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
];
126 struct mem_cgroup_reclaim_iter
{
127 /* css_id of the last scanned hierarchy member */
129 /* scan generation, increased every round-trip */
130 unsigned int generation
;
134 * per-zone information in memory controller.
136 struct mem_cgroup_per_zone
{
137 struct lruvec lruvec
;
138 unsigned long count
[NR_LRU_LISTS
];
140 struct mem_cgroup_reclaim_iter reclaim_iter
[DEF_PRIORITY
+ 1];
142 struct zone_reclaim_stat reclaim_stat
;
143 struct rb_node tree_node
; /* RB tree node */
144 unsigned long long usage_in_excess
;/* Set to the value by which */
145 /* the soft limit is exceeded*/
147 struct mem_cgroup
*mem
; /* Back pointer, we cannot */
148 /* use container_of */
150 /* Macro for accessing counter */
151 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
153 struct mem_cgroup_per_node
{
154 struct mem_cgroup_per_zone zoneinfo
[MAX_NR_ZONES
];
157 struct mem_cgroup_lru_info
{
158 struct mem_cgroup_per_node
*nodeinfo
[MAX_NUMNODES
];
162 * Cgroups above their limits are maintained in a RB-Tree, independent of
163 * their hierarchy representation
166 struct mem_cgroup_tree_per_zone
{
167 struct rb_root rb_root
;
171 struct mem_cgroup_tree_per_node
{
172 struct mem_cgroup_tree_per_zone rb_tree_per_zone
[MAX_NR_ZONES
];
175 struct mem_cgroup_tree
{
176 struct mem_cgroup_tree_per_node
*rb_tree_per_node
[MAX_NUMNODES
];
179 static struct mem_cgroup_tree soft_limit_tree __read_mostly
;
181 struct mem_cgroup_threshold
{
182 struct eventfd_ctx
*eventfd
;
187 struct mem_cgroup_threshold_ary
{
188 /* An array index points to threshold just below usage. */
189 int current_threshold
;
190 /* Size of entries[] */
192 /* Array of thresholds */
193 struct mem_cgroup_threshold entries
[0];
196 struct mem_cgroup_thresholds
{
197 /* Primary thresholds array */
198 struct mem_cgroup_threshold_ary
*primary
;
200 * Spare threshold array.
201 * This is needed to make mem_cgroup_unregister_event() "never fail".
202 * It must be able to store at least primary->size - 1 entries.
204 struct mem_cgroup_threshold_ary
*spare
;
208 struct mem_cgroup_eventfd_list
{
209 struct list_head list
;
210 struct eventfd_ctx
*eventfd
;
213 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
);
214 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
);
217 * The memory controller data structure. The memory controller controls both
218 * page cache and RSS per cgroup. We would eventually like to provide
219 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
220 * to help the administrator determine what knobs to tune.
222 * TODO: Add a water mark for the memory controller. Reclaim will begin when
223 * we hit the water mark. May be even add a low water mark, such that
224 * no reclaim occurs from a cgroup at it's low water mark, this is
225 * a feature that will be implemented much later in the future.
228 struct cgroup_subsys_state css
;
230 * the counter to account for memory usage
232 struct res_counter res
;
234 * the counter to account for mem+swap usage.
236 struct res_counter memsw
;
238 * Per cgroup active and inactive list, similar to the
239 * per zone LRU lists.
241 struct mem_cgroup_lru_info info
;
242 int last_scanned_node
;
244 nodemask_t scan_nodes
;
245 atomic_t numainfo_events
;
246 atomic_t numainfo_updating
;
249 * Should the accounting and control be hierarchical, per subtree?
259 /* OOM-Killer disable */
260 int oom_kill_disable
;
262 /* set when res.limit == memsw.limit */
263 bool memsw_is_minimum
;
265 /* protect arrays of thresholds */
266 struct mutex thresholds_lock
;
268 /* thresholds for memory usage. RCU-protected */
269 struct mem_cgroup_thresholds thresholds
;
271 /* thresholds for mem+swap usage. RCU-protected */
272 struct mem_cgroup_thresholds memsw_thresholds
;
274 /* For oom notifier event fd */
275 struct list_head oom_notify
;
278 * Should we move charges of a task when a task is moved into this
279 * mem_cgroup ? And what type of charges should we move ?
281 unsigned long move_charge_at_immigrate
;
285 struct mem_cgroup_stat_cpu
*stat
;
287 * used when a cpu is offlined or other synchronizations
288 * See mem_cgroup_read_stat().
290 struct mem_cgroup_stat_cpu nocpu_base
;
291 spinlock_t pcp_counter_lock
;
294 struct tcp_memcontrol tcp_mem
;
298 /* Stuffs for move charges at task migration. */
300 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
301 * left-shifted bitmap of these types.
304 MOVE_CHARGE_TYPE_ANON
, /* private anonymous page and swap of it */
305 MOVE_CHARGE_TYPE_FILE
, /* file page(including tmpfs) and swap of it */
309 /* "mc" and its members are protected by cgroup_mutex */
310 static struct move_charge_struct
{
311 spinlock_t lock
; /* for from, to */
312 struct mem_cgroup
*from
;
313 struct mem_cgroup
*to
;
314 unsigned long precharge
;
315 unsigned long moved_charge
;
316 unsigned long moved_swap
;
317 struct task_struct
*moving_task
; /* a task moving charges */
318 wait_queue_head_t waitq
; /* a waitq for other context */
320 .lock
= __SPIN_LOCK_UNLOCKED(mc
.lock
),
321 .waitq
= __WAIT_QUEUE_HEAD_INITIALIZER(mc
.waitq
),
324 static bool move_anon(void)
326 return test_bit(MOVE_CHARGE_TYPE_ANON
,
327 &mc
.to
->move_charge_at_immigrate
);
330 static bool move_file(void)
332 return test_bit(MOVE_CHARGE_TYPE_FILE
,
333 &mc
.to
->move_charge_at_immigrate
);
337 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
338 * limit reclaim to prevent infinite loops, if they ever occur.
340 #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
341 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
344 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
345 MEM_CGROUP_CHARGE_TYPE_MAPPED
,
346 MEM_CGROUP_CHARGE_TYPE_SHMEM
, /* used by page migration of shmem */
347 MEM_CGROUP_CHARGE_TYPE_FORCE
, /* used by force_empty */
348 MEM_CGROUP_CHARGE_TYPE_SWAPOUT
, /* for accounting swapcache */
349 MEM_CGROUP_CHARGE_TYPE_DROP
, /* a page was unused swap cache */
353 /* for encoding cft->private value on file */
356 #define _OOM_TYPE (2)
357 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
358 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
359 #define MEMFILE_ATTR(val) ((val) & 0xffff)
360 /* Used for OOM nofiier */
361 #define OOM_CONTROL (0)
364 * Reclaim flags for mem_cgroup_hierarchical_reclaim
366 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
367 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
368 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
369 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
371 static void mem_cgroup_get(struct mem_cgroup
*memcg
);
372 static void mem_cgroup_put(struct mem_cgroup
*memcg
);
374 /* Writing them here to avoid exposing memcg's inner layout */
375 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
376 #include <net/sock.h>
379 static bool mem_cgroup_is_root(struct mem_cgroup
*memcg
);
380 void sock_update_memcg(struct sock
*sk
)
382 if (mem_cgroup_sockets_enabled
) {
383 struct mem_cgroup
*memcg
;
385 BUG_ON(!sk
->sk_prot
->proto_cgroup
);
387 /* Socket cloning can throw us here with sk_cgrp already
388 * filled. It won't however, necessarily happen from
389 * process context. So the test for root memcg given
390 * the current task's memcg won't help us in this case.
392 * Respecting the original socket's memcg is a better
393 * decision in this case.
396 BUG_ON(mem_cgroup_is_root(sk
->sk_cgrp
->memcg
));
397 mem_cgroup_get(sk
->sk_cgrp
->memcg
);
402 memcg
= mem_cgroup_from_task(current
);
403 if (!mem_cgroup_is_root(memcg
)) {
404 mem_cgroup_get(memcg
);
405 sk
->sk_cgrp
= sk
->sk_prot
->proto_cgroup(memcg
);
410 EXPORT_SYMBOL(sock_update_memcg
);
412 void sock_release_memcg(struct sock
*sk
)
414 if (mem_cgroup_sockets_enabled
&& sk
->sk_cgrp
) {
415 struct mem_cgroup
*memcg
;
416 WARN_ON(!sk
->sk_cgrp
->memcg
);
417 memcg
= sk
->sk_cgrp
->memcg
;
418 mem_cgroup_put(memcg
);
423 struct cg_proto
*tcp_proto_cgroup(struct mem_cgroup
*memcg
)
425 if (!memcg
|| mem_cgroup_is_root(memcg
))
428 return &memcg
->tcp_mem
.cg_proto
;
430 EXPORT_SYMBOL(tcp_proto_cgroup
);
431 #endif /* CONFIG_INET */
432 #endif /* CONFIG_CGROUP_MEM_RES_CTLR_KMEM */
434 static void drain_all_stock_async(struct mem_cgroup
*memcg
);
436 static struct mem_cgroup_per_zone
*
437 mem_cgroup_zoneinfo(struct mem_cgroup
*memcg
, int nid
, int zid
)
439 return &memcg
->info
.nodeinfo
[nid
]->zoneinfo
[zid
];
442 struct cgroup_subsys_state
*mem_cgroup_css(struct mem_cgroup
*memcg
)
447 static struct mem_cgroup_per_zone
*
448 page_cgroup_zoneinfo(struct mem_cgroup
*memcg
, struct page
*page
)
450 int nid
= page_to_nid(page
);
451 int zid
= page_zonenum(page
);
453 return mem_cgroup_zoneinfo(memcg
, nid
, zid
);
456 static struct mem_cgroup_tree_per_zone
*
457 soft_limit_tree_node_zone(int nid
, int zid
)
459 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
462 static struct mem_cgroup_tree_per_zone
*
463 soft_limit_tree_from_page(struct page
*page
)
465 int nid
= page_to_nid(page
);
466 int zid
= page_zonenum(page
);
468 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
472 __mem_cgroup_insert_exceeded(struct mem_cgroup
*memcg
,
473 struct mem_cgroup_per_zone
*mz
,
474 struct mem_cgroup_tree_per_zone
*mctz
,
475 unsigned long long new_usage_in_excess
)
477 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
478 struct rb_node
*parent
= NULL
;
479 struct mem_cgroup_per_zone
*mz_node
;
484 mz
->usage_in_excess
= new_usage_in_excess
;
485 if (!mz
->usage_in_excess
)
489 mz_node
= rb_entry(parent
, struct mem_cgroup_per_zone
,
491 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
)
494 * We can't avoid mem cgroups that are over their soft
495 * limit by the same amount
497 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
500 rb_link_node(&mz
->tree_node
, parent
, p
);
501 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
506 __mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
507 struct mem_cgroup_per_zone
*mz
,
508 struct mem_cgroup_tree_per_zone
*mctz
)
512 rb_erase(&mz
->tree_node
, &mctz
->rb_root
);
517 mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
518 struct mem_cgroup_per_zone
*mz
,
519 struct mem_cgroup_tree_per_zone
*mctz
)
521 spin_lock(&mctz
->lock
);
522 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
523 spin_unlock(&mctz
->lock
);
527 static void mem_cgroup_update_tree(struct mem_cgroup
*memcg
, struct page
*page
)
529 unsigned long long excess
;
530 struct mem_cgroup_per_zone
*mz
;
531 struct mem_cgroup_tree_per_zone
*mctz
;
532 int nid
= page_to_nid(page
);
533 int zid
= page_zonenum(page
);
534 mctz
= soft_limit_tree_from_page(page
);
537 * Necessary to update all ancestors when hierarchy is used.
538 * because their event counter is not touched.
540 for (; memcg
; memcg
= parent_mem_cgroup(memcg
)) {
541 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
542 excess
= res_counter_soft_limit_excess(&memcg
->res
);
544 * We have to update the tree if mz is on RB-tree or
545 * mem is over its softlimit.
547 if (excess
|| mz
->on_tree
) {
548 spin_lock(&mctz
->lock
);
549 /* if on-tree, remove it */
551 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
553 * Insert again. mz->usage_in_excess will be updated.
554 * If excess is 0, no tree ops.
556 __mem_cgroup_insert_exceeded(memcg
, mz
, mctz
, excess
);
557 spin_unlock(&mctz
->lock
);
562 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*memcg
)
565 struct mem_cgroup_per_zone
*mz
;
566 struct mem_cgroup_tree_per_zone
*mctz
;
568 for_each_node(node
) {
569 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
570 mz
= mem_cgroup_zoneinfo(memcg
, node
, zone
);
571 mctz
= soft_limit_tree_node_zone(node
, zone
);
572 mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
577 static struct mem_cgroup_per_zone
*
578 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
580 struct rb_node
*rightmost
= NULL
;
581 struct mem_cgroup_per_zone
*mz
;
585 rightmost
= rb_last(&mctz
->rb_root
);
587 goto done
; /* Nothing to reclaim from */
589 mz
= rb_entry(rightmost
, struct mem_cgroup_per_zone
, tree_node
);
591 * Remove the node now but someone else can add it back,
592 * we will to add it back at the end of reclaim to its correct
593 * position in the tree.
595 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
596 if (!res_counter_soft_limit_excess(&mz
->mem
->res
) ||
597 !css_tryget(&mz
->mem
->css
))
603 static struct mem_cgroup_per_zone
*
604 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
606 struct mem_cgroup_per_zone
*mz
;
608 spin_lock(&mctz
->lock
);
609 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
610 spin_unlock(&mctz
->lock
);
615 * Implementation Note: reading percpu statistics for memcg.
617 * Both of vmstat[] and percpu_counter has threshold and do periodic
618 * synchronization to implement "quick" read. There are trade-off between
619 * reading cost and precision of value. Then, we may have a chance to implement
620 * a periodic synchronizion of counter in memcg's counter.
622 * But this _read() function is used for user interface now. The user accounts
623 * memory usage by memory cgroup and he _always_ requires exact value because
624 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
625 * have to visit all online cpus and make sum. So, for now, unnecessary
626 * synchronization is not implemented. (just implemented for cpu hotplug)
628 * If there are kernel internal actions which can make use of some not-exact
629 * value, and reading all cpu value can be performance bottleneck in some
630 * common workload, threashold and synchonization as vmstat[] should be
633 static long mem_cgroup_read_stat(struct mem_cgroup
*memcg
,
634 enum mem_cgroup_stat_index idx
)
640 for_each_online_cpu(cpu
)
641 val
+= per_cpu(memcg
->stat
->count
[idx
], cpu
);
642 #ifdef CONFIG_HOTPLUG_CPU
643 spin_lock(&memcg
->pcp_counter_lock
);
644 val
+= memcg
->nocpu_base
.count
[idx
];
645 spin_unlock(&memcg
->pcp_counter_lock
);
651 static void mem_cgroup_swap_statistics(struct mem_cgroup
*memcg
,
654 int val
= (charge
) ? 1 : -1;
655 this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_SWAPOUT
], val
);
658 static unsigned long mem_cgroup_read_events(struct mem_cgroup
*memcg
,
659 enum mem_cgroup_events_index idx
)
661 unsigned long val
= 0;
664 for_each_online_cpu(cpu
)
665 val
+= per_cpu(memcg
->stat
->events
[idx
], cpu
);
666 #ifdef CONFIG_HOTPLUG_CPU
667 spin_lock(&memcg
->pcp_counter_lock
);
668 val
+= memcg
->nocpu_base
.events
[idx
];
669 spin_unlock(&memcg
->pcp_counter_lock
);
674 static void mem_cgroup_charge_statistics(struct mem_cgroup
*memcg
,
675 bool file
, int nr_pages
)
680 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_CACHE
],
683 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_RSS
],
686 /* pagein of a big page is an event. So, ignore page size */
688 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGIN
]);
690 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGOUT
]);
691 nr_pages
= -nr_pages
; /* for event */
694 __this_cpu_add(memcg
->stat
->events
[MEM_CGROUP_EVENTS_COUNT
], nr_pages
);
700 mem_cgroup_zone_nr_lru_pages(struct mem_cgroup
*memcg
, int nid
, int zid
,
701 unsigned int lru_mask
)
703 struct mem_cgroup_per_zone
*mz
;
705 unsigned long ret
= 0;
707 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
710 if (BIT(l
) & lru_mask
)
711 ret
+= MEM_CGROUP_ZSTAT(mz
, l
);
717 mem_cgroup_node_nr_lru_pages(struct mem_cgroup
*memcg
,
718 int nid
, unsigned int lru_mask
)
723 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++)
724 total
+= mem_cgroup_zone_nr_lru_pages(memcg
,
730 static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup
*memcg
,
731 unsigned int lru_mask
)
736 for_each_node_state(nid
, N_HIGH_MEMORY
)
737 total
+= mem_cgroup_node_nr_lru_pages(memcg
, nid
, lru_mask
);
741 static bool mem_cgroup_event_ratelimit(struct mem_cgroup
*memcg
,
742 enum mem_cgroup_events_target target
)
744 unsigned long val
, next
;
746 val
= __this_cpu_read(memcg
->stat
->events
[MEM_CGROUP_EVENTS_COUNT
]);
747 next
= __this_cpu_read(memcg
->stat
->targets
[target
]);
748 /* from time_after() in jiffies.h */
749 if ((long)next
- (long)val
< 0) {
751 case MEM_CGROUP_TARGET_THRESH
:
752 next
= val
+ THRESHOLDS_EVENTS_TARGET
;
754 case MEM_CGROUP_TARGET_SOFTLIMIT
:
755 next
= val
+ SOFTLIMIT_EVENTS_TARGET
;
757 case MEM_CGROUP_TARGET_NUMAINFO
:
758 next
= val
+ NUMAINFO_EVENTS_TARGET
;
763 __this_cpu_write(memcg
->stat
->targets
[target
], next
);
770 * Check events in order.
773 static void memcg_check_events(struct mem_cgroup
*memcg
, struct page
*page
)
776 /* threshold event is triggered in finer grain than soft limit */
777 if (unlikely(mem_cgroup_event_ratelimit(memcg
,
778 MEM_CGROUP_TARGET_THRESH
))) {
780 bool do_numainfo __maybe_unused
;
782 do_softlimit
= mem_cgroup_event_ratelimit(memcg
,
783 MEM_CGROUP_TARGET_SOFTLIMIT
);
785 do_numainfo
= mem_cgroup_event_ratelimit(memcg
,
786 MEM_CGROUP_TARGET_NUMAINFO
);
790 mem_cgroup_threshold(memcg
);
791 if (unlikely(do_softlimit
))
792 mem_cgroup_update_tree(memcg
, page
);
794 if (unlikely(do_numainfo
))
795 atomic_inc(&memcg
->numainfo_events
);
801 struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
803 return container_of(cgroup_subsys_state(cont
,
804 mem_cgroup_subsys_id
), struct mem_cgroup
,
808 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
811 * mm_update_next_owner() may clear mm->owner to NULL
812 * if it races with swapoff, page migration, etc.
813 * So this can be called with p == NULL.
818 return container_of(task_subsys_state(p
, mem_cgroup_subsys_id
),
819 struct mem_cgroup
, css
);
822 struct mem_cgroup
*try_get_mem_cgroup_from_mm(struct mm_struct
*mm
)
824 struct mem_cgroup
*memcg
= NULL
;
829 * Because we have no locks, mm->owner's may be being moved to other
830 * cgroup. We use css_tryget() here even if this looks
831 * pessimistic (rather than adding locks here).
835 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
836 if (unlikely(!memcg
))
838 } while (!css_tryget(&memcg
->css
));
844 * mem_cgroup_iter - iterate over memory cgroup hierarchy
845 * @root: hierarchy root
846 * @prev: previously returned memcg, NULL on first invocation
847 * @reclaim: cookie for shared reclaim walks, NULL for full walks
849 * Returns references to children of the hierarchy below @root, or
850 * @root itself, or %NULL after a full round-trip.
852 * Caller must pass the return value in @prev on subsequent
853 * invocations for reference counting, or use mem_cgroup_iter_break()
854 * to cancel a hierarchy walk before the round-trip is complete.
856 * Reclaimers can specify a zone and a priority level in @reclaim to
857 * divide up the memcgs in the hierarchy among all concurrent
858 * reclaimers operating on the same zone and priority.
860 struct mem_cgroup
*mem_cgroup_iter(struct mem_cgroup
*root
,
861 struct mem_cgroup
*prev
,
862 struct mem_cgroup_reclaim_cookie
*reclaim
)
864 struct mem_cgroup
*memcg
= NULL
;
867 if (mem_cgroup_disabled())
871 root
= root_mem_cgroup
;
873 if (prev
&& !reclaim
)
874 id
= css_id(&prev
->css
);
876 if (prev
&& prev
!= root
)
879 if (!root
->use_hierarchy
&& root
!= root_mem_cgroup
) {
886 struct mem_cgroup_reclaim_iter
*uninitialized_var(iter
);
887 struct cgroup_subsys_state
*css
;
890 int nid
= zone_to_nid(reclaim
->zone
);
891 int zid
= zone_idx(reclaim
->zone
);
892 struct mem_cgroup_per_zone
*mz
;
894 mz
= mem_cgroup_zoneinfo(root
, nid
, zid
);
895 iter
= &mz
->reclaim_iter
[reclaim
->priority
];
896 if (prev
&& reclaim
->generation
!= iter
->generation
)
902 css
= css_get_next(&mem_cgroup_subsys
, id
+ 1, &root
->css
, &id
);
904 if (css
== &root
->css
|| css_tryget(css
))
905 memcg
= container_of(css
,
906 struct mem_cgroup
, css
);
915 else if (!prev
&& memcg
)
916 reclaim
->generation
= iter
->generation
;
926 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
927 * @root: hierarchy root
928 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
930 void mem_cgroup_iter_break(struct mem_cgroup
*root
,
931 struct mem_cgroup
*prev
)
934 root
= root_mem_cgroup
;
935 if (prev
&& prev
!= root
)
940 * Iteration constructs for visiting all cgroups (under a tree). If
941 * loops are exited prematurely (break), mem_cgroup_iter_break() must
942 * be used for reference counting.
944 #define for_each_mem_cgroup_tree(iter, root) \
945 for (iter = mem_cgroup_iter(root, NULL, NULL); \
947 iter = mem_cgroup_iter(root, iter, NULL))
949 #define for_each_mem_cgroup(iter) \
950 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
952 iter = mem_cgroup_iter(NULL, iter, NULL))
954 static inline bool mem_cgroup_is_root(struct mem_cgroup
*memcg
)
956 return (memcg
== root_mem_cgroup
);
959 void mem_cgroup_count_vm_event(struct mm_struct
*mm
, enum vm_event_item idx
)
961 struct mem_cgroup
*memcg
;
967 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
968 if (unlikely(!memcg
))
973 this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGFAULT
]);
976 this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGMAJFAULT
]);
984 EXPORT_SYMBOL(mem_cgroup_count_vm_event
);
987 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
988 * @zone: zone of the wanted lruvec
989 * @mem: memcg of the wanted lruvec
991 * Returns the lru list vector holding pages for the given @zone and
992 * @mem. This can be the global zone lruvec, if the memory controller
995 struct lruvec
*mem_cgroup_zone_lruvec(struct zone
*zone
,
996 struct mem_cgroup
*memcg
)
998 struct mem_cgroup_per_zone
*mz
;
1000 if (mem_cgroup_disabled())
1001 return &zone
->lruvec
;
1003 mz
= mem_cgroup_zoneinfo(memcg
, zone_to_nid(zone
), zone_idx(zone
));
1008 * Following LRU functions are allowed to be used without PCG_LOCK.
1009 * Operations are called by routine of global LRU independently from memcg.
1010 * What we have to take care of here is validness of pc->mem_cgroup.
1012 * Changes to pc->mem_cgroup happens when
1015 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1016 * It is added to LRU before charge.
1017 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1018 * When moving account, the page is not on LRU. It's isolated.
1022 * mem_cgroup_lru_add_list - account for adding an lru page and return lruvec
1023 * @zone: zone of the page
1027 * This function accounts for @page being added to @lru, and returns
1028 * the lruvec for the given @zone and the memcg @page is charged to.
1030 * The callsite is then responsible for physically linking the page to
1031 * the returned lruvec->lists[@lru].
1033 struct lruvec
*mem_cgroup_lru_add_list(struct zone
*zone
, struct page
*page
,
1036 struct mem_cgroup_per_zone
*mz
;
1037 struct mem_cgroup
*memcg
;
1038 struct page_cgroup
*pc
;
1040 if (mem_cgroup_disabled())
1041 return &zone
->lruvec
;
1043 pc
= lookup_page_cgroup(page
);
1044 memcg
= pc
->mem_cgroup
;
1047 * Surreptitiously switch any uncharged page to root:
1048 * an uncharged page off lru does nothing to secure
1049 * its former mem_cgroup from sudden removal.
1051 * Our caller holds lru_lock, and PageCgroupUsed is updated
1052 * under page_cgroup lock: between them, they make all uses
1053 * of pc->mem_cgroup safe.
1055 if (!PageCgroupUsed(pc
) && memcg
!= root_mem_cgroup
)
1056 pc
->mem_cgroup
= memcg
= root_mem_cgroup
;
1058 mz
= page_cgroup_zoneinfo(memcg
, page
);
1059 /* compound_order() is stabilized through lru_lock */
1060 MEM_CGROUP_ZSTAT(mz
, lru
) += 1 << compound_order(page
);
1065 * mem_cgroup_lru_del_list - account for removing an lru page
1069 * This function accounts for @page being removed from @lru.
1071 * The callsite is then responsible for physically unlinking
1074 void mem_cgroup_lru_del_list(struct page
*page
, enum lru_list lru
)
1076 struct mem_cgroup_per_zone
*mz
;
1077 struct mem_cgroup
*memcg
;
1078 struct page_cgroup
*pc
;
1080 if (mem_cgroup_disabled())
1083 pc
= lookup_page_cgroup(page
);
1084 memcg
= pc
->mem_cgroup
;
1086 mz
= page_cgroup_zoneinfo(memcg
, page
);
1087 /* huge page split is done under lru_lock. so, we have no races. */
1088 VM_BUG_ON(MEM_CGROUP_ZSTAT(mz
, lru
) < (1 << compound_order(page
)));
1089 MEM_CGROUP_ZSTAT(mz
, lru
) -= 1 << compound_order(page
);
1092 void mem_cgroup_lru_del(struct page
*page
)
1094 mem_cgroup_lru_del_list(page
, page_lru(page
));
1098 * mem_cgroup_lru_move_lists - account for moving a page between lrus
1099 * @zone: zone of the page
1101 * @from: current lru
1104 * This function accounts for @page being moved between the lrus @from
1105 * and @to, and returns the lruvec for the given @zone and the memcg
1106 * @page is charged to.
1108 * The callsite is then responsible for physically relinking
1109 * @page->lru to the returned lruvec->lists[@to].
1111 struct lruvec
*mem_cgroup_lru_move_lists(struct zone
*zone
,
1116 /* XXX: Optimize this, especially for @from == @to */
1117 mem_cgroup_lru_del_list(page
, from
);
1118 return mem_cgroup_lru_add_list(zone
, page
, to
);
1122 * Checks whether given mem is same or in the root_mem_cgroup's
1125 static bool mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_memcg
,
1126 struct mem_cgroup
*memcg
)
1128 if (root_memcg
!= memcg
) {
1129 return (root_memcg
->use_hierarchy
&&
1130 css_is_ancestor(&memcg
->css
, &root_memcg
->css
));
1136 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*memcg
)
1139 struct mem_cgroup
*curr
= NULL
;
1140 struct task_struct
*p
;
1142 p
= find_lock_task_mm(task
);
1144 curr
= try_get_mem_cgroup_from_mm(p
->mm
);
1148 * All threads may have already detached their mm's, but the oom
1149 * killer still needs to detect if they have already been oom
1150 * killed to prevent needlessly killing additional tasks.
1153 curr
= mem_cgroup_from_task(task
);
1155 css_get(&curr
->css
);
1161 * We should check use_hierarchy of "memcg" not "curr". Because checking
1162 * use_hierarchy of "curr" here make this function true if hierarchy is
1163 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1164 * hierarchy(even if use_hierarchy is disabled in "memcg").
1166 ret
= mem_cgroup_same_or_subtree(memcg
, curr
);
1167 css_put(&curr
->css
);
1171 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup
*memcg
, struct zone
*zone
)
1173 unsigned long inactive_ratio
;
1174 int nid
= zone_to_nid(zone
);
1175 int zid
= zone_idx(zone
);
1176 unsigned long inactive
;
1177 unsigned long active
;
1180 inactive
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1181 BIT(LRU_INACTIVE_ANON
));
1182 active
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1183 BIT(LRU_ACTIVE_ANON
));
1185 gb
= (inactive
+ active
) >> (30 - PAGE_SHIFT
);
1187 inactive_ratio
= int_sqrt(10 * gb
);
1191 return inactive
* inactive_ratio
< active
;
1194 int mem_cgroup_inactive_file_is_low(struct mem_cgroup
*memcg
, struct zone
*zone
)
1196 unsigned long active
;
1197 unsigned long inactive
;
1198 int zid
= zone_idx(zone
);
1199 int nid
= zone_to_nid(zone
);
1201 inactive
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1202 BIT(LRU_INACTIVE_FILE
));
1203 active
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1204 BIT(LRU_ACTIVE_FILE
));
1206 return (active
> inactive
);
1209 struct zone_reclaim_stat
*mem_cgroup_get_reclaim_stat(struct mem_cgroup
*memcg
,
1212 int nid
= zone_to_nid(zone
);
1213 int zid
= zone_idx(zone
);
1214 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
1216 return &mz
->reclaim_stat
;
1219 struct zone_reclaim_stat
*
1220 mem_cgroup_get_reclaim_stat_from_page(struct page
*page
)
1222 struct page_cgroup
*pc
;
1223 struct mem_cgroup_per_zone
*mz
;
1225 if (mem_cgroup_disabled())
1228 pc
= lookup_page_cgroup(page
);
1229 if (!PageCgroupUsed(pc
))
1231 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1233 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
1234 return &mz
->reclaim_stat
;
1237 #define mem_cgroup_from_res_counter(counter, member) \
1238 container_of(counter, struct mem_cgroup, member)
1241 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1242 * @mem: the memory cgroup
1244 * Returns the maximum amount of memory @mem can be charged with, in
1247 static unsigned long mem_cgroup_margin(struct mem_cgroup
*memcg
)
1249 unsigned long long margin
;
1251 margin
= res_counter_margin(&memcg
->res
);
1252 if (do_swap_account
)
1253 margin
= min(margin
, res_counter_margin(&memcg
->memsw
));
1254 return margin
>> PAGE_SHIFT
;
1257 int mem_cgroup_swappiness(struct mem_cgroup
*memcg
)
1259 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
1262 if (cgrp
->parent
== NULL
)
1263 return vm_swappiness
;
1265 return memcg
->swappiness
;
1268 static void mem_cgroup_start_move(struct mem_cgroup
*memcg
)
1273 spin_lock(&memcg
->pcp_counter_lock
);
1274 for_each_online_cpu(cpu
)
1275 per_cpu(memcg
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) += 1;
1276 memcg
->nocpu_base
.count
[MEM_CGROUP_ON_MOVE
] += 1;
1277 spin_unlock(&memcg
->pcp_counter_lock
);
1283 static void mem_cgroup_end_move(struct mem_cgroup
*memcg
)
1290 spin_lock(&memcg
->pcp_counter_lock
);
1291 for_each_online_cpu(cpu
)
1292 per_cpu(memcg
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) -= 1;
1293 memcg
->nocpu_base
.count
[MEM_CGROUP_ON_MOVE
] -= 1;
1294 spin_unlock(&memcg
->pcp_counter_lock
);
1298 * 2 routines for checking "mem" is under move_account() or not.
1300 * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used
1301 * for avoiding race in accounting. If true,
1302 * pc->mem_cgroup may be overwritten.
1304 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1305 * under hierarchy of moving cgroups. This is for
1306 * waiting at hith-memory prressure caused by "move".
1309 static bool mem_cgroup_stealed(struct mem_cgroup
*memcg
)
1311 VM_BUG_ON(!rcu_read_lock_held());
1312 return this_cpu_read(memcg
->stat
->count
[MEM_CGROUP_ON_MOVE
]) > 0;
1315 static bool mem_cgroup_under_move(struct mem_cgroup
*memcg
)
1317 struct mem_cgroup
*from
;
1318 struct mem_cgroup
*to
;
1321 * Unlike task_move routines, we access mc.to, mc.from not under
1322 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1324 spin_lock(&mc
.lock
);
1330 ret
= mem_cgroup_same_or_subtree(memcg
, from
)
1331 || mem_cgroup_same_or_subtree(memcg
, to
);
1333 spin_unlock(&mc
.lock
);
1337 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*memcg
)
1339 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1340 if (mem_cgroup_under_move(memcg
)) {
1342 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1343 /* moving charge context might have finished. */
1346 finish_wait(&mc
.waitq
, &wait
);
1354 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1355 * @memcg: The memory cgroup that went over limit
1356 * @p: Task that is going to be killed
1358 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1361 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1363 struct cgroup
*task_cgrp
;
1364 struct cgroup
*mem_cgrp
;
1366 * Need a buffer in BSS, can't rely on allocations. The code relies
1367 * on the assumption that OOM is serialized for memory controller.
1368 * If this assumption is broken, revisit this code.
1370 static char memcg_name
[PATH_MAX
];
1379 mem_cgrp
= memcg
->css
.cgroup
;
1380 task_cgrp
= task_cgroup(p
, mem_cgroup_subsys_id
);
1382 ret
= cgroup_path(task_cgrp
, memcg_name
, PATH_MAX
);
1385 * Unfortunately, we are unable to convert to a useful name
1386 * But we'll still print out the usage information
1393 printk(KERN_INFO
"Task in %s killed", memcg_name
);
1396 ret
= cgroup_path(mem_cgrp
, memcg_name
, PATH_MAX
);
1404 * Continues from above, so we don't need an KERN_ level
1406 printk(KERN_CONT
" as a result of limit of %s\n", memcg_name
);
1409 printk(KERN_INFO
"memory: usage %llukB, limit %llukB, failcnt %llu\n",
1410 res_counter_read_u64(&memcg
->res
, RES_USAGE
) >> 10,
1411 res_counter_read_u64(&memcg
->res
, RES_LIMIT
) >> 10,
1412 res_counter_read_u64(&memcg
->res
, RES_FAILCNT
));
1413 printk(KERN_INFO
"memory+swap: usage %llukB, limit %llukB, "
1415 res_counter_read_u64(&memcg
->memsw
, RES_USAGE
) >> 10,
1416 res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
) >> 10,
1417 res_counter_read_u64(&memcg
->memsw
, RES_FAILCNT
));
1421 * This function returns the number of memcg under hierarchy tree. Returns
1422 * 1(self count) if no children.
1424 static int mem_cgroup_count_children(struct mem_cgroup
*memcg
)
1427 struct mem_cgroup
*iter
;
1429 for_each_mem_cgroup_tree(iter
, memcg
)
1435 * Return the memory (and swap, if configured) limit for a memcg.
1437 u64
mem_cgroup_get_limit(struct mem_cgroup
*memcg
)
1442 limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
1443 limit
+= total_swap_pages
<< PAGE_SHIFT
;
1445 memsw
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
1447 * If memsw is finite and limits the amount of swap space available
1448 * to this memcg, return that limit.
1450 return min(limit
, memsw
);
1453 static unsigned long mem_cgroup_reclaim(struct mem_cgroup
*memcg
,
1455 unsigned long flags
)
1457 unsigned long total
= 0;
1458 bool noswap
= false;
1461 if (flags
& MEM_CGROUP_RECLAIM_NOSWAP
)
1463 if (!(flags
& MEM_CGROUP_RECLAIM_SHRINK
) && memcg
->memsw_is_minimum
)
1466 for (loop
= 0; loop
< MEM_CGROUP_MAX_RECLAIM_LOOPS
; loop
++) {
1468 drain_all_stock_async(memcg
);
1469 total
+= try_to_free_mem_cgroup_pages(memcg
, gfp_mask
, noswap
);
1471 * Allow limit shrinkers, which are triggered directly
1472 * by userspace, to catch signals and stop reclaim
1473 * after minimal progress, regardless of the margin.
1475 if (total
&& (flags
& MEM_CGROUP_RECLAIM_SHRINK
))
1477 if (mem_cgroup_margin(memcg
))
1480 * If nothing was reclaimed after two attempts, there
1481 * may be no reclaimable pages in this hierarchy.
1490 * test_mem_cgroup_node_reclaimable
1491 * @mem: the target memcg
1492 * @nid: the node ID to be checked.
1493 * @noswap : specify true here if the user wants flle only information.
1495 * This function returns whether the specified memcg contains any
1496 * reclaimable pages on a node. Returns true if there are any reclaimable
1497 * pages in the node.
1499 static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup
*memcg
,
1500 int nid
, bool noswap
)
1502 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_FILE
))
1504 if (noswap
|| !total_swap_pages
)
1506 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_ANON
))
1511 #if MAX_NUMNODES > 1
1514 * Always updating the nodemask is not very good - even if we have an empty
1515 * list or the wrong list here, we can start from some node and traverse all
1516 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1519 static void mem_cgroup_may_update_nodemask(struct mem_cgroup
*memcg
)
1523 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1524 * pagein/pageout changes since the last update.
1526 if (!atomic_read(&memcg
->numainfo_events
))
1528 if (atomic_inc_return(&memcg
->numainfo_updating
) > 1)
1531 /* make a nodemask where this memcg uses memory from */
1532 memcg
->scan_nodes
= node_states
[N_HIGH_MEMORY
];
1534 for_each_node_mask(nid
, node_states
[N_HIGH_MEMORY
]) {
1536 if (!test_mem_cgroup_node_reclaimable(memcg
, nid
, false))
1537 node_clear(nid
, memcg
->scan_nodes
);
1540 atomic_set(&memcg
->numainfo_events
, 0);
1541 atomic_set(&memcg
->numainfo_updating
, 0);
1545 * Selecting a node where we start reclaim from. Because what we need is just
1546 * reducing usage counter, start from anywhere is O,K. Considering
1547 * memory reclaim from current node, there are pros. and cons.
1549 * Freeing memory from current node means freeing memory from a node which
1550 * we'll use or we've used. So, it may make LRU bad. And if several threads
1551 * hit limits, it will see a contention on a node. But freeing from remote
1552 * node means more costs for memory reclaim because of memory latency.
1554 * Now, we use round-robin. Better algorithm is welcomed.
1556 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1560 mem_cgroup_may_update_nodemask(memcg
);
1561 node
= memcg
->last_scanned_node
;
1563 node
= next_node(node
, memcg
->scan_nodes
);
1564 if (node
== MAX_NUMNODES
)
1565 node
= first_node(memcg
->scan_nodes
);
1567 * We call this when we hit limit, not when pages are added to LRU.
1568 * No LRU may hold pages because all pages are UNEVICTABLE or
1569 * memcg is too small and all pages are not on LRU. In that case,
1570 * we use curret node.
1572 if (unlikely(node
== MAX_NUMNODES
))
1573 node
= numa_node_id();
1575 memcg
->last_scanned_node
= node
;
1580 * Check all nodes whether it contains reclaimable pages or not.
1581 * For quick scan, we make use of scan_nodes. This will allow us to skip
1582 * unused nodes. But scan_nodes is lazily updated and may not cotain
1583 * enough new information. We need to do double check.
1585 bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1590 * quick check...making use of scan_node.
1591 * We can skip unused nodes.
1593 if (!nodes_empty(memcg
->scan_nodes
)) {
1594 for (nid
= first_node(memcg
->scan_nodes
);
1596 nid
= next_node(nid
, memcg
->scan_nodes
)) {
1598 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1603 * Check rest of nodes.
1605 for_each_node_state(nid
, N_HIGH_MEMORY
) {
1606 if (node_isset(nid
, memcg
->scan_nodes
))
1608 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1615 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1620 bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1622 return test_mem_cgroup_node_reclaimable(memcg
, 0, noswap
);
1626 static int mem_cgroup_soft_reclaim(struct mem_cgroup
*root_memcg
,
1629 unsigned long *total_scanned
)
1631 struct mem_cgroup
*victim
= NULL
;
1634 unsigned long excess
;
1635 unsigned long nr_scanned
;
1636 struct mem_cgroup_reclaim_cookie reclaim
= {
1641 excess
= res_counter_soft_limit_excess(&root_memcg
->res
) >> PAGE_SHIFT
;
1644 victim
= mem_cgroup_iter(root_memcg
, victim
, &reclaim
);
1649 * If we have not been able to reclaim
1650 * anything, it might because there are
1651 * no reclaimable pages under this hierarchy
1656 * We want to do more targeted reclaim.
1657 * excess >> 2 is not to excessive so as to
1658 * reclaim too much, nor too less that we keep
1659 * coming back to reclaim from this cgroup
1661 if (total
>= (excess
>> 2) ||
1662 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
))
1667 if (!mem_cgroup_reclaimable(victim
, false))
1669 total
+= mem_cgroup_shrink_node_zone(victim
, gfp_mask
, false,
1671 *total_scanned
+= nr_scanned
;
1672 if (!res_counter_soft_limit_excess(&root_memcg
->res
))
1675 mem_cgroup_iter_break(root_memcg
, victim
);
1680 * Check OOM-Killer is already running under our hierarchy.
1681 * If someone is running, return false.
1682 * Has to be called with memcg_oom_lock
1684 static bool mem_cgroup_oom_lock(struct mem_cgroup
*memcg
)
1686 struct mem_cgroup
*iter
, *failed
= NULL
;
1688 for_each_mem_cgroup_tree(iter
, memcg
) {
1689 if (iter
->oom_lock
) {
1691 * this subtree of our hierarchy is already locked
1692 * so we cannot give a lock.
1695 mem_cgroup_iter_break(memcg
, iter
);
1698 iter
->oom_lock
= true;
1705 * OK, we failed to lock the whole subtree so we have to clean up
1706 * what we set up to the failing subtree
1708 for_each_mem_cgroup_tree(iter
, memcg
) {
1709 if (iter
== failed
) {
1710 mem_cgroup_iter_break(memcg
, iter
);
1713 iter
->oom_lock
= false;
1719 * Has to be called with memcg_oom_lock
1721 static int mem_cgroup_oom_unlock(struct mem_cgroup
*memcg
)
1723 struct mem_cgroup
*iter
;
1725 for_each_mem_cgroup_tree(iter
, memcg
)
1726 iter
->oom_lock
= false;
1730 static void mem_cgroup_mark_under_oom(struct mem_cgroup
*memcg
)
1732 struct mem_cgroup
*iter
;
1734 for_each_mem_cgroup_tree(iter
, memcg
)
1735 atomic_inc(&iter
->under_oom
);
1738 static void mem_cgroup_unmark_under_oom(struct mem_cgroup
*memcg
)
1740 struct mem_cgroup
*iter
;
1743 * When a new child is created while the hierarchy is under oom,
1744 * mem_cgroup_oom_lock() may not be called. We have to use
1745 * atomic_add_unless() here.
1747 for_each_mem_cgroup_tree(iter
, memcg
)
1748 atomic_add_unless(&iter
->under_oom
, -1, 0);
1751 static DEFINE_SPINLOCK(memcg_oom_lock
);
1752 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1754 struct oom_wait_info
{
1755 struct mem_cgroup
*mem
;
1759 static int memcg_oom_wake_function(wait_queue_t
*wait
,
1760 unsigned mode
, int sync
, void *arg
)
1762 struct mem_cgroup
*wake_memcg
= (struct mem_cgroup
*)arg
,
1764 struct oom_wait_info
*oom_wait_info
;
1766 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1767 oom_wait_memcg
= oom_wait_info
->mem
;
1770 * Both of oom_wait_info->mem and wake_mem are stable under us.
1771 * Then we can use css_is_ancestor without taking care of RCU.
1773 if (!mem_cgroup_same_or_subtree(oom_wait_memcg
, wake_memcg
)
1774 && !mem_cgroup_same_or_subtree(wake_memcg
, oom_wait_memcg
))
1776 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1779 static void memcg_wakeup_oom(struct mem_cgroup
*memcg
)
1781 /* for filtering, pass "memcg" as argument. */
1782 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, memcg
);
1785 static void memcg_oom_recover(struct mem_cgroup
*memcg
)
1787 if (memcg
&& atomic_read(&memcg
->under_oom
))
1788 memcg_wakeup_oom(memcg
);
1792 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1794 bool mem_cgroup_handle_oom(struct mem_cgroup
*memcg
, gfp_t mask
)
1796 struct oom_wait_info owait
;
1797 bool locked
, need_to_kill
;
1800 owait
.wait
.flags
= 0;
1801 owait
.wait
.func
= memcg_oom_wake_function
;
1802 owait
.wait
.private = current
;
1803 INIT_LIST_HEAD(&owait
.wait
.task_list
);
1804 need_to_kill
= true;
1805 mem_cgroup_mark_under_oom(memcg
);
1807 /* At first, try to OOM lock hierarchy under memcg.*/
1808 spin_lock(&memcg_oom_lock
);
1809 locked
= mem_cgroup_oom_lock(memcg
);
1811 * Even if signal_pending(), we can't quit charge() loop without
1812 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1813 * under OOM is always welcomed, use TASK_KILLABLE here.
1815 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1816 if (!locked
|| memcg
->oom_kill_disable
)
1817 need_to_kill
= false;
1819 mem_cgroup_oom_notify(memcg
);
1820 spin_unlock(&memcg_oom_lock
);
1823 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1824 mem_cgroup_out_of_memory(memcg
, mask
);
1827 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1829 spin_lock(&memcg_oom_lock
);
1831 mem_cgroup_oom_unlock(memcg
);
1832 memcg_wakeup_oom(memcg
);
1833 spin_unlock(&memcg_oom_lock
);
1835 mem_cgroup_unmark_under_oom(memcg
);
1837 if (test_thread_flag(TIF_MEMDIE
) || fatal_signal_pending(current
))
1839 /* Give chance to dying process */
1840 schedule_timeout_uninterruptible(1);
1845 * Currently used to update mapped file statistics, but the routine can be
1846 * generalized to update other statistics as well.
1848 * Notes: Race condition
1850 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1851 * it tends to be costly. But considering some conditions, we doesn't need
1852 * to do so _always_.
1854 * Considering "charge", lock_page_cgroup() is not required because all
1855 * file-stat operations happen after a page is attached to radix-tree. There
1856 * are no race with "charge".
1858 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1859 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1860 * if there are race with "uncharge". Statistics itself is properly handled
1863 * Considering "move", this is an only case we see a race. To make the race
1864 * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are
1865 * possibility of race condition. If there is, we take a lock.
1868 void mem_cgroup_update_page_stat(struct page
*page
,
1869 enum mem_cgroup_page_stat_item idx
, int val
)
1871 struct mem_cgroup
*memcg
;
1872 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1873 bool need_unlock
= false;
1874 unsigned long uninitialized_var(flags
);
1876 if (mem_cgroup_disabled())
1880 memcg
= pc
->mem_cgroup
;
1881 if (unlikely(!memcg
|| !PageCgroupUsed(pc
)))
1883 /* pc->mem_cgroup is unstable ? */
1884 if (unlikely(mem_cgroup_stealed(memcg
)) || PageTransHuge(page
)) {
1885 /* take a lock against to access pc->mem_cgroup */
1886 move_lock_page_cgroup(pc
, &flags
);
1888 memcg
= pc
->mem_cgroup
;
1889 if (!memcg
|| !PageCgroupUsed(pc
))
1894 case MEMCG_NR_FILE_MAPPED
:
1896 SetPageCgroupFileMapped(pc
);
1897 else if (!page_mapped(page
))
1898 ClearPageCgroupFileMapped(pc
);
1899 idx
= MEM_CGROUP_STAT_FILE_MAPPED
;
1905 this_cpu_add(memcg
->stat
->count
[idx
], val
);
1908 if (unlikely(need_unlock
))
1909 move_unlock_page_cgroup(pc
, &flags
);
1913 EXPORT_SYMBOL(mem_cgroup_update_page_stat
);
1916 * size of first charge trial. "32" comes from vmscan.c's magic value.
1917 * TODO: maybe necessary to use big numbers in big irons.
1919 #define CHARGE_BATCH 32U
1920 struct memcg_stock_pcp
{
1921 struct mem_cgroup
*cached
; /* this never be root cgroup */
1922 unsigned int nr_pages
;
1923 struct work_struct work
;
1924 unsigned long flags
;
1925 #define FLUSHING_CACHED_CHARGE (0)
1927 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
1928 static DEFINE_MUTEX(percpu_charge_mutex
);
1931 * Try to consume stocked charge on this cpu. If success, one page is consumed
1932 * from local stock and true is returned. If the stock is 0 or charges from a
1933 * cgroup which is not current target, returns false. This stock will be
1936 static bool consume_stock(struct mem_cgroup
*memcg
)
1938 struct memcg_stock_pcp
*stock
;
1941 stock
= &get_cpu_var(memcg_stock
);
1942 if (memcg
== stock
->cached
&& stock
->nr_pages
)
1944 else /* need to call res_counter_charge */
1946 put_cpu_var(memcg_stock
);
1951 * Returns stocks cached in percpu to res_counter and reset cached information.
1953 static void drain_stock(struct memcg_stock_pcp
*stock
)
1955 struct mem_cgroup
*old
= stock
->cached
;
1957 if (stock
->nr_pages
) {
1958 unsigned long bytes
= stock
->nr_pages
* PAGE_SIZE
;
1960 res_counter_uncharge(&old
->res
, bytes
);
1961 if (do_swap_account
)
1962 res_counter_uncharge(&old
->memsw
, bytes
);
1963 stock
->nr_pages
= 0;
1965 stock
->cached
= NULL
;
1969 * This must be called under preempt disabled or must be called by
1970 * a thread which is pinned to local cpu.
1972 static void drain_local_stock(struct work_struct
*dummy
)
1974 struct memcg_stock_pcp
*stock
= &__get_cpu_var(memcg_stock
);
1976 clear_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
);
1980 * Cache charges(val) which is from res_counter, to local per_cpu area.
1981 * This will be consumed by consume_stock() function, later.
1983 static void refill_stock(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
1985 struct memcg_stock_pcp
*stock
= &get_cpu_var(memcg_stock
);
1987 if (stock
->cached
!= memcg
) { /* reset if necessary */
1989 stock
->cached
= memcg
;
1991 stock
->nr_pages
+= nr_pages
;
1992 put_cpu_var(memcg_stock
);
1996 * Drains all per-CPU charge caches for given root_memcg resp. subtree
1997 * of the hierarchy under it. sync flag says whether we should block
1998 * until the work is done.
2000 static void drain_all_stock(struct mem_cgroup
*root_memcg
, bool sync
)
2004 /* Notify other cpus that system-wide "drain" is running */
2007 for_each_online_cpu(cpu
) {
2008 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2009 struct mem_cgroup
*memcg
;
2011 memcg
= stock
->cached
;
2012 if (!memcg
|| !stock
->nr_pages
)
2014 if (!mem_cgroup_same_or_subtree(root_memcg
, memcg
))
2016 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
)) {
2018 drain_local_stock(&stock
->work
);
2020 schedule_work_on(cpu
, &stock
->work
);
2028 for_each_online_cpu(cpu
) {
2029 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2030 if (test_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
))
2031 flush_work(&stock
->work
);
2038 * Tries to drain stocked charges in other cpus. This function is asynchronous
2039 * and just put a work per cpu for draining localy on each cpu. Caller can
2040 * expects some charges will be back to res_counter later but cannot wait for
2043 static void drain_all_stock_async(struct mem_cgroup
*root_memcg
)
2046 * If someone calls draining, avoid adding more kworker runs.
2048 if (!mutex_trylock(&percpu_charge_mutex
))
2050 drain_all_stock(root_memcg
, false);
2051 mutex_unlock(&percpu_charge_mutex
);
2054 /* This is a synchronous drain interface. */
2055 static void drain_all_stock_sync(struct mem_cgroup
*root_memcg
)
2057 /* called when force_empty is called */
2058 mutex_lock(&percpu_charge_mutex
);
2059 drain_all_stock(root_memcg
, true);
2060 mutex_unlock(&percpu_charge_mutex
);
2064 * This function drains percpu counter value from DEAD cpu and
2065 * move it to local cpu. Note that this function can be preempted.
2067 static void mem_cgroup_drain_pcp_counter(struct mem_cgroup
*memcg
, int cpu
)
2071 spin_lock(&memcg
->pcp_counter_lock
);
2072 for (i
= 0; i
< MEM_CGROUP_STAT_DATA
; i
++) {
2073 long x
= per_cpu(memcg
->stat
->count
[i
], cpu
);
2075 per_cpu(memcg
->stat
->count
[i
], cpu
) = 0;
2076 memcg
->nocpu_base
.count
[i
] += x
;
2078 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
2079 unsigned long x
= per_cpu(memcg
->stat
->events
[i
], cpu
);
2081 per_cpu(memcg
->stat
->events
[i
], cpu
) = 0;
2082 memcg
->nocpu_base
.events
[i
] += x
;
2084 /* need to clear ON_MOVE value, works as a kind of lock. */
2085 per_cpu(memcg
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) = 0;
2086 spin_unlock(&memcg
->pcp_counter_lock
);
2089 static void synchronize_mem_cgroup_on_move(struct mem_cgroup
*memcg
, int cpu
)
2091 int idx
= MEM_CGROUP_ON_MOVE
;
2093 spin_lock(&memcg
->pcp_counter_lock
);
2094 per_cpu(memcg
->stat
->count
[idx
], cpu
) = memcg
->nocpu_base
.count
[idx
];
2095 spin_unlock(&memcg
->pcp_counter_lock
);
2098 static int __cpuinit
memcg_cpu_hotplug_callback(struct notifier_block
*nb
,
2099 unsigned long action
,
2102 int cpu
= (unsigned long)hcpu
;
2103 struct memcg_stock_pcp
*stock
;
2104 struct mem_cgroup
*iter
;
2106 if ((action
== CPU_ONLINE
)) {
2107 for_each_mem_cgroup(iter
)
2108 synchronize_mem_cgroup_on_move(iter
, cpu
);
2112 if ((action
!= CPU_DEAD
) || action
!= CPU_DEAD_FROZEN
)
2115 for_each_mem_cgroup(iter
)
2116 mem_cgroup_drain_pcp_counter(iter
, cpu
);
2118 stock
= &per_cpu(memcg_stock
, cpu
);
2124 /* See __mem_cgroup_try_charge() for details */
2126 CHARGE_OK
, /* success */
2127 CHARGE_RETRY
, /* need to retry but retry is not bad */
2128 CHARGE_NOMEM
, /* we can't do more. return -ENOMEM */
2129 CHARGE_WOULDBLOCK
, /* GFP_WAIT wasn't set and no enough res. */
2130 CHARGE_OOM_DIE
, /* the current is killed because of OOM */
2133 static int mem_cgroup_do_charge(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
2134 unsigned int nr_pages
, bool oom_check
)
2136 unsigned long csize
= nr_pages
* PAGE_SIZE
;
2137 struct mem_cgroup
*mem_over_limit
;
2138 struct res_counter
*fail_res
;
2139 unsigned long flags
= 0;
2142 ret
= res_counter_charge(&memcg
->res
, csize
, &fail_res
);
2145 if (!do_swap_account
)
2147 ret
= res_counter_charge(&memcg
->memsw
, csize
, &fail_res
);
2151 res_counter_uncharge(&memcg
->res
, csize
);
2152 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, memsw
);
2153 flags
|= MEM_CGROUP_RECLAIM_NOSWAP
;
2155 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, res
);
2157 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2158 * of regular pages (CHARGE_BATCH), or a single regular page (1).
2160 * Never reclaim on behalf of optional batching, retry with a
2161 * single page instead.
2163 if (nr_pages
== CHARGE_BATCH
)
2164 return CHARGE_RETRY
;
2166 if (!(gfp_mask
& __GFP_WAIT
))
2167 return CHARGE_WOULDBLOCK
;
2169 ret
= mem_cgroup_reclaim(mem_over_limit
, gfp_mask
, flags
);
2170 if (mem_cgroup_margin(mem_over_limit
) >= nr_pages
)
2171 return CHARGE_RETRY
;
2173 * Even though the limit is exceeded at this point, reclaim
2174 * may have been able to free some pages. Retry the charge
2175 * before killing the task.
2177 * Only for regular pages, though: huge pages are rather
2178 * unlikely to succeed so close to the limit, and we fall back
2179 * to regular pages anyway in case of failure.
2181 if (nr_pages
== 1 && ret
)
2182 return CHARGE_RETRY
;
2185 * At task move, charge accounts can be doubly counted. So, it's
2186 * better to wait until the end of task_move if something is going on.
2188 if (mem_cgroup_wait_acct_move(mem_over_limit
))
2189 return CHARGE_RETRY
;
2191 /* If we don't need to call oom-killer at el, return immediately */
2193 return CHARGE_NOMEM
;
2195 if (!mem_cgroup_handle_oom(mem_over_limit
, gfp_mask
))
2196 return CHARGE_OOM_DIE
;
2198 return CHARGE_RETRY
;
2202 * __mem_cgroup_try_charge() does
2203 * 1. detect memcg to be charged against from passed *mm and *ptr,
2204 * 2. update res_counter
2205 * 3. call memory reclaim if necessary.
2207 * In some special case, if the task is fatal, fatal_signal_pending() or
2208 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2209 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2210 * as possible without any hazards. 2: all pages should have a valid
2211 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2212 * pointer, that is treated as a charge to root_mem_cgroup.
2214 * So __mem_cgroup_try_charge() will return
2215 * 0 ... on success, filling *ptr with a valid memcg pointer.
2216 * -ENOMEM ... charge failure because of resource limits.
2217 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2219 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2220 * the oom-killer can be invoked.
2222 static int __mem_cgroup_try_charge(struct mm_struct
*mm
,
2224 unsigned int nr_pages
,
2225 struct mem_cgroup
**ptr
,
2228 unsigned int batch
= max(CHARGE_BATCH
, nr_pages
);
2229 int nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2230 struct mem_cgroup
*memcg
= NULL
;
2234 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2235 * in system level. So, allow to go ahead dying process in addition to
2238 if (unlikely(test_thread_flag(TIF_MEMDIE
)
2239 || fatal_signal_pending(current
)))
2243 * We always charge the cgroup the mm_struct belongs to.
2244 * The mm_struct's mem_cgroup changes on task migration if the
2245 * thread group leader migrates. It's possible that mm is not
2246 * set, if so charge the init_mm (happens for pagecache usage).
2249 *ptr
= root_mem_cgroup
;
2251 if (*ptr
) { /* css should be a valid one */
2253 VM_BUG_ON(css_is_removed(&memcg
->css
));
2254 if (mem_cgroup_is_root(memcg
))
2256 if (nr_pages
== 1 && consume_stock(memcg
))
2258 css_get(&memcg
->css
);
2260 struct task_struct
*p
;
2263 p
= rcu_dereference(mm
->owner
);
2265 * Because we don't have task_lock(), "p" can exit.
2266 * In that case, "memcg" can point to root or p can be NULL with
2267 * race with swapoff. Then, we have small risk of mis-accouning.
2268 * But such kind of mis-account by race always happens because
2269 * we don't have cgroup_mutex(). It's overkill and we allo that
2271 * (*) swapoff at el will charge against mm-struct not against
2272 * task-struct. So, mm->owner can be NULL.
2274 memcg
= mem_cgroup_from_task(p
);
2276 memcg
= root_mem_cgroup
;
2277 if (mem_cgroup_is_root(memcg
)) {
2281 if (nr_pages
== 1 && consume_stock(memcg
)) {
2283 * It seems dagerous to access memcg without css_get().
2284 * But considering how consume_stok works, it's not
2285 * necessary. If consume_stock success, some charges
2286 * from this memcg are cached on this cpu. So, we
2287 * don't need to call css_get()/css_tryget() before
2288 * calling consume_stock().
2293 /* after here, we may be blocked. we need to get refcnt */
2294 if (!css_tryget(&memcg
->css
)) {
2304 /* If killed, bypass charge */
2305 if (fatal_signal_pending(current
)) {
2306 css_put(&memcg
->css
);
2311 if (oom
&& !nr_oom_retries
) {
2313 nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2316 ret
= mem_cgroup_do_charge(memcg
, gfp_mask
, batch
, oom_check
);
2320 case CHARGE_RETRY
: /* not in OOM situation but retry */
2322 css_put(&memcg
->css
);
2325 case CHARGE_WOULDBLOCK
: /* !__GFP_WAIT */
2326 css_put(&memcg
->css
);
2328 case CHARGE_NOMEM
: /* OOM routine works */
2330 css_put(&memcg
->css
);
2333 /* If oom, we never return -ENOMEM */
2336 case CHARGE_OOM_DIE
: /* Killed by OOM Killer */
2337 css_put(&memcg
->css
);
2340 } while (ret
!= CHARGE_OK
);
2342 if (batch
> nr_pages
)
2343 refill_stock(memcg
, batch
- nr_pages
);
2344 css_put(&memcg
->css
);
2352 *ptr
= root_mem_cgroup
;
2357 * Somemtimes we have to undo a charge we got by try_charge().
2358 * This function is for that and do uncharge, put css's refcnt.
2359 * gotten by try_charge().
2361 static void __mem_cgroup_cancel_charge(struct mem_cgroup
*memcg
,
2362 unsigned int nr_pages
)
2364 if (!mem_cgroup_is_root(memcg
)) {
2365 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2367 res_counter_uncharge(&memcg
->res
, bytes
);
2368 if (do_swap_account
)
2369 res_counter_uncharge(&memcg
->memsw
, bytes
);
2374 * A helper function to get mem_cgroup from ID. must be called under
2375 * rcu_read_lock(). The caller must check css_is_removed() or some if
2376 * it's concern. (dropping refcnt from swap can be called against removed
2379 static struct mem_cgroup
*mem_cgroup_lookup(unsigned short id
)
2381 struct cgroup_subsys_state
*css
;
2383 /* ID 0 is unused ID */
2386 css
= css_lookup(&mem_cgroup_subsys
, id
);
2389 return container_of(css
, struct mem_cgroup
, css
);
2392 struct mem_cgroup
*try_get_mem_cgroup_from_page(struct page
*page
)
2394 struct mem_cgroup
*memcg
= NULL
;
2395 struct page_cgroup
*pc
;
2399 VM_BUG_ON(!PageLocked(page
));
2401 pc
= lookup_page_cgroup(page
);
2402 lock_page_cgroup(pc
);
2403 if (PageCgroupUsed(pc
)) {
2404 memcg
= pc
->mem_cgroup
;
2405 if (memcg
&& !css_tryget(&memcg
->css
))
2407 } else if (PageSwapCache(page
)) {
2408 ent
.val
= page_private(page
);
2409 id
= lookup_swap_cgroup_id(ent
);
2411 memcg
= mem_cgroup_lookup(id
);
2412 if (memcg
&& !css_tryget(&memcg
->css
))
2416 unlock_page_cgroup(pc
);
2420 static void __mem_cgroup_commit_charge(struct mem_cgroup
*memcg
,
2422 unsigned int nr_pages
,
2423 struct page_cgroup
*pc
,
2424 enum charge_type ctype
,
2427 struct zone
*uninitialized_var(zone
);
2428 bool was_on_lru
= false;
2430 lock_page_cgroup(pc
);
2431 if (unlikely(PageCgroupUsed(pc
))) {
2432 unlock_page_cgroup(pc
);
2433 __mem_cgroup_cancel_charge(memcg
, nr_pages
);
2437 * we don't need page_cgroup_lock about tail pages, becase they are not
2438 * accessed by any other context at this point.
2442 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2443 * may already be on some other mem_cgroup's LRU. Take care of it.
2446 zone
= page_zone(page
);
2447 spin_lock_irq(&zone
->lru_lock
);
2448 if (PageLRU(page
)) {
2450 del_page_from_lru_list(zone
, page
, page_lru(page
));
2455 pc
->mem_cgroup
= memcg
;
2457 * We access a page_cgroup asynchronously without lock_page_cgroup().
2458 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2459 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2460 * before USED bit, we need memory barrier here.
2461 * See mem_cgroup_add_lru_list(), etc.
2465 case MEM_CGROUP_CHARGE_TYPE_CACHE
:
2466 case MEM_CGROUP_CHARGE_TYPE_SHMEM
:
2467 SetPageCgroupCache(pc
);
2468 SetPageCgroupUsed(pc
);
2470 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2471 ClearPageCgroupCache(pc
);
2472 SetPageCgroupUsed(pc
);
2480 VM_BUG_ON(PageLRU(page
));
2482 add_page_to_lru_list(zone
, page
, page_lru(page
));
2484 spin_unlock_irq(&zone
->lru_lock
);
2487 mem_cgroup_charge_statistics(memcg
, PageCgroupCache(pc
), nr_pages
);
2488 unlock_page_cgroup(pc
);
2491 * "charge_statistics" updated event counter. Then, check it.
2492 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2493 * if they exceeds softlimit.
2495 memcg_check_events(memcg
, page
);
2498 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2500 #define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\
2501 (1 << PCG_MIGRATION))
2503 * Because tail pages are not marked as "used", set it. We're under
2504 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2505 * charge/uncharge will be never happen and move_account() is done under
2506 * compound_lock(), so we don't have to take care of races.
2508 void mem_cgroup_split_huge_fixup(struct page
*head
)
2510 struct page_cgroup
*head_pc
= lookup_page_cgroup(head
);
2511 struct page_cgroup
*pc
;
2514 if (mem_cgroup_disabled())
2516 for (i
= 1; i
< HPAGE_PMD_NR
; i
++) {
2518 pc
->mem_cgroup
= head_pc
->mem_cgroup
;
2519 smp_wmb();/* see __commit_charge() */
2520 pc
->flags
= head_pc
->flags
& ~PCGF_NOCOPY_AT_SPLIT
;
2523 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2526 * mem_cgroup_move_account - move account of the page
2528 * @nr_pages: number of regular pages (>1 for huge pages)
2529 * @pc: page_cgroup of the page.
2530 * @from: mem_cgroup which the page is moved from.
2531 * @to: mem_cgroup which the page is moved to. @from != @to.
2532 * @uncharge: whether we should call uncharge and css_put against @from.
2534 * The caller must confirm following.
2535 * - page is not on LRU (isolate_page() is useful.)
2536 * - compound_lock is held when nr_pages > 1
2538 * This function doesn't do "charge" nor css_get to new cgroup. It should be
2539 * done by a caller(__mem_cgroup_try_charge would be useful). If @uncharge is
2540 * true, this function does "uncharge" from old cgroup, but it doesn't if
2541 * @uncharge is false, so a caller should do "uncharge".
2543 static int mem_cgroup_move_account(struct page
*page
,
2544 unsigned int nr_pages
,
2545 struct page_cgroup
*pc
,
2546 struct mem_cgroup
*from
,
2547 struct mem_cgroup
*to
,
2550 unsigned long flags
;
2553 VM_BUG_ON(from
== to
);
2554 VM_BUG_ON(PageLRU(page
));
2556 * The page is isolated from LRU. So, collapse function
2557 * will not handle this page. But page splitting can happen.
2558 * Do this check under compound_page_lock(). The caller should
2562 if (nr_pages
> 1 && !PageTransHuge(page
))
2565 lock_page_cgroup(pc
);
2568 if (!PageCgroupUsed(pc
) || pc
->mem_cgroup
!= from
)
2571 move_lock_page_cgroup(pc
, &flags
);
2573 if (PageCgroupFileMapped(pc
)) {
2574 /* Update mapped_file data for mem_cgroup */
2576 __this_cpu_dec(from
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2577 __this_cpu_inc(to
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2580 mem_cgroup_charge_statistics(from
, PageCgroupCache(pc
), -nr_pages
);
2582 /* This is not "cancel", but cancel_charge does all we need. */
2583 __mem_cgroup_cancel_charge(from
, nr_pages
);
2585 /* caller should have done css_get */
2586 pc
->mem_cgroup
= to
;
2587 mem_cgroup_charge_statistics(to
, PageCgroupCache(pc
), nr_pages
);
2589 * We charges against "to" which may not have any tasks. Then, "to"
2590 * can be under rmdir(). But in current implementation, caller of
2591 * this function is just force_empty() and move charge, so it's
2592 * guaranteed that "to" is never removed. So, we don't check rmdir
2595 move_unlock_page_cgroup(pc
, &flags
);
2598 unlock_page_cgroup(pc
);
2602 memcg_check_events(to
, page
);
2603 memcg_check_events(from
, page
);
2609 * move charges to its parent.
2612 static int mem_cgroup_move_parent(struct page
*page
,
2613 struct page_cgroup
*pc
,
2614 struct mem_cgroup
*child
,
2617 struct cgroup
*cg
= child
->css
.cgroup
;
2618 struct cgroup
*pcg
= cg
->parent
;
2619 struct mem_cgroup
*parent
;
2620 unsigned int nr_pages
;
2621 unsigned long uninitialized_var(flags
);
2629 if (!get_page_unless_zero(page
))
2631 if (isolate_lru_page(page
))
2634 nr_pages
= hpage_nr_pages(page
);
2636 parent
= mem_cgroup_from_cont(pcg
);
2637 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, nr_pages
, &parent
, false);
2642 flags
= compound_lock_irqsave(page
);
2644 ret
= mem_cgroup_move_account(page
, nr_pages
, pc
, child
, parent
, true);
2646 __mem_cgroup_cancel_charge(parent
, nr_pages
);
2649 compound_unlock_irqrestore(page
, flags
);
2651 putback_lru_page(page
);
2659 * Charge the memory controller for page usage.
2661 * 0 if the charge was successful
2662 * < 0 if the cgroup is over its limit
2664 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
2665 gfp_t gfp_mask
, enum charge_type ctype
)
2667 struct mem_cgroup
*memcg
= NULL
;
2668 unsigned int nr_pages
= 1;
2669 struct page_cgroup
*pc
;
2673 if (PageTransHuge(page
)) {
2674 nr_pages
<<= compound_order(page
);
2675 VM_BUG_ON(!PageTransHuge(page
));
2677 * Never OOM-kill a process for a huge page. The
2678 * fault handler will fall back to regular pages.
2683 pc
= lookup_page_cgroup(page
);
2684 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, nr_pages
, &memcg
, oom
);
2687 __mem_cgroup_commit_charge(memcg
, page
, nr_pages
, pc
, ctype
, false);
2691 int mem_cgroup_newpage_charge(struct page
*page
,
2692 struct mm_struct
*mm
, gfp_t gfp_mask
)
2694 if (mem_cgroup_disabled())
2696 VM_BUG_ON(page_mapped(page
));
2697 VM_BUG_ON(page
->mapping
&& !PageAnon(page
));
2699 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2700 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2704 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2705 enum charge_type ctype
);
2707 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
2710 struct mem_cgroup
*memcg
= NULL
;
2711 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
2714 if (mem_cgroup_disabled())
2716 if (PageCompound(page
))
2721 if (!page_is_file_cache(page
))
2722 type
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
2724 if (!PageSwapCache(page
))
2725 ret
= mem_cgroup_charge_common(page
, mm
, gfp_mask
, type
);
2726 else { /* page is swapcache/shmem */
2727 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &memcg
);
2729 __mem_cgroup_commit_charge_swapin(page
, memcg
, type
);
2735 * While swap-in, try_charge -> commit or cancel, the page is locked.
2736 * And when try_charge() successfully returns, one refcnt to memcg without
2737 * struct page_cgroup is acquired. This refcnt will be consumed by
2738 * "commit()" or removed by "cancel()"
2740 int mem_cgroup_try_charge_swapin(struct mm_struct
*mm
,
2742 gfp_t mask
, struct mem_cgroup
**memcgp
)
2744 struct mem_cgroup
*memcg
;
2749 if (mem_cgroup_disabled())
2752 if (!do_swap_account
)
2755 * A racing thread's fault, or swapoff, may have already updated
2756 * the pte, and even removed page from swap cache: in those cases
2757 * do_swap_page()'s pte_same() test will fail; but there's also a
2758 * KSM case which does need to charge the page.
2760 if (!PageSwapCache(page
))
2762 memcg
= try_get_mem_cgroup_from_page(page
);
2766 ret
= __mem_cgroup_try_charge(NULL
, mask
, 1, memcgp
, true);
2767 css_put(&memcg
->css
);
2774 ret
= __mem_cgroup_try_charge(mm
, mask
, 1, memcgp
, true);
2781 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*memcg
,
2782 enum charge_type ctype
)
2784 struct page_cgroup
*pc
;
2786 if (mem_cgroup_disabled())
2790 cgroup_exclude_rmdir(&memcg
->css
);
2792 pc
= lookup_page_cgroup(page
);
2793 __mem_cgroup_commit_charge(memcg
, page
, 1, pc
, ctype
, true);
2795 * Now swap is on-memory. This means this page may be
2796 * counted both as mem and swap....double count.
2797 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2798 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2799 * may call delete_from_swap_cache() before reach here.
2801 if (do_swap_account
&& PageSwapCache(page
)) {
2802 swp_entry_t ent
= {.val
= page_private(page
)};
2803 struct mem_cgroup
*swap_memcg
;
2806 id
= swap_cgroup_record(ent
, 0);
2808 swap_memcg
= mem_cgroup_lookup(id
);
2811 * This recorded memcg can be obsolete one. So, avoid
2812 * calling css_tryget
2814 if (!mem_cgroup_is_root(swap_memcg
))
2815 res_counter_uncharge(&swap_memcg
->memsw
,
2817 mem_cgroup_swap_statistics(swap_memcg
, false);
2818 mem_cgroup_put(swap_memcg
);
2823 * At swapin, we may charge account against cgroup which has no tasks.
2824 * So, rmdir()->pre_destroy() can be called while we do this charge.
2825 * In that case, we need to call pre_destroy() again. check it here.
2827 cgroup_release_and_wakeup_rmdir(&memcg
->css
);
2830 void mem_cgroup_commit_charge_swapin(struct page
*page
,
2831 struct mem_cgroup
*memcg
)
2833 __mem_cgroup_commit_charge_swapin(page
, memcg
,
2834 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2837 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup
*memcg
)
2839 if (mem_cgroup_disabled())
2843 __mem_cgroup_cancel_charge(memcg
, 1);
2846 static void mem_cgroup_do_uncharge(struct mem_cgroup
*memcg
,
2847 unsigned int nr_pages
,
2848 const enum charge_type ctype
)
2850 struct memcg_batch_info
*batch
= NULL
;
2851 bool uncharge_memsw
= true;
2853 /* If swapout, usage of swap doesn't decrease */
2854 if (!do_swap_account
|| ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
)
2855 uncharge_memsw
= false;
2857 batch
= ¤t
->memcg_batch
;
2859 * In usual, we do css_get() when we remember memcg pointer.
2860 * But in this case, we keep res->usage until end of a series of
2861 * uncharges. Then, it's ok to ignore memcg's refcnt.
2864 batch
->memcg
= memcg
;
2866 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2867 * In those cases, all pages freed continuously can be expected to be in
2868 * the same cgroup and we have chance to coalesce uncharges.
2869 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2870 * because we want to do uncharge as soon as possible.
2873 if (!batch
->do_batch
|| test_thread_flag(TIF_MEMDIE
))
2874 goto direct_uncharge
;
2877 goto direct_uncharge
;
2880 * In typical case, batch->memcg == mem. This means we can
2881 * merge a series of uncharges to an uncharge of res_counter.
2882 * If not, we uncharge res_counter ony by one.
2884 if (batch
->memcg
!= memcg
)
2885 goto direct_uncharge
;
2886 /* remember freed charge and uncharge it later */
2889 batch
->memsw_nr_pages
++;
2892 res_counter_uncharge(&memcg
->res
, nr_pages
* PAGE_SIZE
);
2894 res_counter_uncharge(&memcg
->memsw
, nr_pages
* PAGE_SIZE
);
2895 if (unlikely(batch
->memcg
!= memcg
))
2896 memcg_oom_recover(memcg
);
2901 * uncharge if !page_mapped(page)
2903 static struct mem_cgroup
*
2904 __mem_cgroup_uncharge_common(struct page
*page
, enum charge_type ctype
)
2906 struct mem_cgroup
*memcg
= NULL
;
2907 unsigned int nr_pages
= 1;
2908 struct page_cgroup
*pc
;
2910 if (mem_cgroup_disabled())
2913 if (PageSwapCache(page
))
2916 if (PageTransHuge(page
)) {
2917 nr_pages
<<= compound_order(page
);
2918 VM_BUG_ON(!PageTransHuge(page
));
2921 * Check if our page_cgroup is valid
2923 pc
= lookup_page_cgroup(page
);
2924 if (unlikely(!PageCgroupUsed(pc
)))
2927 lock_page_cgroup(pc
);
2929 memcg
= pc
->mem_cgroup
;
2931 if (!PageCgroupUsed(pc
))
2935 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2936 case MEM_CGROUP_CHARGE_TYPE_DROP
:
2937 /* See mem_cgroup_prepare_migration() */
2938 if (page_mapped(page
) || PageCgroupMigration(pc
))
2941 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT
:
2942 if (!PageAnon(page
)) { /* Shared memory */
2943 if (page
->mapping
&& !page_is_file_cache(page
))
2945 } else if (page_mapped(page
)) /* Anon */
2952 mem_cgroup_charge_statistics(memcg
, PageCgroupCache(pc
), -nr_pages
);
2954 ClearPageCgroupUsed(pc
);
2956 * pc->mem_cgroup is not cleared here. It will be accessed when it's
2957 * freed from LRU. This is safe because uncharged page is expected not
2958 * to be reused (freed soon). Exception is SwapCache, it's handled by
2959 * special functions.
2962 unlock_page_cgroup(pc
);
2964 * even after unlock, we have memcg->res.usage here and this memcg
2965 * will never be freed.
2967 memcg_check_events(memcg
, page
);
2968 if (do_swap_account
&& ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
) {
2969 mem_cgroup_swap_statistics(memcg
, true);
2970 mem_cgroup_get(memcg
);
2972 if (!mem_cgroup_is_root(memcg
))
2973 mem_cgroup_do_uncharge(memcg
, nr_pages
, ctype
);
2978 unlock_page_cgroup(pc
);
2982 void mem_cgroup_uncharge_page(struct page
*page
)
2985 if (page_mapped(page
))
2987 VM_BUG_ON(page
->mapping
&& !PageAnon(page
));
2988 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2991 void mem_cgroup_uncharge_cache_page(struct page
*page
)
2993 VM_BUG_ON(page_mapped(page
));
2994 VM_BUG_ON(page
->mapping
);
2995 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_CACHE
);
2999 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
3000 * In that cases, pages are freed continuously and we can expect pages
3001 * are in the same memcg. All these calls itself limits the number of
3002 * pages freed at once, then uncharge_start/end() is called properly.
3003 * This may be called prural(2) times in a context,
3006 void mem_cgroup_uncharge_start(void)
3008 current
->memcg_batch
.do_batch
++;
3009 /* We can do nest. */
3010 if (current
->memcg_batch
.do_batch
== 1) {
3011 current
->memcg_batch
.memcg
= NULL
;
3012 current
->memcg_batch
.nr_pages
= 0;
3013 current
->memcg_batch
.memsw_nr_pages
= 0;
3017 void mem_cgroup_uncharge_end(void)
3019 struct memcg_batch_info
*batch
= ¤t
->memcg_batch
;
3021 if (!batch
->do_batch
)
3025 if (batch
->do_batch
) /* If stacked, do nothing. */
3031 * This "batch->memcg" is valid without any css_get/put etc...
3032 * bacause we hide charges behind us.
3034 if (batch
->nr_pages
)
3035 res_counter_uncharge(&batch
->memcg
->res
,
3036 batch
->nr_pages
* PAGE_SIZE
);
3037 if (batch
->memsw_nr_pages
)
3038 res_counter_uncharge(&batch
->memcg
->memsw
,
3039 batch
->memsw_nr_pages
* PAGE_SIZE
);
3040 memcg_oom_recover(batch
->memcg
);
3041 /* forget this pointer (for sanity check) */
3042 batch
->memcg
= NULL
;
3047 * called after __delete_from_swap_cache() and drop "page" account.
3048 * memcg information is recorded to swap_cgroup of "ent"
3051 mem_cgroup_uncharge_swapcache(struct page
*page
, swp_entry_t ent
, bool swapout
)
3053 struct mem_cgroup
*memcg
;
3054 int ctype
= MEM_CGROUP_CHARGE_TYPE_SWAPOUT
;
3056 if (!swapout
) /* this was a swap cache but the swap is unused ! */
3057 ctype
= MEM_CGROUP_CHARGE_TYPE_DROP
;
3059 memcg
= __mem_cgroup_uncharge_common(page
, ctype
);
3062 * record memcg information, if swapout && memcg != NULL,
3063 * mem_cgroup_get() was called in uncharge().
3065 if (do_swap_account
&& swapout
&& memcg
)
3066 swap_cgroup_record(ent
, css_id(&memcg
->css
));
3070 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3072 * called from swap_entry_free(). remove record in swap_cgroup and
3073 * uncharge "memsw" account.
3075 void mem_cgroup_uncharge_swap(swp_entry_t ent
)
3077 struct mem_cgroup
*memcg
;
3080 if (!do_swap_account
)
3083 id
= swap_cgroup_record(ent
, 0);
3085 memcg
= mem_cgroup_lookup(id
);
3088 * We uncharge this because swap is freed.
3089 * This memcg can be obsolete one. We avoid calling css_tryget
3091 if (!mem_cgroup_is_root(memcg
))
3092 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
3093 mem_cgroup_swap_statistics(memcg
, false);
3094 mem_cgroup_put(memcg
);
3100 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3101 * @entry: swap entry to be moved
3102 * @from: mem_cgroup which the entry is moved from
3103 * @to: mem_cgroup which the entry is moved to
3104 * @need_fixup: whether we should fixup res_counters and refcounts.
3106 * It succeeds only when the swap_cgroup's record for this entry is the same
3107 * as the mem_cgroup's id of @from.
3109 * Returns 0 on success, -EINVAL on failure.
3111 * The caller must have charged to @to, IOW, called res_counter_charge() about
3112 * both res and memsw, and called css_get().
3114 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
3115 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
3117 unsigned short old_id
, new_id
;
3119 old_id
= css_id(&from
->css
);
3120 new_id
= css_id(&to
->css
);
3122 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
3123 mem_cgroup_swap_statistics(from
, false);
3124 mem_cgroup_swap_statistics(to
, true);
3126 * This function is only called from task migration context now.
3127 * It postpones res_counter and refcount handling till the end
3128 * of task migration(mem_cgroup_clear_mc()) for performance
3129 * improvement. But we cannot postpone mem_cgroup_get(to)
3130 * because if the process that has been moved to @to does
3131 * swap-in, the refcount of @to might be decreased to 0.
3135 if (!mem_cgroup_is_root(from
))
3136 res_counter_uncharge(&from
->memsw
, PAGE_SIZE
);
3137 mem_cgroup_put(from
);
3139 * we charged both to->res and to->memsw, so we should
3142 if (!mem_cgroup_is_root(to
))
3143 res_counter_uncharge(&to
->res
, PAGE_SIZE
);
3150 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
3151 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
3158 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3161 int mem_cgroup_prepare_migration(struct page
*page
,
3162 struct page
*newpage
, struct mem_cgroup
**memcgp
, gfp_t gfp_mask
)
3164 struct mem_cgroup
*memcg
= NULL
;
3165 struct page_cgroup
*pc
;
3166 enum charge_type ctype
;
3171 VM_BUG_ON(PageTransHuge(page
));
3172 if (mem_cgroup_disabled())
3175 pc
= lookup_page_cgroup(page
);
3176 lock_page_cgroup(pc
);
3177 if (PageCgroupUsed(pc
)) {
3178 memcg
= pc
->mem_cgroup
;
3179 css_get(&memcg
->css
);
3181 * At migrating an anonymous page, its mapcount goes down
3182 * to 0 and uncharge() will be called. But, even if it's fully
3183 * unmapped, migration may fail and this page has to be
3184 * charged again. We set MIGRATION flag here and delay uncharge
3185 * until end_migration() is called
3187 * Corner Case Thinking
3189 * When the old page was mapped as Anon and it's unmap-and-freed
3190 * while migration was ongoing.
3191 * If unmap finds the old page, uncharge() of it will be delayed
3192 * until end_migration(). If unmap finds a new page, it's
3193 * uncharged when it make mapcount to be 1->0. If unmap code
3194 * finds swap_migration_entry, the new page will not be mapped
3195 * and end_migration() will find it(mapcount==0).
3198 * When the old page was mapped but migraion fails, the kernel
3199 * remaps it. A charge for it is kept by MIGRATION flag even
3200 * if mapcount goes down to 0. We can do remap successfully
3201 * without charging it again.
3204 * The "old" page is under lock_page() until the end of
3205 * migration, so, the old page itself will not be swapped-out.
3206 * If the new page is swapped out before end_migraton, our
3207 * hook to usual swap-out path will catch the event.
3210 SetPageCgroupMigration(pc
);
3212 unlock_page_cgroup(pc
);
3214 * If the page is not charged at this point,
3221 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, 1, memcgp
, false);
3222 css_put(&memcg
->css
);/* drop extra refcnt */
3224 if (PageAnon(page
)) {
3225 lock_page_cgroup(pc
);
3226 ClearPageCgroupMigration(pc
);
3227 unlock_page_cgroup(pc
);
3229 * The old page may be fully unmapped while we kept it.
3231 mem_cgroup_uncharge_page(page
);
3233 /* we'll need to revisit this error code (we have -EINTR) */
3237 * We charge new page before it's used/mapped. So, even if unlock_page()
3238 * is called before end_migration, we can catch all events on this new
3239 * page. In the case new page is migrated but not remapped, new page's
3240 * mapcount will be finally 0 and we call uncharge in end_migration().
3242 pc
= lookup_page_cgroup(newpage
);
3244 ctype
= MEM_CGROUP_CHARGE_TYPE_MAPPED
;
3245 else if (page_is_file_cache(page
))
3246 ctype
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3248 ctype
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
3249 __mem_cgroup_commit_charge(memcg
, newpage
, 1, pc
, ctype
, false);
3253 /* remove redundant charge if migration failed*/
3254 void mem_cgroup_end_migration(struct mem_cgroup
*memcg
,
3255 struct page
*oldpage
, struct page
*newpage
, bool migration_ok
)
3257 struct page
*used
, *unused
;
3258 struct page_cgroup
*pc
;
3262 /* blocks rmdir() */
3263 cgroup_exclude_rmdir(&memcg
->css
);
3264 if (!migration_ok
) {
3272 * We disallowed uncharge of pages under migration because mapcount
3273 * of the page goes down to zero, temporarly.
3274 * Clear the flag and check the page should be charged.
3276 pc
= lookup_page_cgroup(oldpage
);
3277 lock_page_cgroup(pc
);
3278 ClearPageCgroupMigration(pc
);
3279 unlock_page_cgroup(pc
);
3281 __mem_cgroup_uncharge_common(unused
, MEM_CGROUP_CHARGE_TYPE_FORCE
);
3284 * If a page is a file cache, radix-tree replacement is very atomic
3285 * and we can skip this check. When it was an Anon page, its mapcount
3286 * goes down to 0. But because we added MIGRATION flage, it's not
3287 * uncharged yet. There are several case but page->mapcount check
3288 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3289 * check. (see prepare_charge() also)
3292 mem_cgroup_uncharge_page(used
);
3294 * At migration, we may charge account against cgroup which has no
3296 * So, rmdir()->pre_destroy() can be called while we do this charge.
3297 * In that case, we need to call pre_destroy() again. check it here.
3299 cgroup_release_and_wakeup_rmdir(&memcg
->css
);
3303 * At replace page cache, newpage is not under any memcg but it's on
3304 * LRU. So, this function doesn't touch res_counter but handles LRU
3305 * in correct way. Both pages are locked so we cannot race with uncharge.
3307 void mem_cgroup_replace_page_cache(struct page
*oldpage
,
3308 struct page
*newpage
)
3310 struct mem_cgroup
*memcg
;
3311 struct page_cgroup
*pc
;
3312 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3314 if (mem_cgroup_disabled())
3317 pc
= lookup_page_cgroup(oldpage
);
3318 /* fix accounting on old pages */
3319 lock_page_cgroup(pc
);
3320 memcg
= pc
->mem_cgroup
;
3321 mem_cgroup_charge_statistics(memcg
, PageCgroupCache(pc
), -1);
3322 ClearPageCgroupUsed(pc
);
3323 unlock_page_cgroup(pc
);
3325 if (PageSwapBacked(oldpage
))
3326 type
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
3329 * Even if newpage->mapping was NULL before starting replacement,
3330 * the newpage may be on LRU(or pagevec for LRU) already. We lock
3331 * LRU while we overwrite pc->mem_cgroup.
3333 __mem_cgroup_commit_charge(memcg
, newpage
, 1, pc
, type
, true);
3336 #ifdef CONFIG_DEBUG_VM
3337 static struct page_cgroup
*lookup_page_cgroup_used(struct page
*page
)
3339 struct page_cgroup
*pc
;
3341 pc
= lookup_page_cgroup(page
);
3343 * Can be NULL while feeding pages into the page allocator for
3344 * the first time, i.e. during boot or memory hotplug;
3345 * or when mem_cgroup_disabled().
3347 if (likely(pc
) && PageCgroupUsed(pc
))
3352 bool mem_cgroup_bad_page_check(struct page
*page
)
3354 if (mem_cgroup_disabled())
3357 return lookup_page_cgroup_used(page
) != NULL
;
3360 void mem_cgroup_print_bad_page(struct page
*page
)
3362 struct page_cgroup
*pc
;
3364 pc
= lookup_page_cgroup_used(page
);
3366 printk(KERN_ALERT
"pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
3367 pc
, pc
->flags
, pc
->mem_cgroup
);
3372 static DEFINE_MUTEX(set_limit_mutex
);
3374 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
3375 unsigned long long val
)
3378 u64 memswlimit
, memlimit
;
3380 int children
= mem_cgroup_count_children(memcg
);
3381 u64 curusage
, oldusage
;
3385 * For keeping hierarchical_reclaim simple, how long we should retry
3386 * is depends on callers. We set our retry-count to be function
3387 * of # of children which we should visit in this loop.
3389 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
* children
;
3391 oldusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3394 while (retry_count
) {
3395 if (signal_pending(current
)) {
3400 * Rather than hide all in some function, I do this in
3401 * open coded manner. You see what this really does.
3402 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
3404 mutex_lock(&set_limit_mutex
);
3405 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3406 if (memswlimit
< val
) {
3408 mutex_unlock(&set_limit_mutex
);
3412 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3416 ret
= res_counter_set_limit(&memcg
->res
, val
);
3418 if (memswlimit
== val
)
3419 memcg
->memsw_is_minimum
= true;
3421 memcg
->memsw_is_minimum
= false;
3423 mutex_unlock(&set_limit_mutex
);
3428 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3429 MEM_CGROUP_RECLAIM_SHRINK
);
3430 curusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3431 /* Usage is reduced ? */
3432 if (curusage
>= oldusage
)
3435 oldusage
= curusage
;
3437 if (!ret
&& enlarge
)
3438 memcg_oom_recover(memcg
);
3443 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
3444 unsigned long long val
)
3447 u64 memlimit
, memswlimit
, oldusage
, curusage
;
3448 int children
= mem_cgroup_count_children(memcg
);
3452 /* see mem_cgroup_resize_res_limit */
3453 retry_count
= children
* MEM_CGROUP_RECLAIM_RETRIES
;
3454 oldusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3455 while (retry_count
) {
3456 if (signal_pending(current
)) {
3461 * Rather than hide all in some function, I do this in
3462 * open coded manner. You see what this really does.
3463 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
3465 mutex_lock(&set_limit_mutex
);
3466 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3467 if (memlimit
> val
) {
3469 mutex_unlock(&set_limit_mutex
);
3472 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3473 if (memswlimit
< val
)
3475 ret
= res_counter_set_limit(&memcg
->memsw
, val
);
3477 if (memlimit
== val
)
3478 memcg
->memsw_is_minimum
= true;
3480 memcg
->memsw_is_minimum
= false;
3482 mutex_unlock(&set_limit_mutex
);
3487 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3488 MEM_CGROUP_RECLAIM_NOSWAP
|
3489 MEM_CGROUP_RECLAIM_SHRINK
);
3490 curusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3491 /* Usage is reduced ? */
3492 if (curusage
>= oldusage
)
3495 oldusage
= curusage
;
3497 if (!ret
&& enlarge
)
3498 memcg_oom_recover(memcg
);
3502 unsigned long mem_cgroup_soft_limit_reclaim(struct zone
*zone
, int order
,
3504 unsigned long *total_scanned
)
3506 unsigned long nr_reclaimed
= 0;
3507 struct mem_cgroup_per_zone
*mz
, *next_mz
= NULL
;
3508 unsigned long reclaimed
;
3510 struct mem_cgroup_tree_per_zone
*mctz
;
3511 unsigned long long excess
;
3512 unsigned long nr_scanned
;
3517 mctz
= soft_limit_tree_node_zone(zone_to_nid(zone
), zone_idx(zone
));
3519 * This loop can run a while, specially if mem_cgroup's continuously
3520 * keep exceeding their soft limit and putting the system under
3527 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
3532 reclaimed
= mem_cgroup_soft_reclaim(mz
->mem
, zone
,
3533 gfp_mask
, &nr_scanned
);
3534 nr_reclaimed
+= reclaimed
;
3535 *total_scanned
+= nr_scanned
;
3536 spin_lock(&mctz
->lock
);
3539 * If we failed to reclaim anything from this memory cgroup
3540 * it is time to move on to the next cgroup
3546 * Loop until we find yet another one.
3548 * By the time we get the soft_limit lock
3549 * again, someone might have aded the
3550 * group back on the RB tree. Iterate to
3551 * make sure we get a different mem.
3552 * mem_cgroup_largest_soft_limit_node returns
3553 * NULL if no other cgroup is present on
3557 __mem_cgroup_largest_soft_limit_node(mctz
);
3559 css_put(&next_mz
->mem
->css
);
3560 else /* next_mz == NULL or other memcg */
3564 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
3565 excess
= res_counter_soft_limit_excess(&mz
->mem
->res
);
3567 * One school of thought says that we should not add
3568 * back the node to the tree if reclaim returns 0.
3569 * But our reclaim could return 0, simply because due
3570 * to priority we are exposing a smaller subset of
3571 * memory to reclaim from. Consider this as a longer
3574 /* If excess == 0, no tree ops */
3575 __mem_cgroup_insert_exceeded(mz
->mem
, mz
, mctz
, excess
);
3576 spin_unlock(&mctz
->lock
);
3577 css_put(&mz
->mem
->css
);
3580 * Could not reclaim anything and there are no more
3581 * mem cgroups to try or we seem to be looping without
3582 * reclaiming anything.
3584 if (!nr_reclaimed
&&
3586 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
3588 } while (!nr_reclaimed
);
3590 css_put(&next_mz
->mem
->css
);
3591 return nr_reclaimed
;
3595 * This routine traverse page_cgroup in given list and drop them all.
3596 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3598 static int mem_cgroup_force_empty_list(struct mem_cgroup
*memcg
,
3599 int node
, int zid
, enum lru_list lru
)
3601 struct mem_cgroup_per_zone
*mz
;
3602 unsigned long flags
, loop
;
3603 struct list_head
*list
;
3608 zone
= &NODE_DATA(node
)->node_zones
[zid
];
3609 mz
= mem_cgroup_zoneinfo(memcg
, node
, zid
);
3610 list
= &mz
->lruvec
.lists
[lru
];
3612 loop
= MEM_CGROUP_ZSTAT(mz
, lru
);
3613 /* give some margin against EBUSY etc...*/
3617 struct page_cgroup
*pc
;
3621 spin_lock_irqsave(&zone
->lru_lock
, flags
);
3622 if (list_empty(list
)) {
3623 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3626 page
= list_entry(list
->prev
, struct page
, lru
);
3628 list_move(&page
->lru
, list
);
3630 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3633 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3635 pc
= lookup_page_cgroup(page
);
3637 ret
= mem_cgroup_move_parent(page
, pc
, memcg
, GFP_KERNEL
);
3638 if (ret
== -ENOMEM
|| ret
== -EINTR
)
3641 if (ret
== -EBUSY
|| ret
== -EINVAL
) {
3642 /* found lock contention or "pc" is obsolete. */
3649 if (!ret
&& !list_empty(list
))
3655 * make mem_cgroup's charge to be 0 if there is no task.
3656 * This enables deleting this mem_cgroup.
3658 static int mem_cgroup_force_empty(struct mem_cgroup
*memcg
, bool free_all
)
3661 int node
, zid
, shrink
;
3662 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
3663 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
3665 css_get(&memcg
->css
);
3668 /* should free all ? */
3674 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
))
3677 if (signal_pending(current
))
3679 /* This is for making all *used* pages to be on LRU. */
3680 lru_add_drain_all();
3681 drain_all_stock_sync(memcg
);
3683 mem_cgroup_start_move(memcg
);
3684 for_each_node_state(node
, N_HIGH_MEMORY
) {
3685 for (zid
= 0; !ret
&& zid
< MAX_NR_ZONES
; zid
++) {
3688 ret
= mem_cgroup_force_empty_list(memcg
,
3697 mem_cgroup_end_move(memcg
);
3698 memcg_oom_recover(memcg
);
3699 /* it seems parent cgroup doesn't have enough mem */
3703 /* "ret" should also be checked to ensure all lists are empty. */
3704 } while (memcg
->res
.usage
> 0 || ret
);
3706 css_put(&memcg
->css
);
3710 /* returns EBUSY if there is a task or if we come here twice. */
3711 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
) || shrink
) {
3715 /* we call try-to-free pages for make this cgroup empty */
3716 lru_add_drain_all();
3717 /* try to free all pages in this cgroup */
3719 while (nr_retries
&& memcg
->res
.usage
> 0) {
3722 if (signal_pending(current
)) {
3726 progress
= try_to_free_mem_cgroup_pages(memcg
, GFP_KERNEL
,
3730 /* maybe some writeback is necessary */
3731 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3736 /* try move_account...there may be some *locked* pages. */
3740 int mem_cgroup_force_empty_write(struct cgroup
*cont
, unsigned int event
)
3742 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont
), true);
3746 static u64
mem_cgroup_hierarchy_read(struct cgroup
*cont
, struct cftype
*cft
)
3748 return mem_cgroup_from_cont(cont
)->use_hierarchy
;
3751 static int mem_cgroup_hierarchy_write(struct cgroup
*cont
, struct cftype
*cft
,
3755 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3756 struct cgroup
*parent
= cont
->parent
;
3757 struct mem_cgroup
*parent_memcg
= NULL
;
3760 parent_memcg
= mem_cgroup_from_cont(parent
);
3764 * If parent's use_hierarchy is set, we can't make any modifications
3765 * in the child subtrees. If it is unset, then the change can
3766 * occur, provided the current cgroup has no children.
3768 * For the root cgroup, parent_mem is NULL, we allow value to be
3769 * set if there are no children.
3771 if ((!parent_memcg
|| !parent_memcg
->use_hierarchy
) &&
3772 (val
== 1 || val
== 0)) {
3773 if (list_empty(&cont
->children
))
3774 memcg
->use_hierarchy
= val
;
3785 static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup
*memcg
,
3786 enum mem_cgroup_stat_index idx
)
3788 struct mem_cgroup
*iter
;
3791 /* Per-cpu values can be negative, use a signed accumulator */
3792 for_each_mem_cgroup_tree(iter
, memcg
)
3793 val
+= mem_cgroup_read_stat(iter
, idx
);
3795 if (val
< 0) /* race ? */
3800 static inline u64
mem_cgroup_usage(struct mem_cgroup
*memcg
, bool swap
)
3804 if (!mem_cgroup_is_root(memcg
)) {
3806 return res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3808 return res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3811 val
= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_CACHE
);
3812 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_RSS
);
3815 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_SWAPOUT
);
3817 return val
<< PAGE_SHIFT
;
3820 static u64
mem_cgroup_read(struct cgroup
*cont
, struct cftype
*cft
)
3822 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3826 type
= MEMFILE_TYPE(cft
->private);
3827 name
= MEMFILE_ATTR(cft
->private);
3830 if (name
== RES_USAGE
)
3831 val
= mem_cgroup_usage(memcg
, false);
3833 val
= res_counter_read_u64(&memcg
->res
, name
);
3836 if (name
== RES_USAGE
)
3837 val
= mem_cgroup_usage(memcg
, true);
3839 val
= res_counter_read_u64(&memcg
->memsw
, name
);
3848 * The user of this function is...
3851 static int mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
3854 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3856 unsigned long long val
;
3859 type
= MEMFILE_TYPE(cft
->private);
3860 name
= MEMFILE_ATTR(cft
->private);
3863 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
3867 /* This function does all necessary parse...reuse it */
3868 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3872 ret
= mem_cgroup_resize_limit(memcg
, val
);
3874 ret
= mem_cgroup_resize_memsw_limit(memcg
, val
);
3876 case RES_SOFT_LIMIT
:
3877 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3881 * For memsw, soft limits are hard to implement in terms
3882 * of semantics, for now, we support soft limits for
3883 * control without swap
3886 ret
= res_counter_set_soft_limit(&memcg
->res
, val
);
3891 ret
= -EINVAL
; /* should be BUG() ? */
3897 static void memcg_get_hierarchical_limit(struct mem_cgroup
*memcg
,
3898 unsigned long long *mem_limit
, unsigned long long *memsw_limit
)
3900 struct cgroup
*cgroup
;
3901 unsigned long long min_limit
, min_memsw_limit
, tmp
;
3903 min_limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3904 min_memsw_limit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3905 cgroup
= memcg
->css
.cgroup
;
3906 if (!memcg
->use_hierarchy
)
3909 while (cgroup
->parent
) {
3910 cgroup
= cgroup
->parent
;
3911 memcg
= mem_cgroup_from_cont(cgroup
);
3912 if (!memcg
->use_hierarchy
)
3914 tmp
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3915 min_limit
= min(min_limit
, tmp
);
3916 tmp
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3917 min_memsw_limit
= min(min_memsw_limit
, tmp
);
3920 *mem_limit
= min_limit
;
3921 *memsw_limit
= min_memsw_limit
;
3925 static int mem_cgroup_reset(struct cgroup
*cont
, unsigned int event
)
3927 struct mem_cgroup
*memcg
;
3930 memcg
= mem_cgroup_from_cont(cont
);
3931 type
= MEMFILE_TYPE(event
);
3932 name
= MEMFILE_ATTR(event
);
3936 res_counter_reset_max(&memcg
->res
);
3938 res_counter_reset_max(&memcg
->memsw
);
3942 res_counter_reset_failcnt(&memcg
->res
);
3944 res_counter_reset_failcnt(&memcg
->memsw
);
3951 static u64
mem_cgroup_move_charge_read(struct cgroup
*cgrp
,
3954 return mem_cgroup_from_cont(cgrp
)->move_charge_at_immigrate
;
3958 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
3959 struct cftype
*cft
, u64 val
)
3961 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3963 if (val
>= (1 << NR_MOVE_TYPE
))
3966 * We check this value several times in both in can_attach() and
3967 * attach(), so we need cgroup lock to prevent this value from being
3971 memcg
->move_charge_at_immigrate
= val
;
3977 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
3978 struct cftype
*cft
, u64 val
)
3985 /* For read statistics */
4003 struct mcs_total_stat
{
4004 s64 stat
[NR_MCS_STAT
];
4010 } memcg_stat_strings
[NR_MCS_STAT
] = {
4011 {"cache", "total_cache"},
4012 {"rss", "total_rss"},
4013 {"mapped_file", "total_mapped_file"},
4014 {"pgpgin", "total_pgpgin"},
4015 {"pgpgout", "total_pgpgout"},
4016 {"swap", "total_swap"},
4017 {"pgfault", "total_pgfault"},
4018 {"pgmajfault", "total_pgmajfault"},
4019 {"inactive_anon", "total_inactive_anon"},
4020 {"active_anon", "total_active_anon"},
4021 {"inactive_file", "total_inactive_file"},
4022 {"active_file", "total_active_file"},
4023 {"unevictable", "total_unevictable"}
4028 mem_cgroup_get_local_stat(struct mem_cgroup
*memcg
, struct mcs_total_stat
*s
)
4033 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_CACHE
);
4034 s
->stat
[MCS_CACHE
] += val
* PAGE_SIZE
;
4035 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_RSS
);
4036 s
->stat
[MCS_RSS
] += val
* PAGE_SIZE
;
4037 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_FILE_MAPPED
);
4038 s
->stat
[MCS_FILE_MAPPED
] += val
* PAGE_SIZE
;
4039 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGPGIN
);
4040 s
->stat
[MCS_PGPGIN
] += val
;
4041 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGPGOUT
);
4042 s
->stat
[MCS_PGPGOUT
] += val
;
4043 if (do_swap_account
) {
4044 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_SWAPOUT
);
4045 s
->stat
[MCS_SWAP
] += val
* PAGE_SIZE
;
4047 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGFAULT
);
4048 s
->stat
[MCS_PGFAULT
] += val
;
4049 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGMAJFAULT
);
4050 s
->stat
[MCS_PGMAJFAULT
] += val
;
4053 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_INACTIVE_ANON
));
4054 s
->stat
[MCS_INACTIVE_ANON
] += val
* PAGE_SIZE
;
4055 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_ACTIVE_ANON
));
4056 s
->stat
[MCS_ACTIVE_ANON
] += val
* PAGE_SIZE
;
4057 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_INACTIVE_FILE
));
4058 s
->stat
[MCS_INACTIVE_FILE
] += val
* PAGE_SIZE
;
4059 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_ACTIVE_FILE
));
4060 s
->stat
[MCS_ACTIVE_FILE
] += val
* PAGE_SIZE
;
4061 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_UNEVICTABLE
));
4062 s
->stat
[MCS_UNEVICTABLE
] += val
* PAGE_SIZE
;
4066 mem_cgroup_get_total_stat(struct mem_cgroup
*memcg
, struct mcs_total_stat
*s
)
4068 struct mem_cgroup
*iter
;
4070 for_each_mem_cgroup_tree(iter
, memcg
)
4071 mem_cgroup_get_local_stat(iter
, s
);
4075 static int mem_control_numa_stat_show(struct seq_file
*m
, void *arg
)
4078 unsigned long total_nr
, file_nr
, anon_nr
, unevictable_nr
;
4079 unsigned long node_nr
;
4080 struct cgroup
*cont
= m
->private;
4081 struct mem_cgroup
*mem_cont
= mem_cgroup_from_cont(cont
);
4083 total_nr
= mem_cgroup_nr_lru_pages(mem_cont
, LRU_ALL
);
4084 seq_printf(m
, "total=%lu", total_nr
);
4085 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4086 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
, LRU_ALL
);
4087 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4091 file_nr
= mem_cgroup_nr_lru_pages(mem_cont
, LRU_ALL_FILE
);
4092 seq_printf(m
, "file=%lu", file_nr
);
4093 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4094 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
,
4096 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4100 anon_nr
= mem_cgroup_nr_lru_pages(mem_cont
, LRU_ALL_ANON
);
4101 seq_printf(m
, "anon=%lu", anon_nr
);
4102 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4103 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
,
4105 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4109 unevictable_nr
= mem_cgroup_nr_lru_pages(mem_cont
, BIT(LRU_UNEVICTABLE
));
4110 seq_printf(m
, "unevictable=%lu", unevictable_nr
);
4111 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4112 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
,
4113 BIT(LRU_UNEVICTABLE
));
4114 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4119 #endif /* CONFIG_NUMA */
4121 static int mem_control_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
4122 struct cgroup_map_cb
*cb
)
4124 struct mem_cgroup
*mem_cont
= mem_cgroup_from_cont(cont
);
4125 struct mcs_total_stat mystat
;
4128 memset(&mystat
, 0, sizeof(mystat
));
4129 mem_cgroup_get_local_stat(mem_cont
, &mystat
);
4132 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
4133 if (i
== MCS_SWAP
&& !do_swap_account
)
4135 cb
->fill(cb
, memcg_stat_strings
[i
].local_name
, mystat
.stat
[i
]);
4138 /* Hierarchical information */
4140 unsigned long long limit
, memsw_limit
;
4141 memcg_get_hierarchical_limit(mem_cont
, &limit
, &memsw_limit
);
4142 cb
->fill(cb
, "hierarchical_memory_limit", limit
);
4143 if (do_swap_account
)
4144 cb
->fill(cb
, "hierarchical_memsw_limit", memsw_limit
);
4147 memset(&mystat
, 0, sizeof(mystat
));
4148 mem_cgroup_get_total_stat(mem_cont
, &mystat
);
4149 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
4150 if (i
== MCS_SWAP
&& !do_swap_account
)
4152 cb
->fill(cb
, memcg_stat_strings
[i
].total_name
, mystat
.stat
[i
]);
4155 #ifdef CONFIG_DEBUG_VM
4158 struct mem_cgroup_per_zone
*mz
;
4159 unsigned long recent_rotated
[2] = {0, 0};
4160 unsigned long recent_scanned
[2] = {0, 0};
4162 for_each_online_node(nid
)
4163 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
4164 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
4166 recent_rotated
[0] +=
4167 mz
->reclaim_stat
.recent_rotated
[0];
4168 recent_rotated
[1] +=
4169 mz
->reclaim_stat
.recent_rotated
[1];
4170 recent_scanned
[0] +=
4171 mz
->reclaim_stat
.recent_scanned
[0];
4172 recent_scanned
[1] +=
4173 mz
->reclaim_stat
.recent_scanned
[1];
4175 cb
->fill(cb
, "recent_rotated_anon", recent_rotated
[0]);
4176 cb
->fill(cb
, "recent_rotated_file", recent_rotated
[1]);
4177 cb
->fill(cb
, "recent_scanned_anon", recent_scanned
[0]);
4178 cb
->fill(cb
, "recent_scanned_file", recent_scanned
[1]);
4185 static u64
mem_cgroup_swappiness_read(struct cgroup
*cgrp
, struct cftype
*cft
)
4187 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4189 return mem_cgroup_swappiness(memcg
);
4192 static int mem_cgroup_swappiness_write(struct cgroup
*cgrp
, struct cftype
*cft
,
4195 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4196 struct mem_cgroup
*parent
;
4201 if (cgrp
->parent
== NULL
)
4204 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4208 /* If under hierarchy, only empty-root can set this value */
4209 if ((parent
->use_hierarchy
) ||
4210 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4215 memcg
->swappiness
= val
;
4222 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
4224 struct mem_cgroup_threshold_ary
*t
;
4230 t
= rcu_dereference(memcg
->thresholds
.primary
);
4232 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
4237 usage
= mem_cgroup_usage(memcg
, swap
);
4240 * current_threshold points to threshold just below usage.
4241 * If it's not true, a threshold was crossed after last
4242 * call of __mem_cgroup_threshold().
4244 i
= t
->current_threshold
;
4247 * Iterate backward over array of thresholds starting from
4248 * current_threshold and check if a threshold is crossed.
4249 * If none of thresholds below usage is crossed, we read
4250 * only one element of the array here.
4252 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
4253 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4255 /* i = current_threshold + 1 */
4259 * Iterate forward over array of thresholds starting from
4260 * current_threshold+1 and check if a threshold is crossed.
4261 * If none of thresholds above usage is crossed, we read
4262 * only one element of the array here.
4264 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
4265 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4267 /* Update current_threshold */
4268 t
->current_threshold
= i
- 1;
4273 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
4276 __mem_cgroup_threshold(memcg
, false);
4277 if (do_swap_account
)
4278 __mem_cgroup_threshold(memcg
, true);
4280 memcg
= parent_mem_cgroup(memcg
);
4284 static int compare_thresholds(const void *a
, const void *b
)
4286 const struct mem_cgroup_threshold
*_a
= a
;
4287 const struct mem_cgroup_threshold
*_b
= b
;
4289 return _a
->threshold
- _b
->threshold
;
4292 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*memcg
)
4294 struct mem_cgroup_eventfd_list
*ev
;
4296 list_for_each_entry(ev
, &memcg
->oom_notify
, list
)
4297 eventfd_signal(ev
->eventfd
, 1);
4301 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
)
4303 struct mem_cgroup
*iter
;
4305 for_each_mem_cgroup_tree(iter
, memcg
)
4306 mem_cgroup_oom_notify_cb(iter
);
4309 static int mem_cgroup_usage_register_event(struct cgroup
*cgrp
,
4310 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4312 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4313 struct mem_cgroup_thresholds
*thresholds
;
4314 struct mem_cgroup_threshold_ary
*new;
4315 int type
= MEMFILE_TYPE(cft
->private);
4316 u64 threshold
, usage
;
4319 ret
= res_counter_memparse_write_strategy(args
, &threshold
);
4323 mutex_lock(&memcg
->thresholds_lock
);
4326 thresholds
= &memcg
->thresholds
;
4327 else if (type
== _MEMSWAP
)
4328 thresholds
= &memcg
->memsw_thresholds
;
4332 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4334 /* Check if a threshold crossed before adding a new one */
4335 if (thresholds
->primary
)
4336 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4338 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
4340 /* Allocate memory for new array of thresholds */
4341 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
4349 /* Copy thresholds (if any) to new array */
4350 if (thresholds
->primary
) {
4351 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
4352 sizeof(struct mem_cgroup_threshold
));
4355 /* Add new threshold */
4356 new->entries
[size
- 1].eventfd
= eventfd
;
4357 new->entries
[size
- 1].threshold
= threshold
;
4359 /* Sort thresholds. Registering of new threshold isn't time-critical */
4360 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
4361 compare_thresholds
, NULL
);
4363 /* Find current threshold */
4364 new->current_threshold
= -1;
4365 for (i
= 0; i
< size
; i
++) {
4366 if (new->entries
[i
].threshold
< usage
) {
4368 * new->current_threshold will not be used until
4369 * rcu_assign_pointer(), so it's safe to increment
4372 ++new->current_threshold
;
4376 /* Free old spare buffer and save old primary buffer as spare */
4377 kfree(thresholds
->spare
);
4378 thresholds
->spare
= thresholds
->primary
;
4380 rcu_assign_pointer(thresholds
->primary
, new);
4382 /* To be sure that nobody uses thresholds */
4386 mutex_unlock(&memcg
->thresholds_lock
);
4391 static void mem_cgroup_usage_unregister_event(struct cgroup
*cgrp
,
4392 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4394 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4395 struct mem_cgroup_thresholds
*thresholds
;
4396 struct mem_cgroup_threshold_ary
*new;
4397 int type
= MEMFILE_TYPE(cft
->private);
4401 mutex_lock(&memcg
->thresholds_lock
);
4403 thresholds
= &memcg
->thresholds
;
4404 else if (type
== _MEMSWAP
)
4405 thresholds
= &memcg
->memsw_thresholds
;
4410 * Something went wrong if we trying to unregister a threshold
4411 * if we don't have thresholds
4413 BUG_ON(!thresholds
);
4415 if (!thresholds
->primary
)
4418 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4420 /* Check if a threshold crossed before removing */
4421 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4423 /* Calculate new number of threshold */
4425 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
4426 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
4430 new = thresholds
->spare
;
4432 /* Set thresholds array to NULL if we don't have thresholds */
4441 /* Copy thresholds and find current threshold */
4442 new->current_threshold
= -1;
4443 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
4444 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
4447 new->entries
[j
] = thresholds
->primary
->entries
[i
];
4448 if (new->entries
[j
].threshold
< usage
) {
4450 * new->current_threshold will not be used
4451 * until rcu_assign_pointer(), so it's safe to increment
4454 ++new->current_threshold
;
4460 /* Swap primary and spare array */
4461 thresholds
->spare
= thresholds
->primary
;
4462 rcu_assign_pointer(thresholds
->primary
, new);
4464 /* To be sure that nobody uses thresholds */
4467 mutex_unlock(&memcg
->thresholds_lock
);
4470 static int mem_cgroup_oom_register_event(struct cgroup
*cgrp
,
4471 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4473 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4474 struct mem_cgroup_eventfd_list
*event
;
4475 int type
= MEMFILE_TYPE(cft
->private);
4477 BUG_ON(type
!= _OOM_TYPE
);
4478 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
4482 spin_lock(&memcg_oom_lock
);
4484 event
->eventfd
= eventfd
;
4485 list_add(&event
->list
, &memcg
->oom_notify
);
4487 /* already in OOM ? */
4488 if (atomic_read(&memcg
->under_oom
))
4489 eventfd_signal(eventfd
, 1);
4490 spin_unlock(&memcg_oom_lock
);
4495 static void mem_cgroup_oom_unregister_event(struct cgroup
*cgrp
,
4496 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4498 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4499 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
4500 int type
= MEMFILE_TYPE(cft
->private);
4502 BUG_ON(type
!= _OOM_TYPE
);
4504 spin_lock(&memcg_oom_lock
);
4506 list_for_each_entry_safe(ev
, tmp
, &memcg
->oom_notify
, list
) {
4507 if (ev
->eventfd
== eventfd
) {
4508 list_del(&ev
->list
);
4513 spin_unlock(&memcg_oom_lock
);
4516 static int mem_cgroup_oom_control_read(struct cgroup
*cgrp
,
4517 struct cftype
*cft
, struct cgroup_map_cb
*cb
)
4519 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4521 cb
->fill(cb
, "oom_kill_disable", memcg
->oom_kill_disable
);
4523 if (atomic_read(&memcg
->under_oom
))
4524 cb
->fill(cb
, "under_oom", 1);
4526 cb
->fill(cb
, "under_oom", 0);
4530 static int mem_cgroup_oom_control_write(struct cgroup
*cgrp
,
4531 struct cftype
*cft
, u64 val
)
4533 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4534 struct mem_cgroup
*parent
;
4536 /* cannot set to root cgroup and only 0 and 1 are allowed */
4537 if (!cgrp
->parent
|| !((val
== 0) || (val
== 1)))
4540 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4543 /* oom-kill-disable is a flag for subhierarchy. */
4544 if ((parent
->use_hierarchy
) ||
4545 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4549 memcg
->oom_kill_disable
= val
;
4551 memcg_oom_recover(memcg
);
4557 static const struct file_operations mem_control_numa_stat_file_operations
= {
4559 .llseek
= seq_lseek
,
4560 .release
= single_release
,
4563 static int mem_control_numa_stat_open(struct inode
*unused
, struct file
*file
)
4565 struct cgroup
*cont
= file
->f_dentry
->d_parent
->d_fsdata
;
4567 file
->f_op
= &mem_control_numa_stat_file_operations
;
4568 return single_open(file
, mem_control_numa_stat_show
, cont
);
4570 #endif /* CONFIG_NUMA */
4572 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
4573 static int register_kmem_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4576 * Part of this would be better living in a separate allocation
4577 * function, leaving us with just the cgroup tree population work.
4578 * We, however, depend on state such as network's proto_list that
4579 * is only initialized after cgroup creation. I found the less
4580 * cumbersome way to deal with it to defer it all to populate time
4582 return mem_cgroup_sockets_init(cont
, ss
);
4585 static void kmem_cgroup_destroy(struct cgroup_subsys
*ss
,
4586 struct cgroup
*cont
)
4588 mem_cgroup_sockets_destroy(cont
, ss
);
4591 static int register_kmem_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4596 static void kmem_cgroup_destroy(struct cgroup_subsys
*ss
,
4597 struct cgroup
*cont
)
4602 static struct cftype mem_cgroup_files
[] = {
4604 .name
= "usage_in_bytes",
4605 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
4606 .read_u64
= mem_cgroup_read
,
4607 .register_event
= mem_cgroup_usage_register_event
,
4608 .unregister_event
= mem_cgroup_usage_unregister_event
,
4611 .name
= "max_usage_in_bytes",
4612 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
4613 .trigger
= mem_cgroup_reset
,
4614 .read_u64
= mem_cgroup_read
,
4617 .name
= "limit_in_bytes",
4618 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
4619 .write_string
= mem_cgroup_write
,
4620 .read_u64
= mem_cgroup_read
,
4623 .name
= "soft_limit_in_bytes",
4624 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
4625 .write_string
= mem_cgroup_write
,
4626 .read_u64
= mem_cgroup_read
,
4630 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
4631 .trigger
= mem_cgroup_reset
,
4632 .read_u64
= mem_cgroup_read
,
4636 .read_map
= mem_control_stat_show
,
4639 .name
= "force_empty",
4640 .trigger
= mem_cgroup_force_empty_write
,
4643 .name
= "use_hierarchy",
4644 .write_u64
= mem_cgroup_hierarchy_write
,
4645 .read_u64
= mem_cgroup_hierarchy_read
,
4648 .name
= "swappiness",
4649 .read_u64
= mem_cgroup_swappiness_read
,
4650 .write_u64
= mem_cgroup_swappiness_write
,
4653 .name
= "move_charge_at_immigrate",
4654 .read_u64
= mem_cgroup_move_charge_read
,
4655 .write_u64
= mem_cgroup_move_charge_write
,
4658 .name
= "oom_control",
4659 .read_map
= mem_cgroup_oom_control_read
,
4660 .write_u64
= mem_cgroup_oom_control_write
,
4661 .register_event
= mem_cgroup_oom_register_event
,
4662 .unregister_event
= mem_cgroup_oom_unregister_event
,
4663 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
4667 .name
= "numa_stat",
4668 .open
= mem_control_numa_stat_open
,
4674 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4675 static struct cftype memsw_cgroup_files
[] = {
4677 .name
= "memsw.usage_in_bytes",
4678 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
4679 .read_u64
= mem_cgroup_read
,
4680 .register_event
= mem_cgroup_usage_register_event
,
4681 .unregister_event
= mem_cgroup_usage_unregister_event
,
4684 .name
= "memsw.max_usage_in_bytes",
4685 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
4686 .trigger
= mem_cgroup_reset
,
4687 .read_u64
= mem_cgroup_read
,
4690 .name
= "memsw.limit_in_bytes",
4691 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
4692 .write_string
= mem_cgroup_write
,
4693 .read_u64
= mem_cgroup_read
,
4696 .name
= "memsw.failcnt",
4697 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
4698 .trigger
= mem_cgroup_reset
,
4699 .read_u64
= mem_cgroup_read
,
4703 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4705 if (!do_swap_account
)
4707 return cgroup_add_files(cont
, ss
, memsw_cgroup_files
,
4708 ARRAY_SIZE(memsw_cgroup_files
));
4711 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4717 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4719 struct mem_cgroup_per_node
*pn
;
4720 struct mem_cgroup_per_zone
*mz
;
4722 int zone
, tmp
= node
;
4724 * This routine is called against possible nodes.
4725 * But it's BUG to call kmalloc() against offline node.
4727 * TODO: this routine can waste much memory for nodes which will
4728 * never be onlined. It's better to use memory hotplug callback
4731 if (!node_state(node
, N_NORMAL_MEMORY
))
4733 pn
= kzalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
4737 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4738 mz
= &pn
->zoneinfo
[zone
];
4740 INIT_LIST_HEAD(&mz
->lruvec
.lists
[l
]);
4741 mz
->usage_in_excess
= 0;
4742 mz
->on_tree
= false;
4745 memcg
->info
.nodeinfo
[node
] = pn
;
4749 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4751 kfree(memcg
->info
.nodeinfo
[node
]);
4754 static struct mem_cgroup
*mem_cgroup_alloc(void)
4756 struct mem_cgroup
*mem
;
4757 int size
= sizeof(struct mem_cgroup
);
4759 /* Can be very big if MAX_NUMNODES is very big */
4760 if (size
< PAGE_SIZE
)
4761 mem
= kzalloc(size
, GFP_KERNEL
);
4763 mem
= vzalloc(size
);
4768 mem
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4771 spin_lock_init(&mem
->pcp_counter_lock
);
4775 if (size
< PAGE_SIZE
)
4783 * At destroying mem_cgroup, references from swap_cgroup can remain.
4784 * (scanning all at force_empty is too costly...)
4786 * Instead of clearing all references at force_empty, we remember
4787 * the number of reference from swap_cgroup and free mem_cgroup when
4788 * it goes down to 0.
4790 * Removal of cgroup itself succeeds regardless of refs from swap.
4793 static void __mem_cgroup_free(struct mem_cgroup
*memcg
)
4797 mem_cgroup_remove_from_trees(memcg
);
4798 free_css_id(&mem_cgroup_subsys
, &memcg
->css
);
4801 free_mem_cgroup_per_zone_info(memcg
, node
);
4803 free_percpu(memcg
->stat
);
4804 if (sizeof(struct mem_cgroup
) < PAGE_SIZE
)
4810 static void mem_cgroup_get(struct mem_cgroup
*memcg
)
4812 atomic_inc(&memcg
->refcnt
);
4815 static void __mem_cgroup_put(struct mem_cgroup
*memcg
, int count
)
4817 if (atomic_sub_and_test(count
, &memcg
->refcnt
)) {
4818 struct mem_cgroup
*parent
= parent_mem_cgroup(memcg
);
4819 __mem_cgroup_free(memcg
);
4821 mem_cgroup_put(parent
);
4825 static void mem_cgroup_put(struct mem_cgroup
*memcg
)
4827 __mem_cgroup_put(memcg
, 1);
4831 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4833 struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*memcg
)
4835 if (!memcg
->res
.parent
)
4837 return mem_cgroup_from_res_counter(memcg
->res
.parent
, res
);
4839 EXPORT_SYMBOL(parent_mem_cgroup
);
4841 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4842 static void __init
enable_swap_cgroup(void)
4844 if (!mem_cgroup_disabled() && really_do_swap_account
)
4845 do_swap_account
= 1;
4848 static void __init
enable_swap_cgroup(void)
4853 static int mem_cgroup_soft_limit_tree_init(void)
4855 struct mem_cgroup_tree_per_node
*rtpn
;
4856 struct mem_cgroup_tree_per_zone
*rtpz
;
4857 int tmp
, node
, zone
;
4859 for_each_node(node
) {
4861 if (!node_state(node
, N_NORMAL_MEMORY
))
4863 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
, tmp
);
4867 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
4869 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4870 rtpz
= &rtpn
->rb_tree_per_zone
[zone
];
4871 rtpz
->rb_root
= RB_ROOT
;
4872 spin_lock_init(&rtpz
->lock
);
4878 for_each_node(node
) {
4879 if (!soft_limit_tree
.rb_tree_per_node
[node
])
4881 kfree(soft_limit_tree
.rb_tree_per_node
[node
]);
4882 soft_limit_tree
.rb_tree_per_node
[node
] = NULL
;
4888 static struct cgroup_subsys_state
* __ref
4889 mem_cgroup_create(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
4891 struct mem_cgroup
*memcg
, *parent
;
4892 long error
= -ENOMEM
;
4895 memcg
= mem_cgroup_alloc();
4897 return ERR_PTR(error
);
4900 if (alloc_mem_cgroup_per_zone_info(memcg
, node
))
4904 if (cont
->parent
== NULL
) {
4906 enable_swap_cgroup();
4908 if (mem_cgroup_soft_limit_tree_init())
4910 root_mem_cgroup
= memcg
;
4911 for_each_possible_cpu(cpu
) {
4912 struct memcg_stock_pcp
*stock
=
4913 &per_cpu(memcg_stock
, cpu
);
4914 INIT_WORK(&stock
->work
, drain_local_stock
);
4916 hotcpu_notifier(memcg_cpu_hotplug_callback
, 0);
4918 parent
= mem_cgroup_from_cont(cont
->parent
);
4919 memcg
->use_hierarchy
= parent
->use_hierarchy
;
4920 memcg
->oom_kill_disable
= parent
->oom_kill_disable
;
4923 if (parent
&& parent
->use_hierarchy
) {
4924 res_counter_init(&memcg
->res
, &parent
->res
);
4925 res_counter_init(&memcg
->memsw
, &parent
->memsw
);
4927 * We increment refcnt of the parent to ensure that we can
4928 * safely access it on res_counter_charge/uncharge.
4929 * This refcnt will be decremented when freeing this
4930 * mem_cgroup(see mem_cgroup_put).
4932 mem_cgroup_get(parent
);
4934 res_counter_init(&memcg
->res
, NULL
);
4935 res_counter_init(&memcg
->memsw
, NULL
);
4937 memcg
->last_scanned_node
= MAX_NUMNODES
;
4938 INIT_LIST_HEAD(&memcg
->oom_notify
);
4941 memcg
->swappiness
= mem_cgroup_swappiness(parent
);
4942 atomic_set(&memcg
->refcnt
, 1);
4943 memcg
->move_charge_at_immigrate
= 0;
4944 mutex_init(&memcg
->thresholds_lock
);
4947 __mem_cgroup_free(memcg
);
4948 return ERR_PTR(error
);
4951 static int mem_cgroup_pre_destroy(struct cgroup_subsys
*ss
,
4952 struct cgroup
*cont
)
4954 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4956 return mem_cgroup_force_empty(memcg
, false);
4959 static void mem_cgroup_destroy(struct cgroup_subsys
*ss
,
4960 struct cgroup
*cont
)
4962 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4964 kmem_cgroup_destroy(ss
, cont
);
4966 mem_cgroup_put(memcg
);
4969 static int mem_cgroup_populate(struct cgroup_subsys
*ss
,
4970 struct cgroup
*cont
)
4974 ret
= cgroup_add_files(cont
, ss
, mem_cgroup_files
,
4975 ARRAY_SIZE(mem_cgroup_files
));
4978 ret
= register_memsw_files(cont
, ss
);
4981 ret
= register_kmem_files(cont
, ss
);
4987 /* Handlers for move charge at task migration. */
4988 #define PRECHARGE_COUNT_AT_ONCE 256
4989 static int mem_cgroup_do_precharge(unsigned long count
)
4992 int batch_count
= PRECHARGE_COUNT_AT_ONCE
;
4993 struct mem_cgroup
*memcg
= mc
.to
;
4995 if (mem_cgroup_is_root(memcg
)) {
4996 mc
.precharge
+= count
;
4997 /* we don't need css_get for root */
5000 /* try to charge at once */
5002 struct res_counter
*dummy
;
5004 * "memcg" cannot be under rmdir() because we've already checked
5005 * by cgroup_lock_live_cgroup() that it is not removed and we
5006 * are still under the same cgroup_mutex. So we can postpone
5009 if (res_counter_charge(&memcg
->res
, PAGE_SIZE
* count
, &dummy
))
5011 if (do_swap_account
&& res_counter_charge(&memcg
->memsw
,
5012 PAGE_SIZE
* count
, &dummy
)) {
5013 res_counter_uncharge(&memcg
->res
, PAGE_SIZE
* count
);
5016 mc
.precharge
+= count
;
5020 /* fall back to one by one charge */
5022 if (signal_pending(current
)) {
5026 if (!batch_count
--) {
5027 batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5030 ret
= __mem_cgroup_try_charge(NULL
,
5031 GFP_KERNEL
, 1, &memcg
, false);
5033 /* mem_cgroup_clear_mc() will do uncharge later */
5041 * is_target_pte_for_mc - check a pte whether it is valid for move charge
5042 * @vma: the vma the pte to be checked belongs
5043 * @addr: the address corresponding to the pte to be checked
5044 * @ptent: the pte to be checked
5045 * @target: the pointer the target page or swap ent will be stored(can be NULL)
5048 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5049 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5050 * move charge. if @target is not NULL, the page is stored in target->page
5051 * with extra refcnt got(Callers should handle it).
5052 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5053 * target for charge migration. if @target is not NULL, the entry is stored
5056 * Called with pte lock held.
5063 enum mc_target_type
{
5064 MC_TARGET_NONE
, /* not used */
5069 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
5070 unsigned long addr
, pte_t ptent
)
5072 struct page
*page
= vm_normal_page(vma
, addr
, ptent
);
5074 if (!page
|| !page_mapped(page
))
5076 if (PageAnon(page
)) {
5077 /* we don't move shared anon */
5078 if (!move_anon() || page_mapcount(page
) > 1)
5080 } else if (!move_file())
5081 /* we ignore mapcount for file pages */
5083 if (!get_page_unless_zero(page
))
5089 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5090 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5093 struct page
*page
= NULL
;
5094 swp_entry_t ent
= pte_to_swp_entry(ptent
);
5096 if (!move_anon() || non_swap_entry(ent
))
5098 usage_count
= mem_cgroup_count_swap_user(ent
, &page
);
5099 if (usage_count
> 1) { /* we don't move shared anon */
5104 if (do_swap_account
)
5105 entry
->val
= ent
.val
;
5110 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
5111 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5113 struct page
*page
= NULL
;
5114 struct inode
*inode
;
5115 struct address_space
*mapping
;
5118 if (!vma
->vm_file
) /* anonymous vma */
5123 inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
5124 mapping
= vma
->vm_file
->f_mapping
;
5125 if (pte_none(ptent
))
5126 pgoff
= linear_page_index(vma
, addr
);
5127 else /* pte_file(ptent) is true */
5128 pgoff
= pte_to_pgoff(ptent
);
5130 /* page is moved even if it's not RSS of this task(page-faulted). */
5131 page
= find_get_page(mapping
, pgoff
);
5134 /* shmem/tmpfs may report page out on swap: account for that too. */
5135 if (radix_tree_exceptional_entry(page
)) {
5136 swp_entry_t swap
= radix_to_swp_entry(page
);
5137 if (do_swap_account
)
5139 page
= find_get_page(&swapper_space
, swap
.val
);
5145 static int is_target_pte_for_mc(struct vm_area_struct
*vma
,
5146 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
5148 struct page
*page
= NULL
;
5149 struct page_cgroup
*pc
;
5151 swp_entry_t ent
= { .val
= 0 };
5153 if (pte_present(ptent
))
5154 page
= mc_handle_present_pte(vma
, addr
, ptent
);
5155 else if (is_swap_pte(ptent
))
5156 page
= mc_handle_swap_pte(vma
, addr
, ptent
, &ent
);
5157 else if (pte_none(ptent
) || pte_file(ptent
))
5158 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
5160 if (!page
&& !ent
.val
)
5163 pc
= lookup_page_cgroup(page
);
5165 * Do only loose check w/o page_cgroup lock.
5166 * mem_cgroup_move_account() checks the pc is valid or not under
5169 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5170 ret
= MC_TARGET_PAGE
;
5172 target
->page
= page
;
5174 if (!ret
|| !target
)
5177 /* There is a swap entry and a page doesn't exist or isn't charged */
5178 if (ent
.val
&& !ret
&&
5179 css_id(&mc
.from
->css
) == lookup_swap_cgroup_id(ent
)) {
5180 ret
= MC_TARGET_SWAP
;
5187 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
5188 unsigned long addr
, unsigned long end
,
5189 struct mm_walk
*walk
)
5191 struct vm_area_struct
*vma
= walk
->private;
5195 split_huge_page_pmd(walk
->mm
, pmd
);
5197 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5198 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
5199 if (is_target_pte_for_mc(vma
, addr
, *pte
, NULL
))
5200 mc
.precharge
++; /* increment precharge temporarily */
5201 pte_unmap_unlock(pte
- 1, ptl
);
5207 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
5209 unsigned long precharge
;
5210 struct vm_area_struct
*vma
;
5212 down_read(&mm
->mmap_sem
);
5213 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5214 struct mm_walk mem_cgroup_count_precharge_walk
= {
5215 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
5219 if (is_vm_hugetlb_page(vma
))
5221 walk_page_range(vma
->vm_start
, vma
->vm_end
,
5222 &mem_cgroup_count_precharge_walk
);
5224 up_read(&mm
->mmap_sem
);
5226 precharge
= mc
.precharge
;
5232 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
5234 unsigned long precharge
= mem_cgroup_count_precharge(mm
);
5236 VM_BUG_ON(mc
.moving_task
);
5237 mc
.moving_task
= current
;
5238 return mem_cgroup_do_precharge(precharge
);
5241 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5242 static void __mem_cgroup_clear_mc(void)
5244 struct mem_cgroup
*from
= mc
.from
;
5245 struct mem_cgroup
*to
= mc
.to
;
5247 /* we must uncharge all the leftover precharges from mc.to */
5249 __mem_cgroup_cancel_charge(mc
.to
, mc
.precharge
);
5253 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5254 * we must uncharge here.
5256 if (mc
.moved_charge
) {
5257 __mem_cgroup_cancel_charge(mc
.from
, mc
.moved_charge
);
5258 mc
.moved_charge
= 0;
5260 /* we must fixup refcnts and charges */
5261 if (mc
.moved_swap
) {
5262 /* uncharge swap account from the old cgroup */
5263 if (!mem_cgroup_is_root(mc
.from
))
5264 res_counter_uncharge(&mc
.from
->memsw
,
5265 PAGE_SIZE
* mc
.moved_swap
);
5266 __mem_cgroup_put(mc
.from
, mc
.moved_swap
);
5268 if (!mem_cgroup_is_root(mc
.to
)) {
5270 * we charged both to->res and to->memsw, so we should
5273 res_counter_uncharge(&mc
.to
->res
,
5274 PAGE_SIZE
* mc
.moved_swap
);
5276 /* we've already done mem_cgroup_get(mc.to) */
5279 memcg_oom_recover(from
);
5280 memcg_oom_recover(to
);
5281 wake_up_all(&mc
.waitq
);
5284 static void mem_cgroup_clear_mc(void)
5286 struct mem_cgroup
*from
= mc
.from
;
5289 * we must clear moving_task before waking up waiters at the end of
5292 mc
.moving_task
= NULL
;
5293 __mem_cgroup_clear_mc();
5294 spin_lock(&mc
.lock
);
5297 spin_unlock(&mc
.lock
);
5298 mem_cgroup_end_move(from
);
5301 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
5302 struct cgroup
*cgroup
,
5303 struct cgroup_taskset
*tset
)
5305 struct task_struct
*p
= cgroup_taskset_first(tset
);
5307 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgroup
);
5309 if (memcg
->move_charge_at_immigrate
) {
5310 struct mm_struct
*mm
;
5311 struct mem_cgroup
*from
= mem_cgroup_from_task(p
);
5313 VM_BUG_ON(from
== memcg
);
5315 mm
= get_task_mm(p
);
5318 /* We move charges only when we move a owner of the mm */
5319 if (mm
->owner
== p
) {
5322 VM_BUG_ON(mc
.precharge
);
5323 VM_BUG_ON(mc
.moved_charge
);
5324 VM_BUG_ON(mc
.moved_swap
);
5325 mem_cgroup_start_move(from
);
5326 spin_lock(&mc
.lock
);
5329 spin_unlock(&mc
.lock
);
5330 /* We set mc.moving_task later */
5332 ret
= mem_cgroup_precharge_mc(mm
);
5334 mem_cgroup_clear_mc();
5341 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
5342 struct cgroup
*cgroup
,
5343 struct cgroup_taskset
*tset
)
5345 mem_cgroup_clear_mc();
5348 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
5349 unsigned long addr
, unsigned long end
,
5350 struct mm_walk
*walk
)
5353 struct vm_area_struct
*vma
= walk
->private;
5357 split_huge_page_pmd(walk
->mm
, pmd
);
5359 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5360 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
5361 pte_t ptent
= *(pte
++);
5362 union mc_target target
;
5365 struct page_cgroup
*pc
;
5371 type
= is_target_pte_for_mc(vma
, addr
, ptent
, &target
);
5373 case MC_TARGET_PAGE
:
5375 if (isolate_lru_page(page
))
5377 pc
= lookup_page_cgroup(page
);
5378 if (!mem_cgroup_move_account(page
, 1, pc
,
5379 mc
.from
, mc
.to
, false)) {
5381 /* we uncharge from mc.from later. */
5384 putback_lru_page(page
);
5385 put
: /* is_target_pte_for_mc() gets the page */
5388 case MC_TARGET_SWAP
:
5390 if (!mem_cgroup_move_swap_account(ent
,
5391 mc
.from
, mc
.to
, false)) {
5393 /* we fixup refcnts and charges later. */
5401 pte_unmap_unlock(pte
- 1, ptl
);
5406 * We have consumed all precharges we got in can_attach().
5407 * We try charge one by one, but don't do any additional
5408 * charges to mc.to if we have failed in charge once in attach()
5411 ret
= mem_cgroup_do_precharge(1);
5419 static void mem_cgroup_move_charge(struct mm_struct
*mm
)
5421 struct vm_area_struct
*vma
;
5423 lru_add_drain_all();
5425 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
5427 * Someone who are holding the mmap_sem might be waiting in
5428 * waitq. So we cancel all extra charges, wake up all waiters,
5429 * and retry. Because we cancel precharges, we might not be able
5430 * to move enough charges, but moving charge is a best-effort
5431 * feature anyway, so it wouldn't be a big problem.
5433 __mem_cgroup_clear_mc();
5437 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5439 struct mm_walk mem_cgroup_move_charge_walk
= {
5440 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
5444 if (is_vm_hugetlb_page(vma
))
5446 ret
= walk_page_range(vma
->vm_start
, vma
->vm_end
,
5447 &mem_cgroup_move_charge_walk
);
5450 * means we have consumed all precharges and failed in
5451 * doing additional charge. Just abandon here.
5455 up_read(&mm
->mmap_sem
);
5458 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
5459 struct cgroup
*cont
,
5460 struct cgroup_taskset
*tset
)
5462 struct task_struct
*p
= cgroup_taskset_first(tset
);
5463 struct mm_struct
*mm
= get_task_mm(p
);
5467 mem_cgroup_move_charge(mm
);
5472 mem_cgroup_clear_mc();
5474 #else /* !CONFIG_MMU */
5475 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
5476 struct cgroup
*cgroup
,
5477 struct cgroup_taskset
*tset
)
5481 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
5482 struct cgroup
*cgroup
,
5483 struct cgroup_taskset
*tset
)
5486 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
5487 struct cgroup
*cont
,
5488 struct cgroup_taskset
*tset
)
5493 struct cgroup_subsys mem_cgroup_subsys
= {
5495 .subsys_id
= mem_cgroup_subsys_id
,
5496 .create
= mem_cgroup_create
,
5497 .pre_destroy
= mem_cgroup_pre_destroy
,
5498 .destroy
= mem_cgroup_destroy
,
5499 .populate
= mem_cgroup_populate
,
5500 .can_attach
= mem_cgroup_can_attach
,
5501 .cancel_attach
= mem_cgroup_cancel_attach
,
5502 .attach
= mem_cgroup_move_task
,
5507 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
5508 static int __init
enable_swap_account(char *s
)
5510 /* consider enabled if no parameter or 1 is given */
5511 if (!strcmp(s
, "1"))
5512 really_do_swap_account
= 1;
5513 else if (!strcmp(s
, "0"))
5514 really_do_swap_account
= 0;
5517 __setup("swapaccount=", enable_swap_account
);