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/mutex.h>
37 #include <linux/rbtree.h>
38 #include <linux/shmem_fs.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 <asm/uaccess.h>
56 #include <trace/events/vmscan.h>
58 struct cgroup_subsys mem_cgroup_subsys __read_mostly
;
59 #define MEM_CGROUP_RECLAIM_RETRIES 5
60 struct mem_cgroup
*root_mem_cgroup __read_mostly
;
62 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
63 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
64 int do_swap_account __read_mostly
;
66 /* for remember boot option*/
67 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED
68 static int really_do_swap_account __initdata
= 1;
70 static int really_do_swap_account __initdata
= 0;
74 #define do_swap_account (0)
79 * Statistics for memory cgroup.
81 enum mem_cgroup_stat_index
{
83 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
85 MEM_CGROUP_STAT_CACHE
, /* # of pages charged as cache */
86 MEM_CGROUP_STAT_RSS
, /* # of pages charged as anon rss */
87 MEM_CGROUP_STAT_FILE_MAPPED
, /* # of pages charged as file rss */
88 MEM_CGROUP_STAT_SWAPOUT
, /* # of pages, swapped out */
89 MEM_CGROUP_STAT_DATA
, /* end of data requires synchronization */
90 MEM_CGROUP_ON_MOVE
, /* someone is moving account between groups */
91 MEM_CGROUP_STAT_NSTATS
,
94 enum mem_cgroup_events_index
{
95 MEM_CGROUP_EVENTS_PGPGIN
, /* # of pages paged in */
96 MEM_CGROUP_EVENTS_PGPGOUT
, /* # of pages paged out */
97 MEM_CGROUP_EVENTS_COUNT
, /* # of pages paged in/out */
98 MEM_CGROUP_EVENTS_PGFAULT
, /* # of page-faults */
99 MEM_CGROUP_EVENTS_PGMAJFAULT
, /* # of major page-faults */
100 MEM_CGROUP_EVENTS_NSTATS
,
103 * Per memcg event counter is incremented at every pagein/pageout. With THP,
104 * it will be incremated by the number of pages. This counter is used for
105 * for trigger some periodic events. This is straightforward and better
106 * than using jiffies etc. to handle periodic memcg event.
108 enum mem_cgroup_events_target
{
109 MEM_CGROUP_TARGET_THRESH
,
110 MEM_CGROUP_TARGET_SOFTLIMIT
,
111 MEM_CGROUP_TARGET_NUMAINFO
,
114 #define THRESHOLDS_EVENTS_TARGET (128)
115 #define SOFTLIMIT_EVENTS_TARGET (1024)
116 #define NUMAINFO_EVENTS_TARGET (1024)
118 struct mem_cgroup_stat_cpu
{
119 long count
[MEM_CGROUP_STAT_NSTATS
];
120 unsigned long events
[MEM_CGROUP_EVENTS_NSTATS
];
121 unsigned long targets
[MEM_CGROUP_NTARGETS
];
125 * per-zone information in memory controller.
127 struct mem_cgroup_per_zone
{
129 * spin_lock to protect the per cgroup LRU
131 struct list_head lists
[NR_LRU_LISTS
];
132 unsigned long count
[NR_LRU_LISTS
];
134 struct zone_reclaim_stat reclaim_stat
;
135 struct rb_node tree_node
; /* RB tree node */
136 unsigned long long usage_in_excess
;/* Set to the value by which */
137 /* the soft limit is exceeded*/
139 struct mem_cgroup
*mem
; /* Back pointer, we cannot */
140 /* use container_of */
142 /* Macro for accessing counter */
143 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
145 struct mem_cgroup_per_node
{
146 struct mem_cgroup_per_zone zoneinfo
[MAX_NR_ZONES
];
149 struct mem_cgroup_lru_info
{
150 struct mem_cgroup_per_node
*nodeinfo
[MAX_NUMNODES
];
154 * Cgroups above their limits are maintained in a RB-Tree, independent of
155 * their hierarchy representation
158 struct mem_cgroup_tree_per_zone
{
159 struct rb_root rb_root
;
163 struct mem_cgroup_tree_per_node
{
164 struct mem_cgroup_tree_per_zone rb_tree_per_zone
[MAX_NR_ZONES
];
167 struct mem_cgroup_tree
{
168 struct mem_cgroup_tree_per_node
*rb_tree_per_node
[MAX_NUMNODES
];
171 static struct mem_cgroup_tree soft_limit_tree __read_mostly
;
173 struct mem_cgroup_threshold
{
174 struct eventfd_ctx
*eventfd
;
179 struct mem_cgroup_threshold_ary
{
180 /* An array index points to threshold just below usage. */
181 int current_threshold
;
182 /* Size of entries[] */
184 /* Array of thresholds */
185 struct mem_cgroup_threshold entries
[0];
188 struct mem_cgroup_thresholds
{
189 /* Primary thresholds array */
190 struct mem_cgroup_threshold_ary
*primary
;
192 * Spare threshold array.
193 * This is needed to make mem_cgroup_unregister_event() "never fail".
194 * It must be able to store at least primary->size - 1 entries.
196 struct mem_cgroup_threshold_ary
*spare
;
200 struct mem_cgroup_eventfd_list
{
201 struct list_head list
;
202 struct eventfd_ctx
*eventfd
;
205 static void mem_cgroup_threshold(struct mem_cgroup
*mem
);
206 static void mem_cgroup_oom_notify(struct mem_cgroup
*mem
);
212 SCAN_BY_SHRINK
, /* not recorded now */
231 unsigned long stats
[NR_SCAN_CONTEXT
][NR_SCANSTATS
];
232 unsigned long rootstats
[NR_SCAN_CONTEXT
][NR_SCANSTATS
];
235 const char *scanstat_string
[NR_SCANSTATS
] = {
237 "scanned_anon_pages",
238 "scanned_file_pages",
240 "rotated_anon_pages",
241 "rotated_file_pages",
247 #define SCANSTAT_WORD_LIMIT "_by_limit"
248 #define SCANSTAT_WORD_SYSTEM "_by_system"
249 #define SCANSTAT_WORD_HIERARCHY "_under_hierarchy"
253 * The memory controller data structure. The memory controller controls both
254 * page cache and RSS per cgroup. We would eventually like to provide
255 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
256 * to help the administrator determine what knobs to tune.
258 * TODO: Add a water mark for the memory controller. Reclaim will begin when
259 * we hit the water mark. May be even add a low water mark, such that
260 * no reclaim occurs from a cgroup at it's low water mark, this is
261 * a feature that will be implemented much later in the future.
264 struct cgroup_subsys_state css
;
266 * the counter to account for memory usage
268 struct res_counter res
;
270 * the counter to account for mem+swap usage.
272 struct res_counter memsw
;
274 * Per cgroup active and inactive list, similar to the
275 * per zone LRU lists.
277 struct mem_cgroup_lru_info info
;
279 * While reclaiming in a hierarchy, we cache the last child we
282 int last_scanned_child
;
283 int last_scanned_node
;
285 nodemask_t scan_nodes
;
286 atomic_t numainfo_events
;
287 atomic_t numainfo_updating
;
290 * Should the accounting and control be hierarchical, per subtree?
300 /* OOM-Killer disable */
301 int oom_kill_disable
;
303 /* set when res.limit == memsw.limit */
304 bool memsw_is_minimum
;
306 /* protect arrays of thresholds */
307 struct mutex thresholds_lock
;
309 /* thresholds for memory usage. RCU-protected */
310 struct mem_cgroup_thresholds thresholds
;
312 /* thresholds for mem+swap usage. RCU-protected */
313 struct mem_cgroup_thresholds memsw_thresholds
;
315 /* For oom notifier event fd */
316 struct list_head oom_notify
;
317 /* For recording LRU-scan statistics */
318 struct scanstat scanstat
;
320 * Should we move charges of a task when a task is moved into this
321 * mem_cgroup ? And what type of charges should we move ?
323 unsigned long move_charge_at_immigrate
;
327 struct mem_cgroup_stat_cpu
*stat
;
329 * used when a cpu is offlined or other synchronizations
330 * See mem_cgroup_read_stat().
332 struct mem_cgroup_stat_cpu nocpu_base
;
333 spinlock_t pcp_counter_lock
;
336 /* Stuffs for move charges at task migration. */
338 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
339 * left-shifted bitmap of these types.
342 MOVE_CHARGE_TYPE_ANON
, /* private anonymous page and swap of it */
343 MOVE_CHARGE_TYPE_FILE
, /* file page(including tmpfs) and swap of it */
347 /* "mc" and its members are protected by cgroup_mutex */
348 static struct move_charge_struct
{
349 spinlock_t lock
; /* for from, to */
350 struct mem_cgroup
*from
;
351 struct mem_cgroup
*to
;
352 unsigned long precharge
;
353 unsigned long moved_charge
;
354 unsigned long moved_swap
;
355 struct task_struct
*moving_task
; /* a task moving charges */
356 wait_queue_head_t waitq
; /* a waitq for other context */
358 .lock
= __SPIN_LOCK_UNLOCKED(mc
.lock
),
359 .waitq
= __WAIT_QUEUE_HEAD_INITIALIZER(mc
.waitq
),
362 static bool move_anon(void)
364 return test_bit(MOVE_CHARGE_TYPE_ANON
,
365 &mc
.to
->move_charge_at_immigrate
);
368 static bool move_file(void)
370 return test_bit(MOVE_CHARGE_TYPE_FILE
,
371 &mc
.to
->move_charge_at_immigrate
);
375 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
376 * limit reclaim to prevent infinite loops, if they ever occur.
378 #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
379 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
382 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
383 MEM_CGROUP_CHARGE_TYPE_MAPPED
,
384 MEM_CGROUP_CHARGE_TYPE_SHMEM
, /* used by page migration of shmem */
385 MEM_CGROUP_CHARGE_TYPE_FORCE
, /* used by force_empty */
386 MEM_CGROUP_CHARGE_TYPE_SWAPOUT
, /* for accounting swapcache */
387 MEM_CGROUP_CHARGE_TYPE_DROP
, /* a page was unused swap cache */
391 /* for encoding cft->private value on file */
394 #define _OOM_TYPE (2)
395 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
396 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
397 #define MEMFILE_ATTR(val) ((val) & 0xffff)
398 /* Used for OOM nofiier */
399 #define OOM_CONTROL (0)
402 * Reclaim flags for mem_cgroup_hierarchical_reclaim
404 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
405 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
406 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
407 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
408 #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2
409 #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT)
411 static void mem_cgroup_get(struct mem_cgroup
*mem
);
412 static void mem_cgroup_put(struct mem_cgroup
*mem
);
413 static struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*mem
);
414 static void drain_all_stock_async(struct mem_cgroup
*mem
);
416 static struct mem_cgroup_per_zone
*
417 mem_cgroup_zoneinfo(struct mem_cgroup
*mem
, int nid
, int zid
)
419 return &mem
->info
.nodeinfo
[nid
]->zoneinfo
[zid
];
422 struct cgroup_subsys_state
*mem_cgroup_css(struct mem_cgroup
*mem
)
427 static struct mem_cgroup_per_zone
*
428 page_cgroup_zoneinfo(struct mem_cgroup
*mem
, struct page
*page
)
430 int nid
= page_to_nid(page
);
431 int zid
= page_zonenum(page
);
433 return mem_cgroup_zoneinfo(mem
, nid
, zid
);
436 static struct mem_cgroup_tree_per_zone
*
437 soft_limit_tree_node_zone(int nid
, int zid
)
439 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
442 static struct mem_cgroup_tree_per_zone
*
443 soft_limit_tree_from_page(struct page
*page
)
445 int nid
= page_to_nid(page
);
446 int zid
= page_zonenum(page
);
448 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
452 __mem_cgroup_insert_exceeded(struct mem_cgroup
*mem
,
453 struct mem_cgroup_per_zone
*mz
,
454 struct mem_cgroup_tree_per_zone
*mctz
,
455 unsigned long long new_usage_in_excess
)
457 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
458 struct rb_node
*parent
= NULL
;
459 struct mem_cgroup_per_zone
*mz_node
;
464 mz
->usage_in_excess
= new_usage_in_excess
;
465 if (!mz
->usage_in_excess
)
469 mz_node
= rb_entry(parent
, struct mem_cgroup_per_zone
,
471 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
)
474 * We can't avoid mem cgroups that are over their soft
475 * limit by the same amount
477 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
480 rb_link_node(&mz
->tree_node
, parent
, p
);
481 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
486 __mem_cgroup_remove_exceeded(struct mem_cgroup
*mem
,
487 struct mem_cgroup_per_zone
*mz
,
488 struct mem_cgroup_tree_per_zone
*mctz
)
492 rb_erase(&mz
->tree_node
, &mctz
->rb_root
);
497 mem_cgroup_remove_exceeded(struct mem_cgroup
*mem
,
498 struct mem_cgroup_per_zone
*mz
,
499 struct mem_cgroup_tree_per_zone
*mctz
)
501 spin_lock(&mctz
->lock
);
502 __mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
503 spin_unlock(&mctz
->lock
);
507 static void mem_cgroup_update_tree(struct mem_cgroup
*mem
, struct page
*page
)
509 unsigned long long excess
;
510 struct mem_cgroup_per_zone
*mz
;
511 struct mem_cgroup_tree_per_zone
*mctz
;
512 int nid
= page_to_nid(page
);
513 int zid
= page_zonenum(page
);
514 mctz
= soft_limit_tree_from_page(page
);
517 * Necessary to update all ancestors when hierarchy is used.
518 * because their event counter is not touched.
520 for (; mem
; mem
= parent_mem_cgroup(mem
)) {
521 mz
= mem_cgroup_zoneinfo(mem
, nid
, zid
);
522 excess
= res_counter_soft_limit_excess(&mem
->res
);
524 * We have to update the tree if mz is on RB-tree or
525 * mem is over its softlimit.
527 if (excess
|| mz
->on_tree
) {
528 spin_lock(&mctz
->lock
);
529 /* if on-tree, remove it */
531 __mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
533 * Insert again. mz->usage_in_excess will be updated.
534 * If excess is 0, no tree ops.
536 __mem_cgroup_insert_exceeded(mem
, mz
, mctz
, excess
);
537 spin_unlock(&mctz
->lock
);
542 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*mem
)
545 struct mem_cgroup_per_zone
*mz
;
546 struct mem_cgroup_tree_per_zone
*mctz
;
548 for_each_node_state(node
, N_POSSIBLE
) {
549 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
550 mz
= mem_cgroup_zoneinfo(mem
, node
, zone
);
551 mctz
= soft_limit_tree_node_zone(node
, zone
);
552 mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
557 static struct mem_cgroup_per_zone
*
558 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
560 struct rb_node
*rightmost
= NULL
;
561 struct mem_cgroup_per_zone
*mz
;
565 rightmost
= rb_last(&mctz
->rb_root
);
567 goto done
; /* Nothing to reclaim from */
569 mz
= rb_entry(rightmost
, struct mem_cgroup_per_zone
, tree_node
);
571 * Remove the node now but someone else can add it back,
572 * we will to add it back at the end of reclaim to its correct
573 * position in the tree.
575 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
576 if (!res_counter_soft_limit_excess(&mz
->mem
->res
) ||
577 !css_tryget(&mz
->mem
->css
))
583 static struct mem_cgroup_per_zone
*
584 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
586 struct mem_cgroup_per_zone
*mz
;
588 spin_lock(&mctz
->lock
);
589 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
590 spin_unlock(&mctz
->lock
);
595 * Implementation Note: reading percpu statistics for memcg.
597 * Both of vmstat[] and percpu_counter has threshold and do periodic
598 * synchronization to implement "quick" read. There are trade-off between
599 * reading cost and precision of value. Then, we may have a chance to implement
600 * a periodic synchronizion of counter in memcg's counter.
602 * But this _read() function is used for user interface now. The user accounts
603 * memory usage by memory cgroup and he _always_ requires exact value because
604 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
605 * have to visit all online cpus and make sum. So, for now, unnecessary
606 * synchronization is not implemented. (just implemented for cpu hotplug)
608 * If there are kernel internal actions which can make use of some not-exact
609 * value, and reading all cpu value can be performance bottleneck in some
610 * common workload, threashold and synchonization as vmstat[] should be
613 static long mem_cgroup_read_stat(struct mem_cgroup
*mem
,
614 enum mem_cgroup_stat_index idx
)
620 for_each_online_cpu(cpu
)
621 val
+= per_cpu(mem
->stat
->count
[idx
], cpu
);
622 #ifdef CONFIG_HOTPLUG_CPU
623 spin_lock(&mem
->pcp_counter_lock
);
624 val
+= mem
->nocpu_base
.count
[idx
];
625 spin_unlock(&mem
->pcp_counter_lock
);
631 static void mem_cgroup_swap_statistics(struct mem_cgroup
*mem
,
634 int val
= (charge
) ? 1 : -1;
635 this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_SWAPOUT
], val
);
638 void mem_cgroup_pgfault(struct mem_cgroup
*mem
, int val
)
640 this_cpu_add(mem
->stat
->events
[MEM_CGROUP_EVENTS_PGFAULT
], val
);
643 void mem_cgroup_pgmajfault(struct mem_cgroup
*mem
, int val
)
645 this_cpu_add(mem
->stat
->events
[MEM_CGROUP_EVENTS_PGMAJFAULT
], val
);
648 static unsigned long mem_cgroup_read_events(struct mem_cgroup
*mem
,
649 enum mem_cgroup_events_index idx
)
651 unsigned long val
= 0;
654 for_each_online_cpu(cpu
)
655 val
+= per_cpu(mem
->stat
->events
[idx
], cpu
);
656 #ifdef CONFIG_HOTPLUG_CPU
657 spin_lock(&mem
->pcp_counter_lock
);
658 val
+= mem
->nocpu_base
.events
[idx
];
659 spin_unlock(&mem
->pcp_counter_lock
);
664 static void mem_cgroup_charge_statistics(struct mem_cgroup
*mem
,
665 bool file
, int nr_pages
)
670 __this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_CACHE
], nr_pages
);
672 __this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_RSS
], nr_pages
);
674 /* pagein of a big page is an event. So, ignore page size */
676 __this_cpu_inc(mem
->stat
->events
[MEM_CGROUP_EVENTS_PGPGIN
]);
678 __this_cpu_inc(mem
->stat
->events
[MEM_CGROUP_EVENTS_PGPGOUT
]);
679 nr_pages
= -nr_pages
; /* for event */
682 __this_cpu_add(mem
->stat
->events
[MEM_CGROUP_EVENTS_COUNT
], nr_pages
);
688 mem_cgroup_zone_nr_lru_pages(struct mem_cgroup
*mem
, int nid
, int zid
,
689 unsigned int lru_mask
)
691 struct mem_cgroup_per_zone
*mz
;
693 unsigned long ret
= 0;
695 mz
= mem_cgroup_zoneinfo(mem
, nid
, zid
);
698 if (BIT(l
) & lru_mask
)
699 ret
+= MEM_CGROUP_ZSTAT(mz
, l
);
705 mem_cgroup_node_nr_lru_pages(struct mem_cgroup
*mem
,
706 int nid
, unsigned int lru_mask
)
711 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++)
712 total
+= mem_cgroup_zone_nr_lru_pages(mem
, nid
, zid
, lru_mask
);
717 static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup
*mem
,
718 unsigned int lru_mask
)
723 for_each_node_state(nid
, N_HIGH_MEMORY
)
724 total
+= mem_cgroup_node_nr_lru_pages(mem
, nid
, lru_mask
);
728 static bool __memcg_event_check(struct mem_cgroup
*mem
, int target
)
730 unsigned long val
, next
;
732 val
= this_cpu_read(mem
->stat
->events
[MEM_CGROUP_EVENTS_COUNT
]);
733 next
= this_cpu_read(mem
->stat
->targets
[target
]);
734 /* from time_after() in jiffies.h */
735 return ((long)next
- (long)val
< 0);
738 static void __mem_cgroup_target_update(struct mem_cgroup
*mem
, int target
)
740 unsigned long val
, next
;
742 val
= this_cpu_read(mem
->stat
->events
[MEM_CGROUP_EVENTS_COUNT
]);
745 case MEM_CGROUP_TARGET_THRESH
:
746 next
= val
+ THRESHOLDS_EVENTS_TARGET
;
748 case MEM_CGROUP_TARGET_SOFTLIMIT
:
749 next
= val
+ SOFTLIMIT_EVENTS_TARGET
;
751 case MEM_CGROUP_TARGET_NUMAINFO
:
752 next
= val
+ NUMAINFO_EVENTS_TARGET
;
758 this_cpu_write(mem
->stat
->targets
[target
], next
);
762 * Check events in order.
765 static void memcg_check_events(struct mem_cgroup
*mem
, struct page
*page
)
767 /* threshold event is triggered in finer grain than soft limit */
768 if (unlikely(__memcg_event_check(mem
, MEM_CGROUP_TARGET_THRESH
))) {
769 mem_cgroup_threshold(mem
);
770 __mem_cgroup_target_update(mem
, MEM_CGROUP_TARGET_THRESH
);
771 if (unlikely(__memcg_event_check(mem
,
772 MEM_CGROUP_TARGET_SOFTLIMIT
))) {
773 mem_cgroup_update_tree(mem
, page
);
774 __mem_cgroup_target_update(mem
,
775 MEM_CGROUP_TARGET_SOFTLIMIT
);
778 if (unlikely(__memcg_event_check(mem
,
779 MEM_CGROUP_TARGET_NUMAINFO
))) {
780 atomic_inc(&mem
->numainfo_events
);
781 __mem_cgroup_target_update(mem
,
782 MEM_CGROUP_TARGET_NUMAINFO
);
788 static struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
790 return container_of(cgroup_subsys_state(cont
,
791 mem_cgroup_subsys_id
), struct mem_cgroup
,
795 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
798 * mm_update_next_owner() may clear mm->owner to NULL
799 * if it races with swapoff, page migration, etc.
800 * So this can be called with p == NULL.
805 return container_of(task_subsys_state(p
, mem_cgroup_subsys_id
),
806 struct mem_cgroup
, css
);
809 struct mem_cgroup
*try_get_mem_cgroup_from_mm(struct mm_struct
*mm
)
811 struct mem_cgroup
*mem
= NULL
;
816 * Because we have no locks, mm->owner's may be being moved to other
817 * cgroup. We use css_tryget() here even if this looks
818 * pessimistic (rather than adding locks here).
822 mem
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
825 } while (!css_tryget(&mem
->css
));
830 /* The caller has to guarantee "mem" exists before calling this */
831 static struct mem_cgroup
*mem_cgroup_start_loop(struct mem_cgroup
*mem
)
833 struct cgroup_subsys_state
*css
;
836 if (!mem
) /* ROOT cgroup has the smallest ID */
837 return root_mem_cgroup
; /*css_put/get against root is ignored*/
838 if (!mem
->use_hierarchy
) {
839 if (css_tryget(&mem
->css
))
845 * searching a memory cgroup which has the smallest ID under given
846 * ROOT cgroup. (ID >= 1)
848 css
= css_get_next(&mem_cgroup_subsys
, 1, &mem
->css
, &found
);
849 if (css
&& css_tryget(css
))
850 mem
= container_of(css
, struct mem_cgroup
, css
);
857 static struct mem_cgroup
*mem_cgroup_get_next(struct mem_cgroup
*iter
,
858 struct mem_cgroup
*root
,
861 int nextid
= css_id(&iter
->css
) + 1;
864 struct cgroup_subsys_state
*css
;
866 hierarchy_used
= iter
->use_hierarchy
;
869 /* If no ROOT, walk all, ignore hierarchy */
870 if (!cond
|| (root
&& !hierarchy_used
))
874 root
= root_mem_cgroup
;
880 css
= css_get_next(&mem_cgroup_subsys
, nextid
,
882 if (css
&& css_tryget(css
))
883 iter
= container_of(css
, struct mem_cgroup
, css
);
885 /* If css is NULL, no more cgroups will be found */
887 } while (css
&& !iter
);
892 * for_eacn_mem_cgroup_tree() for visiting all cgroup under tree. Please
893 * be careful that "break" loop is not allowed. We have reference count.
894 * Instead of that modify "cond" to be false and "continue" to exit the loop.
896 #define for_each_mem_cgroup_tree_cond(iter, root, cond) \
897 for (iter = mem_cgroup_start_loop(root);\
899 iter = mem_cgroup_get_next(iter, root, cond))
901 #define for_each_mem_cgroup_tree(iter, root) \
902 for_each_mem_cgroup_tree_cond(iter, root, true)
904 #define for_each_mem_cgroup_all(iter) \
905 for_each_mem_cgroup_tree_cond(iter, NULL, true)
908 static inline bool mem_cgroup_is_root(struct mem_cgroup
*mem
)
910 return (mem
== root_mem_cgroup
);
913 void mem_cgroup_count_vm_event(struct mm_struct
*mm
, enum vm_event_item idx
)
915 struct mem_cgroup
*mem
;
921 mem
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
927 mem_cgroup_pgmajfault(mem
, 1);
930 mem_cgroup_pgfault(mem
, 1);
938 EXPORT_SYMBOL(mem_cgroup_count_vm_event
);
941 * Following LRU functions are allowed to be used without PCG_LOCK.
942 * Operations are called by routine of global LRU independently from memcg.
943 * What we have to take care of here is validness of pc->mem_cgroup.
945 * Changes to pc->mem_cgroup happens when
948 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
949 * It is added to LRU before charge.
950 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
951 * When moving account, the page is not on LRU. It's isolated.
954 void mem_cgroup_del_lru_list(struct page
*page
, enum lru_list lru
)
956 struct page_cgroup
*pc
;
957 struct mem_cgroup_per_zone
*mz
;
959 if (mem_cgroup_disabled())
961 pc
= lookup_page_cgroup(page
);
962 /* can happen while we handle swapcache. */
963 if (!TestClearPageCgroupAcctLRU(pc
))
965 VM_BUG_ON(!pc
->mem_cgroup
);
967 * We don't check PCG_USED bit. It's cleared when the "page" is finally
968 * removed from global LRU.
970 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
971 /* huge page split is done under lru_lock. so, we have no races. */
972 MEM_CGROUP_ZSTAT(mz
, lru
) -= 1 << compound_order(page
);
973 if (mem_cgroup_is_root(pc
->mem_cgroup
))
975 VM_BUG_ON(list_empty(&pc
->lru
));
976 list_del_init(&pc
->lru
);
979 void mem_cgroup_del_lru(struct page
*page
)
981 mem_cgroup_del_lru_list(page
, page_lru(page
));
985 * Writeback is about to end against a page which has been marked for immediate
986 * reclaim. If it still appears to be reclaimable, move it to the tail of the
989 void mem_cgroup_rotate_reclaimable_page(struct page
*page
)
991 struct mem_cgroup_per_zone
*mz
;
992 struct page_cgroup
*pc
;
993 enum lru_list lru
= page_lru(page
);
995 if (mem_cgroup_disabled())
998 pc
= lookup_page_cgroup(page
);
999 /* unused or root page is not rotated. */
1000 if (!PageCgroupUsed(pc
))
1002 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1004 if (mem_cgroup_is_root(pc
->mem_cgroup
))
1006 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
1007 list_move_tail(&pc
->lru
, &mz
->lists
[lru
]);
1010 void mem_cgroup_rotate_lru_list(struct page
*page
, enum lru_list lru
)
1012 struct mem_cgroup_per_zone
*mz
;
1013 struct page_cgroup
*pc
;
1015 if (mem_cgroup_disabled())
1018 pc
= lookup_page_cgroup(page
);
1019 /* unused or root page is not rotated. */
1020 if (!PageCgroupUsed(pc
))
1022 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1024 if (mem_cgroup_is_root(pc
->mem_cgroup
))
1026 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
1027 list_move(&pc
->lru
, &mz
->lists
[lru
]);
1030 void mem_cgroup_add_lru_list(struct page
*page
, enum lru_list lru
)
1032 struct page_cgroup
*pc
;
1033 struct mem_cgroup_per_zone
*mz
;
1035 if (mem_cgroup_disabled())
1037 pc
= lookup_page_cgroup(page
);
1038 VM_BUG_ON(PageCgroupAcctLRU(pc
));
1039 if (!PageCgroupUsed(pc
))
1041 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1043 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
1044 /* huge page split is done under lru_lock. so, we have no races. */
1045 MEM_CGROUP_ZSTAT(mz
, lru
) += 1 << compound_order(page
);
1046 SetPageCgroupAcctLRU(pc
);
1047 if (mem_cgroup_is_root(pc
->mem_cgroup
))
1049 list_add(&pc
->lru
, &mz
->lists
[lru
]);
1053 * At handling SwapCache and other FUSE stuff, pc->mem_cgroup may be changed
1054 * while it's linked to lru because the page may be reused after it's fully
1055 * uncharged. To handle that, unlink page_cgroup from LRU when charge it again.
1056 * It's done under lock_page and expected that zone->lru_lock isnever held.
1058 static void mem_cgroup_lru_del_before_commit(struct page
*page
)
1060 unsigned long flags
;
1061 struct zone
*zone
= page_zone(page
);
1062 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1065 * Doing this check without taking ->lru_lock seems wrong but this
1066 * is safe. Because if page_cgroup's USED bit is unset, the page
1067 * will not be added to any memcg's LRU. If page_cgroup's USED bit is
1068 * set, the commit after this will fail, anyway.
1069 * This all charge/uncharge is done under some mutual execustion.
1070 * So, we don't need to taking care of changes in USED bit.
1072 if (likely(!PageLRU(page
)))
1075 spin_lock_irqsave(&zone
->lru_lock
, flags
);
1077 * Forget old LRU when this page_cgroup is *not* used. This Used bit
1078 * is guarded by lock_page() because the page is SwapCache.
1080 if (!PageCgroupUsed(pc
))
1081 mem_cgroup_del_lru_list(page
, page_lru(page
));
1082 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
1085 static void mem_cgroup_lru_add_after_commit(struct page
*page
)
1087 unsigned long flags
;
1088 struct zone
*zone
= page_zone(page
);
1089 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1091 /* taking care of that the page is added to LRU while we commit it */
1092 if (likely(!PageLRU(page
)))
1094 spin_lock_irqsave(&zone
->lru_lock
, flags
);
1095 /* link when the page is linked to LRU but page_cgroup isn't */
1096 if (PageLRU(page
) && !PageCgroupAcctLRU(pc
))
1097 mem_cgroup_add_lru_list(page
, page_lru(page
));
1098 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
1102 void mem_cgroup_move_lists(struct page
*page
,
1103 enum lru_list from
, enum lru_list to
)
1105 if (mem_cgroup_disabled())
1107 mem_cgroup_del_lru_list(page
, from
);
1108 mem_cgroup_add_lru_list(page
, to
);
1112 * Checks whether given mem is same or in the root_mem's
1115 static bool mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_mem
,
1116 struct mem_cgroup
*mem
)
1118 if (root_mem
!= mem
) {
1119 return (root_mem
->use_hierarchy
&&
1120 css_is_ancestor(&mem
->css
, &root_mem
->css
));
1126 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*mem
)
1129 struct mem_cgroup
*curr
= NULL
;
1130 struct task_struct
*p
;
1132 p
= find_lock_task_mm(task
);
1135 curr
= try_get_mem_cgroup_from_mm(p
->mm
);
1140 * We should check use_hierarchy of "mem" not "curr". Because checking
1141 * use_hierarchy of "curr" here make this function true if hierarchy is
1142 * enabled in "curr" and "curr" is a child of "mem" in *cgroup*
1143 * hierarchy(even if use_hierarchy is disabled in "mem").
1145 ret
= mem_cgroup_same_or_subtree(mem
, curr
);
1146 css_put(&curr
->css
);
1150 static int calc_inactive_ratio(struct mem_cgroup
*memcg
, unsigned long *present_pages
)
1152 unsigned long active
;
1153 unsigned long inactive
;
1155 unsigned long inactive_ratio
;
1157 inactive
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_INACTIVE_ANON
));
1158 active
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_ACTIVE_ANON
));
1160 gb
= (inactive
+ active
) >> (30 - PAGE_SHIFT
);
1162 inactive_ratio
= int_sqrt(10 * gb
);
1166 if (present_pages
) {
1167 present_pages
[0] = inactive
;
1168 present_pages
[1] = active
;
1171 return inactive_ratio
;
1174 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup
*memcg
)
1176 unsigned long active
;
1177 unsigned long inactive
;
1178 unsigned long present_pages
[2];
1179 unsigned long inactive_ratio
;
1181 inactive_ratio
= calc_inactive_ratio(memcg
, present_pages
);
1183 inactive
= present_pages
[0];
1184 active
= present_pages
[1];
1186 if (inactive
* inactive_ratio
< active
)
1192 int mem_cgroup_inactive_file_is_low(struct mem_cgroup
*memcg
)
1194 unsigned long active
;
1195 unsigned long inactive
;
1197 inactive
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_INACTIVE_FILE
));
1198 active
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_ACTIVE_FILE
));
1200 return (active
> inactive
);
1203 struct zone_reclaim_stat
*mem_cgroup_get_reclaim_stat(struct mem_cgroup
*memcg
,
1206 int nid
= zone_to_nid(zone
);
1207 int zid
= zone_idx(zone
);
1208 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
1210 return &mz
->reclaim_stat
;
1213 struct zone_reclaim_stat
*
1214 mem_cgroup_get_reclaim_stat_from_page(struct page
*page
)
1216 struct page_cgroup
*pc
;
1217 struct mem_cgroup_per_zone
*mz
;
1219 if (mem_cgroup_disabled())
1222 pc
= lookup_page_cgroup(page
);
1223 if (!PageCgroupUsed(pc
))
1225 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1227 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
1228 return &mz
->reclaim_stat
;
1231 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan
,
1232 struct list_head
*dst
,
1233 unsigned long *scanned
, int order
,
1234 int mode
, struct zone
*z
,
1235 struct mem_cgroup
*mem_cont
,
1236 int active
, int file
)
1238 unsigned long nr_taken
= 0;
1242 struct list_head
*src
;
1243 struct page_cgroup
*pc
, *tmp
;
1244 int nid
= zone_to_nid(z
);
1245 int zid
= zone_idx(z
);
1246 struct mem_cgroup_per_zone
*mz
;
1247 int lru
= LRU_FILE
* file
+ active
;
1251 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
1252 src
= &mz
->lists
[lru
];
1255 list_for_each_entry_safe_reverse(pc
, tmp
, src
, lru
) {
1256 if (scan
>= nr_to_scan
)
1259 if (unlikely(!PageCgroupUsed(pc
)))
1262 page
= lookup_cgroup_page(pc
);
1264 if (unlikely(!PageLRU(page
)))
1268 ret
= __isolate_lru_page(page
, mode
, file
);
1271 list_move(&page
->lru
, dst
);
1272 mem_cgroup_del_lru(page
);
1273 nr_taken
+= hpage_nr_pages(page
);
1276 /* we don't affect global LRU but rotate in our LRU */
1277 mem_cgroup_rotate_lru_list(page
, page_lru(page
));
1286 trace_mm_vmscan_memcg_isolate(0, nr_to_scan
, scan
, nr_taken
,
1292 #define mem_cgroup_from_res_counter(counter, member) \
1293 container_of(counter, struct mem_cgroup, member)
1296 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1297 * @mem: the memory cgroup
1299 * Returns the maximum amount of memory @mem can be charged with, in
1302 static unsigned long mem_cgroup_margin(struct mem_cgroup
*mem
)
1304 unsigned long long margin
;
1306 margin
= res_counter_margin(&mem
->res
);
1307 if (do_swap_account
)
1308 margin
= min(margin
, res_counter_margin(&mem
->memsw
));
1309 return margin
>> PAGE_SHIFT
;
1312 int mem_cgroup_swappiness(struct mem_cgroup
*memcg
)
1314 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
1317 if (cgrp
->parent
== NULL
)
1318 return vm_swappiness
;
1320 return memcg
->swappiness
;
1323 static void mem_cgroup_start_move(struct mem_cgroup
*mem
)
1328 spin_lock(&mem
->pcp_counter_lock
);
1329 for_each_online_cpu(cpu
)
1330 per_cpu(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) += 1;
1331 mem
->nocpu_base
.count
[MEM_CGROUP_ON_MOVE
] += 1;
1332 spin_unlock(&mem
->pcp_counter_lock
);
1338 static void mem_cgroup_end_move(struct mem_cgroup
*mem
)
1345 spin_lock(&mem
->pcp_counter_lock
);
1346 for_each_online_cpu(cpu
)
1347 per_cpu(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) -= 1;
1348 mem
->nocpu_base
.count
[MEM_CGROUP_ON_MOVE
] -= 1;
1349 spin_unlock(&mem
->pcp_counter_lock
);
1353 * 2 routines for checking "mem" is under move_account() or not.
1355 * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used
1356 * for avoiding race in accounting. If true,
1357 * pc->mem_cgroup may be overwritten.
1359 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1360 * under hierarchy of moving cgroups. This is for
1361 * waiting at hith-memory prressure caused by "move".
1364 static bool mem_cgroup_stealed(struct mem_cgroup
*mem
)
1366 VM_BUG_ON(!rcu_read_lock_held());
1367 return this_cpu_read(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
]) > 0;
1370 static bool mem_cgroup_under_move(struct mem_cgroup
*mem
)
1372 struct mem_cgroup
*from
;
1373 struct mem_cgroup
*to
;
1376 * Unlike task_move routines, we access mc.to, mc.from not under
1377 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1379 spin_lock(&mc
.lock
);
1385 ret
= mem_cgroup_same_or_subtree(mem
, from
)
1386 || mem_cgroup_same_or_subtree(mem
, to
);
1388 spin_unlock(&mc
.lock
);
1392 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*mem
)
1394 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1395 if (mem_cgroup_under_move(mem
)) {
1397 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1398 /* moving charge context might have finished. */
1401 finish_wait(&mc
.waitq
, &wait
);
1409 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1410 * @memcg: The memory cgroup that went over limit
1411 * @p: Task that is going to be killed
1413 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1416 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1418 struct cgroup
*task_cgrp
;
1419 struct cgroup
*mem_cgrp
;
1421 * Need a buffer in BSS, can't rely on allocations. The code relies
1422 * on the assumption that OOM is serialized for memory controller.
1423 * If this assumption is broken, revisit this code.
1425 static char memcg_name
[PATH_MAX
];
1434 mem_cgrp
= memcg
->css
.cgroup
;
1435 task_cgrp
= task_cgroup(p
, mem_cgroup_subsys_id
);
1437 ret
= cgroup_path(task_cgrp
, memcg_name
, PATH_MAX
);
1440 * Unfortunately, we are unable to convert to a useful name
1441 * But we'll still print out the usage information
1448 printk(KERN_INFO
"Task in %s killed", memcg_name
);
1451 ret
= cgroup_path(mem_cgrp
, memcg_name
, PATH_MAX
);
1459 * Continues from above, so we don't need an KERN_ level
1461 printk(KERN_CONT
" as a result of limit of %s\n", memcg_name
);
1464 printk(KERN_INFO
"memory: usage %llukB, limit %llukB, failcnt %llu\n",
1465 res_counter_read_u64(&memcg
->res
, RES_USAGE
) >> 10,
1466 res_counter_read_u64(&memcg
->res
, RES_LIMIT
) >> 10,
1467 res_counter_read_u64(&memcg
->res
, RES_FAILCNT
));
1468 printk(KERN_INFO
"memory+swap: usage %llukB, limit %llukB, "
1470 res_counter_read_u64(&memcg
->memsw
, RES_USAGE
) >> 10,
1471 res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
) >> 10,
1472 res_counter_read_u64(&memcg
->memsw
, RES_FAILCNT
));
1476 * This function returns the number of memcg under hierarchy tree. Returns
1477 * 1(self count) if no children.
1479 static int mem_cgroup_count_children(struct mem_cgroup
*mem
)
1482 struct mem_cgroup
*iter
;
1484 for_each_mem_cgroup_tree(iter
, mem
)
1490 * Return the memory (and swap, if configured) limit for a memcg.
1492 u64
mem_cgroup_get_limit(struct mem_cgroup
*memcg
)
1497 limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
1498 limit
+= total_swap_pages
<< PAGE_SHIFT
;
1500 memsw
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
1502 * If memsw is finite and limits the amount of swap space available
1503 * to this memcg, return that limit.
1505 return min(limit
, memsw
);
1509 * Visit the first child (need not be the first child as per the ordering
1510 * of the cgroup list, since we track last_scanned_child) of @mem and use
1511 * that to reclaim free pages from.
1513 static struct mem_cgroup
*
1514 mem_cgroup_select_victim(struct mem_cgroup
*root_mem
)
1516 struct mem_cgroup
*ret
= NULL
;
1517 struct cgroup_subsys_state
*css
;
1520 if (!root_mem
->use_hierarchy
) {
1521 css_get(&root_mem
->css
);
1527 nextid
= root_mem
->last_scanned_child
+ 1;
1528 css
= css_get_next(&mem_cgroup_subsys
, nextid
, &root_mem
->css
,
1530 if (css
&& css_tryget(css
))
1531 ret
= container_of(css
, struct mem_cgroup
, css
);
1534 /* Updates scanning parameter */
1536 /* this means start scan from ID:1 */
1537 root_mem
->last_scanned_child
= 0;
1539 root_mem
->last_scanned_child
= found
;
1546 * test_mem_cgroup_node_reclaimable
1547 * @mem: the target memcg
1548 * @nid: the node ID to be checked.
1549 * @noswap : specify true here if the user wants flle only information.
1551 * This function returns whether the specified memcg contains any
1552 * reclaimable pages on a node. Returns true if there are any reclaimable
1553 * pages in the node.
1555 static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup
*mem
,
1556 int nid
, bool noswap
)
1558 if (mem_cgroup_node_nr_lru_pages(mem
, nid
, LRU_ALL_FILE
))
1560 if (noswap
|| !total_swap_pages
)
1562 if (mem_cgroup_node_nr_lru_pages(mem
, nid
, LRU_ALL_ANON
))
1567 #if MAX_NUMNODES > 1
1570 * Always updating the nodemask is not very good - even if we have an empty
1571 * list or the wrong list here, we can start from some node and traverse all
1572 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1575 static void mem_cgroup_may_update_nodemask(struct mem_cgroup
*mem
)
1579 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1580 * pagein/pageout changes since the last update.
1582 if (!atomic_read(&mem
->numainfo_events
))
1584 if (atomic_inc_return(&mem
->numainfo_updating
) > 1)
1587 /* make a nodemask where this memcg uses memory from */
1588 mem
->scan_nodes
= node_states
[N_HIGH_MEMORY
];
1590 for_each_node_mask(nid
, node_states
[N_HIGH_MEMORY
]) {
1592 if (!test_mem_cgroup_node_reclaimable(mem
, nid
, false))
1593 node_clear(nid
, mem
->scan_nodes
);
1596 atomic_set(&mem
->numainfo_events
, 0);
1597 atomic_set(&mem
->numainfo_updating
, 0);
1601 * Selecting a node where we start reclaim from. Because what we need is just
1602 * reducing usage counter, start from anywhere is O,K. Considering
1603 * memory reclaim from current node, there are pros. and cons.
1605 * Freeing memory from current node means freeing memory from a node which
1606 * we'll use or we've used. So, it may make LRU bad. And if several threads
1607 * hit limits, it will see a contention on a node. But freeing from remote
1608 * node means more costs for memory reclaim because of memory latency.
1610 * Now, we use round-robin. Better algorithm is welcomed.
1612 int mem_cgroup_select_victim_node(struct mem_cgroup
*mem
)
1616 mem_cgroup_may_update_nodemask(mem
);
1617 node
= mem
->last_scanned_node
;
1619 node
= next_node(node
, mem
->scan_nodes
);
1620 if (node
== MAX_NUMNODES
)
1621 node
= first_node(mem
->scan_nodes
);
1623 * We call this when we hit limit, not when pages are added to LRU.
1624 * No LRU may hold pages because all pages are UNEVICTABLE or
1625 * memcg is too small and all pages are not on LRU. In that case,
1626 * we use curret node.
1628 if (unlikely(node
== MAX_NUMNODES
))
1629 node
= numa_node_id();
1631 mem
->last_scanned_node
= node
;
1636 * Check all nodes whether it contains reclaimable pages or not.
1637 * For quick scan, we make use of scan_nodes. This will allow us to skip
1638 * unused nodes. But scan_nodes is lazily updated and may not cotain
1639 * enough new information. We need to do double check.
1641 bool mem_cgroup_reclaimable(struct mem_cgroup
*mem
, bool noswap
)
1646 * quick check...making use of scan_node.
1647 * We can skip unused nodes.
1649 if (!nodes_empty(mem
->scan_nodes
)) {
1650 for (nid
= first_node(mem
->scan_nodes
);
1652 nid
= next_node(nid
, mem
->scan_nodes
)) {
1654 if (test_mem_cgroup_node_reclaimable(mem
, nid
, noswap
))
1659 * Check rest of nodes.
1661 for_each_node_state(nid
, N_HIGH_MEMORY
) {
1662 if (node_isset(nid
, mem
->scan_nodes
))
1664 if (test_mem_cgroup_node_reclaimable(mem
, nid
, noswap
))
1671 int mem_cgroup_select_victim_node(struct mem_cgroup
*mem
)
1676 bool mem_cgroup_reclaimable(struct mem_cgroup
*mem
, bool noswap
)
1678 return test_mem_cgroup_node_reclaimable(mem
, 0, noswap
);
1682 static void __mem_cgroup_record_scanstat(unsigned long *stats
,
1683 struct memcg_scanrecord
*rec
)
1686 stats
[SCAN
] += rec
->nr_scanned
[0] + rec
->nr_scanned
[1];
1687 stats
[SCAN_ANON
] += rec
->nr_scanned
[0];
1688 stats
[SCAN_FILE
] += rec
->nr_scanned
[1];
1690 stats
[ROTATE
] += rec
->nr_rotated
[0] + rec
->nr_rotated
[1];
1691 stats
[ROTATE_ANON
] += rec
->nr_rotated
[0];
1692 stats
[ROTATE_FILE
] += rec
->nr_rotated
[1];
1694 stats
[FREED
] += rec
->nr_freed
[0] + rec
->nr_freed
[1];
1695 stats
[FREED_ANON
] += rec
->nr_freed
[0];
1696 stats
[FREED_FILE
] += rec
->nr_freed
[1];
1698 stats
[ELAPSED
] += rec
->elapsed
;
1701 static void mem_cgroup_record_scanstat(struct memcg_scanrecord
*rec
)
1703 struct mem_cgroup
*mem
;
1704 int context
= rec
->context
;
1706 if (context
>= NR_SCAN_CONTEXT
)
1710 spin_lock(&mem
->scanstat
.lock
);
1711 __mem_cgroup_record_scanstat(mem
->scanstat
.stats
[context
], rec
);
1712 spin_unlock(&mem
->scanstat
.lock
);
1715 spin_lock(&mem
->scanstat
.lock
);
1716 __mem_cgroup_record_scanstat(mem
->scanstat
.rootstats
[context
], rec
);
1717 spin_unlock(&mem
->scanstat
.lock
);
1721 * Scan the hierarchy if needed to reclaim memory. We remember the last child
1722 * we reclaimed from, so that we don't end up penalizing one child extensively
1723 * based on its position in the children list.
1725 * root_mem is the original ancestor that we've been reclaim from.
1727 * We give up and return to the caller when we visit root_mem twice.
1728 * (other groups can be removed while we're walking....)
1730 * If shrink==true, for avoiding to free too much, this returns immedieately.
1732 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup
*root_mem
,
1735 unsigned long reclaim_options
,
1736 unsigned long *total_scanned
)
1738 struct mem_cgroup
*victim
;
1741 bool noswap
= reclaim_options
& MEM_CGROUP_RECLAIM_NOSWAP
;
1742 bool shrink
= reclaim_options
& MEM_CGROUP_RECLAIM_SHRINK
;
1743 bool check_soft
= reclaim_options
& MEM_CGROUP_RECLAIM_SOFT
;
1744 struct memcg_scanrecord rec
;
1745 unsigned long excess
;
1746 unsigned long scanned
;
1748 excess
= res_counter_soft_limit_excess(&root_mem
->res
) >> PAGE_SHIFT
;
1750 /* If memsw_is_minimum==1, swap-out is of-no-use. */
1751 if (!check_soft
&& !shrink
&& root_mem
->memsw_is_minimum
)
1755 rec
.context
= SCAN_BY_SHRINK
;
1756 else if (check_soft
)
1757 rec
.context
= SCAN_BY_SYSTEM
;
1759 rec
.context
= SCAN_BY_LIMIT
;
1761 rec
.root
= root_mem
;
1764 victim
= mem_cgroup_select_victim(root_mem
);
1765 if (victim
== root_mem
) {
1768 * We are not draining per cpu cached charges during
1769 * soft limit reclaim because global reclaim doesn't
1770 * care about charges. It tries to free some memory and
1771 * charges will not give any.
1773 if (!check_soft
&& loop
>= 1)
1774 drain_all_stock_async(root_mem
);
1777 * If we have not been able to reclaim
1778 * anything, it might because there are
1779 * no reclaimable pages under this hierarchy
1781 if (!check_soft
|| !total
) {
1782 css_put(&victim
->css
);
1786 * We want to do more targeted reclaim.
1787 * excess >> 2 is not to excessive so as to
1788 * reclaim too much, nor too less that we keep
1789 * coming back to reclaim from this cgroup
1791 if (total
>= (excess
>> 2) ||
1792 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
)) {
1793 css_put(&victim
->css
);
1798 if (!mem_cgroup_reclaimable(victim
, noswap
)) {
1799 /* this cgroup's local usage == 0 */
1800 css_put(&victim
->css
);
1804 rec
.nr_scanned
[0] = 0;
1805 rec
.nr_scanned
[1] = 0;
1806 rec
.nr_rotated
[0] = 0;
1807 rec
.nr_rotated
[1] = 0;
1808 rec
.nr_freed
[0] = 0;
1809 rec
.nr_freed
[1] = 0;
1811 /* we use swappiness of local cgroup */
1813 ret
= mem_cgroup_shrink_node_zone(victim
, gfp_mask
,
1814 noswap
, zone
, &rec
, &scanned
);
1815 *total_scanned
+= scanned
;
1817 ret
= try_to_free_mem_cgroup_pages(victim
, gfp_mask
,
1819 mem_cgroup_record_scanstat(&rec
);
1820 css_put(&victim
->css
);
1822 * At shrinking usage, we can't check we should stop here or
1823 * reclaim more. It's depends on callers. last_scanned_child
1824 * will work enough for keeping fairness under tree.
1830 if (!res_counter_soft_limit_excess(&root_mem
->res
))
1832 } else if (mem_cgroup_margin(root_mem
))
1839 * Check OOM-Killer is already running under our hierarchy.
1840 * If someone is running, return false.
1841 * Has to be called with memcg_oom_lock
1843 static bool mem_cgroup_oom_lock(struct mem_cgroup
*mem
)
1845 int lock_count
= -1;
1846 struct mem_cgroup
*iter
, *failed
= NULL
;
1849 for_each_mem_cgroup_tree_cond(iter
, mem
, cond
) {
1850 bool locked
= iter
->oom_lock
;
1852 iter
->oom_lock
= true;
1853 if (lock_count
== -1)
1854 lock_count
= iter
->oom_lock
;
1855 else if (lock_count
!= locked
) {
1857 * this subtree of our hierarchy is already locked
1858 * so we cannot give a lock.
1870 * OK, we failed to lock the whole subtree so we have to clean up
1871 * what we set up to the failing subtree
1874 for_each_mem_cgroup_tree_cond(iter
, mem
, cond
) {
1875 if (iter
== failed
) {
1879 iter
->oom_lock
= false;
1886 * Has to be called with memcg_oom_lock
1888 static int mem_cgroup_oom_unlock(struct mem_cgroup
*mem
)
1890 struct mem_cgroup
*iter
;
1892 for_each_mem_cgroup_tree(iter
, mem
)
1893 iter
->oom_lock
= false;
1897 static void mem_cgroup_mark_under_oom(struct mem_cgroup
*mem
)
1899 struct mem_cgroup
*iter
;
1901 for_each_mem_cgroup_tree(iter
, mem
)
1902 atomic_inc(&iter
->under_oom
);
1905 static void mem_cgroup_unmark_under_oom(struct mem_cgroup
*mem
)
1907 struct mem_cgroup
*iter
;
1910 * When a new child is created while the hierarchy is under oom,
1911 * mem_cgroup_oom_lock() may not be called. We have to use
1912 * atomic_add_unless() here.
1914 for_each_mem_cgroup_tree(iter
, mem
)
1915 atomic_add_unless(&iter
->under_oom
, -1, 0);
1918 static DEFINE_SPINLOCK(memcg_oom_lock
);
1919 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1921 struct oom_wait_info
{
1922 struct mem_cgroup
*mem
;
1926 static int memcg_oom_wake_function(wait_queue_t
*wait
,
1927 unsigned mode
, int sync
, void *arg
)
1929 struct mem_cgroup
*wake_mem
= (struct mem_cgroup
*)arg
,
1931 struct oom_wait_info
*oom_wait_info
;
1933 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1934 oom_wait_mem
= oom_wait_info
->mem
;
1937 * Both of oom_wait_info->mem and wake_mem are stable under us.
1938 * Then we can use css_is_ancestor without taking care of RCU.
1940 if (!mem_cgroup_same_or_subtree(oom_wait_mem
, wake_mem
)
1941 && !mem_cgroup_same_or_subtree(wake_mem
, oom_wait_mem
))
1943 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1946 static void memcg_wakeup_oom(struct mem_cgroup
*mem
)
1948 /* for filtering, pass "mem" as argument. */
1949 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, mem
);
1952 static void memcg_oom_recover(struct mem_cgroup
*mem
)
1954 if (mem
&& atomic_read(&mem
->under_oom
))
1955 memcg_wakeup_oom(mem
);
1959 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1961 bool mem_cgroup_handle_oom(struct mem_cgroup
*mem
, gfp_t mask
)
1963 struct oom_wait_info owait
;
1964 bool locked
, need_to_kill
;
1967 owait
.wait
.flags
= 0;
1968 owait
.wait
.func
= memcg_oom_wake_function
;
1969 owait
.wait
.private = current
;
1970 INIT_LIST_HEAD(&owait
.wait
.task_list
);
1971 need_to_kill
= true;
1972 mem_cgroup_mark_under_oom(mem
);
1974 /* At first, try to OOM lock hierarchy under mem.*/
1975 spin_lock(&memcg_oom_lock
);
1976 locked
= mem_cgroup_oom_lock(mem
);
1978 * Even if signal_pending(), we can't quit charge() loop without
1979 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1980 * under OOM is always welcomed, use TASK_KILLABLE here.
1982 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1983 if (!locked
|| mem
->oom_kill_disable
)
1984 need_to_kill
= false;
1986 mem_cgroup_oom_notify(mem
);
1987 spin_unlock(&memcg_oom_lock
);
1990 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1991 mem_cgroup_out_of_memory(mem
, mask
);
1994 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1996 spin_lock(&memcg_oom_lock
);
1998 mem_cgroup_oom_unlock(mem
);
1999 memcg_wakeup_oom(mem
);
2000 spin_unlock(&memcg_oom_lock
);
2002 mem_cgroup_unmark_under_oom(mem
);
2004 if (test_thread_flag(TIF_MEMDIE
) || fatal_signal_pending(current
))
2006 /* Give chance to dying process */
2007 schedule_timeout(1);
2012 * Currently used to update mapped file statistics, but the routine can be
2013 * generalized to update other statistics as well.
2015 * Notes: Race condition
2017 * We usually use page_cgroup_lock() for accessing page_cgroup member but
2018 * it tends to be costly. But considering some conditions, we doesn't need
2019 * to do so _always_.
2021 * Considering "charge", lock_page_cgroup() is not required because all
2022 * file-stat operations happen after a page is attached to radix-tree. There
2023 * are no race with "charge".
2025 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
2026 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
2027 * if there are race with "uncharge". Statistics itself is properly handled
2030 * Considering "move", this is an only case we see a race. To make the race
2031 * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are
2032 * possibility of race condition. If there is, we take a lock.
2035 void mem_cgroup_update_page_stat(struct page
*page
,
2036 enum mem_cgroup_page_stat_item idx
, int val
)
2038 struct mem_cgroup
*mem
;
2039 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
2040 bool need_unlock
= false;
2041 unsigned long uninitialized_var(flags
);
2047 mem
= pc
->mem_cgroup
;
2048 if (unlikely(!mem
|| !PageCgroupUsed(pc
)))
2050 /* pc->mem_cgroup is unstable ? */
2051 if (unlikely(mem_cgroup_stealed(mem
)) || PageTransHuge(page
)) {
2052 /* take a lock against to access pc->mem_cgroup */
2053 move_lock_page_cgroup(pc
, &flags
);
2055 mem
= pc
->mem_cgroup
;
2056 if (!mem
|| !PageCgroupUsed(pc
))
2061 case MEMCG_NR_FILE_MAPPED
:
2063 SetPageCgroupFileMapped(pc
);
2064 else if (!page_mapped(page
))
2065 ClearPageCgroupFileMapped(pc
);
2066 idx
= MEM_CGROUP_STAT_FILE_MAPPED
;
2072 this_cpu_add(mem
->stat
->count
[idx
], val
);
2075 if (unlikely(need_unlock
))
2076 move_unlock_page_cgroup(pc
, &flags
);
2080 EXPORT_SYMBOL(mem_cgroup_update_page_stat
);
2083 * size of first charge trial. "32" comes from vmscan.c's magic value.
2084 * TODO: maybe necessary to use big numbers in big irons.
2086 #define CHARGE_BATCH 32U
2087 struct memcg_stock_pcp
{
2088 struct mem_cgroup
*cached
; /* this never be root cgroup */
2089 unsigned int nr_pages
;
2090 struct work_struct work
;
2091 unsigned long flags
;
2092 #define FLUSHING_CACHED_CHARGE (0)
2094 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
2095 static DEFINE_MUTEX(percpu_charge_mutex
);
2098 * Try to consume stocked charge on this cpu. If success, one page is consumed
2099 * from local stock and true is returned. If the stock is 0 or charges from a
2100 * cgroup which is not current target, returns false. This stock will be
2103 static bool consume_stock(struct mem_cgroup
*mem
)
2105 struct memcg_stock_pcp
*stock
;
2108 stock
= &get_cpu_var(memcg_stock
);
2109 if (mem
== stock
->cached
&& stock
->nr_pages
)
2111 else /* need to call res_counter_charge */
2113 put_cpu_var(memcg_stock
);
2118 * Returns stocks cached in percpu to res_counter and reset cached information.
2120 static void drain_stock(struct memcg_stock_pcp
*stock
)
2122 struct mem_cgroup
*old
= stock
->cached
;
2124 if (stock
->nr_pages
) {
2125 unsigned long bytes
= stock
->nr_pages
* PAGE_SIZE
;
2127 res_counter_uncharge(&old
->res
, bytes
);
2128 if (do_swap_account
)
2129 res_counter_uncharge(&old
->memsw
, bytes
);
2130 stock
->nr_pages
= 0;
2132 stock
->cached
= NULL
;
2136 * This must be called under preempt disabled or must be called by
2137 * a thread which is pinned to local cpu.
2139 static void drain_local_stock(struct work_struct
*dummy
)
2141 struct memcg_stock_pcp
*stock
= &__get_cpu_var(memcg_stock
);
2143 clear_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
);
2147 * Cache charges(val) which is from res_counter, to local per_cpu area.
2148 * This will be consumed by consume_stock() function, later.
2150 static void refill_stock(struct mem_cgroup
*mem
, unsigned int nr_pages
)
2152 struct memcg_stock_pcp
*stock
= &get_cpu_var(memcg_stock
);
2154 if (stock
->cached
!= mem
) { /* reset if necessary */
2156 stock
->cached
= mem
;
2158 stock
->nr_pages
+= nr_pages
;
2159 put_cpu_var(memcg_stock
);
2163 * Drains all per-CPU charge caches for given root_mem resp. subtree
2164 * of the hierarchy under it. sync flag says whether we should block
2165 * until the work is done.
2167 static void drain_all_stock(struct mem_cgroup
*root_mem
, bool sync
)
2171 /* Notify other cpus that system-wide "drain" is running */
2174 * Get a hint for avoiding draining charges on the current cpu,
2175 * which must be exhausted by our charging. It is not required that
2176 * this be a precise check, so we use raw_smp_processor_id() instead of
2177 * getcpu()/putcpu().
2179 curcpu
= raw_smp_processor_id();
2180 for_each_online_cpu(cpu
) {
2181 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2182 struct mem_cgroup
*mem
;
2184 mem
= stock
->cached
;
2185 if (!mem
|| !stock
->nr_pages
)
2187 if (!mem_cgroup_same_or_subtree(root_mem
, mem
))
2189 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
)) {
2191 drain_local_stock(&stock
->work
);
2193 schedule_work_on(cpu
, &stock
->work
);
2200 for_each_online_cpu(cpu
) {
2201 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2202 if (test_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
))
2203 flush_work(&stock
->work
);
2210 * Tries to drain stocked charges in other cpus. This function is asynchronous
2211 * and just put a work per cpu for draining localy on each cpu. Caller can
2212 * expects some charges will be back to res_counter later but cannot wait for
2215 static void drain_all_stock_async(struct mem_cgroup
*root_mem
)
2218 * If someone calls draining, avoid adding more kworker runs.
2220 if (!mutex_trylock(&percpu_charge_mutex
))
2222 drain_all_stock(root_mem
, false);
2223 mutex_unlock(&percpu_charge_mutex
);
2226 /* This is a synchronous drain interface. */
2227 static void drain_all_stock_sync(struct mem_cgroup
*root_mem
)
2229 /* called when force_empty is called */
2230 mutex_lock(&percpu_charge_mutex
);
2231 drain_all_stock(root_mem
, true);
2232 mutex_unlock(&percpu_charge_mutex
);
2236 * This function drains percpu counter value from DEAD cpu and
2237 * move it to local cpu. Note that this function can be preempted.
2239 static void mem_cgroup_drain_pcp_counter(struct mem_cgroup
*mem
, int cpu
)
2243 spin_lock(&mem
->pcp_counter_lock
);
2244 for (i
= 0; i
< MEM_CGROUP_STAT_DATA
; i
++) {
2245 long x
= per_cpu(mem
->stat
->count
[i
], cpu
);
2247 per_cpu(mem
->stat
->count
[i
], cpu
) = 0;
2248 mem
->nocpu_base
.count
[i
] += x
;
2250 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
2251 unsigned long x
= per_cpu(mem
->stat
->events
[i
], cpu
);
2253 per_cpu(mem
->stat
->events
[i
], cpu
) = 0;
2254 mem
->nocpu_base
.events
[i
] += x
;
2256 /* need to clear ON_MOVE value, works as a kind of lock. */
2257 per_cpu(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) = 0;
2258 spin_unlock(&mem
->pcp_counter_lock
);
2261 static void synchronize_mem_cgroup_on_move(struct mem_cgroup
*mem
, int cpu
)
2263 int idx
= MEM_CGROUP_ON_MOVE
;
2265 spin_lock(&mem
->pcp_counter_lock
);
2266 per_cpu(mem
->stat
->count
[idx
], cpu
) = mem
->nocpu_base
.count
[idx
];
2267 spin_unlock(&mem
->pcp_counter_lock
);
2270 static int __cpuinit
memcg_cpu_hotplug_callback(struct notifier_block
*nb
,
2271 unsigned long action
,
2274 int cpu
= (unsigned long)hcpu
;
2275 struct memcg_stock_pcp
*stock
;
2276 struct mem_cgroup
*iter
;
2278 if ((action
== CPU_ONLINE
)) {
2279 for_each_mem_cgroup_all(iter
)
2280 synchronize_mem_cgroup_on_move(iter
, cpu
);
2284 if ((action
!= CPU_DEAD
) || action
!= CPU_DEAD_FROZEN
)
2287 for_each_mem_cgroup_all(iter
)
2288 mem_cgroup_drain_pcp_counter(iter
, cpu
);
2290 stock
= &per_cpu(memcg_stock
, cpu
);
2296 /* See __mem_cgroup_try_charge() for details */
2298 CHARGE_OK
, /* success */
2299 CHARGE_RETRY
, /* need to retry but retry is not bad */
2300 CHARGE_NOMEM
, /* we can't do more. return -ENOMEM */
2301 CHARGE_WOULDBLOCK
, /* GFP_WAIT wasn't set and no enough res. */
2302 CHARGE_OOM_DIE
, /* the current is killed because of OOM */
2305 static int mem_cgroup_do_charge(struct mem_cgroup
*mem
, gfp_t gfp_mask
,
2306 unsigned int nr_pages
, bool oom_check
)
2308 unsigned long csize
= nr_pages
* PAGE_SIZE
;
2309 struct mem_cgroup
*mem_over_limit
;
2310 struct res_counter
*fail_res
;
2311 unsigned long flags
= 0;
2314 ret
= res_counter_charge(&mem
->res
, csize
, &fail_res
);
2317 if (!do_swap_account
)
2319 ret
= res_counter_charge(&mem
->memsw
, csize
, &fail_res
);
2323 res_counter_uncharge(&mem
->res
, csize
);
2324 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, memsw
);
2325 flags
|= MEM_CGROUP_RECLAIM_NOSWAP
;
2327 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, res
);
2329 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2330 * of regular pages (CHARGE_BATCH), or a single regular page (1).
2332 * Never reclaim on behalf of optional batching, retry with a
2333 * single page instead.
2335 if (nr_pages
== CHARGE_BATCH
)
2336 return CHARGE_RETRY
;
2338 if (!(gfp_mask
& __GFP_WAIT
))
2339 return CHARGE_WOULDBLOCK
;
2341 ret
= mem_cgroup_hierarchical_reclaim(mem_over_limit
, NULL
,
2342 gfp_mask
, flags
, NULL
);
2343 if (mem_cgroup_margin(mem_over_limit
) >= nr_pages
)
2344 return CHARGE_RETRY
;
2346 * Even though the limit is exceeded at this point, reclaim
2347 * may have been able to free some pages. Retry the charge
2348 * before killing the task.
2350 * Only for regular pages, though: huge pages are rather
2351 * unlikely to succeed so close to the limit, and we fall back
2352 * to regular pages anyway in case of failure.
2354 if (nr_pages
== 1 && ret
)
2355 return CHARGE_RETRY
;
2358 * At task move, charge accounts can be doubly counted. So, it's
2359 * better to wait until the end of task_move if something is going on.
2361 if (mem_cgroup_wait_acct_move(mem_over_limit
))
2362 return CHARGE_RETRY
;
2364 /* If we don't need to call oom-killer at el, return immediately */
2366 return CHARGE_NOMEM
;
2368 if (!mem_cgroup_handle_oom(mem_over_limit
, gfp_mask
))
2369 return CHARGE_OOM_DIE
;
2371 return CHARGE_RETRY
;
2375 * Unlike exported interface, "oom" parameter is added. if oom==true,
2376 * oom-killer can be invoked.
2378 static int __mem_cgroup_try_charge(struct mm_struct
*mm
,
2380 unsigned int nr_pages
,
2381 struct mem_cgroup
**memcg
,
2384 unsigned int batch
= max(CHARGE_BATCH
, nr_pages
);
2385 int nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2386 struct mem_cgroup
*mem
= NULL
;
2390 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2391 * in system level. So, allow to go ahead dying process in addition to
2394 if (unlikely(test_thread_flag(TIF_MEMDIE
)
2395 || fatal_signal_pending(current
)))
2399 * We always charge the cgroup the mm_struct belongs to.
2400 * The mm_struct's mem_cgroup changes on task migration if the
2401 * thread group leader migrates. It's possible that mm is not
2402 * set, if so charge the init_mm (happens for pagecache usage).
2407 if (*memcg
) { /* css should be a valid one */
2409 VM_BUG_ON(css_is_removed(&mem
->css
));
2410 if (mem_cgroup_is_root(mem
))
2412 if (nr_pages
== 1 && consume_stock(mem
))
2416 struct task_struct
*p
;
2419 p
= rcu_dereference(mm
->owner
);
2421 * Because we don't have task_lock(), "p" can exit.
2422 * In that case, "mem" can point to root or p can be NULL with
2423 * race with swapoff. Then, we have small risk of mis-accouning.
2424 * But such kind of mis-account by race always happens because
2425 * we don't have cgroup_mutex(). It's overkill and we allo that
2427 * (*) swapoff at el will charge against mm-struct not against
2428 * task-struct. So, mm->owner can be NULL.
2430 mem
= mem_cgroup_from_task(p
);
2431 if (!mem
|| mem_cgroup_is_root(mem
)) {
2435 if (nr_pages
== 1 && consume_stock(mem
)) {
2437 * It seems dagerous to access memcg without css_get().
2438 * But considering how consume_stok works, it's not
2439 * necessary. If consume_stock success, some charges
2440 * from this memcg are cached on this cpu. So, we
2441 * don't need to call css_get()/css_tryget() before
2442 * calling consume_stock().
2447 /* after here, we may be blocked. we need to get refcnt */
2448 if (!css_tryget(&mem
->css
)) {
2458 /* If killed, bypass charge */
2459 if (fatal_signal_pending(current
)) {
2465 if (oom
&& !nr_oom_retries
) {
2467 nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2470 ret
= mem_cgroup_do_charge(mem
, gfp_mask
, batch
, oom_check
);
2474 case CHARGE_RETRY
: /* not in OOM situation but retry */
2479 case CHARGE_WOULDBLOCK
: /* !__GFP_WAIT */
2482 case CHARGE_NOMEM
: /* OOM routine works */
2487 /* If oom, we never return -ENOMEM */
2490 case CHARGE_OOM_DIE
: /* Killed by OOM Killer */
2494 } while (ret
!= CHARGE_OK
);
2496 if (batch
> nr_pages
)
2497 refill_stock(mem
, batch
- nr_pages
);
2511 * Somemtimes we have to undo a charge we got by try_charge().
2512 * This function is for that and do uncharge, put css's refcnt.
2513 * gotten by try_charge().
2515 static void __mem_cgroup_cancel_charge(struct mem_cgroup
*mem
,
2516 unsigned int nr_pages
)
2518 if (!mem_cgroup_is_root(mem
)) {
2519 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2521 res_counter_uncharge(&mem
->res
, bytes
);
2522 if (do_swap_account
)
2523 res_counter_uncharge(&mem
->memsw
, bytes
);
2528 * A helper function to get mem_cgroup from ID. must be called under
2529 * rcu_read_lock(). The caller must check css_is_removed() or some if
2530 * it's concern. (dropping refcnt from swap can be called against removed
2533 static struct mem_cgroup
*mem_cgroup_lookup(unsigned short id
)
2535 struct cgroup_subsys_state
*css
;
2537 /* ID 0 is unused ID */
2540 css
= css_lookup(&mem_cgroup_subsys
, id
);
2543 return container_of(css
, struct mem_cgroup
, css
);
2546 struct mem_cgroup
*try_get_mem_cgroup_from_page(struct page
*page
)
2548 struct mem_cgroup
*mem
= NULL
;
2549 struct page_cgroup
*pc
;
2553 VM_BUG_ON(!PageLocked(page
));
2555 pc
= lookup_page_cgroup(page
);
2556 lock_page_cgroup(pc
);
2557 if (PageCgroupUsed(pc
)) {
2558 mem
= pc
->mem_cgroup
;
2559 if (mem
&& !css_tryget(&mem
->css
))
2561 } else if (PageSwapCache(page
)) {
2562 ent
.val
= page_private(page
);
2563 id
= lookup_swap_cgroup(ent
);
2565 mem
= mem_cgroup_lookup(id
);
2566 if (mem
&& !css_tryget(&mem
->css
))
2570 unlock_page_cgroup(pc
);
2574 static void __mem_cgroup_commit_charge(struct mem_cgroup
*mem
,
2576 unsigned int nr_pages
,
2577 struct page_cgroup
*pc
,
2578 enum charge_type ctype
)
2580 lock_page_cgroup(pc
);
2581 if (unlikely(PageCgroupUsed(pc
))) {
2582 unlock_page_cgroup(pc
);
2583 __mem_cgroup_cancel_charge(mem
, nr_pages
);
2587 * we don't need page_cgroup_lock about tail pages, becase they are not
2588 * accessed by any other context at this point.
2590 pc
->mem_cgroup
= mem
;
2592 * We access a page_cgroup asynchronously without lock_page_cgroup().
2593 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2594 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2595 * before USED bit, we need memory barrier here.
2596 * See mem_cgroup_add_lru_list(), etc.
2600 case MEM_CGROUP_CHARGE_TYPE_CACHE
:
2601 case MEM_CGROUP_CHARGE_TYPE_SHMEM
:
2602 SetPageCgroupCache(pc
);
2603 SetPageCgroupUsed(pc
);
2605 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2606 ClearPageCgroupCache(pc
);
2607 SetPageCgroupUsed(pc
);
2613 mem_cgroup_charge_statistics(mem
, PageCgroupCache(pc
), nr_pages
);
2614 unlock_page_cgroup(pc
);
2616 * "charge_statistics" updated event counter. Then, check it.
2617 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2618 * if they exceeds softlimit.
2620 memcg_check_events(mem
, page
);
2623 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2625 #define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\
2626 (1 << PCG_ACCT_LRU) | (1 << PCG_MIGRATION))
2628 * Because tail pages are not marked as "used", set it. We're under
2629 * zone->lru_lock, 'splitting on pmd' and compund_lock.
2631 void mem_cgroup_split_huge_fixup(struct page
*head
, struct page
*tail
)
2633 struct page_cgroup
*head_pc
= lookup_page_cgroup(head
);
2634 struct page_cgroup
*tail_pc
= lookup_page_cgroup(tail
);
2635 unsigned long flags
;
2637 if (mem_cgroup_disabled())
2640 * We have no races with charge/uncharge but will have races with
2641 * page state accounting.
2643 move_lock_page_cgroup(head_pc
, &flags
);
2645 tail_pc
->mem_cgroup
= head_pc
->mem_cgroup
;
2646 smp_wmb(); /* see __commit_charge() */
2647 if (PageCgroupAcctLRU(head_pc
)) {
2649 struct mem_cgroup_per_zone
*mz
;
2652 * LRU flags cannot be copied because we need to add tail
2653 *.page to LRU by generic call and our hook will be called.
2654 * We hold lru_lock, then, reduce counter directly.
2656 lru
= page_lru(head
);
2657 mz
= page_cgroup_zoneinfo(head_pc
->mem_cgroup
, head
);
2658 MEM_CGROUP_ZSTAT(mz
, lru
) -= 1;
2660 tail_pc
->flags
= head_pc
->flags
& ~PCGF_NOCOPY_AT_SPLIT
;
2661 move_unlock_page_cgroup(head_pc
, &flags
);
2666 * mem_cgroup_move_account - move account of the page
2668 * @nr_pages: number of regular pages (>1 for huge pages)
2669 * @pc: page_cgroup of the page.
2670 * @from: mem_cgroup which the page is moved from.
2671 * @to: mem_cgroup which the page is moved to. @from != @to.
2672 * @uncharge: whether we should call uncharge and css_put against @from.
2674 * The caller must confirm following.
2675 * - page is not on LRU (isolate_page() is useful.)
2676 * - compound_lock is held when nr_pages > 1
2678 * This function doesn't do "charge" nor css_get to new cgroup. It should be
2679 * done by a caller(__mem_cgroup_try_charge would be useful). If @uncharge is
2680 * true, this function does "uncharge" from old cgroup, but it doesn't if
2681 * @uncharge is false, so a caller should do "uncharge".
2683 static int mem_cgroup_move_account(struct page
*page
,
2684 unsigned int nr_pages
,
2685 struct page_cgroup
*pc
,
2686 struct mem_cgroup
*from
,
2687 struct mem_cgroup
*to
,
2690 unsigned long flags
;
2693 VM_BUG_ON(from
== to
);
2694 VM_BUG_ON(PageLRU(page
));
2696 * The page is isolated from LRU. So, collapse function
2697 * will not handle this page. But page splitting can happen.
2698 * Do this check under compound_page_lock(). The caller should
2702 if (nr_pages
> 1 && !PageTransHuge(page
))
2705 lock_page_cgroup(pc
);
2708 if (!PageCgroupUsed(pc
) || pc
->mem_cgroup
!= from
)
2711 move_lock_page_cgroup(pc
, &flags
);
2713 if (PageCgroupFileMapped(pc
)) {
2714 /* Update mapped_file data for mem_cgroup */
2716 __this_cpu_dec(from
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2717 __this_cpu_inc(to
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2720 mem_cgroup_charge_statistics(from
, PageCgroupCache(pc
), -nr_pages
);
2722 /* This is not "cancel", but cancel_charge does all we need. */
2723 __mem_cgroup_cancel_charge(from
, nr_pages
);
2725 /* caller should have done css_get */
2726 pc
->mem_cgroup
= to
;
2727 mem_cgroup_charge_statistics(to
, PageCgroupCache(pc
), nr_pages
);
2729 * We charges against "to" which may not have any tasks. Then, "to"
2730 * can be under rmdir(). But in current implementation, caller of
2731 * this function is just force_empty() and move charge, so it's
2732 * guaranteed that "to" is never removed. So, we don't check rmdir
2735 move_unlock_page_cgroup(pc
, &flags
);
2738 unlock_page_cgroup(pc
);
2742 memcg_check_events(to
, page
);
2743 memcg_check_events(from
, page
);
2749 * move charges to its parent.
2752 static int mem_cgroup_move_parent(struct page
*page
,
2753 struct page_cgroup
*pc
,
2754 struct mem_cgroup
*child
,
2757 struct cgroup
*cg
= child
->css
.cgroup
;
2758 struct cgroup
*pcg
= cg
->parent
;
2759 struct mem_cgroup
*parent
;
2760 unsigned int nr_pages
;
2761 unsigned long uninitialized_var(flags
);
2769 if (!get_page_unless_zero(page
))
2771 if (isolate_lru_page(page
))
2774 nr_pages
= hpage_nr_pages(page
);
2776 parent
= mem_cgroup_from_cont(pcg
);
2777 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, nr_pages
, &parent
, false);
2782 flags
= compound_lock_irqsave(page
);
2784 ret
= mem_cgroup_move_account(page
, nr_pages
, pc
, child
, parent
, true);
2786 __mem_cgroup_cancel_charge(parent
, nr_pages
);
2789 compound_unlock_irqrestore(page
, flags
);
2791 putback_lru_page(page
);
2799 * Charge the memory controller for page usage.
2801 * 0 if the charge was successful
2802 * < 0 if the cgroup is over its limit
2804 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
2805 gfp_t gfp_mask
, enum charge_type ctype
)
2807 struct mem_cgroup
*mem
= NULL
;
2808 unsigned int nr_pages
= 1;
2809 struct page_cgroup
*pc
;
2813 if (PageTransHuge(page
)) {
2814 nr_pages
<<= compound_order(page
);
2815 VM_BUG_ON(!PageTransHuge(page
));
2817 * Never OOM-kill a process for a huge page. The
2818 * fault handler will fall back to regular pages.
2823 pc
= lookup_page_cgroup(page
);
2824 BUG_ON(!pc
); /* XXX: remove this and move pc lookup into commit */
2826 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, nr_pages
, &mem
, oom
);
2830 __mem_cgroup_commit_charge(mem
, page
, nr_pages
, pc
, ctype
);
2834 int mem_cgroup_newpage_charge(struct page
*page
,
2835 struct mm_struct
*mm
, gfp_t gfp_mask
)
2837 if (mem_cgroup_disabled())
2840 * If already mapped, we don't have to account.
2841 * If page cache, page->mapping has address_space.
2842 * But page->mapping may have out-of-use anon_vma pointer,
2843 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
2846 if (page_mapped(page
) || (page
->mapping
&& !PageAnon(page
)))
2850 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2851 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2855 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2856 enum charge_type ctype
);
2859 __mem_cgroup_commit_charge_lrucare(struct page
*page
, struct mem_cgroup
*mem
,
2860 enum charge_type ctype
)
2862 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
2864 * In some case, SwapCache, FUSE(splice_buf->radixtree), the page
2865 * is already on LRU. It means the page may on some other page_cgroup's
2866 * LRU. Take care of it.
2868 mem_cgroup_lru_del_before_commit(page
);
2869 __mem_cgroup_commit_charge(mem
, page
, 1, pc
, ctype
);
2870 mem_cgroup_lru_add_after_commit(page
);
2874 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
2877 struct mem_cgroup
*mem
= NULL
;
2880 if (mem_cgroup_disabled())
2882 if (PageCompound(page
))
2885 * Corner case handling. This is called from add_to_page_cache()
2886 * in usual. But some FS (shmem) precharges this page before calling it
2887 * and call add_to_page_cache() with GFP_NOWAIT.
2889 * For GFP_NOWAIT case, the page may be pre-charged before calling
2890 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
2891 * charge twice. (It works but has to pay a bit larger cost.)
2892 * And when the page is SwapCache, it should take swap information
2893 * into account. This is under lock_page() now.
2895 if (!(gfp_mask
& __GFP_WAIT
)) {
2896 struct page_cgroup
*pc
;
2898 pc
= lookup_page_cgroup(page
);
2901 lock_page_cgroup(pc
);
2902 if (PageCgroupUsed(pc
)) {
2903 unlock_page_cgroup(pc
);
2906 unlock_page_cgroup(pc
);
2912 if (page_is_file_cache(page
)) {
2913 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, 1, &mem
, true);
2918 * FUSE reuses pages without going through the final
2919 * put that would remove them from the LRU list, make
2920 * sure that they get relinked properly.
2922 __mem_cgroup_commit_charge_lrucare(page
, mem
,
2923 MEM_CGROUP_CHARGE_TYPE_CACHE
);
2927 if (PageSwapCache(page
)) {
2928 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &mem
);
2930 __mem_cgroup_commit_charge_swapin(page
, mem
,
2931 MEM_CGROUP_CHARGE_TYPE_SHMEM
);
2933 ret
= mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2934 MEM_CGROUP_CHARGE_TYPE_SHMEM
);
2940 * While swap-in, try_charge -> commit or cancel, the page is locked.
2941 * And when try_charge() successfully returns, one refcnt to memcg without
2942 * struct page_cgroup is acquired. This refcnt will be consumed by
2943 * "commit()" or removed by "cancel()"
2945 int mem_cgroup_try_charge_swapin(struct mm_struct
*mm
,
2947 gfp_t mask
, struct mem_cgroup
**ptr
)
2949 struct mem_cgroup
*mem
;
2954 if (mem_cgroup_disabled())
2957 if (!do_swap_account
)
2960 * A racing thread's fault, or swapoff, may have already updated
2961 * the pte, and even removed page from swap cache: in those cases
2962 * do_swap_page()'s pte_same() test will fail; but there's also a
2963 * KSM case which does need to charge the page.
2965 if (!PageSwapCache(page
))
2967 mem
= try_get_mem_cgroup_from_page(page
);
2971 ret
= __mem_cgroup_try_charge(NULL
, mask
, 1, ptr
, true);
2977 return __mem_cgroup_try_charge(mm
, mask
, 1, ptr
, true);
2981 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2982 enum charge_type ctype
)
2984 if (mem_cgroup_disabled())
2988 cgroup_exclude_rmdir(&ptr
->css
);
2990 __mem_cgroup_commit_charge_lrucare(page
, ptr
, ctype
);
2992 * Now swap is on-memory. This means this page may be
2993 * counted both as mem and swap....double count.
2994 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2995 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2996 * may call delete_from_swap_cache() before reach here.
2998 if (do_swap_account
&& PageSwapCache(page
)) {
2999 swp_entry_t ent
= {.val
= page_private(page
)};
3001 struct mem_cgroup
*memcg
;
3003 id
= swap_cgroup_record(ent
, 0);
3005 memcg
= mem_cgroup_lookup(id
);
3008 * This recorded memcg can be obsolete one. So, avoid
3009 * calling css_tryget
3011 if (!mem_cgroup_is_root(memcg
))
3012 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
3013 mem_cgroup_swap_statistics(memcg
, false);
3014 mem_cgroup_put(memcg
);
3019 * At swapin, we may charge account against cgroup which has no tasks.
3020 * So, rmdir()->pre_destroy() can be called while we do this charge.
3021 * In that case, we need to call pre_destroy() again. check it here.
3023 cgroup_release_and_wakeup_rmdir(&ptr
->css
);
3026 void mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
)
3028 __mem_cgroup_commit_charge_swapin(page
, ptr
,
3029 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
3032 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup
*mem
)
3034 if (mem_cgroup_disabled())
3038 __mem_cgroup_cancel_charge(mem
, 1);
3041 static void mem_cgroup_do_uncharge(struct mem_cgroup
*mem
,
3042 unsigned int nr_pages
,
3043 const enum charge_type ctype
)
3045 struct memcg_batch_info
*batch
= NULL
;
3046 bool uncharge_memsw
= true;
3048 /* If swapout, usage of swap doesn't decrease */
3049 if (!do_swap_account
|| ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
)
3050 uncharge_memsw
= false;
3052 batch
= ¤t
->memcg_batch
;
3054 * In usual, we do css_get() when we remember memcg pointer.
3055 * But in this case, we keep res->usage until end of a series of
3056 * uncharges. Then, it's ok to ignore memcg's refcnt.
3061 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
3062 * In those cases, all pages freed continuously can be expected to be in
3063 * the same cgroup and we have chance to coalesce uncharges.
3064 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
3065 * because we want to do uncharge as soon as possible.
3068 if (!batch
->do_batch
|| test_thread_flag(TIF_MEMDIE
))
3069 goto direct_uncharge
;
3072 goto direct_uncharge
;
3075 * In typical case, batch->memcg == mem. This means we can
3076 * merge a series of uncharges to an uncharge of res_counter.
3077 * If not, we uncharge res_counter ony by one.
3079 if (batch
->memcg
!= mem
)
3080 goto direct_uncharge
;
3081 /* remember freed charge and uncharge it later */
3084 batch
->memsw_nr_pages
++;
3087 res_counter_uncharge(&mem
->res
, nr_pages
* PAGE_SIZE
);
3089 res_counter_uncharge(&mem
->memsw
, nr_pages
* PAGE_SIZE
);
3090 if (unlikely(batch
->memcg
!= mem
))
3091 memcg_oom_recover(mem
);
3096 * uncharge if !page_mapped(page)
3098 static struct mem_cgroup
*
3099 __mem_cgroup_uncharge_common(struct page
*page
, enum charge_type ctype
)
3101 struct mem_cgroup
*mem
= NULL
;
3102 unsigned int nr_pages
= 1;
3103 struct page_cgroup
*pc
;
3105 if (mem_cgroup_disabled())
3108 if (PageSwapCache(page
))
3111 if (PageTransHuge(page
)) {
3112 nr_pages
<<= compound_order(page
);
3113 VM_BUG_ON(!PageTransHuge(page
));
3116 * Check if our page_cgroup is valid
3118 pc
= lookup_page_cgroup(page
);
3119 if (unlikely(!pc
|| !PageCgroupUsed(pc
)))
3122 lock_page_cgroup(pc
);
3124 mem
= pc
->mem_cgroup
;
3126 if (!PageCgroupUsed(pc
))
3130 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
3131 case MEM_CGROUP_CHARGE_TYPE_DROP
:
3132 /* See mem_cgroup_prepare_migration() */
3133 if (page_mapped(page
) || PageCgroupMigration(pc
))
3136 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT
:
3137 if (!PageAnon(page
)) { /* Shared memory */
3138 if (page
->mapping
&& !page_is_file_cache(page
))
3140 } else if (page_mapped(page
)) /* Anon */
3147 mem_cgroup_charge_statistics(mem
, PageCgroupCache(pc
), -nr_pages
);
3149 ClearPageCgroupUsed(pc
);
3151 * pc->mem_cgroup is not cleared here. It will be accessed when it's
3152 * freed from LRU. This is safe because uncharged page is expected not
3153 * to be reused (freed soon). Exception is SwapCache, it's handled by
3154 * special functions.
3157 unlock_page_cgroup(pc
);
3159 * even after unlock, we have mem->res.usage here and this memcg
3160 * will never be freed.
3162 memcg_check_events(mem
, page
);
3163 if (do_swap_account
&& ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
) {
3164 mem_cgroup_swap_statistics(mem
, true);
3165 mem_cgroup_get(mem
);
3167 if (!mem_cgroup_is_root(mem
))
3168 mem_cgroup_do_uncharge(mem
, nr_pages
, ctype
);
3173 unlock_page_cgroup(pc
);
3177 void mem_cgroup_uncharge_page(struct page
*page
)
3180 if (page_mapped(page
))
3182 if (page
->mapping
&& !PageAnon(page
))
3184 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_MAPPED
);
3187 void mem_cgroup_uncharge_cache_page(struct page
*page
)
3189 VM_BUG_ON(page_mapped(page
));
3190 VM_BUG_ON(page
->mapping
);
3191 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_CACHE
);
3195 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
3196 * In that cases, pages are freed continuously and we can expect pages
3197 * are in the same memcg. All these calls itself limits the number of
3198 * pages freed at once, then uncharge_start/end() is called properly.
3199 * This may be called prural(2) times in a context,
3202 void mem_cgroup_uncharge_start(void)
3204 current
->memcg_batch
.do_batch
++;
3205 /* We can do nest. */
3206 if (current
->memcg_batch
.do_batch
== 1) {
3207 current
->memcg_batch
.memcg
= NULL
;
3208 current
->memcg_batch
.nr_pages
= 0;
3209 current
->memcg_batch
.memsw_nr_pages
= 0;
3213 void mem_cgroup_uncharge_end(void)
3215 struct memcg_batch_info
*batch
= ¤t
->memcg_batch
;
3217 if (!batch
->do_batch
)
3221 if (batch
->do_batch
) /* If stacked, do nothing. */
3227 * This "batch->memcg" is valid without any css_get/put etc...
3228 * bacause we hide charges behind us.
3230 if (batch
->nr_pages
)
3231 res_counter_uncharge(&batch
->memcg
->res
,
3232 batch
->nr_pages
* PAGE_SIZE
);
3233 if (batch
->memsw_nr_pages
)
3234 res_counter_uncharge(&batch
->memcg
->memsw
,
3235 batch
->memsw_nr_pages
* PAGE_SIZE
);
3236 memcg_oom_recover(batch
->memcg
);
3237 /* forget this pointer (for sanity check) */
3238 batch
->memcg
= NULL
;
3243 * called after __delete_from_swap_cache() and drop "page" account.
3244 * memcg information is recorded to swap_cgroup of "ent"
3247 mem_cgroup_uncharge_swapcache(struct page
*page
, swp_entry_t ent
, bool swapout
)
3249 struct mem_cgroup
*memcg
;
3250 int ctype
= MEM_CGROUP_CHARGE_TYPE_SWAPOUT
;
3252 if (!swapout
) /* this was a swap cache but the swap is unused ! */
3253 ctype
= MEM_CGROUP_CHARGE_TYPE_DROP
;
3255 memcg
= __mem_cgroup_uncharge_common(page
, ctype
);
3258 * record memcg information, if swapout && memcg != NULL,
3259 * mem_cgroup_get() was called in uncharge().
3261 if (do_swap_account
&& swapout
&& memcg
)
3262 swap_cgroup_record(ent
, css_id(&memcg
->css
));
3266 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3268 * called from swap_entry_free(). remove record in swap_cgroup and
3269 * uncharge "memsw" account.
3271 void mem_cgroup_uncharge_swap(swp_entry_t ent
)
3273 struct mem_cgroup
*memcg
;
3276 if (!do_swap_account
)
3279 id
= swap_cgroup_record(ent
, 0);
3281 memcg
= mem_cgroup_lookup(id
);
3284 * We uncharge this because swap is freed.
3285 * This memcg can be obsolete one. We avoid calling css_tryget
3287 if (!mem_cgroup_is_root(memcg
))
3288 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
3289 mem_cgroup_swap_statistics(memcg
, false);
3290 mem_cgroup_put(memcg
);
3296 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3297 * @entry: swap entry to be moved
3298 * @from: mem_cgroup which the entry is moved from
3299 * @to: mem_cgroup which the entry is moved to
3300 * @need_fixup: whether we should fixup res_counters and refcounts.
3302 * It succeeds only when the swap_cgroup's record for this entry is the same
3303 * as the mem_cgroup's id of @from.
3305 * Returns 0 on success, -EINVAL on failure.
3307 * The caller must have charged to @to, IOW, called res_counter_charge() about
3308 * both res and memsw, and called css_get().
3310 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
3311 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
3313 unsigned short old_id
, new_id
;
3315 old_id
= css_id(&from
->css
);
3316 new_id
= css_id(&to
->css
);
3318 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
3319 mem_cgroup_swap_statistics(from
, false);
3320 mem_cgroup_swap_statistics(to
, true);
3322 * This function is only called from task migration context now.
3323 * It postpones res_counter and refcount handling till the end
3324 * of task migration(mem_cgroup_clear_mc()) for performance
3325 * improvement. But we cannot postpone mem_cgroup_get(to)
3326 * because if the process that has been moved to @to does
3327 * swap-in, the refcount of @to might be decreased to 0.
3331 if (!mem_cgroup_is_root(from
))
3332 res_counter_uncharge(&from
->memsw
, PAGE_SIZE
);
3333 mem_cgroup_put(from
);
3335 * we charged both to->res and to->memsw, so we should
3338 if (!mem_cgroup_is_root(to
))
3339 res_counter_uncharge(&to
->res
, PAGE_SIZE
);
3346 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
3347 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
3354 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3357 int mem_cgroup_prepare_migration(struct page
*page
,
3358 struct page
*newpage
, struct mem_cgroup
**ptr
, gfp_t gfp_mask
)
3360 struct mem_cgroup
*mem
= NULL
;
3361 struct page_cgroup
*pc
;
3362 enum charge_type ctype
;
3367 VM_BUG_ON(PageTransHuge(page
));
3368 if (mem_cgroup_disabled())
3371 pc
= lookup_page_cgroup(page
);
3372 lock_page_cgroup(pc
);
3373 if (PageCgroupUsed(pc
)) {
3374 mem
= pc
->mem_cgroup
;
3377 * At migrating an anonymous page, its mapcount goes down
3378 * to 0 and uncharge() will be called. But, even if it's fully
3379 * unmapped, migration may fail and this page has to be
3380 * charged again. We set MIGRATION flag here and delay uncharge
3381 * until end_migration() is called
3383 * Corner Case Thinking
3385 * When the old page was mapped as Anon and it's unmap-and-freed
3386 * while migration was ongoing.
3387 * If unmap finds the old page, uncharge() of it will be delayed
3388 * until end_migration(). If unmap finds a new page, it's
3389 * uncharged when it make mapcount to be 1->0. If unmap code
3390 * finds swap_migration_entry, the new page will not be mapped
3391 * and end_migration() will find it(mapcount==0).
3394 * When the old page was mapped but migraion fails, the kernel
3395 * remaps it. A charge for it is kept by MIGRATION flag even
3396 * if mapcount goes down to 0. We can do remap successfully
3397 * without charging it again.
3400 * The "old" page is under lock_page() until the end of
3401 * migration, so, the old page itself will not be swapped-out.
3402 * If the new page is swapped out before end_migraton, our
3403 * hook to usual swap-out path will catch the event.
3406 SetPageCgroupMigration(pc
);
3408 unlock_page_cgroup(pc
);
3410 * If the page is not charged at this point,
3417 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, 1, ptr
, false);
3418 css_put(&mem
->css
);/* drop extra refcnt */
3419 if (ret
|| *ptr
== NULL
) {
3420 if (PageAnon(page
)) {
3421 lock_page_cgroup(pc
);
3422 ClearPageCgroupMigration(pc
);
3423 unlock_page_cgroup(pc
);
3425 * The old page may be fully unmapped while we kept it.
3427 mem_cgroup_uncharge_page(page
);
3432 * We charge new page before it's used/mapped. So, even if unlock_page()
3433 * is called before end_migration, we can catch all events on this new
3434 * page. In the case new page is migrated but not remapped, new page's
3435 * mapcount will be finally 0 and we call uncharge in end_migration().
3437 pc
= lookup_page_cgroup(newpage
);
3439 ctype
= MEM_CGROUP_CHARGE_TYPE_MAPPED
;
3440 else if (page_is_file_cache(page
))
3441 ctype
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3443 ctype
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
3444 __mem_cgroup_commit_charge(mem
, page
, 1, pc
, ctype
);
3448 /* remove redundant charge if migration failed*/
3449 void mem_cgroup_end_migration(struct mem_cgroup
*mem
,
3450 struct page
*oldpage
, struct page
*newpage
, bool migration_ok
)
3452 struct page
*used
, *unused
;
3453 struct page_cgroup
*pc
;
3457 /* blocks rmdir() */
3458 cgroup_exclude_rmdir(&mem
->css
);
3459 if (!migration_ok
) {
3467 * We disallowed uncharge of pages under migration because mapcount
3468 * of the page goes down to zero, temporarly.
3469 * Clear the flag and check the page should be charged.
3471 pc
= lookup_page_cgroup(oldpage
);
3472 lock_page_cgroup(pc
);
3473 ClearPageCgroupMigration(pc
);
3474 unlock_page_cgroup(pc
);
3476 __mem_cgroup_uncharge_common(unused
, MEM_CGROUP_CHARGE_TYPE_FORCE
);
3479 * If a page is a file cache, radix-tree replacement is very atomic
3480 * and we can skip this check. When it was an Anon page, its mapcount
3481 * goes down to 0. But because we added MIGRATION flage, it's not
3482 * uncharged yet. There are several case but page->mapcount check
3483 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3484 * check. (see prepare_charge() also)
3487 mem_cgroup_uncharge_page(used
);
3489 * At migration, we may charge account against cgroup which has no
3491 * So, rmdir()->pre_destroy() can be called while we do this charge.
3492 * In that case, we need to call pre_destroy() again. check it here.
3494 cgroup_release_and_wakeup_rmdir(&mem
->css
);
3498 * A call to try to shrink memory usage on charge failure at shmem's swapin.
3499 * Calling hierarchical_reclaim is not enough because we should update
3500 * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM.
3501 * Moreover considering hierarchy, we should reclaim from the mem_over_limit,
3502 * not from the memcg which this page would be charged to.
3503 * try_charge_swapin does all of these works properly.
3505 int mem_cgroup_shmem_charge_fallback(struct page
*page
,
3506 struct mm_struct
*mm
,
3509 struct mem_cgroup
*mem
;
3512 if (mem_cgroup_disabled())
3515 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &mem
);
3517 mem_cgroup_cancel_charge_swapin(mem
); /* it does !mem check */
3522 #ifdef CONFIG_DEBUG_VM
3523 static struct page_cgroup
*lookup_page_cgroup_used(struct page
*page
)
3525 struct page_cgroup
*pc
;
3527 pc
= lookup_page_cgroup(page
);
3528 if (likely(pc
) && PageCgroupUsed(pc
))
3533 bool mem_cgroup_bad_page_check(struct page
*page
)
3535 if (mem_cgroup_disabled())
3538 return lookup_page_cgroup_used(page
) != NULL
;
3541 void mem_cgroup_print_bad_page(struct page
*page
)
3543 struct page_cgroup
*pc
;
3545 pc
= lookup_page_cgroup_used(page
);
3550 printk(KERN_ALERT
"pc:%p pc->flags:%lx pc->mem_cgroup:%p",
3551 pc
, pc
->flags
, pc
->mem_cgroup
);
3553 path
= kmalloc(PATH_MAX
, GFP_KERNEL
);
3556 ret
= cgroup_path(pc
->mem_cgroup
->css
.cgroup
,
3561 printk(KERN_CONT
"(%s)\n",
3562 (ret
< 0) ? "cannot get the path" : path
);
3568 static DEFINE_MUTEX(set_limit_mutex
);
3570 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
3571 unsigned long long val
)
3574 u64 memswlimit
, memlimit
;
3576 int children
= mem_cgroup_count_children(memcg
);
3577 u64 curusage
, oldusage
;
3581 * For keeping hierarchical_reclaim simple, how long we should retry
3582 * is depends on callers. We set our retry-count to be function
3583 * of # of children which we should visit in this loop.
3585 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
* children
;
3587 oldusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3590 while (retry_count
) {
3591 if (signal_pending(current
)) {
3596 * Rather than hide all in some function, I do this in
3597 * open coded manner. You see what this really does.
3598 * We have to guarantee mem->res.limit < mem->memsw.limit.
3600 mutex_lock(&set_limit_mutex
);
3601 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3602 if (memswlimit
< val
) {
3604 mutex_unlock(&set_limit_mutex
);
3608 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3612 ret
= res_counter_set_limit(&memcg
->res
, val
);
3614 if (memswlimit
== val
)
3615 memcg
->memsw_is_minimum
= true;
3617 memcg
->memsw_is_minimum
= false;
3619 mutex_unlock(&set_limit_mutex
);
3624 mem_cgroup_hierarchical_reclaim(memcg
, NULL
, GFP_KERNEL
,
3625 MEM_CGROUP_RECLAIM_SHRINK
,
3627 curusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3628 /* Usage is reduced ? */
3629 if (curusage
>= oldusage
)
3632 oldusage
= curusage
;
3634 if (!ret
&& enlarge
)
3635 memcg_oom_recover(memcg
);
3640 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
3641 unsigned long long val
)
3644 u64 memlimit
, memswlimit
, oldusage
, curusage
;
3645 int children
= mem_cgroup_count_children(memcg
);
3649 /* see mem_cgroup_resize_res_limit */
3650 retry_count
= children
* MEM_CGROUP_RECLAIM_RETRIES
;
3651 oldusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3652 while (retry_count
) {
3653 if (signal_pending(current
)) {
3658 * Rather than hide all in some function, I do this in
3659 * open coded manner. You see what this really does.
3660 * We have to guarantee mem->res.limit < mem->memsw.limit.
3662 mutex_lock(&set_limit_mutex
);
3663 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3664 if (memlimit
> val
) {
3666 mutex_unlock(&set_limit_mutex
);
3669 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3670 if (memswlimit
< val
)
3672 ret
= res_counter_set_limit(&memcg
->memsw
, val
);
3674 if (memlimit
== val
)
3675 memcg
->memsw_is_minimum
= true;
3677 memcg
->memsw_is_minimum
= false;
3679 mutex_unlock(&set_limit_mutex
);
3684 mem_cgroup_hierarchical_reclaim(memcg
, NULL
, GFP_KERNEL
,
3685 MEM_CGROUP_RECLAIM_NOSWAP
|
3686 MEM_CGROUP_RECLAIM_SHRINK
,
3688 curusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3689 /* Usage is reduced ? */
3690 if (curusage
>= oldusage
)
3693 oldusage
= curusage
;
3695 if (!ret
&& enlarge
)
3696 memcg_oom_recover(memcg
);
3700 unsigned long mem_cgroup_soft_limit_reclaim(struct zone
*zone
, int order
,
3702 unsigned long *total_scanned
)
3704 unsigned long nr_reclaimed
= 0;
3705 struct mem_cgroup_per_zone
*mz
, *next_mz
= NULL
;
3706 unsigned long reclaimed
;
3708 struct mem_cgroup_tree_per_zone
*mctz
;
3709 unsigned long long excess
;
3710 unsigned long nr_scanned
;
3715 mctz
= soft_limit_tree_node_zone(zone_to_nid(zone
), zone_idx(zone
));
3717 * This loop can run a while, specially if mem_cgroup's continuously
3718 * keep exceeding their soft limit and putting the system under
3725 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
3730 reclaimed
= mem_cgroup_hierarchical_reclaim(mz
->mem
, zone
,
3732 MEM_CGROUP_RECLAIM_SOFT
,
3734 nr_reclaimed
+= reclaimed
;
3735 *total_scanned
+= nr_scanned
;
3736 spin_lock(&mctz
->lock
);
3739 * If we failed to reclaim anything from this memory cgroup
3740 * it is time to move on to the next cgroup
3746 * Loop until we find yet another one.
3748 * By the time we get the soft_limit lock
3749 * again, someone might have aded the
3750 * group back on the RB tree. Iterate to
3751 * make sure we get a different mem.
3752 * mem_cgroup_largest_soft_limit_node returns
3753 * NULL if no other cgroup is present on
3757 __mem_cgroup_largest_soft_limit_node(mctz
);
3759 css_put(&next_mz
->mem
->css
);
3760 else /* next_mz == NULL or other memcg */
3764 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
3765 excess
= res_counter_soft_limit_excess(&mz
->mem
->res
);
3767 * One school of thought says that we should not add
3768 * back the node to the tree if reclaim returns 0.
3769 * But our reclaim could return 0, simply because due
3770 * to priority we are exposing a smaller subset of
3771 * memory to reclaim from. Consider this as a longer
3774 /* If excess == 0, no tree ops */
3775 __mem_cgroup_insert_exceeded(mz
->mem
, mz
, mctz
, excess
);
3776 spin_unlock(&mctz
->lock
);
3777 css_put(&mz
->mem
->css
);
3780 * Could not reclaim anything and there are no more
3781 * mem cgroups to try or we seem to be looping without
3782 * reclaiming anything.
3784 if (!nr_reclaimed
&&
3786 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
3788 } while (!nr_reclaimed
);
3790 css_put(&next_mz
->mem
->css
);
3791 return nr_reclaimed
;
3795 * This routine traverse page_cgroup in given list and drop them all.
3796 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3798 static int mem_cgroup_force_empty_list(struct mem_cgroup
*mem
,
3799 int node
, int zid
, enum lru_list lru
)
3802 struct mem_cgroup_per_zone
*mz
;
3803 struct page_cgroup
*pc
, *busy
;
3804 unsigned long flags
, loop
;
3805 struct list_head
*list
;
3808 zone
= &NODE_DATA(node
)->node_zones
[zid
];
3809 mz
= mem_cgroup_zoneinfo(mem
, node
, zid
);
3810 list
= &mz
->lists
[lru
];
3812 loop
= MEM_CGROUP_ZSTAT(mz
, lru
);
3813 /* give some margin against EBUSY etc...*/
3820 spin_lock_irqsave(&zone
->lru_lock
, flags
);
3821 if (list_empty(list
)) {
3822 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3825 pc
= list_entry(list
->prev
, struct page_cgroup
, lru
);
3827 list_move(&pc
->lru
, list
);
3829 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3832 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3834 page
= lookup_cgroup_page(pc
);
3836 ret
= mem_cgroup_move_parent(page
, pc
, mem
, GFP_KERNEL
);
3840 if (ret
== -EBUSY
|| ret
== -EINVAL
) {
3841 /* found lock contention or "pc" is obsolete. */
3848 if (!ret
&& !list_empty(list
))
3854 * make mem_cgroup's charge to be 0 if there is no task.
3855 * This enables deleting this mem_cgroup.
3857 static int mem_cgroup_force_empty(struct mem_cgroup
*mem
, bool free_all
)
3860 int node
, zid
, shrink
;
3861 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
3862 struct cgroup
*cgrp
= mem
->css
.cgroup
;
3867 /* should free all ? */
3873 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
))
3876 if (signal_pending(current
))
3878 /* This is for making all *used* pages to be on LRU. */
3879 lru_add_drain_all();
3880 drain_all_stock_sync(mem
);
3882 mem_cgroup_start_move(mem
);
3883 for_each_node_state(node
, N_HIGH_MEMORY
) {
3884 for (zid
= 0; !ret
&& zid
< MAX_NR_ZONES
; zid
++) {
3887 ret
= mem_cgroup_force_empty_list(mem
,
3896 mem_cgroup_end_move(mem
);
3897 memcg_oom_recover(mem
);
3898 /* it seems parent cgroup doesn't have enough mem */
3902 /* "ret" should also be checked to ensure all lists are empty. */
3903 } while (mem
->res
.usage
> 0 || ret
);
3909 /* returns EBUSY if there is a task or if we come here twice. */
3910 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
) || shrink
) {
3914 /* we call try-to-free pages for make this cgroup empty */
3915 lru_add_drain_all();
3916 /* try to free all pages in this cgroup */
3918 while (nr_retries
&& mem
->res
.usage
> 0) {
3919 struct memcg_scanrecord rec
;
3922 if (signal_pending(current
)) {
3926 rec
.context
= SCAN_BY_SHRINK
;
3929 progress
= try_to_free_mem_cgroup_pages(mem
, GFP_KERNEL
,
3933 /* maybe some writeback is necessary */
3934 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3939 /* try move_account...there may be some *locked* pages. */
3943 int mem_cgroup_force_empty_write(struct cgroup
*cont
, unsigned int event
)
3945 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont
), true);
3949 static u64
mem_cgroup_hierarchy_read(struct cgroup
*cont
, struct cftype
*cft
)
3951 return mem_cgroup_from_cont(cont
)->use_hierarchy
;
3954 static int mem_cgroup_hierarchy_write(struct cgroup
*cont
, struct cftype
*cft
,
3958 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
3959 struct cgroup
*parent
= cont
->parent
;
3960 struct mem_cgroup
*parent_mem
= NULL
;
3963 parent_mem
= mem_cgroup_from_cont(parent
);
3967 * If parent's use_hierarchy is set, we can't make any modifications
3968 * in the child subtrees. If it is unset, then the change can
3969 * occur, provided the current cgroup has no children.
3971 * For the root cgroup, parent_mem is NULL, we allow value to be
3972 * set if there are no children.
3974 if ((!parent_mem
|| !parent_mem
->use_hierarchy
) &&
3975 (val
== 1 || val
== 0)) {
3976 if (list_empty(&cont
->children
))
3977 mem
->use_hierarchy
= val
;
3988 static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup
*mem
,
3989 enum mem_cgroup_stat_index idx
)
3991 struct mem_cgroup
*iter
;
3994 /* Per-cpu values can be negative, use a signed accumulator */
3995 for_each_mem_cgroup_tree(iter
, mem
)
3996 val
+= mem_cgroup_read_stat(iter
, idx
);
3998 if (val
< 0) /* race ? */
4003 static inline u64
mem_cgroup_usage(struct mem_cgroup
*mem
, bool swap
)
4007 if (!mem_cgroup_is_root(mem
)) {
4009 return res_counter_read_u64(&mem
->res
, RES_USAGE
);
4011 return res_counter_read_u64(&mem
->memsw
, RES_USAGE
);
4014 val
= mem_cgroup_recursive_stat(mem
, MEM_CGROUP_STAT_CACHE
);
4015 val
+= mem_cgroup_recursive_stat(mem
, MEM_CGROUP_STAT_RSS
);
4018 val
+= mem_cgroup_recursive_stat(mem
, MEM_CGROUP_STAT_SWAPOUT
);
4020 return val
<< PAGE_SHIFT
;
4023 static u64
mem_cgroup_read(struct cgroup
*cont
, struct cftype
*cft
)
4025 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
4029 type
= MEMFILE_TYPE(cft
->private);
4030 name
= MEMFILE_ATTR(cft
->private);
4033 if (name
== RES_USAGE
)
4034 val
= mem_cgroup_usage(mem
, false);
4036 val
= res_counter_read_u64(&mem
->res
, name
);
4039 if (name
== RES_USAGE
)
4040 val
= mem_cgroup_usage(mem
, true);
4042 val
= res_counter_read_u64(&mem
->memsw
, name
);
4051 * The user of this function is...
4054 static int mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
4057 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4059 unsigned long long val
;
4062 type
= MEMFILE_TYPE(cft
->private);
4063 name
= MEMFILE_ATTR(cft
->private);
4066 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
4070 /* This function does all necessary parse...reuse it */
4071 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
4075 ret
= mem_cgroup_resize_limit(memcg
, val
);
4077 ret
= mem_cgroup_resize_memsw_limit(memcg
, val
);
4079 case RES_SOFT_LIMIT
:
4080 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
4084 * For memsw, soft limits are hard to implement in terms
4085 * of semantics, for now, we support soft limits for
4086 * control without swap
4089 ret
= res_counter_set_soft_limit(&memcg
->res
, val
);
4094 ret
= -EINVAL
; /* should be BUG() ? */
4100 static void memcg_get_hierarchical_limit(struct mem_cgroup
*memcg
,
4101 unsigned long long *mem_limit
, unsigned long long *memsw_limit
)
4103 struct cgroup
*cgroup
;
4104 unsigned long long min_limit
, min_memsw_limit
, tmp
;
4106 min_limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
4107 min_memsw_limit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
4108 cgroup
= memcg
->css
.cgroup
;
4109 if (!memcg
->use_hierarchy
)
4112 while (cgroup
->parent
) {
4113 cgroup
= cgroup
->parent
;
4114 memcg
= mem_cgroup_from_cont(cgroup
);
4115 if (!memcg
->use_hierarchy
)
4117 tmp
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
4118 min_limit
= min(min_limit
, tmp
);
4119 tmp
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
4120 min_memsw_limit
= min(min_memsw_limit
, tmp
);
4123 *mem_limit
= min_limit
;
4124 *memsw_limit
= min_memsw_limit
;
4128 static int mem_cgroup_reset(struct cgroup
*cont
, unsigned int event
)
4130 struct mem_cgroup
*mem
;
4133 mem
= mem_cgroup_from_cont(cont
);
4134 type
= MEMFILE_TYPE(event
);
4135 name
= MEMFILE_ATTR(event
);
4139 res_counter_reset_max(&mem
->res
);
4141 res_counter_reset_max(&mem
->memsw
);
4145 res_counter_reset_failcnt(&mem
->res
);
4147 res_counter_reset_failcnt(&mem
->memsw
);
4154 static u64
mem_cgroup_move_charge_read(struct cgroup
*cgrp
,
4157 return mem_cgroup_from_cont(cgrp
)->move_charge_at_immigrate
;
4161 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4162 struct cftype
*cft
, u64 val
)
4164 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
4166 if (val
>= (1 << NR_MOVE_TYPE
))
4169 * We check this value several times in both in can_attach() and
4170 * attach(), so we need cgroup lock to prevent this value from being
4174 mem
->move_charge_at_immigrate
= val
;
4180 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4181 struct cftype
*cft
, u64 val
)
4188 /* For read statistics */
4206 struct mcs_total_stat
{
4207 s64 stat
[NR_MCS_STAT
];
4213 } memcg_stat_strings
[NR_MCS_STAT
] = {
4214 {"cache", "total_cache"},
4215 {"rss", "total_rss"},
4216 {"mapped_file", "total_mapped_file"},
4217 {"pgpgin", "total_pgpgin"},
4218 {"pgpgout", "total_pgpgout"},
4219 {"swap", "total_swap"},
4220 {"pgfault", "total_pgfault"},
4221 {"pgmajfault", "total_pgmajfault"},
4222 {"inactive_anon", "total_inactive_anon"},
4223 {"active_anon", "total_active_anon"},
4224 {"inactive_file", "total_inactive_file"},
4225 {"active_file", "total_active_file"},
4226 {"unevictable", "total_unevictable"}
4231 mem_cgroup_get_local_stat(struct mem_cgroup
*mem
, struct mcs_total_stat
*s
)
4236 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_CACHE
);
4237 s
->stat
[MCS_CACHE
] += val
* PAGE_SIZE
;
4238 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_RSS
);
4239 s
->stat
[MCS_RSS
] += val
* PAGE_SIZE
;
4240 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_FILE_MAPPED
);
4241 s
->stat
[MCS_FILE_MAPPED
] += val
* PAGE_SIZE
;
4242 val
= mem_cgroup_read_events(mem
, MEM_CGROUP_EVENTS_PGPGIN
);
4243 s
->stat
[MCS_PGPGIN
] += val
;
4244 val
= mem_cgroup_read_events(mem
, MEM_CGROUP_EVENTS_PGPGOUT
);
4245 s
->stat
[MCS_PGPGOUT
] += val
;
4246 if (do_swap_account
) {
4247 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_SWAPOUT
);
4248 s
->stat
[MCS_SWAP
] += val
* PAGE_SIZE
;
4250 val
= mem_cgroup_read_events(mem
, MEM_CGROUP_EVENTS_PGFAULT
);
4251 s
->stat
[MCS_PGFAULT
] += val
;
4252 val
= mem_cgroup_read_events(mem
, MEM_CGROUP_EVENTS_PGMAJFAULT
);
4253 s
->stat
[MCS_PGMAJFAULT
] += val
;
4256 val
= mem_cgroup_nr_lru_pages(mem
, BIT(LRU_INACTIVE_ANON
));
4257 s
->stat
[MCS_INACTIVE_ANON
] += val
* PAGE_SIZE
;
4258 val
= mem_cgroup_nr_lru_pages(mem
, BIT(LRU_ACTIVE_ANON
));
4259 s
->stat
[MCS_ACTIVE_ANON
] += val
* PAGE_SIZE
;
4260 val
= mem_cgroup_nr_lru_pages(mem
, BIT(LRU_INACTIVE_FILE
));
4261 s
->stat
[MCS_INACTIVE_FILE
] += val
* PAGE_SIZE
;
4262 val
= mem_cgroup_nr_lru_pages(mem
, BIT(LRU_ACTIVE_FILE
));
4263 s
->stat
[MCS_ACTIVE_FILE
] += val
* PAGE_SIZE
;
4264 val
= mem_cgroup_nr_lru_pages(mem
, BIT(LRU_UNEVICTABLE
));
4265 s
->stat
[MCS_UNEVICTABLE
] += val
* PAGE_SIZE
;
4269 mem_cgroup_get_total_stat(struct mem_cgroup
*mem
, struct mcs_total_stat
*s
)
4271 struct mem_cgroup
*iter
;
4273 for_each_mem_cgroup_tree(iter
, mem
)
4274 mem_cgroup_get_local_stat(iter
, s
);
4278 static int mem_control_numa_stat_show(struct seq_file
*m
, void *arg
)
4281 unsigned long total_nr
, file_nr
, anon_nr
, unevictable_nr
;
4282 unsigned long node_nr
;
4283 struct cgroup
*cont
= m
->private;
4284 struct mem_cgroup
*mem_cont
= mem_cgroup_from_cont(cont
);
4286 total_nr
= mem_cgroup_nr_lru_pages(mem_cont
, LRU_ALL
);
4287 seq_printf(m
, "total=%lu", total_nr
);
4288 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4289 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
, LRU_ALL
);
4290 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4294 file_nr
= mem_cgroup_nr_lru_pages(mem_cont
, LRU_ALL_FILE
);
4295 seq_printf(m
, "file=%lu", file_nr
);
4296 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4297 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
,
4299 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4303 anon_nr
= mem_cgroup_nr_lru_pages(mem_cont
, LRU_ALL_ANON
);
4304 seq_printf(m
, "anon=%lu", anon_nr
);
4305 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4306 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
,
4308 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4312 unevictable_nr
= mem_cgroup_nr_lru_pages(mem_cont
, BIT(LRU_UNEVICTABLE
));
4313 seq_printf(m
, "unevictable=%lu", unevictable_nr
);
4314 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4315 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
,
4316 BIT(LRU_UNEVICTABLE
));
4317 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4322 #endif /* CONFIG_NUMA */
4324 static int mem_control_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
4325 struct cgroup_map_cb
*cb
)
4327 struct mem_cgroup
*mem_cont
= mem_cgroup_from_cont(cont
);
4328 struct mcs_total_stat mystat
;
4331 memset(&mystat
, 0, sizeof(mystat
));
4332 mem_cgroup_get_local_stat(mem_cont
, &mystat
);
4335 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
4336 if (i
== MCS_SWAP
&& !do_swap_account
)
4338 cb
->fill(cb
, memcg_stat_strings
[i
].local_name
, mystat
.stat
[i
]);
4341 /* Hierarchical information */
4343 unsigned long long limit
, memsw_limit
;
4344 memcg_get_hierarchical_limit(mem_cont
, &limit
, &memsw_limit
);
4345 cb
->fill(cb
, "hierarchical_memory_limit", limit
);
4346 if (do_swap_account
)
4347 cb
->fill(cb
, "hierarchical_memsw_limit", memsw_limit
);
4350 memset(&mystat
, 0, sizeof(mystat
));
4351 mem_cgroup_get_total_stat(mem_cont
, &mystat
);
4352 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
4353 if (i
== MCS_SWAP
&& !do_swap_account
)
4355 cb
->fill(cb
, memcg_stat_strings
[i
].total_name
, mystat
.stat
[i
]);
4358 #ifdef CONFIG_DEBUG_VM
4359 cb
->fill(cb
, "inactive_ratio", calc_inactive_ratio(mem_cont
, NULL
));
4363 struct mem_cgroup_per_zone
*mz
;
4364 unsigned long recent_rotated
[2] = {0, 0};
4365 unsigned long recent_scanned
[2] = {0, 0};
4367 for_each_online_node(nid
)
4368 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
4369 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
4371 recent_rotated
[0] +=
4372 mz
->reclaim_stat
.recent_rotated
[0];
4373 recent_rotated
[1] +=
4374 mz
->reclaim_stat
.recent_rotated
[1];
4375 recent_scanned
[0] +=
4376 mz
->reclaim_stat
.recent_scanned
[0];
4377 recent_scanned
[1] +=
4378 mz
->reclaim_stat
.recent_scanned
[1];
4380 cb
->fill(cb
, "recent_rotated_anon", recent_rotated
[0]);
4381 cb
->fill(cb
, "recent_rotated_file", recent_rotated
[1]);
4382 cb
->fill(cb
, "recent_scanned_anon", recent_scanned
[0]);
4383 cb
->fill(cb
, "recent_scanned_file", recent_scanned
[1]);
4390 static u64
mem_cgroup_swappiness_read(struct cgroup
*cgrp
, struct cftype
*cft
)
4392 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4394 return mem_cgroup_swappiness(memcg
);
4397 static int mem_cgroup_swappiness_write(struct cgroup
*cgrp
, struct cftype
*cft
,
4400 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4401 struct mem_cgroup
*parent
;
4406 if (cgrp
->parent
== NULL
)
4409 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4413 /* If under hierarchy, only empty-root can set this value */
4414 if ((parent
->use_hierarchy
) ||
4415 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4420 memcg
->swappiness
= val
;
4427 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
4429 struct mem_cgroup_threshold_ary
*t
;
4435 t
= rcu_dereference(memcg
->thresholds
.primary
);
4437 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
4442 usage
= mem_cgroup_usage(memcg
, swap
);
4445 * current_threshold points to threshold just below usage.
4446 * If it's not true, a threshold was crossed after last
4447 * call of __mem_cgroup_threshold().
4449 i
= t
->current_threshold
;
4452 * Iterate backward over array of thresholds starting from
4453 * current_threshold and check if a threshold is crossed.
4454 * If none of thresholds below usage is crossed, we read
4455 * only one element of the array here.
4457 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
4458 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4460 /* i = current_threshold + 1 */
4464 * Iterate forward over array of thresholds starting from
4465 * current_threshold+1 and check if a threshold is crossed.
4466 * If none of thresholds above usage is crossed, we read
4467 * only one element of the array here.
4469 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
4470 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4472 /* Update current_threshold */
4473 t
->current_threshold
= i
- 1;
4478 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
4481 __mem_cgroup_threshold(memcg
, false);
4482 if (do_swap_account
)
4483 __mem_cgroup_threshold(memcg
, true);
4485 memcg
= parent_mem_cgroup(memcg
);
4489 static int compare_thresholds(const void *a
, const void *b
)
4491 const struct mem_cgroup_threshold
*_a
= a
;
4492 const struct mem_cgroup_threshold
*_b
= b
;
4494 return _a
->threshold
- _b
->threshold
;
4497 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*mem
)
4499 struct mem_cgroup_eventfd_list
*ev
;
4501 list_for_each_entry(ev
, &mem
->oom_notify
, list
)
4502 eventfd_signal(ev
->eventfd
, 1);
4506 static void mem_cgroup_oom_notify(struct mem_cgroup
*mem
)
4508 struct mem_cgroup
*iter
;
4510 for_each_mem_cgroup_tree(iter
, mem
)
4511 mem_cgroup_oom_notify_cb(iter
);
4514 static int mem_cgroup_usage_register_event(struct cgroup
*cgrp
,
4515 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4517 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4518 struct mem_cgroup_thresholds
*thresholds
;
4519 struct mem_cgroup_threshold_ary
*new;
4520 int type
= MEMFILE_TYPE(cft
->private);
4521 u64 threshold
, usage
;
4524 ret
= res_counter_memparse_write_strategy(args
, &threshold
);
4528 mutex_lock(&memcg
->thresholds_lock
);
4531 thresholds
= &memcg
->thresholds
;
4532 else if (type
== _MEMSWAP
)
4533 thresholds
= &memcg
->memsw_thresholds
;
4537 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4539 /* Check if a threshold crossed before adding a new one */
4540 if (thresholds
->primary
)
4541 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4543 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
4545 /* Allocate memory for new array of thresholds */
4546 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
4554 /* Copy thresholds (if any) to new array */
4555 if (thresholds
->primary
) {
4556 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
4557 sizeof(struct mem_cgroup_threshold
));
4560 /* Add new threshold */
4561 new->entries
[size
- 1].eventfd
= eventfd
;
4562 new->entries
[size
- 1].threshold
= threshold
;
4564 /* Sort thresholds. Registering of new threshold isn't time-critical */
4565 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
4566 compare_thresholds
, NULL
);
4568 /* Find current threshold */
4569 new->current_threshold
= -1;
4570 for (i
= 0; i
< size
; i
++) {
4571 if (new->entries
[i
].threshold
< usage
) {
4573 * new->current_threshold will not be used until
4574 * rcu_assign_pointer(), so it's safe to increment
4577 ++new->current_threshold
;
4581 /* Free old spare buffer and save old primary buffer as spare */
4582 kfree(thresholds
->spare
);
4583 thresholds
->spare
= thresholds
->primary
;
4585 rcu_assign_pointer(thresholds
->primary
, new);
4587 /* To be sure that nobody uses thresholds */
4591 mutex_unlock(&memcg
->thresholds_lock
);
4596 static void mem_cgroup_usage_unregister_event(struct cgroup
*cgrp
,
4597 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4599 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4600 struct mem_cgroup_thresholds
*thresholds
;
4601 struct mem_cgroup_threshold_ary
*new;
4602 int type
= MEMFILE_TYPE(cft
->private);
4606 mutex_lock(&memcg
->thresholds_lock
);
4608 thresholds
= &memcg
->thresholds
;
4609 else if (type
== _MEMSWAP
)
4610 thresholds
= &memcg
->memsw_thresholds
;
4615 * Something went wrong if we trying to unregister a threshold
4616 * if we don't have thresholds
4618 BUG_ON(!thresholds
);
4620 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4622 /* Check if a threshold crossed before removing */
4623 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4625 /* Calculate new number of threshold */
4627 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
4628 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
4632 new = thresholds
->spare
;
4634 /* Set thresholds array to NULL if we don't have thresholds */
4643 /* Copy thresholds and find current threshold */
4644 new->current_threshold
= -1;
4645 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
4646 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
4649 new->entries
[j
] = thresholds
->primary
->entries
[i
];
4650 if (new->entries
[j
].threshold
< usage
) {
4652 * new->current_threshold will not be used
4653 * until rcu_assign_pointer(), so it's safe to increment
4656 ++new->current_threshold
;
4662 /* Swap primary and spare array */
4663 thresholds
->spare
= thresholds
->primary
;
4664 rcu_assign_pointer(thresholds
->primary
, new);
4666 /* To be sure that nobody uses thresholds */
4669 mutex_unlock(&memcg
->thresholds_lock
);
4672 static int mem_cgroup_oom_register_event(struct cgroup
*cgrp
,
4673 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4675 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4676 struct mem_cgroup_eventfd_list
*event
;
4677 int type
= MEMFILE_TYPE(cft
->private);
4679 BUG_ON(type
!= _OOM_TYPE
);
4680 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
4684 spin_lock(&memcg_oom_lock
);
4686 event
->eventfd
= eventfd
;
4687 list_add(&event
->list
, &memcg
->oom_notify
);
4689 /* already in OOM ? */
4690 if (atomic_read(&memcg
->under_oom
))
4691 eventfd_signal(eventfd
, 1);
4692 spin_unlock(&memcg_oom_lock
);
4697 static void mem_cgroup_oom_unregister_event(struct cgroup
*cgrp
,
4698 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4700 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
4701 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
4702 int type
= MEMFILE_TYPE(cft
->private);
4704 BUG_ON(type
!= _OOM_TYPE
);
4706 spin_lock(&memcg_oom_lock
);
4708 list_for_each_entry_safe(ev
, tmp
, &mem
->oom_notify
, list
) {
4709 if (ev
->eventfd
== eventfd
) {
4710 list_del(&ev
->list
);
4715 spin_unlock(&memcg_oom_lock
);
4718 static int mem_cgroup_oom_control_read(struct cgroup
*cgrp
,
4719 struct cftype
*cft
, struct cgroup_map_cb
*cb
)
4721 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
4723 cb
->fill(cb
, "oom_kill_disable", mem
->oom_kill_disable
);
4725 if (atomic_read(&mem
->under_oom
))
4726 cb
->fill(cb
, "under_oom", 1);
4728 cb
->fill(cb
, "under_oom", 0);
4732 static int mem_cgroup_oom_control_write(struct cgroup
*cgrp
,
4733 struct cftype
*cft
, u64 val
)
4735 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
4736 struct mem_cgroup
*parent
;
4738 /* cannot set to root cgroup and only 0 and 1 are allowed */
4739 if (!cgrp
->parent
|| !((val
== 0) || (val
== 1)))
4742 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4745 /* oom-kill-disable is a flag for subhierarchy. */
4746 if ((parent
->use_hierarchy
) ||
4747 (mem
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4751 mem
->oom_kill_disable
= val
;
4753 memcg_oom_recover(mem
);
4759 static const struct file_operations mem_control_numa_stat_file_operations
= {
4761 .llseek
= seq_lseek
,
4762 .release
= single_release
,
4765 static int mem_control_numa_stat_open(struct inode
*unused
, struct file
*file
)
4767 struct cgroup
*cont
= file
->f_dentry
->d_parent
->d_fsdata
;
4769 file
->f_op
= &mem_control_numa_stat_file_operations
;
4770 return single_open(file
, mem_control_numa_stat_show
, cont
);
4772 #endif /* CONFIG_NUMA */
4774 static int mem_cgroup_vmscan_stat_read(struct cgroup
*cgrp
,
4776 struct cgroup_map_cb
*cb
)
4778 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
4782 for (i
= 0; i
< NR_SCANSTATS
; i
++) {
4783 strcpy(string
, scanstat_string
[i
]);
4784 strcat(string
, SCANSTAT_WORD_LIMIT
);
4785 cb
->fill(cb
, string
, mem
->scanstat
.stats
[SCAN_BY_LIMIT
][i
]);
4788 for (i
= 0; i
< NR_SCANSTATS
; i
++) {
4789 strcpy(string
, scanstat_string
[i
]);
4790 strcat(string
, SCANSTAT_WORD_SYSTEM
);
4791 cb
->fill(cb
, string
, mem
->scanstat
.stats
[SCAN_BY_SYSTEM
][i
]);
4794 for (i
= 0; i
< NR_SCANSTATS
; i
++) {
4795 strcpy(string
, scanstat_string
[i
]);
4796 strcat(string
, SCANSTAT_WORD_LIMIT
);
4797 strcat(string
, SCANSTAT_WORD_HIERARCHY
);
4798 cb
->fill(cb
, string
, mem
->scanstat
.rootstats
[SCAN_BY_LIMIT
][i
]);
4800 for (i
= 0; i
< NR_SCANSTATS
; i
++) {
4801 strcpy(string
, scanstat_string
[i
]);
4802 strcat(string
, SCANSTAT_WORD_SYSTEM
);
4803 strcat(string
, SCANSTAT_WORD_HIERARCHY
);
4804 cb
->fill(cb
, string
, mem
->scanstat
.rootstats
[SCAN_BY_SYSTEM
][i
]);
4809 static int mem_cgroup_reset_vmscan_stat(struct cgroup
*cgrp
,
4812 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
4814 spin_lock(&mem
->scanstat
.lock
);
4815 memset(&mem
->scanstat
.stats
, 0, sizeof(mem
->scanstat
.stats
));
4816 memset(&mem
->scanstat
.rootstats
, 0, sizeof(mem
->scanstat
.rootstats
));
4817 spin_unlock(&mem
->scanstat
.lock
);
4822 static struct cftype mem_cgroup_files
[] = {
4824 .name
= "usage_in_bytes",
4825 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
4826 .read_u64
= mem_cgroup_read
,
4827 .register_event
= mem_cgroup_usage_register_event
,
4828 .unregister_event
= mem_cgroup_usage_unregister_event
,
4831 .name
= "max_usage_in_bytes",
4832 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
4833 .trigger
= mem_cgroup_reset
,
4834 .read_u64
= mem_cgroup_read
,
4837 .name
= "limit_in_bytes",
4838 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
4839 .write_string
= mem_cgroup_write
,
4840 .read_u64
= mem_cgroup_read
,
4843 .name
= "soft_limit_in_bytes",
4844 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
4845 .write_string
= mem_cgroup_write
,
4846 .read_u64
= mem_cgroup_read
,
4850 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
4851 .trigger
= mem_cgroup_reset
,
4852 .read_u64
= mem_cgroup_read
,
4856 .read_map
= mem_control_stat_show
,
4859 .name
= "force_empty",
4860 .trigger
= mem_cgroup_force_empty_write
,
4863 .name
= "use_hierarchy",
4864 .write_u64
= mem_cgroup_hierarchy_write
,
4865 .read_u64
= mem_cgroup_hierarchy_read
,
4868 .name
= "swappiness",
4869 .read_u64
= mem_cgroup_swappiness_read
,
4870 .write_u64
= mem_cgroup_swappiness_write
,
4873 .name
= "move_charge_at_immigrate",
4874 .read_u64
= mem_cgroup_move_charge_read
,
4875 .write_u64
= mem_cgroup_move_charge_write
,
4878 .name
= "oom_control",
4879 .read_map
= mem_cgroup_oom_control_read
,
4880 .write_u64
= mem_cgroup_oom_control_write
,
4881 .register_event
= mem_cgroup_oom_register_event
,
4882 .unregister_event
= mem_cgroup_oom_unregister_event
,
4883 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
4887 .name
= "numa_stat",
4888 .open
= mem_control_numa_stat_open
,
4893 .name
= "vmscan_stat",
4894 .read_map
= mem_cgroup_vmscan_stat_read
,
4895 .trigger
= mem_cgroup_reset_vmscan_stat
,
4899 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4900 static struct cftype memsw_cgroup_files
[] = {
4902 .name
= "memsw.usage_in_bytes",
4903 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
4904 .read_u64
= mem_cgroup_read
,
4905 .register_event
= mem_cgroup_usage_register_event
,
4906 .unregister_event
= mem_cgroup_usage_unregister_event
,
4909 .name
= "memsw.max_usage_in_bytes",
4910 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
4911 .trigger
= mem_cgroup_reset
,
4912 .read_u64
= mem_cgroup_read
,
4915 .name
= "memsw.limit_in_bytes",
4916 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
4917 .write_string
= mem_cgroup_write
,
4918 .read_u64
= mem_cgroup_read
,
4921 .name
= "memsw.failcnt",
4922 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
4923 .trigger
= mem_cgroup_reset
,
4924 .read_u64
= mem_cgroup_read
,
4928 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4930 if (!do_swap_account
)
4932 return cgroup_add_files(cont
, ss
, memsw_cgroup_files
,
4933 ARRAY_SIZE(memsw_cgroup_files
));
4936 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4942 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*mem
, int node
)
4944 struct mem_cgroup_per_node
*pn
;
4945 struct mem_cgroup_per_zone
*mz
;
4947 int zone
, tmp
= node
;
4949 * This routine is called against possible nodes.
4950 * But it's BUG to call kmalloc() against offline node.
4952 * TODO: this routine can waste much memory for nodes which will
4953 * never be onlined. It's better to use memory hotplug callback
4956 if (!node_state(node
, N_NORMAL_MEMORY
))
4958 pn
= kzalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
4962 mem
->info
.nodeinfo
[node
] = pn
;
4963 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4964 mz
= &pn
->zoneinfo
[zone
];
4966 INIT_LIST_HEAD(&mz
->lists
[l
]);
4967 mz
->usage_in_excess
= 0;
4968 mz
->on_tree
= false;
4974 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*mem
, int node
)
4976 kfree(mem
->info
.nodeinfo
[node
]);
4979 static struct mem_cgroup
*mem_cgroup_alloc(void)
4981 struct mem_cgroup
*mem
;
4982 int size
= sizeof(struct mem_cgroup
);
4984 /* Can be very big if MAX_NUMNODES is very big */
4985 if (size
< PAGE_SIZE
)
4986 mem
= kzalloc(size
, GFP_KERNEL
);
4988 mem
= vzalloc(size
);
4993 mem
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4996 spin_lock_init(&mem
->pcp_counter_lock
);
5000 if (size
< PAGE_SIZE
)
5008 * At destroying mem_cgroup, references from swap_cgroup can remain.
5009 * (scanning all at force_empty is too costly...)
5011 * Instead of clearing all references at force_empty, we remember
5012 * the number of reference from swap_cgroup and free mem_cgroup when
5013 * it goes down to 0.
5015 * Removal of cgroup itself succeeds regardless of refs from swap.
5018 static void __mem_cgroup_free(struct mem_cgroup
*mem
)
5022 mem_cgroup_remove_from_trees(mem
);
5023 free_css_id(&mem_cgroup_subsys
, &mem
->css
);
5025 for_each_node_state(node
, N_POSSIBLE
)
5026 free_mem_cgroup_per_zone_info(mem
, node
);
5028 free_percpu(mem
->stat
);
5029 if (sizeof(struct mem_cgroup
) < PAGE_SIZE
)
5035 static void mem_cgroup_get(struct mem_cgroup
*mem
)
5037 atomic_inc(&mem
->refcnt
);
5040 static void __mem_cgroup_put(struct mem_cgroup
*mem
, int count
)
5042 if (atomic_sub_and_test(count
, &mem
->refcnt
)) {
5043 struct mem_cgroup
*parent
= parent_mem_cgroup(mem
);
5044 __mem_cgroup_free(mem
);
5046 mem_cgroup_put(parent
);
5050 static void mem_cgroup_put(struct mem_cgroup
*mem
)
5052 __mem_cgroup_put(mem
, 1);
5056 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
5058 static struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*mem
)
5060 if (!mem
->res
.parent
)
5062 return mem_cgroup_from_res_counter(mem
->res
.parent
, res
);
5065 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
5066 static void __init
enable_swap_cgroup(void)
5068 if (!mem_cgroup_disabled() && really_do_swap_account
)
5069 do_swap_account
= 1;
5072 static void __init
enable_swap_cgroup(void)
5077 static int mem_cgroup_soft_limit_tree_init(void)
5079 struct mem_cgroup_tree_per_node
*rtpn
;
5080 struct mem_cgroup_tree_per_zone
*rtpz
;
5081 int tmp
, node
, zone
;
5083 for_each_node_state(node
, N_POSSIBLE
) {
5085 if (!node_state(node
, N_NORMAL_MEMORY
))
5087 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
, tmp
);
5091 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
5093 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
5094 rtpz
= &rtpn
->rb_tree_per_zone
[zone
];
5095 rtpz
->rb_root
= RB_ROOT
;
5096 spin_lock_init(&rtpz
->lock
);
5102 static struct cgroup_subsys_state
* __ref
5103 mem_cgroup_create(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
5105 struct mem_cgroup
*mem
, *parent
;
5106 long error
= -ENOMEM
;
5109 mem
= mem_cgroup_alloc();
5111 return ERR_PTR(error
);
5113 for_each_node_state(node
, N_POSSIBLE
)
5114 if (alloc_mem_cgroup_per_zone_info(mem
, node
))
5118 if (cont
->parent
== NULL
) {
5120 enable_swap_cgroup();
5122 root_mem_cgroup
= mem
;
5123 if (mem_cgroup_soft_limit_tree_init())
5125 for_each_possible_cpu(cpu
) {
5126 struct memcg_stock_pcp
*stock
=
5127 &per_cpu(memcg_stock
, cpu
);
5128 INIT_WORK(&stock
->work
, drain_local_stock
);
5130 hotcpu_notifier(memcg_cpu_hotplug_callback
, 0);
5132 parent
= mem_cgroup_from_cont(cont
->parent
);
5133 mem
->use_hierarchy
= parent
->use_hierarchy
;
5134 mem
->oom_kill_disable
= parent
->oom_kill_disable
;
5137 if (parent
&& parent
->use_hierarchy
) {
5138 res_counter_init(&mem
->res
, &parent
->res
);
5139 res_counter_init(&mem
->memsw
, &parent
->memsw
);
5141 * We increment refcnt of the parent to ensure that we can
5142 * safely access it on res_counter_charge/uncharge.
5143 * This refcnt will be decremented when freeing this
5144 * mem_cgroup(see mem_cgroup_put).
5146 mem_cgroup_get(parent
);
5148 res_counter_init(&mem
->res
, NULL
);
5149 res_counter_init(&mem
->memsw
, NULL
);
5151 mem
->last_scanned_child
= 0;
5152 mem
->last_scanned_node
= MAX_NUMNODES
;
5153 INIT_LIST_HEAD(&mem
->oom_notify
);
5156 mem
->swappiness
= mem_cgroup_swappiness(parent
);
5157 atomic_set(&mem
->refcnt
, 1);
5158 mem
->move_charge_at_immigrate
= 0;
5159 mutex_init(&mem
->thresholds_lock
);
5160 spin_lock_init(&mem
->scanstat
.lock
);
5163 __mem_cgroup_free(mem
);
5164 root_mem_cgroup
= NULL
;
5165 return ERR_PTR(error
);
5168 static int mem_cgroup_pre_destroy(struct cgroup_subsys
*ss
,
5169 struct cgroup
*cont
)
5171 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
5173 return mem_cgroup_force_empty(mem
, false);
5176 static void mem_cgroup_destroy(struct cgroup_subsys
*ss
,
5177 struct cgroup
*cont
)
5179 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
5181 mem_cgroup_put(mem
);
5184 static int mem_cgroup_populate(struct cgroup_subsys
*ss
,
5185 struct cgroup
*cont
)
5189 ret
= cgroup_add_files(cont
, ss
, mem_cgroup_files
,
5190 ARRAY_SIZE(mem_cgroup_files
));
5193 ret
= register_memsw_files(cont
, ss
);
5198 /* Handlers for move charge at task migration. */
5199 #define PRECHARGE_COUNT_AT_ONCE 256
5200 static int mem_cgroup_do_precharge(unsigned long count
)
5203 int batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5204 struct mem_cgroup
*mem
= mc
.to
;
5206 if (mem_cgroup_is_root(mem
)) {
5207 mc
.precharge
+= count
;
5208 /* we don't need css_get for root */
5211 /* try to charge at once */
5213 struct res_counter
*dummy
;
5215 * "mem" cannot be under rmdir() because we've already checked
5216 * by cgroup_lock_live_cgroup() that it is not removed and we
5217 * are still under the same cgroup_mutex. So we can postpone
5220 if (res_counter_charge(&mem
->res
, PAGE_SIZE
* count
, &dummy
))
5222 if (do_swap_account
&& res_counter_charge(&mem
->memsw
,
5223 PAGE_SIZE
* count
, &dummy
)) {
5224 res_counter_uncharge(&mem
->res
, PAGE_SIZE
* count
);
5227 mc
.precharge
+= count
;
5231 /* fall back to one by one charge */
5233 if (signal_pending(current
)) {
5237 if (!batch_count
--) {
5238 batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5241 ret
= __mem_cgroup_try_charge(NULL
, GFP_KERNEL
, 1, &mem
, false);
5243 /* mem_cgroup_clear_mc() will do uncharge later */
5251 * is_target_pte_for_mc - check a pte whether it is valid for move charge
5252 * @vma: the vma the pte to be checked belongs
5253 * @addr: the address corresponding to the pte to be checked
5254 * @ptent: the pte to be checked
5255 * @target: the pointer the target page or swap ent will be stored(can be NULL)
5258 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5259 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5260 * move charge. if @target is not NULL, the page is stored in target->page
5261 * with extra refcnt got(Callers should handle it).
5262 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5263 * target for charge migration. if @target is not NULL, the entry is stored
5266 * Called with pte lock held.
5273 enum mc_target_type
{
5274 MC_TARGET_NONE
, /* not used */
5279 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
5280 unsigned long addr
, pte_t ptent
)
5282 struct page
*page
= vm_normal_page(vma
, addr
, ptent
);
5284 if (!page
|| !page_mapped(page
))
5286 if (PageAnon(page
)) {
5287 /* we don't move shared anon */
5288 if (!move_anon() || page_mapcount(page
) > 2)
5290 } else if (!move_file())
5291 /* we ignore mapcount for file pages */
5293 if (!get_page_unless_zero(page
))
5299 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5300 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5303 struct page
*page
= NULL
;
5304 swp_entry_t ent
= pte_to_swp_entry(ptent
);
5306 if (!move_anon() || non_swap_entry(ent
))
5308 usage_count
= mem_cgroup_count_swap_user(ent
, &page
);
5309 if (usage_count
> 1) { /* we don't move shared anon */
5314 if (do_swap_account
)
5315 entry
->val
= ent
.val
;
5320 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
5321 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5323 struct page
*page
= NULL
;
5324 struct inode
*inode
;
5325 struct address_space
*mapping
;
5328 if (!vma
->vm_file
) /* anonymous vma */
5333 inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
5334 mapping
= vma
->vm_file
->f_mapping
;
5335 if (pte_none(ptent
))
5336 pgoff
= linear_page_index(vma
, addr
);
5337 else /* pte_file(ptent) is true */
5338 pgoff
= pte_to_pgoff(ptent
);
5340 /* page is moved even if it's not RSS of this task(page-faulted). */
5341 if (!mapping_cap_swap_backed(mapping
)) { /* normal file */
5342 page
= find_get_page(mapping
, pgoff
);
5343 } else { /* shmem/tmpfs file. we should take account of swap too. */
5345 mem_cgroup_get_shmem_target(inode
, pgoff
, &page
, &ent
);
5346 if (do_swap_account
)
5347 entry
->val
= ent
.val
;
5353 static int is_target_pte_for_mc(struct vm_area_struct
*vma
,
5354 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
5356 struct page
*page
= NULL
;
5357 struct page_cgroup
*pc
;
5359 swp_entry_t ent
= { .val
= 0 };
5361 if (pte_present(ptent
))
5362 page
= mc_handle_present_pte(vma
, addr
, ptent
);
5363 else if (is_swap_pte(ptent
))
5364 page
= mc_handle_swap_pte(vma
, addr
, ptent
, &ent
);
5365 else if (pte_none(ptent
) || pte_file(ptent
))
5366 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
5368 if (!page
&& !ent
.val
)
5371 pc
= lookup_page_cgroup(page
);
5373 * Do only loose check w/o page_cgroup lock.
5374 * mem_cgroup_move_account() checks the pc is valid or not under
5377 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5378 ret
= MC_TARGET_PAGE
;
5380 target
->page
= page
;
5382 if (!ret
|| !target
)
5385 /* There is a swap entry and a page doesn't exist or isn't charged */
5386 if (ent
.val
&& !ret
&&
5387 css_id(&mc
.from
->css
) == lookup_swap_cgroup(ent
)) {
5388 ret
= MC_TARGET_SWAP
;
5395 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
5396 unsigned long addr
, unsigned long end
,
5397 struct mm_walk
*walk
)
5399 struct vm_area_struct
*vma
= walk
->private;
5403 split_huge_page_pmd(walk
->mm
, pmd
);
5405 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5406 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
5407 if (is_target_pte_for_mc(vma
, addr
, *pte
, NULL
))
5408 mc
.precharge
++; /* increment precharge temporarily */
5409 pte_unmap_unlock(pte
- 1, ptl
);
5415 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
5417 unsigned long precharge
;
5418 struct vm_area_struct
*vma
;
5420 down_read(&mm
->mmap_sem
);
5421 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5422 struct mm_walk mem_cgroup_count_precharge_walk
= {
5423 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
5427 if (is_vm_hugetlb_page(vma
))
5429 walk_page_range(vma
->vm_start
, vma
->vm_end
,
5430 &mem_cgroup_count_precharge_walk
);
5432 up_read(&mm
->mmap_sem
);
5434 precharge
= mc
.precharge
;
5440 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
5442 unsigned long precharge
= mem_cgroup_count_precharge(mm
);
5444 VM_BUG_ON(mc
.moving_task
);
5445 mc
.moving_task
= current
;
5446 return mem_cgroup_do_precharge(precharge
);
5449 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5450 static void __mem_cgroup_clear_mc(void)
5452 struct mem_cgroup
*from
= mc
.from
;
5453 struct mem_cgroup
*to
= mc
.to
;
5455 /* we must uncharge all the leftover precharges from mc.to */
5457 __mem_cgroup_cancel_charge(mc
.to
, mc
.precharge
);
5461 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5462 * we must uncharge here.
5464 if (mc
.moved_charge
) {
5465 __mem_cgroup_cancel_charge(mc
.from
, mc
.moved_charge
);
5466 mc
.moved_charge
= 0;
5468 /* we must fixup refcnts and charges */
5469 if (mc
.moved_swap
) {
5470 /* uncharge swap account from the old cgroup */
5471 if (!mem_cgroup_is_root(mc
.from
))
5472 res_counter_uncharge(&mc
.from
->memsw
,
5473 PAGE_SIZE
* mc
.moved_swap
);
5474 __mem_cgroup_put(mc
.from
, mc
.moved_swap
);
5476 if (!mem_cgroup_is_root(mc
.to
)) {
5478 * we charged both to->res and to->memsw, so we should
5481 res_counter_uncharge(&mc
.to
->res
,
5482 PAGE_SIZE
* mc
.moved_swap
);
5484 /* we've already done mem_cgroup_get(mc.to) */
5487 memcg_oom_recover(from
);
5488 memcg_oom_recover(to
);
5489 wake_up_all(&mc
.waitq
);
5492 static void mem_cgroup_clear_mc(void)
5494 struct mem_cgroup
*from
= mc
.from
;
5497 * we must clear moving_task before waking up waiters at the end of
5500 mc
.moving_task
= NULL
;
5501 __mem_cgroup_clear_mc();
5502 spin_lock(&mc
.lock
);
5505 spin_unlock(&mc
.lock
);
5506 mem_cgroup_end_move(from
);
5509 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
5510 struct cgroup
*cgroup
,
5511 struct task_struct
*p
)
5514 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgroup
);
5516 if (mem
->move_charge_at_immigrate
) {
5517 struct mm_struct
*mm
;
5518 struct mem_cgroup
*from
= mem_cgroup_from_task(p
);
5520 VM_BUG_ON(from
== mem
);
5522 mm
= get_task_mm(p
);
5525 /* We move charges only when we move a owner of the mm */
5526 if (mm
->owner
== p
) {
5529 VM_BUG_ON(mc
.precharge
);
5530 VM_BUG_ON(mc
.moved_charge
);
5531 VM_BUG_ON(mc
.moved_swap
);
5532 mem_cgroup_start_move(from
);
5533 spin_lock(&mc
.lock
);
5536 spin_unlock(&mc
.lock
);
5537 /* We set mc.moving_task later */
5539 ret
= mem_cgroup_precharge_mc(mm
);
5541 mem_cgroup_clear_mc();
5548 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
5549 struct cgroup
*cgroup
,
5550 struct task_struct
*p
)
5552 mem_cgroup_clear_mc();
5555 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
5556 unsigned long addr
, unsigned long end
,
5557 struct mm_walk
*walk
)
5560 struct vm_area_struct
*vma
= walk
->private;
5564 split_huge_page_pmd(walk
->mm
, pmd
);
5566 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5567 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
5568 pte_t ptent
= *(pte
++);
5569 union mc_target target
;
5572 struct page_cgroup
*pc
;
5578 type
= is_target_pte_for_mc(vma
, addr
, ptent
, &target
);
5580 case MC_TARGET_PAGE
:
5582 if (isolate_lru_page(page
))
5584 pc
= lookup_page_cgroup(page
);
5585 if (!mem_cgroup_move_account(page
, 1, pc
,
5586 mc
.from
, mc
.to
, false)) {
5588 /* we uncharge from mc.from later. */
5591 putback_lru_page(page
);
5592 put
: /* is_target_pte_for_mc() gets the page */
5595 case MC_TARGET_SWAP
:
5597 if (!mem_cgroup_move_swap_account(ent
,
5598 mc
.from
, mc
.to
, false)) {
5600 /* we fixup refcnts and charges later. */
5608 pte_unmap_unlock(pte
- 1, ptl
);
5613 * We have consumed all precharges we got in can_attach().
5614 * We try charge one by one, but don't do any additional
5615 * charges to mc.to if we have failed in charge once in attach()
5618 ret
= mem_cgroup_do_precharge(1);
5626 static void mem_cgroup_move_charge(struct mm_struct
*mm
)
5628 struct vm_area_struct
*vma
;
5630 lru_add_drain_all();
5632 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
5634 * Someone who are holding the mmap_sem might be waiting in
5635 * waitq. So we cancel all extra charges, wake up all waiters,
5636 * and retry. Because we cancel precharges, we might not be able
5637 * to move enough charges, but moving charge is a best-effort
5638 * feature anyway, so it wouldn't be a big problem.
5640 __mem_cgroup_clear_mc();
5644 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5646 struct mm_walk mem_cgroup_move_charge_walk
= {
5647 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
5651 if (is_vm_hugetlb_page(vma
))
5653 ret
= walk_page_range(vma
->vm_start
, vma
->vm_end
,
5654 &mem_cgroup_move_charge_walk
);
5657 * means we have consumed all precharges and failed in
5658 * doing additional charge. Just abandon here.
5662 up_read(&mm
->mmap_sem
);
5665 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
5666 struct cgroup
*cont
,
5667 struct cgroup
*old_cont
,
5668 struct task_struct
*p
)
5670 struct mm_struct
*mm
= get_task_mm(p
);
5674 mem_cgroup_move_charge(mm
);
5679 mem_cgroup_clear_mc();
5681 #else /* !CONFIG_MMU */
5682 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
5683 struct cgroup
*cgroup
,
5684 struct task_struct
*p
)
5688 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
5689 struct cgroup
*cgroup
,
5690 struct task_struct
*p
)
5693 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
5694 struct cgroup
*cont
,
5695 struct cgroup
*old_cont
,
5696 struct task_struct
*p
)
5701 struct cgroup_subsys mem_cgroup_subsys
= {
5703 .subsys_id
= mem_cgroup_subsys_id
,
5704 .create
= mem_cgroup_create
,
5705 .pre_destroy
= mem_cgroup_pre_destroy
,
5706 .destroy
= mem_cgroup_destroy
,
5707 .populate
= mem_cgroup_populate
,
5708 .can_attach
= mem_cgroup_can_attach
,
5709 .cancel_attach
= mem_cgroup_cancel_attach
,
5710 .attach
= mem_cgroup_move_task
,
5715 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
5716 static int __init
enable_swap_account(char *s
)
5718 /* consider enabled if no parameter or 1 is given */
5719 if (!strcmp(s
, "1"))
5720 really_do_swap_account
= 1;
5721 else if (!strcmp(s
, "0"))
5722 really_do_swap_account
= 0;
5725 __setup("swapaccount=", enable_swap_account
);