1 /* memcontrol.c - Memory Controller
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
6 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
13 * This program is free software; you can redistribute it and/or modify
14 * it under the terms of the GNU General Public License as published by
15 * the Free Software Foundation; either version 2 of the License, or
16 * (at your option) any later version.
18 * This program is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 * GNU General Public License for more details.
24 #include <linux/res_counter.h>
25 #include <linux/memcontrol.h>
26 #include <linux/cgroup.h>
28 #include <linux/hugetlb.h>
29 #include <linux/pagemap.h>
30 #include <linux/smp.h>
31 #include <linux/page-flags.h>
32 #include <linux/backing-dev.h>
33 #include <linux/bit_spinlock.h>
34 #include <linux/rcupdate.h>
35 #include <linux/limits.h>
36 #include <linux/export.h>
37 #include <linux/mutex.h>
38 #include <linux/rbtree.h>
39 #include <linux/slab.h>
40 #include <linux/swap.h>
41 #include <linux/swapops.h>
42 #include <linux/spinlock.h>
43 #include <linux/eventfd.h>
44 #include <linux/sort.h>
46 #include <linux/seq_file.h>
47 #include <linux/vmalloc.h>
48 #include <linux/mm_inline.h>
49 #include <linux/page_cgroup.h>
50 #include <linux/cpu.h>
51 #include <linux/oom.h>
54 #include <net/tcp_memcontrol.h>
56 #include <asm/uaccess.h>
58 #include <trace/events/vmscan.h>
60 struct cgroup_subsys mem_cgroup_subsys __read_mostly
;
61 #define MEM_CGROUP_RECLAIM_RETRIES 5
62 static struct mem_cgroup
*root_mem_cgroup __read_mostly
;
64 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
65 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
66 int do_swap_account __read_mostly
;
68 /* for remember boot option*/
69 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED
70 static int really_do_swap_account __initdata
= 1;
72 static int really_do_swap_account __initdata
= 0;
76 #define do_swap_account 0
81 * Statistics for memory cgroup.
83 enum mem_cgroup_stat_index
{
85 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
87 MEM_CGROUP_STAT_CACHE
, /* # of pages charged as cache */
88 MEM_CGROUP_STAT_RSS
, /* # of pages charged as anon rss */
89 MEM_CGROUP_STAT_FILE_MAPPED
, /* # of pages charged as file rss */
90 MEM_CGROUP_STAT_SWAP
, /* # of pages, swapped out */
91 MEM_CGROUP_STAT_NSTATS
,
94 static const char * const mem_cgroup_stat_names
[] = {
101 enum mem_cgroup_events_index
{
102 MEM_CGROUP_EVENTS_PGPGIN
, /* # of pages paged in */
103 MEM_CGROUP_EVENTS_PGPGOUT
, /* # of pages paged out */
104 MEM_CGROUP_EVENTS_PGFAULT
, /* # of page-faults */
105 MEM_CGROUP_EVENTS_PGMAJFAULT
, /* # of major page-faults */
106 MEM_CGROUP_EVENTS_NSTATS
,
109 static const char * const mem_cgroup_events_names
[] = {
117 * Per memcg event counter is incremented at every pagein/pageout. With THP,
118 * it will be incremated by the number of pages. This counter is used for
119 * for trigger some periodic events. This is straightforward and better
120 * than using jiffies etc. to handle periodic memcg event.
122 enum mem_cgroup_events_target
{
123 MEM_CGROUP_TARGET_THRESH
,
124 MEM_CGROUP_TARGET_SOFTLIMIT
,
125 MEM_CGROUP_TARGET_NUMAINFO
,
128 #define THRESHOLDS_EVENTS_TARGET 128
129 #define SOFTLIMIT_EVENTS_TARGET 1024
130 #define NUMAINFO_EVENTS_TARGET 1024
132 struct mem_cgroup_stat_cpu
{
133 long count
[MEM_CGROUP_STAT_NSTATS
];
134 unsigned long events
[MEM_CGROUP_EVENTS_NSTATS
];
135 unsigned long nr_page_events
;
136 unsigned long targets
[MEM_CGROUP_NTARGETS
];
139 struct mem_cgroup_reclaim_iter
{
140 /* css_id of the last scanned hierarchy member */
142 /* scan generation, increased every round-trip */
143 unsigned int generation
;
147 * per-zone information in memory controller.
149 struct mem_cgroup_per_zone
{
150 struct lruvec lruvec
;
151 unsigned long lru_size
[NR_LRU_LISTS
];
153 struct mem_cgroup_reclaim_iter reclaim_iter
[DEF_PRIORITY
+ 1];
155 struct rb_node tree_node
; /* RB tree node */
156 unsigned long long usage_in_excess
;/* Set to the value by which */
157 /* the soft limit is exceeded*/
159 struct mem_cgroup
*memcg
; /* Back pointer, we cannot */
160 /* use container_of */
163 struct mem_cgroup_per_node
{
164 struct mem_cgroup_per_zone zoneinfo
[MAX_NR_ZONES
];
167 struct mem_cgroup_lru_info
{
168 struct mem_cgroup_per_node
*nodeinfo
[MAX_NUMNODES
];
172 * Cgroups above their limits are maintained in a RB-Tree, independent of
173 * their hierarchy representation
176 struct mem_cgroup_tree_per_zone
{
177 struct rb_root rb_root
;
181 struct mem_cgroup_tree_per_node
{
182 struct mem_cgroup_tree_per_zone rb_tree_per_zone
[MAX_NR_ZONES
];
185 struct mem_cgroup_tree
{
186 struct mem_cgroup_tree_per_node
*rb_tree_per_node
[MAX_NUMNODES
];
189 static struct mem_cgroup_tree soft_limit_tree __read_mostly
;
191 struct mem_cgroup_threshold
{
192 struct eventfd_ctx
*eventfd
;
197 struct mem_cgroup_threshold_ary
{
198 /* An array index points to threshold just below or equal to usage. */
199 int current_threshold
;
200 /* Size of entries[] */
202 /* Array of thresholds */
203 struct mem_cgroup_threshold entries
[0];
206 struct mem_cgroup_thresholds
{
207 /* Primary thresholds array */
208 struct mem_cgroup_threshold_ary
*primary
;
210 * Spare threshold array.
211 * This is needed to make mem_cgroup_unregister_event() "never fail".
212 * It must be able to store at least primary->size - 1 entries.
214 struct mem_cgroup_threshold_ary
*spare
;
218 struct mem_cgroup_eventfd_list
{
219 struct list_head list
;
220 struct eventfd_ctx
*eventfd
;
223 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
);
224 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
);
227 * The memory controller data structure. The memory controller controls both
228 * page cache and RSS per cgroup. We would eventually like to provide
229 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
230 * to help the administrator determine what knobs to tune.
232 * TODO: Add a water mark for the memory controller. Reclaim will begin when
233 * we hit the water mark. May be even add a low water mark, such that
234 * no reclaim occurs from a cgroup at it's low water mark, this is
235 * a feature that will be implemented much later in the future.
238 struct cgroup_subsys_state css
;
240 * the counter to account for memory usage
242 struct res_counter res
;
246 * the counter to account for mem+swap usage.
248 struct res_counter memsw
;
251 * rcu_freeing is used only when freeing struct mem_cgroup,
252 * so put it into a union to avoid wasting more memory.
253 * It must be disjoint from the css field. It could be
254 * in a union with the res field, but res plays a much
255 * larger part in mem_cgroup life than memsw, and might
256 * be of interest, even at time of free, when debugging.
257 * So share rcu_head with the less interesting memsw.
259 struct rcu_head rcu_freeing
;
261 * We also need some space for a worker in deferred freeing.
262 * By the time we call it, rcu_freeing is no longer in use.
264 struct work_struct work_freeing
;
268 * Per cgroup active and inactive list, similar to the
269 * per zone LRU lists.
271 struct mem_cgroup_lru_info info
;
272 int last_scanned_node
;
274 nodemask_t scan_nodes
;
275 atomic_t numainfo_events
;
276 atomic_t numainfo_updating
;
279 * Should the accounting and control be hierarchical, per subtree?
289 /* OOM-Killer disable */
290 int oom_kill_disable
;
292 /* set when res.limit == memsw.limit */
293 bool memsw_is_minimum
;
295 /* protect arrays of thresholds */
296 struct mutex thresholds_lock
;
298 /* thresholds for memory usage. RCU-protected */
299 struct mem_cgroup_thresholds thresholds
;
301 /* thresholds for mem+swap usage. RCU-protected */
302 struct mem_cgroup_thresholds memsw_thresholds
;
304 /* For oom notifier event fd */
305 struct list_head oom_notify
;
308 * Should we move charges of a task when a task is moved into this
309 * mem_cgroup ? And what type of charges should we move ?
311 unsigned long move_charge_at_immigrate
;
313 * set > 0 if pages under this cgroup are moving to other cgroup.
315 atomic_t moving_account
;
316 /* taken only while moving_account > 0 */
317 spinlock_t move_lock
;
321 struct mem_cgroup_stat_cpu __percpu
*stat
;
323 * used when a cpu is offlined or other synchronizations
324 * See mem_cgroup_read_stat().
326 struct mem_cgroup_stat_cpu nocpu_base
;
327 spinlock_t pcp_counter_lock
;
330 struct tcp_memcontrol tcp_mem
;
334 /* Stuffs for move charges at task migration. */
336 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
337 * left-shifted bitmap of these types.
340 MOVE_CHARGE_TYPE_ANON
, /* private anonymous page and swap of it */
341 MOVE_CHARGE_TYPE_FILE
, /* file page(including tmpfs) and swap of it */
345 /* "mc" and its members are protected by cgroup_mutex */
346 static struct move_charge_struct
{
347 spinlock_t lock
; /* for from, to */
348 struct mem_cgroup
*from
;
349 struct mem_cgroup
*to
;
350 unsigned long precharge
;
351 unsigned long moved_charge
;
352 unsigned long moved_swap
;
353 struct task_struct
*moving_task
; /* a task moving charges */
354 wait_queue_head_t waitq
; /* a waitq for other context */
356 .lock
= __SPIN_LOCK_UNLOCKED(mc
.lock
),
357 .waitq
= __WAIT_QUEUE_HEAD_INITIALIZER(mc
.waitq
),
360 static bool move_anon(void)
362 return test_bit(MOVE_CHARGE_TYPE_ANON
,
363 &mc
.to
->move_charge_at_immigrate
);
366 static bool move_file(void)
368 return test_bit(MOVE_CHARGE_TYPE_FILE
,
369 &mc
.to
->move_charge_at_immigrate
);
373 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
374 * limit reclaim to prevent infinite loops, if they ever occur.
376 #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
377 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
380 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
381 MEM_CGROUP_CHARGE_TYPE_ANON
,
382 MEM_CGROUP_CHARGE_TYPE_SHMEM
, /* used by page migration of shmem */
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
);
409 /* Writing them here to avoid exposing memcg's inner layout */
410 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
411 #include <net/sock.h>
414 static bool mem_cgroup_is_root(struct mem_cgroup
*memcg
);
415 void sock_update_memcg(struct sock
*sk
)
417 if (mem_cgroup_sockets_enabled
) {
418 struct mem_cgroup
*memcg
;
419 struct cg_proto
*cg_proto
;
421 BUG_ON(!sk
->sk_prot
->proto_cgroup
);
423 /* Socket cloning can throw us here with sk_cgrp already
424 * filled. It won't however, necessarily happen from
425 * process context. So the test for root memcg given
426 * the current task's memcg won't help us in this case.
428 * Respecting the original socket's memcg is a better
429 * decision in this case.
432 BUG_ON(mem_cgroup_is_root(sk
->sk_cgrp
->memcg
));
433 mem_cgroup_get(sk
->sk_cgrp
->memcg
);
438 memcg
= mem_cgroup_from_task(current
);
439 cg_proto
= sk
->sk_prot
->proto_cgroup(memcg
);
440 if (!mem_cgroup_is_root(memcg
) && memcg_proto_active(cg_proto
)) {
441 mem_cgroup_get(memcg
);
442 sk
->sk_cgrp
= cg_proto
;
447 EXPORT_SYMBOL(sock_update_memcg
);
449 void sock_release_memcg(struct sock
*sk
)
451 if (mem_cgroup_sockets_enabled
&& sk
->sk_cgrp
) {
452 struct mem_cgroup
*memcg
;
453 WARN_ON(!sk
->sk_cgrp
->memcg
);
454 memcg
= sk
->sk_cgrp
->memcg
;
455 mem_cgroup_put(memcg
);
460 struct cg_proto
*tcp_proto_cgroup(struct mem_cgroup
*memcg
)
462 if (!memcg
|| mem_cgroup_is_root(memcg
))
465 return &memcg
->tcp_mem
.cg_proto
;
467 EXPORT_SYMBOL(tcp_proto_cgroup
);
468 #endif /* CONFIG_INET */
469 #endif /* CONFIG_CGROUP_MEM_RES_CTLR_KMEM */
471 #if defined(CONFIG_INET) && defined(CONFIG_CGROUP_MEM_RES_CTLR_KMEM)
472 static void disarm_sock_keys(struct mem_cgroup
*memcg
)
474 if (!memcg_proto_activated(&memcg
->tcp_mem
.cg_proto
))
476 static_key_slow_dec(&memcg_socket_limit_enabled
);
479 static void disarm_sock_keys(struct mem_cgroup
*memcg
)
484 static void drain_all_stock_async(struct mem_cgroup
*memcg
);
486 static struct mem_cgroup_per_zone
*
487 mem_cgroup_zoneinfo(struct mem_cgroup
*memcg
, int nid
, int zid
)
489 return &memcg
->info
.nodeinfo
[nid
]->zoneinfo
[zid
];
492 struct cgroup_subsys_state
*mem_cgroup_css(struct mem_cgroup
*memcg
)
497 static struct mem_cgroup_per_zone
*
498 page_cgroup_zoneinfo(struct mem_cgroup
*memcg
, struct page
*page
)
500 int nid
= page_to_nid(page
);
501 int zid
= page_zonenum(page
);
503 return mem_cgroup_zoneinfo(memcg
, nid
, zid
);
506 static struct mem_cgroup_tree_per_zone
*
507 soft_limit_tree_node_zone(int nid
, int zid
)
509 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
512 static struct mem_cgroup_tree_per_zone
*
513 soft_limit_tree_from_page(struct page
*page
)
515 int nid
= page_to_nid(page
);
516 int zid
= page_zonenum(page
);
518 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
522 __mem_cgroup_insert_exceeded(struct mem_cgroup
*memcg
,
523 struct mem_cgroup_per_zone
*mz
,
524 struct mem_cgroup_tree_per_zone
*mctz
,
525 unsigned long long new_usage_in_excess
)
527 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
528 struct rb_node
*parent
= NULL
;
529 struct mem_cgroup_per_zone
*mz_node
;
534 mz
->usage_in_excess
= new_usage_in_excess
;
535 if (!mz
->usage_in_excess
)
539 mz_node
= rb_entry(parent
, struct mem_cgroup_per_zone
,
541 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
)
544 * We can't avoid mem cgroups that are over their soft
545 * limit by the same amount
547 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
550 rb_link_node(&mz
->tree_node
, parent
, p
);
551 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
556 __mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
557 struct mem_cgroup_per_zone
*mz
,
558 struct mem_cgroup_tree_per_zone
*mctz
)
562 rb_erase(&mz
->tree_node
, &mctz
->rb_root
);
567 mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
568 struct mem_cgroup_per_zone
*mz
,
569 struct mem_cgroup_tree_per_zone
*mctz
)
571 spin_lock(&mctz
->lock
);
572 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
573 spin_unlock(&mctz
->lock
);
577 static void mem_cgroup_update_tree(struct mem_cgroup
*memcg
, struct page
*page
)
579 unsigned long long excess
;
580 struct mem_cgroup_per_zone
*mz
;
581 struct mem_cgroup_tree_per_zone
*mctz
;
582 int nid
= page_to_nid(page
);
583 int zid
= page_zonenum(page
);
584 mctz
= soft_limit_tree_from_page(page
);
587 * Necessary to update all ancestors when hierarchy is used.
588 * because their event counter is not touched.
590 for (; memcg
; memcg
= parent_mem_cgroup(memcg
)) {
591 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
592 excess
= res_counter_soft_limit_excess(&memcg
->res
);
594 * We have to update the tree if mz is on RB-tree or
595 * mem is over its softlimit.
597 if (excess
|| mz
->on_tree
) {
598 spin_lock(&mctz
->lock
);
599 /* if on-tree, remove it */
601 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
603 * Insert again. mz->usage_in_excess will be updated.
604 * If excess is 0, no tree ops.
606 __mem_cgroup_insert_exceeded(memcg
, mz
, mctz
, excess
);
607 spin_unlock(&mctz
->lock
);
612 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*memcg
)
615 struct mem_cgroup_per_zone
*mz
;
616 struct mem_cgroup_tree_per_zone
*mctz
;
618 for_each_node(node
) {
619 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
620 mz
= mem_cgroup_zoneinfo(memcg
, node
, zone
);
621 mctz
= soft_limit_tree_node_zone(node
, zone
);
622 mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
627 static struct mem_cgroup_per_zone
*
628 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
630 struct rb_node
*rightmost
= NULL
;
631 struct mem_cgroup_per_zone
*mz
;
635 rightmost
= rb_last(&mctz
->rb_root
);
637 goto done
; /* Nothing to reclaim from */
639 mz
= rb_entry(rightmost
, struct mem_cgroup_per_zone
, tree_node
);
641 * Remove the node now but someone else can add it back,
642 * we will to add it back at the end of reclaim to its correct
643 * position in the tree.
645 __mem_cgroup_remove_exceeded(mz
->memcg
, mz
, mctz
);
646 if (!res_counter_soft_limit_excess(&mz
->memcg
->res
) ||
647 !css_tryget(&mz
->memcg
->css
))
653 static struct mem_cgroup_per_zone
*
654 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
656 struct mem_cgroup_per_zone
*mz
;
658 spin_lock(&mctz
->lock
);
659 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
660 spin_unlock(&mctz
->lock
);
665 * Implementation Note: reading percpu statistics for memcg.
667 * Both of vmstat[] and percpu_counter has threshold and do periodic
668 * synchronization to implement "quick" read. There are trade-off between
669 * reading cost and precision of value. Then, we may have a chance to implement
670 * a periodic synchronizion of counter in memcg's counter.
672 * But this _read() function is used for user interface now. The user accounts
673 * memory usage by memory cgroup and he _always_ requires exact value because
674 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
675 * have to visit all online cpus and make sum. So, for now, unnecessary
676 * synchronization is not implemented. (just implemented for cpu hotplug)
678 * If there are kernel internal actions which can make use of some not-exact
679 * value, and reading all cpu value can be performance bottleneck in some
680 * common workload, threashold and synchonization as vmstat[] should be
683 static long mem_cgroup_read_stat(struct mem_cgroup
*memcg
,
684 enum mem_cgroup_stat_index idx
)
690 for_each_online_cpu(cpu
)
691 val
+= per_cpu(memcg
->stat
->count
[idx
], cpu
);
692 #ifdef CONFIG_HOTPLUG_CPU
693 spin_lock(&memcg
->pcp_counter_lock
);
694 val
+= memcg
->nocpu_base
.count
[idx
];
695 spin_unlock(&memcg
->pcp_counter_lock
);
701 static void mem_cgroup_swap_statistics(struct mem_cgroup
*memcg
,
704 int val
= (charge
) ? 1 : -1;
705 this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_SWAP
], val
);
708 static unsigned long mem_cgroup_read_events(struct mem_cgroup
*memcg
,
709 enum mem_cgroup_events_index idx
)
711 unsigned long val
= 0;
714 for_each_online_cpu(cpu
)
715 val
+= per_cpu(memcg
->stat
->events
[idx
], cpu
);
716 #ifdef CONFIG_HOTPLUG_CPU
717 spin_lock(&memcg
->pcp_counter_lock
);
718 val
+= memcg
->nocpu_base
.events
[idx
];
719 spin_unlock(&memcg
->pcp_counter_lock
);
724 static void mem_cgroup_charge_statistics(struct mem_cgroup
*memcg
,
725 bool anon
, int nr_pages
)
730 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
731 * counted as CACHE even if it's on ANON LRU.
734 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_RSS
],
737 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_CACHE
],
740 /* pagein of a big page is an event. So, ignore page size */
742 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGIN
]);
744 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGOUT
]);
745 nr_pages
= -nr_pages
; /* for event */
748 __this_cpu_add(memcg
->stat
->nr_page_events
, nr_pages
);
754 mem_cgroup_get_lru_size(struct lruvec
*lruvec
, enum lru_list lru
)
756 struct mem_cgroup_per_zone
*mz
;
758 mz
= container_of(lruvec
, struct mem_cgroup_per_zone
, lruvec
);
759 return mz
->lru_size
[lru
];
763 mem_cgroup_zone_nr_lru_pages(struct mem_cgroup
*memcg
, int nid
, int zid
,
764 unsigned int lru_mask
)
766 struct mem_cgroup_per_zone
*mz
;
768 unsigned long ret
= 0;
770 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
773 if (BIT(lru
) & lru_mask
)
774 ret
+= mz
->lru_size
[lru
];
780 mem_cgroup_node_nr_lru_pages(struct mem_cgroup
*memcg
,
781 int nid
, unsigned int lru_mask
)
786 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++)
787 total
+= mem_cgroup_zone_nr_lru_pages(memcg
,
793 static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup
*memcg
,
794 unsigned int lru_mask
)
799 for_each_node_state(nid
, N_HIGH_MEMORY
)
800 total
+= mem_cgroup_node_nr_lru_pages(memcg
, nid
, lru_mask
);
804 static bool mem_cgroup_event_ratelimit(struct mem_cgroup
*memcg
,
805 enum mem_cgroup_events_target target
)
807 unsigned long val
, next
;
809 val
= __this_cpu_read(memcg
->stat
->nr_page_events
);
810 next
= __this_cpu_read(memcg
->stat
->targets
[target
]);
811 /* from time_after() in jiffies.h */
812 if ((long)next
- (long)val
< 0) {
814 case MEM_CGROUP_TARGET_THRESH
:
815 next
= val
+ THRESHOLDS_EVENTS_TARGET
;
817 case MEM_CGROUP_TARGET_SOFTLIMIT
:
818 next
= val
+ SOFTLIMIT_EVENTS_TARGET
;
820 case MEM_CGROUP_TARGET_NUMAINFO
:
821 next
= val
+ NUMAINFO_EVENTS_TARGET
;
826 __this_cpu_write(memcg
->stat
->targets
[target
], next
);
833 * Check events in order.
836 static void memcg_check_events(struct mem_cgroup
*memcg
, struct page
*page
)
839 /* threshold event is triggered in finer grain than soft limit */
840 if (unlikely(mem_cgroup_event_ratelimit(memcg
,
841 MEM_CGROUP_TARGET_THRESH
))) {
843 bool do_numainfo __maybe_unused
;
845 do_softlimit
= mem_cgroup_event_ratelimit(memcg
,
846 MEM_CGROUP_TARGET_SOFTLIMIT
);
848 do_numainfo
= mem_cgroup_event_ratelimit(memcg
,
849 MEM_CGROUP_TARGET_NUMAINFO
);
853 mem_cgroup_threshold(memcg
);
854 if (unlikely(do_softlimit
))
855 mem_cgroup_update_tree(memcg
, page
);
857 if (unlikely(do_numainfo
))
858 atomic_inc(&memcg
->numainfo_events
);
864 struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
866 return container_of(cgroup_subsys_state(cont
,
867 mem_cgroup_subsys_id
), struct mem_cgroup
,
871 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
874 * mm_update_next_owner() may clear mm->owner to NULL
875 * if it races with swapoff, page migration, etc.
876 * So this can be called with p == NULL.
881 return container_of(task_subsys_state(p
, mem_cgroup_subsys_id
),
882 struct mem_cgroup
, css
);
885 struct mem_cgroup
*try_get_mem_cgroup_from_mm(struct mm_struct
*mm
)
887 struct mem_cgroup
*memcg
= NULL
;
892 * Because we have no locks, mm->owner's may be being moved to other
893 * cgroup. We use css_tryget() here even if this looks
894 * pessimistic (rather than adding locks here).
898 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
899 if (unlikely(!memcg
))
901 } while (!css_tryget(&memcg
->css
));
907 * mem_cgroup_iter - iterate over memory cgroup hierarchy
908 * @root: hierarchy root
909 * @prev: previously returned memcg, NULL on first invocation
910 * @reclaim: cookie for shared reclaim walks, NULL for full walks
912 * Returns references to children of the hierarchy below @root, or
913 * @root itself, or %NULL after a full round-trip.
915 * Caller must pass the return value in @prev on subsequent
916 * invocations for reference counting, or use mem_cgroup_iter_break()
917 * to cancel a hierarchy walk before the round-trip is complete.
919 * Reclaimers can specify a zone and a priority level in @reclaim to
920 * divide up the memcgs in the hierarchy among all concurrent
921 * reclaimers operating on the same zone and priority.
923 struct mem_cgroup
*mem_cgroup_iter(struct mem_cgroup
*root
,
924 struct mem_cgroup
*prev
,
925 struct mem_cgroup_reclaim_cookie
*reclaim
)
927 struct mem_cgroup
*memcg
= NULL
;
930 if (mem_cgroup_disabled())
934 root
= root_mem_cgroup
;
936 if (prev
&& !reclaim
)
937 id
= css_id(&prev
->css
);
939 if (prev
&& prev
!= root
)
942 if (!root
->use_hierarchy
&& root
!= root_mem_cgroup
) {
949 struct mem_cgroup_reclaim_iter
*uninitialized_var(iter
);
950 struct cgroup_subsys_state
*css
;
953 int nid
= zone_to_nid(reclaim
->zone
);
954 int zid
= zone_idx(reclaim
->zone
);
955 struct mem_cgroup_per_zone
*mz
;
957 mz
= mem_cgroup_zoneinfo(root
, nid
, zid
);
958 iter
= &mz
->reclaim_iter
[reclaim
->priority
];
959 if (prev
&& reclaim
->generation
!= iter
->generation
)
965 css
= css_get_next(&mem_cgroup_subsys
, id
+ 1, &root
->css
, &id
);
967 if (css
== &root
->css
|| css_tryget(css
))
968 memcg
= container_of(css
,
969 struct mem_cgroup
, css
);
978 else if (!prev
&& memcg
)
979 reclaim
->generation
= iter
->generation
;
989 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
990 * @root: hierarchy root
991 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
993 void mem_cgroup_iter_break(struct mem_cgroup
*root
,
994 struct mem_cgroup
*prev
)
997 root
= root_mem_cgroup
;
998 if (prev
&& prev
!= root
)
1003 * Iteration constructs for visiting all cgroups (under a tree). If
1004 * loops are exited prematurely (break), mem_cgroup_iter_break() must
1005 * be used for reference counting.
1007 #define for_each_mem_cgroup_tree(iter, root) \
1008 for (iter = mem_cgroup_iter(root, NULL, NULL); \
1010 iter = mem_cgroup_iter(root, iter, NULL))
1012 #define for_each_mem_cgroup(iter) \
1013 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
1015 iter = mem_cgroup_iter(NULL, iter, NULL))
1017 static inline bool mem_cgroup_is_root(struct mem_cgroup
*memcg
)
1019 return (memcg
== root_mem_cgroup
);
1022 void mem_cgroup_count_vm_event(struct mm_struct
*mm
, enum vm_event_item idx
)
1024 struct mem_cgroup
*memcg
;
1030 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
1031 if (unlikely(!memcg
))
1036 this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGFAULT
]);
1039 this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGMAJFAULT
]);
1047 EXPORT_SYMBOL(mem_cgroup_count_vm_event
);
1050 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1051 * @zone: zone of the wanted lruvec
1052 * @memcg: memcg of the wanted lruvec
1054 * Returns the lru list vector holding pages for the given @zone and
1055 * @mem. This can be the global zone lruvec, if the memory controller
1058 struct lruvec
*mem_cgroup_zone_lruvec(struct zone
*zone
,
1059 struct mem_cgroup
*memcg
)
1061 struct mem_cgroup_per_zone
*mz
;
1063 if (mem_cgroup_disabled())
1064 return &zone
->lruvec
;
1066 mz
= mem_cgroup_zoneinfo(memcg
, zone_to_nid(zone
), zone_idx(zone
));
1071 * Following LRU functions are allowed to be used without PCG_LOCK.
1072 * Operations are called by routine of global LRU independently from memcg.
1073 * What we have to take care of here is validness of pc->mem_cgroup.
1075 * Changes to pc->mem_cgroup happens when
1078 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1079 * It is added to LRU before charge.
1080 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1081 * When moving account, the page is not on LRU. It's isolated.
1085 * mem_cgroup_page_lruvec - return lruvec for adding an lru page
1087 * @zone: zone of the page
1089 struct lruvec
*mem_cgroup_page_lruvec(struct page
*page
, struct zone
*zone
)
1091 struct mem_cgroup_per_zone
*mz
;
1092 struct mem_cgroup
*memcg
;
1093 struct page_cgroup
*pc
;
1095 if (mem_cgroup_disabled())
1096 return &zone
->lruvec
;
1098 pc
= lookup_page_cgroup(page
);
1099 memcg
= pc
->mem_cgroup
;
1102 * Surreptitiously switch any uncharged offlist page to root:
1103 * an uncharged page off lru does nothing to secure
1104 * its former mem_cgroup from sudden removal.
1106 * Our caller holds lru_lock, and PageCgroupUsed is updated
1107 * under page_cgroup lock: between them, they make all uses
1108 * of pc->mem_cgroup safe.
1110 if (!PageLRU(page
) && !PageCgroupUsed(pc
) && memcg
!= root_mem_cgroup
)
1111 pc
->mem_cgroup
= memcg
= root_mem_cgroup
;
1113 mz
= page_cgroup_zoneinfo(memcg
, page
);
1118 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1119 * @lruvec: mem_cgroup per zone lru vector
1120 * @lru: index of lru list the page is sitting on
1121 * @nr_pages: positive when adding or negative when removing
1123 * This function must be called when a page is added to or removed from an
1126 void mem_cgroup_update_lru_size(struct lruvec
*lruvec
, enum lru_list lru
,
1129 struct mem_cgroup_per_zone
*mz
;
1130 unsigned long *lru_size
;
1132 if (mem_cgroup_disabled())
1135 mz
= container_of(lruvec
, struct mem_cgroup_per_zone
, lruvec
);
1136 lru_size
= mz
->lru_size
+ lru
;
1137 *lru_size
+= nr_pages
;
1138 VM_BUG_ON((long)(*lru_size
) < 0);
1142 * Checks whether given mem is same or in the root_mem_cgroup's
1145 bool __mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_memcg
,
1146 struct mem_cgroup
*memcg
)
1148 if (root_memcg
== memcg
)
1150 if (!root_memcg
->use_hierarchy
|| !memcg
)
1152 return css_is_ancestor(&memcg
->css
, &root_memcg
->css
);
1155 static bool mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_memcg
,
1156 struct mem_cgroup
*memcg
)
1161 ret
= __mem_cgroup_same_or_subtree(root_memcg
, memcg
);
1166 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*memcg
)
1169 struct mem_cgroup
*curr
= NULL
;
1170 struct task_struct
*p
;
1172 p
= find_lock_task_mm(task
);
1174 curr
= try_get_mem_cgroup_from_mm(p
->mm
);
1178 * All threads may have already detached their mm's, but the oom
1179 * killer still needs to detect if they have already been oom
1180 * killed to prevent needlessly killing additional tasks.
1183 curr
= mem_cgroup_from_task(task
);
1185 css_get(&curr
->css
);
1191 * We should check use_hierarchy of "memcg" not "curr". Because checking
1192 * use_hierarchy of "curr" here make this function true if hierarchy is
1193 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1194 * hierarchy(even if use_hierarchy is disabled in "memcg").
1196 ret
= mem_cgroup_same_or_subtree(memcg
, curr
);
1197 css_put(&curr
->css
);
1201 int mem_cgroup_inactive_anon_is_low(struct lruvec
*lruvec
)
1203 unsigned long inactive_ratio
;
1204 unsigned long inactive
;
1205 unsigned long active
;
1208 inactive
= mem_cgroup_get_lru_size(lruvec
, LRU_INACTIVE_ANON
);
1209 active
= mem_cgroup_get_lru_size(lruvec
, LRU_ACTIVE_ANON
);
1211 gb
= (inactive
+ active
) >> (30 - PAGE_SHIFT
);
1213 inactive_ratio
= int_sqrt(10 * gb
);
1217 return inactive
* inactive_ratio
< active
;
1220 int mem_cgroup_inactive_file_is_low(struct lruvec
*lruvec
)
1222 unsigned long active
;
1223 unsigned long inactive
;
1225 inactive
= mem_cgroup_get_lru_size(lruvec
, LRU_INACTIVE_FILE
);
1226 active
= mem_cgroup_get_lru_size(lruvec
, LRU_ACTIVE_FILE
);
1228 return (active
> inactive
);
1231 #define mem_cgroup_from_res_counter(counter, member) \
1232 container_of(counter, struct mem_cgroup, member)
1235 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1236 * @memcg: the memory cgroup
1238 * Returns the maximum amount of memory @mem can be charged with, in
1241 static unsigned long mem_cgroup_margin(struct mem_cgroup
*memcg
)
1243 unsigned long long margin
;
1245 margin
= res_counter_margin(&memcg
->res
);
1246 if (do_swap_account
)
1247 margin
= min(margin
, res_counter_margin(&memcg
->memsw
));
1248 return margin
>> PAGE_SHIFT
;
1251 int mem_cgroup_swappiness(struct mem_cgroup
*memcg
)
1253 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
1256 if (cgrp
->parent
== NULL
)
1257 return vm_swappiness
;
1259 return memcg
->swappiness
;
1263 * memcg->moving_account is used for checking possibility that some thread is
1264 * calling move_account(). When a thread on CPU-A starts moving pages under
1265 * a memcg, other threads should check memcg->moving_account under
1266 * rcu_read_lock(), like this:
1270 * memcg->moving_account+1 if (memcg->mocing_account)
1272 * synchronize_rcu() update something.
1277 /* for quick checking without looking up memcg */
1278 atomic_t memcg_moving __read_mostly
;
1280 static void mem_cgroup_start_move(struct mem_cgroup
*memcg
)
1282 atomic_inc(&memcg_moving
);
1283 atomic_inc(&memcg
->moving_account
);
1287 static void mem_cgroup_end_move(struct mem_cgroup
*memcg
)
1290 * Now, mem_cgroup_clear_mc() may call this function with NULL.
1291 * We check NULL in callee rather than caller.
1294 atomic_dec(&memcg_moving
);
1295 atomic_dec(&memcg
->moving_account
);
1300 * 2 routines for checking "mem" is under move_account() or not.
1302 * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This
1303 * is used for avoiding races in accounting. If true,
1304 * pc->mem_cgroup may be overwritten.
1306 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1307 * under hierarchy of moving cgroups. This is for
1308 * waiting at hith-memory prressure caused by "move".
1311 static bool mem_cgroup_stolen(struct mem_cgroup
*memcg
)
1313 VM_BUG_ON(!rcu_read_lock_held());
1314 return atomic_read(&memcg
->moving_account
) > 0;
1317 static bool mem_cgroup_under_move(struct mem_cgroup
*memcg
)
1319 struct mem_cgroup
*from
;
1320 struct mem_cgroup
*to
;
1323 * Unlike task_move routines, we access mc.to, mc.from not under
1324 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1326 spin_lock(&mc
.lock
);
1332 ret
= mem_cgroup_same_or_subtree(memcg
, from
)
1333 || mem_cgroup_same_or_subtree(memcg
, to
);
1335 spin_unlock(&mc
.lock
);
1339 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*memcg
)
1341 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1342 if (mem_cgroup_under_move(memcg
)) {
1344 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1345 /* moving charge context might have finished. */
1348 finish_wait(&mc
.waitq
, &wait
);
1356 * Take this lock when
1357 * - a code tries to modify page's memcg while it's USED.
1358 * - a code tries to modify page state accounting in a memcg.
1359 * see mem_cgroup_stolen(), too.
1361 static void move_lock_mem_cgroup(struct mem_cgroup
*memcg
,
1362 unsigned long *flags
)
1364 spin_lock_irqsave(&memcg
->move_lock
, *flags
);
1367 static void move_unlock_mem_cgroup(struct mem_cgroup
*memcg
,
1368 unsigned long *flags
)
1370 spin_unlock_irqrestore(&memcg
->move_lock
, *flags
);
1374 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1375 * @memcg: The memory cgroup that went over limit
1376 * @p: Task that is going to be killed
1378 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1381 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1383 struct cgroup
*task_cgrp
;
1384 struct cgroup
*mem_cgrp
;
1386 * Need a buffer in BSS, can't rely on allocations. The code relies
1387 * on the assumption that OOM is serialized for memory controller.
1388 * If this assumption is broken, revisit this code.
1390 static char memcg_name
[PATH_MAX
];
1398 mem_cgrp
= memcg
->css
.cgroup
;
1399 task_cgrp
= task_cgroup(p
, mem_cgroup_subsys_id
);
1401 ret
= cgroup_path(task_cgrp
, memcg_name
, PATH_MAX
);
1404 * Unfortunately, we are unable to convert to a useful name
1405 * But we'll still print out the usage information
1412 printk(KERN_INFO
"Task in %s killed", memcg_name
);
1415 ret
= cgroup_path(mem_cgrp
, memcg_name
, PATH_MAX
);
1423 * Continues from above, so we don't need an KERN_ level
1425 printk(KERN_CONT
" as a result of limit of %s\n", memcg_name
);
1428 printk(KERN_INFO
"memory: usage %llukB, limit %llukB, failcnt %llu\n",
1429 res_counter_read_u64(&memcg
->res
, RES_USAGE
) >> 10,
1430 res_counter_read_u64(&memcg
->res
, RES_LIMIT
) >> 10,
1431 res_counter_read_u64(&memcg
->res
, RES_FAILCNT
));
1432 printk(KERN_INFO
"memory+swap: usage %llukB, limit %llukB, "
1434 res_counter_read_u64(&memcg
->memsw
, RES_USAGE
) >> 10,
1435 res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
) >> 10,
1436 res_counter_read_u64(&memcg
->memsw
, RES_FAILCNT
));
1440 * This function returns the number of memcg under hierarchy tree. Returns
1441 * 1(self count) if no children.
1443 static int mem_cgroup_count_children(struct mem_cgroup
*memcg
)
1446 struct mem_cgroup
*iter
;
1448 for_each_mem_cgroup_tree(iter
, memcg
)
1454 * Return the memory (and swap, if configured) limit for a memcg.
1456 u64
mem_cgroup_get_limit(struct mem_cgroup
*memcg
)
1461 limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
1462 limit
+= total_swap_pages
<< PAGE_SHIFT
;
1464 memsw
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
1466 * If memsw is finite and limits the amount of swap space available
1467 * to this memcg, return that limit.
1469 return min(limit
, memsw
);
1472 static unsigned long mem_cgroup_reclaim(struct mem_cgroup
*memcg
,
1474 unsigned long flags
)
1476 unsigned long total
= 0;
1477 bool noswap
= false;
1480 if (flags
& MEM_CGROUP_RECLAIM_NOSWAP
)
1482 if (!(flags
& MEM_CGROUP_RECLAIM_SHRINK
) && memcg
->memsw_is_minimum
)
1485 for (loop
= 0; loop
< MEM_CGROUP_MAX_RECLAIM_LOOPS
; loop
++) {
1487 drain_all_stock_async(memcg
);
1488 total
+= try_to_free_mem_cgroup_pages(memcg
, gfp_mask
, noswap
);
1490 * Allow limit shrinkers, which are triggered directly
1491 * by userspace, to catch signals and stop reclaim
1492 * after minimal progress, regardless of the margin.
1494 if (total
&& (flags
& MEM_CGROUP_RECLAIM_SHRINK
))
1496 if (mem_cgroup_margin(memcg
))
1499 * If nothing was reclaimed after two attempts, there
1500 * may be no reclaimable pages in this hierarchy.
1509 * test_mem_cgroup_node_reclaimable
1510 * @memcg: the target memcg
1511 * @nid: the node ID to be checked.
1512 * @noswap : specify true here if the user wants flle only information.
1514 * This function returns whether the specified memcg contains any
1515 * reclaimable pages on a node. Returns true if there are any reclaimable
1516 * pages in the node.
1518 static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup
*memcg
,
1519 int nid
, bool noswap
)
1521 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_FILE
))
1523 if (noswap
|| !total_swap_pages
)
1525 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_ANON
))
1530 #if MAX_NUMNODES > 1
1533 * Always updating the nodemask is not very good - even if we have an empty
1534 * list or the wrong list here, we can start from some node and traverse all
1535 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1538 static void mem_cgroup_may_update_nodemask(struct mem_cgroup
*memcg
)
1542 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1543 * pagein/pageout changes since the last update.
1545 if (!atomic_read(&memcg
->numainfo_events
))
1547 if (atomic_inc_return(&memcg
->numainfo_updating
) > 1)
1550 /* make a nodemask where this memcg uses memory from */
1551 memcg
->scan_nodes
= node_states
[N_HIGH_MEMORY
];
1553 for_each_node_mask(nid
, node_states
[N_HIGH_MEMORY
]) {
1555 if (!test_mem_cgroup_node_reclaimable(memcg
, nid
, false))
1556 node_clear(nid
, memcg
->scan_nodes
);
1559 atomic_set(&memcg
->numainfo_events
, 0);
1560 atomic_set(&memcg
->numainfo_updating
, 0);
1564 * Selecting a node where we start reclaim from. Because what we need is just
1565 * reducing usage counter, start from anywhere is O,K. Considering
1566 * memory reclaim from current node, there are pros. and cons.
1568 * Freeing memory from current node means freeing memory from a node which
1569 * we'll use or we've used. So, it may make LRU bad. And if several threads
1570 * hit limits, it will see a contention on a node. But freeing from remote
1571 * node means more costs for memory reclaim because of memory latency.
1573 * Now, we use round-robin. Better algorithm is welcomed.
1575 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1579 mem_cgroup_may_update_nodemask(memcg
);
1580 node
= memcg
->last_scanned_node
;
1582 node
= next_node(node
, memcg
->scan_nodes
);
1583 if (node
== MAX_NUMNODES
)
1584 node
= first_node(memcg
->scan_nodes
);
1586 * We call this when we hit limit, not when pages are added to LRU.
1587 * No LRU may hold pages because all pages are UNEVICTABLE or
1588 * memcg is too small and all pages are not on LRU. In that case,
1589 * we use curret node.
1591 if (unlikely(node
== MAX_NUMNODES
))
1592 node
= numa_node_id();
1594 memcg
->last_scanned_node
= node
;
1599 * Check all nodes whether it contains reclaimable pages or not.
1600 * For quick scan, we make use of scan_nodes. This will allow us to skip
1601 * unused nodes. But scan_nodes is lazily updated and may not cotain
1602 * enough new information. We need to do double check.
1604 static bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1609 * quick check...making use of scan_node.
1610 * We can skip unused nodes.
1612 if (!nodes_empty(memcg
->scan_nodes
)) {
1613 for (nid
= first_node(memcg
->scan_nodes
);
1615 nid
= next_node(nid
, memcg
->scan_nodes
)) {
1617 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1622 * Check rest of nodes.
1624 for_each_node_state(nid
, N_HIGH_MEMORY
) {
1625 if (node_isset(nid
, memcg
->scan_nodes
))
1627 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1634 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1639 static bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1641 return test_mem_cgroup_node_reclaimable(memcg
, 0, noswap
);
1645 static int mem_cgroup_soft_reclaim(struct mem_cgroup
*root_memcg
,
1648 unsigned long *total_scanned
)
1650 struct mem_cgroup
*victim
= NULL
;
1653 unsigned long excess
;
1654 unsigned long nr_scanned
;
1655 struct mem_cgroup_reclaim_cookie reclaim
= {
1660 excess
= res_counter_soft_limit_excess(&root_memcg
->res
) >> PAGE_SHIFT
;
1663 victim
= mem_cgroup_iter(root_memcg
, victim
, &reclaim
);
1668 * If we have not been able to reclaim
1669 * anything, it might because there are
1670 * no reclaimable pages under this hierarchy
1675 * We want to do more targeted reclaim.
1676 * excess >> 2 is not to excessive so as to
1677 * reclaim too much, nor too less that we keep
1678 * coming back to reclaim from this cgroup
1680 if (total
>= (excess
>> 2) ||
1681 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
))
1686 if (!mem_cgroup_reclaimable(victim
, false))
1688 total
+= mem_cgroup_shrink_node_zone(victim
, gfp_mask
, false,
1690 *total_scanned
+= nr_scanned
;
1691 if (!res_counter_soft_limit_excess(&root_memcg
->res
))
1694 mem_cgroup_iter_break(root_memcg
, victim
);
1699 * Check OOM-Killer is already running under our hierarchy.
1700 * If someone is running, return false.
1701 * Has to be called with memcg_oom_lock
1703 static bool mem_cgroup_oom_lock(struct mem_cgroup
*memcg
)
1705 struct mem_cgroup
*iter
, *failed
= NULL
;
1707 for_each_mem_cgroup_tree(iter
, memcg
) {
1708 if (iter
->oom_lock
) {
1710 * this subtree of our hierarchy is already locked
1711 * so we cannot give a lock.
1714 mem_cgroup_iter_break(memcg
, iter
);
1717 iter
->oom_lock
= true;
1724 * OK, we failed to lock the whole subtree so we have to clean up
1725 * what we set up to the failing subtree
1727 for_each_mem_cgroup_tree(iter
, memcg
) {
1728 if (iter
== failed
) {
1729 mem_cgroup_iter_break(memcg
, iter
);
1732 iter
->oom_lock
= false;
1738 * Has to be called with memcg_oom_lock
1740 static int mem_cgroup_oom_unlock(struct mem_cgroup
*memcg
)
1742 struct mem_cgroup
*iter
;
1744 for_each_mem_cgroup_tree(iter
, memcg
)
1745 iter
->oom_lock
= false;
1749 static void mem_cgroup_mark_under_oom(struct mem_cgroup
*memcg
)
1751 struct mem_cgroup
*iter
;
1753 for_each_mem_cgroup_tree(iter
, memcg
)
1754 atomic_inc(&iter
->under_oom
);
1757 static void mem_cgroup_unmark_under_oom(struct mem_cgroup
*memcg
)
1759 struct mem_cgroup
*iter
;
1762 * When a new child is created while the hierarchy is under oom,
1763 * mem_cgroup_oom_lock() may not be called. We have to use
1764 * atomic_add_unless() here.
1766 for_each_mem_cgroup_tree(iter
, memcg
)
1767 atomic_add_unless(&iter
->under_oom
, -1, 0);
1770 static DEFINE_SPINLOCK(memcg_oom_lock
);
1771 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1773 struct oom_wait_info
{
1774 struct mem_cgroup
*memcg
;
1778 static int memcg_oom_wake_function(wait_queue_t
*wait
,
1779 unsigned mode
, int sync
, void *arg
)
1781 struct mem_cgroup
*wake_memcg
= (struct mem_cgroup
*)arg
;
1782 struct mem_cgroup
*oom_wait_memcg
;
1783 struct oom_wait_info
*oom_wait_info
;
1785 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1786 oom_wait_memcg
= oom_wait_info
->memcg
;
1789 * Both of oom_wait_info->memcg and wake_memcg are stable under us.
1790 * Then we can use css_is_ancestor without taking care of RCU.
1792 if (!mem_cgroup_same_or_subtree(oom_wait_memcg
, wake_memcg
)
1793 && !mem_cgroup_same_or_subtree(wake_memcg
, oom_wait_memcg
))
1795 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1798 static void memcg_wakeup_oom(struct mem_cgroup
*memcg
)
1800 /* for filtering, pass "memcg" as argument. */
1801 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, memcg
);
1804 static void memcg_oom_recover(struct mem_cgroup
*memcg
)
1806 if (memcg
&& atomic_read(&memcg
->under_oom
))
1807 memcg_wakeup_oom(memcg
);
1811 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1813 static bool mem_cgroup_handle_oom(struct mem_cgroup
*memcg
, gfp_t mask
,
1816 struct oom_wait_info owait
;
1817 bool locked
, need_to_kill
;
1819 owait
.memcg
= memcg
;
1820 owait
.wait
.flags
= 0;
1821 owait
.wait
.func
= memcg_oom_wake_function
;
1822 owait
.wait
.private = current
;
1823 INIT_LIST_HEAD(&owait
.wait
.task_list
);
1824 need_to_kill
= true;
1825 mem_cgroup_mark_under_oom(memcg
);
1827 /* At first, try to OOM lock hierarchy under memcg.*/
1828 spin_lock(&memcg_oom_lock
);
1829 locked
= mem_cgroup_oom_lock(memcg
);
1831 * Even if signal_pending(), we can't quit charge() loop without
1832 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1833 * under OOM is always welcomed, use TASK_KILLABLE here.
1835 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1836 if (!locked
|| memcg
->oom_kill_disable
)
1837 need_to_kill
= false;
1839 mem_cgroup_oom_notify(memcg
);
1840 spin_unlock(&memcg_oom_lock
);
1843 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1844 mem_cgroup_out_of_memory(memcg
, mask
, order
);
1847 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1849 spin_lock(&memcg_oom_lock
);
1851 mem_cgroup_oom_unlock(memcg
);
1852 memcg_wakeup_oom(memcg
);
1853 spin_unlock(&memcg_oom_lock
);
1855 mem_cgroup_unmark_under_oom(memcg
);
1857 if (test_thread_flag(TIF_MEMDIE
) || fatal_signal_pending(current
))
1859 /* Give chance to dying process */
1860 schedule_timeout_uninterruptible(1);
1865 * Currently used to update mapped file statistics, but the routine can be
1866 * generalized to update other statistics as well.
1868 * Notes: Race condition
1870 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1871 * it tends to be costly. But considering some conditions, we doesn't need
1872 * to do so _always_.
1874 * Considering "charge", lock_page_cgroup() is not required because all
1875 * file-stat operations happen after a page is attached to radix-tree. There
1876 * are no race with "charge".
1878 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1879 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1880 * if there are race with "uncharge". Statistics itself is properly handled
1883 * Considering "move", this is an only case we see a race. To make the race
1884 * small, we check mm->moving_account and detect there are possibility of race
1885 * If there is, we take a lock.
1888 void __mem_cgroup_begin_update_page_stat(struct page
*page
,
1889 bool *locked
, unsigned long *flags
)
1891 struct mem_cgroup
*memcg
;
1892 struct page_cgroup
*pc
;
1894 pc
= lookup_page_cgroup(page
);
1896 memcg
= pc
->mem_cgroup
;
1897 if (unlikely(!memcg
|| !PageCgroupUsed(pc
)))
1900 * If this memory cgroup is not under account moving, we don't
1901 * need to take move_lock_page_cgroup(). Because we already hold
1902 * rcu_read_lock(), any calls to move_account will be delayed until
1903 * rcu_read_unlock() if mem_cgroup_stolen() == true.
1905 if (!mem_cgroup_stolen(memcg
))
1908 move_lock_mem_cgroup(memcg
, flags
);
1909 if (memcg
!= pc
->mem_cgroup
|| !PageCgroupUsed(pc
)) {
1910 move_unlock_mem_cgroup(memcg
, flags
);
1916 void __mem_cgroup_end_update_page_stat(struct page
*page
, unsigned long *flags
)
1918 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1921 * It's guaranteed that pc->mem_cgroup never changes while
1922 * lock is held because a routine modifies pc->mem_cgroup
1923 * should take move_lock_page_cgroup().
1925 move_unlock_mem_cgroup(pc
->mem_cgroup
, flags
);
1928 void mem_cgroup_update_page_stat(struct page
*page
,
1929 enum mem_cgroup_page_stat_item idx
, int val
)
1931 struct mem_cgroup
*memcg
;
1932 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1933 unsigned long uninitialized_var(flags
);
1935 if (mem_cgroup_disabled())
1938 memcg
= pc
->mem_cgroup
;
1939 if (unlikely(!memcg
|| !PageCgroupUsed(pc
)))
1943 case MEMCG_NR_FILE_MAPPED
:
1944 idx
= MEM_CGROUP_STAT_FILE_MAPPED
;
1950 this_cpu_add(memcg
->stat
->count
[idx
], val
);
1954 * size of first charge trial. "32" comes from vmscan.c's magic value.
1955 * TODO: maybe necessary to use big numbers in big irons.
1957 #define CHARGE_BATCH 32U
1958 struct memcg_stock_pcp
{
1959 struct mem_cgroup
*cached
; /* this never be root cgroup */
1960 unsigned int nr_pages
;
1961 struct work_struct work
;
1962 unsigned long flags
;
1963 #define FLUSHING_CACHED_CHARGE 0
1965 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
1966 static DEFINE_MUTEX(percpu_charge_mutex
);
1969 * Try to consume stocked charge on this cpu. If success, one page is consumed
1970 * from local stock and true is returned. If the stock is 0 or charges from a
1971 * cgroup which is not current target, returns false. This stock will be
1974 static bool consume_stock(struct mem_cgroup
*memcg
)
1976 struct memcg_stock_pcp
*stock
;
1979 stock
= &get_cpu_var(memcg_stock
);
1980 if (memcg
== stock
->cached
&& stock
->nr_pages
)
1982 else /* need to call res_counter_charge */
1984 put_cpu_var(memcg_stock
);
1989 * Returns stocks cached in percpu to res_counter and reset cached information.
1991 static void drain_stock(struct memcg_stock_pcp
*stock
)
1993 struct mem_cgroup
*old
= stock
->cached
;
1995 if (stock
->nr_pages
) {
1996 unsigned long bytes
= stock
->nr_pages
* PAGE_SIZE
;
1998 res_counter_uncharge(&old
->res
, bytes
);
1999 if (do_swap_account
)
2000 res_counter_uncharge(&old
->memsw
, bytes
);
2001 stock
->nr_pages
= 0;
2003 stock
->cached
= NULL
;
2007 * This must be called under preempt disabled or must be called by
2008 * a thread which is pinned to local cpu.
2010 static void drain_local_stock(struct work_struct
*dummy
)
2012 struct memcg_stock_pcp
*stock
= &__get_cpu_var(memcg_stock
);
2014 clear_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
);
2018 * Cache charges(val) which is from res_counter, to local per_cpu area.
2019 * This will be consumed by consume_stock() function, later.
2021 static void refill_stock(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
2023 struct memcg_stock_pcp
*stock
= &get_cpu_var(memcg_stock
);
2025 if (stock
->cached
!= memcg
) { /* reset if necessary */
2027 stock
->cached
= memcg
;
2029 stock
->nr_pages
+= nr_pages
;
2030 put_cpu_var(memcg_stock
);
2034 * Drains all per-CPU charge caches for given root_memcg resp. subtree
2035 * of the hierarchy under it. sync flag says whether we should block
2036 * until the work is done.
2038 static void drain_all_stock(struct mem_cgroup
*root_memcg
, bool sync
)
2042 /* Notify other cpus that system-wide "drain" is running */
2045 for_each_online_cpu(cpu
) {
2046 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2047 struct mem_cgroup
*memcg
;
2049 memcg
= stock
->cached
;
2050 if (!memcg
|| !stock
->nr_pages
)
2052 if (!mem_cgroup_same_or_subtree(root_memcg
, memcg
))
2054 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
)) {
2056 drain_local_stock(&stock
->work
);
2058 schedule_work_on(cpu
, &stock
->work
);
2066 for_each_online_cpu(cpu
) {
2067 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2068 if (test_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
))
2069 flush_work(&stock
->work
);
2076 * Tries to drain stocked charges in other cpus. This function is asynchronous
2077 * and just put a work per cpu for draining localy on each cpu. Caller can
2078 * expects some charges will be back to res_counter later but cannot wait for
2081 static void drain_all_stock_async(struct mem_cgroup
*root_memcg
)
2084 * If someone calls draining, avoid adding more kworker runs.
2086 if (!mutex_trylock(&percpu_charge_mutex
))
2088 drain_all_stock(root_memcg
, false);
2089 mutex_unlock(&percpu_charge_mutex
);
2092 /* This is a synchronous drain interface. */
2093 static void drain_all_stock_sync(struct mem_cgroup
*root_memcg
)
2095 /* called when force_empty is called */
2096 mutex_lock(&percpu_charge_mutex
);
2097 drain_all_stock(root_memcg
, true);
2098 mutex_unlock(&percpu_charge_mutex
);
2102 * This function drains percpu counter value from DEAD cpu and
2103 * move it to local cpu. Note that this function can be preempted.
2105 static void mem_cgroup_drain_pcp_counter(struct mem_cgroup
*memcg
, int cpu
)
2109 spin_lock(&memcg
->pcp_counter_lock
);
2110 for (i
= 0; i
< MEM_CGROUP_STAT_NSTATS
; i
++) {
2111 long x
= per_cpu(memcg
->stat
->count
[i
], cpu
);
2113 per_cpu(memcg
->stat
->count
[i
], cpu
) = 0;
2114 memcg
->nocpu_base
.count
[i
] += x
;
2116 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
2117 unsigned long x
= per_cpu(memcg
->stat
->events
[i
], cpu
);
2119 per_cpu(memcg
->stat
->events
[i
], cpu
) = 0;
2120 memcg
->nocpu_base
.events
[i
] += x
;
2122 spin_unlock(&memcg
->pcp_counter_lock
);
2125 static int __cpuinit
memcg_cpu_hotplug_callback(struct notifier_block
*nb
,
2126 unsigned long action
,
2129 int cpu
= (unsigned long)hcpu
;
2130 struct memcg_stock_pcp
*stock
;
2131 struct mem_cgroup
*iter
;
2133 if (action
== CPU_ONLINE
)
2136 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
2139 for_each_mem_cgroup(iter
)
2140 mem_cgroup_drain_pcp_counter(iter
, cpu
);
2142 stock
= &per_cpu(memcg_stock
, cpu
);
2148 /* See __mem_cgroup_try_charge() for details */
2150 CHARGE_OK
, /* success */
2151 CHARGE_RETRY
, /* need to retry but retry is not bad */
2152 CHARGE_NOMEM
, /* we can't do more. return -ENOMEM */
2153 CHARGE_WOULDBLOCK
, /* GFP_WAIT wasn't set and no enough res. */
2154 CHARGE_OOM_DIE
, /* the current is killed because of OOM */
2157 static int mem_cgroup_do_charge(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
2158 unsigned int nr_pages
, bool oom_check
)
2160 unsigned long csize
= nr_pages
* PAGE_SIZE
;
2161 struct mem_cgroup
*mem_over_limit
;
2162 struct res_counter
*fail_res
;
2163 unsigned long flags
= 0;
2166 ret
= res_counter_charge(&memcg
->res
, csize
, &fail_res
);
2169 if (!do_swap_account
)
2171 ret
= res_counter_charge(&memcg
->memsw
, csize
, &fail_res
);
2175 res_counter_uncharge(&memcg
->res
, csize
);
2176 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, memsw
);
2177 flags
|= MEM_CGROUP_RECLAIM_NOSWAP
;
2179 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, res
);
2181 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2182 * of regular pages (CHARGE_BATCH), or a single regular page (1).
2184 * Never reclaim on behalf of optional batching, retry with a
2185 * single page instead.
2187 if (nr_pages
== CHARGE_BATCH
)
2188 return CHARGE_RETRY
;
2190 if (!(gfp_mask
& __GFP_WAIT
))
2191 return CHARGE_WOULDBLOCK
;
2193 ret
= mem_cgroup_reclaim(mem_over_limit
, gfp_mask
, flags
);
2194 if (mem_cgroup_margin(mem_over_limit
) >= nr_pages
)
2195 return CHARGE_RETRY
;
2197 * Even though the limit is exceeded at this point, reclaim
2198 * may have been able to free some pages. Retry the charge
2199 * before killing the task.
2201 * Only for regular pages, though: huge pages are rather
2202 * unlikely to succeed so close to the limit, and we fall back
2203 * to regular pages anyway in case of failure.
2205 if (nr_pages
== 1 && ret
)
2206 return CHARGE_RETRY
;
2209 * At task move, charge accounts can be doubly counted. So, it's
2210 * better to wait until the end of task_move if something is going on.
2212 if (mem_cgroup_wait_acct_move(mem_over_limit
))
2213 return CHARGE_RETRY
;
2215 /* If we don't need to call oom-killer at el, return immediately */
2217 return CHARGE_NOMEM
;
2219 if (!mem_cgroup_handle_oom(mem_over_limit
, gfp_mask
, get_order(csize
)))
2220 return CHARGE_OOM_DIE
;
2222 return CHARGE_RETRY
;
2226 * __mem_cgroup_try_charge() does
2227 * 1. detect memcg to be charged against from passed *mm and *ptr,
2228 * 2. update res_counter
2229 * 3. call memory reclaim if necessary.
2231 * In some special case, if the task is fatal, fatal_signal_pending() or
2232 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2233 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2234 * as possible without any hazards. 2: all pages should have a valid
2235 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2236 * pointer, that is treated as a charge to root_mem_cgroup.
2238 * So __mem_cgroup_try_charge() will return
2239 * 0 ... on success, filling *ptr with a valid memcg pointer.
2240 * -ENOMEM ... charge failure because of resource limits.
2241 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2243 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2244 * the oom-killer can be invoked.
2246 static int __mem_cgroup_try_charge(struct mm_struct
*mm
,
2248 unsigned int nr_pages
,
2249 struct mem_cgroup
**ptr
,
2252 unsigned int batch
= max(CHARGE_BATCH
, nr_pages
);
2253 int nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2254 struct mem_cgroup
*memcg
= NULL
;
2258 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2259 * in system level. So, allow to go ahead dying process in addition to
2262 if (unlikely(test_thread_flag(TIF_MEMDIE
)
2263 || fatal_signal_pending(current
)))
2267 * We always charge the cgroup the mm_struct belongs to.
2268 * The mm_struct's mem_cgroup changes on task migration if the
2269 * thread group leader migrates. It's possible that mm is not
2270 * set, if so charge the init_mm (happens for pagecache usage).
2273 *ptr
= root_mem_cgroup
;
2275 if (*ptr
) { /* css should be a valid one */
2277 VM_BUG_ON(css_is_removed(&memcg
->css
));
2278 if (mem_cgroup_is_root(memcg
))
2280 if (nr_pages
== 1 && consume_stock(memcg
))
2282 css_get(&memcg
->css
);
2284 struct task_struct
*p
;
2287 p
= rcu_dereference(mm
->owner
);
2289 * Because we don't have task_lock(), "p" can exit.
2290 * In that case, "memcg" can point to root or p can be NULL with
2291 * race with swapoff. Then, we have small risk of mis-accouning.
2292 * But such kind of mis-account by race always happens because
2293 * we don't have cgroup_mutex(). It's overkill and we allo that
2295 * (*) swapoff at el will charge against mm-struct not against
2296 * task-struct. So, mm->owner can be NULL.
2298 memcg
= mem_cgroup_from_task(p
);
2300 memcg
= root_mem_cgroup
;
2301 if (mem_cgroup_is_root(memcg
)) {
2305 if (nr_pages
== 1 && consume_stock(memcg
)) {
2307 * It seems dagerous to access memcg without css_get().
2308 * But considering how consume_stok works, it's not
2309 * necessary. If consume_stock success, some charges
2310 * from this memcg are cached on this cpu. So, we
2311 * don't need to call css_get()/css_tryget() before
2312 * calling consume_stock().
2317 /* after here, we may be blocked. we need to get refcnt */
2318 if (!css_tryget(&memcg
->css
)) {
2328 /* If killed, bypass charge */
2329 if (fatal_signal_pending(current
)) {
2330 css_put(&memcg
->css
);
2335 if (oom
&& !nr_oom_retries
) {
2337 nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2340 ret
= mem_cgroup_do_charge(memcg
, gfp_mask
, batch
, oom_check
);
2344 case CHARGE_RETRY
: /* not in OOM situation but retry */
2346 css_put(&memcg
->css
);
2349 case CHARGE_WOULDBLOCK
: /* !__GFP_WAIT */
2350 css_put(&memcg
->css
);
2352 case CHARGE_NOMEM
: /* OOM routine works */
2354 css_put(&memcg
->css
);
2357 /* If oom, we never return -ENOMEM */
2360 case CHARGE_OOM_DIE
: /* Killed by OOM Killer */
2361 css_put(&memcg
->css
);
2364 } while (ret
!= CHARGE_OK
);
2366 if (batch
> nr_pages
)
2367 refill_stock(memcg
, batch
- nr_pages
);
2368 css_put(&memcg
->css
);
2376 *ptr
= root_mem_cgroup
;
2381 * Somemtimes we have to undo a charge we got by try_charge().
2382 * This function is for that and do uncharge, put css's refcnt.
2383 * gotten by try_charge().
2385 static void __mem_cgroup_cancel_charge(struct mem_cgroup
*memcg
,
2386 unsigned int nr_pages
)
2388 if (!mem_cgroup_is_root(memcg
)) {
2389 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2391 res_counter_uncharge(&memcg
->res
, bytes
);
2392 if (do_swap_account
)
2393 res_counter_uncharge(&memcg
->memsw
, bytes
);
2398 * Cancel chrages in this cgroup....doesn't propagate to parent cgroup.
2399 * This is useful when moving usage to parent cgroup.
2401 static void __mem_cgroup_cancel_local_charge(struct mem_cgroup
*memcg
,
2402 unsigned int nr_pages
)
2404 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2406 if (mem_cgroup_is_root(memcg
))
2409 res_counter_uncharge_until(&memcg
->res
, memcg
->res
.parent
, bytes
);
2410 if (do_swap_account
)
2411 res_counter_uncharge_until(&memcg
->memsw
,
2412 memcg
->memsw
.parent
, bytes
);
2416 * A helper function to get mem_cgroup from ID. must be called under
2417 * rcu_read_lock(). The caller must check css_is_removed() or some if
2418 * it's concern. (dropping refcnt from swap can be called against removed
2421 static struct mem_cgroup
*mem_cgroup_lookup(unsigned short id
)
2423 struct cgroup_subsys_state
*css
;
2425 /* ID 0 is unused ID */
2428 css
= css_lookup(&mem_cgroup_subsys
, id
);
2431 return container_of(css
, struct mem_cgroup
, css
);
2434 struct mem_cgroup
*try_get_mem_cgroup_from_page(struct page
*page
)
2436 struct mem_cgroup
*memcg
= NULL
;
2437 struct page_cgroup
*pc
;
2441 VM_BUG_ON(!PageLocked(page
));
2443 pc
= lookup_page_cgroup(page
);
2444 lock_page_cgroup(pc
);
2445 if (PageCgroupUsed(pc
)) {
2446 memcg
= pc
->mem_cgroup
;
2447 if (memcg
&& !css_tryget(&memcg
->css
))
2449 } else if (PageSwapCache(page
)) {
2450 ent
.val
= page_private(page
);
2451 id
= lookup_swap_cgroup_id(ent
);
2453 memcg
= mem_cgroup_lookup(id
);
2454 if (memcg
&& !css_tryget(&memcg
->css
))
2458 unlock_page_cgroup(pc
);
2462 static void __mem_cgroup_commit_charge(struct mem_cgroup
*memcg
,
2464 unsigned int nr_pages
,
2465 enum charge_type ctype
,
2468 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
2469 struct zone
*uninitialized_var(zone
);
2470 struct lruvec
*lruvec
;
2471 bool was_on_lru
= false;
2474 lock_page_cgroup(pc
);
2475 if (unlikely(PageCgroupUsed(pc
))) {
2476 unlock_page_cgroup(pc
);
2477 __mem_cgroup_cancel_charge(memcg
, nr_pages
);
2481 * we don't need page_cgroup_lock about tail pages, becase they are not
2482 * accessed by any other context at this point.
2486 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2487 * may already be on some other mem_cgroup's LRU. Take care of it.
2490 zone
= page_zone(page
);
2491 spin_lock_irq(&zone
->lru_lock
);
2492 if (PageLRU(page
)) {
2493 lruvec
= mem_cgroup_zone_lruvec(zone
, pc
->mem_cgroup
);
2495 del_page_from_lru_list(page
, lruvec
, page_lru(page
));
2500 pc
->mem_cgroup
= memcg
;
2502 * We access a page_cgroup asynchronously without lock_page_cgroup().
2503 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2504 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2505 * before USED bit, we need memory barrier here.
2506 * See mem_cgroup_add_lru_list(), etc.
2509 SetPageCgroupUsed(pc
);
2513 lruvec
= mem_cgroup_zone_lruvec(zone
, pc
->mem_cgroup
);
2514 VM_BUG_ON(PageLRU(page
));
2516 add_page_to_lru_list(page
, lruvec
, page_lru(page
));
2518 spin_unlock_irq(&zone
->lru_lock
);
2521 if (ctype
== MEM_CGROUP_CHARGE_TYPE_ANON
)
2526 mem_cgroup_charge_statistics(memcg
, anon
, nr_pages
);
2527 unlock_page_cgroup(pc
);
2530 * "charge_statistics" updated event counter. Then, check it.
2531 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2532 * if they exceeds softlimit.
2534 memcg_check_events(memcg
, page
);
2537 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2539 #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION)
2541 * Because tail pages are not marked as "used", set it. We're under
2542 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2543 * charge/uncharge will be never happen and move_account() is done under
2544 * compound_lock(), so we don't have to take care of races.
2546 void mem_cgroup_split_huge_fixup(struct page
*head
)
2548 struct page_cgroup
*head_pc
= lookup_page_cgroup(head
);
2549 struct page_cgroup
*pc
;
2552 if (mem_cgroup_disabled())
2554 for (i
= 1; i
< HPAGE_PMD_NR
; i
++) {
2556 pc
->mem_cgroup
= head_pc
->mem_cgroup
;
2557 smp_wmb();/* see __commit_charge() */
2558 pc
->flags
= head_pc
->flags
& ~PCGF_NOCOPY_AT_SPLIT
;
2561 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2564 * mem_cgroup_move_account - move account of the page
2566 * @nr_pages: number of regular pages (>1 for huge pages)
2567 * @pc: page_cgroup of the page.
2568 * @from: mem_cgroup which the page is moved from.
2569 * @to: mem_cgroup which the page is moved to. @from != @to.
2571 * The caller must confirm following.
2572 * - page is not on LRU (isolate_page() is useful.)
2573 * - compound_lock is held when nr_pages > 1
2575 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
2578 static int mem_cgroup_move_account(struct page
*page
,
2579 unsigned int nr_pages
,
2580 struct page_cgroup
*pc
,
2581 struct mem_cgroup
*from
,
2582 struct mem_cgroup
*to
)
2584 unsigned long flags
;
2586 bool anon
= PageAnon(page
);
2588 VM_BUG_ON(from
== to
);
2589 VM_BUG_ON(PageLRU(page
));
2591 * The page is isolated from LRU. So, collapse function
2592 * will not handle this page. But page splitting can happen.
2593 * Do this check under compound_page_lock(). The caller should
2597 if (nr_pages
> 1 && !PageTransHuge(page
))
2600 lock_page_cgroup(pc
);
2603 if (!PageCgroupUsed(pc
) || pc
->mem_cgroup
!= from
)
2606 move_lock_mem_cgroup(from
, &flags
);
2608 if (!anon
&& page_mapped(page
)) {
2609 /* Update mapped_file data for mem_cgroup */
2611 __this_cpu_dec(from
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2612 __this_cpu_inc(to
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2615 mem_cgroup_charge_statistics(from
, anon
, -nr_pages
);
2617 /* caller should have done css_get */
2618 pc
->mem_cgroup
= to
;
2619 mem_cgroup_charge_statistics(to
, anon
, nr_pages
);
2621 * We charges against "to" which may not have any tasks. Then, "to"
2622 * can be under rmdir(). But in current implementation, caller of
2623 * this function is just force_empty() and move charge, so it's
2624 * guaranteed that "to" is never removed. So, we don't check rmdir
2627 move_unlock_mem_cgroup(from
, &flags
);
2630 unlock_page_cgroup(pc
);
2634 memcg_check_events(to
, page
);
2635 memcg_check_events(from
, page
);
2641 * move charges to its parent.
2644 static int mem_cgroup_move_parent(struct page
*page
,
2645 struct page_cgroup
*pc
,
2646 struct mem_cgroup
*child
,
2649 struct mem_cgroup
*parent
;
2650 unsigned int nr_pages
;
2651 unsigned long uninitialized_var(flags
);
2655 if (mem_cgroup_is_root(child
))
2659 if (!get_page_unless_zero(page
))
2661 if (isolate_lru_page(page
))
2664 nr_pages
= hpage_nr_pages(page
);
2666 parent
= parent_mem_cgroup(child
);
2668 * If no parent, move charges to root cgroup.
2671 parent
= root_mem_cgroup
;
2674 flags
= compound_lock_irqsave(page
);
2676 ret
= mem_cgroup_move_account(page
, nr_pages
,
2679 __mem_cgroup_cancel_local_charge(child
, nr_pages
);
2682 compound_unlock_irqrestore(page
, flags
);
2683 putback_lru_page(page
);
2691 * Charge the memory controller for page usage.
2693 * 0 if the charge was successful
2694 * < 0 if the cgroup is over its limit
2696 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
2697 gfp_t gfp_mask
, enum charge_type ctype
)
2699 struct mem_cgroup
*memcg
= NULL
;
2700 unsigned int nr_pages
= 1;
2704 if (PageTransHuge(page
)) {
2705 nr_pages
<<= compound_order(page
);
2706 VM_BUG_ON(!PageTransHuge(page
));
2708 * Never OOM-kill a process for a huge page. The
2709 * fault handler will fall back to regular pages.
2714 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, nr_pages
, &memcg
, oom
);
2717 __mem_cgroup_commit_charge(memcg
, page
, nr_pages
, ctype
, false);
2721 int mem_cgroup_newpage_charge(struct page
*page
,
2722 struct mm_struct
*mm
, gfp_t gfp_mask
)
2724 if (mem_cgroup_disabled())
2726 VM_BUG_ON(page_mapped(page
));
2727 VM_BUG_ON(page
->mapping
&& !PageAnon(page
));
2729 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2730 MEM_CGROUP_CHARGE_TYPE_ANON
);
2734 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2735 enum charge_type ctype
);
2737 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
2740 struct mem_cgroup
*memcg
= NULL
;
2741 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
2744 if (mem_cgroup_disabled())
2746 if (PageCompound(page
))
2751 if (!page_is_file_cache(page
))
2752 type
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
2754 if (!PageSwapCache(page
))
2755 ret
= mem_cgroup_charge_common(page
, mm
, gfp_mask
, type
);
2756 else { /* page is swapcache/shmem */
2757 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &memcg
);
2759 __mem_cgroup_commit_charge_swapin(page
, memcg
, type
);
2765 * While swap-in, try_charge -> commit or cancel, the page is locked.
2766 * And when try_charge() successfully returns, one refcnt to memcg without
2767 * struct page_cgroup is acquired. This refcnt will be consumed by
2768 * "commit()" or removed by "cancel()"
2770 int mem_cgroup_try_charge_swapin(struct mm_struct
*mm
,
2772 gfp_t mask
, struct mem_cgroup
**memcgp
)
2774 struct mem_cgroup
*memcg
;
2779 if (mem_cgroup_disabled())
2782 if (!do_swap_account
)
2785 * A racing thread's fault, or swapoff, may have already updated
2786 * the pte, and even removed page from swap cache: in those cases
2787 * do_swap_page()'s pte_same() test will fail; but there's also a
2788 * KSM case which does need to charge the page.
2790 if (!PageSwapCache(page
))
2792 memcg
= try_get_mem_cgroup_from_page(page
);
2796 ret
= __mem_cgroup_try_charge(NULL
, mask
, 1, memcgp
, true);
2797 css_put(&memcg
->css
);
2804 ret
= __mem_cgroup_try_charge(mm
, mask
, 1, memcgp
, true);
2811 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*memcg
,
2812 enum charge_type ctype
)
2814 if (mem_cgroup_disabled())
2818 cgroup_exclude_rmdir(&memcg
->css
);
2820 __mem_cgroup_commit_charge(memcg
, page
, 1, ctype
, true);
2822 * Now swap is on-memory. This means this page may be
2823 * counted both as mem and swap....double count.
2824 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2825 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2826 * may call delete_from_swap_cache() before reach here.
2828 if (do_swap_account
&& PageSwapCache(page
)) {
2829 swp_entry_t ent
= {.val
= page_private(page
)};
2830 mem_cgroup_uncharge_swap(ent
);
2833 * At swapin, we may charge account against cgroup which has no tasks.
2834 * So, rmdir()->pre_destroy() can be called while we do this charge.
2835 * In that case, we need to call pre_destroy() again. check it here.
2837 cgroup_release_and_wakeup_rmdir(&memcg
->css
);
2840 void mem_cgroup_commit_charge_swapin(struct page
*page
,
2841 struct mem_cgroup
*memcg
)
2843 __mem_cgroup_commit_charge_swapin(page
, memcg
,
2844 MEM_CGROUP_CHARGE_TYPE_ANON
);
2847 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup
*memcg
)
2849 if (mem_cgroup_disabled())
2853 __mem_cgroup_cancel_charge(memcg
, 1);
2856 static void mem_cgroup_do_uncharge(struct mem_cgroup
*memcg
,
2857 unsigned int nr_pages
,
2858 const enum charge_type ctype
)
2860 struct memcg_batch_info
*batch
= NULL
;
2861 bool uncharge_memsw
= true;
2863 /* If swapout, usage of swap doesn't decrease */
2864 if (!do_swap_account
|| ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
)
2865 uncharge_memsw
= false;
2867 batch
= ¤t
->memcg_batch
;
2869 * In usual, we do css_get() when we remember memcg pointer.
2870 * But in this case, we keep res->usage until end of a series of
2871 * uncharges. Then, it's ok to ignore memcg's refcnt.
2874 batch
->memcg
= memcg
;
2876 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2877 * In those cases, all pages freed continuously can be expected to be in
2878 * the same cgroup and we have chance to coalesce uncharges.
2879 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2880 * because we want to do uncharge as soon as possible.
2883 if (!batch
->do_batch
|| test_thread_flag(TIF_MEMDIE
))
2884 goto direct_uncharge
;
2887 goto direct_uncharge
;
2890 * In typical case, batch->memcg == mem. This means we can
2891 * merge a series of uncharges to an uncharge of res_counter.
2892 * If not, we uncharge res_counter ony by one.
2894 if (batch
->memcg
!= memcg
)
2895 goto direct_uncharge
;
2896 /* remember freed charge and uncharge it later */
2899 batch
->memsw_nr_pages
++;
2902 res_counter_uncharge(&memcg
->res
, nr_pages
* PAGE_SIZE
);
2904 res_counter_uncharge(&memcg
->memsw
, nr_pages
* PAGE_SIZE
);
2905 if (unlikely(batch
->memcg
!= memcg
))
2906 memcg_oom_recover(memcg
);
2910 * uncharge if !page_mapped(page)
2912 static struct mem_cgroup
*
2913 __mem_cgroup_uncharge_common(struct page
*page
, enum charge_type ctype
)
2915 struct mem_cgroup
*memcg
= NULL
;
2916 unsigned int nr_pages
= 1;
2917 struct page_cgroup
*pc
;
2920 if (mem_cgroup_disabled())
2923 if (PageSwapCache(page
))
2926 if (PageTransHuge(page
)) {
2927 nr_pages
<<= compound_order(page
);
2928 VM_BUG_ON(!PageTransHuge(page
));
2931 * Check if our page_cgroup is valid
2933 pc
= lookup_page_cgroup(page
);
2934 if (unlikely(!PageCgroupUsed(pc
)))
2937 lock_page_cgroup(pc
);
2939 memcg
= pc
->mem_cgroup
;
2941 if (!PageCgroupUsed(pc
))
2944 anon
= PageAnon(page
);
2947 case MEM_CGROUP_CHARGE_TYPE_ANON
:
2949 * Generally PageAnon tells if it's the anon statistics to be
2950 * updated; but sometimes e.g. mem_cgroup_uncharge_page() is
2951 * used before page reached the stage of being marked PageAnon.
2955 case MEM_CGROUP_CHARGE_TYPE_DROP
:
2956 /* See mem_cgroup_prepare_migration() */
2957 if (page_mapped(page
) || PageCgroupMigration(pc
))
2960 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT
:
2961 if (!PageAnon(page
)) { /* Shared memory */
2962 if (page
->mapping
&& !page_is_file_cache(page
))
2964 } else if (page_mapped(page
)) /* Anon */
2971 mem_cgroup_charge_statistics(memcg
, anon
, -nr_pages
);
2973 ClearPageCgroupUsed(pc
);
2975 * pc->mem_cgroup is not cleared here. It will be accessed when it's
2976 * freed from LRU. This is safe because uncharged page is expected not
2977 * to be reused (freed soon). Exception is SwapCache, it's handled by
2978 * special functions.
2981 unlock_page_cgroup(pc
);
2983 * even after unlock, we have memcg->res.usage here and this memcg
2984 * will never be freed.
2986 memcg_check_events(memcg
, page
);
2987 if (do_swap_account
&& ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
) {
2988 mem_cgroup_swap_statistics(memcg
, true);
2989 mem_cgroup_get(memcg
);
2991 if (!mem_cgroup_is_root(memcg
))
2992 mem_cgroup_do_uncharge(memcg
, nr_pages
, ctype
);
2997 unlock_page_cgroup(pc
);
3001 void mem_cgroup_uncharge_page(struct page
*page
)
3004 if (page_mapped(page
))
3006 VM_BUG_ON(page
->mapping
&& !PageAnon(page
));
3007 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_ANON
);
3010 void mem_cgroup_uncharge_cache_page(struct page
*page
)
3012 VM_BUG_ON(page_mapped(page
));
3013 VM_BUG_ON(page
->mapping
);
3014 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_CACHE
);
3018 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
3019 * In that cases, pages are freed continuously and we can expect pages
3020 * are in the same memcg. All these calls itself limits the number of
3021 * pages freed at once, then uncharge_start/end() is called properly.
3022 * This may be called prural(2) times in a context,
3025 void mem_cgroup_uncharge_start(void)
3027 current
->memcg_batch
.do_batch
++;
3028 /* We can do nest. */
3029 if (current
->memcg_batch
.do_batch
== 1) {
3030 current
->memcg_batch
.memcg
= NULL
;
3031 current
->memcg_batch
.nr_pages
= 0;
3032 current
->memcg_batch
.memsw_nr_pages
= 0;
3036 void mem_cgroup_uncharge_end(void)
3038 struct memcg_batch_info
*batch
= ¤t
->memcg_batch
;
3040 if (!batch
->do_batch
)
3044 if (batch
->do_batch
) /* If stacked, do nothing. */
3050 * This "batch->memcg" is valid without any css_get/put etc...
3051 * bacause we hide charges behind us.
3053 if (batch
->nr_pages
)
3054 res_counter_uncharge(&batch
->memcg
->res
,
3055 batch
->nr_pages
* PAGE_SIZE
);
3056 if (batch
->memsw_nr_pages
)
3057 res_counter_uncharge(&batch
->memcg
->memsw
,
3058 batch
->memsw_nr_pages
* PAGE_SIZE
);
3059 memcg_oom_recover(batch
->memcg
);
3060 /* forget this pointer (for sanity check) */
3061 batch
->memcg
= NULL
;
3066 * called after __delete_from_swap_cache() and drop "page" account.
3067 * memcg information is recorded to swap_cgroup of "ent"
3070 mem_cgroup_uncharge_swapcache(struct page
*page
, swp_entry_t ent
, bool swapout
)
3072 struct mem_cgroup
*memcg
;
3073 int ctype
= MEM_CGROUP_CHARGE_TYPE_SWAPOUT
;
3075 if (!swapout
) /* this was a swap cache but the swap is unused ! */
3076 ctype
= MEM_CGROUP_CHARGE_TYPE_DROP
;
3078 memcg
= __mem_cgroup_uncharge_common(page
, ctype
);
3081 * record memcg information, if swapout && memcg != NULL,
3082 * mem_cgroup_get() was called in uncharge().
3084 if (do_swap_account
&& swapout
&& memcg
)
3085 swap_cgroup_record(ent
, css_id(&memcg
->css
));
3089 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3091 * called from swap_entry_free(). remove record in swap_cgroup and
3092 * uncharge "memsw" account.
3094 void mem_cgroup_uncharge_swap(swp_entry_t ent
)
3096 struct mem_cgroup
*memcg
;
3099 if (!do_swap_account
)
3102 id
= swap_cgroup_record(ent
, 0);
3104 memcg
= mem_cgroup_lookup(id
);
3107 * We uncharge this because swap is freed.
3108 * This memcg can be obsolete one. We avoid calling css_tryget
3110 if (!mem_cgroup_is_root(memcg
))
3111 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
3112 mem_cgroup_swap_statistics(memcg
, false);
3113 mem_cgroup_put(memcg
);
3119 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3120 * @entry: swap entry to be moved
3121 * @from: mem_cgroup which the entry is moved from
3122 * @to: mem_cgroup which the entry is moved to
3124 * It succeeds only when the swap_cgroup's record for this entry is the same
3125 * as the mem_cgroup's id of @from.
3127 * Returns 0 on success, -EINVAL on failure.
3129 * The caller must have charged to @to, IOW, called res_counter_charge() about
3130 * both res and memsw, and called css_get().
3132 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
3133 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
3135 unsigned short old_id
, new_id
;
3137 old_id
= css_id(&from
->css
);
3138 new_id
= css_id(&to
->css
);
3140 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
3141 mem_cgroup_swap_statistics(from
, false);
3142 mem_cgroup_swap_statistics(to
, true);
3144 * This function is only called from task migration context now.
3145 * It postpones res_counter and refcount handling till the end
3146 * of task migration(mem_cgroup_clear_mc()) for performance
3147 * improvement. But we cannot postpone mem_cgroup_get(to)
3148 * because if the process that has been moved to @to does
3149 * swap-in, the refcount of @to might be decreased to 0.
3157 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
3158 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
3165 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3168 int mem_cgroup_prepare_migration(struct page
*page
,
3169 struct page
*newpage
, struct mem_cgroup
**memcgp
, gfp_t gfp_mask
)
3171 struct mem_cgroup
*memcg
= NULL
;
3172 struct page_cgroup
*pc
;
3173 enum charge_type ctype
;
3178 VM_BUG_ON(PageTransHuge(page
));
3179 if (mem_cgroup_disabled())
3182 pc
= lookup_page_cgroup(page
);
3183 lock_page_cgroup(pc
);
3184 if (PageCgroupUsed(pc
)) {
3185 memcg
= pc
->mem_cgroup
;
3186 css_get(&memcg
->css
);
3188 * At migrating an anonymous page, its mapcount goes down
3189 * to 0 and uncharge() will be called. But, even if it's fully
3190 * unmapped, migration may fail and this page has to be
3191 * charged again. We set MIGRATION flag here and delay uncharge
3192 * until end_migration() is called
3194 * Corner Case Thinking
3196 * When the old page was mapped as Anon and it's unmap-and-freed
3197 * while migration was ongoing.
3198 * If unmap finds the old page, uncharge() of it will be delayed
3199 * until end_migration(). If unmap finds a new page, it's
3200 * uncharged when it make mapcount to be 1->0. If unmap code
3201 * finds swap_migration_entry, the new page will not be mapped
3202 * and end_migration() will find it(mapcount==0).
3205 * When the old page was mapped but migraion fails, the kernel
3206 * remaps it. A charge for it is kept by MIGRATION flag even
3207 * if mapcount goes down to 0. We can do remap successfully
3208 * without charging it again.
3211 * The "old" page is under lock_page() until the end of
3212 * migration, so, the old page itself will not be swapped-out.
3213 * If the new page is swapped out before end_migraton, our
3214 * hook to usual swap-out path will catch the event.
3217 SetPageCgroupMigration(pc
);
3219 unlock_page_cgroup(pc
);
3221 * If the page is not charged at this point,
3228 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, 1, memcgp
, false);
3229 css_put(&memcg
->css
);/* drop extra refcnt */
3231 if (PageAnon(page
)) {
3232 lock_page_cgroup(pc
);
3233 ClearPageCgroupMigration(pc
);
3234 unlock_page_cgroup(pc
);
3236 * The old page may be fully unmapped while we kept it.
3238 mem_cgroup_uncharge_page(page
);
3240 /* we'll need to revisit this error code (we have -EINTR) */
3244 * We charge new page before it's used/mapped. So, even if unlock_page()
3245 * is called before end_migration, we can catch all events on this new
3246 * page. In the case new page is migrated but not remapped, new page's
3247 * mapcount will be finally 0 and we call uncharge in end_migration().
3250 ctype
= MEM_CGROUP_CHARGE_TYPE_ANON
;
3251 else if (page_is_file_cache(page
))
3252 ctype
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3254 ctype
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
3255 __mem_cgroup_commit_charge(memcg
, newpage
, 1, ctype
, false);
3259 /* remove redundant charge if migration failed*/
3260 void mem_cgroup_end_migration(struct mem_cgroup
*memcg
,
3261 struct page
*oldpage
, struct page
*newpage
, bool migration_ok
)
3263 struct page
*used
, *unused
;
3264 struct page_cgroup
*pc
;
3269 /* blocks rmdir() */
3270 cgroup_exclude_rmdir(&memcg
->css
);
3271 if (!migration_ok
) {
3279 * We disallowed uncharge of pages under migration because mapcount
3280 * of the page goes down to zero, temporarly.
3281 * Clear the flag and check the page should be charged.
3283 pc
= lookup_page_cgroup(oldpage
);
3284 lock_page_cgroup(pc
);
3285 ClearPageCgroupMigration(pc
);
3286 unlock_page_cgroup(pc
);
3287 anon
= PageAnon(used
);
3288 __mem_cgroup_uncharge_common(unused
,
3289 anon
? MEM_CGROUP_CHARGE_TYPE_ANON
3290 : MEM_CGROUP_CHARGE_TYPE_CACHE
);
3293 * If a page is a file cache, radix-tree replacement is very atomic
3294 * and we can skip this check. When it was an Anon page, its mapcount
3295 * goes down to 0. But because we added MIGRATION flage, it's not
3296 * uncharged yet. There are several case but page->mapcount check
3297 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3298 * check. (see prepare_charge() also)
3301 mem_cgroup_uncharge_page(used
);
3303 * At migration, we may charge account against cgroup which has no
3305 * So, rmdir()->pre_destroy() can be called while we do this charge.
3306 * In that case, we need to call pre_destroy() again. check it here.
3308 cgroup_release_and_wakeup_rmdir(&memcg
->css
);
3312 * At replace page cache, newpage is not under any memcg but it's on
3313 * LRU. So, this function doesn't touch res_counter but handles LRU
3314 * in correct way. Both pages are locked so we cannot race with uncharge.
3316 void mem_cgroup_replace_page_cache(struct page
*oldpage
,
3317 struct page
*newpage
)
3319 struct mem_cgroup
*memcg
= NULL
;
3320 struct page_cgroup
*pc
;
3321 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3323 if (mem_cgroup_disabled())
3326 pc
= lookup_page_cgroup(oldpage
);
3327 /* fix accounting on old pages */
3328 lock_page_cgroup(pc
);
3329 if (PageCgroupUsed(pc
)) {
3330 memcg
= pc
->mem_cgroup
;
3331 mem_cgroup_charge_statistics(memcg
, false, -1);
3332 ClearPageCgroupUsed(pc
);
3334 unlock_page_cgroup(pc
);
3337 * When called from shmem_replace_page(), in some cases the
3338 * oldpage has already been charged, and in some cases not.
3343 if (PageSwapBacked(oldpage
))
3344 type
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
3347 * Even if newpage->mapping was NULL before starting replacement,
3348 * the newpage may be on LRU(or pagevec for LRU) already. We lock
3349 * LRU while we overwrite pc->mem_cgroup.
3351 __mem_cgroup_commit_charge(memcg
, newpage
, 1, type
, true);
3354 #ifdef CONFIG_DEBUG_VM
3355 static struct page_cgroup
*lookup_page_cgroup_used(struct page
*page
)
3357 struct page_cgroup
*pc
;
3359 pc
= lookup_page_cgroup(page
);
3361 * Can be NULL while feeding pages into the page allocator for
3362 * the first time, i.e. during boot or memory hotplug;
3363 * or when mem_cgroup_disabled().
3365 if (likely(pc
) && PageCgroupUsed(pc
))
3370 bool mem_cgroup_bad_page_check(struct page
*page
)
3372 if (mem_cgroup_disabled())
3375 return lookup_page_cgroup_used(page
) != NULL
;
3378 void mem_cgroup_print_bad_page(struct page
*page
)
3380 struct page_cgroup
*pc
;
3382 pc
= lookup_page_cgroup_used(page
);
3384 printk(KERN_ALERT
"pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
3385 pc
, pc
->flags
, pc
->mem_cgroup
);
3390 static DEFINE_MUTEX(set_limit_mutex
);
3392 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
3393 unsigned long long val
)
3396 u64 memswlimit
, memlimit
;
3398 int children
= mem_cgroup_count_children(memcg
);
3399 u64 curusage
, oldusage
;
3403 * For keeping hierarchical_reclaim simple, how long we should retry
3404 * is depends on callers. We set our retry-count to be function
3405 * of # of children which we should visit in this loop.
3407 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
* children
;
3409 oldusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3412 while (retry_count
) {
3413 if (signal_pending(current
)) {
3418 * Rather than hide all in some function, I do this in
3419 * open coded manner. You see what this really does.
3420 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
3422 mutex_lock(&set_limit_mutex
);
3423 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3424 if (memswlimit
< val
) {
3426 mutex_unlock(&set_limit_mutex
);
3430 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3434 ret
= res_counter_set_limit(&memcg
->res
, val
);
3436 if (memswlimit
== val
)
3437 memcg
->memsw_is_minimum
= true;
3439 memcg
->memsw_is_minimum
= false;
3441 mutex_unlock(&set_limit_mutex
);
3446 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3447 MEM_CGROUP_RECLAIM_SHRINK
);
3448 curusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3449 /* Usage is reduced ? */
3450 if (curusage
>= oldusage
)
3453 oldusage
= curusage
;
3455 if (!ret
&& enlarge
)
3456 memcg_oom_recover(memcg
);
3461 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
3462 unsigned long long val
)
3465 u64 memlimit
, memswlimit
, oldusage
, curusage
;
3466 int children
= mem_cgroup_count_children(memcg
);
3470 /* see mem_cgroup_resize_res_limit */
3471 retry_count
= children
* MEM_CGROUP_RECLAIM_RETRIES
;
3472 oldusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3473 while (retry_count
) {
3474 if (signal_pending(current
)) {
3479 * Rather than hide all in some function, I do this in
3480 * open coded manner. You see what this really does.
3481 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
3483 mutex_lock(&set_limit_mutex
);
3484 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3485 if (memlimit
> val
) {
3487 mutex_unlock(&set_limit_mutex
);
3490 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3491 if (memswlimit
< val
)
3493 ret
= res_counter_set_limit(&memcg
->memsw
, val
);
3495 if (memlimit
== val
)
3496 memcg
->memsw_is_minimum
= true;
3498 memcg
->memsw_is_minimum
= false;
3500 mutex_unlock(&set_limit_mutex
);
3505 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3506 MEM_CGROUP_RECLAIM_NOSWAP
|
3507 MEM_CGROUP_RECLAIM_SHRINK
);
3508 curusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3509 /* Usage is reduced ? */
3510 if (curusage
>= oldusage
)
3513 oldusage
= curusage
;
3515 if (!ret
&& enlarge
)
3516 memcg_oom_recover(memcg
);
3520 unsigned long mem_cgroup_soft_limit_reclaim(struct zone
*zone
, int order
,
3522 unsigned long *total_scanned
)
3524 unsigned long nr_reclaimed
= 0;
3525 struct mem_cgroup_per_zone
*mz
, *next_mz
= NULL
;
3526 unsigned long reclaimed
;
3528 struct mem_cgroup_tree_per_zone
*mctz
;
3529 unsigned long long excess
;
3530 unsigned long nr_scanned
;
3535 mctz
= soft_limit_tree_node_zone(zone_to_nid(zone
), zone_idx(zone
));
3537 * This loop can run a while, specially if mem_cgroup's continuously
3538 * keep exceeding their soft limit and putting the system under
3545 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
3550 reclaimed
= mem_cgroup_soft_reclaim(mz
->memcg
, zone
,
3551 gfp_mask
, &nr_scanned
);
3552 nr_reclaimed
+= reclaimed
;
3553 *total_scanned
+= nr_scanned
;
3554 spin_lock(&mctz
->lock
);
3557 * If we failed to reclaim anything from this memory cgroup
3558 * it is time to move on to the next cgroup
3564 * Loop until we find yet another one.
3566 * By the time we get the soft_limit lock
3567 * again, someone might have aded the
3568 * group back on the RB tree. Iterate to
3569 * make sure we get a different mem.
3570 * mem_cgroup_largest_soft_limit_node returns
3571 * NULL if no other cgroup is present on
3575 __mem_cgroup_largest_soft_limit_node(mctz
);
3577 css_put(&next_mz
->memcg
->css
);
3578 else /* next_mz == NULL or other memcg */
3582 __mem_cgroup_remove_exceeded(mz
->memcg
, mz
, mctz
);
3583 excess
= res_counter_soft_limit_excess(&mz
->memcg
->res
);
3585 * One school of thought says that we should not add
3586 * back the node to the tree if reclaim returns 0.
3587 * But our reclaim could return 0, simply because due
3588 * to priority we are exposing a smaller subset of
3589 * memory to reclaim from. Consider this as a longer
3592 /* If excess == 0, no tree ops */
3593 __mem_cgroup_insert_exceeded(mz
->memcg
, mz
, mctz
, excess
);
3594 spin_unlock(&mctz
->lock
);
3595 css_put(&mz
->memcg
->css
);
3598 * Could not reclaim anything and there are no more
3599 * mem cgroups to try or we seem to be looping without
3600 * reclaiming anything.
3602 if (!nr_reclaimed
&&
3604 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
3606 } while (!nr_reclaimed
);
3608 css_put(&next_mz
->memcg
->css
);
3609 return nr_reclaimed
;
3613 * This routine traverse page_cgroup in given list and drop them all.
3614 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3616 static int mem_cgroup_force_empty_list(struct mem_cgroup
*memcg
,
3617 int node
, int zid
, enum lru_list lru
)
3619 struct mem_cgroup_per_zone
*mz
;
3620 unsigned long flags
, loop
;
3621 struct list_head
*list
;
3626 zone
= &NODE_DATA(node
)->node_zones
[zid
];
3627 mz
= mem_cgroup_zoneinfo(memcg
, node
, zid
);
3628 list
= &mz
->lruvec
.lists
[lru
];
3630 loop
= mz
->lru_size
[lru
];
3631 /* give some margin against EBUSY etc...*/
3635 struct page_cgroup
*pc
;
3639 spin_lock_irqsave(&zone
->lru_lock
, flags
);
3640 if (list_empty(list
)) {
3641 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3644 page
= list_entry(list
->prev
, struct page
, lru
);
3646 list_move(&page
->lru
, list
);
3648 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3651 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3653 pc
= lookup_page_cgroup(page
);
3655 ret
= mem_cgroup_move_parent(page
, pc
, memcg
, GFP_KERNEL
);
3656 if (ret
== -ENOMEM
|| ret
== -EINTR
)
3659 if (ret
== -EBUSY
|| ret
== -EINVAL
) {
3660 /* found lock contention or "pc" is obsolete. */
3667 if (!ret
&& !list_empty(list
))
3673 * make mem_cgroup's charge to be 0 if there is no task.
3674 * This enables deleting this mem_cgroup.
3676 static int mem_cgroup_force_empty(struct mem_cgroup
*memcg
, bool free_all
)
3679 int node
, zid
, shrink
;
3680 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
3681 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
3683 css_get(&memcg
->css
);
3686 /* should free all ? */
3692 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
))
3694 /* This is for making all *used* pages to be on LRU. */
3695 lru_add_drain_all();
3696 drain_all_stock_sync(memcg
);
3698 mem_cgroup_start_move(memcg
);
3699 for_each_node_state(node
, N_HIGH_MEMORY
) {
3700 for (zid
= 0; !ret
&& zid
< MAX_NR_ZONES
; zid
++) {
3703 ret
= mem_cgroup_force_empty_list(memcg
,
3712 mem_cgroup_end_move(memcg
);
3713 memcg_oom_recover(memcg
);
3714 /* it seems parent cgroup doesn't have enough mem */
3718 /* "ret" should also be checked to ensure all lists are empty. */
3719 } while (res_counter_read_u64(&memcg
->res
, RES_USAGE
) > 0 || ret
);
3721 css_put(&memcg
->css
);
3725 /* returns EBUSY if there is a task or if we come here twice. */
3726 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
) || shrink
) {
3730 /* we call try-to-free pages for make this cgroup empty */
3731 lru_add_drain_all();
3732 /* try to free all pages in this cgroup */
3734 while (nr_retries
&& res_counter_read_u64(&memcg
->res
, RES_USAGE
) > 0) {
3737 if (signal_pending(current
)) {
3741 progress
= try_to_free_mem_cgroup_pages(memcg
, GFP_KERNEL
,
3745 /* maybe some writeback is necessary */
3746 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3751 /* try move_account...there may be some *locked* pages. */
3755 static int mem_cgroup_force_empty_write(struct cgroup
*cont
, unsigned int event
)
3757 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont
), true);
3761 static u64
mem_cgroup_hierarchy_read(struct cgroup
*cont
, struct cftype
*cft
)
3763 return mem_cgroup_from_cont(cont
)->use_hierarchy
;
3766 static int mem_cgroup_hierarchy_write(struct cgroup
*cont
, struct cftype
*cft
,
3770 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3771 struct cgroup
*parent
= cont
->parent
;
3772 struct mem_cgroup
*parent_memcg
= NULL
;
3775 parent_memcg
= mem_cgroup_from_cont(parent
);
3779 * If parent's use_hierarchy is set, we can't make any modifications
3780 * in the child subtrees. If it is unset, then the change can
3781 * occur, provided the current cgroup has no children.
3783 * For the root cgroup, parent_mem is NULL, we allow value to be
3784 * set if there are no children.
3786 if ((!parent_memcg
|| !parent_memcg
->use_hierarchy
) &&
3787 (val
== 1 || val
== 0)) {
3788 if (list_empty(&cont
->children
))
3789 memcg
->use_hierarchy
= val
;
3800 static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup
*memcg
,
3801 enum mem_cgroup_stat_index idx
)
3803 struct mem_cgroup
*iter
;
3806 /* Per-cpu values can be negative, use a signed accumulator */
3807 for_each_mem_cgroup_tree(iter
, memcg
)
3808 val
+= mem_cgroup_read_stat(iter
, idx
);
3810 if (val
< 0) /* race ? */
3815 static inline u64
mem_cgroup_usage(struct mem_cgroup
*memcg
, bool swap
)
3819 if (!mem_cgroup_is_root(memcg
)) {
3821 return res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3823 return res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3826 val
= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_CACHE
);
3827 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_RSS
);
3830 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_SWAP
);
3832 return val
<< PAGE_SHIFT
;
3835 static ssize_t
mem_cgroup_read(struct cgroup
*cont
, struct cftype
*cft
,
3836 struct file
*file
, char __user
*buf
,
3837 size_t nbytes
, loff_t
*ppos
)
3839 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3842 int type
, name
, len
;
3844 type
= MEMFILE_TYPE(cft
->private);
3845 name
= MEMFILE_ATTR(cft
->private);
3847 if (!do_swap_account
&& type
== _MEMSWAP
)
3852 if (name
== RES_USAGE
)
3853 val
= mem_cgroup_usage(memcg
, false);
3855 val
= res_counter_read_u64(&memcg
->res
, name
);
3858 if (name
== RES_USAGE
)
3859 val
= mem_cgroup_usage(memcg
, true);
3861 val
= res_counter_read_u64(&memcg
->memsw
, name
);
3867 len
= scnprintf(str
, sizeof(str
), "%llu\n", (unsigned long long)val
);
3868 return simple_read_from_buffer(buf
, nbytes
, ppos
, str
, len
);
3871 * The user of this function is...
3874 static int mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
3877 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3879 unsigned long long val
;
3882 type
= MEMFILE_TYPE(cft
->private);
3883 name
= MEMFILE_ATTR(cft
->private);
3885 if (!do_swap_account
&& type
== _MEMSWAP
)
3890 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
3894 /* This function does all necessary parse...reuse it */
3895 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3899 ret
= mem_cgroup_resize_limit(memcg
, val
);
3901 ret
= mem_cgroup_resize_memsw_limit(memcg
, val
);
3903 case RES_SOFT_LIMIT
:
3904 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3908 * For memsw, soft limits are hard to implement in terms
3909 * of semantics, for now, we support soft limits for
3910 * control without swap
3913 ret
= res_counter_set_soft_limit(&memcg
->res
, val
);
3918 ret
= -EINVAL
; /* should be BUG() ? */
3924 static void memcg_get_hierarchical_limit(struct mem_cgroup
*memcg
,
3925 unsigned long long *mem_limit
, unsigned long long *memsw_limit
)
3927 struct cgroup
*cgroup
;
3928 unsigned long long min_limit
, min_memsw_limit
, tmp
;
3930 min_limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3931 min_memsw_limit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3932 cgroup
= memcg
->css
.cgroup
;
3933 if (!memcg
->use_hierarchy
)
3936 while (cgroup
->parent
) {
3937 cgroup
= cgroup
->parent
;
3938 memcg
= mem_cgroup_from_cont(cgroup
);
3939 if (!memcg
->use_hierarchy
)
3941 tmp
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3942 min_limit
= min(min_limit
, tmp
);
3943 tmp
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3944 min_memsw_limit
= min(min_memsw_limit
, tmp
);
3947 *mem_limit
= min_limit
;
3948 *memsw_limit
= min_memsw_limit
;
3951 static int mem_cgroup_reset(struct cgroup
*cont
, unsigned int event
)
3953 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3956 type
= MEMFILE_TYPE(event
);
3957 name
= MEMFILE_ATTR(event
);
3959 if (!do_swap_account
&& type
== _MEMSWAP
)
3965 res_counter_reset_max(&memcg
->res
);
3967 res_counter_reset_max(&memcg
->memsw
);
3971 res_counter_reset_failcnt(&memcg
->res
);
3973 res_counter_reset_failcnt(&memcg
->memsw
);
3980 static u64
mem_cgroup_move_charge_read(struct cgroup
*cgrp
,
3983 return mem_cgroup_from_cont(cgrp
)->move_charge_at_immigrate
;
3987 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
3988 struct cftype
*cft
, u64 val
)
3990 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3992 if (val
>= (1 << NR_MOVE_TYPE
))
3995 * We check this value several times in both in can_attach() and
3996 * attach(), so we need cgroup lock to prevent this value from being
4000 memcg
->move_charge_at_immigrate
= val
;
4006 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4007 struct cftype
*cft
, u64 val
)
4014 static int mem_control_numa_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
4018 unsigned long total_nr
, file_nr
, anon_nr
, unevictable_nr
;
4019 unsigned long node_nr
;
4020 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4022 total_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL
);
4023 seq_printf(m
, "total=%lu", total_nr
);
4024 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4025 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL
);
4026 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4030 file_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL_FILE
);
4031 seq_printf(m
, "file=%lu", file_nr
);
4032 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4033 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4035 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4039 anon_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL_ANON
);
4040 seq_printf(m
, "anon=%lu", anon_nr
);
4041 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4042 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4044 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4048 unevictable_nr
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_UNEVICTABLE
));
4049 seq_printf(m
, "unevictable=%lu", unevictable_nr
);
4050 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4051 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4052 BIT(LRU_UNEVICTABLE
));
4053 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4058 #endif /* CONFIG_NUMA */
4060 static const char * const mem_cgroup_lru_names
[] = {
4068 static inline void mem_cgroup_lru_names_not_uptodate(void)
4070 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names
) != NR_LRU_LISTS
);
4073 static int mem_control_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
4076 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4077 struct mem_cgroup
*mi
;
4080 for (i
= 0; i
< MEM_CGROUP_STAT_NSTATS
; i
++) {
4081 if (i
== MEM_CGROUP_STAT_SWAP
&& !do_swap_account
)
4083 seq_printf(m
, "%s %ld\n", mem_cgroup_stat_names
[i
],
4084 mem_cgroup_read_stat(memcg
, i
) * PAGE_SIZE
);
4087 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++)
4088 seq_printf(m
, "%s %lu\n", mem_cgroup_events_names
[i
],
4089 mem_cgroup_read_events(memcg
, i
));
4091 for (i
= 0; i
< NR_LRU_LISTS
; i
++)
4092 seq_printf(m
, "%s %lu\n", mem_cgroup_lru_names
[i
],
4093 mem_cgroup_nr_lru_pages(memcg
, BIT(i
)) * PAGE_SIZE
);
4095 /* Hierarchical information */
4097 unsigned long long limit
, memsw_limit
;
4098 memcg_get_hierarchical_limit(memcg
, &limit
, &memsw_limit
);
4099 seq_printf(m
, "hierarchical_memory_limit %llu\n", limit
);
4100 if (do_swap_account
)
4101 seq_printf(m
, "hierarchical_memsw_limit %llu\n",
4105 for (i
= 0; i
< MEM_CGROUP_STAT_NSTATS
; i
++) {
4108 if (i
== MEM_CGROUP_STAT_SWAP
&& !do_swap_account
)
4110 for_each_mem_cgroup_tree(mi
, memcg
)
4111 val
+= mem_cgroup_read_stat(mi
, i
) * PAGE_SIZE
;
4112 seq_printf(m
, "total_%s %lld\n", mem_cgroup_stat_names
[i
], val
);
4115 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
4116 unsigned long long val
= 0;
4118 for_each_mem_cgroup_tree(mi
, memcg
)
4119 val
+= mem_cgroup_read_events(mi
, i
);
4120 seq_printf(m
, "total_%s %llu\n",
4121 mem_cgroup_events_names
[i
], val
);
4124 for (i
= 0; i
< NR_LRU_LISTS
; i
++) {
4125 unsigned long long val
= 0;
4127 for_each_mem_cgroup_tree(mi
, memcg
)
4128 val
+= mem_cgroup_nr_lru_pages(mi
, BIT(i
)) * PAGE_SIZE
;
4129 seq_printf(m
, "total_%s %llu\n", mem_cgroup_lru_names
[i
], val
);
4132 #ifdef CONFIG_DEBUG_VM
4135 struct mem_cgroup_per_zone
*mz
;
4136 struct zone_reclaim_stat
*rstat
;
4137 unsigned long recent_rotated
[2] = {0, 0};
4138 unsigned long recent_scanned
[2] = {0, 0};
4140 for_each_online_node(nid
)
4141 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
4142 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
4143 rstat
= &mz
->lruvec
.reclaim_stat
;
4145 recent_rotated
[0] += rstat
->recent_rotated
[0];
4146 recent_rotated
[1] += rstat
->recent_rotated
[1];
4147 recent_scanned
[0] += rstat
->recent_scanned
[0];
4148 recent_scanned
[1] += rstat
->recent_scanned
[1];
4150 seq_printf(m
, "recent_rotated_anon %lu\n", recent_rotated
[0]);
4151 seq_printf(m
, "recent_rotated_file %lu\n", recent_rotated
[1]);
4152 seq_printf(m
, "recent_scanned_anon %lu\n", recent_scanned
[0]);
4153 seq_printf(m
, "recent_scanned_file %lu\n", recent_scanned
[1]);
4160 static u64
mem_cgroup_swappiness_read(struct cgroup
*cgrp
, struct cftype
*cft
)
4162 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4164 return mem_cgroup_swappiness(memcg
);
4167 static int mem_cgroup_swappiness_write(struct cgroup
*cgrp
, struct cftype
*cft
,
4170 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4171 struct mem_cgroup
*parent
;
4176 if (cgrp
->parent
== NULL
)
4179 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4183 /* If under hierarchy, only empty-root can set this value */
4184 if ((parent
->use_hierarchy
) ||
4185 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4190 memcg
->swappiness
= val
;
4197 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
4199 struct mem_cgroup_threshold_ary
*t
;
4205 t
= rcu_dereference(memcg
->thresholds
.primary
);
4207 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
4212 usage
= mem_cgroup_usage(memcg
, swap
);
4215 * current_threshold points to threshold just below or equal to usage.
4216 * If it's not true, a threshold was crossed after last
4217 * call of __mem_cgroup_threshold().
4219 i
= t
->current_threshold
;
4222 * Iterate backward over array of thresholds starting from
4223 * current_threshold and check if a threshold is crossed.
4224 * If none of thresholds below usage is crossed, we read
4225 * only one element of the array here.
4227 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
4228 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4230 /* i = current_threshold + 1 */
4234 * Iterate forward over array of thresholds starting from
4235 * current_threshold+1 and check if a threshold is crossed.
4236 * If none of thresholds above usage is crossed, we read
4237 * only one element of the array here.
4239 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
4240 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4242 /* Update current_threshold */
4243 t
->current_threshold
= i
- 1;
4248 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
4251 __mem_cgroup_threshold(memcg
, false);
4252 if (do_swap_account
)
4253 __mem_cgroup_threshold(memcg
, true);
4255 memcg
= parent_mem_cgroup(memcg
);
4259 static int compare_thresholds(const void *a
, const void *b
)
4261 const struct mem_cgroup_threshold
*_a
= a
;
4262 const struct mem_cgroup_threshold
*_b
= b
;
4264 return _a
->threshold
- _b
->threshold
;
4267 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*memcg
)
4269 struct mem_cgroup_eventfd_list
*ev
;
4271 list_for_each_entry(ev
, &memcg
->oom_notify
, list
)
4272 eventfd_signal(ev
->eventfd
, 1);
4276 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
)
4278 struct mem_cgroup
*iter
;
4280 for_each_mem_cgroup_tree(iter
, memcg
)
4281 mem_cgroup_oom_notify_cb(iter
);
4284 static int mem_cgroup_usage_register_event(struct cgroup
*cgrp
,
4285 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4287 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4288 struct mem_cgroup_thresholds
*thresholds
;
4289 struct mem_cgroup_threshold_ary
*new;
4290 int type
= MEMFILE_TYPE(cft
->private);
4291 u64 threshold
, usage
;
4294 ret
= res_counter_memparse_write_strategy(args
, &threshold
);
4298 mutex_lock(&memcg
->thresholds_lock
);
4301 thresholds
= &memcg
->thresholds
;
4302 else if (type
== _MEMSWAP
)
4303 thresholds
= &memcg
->memsw_thresholds
;
4307 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4309 /* Check if a threshold crossed before adding a new one */
4310 if (thresholds
->primary
)
4311 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4313 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
4315 /* Allocate memory for new array of thresholds */
4316 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
4324 /* Copy thresholds (if any) to new array */
4325 if (thresholds
->primary
) {
4326 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
4327 sizeof(struct mem_cgroup_threshold
));
4330 /* Add new threshold */
4331 new->entries
[size
- 1].eventfd
= eventfd
;
4332 new->entries
[size
- 1].threshold
= threshold
;
4334 /* Sort thresholds. Registering of new threshold isn't time-critical */
4335 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
4336 compare_thresholds
, NULL
);
4338 /* Find current threshold */
4339 new->current_threshold
= -1;
4340 for (i
= 0; i
< size
; i
++) {
4341 if (new->entries
[i
].threshold
<= usage
) {
4343 * new->current_threshold will not be used until
4344 * rcu_assign_pointer(), so it's safe to increment
4347 ++new->current_threshold
;
4352 /* Free old spare buffer and save old primary buffer as spare */
4353 kfree(thresholds
->spare
);
4354 thresholds
->spare
= thresholds
->primary
;
4356 rcu_assign_pointer(thresholds
->primary
, new);
4358 /* To be sure that nobody uses thresholds */
4362 mutex_unlock(&memcg
->thresholds_lock
);
4367 static void mem_cgroup_usage_unregister_event(struct cgroup
*cgrp
,
4368 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4370 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4371 struct mem_cgroup_thresholds
*thresholds
;
4372 struct mem_cgroup_threshold_ary
*new;
4373 int type
= MEMFILE_TYPE(cft
->private);
4377 mutex_lock(&memcg
->thresholds_lock
);
4379 thresholds
= &memcg
->thresholds
;
4380 else if (type
== _MEMSWAP
)
4381 thresholds
= &memcg
->memsw_thresholds
;
4385 if (!thresholds
->primary
)
4388 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4390 /* Check if a threshold crossed before removing */
4391 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4393 /* Calculate new number of threshold */
4395 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
4396 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
4400 new = thresholds
->spare
;
4402 /* Set thresholds array to NULL if we don't have thresholds */
4411 /* Copy thresholds and find current threshold */
4412 new->current_threshold
= -1;
4413 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
4414 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
4417 new->entries
[j
] = thresholds
->primary
->entries
[i
];
4418 if (new->entries
[j
].threshold
<= usage
) {
4420 * new->current_threshold will not be used
4421 * until rcu_assign_pointer(), so it's safe to increment
4424 ++new->current_threshold
;
4430 /* Swap primary and spare array */
4431 thresholds
->spare
= thresholds
->primary
;
4432 /* If all events are unregistered, free the spare array */
4434 kfree(thresholds
->spare
);
4435 thresholds
->spare
= NULL
;
4438 rcu_assign_pointer(thresholds
->primary
, new);
4440 /* To be sure that nobody uses thresholds */
4443 mutex_unlock(&memcg
->thresholds_lock
);
4446 static int mem_cgroup_oom_register_event(struct cgroup
*cgrp
,
4447 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4449 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4450 struct mem_cgroup_eventfd_list
*event
;
4451 int type
= MEMFILE_TYPE(cft
->private);
4453 BUG_ON(type
!= _OOM_TYPE
);
4454 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
4458 spin_lock(&memcg_oom_lock
);
4460 event
->eventfd
= eventfd
;
4461 list_add(&event
->list
, &memcg
->oom_notify
);
4463 /* already in OOM ? */
4464 if (atomic_read(&memcg
->under_oom
))
4465 eventfd_signal(eventfd
, 1);
4466 spin_unlock(&memcg_oom_lock
);
4471 static void mem_cgroup_oom_unregister_event(struct cgroup
*cgrp
,
4472 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4474 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4475 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
4476 int type
= MEMFILE_TYPE(cft
->private);
4478 BUG_ON(type
!= _OOM_TYPE
);
4480 spin_lock(&memcg_oom_lock
);
4482 list_for_each_entry_safe(ev
, tmp
, &memcg
->oom_notify
, list
) {
4483 if (ev
->eventfd
== eventfd
) {
4484 list_del(&ev
->list
);
4489 spin_unlock(&memcg_oom_lock
);
4492 static int mem_cgroup_oom_control_read(struct cgroup
*cgrp
,
4493 struct cftype
*cft
, struct cgroup_map_cb
*cb
)
4495 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4497 cb
->fill(cb
, "oom_kill_disable", memcg
->oom_kill_disable
);
4499 if (atomic_read(&memcg
->under_oom
))
4500 cb
->fill(cb
, "under_oom", 1);
4502 cb
->fill(cb
, "under_oom", 0);
4506 static int mem_cgroup_oom_control_write(struct cgroup
*cgrp
,
4507 struct cftype
*cft
, u64 val
)
4509 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4510 struct mem_cgroup
*parent
;
4512 /* cannot set to root cgroup and only 0 and 1 are allowed */
4513 if (!cgrp
->parent
|| !((val
== 0) || (val
== 1)))
4516 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4519 /* oom-kill-disable is a flag for subhierarchy. */
4520 if ((parent
->use_hierarchy
) ||
4521 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4525 memcg
->oom_kill_disable
= val
;
4527 memcg_oom_recover(memcg
);
4532 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
4533 static int memcg_init_kmem(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
4535 return mem_cgroup_sockets_init(memcg
, ss
);
4538 static void kmem_cgroup_destroy(struct mem_cgroup
*memcg
)
4540 mem_cgroup_sockets_destroy(memcg
);
4543 static int memcg_init_kmem(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
4548 static void kmem_cgroup_destroy(struct mem_cgroup
*memcg
)
4553 static struct cftype mem_cgroup_files
[] = {
4555 .name
= "usage_in_bytes",
4556 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
4557 .read
= mem_cgroup_read
,
4558 .register_event
= mem_cgroup_usage_register_event
,
4559 .unregister_event
= mem_cgroup_usage_unregister_event
,
4562 .name
= "max_usage_in_bytes",
4563 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
4564 .trigger
= mem_cgroup_reset
,
4565 .read
= mem_cgroup_read
,
4568 .name
= "limit_in_bytes",
4569 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
4570 .write_string
= mem_cgroup_write
,
4571 .read
= mem_cgroup_read
,
4574 .name
= "soft_limit_in_bytes",
4575 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
4576 .write_string
= mem_cgroup_write
,
4577 .read
= mem_cgroup_read
,
4581 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
4582 .trigger
= mem_cgroup_reset
,
4583 .read
= mem_cgroup_read
,
4587 .read_seq_string
= mem_control_stat_show
,
4590 .name
= "force_empty",
4591 .trigger
= mem_cgroup_force_empty_write
,
4594 .name
= "use_hierarchy",
4595 .write_u64
= mem_cgroup_hierarchy_write
,
4596 .read_u64
= mem_cgroup_hierarchy_read
,
4599 .name
= "swappiness",
4600 .read_u64
= mem_cgroup_swappiness_read
,
4601 .write_u64
= mem_cgroup_swappiness_write
,
4604 .name
= "move_charge_at_immigrate",
4605 .read_u64
= mem_cgroup_move_charge_read
,
4606 .write_u64
= mem_cgroup_move_charge_write
,
4609 .name
= "oom_control",
4610 .read_map
= mem_cgroup_oom_control_read
,
4611 .write_u64
= mem_cgroup_oom_control_write
,
4612 .register_event
= mem_cgroup_oom_register_event
,
4613 .unregister_event
= mem_cgroup_oom_unregister_event
,
4614 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
4618 .name
= "numa_stat",
4619 .read_seq_string
= mem_control_numa_stat_show
,
4622 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4624 .name
= "memsw.usage_in_bytes",
4625 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
4626 .read
= mem_cgroup_read
,
4627 .register_event
= mem_cgroup_usage_register_event
,
4628 .unregister_event
= mem_cgroup_usage_unregister_event
,
4631 .name
= "memsw.max_usage_in_bytes",
4632 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
4633 .trigger
= mem_cgroup_reset
,
4634 .read
= mem_cgroup_read
,
4637 .name
= "memsw.limit_in_bytes",
4638 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
4639 .write_string
= mem_cgroup_write
,
4640 .read
= mem_cgroup_read
,
4643 .name
= "memsw.failcnt",
4644 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
4645 .trigger
= mem_cgroup_reset
,
4646 .read
= mem_cgroup_read
,
4649 { }, /* terminate */
4652 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4654 struct mem_cgroup_per_node
*pn
;
4655 struct mem_cgroup_per_zone
*mz
;
4656 int zone
, tmp
= node
;
4658 * This routine is called against possible nodes.
4659 * But it's BUG to call kmalloc() against offline node.
4661 * TODO: this routine can waste much memory for nodes which will
4662 * never be onlined. It's better to use memory hotplug callback
4665 if (!node_state(node
, N_NORMAL_MEMORY
))
4667 pn
= kzalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
4671 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4672 mz
= &pn
->zoneinfo
[zone
];
4673 lruvec_init(&mz
->lruvec
, &NODE_DATA(node
)->node_zones
[zone
]);
4674 mz
->usage_in_excess
= 0;
4675 mz
->on_tree
= false;
4678 memcg
->info
.nodeinfo
[node
] = pn
;
4682 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4684 kfree(memcg
->info
.nodeinfo
[node
]);
4687 static struct mem_cgroup
*mem_cgroup_alloc(void)
4689 struct mem_cgroup
*memcg
;
4690 int size
= sizeof(struct mem_cgroup
);
4692 /* Can be very big if MAX_NUMNODES is very big */
4693 if (size
< PAGE_SIZE
)
4694 memcg
= kzalloc(size
, GFP_KERNEL
);
4696 memcg
= vzalloc(size
);
4701 memcg
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4704 spin_lock_init(&memcg
->pcp_counter_lock
);
4708 if (size
< PAGE_SIZE
)
4716 * Helpers for freeing a kmalloc()ed/vzalloc()ed mem_cgroup by RCU,
4717 * but in process context. The work_freeing structure is overlaid
4718 * on the rcu_freeing structure, which itself is overlaid on memsw.
4720 static void free_work(struct work_struct
*work
)
4722 struct mem_cgroup
*memcg
;
4723 int size
= sizeof(struct mem_cgroup
);
4725 memcg
= container_of(work
, struct mem_cgroup
, work_freeing
);
4727 * We need to make sure that (at least for now), the jump label
4728 * destruction code runs outside of the cgroup lock. This is because
4729 * get_online_cpus(), which is called from the static_branch update,
4730 * can't be called inside the cgroup_lock. cpusets are the ones
4731 * enforcing this dependency, so if they ever change, we might as well.
4733 * schedule_work() will guarantee this happens. Be careful if you need
4734 * to move this code around, and make sure it is outside
4737 disarm_sock_keys(memcg
);
4738 if (size
< PAGE_SIZE
)
4744 static void free_rcu(struct rcu_head
*rcu_head
)
4746 struct mem_cgroup
*memcg
;
4748 memcg
= container_of(rcu_head
, struct mem_cgroup
, rcu_freeing
);
4749 INIT_WORK(&memcg
->work_freeing
, free_work
);
4750 schedule_work(&memcg
->work_freeing
);
4754 * At destroying mem_cgroup, references from swap_cgroup can remain.
4755 * (scanning all at force_empty is too costly...)
4757 * Instead of clearing all references at force_empty, we remember
4758 * the number of reference from swap_cgroup and free mem_cgroup when
4759 * it goes down to 0.
4761 * Removal of cgroup itself succeeds regardless of refs from swap.
4764 static void __mem_cgroup_free(struct mem_cgroup
*memcg
)
4768 mem_cgroup_remove_from_trees(memcg
);
4769 free_css_id(&mem_cgroup_subsys
, &memcg
->css
);
4772 free_mem_cgroup_per_zone_info(memcg
, node
);
4774 free_percpu(memcg
->stat
);
4775 call_rcu(&memcg
->rcu_freeing
, free_rcu
);
4778 static void mem_cgroup_get(struct mem_cgroup
*memcg
)
4780 atomic_inc(&memcg
->refcnt
);
4783 static void __mem_cgroup_put(struct mem_cgroup
*memcg
, int count
)
4785 if (atomic_sub_and_test(count
, &memcg
->refcnt
)) {
4786 struct mem_cgroup
*parent
= parent_mem_cgroup(memcg
);
4787 __mem_cgroup_free(memcg
);
4789 mem_cgroup_put(parent
);
4793 static void mem_cgroup_put(struct mem_cgroup
*memcg
)
4795 __mem_cgroup_put(memcg
, 1);
4799 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4801 struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*memcg
)
4803 if (!memcg
->res
.parent
)
4805 return mem_cgroup_from_res_counter(memcg
->res
.parent
, res
);
4807 EXPORT_SYMBOL(parent_mem_cgroup
);
4809 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4810 static void __init
enable_swap_cgroup(void)
4812 if (!mem_cgroup_disabled() && really_do_swap_account
)
4813 do_swap_account
= 1;
4816 static void __init
enable_swap_cgroup(void)
4821 static int mem_cgroup_soft_limit_tree_init(void)
4823 struct mem_cgroup_tree_per_node
*rtpn
;
4824 struct mem_cgroup_tree_per_zone
*rtpz
;
4825 int tmp
, node
, zone
;
4827 for_each_node(node
) {
4829 if (!node_state(node
, N_NORMAL_MEMORY
))
4831 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
, tmp
);
4835 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
4837 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4838 rtpz
= &rtpn
->rb_tree_per_zone
[zone
];
4839 rtpz
->rb_root
= RB_ROOT
;
4840 spin_lock_init(&rtpz
->lock
);
4846 for_each_node(node
) {
4847 if (!soft_limit_tree
.rb_tree_per_node
[node
])
4849 kfree(soft_limit_tree
.rb_tree_per_node
[node
]);
4850 soft_limit_tree
.rb_tree_per_node
[node
] = NULL
;
4856 static struct cgroup_subsys_state
* __ref
4857 mem_cgroup_create(struct cgroup
*cont
)
4859 struct mem_cgroup
*memcg
, *parent
;
4860 long error
= -ENOMEM
;
4863 memcg
= mem_cgroup_alloc();
4865 return ERR_PTR(error
);
4868 if (alloc_mem_cgroup_per_zone_info(memcg
, node
))
4872 if (cont
->parent
== NULL
) {
4874 enable_swap_cgroup();
4876 if (mem_cgroup_soft_limit_tree_init())
4878 root_mem_cgroup
= memcg
;
4879 for_each_possible_cpu(cpu
) {
4880 struct memcg_stock_pcp
*stock
=
4881 &per_cpu(memcg_stock
, cpu
);
4882 INIT_WORK(&stock
->work
, drain_local_stock
);
4884 hotcpu_notifier(memcg_cpu_hotplug_callback
, 0);
4886 parent
= mem_cgroup_from_cont(cont
->parent
);
4887 memcg
->use_hierarchy
= parent
->use_hierarchy
;
4888 memcg
->oom_kill_disable
= parent
->oom_kill_disable
;
4891 if (parent
&& parent
->use_hierarchy
) {
4892 res_counter_init(&memcg
->res
, &parent
->res
);
4893 res_counter_init(&memcg
->memsw
, &parent
->memsw
);
4895 * We increment refcnt of the parent to ensure that we can
4896 * safely access it on res_counter_charge/uncharge.
4897 * This refcnt will be decremented when freeing this
4898 * mem_cgroup(see mem_cgroup_put).
4900 mem_cgroup_get(parent
);
4902 res_counter_init(&memcg
->res
, NULL
);
4903 res_counter_init(&memcg
->memsw
, NULL
);
4905 memcg
->last_scanned_node
= MAX_NUMNODES
;
4906 INIT_LIST_HEAD(&memcg
->oom_notify
);
4909 memcg
->swappiness
= mem_cgroup_swappiness(parent
);
4910 atomic_set(&memcg
->refcnt
, 1);
4911 memcg
->move_charge_at_immigrate
= 0;
4912 mutex_init(&memcg
->thresholds_lock
);
4913 spin_lock_init(&memcg
->move_lock
);
4915 error
= memcg_init_kmem(memcg
, &mem_cgroup_subsys
);
4918 * We call put now because our (and parent's) refcnts
4919 * are already in place. mem_cgroup_put() will internally
4920 * call __mem_cgroup_free, so return directly
4922 mem_cgroup_put(memcg
);
4923 return ERR_PTR(error
);
4927 __mem_cgroup_free(memcg
);
4928 return ERR_PTR(error
);
4931 static int mem_cgroup_pre_destroy(struct cgroup
*cont
)
4933 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4935 return mem_cgroup_force_empty(memcg
, false);
4938 static void mem_cgroup_destroy(struct cgroup
*cont
)
4940 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4942 kmem_cgroup_destroy(memcg
);
4944 mem_cgroup_put(memcg
);
4948 /* Handlers for move charge at task migration. */
4949 #define PRECHARGE_COUNT_AT_ONCE 256
4950 static int mem_cgroup_do_precharge(unsigned long count
)
4953 int batch_count
= PRECHARGE_COUNT_AT_ONCE
;
4954 struct mem_cgroup
*memcg
= mc
.to
;
4956 if (mem_cgroup_is_root(memcg
)) {
4957 mc
.precharge
+= count
;
4958 /* we don't need css_get for root */
4961 /* try to charge at once */
4963 struct res_counter
*dummy
;
4965 * "memcg" cannot be under rmdir() because we've already checked
4966 * by cgroup_lock_live_cgroup() that it is not removed and we
4967 * are still under the same cgroup_mutex. So we can postpone
4970 if (res_counter_charge(&memcg
->res
, PAGE_SIZE
* count
, &dummy
))
4972 if (do_swap_account
&& res_counter_charge(&memcg
->memsw
,
4973 PAGE_SIZE
* count
, &dummy
)) {
4974 res_counter_uncharge(&memcg
->res
, PAGE_SIZE
* count
);
4977 mc
.precharge
+= count
;
4981 /* fall back to one by one charge */
4983 if (signal_pending(current
)) {
4987 if (!batch_count
--) {
4988 batch_count
= PRECHARGE_COUNT_AT_ONCE
;
4991 ret
= __mem_cgroup_try_charge(NULL
,
4992 GFP_KERNEL
, 1, &memcg
, false);
4994 /* mem_cgroup_clear_mc() will do uncharge later */
5002 * get_mctgt_type - get target type of moving charge
5003 * @vma: the vma the pte to be checked belongs
5004 * @addr: the address corresponding to the pte to be checked
5005 * @ptent: the pte to be checked
5006 * @target: the pointer the target page or swap ent will be stored(can be NULL)
5009 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5010 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5011 * move charge. if @target is not NULL, the page is stored in target->page
5012 * with extra refcnt got(Callers should handle it).
5013 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5014 * target for charge migration. if @target is not NULL, the entry is stored
5017 * Called with pte lock held.
5024 enum mc_target_type
{
5030 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
5031 unsigned long addr
, pte_t ptent
)
5033 struct page
*page
= vm_normal_page(vma
, addr
, ptent
);
5035 if (!page
|| !page_mapped(page
))
5037 if (PageAnon(page
)) {
5038 /* we don't move shared anon */
5041 } else if (!move_file())
5042 /* we ignore mapcount for file pages */
5044 if (!get_page_unless_zero(page
))
5051 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5052 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5054 struct page
*page
= NULL
;
5055 swp_entry_t ent
= pte_to_swp_entry(ptent
);
5057 if (!move_anon() || non_swap_entry(ent
))
5060 * Because lookup_swap_cache() updates some statistics counter,
5061 * we call find_get_page() with swapper_space directly.
5063 page
= find_get_page(&swapper_space
, ent
.val
);
5064 if (do_swap_account
)
5065 entry
->val
= ent
.val
;
5070 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5071 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5077 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
5078 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5080 struct page
*page
= NULL
;
5081 struct address_space
*mapping
;
5084 if (!vma
->vm_file
) /* anonymous vma */
5089 mapping
= vma
->vm_file
->f_mapping
;
5090 if (pte_none(ptent
))
5091 pgoff
= linear_page_index(vma
, addr
);
5092 else /* pte_file(ptent) is true */
5093 pgoff
= pte_to_pgoff(ptent
);
5095 /* page is moved even if it's not RSS of this task(page-faulted). */
5096 page
= find_get_page(mapping
, pgoff
);
5099 /* shmem/tmpfs may report page out on swap: account for that too. */
5100 if (radix_tree_exceptional_entry(page
)) {
5101 swp_entry_t swap
= radix_to_swp_entry(page
);
5102 if (do_swap_account
)
5104 page
= find_get_page(&swapper_space
, swap
.val
);
5110 static enum mc_target_type
get_mctgt_type(struct vm_area_struct
*vma
,
5111 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
5113 struct page
*page
= NULL
;
5114 struct page_cgroup
*pc
;
5115 enum mc_target_type ret
= MC_TARGET_NONE
;
5116 swp_entry_t ent
= { .val
= 0 };
5118 if (pte_present(ptent
))
5119 page
= mc_handle_present_pte(vma
, addr
, ptent
);
5120 else if (is_swap_pte(ptent
))
5121 page
= mc_handle_swap_pte(vma
, addr
, ptent
, &ent
);
5122 else if (pte_none(ptent
) || pte_file(ptent
))
5123 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
5125 if (!page
&& !ent
.val
)
5128 pc
= lookup_page_cgroup(page
);
5130 * Do only loose check w/o page_cgroup lock.
5131 * mem_cgroup_move_account() checks the pc is valid or not under
5134 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5135 ret
= MC_TARGET_PAGE
;
5137 target
->page
= page
;
5139 if (!ret
|| !target
)
5142 /* There is a swap entry and a page doesn't exist or isn't charged */
5143 if (ent
.val
&& !ret
&&
5144 css_id(&mc
.from
->css
) == lookup_swap_cgroup_id(ent
)) {
5145 ret
= MC_TARGET_SWAP
;
5152 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5154 * We don't consider swapping or file mapped pages because THP does not
5155 * support them for now.
5156 * Caller should make sure that pmd_trans_huge(pmd) is true.
5158 static enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
5159 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
5161 struct page
*page
= NULL
;
5162 struct page_cgroup
*pc
;
5163 enum mc_target_type ret
= MC_TARGET_NONE
;
5165 page
= pmd_page(pmd
);
5166 VM_BUG_ON(!page
|| !PageHead(page
));
5169 pc
= lookup_page_cgroup(page
);
5170 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5171 ret
= MC_TARGET_PAGE
;
5174 target
->page
= page
;
5180 static inline enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
5181 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
5183 return MC_TARGET_NONE
;
5187 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
5188 unsigned long addr
, unsigned long end
,
5189 struct mm_walk
*walk
)
5191 struct vm_area_struct
*vma
= walk
->private;
5195 if (pmd_trans_huge_lock(pmd
, vma
) == 1) {
5196 if (get_mctgt_type_thp(vma
, addr
, *pmd
, NULL
) == MC_TARGET_PAGE
)
5197 mc
.precharge
+= HPAGE_PMD_NR
;
5198 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5202 if (pmd_trans_unstable(pmd
))
5204 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5205 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
5206 if (get_mctgt_type(vma
, addr
, *pte
, NULL
))
5207 mc
.precharge
++; /* increment precharge temporarily */
5208 pte_unmap_unlock(pte
- 1, ptl
);
5214 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
5216 unsigned long precharge
;
5217 struct vm_area_struct
*vma
;
5219 down_read(&mm
->mmap_sem
);
5220 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5221 struct mm_walk mem_cgroup_count_precharge_walk
= {
5222 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
5226 if (is_vm_hugetlb_page(vma
))
5228 walk_page_range(vma
->vm_start
, vma
->vm_end
,
5229 &mem_cgroup_count_precharge_walk
);
5231 up_read(&mm
->mmap_sem
);
5233 precharge
= mc
.precharge
;
5239 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
5241 unsigned long precharge
= mem_cgroup_count_precharge(mm
);
5243 VM_BUG_ON(mc
.moving_task
);
5244 mc
.moving_task
= current
;
5245 return mem_cgroup_do_precharge(precharge
);
5248 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5249 static void __mem_cgroup_clear_mc(void)
5251 struct mem_cgroup
*from
= mc
.from
;
5252 struct mem_cgroup
*to
= mc
.to
;
5254 /* we must uncharge all the leftover precharges from mc.to */
5256 __mem_cgroup_cancel_charge(mc
.to
, mc
.precharge
);
5260 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5261 * we must uncharge here.
5263 if (mc
.moved_charge
) {
5264 __mem_cgroup_cancel_charge(mc
.from
, mc
.moved_charge
);
5265 mc
.moved_charge
= 0;
5267 /* we must fixup refcnts and charges */
5268 if (mc
.moved_swap
) {
5269 /* uncharge swap account from the old cgroup */
5270 if (!mem_cgroup_is_root(mc
.from
))
5271 res_counter_uncharge(&mc
.from
->memsw
,
5272 PAGE_SIZE
* mc
.moved_swap
);
5273 __mem_cgroup_put(mc
.from
, mc
.moved_swap
);
5275 if (!mem_cgroup_is_root(mc
.to
)) {
5277 * we charged both to->res and to->memsw, so we should
5280 res_counter_uncharge(&mc
.to
->res
,
5281 PAGE_SIZE
* mc
.moved_swap
);
5283 /* we've already done mem_cgroup_get(mc.to) */
5286 memcg_oom_recover(from
);
5287 memcg_oom_recover(to
);
5288 wake_up_all(&mc
.waitq
);
5291 static void mem_cgroup_clear_mc(void)
5293 struct mem_cgroup
*from
= mc
.from
;
5296 * we must clear moving_task before waking up waiters at the end of
5299 mc
.moving_task
= NULL
;
5300 __mem_cgroup_clear_mc();
5301 spin_lock(&mc
.lock
);
5304 spin_unlock(&mc
.lock
);
5305 mem_cgroup_end_move(from
);
5308 static int mem_cgroup_can_attach(struct cgroup
*cgroup
,
5309 struct cgroup_taskset
*tset
)
5311 struct task_struct
*p
= cgroup_taskset_first(tset
);
5313 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgroup
);
5315 if (memcg
->move_charge_at_immigrate
) {
5316 struct mm_struct
*mm
;
5317 struct mem_cgroup
*from
= mem_cgroup_from_task(p
);
5319 VM_BUG_ON(from
== memcg
);
5321 mm
= get_task_mm(p
);
5324 /* We move charges only when we move a owner of the mm */
5325 if (mm
->owner
== p
) {
5328 VM_BUG_ON(mc
.precharge
);
5329 VM_BUG_ON(mc
.moved_charge
);
5330 VM_BUG_ON(mc
.moved_swap
);
5331 mem_cgroup_start_move(from
);
5332 spin_lock(&mc
.lock
);
5335 spin_unlock(&mc
.lock
);
5336 /* We set mc.moving_task later */
5338 ret
= mem_cgroup_precharge_mc(mm
);
5340 mem_cgroup_clear_mc();
5347 static void mem_cgroup_cancel_attach(struct cgroup
*cgroup
,
5348 struct cgroup_taskset
*tset
)
5350 mem_cgroup_clear_mc();
5353 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
5354 unsigned long addr
, unsigned long end
,
5355 struct mm_walk
*walk
)
5358 struct vm_area_struct
*vma
= walk
->private;
5361 enum mc_target_type target_type
;
5362 union mc_target target
;
5364 struct page_cgroup
*pc
;
5367 * We don't take compound_lock() here but no race with splitting thp
5369 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
5370 * under splitting, which means there's no concurrent thp split,
5371 * - if another thread runs into split_huge_page() just after we
5372 * entered this if-block, the thread must wait for page table lock
5373 * to be unlocked in __split_huge_page_splitting(), where the main
5374 * part of thp split is not executed yet.
5376 if (pmd_trans_huge_lock(pmd
, vma
) == 1) {
5377 if (mc
.precharge
< HPAGE_PMD_NR
) {
5378 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5381 target_type
= get_mctgt_type_thp(vma
, addr
, *pmd
, &target
);
5382 if (target_type
== MC_TARGET_PAGE
) {
5384 if (!isolate_lru_page(page
)) {
5385 pc
= lookup_page_cgroup(page
);
5386 if (!mem_cgroup_move_account(page
, HPAGE_PMD_NR
,
5387 pc
, mc
.from
, mc
.to
)) {
5388 mc
.precharge
-= HPAGE_PMD_NR
;
5389 mc
.moved_charge
+= HPAGE_PMD_NR
;
5391 putback_lru_page(page
);
5395 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5399 if (pmd_trans_unstable(pmd
))
5402 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5403 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
5404 pte_t ptent
= *(pte
++);
5410 switch (get_mctgt_type(vma
, addr
, ptent
, &target
)) {
5411 case MC_TARGET_PAGE
:
5413 if (isolate_lru_page(page
))
5415 pc
= lookup_page_cgroup(page
);
5416 if (!mem_cgroup_move_account(page
, 1, pc
,
5419 /* we uncharge from mc.from later. */
5422 putback_lru_page(page
);
5423 put
: /* get_mctgt_type() gets the page */
5426 case MC_TARGET_SWAP
:
5428 if (!mem_cgroup_move_swap_account(ent
, mc
.from
, mc
.to
)) {
5430 /* we fixup refcnts and charges later. */
5438 pte_unmap_unlock(pte
- 1, ptl
);
5443 * We have consumed all precharges we got in can_attach().
5444 * We try charge one by one, but don't do any additional
5445 * charges to mc.to if we have failed in charge once in attach()
5448 ret
= mem_cgroup_do_precharge(1);
5456 static void mem_cgroup_move_charge(struct mm_struct
*mm
)
5458 struct vm_area_struct
*vma
;
5460 lru_add_drain_all();
5462 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
5464 * Someone who are holding the mmap_sem might be waiting in
5465 * waitq. So we cancel all extra charges, wake up all waiters,
5466 * and retry. Because we cancel precharges, we might not be able
5467 * to move enough charges, but moving charge is a best-effort
5468 * feature anyway, so it wouldn't be a big problem.
5470 __mem_cgroup_clear_mc();
5474 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5476 struct mm_walk mem_cgroup_move_charge_walk
= {
5477 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
5481 if (is_vm_hugetlb_page(vma
))
5483 ret
= walk_page_range(vma
->vm_start
, vma
->vm_end
,
5484 &mem_cgroup_move_charge_walk
);
5487 * means we have consumed all precharges and failed in
5488 * doing additional charge. Just abandon here.
5492 up_read(&mm
->mmap_sem
);
5495 static void mem_cgroup_move_task(struct cgroup
*cont
,
5496 struct cgroup_taskset
*tset
)
5498 struct task_struct
*p
= cgroup_taskset_first(tset
);
5499 struct mm_struct
*mm
= get_task_mm(p
);
5503 mem_cgroup_move_charge(mm
);
5507 mem_cgroup_clear_mc();
5509 #else /* !CONFIG_MMU */
5510 static int mem_cgroup_can_attach(struct cgroup
*cgroup
,
5511 struct cgroup_taskset
*tset
)
5515 static void mem_cgroup_cancel_attach(struct cgroup
*cgroup
,
5516 struct cgroup_taskset
*tset
)
5519 static void mem_cgroup_move_task(struct cgroup
*cont
,
5520 struct cgroup_taskset
*tset
)
5525 struct cgroup_subsys mem_cgroup_subsys
= {
5527 .subsys_id
= mem_cgroup_subsys_id
,
5528 .create
= mem_cgroup_create
,
5529 .pre_destroy
= mem_cgroup_pre_destroy
,
5530 .destroy
= mem_cgroup_destroy
,
5531 .can_attach
= mem_cgroup_can_attach
,
5532 .cancel_attach
= mem_cgroup_cancel_attach
,
5533 .attach
= mem_cgroup_move_task
,
5534 .base_cftypes
= mem_cgroup_files
,
5537 .__DEPRECATED_clear_css_refs
= true,
5540 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
5541 static int __init
enable_swap_account(char *s
)
5543 /* consider enabled if no parameter or 1 is given */
5544 if (!strcmp(s
, "1"))
5545 really_do_swap_account
= 1;
5546 else if (!strcmp(s
, "0"))
5547 really_do_swap_account
= 0;
5550 __setup("swapaccount=", enable_swap_account
);