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 * Kernel Memory Controller
14 * Copyright (C) 2012 Parallels Inc. and Google Inc.
15 * Authors: Glauber Costa and Suleiman Souhlal
18 * Charge lifetime sanitation
19 * Lockless page tracking & accounting
20 * Unified hierarchy configuration model
21 * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner
23 * This program is free software; you can redistribute it and/or modify
24 * it under the terms of the GNU General Public License as published by
25 * the Free Software Foundation; either version 2 of the License, or
26 * (at your option) any later version.
28 * This program is distributed in the hope that it will be useful,
29 * but WITHOUT ANY WARRANTY; without even the implied warranty of
30 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
31 * GNU General Public License for more details.
34 #include <linux/page_counter.h>
35 #include <linux/memcontrol.h>
36 #include <linux/cgroup.h>
38 #include <linux/sched/mm.h>
39 #include <linux/shmem_fs.h>
40 #include <linux/hugetlb.h>
41 #include <linux/pagemap.h>
42 #include <linux/smp.h>
43 #include <linux/page-flags.h>
44 #include <linux/backing-dev.h>
45 #include <linux/bit_spinlock.h>
46 #include <linux/rcupdate.h>
47 #include <linux/limits.h>
48 #include <linux/export.h>
49 #include <linux/mutex.h>
50 #include <linux/rbtree.h>
51 #include <linux/slab.h>
52 #include <linux/swap.h>
53 #include <linux/swapops.h>
54 #include <linux/spinlock.h>
55 #include <linux/eventfd.h>
56 #include <linux/poll.h>
57 #include <linux/sort.h>
59 #include <linux/seq_file.h>
60 #include <linux/vmpressure.h>
61 #include <linux/mm_inline.h>
62 #include <linux/swap_cgroup.h>
63 #include <linux/cpu.h>
64 #include <linux/oom.h>
65 #include <linux/lockdep.h>
66 #include <linux/file.h>
67 #include <linux/tracehook.h>
73 #include <linux/uaccess.h>
75 #include <trace/events/vmscan.h>
77 struct cgroup_subsys memory_cgrp_subsys __read_mostly
;
78 EXPORT_SYMBOL(memory_cgrp_subsys
);
80 struct mem_cgroup
*root_mem_cgroup __read_mostly
;
82 #define MEM_CGROUP_RECLAIM_RETRIES 5
84 /* Socket memory accounting disabled? */
85 static bool cgroup_memory_nosocket
;
87 /* Kernel memory accounting disabled? */
88 static bool cgroup_memory_nokmem
;
90 /* Whether the swap controller is active */
91 #ifdef CONFIG_MEMCG_SWAP
92 int do_swap_account __read_mostly
;
94 #define do_swap_account 0
97 /* Whether legacy memory+swap accounting is active */
98 static bool do_memsw_account(void)
100 return !cgroup_subsys_on_dfl(memory_cgrp_subsys
) && do_swap_account
;
103 static const char *const mem_cgroup_lru_names
[] = {
111 #define THRESHOLDS_EVENTS_TARGET 128
112 #define SOFTLIMIT_EVENTS_TARGET 1024
113 #define NUMAINFO_EVENTS_TARGET 1024
116 * Cgroups above their limits are maintained in a RB-Tree, independent of
117 * their hierarchy representation
120 struct mem_cgroup_tree_per_node
{
121 struct rb_root rb_root
;
122 struct rb_node
*rb_rightmost
;
126 struct mem_cgroup_tree
{
127 struct mem_cgroup_tree_per_node
*rb_tree_per_node
[MAX_NUMNODES
];
130 static struct mem_cgroup_tree soft_limit_tree __read_mostly
;
133 struct mem_cgroup_eventfd_list
{
134 struct list_head list
;
135 struct eventfd_ctx
*eventfd
;
139 * cgroup_event represents events which userspace want to receive.
141 struct mem_cgroup_event
{
143 * memcg which the event belongs to.
145 struct mem_cgroup
*memcg
;
147 * eventfd to signal userspace about the event.
149 struct eventfd_ctx
*eventfd
;
151 * Each of these stored in a list by the cgroup.
153 struct list_head list
;
155 * register_event() callback will be used to add new userspace
156 * waiter for changes related to this event. Use eventfd_signal()
157 * on eventfd to send notification to userspace.
159 int (*register_event
)(struct mem_cgroup
*memcg
,
160 struct eventfd_ctx
*eventfd
, const char *args
);
162 * unregister_event() callback will be called when userspace closes
163 * the eventfd or on cgroup removing. This callback must be set,
164 * if you want provide notification functionality.
166 void (*unregister_event
)(struct mem_cgroup
*memcg
,
167 struct eventfd_ctx
*eventfd
);
169 * All fields below needed to unregister event when
170 * userspace closes eventfd.
173 wait_queue_head_t
*wqh
;
174 wait_queue_entry_t wait
;
175 struct work_struct remove
;
178 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
);
179 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
);
181 /* Stuffs for move charges at task migration. */
183 * Types of charges to be moved.
185 #define MOVE_ANON 0x1U
186 #define MOVE_FILE 0x2U
187 #define MOVE_MASK (MOVE_ANON | MOVE_FILE)
189 /* "mc" and its members are protected by cgroup_mutex */
190 static struct move_charge_struct
{
191 spinlock_t lock
; /* for from, to */
192 struct mm_struct
*mm
;
193 struct mem_cgroup
*from
;
194 struct mem_cgroup
*to
;
196 unsigned long precharge
;
197 unsigned long moved_charge
;
198 unsigned long moved_swap
;
199 struct task_struct
*moving_task
; /* a task moving charges */
200 wait_queue_head_t waitq
; /* a waitq for other context */
202 .lock
= __SPIN_LOCK_UNLOCKED(mc
.lock
),
203 .waitq
= __WAIT_QUEUE_HEAD_INITIALIZER(mc
.waitq
),
207 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
208 * limit reclaim to prevent infinite loops, if they ever occur.
210 #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
211 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
214 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
215 MEM_CGROUP_CHARGE_TYPE_ANON
,
216 MEM_CGROUP_CHARGE_TYPE_SWAPOUT
, /* for accounting swapcache */
217 MEM_CGROUP_CHARGE_TYPE_DROP
, /* a page was unused swap cache */
221 /* for encoding cft->private value on file */
230 #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
231 #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
232 #define MEMFILE_ATTR(val) ((val) & 0xffff)
233 /* Used for OOM nofiier */
234 #define OOM_CONTROL (0)
236 /* Some nice accessors for the vmpressure. */
237 struct vmpressure
*memcg_to_vmpressure(struct mem_cgroup
*memcg
)
240 memcg
= root_mem_cgroup
;
241 return &memcg
->vmpressure
;
244 struct cgroup_subsys_state
*vmpressure_to_css(struct vmpressure
*vmpr
)
246 return &container_of(vmpr
, struct mem_cgroup
, vmpressure
)->css
;
249 static inline bool mem_cgroup_is_root(struct mem_cgroup
*memcg
)
251 return (memcg
== root_mem_cgroup
);
256 * This will be the memcg's index in each cache's ->memcg_params.memcg_caches.
257 * The main reason for not using cgroup id for this:
258 * this works better in sparse environments, where we have a lot of memcgs,
259 * but only a few kmem-limited. Or also, if we have, for instance, 200
260 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
261 * 200 entry array for that.
263 * The current size of the caches array is stored in memcg_nr_cache_ids. It
264 * will double each time we have to increase it.
266 static DEFINE_IDA(memcg_cache_ida
);
267 int memcg_nr_cache_ids
;
269 /* Protects memcg_nr_cache_ids */
270 static DECLARE_RWSEM(memcg_cache_ids_sem
);
272 void memcg_get_cache_ids(void)
274 down_read(&memcg_cache_ids_sem
);
277 void memcg_put_cache_ids(void)
279 up_read(&memcg_cache_ids_sem
);
283 * MIN_SIZE is different than 1, because we would like to avoid going through
284 * the alloc/free process all the time. In a small machine, 4 kmem-limited
285 * cgroups is a reasonable guess. In the future, it could be a parameter or
286 * tunable, but that is strictly not necessary.
288 * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get
289 * this constant directly from cgroup, but it is understandable that this is
290 * better kept as an internal representation in cgroup.c. In any case, the
291 * cgrp_id space is not getting any smaller, and we don't have to necessarily
292 * increase ours as well if it increases.
294 #define MEMCG_CACHES_MIN_SIZE 4
295 #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
298 * A lot of the calls to the cache allocation functions are expected to be
299 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
300 * conditional to this static branch, we'll have to allow modules that does
301 * kmem_cache_alloc and the such to see this symbol as well
303 DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key
);
304 EXPORT_SYMBOL(memcg_kmem_enabled_key
);
306 struct workqueue_struct
*memcg_kmem_cache_wq
;
308 #endif /* !CONFIG_SLOB */
311 * mem_cgroup_css_from_page - css of the memcg associated with a page
312 * @page: page of interest
314 * If memcg is bound to the default hierarchy, css of the memcg associated
315 * with @page is returned. The returned css remains associated with @page
316 * until it is released.
318 * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup
321 struct cgroup_subsys_state
*mem_cgroup_css_from_page(struct page
*page
)
323 struct mem_cgroup
*memcg
;
325 memcg
= page
->mem_cgroup
;
327 if (!memcg
|| !cgroup_subsys_on_dfl(memory_cgrp_subsys
))
328 memcg
= root_mem_cgroup
;
334 * page_cgroup_ino - return inode number of the memcg a page is charged to
337 * Look up the closest online ancestor of the memory cgroup @page is charged to
338 * and return its inode number or 0 if @page is not charged to any cgroup. It
339 * is safe to call this function without holding a reference to @page.
341 * Note, this function is inherently racy, because there is nothing to prevent
342 * the cgroup inode from getting torn down and potentially reallocated a moment
343 * after page_cgroup_ino() returns, so it only should be used by callers that
344 * do not care (such as procfs interfaces).
346 ino_t
page_cgroup_ino(struct page
*page
)
348 struct mem_cgroup
*memcg
;
349 unsigned long ino
= 0;
352 memcg
= READ_ONCE(page
->mem_cgroup
);
353 while (memcg
&& !(memcg
->css
.flags
& CSS_ONLINE
))
354 memcg
= parent_mem_cgroup(memcg
);
356 ino
= cgroup_ino(memcg
->css
.cgroup
);
361 static struct mem_cgroup_per_node
*
362 mem_cgroup_page_nodeinfo(struct mem_cgroup
*memcg
, struct page
*page
)
364 int nid
= page_to_nid(page
);
366 return memcg
->nodeinfo
[nid
];
369 static struct mem_cgroup_tree_per_node
*
370 soft_limit_tree_node(int nid
)
372 return soft_limit_tree
.rb_tree_per_node
[nid
];
375 static struct mem_cgroup_tree_per_node
*
376 soft_limit_tree_from_page(struct page
*page
)
378 int nid
= page_to_nid(page
);
380 return soft_limit_tree
.rb_tree_per_node
[nid
];
383 static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node
*mz
,
384 struct mem_cgroup_tree_per_node
*mctz
,
385 unsigned long new_usage_in_excess
)
387 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
388 struct rb_node
*parent
= NULL
;
389 struct mem_cgroup_per_node
*mz_node
;
390 bool rightmost
= true;
395 mz
->usage_in_excess
= new_usage_in_excess
;
396 if (!mz
->usage_in_excess
)
400 mz_node
= rb_entry(parent
, struct mem_cgroup_per_node
,
402 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
) {
408 * We can't avoid mem cgroups that are over their soft
409 * limit by the same amount
411 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
416 mctz
->rb_rightmost
= &mz
->tree_node
;
418 rb_link_node(&mz
->tree_node
, parent
, p
);
419 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
423 static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node
*mz
,
424 struct mem_cgroup_tree_per_node
*mctz
)
429 if (&mz
->tree_node
== mctz
->rb_rightmost
)
430 mctz
->rb_rightmost
= rb_prev(&mz
->tree_node
);
432 rb_erase(&mz
->tree_node
, &mctz
->rb_root
);
436 static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node
*mz
,
437 struct mem_cgroup_tree_per_node
*mctz
)
441 spin_lock_irqsave(&mctz
->lock
, flags
);
442 __mem_cgroup_remove_exceeded(mz
, mctz
);
443 spin_unlock_irqrestore(&mctz
->lock
, flags
);
446 static unsigned long soft_limit_excess(struct mem_cgroup
*memcg
)
448 unsigned long nr_pages
= page_counter_read(&memcg
->memory
);
449 unsigned long soft_limit
= READ_ONCE(memcg
->soft_limit
);
450 unsigned long excess
= 0;
452 if (nr_pages
> soft_limit
)
453 excess
= nr_pages
- soft_limit
;
458 static void mem_cgroup_update_tree(struct mem_cgroup
*memcg
, struct page
*page
)
460 unsigned long excess
;
461 struct mem_cgroup_per_node
*mz
;
462 struct mem_cgroup_tree_per_node
*mctz
;
464 mctz
= soft_limit_tree_from_page(page
);
468 * Necessary to update all ancestors when hierarchy is used.
469 * because their event counter is not touched.
471 for (; memcg
; memcg
= parent_mem_cgroup(memcg
)) {
472 mz
= mem_cgroup_page_nodeinfo(memcg
, page
);
473 excess
= soft_limit_excess(memcg
);
475 * We have to update the tree if mz is on RB-tree or
476 * mem is over its softlimit.
478 if (excess
|| mz
->on_tree
) {
481 spin_lock_irqsave(&mctz
->lock
, flags
);
482 /* if on-tree, remove it */
484 __mem_cgroup_remove_exceeded(mz
, mctz
);
486 * Insert again. mz->usage_in_excess will be updated.
487 * If excess is 0, no tree ops.
489 __mem_cgroup_insert_exceeded(mz
, mctz
, excess
);
490 spin_unlock_irqrestore(&mctz
->lock
, flags
);
495 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*memcg
)
497 struct mem_cgroup_tree_per_node
*mctz
;
498 struct mem_cgroup_per_node
*mz
;
502 mz
= mem_cgroup_nodeinfo(memcg
, nid
);
503 mctz
= soft_limit_tree_node(nid
);
505 mem_cgroup_remove_exceeded(mz
, mctz
);
509 static struct mem_cgroup_per_node
*
510 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node
*mctz
)
512 struct mem_cgroup_per_node
*mz
;
516 if (!mctz
->rb_rightmost
)
517 goto done
; /* Nothing to reclaim from */
519 mz
= rb_entry(mctz
->rb_rightmost
,
520 struct mem_cgroup_per_node
, tree_node
);
522 * Remove the node now but someone else can add it back,
523 * we will to add it back at the end of reclaim to its correct
524 * position in the tree.
526 __mem_cgroup_remove_exceeded(mz
, mctz
);
527 if (!soft_limit_excess(mz
->memcg
) ||
528 !css_tryget_online(&mz
->memcg
->css
))
534 static struct mem_cgroup_per_node
*
535 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node
*mctz
)
537 struct mem_cgroup_per_node
*mz
;
539 spin_lock_irq(&mctz
->lock
);
540 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
541 spin_unlock_irq(&mctz
->lock
);
546 * Return page count for single (non recursive) @memcg.
548 * Implementation Note: reading percpu statistics for memcg.
550 * Both of vmstat[] and percpu_counter has threshold and do periodic
551 * synchronization to implement "quick" read. There are trade-off between
552 * reading cost and precision of value. Then, we may have a chance to implement
553 * a periodic synchronization of counter in memcg's counter.
555 * But this _read() function is used for user interface now. The user accounts
556 * memory usage by memory cgroup and he _always_ requires exact value because
557 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
558 * have to visit all online cpus and make sum. So, for now, unnecessary
559 * synchronization is not implemented. (just implemented for cpu hotplug)
561 * If there are kernel internal actions which can make use of some not-exact
562 * value, and reading all cpu value can be performance bottleneck in some
563 * common workload, threshold and synchronization as vmstat[] should be
566 * The parameter idx can be of type enum memcg_event_item or vm_event_item.
569 static unsigned long memcg_sum_events(struct mem_cgroup
*memcg
,
572 unsigned long val
= 0;
575 for_each_possible_cpu(cpu
)
576 val
+= per_cpu(memcg
->stat
->events
[event
], cpu
);
580 static void mem_cgroup_charge_statistics(struct mem_cgroup
*memcg
,
582 bool compound
, int nr_pages
)
585 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
586 * counted as CACHE even if it's on ANON LRU.
589 __this_cpu_add(memcg
->stat
->count
[MEMCG_RSS
], nr_pages
);
591 __this_cpu_add(memcg
->stat
->count
[MEMCG_CACHE
], nr_pages
);
592 if (PageSwapBacked(page
))
593 __this_cpu_add(memcg
->stat
->count
[NR_SHMEM
], nr_pages
);
597 VM_BUG_ON_PAGE(!PageTransHuge(page
), page
);
598 __this_cpu_add(memcg
->stat
->count
[MEMCG_RSS_HUGE
], nr_pages
);
601 /* pagein of a big page is an event. So, ignore page size */
603 __this_cpu_inc(memcg
->stat
->events
[PGPGIN
]);
605 __this_cpu_inc(memcg
->stat
->events
[PGPGOUT
]);
606 nr_pages
= -nr_pages
; /* for event */
609 __this_cpu_add(memcg
->stat
->nr_page_events
, nr_pages
);
612 unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup
*memcg
,
613 int nid
, unsigned int lru_mask
)
615 struct lruvec
*lruvec
= mem_cgroup_lruvec(NODE_DATA(nid
), memcg
);
616 unsigned long nr
= 0;
619 VM_BUG_ON((unsigned)nid
>= nr_node_ids
);
622 if (!(BIT(lru
) & lru_mask
))
624 nr
+= mem_cgroup_get_lru_size(lruvec
, lru
);
629 static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup
*memcg
,
630 unsigned int lru_mask
)
632 unsigned long nr
= 0;
635 for_each_node_state(nid
, N_MEMORY
)
636 nr
+= mem_cgroup_node_nr_lru_pages(memcg
, nid
, lru_mask
);
640 static bool mem_cgroup_event_ratelimit(struct mem_cgroup
*memcg
,
641 enum mem_cgroup_events_target target
)
643 unsigned long val
, next
;
645 val
= __this_cpu_read(memcg
->stat
->nr_page_events
);
646 next
= __this_cpu_read(memcg
->stat
->targets
[target
]);
647 /* from time_after() in jiffies.h */
648 if ((long)(next
- val
) < 0) {
650 case MEM_CGROUP_TARGET_THRESH
:
651 next
= val
+ THRESHOLDS_EVENTS_TARGET
;
653 case MEM_CGROUP_TARGET_SOFTLIMIT
:
654 next
= val
+ SOFTLIMIT_EVENTS_TARGET
;
656 case MEM_CGROUP_TARGET_NUMAINFO
:
657 next
= val
+ NUMAINFO_EVENTS_TARGET
;
662 __this_cpu_write(memcg
->stat
->targets
[target
], next
);
669 * Check events in order.
672 static void memcg_check_events(struct mem_cgroup
*memcg
, struct page
*page
)
674 /* threshold event is triggered in finer grain than soft limit */
675 if (unlikely(mem_cgroup_event_ratelimit(memcg
,
676 MEM_CGROUP_TARGET_THRESH
))) {
678 bool do_numainfo __maybe_unused
;
680 do_softlimit
= mem_cgroup_event_ratelimit(memcg
,
681 MEM_CGROUP_TARGET_SOFTLIMIT
);
683 do_numainfo
= mem_cgroup_event_ratelimit(memcg
,
684 MEM_CGROUP_TARGET_NUMAINFO
);
686 mem_cgroup_threshold(memcg
);
687 if (unlikely(do_softlimit
))
688 mem_cgroup_update_tree(memcg
, page
);
690 if (unlikely(do_numainfo
))
691 atomic_inc(&memcg
->numainfo_events
);
696 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
699 * mm_update_next_owner() may clear mm->owner to NULL
700 * if it races with swapoff, page migration, etc.
701 * So this can be called with p == NULL.
706 return mem_cgroup_from_css(task_css(p
, memory_cgrp_id
));
708 EXPORT_SYMBOL(mem_cgroup_from_task
);
710 static struct mem_cgroup
*get_mem_cgroup_from_mm(struct mm_struct
*mm
)
712 struct mem_cgroup
*memcg
= NULL
;
717 * Page cache insertions can happen withou an
718 * actual mm context, e.g. during disk probing
719 * on boot, loopback IO, acct() writes etc.
722 memcg
= root_mem_cgroup
;
724 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
725 if (unlikely(!memcg
))
726 memcg
= root_mem_cgroup
;
728 } while (!css_tryget(&memcg
->css
));
734 * mem_cgroup_iter - iterate over memory cgroup hierarchy
735 * @root: hierarchy root
736 * @prev: previously returned memcg, NULL on first invocation
737 * @reclaim: cookie for shared reclaim walks, NULL for full walks
739 * Returns references to children of the hierarchy below @root, or
740 * @root itself, or %NULL after a full round-trip.
742 * Caller must pass the return value in @prev on subsequent
743 * invocations for reference counting, or use mem_cgroup_iter_break()
744 * to cancel a hierarchy walk before the round-trip is complete.
746 * Reclaimers can specify a zone and a priority level in @reclaim to
747 * divide up the memcgs in the hierarchy among all concurrent
748 * reclaimers operating on the same zone and priority.
750 struct mem_cgroup
*mem_cgroup_iter(struct mem_cgroup
*root
,
751 struct mem_cgroup
*prev
,
752 struct mem_cgroup_reclaim_cookie
*reclaim
)
754 struct mem_cgroup_reclaim_iter
*uninitialized_var(iter
);
755 struct cgroup_subsys_state
*css
= NULL
;
756 struct mem_cgroup
*memcg
= NULL
;
757 struct mem_cgroup
*pos
= NULL
;
759 if (mem_cgroup_disabled())
763 root
= root_mem_cgroup
;
765 if (prev
&& !reclaim
)
768 if (!root
->use_hierarchy
&& root
!= root_mem_cgroup
) {
777 struct mem_cgroup_per_node
*mz
;
779 mz
= mem_cgroup_nodeinfo(root
, reclaim
->pgdat
->node_id
);
780 iter
= &mz
->iter
[reclaim
->priority
];
782 if (prev
&& reclaim
->generation
!= iter
->generation
)
786 pos
= READ_ONCE(iter
->position
);
787 if (!pos
|| css_tryget(&pos
->css
))
790 * css reference reached zero, so iter->position will
791 * be cleared by ->css_released. However, we should not
792 * rely on this happening soon, because ->css_released
793 * is called from a work queue, and by busy-waiting we
794 * might block it. So we clear iter->position right
797 (void)cmpxchg(&iter
->position
, pos
, NULL
);
805 css
= css_next_descendant_pre(css
, &root
->css
);
808 * Reclaimers share the hierarchy walk, and a
809 * new one might jump in right at the end of
810 * the hierarchy - make sure they see at least
811 * one group and restart from the beginning.
819 * Verify the css and acquire a reference. The root
820 * is provided by the caller, so we know it's alive
821 * and kicking, and don't take an extra reference.
823 memcg
= mem_cgroup_from_css(css
);
825 if (css
== &root
->css
)
836 * The position could have already been updated by a competing
837 * thread, so check that the value hasn't changed since we read
838 * it to avoid reclaiming from the same cgroup twice.
840 (void)cmpxchg(&iter
->position
, pos
, memcg
);
848 reclaim
->generation
= iter
->generation
;
854 if (prev
&& prev
!= root
)
861 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
862 * @root: hierarchy root
863 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
865 void mem_cgroup_iter_break(struct mem_cgroup
*root
,
866 struct mem_cgroup
*prev
)
869 root
= root_mem_cgroup
;
870 if (prev
&& prev
!= root
)
874 static void __invalidate_reclaim_iterators(struct mem_cgroup
*from
,
875 struct mem_cgroup
*dead_memcg
)
877 struct mem_cgroup_reclaim_iter
*iter
;
878 struct mem_cgroup_per_node
*mz
;
883 mz
= mem_cgroup_nodeinfo(from
, nid
);
884 for (i
= 0; i
<= DEF_PRIORITY
; i
++) {
886 cmpxchg(&iter
->position
,
892 static void invalidate_reclaim_iterators(struct mem_cgroup
*dead_memcg
)
894 struct mem_cgroup
*memcg
= dead_memcg
;
895 struct mem_cgroup
*last
;
898 __invalidate_reclaim_iterators(memcg
, dead_memcg
);
900 } while ((memcg
= parent_mem_cgroup(memcg
)));
903 * When cgruop1 non-hierarchy mode is used,
904 * parent_mem_cgroup() does not walk all the way up to the
905 * cgroup root (root_mem_cgroup). So we have to handle
906 * dead_memcg from cgroup root separately.
908 if (last
!= root_mem_cgroup
)
909 __invalidate_reclaim_iterators(root_mem_cgroup
,
914 * Iteration constructs for visiting all cgroups (under a tree). If
915 * loops are exited prematurely (break), mem_cgroup_iter_break() must
916 * be used for reference counting.
918 #define for_each_mem_cgroup_tree(iter, root) \
919 for (iter = mem_cgroup_iter(root, NULL, NULL); \
921 iter = mem_cgroup_iter(root, iter, NULL))
923 #define for_each_mem_cgroup(iter) \
924 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
926 iter = mem_cgroup_iter(NULL, iter, NULL))
929 * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy
930 * @memcg: hierarchy root
931 * @fn: function to call for each task
932 * @arg: argument passed to @fn
934 * This function iterates over tasks attached to @memcg or to any of its
935 * descendants and calls @fn for each task. If @fn returns a non-zero
936 * value, the function breaks the iteration loop and returns the value.
937 * Otherwise, it will iterate over all tasks and return 0.
939 * This function must not be called for the root memory cgroup.
941 int mem_cgroup_scan_tasks(struct mem_cgroup
*memcg
,
942 int (*fn
)(struct task_struct
*, void *), void *arg
)
944 struct mem_cgroup
*iter
;
947 BUG_ON(memcg
== root_mem_cgroup
);
949 for_each_mem_cgroup_tree(iter
, memcg
) {
950 struct css_task_iter it
;
951 struct task_struct
*task
;
953 css_task_iter_start(&iter
->css
, 0, &it
);
954 while (!ret
&& (task
= css_task_iter_next(&it
)))
956 css_task_iter_end(&it
);
958 mem_cgroup_iter_break(memcg
, iter
);
966 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
968 * @zone: zone of the page
970 * This function is only safe when following the LRU page isolation
971 * and putback protocol: the LRU lock must be held, and the page must
972 * either be PageLRU() or the caller must have isolated/allocated it.
974 struct lruvec
*mem_cgroup_page_lruvec(struct page
*page
, struct pglist_data
*pgdat
)
976 struct mem_cgroup_per_node
*mz
;
977 struct mem_cgroup
*memcg
;
978 struct lruvec
*lruvec
;
980 if (mem_cgroup_disabled()) {
981 lruvec
= &pgdat
->lruvec
;
985 memcg
= page
->mem_cgroup
;
987 * Swapcache readahead pages are added to the LRU - and
988 * possibly migrated - before they are charged.
991 memcg
= root_mem_cgroup
;
993 mz
= mem_cgroup_page_nodeinfo(memcg
, page
);
994 lruvec
= &mz
->lruvec
;
997 * Since a node can be onlined after the mem_cgroup was created,
998 * we have to be prepared to initialize lruvec->zone here;
999 * and if offlined then reonlined, we need to reinitialize it.
1001 if (unlikely(lruvec
->pgdat
!= pgdat
))
1002 lruvec
->pgdat
= pgdat
;
1007 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1008 * @lruvec: mem_cgroup per zone lru vector
1009 * @lru: index of lru list the page is sitting on
1010 * @zid: zone id of the accounted pages
1011 * @nr_pages: positive when adding or negative when removing
1013 * This function must be called under lru_lock, just before a page is added
1014 * to or just after a page is removed from an lru list (that ordering being
1015 * so as to allow it to check that lru_size 0 is consistent with list_empty).
1017 void mem_cgroup_update_lru_size(struct lruvec
*lruvec
, enum lru_list lru
,
1018 int zid
, int nr_pages
)
1020 struct mem_cgroup_per_node
*mz
;
1021 unsigned long *lru_size
;
1024 if (mem_cgroup_disabled())
1027 mz
= container_of(lruvec
, struct mem_cgroup_per_node
, lruvec
);
1028 lru_size
= &mz
->lru_zone_size
[zid
][lru
];
1031 *lru_size
+= nr_pages
;
1034 if (WARN_ONCE(size
< 0,
1035 "%s(%p, %d, %d): lru_size %ld\n",
1036 __func__
, lruvec
, lru
, nr_pages
, size
)) {
1042 *lru_size
+= nr_pages
;
1045 bool task_in_mem_cgroup(struct task_struct
*task
, struct mem_cgroup
*memcg
)
1047 struct mem_cgroup
*task_memcg
;
1048 struct task_struct
*p
;
1051 p
= find_lock_task_mm(task
);
1053 task_memcg
= get_mem_cgroup_from_mm(p
->mm
);
1057 * All threads may have already detached their mm's, but the oom
1058 * killer still needs to detect if they have already been oom
1059 * killed to prevent needlessly killing additional tasks.
1062 task_memcg
= mem_cgroup_from_task(task
);
1063 css_get(&task_memcg
->css
);
1066 ret
= mem_cgroup_is_descendant(task_memcg
, memcg
);
1067 css_put(&task_memcg
->css
);
1072 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1073 * @memcg: the memory cgroup
1075 * Returns the maximum amount of memory @mem can be charged with, in
1078 static unsigned long mem_cgroup_margin(struct mem_cgroup
*memcg
)
1080 unsigned long margin
= 0;
1081 unsigned long count
;
1082 unsigned long limit
;
1084 count
= page_counter_read(&memcg
->memory
);
1085 limit
= READ_ONCE(memcg
->memory
.limit
);
1087 margin
= limit
- count
;
1089 if (do_memsw_account()) {
1090 count
= page_counter_read(&memcg
->memsw
);
1091 limit
= READ_ONCE(memcg
->memsw
.limit
);
1093 margin
= min(margin
, limit
- count
);
1102 * A routine for checking "mem" is under move_account() or not.
1104 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1105 * moving cgroups. This is for waiting at high-memory pressure
1108 static bool mem_cgroup_under_move(struct mem_cgroup
*memcg
)
1110 struct mem_cgroup
*from
;
1111 struct mem_cgroup
*to
;
1114 * Unlike task_move routines, we access mc.to, mc.from not under
1115 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1117 spin_lock(&mc
.lock
);
1123 ret
= mem_cgroup_is_descendant(from
, memcg
) ||
1124 mem_cgroup_is_descendant(to
, memcg
);
1126 spin_unlock(&mc
.lock
);
1130 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*memcg
)
1132 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1133 if (mem_cgroup_under_move(memcg
)) {
1135 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1136 /* moving charge context might have finished. */
1139 finish_wait(&mc
.waitq
, &wait
);
1146 unsigned int memcg1_stats
[] = {
1157 static const char *const memcg1_stat_names
[] = {
1168 #define K(x) ((x) << (PAGE_SHIFT-10))
1170 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
1171 * @memcg: The memory cgroup that went over limit
1172 * @p: Task that is going to be killed
1174 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1177 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1179 struct mem_cgroup
*iter
;
1185 pr_info("Task in ");
1186 pr_cont_cgroup_path(task_cgroup(p
, memory_cgrp_id
));
1187 pr_cont(" killed as a result of limit of ");
1189 pr_info("Memory limit reached of cgroup ");
1192 pr_cont_cgroup_path(memcg
->css
.cgroup
);
1197 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1198 K((u64
)page_counter_read(&memcg
->memory
)),
1199 K((u64
)memcg
->memory
.limit
), memcg
->memory
.failcnt
);
1200 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1201 K((u64
)page_counter_read(&memcg
->memsw
)),
1202 K((u64
)memcg
->memsw
.limit
), memcg
->memsw
.failcnt
);
1203 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1204 K((u64
)page_counter_read(&memcg
->kmem
)),
1205 K((u64
)memcg
->kmem
.limit
), memcg
->kmem
.failcnt
);
1207 for_each_mem_cgroup_tree(iter
, memcg
) {
1208 pr_info("Memory cgroup stats for ");
1209 pr_cont_cgroup_path(iter
->css
.cgroup
);
1212 for (i
= 0; i
< ARRAY_SIZE(memcg1_stats
); i
++) {
1213 if (memcg1_stats
[i
] == MEMCG_SWAP
&& !do_swap_account
)
1215 pr_cont(" %s:%luKB", memcg1_stat_names
[i
],
1216 K(memcg_page_state(iter
, memcg1_stats
[i
])));
1219 for (i
= 0; i
< NR_LRU_LISTS
; i
++)
1220 pr_cont(" %s:%luKB", mem_cgroup_lru_names
[i
],
1221 K(mem_cgroup_nr_lru_pages(iter
, BIT(i
))));
1228 * This function returns the number of memcg under hierarchy tree. Returns
1229 * 1(self count) if no children.
1231 static int mem_cgroup_count_children(struct mem_cgroup
*memcg
)
1234 struct mem_cgroup
*iter
;
1236 for_each_mem_cgroup_tree(iter
, memcg
)
1242 * Return the memory (and swap, if configured) limit for a memcg.
1244 unsigned long mem_cgroup_get_limit(struct mem_cgroup
*memcg
)
1246 unsigned long limit
;
1248 limit
= memcg
->memory
.limit
;
1249 if (mem_cgroup_swappiness(memcg
)) {
1250 unsigned long memsw_limit
;
1251 unsigned long swap_limit
;
1253 memsw_limit
= memcg
->memsw
.limit
;
1254 swap_limit
= memcg
->swap
.limit
;
1255 swap_limit
= min(swap_limit
, (unsigned long)total_swap_pages
);
1256 limit
= min(limit
+ swap_limit
, memsw_limit
);
1261 static bool mem_cgroup_out_of_memory(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
1264 struct oom_control oc
= {
1268 .gfp_mask
= gfp_mask
,
1273 mutex_lock(&oom_lock
);
1274 ret
= out_of_memory(&oc
);
1275 mutex_unlock(&oom_lock
);
1279 #if MAX_NUMNODES > 1
1282 * test_mem_cgroup_node_reclaimable
1283 * @memcg: the target memcg
1284 * @nid: the node ID to be checked.
1285 * @noswap : specify true here if the user wants flle only information.
1287 * This function returns whether the specified memcg contains any
1288 * reclaimable pages on a node. Returns true if there are any reclaimable
1289 * pages in the node.
1291 static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup
*memcg
,
1292 int nid
, bool noswap
)
1294 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_FILE
))
1296 if (noswap
|| !total_swap_pages
)
1298 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_ANON
))
1305 * Always updating the nodemask is not very good - even if we have an empty
1306 * list or the wrong list here, we can start from some node and traverse all
1307 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1310 static void mem_cgroup_may_update_nodemask(struct mem_cgroup
*memcg
)
1314 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1315 * pagein/pageout changes since the last update.
1317 if (!atomic_read(&memcg
->numainfo_events
))
1319 if (atomic_inc_return(&memcg
->numainfo_updating
) > 1)
1322 /* make a nodemask where this memcg uses memory from */
1323 memcg
->scan_nodes
= node_states
[N_MEMORY
];
1325 for_each_node_mask(nid
, node_states
[N_MEMORY
]) {
1327 if (!test_mem_cgroup_node_reclaimable(memcg
, nid
, false))
1328 node_clear(nid
, memcg
->scan_nodes
);
1331 atomic_set(&memcg
->numainfo_events
, 0);
1332 atomic_set(&memcg
->numainfo_updating
, 0);
1336 * Selecting a node where we start reclaim from. Because what we need is just
1337 * reducing usage counter, start from anywhere is O,K. Considering
1338 * memory reclaim from current node, there are pros. and cons.
1340 * Freeing memory from current node means freeing memory from a node which
1341 * we'll use or we've used. So, it may make LRU bad. And if several threads
1342 * hit limits, it will see a contention on a node. But freeing from remote
1343 * node means more costs for memory reclaim because of memory latency.
1345 * Now, we use round-robin. Better algorithm is welcomed.
1347 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1351 mem_cgroup_may_update_nodemask(memcg
);
1352 node
= memcg
->last_scanned_node
;
1354 node
= next_node_in(node
, memcg
->scan_nodes
);
1356 * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages
1357 * last time it really checked all the LRUs due to rate limiting.
1358 * Fallback to the current node in that case for simplicity.
1360 if (unlikely(node
== MAX_NUMNODES
))
1361 node
= numa_node_id();
1363 memcg
->last_scanned_node
= node
;
1367 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1373 static int mem_cgroup_soft_reclaim(struct mem_cgroup
*root_memcg
,
1376 unsigned long *total_scanned
)
1378 struct mem_cgroup
*victim
= NULL
;
1381 unsigned long excess
;
1382 unsigned long nr_scanned
;
1383 struct mem_cgroup_reclaim_cookie reclaim
= {
1388 excess
= soft_limit_excess(root_memcg
);
1391 victim
= mem_cgroup_iter(root_memcg
, victim
, &reclaim
);
1396 * If we have not been able to reclaim
1397 * anything, it might because there are
1398 * no reclaimable pages under this hierarchy
1403 * We want to do more targeted reclaim.
1404 * excess >> 2 is not to excessive so as to
1405 * reclaim too much, nor too less that we keep
1406 * coming back to reclaim from this cgroup
1408 if (total
>= (excess
>> 2) ||
1409 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
))
1414 total
+= mem_cgroup_shrink_node(victim
, gfp_mask
, false,
1415 pgdat
, &nr_scanned
);
1416 *total_scanned
+= nr_scanned
;
1417 if (!soft_limit_excess(root_memcg
))
1420 mem_cgroup_iter_break(root_memcg
, victim
);
1424 #ifdef CONFIG_LOCKDEP
1425 static struct lockdep_map memcg_oom_lock_dep_map
= {
1426 .name
= "memcg_oom_lock",
1430 static DEFINE_SPINLOCK(memcg_oom_lock
);
1433 * Check OOM-Killer is already running under our hierarchy.
1434 * If someone is running, return false.
1436 static bool mem_cgroup_oom_trylock(struct mem_cgroup
*memcg
)
1438 struct mem_cgroup
*iter
, *failed
= NULL
;
1440 spin_lock(&memcg_oom_lock
);
1442 for_each_mem_cgroup_tree(iter
, memcg
) {
1443 if (iter
->oom_lock
) {
1445 * this subtree of our hierarchy is already locked
1446 * so we cannot give a lock.
1449 mem_cgroup_iter_break(memcg
, iter
);
1452 iter
->oom_lock
= true;
1457 * OK, we failed to lock the whole subtree so we have
1458 * to clean up what we set up to the failing subtree
1460 for_each_mem_cgroup_tree(iter
, memcg
) {
1461 if (iter
== failed
) {
1462 mem_cgroup_iter_break(memcg
, iter
);
1465 iter
->oom_lock
= false;
1468 mutex_acquire(&memcg_oom_lock_dep_map
, 0, 1, _RET_IP_
);
1470 spin_unlock(&memcg_oom_lock
);
1475 static void mem_cgroup_oom_unlock(struct mem_cgroup
*memcg
)
1477 struct mem_cgroup
*iter
;
1479 spin_lock(&memcg_oom_lock
);
1480 mutex_release(&memcg_oom_lock_dep_map
, 1, _RET_IP_
);
1481 for_each_mem_cgroup_tree(iter
, memcg
)
1482 iter
->oom_lock
= false;
1483 spin_unlock(&memcg_oom_lock
);
1486 static void mem_cgroup_mark_under_oom(struct mem_cgroup
*memcg
)
1488 struct mem_cgroup
*iter
;
1490 spin_lock(&memcg_oom_lock
);
1491 for_each_mem_cgroup_tree(iter
, memcg
)
1493 spin_unlock(&memcg_oom_lock
);
1496 static void mem_cgroup_unmark_under_oom(struct mem_cgroup
*memcg
)
1498 struct mem_cgroup
*iter
;
1501 * When a new child is created while the hierarchy is under oom,
1502 * mem_cgroup_oom_lock() may not be called. Watch for underflow.
1504 spin_lock(&memcg_oom_lock
);
1505 for_each_mem_cgroup_tree(iter
, memcg
)
1506 if (iter
->under_oom
> 0)
1508 spin_unlock(&memcg_oom_lock
);
1511 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1513 struct oom_wait_info
{
1514 struct mem_cgroup
*memcg
;
1515 wait_queue_entry_t wait
;
1518 static int memcg_oom_wake_function(wait_queue_entry_t
*wait
,
1519 unsigned mode
, int sync
, void *arg
)
1521 struct mem_cgroup
*wake_memcg
= (struct mem_cgroup
*)arg
;
1522 struct mem_cgroup
*oom_wait_memcg
;
1523 struct oom_wait_info
*oom_wait_info
;
1525 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1526 oom_wait_memcg
= oom_wait_info
->memcg
;
1528 if (!mem_cgroup_is_descendant(wake_memcg
, oom_wait_memcg
) &&
1529 !mem_cgroup_is_descendant(oom_wait_memcg
, wake_memcg
))
1531 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1534 static void memcg_oom_recover(struct mem_cgroup
*memcg
)
1537 * For the following lockless ->under_oom test, the only required
1538 * guarantee is that it must see the state asserted by an OOM when
1539 * this function is called as a result of userland actions
1540 * triggered by the notification of the OOM. This is trivially
1541 * achieved by invoking mem_cgroup_mark_under_oom() before
1542 * triggering notification.
1544 if (memcg
&& memcg
->under_oom
)
1545 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, memcg
);
1548 static void mem_cgroup_oom(struct mem_cgroup
*memcg
, gfp_t mask
, int order
)
1550 if (!current
->memcg_may_oom
)
1553 * We are in the middle of the charge context here, so we
1554 * don't want to block when potentially sitting on a callstack
1555 * that holds all kinds of filesystem and mm locks.
1557 * Also, the caller may handle a failed allocation gracefully
1558 * (like optional page cache readahead) and so an OOM killer
1559 * invocation might not even be necessary.
1561 * That's why we don't do anything here except remember the
1562 * OOM context and then deal with it at the end of the page
1563 * fault when the stack is unwound, the locks are released,
1564 * and when we know whether the fault was overall successful.
1566 css_get(&memcg
->css
);
1567 current
->memcg_in_oom
= memcg
;
1568 current
->memcg_oom_gfp_mask
= mask
;
1569 current
->memcg_oom_order
= order
;
1573 * mem_cgroup_oom_synchronize - complete memcg OOM handling
1574 * @handle: actually kill/wait or just clean up the OOM state
1576 * This has to be called at the end of a page fault if the memcg OOM
1577 * handler was enabled.
1579 * Memcg supports userspace OOM handling where failed allocations must
1580 * sleep on a waitqueue until the userspace task resolves the
1581 * situation. Sleeping directly in the charge context with all kinds
1582 * of locks held is not a good idea, instead we remember an OOM state
1583 * in the task and mem_cgroup_oom_synchronize() has to be called at
1584 * the end of the page fault to complete the OOM handling.
1586 * Returns %true if an ongoing memcg OOM situation was detected and
1587 * completed, %false otherwise.
1589 bool mem_cgroup_oom_synchronize(bool handle
)
1591 struct mem_cgroup
*memcg
= current
->memcg_in_oom
;
1592 struct oom_wait_info owait
;
1595 /* OOM is global, do not handle */
1602 owait
.memcg
= memcg
;
1603 owait
.wait
.flags
= 0;
1604 owait
.wait
.func
= memcg_oom_wake_function
;
1605 owait
.wait
.private = current
;
1606 INIT_LIST_HEAD(&owait
.wait
.entry
);
1608 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1609 mem_cgroup_mark_under_oom(memcg
);
1611 locked
= mem_cgroup_oom_trylock(memcg
);
1614 mem_cgroup_oom_notify(memcg
);
1616 if (locked
&& !memcg
->oom_kill_disable
) {
1617 mem_cgroup_unmark_under_oom(memcg
);
1618 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1619 mem_cgroup_out_of_memory(memcg
, current
->memcg_oom_gfp_mask
,
1620 current
->memcg_oom_order
);
1623 mem_cgroup_unmark_under_oom(memcg
);
1624 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1628 mem_cgroup_oom_unlock(memcg
);
1630 * There is no guarantee that an OOM-lock contender
1631 * sees the wakeups triggered by the OOM kill
1632 * uncharges. Wake any sleepers explicitely.
1634 memcg_oom_recover(memcg
);
1637 current
->memcg_in_oom
= NULL
;
1638 css_put(&memcg
->css
);
1643 * lock_page_memcg - lock a page->mem_cgroup binding
1646 * This function protects unlocked LRU pages from being moved to
1649 * It ensures lifetime of the returned memcg. Caller is responsible
1650 * for the lifetime of the page; __unlock_page_memcg() is available
1651 * when @page might get freed inside the locked section.
1653 struct mem_cgroup
*lock_page_memcg(struct page
*page
)
1655 struct mem_cgroup
*memcg
;
1656 unsigned long flags
;
1659 * The RCU lock is held throughout the transaction. The fast
1660 * path can get away without acquiring the memcg->move_lock
1661 * because page moving starts with an RCU grace period.
1663 * The RCU lock also protects the memcg from being freed when
1664 * the page state that is going to change is the only thing
1665 * preventing the page itself from being freed. E.g. writeback
1666 * doesn't hold a page reference and relies on PG_writeback to
1667 * keep off truncation, migration and so forth.
1671 if (mem_cgroup_disabled())
1674 memcg
= page
->mem_cgroup
;
1675 if (unlikely(!memcg
))
1678 if (atomic_read(&memcg
->moving_account
) <= 0)
1681 spin_lock_irqsave(&memcg
->move_lock
, flags
);
1682 if (memcg
!= page
->mem_cgroup
) {
1683 spin_unlock_irqrestore(&memcg
->move_lock
, flags
);
1688 * When charge migration first begins, we can have locked and
1689 * unlocked page stat updates happening concurrently. Track
1690 * the task who has the lock for unlock_page_memcg().
1692 memcg
->move_lock_task
= current
;
1693 memcg
->move_lock_flags
= flags
;
1697 EXPORT_SYMBOL(lock_page_memcg
);
1700 * __unlock_page_memcg - unlock and unpin a memcg
1703 * Unlock and unpin a memcg returned by lock_page_memcg().
1705 void __unlock_page_memcg(struct mem_cgroup
*memcg
)
1707 if (memcg
&& memcg
->move_lock_task
== current
) {
1708 unsigned long flags
= memcg
->move_lock_flags
;
1710 memcg
->move_lock_task
= NULL
;
1711 memcg
->move_lock_flags
= 0;
1713 spin_unlock_irqrestore(&memcg
->move_lock
, flags
);
1720 * unlock_page_memcg - unlock a page->mem_cgroup binding
1723 void unlock_page_memcg(struct page
*page
)
1725 __unlock_page_memcg(page
->mem_cgroup
);
1727 EXPORT_SYMBOL(unlock_page_memcg
);
1730 * size of first charge trial. "32" comes from vmscan.c's magic value.
1731 * TODO: maybe necessary to use big numbers in big irons.
1733 #define CHARGE_BATCH 32U
1734 struct memcg_stock_pcp
{
1735 struct mem_cgroup
*cached
; /* this never be root cgroup */
1736 unsigned int nr_pages
;
1737 struct work_struct work
;
1738 unsigned long flags
;
1739 #define FLUSHING_CACHED_CHARGE 0
1741 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
1742 static DEFINE_MUTEX(percpu_charge_mutex
);
1745 * consume_stock: Try to consume stocked charge on this cpu.
1746 * @memcg: memcg to consume from.
1747 * @nr_pages: how many pages to charge.
1749 * The charges will only happen if @memcg matches the current cpu's memcg
1750 * stock, and at least @nr_pages are available in that stock. Failure to
1751 * service an allocation will refill the stock.
1753 * returns true if successful, false otherwise.
1755 static bool consume_stock(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
1757 struct memcg_stock_pcp
*stock
;
1758 unsigned long flags
;
1761 if (nr_pages
> CHARGE_BATCH
)
1764 local_irq_save(flags
);
1766 stock
= this_cpu_ptr(&memcg_stock
);
1767 if (memcg
== stock
->cached
&& stock
->nr_pages
>= nr_pages
) {
1768 stock
->nr_pages
-= nr_pages
;
1772 local_irq_restore(flags
);
1778 * Returns stocks cached in percpu and reset cached information.
1780 static void drain_stock(struct memcg_stock_pcp
*stock
)
1782 struct mem_cgroup
*old
= stock
->cached
;
1784 if (stock
->nr_pages
) {
1785 page_counter_uncharge(&old
->memory
, stock
->nr_pages
);
1786 if (do_memsw_account())
1787 page_counter_uncharge(&old
->memsw
, stock
->nr_pages
);
1788 css_put_many(&old
->css
, stock
->nr_pages
);
1789 stock
->nr_pages
= 0;
1791 stock
->cached
= NULL
;
1794 static void drain_local_stock(struct work_struct
*dummy
)
1796 struct memcg_stock_pcp
*stock
;
1797 unsigned long flags
;
1800 * The only protection from memory hotplug vs. drain_stock races is
1801 * that we always operate on local CPU stock here with IRQ disabled
1803 local_irq_save(flags
);
1805 stock
= this_cpu_ptr(&memcg_stock
);
1807 clear_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
);
1809 local_irq_restore(flags
);
1813 * Cache charges(val) to local per_cpu area.
1814 * This will be consumed by consume_stock() function, later.
1816 static void refill_stock(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
1818 struct memcg_stock_pcp
*stock
;
1819 unsigned long flags
;
1821 local_irq_save(flags
);
1823 stock
= this_cpu_ptr(&memcg_stock
);
1824 if (stock
->cached
!= memcg
) { /* reset if necessary */
1826 stock
->cached
= memcg
;
1828 stock
->nr_pages
+= nr_pages
;
1830 if (stock
->nr_pages
> CHARGE_BATCH
)
1833 local_irq_restore(flags
);
1837 * Drains all per-CPU charge caches for given root_memcg resp. subtree
1838 * of the hierarchy under it.
1840 static void drain_all_stock(struct mem_cgroup
*root_memcg
)
1844 /* If someone's already draining, avoid adding running more workers. */
1845 if (!mutex_trylock(&percpu_charge_mutex
))
1848 * Notify other cpus that system-wide "drain" is running
1849 * We do not care about races with the cpu hotplug because cpu down
1850 * as well as workers from this path always operate on the local
1851 * per-cpu data. CPU up doesn't touch memcg_stock at all.
1854 for_each_online_cpu(cpu
) {
1855 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
1856 struct mem_cgroup
*memcg
;
1858 memcg
= stock
->cached
;
1859 if (!memcg
|| !stock
->nr_pages
|| !css_tryget(&memcg
->css
))
1861 if (!mem_cgroup_is_descendant(memcg
, root_memcg
)) {
1862 css_put(&memcg
->css
);
1865 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
)) {
1867 drain_local_stock(&stock
->work
);
1869 schedule_work_on(cpu
, &stock
->work
);
1871 css_put(&memcg
->css
);
1874 mutex_unlock(&percpu_charge_mutex
);
1877 static int memcg_hotplug_cpu_dead(unsigned int cpu
)
1879 struct memcg_stock_pcp
*stock
;
1881 stock
= &per_cpu(memcg_stock
, cpu
);
1886 static void reclaim_high(struct mem_cgroup
*memcg
,
1887 unsigned int nr_pages
,
1891 if (page_counter_read(&memcg
->memory
) <= memcg
->high
)
1893 mem_cgroup_event(memcg
, MEMCG_HIGH
);
1894 try_to_free_mem_cgroup_pages(memcg
, nr_pages
, gfp_mask
, true);
1895 } while ((memcg
= parent_mem_cgroup(memcg
)));
1898 static void high_work_func(struct work_struct
*work
)
1900 struct mem_cgroup
*memcg
;
1902 memcg
= container_of(work
, struct mem_cgroup
, high_work
);
1903 reclaim_high(memcg
, CHARGE_BATCH
, GFP_KERNEL
);
1907 * Scheduled by try_charge() to be executed from the userland return path
1908 * and reclaims memory over the high limit.
1910 void mem_cgroup_handle_over_high(void)
1912 unsigned int nr_pages
= current
->memcg_nr_pages_over_high
;
1913 struct mem_cgroup
*memcg
;
1915 if (likely(!nr_pages
))
1918 memcg
= get_mem_cgroup_from_mm(current
->mm
);
1919 reclaim_high(memcg
, nr_pages
, GFP_KERNEL
);
1920 css_put(&memcg
->css
);
1921 current
->memcg_nr_pages_over_high
= 0;
1924 static int try_charge(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
1925 unsigned int nr_pages
)
1927 unsigned int batch
= max(CHARGE_BATCH
, nr_pages
);
1928 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
1929 struct mem_cgroup
*mem_over_limit
;
1930 struct page_counter
*counter
;
1931 unsigned long nr_reclaimed
;
1932 bool may_swap
= true;
1933 bool drained
= false;
1935 if (mem_cgroup_is_root(memcg
))
1938 if (consume_stock(memcg
, nr_pages
))
1941 if (!do_memsw_account() ||
1942 page_counter_try_charge(&memcg
->memsw
, batch
, &counter
)) {
1943 if (page_counter_try_charge(&memcg
->memory
, batch
, &counter
))
1945 if (do_memsw_account())
1946 page_counter_uncharge(&memcg
->memsw
, batch
);
1947 mem_over_limit
= mem_cgroup_from_counter(counter
, memory
);
1949 mem_over_limit
= mem_cgroup_from_counter(counter
, memsw
);
1953 if (batch
> nr_pages
) {
1959 * Unlike in global OOM situations, memcg is not in a physical
1960 * memory shortage. Allow dying and OOM-killed tasks to
1961 * bypass the last charges so that they can exit quickly and
1962 * free their memory.
1964 if (unlikely(tsk_is_oom_victim(current
) ||
1965 fatal_signal_pending(current
) ||
1966 current
->flags
& PF_EXITING
))
1970 * Prevent unbounded recursion when reclaim operations need to
1971 * allocate memory. This might exceed the limits temporarily,
1972 * but we prefer facilitating memory reclaim and getting back
1973 * under the limit over triggering OOM kills in these cases.
1975 if (unlikely(current
->flags
& PF_MEMALLOC
))
1978 if (unlikely(task_in_memcg_oom(current
)))
1981 if (!gfpflags_allow_blocking(gfp_mask
))
1984 mem_cgroup_event(mem_over_limit
, MEMCG_MAX
);
1986 nr_reclaimed
= try_to_free_mem_cgroup_pages(mem_over_limit
, nr_pages
,
1987 gfp_mask
, may_swap
);
1989 if (mem_cgroup_margin(mem_over_limit
) >= nr_pages
)
1993 drain_all_stock(mem_over_limit
);
1998 if (gfp_mask
& __GFP_NORETRY
)
2001 * Even though the limit is exceeded at this point, reclaim
2002 * may have been able to free some pages. Retry the charge
2003 * before killing the task.
2005 * Only for regular pages, though: huge pages are rather
2006 * unlikely to succeed so close to the limit, and we fall back
2007 * to regular pages anyway in case of failure.
2009 if (nr_reclaimed
&& nr_pages
<= (1 << PAGE_ALLOC_COSTLY_ORDER
))
2012 * At task move, charge accounts can be doubly counted. So, it's
2013 * better to wait until the end of task_move if something is going on.
2015 if (mem_cgroup_wait_acct_move(mem_over_limit
))
2021 if (gfp_mask
& __GFP_NOFAIL
)
2024 if (fatal_signal_pending(current
))
2027 mem_cgroup_event(mem_over_limit
, MEMCG_OOM
);
2029 mem_cgroup_oom(mem_over_limit
, gfp_mask
,
2030 get_order(nr_pages
* PAGE_SIZE
));
2032 if (!(gfp_mask
& __GFP_NOFAIL
))
2036 * The allocation either can't fail or will lead to more memory
2037 * being freed very soon. Allow memory usage go over the limit
2038 * temporarily by force charging it.
2040 page_counter_charge(&memcg
->memory
, nr_pages
);
2041 if (do_memsw_account())
2042 page_counter_charge(&memcg
->memsw
, nr_pages
);
2043 css_get_many(&memcg
->css
, nr_pages
);
2048 css_get_many(&memcg
->css
, batch
);
2049 if (batch
> nr_pages
)
2050 refill_stock(memcg
, batch
- nr_pages
);
2053 * If the hierarchy is above the normal consumption range, schedule
2054 * reclaim on returning to userland. We can perform reclaim here
2055 * if __GFP_RECLAIM but let's always punt for simplicity and so that
2056 * GFP_KERNEL can consistently be used during reclaim. @memcg is
2057 * not recorded as it most likely matches current's and won't
2058 * change in the meantime. As high limit is checked again before
2059 * reclaim, the cost of mismatch is negligible.
2062 if (page_counter_read(&memcg
->memory
) > memcg
->high
) {
2063 /* Don't bother a random interrupted task */
2064 if (in_interrupt()) {
2065 schedule_work(&memcg
->high_work
);
2068 current
->memcg_nr_pages_over_high
+= batch
;
2069 set_notify_resume(current
);
2072 } while ((memcg
= parent_mem_cgroup(memcg
)));
2077 static void cancel_charge(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
2079 if (mem_cgroup_is_root(memcg
))
2082 page_counter_uncharge(&memcg
->memory
, nr_pages
);
2083 if (do_memsw_account())
2084 page_counter_uncharge(&memcg
->memsw
, nr_pages
);
2086 css_put_many(&memcg
->css
, nr_pages
);
2089 static void lock_page_lru(struct page
*page
, int *isolated
)
2091 struct zone
*zone
= page_zone(page
);
2093 spin_lock_irq(zone_lru_lock(zone
));
2094 if (PageLRU(page
)) {
2095 struct lruvec
*lruvec
;
2097 lruvec
= mem_cgroup_page_lruvec(page
, zone
->zone_pgdat
);
2099 del_page_from_lru_list(page
, lruvec
, page_lru(page
));
2105 static void unlock_page_lru(struct page
*page
, int isolated
)
2107 struct zone
*zone
= page_zone(page
);
2110 struct lruvec
*lruvec
;
2112 lruvec
= mem_cgroup_page_lruvec(page
, zone
->zone_pgdat
);
2113 VM_BUG_ON_PAGE(PageLRU(page
), page
);
2115 add_page_to_lru_list(page
, lruvec
, page_lru(page
));
2117 spin_unlock_irq(zone_lru_lock(zone
));
2120 static void commit_charge(struct page
*page
, struct mem_cgroup
*memcg
,
2125 VM_BUG_ON_PAGE(page
->mem_cgroup
, page
);
2128 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2129 * may already be on some other mem_cgroup's LRU. Take care of it.
2132 lock_page_lru(page
, &isolated
);
2135 * Nobody should be changing or seriously looking at
2136 * page->mem_cgroup at this point:
2138 * - the page is uncharged
2140 * - the page is off-LRU
2142 * - an anonymous fault has exclusive page access, except for
2143 * a locked page table
2145 * - a page cache insertion, a swapin fault, or a migration
2146 * have the page locked
2148 page
->mem_cgroup
= memcg
;
2151 unlock_page_lru(page
, isolated
);
2155 static int memcg_alloc_cache_id(void)
2160 id
= ida_simple_get(&memcg_cache_ida
,
2161 0, MEMCG_CACHES_MAX_SIZE
, GFP_KERNEL
);
2165 if (id
< memcg_nr_cache_ids
)
2169 * There's no space for the new id in memcg_caches arrays,
2170 * so we have to grow them.
2172 down_write(&memcg_cache_ids_sem
);
2174 size
= 2 * (id
+ 1);
2175 if (size
< MEMCG_CACHES_MIN_SIZE
)
2176 size
= MEMCG_CACHES_MIN_SIZE
;
2177 else if (size
> MEMCG_CACHES_MAX_SIZE
)
2178 size
= MEMCG_CACHES_MAX_SIZE
;
2180 err
= memcg_update_all_caches(size
);
2182 err
= memcg_update_all_list_lrus(size
);
2184 memcg_nr_cache_ids
= size
;
2186 up_write(&memcg_cache_ids_sem
);
2189 ida_simple_remove(&memcg_cache_ida
, id
);
2195 static void memcg_free_cache_id(int id
)
2197 ida_simple_remove(&memcg_cache_ida
, id
);
2200 struct memcg_kmem_cache_create_work
{
2201 struct mem_cgroup
*memcg
;
2202 struct kmem_cache
*cachep
;
2203 struct work_struct work
;
2206 static void memcg_kmem_cache_create_func(struct work_struct
*w
)
2208 struct memcg_kmem_cache_create_work
*cw
=
2209 container_of(w
, struct memcg_kmem_cache_create_work
, work
);
2210 struct mem_cgroup
*memcg
= cw
->memcg
;
2211 struct kmem_cache
*cachep
= cw
->cachep
;
2213 memcg_create_kmem_cache(memcg
, cachep
);
2215 css_put(&memcg
->css
);
2220 * Enqueue the creation of a per-memcg kmem_cache.
2222 static void __memcg_schedule_kmem_cache_create(struct mem_cgroup
*memcg
,
2223 struct kmem_cache
*cachep
)
2225 struct memcg_kmem_cache_create_work
*cw
;
2227 cw
= kmalloc(sizeof(*cw
), GFP_NOWAIT
| __GFP_NOWARN
);
2231 css_get(&memcg
->css
);
2234 cw
->cachep
= cachep
;
2235 INIT_WORK(&cw
->work
, memcg_kmem_cache_create_func
);
2237 queue_work(memcg_kmem_cache_wq
, &cw
->work
);
2240 static void memcg_schedule_kmem_cache_create(struct mem_cgroup
*memcg
,
2241 struct kmem_cache
*cachep
)
2244 * We need to stop accounting when we kmalloc, because if the
2245 * corresponding kmalloc cache is not yet created, the first allocation
2246 * in __memcg_schedule_kmem_cache_create will recurse.
2248 * However, it is better to enclose the whole function. Depending on
2249 * the debugging options enabled, INIT_WORK(), for instance, can
2250 * trigger an allocation. This too, will make us recurse. Because at
2251 * this point we can't allow ourselves back into memcg_kmem_get_cache,
2252 * the safest choice is to do it like this, wrapping the whole function.
2254 current
->memcg_kmem_skip_account
= 1;
2255 __memcg_schedule_kmem_cache_create(memcg
, cachep
);
2256 current
->memcg_kmem_skip_account
= 0;
2259 static inline bool memcg_kmem_bypass(void)
2261 if (in_interrupt() || !current
->mm
|| (current
->flags
& PF_KTHREAD
))
2267 * memcg_kmem_get_cache: select the correct per-memcg cache for allocation
2268 * @cachep: the original global kmem cache
2270 * Return the kmem_cache we're supposed to use for a slab allocation.
2271 * We try to use the current memcg's version of the cache.
2273 * If the cache does not exist yet, if we are the first user of it, we
2274 * create it asynchronously in a workqueue and let the current allocation
2275 * go through with the original cache.
2277 * This function takes a reference to the cache it returns to assure it
2278 * won't get destroyed while we are working with it. Once the caller is
2279 * done with it, memcg_kmem_put_cache() must be called to release the
2282 struct kmem_cache
*memcg_kmem_get_cache(struct kmem_cache
*cachep
)
2284 struct mem_cgroup
*memcg
;
2285 struct kmem_cache
*memcg_cachep
;
2288 VM_BUG_ON(!is_root_cache(cachep
));
2290 if (memcg_kmem_bypass())
2293 if (current
->memcg_kmem_skip_account
)
2296 memcg
= get_mem_cgroup_from_mm(current
->mm
);
2297 kmemcg_id
= READ_ONCE(memcg
->kmemcg_id
);
2301 memcg_cachep
= cache_from_memcg_idx(cachep
, kmemcg_id
);
2302 if (likely(memcg_cachep
))
2303 return memcg_cachep
;
2306 * If we are in a safe context (can wait, and not in interrupt
2307 * context), we could be be predictable and return right away.
2308 * This would guarantee that the allocation being performed
2309 * already belongs in the new cache.
2311 * However, there are some clashes that can arrive from locking.
2312 * For instance, because we acquire the slab_mutex while doing
2313 * memcg_create_kmem_cache, this means no further allocation
2314 * could happen with the slab_mutex held. So it's better to
2317 memcg_schedule_kmem_cache_create(memcg
, cachep
);
2319 css_put(&memcg
->css
);
2324 * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache
2325 * @cachep: the cache returned by memcg_kmem_get_cache
2327 void memcg_kmem_put_cache(struct kmem_cache
*cachep
)
2329 if (!is_root_cache(cachep
))
2330 css_put(&cachep
->memcg_params
.memcg
->css
);
2334 * memcg_kmem_charge: charge a kmem page
2335 * @page: page to charge
2336 * @gfp: reclaim mode
2337 * @order: allocation order
2338 * @memcg: memory cgroup to charge
2340 * Returns 0 on success, an error code on failure.
2342 int memcg_kmem_charge_memcg(struct page
*page
, gfp_t gfp
, int order
,
2343 struct mem_cgroup
*memcg
)
2345 unsigned int nr_pages
= 1 << order
;
2346 struct page_counter
*counter
;
2349 ret
= try_charge(memcg
, gfp
, nr_pages
);
2353 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys
) &&
2354 !page_counter_try_charge(&memcg
->kmem
, nr_pages
, &counter
)) {
2357 * Enforce __GFP_NOFAIL allocation because callers are not
2358 * prepared to see failures and likely do not have any failure
2361 if (gfp
& __GFP_NOFAIL
) {
2362 page_counter_charge(&memcg
->kmem
, nr_pages
);
2365 cancel_charge(memcg
, nr_pages
);
2369 page
->mem_cgroup
= memcg
;
2375 * memcg_kmem_charge: charge a kmem page to the current memory cgroup
2376 * @page: page to charge
2377 * @gfp: reclaim mode
2378 * @order: allocation order
2380 * Returns 0 on success, an error code on failure.
2382 int memcg_kmem_charge(struct page
*page
, gfp_t gfp
, int order
)
2384 struct mem_cgroup
*memcg
;
2387 if (memcg_kmem_bypass())
2390 memcg
= get_mem_cgroup_from_mm(current
->mm
);
2391 if (!mem_cgroup_is_root(memcg
)) {
2392 ret
= memcg_kmem_charge_memcg(page
, gfp
, order
, memcg
);
2394 __SetPageKmemcg(page
);
2396 css_put(&memcg
->css
);
2400 * memcg_kmem_uncharge: uncharge a kmem page
2401 * @page: page to uncharge
2402 * @order: allocation order
2404 void memcg_kmem_uncharge(struct page
*page
, int order
)
2406 struct mem_cgroup
*memcg
= page
->mem_cgroup
;
2407 unsigned int nr_pages
= 1 << order
;
2412 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg
), page
);
2414 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys
))
2415 page_counter_uncharge(&memcg
->kmem
, nr_pages
);
2417 page_counter_uncharge(&memcg
->memory
, nr_pages
);
2418 if (do_memsw_account())
2419 page_counter_uncharge(&memcg
->memsw
, nr_pages
);
2421 page
->mem_cgroup
= NULL
;
2423 /* slab pages do not have PageKmemcg flag set */
2424 if (PageKmemcg(page
))
2425 __ClearPageKmemcg(page
);
2427 css_put_many(&memcg
->css
, nr_pages
);
2429 #endif /* !CONFIG_SLOB */
2431 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2434 * Because tail pages are not marked as "used", set it. We're under
2435 * zone_lru_lock and migration entries setup in all page mappings.
2437 void mem_cgroup_split_huge_fixup(struct page
*head
)
2441 if (mem_cgroup_disabled())
2444 for (i
= 1; i
< HPAGE_PMD_NR
; i
++)
2445 head
[i
].mem_cgroup
= head
->mem_cgroup
;
2447 __this_cpu_sub(head
->mem_cgroup
->stat
->count
[MEMCG_RSS_HUGE
],
2450 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2452 #ifdef CONFIG_MEMCG_SWAP
2453 static void mem_cgroup_swap_statistics(struct mem_cgroup
*memcg
,
2456 this_cpu_add(memcg
->stat
->count
[MEMCG_SWAP
], nr_entries
);
2460 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2461 * @entry: swap entry to be moved
2462 * @from: mem_cgroup which the entry is moved from
2463 * @to: mem_cgroup which the entry is moved to
2465 * It succeeds only when the swap_cgroup's record for this entry is the same
2466 * as the mem_cgroup's id of @from.
2468 * Returns 0 on success, -EINVAL on failure.
2470 * The caller must have charged to @to, IOW, called page_counter_charge() about
2471 * both res and memsw, and called css_get().
2473 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
2474 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
2476 unsigned short old_id
, new_id
;
2478 old_id
= mem_cgroup_id(from
);
2479 new_id
= mem_cgroup_id(to
);
2481 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
2482 mem_cgroup_swap_statistics(from
, -1);
2483 mem_cgroup_swap_statistics(to
, 1);
2489 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
2490 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
2496 static DEFINE_MUTEX(memcg_limit_mutex
);
2498 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
2499 unsigned long limit
)
2501 unsigned long curusage
;
2502 unsigned long oldusage
;
2503 bool enlarge
= false;
2508 * For keeping hierarchical_reclaim simple, how long we should retry
2509 * is depends on callers. We set our retry-count to be function
2510 * of # of children which we should visit in this loop.
2512 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
*
2513 mem_cgroup_count_children(memcg
);
2515 oldusage
= page_counter_read(&memcg
->memory
);
2518 if (signal_pending(current
)) {
2523 mutex_lock(&memcg_limit_mutex
);
2524 if (limit
> memcg
->memsw
.limit
) {
2525 mutex_unlock(&memcg_limit_mutex
);
2529 if (limit
> memcg
->memory
.limit
)
2531 ret
= page_counter_limit(&memcg
->memory
, limit
);
2532 mutex_unlock(&memcg_limit_mutex
);
2537 try_to_free_mem_cgroup_pages(memcg
, 1, GFP_KERNEL
, true);
2539 curusage
= page_counter_read(&memcg
->memory
);
2540 /* Usage is reduced ? */
2541 if (curusage
>= oldusage
)
2544 oldusage
= curusage
;
2545 } while (retry_count
);
2547 if (!ret
&& enlarge
)
2548 memcg_oom_recover(memcg
);
2553 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
2554 unsigned long limit
)
2556 unsigned long curusage
;
2557 unsigned long oldusage
;
2558 bool enlarge
= false;
2562 /* see mem_cgroup_resize_res_limit */
2563 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
*
2564 mem_cgroup_count_children(memcg
);
2566 oldusage
= page_counter_read(&memcg
->memsw
);
2569 if (signal_pending(current
)) {
2574 mutex_lock(&memcg_limit_mutex
);
2575 if (limit
< memcg
->memory
.limit
) {
2576 mutex_unlock(&memcg_limit_mutex
);
2580 if (limit
> memcg
->memsw
.limit
)
2582 ret
= page_counter_limit(&memcg
->memsw
, limit
);
2583 mutex_unlock(&memcg_limit_mutex
);
2588 try_to_free_mem_cgroup_pages(memcg
, 1, GFP_KERNEL
, false);
2590 curusage
= page_counter_read(&memcg
->memsw
);
2591 /* Usage is reduced ? */
2592 if (curusage
>= oldusage
)
2595 oldusage
= curusage
;
2596 } while (retry_count
);
2598 if (!ret
&& enlarge
)
2599 memcg_oom_recover(memcg
);
2604 unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t
*pgdat
, int order
,
2606 unsigned long *total_scanned
)
2608 unsigned long nr_reclaimed
= 0;
2609 struct mem_cgroup_per_node
*mz
, *next_mz
= NULL
;
2610 unsigned long reclaimed
;
2612 struct mem_cgroup_tree_per_node
*mctz
;
2613 unsigned long excess
;
2614 unsigned long nr_scanned
;
2619 mctz
= soft_limit_tree_node(pgdat
->node_id
);
2622 * Do not even bother to check the largest node if the root
2623 * is empty. Do it lockless to prevent lock bouncing. Races
2624 * are acceptable as soft limit is best effort anyway.
2626 if (!mctz
|| RB_EMPTY_ROOT(&mctz
->rb_root
))
2630 * This loop can run a while, specially if mem_cgroup's continuously
2631 * keep exceeding their soft limit and putting the system under
2638 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
2643 reclaimed
= mem_cgroup_soft_reclaim(mz
->memcg
, pgdat
,
2644 gfp_mask
, &nr_scanned
);
2645 nr_reclaimed
+= reclaimed
;
2646 *total_scanned
+= nr_scanned
;
2647 spin_lock_irq(&mctz
->lock
);
2648 __mem_cgroup_remove_exceeded(mz
, mctz
);
2651 * If we failed to reclaim anything from this memory cgroup
2652 * it is time to move on to the next cgroup
2656 next_mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
2658 excess
= soft_limit_excess(mz
->memcg
);
2660 * One school of thought says that we should not add
2661 * back the node to the tree if reclaim returns 0.
2662 * But our reclaim could return 0, simply because due
2663 * to priority we are exposing a smaller subset of
2664 * memory to reclaim from. Consider this as a longer
2667 /* If excess == 0, no tree ops */
2668 __mem_cgroup_insert_exceeded(mz
, mctz
, excess
);
2669 spin_unlock_irq(&mctz
->lock
);
2670 css_put(&mz
->memcg
->css
);
2673 * Could not reclaim anything and there are no more
2674 * mem cgroups to try or we seem to be looping without
2675 * reclaiming anything.
2677 if (!nr_reclaimed
&&
2679 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
2681 } while (!nr_reclaimed
);
2683 css_put(&next_mz
->memcg
->css
);
2684 return nr_reclaimed
;
2688 * Test whether @memcg has children, dead or alive. Note that this
2689 * function doesn't care whether @memcg has use_hierarchy enabled and
2690 * returns %true if there are child csses according to the cgroup
2691 * hierarchy. Testing use_hierarchy is the caller's responsiblity.
2693 static inline bool memcg_has_children(struct mem_cgroup
*memcg
)
2698 ret
= css_next_child(NULL
, &memcg
->css
);
2704 * Reclaims as many pages from the given memcg as possible.
2706 * Caller is responsible for holding css reference for memcg.
2708 static int mem_cgroup_force_empty(struct mem_cgroup
*memcg
)
2710 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2712 /* we call try-to-free pages for make this cgroup empty */
2713 lru_add_drain_all();
2714 /* try to free all pages in this cgroup */
2715 while (nr_retries
&& page_counter_read(&memcg
->memory
)) {
2718 if (signal_pending(current
))
2721 progress
= try_to_free_mem_cgroup_pages(memcg
, 1,
2725 /* maybe some writeback is necessary */
2726 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
2734 static ssize_t
mem_cgroup_force_empty_write(struct kernfs_open_file
*of
,
2735 char *buf
, size_t nbytes
,
2738 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
2740 if (mem_cgroup_is_root(memcg
))
2742 return mem_cgroup_force_empty(memcg
) ?: nbytes
;
2745 static u64
mem_cgroup_hierarchy_read(struct cgroup_subsys_state
*css
,
2748 return mem_cgroup_from_css(css
)->use_hierarchy
;
2751 static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state
*css
,
2752 struct cftype
*cft
, u64 val
)
2755 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
2756 struct mem_cgroup
*parent_memcg
= mem_cgroup_from_css(memcg
->css
.parent
);
2758 if (memcg
->use_hierarchy
== val
)
2762 * If parent's use_hierarchy is set, we can't make any modifications
2763 * in the child subtrees. If it is unset, then the change can
2764 * occur, provided the current cgroup has no children.
2766 * For the root cgroup, parent_mem is NULL, we allow value to be
2767 * set if there are no children.
2769 if ((!parent_memcg
|| !parent_memcg
->use_hierarchy
) &&
2770 (val
== 1 || val
== 0)) {
2771 if (!memcg_has_children(memcg
))
2772 memcg
->use_hierarchy
= val
;
2781 static void tree_stat(struct mem_cgroup
*memcg
, unsigned long *stat
)
2783 struct mem_cgroup
*iter
;
2786 memset(stat
, 0, sizeof(*stat
) * MEMCG_NR_STAT
);
2788 for_each_mem_cgroup_tree(iter
, memcg
) {
2789 for (i
= 0; i
< MEMCG_NR_STAT
; i
++)
2790 stat
[i
] += memcg_page_state(iter
, i
);
2794 static void tree_events(struct mem_cgroup
*memcg
, unsigned long *events
)
2796 struct mem_cgroup
*iter
;
2799 memset(events
, 0, sizeof(*events
) * MEMCG_NR_EVENTS
);
2801 for_each_mem_cgroup_tree(iter
, memcg
) {
2802 for (i
= 0; i
< MEMCG_NR_EVENTS
; i
++)
2803 events
[i
] += memcg_sum_events(iter
, i
);
2807 static unsigned long mem_cgroup_usage(struct mem_cgroup
*memcg
, bool swap
)
2809 unsigned long val
= 0;
2811 if (mem_cgroup_is_root(memcg
)) {
2812 struct mem_cgroup
*iter
;
2814 for_each_mem_cgroup_tree(iter
, memcg
) {
2815 val
+= memcg_page_state(iter
, MEMCG_CACHE
);
2816 val
+= memcg_page_state(iter
, MEMCG_RSS
);
2818 val
+= memcg_page_state(iter
, MEMCG_SWAP
);
2822 val
= page_counter_read(&memcg
->memory
);
2824 val
= page_counter_read(&memcg
->memsw
);
2837 static u64
mem_cgroup_read_u64(struct cgroup_subsys_state
*css
,
2840 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
2841 struct page_counter
*counter
;
2843 switch (MEMFILE_TYPE(cft
->private)) {
2845 counter
= &memcg
->memory
;
2848 counter
= &memcg
->memsw
;
2851 counter
= &memcg
->kmem
;
2854 counter
= &memcg
->tcpmem
;
2860 switch (MEMFILE_ATTR(cft
->private)) {
2862 if (counter
== &memcg
->memory
)
2863 return (u64
)mem_cgroup_usage(memcg
, false) * PAGE_SIZE
;
2864 if (counter
== &memcg
->memsw
)
2865 return (u64
)mem_cgroup_usage(memcg
, true) * PAGE_SIZE
;
2866 return (u64
)page_counter_read(counter
) * PAGE_SIZE
;
2868 return (u64
)counter
->limit
* PAGE_SIZE
;
2870 return (u64
)counter
->watermark
* PAGE_SIZE
;
2872 return counter
->failcnt
;
2873 case RES_SOFT_LIMIT
:
2874 return (u64
)memcg
->soft_limit
* PAGE_SIZE
;
2881 static int memcg_online_kmem(struct mem_cgroup
*memcg
)
2885 if (cgroup_memory_nokmem
)
2888 BUG_ON(memcg
->kmemcg_id
>= 0);
2889 BUG_ON(memcg
->kmem_state
);
2891 memcg_id
= memcg_alloc_cache_id();
2895 static_branch_inc(&memcg_kmem_enabled_key
);
2897 * A memory cgroup is considered kmem-online as soon as it gets
2898 * kmemcg_id. Setting the id after enabling static branching will
2899 * guarantee no one starts accounting before all call sites are
2902 memcg
->kmemcg_id
= memcg_id
;
2903 memcg
->kmem_state
= KMEM_ONLINE
;
2904 INIT_LIST_HEAD(&memcg
->kmem_caches
);
2909 static void memcg_offline_kmem(struct mem_cgroup
*memcg
)
2911 struct cgroup_subsys_state
*css
;
2912 struct mem_cgroup
*parent
, *child
;
2915 if (memcg
->kmem_state
!= KMEM_ONLINE
)
2918 * Clear the online state before clearing memcg_caches array
2919 * entries. The slab_mutex in memcg_deactivate_kmem_caches()
2920 * guarantees that no cache will be created for this cgroup
2921 * after we are done (see memcg_create_kmem_cache()).
2923 memcg
->kmem_state
= KMEM_ALLOCATED
;
2925 memcg_deactivate_kmem_caches(memcg
);
2927 kmemcg_id
= memcg
->kmemcg_id
;
2928 BUG_ON(kmemcg_id
< 0);
2930 parent
= parent_mem_cgroup(memcg
);
2932 parent
= root_mem_cgroup
;
2935 * Change kmemcg_id of this cgroup and all its descendants to the
2936 * parent's id, and then move all entries from this cgroup's list_lrus
2937 * to ones of the parent. After we have finished, all list_lrus
2938 * corresponding to this cgroup are guaranteed to remain empty. The
2939 * ordering is imposed by list_lru_node->lock taken by
2940 * memcg_drain_all_list_lrus().
2942 rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */
2943 css_for_each_descendant_pre(css
, &memcg
->css
) {
2944 child
= mem_cgroup_from_css(css
);
2945 BUG_ON(child
->kmemcg_id
!= kmemcg_id
);
2946 child
->kmemcg_id
= parent
->kmemcg_id
;
2947 if (!memcg
->use_hierarchy
)
2952 memcg_drain_all_list_lrus(kmemcg_id
, parent
->kmemcg_id
);
2954 memcg_free_cache_id(kmemcg_id
);
2957 static void memcg_free_kmem(struct mem_cgroup
*memcg
)
2959 /* css_alloc() failed, offlining didn't happen */
2960 if (unlikely(memcg
->kmem_state
== KMEM_ONLINE
))
2961 memcg_offline_kmem(memcg
);
2963 if (memcg
->kmem_state
== KMEM_ALLOCATED
) {
2964 memcg_destroy_kmem_caches(memcg
);
2965 static_branch_dec(&memcg_kmem_enabled_key
);
2966 WARN_ON(page_counter_read(&memcg
->kmem
));
2970 static int memcg_online_kmem(struct mem_cgroup
*memcg
)
2974 static void memcg_offline_kmem(struct mem_cgroup
*memcg
)
2977 static void memcg_free_kmem(struct mem_cgroup
*memcg
)
2980 #endif /* !CONFIG_SLOB */
2982 static int memcg_update_kmem_limit(struct mem_cgroup
*memcg
,
2983 unsigned long limit
)
2987 mutex_lock(&memcg_limit_mutex
);
2988 ret
= page_counter_limit(&memcg
->kmem
, limit
);
2989 mutex_unlock(&memcg_limit_mutex
);
2993 static int memcg_update_tcp_limit(struct mem_cgroup
*memcg
, unsigned long limit
)
2997 mutex_lock(&memcg_limit_mutex
);
2999 ret
= page_counter_limit(&memcg
->tcpmem
, limit
);
3003 if (!memcg
->tcpmem_active
) {
3005 * The active flag needs to be written after the static_key
3006 * update. This is what guarantees that the socket activation
3007 * function is the last one to run. See mem_cgroup_sk_alloc()
3008 * for details, and note that we don't mark any socket as
3009 * belonging to this memcg until that flag is up.
3011 * We need to do this, because static_keys will span multiple
3012 * sites, but we can't control their order. If we mark a socket
3013 * as accounted, but the accounting functions are not patched in
3014 * yet, we'll lose accounting.
3016 * We never race with the readers in mem_cgroup_sk_alloc(),
3017 * because when this value change, the code to process it is not
3020 static_branch_inc(&memcg_sockets_enabled_key
);
3021 memcg
->tcpmem_active
= true;
3024 mutex_unlock(&memcg_limit_mutex
);
3029 * The user of this function is...
3032 static ssize_t
mem_cgroup_write(struct kernfs_open_file
*of
,
3033 char *buf
, size_t nbytes
, loff_t off
)
3035 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
3036 unsigned long nr_pages
;
3039 buf
= strstrip(buf
);
3040 ret
= page_counter_memparse(buf
, "-1", &nr_pages
);
3044 switch (MEMFILE_ATTR(of_cft(of
)->private)) {
3046 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
3050 switch (MEMFILE_TYPE(of_cft(of
)->private)) {
3052 ret
= mem_cgroup_resize_limit(memcg
, nr_pages
);
3055 ret
= mem_cgroup_resize_memsw_limit(memcg
, nr_pages
);
3058 ret
= memcg_update_kmem_limit(memcg
, nr_pages
);
3061 ret
= memcg_update_tcp_limit(memcg
, nr_pages
);
3065 case RES_SOFT_LIMIT
:
3066 memcg
->soft_limit
= nr_pages
;
3070 return ret
?: nbytes
;
3073 static ssize_t
mem_cgroup_reset(struct kernfs_open_file
*of
, char *buf
,
3074 size_t nbytes
, loff_t off
)
3076 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
3077 struct page_counter
*counter
;
3079 switch (MEMFILE_TYPE(of_cft(of
)->private)) {
3081 counter
= &memcg
->memory
;
3084 counter
= &memcg
->memsw
;
3087 counter
= &memcg
->kmem
;
3090 counter
= &memcg
->tcpmem
;
3096 switch (MEMFILE_ATTR(of_cft(of
)->private)) {
3098 page_counter_reset_watermark(counter
);
3101 counter
->failcnt
= 0;
3110 static u64
mem_cgroup_move_charge_read(struct cgroup_subsys_state
*css
,
3113 return mem_cgroup_from_css(css
)->move_charge_at_immigrate
;
3117 static int mem_cgroup_move_charge_write(struct cgroup_subsys_state
*css
,
3118 struct cftype
*cft
, u64 val
)
3120 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
3122 if (val
& ~MOVE_MASK
)
3126 * No kind of locking is needed in here, because ->can_attach() will
3127 * check this value once in the beginning of the process, and then carry
3128 * on with stale data. This means that changes to this value will only
3129 * affect task migrations starting after the change.
3131 memcg
->move_charge_at_immigrate
= val
;
3135 static int mem_cgroup_move_charge_write(struct cgroup_subsys_state
*css
,
3136 struct cftype
*cft
, u64 val
)
3143 static int memcg_numa_stat_show(struct seq_file
*m
, void *v
)
3147 unsigned int lru_mask
;
3150 static const struct numa_stat stats
[] = {
3151 { "total", LRU_ALL
},
3152 { "file", LRU_ALL_FILE
},
3153 { "anon", LRU_ALL_ANON
},
3154 { "unevictable", BIT(LRU_UNEVICTABLE
) },
3156 const struct numa_stat
*stat
;
3159 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
3161 for (stat
= stats
; stat
< stats
+ ARRAY_SIZE(stats
); stat
++) {
3162 nr
= mem_cgroup_nr_lru_pages(memcg
, stat
->lru_mask
);
3163 seq_printf(m
, "%s=%lu", stat
->name
, nr
);
3164 for_each_node_state(nid
, N_MEMORY
) {
3165 nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
3167 seq_printf(m
, " N%d=%lu", nid
, nr
);
3172 for (stat
= stats
; stat
< stats
+ ARRAY_SIZE(stats
); stat
++) {
3173 struct mem_cgroup
*iter
;
3176 for_each_mem_cgroup_tree(iter
, memcg
)
3177 nr
+= mem_cgroup_nr_lru_pages(iter
, stat
->lru_mask
);
3178 seq_printf(m
, "hierarchical_%s=%lu", stat
->name
, nr
);
3179 for_each_node_state(nid
, N_MEMORY
) {
3181 for_each_mem_cgroup_tree(iter
, memcg
)
3182 nr
+= mem_cgroup_node_nr_lru_pages(
3183 iter
, nid
, stat
->lru_mask
);
3184 seq_printf(m
, " N%d=%lu", nid
, nr
);
3191 #endif /* CONFIG_NUMA */
3193 /* Universal VM events cgroup1 shows, original sort order */
3194 unsigned int memcg1_events
[] = {
3201 static const char *const memcg1_event_names
[] = {
3208 static int memcg_stat_show(struct seq_file
*m
, void *v
)
3210 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
3211 unsigned long memory
, memsw
;
3212 struct mem_cgroup
*mi
;
3215 BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names
) != ARRAY_SIZE(memcg1_stats
));
3216 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names
) != NR_LRU_LISTS
);
3218 for (i
= 0; i
< ARRAY_SIZE(memcg1_stats
); i
++) {
3219 if (memcg1_stats
[i
] == MEMCG_SWAP
&& !do_memsw_account())
3221 seq_printf(m
, "%s %lu\n", memcg1_stat_names
[i
],
3222 memcg_page_state(memcg
, memcg1_stats
[i
]) *
3226 for (i
= 0; i
< ARRAY_SIZE(memcg1_events
); i
++)
3227 seq_printf(m
, "%s %lu\n", memcg1_event_names
[i
],
3228 memcg_sum_events(memcg
, memcg1_events
[i
]));
3230 for (i
= 0; i
< NR_LRU_LISTS
; i
++)
3231 seq_printf(m
, "%s %lu\n", mem_cgroup_lru_names
[i
],
3232 mem_cgroup_nr_lru_pages(memcg
, BIT(i
)) * PAGE_SIZE
);
3234 /* Hierarchical information */
3235 memory
= memsw
= PAGE_COUNTER_MAX
;
3236 for (mi
= memcg
; mi
; mi
= parent_mem_cgroup(mi
)) {
3237 memory
= min(memory
, mi
->memory
.limit
);
3238 memsw
= min(memsw
, mi
->memsw
.limit
);
3240 seq_printf(m
, "hierarchical_memory_limit %llu\n",
3241 (u64
)memory
* PAGE_SIZE
);
3242 if (do_memsw_account())
3243 seq_printf(m
, "hierarchical_memsw_limit %llu\n",
3244 (u64
)memsw
* PAGE_SIZE
);
3246 for (i
= 0; i
< ARRAY_SIZE(memcg1_stats
); i
++) {
3247 unsigned long long val
= 0;
3249 if (memcg1_stats
[i
] == MEMCG_SWAP
&& !do_memsw_account())
3251 for_each_mem_cgroup_tree(mi
, memcg
)
3252 val
+= memcg_page_state(mi
, memcg1_stats
[i
]) *
3254 seq_printf(m
, "total_%s %llu\n", memcg1_stat_names
[i
], val
);
3257 for (i
= 0; i
< ARRAY_SIZE(memcg1_events
); i
++) {
3258 unsigned long long val
= 0;
3260 for_each_mem_cgroup_tree(mi
, memcg
)
3261 val
+= memcg_sum_events(mi
, memcg1_events
[i
]);
3262 seq_printf(m
, "total_%s %llu\n", memcg1_event_names
[i
], val
);
3265 for (i
= 0; i
< NR_LRU_LISTS
; i
++) {
3266 unsigned long long val
= 0;
3268 for_each_mem_cgroup_tree(mi
, memcg
)
3269 val
+= mem_cgroup_nr_lru_pages(mi
, BIT(i
)) * PAGE_SIZE
;
3270 seq_printf(m
, "total_%s %llu\n", mem_cgroup_lru_names
[i
], val
);
3273 #ifdef CONFIG_DEBUG_VM
3276 struct mem_cgroup_per_node
*mz
;
3277 struct zone_reclaim_stat
*rstat
;
3278 unsigned long recent_rotated
[2] = {0, 0};
3279 unsigned long recent_scanned
[2] = {0, 0};
3281 for_each_online_pgdat(pgdat
) {
3282 mz
= mem_cgroup_nodeinfo(memcg
, pgdat
->node_id
);
3283 rstat
= &mz
->lruvec
.reclaim_stat
;
3285 recent_rotated
[0] += rstat
->recent_rotated
[0];
3286 recent_rotated
[1] += rstat
->recent_rotated
[1];
3287 recent_scanned
[0] += rstat
->recent_scanned
[0];
3288 recent_scanned
[1] += rstat
->recent_scanned
[1];
3290 seq_printf(m
, "recent_rotated_anon %lu\n", recent_rotated
[0]);
3291 seq_printf(m
, "recent_rotated_file %lu\n", recent_rotated
[1]);
3292 seq_printf(m
, "recent_scanned_anon %lu\n", recent_scanned
[0]);
3293 seq_printf(m
, "recent_scanned_file %lu\n", recent_scanned
[1]);
3300 static u64
mem_cgroup_swappiness_read(struct cgroup_subsys_state
*css
,
3303 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
3305 return mem_cgroup_swappiness(memcg
);
3308 static int mem_cgroup_swappiness_write(struct cgroup_subsys_state
*css
,
3309 struct cftype
*cft
, u64 val
)
3311 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
3317 memcg
->swappiness
= val
;
3319 vm_swappiness
= val
;
3324 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
3326 struct mem_cgroup_threshold_ary
*t
;
3327 unsigned long usage
;
3332 t
= rcu_dereference(memcg
->thresholds
.primary
);
3334 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
3339 usage
= mem_cgroup_usage(memcg
, swap
);
3342 * current_threshold points to threshold just below or equal to usage.
3343 * If it's not true, a threshold was crossed after last
3344 * call of __mem_cgroup_threshold().
3346 i
= t
->current_threshold
;
3349 * Iterate backward over array of thresholds starting from
3350 * current_threshold and check if a threshold is crossed.
3351 * If none of thresholds below usage is crossed, we read
3352 * only one element of the array here.
3354 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
3355 eventfd_signal(t
->entries
[i
].eventfd
, 1);
3357 /* i = current_threshold + 1 */
3361 * Iterate forward over array of thresholds starting from
3362 * current_threshold+1 and check if a threshold is crossed.
3363 * If none of thresholds above usage is crossed, we read
3364 * only one element of the array here.
3366 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
3367 eventfd_signal(t
->entries
[i
].eventfd
, 1);
3369 /* Update current_threshold */
3370 t
->current_threshold
= i
- 1;
3375 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
3378 __mem_cgroup_threshold(memcg
, false);
3379 if (do_memsw_account())
3380 __mem_cgroup_threshold(memcg
, true);
3382 memcg
= parent_mem_cgroup(memcg
);
3386 static int compare_thresholds(const void *a
, const void *b
)
3388 const struct mem_cgroup_threshold
*_a
= a
;
3389 const struct mem_cgroup_threshold
*_b
= b
;
3391 if (_a
->threshold
> _b
->threshold
)
3394 if (_a
->threshold
< _b
->threshold
)
3400 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*memcg
)
3402 struct mem_cgroup_eventfd_list
*ev
;
3404 spin_lock(&memcg_oom_lock
);
3406 list_for_each_entry(ev
, &memcg
->oom_notify
, list
)
3407 eventfd_signal(ev
->eventfd
, 1);
3409 spin_unlock(&memcg_oom_lock
);
3413 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
)
3415 struct mem_cgroup
*iter
;
3417 for_each_mem_cgroup_tree(iter
, memcg
)
3418 mem_cgroup_oom_notify_cb(iter
);
3421 static int __mem_cgroup_usage_register_event(struct mem_cgroup
*memcg
,
3422 struct eventfd_ctx
*eventfd
, const char *args
, enum res_type type
)
3424 struct mem_cgroup_thresholds
*thresholds
;
3425 struct mem_cgroup_threshold_ary
*new;
3426 unsigned long threshold
;
3427 unsigned long usage
;
3430 ret
= page_counter_memparse(args
, "-1", &threshold
);
3434 mutex_lock(&memcg
->thresholds_lock
);
3437 thresholds
= &memcg
->thresholds
;
3438 usage
= mem_cgroup_usage(memcg
, false);
3439 } else if (type
== _MEMSWAP
) {
3440 thresholds
= &memcg
->memsw_thresholds
;
3441 usage
= mem_cgroup_usage(memcg
, true);
3445 /* Check if a threshold crossed before adding a new one */
3446 if (thresholds
->primary
)
3447 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
3449 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
3451 /* Allocate memory for new array of thresholds */
3452 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
3460 /* Copy thresholds (if any) to new array */
3461 if (thresholds
->primary
) {
3462 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
3463 sizeof(struct mem_cgroup_threshold
));
3466 /* Add new threshold */
3467 new->entries
[size
- 1].eventfd
= eventfd
;
3468 new->entries
[size
- 1].threshold
= threshold
;
3470 /* Sort thresholds. Registering of new threshold isn't time-critical */
3471 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
3472 compare_thresholds
, NULL
);
3474 /* Find current threshold */
3475 new->current_threshold
= -1;
3476 for (i
= 0; i
< size
; i
++) {
3477 if (new->entries
[i
].threshold
<= usage
) {
3479 * new->current_threshold will not be used until
3480 * rcu_assign_pointer(), so it's safe to increment
3483 ++new->current_threshold
;
3488 /* Free old spare buffer and save old primary buffer as spare */
3489 kfree(thresholds
->spare
);
3490 thresholds
->spare
= thresholds
->primary
;
3492 rcu_assign_pointer(thresholds
->primary
, new);
3494 /* To be sure that nobody uses thresholds */
3498 mutex_unlock(&memcg
->thresholds_lock
);
3503 static int mem_cgroup_usage_register_event(struct mem_cgroup
*memcg
,
3504 struct eventfd_ctx
*eventfd
, const char *args
)
3506 return __mem_cgroup_usage_register_event(memcg
, eventfd
, args
, _MEM
);
3509 static int memsw_cgroup_usage_register_event(struct mem_cgroup
*memcg
,
3510 struct eventfd_ctx
*eventfd
, const char *args
)
3512 return __mem_cgroup_usage_register_event(memcg
, eventfd
, args
, _MEMSWAP
);
3515 static void __mem_cgroup_usage_unregister_event(struct mem_cgroup
*memcg
,
3516 struct eventfd_ctx
*eventfd
, enum res_type type
)
3518 struct mem_cgroup_thresholds
*thresholds
;
3519 struct mem_cgroup_threshold_ary
*new;
3520 unsigned long usage
;
3523 mutex_lock(&memcg
->thresholds_lock
);
3526 thresholds
= &memcg
->thresholds
;
3527 usage
= mem_cgroup_usage(memcg
, false);
3528 } else if (type
== _MEMSWAP
) {
3529 thresholds
= &memcg
->memsw_thresholds
;
3530 usage
= mem_cgroup_usage(memcg
, true);
3534 if (!thresholds
->primary
)
3537 /* Check if a threshold crossed before removing */
3538 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
3540 /* Calculate new number of threshold */
3542 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
3543 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
3547 new = thresholds
->spare
;
3549 /* Set thresholds array to NULL if we don't have thresholds */
3558 /* Copy thresholds and find current threshold */
3559 new->current_threshold
= -1;
3560 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
3561 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
3564 new->entries
[j
] = thresholds
->primary
->entries
[i
];
3565 if (new->entries
[j
].threshold
<= usage
) {
3567 * new->current_threshold will not be used
3568 * until rcu_assign_pointer(), so it's safe to increment
3571 ++new->current_threshold
;
3577 /* Swap primary and spare array */
3578 thresholds
->spare
= thresholds
->primary
;
3580 rcu_assign_pointer(thresholds
->primary
, new);
3582 /* To be sure that nobody uses thresholds */
3585 /* If all events are unregistered, free the spare array */
3587 kfree(thresholds
->spare
);
3588 thresholds
->spare
= NULL
;
3591 mutex_unlock(&memcg
->thresholds_lock
);
3594 static void mem_cgroup_usage_unregister_event(struct mem_cgroup
*memcg
,
3595 struct eventfd_ctx
*eventfd
)
3597 return __mem_cgroup_usage_unregister_event(memcg
, eventfd
, _MEM
);
3600 static void memsw_cgroup_usage_unregister_event(struct mem_cgroup
*memcg
,
3601 struct eventfd_ctx
*eventfd
)
3603 return __mem_cgroup_usage_unregister_event(memcg
, eventfd
, _MEMSWAP
);
3606 static int mem_cgroup_oom_register_event(struct mem_cgroup
*memcg
,
3607 struct eventfd_ctx
*eventfd
, const char *args
)
3609 struct mem_cgroup_eventfd_list
*event
;
3611 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
3615 spin_lock(&memcg_oom_lock
);
3617 event
->eventfd
= eventfd
;
3618 list_add(&event
->list
, &memcg
->oom_notify
);
3620 /* already in OOM ? */
3621 if (memcg
->under_oom
)
3622 eventfd_signal(eventfd
, 1);
3623 spin_unlock(&memcg_oom_lock
);
3628 static void mem_cgroup_oom_unregister_event(struct mem_cgroup
*memcg
,
3629 struct eventfd_ctx
*eventfd
)
3631 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
3633 spin_lock(&memcg_oom_lock
);
3635 list_for_each_entry_safe(ev
, tmp
, &memcg
->oom_notify
, list
) {
3636 if (ev
->eventfd
== eventfd
) {
3637 list_del(&ev
->list
);
3642 spin_unlock(&memcg_oom_lock
);
3645 static int mem_cgroup_oom_control_read(struct seq_file
*sf
, void *v
)
3647 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(sf
));
3649 seq_printf(sf
, "oom_kill_disable %d\n", memcg
->oom_kill_disable
);
3650 seq_printf(sf
, "under_oom %d\n", (bool)memcg
->under_oom
);
3651 seq_printf(sf
, "oom_kill %lu\n", memcg_sum_events(memcg
, OOM_KILL
));
3655 static int mem_cgroup_oom_control_write(struct cgroup_subsys_state
*css
,
3656 struct cftype
*cft
, u64 val
)
3658 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
3660 /* cannot set to root cgroup and only 0 and 1 are allowed */
3661 if (!css
->parent
|| !((val
== 0) || (val
== 1)))
3664 memcg
->oom_kill_disable
= val
;
3666 memcg_oom_recover(memcg
);
3671 #ifdef CONFIG_CGROUP_WRITEBACK
3673 struct list_head
*mem_cgroup_cgwb_list(struct mem_cgroup
*memcg
)
3675 return &memcg
->cgwb_list
;
3678 static int memcg_wb_domain_init(struct mem_cgroup
*memcg
, gfp_t gfp
)
3680 return wb_domain_init(&memcg
->cgwb_domain
, gfp
);
3683 static void memcg_wb_domain_exit(struct mem_cgroup
*memcg
)
3685 wb_domain_exit(&memcg
->cgwb_domain
);
3688 static void memcg_wb_domain_size_changed(struct mem_cgroup
*memcg
)
3690 wb_domain_size_changed(&memcg
->cgwb_domain
);
3693 struct wb_domain
*mem_cgroup_wb_domain(struct bdi_writeback
*wb
)
3695 struct mem_cgroup
*memcg
= mem_cgroup_from_css(wb
->memcg_css
);
3697 if (!memcg
->css
.parent
)
3700 return &memcg
->cgwb_domain
;
3704 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
3705 * @wb: bdi_writeback in question
3706 * @pfilepages: out parameter for number of file pages
3707 * @pheadroom: out parameter for number of allocatable pages according to memcg
3708 * @pdirty: out parameter for number of dirty pages
3709 * @pwriteback: out parameter for number of pages under writeback
3711 * Determine the numbers of file, headroom, dirty, and writeback pages in
3712 * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom
3713 * is a bit more involved.
3715 * A memcg's headroom is "min(max, high) - used". In the hierarchy, the
3716 * headroom is calculated as the lowest headroom of itself and the
3717 * ancestors. Note that this doesn't consider the actual amount of
3718 * available memory in the system. The caller should further cap
3719 * *@pheadroom accordingly.
3721 void mem_cgroup_wb_stats(struct bdi_writeback
*wb
, unsigned long *pfilepages
,
3722 unsigned long *pheadroom
, unsigned long *pdirty
,
3723 unsigned long *pwriteback
)
3725 struct mem_cgroup
*memcg
= mem_cgroup_from_css(wb
->memcg_css
);
3726 struct mem_cgroup
*parent
;
3728 *pdirty
= memcg_page_state(memcg
, NR_FILE_DIRTY
);
3730 /* this should eventually include NR_UNSTABLE_NFS */
3731 *pwriteback
= memcg_page_state(memcg
, NR_WRITEBACK
);
3732 *pfilepages
= mem_cgroup_nr_lru_pages(memcg
, (1 << LRU_INACTIVE_FILE
) |
3733 (1 << LRU_ACTIVE_FILE
));
3734 *pheadroom
= PAGE_COUNTER_MAX
;
3736 while ((parent
= parent_mem_cgroup(memcg
))) {
3737 unsigned long ceiling
= min(memcg
->memory
.limit
, memcg
->high
);
3738 unsigned long used
= page_counter_read(&memcg
->memory
);
3740 *pheadroom
= min(*pheadroom
, ceiling
- min(ceiling
, used
));
3745 #else /* CONFIG_CGROUP_WRITEBACK */
3747 static int memcg_wb_domain_init(struct mem_cgroup
*memcg
, gfp_t gfp
)
3752 static void memcg_wb_domain_exit(struct mem_cgroup
*memcg
)
3756 static void memcg_wb_domain_size_changed(struct mem_cgroup
*memcg
)
3760 #endif /* CONFIG_CGROUP_WRITEBACK */
3763 * DO NOT USE IN NEW FILES.
3765 * "cgroup.event_control" implementation.
3767 * This is way over-engineered. It tries to support fully configurable
3768 * events for each user. Such level of flexibility is completely
3769 * unnecessary especially in the light of the planned unified hierarchy.
3771 * Please deprecate this and replace with something simpler if at all
3776 * Unregister event and free resources.
3778 * Gets called from workqueue.
3780 static void memcg_event_remove(struct work_struct
*work
)
3782 struct mem_cgroup_event
*event
=
3783 container_of(work
, struct mem_cgroup_event
, remove
);
3784 struct mem_cgroup
*memcg
= event
->memcg
;
3786 remove_wait_queue(event
->wqh
, &event
->wait
);
3788 event
->unregister_event(memcg
, event
->eventfd
);
3790 /* Notify userspace the event is going away. */
3791 eventfd_signal(event
->eventfd
, 1);
3793 eventfd_ctx_put(event
->eventfd
);
3795 css_put(&memcg
->css
);
3799 * Gets called on POLLHUP on eventfd when user closes it.
3801 * Called with wqh->lock held and interrupts disabled.
3803 static int memcg_event_wake(wait_queue_entry_t
*wait
, unsigned mode
,
3804 int sync
, void *key
)
3806 struct mem_cgroup_event
*event
=
3807 container_of(wait
, struct mem_cgroup_event
, wait
);
3808 struct mem_cgroup
*memcg
= event
->memcg
;
3809 unsigned long flags
= (unsigned long)key
;
3811 if (flags
& POLLHUP
) {
3813 * If the event has been detached at cgroup removal, we
3814 * can simply return knowing the other side will cleanup
3817 * We can't race against event freeing since the other
3818 * side will require wqh->lock via remove_wait_queue(),
3821 spin_lock(&memcg
->event_list_lock
);
3822 if (!list_empty(&event
->list
)) {
3823 list_del_init(&event
->list
);
3825 * We are in atomic context, but cgroup_event_remove()
3826 * may sleep, so we have to call it in workqueue.
3828 schedule_work(&event
->remove
);
3830 spin_unlock(&memcg
->event_list_lock
);
3836 static void memcg_event_ptable_queue_proc(struct file
*file
,
3837 wait_queue_head_t
*wqh
, poll_table
*pt
)
3839 struct mem_cgroup_event
*event
=
3840 container_of(pt
, struct mem_cgroup_event
, pt
);
3843 add_wait_queue(wqh
, &event
->wait
);
3847 * DO NOT USE IN NEW FILES.
3849 * Parse input and register new cgroup event handler.
3851 * Input must be in format '<event_fd> <control_fd> <args>'.
3852 * Interpretation of args is defined by control file implementation.
3854 static ssize_t
memcg_write_event_control(struct kernfs_open_file
*of
,
3855 char *buf
, size_t nbytes
, loff_t off
)
3857 struct cgroup_subsys_state
*css
= of_css(of
);
3858 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
3859 struct mem_cgroup_event
*event
;
3860 struct cgroup_subsys_state
*cfile_css
;
3861 unsigned int efd
, cfd
;
3868 buf
= strstrip(buf
);
3870 efd
= simple_strtoul(buf
, &endp
, 10);
3875 cfd
= simple_strtoul(buf
, &endp
, 10);
3876 if ((*endp
!= ' ') && (*endp
!= '\0'))
3880 event
= kzalloc(sizeof(*event
), GFP_KERNEL
);
3884 event
->memcg
= memcg
;
3885 INIT_LIST_HEAD(&event
->list
);
3886 init_poll_funcptr(&event
->pt
, memcg_event_ptable_queue_proc
);
3887 init_waitqueue_func_entry(&event
->wait
, memcg_event_wake
);
3888 INIT_WORK(&event
->remove
, memcg_event_remove
);
3896 event
->eventfd
= eventfd_ctx_fileget(efile
.file
);
3897 if (IS_ERR(event
->eventfd
)) {
3898 ret
= PTR_ERR(event
->eventfd
);
3905 goto out_put_eventfd
;
3908 /* the process need read permission on control file */
3909 /* AV: shouldn't we check that it's been opened for read instead? */
3910 ret
= inode_permission(file_inode(cfile
.file
), MAY_READ
);
3915 * Determine the event callbacks and set them in @event. This used
3916 * to be done via struct cftype but cgroup core no longer knows
3917 * about these events. The following is crude but the whole thing
3918 * is for compatibility anyway.
3920 * DO NOT ADD NEW FILES.
3922 name
= cfile
.file
->f_path
.dentry
->d_name
.name
;
3924 if (!strcmp(name
, "memory.usage_in_bytes")) {
3925 event
->register_event
= mem_cgroup_usage_register_event
;
3926 event
->unregister_event
= mem_cgroup_usage_unregister_event
;
3927 } else if (!strcmp(name
, "memory.oom_control")) {
3928 event
->register_event
= mem_cgroup_oom_register_event
;
3929 event
->unregister_event
= mem_cgroup_oom_unregister_event
;
3930 } else if (!strcmp(name
, "memory.pressure_level")) {
3931 event
->register_event
= vmpressure_register_event
;
3932 event
->unregister_event
= vmpressure_unregister_event
;
3933 } else if (!strcmp(name
, "memory.memsw.usage_in_bytes")) {
3934 event
->register_event
= memsw_cgroup_usage_register_event
;
3935 event
->unregister_event
= memsw_cgroup_usage_unregister_event
;
3942 * Verify @cfile should belong to @css. Also, remaining events are
3943 * automatically removed on cgroup destruction but the removal is
3944 * asynchronous, so take an extra ref on @css.
3946 cfile_css
= css_tryget_online_from_dir(cfile
.file
->f_path
.dentry
->d_parent
,
3947 &memory_cgrp_subsys
);
3949 if (IS_ERR(cfile_css
))
3951 if (cfile_css
!= css
) {
3956 ret
= event
->register_event(memcg
, event
->eventfd
, buf
);
3960 efile
.file
->f_op
->poll(efile
.file
, &event
->pt
);
3962 spin_lock(&memcg
->event_list_lock
);
3963 list_add(&event
->list
, &memcg
->event_list
);
3964 spin_unlock(&memcg
->event_list_lock
);
3976 eventfd_ctx_put(event
->eventfd
);
3985 static struct cftype mem_cgroup_legacy_files
[] = {
3987 .name
= "usage_in_bytes",
3988 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
3989 .read_u64
= mem_cgroup_read_u64
,
3992 .name
= "max_usage_in_bytes",
3993 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
3994 .write
= mem_cgroup_reset
,
3995 .read_u64
= mem_cgroup_read_u64
,
3998 .name
= "limit_in_bytes",
3999 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
4000 .write
= mem_cgroup_write
,
4001 .read_u64
= mem_cgroup_read_u64
,
4004 .name
= "soft_limit_in_bytes",
4005 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
4006 .write
= mem_cgroup_write
,
4007 .read_u64
= mem_cgroup_read_u64
,
4011 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
4012 .write
= mem_cgroup_reset
,
4013 .read_u64
= mem_cgroup_read_u64
,
4017 .seq_show
= memcg_stat_show
,
4020 .name
= "force_empty",
4021 .write
= mem_cgroup_force_empty_write
,
4024 .name
= "use_hierarchy",
4025 .write_u64
= mem_cgroup_hierarchy_write
,
4026 .read_u64
= mem_cgroup_hierarchy_read
,
4029 .name
= "cgroup.event_control", /* XXX: for compat */
4030 .write
= memcg_write_event_control
,
4031 .flags
= CFTYPE_NO_PREFIX
| CFTYPE_WORLD_WRITABLE
,
4034 .name
= "swappiness",
4035 .read_u64
= mem_cgroup_swappiness_read
,
4036 .write_u64
= mem_cgroup_swappiness_write
,
4039 .name
= "move_charge_at_immigrate",
4040 .read_u64
= mem_cgroup_move_charge_read
,
4041 .write_u64
= mem_cgroup_move_charge_write
,
4044 .name
= "oom_control",
4045 .seq_show
= mem_cgroup_oom_control_read
,
4046 .write_u64
= mem_cgroup_oom_control_write
,
4047 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
4050 .name
= "pressure_level",
4054 .name
= "numa_stat",
4055 .seq_show
= memcg_numa_stat_show
,
4059 .name
= "kmem.limit_in_bytes",
4060 .private = MEMFILE_PRIVATE(_KMEM
, RES_LIMIT
),
4061 .write
= mem_cgroup_write
,
4062 .read_u64
= mem_cgroup_read_u64
,
4065 .name
= "kmem.usage_in_bytes",
4066 .private = MEMFILE_PRIVATE(_KMEM
, RES_USAGE
),
4067 .read_u64
= mem_cgroup_read_u64
,
4070 .name
= "kmem.failcnt",
4071 .private = MEMFILE_PRIVATE(_KMEM
, RES_FAILCNT
),
4072 .write
= mem_cgroup_reset
,
4073 .read_u64
= mem_cgroup_read_u64
,
4076 .name
= "kmem.max_usage_in_bytes",
4077 .private = MEMFILE_PRIVATE(_KMEM
, RES_MAX_USAGE
),
4078 .write
= mem_cgroup_reset
,
4079 .read_u64
= mem_cgroup_read_u64
,
4081 #ifdef CONFIG_SLABINFO
4083 .name
= "kmem.slabinfo",
4084 .seq_start
= memcg_slab_start
,
4085 .seq_next
= memcg_slab_next
,
4086 .seq_stop
= memcg_slab_stop
,
4087 .seq_show
= memcg_slab_show
,
4091 .name
= "kmem.tcp.limit_in_bytes",
4092 .private = MEMFILE_PRIVATE(_TCP
, RES_LIMIT
),
4093 .write
= mem_cgroup_write
,
4094 .read_u64
= mem_cgroup_read_u64
,
4097 .name
= "kmem.tcp.usage_in_bytes",
4098 .private = MEMFILE_PRIVATE(_TCP
, RES_USAGE
),
4099 .read_u64
= mem_cgroup_read_u64
,
4102 .name
= "kmem.tcp.failcnt",
4103 .private = MEMFILE_PRIVATE(_TCP
, RES_FAILCNT
),
4104 .write
= mem_cgroup_reset
,
4105 .read_u64
= mem_cgroup_read_u64
,
4108 .name
= "kmem.tcp.max_usage_in_bytes",
4109 .private = MEMFILE_PRIVATE(_TCP
, RES_MAX_USAGE
),
4110 .write
= mem_cgroup_reset
,
4111 .read_u64
= mem_cgroup_read_u64
,
4113 { }, /* terminate */
4117 * Private memory cgroup IDR
4119 * Swap-out records and page cache shadow entries need to store memcg
4120 * references in constrained space, so we maintain an ID space that is
4121 * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of
4122 * memory-controlled cgroups to 64k.
4124 * However, there usually are many references to the oflline CSS after
4125 * the cgroup has been destroyed, such as page cache or reclaimable
4126 * slab objects, that don't need to hang on to the ID. We want to keep
4127 * those dead CSS from occupying IDs, or we might quickly exhaust the
4128 * relatively small ID space and prevent the creation of new cgroups
4129 * even when there are much fewer than 64k cgroups - possibly none.
4131 * Maintain a private 16-bit ID space for memcg, and allow the ID to
4132 * be freed and recycled when it's no longer needed, which is usually
4133 * when the CSS is offlined.
4135 * The only exception to that are records of swapped out tmpfs/shmem
4136 * pages that need to be attributed to live ancestors on swapin. But
4137 * those references are manageable from userspace.
4140 static DEFINE_IDR(mem_cgroup_idr
);
4142 static void mem_cgroup_id_remove(struct mem_cgroup
*memcg
)
4144 if (memcg
->id
.id
> 0) {
4145 idr_remove(&mem_cgroup_idr
, memcg
->id
.id
);
4150 static void mem_cgroup_id_get_many(struct mem_cgroup
*memcg
, unsigned int n
)
4152 VM_BUG_ON(atomic_read(&memcg
->id
.ref
) <= 0);
4153 atomic_add(n
, &memcg
->id
.ref
);
4156 static void mem_cgroup_id_put_many(struct mem_cgroup
*memcg
, unsigned int n
)
4158 VM_BUG_ON(atomic_read(&memcg
->id
.ref
) < n
);
4159 if (atomic_sub_and_test(n
, &memcg
->id
.ref
)) {
4160 mem_cgroup_id_remove(memcg
);
4162 /* Memcg ID pins CSS */
4163 css_put(&memcg
->css
);
4167 static inline void mem_cgroup_id_get(struct mem_cgroup
*memcg
)
4169 mem_cgroup_id_get_many(memcg
, 1);
4172 static inline void mem_cgroup_id_put(struct mem_cgroup
*memcg
)
4174 mem_cgroup_id_put_many(memcg
, 1);
4178 * mem_cgroup_from_id - look up a memcg from a memcg id
4179 * @id: the memcg id to look up
4181 * Caller must hold rcu_read_lock().
4183 struct mem_cgroup
*mem_cgroup_from_id(unsigned short id
)
4185 WARN_ON_ONCE(!rcu_read_lock_held());
4186 return idr_find(&mem_cgroup_idr
, id
);
4189 static int alloc_mem_cgroup_per_node_info(struct mem_cgroup
*memcg
, int node
)
4191 struct mem_cgroup_per_node
*pn
;
4194 * This routine is called against possible nodes.
4195 * But it's BUG to call kmalloc() against offline node.
4197 * TODO: this routine can waste much memory for nodes which will
4198 * never be onlined. It's better to use memory hotplug callback
4201 if (!node_state(node
, N_NORMAL_MEMORY
))
4203 pn
= kzalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
4207 pn
->lruvec_stat
= alloc_percpu(struct lruvec_stat
);
4208 if (!pn
->lruvec_stat
) {
4213 lruvec_init(&pn
->lruvec
);
4214 pn
->usage_in_excess
= 0;
4215 pn
->on_tree
= false;
4218 memcg
->nodeinfo
[node
] = pn
;
4222 static void free_mem_cgroup_per_node_info(struct mem_cgroup
*memcg
, int node
)
4224 struct mem_cgroup_per_node
*pn
= memcg
->nodeinfo
[node
];
4229 free_percpu(pn
->lruvec_stat
);
4233 static void __mem_cgroup_free(struct mem_cgroup
*memcg
)
4238 free_mem_cgroup_per_node_info(memcg
, node
);
4239 free_percpu(memcg
->stat
);
4243 static void mem_cgroup_free(struct mem_cgroup
*memcg
)
4245 memcg_wb_domain_exit(memcg
);
4246 __mem_cgroup_free(memcg
);
4249 static struct mem_cgroup
*mem_cgroup_alloc(void)
4251 struct mem_cgroup
*memcg
;
4255 size
= sizeof(struct mem_cgroup
);
4256 size
+= nr_node_ids
* sizeof(struct mem_cgroup_per_node
*);
4258 memcg
= kzalloc(size
, GFP_KERNEL
);
4262 memcg
->id
.id
= idr_alloc(&mem_cgroup_idr
, NULL
,
4263 1, MEM_CGROUP_ID_MAX
,
4265 if (memcg
->id
.id
< 0)
4268 memcg
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4273 if (alloc_mem_cgroup_per_node_info(memcg
, node
))
4276 if (memcg_wb_domain_init(memcg
, GFP_KERNEL
))
4279 INIT_WORK(&memcg
->high_work
, high_work_func
);
4280 memcg
->last_scanned_node
= MAX_NUMNODES
;
4281 INIT_LIST_HEAD(&memcg
->oom_notify
);
4282 mutex_init(&memcg
->thresholds_lock
);
4283 spin_lock_init(&memcg
->move_lock
);
4284 vmpressure_init(&memcg
->vmpressure
);
4285 INIT_LIST_HEAD(&memcg
->event_list
);
4286 spin_lock_init(&memcg
->event_list_lock
);
4287 memcg
->socket_pressure
= jiffies
;
4289 memcg
->kmemcg_id
= -1;
4291 #ifdef CONFIG_CGROUP_WRITEBACK
4292 INIT_LIST_HEAD(&memcg
->cgwb_list
);
4294 idr_replace(&mem_cgroup_idr
, memcg
, memcg
->id
.id
);
4297 mem_cgroup_id_remove(memcg
);
4298 __mem_cgroup_free(memcg
);
4302 static struct cgroup_subsys_state
* __ref
4303 mem_cgroup_css_alloc(struct cgroup_subsys_state
*parent_css
)
4305 struct mem_cgroup
*parent
= mem_cgroup_from_css(parent_css
);
4306 struct mem_cgroup
*memcg
;
4307 long error
= -ENOMEM
;
4309 memcg
= mem_cgroup_alloc();
4311 return ERR_PTR(error
);
4313 memcg
->high
= PAGE_COUNTER_MAX
;
4314 memcg
->soft_limit
= PAGE_COUNTER_MAX
;
4316 memcg
->swappiness
= mem_cgroup_swappiness(parent
);
4317 memcg
->oom_kill_disable
= parent
->oom_kill_disable
;
4319 if (parent
&& parent
->use_hierarchy
) {
4320 memcg
->use_hierarchy
= true;
4321 page_counter_init(&memcg
->memory
, &parent
->memory
);
4322 page_counter_init(&memcg
->swap
, &parent
->swap
);
4323 page_counter_init(&memcg
->memsw
, &parent
->memsw
);
4324 page_counter_init(&memcg
->kmem
, &parent
->kmem
);
4325 page_counter_init(&memcg
->tcpmem
, &parent
->tcpmem
);
4327 page_counter_init(&memcg
->memory
, NULL
);
4328 page_counter_init(&memcg
->swap
, NULL
);
4329 page_counter_init(&memcg
->memsw
, NULL
);
4330 page_counter_init(&memcg
->kmem
, NULL
);
4331 page_counter_init(&memcg
->tcpmem
, NULL
);
4333 * Deeper hierachy with use_hierarchy == false doesn't make
4334 * much sense so let cgroup subsystem know about this
4335 * unfortunate state in our controller.
4337 if (parent
!= root_mem_cgroup
)
4338 memory_cgrp_subsys
.broken_hierarchy
= true;
4341 /* The following stuff does not apply to the root */
4343 root_mem_cgroup
= memcg
;
4347 error
= memcg_online_kmem(memcg
);
4351 if (cgroup_subsys_on_dfl(memory_cgrp_subsys
) && !cgroup_memory_nosocket
)
4352 static_branch_inc(&memcg_sockets_enabled_key
);
4356 mem_cgroup_id_remove(memcg
);
4357 mem_cgroup_free(memcg
);
4358 return ERR_PTR(-ENOMEM
);
4361 static int mem_cgroup_css_online(struct cgroup_subsys_state
*css
)
4363 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
4365 /* Online state pins memcg ID, memcg ID pins CSS */
4366 atomic_set(&memcg
->id
.ref
, 1);
4371 static void mem_cgroup_css_offline(struct cgroup_subsys_state
*css
)
4373 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
4374 struct mem_cgroup_event
*event
, *tmp
;
4377 * Unregister events and notify userspace.
4378 * Notify userspace about cgroup removing only after rmdir of cgroup
4379 * directory to avoid race between userspace and kernelspace.
4381 spin_lock(&memcg
->event_list_lock
);
4382 list_for_each_entry_safe(event
, tmp
, &memcg
->event_list
, list
) {
4383 list_del_init(&event
->list
);
4384 schedule_work(&event
->remove
);
4386 spin_unlock(&memcg
->event_list_lock
);
4390 memcg_offline_kmem(memcg
);
4391 wb_memcg_offline(memcg
);
4393 mem_cgroup_id_put(memcg
);
4396 static void mem_cgroup_css_released(struct cgroup_subsys_state
*css
)
4398 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
4400 invalidate_reclaim_iterators(memcg
);
4403 static void mem_cgroup_css_free(struct cgroup_subsys_state
*css
)
4405 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
4407 if (cgroup_subsys_on_dfl(memory_cgrp_subsys
) && !cgroup_memory_nosocket
)
4408 static_branch_dec(&memcg_sockets_enabled_key
);
4410 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys
) && memcg
->tcpmem_active
)
4411 static_branch_dec(&memcg_sockets_enabled_key
);
4413 vmpressure_cleanup(&memcg
->vmpressure
);
4414 cancel_work_sync(&memcg
->high_work
);
4415 mem_cgroup_remove_from_trees(memcg
);
4416 memcg_free_kmem(memcg
);
4417 mem_cgroup_free(memcg
);
4421 * mem_cgroup_css_reset - reset the states of a mem_cgroup
4422 * @css: the target css
4424 * Reset the states of the mem_cgroup associated with @css. This is
4425 * invoked when the userland requests disabling on the default hierarchy
4426 * but the memcg is pinned through dependency. The memcg should stop
4427 * applying policies and should revert to the vanilla state as it may be
4428 * made visible again.
4430 * The current implementation only resets the essential configurations.
4431 * This needs to be expanded to cover all the visible parts.
4433 static void mem_cgroup_css_reset(struct cgroup_subsys_state
*css
)
4435 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
4437 page_counter_limit(&memcg
->memory
, PAGE_COUNTER_MAX
);
4438 page_counter_limit(&memcg
->swap
, PAGE_COUNTER_MAX
);
4439 page_counter_limit(&memcg
->memsw
, PAGE_COUNTER_MAX
);
4440 page_counter_limit(&memcg
->kmem
, PAGE_COUNTER_MAX
);
4441 page_counter_limit(&memcg
->tcpmem
, PAGE_COUNTER_MAX
);
4443 memcg
->high
= PAGE_COUNTER_MAX
;
4444 memcg
->soft_limit
= PAGE_COUNTER_MAX
;
4445 memcg_wb_domain_size_changed(memcg
);
4449 /* Handlers for move charge at task migration. */
4450 static int mem_cgroup_do_precharge(unsigned long count
)
4454 /* Try a single bulk charge without reclaim first, kswapd may wake */
4455 ret
= try_charge(mc
.to
, GFP_KERNEL
& ~__GFP_DIRECT_RECLAIM
, count
);
4457 mc
.precharge
+= count
;
4461 /* Try charges one by one with reclaim, but do not retry */
4463 ret
= try_charge(mc
.to
, GFP_KERNEL
| __GFP_NORETRY
, 1);
4477 enum mc_target_type
{
4484 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
4485 unsigned long addr
, pte_t ptent
)
4487 struct page
*page
= _vm_normal_page(vma
, addr
, ptent
, true);
4489 if (!page
|| !page_mapped(page
))
4491 if (PageAnon(page
)) {
4492 if (!(mc
.flags
& MOVE_ANON
))
4495 if (!(mc
.flags
& MOVE_FILE
))
4498 if (!get_page_unless_zero(page
))
4504 #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE)
4505 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
4506 pte_t ptent
, swp_entry_t
*entry
)
4508 struct page
*page
= NULL
;
4509 swp_entry_t ent
= pte_to_swp_entry(ptent
);
4511 if (!(mc
.flags
& MOVE_ANON
) || non_swap_entry(ent
))
4515 * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to
4516 * a device and because they are not accessible by CPU they are store
4517 * as special swap entry in the CPU page table.
4519 if (is_device_private_entry(ent
)) {
4520 page
= device_private_entry_to_page(ent
);
4522 * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have
4523 * a refcount of 1 when free (unlike normal page)
4525 if (!page_ref_add_unless(page
, 1, 1))
4531 * Because lookup_swap_cache() updates some statistics counter,
4532 * we call find_get_page() with swapper_space directly.
4534 page
= find_get_page(swap_address_space(ent
), swp_offset(ent
));
4535 if (do_memsw_account())
4536 entry
->val
= ent
.val
;
4541 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
4542 pte_t ptent
, swp_entry_t
*entry
)
4548 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
4549 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
4551 struct page
*page
= NULL
;
4552 struct address_space
*mapping
;
4555 if (!vma
->vm_file
) /* anonymous vma */
4557 if (!(mc
.flags
& MOVE_FILE
))
4560 mapping
= vma
->vm_file
->f_mapping
;
4561 pgoff
= linear_page_index(vma
, addr
);
4563 /* page is moved even if it's not RSS of this task(page-faulted). */
4565 /* shmem/tmpfs may report page out on swap: account for that too. */
4566 if (shmem_mapping(mapping
)) {
4567 page
= find_get_entry(mapping
, pgoff
);
4568 if (radix_tree_exceptional_entry(page
)) {
4569 swp_entry_t swp
= radix_to_swp_entry(page
);
4570 if (do_memsw_account())
4572 page
= find_get_page(swap_address_space(swp
),
4576 page
= find_get_page(mapping
, pgoff
);
4578 page
= find_get_page(mapping
, pgoff
);
4584 * mem_cgroup_move_account - move account of the page
4586 * @compound: charge the page as compound or small page
4587 * @from: mem_cgroup which the page is moved from.
4588 * @to: mem_cgroup which the page is moved to. @from != @to.
4590 * The caller must make sure the page is not on LRU (isolate_page() is useful.)
4592 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
4595 static int mem_cgroup_move_account(struct page
*page
,
4597 struct mem_cgroup
*from
,
4598 struct mem_cgroup
*to
)
4600 unsigned long flags
;
4601 unsigned int nr_pages
= compound
? hpage_nr_pages(page
) : 1;
4605 VM_BUG_ON(from
== to
);
4606 VM_BUG_ON_PAGE(PageLRU(page
), page
);
4607 VM_BUG_ON(compound
&& !PageTransHuge(page
));
4610 * Prevent mem_cgroup_migrate() from looking at
4611 * page->mem_cgroup of its source page while we change it.
4614 if (!trylock_page(page
))
4618 if (page
->mem_cgroup
!= from
)
4621 anon
= PageAnon(page
);
4623 spin_lock_irqsave(&from
->move_lock
, flags
);
4625 if (!anon
&& page_mapped(page
)) {
4626 __this_cpu_sub(from
->stat
->count
[NR_FILE_MAPPED
], nr_pages
);
4627 __this_cpu_add(to
->stat
->count
[NR_FILE_MAPPED
], nr_pages
);
4631 * move_lock grabbed above and caller set from->moving_account, so
4632 * mod_memcg_page_state will serialize updates to PageDirty.
4633 * So mapping should be stable for dirty pages.
4635 if (!anon
&& PageDirty(page
)) {
4636 struct address_space
*mapping
= page_mapping(page
);
4638 if (mapping_cap_account_dirty(mapping
)) {
4639 __this_cpu_sub(from
->stat
->count
[NR_FILE_DIRTY
],
4641 __this_cpu_add(to
->stat
->count
[NR_FILE_DIRTY
],
4646 if (PageWriteback(page
)) {
4647 __this_cpu_sub(from
->stat
->count
[NR_WRITEBACK
], nr_pages
);
4648 __this_cpu_add(to
->stat
->count
[NR_WRITEBACK
], nr_pages
);
4652 * It is safe to change page->mem_cgroup here because the page
4653 * is referenced, charged, and isolated - we can't race with
4654 * uncharging, charging, migration, or LRU putback.
4657 /* caller should have done css_get */
4658 page
->mem_cgroup
= to
;
4659 spin_unlock_irqrestore(&from
->move_lock
, flags
);
4663 local_irq_disable();
4664 mem_cgroup_charge_statistics(to
, page
, compound
, nr_pages
);
4665 memcg_check_events(to
, page
);
4666 mem_cgroup_charge_statistics(from
, page
, compound
, -nr_pages
);
4667 memcg_check_events(from
, page
);
4676 * get_mctgt_type - get target type of moving charge
4677 * @vma: the vma the pte to be checked belongs
4678 * @addr: the address corresponding to the pte to be checked
4679 * @ptent: the pte to be checked
4680 * @target: the pointer the target page or swap ent will be stored(can be NULL)
4683 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4684 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4685 * move charge. if @target is not NULL, the page is stored in target->page
4686 * with extra refcnt got(Callers should handle it).
4687 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4688 * target for charge migration. if @target is not NULL, the entry is stored
4690 * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PUBLIC
4691 * or MEMORY_DEVICE_PRIVATE (so ZONE_DEVICE page and thus not on the lru).
4692 * For now we such page is charge like a regular page would be as for all
4693 * intent and purposes it is just special memory taking the place of a
4696 * See Documentations/vm/hmm.txt and include/linux/hmm.h
4698 * Called with pte lock held.
4701 static enum mc_target_type
get_mctgt_type(struct vm_area_struct
*vma
,
4702 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
4704 struct page
*page
= NULL
;
4705 enum mc_target_type ret
= MC_TARGET_NONE
;
4706 swp_entry_t ent
= { .val
= 0 };
4708 if (pte_present(ptent
))
4709 page
= mc_handle_present_pte(vma
, addr
, ptent
);
4710 else if (is_swap_pte(ptent
))
4711 page
= mc_handle_swap_pte(vma
, ptent
, &ent
);
4712 else if (pte_none(ptent
))
4713 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
4715 if (!page
&& !ent
.val
)
4719 * Do only loose check w/o serialization.
4720 * mem_cgroup_move_account() checks the page is valid or
4721 * not under LRU exclusion.
4723 if (page
->mem_cgroup
== mc
.from
) {
4724 ret
= MC_TARGET_PAGE
;
4725 if (is_device_private_page(page
) ||
4726 is_device_public_page(page
))
4727 ret
= MC_TARGET_DEVICE
;
4729 target
->page
= page
;
4731 if (!ret
|| !target
)
4735 * There is a swap entry and a page doesn't exist or isn't charged.
4736 * But we cannot move a tail-page in a THP.
4738 if (ent
.val
&& !ret
&& (!page
|| !PageTransCompound(page
)) &&
4739 mem_cgroup_id(mc
.from
) == lookup_swap_cgroup_id(ent
)) {
4740 ret
= MC_TARGET_SWAP
;
4747 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4749 * We don't consider PMD mapped swapping or file mapped pages because THP does
4750 * not support them for now.
4751 * Caller should make sure that pmd_trans_huge(pmd) is true.
4753 static enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
4754 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
4756 struct page
*page
= NULL
;
4757 enum mc_target_type ret
= MC_TARGET_NONE
;
4759 if (unlikely(is_swap_pmd(pmd
))) {
4760 VM_BUG_ON(thp_migration_supported() &&
4761 !is_pmd_migration_entry(pmd
));
4764 page
= pmd_page(pmd
);
4765 VM_BUG_ON_PAGE(!page
|| !PageHead(page
), page
);
4766 if (!(mc
.flags
& MOVE_ANON
))
4768 if (page
->mem_cgroup
== mc
.from
) {
4769 ret
= MC_TARGET_PAGE
;
4772 target
->page
= page
;
4778 static inline enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
4779 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
4781 return MC_TARGET_NONE
;
4785 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
4786 unsigned long addr
, unsigned long end
,
4787 struct mm_walk
*walk
)
4789 struct vm_area_struct
*vma
= walk
->vma
;
4793 ptl
= pmd_trans_huge_lock(pmd
, vma
);
4796 * Note their can not be MC_TARGET_DEVICE for now as we do not
4797 * support transparent huge page with MEMORY_DEVICE_PUBLIC or
4798 * MEMORY_DEVICE_PRIVATE but this might change.
4800 if (get_mctgt_type_thp(vma
, addr
, *pmd
, NULL
) == MC_TARGET_PAGE
)
4801 mc
.precharge
+= HPAGE_PMD_NR
;
4806 if (pmd_trans_unstable(pmd
))
4808 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
4809 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
4810 if (get_mctgt_type(vma
, addr
, *pte
, NULL
))
4811 mc
.precharge
++; /* increment precharge temporarily */
4812 pte_unmap_unlock(pte
- 1, ptl
);
4818 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
4820 unsigned long precharge
;
4822 struct mm_walk mem_cgroup_count_precharge_walk
= {
4823 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
4826 down_read(&mm
->mmap_sem
);
4827 walk_page_range(0, mm
->highest_vm_end
,
4828 &mem_cgroup_count_precharge_walk
);
4829 up_read(&mm
->mmap_sem
);
4831 precharge
= mc
.precharge
;
4837 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
4839 unsigned long precharge
= mem_cgroup_count_precharge(mm
);
4841 VM_BUG_ON(mc
.moving_task
);
4842 mc
.moving_task
= current
;
4843 return mem_cgroup_do_precharge(precharge
);
4846 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4847 static void __mem_cgroup_clear_mc(void)
4849 struct mem_cgroup
*from
= mc
.from
;
4850 struct mem_cgroup
*to
= mc
.to
;
4852 /* we must uncharge all the leftover precharges from mc.to */
4854 cancel_charge(mc
.to
, mc
.precharge
);
4858 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4859 * we must uncharge here.
4861 if (mc
.moved_charge
) {
4862 cancel_charge(mc
.from
, mc
.moved_charge
);
4863 mc
.moved_charge
= 0;
4865 /* we must fixup refcnts and charges */
4866 if (mc
.moved_swap
) {
4867 /* uncharge swap account from the old cgroup */
4868 if (!mem_cgroup_is_root(mc
.from
))
4869 page_counter_uncharge(&mc
.from
->memsw
, mc
.moved_swap
);
4871 mem_cgroup_id_put_many(mc
.from
, mc
.moved_swap
);
4874 * we charged both to->memory and to->memsw, so we
4875 * should uncharge to->memory.
4877 if (!mem_cgroup_is_root(mc
.to
))
4878 page_counter_uncharge(&mc
.to
->memory
, mc
.moved_swap
);
4880 mem_cgroup_id_get_many(mc
.to
, mc
.moved_swap
);
4881 css_put_many(&mc
.to
->css
, mc
.moved_swap
);
4885 memcg_oom_recover(from
);
4886 memcg_oom_recover(to
);
4887 wake_up_all(&mc
.waitq
);
4890 static void mem_cgroup_clear_mc(void)
4892 struct mm_struct
*mm
= mc
.mm
;
4895 * we must clear moving_task before waking up waiters at the end of
4898 mc
.moving_task
= NULL
;
4899 __mem_cgroup_clear_mc();
4900 spin_lock(&mc
.lock
);
4904 spin_unlock(&mc
.lock
);
4909 static int mem_cgroup_can_attach(struct cgroup_taskset
*tset
)
4911 struct cgroup_subsys_state
*css
;
4912 struct mem_cgroup
*memcg
= NULL
; /* unneeded init to make gcc happy */
4913 struct mem_cgroup
*from
;
4914 struct task_struct
*leader
, *p
;
4915 struct mm_struct
*mm
;
4916 unsigned long move_flags
;
4919 /* charge immigration isn't supported on the default hierarchy */
4920 if (cgroup_subsys_on_dfl(memory_cgrp_subsys
))
4924 * Multi-process migrations only happen on the default hierarchy
4925 * where charge immigration is not used. Perform charge
4926 * immigration if @tset contains a leader and whine if there are
4930 cgroup_taskset_for_each_leader(leader
, css
, tset
) {
4933 memcg
= mem_cgroup_from_css(css
);
4939 * We are now commited to this value whatever it is. Changes in this
4940 * tunable will only affect upcoming migrations, not the current one.
4941 * So we need to save it, and keep it going.
4943 move_flags
= READ_ONCE(memcg
->move_charge_at_immigrate
);
4947 from
= mem_cgroup_from_task(p
);
4949 VM_BUG_ON(from
== memcg
);
4951 mm
= get_task_mm(p
);
4954 /* We move charges only when we move a owner of the mm */
4955 if (mm
->owner
== p
) {
4958 VM_BUG_ON(mc
.precharge
);
4959 VM_BUG_ON(mc
.moved_charge
);
4960 VM_BUG_ON(mc
.moved_swap
);
4962 spin_lock(&mc
.lock
);
4966 mc
.flags
= move_flags
;
4967 spin_unlock(&mc
.lock
);
4968 /* We set mc.moving_task later */
4970 ret
= mem_cgroup_precharge_mc(mm
);
4972 mem_cgroup_clear_mc();
4979 static void mem_cgroup_cancel_attach(struct cgroup_taskset
*tset
)
4982 mem_cgroup_clear_mc();
4985 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
4986 unsigned long addr
, unsigned long end
,
4987 struct mm_walk
*walk
)
4990 struct vm_area_struct
*vma
= walk
->vma
;
4993 enum mc_target_type target_type
;
4994 union mc_target target
;
4997 ptl
= pmd_trans_huge_lock(pmd
, vma
);
4999 if (mc
.precharge
< HPAGE_PMD_NR
) {
5003 target_type
= get_mctgt_type_thp(vma
, addr
, *pmd
, &target
);
5004 if (target_type
== MC_TARGET_PAGE
) {
5006 if (!isolate_lru_page(page
)) {
5007 if (!mem_cgroup_move_account(page
, true,
5009 mc
.precharge
-= HPAGE_PMD_NR
;
5010 mc
.moved_charge
+= HPAGE_PMD_NR
;
5012 putback_lru_page(page
);
5015 } else if (target_type
== MC_TARGET_DEVICE
) {
5017 if (!mem_cgroup_move_account(page
, true,
5019 mc
.precharge
-= HPAGE_PMD_NR
;
5020 mc
.moved_charge
+= HPAGE_PMD_NR
;
5028 if (pmd_trans_unstable(pmd
))
5031 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5032 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
5033 pte_t ptent
= *(pte
++);
5034 bool device
= false;
5040 switch (get_mctgt_type(vma
, addr
, ptent
, &target
)) {
5041 case MC_TARGET_DEVICE
:
5044 case MC_TARGET_PAGE
:
5047 * We can have a part of the split pmd here. Moving it
5048 * can be done but it would be too convoluted so simply
5049 * ignore such a partial THP and keep it in original
5050 * memcg. There should be somebody mapping the head.
5052 if (PageTransCompound(page
))
5054 if (!device
&& isolate_lru_page(page
))
5056 if (!mem_cgroup_move_account(page
, false,
5059 /* we uncharge from mc.from later. */
5063 putback_lru_page(page
);
5064 put
: /* get_mctgt_type() gets the page */
5067 case MC_TARGET_SWAP
:
5069 if (!mem_cgroup_move_swap_account(ent
, mc
.from
, mc
.to
)) {
5071 /* we fixup refcnts and charges later. */
5079 pte_unmap_unlock(pte
- 1, ptl
);
5084 * We have consumed all precharges we got in can_attach().
5085 * We try charge one by one, but don't do any additional
5086 * charges to mc.to if we have failed in charge once in attach()
5089 ret
= mem_cgroup_do_precharge(1);
5097 static void mem_cgroup_move_charge(void)
5099 struct mm_walk mem_cgroup_move_charge_walk
= {
5100 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
5104 lru_add_drain_all();
5106 * Signal lock_page_memcg() to take the memcg's move_lock
5107 * while we're moving its pages to another memcg. Then wait
5108 * for already started RCU-only updates to finish.
5110 atomic_inc(&mc
.from
->moving_account
);
5113 if (unlikely(!down_read_trylock(&mc
.mm
->mmap_sem
))) {
5115 * Someone who are holding the mmap_sem might be waiting in
5116 * waitq. So we cancel all extra charges, wake up all waiters,
5117 * and retry. Because we cancel precharges, we might not be able
5118 * to move enough charges, but moving charge is a best-effort
5119 * feature anyway, so it wouldn't be a big problem.
5121 __mem_cgroup_clear_mc();
5126 * When we have consumed all precharges and failed in doing
5127 * additional charge, the page walk just aborts.
5129 walk_page_range(0, mc
.mm
->highest_vm_end
, &mem_cgroup_move_charge_walk
);
5131 up_read(&mc
.mm
->mmap_sem
);
5132 atomic_dec(&mc
.from
->moving_account
);
5135 static void mem_cgroup_move_task(void)
5138 mem_cgroup_move_charge();
5139 mem_cgroup_clear_mc();
5142 #else /* !CONFIG_MMU */
5143 static int mem_cgroup_can_attach(struct cgroup_taskset
*tset
)
5147 static void mem_cgroup_cancel_attach(struct cgroup_taskset
*tset
)
5150 static void mem_cgroup_move_task(void)
5156 * Cgroup retains root cgroups across [un]mount cycles making it necessary
5157 * to verify whether we're attached to the default hierarchy on each mount
5160 static void mem_cgroup_bind(struct cgroup_subsys_state
*root_css
)
5163 * use_hierarchy is forced on the default hierarchy. cgroup core
5164 * guarantees that @root doesn't have any children, so turning it
5165 * on for the root memcg is enough.
5167 if (cgroup_subsys_on_dfl(memory_cgrp_subsys
))
5168 root_mem_cgroup
->use_hierarchy
= true;
5170 root_mem_cgroup
->use_hierarchy
= false;
5173 static u64
memory_current_read(struct cgroup_subsys_state
*css
,
5176 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
5178 return (u64
)page_counter_read(&memcg
->memory
) * PAGE_SIZE
;
5181 static int memory_low_show(struct seq_file
*m
, void *v
)
5183 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
5184 unsigned long low
= READ_ONCE(memcg
->low
);
5186 if (low
== PAGE_COUNTER_MAX
)
5187 seq_puts(m
, "max\n");
5189 seq_printf(m
, "%llu\n", (u64
)low
* PAGE_SIZE
);
5194 static ssize_t
memory_low_write(struct kernfs_open_file
*of
,
5195 char *buf
, size_t nbytes
, loff_t off
)
5197 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
5201 buf
= strstrip(buf
);
5202 err
= page_counter_memparse(buf
, "max", &low
);
5211 static int memory_high_show(struct seq_file
*m
, void *v
)
5213 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
5214 unsigned long high
= READ_ONCE(memcg
->high
);
5216 if (high
== PAGE_COUNTER_MAX
)
5217 seq_puts(m
, "max\n");
5219 seq_printf(m
, "%llu\n", (u64
)high
* PAGE_SIZE
);
5224 static ssize_t
memory_high_write(struct kernfs_open_file
*of
,
5225 char *buf
, size_t nbytes
, loff_t off
)
5227 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
5228 unsigned long nr_pages
;
5232 buf
= strstrip(buf
);
5233 err
= page_counter_memparse(buf
, "max", &high
);
5239 nr_pages
= page_counter_read(&memcg
->memory
);
5240 if (nr_pages
> high
)
5241 try_to_free_mem_cgroup_pages(memcg
, nr_pages
- high
,
5244 memcg_wb_domain_size_changed(memcg
);
5248 static int memory_max_show(struct seq_file
*m
, void *v
)
5250 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
5251 unsigned long max
= READ_ONCE(memcg
->memory
.limit
);
5253 if (max
== PAGE_COUNTER_MAX
)
5254 seq_puts(m
, "max\n");
5256 seq_printf(m
, "%llu\n", (u64
)max
* PAGE_SIZE
);
5261 static ssize_t
memory_max_write(struct kernfs_open_file
*of
,
5262 char *buf
, size_t nbytes
, loff_t off
)
5264 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
5265 unsigned int nr_reclaims
= MEM_CGROUP_RECLAIM_RETRIES
;
5266 bool drained
= false;
5270 buf
= strstrip(buf
);
5271 err
= page_counter_memparse(buf
, "max", &max
);
5275 xchg(&memcg
->memory
.limit
, max
);
5278 unsigned long nr_pages
= page_counter_read(&memcg
->memory
);
5280 if (nr_pages
<= max
)
5283 if (signal_pending(current
)) {
5289 drain_all_stock(memcg
);
5295 if (!try_to_free_mem_cgroup_pages(memcg
, nr_pages
- max
,
5301 mem_cgroup_event(memcg
, MEMCG_OOM
);
5302 if (!mem_cgroup_out_of_memory(memcg
, GFP_KERNEL
, 0))
5306 memcg_wb_domain_size_changed(memcg
);
5310 static int memory_events_show(struct seq_file
*m
, void *v
)
5312 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
5314 seq_printf(m
, "low %lu\n", memcg_sum_events(memcg
, MEMCG_LOW
));
5315 seq_printf(m
, "high %lu\n", memcg_sum_events(memcg
, MEMCG_HIGH
));
5316 seq_printf(m
, "max %lu\n", memcg_sum_events(memcg
, MEMCG_MAX
));
5317 seq_printf(m
, "oom %lu\n", memcg_sum_events(memcg
, MEMCG_OOM
));
5318 seq_printf(m
, "oom_kill %lu\n", memcg_sum_events(memcg
, OOM_KILL
));
5323 static int memory_stat_show(struct seq_file
*m
, void *v
)
5325 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
5326 unsigned long stat
[MEMCG_NR_STAT
];
5327 unsigned long events
[MEMCG_NR_EVENTS
];
5331 * Provide statistics on the state of the memory subsystem as
5332 * well as cumulative event counters that show past behavior.
5334 * This list is ordered following a combination of these gradients:
5335 * 1) generic big picture -> specifics and details
5336 * 2) reflecting userspace activity -> reflecting kernel heuristics
5338 * Current memory state:
5341 tree_stat(memcg
, stat
);
5342 tree_events(memcg
, events
);
5344 seq_printf(m
, "anon %llu\n",
5345 (u64
)stat
[MEMCG_RSS
] * PAGE_SIZE
);
5346 seq_printf(m
, "file %llu\n",
5347 (u64
)stat
[MEMCG_CACHE
] * PAGE_SIZE
);
5348 seq_printf(m
, "kernel_stack %llu\n",
5349 (u64
)stat
[MEMCG_KERNEL_STACK_KB
] * 1024);
5350 seq_printf(m
, "slab %llu\n",
5351 (u64
)(stat
[NR_SLAB_RECLAIMABLE
] +
5352 stat
[NR_SLAB_UNRECLAIMABLE
]) * PAGE_SIZE
);
5353 seq_printf(m
, "sock %llu\n",
5354 (u64
)stat
[MEMCG_SOCK
] * PAGE_SIZE
);
5356 seq_printf(m
, "shmem %llu\n",
5357 (u64
)stat
[NR_SHMEM
] * PAGE_SIZE
);
5358 seq_printf(m
, "file_mapped %llu\n",
5359 (u64
)stat
[NR_FILE_MAPPED
] * PAGE_SIZE
);
5360 seq_printf(m
, "file_dirty %llu\n",
5361 (u64
)stat
[NR_FILE_DIRTY
] * PAGE_SIZE
);
5362 seq_printf(m
, "file_writeback %llu\n",
5363 (u64
)stat
[NR_WRITEBACK
] * PAGE_SIZE
);
5365 for (i
= 0; i
< NR_LRU_LISTS
; i
++) {
5366 struct mem_cgroup
*mi
;
5367 unsigned long val
= 0;
5369 for_each_mem_cgroup_tree(mi
, memcg
)
5370 val
+= mem_cgroup_nr_lru_pages(mi
, BIT(i
));
5371 seq_printf(m
, "%s %llu\n",
5372 mem_cgroup_lru_names
[i
], (u64
)val
* PAGE_SIZE
);
5375 seq_printf(m
, "slab_reclaimable %llu\n",
5376 (u64
)stat
[NR_SLAB_RECLAIMABLE
] * PAGE_SIZE
);
5377 seq_printf(m
, "slab_unreclaimable %llu\n",
5378 (u64
)stat
[NR_SLAB_UNRECLAIMABLE
] * PAGE_SIZE
);
5380 /* Accumulated memory events */
5382 seq_printf(m
, "pgfault %lu\n", events
[PGFAULT
]);
5383 seq_printf(m
, "pgmajfault %lu\n", events
[PGMAJFAULT
]);
5385 seq_printf(m
, "pgrefill %lu\n", events
[PGREFILL
]);
5386 seq_printf(m
, "pgscan %lu\n", events
[PGSCAN_KSWAPD
] +
5387 events
[PGSCAN_DIRECT
]);
5388 seq_printf(m
, "pgsteal %lu\n", events
[PGSTEAL_KSWAPD
] +
5389 events
[PGSTEAL_DIRECT
]);
5390 seq_printf(m
, "pgactivate %lu\n", events
[PGACTIVATE
]);
5391 seq_printf(m
, "pgdeactivate %lu\n", events
[PGDEACTIVATE
]);
5392 seq_printf(m
, "pglazyfree %lu\n", events
[PGLAZYFREE
]);
5393 seq_printf(m
, "pglazyfreed %lu\n", events
[PGLAZYFREED
]);
5395 seq_printf(m
, "workingset_refault %lu\n",
5396 stat
[WORKINGSET_REFAULT
]);
5397 seq_printf(m
, "workingset_activate %lu\n",
5398 stat
[WORKINGSET_ACTIVATE
]);
5399 seq_printf(m
, "workingset_nodereclaim %lu\n",
5400 stat
[WORKINGSET_NODERECLAIM
]);
5405 static struct cftype memory_files
[] = {
5408 .flags
= CFTYPE_NOT_ON_ROOT
,
5409 .read_u64
= memory_current_read
,
5413 .flags
= CFTYPE_NOT_ON_ROOT
,
5414 .seq_show
= memory_low_show
,
5415 .write
= memory_low_write
,
5419 .flags
= CFTYPE_NOT_ON_ROOT
,
5420 .seq_show
= memory_high_show
,
5421 .write
= memory_high_write
,
5425 .flags
= CFTYPE_NOT_ON_ROOT
,
5426 .seq_show
= memory_max_show
,
5427 .write
= memory_max_write
,
5431 .flags
= CFTYPE_NOT_ON_ROOT
,
5432 .file_offset
= offsetof(struct mem_cgroup
, events_file
),
5433 .seq_show
= memory_events_show
,
5437 .flags
= CFTYPE_NOT_ON_ROOT
,
5438 .seq_show
= memory_stat_show
,
5443 struct cgroup_subsys memory_cgrp_subsys
= {
5444 .css_alloc
= mem_cgroup_css_alloc
,
5445 .css_online
= mem_cgroup_css_online
,
5446 .css_offline
= mem_cgroup_css_offline
,
5447 .css_released
= mem_cgroup_css_released
,
5448 .css_free
= mem_cgroup_css_free
,
5449 .css_reset
= mem_cgroup_css_reset
,
5450 .can_attach
= mem_cgroup_can_attach
,
5451 .cancel_attach
= mem_cgroup_cancel_attach
,
5452 .post_attach
= mem_cgroup_move_task
,
5453 .bind
= mem_cgroup_bind
,
5454 .dfl_cftypes
= memory_files
,
5455 .legacy_cftypes
= mem_cgroup_legacy_files
,
5460 * mem_cgroup_low - check if memory consumption is below the normal range
5461 * @root: the top ancestor of the sub-tree being checked
5462 * @memcg: the memory cgroup to check
5464 * Returns %true if memory consumption of @memcg, and that of all
5465 * ancestors up to (but not including) @root, is below the normal range.
5467 * @root is exclusive; it is never low when looked at directly and isn't
5468 * checked when traversing the hierarchy.
5470 * Excluding @root enables using memory.low to prioritize memory usage
5471 * between cgroups within a subtree of the hierarchy that is limited by
5472 * memory.high or memory.max.
5474 * For example, given cgroup A with children B and C:
5482 * 1. A/memory.current > A/memory.high
5483 * 2. A/B/memory.current < A/B/memory.low
5484 * 3. A/C/memory.current >= A/C/memory.low
5486 * As 'A' is high, i.e. triggers reclaim from 'A', and 'B' is low, we
5487 * should reclaim from 'C' until 'A' is no longer high or until we can
5488 * no longer reclaim from 'C'. If 'A', i.e. @root, isn't excluded by
5489 * mem_cgroup_low when reclaming from 'A', then 'B' won't be considered
5490 * low and we will reclaim indiscriminately from both 'B' and 'C'.
5492 bool mem_cgroup_low(struct mem_cgroup
*root
, struct mem_cgroup
*memcg
)
5494 if (mem_cgroup_disabled())
5498 root
= root_mem_cgroup
;
5502 for (; memcg
!= root
; memcg
= parent_mem_cgroup(memcg
)) {
5503 if (page_counter_read(&memcg
->memory
) >= memcg
->low
)
5511 * mem_cgroup_try_charge - try charging a page
5512 * @page: page to charge
5513 * @mm: mm context of the victim
5514 * @gfp_mask: reclaim mode
5515 * @memcgp: charged memcg return
5516 * @compound: charge the page as compound or small page
5518 * Try to charge @page to the memcg that @mm belongs to, reclaiming
5519 * pages according to @gfp_mask if necessary.
5521 * Returns 0 on success, with *@memcgp pointing to the charged memcg.
5522 * Otherwise, an error code is returned.
5524 * After page->mapping has been set up, the caller must finalize the
5525 * charge with mem_cgroup_commit_charge(). Or abort the transaction
5526 * with mem_cgroup_cancel_charge() in case page instantiation fails.
5528 int mem_cgroup_try_charge(struct page
*page
, struct mm_struct
*mm
,
5529 gfp_t gfp_mask
, struct mem_cgroup
**memcgp
,
5532 struct mem_cgroup
*memcg
= NULL
;
5533 unsigned int nr_pages
= compound
? hpage_nr_pages(page
) : 1;
5536 if (mem_cgroup_disabled())
5539 if (PageSwapCache(page
)) {
5541 * Every swap fault against a single page tries to charge the
5542 * page, bail as early as possible. shmem_unuse() encounters
5543 * already charged pages, too. The USED bit is protected by
5544 * the page lock, which serializes swap cache removal, which
5545 * in turn serializes uncharging.
5547 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
5548 if (compound_head(page
)->mem_cgroup
)
5551 if (do_swap_account
) {
5552 swp_entry_t ent
= { .val
= page_private(page
), };
5553 unsigned short id
= lookup_swap_cgroup_id(ent
);
5556 memcg
= mem_cgroup_from_id(id
);
5557 if (memcg
&& !css_tryget_online(&memcg
->css
))
5564 memcg
= get_mem_cgroup_from_mm(mm
);
5566 ret
= try_charge(memcg
, gfp_mask
, nr_pages
);
5568 css_put(&memcg
->css
);
5575 * mem_cgroup_commit_charge - commit a page charge
5576 * @page: page to charge
5577 * @memcg: memcg to charge the page to
5578 * @lrucare: page might be on LRU already
5579 * @compound: charge the page as compound or small page
5581 * Finalize a charge transaction started by mem_cgroup_try_charge(),
5582 * after page->mapping has been set up. This must happen atomically
5583 * as part of the page instantiation, i.e. under the page table lock
5584 * for anonymous pages, under the page lock for page and swap cache.
5586 * In addition, the page must not be on the LRU during the commit, to
5587 * prevent racing with task migration. If it might be, use @lrucare.
5589 * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
5591 void mem_cgroup_commit_charge(struct page
*page
, struct mem_cgroup
*memcg
,
5592 bool lrucare
, bool compound
)
5594 unsigned int nr_pages
= compound
? hpage_nr_pages(page
) : 1;
5596 VM_BUG_ON_PAGE(!page
->mapping
, page
);
5597 VM_BUG_ON_PAGE(PageLRU(page
) && !lrucare
, page
);
5599 if (mem_cgroup_disabled())
5602 * Swap faults will attempt to charge the same page multiple
5603 * times. But reuse_swap_page() might have removed the page
5604 * from swapcache already, so we can't check PageSwapCache().
5609 commit_charge(page
, memcg
, lrucare
);
5611 local_irq_disable();
5612 mem_cgroup_charge_statistics(memcg
, page
, compound
, nr_pages
);
5613 memcg_check_events(memcg
, page
);
5616 if (do_memsw_account() && PageSwapCache(page
)) {
5617 swp_entry_t entry
= { .val
= page_private(page
) };
5619 * The swap entry might not get freed for a long time,
5620 * let's not wait for it. The page already received a
5621 * memory+swap charge, drop the swap entry duplicate.
5623 mem_cgroup_uncharge_swap(entry
, nr_pages
);
5628 * mem_cgroup_cancel_charge - cancel a page charge
5629 * @page: page to charge
5630 * @memcg: memcg to charge the page to
5631 * @compound: charge the page as compound or small page
5633 * Cancel a charge transaction started by mem_cgroup_try_charge().
5635 void mem_cgroup_cancel_charge(struct page
*page
, struct mem_cgroup
*memcg
,
5638 unsigned int nr_pages
= compound
? hpage_nr_pages(page
) : 1;
5640 if (mem_cgroup_disabled())
5643 * Swap faults will attempt to charge the same page multiple
5644 * times. But reuse_swap_page() might have removed the page
5645 * from swapcache already, so we can't check PageSwapCache().
5650 cancel_charge(memcg
, nr_pages
);
5653 struct uncharge_gather
{
5654 struct mem_cgroup
*memcg
;
5655 unsigned long pgpgout
;
5656 unsigned long nr_anon
;
5657 unsigned long nr_file
;
5658 unsigned long nr_kmem
;
5659 unsigned long nr_huge
;
5660 unsigned long nr_shmem
;
5661 struct page
*dummy_page
;
5664 static inline void uncharge_gather_clear(struct uncharge_gather
*ug
)
5666 memset(ug
, 0, sizeof(*ug
));
5669 static void uncharge_batch(const struct uncharge_gather
*ug
)
5671 unsigned long nr_pages
= ug
->nr_anon
+ ug
->nr_file
+ ug
->nr_kmem
;
5672 unsigned long flags
;
5674 if (!mem_cgroup_is_root(ug
->memcg
)) {
5675 page_counter_uncharge(&ug
->memcg
->memory
, nr_pages
);
5676 if (do_memsw_account())
5677 page_counter_uncharge(&ug
->memcg
->memsw
, nr_pages
);
5678 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys
) && ug
->nr_kmem
)
5679 page_counter_uncharge(&ug
->memcg
->kmem
, ug
->nr_kmem
);
5680 memcg_oom_recover(ug
->memcg
);
5683 local_irq_save(flags
);
5684 __this_cpu_sub(ug
->memcg
->stat
->count
[MEMCG_RSS
], ug
->nr_anon
);
5685 __this_cpu_sub(ug
->memcg
->stat
->count
[MEMCG_CACHE
], ug
->nr_file
);
5686 __this_cpu_sub(ug
->memcg
->stat
->count
[MEMCG_RSS_HUGE
], ug
->nr_huge
);
5687 __this_cpu_sub(ug
->memcg
->stat
->count
[NR_SHMEM
], ug
->nr_shmem
);
5688 __this_cpu_add(ug
->memcg
->stat
->events
[PGPGOUT
], ug
->pgpgout
);
5689 __this_cpu_add(ug
->memcg
->stat
->nr_page_events
, nr_pages
);
5690 memcg_check_events(ug
->memcg
, ug
->dummy_page
);
5691 local_irq_restore(flags
);
5693 if (!mem_cgroup_is_root(ug
->memcg
))
5694 css_put_many(&ug
->memcg
->css
, nr_pages
);
5697 static void uncharge_page(struct page
*page
, struct uncharge_gather
*ug
)
5699 VM_BUG_ON_PAGE(PageLRU(page
), page
);
5700 VM_BUG_ON_PAGE(page_count(page
) && !is_zone_device_page(page
) &&
5701 !PageHWPoison(page
) , page
);
5703 if (!page
->mem_cgroup
)
5707 * Nobody should be changing or seriously looking at
5708 * page->mem_cgroup at this point, we have fully
5709 * exclusive access to the page.
5712 if (ug
->memcg
!= page
->mem_cgroup
) {
5715 uncharge_gather_clear(ug
);
5717 ug
->memcg
= page
->mem_cgroup
;
5720 if (!PageKmemcg(page
)) {
5721 unsigned int nr_pages
= 1;
5723 if (PageTransHuge(page
)) {
5724 nr_pages
<<= compound_order(page
);
5725 ug
->nr_huge
+= nr_pages
;
5728 ug
->nr_anon
+= nr_pages
;
5730 ug
->nr_file
+= nr_pages
;
5731 if (PageSwapBacked(page
))
5732 ug
->nr_shmem
+= nr_pages
;
5736 ug
->nr_kmem
+= 1 << compound_order(page
);
5737 __ClearPageKmemcg(page
);
5740 ug
->dummy_page
= page
;
5741 page
->mem_cgroup
= NULL
;
5744 static void uncharge_list(struct list_head
*page_list
)
5746 struct uncharge_gather ug
;
5747 struct list_head
*next
;
5749 uncharge_gather_clear(&ug
);
5752 * Note that the list can be a single page->lru; hence the
5753 * do-while loop instead of a simple list_for_each_entry().
5755 next
= page_list
->next
;
5759 page
= list_entry(next
, struct page
, lru
);
5760 next
= page
->lru
.next
;
5762 uncharge_page(page
, &ug
);
5763 } while (next
!= page_list
);
5766 uncharge_batch(&ug
);
5770 * mem_cgroup_uncharge - uncharge a page
5771 * @page: page to uncharge
5773 * Uncharge a page previously charged with mem_cgroup_try_charge() and
5774 * mem_cgroup_commit_charge().
5776 void mem_cgroup_uncharge(struct page
*page
)
5778 struct uncharge_gather ug
;
5780 if (mem_cgroup_disabled())
5783 /* Don't touch page->lru of any random page, pre-check: */
5784 if (!page
->mem_cgroup
)
5787 uncharge_gather_clear(&ug
);
5788 uncharge_page(page
, &ug
);
5789 uncharge_batch(&ug
);
5793 * mem_cgroup_uncharge_list - uncharge a list of page
5794 * @page_list: list of pages to uncharge
5796 * Uncharge a list of pages previously charged with
5797 * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
5799 void mem_cgroup_uncharge_list(struct list_head
*page_list
)
5801 if (mem_cgroup_disabled())
5804 if (!list_empty(page_list
))
5805 uncharge_list(page_list
);
5809 * mem_cgroup_migrate - charge a page's replacement
5810 * @oldpage: currently circulating page
5811 * @newpage: replacement page
5813 * Charge @newpage as a replacement page for @oldpage. @oldpage will
5814 * be uncharged upon free.
5816 * Both pages must be locked, @newpage->mapping must be set up.
5818 void mem_cgroup_migrate(struct page
*oldpage
, struct page
*newpage
)
5820 struct mem_cgroup
*memcg
;
5821 unsigned int nr_pages
;
5823 unsigned long flags
;
5825 VM_BUG_ON_PAGE(!PageLocked(oldpage
), oldpage
);
5826 VM_BUG_ON_PAGE(!PageLocked(newpage
), newpage
);
5827 VM_BUG_ON_PAGE(PageAnon(oldpage
) != PageAnon(newpage
), newpage
);
5828 VM_BUG_ON_PAGE(PageTransHuge(oldpage
) != PageTransHuge(newpage
),
5831 if (mem_cgroup_disabled())
5834 /* Page cache replacement: new page already charged? */
5835 if (newpage
->mem_cgroup
)
5838 /* Swapcache readahead pages can get replaced before being charged */
5839 memcg
= oldpage
->mem_cgroup
;
5843 /* Force-charge the new page. The old one will be freed soon */
5844 compound
= PageTransHuge(newpage
);
5845 nr_pages
= compound
? hpage_nr_pages(newpage
) : 1;
5847 page_counter_charge(&memcg
->memory
, nr_pages
);
5848 if (do_memsw_account())
5849 page_counter_charge(&memcg
->memsw
, nr_pages
);
5850 css_get_many(&memcg
->css
, nr_pages
);
5852 commit_charge(newpage
, memcg
, false);
5854 local_irq_save(flags
);
5855 mem_cgroup_charge_statistics(memcg
, newpage
, compound
, nr_pages
);
5856 memcg_check_events(memcg
, newpage
);
5857 local_irq_restore(flags
);
5860 DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key
);
5861 EXPORT_SYMBOL(memcg_sockets_enabled_key
);
5863 void mem_cgroup_sk_alloc(struct sock
*sk
)
5865 struct mem_cgroup
*memcg
;
5867 if (!mem_cgroup_sockets_enabled
)
5871 * Socket cloning can throw us here with sk_memcg already
5872 * filled. It won't however, necessarily happen from
5873 * process context. So the test for root memcg given
5874 * the current task's memcg won't help us in this case.
5876 * Respecting the original socket's memcg is a better
5877 * decision in this case.
5880 css_get(&sk
->sk_memcg
->css
);
5885 memcg
= mem_cgroup_from_task(current
);
5886 if (memcg
== root_mem_cgroup
)
5888 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys
) && !memcg
->tcpmem_active
)
5890 if (css_tryget_online(&memcg
->css
))
5891 sk
->sk_memcg
= memcg
;
5896 void mem_cgroup_sk_free(struct sock
*sk
)
5899 css_put(&sk
->sk_memcg
->css
);
5903 * mem_cgroup_charge_skmem - charge socket memory
5904 * @memcg: memcg to charge
5905 * @nr_pages: number of pages to charge
5907 * Charges @nr_pages to @memcg. Returns %true if the charge fit within
5908 * @memcg's configured limit, %false if the charge had to be forced.
5910 bool mem_cgroup_charge_skmem(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
5912 gfp_t gfp_mask
= GFP_KERNEL
;
5914 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys
)) {
5915 struct page_counter
*fail
;
5917 if (page_counter_try_charge(&memcg
->tcpmem
, nr_pages
, &fail
)) {
5918 memcg
->tcpmem_pressure
= 0;
5921 page_counter_charge(&memcg
->tcpmem
, nr_pages
);
5922 memcg
->tcpmem_pressure
= 1;
5926 /* Don't block in the packet receive path */
5928 gfp_mask
= GFP_NOWAIT
;
5930 this_cpu_add(memcg
->stat
->count
[MEMCG_SOCK
], nr_pages
);
5932 if (try_charge(memcg
, gfp_mask
, nr_pages
) == 0)
5935 try_charge(memcg
, gfp_mask
|__GFP_NOFAIL
, nr_pages
);
5940 * mem_cgroup_uncharge_skmem - uncharge socket memory
5941 * @memcg - memcg to uncharge
5942 * @nr_pages - number of pages to uncharge
5944 void mem_cgroup_uncharge_skmem(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
5946 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys
)) {
5947 page_counter_uncharge(&memcg
->tcpmem
, nr_pages
);
5951 this_cpu_sub(memcg
->stat
->count
[MEMCG_SOCK
], nr_pages
);
5953 refill_stock(memcg
, nr_pages
);
5956 static int __init
cgroup_memory(char *s
)
5960 while ((token
= strsep(&s
, ",")) != NULL
) {
5963 if (!strcmp(token
, "nosocket"))
5964 cgroup_memory_nosocket
= true;
5965 if (!strcmp(token
, "nokmem"))
5966 cgroup_memory_nokmem
= true;
5970 __setup("cgroup.memory=", cgroup_memory
);
5973 * subsys_initcall() for memory controller.
5975 * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this
5976 * context because of lock dependencies (cgroup_lock -> cpu hotplug) but
5977 * basically everything that doesn't depend on a specific mem_cgroup structure
5978 * should be initialized from here.
5980 static int __init
mem_cgroup_init(void)
5986 * Kmem cache creation is mostly done with the slab_mutex held,
5987 * so use a workqueue with limited concurrency to avoid stalling
5988 * all worker threads in case lots of cgroups are created and
5989 * destroyed simultaneously.
5991 memcg_kmem_cache_wq
= alloc_workqueue("memcg_kmem_cache", 0, 1);
5992 BUG_ON(!memcg_kmem_cache_wq
);
5995 cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD
, "mm/memctrl:dead", NULL
,
5996 memcg_hotplug_cpu_dead
);
5998 for_each_possible_cpu(cpu
)
5999 INIT_WORK(&per_cpu_ptr(&memcg_stock
, cpu
)->work
,
6002 for_each_node(node
) {
6003 struct mem_cgroup_tree_per_node
*rtpn
;
6005 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
,
6006 node_online(node
) ? node
: NUMA_NO_NODE
);
6008 rtpn
->rb_root
= RB_ROOT
;
6009 rtpn
->rb_rightmost
= NULL
;
6010 spin_lock_init(&rtpn
->lock
);
6011 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
6016 subsys_initcall(mem_cgroup_init
);
6018 #ifdef CONFIG_MEMCG_SWAP
6019 static struct mem_cgroup
*mem_cgroup_id_get_online(struct mem_cgroup
*memcg
)
6021 while (!atomic_inc_not_zero(&memcg
->id
.ref
)) {
6023 * The root cgroup cannot be destroyed, so it's refcount must
6026 if (WARN_ON_ONCE(memcg
== root_mem_cgroup
)) {
6030 memcg
= parent_mem_cgroup(memcg
);
6032 memcg
= root_mem_cgroup
;
6038 * mem_cgroup_swapout - transfer a memsw charge to swap
6039 * @page: page whose memsw charge to transfer
6040 * @entry: swap entry to move the charge to
6042 * Transfer the memsw charge of @page to @entry.
6044 void mem_cgroup_swapout(struct page
*page
, swp_entry_t entry
)
6046 struct mem_cgroup
*memcg
, *swap_memcg
;
6047 unsigned int nr_entries
;
6048 unsigned short oldid
;
6050 VM_BUG_ON_PAGE(PageLRU(page
), page
);
6051 VM_BUG_ON_PAGE(page_count(page
), page
);
6053 if (!do_memsw_account())
6056 memcg
= page
->mem_cgroup
;
6058 /* Readahead page, never charged */
6063 * In case the memcg owning these pages has been offlined and doesn't
6064 * have an ID allocated to it anymore, charge the closest online
6065 * ancestor for the swap instead and transfer the memory+swap charge.
6067 swap_memcg
= mem_cgroup_id_get_online(memcg
);
6068 nr_entries
= hpage_nr_pages(page
);
6069 /* Get references for the tail pages, too */
6071 mem_cgroup_id_get_many(swap_memcg
, nr_entries
- 1);
6072 oldid
= swap_cgroup_record(entry
, mem_cgroup_id(swap_memcg
),
6074 VM_BUG_ON_PAGE(oldid
, page
);
6075 mem_cgroup_swap_statistics(swap_memcg
, nr_entries
);
6077 page
->mem_cgroup
= NULL
;
6079 if (!mem_cgroup_is_root(memcg
))
6080 page_counter_uncharge(&memcg
->memory
, nr_entries
);
6082 if (memcg
!= swap_memcg
) {
6083 if (!mem_cgroup_is_root(swap_memcg
))
6084 page_counter_charge(&swap_memcg
->memsw
, nr_entries
);
6085 page_counter_uncharge(&memcg
->memsw
, nr_entries
);
6089 * Interrupts should be disabled here because the caller holds the
6090 * mapping->tree_lock lock which is taken with interrupts-off. It is
6091 * important here to have the interrupts disabled because it is the
6092 * only synchronisation we have for udpating the per-CPU variables.
6094 VM_BUG_ON(!irqs_disabled());
6095 mem_cgroup_charge_statistics(memcg
, page
, PageTransHuge(page
),
6097 memcg_check_events(memcg
, page
);
6099 if (!mem_cgroup_is_root(memcg
))
6100 css_put_many(&memcg
->css
, nr_entries
);
6104 * mem_cgroup_try_charge_swap - try charging swap space for a page
6105 * @page: page being added to swap
6106 * @entry: swap entry to charge
6108 * Try to charge @page's memcg for the swap space at @entry.
6110 * Returns 0 on success, -ENOMEM on failure.
6112 int mem_cgroup_try_charge_swap(struct page
*page
, swp_entry_t entry
)
6114 unsigned int nr_pages
= hpage_nr_pages(page
);
6115 struct page_counter
*counter
;
6116 struct mem_cgroup
*memcg
;
6117 unsigned short oldid
;
6119 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys
) || !do_swap_account
)
6122 memcg
= page
->mem_cgroup
;
6124 /* Readahead page, never charged */
6128 memcg
= mem_cgroup_id_get_online(memcg
);
6130 if (!mem_cgroup_is_root(memcg
) &&
6131 !page_counter_try_charge(&memcg
->swap
, nr_pages
, &counter
)) {
6132 mem_cgroup_id_put(memcg
);
6136 /* Get references for the tail pages, too */
6138 mem_cgroup_id_get_many(memcg
, nr_pages
- 1);
6139 oldid
= swap_cgroup_record(entry
, mem_cgroup_id(memcg
), nr_pages
);
6140 VM_BUG_ON_PAGE(oldid
, page
);
6141 mem_cgroup_swap_statistics(memcg
, nr_pages
);
6147 * mem_cgroup_uncharge_swap - uncharge swap space
6148 * @entry: swap entry to uncharge
6149 * @nr_pages: the amount of swap space to uncharge
6151 void mem_cgroup_uncharge_swap(swp_entry_t entry
, unsigned int nr_pages
)
6153 struct mem_cgroup
*memcg
;
6156 if (!do_swap_account
)
6159 id
= swap_cgroup_record(entry
, 0, nr_pages
);
6161 memcg
= mem_cgroup_from_id(id
);
6163 if (!mem_cgroup_is_root(memcg
)) {
6164 if (cgroup_subsys_on_dfl(memory_cgrp_subsys
))
6165 page_counter_uncharge(&memcg
->swap
, nr_pages
);
6167 page_counter_uncharge(&memcg
->memsw
, nr_pages
);
6169 mem_cgroup_swap_statistics(memcg
, -nr_pages
);
6170 mem_cgroup_id_put_many(memcg
, nr_pages
);
6175 long mem_cgroup_get_nr_swap_pages(struct mem_cgroup
*memcg
)
6177 long nr_swap_pages
= get_nr_swap_pages();
6179 if (!do_swap_account
|| !cgroup_subsys_on_dfl(memory_cgrp_subsys
))
6180 return nr_swap_pages
;
6181 for (; memcg
!= root_mem_cgroup
; memcg
= parent_mem_cgroup(memcg
))
6182 nr_swap_pages
= min_t(long, nr_swap_pages
,
6183 READ_ONCE(memcg
->swap
.limit
) -
6184 page_counter_read(&memcg
->swap
));
6185 return nr_swap_pages
;
6188 bool mem_cgroup_swap_full(struct page
*page
)
6190 struct mem_cgroup
*memcg
;
6192 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
6196 if (!do_swap_account
|| !cgroup_subsys_on_dfl(memory_cgrp_subsys
))
6199 memcg
= page
->mem_cgroup
;
6203 for (; memcg
!= root_mem_cgroup
; memcg
= parent_mem_cgroup(memcg
))
6204 if (page_counter_read(&memcg
->swap
) * 2 >= memcg
->swap
.limit
)
6210 /* for remember boot option*/
6211 #ifdef CONFIG_MEMCG_SWAP_ENABLED
6212 static int really_do_swap_account __initdata
= 1;
6214 static int really_do_swap_account __initdata
;
6217 static int __init
enable_swap_account(char *s
)
6219 if (!strcmp(s
, "1"))
6220 really_do_swap_account
= 1;
6221 else if (!strcmp(s
, "0"))
6222 really_do_swap_account
= 0;
6225 __setup("swapaccount=", enable_swap_account
);
6227 static u64
swap_current_read(struct cgroup_subsys_state
*css
,
6230 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
6232 return (u64
)page_counter_read(&memcg
->swap
) * PAGE_SIZE
;
6235 static int swap_max_show(struct seq_file
*m
, void *v
)
6237 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
6238 unsigned long max
= READ_ONCE(memcg
->swap
.limit
);
6240 if (max
== PAGE_COUNTER_MAX
)
6241 seq_puts(m
, "max\n");
6243 seq_printf(m
, "%llu\n", (u64
)max
* PAGE_SIZE
);
6248 static ssize_t
swap_max_write(struct kernfs_open_file
*of
,
6249 char *buf
, size_t nbytes
, loff_t off
)
6251 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
6255 buf
= strstrip(buf
);
6256 err
= page_counter_memparse(buf
, "max", &max
);
6260 mutex_lock(&memcg_limit_mutex
);
6261 err
= page_counter_limit(&memcg
->swap
, max
);
6262 mutex_unlock(&memcg_limit_mutex
);
6269 static struct cftype swap_files
[] = {
6271 .name
= "swap.current",
6272 .flags
= CFTYPE_NOT_ON_ROOT
,
6273 .read_u64
= swap_current_read
,
6277 .flags
= CFTYPE_NOT_ON_ROOT
,
6278 .seq_show
= swap_max_show
,
6279 .write
= swap_max_write
,
6284 static struct cftype memsw_cgroup_files
[] = {
6286 .name
= "memsw.usage_in_bytes",
6287 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
6288 .read_u64
= mem_cgroup_read_u64
,
6291 .name
= "memsw.max_usage_in_bytes",
6292 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
6293 .write
= mem_cgroup_reset
,
6294 .read_u64
= mem_cgroup_read_u64
,
6297 .name
= "memsw.limit_in_bytes",
6298 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
6299 .write
= mem_cgroup_write
,
6300 .read_u64
= mem_cgroup_read_u64
,
6303 .name
= "memsw.failcnt",
6304 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
6305 .write
= mem_cgroup_reset
,
6306 .read_u64
= mem_cgroup_read_u64
,
6308 { }, /* terminate */
6311 static int __init
mem_cgroup_swap_init(void)
6313 if (!mem_cgroup_disabled() && really_do_swap_account
) {
6314 do_swap_account
= 1;
6315 WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys
,
6317 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys
,
6318 memsw_cgroup_files
));
6322 subsys_initcall(mem_cgroup_swap_init
);
6324 #endif /* CONFIG_MEMCG_SWAP */