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/slab.h>
39 #include <linux/swap.h>
40 #include <linux/swapops.h>
41 #include <linux/spinlock.h>
42 #include <linux/eventfd.h>
43 #include <linux/sort.h>
45 #include <linux/seq_file.h>
46 #include <linux/vmalloc.h>
47 #include <linux/mm_inline.h>
48 #include <linux/page_cgroup.h>
49 #include <linux/cpu.h>
50 #include <linux/oom.h>
53 #include <asm/uaccess.h>
55 #include <trace/events/vmscan.h>
57 struct cgroup_subsys mem_cgroup_subsys __read_mostly
;
58 #define MEM_CGROUP_RECLAIM_RETRIES 5
59 struct mem_cgroup
*root_mem_cgroup __read_mostly
;
61 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
62 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
63 int do_swap_account __read_mostly
;
65 /* for remember boot option*/
66 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED
67 static int really_do_swap_account __initdata
= 1;
69 static int really_do_swap_account __initdata
= 0;
73 #define do_swap_account (0)
77 * Per memcg event counter is incremented at every pagein/pageout. This counter
78 * is used for trigger some periodic events. This is straightforward and better
79 * than using jiffies etc. to handle periodic memcg event.
81 * These values will be used as !((event) & ((1 <<(thresh)) - 1))
83 #define THRESHOLDS_EVENTS_THRESH (7) /* once in 128 */
84 #define SOFTLIMIT_EVENTS_THRESH (10) /* once in 1024 */
87 * Statistics for memory cgroup.
89 enum mem_cgroup_stat_index
{
91 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
93 MEM_CGROUP_STAT_CACHE
, /* # of pages charged as cache */
94 MEM_CGROUP_STAT_RSS
, /* # of pages charged as anon rss */
95 MEM_CGROUP_STAT_FILE_MAPPED
, /* # of pages charged as file rss */
96 MEM_CGROUP_STAT_SWAPOUT
, /* # of pages, swapped out */
97 MEM_CGROUP_STAT_DATA
, /* end of data requires synchronization */
98 MEM_CGROUP_ON_MOVE
, /* someone is moving account between groups */
99 MEM_CGROUP_STAT_NSTATS
,
102 enum mem_cgroup_events_index
{
103 MEM_CGROUP_EVENTS_PGPGIN
, /* # of pages paged in */
104 MEM_CGROUP_EVENTS_PGPGOUT
, /* # of pages paged out */
105 MEM_CGROUP_EVENTS_COUNT
, /* # of pages paged in/out */
106 MEM_CGROUP_EVENTS_NSTATS
,
109 struct mem_cgroup_stat_cpu
{
110 s64 count
[MEM_CGROUP_STAT_NSTATS
];
111 unsigned long events
[MEM_CGROUP_EVENTS_NSTATS
];
115 * per-zone information in memory controller.
117 struct mem_cgroup_per_zone
{
119 * spin_lock to protect the per cgroup LRU
121 struct list_head lists
[NR_LRU_LISTS
];
122 unsigned long count
[NR_LRU_LISTS
];
124 struct zone_reclaim_stat reclaim_stat
;
125 struct rb_node tree_node
; /* RB tree node */
126 unsigned long long usage_in_excess
;/* Set to the value by which */
127 /* the soft limit is exceeded*/
129 struct mem_cgroup
*mem
; /* Back pointer, we cannot */
130 /* use container_of */
132 /* Macro for accessing counter */
133 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
135 struct mem_cgroup_per_node
{
136 struct mem_cgroup_per_zone zoneinfo
[MAX_NR_ZONES
];
139 struct mem_cgroup_lru_info
{
140 struct mem_cgroup_per_node
*nodeinfo
[MAX_NUMNODES
];
144 * Cgroups above their limits are maintained in a RB-Tree, independent of
145 * their hierarchy representation
148 struct mem_cgroup_tree_per_zone
{
149 struct rb_root rb_root
;
153 struct mem_cgroup_tree_per_node
{
154 struct mem_cgroup_tree_per_zone rb_tree_per_zone
[MAX_NR_ZONES
];
157 struct mem_cgroup_tree
{
158 struct mem_cgroup_tree_per_node
*rb_tree_per_node
[MAX_NUMNODES
];
161 static struct mem_cgroup_tree soft_limit_tree __read_mostly
;
163 struct mem_cgroup_threshold
{
164 struct eventfd_ctx
*eventfd
;
169 struct mem_cgroup_threshold_ary
{
170 /* An array index points to threshold just below usage. */
171 int current_threshold
;
172 /* Size of entries[] */
174 /* Array of thresholds */
175 struct mem_cgroup_threshold entries
[0];
178 struct mem_cgroup_thresholds
{
179 /* Primary thresholds array */
180 struct mem_cgroup_threshold_ary
*primary
;
182 * Spare threshold array.
183 * This is needed to make mem_cgroup_unregister_event() "never fail".
184 * It must be able to store at least primary->size - 1 entries.
186 struct mem_cgroup_threshold_ary
*spare
;
190 struct mem_cgroup_eventfd_list
{
191 struct list_head list
;
192 struct eventfd_ctx
*eventfd
;
195 static void mem_cgroup_threshold(struct mem_cgroup
*mem
);
196 static void mem_cgroup_oom_notify(struct mem_cgroup
*mem
);
199 * The memory controller data structure. The memory controller controls both
200 * page cache and RSS per cgroup. We would eventually like to provide
201 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
202 * to help the administrator determine what knobs to tune.
204 * TODO: Add a water mark for the memory controller. Reclaim will begin when
205 * we hit the water mark. May be even add a low water mark, such that
206 * no reclaim occurs from a cgroup at it's low water mark, this is
207 * a feature that will be implemented much later in the future.
210 struct cgroup_subsys_state css
;
212 * the counter to account for memory usage
214 struct res_counter res
;
216 * the counter to account for mem+swap usage.
218 struct res_counter memsw
;
220 * Per cgroup active and inactive list, similar to the
221 * per zone LRU lists.
223 struct mem_cgroup_lru_info info
;
225 * While reclaiming in a hierarchy, we cache the last child we
228 int last_scanned_child
;
230 * Should the accounting and control be hierarchical, per subtree?
236 unsigned int swappiness
;
237 /* OOM-Killer disable */
238 int oom_kill_disable
;
240 /* set when res.limit == memsw.limit */
241 bool memsw_is_minimum
;
243 /* protect arrays of thresholds */
244 struct mutex thresholds_lock
;
246 /* thresholds for memory usage. RCU-protected */
247 struct mem_cgroup_thresholds thresholds
;
249 /* thresholds for mem+swap usage. RCU-protected */
250 struct mem_cgroup_thresholds memsw_thresholds
;
252 /* For oom notifier event fd */
253 struct list_head oom_notify
;
256 * Should we move charges of a task when a task is moved into this
257 * mem_cgroup ? And what type of charges should we move ?
259 unsigned long move_charge_at_immigrate
;
263 struct mem_cgroup_stat_cpu
*stat
;
265 * used when a cpu is offlined or other synchronizations
266 * See mem_cgroup_read_stat().
268 struct mem_cgroup_stat_cpu nocpu_base
;
269 spinlock_t pcp_counter_lock
;
272 /* Stuffs for move charges at task migration. */
274 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
275 * left-shifted bitmap of these types.
278 MOVE_CHARGE_TYPE_ANON
, /* private anonymous page and swap of it */
279 MOVE_CHARGE_TYPE_FILE
, /* file page(including tmpfs) and swap of it */
283 /* "mc" and its members are protected by cgroup_mutex */
284 static struct move_charge_struct
{
285 spinlock_t lock
; /* for from, to */
286 struct mem_cgroup
*from
;
287 struct mem_cgroup
*to
;
288 unsigned long precharge
;
289 unsigned long moved_charge
;
290 unsigned long moved_swap
;
291 struct task_struct
*moving_task
; /* a task moving charges */
292 wait_queue_head_t waitq
; /* a waitq for other context */
294 .lock
= __SPIN_LOCK_UNLOCKED(mc
.lock
),
295 .waitq
= __WAIT_QUEUE_HEAD_INITIALIZER(mc
.waitq
),
298 static bool move_anon(void)
300 return test_bit(MOVE_CHARGE_TYPE_ANON
,
301 &mc
.to
->move_charge_at_immigrate
);
304 static bool move_file(void)
306 return test_bit(MOVE_CHARGE_TYPE_FILE
,
307 &mc
.to
->move_charge_at_immigrate
);
311 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
312 * limit reclaim to prevent infinite loops, if they ever occur.
314 #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
315 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
318 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
319 MEM_CGROUP_CHARGE_TYPE_MAPPED
,
320 MEM_CGROUP_CHARGE_TYPE_SHMEM
, /* used by page migration of shmem */
321 MEM_CGROUP_CHARGE_TYPE_FORCE
, /* used by force_empty */
322 MEM_CGROUP_CHARGE_TYPE_SWAPOUT
, /* for accounting swapcache */
323 MEM_CGROUP_CHARGE_TYPE_DROP
, /* a page was unused swap cache */
327 /* for encoding cft->private value on file */
330 #define _OOM_TYPE (2)
331 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
332 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
333 #define MEMFILE_ATTR(val) ((val) & 0xffff)
334 /* Used for OOM nofiier */
335 #define OOM_CONTROL (0)
338 * Reclaim flags for mem_cgroup_hierarchical_reclaim
340 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
341 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
342 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
343 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
344 #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2
345 #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT)
347 static void mem_cgroup_get(struct mem_cgroup
*mem
);
348 static void mem_cgroup_put(struct mem_cgroup
*mem
);
349 static struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*mem
);
350 static void drain_all_stock_async(void);
352 static struct mem_cgroup_per_zone
*
353 mem_cgroup_zoneinfo(struct mem_cgroup
*mem
, int nid
, int zid
)
355 return &mem
->info
.nodeinfo
[nid
]->zoneinfo
[zid
];
358 struct cgroup_subsys_state
*mem_cgroup_css(struct mem_cgroup
*mem
)
363 static struct mem_cgroup_per_zone
*
364 page_cgroup_zoneinfo(struct mem_cgroup
*mem
, struct page
*page
)
366 int nid
= page_to_nid(page
);
367 int zid
= page_zonenum(page
);
369 return mem_cgroup_zoneinfo(mem
, nid
, zid
);
372 static struct mem_cgroup_tree_per_zone
*
373 soft_limit_tree_node_zone(int nid
, int zid
)
375 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
378 static struct mem_cgroup_tree_per_zone
*
379 soft_limit_tree_from_page(struct page
*page
)
381 int nid
= page_to_nid(page
);
382 int zid
= page_zonenum(page
);
384 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
388 __mem_cgroup_insert_exceeded(struct mem_cgroup
*mem
,
389 struct mem_cgroup_per_zone
*mz
,
390 struct mem_cgroup_tree_per_zone
*mctz
,
391 unsigned long long new_usage_in_excess
)
393 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
394 struct rb_node
*parent
= NULL
;
395 struct mem_cgroup_per_zone
*mz_node
;
400 mz
->usage_in_excess
= new_usage_in_excess
;
401 if (!mz
->usage_in_excess
)
405 mz_node
= rb_entry(parent
, struct mem_cgroup_per_zone
,
407 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
)
410 * We can't avoid mem cgroups that are over their soft
411 * limit by the same amount
413 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
416 rb_link_node(&mz
->tree_node
, parent
, p
);
417 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
422 __mem_cgroup_remove_exceeded(struct mem_cgroup
*mem
,
423 struct mem_cgroup_per_zone
*mz
,
424 struct mem_cgroup_tree_per_zone
*mctz
)
428 rb_erase(&mz
->tree_node
, &mctz
->rb_root
);
433 mem_cgroup_remove_exceeded(struct mem_cgroup
*mem
,
434 struct mem_cgroup_per_zone
*mz
,
435 struct mem_cgroup_tree_per_zone
*mctz
)
437 spin_lock(&mctz
->lock
);
438 __mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
439 spin_unlock(&mctz
->lock
);
443 static void mem_cgroup_update_tree(struct mem_cgroup
*mem
, struct page
*page
)
445 unsigned long long excess
;
446 struct mem_cgroup_per_zone
*mz
;
447 struct mem_cgroup_tree_per_zone
*mctz
;
448 int nid
= page_to_nid(page
);
449 int zid
= page_zonenum(page
);
450 mctz
= soft_limit_tree_from_page(page
);
453 * Necessary to update all ancestors when hierarchy is used.
454 * because their event counter is not touched.
456 for (; mem
; mem
= parent_mem_cgroup(mem
)) {
457 mz
= mem_cgroup_zoneinfo(mem
, nid
, zid
);
458 excess
= res_counter_soft_limit_excess(&mem
->res
);
460 * We have to update the tree if mz is on RB-tree or
461 * mem is over its softlimit.
463 if (excess
|| mz
->on_tree
) {
464 spin_lock(&mctz
->lock
);
465 /* if on-tree, remove it */
467 __mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
469 * Insert again. mz->usage_in_excess will be updated.
470 * If excess is 0, no tree ops.
472 __mem_cgroup_insert_exceeded(mem
, mz
, mctz
, excess
);
473 spin_unlock(&mctz
->lock
);
478 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*mem
)
481 struct mem_cgroup_per_zone
*mz
;
482 struct mem_cgroup_tree_per_zone
*mctz
;
484 for_each_node_state(node
, N_POSSIBLE
) {
485 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
486 mz
= mem_cgroup_zoneinfo(mem
, node
, zone
);
487 mctz
= soft_limit_tree_node_zone(node
, zone
);
488 mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
493 static struct mem_cgroup_per_zone
*
494 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
496 struct rb_node
*rightmost
= NULL
;
497 struct mem_cgroup_per_zone
*mz
;
501 rightmost
= rb_last(&mctz
->rb_root
);
503 goto done
; /* Nothing to reclaim from */
505 mz
= rb_entry(rightmost
, struct mem_cgroup_per_zone
, tree_node
);
507 * Remove the node now but someone else can add it back,
508 * we will to add it back at the end of reclaim to its correct
509 * position in the tree.
511 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
512 if (!res_counter_soft_limit_excess(&mz
->mem
->res
) ||
513 !css_tryget(&mz
->mem
->css
))
519 static struct mem_cgroup_per_zone
*
520 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
522 struct mem_cgroup_per_zone
*mz
;
524 spin_lock(&mctz
->lock
);
525 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
526 spin_unlock(&mctz
->lock
);
531 * Implementation Note: reading percpu statistics for memcg.
533 * Both of vmstat[] and percpu_counter has threshold and do periodic
534 * synchronization to implement "quick" read. There are trade-off between
535 * reading cost and precision of value. Then, we may have a chance to implement
536 * a periodic synchronizion of counter in memcg's counter.
538 * But this _read() function is used for user interface now. The user accounts
539 * memory usage by memory cgroup and he _always_ requires exact value because
540 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
541 * have to visit all online cpus and make sum. So, for now, unnecessary
542 * synchronization is not implemented. (just implemented for cpu hotplug)
544 * If there are kernel internal actions which can make use of some not-exact
545 * value, and reading all cpu value can be performance bottleneck in some
546 * common workload, threashold and synchonization as vmstat[] should be
549 static s64
mem_cgroup_read_stat(struct mem_cgroup
*mem
,
550 enum mem_cgroup_stat_index idx
)
556 for_each_online_cpu(cpu
)
557 val
+= per_cpu(mem
->stat
->count
[idx
], cpu
);
558 #ifdef CONFIG_HOTPLUG_CPU
559 spin_lock(&mem
->pcp_counter_lock
);
560 val
+= mem
->nocpu_base
.count
[idx
];
561 spin_unlock(&mem
->pcp_counter_lock
);
567 static s64
mem_cgroup_local_usage(struct mem_cgroup
*mem
)
571 ret
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_RSS
);
572 ret
+= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_CACHE
);
576 static void mem_cgroup_swap_statistics(struct mem_cgroup
*mem
,
579 int val
= (charge
) ? 1 : -1;
580 this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_SWAPOUT
], val
);
583 static unsigned long mem_cgroup_read_events(struct mem_cgroup
*mem
,
584 enum mem_cgroup_events_index idx
)
586 unsigned long val
= 0;
589 for_each_online_cpu(cpu
)
590 val
+= per_cpu(mem
->stat
->events
[idx
], cpu
);
591 #ifdef CONFIG_HOTPLUG_CPU
592 spin_lock(&mem
->pcp_counter_lock
);
593 val
+= mem
->nocpu_base
.events
[idx
];
594 spin_unlock(&mem
->pcp_counter_lock
);
599 static void mem_cgroup_charge_statistics(struct mem_cgroup
*mem
,
600 bool file
, int nr_pages
)
605 __this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_CACHE
], nr_pages
);
607 __this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_RSS
], nr_pages
);
609 /* pagein of a big page is an event. So, ignore page size */
611 __this_cpu_inc(mem
->stat
->events
[MEM_CGROUP_EVENTS_PGPGIN
]);
613 __this_cpu_inc(mem
->stat
->events
[MEM_CGROUP_EVENTS_PGPGOUT
]);
614 nr_pages
= -nr_pages
; /* for event */
617 __this_cpu_add(mem
->stat
->events
[MEM_CGROUP_EVENTS_COUNT
], nr_pages
);
622 static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup
*mem
,
626 struct mem_cgroup_per_zone
*mz
;
629 for_each_online_node(nid
)
630 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
631 mz
= mem_cgroup_zoneinfo(mem
, nid
, zid
);
632 total
+= MEM_CGROUP_ZSTAT(mz
, idx
);
637 static bool __memcg_event_check(struct mem_cgroup
*mem
, int event_mask_shift
)
641 val
= this_cpu_read(mem
->stat
->events
[MEM_CGROUP_EVENTS_COUNT
]);
643 return !(val
& ((1 << event_mask_shift
) - 1));
647 * Check events in order.
650 static void memcg_check_events(struct mem_cgroup
*mem
, struct page
*page
)
652 /* threshold event is triggered in finer grain than soft limit */
653 if (unlikely(__memcg_event_check(mem
, THRESHOLDS_EVENTS_THRESH
))) {
654 mem_cgroup_threshold(mem
);
655 if (unlikely(__memcg_event_check(mem
, SOFTLIMIT_EVENTS_THRESH
)))
656 mem_cgroup_update_tree(mem
, page
);
660 static struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
662 return container_of(cgroup_subsys_state(cont
,
663 mem_cgroup_subsys_id
), struct mem_cgroup
,
667 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
670 * mm_update_next_owner() may clear mm->owner to NULL
671 * if it races with swapoff, page migration, etc.
672 * So this can be called with p == NULL.
677 return container_of(task_subsys_state(p
, mem_cgroup_subsys_id
),
678 struct mem_cgroup
, css
);
681 static struct mem_cgroup
*try_get_mem_cgroup_from_mm(struct mm_struct
*mm
)
683 struct mem_cgroup
*mem
= NULL
;
688 * Because we have no locks, mm->owner's may be being moved to other
689 * cgroup. We use css_tryget() here even if this looks
690 * pessimistic (rather than adding locks here).
694 mem
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
697 } while (!css_tryget(&mem
->css
));
702 /* The caller has to guarantee "mem" exists before calling this */
703 static struct mem_cgroup
*mem_cgroup_start_loop(struct mem_cgroup
*mem
)
705 struct cgroup_subsys_state
*css
;
708 if (!mem
) /* ROOT cgroup has the smallest ID */
709 return root_mem_cgroup
; /*css_put/get against root is ignored*/
710 if (!mem
->use_hierarchy
) {
711 if (css_tryget(&mem
->css
))
717 * searching a memory cgroup which has the smallest ID under given
718 * ROOT cgroup. (ID >= 1)
720 css
= css_get_next(&mem_cgroup_subsys
, 1, &mem
->css
, &found
);
721 if (css
&& css_tryget(css
))
722 mem
= container_of(css
, struct mem_cgroup
, css
);
729 static struct mem_cgroup
*mem_cgroup_get_next(struct mem_cgroup
*iter
,
730 struct mem_cgroup
*root
,
733 int nextid
= css_id(&iter
->css
) + 1;
736 struct cgroup_subsys_state
*css
;
738 hierarchy_used
= iter
->use_hierarchy
;
741 /* If no ROOT, walk all, ignore hierarchy */
742 if (!cond
|| (root
&& !hierarchy_used
))
746 root
= root_mem_cgroup
;
752 css
= css_get_next(&mem_cgroup_subsys
, nextid
,
754 if (css
&& css_tryget(css
))
755 iter
= container_of(css
, struct mem_cgroup
, css
);
757 /* If css is NULL, no more cgroups will be found */
759 } while (css
&& !iter
);
764 * for_eacn_mem_cgroup_tree() for visiting all cgroup under tree. Please
765 * be careful that "break" loop is not allowed. We have reference count.
766 * Instead of that modify "cond" to be false and "continue" to exit the loop.
768 #define for_each_mem_cgroup_tree_cond(iter, root, cond) \
769 for (iter = mem_cgroup_start_loop(root);\
771 iter = mem_cgroup_get_next(iter, root, cond))
773 #define for_each_mem_cgroup_tree(iter, root) \
774 for_each_mem_cgroup_tree_cond(iter, root, true)
776 #define for_each_mem_cgroup_all(iter) \
777 for_each_mem_cgroup_tree_cond(iter, NULL, true)
780 static inline bool mem_cgroup_is_root(struct mem_cgroup
*mem
)
782 return (mem
== root_mem_cgroup
);
786 * Following LRU functions are allowed to be used without PCG_LOCK.
787 * Operations are called by routine of global LRU independently from memcg.
788 * What we have to take care of here is validness of pc->mem_cgroup.
790 * Changes to pc->mem_cgroup happens when
793 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
794 * It is added to LRU before charge.
795 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
796 * When moving account, the page is not on LRU. It's isolated.
799 void mem_cgroup_del_lru_list(struct page
*page
, enum lru_list lru
)
801 struct page_cgroup
*pc
;
802 struct mem_cgroup_per_zone
*mz
;
804 if (mem_cgroup_disabled())
806 pc
= lookup_page_cgroup(page
);
807 /* can happen while we handle swapcache. */
808 if (!TestClearPageCgroupAcctLRU(pc
))
810 VM_BUG_ON(!pc
->mem_cgroup
);
812 * We don't check PCG_USED bit. It's cleared when the "page" is finally
813 * removed from global LRU.
815 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
816 /* huge page split is done under lru_lock. so, we have no races. */
817 MEM_CGROUP_ZSTAT(mz
, lru
) -= 1 << compound_order(page
);
818 if (mem_cgroup_is_root(pc
->mem_cgroup
))
820 VM_BUG_ON(list_empty(&pc
->lru
));
821 list_del_init(&pc
->lru
);
824 void mem_cgroup_del_lru(struct page
*page
)
826 mem_cgroup_del_lru_list(page
, page_lru(page
));
830 * Writeback is about to end against a page which has been marked for immediate
831 * reclaim. If it still appears to be reclaimable, move it to the tail of the
834 void mem_cgroup_rotate_reclaimable_page(struct page
*page
)
836 struct mem_cgroup_per_zone
*mz
;
837 struct page_cgroup
*pc
;
838 enum lru_list lru
= page_lru(page
);
840 if (mem_cgroup_disabled())
843 pc
= lookup_page_cgroup(page
);
844 /* unused or root page is not rotated. */
845 if (!PageCgroupUsed(pc
))
847 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
849 if (mem_cgroup_is_root(pc
->mem_cgroup
))
851 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
852 list_move_tail(&pc
->lru
, &mz
->lists
[lru
]);
855 void mem_cgroup_rotate_lru_list(struct page
*page
, enum lru_list lru
)
857 struct mem_cgroup_per_zone
*mz
;
858 struct page_cgroup
*pc
;
860 if (mem_cgroup_disabled())
863 pc
= lookup_page_cgroup(page
);
864 /* unused or root page is not rotated. */
865 if (!PageCgroupUsed(pc
))
867 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
869 if (mem_cgroup_is_root(pc
->mem_cgroup
))
871 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
872 list_move(&pc
->lru
, &mz
->lists
[lru
]);
875 void mem_cgroup_add_lru_list(struct page
*page
, enum lru_list lru
)
877 struct page_cgroup
*pc
;
878 struct mem_cgroup_per_zone
*mz
;
880 if (mem_cgroup_disabled())
882 pc
= lookup_page_cgroup(page
);
883 VM_BUG_ON(PageCgroupAcctLRU(pc
));
884 if (!PageCgroupUsed(pc
))
886 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
888 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
889 /* huge page split is done under lru_lock. so, we have no races. */
890 MEM_CGROUP_ZSTAT(mz
, lru
) += 1 << compound_order(page
);
891 SetPageCgroupAcctLRU(pc
);
892 if (mem_cgroup_is_root(pc
->mem_cgroup
))
894 list_add(&pc
->lru
, &mz
->lists
[lru
]);
898 * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to
899 * lru because the page may.be reused after it's fully uncharged (because of
900 * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge
901 * it again. This function is only used to charge SwapCache. It's done under
902 * lock_page and expected that zone->lru_lock is never held.
904 static void mem_cgroup_lru_del_before_commit_swapcache(struct page
*page
)
907 struct zone
*zone
= page_zone(page
);
908 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
910 spin_lock_irqsave(&zone
->lru_lock
, flags
);
912 * Forget old LRU when this page_cgroup is *not* used. This Used bit
913 * is guarded by lock_page() because the page is SwapCache.
915 if (!PageCgroupUsed(pc
))
916 mem_cgroup_del_lru_list(page
, page_lru(page
));
917 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
920 static void mem_cgroup_lru_add_after_commit_swapcache(struct page
*page
)
923 struct zone
*zone
= page_zone(page
);
924 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
926 spin_lock_irqsave(&zone
->lru_lock
, flags
);
927 /* link when the page is linked to LRU but page_cgroup isn't */
928 if (PageLRU(page
) && !PageCgroupAcctLRU(pc
))
929 mem_cgroup_add_lru_list(page
, page_lru(page
));
930 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
934 void mem_cgroup_move_lists(struct page
*page
,
935 enum lru_list from
, enum lru_list to
)
937 if (mem_cgroup_disabled())
939 mem_cgroup_del_lru_list(page
, from
);
940 mem_cgroup_add_lru_list(page
, to
);
943 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*mem
)
946 struct mem_cgroup
*curr
= NULL
;
947 struct task_struct
*p
;
949 p
= find_lock_task_mm(task
);
952 curr
= try_get_mem_cgroup_from_mm(p
->mm
);
957 * We should check use_hierarchy of "mem" not "curr". Because checking
958 * use_hierarchy of "curr" here make this function true if hierarchy is
959 * enabled in "curr" and "curr" is a child of "mem" in *cgroup*
960 * hierarchy(even if use_hierarchy is disabled in "mem").
962 if (mem
->use_hierarchy
)
963 ret
= css_is_ancestor(&curr
->css
, &mem
->css
);
970 static int calc_inactive_ratio(struct mem_cgroup
*memcg
, unsigned long *present_pages
)
972 unsigned long active
;
973 unsigned long inactive
;
975 unsigned long inactive_ratio
;
977 inactive
= mem_cgroup_get_local_zonestat(memcg
, LRU_INACTIVE_ANON
);
978 active
= mem_cgroup_get_local_zonestat(memcg
, LRU_ACTIVE_ANON
);
980 gb
= (inactive
+ active
) >> (30 - PAGE_SHIFT
);
982 inactive_ratio
= int_sqrt(10 * gb
);
987 present_pages
[0] = inactive
;
988 present_pages
[1] = active
;
991 return inactive_ratio
;
994 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup
*memcg
)
996 unsigned long active
;
997 unsigned long inactive
;
998 unsigned long present_pages
[2];
999 unsigned long inactive_ratio
;
1001 inactive_ratio
= calc_inactive_ratio(memcg
, present_pages
);
1003 inactive
= present_pages
[0];
1004 active
= present_pages
[1];
1006 if (inactive
* inactive_ratio
< active
)
1012 int mem_cgroup_inactive_file_is_low(struct mem_cgroup
*memcg
)
1014 unsigned long active
;
1015 unsigned long inactive
;
1017 inactive
= mem_cgroup_get_local_zonestat(memcg
, LRU_INACTIVE_FILE
);
1018 active
= mem_cgroup_get_local_zonestat(memcg
, LRU_ACTIVE_FILE
);
1020 return (active
> inactive
);
1023 unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup
*memcg
,
1027 int nid
= zone_to_nid(zone
);
1028 int zid
= zone_idx(zone
);
1029 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
1031 return MEM_CGROUP_ZSTAT(mz
, lru
);
1034 struct zone_reclaim_stat
*mem_cgroup_get_reclaim_stat(struct mem_cgroup
*memcg
,
1037 int nid
= zone_to_nid(zone
);
1038 int zid
= zone_idx(zone
);
1039 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
1041 return &mz
->reclaim_stat
;
1044 struct zone_reclaim_stat
*
1045 mem_cgroup_get_reclaim_stat_from_page(struct page
*page
)
1047 struct page_cgroup
*pc
;
1048 struct mem_cgroup_per_zone
*mz
;
1050 if (mem_cgroup_disabled())
1053 pc
= lookup_page_cgroup(page
);
1054 if (!PageCgroupUsed(pc
))
1056 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1058 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
1059 return &mz
->reclaim_stat
;
1062 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan
,
1063 struct list_head
*dst
,
1064 unsigned long *scanned
, int order
,
1065 int mode
, struct zone
*z
,
1066 struct mem_cgroup
*mem_cont
,
1067 int active
, int file
)
1069 unsigned long nr_taken
= 0;
1073 struct list_head
*src
;
1074 struct page_cgroup
*pc
, *tmp
;
1075 int nid
= zone_to_nid(z
);
1076 int zid
= zone_idx(z
);
1077 struct mem_cgroup_per_zone
*mz
;
1078 int lru
= LRU_FILE
* file
+ active
;
1082 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
1083 src
= &mz
->lists
[lru
];
1086 list_for_each_entry_safe_reverse(pc
, tmp
, src
, lru
) {
1087 if (scan
>= nr_to_scan
)
1090 if (unlikely(!PageCgroupUsed(pc
)))
1093 page
= lookup_cgroup_page(pc
);
1095 if (unlikely(!PageLRU(page
)))
1099 ret
= __isolate_lru_page(page
, mode
, file
);
1102 list_move(&page
->lru
, dst
);
1103 mem_cgroup_del_lru(page
);
1104 nr_taken
+= hpage_nr_pages(page
);
1107 /* we don't affect global LRU but rotate in our LRU */
1108 mem_cgroup_rotate_lru_list(page
, page_lru(page
));
1117 trace_mm_vmscan_memcg_isolate(0, nr_to_scan
, scan
, nr_taken
,
1123 #define mem_cgroup_from_res_counter(counter, member) \
1124 container_of(counter, struct mem_cgroup, member)
1127 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1128 * @mem: the memory cgroup
1130 * Returns the maximum amount of memory @mem can be charged with, in
1133 static unsigned long mem_cgroup_margin(struct mem_cgroup
*mem
)
1135 unsigned long long margin
;
1137 margin
= res_counter_margin(&mem
->res
);
1138 if (do_swap_account
)
1139 margin
= min(margin
, res_counter_margin(&mem
->memsw
));
1140 return margin
>> PAGE_SHIFT
;
1143 static unsigned int get_swappiness(struct mem_cgroup
*memcg
)
1145 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
1148 if (cgrp
->parent
== NULL
)
1149 return vm_swappiness
;
1151 return memcg
->swappiness
;
1154 static void mem_cgroup_start_move(struct mem_cgroup
*mem
)
1159 spin_lock(&mem
->pcp_counter_lock
);
1160 for_each_online_cpu(cpu
)
1161 per_cpu(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) += 1;
1162 mem
->nocpu_base
.count
[MEM_CGROUP_ON_MOVE
] += 1;
1163 spin_unlock(&mem
->pcp_counter_lock
);
1169 static void mem_cgroup_end_move(struct mem_cgroup
*mem
)
1176 spin_lock(&mem
->pcp_counter_lock
);
1177 for_each_online_cpu(cpu
)
1178 per_cpu(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) -= 1;
1179 mem
->nocpu_base
.count
[MEM_CGROUP_ON_MOVE
] -= 1;
1180 spin_unlock(&mem
->pcp_counter_lock
);
1184 * 2 routines for checking "mem" is under move_account() or not.
1186 * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used
1187 * for avoiding race in accounting. If true,
1188 * pc->mem_cgroup may be overwritten.
1190 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1191 * under hierarchy of moving cgroups. This is for
1192 * waiting at hith-memory prressure caused by "move".
1195 static bool mem_cgroup_stealed(struct mem_cgroup
*mem
)
1197 VM_BUG_ON(!rcu_read_lock_held());
1198 return this_cpu_read(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
]) > 0;
1201 static bool mem_cgroup_under_move(struct mem_cgroup
*mem
)
1203 struct mem_cgroup
*from
;
1204 struct mem_cgroup
*to
;
1207 * Unlike task_move routines, we access mc.to, mc.from not under
1208 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1210 spin_lock(&mc
.lock
);
1215 if (from
== mem
|| to
== mem
1216 || (mem
->use_hierarchy
&& css_is_ancestor(&from
->css
, &mem
->css
))
1217 || (mem
->use_hierarchy
&& css_is_ancestor(&to
->css
, &mem
->css
)))
1220 spin_unlock(&mc
.lock
);
1224 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*mem
)
1226 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1227 if (mem_cgroup_under_move(mem
)) {
1229 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1230 /* moving charge context might have finished. */
1233 finish_wait(&mc
.waitq
, &wait
);
1241 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1242 * @memcg: The memory cgroup that went over limit
1243 * @p: Task that is going to be killed
1245 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1248 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1250 struct cgroup
*task_cgrp
;
1251 struct cgroup
*mem_cgrp
;
1253 * Need a buffer in BSS, can't rely on allocations. The code relies
1254 * on the assumption that OOM is serialized for memory controller.
1255 * If this assumption is broken, revisit this code.
1257 static char memcg_name
[PATH_MAX
];
1266 mem_cgrp
= memcg
->css
.cgroup
;
1267 task_cgrp
= task_cgroup(p
, mem_cgroup_subsys_id
);
1269 ret
= cgroup_path(task_cgrp
, memcg_name
, PATH_MAX
);
1272 * Unfortunately, we are unable to convert to a useful name
1273 * But we'll still print out the usage information
1280 printk(KERN_INFO
"Task in %s killed", memcg_name
);
1283 ret
= cgroup_path(mem_cgrp
, memcg_name
, PATH_MAX
);
1291 * Continues from above, so we don't need an KERN_ level
1293 printk(KERN_CONT
" as a result of limit of %s\n", memcg_name
);
1296 printk(KERN_INFO
"memory: usage %llukB, limit %llukB, failcnt %llu\n",
1297 res_counter_read_u64(&memcg
->res
, RES_USAGE
) >> 10,
1298 res_counter_read_u64(&memcg
->res
, RES_LIMIT
) >> 10,
1299 res_counter_read_u64(&memcg
->res
, RES_FAILCNT
));
1300 printk(KERN_INFO
"memory+swap: usage %llukB, limit %llukB, "
1302 res_counter_read_u64(&memcg
->memsw
, RES_USAGE
) >> 10,
1303 res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
) >> 10,
1304 res_counter_read_u64(&memcg
->memsw
, RES_FAILCNT
));
1308 * This function returns the number of memcg under hierarchy tree. Returns
1309 * 1(self count) if no children.
1311 static int mem_cgroup_count_children(struct mem_cgroup
*mem
)
1314 struct mem_cgroup
*iter
;
1316 for_each_mem_cgroup_tree(iter
, mem
)
1322 * Return the memory (and swap, if configured) limit for a memcg.
1324 u64
mem_cgroup_get_limit(struct mem_cgroup
*memcg
)
1329 limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
1330 limit
+= total_swap_pages
<< PAGE_SHIFT
;
1332 memsw
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
1334 * If memsw is finite and limits the amount of swap space available
1335 * to this memcg, return that limit.
1337 return min(limit
, memsw
);
1341 * Visit the first child (need not be the first child as per the ordering
1342 * of the cgroup list, since we track last_scanned_child) of @mem and use
1343 * that to reclaim free pages from.
1345 static struct mem_cgroup
*
1346 mem_cgroup_select_victim(struct mem_cgroup
*root_mem
)
1348 struct mem_cgroup
*ret
= NULL
;
1349 struct cgroup_subsys_state
*css
;
1352 if (!root_mem
->use_hierarchy
) {
1353 css_get(&root_mem
->css
);
1359 nextid
= root_mem
->last_scanned_child
+ 1;
1360 css
= css_get_next(&mem_cgroup_subsys
, nextid
, &root_mem
->css
,
1362 if (css
&& css_tryget(css
))
1363 ret
= container_of(css
, struct mem_cgroup
, css
);
1366 /* Updates scanning parameter */
1368 /* this means start scan from ID:1 */
1369 root_mem
->last_scanned_child
= 0;
1371 root_mem
->last_scanned_child
= found
;
1378 * Scan the hierarchy if needed to reclaim memory. We remember the last child
1379 * we reclaimed from, so that we don't end up penalizing one child extensively
1380 * based on its position in the children list.
1382 * root_mem is the original ancestor that we've been reclaim from.
1384 * We give up and return to the caller when we visit root_mem twice.
1385 * (other groups can be removed while we're walking....)
1387 * If shrink==true, for avoiding to free too much, this returns immedieately.
1389 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup
*root_mem
,
1392 unsigned long reclaim_options
)
1394 struct mem_cgroup
*victim
;
1397 bool noswap
= reclaim_options
& MEM_CGROUP_RECLAIM_NOSWAP
;
1398 bool shrink
= reclaim_options
& MEM_CGROUP_RECLAIM_SHRINK
;
1399 bool check_soft
= reclaim_options
& MEM_CGROUP_RECLAIM_SOFT
;
1400 unsigned long excess
;
1402 excess
= res_counter_soft_limit_excess(&root_mem
->res
) >> PAGE_SHIFT
;
1404 /* If memsw_is_minimum==1, swap-out is of-no-use. */
1405 if (root_mem
->memsw_is_minimum
)
1409 victim
= mem_cgroup_select_victim(root_mem
);
1410 if (victim
== root_mem
) {
1413 drain_all_stock_async();
1416 * If we have not been able to reclaim
1417 * anything, it might because there are
1418 * no reclaimable pages under this hierarchy
1420 if (!check_soft
|| !total
) {
1421 css_put(&victim
->css
);
1425 * We want to do more targetted reclaim.
1426 * excess >> 2 is not to excessive so as to
1427 * reclaim too much, nor too less that we keep
1428 * coming back to reclaim from this cgroup
1430 if (total
>= (excess
>> 2) ||
1431 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
)) {
1432 css_put(&victim
->css
);
1437 if (!mem_cgroup_local_usage(victim
)) {
1438 /* this cgroup's local usage == 0 */
1439 css_put(&victim
->css
);
1442 /* we use swappiness of local cgroup */
1444 ret
= mem_cgroup_shrink_node_zone(victim
, gfp_mask
,
1445 noswap
, get_swappiness(victim
), zone
);
1447 ret
= try_to_free_mem_cgroup_pages(victim
, gfp_mask
,
1448 noswap
, get_swappiness(victim
));
1449 css_put(&victim
->css
);
1451 * At shrinking usage, we can't check we should stop here or
1452 * reclaim more. It's depends on callers. last_scanned_child
1453 * will work enough for keeping fairness under tree.
1459 if (!res_counter_soft_limit_excess(&root_mem
->res
))
1461 } else if (mem_cgroup_margin(root_mem
))
1468 * Check OOM-Killer is already running under our hierarchy.
1469 * If someone is running, return false.
1471 static bool mem_cgroup_oom_lock(struct mem_cgroup
*mem
)
1473 int x
, lock_count
= 0;
1474 struct mem_cgroup
*iter
;
1476 for_each_mem_cgroup_tree(iter
, mem
) {
1477 x
= atomic_inc_return(&iter
->oom_lock
);
1478 lock_count
= max(x
, lock_count
);
1481 if (lock_count
== 1)
1486 static int mem_cgroup_oom_unlock(struct mem_cgroup
*mem
)
1488 struct mem_cgroup
*iter
;
1491 * When a new child is created while the hierarchy is under oom,
1492 * mem_cgroup_oom_lock() may not be called. We have to use
1493 * atomic_add_unless() here.
1495 for_each_mem_cgroup_tree(iter
, mem
)
1496 atomic_add_unless(&iter
->oom_lock
, -1, 0);
1501 static DEFINE_MUTEX(memcg_oom_mutex
);
1502 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1504 struct oom_wait_info
{
1505 struct mem_cgroup
*mem
;
1509 static int memcg_oom_wake_function(wait_queue_t
*wait
,
1510 unsigned mode
, int sync
, void *arg
)
1512 struct mem_cgroup
*wake_mem
= (struct mem_cgroup
*)arg
;
1513 struct oom_wait_info
*oom_wait_info
;
1515 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1517 if (oom_wait_info
->mem
== wake_mem
)
1519 /* if no hierarchy, no match */
1520 if (!oom_wait_info
->mem
->use_hierarchy
|| !wake_mem
->use_hierarchy
)
1523 * Both of oom_wait_info->mem and wake_mem are stable under us.
1524 * Then we can use css_is_ancestor without taking care of RCU.
1526 if (!css_is_ancestor(&oom_wait_info
->mem
->css
, &wake_mem
->css
) &&
1527 !css_is_ancestor(&wake_mem
->css
, &oom_wait_info
->mem
->css
))
1531 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1534 static void memcg_wakeup_oom(struct mem_cgroup
*mem
)
1536 /* for filtering, pass "mem" as argument. */
1537 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, mem
);
1540 static void memcg_oom_recover(struct mem_cgroup
*mem
)
1542 if (mem
&& atomic_read(&mem
->oom_lock
))
1543 memcg_wakeup_oom(mem
);
1547 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1549 bool mem_cgroup_handle_oom(struct mem_cgroup
*mem
, gfp_t mask
)
1551 struct oom_wait_info owait
;
1552 bool locked
, need_to_kill
;
1555 owait
.wait
.flags
= 0;
1556 owait
.wait
.func
= memcg_oom_wake_function
;
1557 owait
.wait
.private = current
;
1558 INIT_LIST_HEAD(&owait
.wait
.task_list
);
1559 need_to_kill
= true;
1560 /* At first, try to OOM lock hierarchy under mem.*/
1561 mutex_lock(&memcg_oom_mutex
);
1562 locked
= mem_cgroup_oom_lock(mem
);
1564 * Even if signal_pending(), we can't quit charge() loop without
1565 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1566 * under OOM is always welcomed, use TASK_KILLABLE here.
1568 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1569 if (!locked
|| mem
->oom_kill_disable
)
1570 need_to_kill
= false;
1572 mem_cgroup_oom_notify(mem
);
1573 mutex_unlock(&memcg_oom_mutex
);
1576 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1577 mem_cgroup_out_of_memory(mem
, mask
);
1580 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1582 mutex_lock(&memcg_oom_mutex
);
1583 mem_cgroup_oom_unlock(mem
);
1584 memcg_wakeup_oom(mem
);
1585 mutex_unlock(&memcg_oom_mutex
);
1587 if (test_thread_flag(TIF_MEMDIE
) || fatal_signal_pending(current
))
1589 /* Give chance to dying process */
1590 schedule_timeout(1);
1595 * Currently used to update mapped file statistics, but the routine can be
1596 * generalized to update other statistics as well.
1598 * Notes: Race condition
1600 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1601 * it tends to be costly. But considering some conditions, we doesn't need
1602 * to do so _always_.
1604 * Considering "charge", lock_page_cgroup() is not required because all
1605 * file-stat operations happen after a page is attached to radix-tree. There
1606 * are no race with "charge".
1608 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1609 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1610 * if there are race with "uncharge". Statistics itself is properly handled
1613 * Considering "move", this is an only case we see a race. To make the race
1614 * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are
1615 * possibility of race condition. If there is, we take a lock.
1618 void mem_cgroup_update_page_stat(struct page
*page
,
1619 enum mem_cgroup_page_stat_item idx
, int val
)
1621 struct mem_cgroup
*mem
;
1622 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1623 bool need_unlock
= false;
1624 unsigned long uninitialized_var(flags
);
1630 mem
= pc
->mem_cgroup
;
1631 if (unlikely(!mem
|| !PageCgroupUsed(pc
)))
1633 /* pc->mem_cgroup is unstable ? */
1634 if (unlikely(mem_cgroup_stealed(mem
)) || PageTransHuge(page
)) {
1635 /* take a lock against to access pc->mem_cgroup */
1636 move_lock_page_cgroup(pc
, &flags
);
1638 mem
= pc
->mem_cgroup
;
1639 if (!mem
|| !PageCgroupUsed(pc
))
1644 case MEMCG_NR_FILE_MAPPED
:
1646 SetPageCgroupFileMapped(pc
);
1647 else if (!page_mapped(page
))
1648 ClearPageCgroupFileMapped(pc
);
1649 idx
= MEM_CGROUP_STAT_FILE_MAPPED
;
1655 this_cpu_add(mem
->stat
->count
[idx
], val
);
1658 if (unlikely(need_unlock
))
1659 move_unlock_page_cgroup(pc
, &flags
);
1663 EXPORT_SYMBOL(mem_cgroup_update_page_stat
);
1666 * size of first charge trial. "32" comes from vmscan.c's magic value.
1667 * TODO: maybe necessary to use big numbers in big irons.
1669 #define CHARGE_BATCH 32U
1670 struct memcg_stock_pcp
{
1671 struct mem_cgroup
*cached
; /* this never be root cgroup */
1672 unsigned int nr_pages
;
1673 struct work_struct work
;
1675 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
1676 static atomic_t memcg_drain_count
;
1679 * Try to consume stocked charge on this cpu. If success, one page is consumed
1680 * from local stock and true is returned. If the stock is 0 or charges from a
1681 * cgroup which is not current target, returns false. This stock will be
1684 static bool consume_stock(struct mem_cgroup
*mem
)
1686 struct memcg_stock_pcp
*stock
;
1689 stock
= &get_cpu_var(memcg_stock
);
1690 if (mem
== stock
->cached
&& stock
->nr_pages
)
1692 else /* need to call res_counter_charge */
1694 put_cpu_var(memcg_stock
);
1699 * Returns stocks cached in percpu to res_counter and reset cached information.
1701 static void drain_stock(struct memcg_stock_pcp
*stock
)
1703 struct mem_cgroup
*old
= stock
->cached
;
1705 if (stock
->nr_pages
) {
1706 unsigned long bytes
= stock
->nr_pages
* PAGE_SIZE
;
1708 res_counter_uncharge(&old
->res
, bytes
);
1709 if (do_swap_account
)
1710 res_counter_uncharge(&old
->memsw
, bytes
);
1711 stock
->nr_pages
= 0;
1713 stock
->cached
= NULL
;
1717 * This must be called under preempt disabled or must be called by
1718 * a thread which is pinned to local cpu.
1720 static void drain_local_stock(struct work_struct
*dummy
)
1722 struct memcg_stock_pcp
*stock
= &__get_cpu_var(memcg_stock
);
1727 * Cache charges(val) which is from res_counter, to local per_cpu area.
1728 * This will be consumed by consume_stock() function, later.
1730 static void refill_stock(struct mem_cgroup
*mem
, unsigned int nr_pages
)
1732 struct memcg_stock_pcp
*stock
= &get_cpu_var(memcg_stock
);
1734 if (stock
->cached
!= mem
) { /* reset if necessary */
1736 stock
->cached
= mem
;
1738 stock
->nr_pages
+= nr_pages
;
1739 put_cpu_var(memcg_stock
);
1743 * Tries to drain stocked charges in other cpus. This function is asynchronous
1744 * and just put a work per cpu for draining localy on each cpu. Caller can
1745 * expects some charges will be back to res_counter later but cannot wait for
1748 static void drain_all_stock_async(void)
1751 /* This function is for scheduling "drain" in asynchronous way.
1752 * The result of "drain" is not directly handled by callers. Then,
1753 * if someone is calling drain, we don't have to call drain more.
1754 * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if
1755 * there is a race. We just do loose check here.
1757 if (atomic_read(&memcg_drain_count
))
1759 /* Notify other cpus that system-wide "drain" is running */
1760 atomic_inc(&memcg_drain_count
);
1762 for_each_online_cpu(cpu
) {
1763 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
1764 schedule_work_on(cpu
, &stock
->work
);
1767 atomic_dec(&memcg_drain_count
);
1768 /* We don't wait for flush_work */
1771 /* This is a synchronous drain interface. */
1772 static void drain_all_stock_sync(void)
1774 /* called when force_empty is called */
1775 atomic_inc(&memcg_drain_count
);
1776 schedule_on_each_cpu(drain_local_stock
);
1777 atomic_dec(&memcg_drain_count
);
1781 * This function drains percpu counter value from DEAD cpu and
1782 * move it to local cpu. Note that this function can be preempted.
1784 static void mem_cgroup_drain_pcp_counter(struct mem_cgroup
*mem
, int cpu
)
1788 spin_lock(&mem
->pcp_counter_lock
);
1789 for (i
= 0; i
< MEM_CGROUP_STAT_DATA
; i
++) {
1790 s64 x
= per_cpu(mem
->stat
->count
[i
], cpu
);
1792 per_cpu(mem
->stat
->count
[i
], cpu
) = 0;
1793 mem
->nocpu_base
.count
[i
] += x
;
1795 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
1796 unsigned long x
= per_cpu(mem
->stat
->events
[i
], cpu
);
1798 per_cpu(mem
->stat
->events
[i
], cpu
) = 0;
1799 mem
->nocpu_base
.events
[i
] += x
;
1801 /* need to clear ON_MOVE value, works as a kind of lock. */
1802 per_cpu(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) = 0;
1803 spin_unlock(&mem
->pcp_counter_lock
);
1806 static void synchronize_mem_cgroup_on_move(struct mem_cgroup
*mem
, int cpu
)
1808 int idx
= MEM_CGROUP_ON_MOVE
;
1810 spin_lock(&mem
->pcp_counter_lock
);
1811 per_cpu(mem
->stat
->count
[idx
], cpu
) = mem
->nocpu_base
.count
[idx
];
1812 spin_unlock(&mem
->pcp_counter_lock
);
1815 static int __cpuinit
memcg_cpu_hotplug_callback(struct notifier_block
*nb
,
1816 unsigned long action
,
1819 int cpu
= (unsigned long)hcpu
;
1820 struct memcg_stock_pcp
*stock
;
1821 struct mem_cgroup
*iter
;
1823 if ((action
== CPU_ONLINE
)) {
1824 for_each_mem_cgroup_all(iter
)
1825 synchronize_mem_cgroup_on_move(iter
, cpu
);
1829 if ((action
!= CPU_DEAD
) || action
!= CPU_DEAD_FROZEN
)
1832 for_each_mem_cgroup_all(iter
)
1833 mem_cgroup_drain_pcp_counter(iter
, cpu
);
1835 stock
= &per_cpu(memcg_stock
, cpu
);
1841 /* See __mem_cgroup_try_charge() for details */
1843 CHARGE_OK
, /* success */
1844 CHARGE_RETRY
, /* need to retry but retry is not bad */
1845 CHARGE_NOMEM
, /* we can't do more. return -ENOMEM */
1846 CHARGE_WOULDBLOCK
, /* GFP_WAIT wasn't set and no enough res. */
1847 CHARGE_OOM_DIE
, /* the current is killed because of OOM */
1850 static int mem_cgroup_do_charge(struct mem_cgroup
*mem
, gfp_t gfp_mask
,
1851 unsigned int nr_pages
, bool oom_check
)
1853 unsigned long csize
= nr_pages
* PAGE_SIZE
;
1854 struct mem_cgroup
*mem_over_limit
;
1855 struct res_counter
*fail_res
;
1856 unsigned long flags
= 0;
1859 ret
= res_counter_charge(&mem
->res
, csize
, &fail_res
);
1862 if (!do_swap_account
)
1864 ret
= res_counter_charge(&mem
->memsw
, csize
, &fail_res
);
1868 res_counter_uncharge(&mem
->res
, csize
);
1869 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, memsw
);
1870 flags
|= MEM_CGROUP_RECLAIM_NOSWAP
;
1872 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, res
);
1874 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
1875 * of regular pages (CHARGE_BATCH), or a single regular page (1).
1877 * Never reclaim on behalf of optional batching, retry with a
1878 * single page instead.
1880 if (nr_pages
== CHARGE_BATCH
)
1881 return CHARGE_RETRY
;
1883 if (!(gfp_mask
& __GFP_WAIT
))
1884 return CHARGE_WOULDBLOCK
;
1886 ret
= mem_cgroup_hierarchical_reclaim(mem_over_limit
, NULL
,
1888 if (mem_cgroup_margin(mem_over_limit
) >= nr_pages
)
1889 return CHARGE_RETRY
;
1891 * Even though the limit is exceeded at this point, reclaim
1892 * may have been able to free some pages. Retry the charge
1893 * before killing the task.
1895 * Only for regular pages, though: huge pages are rather
1896 * unlikely to succeed so close to the limit, and we fall back
1897 * to regular pages anyway in case of failure.
1899 if (nr_pages
== 1 && ret
)
1900 return CHARGE_RETRY
;
1903 * At task move, charge accounts can be doubly counted. So, it's
1904 * better to wait until the end of task_move if something is going on.
1906 if (mem_cgroup_wait_acct_move(mem_over_limit
))
1907 return CHARGE_RETRY
;
1909 /* If we don't need to call oom-killer at el, return immediately */
1911 return CHARGE_NOMEM
;
1913 if (!mem_cgroup_handle_oom(mem_over_limit
, gfp_mask
))
1914 return CHARGE_OOM_DIE
;
1916 return CHARGE_RETRY
;
1920 * Unlike exported interface, "oom" parameter is added. if oom==true,
1921 * oom-killer can be invoked.
1923 static int __mem_cgroup_try_charge(struct mm_struct
*mm
,
1925 unsigned int nr_pages
,
1926 struct mem_cgroup
**memcg
,
1929 unsigned int batch
= max(CHARGE_BATCH
, nr_pages
);
1930 int nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
1931 struct mem_cgroup
*mem
= NULL
;
1935 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
1936 * in system level. So, allow to go ahead dying process in addition to
1939 if (unlikely(test_thread_flag(TIF_MEMDIE
)
1940 || fatal_signal_pending(current
)))
1944 * We always charge the cgroup the mm_struct belongs to.
1945 * The mm_struct's mem_cgroup changes on task migration if the
1946 * thread group leader migrates. It's possible that mm is not
1947 * set, if so charge the init_mm (happens for pagecache usage).
1952 if (*memcg
) { /* css should be a valid one */
1954 VM_BUG_ON(css_is_removed(&mem
->css
));
1955 if (mem_cgroup_is_root(mem
))
1957 if (nr_pages
== 1 && consume_stock(mem
))
1961 struct task_struct
*p
;
1964 p
= rcu_dereference(mm
->owner
);
1966 * Because we don't have task_lock(), "p" can exit.
1967 * In that case, "mem" can point to root or p can be NULL with
1968 * race with swapoff. Then, we have small risk of mis-accouning.
1969 * But such kind of mis-account by race always happens because
1970 * we don't have cgroup_mutex(). It's overkill and we allo that
1972 * (*) swapoff at el will charge against mm-struct not against
1973 * task-struct. So, mm->owner can be NULL.
1975 mem
= mem_cgroup_from_task(p
);
1976 if (!mem
|| mem_cgroup_is_root(mem
)) {
1980 if (nr_pages
== 1 && consume_stock(mem
)) {
1982 * It seems dagerous to access memcg without css_get().
1983 * But considering how consume_stok works, it's not
1984 * necessary. If consume_stock success, some charges
1985 * from this memcg are cached on this cpu. So, we
1986 * don't need to call css_get()/css_tryget() before
1987 * calling consume_stock().
1992 /* after here, we may be blocked. we need to get refcnt */
1993 if (!css_tryget(&mem
->css
)) {
2003 /* If killed, bypass charge */
2004 if (fatal_signal_pending(current
)) {
2010 if (oom
&& !nr_oom_retries
) {
2012 nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2015 ret
= mem_cgroup_do_charge(mem
, gfp_mask
, batch
, oom_check
);
2019 case CHARGE_RETRY
: /* not in OOM situation but retry */
2024 case CHARGE_WOULDBLOCK
: /* !__GFP_WAIT */
2027 case CHARGE_NOMEM
: /* OOM routine works */
2032 /* If oom, we never return -ENOMEM */
2035 case CHARGE_OOM_DIE
: /* Killed by OOM Killer */
2039 } while (ret
!= CHARGE_OK
);
2041 if (batch
> nr_pages
)
2042 refill_stock(mem
, batch
- nr_pages
);
2056 * Somemtimes we have to undo a charge we got by try_charge().
2057 * This function is for that and do uncharge, put css's refcnt.
2058 * gotten by try_charge().
2060 static void __mem_cgroup_cancel_charge(struct mem_cgroup
*mem
,
2061 unsigned int nr_pages
)
2063 if (!mem_cgroup_is_root(mem
)) {
2064 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2066 res_counter_uncharge(&mem
->res
, bytes
);
2067 if (do_swap_account
)
2068 res_counter_uncharge(&mem
->memsw
, bytes
);
2073 * A helper function to get mem_cgroup from ID. must be called under
2074 * rcu_read_lock(). The caller must check css_is_removed() or some if
2075 * it's concern. (dropping refcnt from swap can be called against removed
2078 static struct mem_cgroup
*mem_cgroup_lookup(unsigned short id
)
2080 struct cgroup_subsys_state
*css
;
2082 /* ID 0 is unused ID */
2085 css
= css_lookup(&mem_cgroup_subsys
, id
);
2088 return container_of(css
, struct mem_cgroup
, css
);
2091 struct mem_cgroup
*try_get_mem_cgroup_from_page(struct page
*page
)
2093 struct mem_cgroup
*mem
= NULL
;
2094 struct page_cgroup
*pc
;
2098 VM_BUG_ON(!PageLocked(page
));
2100 pc
= lookup_page_cgroup(page
);
2101 lock_page_cgroup(pc
);
2102 if (PageCgroupUsed(pc
)) {
2103 mem
= pc
->mem_cgroup
;
2104 if (mem
&& !css_tryget(&mem
->css
))
2106 } else if (PageSwapCache(page
)) {
2107 ent
.val
= page_private(page
);
2108 id
= lookup_swap_cgroup(ent
);
2110 mem
= mem_cgroup_lookup(id
);
2111 if (mem
&& !css_tryget(&mem
->css
))
2115 unlock_page_cgroup(pc
);
2119 static void __mem_cgroup_commit_charge(struct mem_cgroup
*mem
,
2121 unsigned int nr_pages
,
2122 struct page_cgroup
*pc
,
2123 enum charge_type ctype
)
2125 lock_page_cgroup(pc
);
2126 if (unlikely(PageCgroupUsed(pc
))) {
2127 unlock_page_cgroup(pc
);
2128 __mem_cgroup_cancel_charge(mem
, nr_pages
);
2132 * we don't need page_cgroup_lock about tail pages, becase they are not
2133 * accessed by any other context at this point.
2135 pc
->mem_cgroup
= mem
;
2137 * We access a page_cgroup asynchronously without lock_page_cgroup().
2138 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2139 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2140 * before USED bit, we need memory barrier here.
2141 * See mem_cgroup_add_lru_list(), etc.
2145 case MEM_CGROUP_CHARGE_TYPE_CACHE
:
2146 case MEM_CGROUP_CHARGE_TYPE_SHMEM
:
2147 SetPageCgroupCache(pc
);
2148 SetPageCgroupUsed(pc
);
2150 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2151 ClearPageCgroupCache(pc
);
2152 SetPageCgroupUsed(pc
);
2158 mem_cgroup_charge_statistics(mem
, PageCgroupCache(pc
), nr_pages
);
2159 unlock_page_cgroup(pc
);
2161 * "charge_statistics" updated event counter. Then, check it.
2162 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2163 * if they exceeds softlimit.
2165 memcg_check_events(mem
, page
);
2168 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2170 #define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\
2171 (1 << PCG_ACCT_LRU) | (1 << PCG_MIGRATION))
2173 * Because tail pages are not marked as "used", set it. We're under
2174 * zone->lru_lock, 'splitting on pmd' and compund_lock.
2176 void mem_cgroup_split_huge_fixup(struct page
*head
, struct page
*tail
)
2178 struct page_cgroup
*head_pc
= lookup_page_cgroup(head
);
2179 struct page_cgroup
*tail_pc
= lookup_page_cgroup(tail
);
2180 unsigned long flags
;
2182 if (mem_cgroup_disabled())
2185 * We have no races with charge/uncharge but will have races with
2186 * page state accounting.
2188 move_lock_page_cgroup(head_pc
, &flags
);
2190 tail_pc
->mem_cgroup
= head_pc
->mem_cgroup
;
2191 smp_wmb(); /* see __commit_charge() */
2192 if (PageCgroupAcctLRU(head_pc
)) {
2194 struct mem_cgroup_per_zone
*mz
;
2197 * LRU flags cannot be copied because we need to add tail
2198 *.page to LRU by generic call and our hook will be called.
2199 * We hold lru_lock, then, reduce counter directly.
2201 lru
= page_lru(head
);
2202 mz
= page_cgroup_zoneinfo(head_pc
->mem_cgroup
, head
);
2203 MEM_CGROUP_ZSTAT(mz
, lru
) -= 1;
2205 tail_pc
->flags
= head_pc
->flags
& ~PCGF_NOCOPY_AT_SPLIT
;
2206 move_unlock_page_cgroup(head_pc
, &flags
);
2211 * mem_cgroup_move_account - move account of the page
2213 * @nr_pages: number of regular pages (>1 for huge pages)
2214 * @pc: page_cgroup of the page.
2215 * @from: mem_cgroup which the page is moved from.
2216 * @to: mem_cgroup which the page is moved to. @from != @to.
2217 * @uncharge: whether we should call uncharge and css_put against @from.
2219 * The caller must confirm following.
2220 * - page is not on LRU (isolate_page() is useful.)
2221 * - compound_lock is held when nr_pages > 1
2223 * This function doesn't do "charge" nor css_get to new cgroup. It should be
2224 * done by a caller(__mem_cgroup_try_charge would be usefull). If @uncharge is
2225 * true, this function does "uncharge" from old cgroup, but it doesn't if
2226 * @uncharge is false, so a caller should do "uncharge".
2228 static int mem_cgroup_move_account(struct page
*page
,
2229 unsigned int nr_pages
,
2230 struct page_cgroup
*pc
,
2231 struct mem_cgroup
*from
,
2232 struct mem_cgroup
*to
,
2235 unsigned long flags
;
2238 VM_BUG_ON(from
== to
);
2239 VM_BUG_ON(PageLRU(page
));
2241 * The page is isolated from LRU. So, collapse function
2242 * will not handle this page. But page splitting can happen.
2243 * Do this check under compound_page_lock(). The caller should
2247 if (nr_pages
> 1 && !PageTransHuge(page
))
2250 lock_page_cgroup(pc
);
2253 if (!PageCgroupUsed(pc
) || pc
->mem_cgroup
!= from
)
2256 move_lock_page_cgroup(pc
, &flags
);
2258 if (PageCgroupFileMapped(pc
)) {
2259 /* Update mapped_file data for mem_cgroup */
2261 __this_cpu_dec(from
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2262 __this_cpu_inc(to
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2265 mem_cgroup_charge_statistics(from
, PageCgroupCache(pc
), -nr_pages
);
2267 /* This is not "cancel", but cancel_charge does all we need. */
2268 __mem_cgroup_cancel_charge(from
, nr_pages
);
2270 /* caller should have done css_get */
2271 pc
->mem_cgroup
= to
;
2272 mem_cgroup_charge_statistics(to
, PageCgroupCache(pc
), nr_pages
);
2274 * We charges against "to" which may not have any tasks. Then, "to"
2275 * can be under rmdir(). But in current implementation, caller of
2276 * this function is just force_empty() and move charge, so it's
2277 * garanteed that "to" is never removed. So, we don't check rmdir
2280 move_unlock_page_cgroup(pc
, &flags
);
2283 unlock_page_cgroup(pc
);
2287 memcg_check_events(to
, page
);
2288 memcg_check_events(from
, page
);
2294 * move charges to its parent.
2297 static int mem_cgroup_move_parent(struct page
*page
,
2298 struct page_cgroup
*pc
,
2299 struct mem_cgroup
*child
,
2302 struct cgroup
*cg
= child
->css
.cgroup
;
2303 struct cgroup
*pcg
= cg
->parent
;
2304 struct mem_cgroup
*parent
;
2305 unsigned int nr_pages
;
2306 unsigned long flags
;
2314 if (!get_page_unless_zero(page
))
2316 if (isolate_lru_page(page
))
2319 nr_pages
= hpage_nr_pages(page
);
2321 parent
= mem_cgroup_from_cont(pcg
);
2322 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, nr_pages
, &parent
, false);
2327 flags
= compound_lock_irqsave(page
);
2329 ret
= mem_cgroup_move_account(page
, nr_pages
, pc
, child
, parent
, true);
2331 __mem_cgroup_cancel_charge(parent
, nr_pages
);
2334 compound_unlock_irqrestore(page
, flags
);
2336 putback_lru_page(page
);
2344 * Charge the memory controller for page usage.
2346 * 0 if the charge was successful
2347 * < 0 if the cgroup is over its limit
2349 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
2350 gfp_t gfp_mask
, enum charge_type ctype
)
2352 struct mem_cgroup
*mem
= NULL
;
2353 unsigned int nr_pages
= 1;
2354 struct page_cgroup
*pc
;
2358 if (PageTransHuge(page
)) {
2359 nr_pages
<<= compound_order(page
);
2360 VM_BUG_ON(!PageTransHuge(page
));
2362 * Never OOM-kill a process for a huge page. The
2363 * fault handler will fall back to regular pages.
2368 pc
= lookup_page_cgroup(page
);
2369 BUG_ON(!pc
); /* XXX: remove this and move pc lookup into commit */
2371 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, nr_pages
, &mem
, oom
);
2375 __mem_cgroup_commit_charge(mem
, page
, nr_pages
, pc
, ctype
);
2379 int mem_cgroup_newpage_charge(struct page
*page
,
2380 struct mm_struct
*mm
, gfp_t gfp_mask
)
2382 if (mem_cgroup_disabled())
2385 * If already mapped, we don't have to account.
2386 * If page cache, page->mapping has address_space.
2387 * But page->mapping may have out-of-use anon_vma pointer,
2388 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
2391 if (page_mapped(page
) || (page
->mapping
&& !PageAnon(page
)))
2395 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2396 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2400 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2401 enum charge_type ctype
);
2403 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
2408 if (mem_cgroup_disabled())
2410 if (PageCompound(page
))
2413 * Corner case handling. This is called from add_to_page_cache()
2414 * in usual. But some FS (shmem) precharges this page before calling it
2415 * and call add_to_page_cache() with GFP_NOWAIT.
2417 * For GFP_NOWAIT case, the page may be pre-charged before calling
2418 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
2419 * charge twice. (It works but has to pay a bit larger cost.)
2420 * And when the page is SwapCache, it should take swap information
2421 * into account. This is under lock_page() now.
2423 if (!(gfp_mask
& __GFP_WAIT
)) {
2424 struct page_cgroup
*pc
;
2426 pc
= lookup_page_cgroup(page
);
2429 lock_page_cgroup(pc
);
2430 if (PageCgroupUsed(pc
)) {
2431 unlock_page_cgroup(pc
);
2434 unlock_page_cgroup(pc
);
2440 if (page_is_file_cache(page
))
2441 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2442 MEM_CGROUP_CHARGE_TYPE_CACHE
);
2445 if (PageSwapCache(page
)) {
2446 struct mem_cgroup
*mem
;
2448 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &mem
);
2450 __mem_cgroup_commit_charge_swapin(page
, mem
,
2451 MEM_CGROUP_CHARGE_TYPE_SHMEM
);
2453 ret
= mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2454 MEM_CGROUP_CHARGE_TYPE_SHMEM
);
2460 * While swap-in, try_charge -> commit or cancel, the page is locked.
2461 * And when try_charge() successfully returns, one refcnt to memcg without
2462 * struct page_cgroup is acquired. This refcnt will be consumed by
2463 * "commit()" or removed by "cancel()"
2465 int mem_cgroup_try_charge_swapin(struct mm_struct
*mm
,
2467 gfp_t mask
, struct mem_cgroup
**ptr
)
2469 struct mem_cgroup
*mem
;
2474 if (mem_cgroup_disabled())
2477 if (!do_swap_account
)
2480 * A racing thread's fault, or swapoff, may have already updated
2481 * the pte, and even removed page from swap cache: in those cases
2482 * do_swap_page()'s pte_same() test will fail; but there's also a
2483 * KSM case which does need to charge the page.
2485 if (!PageSwapCache(page
))
2487 mem
= try_get_mem_cgroup_from_page(page
);
2491 ret
= __mem_cgroup_try_charge(NULL
, mask
, 1, ptr
, true);
2497 return __mem_cgroup_try_charge(mm
, mask
, 1, ptr
, true);
2501 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2502 enum charge_type ctype
)
2504 struct page_cgroup
*pc
;
2506 if (mem_cgroup_disabled())
2510 cgroup_exclude_rmdir(&ptr
->css
);
2511 pc
= lookup_page_cgroup(page
);
2512 mem_cgroup_lru_del_before_commit_swapcache(page
);
2513 __mem_cgroup_commit_charge(ptr
, page
, 1, pc
, ctype
);
2514 mem_cgroup_lru_add_after_commit_swapcache(page
);
2516 * Now swap is on-memory. This means this page may be
2517 * counted both as mem and swap....double count.
2518 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2519 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2520 * may call delete_from_swap_cache() before reach here.
2522 if (do_swap_account
&& PageSwapCache(page
)) {
2523 swp_entry_t ent
= {.val
= page_private(page
)};
2525 struct mem_cgroup
*memcg
;
2527 id
= swap_cgroup_record(ent
, 0);
2529 memcg
= mem_cgroup_lookup(id
);
2532 * This recorded memcg can be obsolete one. So, avoid
2533 * calling css_tryget
2535 if (!mem_cgroup_is_root(memcg
))
2536 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
2537 mem_cgroup_swap_statistics(memcg
, false);
2538 mem_cgroup_put(memcg
);
2543 * At swapin, we may charge account against cgroup which has no tasks.
2544 * So, rmdir()->pre_destroy() can be called while we do this charge.
2545 * In that case, we need to call pre_destroy() again. check it here.
2547 cgroup_release_and_wakeup_rmdir(&ptr
->css
);
2550 void mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
)
2552 __mem_cgroup_commit_charge_swapin(page
, ptr
,
2553 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2556 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup
*mem
)
2558 if (mem_cgroup_disabled())
2562 __mem_cgroup_cancel_charge(mem
, 1);
2565 static void mem_cgroup_do_uncharge(struct mem_cgroup
*mem
,
2566 unsigned int nr_pages
,
2567 const enum charge_type ctype
)
2569 struct memcg_batch_info
*batch
= NULL
;
2570 bool uncharge_memsw
= true;
2572 /* If swapout, usage of swap doesn't decrease */
2573 if (!do_swap_account
|| ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
)
2574 uncharge_memsw
= false;
2576 batch
= ¤t
->memcg_batch
;
2578 * In usual, we do css_get() when we remember memcg pointer.
2579 * But in this case, we keep res->usage until end of a series of
2580 * uncharges. Then, it's ok to ignore memcg's refcnt.
2585 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2586 * In those cases, all pages freed continously can be expected to be in
2587 * the same cgroup and we have chance to coalesce uncharges.
2588 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2589 * because we want to do uncharge as soon as possible.
2592 if (!batch
->do_batch
|| test_thread_flag(TIF_MEMDIE
))
2593 goto direct_uncharge
;
2596 goto direct_uncharge
;
2599 * In typical case, batch->memcg == mem. This means we can
2600 * merge a series of uncharges to an uncharge of res_counter.
2601 * If not, we uncharge res_counter ony by one.
2603 if (batch
->memcg
!= mem
)
2604 goto direct_uncharge
;
2605 /* remember freed charge and uncharge it later */
2608 batch
->memsw_nr_pages
++;
2611 res_counter_uncharge(&mem
->res
, nr_pages
* PAGE_SIZE
);
2613 res_counter_uncharge(&mem
->memsw
, nr_pages
* PAGE_SIZE
);
2614 if (unlikely(batch
->memcg
!= mem
))
2615 memcg_oom_recover(mem
);
2620 * uncharge if !page_mapped(page)
2622 static struct mem_cgroup
*
2623 __mem_cgroup_uncharge_common(struct page
*page
, enum charge_type ctype
)
2625 struct mem_cgroup
*mem
= NULL
;
2626 unsigned int nr_pages
= 1;
2627 struct page_cgroup
*pc
;
2629 if (mem_cgroup_disabled())
2632 if (PageSwapCache(page
))
2635 if (PageTransHuge(page
)) {
2636 nr_pages
<<= compound_order(page
);
2637 VM_BUG_ON(!PageTransHuge(page
));
2640 * Check if our page_cgroup is valid
2642 pc
= lookup_page_cgroup(page
);
2643 if (unlikely(!pc
|| !PageCgroupUsed(pc
)))
2646 lock_page_cgroup(pc
);
2648 mem
= pc
->mem_cgroup
;
2650 if (!PageCgroupUsed(pc
))
2654 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2655 case MEM_CGROUP_CHARGE_TYPE_DROP
:
2656 /* See mem_cgroup_prepare_migration() */
2657 if (page_mapped(page
) || PageCgroupMigration(pc
))
2660 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT
:
2661 if (!PageAnon(page
)) { /* Shared memory */
2662 if (page
->mapping
&& !page_is_file_cache(page
))
2664 } else if (page_mapped(page
)) /* Anon */
2671 mem_cgroup_charge_statistics(mem
, PageCgroupCache(pc
), -nr_pages
);
2673 ClearPageCgroupUsed(pc
);
2675 * pc->mem_cgroup is not cleared here. It will be accessed when it's
2676 * freed from LRU. This is safe because uncharged page is expected not
2677 * to be reused (freed soon). Exception is SwapCache, it's handled by
2678 * special functions.
2681 unlock_page_cgroup(pc
);
2683 * even after unlock, we have mem->res.usage here and this memcg
2684 * will never be freed.
2686 memcg_check_events(mem
, page
);
2687 if (do_swap_account
&& ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
) {
2688 mem_cgroup_swap_statistics(mem
, true);
2689 mem_cgroup_get(mem
);
2691 if (!mem_cgroup_is_root(mem
))
2692 mem_cgroup_do_uncharge(mem
, nr_pages
, ctype
);
2697 unlock_page_cgroup(pc
);
2701 void mem_cgroup_uncharge_page(struct page
*page
)
2704 if (page_mapped(page
))
2706 if (page
->mapping
&& !PageAnon(page
))
2708 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2711 void mem_cgroup_uncharge_cache_page(struct page
*page
)
2713 VM_BUG_ON(page_mapped(page
));
2714 VM_BUG_ON(page
->mapping
);
2715 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_CACHE
);
2719 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
2720 * In that cases, pages are freed continuously and we can expect pages
2721 * are in the same memcg. All these calls itself limits the number of
2722 * pages freed at once, then uncharge_start/end() is called properly.
2723 * This may be called prural(2) times in a context,
2726 void mem_cgroup_uncharge_start(void)
2728 current
->memcg_batch
.do_batch
++;
2729 /* We can do nest. */
2730 if (current
->memcg_batch
.do_batch
== 1) {
2731 current
->memcg_batch
.memcg
= NULL
;
2732 current
->memcg_batch
.nr_pages
= 0;
2733 current
->memcg_batch
.memsw_nr_pages
= 0;
2737 void mem_cgroup_uncharge_end(void)
2739 struct memcg_batch_info
*batch
= ¤t
->memcg_batch
;
2741 if (!batch
->do_batch
)
2745 if (batch
->do_batch
) /* If stacked, do nothing. */
2751 * This "batch->memcg" is valid without any css_get/put etc...
2752 * bacause we hide charges behind us.
2754 if (batch
->nr_pages
)
2755 res_counter_uncharge(&batch
->memcg
->res
,
2756 batch
->nr_pages
* PAGE_SIZE
);
2757 if (batch
->memsw_nr_pages
)
2758 res_counter_uncharge(&batch
->memcg
->memsw
,
2759 batch
->memsw_nr_pages
* PAGE_SIZE
);
2760 memcg_oom_recover(batch
->memcg
);
2761 /* forget this pointer (for sanity check) */
2762 batch
->memcg
= NULL
;
2767 * called after __delete_from_swap_cache() and drop "page" account.
2768 * memcg information is recorded to swap_cgroup of "ent"
2771 mem_cgroup_uncharge_swapcache(struct page
*page
, swp_entry_t ent
, bool swapout
)
2773 struct mem_cgroup
*memcg
;
2774 int ctype
= MEM_CGROUP_CHARGE_TYPE_SWAPOUT
;
2776 if (!swapout
) /* this was a swap cache but the swap is unused ! */
2777 ctype
= MEM_CGROUP_CHARGE_TYPE_DROP
;
2779 memcg
= __mem_cgroup_uncharge_common(page
, ctype
);
2782 * record memcg information, if swapout && memcg != NULL,
2783 * mem_cgroup_get() was called in uncharge().
2785 if (do_swap_account
&& swapout
&& memcg
)
2786 swap_cgroup_record(ent
, css_id(&memcg
->css
));
2790 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2792 * called from swap_entry_free(). remove record in swap_cgroup and
2793 * uncharge "memsw" account.
2795 void mem_cgroup_uncharge_swap(swp_entry_t ent
)
2797 struct mem_cgroup
*memcg
;
2800 if (!do_swap_account
)
2803 id
= swap_cgroup_record(ent
, 0);
2805 memcg
= mem_cgroup_lookup(id
);
2808 * We uncharge this because swap is freed.
2809 * This memcg can be obsolete one. We avoid calling css_tryget
2811 if (!mem_cgroup_is_root(memcg
))
2812 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
2813 mem_cgroup_swap_statistics(memcg
, false);
2814 mem_cgroup_put(memcg
);
2820 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2821 * @entry: swap entry to be moved
2822 * @from: mem_cgroup which the entry is moved from
2823 * @to: mem_cgroup which the entry is moved to
2824 * @need_fixup: whether we should fixup res_counters and refcounts.
2826 * It succeeds only when the swap_cgroup's record for this entry is the same
2827 * as the mem_cgroup's id of @from.
2829 * Returns 0 on success, -EINVAL on failure.
2831 * The caller must have charged to @to, IOW, called res_counter_charge() about
2832 * both res and memsw, and called css_get().
2834 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
2835 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
2837 unsigned short old_id
, new_id
;
2839 old_id
= css_id(&from
->css
);
2840 new_id
= css_id(&to
->css
);
2842 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
2843 mem_cgroup_swap_statistics(from
, false);
2844 mem_cgroup_swap_statistics(to
, true);
2846 * This function is only called from task migration context now.
2847 * It postpones res_counter and refcount handling till the end
2848 * of task migration(mem_cgroup_clear_mc()) for performance
2849 * improvement. But we cannot postpone mem_cgroup_get(to)
2850 * because if the process that has been moved to @to does
2851 * swap-in, the refcount of @to might be decreased to 0.
2855 if (!mem_cgroup_is_root(from
))
2856 res_counter_uncharge(&from
->memsw
, PAGE_SIZE
);
2857 mem_cgroup_put(from
);
2859 * we charged both to->res and to->memsw, so we should
2862 if (!mem_cgroup_is_root(to
))
2863 res_counter_uncharge(&to
->res
, PAGE_SIZE
);
2870 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
2871 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
2878 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
2881 int mem_cgroup_prepare_migration(struct page
*page
,
2882 struct page
*newpage
, struct mem_cgroup
**ptr
, gfp_t gfp_mask
)
2884 struct mem_cgroup
*mem
= NULL
;
2885 struct page_cgroup
*pc
;
2886 enum charge_type ctype
;
2891 VM_BUG_ON(PageTransHuge(page
));
2892 if (mem_cgroup_disabled())
2895 pc
= lookup_page_cgroup(page
);
2896 lock_page_cgroup(pc
);
2897 if (PageCgroupUsed(pc
)) {
2898 mem
= pc
->mem_cgroup
;
2901 * At migrating an anonymous page, its mapcount goes down
2902 * to 0 and uncharge() will be called. But, even if it's fully
2903 * unmapped, migration may fail and this page has to be
2904 * charged again. We set MIGRATION flag here and delay uncharge
2905 * until end_migration() is called
2907 * Corner Case Thinking
2909 * When the old page was mapped as Anon and it's unmap-and-freed
2910 * while migration was ongoing.
2911 * If unmap finds the old page, uncharge() of it will be delayed
2912 * until end_migration(). If unmap finds a new page, it's
2913 * uncharged when it make mapcount to be 1->0. If unmap code
2914 * finds swap_migration_entry, the new page will not be mapped
2915 * and end_migration() will find it(mapcount==0).
2918 * When the old page was mapped but migraion fails, the kernel
2919 * remaps it. A charge for it is kept by MIGRATION flag even
2920 * if mapcount goes down to 0. We can do remap successfully
2921 * without charging it again.
2924 * The "old" page is under lock_page() until the end of
2925 * migration, so, the old page itself will not be swapped-out.
2926 * If the new page is swapped out before end_migraton, our
2927 * hook to usual swap-out path will catch the event.
2930 SetPageCgroupMigration(pc
);
2932 unlock_page_cgroup(pc
);
2934 * If the page is not charged at this point,
2941 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, 1, ptr
, false);
2942 css_put(&mem
->css
);/* drop extra refcnt */
2943 if (ret
|| *ptr
== NULL
) {
2944 if (PageAnon(page
)) {
2945 lock_page_cgroup(pc
);
2946 ClearPageCgroupMigration(pc
);
2947 unlock_page_cgroup(pc
);
2949 * The old page may be fully unmapped while we kept it.
2951 mem_cgroup_uncharge_page(page
);
2956 * We charge new page before it's used/mapped. So, even if unlock_page()
2957 * is called before end_migration, we can catch all events on this new
2958 * page. In the case new page is migrated but not remapped, new page's
2959 * mapcount will be finally 0 and we call uncharge in end_migration().
2961 pc
= lookup_page_cgroup(newpage
);
2963 ctype
= MEM_CGROUP_CHARGE_TYPE_MAPPED
;
2964 else if (page_is_file_cache(page
))
2965 ctype
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
2967 ctype
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
2968 __mem_cgroup_commit_charge(mem
, page
, 1, pc
, ctype
);
2972 /* remove redundant charge if migration failed*/
2973 void mem_cgroup_end_migration(struct mem_cgroup
*mem
,
2974 struct page
*oldpage
, struct page
*newpage
, bool migration_ok
)
2976 struct page
*used
, *unused
;
2977 struct page_cgroup
*pc
;
2981 /* blocks rmdir() */
2982 cgroup_exclude_rmdir(&mem
->css
);
2983 if (!migration_ok
) {
2991 * We disallowed uncharge of pages under migration because mapcount
2992 * of the page goes down to zero, temporarly.
2993 * Clear the flag and check the page should be charged.
2995 pc
= lookup_page_cgroup(oldpage
);
2996 lock_page_cgroup(pc
);
2997 ClearPageCgroupMigration(pc
);
2998 unlock_page_cgroup(pc
);
3000 __mem_cgroup_uncharge_common(unused
, MEM_CGROUP_CHARGE_TYPE_FORCE
);
3003 * If a page is a file cache, radix-tree replacement is very atomic
3004 * and we can skip this check. When it was an Anon page, its mapcount
3005 * goes down to 0. But because we added MIGRATION flage, it's not
3006 * uncharged yet. There are several case but page->mapcount check
3007 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3008 * check. (see prepare_charge() also)
3011 mem_cgroup_uncharge_page(used
);
3013 * At migration, we may charge account against cgroup which has no
3015 * So, rmdir()->pre_destroy() can be called while we do this charge.
3016 * In that case, we need to call pre_destroy() again. check it here.
3018 cgroup_release_and_wakeup_rmdir(&mem
->css
);
3022 * A call to try to shrink memory usage on charge failure at shmem's swapin.
3023 * Calling hierarchical_reclaim is not enough because we should update
3024 * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM.
3025 * Moreover considering hierarchy, we should reclaim from the mem_over_limit,
3026 * not from the memcg which this page would be charged to.
3027 * try_charge_swapin does all of these works properly.
3029 int mem_cgroup_shmem_charge_fallback(struct page
*page
,
3030 struct mm_struct
*mm
,
3033 struct mem_cgroup
*mem
;
3036 if (mem_cgroup_disabled())
3039 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &mem
);
3041 mem_cgroup_cancel_charge_swapin(mem
); /* it does !mem check */
3046 #ifdef CONFIG_DEBUG_VM
3047 static struct page_cgroup
*lookup_page_cgroup_used(struct page
*page
)
3049 struct page_cgroup
*pc
;
3051 pc
= lookup_page_cgroup(page
);
3052 if (likely(pc
) && PageCgroupUsed(pc
))
3057 bool mem_cgroup_bad_page_check(struct page
*page
)
3059 if (mem_cgroup_disabled())
3062 return lookup_page_cgroup_used(page
) != NULL
;
3065 void mem_cgroup_print_bad_page(struct page
*page
)
3067 struct page_cgroup
*pc
;
3069 pc
= lookup_page_cgroup_used(page
);
3074 printk(KERN_ALERT
"pc:%p pc->flags:%lx pc->mem_cgroup:%p",
3075 pc
, pc
->flags
, pc
->mem_cgroup
);
3077 path
= kmalloc(PATH_MAX
, GFP_KERNEL
);
3080 ret
= cgroup_path(pc
->mem_cgroup
->css
.cgroup
,
3085 printk(KERN_CONT
"(%s)\n",
3086 (ret
< 0) ? "cannot get the path" : path
);
3092 static DEFINE_MUTEX(set_limit_mutex
);
3094 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
3095 unsigned long long val
)
3098 u64 memswlimit
, memlimit
;
3100 int children
= mem_cgroup_count_children(memcg
);
3101 u64 curusage
, oldusage
;
3105 * For keeping hierarchical_reclaim simple, how long we should retry
3106 * is depends on callers. We set our retry-count to be function
3107 * of # of children which we should visit in this loop.
3109 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
* children
;
3111 oldusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3114 while (retry_count
) {
3115 if (signal_pending(current
)) {
3120 * Rather than hide all in some function, I do this in
3121 * open coded manner. You see what this really does.
3122 * We have to guarantee mem->res.limit < mem->memsw.limit.
3124 mutex_lock(&set_limit_mutex
);
3125 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3126 if (memswlimit
< val
) {
3128 mutex_unlock(&set_limit_mutex
);
3132 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3136 ret
= res_counter_set_limit(&memcg
->res
, val
);
3138 if (memswlimit
== val
)
3139 memcg
->memsw_is_minimum
= true;
3141 memcg
->memsw_is_minimum
= false;
3143 mutex_unlock(&set_limit_mutex
);
3148 mem_cgroup_hierarchical_reclaim(memcg
, NULL
, GFP_KERNEL
,
3149 MEM_CGROUP_RECLAIM_SHRINK
);
3150 curusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3151 /* Usage is reduced ? */
3152 if (curusage
>= oldusage
)
3155 oldusage
= curusage
;
3157 if (!ret
&& enlarge
)
3158 memcg_oom_recover(memcg
);
3163 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
3164 unsigned long long val
)
3167 u64 memlimit
, memswlimit
, oldusage
, curusage
;
3168 int children
= mem_cgroup_count_children(memcg
);
3172 /* see mem_cgroup_resize_res_limit */
3173 retry_count
= children
* MEM_CGROUP_RECLAIM_RETRIES
;
3174 oldusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3175 while (retry_count
) {
3176 if (signal_pending(current
)) {
3181 * Rather than hide all in some function, I do this in
3182 * open coded manner. You see what this really does.
3183 * We have to guarantee mem->res.limit < mem->memsw.limit.
3185 mutex_lock(&set_limit_mutex
);
3186 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3187 if (memlimit
> val
) {
3189 mutex_unlock(&set_limit_mutex
);
3192 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3193 if (memswlimit
< val
)
3195 ret
= res_counter_set_limit(&memcg
->memsw
, val
);
3197 if (memlimit
== val
)
3198 memcg
->memsw_is_minimum
= true;
3200 memcg
->memsw_is_minimum
= false;
3202 mutex_unlock(&set_limit_mutex
);
3207 mem_cgroup_hierarchical_reclaim(memcg
, NULL
, GFP_KERNEL
,
3208 MEM_CGROUP_RECLAIM_NOSWAP
|
3209 MEM_CGROUP_RECLAIM_SHRINK
);
3210 curusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3211 /* Usage is reduced ? */
3212 if (curusage
>= oldusage
)
3215 oldusage
= curusage
;
3217 if (!ret
&& enlarge
)
3218 memcg_oom_recover(memcg
);
3222 unsigned long mem_cgroup_soft_limit_reclaim(struct zone
*zone
, int order
,
3225 unsigned long nr_reclaimed
= 0;
3226 struct mem_cgroup_per_zone
*mz
, *next_mz
= NULL
;
3227 unsigned long reclaimed
;
3229 struct mem_cgroup_tree_per_zone
*mctz
;
3230 unsigned long long excess
;
3235 mctz
= soft_limit_tree_node_zone(zone_to_nid(zone
), zone_idx(zone
));
3237 * This loop can run a while, specially if mem_cgroup's continuously
3238 * keep exceeding their soft limit and putting the system under
3245 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
3249 reclaimed
= mem_cgroup_hierarchical_reclaim(mz
->mem
, zone
,
3251 MEM_CGROUP_RECLAIM_SOFT
);
3252 nr_reclaimed
+= reclaimed
;
3253 spin_lock(&mctz
->lock
);
3256 * If we failed to reclaim anything from this memory cgroup
3257 * it is time to move on to the next cgroup
3263 * Loop until we find yet another one.
3265 * By the time we get the soft_limit lock
3266 * again, someone might have aded the
3267 * group back on the RB tree. Iterate to
3268 * make sure we get a different mem.
3269 * mem_cgroup_largest_soft_limit_node returns
3270 * NULL if no other cgroup is present on
3274 __mem_cgroup_largest_soft_limit_node(mctz
);
3275 if (next_mz
== mz
) {
3276 css_put(&next_mz
->mem
->css
);
3278 } else /* next_mz == NULL or other memcg */
3282 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
3283 excess
= res_counter_soft_limit_excess(&mz
->mem
->res
);
3285 * One school of thought says that we should not add
3286 * back the node to the tree if reclaim returns 0.
3287 * But our reclaim could return 0, simply because due
3288 * to priority we are exposing a smaller subset of
3289 * memory to reclaim from. Consider this as a longer
3292 /* If excess == 0, no tree ops */
3293 __mem_cgroup_insert_exceeded(mz
->mem
, mz
, mctz
, excess
);
3294 spin_unlock(&mctz
->lock
);
3295 css_put(&mz
->mem
->css
);
3298 * Could not reclaim anything and there are no more
3299 * mem cgroups to try or we seem to be looping without
3300 * reclaiming anything.
3302 if (!nr_reclaimed
&&
3304 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
3306 } while (!nr_reclaimed
);
3308 css_put(&next_mz
->mem
->css
);
3309 return nr_reclaimed
;
3313 * This routine traverse page_cgroup in given list and drop them all.
3314 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3316 static int mem_cgroup_force_empty_list(struct mem_cgroup
*mem
,
3317 int node
, int zid
, enum lru_list lru
)
3320 struct mem_cgroup_per_zone
*mz
;
3321 struct page_cgroup
*pc
, *busy
;
3322 unsigned long flags
, loop
;
3323 struct list_head
*list
;
3326 zone
= &NODE_DATA(node
)->node_zones
[zid
];
3327 mz
= mem_cgroup_zoneinfo(mem
, node
, zid
);
3328 list
= &mz
->lists
[lru
];
3330 loop
= MEM_CGROUP_ZSTAT(mz
, lru
);
3331 /* give some margin against EBUSY etc...*/
3338 spin_lock_irqsave(&zone
->lru_lock
, flags
);
3339 if (list_empty(list
)) {
3340 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3343 pc
= list_entry(list
->prev
, struct page_cgroup
, lru
);
3345 list_move(&pc
->lru
, list
);
3347 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3350 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3352 page
= lookup_cgroup_page(pc
);
3354 ret
= mem_cgroup_move_parent(page
, pc
, mem
, GFP_KERNEL
);
3358 if (ret
== -EBUSY
|| ret
== -EINVAL
) {
3359 /* found lock contention or "pc" is obsolete. */
3366 if (!ret
&& !list_empty(list
))
3372 * make mem_cgroup's charge to be 0 if there is no task.
3373 * This enables deleting this mem_cgroup.
3375 static int mem_cgroup_force_empty(struct mem_cgroup
*mem
, bool free_all
)
3378 int node
, zid
, shrink
;
3379 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
3380 struct cgroup
*cgrp
= mem
->css
.cgroup
;
3385 /* should free all ? */
3391 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
))
3394 if (signal_pending(current
))
3396 /* This is for making all *used* pages to be on LRU. */
3397 lru_add_drain_all();
3398 drain_all_stock_sync();
3400 mem_cgroup_start_move(mem
);
3401 for_each_node_state(node
, N_HIGH_MEMORY
) {
3402 for (zid
= 0; !ret
&& zid
< MAX_NR_ZONES
; zid
++) {
3405 ret
= mem_cgroup_force_empty_list(mem
,
3414 mem_cgroup_end_move(mem
);
3415 memcg_oom_recover(mem
);
3416 /* it seems parent cgroup doesn't have enough mem */
3420 /* "ret" should also be checked to ensure all lists are empty. */
3421 } while (mem
->res
.usage
> 0 || ret
);
3427 /* returns EBUSY if there is a task or if we come here twice. */
3428 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
) || shrink
) {
3432 /* we call try-to-free pages for make this cgroup empty */
3433 lru_add_drain_all();
3434 /* try to free all pages in this cgroup */
3436 while (nr_retries
&& mem
->res
.usage
> 0) {
3439 if (signal_pending(current
)) {
3443 progress
= try_to_free_mem_cgroup_pages(mem
, GFP_KERNEL
,
3444 false, get_swappiness(mem
));
3447 /* maybe some writeback is necessary */
3448 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3453 /* try move_account...there may be some *locked* pages. */
3457 int mem_cgroup_force_empty_write(struct cgroup
*cont
, unsigned int event
)
3459 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont
), true);
3463 static u64
mem_cgroup_hierarchy_read(struct cgroup
*cont
, struct cftype
*cft
)
3465 return mem_cgroup_from_cont(cont
)->use_hierarchy
;
3468 static int mem_cgroup_hierarchy_write(struct cgroup
*cont
, struct cftype
*cft
,
3472 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
3473 struct cgroup
*parent
= cont
->parent
;
3474 struct mem_cgroup
*parent_mem
= NULL
;
3477 parent_mem
= mem_cgroup_from_cont(parent
);
3481 * If parent's use_hierarchy is set, we can't make any modifications
3482 * in the child subtrees. If it is unset, then the change can
3483 * occur, provided the current cgroup has no children.
3485 * For the root cgroup, parent_mem is NULL, we allow value to be
3486 * set if there are no children.
3488 if ((!parent_mem
|| !parent_mem
->use_hierarchy
) &&
3489 (val
== 1 || val
== 0)) {
3490 if (list_empty(&cont
->children
))
3491 mem
->use_hierarchy
= val
;
3502 static u64
mem_cgroup_get_recursive_idx_stat(struct mem_cgroup
*mem
,
3503 enum mem_cgroup_stat_index idx
)
3505 struct mem_cgroup
*iter
;
3508 /* each per cpu's value can be minus.Then, use s64 */
3509 for_each_mem_cgroup_tree(iter
, mem
)
3510 val
+= mem_cgroup_read_stat(iter
, idx
);
3512 if (val
< 0) /* race ? */
3517 static inline u64
mem_cgroup_usage(struct mem_cgroup
*mem
, bool swap
)
3521 if (!mem_cgroup_is_root(mem
)) {
3523 return res_counter_read_u64(&mem
->res
, RES_USAGE
);
3525 return res_counter_read_u64(&mem
->memsw
, RES_USAGE
);
3528 val
= mem_cgroup_get_recursive_idx_stat(mem
, MEM_CGROUP_STAT_CACHE
);
3529 val
+= mem_cgroup_get_recursive_idx_stat(mem
, MEM_CGROUP_STAT_RSS
);
3532 val
+= mem_cgroup_get_recursive_idx_stat(mem
,
3533 MEM_CGROUP_STAT_SWAPOUT
);
3535 return val
<< PAGE_SHIFT
;
3538 static u64
mem_cgroup_read(struct cgroup
*cont
, struct cftype
*cft
)
3540 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
3544 type
= MEMFILE_TYPE(cft
->private);
3545 name
= MEMFILE_ATTR(cft
->private);
3548 if (name
== RES_USAGE
)
3549 val
= mem_cgroup_usage(mem
, false);
3551 val
= res_counter_read_u64(&mem
->res
, name
);
3554 if (name
== RES_USAGE
)
3555 val
= mem_cgroup_usage(mem
, true);
3557 val
= res_counter_read_u64(&mem
->memsw
, name
);
3566 * The user of this function is...
3569 static int mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
3572 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3574 unsigned long long val
;
3577 type
= MEMFILE_TYPE(cft
->private);
3578 name
= MEMFILE_ATTR(cft
->private);
3581 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
3585 /* This function does all necessary parse...reuse it */
3586 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3590 ret
= mem_cgroup_resize_limit(memcg
, val
);
3592 ret
= mem_cgroup_resize_memsw_limit(memcg
, val
);
3594 case RES_SOFT_LIMIT
:
3595 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3599 * For memsw, soft limits are hard to implement in terms
3600 * of semantics, for now, we support soft limits for
3601 * control without swap
3604 ret
= res_counter_set_soft_limit(&memcg
->res
, val
);
3609 ret
= -EINVAL
; /* should be BUG() ? */
3615 static void memcg_get_hierarchical_limit(struct mem_cgroup
*memcg
,
3616 unsigned long long *mem_limit
, unsigned long long *memsw_limit
)
3618 struct cgroup
*cgroup
;
3619 unsigned long long min_limit
, min_memsw_limit
, tmp
;
3621 min_limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3622 min_memsw_limit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3623 cgroup
= memcg
->css
.cgroup
;
3624 if (!memcg
->use_hierarchy
)
3627 while (cgroup
->parent
) {
3628 cgroup
= cgroup
->parent
;
3629 memcg
= mem_cgroup_from_cont(cgroup
);
3630 if (!memcg
->use_hierarchy
)
3632 tmp
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3633 min_limit
= min(min_limit
, tmp
);
3634 tmp
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3635 min_memsw_limit
= min(min_memsw_limit
, tmp
);
3638 *mem_limit
= min_limit
;
3639 *memsw_limit
= min_memsw_limit
;
3643 static int mem_cgroup_reset(struct cgroup
*cont
, unsigned int event
)
3645 struct mem_cgroup
*mem
;
3648 mem
= mem_cgroup_from_cont(cont
);
3649 type
= MEMFILE_TYPE(event
);
3650 name
= MEMFILE_ATTR(event
);
3654 res_counter_reset_max(&mem
->res
);
3656 res_counter_reset_max(&mem
->memsw
);
3660 res_counter_reset_failcnt(&mem
->res
);
3662 res_counter_reset_failcnt(&mem
->memsw
);
3669 static u64
mem_cgroup_move_charge_read(struct cgroup
*cgrp
,
3672 return mem_cgroup_from_cont(cgrp
)->move_charge_at_immigrate
;
3676 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
3677 struct cftype
*cft
, u64 val
)
3679 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
3681 if (val
>= (1 << NR_MOVE_TYPE
))
3684 * We check this value several times in both in can_attach() and
3685 * attach(), so we need cgroup lock to prevent this value from being
3689 mem
->move_charge_at_immigrate
= val
;
3695 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
3696 struct cftype
*cft
, u64 val
)
3703 /* For read statistics */
3719 struct mcs_total_stat
{
3720 s64 stat
[NR_MCS_STAT
];
3726 } memcg_stat_strings
[NR_MCS_STAT
] = {
3727 {"cache", "total_cache"},
3728 {"rss", "total_rss"},
3729 {"mapped_file", "total_mapped_file"},
3730 {"pgpgin", "total_pgpgin"},
3731 {"pgpgout", "total_pgpgout"},
3732 {"swap", "total_swap"},
3733 {"inactive_anon", "total_inactive_anon"},
3734 {"active_anon", "total_active_anon"},
3735 {"inactive_file", "total_inactive_file"},
3736 {"active_file", "total_active_file"},
3737 {"unevictable", "total_unevictable"}
3742 mem_cgroup_get_local_stat(struct mem_cgroup
*mem
, struct mcs_total_stat
*s
)
3747 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_CACHE
);
3748 s
->stat
[MCS_CACHE
] += val
* PAGE_SIZE
;
3749 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_RSS
);
3750 s
->stat
[MCS_RSS
] += val
* PAGE_SIZE
;
3751 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_FILE_MAPPED
);
3752 s
->stat
[MCS_FILE_MAPPED
] += val
* PAGE_SIZE
;
3753 val
= mem_cgroup_read_events(mem
, MEM_CGROUP_EVENTS_PGPGIN
);
3754 s
->stat
[MCS_PGPGIN
] += val
;
3755 val
= mem_cgroup_read_events(mem
, MEM_CGROUP_EVENTS_PGPGOUT
);
3756 s
->stat
[MCS_PGPGOUT
] += val
;
3757 if (do_swap_account
) {
3758 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_SWAPOUT
);
3759 s
->stat
[MCS_SWAP
] += val
* PAGE_SIZE
;
3763 val
= mem_cgroup_get_local_zonestat(mem
, LRU_INACTIVE_ANON
);
3764 s
->stat
[MCS_INACTIVE_ANON
] += val
* PAGE_SIZE
;
3765 val
= mem_cgroup_get_local_zonestat(mem
, LRU_ACTIVE_ANON
);
3766 s
->stat
[MCS_ACTIVE_ANON
] += val
* PAGE_SIZE
;
3767 val
= mem_cgroup_get_local_zonestat(mem
, LRU_INACTIVE_FILE
);
3768 s
->stat
[MCS_INACTIVE_FILE
] += val
* PAGE_SIZE
;
3769 val
= mem_cgroup_get_local_zonestat(mem
, LRU_ACTIVE_FILE
);
3770 s
->stat
[MCS_ACTIVE_FILE
] += val
* PAGE_SIZE
;
3771 val
= mem_cgroup_get_local_zonestat(mem
, LRU_UNEVICTABLE
);
3772 s
->stat
[MCS_UNEVICTABLE
] += val
* PAGE_SIZE
;
3776 mem_cgroup_get_total_stat(struct mem_cgroup
*mem
, struct mcs_total_stat
*s
)
3778 struct mem_cgroup
*iter
;
3780 for_each_mem_cgroup_tree(iter
, mem
)
3781 mem_cgroup_get_local_stat(iter
, s
);
3784 static int mem_control_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
3785 struct cgroup_map_cb
*cb
)
3787 struct mem_cgroup
*mem_cont
= mem_cgroup_from_cont(cont
);
3788 struct mcs_total_stat mystat
;
3791 memset(&mystat
, 0, sizeof(mystat
));
3792 mem_cgroup_get_local_stat(mem_cont
, &mystat
);
3794 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
3795 if (i
== MCS_SWAP
&& !do_swap_account
)
3797 cb
->fill(cb
, memcg_stat_strings
[i
].local_name
, mystat
.stat
[i
]);
3800 /* Hierarchical information */
3802 unsigned long long limit
, memsw_limit
;
3803 memcg_get_hierarchical_limit(mem_cont
, &limit
, &memsw_limit
);
3804 cb
->fill(cb
, "hierarchical_memory_limit", limit
);
3805 if (do_swap_account
)
3806 cb
->fill(cb
, "hierarchical_memsw_limit", memsw_limit
);
3809 memset(&mystat
, 0, sizeof(mystat
));
3810 mem_cgroup_get_total_stat(mem_cont
, &mystat
);
3811 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
3812 if (i
== MCS_SWAP
&& !do_swap_account
)
3814 cb
->fill(cb
, memcg_stat_strings
[i
].total_name
, mystat
.stat
[i
]);
3817 #ifdef CONFIG_DEBUG_VM
3818 cb
->fill(cb
, "inactive_ratio", calc_inactive_ratio(mem_cont
, NULL
));
3822 struct mem_cgroup_per_zone
*mz
;
3823 unsigned long recent_rotated
[2] = {0, 0};
3824 unsigned long recent_scanned
[2] = {0, 0};
3826 for_each_online_node(nid
)
3827 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
3828 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
3830 recent_rotated
[0] +=
3831 mz
->reclaim_stat
.recent_rotated
[0];
3832 recent_rotated
[1] +=
3833 mz
->reclaim_stat
.recent_rotated
[1];
3834 recent_scanned
[0] +=
3835 mz
->reclaim_stat
.recent_scanned
[0];
3836 recent_scanned
[1] +=
3837 mz
->reclaim_stat
.recent_scanned
[1];
3839 cb
->fill(cb
, "recent_rotated_anon", recent_rotated
[0]);
3840 cb
->fill(cb
, "recent_rotated_file", recent_rotated
[1]);
3841 cb
->fill(cb
, "recent_scanned_anon", recent_scanned
[0]);
3842 cb
->fill(cb
, "recent_scanned_file", recent_scanned
[1]);
3849 static u64
mem_cgroup_swappiness_read(struct cgroup
*cgrp
, struct cftype
*cft
)
3851 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3853 return get_swappiness(memcg
);
3856 static int mem_cgroup_swappiness_write(struct cgroup
*cgrp
, struct cftype
*cft
,
3859 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3860 struct mem_cgroup
*parent
;
3865 if (cgrp
->parent
== NULL
)
3868 parent
= mem_cgroup_from_cont(cgrp
->parent
);
3872 /* If under hierarchy, only empty-root can set this value */
3873 if ((parent
->use_hierarchy
) ||
3874 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
3879 memcg
->swappiness
= val
;
3886 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
3888 struct mem_cgroup_threshold_ary
*t
;
3894 t
= rcu_dereference(memcg
->thresholds
.primary
);
3896 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
3901 usage
= mem_cgroup_usage(memcg
, swap
);
3904 * current_threshold points to threshold just below usage.
3905 * If it's not true, a threshold was crossed after last
3906 * call of __mem_cgroup_threshold().
3908 i
= t
->current_threshold
;
3911 * Iterate backward over array of thresholds starting from
3912 * current_threshold and check if a threshold is crossed.
3913 * If none of thresholds below usage is crossed, we read
3914 * only one element of the array here.
3916 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
3917 eventfd_signal(t
->entries
[i
].eventfd
, 1);
3919 /* i = current_threshold + 1 */
3923 * Iterate forward over array of thresholds starting from
3924 * current_threshold+1 and check if a threshold is crossed.
3925 * If none of thresholds above usage is crossed, we read
3926 * only one element of the array here.
3928 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
3929 eventfd_signal(t
->entries
[i
].eventfd
, 1);
3931 /* Update current_threshold */
3932 t
->current_threshold
= i
- 1;
3937 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
3940 __mem_cgroup_threshold(memcg
, false);
3941 if (do_swap_account
)
3942 __mem_cgroup_threshold(memcg
, true);
3944 memcg
= parent_mem_cgroup(memcg
);
3948 static int compare_thresholds(const void *a
, const void *b
)
3950 const struct mem_cgroup_threshold
*_a
= a
;
3951 const struct mem_cgroup_threshold
*_b
= b
;
3953 return _a
->threshold
- _b
->threshold
;
3956 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*mem
)
3958 struct mem_cgroup_eventfd_list
*ev
;
3960 list_for_each_entry(ev
, &mem
->oom_notify
, list
)
3961 eventfd_signal(ev
->eventfd
, 1);
3965 static void mem_cgroup_oom_notify(struct mem_cgroup
*mem
)
3967 struct mem_cgroup
*iter
;
3969 for_each_mem_cgroup_tree(iter
, mem
)
3970 mem_cgroup_oom_notify_cb(iter
);
3973 static int mem_cgroup_usage_register_event(struct cgroup
*cgrp
,
3974 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
3976 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3977 struct mem_cgroup_thresholds
*thresholds
;
3978 struct mem_cgroup_threshold_ary
*new;
3979 int type
= MEMFILE_TYPE(cft
->private);
3980 u64 threshold
, usage
;
3983 ret
= res_counter_memparse_write_strategy(args
, &threshold
);
3987 mutex_lock(&memcg
->thresholds_lock
);
3990 thresholds
= &memcg
->thresholds
;
3991 else if (type
== _MEMSWAP
)
3992 thresholds
= &memcg
->memsw_thresholds
;
3996 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
3998 /* Check if a threshold crossed before adding a new one */
3999 if (thresholds
->primary
)
4000 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4002 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
4004 /* Allocate memory for new array of thresholds */
4005 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
4013 /* Copy thresholds (if any) to new array */
4014 if (thresholds
->primary
) {
4015 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
4016 sizeof(struct mem_cgroup_threshold
));
4019 /* Add new threshold */
4020 new->entries
[size
- 1].eventfd
= eventfd
;
4021 new->entries
[size
- 1].threshold
= threshold
;
4023 /* Sort thresholds. Registering of new threshold isn't time-critical */
4024 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
4025 compare_thresholds
, NULL
);
4027 /* Find current threshold */
4028 new->current_threshold
= -1;
4029 for (i
= 0; i
< size
; i
++) {
4030 if (new->entries
[i
].threshold
< usage
) {
4032 * new->current_threshold will not be used until
4033 * rcu_assign_pointer(), so it's safe to increment
4036 ++new->current_threshold
;
4040 /* Free old spare buffer and save old primary buffer as spare */
4041 kfree(thresholds
->spare
);
4042 thresholds
->spare
= thresholds
->primary
;
4044 rcu_assign_pointer(thresholds
->primary
, new);
4046 /* To be sure that nobody uses thresholds */
4050 mutex_unlock(&memcg
->thresholds_lock
);
4055 static void mem_cgroup_usage_unregister_event(struct cgroup
*cgrp
,
4056 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4058 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4059 struct mem_cgroup_thresholds
*thresholds
;
4060 struct mem_cgroup_threshold_ary
*new;
4061 int type
= MEMFILE_TYPE(cft
->private);
4065 mutex_lock(&memcg
->thresholds_lock
);
4067 thresholds
= &memcg
->thresholds
;
4068 else if (type
== _MEMSWAP
)
4069 thresholds
= &memcg
->memsw_thresholds
;
4074 * Something went wrong if we trying to unregister a threshold
4075 * if we don't have thresholds
4077 BUG_ON(!thresholds
);
4079 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4081 /* Check if a threshold crossed before removing */
4082 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4084 /* Calculate new number of threshold */
4086 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
4087 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
4091 new = thresholds
->spare
;
4093 /* Set thresholds array to NULL if we don't have thresholds */
4102 /* Copy thresholds and find current threshold */
4103 new->current_threshold
= -1;
4104 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
4105 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
4108 new->entries
[j
] = thresholds
->primary
->entries
[i
];
4109 if (new->entries
[j
].threshold
< usage
) {
4111 * new->current_threshold will not be used
4112 * until rcu_assign_pointer(), so it's safe to increment
4115 ++new->current_threshold
;
4121 /* Swap primary and spare array */
4122 thresholds
->spare
= thresholds
->primary
;
4123 rcu_assign_pointer(thresholds
->primary
, new);
4125 /* To be sure that nobody uses thresholds */
4128 mutex_unlock(&memcg
->thresholds_lock
);
4131 static int mem_cgroup_oom_register_event(struct cgroup
*cgrp
,
4132 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4134 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4135 struct mem_cgroup_eventfd_list
*event
;
4136 int type
= MEMFILE_TYPE(cft
->private);
4138 BUG_ON(type
!= _OOM_TYPE
);
4139 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
4143 mutex_lock(&memcg_oom_mutex
);
4145 event
->eventfd
= eventfd
;
4146 list_add(&event
->list
, &memcg
->oom_notify
);
4148 /* already in OOM ? */
4149 if (atomic_read(&memcg
->oom_lock
))
4150 eventfd_signal(eventfd
, 1);
4151 mutex_unlock(&memcg_oom_mutex
);
4156 static void mem_cgroup_oom_unregister_event(struct cgroup
*cgrp
,
4157 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4159 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
4160 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
4161 int type
= MEMFILE_TYPE(cft
->private);
4163 BUG_ON(type
!= _OOM_TYPE
);
4165 mutex_lock(&memcg_oom_mutex
);
4167 list_for_each_entry_safe(ev
, tmp
, &mem
->oom_notify
, list
) {
4168 if (ev
->eventfd
== eventfd
) {
4169 list_del(&ev
->list
);
4174 mutex_unlock(&memcg_oom_mutex
);
4177 static int mem_cgroup_oom_control_read(struct cgroup
*cgrp
,
4178 struct cftype
*cft
, struct cgroup_map_cb
*cb
)
4180 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
4182 cb
->fill(cb
, "oom_kill_disable", mem
->oom_kill_disable
);
4184 if (atomic_read(&mem
->oom_lock
))
4185 cb
->fill(cb
, "under_oom", 1);
4187 cb
->fill(cb
, "under_oom", 0);
4191 static int mem_cgroup_oom_control_write(struct cgroup
*cgrp
,
4192 struct cftype
*cft
, u64 val
)
4194 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
4195 struct mem_cgroup
*parent
;
4197 /* cannot set to root cgroup and only 0 and 1 are allowed */
4198 if (!cgrp
->parent
|| !((val
== 0) || (val
== 1)))
4201 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4204 /* oom-kill-disable is a flag for subhierarchy. */
4205 if ((parent
->use_hierarchy
) ||
4206 (mem
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4210 mem
->oom_kill_disable
= val
;
4212 memcg_oom_recover(mem
);
4217 static struct cftype mem_cgroup_files
[] = {
4219 .name
= "usage_in_bytes",
4220 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
4221 .read_u64
= mem_cgroup_read
,
4222 .register_event
= mem_cgroup_usage_register_event
,
4223 .unregister_event
= mem_cgroup_usage_unregister_event
,
4226 .name
= "max_usage_in_bytes",
4227 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
4228 .trigger
= mem_cgroup_reset
,
4229 .read_u64
= mem_cgroup_read
,
4232 .name
= "limit_in_bytes",
4233 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
4234 .write_string
= mem_cgroup_write
,
4235 .read_u64
= mem_cgroup_read
,
4238 .name
= "soft_limit_in_bytes",
4239 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
4240 .write_string
= mem_cgroup_write
,
4241 .read_u64
= mem_cgroup_read
,
4245 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
4246 .trigger
= mem_cgroup_reset
,
4247 .read_u64
= mem_cgroup_read
,
4251 .read_map
= mem_control_stat_show
,
4254 .name
= "force_empty",
4255 .trigger
= mem_cgroup_force_empty_write
,
4258 .name
= "use_hierarchy",
4259 .write_u64
= mem_cgroup_hierarchy_write
,
4260 .read_u64
= mem_cgroup_hierarchy_read
,
4263 .name
= "swappiness",
4264 .read_u64
= mem_cgroup_swappiness_read
,
4265 .write_u64
= mem_cgroup_swappiness_write
,
4268 .name
= "move_charge_at_immigrate",
4269 .read_u64
= mem_cgroup_move_charge_read
,
4270 .write_u64
= mem_cgroup_move_charge_write
,
4273 .name
= "oom_control",
4274 .read_map
= mem_cgroup_oom_control_read
,
4275 .write_u64
= mem_cgroup_oom_control_write
,
4276 .register_event
= mem_cgroup_oom_register_event
,
4277 .unregister_event
= mem_cgroup_oom_unregister_event
,
4278 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
4282 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4283 static struct cftype memsw_cgroup_files
[] = {
4285 .name
= "memsw.usage_in_bytes",
4286 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
4287 .read_u64
= mem_cgroup_read
,
4288 .register_event
= mem_cgroup_usage_register_event
,
4289 .unregister_event
= mem_cgroup_usage_unregister_event
,
4292 .name
= "memsw.max_usage_in_bytes",
4293 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
4294 .trigger
= mem_cgroup_reset
,
4295 .read_u64
= mem_cgroup_read
,
4298 .name
= "memsw.limit_in_bytes",
4299 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
4300 .write_string
= mem_cgroup_write
,
4301 .read_u64
= mem_cgroup_read
,
4304 .name
= "memsw.failcnt",
4305 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
4306 .trigger
= mem_cgroup_reset
,
4307 .read_u64
= mem_cgroup_read
,
4311 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4313 if (!do_swap_account
)
4315 return cgroup_add_files(cont
, ss
, memsw_cgroup_files
,
4316 ARRAY_SIZE(memsw_cgroup_files
));
4319 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4325 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*mem
, int node
)
4327 struct mem_cgroup_per_node
*pn
;
4328 struct mem_cgroup_per_zone
*mz
;
4330 int zone
, tmp
= node
;
4332 * This routine is called against possible nodes.
4333 * But it's BUG to call kmalloc() against offline node.
4335 * TODO: this routine can waste much memory for nodes which will
4336 * never be onlined. It's better to use memory hotplug callback
4339 if (!node_state(node
, N_NORMAL_MEMORY
))
4341 pn
= kzalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
4345 mem
->info
.nodeinfo
[node
] = pn
;
4346 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4347 mz
= &pn
->zoneinfo
[zone
];
4349 INIT_LIST_HEAD(&mz
->lists
[l
]);
4350 mz
->usage_in_excess
= 0;
4351 mz
->on_tree
= false;
4357 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*mem
, int node
)
4359 kfree(mem
->info
.nodeinfo
[node
]);
4362 static struct mem_cgroup
*mem_cgroup_alloc(void)
4364 struct mem_cgroup
*mem
;
4365 int size
= sizeof(struct mem_cgroup
);
4367 /* Can be very big if MAX_NUMNODES is very big */
4368 if (size
< PAGE_SIZE
)
4369 mem
= kzalloc(size
, GFP_KERNEL
);
4371 mem
= vzalloc(size
);
4376 mem
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4379 spin_lock_init(&mem
->pcp_counter_lock
);
4383 if (size
< PAGE_SIZE
)
4391 * At destroying mem_cgroup, references from swap_cgroup can remain.
4392 * (scanning all at force_empty is too costly...)
4394 * Instead of clearing all references at force_empty, we remember
4395 * the number of reference from swap_cgroup and free mem_cgroup when
4396 * it goes down to 0.
4398 * Removal of cgroup itself succeeds regardless of refs from swap.
4401 static void __mem_cgroup_free(struct mem_cgroup
*mem
)
4405 mem_cgroup_remove_from_trees(mem
);
4406 free_css_id(&mem_cgroup_subsys
, &mem
->css
);
4408 for_each_node_state(node
, N_POSSIBLE
)
4409 free_mem_cgroup_per_zone_info(mem
, node
);
4411 free_percpu(mem
->stat
);
4412 if (sizeof(struct mem_cgroup
) < PAGE_SIZE
)
4418 static void mem_cgroup_get(struct mem_cgroup
*mem
)
4420 atomic_inc(&mem
->refcnt
);
4423 static void __mem_cgroup_put(struct mem_cgroup
*mem
, int count
)
4425 if (atomic_sub_and_test(count
, &mem
->refcnt
)) {
4426 struct mem_cgroup
*parent
= parent_mem_cgroup(mem
);
4427 __mem_cgroup_free(mem
);
4429 mem_cgroup_put(parent
);
4433 static void mem_cgroup_put(struct mem_cgroup
*mem
)
4435 __mem_cgroup_put(mem
, 1);
4439 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4441 static struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*mem
)
4443 if (!mem
->res
.parent
)
4445 return mem_cgroup_from_res_counter(mem
->res
.parent
, res
);
4448 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4449 static void __init
enable_swap_cgroup(void)
4451 if (!mem_cgroup_disabled() && really_do_swap_account
)
4452 do_swap_account
= 1;
4455 static void __init
enable_swap_cgroup(void)
4460 static int mem_cgroup_soft_limit_tree_init(void)
4462 struct mem_cgroup_tree_per_node
*rtpn
;
4463 struct mem_cgroup_tree_per_zone
*rtpz
;
4464 int tmp
, node
, zone
;
4466 for_each_node_state(node
, N_POSSIBLE
) {
4468 if (!node_state(node
, N_NORMAL_MEMORY
))
4470 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
, tmp
);
4474 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
4476 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4477 rtpz
= &rtpn
->rb_tree_per_zone
[zone
];
4478 rtpz
->rb_root
= RB_ROOT
;
4479 spin_lock_init(&rtpz
->lock
);
4485 static struct cgroup_subsys_state
* __ref
4486 mem_cgroup_create(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
4488 struct mem_cgroup
*mem
, *parent
;
4489 long error
= -ENOMEM
;
4492 mem
= mem_cgroup_alloc();
4494 return ERR_PTR(error
);
4496 for_each_node_state(node
, N_POSSIBLE
)
4497 if (alloc_mem_cgroup_per_zone_info(mem
, node
))
4501 if (cont
->parent
== NULL
) {
4503 enable_swap_cgroup();
4505 root_mem_cgroup
= mem
;
4506 if (mem_cgroup_soft_limit_tree_init())
4508 for_each_possible_cpu(cpu
) {
4509 struct memcg_stock_pcp
*stock
=
4510 &per_cpu(memcg_stock
, cpu
);
4511 INIT_WORK(&stock
->work
, drain_local_stock
);
4513 hotcpu_notifier(memcg_cpu_hotplug_callback
, 0);
4515 parent
= mem_cgroup_from_cont(cont
->parent
);
4516 mem
->use_hierarchy
= parent
->use_hierarchy
;
4517 mem
->oom_kill_disable
= parent
->oom_kill_disable
;
4520 if (parent
&& parent
->use_hierarchy
) {
4521 res_counter_init(&mem
->res
, &parent
->res
);
4522 res_counter_init(&mem
->memsw
, &parent
->memsw
);
4524 * We increment refcnt of the parent to ensure that we can
4525 * safely access it on res_counter_charge/uncharge.
4526 * This refcnt will be decremented when freeing this
4527 * mem_cgroup(see mem_cgroup_put).
4529 mem_cgroup_get(parent
);
4531 res_counter_init(&mem
->res
, NULL
);
4532 res_counter_init(&mem
->memsw
, NULL
);
4534 mem
->last_scanned_child
= 0;
4535 INIT_LIST_HEAD(&mem
->oom_notify
);
4538 mem
->swappiness
= get_swappiness(parent
);
4539 atomic_set(&mem
->refcnt
, 1);
4540 mem
->move_charge_at_immigrate
= 0;
4541 mutex_init(&mem
->thresholds_lock
);
4544 __mem_cgroup_free(mem
);
4545 root_mem_cgroup
= NULL
;
4546 return ERR_PTR(error
);
4549 static int mem_cgroup_pre_destroy(struct cgroup_subsys
*ss
,
4550 struct cgroup
*cont
)
4552 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
4554 return mem_cgroup_force_empty(mem
, false);
4557 static void mem_cgroup_destroy(struct cgroup_subsys
*ss
,
4558 struct cgroup
*cont
)
4560 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
4562 mem_cgroup_put(mem
);
4565 static int mem_cgroup_populate(struct cgroup_subsys
*ss
,
4566 struct cgroup
*cont
)
4570 ret
= cgroup_add_files(cont
, ss
, mem_cgroup_files
,
4571 ARRAY_SIZE(mem_cgroup_files
));
4574 ret
= register_memsw_files(cont
, ss
);
4579 /* Handlers for move charge at task migration. */
4580 #define PRECHARGE_COUNT_AT_ONCE 256
4581 static int mem_cgroup_do_precharge(unsigned long count
)
4584 int batch_count
= PRECHARGE_COUNT_AT_ONCE
;
4585 struct mem_cgroup
*mem
= mc
.to
;
4587 if (mem_cgroup_is_root(mem
)) {
4588 mc
.precharge
+= count
;
4589 /* we don't need css_get for root */
4592 /* try to charge at once */
4594 struct res_counter
*dummy
;
4596 * "mem" cannot be under rmdir() because we've already checked
4597 * by cgroup_lock_live_cgroup() that it is not removed and we
4598 * are still under the same cgroup_mutex. So we can postpone
4601 if (res_counter_charge(&mem
->res
, PAGE_SIZE
* count
, &dummy
))
4603 if (do_swap_account
&& res_counter_charge(&mem
->memsw
,
4604 PAGE_SIZE
* count
, &dummy
)) {
4605 res_counter_uncharge(&mem
->res
, PAGE_SIZE
* count
);
4608 mc
.precharge
+= count
;
4612 /* fall back to one by one charge */
4614 if (signal_pending(current
)) {
4618 if (!batch_count
--) {
4619 batch_count
= PRECHARGE_COUNT_AT_ONCE
;
4622 ret
= __mem_cgroup_try_charge(NULL
, GFP_KERNEL
, 1, &mem
, false);
4624 /* mem_cgroup_clear_mc() will do uncharge later */
4632 * is_target_pte_for_mc - check a pte whether it is valid for move charge
4633 * @vma: the vma the pte to be checked belongs
4634 * @addr: the address corresponding to the pte to be checked
4635 * @ptent: the pte to be checked
4636 * @target: the pointer the target page or swap ent will be stored(can be NULL)
4639 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4640 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4641 * move charge. if @target is not NULL, the page is stored in target->page
4642 * with extra refcnt got(Callers should handle it).
4643 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4644 * target for charge migration. if @target is not NULL, the entry is stored
4647 * Called with pte lock held.
4654 enum mc_target_type
{
4655 MC_TARGET_NONE
, /* not used */
4660 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
4661 unsigned long addr
, pte_t ptent
)
4663 struct page
*page
= vm_normal_page(vma
, addr
, ptent
);
4665 if (!page
|| !page_mapped(page
))
4667 if (PageAnon(page
)) {
4668 /* we don't move shared anon */
4669 if (!move_anon() || page_mapcount(page
) > 2)
4671 } else if (!move_file())
4672 /* we ignore mapcount for file pages */
4674 if (!get_page_unless_zero(page
))
4680 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
4681 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
4684 struct page
*page
= NULL
;
4685 swp_entry_t ent
= pte_to_swp_entry(ptent
);
4687 if (!move_anon() || non_swap_entry(ent
))
4689 usage_count
= mem_cgroup_count_swap_user(ent
, &page
);
4690 if (usage_count
> 1) { /* we don't move shared anon */
4695 if (do_swap_account
)
4696 entry
->val
= ent
.val
;
4701 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
4702 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
4704 struct page
*page
= NULL
;
4705 struct inode
*inode
;
4706 struct address_space
*mapping
;
4709 if (!vma
->vm_file
) /* anonymous vma */
4714 inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
4715 mapping
= vma
->vm_file
->f_mapping
;
4716 if (pte_none(ptent
))
4717 pgoff
= linear_page_index(vma
, addr
);
4718 else /* pte_file(ptent) is true */
4719 pgoff
= pte_to_pgoff(ptent
);
4721 /* page is moved even if it's not RSS of this task(page-faulted). */
4722 if (!mapping_cap_swap_backed(mapping
)) { /* normal file */
4723 page
= find_get_page(mapping
, pgoff
);
4724 } else { /* shmem/tmpfs file. we should take account of swap too. */
4726 mem_cgroup_get_shmem_target(inode
, pgoff
, &page
, &ent
);
4727 if (do_swap_account
)
4728 entry
->val
= ent
.val
;
4734 static int is_target_pte_for_mc(struct vm_area_struct
*vma
,
4735 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
4737 struct page
*page
= NULL
;
4738 struct page_cgroup
*pc
;
4740 swp_entry_t ent
= { .val
= 0 };
4742 if (pte_present(ptent
))
4743 page
= mc_handle_present_pte(vma
, addr
, ptent
);
4744 else if (is_swap_pte(ptent
))
4745 page
= mc_handle_swap_pte(vma
, addr
, ptent
, &ent
);
4746 else if (pte_none(ptent
) || pte_file(ptent
))
4747 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
4749 if (!page
&& !ent
.val
)
4752 pc
= lookup_page_cgroup(page
);
4754 * Do only loose check w/o page_cgroup lock.
4755 * mem_cgroup_move_account() checks the pc is valid or not under
4758 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
4759 ret
= MC_TARGET_PAGE
;
4761 target
->page
= page
;
4763 if (!ret
|| !target
)
4766 /* There is a swap entry and a page doesn't exist or isn't charged */
4767 if (ent
.val
&& !ret
&&
4768 css_id(&mc
.from
->css
) == lookup_swap_cgroup(ent
)) {
4769 ret
= MC_TARGET_SWAP
;
4776 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
4777 unsigned long addr
, unsigned long end
,
4778 struct mm_walk
*walk
)
4780 struct vm_area_struct
*vma
= walk
->private;
4784 split_huge_page_pmd(walk
->mm
, pmd
);
4786 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
4787 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
4788 if (is_target_pte_for_mc(vma
, addr
, *pte
, NULL
))
4789 mc
.precharge
++; /* increment precharge temporarily */
4790 pte_unmap_unlock(pte
- 1, ptl
);
4796 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
4798 unsigned long precharge
;
4799 struct vm_area_struct
*vma
;
4801 down_read(&mm
->mmap_sem
);
4802 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
4803 struct mm_walk mem_cgroup_count_precharge_walk
= {
4804 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
4808 if (is_vm_hugetlb_page(vma
))
4810 walk_page_range(vma
->vm_start
, vma
->vm_end
,
4811 &mem_cgroup_count_precharge_walk
);
4813 up_read(&mm
->mmap_sem
);
4815 precharge
= mc
.precharge
;
4821 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
4823 unsigned long precharge
= mem_cgroup_count_precharge(mm
);
4825 VM_BUG_ON(mc
.moving_task
);
4826 mc
.moving_task
= current
;
4827 return mem_cgroup_do_precharge(precharge
);
4830 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4831 static void __mem_cgroup_clear_mc(void)
4833 struct mem_cgroup
*from
= mc
.from
;
4834 struct mem_cgroup
*to
= mc
.to
;
4836 /* we must uncharge all the leftover precharges from mc.to */
4838 __mem_cgroup_cancel_charge(mc
.to
, mc
.precharge
);
4842 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4843 * we must uncharge here.
4845 if (mc
.moved_charge
) {
4846 __mem_cgroup_cancel_charge(mc
.from
, mc
.moved_charge
);
4847 mc
.moved_charge
= 0;
4849 /* we must fixup refcnts and charges */
4850 if (mc
.moved_swap
) {
4851 /* uncharge swap account from the old cgroup */
4852 if (!mem_cgroup_is_root(mc
.from
))
4853 res_counter_uncharge(&mc
.from
->memsw
,
4854 PAGE_SIZE
* mc
.moved_swap
);
4855 __mem_cgroup_put(mc
.from
, mc
.moved_swap
);
4857 if (!mem_cgroup_is_root(mc
.to
)) {
4859 * we charged both to->res and to->memsw, so we should
4862 res_counter_uncharge(&mc
.to
->res
,
4863 PAGE_SIZE
* mc
.moved_swap
);
4865 /* we've already done mem_cgroup_get(mc.to) */
4868 memcg_oom_recover(from
);
4869 memcg_oom_recover(to
);
4870 wake_up_all(&mc
.waitq
);
4873 static void mem_cgroup_clear_mc(void)
4875 struct mem_cgroup
*from
= mc
.from
;
4878 * we must clear moving_task before waking up waiters at the end of
4881 mc
.moving_task
= NULL
;
4882 __mem_cgroup_clear_mc();
4883 spin_lock(&mc
.lock
);
4886 spin_unlock(&mc
.lock
);
4887 mem_cgroup_end_move(from
);
4890 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
4891 struct cgroup
*cgroup
,
4892 struct task_struct
*p
,
4896 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgroup
);
4898 if (mem
->move_charge_at_immigrate
) {
4899 struct mm_struct
*mm
;
4900 struct mem_cgroup
*from
= mem_cgroup_from_task(p
);
4902 VM_BUG_ON(from
== mem
);
4904 mm
= get_task_mm(p
);
4907 /* We move charges only when we move a owner of the mm */
4908 if (mm
->owner
== p
) {
4911 VM_BUG_ON(mc
.precharge
);
4912 VM_BUG_ON(mc
.moved_charge
);
4913 VM_BUG_ON(mc
.moved_swap
);
4914 mem_cgroup_start_move(from
);
4915 spin_lock(&mc
.lock
);
4918 spin_unlock(&mc
.lock
);
4919 /* We set mc.moving_task later */
4921 ret
= mem_cgroup_precharge_mc(mm
);
4923 mem_cgroup_clear_mc();
4930 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
4931 struct cgroup
*cgroup
,
4932 struct task_struct
*p
,
4935 mem_cgroup_clear_mc();
4938 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
4939 unsigned long addr
, unsigned long end
,
4940 struct mm_walk
*walk
)
4943 struct vm_area_struct
*vma
= walk
->private;
4947 split_huge_page_pmd(walk
->mm
, pmd
);
4949 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
4950 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
4951 pte_t ptent
= *(pte
++);
4952 union mc_target target
;
4955 struct page_cgroup
*pc
;
4961 type
= is_target_pte_for_mc(vma
, addr
, ptent
, &target
);
4963 case MC_TARGET_PAGE
:
4965 if (isolate_lru_page(page
))
4967 pc
= lookup_page_cgroup(page
);
4968 if (!mem_cgroup_move_account(page
, 1, pc
,
4969 mc
.from
, mc
.to
, false)) {
4971 /* we uncharge from mc.from later. */
4974 putback_lru_page(page
);
4975 put
: /* is_target_pte_for_mc() gets the page */
4978 case MC_TARGET_SWAP
:
4980 if (!mem_cgroup_move_swap_account(ent
,
4981 mc
.from
, mc
.to
, false)) {
4983 /* we fixup refcnts and charges later. */
4991 pte_unmap_unlock(pte
- 1, ptl
);
4996 * We have consumed all precharges we got in can_attach().
4997 * We try charge one by one, but don't do any additional
4998 * charges to mc.to if we have failed in charge once in attach()
5001 ret
= mem_cgroup_do_precharge(1);
5009 static void mem_cgroup_move_charge(struct mm_struct
*mm
)
5011 struct vm_area_struct
*vma
;
5013 lru_add_drain_all();
5015 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
5017 * Someone who are holding the mmap_sem might be waiting in
5018 * waitq. So we cancel all extra charges, wake up all waiters,
5019 * and retry. Because we cancel precharges, we might not be able
5020 * to move enough charges, but moving charge is a best-effort
5021 * feature anyway, so it wouldn't be a big problem.
5023 __mem_cgroup_clear_mc();
5027 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5029 struct mm_walk mem_cgroup_move_charge_walk
= {
5030 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
5034 if (is_vm_hugetlb_page(vma
))
5036 ret
= walk_page_range(vma
->vm_start
, vma
->vm_end
,
5037 &mem_cgroup_move_charge_walk
);
5040 * means we have consumed all precharges and failed in
5041 * doing additional charge. Just abandon here.
5045 up_read(&mm
->mmap_sem
);
5048 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
5049 struct cgroup
*cont
,
5050 struct cgroup
*old_cont
,
5051 struct task_struct
*p
,
5054 struct mm_struct
*mm
;
5057 /* no need to move charge */
5060 mm
= get_task_mm(p
);
5062 mem_cgroup_move_charge(mm
);
5065 mem_cgroup_clear_mc();
5067 #else /* !CONFIG_MMU */
5068 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
5069 struct cgroup
*cgroup
,
5070 struct task_struct
*p
,
5075 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
5076 struct cgroup
*cgroup
,
5077 struct task_struct
*p
,
5081 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
5082 struct cgroup
*cont
,
5083 struct cgroup
*old_cont
,
5084 struct task_struct
*p
,
5090 struct cgroup_subsys mem_cgroup_subsys
= {
5092 .subsys_id
= mem_cgroup_subsys_id
,
5093 .create
= mem_cgroup_create
,
5094 .pre_destroy
= mem_cgroup_pre_destroy
,
5095 .destroy
= mem_cgroup_destroy
,
5096 .populate
= mem_cgroup_populate
,
5097 .can_attach
= mem_cgroup_can_attach
,
5098 .cancel_attach
= mem_cgroup_cancel_attach
,
5099 .attach
= mem_cgroup_move_task
,
5104 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
5105 static int __init
enable_swap_account(char *s
)
5107 /* consider enabled if no parameter or 1 is given */
5108 if (!(*s
) || !strcmp(s
, "=1"))
5109 really_do_swap_account
= 1;
5110 else if (!strcmp(s
, "=0"))
5111 really_do_swap_account
= 0;
5114 __setup("swapaccount", enable_swap_account
);
5116 static int __init
disable_swap_account(char *s
)
5118 printk_once("noswapaccount is deprecated and will be removed in 2.6.40. Use swapaccount=0 instead\n");
5119 enable_swap_account("=0");
5122 __setup("noswapaccount", disable_swap_account
);