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
377 #include <net/sock.h>
380 static bool mem_cgroup_is_root(struct mem_cgroup
*memcg
);
381 void sock_update_memcg(struct sock
*sk
)
383 if (static_branch(&memcg_socket_limit_enabled
)) {
384 struct mem_cgroup
*memcg
;
386 BUG_ON(!sk
->sk_prot
->proto_cgroup
);
388 /* Socket cloning can throw us here with sk_cgrp already
389 * filled. It won't however, necessarily happen from
390 * process context. So the test for root memcg given
391 * the current task's memcg won't help us in this case.
393 * Respecting the original socket's memcg is a better
394 * decision in this case.
397 BUG_ON(mem_cgroup_is_root(sk
->sk_cgrp
->memcg
));
398 mem_cgroup_get(sk
->sk_cgrp
->memcg
);
403 memcg
= mem_cgroup_from_task(current
);
404 if (!mem_cgroup_is_root(memcg
)) {
405 mem_cgroup_get(memcg
);
406 sk
->sk_cgrp
= sk
->sk_prot
->proto_cgroup(memcg
);
411 EXPORT_SYMBOL(sock_update_memcg
);
413 void sock_release_memcg(struct sock
*sk
)
415 if (static_branch(&memcg_socket_limit_enabled
) && sk
->sk_cgrp
) {
416 struct mem_cgroup
*memcg
;
417 WARN_ON(!sk
->sk_cgrp
->memcg
);
418 memcg
= sk
->sk_cgrp
->memcg
;
419 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
))) {
779 bool do_softlimit
, do_numainfo
;
781 do_softlimit
= mem_cgroup_event_ratelimit(memcg
,
782 MEM_CGROUP_TARGET_SOFTLIMIT
);
784 do_numainfo
= mem_cgroup_event_ratelimit(memcg
,
785 MEM_CGROUP_TARGET_NUMAINFO
);
789 mem_cgroup_threshold(memcg
);
790 if (unlikely(do_softlimit
))
791 mem_cgroup_update_tree(memcg
, page
);
793 if (unlikely(do_numainfo
))
794 atomic_inc(&memcg
->numainfo_events
);
800 struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
802 return container_of(cgroup_subsys_state(cont
,
803 mem_cgroup_subsys_id
), struct mem_cgroup
,
807 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
810 * mm_update_next_owner() may clear mm->owner to NULL
811 * if it races with swapoff, page migration, etc.
812 * So this can be called with p == NULL.
817 return container_of(task_subsys_state(p
, mem_cgroup_subsys_id
),
818 struct mem_cgroup
, css
);
821 struct mem_cgroup
*try_get_mem_cgroup_from_mm(struct mm_struct
*mm
)
823 struct mem_cgroup
*memcg
= NULL
;
828 * Because we have no locks, mm->owner's may be being moved to other
829 * cgroup. We use css_tryget() here even if this looks
830 * pessimistic (rather than adding locks here).
834 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
835 if (unlikely(!memcg
))
837 } while (!css_tryget(&memcg
->css
));
843 * mem_cgroup_iter - iterate over memory cgroup hierarchy
844 * @root: hierarchy root
845 * @prev: previously returned memcg, NULL on first invocation
846 * @reclaim: cookie for shared reclaim walks, NULL for full walks
848 * Returns references to children of the hierarchy below @root, or
849 * @root itself, or %NULL after a full round-trip.
851 * Caller must pass the return value in @prev on subsequent
852 * invocations for reference counting, or use mem_cgroup_iter_break()
853 * to cancel a hierarchy walk before the round-trip is complete.
855 * Reclaimers can specify a zone and a priority level in @reclaim to
856 * divide up the memcgs in the hierarchy among all concurrent
857 * reclaimers operating on the same zone and priority.
859 struct mem_cgroup
*mem_cgroup_iter(struct mem_cgroup
*root
,
860 struct mem_cgroup
*prev
,
861 struct mem_cgroup_reclaim_cookie
*reclaim
)
863 struct mem_cgroup
*memcg
= NULL
;
866 if (mem_cgroup_disabled())
870 root
= root_mem_cgroup
;
872 if (prev
&& !reclaim
)
873 id
= css_id(&prev
->css
);
875 if (prev
&& prev
!= root
)
878 if (!root
->use_hierarchy
&& root
!= root_mem_cgroup
) {
885 struct mem_cgroup_reclaim_iter
*uninitialized_var(iter
);
886 struct cgroup_subsys_state
*css
;
889 int nid
= zone_to_nid(reclaim
->zone
);
890 int zid
= zone_idx(reclaim
->zone
);
891 struct mem_cgroup_per_zone
*mz
;
893 mz
= mem_cgroup_zoneinfo(root
, nid
, zid
);
894 iter
= &mz
->reclaim_iter
[reclaim
->priority
];
895 if (prev
&& reclaim
->generation
!= iter
->generation
)
901 css
= css_get_next(&mem_cgroup_subsys
, id
+ 1, &root
->css
, &id
);
903 if (css
== &root
->css
|| css_tryget(css
))
904 memcg
= container_of(css
,
905 struct mem_cgroup
, css
);
914 else if (!prev
&& memcg
)
915 reclaim
->generation
= iter
->generation
;
925 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
926 * @root: hierarchy root
927 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
929 void mem_cgroup_iter_break(struct mem_cgroup
*root
,
930 struct mem_cgroup
*prev
)
933 root
= root_mem_cgroup
;
934 if (prev
&& prev
!= root
)
939 * Iteration constructs for visiting all cgroups (under a tree). If
940 * loops are exited prematurely (break), mem_cgroup_iter_break() must
941 * be used for reference counting.
943 #define for_each_mem_cgroup_tree(iter, root) \
944 for (iter = mem_cgroup_iter(root, NULL, NULL); \
946 iter = mem_cgroup_iter(root, iter, NULL))
948 #define for_each_mem_cgroup(iter) \
949 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
951 iter = mem_cgroup_iter(NULL, iter, NULL))
953 static inline bool mem_cgroup_is_root(struct mem_cgroup
*memcg
)
955 return (memcg
== root_mem_cgroup
);
958 void mem_cgroup_count_vm_event(struct mm_struct
*mm
, enum vm_event_item idx
)
960 struct mem_cgroup
*memcg
;
966 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
967 if (unlikely(!memcg
))
972 this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGFAULT
]);
975 this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGMAJFAULT
]);
983 EXPORT_SYMBOL(mem_cgroup_count_vm_event
);
986 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
987 * @zone: zone of the wanted lruvec
988 * @mem: memcg of the wanted lruvec
990 * Returns the lru list vector holding pages for the given @zone and
991 * @mem. This can be the global zone lruvec, if the memory controller
994 struct lruvec
*mem_cgroup_zone_lruvec(struct zone
*zone
,
995 struct mem_cgroup
*memcg
)
997 struct mem_cgroup_per_zone
*mz
;
999 if (mem_cgroup_disabled())
1000 return &zone
->lruvec
;
1002 mz
= mem_cgroup_zoneinfo(memcg
, zone_to_nid(zone
), zone_idx(zone
));
1007 * Following LRU functions are allowed to be used without PCG_LOCK.
1008 * Operations are called by routine of global LRU independently from memcg.
1009 * What we have to take care of here is validness of pc->mem_cgroup.
1011 * Changes to pc->mem_cgroup happens when
1014 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1015 * It is added to LRU before charge.
1016 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1017 * When moving account, the page is not on LRU. It's isolated.
1021 * mem_cgroup_lru_add_list - account for adding an lru page and return lruvec
1022 * @zone: zone of the page
1026 * This function accounts for @page being added to @lru, and returns
1027 * the lruvec for the given @zone and the memcg @page is charged to.
1029 * The callsite is then responsible for physically linking the page to
1030 * the returned lruvec->lists[@lru].
1032 struct lruvec
*mem_cgroup_lru_add_list(struct zone
*zone
, struct page
*page
,
1035 struct mem_cgroup_per_zone
*mz
;
1036 struct mem_cgroup
*memcg
;
1037 struct page_cgroup
*pc
;
1039 if (mem_cgroup_disabled())
1040 return &zone
->lruvec
;
1042 pc
= lookup_page_cgroup(page
);
1043 memcg
= pc
->mem_cgroup
;
1044 mz
= page_cgroup_zoneinfo(memcg
, page
);
1045 /* compound_order() is stabilized through lru_lock */
1046 MEM_CGROUP_ZSTAT(mz
, lru
) += 1 << compound_order(page
);
1051 * mem_cgroup_lru_del_list - account for removing an lru page
1055 * This function accounts for @page being removed from @lru.
1057 * The callsite is then responsible for physically unlinking
1060 void mem_cgroup_lru_del_list(struct page
*page
, enum lru_list lru
)
1062 struct mem_cgroup_per_zone
*mz
;
1063 struct mem_cgroup
*memcg
;
1064 struct page_cgroup
*pc
;
1066 if (mem_cgroup_disabled())
1069 pc
= lookup_page_cgroup(page
);
1070 memcg
= pc
->mem_cgroup
;
1072 mz
= page_cgroup_zoneinfo(memcg
, page
);
1073 /* huge page split is done under lru_lock. so, we have no races. */
1074 MEM_CGROUP_ZSTAT(mz
, lru
) -= 1 << compound_order(page
);
1077 void mem_cgroup_lru_del(struct page
*page
)
1079 mem_cgroup_lru_del_list(page
, page_lru(page
));
1083 * mem_cgroup_lru_move_lists - account for moving a page between lrus
1084 * @zone: zone of the page
1086 * @from: current lru
1089 * This function accounts for @page being moved between the lrus @from
1090 * and @to, and returns the lruvec for the given @zone and the memcg
1091 * @page is charged to.
1093 * The callsite is then responsible for physically relinking
1094 * @page->lru to the returned lruvec->lists[@to].
1096 struct lruvec
*mem_cgroup_lru_move_lists(struct zone
*zone
,
1101 /* XXX: Optimize this, especially for @from == @to */
1102 mem_cgroup_lru_del_list(page
, from
);
1103 return mem_cgroup_lru_add_list(zone
, page
, to
);
1107 * Checks whether given mem is same or in the root_mem_cgroup's
1110 static bool mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_memcg
,
1111 struct mem_cgroup
*memcg
)
1113 if (root_memcg
!= memcg
) {
1114 return (root_memcg
->use_hierarchy
&&
1115 css_is_ancestor(&memcg
->css
, &root_memcg
->css
));
1121 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*memcg
)
1124 struct mem_cgroup
*curr
= NULL
;
1125 struct task_struct
*p
;
1127 p
= find_lock_task_mm(task
);
1129 curr
= try_get_mem_cgroup_from_mm(p
->mm
);
1133 * All threads may have already detached their mm's, but the oom
1134 * killer still needs to detect if they have already been oom
1135 * killed to prevent needlessly killing additional tasks.
1138 curr
= mem_cgroup_from_task(task
);
1140 css_get(&curr
->css
);
1146 * We should check use_hierarchy of "memcg" not "curr". Because checking
1147 * use_hierarchy of "curr" here make this function true if hierarchy is
1148 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1149 * hierarchy(even if use_hierarchy is disabled in "memcg").
1151 ret
= mem_cgroup_same_or_subtree(memcg
, curr
);
1152 css_put(&curr
->css
);
1156 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup
*memcg
, struct zone
*zone
)
1158 unsigned long inactive_ratio
;
1159 int nid
= zone_to_nid(zone
);
1160 int zid
= zone_idx(zone
);
1161 unsigned long inactive
;
1162 unsigned long active
;
1165 inactive
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1166 BIT(LRU_INACTIVE_ANON
));
1167 active
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1168 BIT(LRU_ACTIVE_ANON
));
1170 gb
= (inactive
+ active
) >> (30 - PAGE_SHIFT
);
1172 inactive_ratio
= int_sqrt(10 * gb
);
1176 return inactive
* inactive_ratio
< active
;
1179 int mem_cgroup_inactive_file_is_low(struct mem_cgroup
*memcg
, struct zone
*zone
)
1181 unsigned long active
;
1182 unsigned long inactive
;
1183 int zid
= zone_idx(zone
);
1184 int nid
= zone_to_nid(zone
);
1186 inactive
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1187 BIT(LRU_INACTIVE_FILE
));
1188 active
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1189 BIT(LRU_ACTIVE_FILE
));
1191 return (active
> inactive
);
1194 struct zone_reclaim_stat
*mem_cgroup_get_reclaim_stat(struct mem_cgroup
*memcg
,
1197 int nid
= zone_to_nid(zone
);
1198 int zid
= zone_idx(zone
);
1199 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
1201 return &mz
->reclaim_stat
;
1204 struct zone_reclaim_stat
*
1205 mem_cgroup_get_reclaim_stat_from_page(struct page
*page
)
1207 struct page_cgroup
*pc
;
1208 struct mem_cgroup_per_zone
*mz
;
1210 if (mem_cgroup_disabled())
1213 pc
= lookup_page_cgroup(page
);
1214 if (!PageCgroupUsed(pc
))
1216 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1218 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
1219 return &mz
->reclaim_stat
;
1222 #define mem_cgroup_from_res_counter(counter, member) \
1223 container_of(counter, struct mem_cgroup, member)
1226 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1227 * @mem: the memory cgroup
1229 * Returns the maximum amount of memory @mem can be charged with, in
1232 static unsigned long mem_cgroup_margin(struct mem_cgroup
*memcg
)
1234 unsigned long long margin
;
1236 margin
= res_counter_margin(&memcg
->res
);
1237 if (do_swap_account
)
1238 margin
= min(margin
, res_counter_margin(&memcg
->memsw
));
1239 return margin
>> PAGE_SHIFT
;
1242 int mem_cgroup_swappiness(struct mem_cgroup
*memcg
)
1244 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
1247 if (cgrp
->parent
== NULL
)
1248 return vm_swappiness
;
1250 return memcg
->swappiness
;
1253 static void mem_cgroup_start_move(struct mem_cgroup
*memcg
)
1258 spin_lock(&memcg
->pcp_counter_lock
);
1259 for_each_online_cpu(cpu
)
1260 per_cpu(memcg
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) += 1;
1261 memcg
->nocpu_base
.count
[MEM_CGROUP_ON_MOVE
] += 1;
1262 spin_unlock(&memcg
->pcp_counter_lock
);
1268 static void mem_cgroup_end_move(struct mem_cgroup
*memcg
)
1275 spin_lock(&memcg
->pcp_counter_lock
);
1276 for_each_online_cpu(cpu
)
1277 per_cpu(memcg
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) -= 1;
1278 memcg
->nocpu_base
.count
[MEM_CGROUP_ON_MOVE
] -= 1;
1279 spin_unlock(&memcg
->pcp_counter_lock
);
1283 * 2 routines for checking "mem" is under move_account() or not.
1285 * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used
1286 * for avoiding race in accounting. If true,
1287 * pc->mem_cgroup may be overwritten.
1289 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1290 * under hierarchy of moving cgroups. This is for
1291 * waiting at hith-memory prressure caused by "move".
1294 static bool mem_cgroup_stealed(struct mem_cgroup
*memcg
)
1296 VM_BUG_ON(!rcu_read_lock_held());
1297 return this_cpu_read(memcg
->stat
->count
[MEM_CGROUP_ON_MOVE
]) > 0;
1300 static bool mem_cgroup_under_move(struct mem_cgroup
*memcg
)
1302 struct mem_cgroup
*from
;
1303 struct mem_cgroup
*to
;
1306 * Unlike task_move routines, we access mc.to, mc.from not under
1307 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1309 spin_lock(&mc
.lock
);
1315 ret
= mem_cgroup_same_or_subtree(memcg
, from
)
1316 || mem_cgroup_same_or_subtree(memcg
, to
);
1318 spin_unlock(&mc
.lock
);
1322 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*memcg
)
1324 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1325 if (mem_cgroup_under_move(memcg
)) {
1327 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1328 /* moving charge context might have finished. */
1331 finish_wait(&mc
.waitq
, &wait
);
1339 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1340 * @memcg: The memory cgroup that went over limit
1341 * @p: Task that is going to be killed
1343 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1346 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1348 struct cgroup
*task_cgrp
;
1349 struct cgroup
*mem_cgrp
;
1351 * Need a buffer in BSS, can't rely on allocations. The code relies
1352 * on the assumption that OOM is serialized for memory controller.
1353 * If this assumption is broken, revisit this code.
1355 static char memcg_name
[PATH_MAX
];
1364 mem_cgrp
= memcg
->css
.cgroup
;
1365 task_cgrp
= task_cgroup(p
, mem_cgroup_subsys_id
);
1367 ret
= cgroup_path(task_cgrp
, memcg_name
, PATH_MAX
);
1370 * Unfortunately, we are unable to convert to a useful name
1371 * But we'll still print out the usage information
1378 printk(KERN_INFO
"Task in %s killed", memcg_name
);
1381 ret
= cgroup_path(mem_cgrp
, memcg_name
, PATH_MAX
);
1389 * Continues from above, so we don't need an KERN_ level
1391 printk(KERN_CONT
" as a result of limit of %s\n", memcg_name
);
1394 printk(KERN_INFO
"memory: usage %llukB, limit %llukB, failcnt %llu\n",
1395 res_counter_read_u64(&memcg
->res
, RES_USAGE
) >> 10,
1396 res_counter_read_u64(&memcg
->res
, RES_LIMIT
) >> 10,
1397 res_counter_read_u64(&memcg
->res
, RES_FAILCNT
));
1398 printk(KERN_INFO
"memory+swap: usage %llukB, limit %llukB, "
1400 res_counter_read_u64(&memcg
->memsw
, RES_USAGE
) >> 10,
1401 res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
) >> 10,
1402 res_counter_read_u64(&memcg
->memsw
, RES_FAILCNT
));
1406 * This function returns the number of memcg under hierarchy tree. Returns
1407 * 1(self count) if no children.
1409 static int mem_cgroup_count_children(struct mem_cgroup
*memcg
)
1412 struct mem_cgroup
*iter
;
1414 for_each_mem_cgroup_tree(iter
, memcg
)
1420 * Return the memory (and swap, if configured) limit for a memcg.
1422 u64
mem_cgroup_get_limit(struct mem_cgroup
*memcg
)
1427 limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
1428 limit
+= total_swap_pages
<< PAGE_SHIFT
;
1430 memsw
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
1432 * If memsw is finite and limits the amount of swap space available
1433 * to this memcg, return that limit.
1435 return min(limit
, memsw
);
1438 static unsigned long mem_cgroup_reclaim(struct mem_cgroup
*memcg
,
1440 unsigned long flags
)
1442 unsigned long total
= 0;
1443 bool noswap
= false;
1446 if (flags
& MEM_CGROUP_RECLAIM_NOSWAP
)
1448 if (!(flags
& MEM_CGROUP_RECLAIM_SHRINK
) && memcg
->memsw_is_minimum
)
1451 for (loop
= 0; loop
< MEM_CGROUP_MAX_RECLAIM_LOOPS
; loop
++) {
1453 drain_all_stock_async(memcg
);
1454 total
+= try_to_free_mem_cgroup_pages(memcg
, gfp_mask
, noswap
);
1456 * Allow limit shrinkers, which are triggered directly
1457 * by userspace, to catch signals and stop reclaim
1458 * after minimal progress, regardless of the margin.
1460 if (total
&& (flags
& MEM_CGROUP_RECLAIM_SHRINK
))
1462 if (mem_cgroup_margin(memcg
))
1465 * If nothing was reclaimed after two attempts, there
1466 * may be no reclaimable pages in this hierarchy.
1475 * test_mem_cgroup_node_reclaimable
1476 * @mem: the target memcg
1477 * @nid: the node ID to be checked.
1478 * @noswap : specify true here if the user wants flle only information.
1480 * This function returns whether the specified memcg contains any
1481 * reclaimable pages on a node. Returns true if there are any reclaimable
1482 * pages in the node.
1484 static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup
*memcg
,
1485 int nid
, bool noswap
)
1487 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_FILE
))
1489 if (noswap
|| !total_swap_pages
)
1491 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_ANON
))
1496 #if MAX_NUMNODES > 1
1499 * Always updating the nodemask is not very good - even if we have an empty
1500 * list or the wrong list here, we can start from some node and traverse all
1501 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1504 static void mem_cgroup_may_update_nodemask(struct mem_cgroup
*memcg
)
1508 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1509 * pagein/pageout changes since the last update.
1511 if (!atomic_read(&memcg
->numainfo_events
))
1513 if (atomic_inc_return(&memcg
->numainfo_updating
) > 1)
1516 /* make a nodemask where this memcg uses memory from */
1517 memcg
->scan_nodes
= node_states
[N_HIGH_MEMORY
];
1519 for_each_node_mask(nid
, node_states
[N_HIGH_MEMORY
]) {
1521 if (!test_mem_cgroup_node_reclaimable(memcg
, nid
, false))
1522 node_clear(nid
, memcg
->scan_nodes
);
1525 atomic_set(&memcg
->numainfo_events
, 0);
1526 atomic_set(&memcg
->numainfo_updating
, 0);
1530 * Selecting a node where we start reclaim from. Because what we need is just
1531 * reducing usage counter, start from anywhere is O,K. Considering
1532 * memory reclaim from current node, there are pros. and cons.
1534 * Freeing memory from current node means freeing memory from a node which
1535 * we'll use or we've used. So, it may make LRU bad. And if several threads
1536 * hit limits, it will see a contention on a node. But freeing from remote
1537 * node means more costs for memory reclaim because of memory latency.
1539 * Now, we use round-robin. Better algorithm is welcomed.
1541 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1545 mem_cgroup_may_update_nodemask(memcg
);
1546 node
= memcg
->last_scanned_node
;
1548 node
= next_node(node
, memcg
->scan_nodes
);
1549 if (node
== MAX_NUMNODES
)
1550 node
= first_node(memcg
->scan_nodes
);
1552 * We call this when we hit limit, not when pages are added to LRU.
1553 * No LRU may hold pages because all pages are UNEVICTABLE or
1554 * memcg is too small and all pages are not on LRU. In that case,
1555 * we use curret node.
1557 if (unlikely(node
== MAX_NUMNODES
))
1558 node
= numa_node_id();
1560 memcg
->last_scanned_node
= node
;
1565 * Check all nodes whether it contains reclaimable pages or not.
1566 * For quick scan, we make use of scan_nodes. This will allow us to skip
1567 * unused nodes. But scan_nodes is lazily updated and may not cotain
1568 * enough new information. We need to do double check.
1570 bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1575 * quick check...making use of scan_node.
1576 * We can skip unused nodes.
1578 if (!nodes_empty(memcg
->scan_nodes
)) {
1579 for (nid
= first_node(memcg
->scan_nodes
);
1581 nid
= next_node(nid
, memcg
->scan_nodes
)) {
1583 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1588 * Check rest of nodes.
1590 for_each_node_state(nid
, N_HIGH_MEMORY
) {
1591 if (node_isset(nid
, memcg
->scan_nodes
))
1593 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1600 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1605 bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1607 return test_mem_cgroup_node_reclaimable(memcg
, 0, noswap
);
1611 static int mem_cgroup_soft_reclaim(struct mem_cgroup
*root_memcg
,
1614 unsigned long *total_scanned
)
1616 struct mem_cgroup
*victim
= NULL
;
1619 unsigned long excess
;
1620 unsigned long nr_scanned
;
1621 struct mem_cgroup_reclaim_cookie reclaim
= {
1626 excess
= res_counter_soft_limit_excess(&root_memcg
->res
) >> PAGE_SHIFT
;
1629 victim
= mem_cgroup_iter(root_memcg
, victim
, &reclaim
);
1634 * If we have not been able to reclaim
1635 * anything, it might because there are
1636 * no reclaimable pages under this hierarchy
1641 * We want to do more targeted reclaim.
1642 * excess >> 2 is not to excessive so as to
1643 * reclaim too much, nor too less that we keep
1644 * coming back to reclaim from this cgroup
1646 if (total
>= (excess
>> 2) ||
1647 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
))
1652 if (!mem_cgroup_reclaimable(victim
, false))
1654 total
+= mem_cgroup_shrink_node_zone(victim
, gfp_mask
, false,
1656 *total_scanned
+= nr_scanned
;
1657 if (!res_counter_soft_limit_excess(&root_memcg
->res
))
1660 mem_cgroup_iter_break(root_memcg
, victim
);
1665 * Check OOM-Killer is already running under our hierarchy.
1666 * If someone is running, return false.
1667 * Has to be called with memcg_oom_lock
1669 static bool mem_cgroup_oom_lock(struct mem_cgroup
*memcg
)
1671 struct mem_cgroup
*iter
, *failed
= NULL
;
1673 for_each_mem_cgroup_tree(iter
, memcg
) {
1674 if (iter
->oom_lock
) {
1676 * this subtree of our hierarchy is already locked
1677 * so we cannot give a lock.
1680 mem_cgroup_iter_break(memcg
, iter
);
1683 iter
->oom_lock
= true;
1690 * OK, we failed to lock the whole subtree so we have to clean up
1691 * what we set up to the failing subtree
1693 for_each_mem_cgroup_tree(iter
, memcg
) {
1694 if (iter
== failed
) {
1695 mem_cgroup_iter_break(memcg
, iter
);
1698 iter
->oom_lock
= false;
1704 * Has to be called with memcg_oom_lock
1706 static int mem_cgroup_oom_unlock(struct mem_cgroup
*memcg
)
1708 struct mem_cgroup
*iter
;
1710 for_each_mem_cgroup_tree(iter
, memcg
)
1711 iter
->oom_lock
= false;
1715 static void mem_cgroup_mark_under_oom(struct mem_cgroup
*memcg
)
1717 struct mem_cgroup
*iter
;
1719 for_each_mem_cgroup_tree(iter
, memcg
)
1720 atomic_inc(&iter
->under_oom
);
1723 static void mem_cgroup_unmark_under_oom(struct mem_cgroup
*memcg
)
1725 struct mem_cgroup
*iter
;
1728 * When a new child is created while the hierarchy is under oom,
1729 * mem_cgroup_oom_lock() may not be called. We have to use
1730 * atomic_add_unless() here.
1732 for_each_mem_cgroup_tree(iter
, memcg
)
1733 atomic_add_unless(&iter
->under_oom
, -1, 0);
1736 static DEFINE_SPINLOCK(memcg_oom_lock
);
1737 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1739 struct oom_wait_info
{
1740 struct mem_cgroup
*mem
;
1744 static int memcg_oom_wake_function(wait_queue_t
*wait
,
1745 unsigned mode
, int sync
, void *arg
)
1747 struct mem_cgroup
*wake_memcg
= (struct mem_cgroup
*)arg
,
1749 struct oom_wait_info
*oom_wait_info
;
1751 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1752 oom_wait_memcg
= oom_wait_info
->mem
;
1755 * Both of oom_wait_info->mem and wake_mem are stable under us.
1756 * Then we can use css_is_ancestor without taking care of RCU.
1758 if (!mem_cgroup_same_or_subtree(oom_wait_memcg
, wake_memcg
)
1759 && !mem_cgroup_same_or_subtree(wake_memcg
, oom_wait_memcg
))
1761 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1764 static void memcg_wakeup_oom(struct mem_cgroup
*memcg
)
1766 /* for filtering, pass "memcg" as argument. */
1767 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, memcg
);
1770 static void memcg_oom_recover(struct mem_cgroup
*memcg
)
1772 if (memcg
&& atomic_read(&memcg
->under_oom
))
1773 memcg_wakeup_oom(memcg
);
1777 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1779 bool mem_cgroup_handle_oom(struct mem_cgroup
*memcg
, gfp_t mask
)
1781 struct oom_wait_info owait
;
1782 bool locked
, need_to_kill
;
1785 owait
.wait
.flags
= 0;
1786 owait
.wait
.func
= memcg_oom_wake_function
;
1787 owait
.wait
.private = current
;
1788 INIT_LIST_HEAD(&owait
.wait
.task_list
);
1789 need_to_kill
= true;
1790 mem_cgroup_mark_under_oom(memcg
);
1792 /* At first, try to OOM lock hierarchy under memcg.*/
1793 spin_lock(&memcg_oom_lock
);
1794 locked
= mem_cgroup_oom_lock(memcg
);
1796 * Even if signal_pending(), we can't quit charge() loop without
1797 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1798 * under OOM is always welcomed, use TASK_KILLABLE here.
1800 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1801 if (!locked
|| memcg
->oom_kill_disable
)
1802 need_to_kill
= false;
1804 mem_cgroup_oom_notify(memcg
);
1805 spin_unlock(&memcg_oom_lock
);
1808 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1809 mem_cgroup_out_of_memory(memcg
, mask
);
1812 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1814 spin_lock(&memcg_oom_lock
);
1816 mem_cgroup_oom_unlock(memcg
);
1817 memcg_wakeup_oom(memcg
);
1818 spin_unlock(&memcg_oom_lock
);
1820 mem_cgroup_unmark_under_oom(memcg
);
1822 if (test_thread_flag(TIF_MEMDIE
) || fatal_signal_pending(current
))
1824 /* Give chance to dying process */
1825 schedule_timeout_uninterruptible(1);
1830 * Currently used to update mapped file statistics, but the routine can be
1831 * generalized to update other statistics as well.
1833 * Notes: Race condition
1835 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1836 * it tends to be costly. But considering some conditions, we doesn't need
1837 * to do so _always_.
1839 * Considering "charge", lock_page_cgroup() is not required because all
1840 * file-stat operations happen after a page is attached to radix-tree. There
1841 * are no race with "charge".
1843 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1844 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1845 * if there are race with "uncharge". Statistics itself is properly handled
1848 * Considering "move", this is an only case we see a race. To make the race
1849 * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are
1850 * possibility of race condition. If there is, we take a lock.
1853 void mem_cgroup_update_page_stat(struct page
*page
,
1854 enum mem_cgroup_page_stat_item idx
, int val
)
1856 struct mem_cgroup
*memcg
;
1857 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1858 bool need_unlock
= false;
1859 unsigned long uninitialized_var(flags
);
1861 if (mem_cgroup_disabled())
1865 memcg
= pc
->mem_cgroup
;
1866 if (unlikely(!memcg
|| !PageCgroupUsed(pc
)))
1868 /* pc->mem_cgroup is unstable ? */
1869 if (unlikely(mem_cgroup_stealed(memcg
)) || PageTransHuge(page
)) {
1870 /* take a lock against to access pc->mem_cgroup */
1871 move_lock_page_cgroup(pc
, &flags
);
1873 memcg
= pc
->mem_cgroup
;
1874 if (!memcg
|| !PageCgroupUsed(pc
))
1879 case MEMCG_NR_FILE_MAPPED
:
1881 SetPageCgroupFileMapped(pc
);
1882 else if (!page_mapped(page
))
1883 ClearPageCgroupFileMapped(pc
);
1884 idx
= MEM_CGROUP_STAT_FILE_MAPPED
;
1890 this_cpu_add(memcg
->stat
->count
[idx
], val
);
1893 if (unlikely(need_unlock
))
1894 move_unlock_page_cgroup(pc
, &flags
);
1898 EXPORT_SYMBOL(mem_cgroup_update_page_stat
);
1901 * size of first charge trial. "32" comes from vmscan.c's magic value.
1902 * TODO: maybe necessary to use big numbers in big irons.
1904 #define CHARGE_BATCH 32U
1905 struct memcg_stock_pcp
{
1906 struct mem_cgroup
*cached
; /* this never be root cgroup */
1907 unsigned int nr_pages
;
1908 struct work_struct work
;
1909 unsigned long flags
;
1910 #define FLUSHING_CACHED_CHARGE (0)
1912 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
1913 static DEFINE_MUTEX(percpu_charge_mutex
);
1916 * Try to consume stocked charge on this cpu. If success, one page is consumed
1917 * from local stock and true is returned. If the stock is 0 or charges from a
1918 * cgroup which is not current target, returns false. This stock will be
1921 static bool consume_stock(struct mem_cgroup
*memcg
)
1923 struct memcg_stock_pcp
*stock
;
1926 stock
= &get_cpu_var(memcg_stock
);
1927 if (memcg
== stock
->cached
&& stock
->nr_pages
)
1929 else /* need to call res_counter_charge */
1931 put_cpu_var(memcg_stock
);
1936 * Returns stocks cached in percpu to res_counter and reset cached information.
1938 static void drain_stock(struct memcg_stock_pcp
*stock
)
1940 struct mem_cgroup
*old
= stock
->cached
;
1942 if (stock
->nr_pages
) {
1943 unsigned long bytes
= stock
->nr_pages
* PAGE_SIZE
;
1945 res_counter_uncharge(&old
->res
, bytes
);
1946 if (do_swap_account
)
1947 res_counter_uncharge(&old
->memsw
, bytes
);
1948 stock
->nr_pages
= 0;
1950 stock
->cached
= NULL
;
1954 * This must be called under preempt disabled or must be called by
1955 * a thread which is pinned to local cpu.
1957 static void drain_local_stock(struct work_struct
*dummy
)
1959 struct memcg_stock_pcp
*stock
= &__get_cpu_var(memcg_stock
);
1961 clear_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
);
1965 * Cache charges(val) which is from res_counter, to local per_cpu area.
1966 * This will be consumed by consume_stock() function, later.
1968 static void refill_stock(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
1970 struct memcg_stock_pcp
*stock
= &get_cpu_var(memcg_stock
);
1972 if (stock
->cached
!= memcg
) { /* reset if necessary */
1974 stock
->cached
= memcg
;
1976 stock
->nr_pages
+= nr_pages
;
1977 put_cpu_var(memcg_stock
);
1981 * Drains all per-CPU charge caches for given root_memcg resp. subtree
1982 * of the hierarchy under it. sync flag says whether we should block
1983 * until the work is done.
1985 static void drain_all_stock(struct mem_cgroup
*root_memcg
, bool sync
)
1989 /* Notify other cpus that system-wide "drain" is running */
1992 for_each_online_cpu(cpu
) {
1993 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
1994 struct mem_cgroup
*memcg
;
1996 memcg
= stock
->cached
;
1997 if (!memcg
|| !stock
->nr_pages
)
1999 if (!mem_cgroup_same_or_subtree(root_memcg
, memcg
))
2001 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
)) {
2003 drain_local_stock(&stock
->work
);
2005 schedule_work_on(cpu
, &stock
->work
);
2013 for_each_online_cpu(cpu
) {
2014 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2015 if (test_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
))
2016 flush_work(&stock
->work
);
2023 * Tries to drain stocked charges in other cpus. This function is asynchronous
2024 * and just put a work per cpu for draining localy on each cpu. Caller can
2025 * expects some charges will be back to res_counter later but cannot wait for
2028 static void drain_all_stock_async(struct mem_cgroup
*root_memcg
)
2031 * If someone calls draining, avoid adding more kworker runs.
2033 if (!mutex_trylock(&percpu_charge_mutex
))
2035 drain_all_stock(root_memcg
, false);
2036 mutex_unlock(&percpu_charge_mutex
);
2039 /* This is a synchronous drain interface. */
2040 static void drain_all_stock_sync(struct mem_cgroup
*root_memcg
)
2042 /* called when force_empty is called */
2043 mutex_lock(&percpu_charge_mutex
);
2044 drain_all_stock(root_memcg
, true);
2045 mutex_unlock(&percpu_charge_mutex
);
2049 * This function drains percpu counter value from DEAD cpu and
2050 * move it to local cpu. Note that this function can be preempted.
2052 static void mem_cgroup_drain_pcp_counter(struct mem_cgroup
*memcg
, int cpu
)
2056 spin_lock(&memcg
->pcp_counter_lock
);
2057 for (i
= 0; i
< MEM_CGROUP_STAT_DATA
; i
++) {
2058 long x
= per_cpu(memcg
->stat
->count
[i
], cpu
);
2060 per_cpu(memcg
->stat
->count
[i
], cpu
) = 0;
2061 memcg
->nocpu_base
.count
[i
] += x
;
2063 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
2064 unsigned long x
= per_cpu(memcg
->stat
->events
[i
], cpu
);
2066 per_cpu(memcg
->stat
->events
[i
], cpu
) = 0;
2067 memcg
->nocpu_base
.events
[i
] += x
;
2069 /* need to clear ON_MOVE value, works as a kind of lock. */
2070 per_cpu(memcg
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) = 0;
2071 spin_unlock(&memcg
->pcp_counter_lock
);
2074 static void synchronize_mem_cgroup_on_move(struct mem_cgroup
*memcg
, int cpu
)
2076 int idx
= MEM_CGROUP_ON_MOVE
;
2078 spin_lock(&memcg
->pcp_counter_lock
);
2079 per_cpu(memcg
->stat
->count
[idx
], cpu
) = memcg
->nocpu_base
.count
[idx
];
2080 spin_unlock(&memcg
->pcp_counter_lock
);
2083 static int __cpuinit
memcg_cpu_hotplug_callback(struct notifier_block
*nb
,
2084 unsigned long action
,
2087 int cpu
= (unsigned long)hcpu
;
2088 struct memcg_stock_pcp
*stock
;
2089 struct mem_cgroup
*iter
;
2091 if ((action
== CPU_ONLINE
)) {
2092 for_each_mem_cgroup(iter
)
2093 synchronize_mem_cgroup_on_move(iter
, cpu
);
2097 if ((action
!= CPU_DEAD
) || action
!= CPU_DEAD_FROZEN
)
2100 for_each_mem_cgroup(iter
)
2101 mem_cgroup_drain_pcp_counter(iter
, cpu
);
2103 stock
= &per_cpu(memcg_stock
, cpu
);
2109 /* See __mem_cgroup_try_charge() for details */
2111 CHARGE_OK
, /* success */
2112 CHARGE_RETRY
, /* need to retry but retry is not bad */
2113 CHARGE_NOMEM
, /* we can't do more. return -ENOMEM */
2114 CHARGE_WOULDBLOCK
, /* GFP_WAIT wasn't set and no enough res. */
2115 CHARGE_OOM_DIE
, /* the current is killed because of OOM */
2118 static int mem_cgroup_do_charge(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
2119 unsigned int nr_pages
, bool oom_check
)
2121 unsigned long csize
= nr_pages
* PAGE_SIZE
;
2122 struct mem_cgroup
*mem_over_limit
;
2123 struct res_counter
*fail_res
;
2124 unsigned long flags
= 0;
2127 ret
= res_counter_charge(&memcg
->res
, csize
, &fail_res
);
2130 if (!do_swap_account
)
2132 ret
= res_counter_charge(&memcg
->memsw
, csize
, &fail_res
);
2136 res_counter_uncharge(&memcg
->res
, csize
);
2137 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, memsw
);
2138 flags
|= MEM_CGROUP_RECLAIM_NOSWAP
;
2140 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, res
);
2142 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2143 * of regular pages (CHARGE_BATCH), or a single regular page (1).
2145 * Never reclaim on behalf of optional batching, retry with a
2146 * single page instead.
2148 if (nr_pages
== CHARGE_BATCH
)
2149 return CHARGE_RETRY
;
2151 if (!(gfp_mask
& __GFP_WAIT
))
2152 return CHARGE_WOULDBLOCK
;
2154 ret
= mem_cgroup_reclaim(mem_over_limit
, gfp_mask
, flags
);
2155 if (mem_cgroup_margin(mem_over_limit
) >= nr_pages
)
2156 return CHARGE_RETRY
;
2158 * Even though the limit is exceeded at this point, reclaim
2159 * may have been able to free some pages. Retry the charge
2160 * before killing the task.
2162 * Only for regular pages, though: huge pages are rather
2163 * unlikely to succeed so close to the limit, and we fall back
2164 * to regular pages anyway in case of failure.
2166 if (nr_pages
== 1 && ret
)
2167 return CHARGE_RETRY
;
2170 * At task move, charge accounts can be doubly counted. So, it's
2171 * better to wait until the end of task_move if something is going on.
2173 if (mem_cgroup_wait_acct_move(mem_over_limit
))
2174 return CHARGE_RETRY
;
2176 /* If we don't need to call oom-killer at el, return immediately */
2178 return CHARGE_NOMEM
;
2180 if (!mem_cgroup_handle_oom(mem_over_limit
, gfp_mask
))
2181 return CHARGE_OOM_DIE
;
2183 return CHARGE_RETRY
;
2187 * __mem_cgroup_try_charge() does
2188 * 1. detect memcg to be charged against from passed *mm and *ptr,
2189 * 2. update res_counter
2190 * 3. call memory reclaim if necessary.
2192 * In some special case, if the task is fatal, fatal_signal_pending() or
2193 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2194 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2195 * as possible without any hazards. 2: all pages should have a valid
2196 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2197 * pointer, that is treated as a charge to root_mem_cgroup.
2199 * So __mem_cgroup_try_charge() will return
2200 * 0 ... on success, filling *ptr with a valid memcg pointer.
2201 * -ENOMEM ... charge failure because of resource limits.
2202 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2204 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2205 * the oom-killer can be invoked.
2207 static int __mem_cgroup_try_charge(struct mm_struct
*mm
,
2209 unsigned int nr_pages
,
2210 struct mem_cgroup
**ptr
,
2213 unsigned int batch
= max(CHARGE_BATCH
, nr_pages
);
2214 int nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2215 struct mem_cgroup
*memcg
= NULL
;
2219 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2220 * in system level. So, allow to go ahead dying process in addition to
2223 if (unlikely(test_thread_flag(TIF_MEMDIE
)
2224 || fatal_signal_pending(current
)))
2228 * We always charge the cgroup the mm_struct belongs to.
2229 * The mm_struct's mem_cgroup changes on task migration if the
2230 * thread group leader migrates. It's possible that mm is not
2231 * set, if so charge the init_mm (happens for pagecache usage).
2234 *ptr
= root_mem_cgroup
;
2236 if (*ptr
) { /* css should be a valid one */
2238 VM_BUG_ON(css_is_removed(&memcg
->css
));
2239 if (mem_cgroup_is_root(memcg
))
2241 if (nr_pages
== 1 && consume_stock(memcg
))
2243 css_get(&memcg
->css
);
2245 struct task_struct
*p
;
2248 p
= rcu_dereference(mm
->owner
);
2250 * Because we don't have task_lock(), "p" can exit.
2251 * In that case, "memcg" can point to root or p can be NULL with
2252 * race with swapoff. Then, we have small risk of mis-accouning.
2253 * But such kind of mis-account by race always happens because
2254 * we don't have cgroup_mutex(). It's overkill and we allo that
2256 * (*) swapoff at el will charge against mm-struct not against
2257 * task-struct. So, mm->owner can be NULL.
2259 memcg
= mem_cgroup_from_task(p
);
2261 memcg
= root_mem_cgroup
;
2262 if (mem_cgroup_is_root(memcg
)) {
2266 if (nr_pages
== 1 && consume_stock(memcg
)) {
2268 * It seems dagerous to access memcg without css_get().
2269 * But considering how consume_stok works, it's not
2270 * necessary. If consume_stock success, some charges
2271 * from this memcg are cached on this cpu. So, we
2272 * don't need to call css_get()/css_tryget() before
2273 * calling consume_stock().
2278 /* after here, we may be blocked. we need to get refcnt */
2279 if (!css_tryget(&memcg
->css
)) {
2289 /* If killed, bypass charge */
2290 if (fatal_signal_pending(current
)) {
2291 css_put(&memcg
->css
);
2296 if (oom
&& !nr_oom_retries
) {
2298 nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2301 ret
= mem_cgroup_do_charge(memcg
, gfp_mask
, batch
, oom_check
);
2305 case CHARGE_RETRY
: /* not in OOM situation but retry */
2307 css_put(&memcg
->css
);
2310 case CHARGE_WOULDBLOCK
: /* !__GFP_WAIT */
2311 css_put(&memcg
->css
);
2313 case CHARGE_NOMEM
: /* OOM routine works */
2315 css_put(&memcg
->css
);
2318 /* If oom, we never return -ENOMEM */
2321 case CHARGE_OOM_DIE
: /* Killed by OOM Killer */
2322 css_put(&memcg
->css
);
2325 } while (ret
!= CHARGE_OK
);
2327 if (batch
> nr_pages
)
2328 refill_stock(memcg
, batch
- nr_pages
);
2329 css_put(&memcg
->css
);
2337 *ptr
= root_mem_cgroup
;
2342 * Somemtimes we have to undo a charge we got by try_charge().
2343 * This function is for that and do uncharge, put css's refcnt.
2344 * gotten by try_charge().
2346 static void __mem_cgroup_cancel_charge(struct mem_cgroup
*memcg
,
2347 unsigned int nr_pages
)
2349 if (!mem_cgroup_is_root(memcg
)) {
2350 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2352 res_counter_uncharge(&memcg
->res
, bytes
);
2353 if (do_swap_account
)
2354 res_counter_uncharge(&memcg
->memsw
, bytes
);
2359 * A helper function to get mem_cgroup from ID. must be called under
2360 * rcu_read_lock(). The caller must check css_is_removed() or some if
2361 * it's concern. (dropping refcnt from swap can be called against removed
2364 static struct mem_cgroup
*mem_cgroup_lookup(unsigned short id
)
2366 struct cgroup_subsys_state
*css
;
2368 /* ID 0 is unused ID */
2371 css
= css_lookup(&mem_cgroup_subsys
, id
);
2374 return container_of(css
, struct mem_cgroup
, css
);
2377 struct mem_cgroup
*try_get_mem_cgroup_from_page(struct page
*page
)
2379 struct mem_cgroup
*memcg
= NULL
;
2380 struct page_cgroup
*pc
;
2384 VM_BUG_ON(!PageLocked(page
));
2386 pc
= lookup_page_cgroup(page
);
2387 lock_page_cgroup(pc
);
2388 if (PageCgroupUsed(pc
)) {
2389 memcg
= pc
->mem_cgroup
;
2390 if (memcg
&& !css_tryget(&memcg
->css
))
2392 } else if (PageSwapCache(page
)) {
2393 ent
.val
= page_private(page
);
2394 id
= lookup_swap_cgroup_id(ent
);
2396 memcg
= mem_cgroup_lookup(id
);
2397 if (memcg
&& !css_tryget(&memcg
->css
))
2401 unlock_page_cgroup(pc
);
2405 static void __mem_cgroup_commit_charge(struct mem_cgroup
*memcg
,
2407 unsigned int nr_pages
,
2408 struct page_cgroup
*pc
,
2409 enum charge_type ctype
)
2411 lock_page_cgroup(pc
);
2412 if (unlikely(PageCgroupUsed(pc
))) {
2413 unlock_page_cgroup(pc
);
2414 __mem_cgroup_cancel_charge(memcg
, nr_pages
);
2418 * we don't need page_cgroup_lock about tail pages, becase they are not
2419 * accessed by any other context at this point.
2421 pc
->mem_cgroup
= memcg
;
2423 * We access a page_cgroup asynchronously without lock_page_cgroup().
2424 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2425 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2426 * before USED bit, we need memory barrier here.
2427 * See mem_cgroup_add_lru_list(), etc.
2431 case MEM_CGROUP_CHARGE_TYPE_CACHE
:
2432 case MEM_CGROUP_CHARGE_TYPE_SHMEM
:
2433 SetPageCgroupCache(pc
);
2434 SetPageCgroupUsed(pc
);
2436 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2437 ClearPageCgroupCache(pc
);
2438 SetPageCgroupUsed(pc
);
2444 mem_cgroup_charge_statistics(memcg
, PageCgroupCache(pc
), nr_pages
);
2445 unlock_page_cgroup(pc
);
2446 WARN_ON_ONCE(PageLRU(page
));
2448 * "charge_statistics" updated event counter. Then, check it.
2449 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2450 * if they exceeds softlimit.
2452 memcg_check_events(memcg
, page
);
2455 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2457 #define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\
2458 (1 << PCG_MIGRATION))
2460 * Because tail pages are not marked as "used", set it. We're under
2461 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2462 * charge/uncharge will be never happen and move_account() is done under
2463 * compound_lock(), so we don't have to take care of races.
2465 void mem_cgroup_split_huge_fixup(struct page
*head
)
2467 struct page_cgroup
*head_pc
= lookup_page_cgroup(head
);
2468 struct mem_cgroup_per_zone
*mz
;
2469 struct page_cgroup
*pc
;
2473 if (mem_cgroup_disabled())
2475 for (i
= 1; i
< HPAGE_PMD_NR
; i
++) {
2477 pc
->mem_cgroup
= head_pc
->mem_cgroup
;
2478 smp_wmb();/* see __commit_charge() */
2479 pc
->flags
= head_pc
->flags
& ~PCGF_NOCOPY_AT_SPLIT
;
2482 * Tail pages will be added to LRU.
2483 * We hold lru_lock,then,reduce counter directly.
2485 lru
= page_lru(head
);
2486 mz
= page_cgroup_zoneinfo(head_pc
->mem_cgroup
, head
);
2487 MEM_CGROUP_ZSTAT(mz
, lru
) -= HPAGE_PMD_NR
- 1;
2492 * mem_cgroup_move_account - move account of the page
2494 * @nr_pages: number of regular pages (>1 for huge pages)
2495 * @pc: page_cgroup of the page.
2496 * @from: mem_cgroup which the page is moved from.
2497 * @to: mem_cgroup which the page is moved to. @from != @to.
2498 * @uncharge: whether we should call uncharge and css_put against @from.
2500 * The caller must confirm following.
2501 * - page is not on LRU (isolate_page() is useful.)
2502 * - compound_lock is held when nr_pages > 1
2504 * This function doesn't do "charge" nor css_get to new cgroup. It should be
2505 * done by a caller(__mem_cgroup_try_charge would be useful). If @uncharge is
2506 * true, this function does "uncharge" from old cgroup, but it doesn't if
2507 * @uncharge is false, so a caller should do "uncharge".
2509 static int mem_cgroup_move_account(struct page
*page
,
2510 unsigned int nr_pages
,
2511 struct page_cgroup
*pc
,
2512 struct mem_cgroup
*from
,
2513 struct mem_cgroup
*to
,
2516 unsigned long flags
;
2519 VM_BUG_ON(from
== to
);
2520 VM_BUG_ON(PageLRU(page
));
2522 * The page is isolated from LRU. So, collapse function
2523 * will not handle this page. But page splitting can happen.
2524 * Do this check under compound_page_lock(). The caller should
2528 if (nr_pages
> 1 && !PageTransHuge(page
))
2531 lock_page_cgroup(pc
);
2534 if (!PageCgroupUsed(pc
) || pc
->mem_cgroup
!= from
)
2537 move_lock_page_cgroup(pc
, &flags
);
2539 if (PageCgroupFileMapped(pc
)) {
2540 /* Update mapped_file data for mem_cgroup */
2542 __this_cpu_dec(from
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2543 __this_cpu_inc(to
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2546 mem_cgroup_charge_statistics(from
, PageCgroupCache(pc
), -nr_pages
);
2548 /* This is not "cancel", but cancel_charge does all we need. */
2549 __mem_cgroup_cancel_charge(from
, nr_pages
);
2551 /* caller should have done css_get */
2552 pc
->mem_cgroup
= to
;
2553 mem_cgroup_charge_statistics(to
, PageCgroupCache(pc
), nr_pages
);
2555 * We charges against "to" which may not have any tasks. Then, "to"
2556 * can be under rmdir(). But in current implementation, caller of
2557 * this function is just force_empty() and move charge, so it's
2558 * guaranteed that "to" is never removed. So, we don't check rmdir
2561 move_unlock_page_cgroup(pc
, &flags
);
2564 unlock_page_cgroup(pc
);
2568 memcg_check_events(to
, page
);
2569 memcg_check_events(from
, page
);
2575 * move charges to its parent.
2578 static int mem_cgroup_move_parent(struct page
*page
,
2579 struct page_cgroup
*pc
,
2580 struct mem_cgroup
*child
,
2583 struct cgroup
*cg
= child
->css
.cgroup
;
2584 struct cgroup
*pcg
= cg
->parent
;
2585 struct mem_cgroup
*parent
;
2586 unsigned int nr_pages
;
2587 unsigned long uninitialized_var(flags
);
2595 if (!get_page_unless_zero(page
))
2597 if (isolate_lru_page(page
))
2600 nr_pages
= hpage_nr_pages(page
);
2602 parent
= mem_cgroup_from_cont(pcg
);
2603 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, nr_pages
, &parent
, false);
2608 flags
= compound_lock_irqsave(page
);
2610 ret
= mem_cgroup_move_account(page
, nr_pages
, pc
, child
, parent
, true);
2612 __mem_cgroup_cancel_charge(parent
, nr_pages
);
2615 compound_unlock_irqrestore(page
, flags
);
2617 putback_lru_page(page
);
2625 * Charge the memory controller for page usage.
2627 * 0 if the charge was successful
2628 * < 0 if the cgroup is over its limit
2630 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
2631 gfp_t gfp_mask
, enum charge_type ctype
)
2633 struct mem_cgroup
*memcg
= NULL
;
2634 unsigned int nr_pages
= 1;
2635 struct page_cgroup
*pc
;
2639 if (PageTransHuge(page
)) {
2640 nr_pages
<<= compound_order(page
);
2641 VM_BUG_ON(!PageTransHuge(page
));
2643 * Never OOM-kill a process for a huge page. The
2644 * fault handler will fall back to regular pages.
2649 pc
= lookup_page_cgroup(page
);
2650 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, nr_pages
, &memcg
, oom
);
2653 __mem_cgroup_commit_charge(memcg
, page
, nr_pages
, pc
, ctype
);
2657 int mem_cgroup_newpage_charge(struct page
*page
,
2658 struct mm_struct
*mm
, gfp_t gfp_mask
)
2660 if (mem_cgroup_disabled())
2662 VM_BUG_ON(page_mapped(page
));
2663 VM_BUG_ON(page
->mapping
&& !PageAnon(page
));
2665 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2666 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2670 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2671 enum charge_type ctype
);
2674 __mem_cgroup_commit_charge_lrucare(struct page
*page
, struct mem_cgroup
*memcg
,
2675 enum charge_type ctype
)
2677 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
2678 struct zone
*zone
= page_zone(page
);
2679 unsigned long flags
;
2680 bool removed
= false;
2683 * In some case, SwapCache, FUSE(splice_buf->radixtree), the page
2684 * is already on LRU. It means the page may on some other page_cgroup's
2685 * LRU. Take care of it.
2687 spin_lock_irqsave(&zone
->lru_lock
, flags
);
2688 if (PageLRU(page
)) {
2689 del_page_from_lru_list(zone
, page
, page_lru(page
));
2693 __mem_cgroup_commit_charge(memcg
, page
, 1, pc
, ctype
);
2695 add_page_to_lru_list(zone
, page
, page_lru(page
));
2698 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
2702 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
2705 struct mem_cgroup
*memcg
= NULL
;
2706 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
2709 if (mem_cgroup_disabled())
2711 if (PageCompound(page
))
2716 if (!page_is_file_cache(page
))
2717 type
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
2719 if (!PageSwapCache(page
))
2720 ret
= mem_cgroup_charge_common(page
, mm
, gfp_mask
, type
);
2721 else { /* page is swapcache/shmem */
2722 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &memcg
);
2724 __mem_cgroup_commit_charge_swapin(page
, memcg
, type
);
2730 * While swap-in, try_charge -> commit or cancel, the page is locked.
2731 * And when try_charge() successfully returns, one refcnt to memcg without
2732 * struct page_cgroup is acquired. This refcnt will be consumed by
2733 * "commit()" or removed by "cancel()"
2735 int mem_cgroup_try_charge_swapin(struct mm_struct
*mm
,
2737 gfp_t mask
, struct mem_cgroup
**memcgp
)
2739 struct mem_cgroup
*memcg
;
2744 if (mem_cgroup_disabled())
2747 if (!do_swap_account
)
2750 * A racing thread's fault, or swapoff, may have already updated
2751 * the pte, and even removed page from swap cache: in those cases
2752 * do_swap_page()'s pte_same() test will fail; but there's also a
2753 * KSM case which does need to charge the page.
2755 if (!PageSwapCache(page
))
2757 memcg
= try_get_mem_cgroup_from_page(page
);
2761 ret
= __mem_cgroup_try_charge(NULL
, mask
, 1, memcgp
, true);
2762 css_put(&memcg
->css
);
2769 ret
= __mem_cgroup_try_charge(mm
, mask
, 1, memcgp
, true);
2776 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*memcg
,
2777 enum charge_type ctype
)
2779 if (mem_cgroup_disabled())
2783 cgroup_exclude_rmdir(&memcg
->css
);
2785 __mem_cgroup_commit_charge_lrucare(page
, memcg
, ctype
);
2787 * Now swap is on-memory. This means this page may be
2788 * counted both as mem and swap....double count.
2789 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2790 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2791 * may call delete_from_swap_cache() before reach here.
2793 if (do_swap_account
&& PageSwapCache(page
)) {
2794 swp_entry_t ent
= {.val
= page_private(page
)};
2795 struct mem_cgroup
*swap_memcg
;
2798 id
= swap_cgroup_record(ent
, 0);
2800 swap_memcg
= mem_cgroup_lookup(id
);
2803 * This recorded memcg can be obsolete one. So, avoid
2804 * calling css_tryget
2806 if (!mem_cgroup_is_root(swap_memcg
))
2807 res_counter_uncharge(&swap_memcg
->memsw
,
2809 mem_cgroup_swap_statistics(swap_memcg
, false);
2810 mem_cgroup_put(swap_memcg
);
2815 * At swapin, we may charge account against cgroup which has no tasks.
2816 * So, rmdir()->pre_destroy() can be called while we do this charge.
2817 * In that case, we need to call pre_destroy() again. check it here.
2819 cgroup_release_and_wakeup_rmdir(&memcg
->css
);
2822 void mem_cgroup_commit_charge_swapin(struct page
*page
,
2823 struct mem_cgroup
*memcg
)
2825 __mem_cgroup_commit_charge_swapin(page
, memcg
,
2826 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2829 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup
*memcg
)
2831 if (mem_cgroup_disabled())
2835 __mem_cgroup_cancel_charge(memcg
, 1);
2838 static void mem_cgroup_do_uncharge(struct mem_cgroup
*memcg
,
2839 unsigned int nr_pages
,
2840 const enum charge_type ctype
)
2842 struct memcg_batch_info
*batch
= NULL
;
2843 bool uncharge_memsw
= true;
2845 /* If swapout, usage of swap doesn't decrease */
2846 if (!do_swap_account
|| ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
)
2847 uncharge_memsw
= false;
2849 batch
= ¤t
->memcg_batch
;
2851 * In usual, we do css_get() when we remember memcg pointer.
2852 * But in this case, we keep res->usage until end of a series of
2853 * uncharges. Then, it's ok to ignore memcg's refcnt.
2856 batch
->memcg
= memcg
;
2858 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2859 * In those cases, all pages freed continuously can be expected to be in
2860 * the same cgroup and we have chance to coalesce uncharges.
2861 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2862 * because we want to do uncharge as soon as possible.
2865 if (!batch
->do_batch
|| test_thread_flag(TIF_MEMDIE
))
2866 goto direct_uncharge
;
2869 goto direct_uncharge
;
2872 * In typical case, batch->memcg == mem. This means we can
2873 * merge a series of uncharges to an uncharge of res_counter.
2874 * If not, we uncharge res_counter ony by one.
2876 if (batch
->memcg
!= memcg
)
2877 goto direct_uncharge
;
2878 /* remember freed charge and uncharge it later */
2881 batch
->memsw_nr_pages
++;
2884 res_counter_uncharge(&memcg
->res
, nr_pages
* PAGE_SIZE
);
2886 res_counter_uncharge(&memcg
->memsw
, nr_pages
* PAGE_SIZE
);
2887 if (unlikely(batch
->memcg
!= memcg
))
2888 memcg_oom_recover(memcg
);
2893 * uncharge if !page_mapped(page)
2895 static struct mem_cgroup
*
2896 __mem_cgroup_uncharge_common(struct page
*page
, enum charge_type ctype
)
2898 struct mem_cgroup
*memcg
= NULL
;
2899 unsigned int nr_pages
= 1;
2900 struct page_cgroup
*pc
;
2902 if (mem_cgroup_disabled())
2905 if (PageSwapCache(page
))
2908 if (PageTransHuge(page
)) {
2909 nr_pages
<<= compound_order(page
);
2910 VM_BUG_ON(!PageTransHuge(page
));
2913 * Check if our page_cgroup is valid
2915 pc
= lookup_page_cgroup(page
);
2916 if (unlikely(!PageCgroupUsed(pc
)))
2919 lock_page_cgroup(pc
);
2921 memcg
= pc
->mem_cgroup
;
2923 if (!PageCgroupUsed(pc
))
2927 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2928 case MEM_CGROUP_CHARGE_TYPE_DROP
:
2929 /* See mem_cgroup_prepare_migration() */
2930 if (page_mapped(page
) || PageCgroupMigration(pc
))
2933 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT
:
2934 if (!PageAnon(page
)) { /* Shared memory */
2935 if (page
->mapping
&& !page_is_file_cache(page
))
2937 } else if (page_mapped(page
)) /* Anon */
2944 mem_cgroup_charge_statistics(memcg
, PageCgroupCache(pc
), -nr_pages
);
2946 ClearPageCgroupUsed(pc
);
2948 * pc->mem_cgroup is not cleared here. It will be accessed when it's
2949 * freed from LRU. This is safe because uncharged page is expected not
2950 * to be reused (freed soon). Exception is SwapCache, it's handled by
2951 * special functions.
2954 unlock_page_cgroup(pc
);
2956 * even after unlock, we have memcg->res.usage here and this memcg
2957 * will never be freed.
2959 memcg_check_events(memcg
, page
);
2960 if (do_swap_account
&& ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
) {
2961 mem_cgroup_swap_statistics(memcg
, true);
2962 mem_cgroup_get(memcg
);
2964 if (!mem_cgroup_is_root(memcg
))
2965 mem_cgroup_do_uncharge(memcg
, nr_pages
, ctype
);
2970 unlock_page_cgroup(pc
);
2974 void mem_cgroup_uncharge_page(struct page
*page
)
2977 if (page_mapped(page
))
2979 VM_BUG_ON(page
->mapping
&& !PageAnon(page
));
2980 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2983 void mem_cgroup_uncharge_cache_page(struct page
*page
)
2985 VM_BUG_ON(page_mapped(page
));
2986 VM_BUG_ON(page
->mapping
);
2987 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_CACHE
);
2991 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
2992 * In that cases, pages are freed continuously and we can expect pages
2993 * are in the same memcg. All these calls itself limits the number of
2994 * pages freed at once, then uncharge_start/end() is called properly.
2995 * This may be called prural(2) times in a context,
2998 void mem_cgroup_uncharge_start(void)
3000 current
->memcg_batch
.do_batch
++;
3001 /* We can do nest. */
3002 if (current
->memcg_batch
.do_batch
== 1) {
3003 current
->memcg_batch
.memcg
= NULL
;
3004 current
->memcg_batch
.nr_pages
= 0;
3005 current
->memcg_batch
.memsw_nr_pages
= 0;
3009 void mem_cgroup_uncharge_end(void)
3011 struct memcg_batch_info
*batch
= ¤t
->memcg_batch
;
3013 if (!batch
->do_batch
)
3017 if (batch
->do_batch
) /* If stacked, do nothing. */
3023 * This "batch->memcg" is valid without any css_get/put etc...
3024 * bacause we hide charges behind us.
3026 if (batch
->nr_pages
)
3027 res_counter_uncharge(&batch
->memcg
->res
,
3028 batch
->nr_pages
* PAGE_SIZE
);
3029 if (batch
->memsw_nr_pages
)
3030 res_counter_uncharge(&batch
->memcg
->memsw
,
3031 batch
->memsw_nr_pages
* PAGE_SIZE
);
3032 memcg_oom_recover(batch
->memcg
);
3033 /* forget this pointer (for sanity check) */
3034 batch
->memcg
= NULL
;
3038 * A function for resetting pc->mem_cgroup for newly allocated pages.
3039 * This function should be called if the newpage will be added to LRU
3040 * before start accounting.
3042 void mem_cgroup_reset_owner(struct page
*newpage
)
3044 struct page_cgroup
*pc
;
3046 if (mem_cgroup_disabled())
3049 pc
= lookup_page_cgroup(newpage
);
3050 VM_BUG_ON(PageCgroupUsed(pc
));
3051 pc
->mem_cgroup
= root_mem_cgroup
;
3056 * called after __delete_from_swap_cache() and drop "page" account.
3057 * memcg information is recorded to swap_cgroup of "ent"
3060 mem_cgroup_uncharge_swapcache(struct page
*page
, swp_entry_t ent
, bool swapout
)
3062 struct mem_cgroup
*memcg
;
3063 int ctype
= MEM_CGROUP_CHARGE_TYPE_SWAPOUT
;
3065 if (!swapout
) /* this was a swap cache but the swap is unused ! */
3066 ctype
= MEM_CGROUP_CHARGE_TYPE_DROP
;
3068 memcg
= __mem_cgroup_uncharge_common(page
, ctype
);
3071 * record memcg information, if swapout && memcg != NULL,
3072 * mem_cgroup_get() was called in uncharge().
3074 if (do_swap_account
&& swapout
&& memcg
)
3075 swap_cgroup_record(ent
, css_id(&memcg
->css
));
3079 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3081 * called from swap_entry_free(). remove record in swap_cgroup and
3082 * uncharge "memsw" account.
3084 void mem_cgroup_uncharge_swap(swp_entry_t ent
)
3086 struct mem_cgroup
*memcg
;
3089 if (!do_swap_account
)
3092 id
= swap_cgroup_record(ent
, 0);
3094 memcg
= mem_cgroup_lookup(id
);
3097 * We uncharge this because swap is freed.
3098 * This memcg can be obsolete one. We avoid calling css_tryget
3100 if (!mem_cgroup_is_root(memcg
))
3101 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
3102 mem_cgroup_swap_statistics(memcg
, false);
3103 mem_cgroup_put(memcg
);
3109 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3110 * @entry: swap entry to be moved
3111 * @from: mem_cgroup which the entry is moved from
3112 * @to: mem_cgroup which the entry is moved to
3113 * @need_fixup: whether we should fixup res_counters and refcounts.
3115 * It succeeds only when the swap_cgroup's record for this entry is the same
3116 * as the mem_cgroup's id of @from.
3118 * Returns 0 on success, -EINVAL on failure.
3120 * The caller must have charged to @to, IOW, called res_counter_charge() about
3121 * both res and memsw, and called css_get().
3123 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
3124 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
3126 unsigned short old_id
, new_id
;
3128 old_id
= css_id(&from
->css
);
3129 new_id
= css_id(&to
->css
);
3131 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
3132 mem_cgroup_swap_statistics(from
, false);
3133 mem_cgroup_swap_statistics(to
, true);
3135 * This function is only called from task migration context now.
3136 * It postpones res_counter and refcount handling till the end
3137 * of task migration(mem_cgroup_clear_mc()) for performance
3138 * improvement. But we cannot postpone mem_cgroup_get(to)
3139 * because if the process that has been moved to @to does
3140 * swap-in, the refcount of @to might be decreased to 0.
3144 if (!mem_cgroup_is_root(from
))
3145 res_counter_uncharge(&from
->memsw
, PAGE_SIZE
);
3146 mem_cgroup_put(from
);
3148 * we charged both to->res and to->memsw, so we should
3151 if (!mem_cgroup_is_root(to
))
3152 res_counter_uncharge(&to
->res
, PAGE_SIZE
);
3159 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
3160 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
3167 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3170 int mem_cgroup_prepare_migration(struct page
*page
,
3171 struct page
*newpage
, struct mem_cgroup
**memcgp
, gfp_t gfp_mask
)
3173 struct mem_cgroup
*memcg
= NULL
;
3174 struct page_cgroup
*pc
;
3175 enum charge_type ctype
;
3180 VM_BUG_ON(PageTransHuge(page
));
3181 if (mem_cgroup_disabled())
3184 pc
= lookup_page_cgroup(page
);
3185 lock_page_cgroup(pc
);
3186 if (PageCgroupUsed(pc
)) {
3187 memcg
= pc
->mem_cgroup
;
3188 css_get(&memcg
->css
);
3190 * At migrating an anonymous page, its mapcount goes down
3191 * to 0 and uncharge() will be called. But, even if it's fully
3192 * unmapped, migration may fail and this page has to be
3193 * charged again. We set MIGRATION flag here and delay uncharge
3194 * until end_migration() is called
3196 * Corner Case Thinking
3198 * When the old page was mapped as Anon and it's unmap-and-freed
3199 * while migration was ongoing.
3200 * If unmap finds the old page, uncharge() of it will be delayed
3201 * until end_migration(). If unmap finds a new page, it's
3202 * uncharged when it make mapcount to be 1->0. If unmap code
3203 * finds swap_migration_entry, the new page will not be mapped
3204 * and end_migration() will find it(mapcount==0).
3207 * When the old page was mapped but migraion fails, the kernel
3208 * remaps it. A charge for it is kept by MIGRATION flag even
3209 * if mapcount goes down to 0. We can do remap successfully
3210 * without charging it again.
3213 * The "old" page is under lock_page() until the end of
3214 * migration, so, the old page itself will not be swapped-out.
3215 * If the new page is swapped out before end_migraton, our
3216 * hook to usual swap-out path will catch the event.
3219 SetPageCgroupMigration(pc
);
3221 unlock_page_cgroup(pc
);
3223 * If the page is not charged at this point,
3230 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, 1, memcgp
, false);
3231 css_put(&memcg
->css
);/* drop extra refcnt */
3233 if (PageAnon(page
)) {
3234 lock_page_cgroup(pc
);
3235 ClearPageCgroupMigration(pc
);
3236 unlock_page_cgroup(pc
);
3238 * The old page may be fully unmapped while we kept it.
3240 mem_cgroup_uncharge_page(page
);
3242 /* we'll need to revisit this error code (we have -EINTR) */
3246 * We charge new page before it's used/mapped. So, even if unlock_page()
3247 * is called before end_migration, we can catch all events on this new
3248 * page. In the case new page is migrated but not remapped, new page's
3249 * mapcount will be finally 0 and we call uncharge in end_migration().
3251 pc
= lookup_page_cgroup(newpage
);
3253 ctype
= MEM_CGROUP_CHARGE_TYPE_MAPPED
;
3254 else if (page_is_file_cache(page
))
3255 ctype
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3257 ctype
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
3258 __mem_cgroup_commit_charge(memcg
, page
, 1, pc
, ctype
);
3262 /* remove redundant charge if migration failed*/
3263 void mem_cgroup_end_migration(struct mem_cgroup
*memcg
,
3264 struct page
*oldpage
, struct page
*newpage
, bool migration_ok
)
3266 struct page
*used
, *unused
;
3267 struct page_cgroup
*pc
;
3271 /* blocks rmdir() */
3272 cgroup_exclude_rmdir(&memcg
->css
);
3273 if (!migration_ok
) {
3281 * We disallowed uncharge of pages under migration because mapcount
3282 * of the page goes down to zero, temporarly.
3283 * Clear the flag and check the page should be charged.
3285 pc
= lookup_page_cgroup(oldpage
);
3286 lock_page_cgroup(pc
);
3287 ClearPageCgroupMigration(pc
);
3288 unlock_page_cgroup(pc
);
3290 __mem_cgroup_uncharge_common(unused
, MEM_CGROUP_CHARGE_TYPE_FORCE
);
3293 * If a page is a file cache, radix-tree replacement is very atomic
3294 * and we can skip this check. When it was an Anon page, its mapcount
3295 * goes down to 0. But because we added MIGRATION flage, it's not
3296 * uncharged yet. There are several case but page->mapcount check
3297 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3298 * check. (see prepare_charge() also)
3301 mem_cgroup_uncharge_page(used
);
3303 * At migration, we may charge account against cgroup which has no
3305 * So, rmdir()->pre_destroy() can be called while we do this charge.
3306 * In that case, we need to call pre_destroy() again. check it here.
3308 cgroup_release_and_wakeup_rmdir(&memcg
->css
);
3312 * At replace page cache, newpage is not under any memcg but it's on
3313 * LRU. So, this function doesn't touch res_counter but handles LRU
3314 * in correct way. Both pages are locked so we cannot race with uncharge.
3316 void mem_cgroup_replace_page_cache(struct page
*oldpage
,
3317 struct page
*newpage
)
3319 struct mem_cgroup
*memcg
;
3320 struct page_cgroup
*pc
;
3321 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3323 if (mem_cgroup_disabled())
3326 pc
= lookup_page_cgroup(oldpage
);
3327 /* fix accounting on old pages */
3328 lock_page_cgroup(pc
);
3329 memcg
= pc
->mem_cgroup
;
3330 mem_cgroup_charge_statistics(memcg
, PageCgroupCache(pc
), -1);
3331 ClearPageCgroupUsed(pc
);
3332 unlock_page_cgroup(pc
);
3334 if (PageSwapBacked(oldpage
))
3335 type
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
3338 * Even if newpage->mapping was NULL before starting replacement,
3339 * the newpage may be on LRU(or pagevec for LRU) already. We lock
3340 * LRU while we overwrite pc->mem_cgroup.
3342 __mem_cgroup_commit_charge_lrucare(newpage
, memcg
, type
);
3345 #ifdef CONFIG_DEBUG_VM
3346 static struct page_cgroup
*lookup_page_cgroup_used(struct page
*page
)
3348 struct page_cgroup
*pc
;
3350 pc
= lookup_page_cgroup(page
);
3352 * Can be NULL while feeding pages into the page allocator for
3353 * the first time, i.e. during boot or memory hotplug;
3354 * or when mem_cgroup_disabled().
3356 if (likely(pc
) && PageCgroupUsed(pc
))
3361 bool mem_cgroup_bad_page_check(struct page
*page
)
3363 if (mem_cgroup_disabled())
3366 return lookup_page_cgroup_used(page
) != NULL
;
3369 void mem_cgroup_print_bad_page(struct page
*page
)
3371 struct page_cgroup
*pc
;
3373 pc
= lookup_page_cgroup_used(page
);
3378 printk(KERN_ALERT
"pc:%p pc->flags:%lx pc->mem_cgroup:%p",
3379 pc
, pc
->flags
, pc
->mem_cgroup
);
3381 path
= kmalloc(PATH_MAX
, GFP_KERNEL
);
3384 ret
= cgroup_path(pc
->mem_cgroup
->css
.cgroup
,
3389 printk(KERN_CONT
"(%s)\n",
3390 (ret
< 0) ? "cannot get the path" : path
);
3396 static DEFINE_MUTEX(set_limit_mutex
);
3398 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
3399 unsigned long long val
)
3402 u64 memswlimit
, memlimit
;
3404 int children
= mem_cgroup_count_children(memcg
);
3405 u64 curusage
, oldusage
;
3409 * For keeping hierarchical_reclaim simple, how long we should retry
3410 * is depends on callers. We set our retry-count to be function
3411 * of # of children which we should visit in this loop.
3413 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
* children
;
3415 oldusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3418 while (retry_count
) {
3419 if (signal_pending(current
)) {
3424 * Rather than hide all in some function, I do this in
3425 * open coded manner. You see what this really does.
3426 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
3428 mutex_lock(&set_limit_mutex
);
3429 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3430 if (memswlimit
< val
) {
3432 mutex_unlock(&set_limit_mutex
);
3436 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3440 ret
= res_counter_set_limit(&memcg
->res
, val
);
3442 if (memswlimit
== val
)
3443 memcg
->memsw_is_minimum
= true;
3445 memcg
->memsw_is_minimum
= false;
3447 mutex_unlock(&set_limit_mutex
);
3452 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3453 MEM_CGROUP_RECLAIM_SHRINK
);
3454 curusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3455 /* Usage is reduced ? */
3456 if (curusage
>= oldusage
)
3459 oldusage
= curusage
;
3461 if (!ret
&& enlarge
)
3462 memcg_oom_recover(memcg
);
3467 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
3468 unsigned long long val
)
3471 u64 memlimit
, memswlimit
, oldusage
, curusage
;
3472 int children
= mem_cgroup_count_children(memcg
);
3476 /* see mem_cgroup_resize_res_limit */
3477 retry_count
= children
* MEM_CGROUP_RECLAIM_RETRIES
;
3478 oldusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3479 while (retry_count
) {
3480 if (signal_pending(current
)) {
3485 * Rather than hide all in some function, I do this in
3486 * open coded manner. You see what this really does.
3487 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
3489 mutex_lock(&set_limit_mutex
);
3490 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3491 if (memlimit
> val
) {
3493 mutex_unlock(&set_limit_mutex
);
3496 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3497 if (memswlimit
< val
)
3499 ret
= res_counter_set_limit(&memcg
->memsw
, val
);
3501 if (memlimit
== val
)
3502 memcg
->memsw_is_minimum
= true;
3504 memcg
->memsw_is_minimum
= false;
3506 mutex_unlock(&set_limit_mutex
);
3511 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3512 MEM_CGROUP_RECLAIM_NOSWAP
|
3513 MEM_CGROUP_RECLAIM_SHRINK
);
3514 curusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3515 /* Usage is reduced ? */
3516 if (curusage
>= oldusage
)
3519 oldusage
= curusage
;
3521 if (!ret
&& enlarge
)
3522 memcg_oom_recover(memcg
);
3526 unsigned long mem_cgroup_soft_limit_reclaim(struct zone
*zone
, int order
,
3528 unsigned long *total_scanned
)
3530 unsigned long nr_reclaimed
= 0;
3531 struct mem_cgroup_per_zone
*mz
, *next_mz
= NULL
;
3532 unsigned long reclaimed
;
3534 struct mem_cgroup_tree_per_zone
*mctz
;
3535 unsigned long long excess
;
3536 unsigned long nr_scanned
;
3541 mctz
= soft_limit_tree_node_zone(zone_to_nid(zone
), zone_idx(zone
));
3543 * This loop can run a while, specially if mem_cgroup's continuously
3544 * keep exceeding their soft limit and putting the system under
3551 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
3556 reclaimed
= mem_cgroup_soft_reclaim(mz
->mem
, zone
,
3557 gfp_mask
, &nr_scanned
);
3558 nr_reclaimed
+= reclaimed
;
3559 *total_scanned
+= nr_scanned
;
3560 spin_lock(&mctz
->lock
);
3563 * If we failed to reclaim anything from this memory cgroup
3564 * it is time to move on to the next cgroup
3570 * Loop until we find yet another one.
3572 * By the time we get the soft_limit lock
3573 * again, someone might have aded the
3574 * group back on the RB tree. Iterate to
3575 * make sure we get a different mem.
3576 * mem_cgroup_largest_soft_limit_node returns
3577 * NULL if no other cgroup is present on
3581 __mem_cgroup_largest_soft_limit_node(mctz
);
3583 css_put(&next_mz
->mem
->css
);
3584 else /* next_mz == NULL or other memcg */
3588 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
3589 excess
= res_counter_soft_limit_excess(&mz
->mem
->res
);
3591 * One school of thought says that we should not add
3592 * back the node to the tree if reclaim returns 0.
3593 * But our reclaim could return 0, simply because due
3594 * to priority we are exposing a smaller subset of
3595 * memory to reclaim from. Consider this as a longer
3598 /* If excess == 0, no tree ops */
3599 __mem_cgroup_insert_exceeded(mz
->mem
, mz
, mctz
, excess
);
3600 spin_unlock(&mctz
->lock
);
3601 css_put(&mz
->mem
->css
);
3604 * Could not reclaim anything and there are no more
3605 * mem cgroups to try or we seem to be looping without
3606 * reclaiming anything.
3608 if (!nr_reclaimed
&&
3610 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
3612 } while (!nr_reclaimed
);
3614 css_put(&next_mz
->mem
->css
);
3615 return nr_reclaimed
;
3619 * This routine traverse page_cgroup in given list and drop them all.
3620 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3622 static int mem_cgroup_force_empty_list(struct mem_cgroup
*memcg
,
3623 int node
, int zid
, enum lru_list lru
)
3625 struct mem_cgroup_per_zone
*mz
;
3626 unsigned long flags
, loop
;
3627 struct list_head
*list
;
3632 zone
= &NODE_DATA(node
)->node_zones
[zid
];
3633 mz
= mem_cgroup_zoneinfo(memcg
, node
, zid
);
3634 list
= &mz
->lruvec
.lists
[lru
];
3636 loop
= MEM_CGROUP_ZSTAT(mz
, lru
);
3637 /* give some margin against EBUSY etc...*/
3641 struct page_cgroup
*pc
;
3645 spin_lock_irqsave(&zone
->lru_lock
, flags
);
3646 if (list_empty(list
)) {
3647 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3650 page
= list_entry(list
->prev
, struct page
, lru
);
3652 list_move(&page
->lru
, list
);
3654 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3657 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3659 pc
= lookup_page_cgroup(page
);
3661 ret
= mem_cgroup_move_parent(page
, pc
, memcg
, GFP_KERNEL
);
3662 if (ret
== -ENOMEM
|| ret
== -EINTR
)
3665 if (ret
== -EBUSY
|| ret
== -EINVAL
) {
3666 /* found lock contention or "pc" is obsolete. */
3673 if (!ret
&& !list_empty(list
))
3679 * make mem_cgroup's charge to be 0 if there is no task.
3680 * This enables deleting this mem_cgroup.
3682 static int mem_cgroup_force_empty(struct mem_cgroup
*memcg
, bool free_all
)
3685 int node
, zid
, shrink
;
3686 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
3687 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
3689 css_get(&memcg
->css
);
3692 /* should free all ? */
3698 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
))
3701 if (signal_pending(current
))
3703 /* This is for making all *used* pages to be on LRU. */
3704 lru_add_drain_all();
3705 drain_all_stock_sync(memcg
);
3707 mem_cgroup_start_move(memcg
);
3708 for_each_node_state(node
, N_HIGH_MEMORY
) {
3709 for (zid
= 0; !ret
&& zid
< MAX_NR_ZONES
; zid
++) {
3712 ret
= mem_cgroup_force_empty_list(memcg
,
3721 mem_cgroup_end_move(memcg
);
3722 memcg_oom_recover(memcg
);
3723 /* it seems parent cgroup doesn't have enough mem */
3727 /* "ret" should also be checked to ensure all lists are empty. */
3728 } while (memcg
->res
.usage
> 0 || ret
);
3730 css_put(&memcg
->css
);
3734 /* returns EBUSY if there is a task or if we come here twice. */
3735 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
) || shrink
) {
3739 /* we call try-to-free pages for make this cgroup empty */
3740 lru_add_drain_all();
3741 /* try to free all pages in this cgroup */
3743 while (nr_retries
&& memcg
->res
.usage
> 0) {
3746 if (signal_pending(current
)) {
3750 progress
= try_to_free_mem_cgroup_pages(memcg
, GFP_KERNEL
,
3754 /* maybe some writeback is necessary */
3755 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3760 /* try move_account...there may be some *locked* pages. */
3764 int mem_cgroup_force_empty_write(struct cgroup
*cont
, unsigned int event
)
3766 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont
), true);
3770 static u64
mem_cgroup_hierarchy_read(struct cgroup
*cont
, struct cftype
*cft
)
3772 return mem_cgroup_from_cont(cont
)->use_hierarchy
;
3775 static int mem_cgroup_hierarchy_write(struct cgroup
*cont
, struct cftype
*cft
,
3779 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3780 struct cgroup
*parent
= cont
->parent
;
3781 struct mem_cgroup
*parent_memcg
= NULL
;
3784 parent_memcg
= mem_cgroup_from_cont(parent
);
3788 * If parent's use_hierarchy is set, we can't make any modifications
3789 * in the child subtrees. If it is unset, then the change can
3790 * occur, provided the current cgroup has no children.
3792 * For the root cgroup, parent_mem is NULL, we allow value to be
3793 * set if there are no children.
3795 if ((!parent_memcg
|| !parent_memcg
->use_hierarchy
) &&
3796 (val
== 1 || val
== 0)) {
3797 if (list_empty(&cont
->children
))
3798 memcg
->use_hierarchy
= val
;
3809 static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup
*memcg
,
3810 enum mem_cgroup_stat_index idx
)
3812 struct mem_cgroup
*iter
;
3815 /* Per-cpu values can be negative, use a signed accumulator */
3816 for_each_mem_cgroup_tree(iter
, memcg
)
3817 val
+= mem_cgroup_read_stat(iter
, idx
);
3819 if (val
< 0) /* race ? */
3824 static inline u64
mem_cgroup_usage(struct mem_cgroup
*memcg
, bool swap
)
3828 if (!mem_cgroup_is_root(memcg
)) {
3830 return res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3832 return res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3835 val
= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_CACHE
);
3836 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_RSS
);
3839 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_SWAPOUT
);
3841 return val
<< PAGE_SHIFT
;
3844 static u64
mem_cgroup_read(struct cgroup
*cont
, struct cftype
*cft
)
3846 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3850 type
= MEMFILE_TYPE(cft
->private);
3851 name
= MEMFILE_ATTR(cft
->private);
3854 if (name
== RES_USAGE
)
3855 val
= mem_cgroup_usage(memcg
, false);
3857 val
= res_counter_read_u64(&memcg
->res
, name
);
3860 if (name
== RES_USAGE
)
3861 val
= mem_cgroup_usage(memcg
, true);
3863 val
= res_counter_read_u64(&memcg
->memsw
, name
);
3872 * The user of this function is...
3875 static int mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
3878 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3880 unsigned long long val
;
3883 type
= MEMFILE_TYPE(cft
->private);
3884 name
= MEMFILE_ATTR(cft
->private);
3887 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
3891 /* This function does all necessary parse...reuse it */
3892 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3896 ret
= mem_cgroup_resize_limit(memcg
, val
);
3898 ret
= mem_cgroup_resize_memsw_limit(memcg
, val
);
3900 case RES_SOFT_LIMIT
:
3901 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3905 * For memsw, soft limits are hard to implement in terms
3906 * of semantics, for now, we support soft limits for
3907 * control without swap
3910 ret
= res_counter_set_soft_limit(&memcg
->res
, val
);
3915 ret
= -EINVAL
; /* should be BUG() ? */
3921 static void memcg_get_hierarchical_limit(struct mem_cgroup
*memcg
,
3922 unsigned long long *mem_limit
, unsigned long long *memsw_limit
)
3924 struct cgroup
*cgroup
;
3925 unsigned long long min_limit
, min_memsw_limit
, tmp
;
3927 min_limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3928 min_memsw_limit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3929 cgroup
= memcg
->css
.cgroup
;
3930 if (!memcg
->use_hierarchy
)
3933 while (cgroup
->parent
) {
3934 cgroup
= cgroup
->parent
;
3935 memcg
= mem_cgroup_from_cont(cgroup
);
3936 if (!memcg
->use_hierarchy
)
3938 tmp
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3939 min_limit
= min(min_limit
, tmp
);
3940 tmp
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3941 min_memsw_limit
= min(min_memsw_limit
, tmp
);
3944 *mem_limit
= min_limit
;
3945 *memsw_limit
= min_memsw_limit
;
3949 static int mem_cgroup_reset(struct cgroup
*cont
, unsigned int event
)
3951 struct mem_cgroup
*memcg
;
3954 memcg
= mem_cgroup_from_cont(cont
);
3955 type
= MEMFILE_TYPE(event
);
3956 name
= MEMFILE_ATTR(event
);
3960 res_counter_reset_max(&memcg
->res
);
3962 res_counter_reset_max(&memcg
->memsw
);
3966 res_counter_reset_failcnt(&memcg
->res
);
3968 res_counter_reset_failcnt(&memcg
->memsw
);
3975 static u64
mem_cgroup_move_charge_read(struct cgroup
*cgrp
,
3978 return mem_cgroup_from_cont(cgrp
)->move_charge_at_immigrate
;
3982 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
3983 struct cftype
*cft
, u64 val
)
3985 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3987 if (val
>= (1 << NR_MOVE_TYPE
))
3990 * We check this value several times in both in can_attach() and
3991 * attach(), so we need cgroup lock to prevent this value from being
3995 memcg
->move_charge_at_immigrate
= val
;
4001 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4002 struct cftype
*cft
, u64 val
)
4009 /* For read statistics */
4027 struct mcs_total_stat
{
4028 s64 stat
[NR_MCS_STAT
];
4034 } memcg_stat_strings
[NR_MCS_STAT
] = {
4035 {"cache", "total_cache"},
4036 {"rss", "total_rss"},
4037 {"mapped_file", "total_mapped_file"},
4038 {"pgpgin", "total_pgpgin"},
4039 {"pgpgout", "total_pgpgout"},
4040 {"swap", "total_swap"},
4041 {"pgfault", "total_pgfault"},
4042 {"pgmajfault", "total_pgmajfault"},
4043 {"inactive_anon", "total_inactive_anon"},
4044 {"active_anon", "total_active_anon"},
4045 {"inactive_file", "total_inactive_file"},
4046 {"active_file", "total_active_file"},
4047 {"unevictable", "total_unevictable"}
4052 mem_cgroup_get_local_stat(struct mem_cgroup
*memcg
, struct mcs_total_stat
*s
)
4057 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_CACHE
);
4058 s
->stat
[MCS_CACHE
] += val
* PAGE_SIZE
;
4059 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_RSS
);
4060 s
->stat
[MCS_RSS
] += val
* PAGE_SIZE
;
4061 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_FILE_MAPPED
);
4062 s
->stat
[MCS_FILE_MAPPED
] += val
* PAGE_SIZE
;
4063 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGPGIN
);
4064 s
->stat
[MCS_PGPGIN
] += val
;
4065 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGPGOUT
);
4066 s
->stat
[MCS_PGPGOUT
] += val
;
4067 if (do_swap_account
) {
4068 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_SWAPOUT
);
4069 s
->stat
[MCS_SWAP
] += val
* PAGE_SIZE
;
4071 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGFAULT
);
4072 s
->stat
[MCS_PGFAULT
] += val
;
4073 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGMAJFAULT
);
4074 s
->stat
[MCS_PGMAJFAULT
] += val
;
4077 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_INACTIVE_ANON
));
4078 s
->stat
[MCS_INACTIVE_ANON
] += val
* PAGE_SIZE
;
4079 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_ACTIVE_ANON
));
4080 s
->stat
[MCS_ACTIVE_ANON
] += val
* PAGE_SIZE
;
4081 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_INACTIVE_FILE
));
4082 s
->stat
[MCS_INACTIVE_FILE
] += val
* PAGE_SIZE
;
4083 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_ACTIVE_FILE
));
4084 s
->stat
[MCS_ACTIVE_FILE
] += val
* PAGE_SIZE
;
4085 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_UNEVICTABLE
));
4086 s
->stat
[MCS_UNEVICTABLE
] += val
* PAGE_SIZE
;
4090 mem_cgroup_get_total_stat(struct mem_cgroup
*memcg
, struct mcs_total_stat
*s
)
4092 struct mem_cgroup
*iter
;
4094 for_each_mem_cgroup_tree(iter
, memcg
)
4095 mem_cgroup_get_local_stat(iter
, s
);
4099 static int mem_control_numa_stat_show(struct seq_file
*m
, void *arg
)
4102 unsigned long total_nr
, file_nr
, anon_nr
, unevictable_nr
;
4103 unsigned long node_nr
;
4104 struct cgroup
*cont
= m
->private;
4105 struct mem_cgroup
*mem_cont
= mem_cgroup_from_cont(cont
);
4107 total_nr
= mem_cgroup_nr_lru_pages(mem_cont
, LRU_ALL
);
4108 seq_printf(m
, "total=%lu", total_nr
);
4109 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4110 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
, LRU_ALL
);
4111 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4115 file_nr
= mem_cgroup_nr_lru_pages(mem_cont
, LRU_ALL_FILE
);
4116 seq_printf(m
, "file=%lu", file_nr
);
4117 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4118 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
,
4120 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4124 anon_nr
= mem_cgroup_nr_lru_pages(mem_cont
, LRU_ALL_ANON
);
4125 seq_printf(m
, "anon=%lu", anon_nr
);
4126 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4127 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
,
4129 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4133 unevictable_nr
= mem_cgroup_nr_lru_pages(mem_cont
, BIT(LRU_UNEVICTABLE
));
4134 seq_printf(m
, "unevictable=%lu", unevictable_nr
);
4135 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4136 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
,
4137 BIT(LRU_UNEVICTABLE
));
4138 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4143 #endif /* CONFIG_NUMA */
4145 static int mem_control_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
4146 struct cgroup_map_cb
*cb
)
4148 struct mem_cgroup
*mem_cont
= mem_cgroup_from_cont(cont
);
4149 struct mcs_total_stat mystat
;
4152 memset(&mystat
, 0, sizeof(mystat
));
4153 mem_cgroup_get_local_stat(mem_cont
, &mystat
);
4156 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
4157 if (i
== MCS_SWAP
&& !do_swap_account
)
4159 cb
->fill(cb
, memcg_stat_strings
[i
].local_name
, mystat
.stat
[i
]);
4162 /* Hierarchical information */
4164 unsigned long long limit
, memsw_limit
;
4165 memcg_get_hierarchical_limit(mem_cont
, &limit
, &memsw_limit
);
4166 cb
->fill(cb
, "hierarchical_memory_limit", limit
);
4167 if (do_swap_account
)
4168 cb
->fill(cb
, "hierarchical_memsw_limit", memsw_limit
);
4171 memset(&mystat
, 0, sizeof(mystat
));
4172 mem_cgroup_get_total_stat(mem_cont
, &mystat
);
4173 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
4174 if (i
== MCS_SWAP
&& !do_swap_account
)
4176 cb
->fill(cb
, memcg_stat_strings
[i
].total_name
, mystat
.stat
[i
]);
4179 #ifdef CONFIG_DEBUG_VM
4182 struct mem_cgroup_per_zone
*mz
;
4183 unsigned long recent_rotated
[2] = {0, 0};
4184 unsigned long recent_scanned
[2] = {0, 0};
4186 for_each_online_node(nid
)
4187 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
4188 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
4190 recent_rotated
[0] +=
4191 mz
->reclaim_stat
.recent_rotated
[0];
4192 recent_rotated
[1] +=
4193 mz
->reclaim_stat
.recent_rotated
[1];
4194 recent_scanned
[0] +=
4195 mz
->reclaim_stat
.recent_scanned
[0];
4196 recent_scanned
[1] +=
4197 mz
->reclaim_stat
.recent_scanned
[1];
4199 cb
->fill(cb
, "recent_rotated_anon", recent_rotated
[0]);
4200 cb
->fill(cb
, "recent_rotated_file", recent_rotated
[1]);
4201 cb
->fill(cb
, "recent_scanned_anon", recent_scanned
[0]);
4202 cb
->fill(cb
, "recent_scanned_file", recent_scanned
[1]);
4209 static u64
mem_cgroup_swappiness_read(struct cgroup
*cgrp
, struct cftype
*cft
)
4211 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4213 return mem_cgroup_swappiness(memcg
);
4216 static int mem_cgroup_swappiness_write(struct cgroup
*cgrp
, struct cftype
*cft
,
4219 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4220 struct mem_cgroup
*parent
;
4225 if (cgrp
->parent
== NULL
)
4228 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4232 /* If under hierarchy, only empty-root can set this value */
4233 if ((parent
->use_hierarchy
) ||
4234 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4239 memcg
->swappiness
= val
;
4246 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
4248 struct mem_cgroup_threshold_ary
*t
;
4254 t
= rcu_dereference(memcg
->thresholds
.primary
);
4256 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
4261 usage
= mem_cgroup_usage(memcg
, swap
);
4264 * current_threshold points to threshold just below usage.
4265 * If it's not true, a threshold was crossed after last
4266 * call of __mem_cgroup_threshold().
4268 i
= t
->current_threshold
;
4271 * Iterate backward over array of thresholds starting from
4272 * current_threshold and check if a threshold is crossed.
4273 * If none of thresholds below usage is crossed, we read
4274 * only one element of the array here.
4276 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
4277 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4279 /* i = current_threshold + 1 */
4283 * Iterate forward over array of thresholds starting from
4284 * current_threshold+1 and check if a threshold is crossed.
4285 * If none of thresholds above usage is crossed, we read
4286 * only one element of the array here.
4288 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
4289 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4291 /* Update current_threshold */
4292 t
->current_threshold
= i
- 1;
4297 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
4300 __mem_cgroup_threshold(memcg
, false);
4301 if (do_swap_account
)
4302 __mem_cgroup_threshold(memcg
, true);
4304 memcg
= parent_mem_cgroup(memcg
);
4308 static int compare_thresholds(const void *a
, const void *b
)
4310 const struct mem_cgroup_threshold
*_a
= a
;
4311 const struct mem_cgroup_threshold
*_b
= b
;
4313 return _a
->threshold
- _b
->threshold
;
4316 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*memcg
)
4318 struct mem_cgroup_eventfd_list
*ev
;
4320 list_for_each_entry(ev
, &memcg
->oom_notify
, list
)
4321 eventfd_signal(ev
->eventfd
, 1);
4325 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
)
4327 struct mem_cgroup
*iter
;
4329 for_each_mem_cgroup_tree(iter
, memcg
)
4330 mem_cgroup_oom_notify_cb(iter
);
4333 static int mem_cgroup_usage_register_event(struct cgroup
*cgrp
,
4334 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4336 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4337 struct mem_cgroup_thresholds
*thresholds
;
4338 struct mem_cgroup_threshold_ary
*new;
4339 int type
= MEMFILE_TYPE(cft
->private);
4340 u64 threshold
, usage
;
4343 ret
= res_counter_memparse_write_strategy(args
, &threshold
);
4347 mutex_lock(&memcg
->thresholds_lock
);
4350 thresholds
= &memcg
->thresholds
;
4351 else if (type
== _MEMSWAP
)
4352 thresholds
= &memcg
->memsw_thresholds
;
4356 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4358 /* Check if a threshold crossed before adding a new one */
4359 if (thresholds
->primary
)
4360 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4362 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
4364 /* Allocate memory for new array of thresholds */
4365 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
4373 /* Copy thresholds (if any) to new array */
4374 if (thresholds
->primary
) {
4375 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
4376 sizeof(struct mem_cgroup_threshold
));
4379 /* Add new threshold */
4380 new->entries
[size
- 1].eventfd
= eventfd
;
4381 new->entries
[size
- 1].threshold
= threshold
;
4383 /* Sort thresholds. Registering of new threshold isn't time-critical */
4384 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
4385 compare_thresholds
, NULL
);
4387 /* Find current threshold */
4388 new->current_threshold
= -1;
4389 for (i
= 0; i
< size
; i
++) {
4390 if (new->entries
[i
].threshold
< usage
) {
4392 * new->current_threshold will not be used until
4393 * rcu_assign_pointer(), so it's safe to increment
4396 ++new->current_threshold
;
4400 /* Free old spare buffer and save old primary buffer as spare */
4401 kfree(thresholds
->spare
);
4402 thresholds
->spare
= thresholds
->primary
;
4404 rcu_assign_pointer(thresholds
->primary
, new);
4406 /* To be sure that nobody uses thresholds */
4410 mutex_unlock(&memcg
->thresholds_lock
);
4415 static void mem_cgroup_usage_unregister_event(struct cgroup
*cgrp
,
4416 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4418 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4419 struct mem_cgroup_thresholds
*thresholds
;
4420 struct mem_cgroup_threshold_ary
*new;
4421 int type
= MEMFILE_TYPE(cft
->private);
4425 mutex_lock(&memcg
->thresholds_lock
);
4427 thresholds
= &memcg
->thresholds
;
4428 else if (type
== _MEMSWAP
)
4429 thresholds
= &memcg
->memsw_thresholds
;
4434 * Something went wrong if we trying to unregister a threshold
4435 * if we don't have thresholds
4437 BUG_ON(!thresholds
);
4439 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4441 /* Check if a threshold crossed before removing */
4442 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4444 /* Calculate new number of threshold */
4446 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
4447 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
4451 new = thresholds
->spare
;
4453 /* Set thresholds array to NULL if we don't have thresholds */
4462 /* Copy thresholds and find current threshold */
4463 new->current_threshold
= -1;
4464 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
4465 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
4468 new->entries
[j
] = thresholds
->primary
->entries
[i
];
4469 if (new->entries
[j
].threshold
< usage
) {
4471 * new->current_threshold will not be used
4472 * until rcu_assign_pointer(), so it's safe to increment
4475 ++new->current_threshold
;
4481 /* Swap primary and spare array */
4482 thresholds
->spare
= thresholds
->primary
;
4483 rcu_assign_pointer(thresholds
->primary
, new);
4485 /* To be sure that nobody uses thresholds */
4488 mutex_unlock(&memcg
->thresholds_lock
);
4491 static int mem_cgroup_oom_register_event(struct cgroup
*cgrp
,
4492 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4494 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4495 struct mem_cgroup_eventfd_list
*event
;
4496 int type
= MEMFILE_TYPE(cft
->private);
4498 BUG_ON(type
!= _OOM_TYPE
);
4499 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
4503 spin_lock(&memcg_oom_lock
);
4505 event
->eventfd
= eventfd
;
4506 list_add(&event
->list
, &memcg
->oom_notify
);
4508 /* already in OOM ? */
4509 if (atomic_read(&memcg
->under_oom
))
4510 eventfd_signal(eventfd
, 1);
4511 spin_unlock(&memcg_oom_lock
);
4516 static void mem_cgroup_oom_unregister_event(struct cgroup
*cgrp
,
4517 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4519 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4520 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
4521 int type
= MEMFILE_TYPE(cft
->private);
4523 BUG_ON(type
!= _OOM_TYPE
);
4525 spin_lock(&memcg_oom_lock
);
4527 list_for_each_entry_safe(ev
, tmp
, &memcg
->oom_notify
, list
) {
4528 if (ev
->eventfd
== eventfd
) {
4529 list_del(&ev
->list
);
4534 spin_unlock(&memcg_oom_lock
);
4537 static int mem_cgroup_oom_control_read(struct cgroup
*cgrp
,
4538 struct cftype
*cft
, struct cgroup_map_cb
*cb
)
4540 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4542 cb
->fill(cb
, "oom_kill_disable", memcg
->oom_kill_disable
);
4544 if (atomic_read(&memcg
->under_oom
))
4545 cb
->fill(cb
, "under_oom", 1);
4547 cb
->fill(cb
, "under_oom", 0);
4551 static int mem_cgroup_oom_control_write(struct cgroup
*cgrp
,
4552 struct cftype
*cft
, u64 val
)
4554 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4555 struct mem_cgroup
*parent
;
4557 /* cannot set to root cgroup and only 0 and 1 are allowed */
4558 if (!cgrp
->parent
|| !((val
== 0) || (val
== 1)))
4561 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4564 /* oom-kill-disable is a flag for subhierarchy. */
4565 if ((parent
->use_hierarchy
) ||
4566 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4570 memcg
->oom_kill_disable
= val
;
4572 memcg_oom_recover(memcg
);
4578 static const struct file_operations mem_control_numa_stat_file_operations
= {
4580 .llseek
= seq_lseek
,
4581 .release
= single_release
,
4584 static int mem_control_numa_stat_open(struct inode
*unused
, struct file
*file
)
4586 struct cgroup
*cont
= file
->f_dentry
->d_parent
->d_fsdata
;
4588 file
->f_op
= &mem_control_numa_stat_file_operations
;
4589 return single_open(file
, mem_control_numa_stat_show
, cont
);
4591 #endif /* CONFIG_NUMA */
4593 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
4594 static int register_kmem_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4597 * Part of this would be better living in a separate allocation
4598 * function, leaving us with just the cgroup tree population work.
4599 * We, however, depend on state such as network's proto_list that
4600 * is only initialized after cgroup creation. I found the less
4601 * cumbersome way to deal with it to defer it all to populate time
4603 return mem_cgroup_sockets_init(cont
, ss
);
4606 static void kmem_cgroup_destroy(struct cgroup_subsys
*ss
,
4607 struct cgroup
*cont
)
4609 mem_cgroup_sockets_destroy(cont
, ss
);
4612 static int register_kmem_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4617 static void kmem_cgroup_destroy(struct cgroup_subsys
*ss
,
4618 struct cgroup
*cont
)
4623 static struct cftype mem_cgroup_files
[] = {
4625 .name
= "usage_in_bytes",
4626 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
4627 .read_u64
= mem_cgroup_read
,
4628 .register_event
= mem_cgroup_usage_register_event
,
4629 .unregister_event
= mem_cgroup_usage_unregister_event
,
4632 .name
= "max_usage_in_bytes",
4633 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
4634 .trigger
= mem_cgroup_reset
,
4635 .read_u64
= mem_cgroup_read
,
4638 .name
= "limit_in_bytes",
4639 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
4640 .write_string
= mem_cgroup_write
,
4641 .read_u64
= mem_cgroup_read
,
4644 .name
= "soft_limit_in_bytes",
4645 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
4646 .write_string
= mem_cgroup_write
,
4647 .read_u64
= mem_cgroup_read
,
4651 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
4652 .trigger
= mem_cgroup_reset
,
4653 .read_u64
= mem_cgroup_read
,
4657 .read_map
= mem_control_stat_show
,
4660 .name
= "force_empty",
4661 .trigger
= mem_cgroup_force_empty_write
,
4664 .name
= "use_hierarchy",
4665 .write_u64
= mem_cgroup_hierarchy_write
,
4666 .read_u64
= mem_cgroup_hierarchy_read
,
4669 .name
= "swappiness",
4670 .read_u64
= mem_cgroup_swappiness_read
,
4671 .write_u64
= mem_cgroup_swappiness_write
,
4674 .name
= "move_charge_at_immigrate",
4675 .read_u64
= mem_cgroup_move_charge_read
,
4676 .write_u64
= mem_cgroup_move_charge_write
,
4679 .name
= "oom_control",
4680 .read_map
= mem_cgroup_oom_control_read
,
4681 .write_u64
= mem_cgroup_oom_control_write
,
4682 .register_event
= mem_cgroup_oom_register_event
,
4683 .unregister_event
= mem_cgroup_oom_unregister_event
,
4684 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
4688 .name
= "numa_stat",
4689 .open
= mem_control_numa_stat_open
,
4695 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4696 static struct cftype memsw_cgroup_files
[] = {
4698 .name
= "memsw.usage_in_bytes",
4699 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
4700 .read_u64
= mem_cgroup_read
,
4701 .register_event
= mem_cgroup_usage_register_event
,
4702 .unregister_event
= mem_cgroup_usage_unregister_event
,
4705 .name
= "memsw.max_usage_in_bytes",
4706 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
4707 .trigger
= mem_cgroup_reset
,
4708 .read_u64
= mem_cgroup_read
,
4711 .name
= "memsw.limit_in_bytes",
4712 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
4713 .write_string
= mem_cgroup_write
,
4714 .read_u64
= mem_cgroup_read
,
4717 .name
= "memsw.failcnt",
4718 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
4719 .trigger
= mem_cgroup_reset
,
4720 .read_u64
= mem_cgroup_read
,
4724 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4726 if (!do_swap_account
)
4728 return cgroup_add_files(cont
, ss
, memsw_cgroup_files
,
4729 ARRAY_SIZE(memsw_cgroup_files
));
4732 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4738 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4740 struct mem_cgroup_per_node
*pn
;
4741 struct mem_cgroup_per_zone
*mz
;
4743 int zone
, tmp
= node
;
4745 * This routine is called against possible nodes.
4746 * But it's BUG to call kmalloc() against offline node.
4748 * TODO: this routine can waste much memory for nodes which will
4749 * never be onlined. It's better to use memory hotplug callback
4752 if (!node_state(node
, N_NORMAL_MEMORY
))
4754 pn
= kzalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
4758 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4759 mz
= &pn
->zoneinfo
[zone
];
4761 INIT_LIST_HEAD(&mz
->lruvec
.lists
[l
]);
4762 mz
->usage_in_excess
= 0;
4763 mz
->on_tree
= false;
4766 memcg
->info
.nodeinfo
[node
] = pn
;
4770 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4772 kfree(memcg
->info
.nodeinfo
[node
]);
4775 static struct mem_cgroup
*mem_cgroup_alloc(void)
4777 struct mem_cgroup
*mem
;
4778 int size
= sizeof(struct mem_cgroup
);
4780 /* Can be very big if MAX_NUMNODES is very big */
4781 if (size
< PAGE_SIZE
)
4782 mem
= kzalloc(size
, GFP_KERNEL
);
4784 mem
= vzalloc(size
);
4789 mem
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4792 spin_lock_init(&mem
->pcp_counter_lock
);
4796 if (size
< PAGE_SIZE
)
4804 * At destroying mem_cgroup, references from swap_cgroup can remain.
4805 * (scanning all at force_empty is too costly...)
4807 * Instead of clearing all references at force_empty, we remember
4808 * the number of reference from swap_cgroup and free mem_cgroup when
4809 * it goes down to 0.
4811 * Removal of cgroup itself succeeds regardless of refs from swap.
4814 static void __mem_cgroup_free(struct mem_cgroup
*memcg
)
4818 mem_cgroup_remove_from_trees(memcg
);
4819 free_css_id(&mem_cgroup_subsys
, &memcg
->css
);
4822 free_mem_cgroup_per_zone_info(memcg
, node
);
4824 free_percpu(memcg
->stat
);
4825 if (sizeof(struct mem_cgroup
) < PAGE_SIZE
)
4831 static void mem_cgroup_get(struct mem_cgroup
*memcg
)
4833 atomic_inc(&memcg
->refcnt
);
4836 static void __mem_cgroup_put(struct mem_cgroup
*memcg
, int count
)
4838 if (atomic_sub_and_test(count
, &memcg
->refcnt
)) {
4839 struct mem_cgroup
*parent
= parent_mem_cgroup(memcg
);
4840 __mem_cgroup_free(memcg
);
4842 mem_cgroup_put(parent
);
4846 static void mem_cgroup_put(struct mem_cgroup
*memcg
)
4848 __mem_cgroup_put(memcg
, 1);
4852 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4854 struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*memcg
)
4856 if (!memcg
->res
.parent
)
4858 return mem_cgroup_from_res_counter(memcg
->res
.parent
, res
);
4860 EXPORT_SYMBOL(parent_mem_cgroup
);
4862 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4863 static void __init
enable_swap_cgroup(void)
4865 if (!mem_cgroup_disabled() && really_do_swap_account
)
4866 do_swap_account
= 1;
4869 static void __init
enable_swap_cgroup(void)
4874 static int mem_cgroup_soft_limit_tree_init(void)
4876 struct mem_cgroup_tree_per_node
*rtpn
;
4877 struct mem_cgroup_tree_per_zone
*rtpz
;
4878 int tmp
, node
, zone
;
4880 for_each_node(node
) {
4882 if (!node_state(node
, N_NORMAL_MEMORY
))
4884 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
, tmp
);
4888 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
4890 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4891 rtpz
= &rtpn
->rb_tree_per_zone
[zone
];
4892 rtpz
->rb_root
= RB_ROOT
;
4893 spin_lock_init(&rtpz
->lock
);
4899 for_each_node(node
) {
4900 if (!soft_limit_tree
.rb_tree_per_node
[node
])
4902 kfree(soft_limit_tree
.rb_tree_per_node
[node
]);
4903 soft_limit_tree
.rb_tree_per_node
[node
] = NULL
;
4909 static struct cgroup_subsys_state
* __ref
4910 mem_cgroup_create(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
4912 struct mem_cgroup
*memcg
, *parent
;
4913 long error
= -ENOMEM
;
4916 memcg
= mem_cgroup_alloc();
4918 return ERR_PTR(error
);
4921 if (alloc_mem_cgroup_per_zone_info(memcg
, node
))
4925 if (cont
->parent
== NULL
) {
4927 enable_swap_cgroup();
4929 if (mem_cgroup_soft_limit_tree_init())
4931 root_mem_cgroup
= memcg
;
4932 for_each_possible_cpu(cpu
) {
4933 struct memcg_stock_pcp
*stock
=
4934 &per_cpu(memcg_stock
, cpu
);
4935 INIT_WORK(&stock
->work
, drain_local_stock
);
4937 hotcpu_notifier(memcg_cpu_hotplug_callback
, 0);
4939 parent
= mem_cgroup_from_cont(cont
->parent
);
4940 memcg
->use_hierarchy
= parent
->use_hierarchy
;
4941 memcg
->oom_kill_disable
= parent
->oom_kill_disable
;
4944 if (parent
&& parent
->use_hierarchy
) {
4945 res_counter_init(&memcg
->res
, &parent
->res
);
4946 res_counter_init(&memcg
->memsw
, &parent
->memsw
);
4948 * We increment refcnt of the parent to ensure that we can
4949 * safely access it on res_counter_charge/uncharge.
4950 * This refcnt will be decremented when freeing this
4951 * mem_cgroup(see mem_cgroup_put).
4953 mem_cgroup_get(parent
);
4955 res_counter_init(&memcg
->res
, NULL
);
4956 res_counter_init(&memcg
->memsw
, NULL
);
4958 memcg
->last_scanned_node
= MAX_NUMNODES
;
4959 INIT_LIST_HEAD(&memcg
->oom_notify
);
4962 memcg
->swappiness
= mem_cgroup_swappiness(parent
);
4963 atomic_set(&memcg
->refcnt
, 1);
4964 memcg
->move_charge_at_immigrate
= 0;
4965 mutex_init(&memcg
->thresholds_lock
);
4968 __mem_cgroup_free(memcg
);
4969 return ERR_PTR(error
);
4972 static int mem_cgroup_pre_destroy(struct cgroup_subsys
*ss
,
4973 struct cgroup
*cont
)
4975 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4977 return mem_cgroup_force_empty(memcg
, false);
4980 static void mem_cgroup_destroy(struct cgroup_subsys
*ss
,
4981 struct cgroup
*cont
)
4983 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4985 kmem_cgroup_destroy(ss
, cont
);
4987 mem_cgroup_put(memcg
);
4990 static int mem_cgroup_populate(struct cgroup_subsys
*ss
,
4991 struct cgroup
*cont
)
4995 ret
= cgroup_add_files(cont
, ss
, mem_cgroup_files
,
4996 ARRAY_SIZE(mem_cgroup_files
));
4999 ret
= register_memsw_files(cont
, ss
);
5002 ret
= register_kmem_files(cont
, ss
);
5008 /* Handlers for move charge at task migration. */
5009 #define PRECHARGE_COUNT_AT_ONCE 256
5010 static int mem_cgroup_do_precharge(unsigned long count
)
5013 int batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5014 struct mem_cgroup
*memcg
= mc
.to
;
5016 if (mem_cgroup_is_root(memcg
)) {
5017 mc
.precharge
+= count
;
5018 /* we don't need css_get for root */
5021 /* try to charge at once */
5023 struct res_counter
*dummy
;
5025 * "memcg" cannot be under rmdir() because we've already checked
5026 * by cgroup_lock_live_cgroup() that it is not removed and we
5027 * are still under the same cgroup_mutex. So we can postpone
5030 if (res_counter_charge(&memcg
->res
, PAGE_SIZE
* count
, &dummy
))
5032 if (do_swap_account
&& res_counter_charge(&memcg
->memsw
,
5033 PAGE_SIZE
* count
, &dummy
)) {
5034 res_counter_uncharge(&memcg
->res
, PAGE_SIZE
* count
);
5037 mc
.precharge
+= count
;
5041 /* fall back to one by one charge */
5043 if (signal_pending(current
)) {
5047 if (!batch_count
--) {
5048 batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5051 ret
= __mem_cgroup_try_charge(NULL
,
5052 GFP_KERNEL
, 1, &memcg
, false);
5054 /* mem_cgroup_clear_mc() will do uncharge later */
5062 * is_target_pte_for_mc - check a pte whether it is valid for move charge
5063 * @vma: the vma the pte to be checked belongs
5064 * @addr: the address corresponding to the pte to be checked
5065 * @ptent: the pte to be checked
5066 * @target: the pointer the target page or swap ent will be stored(can be NULL)
5069 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5070 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5071 * move charge. if @target is not NULL, the page is stored in target->page
5072 * with extra refcnt got(Callers should handle it).
5073 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5074 * target for charge migration. if @target is not NULL, the entry is stored
5077 * Called with pte lock held.
5084 enum mc_target_type
{
5085 MC_TARGET_NONE
, /* not used */
5090 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
5091 unsigned long addr
, pte_t ptent
)
5093 struct page
*page
= vm_normal_page(vma
, addr
, ptent
);
5095 if (!page
|| !page_mapped(page
))
5097 if (PageAnon(page
)) {
5098 /* we don't move shared anon */
5099 if (!move_anon() || page_mapcount(page
) > 2)
5101 } else if (!move_file())
5102 /* we ignore mapcount for file pages */
5104 if (!get_page_unless_zero(page
))
5110 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5111 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5114 struct page
*page
= NULL
;
5115 swp_entry_t ent
= pte_to_swp_entry(ptent
);
5117 if (!move_anon() || non_swap_entry(ent
))
5119 usage_count
= mem_cgroup_count_swap_user(ent
, &page
);
5120 if (usage_count
> 1) { /* we don't move shared anon */
5125 if (do_swap_account
)
5126 entry
->val
= ent
.val
;
5131 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
5132 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5134 struct page
*page
= NULL
;
5135 struct inode
*inode
;
5136 struct address_space
*mapping
;
5139 if (!vma
->vm_file
) /* anonymous vma */
5144 inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
5145 mapping
= vma
->vm_file
->f_mapping
;
5146 if (pte_none(ptent
))
5147 pgoff
= linear_page_index(vma
, addr
);
5148 else /* pte_file(ptent) is true */
5149 pgoff
= pte_to_pgoff(ptent
);
5151 /* page is moved even if it's not RSS of this task(page-faulted). */
5152 page
= find_get_page(mapping
, pgoff
);
5155 /* shmem/tmpfs may report page out on swap: account for that too. */
5156 if (radix_tree_exceptional_entry(page
)) {
5157 swp_entry_t swap
= radix_to_swp_entry(page
);
5158 if (do_swap_account
)
5160 page
= find_get_page(&swapper_space
, swap
.val
);
5166 static int is_target_pte_for_mc(struct vm_area_struct
*vma
,
5167 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
5169 struct page
*page
= NULL
;
5170 struct page_cgroup
*pc
;
5172 swp_entry_t ent
= { .val
= 0 };
5174 if (pte_present(ptent
))
5175 page
= mc_handle_present_pte(vma
, addr
, ptent
);
5176 else if (is_swap_pte(ptent
))
5177 page
= mc_handle_swap_pte(vma
, addr
, ptent
, &ent
);
5178 else if (pte_none(ptent
) || pte_file(ptent
))
5179 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
5181 if (!page
&& !ent
.val
)
5184 pc
= lookup_page_cgroup(page
);
5186 * Do only loose check w/o page_cgroup lock.
5187 * mem_cgroup_move_account() checks the pc is valid or not under
5190 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5191 ret
= MC_TARGET_PAGE
;
5193 target
->page
= page
;
5195 if (!ret
|| !target
)
5198 /* There is a swap entry and a page doesn't exist or isn't charged */
5199 if (ent
.val
&& !ret
&&
5200 css_id(&mc
.from
->css
) == lookup_swap_cgroup_id(ent
)) {
5201 ret
= MC_TARGET_SWAP
;
5208 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
5209 unsigned long addr
, unsigned long end
,
5210 struct mm_walk
*walk
)
5212 struct vm_area_struct
*vma
= walk
->private;
5216 split_huge_page_pmd(walk
->mm
, pmd
);
5218 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5219 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
5220 if (is_target_pte_for_mc(vma
, addr
, *pte
, NULL
))
5221 mc
.precharge
++; /* increment precharge temporarily */
5222 pte_unmap_unlock(pte
- 1, ptl
);
5228 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
5230 unsigned long precharge
;
5231 struct vm_area_struct
*vma
;
5233 down_read(&mm
->mmap_sem
);
5234 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5235 struct mm_walk mem_cgroup_count_precharge_walk
= {
5236 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
5240 if (is_vm_hugetlb_page(vma
))
5242 walk_page_range(vma
->vm_start
, vma
->vm_end
,
5243 &mem_cgroup_count_precharge_walk
);
5245 up_read(&mm
->mmap_sem
);
5247 precharge
= mc
.precharge
;
5253 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
5255 unsigned long precharge
= mem_cgroup_count_precharge(mm
);
5257 VM_BUG_ON(mc
.moving_task
);
5258 mc
.moving_task
= current
;
5259 return mem_cgroup_do_precharge(precharge
);
5262 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5263 static void __mem_cgroup_clear_mc(void)
5265 struct mem_cgroup
*from
= mc
.from
;
5266 struct mem_cgroup
*to
= mc
.to
;
5268 /* we must uncharge all the leftover precharges from mc.to */
5270 __mem_cgroup_cancel_charge(mc
.to
, mc
.precharge
);
5274 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5275 * we must uncharge here.
5277 if (mc
.moved_charge
) {
5278 __mem_cgroup_cancel_charge(mc
.from
, mc
.moved_charge
);
5279 mc
.moved_charge
= 0;
5281 /* we must fixup refcnts and charges */
5282 if (mc
.moved_swap
) {
5283 /* uncharge swap account from the old cgroup */
5284 if (!mem_cgroup_is_root(mc
.from
))
5285 res_counter_uncharge(&mc
.from
->memsw
,
5286 PAGE_SIZE
* mc
.moved_swap
);
5287 __mem_cgroup_put(mc
.from
, mc
.moved_swap
);
5289 if (!mem_cgroup_is_root(mc
.to
)) {
5291 * we charged both to->res and to->memsw, so we should
5294 res_counter_uncharge(&mc
.to
->res
,
5295 PAGE_SIZE
* mc
.moved_swap
);
5297 /* we've already done mem_cgroup_get(mc.to) */
5300 memcg_oom_recover(from
);
5301 memcg_oom_recover(to
);
5302 wake_up_all(&mc
.waitq
);
5305 static void mem_cgroup_clear_mc(void)
5307 struct mem_cgroup
*from
= mc
.from
;
5310 * we must clear moving_task before waking up waiters at the end of
5313 mc
.moving_task
= NULL
;
5314 __mem_cgroup_clear_mc();
5315 spin_lock(&mc
.lock
);
5318 spin_unlock(&mc
.lock
);
5319 mem_cgroup_end_move(from
);
5322 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
5323 struct cgroup
*cgroup
,
5324 struct cgroup_taskset
*tset
)
5326 struct task_struct
*p
= cgroup_taskset_first(tset
);
5328 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgroup
);
5330 if (memcg
->move_charge_at_immigrate
) {
5331 struct mm_struct
*mm
;
5332 struct mem_cgroup
*from
= mem_cgroup_from_task(p
);
5334 VM_BUG_ON(from
== memcg
);
5336 mm
= get_task_mm(p
);
5339 /* We move charges only when we move a owner of the mm */
5340 if (mm
->owner
== p
) {
5343 VM_BUG_ON(mc
.precharge
);
5344 VM_BUG_ON(mc
.moved_charge
);
5345 VM_BUG_ON(mc
.moved_swap
);
5346 mem_cgroup_start_move(from
);
5347 spin_lock(&mc
.lock
);
5350 spin_unlock(&mc
.lock
);
5351 /* We set mc.moving_task later */
5353 ret
= mem_cgroup_precharge_mc(mm
);
5355 mem_cgroup_clear_mc();
5362 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
5363 struct cgroup
*cgroup
,
5364 struct cgroup_taskset
*tset
)
5366 mem_cgroup_clear_mc();
5369 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
5370 unsigned long addr
, unsigned long end
,
5371 struct mm_walk
*walk
)
5374 struct vm_area_struct
*vma
= walk
->private;
5378 split_huge_page_pmd(walk
->mm
, pmd
);
5380 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5381 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
5382 pte_t ptent
= *(pte
++);
5383 union mc_target target
;
5386 struct page_cgroup
*pc
;
5392 type
= is_target_pte_for_mc(vma
, addr
, ptent
, &target
);
5394 case MC_TARGET_PAGE
:
5396 if (isolate_lru_page(page
))
5398 pc
= lookup_page_cgroup(page
);
5399 if (!mem_cgroup_move_account(page
, 1, pc
,
5400 mc
.from
, mc
.to
, false)) {
5402 /* we uncharge from mc.from later. */
5405 putback_lru_page(page
);
5406 put
: /* is_target_pte_for_mc() gets the page */
5409 case MC_TARGET_SWAP
:
5411 if (!mem_cgroup_move_swap_account(ent
,
5412 mc
.from
, mc
.to
, false)) {
5414 /* we fixup refcnts and charges later. */
5422 pte_unmap_unlock(pte
- 1, ptl
);
5427 * We have consumed all precharges we got in can_attach().
5428 * We try charge one by one, but don't do any additional
5429 * charges to mc.to if we have failed in charge once in attach()
5432 ret
= mem_cgroup_do_precharge(1);
5440 static void mem_cgroup_move_charge(struct mm_struct
*mm
)
5442 struct vm_area_struct
*vma
;
5444 lru_add_drain_all();
5446 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
5448 * Someone who are holding the mmap_sem might be waiting in
5449 * waitq. So we cancel all extra charges, wake up all waiters,
5450 * and retry. Because we cancel precharges, we might not be able
5451 * to move enough charges, but moving charge is a best-effort
5452 * feature anyway, so it wouldn't be a big problem.
5454 __mem_cgroup_clear_mc();
5458 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5460 struct mm_walk mem_cgroup_move_charge_walk
= {
5461 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
5465 if (is_vm_hugetlb_page(vma
))
5467 ret
= walk_page_range(vma
->vm_start
, vma
->vm_end
,
5468 &mem_cgroup_move_charge_walk
);
5471 * means we have consumed all precharges and failed in
5472 * doing additional charge. Just abandon here.
5476 up_read(&mm
->mmap_sem
);
5479 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
5480 struct cgroup
*cont
,
5481 struct cgroup_taskset
*tset
)
5483 struct task_struct
*p
= cgroup_taskset_first(tset
);
5484 struct mm_struct
*mm
= get_task_mm(p
);
5488 mem_cgroup_move_charge(mm
);
5493 mem_cgroup_clear_mc();
5495 #else /* !CONFIG_MMU */
5496 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
5497 struct cgroup
*cgroup
,
5498 struct cgroup_taskset
*tset
)
5502 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
5503 struct cgroup
*cgroup
,
5504 struct cgroup_taskset
*tset
)
5507 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
5508 struct cgroup
*cont
,
5509 struct cgroup_taskset
*tset
)
5514 struct cgroup_subsys mem_cgroup_subsys
= {
5516 .subsys_id
= mem_cgroup_subsys_id
,
5517 .create
= mem_cgroup_create
,
5518 .pre_destroy
= mem_cgroup_pre_destroy
,
5519 .destroy
= mem_cgroup_destroy
,
5520 .populate
= mem_cgroup_populate
,
5521 .can_attach
= mem_cgroup_can_attach
,
5522 .cancel_attach
= mem_cgroup_cancel_attach
,
5523 .attach
= mem_cgroup_move_task
,
5528 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
5529 static int __init
enable_swap_account(char *s
)
5531 /* consider enabled if no parameter or 1 is given */
5532 if (!strcmp(s
, "1"))
5533 really_do_swap_account
= 1;
5534 else if (!strcmp(s
, "0"))
5535 really_do_swap_account
= 0;
5538 __setup("swapaccount=", enable_swap_account
);