1 /* memcontrol.c - Memory Controller
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
6 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
13 * This program is free software; you can redistribute it and/or modify
14 * it under the terms of the GNU General Public License as published by
15 * the Free Software Foundation; either version 2 of the License, or
16 * (at your option) any later version.
18 * This program is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 * GNU General Public License for more details.
24 #include <linux/res_counter.h>
25 #include <linux/memcontrol.h>
26 #include <linux/cgroup.h>
28 #include <linux/hugetlb.h>
29 #include <linux/pagemap.h>
30 #include <linux/smp.h>
31 #include <linux/page-flags.h>
32 #include <linux/backing-dev.h>
33 #include <linux/bit_spinlock.h>
34 #include <linux/rcupdate.h>
35 #include <linux/limits.h>
36 #include <linux/export.h>
37 #include <linux/mutex.h>
38 #include <linux/rbtree.h>
39 #include <linux/slab.h>
40 #include <linux/swap.h>
41 #include <linux/swapops.h>
42 #include <linux/spinlock.h>
43 #include <linux/eventfd.h>
44 #include <linux/sort.h>
46 #include <linux/seq_file.h>
47 #include <linux/vmalloc.h>
48 #include <linux/mm_inline.h>
49 #include <linux/page_cgroup.h>
50 #include <linux/cpu.h>
51 #include <linux/oom.h>
54 #include <net/tcp_memcontrol.h>
56 #include <asm/uaccess.h>
58 #include <trace/events/vmscan.h>
60 struct cgroup_subsys mem_cgroup_subsys __read_mostly
;
61 #define MEM_CGROUP_RECLAIM_RETRIES 5
62 static struct mem_cgroup
*root_mem_cgroup __read_mostly
;
64 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
65 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
66 int do_swap_account __read_mostly
;
68 /* for remember boot option*/
69 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED
70 static int really_do_swap_account __initdata
= 1;
72 static int really_do_swap_account __initdata
= 0;
76 #define do_swap_account 0
81 * Statistics for memory cgroup.
83 enum mem_cgroup_stat_index
{
85 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
87 MEM_CGROUP_STAT_CACHE
, /* # of pages charged as cache */
88 MEM_CGROUP_STAT_RSS
, /* # of pages charged as anon rss */
89 MEM_CGROUP_STAT_FILE_MAPPED
, /* # of pages charged as file rss */
90 MEM_CGROUP_STAT_SWAPOUT
, /* # of pages, swapped out */
91 MEM_CGROUP_STAT_NSTATS
,
94 static const char * const mem_cgroup_stat_names
[] = {
101 enum mem_cgroup_events_index
{
102 MEM_CGROUP_EVENTS_PGPGIN
, /* # of pages paged in */
103 MEM_CGROUP_EVENTS_PGPGOUT
, /* # of pages paged out */
104 MEM_CGROUP_EVENTS_PGFAULT
, /* # of page-faults */
105 MEM_CGROUP_EVENTS_PGMAJFAULT
, /* # of major page-faults */
106 MEM_CGROUP_EVENTS_NSTATS
,
109 static const char * const mem_cgroup_events_names
[] = {
117 * Per memcg event counter is incremented at every pagein/pageout. With THP,
118 * it will be incremated by the number of pages. This counter is used for
119 * for trigger some periodic events. This is straightforward and better
120 * than using jiffies etc. to handle periodic memcg event.
122 enum mem_cgroup_events_target
{
123 MEM_CGROUP_TARGET_THRESH
,
124 MEM_CGROUP_TARGET_SOFTLIMIT
,
125 MEM_CGROUP_TARGET_NUMAINFO
,
128 #define THRESHOLDS_EVENTS_TARGET 128
129 #define SOFTLIMIT_EVENTS_TARGET 1024
130 #define NUMAINFO_EVENTS_TARGET 1024
132 struct mem_cgroup_stat_cpu
{
133 long count
[MEM_CGROUP_STAT_NSTATS
];
134 unsigned long events
[MEM_CGROUP_EVENTS_NSTATS
];
135 unsigned long nr_page_events
;
136 unsigned long targets
[MEM_CGROUP_NTARGETS
];
139 struct mem_cgroup_reclaim_iter
{
140 /* css_id of the last scanned hierarchy member */
142 /* scan generation, increased every round-trip */
143 unsigned int generation
;
147 * per-zone information in memory controller.
149 struct mem_cgroup_per_zone
{
150 struct lruvec lruvec
;
151 unsigned long lru_size
[NR_LRU_LISTS
];
153 struct mem_cgroup_reclaim_iter reclaim_iter
[DEF_PRIORITY
+ 1];
155 struct rb_node tree_node
; /* RB tree node */
156 unsigned long long usage_in_excess
;/* Set to the value by which */
157 /* the soft limit is exceeded*/
159 struct mem_cgroup
*memcg
; /* Back pointer, we cannot */
160 /* use container_of */
163 struct mem_cgroup_per_node
{
164 struct mem_cgroup_per_zone zoneinfo
[MAX_NR_ZONES
];
167 struct mem_cgroup_lru_info
{
168 struct mem_cgroup_per_node
*nodeinfo
[MAX_NUMNODES
];
172 * Cgroups above their limits are maintained in a RB-Tree, independent of
173 * their hierarchy representation
176 struct mem_cgroup_tree_per_zone
{
177 struct rb_root rb_root
;
181 struct mem_cgroup_tree_per_node
{
182 struct mem_cgroup_tree_per_zone rb_tree_per_zone
[MAX_NR_ZONES
];
185 struct mem_cgroup_tree
{
186 struct mem_cgroup_tree_per_node
*rb_tree_per_node
[MAX_NUMNODES
];
189 static struct mem_cgroup_tree soft_limit_tree __read_mostly
;
191 struct mem_cgroup_threshold
{
192 struct eventfd_ctx
*eventfd
;
197 struct mem_cgroup_threshold_ary
{
198 /* An array index points to threshold just below or equal to usage. */
199 int current_threshold
;
200 /* Size of entries[] */
202 /* Array of thresholds */
203 struct mem_cgroup_threshold entries
[0];
206 struct mem_cgroup_thresholds
{
207 /* Primary thresholds array */
208 struct mem_cgroup_threshold_ary
*primary
;
210 * Spare threshold array.
211 * This is needed to make mem_cgroup_unregister_event() "never fail".
212 * It must be able to store at least primary->size - 1 entries.
214 struct mem_cgroup_threshold_ary
*spare
;
218 struct mem_cgroup_eventfd_list
{
219 struct list_head list
;
220 struct eventfd_ctx
*eventfd
;
223 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
);
224 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
);
227 * The memory controller data structure. The memory controller controls both
228 * page cache and RSS per cgroup. We would eventually like to provide
229 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
230 * to help the administrator determine what knobs to tune.
232 * TODO: Add a water mark for the memory controller. Reclaim will begin when
233 * we hit the water mark. May be even add a low water mark, such that
234 * no reclaim occurs from a cgroup at it's low water mark, this is
235 * a feature that will be implemented much later in the future.
238 struct cgroup_subsys_state css
;
240 * the counter to account for memory usage
242 struct res_counter res
;
246 * the counter to account for mem+swap usage.
248 struct res_counter memsw
;
251 * rcu_freeing is used only when freeing struct mem_cgroup,
252 * so put it into a union to avoid wasting more memory.
253 * It must be disjoint from the css field. It could be
254 * in a union with the res field, but res plays a much
255 * larger part in mem_cgroup life than memsw, and might
256 * be of interest, even at time of free, when debugging.
257 * So share rcu_head with the less interesting memsw.
259 struct rcu_head rcu_freeing
;
261 * But when using vfree(), that cannot be done at
262 * interrupt time, so we must then queue the work.
264 struct work_struct work_freeing
;
268 * Per cgroup active and inactive list, similar to the
269 * per zone LRU lists.
271 struct mem_cgroup_lru_info info
;
272 int last_scanned_node
;
274 nodemask_t scan_nodes
;
275 atomic_t numainfo_events
;
276 atomic_t numainfo_updating
;
279 * Should the accounting and control be hierarchical, per subtree?
289 /* OOM-Killer disable */
290 int oom_kill_disable
;
292 /* set when res.limit == memsw.limit */
293 bool memsw_is_minimum
;
295 /* protect arrays of thresholds */
296 struct mutex thresholds_lock
;
298 /* thresholds for memory usage. RCU-protected */
299 struct mem_cgroup_thresholds thresholds
;
301 /* thresholds for mem+swap usage. RCU-protected */
302 struct mem_cgroup_thresholds memsw_thresholds
;
304 /* For oom notifier event fd */
305 struct list_head oom_notify
;
308 * Should we move charges of a task when a task is moved into this
309 * mem_cgroup ? And what type of charges should we move ?
311 unsigned long move_charge_at_immigrate
;
313 * set > 0 if pages under this cgroup are moving to other cgroup.
315 atomic_t moving_account
;
316 /* taken only while moving_account > 0 */
317 spinlock_t move_lock
;
321 struct mem_cgroup_stat_cpu __percpu
*stat
;
323 * used when a cpu is offlined or other synchronizations
324 * See mem_cgroup_read_stat().
326 struct mem_cgroup_stat_cpu nocpu_base
;
327 spinlock_t pcp_counter_lock
;
330 struct tcp_memcontrol tcp_mem
;
334 /* Stuffs for move charges at task migration. */
336 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
337 * left-shifted bitmap of these types.
340 MOVE_CHARGE_TYPE_ANON
, /* private anonymous page and swap of it */
341 MOVE_CHARGE_TYPE_FILE
, /* file page(including tmpfs) and swap of it */
345 /* "mc" and its members are protected by cgroup_mutex */
346 static struct move_charge_struct
{
347 spinlock_t lock
; /* for from, to */
348 struct mem_cgroup
*from
;
349 struct mem_cgroup
*to
;
350 unsigned long precharge
;
351 unsigned long moved_charge
;
352 unsigned long moved_swap
;
353 struct task_struct
*moving_task
; /* a task moving charges */
354 wait_queue_head_t waitq
; /* a waitq for other context */
356 .lock
= __SPIN_LOCK_UNLOCKED(mc
.lock
),
357 .waitq
= __WAIT_QUEUE_HEAD_INITIALIZER(mc
.waitq
),
360 static bool move_anon(void)
362 return test_bit(MOVE_CHARGE_TYPE_ANON
,
363 &mc
.to
->move_charge_at_immigrate
);
366 static bool move_file(void)
368 return test_bit(MOVE_CHARGE_TYPE_FILE
,
369 &mc
.to
->move_charge_at_immigrate
);
373 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
374 * limit reclaim to prevent infinite loops, if they ever occur.
376 #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
377 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
380 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
381 MEM_CGROUP_CHARGE_TYPE_MAPPED
,
382 MEM_CGROUP_CHARGE_TYPE_SHMEM
, /* used by page migration of shmem */
383 MEM_CGROUP_CHARGE_TYPE_FORCE
, /* used by force_empty */
384 MEM_CGROUP_CHARGE_TYPE_SWAPOUT
, /* for accounting swapcache */
385 MEM_CGROUP_CHARGE_TYPE_DROP
, /* a page was unused swap cache */
389 /* for encoding cft->private value on file */
392 #define _OOM_TYPE (2)
393 #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
394 #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
395 #define MEMFILE_ATTR(val) ((val) & 0xffff)
396 /* Used for OOM nofiier */
397 #define OOM_CONTROL (0)
400 * Reclaim flags for mem_cgroup_hierarchical_reclaim
402 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
403 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
404 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
405 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
407 static void mem_cgroup_get(struct mem_cgroup
*memcg
);
408 static void mem_cgroup_put(struct mem_cgroup
*memcg
);
410 /* Writing them here to avoid exposing memcg's inner layout */
411 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
412 #include <net/sock.h>
415 static bool mem_cgroup_is_root(struct mem_cgroup
*memcg
);
416 void sock_update_memcg(struct sock
*sk
)
418 if (mem_cgroup_sockets_enabled
) {
419 struct mem_cgroup
*memcg
;
421 BUG_ON(!sk
->sk_prot
->proto_cgroup
);
423 /* Socket cloning can throw us here with sk_cgrp already
424 * filled. It won't however, necessarily happen from
425 * process context. So the test for root memcg given
426 * the current task's memcg won't help us in this case.
428 * Respecting the original socket's memcg is a better
429 * decision in this case.
432 BUG_ON(mem_cgroup_is_root(sk
->sk_cgrp
->memcg
));
433 mem_cgroup_get(sk
->sk_cgrp
->memcg
);
438 memcg
= mem_cgroup_from_task(current
);
439 if (!mem_cgroup_is_root(memcg
)) {
440 mem_cgroup_get(memcg
);
441 sk
->sk_cgrp
= sk
->sk_prot
->proto_cgroup(memcg
);
446 EXPORT_SYMBOL(sock_update_memcg
);
448 void sock_release_memcg(struct sock
*sk
)
450 if (mem_cgroup_sockets_enabled
&& sk
->sk_cgrp
) {
451 struct mem_cgroup
*memcg
;
452 WARN_ON(!sk
->sk_cgrp
->memcg
);
453 memcg
= sk
->sk_cgrp
->memcg
;
454 mem_cgroup_put(memcg
);
459 struct cg_proto
*tcp_proto_cgroup(struct mem_cgroup
*memcg
)
461 if (!memcg
|| mem_cgroup_is_root(memcg
))
464 return &memcg
->tcp_mem
.cg_proto
;
466 EXPORT_SYMBOL(tcp_proto_cgroup
);
467 #endif /* CONFIG_INET */
468 #endif /* CONFIG_CGROUP_MEM_RES_CTLR_KMEM */
470 static void drain_all_stock_async(struct mem_cgroup
*memcg
);
472 static struct mem_cgroup_per_zone
*
473 mem_cgroup_zoneinfo(struct mem_cgroup
*memcg
, int nid
, int zid
)
475 return &memcg
->info
.nodeinfo
[nid
]->zoneinfo
[zid
];
478 struct cgroup_subsys_state
*mem_cgroup_css(struct mem_cgroup
*memcg
)
483 static struct mem_cgroup_per_zone
*
484 page_cgroup_zoneinfo(struct mem_cgroup
*memcg
, struct page
*page
)
486 int nid
= page_to_nid(page
);
487 int zid
= page_zonenum(page
);
489 return mem_cgroup_zoneinfo(memcg
, nid
, zid
);
492 static struct mem_cgroup_tree_per_zone
*
493 soft_limit_tree_node_zone(int nid
, int zid
)
495 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
498 static struct mem_cgroup_tree_per_zone
*
499 soft_limit_tree_from_page(struct page
*page
)
501 int nid
= page_to_nid(page
);
502 int zid
= page_zonenum(page
);
504 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
508 __mem_cgroup_insert_exceeded(struct mem_cgroup
*memcg
,
509 struct mem_cgroup_per_zone
*mz
,
510 struct mem_cgroup_tree_per_zone
*mctz
,
511 unsigned long long new_usage_in_excess
)
513 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
514 struct rb_node
*parent
= NULL
;
515 struct mem_cgroup_per_zone
*mz_node
;
520 mz
->usage_in_excess
= new_usage_in_excess
;
521 if (!mz
->usage_in_excess
)
525 mz_node
= rb_entry(parent
, struct mem_cgroup_per_zone
,
527 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
)
530 * We can't avoid mem cgroups that are over their soft
531 * limit by the same amount
533 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
536 rb_link_node(&mz
->tree_node
, parent
, p
);
537 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
542 __mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
543 struct mem_cgroup_per_zone
*mz
,
544 struct mem_cgroup_tree_per_zone
*mctz
)
548 rb_erase(&mz
->tree_node
, &mctz
->rb_root
);
553 mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
554 struct mem_cgroup_per_zone
*mz
,
555 struct mem_cgroup_tree_per_zone
*mctz
)
557 spin_lock(&mctz
->lock
);
558 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
559 spin_unlock(&mctz
->lock
);
563 static void mem_cgroup_update_tree(struct mem_cgroup
*memcg
, struct page
*page
)
565 unsigned long long excess
;
566 struct mem_cgroup_per_zone
*mz
;
567 struct mem_cgroup_tree_per_zone
*mctz
;
568 int nid
= page_to_nid(page
);
569 int zid
= page_zonenum(page
);
570 mctz
= soft_limit_tree_from_page(page
);
573 * Necessary to update all ancestors when hierarchy is used.
574 * because their event counter is not touched.
576 for (; memcg
; memcg
= parent_mem_cgroup(memcg
)) {
577 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
578 excess
= res_counter_soft_limit_excess(&memcg
->res
);
580 * We have to update the tree if mz is on RB-tree or
581 * mem is over its softlimit.
583 if (excess
|| mz
->on_tree
) {
584 spin_lock(&mctz
->lock
);
585 /* if on-tree, remove it */
587 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
589 * Insert again. mz->usage_in_excess will be updated.
590 * If excess is 0, no tree ops.
592 __mem_cgroup_insert_exceeded(memcg
, mz
, mctz
, excess
);
593 spin_unlock(&mctz
->lock
);
598 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*memcg
)
601 struct mem_cgroup_per_zone
*mz
;
602 struct mem_cgroup_tree_per_zone
*mctz
;
604 for_each_node(node
) {
605 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
606 mz
= mem_cgroup_zoneinfo(memcg
, node
, zone
);
607 mctz
= soft_limit_tree_node_zone(node
, zone
);
608 mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
613 static struct mem_cgroup_per_zone
*
614 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
616 struct rb_node
*rightmost
= NULL
;
617 struct mem_cgroup_per_zone
*mz
;
621 rightmost
= rb_last(&mctz
->rb_root
);
623 goto done
; /* Nothing to reclaim from */
625 mz
= rb_entry(rightmost
, struct mem_cgroup_per_zone
, tree_node
);
627 * Remove the node now but someone else can add it back,
628 * we will to add it back at the end of reclaim to its correct
629 * position in the tree.
631 __mem_cgroup_remove_exceeded(mz
->memcg
, mz
, mctz
);
632 if (!res_counter_soft_limit_excess(&mz
->memcg
->res
) ||
633 !css_tryget(&mz
->memcg
->css
))
639 static struct mem_cgroup_per_zone
*
640 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
642 struct mem_cgroup_per_zone
*mz
;
644 spin_lock(&mctz
->lock
);
645 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
646 spin_unlock(&mctz
->lock
);
651 * Implementation Note: reading percpu statistics for memcg.
653 * Both of vmstat[] and percpu_counter has threshold and do periodic
654 * synchronization to implement "quick" read. There are trade-off between
655 * reading cost and precision of value. Then, we may have a chance to implement
656 * a periodic synchronizion of counter in memcg's counter.
658 * But this _read() function is used for user interface now. The user accounts
659 * memory usage by memory cgroup and he _always_ requires exact value because
660 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
661 * have to visit all online cpus and make sum. So, for now, unnecessary
662 * synchronization is not implemented. (just implemented for cpu hotplug)
664 * If there are kernel internal actions which can make use of some not-exact
665 * value, and reading all cpu value can be performance bottleneck in some
666 * common workload, threashold and synchonization as vmstat[] should be
669 static long mem_cgroup_read_stat(struct mem_cgroup
*memcg
,
670 enum mem_cgroup_stat_index idx
)
676 for_each_online_cpu(cpu
)
677 val
+= per_cpu(memcg
->stat
->count
[idx
], cpu
);
678 #ifdef CONFIG_HOTPLUG_CPU
679 spin_lock(&memcg
->pcp_counter_lock
);
680 val
+= memcg
->nocpu_base
.count
[idx
];
681 spin_unlock(&memcg
->pcp_counter_lock
);
687 static void mem_cgroup_swap_statistics(struct mem_cgroup
*memcg
,
690 int val
= (charge
) ? 1 : -1;
691 this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_SWAPOUT
], val
);
694 static unsigned long mem_cgroup_read_events(struct mem_cgroup
*memcg
,
695 enum mem_cgroup_events_index idx
)
697 unsigned long val
= 0;
700 for_each_online_cpu(cpu
)
701 val
+= per_cpu(memcg
->stat
->events
[idx
], cpu
);
702 #ifdef CONFIG_HOTPLUG_CPU
703 spin_lock(&memcg
->pcp_counter_lock
);
704 val
+= memcg
->nocpu_base
.events
[idx
];
705 spin_unlock(&memcg
->pcp_counter_lock
);
710 static void mem_cgroup_charge_statistics(struct mem_cgroup
*memcg
,
711 bool anon
, int nr_pages
)
716 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
717 * counted as CACHE even if it's on ANON LRU.
720 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_RSS
],
723 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_CACHE
],
726 /* pagein of a big page is an event. So, ignore page size */
728 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGIN
]);
730 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGOUT
]);
731 nr_pages
= -nr_pages
; /* for event */
734 __this_cpu_add(memcg
->stat
->nr_page_events
, nr_pages
);
740 mem_cgroup_get_lru_size(struct lruvec
*lruvec
, enum lru_list lru
)
742 struct mem_cgroup_per_zone
*mz
;
744 mz
= container_of(lruvec
, struct mem_cgroup_per_zone
, lruvec
);
745 return mz
->lru_size
[lru
];
749 mem_cgroup_zone_nr_lru_pages(struct mem_cgroup
*memcg
, int nid
, int zid
,
750 unsigned int lru_mask
)
752 struct mem_cgroup_per_zone
*mz
;
754 unsigned long ret
= 0;
756 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
759 if (BIT(lru
) & lru_mask
)
760 ret
+= mz
->lru_size
[lru
];
766 mem_cgroup_node_nr_lru_pages(struct mem_cgroup
*memcg
,
767 int nid
, unsigned int lru_mask
)
772 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++)
773 total
+= mem_cgroup_zone_nr_lru_pages(memcg
,
779 static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup
*memcg
,
780 unsigned int lru_mask
)
785 for_each_node_state(nid
, N_HIGH_MEMORY
)
786 total
+= mem_cgroup_node_nr_lru_pages(memcg
, nid
, lru_mask
);
790 static bool mem_cgroup_event_ratelimit(struct mem_cgroup
*memcg
,
791 enum mem_cgroup_events_target target
)
793 unsigned long val
, next
;
795 val
= __this_cpu_read(memcg
->stat
->nr_page_events
);
796 next
= __this_cpu_read(memcg
->stat
->targets
[target
]);
797 /* from time_after() in jiffies.h */
798 if ((long)next
- (long)val
< 0) {
800 case MEM_CGROUP_TARGET_THRESH
:
801 next
= val
+ THRESHOLDS_EVENTS_TARGET
;
803 case MEM_CGROUP_TARGET_SOFTLIMIT
:
804 next
= val
+ SOFTLIMIT_EVENTS_TARGET
;
806 case MEM_CGROUP_TARGET_NUMAINFO
:
807 next
= val
+ NUMAINFO_EVENTS_TARGET
;
812 __this_cpu_write(memcg
->stat
->targets
[target
], next
);
819 * Check events in order.
822 static void memcg_check_events(struct mem_cgroup
*memcg
, struct page
*page
)
825 /* threshold event is triggered in finer grain than soft limit */
826 if (unlikely(mem_cgroup_event_ratelimit(memcg
,
827 MEM_CGROUP_TARGET_THRESH
))) {
829 bool do_numainfo __maybe_unused
;
831 do_softlimit
= mem_cgroup_event_ratelimit(memcg
,
832 MEM_CGROUP_TARGET_SOFTLIMIT
);
834 do_numainfo
= mem_cgroup_event_ratelimit(memcg
,
835 MEM_CGROUP_TARGET_NUMAINFO
);
839 mem_cgroup_threshold(memcg
);
840 if (unlikely(do_softlimit
))
841 mem_cgroup_update_tree(memcg
, page
);
843 if (unlikely(do_numainfo
))
844 atomic_inc(&memcg
->numainfo_events
);
850 struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
852 return container_of(cgroup_subsys_state(cont
,
853 mem_cgroup_subsys_id
), struct mem_cgroup
,
857 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
860 * mm_update_next_owner() may clear mm->owner to NULL
861 * if it races with swapoff, page migration, etc.
862 * So this can be called with p == NULL.
867 return container_of(task_subsys_state(p
, mem_cgroup_subsys_id
),
868 struct mem_cgroup
, css
);
871 struct mem_cgroup
*try_get_mem_cgroup_from_mm(struct mm_struct
*mm
)
873 struct mem_cgroup
*memcg
= NULL
;
878 * Because we have no locks, mm->owner's may be being moved to other
879 * cgroup. We use css_tryget() here even if this looks
880 * pessimistic (rather than adding locks here).
884 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
885 if (unlikely(!memcg
))
887 } while (!css_tryget(&memcg
->css
));
893 * mem_cgroup_iter - iterate over memory cgroup hierarchy
894 * @root: hierarchy root
895 * @prev: previously returned memcg, NULL on first invocation
896 * @reclaim: cookie for shared reclaim walks, NULL for full walks
898 * Returns references to children of the hierarchy below @root, or
899 * @root itself, or %NULL after a full round-trip.
901 * Caller must pass the return value in @prev on subsequent
902 * invocations for reference counting, or use mem_cgroup_iter_break()
903 * to cancel a hierarchy walk before the round-trip is complete.
905 * Reclaimers can specify a zone and a priority level in @reclaim to
906 * divide up the memcgs in the hierarchy among all concurrent
907 * reclaimers operating on the same zone and priority.
909 struct mem_cgroup
*mem_cgroup_iter(struct mem_cgroup
*root
,
910 struct mem_cgroup
*prev
,
911 struct mem_cgroup_reclaim_cookie
*reclaim
)
913 struct mem_cgroup
*memcg
= NULL
;
916 if (mem_cgroup_disabled())
920 root
= root_mem_cgroup
;
922 if (prev
&& !reclaim
)
923 id
= css_id(&prev
->css
);
925 if (prev
&& prev
!= root
)
928 if (!root
->use_hierarchy
&& root
!= root_mem_cgroup
) {
935 struct mem_cgroup_reclaim_iter
*uninitialized_var(iter
);
936 struct cgroup_subsys_state
*css
;
939 int nid
= zone_to_nid(reclaim
->zone
);
940 int zid
= zone_idx(reclaim
->zone
);
941 struct mem_cgroup_per_zone
*mz
;
943 mz
= mem_cgroup_zoneinfo(root
, nid
, zid
);
944 iter
= &mz
->reclaim_iter
[reclaim
->priority
];
945 if (prev
&& reclaim
->generation
!= iter
->generation
)
951 css
= css_get_next(&mem_cgroup_subsys
, id
+ 1, &root
->css
, &id
);
953 if (css
== &root
->css
|| css_tryget(css
))
954 memcg
= container_of(css
,
955 struct mem_cgroup
, css
);
964 else if (!prev
&& memcg
)
965 reclaim
->generation
= iter
->generation
;
975 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
976 * @root: hierarchy root
977 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
979 void mem_cgroup_iter_break(struct mem_cgroup
*root
,
980 struct mem_cgroup
*prev
)
983 root
= root_mem_cgroup
;
984 if (prev
&& prev
!= root
)
989 * Iteration constructs for visiting all cgroups (under a tree). If
990 * loops are exited prematurely (break), mem_cgroup_iter_break() must
991 * be used for reference counting.
993 #define for_each_mem_cgroup_tree(iter, root) \
994 for (iter = mem_cgroup_iter(root, NULL, NULL); \
996 iter = mem_cgroup_iter(root, iter, NULL))
998 #define for_each_mem_cgroup(iter) \
999 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
1001 iter = mem_cgroup_iter(NULL, iter, NULL))
1003 static inline bool mem_cgroup_is_root(struct mem_cgroup
*memcg
)
1005 return (memcg
== root_mem_cgroup
);
1008 void mem_cgroup_count_vm_event(struct mm_struct
*mm
, enum vm_event_item idx
)
1010 struct mem_cgroup
*memcg
;
1016 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
1017 if (unlikely(!memcg
))
1022 this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGFAULT
]);
1025 this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGMAJFAULT
]);
1033 EXPORT_SYMBOL(mem_cgroup_count_vm_event
);
1036 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1037 * @zone: zone of the wanted lruvec
1038 * @mem: memcg of the wanted lruvec
1040 * Returns the lru list vector holding pages for the given @zone and
1041 * @mem. This can be the global zone lruvec, if the memory controller
1044 struct lruvec
*mem_cgroup_zone_lruvec(struct zone
*zone
,
1045 struct mem_cgroup
*memcg
)
1047 struct mem_cgroup_per_zone
*mz
;
1049 if (mem_cgroup_disabled())
1050 return &zone
->lruvec
;
1052 mz
= mem_cgroup_zoneinfo(memcg
, zone_to_nid(zone
), zone_idx(zone
));
1057 * Following LRU functions are allowed to be used without PCG_LOCK.
1058 * Operations are called by routine of global LRU independently from memcg.
1059 * What we have to take care of here is validness of pc->mem_cgroup.
1061 * Changes to pc->mem_cgroup happens when
1064 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1065 * It is added to LRU before charge.
1066 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1067 * When moving account, the page is not on LRU. It's isolated.
1071 * mem_cgroup_lru_add_list - account for adding an lru page and return lruvec
1072 * @zone: zone of the page
1076 * This function accounts for @page being added to @lru, and returns
1077 * the lruvec for the given @zone and the memcg @page is charged to.
1079 * The callsite is then responsible for physically linking the page to
1080 * the returned lruvec->lists[@lru].
1082 struct lruvec
*mem_cgroup_lru_add_list(struct zone
*zone
, struct page
*page
,
1085 struct mem_cgroup_per_zone
*mz
;
1086 struct mem_cgroup
*memcg
;
1087 struct page_cgroup
*pc
;
1089 if (mem_cgroup_disabled())
1090 return &zone
->lruvec
;
1092 pc
= lookup_page_cgroup(page
);
1093 memcg
= pc
->mem_cgroup
;
1096 * Surreptitiously switch any uncharged page to root:
1097 * an uncharged page off lru does nothing to secure
1098 * its former mem_cgroup from sudden removal.
1100 * Our caller holds lru_lock, and PageCgroupUsed is updated
1101 * under page_cgroup lock: between them, they make all uses
1102 * of pc->mem_cgroup safe.
1104 if (!PageCgroupUsed(pc
) && memcg
!= root_mem_cgroup
)
1105 pc
->mem_cgroup
= memcg
= root_mem_cgroup
;
1107 mz
= page_cgroup_zoneinfo(memcg
, page
);
1108 /* compound_order() is stabilized through lru_lock */
1109 mz
->lru_size
[lru
] += 1 << compound_order(page
);
1114 * mem_cgroup_lru_del_list - account for removing an lru page
1118 * This function accounts for @page being removed from @lru.
1120 * The callsite is then responsible for physically unlinking
1123 void mem_cgroup_lru_del_list(struct page
*page
, enum lru_list lru
)
1125 struct mem_cgroup_per_zone
*mz
;
1126 struct mem_cgroup
*memcg
;
1127 struct page_cgroup
*pc
;
1129 if (mem_cgroup_disabled())
1132 pc
= lookup_page_cgroup(page
);
1133 memcg
= pc
->mem_cgroup
;
1135 mz
= page_cgroup_zoneinfo(memcg
, page
);
1136 /* huge page split is done under lru_lock. so, we have no races. */
1137 VM_BUG_ON(mz
->lru_size
[lru
] < (1 << compound_order(page
)));
1138 mz
->lru_size
[lru
] -= 1 << compound_order(page
);
1142 * mem_cgroup_lru_move_lists - account for moving a page between lrus
1143 * @zone: zone of the page
1145 * @from: current lru
1148 * This function accounts for @page being moved between the lrus @from
1149 * and @to, and returns the lruvec for the given @zone and the memcg
1150 * @page is charged to.
1152 * The callsite is then responsible for physically relinking
1153 * @page->lru to the returned lruvec->lists[@to].
1155 struct lruvec
*mem_cgroup_lru_move_lists(struct zone
*zone
,
1160 /* XXX: Optimize this, especially for @from == @to */
1161 mem_cgroup_lru_del_list(page
, from
);
1162 return mem_cgroup_lru_add_list(zone
, page
, to
);
1166 * Checks whether given mem is same or in the root_mem_cgroup's
1169 bool __mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_memcg
,
1170 struct mem_cgroup
*memcg
)
1172 if (root_memcg
== memcg
)
1174 if (!root_memcg
->use_hierarchy
)
1176 return css_is_ancestor(&memcg
->css
, &root_memcg
->css
);
1179 static bool mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_memcg
,
1180 struct mem_cgroup
*memcg
)
1185 ret
= __mem_cgroup_same_or_subtree(root_memcg
, memcg
);
1190 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*memcg
)
1193 struct mem_cgroup
*curr
= NULL
;
1194 struct task_struct
*p
;
1196 p
= find_lock_task_mm(task
);
1198 curr
= try_get_mem_cgroup_from_mm(p
->mm
);
1202 * All threads may have already detached their mm's, but the oom
1203 * killer still needs to detect if they have already been oom
1204 * killed to prevent needlessly killing additional tasks.
1207 curr
= mem_cgroup_from_task(task
);
1209 css_get(&curr
->css
);
1215 * We should check use_hierarchy of "memcg" not "curr". Because checking
1216 * use_hierarchy of "curr" here make this function true if hierarchy is
1217 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1218 * hierarchy(even if use_hierarchy is disabled in "memcg").
1220 ret
= mem_cgroup_same_or_subtree(memcg
, curr
);
1221 css_put(&curr
->css
);
1225 int mem_cgroup_inactive_anon_is_low(struct lruvec
*lruvec
)
1227 unsigned long inactive_ratio
;
1228 unsigned long inactive
;
1229 unsigned long active
;
1232 inactive
= mem_cgroup_get_lru_size(lruvec
, LRU_INACTIVE_ANON
);
1233 active
= mem_cgroup_get_lru_size(lruvec
, LRU_ACTIVE_ANON
);
1235 gb
= (inactive
+ active
) >> (30 - PAGE_SHIFT
);
1237 inactive_ratio
= int_sqrt(10 * gb
);
1241 return inactive
* inactive_ratio
< active
;
1244 int mem_cgroup_inactive_file_is_low(struct lruvec
*lruvec
)
1246 unsigned long active
;
1247 unsigned long inactive
;
1249 inactive
= mem_cgroup_get_lru_size(lruvec
, LRU_INACTIVE_FILE
);
1250 active
= mem_cgroup_get_lru_size(lruvec
, LRU_ACTIVE_FILE
);
1252 return (active
> inactive
);
1255 struct zone_reclaim_stat
*
1256 mem_cgroup_get_reclaim_stat_from_page(struct page
*page
)
1258 struct page_cgroup
*pc
;
1259 struct mem_cgroup_per_zone
*mz
;
1261 if (mem_cgroup_disabled())
1264 pc
= lookup_page_cgroup(page
);
1265 if (!PageCgroupUsed(pc
))
1267 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1269 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
1270 return &mz
->lruvec
.reclaim_stat
;
1273 #define mem_cgroup_from_res_counter(counter, member) \
1274 container_of(counter, struct mem_cgroup, member)
1277 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1278 * @mem: the memory cgroup
1280 * Returns the maximum amount of memory @mem can be charged with, in
1283 static unsigned long mem_cgroup_margin(struct mem_cgroup
*memcg
)
1285 unsigned long long margin
;
1287 margin
= res_counter_margin(&memcg
->res
);
1288 if (do_swap_account
)
1289 margin
= min(margin
, res_counter_margin(&memcg
->memsw
));
1290 return margin
>> PAGE_SHIFT
;
1293 int mem_cgroup_swappiness(struct mem_cgroup
*memcg
)
1295 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
1298 if (cgrp
->parent
== NULL
)
1299 return vm_swappiness
;
1301 return memcg
->swappiness
;
1305 * memcg->moving_account is used for checking possibility that some thread is
1306 * calling move_account(). When a thread on CPU-A starts moving pages under
1307 * a memcg, other threads should check memcg->moving_account under
1308 * rcu_read_lock(), like this:
1312 * memcg->moving_account+1 if (memcg->mocing_account)
1314 * synchronize_rcu() update something.
1319 /* for quick checking without looking up memcg */
1320 atomic_t memcg_moving __read_mostly
;
1322 static void mem_cgroup_start_move(struct mem_cgroup
*memcg
)
1324 atomic_inc(&memcg_moving
);
1325 atomic_inc(&memcg
->moving_account
);
1329 static void mem_cgroup_end_move(struct mem_cgroup
*memcg
)
1332 * Now, mem_cgroup_clear_mc() may call this function with NULL.
1333 * We check NULL in callee rather than caller.
1336 atomic_dec(&memcg_moving
);
1337 atomic_dec(&memcg
->moving_account
);
1342 * 2 routines for checking "mem" is under move_account() or not.
1344 * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This
1345 * is used for avoiding races in accounting. If true,
1346 * pc->mem_cgroup may be overwritten.
1348 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1349 * under hierarchy of moving cgroups. This is for
1350 * waiting at hith-memory prressure caused by "move".
1353 static bool mem_cgroup_stolen(struct mem_cgroup
*memcg
)
1355 VM_BUG_ON(!rcu_read_lock_held());
1356 return atomic_read(&memcg
->moving_account
) > 0;
1359 static bool mem_cgroup_under_move(struct mem_cgroup
*memcg
)
1361 struct mem_cgroup
*from
;
1362 struct mem_cgroup
*to
;
1365 * Unlike task_move routines, we access mc.to, mc.from not under
1366 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1368 spin_lock(&mc
.lock
);
1374 ret
= mem_cgroup_same_or_subtree(memcg
, from
)
1375 || mem_cgroup_same_or_subtree(memcg
, to
);
1377 spin_unlock(&mc
.lock
);
1381 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*memcg
)
1383 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1384 if (mem_cgroup_under_move(memcg
)) {
1386 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1387 /* moving charge context might have finished. */
1390 finish_wait(&mc
.waitq
, &wait
);
1398 * Take this lock when
1399 * - a code tries to modify page's memcg while it's USED.
1400 * - a code tries to modify page state accounting in a memcg.
1401 * see mem_cgroup_stolen(), too.
1403 static void move_lock_mem_cgroup(struct mem_cgroup
*memcg
,
1404 unsigned long *flags
)
1406 spin_lock_irqsave(&memcg
->move_lock
, *flags
);
1409 static void move_unlock_mem_cgroup(struct mem_cgroup
*memcg
,
1410 unsigned long *flags
)
1412 spin_unlock_irqrestore(&memcg
->move_lock
, *flags
);
1416 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1417 * @memcg: The memory cgroup that went over limit
1418 * @p: Task that is going to be killed
1420 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1423 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1425 struct cgroup
*task_cgrp
;
1426 struct cgroup
*mem_cgrp
;
1428 * Need a buffer in BSS, can't rely on allocations. The code relies
1429 * on the assumption that OOM is serialized for memory controller.
1430 * If this assumption is broken, revisit this code.
1432 static char memcg_name
[PATH_MAX
];
1440 mem_cgrp
= memcg
->css
.cgroup
;
1441 task_cgrp
= task_cgroup(p
, mem_cgroup_subsys_id
);
1443 ret
= cgroup_path(task_cgrp
, memcg_name
, PATH_MAX
);
1446 * Unfortunately, we are unable to convert to a useful name
1447 * But we'll still print out the usage information
1454 printk(KERN_INFO
"Task in %s killed", memcg_name
);
1457 ret
= cgroup_path(mem_cgrp
, memcg_name
, PATH_MAX
);
1465 * Continues from above, so we don't need an KERN_ level
1467 printk(KERN_CONT
" as a result of limit of %s\n", memcg_name
);
1470 printk(KERN_INFO
"memory: usage %llukB, limit %llukB, failcnt %llu\n",
1471 res_counter_read_u64(&memcg
->res
, RES_USAGE
) >> 10,
1472 res_counter_read_u64(&memcg
->res
, RES_LIMIT
) >> 10,
1473 res_counter_read_u64(&memcg
->res
, RES_FAILCNT
));
1474 printk(KERN_INFO
"memory+swap: usage %llukB, limit %llukB, "
1476 res_counter_read_u64(&memcg
->memsw
, RES_USAGE
) >> 10,
1477 res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
) >> 10,
1478 res_counter_read_u64(&memcg
->memsw
, RES_FAILCNT
));
1482 * This function returns the number of memcg under hierarchy tree. Returns
1483 * 1(self count) if no children.
1485 static int mem_cgroup_count_children(struct mem_cgroup
*memcg
)
1488 struct mem_cgroup
*iter
;
1490 for_each_mem_cgroup_tree(iter
, memcg
)
1496 * Return the memory (and swap, if configured) limit for a memcg.
1498 u64
mem_cgroup_get_limit(struct mem_cgroup
*memcg
)
1503 limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
1504 limit
+= total_swap_pages
<< PAGE_SHIFT
;
1506 memsw
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
1508 * If memsw is finite and limits the amount of swap space available
1509 * to this memcg, return that limit.
1511 return min(limit
, memsw
);
1514 static unsigned long mem_cgroup_reclaim(struct mem_cgroup
*memcg
,
1516 unsigned long flags
)
1518 unsigned long total
= 0;
1519 bool noswap
= false;
1522 if (flags
& MEM_CGROUP_RECLAIM_NOSWAP
)
1524 if (!(flags
& MEM_CGROUP_RECLAIM_SHRINK
) && memcg
->memsw_is_minimum
)
1527 for (loop
= 0; loop
< MEM_CGROUP_MAX_RECLAIM_LOOPS
; loop
++) {
1529 drain_all_stock_async(memcg
);
1530 total
+= try_to_free_mem_cgroup_pages(memcg
, gfp_mask
, noswap
);
1532 * Allow limit shrinkers, which are triggered directly
1533 * by userspace, to catch signals and stop reclaim
1534 * after minimal progress, regardless of the margin.
1536 if (total
&& (flags
& MEM_CGROUP_RECLAIM_SHRINK
))
1538 if (mem_cgroup_margin(memcg
))
1541 * If nothing was reclaimed after two attempts, there
1542 * may be no reclaimable pages in this hierarchy.
1551 * test_mem_cgroup_node_reclaimable
1552 * @mem: the target memcg
1553 * @nid: the node ID to be checked.
1554 * @noswap : specify true here if the user wants flle only information.
1556 * This function returns whether the specified memcg contains any
1557 * reclaimable pages on a node. Returns true if there are any reclaimable
1558 * pages in the node.
1560 static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup
*memcg
,
1561 int nid
, bool noswap
)
1563 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_FILE
))
1565 if (noswap
|| !total_swap_pages
)
1567 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_ANON
))
1572 #if MAX_NUMNODES > 1
1575 * Always updating the nodemask is not very good - even if we have an empty
1576 * list or the wrong list here, we can start from some node and traverse all
1577 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1580 static void mem_cgroup_may_update_nodemask(struct mem_cgroup
*memcg
)
1584 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1585 * pagein/pageout changes since the last update.
1587 if (!atomic_read(&memcg
->numainfo_events
))
1589 if (atomic_inc_return(&memcg
->numainfo_updating
) > 1)
1592 /* make a nodemask where this memcg uses memory from */
1593 memcg
->scan_nodes
= node_states
[N_HIGH_MEMORY
];
1595 for_each_node_mask(nid
, node_states
[N_HIGH_MEMORY
]) {
1597 if (!test_mem_cgroup_node_reclaimable(memcg
, nid
, false))
1598 node_clear(nid
, memcg
->scan_nodes
);
1601 atomic_set(&memcg
->numainfo_events
, 0);
1602 atomic_set(&memcg
->numainfo_updating
, 0);
1606 * Selecting a node where we start reclaim from. Because what we need is just
1607 * reducing usage counter, start from anywhere is O,K. Considering
1608 * memory reclaim from current node, there are pros. and cons.
1610 * Freeing memory from current node means freeing memory from a node which
1611 * we'll use or we've used. So, it may make LRU bad. And if several threads
1612 * hit limits, it will see a contention on a node. But freeing from remote
1613 * node means more costs for memory reclaim because of memory latency.
1615 * Now, we use round-robin. Better algorithm is welcomed.
1617 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1621 mem_cgroup_may_update_nodemask(memcg
);
1622 node
= memcg
->last_scanned_node
;
1624 node
= next_node(node
, memcg
->scan_nodes
);
1625 if (node
== MAX_NUMNODES
)
1626 node
= first_node(memcg
->scan_nodes
);
1628 * We call this when we hit limit, not when pages are added to LRU.
1629 * No LRU may hold pages because all pages are UNEVICTABLE or
1630 * memcg is too small and all pages are not on LRU. In that case,
1631 * we use curret node.
1633 if (unlikely(node
== MAX_NUMNODES
))
1634 node
= numa_node_id();
1636 memcg
->last_scanned_node
= node
;
1641 * Check all nodes whether it contains reclaimable pages or not.
1642 * For quick scan, we make use of scan_nodes. This will allow us to skip
1643 * unused nodes. But scan_nodes is lazily updated and may not cotain
1644 * enough new information. We need to do double check.
1646 static bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1651 * quick check...making use of scan_node.
1652 * We can skip unused nodes.
1654 if (!nodes_empty(memcg
->scan_nodes
)) {
1655 for (nid
= first_node(memcg
->scan_nodes
);
1657 nid
= next_node(nid
, memcg
->scan_nodes
)) {
1659 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1664 * Check rest of nodes.
1666 for_each_node_state(nid
, N_HIGH_MEMORY
) {
1667 if (node_isset(nid
, memcg
->scan_nodes
))
1669 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1676 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1681 static bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1683 return test_mem_cgroup_node_reclaimable(memcg
, 0, noswap
);
1687 static int mem_cgroup_soft_reclaim(struct mem_cgroup
*root_memcg
,
1690 unsigned long *total_scanned
)
1692 struct mem_cgroup
*victim
= NULL
;
1695 unsigned long excess
;
1696 unsigned long nr_scanned
;
1697 struct mem_cgroup_reclaim_cookie reclaim
= {
1702 excess
= res_counter_soft_limit_excess(&root_memcg
->res
) >> PAGE_SHIFT
;
1705 victim
= mem_cgroup_iter(root_memcg
, victim
, &reclaim
);
1710 * If we have not been able to reclaim
1711 * anything, it might because there are
1712 * no reclaimable pages under this hierarchy
1717 * We want to do more targeted reclaim.
1718 * excess >> 2 is not to excessive so as to
1719 * reclaim too much, nor too less that we keep
1720 * coming back to reclaim from this cgroup
1722 if (total
>= (excess
>> 2) ||
1723 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
))
1728 if (!mem_cgroup_reclaimable(victim
, false))
1730 total
+= mem_cgroup_shrink_node_zone(victim
, gfp_mask
, false,
1732 *total_scanned
+= nr_scanned
;
1733 if (!res_counter_soft_limit_excess(&root_memcg
->res
))
1736 mem_cgroup_iter_break(root_memcg
, victim
);
1741 * Check OOM-Killer is already running under our hierarchy.
1742 * If someone is running, return false.
1743 * Has to be called with memcg_oom_lock
1745 static bool mem_cgroup_oom_lock(struct mem_cgroup
*memcg
)
1747 struct mem_cgroup
*iter
, *failed
= NULL
;
1749 for_each_mem_cgroup_tree(iter
, memcg
) {
1750 if (iter
->oom_lock
) {
1752 * this subtree of our hierarchy is already locked
1753 * so we cannot give a lock.
1756 mem_cgroup_iter_break(memcg
, iter
);
1759 iter
->oom_lock
= true;
1766 * OK, we failed to lock the whole subtree so we have to clean up
1767 * what we set up to the failing subtree
1769 for_each_mem_cgroup_tree(iter
, memcg
) {
1770 if (iter
== failed
) {
1771 mem_cgroup_iter_break(memcg
, iter
);
1774 iter
->oom_lock
= false;
1780 * Has to be called with memcg_oom_lock
1782 static int mem_cgroup_oom_unlock(struct mem_cgroup
*memcg
)
1784 struct mem_cgroup
*iter
;
1786 for_each_mem_cgroup_tree(iter
, memcg
)
1787 iter
->oom_lock
= false;
1791 static void mem_cgroup_mark_under_oom(struct mem_cgroup
*memcg
)
1793 struct mem_cgroup
*iter
;
1795 for_each_mem_cgroup_tree(iter
, memcg
)
1796 atomic_inc(&iter
->under_oom
);
1799 static void mem_cgroup_unmark_under_oom(struct mem_cgroup
*memcg
)
1801 struct mem_cgroup
*iter
;
1804 * When a new child is created while the hierarchy is under oom,
1805 * mem_cgroup_oom_lock() may not be called. We have to use
1806 * atomic_add_unless() here.
1808 for_each_mem_cgroup_tree(iter
, memcg
)
1809 atomic_add_unless(&iter
->under_oom
, -1, 0);
1812 static DEFINE_SPINLOCK(memcg_oom_lock
);
1813 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1815 struct oom_wait_info
{
1816 struct mem_cgroup
*memcg
;
1820 static int memcg_oom_wake_function(wait_queue_t
*wait
,
1821 unsigned mode
, int sync
, void *arg
)
1823 struct mem_cgroup
*wake_memcg
= (struct mem_cgroup
*)arg
;
1824 struct mem_cgroup
*oom_wait_memcg
;
1825 struct oom_wait_info
*oom_wait_info
;
1827 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1828 oom_wait_memcg
= oom_wait_info
->memcg
;
1831 * Both of oom_wait_info->memcg and wake_memcg are stable under us.
1832 * Then we can use css_is_ancestor without taking care of RCU.
1834 if (!mem_cgroup_same_or_subtree(oom_wait_memcg
, wake_memcg
)
1835 && !mem_cgroup_same_or_subtree(wake_memcg
, oom_wait_memcg
))
1837 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1840 static void memcg_wakeup_oom(struct mem_cgroup
*memcg
)
1842 /* for filtering, pass "memcg" as argument. */
1843 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, memcg
);
1846 static void memcg_oom_recover(struct mem_cgroup
*memcg
)
1848 if (memcg
&& atomic_read(&memcg
->under_oom
))
1849 memcg_wakeup_oom(memcg
);
1853 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1855 static bool mem_cgroup_handle_oom(struct mem_cgroup
*memcg
, gfp_t mask
,
1858 struct oom_wait_info owait
;
1859 bool locked
, need_to_kill
;
1861 owait
.memcg
= memcg
;
1862 owait
.wait
.flags
= 0;
1863 owait
.wait
.func
= memcg_oom_wake_function
;
1864 owait
.wait
.private = current
;
1865 INIT_LIST_HEAD(&owait
.wait
.task_list
);
1866 need_to_kill
= true;
1867 mem_cgroup_mark_under_oom(memcg
);
1869 /* At first, try to OOM lock hierarchy under memcg.*/
1870 spin_lock(&memcg_oom_lock
);
1871 locked
= mem_cgroup_oom_lock(memcg
);
1873 * Even if signal_pending(), we can't quit charge() loop without
1874 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1875 * under OOM is always welcomed, use TASK_KILLABLE here.
1877 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1878 if (!locked
|| memcg
->oom_kill_disable
)
1879 need_to_kill
= false;
1881 mem_cgroup_oom_notify(memcg
);
1882 spin_unlock(&memcg_oom_lock
);
1885 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1886 mem_cgroup_out_of_memory(memcg
, mask
, order
);
1889 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1891 spin_lock(&memcg_oom_lock
);
1893 mem_cgroup_oom_unlock(memcg
);
1894 memcg_wakeup_oom(memcg
);
1895 spin_unlock(&memcg_oom_lock
);
1897 mem_cgroup_unmark_under_oom(memcg
);
1899 if (test_thread_flag(TIF_MEMDIE
) || fatal_signal_pending(current
))
1901 /* Give chance to dying process */
1902 schedule_timeout_uninterruptible(1);
1907 * Currently used to update mapped file statistics, but the routine can be
1908 * generalized to update other statistics as well.
1910 * Notes: Race condition
1912 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1913 * it tends to be costly. But considering some conditions, we doesn't need
1914 * to do so _always_.
1916 * Considering "charge", lock_page_cgroup() is not required because all
1917 * file-stat operations happen after a page is attached to radix-tree. There
1918 * are no race with "charge".
1920 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1921 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1922 * if there are race with "uncharge". Statistics itself is properly handled
1925 * Considering "move", this is an only case we see a race. To make the race
1926 * small, we check mm->moving_account and detect there are possibility of race
1927 * If there is, we take a lock.
1930 void __mem_cgroup_begin_update_page_stat(struct page
*page
,
1931 bool *locked
, unsigned long *flags
)
1933 struct mem_cgroup
*memcg
;
1934 struct page_cgroup
*pc
;
1936 pc
= lookup_page_cgroup(page
);
1938 memcg
= pc
->mem_cgroup
;
1939 if (unlikely(!memcg
|| !PageCgroupUsed(pc
)))
1942 * If this memory cgroup is not under account moving, we don't
1943 * need to take move_lock_page_cgroup(). Because we already hold
1944 * rcu_read_lock(), any calls to move_account will be delayed until
1945 * rcu_read_unlock() if mem_cgroup_stolen() == true.
1947 if (!mem_cgroup_stolen(memcg
))
1950 move_lock_mem_cgroup(memcg
, flags
);
1951 if (memcg
!= pc
->mem_cgroup
|| !PageCgroupUsed(pc
)) {
1952 move_unlock_mem_cgroup(memcg
, flags
);
1958 void __mem_cgroup_end_update_page_stat(struct page
*page
, unsigned long *flags
)
1960 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1963 * It's guaranteed that pc->mem_cgroup never changes while
1964 * lock is held because a routine modifies pc->mem_cgroup
1965 * should take move_lock_page_cgroup().
1967 move_unlock_mem_cgroup(pc
->mem_cgroup
, flags
);
1970 void mem_cgroup_update_page_stat(struct page
*page
,
1971 enum mem_cgroup_page_stat_item idx
, int val
)
1973 struct mem_cgroup
*memcg
;
1974 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1975 unsigned long uninitialized_var(flags
);
1977 if (mem_cgroup_disabled())
1980 memcg
= pc
->mem_cgroup
;
1981 if (unlikely(!memcg
|| !PageCgroupUsed(pc
)))
1985 case MEMCG_NR_FILE_MAPPED
:
1986 idx
= MEM_CGROUP_STAT_FILE_MAPPED
;
1992 this_cpu_add(memcg
->stat
->count
[idx
], val
);
1996 * size of first charge trial. "32" comes from vmscan.c's magic value.
1997 * TODO: maybe necessary to use big numbers in big irons.
1999 #define CHARGE_BATCH 32U
2000 struct memcg_stock_pcp
{
2001 struct mem_cgroup
*cached
; /* this never be root cgroup */
2002 unsigned int nr_pages
;
2003 struct work_struct work
;
2004 unsigned long flags
;
2005 #define FLUSHING_CACHED_CHARGE 0
2007 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
2008 static DEFINE_MUTEX(percpu_charge_mutex
);
2011 * Try to consume stocked charge on this cpu. If success, one page is consumed
2012 * from local stock and true is returned. If the stock is 0 or charges from a
2013 * cgroup which is not current target, returns false. This stock will be
2016 static bool consume_stock(struct mem_cgroup
*memcg
)
2018 struct memcg_stock_pcp
*stock
;
2021 stock
= &get_cpu_var(memcg_stock
);
2022 if (memcg
== stock
->cached
&& stock
->nr_pages
)
2024 else /* need to call res_counter_charge */
2026 put_cpu_var(memcg_stock
);
2031 * Returns stocks cached in percpu to res_counter and reset cached information.
2033 static void drain_stock(struct memcg_stock_pcp
*stock
)
2035 struct mem_cgroup
*old
= stock
->cached
;
2037 if (stock
->nr_pages
) {
2038 unsigned long bytes
= stock
->nr_pages
* PAGE_SIZE
;
2040 res_counter_uncharge(&old
->res
, bytes
);
2041 if (do_swap_account
)
2042 res_counter_uncharge(&old
->memsw
, bytes
);
2043 stock
->nr_pages
= 0;
2045 stock
->cached
= NULL
;
2049 * This must be called under preempt disabled or must be called by
2050 * a thread which is pinned to local cpu.
2052 static void drain_local_stock(struct work_struct
*dummy
)
2054 struct memcg_stock_pcp
*stock
= &__get_cpu_var(memcg_stock
);
2056 clear_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
);
2060 * Cache charges(val) which is from res_counter, to local per_cpu area.
2061 * This will be consumed by consume_stock() function, later.
2063 static void refill_stock(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
2065 struct memcg_stock_pcp
*stock
= &get_cpu_var(memcg_stock
);
2067 if (stock
->cached
!= memcg
) { /* reset if necessary */
2069 stock
->cached
= memcg
;
2071 stock
->nr_pages
+= nr_pages
;
2072 put_cpu_var(memcg_stock
);
2076 * Drains all per-CPU charge caches for given root_memcg resp. subtree
2077 * of the hierarchy under it. sync flag says whether we should block
2078 * until the work is done.
2080 static void drain_all_stock(struct mem_cgroup
*root_memcg
, bool sync
)
2084 /* Notify other cpus that system-wide "drain" is running */
2087 for_each_online_cpu(cpu
) {
2088 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2089 struct mem_cgroup
*memcg
;
2091 memcg
= stock
->cached
;
2092 if (!memcg
|| !stock
->nr_pages
)
2094 if (!mem_cgroup_same_or_subtree(root_memcg
, memcg
))
2096 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
)) {
2098 drain_local_stock(&stock
->work
);
2100 schedule_work_on(cpu
, &stock
->work
);
2108 for_each_online_cpu(cpu
) {
2109 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2110 if (test_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
))
2111 flush_work(&stock
->work
);
2118 * Tries to drain stocked charges in other cpus. This function is asynchronous
2119 * and just put a work per cpu for draining localy on each cpu. Caller can
2120 * expects some charges will be back to res_counter later but cannot wait for
2123 static void drain_all_stock_async(struct mem_cgroup
*root_memcg
)
2126 * If someone calls draining, avoid adding more kworker runs.
2128 if (!mutex_trylock(&percpu_charge_mutex
))
2130 drain_all_stock(root_memcg
, false);
2131 mutex_unlock(&percpu_charge_mutex
);
2134 /* This is a synchronous drain interface. */
2135 static void drain_all_stock_sync(struct mem_cgroup
*root_memcg
)
2137 /* called when force_empty is called */
2138 mutex_lock(&percpu_charge_mutex
);
2139 drain_all_stock(root_memcg
, true);
2140 mutex_unlock(&percpu_charge_mutex
);
2144 * This function drains percpu counter value from DEAD cpu and
2145 * move it to local cpu. Note that this function can be preempted.
2147 static void mem_cgroup_drain_pcp_counter(struct mem_cgroup
*memcg
, int cpu
)
2151 spin_lock(&memcg
->pcp_counter_lock
);
2152 for (i
= 0; i
< MEM_CGROUP_STAT_NSTATS
; i
++) {
2153 long x
= per_cpu(memcg
->stat
->count
[i
], cpu
);
2155 per_cpu(memcg
->stat
->count
[i
], cpu
) = 0;
2156 memcg
->nocpu_base
.count
[i
] += x
;
2158 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
2159 unsigned long x
= per_cpu(memcg
->stat
->events
[i
], cpu
);
2161 per_cpu(memcg
->stat
->events
[i
], cpu
) = 0;
2162 memcg
->nocpu_base
.events
[i
] += x
;
2164 spin_unlock(&memcg
->pcp_counter_lock
);
2167 static int __cpuinit
memcg_cpu_hotplug_callback(struct notifier_block
*nb
,
2168 unsigned long action
,
2171 int cpu
= (unsigned long)hcpu
;
2172 struct memcg_stock_pcp
*stock
;
2173 struct mem_cgroup
*iter
;
2175 if (action
== CPU_ONLINE
)
2178 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
2181 for_each_mem_cgroup(iter
)
2182 mem_cgroup_drain_pcp_counter(iter
, cpu
);
2184 stock
= &per_cpu(memcg_stock
, cpu
);
2190 /* See __mem_cgroup_try_charge() for details */
2192 CHARGE_OK
, /* success */
2193 CHARGE_RETRY
, /* need to retry but retry is not bad */
2194 CHARGE_NOMEM
, /* we can't do more. return -ENOMEM */
2195 CHARGE_WOULDBLOCK
, /* GFP_WAIT wasn't set and no enough res. */
2196 CHARGE_OOM_DIE
, /* the current is killed because of OOM */
2199 static int mem_cgroup_do_charge(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
2200 unsigned int nr_pages
, bool oom_check
)
2202 unsigned long csize
= nr_pages
* PAGE_SIZE
;
2203 struct mem_cgroup
*mem_over_limit
;
2204 struct res_counter
*fail_res
;
2205 unsigned long flags
= 0;
2208 ret
= res_counter_charge(&memcg
->res
, csize
, &fail_res
);
2211 if (!do_swap_account
)
2213 ret
= res_counter_charge(&memcg
->memsw
, csize
, &fail_res
);
2217 res_counter_uncharge(&memcg
->res
, csize
);
2218 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, memsw
);
2219 flags
|= MEM_CGROUP_RECLAIM_NOSWAP
;
2221 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, res
);
2223 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2224 * of regular pages (CHARGE_BATCH), or a single regular page (1).
2226 * Never reclaim on behalf of optional batching, retry with a
2227 * single page instead.
2229 if (nr_pages
== CHARGE_BATCH
)
2230 return CHARGE_RETRY
;
2232 if (!(gfp_mask
& __GFP_WAIT
))
2233 return CHARGE_WOULDBLOCK
;
2235 ret
= mem_cgroup_reclaim(mem_over_limit
, gfp_mask
, flags
);
2236 if (mem_cgroup_margin(mem_over_limit
) >= nr_pages
)
2237 return CHARGE_RETRY
;
2239 * Even though the limit is exceeded at this point, reclaim
2240 * may have been able to free some pages. Retry the charge
2241 * before killing the task.
2243 * Only for regular pages, though: huge pages are rather
2244 * unlikely to succeed so close to the limit, and we fall back
2245 * to regular pages anyway in case of failure.
2247 if (nr_pages
== 1 && ret
)
2248 return CHARGE_RETRY
;
2251 * At task move, charge accounts can be doubly counted. So, it's
2252 * better to wait until the end of task_move if something is going on.
2254 if (mem_cgroup_wait_acct_move(mem_over_limit
))
2255 return CHARGE_RETRY
;
2257 /* If we don't need to call oom-killer at el, return immediately */
2259 return CHARGE_NOMEM
;
2261 if (!mem_cgroup_handle_oom(mem_over_limit
, gfp_mask
, get_order(csize
)))
2262 return CHARGE_OOM_DIE
;
2264 return CHARGE_RETRY
;
2268 * __mem_cgroup_try_charge() does
2269 * 1. detect memcg to be charged against from passed *mm and *ptr,
2270 * 2. update res_counter
2271 * 3. call memory reclaim if necessary.
2273 * In some special case, if the task is fatal, fatal_signal_pending() or
2274 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2275 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2276 * as possible without any hazards. 2: all pages should have a valid
2277 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2278 * pointer, that is treated as a charge to root_mem_cgroup.
2280 * So __mem_cgroup_try_charge() will return
2281 * 0 ... on success, filling *ptr with a valid memcg pointer.
2282 * -ENOMEM ... charge failure because of resource limits.
2283 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2285 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2286 * the oom-killer can be invoked.
2288 static int __mem_cgroup_try_charge(struct mm_struct
*mm
,
2290 unsigned int nr_pages
,
2291 struct mem_cgroup
**ptr
,
2294 unsigned int batch
= max(CHARGE_BATCH
, nr_pages
);
2295 int nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2296 struct mem_cgroup
*memcg
= NULL
;
2300 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2301 * in system level. So, allow to go ahead dying process in addition to
2304 if (unlikely(test_thread_flag(TIF_MEMDIE
)
2305 || fatal_signal_pending(current
)))
2309 * We always charge the cgroup the mm_struct belongs to.
2310 * The mm_struct's mem_cgroup changes on task migration if the
2311 * thread group leader migrates. It's possible that mm is not
2312 * set, if so charge the init_mm (happens for pagecache usage).
2315 *ptr
= root_mem_cgroup
;
2317 if (*ptr
) { /* css should be a valid one */
2319 VM_BUG_ON(css_is_removed(&memcg
->css
));
2320 if (mem_cgroup_is_root(memcg
))
2322 if (nr_pages
== 1 && consume_stock(memcg
))
2324 css_get(&memcg
->css
);
2326 struct task_struct
*p
;
2329 p
= rcu_dereference(mm
->owner
);
2331 * Because we don't have task_lock(), "p" can exit.
2332 * In that case, "memcg" can point to root or p can be NULL with
2333 * race with swapoff. Then, we have small risk of mis-accouning.
2334 * But such kind of mis-account by race always happens because
2335 * we don't have cgroup_mutex(). It's overkill and we allo that
2337 * (*) swapoff at el will charge against mm-struct not against
2338 * task-struct. So, mm->owner can be NULL.
2340 memcg
= mem_cgroup_from_task(p
);
2342 memcg
= root_mem_cgroup
;
2343 if (mem_cgroup_is_root(memcg
)) {
2347 if (nr_pages
== 1 && consume_stock(memcg
)) {
2349 * It seems dagerous to access memcg without css_get().
2350 * But considering how consume_stok works, it's not
2351 * necessary. If consume_stock success, some charges
2352 * from this memcg are cached on this cpu. So, we
2353 * don't need to call css_get()/css_tryget() before
2354 * calling consume_stock().
2359 /* after here, we may be blocked. we need to get refcnt */
2360 if (!css_tryget(&memcg
->css
)) {
2370 /* If killed, bypass charge */
2371 if (fatal_signal_pending(current
)) {
2372 css_put(&memcg
->css
);
2377 if (oom
&& !nr_oom_retries
) {
2379 nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2382 ret
= mem_cgroup_do_charge(memcg
, gfp_mask
, batch
, oom_check
);
2386 case CHARGE_RETRY
: /* not in OOM situation but retry */
2388 css_put(&memcg
->css
);
2391 case CHARGE_WOULDBLOCK
: /* !__GFP_WAIT */
2392 css_put(&memcg
->css
);
2394 case CHARGE_NOMEM
: /* OOM routine works */
2396 css_put(&memcg
->css
);
2399 /* If oom, we never return -ENOMEM */
2402 case CHARGE_OOM_DIE
: /* Killed by OOM Killer */
2403 css_put(&memcg
->css
);
2406 } while (ret
!= CHARGE_OK
);
2408 if (batch
> nr_pages
)
2409 refill_stock(memcg
, batch
- nr_pages
);
2410 css_put(&memcg
->css
);
2418 *ptr
= root_mem_cgroup
;
2423 * Somemtimes we have to undo a charge we got by try_charge().
2424 * This function is for that and do uncharge, put css's refcnt.
2425 * gotten by try_charge().
2427 static void __mem_cgroup_cancel_charge(struct mem_cgroup
*memcg
,
2428 unsigned int nr_pages
)
2430 if (!mem_cgroup_is_root(memcg
)) {
2431 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2433 res_counter_uncharge(&memcg
->res
, bytes
);
2434 if (do_swap_account
)
2435 res_counter_uncharge(&memcg
->memsw
, bytes
);
2440 * Cancel chrages in this cgroup....doesn't propagate to parent cgroup.
2441 * This is useful when moving usage to parent cgroup.
2443 static void __mem_cgroup_cancel_local_charge(struct mem_cgroup
*memcg
,
2444 unsigned int nr_pages
)
2446 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2448 if (mem_cgroup_is_root(memcg
))
2451 res_counter_uncharge_until(&memcg
->res
, memcg
->res
.parent
, bytes
);
2452 if (do_swap_account
)
2453 res_counter_uncharge_until(&memcg
->memsw
,
2454 memcg
->memsw
.parent
, bytes
);
2458 * A helper function to get mem_cgroup from ID. must be called under
2459 * rcu_read_lock(). The caller must check css_is_removed() or some if
2460 * it's concern. (dropping refcnt from swap can be called against removed
2463 static struct mem_cgroup
*mem_cgroup_lookup(unsigned short id
)
2465 struct cgroup_subsys_state
*css
;
2467 /* ID 0 is unused ID */
2470 css
= css_lookup(&mem_cgroup_subsys
, id
);
2473 return container_of(css
, struct mem_cgroup
, css
);
2476 struct mem_cgroup
*try_get_mem_cgroup_from_page(struct page
*page
)
2478 struct mem_cgroup
*memcg
= NULL
;
2479 struct page_cgroup
*pc
;
2483 VM_BUG_ON(!PageLocked(page
));
2485 pc
= lookup_page_cgroup(page
);
2486 lock_page_cgroup(pc
);
2487 if (PageCgroupUsed(pc
)) {
2488 memcg
= pc
->mem_cgroup
;
2489 if (memcg
&& !css_tryget(&memcg
->css
))
2491 } else if (PageSwapCache(page
)) {
2492 ent
.val
= page_private(page
);
2493 id
= lookup_swap_cgroup_id(ent
);
2495 memcg
= mem_cgroup_lookup(id
);
2496 if (memcg
&& !css_tryget(&memcg
->css
))
2500 unlock_page_cgroup(pc
);
2504 static void __mem_cgroup_commit_charge(struct mem_cgroup
*memcg
,
2506 unsigned int nr_pages
,
2507 enum charge_type ctype
,
2510 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
2511 struct zone
*uninitialized_var(zone
);
2512 bool was_on_lru
= false;
2515 lock_page_cgroup(pc
);
2516 if (unlikely(PageCgroupUsed(pc
))) {
2517 unlock_page_cgroup(pc
);
2518 __mem_cgroup_cancel_charge(memcg
, nr_pages
);
2522 * we don't need page_cgroup_lock about tail pages, becase they are not
2523 * accessed by any other context at this point.
2527 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2528 * may already be on some other mem_cgroup's LRU. Take care of it.
2531 zone
= page_zone(page
);
2532 spin_lock_irq(&zone
->lru_lock
);
2533 if (PageLRU(page
)) {
2535 del_page_from_lru_list(zone
, page
, page_lru(page
));
2540 pc
->mem_cgroup
= memcg
;
2542 * We access a page_cgroup asynchronously without lock_page_cgroup().
2543 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2544 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2545 * before USED bit, we need memory barrier here.
2546 * See mem_cgroup_add_lru_list(), etc.
2549 SetPageCgroupUsed(pc
);
2553 VM_BUG_ON(PageLRU(page
));
2555 add_page_to_lru_list(zone
, page
, page_lru(page
));
2557 spin_unlock_irq(&zone
->lru_lock
);
2560 if (ctype
== MEM_CGROUP_CHARGE_TYPE_MAPPED
)
2565 mem_cgroup_charge_statistics(memcg
, anon
, nr_pages
);
2566 unlock_page_cgroup(pc
);
2569 * "charge_statistics" updated event counter. Then, check it.
2570 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2571 * if they exceeds softlimit.
2573 memcg_check_events(memcg
, page
);
2576 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2578 #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION)
2580 * Because tail pages are not marked as "used", set it. We're under
2581 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2582 * charge/uncharge will be never happen and move_account() is done under
2583 * compound_lock(), so we don't have to take care of races.
2585 void mem_cgroup_split_huge_fixup(struct page
*head
)
2587 struct page_cgroup
*head_pc
= lookup_page_cgroup(head
);
2588 struct page_cgroup
*pc
;
2591 if (mem_cgroup_disabled())
2593 for (i
= 1; i
< HPAGE_PMD_NR
; i
++) {
2595 pc
->mem_cgroup
= head_pc
->mem_cgroup
;
2596 smp_wmb();/* see __commit_charge() */
2597 pc
->flags
= head_pc
->flags
& ~PCGF_NOCOPY_AT_SPLIT
;
2600 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2603 * mem_cgroup_move_account - move account of the page
2605 * @nr_pages: number of regular pages (>1 for huge pages)
2606 * @pc: page_cgroup of the page.
2607 * @from: mem_cgroup which the page is moved from.
2608 * @to: mem_cgroup which the page is moved to. @from != @to.
2610 * The caller must confirm following.
2611 * - page is not on LRU (isolate_page() is useful.)
2612 * - compound_lock is held when nr_pages > 1
2614 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
2617 static int mem_cgroup_move_account(struct page
*page
,
2618 unsigned int nr_pages
,
2619 struct page_cgroup
*pc
,
2620 struct mem_cgroup
*from
,
2621 struct mem_cgroup
*to
)
2623 unsigned long flags
;
2625 bool anon
= PageAnon(page
);
2627 VM_BUG_ON(from
== to
);
2628 VM_BUG_ON(PageLRU(page
));
2630 * The page is isolated from LRU. So, collapse function
2631 * will not handle this page. But page splitting can happen.
2632 * Do this check under compound_page_lock(). The caller should
2636 if (nr_pages
> 1 && !PageTransHuge(page
))
2639 lock_page_cgroup(pc
);
2642 if (!PageCgroupUsed(pc
) || pc
->mem_cgroup
!= from
)
2645 move_lock_mem_cgroup(from
, &flags
);
2647 if (!anon
&& page_mapped(page
)) {
2648 /* Update mapped_file data for mem_cgroup */
2650 __this_cpu_dec(from
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2651 __this_cpu_inc(to
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2654 mem_cgroup_charge_statistics(from
, anon
, -nr_pages
);
2656 /* caller should have done css_get */
2657 pc
->mem_cgroup
= to
;
2658 mem_cgroup_charge_statistics(to
, anon
, nr_pages
);
2660 * We charges against "to" which may not have any tasks. Then, "to"
2661 * can be under rmdir(). But in current implementation, caller of
2662 * this function is just force_empty() and move charge, so it's
2663 * guaranteed that "to" is never removed. So, we don't check rmdir
2666 move_unlock_mem_cgroup(from
, &flags
);
2669 unlock_page_cgroup(pc
);
2673 memcg_check_events(to
, page
);
2674 memcg_check_events(from
, page
);
2680 * move charges to its parent.
2683 static int mem_cgroup_move_parent(struct page
*page
,
2684 struct page_cgroup
*pc
,
2685 struct mem_cgroup
*child
,
2688 struct mem_cgroup
*parent
;
2689 unsigned int nr_pages
;
2690 unsigned long uninitialized_var(flags
);
2694 if (mem_cgroup_is_root(child
))
2698 if (!get_page_unless_zero(page
))
2700 if (isolate_lru_page(page
))
2703 nr_pages
= hpage_nr_pages(page
);
2705 parent
= parent_mem_cgroup(child
);
2707 * If no parent, move charges to root cgroup.
2710 parent
= root_mem_cgroup
;
2713 flags
= compound_lock_irqsave(page
);
2715 ret
= mem_cgroup_move_account(page
, nr_pages
,
2718 __mem_cgroup_cancel_local_charge(child
, nr_pages
);
2721 compound_unlock_irqrestore(page
, flags
);
2722 putback_lru_page(page
);
2730 * Charge the memory controller for page usage.
2732 * 0 if the charge was successful
2733 * < 0 if the cgroup is over its limit
2735 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
2736 gfp_t gfp_mask
, enum charge_type ctype
)
2738 struct mem_cgroup
*memcg
= NULL
;
2739 unsigned int nr_pages
= 1;
2743 if (PageTransHuge(page
)) {
2744 nr_pages
<<= compound_order(page
);
2745 VM_BUG_ON(!PageTransHuge(page
));
2747 * Never OOM-kill a process for a huge page. The
2748 * fault handler will fall back to regular pages.
2753 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, nr_pages
, &memcg
, oom
);
2756 __mem_cgroup_commit_charge(memcg
, page
, nr_pages
, ctype
, false);
2760 int mem_cgroup_newpage_charge(struct page
*page
,
2761 struct mm_struct
*mm
, gfp_t gfp_mask
)
2763 if (mem_cgroup_disabled())
2765 VM_BUG_ON(page_mapped(page
));
2766 VM_BUG_ON(page
->mapping
&& !PageAnon(page
));
2768 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2769 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2773 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2774 enum charge_type ctype
);
2776 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
2779 struct mem_cgroup
*memcg
= NULL
;
2780 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
2783 if (mem_cgroup_disabled())
2785 if (PageCompound(page
))
2790 if (!page_is_file_cache(page
))
2791 type
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
2793 if (!PageSwapCache(page
))
2794 ret
= mem_cgroup_charge_common(page
, mm
, gfp_mask
, type
);
2795 else { /* page is swapcache/shmem */
2796 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &memcg
);
2798 __mem_cgroup_commit_charge_swapin(page
, memcg
, type
);
2804 * While swap-in, try_charge -> commit or cancel, the page is locked.
2805 * And when try_charge() successfully returns, one refcnt to memcg without
2806 * struct page_cgroup is acquired. This refcnt will be consumed by
2807 * "commit()" or removed by "cancel()"
2809 int mem_cgroup_try_charge_swapin(struct mm_struct
*mm
,
2811 gfp_t mask
, struct mem_cgroup
**memcgp
)
2813 struct mem_cgroup
*memcg
;
2818 if (mem_cgroup_disabled())
2821 if (!do_swap_account
)
2824 * A racing thread's fault, or swapoff, may have already updated
2825 * the pte, and even removed page from swap cache: in those cases
2826 * do_swap_page()'s pte_same() test will fail; but there's also a
2827 * KSM case which does need to charge the page.
2829 if (!PageSwapCache(page
))
2831 memcg
= try_get_mem_cgroup_from_page(page
);
2835 ret
= __mem_cgroup_try_charge(NULL
, mask
, 1, memcgp
, true);
2836 css_put(&memcg
->css
);
2843 ret
= __mem_cgroup_try_charge(mm
, mask
, 1, memcgp
, true);
2850 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*memcg
,
2851 enum charge_type ctype
)
2853 if (mem_cgroup_disabled())
2857 cgroup_exclude_rmdir(&memcg
->css
);
2859 __mem_cgroup_commit_charge(memcg
, page
, 1, ctype
, true);
2861 * Now swap is on-memory. This means this page may be
2862 * counted both as mem and swap....double count.
2863 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2864 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2865 * may call delete_from_swap_cache() before reach here.
2867 if (do_swap_account
&& PageSwapCache(page
)) {
2868 swp_entry_t ent
= {.val
= page_private(page
)};
2869 mem_cgroup_uncharge_swap(ent
);
2872 * At swapin, we may charge account against cgroup which has no tasks.
2873 * So, rmdir()->pre_destroy() can be called while we do this charge.
2874 * In that case, we need to call pre_destroy() again. check it here.
2876 cgroup_release_and_wakeup_rmdir(&memcg
->css
);
2879 void mem_cgroup_commit_charge_swapin(struct page
*page
,
2880 struct mem_cgroup
*memcg
)
2882 __mem_cgroup_commit_charge_swapin(page
, memcg
,
2883 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2886 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup
*memcg
)
2888 if (mem_cgroup_disabled())
2892 __mem_cgroup_cancel_charge(memcg
, 1);
2895 static void mem_cgroup_do_uncharge(struct mem_cgroup
*memcg
,
2896 unsigned int nr_pages
,
2897 const enum charge_type ctype
)
2899 struct memcg_batch_info
*batch
= NULL
;
2900 bool uncharge_memsw
= true;
2902 /* If swapout, usage of swap doesn't decrease */
2903 if (!do_swap_account
|| ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
)
2904 uncharge_memsw
= false;
2906 batch
= ¤t
->memcg_batch
;
2908 * In usual, we do css_get() when we remember memcg pointer.
2909 * But in this case, we keep res->usage until end of a series of
2910 * uncharges. Then, it's ok to ignore memcg's refcnt.
2913 batch
->memcg
= memcg
;
2915 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2916 * In those cases, all pages freed continuously can be expected to be in
2917 * the same cgroup and we have chance to coalesce uncharges.
2918 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2919 * because we want to do uncharge as soon as possible.
2922 if (!batch
->do_batch
|| test_thread_flag(TIF_MEMDIE
))
2923 goto direct_uncharge
;
2926 goto direct_uncharge
;
2929 * In typical case, batch->memcg == mem. This means we can
2930 * merge a series of uncharges to an uncharge of res_counter.
2931 * If not, we uncharge res_counter ony by one.
2933 if (batch
->memcg
!= memcg
)
2934 goto direct_uncharge
;
2935 /* remember freed charge and uncharge it later */
2938 batch
->memsw_nr_pages
++;
2941 res_counter_uncharge(&memcg
->res
, nr_pages
* PAGE_SIZE
);
2943 res_counter_uncharge(&memcg
->memsw
, nr_pages
* PAGE_SIZE
);
2944 if (unlikely(batch
->memcg
!= memcg
))
2945 memcg_oom_recover(memcg
);
2949 * uncharge if !page_mapped(page)
2951 static struct mem_cgroup
*
2952 __mem_cgroup_uncharge_common(struct page
*page
, enum charge_type ctype
)
2954 struct mem_cgroup
*memcg
= NULL
;
2955 unsigned int nr_pages
= 1;
2956 struct page_cgroup
*pc
;
2959 if (mem_cgroup_disabled())
2962 if (PageSwapCache(page
))
2965 if (PageTransHuge(page
)) {
2966 nr_pages
<<= compound_order(page
);
2967 VM_BUG_ON(!PageTransHuge(page
));
2970 * Check if our page_cgroup is valid
2972 pc
= lookup_page_cgroup(page
);
2973 if (unlikely(!PageCgroupUsed(pc
)))
2976 lock_page_cgroup(pc
);
2978 memcg
= pc
->mem_cgroup
;
2980 if (!PageCgroupUsed(pc
))
2983 anon
= PageAnon(page
);
2986 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2988 * Generally PageAnon tells if it's the anon statistics to be
2989 * updated; but sometimes e.g. mem_cgroup_uncharge_page() is
2990 * used before page reached the stage of being marked PageAnon.
2994 case MEM_CGROUP_CHARGE_TYPE_DROP
:
2995 /* See mem_cgroup_prepare_migration() */
2996 if (page_mapped(page
) || PageCgroupMigration(pc
))
2999 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT
:
3000 if (!PageAnon(page
)) { /* Shared memory */
3001 if (page
->mapping
&& !page_is_file_cache(page
))
3003 } else if (page_mapped(page
)) /* Anon */
3010 mem_cgroup_charge_statistics(memcg
, anon
, -nr_pages
);
3012 ClearPageCgroupUsed(pc
);
3014 * pc->mem_cgroup is not cleared here. It will be accessed when it's
3015 * freed from LRU. This is safe because uncharged page is expected not
3016 * to be reused (freed soon). Exception is SwapCache, it's handled by
3017 * special functions.
3020 unlock_page_cgroup(pc
);
3022 * even after unlock, we have memcg->res.usage here and this memcg
3023 * will never be freed.
3025 memcg_check_events(memcg
, page
);
3026 if (do_swap_account
&& ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
) {
3027 mem_cgroup_swap_statistics(memcg
, true);
3028 mem_cgroup_get(memcg
);
3030 if (!mem_cgroup_is_root(memcg
))
3031 mem_cgroup_do_uncharge(memcg
, nr_pages
, ctype
);
3036 unlock_page_cgroup(pc
);
3040 void mem_cgroup_uncharge_page(struct page
*page
)
3043 if (page_mapped(page
))
3045 VM_BUG_ON(page
->mapping
&& !PageAnon(page
));
3046 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_MAPPED
);
3049 void mem_cgroup_uncharge_cache_page(struct page
*page
)
3051 VM_BUG_ON(page_mapped(page
));
3052 VM_BUG_ON(page
->mapping
);
3053 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_CACHE
);
3057 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
3058 * In that cases, pages are freed continuously and we can expect pages
3059 * are in the same memcg. All these calls itself limits the number of
3060 * pages freed at once, then uncharge_start/end() is called properly.
3061 * This may be called prural(2) times in a context,
3064 void mem_cgroup_uncharge_start(void)
3066 current
->memcg_batch
.do_batch
++;
3067 /* We can do nest. */
3068 if (current
->memcg_batch
.do_batch
== 1) {
3069 current
->memcg_batch
.memcg
= NULL
;
3070 current
->memcg_batch
.nr_pages
= 0;
3071 current
->memcg_batch
.memsw_nr_pages
= 0;
3075 void mem_cgroup_uncharge_end(void)
3077 struct memcg_batch_info
*batch
= ¤t
->memcg_batch
;
3079 if (!batch
->do_batch
)
3083 if (batch
->do_batch
) /* If stacked, do nothing. */
3089 * This "batch->memcg" is valid without any css_get/put etc...
3090 * bacause we hide charges behind us.
3092 if (batch
->nr_pages
)
3093 res_counter_uncharge(&batch
->memcg
->res
,
3094 batch
->nr_pages
* PAGE_SIZE
);
3095 if (batch
->memsw_nr_pages
)
3096 res_counter_uncharge(&batch
->memcg
->memsw
,
3097 batch
->memsw_nr_pages
* PAGE_SIZE
);
3098 memcg_oom_recover(batch
->memcg
);
3099 /* forget this pointer (for sanity check) */
3100 batch
->memcg
= NULL
;
3105 * called after __delete_from_swap_cache() and drop "page" account.
3106 * memcg information is recorded to swap_cgroup of "ent"
3109 mem_cgroup_uncharge_swapcache(struct page
*page
, swp_entry_t ent
, bool swapout
)
3111 struct mem_cgroup
*memcg
;
3112 int ctype
= MEM_CGROUP_CHARGE_TYPE_SWAPOUT
;
3114 if (!swapout
) /* this was a swap cache but the swap is unused ! */
3115 ctype
= MEM_CGROUP_CHARGE_TYPE_DROP
;
3117 memcg
= __mem_cgroup_uncharge_common(page
, ctype
);
3120 * record memcg information, if swapout && memcg != NULL,
3121 * mem_cgroup_get() was called in uncharge().
3123 if (do_swap_account
&& swapout
&& memcg
)
3124 swap_cgroup_record(ent
, css_id(&memcg
->css
));
3128 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3130 * called from swap_entry_free(). remove record in swap_cgroup and
3131 * uncharge "memsw" account.
3133 void mem_cgroup_uncharge_swap(swp_entry_t ent
)
3135 struct mem_cgroup
*memcg
;
3138 if (!do_swap_account
)
3141 id
= swap_cgroup_record(ent
, 0);
3143 memcg
= mem_cgroup_lookup(id
);
3146 * We uncharge this because swap is freed.
3147 * This memcg can be obsolete one. We avoid calling css_tryget
3149 if (!mem_cgroup_is_root(memcg
))
3150 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
3151 mem_cgroup_swap_statistics(memcg
, false);
3152 mem_cgroup_put(memcg
);
3158 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3159 * @entry: swap entry to be moved
3160 * @from: mem_cgroup which the entry is moved from
3161 * @to: mem_cgroup which the entry is moved to
3163 * It succeeds only when the swap_cgroup's record for this entry is the same
3164 * as the mem_cgroup's id of @from.
3166 * Returns 0 on success, -EINVAL on failure.
3168 * The caller must have charged to @to, IOW, called res_counter_charge() about
3169 * both res and memsw, and called css_get().
3171 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
3172 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
3174 unsigned short old_id
, new_id
;
3176 old_id
= css_id(&from
->css
);
3177 new_id
= css_id(&to
->css
);
3179 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
3180 mem_cgroup_swap_statistics(from
, false);
3181 mem_cgroup_swap_statistics(to
, true);
3183 * This function is only called from task migration context now.
3184 * It postpones res_counter and refcount handling till the end
3185 * of task migration(mem_cgroup_clear_mc()) for performance
3186 * improvement. But we cannot postpone mem_cgroup_get(to)
3187 * because if the process that has been moved to @to does
3188 * swap-in, the refcount of @to might be decreased to 0.
3196 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
3197 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
3204 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3207 int mem_cgroup_prepare_migration(struct page
*page
,
3208 struct page
*newpage
, struct mem_cgroup
**memcgp
, gfp_t gfp_mask
)
3210 struct mem_cgroup
*memcg
= NULL
;
3211 struct page_cgroup
*pc
;
3212 enum charge_type ctype
;
3217 VM_BUG_ON(PageTransHuge(page
));
3218 if (mem_cgroup_disabled())
3221 pc
= lookup_page_cgroup(page
);
3222 lock_page_cgroup(pc
);
3223 if (PageCgroupUsed(pc
)) {
3224 memcg
= pc
->mem_cgroup
;
3225 css_get(&memcg
->css
);
3227 * At migrating an anonymous page, its mapcount goes down
3228 * to 0 and uncharge() will be called. But, even if it's fully
3229 * unmapped, migration may fail and this page has to be
3230 * charged again. We set MIGRATION flag here and delay uncharge
3231 * until end_migration() is called
3233 * Corner Case Thinking
3235 * When the old page was mapped as Anon and it's unmap-and-freed
3236 * while migration was ongoing.
3237 * If unmap finds the old page, uncharge() of it will be delayed
3238 * until end_migration(). If unmap finds a new page, it's
3239 * uncharged when it make mapcount to be 1->0. If unmap code
3240 * finds swap_migration_entry, the new page will not be mapped
3241 * and end_migration() will find it(mapcount==0).
3244 * When the old page was mapped but migraion fails, the kernel
3245 * remaps it. A charge for it is kept by MIGRATION flag even
3246 * if mapcount goes down to 0. We can do remap successfully
3247 * without charging it again.
3250 * The "old" page is under lock_page() until the end of
3251 * migration, so, the old page itself will not be swapped-out.
3252 * If the new page is swapped out before end_migraton, our
3253 * hook to usual swap-out path will catch the event.
3256 SetPageCgroupMigration(pc
);
3258 unlock_page_cgroup(pc
);
3260 * If the page is not charged at this point,
3267 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, 1, memcgp
, false);
3268 css_put(&memcg
->css
);/* drop extra refcnt */
3270 if (PageAnon(page
)) {
3271 lock_page_cgroup(pc
);
3272 ClearPageCgroupMigration(pc
);
3273 unlock_page_cgroup(pc
);
3275 * The old page may be fully unmapped while we kept it.
3277 mem_cgroup_uncharge_page(page
);
3279 /* we'll need to revisit this error code (we have -EINTR) */
3283 * We charge new page before it's used/mapped. So, even if unlock_page()
3284 * is called before end_migration, we can catch all events on this new
3285 * page. In the case new page is migrated but not remapped, new page's
3286 * mapcount will be finally 0 and we call uncharge in end_migration().
3289 ctype
= MEM_CGROUP_CHARGE_TYPE_MAPPED
;
3290 else if (page_is_file_cache(page
))
3291 ctype
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3293 ctype
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
3294 __mem_cgroup_commit_charge(memcg
, newpage
, 1, ctype
, false);
3298 /* remove redundant charge if migration failed*/
3299 void mem_cgroup_end_migration(struct mem_cgroup
*memcg
,
3300 struct page
*oldpage
, struct page
*newpage
, bool migration_ok
)
3302 struct page
*used
, *unused
;
3303 struct page_cgroup
*pc
;
3308 /* blocks rmdir() */
3309 cgroup_exclude_rmdir(&memcg
->css
);
3310 if (!migration_ok
) {
3318 * We disallowed uncharge of pages under migration because mapcount
3319 * of the page goes down to zero, temporarly.
3320 * Clear the flag and check the page should be charged.
3322 pc
= lookup_page_cgroup(oldpage
);
3323 lock_page_cgroup(pc
);
3324 ClearPageCgroupMigration(pc
);
3325 unlock_page_cgroup(pc
);
3326 anon
= PageAnon(used
);
3327 __mem_cgroup_uncharge_common(unused
,
3328 anon
? MEM_CGROUP_CHARGE_TYPE_MAPPED
3329 : MEM_CGROUP_CHARGE_TYPE_CACHE
);
3332 * If a page is a file cache, radix-tree replacement is very atomic
3333 * and we can skip this check. When it was an Anon page, its mapcount
3334 * goes down to 0. But because we added MIGRATION flage, it's not
3335 * uncharged yet. There are several case but page->mapcount check
3336 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3337 * check. (see prepare_charge() also)
3340 mem_cgroup_uncharge_page(used
);
3342 * At migration, we may charge account against cgroup which has no
3344 * So, rmdir()->pre_destroy() can be called while we do this charge.
3345 * In that case, we need to call pre_destroy() again. check it here.
3347 cgroup_release_and_wakeup_rmdir(&memcg
->css
);
3351 * At replace page cache, newpage is not under any memcg but it's on
3352 * LRU. So, this function doesn't touch res_counter but handles LRU
3353 * in correct way. Both pages are locked so we cannot race with uncharge.
3355 void mem_cgroup_replace_page_cache(struct page
*oldpage
,
3356 struct page
*newpage
)
3358 struct mem_cgroup
*memcg
= NULL
;
3359 struct page_cgroup
*pc
;
3360 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3362 if (mem_cgroup_disabled())
3365 pc
= lookup_page_cgroup(oldpage
);
3366 /* fix accounting on old pages */
3367 lock_page_cgroup(pc
);
3368 if (PageCgroupUsed(pc
)) {
3369 memcg
= pc
->mem_cgroup
;
3370 mem_cgroup_charge_statistics(memcg
, false, -1);
3371 ClearPageCgroupUsed(pc
);
3373 unlock_page_cgroup(pc
);
3376 * When called from shmem_replace_page(), in some cases the
3377 * oldpage has already been charged, and in some cases not.
3382 if (PageSwapBacked(oldpage
))
3383 type
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
3386 * Even if newpage->mapping was NULL before starting replacement,
3387 * the newpage may be on LRU(or pagevec for LRU) already. We lock
3388 * LRU while we overwrite pc->mem_cgroup.
3390 __mem_cgroup_commit_charge(memcg
, newpage
, 1, type
, true);
3393 #ifdef CONFIG_DEBUG_VM
3394 static struct page_cgroup
*lookup_page_cgroup_used(struct page
*page
)
3396 struct page_cgroup
*pc
;
3398 pc
= lookup_page_cgroup(page
);
3400 * Can be NULL while feeding pages into the page allocator for
3401 * the first time, i.e. during boot or memory hotplug;
3402 * or when mem_cgroup_disabled().
3404 if (likely(pc
) && PageCgroupUsed(pc
))
3409 bool mem_cgroup_bad_page_check(struct page
*page
)
3411 if (mem_cgroup_disabled())
3414 return lookup_page_cgroup_used(page
) != NULL
;
3417 void mem_cgroup_print_bad_page(struct page
*page
)
3419 struct page_cgroup
*pc
;
3421 pc
= lookup_page_cgroup_used(page
);
3423 printk(KERN_ALERT
"pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
3424 pc
, pc
->flags
, pc
->mem_cgroup
);
3429 static DEFINE_MUTEX(set_limit_mutex
);
3431 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
3432 unsigned long long val
)
3435 u64 memswlimit
, memlimit
;
3437 int children
= mem_cgroup_count_children(memcg
);
3438 u64 curusage
, oldusage
;
3442 * For keeping hierarchical_reclaim simple, how long we should retry
3443 * is depends on callers. We set our retry-count to be function
3444 * of # of children which we should visit in this loop.
3446 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
* children
;
3448 oldusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3451 while (retry_count
) {
3452 if (signal_pending(current
)) {
3457 * Rather than hide all in some function, I do this in
3458 * open coded manner. You see what this really does.
3459 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
3461 mutex_lock(&set_limit_mutex
);
3462 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3463 if (memswlimit
< val
) {
3465 mutex_unlock(&set_limit_mutex
);
3469 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3473 ret
= res_counter_set_limit(&memcg
->res
, val
);
3475 if (memswlimit
== val
)
3476 memcg
->memsw_is_minimum
= true;
3478 memcg
->memsw_is_minimum
= false;
3480 mutex_unlock(&set_limit_mutex
);
3485 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3486 MEM_CGROUP_RECLAIM_SHRINK
);
3487 curusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3488 /* Usage is reduced ? */
3489 if (curusage
>= oldusage
)
3492 oldusage
= curusage
;
3494 if (!ret
&& enlarge
)
3495 memcg_oom_recover(memcg
);
3500 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
3501 unsigned long long val
)
3504 u64 memlimit
, memswlimit
, oldusage
, curusage
;
3505 int children
= mem_cgroup_count_children(memcg
);
3509 /* see mem_cgroup_resize_res_limit */
3510 retry_count
= children
* MEM_CGROUP_RECLAIM_RETRIES
;
3511 oldusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3512 while (retry_count
) {
3513 if (signal_pending(current
)) {
3518 * Rather than hide all in some function, I do this in
3519 * open coded manner. You see what this really does.
3520 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
3522 mutex_lock(&set_limit_mutex
);
3523 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3524 if (memlimit
> val
) {
3526 mutex_unlock(&set_limit_mutex
);
3529 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3530 if (memswlimit
< val
)
3532 ret
= res_counter_set_limit(&memcg
->memsw
, val
);
3534 if (memlimit
== val
)
3535 memcg
->memsw_is_minimum
= true;
3537 memcg
->memsw_is_minimum
= false;
3539 mutex_unlock(&set_limit_mutex
);
3544 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3545 MEM_CGROUP_RECLAIM_NOSWAP
|
3546 MEM_CGROUP_RECLAIM_SHRINK
);
3547 curusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3548 /* Usage is reduced ? */
3549 if (curusage
>= oldusage
)
3552 oldusage
= curusage
;
3554 if (!ret
&& enlarge
)
3555 memcg_oom_recover(memcg
);
3559 unsigned long mem_cgroup_soft_limit_reclaim(struct zone
*zone
, int order
,
3561 unsigned long *total_scanned
)
3563 unsigned long nr_reclaimed
= 0;
3564 struct mem_cgroup_per_zone
*mz
, *next_mz
= NULL
;
3565 unsigned long reclaimed
;
3567 struct mem_cgroup_tree_per_zone
*mctz
;
3568 unsigned long long excess
;
3569 unsigned long nr_scanned
;
3574 mctz
= soft_limit_tree_node_zone(zone_to_nid(zone
), zone_idx(zone
));
3576 * This loop can run a while, specially if mem_cgroup's continuously
3577 * keep exceeding their soft limit and putting the system under
3584 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
3589 reclaimed
= mem_cgroup_soft_reclaim(mz
->memcg
, zone
,
3590 gfp_mask
, &nr_scanned
);
3591 nr_reclaimed
+= reclaimed
;
3592 *total_scanned
+= nr_scanned
;
3593 spin_lock(&mctz
->lock
);
3596 * If we failed to reclaim anything from this memory cgroup
3597 * it is time to move on to the next cgroup
3603 * Loop until we find yet another one.
3605 * By the time we get the soft_limit lock
3606 * again, someone might have aded the
3607 * group back on the RB tree. Iterate to
3608 * make sure we get a different mem.
3609 * mem_cgroup_largest_soft_limit_node returns
3610 * NULL if no other cgroup is present on
3614 __mem_cgroup_largest_soft_limit_node(mctz
);
3616 css_put(&next_mz
->memcg
->css
);
3617 else /* next_mz == NULL or other memcg */
3621 __mem_cgroup_remove_exceeded(mz
->memcg
, mz
, mctz
);
3622 excess
= res_counter_soft_limit_excess(&mz
->memcg
->res
);
3624 * One school of thought says that we should not add
3625 * back the node to the tree if reclaim returns 0.
3626 * But our reclaim could return 0, simply because due
3627 * to priority we are exposing a smaller subset of
3628 * memory to reclaim from. Consider this as a longer
3631 /* If excess == 0, no tree ops */
3632 __mem_cgroup_insert_exceeded(mz
->memcg
, mz
, mctz
, excess
);
3633 spin_unlock(&mctz
->lock
);
3634 css_put(&mz
->memcg
->css
);
3637 * Could not reclaim anything and there are no more
3638 * mem cgroups to try or we seem to be looping without
3639 * reclaiming anything.
3641 if (!nr_reclaimed
&&
3643 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
3645 } while (!nr_reclaimed
);
3647 css_put(&next_mz
->memcg
->css
);
3648 return nr_reclaimed
;
3652 * This routine traverse page_cgroup in given list and drop them all.
3653 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3655 static int mem_cgroup_force_empty_list(struct mem_cgroup
*memcg
,
3656 int node
, int zid
, enum lru_list lru
)
3658 struct mem_cgroup_per_zone
*mz
;
3659 unsigned long flags
, loop
;
3660 struct list_head
*list
;
3665 zone
= &NODE_DATA(node
)->node_zones
[zid
];
3666 mz
= mem_cgroup_zoneinfo(memcg
, node
, zid
);
3667 list
= &mz
->lruvec
.lists
[lru
];
3669 loop
= mz
->lru_size
[lru
];
3670 /* give some margin against EBUSY etc...*/
3674 struct page_cgroup
*pc
;
3678 spin_lock_irqsave(&zone
->lru_lock
, flags
);
3679 if (list_empty(list
)) {
3680 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3683 page
= list_entry(list
->prev
, struct page
, lru
);
3685 list_move(&page
->lru
, list
);
3687 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3690 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3692 pc
= lookup_page_cgroup(page
);
3694 ret
= mem_cgroup_move_parent(page
, pc
, memcg
, GFP_KERNEL
);
3695 if (ret
== -ENOMEM
|| ret
== -EINTR
)
3698 if (ret
== -EBUSY
|| ret
== -EINVAL
) {
3699 /* found lock contention or "pc" is obsolete. */
3706 if (!ret
&& !list_empty(list
))
3712 * make mem_cgroup's charge to be 0 if there is no task.
3713 * This enables deleting this mem_cgroup.
3715 static int mem_cgroup_force_empty(struct mem_cgroup
*memcg
, bool free_all
)
3718 int node
, zid
, shrink
;
3719 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
3720 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
3722 css_get(&memcg
->css
);
3725 /* should free all ? */
3731 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
))
3734 if (signal_pending(current
))
3736 /* This is for making all *used* pages to be on LRU. */
3737 lru_add_drain_all();
3738 drain_all_stock_sync(memcg
);
3740 mem_cgroup_start_move(memcg
);
3741 for_each_node_state(node
, N_HIGH_MEMORY
) {
3742 for (zid
= 0; !ret
&& zid
< MAX_NR_ZONES
; zid
++) {
3745 ret
= mem_cgroup_force_empty_list(memcg
,
3754 mem_cgroup_end_move(memcg
);
3755 memcg_oom_recover(memcg
);
3756 /* it seems parent cgroup doesn't have enough mem */
3760 /* "ret" should also be checked to ensure all lists are empty. */
3761 } while (res_counter_read_u64(&memcg
->res
, RES_USAGE
) > 0 || ret
);
3763 css_put(&memcg
->css
);
3767 /* returns EBUSY if there is a task or if we come here twice. */
3768 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
) || shrink
) {
3772 /* we call try-to-free pages for make this cgroup empty */
3773 lru_add_drain_all();
3774 /* try to free all pages in this cgroup */
3776 while (nr_retries
&& res_counter_read_u64(&memcg
->res
, RES_USAGE
) > 0) {
3779 if (signal_pending(current
)) {
3783 progress
= try_to_free_mem_cgroup_pages(memcg
, GFP_KERNEL
,
3787 /* maybe some writeback is necessary */
3788 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3793 /* try move_account...there may be some *locked* pages. */
3797 static int mem_cgroup_force_empty_write(struct cgroup
*cont
, unsigned int event
)
3799 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont
), true);
3803 static u64
mem_cgroup_hierarchy_read(struct cgroup
*cont
, struct cftype
*cft
)
3805 return mem_cgroup_from_cont(cont
)->use_hierarchy
;
3808 static int mem_cgroup_hierarchy_write(struct cgroup
*cont
, struct cftype
*cft
,
3812 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3813 struct cgroup
*parent
= cont
->parent
;
3814 struct mem_cgroup
*parent_memcg
= NULL
;
3817 parent_memcg
= mem_cgroup_from_cont(parent
);
3821 * If parent's use_hierarchy is set, we can't make any modifications
3822 * in the child subtrees. If it is unset, then the change can
3823 * occur, provided the current cgroup has no children.
3825 * For the root cgroup, parent_mem is NULL, we allow value to be
3826 * set if there are no children.
3828 if ((!parent_memcg
|| !parent_memcg
->use_hierarchy
) &&
3829 (val
== 1 || val
== 0)) {
3830 if (list_empty(&cont
->children
))
3831 memcg
->use_hierarchy
= val
;
3842 static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup
*memcg
,
3843 enum mem_cgroup_stat_index idx
)
3845 struct mem_cgroup
*iter
;
3848 /* Per-cpu values can be negative, use a signed accumulator */
3849 for_each_mem_cgroup_tree(iter
, memcg
)
3850 val
+= mem_cgroup_read_stat(iter
, idx
);
3852 if (val
< 0) /* race ? */
3857 static inline u64
mem_cgroup_usage(struct mem_cgroup
*memcg
, bool swap
)
3861 if (!mem_cgroup_is_root(memcg
)) {
3863 return res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3865 return res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3868 val
= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_CACHE
);
3869 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_RSS
);
3872 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_SWAPOUT
);
3874 return val
<< PAGE_SHIFT
;
3877 static ssize_t
mem_cgroup_read(struct cgroup
*cont
, struct cftype
*cft
,
3878 struct file
*file
, char __user
*buf
,
3879 size_t nbytes
, loff_t
*ppos
)
3881 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3884 int type
, name
, len
;
3886 type
= MEMFILE_TYPE(cft
->private);
3887 name
= MEMFILE_ATTR(cft
->private);
3889 if (!do_swap_account
&& type
== _MEMSWAP
)
3894 if (name
== RES_USAGE
)
3895 val
= mem_cgroup_usage(memcg
, false);
3897 val
= res_counter_read_u64(&memcg
->res
, name
);
3900 if (name
== RES_USAGE
)
3901 val
= mem_cgroup_usage(memcg
, true);
3903 val
= res_counter_read_u64(&memcg
->memsw
, name
);
3909 len
= scnprintf(str
, sizeof(str
), "%llu\n", (unsigned long long)val
);
3910 return simple_read_from_buffer(buf
, nbytes
, ppos
, str
, len
);
3913 * The user of this function is...
3916 static int mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
3919 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3921 unsigned long long val
;
3924 type
= MEMFILE_TYPE(cft
->private);
3925 name
= MEMFILE_ATTR(cft
->private);
3927 if (!do_swap_account
&& type
== _MEMSWAP
)
3932 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
3936 /* This function does all necessary parse...reuse it */
3937 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3941 ret
= mem_cgroup_resize_limit(memcg
, val
);
3943 ret
= mem_cgroup_resize_memsw_limit(memcg
, val
);
3945 case RES_SOFT_LIMIT
:
3946 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3950 * For memsw, soft limits are hard to implement in terms
3951 * of semantics, for now, we support soft limits for
3952 * control without swap
3955 ret
= res_counter_set_soft_limit(&memcg
->res
, val
);
3960 ret
= -EINVAL
; /* should be BUG() ? */
3966 static void memcg_get_hierarchical_limit(struct mem_cgroup
*memcg
,
3967 unsigned long long *mem_limit
, unsigned long long *memsw_limit
)
3969 struct cgroup
*cgroup
;
3970 unsigned long long min_limit
, min_memsw_limit
, tmp
;
3972 min_limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3973 min_memsw_limit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3974 cgroup
= memcg
->css
.cgroup
;
3975 if (!memcg
->use_hierarchy
)
3978 while (cgroup
->parent
) {
3979 cgroup
= cgroup
->parent
;
3980 memcg
= mem_cgroup_from_cont(cgroup
);
3981 if (!memcg
->use_hierarchy
)
3983 tmp
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3984 min_limit
= min(min_limit
, tmp
);
3985 tmp
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3986 min_memsw_limit
= min(min_memsw_limit
, tmp
);
3989 *mem_limit
= min_limit
;
3990 *memsw_limit
= min_memsw_limit
;
3993 static int mem_cgroup_reset(struct cgroup
*cont
, unsigned int event
)
3995 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3998 type
= MEMFILE_TYPE(event
);
3999 name
= MEMFILE_ATTR(event
);
4001 if (!do_swap_account
&& type
== _MEMSWAP
)
4007 res_counter_reset_max(&memcg
->res
);
4009 res_counter_reset_max(&memcg
->memsw
);
4013 res_counter_reset_failcnt(&memcg
->res
);
4015 res_counter_reset_failcnt(&memcg
->memsw
);
4022 static u64
mem_cgroup_move_charge_read(struct cgroup
*cgrp
,
4025 return mem_cgroup_from_cont(cgrp
)->move_charge_at_immigrate
;
4029 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4030 struct cftype
*cft
, u64 val
)
4032 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4034 if (val
>= (1 << NR_MOVE_TYPE
))
4037 * We check this value several times in both in can_attach() and
4038 * attach(), so we need cgroup lock to prevent this value from being
4042 memcg
->move_charge_at_immigrate
= val
;
4048 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4049 struct cftype
*cft
, u64 val
)
4056 static int mem_control_numa_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
4060 unsigned long total_nr
, file_nr
, anon_nr
, unevictable_nr
;
4061 unsigned long node_nr
;
4062 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4064 total_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL
);
4065 seq_printf(m
, "total=%lu", total_nr
);
4066 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4067 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL
);
4068 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4072 file_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL_FILE
);
4073 seq_printf(m
, "file=%lu", file_nr
);
4074 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4075 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4077 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4081 anon_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL_ANON
);
4082 seq_printf(m
, "anon=%lu", anon_nr
);
4083 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4084 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4086 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4090 unevictable_nr
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_UNEVICTABLE
));
4091 seq_printf(m
, "unevictable=%lu", unevictable_nr
);
4092 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4093 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4094 BIT(LRU_UNEVICTABLE
));
4095 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4100 #endif /* CONFIG_NUMA */
4102 static const char * const mem_cgroup_lru_names
[] = {
4110 static inline void mem_cgroup_lru_names_not_uptodate(void)
4112 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names
) != NR_LRU_LISTS
);
4115 static int mem_control_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
4118 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4119 struct mem_cgroup
*mi
;
4122 for (i
= 0; i
< MEM_CGROUP_STAT_NSTATS
; i
++) {
4123 if (i
== MEM_CGROUP_STAT_SWAPOUT
&& !do_swap_account
)
4125 seq_printf(m
, "%s %ld\n", mem_cgroup_stat_names
[i
],
4126 mem_cgroup_read_stat(memcg
, i
) * PAGE_SIZE
);
4129 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++)
4130 seq_printf(m
, "%s %lu\n", mem_cgroup_events_names
[i
],
4131 mem_cgroup_read_events(memcg
, i
));
4133 for (i
= 0; i
< NR_LRU_LISTS
; i
++)
4134 seq_printf(m
, "%s %lu\n", mem_cgroup_lru_names
[i
],
4135 mem_cgroup_nr_lru_pages(memcg
, BIT(i
)) * PAGE_SIZE
);
4137 /* Hierarchical information */
4139 unsigned long long limit
, memsw_limit
;
4140 memcg_get_hierarchical_limit(memcg
, &limit
, &memsw_limit
);
4141 seq_printf(m
, "hierarchical_memory_limit %llu\n", limit
);
4142 if (do_swap_account
)
4143 seq_printf(m
, "hierarchical_memsw_limit %llu\n",
4147 for (i
= 0; i
< MEM_CGROUP_STAT_NSTATS
; i
++) {
4150 if (i
== MEM_CGROUP_STAT_SWAPOUT
&& !do_swap_account
)
4152 for_each_mem_cgroup_tree(mi
, memcg
)
4153 val
+= mem_cgroup_read_stat(mi
, i
) * PAGE_SIZE
;
4154 seq_printf(m
, "total_%s %lld\n", mem_cgroup_stat_names
[i
], val
);
4157 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
4158 unsigned long long val
= 0;
4160 for_each_mem_cgroup_tree(mi
, memcg
)
4161 val
+= mem_cgroup_read_events(mi
, i
);
4162 seq_printf(m
, "total_%s %llu\n",
4163 mem_cgroup_events_names
[i
], val
);
4166 for (i
= 0; i
< NR_LRU_LISTS
; i
++) {
4167 unsigned long long val
= 0;
4169 for_each_mem_cgroup_tree(mi
, memcg
)
4170 val
+= mem_cgroup_nr_lru_pages(mi
, BIT(i
)) * PAGE_SIZE
;
4171 seq_printf(m
, "total_%s %llu\n", mem_cgroup_lru_names
[i
], val
);
4174 #ifdef CONFIG_DEBUG_VM
4177 struct mem_cgroup_per_zone
*mz
;
4178 struct zone_reclaim_stat
*rstat
;
4179 unsigned long recent_rotated
[2] = {0, 0};
4180 unsigned long recent_scanned
[2] = {0, 0};
4182 for_each_online_node(nid
)
4183 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
4184 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
4185 rstat
= &mz
->lruvec
.reclaim_stat
;
4187 recent_rotated
[0] += rstat
->recent_rotated
[0];
4188 recent_rotated
[1] += rstat
->recent_rotated
[1];
4189 recent_scanned
[0] += rstat
->recent_scanned
[0];
4190 recent_scanned
[1] += rstat
->recent_scanned
[1];
4192 seq_printf(m
, "recent_rotated_anon %lu\n", recent_rotated
[0]);
4193 seq_printf(m
, "recent_rotated_file %lu\n", recent_rotated
[1]);
4194 seq_printf(m
, "recent_scanned_anon %lu\n", recent_scanned
[0]);
4195 seq_printf(m
, "recent_scanned_file %lu\n", recent_scanned
[1]);
4202 static u64
mem_cgroup_swappiness_read(struct cgroup
*cgrp
, struct cftype
*cft
)
4204 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4206 return mem_cgroup_swappiness(memcg
);
4209 static int mem_cgroup_swappiness_write(struct cgroup
*cgrp
, struct cftype
*cft
,
4212 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4213 struct mem_cgroup
*parent
;
4218 if (cgrp
->parent
== NULL
)
4221 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4225 /* If under hierarchy, only empty-root can set this value */
4226 if ((parent
->use_hierarchy
) ||
4227 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4232 memcg
->swappiness
= val
;
4239 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
4241 struct mem_cgroup_threshold_ary
*t
;
4247 t
= rcu_dereference(memcg
->thresholds
.primary
);
4249 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
4254 usage
= mem_cgroup_usage(memcg
, swap
);
4257 * current_threshold points to threshold just below or equal to usage.
4258 * If it's not true, a threshold was crossed after last
4259 * call of __mem_cgroup_threshold().
4261 i
= t
->current_threshold
;
4264 * Iterate backward over array of thresholds starting from
4265 * current_threshold and check if a threshold is crossed.
4266 * If none of thresholds below usage is crossed, we read
4267 * only one element of the array here.
4269 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
4270 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4272 /* i = current_threshold + 1 */
4276 * Iterate forward over array of thresholds starting from
4277 * current_threshold+1 and check if a threshold is crossed.
4278 * If none of thresholds above usage is crossed, we read
4279 * only one element of the array here.
4281 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
4282 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4284 /* Update current_threshold */
4285 t
->current_threshold
= i
- 1;
4290 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
4293 __mem_cgroup_threshold(memcg
, false);
4294 if (do_swap_account
)
4295 __mem_cgroup_threshold(memcg
, true);
4297 memcg
= parent_mem_cgroup(memcg
);
4301 static int compare_thresholds(const void *a
, const void *b
)
4303 const struct mem_cgroup_threshold
*_a
= a
;
4304 const struct mem_cgroup_threshold
*_b
= b
;
4306 return _a
->threshold
- _b
->threshold
;
4309 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*memcg
)
4311 struct mem_cgroup_eventfd_list
*ev
;
4313 list_for_each_entry(ev
, &memcg
->oom_notify
, list
)
4314 eventfd_signal(ev
->eventfd
, 1);
4318 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
)
4320 struct mem_cgroup
*iter
;
4322 for_each_mem_cgroup_tree(iter
, memcg
)
4323 mem_cgroup_oom_notify_cb(iter
);
4326 static int mem_cgroup_usage_register_event(struct cgroup
*cgrp
,
4327 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4329 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4330 struct mem_cgroup_thresholds
*thresholds
;
4331 struct mem_cgroup_threshold_ary
*new;
4332 int type
= MEMFILE_TYPE(cft
->private);
4333 u64 threshold
, usage
;
4336 ret
= res_counter_memparse_write_strategy(args
, &threshold
);
4340 mutex_lock(&memcg
->thresholds_lock
);
4343 thresholds
= &memcg
->thresholds
;
4344 else if (type
== _MEMSWAP
)
4345 thresholds
= &memcg
->memsw_thresholds
;
4349 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4351 /* Check if a threshold crossed before adding a new one */
4352 if (thresholds
->primary
)
4353 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4355 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
4357 /* Allocate memory for new array of thresholds */
4358 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
4366 /* Copy thresholds (if any) to new array */
4367 if (thresholds
->primary
) {
4368 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
4369 sizeof(struct mem_cgroup_threshold
));
4372 /* Add new threshold */
4373 new->entries
[size
- 1].eventfd
= eventfd
;
4374 new->entries
[size
- 1].threshold
= threshold
;
4376 /* Sort thresholds. Registering of new threshold isn't time-critical */
4377 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
4378 compare_thresholds
, NULL
);
4380 /* Find current threshold */
4381 new->current_threshold
= -1;
4382 for (i
= 0; i
< size
; i
++) {
4383 if (new->entries
[i
].threshold
<= usage
) {
4385 * new->current_threshold will not be used until
4386 * rcu_assign_pointer(), so it's safe to increment
4389 ++new->current_threshold
;
4394 /* Free old spare buffer and save old primary buffer as spare */
4395 kfree(thresholds
->spare
);
4396 thresholds
->spare
= thresholds
->primary
;
4398 rcu_assign_pointer(thresholds
->primary
, new);
4400 /* To be sure that nobody uses thresholds */
4404 mutex_unlock(&memcg
->thresholds_lock
);
4409 static void mem_cgroup_usage_unregister_event(struct cgroup
*cgrp
,
4410 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4412 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4413 struct mem_cgroup_thresholds
*thresholds
;
4414 struct mem_cgroup_threshold_ary
*new;
4415 int type
= MEMFILE_TYPE(cft
->private);
4419 mutex_lock(&memcg
->thresholds_lock
);
4421 thresholds
= &memcg
->thresholds
;
4422 else if (type
== _MEMSWAP
)
4423 thresholds
= &memcg
->memsw_thresholds
;
4427 if (!thresholds
->primary
)
4430 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4432 /* Check if a threshold crossed before removing */
4433 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4435 /* Calculate new number of threshold */
4437 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
4438 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
4442 new = thresholds
->spare
;
4444 /* Set thresholds array to NULL if we don't have thresholds */
4453 /* Copy thresholds and find current threshold */
4454 new->current_threshold
= -1;
4455 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
4456 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
4459 new->entries
[j
] = thresholds
->primary
->entries
[i
];
4460 if (new->entries
[j
].threshold
<= usage
) {
4462 * new->current_threshold will not be used
4463 * until rcu_assign_pointer(), so it's safe to increment
4466 ++new->current_threshold
;
4472 /* Swap primary and spare array */
4473 thresholds
->spare
= thresholds
->primary
;
4474 /* If all events are unregistered, free the spare array */
4476 kfree(thresholds
->spare
);
4477 thresholds
->spare
= NULL
;
4480 rcu_assign_pointer(thresholds
->primary
, new);
4482 /* To be sure that nobody uses thresholds */
4485 mutex_unlock(&memcg
->thresholds_lock
);
4488 static int mem_cgroup_oom_register_event(struct cgroup
*cgrp
,
4489 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4491 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4492 struct mem_cgroup_eventfd_list
*event
;
4493 int type
= MEMFILE_TYPE(cft
->private);
4495 BUG_ON(type
!= _OOM_TYPE
);
4496 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
4500 spin_lock(&memcg_oom_lock
);
4502 event
->eventfd
= eventfd
;
4503 list_add(&event
->list
, &memcg
->oom_notify
);
4505 /* already in OOM ? */
4506 if (atomic_read(&memcg
->under_oom
))
4507 eventfd_signal(eventfd
, 1);
4508 spin_unlock(&memcg_oom_lock
);
4513 static void mem_cgroup_oom_unregister_event(struct cgroup
*cgrp
,
4514 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4516 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4517 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
4518 int type
= MEMFILE_TYPE(cft
->private);
4520 BUG_ON(type
!= _OOM_TYPE
);
4522 spin_lock(&memcg_oom_lock
);
4524 list_for_each_entry_safe(ev
, tmp
, &memcg
->oom_notify
, list
) {
4525 if (ev
->eventfd
== eventfd
) {
4526 list_del(&ev
->list
);
4531 spin_unlock(&memcg_oom_lock
);
4534 static int mem_cgroup_oom_control_read(struct cgroup
*cgrp
,
4535 struct cftype
*cft
, struct cgroup_map_cb
*cb
)
4537 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4539 cb
->fill(cb
, "oom_kill_disable", memcg
->oom_kill_disable
);
4541 if (atomic_read(&memcg
->under_oom
))
4542 cb
->fill(cb
, "under_oom", 1);
4544 cb
->fill(cb
, "under_oom", 0);
4548 static int mem_cgroup_oom_control_write(struct cgroup
*cgrp
,
4549 struct cftype
*cft
, u64 val
)
4551 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4552 struct mem_cgroup
*parent
;
4554 /* cannot set to root cgroup and only 0 and 1 are allowed */
4555 if (!cgrp
->parent
|| !((val
== 0) || (val
== 1)))
4558 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4561 /* oom-kill-disable is a flag for subhierarchy. */
4562 if ((parent
->use_hierarchy
) ||
4563 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4567 memcg
->oom_kill_disable
= val
;
4569 memcg_oom_recover(memcg
);
4574 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
4575 static int memcg_init_kmem(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
4577 return mem_cgroup_sockets_init(memcg
, ss
);
4580 static void kmem_cgroup_destroy(struct mem_cgroup
*memcg
)
4582 mem_cgroup_sockets_destroy(memcg
);
4585 static int memcg_init_kmem(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
4590 static void kmem_cgroup_destroy(struct mem_cgroup
*memcg
)
4595 static struct cftype mem_cgroup_files
[] = {
4597 .name
= "usage_in_bytes",
4598 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
4599 .read
= mem_cgroup_read
,
4600 .register_event
= mem_cgroup_usage_register_event
,
4601 .unregister_event
= mem_cgroup_usage_unregister_event
,
4604 .name
= "max_usage_in_bytes",
4605 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
4606 .trigger
= mem_cgroup_reset
,
4607 .read
= mem_cgroup_read
,
4610 .name
= "limit_in_bytes",
4611 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
4612 .write_string
= mem_cgroup_write
,
4613 .read
= mem_cgroup_read
,
4616 .name
= "soft_limit_in_bytes",
4617 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
4618 .write_string
= mem_cgroup_write
,
4619 .read
= mem_cgroup_read
,
4623 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
4624 .trigger
= mem_cgroup_reset
,
4625 .read
= mem_cgroup_read
,
4629 .read_seq_string
= mem_control_stat_show
,
4632 .name
= "force_empty",
4633 .trigger
= mem_cgroup_force_empty_write
,
4636 .name
= "use_hierarchy",
4637 .write_u64
= mem_cgroup_hierarchy_write
,
4638 .read_u64
= mem_cgroup_hierarchy_read
,
4641 .name
= "swappiness",
4642 .read_u64
= mem_cgroup_swappiness_read
,
4643 .write_u64
= mem_cgroup_swappiness_write
,
4646 .name
= "move_charge_at_immigrate",
4647 .read_u64
= mem_cgroup_move_charge_read
,
4648 .write_u64
= mem_cgroup_move_charge_write
,
4651 .name
= "oom_control",
4652 .read_map
= mem_cgroup_oom_control_read
,
4653 .write_u64
= mem_cgroup_oom_control_write
,
4654 .register_event
= mem_cgroup_oom_register_event
,
4655 .unregister_event
= mem_cgroup_oom_unregister_event
,
4656 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
4660 .name
= "numa_stat",
4661 .read_seq_string
= mem_control_numa_stat_show
,
4664 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4666 .name
= "memsw.usage_in_bytes",
4667 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
4668 .read
= mem_cgroup_read
,
4669 .register_event
= mem_cgroup_usage_register_event
,
4670 .unregister_event
= mem_cgroup_usage_unregister_event
,
4673 .name
= "memsw.max_usage_in_bytes",
4674 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
4675 .trigger
= mem_cgroup_reset
,
4676 .read
= mem_cgroup_read
,
4679 .name
= "memsw.limit_in_bytes",
4680 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
4681 .write_string
= mem_cgroup_write
,
4682 .read
= mem_cgroup_read
,
4685 .name
= "memsw.failcnt",
4686 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
4687 .trigger
= mem_cgroup_reset
,
4688 .read
= mem_cgroup_read
,
4691 { }, /* terminate */
4694 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4696 struct mem_cgroup_per_node
*pn
;
4697 struct mem_cgroup_per_zone
*mz
;
4698 int zone
, tmp
= node
;
4700 * This routine is called against possible nodes.
4701 * But it's BUG to call kmalloc() against offline node.
4703 * TODO: this routine can waste much memory for nodes which will
4704 * never be onlined. It's better to use memory hotplug callback
4707 if (!node_state(node
, N_NORMAL_MEMORY
))
4709 pn
= kzalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
4713 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4714 mz
= &pn
->zoneinfo
[zone
];
4715 lruvec_init(&mz
->lruvec
, &NODE_DATA(node
)->node_zones
[zone
]);
4716 mz
->usage_in_excess
= 0;
4717 mz
->on_tree
= false;
4720 memcg
->info
.nodeinfo
[node
] = pn
;
4724 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4726 kfree(memcg
->info
.nodeinfo
[node
]);
4729 static struct mem_cgroup
*mem_cgroup_alloc(void)
4731 struct mem_cgroup
*memcg
;
4732 int size
= sizeof(struct mem_cgroup
);
4734 /* Can be very big if MAX_NUMNODES is very big */
4735 if (size
< PAGE_SIZE
)
4736 memcg
= kzalloc(size
, GFP_KERNEL
);
4738 memcg
= vzalloc(size
);
4743 memcg
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4746 spin_lock_init(&memcg
->pcp_counter_lock
);
4750 if (size
< PAGE_SIZE
)
4758 * Helpers for freeing a vzalloc()ed mem_cgroup by RCU,
4759 * but in process context. The work_freeing structure is overlaid
4760 * on the rcu_freeing structure, which itself is overlaid on memsw.
4762 static void vfree_work(struct work_struct
*work
)
4764 struct mem_cgroup
*memcg
;
4766 memcg
= container_of(work
, struct mem_cgroup
, work_freeing
);
4769 static void vfree_rcu(struct rcu_head
*rcu_head
)
4771 struct mem_cgroup
*memcg
;
4773 memcg
= container_of(rcu_head
, struct mem_cgroup
, rcu_freeing
);
4774 INIT_WORK(&memcg
->work_freeing
, vfree_work
);
4775 schedule_work(&memcg
->work_freeing
);
4779 * At destroying mem_cgroup, references from swap_cgroup can remain.
4780 * (scanning all at force_empty is too costly...)
4782 * Instead of clearing all references at force_empty, we remember
4783 * the number of reference from swap_cgroup and free mem_cgroup when
4784 * it goes down to 0.
4786 * Removal of cgroup itself succeeds regardless of refs from swap.
4789 static void __mem_cgroup_free(struct mem_cgroup
*memcg
)
4793 mem_cgroup_remove_from_trees(memcg
);
4794 free_css_id(&mem_cgroup_subsys
, &memcg
->css
);
4797 free_mem_cgroup_per_zone_info(memcg
, node
);
4799 free_percpu(memcg
->stat
);
4800 if (sizeof(struct mem_cgroup
) < PAGE_SIZE
)
4801 kfree_rcu(memcg
, rcu_freeing
);
4803 call_rcu(&memcg
->rcu_freeing
, vfree_rcu
);
4806 static void mem_cgroup_get(struct mem_cgroup
*memcg
)
4808 atomic_inc(&memcg
->refcnt
);
4811 static void __mem_cgroup_put(struct mem_cgroup
*memcg
, int count
)
4813 if (atomic_sub_and_test(count
, &memcg
->refcnt
)) {
4814 struct mem_cgroup
*parent
= parent_mem_cgroup(memcg
);
4815 __mem_cgroup_free(memcg
);
4817 mem_cgroup_put(parent
);
4821 static void mem_cgroup_put(struct mem_cgroup
*memcg
)
4823 __mem_cgroup_put(memcg
, 1);
4827 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4829 struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*memcg
)
4831 if (!memcg
->res
.parent
)
4833 return mem_cgroup_from_res_counter(memcg
->res
.parent
, res
);
4835 EXPORT_SYMBOL(parent_mem_cgroup
);
4837 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4838 static void __init
enable_swap_cgroup(void)
4840 if (!mem_cgroup_disabled() && really_do_swap_account
)
4841 do_swap_account
= 1;
4844 static void __init
enable_swap_cgroup(void)
4849 static int mem_cgroup_soft_limit_tree_init(void)
4851 struct mem_cgroup_tree_per_node
*rtpn
;
4852 struct mem_cgroup_tree_per_zone
*rtpz
;
4853 int tmp
, node
, zone
;
4855 for_each_node(node
) {
4857 if (!node_state(node
, N_NORMAL_MEMORY
))
4859 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
, tmp
);
4863 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
4865 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4866 rtpz
= &rtpn
->rb_tree_per_zone
[zone
];
4867 rtpz
->rb_root
= RB_ROOT
;
4868 spin_lock_init(&rtpz
->lock
);
4874 for_each_node(node
) {
4875 if (!soft_limit_tree
.rb_tree_per_node
[node
])
4877 kfree(soft_limit_tree
.rb_tree_per_node
[node
]);
4878 soft_limit_tree
.rb_tree_per_node
[node
] = NULL
;
4884 static struct cgroup_subsys_state
* __ref
4885 mem_cgroup_create(struct cgroup
*cont
)
4887 struct mem_cgroup
*memcg
, *parent
;
4888 long error
= -ENOMEM
;
4891 memcg
= mem_cgroup_alloc();
4893 return ERR_PTR(error
);
4896 if (alloc_mem_cgroup_per_zone_info(memcg
, node
))
4900 if (cont
->parent
== NULL
) {
4902 enable_swap_cgroup();
4904 if (mem_cgroup_soft_limit_tree_init())
4906 root_mem_cgroup
= memcg
;
4907 for_each_possible_cpu(cpu
) {
4908 struct memcg_stock_pcp
*stock
=
4909 &per_cpu(memcg_stock
, cpu
);
4910 INIT_WORK(&stock
->work
, drain_local_stock
);
4912 hotcpu_notifier(memcg_cpu_hotplug_callback
, 0);
4914 parent
= mem_cgroup_from_cont(cont
->parent
);
4915 memcg
->use_hierarchy
= parent
->use_hierarchy
;
4916 memcg
->oom_kill_disable
= parent
->oom_kill_disable
;
4919 if (parent
&& parent
->use_hierarchy
) {
4920 res_counter_init(&memcg
->res
, &parent
->res
);
4921 res_counter_init(&memcg
->memsw
, &parent
->memsw
);
4923 * We increment refcnt of the parent to ensure that we can
4924 * safely access it on res_counter_charge/uncharge.
4925 * This refcnt will be decremented when freeing this
4926 * mem_cgroup(see mem_cgroup_put).
4928 mem_cgroup_get(parent
);
4930 res_counter_init(&memcg
->res
, NULL
);
4931 res_counter_init(&memcg
->memsw
, NULL
);
4933 memcg
->last_scanned_node
= MAX_NUMNODES
;
4934 INIT_LIST_HEAD(&memcg
->oom_notify
);
4937 memcg
->swappiness
= mem_cgroup_swappiness(parent
);
4938 atomic_set(&memcg
->refcnt
, 1);
4939 memcg
->move_charge_at_immigrate
= 0;
4940 mutex_init(&memcg
->thresholds_lock
);
4941 spin_lock_init(&memcg
->move_lock
);
4943 error
= memcg_init_kmem(memcg
, &mem_cgroup_subsys
);
4946 * We call put now because our (and parent's) refcnts
4947 * are already in place. mem_cgroup_put() will internally
4948 * call __mem_cgroup_free, so return directly
4950 mem_cgroup_put(memcg
);
4951 return ERR_PTR(error
);
4955 __mem_cgroup_free(memcg
);
4956 return ERR_PTR(error
);
4959 static int mem_cgroup_pre_destroy(struct cgroup
*cont
)
4961 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4963 return mem_cgroup_force_empty(memcg
, false);
4966 static void mem_cgroup_destroy(struct cgroup
*cont
)
4968 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4970 kmem_cgroup_destroy(memcg
);
4972 mem_cgroup_put(memcg
);
4976 /* Handlers for move charge at task migration. */
4977 #define PRECHARGE_COUNT_AT_ONCE 256
4978 static int mem_cgroup_do_precharge(unsigned long count
)
4981 int batch_count
= PRECHARGE_COUNT_AT_ONCE
;
4982 struct mem_cgroup
*memcg
= mc
.to
;
4984 if (mem_cgroup_is_root(memcg
)) {
4985 mc
.precharge
+= count
;
4986 /* we don't need css_get for root */
4989 /* try to charge at once */
4991 struct res_counter
*dummy
;
4993 * "memcg" cannot be under rmdir() because we've already checked
4994 * by cgroup_lock_live_cgroup() that it is not removed and we
4995 * are still under the same cgroup_mutex. So we can postpone
4998 if (res_counter_charge(&memcg
->res
, PAGE_SIZE
* count
, &dummy
))
5000 if (do_swap_account
&& res_counter_charge(&memcg
->memsw
,
5001 PAGE_SIZE
* count
, &dummy
)) {
5002 res_counter_uncharge(&memcg
->res
, PAGE_SIZE
* count
);
5005 mc
.precharge
+= count
;
5009 /* fall back to one by one charge */
5011 if (signal_pending(current
)) {
5015 if (!batch_count
--) {
5016 batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5019 ret
= __mem_cgroup_try_charge(NULL
,
5020 GFP_KERNEL
, 1, &memcg
, false);
5022 /* mem_cgroup_clear_mc() will do uncharge later */
5030 * get_mctgt_type - get target type of moving charge
5031 * @vma: the vma the pte to be checked belongs
5032 * @addr: the address corresponding to the pte to be checked
5033 * @ptent: the pte to be checked
5034 * @target: the pointer the target page or swap ent will be stored(can be NULL)
5037 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5038 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5039 * move charge. if @target is not NULL, the page is stored in target->page
5040 * with extra refcnt got(Callers should handle it).
5041 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5042 * target for charge migration. if @target is not NULL, the entry is stored
5045 * Called with pte lock held.
5052 enum mc_target_type
{
5058 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
5059 unsigned long addr
, pte_t ptent
)
5061 struct page
*page
= vm_normal_page(vma
, addr
, ptent
);
5063 if (!page
|| !page_mapped(page
))
5065 if (PageAnon(page
)) {
5066 /* we don't move shared anon */
5069 } else if (!move_file())
5070 /* we ignore mapcount for file pages */
5072 if (!get_page_unless_zero(page
))
5079 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5080 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5082 struct page
*page
= NULL
;
5083 swp_entry_t ent
= pte_to_swp_entry(ptent
);
5085 if (!move_anon() || non_swap_entry(ent
))
5088 * Because lookup_swap_cache() updates some statistics counter,
5089 * we call find_get_page() with swapper_space directly.
5091 page
= find_get_page(&swapper_space
, ent
.val
);
5092 if (do_swap_account
)
5093 entry
->val
= ent
.val
;
5098 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5099 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5105 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
5106 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5108 struct page
*page
= NULL
;
5109 struct address_space
*mapping
;
5112 if (!vma
->vm_file
) /* anonymous vma */
5117 mapping
= vma
->vm_file
->f_mapping
;
5118 if (pte_none(ptent
))
5119 pgoff
= linear_page_index(vma
, addr
);
5120 else /* pte_file(ptent) is true */
5121 pgoff
= pte_to_pgoff(ptent
);
5123 /* page is moved even if it's not RSS of this task(page-faulted). */
5124 page
= find_get_page(mapping
, pgoff
);
5127 /* shmem/tmpfs may report page out on swap: account for that too. */
5128 if (radix_tree_exceptional_entry(page
)) {
5129 swp_entry_t swap
= radix_to_swp_entry(page
);
5130 if (do_swap_account
)
5132 page
= find_get_page(&swapper_space
, swap
.val
);
5138 static enum mc_target_type
get_mctgt_type(struct vm_area_struct
*vma
,
5139 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
5141 struct page
*page
= NULL
;
5142 struct page_cgroup
*pc
;
5143 enum mc_target_type ret
= MC_TARGET_NONE
;
5144 swp_entry_t ent
= { .val
= 0 };
5146 if (pte_present(ptent
))
5147 page
= mc_handle_present_pte(vma
, addr
, ptent
);
5148 else if (is_swap_pte(ptent
))
5149 page
= mc_handle_swap_pte(vma
, addr
, ptent
, &ent
);
5150 else if (pte_none(ptent
) || pte_file(ptent
))
5151 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
5153 if (!page
&& !ent
.val
)
5156 pc
= lookup_page_cgroup(page
);
5158 * Do only loose check w/o page_cgroup lock.
5159 * mem_cgroup_move_account() checks the pc is valid or not under
5162 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5163 ret
= MC_TARGET_PAGE
;
5165 target
->page
= page
;
5167 if (!ret
|| !target
)
5170 /* There is a swap entry and a page doesn't exist or isn't charged */
5171 if (ent
.val
&& !ret
&&
5172 css_id(&mc
.from
->css
) == lookup_swap_cgroup_id(ent
)) {
5173 ret
= MC_TARGET_SWAP
;
5180 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5182 * We don't consider swapping or file mapped pages because THP does not
5183 * support them for now.
5184 * Caller should make sure that pmd_trans_huge(pmd) is true.
5186 static enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
5187 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
5189 struct page
*page
= NULL
;
5190 struct page_cgroup
*pc
;
5191 enum mc_target_type ret
= MC_TARGET_NONE
;
5193 page
= pmd_page(pmd
);
5194 VM_BUG_ON(!page
|| !PageHead(page
));
5197 pc
= lookup_page_cgroup(page
);
5198 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5199 ret
= MC_TARGET_PAGE
;
5202 target
->page
= page
;
5208 static inline enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
5209 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
5211 return MC_TARGET_NONE
;
5215 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
5216 unsigned long addr
, unsigned long end
,
5217 struct mm_walk
*walk
)
5219 struct vm_area_struct
*vma
= walk
->private;
5223 if (pmd_trans_huge_lock(pmd
, vma
) == 1) {
5224 if (get_mctgt_type_thp(vma
, addr
, *pmd
, NULL
) == MC_TARGET_PAGE
)
5225 mc
.precharge
+= HPAGE_PMD_NR
;
5226 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5230 if (pmd_trans_unstable(pmd
))
5232 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5233 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
5234 if (get_mctgt_type(vma
, addr
, *pte
, NULL
))
5235 mc
.precharge
++; /* increment precharge temporarily */
5236 pte_unmap_unlock(pte
- 1, ptl
);
5242 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
5244 unsigned long precharge
;
5245 struct vm_area_struct
*vma
;
5247 down_read(&mm
->mmap_sem
);
5248 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5249 struct mm_walk mem_cgroup_count_precharge_walk
= {
5250 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
5254 if (is_vm_hugetlb_page(vma
))
5256 walk_page_range(vma
->vm_start
, vma
->vm_end
,
5257 &mem_cgroup_count_precharge_walk
);
5259 up_read(&mm
->mmap_sem
);
5261 precharge
= mc
.precharge
;
5267 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
5269 unsigned long precharge
= mem_cgroup_count_precharge(mm
);
5271 VM_BUG_ON(mc
.moving_task
);
5272 mc
.moving_task
= current
;
5273 return mem_cgroup_do_precharge(precharge
);
5276 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5277 static void __mem_cgroup_clear_mc(void)
5279 struct mem_cgroup
*from
= mc
.from
;
5280 struct mem_cgroup
*to
= mc
.to
;
5282 /* we must uncharge all the leftover precharges from mc.to */
5284 __mem_cgroup_cancel_charge(mc
.to
, mc
.precharge
);
5288 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5289 * we must uncharge here.
5291 if (mc
.moved_charge
) {
5292 __mem_cgroup_cancel_charge(mc
.from
, mc
.moved_charge
);
5293 mc
.moved_charge
= 0;
5295 /* we must fixup refcnts and charges */
5296 if (mc
.moved_swap
) {
5297 /* uncharge swap account from the old cgroup */
5298 if (!mem_cgroup_is_root(mc
.from
))
5299 res_counter_uncharge(&mc
.from
->memsw
,
5300 PAGE_SIZE
* mc
.moved_swap
);
5301 __mem_cgroup_put(mc
.from
, mc
.moved_swap
);
5303 if (!mem_cgroup_is_root(mc
.to
)) {
5305 * we charged both to->res and to->memsw, so we should
5308 res_counter_uncharge(&mc
.to
->res
,
5309 PAGE_SIZE
* mc
.moved_swap
);
5311 /* we've already done mem_cgroup_get(mc.to) */
5314 memcg_oom_recover(from
);
5315 memcg_oom_recover(to
);
5316 wake_up_all(&mc
.waitq
);
5319 static void mem_cgroup_clear_mc(void)
5321 struct mem_cgroup
*from
= mc
.from
;
5324 * we must clear moving_task before waking up waiters at the end of
5327 mc
.moving_task
= NULL
;
5328 __mem_cgroup_clear_mc();
5329 spin_lock(&mc
.lock
);
5332 spin_unlock(&mc
.lock
);
5333 mem_cgroup_end_move(from
);
5336 static int mem_cgroup_can_attach(struct cgroup
*cgroup
,
5337 struct cgroup_taskset
*tset
)
5339 struct task_struct
*p
= cgroup_taskset_first(tset
);
5341 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgroup
);
5343 if (memcg
->move_charge_at_immigrate
) {
5344 struct mm_struct
*mm
;
5345 struct mem_cgroup
*from
= mem_cgroup_from_task(p
);
5347 VM_BUG_ON(from
== memcg
);
5349 mm
= get_task_mm(p
);
5352 /* We move charges only when we move a owner of the mm */
5353 if (mm
->owner
== p
) {
5356 VM_BUG_ON(mc
.precharge
);
5357 VM_BUG_ON(mc
.moved_charge
);
5358 VM_BUG_ON(mc
.moved_swap
);
5359 mem_cgroup_start_move(from
);
5360 spin_lock(&mc
.lock
);
5363 spin_unlock(&mc
.lock
);
5364 /* We set mc.moving_task later */
5366 ret
= mem_cgroup_precharge_mc(mm
);
5368 mem_cgroup_clear_mc();
5375 static void mem_cgroup_cancel_attach(struct cgroup
*cgroup
,
5376 struct cgroup_taskset
*tset
)
5378 mem_cgroup_clear_mc();
5381 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
5382 unsigned long addr
, unsigned long end
,
5383 struct mm_walk
*walk
)
5386 struct vm_area_struct
*vma
= walk
->private;
5389 enum mc_target_type target_type
;
5390 union mc_target target
;
5392 struct page_cgroup
*pc
;
5395 * We don't take compound_lock() here but no race with splitting thp
5397 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
5398 * under splitting, which means there's no concurrent thp split,
5399 * - if another thread runs into split_huge_page() just after we
5400 * entered this if-block, the thread must wait for page table lock
5401 * to be unlocked in __split_huge_page_splitting(), where the main
5402 * part of thp split is not executed yet.
5404 if (pmd_trans_huge_lock(pmd
, vma
) == 1) {
5405 if (mc
.precharge
< HPAGE_PMD_NR
) {
5406 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5409 target_type
= get_mctgt_type_thp(vma
, addr
, *pmd
, &target
);
5410 if (target_type
== MC_TARGET_PAGE
) {
5412 if (!isolate_lru_page(page
)) {
5413 pc
= lookup_page_cgroup(page
);
5414 if (!mem_cgroup_move_account(page
, HPAGE_PMD_NR
,
5415 pc
, mc
.from
, mc
.to
)) {
5416 mc
.precharge
-= HPAGE_PMD_NR
;
5417 mc
.moved_charge
+= HPAGE_PMD_NR
;
5419 putback_lru_page(page
);
5423 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5427 if (pmd_trans_unstable(pmd
))
5430 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5431 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
5432 pte_t ptent
= *(pte
++);
5438 switch (get_mctgt_type(vma
, addr
, ptent
, &target
)) {
5439 case MC_TARGET_PAGE
:
5441 if (isolate_lru_page(page
))
5443 pc
= lookup_page_cgroup(page
);
5444 if (!mem_cgroup_move_account(page
, 1, pc
,
5447 /* we uncharge from mc.from later. */
5450 putback_lru_page(page
);
5451 put
: /* get_mctgt_type() gets the page */
5454 case MC_TARGET_SWAP
:
5456 if (!mem_cgroup_move_swap_account(ent
, mc
.from
, mc
.to
)) {
5458 /* we fixup refcnts and charges later. */
5466 pte_unmap_unlock(pte
- 1, ptl
);
5471 * We have consumed all precharges we got in can_attach().
5472 * We try charge one by one, but don't do any additional
5473 * charges to mc.to if we have failed in charge once in attach()
5476 ret
= mem_cgroup_do_precharge(1);
5484 static void mem_cgroup_move_charge(struct mm_struct
*mm
)
5486 struct vm_area_struct
*vma
;
5488 lru_add_drain_all();
5490 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
5492 * Someone who are holding the mmap_sem might be waiting in
5493 * waitq. So we cancel all extra charges, wake up all waiters,
5494 * and retry. Because we cancel precharges, we might not be able
5495 * to move enough charges, but moving charge is a best-effort
5496 * feature anyway, so it wouldn't be a big problem.
5498 __mem_cgroup_clear_mc();
5502 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5504 struct mm_walk mem_cgroup_move_charge_walk
= {
5505 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
5509 if (is_vm_hugetlb_page(vma
))
5511 ret
= walk_page_range(vma
->vm_start
, vma
->vm_end
,
5512 &mem_cgroup_move_charge_walk
);
5515 * means we have consumed all precharges and failed in
5516 * doing additional charge. Just abandon here.
5520 up_read(&mm
->mmap_sem
);
5523 static void mem_cgroup_move_task(struct cgroup
*cont
,
5524 struct cgroup_taskset
*tset
)
5526 struct task_struct
*p
= cgroup_taskset_first(tset
);
5527 struct mm_struct
*mm
= get_task_mm(p
);
5531 mem_cgroup_move_charge(mm
);
5535 mem_cgroup_clear_mc();
5537 #else /* !CONFIG_MMU */
5538 static int mem_cgroup_can_attach(struct cgroup
*cgroup
,
5539 struct cgroup_taskset
*tset
)
5543 static void mem_cgroup_cancel_attach(struct cgroup
*cgroup
,
5544 struct cgroup_taskset
*tset
)
5547 static void mem_cgroup_move_task(struct cgroup
*cont
,
5548 struct cgroup_taskset
*tset
)
5553 struct cgroup_subsys mem_cgroup_subsys
= {
5555 .subsys_id
= mem_cgroup_subsys_id
,
5556 .create
= mem_cgroup_create
,
5557 .pre_destroy
= mem_cgroup_pre_destroy
,
5558 .destroy
= mem_cgroup_destroy
,
5559 .can_attach
= mem_cgroup_can_attach
,
5560 .cancel_attach
= mem_cgroup_cancel_attach
,
5561 .attach
= mem_cgroup_move_task
,
5562 .base_cftypes
= mem_cgroup_files
,
5565 .__DEPRECATED_clear_css_refs
= true,
5568 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
5569 static int __init
enable_swap_account(char *s
)
5571 /* consider enabled if no parameter or 1 is given */
5572 if (!strcmp(s
, "1"))
5573 really_do_swap_account
= 1;
5574 else if (!strcmp(s
, "0"))
5575 really_do_swap_account
= 0;
5578 __setup("swapaccount=", enable_swap_account
);