1 /* memcontrol.c - Memory Controller
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
6 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
13 * This program is free software; you can redistribute it and/or modify
14 * it under the terms of the GNU General Public License as published by
15 * the Free Software Foundation; either version 2 of the License, or
16 * (at your option) any later version.
18 * This program is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 * GNU General Public License for more details.
24 #include <linux/res_counter.h>
25 #include <linux/memcontrol.h>
26 #include <linux/cgroup.h>
28 #include <linux/hugetlb.h>
29 #include <linux/pagemap.h>
30 #include <linux/smp.h>
31 #include <linux/page-flags.h>
32 #include <linux/backing-dev.h>
33 #include <linux/bit_spinlock.h>
34 #include <linux/rcupdate.h>
35 #include <linux/limits.h>
36 #include <linux/export.h>
37 #include <linux/mutex.h>
38 #include <linux/rbtree.h>
39 #include <linux/slab.h>
40 #include <linux/swap.h>
41 #include <linux/swapops.h>
42 #include <linux/spinlock.h>
43 #include <linux/eventfd.h>
44 #include <linux/sort.h>
46 #include <linux/seq_file.h>
47 #include <linux/vmalloc.h>
48 #include <linux/mm_inline.h>
49 #include <linux/page_cgroup.h>
50 #include <linux/cpu.h>
51 #include <linux/oom.h>
55 #include <net/tcp_memcontrol.h>
57 #include <asm/uaccess.h>
59 #include <trace/events/vmscan.h>
61 struct cgroup_subsys mem_cgroup_subsys __read_mostly
;
62 #define MEM_CGROUP_RECLAIM_RETRIES 5
63 static struct mem_cgroup
*root_mem_cgroup __read_mostly
;
65 #ifdef CONFIG_MEMCG_SWAP
66 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
67 int do_swap_account __read_mostly
;
69 /* for remember boot option*/
70 #ifdef CONFIG_MEMCG_SWAP_ENABLED
71 static int really_do_swap_account __initdata
= 1;
73 static int really_do_swap_account __initdata
= 0;
77 #define do_swap_account 0
82 * Statistics for memory cgroup.
84 enum mem_cgroup_stat_index
{
86 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
88 MEM_CGROUP_STAT_CACHE
, /* # of pages charged as cache */
89 MEM_CGROUP_STAT_RSS
, /* # of pages charged as anon rss */
90 MEM_CGROUP_STAT_FILE_MAPPED
, /* # of pages charged as file rss */
91 MEM_CGROUP_STAT_SWAP
, /* # of pages, swapped out */
92 MEM_CGROUP_STAT_NSTATS
,
95 static const char * const mem_cgroup_stat_names
[] = {
102 enum mem_cgroup_events_index
{
103 MEM_CGROUP_EVENTS_PGPGIN
, /* # of pages paged in */
104 MEM_CGROUP_EVENTS_PGPGOUT
, /* # of pages paged out */
105 MEM_CGROUP_EVENTS_PGFAULT
, /* # of page-faults */
106 MEM_CGROUP_EVENTS_PGMAJFAULT
, /* # of major page-faults */
107 MEM_CGROUP_EVENTS_NSTATS
,
110 static const char * const mem_cgroup_events_names
[] = {
118 * Per memcg event counter is incremented at every pagein/pageout. With THP,
119 * it will be incremated by the number of pages. This counter is used for
120 * for trigger some periodic events. This is straightforward and better
121 * than using jiffies etc. to handle periodic memcg event.
123 enum mem_cgroup_events_target
{
124 MEM_CGROUP_TARGET_THRESH
,
125 MEM_CGROUP_TARGET_SOFTLIMIT
,
126 MEM_CGROUP_TARGET_NUMAINFO
,
129 #define THRESHOLDS_EVENTS_TARGET 128
130 #define SOFTLIMIT_EVENTS_TARGET 1024
131 #define NUMAINFO_EVENTS_TARGET 1024
133 struct mem_cgroup_stat_cpu
{
134 long count
[MEM_CGROUP_STAT_NSTATS
];
135 unsigned long events
[MEM_CGROUP_EVENTS_NSTATS
];
136 unsigned long nr_page_events
;
137 unsigned long targets
[MEM_CGROUP_NTARGETS
];
140 struct mem_cgroup_reclaim_iter
{
141 /* css_id of the last scanned hierarchy member */
143 /* scan generation, increased every round-trip */
144 unsigned int generation
;
148 * per-zone information in memory controller.
150 struct mem_cgroup_per_zone
{
151 struct lruvec lruvec
;
152 unsigned long lru_size
[NR_LRU_LISTS
];
154 struct mem_cgroup_reclaim_iter reclaim_iter
[DEF_PRIORITY
+ 1];
156 struct rb_node tree_node
; /* RB tree node */
157 unsigned long long usage_in_excess
;/* Set to the value by which */
158 /* the soft limit is exceeded*/
160 struct mem_cgroup
*memcg
; /* Back pointer, we cannot */
161 /* use container_of */
164 struct mem_cgroup_per_node
{
165 struct mem_cgroup_per_zone zoneinfo
[MAX_NR_ZONES
];
168 struct mem_cgroup_lru_info
{
169 struct mem_cgroup_per_node
*nodeinfo
[MAX_NUMNODES
];
173 * Cgroups above their limits are maintained in a RB-Tree, independent of
174 * their hierarchy representation
177 struct mem_cgroup_tree_per_zone
{
178 struct rb_root rb_root
;
182 struct mem_cgroup_tree_per_node
{
183 struct mem_cgroup_tree_per_zone rb_tree_per_zone
[MAX_NR_ZONES
];
186 struct mem_cgroup_tree
{
187 struct mem_cgroup_tree_per_node
*rb_tree_per_node
[MAX_NUMNODES
];
190 static struct mem_cgroup_tree soft_limit_tree __read_mostly
;
192 struct mem_cgroup_threshold
{
193 struct eventfd_ctx
*eventfd
;
198 struct mem_cgroup_threshold_ary
{
199 /* An array index points to threshold just below or equal to usage. */
200 int current_threshold
;
201 /* Size of entries[] */
203 /* Array of thresholds */
204 struct mem_cgroup_threshold entries
[0];
207 struct mem_cgroup_thresholds
{
208 /* Primary thresholds array */
209 struct mem_cgroup_threshold_ary
*primary
;
211 * Spare threshold array.
212 * This is needed to make mem_cgroup_unregister_event() "never fail".
213 * It must be able to store at least primary->size - 1 entries.
215 struct mem_cgroup_threshold_ary
*spare
;
219 struct mem_cgroup_eventfd_list
{
220 struct list_head list
;
221 struct eventfd_ctx
*eventfd
;
224 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
);
225 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
);
228 * The memory controller data structure. The memory controller controls both
229 * page cache and RSS per cgroup. We would eventually like to provide
230 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
231 * to help the administrator determine what knobs to tune.
233 * TODO: Add a water mark for the memory controller. Reclaim will begin when
234 * we hit the water mark. May be even add a low water mark, such that
235 * no reclaim occurs from a cgroup at it's low water mark, this is
236 * a feature that will be implemented much later in the future.
239 struct cgroup_subsys_state css
;
241 * the counter to account for memory usage
243 struct res_counter res
;
247 * the counter to account for mem+swap usage.
249 struct res_counter memsw
;
252 * rcu_freeing is used only when freeing struct mem_cgroup,
253 * so put it into a union to avoid wasting more memory.
254 * It must be disjoint from the css field. It could be
255 * in a union with the res field, but res plays a much
256 * larger part in mem_cgroup life than memsw, and might
257 * be of interest, even at time of free, when debugging.
258 * So share rcu_head with the less interesting memsw.
260 struct rcu_head rcu_freeing
;
262 * We also need some space for a worker in deferred freeing.
263 * By the time we call it, rcu_freeing is no longer in use.
265 struct work_struct work_freeing
;
269 * Per cgroup active and inactive list, similar to the
270 * per zone LRU lists.
272 struct mem_cgroup_lru_info info
;
273 int last_scanned_node
;
275 nodemask_t scan_nodes
;
276 atomic_t numainfo_events
;
277 atomic_t numainfo_updating
;
280 * Should the accounting and control be hierarchical, per subtree?
290 /* OOM-Killer disable */
291 int oom_kill_disable
;
293 /* set when res.limit == memsw.limit */
294 bool memsw_is_minimum
;
296 /* protect arrays of thresholds */
297 struct mutex thresholds_lock
;
299 /* thresholds for memory usage. RCU-protected */
300 struct mem_cgroup_thresholds thresholds
;
302 /* thresholds for mem+swap usage. RCU-protected */
303 struct mem_cgroup_thresholds memsw_thresholds
;
305 /* For oom notifier event fd */
306 struct list_head oom_notify
;
309 * Should we move charges of a task when a task is moved into this
310 * mem_cgroup ? And what type of charges should we move ?
312 unsigned long move_charge_at_immigrate
;
314 * set > 0 if pages under this cgroup are moving to other cgroup.
316 atomic_t moving_account
;
317 /* taken only while moving_account > 0 */
318 spinlock_t move_lock
;
322 struct mem_cgroup_stat_cpu __percpu
*stat
;
324 * used when a cpu is offlined or other synchronizations
325 * See mem_cgroup_read_stat().
327 struct mem_cgroup_stat_cpu nocpu_base
;
328 spinlock_t pcp_counter_lock
;
330 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET)
331 struct tcp_memcontrol tcp_mem
;
335 /* Stuffs for move charges at task migration. */
337 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
338 * left-shifted bitmap of these types.
341 MOVE_CHARGE_TYPE_ANON
, /* private anonymous page and swap of it */
342 MOVE_CHARGE_TYPE_FILE
, /* file page(including tmpfs) and swap of it */
346 /* "mc" and its members are protected by cgroup_mutex */
347 static struct move_charge_struct
{
348 spinlock_t lock
; /* for from, to */
349 struct mem_cgroup
*from
;
350 struct mem_cgroup
*to
;
351 unsigned long precharge
;
352 unsigned long moved_charge
;
353 unsigned long moved_swap
;
354 struct task_struct
*moving_task
; /* a task moving charges */
355 wait_queue_head_t waitq
; /* a waitq for other context */
357 .lock
= __SPIN_LOCK_UNLOCKED(mc
.lock
),
358 .waitq
= __WAIT_QUEUE_HEAD_INITIALIZER(mc
.waitq
),
361 static bool move_anon(void)
363 return test_bit(MOVE_CHARGE_TYPE_ANON
,
364 &mc
.to
->move_charge_at_immigrate
);
367 static bool move_file(void)
369 return test_bit(MOVE_CHARGE_TYPE_FILE
,
370 &mc
.to
->move_charge_at_immigrate
);
374 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
375 * limit reclaim to prevent infinite loops, if they ever occur.
377 #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
378 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
381 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
382 MEM_CGROUP_CHARGE_TYPE_ANON
,
383 MEM_CGROUP_CHARGE_TYPE_SWAPOUT
, /* for accounting swapcache */
384 MEM_CGROUP_CHARGE_TYPE_DROP
, /* a page was unused swap cache */
388 /* for encoding cft->private value on file */
391 #define _OOM_TYPE (2)
392 #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
393 #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
394 #define MEMFILE_ATTR(val) ((val) & 0xffff)
395 /* Used for OOM nofiier */
396 #define OOM_CONTROL (0)
399 * Reclaim flags for mem_cgroup_hierarchical_reclaim
401 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
402 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
403 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
404 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
406 static void mem_cgroup_get(struct mem_cgroup
*memcg
);
407 static void mem_cgroup_put(struct mem_cgroup
*memcg
);
410 struct mem_cgroup
*mem_cgroup_from_css(struct cgroup_subsys_state
*s
)
412 return container_of(s
, struct mem_cgroup
, css
);
415 static inline bool mem_cgroup_is_root(struct mem_cgroup
*memcg
)
417 return (memcg
== root_mem_cgroup
);
420 /* Writing them here to avoid exposing memcg's inner layout */
421 #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
423 void sock_update_memcg(struct sock
*sk
)
425 if (mem_cgroup_sockets_enabled
) {
426 struct mem_cgroup
*memcg
;
427 struct cg_proto
*cg_proto
;
429 BUG_ON(!sk
->sk_prot
->proto_cgroup
);
431 /* Socket cloning can throw us here with sk_cgrp already
432 * filled. It won't however, necessarily happen from
433 * process context. So the test for root memcg given
434 * the current task's memcg won't help us in this case.
436 * Respecting the original socket's memcg is a better
437 * decision in this case.
440 BUG_ON(mem_cgroup_is_root(sk
->sk_cgrp
->memcg
));
441 mem_cgroup_get(sk
->sk_cgrp
->memcg
);
446 memcg
= mem_cgroup_from_task(current
);
447 cg_proto
= sk
->sk_prot
->proto_cgroup(memcg
);
448 if (!mem_cgroup_is_root(memcg
) && memcg_proto_active(cg_proto
)) {
449 mem_cgroup_get(memcg
);
450 sk
->sk_cgrp
= cg_proto
;
455 EXPORT_SYMBOL(sock_update_memcg
);
457 void sock_release_memcg(struct sock
*sk
)
459 if (mem_cgroup_sockets_enabled
&& sk
->sk_cgrp
) {
460 struct mem_cgroup
*memcg
;
461 WARN_ON(!sk
->sk_cgrp
->memcg
);
462 memcg
= sk
->sk_cgrp
->memcg
;
463 mem_cgroup_put(memcg
);
467 struct cg_proto
*tcp_proto_cgroup(struct mem_cgroup
*memcg
)
469 if (!memcg
|| mem_cgroup_is_root(memcg
))
472 return &memcg
->tcp_mem
.cg_proto
;
474 EXPORT_SYMBOL(tcp_proto_cgroup
);
476 static void disarm_sock_keys(struct mem_cgroup
*memcg
)
478 if (!memcg_proto_activated(&memcg
->tcp_mem
.cg_proto
))
480 static_key_slow_dec(&memcg_socket_limit_enabled
);
483 static void disarm_sock_keys(struct mem_cgroup
*memcg
)
488 static void drain_all_stock_async(struct mem_cgroup
*memcg
);
490 static struct mem_cgroup_per_zone
*
491 mem_cgroup_zoneinfo(struct mem_cgroup
*memcg
, int nid
, int zid
)
493 return &memcg
->info
.nodeinfo
[nid
]->zoneinfo
[zid
];
496 struct cgroup_subsys_state
*mem_cgroup_css(struct mem_cgroup
*memcg
)
501 static struct mem_cgroup_per_zone
*
502 page_cgroup_zoneinfo(struct mem_cgroup
*memcg
, struct page
*page
)
504 int nid
= page_to_nid(page
);
505 int zid
= page_zonenum(page
);
507 return mem_cgroup_zoneinfo(memcg
, nid
, zid
);
510 static struct mem_cgroup_tree_per_zone
*
511 soft_limit_tree_node_zone(int nid
, int zid
)
513 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
516 static struct mem_cgroup_tree_per_zone
*
517 soft_limit_tree_from_page(struct page
*page
)
519 int nid
= page_to_nid(page
);
520 int zid
= page_zonenum(page
);
522 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
526 __mem_cgroup_insert_exceeded(struct mem_cgroup
*memcg
,
527 struct mem_cgroup_per_zone
*mz
,
528 struct mem_cgroup_tree_per_zone
*mctz
,
529 unsigned long long new_usage_in_excess
)
531 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
532 struct rb_node
*parent
= NULL
;
533 struct mem_cgroup_per_zone
*mz_node
;
538 mz
->usage_in_excess
= new_usage_in_excess
;
539 if (!mz
->usage_in_excess
)
543 mz_node
= rb_entry(parent
, struct mem_cgroup_per_zone
,
545 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
)
548 * We can't avoid mem cgroups that are over their soft
549 * limit by the same amount
551 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
554 rb_link_node(&mz
->tree_node
, parent
, p
);
555 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
560 __mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
561 struct mem_cgroup_per_zone
*mz
,
562 struct mem_cgroup_tree_per_zone
*mctz
)
566 rb_erase(&mz
->tree_node
, &mctz
->rb_root
);
571 mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
572 struct mem_cgroup_per_zone
*mz
,
573 struct mem_cgroup_tree_per_zone
*mctz
)
575 spin_lock(&mctz
->lock
);
576 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
577 spin_unlock(&mctz
->lock
);
581 static void mem_cgroup_update_tree(struct mem_cgroup
*memcg
, struct page
*page
)
583 unsigned long long excess
;
584 struct mem_cgroup_per_zone
*mz
;
585 struct mem_cgroup_tree_per_zone
*mctz
;
586 int nid
= page_to_nid(page
);
587 int zid
= page_zonenum(page
);
588 mctz
= soft_limit_tree_from_page(page
);
591 * Necessary to update all ancestors when hierarchy is used.
592 * because their event counter is not touched.
594 for (; memcg
; memcg
= parent_mem_cgroup(memcg
)) {
595 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
596 excess
= res_counter_soft_limit_excess(&memcg
->res
);
598 * We have to update the tree if mz is on RB-tree or
599 * mem is over its softlimit.
601 if (excess
|| mz
->on_tree
) {
602 spin_lock(&mctz
->lock
);
603 /* if on-tree, remove it */
605 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
607 * Insert again. mz->usage_in_excess will be updated.
608 * If excess is 0, no tree ops.
610 __mem_cgroup_insert_exceeded(memcg
, mz
, mctz
, excess
);
611 spin_unlock(&mctz
->lock
);
616 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*memcg
)
619 struct mem_cgroup_per_zone
*mz
;
620 struct mem_cgroup_tree_per_zone
*mctz
;
622 for_each_node(node
) {
623 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
624 mz
= mem_cgroup_zoneinfo(memcg
, node
, zone
);
625 mctz
= soft_limit_tree_node_zone(node
, zone
);
626 mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
631 static struct mem_cgroup_per_zone
*
632 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
634 struct rb_node
*rightmost
= NULL
;
635 struct mem_cgroup_per_zone
*mz
;
639 rightmost
= rb_last(&mctz
->rb_root
);
641 goto done
; /* Nothing to reclaim from */
643 mz
= rb_entry(rightmost
, struct mem_cgroup_per_zone
, tree_node
);
645 * Remove the node now but someone else can add it back,
646 * we will to add it back at the end of reclaim to its correct
647 * position in the tree.
649 __mem_cgroup_remove_exceeded(mz
->memcg
, mz
, mctz
);
650 if (!res_counter_soft_limit_excess(&mz
->memcg
->res
) ||
651 !css_tryget(&mz
->memcg
->css
))
657 static struct mem_cgroup_per_zone
*
658 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
660 struct mem_cgroup_per_zone
*mz
;
662 spin_lock(&mctz
->lock
);
663 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
664 spin_unlock(&mctz
->lock
);
669 * Implementation Note: reading percpu statistics for memcg.
671 * Both of vmstat[] and percpu_counter has threshold and do periodic
672 * synchronization to implement "quick" read. There are trade-off between
673 * reading cost and precision of value. Then, we may have a chance to implement
674 * a periodic synchronizion of counter in memcg's counter.
676 * But this _read() function is used for user interface now. The user accounts
677 * memory usage by memory cgroup and he _always_ requires exact value because
678 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
679 * have to visit all online cpus and make sum. So, for now, unnecessary
680 * synchronization is not implemented. (just implemented for cpu hotplug)
682 * If there are kernel internal actions which can make use of some not-exact
683 * value, and reading all cpu value can be performance bottleneck in some
684 * common workload, threashold and synchonization as vmstat[] should be
687 static long mem_cgroup_read_stat(struct mem_cgroup
*memcg
,
688 enum mem_cgroup_stat_index idx
)
694 for_each_online_cpu(cpu
)
695 val
+= per_cpu(memcg
->stat
->count
[idx
], cpu
);
696 #ifdef CONFIG_HOTPLUG_CPU
697 spin_lock(&memcg
->pcp_counter_lock
);
698 val
+= memcg
->nocpu_base
.count
[idx
];
699 spin_unlock(&memcg
->pcp_counter_lock
);
705 static void mem_cgroup_swap_statistics(struct mem_cgroup
*memcg
,
708 int val
= (charge
) ? 1 : -1;
709 this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_SWAP
], val
);
712 static unsigned long mem_cgroup_read_events(struct mem_cgroup
*memcg
,
713 enum mem_cgroup_events_index idx
)
715 unsigned long val
= 0;
718 for_each_online_cpu(cpu
)
719 val
+= per_cpu(memcg
->stat
->events
[idx
], cpu
);
720 #ifdef CONFIG_HOTPLUG_CPU
721 spin_lock(&memcg
->pcp_counter_lock
);
722 val
+= memcg
->nocpu_base
.events
[idx
];
723 spin_unlock(&memcg
->pcp_counter_lock
);
728 static void mem_cgroup_charge_statistics(struct mem_cgroup
*memcg
,
729 bool anon
, int nr_pages
)
734 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
735 * counted as CACHE even if it's on ANON LRU.
738 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_RSS
],
741 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_CACHE
],
744 /* pagein of a big page is an event. So, ignore page size */
746 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGIN
]);
748 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGOUT
]);
749 nr_pages
= -nr_pages
; /* for event */
752 __this_cpu_add(memcg
->stat
->nr_page_events
, nr_pages
);
758 mem_cgroup_get_lru_size(struct lruvec
*lruvec
, enum lru_list lru
)
760 struct mem_cgroup_per_zone
*mz
;
762 mz
= container_of(lruvec
, struct mem_cgroup_per_zone
, lruvec
);
763 return mz
->lru_size
[lru
];
767 mem_cgroup_zone_nr_lru_pages(struct mem_cgroup
*memcg
, int nid
, int zid
,
768 unsigned int lru_mask
)
770 struct mem_cgroup_per_zone
*mz
;
772 unsigned long ret
= 0;
774 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
777 if (BIT(lru
) & lru_mask
)
778 ret
+= mz
->lru_size
[lru
];
784 mem_cgroup_node_nr_lru_pages(struct mem_cgroup
*memcg
,
785 int nid
, unsigned int lru_mask
)
790 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++)
791 total
+= mem_cgroup_zone_nr_lru_pages(memcg
,
797 static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup
*memcg
,
798 unsigned int lru_mask
)
803 for_each_node_state(nid
, N_HIGH_MEMORY
)
804 total
+= mem_cgroup_node_nr_lru_pages(memcg
, nid
, lru_mask
);
808 static bool mem_cgroup_event_ratelimit(struct mem_cgroup
*memcg
,
809 enum mem_cgroup_events_target target
)
811 unsigned long val
, next
;
813 val
= __this_cpu_read(memcg
->stat
->nr_page_events
);
814 next
= __this_cpu_read(memcg
->stat
->targets
[target
]);
815 /* from time_after() in jiffies.h */
816 if ((long)next
- (long)val
< 0) {
818 case MEM_CGROUP_TARGET_THRESH
:
819 next
= val
+ THRESHOLDS_EVENTS_TARGET
;
821 case MEM_CGROUP_TARGET_SOFTLIMIT
:
822 next
= val
+ SOFTLIMIT_EVENTS_TARGET
;
824 case MEM_CGROUP_TARGET_NUMAINFO
:
825 next
= val
+ NUMAINFO_EVENTS_TARGET
;
830 __this_cpu_write(memcg
->stat
->targets
[target
], next
);
837 * Check events in order.
840 static void memcg_check_events(struct mem_cgroup
*memcg
, struct page
*page
)
843 /* threshold event is triggered in finer grain than soft limit */
844 if (unlikely(mem_cgroup_event_ratelimit(memcg
,
845 MEM_CGROUP_TARGET_THRESH
))) {
847 bool do_numainfo __maybe_unused
;
849 do_softlimit
= mem_cgroup_event_ratelimit(memcg
,
850 MEM_CGROUP_TARGET_SOFTLIMIT
);
852 do_numainfo
= mem_cgroup_event_ratelimit(memcg
,
853 MEM_CGROUP_TARGET_NUMAINFO
);
857 mem_cgroup_threshold(memcg
);
858 if (unlikely(do_softlimit
))
859 mem_cgroup_update_tree(memcg
, page
);
861 if (unlikely(do_numainfo
))
862 atomic_inc(&memcg
->numainfo_events
);
868 struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
870 return mem_cgroup_from_css(
871 cgroup_subsys_state(cont
, mem_cgroup_subsys_id
));
874 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
877 * mm_update_next_owner() may clear mm->owner to NULL
878 * if it races with swapoff, page migration, etc.
879 * So this can be called with p == NULL.
884 return mem_cgroup_from_css(task_subsys_state(p
, mem_cgroup_subsys_id
));
887 struct mem_cgroup
*try_get_mem_cgroup_from_mm(struct mm_struct
*mm
)
889 struct mem_cgroup
*memcg
= NULL
;
894 * Because we have no locks, mm->owner's may be being moved to other
895 * cgroup. We use css_tryget() here even if this looks
896 * pessimistic (rather than adding locks here).
900 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
901 if (unlikely(!memcg
))
903 } while (!css_tryget(&memcg
->css
));
909 * mem_cgroup_iter - iterate over memory cgroup hierarchy
910 * @root: hierarchy root
911 * @prev: previously returned memcg, NULL on first invocation
912 * @reclaim: cookie for shared reclaim walks, NULL for full walks
914 * Returns references to children of the hierarchy below @root, or
915 * @root itself, or %NULL after a full round-trip.
917 * Caller must pass the return value in @prev on subsequent
918 * invocations for reference counting, or use mem_cgroup_iter_break()
919 * to cancel a hierarchy walk before the round-trip is complete.
921 * Reclaimers can specify a zone and a priority level in @reclaim to
922 * divide up the memcgs in the hierarchy among all concurrent
923 * reclaimers operating on the same zone and priority.
925 struct mem_cgroup
*mem_cgroup_iter(struct mem_cgroup
*root
,
926 struct mem_cgroup
*prev
,
927 struct mem_cgroup_reclaim_cookie
*reclaim
)
929 struct mem_cgroup
*memcg
= NULL
;
932 if (mem_cgroup_disabled())
936 root
= root_mem_cgroup
;
938 if (prev
&& !reclaim
)
939 id
= css_id(&prev
->css
);
941 if (prev
&& prev
!= root
)
944 if (!root
->use_hierarchy
&& root
!= root_mem_cgroup
) {
951 struct mem_cgroup_reclaim_iter
*uninitialized_var(iter
);
952 struct cgroup_subsys_state
*css
;
955 int nid
= zone_to_nid(reclaim
->zone
);
956 int zid
= zone_idx(reclaim
->zone
);
957 struct mem_cgroup_per_zone
*mz
;
959 mz
= mem_cgroup_zoneinfo(root
, nid
, zid
);
960 iter
= &mz
->reclaim_iter
[reclaim
->priority
];
961 if (prev
&& reclaim
->generation
!= iter
->generation
)
967 css
= css_get_next(&mem_cgroup_subsys
, id
+ 1, &root
->css
, &id
);
969 if (css
== &root
->css
|| css_tryget(css
))
970 memcg
= mem_cgroup_from_css(css
);
979 else if (!prev
&& memcg
)
980 reclaim
->generation
= iter
->generation
;
990 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
991 * @root: hierarchy root
992 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
994 void mem_cgroup_iter_break(struct mem_cgroup
*root
,
995 struct mem_cgroup
*prev
)
998 root
= root_mem_cgroup
;
999 if (prev
&& prev
!= root
)
1000 css_put(&prev
->css
);
1004 * Iteration constructs for visiting all cgroups (under a tree). If
1005 * loops are exited prematurely (break), mem_cgroup_iter_break() must
1006 * be used for reference counting.
1008 #define for_each_mem_cgroup_tree(iter, root) \
1009 for (iter = mem_cgroup_iter(root, NULL, NULL); \
1011 iter = mem_cgroup_iter(root, iter, NULL))
1013 #define for_each_mem_cgroup(iter) \
1014 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
1016 iter = mem_cgroup_iter(NULL, iter, NULL))
1018 void mem_cgroup_count_vm_event(struct mm_struct
*mm
, enum vm_event_item idx
)
1020 struct mem_cgroup
*memcg
;
1026 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
1027 if (unlikely(!memcg
))
1032 this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGFAULT
]);
1035 this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGMAJFAULT
]);
1043 EXPORT_SYMBOL(mem_cgroup_count_vm_event
);
1046 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1047 * @zone: zone of the wanted lruvec
1048 * @memcg: memcg of the wanted lruvec
1050 * Returns the lru list vector holding pages for the given @zone and
1051 * @mem. This can be the global zone lruvec, if the memory controller
1054 struct lruvec
*mem_cgroup_zone_lruvec(struct zone
*zone
,
1055 struct mem_cgroup
*memcg
)
1057 struct mem_cgroup_per_zone
*mz
;
1059 if (mem_cgroup_disabled())
1060 return &zone
->lruvec
;
1062 mz
= mem_cgroup_zoneinfo(memcg
, zone_to_nid(zone
), zone_idx(zone
));
1067 * Following LRU functions are allowed to be used without PCG_LOCK.
1068 * Operations are called by routine of global LRU independently from memcg.
1069 * What we have to take care of here is validness of pc->mem_cgroup.
1071 * Changes to pc->mem_cgroup happens when
1074 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1075 * It is added to LRU before charge.
1076 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1077 * When moving account, the page is not on LRU. It's isolated.
1081 * mem_cgroup_page_lruvec - return lruvec for adding an lru page
1083 * @zone: zone of the page
1085 struct lruvec
*mem_cgroup_page_lruvec(struct page
*page
, struct zone
*zone
)
1087 struct mem_cgroup_per_zone
*mz
;
1088 struct mem_cgroup
*memcg
;
1089 struct page_cgroup
*pc
;
1091 if (mem_cgroup_disabled())
1092 return &zone
->lruvec
;
1094 pc
= lookup_page_cgroup(page
);
1095 memcg
= pc
->mem_cgroup
;
1098 * Surreptitiously switch any uncharged offlist page to root:
1099 * an uncharged page off lru does nothing to secure
1100 * its former mem_cgroup from sudden removal.
1102 * Our caller holds lru_lock, and PageCgroupUsed is updated
1103 * under page_cgroup lock: between them, they make all uses
1104 * of pc->mem_cgroup safe.
1106 if (!PageLRU(page
) && !PageCgroupUsed(pc
) && memcg
!= root_mem_cgroup
)
1107 pc
->mem_cgroup
= memcg
= root_mem_cgroup
;
1109 mz
= page_cgroup_zoneinfo(memcg
, page
);
1114 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1115 * @lruvec: mem_cgroup per zone lru vector
1116 * @lru: index of lru list the page is sitting on
1117 * @nr_pages: positive when adding or negative when removing
1119 * This function must be called when a page is added to or removed from an
1122 void mem_cgroup_update_lru_size(struct lruvec
*lruvec
, enum lru_list lru
,
1125 struct mem_cgroup_per_zone
*mz
;
1126 unsigned long *lru_size
;
1128 if (mem_cgroup_disabled())
1131 mz
= container_of(lruvec
, struct mem_cgroup_per_zone
, lruvec
);
1132 lru_size
= mz
->lru_size
+ lru
;
1133 *lru_size
+= nr_pages
;
1134 VM_BUG_ON((long)(*lru_size
) < 0);
1138 * Checks whether given mem is same or in the root_mem_cgroup's
1141 bool __mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_memcg
,
1142 struct mem_cgroup
*memcg
)
1144 if (root_memcg
== memcg
)
1146 if (!root_memcg
->use_hierarchy
|| !memcg
)
1148 return css_is_ancestor(&memcg
->css
, &root_memcg
->css
);
1151 static bool mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_memcg
,
1152 struct mem_cgroup
*memcg
)
1157 ret
= __mem_cgroup_same_or_subtree(root_memcg
, memcg
);
1162 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*memcg
)
1165 struct mem_cgroup
*curr
= NULL
;
1166 struct task_struct
*p
;
1168 p
= find_lock_task_mm(task
);
1170 curr
= try_get_mem_cgroup_from_mm(p
->mm
);
1174 * All threads may have already detached their mm's, but the oom
1175 * killer still needs to detect if they have already been oom
1176 * killed to prevent needlessly killing additional tasks.
1179 curr
= mem_cgroup_from_task(task
);
1181 css_get(&curr
->css
);
1187 * We should check use_hierarchy of "memcg" not "curr". Because checking
1188 * use_hierarchy of "curr" here make this function true if hierarchy is
1189 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1190 * hierarchy(even if use_hierarchy is disabled in "memcg").
1192 ret
= mem_cgroup_same_or_subtree(memcg
, curr
);
1193 css_put(&curr
->css
);
1197 int mem_cgroup_inactive_anon_is_low(struct lruvec
*lruvec
)
1199 unsigned long inactive_ratio
;
1200 unsigned long inactive
;
1201 unsigned long active
;
1204 inactive
= mem_cgroup_get_lru_size(lruvec
, LRU_INACTIVE_ANON
);
1205 active
= mem_cgroup_get_lru_size(lruvec
, LRU_ACTIVE_ANON
);
1207 gb
= (inactive
+ active
) >> (30 - PAGE_SHIFT
);
1209 inactive_ratio
= int_sqrt(10 * gb
);
1213 return inactive
* inactive_ratio
< active
;
1216 int mem_cgroup_inactive_file_is_low(struct lruvec
*lruvec
)
1218 unsigned long active
;
1219 unsigned long inactive
;
1221 inactive
= mem_cgroup_get_lru_size(lruvec
, LRU_INACTIVE_FILE
);
1222 active
= mem_cgroup_get_lru_size(lruvec
, LRU_ACTIVE_FILE
);
1224 return (active
> inactive
);
1227 #define mem_cgroup_from_res_counter(counter, member) \
1228 container_of(counter, struct mem_cgroup, member)
1231 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1232 * @memcg: the memory cgroup
1234 * Returns the maximum amount of memory @mem can be charged with, in
1237 static unsigned long mem_cgroup_margin(struct mem_cgroup
*memcg
)
1239 unsigned long long margin
;
1241 margin
= res_counter_margin(&memcg
->res
);
1242 if (do_swap_account
)
1243 margin
= min(margin
, res_counter_margin(&memcg
->memsw
));
1244 return margin
>> PAGE_SHIFT
;
1247 int mem_cgroup_swappiness(struct mem_cgroup
*memcg
)
1249 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
1252 if (cgrp
->parent
== NULL
)
1253 return vm_swappiness
;
1255 return memcg
->swappiness
;
1259 * memcg->moving_account is used for checking possibility that some thread is
1260 * calling move_account(). When a thread on CPU-A starts moving pages under
1261 * a memcg, other threads should check memcg->moving_account under
1262 * rcu_read_lock(), like this:
1266 * memcg->moving_account+1 if (memcg->mocing_account)
1268 * synchronize_rcu() update something.
1273 /* for quick checking without looking up memcg */
1274 atomic_t memcg_moving __read_mostly
;
1276 static void mem_cgroup_start_move(struct mem_cgroup
*memcg
)
1278 atomic_inc(&memcg_moving
);
1279 atomic_inc(&memcg
->moving_account
);
1283 static void mem_cgroup_end_move(struct mem_cgroup
*memcg
)
1286 * Now, mem_cgroup_clear_mc() may call this function with NULL.
1287 * We check NULL in callee rather than caller.
1290 atomic_dec(&memcg_moving
);
1291 atomic_dec(&memcg
->moving_account
);
1296 * 2 routines for checking "mem" is under move_account() or not.
1298 * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This
1299 * is used for avoiding races in accounting. If true,
1300 * pc->mem_cgroup may be overwritten.
1302 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1303 * under hierarchy of moving cgroups. This is for
1304 * waiting at hith-memory prressure caused by "move".
1307 static bool mem_cgroup_stolen(struct mem_cgroup
*memcg
)
1309 VM_BUG_ON(!rcu_read_lock_held());
1310 return atomic_read(&memcg
->moving_account
) > 0;
1313 static bool mem_cgroup_under_move(struct mem_cgroup
*memcg
)
1315 struct mem_cgroup
*from
;
1316 struct mem_cgroup
*to
;
1319 * Unlike task_move routines, we access mc.to, mc.from not under
1320 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1322 spin_lock(&mc
.lock
);
1328 ret
= mem_cgroup_same_or_subtree(memcg
, from
)
1329 || mem_cgroup_same_or_subtree(memcg
, to
);
1331 spin_unlock(&mc
.lock
);
1335 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*memcg
)
1337 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1338 if (mem_cgroup_under_move(memcg
)) {
1340 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1341 /* moving charge context might have finished. */
1344 finish_wait(&mc
.waitq
, &wait
);
1352 * Take this lock when
1353 * - a code tries to modify page's memcg while it's USED.
1354 * - a code tries to modify page state accounting in a memcg.
1355 * see mem_cgroup_stolen(), too.
1357 static void move_lock_mem_cgroup(struct mem_cgroup
*memcg
,
1358 unsigned long *flags
)
1360 spin_lock_irqsave(&memcg
->move_lock
, *flags
);
1363 static void move_unlock_mem_cgroup(struct mem_cgroup
*memcg
,
1364 unsigned long *flags
)
1366 spin_unlock_irqrestore(&memcg
->move_lock
, *flags
);
1370 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1371 * @memcg: The memory cgroup that went over limit
1372 * @p: Task that is going to be killed
1374 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1377 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1379 struct cgroup
*task_cgrp
;
1380 struct cgroup
*mem_cgrp
;
1382 * Need a buffer in BSS, can't rely on allocations. The code relies
1383 * on the assumption that OOM is serialized for memory controller.
1384 * If this assumption is broken, revisit this code.
1386 static char memcg_name
[PATH_MAX
];
1394 mem_cgrp
= memcg
->css
.cgroup
;
1395 task_cgrp
= task_cgroup(p
, mem_cgroup_subsys_id
);
1397 ret
= cgroup_path(task_cgrp
, memcg_name
, PATH_MAX
);
1400 * Unfortunately, we are unable to convert to a useful name
1401 * But we'll still print out the usage information
1408 printk(KERN_INFO
"Task in %s killed", memcg_name
);
1411 ret
= cgroup_path(mem_cgrp
, memcg_name
, PATH_MAX
);
1419 * Continues from above, so we don't need an KERN_ level
1421 printk(KERN_CONT
" as a result of limit of %s\n", memcg_name
);
1424 printk(KERN_INFO
"memory: usage %llukB, limit %llukB, failcnt %llu\n",
1425 res_counter_read_u64(&memcg
->res
, RES_USAGE
) >> 10,
1426 res_counter_read_u64(&memcg
->res
, RES_LIMIT
) >> 10,
1427 res_counter_read_u64(&memcg
->res
, RES_FAILCNT
));
1428 printk(KERN_INFO
"memory+swap: usage %llukB, limit %llukB, "
1430 res_counter_read_u64(&memcg
->memsw
, RES_USAGE
) >> 10,
1431 res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
) >> 10,
1432 res_counter_read_u64(&memcg
->memsw
, RES_FAILCNT
));
1436 * This function returns the number of memcg under hierarchy tree. Returns
1437 * 1(self count) if no children.
1439 static int mem_cgroup_count_children(struct mem_cgroup
*memcg
)
1442 struct mem_cgroup
*iter
;
1444 for_each_mem_cgroup_tree(iter
, memcg
)
1450 * Return the memory (and swap, if configured) limit for a memcg.
1452 static u64
mem_cgroup_get_limit(struct mem_cgroup
*memcg
)
1457 limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
1458 limit
+= total_swap_pages
<< PAGE_SHIFT
;
1460 memsw
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
1462 * If memsw is finite and limits the amount of swap space available
1463 * to this memcg, return that limit.
1465 return min(limit
, memsw
);
1468 void mem_cgroup_out_of_memory(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
1471 struct mem_cgroup
*iter
;
1472 unsigned long chosen_points
= 0;
1473 unsigned long totalpages
;
1474 unsigned int points
= 0;
1475 struct task_struct
*chosen
= NULL
;
1478 * If current has a pending SIGKILL, then automatically select it. The
1479 * goal is to allow it to allocate so that it may quickly exit and free
1482 if (fatal_signal_pending(current
)) {
1483 set_thread_flag(TIF_MEMDIE
);
1487 check_panic_on_oom(CONSTRAINT_MEMCG
, gfp_mask
, order
, NULL
);
1488 totalpages
= mem_cgroup_get_limit(memcg
) >> PAGE_SHIFT
? : 1;
1489 for_each_mem_cgroup_tree(iter
, memcg
) {
1490 struct cgroup
*cgroup
= iter
->css
.cgroup
;
1491 struct cgroup_iter it
;
1492 struct task_struct
*task
;
1494 cgroup_iter_start(cgroup
, &it
);
1495 while ((task
= cgroup_iter_next(cgroup
, &it
))) {
1496 switch (oom_scan_process_thread(task
, totalpages
, NULL
,
1498 case OOM_SCAN_SELECT
:
1500 put_task_struct(chosen
);
1502 chosen_points
= ULONG_MAX
;
1503 get_task_struct(chosen
);
1505 case OOM_SCAN_CONTINUE
:
1507 case OOM_SCAN_ABORT
:
1508 cgroup_iter_end(cgroup
, &it
);
1509 mem_cgroup_iter_break(memcg
, iter
);
1511 put_task_struct(chosen
);
1516 points
= oom_badness(task
, memcg
, NULL
, totalpages
);
1517 if (points
> chosen_points
) {
1519 put_task_struct(chosen
);
1521 chosen_points
= points
;
1522 get_task_struct(chosen
);
1525 cgroup_iter_end(cgroup
, &it
);
1530 points
= chosen_points
* 1000 / totalpages
;
1531 oom_kill_process(chosen
, gfp_mask
, order
, points
, totalpages
, memcg
,
1532 NULL
, "Memory cgroup out of memory");
1535 static unsigned long mem_cgroup_reclaim(struct mem_cgroup
*memcg
,
1537 unsigned long flags
)
1539 unsigned long total
= 0;
1540 bool noswap
= false;
1543 if (flags
& MEM_CGROUP_RECLAIM_NOSWAP
)
1545 if (!(flags
& MEM_CGROUP_RECLAIM_SHRINK
) && memcg
->memsw_is_minimum
)
1548 for (loop
= 0; loop
< MEM_CGROUP_MAX_RECLAIM_LOOPS
; loop
++) {
1550 drain_all_stock_async(memcg
);
1551 total
+= try_to_free_mem_cgroup_pages(memcg
, gfp_mask
, noswap
);
1553 * Allow limit shrinkers, which are triggered directly
1554 * by userspace, to catch signals and stop reclaim
1555 * after minimal progress, regardless of the margin.
1557 if (total
&& (flags
& MEM_CGROUP_RECLAIM_SHRINK
))
1559 if (mem_cgroup_margin(memcg
))
1562 * If nothing was reclaimed after two attempts, there
1563 * may be no reclaimable pages in this hierarchy.
1572 * test_mem_cgroup_node_reclaimable
1573 * @memcg: the target memcg
1574 * @nid: the node ID to be checked.
1575 * @noswap : specify true here if the user wants flle only information.
1577 * This function returns whether the specified memcg contains any
1578 * reclaimable pages on a node. Returns true if there are any reclaimable
1579 * pages in the node.
1581 static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup
*memcg
,
1582 int nid
, bool noswap
)
1584 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_FILE
))
1586 if (noswap
|| !total_swap_pages
)
1588 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_ANON
))
1593 #if MAX_NUMNODES > 1
1596 * Always updating the nodemask is not very good - even if we have an empty
1597 * list or the wrong list here, we can start from some node and traverse all
1598 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1601 static void mem_cgroup_may_update_nodemask(struct mem_cgroup
*memcg
)
1605 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1606 * pagein/pageout changes since the last update.
1608 if (!atomic_read(&memcg
->numainfo_events
))
1610 if (atomic_inc_return(&memcg
->numainfo_updating
) > 1)
1613 /* make a nodemask where this memcg uses memory from */
1614 memcg
->scan_nodes
= node_states
[N_HIGH_MEMORY
];
1616 for_each_node_mask(nid
, node_states
[N_HIGH_MEMORY
]) {
1618 if (!test_mem_cgroup_node_reclaimable(memcg
, nid
, false))
1619 node_clear(nid
, memcg
->scan_nodes
);
1622 atomic_set(&memcg
->numainfo_events
, 0);
1623 atomic_set(&memcg
->numainfo_updating
, 0);
1627 * Selecting a node where we start reclaim from. Because what we need is just
1628 * reducing usage counter, start from anywhere is O,K. Considering
1629 * memory reclaim from current node, there are pros. and cons.
1631 * Freeing memory from current node means freeing memory from a node which
1632 * we'll use or we've used. So, it may make LRU bad. And if several threads
1633 * hit limits, it will see a contention on a node. But freeing from remote
1634 * node means more costs for memory reclaim because of memory latency.
1636 * Now, we use round-robin. Better algorithm is welcomed.
1638 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1642 mem_cgroup_may_update_nodemask(memcg
);
1643 node
= memcg
->last_scanned_node
;
1645 node
= next_node(node
, memcg
->scan_nodes
);
1646 if (node
== MAX_NUMNODES
)
1647 node
= first_node(memcg
->scan_nodes
);
1649 * We call this when we hit limit, not when pages are added to LRU.
1650 * No LRU may hold pages because all pages are UNEVICTABLE or
1651 * memcg is too small and all pages are not on LRU. In that case,
1652 * we use curret node.
1654 if (unlikely(node
== MAX_NUMNODES
))
1655 node
= numa_node_id();
1657 memcg
->last_scanned_node
= node
;
1662 * Check all nodes whether it contains reclaimable pages or not.
1663 * For quick scan, we make use of scan_nodes. This will allow us to skip
1664 * unused nodes. But scan_nodes is lazily updated and may not cotain
1665 * enough new information. We need to do double check.
1667 static bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1672 * quick check...making use of scan_node.
1673 * We can skip unused nodes.
1675 if (!nodes_empty(memcg
->scan_nodes
)) {
1676 for (nid
= first_node(memcg
->scan_nodes
);
1678 nid
= next_node(nid
, memcg
->scan_nodes
)) {
1680 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1685 * Check rest of nodes.
1687 for_each_node_state(nid
, N_HIGH_MEMORY
) {
1688 if (node_isset(nid
, memcg
->scan_nodes
))
1690 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1697 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1702 static bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1704 return test_mem_cgroup_node_reclaimable(memcg
, 0, noswap
);
1708 static int mem_cgroup_soft_reclaim(struct mem_cgroup
*root_memcg
,
1711 unsigned long *total_scanned
)
1713 struct mem_cgroup
*victim
= NULL
;
1716 unsigned long excess
;
1717 unsigned long nr_scanned
;
1718 struct mem_cgroup_reclaim_cookie reclaim
= {
1723 excess
= res_counter_soft_limit_excess(&root_memcg
->res
) >> PAGE_SHIFT
;
1726 victim
= mem_cgroup_iter(root_memcg
, victim
, &reclaim
);
1731 * If we have not been able to reclaim
1732 * anything, it might because there are
1733 * no reclaimable pages under this hierarchy
1738 * We want to do more targeted reclaim.
1739 * excess >> 2 is not to excessive so as to
1740 * reclaim too much, nor too less that we keep
1741 * coming back to reclaim from this cgroup
1743 if (total
>= (excess
>> 2) ||
1744 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
))
1749 if (!mem_cgroup_reclaimable(victim
, false))
1751 total
+= mem_cgroup_shrink_node_zone(victim
, gfp_mask
, false,
1753 *total_scanned
+= nr_scanned
;
1754 if (!res_counter_soft_limit_excess(&root_memcg
->res
))
1757 mem_cgroup_iter_break(root_memcg
, victim
);
1762 * Check OOM-Killer is already running under our hierarchy.
1763 * If someone is running, return false.
1764 * Has to be called with memcg_oom_lock
1766 static bool mem_cgroup_oom_lock(struct mem_cgroup
*memcg
)
1768 struct mem_cgroup
*iter
, *failed
= NULL
;
1770 for_each_mem_cgroup_tree(iter
, memcg
) {
1771 if (iter
->oom_lock
) {
1773 * this subtree of our hierarchy is already locked
1774 * so we cannot give a lock.
1777 mem_cgroup_iter_break(memcg
, iter
);
1780 iter
->oom_lock
= true;
1787 * OK, we failed to lock the whole subtree so we have to clean up
1788 * what we set up to the failing subtree
1790 for_each_mem_cgroup_tree(iter
, memcg
) {
1791 if (iter
== failed
) {
1792 mem_cgroup_iter_break(memcg
, iter
);
1795 iter
->oom_lock
= false;
1801 * Has to be called with memcg_oom_lock
1803 static int mem_cgroup_oom_unlock(struct mem_cgroup
*memcg
)
1805 struct mem_cgroup
*iter
;
1807 for_each_mem_cgroup_tree(iter
, memcg
)
1808 iter
->oom_lock
= false;
1812 static void mem_cgroup_mark_under_oom(struct mem_cgroup
*memcg
)
1814 struct mem_cgroup
*iter
;
1816 for_each_mem_cgroup_tree(iter
, memcg
)
1817 atomic_inc(&iter
->under_oom
);
1820 static void mem_cgroup_unmark_under_oom(struct mem_cgroup
*memcg
)
1822 struct mem_cgroup
*iter
;
1825 * When a new child is created while the hierarchy is under oom,
1826 * mem_cgroup_oom_lock() may not be called. We have to use
1827 * atomic_add_unless() here.
1829 for_each_mem_cgroup_tree(iter
, memcg
)
1830 atomic_add_unless(&iter
->under_oom
, -1, 0);
1833 static DEFINE_SPINLOCK(memcg_oom_lock
);
1834 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1836 struct oom_wait_info
{
1837 struct mem_cgroup
*memcg
;
1841 static int memcg_oom_wake_function(wait_queue_t
*wait
,
1842 unsigned mode
, int sync
, void *arg
)
1844 struct mem_cgroup
*wake_memcg
= (struct mem_cgroup
*)arg
;
1845 struct mem_cgroup
*oom_wait_memcg
;
1846 struct oom_wait_info
*oom_wait_info
;
1848 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1849 oom_wait_memcg
= oom_wait_info
->memcg
;
1852 * Both of oom_wait_info->memcg and wake_memcg are stable under us.
1853 * Then we can use css_is_ancestor without taking care of RCU.
1855 if (!mem_cgroup_same_or_subtree(oom_wait_memcg
, wake_memcg
)
1856 && !mem_cgroup_same_or_subtree(wake_memcg
, oom_wait_memcg
))
1858 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1861 static void memcg_wakeup_oom(struct mem_cgroup
*memcg
)
1863 /* for filtering, pass "memcg" as argument. */
1864 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, memcg
);
1867 static void memcg_oom_recover(struct mem_cgroup
*memcg
)
1869 if (memcg
&& atomic_read(&memcg
->under_oom
))
1870 memcg_wakeup_oom(memcg
);
1874 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1876 static bool mem_cgroup_handle_oom(struct mem_cgroup
*memcg
, gfp_t mask
,
1879 struct oom_wait_info owait
;
1880 bool locked
, need_to_kill
;
1882 owait
.memcg
= memcg
;
1883 owait
.wait
.flags
= 0;
1884 owait
.wait
.func
= memcg_oom_wake_function
;
1885 owait
.wait
.private = current
;
1886 INIT_LIST_HEAD(&owait
.wait
.task_list
);
1887 need_to_kill
= true;
1888 mem_cgroup_mark_under_oom(memcg
);
1890 /* At first, try to OOM lock hierarchy under memcg.*/
1891 spin_lock(&memcg_oom_lock
);
1892 locked
= mem_cgroup_oom_lock(memcg
);
1894 * Even if signal_pending(), we can't quit charge() loop without
1895 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1896 * under OOM is always welcomed, use TASK_KILLABLE here.
1898 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1899 if (!locked
|| memcg
->oom_kill_disable
)
1900 need_to_kill
= false;
1902 mem_cgroup_oom_notify(memcg
);
1903 spin_unlock(&memcg_oom_lock
);
1906 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1907 mem_cgroup_out_of_memory(memcg
, mask
, order
);
1910 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1912 spin_lock(&memcg_oom_lock
);
1914 mem_cgroup_oom_unlock(memcg
);
1915 memcg_wakeup_oom(memcg
);
1916 spin_unlock(&memcg_oom_lock
);
1918 mem_cgroup_unmark_under_oom(memcg
);
1920 if (test_thread_flag(TIF_MEMDIE
) || fatal_signal_pending(current
))
1922 /* Give chance to dying process */
1923 schedule_timeout_uninterruptible(1);
1928 * Currently used to update mapped file statistics, but the routine can be
1929 * generalized to update other statistics as well.
1931 * Notes: Race condition
1933 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1934 * it tends to be costly. But considering some conditions, we doesn't need
1935 * to do so _always_.
1937 * Considering "charge", lock_page_cgroup() is not required because all
1938 * file-stat operations happen after a page is attached to radix-tree. There
1939 * are no race with "charge".
1941 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1942 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1943 * if there are race with "uncharge". Statistics itself is properly handled
1946 * Considering "move", this is an only case we see a race. To make the race
1947 * small, we check mm->moving_account and detect there are possibility of race
1948 * If there is, we take a lock.
1951 void __mem_cgroup_begin_update_page_stat(struct page
*page
,
1952 bool *locked
, unsigned long *flags
)
1954 struct mem_cgroup
*memcg
;
1955 struct page_cgroup
*pc
;
1957 pc
= lookup_page_cgroup(page
);
1959 memcg
= pc
->mem_cgroup
;
1960 if (unlikely(!memcg
|| !PageCgroupUsed(pc
)))
1963 * If this memory cgroup is not under account moving, we don't
1964 * need to take move_lock_mem_cgroup(). Because we already hold
1965 * rcu_read_lock(), any calls to move_account will be delayed until
1966 * rcu_read_unlock() if mem_cgroup_stolen() == true.
1968 if (!mem_cgroup_stolen(memcg
))
1971 move_lock_mem_cgroup(memcg
, flags
);
1972 if (memcg
!= pc
->mem_cgroup
|| !PageCgroupUsed(pc
)) {
1973 move_unlock_mem_cgroup(memcg
, flags
);
1979 void __mem_cgroup_end_update_page_stat(struct page
*page
, unsigned long *flags
)
1981 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1984 * It's guaranteed that pc->mem_cgroup never changes while
1985 * lock is held because a routine modifies pc->mem_cgroup
1986 * should take move_lock_mem_cgroup().
1988 move_unlock_mem_cgroup(pc
->mem_cgroup
, flags
);
1991 void mem_cgroup_update_page_stat(struct page
*page
,
1992 enum mem_cgroup_page_stat_item idx
, int val
)
1994 struct mem_cgroup
*memcg
;
1995 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1996 unsigned long uninitialized_var(flags
);
1998 if (mem_cgroup_disabled())
2001 memcg
= pc
->mem_cgroup
;
2002 if (unlikely(!memcg
|| !PageCgroupUsed(pc
)))
2006 case MEMCG_NR_FILE_MAPPED
:
2007 idx
= MEM_CGROUP_STAT_FILE_MAPPED
;
2013 this_cpu_add(memcg
->stat
->count
[idx
], val
);
2017 * size of first charge trial. "32" comes from vmscan.c's magic value.
2018 * TODO: maybe necessary to use big numbers in big irons.
2020 #define CHARGE_BATCH 32U
2021 struct memcg_stock_pcp
{
2022 struct mem_cgroup
*cached
; /* this never be root cgroup */
2023 unsigned int nr_pages
;
2024 struct work_struct work
;
2025 unsigned long flags
;
2026 #define FLUSHING_CACHED_CHARGE 0
2028 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
2029 static DEFINE_MUTEX(percpu_charge_mutex
);
2032 * Try to consume stocked charge on this cpu. If success, one page is consumed
2033 * from local stock and true is returned. If the stock is 0 or charges from a
2034 * cgroup which is not current target, returns false. This stock will be
2037 static bool consume_stock(struct mem_cgroup
*memcg
)
2039 struct memcg_stock_pcp
*stock
;
2042 stock
= &get_cpu_var(memcg_stock
);
2043 if (memcg
== stock
->cached
&& stock
->nr_pages
)
2045 else /* need to call res_counter_charge */
2047 put_cpu_var(memcg_stock
);
2052 * Returns stocks cached in percpu to res_counter and reset cached information.
2054 static void drain_stock(struct memcg_stock_pcp
*stock
)
2056 struct mem_cgroup
*old
= stock
->cached
;
2058 if (stock
->nr_pages
) {
2059 unsigned long bytes
= stock
->nr_pages
* PAGE_SIZE
;
2061 res_counter_uncharge(&old
->res
, bytes
);
2062 if (do_swap_account
)
2063 res_counter_uncharge(&old
->memsw
, bytes
);
2064 stock
->nr_pages
= 0;
2066 stock
->cached
= NULL
;
2070 * This must be called under preempt disabled or must be called by
2071 * a thread which is pinned to local cpu.
2073 static void drain_local_stock(struct work_struct
*dummy
)
2075 struct memcg_stock_pcp
*stock
= &__get_cpu_var(memcg_stock
);
2077 clear_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
);
2081 * Cache charges(val) which is from res_counter, to local per_cpu area.
2082 * This will be consumed by consume_stock() function, later.
2084 static void refill_stock(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
2086 struct memcg_stock_pcp
*stock
= &get_cpu_var(memcg_stock
);
2088 if (stock
->cached
!= memcg
) { /* reset if necessary */
2090 stock
->cached
= memcg
;
2092 stock
->nr_pages
+= nr_pages
;
2093 put_cpu_var(memcg_stock
);
2097 * Drains all per-CPU charge caches for given root_memcg resp. subtree
2098 * of the hierarchy under it. sync flag says whether we should block
2099 * until the work is done.
2101 static void drain_all_stock(struct mem_cgroup
*root_memcg
, bool sync
)
2105 /* Notify other cpus that system-wide "drain" is running */
2108 for_each_online_cpu(cpu
) {
2109 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2110 struct mem_cgroup
*memcg
;
2112 memcg
= stock
->cached
;
2113 if (!memcg
|| !stock
->nr_pages
)
2115 if (!mem_cgroup_same_or_subtree(root_memcg
, memcg
))
2117 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
)) {
2119 drain_local_stock(&stock
->work
);
2121 schedule_work_on(cpu
, &stock
->work
);
2129 for_each_online_cpu(cpu
) {
2130 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2131 if (test_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
))
2132 flush_work(&stock
->work
);
2139 * Tries to drain stocked charges in other cpus. This function is asynchronous
2140 * and just put a work per cpu for draining localy on each cpu. Caller can
2141 * expects some charges will be back to res_counter later but cannot wait for
2144 static void drain_all_stock_async(struct mem_cgroup
*root_memcg
)
2147 * If someone calls draining, avoid adding more kworker runs.
2149 if (!mutex_trylock(&percpu_charge_mutex
))
2151 drain_all_stock(root_memcg
, false);
2152 mutex_unlock(&percpu_charge_mutex
);
2155 /* This is a synchronous drain interface. */
2156 static void drain_all_stock_sync(struct mem_cgroup
*root_memcg
)
2158 /* called when force_empty is called */
2159 mutex_lock(&percpu_charge_mutex
);
2160 drain_all_stock(root_memcg
, true);
2161 mutex_unlock(&percpu_charge_mutex
);
2165 * This function drains percpu counter value from DEAD cpu and
2166 * move it to local cpu. Note that this function can be preempted.
2168 static void mem_cgroup_drain_pcp_counter(struct mem_cgroup
*memcg
, int cpu
)
2172 spin_lock(&memcg
->pcp_counter_lock
);
2173 for (i
= 0; i
< MEM_CGROUP_STAT_NSTATS
; i
++) {
2174 long x
= per_cpu(memcg
->stat
->count
[i
], cpu
);
2176 per_cpu(memcg
->stat
->count
[i
], cpu
) = 0;
2177 memcg
->nocpu_base
.count
[i
] += x
;
2179 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
2180 unsigned long x
= per_cpu(memcg
->stat
->events
[i
], cpu
);
2182 per_cpu(memcg
->stat
->events
[i
], cpu
) = 0;
2183 memcg
->nocpu_base
.events
[i
] += x
;
2185 spin_unlock(&memcg
->pcp_counter_lock
);
2188 static int __cpuinit
memcg_cpu_hotplug_callback(struct notifier_block
*nb
,
2189 unsigned long action
,
2192 int cpu
= (unsigned long)hcpu
;
2193 struct memcg_stock_pcp
*stock
;
2194 struct mem_cgroup
*iter
;
2196 if (action
== CPU_ONLINE
)
2199 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
2202 for_each_mem_cgroup(iter
)
2203 mem_cgroup_drain_pcp_counter(iter
, cpu
);
2205 stock
= &per_cpu(memcg_stock
, cpu
);
2211 /* See __mem_cgroup_try_charge() for details */
2213 CHARGE_OK
, /* success */
2214 CHARGE_RETRY
, /* need to retry but retry is not bad */
2215 CHARGE_NOMEM
, /* we can't do more. return -ENOMEM */
2216 CHARGE_WOULDBLOCK
, /* GFP_WAIT wasn't set and no enough res. */
2217 CHARGE_OOM_DIE
, /* the current is killed because of OOM */
2220 static int mem_cgroup_do_charge(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
2221 unsigned int nr_pages
, bool oom_check
)
2223 unsigned long csize
= nr_pages
* PAGE_SIZE
;
2224 struct mem_cgroup
*mem_over_limit
;
2225 struct res_counter
*fail_res
;
2226 unsigned long flags
= 0;
2229 ret
= res_counter_charge(&memcg
->res
, csize
, &fail_res
);
2232 if (!do_swap_account
)
2234 ret
= res_counter_charge(&memcg
->memsw
, csize
, &fail_res
);
2238 res_counter_uncharge(&memcg
->res
, csize
);
2239 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, memsw
);
2240 flags
|= MEM_CGROUP_RECLAIM_NOSWAP
;
2242 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, res
);
2244 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2245 * of regular pages (CHARGE_BATCH), or a single regular page (1).
2247 * Never reclaim on behalf of optional batching, retry with a
2248 * single page instead.
2250 if (nr_pages
== CHARGE_BATCH
)
2251 return CHARGE_RETRY
;
2253 if (!(gfp_mask
& __GFP_WAIT
))
2254 return CHARGE_WOULDBLOCK
;
2256 ret
= mem_cgroup_reclaim(mem_over_limit
, gfp_mask
, flags
);
2257 if (mem_cgroup_margin(mem_over_limit
) >= nr_pages
)
2258 return CHARGE_RETRY
;
2260 * Even though the limit is exceeded at this point, reclaim
2261 * may have been able to free some pages. Retry the charge
2262 * before killing the task.
2264 * Only for regular pages, though: huge pages are rather
2265 * unlikely to succeed so close to the limit, and we fall back
2266 * to regular pages anyway in case of failure.
2268 if (nr_pages
== 1 && ret
)
2269 return CHARGE_RETRY
;
2272 * At task move, charge accounts can be doubly counted. So, it's
2273 * better to wait until the end of task_move if something is going on.
2275 if (mem_cgroup_wait_acct_move(mem_over_limit
))
2276 return CHARGE_RETRY
;
2278 /* If we don't need to call oom-killer at el, return immediately */
2280 return CHARGE_NOMEM
;
2282 if (!mem_cgroup_handle_oom(mem_over_limit
, gfp_mask
, get_order(csize
)))
2283 return CHARGE_OOM_DIE
;
2285 return CHARGE_RETRY
;
2289 * __mem_cgroup_try_charge() does
2290 * 1. detect memcg to be charged against from passed *mm and *ptr,
2291 * 2. update res_counter
2292 * 3. call memory reclaim if necessary.
2294 * In some special case, if the task is fatal, fatal_signal_pending() or
2295 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2296 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2297 * as possible without any hazards. 2: all pages should have a valid
2298 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2299 * pointer, that is treated as a charge to root_mem_cgroup.
2301 * So __mem_cgroup_try_charge() will return
2302 * 0 ... on success, filling *ptr with a valid memcg pointer.
2303 * -ENOMEM ... charge failure because of resource limits.
2304 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2306 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2307 * the oom-killer can be invoked.
2309 static int __mem_cgroup_try_charge(struct mm_struct
*mm
,
2311 unsigned int nr_pages
,
2312 struct mem_cgroup
**ptr
,
2315 unsigned int batch
= max(CHARGE_BATCH
, nr_pages
);
2316 int nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2317 struct mem_cgroup
*memcg
= NULL
;
2321 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2322 * in system level. So, allow to go ahead dying process in addition to
2325 if (unlikely(test_thread_flag(TIF_MEMDIE
)
2326 || fatal_signal_pending(current
)))
2330 * We always charge the cgroup the mm_struct belongs to.
2331 * The mm_struct's mem_cgroup changes on task migration if the
2332 * thread group leader migrates. It's possible that mm is not
2333 * set, if so charge the root memcg (happens for pagecache usage).
2336 *ptr
= root_mem_cgroup
;
2338 if (*ptr
) { /* css should be a valid one */
2340 if (mem_cgroup_is_root(memcg
))
2342 if (nr_pages
== 1 && consume_stock(memcg
))
2344 css_get(&memcg
->css
);
2346 struct task_struct
*p
;
2349 p
= rcu_dereference(mm
->owner
);
2351 * Because we don't have task_lock(), "p" can exit.
2352 * In that case, "memcg" can point to root or p can be NULL with
2353 * race with swapoff. Then, we have small risk of mis-accouning.
2354 * But such kind of mis-account by race always happens because
2355 * we don't have cgroup_mutex(). It's overkill and we allo that
2357 * (*) swapoff at el will charge against mm-struct not against
2358 * task-struct. So, mm->owner can be NULL.
2360 memcg
= mem_cgroup_from_task(p
);
2362 memcg
= root_mem_cgroup
;
2363 if (mem_cgroup_is_root(memcg
)) {
2367 if (nr_pages
== 1 && consume_stock(memcg
)) {
2369 * It seems dagerous to access memcg without css_get().
2370 * But considering how consume_stok works, it's not
2371 * necessary. If consume_stock success, some charges
2372 * from this memcg are cached on this cpu. So, we
2373 * don't need to call css_get()/css_tryget() before
2374 * calling consume_stock().
2379 /* after here, we may be blocked. we need to get refcnt */
2380 if (!css_tryget(&memcg
->css
)) {
2390 /* If killed, bypass charge */
2391 if (fatal_signal_pending(current
)) {
2392 css_put(&memcg
->css
);
2397 if (oom
&& !nr_oom_retries
) {
2399 nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2402 ret
= mem_cgroup_do_charge(memcg
, gfp_mask
, batch
, oom_check
);
2406 case CHARGE_RETRY
: /* not in OOM situation but retry */
2408 css_put(&memcg
->css
);
2411 case CHARGE_WOULDBLOCK
: /* !__GFP_WAIT */
2412 css_put(&memcg
->css
);
2414 case CHARGE_NOMEM
: /* OOM routine works */
2416 css_put(&memcg
->css
);
2419 /* If oom, we never return -ENOMEM */
2422 case CHARGE_OOM_DIE
: /* Killed by OOM Killer */
2423 css_put(&memcg
->css
);
2426 } while (ret
!= CHARGE_OK
);
2428 if (batch
> nr_pages
)
2429 refill_stock(memcg
, batch
- nr_pages
);
2430 css_put(&memcg
->css
);
2438 *ptr
= root_mem_cgroup
;
2443 * Somemtimes we have to undo a charge we got by try_charge().
2444 * This function is for that and do uncharge, put css's refcnt.
2445 * gotten by try_charge().
2447 static void __mem_cgroup_cancel_charge(struct mem_cgroup
*memcg
,
2448 unsigned int nr_pages
)
2450 if (!mem_cgroup_is_root(memcg
)) {
2451 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2453 res_counter_uncharge(&memcg
->res
, bytes
);
2454 if (do_swap_account
)
2455 res_counter_uncharge(&memcg
->memsw
, bytes
);
2460 * Cancel chrages in this cgroup....doesn't propagate to parent cgroup.
2461 * This is useful when moving usage to parent cgroup.
2463 static void __mem_cgroup_cancel_local_charge(struct mem_cgroup
*memcg
,
2464 unsigned int nr_pages
)
2466 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2468 if (mem_cgroup_is_root(memcg
))
2471 res_counter_uncharge_until(&memcg
->res
, memcg
->res
.parent
, bytes
);
2472 if (do_swap_account
)
2473 res_counter_uncharge_until(&memcg
->memsw
,
2474 memcg
->memsw
.parent
, bytes
);
2478 * A helper function to get mem_cgroup from ID. must be called under
2479 * rcu_read_lock(). The caller is responsible for calling css_tryget if
2480 * the mem_cgroup is used for charging. (dropping refcnt from swap can be
2481 * called against removed memcg.)
2483 static struct mem_cgroup
*mem_cgroup_lookup(unsigned short id
)
2485 struct cgroup_subsys_state
*css
;
2487 /* ID 0 is unused ID */
2490 css
= css_lookup(&mem_cgroup_subsys
, id
);
2493 return mem_cgroup_from_css(css
);
2496 struct mem_cgroup
*try_get_mem_cgroup_from_page(struct page
*page
)
2498 struct mem_cgroup
*memcg
= NULL
;
2499 struct page_cgroup
*pc
;
2503 VM_BUG_ON(!PageLocked(page
));
2505 pc
= lookup_page_cgroup(page
);
2506 lock_page_cgroup(pc
);
2507 if (PageCgroupUsed(pc
)) {
2508 memcg
= pc
->mem_cgroup
;
2509 if (memcg
&& !css_tryget(&memcg
->css
))
2511 } else if (PageSwapCache(page
)) {
2512 ent
.val
= page_private(page
);
2513 id
= lookup_swap_cgroup_id(ent
);
2515 memcg
= mem_cgroup_lookup(id
);
2516 if (memcg
&& !css_tryget(&memcg
->css
))
2520 unlock_page_cgroup(pc
);
2524 static void __mem_cgroup_commit_charge(struct mem_cgroup
*memcg
,
2526 unsigned int nr_pages
,
2527 enum charge_type ctype
,
2530 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
2531 struct zone
*uninitialized_var(zone
);
2532 struct lruvec
*lruvec
;
2533 bool was_on_lru
= false;
2536 lock_page_cgroup(pc
);
2537 VM_BUG_ON(PageCgroupUsed(pc
));
2539 * we don't need page_cgroup_lock about tail pages, becase they are not
2540 * accessed by any other context at this point.
2544 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2545 * may already be on some other mem_cgroup's LRU. Take care of it.
2548 zone
= page_zone(page
);
2549 spin_lock_irq(&zone
->lru_lock
);
2550 if (PageLRU(page
)) {
2551 lruvec
= mem_cgroup_zone_lruvec(zone
, pc
->mem_cgroup
);
2553 del_page_from_lru_list(page
, lruvec
, page_lru(page
));
2558 pc
->mem_cgroup
= memcg
;
2560 * We access a page_cgroup asynchronously without lock_page_cgroup().
2561 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2562 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2563 * before USED bit, we need memory barrier here.
2564 * See mem_cgroup_add_lru_list(), etc.
2567 SetPageCgroupUsed(pc
);
2571 lruvec
= mem_cgroup_zone_lruvec(zone
, pc
->mem_cgroup
);
2572 VM_BUG_ON(PageLRU(page
));
2574 add_page_to_lru_list(page
, lruvec
, page_lru(page
));
2576 spin_unlock_irq(&zone
->lru_lock
);
2579 if (ctype
== MEM_CGROUP_CHARGE_TYPE_ANON
)
2584 mem_cgroup_charge_statistics(memcg
, anon
, nr_pages
);
2585 unlock_page_cgroup(pc
);
2588 * "charge_statistics" updated event counter. Then, check it.
2589 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2590 * if they exceeds softlimit.
2592 memcg_check_events(memcg
, page
);
2595 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2597 #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION)
2599 * Because tail pages are not marked as "used", set it. We're under
2600 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2601 * charge/uncharge will be never happen and move_account() is done under
2602 * compound_lock(), so we don't have to take care of races.
2604 void mem_cgroup_split_huge_fixup(struct page
*head
)
2606 struct page_cgroup
*head_pc
= lookup_page_cgroup(head
);
2607 struct page_cgroup
*pc
;
2610 if (mem_cgroup_disabled())
2612 for (i
= 1; i
< HPAGE_PMD_NR
; i
++) {
2614 pc
->mem_cgroup
= head_pc
->mem_cgroup
;
2615 smp_wmb();/* see __commit_charge() */
2616 pc
->flags
= head_pc
->flags
& ~PCGF_NOCOPY_AT_SPLIT
;
2619 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2622 * mem_cgroup_move_account - move account of the page
2624 * @nr_pages: number of regular pages (>1 for huge pages)
2625 * @pc: page_cgroup of the page.
2626 * @from: mem_cgroup which the page is moved from.
2627 * @to: mem_cgroup which the page is moved to. @from != @to.
2629 * The caller must confirm following.
2630 * - page is not on LRU (isolate_page() is useful.)
2631 * - compound_lock is held when nr_pages > 1
2633 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
2636 static int mem_cgroup_move_account(struct page
*page
,
2637 unsigned int nr_pages
,
2638 struct page_cgroup
*pc
,
2639 struct mem_cgroup
*from
,
2640 struct mem_cgroup
*to
)
2642 unsigned long flags
;
2644 bool anon
= PageAnon(page
);
2646 VM_BUG_ON(from
== to
);
2647 VM_BUG_ON(PageLRU(page
));
2649 * The page is isolated from LRU. So, collapse function
2650 * will not handle this page. But page splitting can happen.
2651 * Do this check under compound_page_lock(). The caller should
2655 if (nr_pages
> 1 && !PageTransHuge(page
))
2658 lock_page_cgroup(pc
);
2661 if (!PageCgroupUsed(pc
) || pc
->mem_cgroup
!= from
)
2664 move_lock_mem_cgroup(from
, &flags
);
2666 if (!anon
&& page_mapped(page
)) {
2667 /* Update mapped_file data for mem_cgroup */
2669 __this_cpu_dec(from
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2670 __this_cpu_inc(to
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2673 mem_cgroup_charge_statistics(from
, anon
, -nr_pages
);
2675 /* caller should have done css_get */
2676 pc
->mem_cgroup
= to
;
2677 mem_cgroup_charge_statistics(to
, anon
, nr_pages
);
2678 move_unlock_mem_cgroup(from
, &flags
);
2681 unlock_page_cgroup(pc
);
2685 memcg_check_events(to
, page
);
2686 memcg_check_events(from
, page
);
2692 * mem_cgroup_move_parent - moves page to the parent group
2693 * @page: the page to move
2694 * @pc: page_cgroup of the page
2695 * @child: page's cgroup
2697 * move charges to its parent or the root cgroup if the group has no
2698 * parent (aka use_hierarchy==0).
2699 * Although this might fail (get_page_unless_zero, isolate_lru_page or
2700 * mem_cgroup_move_account fails) the failure is always temporary and
2701 * it signals a race with a page removal/uncharge or migration. In the
2702 * first case the page is on the way out and it will vanish from the LRU
2703 * on the next attempt and the call should be retried later.
2704 * Isolation from the LRU fails only if page has been isolated from
2705 * the LRU since we looked at it and that usually means either global
2706 * reclaim or migration going on. The page will either get back to the
2708 * Finaly mem_cgroup_move_account fails only if the page got uncharged
2709 * (!PageCgroupUsed) or moved to a different group. The page will
2710 * disappear in the next attempt.
2712 static int mem_cgroup_move_parent(struct page
*page
,
2713 struct page_cgroup
*pc
,
2714 struct mem_cgroup
*child
)
2716 struct mem_cgroup
*parent
;
2717 unsigned int nr_pages
;
2718 unsigned long uninitialized_var(flags
);
2721 VM_BUG_ON(mem_cgroup_is_root(child
));
2724 if (!get_page_unless_zero(page
))
2726 if (isolate_lru_page(page
))
2729 nr_pages
= hpage_nr_pages(page
);
2731 parent
= parent_mem_cgroup(child
);
2733 * If no parent, move charges to root cgroup.
2736 parent
= root_mem_cgroup
;
2739 VM_BUG_ON(!PageTransHuge(page
));
2740 flags
= compound_lock_irqsave(page
);
2743 ret
= mem_cgroup_move_account(page
, nr_pages
,
2746 __mem_cgroup_cancel_local_charge(child
, nr_pages
);
2749 compound_unlock_irqrestore(page
, flags
);
2750 putback_lru_page(page
);
2758 * Charge the memory controller for page usage.
2760 * 0 if the charge was successful
2761 * < 0 if the cgroup is over its limit
2763 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
2764 gfp_t gfp_mask
, enum charge_type ctype
)
2766 struct mem_cgroup
*memcg
= NULL
;
2767 unsigned int nr_pages
= 1;
2771 if (PageTransHuge(page
)) {
2772 nr_pages
<<= compound_order(page
);
2773 VM_BUG_ON(!PageTransHuge(page
));
2775 * Never OOM-kill a process for a huge page. The
2776 * fault handler will fall back to regular pages.
2781 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, nr_pages
, &memcg
, oom
);
2784 __mem_cgroup_commit_charge(memcg
, page
, nr_pages
, ctype
, false);
2788 int mem_cgroup_newpage_charge(struct page
*page
,
2789 struct mm_struct
*mm
, gfp_t gfp_mask
)
2791 if (mem_cgroup_disabled())
2793 VM_BUG_ON(page_mapped(page
));
2794 VM_BUG_ON(page
->mapping
&& !PageAnon(page
));
2796 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2797 MEM_CGROUP_CHARGE_TYPE_ANON
);
2801 * While swap-in, try_charge -> commit or cancel, the page is locked.
2802 * And when try_charge() successfully returns, one refcnt to memcg without
2803 * struct page_cgroup is acquired. This refcnt will be consumed by
2804 * "commit()" or removed by "cancel()"
2806 static int __mem_cgroup_try_charge_swapin(struct mm_struct
*mm
,
2809 struct mem_cgroup
**memcgp
)
2811 struct mem_cgroup
*memcg
;
2812 struct page_cgroup
*pc
;
2815 pc
= lookup_page_cgroup(page
);
2817 * Every swap fault against a single page tries to charge the
2818 * page, bail as early as possible. shmem_unuse() encounters
2819 * already charged pages, too. The USED bit is protected by
2820 * the page lock, which serializes swap cache removal, which
2821 * in turn serializes uncharging.
2823 if (PageCgroupUsed(pc
))
2825 if (!do_swap_account
)
2827 memcg
= try_get_mem_cgroup_from_page(page
);
2831 ret
= __mem_cgroup_try_charge(NULL
, mask
, 1, memcgp
, true);
2832 css_put(&memcg
->css
);
2837 ret
= __mem_cgroup_try_charge(mm
, mask
, 1, memcgp
, true);
2843 int mem_cgroup_try_charge_swapin(struct mm_struct
*mm
, struct page
*page
,
2844 gfp_t gfp_mask
, struct mem_cgroup
**memcgp
)
2847 if (mem_cgroup_disabled())
2850 * A racing thread's fault, or swapoff, may have already
2851 * updated the pte, and even removed page from swap cache: in
2852 * those cases unuse_pte()'s pte_same() test will fail; but
2853 * there's also a KSM case which does need to charge the page.
2855 if (!PageSwapCache(page
)) {
2858 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, 1, memcgp
, true);
2863 return __mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, memcgp
);
2866 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup
*memcg
)
2868 if (mem_cgroup_disabled())
2872 __mem_cgroup_cancel_charge(memcg
, 1);
2876 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*memcg
,
2877 enum charge_type ctype
)
2879 if (mem_cgroup_disabled())
2884 __mem_cgroup_commit_charge(memcg
, page
, 1, ctype
, true);
2886 * Now swap is on-memory. This means this page may be
2887 * counted both as mem and swap....double count.
2888 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2889 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2890 * may call delete_from_swap_cache() before reach here.
2892 if (do_swap_account
&& PageSwapCache(page
)) {
2893 swp_entry_t ent
= {.val
= page_private(page
)};
2894 mem_cgroup_uncharge_swap(ent
);
2898 void mem_cgroup_commit_charge_swapin(struct page
*page
,
2899 struct mem_cgroup
*memcg
)
2901 __mem_cgroup_commit_charge_swapin(page
, memcg
,
2902 MEM_CGROUP_CHARGE_TYPE_ANON
);
2905 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
2908 struct mem_cgroup
*memcg
= NULL
;
2909 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
2912 if (mem_cgroup_disabled())
2914 if (PageCompound(page
))
2917 if (!PageSwapCache(page
))
2918 ret
= mem_cgroup_charge_common(page
, mm
, gfp_mask
, type
);
2919 else { /* page is swapcache/shmem */
2920 ret
= __mem_cgroup_try_charge_swapin(mm
, page
,
2923 __mem_cgroup_commit_charge_swapin(page
, memcg
, type
);
2928 static void mem_cgroup_do_uncharge(struct mem_cgroup
*memcg
,
2929 unsigned int nr_pages
,
2930 const enum charge_type ctype
)
2932 struct memcg_batch_info
*batch
= NULL
;
2933 bool uncharge_memsw
= true;
2935 /* If swapout, usage of swap doesn't decrease */
2936 if (!do_swap_account
|| ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
)
2937 uncharge_memsw
= false;
2939 batch
= ¤t
->memcg_batch
;
2941 * In usual, we do css_get() when we remember memcg pointer.
2942 * But in this case, we keep res->usage until end of a series of
2943 * uncharges. Then, it's ok to ignore memcg's refcnt.
2946 batch
->memcg
= memcg
;
2948 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2949 * In those cases, all pages freed continuously can be expected to be in
2950 * the same cgroup and we have chance to coalesce uncharges.
2951 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2952 * because we want to do uncharge as soon as possible.
2955 if (!batch
->do_batch
|| test_thread_flag(TIF_MEMDIE
))
2956 goto direct_uncharge
;
2959 goto direct_uncharge
;
2962 * In typical case, batch->memcg == mem. This means we can
2963 * merge a series of uncharges to an uncharge of res_counter.
2964 * If not, we uncharge res_counter ony by one.
2966 if (batch
->memcg
!= memcg
)
2967 goto direct_uncharge
;
2968 /* remember freed charge and uncharge it later */
2971 batch
->memsw_nr_pages
++;
2974 res_counter_uncharge(&memcg
->res
, nr_pages
* PAGE_SIZE
);
2976 res_counter_uncharge(&memcg
->memsw
, nr_pages
* PAGE_SIZE
);
2977 if (unlikely(batch
->memcg
!= memcg
))
2978 memcg_oom_recover(memcg
);
2982 * uncharge if !page_mapped(page)
2984 static struct mem_cgroup
*
2985 __mem_cgroup_uncharge_common(struct page
*page
, enum charge_type ctype
,
2988 struct mem_cgroup
*memcg
= NULL
;
2989 unsigned int nr_pages
= 1;
2990 struct page_cgroup
*pc
;
2993 if (mem_cgroup_disabled())
2996 VM_BUG_ON(PageSwapCache(page
));
2998 if (PageTransHuge(page
)) {
2999 nr_pages
<<= compound_order(page
);
3000 VM_BUG_ON(!PageTransHuge(page
));
3003 * Check if our page_cgroup is valid
3005 pc
= lookup_page_cgroup(page
);
3006 if (unlikely(!PageCgroupUsed(pc
)))
3009 lock_page_cgroup(pc
);
3011 memcg
= pc
->mem_cgroup
;
3013 if (!PageCgroupUsed(pc
))
3016 anon
= PageAnon(page
);
3019 case MEM_CGROUP_CHARGE_TYPE_ANON
:
3021 * Generally PageAnon tells if it's the anon statistics to be
3022 * updated; but sometimes e.g. mem_cgroup_uncharge_page() is
3023 * used before page reached the stage of being marked PageAnon.
3027 case MEM_CGROUP_CHARGE_TYPE_DROP
:
3028 /* See mem_cgroup_prepare_migration() */
3029 if (page_mapped(page
))
3032 * Pages under migration may not be uncharged. But
3033 * end_migration() /must/ be the one uncharging the
3034 * unused post-migration page and so it has to call
3035 * here with the migration bit still set. See the
3036 * res_counter handling below.
3038 if (!end_migration
&& PageCgroupMigration(pc
))
3041 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT
:
3042 if (!PageAnon(page
)) { /* Shared memory */
3043 if (page
->mapping
&& !page_is_file_cache(page
))
3045 } else if (page_mapped(page
)) /* Anon */
3052 mem_cgroup_charge_statistics(memcg
, anon
, -nr_pages
);
3054 ClearPageCgroupUsed(pc
);
3056 * pc->mem_cgroup is not cleared here. It will be accessed when it's
3057 * freed from LRU. This is safe because uncharged page is expected not
3058 * to be reused (freed soon). Exception is SwapCache, it's handled by
3059 * special functions.
3062 unlock_page_cgroup(pc
);
3064 * even after unlock, we have memcg->res.usage here and this memcg
3065 * will never be freed.
3067 memcg_check_events(memcg
, page
);
3068 if (do_swap_account
&& ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
) {
3069 mem_cgroup_swap_statistics(memcg
, true);
3070 mem_cgroup_get(memcg
);
3073 * Migration does not charge the res_counter for the
3074 * replacement page, so leave it alone when phasing out the
3075 * page that is unused after the migration.
3077 if (!end_migration
&& !mem_cgroup_is_root(memcg
))
3078 mem_cgroup_do_uncharge(memcg
, nr_pages
, ctype
);
3083 unlock_page_cgroup(pc
);
3087 void mem_cgroup_uncharge_page(struct page
*page
)
3090 if (page_mapped(page
))
3092 VM_BUG_ON(page
->mapping
&& !PageAnon(page
));
3093 if (PageSwapCache(page
))
3095 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_ANON
, false);
3098 void mem_cgroup_uncharge_cache_page(struct page
*page
)
3100 VM_BUG_ON(page_mapped(page
));
3101 VM_BUG_ON(page
->mapping
);
3102 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_CACHE
, false);
3106 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
3107 * In that cases, pages are freed continuously and we can expect pages
3108 * are in the same memcg. All these calls itself limits the number of
3109 * pages freed at once, then uncharge_start/end() is called properly.
3110 * This may be called prural(2) times in a context,
3113 void mem_cgroup_uncharge_start(void)
3115 current
->memcg_batch
.do_batch
++;
3116 /* We can do nest. */
3117 if (current
->memcg_batch
.do_batch
== 1) {
3118 current
->memcg_batch
.memcg
= NULL
;
3119 current
->memcg_batch
.nr_pages
= 0;
3120 current
->memcg_batch
.memsw_nr_pages
= 0;
3124 void mem_cgroup_uncharge_end(void)
3126 struct memcg_batch_info
*batch
= ¤t
->memcg_batch
;
3128 if (!batch
->do_batch
)
3132 if (batch
->do_batch
) /* If stacked, do nothing. */
3138 * This "batch->memcg" is valid without any css_get/put etc...
3139 * bacause we hide charges behind us.
3141 if (batch
->nr_pages
)
3142 res_counter_uncharge(&batch
->memcg
->res
,
3143 batch
->nr_pages
* PAGE_SIZE
);
3144 if (batch
->memsw_nr_pages
)
3145 res_counter_uncharge(&batch
->memcg
->memsw
,
3146 batch
->memsw_nr_pages
* PAGE_SIZE
);
3147 memcg_oom_recover(batch
->memcg
);
3148 /* forget this pointer (for sanity check) */
3149 batch
->memcg
= NULL
;
3154 * called after __delete_from_swap_cache() and drop "page" account.
3155 * memcg information is recorded to swap_cgroup of "ent"
3158 mem_cgroup_uncharge_swapcache(struct page
*page
, swp_entry_t ent
, bool swapout
)
3160 struct mem_cgroup
*memcg
;
3161 int ctype
= MEM_CGROUP_CHARGE_TYPE_SWAPOUT
;
3163 if (!swapout
) /* this was a swap cache but the swap is unused ! */
3164 ctype
= MEM_CGROUP_CHARGE_TYPE_DROP
;
3166 memcg
= __mem_cgroup_uncharge_common(page
, ctype
, false);
3169 * record memcg information, if swapout && memcg != NULL,
3170 * mem_cgroup_get() was called in uncharge().
3172 if (do_swap_account
&& swapout
&& memcg
)
3173 swap_cgroup_record(ent
, css_id(&memcg
->css
));
3177 #ifdef CONFIG_MEMCG_SWAP
3179 * called from swap_entry_free(). remove record in swap_cgroup and
3180 * uncharge "memsw" account.
3182 void mem_cgroup_uncharge_swap(swp_entry_t ent
)
3184 struct mem_cgroup
*memcg
;
3187 if (!do_swap_account
)
3190 id
= swap_cgroup_record(ent
, 0);
3192 memcg
= mem_cgroup_lookup(id
);
3195 * We uncharge this because swap is freed.
3196 * This memcg can be obsolete one. We avoid calling css_tryget
3198 if (!mem_cgroup_is_root(memcg
))
3199 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
3200 mem_cgroup_swap_statistics(memcg
, false);
3201 mem_cgroup_put(memcg
);
3207 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3208 * @entry: swap entry to be moved
3209 * @from: mem_cgroup which the entry is moved from
3210 * @to: mem_cgroup which the entry is moved to
3212 * It succeeds only when the swap_cgroup's record for this entry is the same
3213 * as the mem_cgroup's id of @from.
3215 * Returns 0 on success, -EINVAL on failure.
3217 * The caller must have charged to @to, IOW, called res_counter_charge() about
3218 * both res and memsw, and called css_get().
3220 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
3221 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
3223 unsigned short old_id
, new_id
;
3225 old_id
= css_id(&from
->css
);
3226 new_id
= css_id(&to
->css
);
3228 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
3229 mem_cgroup_swap_statistics(from
, false);
3230 mem_cgroup_swap_statistics(to
, true);
3232 * This function is only called from task migration context now.
3233 * It postpones res_counter and refcount handling till the end
3234 * of task migration(mem_cgroup_clear_mc()) for performance
3235 * improvement. But we cannot postpone mem_cgroup_get(to)
3236 * because if the process that has been moved to @to does
3237 * swap-in, the refcount of @to might be decreased to 0.
3245 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
3246 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
3253 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3256 void mem_cgroup_prepare_migration(struct page
*page
, struct page
*newpage
,
3257 struct mem_cgroup
**memcgp
)
3259 struct mem_cgroup
*memcg
= NULL
;
3260 struct page_cgroup
*pc
;
3261 enum charge_type ctype
;
3265 VM_BUG_ON(PageTransHuge(page
));
3266 if (mem_cgroup_disabled())
3269 pc
= lookup_page_cgroup(page
);
3270 lock_page_cgroup(pc
);
3271 if (PageCgroupUsed(pc
)) {
3272 memcg
= pc
->mem_cgroup
;
3273 css_get(&memcg
->css
);
3275 * At migrating an anonymous page, its mapcount goes down
3276 * to 0 and uncharge() will be called. But, even if it's fully
3277 * unmapped, migration may fail and this page has to be
3278 * charged again. We set MIGRATION flag here and delay uncharge
3279 * until end_migration() is called
3281 * Corner Case Thinking
3283 * When the old page was mapped as Anon and it's unmap-and-freed
3284 * while migration was ongoing.
3285 * If unmap finds the old page, uncharge() of it will be delayed
3286 * until end_migration(). If unmap finds a new page, it's
3287 * uncharged when it make mapcount to be 1->0. If unmap code
3288 * finds swap_migration_entry, the new page will not be mapped
3289 * and end_migration() will find it(mapcount==0).
3292 * When the old page was mapped but migraion fails, the kernel
3293 * remaps it. A charge for it is kept by MIGRATION flag even
3294 * if mapcount goes down to 0. We can do remap successfully
3295 * without charging it again.
3298 * The "old" page is under lock_page() until the end of
3299 * migration, so, the old page itself will not be swapped-out.
3300 * If the new page is swapped out before end_migraton, our
3301 * hook to usual swap-out path will catch the event.
3304 SetPageCgroupMigration(pc
);
3306 unlock_page_cgroup(pc
);
3308 * If the page is not charged at this point,
3316 * We charge new page before it's used/mapped. So, even if unlock_page()
3317 * is called before end_migration, we can catch all events on this new
3318 * page. In the case new page is migrated but not remapped, new page's
3319 * mapcount will be finally 0 and we call uncharge in end_migration().
3322 ctype
= MEM_CGROUP_CHARGE_TYPE_ANON
;
3324 ctype
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3326 * The page is committed to the memcg, but it's not actually
3327 * charged to the res_counter since we plan on replacing the
3328 * old one and only one page is going to be left afterwards.
3330 __mem_cgroup_commit_charge(memcg
, newpage
, 1, ctype
, false);
3333 /* remove redundant charge if migration failed*/
3334 void mem_cgroup_end_migration(struct mem_cgroup
*memcg
,
3335 struct page
*oldpage
, struct page
*newpage
, bool migration_ok
)
3337 struct page
*used
, *unused
;
3338 struct page_cgroup
*pc
;
3344 if (!migration_ok
) {
3351 anon
= PageAnon(used
);
3352 __mem_cgroup_uncharge_common(unused
,
3353 anon
? MEM_CGROUP_CHARGE_TYPE_ANON
3354 : MEM_CGROUP_CHARGE_TYPE_CACHE
,
3356 css_put(&memcg
->css
);
3358 * We disallowed uncharge of pages under migration because mapcount
3359 * of the page goes down to zero, temporarly.
3360 * Clear the flag and check the page should be charged.
3362 pc
= lookup_page_cgroup(oldpage
);
3363 lock_page_cgroup(pc
);
3364 ClearPageCgroupMigration(pc
);
3365 unlock_page_cgroup(pc
);
3368 * If a page is a file cache, radix-tree replacement is very atomic
3369 * and we can skip this check. When it was an Anon page, its mapcount
3370 * goes down to 0. But because we added MIGRATION flage, it's not
3371 * uncharged yet. There are several case but page->mapcount check
3372 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3373 * check. (see prepare_charge() also)
3376 mem_cgroup_uncharge_page(used
);
3380 * At replace page cache, newpage is not under any memcg but it's on
3381 * LRU. So, this function doesn't touch res_counter but handles LRU
3382 * in correct way. Both pages are locked so we cannot race with uncharge.
3384 void mem_cgroup_replace_page_cache(struct page
*oldpage
,
3385 struct page
*newpage
)
3387 struct mem_cgroup
*memcg
= NULL
;
3388 struct page_cgroup
*pc
;
3389 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3391 if (mem_cgroup_disabled())
3394 pc
= lookup_page_cgroup(oldpage
);
3395 /* fix accounting on old pages */
3396 lock_page_cgroup(pc
);
3397 if (PageCgroupUsed(pc
)) {
3398 memcg
= pc
->mem_cgroup
;
3399 mem_cgroup_charge_statistics(memcg
, false, -1);
3400 ClearPageCgroupUsed(pc
);
3402 unlock_page_cgroup(pc
);
3405 * When called from shmem_replace_page(), in some cases the
3406 * oldpage has already been charged, and in some cases not.
3411 * Even if newpage->mapping was NULL before starting replacement,
3412 * the newpage may be on LRU(or pagevec for LRU) already. We lock
3413 * LRU while we overwrite pc->mem_cgroup.
3415 __mem_cgroup_commit_charge(memcg
, newpage
, 1, type
, true);
3418 #ifdef CONFIG_DEBUG_VM
3419 static struct page_cgroup
*lookup_page_cgroup_used(struct page
*page
)
3421 struct page_cgroup
*pc
;
3423 pc
= lookup_page_cgroup(page
);
3425 * Can be NULL while feeding pages into the page allocator for
3426 * the first time, i.e. during boot or memory hotplug;
3427 * or when mem_cgroup_disabled().
3429 if (likely(pc
) && PageCgroupUsed(pc
))
3434 bool mem_cgroup_bad_page_check(struct page
*page
)
3436 if (mem_cgroup_disabled())
3439 return lookup_page_cgroup_used(page
) != NULL
;
3442 void mem_cgroup_print_bad_page(struct page
*page
)
3444 struct page_cgroup
*pc
;
3446 pc
= lookup_page_cgroup_used(page
);
3448 printk(KERN_ALERT
"pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
3449 pc
, pc
->flags
, pc
->mem_cgroup
);
3454 static DEFINE_MUTEX(set_limit_mutex
);
3456 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
3457 unsigned long long val
)
3460 u64 memswlimit
, memlimit
;
3462 int children
= mem_cgroup_count_children(memcg
);
3463 u64 curusage
, oldusage
;
3467 * For keeping hierarchical_reclaim simple, how long we should retry
3468 * is depends on callers. We set our retry-count to be function
3469 * of # of children which we should visit in this loop.
3471 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
* children
;
3473 oldusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3476 while (retry_count
) {
3477 if (signal_pending(current
)) {
3482 * Rather than hide all in some function, I do this in
3483 * open coded manner. You see what this really does.
3484 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
3486 mutex_lock(&set_limit_mutex
);
3487 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3488 if (memswlimit
< val
) {
3490 mutex_unlock(&set_limit_mutex
);
3494 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3498 ret
= res_counter_set_limit(&memcg
->res
, val
);
3500 if (memswlimit
== val
)
3501 memcg
->memsw_is_minimum
= true;
3503 memcg
->memsw_is_minimum
= false;
3505 mutex_unlock(&set_limit_mutex
);
3510 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3511 MEM_CGROUP_RECLAIM_SHRINK
);
3512 curusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3513 /* Usage is reduced ? */
3514 if (curusage
>= oldusage
)
3517 oldusage
= curusage
;
3519 if (!ret
&& enlarge
)
3520 memcg_oom_recover(memcg
);
3525 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
3526 unsigned long long val
)
3529 u64 memlimit
, memswlimit
, oldusage
, curusage
;
3530 int children
= mem_cgroup_count_children(memcg
);
3534 /* see mem_cgroup_resize_res_limit */
3535 retry_count
= children
* MEM_CGROUP_RECLAIM_RETRIES
;
3536 oldusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3537 while (retry_count
) {
3538 if (signal_pending(current
)) {
3543 * Rather than hide all in some function, I do this in
3544 * open coded manner. You see what this really does.
3545 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
3547 mutex_lock(&set_limit_mutex
);
3548 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3549 if (memlimit
> val
) {
3551 mutex_unlock(&set_limit_mutex
);
3554 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3555 if (memswlimit
< val
)
3557 ret
= res_counter_set_limit(&memcg
->memsw
, val
);
3559 if (memlimit
== val
)
3560 memcg
->memsw_is_minimum
= true;
3562 memcg
->memsw_is_minimum
= false;
3564 mutex_unlock(&set_limit_mutex
);
3569 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3570 MEM_CGROUP_RECLAIM_NOSWAP
|
3571 MEM_CGROUP_RECLAIM_SHRINK
);
3572 curusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3573 /* Usage is reduced ? */
3574 if (curusage
>= oldusage
)
3577 oldusage
= curusage
;
3579 if (!ret
&& enlarge
)
3580 memcg_oom_recover(memcg
);
3584 unsigned long mem_cgroup_soft_limit_reclaim(struct zone
*zone
, int order
,
3586 unsigned long *total_scanned
)
3588 unsigned long nr_reclaimed
= 0;
3589 struct mem_cgroup_per_zone
*mz
, *next_mz
= NULL
;
3590 unsigned long reclaimed
;
3592 struct mem_cgroup_tree_per_zone
*mctz
;
3593 unsigned long long excess
;
3594 unsigned long nr_scanned
;
3599 mctz
= soft_limit_tree_node_zone(zone_to_nid(zone
), zone_idx(zone
));
3601 * This loop can run a while, specially if mem_cgroup's continuously
3602 * keep exceeding their soft limit and putting the system under
3609 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
3614 reclaimed
= mem_cgroup_soft_reclaim(mz
->memcg
, zone
,
3615 gfp_mask
, &nr_scanned
);
3616 nr_reclaimed
+= reclaimed
;
3617 *total_scanned
+= nr_scanned
;
3618 spin_lock(&mctz
->lock
);
3621 * If we failed to reclaim anything from this memory cgroup
3622 * it is time to move on to the next cgroup
3628 * Loop until we find yet another one.
3630 * By the time we get the soft_limit lock
3631 * again, someone might have aded the
3632 * group back on the RB tree. Iterate to
3633 * make sure we get a different mem.
3634 * mem_cgroup_largest_soft_limit_node returns
3635 * NULL if no other cgroup is present on
3639 __mem_cgroup_largest_soft_limit_node(mctz
);
3641 css_put(&next_mz
->memcg
->css
);
3642 else /* next_mz == NULL or other memcg */
3646 __mem_cgroup_remove_exceeded(mz
->memcg
, mz
, mctz
);
3647 excess
= res_counter_soft_limit_excess(&mz
->memcg
->res
);
3649 * One school of thought says that we should not add
3650 * back the node to the tree if reclaim returns 0.
3651 * But our reclaim could return 0, simply because due
3652 * to priority we are exposing a smaller subset of
3653 * memory to reclaim from. Consider this as a longer
3656 /* If excess == 0, no tree ops */
3657 __mem_cgroup_insert_exceeded(mz
->memcg
, mz
, mctz
, excess
);
3658 spin_unlock(&mctz
->lock
);
3659 css_put(&mz
->memcg
->css
);
3662 * Could not reclaim anything and there are no more
3663 * mem cgroups to try or we seem to be looping without
3664 * reclaiming anything.
3666 if (!nr_reclaimed
&&
3668 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
3670 } while (!nr_reclaimed
);
3672 css_put(&next_mz
->memcg
->css
);
3673 return nr_reclaimed
;
3677 * mem_cgroup_force_empty_list - clears LRU of a group
3678 * @memcg: group to clear
3681 * @lru: lru to to clear
3683 * Traverse a specified page_cgroup list and try to drop them all. This doesn't
3684 * reclaim the pages page themselves - pages are moved to the parent (or root)
3687 static void mem_cgroup_force_empty_list(struct mem_cgroup
*memcg
,
3688 int node
, int zid
, enum lru_list lru
)
3690 struct mem_cgroup_per_zone
*mz
;
3691 unsigned long flags
;
3692 struct list_head
*list
;
3696 zone
= &NODE_DATA(node
)->node_zones
[zid
];
3697 mz
= mem_cgroup_zoneinfo(memcg
, node
, zid
);
3698 list
= &mz
->lruvec
.lists
[lru
];
3702 struct page_cgroup
*pc
;
3705 spin_lock_irqsave(&zone
->lru_lock
, flags
);
3706 if (list_empty(list
)) {
3707 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3710 page
= list_entry(list
->prev
, struct page
, lru
);
3712 list_move(&page
->lru
, list
);
3714 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3717 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3719 pc
= lookup_page_cgroup(page
);
3721 if (mem_cgroup_move_parent(page
, pc
, memcg
)) {
3722 /* found lock contention or "pc" is obsolete. */
3727 } while (!list_empty(list
));
3731 * make mem_cgroup's charge to be 0 if there is no task by moving
3732 * all the charges and pages to the parent.
3733 * This enables deleting this mem_cgroup.
3735 * Caller is responsible for holding css reference on the memcg.
3737 static void mem_cgroup_reparent_charges(struct mem_cgroup
*memcg
)
3742 /* This is for making all *used* pages to be on LRU. */
3743 lru_add_drain_all();
3744 drain_all_stock_sync(memcg
);
3745 mem_cgroup_start_move(memcg
);
3746 for_each_node_state(node
, N_HIGH_MEMORY
) {
3747 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
3750 mem_cgroup_force_empty_list(memcg
,
3755 mem_cgroup_end_move(memcg
);
3756 memcg_oom_recover(memcg
);
3760 * This is a safety check because mem_cgroup_force_empty_list
3761 * could have raced with mem_cgroup_replace_page_cache callers
3762 * so the lru seemed empty but the page could have been added
3763 * right after the check. RES_USAGE should be safe as we always
3764 * charge before adding to the LRU.
3766 } while (res_counter_read_u64(&memcg
->res
, RES_USAGE
) > 0);
3770 * Reclaims as many pages from the given memcg as possible and moves
3771 * the rest to the parent.
3773 * Caller is responsible for holding css reference for memcg.
3775 static int mem_cgroup_force_empty(struct mem_cgroup
*memcg
)
3777 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
3778 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
3780 /* returns EBUSY if there is a task or if we come here twice. */
3781 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
))
3784 /* we call try-to-free pages for make this cgroup empty */
3785 lru_add_drain_all();
3786 /* try to free all pages in this cgroup */
3787 while (nr_retries
&& res_counter_read_u64(&memcg
->res
, RES_USAGE
) > 0) {
3790 if (signal_pending(current
))
3793 progress
= try_to_free_mem_cgroup_pages(memcg
, GFP_KERNEL
,
3797 /* maybe some writeback is necessary */
3798 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3803 mem_cgroup_reparent_charges(memcg
);
3808 static int mem_cgroup_force_empty_write(struct cgroup
*cont
, unsigned int event
)
3810 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3813 if (mem_cgroup_is_root(memcg
))
3815 css_get(&memcg
->css
);
3816 ret
= mem_cgroup_force_empty(memcg
);
3817 css_put(&memcg
->css
);
3823 static u64
mem_cgroup_hierarchy_read(struct cgroup
*cont
, struct cftype
*cft
)
3825 return mem_cgroup_from_cont(cont
)->use_hierarchy
;
3828 static int mem_cgroup_hierarchy_write(struct cgroup
*cont
, struct cftype
*cft
,
3832 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3833 struct cgroup
*parent
= cont
->parent
;
3834 struct mem_cgroup
*parent_memcg
= NULL
;
3837 parent_memcg
= mem_cgroup_from_cont(parent
);
3841 if (memcg
->use_hierarchy
== val
)
3845 * If parent's use_hierarchy is set, we can't make any modifications
3846 * in the child subtrees. If it is unset, then the change can
3847 * occur, provided the current cgroup has no children.
3849 * For the root cgroup, parent_mem is NULL, we allow value to be
3850 * set if there are no children.
3852 if ((!parent_memcg
|| !parent_memcg
->use_hierarchy
) &&
3853 (val
== 1 || val
== 0)) {
3854 if (list_empty(&cont
->children
))
3855 memcg
->use_hierarchy
= val
;
3868 static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup
*memcg
,
3869 enum mem_cgroup_stat_index idx
)
3871 struct mem_cgroup
*iter
;
3874 /* Per-cpu values can be negative, use a signed accumulator */
3875 for_each_mem_cgroup_tree(iter
, memcg
)
3876 val
+= mem_cgroup_read_stat(iter
, idx
);
3878 if (val
< 0) /* race ? */
3883 static inline u64
mem_cgroup_usage(struct mem_cgroup
*memcg
, bool swap
)
3887 if (!mem_cgroup_is_root(memcg
)) {
3889 return res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3891 return res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3894 val
= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_CACHE
);
3895 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_RSS
);
3898 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_SWAP
);
3900 return val
<< PAGE_SHIFT
;
3903 static ssize_t
mem_cgroup_read(struct cgroup
*cont
, struct cftype
*cft
,
3904 struct file
*file
, char __user
*buf
,
3905 size_t nbytes
, loff_t
*ppos
)
3907 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3910 int type
, name
, len
;
3912 type
= MEMFILE_TYPE(cft
->private);
3913 name
= MEMFILE_ATTR(cft
->private);
3915 if (!do_swap_account
&& type
== _MEMSWAP
)
3920 if (name
== RES_USAGE
)
3921 val
= mem_cgroup_usage(memcg
, false);
3923 val
= res_counter_read_u64(&memcg
->res
, name
);
3926 if (name
== RES_USAGE
)
3927 val
= mem_cgroup_usage(memcg
, true);
3929 val
= res_counter_read_u64(&memcg
->memsw
, name
);
3935 len
= scnprintf(str
, sizeof(str
), "%llu\n", (unsigned long long)val
);
3936 return simple_read_from_buffer(buf
, nbytes
, ppos
, str
, len
);
3939 * The user of this function is...
3942 static int mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
3945 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3947 unsigned long long val
;
3950 type
= MEMFILE_TYPE(cft
->private);
3951 name
= MEMFILE_ATTR(cft
->private);
3953 if (!do_swap_account
&& type
== _MEMSWAP
)
3958 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
3962 /* This function does all necessary parse...reuse it */
3963 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3967 ret
= mem_cgroup_resize_limit(memcg
, val
);
3969 ret
= mem_cgroup_resize_memsw_limit(memcg
, val
);
3971 case RES_SOFT_LIMIT
:
3972 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3976 * For memsw, soft limits are hard to implement in terms
3977 * of semantics, for now, we support soft limits for
3978 * control without swap
3981 ret
= res_counter_set_soft_limit(&memcg
->res
, val
);
3986 ret
= -EINVAL
; /* should be BUG() ? */
3992 static void memcg_get_hierarchical_limit(struct mem_cgroup
*memcg
,
3993 unsigned long long *mem_limit
, unsigned long long *memsw_limit
)
3995 struct cgroup
*cgroup
;
3996 unsigned long long min_limit
, min_memsw_limit
, tmp
;
3998 min_limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3999 min_memsw_limit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
4000 cgroup
= memcg
->css
.cgroup
;
4001 if (!memcg
->use_hierarchy
)
4004 while (cgroup
->parent
) {
4005 cgroup
= cgroup
->parent
;
4006 memcg
= mem_cgroup_from_cont(cgroup
);
4007 if (!memcg
->use_hierarchy
)
4009 tmp
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
4010 min_limit
= min(min_limit
, tmp
);
4011 tmp
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
4012 min_memsw_limit
= min(min_memsw_limit
, tmp
);
4015 *mem_limit
= min_limit
;
4016 *memsw_limit
= min_memsw_limit
;
4019 static int mem_cgroup_reset(struct cgroup
*cont
, unsigned int event
)
4021 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4024 type
= MEMFILE_TYPE(event
);
4025 name
= MEMFILE_ATTR(event
);
4027 if (!do_swap_account
&& type
== _MEMSWAP
)
4033 res_counter_reset_max(&memcg
->res
);
4035 res_counter_reset_max(&memcg
->memsw
);
4039 res_counter_reset_failcnt(&memcg
->res
);
4041 res_counter_reset_failcnt(&memcg
->memsw
);
4048 static u64
mem_cgroup_move_charge_read(struct cgroup
*cgrp
,
4051 return mem_cgroup_from_cont(cgrp
)->move_charge_at_immigrate
;
4055 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4056 struct cftype
*cft
, u64 val
)
4058 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4060 if (val
>= (1 << NR_MOVE_TYPE
))
4063 * We check this value several times in both in can_attach() and
4064 * attach(), so we need cgroup lock to prevent this value from being
4068 memcg
->move_charge_at_immigrate
= val
;
4074 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4075 struct cftype
*cft
, u64 val
)
4082 static int memcg_numa_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
4086 unsigned long total_nr
, file_nr
, anon_nr
, unevictable_nr
;
4087 unsigned long node_nr
;
4088 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4090 total_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL
);
4091 seq_printf(m
, "total=%lu", total_nr
);
4092 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4093 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL
);
4094 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4098 file_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL_FILE
);
4099 seq_printf(m
, "file=%lu", file_nr
);
4100 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4101 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4103 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4107 anon_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL_ANON
);
4108 seq_printf(m
, "anon=%lu", anon_nr
);
4109 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4110 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4112 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4116 unevictable_nr
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_UNEVICTABLE
));
4117 seq_printf(m
, "unevictable=%lu", unevictable_nr
);
4118 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4119 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4120 BIT(LRU_UNEVICTABLE
));
4121 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4126 #endif /* CONFIG_NUMA */
4128 static const char * const mem_cgroup_lru_names
[] = {
4136 static inline void mem_cgroup_lru_names_not_uptodate(void)
4138 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names
) != NR_LRU_LISTS
);
4141 static int memcg_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
4144 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4145 struct mem_cgroup
*mi
;
4148 for (i
= 0; i
< MEM_CGROUP_STAT_NSTATS
; i
++) {
4149 if (i
== MEM_CGROUP_STAT_SWAP
&& !do_swap_account
)
4151 seq_printf(m
, "%s %ld\n", mem_cgroup_stat_names
[i
],
4152 mem_cgroup_read_stat(memcg
, i
) * PAGE_SIZE
);
4155 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++)
4156 seq_printf(m
, "%s %lu\n", mem_cgroup_events_names
[i
],
4157 mem_cgroup_read_events(memcg
, i
));
4159 for (i
= 0; i
< NR_LRU_LISTS
; i
++)
4160 seq_printf(m
, "%s %lu\n", mem_cgroup_lru_names
[i
],
4161 mem_cgroup_nr_lru_pages(memcg
, BIT(i
)) * PAGE_SIZE
);
4163 /* Hierarchical information */
4165 unsigned long long limit
, memsw_limit
;
4166 memcg_get_hierarchical_limit(memcg
, &limit
, &memsw_limit
);
4167 seq_printf(m
, "hierarchical_memory_limit %llu\n", limit
);
4168 if (do_swap_account
)
4169 seq_printf(m
, "hierarchical_memsw_limit %llu\n",
4173 for (i
= 0; i
< MEM_CGROUP_STAT_NSTATS
; i
++) {
4176 if (i
== MEM_CGROUP_STAT_SWAP
&& !do_swap_account
)
4178 for_each_mem_cgroup_tree(mi
, memcg
)
4179 val
+= mem_cgroup_read_stat(mi
, i
) * PAGE_SIZE
;
4180 seq_printf(m
, "total_%s %lld\n", mem_cgroup_stat_names
[i
], val
);
4183 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
4184 unsigned long long val
= 0;
4186 for_each_mem_cgroup_tree(mi
, memcg
)
4187 val
+= mem_cgroup_read_events(mi
, i
);
4188 seq_printf(m
, "total_%s %llu\n",
4189 mem_cgroup_events_names
[i
], val
);
4192 for (i
= 0; i
< NR_LRU_LISTS
; i
++) {
4193 unsigned long long val
= 0;
4195 for_each_mem_cgroup_tree(mi
, memcg
)
4196 val
+= mem_cgroup_nr_lru_pages(mi
, BIT(i
)) * PAGE_SIZE
;
4197 seq_printf(m
, "total_%s %llu\n", mem_cgroup_lru_names
[i
], val
);
4200 #ifdef CONFIG_DEBUG_VM
4203 struct mem_cgroup_per_zone
*mz
;
4204 struct zone_reclaim_stat
*rstat
;
4205 unsigned long recent_rotated
[2] = {0, 0};
4206 unsigned long recent_scanned
[2] = {0, 0};
4208 for_each_online_node(nid
)
4209 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
4210 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
4211 rstat
= &mz
->lruvec
.reclaim_stat
;
4213 recent_rotated
[0] += rstat
->recent_rotated
[0];
4214 recent_rotated
[1] += rstat
->recent_rotated
[1];
4215 recent_scanned
[0] += rstat
->recent_scanned
[0];
4216 recent_scanned
[1] += rstat
->recent_scanned
[1];
4218 seq_printf(m
, "recent_rotated_anon %lu\n", recent_rotated
[0]);
4219 seq_printf(m
, "recent_rotated_file %lu\n", recent_rotated
[1]);
4220 seq_printf(m
, "recent_scanned_anon %lu\n", recent_scanned
[0]);
4221 seq_printf(m
, "recent_scanned_file %lu\n", recent_scanned
[1]);
4228 static u64
mem_cgroup_swappiness_read(struct cgroup
*cgrp
, struct cftype
*cft
)
4230 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4232 return mem_cgroup_swappiness(memcg
);
4235 static int mem_cgroup_swappiness_write(struct cgroup
*cgrp
, struct cftype
*cft
,
4238 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4239 struct mem_cgroup
*parent
;
4244 if (cgrp
->parent
== NULL
)
4247 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4251 /* If under hierarchy, only empty-root can set this value */
4252 if ((parent
->use_hierarchy
) ||
4253 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4258 memcg
->swappiness
= val
;
4265 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
4267 struct mem_cgroup_threshold_ary
*t
;
4273 t
= rcu_dereference(memcg
->thresholds
.primary
);
4275 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
4280 usage
= mem_cgroup_usage(memcg
, swap
);
4283 * current_threshold points to threshold just below or equal to usage.
4284 * If it's not true, a threshold was crossed after last
4285 * call of __mem_cgroup_threshold().
4287 i
= t
->current_threshold
;
4290 * Iterate backward over array of thresholds starting from
4291 * current_threshold and check if a threshold is crossed.
4292 * If none of thresholds below usage is crossed, we read
4293 * only one element of the array here.
4295 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
4296 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4298 /* i = current_threshold + 1 */
4302 * Iterate forward over array of thresholds starting from
4303 * current_threshold+1 and check if a threshold is crossed.
4304 * If none of thresholds above usage is crossed, we read
4305 * only one element of the array here.
4307 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
4308 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4310 /* Update current_threshold */
4311 t
->current_threshold
= i
- 1;
4316 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
4319 __mem_cgroup_threshold(memcg
, false);
4320 if (do_swap_account
)
4321 __mem_cgroup_threshold(memcg
, true);
4323 memcg
= parent_mem_cgroup(memcg
);
4327 static int compare_thresholds(const void *a
, const void *b
)
4329 const struct mem_cgroup_threshold
*_a
= a
;
4330 const struct mem_cgroup_threshold
*_b
= b
;
4332 return _a
->threshold
- _b
->threshold
;
4335 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*memcg
)
4337 struct mem_cgroup_eventfd_list
*ev
;
4339 list_for_each_entry(ev
, &memcg
->oom_notify
, list
)
4340 eventfd_signal(ev
->eventfd
, 1);
4344 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
)
4346 struct mem_cgroup
*iter
;
4348 for_each_mem_cgroup_tree(iter
, memcg
)
4349 mem_cgroup_oom_notify_cb(iter
);
4352 static int mem_cgroup_usage_register_event(struct cgroup
*cgrp
,
4353 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4355 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4356 struct mem_cgroup_thresholds
*thresholds
;
4357 struct mem_cgroup_threshold_ary
*new;
4358 int type
= MEMFILE_TYPE(cft
->private);
4359 u64 threshold
, usage
;
4362 ret
= res_counter_memparse_write_strategy(args
, &threshold
);
4366 mutex_lock(&memcg
->thresholds_lock
);
4369 thresholds
= &memcg
->thresholds
;
4370 else if (type
== _MEMSWAP
)
4371 thresholds
= &memcg
->memsw_thresholds
;
4375 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4377 /* Check if a threshold crossed before adding a new one */
4378 if (thresholds
->primary
)
4379 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4381 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
4383 /* Allocate memory for new array of thresholds */
4384 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
4392 /* Copy thresholds (if any) to new array */
4393 if (thresholds
->primary
) {
4394 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
4395 sizeof(struct mem_cgroup_threshold
));
4398 /* Add new threshold */
4399 new->entries
[size
- 1].eventfd
= eventfd
;
4400 new->entries
[size
- 1].threshold
= threshold
;
4402 /* Sort thresholds. Registering of new threshold isn't time-critical */
4403 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
4404 compare_thresholds
, NULL
);
4406 /* Find current threshold */
4407 new->current_threshold
= -1;
4408 for (i
= 0; i
< size
; i
++) {
4409 if (new->entries
[i
].threshold
<= usage
) {
4411 * new->current_threshold will not be used until
4412 * rcu_assign_pointer(), so it's safe to increment
4415 ++new->current_threshold
;
4420 /* Free old spare buffer and save old primary buffer as spare */
4421 kfree(thresholds
->spare
);
4422 thresholds
->spare
= thresholds
->primary
;
4424 rcu_assign_pointer(thresholds
->primary
, new);
4426 /* To be sure that nobody uses thresholds */
4430 mutex_unlock(&memcg
->thresholds_lock
);
4435 static void mem_cgroup_usage_unregister_event(struct cgroup
*cgrp
,
4436 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4438 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4439 struct mem_cgroup_thresholds
*thresholds
;
4440 struct mem_cgroup_threshold_ary
*new;
4441 int type
= MEMFILE_TYPE(cft
->private);
4445 mutex_lock(&memcg
->thresholds_lock
);
4447 thresholds
= &memcg
->thresholds
;
4448 else if (type
== _MEMSWAP
)
4449 thresholds
= &memcg
->memsw_thresholds
;
4453 if (!thresholds
->primary
)
4456 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4458 /* Check if a threshold crossed before removing */
4459 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4461 /* Calculate new number of threshold */
4463 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
4464 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
4468 new = thresholds
->spare
;
4470 /* Set thresholds array to NULL if we don't have thresholds */
4479 /* Copy thresholds and find current threshold */
4480 new->current_threshold
= -1;
4481 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
4482 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
4485 new->entries
[j
] = thresholds
->primary
->entries
[i
];
4486 if (new->entries
[j
].threshold
<= usage
) {
4488 * new->current_threshold will not be used
4489 * until rcu_assign_pointer(), so it's safe to increment
4492 ++new->current_threshold
;
4498 /* Swap primary and spare array */
4499 thresholds
->spare
= thresholds
->primary
;
4500 /* If all events are unregistered, free the spare array */
4502 kfree(thresholds
->spare
);
4503 thresholds
->spare
= NULL
;
4506 rcu_assign_pointer(thresholds
->primary
, new);
4508 /* To be sure that nobody uses thresholds */
4511 mutex_unlock(&memcg
->thresholds_lock
);
4514 static int mem_cgroup_oom_register_event(struct cgroup
*cgrp
,
4515 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4517 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4518 struct mem_cgroup_eventfd_list
*event
;
4519 int type
= MEMFILE_TYPE(cft
->private);
4521 BUG_ON(type
!= _OOM_TYPE
);
4522 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
4526 spin_lock(&memcg_oom_lock
);
4528 event
->eventfd
= eventfd
;
4529 list_add(&event
->list
, &memcg
->oom_notify
);
4531 /* already in OOM ? */
4532 if (atomic_read(&memcg
->under_oom
))
4533 eventfd_signal(eventfd
, 1);
4534 spin_unlock(&memcg_oom_lock
);
4539 static void mem_cgroup_oom_unregister_event(struct cgroup
*cgrp
,
4540 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4542 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4543 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
4544 int type
= MEMFILE_TYPE(cft
->private);
4546 BUG_ON(type
!= _OOM_TYPE
);
4548 spin_lock(&memcg_oom_lock
);
4550 list_for_each_entry_safe(ev
, tmp
, &memcg
->oom_notify
, list
) {
4551 if (ev
->eventfd
== eventfd
) {
4552 list_del(&ev
->list
);
4557 spin_unlock(&memcg_oom_lock
);
4560 static int mem_cgroup_oom_control_read(struct cgroup
*cgrp
,
4561 struct cftype
*cft
, struct cgroup_map_cb
*cb
)
4563 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4565 cb
->fill(cb
, "oom_kill_disable", memcg
->oom_kill_disable
);
4567 if (atomic_read(&memcg
->under_oom
))
4568 cb
->fill(cb
, "under_oom", 1);
4570 cb
->fill(cb
, "under_oom", 0);
4574 static int mem_cgroup_oom_control_write(struct cgroup
*cgrp
,
4575 struct cftype
*cft
, u64 val
)
4577 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4578 struct mem_cgroup
*parent
;
4580 /* cannot set to root cgroup and only 0 and 1 are allowed */
4581 if (!cgrp
->parent
|| !((val
== 0) || (val
== 1)))
4584 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4587 /* oom-kill-disable is a flag for subhierarchy. */
4588 if ((parent
->use_hierarchy
) ||
4589 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4593 memcg
->oom_kill_disable
= val
;
4595 memcg_oom_recover(memcg
);
4600 #ifdef CONFIG_MEMCG_KMEM
4601 static int memcg_init_kmem(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
4603 return mem_cgroup_sockets_init(memcg
, ss
);
4606 static void kmem_cgroup_destroy(struct mem_cgroup
*memcg
)
4608 mem_cgroup_sockets_destroy(memcg
);
4611 static int memcg_init_kmem(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
4616 static void kmem_cgroup_destroy(struct mem_cgroup
*memcg
)
4621 static struct cftype mem_cgroup_files
[] = {
4623 .name
= "usage_in_bytes",
4624 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
4625 .read
= mem_cgroup_read
,
4626 .register_event
= mem_cgroup_usage_register_event
,
4627 .unregister_event
= mem_cgroup_usage_unregister_event
,
4630 .name
= "max_usage_in_bytes",
4631 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
4632 .trigger
= mem_cgroup_reset
,
4633 .read
= mem_cgroup_read
,
4636 .name
= "limit_in_bytes",
4637 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
4638 .write_string
= mem_cgroup_write
,
4639 .read
= mem_cgroup_read
,
4642 .name
= "soft_limit_in_bytes",
4643 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
4644 .write_string
= mem_cgroup_write
,
4645 .read
= mem_cgroup_read
,
4649 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
4650 .trigger
= mem_cgroup_reset
,
4651 .read
= mem_cgroup_read
,
4655 .read_seq_string
= memcg_stat_show
,
4658 .name
= "force_empty",
4659 .trigger
= mem_cgroup_force_empty_write
,
4662 .name
= "use_hierarchy",
4663 .write_u64
= mem_cgroup_hierarchy_write
,
4664 .read_u64
= mem_cgroup_hierarchy_read
,
4667 .name
= "swappiness",
4668 .read_u64
= mem_cgroup_swappiness_read
,
4669 .write_u64
= mem_cgroup_swappiness_write
,
4672 .name
= "move_charge_at_immigrate",
4673 .read_u64
= mem_cgroup_move_charge_read
,
4674 .write_u64
= mem_cgroup_move_charge_write
,
4677 .name
= "oom_control",
4678 .read_map
= mem_cgroup_oom_control_read
,
4679 .write_u64
= mem_cgroup_oom_control_write
,
4680 .register_event
= mem_cgroup_oom_register_event
,
4681 .unregister_event
= mem_cgroup_oom_unregister_event
,
4682 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
4686 .name
= "numa_stat",
4687 .read_seq_string
= memcg_numa_stat_show
,
4690 #ifdef CONFIG_MEMCG_SWAP
4692 .name
= "memsw.usage_in_bytes",
4693 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
4694 .read
= mem_cgroup_read
,
4695 .register_event
= mem_cgroup_usage_register_event
,
4696 .unregister_event
= mem_cgroup_usage_unregister_event
,
4699 .name
= "memsw.max_usage_in_bytes",
4700 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
4701 .trigger
= mem_cgroup_reset
,
4702 .read
= mem_cgroup_read
,
4705 .name
= "memsw.limit_in_bytes",
4706 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
4707 .write_string
= mem_cgroup_write
,
4708 .read
= mem_cgroup_read
,
4711 .name
= "memsw.failcnt",
4712 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
4713 .trigger
= mem_cgroup_reset
,
4714 .read
= mem_cgroup_read
,
4717 { }, /* terminate */
4720 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4722 struct mem_cgroup_per_node
*pn
;
4723 struct mem_cgroup_per_zone
*mz
;
4724 int zone
, tmp
= node
;
4726 * This routine is called against possible nodes.
4727 * But it's BUG to call kmalloc() against offline node.
4729 * TODO: this routine can waste much memory for nodes which will
4730 * never be onlined. It's better to use memory hotplug callback
4733 if (!node_state(node
, N_NORMAL_MEMORY
))
4735 pn
= kzalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
4739 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4740 mz
= &pn
->zoneinfo
[zone
];
4741 lruvec_init(&mz
->lruvec
, &NODE_DATA(node
)->node_zones
[zone
]);
4742 mz
->usage_in_excess
= 0;
4743 mz
->on_tree
= false;
4746 memcg
->info
.nodeinfo
[node
] = pn
;
4750 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4752 kfree(memcg
->info
.nodeinfo
[node
]);
4755 static struct mem_cgroup
*mem_cgroup_alloc(void)
4757 struct mem_cgroup
*memcg
;
4758 int size
= sizeof(struct mem_cgroup
);
4760 /* Can be very big if MAX_NUMNODES is very big */
4761 if (size
< PAGE_SIZE
)
4762 memcg
= kzalloc(size
, GFP_KERNEL
);
4764 memcg
= vzalloc(size
);
4769 memcg
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4772 spin_lock_init(&memcg
->pcp_counter_lock
);
4776 if (size
< PAGE_SIZE
)
4784 * Helpers for freeing a kmalloc()ed/vzalloc()ed mem_cgroup by RCU,
4785 * but in process context. The work_freeing structure is overlaid
4786 * on the rcu_freeing structure, which itself is overlaid on memsw.
4788 static void free_work(struct work_struct
*work
)
4790 struct mem_cgroup
*memcg
;
4791 int size
= sizeof(struct mem_cgroup
);
4793 memcg
= container_of(work
, struct mem_cgroup
, work_freeing
);
4795 * We need to make sure that (at least for now), the jump label
4796 * destruction code runs outside of the cgroup lock. This is because
4797 * get_online_cpus(), which is called from the static_branch update,
4798 * can't be called inside the cgroup_lock. cpusets are the ones
4799 * enforcing this dependency, so if they ever change, we might as well.
4801 * schedule_work() will guarantee this happens. Be careful if you need
4802 * to move this code around, and make sure it is outside
4805 disarm_sock_keys(memcg
);
4806 if (size
< PAGE_SIZE
)
4812 static void free_rcu(struct rcu_head
*rcu_head
)
4814 struct mem_cgroup
*memcg
;
4816 memcg
= container_of(rcu_head
, struct mem_cgroup
, rcu_freeing
);
4817 INIT_WORK(&memcg
->work_freeing
, free_work
);
4818 schedule_work(&memcg
->work_freeing
);
4822 * At destroying mem_cgroup, references from swap_cgroup can remain.
4823 * (scanning all at force_empty is too costly...)
4825 * Instead of clearing all references at force_empty, we remember
4826 * the number of reference from swap_cgroup and free mem_cgroup when
4827 * it goes down to 0.
4829 * Removal of cgroup itself succeeds regardless of refs from swap.
4832 static void __mem_cgroup_free(struct mem_cgroup
*memcg
)
4836 mem_cgroup_remove_from_trees(memcg
);
4837 free_css_id(&mem_cgroup_subsys
, &memcg
->css
);
4840 free_mem_cgroup_per_zone_info(memcg
, node
);
4842 free_percpu(memcg
->stat
);
4843 call_rcu(&memcg
->rcu_freeing
, free_rcu
);
4846 static void mem_cgroup_get(struct mem_cgroup
*memcg
)
4848 atomic_inc(&memcg
->refcnt
);
4851 static void __mem_cgroup_put(struct mem_cgroup
*memcg
, int count
)
4853 if (atomic_sub_and_test(count
, &memcg
->refcnt
)) {
4854 struct mem_cgroup
*parent
= parent_mem_cgroup(memcg
);
4855 __mem_cgroup_free(memcg
);
4857 mem_cgroup_put(parent
);
4861 static void mem_cgroup_put(struct mem_cgroup
*memcg
)
4863 __mem_cgroup_put(memcg
, 1);
4867 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4869 struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*memcg
)
4871 if (!memcg
->res
.parent
)
4873 return mem_cgroup_from_res_counter(memcg
->res
.parent
, res
);
4875 EXPORT_SYMBOL(parent_mem_cgroup
);
4877 #ifdef CONFIG_MEMCG_SWAP
4878 static void __init
enable_swap_cgroup(void)
4880 if (!mem_cgroup_disabled() && really_do_swap_account
)
4881 do_swap_account
= 1;
4884 static void __init
enable_swap_cgroup(void)
4889 static int mem_cgroup_soft_limit_tree_init(void)
4891 struct mem_cgroup_tree_per_node
*rtpn
;
4892 struct mem_cgroup_tree_per_zone
*rtpz
;
4893 int tmp
, node
, zone
;
4895 for_each_node(node
) {
4897 if (!node_state(node
, N_NORMAL_MEMORY
))
4899 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
, tmp
);
4903 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
4905 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4906 rtpz
= &rtpn
->rb_tree_per_zone
[zone
];
4907 rtpz
->rb_root
= RB_ROOT
;
4908 spin_lock_init(&rtpz
->lock
);
4914 for_each_node(node
) {
4915 if (!soft_limit_tree
.rb_tree_per_node
[node
])
4917 kfree(soft_limit_tree
.rb_tree_per_node
[node
]);
4918 soft_limit_tree
.rb_tree_per_node
[node
] = NULL
;
4924 static struct cgroup_subsys_state
* __ref
4925 mem_cgroup_css_alloc(struct cgroup
*cont
)
4927 struct mem_cgroup
*memcg
, *parent
;
4928 long error
= -ENOMEM
;
4931 memcg
= mem_cgroup_alloc();
4933 return ERR_PTR(error
);
4936 if (alloc_mem_cgroup_per_zone_info(memcg
, node
))
4940 if (cont
->parent
== NULL
) {
4942 enable_swap_cgroup();
4944 if (mem_cgroup_soft_limit_tree_init())
4946 root_mem_cgroup
= memcg
;
4947 for_each_possible_cpu(cpu
) {
4948 struct memcg_stock_pcp
*stock
=
4949 &per_cpu(memcg_stock
, cpu
);
4950 INIT_WORK(&stock
->work
, drain_local_stock
);
4952 hotcpu_notifier(memcg_cpu_hotplug_callback
, 0);
4954 parent
= mem_cgroup_from_cont(cont
->parent
);
4955 memcg
->use_hierarchy
= parent
->use_hierarchy
;
4956 memcg
->oom_kill_disable
= parent
->oom_kill_disable
;
4959 if (parent
&& parent
->use_hierarchy
) {
4960 res_counter_init(&memcg
->res
, &parent
->res
);
4961 res_counter_init(&memcg
->memsw
, &parent
->memsw
);
4963 * We increment refcnt of the parent to ensure that we can
4964 * safely access it on res_counter_charge/uncharge.
4965 * This refcnt will be decremented when freeing this
4966 * mem_cgroup(see mem_cgroup_put).
4968 mem_cgroup_get(parent
);
4970 res_counter_init(&memcg
->res
, NULL
);
4971 res_counter_init(&memcg
->memsw
, NULL
);
4973 * Deeper hierachy with use_hierarchy == false doesn't make
4974 * much sense so let cgroup subsystem know about this
4975 * unfortunate state in our controller.
4977 if (parent
&& parent
!= root_mem_cgroup
)
4978 mem_cgroup_subsys
.broken_hierarchy
= true;
4980 memcg
->last_scanned_node
= MAX_NUMNODES
;
4981 INIT_LIST_HEAD(&memcg
->oom_notify
);
4984 memcg
->swappiness
= mem_cgroup_swappiness(parent
);
4985 atomic_set(&memcg
->refcnt
, 1);
4986 memcg
->move_charge_at_immigrate
= 0;
4987 mutex_init(&memcg
->thresholds_lock
);
4988 spin_lock_init(&memcg
->move_lock
);
4990 error
= memcg_init_kmem(memcg
, &mem_cgroup_subsys
);
4993 * We call put now because our (and parent's) refcnts
4994 * are already in place. mem_cgroup_put() will internally
4995 * call __mem_cgroup_free, so return directly
4997 mem_cgroup_put(memcg
);
4998 return ERR_PTR(error
);
5002 __mem_cgroup_free(memcg
);
5003 return ERR_PTR(error
);
5006 static void mem_cgroup_css_offline(struct cgroup
*cont
)
5008 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
5010 mem_cgroup_reparent_charges(memcg
);
5013 static void mem_cgroup_css_free(struct cgroup
*cont
)
5015 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
5017 kmem_cgroup_destroy(memcg
);
5019 mem_cgroup_put(memcg
);
5023 /* Handlers for move charge at task migration. */
5024 #define PRECHARGE_COUNT_AT_ONCE 256
5025 static int mem_cgroup_do_precharge(unsigned long count
)
5028 int batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5029 struct mem_cgroup
*memcg
= mc
.to
;
5031 if (mem_cgroup_is_root(memcg
)) {
5032 mc
.precharge
+= count
;
5033 /* we don't need css_get for root */
5036 /* try to charge at once */
5038 struct res_counter
*dummy
;
5040 * "memcg" cannot be under rmdir() because we've already checked
5041 * by cgroup_lock_live_cgroup() that it is not removed and we
5042 * are still under the same cgroup_mutex. So we can postpone
5045 if (res_counter_charge(&memcg
->res
, PAGE_SIZE
* count
, &dummy
))
5047 if (do_swap_account
&& res_counter_charge(&memcg
->memsw
,
5048 PAGE_SIZE
* count
, &dummy
)) {
5049 res_counter_uncharge(&memcg
->res
, PAGE_SIZE
* count
);
5052 mc
.precharge
+= count
;
5056 /* fall back to one by one charge */
5058 if (signal_pending(current
)) {
5062 if (!batch_count
--) {
5063 batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5066 ret
= __mem_cgroup_try_charge(NULL
,
5067 GFP_KERNEL
, 1, &memcg
, false);
5069 /* mem_cgroup_clear_mc() will do uncharge later */
5077 * get_mctgt_type - get target type of moving charge
5078 * @vma: the vma the pte to be checked belongs
5079 * @addr: the address corresponding to the pte to be checked
5080 * @ptent: the pte to be checked
5081 * @target: the pointer the target page or swap ent will be stored(can be NULL)
5084 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5085 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5086 * move charge. if @target is not NULL, the page is stored in target->page
5087 * with extra refcnt got(Callers should handle it).
5088 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5089 * target for charge migration. if @target is not NULL, the entry is stored
5092 * Called with pte lock held.
5099 enum mc_target_type
{
5105 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
5106 unsigned long addr
, pte_t ptent
)
5108 struct page
*page
= vm_normal_page(vma
, addr
, ptent
);
5110 if (!page
|| !page_mapped(page
))
5112 if (PageAnon(page
)) {
5113 /* we don't move shared anon */
5116 } else if (!move_file())
5117 /* we ignore mapcount for file pages */
5119 if (!get_page_unless_zero(page
))
5126 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5127 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5129 struct page
*page
= NULL
;
5130 swp_entry_t ent
= pte_to_swp_entry(ptent
);
5132 if (!move_anon() || non_swap_entry(ent
))
5135 * Because lookup_swap_cache() updates some statistics counter,
5136 * we call find_get_page() with swapper_space directly.
5138 page
= find_get_page(&swapper_space
, ent
.val
);
5139 if (do_swap_account
)
5140 entry
->val
= ent
.val
;
5145 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5146 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5152 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
5153 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5155 struct page
*page
= NULL
;
5156 struct address_space
*mapping
;
5159 if (!vma
->vm_file
) /* anonymous vma */
5164 mapping
= vma
->vm_file
->f_mapping
;
5165 if (pte_none(ptent
))
5166 pgoff
= linear_page_index(vma
, addr
);
5167 else /* pte_file(ptent) is true */
5168 pgoff
= pte_to_pgoff(ptent
);
5170 /* page is moved even if it's not RSS of this task(page-faulted). */
5171 page
= find_get_page(mapping
, pgoff
);
5174 /* shmem/tmpfs may report page out on swap: account for that too. */
5175 if (radix_tree_exceptional_entry(page
)) {
5176 swp_entry_t swap
= radix_to_swp_entry(page
);
5177 if (do_swap_account
)
5179 page
= find_get_page(&swapper_space
, swap
.val
);
5185 static enum mc_target_type
get_mctgt_type(struct vm_area_struct
*vma
,
5186 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
5188 struct page
*page
= NULL
;
5189 struct page_cgroup
*pc
;
5190 enum mc_target_type ret
= MC_TARGET_NONE
;
5191 swp_entry_t ent
= { .val
= 0 };
5193 if (pte_present(ptent
))
5194 page
= mc_handle_present_pte(vma
, addr
, ptent
);
5195 else if (is_swap_pte(ptent
))
5196 page
= mc_handle_swap_pte(vma
, addr
, ptent
, &ent
);
5197 else if (pte_none(ptent
) || pte_file(ptent
))
5198 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
5200 if (!page
&& !ent
.val
)
5203 pc
= lookup_page_cgroup(page
);
5205 * Do only loose check w/o page_cgroup lock.
5206 * mem_cgroup_move_account() checks the pc is valid or not under
5209 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5210 ret
= MC_TARGET_PAGE
;
5212 target
->page
= page
;
5214 if (!ret
|| !target
)
5217 /* There is a swap entry and a page doesn't exist or isn't charged */
5218 if (ent
.val
&& !ret
&&
5219 css_id(&mc
.from
->css
) == lookup_swap_cgroup_id(ent
)) {
5220 ret
= MC_TARGET_SWAP
;
5227 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5229 * We don't consider swapping or file mapped pages because THP does not
5230 * support them for now.
5231 * Caller should make sure that pmd_trans_huge(pmd) is true.
5233 static enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
5234 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
5236 struct page
*page
= NULL
;
5237 struct page_cgroup
*pc
;
5238 enum mc_target_type ret
= MC_TARGET_NONE
;
5240 page
= pmd_page(pmd
);
5241 VM_BUG_ON(!page
|| !PageHead(page
));
5244 pc
= lookup_page_cgroup(page
);
5245 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5246 ret
= MC_TARGET_PAGE
;
5249 target
->page
= page
;
5255 static inline enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
5256 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
5258 return MC_TARGET_NONE
;
5262 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
5263 unsigned long addr
, unsigned long end
,
5264 struct mm_walk
*walk
)
5266 struct vm_area_struct
*vma
= walk
->private;
5270 if (pmd_trans_huge_lock(pmd
, vma
) == 1) {
5271 if (get_mctgt_type_thp(vma
, addr
, *pmd
, NULL
) == MC_TARGET_PAGE
)
5272 mc
.precharge
+= HPAGE_PMD_NR
;
5273 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5277 if (pmd_trans_unstable(pmd
))
5279 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5280 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
5281 if (get_mctgt_type(vma
, addr
, *pte
, NULL
))
5282 mc
.precharge
++; /* increment precharge temporarily */
5283 pte_unmap_unlock(pte
- 1, ptl
);
5289 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
5291 unsigned long precharge
;
5292 struct vm_area_struct
*vma
;
5294 down_read(&mm
->mmap_sem
);
5295 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5296 struct mm_walk mem_cgroup_count_precharge_walk
= {
5297 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
5301 if (is_vm_hugetlb_page(vma
))
5303 walk_page_range(vma
->vm_start
, vma
->vm_end
,
5304 &mem_cgroup_count_precharge_walk
);
5306 up_read(&mm
->mmap_sem
);
5308 precharge
= mc
.precharge
;
5314 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
5316 unsigned long precharge
= mem_cgroup_count_precharge(mm
);
5318 VM_BUG_ON(mc
.moving_task
);
5319 mc
.moving_task
= current
;
5320 return mem_cgroup_do_precharge(precharge
);
5323 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5324 static void __mem_cgroup_clear_mc(void)
5326 struct mem_cgroup
*from
= mc
.from
;
5327 struct mem_cgroup
*to
= mc
.to
;
5329 /* we must uncharge all the leftover precharges from mc.to */
5331 __mem_cgroup_cancel_charge(mc
.to
, mc
.precharge
);
5335 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5336 * we must uncharge here.
5338 if (mc
.moved_charge
) {
5339 __mem_cgroup_cancel_charge(mc
.from
, mc
.moved_charge
);
5340 mc
.moved_charge
= 0;
5342 /* we must fixup refcnts and charges */
5343 if (mc
.moved_swap
) {
5344 /* uncharge swap account from the old cgroup */
5345 if (!mem_cgroup_is_root(mc
.from
))
5346 res_counter_uncharge(&mc
.from
->memsw
,
5347 PAGE_SIZE
* mc
.moved_swap
);
5348 __mem_cgroup_put(mc
.from
, mc
.moved_swap
);
5350 if (!mem_cgroup_is_root(mc
.to
)) {
5352 * we charged both to->res and to->memsw, so we should
5355 res_counter_uncharge(&mc
.to
->res
,
5356 PAGE_SIZE
* mc
.moved_swap
);
5358 /* we've already done mem_cgroup_get(mc.to) */
5361 memcg_oom_recover(from
);
5362 memcg_oom_recover(to
);
5363 wake_up_all(&mc
.waitq
);
5366 static void mem_cgroup_clear_mc(void)
5368 struct mem_cgroup
*from
= mc
.from
;
5371 * we must clear moving_task before waking up waiters at the end of
5374 mc
.moving_task
= NULL
;
5375 __mem_cgroup_clear_mc();
5376 spin_lock(&mc
.lock
);
5379 spin_unlock(&mc
.lock
);
5380 mem_cgroup_end_move(from
);
5383 static int mem_cgroup_can_attach(struct cgroup
*cgroup
,
5384 struct cgroup_taskset
*tset
)
5386 struct task_struct
*p
= cgroup_taskset_first(tset
);
5388 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgroup
);
5390 if (memcg
->move_charge_at_immigrate
) {
5391 struct mm_struct
*mm
;
5392 struct mem_cgroup
*from
= mem_cgroup_from_task(p
);
5394 VM_BUG_ON(from
== memcg
);
5396 mm
= get_task_mm(p
);
5399 /* We move charges only when we move a owner of the mm */
5400 if (mm
->owner
== p
) {
5403 VM_BUG_ON(mc
.precharge
);
5404 VM_BUG_ON(mc
.moved_charge
);
5405 VM_BUG_ON(mc
.moved_swap
);
5406 mem_cgroup_start_move(from
);
5407 spin_lock(&mc
.lock
);
5410 spin_unlock(&mc
.lock
);
5411 /* We set mc.moving_task later */
5413 ret
= mem_cgroup_precharge_mc(mm
);
5415 mem_cgroup_clear_mc();
5422 static void mem_cgroup_cancel_attach(struct cgroup
*cgroup
,
5423 struct cgroup_taskset
*tset
)
5425 mem_cgroup_clear_mc();
5428 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
5429 unsigned long addr
, unsigned long end
,
5430 struct mm_walk
*walk
)
5433 struct vm_area_struct
*vma
= walk
->private;
5436 enum mc_target_type target_type
;
5437 union mc_target target
;
5439 struct page_cgroup
*pc
;
5442 * We don't take compound_lock() here but no race with splitting thp
5444 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
5445 * under splitting, which means there's no concurrent thp split,
5446 * - if another thread runs into split_huge_page() just after we
5447 * entered this if-block, the thread must wait for page table lock
5448 * to be unlocked in __split_huge_page_splitting(), where the main
5449 * part of thp split is not executed yet.
5451 if (pmd_trans_huge_lock(pmd
, vma
) == 1) {
5452 if (mc
.precharge
< HPAGE_PMD_NR
) {
5453 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5456 target_type
= get_mctgt_type_thp(vma
, addr
, *pmd
, &target
);
5457 if (target_type
== MC_TARGET_PAGE
) {
5459 if (!isolate_lru_page(page
)) {
5460 pc
= lookup_page_cgroup(page
);
5461 if (!mem_cgroup_move_account(page
, HPAGE_PMD_NR
,
5462 pc
, mc
.from
, mc
.to
)) {
5463 mc
.precharge
-= HPAGE_PMD_NR
;
5464 mc
.moved_charge
+= HPAGE_PMD_NR
;
5466 putback_lru_page(page
);
5470 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5474 if (pmd_trans_unstable(pmd
))
5477 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5478 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
5479 pte_t ptent
= *(pte
++);
5485 switch (get_mctgt_type(vma
, addr
, ptent
, &target
)) {
5486 case MC_TARGET_PAGE
:
5488 if (isolate_lru_page(page
))
5490 pc
= lookup_page_cgroup(page
);
5491 if (!mem_cgroup_move_account(page
, 1, pc
,
5494 /* we uncharge from mc.from later. */
5497 putback_lru_page(page
);
5498 put
: /* get_mctgt_type() gets the page */
5501 case MC_TARGET_SWAP
:
5503 if (!mem_cgroup_move_swap_account(ent
, mc
.from
, mc
.to
)) {
5505 /* we fixup refcnts and charges later. */
5513 pte_unmap_unlock(pte
- 1, ptl
);
5518 * We have consumed all precharges we got in can_attach().
5519 * We try charge one by one, but don't do any additional
5520 * charges to mc.to if we have failed in charge once in attach()
5523 ret
= mem_cgroup_do_precharge(1);
5531 static void mem_cgroup_move_charge(struct mm_struct
*mm
)
5533 struct vm_area_struct
*vma
;
5535 lru_add_drain_all();
5537 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
5539 * Someone who are holding the mmap_sem might be waiting in
5540 * waitq. So we cancel all extra charges, wake up all waiters,
5541 * and retry. Because we cancel precharges, we might not be able
5542 * to move enough charges, but moving charge is a best-effort
5543 * feature anyway, so it wouldn't be a big problem.
5545 __mem_cgroup_clear_mc();
5549 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5551 struct mm_walk mem_cgroup_move_charge_walk
= {
5552 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
5556 if (is_vm_hugetlb_page(vma
))
5558 ret
= walk_page_range(vma
->vm_start
, vma
->vm_end
,
5559 &mem_cgroup_move_charge_walk
);
5562 * means we have consumed all precharges and failed in
5563 * doing additional charge. Just abandon here.
5567 up_read(&mm
->mmap_sem
);
5570 static void mem_cgroup_move_task(struct cgroup
*cont
,
5571 struct cgroup_taskset
*tset
)
5573 struct task_struct
*p
= cgroup_taskset_first(tset
);
5574 struct mm_struct
*mm
= get_task_mm(p
);
5578 mem_cgroup_move_charge(mm
);
5582 mem_cgroup_clear_mc();
5584 #else /* !CONFIG_MMU */
5585 static int mem_cgroup_can_attach(struct cgroup
*cgroup
,
5586 struct cgroup_taskset
*tset
)
5590 static void mem_cgroup_cancel_attach(struct cgroup
*cgroup
,
5591 struct cgroup_taskset
*tset
)
5594 static void mem_cgroup_move_task(struct cgroup
*cont
,
5595 struct cgroup_taskset
*tset
)
5600 struct cgroup_subsys mem_cgroup_subsys
= {
5602 .subsys_id
= mem_cgroup_subsys_id
,
5603 .css_alloc
= mem_cgroup_css_alloc
,
5604 .css_offline
= mem_cgroup_css_offline
,
5605 .css_free
= mem_cgroup_css_free
,
5606 .can_attach
= mem_cgroup_can_attach
,
5607 .cancel_attach
= mem_cgroup_cancel_attach
,
5608 .attach
= mem_cgroup_move_task
,
5609 .base_cftypes
= mem_cgroup_files
,
5614 #ifdef CONFIG_MEMCG_SWAP
5615 static int __init
enable_swap_account(char *s
)
5617 /* consider enabled if no parameter or 1 is given */
5618 if (!strcmp(s
, "1"))
5619 really_do_swap_account
= 1;
5620 else if (!strcmp(s
, "0"))
5621 really_do_swap_account
= 0;
5624 __setup("swapaccount=", enable_swap_account
);