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/export.h>
37 #include <linux/mutex.h>
38 #include <linux/rbtree.h>
39 #include <linux/slab.h>
40 #include <linux/swap.h>
41 #include <linux/swapops.h>
42 #include <linux/spinlock.h>
43 #include <linux/eventfd.h>
44 #include <linux/sort.h>
46 #include <linux/seq_file.h>
47 #include <linux/vmalloc.h>
48 #include <linux/mm_inline.h>
49 #include <linux/page_cgroup.h>
50 #include <linux/cpu.h>
51 #include <linux/oom.h>
55 #include <net/tcp_memcontrol.h>
57 #include <asm/uaccess.h>
59 #include <trace/events/vmscan.h>
61 struct cgroup_subsys mem_cgroup_subsys __read_mostly
;
62 #define MEM_CGROUP_RECLAIM_RETRIES 5
63 static struct mem_cgroup
*root_mem_cgroup __read_mostly
;
65 #ifdef CONFIG_MEMCG_SWAP
66 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
67 int do_swap_account __read_mostly
;
69 /* for remember boot option*/
70 #ifdef CONFIG_MEMCG_SWAP_ENABLED
71 static int really_do_swap_account __initdata
= 1;
73 static int really_do_swap_account __initdata
= 0;
77 #define do_swap_account 0
82 * Statistics for memory cgroup.
84 enum mem_cgroup_stat_index
{
86 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
88 MEM_CGROUP_STAT_CACHE
, /* # of pages charged as cache */
89 MEM_CGROUP_STAT_RSS
, /* # of pages charged as anon rss */
90 MEM_CGROUP_STAT_FILE_MAPPED
, /* # of pages charged as file rss */
91 MEM_CGROUP_STAT_SWAP
, /* # of pages, swapped out */
92 MEM_CGROUP_STAT_NSTATS
,
95 static const char * const mem_cgroup_stat_names
[] = {
102 enum mem_cgroup_events_index
{
103 MEM_CGROUP_EVENTS_PGPGIN
, /* # of pages paged in */
104 MEM_CGROUP_EVENTS_PGPGOUT
, /* # of pages paged out */
105 MEM_CGROUP_EVENTS_PGFAULT
, /* # of page-faults */
106 MEM_CGROUP_EVENTS_PGMAJFAULT
, /* # of major page-faults */
107 MEM_CGROUP_EVENTS_NSTATS
,
110 static const char * const mem_cgroup_events_names
[] = {
118 * Per memcg event counter is incremented at every pagein/pageout. With THP,
119 * it will be incremated by the number of pages. This counter is used for
120 * for trigger some periodic events. This is straightforward and better
121 * than using jiffies etc. to handle periodic memcg event.
123 enum mem_cgroup_events_target
{
124 MEM_CGROUP_TARGET_THRESH
,
125 MEM_CGROUP_TARGET_SOFTLIMIT
,
126 MEM_CGROUP_TARGET_NUMAINFO
,
129 #define THRESHOLDS_EVENTS_TARGET 128
130 #define SOFTLIMIT_EVENTS_TARGET 1024
131 #define NUMAINFO_EVENTS_TARGET 1024
133 struct mem_cgroup_stat_cpu
{
134 long count
[MEM_CGROUP_STAT_NSTATS
];
135 unsigned long events
[MEM_CGROUP_EVENTS_NSTATS
];
136 unsigned long nr_page_events
;
137 unsigned long targets
[MEM_CGROUP_NTARGETS
];
140 struct mem_cgroup_reclaim_iter
{
141 /* css_id of the last scanned hierarchy member */
143 /* scan generation, increased every round-trip */
144 unsigned int generation
;
148 * per-zone information in memory controller.
150 struct mem_cgroup_per_zone
{
151 struct lruvec lruvec
;
152 unsigned long lru_size
[NR_LRU_LISTS
];
154 struct mem_cgroup_reclaim_iter reclaim_iter
[DEF_PRIORITY
+ 1];
156 struct rb_node tree_node
; /* RB tree node */
157 unsigned long long usage_in_excess
;/* Set to the value by which */
158 /* the soft limit is exceeded*/
160 struct mem_cgroup
*memcg
; /* Back pointer, we cannot */
161 /* use container_of */
164 struct mem_cgroup_per_node
{
165 struct mem_cgroup_per_zone zoneinfo
[MAX_NR_ZONES
];
168 struct mem_cgroup_lru_info
{
169 struct mem_cgroup_per_node
*nodeinfo
[MAX_NUMNODES
];
173 * Cgroups above their limits are maintained in a RB-Tree, independent of
174 * their hierarchy representation
177 struct mem_cgroup_tree_per_zone
{
178 struct rb_root rb_root
;
182 struct mem_cgroup_tree_per_node
{
183 struct mem_cgroup_tree_per_zone rb_tree_per_zone
[MAX_NR_ZONES
];
186 struct mem_cgroup_tree
{
187 struct mem_cgroup_tree_per_node
*rb_tree_per_node
[MAX_NUMNODES
];
190 static struct mem_cgroup_tree soft_limit_tree __read_mostly
;
192 struct mem_cgroup_threshold
{
193 struct eventfd_ctx
*eventfd
;
198 struct mem_cgroup_threshold_ary
{
199 /* An array index points to threshold just below or equal to usage. */
200 int current_threshold
;
201 /* Size of entries[] */
203 /* Array of thresholds */
204 struct mem_cgroup_threshold entries
[0];
207 struct mem_cgroup_thresholds
{
208 /* Primary thresholds array */
209 struct mem_cgroup_threshold_ary
*primary
;
211 * Spare threshold array.
212 * This is needed to make mem_cgroup_unregister_event() "never fail".
213 * It must be able to store at least primary->size - 1 entries.
215 struct mem_cgroup_threshold_ary
*spare
;
219 struct mem_cgroup_eventfd_list
{
220 struct list_head list
;
221 struct eventfd_ctx
*eventfd
;
224 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
);
225 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
);
228 * The memory controller data structure. The memory controller controls both
229 * page cache and RSS per cgroup. We would eventually like to provide
230 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
231 * to help the administrator determine what knobs to tune.
233 * TODO: Add a water mark for the memory controller. Reclaim will begin when
234 * we hit the water mark. May be even add a low water mark, such that
235 * no reclaim occurs from a cgroup at it's low water mark, this is
236 * a feature that will be implemented much later in the future.
239 struct cgroup_subsys_state css
;
241 * the counter to account for memory usage
243 struct res_counter res
;
247 * the counter to account for mem+swap usage.
249 struct res_counter memsw
;
252 * rcu_freeing is used only when freeing struct mem_cgroup,
253 * so put it into a union to avoid wasting more memory.
254 * It must be disjoint from the css field. It could be
255 * in a union with the res field, but res plays a much
256 * larger part in mem_cgroup life than memsw, and might
257 * be of interest, even at time of free, when debugging.
258 * So share rcu_head with the less interesting memsw.
260 struct rcu_head rcu_freeing
;
262 * We also need some space for a worker in deferred freeing.
263 * By the time we call it, rcu_freeing is no longer in use.
265 struct work_struct work_freeing
;
269 * Per cgroup active and inactive list, similar to the
270 * per zone LRU lists.
272 struct mem_cgroup_lru_info info
;
273 int last_scanned_node
;
275 nodemask_t scan_nodes
;
276 atomic_t numainfo_events
;
277 atomic_t numainfo_updating
;
280 * Should the accounting and control be hierarchical, per subtree?
290 /* OOM-Killer disable */
291 int oom_kill_disable
;
293 /* set when res.limit == memsw.limit */
294 bool memsw_is_minimum
;
296 /* protect arrays of thresholds */
297 struct mutex thresholds_lock
;
299 /* thresholds for memory usage. RCU-protected */
300 struct mem_cgroup_thresholds thresholds
;
302 /* thresholds for mem+swap usage. RCU-protected */
303 struct mem_cgroup_thresholds memsw_thresholds
;
305 /* For oom notifier event fd */
306 struct list_head oom_notify
;
309 * Should we move charges of a task when a task is moved into this
310 * mem_cgroup ? And what type of charges should we move ?
312 unsigned long move_charge_at_immigrate
;
314 * set > 0 if pages under this cgroup are moving to other cgroup.
316 atomic_t moving_account
;
317 /* taken only while moving_account > 0 */
318 spinlock_t move_lock
;
322 struct mem_cgroup_stat_cpu __percpu
*stat
;
324 * used when a cpu is offlined or other synchronizations
325 * See mem_cgroup_read_stat().
327 struct mem_cgroup_stat_cpu nocpu_base
;
328 spinlock_t pcp_counter_lock
;
330 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET)
331 struct tcp_memcontrol tcp_mem
;
335 /* Stuffs for move charges at task migration. */
337 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
338 * left-shifted bitmap of these types.
341 MOVE_CHARGE_TYPE_ANON
, /* private anonymous page and swap of it */
342 MOVE_CHARGE_TYPE_FILE
, /* file page(including tmpfs) and swap of it */
346 /* "mc" and its members are protected by cgroup_mutex */
347 static struct move_charge_struct
{
348 spinlock_t lock
; /* for from, to */
349 struct mem_cgroup
*from
;
350 struct mem_cgroup
*to
;
351 unsigned long precharge
;
352 unsigned long moved_charge
;
353 unsigned long moved_swap
;
354 struct task_struct
*moving_task
; /* a task moving charges */
355 wait_queue_head_t waitq
; /* a waitq for other context */
357 .lock
= __SPIN_LOCK_UNLOCKED(mc
.lock
),
358 .waitq
= __WAIT_QUEUE_HEAD_INITIALIZER(mc
.waitq
),
361 static bool move_anon(void)
363 return test_bit(MOVE_CHARGE_TYPE_ANON
,
364 &mc
.to
->move_charge_at_immigrate
);
367 static bool move_file(void)
369 return test_bit(MOVE_CHARGE_TYPE_FILE
,
370 &mc
.to
->move_charge_at_immigrate
);
374 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
375 * limit reclaim to prevent infinite loops, if they ever occur.
377 #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
378 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
381 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
382 MEM_CGROUP_CHARGE_TYPE_ANON
,
383 MEM_CGROUP_CHARGE_TYPE_SWAPOUT
, /* for accounting swapcache */
384 MEM_CGROUP_CHARGE_TYPE_DROP
, /* a page was unused swap cache */
388 /* for encoding cft->private value on file */
391 #define _OOM_TYPE (2)
392 #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
393 #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
394 #define MEMFILE_ATTR(val) ((val) & 0xffff)
395 /* Used for OOM nofiier */
396 #define OOM_CONTROL (0)
399 * Reclaim flags for mem_cgroup_hierarchical_reclaim
401 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
402 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
403 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
404 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
406 static void mem_cgroup_get(struct mem_cgroup
*memcg
);
407 static void mem_cgroup_put(struct mem_cgroup
*memcg
);
410 struct mem_cgroup
*mem_cgroup_from_css(struct cgroup_subsys_state
*s
)
412 return container_of(s
, struct mem_cgroup
, css
);
415 static inline bool mem_cgroup_is_root(struct mem_cgroup
*memcg
)
417 return (memcg
== root_mem_cgroup
);
420 /* Writing them here to avoid exposing memcg's inner layout */
421 #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
423 void sock_update_memcg(struct sock
*sk
)
425 if (mem_cgroup_sockets_enabled
) {
426 struct mem_cgroup
*memcg
;
427 struct cg_proto
*cg_proto
;
429 BUG_ON(!sk
->sk_prot
->proto_cgroup
);
431 /* Socket cloning can throw us here with sk_cgrp already
432 * filled. It won't however, necessarily happen from
433 * process context. So the test for root memcg given
434 * the current task's memcg won't help us in this case.
436 * Respecting the original socket's memcg is a better
437 * decision in this case.
440 BUG_ON(mem_cgroup_is_root(sk
->sk_cgrp
->memcg
));
441 mem_cgroup_get(sk
->sk_cgrp
->memcg
);
446 memcg
= mem_cgroup_from_task(current
);
447 cg_proto
= sk
->sk_prot
->proto_cgroup(memcg
);
448 if (!mem_cgroup_is_root(memcg
) && memcg_proto_active(cg_proto
)) {
449 mem_cgroup_get(memcg
);
450 sk
->sk_cgrp
= cg_proto
;
455 EXPORT_SYMBOL(sock_update_memcg
);
457 void sock_release_memcg(struct sock
*sk
)
459 if (mem_cgroup_sockets_enabled
&& sk
->sk_cgrp
) {
460 struct mem_cgroup
*memcg
;
461 WARN_ON(!sk
->sk_cgrp
->memcg
);
462 memcg
= sk
->sk_cgrp
->memcg
;
463 mem_cgroup_put(memcg
);
467 struct cg_proto
*tcp_proto_cgroup(struct mem_cgroup
*memcg
)
469 if (!memcg
|| mem_cgroup_is_root(memcg
))
472 return &memcg
->tcp_mem
.cg_proto
;
474 EXPORT_SYMBOL(tcp_proto_cgroup
);
476 static void disarm_sock_keys(struct mem_cgroup
*memcg
)
478 if (!memcg_proto_activated(&memcg
->tcp_mem
.cg_proto
))
480 static_key_slow_dec(&memcg_socket_limit_enabled
);
483 static void disarm_sock_keys(struct mem_cgroup
*memcg
)
488 static void drain_all_stock_async(struct mem_cgroup
*memcg
);
490 static struct mem_cgroup_per_zone
*
491 mem_cgroup_zoneinfo(struct mem_cgroup
*memcg
, int nid
, int zid
)
493 return &memcg
->info
.nodeinfo
[nid
]->zoneinfo
[zid
];
496 struct cgroup_subsys_state
*mem_cgroup_css(struct mem_cgroup
*memcg
)
501 static struct mem_cgroup_per_zone
*
502 page_cgroup_zoneinfo(struct mem_cgroup
*memcg
, struct page
*page
)
504 int nid
= page_to_nid(page
);
505 int zid
= page_zonenum(page
);
507 return mem_cgroup_zoneinfo(memcg
, nid
, zid
);
510 static struct mem_cgroup_tree_per_zone
*
511 soft_limit_tree_node_zone(int nid
, int zid
)
513 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
516 static struct mem_cgroup_tree_per_zone
*
517 soft_limit_tree_from_page(struct page
*page
)
519 int nid
= page_to_nid(page
);
520 int zid
= page_zonenum(page
);
522 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
526 __mem_cgroup_insert_exceeded(struct mem_cgroup
*memcg
,
527 struct mem_cgroup_per_zone
*mz
,
528 struct mem_cgroup_tree_per_zone
*mctz
,
529 unsigned long long new_usage_in_excess
)
531 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
532 struct rb_node
*parent
= NULL
;
533 struct mem_cgroup_per_zone
*mz_node
;
538 mz
->usage_in_excess
= new_usage_in_excess
;
539 if (!mz
->usage_in_excess
)
543 mz_node
= rb_entry(parent
, struct mem_cgroup_per_zone
,
545 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
)
548 * We can't avoid mem cgroups that are over their soft
549 * limit by the same amount
551 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
554 rb_link_node(&mz
->tree_node
, parent
, p
);
555 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
560 __mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
561 struct mem_cgroup_per_zone
*mz
,
562 struct mem_cgroup_tree_per_zone
*mctz
)
566 rb_erase(&mz
->tree_node
, &mctz
->rb_root
);
571 mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
572 struct mem_cgroup_per_zone
*mz
,
573 struct mem_cgroup_tree_per_zone
*mctz
)
575 spin_lock(&mctz
->lock
);
576 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
577 spin_unlock(&mctz
->lock
);
581 static void mem_cgroup_update_tree(struct mem_cgroup
*memcg
, struct page
*page
)
583 unsigned long long excess
;
584 struct mem_cgroup_per_zone
*mz
;
585 struct mem_cgroup_tree_per_zone
*mctz
;
586 int nid
= page_to_nid(page
);
587 int zid
= page_zonenum(page
);
588 mctz
= soft_limit_tree_from_page(page
);
591 * Necessary to update all ancestors when hierarchy is used.
592 * because their event counter is not touched.
594 for (; memcg
; memcg
= parent_mem_cgroup(memcg
)) {
595 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
596 excess
= res_counter_soft_limit_excess(&memcg
->res
);
598 * We have to update the tree if mz is on RB-tree or
599 * mem is over its softlimit.
601 if (excess
|| mz
->on_tree
) {
602 spin_lock(&mctz
->lock
);
603 /* if on-tree, remove it */
605 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
607 * Insert again. mz->usage_in_excess will be updated.
608 * If excess is 0, no tree ops.
610 __mem_cgroup_insert_exceeded(memcg
, mz
, mctz
, excess
);
611 spin_unlock(&mctz
->lock
);
616 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*memcg
)
619 struct mem_cgroup_per_zone
*mz
;
620 struct mem_cgroup_tree_per_zone
*mctz
;
622 for_each_node(node
) {
623 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
624 mz
= mem_cgroup_zoneinfo(memcg
, node
, zone
);
625 mctz
= soft_limit_tree_node_zone(node
, zone
);
626 mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
631 static struct mem_cgroup_per_zone
*
632 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
634 struct rb_node
*rightmost
= NULL
;
635 struct mem_cgroup_per_zone
*mz
;
639 rightmost
= rb_last(&mctz
->rb_root
);
641 goto done
; /* Nothing to reclaim from */
643 mz
= rb_entry(rightmost
, struct mem_cgroup_per_zone
, tree_node
);
645 * Remove the node now but someone else can add it back,
646 * we will to add it back at the end of reclaim to its correct
647 * position in the tree.
649 __mem_cgroup_remove_exceeded(mz
->memcg
, mz
, mctz
);
650 if (!res_counter_soft_limit_excess(&mz
->memcg
->res
) ||
651 !css_tryget(&mz
->memcg
->css
))
657 static struct mem_cgroup_per_zone
*
658 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
660 struct mem_cgroup_per_zone
*mz
;
662 spin_lock(&mctz
->lock
);
663 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
664 spin_unlock(&mctz
->lock
);
669 * Implementation Note: reading percpu statistics for memcg.
671 * Both of vmstat[] and percpu_counter has threshold and do periodic
672 * synchronization to implement "quick" read. There are trade-off between
673 * reading cost and precision of value. Then, we may have a chance to implement
674 * a periodic synchronizion of counter in memcg's counter.
676 * But this _read() function is used for user interface now. The user accounts
677 * memory usage by memory cgroup and he _always_ requires exact value because
678 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
679 * have to visit all online cpus and make sum. So, for now, unnecessary
680 * synchronization is not implemented. (just implemented for cpu hotplug)
682 * If there are kernel internal actions which can make use of some not-exact
683 * value, and reading all cpu value can be performance bottleneck in some
684 * common workload, threashold and synchonization as vmstat[] should be
687 static long mem_cgroup_read_stat(struct mem_cgroup
*memcg
,
688 enum mem_cgroup_stat_index idx
)
694 for_each_online_cpu(cpu
)
695 val
+= per_cpu(memcg
->stat
->count
[idx
], cpu
);
696 #ifdef CONFIG_HOTPLUG_CPU
697 spin_lock(&memcg
->pcp_counter_lock
);
698 val
+= memcg
->nocpu_base
.count
[idx
];
699 spin_unlock(&memcg
->pcp_counter_lock
);
705 static void mem_cgroup_swap_statistics(struct mem_cgroup
*memcg
,
708 int val
= (charge
) ? 1 : -1;
709 this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_SWAP
], val
);
712 static unsigned long mem_cgroup_read_events(struct mem_cgroup
*memcg
,
713 enum mem_cgroup_events_index idx
)
715 unsigned long val
= 0;
718 for_each_online_cpu(cpu
)
719 val
+= per_cpu(memcg
->stat
->events
[idx
], cpu
);
720 #ifdef CONFIG_HOTPLUG_CPU
721 spin_lock(&memcg
->pcp_counter_lock
);
722 val
+= memcg
->nocpu_base
.events
[idx
];
723 spin_unlock(&memcg
->pcp_counter_lock
);
728 static void mem_cgroup_charge_statistics(struct mem_cgroup
*memcg
,
729 bool anon
, int nr_pages
)
734 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
735 * counted as CACHE even if it's on ANON LRU.
738 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_RSS
],
741 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_CACHE
],
744 /* pagein of a big page is an event. So, ignore page size */
746 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGIN
]);
748 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGOUT
]);
749 nr_pages
= -nr_pages
; /* for event */
752 __this_cpu_add(memcg
->stat
->nr_page_events
, nr_pages
);
758 mem_cgroup_get_lru_size(struct lruvec
*lruvec
, enum lru_list lru
)
760 struct mem_cgroup_per_zone
*mz
;
762 mz
= container_of(lruvec
, struct mem_cgroup_per_zone
, lruvec
);
763 return mz
->lru_size
[lru
];
767 mem_cgroup_zone_nr_lru_pages(struct mem_cgroup
*memcg
, int nid
, int zid
,
768 unsigned int lru_mask
)
770 struct mem_cgroup_per_zone
*mz
;
772 unsigned long ret
= 0;
774 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
777 if (BIT(lru
) & lru_mask
)
778 ret
+= mz
->lru_size
[lru
];
784 mem_cgroup_node_nr_lru_pages(struct mem_cgroup
*memcg
,
785 int nid
, unsigned int lru_mask
)
790 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++)
791 total
+= mem_cgroup_zone_nr_lru_pages(memcg
,
797 static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup
*memcg
,
798 unsigned int lru_mask
)
803 for_each_node_state(nid
, N_HIGH_MEMORY
)
804 total
+= mem_cgroup_node_nr_lru_pages(memcg
, nid
, lru_mask
);
808 static bool mem_cgroup_event_ratelimit(struct mem_cgroup
*memcg
,
809 enum mem_cgroup_events_target target
)
811 unsigned long val
, next
;
813 val
= __this_cpu_read(memcg
->stat
->nr_page_events
);
814 next
= __this_cpu_read(memcg
->stat
->targets
[target
]);
815 /* from time_after() in jiffies.h */
816 if ((long)next
- (long)val
< 0) {
818 case MEM_CGROUP_TARGET_THRESH
:
819 next
= val
+ THRESHOLDS_EVENTS_TARGET
;
821 case MEM_CGROUP_TARGET_SOFTLIMIT
:
822 next
= val
+ SOFTLIMIT_EVENTS_TARGET
;
824 case MEM_CGROUP_TARGET_NUMAINFO
:
825 next
= val
+ NUMAINFO_EVENTS_TARGET
;
830 __this_cpu_write(memcg
->stat
->targets
[target
], next
);
837 * Check events in order.
840 static void memcg_check_events(struct mem_cgroup
*memcg
, struct page
*page
)
843 /* threshold event is triggered in finer grain than soft limit */
844 if (unlikely(mem_cgroup_event_ratelimit(memcg
,
845 MEM_CGROUP_TARGET_THRESH
))) {
847 bool do_numainfo __maybe_unused
;
849 do_softlimit
= mem_cgroup_event_ratelimit(memcg
,
850 MEM_CGROUP_TARGET_SOFTLIMIT
);
852 do_numainfo
= mem_cgroup_event_ratelimit(memcg
,
853 MEM_CGROUP_TARGET_NUMAINFO
);
857 mem_cgroup_threshold(memcg
);
858 if (unlikely(do_softlimit
))
859 mem_cgroup_update_tree(memcg
, page
);
861 if (unlikely(do_numainfo
))
862 atomic_inc(&memcg
->numainfo_events
);
868 struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
870 return mem_cgroup_from_css(
871 cgroup_subsys_state(cont
, mem_cgroup_subsys_id
));
874 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
877 * mm_update_next_owner() may clear mm->owner to NULL
878 * if it races with swapoff, page migration, etc.
879 * So this can be called with p == NULL.
884 return mem_cgroup_from_css(task_subsys_state(p
, mem_cgroup_subsys_id
));
887 struct mem_cgroup
*try_get_mem_cgroup_from_mm(struct mm_struct
*mm
)
889 struct mem_cgroup
*memcg
= NULL
;
894 * Because we have no locks, mm->owner's may be being moved to other
895 * cgroup. We use css_tryget() here even if this looks
896 * pessimistic (rather than adding locks here).
900 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
901 if (unlikely(!memcg
))
903 } while (!css_tryget(&memcg
->css
));
909 * mem_cgroup_iter - iterate over memory cgroup hierarchy
910 * @root: hierarchy root
911 * @prev: previously returned memcg, NULL on first invocation
912 * @reclaim: cookie for shared reclaim walks, NULL for full walks
914 * Returns references to children of the hierarchy below @root, or
915 * @root itself, or %NULL after a full round-trip.
917 * Caller must pass the return value in @prev on subsequent
918 * invocations for reference counting, or use mem_cgroup_iter_break()
919 * to cancel a hierarchy walk before the round-trip is complete.
921 * Reclaimers can specify a zone and a priority level in @reclaim to
922 * divide up the memcgs in the hierarchy among all concurrent
923 * reclaimers operating on the same zone and priority.
925 struct mem_cgroup
*mem_cgroup_iter(struct mem_cgroup
*root
,
926 struct mem_cgroup
*prev
,
927 struct mem_cgroup_reclaim_cookie
*reclaim
)
929 struct mem_cgroup
*memcg
= NULL
;
932 if (mem_cgroup_disabled())
936 root
= root_mem_cgroup
;
938 if (prev
&& !reclaim
)
939 id
= css_id(&prev
->css
);
941 if (prev
&& prev
!= root
)
944 if (!root
->use_hierarchy
&& root
!= root_mem_cgroup
) {
951 struct mem_cgroup_reclaim_iter
*uninitialized_var(iter
);
952 struct cgroup_subsys_state
*css
;
955 int nid
= zone_to_nid(reclaim
->zone
);
956 int zid
= zone_idx(reclaim
->zone
);
957 struct mem_cgroup_per_zone
*mz
;
959 mz
= mem_cgroup_zoneinfo(root
, nid
, zid
);
960 iter
= &mz
->reclaim_iter
[reclaim
->priority
];
961 if (prev
&& reclaim
->generation
!= iter
->generation
)
967 css
= css_get_next(&mem_cgroup_subsys
, id
+ 1, &root
->css
, &id
);
969 if (css
== &root
->css
|| css_tryget(css
))
970 memcg
= mem_cgroup_from_css(css
);
979 else if (!prev
&& memcg
)
980 reclaim
->generation
= iter
->generation
;
990 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
991 * @root: hierarchy root
992 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
994 void mem_cgroup_iter_break(struct mem_cgroup
*root
,
995 struct mem_cgroup
*prev
)
998 root
= root_mem_cgroup
;
999 if (prev
&& prev
!= root
)
1000 css_put(&prev
->css
);
1004 * Iteration constructs for visiting all cgroups (under a tree). If
1005 * loops are exited prematurely (break), mem_cgroup_iter_break() must
1006 * be used for reference counting.
1008 #define for_each_mem_cgroup_tree(iter, root) \
1009 for (iter = mem_cgroup_iter(root, NULL, NULL); \
1011 iter = mem_cgroup_iter(root, iter, NULL))
1013 #define for_each_mem_cgroup(iter) \
1014 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
1016 iter = mem_cgroup_iter(NULL, iter, NULL))
1018 void mem_cgroup_count_vm_event(struct mm_struct
*mm
, enum vm_event_item idx
)
1020 struct mem_cgroup
*memcg
;
1026 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
1027 if (unlikely(!memcg
))
1032 this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGFAULT
]);
1035 this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGMAJFAULT
]);
1043 EXPORT_SYMBOL(mem_cgroup_count_vm_event
);
1046 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1047 * @zone: zone of the wanted lruvec
1048 * @memcg: memcg of the wanted lruvec
1050 * Returns the lru list vector holding pages for the given @zone and
1051 * @mem. This can be the global zone lruvec, if the memory controller
1054 struct lruvec
*mem_cgroup_zone_lruvec(struct zone
*zone
,
1055 struct mem_cgroup
*memcg
)
1057 struct mem_cgroup_per_zone
*mz
;
1059 if (mem_cgroup_disabled())
1060 return &zone
->lruvec
;
1062 mz
= mem_cgroup_zoneinfo(memcg
, zone_to_nid(zone
), zone_idx(zone
));
1067 * Following LRU functions are allowed to be used without PCG_LOCK.
1068 * Operations are called by routine of global LRU independently from memcg.
1069 * What we have to take care of here is validness of pc->mem_cgroup.
1071 * Changes to pc->mem_cgroup happens when
1074 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1075 * It is added to LRU before charge.
1076 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1077 * When moving account, the page is not on LRU. It's isolated.
1081 * mem_cgroup_page_lruvec - return lruvec for adding an lru page
1083 * @zone: zone of the page
1085 struct lruvec
*mem_cgroup_page_lruvec(struct page
*page
, struct zone
*zone
)
1087 struct mem_cgroup_per_zone
*mz
;
1088 struct mem_cgroup
*memcg
;
1089 struct page_cgroup
*pc
;
1091 if (mem_cgroup_disabled())
1092 return &zone
->lruvec
;
1094 pc
= lookup_page_cgroup(page
);
1095 memcg
= pc
->mem_cgroup
;
1098 * Surreptitiously switch any uncharged offlist page to root:
1099 * an uncharged page off lru does nothing to secure
1100 * its former mem_cgroup from sudden removal.
1102 * Our caller holds lru_lock, and PageCgroupUsed is updated
1103 * under page_cgroup lock: between them, they make all uses
1104 * of pc->mem_cgroup safe.
1106 if (!PageLRU(page
) && !PageCgroupUsed(pc
) && memcg
!= root_mem_cgroup
)
1107 pc
->mem_cgroup
= memcg
= root_mem_cgroup
;
1109 mz
= page_cgroup_zoneinfo(memcg
, page
);
1114 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1115 * @lruvec: mem_cgroup per zone lru vector
1116 * @lru: index of lru list the page is sitting on
1117 * @nr_pages: positive when adding or negative when removing
1119 * This function must be called when a page is added to or removed from an
1122 void mem_cgroup_update_lru_size(struct lruvec
*lruvec
, enum lru_list lru
,
1125 struct mem_cgroup_per_zone
*mz
;
1126 unsigned long *lru_size
;
1128 if (mem_cgroup_disabled())
1131 mz
= container_of(lruvec
, struct mem_cgroup_per_zone
, lruvec
);
1132 lru_size
= mz
->lru_size
+ lru
;
1133 *lru_size
+= nr_pages
;
1134 VM_BUG_ON((long)(*lru_size
) < 0);
1138 * Checks whether given mem is same or in the root_mem_cgroup's
1141 bool __mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_memcg
,
1142 struct mem_cgroup
*memcg
)
1144 if (root_memcg
== memcg
)
1146 if (!root_memcg
->use_hierarchy
|| !memcg
)
1148 return css_is_ancestor(&memcg
->css
, &root_memcg
->css
);
1151 static bool mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_memcg
,
1152 struct mem_cgroup
*memcg
)
1157 ret
= __mem_cgroup_same_or_subtree(root_memcg
, memcg
);
1162 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*memcg
)
1165 struct mem_cgroup
*curr
= NULL
;
1166 struct task_struct
*p
;
1168 p
= find_lock_task_mm(task
);
1170 curr
= try_get_mem_cgroup_from_mm(p
->mm
);
1174 * All threads may have already detached their mm's, but the oom
1175 * killer still needs to detect if they have already been oom
1176 * killed to prevent needlessly killing additional tasks.
1179 curr
= mem_cgroup_from_task(task
);
1181 css_get(&curr
->css
);
1187 * We should check use_hierarchy of "memcg" not "curr". Because checking
1188 * use_hierarchy of "curr" here make this function true if hierarchy is
1189 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1190 * hierarchy(even if use_hierarchy is disabled in "memcg").
1192 ret
= mem_cgroup_same_or_subtree(memcg
, curr
);
1193 css_put(&curr
->css
);
1197 int mem_cgroup_inactive_anon_is_low(struct lruvec
*lruvec
)
1199 unsigned long inactive_ratio
;
1200 unsigned long inactive
;
1201 unsigned long active
;
1204 inactive
= mem_cgroup_get_lru_size(lruvec
, LRU_INACTIVE_ANON
);
1205 active
= mem_cgroup_get_lru_size(lruvec
, LRU_ACTIVE_ANON
);
1207 gb
= (inactive
+ active
) >> (30 - PAGE_SHIFT
);
1209 inactive_ratio
= int_sqrt(10 * gb
);
1213 return inactive
* inactive_ratio
< active
;
1216 int mem_cgroup_inactive_file_is_low(struct lruvec
*lruvec
)
1218 unsigned long active
;
1219 unsigned long inactive
;
1221 inactive
= mem_cgroup_get_lru_size(lruvec
, LRU_INACTIVE_FILE
);
1222 active
= mem_cgroup_get_lru_size(lruvec
, LRU_ACTIVE_FILE
);
1224 return (active
> inactive
);
1227 #define mem_cgroup_from_res_counter(counter, member) \
1228 container_of(counter, struct mem_cgroup, member)
1231 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1232 * @memcg: the memory cgroup
1234 * Returns the maximum amount of memory @mem can be charged with, in
1237 static unsigned long mem_cgroup_margin(struct mem_cgroup
*memcg
)
1239 unsigned long long margin
;
1241 margin
= res_counter_margin(&memcg
->res
);
1242 if (do_swap_account
)
1243 margin
= min(margin
, res_counter_margin(&memcg
->memsw
));
1244 return margin
>> PAGE_SHIFT
;
1247 int mem_cgroup_swappiness(struct mem_cgroup
*memcg
)
1249 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
1252 if (cgrp
->parent
== NULL
)
1253 return vm_swappiness
;
1255 return memcg
->swappiness
;
1259 * memcg->moving_account is used for checking possibility that some thread is
1260 * calling move_account(). When a thread on CPU-A starts moving pages under
1261 * a memcg, other threads should check memcg->moving_account under
1262 * rcu_read_lock(), like this:
1266 * memcg->moving_account+1 if (memcg->mocing_account)
1268 * synchronize_rcu() update something.
1273 /* for quick checking without looking up memcg */
1274 atomic_t memcg_moving __read_mostly
;
1276 static void mem_cgroup_start_move(struct mem_cgroup
*memcg
)
1278 atomic_inc(&memcg_moving
);
1279 atomic_inc(&memcg
->moving_account
);
1283 static void mem_cgroup_end_move(struct mem_cgroup
*memcg
)
1286 * Now, mem_cgroup_clear_mc() may call this function with NULL.
1287 * We check NULL in callee rather than caller.
1290 atomic_dec(&memcg_moving
);
1291 atomic_dec(&memcg
->moving_account
);
1296 * 2 routines for checking "mem" is under move_account() or not.
1298 * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This
1299 * is used for avoiding races in accounting. If true,
1300 * pc->mem_cgroup may be overwritten.
1302 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1303 * under hierarchy of moving cgroups. This is for
1304 * waiting at hith-memory prressure caused by "move".
1307 static bool mem_cgroup_stolen(struct mem_cgroup
*memcg
)
1309 VM_BUG_ON(!rcu_read_lock_held());
1310 return atomic_read(&memcg
->moving_account
) > 0;
1313 static bool mem_cgroup_under_move(struct mem_cgroup
*memcg
)
1315 struct mem_cgroup
*from
;
1316 struct mem_cgroup
*to
;
1319 * Unlike task_move routines, we access mc.to, mc.from not under
1320 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1322 spin_lock(&mc
.lock
);
1328 ret
= mem_cgroup_same_or_subtree(memcg
, from
)
1329 || mem_cgroup_same_or_subtree(memcg
, to
);
1331 spin_unlock(&mc
.lock
);
1335 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*memcg
)
1337 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1338 if (mem_cgroup_under_move(memcg
)) {
1340 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1341 /* moving charge context might have finished. */
1344 finish_wait(&mc
.waitq
, &wait
);
1352 * Take this lock when
1353 * - a code tries to modify page's memcg while it's USED.
1354 * - a code tries to modify page state accounting in a memcg.
1355 * see mem_cgroup_stolen(), too.
1357 static void move_lock_mem_cgroup(struct mem_cgroup
*memcg
,
1358 unsigned long *flags
)
1360 spin_lock_irqsave(&memcg
->move_lock
, *flags
);
1363 static void move_unlock_mem_cgroup(struct mem_cgroup
*memcg
,
1364 unsigned long *flags
)
1366 spin_unlock_irqrestore(&memcg
->move_lock
, *flags
);
1370 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1371 * @memcg: The memory cgroup that went over limit
1372 * @p: Task that is going to be killed
1374 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1377 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1379 struct cgroup
*task_cgrp
;
1380 struct cgroup
*mem_cgrp
;
1382 * Need a buffer in BSS, can't rely on allocations. The code relies
1383 * on the assumption that OOM is serialized for memory controller.
1384 * If this assumption is broken, revisit this code.
1386 static char memcg_name
[PATH_MAX
];
1394 mem_cgrp
= memcg
->css
.cgroup
;
1395 task_cgrp
= task_cgroup(p
, mem_cgroup_subsys_id
);
1397 ret
= cgroup_path(task_cgrp
, memcg_name
, PATH_MAX
);
1400 * Unfortunately, we are unable to convert to a useful name
1401 * But we'll still print out the usage information
1408 printk(KERN_INFO
"Task in %s killed", memcg_name
);
1411 ret
= cgroup_path(mem_cgrp
, memcg_name
, PATH_MAX
);
1419 * Continues from above, so we don't need an KERN_ level
1421 printk(KERN_CONT
" as a result of limit of %s\n", memcg_name
);
1424 printk(KERN_INFO
"memory: usage %llukB, limit %llukB, failcnt %llu\n",
1425 res_counter_read_u64(&memcg
->res
, RES_USAGE
) >> 10,
1426 res_counter_read_u64(&memcg
->res
, RES_LIMIT
) >> 10,
1427 res_counter_read_u64(&memcg
->res
, RES_FAILCNT
));
1428 printk(KERN_INFO
"memory+swap: usage %llukB, limit %llukB, "
1430 res_counter_read_u64(&memcg
->memsw
, RES_USAGE
) >> 10,
1431 res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
) >> 10,
1432 res_counter_read_u64(&memcg
->memsw
, RES_FAILCNT
));
1436 * This function returns the number of memcg under hierarchy tree. Returns
1437 * 1(self count) if no children.
1439 static int mem_cgroup_count_children(struct mem_cgroup
*memcg
)
1442 struct mem_cgroup
*iter
;
1444 for_each_mem_cgroup_tree(iter
, memcg
)
1450 * Return the memory (and swap, if configured) limit for a memcg.
1452 static u64
mem_cgroup_get_limit(struct mem_cgroup
*memcg
)
1456 limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
1459 * Do not consider swap space if we cannot swap due to swappiness
1461 if (mem_cgroup_swappiness(memcg
)) {
1464 limit
+= total_swap_pages
<< PAGE_SHIFT
;
1465 memsw
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
1468 * If memsw is finite and limits the amount of swap space
1469 * available to this memcg, return that limit.
1471 limit
= min(limit
, memsw
);
1477 void mem_cgroup_out_of_memory(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
1480 struct mem_cgroup
*iter
;
1481 unsigned long chosen_points
= 0;
1482 unsigned long totalpages
;
1483 unsigned int points
= 0;
1484 struct task_struct
*chosen
= NULL
;
1487 * If current has a pending SIGKILL, then automatically select it. The
1488 * goal is to allow it to allocate so that it may quickly exit and free
1491 if (fatal_signal_pending(current
)) {
1492 set_thread_flag(TIF_MEMDIE
);
1496 check_panic_on_oom(CONSTRAINT_MEMCG
, gfp_mask
, order
, NULL
);
1497 totalpages
= mem_cgroup_get_limit(memcg
) >> PAGE_SHIFT
? : 1;
1498 for_each_mem_cgroup_tree(iter
, memcg
) {
1499 struct cgroup
*cgroup
= iter
->css
.cgroup
;
1500 struct cgroup_iter it
;
1501 struct task_struct
*task
;
1503 cgroup_iter_start(cgroup
, &it
);
1504 while ((task
= cgroup_iter_next(cgroup
, &it
))) {
1505 switch (oom_scan_process_thread(task
, totalpages
, NULL
,
1507 case OOM_SCAN_SELECT
:
1509 put_task_struct(chosen
);
1511 chosen_points
= ULONG_MAX
;
1512 get_task_struct(chosen
);
1514 case OOM_SCAN_CONTINUE
:
1516 case OOM_SCAN_ABORT
:
1517 cgroup_iter_end(cgroup
, &it
);
1518 mem_cgroup_iter_break(memcg
, iter
);
1520 put_task_struct(chosen
);
1525 points
= oom_badness(task
, memcg
, NULL
, totalpages
);
1526 if (points
> chosen_points
) {
1528 put_task_struct(chosen
);
1530 chosen_points
= points
;
1531 get_task_struct(chosen
);
1534 cgroup_iter_end(cgroup
, &it
);
1539 points
= chosen_points
* 1000 / totalpages
;
1540 oom_kill_process(chosen
, gfp_mask
, order
, points
, totalpages
, memcg
,
1541 NULL
, "Memory cgroup out of memory");
1544 static unsigned long mem_cgroup_reclaim(struct mem_cgroup
*memcg
,
1546 unsigned long flags
)
1548 unsigned long total
= 0;
1549 bool noswap
= false;
1552 if (flags
& MEM_CGROUP_RECLAIM_NOSWAP
)
1554 if (!(flags
& MEM_CGROUP_RECLAIM_SHRINK
) && memcg
->memsw_is_minimum
)
1557 for (loop
= 0; loop
< MEM_CGROUP_MAX_RECLAIM_LOOPS
; loop
++) {
1559 drain_all_stock_async(memcg
);
1560 total
+= try_to_free_mem_cgroup_pages(memcg
, gfp_mask
, noswap
);
1562 * Allow limit shrinkers, which are triggered directly
1563 * by userspace, to catch signals and stop reclaim
1564 * after minimal progress, regardless of the margin.
1566 if (total
&& (flags
& MEM_CGROUP_RECLAIM_SHRINK
))
1568 if (mem_cgroup_margin(memcg
))
1571 * If nothing was reclaimed after two attempts, there
1572 * may be no reclaimable pages in this hierarchy.
1581 * test_mem_cgroup_node_reclaimable
1582 * @memcg: the target memcg
1583 * @nid: the node ID to be checked.
1584 * @noswap : specify true here if the user wants flle only information.
1586 * This function returns whether the specified memcg contains any
1587 * reclaimable pages on a node. Returns true if there are any reclaimable
1588 * pages in the node.
1590 static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup
*memcg
,
1591 int nid
, bool noswap
)
1593 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_FILE
))
1595 if (noswap
|| !total_swap_pages
)
1597 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_ANON
))
1602 #if MAX_NUMNODES > 1
1605 * Always updating the nodemask is not very good - even if we have an empty
1606 * list or the wrong list here, we can start from some node and traverse all
1607 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1610 static void mem_cgroup_may_update_nodemask(struct mem_cgroup
*memcg
)
1614 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1615 * pagein/pageout changes since the last update.
1617 if (!atomic_read(&memcg
->numainfo_events
))
1619 if (atomic_inc_return(&memcg
->numainfo_updating
) > 1)
1622 /* make a nodemask where this memcg uses memory from */
1623 memcg
->scan_nodes
= node_states
[N_HIGH_MEMORY
];
1625 for_each_node_mask(nid
, node_states
[N_HIGH_MEMORY
]) {
1627 if (!test_mem_cgroup_node_reclaimable(memcg
, nid
, false))
1628 node_clear(nid
, memcg
->scan_nodes
);
1631 atomic_set(&memcg
->numainfo_events
, 0);
1632 atomic_set(&memcg
->numainfo_updating
, 0);
1636 * Selecting a node where we start reclaim from. Because what we need is just
1637 * reducing usage counter, start from anywhere is O,K. Considering
1638 * memory reclaim from current node, there are pros. and cons.
1640 * Freeing memory from current node means freeing memory from a node which
1641 * we'll use or we've used. So, it may make LRU bad. And if several threads
1642 * hit limits, it will see a contention on a node. But freeing from remote
1643 * node means more costs for memory reclaim because of memory latency.
1645 * Now, we use round-robin. Better algorithm is welcomed.
1647 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1651 mem_cgroup_may_update_nodemask(memcg
);
1652 node
= memcg
->last_scanned_node
;
1654 node
= next_node(node
, memcg
->scan_nodes
);
1655 if (node
== MAX_NUMNODES
)
1656 node
= first_node(memcg
->scan_nodes
);
1658 * We call this when we hit limit, not when pages are added to LRU.
1659 * No LRU may hold pages because all pages are UNEVICTABLE or
1660 * memcg is too small and all pages are not on LRU. In that case,
1661 * we use curret node.
1663 if (unlikely(node
== MAX_NUMNODES
))
1664 node
= numa_node_id();
1666 memcg
->last_scanned_node
= node
;
1671 * Check all nodes whether it contains reclaimable pages or not.
1672 * For quick scan, we make use of scan_nodes. This will allow us to skip
1673 * unused nodes. But scan_nodes is lazily updated and may not cotain
1674 * enough new information. We need to do double check.
1676 static bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1681 * quick check...making use of scan_node.
1682 * We can skip unused nodes.
1684 if (!nodes_empty(memcg
->scan_nodes
)) {
1685 for (nid
= first_node(memcg
->scan_nodes
);
1687 nid
= next_node(nid
, memcg
->scan_nodes
)) {
1689 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1694 * Check rest of nodes.
1696 for_each_node_state(nid
, N_HIGH_MEMORY
) {
1697 if (node_isset(nid
, memcg
->scan_nodes
))
1699 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1706 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1711 static bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1713 return test_mem_cgroup_node_reclaimable(memcg
, 0, noswap
);
1717 static int mem_cgroup_soft_reclaim(struct mem_cgroup
*root_memcg
,
1720 unsigned long *total_scanned
)
1722 struct mem_cgroup
*victim
= NULL
;
1725 unsigned long excess
;
1726 unsigned long nr_scanned
;
1727 struct mem_cgroup_reclaim_cookie reclaim
= {
1732 excess
= res_counter_soft_limit_excess(&root_memcg
->res
) >> PAGE_SHIFT
;
1735 victim
= mem_cgroup_iter(root_memcg
, victim
, &reclaim
);
1740 * If we have not been able to reclaim
1741 * anything, it might because there are
1742 * no reclaimable pages under this hierarchy
1747 * We want to do more targeted reclaim.
1748 * excess >> 2 is not to excessive so as to
1749 * reclaim too much, nor too less that we keep
1750 * coming back to reclaim from this cgroup
1752 if (total
>= (excess
>> 2) ||
1753 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
))
1758 if (!mem_cgroup_reclaimable(victim
, false))
1760 total
+= mem_cgroup_shrink_node_zone(victim
, gfp_mask
, false,
1762 *total_scanned
+= nr_scanned
;
1763 if (!res_counter_soft_limit_excess(&root_memcg
->res
))
1766 mem_cgroup_iter_break(root_memcg
, victim
);
1771 * Check OOM-Killer is already running under our hierarchy.
1772 * If someone is running, return false.
1773 * Has to be called with memcg_oom_lock
1775 static bool mem_cgroup_oom_lock(struct mem_cgroup
*memcg
)
1777 struct mem_cgroup
*iter
, *failed
= NULL
;
1779 for_each_mem_cgroup_tree(iter
, memcg
) {
1780 if (iter
->oom_lock
) {
1782 * this subtree of our hierarchy is already locked
1783 * so we cannot give a lock.
1786 mem_cgroup_iter_break(memcg
, iter
);
1789 iter
->oom_lock
= true;
1796 * OK, we failed to lock the whole subtree so we have to clean up
1797 * what we set up to the failing subtree
1799 for_each_mem_cgroup_tree(iter
, memcg
) {
1800 if (iter
== failed
) {
1801 mem_cgroup_iter_break(memcg
, iter
);
1804 iter
->oom_lock
= false;
1810 * Has to be called with memcg_oom_lock
1812 static int mem_cgroup_oom_unlock(struct mem_cgroup
*memcg
)
1814 struct mem_cgroup
*iter
;
1816 for_each_mem_cgroup_tree(iter
, memcg
)
1817 iter
->oom_lock
= false;
1821 static void mem_cgroup_mark_under_oom(struct mem_cgroup
*memcg
)
1823 struct mem_cgroup
*iter
;
1825 for_each_mem_cgroup_tree(iter
, memcg
)
1826 atomic_inc(&iter
->under_oom
);
1829 static void mem_cgroup_unmark_under_oom(struct mem_cgroup
*memcg
)
1831 struct mem_cgroup
*iter
;
1834 * When a new child is created while the hierarchy is under oom,
1835 * mem_cgroup_oom_lock() may not be called. We have to use
1836 * atomic_add_unless() here.
1838 for_each_mem_cgroup_tree(iter
, memcg
)
1839 atomic_add_unless(&iter
->under_oom
, -1, 0);
1842 static DEFINE_SPINLOCK(memcg_oom_lock
);
1843 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1845 struct oom_wait_info
{
1846 struct mem_cgroup
*memcg
;
1850 static int memcg_oom_wake_function(wait_queue_t
*wait
,
1851 unsigned mode
, int sync
, void *arg
)
1853 struct mem_cgroup
*wake_memcg
= (struct mem_cgroup
*)arg
;
1854 struct mem_cgroup
*oom_wait_memcg
;
1855 struct oom_wait_info
*oom_wait_info
;
1857 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1858 oom_wait_memcg
= oom_wait_info
->memcg
;
1861 * Both of oom_wait_info->memcg and wake_memcg are stable under us.
1862 * Then we can use css_is_ancestor without taking care of RCU.
1864 if (!mem_cgroup_same_or_subtree(oom_wait_memcg
, wake_memcg
)
1865 && !mem_cgroup_same_or_subtree(wake_memcg
, oom_wait_memcg
))
1867 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1870 static void memcg_wakeup_oom(struct mem_cgroup
*memcg
)
1872 /* for filtering, pass "memcg" as argument. */
1873 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, memcg
);
1876 static void memcg_oom_recover(struct mem_cgroup
*memcg
)
1878 if (memcg
&& atomic_read(&memcg
->under_oom
))
1879 memcg_wakeup_oom(memcg
);
1883 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1885 static bool mem_cgroup_handle_oom(struct mem_cgroup
*memcg
, gfp_t mask
,
1888 struct oom_wait_info owait
;
1889 bool locked
, need_to_kill
;
1891 owait
.memcg
= memcg
;
1892 owait
.wait
.flags
= 0;
1893 owait
.wait
.func
= memcg_oom_wake_function
;
1894 owait
.wait
.private = current
;
1895 INIT_LIST_HEAD(&owait
.wait
.task_list
);
1896 need_to_kill
= true;
1897 mem_cgroup_mark_under_oom(memcg
);
1899 /* At first, try to OOM lock hierarchy under memcg.*/
1900 spin_lock(&memcg_oom_lock
);
1901 locked
= mem_cgroup_oom_lock(memcg
);
1903 * Even if signal_pending(), we can't quit charge() loop without
1904 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1905 * under OOM is always welcomed, use TASK_KILLABLE here.
1907 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1908 if (!locked
|| memcg
->oom_kill_disable
)
1909 need_to_kill
= false;
1911 mem_cgroup_oom_notify(memcg
);
1912 spin_unlock(&memcg_oom_lock
);
1915 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1916 mem_cgroup_out_of_memory(memcg
, mask
, order
);
1919 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1921 spin_lock(&memcg_oom_lock
);
1923 mem_cgroup_oom_unlock(memcg
);
1924 memcg_wakeup_oom(memcg
);
1925 spin_unlock(&memcg_oom_lock
);
1927 mem_cgroup_unmark_under_oom(memcg
);
1929 if (test_thread_flag(TIF_MEMDIE
) || fatal_signal_pending(current
))
1931 /* Give chance to dying process */
1932 schedule_timeout_uninterruptible(1);
1937 * Currently used to update mapped file statistics, but the routine can be
1938 * generalized to update other statistics as well.
1940 * Notes: Race condition
1942 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1943 * it tends to be costly. But considering some conditions, we doesn't need
1944 * to do so _always_.
1946 * Considering "charge", lock_page_cgroup() is not required because all
1947 * file-stat operations happen after a page is attached to radix-tree. There
1948 * are no race with "charge".
1950 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1951 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1952 * if there are race with "uncharge". Statistics itself is properly handled
1955 * Considering "move", this is an only case we see a race. To make the race
1956 * small, we check mm->moving_account and detect there are possibility of race
1957 * If there is, we take a lock.
1960 void __mem_cgroup_begin_update_page_stat(struct page
*page
,
1961 bool *locked
, unsigned long *flags
)
1963 struct mem_cgroup
*memcg
;
1964 struct page_cgroup
*pc
;
1966 pc
= lookup_page_cgroup(page
);
1968 memcg
= pc
->mem_cgroup
;
1969 if (unlikely(!memcg
|| !PageCgroupUsed(pc
)))
1972 * If this memory cgroup is not under account moving, we don't
1973 * need to take move_lock_mem_cgroup(). Because we already hold
1974 * rcu_read_lock(), any calls to move_account will be delayed until
1975 * rcu_read_unlock() if mem_cgroup_stolen() == true.
1977 if (!mem_cgroup_stolen(memcg
))
1980 move_lock_mem_cgroup(memcg
, flags
);
1981 if (memcg
!= pc
->mem_cgroup
|| !PageCgroupUsed(pc
)) {
1982 move_unlock_mem_cgroup(memcg
, flags
);
1988 void __mem_cgroup_end_update_page_stat(struct page
*page
, unsigned long *flags
)
1990 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1993 * It's guaranteed that pc->mem_cgroup never changes while
1994 * lock is held because a routine modifies pc->mem_cgroup
1995 * should take move_lock_mem_cgroup().
1997 move_unlock_mem_cgroup(pc
->mem_cgroup
, flags
);
2000 void mem_cgroup_update_page_stat(struct page
*page
,
2001 enum mem_cgroup_page_stat_item idx
, int val
)
2003 struct mem_cgroup
*memcg
;
2004 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
2005 unsigned long uninitialized_var(flags
);
2007 if (mem_cgroup_disabled())
2010 memcg
= pc
->mem_cgroup
;
2011 if (unlikely(!memcg
|| !PageCgroupUsed(pc
)))
2015 case MEMCG_NR_FILE_MAPPED
:
2016 idx
= MEM_CGROUP_STAT_FILE_MAPPED
;
2022 this_cpu_add(memcg
->stat
->count
[idx
], val
);
2026 * size of first charge trial. "32" comes from vmscan.c's magic value.
2027 * TODO: maybe necessary to use big numbers in big irons.
2029 #define CHARGE_BATCH 32U
2030 struct memcg_stock_pcp
{
2031 struct mem_cgroup
*cached
; /* this never be root cgroup */
2032 unsigned int nr_pages
;
2033 struct work_struct work
;
2034 unsigned long flags
;
2035 #define FLUSHING_CACHED_CHARGE 0
2037 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
2038 static DEFINE_MUTEX(percpu_charge_mutex
);
2041 * Try to consume stocked charge on this cpu. If success, one page is consumed
2042 * from local stock and true is returned. If the stock is 0 or charges from a
2043 * cgroup which is not current target, returns false. This stock will be
2046 static bool consume_stock(struct mem_cgroup
*memcg
)
2048 struct memcg_stock_pcp
*stock
;
2051 stock
= &get_cpu_var(memcg_stock
);
2052 if (memcg
== stock
->cached
&& stock
->nr_pages
)
2054 else /* need to call res_counter_charge */
2056 put_cpu_var(memcg_stock
);
2061 * Returns stocks cached in percpu to res_counter and reset cached information.
2063 static void drain_stock(struct memcg_stock_pcp
*stock
)
2065 struct mem_cgroup
*old
= stock
->cached
;
2067 if (stock
->nr_pages
) {
2068 unsigned long bytes
= stock
->nr_pages
* PAGE_SIZE
;
2070 res_counter_uncharge(&old
->res
, bytes
);
2071 if (do_swap_account
)
2072 res_counter_uncharge(&old
->memsw
, bytes
);
2073 stock
->nr_pages
= 0;
2075 stock
->cached
= NULL
;
2079 * This must be called under preempt disabled or must be called by
2080 * a thread which is pinned to local cpu.
2082 static void drain_local_stock(struct work_struct
*dummy
)
2084 struct memcg_stock_pcp
*stock
= &__get_cpu_var(memcg_stock
);
2086 clear_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
);
2090 * Cache charges(val) which is from res_counter, to local per_cpu area.
2091 * This will be consumed by consume_stock() function, later.
2093 static void refill_stock(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
2095 struct memcg_stock_pcp
*stock
= &get_cpu_var(memcg_stock
);
2097 if (stock
->cached
!= memcg
) { /* reset if necessary */
2099 stock
->cached
= memcg
;
2101 stock
->nr_pages
+= nr_pages
;
2102 put_cpu_var(memcg_stock
);
2106 * Drains all per-CPU charge caches for given root_memcg resp. subtree
2107 * of the hierarchy under it. sync flag says whether we should block
2108 * until the work is done.
2110 static void drain_all_stock(struct mem_cgroup
*root_memcg
, bool sync
)
2114 /* Notify other cpus that system-wide "drain" is running */
2117 for_each_online_cpu(cpu
) {
2118 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2119 struct mem_cgroup
*memcg
;
2121 memcg
= stock
->cached
;
2122 if (!memcg
|| !stock
->nr_pages
)
2124 if (!mem_cgroup_same_or_subtree(root_memcg
, memcg
))
2126 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
)) {
2128 drain_local_stock(&stock
->work
);
2130 schedule_work_on(cpu
, &stock
->work
);
2138 for_each_online_cpu(cpu
) {
2139 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2140 if (test_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
))
2141 flush_work(&stock
->work
);
2148 * Tries to drain stocked charges in other cpus. This function is asynchronous
2149 * and just put a work per cpu for draining localy on each cpu. Caller can
2150 * expects some charges will be back to res_counter later but cannot wait for
2153 static void drain_all_stock_async(struct mem_cgroup
*root_memcg
)
2156 * If someone calls draining, avoid adding more kworker runs.
2158 if (!mutex_trylock(&percpu_charge_mutex
))
2160 drain_all_stock(root_memcg
, false);
2161 mutex_unlock(&percpu_charge_mutex
);
2164 /* This is a synchronous drain interface. */
2165 static void drain_all_stock_sync(struct mem_cgroup
*root_memcg
)
2167 /* called when force_empty is called */
2168 mutex_lock(&percpu_charge_mutex
);
2169 drain_all_stock(root_memcg
, true);
2170 mutex_unlock(&percpu_charge_mutex
);
2174 * This function drains percpu counter value from DEAD cpu and
2175 * move it to local cpu. Note that this function can be preempted.
2177 static void mem_cgroup_drain_pcp_counter(struct mem_cgroup
*memcg
, int cpu
)
2181 spin_lock(&memcg
->pcp_counter_lock
);
2182 for (i
= 0; i
< MEM_CGROUP_STAT_NSTATS
; i
++) {
2183 long x
= per_cpu(memcg
->stat
->count
[i
], cpu
);
2185 per_cpu(memcg
->stat
->count
[i
], cpu
) = 0;
2186 memcg
->nocpu_base
.count
[i
] += x
;
2188 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
2189 unsigned long x
= per_cpu(memcg
->stat
->events
[i
], cpu
);
2191 per_cpu(memcg
->stat
->events
[i
], cpu
) = 0;
2192 memcg
->nocpu_base
.events
[i
] += x
;
2194 spin_unlock(&memcg
->pcp_counter_lock
);
2197 static int __cpuinit
memcg_cpu_hotplug_callback(struct notifier_block
*nb
,
2198 unsigned long action
,
2201 int cpu
= (unsigned long)hcpu
;
2202 struct memcg_stock_pcp
*stock
;
2203 struct mem_cgroup
*iter
;
2205 if (action
== CPU_ONLINE
)
2208 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
2211 for_each_mem_cgroup(iter
)
2212 mem_cgroup_drain_pcp_counter(iter
, cpu
);
2214 stock
= &per_cpu(memcg_stock
, cpu
);
2220 /* See __mem_cgroup_try_charge() for details */
2222 CHARGE_OK
, /* success */
2223 CHARGE_RETRY
, /* need to retry but retry is not bad */
2224 CHARGE_NOMEM
, /* we can't do more. return -ENOMEM */
2225 CHARGE_WOULDBLOCK
, /* GFP_WAIT wasn't set and no enough res. */
2226 CHARGE_OOM_DIE
, /* the current is killed because of OOM */
2229 static int mem_cgroup_do_charge(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
2230 unsigned int nr_pages
, bool oom_check
)
2232 unsigned long csize
= nr_pages
* PAGE_SIZE
;
2233 struct mem_cgroup
*mem_over_limit
;
2234 struct res_counter
*fail_res
;
2235 unsigned long flags
= 0;
2238 ret
= res_counter_charge(&memcg
->res
, csize
, &fail_res
);
2241 if (!do_swap_account
)
2243 ret
= res_counter_charge(&memcg
->memsw
, csize
, &fail_res
);
2247 res_counter_uncharge(&memcg
->res
, csize
);
2248 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, memsw
);
2249 flags
|= MEM_CGROUP_RECLAIM_NOSWAP
;
2251 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, res
);
2253 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2254 * of regular pages (CHARGE_BATCH), or a single regular page (1).
2256 * Never reclaim on behalf of optional batching, retry with a
2257 * single page instead.
2259 if (nr_pages
== CHARGE_BATCH
)
2260 return CHARGE_RETRY
;
2262 if (!(gfp_mask
& __GFP_WAIT
))
2263 return CHARGE_WOULDBLOCK
;
2265 ret
= mem_cgroup_reclaim(mem_over_limit
, gfp_mask
, flags
);
2266 if (mem_cgroup_margin(mem_over_limit
) >= nr_pages
)
2267 return CHARGE_RETRY
;
2269 * Even though the limit is exceeded at this point, reclaim
2270 * may have been able to free some pages. Retry the charge
2271 * before killing the task.
2273 * Only for regular pages, though: huge pages are rather
2274 * unlikely to succeed so close to the limit, and we fall back
2275 * to regular pages anyway in case of failure.
2277 if (nr_pages
== 1 && ret
)
2278 return CHARGE_RETRY
;
2281 * At task move, charge accounts can be doubly counted. So, it's
2282 * better to wait until the end of task_move if something is going on.
2284 if (mem_cgroup_wait_acct_move(mem_over_limit
))
2285 return CHARGE_RETRY
;
2287 /* If we don't need to call oom-killer at el, return immediately */
2289 return CHARGE_NOMEM
;
2291 if (!mem_cgroup_handle_oom(mem_over_limit
, gfp_mask
, get_order(csize
)))
2292 return CHARGE_OOM_DIE
;
2294 return CHARGE_RETRY
;
2298 * __mem_cgroup_try_charge() does
2299 * 1. detect memcg to be charged against from passed *mm and *ptr,
2300 * 2. update res_counter
2301 * 3. call memory reclaim if necessary.
2303 * In some special case, if the task is fatal, fatal_signal_pending() or
2304 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2305 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2306 * as possible without any hazards. 2: all pages should have a valid
2307 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2308 * pointer, that is treated as a charge to root_mem_cgroup.
2310 * So __mem_cgroup_try_charge() will return
2311 * 0 ... on success, filling *ptr with a valid memcg pointer.
2312 * -ENOMEM ... charge failure because of resource limits.
2313 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2315 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2316 * the oom-killer can be invoked.
2318 static int __mem_cgroup_try_charge(struct mm_struct
*mm
,
2320 unsigned int nr_pages
,
2321 struct mem_cgroup
**ptr
,
2324 unsigned int batch
= max(CHARGE_BATCH
, nr_pages
);
2325 int nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2326 struct mem_cgroup
*memcg
= NULL
;
2330 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2331 * in system level. So, allow to go ahead dying process in addition to
2334 if (unlikely(test_thread_flag(TIF_MEMDIE
)
2335 || fatal_signal_pending(current
)))
2339 * We always charge the cgroup the mm_struct belongs to.
2340 * The mm_struct's mem_cgroup changes on task migration if the
2341 * thread group leader migrates. It's possible that mm is not
2342 * set, if so charge the root memcg (happens for pagecache usage).
2345 *ptr
= root_mem_cgroup
;
2347 if (*ptr
) { /* css should be a valid one */
2349 VM_BUG_ON(css_is_removed(&memcg
->css
));
2350 if (mem_cgroup_is_root(memcg
))
2352 if (nr_pages
== 1 && consume_stock(memcg
))
2354 css_get(&memcg
->css
);
2356 struct task_struct
*p
;
2359 p
= rcu_dereference(mm
->owner
);
2361 * Because we don't have task_lock(), "p" can exit.
2362 * In that case, "memcg" can point to root or p can be NULL with
2363 * race with swapoff. Then, we have small risk of mis-accouning.
2364 * But such kind of mis-account by race always happens because
2365 * we don't have cgroup_mutex(). It's overkill and we allo that
2367 * (*) swapoff at el will charge against mm-struct not against
2368 * task-struct. So, mm->owner can be NULL.
2370 memcg
= mem_cgroup_from_task(p
);
2372 memcg
= root_mem_cgroup
;
2373 if (mem_cgroup_is_root(memcg
)) {
2377 if (nr_pages
== 1 && consume_stock(memcg
)) {
2379 * It seems dagerous to access memcg without css_get().
2380 * But considering how consume_stok works, it's not
2381 * necessary. If consume_stock success, some charges
2382 * from this memcg are cached on this cpu. So, we
2383 * don't need to call css_get()/css_tryget() before
2384 * calling consume_stock().
2389 /* after here, we may be blocked. we need to get refcnt */
2390 if (!css_tryget(&memcg
->css
)) {
2400 /* If killed, bypass charge */
2401 if (fatal_signal_pending(current
)) {
2402 css_put(&memcg
->css
);
2407 if (oom
&& !nr_oom_retries
) {
2409 nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2412 ret
= mem_cgroup_do_charge(memcg
, gfp_mask
, batch
, oom_check
);
2416 case CHARGE_RETRY
: /* not in OOM situation but retry */
2418 css_put(&memcg
->css
);
2421 case CHARGE_WOULDBLOCK
: /* !__GFP_WAIT */
2422 css_put(&memcg
->css
);
2424 case CHARGE_NOMEM
: /* OOM routine works */
2426 css_put(&memcg
->css
);
2429 /* If oom, we never return -ENOMEM */
2432 case CHARGE_OOM_DIE
: /* Killed by OOM Killer */
2433 css_put(&memcg
->css
);
2436 } while (ret
!= CHARGE_OK
);
2438 if (batch
> nr_pages
)
2439 refill_stock(memcg
, batch
- nr_pages
);
2440 css_put(&memcg
->css
);
2448 *ptr
= root_mem_cgroup
;
2453 * Somemtimes we have to undo a charge we got by try_charge().
2454 * This function is for that and do uncharge, put css's refcnt.
2455 * gotten by try_charge().
2457 static void __mem_cgroup_cancel_charge(struct mem_cgroup
*memcg
,
2458 unsigned int nr_pages
)
2460 if (!mem_cgroup_is_root(memcg
)) {
2461 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2463 res_counter_uncharge(&memcg
->res
, bytes
);
2464 if (do_swap_account
)
2465 res_counter_uncharge(&memcg
->memsw
, bytes
);
2470 * Cancel chrages in this cgroup....doesn't propagate to parent cgroup.
2471 * This is useful when moving usage to parent cgroup.
2473 static void __mem_cgroup_cancel_local_charge(struct mem_cgroup
*memcg
,
2474 unsigned int nr_pages
)
2476 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2478 if (mem_cgroup_is_root(memcg
))
2481 res_counter_uncharge_until(&memcg
->res
, memcg
->res
.parent
, bytes
);
2482 if (do_swap_account
)
2483 res_counter_uncharge_until(&memcg
->memsw
,
2484 memcg
->memsw
.parent
, bytes
);
2488 * A helper function to get mem_cgroup from ID. must be called under
2489 * rcu_read_lock(). The caller must check css_is_removed() or some if
2490 * it's concern. (dropping refcnt from swap can be called against removed
2493 static struct mem_cgroup
*mem_cgroup_lookup(unsigned short id
)
2495 struct cgroup_subsys_state
*css
;
2497 /* ID 0 is unused ID */
2500 css
= css_lookup(&mem_cgroup_subsys
, id
);
2503 return mem_cgroup_from_css(css
);
2506 struct mem_cgroup
*try_get_mem_cgroup_from_page(struct page
*page
)
2508 struct mem_cgroup
*memcg
= NULL
;
2509 struct page_cgroup
*pc
;
2513 VM_BUG_ON(!PageLocked(page
));
2515 pc
= lookup_page_cgroup(page
);
2516 lock_page_cgroup(pc
);
2517 if (PageCgroupUsed(pc
)) {
2518 memcg
= pc
->mem_cgroup
;
2519 if (memcg
&& !css_tryget(&memcg
->css
))
2521 } else if (PageSwapCache(page
)) {
2522 ent
.val
= page_private(page
);
2523 id
= lookup_swap_cgroup_id(ent
);
2525 memcg
= mem_cgroup_lookup(id
);
2526 if (memcg
&& !css_tryget(&memcg
->css
))
2530 unlock_page_cgroup(pc
);
2534 static void __mem_cgroup_commit_charge(struct mem_cgroup
*memcg
,
2536 unsigned int nr_pages
,
2537 enum charge_type ctype
,
2540 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
2541 struct zone
*uninitialized_var(zone
);
2542 struct lruvec
*lruvec
;
2543 bool was_on_lru
= false;
2546 lock_page_cgroup(pc
);
2547 VM_BUG_ON(PageCgroupUsed(pc
));
2549 * we don't need page_cgroup_lock about tail pages, becase they are not
2550 * accessed by any other context at this point.
2554 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2555 * may already be on some other mem_cgroup's LRU. Take care of it.
2558 zone
= page_zone(page
);
2559 spin_lock_irq(&zone
->lru_lock
);
2560 if (PageLRU(page
)) {
2561 lruvec
= mem_cgroup_zone_lruvec(zone
, pc
->mem_cgroup
);
2563 del_page_from_lru_list(page
, lruvec
, page_lru(page
));
2568 pc
->mem_cgroup
= memcg
;
2570 * We access a page_cgroup asynchronously without lock_page_cgroup().
2571 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2572 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2573 * before USED bit, we need memory barrier here.
2574 * See mem_cgroup_add_lru_list(), etc.
2577 SetPageCgroupUsed(pc
);
2581 lruvec
= mem_cgroup_zone_lruvec(zone
, pc
->mem_cgroup
);
2582 VM_BUG_ON(PageLRU(page
));
2584 add_page_to_lru_list(page
, lruvec
, page_lru(page
));
2586 spin_unlock_irq(&zone
->lru_lock
);
2589 if (ctype
== MEM_CGROUP_CHARGE_TYPE_ANON
)
2594 mem_cgroup_charge_statistics(memcg
, anon
, nr_pages
);
2595 unlock_page_cgroup(pc
);
2598 * "charge_statistics" updated event counter. Then, check it.
2599 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2600 * if they exceeds softlimit.
2602 memcg_check_events(memcg
, page
);
2605 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2607 #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION)
2609 * Because tail pages are not marked as "used", set it. We're under
2610 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2611 * charge/uncharge will be never happen and move_account() is done under
2612 * compound_lock(), so we don't have to take care of races.
2614 void mem_cgroup_split_huge_fixup(struct page
*head
)
2616 struct page_cgroup
*head_pc
= lookup_page_cgroup(head
);
2617 struct page_cgroup
*pc
;
2620 if (mem_cgroup_disabled())
2622 for (i
= 1; i
< HPAGE_PMD_NR
; i
++) {
2624 pc
->mem_cgroup
= head_pc
->mem_cgroup
;
2625 smp_wmb();/* see __commit_charge() */
2626 pc
->flags
= head_pc
->flags
& ~PCGF_NOCOPY_AT_SPLIT
;
2629 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2632 * mem_cgroup_move_account - move account of the page
2634 * @nr_pages: number of regular pages (>1 for huge pages)
2635 * @pc: page_cgroup of the page.
2636 * @from: mem_cgroup which the page is moved from.
2637 * @to: mem_cgroup which the page is moved to. @from != @to.
2639 * The caller must confirm following.
2640 * - page is not on LRU (isolate_page() is useful.)
2641 * - compound_lock is held when nr_pages > 1
2643 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
2646 static int mem_cgroup_move_account(struct page
*page
,
2647 unsigned int nr_pages
,
2648 struct page_cgroup
*pc
,
2649 struct mem_cgroup
*from
,
2650 struct mem_cgroup
*to
)
2652 unsigned long flags
;
2654 bool anon
= PageAnon(page
);
2656 VM_BUG_ON(from
== to
);
2657 VM_BUG_ON(PageLRU(page
));
2659 * The page is isolated from LRU. So, collapse function
2660 * will not handle this page. But page splitting can happen.
2661 * Do this check under compound_page_lock(). The caller should
2665 if (nr_pages
> 1 && !PageTransHuge(page
))
2668 lock_page_cgroup(pc
);
2671 if (!PageCgroupUsed(pc
) || pc
->mem_cgroup
!= from
)
2674 move_lock_mem_cgroup(from
, &flags
);
2676 if (!anon
&& page_mapped(page
)) {
2677 /* Update mapped_file data for mem_cgroup */
2679 __this_cpu_dec(from
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2680 __this_cpu_inc(to
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2683 mem_cgroup_charge_statistics(from
, anon
, -nr_pages
);
2685 /* caller should have done css_get */
2686 pc
->mem_cgroup
= to
;
2687 mem_cgroup_charge_statistics(to
, anon
, nr_pages
);
2689 * We charges against "to" which may not have any tasks. Then, "to"
2690 * can be under rmdir(). But in current implementation, caller of
2691 * this function is just force_empty() and move charge, so it's
2692 * guaranteed that "to" is never removed. So, we don't check rmdir
2695 move_unlock_mem_cgroup(from
, &flags
);
2698 unlock_page_cgroup(pc
);
2702 memcg_check_events(to
, page
);
2703 memcg_check_events(from
, page
);
2709 * move charges to its parent.
2712 static int mem_cgroup_move_parent(struct page
*page
,
2713 struct page_cgroup
*pc
,
2714 struct mem_cgroup
*child
)
2716 struct mem_cgroup
*parent
;
2717 unsigned int nr_pages
;
2718 unsigned long uninitialized_var(flags
);
2722 if (mem_cgroup_is_root(child
))
2726 if (!get_page_unless_zero(page
))
2728 if (isolate_lru_page(page
))
2731 nr_pages
= hpage_nr_pages(page
);
2733 parent
= parent_mem_cgroup(child
);
2735 * If no parent, move charges to root cgroup.
2738 parent
= root_mem_cgroup
;
2741 flags
= compound_lock_irqsave(page
);
2743 ret
= mem_cgroup_move_account(page
, nr_pages
,
2746 __mem_cgroup_cancel_local_charge(child
, nr_pages
);
2749 compound_unlock_irqrestore(page
, flags
);
2750 putback_lru_page(page
);
2758 * Charge the memory controller for page usage.
2760 * 0 if the charge was successful
2761 * < 0 if the cgroup is over its limit
2763 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
2764 gfp_t gfp_mask
, enum charge_type ctype
)
2766 struct mem_cgroup
*memcg
= NULL
;
2767 unsigned int nr_pages
= 1;
2771 if (PageTransHuge(page
)) {
2772 nr_pages
<<= compound_order(page
);
2773 VM_BUG_ON(!PageTransHuge(page
));
2775 * Never OOM-kill a process for a huge page. The
2776 * fault handler will fall back to regular pages.
2781 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, nr_pages
, &memcg
, oom
);
2784 __mem_cgroup_commit_charge(memcg
, page
, nr_pages
, ctype
, false);
2788 int mem_cgroup_newpage_charge(struct page
*page
,
2789 struct mm_struct
*mm
, gfp_t gfp_mask
)
2791 if (mem_cgroup_disabled())
2793 VM_BUG_ON(page_mapped(page
));
2794 VM_BUG_ON(page
->mapping
&& !PageAnon(page
));
2796 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2797 MEM_CGROUP_CHARGE_TYPE_ANON
);
2801 * While swap-in, try_charge -> commit or cancel, the page is locked.
2802 * And when try_charge() successfully returns, one refcnt to memcg without
2803 * struct page_cgroup is acquired. This refcnt will be consumed by
2804 * "commit()" or removed by "cancel()"
2806 static int __mem_cgroup_try_charge_swapin(struct mm_struct
*mm
,
2809 struct mem_cgroup
**memcgp
)
2811 struct mem_cgroup
*memcg
;
2812 struct page_cgroup
*pc
;
2815 pc
= lookup_page_cgroup(page
);
2817 * Every swap fault against a single page tries to charge the
2818 * page, bail as early as possible. shmem_unuse() encounters
2819 * already charged pages, too. The USED bit is protected by
2820 * the page lock, which serializes swap cache removal, which
2821 * in turn serializes uncharging.
2823 if (PageCgroupUsed(pc
))
2825 if (!do_swap_account
)
2827 memcg
= try_get_mem_cgroup_from_page(page
);
2831 ret
= __mem_cgroup_try_charge(NULL
, mask
, 1, memcgp
, true);
2832 css_put(&memcg
->css
);
2837 ret
= __mem_cgroup_try_charge(mm
, mask
, 1, memcgp
, true);
2843 int mem_cgroup_try_charge_swapin(struct mm_struct
*mm
, struct page
*page
,
2844 gfp_t gfp_mask
, struct mem_cgroup
**memcgp
)
2847 if (mem_cgroup_disabled())
2850 * A racing thread's fault, or swapoff, may have already
2851 * updated the pte, and even removed page from swap cache: in
2852 * those cases unuse_pte()'s pte_same() test will fail; but
2853 * there's also a KSM case which does need to charge the page.
2855 if (!PageSwapCache(page
)) {
2858 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, 1, memcgp
, true);
2863 return __mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, memcgp
);
2866 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup
*memcg
)
2868 if (mem_cgroup_disabled())
2872 __mem_cgroup_cancel_charge(memcg
, 1);
2876 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*memcg
,
2877 enum charge_type ctype
)
2879 if (mem_cgroup_disabled())
2883 cgroup_exclude_rmdir(&memcg
->css
);
2885 __mem_cgroup_commit_charge(memcg
, page
, 1, ctype
, true);
2887 * Now swap is on-memory. This means this page may be
2888 * counted both as mem and swap....double count.
2889 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2890 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2891 * may call delete_from_swap_cache() before reach here.
2893 if (do_swap_account
&& PageSwapCache(page
)) {
2894 swp_entry_t ent
= {.val
= page_private(page
)};
2895 mem_cgroup_uncharge_swap(ent
);
2898 * At swapin, we may charge account against cgroup which has no tasks.
2899 * So, rmdir()->pre_destroy() can be called while we do this charge.
2900 * In that case, we need to call pre_destroy() again. check it here.
2902 cgroup_release_and_wakeup_rmdir(&memcg
->css
);
2905 void mem_cgroup_commit_charge_swapin(struct page
*page
,
2906 struct mem_cgroup
*memcg
)
2908 __mem_cgroup_commit_charge_swapin(page
, memcg
,
2909 MEM_CGROUP_CHARGE_TYPE_ANON
);
2912 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
2915 struct mem_cgroup
*memcg
= NULL
;
2916 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
2919 if (mem_cgroup_disabled())
2921 if (PageCompound(page
))
2924 if (!PageSwapCache(page
))
2925 ret
= mem_cgroup_charge_common(page
, mm
, gfp_mask
, type
);
2926 else { /* page is swapcache/shmem */
2927 ret
= __mem_cgroup_try_charge_swapin(mm
, page
,
2930 __mem_cgroup_commit_charge_swapin(page
, memcg
, type
);
2935 static void mem_cgroup_do_uncharge(struct mem_cgroup
*memcg
,
2936 unsigned int nr_pages
,
2937 const enum charge_type ctype
)
2939 struct memcg_batch_info
*batch
= NULL
;
2940 bool uncharge_memsw
= true;
2942 /* If swapout, usage of swap doesn't decrease */
2943 if (!do_swap_account
|| ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
)
2944 uncharge_memsw
= false;
2946 batch
= ¤t
->memcg_batch
;
2948 * In usual, we do css_get() when we remember memcg pointer.
2949 * But in this case, we keep res->usage until end of a series of
2950 * uncharges. Then, it's ok to ignore memcg's refcnt.
2953 batch
->memcg
= memcg
;
2955 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2956 * In those cases, all pages freed continuously can be expected to be in
2957 * the same cgroup and we have chance to coalesce uncharges.
2958 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2959 * because we want to do uncharge as soon as possible.
2962 if (!batch
->do_batch
|| test_thread_flag(TIF_MEMDIE
))
2963 goto direct_uncharge
;
2966 goto direct_uncharge
;
2969 * In typical case, batch->memcg == mem. This means we can
2970 * merge a series of uncharges to an uncharge of res_counter.
2971 * If not, we uncharge res_counter ony by one.
2973 if (batch
->memcg
!= memcg
)
2974 goto direct_uncharge
;
2975 /* remember freed charge and uncharge it later */
2978 batch
->memsw_nr_pages
++;
2981 res_counter_uncharge(&memcg
->res
, nr_pages
* PAGE_SIZE
);
2983 res_counter_uncharge(&memcg
->memsw
, nr_pages
* PAGE_SIZE
);
2984 if (unlikely(batch
->memcg
!= memcg
))
2985 memcg_oom_recover(memcg
);
2989 * uncharge if !page_mapped(page)
2991 static struct mem_cgroup
*
2992 __mem_cgroup_uncharge_common(struct page
*page
, enum charge_type ctype
,
2995 struct mem_cgroup
*memcg
= NULL
;
2996 unsigned int nr_pages
= 1;
2997 struct page_cgroup
*pc
;
3000 if (mem_cgroup_disabled())
3003 VM_BUG_ON(PageSwapCache(page
));
3005 if (PageTransHuge(page
)) {
3006 nr_pages
<<= compound_order(page
);
3007 VM_BUG_ON(!PageTransHuge(page
));
3010 * Check if our page_cgroup is valid
3012 pc
= lookup_page_cgroup(page
);
3013 if (unlikely(!PageCgroupUsed(pc
)))
3016 lock_page_cgroup(pc
);
3018 memcg
= pc
->mem_cgroup
;
3020 if (!PageCgroupUsed(pc
))
3023 anon
= PageAnon(page
);
3026 case MEM_CGROUP_CHARGE_TYPE_ANON
:
3028 * Generally PageAnon tells if it's the anon statistics to be
3029 * updated; but sometimes e.g. mem_cgroup_uncharge_page() is
3030 * used before page reached the stage of being marked PageAnon.
3034 case MEM_CGROUP_CHARGE_TYPE_DROP
:
3035 /* See mem_cgroup_prepare_migration() */
3036 if (page_mapped(page
))
3039 * Pages under migration may not be uncharged. But
3040 * end_migration() /must/ be the one uncharging the
3041 * unused post-migration page and so it has to call
3042 * here with the migration bit still set. See the
3043 * res_counter handling below.
3045 if (!end_migration
&& PageCgroupMigration(pc
))
3048 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT
:
3049 if (!PageAnon(page
)) { /* Shared memory */
3050 if (page
->mapping
&& !page_is_file_cache(page
))
3052 } else if (page_mapped(page
)) /* Anon */
3059 mem_cgroup_charge_statistics(memcg
, anon
, -nr_pages
);
3061 ClearPageCgroupUsed(pc
);
3063 * pc->mem_cgroup is not cleared here. It will be accessed when it's
3064 * freed from LRU. This is safe because uncharged page is expected not
3065 * to be reused (freed soon). Exception is SwapCache, it's handled by
3066 * special functions.
3069 unlock_page_cgroup(pc
);
3071 * even after unlock, we have memcg->res.usage here and this memcg
3072 * will never be freed.
3074 memcg_check_events(memcg
, page
);
3075 if (do_swap_account
&& ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
) {
3076 mem_cgroup_swap_statistics(memcg
, true);
3077 mem_cgroup_get(memcg
);
3080 * Migration does not charge the res_counter for the
3081 * replacement page, so leave it alone when phasing out the
3082 * page that is unused after the migration.
3084 if (!end_migration
&& !mem_cgroup_is_root(memcg
))
3085 mem_cgroup_do_uncharge(memcg
, nr_pages
, ctype
);
3090 unlock_page_cgroup(pc
);
3094 void mem_cgroup_uncharge_page(struct page
*page
)
3097 if (page_mapped(page
))
3099 VM_BUG_ON(page
->mapping
&& !PageAnon(page
));
3100 if (PageSwapCache(page
))
3102 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_ANON
, false);
3105 void mem_cgroup_uncharge_cache_page(struct page
*page
)
3107 VM_BUG_ON(page_mapped(page
));
3108 VM_BUG_ON(page
->mapping
);
3109 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_CACHE
, false);
3113 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
3114 * In that cases, pages are freed continuously and we can expect pages
3115 * are in the same memcg. All these calls itself limits the number of
3116 * pages freed at once, then uncharge_start/end() is called properly.
3117 * This may be called prural(2) times in a context,
3120 void mem_cgroup_uncharge_start(void)
3122 current
->memcg_batch
.do_batch
++;
3123 /* We can do nest. */
3124 if (current
->memcg_batch
.do_batch
== 1) {
3125 current
->memcg_batch
.memcg
= NULL
;
3126 current
->memcg_batch
.nr_pages
= 0;
3127 current
->memcg_batch
.memsw_nr_pages
= 0;
3131 void mem_cgroup_uncharge_end(void)
3133 struct memcg_batch_info
*batch
= ¤t
->memcg_batch
;
3135 if (!batch
->do_batch
)
3139 if (batch
->do_batch
) /* If stacked, do nothing. */
3145 * This "batch->memcg" is valid without any css_get/put etc...
3146 * bacause we hide charges behind us.
3148 if (batch
->nr_pages
)
3149 res_counter_uncharge(&batch
->memcg
->res
,
3150 batch
->nr_pages
* PAGE_SIZE
);
3151 if (batch
->memsw_nr_pages
)
3152 res_counter_uncharge(&batch
->memcg
->memsw
,
3153 batch
->memsw_nr_pages
* PAGE_SIZE
);
3154 memcg_oom_recover(batch
->memcg
);
3155 /* forget this pointer (for sanity check) */
3156 batch
->memcg
= NULL
;
3161 * called after __delete_from_swap_cache() and drop "page" account.
3162 * memcg information is recorded to swap_cgroup of "ent"
3165 mem_cgroup_uncharge_swapcache(struct page
*page
, swp_entry_t ent
, bool swapout
)
3167 struct mem_cgroup
*memcg
;
3168 int ctype
= MEM_CGROUP_CHARGE_TYPE_SWAPOUT
;
3170 if (!swapout
) /* this was a swap cache but the swap is unused ! */
3171 ctype
= MEM_CGROUP_CHARGE_TYPE_DROP
;
3173 memcg
= __mem_cgroup_uncharge_common(page
, ctype
, false);
3176 * record memcg information, if swapout && memcg != NULL,
3177 * mem_cgroup_get() was called in uncharge().
3179 if (do_swap_account
&& swapout
&& memcg
)
3180 swap_cgroup_record(ent
, css_id(&memcg
->css
));
3184 #ifdef CONFIG_MEMCG_SWAP
3186 * called from swap_entry_free(). remove record in swap_cgroup and
3187 * uncharge "memsw" account.
3189 void mem_cgroup_uncharge_swap(swp_entry_t ent
)
3191 struct mem_cgroup
*memcg
;
3194 if (!do_swap_account
)
3197 id
= swap_cgroup_record(ent
, 0);
3199 memcg
= mem_cgroup_lookup(id
);
3202 * We uncharge this because swap is freed.
3203 * This memcg can be obsolete one. We avoid calling css_tryget
3205 if (!mem_cgroup_is_root(memcg
))
3206 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
3207 mem_cgroup_swap_statistics(memcg
, false);
3208 mem_cgroup_put(memcg
);
3214 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3215 * @entry: swap entry to be moved
3216 * @from: mem_cgroup which the entry is moved from
3217 * @to: mem_cgroup which the entry is moved to
3219 * It succeeds only when the swap_cgroup's record for this entry is the same
3220 * as the mem_cgroup's id of @from.
3222 * Returns 0 on success, -EINVAL on failure.
3224 * The caller must have charged to @to, IOW, called res_counter_charge() about
3225 * both res and memsw, and called css_get().
3227 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
3228 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
3230 unsigned short old_id
, new_id
;
3232 old_id
= css_id(&from
->css
);
3233 new_id
= css_id(&to
->css
);
3235 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
3236 mem_cgroup_swap_statistics(from
, false);
3237 mem_cgroup_swap_statistics(to
, true);
3239 * This function is only called from task migration context now.
3240 * It postpones res_counter and refcount handling till the end
3241 * of task migration(mem_cgroup_clear_mc()) for performance
3242 * improvement. But we cannot postpone mem_cgroup_get(to)
3243 * because if the process that has been moved to @to does
3244 * swap-in, the refcount of @to might be decreased to 0.
3252 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
3253 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
3260 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3263 void mem_cgroup_prepare_migration(struct page
*page
, struct page
*newpage
,
3264 struct mem_cgroup
**memcgp
)
3266 struct mem_cgroup
*memcg
= NULL
;
3267 struct page_cgroup
*pc
;
3268 enum charge_type ctype
;
3272 VM_BUG_ON(PageTransHuge(page
));
3273 if (mem_cgroup_disabled())
3276 pc
= lookup_page_cgroup(page
);
3277 lock_page_cgroup(pc
);
3278 if (PageCgroupUsed(pc
)) {
3279 memcg
= pc
->mem_cgroup
;
3280 css_get(&memcg
->css
);
3282 * At migrating an anonymous page, its mapcount goes down
3283 * to 0 and uncharge() will be called. But, even if it's fully
3284 * unmapped, migration may fail and this page has to be
3285 * charged again. We set MIGRATION flag here and delay uncharge
3286 * until end_migration() is called
3288 * Corner Case Thinking
3290 * When the old page was mapped as Anon and it's unmap-and-freed
3291 * while migration was ongoing.
3292 * If unmap finds the old page, uncharge() of it will be delayed
3293 * until end_migration(). If unmap finds a new page, it's
3294 * uncharged when it make mapcount to be 1->0. If unmap code
3295 * finds swap_migration_entry, the new page will not be mapped
3296 * and end_migration() will find it(mapcount==0).
3299 * When the old page was mapped but migraion fails, the kernel
3300 * remaps it. A charge for it is kept by MIGRATION flag even
3301 * if mapcount goes down to 0. We can do remap successfully
3302 * without charging it again.
3305 * The "old" page is under lock_page() until the end of
3306 * migration, so, the old page itself will not be swapped-out.
3307 * If the new page is swapped out before end_migraton, our
3308 * hook to usual swap-out path will catch the event.
3311 SetPageCgroupMigration(pc
);
3313 unlock_page_cgroup(pc
);
3315 * If the page is not charged at this point,
3323 * We charge new page before it's used/mapped. So, even if unlock_page()
3324 * is called before end_migration, we can catch all events on this new
3325 * page. In the case new page is migrated but not remapped, new page's
3326 * mapcount will be finally 0 and we call uncharge in end_migration().
3329 ctype
= MEM_CGROUP_CHARGE_TYPE_ANON
;
3331 ctype
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3333 * The page is committed to the memcg, but it's not actually
3334 * charged to the res_counter since we plan on replacing the
3335 * old one and only one page is going to be left afterwards.
3337 __mem_cgroup_commit_charge(memcg
, newpage
, 1, ctype
, false);
3340 /* remove redundant charge if migration failed*/
3341 void mem_cgroup_end_migration(struct mem_cgroup
*memcg
,
3342 struct page
*oldpage
, struct page
*newpage
, bool migration_ok
)
3344 struct page
*used
, *unused
;
3345 struct page_cgroup
*pc
;
3350 /* blocks rmdir() */
3351 cgroup_exclude_rmdir(&memcg
->css
);
3352 if (!migration_ok
) {
3359 anon
= PageAnon(used
);
3360 __mem_cgroup_uncharge_common(unused
,
3361 anon
? MEM_CGROUP_CHARGE_TYPE_ANON
3362 : MEM_CGROUP_CHARGE_TYPE_CACHE
,
3364 css_put(&memcg
->css
);
3366 * We disallowed uncharge of pages under migration because mapcount
3367 * of the page goes down to zero, temporarly.
3368 * Clear the flag and check the page should be charged.
3370 pc
= lookup_page_cgroup(oldpage
);
3371 lock_page_cgroup(pc
);
3372 ClearPageCgroupMigration(pc
);
3373 unlock_page_cgroup(pc
);
3376 * If a page is a file cache, radix-tree replacement is very atomic
3377 * and we can skip this check. When it was an Anon page, its mapcount
3378 * goes down to 0. But because we added MIGRATION flage, it's not
3379 * uncharged yet. There are several case but page->mapcount check
3380 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3381 * check. (see prepare_charge() also)
3384 mem_cgroup_uncharge_page(used
);
3386 * At migration, we may charge account against cgroup which has no
3388 * So, rmdir()->pre_destroy() can be called while we do this charge.
3389 * In that case, we need to call pre_destroy() again. check it here.
3391 cgroup_release_and_wakeup_rmdir(&memcg
->css
);
3395 * At replace page cache, newpage is not under any memcg but it's on
3396 * LRU. So, this function doesn't touch res_counter but handles LRU
3397 * in correct way. Both pages are locked so we cannot race with uncharge.
3399 void mem_cgroup_replace_page_cache(struct page
*oldpage
,
3400 struct page
*newpage
)
3402 struct mem_cgroup
*memcg
= NULL
;
3403 struct page_cgroup
*pc
;
3404 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3406 if (mem_cgroup_disabled())
3409 pc
= lookup_page_cgroup(oldpage
);
3410 /* fix accounting on old pages */
3411 lock_page_cgroup(pc
);
3412 if (PageCgroupUsed(pc
)) {
3413 memcg
= pc
->mem_cgroup
;
3414 mem_cgroup_charge_statistics(memcg
, false, -1);
3415 ClearPageCgroupUsed(pc
);
3417 unlock_page_cgroup(pc
);
3420 * When called from shmem_replace_page(), in some cases the
3421 * oldpage has already been charged, and in some cases not.
3426 * Even if newpage->mapping was NULL before starting replacement,
3427 * the newpage may be on LRU(or pagevec for LRU) already. We lock
3428 * LRU while we overwrite pc->mem_cgroup.
3430 __mem_cgroup_commit_charge(memcg
, newpage
, 1, type
, true);
3433 #ifdef CONFIG_DEBUG_VM
3434 static struct page_cgroup
*lookup_page_cgroup_used(struct page
*page
)
3436 struct page_cgroup
*pc
;
3438 pc
= lookup_page_cgroup(page
);
3440 * Can be NULL while feeding pages into the page allocator for
3441 * the first time, i.e. during boot or memory hotplug;
3442 * or when mem_cgroup_disabled().
3444 if (likely(pc
) && PageCgroupUsed(pc
))
3449 bool mem_cgroup_bad_page_check(struct page
*page
)
3451 if (mem_cgroup_disabled())
3454 return lookup_page_cgroup_used(page
) != NULL
;
3457 void mem_cgroup_print_bad_page(struct page
*page
)
3459 struct page_cgroup
*pc
;
3461 pc
= lookup_page_cgroup_used(page
);
3463 printk(KERN_ALERT
"pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
3464 pc
, pc
->flags
, pc
->mem_cgroup
);
3469 static DEFINE_MUTEX(set_limit_mutex
);
3471 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
3472 unsigned long long val
)
3475 u64 memswlimit
, memlimit
;
3477 int children
= mem_cgroup_count_children(memcg
);
3478 u64 curusage
, oldusage
;
3482 * For keeping hierarchical_reclaim simple, how long we should retry
3483 * is depends on callers. We set our retry-count to be function
3484 * of # of children which we should visit in this loop.
3486 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
* children
;
3488 oldusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3491 while (retry_count
) {
3492 if (signal_pending(current
)) {
3497 * Rather than hide all in some function, I do this in
3498 * open coded manner. You see what this really does.
3499 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
3501 mutex_lock(&set_limit_mutex
);
3502 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3503 if (memswlimit
< val
) {
3505 mutex_unlock(&set_limit_mutex
);
3509 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3513 ret
= res_counter_set_limit(&memcg
->res
, val
);
3515 if (memswlimit
== val
)
3516 memcg
->memsw_is_minimum
= true;
3518 memcg
->memsw_is_minimum
= false;
3520 mutex_unlock(&set_limit_mutex
);
3525 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3526 MEM_CGROUP_RECLAIM_SHRINK
);
3527 curusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3528 /* Usage is reduced ? */
3529 if (curusage
>= oldusage
)
3532 oldusage
= curusage
;
3534 if (!ret
&& enlarge
)
3535 memcg_oom_recover(memcg
);
3540 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
3541 unsigned long long val
)
3544 u64 memlimit
, memswlimit
, oldusage
, curusage
;
3545 int children
= mem_cgroup_count_children(memcg
);
3549 /* see mem_cgroup_resize_res_limit */
3550 retry_count
= children
* MEM_CGROUP_RECLAIM_RETRIES
;
3551 oldusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3552 while (retry_count
) {
3553 if (signal_pending(current
)) {
3558 * Rather than hide all in some function, I do this in
3559 * open coded manner. You see what this really does.
3560 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
3562 mutex_lock(&set_limit_mutex
);
3563 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3564 if (memlimit
> val
) {
3566 mutex_unlock(&set_limit_mutex
);
3569 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3570 if (memswlimit
< val
)
3572 ret
= res_counter_set_limit(&memcg
->memsw
, val
);
3574 if (memlimit
== val
)
3575 memcg
->memsw_is_minimum
= true;
3577 memcg
->memsw_is_minimum
= false;
3579 mutex_unlock(&set_limit_mutex
);
3584 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3585 MEM_CGROUP_RECLAIM_NOSWAP
|
3586 MEM_CGROUP_RECLAIM_SHRINK
);
3587 curusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3588 /* Usage is reduced ? */
3589 if (curusage
>= oldusage
)
3592 oldusage
= curusage
;
3594 if (!ret
&& enlarge
)
3595 memcg_oom_recover(memcg
);
3599 unsigned long mem_cgroup_soft_limit_reclaim(struct zone
*zone
, int order
,
3601 unsigned long *total_scanned
)
3603 unsigned long nr_reclaimed
= 0;
3604 struct mem_cgroup_per_zone
*mz
, *next_mz
= NULL
;
3605 unsigned long reclaimed
;
3607 struct mem_cgroup_tree_per_zone
*mctz
;
3608 unsigned long long excess
;
3609 unsigned long nr_scanned
;
3614 mctz
= soft_limit_tree_node_zone(zone_to_nid(zone
), zone_idx(zone
));
3616 * This loop can run a while, specially if mem_cgroup's continuously
3617 * keep exceeding their soft limit and putting the system under
3624 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
3629 reclaimed
= mem_cgroup_soft_reclaim(mz
->memcg
, zone
,
3630 gfp_mask
, &nr_scanned
);
3631 nr_reclaimed
+= reclaimed
;
3632 *total_scanned
+= nr_scanned
;
3633 spin_lock(&mctz
->lock
);
3636 * If we failed to reclaim anything from this memory cgroup
3637 * it is time to move on to the next cgroup
3643 * Loop until we find yet another one.
3645 * By the time we get the soft_limit lock
3646 * again, someone might have aded the
3647 * group back on the RB tree. Iterate to
3648 * make sure we get a different mem.
3649 * mem_cgroup_largest_soft_limit_node returns
3650 * NULL if no other cgroup is present on
3654 __mem_cgroup_largest_soft_limit_node(mctz
);
3656 css_put(&next_mz
->memcg
->css
);
3657 else /* next_mz == NULL or other memcg */
3661 __mem_cgroup_remove_exceeded(mz
->memcg
, mz
, mctz
);
3662 excess
= res_counter_soft_limit_excess(&mz
->memcg
->res
);
3664 * One school of thought says that we should not add
3665 * back the node to the tree if reclaim returns 0.
3666 * But our reclaim could return 0, simply because due
3667 * to priority we are exposing a smaller subset of
3668 * memory to reclaim from. Consider this as a longer
3671 /* If excess == 0, no tree ops */
3672 __mem_cgroup_insert_exceeded(mz
->memcg
, mz
, mctz
, excess
);
3673 spin_unlock(&mctz
->lock
);
3674 css_put(&mz
->memcg
->css
);
3677 * Could not reclaim anything and there are no more
3678 * mem cgroups to try or we seem to be looping without
3679 * reclaiming anything.
3681 if (!nr_reclaimed
&&
3683 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
3685 } while (!nr_reclaimed
);
3687 css_put(&next_mz
->memcg
->css
);
3688 return nr_reclaimed
;
3692 * Traverse a specified page_cgroup list and try to drop them all. This doesn't
3693 * reclaim the pages page themselves - it just removes the page_cgroups.
3694 * Returns true if some page_cgroups were not freed, indicating that the caller
3695 * must retry this operation.
3697 static bool mem_cgroup_force_empty_list(struct mem_cgroup
*memcg
,
3698 int node
, int zid
, enum lru_list lru
)
3700 struct mem_cgroup_per_zone
*mz
;
3701 unsigned long flags
, loop
;
3702 struct list_head
*list
;
3706 zone
= &NODE_DATA(node
)->node_zones
[zid
];
3707 mz
= mem_cgroup_zoneinfo(memcg
, node
, zid
);
3708 list
= &mz
->lruvec
.lists
[lru
];
3710 loop
= mz
->lru_size
[lru
];
3711 /* give some margin against EBUSY etc...*/
3715 struct page_cgroup
*pc
;
3718 spin_lock_irqsave(&zone
->lru_lock
, flags
);
3719 if (list_empty(list
)) {
3720 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3723 page
= list_entry(list
->prev
, struct page
, lru
);
3725 list_move(&page
->lru
, list
);
3727 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3730 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3732 pc
= lookup_page_cgroup(page
);
3734 if (mem_cgroup_move_parent(page
, pc
, memcg
)) {
3735 /* found lock contention or "pc" is obsolete. */
3741 return !list_empty(list
);
3745 * make mem_cgroup's charge to be 0 if there is no task.
3746 * This enables deleting this mem_cgroup.
3748 static int mem_cgroup_force_empty(struct mem_cgroup
*memcg
, bool free_all
)
3751 int node
, zid
, shrink
;
3752 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
3753 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
3755 css_get(&memcg
->css
);
3758 /* should free all ? */
3764 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
))
3766 /* This is for making all *used* pages to be on LRU. */
3767 lru_add_drain_all();
3768 drain_all_stock_sync(memcg
);
3770 mem_cgroup_start_move(memcg
);
3771 for_each_node_state(node
, N_HIGH_MEMORY
) {
3772 for (zid
= 0; !ret
&& zid
< MAX_NR_ZONES
; zid
++) {
3775 ret
= mem_cgroup_force_empty_list(memcg
,
3784 mem_cgroup_end_move(memcg
);
3785 memcg_oom_recover(memcg
);
3787 /* "ret" should also be checked to ensure all lists are empty. */
3788 } while (res_counter_read_u64(&memcg
->res
, RES_USAGE
) > 0 || ret
);
3790 css_put(&memcg
->css
);
3794 /* returns EBUSY if there is a task or if we come here twice. */
3795 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
) || shrink
) {
3799 /* we call try-to-free pages for make this cgroup empty */
3800 lru_add_drain_all();
3801 /* try to free all pages in this cgroup */
3803 while (nr_retries
&& res_counter_read_u64(&memcg
->res
, RES_USAGE
) > 0) {
3806 if (signal_pending(current
)) {
3810 progress
= try_to_free_mem_cgroup_pages(memcg
, GFP_KERNEL
,
3814 /* maybe some writeback is necessary */
3815 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3820 /* try move_account...there may be some *locked* pages. */
3824 static int mem_cgroup_force_empty_write(struct cgroup
*cont
, unsigned int event
)
3826 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont
), true);
3830 static u64
mem_cgroup_hierarchy_read(struct cgroup
*cont
, struct cftype
*cft
)
3832 return mem_cgroup_from_cont(cont
)->use_hierarchy
;
3835 static int mem_cgroup_hierarchy_write(struct cgroup
*cont
, struct cftype
*cft
,
3839 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3840 struct cgroup
*parent
= cont
->parent
;
3841 struct mem_cgroup
*parent_memcg
= NULL
;
3844 parent_memcg
= mem_cgroup_from_cont(parent
);
3848 if (memcg
->use_hierarchy
== val
)
3852 * If parent's use_hierarchy is set, we can't make any modifications
3853 * in the child subtrees. If it is unset, then the change can
3854 * occur, provided the current cgroup has no children.
3856 * For the root cgroup, parent_mem is NULL, we allow value to be
3857 * set if there are no children.
3859 if ((!parent_memcg
|| !parent_memcg
->use_hierarchy
) &&
3860 (val
== 1 || val
== 0)) {
3861 if (list_empty(&cont
->children
))
3862 memcg
->use_hierarchy
= val
;
3875 static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup
*memcg
,
3876 enum mem_cgroup_stat_index idx
)
3878 struct mem_cgroup
*iter
;
3881 /* Per-cpu values can be negative, use a signed accumulator */
3882 for_each_mem_cgroup_tree(iter
, memcg
)
3883 val
+= mem_cgroup_read_stat(iter
, idx
);
3885 if (val
< 0) /* race ? */
3890 static inline u64
mem_cgroup_usage(struct mem_cgroup
*memcg
, bool swap
)
3894 if (!mem_cgroup_is_root(memcg
)) {
3896 return res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3898 return res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3901 val
= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_CACHE
);
3902 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_RSS
);
3905 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_SWAP
);
3907 return val
<< PAGE_SHIFT
;
3910 static ssize_t
mem_cgroup_read(struct cgroup
*cont
, struct cftype
*cft
,
3911 struct file
*file
, char __user
*buf
,
3912 size_t nbytes
, loff_t
*ppos
)
3914 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3917 int type
, name
, len
;
3919 type
= MEMFILE_TYPE(cft
->private);
3920 name
= MEMFILE_ATTR(cft
->private);
3922 if (!do_swap_account
&& type
== _MEMSWAP
)
3927 if (name
== RES_USAGE
)
3928 val
= mem_cgroup_usage(memcg
, false);
3930 val
= res_counter_read_u64(&memcg
->res
, name
);
3933 if (name
== RES_USAGE
)
3934 val
= mem_cgroup_usage(memcg
, true);
3936 val
= res_counter_read_u64(&memcg
->memsw
, name
);
3942 len
= scnprintf(str
, sizeof(str
), "%llu\n", (unsigned long long)val
);
3943 return simple_read_from_buffer(buf
, nbytes
, ppos
, str
, len
);
3946 * The user of this function is...
3949 static int mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
3952 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3954 unsigned long long val
;
3957 type
= MEMFILE_TYPE(cft
->private);
3958 name
= MEMFILE_ATTR(cft
->private);
3960 if (!do_swap_account
&& type
== _MEMSWAP
)
3965 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
3969 /* This function does all necessary parse...reuse it */
3970 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3974 ret
= mem_cgroup_resize_limit(memcg
, val
);
3976 ret
= mem_cgroup_resize_memsw_limit(memcg
, val
);
3978 case RES_SOFT_LIMIT
:
3979 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3983 * For memsw, soft limits are hard to implement in terms
3984 * of semantics, for now, we support soft limits for
3985 * control without swap
3988 ret
= res_counter_set_soft_limit(&memcg
->res
, val
);
3993 ret
= -EINVAL
; /* should be BUG() ? */
3999 static void memcg_get_hierarchical_limit(struct mem_cgroup
*memcg
,
4000 unsigned long long *mem_limit
, unsigned long long *memsw_limit
)
4002 struct cgroup
*cgroup
;
4003 unsigned long long min_limit
, min_memsw_limit
, tmp
;
4005 min_limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
4006 min_memsw_limit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
4007 cgroup
= memcg
->css
.cgroup
;
4008 if (!memcg
->use_hierarchy
)
4011 while (cgroup
->parent
) {
4012 cgroup
= cgroup
->parent
;
4013 memcg
= mem_cgroup_from_cont(cgroup
);
4014 if (!memcg
->use_hierarchy
)
4016 tmp
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
4017 min_limit
= min(min_limit
, tmp
);
4018 tmp
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
4019 min_memsw_limit
= min(min_memsw_limit
, tmp
);
4022 *mem_limit
= min_limit
;
4023 *memsw_limit
= min_memsw_limit
;
4026 static int mem_cgroup_reset(struct cgroup
*cont
, unsigned int event
)
4028 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4031 type
= MEMFILE_TYPE(event
);
4032 name
= MEMFILE_ATTR(event
);
4034 if (!do_swap_account
&& type
== _MEMSWAP
)
4040 res_counter_reset_max(&memcg
->res
);
4042 res_counter_reset_max(&memcg
->memsw
);
4046 res_counter_reset_failcnt(&memcg
->res
);
4048 res_counter_reset_failcnt(&memcg
->memsw
);
4055 static u64
mem_cgroup_move_charge_read(struct cgroup
*cgrp
,
4058 return mem_cgroup_from_cont(cgrp
)->move_charge_at_immigrate
;
4062 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4063 struct cftype
*cft
, u64 val
)
4065 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4067 if (val
>= (1 << NR_MOVE_TYPE
))
4070 * We check this value several times in both in can_attach() and
4071 * attach(), so we need cgroup lock to prevent this value from being
4075 memcg
->move_charge_at_immigrate
= val
;
4081 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4082 struct cftype
*cft
, u64 val
)
4089 static int memcg_numa_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
4093 unsigned long total_nr
, file_nr
, anon_nr
, unevictable_nr
;
4094 unsigned long node_nr
;
4095 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4097 total_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL
);
4098 seq_printf(m
, "total=%lu", total_nr
);
4099 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4100 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL
);
4101 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4105 file_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL_FILE
);
4106 seq_printf(m
, "file=%lu", file_nr
);
4107 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4108 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4110 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4114 anon_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL_ANON
);
4115 seq_printf(m
, "anon=%lu", anon_nr
);
4116 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4117 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4119 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4123 unevictable_nr
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_UNEVICTABLE
));
4124 seq_printf(m
, "unevictable=%lu", unevictable_nr
);
4125 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4126 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4127 BIT(LRU_UNEVICTABLE
));
4128 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4133 #endif /* CONFIG_NUMA */
4135 static const char * const mem_cgroup_lru_names
[] = {
4143 static inline void mem_cgroup_lru_names_not_uptodate(void)
4145 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names
) != NR_LRU_LISTS
);
4148 static int memcg_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
4151 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4152 struct mem_cgroup
*mi
;
4155 for (i
= 0; i
< MEM_CGROUP_STAT_NSTATS
; i
++) {
4156 if (i
== MEM_CGROUP_STAT_SWAP
&& !do_swap_account
)
4158 seq_printf(m
, "%s %ld\n", mem_cgroup_stat_names
[i
],
4159 mem_cgroup_read_stat(memcg
, i
) * PAGE_SIZE
);
4162 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++)
4163 seq_printf(m
, "%s %lu\n", mem_cgroup_events_names
[i
],
4164 mem_cgroup_read_events(memcg
, i
));
4166 for (i
= 0; i
< NR_LRU_LISTS
; i
++)
4167 seq_printf(m
, "%s %lu\n", mem_cgroup_lru_names
[i
],
4168 mem_cgroup_nr_lru_pages(memcg
, BIT(i
)) * PAGE_SIZE
);
4170 /* Hierarchical information */
4172 unsigned long long limit
, memsw_limit
;
4173 memcg_get_hierarchical_limit(memcg
, &limit
, &memsw_limit
);
4174 seq_printf(m
, "hierarchical_memory_limit %llu\n", limit
);
4175 if (do_swap_account
)
4176 seq_printf(m
, "hierarchical_memsw_limit %llu\n",
4180 for (i
= 0; i
< MEM_CGROUP_STAT_NSTATS
; i
++) {
4183 if (i
== MEM_CGROUP_STAT_SWAP
&& !do_swap_account
)
4185 for_each_mem_cgroup_tree(mi
, memcg
)
4186 val
+= mem_cgroup_read_stat(mi
, i
) * PAGE_SIZE
;
4187 seq_printf(m
, "total_%s %lld\n", mem_cgroup_stat_names
[i
], val
);
4190 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
4191 unsigned long long val
= 0;
4193 for_each_mem_cgroup_tree(mi
, memcg
)
4194 val
+= mem_cgroup_read_events(mi
, i
);
4195 seq_printf(m
, "total_%s %llu\n",
4196 mem_cgroup_events_names
[i
], val
);
4199 for (i
= 0; i
< NR_LRU_LISTS
; i
++) {
4200 unsigned long long val
= 0;
4202 for_each_mem_cgroup_tree(mi
, memcg
)
4203 val
+= mem_cgroup_nr_lru_pages(mi
, BIT(i
)) * PAGE_SIZE
;
4204 seq_printf(m
, "total_%s %llu\n", mem_cgroup_lru_names
[i
], val
);
4207 #ifdef CONFIG_DEBUG_VM
4210 struct mem_cgroup_per_zone
*mz
;
4211 struct zone_reclaim_stat
*rstat
;
4212 unsigned long recent_rotated
[2] = {0, 0};
4213 unsigned long recent_scanned
[2] = {0, 0};
4215 for_each_online_node(nid
)
4216 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
4217 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
4218 rstat
= &mz
->lruvec
.reclaim_stat
;
4220 recent_rotated
[0] += rstat
->recent_rotated
[0];
4221 recent_rotated
[1] += rstat
->recent_rotated
[1];
4222 recent_scanned
[0] += rstat
->recent_scanned
[0];
4223 recent_scanned
[1] += rstat
->recent_scanned
[1];
4225 seq_printf(m
, "recent_rotated_anon %lu\n", recent_rotated
[0]);
4226 seq_printf(m
, "recent_rotated_file %lu\n", recent_rotated
[1]);
4227 seq_printf(m
, "recent_scanned_anon %lu\n", recent_scanned
[0]);
4228 seq_printf(m
, "recent_scanned_file %lu\n", recent_scanned
[1]);
4235 static u64
mem_cgroup_swappiness_read(struct cgroup
*cgrp
, struct cftype
*cft
)
4237 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4239 return mem_cgroup_swappiness(memcg
);
4242 static int mem_cgroup_swappiness_write(struct cgroup
*cgrp
, struct cftype
*cft
,
4245 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4246 struct mem_cgroup
*parent
;
4251 if (cgrp
->parent
== NULL
)
4254 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4258 /* If under hierarchy, only empty-root can set this value */
4259 if ((parent
->use_hierarchy
) ||
4260 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4265 memcg
->swappiness
= val
;
4272 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
4274 struct mem_cgroup_threshold_ary
*t
;
4280 t
= rcu_dereference(memcg
->thresholds
.primary
);
4282 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
4287 usage
= mem_cgroup_usage(memcg
, swap
);
4290 * current_threshold points to threshold just below or equal to usage.
4291 * If it's not true, a threshold was crossed after last
4292 * call of __mem_cgroup_threshold().
4294 i
= t
->current_threshold
;
4297 * Iterate backward over array of thresholds starting from
4298 * current_threshold and check if a threshold is crossed.
4299 * If none of thresholds below usage is crossed, we read
4300 * only one element of the array here.
4302 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
4303 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4305 /* i = current_threshold + 1 */
4309 * Iterate forward over array of thresholds starting from
4310 * current_threshold+1 and check if a threshold is crossed.
4311 * If none of thresholds above usage is crossed, we read
4312 * only one element of the array here.
4314 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
4315 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4317 /* Update current_threshold */
4318 t
->current_threshold
= i
- 1;
4323 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
4326 __mem_cgroup_threshold(memcg
, false);
4327 if (do_swap_account
)
4328 __mem_cgroup_threshold(memcg
, true);
4330 memcg
= parent_mem_cgroup(memcg
);
4334 static int compare_thresholds(const void *a
, const void *b
)
4336 const struct mem_cgroup_threshold
*_a
= a
;
4337 const struct mem_cgroup_threshold
*_b
= b
;
4339 return _a
->threshold
- _b
->threshold
;
4342 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*memcg
)
4344 struct mem_cgroup_eventfd_list
*ev
;
4346 list_for_each_entry(ev
, &memcg
->oom_notify
, list
)
4347 eventfd_signal(ev
->eventfd
, 1);
4351 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
)
4353 struct mem_cgroup
*iter
;
4355 for_each_mem_cgroup_tree(iter
, memcg
)
4356 mem_cgroup_oom_notify_cb(iter
);
4359 static int mem_cgroup_usage_register_event(struct cgroup
*cgrp
,
4360 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4362 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4363 struct mem_cgroup_thresholds
*thresholds
;
4364 struct mem_cgroup_threshold_ary
*new;
4365 int type
= MEMFILE_TYPE(cft
->private);
4366 u64 threshold
, usage
;
4369 ret
= res_counter_memparse_write_strategy(args
, &threshold
);
4373 mutex_lock(&memcg
->thresholds_lock
);
4376 thresholds
= &memcg
->thresholds
;
4377 else if (type
== _MEMSWAP
)
4378 thresholds
= &memcg
->memsw_thresholds
;
4382 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4384 /* Check if a threshold crossed before adding a new one */
4385 if (thresholds
->primary
)
4386 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4388 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
4390 /* Allocate memory for new array of thresholds */
4391 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
4399 /* Copy thresholds (if any) to new array */
4400 if (thresholds
->primary
) {
4401 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
4402 sizeof(struct mem_cgroup_threshold
));
4405 /* Add new threshold */
4406 new->entries
[size
- 1].eventfd
= eventfd
;
4407 new->entries
[size
- 1].threshold
= threshold
;
4409 /* Sort thresholds. Registering of new threshold isn't time-critical */
4410 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
4411 compare_thresholds
, NULL
);
4413 /* Find current threshold */
4414 new->current_threshold
= -1;
4415 for (i
= 0; i
< size
; i
++) {
4416 if (new->entries
[i
].threshold
<= usage
) {
4418 * new->current_threshold will not be used until
4419 * rcu_assign_pointer(), so it's safe to increment
4422 ++new->current_threshold
;
4427 /* Free old spare buffer and save old primary buffer as spare */
4428 kfree(thresholds
->spare
);
4429 thresholds
->spare
= thresholds
->primary
;
4431 rcu_assign_pointer(thresholds
->primary
, new);
4433 /* To be sure that nobody uses thresholds */
4437 mutex_unlock(&memcg
->thresholds_lock
);
4442 static void mem_cgroup_usage_unregister_event(struct cgroup
*cgrp
,
4443 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4445 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4446 struct mem_cgroup_thresholds
*thresholds
;
4447 struct mem_cgroup_threshold_ary
*new;
4448 int type
= MEMFILE_TYPE(cft
->private);
4452 mutex_lock(&memcg
->thresholds_lock
);
4454 thresholds
= &memcg
->thresholds
;
4455 else if (type
== _MEMSWAP
)
4456 thresholds
= &memcg
->memsw_thresholds
;
4460 if (!thresholds
->primary
)
4463 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4465 /* Check if a threshold crossed before removing */
4466 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4468 /* Calculate new number of threshold */
4470 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
4471 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
4475 new = thresholds
->spare
;
4477 /* Set thresholds array to NULL if we don't have thresholds */
4486 /* Copy thresholds and find current threshold */
4487 new->current_threshold
= -1;
4488 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
4489 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
4492 new->entries
[j
] = thresholds
->primary
->entries
[i
];
4493 if (new->entries
[j
].threshold
<= usage
) {
4495 * new->current_threshold will not be used
4496 * until rcu_assign_pointer(), so it's safe to increment
4499 ++new->current_threshold
;
4505 /* Swap primary and spare array */
4506 thresholds
->spare
= thresholds
->primary
;
4507 /* If all events are unregistered, free the spare array */
4509 kfree(thresholds
->spare
);
4510 thresholds
->spare
= NULL
;
4513 rcu_assign_pointer(thresholds
->primary
, new);
4515 /* To be sure that nobody uses thresholds */
4518 mutex_unlock(&memcg
->thresholds_lock
);
4521 static int mem_cgroup_oom_register_event(struct cgroup
*cgrp
,
4522 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4524 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4525 struct mem_cgroup_eventfd_list
*event
;
4526 int type
= MEMFILE_TYPE(cft
->private);
4528 BUG_ON(type
!= _OOM_TYPE
);
4529 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
4533 spin_lock(&memcg_oom_lock
);
4535 event
->eventfd
= eventfd
;
4536 list_add(&event
->list
, &memcg
->oom_notify
);
4538 /* already in OOM ? */
4539 if (atomic_read(&memcg
->under_oom
))
4540 eventfd_signal(eventfd
, 1);
4541 spin_unlock(&memcg_oom_lock
);
4546 static void mem_cgroup_oom_unregister_event(struct cgroup
*cgrp
,
4547 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4549 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4550 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
4551 int type
= MEMFILE_TYPE(cft
->private);
4553 BUG_ON(type
!= _OOM_TYPE
);
4555 spin_lock(&memcg_oom_lock
);
4557 list_for_each_entry_safe(ev
, tmp
, &memcg
->oom_notify
, list
) {
4558 if (ev
->eventfd
== eventfd
) {
4559 list_del(&ev
->list
);
4564 spin_unlock(&memcg_oom_lock
);
4567 static int mem_cgroup_oom_control_read(struct cgroup
*cgrp
,
4568 struct cftype
*cft
, struct cgroup_map_cb
*cb
)
4570 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4572 cb
->fill(cb
, "oom_kill_disable", memcg
->oom_kill_disable
);
4574 if (atomic_read(&memcg
->under_oom
))
4575 cb
->fill(cb
, "under_oom", 1);
4577 cb
->fill(cb
, "under_oom", 0);
4581 static int mem_cgroup_oom_control_write(struct cgroup
*cgrp
,
4582 struct cftype
*cft
, u64 val
)
4584 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4585 struct mem_cgroup
*parent
;
4587 /* cannot set to root cgroup and only 0 and 1 are allowed */
4588 if (!cgrp
->parent
|| !((val
== 0) || (val
== 1)))
4591 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4594 /* oom-kill-disable is a flag for subhierarchy. */
4595 if ((parent
->use_hierarchy
) ||
4596 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4600 memcg
->oom_kill_disable
= val
;
4602 memcg_oom_recover(memcg
);
4607 #ifdef CONFIG_MEMCG_KMEM
4608 static int memcg_init_kmem(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
4610 return mem_cgroup_sockets_init(memcg
, ss
);
4613 static void kmem_cgroup_destroy(struct mem_cgroup
*memcg
)
4615 mem_cgroup_sockets_destroy(memcg
);
4618 static int memcg_init_kmem(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
4623 static void kmem_cgroup_destroy(struct mem_cgroup
*memcg
)
4628 static struct cftype mem_cgroup_files
[] = {
4630 .name
= "usage_in_bytes",
4631 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
4632 .read
= mem_cgroup_read
,
4633 .register_event
= mem_cgroup_usage_register_event
,
4634 .unregister_event
= mem_cgroup_usage_unregister_event
,
4637 .name
= "max_usage_in_bytes",
4638 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
4639 .trigger
= mem_cgroup_reset
,
4640 .read
= mem_cgroup_read
,
4643 .name
= "limit_in_bytes",
4644 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
4645 .write_string
= mem_cgroup_write
,
4646 .read
= mem_cgroup_read
,
4649 .name
= "soft_limit_in_bytes",
4650 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
4651 .write_string
= mem_cgroup_write
,
4652 .read
= mem_cgroup_read
,
4656 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
4657 .trigger
= mem_cgroup_reset
,
4658 .read
= mem_cgroup_read
,
4662 .read_seq_string
= memcg_stat_show
,
4665 .name
= "force_empty",
4666 .trigger
= mem_cgroup_force_empty_write
,
4669 .name
= "use_hierarchy",
4670 .write_u64
= mem_cgroup_hierarchy_write
,
4671 .read_u64
= mem_cgroup_hierarchy_read
,
4674 .name
= "swappiness",
4675 .read_u64
= mem_cgroup_swappiness_read
,
4676 .write_u64
= mem_cgroup_swappiness_write
,
4679 .name
= "move_charge_at_immigrate",
4680 .read_u64
= mem_cgroup_move_charge_read
,
4681 .write_u64
= mem_cgroup_move_charge_write
,
4684 .name
= "oom_control",
4685 .read_map
= mem_cgroup_oom_control_read
,
4686 .write_u64
= mem_cgroup_oom_control_write
,
4687 .register_event
= mem_cgroup_oom_register_event
,
4688 .unregister_event
= mem_cgroup_oom_unregister_event
,
4689 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
4693 .name
= "numa_stat",
4694 .read_seq_string
= memcg_numa_stat_show
,
4697 #ifdef CONFIG_MEMCG_SWAP
4699 .name
= "memsw.usage_in_bytes",
4700 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
4701 .read
= mem_cgroup_read
,
4702 .register_event
= mem_cgroup_usage_register_event
,
4703 .unregister_event
= mem_cgroup_usage_unregister_event
,
4706 .name
= "memsw.max_usage_in_bytes",
4707 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
4708 .trigger
= mem_cgroup_reset
,
4709 .read
= mem_cgroup_read
,
4712 .name
= "memsw.limit_in_bytes",
4713 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
4714 .write_string
= mem_cgroup_write
,
4715 .read
= mem_cgroup_read
,
4718 .name
= "memsw.failcnt",
4719 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
4720 .trigger
= mem_cgroup_reset
,
4721 .read
= mem_cgroup_read
,
4724 { }, /* terminate */
4727 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4729 struct mem_cgroup_per_node
*pn
;
4730 struct mem_cgroup_per_zone
*mz
;
4731 int zone
, tmp
= node
;
4733 * This routine is called against possible nodes.
4734 * But it's BUG to call kmalloc() against offline node.
4736 * TODO: this routine can waste much memory for nodes which will
4737 * never be onlined. It's better to use memory hotplug callback
4740 if (!node_state(node
, N_NORMAL_MEMORY
))
4742 pn
= kzalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
4746 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4747 mz
= &pn
->zoneinfo
[zone
];
4748 lruvec_init(&mz
->lruvec
, &NODE_DATA(node
)->node_zones
[zone
]);
4749 mz
->usage_in_excess
= 0;
4750 mz
->on_tree
= false;
4753 memcg
->info
.nodeinfo
[node
] = pn
;
4757 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4759 kfree(memcg
->info
.nodeinfo
[node
]);
4762 static struct mem_cgroup
*mem_cgroup_alloc(void)
4764 struct mem_cgroup
*memcg
;
4765 int size
= sizeof(struct mem_cgroup
);
4767 /* Can be very big if MAX_NUMNODES is very big */
4768 if (size
< PAGE_SIZE
)
4769 memcg
= kzalloc(size
, GFP_KERNEL
);
4771 memcg
= vzalloc(size
);
4776 memcg
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4779 spin_lock_init(&memcg
->pcp_counter_lock
);
4783 if (size
< PAGE_SIZE
)
4791 * Helpers for freeing a kmalloc()ed/vzalloc()ed mem_cgroup by RCU,
4792 * but in process context. The work_freeing structure is overlaid
4793 * on the rcu_freeing structure, which itself is overlaid on memsw.
4795 static void free_work(struct work_struct
*work
)
4797 struct mem_cgroup
*memcg
;
4798 int size
= sizeof(struct mem_cgroup
);
4800 memcg
= container_of(work
, struct mem_cgroup
, work_freeing
);
4802 * We need to make sure that (at least for now), the jump label
4803 * destruction code runs outside of the cgroup lock. This is because
4804 * get_online_cpus(), which is called from the static_branch update,
4805 * can't be called inside the cgroup_lock. cpusets are the ones
4806 * enforcing this dependency, so if they ever change, we might as well.
4808 * schedule_work() will guarantee this happens. Be careful if you need
4809 * to move this code around, and make sure it is outside
4812 disarm_sock_keys(memcg
);
4813 if (size
< PAGE_SIZE
)
4819 static void free_rcu(struct rcu_head
*rcu_head
)
4821 struct mem_cgroup
*memcg
;
4823 memcg
= container_of(rcu_head
, struct mem_cgroup
, rcu_freeing
);
4824 INIT_WORK(&memcg
->work_freeing
, free_work
);
4825 schedule_work(&memcg
->work_freeing
);
4829 * At destroying mem_cgroup, references from swap_cgroup can remain.
4830 * (scanning all at force_empty is too costly...)
4832 * Instead of clearing all references at force_empty, we remember
4833 * the number of reference from swap_cgroup and free mem_cgroup when
4834 * it goes down to 0.
4836 * Removal of cgroup itself succeeds regardless of refs from swap.
4839 static void __mem_cgroup_free(struct mem_cgroup
*memcg
)
4843 mem_cgroup_remove_from_trees(memcg
);
4844 free_css_id(&mem_cgroup_subsys
, &memcg
->css
);
4847 free_mem_cgroup_per_zone_info(memcg
, node
);
4849 free_percpu(memcg
->stat
);
4850 call_rcu(&memcg
->rcu_freeing
, free_rcu
);
4853 static void mem_cgroup_get(struct mem_cgroup
*memcg
)
4855 atomic_inc(&memcg
->refcnt
);
4858 static void __mem_cgroup_put(struct mem_cgroup
*memcg
, int count
)
4860 if (atomic_sub_and_test(count
, &memcg
->refcnt
)) {
4861 struct mem_cgroup
*parent
= parent_mem_cgroup(memcg
);
4862 __mem_cgroup_free(memcg
);
4864 mem_cgroup_put(parent
);
4868 static void mem_cgroup_put(struct mem_cgroup
*memcg
)
4870 __mem_cgroup_put(memcg
, 1);
4874 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4876 struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*memcg
)
4878 if (!memcg
->res
.parent
)
4880 return mem_cgroup_from_res_counter(memcg
->res
.parent
, res
);
4882 EXPORT_SYMBOL(parent_mem_cgroup
);
4884 #ifdef CONFIG_MEMCG_SWAP
4885 static void __init
enable_swap_cgroup(void)
4887 if (!mem_cgroup_disabled() && really_do_swap_account
)
4888 do_swap_account
= 1;
4891 static void __init
enable_swap_cgroup(void)
4896 static int mem_cgroup_soft_limit_tree_init(void)
4898 struct mem_cgroup_tree_per_node
*rtpn
;
4899 struct mem_cgroup_tree_per_zone
*rtpz
;
4900 int tmp
, node
, zone
;
4902 for_each_node(node
) {
4904 if (!node_state(node
, N_NORMAL_MEMORY
))
4906 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
, tmp
);
4910 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
4912 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4913 rtpz
= &rtpn
->rb_tree_per_zone
[zone
];
4914 rtpz
->rb_root
= RB_ROOT
;
4915 spin_lock_init(&rtpz
->lock
);
4921 for_each_node(node
) {
4922 if (!soft_limit_tree
.rb_tree_per_node
[node
])
4924 kfree(soft_limit_tree
.rb_tree_per_node
[node
]);
4925 soft_limit_tree
.rb_tree_per_node
[node
] = NULL
;
4931 static struct cgroup_subsys_state
* __ref
4932 mem_cgroup_create(struct cgroup
*cont
)
4934 struct mem_cgroup
*memcg
, *parent
;
4935 long error
= -ENOMEM
;
4938 memcg
= mem_cgroup_alloc();
4940 return ERR_PTR(error
);
4943 if (alloc_mem_cgroup_per_zone_info(memcg
, node
))
4947 if (cont
->parent
== NULL
) {
4949 enable_swap_cgroup();
4951 if (mem_cgroup_soft_limit_tree_init())
4953 root_mem_cgroup
= memcg
;
4954 for_each_possible_cpu(cpu
) {
4955 struct memcg_stock_pcp
*stock
=
4956 &per_cpu(memcg_stock
, cpu
);
4957 INIT_WORK(&stock
->work
, drain_local_stock
);
4959 hotcpu_notifier(memcg_cpu_hotplug_callback
, 0);
4961 parent
= mem_cgroup_from_cont(cont
->parent
);
4962 memcg
->use_hierarchy
= parent
->use_hierarchy
;
4963 memcg
->oom_kill_disable
= parent
->oom_kill_disable
;
4966 if (parent
&& parent
->use_hierarchy
) {
4967 res_counter_init(&memcg
->res
, &parent
->res
);
4968 res_counter_init(&memcg
->memsw
, &parent
->memsw
);
4970 * We increment refcnt of the parent to ensure that we can
4971 * safely access it on res_counter_charge/uncharge.
4972 * This refcnt will be decremented when freeing this
4973 * mem_cgroup(see mem_cgroup_put).
4975 mem_cgroup_get(parent
);
4977 res_counter_init(&memcg
->res
, NULL
);
4978 res_counter_init(&memcg
->memsw
, NULL
);
4980 * Deeper hierachy with use_hierarchy == false doesn't make
4981 * much sense so let cgroup subsystem know about this
4982 * unfortunate state in our controller.
4984 if (parent
&& parent
!= root_mem_cgroup
)
4985 mem_cgroup_subsys
.broken_hierarchy
= true;
4987 memcg
->last_scanned_node
= MAX_NUMNODES
;
4988 INIT_LIST_HEAD(&memcg
->oom_notify
);
4991 memcg
->swappiness
= mem_cgroup_swappiness(parent
);
4992 atomic_set(&memcg
->refcnt
, 1);
4993 memcg
->move_charge_at_immigrate
= 0;
4994 mutex_init(&memcg
->thresholds_lock
);
4995 spin_lock_init(&memcg
->move_lock
);
4997 error
= memcg_init_kmem(memcg
, &mem_cgroup_subsys
);
5000 * We call put now because our (and parent's) refcnts
5001 * are already in place. mem_cgroup_put() will internally
5002 * call __mem_cgroup_free, so return directly
5004 mem_cgroup_put(memcg
);
5005 return ERR_PTR(error
);
5009 __mem_cgroup_free(memcg
);
5010 return ERR_PTR(error
);
5013 static int mem_cgroup_pre_destroy(struct cgroup
*cont
)
5015 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
5017 return mem_cgroup_force_empty(memcg
, false);
5020 static void mem_cgroup_destroy(struct cgroup
*cont
)
5022 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
5024 kmem_cgroup_destroy(memcg
);
5026 mem_cgroup_put(memcg
);
5030 /* Handlers for move charge at task migration. */
5031 #define PRECHARGE_COUNT_AT_ONCE 256
5032 static int mem_cgroup_do_precharge(unsigned long count
)
5035 int batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5036 struct mem_cgroup
*memcg
= mc
.to
;
5038 if (mem_cgroup_is_root(memcg
)) {
5039 mc
.precharge
+= count
;
5040 /* we don't need css_get for root */
5043 /* try to charge at once */
5045 struct res_counter
*dummy
;
5047 * "memcg" cannot be under rmdir() because we've already checked
5048 * by cgroup_lock_live_cgroup() that it is not removed and we
5049 * are still under the same cgroup_mutex. So we can postpone
5052 if (res_counter_charge(&memcg
->res
, PAGE_SIZE
* count
, &dummy
))
5054 if (do_swap_account
&& res_counter_charge(&memcg
->memsw
,
5055 PAGE_SIZE
* count
, &dummy
)) {
5056 res_counter_uncharge(&memcg
->res
, PAGE_SIZE
* count
);
5059 mc
.precharge
+= count
;
5063 /* fall back to one by one charge */
5065 if (signal_pending(current
)) {
5069 if (!batch_count
--) {
5070 batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5073 ret
= __mem_cgroup_try_charge(NULL
,
5074 GFP_KERNEL
, 1, &memcg
, false);
5076 /* mem_cgroup_clear_mc() will do uncharge later */
5084 * get_mctgt_type - get target type of moving charge
5085 * @vma: the vma the pte to be checked belongs
5086 * @addr: the address corresponding to the pte to be checked
5087 * @ptent: the pte to be checked
5088 * @target: the pointer the target page or swap ent will be stored(can be NULL)
5091 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5092 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5093 * move charge. if @target is not NULL, the page is stored in target->page
5094 * with extra refcnt got(Callers should handle it).
5095 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5096 * target for charge migration. if @target is not NULL, the entry is stored
5099 * Called with pte lock held.
5106 enum mc_target_type
{
5112 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
5113 unsigned long addr
, pte_t ptent
)
5115 struct page
*page
= vm_normal_page(vma
, addr
, ptent
);
5117 if (!page
|| !page_mapped(page
))
5119 if (PageAnon(page
)) {
5120 /* we don't move shared anon */
5123 } else if (!move_file())
5124 /* we ignore mapcount for file pages */
5126 if (!get_page_unless_zero(page
))
5133 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5134 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5136 struct page
*page
= NULL
;
5137 swp_entry_t ent
= pte_to_swp_entry(ptent
);
5139 if (!move_anon() || non_swap_entry(ent
))
5142 * Because lookup_swap_cache() updates some statistics counter,
5143 * we call find_get_page() with swapper_space directly.
5145 page
= find_get_page(&swapper_space
, ent
.val
);
5146 if (do_swap_account
)
5147 entry
->val
= ent
.val
;
5152 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5153 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5159 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
5160 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5162 struct page
*page
= NULL
;
5163 struct address_space
*mapping
;
5166 if (!vma
->vm_file
) /* anonymous vma */
5171 mapping
= vma
->vm_file
->f_mapping
;
5172 if (pte_none(ptent
))
5173 pgoff
= linear_page_index(vma
, addr
);
5174 else /* pte_file(ptent) is true */
5175 pgoff
= pte_to_pgoff(ptent
);
5177 /* page is moved even if it's not RSS of this task(page-faulted). */
5178 page
= find_get_page(mapping
, pgoff
);
5181 /* shmem/tmpfs may report page out on swap: account for that too. */
5182 if (radix_tree_exceptional_entry(page
)) {
5183 swp_entry_t swap
= radix_to_swp_entry(page
);
5184 if (do_swap_account
)
5186 page
= find_get_page(&swapper_space
, swap
.val
);
5192 static enum mc_target_type
get_mctgt_type(struct vm_area_struct
*vma
,
5193 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
5195 struct page
*page
= NULL
;
5196 struct page_cgroup
*pc
;
5197 enum mc_target_type ret
= MC_TARGET_NONE
;
5198 swp_entry_t ent
= { .val
= 0 };
5200 if (pte_present(ptent
))
5201 page
= mc_handle_present_pte(vma
, addr
, ptent
);
5202 else if (is_swap_pte(ptent
))
5203 page
= mc_handle_swap_pte(vma
, addr
, ptent
, &ent
);
5204 else if (pte_none(ptent
) || pte_file(ptent
))
5205 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
5207 if (!page
&& !ent
.val
)
5210 pc
= lookup_page_cgroup(page
);
5212 * Do only loose check w/o page_cgroup lock.
5213 * mem_cgroup_move_account() checks the pc is valid or not under
5216 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5217 ret
= MC_TARGET_PAGE
;
5219 target
->page
= page
;
5221 if (!ret
|| !target
)
5224 /* There is a swap entry and a page doesn't exist or isn't charged */
5225 if (ent
.val
&& !ret
&&
5226 css_id(&mc
.from
->css
) == lookup_swap_cgroup_id(ent
)) {
5227 ret
= MC_TARGET_SWAP
;
5234 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5236 * We don't consider swapping or file mapped pages because THP does not
5237 * support them for now.
5238 * Caller should make sure that pmd_trans_huge(pmd) is true.
5240 static enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
5241 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
5243 struct page
*page
= NULL
;
5244 struct page_cgroup
*pc
;
5245 enum mc_target_type ret
= MC_TARGET_NONE
;
5247 page
= pmd_page(pmd
);
5248 VM_BUG_ON(!page
|| !PageHead(page
));
5251 pc
= lookup_page_cgroup(page
);
5252 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5253 ret
= MC_TARGET_PAGE
;
5256 target
->page
= page
;
5262 static inline enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
5263 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
5265 return MC_TARGET_NONE
;
5269 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
5270 unsigned long addr
, unsigned long end
,
5271 struct mm_walk
*walk
)
5273 struct vm_area_struct
*vma
= walk
->private;
5277 if (pmd_trans_huge_lock(pmd
, vma
) == 1) {
5278 if (get_mctgt_type_thp(vma
, addr
, *pmd
, NULL
) == MC_TARGET_PAGE
)
5279 mc
.precharge
+= HPAGE_PMD_NR
;
5280 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5284 if (pmd_trans_unstable(pmd
))
5286 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5287 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
5288 if (get_mctgt_type(vma
, addr
, *pte
, NULL
))
5289 mc
.precharge
++; /* increment precharge temporarily */
5290 pte_unmap_unlock(pte
- 1, ptl
);
5296 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
5298 unsigned long precharge
;
5299 struct vm_area_struct
*vma
;
5301 down_read(&mm
->mmap_sem
);
5302 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5303 struct mm_walk mem_cgroup_count_precharge_walk
= {
5304 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
5308 if (is_vm_hugetlb_page(vma
))
5310 walk_page_range(vma
->vm_start
, vma
->vm_end
,
5311 &mem_cgroup_count_precharge_walk
);
5313 up_read(&mm
->mmap_sem
);
5315 precharge
= mc
.precharge
;
5321 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
5323 unsigned long precharge
= mem_cgroup_count_precharge(mm
);
5325 VM_BUG_ON(mc
.moving_task
);
5326 mc
.moving_task
= current
;
5327 return mem_cgroup_do_precharge(precharge
);
5330 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5331 static void __mem_cgroup_clear_mc(void)
5333 struct mem_cgroup
*from
= mc
.from
;
5334 struct mem_cgroup
*to
= mc
.to
;
5336 /* we must uncharge all the leftover precharges from mc.to */
5338 __mem_cgroup_cancel_charge(mc
.to
, mc
.precharge
);
5342 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5343 * we must uncharge here.
5345 if (mc
.moved_charge
) {
5346 __mem_cgroup_cancel_charge(mc
.from
, mc
.moved_charge
);
5347 mc
.moved_charge
= 0;
5349 /* we must fixup refcnts and charges */
5350 if (mc
.moved_swap
) {
5351 /* uncharge swap account from the old cgroup */
5352 if (!mem_cgroup_is_root(mc
.from
))
5353 res_counter_uncharge(&mc
.from
->memsw
,
5354 PAGE_SIZE
* mc
.moved_swap
);
5355 __mem_cgroup_put(mc
.from
, mc
.moved_swap
);
5357 if (!mem_cgroup_is_root(mc
.to
)) {
5359 * we charged both to->res and to->memsw, so we should
5362 res_counter_uncharge(&mc
.to
->res
,
5363 PAGE_SIZE
* mc
.moved_swap
);
5365 /* we've already done mem_cgroup_get(mc.to) */
5368 memcg_oom_recover(from
);
5369 memcg_oom_recover(to
);
5370 wake_up_all(&mc
.waitq
);
5373 static void mem_cgroup_clear_mc(void)
5375 struct mem_cgroup
*from
= mc
.from
;
5378 * we must clear moving_task before waking up waiters at the end of
5381 mc
.moving_task
= NULL
;
5382 __mem_cgroup_clear_mc();
5383 spin_lock(&mc
.lock
);
5386 spin_unlock(&mc
.lock
);
5387 mem_cgroup_end_move(from
);
5390 static int mem_cgroup_can_attach(struct cgroup
*cgroup
,
5391 struct cgroup_taskset
*tset
)
5393 struct task_struct
*p
= cgroup_taskset_first(tset
);
5395 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgroup
);
5397 if (memcg
->move_charge_at_immigrate
) {
5398 struct mm_struct
*mm
;
5399 struct mem_cgroup
*from
= mem_cgroup_from_task(p
);
5401 VM_BUG_ON(from
== memcg
);
5403 mm
= get_task_mm(p
);
5406 /* We move charges only when we move a owner of the mm */
5407 if (mm
->owner
== p
) {
5410 VM_BUG_ON(mc
.precharge
);
5411 VM_BUG_ON(mc
.moved_charge
);
5412 VM_BUG_ON(mc
.moved_swap
);
5413 mem_cgroup_start_move(from
);
5414 spin_lock(&mc
.lock
);
5417 spin_unlock(&mc
.lock
);
5418 /* We set mc.moving_task later */
5420 ret
= mem_cgroup_precharge_mc(mm
);
5422 mem_cgroup_clear_mc();
5429 static void mem_cgroup_cancel_attach(struct cgroup
*cgroup
,
5430 struct cgroup_taskset
*tset
)
5432 mem_cgroup_clear_mc();
5435 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
5436 unsigned long addr
, unsigned long end
,
5437 struct mm_walk
*walk
)
5440 struct vm_area_struct
*vma
= walk
->private;
5443 enum mc_target_type target_type
;
5444 union mc_target target
;
5446 struct page_cgroup
*pc
;
5449 * We don't take compound_lock() here but no race with splitting thp
5451 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
5452 * under splitting, which means there's no concurrent thp split,
5453 * - if another thread runs into split_huge_page() just after we
5454 * entered this if-block, the thread must wait for page table lock
5455 * to be unlocked in __split_huge_page_splitting(), where the main
5456 * part of thp split is not executed yet.
5458 if (pmd_trans_huge_lock(pmd
, vma
) == 1) {
5459 if (mc
.precharge
< HPAGE_PMD_NR
) {
5460 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5463 target_type
= get_mctgt_type_thp(vma
, addr
, *pmd
, &target
);
5464 if (target_type
== MC_TARGET_PAGE
) {
5466 if (!isolate_lru_page(page
)) {
5467 pc
= lookup_page_cgroup(page
);
5468 if (!mem_cgroup_move_account(page
, HPAGE_PMD_NR
,
5469 pc
, mc
.from
, mc
.to
)) {
5470 mc
.precharge
-= HPAGE_PMD_NR
;
5471 mc
.moved_charge
+= HPAGE_PMD_NR
;
5473 putback_lru_page(page
);
5477 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5481 if (pmd_trans_unstable(pmd
))
5484 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5485 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
5486 pte_t ptent
= *(pte
++);
5492 switch (get_mctgt_type(vma
, addr
, ptent
, &target
)) {
5493 case MC_TARGET_PAGE
:
5495 if (isolate_lru_page(page
))
5497 pc
= lookup_page_cgroup(page
);
5498 if (!mem_cgroup_move_account(page
, 1, pc
,
5501 /* we uncharge from mc.from later. */
5504 putback_lru_page(page
);
5505 put
: /* get_mctgt_type() gets the page */
5508 case MC_TARGET_SWAP
:
5510 if (!mem_cgroup_move_swap_account(ent
, mc
.from
, mc
.to
)) {
5512 /* we fixup refcnts and charges later. */
5520 pte_unmap_unlock(pte
- 1, ptl
);
5525 * We have consumed all precharges we got in can_attach().
5526 * We try charge one by one, but don't do any additional
5527 * charges to mc.to if we have failed in charge once in attach()
5530 ret
= mem_cgroup_do_precharge(1);
5538 static void mem_cgroup_move_charge(struct mm_struct
*mm
)
5540 struct vm_area_struct
*vma
;
5542 lru_add_drain_all();
5544 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
5546 * Someone who are holding the mmap_sem might be waiting in
5547 * waitq. So we cancel all extra charges, wake up all waiters,
5548 * and retry. Because we cancel precharges, we might not be able
5549 * to move enough charges, but moving charge is a best-effort
5550 * feature anyway, so it wouldn't be a big problem.
5552 __mem_cgroup_clear_mc();
5556 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5558 struct mm_walk mem_cgroup_move_charge_walk
= {
5559 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
5563 if (is_vm_hugetlb_page(vma
))
5565 ret
= walk_page_range(vma
->vm_start
, vma
->vm_end
,
5566 &mem_cgroup_move_charge_walk
);
5569 * means we have consumed all precharges and failed in
5570 * doing additional charge. Just abandon here.
5574 up_read(&mm
->mmap_sem
);
5577 static void mem_cgroup_move_task(struct cgroup
*cont
,
5578 struct cgroup_taskset
*tset
)
5580 struct task_struct
*p
= cgroup_taskset_first(tset
);
5581 struct mm_struct
*mm
= get_task_mm(p
);
5585 mem_cgroup_move_charge(mm
);
5589 mem_cgroup_clear_mc();
5591 #else /* !CONFIG_MMU */
5592 static int mem_cgroup_can_attach(struct cgroup
*cgroup
,
5593 struct cgroup_taskset
*tset
)
5597 static void mem_cgroup_cancel_attach(struct cgroup
*cgroup
,
5598 struct cgroup_taskset
*tset
)
5601 static void mem_cgroup_move_task(struct cgroup
*cont
,
5602 struct cgroup_taskset
*tset
)
5607 struct cgroup_subsys mem_cgroup_subsys
= {
5609 .subsys_id
= mem_cgroup_subsys_id
,
5610 .create
= mem_cgroup_create
,
5611 .pre_destroy
= mem_cgroup_pre_destroy
,
5612 .destroy
= mem_cgroup_destroy
,
5613 .can_attach
= mem_cgroup_can_attach
,
5614 .cancel_attach
= mem_cgroup_cancel_attach
,
5615 .attach
= mem_cgroup_move_task
,
5616 .base_cftypes
= mem_cgroup_files
,
5619 .__DEPRECATED_clear_css_refs
= true,
5622 #ifdef CONFIG_MEMCG_SWAP
5623 static int __init
enable_swap_account(char *s
)
5625 /* consider enabled if no parameter or 1 is given */
5626 if (!strcmp(s
, "1"))
5627 really_do_swap_account
= 1;
5628 else if (!strcmp(s
, "0"))
5629 really_do_swap_account
= 0;
5632 __setup("swapaccount=", enable_swap_account
);