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 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_DATA
, /* end of data requires synchronization */
92 MEM_CGROUP_ON_MOVE
, /* someone is moving account between groups */
93 MEM_CGROUP_STAT_NSTATS
,
96 enum mem_cgroup_events_index
{
97 MEM_CGROUP_EVENTS_PGPGIN
, /* # of pages paged in */
98 MEM_CGROUP_EVENTS_PGPGOUT
, /* # of pages paged out */
99 MEM_CGROUP_EVENTS_COUNT
, /* # of pages paged in/out */
100 MEM_CGROUP_EVENTS_PGFAULT
, /* # of page-faults */
101 MEM_CGROUP_EVENTS_PGMAJFAULT
, /* # of major page-faults */
102 MEM_CGROUP_EVENTS_NSTATS
,
105 * Per memcg event counter is incremented at every pagein/pageout. With THP,
106 * it will be incremated by the number of pages. This counter is used for
107 * for trigger some periodic events. This is straightforward and better
108 * than using jiffies etc. to handle periodic memcg event.
110 enum mem_cgroup_events_target
{
111 MEM_CGROUP_TARGET_THRESH
,
112 MEM_CGROUP_TARGET_SOFTLIMIT
,
113 MEM_CGROUP_TARGET_NUMAINFO
,
116 #define THRESHOLDS_EVENTS_TARGET (128)
117 #define SOFTLIMIT_EVENTS_TARGET (1024)
118 #define NUMAINFO_EVENTS_TARGET (1024)
120 struct mem_cgroup_stat_cpu
{
121 long count
[MEM_CGROUP_STAT_NSTATS
];
122 unsigned long events
[MEM_CGROUP_EVENTS_NSTATS
];
123 unsigned long targets
[MEM_CGROUP_NTARGETS
];
127 * per-zone information in memory controller.
129 struct mem_cgroup_per_zone
{
131 * spin_lock to protect the per cgroup LRU
133 struct list_head lists
[NR_LRU_LISTS
];
134 unsigned long count
[NR_LRU_LISTS
];
136 struct zone_reclaim_stat reclaim_stat
;
137 struct rb_node tree_node
; /* RB tree node */
138 unsigned long long usage_in_excess
;/* Set to the value by which */
139 /* the soft limit is exceeded*/
141 struct mem_cgroup
*mem
; /* Back pointer, we cannot */
142 /* use container_of */
144 /* Macro for accessing counter */
145 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
147 struct mem_cgroup_per_node
{
148 struct mem_cgroup_per_zone zoneinfo
[MAX_NR_ZONES
];
151 struct mem_cgroup_lru_info
{
152 struct mem_cgroup_per_node
*nodeinfo
[MAX_NUMNODES
];
156 * Cgroups above their limits are maintained in a RB-Tree, independent of
157 * their hierarchy representation
160 struct mem_cgroup_tree_per_zone
{
161 struct rb_root rb_root
;
165 struct mem_cgroup_tree_per_node
{
166 struct mem_cgroup_tree_per_zone rb_tree_per_zone
[MAX_NR_ZONES
];
169 struct mem_cgroup_tree
{
170 struct mem_cgroup_tree_per_node
*rb_tree_per_node
[MAX_NUMNODES
];
173 static struct mem_cgroup_tree soft_limit_tree __read_mostly
;
175 struct mem_cgroup_threshold
{
176 struct eventfd_ctx
*eventfd
;
181 struct mem_cgroup_threshold_ary
{
182 /* An array index points to threshold just below usage. */
183 int current_threshold
;
184 /* Size of entries[] */
186 /* Array of thresholds */
187 struct mem_cgroup_threshold entries
[0];
190 struct mem_cgroup_thresholds
{
191 /* Primary thresholds array */
192 struct mem_cgroup_threshold_ary
*primary
;
194 * Spare threshold array.
195 * This is needed to make mem_cgroup_unregister_event() "never fail".
196 * It must be able to store at least primary->size - 1 entries.
198 struct mem_cgroup_threshold_ary
*spare
;
202 struct mem_cgroup_eventfd_list
{
203 struct list_head list
;
204 struct eventfd_ctx
*eventfd
;
207 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
);
208 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
);
211 * The memory controller data structure. The memory controller controls both
212 * page cache and RSS per cgroup. We would eventually like to provide
213 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
214 * to help the administrator determine what knobs to tune.
216 * TODO: Add a water mark for the memory controller. Reclaim will begin when
217 * we hit the water mark. May be even add a low water mark, such that
218 * no reclaim occurs from a cgroup at it's low water mark, this is
219 * a feature that will be implemented much later in the future.
222 struct cgroup_subsys_state css
;
224 * the counter to account for memory usage
226 struct res_counter res
;
228 * the counter to account for mem+swap usage.
230 struct res_counter memsw
;
232 * the counter to account for kmem usage.
234 struct res_counter kmem
;
236 * Per cgroup active and inactive list, similar to the
237 * per zone LRU lists.
239 struct mem_cgroup_lru_info info
;
241 * While reclaiming in a hierarchy, we cache the last child we
244 int last_scanned_child
;
245 int last_scanned_node
;
247 nodemask_t scan_nodes
;
248 atomic_t numainfo_events
;
249 atomic_t numainfo_updating
;
252 * Should the accounting and control be hierarchical, per subtree?
262 /* OOM-Killer disable */
263 int oom_kill_disable
;
265 /* set when res.limit == memsw.limit */
266 bool memsw_is_minimum
;
268 /* protect arrays of thresholds */
269 struct mutex thresholds_lock
;
271 /* thresholds for memory usage. RCU-protected */
272 struct mem_cgroup_thresholds thresholds
;
274 /* thresholds for mem+swap usage. RCU-protected */
275 struct mem_cgroup_thresholds memsw_thresholds
;
277 /* For oom notifier event fd */
278 struct list_head oom_notify
;
281 * Should we move charges of a task when a task is moved into this
282 * mem_cgroup ? And what type of charges should we move ?
284 unsigned long move_charge_at_immigrate
;
286 * Should kernel memory limits be stabilished independently
289 int kmem_independent_accounting
;
293 struct mem_cgroup_stat_cpu
*stat
;
295 * used when a cpu is offlined or other synchronizations
296 * See mem_cgroup_read_stat().
298 struct mem_cgroup_stat_cpu nocpu_base
;
299 spinlock_t pcp_counter_lock
;
302 struct tcp_memcontrol tcp_mem
;
306 /* Stuffs for move charges at task migration. */
308 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
309 * left-shifted bitmap of these types.
312 MOVE_CHARGE_TYPE_ANON
, /* private anonymous page and swap of it */
313 MOVE_CHARGE_TYPE_FILE
, /* file page(including tmpfs) and swap of it */
317 /* "mc" and its members are protected by cgroup_mutex */
318 static struct move_charge_struct
{
319 spinlock_t lock
; /* for from, to */
320 struct mem_cgroup
*from
;
321 struct mem_cgroup
*to
;
322 unsigned long precharge
;
323 unsigned long moved_charge
;
324 unsigned long moved_swap
;
325 struct task_struct
*moving_task
; /* a task moving charges */
326 wait_queue_head_t waitq
; /* a waitq for other context */
328 .lock
= __SPIN_LOCK_UNLOCKED(mc
.lock
),
329 .waitq
= __WAIT_QUEUE_HEAD_INITIALIZER(mc
.waitq
),
332 static bool move_anon(void)
334 return test_bit(MOVE_CHARGE_TYPE_ANON
,
335 &mc
.to
->move_charge_at_immigrate
);
338 static bool move_file(void)
340 return test_bit(MOVE_CHARGE_TYPE_FILE
,
341 &mc
.to
->move_charge_at_immigrate
);
345 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
346 * limit reclaim to prevent infinite loops, if they ever occur.
348 #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
349 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
352 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
353 MEM_CGROUP_CHARGE_TYPE_MAPPED
,
354 MEM_CGROUP_CHARGE_TYPE_SHMEM
, /* used by page migration of shmem */
355 MEM_CGROUP_CHARGE_TYPE_FORCE
, /* used by force_empty */
356 MEM_CGROUP_CHARGE_TYPE_SWAPOUT
, /* for accounting swapcache */
357 MEM_CGROUP_CHARGE_TYPE_DROP
, /* a page was unused swap cache */
361 /* for encoding cft->private value on file */
370 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
371 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
372 #define MEMFILE_ATTR(val) ((val) & 0xffff)
373 /* Used for OOM nofiier */
374 #define OOM_CONTROL (0)
377 * Reclaim flags for mem_cgroup_hierarchical_reclaim
379 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
380 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
381 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
382 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
383 #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2
384 #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT)
386 static void mem_cgroup_get(struct mem_cgroup
*memcg
);
387 static void mem_cgroup_put(struct mem_cgroup
*memcg
);
389 /* Writing them here to avoid exposing memcg's inner layout */
390 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
392 #include <net/sock.h>
395 static bool mem_cgroup_is_root(struct mem_cgroup
*memcg
);
396 void sock_update_memcg(struct sock
*sk
)
398 /* A socket spends its whole life in the same cgroup */
403 if (static_branch(&memcg_socket_limit_enabled
)) {
404 struct mem_cgroup
*memcg
;
406 BUG_ON(!sk
->sk_prot
->proto_cgroup
);
409 memcg
= mem_cgroup_from_task(current
);
410 if (!mem_cgroup_is_root(memcg
)) {
411 mem_cgroup_get(memcg
);
412 sk
->sk_cgrp
= sk
->sk_prot
->proto_cgroup(memcg
);
417 EXPORT_SYMBOL(sock_update_memcg
);
419 void sock_release_memcg(struct sock
*sk
)
421 if (static_branch(&memcg_socket_limit_enabled
) && sk
->sk_cgrp
) {
422 struct mem_cgroup
*memcg
;
423 WARN_ON(!sk
->sk_cgrp
->memcg
);
424 memcg
= sk
->sk_cgrp
->memcg
;
425 mem_cgroup_put(memcg
);
429 struct cg_proto
*tcp_proto_cgroup(struct mem_cgroup
*memcg
)
431 if (!memcg
|| mem_cgroup_is_root(memcg
))
434 return &memcg
->tcp_mem
.cg_proto
;
436 EXPORT_SYMBOL(tcp_proto_cgroup
);
437 #endif /* CONFIG_INET */
438 #endif /* CONFIG_CGROUP_MEM_RES_CTLR_KMEM */
440 static void drain_all_stock_async(struct mem_cgroup
*memcg
);
442 static struct mem_cgroup_per_zone
*
443 mem_cgroup_zoneinfo(struct mem_cgroup
*memcg
, int nid
, int zid
)
445 return &memcg
->info
.nodeinfo
[nid
]->zoneinfo
[zid
];
448 struct cgroup_subsys_state
*mem_cgroup_css(struct mem_cgroup
*memcg
)
453 static struct mem_cgroup_per_zone
*
454 page_cgroup_zoneinfo(struct mem_cgroup
*memcg
, struct page
*page
)
456 int nid
= page_to_nid(page
);
457 int zid
= page_zonenum(page
);
459 return mem_cgroup_zoneinfo(memcg
, nid
, zid
);
462 static struct mem_cgroup_tree_per_zone
*
463 soft_limit_tree_node_zone(int nid
, int zid
)
465 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
468 static struct mem_cgroup_tree_per_zone
*
469 soft_limit_tree_from_page(struct page
*page
)
471 int nid
= page_to_nid(page
);
472 int zid
= page_zonenum(page
);
474 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
478 __mem_cgroup_insert_exceeded(struct mem_cgroup
*memcg
,
479 struct mem_cgroup_per_zone
*mz
,
480 struct mem_cgroup_tree_per_zone
*mctz
,
481 unsigned long long new_usage_in_excess
)
483 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
484 struct rb_node
*parent
= NULL
;
485 struct mem_cgroup_per_zone
*mz_node
;
490 mz
->usage_in_excess
= new_usage_in_excess
;
491 if (!mz
->usage_in_excess
)
495 mz_node
= rb_entry(parent
, struct mem_cgroup_per_zone
,
497 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
)
500 * We can't avoid mem cgroups that are over their soft
501 * limit by the same amount
503 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
506 rb_link_node(&mz
->tree_node
, parent
, p
);
507 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
512 __mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
513 struct mem_cgroup_per_zone
*mz
,
514 struct mem_cgroup_tree_per_zone
*mctz
)
518 rb_erase(&mz
->tree_node
, &mctz
->rb_root
);
523 mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
524 struct mem_cgroup_per_zone
*mz
,
525 struct mem_cgroup_tree_per_zone
*mctz
)
527 spin_lock(&mctz
->lock
);
528 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
529 spin_unlock(&mctz
->lock
);
533 static void mem_cgroup_update_tree(struct mem_cgroup
*memcg
, struct page
*page
)
535 unsigned long long excess
;
536 struct mem_cgroup_per_zone
*mz
;
537 struct mem_cgroup_tree_per_zone
*mctz
;
538 int nid
= page_to_nid(page
);
539 int zid
= page_zonenum(page
);
540 mctz
= soft_limit_tree_from_page(page
);
543 * Necessary to update all ancestors when hierarchy is used.
544 * because their event counter is not touched.
546 for (; memcg
; memcg
= parent_mem_cgroup(memcg
)) {
547 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
548 excess
= res_counter_soft_limit_excess(&memcg
->res
);
550 * We have to update the tree if mz is on RB-tree or
551 * mem is over its softlimit.
553 if (excess
|| mz
->on_tree
) {
554 spin_lock(&mctz
->lock
);
555 /* if on-tree, remove it */
557 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
559 * Insert again. mz->usage_in_excess will be updated.
560 * If excess is 0, no tree ops.
562 __mem_cgroup_insert_exceeded(memcg
, mz
, mctz
, excess
);
563 spin_unlock(&mctz
->lock
);
568 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*memcg
)
571 struct mem_cgroup_per_zone
*mz
;
572 struct mem_cgroup_tree_per_zone
*mctz
;
574 for_each_node_state(node
, N_POSSIBLE
) {
575 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
576 mz
= mem_cgroup_zoneinfo(memcg
, node
, zone
);
577 mctz
= soft_limit_tree_node_zone(node
, zone
);
578 mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
583 static struct mem_cgroup_per_zone
*
584 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
586 struct rb_node
*rightmost
= NULL
;
587 struct mem_cgroup_per_zone
*mz
;
591 rightmost
= rb_last(&mctz
->rb_root
);
593 goto done
; /* Nothing to reclaim from */
595 mz
= rb_entry(rightmost
, struct mem_cgroup_per_zone
, tree_node
);
597 * Remove the node now but someone else can add it back,
598 * we will to add it back at the end of reclaim to its correct
599 * position in the tree.
601 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
602 if (!res_counter_soft_limit_excess(&mz
->mem
->res
) ||
603 !css_tryget(&mz
->mem
->css
))
609 static struct mem_cgroup_per_zone
*
610 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
612 struct mem_cgroup_per_zone
*mz
;
614 spin_lock(&mctz
->lock
);
615 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
616 spin_unlock(&mctz
->lock
);
621 * Implementation Note: reading percpu statistics for memcg.
623 * Both of vmstat[] and percpu_counter has threshold and do periodic
624 * synchronization to implement "quick" read. There are trade-off between
625 * reading cost and precision of value. Then, we may have a chance to implement
626 * a periodic synchronizion of counter in memcg's counter.
628 * But this _read() function is used for user interface now. The user accounts
629 * memory usage by memory cgroup and he _always_ requires exact value because
630 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
631 * have to visit all online cpus and make sum. So, for now, unnecessary
632 * synchronization is not implemented. (just implemented for cpu hotplug)
634 * If there are kernel internal actions which can make use of some not-exact
635 * value, and reading all cpu value can be performance bottleneck in some
636 * common workload, threashold and synchonization as vmstat[] should be
639 static long mem_cgroup_read_stat(struct mem_cgroup
*memcg
,
640 enum mem_cgroup_stat_index idx
)
646 for_each_online_cpu(cpu
)
647 val
+= per_cpu(memcg
->stat
->count
[idx
], cpu
);
648 #ifdef CONFIG_HOTPLUG_CPU
649 spin_lock(&memcg
->pcp_counter_lock
);
650 val
+= memcg
->nocpu_base
.count
[idx
];
651 spin_unlock(&memcg
->pcp_counter_lock
);
657 static void mem_cgroup_swap_statistics(struct mem_cgroup
*memcg
,
660 int val
= (charge
) ? 1 : -1;
661 this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_SWAPOUT
], val
);
664 void mem_cgroup_pgfault(struct mem_cgroup
*memcg
, int val
)
666 this_cpu_add(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGFAULT
], val
);
669 void mem_cgroup_pgmajfault(struct mem_cgroup
*memcg
, int val
)
671 this_cpu_add(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGMAJFAULT
], val
);
674 static unsigned long mem_cgroup_read_events(struct mem_cgroup
*memcg
,
675 enum mem_cgroup_events_index idx
)
677 unsigned long val
= 0;
680 for_each_online_cpu(cpu
)
681 val
+= per_cpu(memcg
->stat
->events
[idx
], cpu
);
682 #ifdef CONFIG_HOTPLUG_CPU
683 spin_lock(&memcg
->pcp_counter_lock
);
684 val
+= memcg
->nocpu_base
.events
[idx
];
685 spin_unlock(&memcg
->pcp_counter_lock
);
690 static void mem_cgroup_charge_statistics(struct mem_cgroup
*memcg
,
691 bool file
, int nr_pages
)
696 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_CACHE
],
699 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_RSS
],
702 /* pagein of a big page is an event. So, ignore page size */
704 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGIN
]);
706 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGOUT
]);
707 nr_pages
= -nr_pages
; /* for event */
710 __this_cpu_add(memcg
->stat
->events
[MEM_CGROUP_EVENTS_COUNT
], nr_pages
);
716 mem_cgroup_zone_nr_lru_pages(struct mem_cgroup
*memcg
, int nid
, int zid
,
717 unsigned int lru_mask
)
719 struct mem_cgroup_per_zone
*mz
;
721 unsigned long ret
= 0;
723 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
726 if (BIT(l
) & lru_mask
)
727 ret
+= MEM_CGROUP_ZSTAT(mz
, l
);
733 mem_cgroup_node_nr_lru_pages(struct mem_cgroup
*memcg
,
734 int nid
, unsigned int lru_mask
)
739 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++)
740 total
+= mem_cgroup_zone_nr_lru_pages(memcg
,
746 static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup
*memcg
,
747 unsigned int lru_mask
)
752 for_each_node_state(nid
, N_HIGH_MEMORY
)
753 total
+= mem_cgroup_node_nr_lru_pages(memcg
, nid
, lru_mask
);
757 static bool __memcg_event_check(struct mem_cgroup
*memcg
, int target
)
759 unsigned long val
, next
;
761 val
= __this_cpu_read(memcg
->stat
->events
[MEM_CGROUP_EVENTS_COUNT
]);
762 next
= __this_cpu_read(memcg
->stat
->targets
[target
]);
763 /* from time_after() in jiffies.h */
764 return ((long)next
- (long)val
< 0);
767 static void __mem_cgroup_target_update(struct mem_cgroup
*memcg
, int target
)
769 unsigned long val
, next
;
771 val
= __this_cpu_read(memcg
->stat
->events
[MEM_CGROUP_EVENTS_COUNT
]);
774 case MEM_CGROUP_TARGET_THRESH
:
775 next
= val
+ THRESHOLDS_EVENTS_TARGET
;
777 case MEM_CGROUP_TARGET_SOFTLIMIT
:
778 next
= val
+ SOFTLIMIT_EVENTS_TARGET
;
780 case MEM_CGROUP_TARGET_NUMAINFO
:
781 next
= val
+ NUMAINFO_EVENTS_TARGET
;
787 __this_cpu_write(memcg
->stat
->targets
[target
], next
);
791 * Check events in order.
794 static void memcg_check_events(struct mem_cgroup
*memcg
, struct page
*page
)
797 /* threshold event is triggered in finer grain than soft limit */
798 if (unlikely(__memcg_event_check(memcg
, MEM_CGROUP_TARGET_THRESH
))) {
799 mem_cgroup_threshold(memcg
);
800 __mem_cgroup_target_update(memcg
, MEM_CGROUP_TARGET_THRESH
);
801 if (unlikely(__memcg_event_check(memcg
,
802 MEM_CGROUP_TARGET_SOFTLIMIT
))) {
803 mem_cgroup_update_tree(memcg
, page
);
804 __mem_cgroup_target_update(memcg
,
805 MEM_CGROUP_TARGET_SOFTLIMIT
);
808 if (unlikely(__memcg_event_check(memcg
,
809 MEM_CGROUP_TARGET_NUMAINFO
))) {
810 atomic_inc(&memcg
->numainfo_events
);
811 __mem_cgroup_target_update(memcg
,
812 MEM_CGROUP_TARGET_NUMAINFO
);
819 struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
821 return container_of(cgroup_subsys_state(cont
,
822 mem_cgroup_subsys_id
), struct mem_cgroup
,
826 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
829 * mm_update_next_owner() may clear mm->owner to NULL
830 * if it races with swapoff, page migration, etc.
831 * So this can be called with p == NULL.
836 return container_of(task_subsys_state(p
, mem_cgroup_subsys_id
),
837 struct mem_cgroup
, css
);
840 struct mem_cgroup
*try_get_mem_cgroup_from_mm(struct mm_struct
*mm
)
842 struct mem_cgroup
*memcg
= NULL
;
847 * Because we have no locks, mm->owner's may be being moved to other
848 * cgroup. We use css_tryget() here even if this looks
849 * pessimistic (rather than adding locks here).
853 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
854 if (unlikely(!memcg
))
856 } while (!css_tryget(&memcg
->css
));
861 /* The caller has to guarantee "mem" exists before calling this */
862 static struct mem_cgroup
*mem_cgroup_start_loop(struct mem_cgroup
*memcg
)
864 struct cgroup_subsys_state
*css
;
867 if (!memcg
) /* ROOT cgroup has the smallest ID */
868 return root_mem_cgroup
; /*css_put/get against root is ignored*/
869 if (!memcg
->use_hierarchy
) {
870 if (css_tryget(&memcg
->css
))
876 * searching a memory cgroup which has the smallest ID under given
877 * ROOT cgroup. (ID >= 1)
879 css
= css_get_next(&mem_cgroup_subsys
, 1, &memcg
->css
, &found
);
880 if (css
&& css_tryget(css
))
881 memcg
= container_of(css
, struct mem_cgroup
, css
);
888 static struct mem_cgroup
*mem_cgroup_get_next(struct mem_cgroup
*iter
,
889 struct mem_cgroup
*root
,
892 int nextid
= css_id(&iter
->css
) + 1;
895 struct cgroup_subsys_state
*css
;
897 hierarchy_used
= iter
->use_hierarchy
;
900 /* If no ROOT, walk all, ignore hierarchy */
901 if (!cond
|| (root
&& !hierarchy_used
))
905 root
= root_mem_cgroup
;
911 css
= css_get_next(&mem_cgroup_subsys
, nextid
,
913 if (css
&& css_tryget(css
))
914 iter
= container_of(css
, struct mem_cgroup
, css
);
916 /* If css is NULL, no more cgroups will be found */
918 } while (css
&& !iter
);
923 * for_eacn_mem_cgroup_tree() for visiting all cgroup under tree. Please
924 * be careful that "break" loop is not allowed. We have reference count.
925 * Instead of that modify "cond" to be false and "continue" to exit the loop.
927 #define for_each_mem_cgroup_tree_cond(iter, root, cond) \
928 for (iter = mem_cgroup_start_loop(root);\
930 iter = mem_cgroup_get_next(iter, root, cond))
932 #define for_each_mem_cgroup_tree(iter, root) \
933 for_each_mem_cgroup_tree_cond(iter, root, true)
935 #define for_each_mem_cgroup_all(iter) \
936 for_each_mem_cgroup_tree_cond(iter, NULL, true)
939 static inline bool mem_cgroup_is_root(struct mem_cgroup
*memcg
)
941 return (memcg
== root_mem_cgroup
);
944 void mem_cgroup_count_vm_event(struct mm_struct
*mm
, enum vm_event_item idx
)
946 struct mem_cgroup
*memcg
;
952 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
953 if (unlikely(!memcg
))
958 mem_cgroup_pgmajfault(memcg
, 1);
961 mem_cgroup_pgfault(memcg
, 1);
969 EXPORT_SYMBOL(mem_cgroup_count_vm_event
);
972 * Following LRU functions are allowed to be used without PCG_LOCK.
973 * Operations are called by routine of global LRU independently from memcg.
974 * What we have to take care of here is validness of pc->mem_cgroup.
976 * Changes to pc->mem_cgroup happens when
979 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
980 * It is added to LRU before charge.
981 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
982 * When moving account, the page is not on LRU. It's isolated.
985 void mem_cgroup_del_lru_list(struct page
*page
, enum lru_list lru
)
987 struct page_cgroup
*pc
;
988 struct mem_cgroup_per_zone
*mz
;
990 if (mem_cgroup_disabled())
992 pc
= lookup_page_cgroup(page
);
993 /* can happen while we handle swapcache. */
994 if (!TestClearPageCgroupAcctLRU(pc
))
996 VM_BUG_ON(!pc
->mem_cgroup
);
998 * We don't check PCG_USED bit. It's cleared when the "page" is finally
999 * removed from global LRU.
1001 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
1002 /* huge page split is done under lru_lock. so, we have no races. */
1003 MEM_CGROUP_ZSTAT(mz
, lru
) -= 1 << compound_order(page
);
1004 if (mem_cgroup_is_root(pc
->mem_cgroup
))
1006 VM_BUG_ON(list_empty(&pc
->lru
));
1007 list_del_init(&pc
->lru
);
1010 void mem_cgroup_del_lru(struct page
*page
)
1012 mem_cgroup_del_lru_list(page
, page_lru(page
));
1016 * Writeback is about to end against a page which has been marked for immediate
1017 * reclaim. If it still appears to be reclaimable, move it to the tail of the
1020 void mem_cgroup_rotate_reclaimable_page(struct page
*page
)
1022 struct mem_cgroup_per_zone
*mz
;
1023 struct page_cgroup
*pc
;
1024 enum lru_list lru
= page_lru(page
);
1026 if (mem_cgroup_disabled())
1029 pc
= lookup_page_cgroup(page
);
1030 /* unused or root page is not rotated. */
1031 if (!PageCgroupUsed(pc
))
1033 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1035 if (mem_cgroup_is_root(pc
->mem_cgroup
))
1037 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
1038 list_move_tail(&pc
->lru
, &mz
->lists
[lru
]);
1041 void mem_cgroup_rotate_lru_list(struct page
*page
, enum lru_list lru
)
1043 struct mem_cgroup_per_zone
*mz
;
1044 struct page_cgroup
*pc
;
1046 if (mem_cgroup_disabled())
1049 pc
= lookup_page_cgroup(page
);
1050 /* unused or root page is not rotated. */
1051 if (!PageCgroupUsed(pc
))
1053 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1055 if (mem_cgroup_is_root(pc
->mem_cgroup
))
1057 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
1058 list_move(&pc
->lru
, &mz
->lists
[lru
]);
1061 void mem_cgroup_add_lru_list(struct page
*page
, enum lru_list lru
)
1063 struct page_cgroup
*pc
;
1064 struct mem_cgroup_per_zone
*mz
;
1066 if (mem_cgroup_disabled())
1068 pc
= lookup_page_cgroup(page
);
1069 VM_BUG_ON(PageCgroupAcctLRU(pc
));
1072 * SetPageLRU SetPageCgroupUsed
1074 * PageCgroupUsed && add to memcg LRU PageLRU && add to memcg LRU
1076 * Ensure that one of the two sides adds the page to the memcg
1077 * LRU during a race.
1080 if (!PageCgroupUsed(pc
))
1082 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1084 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
1085 /* huge page split is done under lru_lock. so, we have no races. */
1086 MEM_CGROUP_ZSTAT(mz
, lru
) += 1 << compound_order(page
);
1087 SetPageCgroupAcctLRU(pc
);
1088 if (mem_cgroup_is_root(pc
->mem_cgroup
))
1090 list_add(&pc
->lru
, &mz
->lists
[lru
]);
1094 * At handling SwapCache and other FUSE stuff, pc->mem_cgroup may be changed
1095 * while it's linked to lru because the page may be reused after it's fully
1096 * uncharged. To handle that, unlink page_cgroup from LRU when charge it again.
1097 * It's done under lock_page and expected that zone->lru_lock isnever held.
1099 static void mem_cgroup_lru_del_before_commit(struct page
*page
)
1101 unsigned long flags
;
1102 struct zone
*zone
= page_zone(page
);
1103 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1106 * Doing this check without taking ->lru_lock seems wrong but this
1107 * is safe. Because if page_cgroup's USED bit is unset, the page
1108 * will not be added to any memcg's LRU. If page_cgroup's USED bit is
1109 * set, the commit after this will fail, anyway.
1110 * This all charge/uncharge is done under some mutual execustion.
1111 * So, we don't need to taking care of changes in USED bit.
1113 if (likely(!PageLRU(page
)))
1116 spin_lock_irqsave(&zone
->lru_lock
, flags
);
1118 * Forget old LRU when this page_cgroup is *not* used. This Used bit
1119 * is guarded by lock_page() because the page is SwapCache.
1121 if (!PageCgroupUsed(pc
))
1122 mem_cgroup_del_lru_list(page
, page_lru(page
));
1123 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
1126 static void mem_cgroup_lru_add_after_commit(struct page
*page
)
1128 unsigned long flags
;
1129 struct zone
*zone
= page_zone(page
);
1130 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1133 * SetPageLRU SetPageCgroupUsed
1135 * PageCgroupUsed && add to memcg LRU PageLRU && add to memcg LRU
1137 * Ensure that one of the two sides adds the page to the memcg
1138 * LRU during a race.
1141 /* taking care of that the page is added to LRU while we commit it */
1142 if (likely(!PageLRU(page
)))
1144 spin_lock_irqsave(&zone
->lru_lock
, flags
);
1145 /* link when the page is linked to LRU but page_cgroup isn't */
1146 if (PageLRU(page
) && !PageCgroupAcctLRU(pc
))
1147 mem_cgroup_add_lru_list(page
, page_lru(page
));
1148 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
1152 void mem_cgroup_move_lists(struct page
*page
,
1153 enum lru_list from
, enum lru_list to
)
1155 if (mem_cgroup_disabled())
1157 mem_cgroup_del_lru_list(page
, from
);
1158 mem_cgroup_add_lru_list(page
, to
);
1162 * Checks whether given mem is same or in the root_mem_cgroup's
1165 static bool mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_memcg
,
1166 struct mem_cgroup
*memcg
)
1168 if (root_memcg
!= memcg
) {
1169 return (root_memcg
->use_hierarchy
&&
1170 css_is_ancestor(&memcg
->css
, &root_memcg
->css
));
1176 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*memcg
)
1179 struct mem_cgroup
*curr
= NULL
;
1180 struct task_struct
*p
;
1182 p
= find_lock_task_mm(task
);
1185 curr
= try_get_mem_cgroup_from_mm(p
->mm
);
1190 * We should check use_hierarchy of "memcg" not "curr". Because checking
1191 * use_hierarchy of "curr" here make this function true if hierarchy is
1192 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1193 * hierarchy(even if use_hierarchy is disabled in "memcg").
1195 ret
= mem_cgroup_same_or_subtree(memcg
, curr
);
1196 css_put(&curr
->css
);
1200 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup
*memcg
, struct zone
*zone
)
1202 unsigned long inactive_ratio
;
1203 int nid
= zone_to_nid(zone
);
1204 int zid
= zone_idx(zone
);
1205 unsigned long inactive
;
1206 unsigned long active
;
1209 inactive
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1210 BIT(LRU_INACTIVE_ANON
));
1211 active
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1212 BIT(LRU_ACTIVE_ANON
));
1214 gb
= (inactive
+ active
) >> (30 - PAGE_SHIFT
);
1216 inactive_ratio
= int_sqrt(10 * gb
);
1220 return inactive
* inactive_ratio
< active
;
1223 int mem_cgroup_inactive_file_is_low(struct mem_cgroup
*memcg
, struct zone
*zone
)
1225 unsigned long active
;
1226 unsigned long inactive
;
1227 int zid
= zone_idx(zone
);
1228 int nid
= zone_to_nid(zone
);
1230 inactive
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1231 BIT(LRU_INACTIVE_FILE
));
1232 active
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1233 BIT(LRU_ACTIVE_FILE
));
1235 return (active
> inactive
);
1238 struct zone_reclaim_stat
*mem_cgroup_get_reclaim_stat(struct mem_cgroup
*memcg
,
1241 int nid
= zone_to_nid(zone
);
1242 int zid
= zone_idx(zone
);
1243 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
1245 return &mz
->reclaim_stat
;
1248 struct zone_reclaim_stat
*
1249 mem_cgroup_get_reclaim_stat_from_page(struct page
*page
)
1251 struct page_cgroup
*pc
;
1252 struct mem_cgroup_per_zone
*mz
;
1254 if (mem_cgroup_disabled())
1257 pc
= lookup_page_cgroup(page
);
1258 if (!PageCgroupUsed(pc
))
1260 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1262 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
1263 return &mz
->reclaim_stat
;
1266 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan
,
1267 struct list_head
*dst
,
1268 unsigned long *scanned
, int order
,
1269 isolate_mode_t mode
,
1271 struct mem_cgroup
*mem_cont
,
1272 int active
, int file
)
1274 unsigned long nr_taken
= 0;
1278 struct list_head
*src
;
1279 struct page_cgroup
*pc
, *tmp
;
1280 int nid
= zone_to_nid(z
);
1281 int zid
= zone_idx(z
);
1282 struct mem_cgroup_per_zone
*mz
;
1283 int lru
= LRU_FILE
* file
+ active
;
1287 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
1288 src
= &mz
->lists
[lru
];
1291 list_for_each_entry_safe_reverse(pc
, tmp
, src
, lru
) {
1292 if (scan
>= nr_to_scan
)
1295 if (unlikely(!PageCgroupUsed(pc
)))
1298 page
= lookup_cgroup_page(pc
);
1300 if (unlikely(!PageLRU(page
)))
1304 ret
= __isolate_lru_page(page
, mode
, file
);
1307 list_move(&page
->lru
, dst
);
1308 mem_cgroup_del_lru(page
);
1309 nr_taken
+= hpage_nr_pages(page
);
1312 /* we don't affect global LRU but rotate in our LRU */
1313 mem_cgroup_rotate_lru_list(page
, page_lru(page
));
1322 trace_mm_vmscan_memcg_isolate(0, nr_to_scan
, scan
, nr_taken
,
1328 #define mem_cgroup_from_res_counter(counter, member) \
1329 container_of(counter, struct mem_cgroup, member)
1332 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1333 * @mem: the memory cgroup
1335 * Returns the maximum amount of memory @mem can be charged with, in
1338 static unsigned long mem_cgroup_margin(struct mem_cgroup
*memcg
)
1340 unsigned long long margin
;
1342 margin
= res_counter_margin(&memcg
->res
);
1343 if (do_swap_account
)
1344 margin
= min(margin
, res_counter_margin(&memcg
->memsw
));
1345 return margin
>> PAGE_SHIFT
;
1348 int mem_cgroup_swappiness(struct mem_cgroup
*memcg
)
1350 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
1353 if (cgrp
->parent
== NULL
)
1354 return vm_swappiness
;
1356 return memcg
->swappiness
;
1359 static void mem_cgroup_start_move(struct mem_cgroup
*memcg
)
1364 spin_lock(&memcg
->pcp_counter_lock
);
1365 for_each_online_cpu(cpu
)
1366 per_cpu(memcg
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) += 1;
1367 memcg
->nocpu_base
.count
[MEM_CGROUP_ON_MOVE
] += 1;
1368 spin_unlock(&memcg
->pcp_counter_lock
);
1374 static void mem_cgroup_end_move(struct mem_cgroup
*memcg
)
1381 spin_lock(&memcg
->pcp_counter_lock
);
1382 for_each_online_cpu(cpu
)
1383 per_cpu(memcg
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) -= 1;
1384 memcg
->nocpu_base
.count
[MEM_CGROUP_ON_MOVE
] -= 1;
1385 spin_unlock(&memcg
->pcp_counter_lock
);
1389 * 2 routines for checking "mem" is under move_account() or not.
1391 * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used
1392 * for avoiding race in accounting. If true,
1393 * pc->mem_cgroup may be overwritten.
1395 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1396 * under hierarchy of moving cgroups. This is for
1397 * waiting at hith-memory prressure caused by "move".
1400 static bool mem_cgroup_stealed(struct mem_cgroup
*memcg
)
1402 VM_BUG_ON(!rcu_read_lock_held());
1403 return this_cpu_read(memcg
->stat
->count
[MEM_CGROUP_ON_MOVE
]) > 0;
1406 static bool mem_cgroup_under_move(struct mem_cgroup
*memcg
)
1408 struct mem_cgroup
*from
;
1409 struct mem_cgroup
*to
;
1412 * Unlike task_move routines, we access mc.to, mc.from not under
1413 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1415 spin_lock(&mc
.lock
);
1421 ret
= mem_cgroup_same_or_subtree(memcg
, from
)
1422 || mem_cgroup_same_or_subtree(memcg
, to
);
1424 spin_unlock(&mc
.lock
);
1428 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*memcg
)
1430 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1431 if (mem_cgroup_under_move(memcg
)) {
1433 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1434 /* moving charge context might have finished. */
1437 finish_wait(&mc
.waitq
, &wait
);
1445 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1446 * @memcg: The memory cgroup that went over limit
1447 * @p: Task that is going to be killed
1449 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1452 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1454 struct cgroup
*task_cgrp
;
1455 struct cgroup
*mem_cgrp
;
1457 * Need a buffer in BSS, can't rely on allocations. The code relies
1458 * on the assumption that OOM is serialized for memory controller.
1459 * If this assumption is broken, revisit this code.
1461 static char memcg_name
[PATH_MAX
];
1470 mem_cgrp
= memcg
->css
.cgroup
;
1471 task_cgrp
= task_cgroup(p
, mem_cgroup_subsys_id
);
1473 ret
= cgroup_path(task_cgrp
, memcg_name
, PATH_MAX
);
1476 * Unfortunately, we are unable to convert to a useful name
1477 * But we'll still print out the usage information
1484 printk(KERN_INFO
"Task in %s killed", memcg_name
);
1487 ret
= cgroup_path(mem_cgrp
, memcg_name
, PATH_MAX
);
1495 * Continues from above, so we don't need an KERN_ level
1497 printk(KERN_CONT
" as a result of limit of %s\n", memcg_name
);
1500 printk(KERN_INFO
"memory: usage %llukB, limit %llukB, failcnt %llu\n",
1501 res_counter_read_u64(&memcg
->res
, RES_USAGE
) >> 10,
1502 res_counter_read_u64(&memcg
->res
, RES_LIMIT
) >> 10,
1503 res_counter_read_u64(&memcg
->res
, RES_FAILCNT
));
1504 printk(KERN_INFO
"memory+swap: usage %llukB, limit %llukB, "
1506 res_counter_read_u64(&memcg
->memsw
, RES_USAGE
) >> 10,
1507 res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
) >> 10,
1508 res_counter_read_u64(&memcg
->memsw
, RES_FAILCNT
));
1512 * This function returns the number of memcg under hierarchy tree. Returns
1513 * 1(self count) if no children.
1515 static int mem_cgroup_count_children(struct mem_cgroup
*memcg
)
1518 struct mem_cgroup
*iter
;
1520 for_each_mem_cgroup_tree(iter
, memcg
)
1526 * Return the memory (and swap, if configured) limit for a memcg.
1528 u64
mem_cgroup_get_limit(struct mem_cgroup
*memcg
)
1533 limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
1534 limit
+= total_swap_pages
<< PAGE_SHIFT
;
1536 memsw
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
1538 * If memsw is finite and limits the amount of swap space available
1539 * to this memcg, return that limit.
1541 return min(limit
, memsw
);
1545 * Visit the first child (need not be the first child as per the ordering
1546 * of the cgroup list, since we track last_scanned_child) of @mem and use
1547 * that to reclaim free pages from.
1549 static struct mem_cgroup
*
1550 mem_cgroup_select_victim(struct mem_cgroup
*root_memcg
)
1552 struct mem_cgroup
*ret
= NULL
;
1553 struct cgroup_subsys_state
*css
;
1556 if (!root_memcg
->use_hierarchy
) {
1557 css_get(&root_memcg
->css
);
1563 nextid
= root_memcg
->last_scanned_child
+ 1;
1564 css
= css_get_next(&mem_cgroup_subsys
, nextid
, &root_memcg
->css
,
1566 if (css
&& css_tryget(css
))
1567 ret
= container_of(css
, struct mem_cgroup
, css
);
1570 /* Updates scanning parameter */
1572 /* this means start scan from ID:1 */
1573 root_memcg
->last_scanned_child
= 0;
1575 root_memcg
->last_scanned_child
= found
;
1582 * test_mem_cgroup_node_reclaimable
1583 * @mem: the target memcg
1584 * @nid: the node ID to be checked.
1585 * @noswap : specify true here if the user wants flle only information.
1587 * This function returns whether the specified memcg contains any
1588 * reclaimable pages on a node. Returns true if there are any reclaimable
1589 * pages in the node.
1591 static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup
*memcg
,
1592 int nid
, bool noswap
)
1594 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_FILE
))
1596 if (noswap
|| !total_swap_pages
)
1598 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_ANON
))
1603 #if MAX_NUMNODES > 1
1606 * Always updating the nodemask is not very good - even if we have an empty
1607 * list or the wrong list here, we can start from some node and traverse all
1608 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1611 static void mem_cgroup_may_update_nodemask(struct mem_cgroup
*memcg
)
1615 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1616 * pagein/pageout changes since the last update.
1618 if (!atomic_read(&memcg
->numainfo_events
))
1620 if (atomic_inc_return(&memcg
->numainfo_updating
) > 1)
1623 /* make a nodemask where this memcg uses memory from */
1624 memcg
->scan_nodes
= node_states
[N_HIGH_MEMORY
];
1626 for_each_node_mask(nid
, node_states
[N_HIGH_MEMORY
]) {
1628 if (!test_mem_cgroup_node_reclaimable(memcg
, nid
, false))
1629 node_clear(nid
, memcg
->scan_nodes
);
1632 atomic_set(&memcg
->numainfo_events
, 0);
1633 atomic_set(&memcg
->numainfo_updating
, 0);
1637 * Selecting a node where we start reclaim from. Because what we need is just
1638 * reducing usage counter, start from anywhere is O,K. Considering
1639 * memory reclaim from current node, there are pros. and cons.
1641 * Freeing memory from current node means freeing memory from a node which
1642 * we'll use or we've used. So, it may make LRU bad. And if several threads
1643 * hit limits, it will see a contention on a node. But freeing from remote
1644 * node means more costs for memory reclaim because of memory latency.
1646 * Now, we use round-robin. Better algorithm is welcomed.
1648 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1652 mem_cgroup_may_update_nodemask(memcg
);
1653 node
= memcg
->last_scanned_node
;
1655 node
= next_node(node
, memcg
->scan_nodes
);
1656 if (node
== MAX_NUMNODES
)
1657 node
= first_node(memcg
->scan_nodes
);
1659 * We call this when we hit limit, not when pages are added to LRU.
1660 * No LRU may hold pages because all pages are UNEVICTABLE or
1661 * memcg is too small and all pages are not on LRU. In that case,
1662 * we use curret node.
1664 if (unlikely(node
== MAX_NUMNODES
))
1665 node
= numa_node_id();
1667 memcg
->last_scanned_node
= node
;
1672 * Check all nodes whether it contains reclaimable pages or not.
1673 * For quick scan, we make use of scan_nodes. This will allow us to skip
1674 * unused nodes. But scan_nodes is lazily updated and may not cotain
1675 * enough new information. We need to do double check.
1677 bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1682 * quick check...making use of scan_node.
1683 * We can skip unused nodes.
1685 if (!nodes_empty(memcg
->scan_nodes
)) {
1686 for (nid
= first_node(memcg
->scan_nodes
);
1688 nid
= next_node(nid
, memcg
->scan_nodes
)) {
1690 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1695 * Check rest of nodes.
1697 for_each_node_state(nid
, N_HIGH_MEMORY
) {
1698 if (node_isset(nid
, memcg
->scan_nodes
))
1700 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1707 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1712 bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1714 return test_mem_cgroup_node_reclaimable(memcg
, 0, noswap
);
1719 * Scan the hierarchy if needed to reclaim memory. We remember the last child
1720 * we reclaimed from, so that we don't end up penalizing one child extensively
1721 * based on its position in the children list.
1723 * root_memcg is the original ancestor that we've been reclaim from.
1725 * We give up and return to the caller when we visit root_memcg twice.
1726 * (other groups can be removed while we're walking....)
1728 * If shrink==true, for avoiding to free too much, this returns immedieately.
1730 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup
*root_memcg
,
1733 unsigned long reclaim_options
,
1734 unsigned long *total_scanned
)
1736 struct mem_cgroup
*victim
;
1739 bool noswap
= reclaim_options
& MEM_CGROUP_RECLAIM_NOSWAP
;
1740 bool shrink
= reclaim_options
& MEM_CGROUP_RECLAIM_SHRINK
;
1741 bool check_soft
= reclaim_options
& MEM_CGROUP_RECLAIM_SOFT
;
1742 unsigned long excess
;
1743 unsigned long nr_scanned
;
1745 excess
= res_counter_soft_limit_excess(&root_memcg
->res
) >> PAGE_SHIFT
;
1747 /* If memsw_is_minimum==1, swap-out is of-no-use. */
1748 if (!check_soft
&& !shrink
&& root_memcg
->memsw_is_minimum
)
1752 victim
= mem_cgroup_select_victim(root_memcg
);
1753 if (victim
== root_memcg
) {
1756 * We are not draining per cpu cached charges during
1757 * soft limit reclaim because global reclaim doesn't
1758 * care about charges. It tries to free some memory and
1759 * charges will not give any.
1761 if (!check_soft
&& loop
>= 1)
1762 drain_all_stock_async(root_memcg
);
1765 * If we have not been able to reclaim
1766 * anything, it might because there are
1767 * no reclaimable pages under this hierarchy
1769 if (!check_soft
|| !total
) {
1770 css_put(&victim
->css
);
1774 * We want to do more targeted reclaim.
1775 * excess >> 2 is not to excessive so as to
1776 * reclaim too much, nor too less that we keep
1777 * coming back to reclaim from this cgroup
1779 if (total
>= (excess
>> 2) ||
1780 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
)) {
1781 css_put(&victim
->css
);
1786 if (!mem_cgroup_reclaimable(victim
, noswap
)) {
1787 /* this cgroup's local usage == 0 */
1788 css_put(&victim
->css
);
1791 /* we use swappiness of local cgroup */
1793 ret
= mem_cgroup_shrink_node_zone(victim
, gfp_mask
,
1794 noswap
, zone
, &nr_scanned
);
1795 *total_scanned
+= nr_scanned
;
1797 ret
= try_to_free_mem_cgroup_pages(victim
, gfp_mask
,
1799 css_put(&victim
->css
);
1801 * At shrinking usage, we can't check we should stop here or
1802 * reclaim more. It's depends on callers. last_scanned_child
1803 * will work enough for keeping fairness under tree.
1809 if (!res_counter_soft_limit_excess(&root_memcg
->res
))
1811 } else if (mem_cgroup_margin(root_memcg
))
1818 * Check OOM-Killer is already running under our hierarchy.
1819 * If someone is running, return false.
1820 * Has to be called with memcg_oom_lock
1822 static bool mem_cgroup_oom_lock(struct mem_cgroup
*memcg
)
1824 struct mem_cgroup
*iter
, *failed
= NULL
;
1827 for_each_mem_cgroup_tree_cond(iter
, memcg
, cond
) {
1828 if (iter
->oom_lock
) {
1830 * this subtree of our hierarchy is already locked
1831 * so we cannot give a lock.
1836 iter
->oom_lock
= true;
1843 * OK, we failed to lock the whole subtree so we have to clean up
1844 * what we set up to the failing subtree
1847 for_each_mem_cgroup_tree_cond(iter
, memcg
, cond
) {
1848 if (iter
== failed
) {
1852 iter
->oom_lock
= false;
1858 * Has to be called with memcg_oom_lock
1860 static int mem_cgroup_oom_unlock(struct mem_cgroup
*memcg
)
1862 struct mem_cgroup
*iter
;
1864 for_each_mem_cgroup_tree(iter
, memcg
)
1865 iter
->oom_lock
= false;
1869 static void mem_cgroup_mark_under_oom(struct mem_cgroup
*memcg
)
1871 struct mem_cgroup
*iter
;
1873 for_each_mem_cgroup_tree(iter
, memcg
)
1874 atomic_inc(&iter
->under_oom
);
1877 static void mem_cgroup_unmark_under_oom(struct mem_cgroup
*memcg
)
1879 struct mem_cgroup
*iter
;
1882 * When a new child is created while the hierarchy is under oom,
1883 * mem_cgroup_oom_lock() may not be called. We have to use
1884 * atomic_add_unless() here.
1886 for_each_mem_cgroup_tree(iter
, memcg
)
1887 atomic_add_unless(&iter
->under_oom
, -1, 0);
1890 static DEFINE_SPINLOCK(memcg_oom_lock
);
1891 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1893 struct oom_wait_info
{
1894 struct mem_cgroup
*mem
;
1898 static int memcg_oom_wake_function(wait_queue_t
*wait
,
1899 unsigned mode
, int sync
, void *arg
)
1901 struct mem_cgroup
*wake_memcg
= (struct mem_cgroup
*)arg
,
1903 struct oom_wait_info
*oom_wait_info
;
1905 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1906 oom_wait_memcg
= oom_wait_info
->mem
;
1909 * Both of oom_wait_info->mem and wake_mem are stable under us.
1910 * Then we can use css_is_ancestor without taking care of RCU.
1912 if (!mem_cgroup_same_or_subtree(oom_wait_memcg
, wake_memcg
)
1913 && !mem_cgroup_same_or_subtree(wake_memcg
, oom_wait_memcg
))
1915 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1918 static void memcg_wakeup_oom(struct mem_cgroup
*memcg
)
1920 /* for filtering, pass "memcg" as argument. */
1921 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, memcg
);
1924 static void memcg_oom_recover(struct mem_cgroup
*memcg
)
1926 if (memcg
&& atomic_read(&memcg
->under_oom
))
1927 memcg_wakeup_oom(memcg
);
1931 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1933 bool mem_cgroup_handle_oom(struct mem_cgroup
*memcg
, gfp_t mask
)
1935 struct oom_wait_info owait
;
1936 bool locked
, need_to_kill
;
1939 owait
.wait
.flags
= 0;
1940 owait
.wait
.func
= memcg_oom_wake_function
;
1941 owait
.wait
.private = current
;
1942 INIT_LIST_HEAD(&owait
.wait
.task_list
);
1943 need_to_kill
= true;
1944 mem_cgroup_mark_under_oom(memcg
);
1946 /* At first, try to OOM lock hierarchy under memcg.*/
1947 spin_lock(&memcg_oom_lock
);
1948 locked
= mem_cgroup_oom_lock(memcg
);
1950 * Even if signal_pending(), we can't quit charge() loop without
1951 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1952 * under OOM is always welcomed, use TASK_KILLABLE here.
1954 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1955 if (!locked
|| memcg
->oom_kill_disable
)
1956 need_to_kill
= false;
1958 mem_cgroup_oom_notify(memcg
);
1959 spin_unlock(&memcg_oom_lock
);
1962 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1963 mem_cgroup_out_of_memory(memcg
, mask
);
1966 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1968 spin_lock(&memcg_oom_lock
);
1970 mem_cgroup_oom_unlock(memcg
);
1971 memcg_wakeup_oom(memcg
);
1972 spin_unlock(&memcg_oom_lock
);
1974 mem_cgroup_unmark_under_oom(memcg
);
1976 if (test_thread_flag(TIF_MEMDIE
) || fatal_signal_pending(current
))
1978 /* Give chance to dying process */
1979 schedule_timeout_uninterruptible(1);
1984 * Currently used to update mapped file statistics, but the routine can be
1985 * generalized to update other statistics as well.
1987 * Notes: Race condition
1989 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1990 * it tends to be costly. But considering some conditions, we doesn't need
1991 * to do so _always_.
1993 * Considering "charge", lock_page_cgroup() is not required because all
1994 * file-stat operations happen after a page is attached to radix-tree. There
1995 * are no race with "charge".
1997 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1998 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1999 * if there are race with "uncharge". Statistics itself is properly handled
2002 * Considering "move", this is an only case we see a race. To make the race
2003 * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are
2004 * possibility of race condition. If there is, we take a lock.
2007 void mem_cgroup_update_page_stat(struct page
*page
,
2008 enum mem_cgroup_page_stat_item idx
, int val
)
2010 struct mem_cgroup
*memcg
;
2011 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
2012 bool need_unlock
= false;
2013 unsigned long uninitialized_var(flags
);
2019 memcg
= pc
->mem_cgroup
;
2020 if (unlikely(!memcg
|| !PageCgroupUsed(pc
)))
2022 /* pc->mem_cgroup is unstable ? */
2023 if (unlikely(mem_cgroup_stealed(memcg
)) || PageTransHuge(page
)) {
2024 /* take a lock against to access pc->mem_cgroup */
2025 move_lock_page_cgroup(pc
, &flags
);
2027 memcg
= pc
->mem_cgroup
;
2028 if (!memcg
|| !PageCgroupUsed(pc
))
2033 case MEMCG_NR_FILE_MAPPED
:
2035 SetPageCgroupFileMapped(pc
);
2036 else if (!page_mapped(page
))
2037 ClearPageCgroupFileMapped(pc
);
2038 idx
= MEM_CGROUP_STAT_FILE_MAPPED
;
2044 this_cpu_add(memcg
->stat
->count
[idx
], val
);
2047 if (unlikely(need_unlock
))
2048 move_unlock_page_cgroup(pc
, &flags
);
2052 EXPORT_SYMBOL(mem_cgroup_update_page_stat
);
2055 * size of first charge trial. "32" comes from vmscan.c's magic value.
2056 * TODO: maybe necessary to use big numbers in big irons.
2058 #define CHARGE_BATCH 32U
2059 struct memcg_stock_pcp
{
2060 struct mem_cgroup
*cached
; /* this never be root cgroup */
2061 unsigned int nr_pages
;
2062 struct work_struct work
;
2063 unsigned long flags
;
2064 #define FLUSHING_CACHED_CHARGE (0)
2066 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
2067 static DEFINE_MUTEX(percpu_charge_mutex
);
2070 * Try to consume stocked charge on this cpu. If success, one page is consumed
2071 * from local stock and true is returned. If the stock is 0 or charges from a
2072 * cgroup which is not current target, returns false. This stock will be
2075 static bool consume_stock(struct mem_cgroup
*memcg
)
2077 struct memcg_stock_pcp
*stock
;
2080 stock
= &get_cpu_var(memcg_stock
);
2081 if (memcg
== stock
->cached
&& stock
->nr_pages
)
2083 else /* need to call res_counter_charge */
2085 put_cpu_var(memcg_stock
);
2090 * Returns stocks cached in percpu to res_counter and reset cached information.
2092 static void drain_stock(struct memcg_stock_pcp
*stock
)
2094 struct mem_cgroup
*old
= stock
->cached
;
2096 if (stock
->nr_pages
) {
2097 unsigned long bytes
= stock
->nr_pages
* PAGE_SIZE
;
2099 res_counter_uncharge(&old
->res
, bytes
);
2100 if (do_swap_account
)
2101 res_counter_uncharge(&old
->memsw
, bytes
);
2102 stock
->nr_pages
= 0;
2104 stock
->cached
= NULL
;
2108 * This must be called under preempt disabled or must be called by
2109 * a thread which is pinned to local cpu.
2111 static void drain_local_stock(struct work_struct
*dummy
)
2113 struct memcg_stock_pcp
*stock
= &__get_cpu_var(memcg_stock
);
2115 clear_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
);
2119 * Cache charges(val) which is from res_counter, to local per_cpu area.
2120 * This will be consumed by consume_stock() function, later.
2122 static void refill_stock(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
2124 struct memcg_stock_pcp
*stock
= &get_cpu_var(memcg_stock
);
2126 if (stock
->cached
!= memcg
) { /* reset if necessary */
2128 stock
->cached
= memcg
;
2130 stock
->nr_pages
+= nr_pages
;
2131 put_cpu_var(memcg_stock
);
2135 * Drains all per-CPU charge caches for given root_memcg resp. subtree
2136 * of the hierarchy under it. sync flag says whether we should block
2137 * until the work is done.
2139 static void drain_all_stock(struct mem_cgroup
*root_memcg
, bool sync
)
2143 /* Notify other cpus that system-wide "drain" is running */
2146 for_each_online_cpu(cpu
) {
2147 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2148 struct mem_cgroup
*memcg
;
2150 memcg
= stock
->cached
;
2151 if (!memcg
|| !stock
->nr_pages
)
2153 if (!mem_cgroup_same_or_subtree(root_memcg
, memcg
))
2155 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
)) {
2157 drain_local_stock(&stock
->work
);
2159 schedule_work_on(cpu
, &stock
->work
);
2167 for_each_online_cpu(cpu
) {
2168 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2169 if (test_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
))
2170 flush_work(&stock
->work
);
2177 * Tries to drain stocked charges in other cpus. This function is asynchronous
2178 * and just put a work per cpu for draining localy on each cpu. Caller can
2179 * expects some charges will be back to res_counter later but cannot wait for
2182 static void drain_all_stock_async(struct mem_cgroup
*root_memcg
)
2185 * If someone calls draining, avoid adding more kworker runs.
2187 if (!mutex_trylock(&percpu_charge_mutex
))
2189 drain_all_stock(root_memcg
, false);
2190 mutex_unlock(&percpu_charge_mutex
);
2193 /* This is a synchronous drain interface. */
2194 static void drain_all_stock_sync(struct mem_cgroup
*root_memcg
)
2196 /* called when force_empty is called */
2197 mutex_lock(&percpu_charge_mutex
);
2198 drain_all_stock(root_memcg
, true);
2199 mutex_unlock(&percpu_charge_mutex
);
2203 * This function drains percpu counter value from DEAD cpu and
2204 * move it to local cpu. Note that this function can be preempted.
2206 static void mem_cgroup_drain_pcp_counter(struct mem_cgroup
*memcg
, int cpu
)
2210 spin_lock(&memcg
->pcp_counter_lock
);
2211 for (i
= 0; i
< MEM_CGROUP_STAT_DATA
; i
++) {
2212 long x
= per_cpu(memcg
->stat
->count
[i
], cpu
);
2214 per_cpu(memcg
->stat
->count
[i
], cpu
) = 0;
2215 memcg
->nocpu_base
.count
[i
] += x
;
2217 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
2218 unsigned long x
= per_cpu(memcg
->stat
->events
[i
], cpu
);
2220 per_cpu(memcg
->stat
->events
[i
], cpu
) = 0;
2221 memcg
->nocpu_base
.events
[i
] += x
;
2223 /* need to clear ON_MOVE value, works as a kind of lock. */
2224 per_cpu(memcg
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) = 0;
2225 spin_unlock(&memcg
->pcp_counter_lock
);
2228 static void synchronize_mem_cgroup_on_move(struct mem_cgroup
*memcg
, int cpu
)
2230 int idx
= MEM_CGROUP_ON_MOVE
;
2232 spin_lock(&memcg
->pcp_counter_lock
);
2233 per_cpu(memcg
->stat
->count
[idx
], cpu
) = memcg
->nocpu_base
.count
[idx
];
2234 spin_unlock(&memcg
->pcp_counter_lock
);
2237 static int __cpuinit
memcg_cpu_hotplug_callback(struct notifier_block
*nb
,
2238 unsigned long action
,
2241 int cpu
= (unsigned long)hcpu
;
2242 struct memcg_stock_pcp
*stock
;
2243 struct mem_cgroup
*iter
;
2245 if ((action
== CPU_ONLINE
)) {
2246 for_each_mem_cgroup_all(iter
)
2247 synchronize_mem_cgroup_on_move(iter
, cpu
);
2251 if ((action
!= CPU_DEAD
) || action
!= CPU_DEAD_FROZEN
)
2254 for_each_mem_cgroup_all(iter
)
2255 mem_cgroup_drain_pcp_counter(iter
, cpu
);
2257 stock
= &per_cpu(memcg_stock
, cpu
);
2263 /* See __mem_cgroup_try_charge() for details */
2265 CHARGE_OK
, /* success */
2266 CHARGE_RETRY
, /* need to retry but retry is not bad */
2267 CHARGE_NOMEM
, /* we can't do more. return -ENOMEM */
2268 CHARGE_WOULDBLOCK
, /* GFP_WAIT wasn't set and no enough res. */
2269 CHARGE_OOM_DIE
, /* the current is killed because of OOM */
2272 static int mem_cgroup_do_charge(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
2273 unsigned int nr_pages
, bool oom_check
)
2275 unsigned long csize
= nr_pages
* PAGE_SIZE
;
2276 struct mem_cgroup
*mem_over_limit
;
2277 struct res_counter
*fail_res
;
2278 unsigned long flags
= 0;
2281 ret
= res_counter_charge(&memcg
->res
, csize
, &fail_res
);
2284 if (!do_swap_account
)
2286 ret
= res_counter_charge(&memcg
->memsw
, csize
, &fail_res
);
2290 res_counter_uncharge(&memcg
->res
, csize
);
2291 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, memsw
);
2292 flags
|= MEM_CGROUP_RECLAIM_NOSWAP
;
2294 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, res
);
2296 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2297 * of regular pages (CHARGE_BATCH), or a single regular page (1).
2299 * Never reclaim on behalf of optional batching, retry with a
2300 * single page instead.
2302 if (nr_pages
== CHARGE_BATCH
)
2303 return CHARGE_RETRY
;
2305 if (!(gfp_mask
& __GFP_WAIT
))
2306 return CHARGE_WOULDBLOCK
;
2308 ret
= mem_cgroup_hierarchical_reclaim(mem_over_limit
, NULL
,
2309 gfp_mask
, flags
, NULL
);
2310 if (mem_cgroup_margin(mem_over_limit
) >= nr_pages
)
2311 return CHARGE_RETRY
;
2313 * Even though the limit is exceeded at this point, reclaim
2314 * may have been able to free some pages. Retry the charge
2315 * before killing the task.
2317 * Only for regular pages, though: huge pages are rather
2318 * unlikely to succeed so close to the limit, and we fall back
2319 * to regular pages anyway in case of failure.
2321 if (nr_pages
== 1 && ret
)
2322 return CHARGE_RETRY
;
2325 * At task move, charge accounts can be doubly counted. So, it's
2326 * better to wait until the end of task_move if something is going on.
2328 if (mem_cgroup_wait_acct_move(mem_over_limit
))
2329 return CHARGE_RETRY
;
2331 /* If we don't need to call oom-killer at el, return immediately */
2333 return CHARGE_NOMEM
;
2335 if (!mem_cgroup_handle_oom(mem_over_limit
, gfp_mask
))
2336 return CHARGE_OOM_DIE
;
2338 return CHARGE_RETRY
;
2342 * Unlike exported interface, "oom" parameter is added. if oom==true,
2343 * oom-killer can be invoked.
2345 static int __mem_cgroup_try_charge(struct mm_struct
*mm
,
2347 unsigned int nr_pages
,
2348 struct mem_cgroup
**ptr
,
2351 unsigned int batch
= max(CHARGE_BATCH
, nr_pages
);
2352 int nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2353 struct mem_cgroup
*memcg
= NULL
;
2357 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2358 * in system level. So, allow to go ahead dying process in addition to
2361 if (unlikely(test_thread_flag(TIF_MEMDIE
)
2362 || fatal_signal_pending(current
)))
2366 * We always charge the cgroup the mm_struct belongs to.
2367 * The mm_struct's mem_cgroup changes on task migration if the
2368 * thread group leader migrates. It's possible that mm is not
2369 * set, if so charge the init_mm (happens for pagecache usage).
2374 if (*ptr
) { /* css should be a valid one */
2376 VM_BUG_ON(css_is_removed(&memcg
->css
));
2377 if (mem_cgroup_is_root(memcg
))
2379 if (nr_pages
== 1 && consume_stock(memcg
))
2381 css_get(&memcg
->css
);
2383 struct task_struct
*p
;
2386 p
= rcu_dereference(mm
->owner
);
2388 * Because we don't have task_lock(), "p" can exit.
2389 * In that case, "memcg" can point to root or p can be NULL with
2390 * race with swapoff. Then, we have small risk of mis-accouning.
2391 * But such kind of mis-account by race always happens because
2392 * we don't have cgroup_mutex(). It's overkill and we allo that
2394 * (*) swapoff at el will charge against mm-struct not against
2395 * task-struct. So, mm->owner can be NULL.
2397 memcg
= mem_cgroup_from_task(p
);
2398 if (!memcg
|| mem_cgroup_is_root(memcg
)) {
2402 if (nr_pages
== 1 && consume_stock(memcg
)) {
2404 * It seems dagerous to access memcg without css_get().
2405 * But considering how consume_stok works, it's not
2406 * necessary. If consume_stock success, some charges
2407 * from this memcg are cached on this cpu. So, we
2408 * don't need to call css_get()/css_tryget() before
2409 * calling consume_stock().
2414 /* after here, we may be blocked. we need to get refcnt */
2415 if (!css_tryget(&memcg
->css
)) {
2425 /* If killed, bypass charge */
2426 if (fatal_signal_pending(current
)) {
2427 css_put(&memcg
->css
);
2432 if (oom
&& !nr_oom_retries
) {
2434 nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2437 ret
= mem_cgroup_do_charge(memcg
, gfp_mask
, batch
, oom_check
);
2441 case CHARGE_RETRY
: /* not in OOM situation but retry */
2443 css_put(&memcg
->css
);
2446 case CHARGE_WOULDBLOCK
: /* !__GFP_WAIT */
2447 css_put(&memcg
->css
);
2449 case CHARGE_NOMEM
: /* OOM routine works */
2451 css_put(&memcg
->css
);
2454 /* If oom, we never return -ENOMEM */
2457 case CHARGE_OOM_DIE
: /* Killed by OOM Killer */
2458 css_put(&memcg
->css
);
2461 } while (ret
!= CHARGE_OK
);
2463 if (batch
> nr_pages
)
2464 refill_stock(memcg
, batch
- nr_pages
);
2465 css_put(&memcg
->css
);
2478 * Somemtimes we have to undo a charge we got by try_charge().
2479 * This function is for that and do uncharge, put css's refcnt.
2480 * gotten by try_charge().
2482 static void __mem_cgroup_cancel_charge(struct mem_cgroup
*memcg
,
2483 unsigned int nr_pages
)
2485 if (!mem_cgroup_is_root(memcg
)) {
2486 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2488 res_counter_uncharge(&memcg
->res
, bytes
);
2489 if (do_swap_account
)
2490 res_counter_uncharge(&memcg
->memsw
, bytes
);
2495 * A helper function to get mem_cgroup from ID. must be called under
2496 * rcu_read_lock(). The caller must check css_is_removed() or some if
2497 * it's concern. (dropping refcnt from swap can be called against removed
2500 static struct mem_cgroup
*mem_cgroup_lookup(unsigned short id
)
2502 struct cgroup_subsys_state
*css
;
2504 /* ID 0 is unused ID */
2507 css
= css_lookup(&mem_cgroup_subsys
, id
);
2510 return container_of(css
, struct mem_cgroup
, css
);
2513 struct mem_cgroup
*try_get_mem_cgroup_from_page(struct page
*page
)
2515 struct mem_cgroup
*memcg
= NULL
;
2516 struct page_cgroup
*pc
;
2520 VM_BUG_ON(!PageLocked(page
));
2522 pc
= lookup_page_cgroup(page
);
2523 lock_page_cgroup(pc
);
2524 if (PageCgroupUsed(pc
)) {
2525 memcg
= pc
->mem_cgroup
;
2526 if (memcg
&& !css_tryget(&memcg
->css
))
2528 } else if (PageSwapCache(page
)) {
2529 ent
.val
= page_private(page
);
2530 id
= lookup_swap_cgroup(ent
);
2532 memcg
= mem_cgroup_lookup(id
);
2533 if (memcg
&& !css_tryget(&memcg
->css
))
2537 unlock_page_cgroup(pc
);
2541 static void __mem_cgroup_commit_charge(struct mem_cgroup
*memcg
,
2543 unsigned int nr_pages
,
2544 struct page_cgroup
*pc
,
2545 enum charge_type ctype
)
2547 lock_page_cgroup(pc
);
2548 if (unlikely(PageCgroupUsed(pc
))) {
2549 unlock_page_cgroup(pc
);
2550 __mem_cgroup_cancel_charge(memcg
, nr_pages
);
2554 * we don't need page_cgroup_lock about tail pages, becase they are not
2555 * accessed by any other context at this point.
2557 pc
->mem_cgroup
= memcg
;
2559 * We access a page_cgroup asynchronously without lock_page_cgroup().
2560 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2561 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2562 * before USED bit, we need memory barrier here.
2563 * See mem_cgroup_add_lru_list(), etc.
2567 case MEM_CGROUP_CHARGE_TYPE_CACHE
:
2568 case MEM_CGROUP_CHARGE_TYPE_SHMEM
:
2569 SetPageCgroupCache(pc
);
2570 SetPageCgroupUsed(pc
);
2572 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2573 ClearPageCgroupCache(pc
);
2574 SetPageCgroupUsed(pc
);
2580 mem_cgroup_charge_statistics(memcg
, PageCgroupCache(pc
), nr_pages
);
2581 unlock_page_cgroup(pc
);
2583 * "charge_statistics" updated event counter. Then, check it.
2584 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2585 * if they exceeds softlimit.
2587 memcg_check_events(memcg
, page
);
2590 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2592 #define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\
2593 (1 << PCG_ACCT_LRU) | (1 << PCG_MIGRATION))
2595 * Because tail pages are not marked as "used", set it. We're under
2596 * zone->lru_lock, 'splitting on pmd' and compund_lock.
2598 void mem_cgroup_split_huge_fixup(struct page
*head
, struct page
*tail
)
2600 struct page_cgroup
*head_pc
= lookup_page_cgroup(head
);
2601 struct page_cgroup
*tail_pc
= lookup_page_cgroup(tail
);
2602 unsigned long flags
;
2604 if (mem_cgroup_disabled())
2607 * We have no races with charge/uncharge but will have races with
2608 * page state accounting.
2610 move_lock_page_cgroup(head_pc
, &flags
);
2612 tail_pc
->mem_cgroup
= head_pc
->mem_cgroup
;
2613 smp_wmb(); /* see __commit_charge() */
2614 if (PageCgroupAcctLRU(head_pc
)) {
2616 struct mem_cgroup_per_zone
*mz
;
2619 * LRU flags cannot be copied because we need to add tail
2620 *.page to LRU by generic call and our hook will be called.
2621 * We hold lru_lock, then, reduce counter directly.
2623 lru
= page_lru(head
);
2624 mz
= page_cgroup_zoneinfo(head_pc
->mem_cgroup
, head
);
2625 MEM_CGROUP_ZSTAT(mz
, lru
) -= 1;
2627 tail_pc
->flags
= head_pc
->flags
& ~PCGF_NOCOPY_AT_SPLIT
;
2628 move_unlock_page_cgroup(head_pc
, &flags
);
2633 * mem_cgroup_move_account - move account of the page
2635 * @nr_pages: number of regular pages (>1 for huge pages)
2636 * @pc: page_cgroup of the page.
2637 * @from: mem_cgroup which the page is moved from.
2638 * @to: mem_cgroup which the page is moved to. @from != @to.
2639 * @uncharge: whether we should call uncharge and css_put against @from.
2641 * The caller must confirm following.
2642 * - page is not on LRU (isolate_page() is useful.)
2643 * - compound_lock is held when nr_pages > 1
2645 * This function doesn't do "charge" nor css_get to new cgroup. It should be
2646 * done by a caller(__mem_cgroup_try_charge would be useful). If @uncharge is
2647 * true, this function does "uncharge" from old cgroup, but it doesn't if
2648 * @uncharge is false, so a caller should do "uncharge".
2650 static int mem_cgroup_move_account(struct page
*page
,
2651 unsigned int nr_pages
,
2652 struct page_cgroup
*pc
,
2653 struct mem_cgroup
*from
,
2654 struct mem_cgroup
*to
,
2657 unsigned long flags
;
2660 VM_BUG_ON(from
== to
);
2661 VM_BUG_ON(PageLRU(page
));
2663 * The page is isolated from LRU. So, collapse function
2664 * will not handle this page. But page splitting can happen.
2665 * Do this check under compound_page_lock(). The caller should
2669 if (nr_pages
> 1 && !PageTransHuge(page
))
2672 lock_page_cgroup(pc
);
2675 if (!PageCgroupUsed(pc
) || pc
->mem_cgroup
!= from
)
2678 move_lock_page_cgroup(pc
, &flags
);
2680 if (PageCgroupFileMapped(pc
)) {
2681 /* Update mapped_file data for mem_cgroup */
2683 __this_cpu_dec(from
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2684 __this_cpu_inc(to
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2687 mem_cgroup_charge_statistics(from
, PageCgroupCache(pc
), -nr_pages
);
2689 /* This is not "cancel", but cancel_charge does all we need. */
2690 __mem_cgroup_cancel_charge(from
, nr_pages
);
2692 /* caller should have done css_get */
2693 pc
->mem_cgroup
= to
;
2694 mem_cgroup_charge_statistics(to
, PageCgroupCache(pc
), nr_pages
);
2696 * We charges against "to" which may not have any tasks. Then, "to"
2697 * can be under rmdir(). But in current implementation, caller of
2698 * this function is just force_empty() and move charge, so it's
2699 * guaranteed that "to" is never removed. So, we don't check rmdir
2702 move_unlock_page_cgroup(pc
, &flags
);
2705 unlock_page_cgroup(pc
);
2709 memcg_check_events(to
, page
);
2710 memcg_check_events(from
, page
);
2716 * move charges to its parent.
2719 static int mem_cgroup_move_parent(struct page
*page
,
2720 struct page_cgroup
*pc
,
2721 struct mem_cgroup
*child
,
2724 struct cgroup
*cg
= child
->css
.cgroup
;
2725 struct cgroup
*pcg
= cg
->parent
;
2726 struct mem_cgroup
*parent
;
2727 unsigned int nr_pages
;
2728 unsigned long uninitialized_var(flags
);
2736 if (!get_page_unless_zero(page
))
2738 if (isolate_lru_page(page
))
2741 nr_pages
= hpage_nr_pages(page
);
2743 parent
= mem_cgroup_from_cont(pcg
);
2744 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, nr_pages
, &parent
, false);
2749 flags
= compound_lock_irqsave(page
);
2751 ret
= mem_cgroup_move_account(page
, nr_pages
, pc
, child
, parent
, true);
2753 __mem_cgroup_cancel_charge(parent
, nr_pages
);
2756 compound_unlock_irqrestore(page
, flags
);
2758 putback_lru_page(page
);
2766 * Charge the memory controller for page usage.
2768 * 0 if the charge was successful
2769 * < 0 if the cgroup is over its limit
2771 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
2772 gfp_t gfp_mask
, enum charge_type ctype
)
2774 struct mem_cgroup
*memcg
= NULL
;
2775 unsigned int nr_pages
= 1;
2776 struct page_cgroup
*pc
;
2780 if (PageTransHuge(page
)) {
2781 nr_pages
<<= compound_order(page
);
2782 VM_BUG_ON(!PageTransHuge(page
));
2784 * Never OOM-kill a process for a huge page. The
2785 * fault handler will fall back to regular pages.
2790 pc
= lookup_page_cgroup(page
);
2791 BUG_ON(!pc
); /* XXX: remove this and move pc lookup into commit */
2793 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, nr_pages
, &memcg
, oom
);
2797 __mem_cgroup_commit_charge(memcg
, page
, nr_pages
, pc
, ctype
);
2801 int mem_cgroup_newpage_charge(struct page
*page
,
2802 struct mm_struct
*mm
, gfp_t gfp_mask
)
2804 if (mem_cgroup_disabled())
2807 * If already mapped, we don't have to account.
2808 * If page cache, page->mapping has address_space.
2809 * But page->mapping may have out-of-use anon_vma pointer,
2810 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
2813 if (page_mapped(page
) || (page
->mapping
&& !PageAnon(page
)))
2817 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2818 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2822 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2823 enum charge_type ctype
);
2826 __mem_cgroup_commit_charge_lrucare(struct page
*page
, struct mem_cgroup
*memcg
,
2827 enum charge_type ctype
)
2829 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
2831 * In some case, SwapCache, FUSE(splice_buf->radixtree), the page
2832 * is already on LRU. It means the page may on some other page_cgroup's
2833 * LRU. Take care of it.
2835 mem_cgroup_lru_del_before_commit(page
);
2836 __mem_cgroup_commit_charge(memcg
, page
, 1, pc
, ctype
);
2837 mem_cgroup_lru_add_after_commit(page
);
2841 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
2844 struct mem_cgroup
*memcg
= NULL
;
2847 if (mem_cgroup_disabled())
2849 if (PageCompound(page
))
2855 if (page_is_file_cache(page
)) {
2856 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, 1, &memcg
, true);
2861 * FUSE reuses pages without going through the final
2862 * put that would remove them from the LRU list, make
2863 * sure that they get relinked properly.
2865 __mem_cgroup_commit_charge_lrucare(page
, memcg
,
2866 MEM_CGROUP_CHARGE_TYPE_CACHE
);
2870 if (PageSwapCache(page
)) {
2871 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &memcg
);
2873 __mem_cgroup_commit_charge_swapin(page
, memcg
,
2874 MEM_CGROUP_CHARGE_TYPE_SHMEM
);
2876 ret
= mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2877 MEM_CGROUP_CHARGE_TYPE_SHMEM
);
2883 * While swap-in, try_charge -> commit or cancel, the page is locked.
2884 * And when try_charge() successfully returns, one refcnt to memcg without
2885 * struct page_cgroup is acquired. This refcnt will be consumed by
2886 * "commit()" or removed by "cancel()"
2888 int mem_cgroup_try_charge_swapin(struct mm_struct
*mm
,
2890 gfp_t mask
, struct mem_cgroup
**ptr
)
2892 struct mem_cgroup
*memcg
;
2897 if (mem_cgroup_disabled())
2900 if (!do_swap_account
)
2903 * A racing thread's fault, or swapoff, may have already updated
2904 * the pte, and even removed page from swap cache: in those cases
2905 * do_swap_page()'s pte_same() test will fail; but there's also a
2906 * KSM case which does need to charge the page.
2908 if (!PageSwapCache(page
))
2910 memcg
= try_get_mem_cgroup_from_page(page
);
2914 ret
= __mem_cgroup_try_charge(NULL
, mask
, 1, ptr
, true);
2915 css_put(&memcg
->css
);
2920 return __mem_cgroup_try_charge(mm
, mask
, 1, ptr
, true);
2924 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2925 enum charge_type ctype
)
2927 if (mem_cgroup_disabled())
2931 cgroup_exclude_rmdir(&ptr
->css
);
2933 __mem_cgroup_commit_charge_lrucare(page
, ptr
, ctype
);
2935 * Now swap is on-memory. This means this page may be
2936 * counted both as mem and swap....double count.
2937 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2938 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2939 * may call delete_from_swap_cache() before reach here.
2941 if (do_swap_account
&& PageSwapCache(page
)) {
2942 swp_entry_t ent
= {.val
= page_private(page
)};
2944 struct mem_cgroup
*memcg
;
2946 id
= swap_cgroup_record(ent
, 0);
2948 memcg
= mem_cgroup_lookup(id
);
2951 * This recorded memcg can be obsolete one. So, avoid
2952 * calling css_tryget
2954 if (!mem_cgroup_is_root(memcg
))
2955 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
2956 mem_cgroup_swap_statistics(memcg
, false);
2957 mem_cgroup_put(memcg
);
2962 * At swapin, we may charge account against cgroup which has no tasks.
2963 * So, rmdir()->pre_destroy() can be called while we do this charge.
2964 * In that case, we need to call pre_destroy() again. check it here.
2966 cgroup_release_and_wakeup_rmdir(&ptr
->css
);
2969 void mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
)
2971 __mem_cgroup_commit_charge_swapin(page
, ptr
,
2972 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2975 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup
*memcg
)
2977 if (mem_cgroup_disabled())
2981 __mem_cgroup_cancel_charge(memcg
, 1);
2984 static void mem_cgroup_do_uncharge(struct mem_cgroup
*memcg
,
2985 unsigned int nr_pages
,
2986 const enum charge_type ctype
)
2988 struct memcg_batch_info
*batch
= NULL
;
2989 bool uncharge_memsw
= true;
2991 /* If swapout, usage of swap doesn't decrease */
2992 if (!do_swap_account
|| ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
)
2993 uncharge_memsw
= false;
2995 batch
= ¤t
->memcg_batch
;
2997 * In usual, we do css_get() when we remember memcg pointer.
2998 * But in this case, we keep res->usage until end of a series of
2999 * uncharges. Then, it's ok to ignore memcg's refcnt.
3002 batch
->memcg
= memcg
;
3004 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
3005 * In those cases, all pages freed continuously can be expected to be in
3006 * the same cgroup and we have chance to coalesce uncharges.
3007 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
3008 * because we want to do uncharge as soon as possible.
3011 if (!batch
->do_batch
|| test_thread_flag(TIF_MEMDIE
))
3012 goto direct_uncharge
;
3015 goto direct_uncharge
;
3018 * In typical case, batch->memcg == mem. This means we can
3019 * merge a series of uncharges to an uncharge of res_counter.
3020 * If not, we uncharge res_counter ony by one.
3022 if (batch
->memcg
!= memcg
)
3023 goto direct_uncharge
;
3024 /* remember freed charge and uncharge it later */
3027 batch
->memsw_nr_pages
++;
3030 res_counter_uncharge(&memcg
->res
, nr_pages
* PAGE_SIZE
);
3032 res_counter_uncharge(&memcg
->memsw
, nr_pages
* PAGE_SIZE
);
3033 if (unlikely(batch
->memcg
!= memcg
))
3034 memcg_oom_recover(memcg
);
3039 * uncharge if !page_mapped(page)
3041 static struct mem_cgroup
*
3042 __mem_cgroup_uncharge_common(struct page
*page
, enum charge_type ctype
)
3044 struct mem_cgroup
*memcg
= NULL
;
3045 unsigned int nr_pages
= 1;
3046 struct page_cgroup
*pc
;
3048 if (mem_cgroup_disabled())
3051 if (PageSwapCache(page
))
3054 if (PageTransHuge(page
)) {
3055 nr_pages
<<= compound_order(page
);
3056 VM_BUG_ON(!PageTransHuge(page
));
3059 * Check if our page_cgroup is valid
3061 pc
= lookup_page_cgroup(page
);
3062 if (unlikely(!pc
|| !PageCgroupUsed(pc
)))
3065 lock_page_cgroup(pc
);
3067 memcg
= pc
->mem_cgroup
;
3069 if (!PageCgroupUsed(pc
))
3073 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
3074 case MEM_CGROUP_CHARGE_TYPE_DROP
:
3075 /* See mem_cgroup_prepare_migration() */
3076 if (page_mapped(page
) || PageCgroupMigration(pc
))
3079 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT
:
3080 if (!PageAnon(page
)) { /* Shared memory */
3081 if (page
->mapping
&& !page_is_file_cache(page
))
3083 } else if (page_mapped(page
)) /* Anon */
3090 mem_cgroup_charge_statistics(memcg
, PageCgroupCache(pc
), -nr_pages
);
3092 ClearPageCgroupUsed(pc
);
3094 * pc->mem_cgroup is not cleared here. It will be accessed when it's
3095 * freed from LRU. This is safe because uncharged page is expected not
3096 * to be reused (freed soon). Exception is SwapCache, it's handled by
3097 * special functions.
3100 unlock_page_cgroup(pc
);
3102 * even after unlock, we have memcg->res.usage here and this memcg
3103 * will never be freed.
3105 memcg_check_events(memcg
, page
);
3106 if (do_swap_account
&& ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
) {
3107 mem_cgroup_swap_statistics(memcg
, true);
3108 mem_cgroup_get(memcg
);
3110 if (!mem_cgroup_is_root(memcg
))
3111 mem_cgroup_do_uncharge(memcg
, nr_pages
, ctype
);
3116 unlock_page_cgroup(pc
);
3120 void mem_cgroup_uncharge_page(struct page
*page
)
3123 if (page_mapped(page
))
3125 if (page
->mapping
&& !PageAnon(page
))
3127 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_MAPPED
);
3130 void mem_cgroup_uncharge_cache_page(struct page
*page
)
3132 VM_BUG_ON(page_mapped(page
));
3133 VM_BUG_ON(page
->mapping
);
3134 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_CACHE
);
3138 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
3139 * In that cases, pages are freed continuously and we can expect pages
3140 * are in the same memcg. All these calls itself limits the number of
3141 * pages freed at once, then uncharge_start/end() is called properly.
3142 * This may be called prural(2) times in a context,
3145 void mem_cgroup_uncharge_start(void)
3147 current
->memcg_batch
.do_batch
++;
3148 /* We can do nest. */
3149 if (current
->memcg_batch
.do_batch
== 1) {
3150 current
->memcg_batch
.memcg
= NULL
;
3151 current
->memcg_batch
.nr_pages
= 0;
3152 current
->memcg_batch
.memsw_nr_pages
= 0;
3156 void mem_cgroup_uncharge_end(void)
3158 struct memcg_batch_info
*batch
= ¤t
->memcg_batch
;
3160 if (!batch
->do_batch
)
3164 if (batch
->do_batch
) /* If stacked, do nothing. */
3170 * This "batch->memcg" is valid without any css_get/put etc...
3171 * bacause we hide charges behind us.
3173 if (batch
->nr_pages
)
3174 res_counter_uncharge(&batch
->memcg
->res
,
3175 batch
->nr_pages
* PAGE_SIZE
);
3176 if (batch
->memsw_nr_pages
)
3177 res_counter_uncharge(&batch
->memcg
->memsw
,
3178 batch
->memsw_nr_pages
* PAGE_SIZE
);
3179 memcg_oom_recover(batch
->memcg
);
3180 /* forget this pointer (for sanity check) */
3181 batch
->memcg
= NULL
;
3186 * called after __delete_from_swap_cache() and drop "page" account.
3187 * memcg information is recorded to swap_cgroup of "ent"
3190 mem_cgroup_uncharge_swapcache(struct page
*page
, swp_entry_t ent
, bool swapout
)
3192 struct mem_cgroup
*memcg
;
3193 int ctype
= MEM_CGROUP_CHARGE_TYPE_SWAPOUT
;
3195 if (!swapout
) /* this was a swap cache but the swap is unused ! */
3196 ctype
= MEM_CGROUP_CHARGE_TYPE_DROP
;
3198 memcg
= __mem_cgroup_uncharge_common(page
, ctype
);
3201 * record memcg information, if swapout && memcg != NULL,
3202 * mem_cgroup_get() was called in uncharge().
3204 if (do_swap_account
&& swapout
&& memcg
)
3205 swap_cgroup_record(ent
, css_id(&memcg
->css
));
3209 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3211 * called from swap_entry_free(). remove record in swap_cgroup and
3212 * uncharge "memsw" account.
3214 void mem_cgroup_uncharge_swap(swp_entry_t ent
)
3216 struct mem_cgroup
*memcg
;
3219 if (!do_swap_account
)
3222 id
= swap_cgroup_record(ent
, 0);
3224 memcg
= mem_cgroup_lookup(id
);
3227 * We uncharge this because swap is freed.
3228 * This memcg can be obsolete one. We avoid calling css_tryget
3230 if (!mem_cgroup_is_root(memcg
))
3231 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
3232 mem_cgroup_swap_statistics(memcg
, false);
3233 mem_cgroup_put(memcg
);
3239 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3240 * @entry: swap entry to be moved
3241 * @from: mem_cgroup which the entry is moved from
3242 * @to: mem_cgroup which the entry is moved to
3243 * @need_fixup: whether we should fixup res_counters and refcounts.
3245 * It succeeds only when the swap_cgroup's record for this entry is the same
3246 * as the mem_cgroup's id of @from.
3248 * Returns 0 on success, -EINVAL on failure.
3250 * The caller must have charged to @to, IOW, called res_counter_charge() about
3251 * both res and memsw, and called css_get().
3253 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
3254 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
3256 unsigned short old_id
, new_id
;
3258 old_id
= css_id(&from
->css
);
3259 new_id
= css_id(&to
->css
);
3261 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
3262 mem_cgroup_swap_statistics(from
, false);
3263 mem_cgroup_swap_statistics(to
, true);
3265 * This function is only called from task migration context now.
3266 * It postpones res_counter and refcount handling till the end
3267 * of task migration(mem_cgroup_clear_mc()) for performance
3268 * improvement. But we cannot postpone mem_cgroup_get(to)
3269 * because if the process that has been moved to @to does
3270 * swap-in, the refcount of @to might be decreased to 0.
3274 if (!mem_cgroup_is_root(from
))
3275 res_counter_uncharge(&from
->memsw
, PAGE_SIZE
);
3276 mem_cgroup_put(from
);
3278 * we charged both to->res and to->memsw, so we should
3281 if (!mem_cgroup_is_root(to
))
3282 res_counter_uncharge(&to
->res
, PAGE_SIZE
);
3289 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
3290 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
3297 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3300 int mem_cgroup_prepare_migration(struct page
*page
,
3301 struct page
*newpage
, struct mem_cgroup
**ptr
, gfp_t gfp_mask
)
3303 struct mem_cgroup
*memcg
= NULL
;
3304 struct page_cgroup
*pc
;
3305 enum charge_type ctype
;
3310 VM_BUG_ON(PageTransHuge(page
));
3311 if (mem_cgroup_disabled())
3314 pc
= lookup_page_cgroup(page
);
3315 lock_page_cgroup(pc
);
3316 if (PageCgroupUsed(pc
)) {
3317 memcg
= pc
->mem_cgroup
;
3318 css_get(&memcg
->css
);
3320 * At migrating an anonymous page, its mapcount goes down
3321 * to 0 and uncharge() will be called. But, even if it's fully
3322 * unmapped, migration may fail and this page has to be
3323 * charged again. We set MIGRATION flag here and delay uncharge
3324 * until end_migration() is called
3326 * Corner Case Thinking
3328 * When the old page was mapped as Anon and it's unmap-and-freed
3329 * while migration was ongoing.
3330 * If unmap finds the old page, uncharge() of it will be delayed
3331 * until end_migration(). If unmap finds a new page, it's
3332 * uncharged when it make mapcount to be 1->0. If unmap code
3333 * finds swap_migration_entry, the new page will not be mapped
3334 * and end_migration() will find it(mapcount==0).
3337 * When the old page was mapped but migraion fails, the kernel
3338 * remaps it. A charge for it is kept by MIGRATION flag even
3339 * if mapcount goes down to 0. We can do remap successfully
3340 * without charging it again.
3343 * The "old" page is under lock_page() until the end of
3344 * migration, so, the old page itself will not be swapped-out.
3345 * If the new page is swapped out before end_migraton, our
3346 * hook to usual swap-out path will catch the event.
3349 SetPageCgroupMigration(pc
);
3351 unlock_page_cgroup(pc
);
3353 * If the page is not charged at this point,
3360 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, 1, ptr
, false);
3361 css_put(&memcg
->css
);/* drop extra refcnt */
3362 if (ret
|| *ptr
== NULL
) {
3363 if (PageAnon(page
)) {
3364 lock_page_cgroup(pc
);
3365 ClearPageCgroupMigration(pc
);
3366 unlock_page_cgroup(pc
);
3368 * The old page may be fully unmapped while we kept it.
3370 mem_cgroup_uncharge_page(page
);
3375 * We charge new page before it's used/mapped. So, even if unlock_page()
3376 * is called before end_migration, we can catch all events on this new
3377 * page. In the case new page is migrated but not remapped, new page's
3378 * mapcount will be finally 0 and we call uncharge in end_migration().
3380 pc
= lookup_page_cgroup(newpage
);
3382 ctype
= MEM_CGROUP_CHARGE_TYPE_MAPPED
;
3383 else if (page_is_file_cache(page
))
3384 ctype
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3386 ctype
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
3387 __mem_cgroup_commit_charge(memcg
, page
, 1, pc
, ctype
);
3391 /* remove redundant charge if migration failed*/
3392 void mem_cgroup_end_migration(struct mem_cgroup
*memcg
,
3393 struct page
*oldpage
, struct page
*newpage
, bool migration_ok
)
3395 struct page
*used
, *unused
;
3396 struct page_cgroup
*pc
;
3400 /* blocks rmdir() */
3401 cgroup_exclude_rmdir(&memcg
->css
);
3402 if (!migration_ok
) {
3410 * We disallowed uncharge of pages under migration because mapcount
3411 * of the page goes down to zero, temporarly.
3412 * Clear the flag and check the page should be charged.
3414 pc
= lookup_page_cgroup(oldpage
);
3415 lock_page_cgroup(pc
);
3416 ClearPageCgroupMigration(pc
);
3417 unlock_page_cgroup(pc
);
3419 __mem_cgroup_uncharge_common(unused
, MEM_CGROUP_CHARGE_TYPE_FORCE
);
3422 * If a page is a file cache, radix-tree replacement is very atomic
3423 * and we can skip this check. When it was an Anon page, its mapcount
3424 * goes down to 0. But because we added MIGRATION flage, it's not
3425 * uncharged yet. There are several case but page->mapcount check
3426 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3427 * check. (see prepare_charge() also)
3430 mem_cgroup_uncharge_page(used
);
3432 * At migration, we may charge account against cgroup which has no
3434 * So, rmdir()->pre_destroy() can be called while we do this charge.
3435 * In that case, we need to call pre_destroy() again. check it here.
3437 cgroup_release_and_wakeup_rmdir(&memcg
->css
);
3440 #ifdef CONFIG_DEBUG_VM
3441 static struct page_cgroup
*lookup_page_cgroup_used(struct page
*page
)
3443 struct page_cgroup
*pc
;
3445 pc
= lookup_page_cgroup(page
);
3446 if (likely(pc
) && PageCgroupUsed(pc
))
3451 bool mem_cgroup_bad_page_check(struct page
*page
)
3453 if (mem_cgroup_disabled())
3456 return lookup_page_cgroup_used(page
) != NULL
;
3459 void mem_cgroup_print_bad_page(struct page
*page
)
3461 struct page_cgroup
*pc
;
3463 pc
= lookup_page_cgroup_used(page
);
3468 printk(KERN_ALERT
"pc:%p pc->flags:%lx pc->mem_cgroup:%p",
3469 pc
, pc
->flags
, pc
->mem_cgroup
);
3471 path
= kmalloc(PATH_MAX
, GFP_KERNEL
);
3474 ret
= cgroup_path(pc
->mem_cgroup
->css
.cgroup
,
3479 printk(KERN_CONT
"(%s)\n",
3480 (ret
< 0) ? "cannot get the path" : path
);
3486 static DEFINE_MUTEX(set_limit_mutex
);
3488 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
3489 unsigned long long val
)
3492 u64 memswlimit
, memlimit
;
3494 int children
= mem_cgroup_count_children(memcg
);
3495 u64 curusage
, oldusage
;
3499 * For keeping hierarchical_reclaim simple, how long we should retry
3500 * is depends on callers. We set our retry-count to be function
3501 * of # of children which we should visit in this loop.
3503 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
* children
;
3505 oldusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3508 while (retry_count
) {
3509 if (signal_pending(current
)) {
3514 * Rather than hide all in some function, I do this in
3515 * open coded manner. You see what this really does.
3516 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
3518 mutex_lock(&set_limit_mutex
);
3519 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3520 if (memswlimit
< val
) {
3522 mutex_unlock(&set_limit_mutex
);
3526 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3530 ret
= res_counter_set_limit(&memcg
->res
, val
);
3532 if (memswlimit
== val
)
3533 memcg
->memsw_is_minimum
= true;
3535 memcg
->memsw_is_minimum
= false;
3537 mutex_unlock(&set_limit_mutex
);
3542 mem_cgroup_hierarchical_reclaim(memcg
, NULL
, GFP_KERNEL
,
3543 MEM_CGROUP_RECLAIM_SHRINK
,
3545 curusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3546 /* Usage is reduced ? */
3547 if (curusage
>= oldusage
)
3550 oldusage
= curusage
;
3552 if (!ret
&& enlarge
)
3553 memcg_oom_recover(memcg
);
3558 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
3559 unsigned long long val
)
3562 u64 memlimit
, memswlimit
, oldusage
, curusage
;
3563 int children
= mem_cgroup_count_children(memcg
);
3567 /* see mem_cgroup_resize_res_limit */
3568 retry_count
= children
* MEM_CGROUP_RECLAIM_RETRIES
;
3569 oldusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3570 while (retry_count
) {
3571 if (signal_pending(current
)) {
3576 * Rather than hide all in some function, I do this in
3577 * open coded manner. You see what this really does.
3578 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
3580 mutex_lock(&set_limit_mutex
);
3581 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3582 if (memlimit
> val
) {
3584 mutex_unlock(&set_limit_mutex
);
3587 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3588 if (memswlimit
< val
)
3590 ret
= res_counter_set_limit(&memcg
->memsw
, val
);
3592 if (memlimit
== val
)
3593 memcg
->memsw_is_minimum
= true;
3595 memcg
->memsw_is_minimum
= false;
3597 mutex_unlock(&set_limit_mutex
);
3602 mem_cgroup_hierarchical_reclaim(memcg
, NULL
, GFP_KERNEL
,
3603 MEM_CGROUP_RECLAIM_NOSWAP
|
3604 MEM_CGROUP_RECLAIM_SHRINK
,
3606 curusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3607 /* Usage is reduced ? */
3608 if (curusage
>= oldusage
)
3611 oldusage
= curusage
;
3613 if (!ret
&& enlarge
)
3614 memcg_oom_recover(memcg
);
3618 unsigned long mem_cgroup_soft_limit_reclaim(struct zone
*zone
, int order
,
3620 unsigned long *total_scanned
)
3622 unsigned long nr_reclaimed
= 0;
3623 struct mem_cgroup_per_zone
*mz
, *next_mz
= NULL
;
3624 unsigned long reclaimed
;
3626 struct mem_cgroup_tree_per_zone
*mctz
;
3627 unsigned long long excess
;
3628 unsigned long nr_scanned
;
3633 mctz
= soft_limit_tree_node_zone(zone_to_nid(zone
), zone_idx(zone
));
3635 * This loop can run a while, specially if mem_cgroup's continuously
3636 * keep exceeding their soft limit and putting the system under
3643 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
3648 reclaimed
= mem_cgroup_hierarchical_reclaim(mz
->mem
, zone
,
3650 MEM_CGROUP_RECLAIM_SOFT
,
3652 nr_reclaimed
+= reclaimed
;
3653 *total_scanned
+= nr_scanned
;
3654 spin_lock(&mctz
->lock
);
3657 * If we failed to reclaim anything from this memory cgroup
3658 * it is time to move on to the next cgroup
3664 * Loop until we find yet another one.
3666 * By the time we get the soft_limit lock
3667 * again, someone might have aded the
3668 * group back on the RB tree. Iterate to
3669 * make sure we get a different mem.
3670 * mem_cgroup_largest_soft_limit_node returns
3671 * NULL if no other cgroup is present on
3675 __mem_cgroup_largest_soft_limit_node(mctz
);
3677 css_put(&next_mz
->mem
->css
);
3678 else /* next_mz == NULL or other memcg */
3682 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
3683 excess
= res_counter_soft_limit_excess(&mz
->mem
->res
);
3685 * One school of thought says that we should not add
3686 * back the node to the tree if reclaim returns 0.
3687 * But our reclaim could return 0, simply because due
3688 * to priority we are exposing a smaller subset of
3689 * memory to reclaim from. Consider this as a longer
3692 /* If excess == 0, no tree ops */
3693 __mem_cgroup_insert_exceeded(mz
->mem
, mz
, mctz
, excess
);
3694 spin_unlock(&mctz
->lock
);
3695 css_put(&mz
->mem
->css
);
3698 * Could not reclaim anything and there are no more
3699 * mem cgroups to try or we seem to be looping without
3700 * reclaiming anything.
3702 if (!nr_reclaimed
&&
3704 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
3706 } while (!nr_reclaimed
);
3708 css_put(&next_mz
->mem
->css
);
3709 return nr_reclaimed
;
3713 * This routine traverse page_cgroup in given list and drop them all.
3714 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3716 static int mem_cgroup_force_empty_list(struct mem_cgroup
*memcg
,
3717 int node
, int zid
, enum lru_list lru
)
3720 struct mem_cgroup_per_zone
*mz
;
3721 struct page_cgroup
*pc
, *busy
;
3722 unsigned long flags
, loop
;
3723 struct list_head
*list
;
3726 zone
= &NODE_DATA(node
)->node_zones
[zid
];
3727 mz
= mem_cgroup_zoneinfo(memcg
, node
, zid
);
3728 list
= &mz
->lists
[lru
];
3730 loop
= MEM_CGROUP_ZSTAT(mz
, lru
);
3731 /* give some margin against EBUSY etc...*/
3738 spin_lock_irqsave(&zone
->lru_lock
, flags
);
3739 if (list_empty(list
)) {
3740 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3743 pc
= list_entry(list
->prev
, struct page_cgroup
, lru
);
3745 list_move(&pc
->lru
, list
);
3747 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3750 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3752 page
= lookup_cgroup_page(pc
);
3754 ret
= mem_cgroup_move_parent(page
, pc
, memcg
, GFP_KERNEL
);
3758 if (ret
== -EBUSY
|| ret
== -EINVAL
) {
3759 /* found lock contention or "pc" is obsolete. */
3766 if (!ret
&& !list_empty(list
))
3772 * make mem_cgroup's charge to be 0 if there is no task.
3773 * This enables deleting this mem_cgroup.
3775 static int mem_cgroup_force_empty(struct mem_cgroup
*memcg
, bool free_all
)
3778 int node
, zid
, shrink
;
3779 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
3780 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
3782 css_get(&memcg
->css
);
3785 /* should free all ? */
3791 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
))
3794 if (signal_pending(current
))
3796 /* This is for making all *used* pages to be on LRU. */
3797 lru_add_drain_all();
3798 drain_all_stock_sync(memcg
);
3800 mem_cgroup_start_move(memcg
);
3801 for_each_node_state(node
, N_HIGH_MEMORY
) {
3802 for (zid
= 0; !ret
&& zid
< MAX_NR_ZONES
; zid
++) {
3805 ret
= mem_cgroup_force_empty_list(memcg
,
3814 mem_cgroup_end_move(memcg
);
3815 memcg_oom_recover(memcg
);
3816 /* it seems parent cgroup doesn't have enough mem */
3820 /* "ret" should also be checked to ensure all lists are empty. */
3821 } while (memcg
->res
.usage
> 0 || ret
);
3823 css_put(&memcg
->css
);
3827 /* returns EBUSY if there is a task or if we come here twice. */
3828 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
) || shrink
) {
3832 /* we call try-to-free pages for make this cgroup empty */
3833 lru_add_drain_all();
3834 /* try to free all pages in this cgroup */
3836 while (nr_retries
&& memcg
->res
.usage
> 0) {
3839 if (signal_pending(current
)) {
3843 progress
= try_to_free_mem_cgroup_pages(memcg
, GFP_KERNEL
,
3847 /* maybe some writeback is necessary */
3848 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3853 /* try move_account...there may be some *locked* pages. */
3857 int mem_cgroup_force_empty_write(struct cgroup
*cont
, unsigned int event
)
3859 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont
), true);
3863 static u64
mem_cgroup_hierarchy_read(struct cgroup
*cont
, struct cftype
*cft
)
3865 return mem_cgroup_from_cont(cont
)->use_hierarchy
;
3868 static int mem_cgroup_hierarchy_write(struct cgroup
*cont
, struct cftype
*cft
,
3872 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3873 struct cgroup
*parent
= cont
->parent
;
3874 struct mem_cgroup
*parent_memcg
= NULL
;
3877 parent_memcg
= mem_cgroup_from_cont(parent
);
3881 * If parent's use_hierarchy is set, we can't make any modifications
3882 * in the child subtrees. If it is unset, then the change can
3883 * occur, provided the current cgroup has no children.
3885 * For the root cgroup, parent_mem is NULL, we allow value to be
3886 * set if there are no children.
3888 if ((!parent_memcg
|| !parent_memcg
->use_hierarchy
) &&
3889 (val
== 1 || val
== 0)) {
3890 if (list_empty(&cont
->children
))
3891 memcg
->use_hierarchy
= val
;
3902 static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup
*memcg
,
3903 enum mem_cgroup_stat_index idx
)
3905 struct mem_cgroup
*iter
;
3908 /* Per-cpu values can be negative, use a signed accumulator */
3909 for_each_mem_cgroup_tree(iter
, memcg
)
3910 val
+= mem_cgroup_read_stat(iter
, idx
);
3912 if (val
< 0) /* race ? */
3917 static inline u64
mem_cgroup_usage(struct mem_cgroup
*memcg
, bool swap
)
3921 if (!mem_cgroup_is_root(memcg
)) {
3923 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
3924 if (!memcg
->kmem_independent_accounting
)
3925 val
= res_counter_read_u64(&memcg
->kmem
, RES_USAGE
);
3928 val
+= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3930 val
+= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3935 val
= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_CACHE
);
3936 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_RSS
);
3939 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_SWAPOUT
);
3941 return val
<< PAGE_SHIFT
;
3944 static u64
mem_cgroup_read(struct cgroup
*cont
, struct cftype
*cft
)
3946 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3950 type
= MEMFILE_TYPE(cft
->private);
3951 name
= MEMFILE_ATTR(cft
->private);
3954 if (name
== RES_USAGE
)
3955 val
= mem_cgroup_usage(memcg
, false);
3957 val
= res_counter_read_u64(&memcg
->res
, name
);
3960 if (name
== RES_USAGE
)
3961 val
= mem_cgroup_usage(memcg
, true);
3963 val
= res_counter_read_u64(&memcg
->memsw
, name
);
3965 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
3967 val
= res_counter_read_u64(&memcg
->kmem
, name
);
3977 * The user of this function is...
3980 static int mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
3983 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3985 unsigned long long val
;
3988 type
= MEMFILE_TYPE(cft
->private);
3989 name
= MEMFILE_ATTR(cft
->private);
3992 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
3996 /* This function does all necessary parse...reuse it */
3997 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
4001 ret
= mem_cgroup_resize_limit(memcg
, val
);
4003 ret
= mem_cgroup_resize_memsw_limit(memcg
, val
);
4005 case RES_SOFT_LIMIT
:
4006 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
4010 * For memsw, soft limits are hard to implement in terms
4011 * of semantics, for now, we support soft limits for
4012 * control without swap
4015 ret
= res_counter_set_soft_limit(&memcg
->res
, val
);
4020 ret
= -EINVAL
; /* should be BUG() ? */
4026 static void memcg_get_hierarchical_limit(struct mem_cgroup
*memcg
,
4027 unsigned long long *mem_limit
, unsigned long long *memsw_limit
)
4029 struct cgroup
*cgroup
;
4030 unsigned long long min_limit
, min_memsw_limit
, tmp
;
4032 min_limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
4033 min_memsw_limit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
4034 cgroup
= memcg
->css
.cgroup
;
4035 if (!memcg
->use_hierarchy
)
4038 while (cgroup
->parent
) {
4039 cgroup
= cgroup
->parent
;
4040 memcg
= mem_cgroup_from_cont(cgroup
);
4041 if (!memcg
->use_hierarchy
)
4043 tmp
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
4044 min_limit
= min(min_limit
, tmp
);
4045 tmp
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
4046 min_memsw_limit
= min(min_memsw_limit
, tmp
);
4049 *mem_limit
= min_limit
;
4050 *memsw_limit
= min_memsw_limit
;
4054 static int mem_cgroup_reset(struct cgroup
*cont
, unsigned int event
)
4056 struct mem_cgroup
*memcg
;
4059 memcg
= mem_cgroup_from_cont(cont
);
4060 type
= MEMFILE_TYPE(event
);
4061 name
= MEMFILE_ATTR(event
);
4065 res_counter_reset_max(&memcg
->res
);
4067 res_counter_reset_max(&memcg
->memsw
);
4071 res_counter_reset_failcnt(&memcg
->res
);
4073 res_counter_reset_failcnt(&memcg
->memsw
);
4080 static u64
mem_cgroup_move_charge_read(struct cgroup
*cgrp
,
4083 return mem_cgroup_from_cont(cgrp
)->move_charge_at_immigrate
;
4087 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4088 struct cftype
*cft
, u64 val
)
4090 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4092 if (val
>= (1 << NR_MOVE_TYPE
))
4095 * We check this value several times in both in can_attach() and
4096 * attach(), so we need cgroup lock to prevent this value from being
4100 memcg
->move_charge_at_immigrate
= val
;
4106 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4107 struct cftype
*cft
, u64 val
)
4114 /* For read statistics */
4132 struct mcs_total_stat
{
4133 s64 stat
[NR_MCS_STAT
];
4139 } memcg_stat_strings
[NR_MCS_STAT
] = {
4140 {"cache", "total_cache"},
4141 {"rss", "total_rss"},
4142 {"mapped_file", "total_mapped_file"},
4143 {"pgpgin", "total_pgpgin"},
4144 {"pgpgout", "total_pgpgout"},
4145 {"swap", "total_swap"},
4146 {"pgfault", "total_pgfault"},
4147 {"pgmajfault", "total_pgmajfault"},
4148 {"inactive_anon", "total_inactive_anon"},
4149 {"active_anon", "total_active_anon"},
4150 {"inactive_file", "total_inactive_file"},
4151 {"active_file", "total_active_file"},
4152 {"unevictable", "total_unevictable"}
4157 mem_cgroup_get_local_stat(struct mem_cgroup
*memcg
, struct mcs_total_stat
*s
)
4162 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_CACHE
);
4163 s
->stat
[MCS_CACHE
] += val
* PAGE_SIZE
;
4164 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_RSS
);
4165 s
->stat
[MCS_RSS
] += val
* PAGE_SIZE
;
4166 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_FILE_MAPPED
);
4167 s
->stat
[MCS_FILE_MAPPED
] += val
* PAGE_SIZE
;
4168 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGPGIN
);
4169 s
->stat
[MCS_PGPGIN
] += val
;
4170 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGPGOUT
);
4171 s
->stat
[MCS_PGPGOUT
] += val
;
4172 if (do_swap_account
) {
4173 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_SWAPOUT
);
4174 s
->stat
[MCS_SWAP
] += val
* PAGE_SIZE
;
4176 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGFAULT
);
4177 s
->stat
[MCS_PGFAULT
] += val
;
4178 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGMAJFAULT
);
4179 s
->stat
[MCS_PGMAJFAULT
] += val
;
4182 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_INACTIVE_ANON
));
4183 s
->stat
[MCS_INACTIVE_ANON
] += val
* PAGE_SIZE
;
4184 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_ACTIVE_ANON
));
4185 s
->stat
[MCS_ACTIVE_ANON
] += val
* PAGE_SIZE
;
4186 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_INACTIVE_FILE
));
4187 s
->stat
[MCS_INACTIVE_FILE
] += val
* PAGE_SIZE
;
4188 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_ACTIVE_FILE
));
4189 s
->stat
[MCS_ACTIVE_FILE
] += val
* PAGE_SIZE
;
4190 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_UNEVICTABLE
));
4191 s
->stat
[MCS_UNEVICTABLE
] += val
* PAGE_SIZE
;
4195 mem_cgroup_get_total_stat(struct mem_cgroup
*memcg
, struct mcs_total_stat
*s
)
4197 struct mem_cgroup
*iter
;
4199 for_each_mem_cgroup_tree(iter
, memcg
)
4200 mem_cgroup_get_local_stat(iter
, s
);
4204 static int mem_control_numa_stat_show(struct seq_file
*m
, void *arg
)
4207 unsigned long total_nr
, file_nr
, anon_nr
, unevictable_nr
;
4208 unsigned long node_nr
;
4209 struct cgroup
*cont
= m
->private;
4210 struct mem_cgroup
*mem_cont
= mem_cgroup_from_cont(cont
);
4212 total_nr
= mem_cgroup_nr_lru_pages(mem_cont
, LRU_ALL
);
4213 seq_printf(m
, "total=%lu", total_nr
);
4214 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4215 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
, LRU_ALL
);
4216 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4220 file_nr
= mem_cgroup_nr_lru_pages(mem_cont
, LRU_ALL_FILE
);
4221 seq_printf(m
, "file=%lu", file_nr
);
4222 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4223 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
,
4225 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4229 anon_nr
= mem_cgroup_nr_lru_pages(mem_cont
, LRU_ALL_ANON
);
4230 seq_printf(m
, "anon=%lu", anon_nr
);
4231 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4232 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
,
4234 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4238 unevictable_nr
= mem_cgroup_nr_lru_pages(mem_cont
, BIT(LRU_UNEVICTABLE
));
4239 seq_printf(m
, "unevictable=%lu", unevictable_nr
);
4240 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4241 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
,
4242 BIT(LRU_UNEVICTABLE
));
4243 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4248 #endif /* CONFIG_NUMA */
4250 static int mem_control_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
4251 struct cgroup_map_cb
*cb
)
4253 struct mem_cgroup
*mem_cont
= mem_cgroup_from_cont(cont
);
4254 struct mcs_total_stat mystat
;
4257 memset(&mystat
, 0, sizeof(mystat
));
4258 mem_cgroup_get_local_stat(mem_cont
, &mystat
);
4261 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
4262 if (i
== MCS_SWAP
&& !do_swap_account
)
4264 cb
->fill(cb
, memcg_stat_strings
[i
].local_name
, mystat
.stat
[i
]);
4267 /* Hierarchical information */
4269 unsigned long long limit
, memsw_limit
;
4270 memcg_get_hierarchical_limit(mem_cont
, &limit
, &memsw_limit
);
4271 cb
->fill(cb
, "hierarchical_memory_limit", limit
);
4272 if (do_swap_account
)
4273 cb
->fill(cb
, "hierarchical_memsw_limit", memsw_limit
);
4276 memset(&mystat
, 0, sizeof(mystat
));
4277 mem_cgroup_get_total_stat(mem_cont
, &mystat
);
4278 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
4279 if (i
== MCS_SWAP
&& !do_swap_account
)
4281 cb
->fill(cb
, memcg_stat_strings
[i
].total_name
, mystat
.stat
[i
]);
4284 #ifdef CONFIG_DEBUG_VM
4287 struct mem_cgroup_per_zone
*mz
;
4288 unsigned long recent_rotated
[2] = {0, 0};
4289 unsigned long recent_scanned
[2] = {0, 0};
4291 for_each_online_node(nid
)
4292 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
4293 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
4295 recent_rotated
[0] +=
4296 mz
->reclaim_stat
.recent_rotated
[0];
4297 recent_rotated
[1] +=
4298 mz
->reclaim_stat
.recent_rotated
[1];
4299 recent_scanned
[0] +=
4300 mz
->reclaim_stat
.recent_scanned
[0];
4301 recent_scanned
[1] +=
4302 mz
->reclaim_stat
.recent_scanned
[1];
4304 cb
->fill(cb
, "recent_rotated_anon", recent_rotated
[0]);
4305 cb
->fill(cb
, "recent_rotated_file", recent_rotated
[1]);
4306 cb
->fill(cb
, "recent_scanned_anon", recent_scanned
[0]);
4307 cb
->fill(cb
, "recent_scanned_file", recent_scanned
[1]);
4314 static u64
mem_cgroup_swappiness_read(struct cgroup
*cgrp
, struct cftype
*cft
)
4316 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4318 return mem_cgroup_swappiness(memcg
);
4321 static int mem_cgroup_swappiness_write(struct cgroup
*cgrp
, struct cftype
*cft
,
4324 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4325 struct mem_cgroup
*parent
;
4330 if (cgrp
->parent
== NULL
)
4333 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4337 /* If under hierarchy, only empty-root can set this value */
4338 if ((parent
->use_hierarchy
) ||
4339 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4344 memcg
->swappiness
= val
;
4351 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
4353 struct mem_cgroup_threshold_ary
*t
;
4359 t
= rcu_dereference(memcg
->thresholds
.primary
);
4361 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
4366 usage
= mem_cgroup_usage(memcg
, swap
);
4369 * current_threshold points to threshold just below usage.
4370 * If it's not true, a threshold was crossed after last
4371 * call of __mem_cgroup_threshold().
4373 i
= t
->current_threshold
;
4376 * Iterate backward over array of thresholds starting from
4377 * current_threshold and check if a threshold is crossed.
4378 * If none of thresholds below usage is crossed, we read
4379 * only one element of the array here.
4381 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
4382 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4384 /* i = current_threshold + 1 */
4388 * Iterate forward over array of thresholds starting from
4389 * current_threshold+1 and check if a threshold is crossed.
4390 * If none of thresholds above usage is crossed, we read
4391 * only one element of the array here.
4393 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
4394 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4396 /* Update current_threshold */
4397 t
->current_threshold
= i
- 1;
4402 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
4405 __mem_cgroup_threshold(memcg
, false);
4406 if (do_swap_account
)
4407 __mem_cgroup_threshold(memcg
, true);
4409 memcg
= parent_mem_cgroup(memcg
);
4413 static int compare_thresholds(const void *a
, const void *b
)
4415 const struct mem_cgroup_threshold
*_a
= a
;
4416 const struct mem_cgroup_threshold
*_b
= b
;
4418 return _a
->threshold
- _b
->threshold
;
4421 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*memcg
)
4423 struct mem_cgroup_eventfd_list
*ev
;
4425 list_for_each_entry(ev
, &memcg
->oom_notify
, list
)
4426 eventfd_signal(ev
->eventfd
, 1);
4430 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
)
4432 struct mem_cgroup
*iter
;
4434 for_each_mem_cgroup_tree(iter
, memcg
)
4435 mem_cgroup_oom_notify_cb(iter
);
4438 static int mem_cgroup_usage_register_event(struct cgroup
*cgrp
,
4439 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4441 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4442 struct mem_cgroup_thresholds
*thresholds
;
4443 struct mem_cgroup_threshold_ary
*new;
4444 int type
= MEMFILE_TYPE(cft
->private);
4445 u64 threshold
, usage
;
4448 ret
= res_counter_memparse_write_strategy(args
, &threshold
);
4452 mutex_lock(&memcg
->thresholds_lock
);
4455 thresholds
= &memcg
->thresholds
;
4456 else if (type
== _MEMSWAP
)
4457 thresholds
= &memcg
->memsw_thresholds
;
4461 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4463 /* Check if a threshold crossed before adding a new one */
4464 if (thresholds
->primary
)
4465 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4467 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
4469 /* Allocate memory for new array of thresholds */
4470 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
4478 /* Copy thresholds (if any) to new array */
4479 if (thresholds
->primary
) {
4480 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
4481 sizeof(struct mem_cgroup_threshold
));
4484 /* Add new threshold */
4485 new->entries
[size
- 1].eventfd
= eventfd
;
4486 new->entries
[size
- 1].threshold
= threshold
;
4488 /* Sort thresholds. Registering of new threshold isn't time-critical */
4489 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
4490 compare_thresholds
, NULL
);
4492 /* Find current threshold */
4493 new->current_threshold
= -1;
4494 for (i
= 0; i
< size
; i
++) {
4495 if (new->entries
[i
].threshold
< usage
) {
4497 * new->current_threshold will not be used until
4498 * rcu_assign_pointer(), so it's safe to increment
4501 ++new->current_threshold
;
4505 /* Free old spare buffer and save old primary buffer as spare */
4506 kfree(thresholds
->spare
);
4507 thresholds
->spare
= thresholds
->primary
;
4509 rcu_assign_pointer(thresholds
->primary
, new);
4511 /* To be sure that nobody uses thresholds */
4515 mutex_unlock(&memcg
->thresholds_lock
);
4520 static void mem_cgroup_usage_unregister_event(struct cgroup
*cgrp
,
4521 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4523 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4524 struct mem_cgroup_thresholds
*thresholds
;
4525 struct mem_cgroup_threshold_ary
*new;
4526 int type
= MEMFILE_TYPE(cft
->private);
4530 mutex_lock(&memcg
->thresholds_lock
);
4532 thresholds
= &memcg
->thresholds
;
4533 else if (type
== _MEMSWAP
)
4534 thresholds
= &memcg
->memsw_thresholds
;
4539 * Something went wrong if we trying to unregister a threshold
4540 * if we don't have thresholds
4542 BUG_ON(!thresholds
);
4544 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4546 /* Check if a threshold crossed before removing */
4547 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4549 /* Calculate new number of threshold */
4551 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
4552 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
4556 new = thresholds
->spare
;
4558 /* Set thresholds array to NULL if we don't have thresholds */
4567 /* Copy thresholds and find current threshold */
4568 new->current_threshold
= -1;
4569 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
4570 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
4573 new->entries
[j
] = thresholds
->primary
->entries
[i
];
4574 if (new->entries
[j
].threshold
< usage
) {
4576 * new->current_threshold will not be used
4577 * until rcu_assign_pointer(), so it's safe to increment
4580 ++new->current_threshold
;
4586 /* Swap primary and spare array */
4587 thresholds
->spare
= thresholds
->primary
;
4588 rcu_assign_pointer(thresholds
->primary
, new);
4590 /* To be sure that nobody uses thresholds */
4593 mutex_unlock(&memcg
->thresholds_lock
);
4596 static int mem_cgroup_oom_register_event(struct cgroup
*cgrp
,
4597 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4599 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4600 struct mem_cgroup_eventfd_list
*event
;
4601 int type
= MEMFILE_TYPE(cft
->private);
4603 BUG_ON(type
!= _OOM_TYPE
);
4604 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
4608 spin_lock(&memcg_oom_lock
);
4610 event
->eventfd
= eventfd
;
4611 list_add(&event
->list
, &memcg
->oom_notify
);
4613 /* already in OOM ? */
4614 if (atomic_read(&memcg
->under_oom
))
4615 eventfd_signal(eventfd
, 1);
4616 spin_unlock(&memcg_oom_lock
);
4621 static void mem_cgroup_oom_unregister_event(struct cgroup
*cgrp
,
4622 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4624 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4625 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
4626 int type
= MEMFILE_TYPE(cft
->private);
4628 BUG_ON(type
!= _OOM_TYPE
);
4630 spin_lock(&memcg_oom_lock
);
4632 list_for_each_entry_safe(ev
, tmp
, &memcg
->oom_notify
, list
) {
4633 if (ev
->eventfd
== eventfd
) {
4634 list_del(&ev
->list
);
4639 spin_unlock(&memcg_oom_lock
);
4642 static int mem_cgroup_oom_control_read(struct cgroup
*cgrp
,
4643 struct cftype
*cft
, struct cgroup_map_cb
*cb
)
4645 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4647 cb
->fill(cb
, "oom_kill_disable", memcg
->oom_kill_disable
);
4649 if (atomic_read(&memcg
->under_oom
))
4650 cb
->fill(cb
, "under_oom", 1);
4652 cb
->fill(cb
, "under_oom", 0);
4656 static int mem_cgroup_oom_control_write(struct cgroup
*cgrp
,
4657 struct cftype
*cft
, u64 val
)
4659 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4660 struct mem_cgroup
*parent
;
4662 /* cannot set to root cgroup and only 0 and 1 are allowed */
4663 if (!cgrp
->parent
|| !((val
== 0) || (val
== 1)))
4666 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4669 /* oom-kill-disable is a flag for subhierarchy. */
4670 if ((parent
->use_hierarchy
) ||
4671 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4675 memcg
->oom_kill_disable
= val
;
4677 memcg_oom_recover(memcg
);
4683 static const struct file_operations mem_control_numa_stat_file_operations
= {
4685 .llseek
= seq_lseek
,
4686 .release
= single_release
,
4689 static int mem_control_numa_stat_open(struct inode
*unused
, struct file
*file
)
4691 struct cgroup
*cont
= file
->f_dentry
->d_parent
->d_fsdata
;
4693 file
->f_op
= &mem_control_numa_stat_file_operations
;
4694 return single_open(file
, mem_control_numa_stat_show
, cont
);
4696 #endif /* CONFIG_NUMA */
4698 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
4699 static u64
kmem_limit_independent_read(struct cgroup
*cgroup
, struct cftype
*cft
)
4701 return mem_cgroup_from_cont(cgroup
)->kmem_independent_accounting
;
4704 static int kmem_limit_independent_write(struct cgroup
*cgroup
, struct cftype
*cft
,
4707 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgroup
);
4708 struct mem_cgroup
*parent
= parent_mem_cgroup(memcg
);
4713 * This follows the same hierarchy restrictions than
4714 * mem_cgroup_hierarchy_write()
4716 if (!parent
|| !parent
->use_hierarchy
) {
4717 if (list_empty(&cgroup
->children
))
4718 memcg
->kmem_independent_accounting
= val
;
4727 static struct cftype kmem_cgroup_files
[] = {
4729 .name
= "independent_kmem_limit",
4730 .read_u64
= kmem_limit_independent_read
,
4731 .write_u64
= kmem_limit_independent_write
,
4734 .name
= "kmem.usage_in_bytes",
4735 .private = MEMFILE_PRIVATE(_KMEM
, RES_USAGE
),
4736 .read_u64
= mem_cgroup_read
,
4739 .name
= "kmem.limit_in_bytes",
4740 .private = MEMFILE_PRIVATE(_KMEM
, RES_LIMIT
),
4741 .read_u64
= mem_cgroup_read
,
4745 static int register_kmem_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4749 ret
= cgroup_add_files(cont
, ss
, kmem_cgroup_files
,
4750 ARRAY_SIZE(kmem_cgroup_files
));
4753 * Part of this would be better living in a separate allocation
4754 * function, leaving us with just the cgroup tree population work.
4755 * We, however, depend on state such as network's proto_list that
4756 * is only initialized after cgroup creation. I found the less
4757 * cumbersome way to deal with it to defer it all to populate time
4760 ret
= mem_cgroup_sockets_init(cont
, ss
);
4764 static void kmem_cgroup_destroy(struct cgroup_subsys
*ss
,
4765 struct cgroup
*cont
)
4767 mem_cgroup_sockets_destroy(cont
, ss
);
4770 static int register_kmem_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4775 static void kmem_cgroup_destroy(struct cgroup_subsys
*ss
,
4776 struct cgroup
*cont
)
4781 static struct cftype mem_cgroup_files
[] = {
4783 .name
= "usage_in_bytes",
4784 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
4785 .read_u64
= mem_cgroup_read
,
4786 .register_event
= mem_cgroup_usage_register_event
,
4787 .unregister_event
= mem_cgroup_usage_unregister_event
,
4790 .name
= "max_usage_in_bytes",
4791 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
4792 .trigger
= mem_cgroup_reset
,
4793 .read_u64
= mem_cgroup_read
,
4796 .name
= "limit_in_bytes",
4797 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
4798 .write_string
= mem_cgroup_write
,
4799 .read_u64
= mem_cgroup_read
,
4802 .name
= "soft_limit_in_bytes",
4803 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
4804 .write_string
= mem_cgroup_write
,
4805 .read_u64
= mem_cgroup_read
,
4809 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
4810 .trigger
= mem_cgroup_reset
,
4811 .read_u64
= mem_cgroup_read
,
4815 .read_map
= mem_control_stat_show
,
4818 .name
= "force_empty",
4819 .trigger
= mem_cgroup_force_empty_write
,
4822 .name
= "use_hierarchy",
4823 .write_u64
= mem_cgroup_hierarchy_write
,
4824 .read_u64
= mem_cgroup_hierarchy_read
,
4827 .name
= "swappiness",
4828 .read_u64
= mem_cgroup_swappiness_read
,
4829 .write_u64
= mem_cgroup_swappiness_write
,
4832 .name
= "move_charge_at_immigrate",
4833 .read_u64
= mem_cgroup_move_charge_read
,
4834 .write_u64
= mem_cgroup_move_charge_write
,
4837 .name
= "oom_control",
4838 .read_map
= mem_cgroup_oom_control_read
,
4839 .write_u64
= mem_cgroup_oom_control_write
,
4840 .register_event
= mem_cgroup_oom_register_event
,
4841 .unregister_event
= mem_cgroup_oom_unregister_event
,
4842 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
4846 .name
= "numa_stat",
4847 .open
= mem_control_numa_stat_open
,
4853 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4854 static struct cftype memsw_cgroup_files
[] = {
4856 .name
= "memsw.usage_in_bytes",
4857 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
4858 .read_u64
= mem_cgroup_read
,
4859 .register_event
= mem_cgroup_usage_register_event
,
4860 .unregister_event
= mem_cgroup_usage_unregister_event
,
4863 .name
= "memsw.max_usage_in_bytes",
4864 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
4865 .trigger
= mem_cgroup_reset
,
4866 .read_u64
= mem_cgroup_read
,
4869 .name
= "memsw.limit_in_bytes",
4870 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
4871 .write_string
= mem_cgroup_write
,
4872 .read_u64
= mem_cgroup_read
,
4875 .name
= "memsw.failcnt",
4876 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
4877 .trigger
= mem_cgroup_reset
,
4878 .read_u64
= mem_cgroup_read
,
4882 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4884 if (!do_swap_account
)
4886 return cgroup_add_files(cont
, ss
, memsw_cgroup_files
,
4887 ARRAY_SIZE(memsw_cgroup_files
));
4890 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4896 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4898 struct mem_cgroup_per_node
*pn
;
4899 struct mem_cgroup_per_zone
*mz
;
4901 int zone
, tmp
= node
;
4903 * This routine is called against possible nodes.
4904 * But it's BUG to call kmalloc() against offline node.
4906 * TODO: this routine can waste much memory for nodes which will
4907 * never be onlined. It's better to use memory hotplug callback
4910 if (!node_state(node
, N_NORMAL_MEMORY
))
4912 pn
= kzalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
4916 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4917 mz
= &pn
->zoneinfo
[zone
];
4919 INIT_LIST_HEAD(&mz
->lists
[l
]);
4920 mz
->usage_in_excess
= 0;
4921 mz
->on_tree
= false;
4924 memcg
->info
.nodeinfo
[node
] = pn
;
4928 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4930 kfree(memcg
->info
.nodeinfo
[node
]);
4933 static struct mem_cgroup
*mem_cgroup_alloc(void)
4935 struct mem_cgroup
*mem
;
4936 int size
= sizeof(struct mem_cgroup
);
4938 /* Can be very big if MAX_NUMNODES is very big */
4939 if (size
< PAGE_SIZE
)
4940 mem
= kzalloc(size
, GFP_KERNEL
);
4942 mem
= vzalloc(size
);
4947 mem
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4950 spin_lock_init(&mem
->pcp_counter_lock
);
4954 if (size
< PAGE_SIZE
)
4962 * At destroying mem_cgroup, references from swap_cgroup can remain.
4963 * (scanning all at force_empty is too costly...)
4965 * Instead of clearing all references at force_empty, we remember
4966 * the number of reference from swap_cgroup and free mem_cgroup when
4967 * it goes down to 0.
4969 * Removal of cgroup itself succeeds regardless of refs from swap.
4972 static void __mem_cgroup_free(struct mem_cgroup
*memcg
)
4976 mem_cgroup_remove_from_trees(memcg
);
4977 free_css_id(&mem_cgroup_subsys
, &memcg
->css
);
4979 for_each_node_state(node
, N_POSSIBLE
)
4980 free_mem_cgroup_per_zone_info(memcg
, node
);
4982 free_percpu(memcg
->stat
);
4983 if (sizeof(struct mem_cgroup
) < PAGE_SIZE
)
4989 static void mem_cgroup_get(struct mem_cgroup
*memcg
)
4991 atomic_inc(&memcg
->refcnt
);
4994 static void __mem_cgroup_put(struct mem_cgroup
*memcg
, int count
)
4996 if (atomic_sub_and_test(count
, &memcg
->refcnt
)) {
4997 struct mem_cgroup
*parent
= parent_mem_cgroup(memcg
);
4998 __mem_cgroup_free(memcg
);
5000 mem_cgroup_put(parent
);
5004 static void mem_cgroup_put(struct mem_cgroup
*memcg
)
5006 __mem_cgroup_put(memcg
, 1);
5010 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
5012 struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*memcg
)
5014 if (!memcg
->res
.parent
)
5016 return mem_cgroup_from_res_counter(memcg
->res
.parent
, res
);
5018 EXPORT_SYMBOL(parent_mem_cgroup
);
5020 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
5021 static void __init
enable_swap_cgroup(void)
5023 if (!mem_cgroup_disabled() && really_do_swap_account
)
5024 do_swap_account
= 1;
5027 static void __init
enable_swap_cgroup(void)
5032 static int mem_cgroup_soft_limit_tree_init(void)
5034 struct mem_cgroup_tree_per_node
*rtpn
;
5035 struct mem_cgroup_tree_per_zone
*rtpz
;
5036 int tmp
, node
, zone
;
5038 for_each_node_state(node
, N_POSSIBLE
) {
5040 if (!node_state(node
, N_NORMAL_MEMORY
))
5042 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
, tmp
);
5046 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
5048 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
5049 rtpz
= &rtpn
->rb_tree_per_zone
[zone
];
5050 rtpz
->rb_root
= RB_ROOT
;
5051 spin_lock_init(&rtpz
->lock
);
5057 static struct cgroup_subsys_state
* __ref
5058 mem_cgroup_create(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
5060 struct mem_cgroup
*memcg
, *parent
;
5061 long error
= -ENOMEM
;
5064 memcg
= mem_cgroup_alloc();
5066 return ERR_PTR(error
);
5068 for_each_node_state(node
, N_POSSIBLE
)
5069 if (alloc_mem_cgroup_per_zone_info(memcg
, node
))
5073 if (cont
->parent
== NULL
) {
5075 enable_swap_cgroup();
5077 root_mem_cgroup
= memcg
;
5078 if (mem_cgroup_soft_limit_tree_init())
5080 for_each_possible_cpu(cpu
) {
5081 struct memcg_stock_pcp
*stock
=
5082 &per_cpu(memcg_stock
, cpu
);
5083 INIT_WORK(&stock
->work
, drain_local_stock
);
5085 hotcpu_notifier(memcg_cpu_hotplug_callback
, 0);
5087 parent
= mem_cgroup_from_cont(cont
->parent
);
5088 memcg
->use_hierarchy
= parent
->use_hierarchy
;
5089 memcg
->oom_kill_disable
= parent
->oom_kill_disable
;
5092 if (parent
&& parent
->use_hierarchy
) {
5093 res_counter_init(&memcg
->res
, &parent
->res
);
5094 res_counter_init(&memcg
->memsw
, &parent
->memsw
);
5095 res_counter_init(&memcg
->kmem
, &parent
->kmem
);
5097 * We increment refcnt of the parent to ensure that we can
5098 * safely access it on res_counter_charge/uncharge.
5099 * This refcnt will be decremented when freeing this
5100 * mem_cgroup(see mem_cgroup_put).
5102 mem_cgroup_get(parent
);
5104 res_counter_init(&memcg
->res
, NULL
);
5105 res_counter_init(&memcg
->memsw
, NULL
);
5106 res_counter_init(&memcg
->kmem
, NULL
);
5108 memcg
->last_scanned_child
= 0;
5109 memcg
->last_scanned_node
= MAX_NUMNODES
;
5110 INIT_LIST_HEAD(&memcg
->oom_notify
);
5113 memcg
->swappiness
= mem_cgroup_swappiness(parent
);
5114 atomic_set(&memcg
->refcnt
, 1);
5115 memcg
->move_charge_at_immigrate
= 0;
5116 mutex_init(&memcg
->thresholds_lock
);
5119 __mem_cgroup_free(memcg
);
5120 root_mem_cgroup
= NULL
;
5121 return ERR_PTR(error
);
5124 static int mem_cgroup_pre_destroy(struct cgroup_subsys
*ss
,
5125 struct cgroup
*cont
)
5127 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
5129 return mem_cgroup_force_empty(memcg
, false);
5132 static void mem_cgroup_destroy(struct cgroup_subsys
*ss
,
5133 struct cgroup
*cont
)
5135 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
5137 kmem_cgroup_destroy(ss
, cont
);
5139 mem_cgroup_put(memcg
);
5142 static int mem_cgroup_populate(struct cgroup_subsys
*ss
,
5143 struct cgroup
*cont
)
5147 ret
= cgroup_add_files(cont
, ss
, mem_cgroup_files
,
5148 ARRAY_SIZE(mem_cgroup_files
));
5151 ret
= register_memsw_files(cont
, ss
);
5154 ret
= register_kmem_files(cont
, ss
);
5160 /* Handlers for move charge at task migration. */
5161 #define PRECHARGE_COUNT_AT_ONCE 256
5162 static int mem_cgroup_do_precharge(unsigned long count
)
5165 int batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5166 struct mem_cgroup
*memcg
= mc
.to
;
5168 if (mem_cgroup_is_root(memcg
)) {
5169 mc
.precharge
+= count
;
5170 /* we don't need css_get for root */
5173 /* try to charge at once */
5175 struct res_counter
*dummy
;
5177 * "memcg" cannot be under rmdir() because we've already checked
5178 * by cgroup_lock_live_cgroup() that it is not removed and we
5179 * are still under the same cgroup_mutex. So we can postpone
5182 if (res_counter_charge(&memcg
->res
, PAGE_SIZE
* count
, &dummy
))
5184 if (do_swap_account
&& res_counter_charge(&memcg
->memsw
,
5185 PAGE_SIZE
* count
, &dummy
)) {
5186 res_counter_uncharge(&memcg
->res
, PAGE_SIZE
* count
);
5189 mc
.precharge
+= count
;
5193 /* fall back to one by one charge */
5195 if (signal_pending(current
)) {
5199 if (!batch_count
--) {
5200 batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5203 ret
= __mem_cgroup_try_charge(NULL
,
5204 GFP_KERNEL
, 1, &memcg
, false);
5206 /* mem_cgroup_clear_mc() will do uncharge later */
5214 * is_target_pte_for_mc - check a pte whether it is valid for move charge
5215 * @vma: the vma the pte to be checked belongs
5216 * @addr: the address corresponding to the pte to be checked
5217 * @ptent: the pte to be checked
5218 * @target: the pointer the target page or swap ent will be stored(can be NULL)
5221 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5222 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5223 * move charge. if @target is not NULL, the page is stored in target->page
5224 * with extra refcnt got(Callers should handle it).
5225 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5226 * target for charge migration. if @target is not NULL, the entry is stored
5229 * Called with pte lock held.
5236 enum mc_target_type
{
5237 MC_TARGET_NONE
, /* not used */
5242 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
5243 unsigned long addr
, pte_t ptent
)
5245 struct page
*page
= vm_normal_page(vma
, addr
, ptent
);
5247 if (!page
|| !page_mapped(page
))
5249 if (PageAnon(page
)) {
5250 /* we don't move shared anon */
5251 if (!move_anon() || page_mapcount(page
) > 2)
5253 } else if (!move_file())
5254 /* we ignore mapcount for file pages */
5256 if (!get_page_unless_zero(page
))
5262 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5263 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5266 struct page
*page
= NULL
;
5267 swp_entry_t ent
= pte_to_swp_entry(ptent
);
5269 if (!move_anon() || non_swap_entry(ent
))
5271 usage_count
= mem_cgroup_count_swap_user(ent
, &page
);
5272 if (usage_count
> 1) { /* we don't move shared anon */
5277 if (do_swap_account
)
5278 entry
->val
= ent
.val
;
5283 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
5284 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5286 struct page
*page
= NULL
;
5287 struct inode
*inode
;
5288 struct address_space
*mapping
;
5291 if (!vma
->vm_file
) /* anonymous vma */
5296 inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
5297 mapping
= vma
->vm_file
->f_mapping
;
5298 if (pte_none(ptent
))
5299 pgoff
= linear_page_index(vma
, addr
);
5300 else /* pte_file(ptent) is true */
5301 pgoff
= pte_to_pgoff(ptent
);
5303 /* page is moved even if it's not RSS of this task(page-faulted). */
5304 page
= find_get_page(mapping
, pgoff
);
5307 /* shmem/tmpfs may report page out on swap: account for that too. */
5308 if (radix_tree_exceptional_entry(page
)) {
5309 swp_entry_t swap
= radix_to_swp_entry(page
);
5310 if (do_swap_account
)
5312 page
= find_get_page(&swapper_space
, swap
.val
);
5318 static int is_target_pte_for_mc(struct vm_area_struct
*vma
,
5319 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
5321 struct page
*page
= NULL
;
5322 struct page_cgroup
*pc
;
5324 swp_entry_t ent
= { .val
= 0 };
5326 if (pte_present(ptent
))
5327 page
= mc_handle_present_pte(vma
, addr
, ptent
);
5328 else if (is_swap_pte(ptent
))
5329 page
= mc_handle_swap_pte(vma
, addr
, ptent
, &ent
);
5330 else if (pte_none(ptent
) || pte_file(ptent
))
5331 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
5333 if (!page
&& !ent
.val
)
5336 pc
= lookup_page_cgroup(page
);
5338 * Do only loose check w/o page_cgroup lock.
5339 * mem_cgroup_move_account() checks the pc is valid or not under
5342 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5343 ret
= MC_TARGET_PAGE
;
5345 target
->page
= page
;
5347 if (!ret
|| !target
)
5350 /* There is a swap entry and a page doesn't exist or isn't charged */
5351 if (ent
.val
&& !ret
&&
5352 css_id(&mc
.from
->css
) == lookup_swap_cgroup(ent
)) {
5353 ret
= MC_TARGET_SWAP
;
5360 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
5361 unsigned long addr
, unsigned long end
,
5362 struct mm_walk
*walk
)
5364 struct vm_area_struct
*vma
= walk
->private;
5368 split_huge_page_pmd(walk
->mm
, pmd
);
5370 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5371 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
5372 if (is_target_pte_for_mc(vma
, addr
, *pte
, NULL
))
5373 mc
.precharge
++; /* increment precharge temporarily */
5374 pte_unmap_unlock(pte
- 1, ptl
);
5380 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
5382 unsigned long precharge
;
5383 struct vm_area_struct
*vma
;
5385 down_read(&mm
->mmap_sem
);
5386 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5387 struct mm_walk mem_cgroup_count_precharge_walk
= {
5388 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
5392 if (is_vm_hugetlb_page(vma
))
5394 walk_page_range(vma
->vm_start
, vma
->vm_end
,
5395 &mem_cgroup_count_precharge_walk
);
5397 up_read(&mm
->mmap_sem
);
5399 precharge
= mc
.precharge
;
5405 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
5407 unsigned long precharge
= mem_cgroup_count_precharge(mm
);
5409 VM_BUG_ON(mc
.moving_task
);
5410 mc
.moving_task
= current
;
5411 return mem_cgroup_do_precharge(precharge
);
5414 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5415 static void __mem_cgroup_clear_mc(void)
5417 struct mem_cgroup
*from
= mc
.from
;
5418 struct mem_cgroup
*to
= mc
.to
;
5420 /* we must uncharge all the leftover precharges from mc.to */
5422 __mem_cgroup_cancel_charge(mc
.to
, mc
.precharge
);
5426 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5427 * we must uncharge here.
5429 if (mc
.moved_charge
) {
5430 __mem_cgroup_cancel_charge(mc
.from
, mc
.moved_charge
);
5431 mc
.moved_charge
= 0;
5433 /* we must fixup refcnts and charges */
5434 if (mc
.moved_swap
) {
5435 /* uncharge swap account from the old cgroup */
5436 if (!mem_cgroup_is_root(mc
.from
))
5437 res_counter_uncharge(&mc
.from
->memsw
,
5438 PAGE_SIZE
* mc
.moved_swap
);
5439 __mem_cgroup_put(mc
.from
, mc
.moved_swap
);
5441 if (!mem_cgroup_is_root(mc
.to
)) {
5443 * we charged both to->res and to->memsw, so we should
5446 res_counter_uncharge(&mc
.to
->res
,
5447 PAGE_SIZE
* mc
.moved_swap
);
5449 /* we've already done mem_cgroup_get(mc.to) */
5452 memcg_oom_recover(from
);
5453 memcg_oom_recover(to
);
5454 wake_up_all(&mc
.waitq
);
5457 static void mem_cgroup_clear_mc(void)
5459 struct mem_cgroup
*from
= mc
.from
;
5462 * we must clear moving_task before waking up waiters at the end of
5465 mc
.moving_task
= NULL
;
5466 __mem_cgroup_clear_mc();
5467 spin_lock(&mc
.lock
);
5470 spin_unlock(&mc
.lock
);
5471 mem_cgroup_end_move(from
);
5474 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
5475 struct cgroup
*cgroup
,
5476 struct task_struct
*p
)
5479 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgroup
);
5481 if (memcg
->move_charge_at_immigrate
) {
5482 struct mm_struct
*mm
;
5483 struct mem_cgroup
*from
= mem_cgroup_from_task(p
);
5485 VM_BUG_ON(from
== memcg
);
5487 mm
= get_task_mm(p
);
5490 /* We move charges only when we move a owner of the mm */
5491 if (mm
->owner
== p
) {
5494 VM_BUG_ON(mc
.precharge
);
5495 VM_BUG_ON(mc
.moved_charge
);
5496 VM_BUG_ON(mc
.moved_swap
);
5497 mem_cgroup_start_move(from
);
5498 spin_lock(&mc
.lock
);
5501 spin_unlock(&mc
.lock
);
5502 /* We set mc.moving_task later */
5504 ret
= mem_cgroup_precharge_mc(mm
);
5506 mem_cgroup_clear_mc();
5513 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
5514 struct cgroup
*cgroup
,
5515 struct task_struct
*p
)
5517 mem_cgroup_clear_mc();
5520 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
5521 unsigned long addr
, unsigned long end
,
5522 struct mm_walk
*walk
)
5525 struct vm_area_struct
*vma
= walk
->private;
5529 split_huge_page_pmd(walk
->mm
, pmd
);
5531 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5532 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
5533 pte_t ptent
= *(pte
++);
5534 union mc_target target
;
5537 struct page_cgroup
*pc
;
5543 type
= is_target_pte_for_mc(vma
, addr
, ptent
, &target
);
5545 case MC_TARGET_PAGE
:
5547 if (isolate_lru_page(page
))
5549 pc
= lookup_page_cgroup(page
);
5550 if (!mem_cgroup_move_account(page
, 1, pc
,
5551 mc
.from
, mc
.to
, false)) {
5553 /* we uncharge from mc.from later. */
5556 putback_lru_page(page
);
5557 put
: /* is_target_pte_for_mc() gets the page */
5560 case MC_TARGET_SWAP
:
5562 if (!mem_cgroup_move_swap_account(ent
,
5563 mc
.from
, mc
.to
, false)) {
5565 /* we fixup refcnts and charges later. */
5573 pte_unmap_unlock(pte
- 1, ptl
);
5578 * We have consumed all precharges we got in can_attach().
5579 * We try charge one by one, but don't do any additional
5580 * charges to mc.to if we have failed in charge once in attach()
5583 ret
= mem_cgroup_do_precharge(1);
5591 static void mem_cgroup_move_charge(struct mm_struct
*mm
)
5593 struct vm_area_struct
*vma
;
5595 lru_add_drain_all();
5597 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
5599 * Someone who are holding the mmap_sem might be waiting in
5600 * waitq. So we cancel all extra charges, wake up all waiters,
5601 * and retry. Because we cancel precharges, we might not be able
5602 * to move enough charges, but moving charge is a best-effort
5603 * feature anyway, so it wouldn't be a big problem.
5605 __mem_cgroup_clear_mc();
5609 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5611 struct mm_walk mem_cgroup_move_charge_walk
= {
5612 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
5616 if (is_vm_hugetlb_page(vma
))
5618 ret
= walk_page_range(vma
->vm_start
, vma
->vm_end
,
5619 &mem_cgroup_move_charge_walk
);
5622 * means we have consumed all precharges and failed in
5623 * doing additional charge. Just abandon here.
5627 up_read(&mm
->mmap_sem
);
5630 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
5631 struct cgroup
*cont
,
5632 struct cgroup
*old_cont
,
5633 struct task_struct
*p
)
5635 struct mm_struct
*mm
= get_task_mm(p
);
5639 mem_cgroup_move_charge(mm
);
5644 mem_cgroup_clear_mc();
5646 #else /* !CONFIG_MMU */
5647 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
5648 struct cgroup
*cgroup
,
5649 struct task_struct
*p
)
5653 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
5654 struct cgroup
*cgroup
,
5655 struct task_struct
*p
)
5658 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
5659 struct cgroup
*cont
,
5660 struct cgroup
*old_cont
,
5661 struct task_struct
*p
)
5666 struct cgroup_subsys mem_cgroup_subsys
= {
5668 .subsys_id
= mem_cgroup_subsys_id
,
5669 .create
= mem_cgroup_create
,
5670 .pre_destroy
= mem_cgroup_pre_destroy
,
5671 .destroy
= mem_cgroup_destroy
,
5672 .populate
= mem_cgroup_populate
,
5673 .can_attach
= mem_cgroup_can_attach
,
5674 .cancel_attach
= mem_cgroup_cancel_attach
,
5675 .attach
= mem_cgroup_move_task
,
5680 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
5681 static int __init
enable_swap_account(char *s
)
5683 /* consider enabled if no parameter or 1 is given */
5684 if (!strcmp(s
, "1"))
5685 really_do_swap_account
= 1;
5686 else if (!strcmp(s
, "0"))
5687 really_do_swap_account
= 0;
5690 __setup("swapaccount=", enable_swap_account
);