_cgroup_clear_mc();
5057 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5059 struct mm_walk mem_cgroup_move_charge_walk
= {
5060 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
5064 if (is_vm_hugetlb_page(vma
))
5066 ret
= walk_page_range(vma
->vm_start
, vma
->vm_end
,
5067 &mem_cgroup_move_charge_walk
);
5070 * means we have consumed all precharges and failed in
5071 * doing additional charge. Just abandon here.
5075 up_read(&mm
->mmap_sem
);
5078 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
5079 struct cgroup
*cont
,
5080 struct cgroup
*old_cont
,
5081 struct task_struct
*p
,
5084 struct mm_struct
*mm
;
5087 /* no need to move charge */
5090 mm
= get_task_mm(p
);
5092 mem_cgroup_move_charge(mm
);
5095 mem_cgroup_clear_mc();
5097 #else /* !CONFIG_MMU */
5098 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
5099 struct cgroup
*cgroup
,
5100 struct task_struct
*p
,
5105 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
5106 struct cgroup
*cgroup
,
5107 struct task_struct
*p
,
5111 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
5112 struct cgroup
*cont
,
5113 struct cgroup
*old_cont
,
5114 struct task_struct
*p
,
5120 struct cgroup_subsys mem_cgroup_subsys
= {
5122 .subsys_id
= mem_cgroup_subsys_id
,
5123 .create
= mem_cgroup_create
,
5124 .pre_destroy
= mem_cgroup_pre_destroy
,
5125 .destroy
= mem_cgroup_destroy
,
5126 .populate
= mem_cgroup_populate
,
5127 .can_attach
= mem_cgroup_can_attach
,
5128 .cancel_attach
= mem_cgroup_cancel_attach
,
5129 .attach
= mem_cgroup_move_task
,
5134 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
5135 static int __init
enable_swap_account(char *s
)
5137 /* consider enabled if no parameter or 1 is given */
5138 if (!(*s
) || !strcmp(s
, "=1"))
5139 really_do_swap_account
= 1;
5140 else if (!strcmp(s
, "=0"))
5141 really_do_swap_account
= 0;
5144 __setup("swapaccount", enable_swap_account
);
5146 static int __init
disable_swap_account(char *s
)
5148 printk_once("noswapaccount is deprecated and will be removed in 2.6.40. Use swapaccount=0 instead\n");
5149 enable_swap_account("=0");
5152 __setup("noswapaccount", disable_swap_account
);