1 /* memcontrol.c - Memory Controller
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
6 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
13 * This program is free software; you can redistribute it and/or modify
14 * it under the terms of the GNU General Public License as published by
15 * the Free Software Foundation; either version 2 of the License, or
16 * (at your option) any later version.
18 * This program is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 * GNU General Public License for more details.
24 #include <linux/res_counter.h>
25 #include <linux/memcontrol.h>
26 #include <linux/cgroup.h>
28 #include <linux/hugetlb.h>
29 #include <linux/pagemap.h>
30 #include <linux/smp.h>
31 #include <linux/page-flags.h>
32 #include <linux/backing-dev.h>
33 #include <linux/bit_spinlock.h>
34 #include <linux/rcupdate.h>
35 #include <linux/limits.h>
36 #include <linux/export.h>
37 #include <linux/mutex.h>
38 #include <linux/rbtree.h>
39 #include <linux/slab.h>
40 #include <linux/swap.h>
41 #include <linux/swapops.h>
42 #include <linux/spinlock.h>
43 #include <linux/eventfd.h>
44 #include <linux/sort.h>
46 #include <linux/seq_file.h>
47 #include <linux/vmalloc.h>
48 #include <linux/mm_inline.h>
49 #include <linux/page_cgroup.h>
50 #include <linux/cpu.h>
51 #include <linux/oom.h>
55 #include <net/tcp_memcontrol.h>
57 #include <asm/uaccess.h>
59 #include <trace/events/vmscan.h>
61 struct cgroup_subsys mem_cgroup_subsys __read_mostly
;
62 EXPORT_SYMBOL(mem_cgroup_subsys
);
64 #define MEM_CGROUP_RECLAIM_RETRIES 5
65 static struct mem_cgroup
*root_mem_cgroup __read_mostly
;
67 #ifdef CONFIG_MEMCG_SWAP
68 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
69 int do_swap_account __read_mostly
;
71 /* for remember boot option*/
72 #ifdef CONFIG_MEMCG_SWAP_ENABLED
73 static int really_do_swap_account __initdata
= 1;
75 static int really_do_swap_account __initdata
= 0;
79 #define do_swap_account 0
84 * Statistics for memory cgroup.
86 enum mem_cgroup_stat_index
{
88 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
90 MEM_CGROUP_STAT_CACHE
, /* # of pages charged as cache */
91 MEM_CGROUP_STAT_RSS
, /* # of pages charged as anon rss */
92 MEM_CGROUP_STAT_FILE_MAPPED
, /* # of pages charged as file rss */
93 MEM_CGROUP_STAT_SWAP
, /* # of pages, swapped out */
94 MEM_CGROUP_STAT_NSTATS
,
97 static const char * const mem_cgroup_stat_names
[] = {
104 enum mem_cgroup_events_index
{
105 MEM_CGROUP_EVENTS_PGPGIN
, /* # of pages paged in */
106 MEM_CGROUP_EVENTS_PGPGOUT
, /* # of pages paged out */
107 MEM_CGROUP_EVENTS_PGFAULT
, /* # of page-faults */
108 MEM_CGROUP_EVENTS_PGMAJFAULT
, /* # of major page-faults */
109 MEM_CGROUP_EVENTS_NSTATS
,
112 static const char * const mem_cgroup_events_names
[] = {
120 * Per memcg event counter is incremented at every pagein/pageout. With THP,
121 * it will be incremated by the number of pages. This counter is used for
122 * for trigger some periodic events. This is straightforward and better
123 * than using jiffies etc. to handle periodic memcg event.
125 enum mem_cgroup_events_target
{
126 MEM_CGROUP_TARGET_THRESH
,
127 MEM_CGROUP_TARGET_SOFTLIMIT
,
128 MEM_CGROUP_TARGET_NUMAINFO
,
131 #define THRESHOLDS_EVENTS_TARGET 128
132 #define SOFTLIMIT_EVENTS_TARGET 1024
133 #define NUMAINFO_EVENTS_TARGET 1024
135 struct mem_cgroup_stat_cpu
{
136 long count
[MEM_CGROUP_STAT_NSTATS
];
137 unsigned long events
[MEM_CGROUP_EVENTS_NSTATS
];
138 unsigned long nr_page_events
;
139 unsigned long targets
[MEM_CGROUP_NTARGETS
];
142 struct mem_cgroup_reclaim_iter
{
143 /* css_id of the last scanned hierarchy member */
145 /* scan generation, increased every round-trip */
146 unsigned int generation
;
150 * per-zone information in memory controller.
152 struct mem_cgroup_per_zone
{
153 struct lruvec lruvec
;
154 unsigned long lru_size
[NR_LRU_LISTS
];
156 struct mem_cgroup_reclaim_iter reclaim_iter
[DEF_PRIORITY
+ 1];
158 struct rb_node tree_node
; /* RB tree node */
159 unsigned long long usage_in_excess
;/* Set to the value by which */
160 /* the soft limit is exceeded*/
162 struct mem_cgroup
*memcg
; /* Back pointer, we cannot */
163 /* use container_of */
166 struct mem_cgroup_per_node
{
167 struct mem_cgroup_per_zone zoneinfo
[MAX_NR_ZONES
];
170 struct mem_cgroup_lru_info
{
171 struct mem_cgroup_per_node
*nodeinfo
[MAX_NUMNODES
];
175 * Cgroups above their limits are maintained in a RB-Tree, independent of
176 * their hierarchy representation
179 struct mem_cgroup_tree_per_zone
{
180 struct rb_root rb_root
;
184 struct mem_cgroup_tree_per_node
{
185 struct mem_cgroup_tree_per_zone rb_tree_per_zone
[MAX_NR_ZONES
];
188 struct mem_cgroup_tree
{
189 struct mem_cgroup_tree_per_node
*rb_tree_per_node
[MAX_NUMNODES
];
192 static struct mem_cgroup_tree soft_limit_tree __read_mostly
;
194 struct mem_cgroup_threshold
{
195 struct eventfd_ctx
*eventfd
;
200 struct mem_cgroup_threshold_ary
{
201 /* An array index points to threshold just below or equal to usage. */
202 int current_threshold
;
203 /* Size of entries[] */
205 /* Array of thresholds */
206 struct mem_cgroup_threshold entries
[0];
209 struct mem_cgroup_thresholds
{
210 /* Primary thresholds array */
211 struct mem_cgroup_threshold_ary
*primary
;
213 * Spare threshold array.
214 * This is needed to make mem_cgroup_unregister_event() "never fail".
215 * It must be able to store at least primary->size - 1 entries.
217 struct mem_cgroup_threshold_ary
*spare
;
221 struct mem_cgroup_eventfd_list
{
222 struct list_head list
;
223 struct eventfd_ctx
*eventfd
;
226 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
);
227 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
);
230 * The memory controller data structure. The memory controller controls both
231 * page cache and RSS per cgroup. We would eventually like to provide
232 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
233 * to help the administrator determine what knobs to tune.
235 * TODO: Add a water mark for the memory controller. Reclaim will begin when
236 * we hit the water mark. May be even add a low water mark, such that
237 * no reclaim occurs from a cgroup at it's low water mark, this is
238 * a feature that will be implemented much later in the future.
241 struct cgroup_subsys_state css
;
243 * the counter to account for memory usage
245 struct res_counter res
;
249 * the counter to account for mem+swap usage.
251 struct res_counter memsw
;
254 * rcu_freeing is used only when freeing struct mem_cgroup,
255 * so put it into a union to avoid wasting more memory.
256 * It must be disjoint from the css field. It could be
257 * in a union with the res field, but res plays a much
258 * larger part in mem_cgroup life than memsw, and might
259 * be of interest, even at time of free, when debugging.
260 * So share rcu_head with the less interesting memsw.
262 struct rcu_head rcu_freeing
;
264 * We also need some space for a worker in deferred freeing.
265 * By the time we call it, rcu_freeing is no longer in use.
267 struct work_struct work_freeing
;
271 * Per cgroup active and inactive list, similar to the
272 * per zone LRU lists.
274 struct mem_cgroup_lru_info info
;
275 int last_scanned_node
;
277 nodemask_t scan_nodes
;
278 atomic_t numainfo_events
;
279 atomic_t numainfo_updating
;
282 * Should the accounting and control be hierarchical, per subtree?
292 /* OOM-Killer disable */
293 int oom_kill_disable
;
295 /* set when res.limit == memsw.limit */
296 bool memsw_is_minimum
;
298 /* protect arrays of thresholds */
299 struct mutex thresholds_lock
;
301 /* thresholds for memory usage. RCU-protected */
302 struct mem_cgroup_thresholds thresholds
;
304 /* thresholds for mem+swap usage. RCU-protected */
305 struct mem_cgroup_thresholds memsw_thresholds
;
307 /* For oom notifier event fd */
308 struct list_head oom_notify
;
311 * Should we move charges of a task when a task is moved into this
312 * mem_cgroup ? And what type of charges should we move ?
314 unsigned long move_charge_at_immigrate
;
316 * set > 0 if pages under this cgroup are moving to other cgroup.
318 atomic_t moving_account
;
319 /* taken only while moving_account > 0 */
320 spinlock_t move_lock
;
324 struct mem_cgroup_stat_cpu __percpu
*stat
;
326 * used when a cpu is offlined or other synchronizations
327 * See mem_cgroup_read_stat().
329 struct mem_cgroup_stat_cpu nocpu_base
;
330 spinlock_t pcp_counter_lock
;
332 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET)
333 struct tcp_memcontrol tcp_mem
;
337 /* Stuffs for move charges at task migration. */
339 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
340 * left-shifted bitmap of these types.
343 MOVE_CHARGE_TYPE_ANON
, /* private anonymous page and swap of it */
344 MOVE_CHARGE_TYPE_FILE
, /* file page(including tmpfs) and swap of it */
348 /* "mc" and its members are protected by cgroup_mutex */
349 static struct move_charge_struct
{
350 spinlock_t lock
; /* for from, to */
351 struct mem_cgroup
*from
;
352 struct mem_cgroup
*to
;
353 unsigned long precharge
;
354 unsigned long moved_charge
;
355 unsigned long moved_swap
;
356 struct task_struct
*moving_task
; /* a task moving charges */
357 wait_queue_head_t waitq
; /* a waitq for other context */
359 .lock
= __SPIN_LOCK_UNLOCKED(mc
.lock
),
360 .waitq
= __WAIT_QUEUE_HEAD_INITIALIZER(mc
.waitq
),
363 static bool move_anon(void)
365 return test_bit(MOVE_CHARGE_TYPE_ANON
,
366 &mc
.to
->move_charge_at_immigrate
);
369 static bool move_file(void)
371 return test_bit(MOVE_CHARGE_TYPE_FILE
,
372 &mc
.to
->move_charge_at_immigrate
);
376 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
377 * limit reclaim to prevent infinite loops, if they ever occur.
379 #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
380 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
383 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
384 MEM_CGROUP_CHARGE_TYPE_ANON
,
385 MEM_CGROUP_CHARGE_TYPE_SWAPOUT
, /* for accounting swapcache */
386 MEM_CGROUP_CHARGE_TYPE_DROP
, /* a page was unused swap cache */
390 /* for encoding cft->private value on file */
393 #define _OOM_TYPE (2)
394 #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
395 #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
396 #define MEMFILE_ATTR(val) ((val) & 0xffff)
397 /* Used for OOM nofiier */
398 #define OOM_CONTROL (0)
401 * Reclaim flags for mem_cgroup_hierarchical_reclaim
403 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
404 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
405 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
406 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
408 static void mem_cgroup_get(struct mem_cgroup
*memcg
);
409 static void mem_cgroup_put(struct mem_cgroup
*memcg
);
412 struct mem_cgroup
*mem_cgroup_from_css(struct cgroup_subsys_state
*s
)
414 return container_of(s
, struct mem_cgroup
, css
);
417 static inline bool mem_cgroup_is_root(struct mem_cgroup
*memcg
)
419 return (memcg
== root_mem_cgroup
);
422 /* Writing them here to avoid exposing memcg's inner layout */
423 #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
425 void sock_update_memcg(struct sock
*sk
)
427 if (mem_cgroup_sockets_enabled
) {
428 struct mem_cgroup
*memcg
;
429 struct cg_proto
*cg_proto
;
431 BUG_ON(!sk
->sk_prot
->proto_cgroup
);
433 /* Socket cloning can throw us here with sk_cgrp already
434 * filled. It won't however, necessarily happen from
435 * process context. So the test for root memcg given
436 * the current task's memcg won't help us in this case.
438 * Respecting the original socket's memcg is a better
439 * decision in this case.
442 BUG_ON(mem_cgroup_is_root(sk
->sk_cgrp
->memcg
));
443 mem_cgroup_get(sk
->sk_cgrp
->memcg
);
448 memcg
= mem_cgroup_from_task(current
);
449 cg_proto
= sk
->sk_prot
->proto_cgroup(memcg
);
450 if (!mem_cgroup_is_root(memcg
) && memcg_proto_active(cg_proto
)) {
451 mem_cgroup_get(memcg
);
452 sk
->sk_cgrp
= cg_proto
;
457 EXPORT_SYMBOL(sock_update_memcg
);
459 void sock_release_memcg(struct sock
*sk
)
461 if (mem_cgroup_sockets_enabled
&& sk
->sk_cgrp
) {
462 struct mem_cgroup
*memcg
;
463 WARN_ON(!sk
->sk_cgrp
->memcg
);
464 memcg
= sk
->sk_cgrp
->memcg
;
465 mem_cgroup_put(memcg
);
469 struct cg_proto
*tcp_proto_cgroup(struct mem_cgroup
*memcg
)
471 if (!memcg
|| mem_cgroup_is_root(memcg
))
474 return &memcg
->tcp_mem
.cg_proto
;
476 EXPORT_SYMBOL(tcp_proto_cgroup
);
478 static void disarm_sock_keys(struct mem_cgroup
*memcg
)
480 if (!memcg_proto_activated(&memcg
->tcp_mem
.cg_proto
))
482 static_key_slow_dec(&memcg_socket_limit_enabled
);
485 static void disarm_sock_keys(struct mem_cgroup
*memcg
)
490 static void drain_all_stock_async(struct mem_cgroup
*memcg
);
492 static struct mem_cgroup_per_zone
*
493 mem_cgroup_zoneinfo(struct mem_cgroup
*memcg
, int nid
, int zid
)
495 return &memcg
->info
.nodeinfo
[nid
]->zoneinfo
[zid
];
498 struct cgroup_subsys_state
*mem_cgroup_css(struct mem_cgroup
*memcg
)
503 static struct mem_cgroup_per_zone
*
504 page_cgroup_zoneinfo(struct mem_cgroup
*memcg
, struct page
*page
)
506 int nid
= page_to_nid(page
);
507 int zid
= page_zonenum(page
);
509 return mem_cgroup_zoneinfo(memcg
, nid
, zid
);
512 static struct mem_cgroup_tree_per_zone
*
513 soft_limit_tree_node_zone(int nid
, int zid
)
515 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
518 static struct mem_cgroup_tree_per_zone
*
519 soft_limit_tree_from_page(struct page
*page
)
521 int nid
= page_to_nid(page
);
522 int zid
= page_zonenum(page
);
524 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
528 __mem_cgroup_insert_exceeded(struct mem_cgroup
*memcg
,
529 struct mem_cgroup_per_zone
*mz
,
530 struct mem_cgroup_tree_per_zone
*mctz
,
531 unsigned long long new_usage_in_excess
)
533 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
534 struct rb_node
*parent
= NULL
;
535 struct mem_cgroup_per_zone
*mz_node
;
540 mz
->usage_in_excess
= new_usage_in_excess
;
541 if (!mz
->usage_in_excess
)
545 mz_node
= rb_entry(parent
, struct mem_cgroup_per_zone
,
547 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
)
550 * We can't avoid mem cgroups that are over their soft
551 * limit by the same amount
553 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
556 rb_link_node(&mz
->tree_node
, parent
, p
);
557 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
562 __mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
563 struct mem_cgroup_per_zone
*mz
,
564 struct mem_cgroup_tree_per_zone
*mctz
)
568 rb_erase(&mz
->tree_node
, &mctz
->rb_root
);
573 mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
574 struct mem_cgroup_per_zone
*mz
,
575 struct mem_cgroup_tree_per_zone
*mctz
)
577 spin_lock(&mctz
->lock
);
578 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
579 spin_unlock(&mctz
->lock
);
583 static void mem_cgroup_update_tree(struct mem_cgroup
*memcg
, struct page
*page
)
585 unsigned long long excess
;
586 struct mem_cgroup_per_zone
*mz
;
587 struct mem_cgroup_tree_per_zone
*mctz
;
588 int nid
= page_to_nid(page
);
589 int zid
= page_zonenum(page
);
590 mctz
= soft_limit_tree_from_page(page
);
593 * Necessary to update all ancestors when hierarchy is used.
594 * because their event counter is not touched.
596 for (; memcg
; memcg
= parent_mem_cgroup(memcg
)) {
597 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
598 excess
= res_counter_soft_limit_excess(&memcg
->res
);
600 * We have to update the tree if mz is on RB-tree or
601 * mem is over its softlimit.
603 if (excess
|| mz
->on_tree
) {
604 spin_lock(&mctz
->lock
);
605 /* if on-tree, remove it */
607 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
609 * Insert again. mz->usage_in_excess will be updated.
610 * If excess is 0, no tree ops.
612 __mem_cgroup_insert_exceeded(memcg
, mz
, mctz
, excess
);
613 spin_unlock(&mctz
->lock
);
618 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*memcg
)
621 struct mem_cgroup_per_zone
*mz
;
622 struct mem_cgroup_tree_per_zone
*mctz
;
624 for_each_node(node
) {
625 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
626 mz
= mem_cgroup_zoneinfo(memcg
, node
, zone
);
627 mctz
= soft_limit_tree_node_zone(node
, zone
);
628 mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
633 static struct mem_cgroup_per_zone
*
634 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
636 struct rb_node
*rightmost
= NULL
;
637 struct mem_cgroup_per_zone
*mz
;
641 rightmost
= rb_last(&mctz
->rb_root
);
643 goto done
; /* Nothing to reclaim from */
645 mz
= rb_entry(rightmost
, struct mem_cgroup_per_zone
, tree_node
);
647 * Remove the node now but someone else can add it back,
648 * we will to add it back at the end of reclaim to its correct
649 * position in the tree.
651 __mem_cgroup_remove_exceeded(mz
->memcg
, mz
, mctz
);
652 if (!res_counter_soft_limit_excess(&mz
->memcg
->res
) ||
653 !css_tryget(&mz
->memcg
->css
))
659 static struct mem_cgroup_per_zone
*
660 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
662 struct mem_cgroup_per_zone
*mz
;
664 spin_lock(&mctz
->lock
);
665 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
666 spin_unlock(&mctz
->lock
);
671 * Implementation Note: reading percpu statistics for memcg.
673 * Both of vmstat[] and percpu_counter has threshold and do periodic
674 * synchronization to implement "quick" read. There are trade-off between
675 * reading cost and precision of value. Then, we may have a chance to implement
676 * a periodic synchronizion of counter in memcg's counter.
678 * But this _read() function is used for user interface now. The user accounts
679 * memory usage by memory cgroup and he _always_ requires exact value because
680 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
681 * have to visit all online cpus and make sum. So, for now, unnecessary
682 * synchronization is not implemented. (just implemented for cpu hotplug)
684 * If there are kernel internal actions which can make use of some not-exact
685 * value, and reading all cpu value can be performance bottleneck in some
686 * common workload, threashold and synchonization as vmstat[] should be
689 static long mem_cgroup_read_stat(struct mem_cgroup
*memcg
,
690 enum mem_cgroup_stat_index idx
)
696 for_each_online_cpu(cpu
)
697 val
+= per_cpu(memcg
->stat
->count
[idx
], cpu
);
698 #ifdef CONFIG_HOTPLUG_CPU
699 spin_lock(&memcg
->pcp_counter_lock
);
700 val
+= memcg
->nocpu_base
.count
[idx
];
701 spin_unlock(&memcg
->pcp_counter_lock
);
707 static void mem_cgroup_swap_statistics(struct mem_cgroup
*memcg
,
710 int val
= (charge
) ? 1 : -1;
711 this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_SWAP
], val
);
714 static unsigned long mem_cgroup_read_events(struct mem_cgroup
*memcg
,
715 enum mem_cgroup_events_index idx
)
717 unsigned long val
= 0;
720 for_each_online_cpu(cpu
)
721 val
+= per_cpu(memcg
->stat
->events
[idx
], cpu
);
722 #ifdef CONFIG_HOTPLUG_CPU
723 spin_lock(&memcg
->pcp_counter_lock
);
724 val
+= memcg
->nocpu_base
.events
[idx
];
725 spin_unlock(&memcg
->pcp_counter_lock
);
730 static void mem_cgroup_charge_statistics(struct mem_cgroup
*memcg
,
731 bool anon
, int nr_pages
)
736 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
737 * counted as CACHE even if it's on ANON LRU.
740 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_RSS
],
743 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_CACHE
],
746 /* pagein of a big page is an event. So, ignore page size */
748 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGIN
]);
750 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGOUT
]);
751 nr_pages
= -nr_pages
; /* for event */
754 __this_cpu_add(memcg
->stat
->nr_page_events
, nr_pages
);
760 mem_cgroup_get_lru_size(struct lruvec
*lruvec
, enum lru_list lru
)
762 struct mem_cgroup_per_zone
*mz
;
764 mz
= container_of(lruvec
, struct mem_cgroup_per_zone
, lruvec
);
765 return mz
->lru_size
[lru
];
769 mem_cgroup_zone_nr_lru_pages(struct mem_cgroup
*memcg
, int nid
, int zid
,
770 unsigned int lru_mask
)
772 struct mem_cgroup_per_zone
*mz
;
774 unsigned long ret
= 0;
776 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
779 if (BIT(lru
) & lru_mask
)
780 ret
+= mz
->lru_size
[lru
];
786 mem_cgroup_node_nr_lru_pages(struct mem_cgroup
*memcg
,
787 int nid
, unsigned int lru_mask
)
792 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++)
793 total
+= mem_cgroup_zone_nr_lru_pages(memcg
,
799 static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup
*memcg
,
800 unsigned int lru_mask
)
805 for_each_node_state(nid
, N_MEMORY
)
806 total
+= mem_cgroup_node_nr_lru_pages(memcg
, nid
, lru_mask
);
810 static bool mem_cgroup_event_ratelimit(struct mem_cgroup
*memcg
,
811 enum mem_cgroup_events_target target
)
813 unsigned long val
, next
;
815 val
= __this_cpu_read(memcg
->stat
->nr_page_events
);
816 next
= __this_cpu_read(memcg
->stat
->targets
[target
]);
817 /* from time_after() in jiffies.h */
818 if ((long)next
- (long)val
< 0) {
820 case MEM_CGROUP_TARGET_THRESH
:
821 next
= val
+ THRESHOLDS_EVENTS_TARGET
;
823 case MEM_CGROUP_TARGET_SOFTLIMIT
:
824 next
= val
+ SOFTLIMIT_EVENTS_TARGET
;
826 case MEM_CGROUP_TARGET_NUMAINFO
:
827 next
= val
+ NUMAINFO_EVENTS_TARGET
;
832 __this_cpu_write(memcg
->stat
->targets
[target
], next
);
839 * Check events in order.
842 static void memcg_check_events(struct mem_cgroup
*memcg
, struct page
*page
)
845 /* threshold event is triggered in finer grain than soft limit */
846 if (unlikely(mem_cgroup_event_ratelimit(memcg
,
847 MEM_CGROUP_TARGET_THRESH
))) {
849 bool do_numainfo __maybe_unused
;
851 do_softlimit
= mem_cgroup_event_ratelimit(memcg
,
852 MEM_CGROUP_TARGET_SOFTLIMIT
);
854 do_numainfo
= mem_cgroup_event_ratelimit(memcg
,
855 MEM_CGROUP_TARGET_NUMAINFO
);
859 mem_cgroup_threshold(memcg
);
860 if (unlikely(do_softlimit
))
861 mem_cgroup_update_tree(memcg
, page
);
863 if (unlikely(do_numainfo
))
864 atomic_inc(&memcg
->numainfo_events
);
870 struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
872 return mem_cgroup_from_css(
873 cgroup_subsys_state(cont
, mem_cgroup_subsys_id
));
876 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
879 * mm_update_next_owner() may clear mm->owner to NULL
880 * if it races with swapoff, page migration, etc.
881 * So this can be called with p == NULL.
886 return mem_cgroup_from_css(task_subsys_state(p
, mem_cgroup_subsys_id
));
889 struct mem_cgroup
*try_get_mem_cgroup_from_mm(struct mm_struct
*mm
)
891 struct mem_cgroup
*memcg
= NULL
;
896 * Because we have no locks, mm->owner's may be being moved to other
897 * cgroup. We use css_tryget() here even if this looks
898 * pessimistic (rather than adding locks here).
902 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
903 if (unlikely(!memcg
))
905 } while (!css_tryget(&memcg
->css
));
911 * mem_cgroup_iter - iterate over memory cgroup hierarchy
912 * @root: hierarchy root
913 * @prev: previously returned memcg, NULL on first invocation
914 * @reclaim: cookie for shared reclaim walks, NULL for full walks
916 * Returns references to children of the hierarchy below @root, or
917 * @root itself, or %NULL after a full round-trip.
919 * Caller must pass the return value in @prev on subsequent
920 * invocations for reference counting, or use mem_cgroup_iter_break()
921 * to cancel a hierarchy walk before the round-trip is complete.
923 * Reclaimers can specify a zone and a priority level in @reclaim to
924 * divide up the memcgs in the hierarchy among all concurrent
925 * reclaimers operating on the same zone and priority.
927 struct mem_cgroup
*mem_cgroup_iter(struct mem_cgroup
*root
,
928 struct mem_cgroup
*prev
,
929 struct mem_cgroup_reclaim_cookie
*reclaim
)
931 struct mem_cgroup
*memcg
= NULL
;
934 if (mem_cgroup_disabled())
938 root
= root_mem_cgroup
;
940 if (prev
&& !reclaim
)
941 id
= css_id(&prev
->css
);
943 if (prev
&& prev
!= root
)
946 if (!root
->use_hierarchy
&& root
!= root_mem_cgroup
) {
953 struct mem_cgroup_reclaim_iter
*uninitialized_var(iter
);
954 struct cgroup_subsys_state
*css
;
957 int nid
= zone_to_nid(reclaim
->zone
);
958 int zid
= zone_idx(reclaim
->zone
);
959 struct mem_cgroup_per_zone
*mz
;
961 mz
= mem_cgroup_zoneinfo(root
, nid
, zid
);
962 iter
= &mz
->reclaim_iter
[reclaim
->priority
];
963 if (prev
&& reclaim
->generation
!= iter
->generation
)
969 css
= css_get_next(&mem_cgroup_subsys
, id
+ 1, &root
->css
, &id
);
971 if (css
== &root
->css
|| css_tryget(css
))
972 memcg
= mem_cgroup_from_css(css
);
981 else if (!prev
&& memcg
)
982 reclaim
->generation
= iter
->generation
;
992 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
993 * @root: hierarchy root
994 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
996 void mem_cgroup_iter_break(struct mem_cgroup
*root
,
997 struct mem_cgroup
*prev
)
1000 root
= root_mem_cgroup
;
1001 if (prev
&& prev
!= root
)
1002 css_put(&prev
->css
);
1006 * Iteration constructs for visiting all cgroups (under a tree). If
1007 * loops are exited prematurely (break), mem_cgroup_iter_break() must
1008 * be used for reference counting.
1010 #define for_each_mem_cgroup_tree(iter, root) \
1011 for (iter = mem_cgroup_iter(root, NULL, NULL); \
1013 iter = mem_cgroup_iter(root, iter, NULL))
1015 #define for_each_mem_cgroup(iter) \
1016 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
1018 iter = mem_cgroup_iter(NULL, iter, NULL))
1020 void __mem_cgroup_count_vm_event(struct mm_struct
*mm
, enum vm_event_item idx
)
1022 struct mem_cgroup
*memcg
;
1025 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
1026 if (unlikely(!memcg
))
1031 this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGFAULT
]);
1034 this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGMAJFAULT
]);
1042 EXPORT_SYMBOL(__mem_cgroup_count_vm_event
);
1045 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1046 * @zone: zone of the wanted lruvec
1047 * @memcg: memcg of the wanted lruvec
1049 * Returns the lru list vector holding pages for the given @zone and
1050 * @mem. This can be the global zone lruvec, if the memory controller
1053 struct lruvec
*mem_cgroup_zone_lruvec(struct zone
*zone
,
1054 struct mem_cgroup
*memcg
)
1056 struct mem_cgroup_per_zone
*mz
;
1057 struct lruvec
*lruvec
;
1059 if (mem_cgroup_disabled()) {
1060 lruvec
= &zone
->lruvec
;
1064 mz
= mem_cgroup_zoneinfo(memcg
, zone_to_nid(zone
), zone_idx(zone
));
1065 lruvec
= &mz
->lruvec
;
1068 * Since a node can be onlined after the mem_cgroup was created,
1069 * we have to be prepared to initialize lruvec->zone here;
1070 * and if offlined then reonlined, we need to reinitialize it.
1072 if (unlikely(lruvec
->zone
!= zone
))
1073 lruvec
->zone
= zone
;
1078 * Following LRU functions are allowed to be used without PCG_LOCK.
1079 * Operations are called by routine of global LRU independently from memcg.
1080 * What we have to take care of here is validness of pc->mem_cgroup.
1082 * Changes to pc->mem_cgroup happens when
1085 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1086 * It is added to LRU before charge.
1087 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1088 * When moving account, the page is not on LRU. It's isolated.
1092 * mem_cgroup_page_lruvec - return lruvec for adding an lru page
1094 * @zone: zone of the page
1096 struct lruvec
*mem_cgroup_page_lruvec(struct page
*page
, struct zone
*zone
)
1098 struct mem_cgroup_per_zone
*mz
;
1099 struct mem_cgroup
*memcg
;
1100 struct page_cgroup
*pc
;
1101 struct lruvec
*lruvec
;
1103 if (mem_cgroup_disabled()) {
1104 lruvec
= &zone
->lruvec
;
1108 pc
= lookup_page_cgroup(page
);
1109 memcg
= pc
->mem_cgroup
;
1112 * Surreptitiously switch any uncharged offlist page to root:
1113 * an uncharged page off lru does nothing to secure
1114 * its former mem_cgroup from sudden removal.
1116 * Our caller holds lru_lock, and PageCgroupUsed is updated
1117 * under page_cgroup lock: between them, they make all uses
1118 * of pc->mem_cgroup safe.
1120 if (!PageLRU(page
) && !PageCgroupUsed(pc
) && memcg
!= root_mem_cgroup
)
1121 pc
->mem_cgroup
= memcg
= root_mem_cgroup
;
1123 mz
= page_cgroup_zoneinfo(memcg
, page
);
1124 lruvec
= &mz
->lruvec
;
1127 * Since a node can be onlined after the mem_cgroup was created,
1128 * we have to be prepared to initialize lruvec->zone here;
1129 * and if offlined then reonlined, we need to reinitialize it.
1131 if (unlikely(lruvec
->zone
!= zone
))
1132 lruvec
->zone
= zone
;
1137 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1138 * @lruvec: mem_cgroup per zone lru vector
1139 * @lru: index of lru list the page is sitting on
1140 * @nr_pages: positive when adding or negative when removing
1142 * This function must be called when a page is added to or removed from an
1145 void mem_cgroup_update_lru_size(struct lruvec
*lruvec
, enum lru_list lru
,
1148 struct mem_cgroup_per_zone
*mz
;
1149 unsigned long *lru_size
;
1151 if (mem_cgroup_disabled())
1154 mz
= container_of(lruvec
, struct mem_cgroup_per_zone
, lruvec
);
1155 lru_size
= mz
->lru_size
+ lru
;
1156 *lru_size
+= nr_pages
;
1157 VM_BUG_ON((long)(*lru_size
) < 0);
1161 * Checks whether given mem is same or in the root_mem_cgroup's
1164 bool __mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_memcg
,
1165 struct mem_cgroup
*memcg
)
1167 if (root_memcg
== memcg
)
1169 if (!root_memcg
->use_hierarchy
|| !memcg
)
1171 return css_is_ancestor(&memcg
->css
, &root_memcg
->css
);
1174 static bool mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_memcg
,
1175 struct mem_cgroup
*memcg
)
1180 ret
= __mem_cgroup_same_or_subtree(root_memcg
, memcg
);
1185 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*memcg
)
1188 struct mem_cgroup
*curr
= NULL
;
1189 struct task_struct
*p
;
1191 p
= find_lock_task_mm(task
);
1193 curr
= try_get_mem_cgroup_from_mm(p
->mm
);
1197 * All threads may have already detached their mm's, but the oom
1198 * killer still needs to detect if they have already been oom
1199 * killed to prevent needlessly killing additional tasks.
1202 curr
= mem_cgroup_from_task(task
);
1204 css_get(&curr
->css
);
1210 * We should check use_hierarchy of "memcg" not "curr". Because checking
1211 * use_hierarchy of "curr" here make this function true if hierarchy is
1212 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1213 * hierarchy(even if use_hierarchy is disabled in "memcg").
1215 ret
= mem_cgroup_same_or_subtree(memcg
, curr
);
1216 css_put(&curr
->css
);
1220 int mem_cgroup_inactive_anon_is_low(struct lruvec
*lruvec
)
1222 unsigned long inactive_ratio
;
1223 unsigned long inactive
;
1224 unsigned long active
;
1227 inactive
= mem_cgroup_get_lru_size(lruvec
, LRU_INACTIVE_ANON
);
1228 active
= mem_cgroup_get_lru_size(lruvec
, LRU_ACTIVE_ANON
);
1230 gb
= (inactive
+ active
) >> (30 - PAGE_SHIFT
);
1232 inactive_ratio
= int_sqrt(10 * gb
);
1236 return inactive
* inactive_ratio
< active
;
1239 int mem_cgroup_inactive_file_is_low(struct lruvec
*lruvec
)
1241 unsigned long active
;
1242 unsigned long inactive
;
1244 inactive
= mem_cgroup_get_lru_size(lruvec
, LRU_INACTIVE_FILE
);
1245 active
= mem_cgroup_get_lru_size(lruvec
, LRU_ACTIVE_FILE
);
1247 return (active
> inactive
);
1250 #define mem_cgroup_from_res_counter(counter, member) \
1251 container_of(counter, struct mem_cgroup, member)
1254 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1255 * @memcg: the memory cgroup
1257 * Returns the maximum amount of memory @mem can be charged with, in
1260 static unsigned long mem_cgroup_margin(struct mem_cgroup
*memcg
)
1262 unsigned long long margin
;
1264 margin
= res_counter_margin(&memcg
->res
);
1265 if (do_swap_account
)
1266 margin
= min(margin
, res_counter_margin(&memcg
->memsw
));
1267 return margin
>> PAGE_SHIFT
;
1270 int mem_cgroup_swappiness(struct mem_cgroup
*memcg
)
1272 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
1275 if (cgrp
->parent
== NULL
)
1276 return vm_swappiness
;
1278 return memcg
->swappiness
;
1282 * memcg->moving_account is used for checking possibility that some thread is
1283 * calling move_account(). When a thread on CPU-A starts moving pages under
1284 * a memcg, other threads should check memcg->moving_account under
1285 * rcu_read_lock(), like this:
1289 * memcg->moving_account+1 if (memcg->mocing_account)
1291 * synchronize_rcu() update something.
1296 /* for quick checking without looking up memcg */
1297 atomic_t memcg_moving __read_mostly
;
1299 static void mem_cgroup_start_move(struct mem_cgroup
*memcg
)
1301 atomic_inc(&memcg_moving
);
1302 atomic_inc(&memcg
->moving_account
);
1306 static void mem_cgroup_end_move(struct mem_cgroup
*memcg
)
1309 * Now, mem_cgroup_clear_mc() may call this function with NULL.
1310 * We check NULL in callee rather than caller.
1313 atomic_dec(&memcg_moving
);
1314 atomic_dec(&memcg
->moving_account
);
1319 * 2 routines for checking "mem" is under move_account() or not.
1321 * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This
1322 * is used for avoiding races in accounting. If true,
1323 * pc->mem_cgroup may be overwritten.
1325 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1326 * under hierarchy of moving cgroups. This is for
1327 * waiting at hith-memory prressure caused by "move".
1330 static bool mem_cgroup_stolen(struct mem_cgroup
*memcg
)
1332 VM_BUG_ON(!rcu_read_lock_held());
1333 return atomic_read(&memcg
->moving_account
) > 0;
1336 static bool mem_cgroup_under_move(struct mem_cgroup
*memcg
)
1338 struct mem_cgroup
*from
;
1339 struct mem_cgroup
*to
;
1342 * Unlike task_move routines, we access mc.to, mc.from not under
1343 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1345 spin_lock(&mc
.lock
);
1351 ret
= mem_cgroup_same_or_subtree(memcg
, from
)
1352 || mem_cgroup_same_or_subtree(memcg
, to
);
1354 spin_unlock(&mc
.lock
);
1358 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*memcg
)
1360 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1361 if (mem_cgroup_under_move(memcg
)) {
1363 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1364 /* moving charge context might have finished. */
1367 finish_wait(&mc
.waitq
, &wait
);
1375 * Take this lock when
1376 * - a code tries to modify page's memcg while it's USED.
1377 * - a code tries to modify page state accounting in a memcg.
1378 * see mem_cgroup_stolen(), too.
1380 static void move_lock_mem_cgroup(struct mem_cgroup
*memcg
,
1381 unsigned long *flags
)
1383 spin_lock_irqsave(&memcg
->move_lock
, *flags
);
1386 static void move_unlock_mem_cgroup(struct mem_cgroup
*memcg
,
1387 unsigned long *flags
)
1389 spin_unlock_irqrestore(&memcg
->move_lock
, *flags
);
1393 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1394 * @memcg: The memory cgroup that went over limit
1395 * @p: Task that is going to be killed
1397 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1400 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1402 struct cgroup
*task_cgrp
;
1403 struct cgroup
*mem_cgrp
;
1405 * Need a buffer in BSS, can't rely on allocations. The code relies
1406 * on the assumption that OOM is serialized for memory controller.
1407 * If this assumption is broken, revisit this code.
1409 static char memcg_name
[PATH_MAX
];
1417 mem_cgrp
= memcg
->css
.cgroup
;
1418 task_cgrp
= task_cgroup(p
, mem_cgroup_subsys_id
);
1420 ret
= cgroup_path(task_cgrp
, memcg_name
, PATH_MAX
);
1423 * Unfortunately, we are unable to convert to a useful name
1424 * But we'll still print out the usage information
1431 printk(KERN_INFO
"Task in %s killed", memcg_name
);
1434 ret
= cgroup_path(mem_cgrp
, memcg_name
, PATH_MAX
);
1442 * Continues from above, so we don't need an KERN_ level
1444 printk(KERN_CONT
" as a result of limit of %s\n", memcg_name
);
1447 printk(KERN_INFO
"memory: usage %llukB, limit %llukB, failcnt %llu\n",
1448 res_counter_read_u64(&memcg
->res
, RES_USAGE
) >> 10,
1449 res_counter_read_u64(&memcg
->res
, RES_LIMIT
) >> 10,
1450 res_counter_read_u64(&memcg
->res
, RES_FAILCNT
));
1451 printk(KERN_INFO
"memory+swap: usage %llukB, limit %llukB, "
1453 res_counter_read_u64(&memcg
->memsw
, RES_USAGE
) >> 10,
1454 res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
) >> 10,
1455 res_counter_read_u64(&memcg
->memsw
, RES_FAILCNT
));
1459 * This function returns the number of memcg under hierarchy tree. Returns
1460 * 1(self count) if no children.
1462 static int mem_cgroup_count_children(struct mem_cgroup
*memcg
)
1465 struct mem_cgroup
*iter
;
1467 for_each_mem_cgroup_tree(iter
, memcg
)
1473 * Return the memory (and swap, if configured) limit for a memcg.
1475 static u64
mem_cgroup_get_limit(struct mem_cgroup
*memcg
)
1479 limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
1482 * Do not consider swap space if we cannot swap due to swappiness
1484 if (mem_cgroup_swappiness(memcg
)) {
1487 limit
+= total_swap_pages
<< PAGE_SHIFT
;
1488 memsw
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
1491 * If memsw is finite and limits the amount of swap space
1492 * available to this memcg, return that limit.
1494 limit
= min(limit
, memsw
);
1500 static void mem_cgroup_out_of_memory(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
1503 struct mem_cgroup
*iter
;
1504 unsigned long chosen_points
= 0;
1505 unsigned long totalpages
;
1506 unsigned int points
= 0;
1507 struct task_struct
*chosen
= NULL
;
1510 * If current has a pending SIGKILL, then automatically select it. The
1511 * goal is to allow it to allocate so that it may quickly exit and free
1514 if (fatal_signal_pending(current
)) {
1515 set_thread_flag(TIF_MEMDIE
);
1519 check_panic_on_oom(CONSTRAINT_MEMCG
, gfp_mask
, order
, NULL
);
1520 totalpages
= mem_cgroup_get_limit(memcg
) >> PAGE_SHIFT
? : 1;
1521 for_each_mem_cgroup_tree(iter
, memcg
) {
1522 struct cgroup
*cgroup
= iter
->css
.cgroup
;
1523 struct cgroup_iter it
;
1524 struct task_struct
*task
;
1526 cgroup_iter_start(cgroup
, &it
);
1527 while ((task
= cgroup_iter_next(cgroup
, &it
))) {
1528 switch (oom_scan_process_thread(task
, totalpages
, NULL
,
1530 case OOM_SCAN_SELECT
:
1532 put_task_struct(chosen
);
1534 chosen_points
= ULONG_MAX
;
1535 get_task_struct(chosen
);
1537 case OOM_SCAN_CONTINUE
:
1539 case OOM_SCAN_ABORT
:
1540 cgroup_iter_end(cgroup
, &it
);
1541 mem_cgroup_iter_break(memcg
, iter
);
1543 put_task_struct(chosen
);
1548 points
= oom_badness(task
, memcg
, NULL
, totalpages
);
1549 if (points
> chosen_points
) {
1551 put_task_struct(chosen
);
1553 chosen_points
= points
;
1554 get_task_struct(chosen
);
1557 cgroup_iter_end(cgroup
, &it
);
1562 points
= chosen_points
* 1000 / totalpages
;
1563 oom_kill_process(chosen
, gfp_mask
, order
, points
, totalpages
, memcg
,
1564 NULL
, "Memory cgroup out of memory");
1567 static unsigned long mem_cgroup_reclaim(struct mem_cgroup
*memcg
,
1569 unsigned long flags
)
1571 unsigned long total
= 0;
1572 bool noswap
= false;
1575 if (flags
& MEM_CGROUP_RECLAIM_NOSWAP
)
1577 if (!(flags
& MEM_CGROUP_RECLAIM_SHRINK
) && memcg
->memsw_is_minimum
)
1580 for (loop
= 0; loop
< MEM_CGROUP_MAX_RECLAIM_LOOPS
; loop
++) {
1582 drain_all_stock_async(memcg
);
1583 total
+= try_to_free_mem_cgroup_pages(memcg
, gfp_mask
, noswap
);
1585 * Allow limit shrinkers, which are triggered directly
1586 * by userspace, to catch signals and stop reclaim
1587 * after minimal progress, regardless of the margin.
1589 if (total
&& (flags
& MEM_CGROUP_RECLAIM_SHRINK
))
1591 if (mem_cgroup_margin(memcg
))
1594 * If nothing was reclaimed after two attempts, there
1595 * may be no reclaimable pages in this hierarchy.
1604 * test_mem_cgroup_node_reclaimable
1605 * @memcg: the target memcg
1606 * @nid: the node ID to be checked.
1607 * @noswap : specify true here if the user wants flle only information.
1609 * This function returns whether the specified memcg contains any
1610 * reclaimable pages on a node. Returns true if there are any reclaimable
1611 * pages in the node.
1613 static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup
*memcg
,
1614 int nid
, bool noswap
)
1616 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_FILE
))
1618 if (noswap
|| !total_swap_pages
)
1620 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_ANON
))
1625 #if MAX_NUMNODES > 1
1628 * Always updating the nodemask is not very good - even if we have an empty
1629 * list or the wrong list here, we can start from some node and traverse all
1630 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1633 static void mem_cgroup_may_update_nodemask(struct mem_cgroup
*memcg
)
1637 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1638 * pagein/pageout changes since the last update.
1640 if (!atomic_read(&memcg
->numainfo_events
))
1642 if (atomic_inc_return(&memcg
->numainfo_updating
) > 1)
1645 /* make a nodemask where this memcg uses memory from */
1646 memcg
->scan_nodes
= node_states
[N_MEMORY
];
1648 for_each_node_mask(nid
, node_states
[N_MEMORY
]) {
1650 if (!test_mem_cgroup_node_reclaimable(memcg
, nid
, false))
1651 node_clear(nid
, memcg
->scan_nodes
);
1654 atomic_set(&memcg
->numainfo_events
, 0);
1655 atomic_set(&memcg
->numainfo_updating
, 0);
1659 * Selecting a node where we start reclaim from. Because what we need is just
1660 * reducing usage counter, start from anywhere is O,K. Considering
1661 * memory reclaim from current node, there are pros. and cons.
1663 * Freeing memory from current node means freeing memory from a node which
1664 * we'll use or we've used. So, it may make LRU bad. And if several threads
1665 * hit limits, it will see a contention on a node. But freeing from remote
1666 * node means more costs for memory reclaim because of memory latency.
1668 * Now, we use round-robin. Better algorithm is welcomed.
1670 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1674 mem_cgroup_may_update_nodemask(memcg
);
1675 node
= memcg
->last_scanned_node
;
1677 node
= next_node(node
, memcg
->scan_nodes
);
1678 if (node
== MAX_NUMNODES
)
1679 node
= first_node(memcg
->scan_nodes
);
1681 * We call this when we hit limit, not when pages are added to LRU.
1682 * No LRU may hold pages because all pages are UNEVICTABLE or
1683 * memcg is too small and all pages are not on LRU. In that case,
1684 * we use curret node.
1686 if (unlikely(node
== MAX_NUMNODES
))
1687 node
= numa_node_id();
1689 memcg
->last_scanned_node
= node
;
1694 * Check all nodes whether it contains reclaimable pages or not.
1695 * For quick scan, we make use of scan_nodes. This will allow us to skip
1696 * unused nodes. But scan_nodes is lazily updated and may not cotain
1697 * enough new information. We need to do double check.
1699 static bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1704 * quick check...making use of scan_node.
1705 * We can skip unused nodes.
1707 if (!nodes_empty(memcg
->scan_nodes
)) {
1708 for (nid
= first_node(memcg
->scan_nodes
);
1710 nid
= next_node(nid
, memcg
->scan_nodes
)) {
1712 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1717 * Check rest of nodes.
1719 for_each_node_state(nid
, N_MEMORY
) {
1720 if (node_isset(nid
, memcg
->scan_nodes
))
1722 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1729 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1734 static bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1736 return test_mem_cgroup_node_reclaimable(memcg
, 0, noswap
);
1740 static int mem_cgroup_soft_reclaim(struct mem_cgroup
*root_memcg
,
1743 unsigned long *total_scanned
)
1745 struct mem_cgroup
*victim
= NULL
;
1748 unsigned long excess
;
1749 unsigned long nr_scanned
;
1750 struct mem_cgroup_reclaim_cookie reclaim
= {
1755 excess
= res_counter_soft_limit_excess(&root_memcg
->res
) >> PAGE_SHIFT
;
1758 victim
= mem_cgroup_iter(root_memcg
, victim
, &reclaim
);
1763 * If we have not been able to reclaim
1764 * anything, it might because there are
1765 * no reclaimable pages under this hierarchy
1770 * We want to do more targeted reclaim.
1771 * excess >> 2 is not to excessive so as to
1772 * reclaim too much, nor too less that we keep
1773 * coming back to reclaim from this cgroup
1775 if (total
>= (excess
>> 2) ||
1776 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
))
1781 if (!mem_cgroup_reclaimable(victim
, false))
1783 total
+= mem_cgroup_shrink_node_zone(victim
, gfp_mask
, false,
1785 *total_scanned
+= nr_scanned
;
1786 if (!res_counter_soft_limit_excess(&root_memcg
->res
))
1789 mem_cgroup_iter_break(root_memcg
, victim
);
1794 * Check OOM-Killer is already running under our hierarchy.
1795 * If someone is running, return false.
1796 * Has to be called with memcg_oom_lock
1798 static bool mem_cgroup_oom_lock(struct mem_cgroup
*memcg
)
1800 struct mem_cgroup
*iter
, *failed
= NULL
;
1802 for_each_mem_cgroup_tree(iter
, memcg
) {
1803 if (iter
->oom_lock
) {
1805 * this subtree of our hierarchy is already locked
1806 * so we cannot give a lock.
1809 mem_cgroup_iter_break(memcg
, iter
);
1812 iter
->oom_lock
= true;
1819 * OK, we failed to lock the whole subtree so we have to clean up
1820 * what we set up to the failing subtree
1822 for_each_mem_cgroup_tree(iter
, memcg
) {
1823 if (iter
== failed
) {
1824 mem_cgroup_iter_break(memcg
, iter
);
1827 iter
->oom_lock
= false;
1833 * Has to be called with memcg_oom_lock
1835 static int mem_cgroup_oom_unlock(struct mem_cgroup
*memcg
)
1837 struct mem_cgroup
*iter
;
1839 for_each_mem_cgroup_tree(iter
, memcg
)
1840 iter
->oom_lock
= false;
1844 static void mem_cgroup_mark_under_oom(struct mem_cgroup
*memcg
)
1846 struct mem_cgroup
*iter
;
1848 for_each_mem_cgroup_tree(iter
, memcg
)
1849 atomic_inc(&iter
->under_oom
);
1852 static void mem_cgroup_unmark_under_oom(struct mem_cgroup
*memcg
)
1854 struct mem_cgroup
*iter
;
1857 * When a new child is created while the hierarchy is under oom,
1858 * mem_cgroup_oom_lock() may not be called. We have to use
1859 * atomic_add_unless() here.
1861 for_each_mem_cgroup_tree(iter
, memcg
)
1862 atomic_add_unless(&iter
->under_oom
, -1, 0);
1865 static DEFINE_SPINLOCK(memcg_oom_lock
);
1866 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1868 struct oom_wait_info
{
1869 struct mem_cgroup
*memcg
;
1873 static int memcg_oom_wake_function(wait_queue_t
*wait
,
1874 unsigned mode
, int sync
, void *arg
)
1876 struct mem_cgroup
*wake_memcg
= (struct mem_cgroup
*)arg
;
1877 struct mem_cgroup
*oom_wait_memcg
;
1878 struct oom_wait_info
*oom_wait_info
;
1880 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1881 oom_wait_memcg
= oom_wait_info
->memcg
;
1884 * Both of oom_wait_info->memcg and wake_memcg are stable under us.
1885 * Then we can use css_is_ancestor without taking care of RCU.
1887 if (!mem_cgroup_same_or_subtree(oom_wait_memcg
, wake_memcg
)
1888 && !mem_cgroup_same_or_subtree(wake_memcg
, oom_wait_memcg
))
1890 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1893 static void memcg_wakeup_oom(struct mem_cgroup
*memcg
)
1895 /* for filtering, pass "memcg" as argument. */
1896 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, memcg
);
1899 static void memcg_oom_recover(struct mem_cgroup
*memcg
)
1901 if (memcg
&& atomic_read(&memcg
->under_oom
))
1902 memcg_wakeup_oom(memcg
);
1906 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1908 static bool mem_cgroup_handle_oom(struct mem_cgroup
*memcg
, gfp_t mask
,
1911 struct oom_wait_info owait
;
1912 bool locked
, need_to_kill
;
1914 owait
.memcg
= memcg
;
1915 owait
.wait
.flags
= 0;
1916 owait
.wait
.func
= memcg_oom_wake_function
;
1917 owait
.wait
.private = current
;
1918 INIT_LIST_HEAD(&owait
.wait
.task_list
);
1919 need_to_kill
= true;
1920 mem_cgroup_mark_under_oom(memcg
);
1922 /* At first, try to OOM lock hierarchy under memcg.*/
1923 spin_lock(&memcg_oom_lock
);
1924 locked
= mem_cgroup_oom_lock(memcg
);
1926 * Even if signal_pending(), we can't quit charge() loop without
1927 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1928 * under OOM is always welcomed, use TASK_KILLABLE here.
1930 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1931 if (!locked
|| memcg
->oom_kill_disable
)
1932 need_to_kill
= false;
1934 mem_cgroup_oom_notify(memcg
);
1935 spin_unlock(&memcg_oom_lock
);
1938 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1939 mem_cgroup_out_of_memory(memcg
, mask
, order
);
1942 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1944 spin_lock(&memcg_oom_lock
);
1946 mem_cgroup_oom_unlock(memcg
);
1947 memcg_wakeup_oom(memcg
);
1948 spin_unlock(&memcg_oom_lock
);
1950 mem_cgroup_unmark_under_oom(memcg
);
1952 if (test_thread_flag(TIF_MEMDIE
) || fatal_signal_pending(current
))
1954 /* Give chance to dying process */
1955 schedule_timeout_uninterruptible(1);
1960 * Currently used to update mapped file statistics, but the routine can be
1961 * generalized to update other statistics as well.
1963 * Notes: Race condition
1965 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1966 * it tends to be costly. But considering some conditions, we doesn't need
1967 * to do so _always_.
1969 * Considering "charge", lock_page_cgroup() is not required because all
1970 * file-stat operations happen after a page is attached to radix-tree. There
1971 * are no race with "charge".
1973 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1974 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1975 * if there are race with "uncharge". Statistics itself is properly handled
1978 * Considering "move", this is an only case we see a race. To make the race
1979 * small, we check mm->moving_account and detect there are possibility of race
1980 * If there is, we take a lock.
1983 void __mem_cgroup_begin_update_page_stat(struct page
*page
,
1984 bool *locked
, unsigned long *flags
)
1986 struct mem_cgroup
*memcg
;
1987 struct page_cgroup
*pc
;
1989 pc
= lookup_page_cgroup(page
);
1991 memcg
= pc
->mem_cgroup
;
1992 if (unlikely(!memcg
|| !PageCgroupUsed(pc
)))
1995 * If this memory cgroup is not under account moving, we don't
1996 * need to take move_lock_mem_cgroup(). Because we already hold
1997 * rcu_read_lock(), any calls to move_account will be delayed until
1998 * rcu_read_unlock() if mem_cgroup_stolen() == true.
2000 if (!mem_cgroup_stolen(memcg
))
2003 move_lock_mem_cgroup(memcg
, flags
);
2004 if (memcg
!= pc
->mem_cgroup
|| !PageCgroupUsed(pc
)) {
2005 move_unlock_mem_cgroup(memcg
, flags
);
2011 void __mem_cgroup_end_update_page_stat(struct page
*page
, unsigned long *flags
)
2013 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
2016 * It's guaranteed that pc->mem_cgroup never changes while
2017 * lock is held because a routine modifies pc->mem_cgroup
2018 * should take move_lock_mem_cgroup().
2020 move_unlock_mem_cgroup(pc
->mem_cgroup
, flags
);
2023 void mem_cgroup_update_page_stat(struct page
*page
,
2024 enum mem_cgroup_page_stat_item idx
, int val
)
2026 struct mem_cgroup
*memcg
;
2027 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
2028 unsigned long uninitialized_var(flags
);
2030 if (mem_cgroup_disabled())
2033 memcg
= pc
->mem_cgroup
;
2034 if (unlikely(!memcg
|| !PageCgroupUsed(pc
)))
2038 case MEMCG_NR_FILE_MAPPED
:
2039 idx
= MEM_CGROUP_STAT_FILE_MAPPED
;
2045 this_cpu_add(memcg
->stat
->count
[idx
], val
);
2049 * size of first charge trial. "32" comes from vmscan.c's magic value.
2050 * TODO: maybe necessary to use big numbers in big irons.
2052 #define CHARGE_BATCH 32U
2053 struct memcg_stock_pcp
{
2054 struct mem_cgroup
*cached
; /* this never be root cgroup */
2055 unsigned int nr_pages
;
2056 struct work_struct work
;
2057 unsigned long flags
;
2058 #define FLUSHING_CACHED_CHARGE 0
2060 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
2061 static DEFINE_MUTEX(percpu_charge_mutex
);
2064 * Try to consume stocked charge on this cpu. If success, one page is consumed
2065 * from local stock and true is returned. If the stock is 0 or charges from a
2066 * cgroup which is not current target, returns false. This stock will be
2069 static bool consume_stock(struct mem_cgroup
*memcg
)
2071 struct memcg_stock_pcp
*stock
;
2074 stock
= &get_cpu_var(memcg_stock
);
2075 if (memcg
== stock
->cached
&& stock
->nr_pages
)
2077 else /* need to call res_counter_charge */
2079 put_cpu_var(memcg_stock
);
2084 * Returns stocks cached in percpu to res_counter and reset cached information.
2086 static void drain_stock(struct memcg_stock_pcp
*stock
)
2088 struct mem_cgroup
*old
= stock
->cached
;
2090 if (stock
->nr_pages
) {
2091 unsigned long bytes
= stock
->nr_pages
* PAGE_SIZE
;
2093 res_counter_uncharge(&old
->res
, bytes
);
2094 if (do_swap_account
)
2095 res_counter_uncharge(&old
->memsw
, bytes
);
2096 stock
->nr_pages
= 0;
2098 stock
->cached
= NULL
;
2102 * This must be called under preempt disabled or must be called by
2103 * a thread which is pinned to local cpu.
2105 static void drain_local_stock(struct work_struct
*dummy
)
2107 struct memcg_stock_pcp
*stock
= &__get_cpu_var(memcg_stock
);
2109 clear_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
);
2113 * Cache charges(val) which is from res_counter, to local per_cpu area.
2114 * This will be consumed by consume_stock() function, later.
2116 static void refill_stock(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
2118 struct memcg_stock_pcp
*stock
= &get_cpu_var(memcg_stock
);
2120 if (stock
->cached
!= memcg
) { /* reset if necessary */
2122 stock
->cached
= memcg
;
2124 stock
->nr_pages
+= nr_pages
;
2125 put_cpu_var(memcg_stock
);
2129 * Drains all per-CPU charge caches for given root_memcg resp. subtree
2130 * of the hierarchy under it. sync flag says whether we should block
2131 * until the work is done.
2133 static void drain_all_stock(struct mem_cgroup
*root_memcg
, bool sync
)
2137 /* Notify other cpus that system-wide "drain" is running */
2140 for_each_online_cpu(cpu
) {
2141 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2142 struct mem_cgroup
*memcg
;
2144 memcg
= stock
->cached
;
2145 if (!memcg
|| !stock
->nr_pages
)
2147 if (!mem_cgroup_same_or_subtree(root_memcg
, memcg
))
2149 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
)) {
2151 drain_local_stock(&stock
->work
);
2153 schedule_work_on(cpu
, &stock
->work
);
2161 for_each_online_cpu(cpu
) {
2162 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2163 if (test_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
))
2164 flush_work(&stock
->work
);
2171 * Tries to drain stocked charges in other cpus. This function is asynchronous
2172 * and just put a work per cpu for draining localy on each cpu. Caller can
2173 * expects some charges will be back to res_counter later but cannot wait for
2176 static void drain_all_stock_async(struct mem_cgroup
*root_memcg
)
2179 * If someone calls draining, avoid adding more kworker runs.
2181 if (!mutex_trylock(&percpu_charge_mutex
))
2183 drain_all_stock(root_memcg
, false);
2184 mutex_unlock(&percpu_charge_mutex
);
2187 /* This is a synchronous drain interface. */
2188 static void drain_all_stock_sync(struct mem_cgroup
*root_memcg
)
2190 /* called when force_empty is called */
2191 mutex_lock(&percpu_charge_mutex
);
2192 drain_all_stock(root_memcg
, true);
2193 mutex_unlock(&percpu_charge_mutex
);
2197 * This function drains percpu counter value from DEAD cpu and
2198 * move it to local cpu. Note that this function can be preempted.
2200 static void mem_cgroup_drain_pcp_counter(struct mem_cgroup
*memcg
, int cpu
)
2204 spin_lock(&memcg
->pcp_counter_lock
);
2205 for (i
= 0; i
< MEM_CGROUP_STAT_NSTATS
; i
++) {
2206 long x
= per_cpu(memcg
->stat
->count
[i
], cpu
);
2208 per_cpu(memcg
->stat
->count
[i
], cpu
) = 0;
2209 memcg
->nocpu_base
.count
[i
] += x
;
2211 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
2212 unsigned long x
= per_cpu(memcg
->stat
->events
[i
], cpu
);
2214 per_cpu(memcg
->stat
->events
[i
], cpu
) = 0;
2215 memcg
->nocpu_base
.events
[i
] += x
;
2217 spin_unlock(&memcg
->pcp_counter_lock
);
2220 static int __cpuinit
memcg_cpu_hotplug_callback(struct notifier_block
*nb
,
2221 unsigned long action
,
2224 int cpu
= (unsigned long)hcpu
;
2225 struct memcg_stock_pcp
*stock
;
2226 struct mem_cgroup
*iter
;
2228 if (action
== CPU_ONLINE
)
2231 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
2234 for_each_mem_cgroup(iter
)
2235 mem_cgroup_drain_pcp_counter(iter
, cpu
);
2237 stock
= &per_cpu(memcg_stock
, cpu
);
2243 /* See __mem_cgroup_try_charge() for details */
2245 CHARGE_OK
, /* success */
2246 CHARGE_RETRY
, /* need to retry but retry is not bad */
2247 CHARGE_NOMEM
, /* we can't do more. return -ENOMEM */
2248 CHARGE_WOULDBLOCK
, /* GFP_WAIT wasn't set and no enough res. */
2249 CHARGE_OOM_DIE
, /* the current is killed because of OOM */
2252 static int mem_cgroup_do_charge(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
2253 unsigned int nr_pages
, bool oom_check
)
2255 unsigned long csize
= nr_pages
* PAGE_SIZE
;
2256 struct mem_cgroup
*mem_over_limit
;
2257 struct res_counter
*fail_res
;
2258 unsigned long flags
= 0;
2261 ret
= res_counter_charge(&memcg
->res
, csize
, &fail_res
);
2264 if (!do_swap_account
)
2266 ret
= res_counter_charge(&memcg
->memsw
, csize
, &fail_res
);
2270 res_counter_uncharge(&memcg
->res
, csize
);
2271 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, memsw
);
2272 flags
|= MEM_CGROUP_RECLAIM_NOSWAP
;
2274 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, res
);
2276 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2277 * of regular pages (CHARGE_BATCH), or a single regular page (1).
2279 * Never reclaim on behalf of optional batching, retry with a
2280 * single page instead.
2282 if (nr_pages
== CHARGE_BATCH
)
2283 return CHARGE_RETRY
;
2285 if (!(gfp_mask
& __GFP_WAIT
))
2286 return CHARGE_WOULDBLOCK
;
2288 ret
= mem_cgroup_reclaim(mem_over_limit
, gfp_mask
, flags
);
2289 if (mem_cgroup_margin(mem_over_limit
) >= nr_pages
)
2290 return CHARGE_RETRY
;
2292 * Even though the limit is exceeded at this point, reclaim
2293 * may have been able to free some pages. Retry the charge
2294 * before killing the task.
2296 * Only for regular pages, though: huge pages are rather
2297 * unlikely to succeed so close to the limit, and we fall back
2298 * to regular pages anyway in case of failure.
2300 if (nr_pages
== 1 && ret
)
2301 return CHARGE_RETRY
;
2304 * At task move, charge accounts can be doubly counted. So, it's
2305 * better to wait until the end of task_move if something is going on.
2307 if (mem_cgroup_wait_acct_move(mem_over_limit
))
2308 return CHARGE_RETRY
;
2310 /* If we don't need to call oom-killer at el, return immediately */
2312 return CHARGE_NOMEM
;
2314 if (!mem_cgroup_handle_oom(mem_over_limit
, gfp_mask
, get_order(csize
)))
2315 return CHARGE_OOM_DIE
;
2317 return CHARGE_RETRY
;
2321 * __mem_cgroup_try_charge() does
2322 * 1. detect memcg to be charged against from passed *mm and *ptr,
2323 * 2. update res_counter
2324 * 3. call memory reclaim if necessary.
2326 * In some special case, if the task is fatal, fatal_signal_pending() or
2327 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2328 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2329 * as possible without any hazards. 2: all pages should have a valid
2330 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2331 * pointer, that is treated as a charge to root_mem_cgroup.
2333 * So __mem_cgroup_try_charge() will return
2334 * 0 ... on success, filling *ptr with a valid memcg pointer.
2335 * -ENOMEM ... charge failure because of resource limits.
2336 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2338 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2339 * the oom-killer can be invoked.
2341 static int __mem_cgroup_try_charge(struct mm_struct
*mm
,
2343 unsigned int nr_pages
,
2344 struct mem_cgroup
**ptr
,
2347 unsigned int batch
= max(CHARGE_BATCH
, nr_pages
);
2348 int nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2349 struct mem_cgroup
*memcg
= NULL
;
2353 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2354 * in system level. So, allow to go ahead dying process in addition to
2357 if (unlikely(test_thread_flag(TIF_MEMDIE
)
2358 || fatal_signal_pending(current
)))
2362 * We always charge the cgroup the mm_struct belongs to.
2363 * The mm_struct's mem_cgroup changes on task migration if the
2364 * thread group leader migrates. It's possible that mm is not
2365 * set, if so charge the root memcg (happens for pagecache usage).
2368 *ptr
= root_mem_cgroup
;
2370 if (*ptr
) { /* css should be a valid one */
2372 if (mem_cgroup_is_root(memcg
))
2374 if (nr_pages
== 1 && consume_stock(memcg
))
2376 css_get(&memcg
->css
);
2378 struct task_struct
*p
;
2381 p
= rcu_dereference(mm
->owner
);
2383 * Because we don't have task_lock(), "p" can exit.
2384 * In that case, "memcg" can point to root or p can be NULL with
2385 * race with swapoff. Then, we have small risk of mis-accouning.
2386 * But such kind of mis-account by race always happens because
2387 * we don't have cgroup_mutex(). It's overkill and we allo that
2389 * (*) swapoff at el will charge against mm-struct not against
2390 * task-struct. So, mm->owner can be NULL.
2392 memcg
= mem_cgroup_from_task(p
);
2394 memcg
= root_mem_cgroup
;
2395 if (mem_cgroup_is_root(memcg
)) {
2399 if (nr_pages
== 1 && consume_stock(memcg
)) {
2401 * It seems dagerous to access memcg without css_get().
2402 * But considering how consume_stok works, it's not
2403 * necessary. If consume_stock success, some charges
2404 * from this memcg are cached on this cpu. So, we
2405 * don't need to call css_get()/css_tryget() before
2406 * calling consume_stock().
2411 /* after here, we may be blocked. we need to get refcnt */
2412 if (!css_tryget(&memcg
->css
)) {
2422 /* If killed, bypass charge */
2423 if (fatal_signal_pending(current
)) {
2424 css_put(&memcg
->css
);
2429 if (oom
&& !nr_oom_retries
) {
2431 nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2434 ret
= mem_cgroup_do_charge(memcg
, gfp_mask
, batch
, oom_check
);
2438 case CHARGE_RETRY
: /* not in OOM situation but retry */
2440 css_put(&memcg
->css
);
2443 case CHARGE_WOULDBLOCK
: /* !__GFP_WAIT */
2444 css_put(&memcg
->css
);
2446 case CHARGE_NOMEM
: /* OOM routine works */
2448 css_put(&memcg
->css
);
2451 /* If oom, we never return -ENOMEM */
2454 case CHARGE_OOM_DIE
: /* Killed by OOM Killer */
2455 css_put(&memcg
->css
);
2458 } while (ret
!= CHARGE_OK
);
2460 if (batch
> nr_pages
)
2461 refill_stock(memcg
, batch
- nr_pages
);
2462 css_put(&memcg
->css
);
2470 *ptr
= root_mem_cgroup
;
2475 * Somemtimes we have to undo a charge we got by try_charge().
2476 * This function is for that and do uncharge, put css's refcnt.
2477 * gotten by try_charge().
2479 static void __mem_cgroup_cancel_charge(struct mem_cgroup
*memcg
,
2480 unsigned int nr_pages
)
2482 if (!mem_cgroup_is_root(memcg
)) {
2483 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2485 res_counter_uncharge(&memcg
->res
, bytes
);
2486 if (do_swap_account
)
2487 res_counter_uncharge(&memcg
->memsw
, bytes
);
2492 * Cancel chrages in this cgroup....doesn't propagate to parent cgroup.
2493 * This is useful when moving usage to parent cgroup.
2495 static void __mem_cgroup_cancel_local_charge(struct mem_cgroup
*memcg
,
2496 unsigned int nr_pages
)
2498 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2500 if (mem_cgroup_is_root(memcg
))
2503 res_counter_uncharge_until(&memcg
->res
, memcg
->res
.parent
, bytes
);
2504 if (do_swap_account
)
2505 res_counter_uncharge_until(&memcg
->memsw
,
2506 memcg
->memsw
.parent
, bytes
);
2510 * A helper function to get mem_cgroup from ID. must be called under
2511 * rcu_read_lock(). The caller is responsible for calling css_tryget if
2512 * the mem_cgroup is used for charging. (dropping refcnt from swap can be
2513 * called against removed memcg.)
2515 static struct mem_cgroup
*mem_cgroup_lookup(unsigned short id
)
2517 struct cgroup_subsys_state
*css
;
2519 /* ID 0 is unused ID */
2522 css
= css_lookup(&mem_cgroup_subsys
, id
);
2525 return mem_cgroup_from_css(css
);
2528 struct mem_cgroup
*try_get_mem_cgroup_from_page(struct page
*page
)
2530 struct mem_cgroup
*memcg
= NULL
;
2531 struct page_cgroup
*pc
;
2535 VM_BUG_ON(!PageLocked(page
));
2537 pc
= lookup_page_cgroup(page
);
2538 lock_page_cgroup(pc
);
2539 if (PageCgroupUsed(pc
)) {
2540 memcg
= pc
->mem_cgroup
;
2541 if (memcg
&& !css_tryget(&memcg
->css
))
2543 } else if (PageSwapCache(page
)) {
2544 ent
.val
= page_private(page
);
2545 id
= lookup_swap_cgroup_id(ent
);
2547 memcg
= mem_cgroup_lookup(id
);
2548 if (memcg
&& !css_tryget(&memcg
->css
))
2552 unlock_page_cgroup(pc
);
2556 static void __mem_cgroup_commit_charge(struct mem_cgroup
*memcg
,
2558 unsigned int nr_pages
,
2559 enum charge_type ctype
,
2562 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
2563 struct zone
*uninitialized_var(zone
);
2564 struct lruvec
*lruvec
;
2565 bool was_on_lru
= false;
2568 lock_page_cgroup(pc
);
2569 VM_BUG_ON(PageCgroupUsed(pc
));
2571 * we don't need page_cgroup_lock about tail pages, becase they are not
2572 * accessed by any other context at this point.
2576 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2577 * may already be on some other mem_cgroup's LRU. Take care of it.
2580 zone
= page_zone(page
);
2581 spin_lock_irq(&zone
->lru_lock
);
2582 if (PageLRU(page
)) {
2583 lruvec
= mem_cgroup_zone_lruvec(zone
, pc
->mem_cgroup
);
2585 del_page_from_lru_list(page
, lruvec
, page_lru(page
));
2590 pc
->mem_cgroup
= memcg
;
2592 * We access a page_cgroup asynchronously without lock_page_cgroup().
2593 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2594 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2595 * before USED bit, we need memory barrier here.
2596 * See mem_cgroup_add_lru_list(), etc.
2599 SetPageCgroupUsed(pc
);
2603 lruvec
= mem_cgroup_zone_lruvec(zone
, pc
->mem_cgroup
);
2604 VM_BUG_ON(PageLRU(page
));
2606 add_page_to_lru_list(page
, lruvec
, page_lru(page
));
2608 spin_unlock_irq(&zone
->lru_lock
);
2611 if (ctype
== MEM_CGROUP_CHARGE_TYPE_ANON
)
2616 mem_cgroup_charge_statistics(memcg
, anon
, nr_pages
);
2617 unlock_page_cgroup(pc
);
2620 * "charge_statistics" updated event counter. Then, check it.
2621 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2622 * if they exceeds softlimit.
2624 memcg_check_events(memcg
, page
);
2627 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2629 #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION)
2631 * Because tail pages are not marked as "used", set it. We're under
2632 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2633 * charge/uncharge will be never happen and move_account() is done under
2634 * compound_lock(), so we don't have to take care of races.
2636 void mem_cgroup_split_huge_fixup(struct page
*head
)
2638 struct page_cgroup
*head_pc
= lookup_page_cgroup(head
);
2639 struct page_cgroup
*pc
;
2642 if (mem_cgroup_disabled())
2644 for (i
= 1; i
< HPAGE_PMD_NR
; i
++) {
2646 pc
->mem_cgroup
= head_pc
->mem_cgroup
;
2647 smp_wmb();/* see __commit_charge() */
2648 pc
->flags
= head_pc
->flags
& ~PCGF_NOCOPY_AT_SPLIT
;
2651 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2654 * mem_cgroup_move_account - move account of the page
2656 * @nr_pages: number of regular pages (>1 for huge pages)
2657 * @pc: page_cgroup of the page.
2658 * @from: mem_cgroup which the page is moved from.
2659 * @to: mem_cgroup which the page is moved to. @from != @to.
2661 * The caller must confirm following.
2662 * - page is not on LRU (isolate_page() is useful.)
2663 * - compound_lock is held when nr_pages > 1
2665 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
2668 static int mem_cgroup_move_account(struct page
*page
,
2669 unsigned int nr_pages
,
2670 struct page_cgroup
*pc
,
2671 struct mem_cgroup
*from
,
2672 struct mem_cgroup
*to
)
2674 unsigned long flags
;
2676 bool anon
= PageAnon(page
);
2678 VM_BUG_ON(from
== to
);
2679 VM_BUG_ON(PageLRU(page
));
2681 * The page is isolated from LRU. So, collapse function
2682 * will not handle this page. But page splitting can happen.
2683 * Do this check under compound_page_lock(). The caller should
2687 if (nr_pages
> 1 && !PageTransHuge(page
))
2690 lock_page_cgroup(pc
);
2693 if (!PageCgroupUsed(pc
) || pc
->mem_cgroup
!= from
)
2696 move_lock_mem_cgroup(from
, &flags
);
2698 if (!anon
&& page_mapped(page
)) {
2699 /* Update mapped_file data for mem_cgroup */
2701 __this_cpu_dec(from
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2702 __this_cpu_inc(to
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2705 mem_cgroup_charge_statistics(from
, anon
, -nr_pages
);
2707 /* caller should have done css_get */
2708 pc
->mem_cgroup
= to
;
2709 mem_cgroup_charge_statistics(to
, anon
, nr_pages
);
2710 move_unlock_mem_cgroup(from
, &flags
);
2713 unlock_page_cgroup(pc
);
2717 memcg_check_events(to
, page
);
2718 memcg_check_events(from
, page
);
2724 * mem_cgroup_move_parent - moves page to the parent group
2725 * @page: the page to move
2726 * @pc: page_cgroup of the page
2727 * @child: page's cgroup
2729 * move charges to its parent or the root cgroup if the group has no
2730 * parent (aka use_hierarchy==0).
2731 * Although this might fail (get_page_unless_zero, isolate_lru_page or
2732 * mem_cgroup_move_account fails) the failure is always temporary and
2733 * it signals a race with a page removal/uncharge or migration. In the
2734 * first case the page is on the way out and it will vanish from the LRU
2735 * on the next attempt and the call should be retried later.
2736 * Isolation from the LRU fails only if page has been isolated from
2737 * the LRU since we looked at it and that usually means either global
2738 * reclaim or migration going on. The page will either get back to the
2740 * Finaly mem_cgroup_move_account fails only if the page got uncharged
2741 * (!PageCgroupUsed) or moved to a different group. The page will
2742 * disappear in the next attempt.
2744 static int mem_cgroup_move_parent(struct page
*page
,
2745 struct page_cgroup
*pc
,
2746 struct mem_cgroup
*child
)
2748 struct mem_cgroup
*parent
;
2749 unsigned int nr_pages
;
2750 unsigned long uninitialized_var(flags
);
2753 VM_BUG_ON(mem_cgroup_is_root(child
));
2756 if (!get_page_unless_zero(page
))
2758 if (isolate_lru_page(page
))
2761 nr_pages
= hpage_nr_pages(page
);
2763 parent
= parent_mem_cgroup(child
);
2765 * If no parent, move charges to root cgroup.
2768 parent
= root_mem_cgroup
;
2771 VM_BUG_ON(!PageTransHuge(page
));
2772 flags
= compound_lock_irqsave(page
);
2775 ret
= mem_cgroup_move_account(page
, nr_pages
,
2778 __mem_cgroup_cancel_local_charge(child
, nr_pages
);
2781 compound_unlock_irqrestore(page
, flags
);
2782 putback_lru_page(page
);
2790 * Charge the memory controller for page usage.
2792 * 0 if the charge was successful
2793 * < 0 if the cgroup is over its limit
2795 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
2796 gfp_t gfp_mask
, enum charge_type ctype
)
2798 struct mem_cgroup
*memcg
= NULL
;
2799 unsigned int nr_pages
= 1;
2803 if (PageTransHuge(page
)) {
2804 nr_pages
<<= compound_order(page
);
2805 VM_BUG_ON(!PageTransHuge(page
));
2807 * Never OOM-kill a process for a huge page. The
2808 * fault handler will fall back to regular pages.
2813 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, nr_pages
, &memcg
, oom
);
2816 __mem_cgroup_commit_charge(memcg
, page
, nr_pages
, ctype
, false);
2820 int mem_cgroup_newpage_charge(struct page
*page
,
2821 struct mm_struct
*mm
, gfp_t gfp_mask
)
2823 if (mem_cgroup_disabled())
2825 VM_BUG_ON(page_mapped(page
));
2826 VM_BUG_ON(page
->mapping
&& !PageAnon(page
));
2828 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2829 MEM_CGROUP_CHARGE_TYPE_ANON
);
2833 * While swap-in, try_charge -> commit or cancel, the page is locked.
2834 * And when try_charge() successfully returns, one refcnt to memcg without
2835 * struct page_cgroup is acquired. This refcnt will be consumed by
2836 * "commit()" or removed by "cancel()"
2838 static int __mem_cgroup_try_charge_swapin(struct mm_struct
*mm
,
2841 struct mem_cgroup
**memcgp
)
2843 struct mem_cgroup
*memcg
;
2844 struct page_cgroup
*pc
;
2847 pc
= lookup_page_cgroup(page
);
2849 * Every swap fault against a single page tries to charge the
2850 * page, bail as early as possible. shmem_unuse() encounters
2851 * already charged pages, too. The USED bit is protected by
2852 * the page lock, which serializes swap cache removal, which
2853 * in turn serializes uncharging.
2855 if (PageCgroupUsed(pc
))
2857 if (!do_swap_account
)
2859 memcg
= try_get_mem_cgroup_from_page(page
);
2863 ret
= __mem_cgroup_try_charge(NULL
, mask
, 1, memcgp
, true);
2864 css_put(&memcg
->css
);
2869 ret
= __mem_cgroup_try_charge(mm
, mask
, 1, memcgp
, true);
2875 int mem_cgroup_try_charge_swapin(struct mm_struct
*mm
, struct page
*page
,
2876 gfp_t gfp_mask
, struct mem_cgroup
**memcgp
)
2879 if (mem_cgroup_disabled())
2882 * A racing thread's fault, or swapoff, may have already
2883 * updated the pte, and even removed page from swap cache: in
2884 * those cases unuse_pte()'s pte_same() test will fail; but
2885 * there's also a KSM case which does need to charge the page.
2887 if (!PageSwapCache(page
)) {
2890 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, 1, memcgp
, true);
2895 return __mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, memcgp
);
2898 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup
*memcg
)
2900 if (mem_cgroup_disabled())
2904 __mem_cgroup_cancel_charge(memcg
, 1);
2908 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*memcg
,
2909 enum charge_type ctype
)
2911 if (mem_cgroup_disabled())
2916 __mem_cgroup_commit_charge(memcg
, page
, 1, ctype
, true);
2918 * Now swap is on-memory. This means this page may be
2919 * counted both as mem and swap....double count.
2920 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2921 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2922 * may call delete_from_swap_cache() before reach here.
2924 if (do_swap_account
&& PageSwapCache(page
)) {
2925 swp_entry_t ent
= {.val
= page_private(page
)};
2926 mem_cgroup_uncharge_swap(ent
);
2930 void mem_cgroup_commit_charge_swapin(struct page
*page
,
2931 struct mem_cgroup
*memcg
)
2933 __mem_cgroup_commit_charge_swapin(page
, memcg
,
2934 MEM_CGROUP_CHARGE_TYPE_ANON
);
2937 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
2940 struct mem_cgroup
*memcg
= NULL
;
2941 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
2944 if (mem_cgroup_disabled())
2946 if (PageCompound(page
))
2949 if (!PageSwapCache(page
))
2950 ret
= mem_cgroup_charge_common(page
, mm
, gfp_mask
, type
);
2951 else { /* page is swapcache/shmem */
2952 ret
= __mem_cgroup_try_charge_swapin(mm
, page
,
2955 __mem_cgroup_commit_charge_swapin(page
, memcg
, type
);
2960 static void mem_cgroup_do_uncharge(struct mem_cgroup
*memcg
,
2961 unsigned int nr_pages
,
2962 const enum charge_type ctype
)
2964 struct memcg_batch_info
*batch
= NULL
;
2965 bool uncharge_memsw
= true;
2967 /* If swapout, usage of swap doesn't decrease */
2968 if (!do_swap_account
|| ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
)
2969 uncharge_memsw
= false;
2971 batch
= ¤t
->memcg_batch
;
2973 * In usual, we do css_get() when we remember memcg pointer.
2974 * But in this case, we keep res->usage until end of a series of
2975 * uncharges. Then, it's ok to ignore memcg's refcnt.
2978 batch
->memcg
= memcg
;
2980 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2981 * In those cases, all pages freed continuously can be expected to be in
2982 * the same cgroup and we have chance to coalesce uncharges.
2983 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2984 * because we want to do uncharge as soon as possible.
2987 if (!batch
->do_batch
|| test_thread_flag(TIF_MEMDIE
))
2988 goto direct_uncharge
;
2991 goto direct_uncharge
;
2994 * In typical case, batch->memcg == mem. This means we can
2995 * merge a series of uncharges to an uncharge of res_counter.
2996 * If not, we uncharge res_counter ony by one.
2998 if (batch
->memcg
!= memcg
)
2999 goto direct_uncharge
;
3000 /* remember freed charge and uncharge it later */
3003 batch
->memsw_nr_pages
++;
3006 res_counter_uncharge(&memcg
->res
, nr_pages
* PAGE_SIZE
);
3008 res_counter_uncharge(&memcg
->memsw
, nr_pages
* PAGE_SIZE
);
3009 if (unlikely(batch
->memcg
!= memcg
))
3010 memcg_oom_recover(memcg
);
3014 * uncharge if !page_mapped(page)
3016 static struct mem_cgroup
*
3017 __mem_cgroup_uncharge_common(struct page
*page
, enum charge_type ctype
,
3020 struct mem_cgroup
*memcg
= NULL
;
3021 unsigned int nr_pages
= 1;
3022 struct page_cgroup
*pc
;
3025 if (mem_cgroup_disabled())
3028 VM_BUG_ON(PageSwapCache(page
));
3030 if (PageTransHuge(page
)) {
3031 nr_pages
<<= compound_order(page
);
3032 VM_BUG_ON(!PageTransHuge(page
));
3035 * Check if our page_cgroup is valid
3037 pc
= lookup_page_cgroup(page
);
3038 if (unlikely(!PageCgroupUsed(pc
)))
3041 lock_page_cgroup(pc
);
3043 memcg
= pc
->mem_cgroup
;
3045 if (!PageCgroupUsed(pc
))
3048 anon
= PageAnon(page
);
3051 case MEM_CGROUP_CHARGE_TYPE_ANON
:
3053 * Generally PageAnon tells if it's the anon statistics to be
3054 * updated; but sometimes e.g. mem_cgroup_uncharge_page() is
3055 * used before page reached the stage of being marked PageAnon.
3059 case MEM_CGROUP_CHARGE_TYPE_DROP
:
3060 /* See mem_cgroup_prepare_migration() */
3061 if (page_mapped(page
))
3064 * Pages under migration may not be uncharged. But
3065 * end_migration() /must/ be the one uncharging the
3066 * unused post-migration page and so it has to call
3067 * here with the migration bit still set. See the
3068 * res_counter handling below.
3070 if (!end_migration
&& PageCgroupMigration(pc
))
3073 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT
:
3074 if (!PageAnon(page
)) { /* Shared memory */
3075 if (page
->mapping
&& !page_is_file_cache(page
))
3077 } else if (page_mapped(page
)) /* Anon */
3084 mem_cgroup_charge_statistics(memcg
, anon
, -nr_pages
);
3086 ClearPageCgroupUsed(pc
);
3088 * pc->mem_cgroup is not cleared here. It will be accessed when it's
3089 * freed from LRU. This is safe because uncharged page is expected not
3090 * to be reused (freed soon). Exception is SwapCache, it's handled by
3091 * special functions.
3094 unlock_page_cgroup(pc
);
3096 * even after unlock, we have memcg->res.usage here and this memcg
3097 * will never be freed.
3099 memcg_check_events(memcg
, page
);
3100 if (do_swap_account
&& ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
) {
3101 mem_cgroup_swap_statistics(memcg
, true);
3102 mem_cgroup_get(memcg
);
3105 * Migration does not charge the res_counter for the
3106 * replacement page, so leave it alone when phasing out the
3107 * page that is unused after the migration.
3109 if (!end_migration
&& !mem_cgroup_is_root(memcg
))
3110 mem_cgroup_do_uncharge(memcg
, nr_pages
, ctype
);
3115 unlock_page_cgroup(pc
);
3119 void mem_cgroup_uncharge_page(struct page
*page
)
3122 if (page_mapped(page
))
3124 VM_BUG_ON(page
->mapping
&& !PageAnon(page
));
3125 if (PageSwapCache(page
))
3127 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_ANON
, false);
3130 void mem_cgroup_uncharge_cache_page(struct page
*page
)
3132 VM_BUG_ON(page_mapped(page
));
3133 VM_BUG_ON(page
->mapping
);
3134 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_CACHE
, false);
3138 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
3139 * In that cases, pages are freed continuously and we can expect pages
3140 * are in the same memcg. All these calls itself limits the number of
3141 * pages freed at once, then uncharge_start/end() is called properly.
3142 * This may be called prural(2) times in a context,
3145 void mem_cgroup_uncharge_start(void)
3147 current
->memcg_batch
.do_batch
++;
3148 /* We can do nest. */
3149 if (current
->memcg_batch
.do_batch
== 1) {
3150 current
->memcg_batch
.memcg
= NULL
;
3151 current
->memcg_batch
.nr_pages
= 0;
3152 current
->memcg_batch
.memsw_nr_pages
= 0;
3156 void mem_cgroup_uncharge_end(void)
3158 struct memcg_batch_info
*batch
= ¤t
->memcg_batch
;
3160 if (!batch
->do_batch
)
3164 if (batch
->do_batch
) /* If stacked, do nothing. */
3170 * This "batch->memcg" is valid without any css_get/put etc...
3171 * bacause we hide charges behind us.
3173 if (batch
->nr_pages
)
3174 res_counter_uncharge(&batch
->memcg
->res
,
3175 batch
->nr_pages
* PAGE_SIZE
);
3176 if (batch
->memsw_nr_pages
)
3177 res_counter_uncharge(&batch
->memcg
->memsw
,
3178 batch
->memsw_nr_pages
* PAGE_SIZE
);
3179 memcg_oom_recover(batch
->memcg
);
3180 /* forget this pointer (for sanity check) */
3181 batch
->memcg
= NULL
;
3186 * called after __delete_from_swap_cache() and drop "page" account.
3187 * memcg information is recorded to swap_cgroup of "ent"
3190 mem_cgroup_uncharge_swapcache(struct page
*page
, swp_entry_t ent
, bool swapout
)
3192 struct mem_cgroup
*memcg
;
3193 int ctype
= MEM_CGROUP_CHARGE_TYPE_SWAPOUT
;
3195 if (!swapout
) /* this was a swap cache but the swap is unused ! */
3196 ctype
= MEM_CGROUP_CHARGE_TYPE_DROP
;
3198 memcg
= __mem_cgroup_uncharge_common(page
, ctype
, false);
3201 * record memcg information, if swapout && memcg != NULL,
3202 * mem_cgroup_get() was called in uncharge().
3204 if (do_swap_account
&& swapout
&& memcg
)
3205 swap_cgroup_record(ent
, css_id(&memcg
->css
));
3209 #ifdef CONFIG_MEMCG_SWAP
3211 * called from swap_entry_free(). remove record in swap_cgroup and
3212 * uncharge "memsw" account.
3214 void mem_cgroup_uncharge_swap(swp_entry_t ent
)
3216 struct mem_cgroup
*memcg
;
3219 if (!do_swap_account
)
3222 id
= swap_cgroup_record(ent
, 0);
3224 memcg
= mem_cgroup_lookup(id
);
3227 * We uncharge this because swap is freed.
3228 * This memcg can be obsolete one. We avoid calling css_tryget
3230 if (!mem_cgroup_is_root(memcg
))
3231 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
3232 mem_cgroup_swap_statistics(memcg
, false);
3233 mem_cgroup_put(memcg
);
3239 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3240 * @entry: swap entry to be moved
3241 * @from: mem_cgroup which the entry is moved from
3242 * @to: mem_cgroup which the entry is moved to
3244 * It succeeds only when the swap_cgroup's record for this entry is the same
3245 * as the mem_cgroup's id of @from.
3247 * Returns 0 on success, -EINVAL on failure.
3249 * The caller must have charged to @to, IOW, called res_counter_charge() about
3250 * both res and memsw, and called css_get().
3252 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
3253 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
3255 unsigned short old_id
, new_id
;
3257 old_id
= css_id(&from
->css
);
3258 new_id
= css_id(&to
->css
);
3260 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
3261 mem_cgroup_swap_statistics(from
, false);
3262 mem_cgroup_swap_statistics(to
, true);
3264 * This function is only called from task migration context now.
3265 * It postpones res_counter and refcount handling till the end
3266 * of task migration(mem_cgroup_clear_mc()) for performance
3267 * improvement. But we cannot postpone mem_cgroup_get(to)
3268 * because if the process that has been moved to @to does
3269 * swap-in, the refcount of @to might be decreased to 0.
3277 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
3278 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
3285 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3288 void mem_cgroup_prepare_migration(struct page
*page
, struct page
*newpage
,
3289 struct mem_cgroup
**memcgp
)
3291 struct mem_cgroup
*memcg
= NULL
;
3292 unsigned int nr_pages
= 1;
3293 struct page_cgroup
*pc
;
3294 enum charge_type ctype
;
3298 if (mem_cgroup_disabled())
3301 if (PageTransHuge(page
))
3302 nr_pages
<<= compound_order(page
);
3304 pc
= lookup_page_cgroup(page
);
3305 lock_page_cgroup(pc
);
3306 if (PageCgroupUsed(pc
)) {
3307 memcg
= pc
->mem_cgroup
;
3308 css_get(&memcg
->css
);
3310 * At migrating an anonymous page, its mapcount goes down
3311 * to 0 and uncharge() will be called. But, even if it's fully
3312 * unmapped, migration may fail and this page has to be
3313 * charged again. We set MIGRATION flag here and delay uncharge
3314 * until end_migration() is called
3316 * Corner Case Thinking
3318 * When the old page was mapped as Anon and it's unmap-and-freed
3319 * while migration was ongoing.
3320 * If unmap finds the old page, uncharge() of it will be delayed
3321 * until end_migration(). If unmap finds a new page, it's
3322 * uncharged when it make mapcount to be 1->0. If unmap code
3323 * finds swap_migration_entry, the new page will not be mapped
3324 * and end_migration() will find it(mapcount==0).
3327 * When the old page was mapped but migraion fails, the kernel
3328 * remaps it. A charge for it is kept by MIGRATION flag even
3329 * if mapcount goes down to 0. We can do remap successfully
3330 * without charging it again.
3333 * The "old" page is under lock_page() until the end of
3334 * migration, so, the old page itself will not be swapped-out.
3335 * If the new page is swapped out before end_migraton, our
3336 * hook to usual swap-out path will catch the event.
3339 SetPageCgroupMigration(pc
);
3341 unlock_page_cgroup(pc
);
3343 * If the page is not charged at this point,
3351 * We charge new page before it's used/mapped. So, even if unlock_page()
3352 * is called before end_migration, we can catch all events on this new
3353 * page. In the case new page is migrated but not remapped, new page's
3354 * mapcount will be finally 0 and we call uncharge in end_migration().
3357 ctype
= MEM_CGROUP_CHARGE_TYPE_ANON
;
3359 ctype
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3361 * The page is committed to the memcg, but it's not actually
3362 * charged to the res_counter since we plan on replacing the
3363 * old one and only one page is going to be left afterwards.
3365 __mem_cgroup_commit_charge(memcg
, newpage
, nr_pages
, ctype
, false);
3368 /* remove redundant charge if migration failed*/
3369 void mem_cgroup_end_migration(struct mem_cgroup
*memcg
,
3370 struct page
*oldpage
, struct page
*newpage
, bool migration_ok
)
3372 struct page
*used
, *unused
;
3373 struct page_cgroup
*pc
;
3379 if (!migration_ok
) {
3386 anon
= PageAnon(used
);
3387 __mem_cgroup_uncharge_common(unused
,
3388 anon
? MEM_CGROUP_CHARGE_TYPE_ANON
3389 : MEM_CGROUP_CHARGE_TYPE_CACHE
,
3391 css_put(&memcg
->css
);
3393 * We disallowed uncharge of pages under migration because mapcount
3394 * of the page goes down to zero, temporarly.
3395 * Clear the flag and check the page should be charged.
3397 pc
= lookup_page_cgroup(oldpage
);
3398 lock_page_cgroup(pc
);
3399 ClearPageCgroupMigration(pc
);
3400 unlock_page_cgroup(pc
);
3403 * If a page is a file cache, radix-tree replacement is very atomic
3404 * and we can skip this check. When it was an Anon page, its mapcount
3405 * goes down to 0. But because we added MIGRATION flage, it's not
3406 * uncharged yet. There are several case but page->mapcount check
3407 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3408 * check. (see prepare_charge() also)
3411 mem_cgroup_uncharge_page(used
);
3415 * At replace page cache, newpage is not under any memcg but it's on
3416 * LRU. So, this function doesn't touch res_counter but handles LRU
3417 * in correct way. Both pages are locked so we cannot race with uncharge.
3419 void mem_cgroup_replace_page_cache(struct page
*oldpage
,
3420 struct page
*newpage
)
3422 struct mem_cgroup
*memcg
= NULL
;
3423 struct page_cgroup
*pc
;
3424 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3426 if (mem_cgroup_disabled())
3429 pc
= lookup_page_cgroup(oldpage
);
3430 /* fix accounting on old pages */
3431 lock_page_cgroup(pc
);
3432 if (PageCgroupUsed(pc
)) {
3433 memcg
= pc
->mem_cgroup
;
3434 mem_cgroup_charge_statistics(memcg
, false, -1);
3435 ClearPageCgroupUsed(pc
);
3437 unlock_page_cgroup(pc
);
3440 * When called from shmem_replace_page(), in some cases the
3441 * oldpage has already been charged, and in some cases not.
3446 * Even if newpage->mapping was NULL before starting replacement,
3447 * the newpage may be on LRU(or pagevec for LRU) already. We lock
3448 * LRU while we overwrite pc->mem_cgroup.
3450 __mem_cgroup_commit_charge(memcg
, newpage
, 1, type
, true);
3453 #ifdef CONFIG_DEBUG_VM
3454 static struct page_cgroup
*lookup_page_cgroup_used(struct page
*page
)
3456 struct page_cgroup
*pc
;
3458 pc
= lookup_page_cgroup(page
);
3460 * Can be NULL while feeding pages into the page allocator for
3461 * the first time, i.e. during boot or memory hotplug;
3462 * or when mem_cgroup_disabled().
3464 if (likely(pc
) && PageCgroupUsed(pc
))
3469 bool mem_cgroup_bad_page_check(struct page
*page
)
3471 if (mem_cgroup_disabled())
3474 return lookup_page_cgroup_used(page
) != NULL
;
3477 void mem_cgroup_print_bad_page(struct page
*page
)
3479 struct page_cgroup
*pc
;
3481 pc
= lookup_page_cgroup_used(page
);
3483 printk(KERN_ALERT
"pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
3484 pc
, pc
->flags
, pc
->mem_cgroup
);
3489 static DEFINE_MUTEX(set_limit_mutex
);
3491 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
3492 unsigned long long val
)
3495 u64 memswlimit
, memlimit
;
3497 int children
= mem_cgroup_count_children(memcg
);
3498 u64 curusage
, oldusage
;
3502 * For keeping hierarchical_reclaim simple, how long we should retry
3503 * is depends on callers. We set our retry-count to be function
3504 * of # of children which we should visit in this loop.
3506 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
* children
;
3508 oldusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3511 while (retry_count
) {
3512 if (signal_pending(current
)) {
3517 * Rather than hide all in some function, I do this in
3518 * open coded manner. You see what this really does.
3519 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
3521 mutex_lock(&set_limit_mutex
);
3522 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3523 if (memswlimit
< val
) {
3525 mutex_unlock(&set_limit_mutex
);
3529 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3533 ret
= res_counter_set_limit(&memcg
->res
, val
);
3535 if (memswlimit
== val
)
3536 memcg
->memsw_is_minimum
= true;
3538 memcg
->memsw_is_minimum
= false;
3540 mutex_unlock(&set_limit_mutex
);
3545 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3546 MEM_CGROUP_RECLAIM_SHRINK
);
3547 curusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3548 /* Usage is reduced ? */
3549 if (curusage
>= oldusage
)
3552 oldusage
= curusage
;
3554 if (!ret
&& enlarge
)
3555 memcg_oom_recover(memcg
);
3560 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
3561 unsigned long long val
)
3564 u64 memlimit
, memswlimit
, oldusage
, curusage
;
3565 int children
= mem_cgroup_count_children(memcg
);
3569 /* see mem_cgroup_resize_res_limit */
3570 retry_count
= children
* MEM_CGROUP_RECLAIM_RETRIES
;
3571 oldusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3572 while (retry_count
) {
3573 if (signal_pending(current
)) {
3578 * Rather than hide all in some function, I do this in
3579 * open coded manner. You see what this really does.
3580 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
3582 mutex_lock(&set_limit_mutex
);
3583 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3584 if (memlimit
> val
) {
3586 mutex_unlock(&set_limit_mutex
);
3589 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3590 if (memswlimit
< val
)
3592 ret
= res_counter_set_limit(&memcg
->memsw
, val
);
3594 if (memlimit
== val
)
3595 memcg
->memsw_is_minimum
= true;
3597 memcg
->memsw_is_minimum
= false;
3599 mutex_unlock(&set_limit_mutex
);
3604 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3605 MEM_CGROUP_RECLAIM_NOSWAP
|
3606 MEM_CGROUP_RECLAIM_SHRINK
);
3607 curusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3608 /* Usage is reduced ? */
3609 if (curusage
>= oldusage
)
3612 oldusage
= curusage
;
3614 if (!ret
&& enlarge
)
3615 memcg_oom_recover(memcg
);
3619 unsigned long mem_cgroup_soft_limit_reclaim(struct zone
*zone
, int order
,
3621 unsigned long *total_scanned
)
3623 unsigned long nr_reclaimed
= 0;
3624 struct mem_cgroup_per_zone
*mz
, *next_mz
= NULL
;
3625 unsigned long reclaimed
;
3627 struct mem_cgroup_tree_per_zone
*mctz
;
3628 unsigned long long excess
;
3629 unsigned long nr_scanned
;
3634 mctz
= soft_limit_tree_node_zone(zone_to_nid(zone
), zone_idx(zone
));
3636 * This loop can run a while, specially if mem_cgroup's continuously
3637 * keep exceeding their soft limit and putting the system under
3644 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
3649 reclaimed
= mem_cgroup_soft_reclaim(mz
->memcg
, zone
,
3650 gfp_mask
, &nr_scanned
);
3651 nr_reclaimed
+= reclaimed
;
3652 *total_scanned
+= nr_scanned
;
3653 spin_lock(&mctz
->lock
);
3656 * If we failed to reclaim anything from this memory cgroup
3657 * it is time to move on to the next cgroup
3663 * Loop until we find yet another one.
3665 * By the time we get the soft_limit lock
3666 * again, someone might have aded the
3667 * group back on the RB tree. Iterate to
3668 * make sure we get a different mem.
3669 * mem_cgroup_largest_soft_limit_node returns
3670 * NULL if no other cgroup is present on
3674 __mem_cgroup_largest_soft_limit_node(mctz
);
3676 css_put(&next_mz
->memcg
->css
);
3677 else /* next_mz == NULL or other memcg */
3681 __mem_cgroup_remove_exceeded(mz
->memcg
, mz
, mctz
);
3682 excess
= res_counter_soft_limit_excess(&mz
->memcg
->res
);
3684 * One school of thought says that we should not add
3685 * back the node to the tree if reclaim returns 0.
3686 * But our reclaim could return 0, simply because due
3687 * to priority we are exposing a smaller subset of
3688 * memory to reclaim from. Consider this as a longer
3691 /* If excess == 0, no tree ops */
3692 __mem_cgroup_insert_exceeded(mz
->memcg
, mz
, mctz
, excess
);
3693 spin_unlock(&mctz
->lock
);
3694 css_put(&mz
->memcg
->css
);
3697 * Could not reclaim anything and there are no more
3698 * mem cgroups to try or we seem to be looping without
3699 * reclaiming anything.
3701 if (!nr_reclaimed
&&
3703 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
3705 } while (!nr_reclaimed
);
3707 css_put(&next_mz
->memcg
->css
);
3708 return nr_reclaimed
;
3712 * mem_cgroup_force_empty_list - clears LRU of a group
3713 * @memcg: group to clear
3716 * @lru: lru to to clear
3718 * Traverse a specified page_cgroup list and try to drop them all. This doesn't
3719 * reclaim the pages page themselves - pages are moved to the parent (or root)
3722 static void mem_cgroup_force_empty_list(struct mem_cgroup
*memcg
,
3723 int node
, int zid
, enum lru_list lru
)
3725 struct lruvec
*lruvec
;
3726 unsigned long flags
;
3727 struct list_head
*list
;
3731 zone
= &NODE_DATA(node
)->node_zones
[zid
];
3732 lruvec
= mem_cgroup_zone_lruvec(zone
, memcg
);
3733 list
= &lruvec
->lists
[lru
];
3737 struct page_cgroup
*pc
;
3740 spin_lock_irqsave(&zone
->lru_lock
, flags
);
3741 if (list_empty(list
)) {
3742 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3745 page
= list_entry(list
->prev
, struct page
, lru
);
3747 list_move(&page
->lru
, list
);
3749 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3752 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3754 pc
= lookup_page_cgroup(page
);
3756 if (mem_cgroup_move_parent(page
, pc
, memcg
)) {
3757 /* found lock contention or "pc" is obsolete. */
3762 } while (!list_empty(list
));
3766 * make mem_cgroup's charge to be 0 if there is no task by moving
3767 * all the charges and pages to the parent.
3768 * This enables deleting this mem_cgroup.
3770 * Caller is responsible for holding css reference on the memcg.
3772 static void mem_cgroup_reparent_charges(struct mem_cgroup
*memcg
)
3777 /* This is for making all *used* pages to be on LRU. */
3778 lru_add_drain_all();
3779 drain_all_stock_sync(memcg
);
3780 mem_cgroup_start_move(memcg
);
3781 for_each_node_state(node
, N_MEMORY
) {
3782 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
3785 mem_cgroup_force_empty_list(memcg
,
3790 mem_cgroup_end_move(memcg
);
3791 memcg_oom_recover(memcg
);
3795 * This is a safety check because mem_cgroup_force_empty_list
3796 * could have raced with mem_cgroup_replace_page_cache callers
3797 * so the lru seemed empty but the page could have been added
3798 * right after the check. RES_USAGE should be safe as we always
3799 * charge before adding to the LRU.
3801 } while (res_counter_read_u64(&memcg
->res
, RES_USAGE
) > 0);
3805 * Reclaims as many pages from the given memcg as possible and moves
3806 * the rest to the parent.
3808 * Caller is responsible for holding css reference for memcg.
3810 static int mem_cgroup_force_empty(struct mem_cgroup
*memcg
)
3812 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
3813 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
3815 /* returns EBUSY if there is a task or if we come here twice. */
3816 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
))
3819 /* we call try-to-free pages for make this cgroup empty */
3820 lru_add_drain_all();
3821 /* try to free all pages in this cgroup */
3822 while (nr_retries
&& res_counter_read_u64(&memcg
->res
, RES_USAGE
) > 0) {
3825 if (signal_pending(current
))
3828 progress
= try_to_free_mem_cgroup_pages(memcg
, GFP_KERNEL
,
3832 /* maybe some writeback is necessary */
3833 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3838 mem_cgroup_reparent_charges(memcg
);
3843 static int mem_cgroup_force_empty_write(struct cgroup
*cont
, unsigned int event
)
3845 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3848 if (mem_cgroup_is_root(memcg
))
3850 css_get(&memcg
->css
);
3851 ret
= mem_cgroup_force_empty(memcg
);
3852 css_put(&memcg
->css
);
3858 static u64
mem_cgroup_hierarchy_read(struct cgroup
*cont
, struct cftype
*cft
)
3860 return mem_cgroup_from_cont(cont
)->use_hierarchy
;
3863 static int mem_cgroup_hierarchy_write(struct cgroup
*cont
, struct cftype
*cft
,
3867 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3868 struct cgroup
*parent
= cont
->parent
;
3869 struct mem_cgroup
*parent_memcg
= NULL
;
3872 parent_memcg
= mem_cgroup_from_cont(parent
);
3876 if (memcg
->use_hierarchy
== val
)
3880 * If parent's use_hierarchy is set, we can't make any modifications
3881 * in the child subtrees. If it is unset, then the change can
3882 * occur, provided the current cgroup has no children.
3884 * For the root cgroup, parent_mem is NULL, we allow value to be
3885 * set if there are no children.
3887 if ((!parent_memcg
|| !parent_memcg
->use_hierarchy
) &&
3888 (val
== 1 || val
== 0)) {
3889 if (list_empty(&cont
->children
))
3890 memcg
->use_hierarchy
= val
;
3903 static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup
*memcg
,
3904 enum mem_cgroup_stat_index idx
)
3906 struct mem_cgroup
*iter
;
3909 /* Per-cpu values can be negative, use a signed accumulator */
3910 for_each_mem_cgroup_tree(iter
, memcg
)
3911 val
+= mem_cgroup_read_stat(iter
, idx
);
3913 if (val
< 0) /* race ? */
3918 static inline u64
mem_cgroup_usage(struct mem_cgroup
*memcg
, bool swap
)
3922 if (!mem_cgroup_is_root(memcg
)) {
3924 return res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3926 return res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3929 val
= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_CACHE
);
3930 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_RSS
);
3933 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_SWAP
);
3935 return val
<< PAGE_SHIFT
;
3938 static ssize_t
mem_cgroup_read(struct cgroup
*cont
, struct cftype
*cft
,
3939 struct file
*file
, char __user
*buf
,
3940 size_t nbytes
, loff_t
*ppos
)
3942 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3945 int type
, name
, len
;
3947 type
= MEMFILE_TYPE(cft
->private);
3948 name
= MEMFILE_ATTR(cft
->private);
3950 if (!do_swap_account
&& type
== _MEMSWAP
)
3955 if (name
== RES_USAGE
)
3956 val
= mem_cgroup_usage(memcg
, false);
3958 val
= res_counter_read_u64(&memcg
->res
, name
);
3961 if (name
== RES_USAGE
)
3962 val
= mem_cgroup_usage(memcg
, true);
3964 val
= res_counter_read_u64(&memcg
->memsw
, name
);
3970 len
= scnprintf(str
, sizeof(str
), "%llu\n", (unsigned long long)val
);
3971 return simple_read_from_buffer(buf
, nbytes
, ppos
, str
, len
);
3974 * The user of this function is...
3977 static int mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
3980 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3982 unsigned long long val
;
3985 type
= MEMFILE_TYPE(cft
->private);
3986 name
= MEMFILE_ATTR(cft
->private);
3988 if (!do_swap_account
&& type
== _MEMSWAP
)
3993 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
3997 /* This function does all necessary parse...reuse it */
3998 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
4002 ret
= mem_cgroup_resize_limit(memcg
, val
);
4004 ret
= mem_cgroup_resize_memsw_limit(memcg
, val
);
4006 case RES_SOFT_LIMIT
:
4007 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
4011 * For memsw, soft limits are hard to implement in terms
4012 * of semantics, for now, we support soft limits for
4013 * control without swap
4016 ret
= res_counter_set_soft_limit(&memcg
->res
, val
);
4021 ret
= -EINVAL
; /* should be BUG() ? */
4027 static void memcg_get_hierarchical_limit(struct mem_cgroup
*memcg
,
4028 unsigned long long *mem_limit
, unsigned long long *memsw_limit
)
4030 struct cgroup
*cgroup
;
4031 unsigned long long min_limit
, min_memsw_limit
, tmp
;
4033 min_limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
4034 min_memsw_limit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
4035 cgroup
= memcg
->css
.cgroup
;
4036 if (!memcg
->use_hierarchy
)
4039 while (cgroup
->parent
) {
4040 cgroup
= cgroup
->parent
;
4041 memcg
= mem_cgroup_from_cont(cgroup
);
4042 if (!memcg
->use_hierarchy
)
4044 tmp
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
4045 min_limit
= min(min_limit
, tmp
);
4046 tmp
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
4047 min_memsw_limit
= min(min_memsw_limit
, tmp
);
4050 *mem_limit
= min_limit
;
4051 *memsw_limit
= min_memsw_limit
;
4054 static int mem_cgroup_reset(struct cgroup
*cont
, unsigned int event
)
4056 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4059 type
= MEMFILE_TYPE(event
);
4060 name
= MEMFILE_ATTR(event
);
4062 if (!do_swap_account
&& type
== _MEMSWAP
)
4068 res_counter_reset_max(&memcg
->res
);
4070 res_counter_reset_max(&memcg
->memsw
);
4074 res_counter_reset_failcnt(&memcg
->res
);
4076 res_counter_reset_failcnt(&memcg
->memsw
);
4083 static u64
mem_cgroup_move_charge_read(struct cgroup
*cgrp
,
4086 return mem_cgroup_from_cont(cgrp
)->move_charge_at_immigrate
;
4090 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4091 struct cftype
*cft
, u64 val
)
4093 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4095 if (val
>= (1 << NR_MOVE_TYPE
))
4098 * We check this value several times in both in can_attach() and
4099 * attach(), so we need cgroup lock to prevent this value from being
4103 memcg
->move_charge_at_immigrate
= val
;
4109 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4110 struct cftype
*cft
, u64 val
)
4117 static int memcg_numa_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
4121 unsigned long total_nr
, file_nr
, anon_nr
, unevictable_nr
;
4122 unsigned long node_nr
;
4123 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4125 total_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL
);
4126 seq_printf(m
, "total=%lu", total_nr
);
4127 for_each_node_state(nid
, N_MEMORY
) {
4128 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL
);
4129 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4133 file_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL_FILE
);
4134 seq_printf(m
, "file=%lu", file_nr
);
4135 for_each_node_state(nid
, N_MEMORY
) {
4136 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4138 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4142 anon_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL_ANON
);
4143 seq_printf(m
, "anon=%lu", anon_nr
);
4144 for_each_node_state(nid
, N_MEMORY
) {
4145 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4147 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4151 unevictable_nr
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_UNEVICTABLE
));
4152 seq_printf(m
, "unevictable=%lu", unevictable_nr
);
4153 for_each_node_state(nid
, N_MEMORY
) {
4154 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4155 BIT(LRU_UNEVICTABLE
));
4156 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4161 #endif /* CONFIG_NUMA */
4163 static const char * const mem_cgroup_lru_names
[] = {
4171 static inline void mem_cgroup_lru_names_not_uptodate(void)
4173 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names
) != NR_LRU_LISTS
);
4176 static int memcg_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
4179 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4180 struct mem_cgroup
*mi
;
4183 for (i
= 0; i
< MEM_CGROUP_STAT_NSTATS
; i
++) {
4184 if (i
== MEM_CGROUP_STAT_SWAP
&& !do_swap_account
)
4186 seq_printf(m
, "%s %ld\n", mem_cgroup_stat_names
[i
],
4187 mem_cgroup_read_stat(memcg
, i
) * PAGE_SIZE
);
4190 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++)
4191 seq_printf(m
, "%s %lu\n", mem_cgroup_events_names
[i
],
4192 mem_cgroup_read_events(memcg
, i
));
4194 for (i
= 0; i
< NR_LRU_LISTS
; i
++)
4195 seq_printf(m
, "%s %lu\n", mem_cgroup_lru_names
[i
],
4196 mem_cgroup_nr_lru_pages(memcg
, BIT(i
)) * PAGE_SIZE
);
4198 /* Hierarchical information */
4200 unsigned long long limit
, memsw_limit
;
4201 memcg_get_hierarchical_limit(memcg
, &limit
, &memsw_limit
);
4202 seq_printf(m
, "hierarchical_memory_limit %llu\n", limit
);
4203 if (do_swap_account
)
4204 seq_printf(m
, "hierarchical_memsw_limit %llu\n",
4208 for (i
= 0; i
< MEM_CGROUP_STAT_NSTATS
; i
++) {
4211 if (i
== MEM_CGROUP_STAT_SWAP
&& !do_swap_account
)
4213 for_each_mem_cgroup_tree(mi
, memcg
)
4214 val
+= mem_cgroup_read_stat(mi
, i
) * PAGE_SIZE
;
4215 seq_printf(m
, "total_%s %lld\n", mem_cgroup_stat_names
[i
], val
);
4218 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
4219 unsigned long long val
= 0;
4221 for_each_mem_cgroup_tree(mi
, memcg
)
4222 val
+= mem_cgroup_read_events(mi
, i
);
4223 seq_printf(m
, "total_%s %llu\n",
4224 mem_cgroup_events_names
[i
], val
);
4227 for (i
= 0; i
< NR_LRU_LISTS
; i
++) {
4228 unsigned long long val
= 0;
4230 for_each_mem_cgroup_tree(mi
, memcg
)
4231 val
+= mem_cgroup_nr_lru_pages(mi
, BIT(i
)) * PAGE_SIZE
;
4232 seq_printf(m
, "total_%s %llu\n", mem_cgroup_lru_names
[i
], val
);
4235 #ifdef CONFIG_DEBUG_VM
4238 struct mem_cgroup_per_zone
*mz
;
4239 struct zone_reclaim_stat
*rstat
;
4240 unsigned long recent_rotated
[2] = {0, 0};
4241 unsigned long recent_scanned
[2] = {0, 0};
4243 for_each_online_node(nid
)
4244 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
4245 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
4246 rstat
= &mz
->lruvec
.reclaim_stat
;
4248 recent_rotated
[0] += rstat
->recent_rotated
[0];
4249 recent_rotated
[1] += rstat
->recent_rotated
[1];
4250 recent_scanned
[0] += rstat
->recent_scanned
[0];
4251 recent_scanned
[1] += rstat
->recent_scanned
[1];
4253 seq_printf(m
, "recent_rotated_anon %lu\n", recent_rotated
[0]);
4254 seq_printf(m
, "recent_rotated_file %lu\n", recent_rotated
[1]);
4255 seq_printf(m
, "recent_scanned_anon %lu\n", recent_scanned
[0]);
4256 seq_printf(m
, "recent_scanned_file %lu\n", recent_scanned
[1]);
4263 static u64
mem_cgroup_swappiness_read(struct cgroup
*cgrp
, struct cftype
*cft
)
4265 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4267 return mem_cgroup_swappiness(memcg
);
4270 static int mem_cgroup_swappiness_write(struct cgroup
*cgrp
, struct cftype
*cft
,
4273 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4274 struct mem_cgroup
*parent
;
4279 if (cgrp
->parent
== NULL
)
4282 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4286 /* If under hierarchy, only empty-root can set this value */
4287 if ((parent
->use_hierarchy
) ||
4288 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4293 memcg
->swappiness
= val
;
4300 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
4302 struct mem_cgroup_threshold_ary
*t
;
4308 t
= rcu_dereference(memcg
->thresholds
.primary
);
4310 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
4315 usage
= mem_cgroup_usage(memcg
, swap
);
4318 * current_threshold points to threshold just below or equal to usage.
4319 * If it's not true, a threshold was crossed after last
4320 * call of __mem_cgroup_threshold().
4322 i
= t
->current_threshold
;
4325 * Iterate backward over array of thresholds starting from
4326 * current_threshold and check if a threshold is crossed.
4327 * If none of thresholds below usage is crossed, we read
4328 * only one element of the array here.
4330 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
4331 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4333 /* i = current_threshold + 1 */
4337 * Iterate forward over array of thresholds starting from
4338 * current_threshold+1 and check if a threshold is crossed.
4339 * If none of thresholds above usage is crossed, we read
4340 * only one element of the array here.
4342 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
4343 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4345 /* Update current_threshold */
4346 t
->current_threshold
= i
- 1;
4351 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
4354 __mem_cgroup_threshold(memcg
, false);
4355 if (do_swap_account
)
4356 __mem_cgroup_threshold(memcg
, true);
4358 memcg
= parent_mem_cgroup(memcg
);
4362 static int compare_thresholds(const void *a
, const void *b
)
4364 const struct mem_cgroup_threshold
*_a
= a
;
4365 const struct mem_cgroup_threshold
*_b
= b
;
4367 return _a
->threshold
- _b
->threshold
;
4370 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*memcg
)
4372 struct mem_cgroup_eventfd_list
*ev
;
4374 list_for_each_entry(ev
, &memcg
->oom_notify
, list
)
4375 eventfd_signal(ev
->eventfd
, 1);
4379 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
)
4381 struct mem_cgroup
*iter
;
4383 for_each_mem_cgroup_tree(iter
, memcg
)
4384 mem_cgroup_oom_notify_cb(iter
);
4387 static int mem_cgroup_usage_register_event(struct cgroup
*cgrp
,
4388 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4390 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4391 struct mem_cgroup_thresholds
*thresholds
;
4392 struct mem_cgroup_threshold_ary
*new;
4393 int type
= MEMFILE_TYPE(cft
->private);
4394 u64 threshold
, usage
;
4397 ret
= res_counter_memparse_write_strategy(args
, &threshold
);
4401 mutex_lock(&memcg
->thresholds_lock
);
4404 thresholds
= &memcg
->thresholds
;
4405 else if (type
== _MEMSWAP
)
4406 thresholds
= &memcg
->memsw_thresholds
;
4410 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4412 /* Check if a threshold crossed before adding a new one */
4413 if (thresholds
->primary
)
4414 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4416 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
4418 /* Allocate memory for new array of thresholds */
4419 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
4427 /* Copy thresholds (if any) to new array */
4428 if (thresholds
->primary
) {
4429 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
4430 sizeof(struct mem_cgroup_threshold
));
4433 /* Add new threshold */
4434 new->entries
[size
- 1].eventfd
= eventfd
;
4435 new->entries
[size
- 1].threshold
= threshold
;
4437 /* Sort thresholds. Registering of new threshold isn't time-critical */
4438 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
4439 compare_thresholds
, NULL
);
4441 /* Find current threshold */
4442 new->current_threshold
= -1;
4443 for (i
= 0; i
< size
; i
++) {
4444 if (new->entries
[i
].threshold
<= usage
) {
4446 * new->current_threshold will not be used until
4447 * rcu_assign_pointer(), so it's safe to increment
4450 ++new->current_threshold
;
4455 /* Free old spare buffer and save old primary buffer as spare */
4456 kfree(thresholds
->spare
);
4457 thresholds
->spare
= thresholds
->primary
;
4459 rcu_assign_pointer(thresholds
->primary
, new);
4461 /* To be sure that nobody uses thresholds */
4465 mutex_unlock(&memcg
->thresholds_lock
);
4470 static void mem_cgroup_usage_unregister_event(struct cgroup
*cgrp
,
4471 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4473 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4474 struct mem_cgroup_thresholds
*thresholds
;
4475 struct mem_cgroup_threshold_ary
*new;
4476 int type
= MEMFILE_TYPE(cft
->private);
4480 mutex_lock(&memcg
->thresholds_lock
);
4482 thresholds
= &memcg
->thresholds
;
4483 else if (type
== _MEMSWAP
)
4484 thresholds
= &memcg
->memsw_thresholds
;
4488 if (!thresholds
->primary
)
4491 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4493 /* Check if a threshold crossed before removing */
4494 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4496 /* Calculate new number of threshold */
4498 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
4499 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
4503 new = thresholds
->spare
;
4505 /* Set thresholds array to NULL if we don't have thresholds */
4514 /* Copy thresholds and find current threshold */
4515 new->current_threshold
= -1;
4516 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
4517 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
4520 new->entries
[j
] = thresholds
->primary
->entries
[i
];
4521 if (new->entries
[j
].threshold
<= usage
) {
4523 * new->current_threshold will not be used
4524 * until rcu_assign_pointer(), so it's safe to increment
4527 ++new->current_threshold
;
4533 /* Swap primary and spare array */
4534 thresholds
->spare
= thresholds
->primary
;
4535 /* If all events are unregistered, free the spare array */
4537 kfree(thresholds
->spare
);
4538 thresholds
->spare
= NULL
;
4541 rcu_assign_pointer(thresholds
->primary
, new);
4543 /* To be sure that nobody uses thresholds */
4546 mutex_unlock(&memcg
->thresholds_lock
);
4549 static int mem_cgroup_oom_register_event(struct cgroup
*cgrp
,
4550 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4552 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4553 struct mem_cgroup_eventfd_list
*event
;
4554 int type
= MEMFILE_TYPE(cft
->private);
4556 BUG_ON(type
!= _OOM_TYPE
);
4557 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
4561 spin_lock(&memcg_oom_lock
);
4563 event
->eventfd
= eventfd
;
4564 list_add(&event
->list
, &memcg
->oom_notify
);
4566 /* already in OOM ? */
4567 if (atomic_read(&memcg
->under_oom
))
4568 eventfd_signal(eventfd
, 1);
4569 spin_unlock(&memcg_oom_lock
);
4574 static void mem_cgroup_oom_unregister_event(struct cgroup
*cgrp
,
4575 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4577 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4578 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
4579 int type
= MEMFILE_TYPE(cft
->private);
4581 BUG_ON(type
!= _OOM_TYPE
);
4583 spin_lock(&memcg_oom_lock
);
4585 list_for_each_entry_safe(ev
, tmp
, &memcg
->oom_notify
, list
) {
4586 if (ev
->eventfd
== eventfd
) {
4587 list_del(&ev
->list
);
4592 spin_unlock(&memcg_oom_lock
);
4595 static int mem_cgroup_oom_control_read(struct cgroup
*cgrp
,
4596 struct cftype
*cft
, struct cgroup_map_cb
*cb
)
4598 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4600 cb
->fill(cb
, "oom_kill_disable", memcg
->oom_kill_disable
);
4602 if (atomic_read(&memcg
->under_oom
))
4603 cb
->fill(cb
, "under_oom", 1);
4605 cb
->fill(cb
, "under_oom", 0);
4609 static int mem_cgroup_oom_control_write(struct cgroup
*cgrp
,
4610 struct cftype
*cft
, u64 val
)
4612 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4613 struct mem_cgroup
*parent
;
4615 /* cannot set to root cgroup and only 0 and 1 are allowed */
4616 if (!cgrp
->parent
|| !((val
== 0) || (val
== 1)))
4619 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4622 /* oom-kill-disable is a flag for subhierarchy. */
4623 if ((parent
->use_hierarchy
) ||
4624 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4628 memcg
->oom_kill_disable
= val
;
4630 memcg_oom_recover(memcg
);
4635 #ifdef CONFIG_MEMCG_KMEM
4636 static int memcg_init_kmem(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
4638 return mem_cgroup_sockets_init(memcg
, ss
);
4641 static void kmem_cgroup_destroy(struct mem_cgroup
*memcg
)
4643 mem_cgroup_sockets_destroy(memcg
);
4646 static int memcg_init_kmem(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
4651 static void kmem_cgroup_destroy(struct mem_cgroup
*memcg
)
4656 static struct cftype mem_cgroup_files
[] = {
4658 .name
= "usage_in_bytes",
4659 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
4660 .read
= mem_cgroup_read
,
4661 .register_event
= mem_cgroup_usage_register_event
,
4662 .unregister_event
= mem_cgroup_usage_unregister_event
,
4665 .name
= "max_usage_in_bytes",
4666 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
4667 .trigger
= mem_cgroup_reset
,
4668 .read
= mem_cgroup_read
,
4671 .name
= "limit_in_bytes",
4672 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
4673 .write_string
= mem_cgroup_write
,
4674 .read
= mem_cgroup_read
,
4677 .name
= "soft_limit_in_bytes",
4678 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
4679 .write_string
= mem_cgroup_write
,
4680 .read
= mem_cgroup_read
,
4684 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
4685 .trigger
= mem_cgroup_reset
,
4686 .read
= mem_cgroup_read
,
4690 .read_seq_string
= memcg_stat_show
,
4693 .name
= "force_empty",
4694 .trigger
= mem_cgroup_force_empty_write
,
4697 .name
= "use_hierarchy",
4698 .write_u64
= mem_cgroup_hierarchy_write
,
4699 .read_u64
= mem_cgroup_hierarchy_read
,
4702 .name
= "swappiness",
4703 .read_u64
= mem_cgroup_swappiness_read
,
4704 .write_u64
= mem_cgroup_swappiness_write
,
4707 .name
= "move_charge_at_immigrate",
4708 .read_u64
= mem_cgroup_move_charge_read
,
4709 .write_u64
= mem_cgroup_move_charge_write
,
4712 .name
= "oom_control",
4713 .read_map
= mem_cgroup_oom_control_read
,
4714 .write_u64
= mem_cgroup_oom_control_write
,
4715 .register_event
= mem_cgroup_oom_register_event
,
4716 .unregister_event
= mem_cgroup_oom_unregister_event
,
4717 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
4721 .name
= "numa_stat",
4722 .read_seq_string
= memcg_numa_stat_show
,
4725 #ifdef CONFIG_MEMCG_SWAP
4727 .name
= "memsw.usage_in_bytes",
4728 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
4729 .read
= mem_cgroup_read
,
4730 .register_event
= mem_cgroup_usage_register_event
,
4731 .unregister_event
= mem_cgroup_usage_unregister_event
,
4734 .name
= "memsw.max_usage_in_bytes",
4735 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
4736 .trigger
= mem_cgroup_reset
,
4737 .read
= mem_cgroup_read
,
4740 .name
= "memsw.limit_in_bytes",
4741 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
4742 .write_string
= mem_cgroup_write
,
4743 .read
= mem_cgroup_read
,
4746 .name
= "memsw.failcnt",
4747 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
4748 .trigger
= mem_cgroup_reset
,
4749 .read
= mem_cgroup_read
,
4752 { }, /* terminate */
4755 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4757 struct mem_cgroup_per_node
*pn
;
4758 struct mem_cgroup_per_zone
*mz
;
4759 int zone
, tmp
= node
;
4761 * This routine is called against possible nodes.
4762 * But it's BUG to call kmalloc() against offline node.
4764 * TODO: this routine can waste much memory for nodes which will
4765 * never be onlined. It's better to use memory hotplug callback
4768 if (!node_state(node
, N_NORMAL_MEMORY
))
4770 pn
= kzalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
4774 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4775 mz
= &pn
->zoneinfo
[zone
];
4776 lruvec_init(&mz
->lruvec
);
4777 mz
->usage_in_excess
= 0;
4778 mz
->on_tree
= false;
4781 memcg
->info
.nodeinfo
[node
] = pn
;
4785 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4787 kfree(memcg
->info
.nodeinfo
[node
]);
4790 static struct mem_cgroup
*mem_cgroup_alloc(void)
4792 struct mem_cgroup
*memcg
;
4793 int size
= sizeof(struct mem_cgroup
);
4795 /* Can be very big if MAX_NUMNODES is very big */
4796 if (size
< PAGE_SIZE
)
4797 memcg
= kzalloc(size
, GFP_KERNEL
);
4799 memcg
= vzalloc(size
);
4804 memcg
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4807 spin_lock_init(&memcg
->pcp_counter_lock
);
4811 if (size
< PAGE_SIZE
)
4819 * Helpers for freeing a kmalloc()ed/vzalloc()ed mem_cgroup by RCU,
4820 * but in process context. The work_freeing structure is overlaid
4821 * on the rcu_freeing structure, which itself is overlaid on memsw.
4823 static void free_work(struct work_struct
*work
)
4825 struct mem_cgroup
*memcg
;
4826 int size
= sizeof(struct mem_cgroup
);
4828 memcg
= container_of(work
, struct mem_cgroup
, work_freeing
);
4830 * We need to make sure that (at least for now), the jump label
4831 * destruction code runs outside of the cgroup lock. This is because
4832 * get_online_cpus(), which is called from the static_branch update,
4833 * can't be called inside the cgroup_lock. cpusets are the ones
4834 * enforcing this dependency, so if they ever change, we might as well.
4836 * schedule_work() will guarantee this happens. Be careful if you need
4837 * to move this code around, and make sure it is outside
4840 disarm_sock_keys(memcg
);
4841 if (size
< PAGE_SIZE
)
4847 static void free_rcu(struct rcu_head
*rcu_head
)
4849 struct mem_cgroup
*memcg
;
4851 memcg
= container_of(rcu_head
, struct mem_cgroup
, rcu_freeing
);
4852 INIT_WORK(&memcg
->work_freeing
, free_work
);
4853 schedule_work(&memcg
->work_freeing
);
4857 * At destroying mem_cgroup, references from swap_cgroup can remain.
4858 * (scanning all at force_empty is too costly...)
4860 * Instead of clearing all references at force_empty, we remember
4861 * the number of reference from swap_cgroup and free mem_cgroup when
4862 * it goes down to 0.
4864 * Removal of cgroup itself succeeds regardless of refs from swap.
4867 static void __mem_cgroup_free(struct mem_cgroup
*memcg
)
4871 mem_cgroup_remove_from_trees(memcg
);
4872 free_css_id(&mem_cgroup_subsys
, &memcg
->css
);
4875 free_mem_cgroup_per_zone_info(memcg
, node
);
4877 free_percpu(memcg
->stat
);
4878 call_rcu(&memcg
->rcu_freeing
, free_rcu
);
4881 static void mem_cgroup_get(struct mem_cgroup
*memcg
)
4883 atomic_inc(&memcg
->refcnt
);
4886 static void __mem_cgroup_put(struct mem_cgroup
*memcg
, int count
)
4888 if (atomic_sub_and_test(count
, &memcg
->refcnt
)) {
4889 struct mem_cgroup
*parent
= parent_mem_cgroup(memcg
);
4890 __mem_cgroup_free(memcg
);
4892 mem_cgroup_put(parent
);
4896 static void mem_cgroup_put(struct mem_cgroup
*memcg
)
4898 __mem_cgroup_put(memcg
, 1);
4902 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4904 struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*memcg
)
4906 if (!memcg
->res
.parent
)
4908 return mem_cgroup_from_res_counter(memcg
->res
.parent
, res
);
4910 EXPORT_SYMBOL(parent_mem_cgroup
);
4912 #ifdef CONFIG_MEMCG_SWAP
4913 static void __init
enable_swap_cgroup(void)
4915 if (!mem_cgroup_disabled() && really_do_swap_account
)
4916 do_swap_account
= 1;
4919 static void __init
enable_swap_cgroup(void)
4924 static int mem_cgroup_soft_limit_tree_init(void)
4926 struct mem_cgroup_tree_per_node
*rtpn
;
4927 struct mem_cgroup_tree_per_zone
*rtpz
;
4928 int tmp
, node
, zone
;
4930 for_each_node(node
) {
4932 if (!node_state(node
, N_NORMAL_MEMORY
))
4934 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
, tmp
);
4938 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
4940 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4941 rtpz
= &rtpn
->rb_tree_per_zone
[zone
];
4942 rtpz
->rb_root
= RB_ROOT
;
4943 spin_lock_init(&rtpz
->lock
);
4949 for_each_node(node
) {
4950 if (!soft_limit_tree
.rb_tree_per_node
[node
])
4952 kfree(soft_limit_tree
.rb_tree_per_node
[node
]);
4953 soft_limit_tree
.rb_tree_per_node
[node
] = NULL
;
4959 static struct cgroup_subsys_state
* __ref
4960 mem_cgroup_css_alloc(struct cgroup
*cont
)
4962 struct mem_cgroup
*memcg
, *parent
;
4963 long error
= -ENOMEM
;
4966 memcg
= mem_cgroup_alloc();
4968 return ERR_PTR(error
);
4971 if (alloc_mem_cgroup_per_zone_info(memcg
, node
))
4975 if (cont
->parent
== NULL
) {
4977 enable_swap_cgroup();
4979 if (mem_cgroup_soft_limit_tree_init())
4981 root_mem_cgroup
= memcg
;
4982 for_each_possible_cpu(cpu
) {
4983 struct memcg_stock_pcp
*stock
=
4984 &per_cpu(memcg_stock
, cpu
);
4985 INIT_WORK(&stock
->work
, drain_local_stock
);
4987 hotcpu_notifier(memcg_cpu_hotplug_callback
, 0);
4989 parent
= mem_cgroup_from_cont(cont
->parent
);
4990 memcg
->use_hierarchy
= parent
->use_hierarchy
;
4991 memcg
->oom_kill_disable
= parent
->oom_kill_disable
;
4994 if (parent
&& parent
->use_hierarchy
) {
4995 res_counter_init(&memcg
->res
, &parent
->res
);
4996 res_counter_init(&memcg
->memsw
, &parent
->memsw
);
4998 * We increment refcnt of the parent to ensure that we can
4999 * safely access it on res_counter_charge/uncharge.
5000 * This refcnt will be decremented when freeing this
5001 * mem_cgroup(see mem_cgroup_put).
5003 mem_cgroup_get(parent
);
5005 res_counter_init(&memcg
->res
, NULL
);
5006 res_counter_init(&memcg
->memsw
, NULL
);
5008 * Deeper hierachy with use_hierarchy == false doesn't make
5009 * much sense so let cgroup subsystem know about this
5010 * unfortunate state in our controller.
5012 if (parent
&& parent
!= root_mem_cgroup
)
5013 mem_cgroup_subsys
.broken_hierarchy
= true;
5015 memcg
->last_scanned_node
= MAX_NUMNODES
;
5016 INIT_LIST_HEAD(&memcg
->oom_notify
);
5019 memcg
->swappiness
= mem_cgroup_swappiness(parent
);
5020 atomic_set(&memcg
->refcnt
, 1);
5021 memcg
->move_charge_at_immigrate
= 0;
5022 mutex_init(&memcg
->thresholds_lock
);
5023 spin_lock_init(&memcg
->move_lock
);
5025 error
= memcg_init_kmem(memcg
, &mem_cgroup_subsys
);
5028 * We call put now because our (and parent's) refcnts
5029 * are already in place. mem_cgroup_put() will internally
5030 * call __mem_cgroup_free, so return directly
5032 mem_cgroup_put(memcg
);
5033 return ERR_PTR(error
);
5037 __mem_cgroup_free(memcg
);
5038 return ERR_PTR(error
);
5041 static void mem_cgroup_css_offline(struct cgroup
*cont
)
5043 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
5045 mem_cgroup_reparent_charges(memcg
);
5048 static void mem_cgroup_css_free(struct cgroup
*cont
)
5050 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
5052 kmem_cgroup_destroy(memcg
);
5054 mem_cgroup_put(memcg
);
5058 /* Handlers for move charge at task migration. */
5059 #define PRECHARGE_COUNT_AT_ONCE 256
5060 static int mem_cgroup_do_precharge(unsigned long count
)
5063 int batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5064 struct mem_cgroup
*memcg
= mc
.to
;
5066 if (mem_cgroup_is_root(memcg
)) {
5067 mc
.precharge
+= count
;
5068 /* we don't need css_get for root */
5071 /* try to charge at once */
5073 struct res_counter
*dummy
;
5075 * "memcg" cannot be under rmdir() because we've already checked
5076 * by cgroup_lock_live_cgroup() that it is not removed and we
5077 * are still under the same cgroup_mutex. So we can postpone
5080 if (res_counter_charge(&memcg
->res
, PAGE_SIZE
* count
, &dummy
))
5082 if (do_swap_account
&& res_counter_charge(&memcg
->memsw
,
5083 PAGE_SIZE
* count
, &dummy
)) {
5084 res_counter_uncharge(&memcg
->res
, PAGE_SIZE
* count
);
5087 mc
.precharge
+= count
;
5091 /* fall back to one by one charge */
5093 if (signal_pending(current
)) {
5097 if (!batch_count
--) {
5098 batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5101 ret
= __mem_cgroup_try_charge(NULL
,
5102 GFP_KERNEL
, 1, &memcg
, false);
5104 /* mem_cgroup_clear_mc() will do uncharge later */
5112 * get_mctgt_type - get target type of moving charge
5113 * @vma: the vma the pte to be checked belongs
5114 * @addr: the address corresponding to the pte to be checked
5115 * @ptent: the pte to be checked
5116 * @target: the pointer the target page or swap ent will be stored(can be NULL)
5119 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5120 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5121 * move charge. if @target is not NULL, the page is stored in target->page
5122 * with extra refcnt got(Callers should handle it).
5123 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5124 * target for charge migration. if @target is not NULL, the entry is stored
5127 * Called with pte lock held.
5134 enum mc_target_type
{
5140 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
5141 unsigned long addr
, pte_t ptent
)
5143 struct page
*page
= vm_normal_page(vma
, addr
, ptent
);
5145 if (!page
|| !page_mapped(page
))
5147 if (PageAnon(page
)) {
5148 /* we don't move shared anon */
5151 } else if (!move_file())
5152 /* we ignore mapcount for file pages */
5154 if (!get_page_unless_zero(page
))
5161 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5162 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5164 struct page
*page
= NULL
;
5165 swp_entry_t ent
= pte_to_swp_entry(ptent
);
5167 if (!move_anon() || non_swap_entry(ent
))
5170 * Because lookup_swap_cache() updates some statistics counter,
5171 * we call find_get_page() with swapper_space directly.
5173 page
= find_get_page(&swapper_space
, ent
.val
);
5174 if (do_swap_account
)
5175 entry
->val
= ent
.val
;
5180 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5181 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5187 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
5188 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5190 struct page
*page
= NULL
;
5191 struct address_space
*mapping
;
5194 if (!vma
->vm_file
) /* anonymous vma */
5199 mapping
= vma
->vm_file
->f_mapping
;
5200 if (pte_none(ptent
))
5201 pgoff
= linear_page_index(vma
, addr
);
5202 else /* pte_file(ptent) is true */
5203 pgoff
= pte_to_pgoff(ptent
);
5205 /* page is moved even if it's not RSS of this task(page-faulted). */
5206 page
= find_get_page(mapping
, pgoff
);
5209 /* shmem/tmpfs may report page out on swap: account for that too. */
5210 if (radix_tree_exceptional_entry(page
)) {
5211 swp_entry_t swap
= radix_to_swp_entry(page
);
5212 if (do_swap_account
)
5214 page
= find_get_page(&swapper_space
, swap
.val
);
5220 static enum mc_target_type
get_mctgt_type(struct vm_area_struct
*vma
,
5221 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
5223 struct page
*page
= NULL
;
5224 struct page_cgroup
*pc
;
5225 enum mc_target_type ret
= MC_TARGET_NONE
;
5226 swp_entry_t ent
= { .val
= 0 };
5228 if (pte_present(ptent
))
5229 page
= mc_handle_present_pte(vma
, addr
, ptent
);
5230 else if (is_swap_pte(ptent
))
5231 page
= mc_handle_swap_pte(vma
, addr
, ptent
, &ent
);
5232 else if (pte_none(ptent
) || pte_file(ptent
))
5233 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
5235 if (!page
&& !ent
.val
)
5238 pc
= lookup_page_cgroup(page
);
5240 * Do only loose check w/o page_cgroup lock.
5241 * mem_cgroup_move_account() checks the pc is valid or not under
5244 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5245 ret
= MC_TARGET_PAGE
;
5247 target
->page
= page
;
5249 if (!ret
|| !target
)
5252 /* There is a swap entry and a page doesn't exist or isn't charged */
5253 if (ent
.val
&& !ret
&&
5254 css_id(&mc
.from
->css
) == lookup_swap_cgroup_id(ent
)) {
5255 ret
= MC_TARGET_SWAP
;
5262 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5264 * We don't consider swapping or file mapped pages because THP does not
5265 * support them for now.
5266 * Caller should make sure that pmd_trans_huge(pmd) is true.
5268 static enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
5269 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
5271 struct page
*page
= NULL
;
5272 struct page_cgroup
*pc
;
5273 enum mc_target_type ret
= MC_TARGET_NONE
;
5275 page
= pmd_page(pmd
);
5276 VM_BUG_ON(!page
|| !PageHead(page
));
5279 pc
= lookup_page_cgroup(page
);
5280 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5281 ret
= MC_TARGET_PAGE
;
5284 target
->page
= page
;
5290 static inline enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
5291 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
5293 return MC_TARGET_NONE
;
5297 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
5298 unsigned long addr
, unsigned long end
,
5299 struct mm_walk
*walk
)
5301 struct vm_area_struct
*vma
= walk
->private;
5305 if (pmd_trans_huge_lock(pmd
, vma
) == 1) {
5306 if (get_mctgt_type_thp(vma
, addr
, *pmd
, NULL
) == MC_TARGET_PAGE
)
5307 mc
.precharge
+= HPAGE_PMD_NR
;
5308 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5312 if (pmd_trans_unstable(pmd
))
5314 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5315 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
5316 if (get_mctgt_type(vma
, addr
, *pte
, NULL
))
5317 mc
.precharge
++; /* increment precharge temporarily */
5318 pte_unmap_unlock(pte
- 1, ptl
);
5324 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
5326 unsigned long precharge
;
5327 struct vm_area_struct
*vma
;
5329 down_read(&mm
->mmap_sem
);
5330 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5331 struct mm_walk mem_cgroup_count_precharge_walk
= {
5332 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
5336 if (is_vm_hugetlb_page(vma
))
5338 walk_page_range(vma
->vm_start
, vma
->vm_end
,
5339 &mem_cgroup_count_precharge_walk
);
5341 up_read(&mm
->mmap_sem
);
5343 precharge
= mc
.precharge
;
5349 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
5351 unsigned long precharge
= mem_cgroup_count_precharge(mm
);
5353 VM_BUG_ON(mc
.moving_task
);
5354 mc
.moving_task
= current
;
5355 return mem_cgroup_do_precharge(precharge
);
5358 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5359 static void __mem_cgroup_clear_mc(void)
5361 struct mem_cgroup
*from
= mc
.from
;
5362 struct mem_cgroup
*to
= mc
.to
;
5364 /* we must uncharge all the leftover precharges from mc.to */
5366 __mem_cgroup_cancel_charge(mc
.to
, mc
.precharge
);
5370 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5371 * we must uncharge here.
5373 if (mc
.moved_charge
) {
5374 __mem_cgroup_cancel_charge(mc
.from
, mc
.moved_charge
);
5375 mc
.moved_charge
= 0;
5377 /* we must fixup refcnts and charges */
5378 if (mc
.moved_swap
) {
5379 /* uncharge swap account from the old cgroup */
5380 if (!mem_cgroup_is_root(mc
.from
))
5381 res_counter_uncharge(&mc
.from
->memsw
,
5382 PAGE_SIZE
* mc
.moved_swap
);
5383 __mem_cgroup_put(mc
.from
, mc
.moved_swap
);
5385 if (!mem_cgroup_is_root(mc
.to
)) {
5387 * we charged both to->res and to->memsw, so we should
5390 res_counter_uncharge(&mc
.to
->res
,
5391 PAGE_SIZE
* mc
.moved_swap
);
5393 /* we've already done mem_cgroup_get(mc.to) */
5396 memcg_oom_recover(from
);
5397 memcg_oom_recover(to
);
5398 wake_up_all(&mc
.waitq
);
5401 static void mem_cgroup_clear_mc(void)
5403 struct mem_cgroup
*from
= mc
.from
;
5406 * we must clear moving_task before waking up waiters at the end of
5409 mc
.moving_task
= NULL
;
5410 __mem_cgroup_clear_mc();
5411 spin_lock(&mc
.lock
);
5414 spin_unlock(&mc
.lock
);
5415 mem_cgroup_end_move(from
);
5418 static int mem_cgroup_can_attach(struct cgroup
*cgroup
,
5419 struct cgroup_taskset
*tset
)
5421 struct task_struct
*p
= cgroup_taskset_first(tset
);
5423 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgroup
);
5425 if (memcg
->move_charge_at_immigrate
) {
5426 struct mm_struct
*mm
;
5427 struct mem_cgroup
*from
= mem_cgroup_from_task(p
);
5429 VM_BUG_ON(from
== memcg
);
5431 mm
= get_task_mm(p
);
5434 /* We move charges only when we move a owner of the mm */
5435 if (mm
->owner
== p
) {
5438 VM_BUG_ON(mc
.precharge
);
5439 VM_BUG_ON(mc
.moved_charge
);
5440 VM_BUG_ON(mc
.moved_swap
);
5441 mem_cgroup_start_move(from
);
5442 spin_lock(&mc
.lock
);
5445 spin_unlock(&mc
.lock
);
5446 /* We set mc.moving_task later */
5448 ret
= mem_cgroup_precharge_mc(mm
);
5450 mem_cgroup_clear_mc();
5457 static void mem_cgroup_cancel_attach(struct cgroup
*cgroup
,
5458 struct cgroup_taskset
*tset
)
5460 mem_cgroup_clear_mc();
5463 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
5464 unsigned long addr
, unsigned long end
,
5465 struct mm_walk
*walk
)
5468 struct vm_area_struct
*vma
= walk
->private;
5471 enum mc_target_type target_type
;
5472 union mc_target target
;
5474 struct page_cgroup
*pc
;
5477 * We don't take compound_lock() here but no race with splitting thp
5479 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
5480 * under splitting, which means there's no concurrent thp split,
5481 * - if another thread runs into split_huge_page() just after we
5482 * entered this if-block, the thread must wait for page table lock
5483 * to be unlocked in __split_huge_page_splitting(), where the main
5484 * part of thp split is not executed yet.
5486 if (pmd_trans_huge_lock(pmd
, vma
) == 1) {
5487 if (mc
.precharge
< HPAGE_PMD_NR
) {
5488 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5491 target_type
= get_mctgt_type_thp(vma
, addr
, *pmd
, &target
);
5492 if (target_type
== MC_TARGET_PAGE
) {
5494 if (!isolate_lru_page(page
)) {
5495 pc
= lookup_page_cgroup(page
);
5496 if (!mem_cgroup_move_account(page
, HPAGE_PMD_NR
,
5497 pc
, mc
.from
, mc
.to
)) {
5498 mc
.precharge
-= HPAGE_PMD_NR
;
5499 mc
.moved_charge
+= HPAGE_PMD_NR
;
5501 putback_lru_page(page
);
5505 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5509 if (pmd_trans_unstable(pmd
))
5512 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5513 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
5514 pte_t ptent
= *(pte
++);
5520 switch (get_mctgt_type(vma
, addr
, ptent
, &target
)) {
5521 case MC_TARGET_PAGE
:
5523 if (isolate_lru_page(page
))
5525 pc
= lookup_page_cgroup(page
);
5526 if (!mem_cgroup_move_account(page
, 1, pc
,
5529 /* we uncharge from mc.from later. */
5532 putback_lru_page(page
);
5533 put
: /* get_mctgt_type() gets the page */
5536 case MC_TARGET_SWAP
:
5538 if (!mem_cgroup_move_swap_account(ent
, mc
.from
, mc
.to
)) {
5540 /* we fixup refcnts and charges later. */
5548 pte_unmap_unlock(pte
- 1, ptl
);
5553 * We have consumed all precharges we got in can_attach().
5554 * We try charge one by one, but don't do any additional
5555 * charges to mc.to if we have failed in charge once in attach()
5558 ret
= mem_cgroup_do_precharge(1);
5566 static void mem_cgroup_move_charge(struct mm_struct
*mm
)
5568 struct vm_area_struct
*vma
;
5570 lru_add_drain_all();
5572 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
5574 * Someone who are holding the mmap_sem might be waiting in
5575 * waitq. So we cancel all extra charges, wake up all waiters,
5576 * and retry. Because we cancel precharges, we might not be able
5577 * to move enough charges, but moving charge is a best-effort
5578 * feature anyway, so it wouldn't be a big problem.
5580 __mem_cgroup_clear_mc();
5584 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5586 struct mm_walk mem_cgroup_move_charge_walk
= {
5587 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
5591 if (is_vm_hugetlb_page(vma
))
5593 ret
= walk_page_range(vma
->vm_start
, vma
->vm_end
,
5594 &mem_cgroup_move_charge_walk
);
5597 * means we have consumed all precharges and failed in
5598 * doing additional charge. Just abandon here.
5602 up_read(&mm
->mmap_sem
);
5605 static void mem_cgroup_move_task(struct cgroup
*cont
,
5606 struct cgroup_taskset
*tset
)
5608 struct task_struct
*p
= cgroup_taskset_first(tset
);
5609 struct mm_struct
*mm
= get_task_mm(p
);
5613 mem_cgroup_move_charge(mm
);
5617 mem_cgroup_clear_mc();
5619 #else /* !CONFIG_MMU */
5620 static int mem_cgroup_can_attach(struct cgroup
*cgroup
,
5621 struct cgroup_taskset
*tset
)
5625 static void mem_cgroup_cancel_attach(struct cgroup
*cgroup
,
5626 struct cgroup_taskset
*tset
)
5629 static void mem_cgroup_move_task(struct cgroup
*cont
,
5630 struct cgroup_taskset
*tset
)
5635 struct cgroup_subsys mem_cgroup_subsys
= {
5637 .subsys_id
= mem_cgroup_subsys_id
,
5638 .css_alloc
= mem_cgroup_css_alloc
,
5639 .css_offline
= mem_cgroup_css_offline
,
5640 .css_free
= mem_cgroup_css_free
,
5641 .can_attach
= mem_cgroup_can_attach
,
5642 .cancel_attach
= mem_cgroup_cancel_attach
,
5643 .attach
= mem_cgroup_move_task
,
5644 .base_cftypes
= mem_cgroup_files
,
5649 #ifdef CONFIG_MEMCG_SWAP
5650 static int __init
enable_swap_account(char *s
)
5652 /* consider enabled if no parameter or 1 is given */
5653 if (!strcmp(s
, "1"))
5654 really_do_swap_account
= 1;
5655 else if (!strcmp(s
, "0"))
5656 really_do_swap_account
= 0;
5659 __setup("swapaccount=", enable_swap_account
);