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>
54 #include <net/tcp_memcontrol.h>
56 #include <asm/uaccess.h>
58 #include <trace/events/vmscan.h>
60 struct cgroup_subsys mem_cgroup_subsys __read_mostly
;
61 #define MEM_CGROUP_RECLAIM_RETRIES 5
62 static struct mem_cgroup
*root_mem_cgroup __read_mostly
;
64 #ifdef CONFIG_MEMCG_SWAP
65 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
66 int do_swap_account __read_mostly
;
68 /* for remember boot option*/
69 #ifdef CONFIG_MEMCG_SWAP_ENABLED
70 static int really_do_swap_account __initdata
= 1;
72 static int really_do_swap_account __initdata
= 0;
76 #define do_swap_account 0
81 * Statistics for memory cgroup.
83 enum mem_cgroup_stat_index
{
85 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
87 MEM_CGROUP_STAT_CACHE
, /* # of pages charged as cache */
88 MEM_CGROUP_STAT_RSS
, /* # of pages charged as anon rss */
89 MEM_CGROUP_STAT_FILE_MAPPED
, /* # of pages charged as file rss */
90 MEM_CGROUP_STAT_SWAP
, /* # of pages, swapped out */
91 MEM_CGROUP_STAT_NSTATS
,
94 static const char * const mem_cgroup_stat_names
[] = {
101 enum mem_cgroup_events_index
{
102 MEM_CGROUP_EVENTS_PGPGIN
, /* # of pages paged in */
103 MEM_CGROUP_EVENTS_PGPGOUT
, /* # of pages paged out */
104 MEM_CGROUP_EVENTS_PGFAULT
, /* # of page-faults */
105 MEM_CGROUP_EVENTS_PGMAJFAULT
, /* # of major page-faults */
106 MEM_CGROUP_EVENTS_NSTATS
,
109 static const char * const mem_cgroup_events_names
[] = {
117 * Per memcg event counter is incremented at every pagein/pageout. With THP,
118 * it will be incremated by the number of pages. This counter is used for
119 * for trigger some periodic events. This is straightforward and better
120 * than using jiffies etc. to handle periodic memcg event.
122 enum mem_cgroup_events_target
{
123 MEM_CGROUP_TARGET_THRESH
,
124 MEM_CGROUP_TARGET_SOFTLIMIT
,
125 MEM_CGROUP_TARGET_NUMAINFO
,
128 #define THRESHOLDS_EVENTS_TARGET 128
129 #define SOFTLIMIT_EVENTS_TARGET 1024
130 #define NUMAINFO_EVENTS_TARGET 1024
132 struct mem_cgroup_stat_cpu
{
133 long count
[MEM_CGROUP_STAT_NSTATS
];
134 unsigned long events
[MEM_CGROUP_EVENTS_NSTATS
];
135 unsigned long nr_page_events
;
136 unsigned long targets
[MEM_CGROUP_NTARGETS
];
139 struct mem_cgroup_reclaim_iter
{
140 /* css_id of the last scanned hierarchy member */
142 /* scan generation, increased every round-trip */
143 unsigned int generation
;
147 * per-zone information in memory controller.
149 struct mem_cgroup_per_zone
{
150 struct lruvec lruvec
;
151 unsigned long lru_size
[NR_LRU_LISTS
];
153 struct mem_cgroup_reclaim_iter reclaim_iter
[DEF_PRIORITY
+ 1];
155 struct rb_node tree_node
; /* RB tree node */
156 unsigned long long usage_in_excess
;/* Set to the value by which */
157 /* the soft limit is exceeded*/
159 struct mem_cgroup
*memcg
; /* Back pointer, we cannot */
160 /* use container_of */
163 struct mem_cgroup_per_node
{
164 struct mem_cgroup_per_zone zoneinfo
[MAX_NR_ZONES
];
167 struct mem_cgroup_lru_info
{
168 struct mem_cgroup_per_node
*nodeinfo
[MAX_NUMNODES
];
172 * Cgroups above their limits are maintained in a RB-Tree, independent of
173 * their hierarchy representation
176 struct mem_cgroup_tree_per_zone
{
177 struct rb_root rb_root
;
181 struct mem_cgroup_tree_per_node
{
182 struct mem_cgroup_tree_per_zone rb_tree_per_zone
[MAX_NR_ZONES
];
185 struct mem_cgroup_tree
{
186 struct mem_cgroup_tree_per_node
*rb_tree_per_node
[MAX_NUMNODES
];
189 static struct mem_cgroup_tree soft_limit_tree __read_mostly
;
191 struct mem_cgroup_threshold
{
192 struct eventfd_ctx
*eventfd
;
197 struct mem_cgroup_threshold_ary
{
198 /* An array index points to threshold just below or equal to usage. */
199 int current_threshold
;
200 /* Size of entries[] */
202 /* Array of thresholds */
203 struct mem_cgroup_threshold entries
[0];
206 struct mem_cgroup_thresholds
{
207 /* Primary thresholds array */
208 struct mem_cgroup_threshold_ary
*primary
;
210 * Spare threshold array.
211 * This is needed to make mem_cgroup_unregister_event() "never fail".
212 * It must be able to store at least primary->size - 1 entries.
214 struct mem_cgroup_threshold_ary
*spare
;
218 struct mem_cgroup_eventfd_list
{
219 struct list_head list
;
220 struct eventfd_ctx
*eventfd
;
223 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
);
224 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
);
227 * The memory controller data structure. The memory controller controls both
228 * page cache and RSS per cgroup. We would eventually like to provide
229 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
230 * to help the administrator determine what knobs to tune.
232 * TODO: Add a water mark for the memory controller. Reclaim will begin when
233 * we hit the water mark. May be even add a low water mark, such that
234 * no reclaim occurs from a cgroup at it's low water mark, this is
235 * a feature that will be implemented much later in the future.
238 struct cgroup_subsys_state css
;
240 * the counter to account for memory usage
242 struct res_counter res
;
246 * the counter to account for mem+swap usage.
248 struct res_counter memsw
;
251 * rcu_freeing is used only when freeing struct mem_cgroup,
252 * so put it into a union to avoid wasting more memory.
253 * It must be disjoint from the css field. It could be
254 * in a union with the res field, but res plays a much
255 * larger part in mem_cgroup life than memsw, and might
256 * be of interest, even at time of free, when debugging.
257 * So share rcu_head with the less interesting memsw.
259 struct rcu_head rcu_freeing
;
261 * We also need some space for a worker in deferred freeing.
262 * By the time we call it, rcu_freeing is no longer in use.
264 struct work_struct work_freeing
;
268 * Per cgroup active and inactive list, similar to the
269 * per zone LRU lists.
271 struct mem_cgroup_lru_info info
;
272 int last_scanned_node
;
274 nodemask_t scan_nodes
;
275 atomic_t numainfo_events
;
276 atomic_t numainfo_updating
;
279 * Should the accounting and control be hierarchical, per subtree?
289 /* OOM-Killer disable */
290 int oom_kill_disable
;
292 /* set when res.limit == memsw.limit */
293 bool memsw_is_minimum
;
295 /* protect arrays of thresholds */
296 struct mutex thresholds_lock
;
298 /* thresholds for memory usage. RCU-protected */
299 struct mem_cgroup_thresholds thresholds
;
301 /* thresholds for mem+swap usage. RCU-protected */
302 struct mem_cgroup_thresholds memsw_thresholds
;
304 /* For oom notifier event fd */
305 struct list_head oom_notify
;
308 * Should we move charges of a task when a task is moved into this
309 * mem_cgroup ? And what type of charges should we move ?
311 unsigned long move_charge_at_immigrate
;
313 * set > 0 if pages under this cgroup are moving to other cgroup.
315 atomic_t moving_account
;
316 /* taken only while moving_account > 0 */
317 spinlock_t move_lock
;
321 struct mem_cgroup_stat_cpu __percpu
*stat
;
323 * used when a cpu is offlined or other synchronizations
324 * See mem_cgroup_read_stat().
326 struct mem_cgroup_stat_cpu nocpu_base
;
327 spinlock_t pcp_counter_lock
;
330 struct tcp_memcontrol tcp_mem
;
334 /* Stuffs for move charges at task migration. */
336 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
337 * left-shifted bitmap of these types.
340 MOVE_CHARGE_TYPE_ANON
, /* private anonymous page and swap of it */
341 MOVE_CHARGE_TYPE_FILE
, /* file page(including tmpfs) and swap of it */
345 /* "mc" and its members are protected by cgroup_mutex */
346 static struct move_charge_struct
{
347 spinlock_t lock
; /* for from, to */
348 struct mem_cgroup
*from
;
349 struct mem_cgroup
*to
;
350 unsigned long precharge
;
351 unsigned long moved_charge
;
352 unsigned long moved_swap
;
353 struct task_struct
*moving_task
; /* a task moving charges */
354 wait_queue_head_t waitq
; /* a waitq for other context */
356 .lock
= __SPIN_LOCK_UNLOCKED(mc
.lock
),
357 .waitq
= __WAIT_QUEUE_HEAD_INITIALIZER(mc
.waitq
),
360 static bool move_anon(void)
362 return test_bit(MOVE_CHARGE_TYPE_ANON
,
363 &mc
.to
->move_charge_at_immigrate
);
366 static bool move_file(void)
368 return test_bit(MOVE_CHARGE_TYPE_FILE
,
369 &mc
.to
->move_charge_at_immigrate
);
373 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
374 * limit reclaim to prevent infinite loops, if they ever occur.
376 #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
377 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
380 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
381 MEM_CGROUP_CHARGE_TYPE_ANON
,
382 MEM_CGROUP_CHARGE_TYPE_SWAPOUT
, /* for accounting swapcache */
383 MEM_CGROUP_CHARGE_TYPE_DROP
, /* a page was unused swap cache */
387 /* for encoding cft->private value on file */
390 #define _OOM_TYPE (2)
391 #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
392 #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
393 #define MEMFILE_ATTR(val) ((val) & 0xffff)
394 /* Used for OOM nofiier */
395 #define OOM_CONTROL (0)
398 * Reclaim flags for mem_cgroup_hierarchical_reclaim
400 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
401 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
402 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
403 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
405 static void mem_cgroup_get(struct mem_cgroup
*memcg
);
406 static void mem_cgroup_put(struct mem_cgroup
*memcg
);
409 struct mem_cgroup
*mem_cgroup_from_css(struct cgroup_subsys_state
*s
)
411 return container_of(s
, struct mem_cgroup
, css
);
414 /* Writing them here to avoid exposing memcg's inner layout */
415 #ifdef CONFIG_MEMCG_KMEM
416 #include <net/sock.h>
419 static bool mem_cgroup_is_root(struct mem_cgroup
*memcg
);
420 void sock_update_memcg(struct sock
*sk
)
422 if (mem_cgroup_sockets_enabled
) {
423 struct mem_cgroup
*memcg
;
424 struct cg_proto
*cg_proto
;
426 BUG_ON(!sk
->sk_prot
->proto_cgroup
);
428 /* Socket cloning can throw us here with sk_cgrp already
429 * filled. It won't however, necessarily happen from
430 * process context. So the test for root memcg given
431 * the current task's memcg won't help us in this case.
433 * Respecting the original socket's memcg is a better
434 * decision in this case.
437 BUG_ON(mem_cgroup_is_root(sk
->sk_cgrp
->memcg
));
438 mem_cgroup_get(sk
->sk_cgrp
->memcg
);
443 memcg
= mem_cgroup_from_task(current
);
444 cg_proto
= sk
->sk_prot
->proto_cgroup(memcg
);
445 if (!mem_cgroup_is_root(memcg
) && memcg_proto_active(cg_proto
)) {
446 mem_cgroup_get(memcg
);
447 sk
->sk_cgrp
= cg_proto
;
452 EXPORT_SYMBOL(sock_update_memcg
);
454 void sock_release_memcg(struct sock
*sk
)
456 if (mem_cgroup_sockets_enabled
&& sk
->sk_cgrp
) {
457 struct mem_cgroup
*memcg
;
458 WARN_ON(!sk
->sk_cgrp
->memcg
);
459 memcg
= sk
->sk_cgrp
->memcg
;
460 mem_cgroup_put(memcg
);
465 struct cg_proto
*tcp_proto_cgroup(struct mem_cgroup
*memcg
)
467 if (!memcg
|| mem_cgroup_is_root(memcg
))
470 return &memcg
->tcp_mem
.cg_proto
;
472 EXPORT_SYMBOL(tcp_proto_cgroup
);
473 #endif /* CONFIG_INET */
474 #endif /* CONFIG_MEMCG_KMEM */
476 #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
477 static void disarm_sock_keys(struct mem_cgroup
*memcg
)
479 if (!memcg_proto_activated(&memcg
->tcp_mem
.cg_proto
))
481 static_key_slow_dec(&memcg_socket_limit_enabled
);
484 static void disarm_sock_keys(struct mem_cgroup
*memcg
)
489 static void drain_all_stock_async(struct mem_cgroup
*memcg
);
491 static struct mem_cgroup_per_zone
*
492 mem_cgroup_zoneinfo(struct mem_cgroup
*memcg
, int nid
, int zid
)
494 return &memcg
->info
.nodeinfo
[nid
]->zoneinfo
[zid
];
497 struct cgroup_subsys_state
*mem_cgroup_css(struct mem_cgroup
*memcg
)
502 static struct mem_cgroup_per_zone
*
503 page_cgroup_zoneinfo(struct mem_cgroup
*memcg
, struct page
*page
)
505 int nid
= page_to_nid(page
);
506 int zid
= page_zonenum(page
);
508 return mem_cgroup_zoneinfo(memcg
, nid
, zid
);
511 static struct mem_cgroup_tree_per_zone
*
512 soft_limit_tree_node_zone(int nid
, int zid
)
514 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
517 static struct mem_cgroup_tree_per_zone
*
518 soft_limit_tree_from_page(struct page
*page
)
520 int nid
= page_to_nid(page
);
521 int zid
= page_zonenum(page
);
523 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
527 __mem_cgroup_insert_exceeded(struct mem_cgroup
*memcg
,
528 struct mem_cgroup_per_zone
*mz
,
529 struct mem_cgroup_tree_per_zone
*mctz
,
530 unsigned long long new_usage_in_excess
)
532 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
533 struct rb_node
*parent
= NULL
;
534 struct mem_cgroup_per_zone
*mz_node
;
539 mz
->usage_in_excess
= new_usage_in_excess
;
540 if (!mz
->usage_in_excess
)
544 mz_node
= rb_entry(parent
, struct mem_cgroup_per_zone
,
546 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
)
549 * We can't avoid mem cgroups that are over their soft
550 * limit by the same amount
552 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
555 rb_link_node(&mz
->tree_node
, parent
, p
);
556 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
561 __mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
562 struct mem_cgroup_per_zone
*mz
,
563 struct mem_cgroup_tree_per_zone
*mctz
)
567 rb_erase(&mz
->tree_node
, &mctz
->rb_root
);
572 mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
573 struct mem_cgroup_per_zone
*mz
,
574 struct mem_cgroup_tree_per_zone
*mctz
)
576 spin_lock(&mctz
->lock
);
577 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
578 spin_unlock(&mctz
->lock
);
582 static void mem_cgroup_update_tree(struct mem_cgroup
*memcg
, struct page
*page
)
584 unsigned long long excess
;
585 struct mem_cgroup_per_zone
*mz
;
586 struct mem_cgroup_tree_per_zone
*mctz
;
587 int nid
= page_to_nid(page
);
588 int zid
= page_zonenum(page
);
589 mctz
= soft_limit_tree_from_page(page
);
592 * Necessary to update all ancestors when hierarchy is used.
593 * because their event counter is not touched.
595 for (; memcg
; memcg
= parent_mem_cgroup(memcg
)) {
596 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
597 excess
= res_counter_soft_limit_excess(&memcg
->res
);
599 * We have to update the tree if mz is on RB-tree or
600 * mem is over its softlimit.
602 if (excess
|| mz
->on_tree
) {
603 spin_lock(&mctz
->lock
);
604 /* if on-tree, remove it */
606 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
608 * Insert again. mz->usage_in_excess will be updated.
609 * If excess is 0, no tree ops.
611 __mem_cgroup_insert_exceeded(memcg
, mz
, mctz
, excess
);
612 spin_unlock(&mctz
->lock
);
617 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*memcg
)
620 struct mem_cgroup_per_zone
*mz
;
621 struct mem_cgroup_tree_per_zone
*mctz
;
623 for_each_node(node
) {
624 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
625 mz
= mem_cgroup_zoneinfo(memcg
, node
, zone
);
626 mctz
= soft_limit_tree_node_zone(node
, zone
);
627 mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
632 static struct mem_cgroup_per_zone
*
633 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
635 struct rb_node
*rightmost
= NULL
;
636 struct mem_cgroup_per_zone
*mz
;
640 rightmost
= rb_last(&mctz
->rb_root
);
642 goto done
; /* Nothing to reclaim from */
644 mz
= rb_entry(rightmost
, struct mem_cgroup_per_zone
, tree_node
);
646 * Remove the node now but someone else can add it back,
647 * we will to add it back at the end of reclaim to its correct
648 * position in the tree.
650 __mem_cgroup_remove_exceeded(mz
->memcg
, mz
, mctz
);
651 if (!res_counter_soft_limit_excess(&mz
->memcg
->res
) ||
652 !css_tryget(&mz
->memcg
->css
))
658 static struct mem_cgroup_per_zone
*
659 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
661 struct mem_cgroup_per_zone
*mz
;
663 spin_lock(&mctz
->lock
);
664 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
665 spin_unlock(&mctz
->lock
);
670 * Implementation Note: reading percpu statistics for memcg.
672 * Both of vmstat[] and percpu_counter has threshold and do periodic
673 * synchronization to implement "quick" read. There are trade-off between
674 * reading cost and precision of value. Then, we may have a chance to implement
675 * a periodic synchronizion of counter in memcg's counter.
677 * But this _read() function is used for user interface now. The user accounts
678 * memory usage by memory cgroup and he _always_ requires exact value because
679 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
680 * have to visit all online cpus and make sum. So, for now, unnecessary
681 * synchronization is not implemented. (just implemented for cpu hotplug)
683 * If there are kernel internal actions which can make use of some not-exact
684 * value, and reading all cpu value can be performance bottleneck in some
685 * common workload, threashold and synchonization as vmstat[] should be
688 static long mem_cgroup_read_stat(struct mem_cgroup
*memcg
,
689 enum mem_cgroup_stat_index idx
)
695 for_each_online_cpu(cpu
)
696 val
+= per_cpu(memcg
->stat
->count
[idx
], cpu
);
697 #ifdef CONFIG_HOTPLUG_CPU
698 spin_lock(&memcg
->pcp_counter_lock
);
699 val
+= memcg
->nocpu_base
.count
[idx
];
700 spin_unlock(&memcg
->pcp_counter_lock
);
706 static void mem_cgroup_swap_statistics(struct mem_cgroup
*memcg
,
709 int val
= (charge
) ? 1 : -1;
710 this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_SWAP
], val
);
713 static unsigned long mem_cgroup_read_events(struct mem_cgroup
*memcg
,
714 enum mem_cgroup_events_index idx
)
716 unsigned long val
= 0;
719 for_each_online_cpu(cpu
)
720 val
+= per_cpu(memcg
->stat
->events
[idx
], cpu
);
721 #ifdef CONFIG_HOTPLUG_CPU
722 spin_lock(&memcg
->pcp_counter_lock
);
723 val
+= memcg
->nocpu_base
.events
[idx
];
724 spin_unlock(&memcg
->pcp_counter_lock
);
729 static void mem_cgroup_charge_statistics(struct mem_cgroup
*memcg
,
730 bool anon
, int nr_pages
)
735 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
736 * counted as CACHE even if it's on ANON LRU.
739 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_RSS
],
742 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_CACHE
],
745 /* pagein of a big page is an event. So, ignore page size */
747 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGIN
]);
749 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGOUT
]);
750 nr_pages
= -nr_pages
; /* for event */
753 __this_cpu_add(memcg
->stat
->nr_page_events
, nr_pages
);
759 mem_cgroup_get_lru_size(struct lruvec
*lruvec
, enum lru_list lru
)
761 struct mem_cgroup_per_zone
*mz
;
763 mz
= container_of(lruvec
, struct mem_cgroup_per_zone
, lruvec
);
764 return mz
->lru_size
[lru
];
768 mem_cgroup_zone_nr_lru_pages(struct mem_cgroup
*memcg
, int nid
, int zid
,
769 unsigned int lru_mask
)
771 struct mem_cgroup_per_zone
*mz
;
773 unsigned long ret
= 0;
775 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
778 if (BIT(lru
) & lru_mask
)
779 ret
+= mz
->lru_size
[lru
];
785 mem_cgroup_node_nr_lru_pages(struct mem_cgroup
*memcg
,
786 int nid
, unsigned int lru_mask
)
791 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++)
792 total
+= mem_cgroup_zone_nr_lru_pages(memcg
,
798 static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup
*memcg
,
799 unsigned int lru_mask
)
804 for_each_node_state(nid
, N_HIGH_MEMORY
)
805 total
+= mem_cgroup_node_nr_lru_pages(memcg
, nid
, lru_mask
);
809 static bool mem_cgroup_event_ratelimit(struct mem_cgroup
*memcg
,
810 enum mem_cgroup_events_target target
)
812 unsigned long val
, next
;
814 val
= __this_cpu_read(memcg
->stat
->nr_page_events
);
815 next
= __this_cpu_read(memcg
->stat
->targets
[target
]);
816 /* from time_after() in jiffies.h */
817 if ((long)next
- (long)val
< 0) {
819 case MEM_CGROUP_TARGET_THRESH
:
820 next
= val
+ THRESHOLDS_EVENTS_TARGET
;
822 case MEM_CGROUP_TARGET_SOFTLIMIT
:
823 next
= val
+ SOFTLIMIT_EVENTS_TARGET
;
825 case MEM_CGROUP_TARGET_NUMAINFO
:
826 next
= val
+ NUMAINFO_EVENTS_TARGET
;
831 __this_cpu_write(memcg
->stat
->targets
[target
], next
);
838 * Check events in order.
841 static void memcg_check_events(struct mem_cgroup
*memcg
, struct page
*page
)
844 /* threshold event is triggered in finer grain than soft limit */
845 if (unlikely(mem_cgroup_event_ratelimit(memcg
,
846 MEM_CGROUP_TARGET_THRESH
))) {
848 bool do_numainfo __maybe_unused
;
850 do_softlimit
= mem_cgroup_event_ratelimit(memcg
,
851 MEM_CGROUP_TARGET_SOFTLIMIT
);
853 do_numainfo
= mem_cgroup_event_ratelimit(memcg
,
854 MEM_CGROUP_TARGET_NUMAINFO
);
858 mem_cgroup_threshold(memcg
);
859 if (unlikely(do_softlimit
))
860 mem_cgroup_update_tree(memcg
, page
);
862 if (unlikely(do_numainfo
))
863 atomic_inc(&memcg
->numainfo_events
);
869 struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
871 return mem_cgroup_from_css(
872 cgroup_subsys_state(cont
, mem_cgroup_subsys_id
));
875 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
878 * mm_update_next_owner() may clear mm->owner to NULL
879 * if it races with swapoff, page migration, etc.
880 * So this can be called with p == NULL.
885 return mem_cgroup_from_css(task_subsys_state(p
, mem_cgroup_subsys_id
));
888 struct mem_cgroup
*try_get_mem_cgroup_from_mm(struct mm_struct
*mm
)
890 struct mem_cgroup
*memcg
= NULL
;
895 * Because we have no locks, mm->owner's may be being moved to other
896 * cgroup. We use css_tryget() here even if this looks
897 * pessimistic (rather than adding locks here).
901 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
902 if (unlikely(!memcg
))
904 } while (!css_tryget(&memcg
->css
));
910 * mem_cgroup_iter - iterate over memory cgroup hierarchy
911 * @root: hierarchy root
912 * @prev: previously returned memcg, NULL on first invocation
913 * @reclaim: cookie for shared reclaim walks, NULL for full walks
915 * Returns references to children of the hierarchy below @root, or
916 * @root itself, or %NULL after a full round-trip.
918 * Caller must pass the return value in @prev on subsequent
919 * invocations for reference counting, or use mem_cgroup_iter_break()
920 * to cancel a hierarchy walk before the round-trip is complete.
922 * Reclaimers can specify a zone and a priority level in @reclaim to
923 * divide up the memcgs in the hierarchy among all concurrent
924 * reclaimers operating on the same zone and priority.
926 struct mem_cgroup
*mem_cgroup_iter(struct mem_cgroup
*root
,
927 struct mem_cgroup
*prev
,
928 struct mem_cgroup_reclaim_cookie
*reclaim
)
930 struct mem_cgroup
*memcg
= NULL
;
933 if (mem_cgroup_disabled())
937 root
= root_mem_cgroup
;
939 if (prev
&& !reclaim
)
940 id
= css_id(&prev
->css
);
942 if (prev
&& prev
!= root
)
945 if (!root
->use_hierarchy
&& root
!= root_mem_cgroup
) {
952 struct mem_cgroup_reclaim_iter
*uninitialized_var(iter
);
953 struct cgroup_subsys_state
*css
;
956 int nid
= zone_to_nid(reclaim
->zone
);
957 int zid
= zone_idx(reclaim
->zone
);
958 struct mem_cgroup_per_zone
*mz
;
960 mz
= mem_cgroup_zoneinfo(root
, nid
, zid
);
961 iter
= &mz
->reclaim_iter
[reclaim
->priority
];
962 if (prev
&& reclaim
->generation
!= iter
->generation
)
968 css
= css_get_next(&mem_cgroup_subsys
, id
+ 1, &root
->css
, &id
);
970 if (css
== &root
->css
|| css_tryget(css
))
971 memcg
= mem_cgroup_from_css(css
);
980 else if (!prev
&& memcg
)
981 reclaim
->generation
= iter
->generation
;
991 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
992 * @root: hierarchy root
993 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
995 void mem_cgroup_iter_break(struct mem_cgroup
*root
,
996 struct mem_cgroup
*prev
)
999 root
= root_mem_cgroup
;
1000 if (prev
&& prev
!= root
)
1001 css_put(&prev
->css
);
1005 * Iteration constructs for visiting all cgroups (under a tree). If
1006 * loops are exited prematurely (break), mem_cgroup_iter_break() must
1007 * be used for reference counting.
1009 #define for_each_mem_cgroup_tree(iter, root) \
1010 for (iter = mem_cgroup_iter(root, NULL, NULL); \
1012 iter = mem_cgroup_iter(root, iter, NULL))
1014 #define for_each_mem_cgroup(iter) \
1015 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
1017 iter = mem_cgroup_iter(NULL, iter, NULL))
1019 static inline bool mem_cgroup_is_root(struct mem_cgroup
*memcg
)
1021 return (memcg
== root_mem_cgroup
);
1024 void mem_cgroup_count_vm_event(struct mm_struct
*mm
, enum vm_event_item idx
)
1026 struct mem_cgroup
*memcg
;
1032 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
1033 if (unlikely(!memcg
))
1038 this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGFAULT
]);
1041 this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGMAJFAULT
]);
1049 EXPORT_SYMBOL(mem_cgroup_count_vm_event
);
1052 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1053 * @zone: zone of the wanted lruvec
1054 * @memcg: memcg of the wanted lruvec
1056 * Returns the lru list vector holding pages for the given @zone and
1057 * @mem. This can be the global zone lruvec, if the memory controller
1060 struct lruvec
*mem_cgroup_zone_lruvec(struct zone
*zone
,
1061 struct mem_cgroup
*memcg
)
1063 struct mem_cgroup_per_zone
*mz
;
1065 if (mem_cgroup_disabled())
1066 return &zone
->lruvec
;
1068 mz
= mem_cgroup_zoneinfo(memcg
, zone_to_nid(zone
), zone_idx(zone
));
1073 * Following LRU functions are allowed to be used without PCG_LOCK.
1074 * Operations are called by routine of global LRU independently from memcg.
1075 * What we have to take care of here is validness of pc->mem_cgroup.
1077 * Changes to pc->mem_cgroup happens when
1080 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1081 * It is added to LRU before charge.
1082 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1083 * When moving account, the page is not on LRU. It's isolated.
1087 * mem_cgroup_page_lruvec - return lruvec for adding an lru page
1089 * @zone: zone of the page
1091 struct lruvec
*mem_cgroup_page_lruvec(struct page
*page
, struct zone
*zone
)
1093 struct mem_cgroup_per_zone
*mz
;
1094 struct mem_cgroup
*memcg
;
1095 struct page_cgroup
*pc
;
1097 if (mem_cgroup_disabled())
1098 return &zone
->lruvec
;
1100 pc
= lookup_page_cgroup(page
);
1101 memcg
= pc
->mem_cgroup
;
1104 * Surreptitiously switch any uncharged offlist page to root:
1105 * an uncharged page off lru does nothing to secure
1106 * its former mem_cgroup from sudden removal.
1108 * Our caller holds lru_lock, and PageCgroupUsed is updated
1109 * under page_cgroup lock: between them, they make all uses
1110 * of pc->mem_cgroup safe.
1112 if (!PageLRU(page
) && !PageCgroupUsed(pc
) && memcg
!= root_mem_cgroup
)
1113 pc
->mem_cgroup
= memcg
= root_mem_cgroup
;
1115 mz
= page_cgroup_zoneinfo(memcg
, page
);
1120 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1121 * @lruvec: mem_cgroup per zone lru vector
1122 * @lru: index of lru list the page is sitting on
1123 * @nr_pages: positive when adding or negative when removing
1125 * This function must be called when a page is added to or removed from an
1128 void mem_cgroup_update_lru_size(struct lruvec
*lruvec
, enum lru_list lru
,
1131 struct mem_cgroup_per_zone
*mz
;
1132 unsigned long *lru_size
;
1134 if (mem_cgroup_disabled())
1137 mz
= container_of(lruvec
, struct mem_cgroup_per_zone
, lruvec
);
1138 lru_size
= mz
->lru_size
+ lru
;
1139 *lru_size
+= nr_pages
;
1140 VM_BUG_ON((long)(*lru_size
) < 0);
1144 * Checks whether given mem is same or in the root_mem_cgroup's
1147 bool __mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_memcg
,
1148 struct mem_cgroup
*memcg
)
1150 if (root_memcg
== memcg
)
1152 if (!root_memcg
->use_hierarchy
|| !memcg
)
1154 return css_is_ancestor(&memcg
->css
, &root_memcg
->css
);
1157 static bool mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_memcg
,
1158 struct mem_cgroup
*memcg
)
1163 ret
= __mem_cgroup_same_or_subtree(root_memcg
, memcg
);
1168 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*memcg
)
1171 struct mem_cgroup
*curr
= NULL
;
1172 struct task_struct
*p
;
1174 p
= find_lock_task_mm(task
);
1176 curr
= try_get_mem_cgroup_from_mm(p
->mm
);
1180 * All threads may have already detached their mm's, but the oom
1181 * killer still needs to detect if they have already been oom
1182 * killed to prevent needlessly killing additional tasks.
1185 curr
= mem_cgroup_from_task(task
);
1187 css_get(&curr
->css
);
1193 * We should check use_hierarchy of "memcg" not "curr". Because checking
1194 * use_hierarchy of "curr" here make this function true if hierarchy is
1195 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1196 * hierarchy(even if use_hierarchy is disabled in "memcg").
1198 ret
= mem_cgroup_same_or_subtree(memcg
, curr
);
1199 css_put(&curr
->css
);
1203 int mem_cgroup_inactive_anon_is_low(struct lruvec
*lruvec
)
1205 unsigned long inactive_ratio
;
1206 unsigned long inactive
;
1207 unsigned long active
;
1210 inactive
= mem_cgroup_get_lru_size(lruvec
, LRU_INACTIVE_ANON
);
1211 active
= mem_cgroup_get_lru_size(lruvec
, LRU_ACTIVE_ANON
);
1213 gb
= (inactive
+ active
) >> (30 - PAGE_SHIFT
);
1215 inactive_ratio
= int_sqrt(10 * gb
);
1219 return inactive
* inactive_ratio
< active
;
1222 int mem_cgroup_inactive_file_is_low(struct lruvec
*lruvec
)
1224 unsigned long active
;
1225 unsigned long inactive
;
1227 inactive
= mem_cgroup_get_lru_size(lruvec
, LRU_INACTIVE_FILE
);
1228 active
= mem_cgroup_get_lru_size(lruvec
, LRU_ACTIVE_FILE
);
1230 return (active
> inactive
);
1233 #define mem_cgroup_from_res_counter(counter, member) \
1234 container_of(counter, struct mem_cgroup, member)
1237 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1238 * @memcg: the memory cgroup
1240 * Returns the maximum amount of memory @mem can be charged with, in
1243 static unsigned long mem_cgroup_margin(struct mem_cgroup
*memcg
)
1245 unsigned long long margin
;
1247 margin
= res_counter_margin(&memcg
->res
);
1248 if (do_swap_account
)
1249 margin
= min(margin
, res_counter_margin(&memcg
->memsw
));
1250 return margin
>> PAGE_SHIFT
;
1253 int mem_cgroup_swappiness(struct mem_cgroup
*memcg
)
1255 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
1258 if (cgrp
->parent
== NULL
)
1259 return vm_swappiness
;
1261 return memcg
->swappiness
;
1265 * memcg->moving_account is used for checking possibility that some thread is
1266 * calling move_account(). When a thread on CPU-A starts moving pages under
1267 * a memcg, other threads should check memcg->moving_account under
1268 * rcu_read_lock(), like this:
1272 * memcg->moving_account+1 if (memcg->mocing_account)
1274 * synchronize_rcu() update something.
1279 /* for quick checking without looking up memcg */
1280 atomic_t memcg_moving __read_mostly
;
1282 static void mem_cgroup_start_move(struct mem_cgroup
*memcg
)
1284 atomic_inc(&memcg_moving
);
1285 atomic_inc(&memcg
->moving_account
);
1289 static void mem_cgroup_end_move(struct mem_cgroup
*memcg
)
1292 * Now, mem_cgroup_clear_mc() may call this function with NULL.
1293 * We check NULL in callee rather than caller.
1296 atomic_dec(&memcg_moving
);
1297 atomic_dec(&memcg
->moving_account
);
1302 * 2 routines for checking "mem" is under move_account() or not.
1304 * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This
1305 * is used for avoiding races in accounting. If true,
1306 * pc->mem_cgroup may be overwritten.
1308 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1309 * under hierarchy of moving cgroups. This is for
1310 * waiting at hith-memory prressure caused by "move".
1313 static bool mem_cgroup_stolen(struct mem_cgroup
*memcg
)
1315 VM_BUG_ON(!rcu_read_lock_held());
1316 return atomic_read(&memcg
->moving_account
) > 0;
1319 static bool mem_cgroup_under_move(struct mem_cgroup
*memcg
)
1321 struct mem_cgroup
*from
;
1322 struct mem_cgroup
*to
;
1325 * Unlike task_move routines, we access mc.to, mc.from not under
1326 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1328 spin_lock(&mc
.lock
);
1334 ret
= mem_cgroup_same_or_subtree(memcg
, from
)
1335 || mem_cgroup_same_or_subtree(memcg
, to
);
1337 spin_unlock(&mc
.lock
);
1341 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*memcg
)
1343 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1344 if (mem_cgroup_under_move(memcg
)) {
1346 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1347 /* moving charge context might have finished. */
1350 finish_wait(&mc
.waitq
, &wait
);
1358 * Take this lock when
1359 * - a code tries to modify page's memcg while it's USED.
1360 * - a code tries to modify page state accounting in a memcg.
1361 * see mem_cgroup_stolen(), too.
1363 static void move_lock_mem_cgroup(struct mem_cgroup
*memcg
,
1364 unsigned long *flags
)
1366 spin_lock_irqsave(&memcg
->move_lock
, *flags
);
1369 static void move_unlock_mem_cgroup(struct mem_cgroup
*memcg
,
1370 unsigned long *flags
)
1372 spin_unlock_irqrestore(&memcg
->move_lock
, *flags
);
1376 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1377 * @memcg: The memory cgroup that went over limit
1378 * @p: Task that is going to be killed
1380 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1383 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1385 struct cgroup
*task_cgrp
;
1386 struct cgroup
*mem_cgrp
;
1388 * Need a buffer in BSS, can't rely on allocations. The code relies
1389 * on the assumption that OOM is serialized for memory controller.
1390 * If this assumption is broken, revisit this code.
1392 static char memcg_name
[PATH_MAX
];
1400 mem_cgrp
= memcg
->css
.cgroup
;
1401 task_cgrp
= task_cgroup(p
, mem_cgroup_subsys_id
);
1403 ret
= cgroup_path(task_cgrp
, memcg_name
, PATH_MAX
);
1406 * Unfortunately, we are unable to convert to a useful name
1407 * But we'll still print out the usage information
1414 printk(KERN_INFO
"Task in %s killed", memcg_name
);
1417 ret
= cgroup_path(mem_cgrp
, memcg_name
, PATH_MAX
);
1425 * Continues from above, so we don't need an KERN_ level
1427 printk(KERN_CONT
" as a result of limit of %s\n", memcg_name
);
1430 printk(KERN_INFO
"memory: usage %llukB, limit %llukB, failcnt %llu\n",
1431 res_counter_read_u64(&memcg
->res
, RES_USAGE
) >> 10,
1432 res_counter_read_u64(&memcg
->res
, RES_LIMIT
) >> 10,
1433 res_counter_read_u64(&memcg
->res
, RES_FAILCNT
));
1434 printk(KERN_INFO
"memory+swap: usage %llukB, limit %llukB, "
1436 res_counter_read_u64(&memcg
->memsw
, RES_USAGE
) >> 10,
1437 res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
) >> 10,
1438 res_counter_read_u64(&memcg
->memsw
, RES_FAILCNT
));
1442 * This function returns the number of memcg under hierarchy tree. Returns
1443 * 1(self count) if no children.
1445 static int mem_cgroup_count_children(struct mem_cgroup
*memcg
)
1448 struct mem_cgroup
*iter
;
1450 for_each_mem_cgroup_tree(iter
, memcg
)
1456 * Return the memory (and swap, if configured) limit for a memcg.
1458 static u64
mem_cgroup_get_limit(struct mem_cgroup
*memcg
)
1463 limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
1464 limit
+= total_swap_pages
<< PAGE_SHIFT
;
1466 memsw
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
1468 * If memsw is finite and limits the amount of swap space available
1469 * to this memcg, return that limit.
1471 return min(limit
, memsw
);
1474 void mem_cgroup_out_of_memory(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
1477 struct mem_cgroup
*iter
;
1478 unsigned long chosen_points
= 0;
1479 unsigned long totalpages
;
1480 unsigned int points
= 0;
1481 struct task_struct
*chosen
= NULL
;
1484 * If current has a pending SIGKILL, then automatically select it. The
1485 * goal is to allow it to allocate so that it may quickly exit and free
1488 if (fatal_signal_pending(current
)) {
1489 set_thread_flag(TIF_MEMDIE
);
1493 check_panic_on_oom(CONSTRAINT_MEMCG
, gfp_mask
, order
, NULL
);
1494 totalpages
= mem_cgroup_get_limit(memcg
) >> PAGE_SHIFT
? : 1;
1495 for_each_mem_cgroup_tree(iter
, memcg
) {
1496 struct cgroup
*cgroup
= iter
->css
.cgroup
;
1497 struct cgroup_iter it
;
1498 struct task_struct
*task
;
1500 cgroup_iter_start(cgroup
, &it
);
1501 while ((task
= cgroup_iter_next(cgroup
, &it
))) {
1502 switch (oom_scan_process_thread(task
, totalpages
, NULL
,
1504 case OOM_SCAN_SELECT
:
1506 put_task_struct(chosen
);
1508 chosen_points
= ULONG_MAX
;
1509 get_task_struct(chosen
);
1511 case OOM_SCAN_CONTINUE
:
1513 case OOM_SCAN_ABORT
:
1514 cgroup_iter_end(cgroup
, &it
);
1515 mem_cgroup_iter_break(memcg
, iter
);
1517 put_task_struct(chosen
);
1522 points
= oom_badness(task
, memcg
, NULL
, totalpages
);
1523 if (points
> chosen_points
) {
1525 put_task_struct(chosen
);
1527 chosen_points
= points
;
1528 get_task_struct(chosen
);
1531 cgroup_iter_end(cgroup
, &it
);
1536 points
= chosen_points
* 1000 / totalpages
;
1537 oom_kill_process(chosen
, gfp_mask
, order
, points
, totalpages
, memcg
,
1538 NULL
, "Memory cgroup out of memory");
1541 static unsigned long mem_cgroup_reclaim(struct mem_cgroup
*memcg
,
1543 unsigned long flags
)
1545 unsigned long total
= 0;
1546 bool noswap
= false;
1549 if (flags
& MEM_CGROUP_RECLAIM_NOSWAP
)
1551 if (!(flags
& MEM_CGROUP_RECLAIM_SHRINK
) && memcg
->memsw_is_minimum
)
1554 for (loop
= 0; loop
< MEM_CGROUP_MAX_RECLAIM_LOOPS
; loop
++) {
1556 drain_all_stock_async(memcg
);
1557 total
+= try_to_free_mem_cgroup_pages(memcg
, gfp_mask
, noswap
);
1559 * Allow limit shrinkers, which are triggered directly
1560 * by userspace, to catch signals and stop reclaim
1561 * after minimal progress, regardless of the margin.
1563 if (total
&& (flags
& MEM_CGROUP_RECLAIM_SHRINK
))
1565 if (mem_cgroup_margin(memcg
))
1568 * If nothing was reclaimed after two attempts, there
1569 * may be no reclaimable pages in this hierarchy.
1578 * test_mem_cgroup_node_reclaimable
1579 * @memcg: the target memcg
1580 * @nid: the node ID to be checked.
1581 * @noswap : specify true here if the user wants flle only information.
1583 * This function returns whether the specified memcg contains any
1584 * reclaimable pages on a node. Returns true if there are any reclaimable
1585 * pages in the node.
1587 static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup
*memcg
,
1588 int nid
, bool noswap
)
1590 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_FILE
))
1592 if (noswap
|| !total_swap_pages
)
1594 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_ANON
))
1599 #if MAX_NUMNODES > 1
1602 * Always updating the nodemask is not very good - even if we have an empty
1603 * list or the wrong list here, we can start from some node and traverse all
1604 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1607 static void mem_cgroup_may_update_nodemask(struct mem_cgroup
*memcg
)
1611 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1612 * pagein/pageout changes since the last update.
1614 if (!atomic_read(&memcg
->numainfo_events
))
1616 if (atomic_inc_return(&memcg
->numainfo_updating
) > 1)
1619 /* make a nodemask where this memcg uses memory from */
1620 memcg
->scan_nodes
= node_states
[N_HIGH_MEMORY
];
1622 for_each_node_mask(nid
, node_states
[N_HIGH_MEMORY
]) {
1624 if (!test_mem_cgroup_node_reclaimable(memcg
, nid
, false))
1625 node_clear(nid
, memcg
->scan_nodes
);
1628 atomic_set(&memcg
->numainfo_events
, 0);
1629 atomic_set(&memcg
->numainfo_updating
, 0);
1633 * Selecting a node where we start reclaim from. Because what we need is just
1634 * reducing usage counter, start from anywhere is O,K. Considering
1635 * memory reclaim from current node, there are pros. and cons.
1637 * Freeing memory from current node means freeing memory from a node which
1638 * we'll use or we've used. So, it may make LRU bad. And if several threads
1639 * hit limits, it will see a contention on a node. But freeing from remote
1640 * node means more costs for memory reclaim because of memory latency.
1642 * Now, we use round-robin. Better algorithm is welcomed.
1644 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1648 mem_cgroup_may_update_nodemask(memcg
);
1649 node
= memcg
->last_scanned_node
;
1651 node
= next_node(node
, memcg
->scan_nodes
);
1652 if (node
== MAX_NUMNODES
)
1653 node
= first_node(memcg
->scan_nodes
);
1655 * We call this when we hit limit, not when pages are added to LRU.
1656 * No LRU may hold pages because all pages are UNEVICTABLE or
1657 * memcg is too small and all pages are not on LRU. In that case,
1658 * we use curret node.
1660 if (unlikely(node
== MAX_NUMNODES
))
1661 node
= numa_node_id();
1663 memcg
->last_scanned_node
= node
;
1668 * Check all nodes whether it contains reclaimable pages or not.
1669 * For quick scan, we make use of scan_nodes. This will allow us to skip
1670 * unused nodes. But scan_nodes is lazily updated and may not cotain
1671 * enough new information. We need to do double check.
1673 static bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1678 * quick check...making use of scan_node.
1679 * We can skip unused nodes.
1681 if (!nodes_empty(memcg
->scan_nodes
)) {
1682 for (nid
= first_node(memcg
->scan_nodes
);
1684 nid
= next_node(nid
, memcg
->scan_nodes
)) {
1686 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1691 * Check rest of nodes.
1693 for_each_node_state(nid
, N_HIGH_MEMORY
) {
1694 if (node_isset(nid
, memcg
->scan_nodes
))
1696 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1703 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1708 static bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1710 return test_mem_cgroup_node_reclaimable(memcg
, 0, noswap
);
1714 static int mem_cgroup_soft_reclaim(struct mem_cgroup
*root_memcg
,
1717 unsigned long *total_scanned
)
1719 struct mem_cgroup
*victim
= NULL
;
1722 unsigned long excess
;
1723 unsigned long nr_scanned
;
1724 struct mem_cgroup_reclaim_cookie reclaim
= {
1729 excess
= res_counter_soft_limit_excess(&root_memcg
->res
) >> PAGE_SHIFT
;
1732 victim
= mem_cgroup_iter(root_memcg
, victim
, &reclaim
);
1737 * If we have not been able to reclaim
1738 * anything, it might because there are
1739 * no reclaimable pages under this hierarchy
1744 * We want to do more targeted reclaim.
1745 * excess >> 2 is not to excessive so as to
1746 * reclaim too much, nor too less that we keep
1747 * coming back to reclaim from this cgroup
1749 if (total
>= (excess
>> 2) ||
1750 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
))
1755 if (!mem_cgroup_reclaimable(victim
, false))
1757 total
+= mem_cgroup_shrink_node_zone(victim
, gfp_mask
, false,
1759 *total_scanned
+= nr_scanned
;
1760 if (!res_counter_soft_limit_excess(&root_memcg
->res
))
1763 mem_cgroup_iter_break(root_memcg
, victim
);
1768 * Check OOM-Killer is already running under our hierarchy.
1769 * If someone is running, return false.
1770 * Has to be called with memcg_oom_lock
1772 static bool mem_cgroup_oom_lock(struct mem_cgroup
*memcg
)
1774 struct mem_cgroup
*iter
, *failed
= NULL
;
1776 for_each_mem_cgroup_tree(iter
, memcg
) {
1777 if (iter
->oom_lock
) {
1779 * this subtree of our hierarchy is already locked
1780 * so we cannot give a lock.
1783 mem_cgroup_iter_break(memcg
, iter
);
1786 iter
->oom_lock
= true;
1793 * OK, we failed to lock the whole subtree so we have to clean up
1794 * what we set up to the failing subtree
1796 for_each_mem_cgroup_tree(iter
, memcg
) {
1797 if (iter
== failed
) {
1798 mem_cgroup_iter_break(memcg
, iter
);
1801 iter
->oom_lock
= false;
1807 * Has to be called with memcg_oom_lock
1809 static int mem_cgroup_oom_unlock(struct mem_cgroup
*memcg
)
1811 struct mem_cgroup
*iter
;
1813 for_each_mem_cgroup_tree(iter
, memcg
)
1814 iter
->oom_lock
= false;
1818 static void mem_cgroup_mark_under_oom(struct mem_cgroup
*memcg
)
1820 struct mem_cgroup
*iter
;
1822 for_each_mem_cgroup_tree(iter
, memcg
)
1823 atomic_inc(&iter
->under_oom
);
1826 static void mem_cgroup_unmark_under_oom(struct mem_cgroup
*memcg
)
1828 struct mem_cgroup
*iter
;
1831 * When a new child is created while the hierarchy is under oom,
1832 * mem_cgroup_oom_lock() may not be called. We have to use
1833 * atomic_add_unless() here.
1835 for_each_mem_cgroup_tree(iter
, memcg
)
1836 atomic_add_unless(&iter
->under_oom
, -1, 0);
1839 static DEFINE_SPINLOCK(memcg_oom_lock
);
1840 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1842 struct oom_wait_info
{
1843 struct mem_cgroup
*memcg
;
1847 static int memcg_oom_wake_function(wait_queue_t
*wait
,
1848 unsigned mode
, int sync
, void *arg
)
1850 struct mem_cgroup
*wake_memcg
= (struct mem_cgroup
*)arg
;
1851 struct mem_cgroup
*oom_wait_memcg
;
1852 struct oom_wait_info
*oom_wait_info
;
1854 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1855 oom_wait_memcg
= oom_wait_info
->memcg
;
1858 * Both of oom_wait_info->memcg and wake_memcg are stable under us.
1859 * Then we can use css_is_ancestor without taking care of RCU.
1861 if (!mem_cgroup_same_or_subtree(oom_wait_memcg
, wake_memcg
)
1862 && !mem_cgroup_same_or_subtree(wake_memcg
, oom_wait_memcg
))
1864 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1867 static void memcg_wakeup_oom(struct mem_cgroup
*memcg
)
1869 /* for filtering, pass "memcg" as argument. */
1870 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, memcg
);
1873 static void memcg_oom_recover(struct mem_cgroup
*memcg
)
1875 if (memcg
&& atomic_read(&memcg
->under_oom
))
1876 memcg_wakeup_oom(memcg
);
1880 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1882 static bool mem_cgroup_handle_oom(struct mem_cgroup
*memcg
, gfp_t mask
,
1885 struct oom_wait_info owait
;
1886 bool locked
, need_to_kill
;
1888 owait
.memcg
= memcg
;
1889 owait
.wait
.flags
= 0;
1890 owait
.wait
.func
= memcg_oom_wake_function
;
1891 owait
.wait
.private = current
;
1892 INIT_LIST_HEAD(&owait
.wait
.task_list
);
1893 need_to_kill
= true;
1894 mem_cgroup_mark_under_oom(memcg
);
1896 /* At first, try to OOM lock hierarchy under memcg.*/
1897 spin_lock(&memcg_oom_lock
);
1898 locked
= mem_cgroup_oom_lock(memcg
);
1900 * Even if signal_pending(), we can't quit charge() loop without
1901 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1902 * under OOM is always welcomed, use TASK_KILLABLE here.
1904 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1905 if (!locked
|| memcg
->oom_kill_disable
)
1906 need_to_kill
= false;
1908 mem_cgroup_oom_notify(memcg
);
1909 spin_unlock(&memcg_oom_lock
);
1912 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1913 mem_cgroup_out_of_memory(memcg
, mask
, order
);
1916 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1918 spin_lock(&memcg_oom_lock
);
1920 mem_cgroup_oom_unlock(memcg
);
1921 memcg_wakeup_oom(memcg
);
1922 spin_unlock(&memcg_oom_lock
);
1924 mem_cgroup_unmark_under_oom(memcg
);
1926 if (test_thread_flag(TIF_MEMDIE
) || fatal_signal_pending(current
))
1928 /* Give chance to dying process */
1929 schedule_timeout_uninterruptible(1);
1934 * Currently used to update mapped file statistics, but the routine can be
1935 * generalized to update other statistics as well.
1937 * Notes: Race condition
1939 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1940 * it tends to be costly. But considering some conditions, we doesn't need
1941 * to do so _always_.
1943 * Considering "charge", lock_page_cgroup() is not required because all
1944 * file-stat operations happen after a page is attached to radix-tree. There
1945 * are no race with "charge".
1947 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1948 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1949 * if there are race with "uncharge". Statistics itself is properly handled
1952 * Considering "move", this is an only case we see a race. To make the race
1953 * small, we check mm->moving_account and detect there are possibility of race
1954 * If there is, we take a lock.
1957 void __mem_cgroup_begin_update_page_stat(struct page
*page
,
1958 bool *locked
, unsigned long *flags
)
1960 struct mem_cgroup
*memcg
;
1961 struct page_cgroup
*pc
;
1963 pc
= lookup_page_cgroup(page
);
1965 memcg
= pc
->mem_cgroup
;
1966 if (unlikely(!memcg
|| !PageCgroupUsed(pc
)))
1969 * If this memory cgroup is not under account moving, we don't
1970 * need to take move_lock_mem_cgroup(). Because we already hold
1971 * rcu_read_lock(), any calls to move_account will be delayed until
1972 * rcu_read_unlock() if mem_cgroup_stolen() == true.
1974 if (!mem_cgroup_stolen(memcg
))
1977 move_lock_mem_cgroup(memcg
, flags
);
1978 if (memcg
!= pc
->mem_cgroup
|| !PageCgroupUsed(pc
)) {
1979 move_unlock_mem_cgroup(memcg
, flags
);
1985 void __mem_cgroup_end_update_page_stat(struct page
*page
, unsigned long *flags
)
1987 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1990 * It's guaranteed that pc->mem_cgroup never changes while
1991 * lock is held because a routine modifies pc->mem_cgroup
1992 * should take move_lock_mem_cgroup().
1994 move_unlock_mem_cgroup(pc
->mem_cgroup
, flags
);
1997 void mem_cgroup_update_page_stat(struct page
*page
,
1998 enum mem_cgroup_page_stat_item idx
, int val
)
2000 struct mem_cgroup
*memcg
;
2001 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
2002 unsigned long uninitialized_var(flags
);
2004 if (mem_cgroup_disabled())
2007 memcg
= pc
->mem_cgroup
;
2008 if (unlikely(!memcg
|| !PageCgroupUsed(pc
)))
2012 case MEMCG_NR_FILE_MAPPED
:
2013 idx
= MEM_CGROUP_STAT_FILE_MAPPED
;
2019 this_cpu_add(memcg
->stat
->count
[idx
], val
);
2023 * size of first charge trial. "32" comes from vmscan.c's magic value.
2024 * TODO: maybe necessary to use big numbers in big irons.
2026 #define CHARGE_BATCH 32U
2027 struct memcg_stock_pcp
{
2028 struct mem_cgroup
*cached
; /* this never be root cgroup */
2029 unsigned int nr_pages
;
2030 struct work_struct work
;
2031 unsigned long flags
;
2032 #define FLUSHING_CACHED_CHARGE 0
2034 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
2035 static DEFINE_MUTEX(percpu_charge_mutex
);
2038 * Try to consume stocked charge on this cpu. If success, one page is consumed
2039 * from local stock and true is returned. If the stock is 0 or charges from a
2040 * cgroup which is not current target, returns false. This stock will be
2043 static bool consume_stock(struct mem_cgroup
*memcg
)
2045 struct memcg_stock_pcp
*stock
;
2048 stock
= &get_cpu_var(memcg_stock
);
2049 if (memcg
== stock
->cached
&& stock
->nr_pages
)
2051 else /* need to call res_counter_charge */
2053 put_cpu_var(memcg_stock
);
2058 * Returns stocks cached in percpu to res_counter and reset cached information.
2060 static void drain_stock(struct memcg_stock_pcp
*stock
)
2062 struct mem_cgroup
*old
= stock
->cached
;
2064 if (stock
->nr_pages
) {
2065 unsigned long bytes
= stock
->nr_pages
* PAGE_SIZE
;
2067 res_counter_uncharge(&old
->res
, bytes
);
2068 if (do_swap_account
)
2069 res_counter_uncharge(&old
->memsw
, bytes
);
2070 stock
->nr_pages
= 0;
2072 stock
->cached
= NULL
;
2076 * This must be called under preempt disabled or must be called by
2077 * a thread which is pinned to local cpu.
2079 static void drain_local_stock(struct work_struct
*dummy
)
2081 struct memcg_stock_pcp
*stock
= &__get_cpu_var(memcg_stock
);
2083 clear_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
);
2087 * Cache charges(val) which is from res_counter, to local per_cpu area.
2088 * This will be consumed by consume_stock() function, later.
2090 static void refill_stock(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
2092 struct memcg_stock_pcp
*stock
= &get_cpu_var(memcg_stock
);
2094 if (stock
->cached
!= memcg
) { /* reset if necessary */
2096 stock
->cached
= memcg
;
2098 stock
->nr_pages
+= nr_pages
;
2099 put_cpu_var(memcg_stock
);
2103 * Drains all per-CPU charge caches for given root_memcg resp. subtree
2104 * of the hierarchy under it. sync flag says whether we should block
2105 * until the work is done.
2107 static void drain_all_stock(struct mem_cgroup
*root_memcg
, bool sync
)
2111 /* Notify other cpus that system-wide "drain" is running */
2114 for_each_online_cpu(cpu
) {
2115 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2116 struct mem_cgroup
*memcg
;
2118 memcg
= stock
->cached
;
2119 if (!memcg
|| !stock
->nr_pages
)
2121 if (!mem_cgroup_same_or_subtree(root_memcg
, memcg
))
2123 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
)) {
2125 drain_local_stock(&stock
->work
);
2127 schedule_work_on(cpu
, &stock
->work
);
2135 for_each_online_cpu(cpu
) {
2136 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2137 if (test_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
))
2138 flush_work(&stock
->work
);
2145 * Tries to drain stocked charges in other cpus. This function is asynchronous
2146 * and just put a work per cpu for draining localy on each cpu. Caller can
2147 * expects some charges will be back to res_counter later but cannot wait for
2150 static void drain_all_stock_async(struct mem_cgroup
*root_memcg
)
2153 * If someone calls draining, avoid adding more kworker runs.
2155 if (!mutex_trylock(&percpu_charge_mutex
))
2157 drain_all_stock(root_memcg
, false);
2158 mutex_unlock(&percpu_charge_mutex
);
2161 /* This is a synchronous drain interface. */
2162 static void drain_all_stock_sync(struct mem_cgroup
*root_memcg
)
2164 /* called when force_empty is called */
2165 mutex_lock(&percpu_charge_mutex
);
2166 drain_all_stock(root_memcg
, true);
2167 mutex_unlock(&percpu_charge_mutex
);
2171 * This function drains percpu counter value from DEAD cpu and
2172 * move it to local cpu. Note that this function can be preempted.
2174 static void mem_cgroup_drain_pcp_counter(struct mem_cgroup
*memcg
, int cpu
)
2178 spin_lock(&memcg
->pcp_counter_lock
);
2179 for (i
= 0; i
< MEM_CGROUP_STAT_NSTATS
; i
++) {
2180 long x
= per_cpu(memcg
->stat
->count
[i
], cpu
);
2182 per_cpu(memcg
->stat
->count
[i
], cpu
) = 0;
2183 memcg
->nocpu_base
.count
[i
] += x
;
2185 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
2186 unsigned long x
= per_cpu(memcg
->stat
->events
[i
], cpu
);
2188 per_cpu(memcg
->stat
->events
[i
], cpu
) = 0;
2189 memcg
->nocpu_base
.events
[i
] += x
;
2191 spin_unlock(&memcg
->pcp_counter_lock
);
2194 static int __cpuinit
memcg_cpu_hotplug_callback(struct notifier_block
*nb
,
2195 unsigned long action
,
2198 int cpu
= (unsigned long)hcpu
;
2199 struct memcg_stock_pcp
*stock
;
2200 struct mem_cgroup
*iter
;
2202 if (action
== CPU_ONLINE
)
2205 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
2208 for_each_mem_cgroup(iter
)
2209 mem_cgroup_drain_pcp_counter(iter
, cpu
);
2211 stock
= &per_cpu(memcg_stock
, cpu
);
2217 /* See __mem_cgroup_try_charge() for details */
2219 CHARGE_OK
, /* success */
2220 CHARGE_RETRY
, /* need to retry but retry is not bad */
2221 CHARGE_NOMEM
, /* we can't do more. return -ENOMEM */
2222 CHARGE_WOULDBLOCK
, /* GFP_WAIT wasn't set and no enough res. */
2223 CHARGE_OOM_DIE
, /* the current is killed because of OOM */
2226 static int mem_cgroup_do_charge(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
2227 unsigned int nr_pages
, bool oom_check
)
2229 unsigned long csize
= nr_pages
* PAGE_SIZE
;
2230 struct mem_cgroup
*mem_over_limit
;
2231 struct res_counter
*fail_res
;
2232 unsigned long flags
= 0;
2235 ret
= res_counter_charge(&memcg
->res
, csize
, &fail_res
);
2238 if (!do_swap_account
)
2240 ret
= res_counter_charge(&memcg
->memsw
, csize
, &fail_res
);
2244 res_counter_uncharge(&memcg
->res
, csize
);
2245 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, memsw
);
2246 flags
|= MEM_CGROUP_RECLAIM_NOSWAP
;
2248 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, res
);
2250 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2251 * of regular pages (CHARGE_BATCH), or a single regular page (1).
2253 * Never reclaim on behalf of optional batching, retry with a
2254 * single page instead.
2256 if (nr_pages
== CHARGE_BATCH
)
2257 return CHARGE_RETRY
;
2259 if (!(gfp_mask
& __GFP_WAIT
))
2260 return CHARGE_WOULDBLOCK
;
2262 ret
= mem_cgroup_reclaim(mem_over_limit
, gfp_mask
, flags
);
2263 if (mem_cgroup_margin(mem_over_limit
) >= nr_pages
)
2264 return CHARGE_RETRY
;
2266 * Even though the limit is exceeded at this point, reclaim
2267 * may have been able to free some pages. Retry the charge
2268 * before killing the task.
2270 * Only for regular pages, though: huge pages are rather
2271 * unlikely to succeed so close to the limit, and we fall back
2272 * to regular pages anyway in case of failure.
2274 if (nr_pages
== 1 && ret
)
2275 return CHARGE_RETRY
;
2278 * At task move, charge accounts can be doubly counted. So, it's
2279 * better to wait until the end of task_move if something is going on.
2281 if (mem_cgroup_wait_acct_move(mem_over_limit
))
2282 return CHARGE_RETRY
;
2284 /* If we don't need to call oom-killer at el, return immediately */
2286 return CHARGE_NOMEM
;
2288 if (!mem_cgroup_handle_oom(mem_over_limit
, gfp_mask
, get_order(csize
)))
2289 return CHARGE_OOM_DIE
;
2291 return CHARGE_RETRY
;
2295 * __mem_cgroup_try_charge() does
2296 * 1. detect memcg to be charged against from passed *mm and *ptr,
2297 * 2. update res_counter
2298 * 3. call memory reclaim if necessary.
2300 * In some special case, if the task is fatal, fatal_signal_pending() or
2301 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2302 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2303 * as possible without any hazards. 2: all pages should have a valid
2304 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2305 * pointer, that is treated as a charge to root_mem_cgroup.
2307 * So __mem_cgroup_try_charge() will return
2308 * 0 ... on success, filling *ptr with a valid memcg pointer.
2309 * -ENOMEM ... charge failure because of resource limits.
2310 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2312 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2313 * the oom-killer can be invoked.
2315 static int __mem_cgroup_try_charge(struct mm_struct
*mm
,
2317 unsigned int nr_pages
,
2318 struct mem_cgroup
**ptr
,
2321 unsigned int batch
= max(CHARGE_BATCH
, nr_pages
);
2322 int nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2323 struct mem_cgroup
*memcg
= NULL
;
2327 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2328 * in system level. So, allow to go ahead dying process in addition to
2331 if (unlikely(test_thread_flag(TIF_MEMDIE
)
2332 || fatal_signal_pending(current
)))
2336 * We always charge the cgroup the mm_struct belongs to.
2337 * The mm_struct's mem_cgroup changes on task migration if the
2338 * thread group leader migrates. It's possible that mm is not
2339 * set, if so charge the root memcg (happens for pagecache usage).
2342 *ptr
= root_mem_cgroup
;
2344 if (*ptr
) { /* css should be a valid one */
2346 VM_BUG_ON(css_is_removed(&memcg
->css
));
2347 if (mem_cgroup_is_root(memcg
))
2349 if (nr_pages
== 1 && consume_stock(memcg
))
2351 css_get(&memcg
->css
);
2353 struct task_struct
*p
;
2356 p
= rcu_dereference(mm
->owner
);
2358 * Because we don't have task_lock(), "p" can exit.
2359 * In that case, "memcg" can point to root or p can be NULL with
2360 * race with swapoff. Then, we have small risk of mis-accouning.
2361 * But such kind of mis-account by race always happens because
2362 * we don't have cgroup_mutex(). It's overkill and we allo that
2364 * (*) swapoff at el will charge against mm-struct not against
2365 * task-struct. So, mm->owner can be NULL.
2367 memcg
= mem_cgroup_from_task(p
);
2369 memcg
= root_mem_cgroup
;
2370 if (mem_cgroup_is_root(memcg
)) {
2374 if (nr_pages
== 1 && consume_stock(memcg
)) {
2376 * It seems dagerous to access memcg without css_get().
2377 * But considering how consume_stok works, it's not
2378 * necessary. If consume_stock success, some charges
2379 * from this memcg are cached on this cpu. So, we
2380 * don't need to call css_get()/css_tryget() before
2381 * calling consume_stock().
2386 /* after here, we may be blocked. we need to get refcnt */
2387 if (!css_tryget(&memcg
->css
)) {
2397 /* If killed, bypass charge */
2398 if (fatal_signal_pending(current
)) {
2399 css_put(&memcg
->css
);
2404 if (oom
&& !nr_oom_retries
) {
2406 nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2409 ret
= mem_cgroup_do_charge(memcg
, gfp_mask
, batch
, oom_check
);
2413 case CHARGE_RETRY
: /* not in OOM situation but retry */
2415 css_put(&memcg
->css
);
2418 case CHARGE_WOULDBLOCK
: /* !__GFP_WAIT */
2419 css_put(&memcg
->css
);
2421 case CHARGE_NOMEM
: /* OOM routine works */
2423 css_put(&memcg
->css
);
2426 /* If oom, we never return -ENOMEM */
2429 case CHARGE_OOM_DIE
: /* Killed by OOM Killer */
2430 css_put(&memcg
->css
);
2433 } while (ret
!= CHARGE_OK
);
2435 if (batch
> nr_pages
)
2436 refill_stock(memcg
, batch
- nr_pages
);
2437 css_put(&memcg
->css
);
2445 *ptr
= root_mem_cgroup
;
2450 * Somemtimes we have to undo a charge we got by try_charge().
2451 * This function is for that and do uncharge, put css's refcnt.
2452 * gotten by try_charge().
2454 static void __mem_cgroup_cancel_charge(struct mem_cgroup
*memcg
,
2455 unsigned int nr_pages
)
2457 if (!mem_cgroup_is_root(memcg
)) {
2458 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2460 res_counter_uncharge(&memcg
->res
, bytes
);
2461 if (do_swap_account
)
2462 res_counter_uncharge(&memcg
->memsw
, bytes
);
2467 * Cancel chrages in this cgroup....doesn't propagate to parent cgroup.
2468 * This is useful when moving usage to parent cgroup.
2470 static void __mem_cgroup_cancel_local_charge(struct mem_cgroup
*memcg
,
2471 unsigned int nr_pages
)
2473 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2475 if (mem_cgroup_is_root(memcg
))
2478 res_counter_uncharge_until(&memcg
->res
, memcg
->res
.parent
, bytes
);
2479 if (do_swap_account
)
2480 res_counter_uncharge_until(&memcg
->memsw
,
2481 memcg
->memsw
.parent
, bytes
);
2485 * A helper function to get mem_cgroup from ID. must be called under
2486 * rcu_read_lock(). The caller must check css_is_removed() or some if
2487 * it's concern. (dropping refcnt from swap can be called against removed
2490 static struct mem_cgroup
*mem_cgroup_lookup(unsigned short id
)
2492 struct cgroup_subsys_state
*css
;
2494 /* ID 0 is unused ID */
2497 css
= css_lookup(&mem_cgroup_subsys
, id
);
2500 return mem_cgroup_from_css(css
);
2503 struct mem_cgroup
*try_get_mem_cgroup_from_page(struct page
*page
)
2505 struct mem_cgroup
*memcg
= NULL
;
2506 struct page_cgroup
*pc
;
2510 VM_BUG_ON(!PageLocked(page
));
2512 pc
= lookup_page_cgroup(page
);
2513 lock_page_cgroup(pc
);
2514 if (PageCgroupUsed(pc
)) {
2515 memcg
= pc
->mem_cgroup
;
2516 if (memcg
&& !css_tryget(&memcg
->css
))
2518 } else if (PageSwapCache(page
)) {
2519 ent
.val
= page_private(page
);
2520 id
= lookup_swap_cgroup_id(ent
);
2522 memcg
= mem_cgroup_lookup(id
);
2523 if (memcg
&& !css_tryget(&memcg
->css
))
2527 unlock_page_cgroup(pc
);
2531 static void __mem_cgroup_commit_charge(struct mem_cgroup
*memcg
,
2533 unsigned int nr_pages
,
2534 enum charge_type ctype
,
2537 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
2538 struct zone
*uninitialized_var(zone
);
2539 struct lruvec
*lruvec
;
2540 bool was_on_lru
= false;
2543 lock_page_cgroup(pc
);
2544 VM_BUG_ON(PageCgroupUsed(pc
));
2546 * we don't need page_cgroup_lock about tail pages, becase they are not
2547 * accessed by any other context at this point.
2551 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2552 * may already be on some other mem_cgroup's LRU. Take care of it.
2555 zone
= page_zone(page
);
2556 spin_lock_irq(&zone
->lru_lock
);
2557 if (PageLRU(page
)) {
2558 lruvec
= mem_cgroup_zone_lruvec(zone
, pc
->mem_cgroup
);
2560 del_page_from_lru_list(page
, lruvec
, page_lru(page
));
2565 pc
->mem_cgroup
= memcg
;
2567 * We access a page_cgroup asynchronously without lock_page_cgroup().
2568 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2569 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2570 * before USED bit, we need memory barrier here.
2571 * See mem_cgroup_add_lru_list(), etc.
2574 SetPageCgroupUsed(pc
);
2578 lruvec
= mem_cgroup_zone_lruvec(zone
, pc
->mem_cgroup
);
2579 VM_BUG_ON(PageLRU(page
));
2581 add_page_to_lru_list(page
, lruvec
, page_lru(page
));
2583 spin_unlock_irq(&zone
->lru_lock
);
2586 if (ctype
== MEM_CGROUP_CHARGE_TYPE_ANON
)
2591 mem_cgroup_charge_statistics(memcg
, anon
, nr_pages
);
2592 unlock_page_cgroup(pc
);
2595 * "charge_statistics" updated event counter. Then, check it.
2596 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2597 * if they exceeds softlimit.
2599 memcg_check_events(memcg
, page
);
2602 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2604 #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION)
2606 * Because tail pages are not marked as "used", set it. We're under
2607 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2608 * charge/uncharge will be never happen and move_account() is done under
2609 * compound_lock(), so we don't have to take care of races.
2611 void mem_cgroup_split_huge_fixup(struct page
*head
)
2613 struct page_cgroup
*head_pc
= lookup_page_cgroup(head
);
2614 struct page_cgroup
*pc
;
2617 if (mem_cgroup_disabled())
2619 for (i
= 1; i
< HPAGE_PMD_NR
; i
++) {
2621 pc
->mem_cgroup
= head_pc
->mem_cgroup
;
2622 smp_wmb();/* see __commit_charge() */
2623 pc
->flags
= head_pc
->flags
& ~PCGF_NOCOPY_AT_SPLIT
;
2626 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2629 * mem_cgroup_move_account - move account of the page
2631 * @nr_pages: number of regular pages (>1 for huge pages)
2632 * @pc: page_cgroup of the page.
2633 * @from: mem_cgroup which the page is moved from.
2634 * @to: mem_cgroup which the page is moved to. @from != @to.
2636 * The caller must confirm following.
2637 * - page is not on LRU (isolate_page() is useful.)
2638 * - compound_lock is held when nr_pages > 1
2640 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
2643 static int mem_cgroup_move_account(struct page
*page
,
2644 unsigned int nr_pages
,
2645 struct page_cgroup
*pc
,
2646 struct mem_cgroup
*from
,
2647 struct mem_cgroup
*to
)
2649 unsigned long flags
;
2651 bool anon
= PageAnon(page
);
2653 VM_BUG_ON(from
== to
);
2654 VM_BUG_ON(PageLRU(page
));
2656 * The page is isolated from LRU. So, collapse function
2657 * will not handle this page. But page splitting can happen.
2658 * Do this check under compound_page_lock(). The caller should
2662 if (nr_pages
> 1 && !PageTransHuge(page
))
2665 lock_page_cgroup(pc
);
2668 if (!PageCgroupUsed(pc
) || pc
->mem_cgroup
!= from
)
2671 move_lock_mem_cgroup(from
, &flags
);
2673 if (!anon
&& page_mapped(page
)) {
2674 /* Update mapped_file data for mem_cgroup */
2676 __this_cpu_dec(from
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2677 __this_cpu_inc(to
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2680 mem_cgroup_charge_statistics(from
, anon
, -nr_pages
);
2682 /* caller should have done css_get */
2683 pc
->mem_cgroup
= to
;
2684 mem_cgroup_charge_statistics(to
, anon
, nr_pages
);
2686 * We charges against "to" which may not have any tasks. Then, "to"
2687 * can be under rmdir(). But in current implementation, caller of
2688 * this function is just force_empty() and move charge, so it's
2689 * guaranteed that "to" is never removed. So, we don't check rmdir
2692 move_unlock_mem_cgroup(from
, &flags
);
2695 unlock_page_cgroup(pc
);
2699 memcg_check_events(to
, page
);
2700 memcg_check_events(from
, page
);
2706 * move charges to its parent.
2709 static int mem_cgroup_move_parent(struct page
*page
,
2710 struct page_cgroup
*pc
,
2711 struct mem_cgroup
*child
)
2713 struct mem_cgroup
*parent
;
2714 unsigned int nr_pages
;
2715 unsigned long uninitialized_var(flags
);
2719 if (mem_cgroup_is_root(child
))
2723 if (!get_page_unless_zero(page
))
2725 if (isolate_lru_page(page
))
2728 nr_pages
= hpage_nr_pages(page
);
2730 parent
= parent_mem_cgroup(child
);
2732 * If no parent, move charges to root cgroup.
2735 parent
= root_mem_cgroup
;
2738 flags
= compound_lock_irqsave(page
);
2740 ret
= mem_cgroup_move_account(page
, nr_pages
,
2743 __mem_cgroup_cancel_local_charge(child
, nr_pages
);
2746 compound_unlock_irqrestore(page
, flags
);
2747 putback_lru_page(page
);
2755 * Charge the memory controller for page usage.
2757 * 0 if the charge was successful
2758 * < 0 if the cgroup is over its limit
2760 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
2761 gfp_t gfp_mask
, enum charge_type ctype
)
2763 struct mem_cgroup
*memcg
= NULL
;
2764 unsigned int nr_pages
= 1;
2768 if (PageTransHuge(page
)) {
2769 nr_pages
<<= compound_order(page
);
2770 VM_BUG_ON(!PageTransHuge(page
));
2772 * Never OOM-kill a process for a huge page. The
2773 * fault handler will fall back to regular pages.
2778 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, nr_pages
, &memcg
, oom
);
2781 __mem_cgroup_commit_charge(memcg
, page
, nr_pages
, ctype
, false);
2785 int mem_cgroup_newpage_charge(struct page
*page
,
2786 struct mm_struct
*mm
, gfp_t gfp_mask
)
2788 if (mem_cgroup_disabled())
2790 VM_BUG_ON(page_mapped(page
));
2791 VM_BUG_ON(page
->mapping
&& !PageAnon(page
));
2793 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2794 MEM_CGROUP_CHARGE_TYPE_ANON
);
2798 * While swap-in, try_charge -> commit or cancel, the page is locked.
2799 * And when try_charge() successfully returns, one refcnt to memcg without
2800 * struct page_cgroup is acquired. This refcnt will be consumed by
2801 * "commit()" or removed by "cancel()"
2803 static int __mem_cgroup_try_charge_swapin(struct mm_struct
*mm
,
2806 struct mem_cgroup
**memcgp
)
2808 struct mem_cgroup
*memcg
;
2809 struct page_cgroup
*pc
;
2812 pc
= lookup_page_cgroup(page
);
2814 * Every swap fault against a single page tries to charge the
2815 * page, bail as early as possible. shmem_unuse() encounters
2816 * already charged pages, too. The USED bit is protected by
2817 * the page lock, which serializes swap cache removal, which
2818 * in turn serializes uncharging.
2820 if (PageCgroupUsed(pc
))
2822 if (!do_swap_account
)
2824 memcg
= try_get_mem_cgroup_from_page(page
);
2828 ret
= __mem_cgroup_try_charge(NULL
, mask
, 1, memcgp
, true);
2829 css_put(&memcg
->css
);
2834 ret
= __mem_cgroup_try_charge(mm
, mask
, 1, memcgp
, true);
2840 int mem_cgroup_try_charge_swapin(struct mm_struct
*mm
, struct page
*page
,
2841 gfp_t gfp_mask
, struct mem_cgroup
**memcgp
)
2844 if (mem_cgroup_disabled())
2847 * A racing thread's fault, or swapoff, may have already
2848 * updated the pte, and even removed page from swap cache: in
2849 * those cases unuse_pte()'s pte_same() test will fail; but
2850 * there's also a KSM case which does need to charge the page.
2852 if (!PageSwapCache(page
)) {
2855 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, 1, memcgp
, true);
2860 return __mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, memcgp
);
2863 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup
*memcg
)
2865 if (mem_cgroup_disabled())
2869 __mem_cgroup_cancel_charge(memcg
, 1);
2873 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*memcg
,
2874 enum charge_type ctype
)
2876 if (mem_cgroup_disabled())
2880 cgroup_exclude_rmdir(&memcg
->css
);
2882 __mem_cgroup_commit_charge(memcg
, page
, 1, ctype
, true);
2884 * Now swap is on-memory. This means this page may be
2885 * counted both as mem and swap....double count.
2886 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2887 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2888 * may call delete_from_swap_cache() before reach here.
2890 if (do_swap_account
&& PageSwapCache(page
)) {
2891 swp_entry_t ent
= {.val
= page_private(page
)};
2892 mem_cgroup_uncharge_swap(ent
);
2895 * At swapin, we may charge account against cgroup which has no tasks.
2896 * So, rmdir()->pre_destroy() can be called while we do this charge.
2897 * In that case, we need to call pre_destroy() again. check it here.
2899 cgroup_release_and_wakeup_rmdir(&memcg
->css
);
2902 void mem_cgroup_commit_charge_swapin(struct page
*page
,
2903 struct mem_cgroup
*memcg
)
2905 __mem_cgroup_commit_charge_swapin(page
, memcg
,
2906 MEM_CGROUP_CHARGE_TYPE_ANON
);
2909 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
2912 struct mem_cgroup
*memcg
= NULL
;
2913 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
2916 if (mem_cgroup_disabled())
2918 if (PageCompound(page
))
2921 if (!PageSwapCache(page
))
2922 ret
= mem_cgroup_charge_common(page
, mm
, gfp_mask
, type
);
2923 else { /* page is swapcache/shmem */
2924 ret
= __mem_cgroup_try_charge_swapin(mm
, page
,
2927 __mem_cgroup_commit_charge_swapin(page
, memcg
, type
);
2932 static void mem_cgroup_do_uncharge(struct mem_cgroup
*memcg
,
2933 unsigned int nr_pages
,
2934 const enum charge_type ctype
)
2936 struct memcg_batch_info
*batch
= NULL
;
2937 bool uncharge_memsw
= true;
2939 /* If swapout, usage of swap doesn't decrease */
2940 if (!do_swap_account
|| ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
)
2941 uncharge_memsw
= false;
2943 batch
= ¤t
->memcg_batch
;
2945 * In usual, we do css_get() when we remember memcg pointer.
2946 * But in this case, we keep res->usage until end of a series of
2947 * uncharges. Then, it's ok to ignore memcg's refcnt.
2950 batch
->memcg
= memcg
;
2952 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2953 * In those cases, all pages freed continuously can be expected to be in
2954 * the same cgroup and we have chance to coalesce uncharges.
2955 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2956 * because we want to do uncharge as soon as possible.
2959 if (!batch
->do_batch
|| test_thread_flag(TIF_MEMDIE
))
2960 goto direct_uncharge
;
2963 goto direct_uncharge
;
2966 * In typical case, batch->memcg == mem. This means we can
2967 * merge a series of uncharges to an uncharge of res_counter.
2968 * If not, we uncharge res_counter ony by one.
2970 if (batch
->memcg
!= memcg
)
2971 goto direct_uncharge
;
2972 /* remember freed charge and uncharge it later */
2975 batch
->memsw_nr_pages
++;
2978 res_counter_uncharge(&memcg
->res
, nr_pages
* PAGE_SIZE
);
2980 res_counter_uncharge(&memcg
->memsw
, nr_pages
* PAGE_SIZE
);
2981 if (unlikely(batch
->memcg
!= memcg
))
2982 memcg_oom_recover(memcg
);
2986 * uncharge if !page_mapped(page)
2988 static struct mem_cgroup
*
2989 __mem_cgroup_uncharge_common(struct page
*page
, enum charge_type ctype
,
2992 struct mem_cgroup
*memcg
= NULL
;
2993 unsigned int nr_pages
= 1;
2994 struct page_cgroup
*pc
;
2997 if (mem_cgroup_disabled())
3000 VM_BUG_ON(PageSwapCache(page
));
3002 if (PageTransHuge(page
)) {
3003 nr_pages
<<= compound_order(page
);
3004 VM_BUG_ON(!PageTransHuge(page
));
3007 * Check if our page_cgroup is valid
3009 pc
= lookup_page_cgroup(page
);
3010 if (unlikely(!PageCgroupUsed(pc
)))
3013 lock_page_cgroup(pc
);
3015 memcg
= pc
->mem_cgroup
;
3017 if (!PageCgroupUsed(pc
))
3020 anon
= PageAnon(page
);
3023 case MEM_CGROUP_CHARGE_TYPE_ANON
:
3025 * Generally PageAnon tells if it's the anon statistics to be
3026 * updated; but sometimes e.g. mem_cgroup_uncharge_page() is
3027 * used before page reached the stage of being marked PageAnon.
3031 case MEM_CGROUP_CHARGE_TYPE_DROP
:
3032 /* See mem_cgroup_prepare_migration() */
3033 if (page_mapped(page
))
3036 * Pages under migration may not be uncharged. But
3037 * end_migration() /must/ be the one uncharging the
3038 * unused post-migration page and so it has to call
3039 * here with the migration bit still set. See the
3040 * res_counter handling below.
3042 if (!end_migration
&& PageCgroupMigration(pc
))
3045 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT
:
3046 if (!PageAnon(page
)) { /* Shared memory */
3047 if (page
->mapping
&& !page_is_file_cache(page
))
3049 } else if (page_mapped(page
)) /* Anon */
3056 mem_cgroup_charge_statistics(memcg
, anon
, -nr_pages
);
3058 ClearPageCgroupUsed(pc
);
3060 * pc->mem_cgroup is not cleared here. It will be accessed when it's
3061 * freed from LRU. This is safe because uncharged page is expected not
3062 * to be reused (freed soon). Exception is SwapCache, it's handled by
3063 * special functions.
3066 unlock_page_cgroup(pc
);
3068 * even after unlock, we have memcg->res.usage here and this memcg
3069 * will never be freed.
3071 memcg_check_events(memcg
, page
);
3072 if (do_swap_account
&& ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
) {
3073 mem_cgroup_swap_statistics(memcg
, true);
3074 mem_cgroup_get(memcg
);
3077 * Migration does not charge the res_counter for the
3078 * replacement page, so leave it alone when phasing out the
3079 * page that is unused after the migration.
3081 if (!end_migration
&& !mem_cgroup_is_root(memcg
))
3082 mem_cgroup_do_uncharge(memcg
, nr_pages
, ctype
);
3087 unlock_page_cgroup(pc
);
3091 void mem_cgroup_uncharge_page(struct page
*page
)
3094 if (page_mapped(page
))
3096 VM_BUG_ON(page
->mapping
&& !PageAnon(page
));
3097 if (PageSwapCache(page
))
3099 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_ANON
, false);
3102 void mem_cgroup_uncharge_cache_page(struct page
*page
)
3104 VM_BUG_ON(page_mapped(page
));
3105 VM_BUG_ON(page
->mapping
);
3106 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_CACHE
, false);
3110 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
3111 * In that cases, pages are freed continuously and we can expect pages
3112 * are in the same memcg. All these calls itself limits the number of
3113 * pages freed at once, then uncharge_start/end() is called properly.
3114 * This may be called prural(2) times in a context,
3117 void mem_cgroup_uncharge_start(void)
3119 current
->memcg_batch
.do_batch
++;
3120 /* We can do nest. */
3121 if (current
->memcg_batch
.do_batch
== 1) {
3122 current
->memcg_batch
.memcg
= NULL
;
3123 current
->memcg_batch
.nr_pages
= 0;
3124 current
->memcg_batch
.memsw_nr_pages
= 0;
3128 void mem_cgroup_uncharge_end(void)
3130 struct memcg_batch_info
*batch
= ¤t
->memcg_batch
;
3132 if (!batch
->do_batch
)
3136 if (batch
->do_batch
) /* If stacked, do nothing. */
3142 * This "batch->memcg" is valid without any css_get/put etc...
3143 * bacause we hide charges behind us.
3145 if (batch
->nr_pages
)
3146 res_counter_uncharge(&batch
->memcg
->res
,
3147 batch
->nr_pages
* PAGE_SIZE
);
3148 if (batch
->memsw_nr_pages
)
3149 res_counter_uncharge(&batch
->memcg
->memsw
,
3150 batch
->memsw_nr_pages
* PAGE_SIZE
);
3151 memcg_oom_recover(batch
->memcg
);
3152 /* forget this pointer (for sanity check) */
3153 batch
->memcg
= NULL
;
3158 * called after __delete_from_swap_cache() and drop "page" account.
3159 * memcg information is recorded to swap_cgroup of "ent"
3162 mem_cgroup_uncharge_swapcache(struct page
*page
, swp_entry_t ent
, bool swapout
)
3164 struct mem_cgroup
*memcg
;
3165 int ctype
= MEM_CGROUP_CHARGE_TYPE_SWAPOUT
;
3167 if (!swapout
) /* this was a swap cache but the swap is unused ! */
3168 ctype
= MEM_CGROUP_CHARGE_TYPE_DROP
;
3170 memcg
= __mem_cgroup_uncharge_common(page
, ctype
, false);
3173 * record memcg information, if swapout && memcg != NULL,
3174 * mem_cgroup_get() was called in uncharge().
3176 if (do_swap_account
&& swapout
&& memcg
)
3177 swap_cgroup_record(ent
, css_id(&memcg
->css
));
3181 #ifdef CONFIG_MEMCG_SWAP
3183 * called from swap_entry_free(). remove record in swap_cgroup and
3184 * uncharge "memsw" account.
3186 void mem_cgroup_uncharge_swap(swp_entry_t ent
)
3188 struct mem_cgroup
*memcg
;
3191 if (!do_swap_account
)
3194 id
= swap_cgroup_record(ent
, 0);
3196 memcg
= mem_cgroup_lookup(id
);
3199 * We uncharge this because swap is freed.
3200 * This memcg can be obsolete one. We avoid calling css_tryget
3202 if (!mem_cgroup_is_root(memcg
))
3203 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
3204 mem_cgroup_swap_statistics(memcg
, false);
3205 mem_cgroup_put(memcg
);
3211 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3212 * @entry: swap entry to be moved
3213 * @from: mem_cgroup which the entry is moved from
3214 * @to: mem_cgroup which the entry is moved to
3216 * It succeeds only when the swap_cgroup's record for this entry is the same
3217 * as the mem_cgroup's id of @from.
3219 * Returns 0 on success, -EINVAL on failure.
3221 * The caller must have charged to @to, IOW, called res_counter_charge() about
3222 * both res and memsw, and called css_get().
3224 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
3225 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
3227 unsigned short old_id
, new_id
;
3229 old_id
= css_id(&from
->css
);
3230 new_id
= css_id(&to
->css
);
3232 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
3233 mem_cgroup_swap_statistics(from
, false);
3234 mem_cgroup_swap_statistics(to
, true);
3236 * This function is only called from task migration context now.
3237 * It postpones res_counter and refcount handling till the end
3238 * of task migration(mem_cgroup_clear_mc()) for performance
3239 * improvement. But we cannot postpone mem_cgroup_get(to)
3240 * because if the process that has been moved to @to does
3241 * swap-in, the refcount of @to might be decreased to 0.
3249 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
3250 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
3257 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3260 void mem_cgroup_prepare_migration(struct page
*page
, struct page
*newpage
,
3261 struct mem_cgroup
**memcgp
)
3263 struct mem_cgroup
*memcg
= NULL
;
3264 struct page_cgroup
*pc
;
3265 enum charge_type ctype
;
3269 VM_BUG_ON(PageTransHuge(page
));
3270 if (mem_cgroup_disabled())
3273 pc
= lookup_page_cgroup(page
);
3274 lock_page_cgroup(pc
);
3275 if (PageCgroupUsed(pc
)) {
3276 memcg
= pc
->mem_cgroup
;
3277 css_get(&memcg
->css
);
3279 * At migrating an anonymous page, its mapcount goes down
3280 * to 0 and uncharge() will be called. But, even if it's fully
3281 * unmapped, migration may fail and this page has to be
3282 * charged again. We set MIGRATION flag here and delay uncharge
3283 * until end_migration() is called
3285 * Corner Case Thinking
3287 * When the old page was mapped as Anon and it's unmap-and-freed
3288 * while migration was ongoing.
3289 * If unmap finds the old page, uncharge() of it will be delayed
3290 * until end_migration(). If unmap finds a new page, it's
3291 * uncharged when it make mapcount to be 1->0. If unmap code
3292 * finds swap_migration_entry, the new page will not be mapped
3293 * and end_migration() will find it(mapcount==0).
3296 * When the old page was mapped but migraion fails, the kernel
3297 * remaps it. A charge for it is kept by MIGRATION flag even
3298 * if mapcount goes down to 0. We can do remap successfully
3299 * without charging it again.
3302 * The "old" page is under lock_page() until the end of
3303 * migration, so, the old page itself will not be swapped-out.
3304 * If the new page is swapped out before end_migraton, our
3305 * hook to usual swap-out path will catch the event.
3308 SetPageCgroupMigration(pc
);
3310 unlock_page_cgroup(pc
);
3312 * If the page is not charged at this point,
3320 * We charge new page before it's used/mapped. So, even if unlock_page()
3321 * is called before end_migration, we can catch all events on this new
3322 * page. In the case new page is migrated but not remapped, new page's
3323 * mapcount will be finally 0 and we call uncharge in end_migration().
3326 ctype
= MEM_CGROUP_CHARGE_TYPE_ANON
;
3328 ctype
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3330 * The page is committed to the memcg, but it's not actually
3331 * charged to the res_counter since we plan on replacing the
3332 * old one and only one page is going to be left afterwards.
3334 __mem_cgroup_commit_charge(memcg
, newpage
, 1, ctype
, false);
3337 /* remove redundant charge if migration failed*/
3338 void mem_cgroup_end_migration(struct mem_cgroup
*memcg
,
3339 struct page
*oldpage
, struct page
*newpage
, bool migration_ok
)
3341 struct page
*used
, *unused
;
3342 struct page_cgroup
*pc
;
3347 /* blocks rmdir() */
3348 cgroup_exclude_rmdir(&memcg
->css
);
3349 if (!migration_ok
) {
3356 anon
= PageAnon(used
);
3357 __mem_cgroup_uncharge_common(unused
,
3358 anon
? MEM_CGROUP_CHARGE_TYPE_ANON
3359 : MEM_CGROUP_CHARGE_TYPE_CACHE
,
3361 css_put(&memcg
->css
);
3363 * We disallowed uncharge of pages under migration because mapcount
3364 * of the page goes down to zero, temporarly.
3365 * Clear the flag and check the page should be charged.
3367 pc
= lookup_page_cgroup(oldpage
);
3368 lock_page_cgroup(pc
);
3369 ClearPageCgroupMigration(pc
);
3370 unlock_page_cgroup(pc
);
3373 * If a page is a file cache, radix-tree replacement is very atomic
3374 * and we can skip this check. When it was an Anon page, its mapcount
3375 * goes down to 0. But because we added MIGRATION flage, it's not
3376 * uncharged yet. There are several case but page->mapcount check
3377 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3378 * check. (see prepare_charge() also)
3381 mem_cgroup_uncharge_page(used
);
3383 * At migration, we may charge account against cgroup which has no
3385 * So, rmdir()->pre_destroy() can be called while we do this charge.
3386 * In that case, we need to call pre_destroy() again. check it here.
3388 cgroup_release_and_wakeup_rmdir(&memcg
->css
);
3392 * At replace page cache, newpage is not under any memcg but it's on
3393 * LRU. So, this function doesn't touch res_counter but handles LRU
3394 * in correct way. Both pages are locked so we cannot race with uncharge.
3396 void mem_cgroup_replace_page_cache(struct page
*oldpage
,
3397 struct page
*newpage
)
3399 struct mem_cgroup
*memcg
= NULL
;
3400 struct page_cgroup
*pc
;
3401 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3403 if (mem_cgroup_disabled())
3406 pc
= lookup_page_cgroup(oldpage
);
3407 /* fix accounting on old pages */
3408 lock_page_cgroup(pc
);
3409 if (PageCgroupUsed(pc
)) {
3410 memcg
= pc
->mem_cgroup
;
3411 mem_cgroup_charge_statistics(memcg
, false, -1);
3412 ClearPageCgroupUsed(pc
);
3414 unlock_page_cgroup(pc
);
3417 * When called from shmem_replace_page(), in some cases the
3418 * oldpage has already been charged, and in some cases not.
3423 * Even if newpage->mapping was NULL before starting replacement,
3424 * the newpage may be on LRU(or pagevec for LRU) already. We lock
3425 * LRU while we overwrite pc->mem_cgroup.
3427 __mem_cgroup_commit_charge(memcg
, newpage
, 1, type
, true);
3430 #ifdef CONFIG_DEBUG_VM
3431 static struct page_cgroup
*lookup_page_cgroup_used(struct page
*page
)
3433 struct page_cgroup
*pc
;
3435 pc
= lookup_page_cgroup(page
);
3437 * Can be NULL while feeding pages into the page allocator for
3438 * the first time, i.e. during boot or memory hotplug;
3439 * or when mem_cgroup_disabled().
3441 if (likely(pc
) && PageCgroupUsed(pc
))
3446 bool mem_cgroup_bad_page_check(struct page
*page
)
3448 if (mem_cgroup_disabled())
3451 return lookup_page_cgroup_used(page
) != NULL
;
3454 void mem_cgroup_print_bad_page(struct page
*page
)
3456 struct page_cgroup
*pc
;
3458 pc
= lookup_page_cgroup_used(page
);
3460 printk(KERN_ALERT
"pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
3461 pc
, pc
->flags
, pc
->mem_cgroup
);
3466 static DEFINE_MUTEX(set_limit_mutex
);
3468 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
3469 unsigned long long val
)
3472 u64 memswlimit
, memlimit
;
3474 int children
= mem_cgroup_count_children(memcg
);
3475 u64 curusage
, oldusage
;
3479 * For keeping hierarchical_reclaim simple, how long we should retry
3480 * is depends on callers. We set our retry-count to be function
3481 * of # of children which we should visit in this loop.
3483 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
* children
;
3485 oldusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3488 while (retry_count
) {
3489 if (signal_pending(current
)) {
3494 * Rather than hide all in some function, I do this in
3495 * open coded manner. You see what this really does.
3496 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
3498 mutex_lock(&set_limit_mutex
);
3499 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3500 if (memswlimit
< val
) {
3502 mutex_unlock(&set_limit_mutex
);
3506 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3510 ret
= res_counter_set_limit(&memcg
->res
, val
);
3512 if (memswlimit
== val
)
3513 memcg
->memsw_is_minimum
= true;
3515 memcg
->memsw_is_minimum
= false;
3517 mutex_unlock(&set_limit_mutex
);
3522 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3523 MEM_CGROUP_RECLAIM_SHRINK
);
3524 curusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3525 /* Usage is reduced ? */
3526 if (curusage
>= oldusage
)
3529 oldusage
= curusage
;
3531 if (!ret
&& enlarge
)
3532 memcg_oom_recover(memcg
);
3537 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
3538 unsigned long long val
)
3541 u64 memlimit
, memswlimit
, oldusage
, curusage
;
3542 int children
= mem_cgroup_count_children(memcg
);
3546 /* see mem_cgroup_resize_res_limit */
3547 retry_count
= children
* MEM_CGROUP_RECLAIM_RETRIES
;
3548 oldusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3549 while (retry_count
) {
3550 if (signal_pending(current
)) {
3555 * Rather than hide all in some function, I do this in
3556 * open coded manner. You see what this really does.
3557 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
3559 mutex_lock(&set_limit_mutex
);
3560 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3561 if (memlimit
> val
) {
3563 mutex_unlock(&set_limit_mutex
);
3566 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3567 if (memswlimit
< val
)
3569 ret
= res_counter_set_limit(&memcg
->memsw
, val
);
3571 if (memlimit
== val
)
3572 memcg
->memsw_is_minimum
= true;
3574 memcg
->memsw_is_minimum
= false;
3576 mutex_unlock(&set_limit_mutex
);
3581 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3582 MEM_CGROUP_RECLAIM_NOSWAP
|
3583 MEM_CGROUP_RECLAIM_SHRINK
);
3584 curusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3585 /* Usage is reduced ? */
3586 if (curusage
>= oldusage
)
3589 oldusage
= curusage
;
3591 if (!ret
&& enlarge
)
3592 memcg_oom_recover(memcg
);
3596 unsigned long mem_cgroup_soft_limit_reclaim(struct zone
*zone
, int order
,
3598 unsigned long *total_scanned
)
3600 unsigned long nr_reclaimed
= 0;
3601 struct mem_cgroup_per_zone
*mz
, *next_mz
= NULL
;
3602 unsigned long reclaimed
;
3604 struct mem_cgroup_tree_per_zone
*mctz
;
3605 unsigned long long excess
;
3606 unsigned long nr_scanned
;
3611 mctz
= soft_limit_tree_node_zone(zone_to_nid(zone
), zone_idx(zone
));
3613 * This loop can run a while, specially if mem_cgroup's continuously
3614 * keep exceeding their soft limit and putting the system under
3621 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
3626 reclaimed
= mem_cgroup_soft_reclaim(mz
->memcg
, zone
,
3627 gfp_mask
, &nr_scanned
);
3628 nr_reclaimed
+= reclaimed
;
3629 *total_scanned
+= nr_scanned
;
3630 spin_lock(&mctz
->lock
);
3633 * If we failed to reclaim anything from this memory cgroup
3634 * it is time to move on to the next cgroup
3640 * Loop until we find yet another one.
3642 * By the time we get the soft_limit lock
3643 * again, someone might have aded the
3644 * group back on the RB tree. Iterate to
3645 * make sure we get a different mem.
3646 * mem_cgroup_largest_soft_limit_node returns
3647 * NULL if no other cgroup is present on
3651 __mem_cgroup_largest_soft_limit_node(mctz
);
3653 css_put(&next_mz
->memcg
->css
);
3654 else /* next_mz == NULL or other memcg */
3658 __mem_cgroup_remove_exceeded(mz
->memcg
, mz
, mctz
);
3659 excess
= res_counter_soft_limit_excess(&mz
->memcg
->res
);
3661 * One school of thought says that we should not add
3662 * back the node to the tree if reclaim returns 0.
3663 * But our reclaim could return 0, simply because due
3664 * to priority we are exposing a smaller subset of
3665 * memory to reclaim from. Consider this as a longer
3668 /* If excess == 0, no tree ops */
3669 __mem_cgroup_insert_exceeded(mz
->memcg
, mz
, mctz
, excess
);
3670 spin_unlock(&mctz
->lock
);
3671 css_put(&mz
->memcg
->css
);
3674 * Could not reclaim anything and there are no more
3675 * mem cgroups to try or we seem to be looping without
3676 * reclaiming anything.
3678 if (!nr_reclaimed
&&
3680 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
3682 } while (!nr_reclaimed
);
3684 css_put(&next_mz
->memcg
->css
);
3685 return nr_reclaimed
;
3689 * Traverse a specified page_cgroup list and try to drop them all. This doesn't
3690 * reclaim the pages page themselves - it just removes the page_cgroups.
3691 * Returns true if some page_cgroups were not freed, indicating that the caller
3692 * must retry this operation.
3694 static bool mem_cgroup_force_empty_list(struct mem_cgroup
*memcg
,
3695 int node
, int zid
, enum lru_list lru
)
3697 struct mem_cgroup_per_zone
*mz
;
3698 unsigned long flags
, loop
;
3699 struct list_head
*list
;
3703 zone
= &NODE_DATA(node
)->node_zones
[zid
];
3704 mz
= mem_cgroup_zoneinfo(memcg
, node
, zid
);
3705 list
= &mz
->lruvec
.lists
[lru
];
3707 loop
= mz
->lru_size
[lru
];
3708 /* give some margin against EBUSY etc...*/
3712 struct page_cgroup
*pc
;
3715 spin_lock_irqsave(&zone
->lru_lock
, flags
);
3716 if (list_empty(list
)) {
3717 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3720 page
= list_entry(list
->prev
, struct page
, lru
);
3722 list_move(&page
->lru
, list
);
3724 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3727 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3729 pc
= lookup_page_cgroup(page
);
3731 if (mem_cgroup_move_parent(page
, pc
, memcg
)) {
3732 /* found lock contention or "pc" is obsolete. */
3738 return !list_empty(list
);
3742 * make mem_cgroup's charge to be 0 if there is no task by moving
3743 * all the charges and pages to the parent.
3744 * This enables deleting this mem_cgroup.
3746 * Caller is responsible for holding css reference on the memcg.
3748 static int mem_cgroup_reparent_charges(struct mem_cgroup
*memcg
)
3750 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
3755 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
))
3757 /* This is for making all *used* pages to be on LRU. */
3758 lru_add_drain_all();
3759 drain_all_stock_sync(memcg
);
3761 mem_cgroup_start_move(memcg
);
3762 for_each_node_state(node
, N_HIGH_MEMORY
) {
3763 for (zid
= 0; !ret
&& zid
< MAX_NR_ZONES
; zid
++) {
3766 ret
= mem_cgroup_force_empty_list(memcg
,
3775 mem_cgroup_end_move(memcg
);
3776 memcg_oom_recover(memcg
);
3778 /* "ret" should also be checked to ensure all lists are empty. */
3779 } while (res_counter_read_u64(&memcg
->res
, RES_USAGE
) > 0 || ret
);
3785 * Reclaims as many pages from the given memcg as possible and moves
3786 * the rest to the parent.
3788 * Caller is responsible for holding css reference for memcg.
3790 static int mem_cgroup_force_empty(struct mem_cgroup
*memcg
)
3792 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
3793 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
3795 /* returns EBUSY if there is a task or if we come here twice. */
3796 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
))
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 */
3802 while (nr_retries
&& res_counter_read_u64(&memcg
->res
, RES_USAGE
) > 0) {
3805 if (signal_pending(current
))
3808 progress
= try_to_free_mem_cgroup_pages(memcg
, GFP_KERNEL
,
3812 /* maybe some writeback is necessary */
3813 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3818 return mem_cgroup_reparent_charges(memcg
);
3821 static int mem_cgroup_force_empty_write(struct cgroup
*cont
, unsigned int event
)
3823 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3826 css_get(&memcg
->css
);
3827 ret
= mem_cgroup_force_empty(memcg
);
3828 css_put(&memcg
->css
);
3834 static u64
mem_cgroup_hierarchy_read(struct cgroup
*cont
, struct cftype
*cft
)
3836 return mem_cgroup_from_cont(cont
)->use_hierarchy
;
3839 static int mem_cgroup_hierarchy_write(struct cgroup
*cont
, struct cftype
*cft
,
3843 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3844 struct cgroup
*parent
= cont
->parent
;
3845 struct mem_cgroup
*parent_memcg
= NULL
;
3848 parent_memcg
= mem_cgroup_from_cont(parent
);
3852 if (memcg
->use_hierarchy
== val
)
3856 * If parent's use_hierarchy is set, we can't make any modifications
3857 * in the child subtrees. If it is unset, then the change can
3858 * occur, provided the current cgroup has no children.
3860 * For the root cgroup, parent_mem is NULL, we allow value to be
3861 * set if there are no children.
3863 if ((!parent_memcg
|| !parent_memcg
->use_hierarchy
) &&
3864 (val
== 1 || val
== 0)) {
3865 if (list_empty(&cont
->children
))
3866 memcg
->use_hierarchy
= val
;
3879 static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup
*memcg
,
3880 enum mem_cgroup_stat_index idx
)
3882 struct mem_cgroup
*iter
;
3885 /* Per-cpu values can be negative, use a signed accumulator */
3886 for_each_mem_cgroup_tree(iter
, memcg
)
3887 val
+= mem_cgroup_read_stat(iter
, idx
);
3889 if (val
< 0) /* race ? */
3894 static inline u64
mem_cgroup_usage(struct mem_cgroup
*memcg
, bool swap
)
3898 if (!mem_cgroup_is_root(memcg
)) {
3900 return res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3902 return res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3905 val
= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_CACHE
);
3906 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_RSS
);
3909 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_SWAP
);
3911 return val
<< PAGE_SHIFT
;
3914 static ssize_t
mem_cgroup_read(struct cgroup
*cont
, struct cftype
*cft
,
3915 struct file
*file
, char __user
*buf
,
3916 size_t nbytes
, loff_t
*ppos
)
3918 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3921 int type
, name
, len
;
3923 type
= MEMFILE_TYPE(cft
->private);
3924 name
= MEMFILE_ATTR(cft
->private);
3926 if (!do_swap_account
&& type
== _MEMSWAP
)
3931 if (name
== RES_USAGE
)
3932 val
= mem_cgroup_usage(memcg
, false);
3934 val
= res_counter_read_u64(&memcg
->res
, name
);
3937 if (name
== RES_USAGE
)
3938 val
= mem_cgroup_usage(memcg
, true);
3940 val
= res_counter_read_u64(&memcg
->memsw
, name
);
3946 len
= scnprintf(str
, sizeof(str
), "%llu\n", (unsigned long long)val
);
3947 return simple_read_from_buffer(buf
, nbytes
, ppos
, str
, len
);
3950 * The user of this function is...
3953 static int mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
3956 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3958 unsigned long long val
;
3961 type
= MEMFILE_TYPE(cft
->private);
3962 name
= MEMFILE_ATTR(cft
->private);
3964 if (!do_swap_account
&& type
== _MEMSWAP
)
3969 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
3973 /* This function does all necessary parse...reuse it */
3974 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3978 ret
= mem_cgroup_resize_limit(memcg
, val
);
3980 ret
= mem_cgroup_resize_memsw_limit(memcg
, val
);
3982 case RES_SOFT_LIMIT
:
3983 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3987 * For memsw, soft limits are hard to implement in terms
3988 * of semantics, for now, we support soft limits for
3989 * control without swap
3992 ret
= res_counter_set_soft_limit(&memcg
->res
, val
);
3997 ret
= -EINVAL
; /* should be BUG() ? */
4003 static void memcg_get_hierarchical_limit(struct mem_cgroup
*memcg
,
4004 unsigned long long *mem_limit
, unsigned long long *memsw_limit
)
4006 struct cgroup
*cgroup
;
4007 unsigned long long min_limit
, min_memsw_limit
, tmp
;
4009 min_limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
4010 min_memsw_limit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
4011 cgroup
= memcg
->css
.cgroup
;
4012 if (!memcg
->use_hierarchy
)
4015 while (cgroup
->parent
) {
4016 cgroup
= cgroup
->parent
;
4017 memcg
= mem_cgroup_from_cont(cgroup
);
4018 if (!memcg
->use_hierarchy
)
4020 tmp
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
4021 min_limit
= min(min_limit
, tmp
);
4022 tmp
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
4023 min_memsw_limit
= min(min_memsw_limit
, tmp
);
4026 *mem_limit
= min_limit
;
4027 *memsw_limit
= min_memsw_limit
;
4030 static int mem_cgroup_reset(struct cgroup
*cont
, unsigned int event
)
4032 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4035 type
= MEMFILE_TYPE(event
);
4036 name
= MEMFILE_ATTR(event
);
4038 if (!do_swap_account
&& type
== _MEMSWAP
)
4044 res_counter_reset_max(&memcg
->res
);
4046 res_counter_reset_max(&memcg
->memsw
);
4050 res_counter_reset_failcnt(&memcg
->res
);
4052 res_counter_reset_failcnt(&memcg
->memsw
);
4059 static u64
mem_cgroup_move_charge_read(struct cgroup
*cgrp
,
4062 return mem_cgroup_from_cont(cgrp
)->move_charge_at_immigrate
;
4066 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4067 struct cftype
*cft
, u64 val
)
4069 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4071 if (val
>= (1 << NR_MOVE_TYPE
))
4074 * We check this value several times in both in can_attach() and
4075 * attach(), so we need cgroup lock to prevent this value from being
4079 memcg
->move_charge_at_immigrate
= val
;
4085 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4086 struct cftype
*cft
, u64 val
)
4093 static int memcg_numa_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
4097 unsigned long total_nr
, file_nr
, anon_nr
, unevictable_nr
;
4098 unsigned long node_nr
;
4099 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4101 total_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL
);
4102 seq_printf(m
, "total=%lu", total_nr
);
4103 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4104 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL
);
4105 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4109 file_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL_FILE
);
4110 seq_printf(m
, "file=%lu", file_nr
);
4111 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4112 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4114 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4118 anon_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL_ANON
);
4119 seq_printf(m
, "anon=%lu", anon_nr
);
4120 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4121 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4123 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4127 unevictable_nr
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_UNEVICTABLE
));
4128 seq_printf(m
, "unevictable=%lu", unevictable_nr
);
4129 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4130 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4131 BIT(LRU_UNEVICTABLE
));
4132 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4137 #endif /* CONFIG_NUMA */
4139 static const char * const mem_cgroup_lru_names
[] = {
4147 static inline void mem_cgroup_lru_names_not_uptodate(void)
4149 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names
) != NR_LRU_LISTS
);
4152 static int memcg_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
4155 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4156 struct mem_cgroup
*mi
;
4159 for (i
= 0; i
< MEM_CGROUP_STAT_NSTATS
; i
++) {
4160 if (i
== MEM_CGROUP_STAT_SWAP
&& !do_swap_account
)
4162 seq_printf(m
, "%s %ld\n", mem_cgroup_stat_names
[i
],
4163 mem_cgroup_read_stat(memcg
, i
) * PAGE_SIZE
);
4166 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++)
4167 seq_printf(m
, "%s %lu\n", mem_cgroup_events_names
[i
],
4168 mem_cgroup_read_events(memcg
, i
));
4170 for (i
= 0; i
< NR_LRU_LISTS
; i
++)
4171 seq_printf(m
, "%s %lu\n", mem_cgroup_lru_names
[i
],
4172 mem_cgroup_nr_lru_pages(memcg
, BIT(i
)) * PAGE_SIZE
);
4174 /* Hierarchical information */
4176 unsigned long long limit
, memsw_limit
;
4177 memcg_get_hierarchical_limit(memcg
, &limit
, &memsw_limit
);
4178 seq_printf(m
, "hierarchical_memory_limit %llu\n", limit
);
4179 if (do_swap_account
)
4180 seq_printf(m
, "hierarchical_memsw_limit %llu\n",
4184 for (i
= 0; i
< MEM_CGROUP_STAT_NSTATS
; i
++) {
4187 if (i
== MEM_CGROUP_STAT_SWAP
&& !do_swap_account
)
4189 for_each_mem_cgroup_tree(mi
, memcg
)
4190 val
+= mem_cgroup_read_stat(mi
, i
) * PAGE_SIZE
;
4191 seq_printf(m
, "total_%s %lld\n", mem_cgroup_stat_names
[i
], val
);
4194 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
4195 unsigned long long val
= 0;
4197 for_each_mem_cgroup_tree(mi
, memcg
)
4198 val
+= mem_cgroup_read_events(mi
, i
);
4199 seq_printf(m
, "total_%s %llu\n",
4200 mem_cgroup_events_names
[i
], val
);
4203 for (i
= 0; i
< NR_LRU_LISTS
; i
++) {
4204 unsigned long long val
= 0;
4206 for_each_mem_cgroup_tree(mi
, memcg
)
4207 val
+= mem_cgroup_nr_lru_pages(mi
, BIT(i
)) * PAGE_SIZE
;
4208 seq_printf(m
, "total_%s %llu\n", mem_cgroup_lru_names
[i
], val
);
4211 #ifdef CONFIG_DEBUG_VM
4214 struct mem_cgroup_per_zone
*mz
;
4215 struct zone_reclaim_stat
*rstat
;
4216 unsigned long recent_rotated
[2] = {0, 0};
4217 unsigned long recent_scanned
[2] = {0, 0};
4219 for_each_online_node(nid
)
4220 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
4221 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
4222 rstat
= &mz
->lruvec
.reclaim_stat
;
4224 recent_rotated
[0] += rstat
->recent_rotated
[0];
4225 recent_rotated
[1] += rstat
->recent_rotated
[1];
4226 recent_scanned
[0] += rstat
->recent_scanned
[0];
4227 recent_scanned
[1] += rstat
->recent_scanned
[1];
4229 seq_printf(m
, "recent_rotated_anon %lu\n", recent_rotated
[0]);
4230 seq_printf(m
, "recent_rotated_file %lu\n", recent_rotated
[1]);
4231 seq_printf(m
, "recent_scanned_anon %lu\n", recent_scanned
[0]);
4232 seq_printf(m
, "recent_scanned_file %lu\n", recent_scanned
[1]);
4239 static u64
mem_cgroup_swappiness_read(struct cgroup
*cgrp
, struct cftype
*cft
)
4241 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4243 return mem_cgroup_swappiness(memcg
);
4246 static int mem_cgroup_swappiness_write(struct cgroup
*cgrp
, struct cftype
*cft
,
4249 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4250 struct mem_cgroup
*parent
;
4255 if (cgrp
->parent
== NULL
)
4258 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4262 /* If under hierarchy, only empty-root can set this value */
4263 if ((parent
->use_hierarchy
) ||
4264 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4269 memcg
->swappiness
= val
;
4276 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
4278 struct mem_cgroup_threshold_ary
*t
;
4284 t
= rcu_dereference(memcg
->thresholds
.primary
);
4286 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
4291 usage
= mem_cgroup_usage(memcg
, swap
);
4294 * current_threshold points to threshold just below or equal to usage.
4295 * If it's not true, a threshold was crossed after last
4296 * call of __mem_cgroup_threshold().
4298 i
= t
->current_threshold
;
4301 * Iterate backward over array of thresholds starting from
4302 * current_threshold and check if a threshold is crossed.
4303 * If none of thresholds below usage is crossed, we read
4304 * only one element of the array here.
4306 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
4307 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4309 /* i = current_threshold + 1 */
4313 * Iterate forward over array of thresholds starting from
4314 * current_threshold+1 and check if a threshold is crossed.
4315 * If none of thresholds above usage is crossed, we read
4316 * only one element of the array here.
4318 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
4319 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4321 /* Update current_threshold */
4322 t
->current_threshold
= i
- 1;
4327 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
4330 __mem_cgroup_threshold(memcg
, false);
4331 if (do_swap_account
)
4332 __mem_cgroup_threshold(memcg
, true);
4334 memcg
= parent_mem_cgroup(memcg
);
4338 static int compare_thresholds(const void *a
, const void *b
)
4340 const struct mem_cgroup_threshold
*_a
= a
;
4341 const struct mem_cgroup_threshold
*_b
= b
;
4343 return _a
->threshold
- _b
->threshold
;
4346 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*memcg
)
4348 struct mem_cgroup_eventfd_list
*ev
;
4350 list_for_each_entry(ev
, &memcg
->oom_notify
, list
)
4351 eventfd_signal(ev
->eventfd
, 1);
4355 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
)
4357 struct mem_cgroup
*iter
;
4359 for_each_mem_cgroup_tree(iter
, memcg
)
4360 mem_cgroup_oom_notify_cb(iter
);
4363 static int mem_cgroup_usage_register_event(struct cgroup
*cgrp
,
4364 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4366 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4367 struct mem_cgroup_thresholds
*thresholds
;
4368 struct mem_cgroup_threshold_ary
*new;
4369 int type
= MEMFILE_TYPE(cft
->private);
4370 u64 threshold
, usage
;
4373 ret
= res_counter_memparse_write_strategy(args
, &threshold
);
4377 mutex_lock(&memcg
->thresholds_lock
);
4380 thresholds
= &memcg
->thresholds
;
4381 else if (type
== _MEMSWAP
)
4382 thresholds
= &memcg
->memsw_thresholds
;
4386 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4388 /* Check if a threshold crossed before adding a new one */
4389 if (thresholds
->primary
)
4390 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4392 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
4394 /* Allocate memory for new array of thresholds */
4395 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
4403 /* Copy thresholds (if any) to new array */
4404 if (thresholds
->primary
) {
4405 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
4406 sizeof(struct mem_cgroup_threshold
));
4409 /* Add new threshold */
4410 new->entries
[size
- 1].eventfd
= eventfd
;
4411 new->entries
[size
- 1].threshold
= threshold
;
4413 /* Sort thresholds. Registering of new threshold isn't time-critical */
4414 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
4415 compare_thresholds
, NULL
);
4417 /* Find current threshold */
4418 new->current_threshold
= -1;
4419 for (i
= 0; i
< size
; i
++) {
4420 if (new->entries
[i
].threshold
<= usage
) {
4422 * new->current_threshold will not be used until
4423 * rcu_assign_pointer(), so it's safe to increment
4426 ++new->current_threshold
;
4431 /* Free old spare buffer and save old primary buffer as spare */
4432 kfree(thresholds
->spare
);
4433 thresholds
->spare
= thresholds
->primary
;
4435 rcu_assign_pointer(thresholds
->primary
, new);
4437 /* To be sure that nobody uses thresholds */
4441 mutex_unlock(&memcg
->thresholds_lock
);
4446 static void mem_cgroup_usage_unregister_event(struct cgroup
*cgrp
,
4447 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4449 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4450 struct mem_cgroup_thresholds
*thresholds
;
4451 struct mem_cgroup_threshold_ary
*new;
4452 int type
= MEMFILE_TYPE(cft
->private);
4456 mutex_lock(&memcg
->thresholds_lock
);
4458 thresholds
= &memcg
->thresholds
;
4459 else if (type
== _MEMSWAP
)
4460 thresholds
= &memcg
->memsw_thresholds
;
4464 if (!thresholds
->primary
)
4467 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4469 /* Check if a threshold crossed before removing */
4470 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4472 /* Calculate new number of threshold */
4474 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
4475 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
4479 new = thresholds
->spare
;
4481 /* Set thresholds array to NULL if we don't have thresholds */
4490 /* Copy thresholds and find current threshold */
4491 new->current_threshold
= -1;
4492 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
4493 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
4496 new->entries
[j
] = thresholds
->primary
->entries
[i
];
4497 if (new->entries
[j
].threshold
<= usage
) {
4499 * new->current_threshold will not be used
4500 * until rcu_assign_pointer(), so it's safe to increment
4503 ++new->current_threshold
;
4509 /* Swap primary and spare array */
4510 thresholds
->spare
= thresholds
->primary
;
4511 /* If all events are unregistered, free the spare array */
4513 kfree(thresholds
->spare
);
4514 thresholds
->spare
= NULL
;
4517 rcu_assign_pointer(thresholds
->primary
, new);
4519 /* To be sure that nobody uses thresholds */
4522 mutex_unlock(&memcg
->thresholds_lock
);
4525 static int mem_cgroup_oom_register_event(struct cgroup
*cgrp
,
4526 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4528 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4529 struct mem_cgroup_eventfd_list
*event
;
4530 int type
= MEMFILE_TYPE(cft
->private);
4532 BUG_ON(type
!= _OOM_TYPE
);
4533 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
4537 spin_lock(&memcg_oom_lock
);
4539 event
->eventfd
= eventfd
;
4540 list_add(&event
->list
, &memcg
->oom_notify
);
4542 /* already in OOM ? */
4543 if (atomic_read(&memcg
->under_oom
))
4544 eventfd_signal(eventfd
, 1);
4545 spin_unlock(&memcg_oom_lock
);
4550 static void mem_cgroup_oom_unregister_event(struct cgroup
*cgrp
,
4551 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4553 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4554 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
4555 int type
= MEMFILE_TYPE(cft
->private);
4557 BUG_ON(type
!= _OOM_TYPE
);
4559 spin_lock(&memcg_oom_lock
);
4561 list_for_each_entry_safe(ev
, tmp
, &memcg
->oom_notify
, list
) {
4562 if (ev
->eventfd
== eventfd
) {
4563 list_del(&ev
->list
);
4568 spin_unlock(&memcg_oom_lock
);
4571 static int mem_cgroup_oom_control_read(struct cgroup
*cgrp
,
4572 struct cftype
*cft
, struct cgroup_map_cb
*cb
)
4574 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4576 cb
->fill(cb
, "oom_kill_disable", memcg
->oom_kill_disable
);
4578 if (atomic_read(&memcg
->under_oom
))
4579 cb
->fill(cb
, "under_oom", 1);
4581 cb
->fill(cb
, "under_oom", 0);
4585 static int mem_cgroup_oom_control_write(struct cgroup
*cgrp
,
4586 struct cftype
*cft
, u64 val
)
4588 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4589 struct mem_cgroup
*parent
;
4591 /* cannot set to root cgroup and only 0 and 1 are allowed */
4592 if (!cgrp
->parent
|| !((val
== 0) || (val
== 1)))
4595 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4598 /* oom-kill-disable is a flag for subhierarchy. */
4599 if ((parent
->use_hierarchy
) ||
4600 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4604 memcg
->oom_kill_disable
= val
;
4606 memcg_oom_recover(memcg
);
4611 #ifdef CONFIG_MEMCG_KMEM
4612 static int memcg_init_kmem(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
4614 return mem_cgroup_sockets_init(memcg
, ss
);
4617 static void kmem_cgroup_destroy(struct mem_cgroup
*memcg
)
4619 mem_cgroup_sockets_destroy(memcg
);
4622 static int memcg_init_kmem(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
4627 static void kmem_cgroup_destroy(struct mem_cgroup
*memcg
)
4632 static struct cftype mem_cgroup_files
[] = {
4634 .name
= "usage_in_bytes",
4635 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
4636 .read
= mem_cgroup_read
,
4637 .register_event
= mem_cgroup_usage_register_event
,
4638 .unregister_event
= mem_cgroup_usage_unregister_event
,
4641 .name
= "max_usage_in_bytes",
4642 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
4643 .trigger
= mem_cgroup_reset
,
4644 .read
= mem_cgroup_read
,
4647 .name
= "limit_in_bytes",
4648 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
4649 .write_string
= mem_cgroup_write
,
4650 .read
= mem_cgroup_read
,
4653 .name
= "soft_limit_in_bytes",
4654 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
4655 .write_string
= mem_cgroup_write
,
4656 .read
= mem_cgroup_read
,
4660 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
4661 .trigger
= mem_cgroup_reset
,
4662 .read
= mem_cgroup_read
,
4666 .read_seq_string
= memcg_stat_show
,
4669 .name
= "force_empty",
4670 .trigger
= mem_cgroup_force_empty_write
,
4673 .name
= "use_hierarchy",
4674 .write_u64
= mem_cgroup_hierarchy_write
,
4675 .read_u64
= mem_cgroup_hierarchy_read
,
4678 .name
= "swappiness",
4679 .read_u64
= mem_cgroup_swappiness_read
,
4680 .write_u64
= mem_cgroup_swappiness_write
,
4683 .name
= "move_charge_at_immigrate",
4684 .read_u64
= mem_cgroup_move_charge_read
,
4685 .write_u64
= mem_cgroup_move_charge_write
,
4688 .name
= "oom_control",
4689 .read_map
= mem_cgroup_oom_control_read
,
4690 .write_u64
= mem_cgroup_oom_control_write
,
4691 .register_event
= mem_cgroup_oom_register_event
,
4692 .unregister_event
= mem_cgroup_oom_unregister_event
,
4693 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
4697 .name
= "numa_stat",
4698 .read_seq_string
= memcg_numa_stat_show
,
4701 #ifdef CONFIG_MEMCG_SWAP
4703 .name
= "memsw.usage_in_bytes",
4704 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
4705 .read
= mem_cgroup_read
,
4706 .register_event
= mem_cgroup_usage_register_event
,
4707 .unregister_event
= mem_cgroup_usage_unregister_event
,
4710 .name
= "memsw.max_usage_in_bytes",
4711 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
4712 .trigger
= mem_cgroup_reset
,
4713 .read
= mem_cgroup_read
,
4716 .name
= "memsw.limit_in_bytes",
4717 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
4718 .write_string
= mem_cgroup_write
,
4719 .read
= mem_cgroup_read
,
4722 .name
= "memsw.failcnt",
4723 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
4724 .trigger
= mem_cgroup_reset
,
4725 .read
= mem_cgroup_read
,
4728 { }, /* terminate */
4731 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4733 struct mem_cgroup_per_node
*pn
;
4734 struct mem_cgroup_per_zone
*mz
;
4735 int zone
, tmp
= node
;
4737 * This routine is called against possible nodes.
4738 * But it's BUG to call kmalloc() against offline node.
4740 * TODO: this routine can waste much memory for nodes which will
4741 * never be onlined. It's better to use memory hotplug callback
4744 if (!node_state(node
, N_NORMAL_MEMORY
))
4746 pn
= kzalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
4750 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4751 mz
= &pn
->zoneinfo
[zone
];
4752 lruvec_init(&mz
->lruvec
, &NODE_DATA(node
)->node_zones
[zone
]);
4753 mz
->usage_in_excess
= 0;
4754 mz
->on_tree
= false;
4757 memcg
->info
.nodeinfo
[node
] = pn
;
4761 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4763 kfree(memcg
->info
.nodeinfo
[node
]);
4766 static struct mem_cgroup
*mem_cgroup_alloc(void)
4768 struct mem_cgroup
*memcg
;
4769 int size
= sizeof(struct mem_cgroup
);
4771 /* Can be very big if MAX_NUMNODES is very big */
4772 if (size
< PAGE_SIZE
)
4773 memcg
= kzalloc(size
, GFP_KERNEL
);
4775 memcg
= vzalloc(size
);
4780 memcg
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4783 spin_lock_init(&memcg
->pcp_counter_lock
);
4787 if (size
< PAGE_SIZE
)
4795 * Helpers for freeing a kmalloc()ed/vzalloc()ed mem_cgroup by RCU,
4796 * but in process context. The work_freeing structure is overlaid
4797 * on the rcu_freeing structure, which itself is overlaid on memsw.
4799 static void free_work(struct work_struct
*work
)
4801 struct mem_cgroup
*memcg
;
4802 int size
= sizeof(struct mem_cgroup
);
4804 memcg
= container_of(work
, struct mem_cgroup
, work_freeing
);
4806 * We need to make sure that (at least for now), the jump label
4807 * destruction code runs outside of the cgroup lock. This is because
4808 * get_online_cpus(), which is called from the static_branch update,
4809 * can't be called inside the cgroup_lock. cpusets are the ones
4810 * enforcing this dependency, so if they ever change, we might as well.
4812 * schedule_work() will guarantee this happens. Be careful if you need
4813 * to move this code around, and make sure it is outside
4816 disarm_sock_keys(memcg
);
4817 if (size
< PAGE_SIZE
)
4823 static void free_rcu(struct rcu_head
*rcu_head
)
4825 struct mem_cgroup
*memcg
;
4827 memcg
= container_of(rcu_head
, struct mem_cgroup
, rcu_freeing
);
4828 INIT_WORK(&memcg
->work_freeing
, free_work
);
4829 schedule_work(&memcg
->work_freeing
);
4833 * At destroying mem_cgroup, references from swap_cgroup can remain.
4834 * (scanning all at force_empty is too costly...)
4836 * Instead of clearing all references at force_empty, we remember
4837 * the number of reference from swap_cgroup and free mem_cgroup when
4838 * it goes down to 0.
4840 * Removal of cgroup itself succeeds regardless of refs from swap.
4843 static void __mem_cgroup_free(struct mem_cgroup
*memcg
)
4847 mem_cgroup_remove_from_trees(memcg
);
4848 free_css_id(&mem_cgroup_subsys
, &memcg
->css
);
4851 free_mem_cgroup_per_zone_info(memcg
, node
);
4853 free_percpu(memcg
->stat
);
4854 call_rcu(&memcg
->rcu_freeing
, free_rcu
);
4857 static void mem_cgroup_get(struct mem_cgroup
*memcg
)
4859 atomic_inc(&memcg
->refcnt
);
4862 static void __mem_cgroup_put(struct mem_cgroup
*memcg
, int count
)
4864 if (atomic_sub_and_test(count
, &memcg
->refcnt
)) {
4865 struct mem_cgroup
*parent
= parent_mem_cgroup(memcg
);
4866 __mem_cgroup_free(memcg
);
4868 mem_cgroup_put(parent
);
4872 static void mem_cgroup_put(struct mem_cgroup
*memcg
)
4874 __mem_cgroup_put(memcg
, 1);
4878 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4880 struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*memcg
)
4882 if (!memcg
->res
.parent
)
4884 return mem_cgroup_from_res_counter(memcg
->res
.parent
, res
);
4886 EXPORT_SYMBOL(parent_mem_cgroup
);
4888 #ifdef CONFIG_MEMCG_SWAP
4889 static void __init
enable_swap_cgroup(void)
4891 if (!mem_cgroup_disabled() && really_do_swap_account
)
4892 do_swap_account
= 1;
4895 static void __init
enable_swap_cgroup(void)
4900 static int mem_cgroup_soft_limit_tree_init(void)
4902 struct mem_cgroup_tree_per_node
*rtpn
;
4903 struct mem_cgroup_tree_per_zone
*rtpz
;
4904 int tmp
, node
, zone
;
4906 for_each_node(node
) {
4908 if (!node_state(node
, N_NORMAL_MEMORY
))
4910 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
, tmp
);
4914 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
4916 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4917 rtpz
= &rtpn
->rb_tree_per_zone
[zone
];
4918 rtpz
->rb_root
= RB_ROOT
;
4919 spin_lock_init(&rtpz
->lock
);
4925 for_each_node(node
) {
4926 if (!soft_limit_tree
.rb_tree_per_node
[node
])
4928 kfree(soft_limit_tree
.rb_tree_per_node
[node
]);
4929 soft_limit_tree
.rb_tree_per_node
[node
] = NULL
;
4935 static struct cgroup_subsys_state
* __ref
4936 mem_cgroup_create(struct cgroup
*cont
)
4938 struct mem_cgroup
*memcg
, *parent
;
4939 long error
= -ENOMEM
;
4942 memcg
= mem_cgroup_alloc();
4944 return ERR_PTR(error
);
4947 if (alloc_mem_cgroup_per_zone_info(memcg
, node
))
4951 if (cont
->parent
== NULL
) {
4953 enable_swap_cgroup();
4955 if (mem_cgroup_soft_limit_tree_init())
4957 root_mem_cgroup
= memcg
;
4958 for_each_possible_cpu(cpu
) {
4959 struct memcg_stock_pcp
*stock
=
4960 &per_cpu(memcg_stock
, cpu
);
4961 INIT_WORK(&stock
->work
, drain_local_stock
);
4963 hotcpu_notifier(memcg_cpu_hotplug_callback
, 0);
4965 parent
= mem_cgroup_from_cont(cont
->parent
);
4966 memcg
->use_hierarchy
= parent
->use_hierarchy
;
4967 memcg
->oom_kill_disable
= parent
->oom_kill_disable
;
4970 if (parent
&& parent
->use_hierarchy
) {
4971 res_counter_init(&memcg
->res
, &parent
->res
);
4972 res_counter_init(&memcg
->memsw
, &parent
->memsw
);
4974 * We increment refcnt of the parent to ensure that we can
4975 * safely access it on res_counter_charge/uncharge.
4976 * This refcnt will be decremented when freeing this
4977 * mem_cgroup(see mem_cgroup_put).
4979 mem_cgroup_get(parent
);
4981 res_counter_init(&memcg
->res
, NULL
);
4982 res_counter_init(&memcg
->memsw
, NULL
);
4984 memcg
->last_scanned_node
= MAX_NUMNODES
;
4985 INIT_LIST_HEAD(&memcg
->oom_notify
);
4988 memcg
->swappiness
= mem_cgroup_swappiness(parent
);
4989 atomic_set(&memcg
->refcnt
, 1);
4990 memcg
->move_charge_at_immigrate
= 0;
4991 mutex_init(&memcg
->thresholds_lock
);
4992 spin_lock_init(&memcg
->move_lock
);
4994 error
= memcg_init_kmem(memcg
, &mem_cgroup_subsys
);
4997 * We call put now because our (and parent's) refcnts
4998 * are already in place. mem_cgroup_put() will internally
4999 * call __mem_cgroup_free, so return directly
5001 mem_cgroup_put(memcg
);
5002 return ERR_PTR(error
);
5006 __mem_cgroup_free(memcg
);
5007 return ERR_PTR(error
);
5010 static int mem_cgroup_pre_destroy(struct cgroup
*cont
)
5012 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
5015 css_get(&memcg
->css
);
5016 ret
= mem_cgroup_reparent_charges(memcg
);
5017 css_put(&memcg
->css
);
5022 static void mem_cgroup_destroy(struct cgroup
*cont
)
5024 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
5026 kmem_cgroup_destroy(memcg
);
5028 mem_cgroup_put(memcg
);
5032 /* Handlers for move charge at task migration. */
5033 #define PRECHARGE_COUNT_AT_ONCE 256
5034 static int mem_cgroup_do_precharge(unsigned long count
)
5037 int batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5038 struct mem_cgroup
*memcg
= mc
.to
;
5040 if (mem_cgroup_is_root(memcg
)) {
5041 mc
.precharge
+= count
;
5042 /* we don't need css_get for root */
5045 /* try to charge at once */
5047 struct res_counter
*dummy
;
5049 * "memcg" cannot be under rmdir() because we've already checked
5050 * by cgroup_lock_live_cgroup() that it is not removed and we
5051 * are still under the same cgroup_mutex. So we can postpone
5054 if (res_counter_charge(&memcg
->res
, PAGE_SIZE
* count
, &dummy
))
5056 if (do_swap_account
&& res_counter_charge(&memcg
->memsw
,
5057 PAGE_SIZE
* count
, &dummy
)) {
5058 res_counter_uncharge(&memcg
->res
, PAGE_SIZE
* count
);
5061 mc
.precharge
+= count
;
5065 /* fall back to one by one charge */
5067 if (signal_pending(current
)) {
5071 if (!batch_count
--) {
5072 batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5075 ret
= __mem_cgroup_try_charge(NULL
,
5076 GFP_KERNEL
, 1, &memcg
, false);
5078 /* mem_cgroup_clear_mc() will do uncharge later */
5086 * get_mctgt_type - get target type of moving charge
5087 * @vma: the vma the pte to be checked belongs
5088 * @addr: the address corresponding to the pte to be checked
5089 * @ptent: the pte to be checked
5090 * @target: the pointer the target page or swap ent will be stored(can be NULL)
5093 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5094 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5095 * move charge. if @target is not NULL, the page is stored in target->page
5096 * with extra refcnt got(Callers should handle it).
5097 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5098 * target for charge migration. if @target is not NULL, the entry is stored
5101 * Called with pte lock held.
5108 enum mc_target_type
{
5114 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
5115 unsigned long addr
, pte_t ptent
)
5117 struct page
*page
= vm_normal_page(vma
, addr
, ptent
);
5119 if (!page
|| !page_mapped(page
))
5121 if (PageAnon(page
)) {
5122 /* we don't move shared anon */
5125 } else if (!move_file())
5126 /* we ignore mapcount for file pages */
5128 if (!get_page_unless_zero(page
))
5135 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5136 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5138 struct page
*page
= NULL
;
5139 swp_entry_t ent
= pte_to_swp_entry(ptent
);
5141 if (!move_anon() || non_swap_entry(ent
))
5144 * Because lookup_swap_cache() updates some statistics counter,
5145 * we call find_get_page() with swapper_space directly.
5147 page
= find_get_page(&swapper_space
, ent
.val
);
5148 if (do_swap_account
)
5149 entry
->val
= ent
.val
;
5154 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5155 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5161 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
5162 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5164 struct page
*page
= NULL
;
5165 struct address_space
*mapping
;
5168 if (!vma
->vm_file
) /* anonymous vma */
5173 mapping
= vma
->vm_file
->f_mapping
;
5174 if (pte_none(ptent
))
5175 pgoff
= linear_page_index(vma
, addr
);
5176 else /* pte_file(ptent) is true */
5177 pgoff
= pte_to_pgoff(ptent
);
5179 /* page is moved even if it's not RSS of this task(page-faulted). */
5180 page
= find_get_page(mapping
, pgoff
);
5183 /* shmem/tmpfs may report page out on swap: account for that too. */
5184 if (radix_tree_exceptional_entry(page
)) {
5185 swp_entry_t swap
= radix_to_swp_entry(page
);
5186 if (do_swap_account
)
5188 page
= find_get_page(&swapper_space
, swap
.val
);
5194 static enum mc_target_type
get_mctgt_type(struct vm_area_struct
*vma
,
5195 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
5197 struct page
*page
= NULL
;
5198 struct page_cgroup
*pc
;
5199 enum mc_target_type ret
= MC_TARGET_NONE
;
5200 swp_entry_t ent
= { .val
= 0 };
5202 if (pte_present(ptent
))
5203 page
= mc_handle_present_pte(vma
, addr
, ptent
);
5204 else if (is_swap_pte(ptent
))
5205 page
= mc_handle_swap_pte(vma
, addr
, ptent
, &ent
);
5206 else if (pte_none(ptent
) || pte_file(ptent
))
5207 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
5209 if (!page
&& !ent
.val
)
5212 pc
= lookup_page_cgroup(page
);
5214 * Do only loose check w/o page_cgroup lock.
5215 * mem_cgroup_move_account() checks the pc is valid or not under
5218 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5219 ret
= MC_TARGET_PAGE
;
5221 target
->page
= page
;
5223 if (!ret
|| !target
)
5226 /* There is a swap entry and a page doesn't exist or isn't charged */
5227 if (ent
.val
&& !ret
&&
5228 css_id(&mc
.from
->css
) == lookup_swap_cgroup_id(ent
)) {
5229 ret
= MC_TARGET_SWAP
;
5236 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5238 * We don't consider swapping or file mapped pages because THP does not
5239 * support them for now.
5240 * Caller should make sure that pmd_trans_huge(pmd) is true.
5242 static enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
5243 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
5245 struct page
*page
= NULL
;
5246 struct page_cgroup
*pc
;
5247 enum mc_target_type ret
= MC_TARGET_NONE
;
5249 page
= pmd_page(pmd
);
5250 VM_BUG_ON(!page
|| !PageHead(page
));
5253 pc
= lookup_page_cgroup(page
);
5254 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5255 ret
= MC_TARGET_PAGE
;
5258 target
->page
= page
;
5264 static inline enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
5265 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
5267 return MC_TARGET_NONE
;
5271 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
5272 unsigned long addr
, unsigned long end
,
5273 struct mm_walk
*walk
)
5275 struct vm_area_struct
*vma
= walk
->private;
5279 if (pmd_trans_huge_lock(pmd
, vma
) == 1) {
5280 if (get_mctgt_type_thp(vma
, addr
, *pmd
, NULL
) == MC_TARGET_PAGE
)
5281 mc
.precharge
+= HPAGE_PMD_NR
;
5282 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5286 if (pmd_trans_unstable(pmd
))
5288 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5289 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
5290 if (get_mctgt_type(vma
, addr
, *pte
, NULL
))
5291 mc
.precharge
++; /* increment precharge temporarily */
5292 pte_unmap_unlock(pte
- 1, ptl
);
5298 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
5300 unsigned long precharge
;
5301 struct vm_area_struct
*vma
;
5303 down_read(&mm
->mmap_sem
);
5304 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5305 struct mm_walk mem_cgroup_count_precharge_walk
= {
5306 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
5310 if (is_vm_hugetlb_page(vma
))
5312 walk_page_range(vma
->vm_start
, vma
->vm_end
,
5313 &mem_cgroup_count_precharge_walk
);
5315 up_read(&mm
->mmap_sem
);
5317 precharge
= mc
.precharge
;
5323 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
5325 unsigned long precharge
= mem_cgroup_count_precharge(mm
);
5327 VM_BUG_ON(mc
.moving_task
);
5328 mc
.moving_task
= current
;
5329 return mem_cgroup_do_precharge(precharge
);
5332 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5333 static void __mem_cgroup_clear_mc(void)
5335 struct mem_cgroup
*from
= mc
.from
;
5336 struct mem_cgroup
*to
= mc
.to
;
5338 /* we must uncharge all the leftover precharges from mc.to */
5340 __mem_cgroup_cancel_charge(mc
.to
, mc
.precharge
);
5344 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5345 * we must uncharge here.
5347 if (mc
.moved_charge
) {
5348 __mem_cgroup_cancel_charge(mc
.from
, mc
.moved_charge
);
5349 mc
.moved_charge
= 0;
5351 /* we must fixup refcnts and charges */
5352 if (mc
.moved_swap
) {
5353 /* uncharge swap account from the old cgroup */
5354 if (!mem_cgroup_is_root(mc
.from
))
5355 res_counter_uncharge(&mc
.from
->memsw
,
5356 PAGE_SIZE
* mc
.moved_swap
);
5357 __mem_cgroup_put(mc
.from
, mc
.moved_swap
);
5359 if (!mem_cgroup_is_root(mc
.to
)) {
5361 * we charged both to->res and to->memsw, so we should
5364 res_counter_uncharge(&mc
.to
->res
,
5365 PAGE_SIZE
* mc
.moved_swap
);
5367 /* we've already done mem_cgroup_get(mc.to) */
5370 memcg_oom_recover(from
);
5371 memcg_oom_recover(to
);
5372 wake_up_all(&mc
.waitq
);
5375 static void mem_cgroup_clear_mc(void)
5377 struct mem_cgroup
*from
= mc
.from
;
5380 * we must clear moving_task before waking up waiters at the end of
5383 mc
.moving_task
= NULL
;
5384 __mem_cgroup_clear_mc();
5385 spin_lock(&mc
.lock
);
5388 spin_unlock(&mc
.lock
);
5389 mem_cgroup_end_move(from
);
5392 static int mem_cgroup_can_attach(struct cgroup
*cgroup
,
5393 struct cgroup_taskset
*tset
)
5395 struct task_struct
*p
= cgroup_taskset_first(tset
);
5397 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgroup
);
5399 if (memcg
->move_charge_at_immigrate
) {
5400 struct mm_struct
*mm
;
5401 struct mem_cgroup
*from
= mem_cgroup_from_task(p
);
5403 VM_BUG_ON(from
== memcg
);
5405 mm
= get_task_mm(p
);
5408 /* We move charges only when we move a owner of the mm */
5409 if (mm
->owner
== p
) {
5412 VM_BUG_ON(mc
.precharge
);
5413 VM_BUG_ON(mc
.moved_charge
);
5414 VM_BUG_ON(mc
.moved_swap
);
5415 mem_cgroup_start_move(from
);
5416 spin_lock(&mc
.lock
);
5419 spin_unlock(&mc
.lock
);
5420 /* We set mc.moving_task later */
5422 ret
= mem_cgroup_precharge_mc(mm
);
5424 mem_cgroup_clear_mc();
5431 static void mem_cgroup_cancel_attach(struct cgroup
*cgroup
,
5432 struct cgroup_taskset
*tset
)
5434 mem_cgroup_clear_mc();
5437 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
5438 unsigned long addr
, unsigned long end
,
5439 struct mm_walk
*walk
)
5442 struct vm_area_struct
*vma
= walk
->private;
5445 enum mc_target_type target_type
;
5446 union mc_target target
;
5448 struct page_cgroup
*pc
;
5451 * We don't take compound_lock() here but no race with splitting thp
5453 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
5454 * under splitting, which means there's no concurrent thp split,
5455 * - if another thread runs into split_huge_page() just after we
5456 * entered this if-block, the thread must wait for page table lock
5457 * to be unlocked in __split_huge_page_splitting(), where the main
5458 * part of thp split is not executed yet.
5460 if (pmd_trans_huge_lock(pmd
, vma
) == 1) {
5461 if (mc
.precharge
< HPAGE_PMD_NR
) {
5462 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5465 target_type
= get_mctgt_type_thp(vma
, addr
, *pmd
, &target
);
5466 if (target_type
== MC_TARGET_PAGE
) {
5468 if (!isolate_lru_page(page
)) {
5469 pc
= lookup_page_cgroup(page
);
5470 if (!mem_cgroup_move_account(page
, HPAGE_PMD_NR
,
5471 pc
, mc
.from
, mc
.to
)) {
5472 mc
.precharge
-= HPAGE_PMD_NR
;
5473 mc
.moved_charge
+= HPAGE_PMD_NR
;
5475 putback_lru_page(page
);
5479 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5483 if (pmd_trans_unstable(pmd
))
5486 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5487 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
5488 pte_t ptent
= *(pte
++);
5494 switch (get_mctgt_type(vma
, addr
, ptent
, &target
)) {
5495 case MC_TARGET_PAGE
:
5497 if (isolate_lru_page(page
))
5499 pc
= lookup_page_cgroup(page
);
5500 if (!mem_cgroup_move_account(page
, 1, pc
,
5503 /* we uncharge from mc.from later. */
5506 putback_lru_page(page
);
5507 put
: /* get_mctgt_type() gets the page */
5510 case MC_TARGET_SWAP
:
5512 if (!mem_cgroup_move_swap_account(ent
, mc
.from
, mc
.to
)) {
5514 /* we fixup refcnts and charges later. */
5522 pte_unmap_unlock(pte
- 1, ptl
);
5527 * We have consumed all precharges we got in can_attach().
5528 * We try charge one by one, but don't do any additional
5529 * charges to mc.to if we have failed in charge once in attach()
5532 ret
= mem_cgroup_do_precharge(1);
5540 static void mem_cgroup_move_charge(struct mm_struct
*mm
)
5542 struct vm_area_struct
*vma
;
5544 lru_add_drain_all();
5546 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
5548 * Someone who are holding the mmap_sem might be waiting in
5549 * waitq. So we cancel all extra charges, wake up all waiters,
5550 * and retry. Because we cancel precharges, we might not be able
5551 * to move enough charges, but moving charge is a best-effort
5552 * feature anyway, so it wouldn't be a big problem.
5554 __mem_cgroup_clear_mc();
5558 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5560 struct mm_walk mem_cgroup_move_charge_walk
= {
5561 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
5565 if (is_vm_hugetlb_page(vma
))
5567 ret
= walk_page_range(vma
->vm_start
, vma
->vm_end
,
5568 &mem_cgroup_move_charge_walk
);
5571 * means we have consumed all precharges and failed in
5572 * doing additional charge. Just abandon here.
5576 up_read(&mm
->mmap_sem
);
5579 static void mem_cgroup_move_task(struct cgroup
*cont
,
5580 struct cgroup_taskset
*tset
)
5582 struct task_struct
*p
= cgroup_taskset_first(tset
);
5583 struct mm_struct
*mm
= get_task_mm(p
);
5587 mem_cgroup_move_charge(mm
);
5591 mem_cgroup_clear_mc();
5593 #else /* !CONFIG_MMU */
5594 static int mem_cgroup_can_attach(struct cgroup
*cgroup
,
5595 struct cgroup_taskset
*tset
)
5599 static void mem_cgroup_cancel_attach(struct cgroup
*cgroup
,
5600 struct cgroup_taskset
*tset
)
5603 static void mem_cgroup_move_task(struct cgroup
*cont
,
5604 struct cgroup_taskset
*tset
)
5609 struct cgroup_subsys mem_cgroup_subsys
= {
5611 .subsys_id
= mem_cgroup_subsys_id
,
5612 .create
= mem_cgroup_create
,
5613 .pre_destroy
= mem_cgroup_pre_destroy
,
5614 .destroy
= mem_cgroup_destroy
,
5615 .can_attach
= mem_cgroup_can_attach
,
5616 .cancel_attach
= mem_cgroup_cancel_attach
,
5617 .attach
= mem_cgroup_move_task
,
5618 .base_cftypes
= mem_cgroup_files
,
5621 .__DEPRECATED_clear_css_refs
= true,
5624 #ifdef CONFIG_MEMCG_SWAP
5625 static int __init
enable_swap_account(char *s
)
5627 /* consider enabled if no parameter or 1 is given */
5628 if (!strcmp(s
, "1"))
5629 really_do_swap_account
= 1;
5630 else if (!strcmp(s
, "0"))
5631 really_do_swap_account
= 0;
5634 __setup("swapaccount=", enable_swap_account
);