Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 1 | /* memcontrol.c - Memory Controller |
| 2 | * |
| 3 | * Copyright IBM Corporation, 2007 |
| 4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> |
| 5 | * |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 6 | * Copyright 2007 OpenVZ SWsoft Inc |
| 7 | * Author: Pavel Emelianov <xemul@openvz.org> |
| 8 | * |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 9 | * Memory thresholds |
| 10 | * Copyright (C) 2009 Nokia Corporation |
| 11 | * Author: Kirill A. Shutemov |
| 12 | * |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 13 | * Kernel Memory Controller |
| 14 | * Copyright (C) 2012 Parallels Inc. and Google Inc. |
| 15 | * Authors: Glauber Costa and Suleiman Souhlal |
| 16 | * |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 17 | * This program is free software; you can redistribute it and/or modify |
| 18 | * it under the terms of the GNU General Public License as published by |
| 19 | * the Free Software Foundation; either version 2 of the License, or |
| 20 | * (at your option) any later version. |
| 21 | * |
| 22 | * This program is distributed in the hope that it will be useful, |
| 23 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 24 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 25 | * GNU General Public License for more details. |
| 26 | */ |
| 27 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 28 | #include <linux/page_counter.h> |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 29 | #include <linux/memcontrol.h> |
| 30 | #include <linux/cgroup.h> |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 31 | #include <linux/mm.h> |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 32 | #include <linux/hugetlb.h> |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 33 | #include <linux/pagemap.h> |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 34 | #include <linux/smp.h> |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 35 | #include <linux/page-flags.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 36 | #include <linux/backing-dev.h> |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 37 | #include <linux/bit_spinlock.h> |
| 38 | #include <linux/rcupdate.h> |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 39 | #include <linux/limits.h> |
Paul Gortmaker | b9e15ba | 2011-05-26 16:00:52 -0400 | [diff] [blame] | 40 | #include <linux/export.h> |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 41 | #include <linux/mutex.h> |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 42 | #include <linux/rbtree.h> |
Balbir Singh | b6ac57d | 2008-04-29 01:00:19 -0700 | [diff] [blame] | 43 | #include <linux/slab.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 44 | #include <linux/swap.h> |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 45 | #include <linux/swapops.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 46 | #include <linux/spinlock.h> |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 47 | #include <linux/eventfd.h> |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 48 | #include <linux/poll.h> |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 49 | #include <linux/sort.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 50 | #include <linux/fs.h> |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 51 | #include <linux/seq_file.h> |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 52 | #include <linux/vmpressure.h> |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 53 | #include <linux/mm_inline.h> |
Johannes Weiner | 5d1ea48 | 2014-12-10 15:44:55 -0800 | [diff] [blame] | 54 | #include <linux/swap_cgroup.h> |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 55 | #include <linux/cpu.h> |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 56 | #include <linux/oom.h> |
Johannes Weiner | 0056f4e | 2013-10-31 16:34:14 -0700 | [diff] [blame] | 57 | #include <linux/lockdep.h> |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 58 | #include <linux/file.h> |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 59 | #include "internal.h" |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 60 | #include <net/sock.h> |
Michal Hocko | 4bd2c1e | 2012-10-08 16:33:10 -0700 | [diff] [blame] | 61 | #include <net/ip.h> |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 62 | #include <net/tcp_memcontrol.h> |
Qiang Huang | f35c3a8 | 2013-11-12 15:08:22 -0800 | [diff] [blame] | 63 | #include "slab.h" |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 64 | |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 65 | #include <asm/uaccess.h> |
| 66 | |
KOSAKI Motohiro | cc8e970 | 2010-08-09 17:19:57 -0700 | [diff] [blame] | 67 | #include <trace/events/vmscan.h> |
| 68 | |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 69 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
| 70 | EXPORT_SYMBOL(memory_cgrp_subsys); |
David Rientjes | 68ae564 | 2012-12-12 13:51:57 -0800 | [diff] [blame] | 71 | |
KAMEZAWA Hiroyuki | a181b0e | 2008-07-25 01:47:08 -0700 | [diff] [blame] | 72 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
Kirill A. Shutemov | 6bbda35 | 2012-05-29 15:06:55 -0700 | [diff] [blame] | 73 | static struct mem_cgroup *root_mem_cgroup __read_mostly; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 74 | |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 75 | /* Whether the swap controller is active */ |
Andrew Morton | c255a45 | 2012-07-31 16:43:02 -0700 | [diff] [blame] | 76 | #ifdef CONFIG_MEMCG_SWAP |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 77 | int do_swap_account __read_mostly; |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 78 | #else |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 79 | #define do_swap_account 0 |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 80 | #endif |
| 81 | |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 82 | static const char * const mem_cgroup_stat_names[] = { |
| 83 | "cache", |
| 84 | "rss", |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 85 | "rss_huge", |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 86 | "mapped_file", |
Sha Zhengju | 3ea67d0 | 2013-09-12 15:13:53 -0700 | [diff] [blame] | 87 | "writeback", |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 88 | "swap", |
| 89 | }; |
| 90 | |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 91 | static const char * const mem_cgroup_events_names[] = { |
| 92 | "pgpgin", |
| 93 | "pgpgout", |
| 94 | "pgfault", |
| 95 | "pgmajfault", |
| 96 | }; |
| 97 | |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 98 | static const char * const mem_cgroup_lru_names[] = { |
| 99 | "inactive_anon", |
| 100 | "active_anon", |
| 101 | "inactive_file", |
| 102 | "active_file", |
| 103 | "unevictable", |
| 104 | }; |
| 105 | |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 106 | /* |
| 107 | * Per memcg event counter is incremented at every pagein/pageout. With THP, |
| 108 | * it will be incremated by the number of pages. This counter is used for |
| 109 | * for trigger some periodic events. This is straightforward and better |
| 110 | * than using jiffies etc. to handle periodic memcg event. |
| 111 | */ |
| 112 | enum mem_cgroup_events_target { |
| 113 | MEM_CGROUP_TARGET_THRESH, |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 114 | MEM_CGROUP_TARGET_SOFTLIMIT, |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 115 | MEM_CGROUP_TARGET_NUMAINFO, |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 116 | MEM_CGROUP_NTARGETS, |
| 117 | }; |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 118 | #define THRESHOLDS_EVENTS_TARGET 128 |
| 119 | #define SOFTLIMIT_EVENTS_TARGET 1024 |
| 120 | #define NUMAINFO_EVENTS_TARGET 1024 |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 121 | |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 122 | struct mem_cgroup_stat_cpu { |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 123 | long count[MEM_CGROUP_STAT_NSTATS]; |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 124 | unsigned long events[MEMCG_NR_EVENTS]; |
Johannes Weiner | 1311471 | 2012-05-29 15:07:07 -0700 | [diff] [blame] | 125 | unsigned long nr_page_events; |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 126 | unsigned long targets[MEM_CGROUP_NTARGETS]; |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 127 | }; |
| 128 | |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 129 | struct reclaim_iter { |
| 130 | struct mem_cgroup *position; |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 131 | /* scan generation, increased every round-trip */ |
| 132 | unsigned int generation; |
| 133 | }; |
| 134 | |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 135 | /* |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 136 | * per-zone information in memory controller. |
| 137 | */ |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 138 | struct mem_cgroup_per_zone { |
Johannes Weiner | 6290df5 | 2012-01-12 17:18:10 -0800 | [diff] [blame] | 139 | struct lruvec lruvec; |
Hugh Dickins | 1eb4927 | 2012-03-21 16:34:19 -0700 | [diff] [blame] | 140 | unsigned long lru_size[NR_LRU_LISTS]; |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 141 | |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 142 | struct reclaim_iter iter[DEF_PRIORITY + 1]; |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 143 | |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 144 | struct rb_node tree_node; /* RB tree node */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 145 | unsigned long usage_in_excess;/* Set to the value by which */ |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 146 | /* the soft limit is exceeded*/ |
| 147 | bool on_tree; |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 148 | struct mem_cgroup *memcg; /* Back pointer, we cannot */ |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 149 | /* use container_of */ |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 150 | }; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 151 | |
| 152 | struct mem_cgroup_per_node { |
| 153 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; |
| 154 | }; |
| 155 | |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 156 | /* |
| 157 | * Cgroups above their limits are maintained in a RB-Tree, independent of |
| 158 | * their hierarchy representation |
| 159 | */ |
| 160 | |
| 161 | struct mem_cgroup_tree_per_zone { |
| 162 | struct rb_root rb_root; |
| 163 | spinlock_t lock; |
| 164 | }; |
| 165 | |
| 166 | struct mem_cgroup_tree_per_node { |
| 167 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; |
| 168 | }; |
| 169 | |
| 170 | struct mem_cgroup_tree { |
| 171 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; |
| 172 | }; |
| 173 | |
| 174 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; |
| 175 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 176 | struct mem_cgroup_threshold { |
| 177 | struct eventfd_ctx *eventfd; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 178 | unsigned long threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 179 | }; |
| 180 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 181 | /* For threshold */ |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 182 | struct mem_cgroup_threshold_ary { |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 183 | /* An array index points to threshold just below or equal to usage. */ |
Phil Carmody | 5407a56 | 2010-05-26 14:42:42 -0700 | [diff] [blame] | 184 | int current_threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 185 | /* Size of entries[] */ |
| 186 | unsigned int size; |
| 187 | /* Array of thresholds */ |
| 188 | struct mem_cgroup_threshold entries[0]; |
| 189 | }; |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 190 | |
| 191 | struct mem_cgroup_thresholds { |
| 192 | /* Primary thresholds array */ |
| 193 | struct mem_cgroup_threshold_ary *primary; |
| 194 | /* |
| 195 | * Spare threshold array. |
| 196 | * This is needed to make mem_cgroup_unregister_event() "never fail". |
| 197 | * It must be able to store at least primary->size - 1 entries. |
| 198 | */ |
| 199 | struct mem_cgroup_threshold_ary *spare; |
| 200 | }; |
| 201 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 202 | /* for OOM */ |
| 203 | struct mem_cgroup_eventfd_list { |
| 204 | struct list_head list; |
| 205 | struct eventfd_ctx *eventfd; |
| 206 | }; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 207 | |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 208 | /* |
| 209 | * cgroup_event represents events which userspace want to receive. |
| 210 | */ |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 211 | struct mem_cgroup_event { |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 212 | /* |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 213 | * memcg which the event belongs to. |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 214 | */ |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 215 | struct mem_cgroup *memcg; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 216 | /* |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 217 | * eventfd to signal userspace about the event. |
| 218 | */ |
| 219 | struct eventfd_ctx *eventfd; |
| 220 | /* |
| 221 | * Each of these stored in a list by the cgroup. |
| 222 | */ |
| 223 | struct list_head list; |
| 224 | /* |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 225 | * register_event() callback will be used to add new userspace |
| 226 | * waiter for changes related to this event. Use eventfd_signal() |
| 227 | * on eventfd to send notification to userspace. |
| 228 | */ |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 229 | int (*register_event)(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 230 | struct eventfd_ctx *eventfd, const char *args); |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 231 | /* |
| 232 | * unregister_event() callback will be called when userspace closes |
| 233 | * the eventfd or on cgroup removing. This callback must be set, |
| 234 | * if you want provide notification functionality. |
| 235 | */ |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 236 | void (*unregister_event)(struct mem_cgroup *memcg, |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 237 | struct eventfd_ctx *eventfd); |
| 238 | /* |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 239 | * All fields below needed to unregister event when |
| 240 | * userspace closes eventfd. |
| 241 | */ |
| 242 | poll_table pt; |
| 243 | wait_queue_head_t *wqh; |
| 244 | wait_queue_t wait; |
| 245 | struct work_struct remove; |
| 246 | }; |
| 247 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 248 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
| 249 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 250 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 251 | /* |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 252 | * The memory controller data structure. The memory controller controls both |
| 253 | * page cache and RSS per cgroup. We would eventually like to provide |
| 254 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, |
| 255 | * to help the administrator determine what knobs to tune. |
| 256 | * |
| 257 | * TODO: Add a water mark for the memory controller. Reclaim will begin when |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 258 | * we hit the water mark. May be even add a low water mark, such that |
| 259 | * no reclaim occurs from a cgroup at it's low water mark, this is |
| 260 | * a feature that will be implemented much later in the future. |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 261 | */ |
| 262 | struct mem_cgroup { |
| 263 | struct cgroup_subsys_state css; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 264 | |
| 265 | /* Accounted resources */ |
| 266 | struct page_counter memory; |
| 267 | struct page_counter memsw; |
| 268 | struct page_counter kmem; |
| 269 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 270 | /* Normal memory consumption range */ |
| 271 | unsigned long low; |
| 272 | unsigned long high; |
| 273 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 274 | unsigned long soft_limit; |
Hugh Dickins | 59927fb | 2012-03-15 15:17:07 -0700 | [diff] [blame] | 275 | |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 276 | /* vmpressure notifications */ |
| 277 | struct vmpressure vmpressure; |
| 278 | |
Johannes Weiner | 2f7dd7a | 2014-10-02 16:16:57 -0700 | [diff] [blame] | 279 | /* css_online() has been completed */ |
| 280 | int initialized; |
| 281 | |
Li Zefan | 465939a | 2013-07-08 16:00:38 -0700 | [diff] [blame] | 282 | /* |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 283 | * Should the accounting and control be hierarchical, per subtree? |
| 284 | */ |
| 285 | bool use_hierarchy; |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 286 | |
| 287 | bool oom_lock; |
| 288 | atomic_t under_oom; |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 289 | atomic_t oom_wakeups; |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 290 | |
KAMEZAWA Hiroyuki | 1f4c025 | 2011-07-26 16:08:21 -0700 | [diff] [blame] | 291 | int swappiness; |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 292 | /* OOM-Killer disable */ |
| 293 | int oom_kill_disable; |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 294 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 295 | /* protect arrays of thresholds */ |
| 296 | struct mutex thresholds_lock; |
| 297 | |
| 298 | /* thresholds for memory usage. RCU-protected */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 299 | struct mem_cgroup_thresholds thresholds; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 300 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 301 | /* thresholds for mem+swap usage. RCU-protected */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 302 | struct mem_cgroup_thresholds memsw_thresholds; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 303 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 304 | /* For oom notifier event fd */ |
| 305 | struct list_head oom_notify; |
Johannes Weiner | 185efc0 | 2011-09-14 16:21:58 -0700 | [diff] [blame] | 306 | |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 307 | /* |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 308 | * Should we move charges of a task when a task is moved into this |
| 309 | * mem_cgroup ? And what type of charges should we move ? |
| 310 | */ |
Andrew Morton | f894ffa | 2013-09-12 15:13:35 -0700 | [diff] [blame] | 311 | unsigned long move_charge_at_immigrate; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 312 | /* |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 313 | * set > 0 if pages under this cgroup are moving to other cgroup. |
| 314 | */ |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 315 | atomic_t moving_account; |
KAMEZAWA Hiroyuki | 312734c0 | 2012-03-21 16:34:24 -0700 | [diff] [blame] | 316 | /* taken only while moving_account > 0 */ |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 317 | spinlock_t move_lock; |
| 318 | struct task_struct *move_lock_task; |
| 319 | unsigned long move_lock_flags; |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 320 | /* |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 321 | * percpu counter. |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 322 | */ |
Kirill A. Shutemov | 3a7951b | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 323 | struct mem_cgroup_stat_cpu __percpu *stat; |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 324 | /* |
| 325 | * used when a cpu is offlined or other synchronizations |
| 326 | * See mem_cgroup_read_stat(). |
| 327 | */ |
| 328 | struct mem_cgroup_stat_cpu nocpu_base; |
| 329 | spinlock_t pcp_counter_lock; |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 330 | |
Michal Hocko | 4bd2c1e | 2012-10-08 16:33:10 -0700 | [diff] [blame] | 331 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET) |
Eric W. Biederman | 2e685cad | 2013-10-19 16:26:19 -0700 | [diff] [blame] | 332 | struct cg_proto tcp_mem; |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 333 | #endif |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 334 | #if defined(CONFIG_MEMCG_KMEM) |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 335 | /* Index in the kmem_cache->memcg_params.memcg_caches array */ |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 336 | int kmemcg_id; |
Vladimir Davydov | 2788cf0 | 2015-02-12 14:59:38 -0800 | [diff] [blame] | 337 | bool kmem_acct_activated; |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 338 | bool kmem_acct_active; |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 339 | #endif |
Glauber Costa | 45cf7eb | 2013-02-22 16:34:49 -0800 | [diff] [blame] | 340 | |
| 341 | int last_scanned_node; |
| 342 | #if MAX_NUMNODES > 1 |
| 343 | nodemask_t scan_nodes; |
| 344 | atomic_t numainfo_events; |
| 345 | atomic_t numainfo_updating; |
| 346 | #endif |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 347 | |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 348 | /* List of events which userspace want to receive */ |
| 349 | struct list_head event_list; |
| 350 | spinlock_t event_list_lock; |
| 351 | |
Johannes Weiner | 54f72fe | 2013-07-08 15:59:49 -0700 | [diff] [blame] | 352 | struct mem_cgroup_per_node *nodeinfo[0]; |
| 353 | /* WARNING: nodeinfo must be the last member here */ |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 354 | }; |
| 355 | |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 356 | #ifdef CONFIG_MEMCG_KMEM |
Vladimir Davydov | cb731d6 | 2015-02-12 14:58:54 -0800 | [diff] [blame] | 357 | bool memcg_kmem_is_active(struct mem_cgroup *memcg) |
Glauber Costa | 7de3768 | 2012-12-18 14:22:07 -0800 | [diff] [blame] | 358 | { |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 359 | return memcg->kmem_acct_active; |
Glauber Costa | 7de3768 | 2012-12-18 14:22:07 -0800 | [diff] [blame] | 360 | } |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 361 | #endif |
| 362 | |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 363 | /* Stuffs for move charges at task migration. */ |
| 364 | /* |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 365 | * Types of charges to be moved. |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 366 | */ |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 367 | #define MOVE_ANON 0x1U |
| 368 | #define MOVE_FILE 0x2U |
| 369 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 370 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 371 | /* "mc" and its members are protected by cgroup_mutex */ |
| 372 | static struct move_charge_struct { |
Daisuke Nishimura | b1dd693 | 2010-11-24 12:57:06 -0800 | [diff] [blame] | 373 | spinlock_t lock; /* for from, to */ |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 374 | struct mem_cgroup *from; |
| 375 | struct mem_cgroup *to; |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 376 | unsigned long flags; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 377 | unsigned long precharge; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 378 | unsigned long moved_charge; |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 379 | unsigned long moved_swap; |
Daisuke Nishimura | 8033b97 | 2010-03-10 15:22:16 -0800 | [diff] [blame] | 380 | struct task_struct *moving_task; /* a task moving charges */ |
| 381 | wait_queue_head_t waitq; /* a waitq for other context */ |
| 382 | } mc = { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 383 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
Daisuke Nishimura | 8033b97 | 2010-03-10 15:22:16 -0800 | [diff] [blame] | 384 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
| 385 | }; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 386 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 387 | /* |
| 388 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft |
| 389 | * limit reclaim to prevent infinite loops, if they ever occur. |
| 390 | */ |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 391 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 392 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 393 | |
KAMEZAWA Hiroyuki | 217bc31 | 2008-02-07 00:14:17 -0800 | [diff] [blame] | 394 | enum charge_type { |
| 395 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, |
Kamezawa Hiroyuki | 41326c1 | 2012-07-31 16:41:40 -0700 | [diff] [blame] | 396 | MEM_CGROUP_CHARGE_TYPE_ANON, |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 397 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
KAMEZAWA Hiroyuki | 8a9478c | 2009-06-17 16:27:17 -0700 | [diff] [blame] | 398 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
KAMEZAWA Hiroyuki | c05555b | 2008-10-18 20:28:11 -0700 | [diff] [blame] | 399 | NR_CHARGE_TYPE, |
| 400 | }; |
| 401 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 402 | /* for encoding cft->private value on file */ |
Glauber Costa | 86ae53e | 2012-12-18 14:21:45 -0800 | [diff] [blame] | 403 | enum res_type { |
| 404 | _MEM, |
| 405 | _MEMSWAP, |
| 406 | _OOM_TYPE, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 407 | _KMEM, |
Glauber Costa | 86ae53e | 2012-12-18 14:21:45 -0800 | [diff] [blame] | 408 | }; |
| 409 | |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 410 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
| 411 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 412 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 413 | /* Used for OOM nofiier */ |
| 414 | #define OOM_CONTROL (0) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 415 | |
Balbir Singh | 75822b4 | 2009-09-23 15:56:38 -0700 | [diff] [blame] | 416 | /* |
Glauber Costa | 0999821 | 2013-02-22 16:34:55 -0800 | [diff] [blame] | 417 | * The memcg_create_mutex will be held whenever a new cgroup is created. |
| 418 | * As a consequence, any change that needs to protect against new child cgroups |
| 419 | * appearing has to hold it as well. |
| 420 | */ |
| 421 | static DEFINE_MUTEX(memcg_create_mutex); |
| 422 | |
Wanpeng Li | b214514 | 2012-07-31 16:46:01 -0700 | [diff] [blame] | 423 | struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s) |
| 424 | { |
Tejun Heo | a7c6d55 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 425 | return s ? container_of(s, struct mem_cgroup, css) : NULL; |
Wanpeng Li | b214514 | 2012-07-31 16:46:01 -0700 | [diff] [blame] | 426 | } |
| 427 | |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 428 | /* Some nice accessors for the vmpressure. */ |
| 429 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) |
| 430 | { |
| 431 | if (!memcg) |
| 432 | memcg = root_mem_cgroup; |
| 433 | return &memcg->vmpressure; |
| 434 | } |
| 435 | |
| 436 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) |
| 437 | { |
| 438 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; |
| 439 | } |
| 440 | |
Michal Hocko | 7ffc0ed | 2012-10-08 16:33:13 -0700 | [diff] [blame] | 441 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
| 442 | { |
| 443 | return (memcg == root_mem_cgroup); |
| 444 | } |
| 445 | |
Li Zefan | 4219b2d | 2013-09-23 16:56:29 +0800 | [diff] [blame] | 446 | /* |
| 447 | * We restrict the id in the range of [1, 65535], so it can fit into |
| 448 | * an unsigned short. |
| 449 | */ |
| 450 | #define MEM_CGROUP_ID_MAX USHRT_MAX |
| 451 | |
Li Zefan | 34c00c3 | 2013-09-23 16:56:01 +0800 | [diff] [blame] | 452 | static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) |
| 453 | { |
Tejun Heo | 15a4c83 | 2014-05-04 15:09:14 -0400 | [diff] [blame] | 454 | return memcg->css.id; |
Li Zefan | 34c00c3 | 2013-09-23 16:56:01 +0800 | [diff] [blame] | 455 | } |
| 456 | |
| 457 | static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) |
| 458 | { |
| 459 | struct cgroup_subsys_state *css; |
| 460 | |
Tejun Heo | 7d699dd | 2014-05-04 15:09:13 -0400 | [diff] [blame] | 461 | css = css_from_id(id, &memory_cgrp_subsys); |
Li Zefan | 34c00c3 | 2013-09-23 16:56:01 +0800 | [diff] [blame] | 462 | return mem_cgroup_from_css(css); |
| 463 | } |
| 464 | |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 465 | /* Writing them here to avoid exposing memcg's inner layout */ |
Michal Hocko | 4bd2c1e | 2012-10-08 16:33:10 -0700 | [diff] [blame] | 466 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 467 | |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 468 | void sock_update_memcg(struct sock *sk) |
| 469 | { |
Glauber Costa | 376be5f | 2012-01-20 04:57:14 +0000 | [diff] [blame] | 470 | if (mem_cgroup_sockets_enabled) { |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 471 | struct mem_cgroup *memcg; |
Glauber Costa | 3f13461 | 2012-05-29 15:07:11 -0700 | [diff] [blame] | 472 | struct cg_proto *cg_proto; |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 473 | |
| 474 | BUG_ON(!sk->sk_prot->proto_cgroup); |
| 475 | |
Glauber Costa | f3f511e | 2012-01-05 20:16:39 +0000 | [diff] [blame] | 476 | /* Socket cloning can throw us here with sk_cgrp already |
| 477 | * filled. It won't however, necessarily happen from |
| 478 | * process context. So the test for root memcg given |
| 479 | * the current task's memcg won't help us in this case. |
| 480 | * |
| 481 | * Respecting the original socket's memcg is a better |
| 482 | * decision in this case. |
| 483 | */ |
| 484 | if (sk->sk_cgrp) { |
| 485 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); |
Li Zefan | 5347e5a | 2013-07-08 16:00:30 -0700 | [diff] [blame] | 486 | css_get(&sk->sk_cgrp->memcg->css); |
Glauber Costa | f3f511e | 2012-01-05 20:16:39 +0000 | [diff] [blame] | 487 | return; |
| 488 | } |
| 489 | |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 490 | rcu_read_lock(); |
| 491 | memcg = mem_cgroup_from_task(current); |
Glauber Costa | 3f13461 | 2012-05-29 15:07:11 -0700 | [diff] [blame] | 492 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
Li Zefan | 5347e5a | 2013-07-08 16:00:30 -0700 | [diff] [blame] | 493 | if (!mem_cgroup_is_root(memcg) && |
Tejun Heo | ec903c0 | 2014-05-13 12:11:01 -0400 | [diff] [blame] | 494 | memcg_proto_active(cg_proto) && |
| 495 | css_tryget_online(&memcg->css)) { |
Glauber Costa | 3f13461 | 2012-05-29 15:07:11 -0700 | [diff] [blame] | 496 | sk->sk_cgrp = cg_proto; |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 497 | } |
| 498 | rcu_read_unlock(); |
| 499 | } |
| 500 | } |
| 501 | EXPORT_SYMBOL(sock_update_memcg); |
| 502 | |
| 503 | void sock_release_memcg(struct sock *sk) |
| 504 | { |
Glauber Costa | 376be5f | 2012-01-20 04:57:14 +0000 | [diff] [blame] | 505 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 506 | struct mem_cgroup *memcg; |
| 507 | WARN_ON(!sk->sk_cgrp->memcg); |
| 508 | memcg = sk->sk_cgrp->memcg; |
Li Zefan | 5347e5a | 2013-07-08 16:00:30 -0700 | [diff] [blame] | 509 | css_put(&sk->sk_cgrp->memcg->css); |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 510 | } |
| 511 | } |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 512 | |
| 513 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) |
| 514 | { |
| 515 | if (!memcg || mem_cgroup_is_root(memcg)) |
| 516 | return NULL; |
| 517 | |
Eric W. Biederman | 2e685cad | 2013-10-19 16:26:19 -0700 | [diff] [blame] | 518 | return &memcg->tcp_mem; |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 519 | } |
| 520 | EXPORT_SYMBOL(tcp_proto_cgroup); |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 521 | |
Glauber Costa | 3f13461 | 2012-05-29 15:07:11 -0700 | [diff] [blame] | 522 | #endif |
| 523 | |
Glauber Costa | a8964b9 | 2012-12-18 14:22:09 -0800 | [diff] [blame] | 524 | #ifdef CONFIG_MEMCG_KMEM |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 525 | /* |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 526 | * This will be the memcg's index in each cache's ->memcg_params.memcg_caches. |
Li Zefan | b862783 | 2013-09-23 16:56:47 +0800 | [diff] [blame] | 527 | * The main reason for not using cgroup id for this: |
| 528 | * this works better in sparse environments, where we have a lot of memcgs, |
| 529 | * but only a few kmem-limited. Or also, if we have, for instance, 200 |
| 530 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a |
| 531 | * 200 entry array for that. |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 532 | * |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 533 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
| 534 | * will double each time we have to increase it. |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 535 | */ |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 536 | static DEFINE_IDA(memcg_cache_ida); |
| 537 | int memcg_nr_cache_ids; |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 538 | |
Vladimir Davydov | 05257a1 | 2015-02-12 14:59:01 -0800 | [diff] [blame] | 539 | /* Protects memcg_nr_cache_ids */ |
| 540 | static DECLARE_RWSEM(memcg_cache_ids_sem); |
| 541 | |
| 542 | void memcg_get_cache_ids(void) |
| 543 | { |
| 544 | down_read(&memcg_cache_ids_sem); |
| 545 | } |
| 546 | |
| 547 | void memcg_put_cache_ids(void) |
| 548 | { |
| 549 | up_read(&memcg_cache_ids_sem); |
| 550 | } |
| 551 | |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 552 | /* |
| 553 | * MIN_SIZE is different than 1, because we would like to avoid going through |
| 554 | * the alloc/free process all the time. In a small machine, 4 kmem-limited |
| 555 | * cgroups is a reasonable guess. In the future, it could be a parameter or |
| 556 | * tunable, but that is strictly not necessary. |
| 557 | * |
Li Zefan | b862783 | 2013-09-23 16:56:47 +0800 | [diff] [blame] | 558 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 559 | * this constant directly from cgroup, but it is understandable that this is |
| 560 | * better kept as an internal representation in cgroup.c. In any case, the |
Li Zefan | b862783 | 2013-09-23 16:56:47 +0800 | [diff] [blame] | 561 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 562 | * increase ours as well if it increases. |
| 563 | */ |
| 564 | #define MEMCG_CACHES_MIN_SIZE 4 |
Li Zefan | b862783 | 2013-09-23 16:56:47 +0800 | [diff] [blame] | 565 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 566 | |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 567 | /* |
| 568 | * A lot of the calls to the cache allocation functions are expected to be |
| 569 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are |
| 570 | * conditional to this static branch, we'll have to allow modules that does |
| 571 | * kmem_cache_alloc and the such to see this symbol as well |
| 572 | */ |
Glauber Costa | a8964b9 | 2012-12-18 14:22:09 -0800 | [diff] [blame] | 573 | struct static_key memcg_kmem_enabled_key; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 574 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
Glauber Costa | a8964b9 | 2012-12-18 14:22:09 -0800 | [diff] [blame] | 575 | |
Glauber Costa | a8964b9 | 2012-12-18 14:22:09 -0800 | [diff] [blame] | 576 | #endif /* CONFIG_MEMCG_KMEM */ |
| 577 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 578 | static struct mem_cgroup_per_zone * |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 579 | mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone) |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 580 | { |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 581 | int nid = zone_to_nid(zone); |
| 582 | int zid = zone_idx(zone); |
| 583 | |
Johannes Weiner | 54f72fe | 2013-07-08 15:59:49 -0700 | [diff] [blame] | 584 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 585 | } |
| 586 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 587 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
Wu Fengguang | d324236 | 2009-12-16 12:19:59 +0100 | [diff] [blame] | 588 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 589 | return &memcg->css; |
Wu Fengguang | d324236 | 2009-12-16 12:19:59 +0100 | [diff] [blame] | 590 | } |
| 591 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 592 | static struct mem_cgroup_per_zone * |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 593 | mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 594 | { |
Johannes Weiner | 97a6c37 | 2011-03-23 16:42:27 -0700 | [diff] [blame] | 595 | int nid = page_to_nid(page); |
| 596 | int zid = page_zonenum(page); |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 597 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 598 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 599 | } |
| 600 | |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 601 | static struct mem_cgroup_tree_per_zone * |
| 602 | soft_limit_tree_node_zone(int nid, int zid) |
| 603 | { |
| 604 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; |
| 605 | } |
| 606 | |
| 607 | static struct mem_cgroup_tree_per_zone * |
| 608 | soft_limit_tree_from_page(struct page *page) |
| 609 | { |
| 610 | int nid = page_to_nid(page); |
| 611 | int zid = page_zonenum(page); |
| 612 | |
| 613 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; |
| 614 | } |
| 615 | |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 616 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz, |
| 617 | struct mem_cgroup_tree_per_zone *mctz, |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 618 | unsigned long new_usage_in_excess) |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 619 | { |
| 620 | struct rb_node **p = &mctz->rb_root.rb_node; |
| 621 | struct rb_node *parent = NULL; |
| 622 | struct mem_cgroup_per_zone *mz_node; |
| 623 | |
| 624 | if (mz->on_tree) |
| 625 | return; |
| 626 | |
| 627 | mz->usage_in_excess = new_usage_in_excess; |
| 628 | if (!mz->usage_in_excess) |
| 629 | return; |
| 630 | while (*p) { |
| 631 | parent = *p; |
| 632 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, |
| 633 | tree_node); |
| 634 | if (mz->usage_in_excess < mz_node->usage_in_excess) |
| 635 | p = &(*p)->rb_left; |
| 636 | /* |
| 637 | * We can't avoid mem cgroups that are over their soft |
| 638 | * limit by the same amount |
| 639 | */ |
| 640 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) |
| 641 | p = &(*p)->rb_right; |
| 642 | } |
| 643 | rb_link_node(&mz->tree_node, parent, p); |
| 644 | rb_insert_color(&mz->tree_node, &mctz->rb_root); |
| 645 | mz->on_tree = true; |
| 646 | } |
| 647 | |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 648 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
| 649 | struct mem_cgroup_tree_per_zone *mctz) |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 650 | { |
| 651 | if (!mz->on_tree) |
| 652 | return; |
| 653 | rb_erase(&mz->tree_node, &mctz->rb_root); |
| 654 | mz->on_tree = false; |
| 655 | } |
| 656 | |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 657 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
| 658 | struct mem_cgroup_tree_per_zone *mctz) |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 659 | { |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 660 | unsigned long flags; |
| 661 | |
| 662 | spin_lock_irqsave(&mctz->lock, flags); |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 663 | __mem_cgroup_remove_exceeded(mz, mctz); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 664 | spin_unlock_irqrestore(&mctz->lock, flags); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 665 | } |
| 666 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 667 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
| 668 | { |
| 669 | unsigned long nr_pages = page_counter_read(&memcg->memory); |
| 670 | unsigned long soft_limit = ACCESS_ONCE(memcg->soft_limit); |
| 671 | unsigned long excess = 0; |
| 672 | |
| 673 | if (nr_pages > soft_limit) |
| 674 | excess = nr_pages - soft_limit; |
| 675 | |
| 676 | return excess; |
| 677 | } |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 678 | |
| 679 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) |
| 680 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 681 | unsigned long excess; |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 682 | struct mem_cgroup_per_zone *mz; |
| 683 | struct mem_cgroup_tree_per_zone *mctz; |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 684 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 685 | mctz = soft_limit_tree_from_page(page); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 686 | /* |
| 687 | * Necessary to update all ancestors when hierarchy is used. |
| 688 | * because their event counter is not touched. |
| 689 | */ |
| 690 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 691 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 692 | excess = soft_limit_excess(memcg); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 693 | /* |
| 694 | * We have to update the tree if mz is on RB-tree or |
| 695 | * mem is over its softlimit. |
| 696 | */ |
| 697 | if (excess || mz->on_tree) { |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 698 | unsigned long flags; |
| 699 | |
| 700 | spin_lock_irqsave(&mctz->lock, flags); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 701 | /* if on-tree, remove it */ |
| 702 | if (mz->on_tree) |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 703 | __mem_cgroup_remove_exceeded(mz, mctz); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 704 | /* |
| 705 | * Insert again. mz->usage_in_excess will be updated. |
| 706 | * If excess is 0, no tree ops. |
| 707 | */ |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 708 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 709 | spin_unlock_irqrestore(&mctz->lock, flags); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 710 | } |
| 711 | } |
| 712 | } |
| 713 | |
| 714 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) |
| 715 | { |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 716 | struct mem_cgroup_tree_per_zone *mctz; |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 717 | struct mem_cgroup_per_zone *mz; |
| 718 | int nid, zid; |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 719 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 720 | for_each_node(nid) { |
| 721 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
| 722 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
| 723 | mctz = soft_limit_tree_node_zone(nid, zid); |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 724 | mem_cgroup_remove_exceeded(mz, mctz); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 725 | } |
| 726 | } |
| 727 | } |
| 728 | |
| 729 | static struct mem_cgroup_per_zone * |
| 730 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) |
| 731 | { |
| 732 | struct rb_node *rightmost = NULL; |
| 733 | struct mem_cgroup_per_zone *mz; |
| 734 | |
| 735 | retry: |
| 736 | mz = NULL; |
| 737 | rightmost = rb_last(&mctz->rb_root); |
| 738 | if (!rightmost) |
| 739 | goto done; /* Nothing to reclaim from */ |
| 740 | |
| 741 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); |
| 742 | /* |
| 743 | * Remove the node now but someone else can add it back, |
| 744 | * we will to add it back at the end of reclaim to its correct |
| 745 | * position in the tree. |
| 746 | */ |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 747 | __mem_cgroup_remove_exceeded(mz, mctz); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 748 | if (!soft_limit_excess(mz->memcg) || |
Tejun Heo | ec903c0 | 2014-05-13 12:11:01 -0400 | [diff] [blame] | 749 | !css_tryget_online(&mz->memcg->css)) |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 750 | goto retry; |
| 751 | done: |
| 752 | return mz; |
| 753 | } |
| 754 | |
| 755 | static struct mem_cgroup_per_zone * |
| 756 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) |
| 757 | { |
| 758 | struct mem_cgroup_per_zone *mz; |
| 759 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 760 | spin_lock_irq(&mctz->lock); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 761 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 762 | spin_unlock_irq(&mctz->lock); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 763 | return mz; |
| 764 | } |
| 765 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 766 | /* |
| 767 | * Implementation Note: reading percpu statistics for memcg. |
| 768 | * |
| 769 | * Both of vmstat[] and percpu_counter has threshold and do periodic |
| 770 | * synchronization to implement "quick" read. There are trade-off between |
| 771 | * reading cost and precision of value. Then, we may have a chance to implement |
| 772 | * a periodic synchronizion of counter in memcg's counter. |
| 773 | * |
| 774 | * But this _read() function is used for user interface now. The user accounts |
| 775 | * memory usage by memory cgroup and he _always_ requires exact value because |
| 776 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always |
| 777 | * have to visit all online cpus and make sum. So, for now, unnecessary |
| 778 | * synchronization is not implemented. (just implemented for cpu hotplug) |
| 779 | * |
| 780 | * If there are kernel internal actions which can make use of some not-exact |
| 781 | * value, and reading all cpu value can be performance bottleneck in some |
| 782 | * common workload, threashold and synchonization as vmstat[] should be |
| 783 | * implemented. |
| 784 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 785 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 786 | enum mem_cgroup_stat_index idx) |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 787 | { |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 788 | long val = 0; |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 789 | int cpu; |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 790 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 791 | get_online_cpus(); |
| 792 | for_each_online_cpu(cpu) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 793 | val += per_cpu(memcg->stat->count[idx], cpu); |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 794 | #ifdef CONFIG_HOTPLUG_CPU |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 795 | spin_lock(&memcg->pcp_counter_lock); |
| 796 | val += memcg->nocpu_base.count[idx]; |
| 797 | spin_unlock(&memcg->pcp_counter_lock); |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 798 | #endif |
| 799 | put_online_cpus(); |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 800 | return val; |
| 801 | } |
| 802 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 803 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 804 | enum mem_cgroup_events_index idx) |
| 805 | { |
| 806 | unsigned long val = 0; |
| 807 | int cpu; |
| 808 | |
David Rientjes | 9c56751 | 2013-10-16 13:46:43 -0700 | [diff] [blame] | 809 | get_online_cpus(); |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 810 | for_each_online_cpu(cpu) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 811 | val += per_cpu(memcg->stat->events[idx], cpu); |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 812 | #ifdef CONFIG_HOTPLUG_CPU |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 813 | spin_lock(&memcg->pcp_counter_lock); |
| 814 | val += memcg->nocpu_base.events[idx]; |
| 815 | spin_unlock(&memcg->pcp_counter_lock); |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 816 | #endif |
David Rientjes | 9c56751 | 2013-10-16 13:46:43 -0700 | [diff] [blame] | 817 | put_online_cpus(); |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 818 | return val; |
| 819 | } |
| 820 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 821 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 822 | struct page *page, |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 823 | int nr_pages) |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 824 | { |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 825 | /* |
| 826 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is |
| 827 | * counted as CACHE even if it's on ANON LRU. |
| 828 | */ |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 829 | if (PageAnon(page)) |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 830 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 831 | nr_pages); |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 832 | else |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 833 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 834 | nr_pages); |
Balaji Rao | 55e462b | 2008-05-01 04:35:12 -0700 | [diff] [blame] | 835 | |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 836 | if (PageTransHuge(page)) |
| 837 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
| 838 | nr_pages); |
| 839 | |
KAMEZAWA Hiroyuki | e401f176 | 2011-01-20 14:44:23 -0800 | [diff] [blame] | 840 | /* pagein of a big page is an event. So, ignore page size */ |
| 841 | if (nr_pages > 0) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 842 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
KAMEZAWA Hiroyuki | 3751d60 | 2011-02-01 15:52:45 -0800 | [diff] [blame] | 843 | else { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 844 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
KAMEZAWA Hiroyuki | 3751d60 | 2011-02-01 15:52:45 -0800 | [diff] [blame] | 845 | nr_pages = -nr_pages; /* for event */ |
| 846 | } |
KAMEZAWA Hiroyuki | e401f176 | 2011-01-20 14:44:23 -0800 | [diff] [blame] | 847 | |
Johannes Weiner | 1311471 | 2012-05-29 15:07:07 -0700 | [diff] [blame] | 848 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 849 | } |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 850 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 851 | unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
Konstantin Khlebnikov | 074291f | 2012-05-29 15:07:00 -0700 | [diff] [blame] | 852 | { |
| 853 | struct mem_cgroup_per_zone *mz; |
| 854 | |
| 855 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
| 856 | return mz->lru_size[lru]; |
| 857 | } |
| 858 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 859 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
| 860 | int nid, |
| 861 | unsigned int lru_mask) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 862 | { |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 863 | unsigned long nr = 0; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 864 | int zid; |
| 865 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 866 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 867 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 868 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
| 869 | struct mem_cgroup_per_zone *mz; |
| 870 | enum lru_list lru; |
| 871 | |
| 872 | for_each_lru(lru) { |
| 873 | if (!(BIT(lru) & lru_mask)) |
| 874 | continue; |
| 875 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
| 876 | nr += mz->lru_size[lru]; |
| 877 | } |
| 878 | } |
| 879 | return nr; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 880 | } |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 881 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 882 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 883 | unsigned int lru_mask) |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 884 | { |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 885 | unsigned long nr = 0; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 886 | int nid; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 887 | |
Lai Jiangshan | 31aaea4 | 2012-12-12 13:51:27 -0800 | [diff] [blame] | 888 | for_each_node_state(nid, N_MEMORY) |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 889 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
| 890 | return nr; |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 891 | } |
| 892 | |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 893 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
| 894 | enum mem_cgroup_events_target target) |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 895 | { |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 896 | unsigned long val, next; |
| 897 | |
Johannes Weiner | 1311471 | 2012-05-29 15:07:07 -0700 | [diff] [blame] | 898 | val = __this_cpu_read(memcg->stat->nr_page_events); |
Steven Rostedt | 4799401 | 2011-11-02 13:38:33 -0700 | [diff] [blame] | 899 | next = __this_cpu_read(memcg->stat->targets[target]); |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 900 | /* from time_after() in jiffies.h */ |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 901 | if ((long)next - (long)val < 0) { |
| 902 | switch (target) { |
| 903 | case MEM_CGROUP_TARGET_THRESH: |
| 904 | next = val + THRESHOLDS_EVENTS_TARGET; |
| 905 | break; |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 906 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
| 907 | next = val + SOFTLIMIT_EVENTS_TARGET; |
| 908 | break; |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 909 | case MEM_CGROUP_TARGET_NUMAINFO: |
| 910 | next = val + NUMAINFO_EVENTS_TARGET; |
| 911 | break; |
| 912 | default: |
| 913 | break; |
| 914 | } |
| 915 | __this_cpu_write(memcg->stat->targets[target], next); |
| 916 | return true; |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 917 | } |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 918 | return false; |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 919 | } |
| 920 | |
| 921 | /* |
| 922 | * Check events in order. |
| 923 | * |
| 924 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 925 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 926 | { |
| 927 | /* threshold event is triggered in finer grain than soft limit */ |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 928 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
| 929 | MEM_CGROUP_TARGET_THRESH))) { |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 930 | bool do_softlimit; |
Andrew Morton | 82b3f2a | 2012-02-03 15:37:14 -0800 | [diff] [blame] | 931 | bool do_numainfo __maybe_unused; |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 932 | |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 933 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
| 934 | MEM_CGROUP_TARGET_SOFTLIMIT); |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 935 | #if MAX_NUMNODES > 1 |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 936 | do_numainfo = mem_cgroup_event_ratelimit(memcg, |
| 937 | MEM_CGROUP_TARGET_NUMAINFO); |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 938 | #endif |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 939 | mem_cgroup_threshold(memcg); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 940 | if (unlikely(do_softlimit)) |
| 941 | mem_cgroup_update_tree(memcg, page); |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 942 | #if MAX_NUMNODES > 1 |
| 943 | if (unlikely(do_numainfo)) |
| 944 | atomic_inc(&memcg->numainfo_events); |
| 945 | #endif |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 946 | } |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 947 | } |
| 948 | |
Balbir Singh | cf475ad | 2008-04-29 01:00:16 -0700 | [diff] [blame] | 949 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 950 | { |
Balbir Singh | 31a78f2 | 2008-09-28 23:09:31 +0100 | [diff] [blame] | 951 | /* |
| 952 | * mm_update_next_owner() may clear mm->owner to NULL |
| 953 | * if it races with swapoff, page migration, etc. |
| 954 | * So this can be called with p == NULL. |
| 955 | */ |
| 956 | if (unlikely(!p)) |
| 957 | return NULL; |
| 958 | |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 959 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 960 | } |
| 961 | |
Johannes Weiner | df38197 | 2014-04-07 15:37:43 -0700 | [diff] [blame] | 962 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 963 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 964 | struct mem_cgroup *memcg = NULL; |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 965 | |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 966 | rcu_read_lock(); |
| 967 | do { |
Michal Hocko | 6f6acb0 | 2014-05-22 11:54:19 -0700 | [diff] [blame] | 968 | /* |
| 969 | * Page cache insertions can happen withou an |
| 970 | * actual mm context, e.g. during disk probing |
| 971 | * on boot, loopback IO, acct() writes etc. |
| 972 | */ |
| 973 | if (unlikely(!mm)) |
Johannes Weiner | df38197 | 2014-04-07 15:37:43 -0700 | [diff] [blame] | 974 | memcg = root_mem_cgroup; |
Michal Hocko | 6f6acb0 | 2014-05-22 11:54:19 -0700 | [diff] [blame] | 975 | else { |
| 976 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
| 977 | if (unlikely(!memcg)) |
| 978 | memcg = root_mem_cgroup; |
| 979 | } |
Tejun Heo | ec903c0 | 2014-05-13 12:11:01 -0400 | [diff] [blame] | 980 | } while (!css_tryget_online(&memcg->css)); |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 981 | rcu_read_unlock(); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 982 | return memcg; |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 983 | } |
| 984 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 985 | /** |
| 986 | * mem_cgroup_iter - iterate over memory cgroup hierarchy |
| 987 | * @root: hierarchy root |
| 988 | * @prev: previously returned memcg, NULL on first invocation |
| 989 | * @reclaim: cookie for shared reclaim walks, NULL for full walks |
| 990 | * |
| 991 | * Returns references to children of the hierarchy below @root, or |
| 992 | * @root itself, or %NULL after a full round-trip. |
| 993 | * |
| 994 | * Caller must pass the return value in @prev on subsequent |
| 995 | * invocations for reference counting, or use mem_cgroup_iter_break() |
| 996 | * to cancel a hierarchy walk before the round-trip is complete. |
| 997 | * |
| 998 | * Reclaimers can specify a zone and a priority level in @reclaim to |
| 999 | * divide up the memcgs in the hierarchy among all concurrent |
| 1000 | * reclaimers operating on the same zone and priority. |
| 1001 | */ |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 1002 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1003 | struct mem_cgroup *prev, |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 1004 | struct mem_cgroup_reclaim_cookie *reclaim) |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 1005 | { |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1006 | struct reclaim_iter *uninitialized_var(iter); |
| 1007 | struct cgroup_subsys_state *css = NULL; |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1008 | struct mem_cgroup *memcg = NULL; |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1009 | struct mem_cgroup *pos = NULL; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1010 | |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 1011 | if (mem_cgroup_disabled()) |
| 1012 | return NULL; |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1013 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1014 | if (!root) |
| 1015 | root = root_mem_cgroup; |
| 1016 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1017 | if (prev && !reclaim) |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1018 | pos = prev; |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1019 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1020 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
| 1021 | if (prev) |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1022 | goto out; |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 1023 | return root; |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1024 | } |
| 1025 | |
Michal Hocko | 542f85f | 2013-04-29 15:07:15 -0700 | [diff] [blame] | 1026 | rcu_read_lock(); |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1027 | |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1028 | if (reclaim) { |
| 1029 | struct mem_cgroup_per_zone *mz; |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 1030 | |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1031 | mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone); |
| 1032 | iter = &mz->iter[reclaim->priority]; |
Michal Hocko | 5f57816 | 2013-04-29 15:07:17 -0700 | [diff] [blame] | 1033 | |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1034 | if (prev && reclaim->generation != iter->generation) |
Michal Hocko | 542f85f | 2013-04-29 15:07:15 -0700 | [diff] [blame] | 1035 | goto out_unlock; |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1036 | |
| 1037 | do { |
| 1038 | pos = ACCESS_ONCE(iter->position); |
| 1039 | /* |
| 1040 | * A racing update may change the position and |
| 1041 | * put the last reference, hence css_tryget(), |
| 1042 | * or retry to see the updated position. |
| 1043 | */ |
| 1044 | } while (pos && !css_tryget(&pos->css)); |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1045 | } |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1046 | |
| 1047 | if (pos) |
| 1048 | css = &pos->css; |
| 1049 | |
| 1050 | for (;;) { |
| 1051 | css = css_next_descendant_pre(css, &root->css); |
| 1052 | if (!css) { |
| 1053 | /* |
| 1054 | * Reclaimers share the hierarchy walk, and a |
| 1055 | * new one might jump in right at the end of |
| 1056 | * the hierarchy - make sure they see at least |
| 1057 | * one group and restart from the beginning. |
| 1058 | */ |
| 1059 | if (!prev) |
| 1060 | continue; |
| 1061 | break; |
| 1062 | } |
| 1063 | |
| 1064 | /* |
| 1065 | * Verify the css and acquire a reference. The root |
| 1066 | * is provided by the caller, so we know it's alive |
| 1067 | * and kicking, and don't take an extra reference. |
| 1068 | */ |
| 1069 | memcg = mem_cgroup_from_css(css); |
| 1070 | |
| 1071 | if (css == &root->css) |
| 1072 | break; |
| 1073 | |
Johannes Weiner | b205256 | 2014-12-10 15:42:48 -0800 | [diff] [blame] | 1074 | if (css_tryget(css)) { |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1075 | /* |
| 1076 | * Make sure the memcg is initialized: |
| 1077 | * mem_cgroup_css_online() orders the the |
| 1078 | * initialization against setting the flag. |
| 1079 | */ |
| 1080 | if (smp_load_acquire(&memcg->initialized)) |
| 1081 | break; |
| 1082 | |
| 1083 | css_put(css); |
| 1084 | } |
| 1085 | |
| 1086 | memcg = NULL; |
| 1087 | } |
| 1088 | |
| 1089 | if (reclaim) { |
| 1090 | if (cmpxchg(&iter->position, pos, memcg) == pos) { |
| 1091 | if (memcg) |
| 1092 | css_get(&memcg->css); |
| 1093 | if (pos) |
| 1094 | css_put(&pos->css); |
| 1095 | } |
| 1096 | |
| 1097 | /* |
| 1098 | * pairs with css_tryget when dereferencing iter->position |
| 1099 | * above. |
| 1100 | */ |
| 1101 | if (pos) |
| 1102 | css_put(&pos->css); |
| 1103 | |
| 1104 | if (!memcg) |
| 1105 | iter->generation++; |
| 1106 | else if (!prev) |
| 1107 | reclaim->generation = iter->generation; |
| 1108 | } |
| 1109 | |
Michal Hocko | 542f85f | 2013-04-29 15:07:15 -0700 | [diff] [blame] | 1110 | out_unlock: |
| 1111 | rcu_read_unlock(); |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1112 | out: |
Michal Hocko | c40046f | 2013-04-29 15:07:14 -0700 | [diff] [blame] | 1113 | if (prev && prev != root) |
| 1114 | css_put(&prev->css); |
| 1115 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1116 | return memcg; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1117 | } |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1118 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1119 | /** |
| 1120 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely |
| 1121 | * @root: hierarchy root |
| 1122 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() |
| 1123 | */ |
| 1124 | void mem_cgroup_iter_break(struct mem_cgroup *root, |
| 1125 | struct mem_cgroup *prev) |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1126 | { |
| 1127 | if (!root) |
| 1128 | root = root_mem_cgroup; |
| 1129 | if (prev && prev != root) |
| 1130 | css_put(&prev->css); |
| 1131 | } |
| 1132 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1133 | /* |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1134 | * Iteration constructs for visiting all cgroups (under a tree). If |
| 1135 | * loops are exited prematurely (break), mem_cgroup_iter_break() must |
| 1136 | * be used for reference counting. |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1137 | */ |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1138 | #define for_each_mem_cgroup_tree(iter, root) \ |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 1139 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1140 | iter != NULL; \ |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 1141 | iter = mem_cgroup_iter(root, iter, NULL)) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1142 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1143 | #define for_each_mem_cgroup(iter) \ |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 1144 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1145 | iter != NULL; \ |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 1146 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1147 | |
David Rientjes | 68ae564 | 2012-12-12 13:51:57 -0800 | [diff] [blame] | 1148 | void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 1149 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1150 | struct mem_cgroup *memcg; |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 1151 | |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 1152 | rcu_read_lock(); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1153 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
| 1154 | if (unlikely(!memcg)) |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 1155 | goto out; |
| 1156 | |
| 1157 | switch (idx) { |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 1158 | case PGFAULT: |
Johannes Weiner | 0e574a9 | 2012-01-12 17:18:35 -0800 | [diff] [blame] | 1159 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
| 1160 | break; |
| 1161 | case PGMAJFAULT: |
| 1162 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 1163 | break; |
| 1164 | default: |
| 1165 | BUG(); |
| 1166 | } |
| 1167 | out: |
| 1168 | rcu_read_unlock(); |
| 1169 | } |
David Rientjes | 68ae564 | 2012-12-12 13:51:57 -0800 | [diff] [blame] | 1170 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 1171 | |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1172 | /** |
| 1173 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg |
| 1174 | * @zone: zone of the wanted lruvec |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1175 | * @memcg: memcg of the wanted lruvec |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1176 | * |
| 1177 | * Returns the lru list vector holding pages for the given @zone and |
| 1178 | * @mem. This can be the global zone lruvec, if the memory controller |
| 1179 | * is disabled. |
| 1180 | */ |
| 1181 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, |
| 1182 | struct mem_cgroup *memcg) |
| 1183 | { |
| 1184 | struct mem_cgroup_per_zone *mz; |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 1185 | struct lruvec *lruvec; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1186 | |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 1187 | if (mem_cgroup_disabled()) { |
| 1188 | lruvec = &zone->lruvec; |
| 1189 | goto out; |
| 1190 | } |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1191 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 1192 | mz = mem_cgroup_zone_zoneinfo(memcg, zone); |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 1193 | lruvec = &mz->lruvec; |
| 1194 | out: |
| 1195 | /* |
| 1196 | * Since a node can be onlined after the mem_cgroup was created, |
| 1197 | * we have to be prepared to initialize lruvec->zone here; |
| 1198 | * and if offlined then reonlined, we need to reinitialize it. |
| 1199 | */ |
| 1200 | if (unlikely(lruvec->zone != zone)) |
| 1201 | lruvec->zone = zone; |
| 1202 | return lruvec; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1203 | } |
| 1204 | |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1205 | /** |
Johannes Weiner | dfe0e77 | 2014-12-10 15:43:43 -0800 | [diff] [blame] | 1206 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1207 | * @page: the page |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1208 | * @zone: zone of the page |
Johannes Weiner | dfe0e77 | 2014-12-10 15:43:43 -0800 | [diff] [blame] | 1209 | * |
| 1210 | * This function is only safe when following the LRU page isolation |
| 1211 | * and putback protocol: the LRU lock must be held, and the page must |
| 1212 | * either be PageLRU() or the caller must have isolated/allocated it. |
Minchan Kim | 3f58a82 | 2011-03-22 16:32:53 -0700 | [diff] [blame] | 1213 | */ |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1214 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
Minchan Kim | 3f58a82 | 2011-03-22 16:32:53 -0700 | [diff] [blame] | 1215 | { |
| 1216 | struct mem_cgroup_per_zone *mz; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1217 | struct mem_cgroup *memcg; |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 1218 | struct lruvec *lruvec; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 1219 | |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 1220 | if (mem_cgroup_disabled()) { |
| 1221 | lruvec = &zone->lruvec; |
| 1222 | goto out; |
| 1223 | } |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1224 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 1225 | memcg = page->mem_cgroup; |
Hugh Dickins | 7512102 | 2012-03-05 14:59:18 -0800 | [diff] [blame] | 1226 | /* |
Johannes Weiner | dfe0e77 | 2014-12-10 15:43:43 -0800 | [diff] [blame] | 1227 | * Swapcache readahead pages are added to the LRU - and |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 1228 | * possibly migrated - before they are charged. |
Hugh Dickins | 7512102 | 2012-03-05 14:59:18 -0800 | [diff] [blame] | 1229 | */ |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 1230 | if (!memcg) |
| 1231 | memcg = root_mem_cgroup; |
Hugh Dickins | 7512102 | 2012-03-05 14:59:18 -0800 | [diff] [blame] | 1232 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 1233 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 1234 | lruvec = &mz->lruvec; |
| 1235 | out: |
| 1236 | /* |
| 1237 | * Since a node can be onlined after the mem_cgroup was created, |
| 1238 | * we have to be prepared to initialize lruvec->zone here; |
| 1239 | * and if offlined then reonlined, we need to reinitialize it. |
| 1240 | */ |
| 1241 | if (unlikely(lruvec->zone != zone)) |
| 1242 | lruvec->zone = zone; |
| 1243 | return lruvec; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1244 | } |
| 1245 | |
| 1246 | /** |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1247 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
| 1248 | * @lruvec: mem_cgroup per zone lru vector |
| 1249 | * @lru: index of lru list the page is sitting on |
| 1250 | * @nr_pages: positive when adding or negative when removing |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1251 | * |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1252 | * This function must be called when a page is added to or removed from an |
| 1253 | * lru list. |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1254 | */ |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1255 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
| 1256 | int nr_pages) |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1257 | { |
| 1258 | struct mem_cgroup_per_zone *mz; |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1259 | unsigned long *lru_size; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1260 | |
| 1261 | if (mem_cgroup_disabled()) |
| 1262 | return; |
| 1263 | |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1264 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
| 1265 | lru_size = mz->lru_size + lru; |
| 1266 | *lru_size += nr_pages; |
| 1267 | VM_BUG_ON((long)(*lru_size) < 0); |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 1268 | } |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 1269 | |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1270 | bool mem_cgroup_is_descendant(struct mem_cgroup *memcg, struct mem_cgroup *root) |
Michal Hocko | 3e92041 | 2011-07-26 16:08:29 -0700 | [diff] [blame] | 1271 | { |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1272 | if (root == memcg) |
Johannes Weiner | 91c63734 | 2012-05-29 15:06:24 -0700 | [diff] [blame] | 1273 | return true; |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1274 | if (!root->use_hierarchy) |
Johannes Weiner | 91c63734 | 2012-05-29 15:06:24 -0700 | [diff] [blame] | 1275 | return false; |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1276 | return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup); |
Johannes Weiner | c3ac9a8 | 2012-05-29 15:06:25 -0700 | [diff] [blame] | 1277 | } |
| 1278 | |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1279 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
Johannes Weiner | c3ac9a8 | 2012-05-29 15:06:25 -0700 | [diff] [blame] | 1280 | { |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1281 | struct mem_cgroup *task_memcg; |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 1282 | struct task_struct *p; |
David Rientjes | ffbdccf | 2013-07-03 15:01:23 -0700 | [diff] [blame] | 1283 | bool ret; |
David Rientjes | 4c4a221 | 2008-02-07 00:14:06 -0800 | [diff] [blame] | 1284 | |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 1285 | p = find_lock_task_mm(task); |
David Rientjes | de077d2 | 2012-01-12 17:18:52 -0800 | [diff] [blame] | 1286 | if (p) { |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1287 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
David Rientjes | de077d2 | 2012-01-12 17:18:52 -0800 | [diff] [blame] | 1288 | task_unlock(p); |
| 1289 | } else { |
| 1290 | /* |
| 1291 | * All threads may have already detached their mm's, but the oom |
| 1292 | * killer still needs to detect if they have already been oom |
| 1293 | * killed to prevent needlessly killing additional tasks. |
| 1294 | */ |
David Rientjes | ffbdccf | 2013-07-03 15:01:23 -0700 | [diff] [blame] | 1295 | rcu_read_lock(); |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1296 | task_memcg = mem_cgroup_from_task(task); |
| 1297 | css_get(&task_memcg->css); |
David Rientjes | ffbdccf | 2013-07-03 15:01:23 -0700 | [diff] [blame] | 1298 | rcu_read_unlock(); |
David Rientjes | de077d2 | 2012-01-12 17:18:52 -0800 | [diff] [blame] | 1299 | } |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1300 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
| 1301 | css_put(&task_memcg->css); |
David Rientjes | 4c4a221 | 2008-02-07 00:14:06 -0800 | [diff] [blame] | 1302 | return ret; |
| 1303 | } |
| 1304 | |
Konstantin Khlebnikov | c56d5c7 | 2012-05-29 15:07:00 -0700 | [diff] [blame] | 1305 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1306 | { |
KOSAKI Motohiro | c772be9 | 2009-01-07 18:08:25 -0800 | [diff] [blame] | 1307 | unsigned long inactive_ratio; |
Johannes Weiner | 9b27297 | 2011-11-02 13:38:23 -0700 | [diff] [blame] | 1308 | unsigned long inactive; |
| 1309 | unsigned long active; |
| 1310 | unsigned long gb; |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1311 | |
Hugh Dickins | 4d7dcca | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 1312 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
| 1313 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1314 | |
KOSAKI Motohiro | c772be9 | 2009-01-07 18:08:25 -0800 | [diff] [blame] | 1315 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
| 1316 | if (gb) |
| 1317 | inactive_ratio = int_sqrt(10 * gb); |
| 1318 | else |
| 1319 | inactive_ratio = 1; |
| 1320 | |
Johannes Weiner | 9b27297 | 2011-11-02 13:38:23 -0700 | [diff] [blame] | 1321 | return inactive * inactive_ratio < active; |
KOSAKI Motohiro | c772be9 | 2009-01-07 18:08:25 -0800 | [diff] [blame] | 1322 | } |
| 1323 | |
Vladimir Davydov | 90cbc25 | 2015-02-11 15:25:55 -0800 | [diff] [blame] | 1324 | bool mem_cgroup_lruvec_online(struct lruvec *lruvec) |
| 1325 | { |
| 1326 | struct mem_cgroup_per_zone *mz; |
| 1327 | struct mem_cgroup *memcg; |
| 1328 | |
| 1329 | if (mem_cgroup_disabled()) |
| 1330 | return true; |
| 1331 | |
| 1332 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
| 1333 | memcg = mz->memcg; |
| 1334 | |
| 1335 | return !!(memcg->css.flags & CSS_ONLINE); |
| 1336 | } |
| 1337 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1338 | #define mem_cgroup_from_counter(counter, member) \ |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1339 | container_of(counter, struct mem_cgroup, member) |
| 1340 | |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1341 | /** |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1342 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
Wanpeng Li | dad7557 | 2012-06-20 12:53:01 -0700 | [diff] [blame] | 1343 | * @memcg: the memory cgroup |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1344 | * |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1345 | * Returns the maximum amount of memory @mem can be charged with, in |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 1346 | * pages. |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1347 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1348 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1349 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1350 | unsigned long margin = 0; |
| 1351 | unsigned long count; |
| 1352 | unsigned long limit; |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1353 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1354 | count = page_counter_read(&memcg->memory); |
| 1355 | limit = ACCESS_ONCE(memcg->memory.limit); |
| 1356 | if (count < limit) |
| 1357 | margin = limit - count; |
| 1358 | |
| 1359 | if (do_swap_account) { |
| 1360 | count = page_counter_read(&memcg->memsw); |
| 1361 | limit = ACCESS_ONCE(memcg->memsw.limit); |
| 1362 | if (count <= limit) |
| 1363 | margin = min(margin, limit - count); |
| 1364 | } |
| 1365 | |
| 1366 | return margin; |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1367 | } |
| 1368 | |
KAMEZAWA Hiroyuki | 1f4c025 | 2011-07-26 16:08:21 -0700 | [diff] [blame] | 1369 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 1370 | { |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 1371 | /* root ? */ |
Linus Torvalds | 14208b0 | 2014-06-09 15:03:33 -0700 | [diff] [blame] | 1372 | if (mem_cgroup_disabled() || !memcg->css.parent) |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 1373 | return vm_swappiness; |
| 1374 | |
Johannes Weiner | bf1ff26 | 2011-03-23 16:42:32 -0700 | [diff] [blame] | 1375 | return memcg->swappiness; |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 1376 | } |
| 1377 | |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 1378 | /* |
Qiang Huang | bdcbb65 | 2014-06-04 16:08:21 -0700 | [diff] [blame] | 1379 | * A routine for checking "mem" is under move_account() or not. |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1380 | * |
Qiang Huang | bdcbb65 | 2014-06-04 16:08:21 -0700 | [diff] [blame] | 1381 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
| 1382 | * moving cgroups. This is for waiting at high-memory pressure |
| 1383 | * caused by "move". |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1384 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1385 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1386 | { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1387 | struct mem_cgroup *from; |
| 1388 | struct mem_cgroup *to; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1389 | bool ret = false; |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1390 | /* |
| 1391 | * Unlike task_move routines, we access mc.to, mc.from not under |
| 1392 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. |
| 1393 | */ |
| 1394 | spin_lock(&mc.lock); |
| 1395 | from = mc.from; |
| 1396 | to = mc.to; |
| 1397 | if (!from) |
| 1398 | goto unlock; |
Michal Hocko | 3e92041 | 2011-07-26 16:08:29 -0700 | [diff] [blame] | 1399 | |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1400 | ret = mem_cgroup_is_descendant(from, memcg) || |
| 1401 | mem_cgroup_is_descendant(to, memcg); |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1402 | unlock: |
| 1403 | spin_unlock(&mc.lock); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1404 | return ret; |
| 1405 | } |
| 1406 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1407 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1408 | { |
| 1409 | if (mc.moving_task && current != mc.moving_task) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1410 | if (mem_cgroup_under_move(memcg)) { |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1411 | DEFINE_WAIT(wait); |
| 1412 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); |
| 1413 | /* moving charge context might have finished. */ |
| 1414 | if (mc.moving_task) |
| 1415 | schedule(); |
| 1416 | finish_wait(&mc.waitq, &wait); |
| 1417 | return true; |
| 1418 | } |
| 1419 | } |
| 1420 | return false; |
| 1421 | } |
| 1422 | |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1423 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1424 | /** |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1425 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1426 | * @memcg: The memory cgroup that went over limit |
| 1427 | * @p: Task that is going to be killed |
| 1428 | * |
| 1429 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is |
| 1430 | * enabled |
| 1431 | */ |
| 1432 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) |
| 1433 | { |
Tejun Heo | e61734c | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 1434 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
Michal Hocko | 08088cb | 2014-02-25 15:01:44 -0800 | [diff] [blame] | 1435 | static DEFINE_MUTEX(oom_info_lock); |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1436 | struct mem_cgroup *iter; |
| 1437 | unsigned int i; |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1438 | |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1439 | if (!p) |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1440 | return; |
| 1441 | |
Michal Hocko | 08088cb | 2014-02-25 15:01:44 -0800 | [diff] [blame] | 1442 | mutex_lock(&oom_info_lock); |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1443 | rcu_read_lock(); |
| 1444 | |
Tejun Heo | e61734c | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 1445 | pr_info("Task in "); |
| 1446 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
Greg Thelen | 0346dad | 2015-01-26 12:58:38 -0800 | [diff] [blame] | 1447 | pr_cont(" killed as a result of limit of "); |
Tejun Heo | e61734c | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 1448 | pr_cont_cgroup_path(memcg->css.cgroup); |
Greg Thelen | 0346dad | 2015-01-26 12:58:38 -0800 | [diff] [blame] | 1449 | pr_cont("\n"); |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1450 | |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1451 | rcu_read_unlock(); |
| 1452 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1453 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
| 1454 | K((u64)page_counter_read(&memcg->memory)), |
| 1455 | K((u64)memcg->memory.limit), memcg->memory.failcnt); |
| 1456 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", |
| 1457 | K((u64)page_counter_read(&memcg->memsw)), |
| 1458 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); |
| 1459 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", |
| 1460 | K((u64)page_counter_read(&memcg->kmem)), |
| 1461 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1462 | |
| 1463 | for_each_mem_cgroup_tree(iter, memcg) { |
Tejun Heo | e61734c | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 1464 | pr_info("Memory cgroup stats for "); |
| 1465 | pr_cont_cgroup_path(iter->css.cgroup); |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1466 | pr_cont(":"); |
| 1467 | |
| 1468 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
| 1469 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
| 1470 | continue; |
| 1471 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], |
| 1472 | K(mem_cgroup_read_stat(iter, i))); |
| 1473 | } |
| 1474 | |
| 1475 | for (i = 0; i < NR_LRU_LISTS; i++) |
| 1476 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], |
| 1477 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); |
| 1478 | |
| 1479 | pr_cont("\n"); |
| 1480 | } |
Michal Hocko | 08088cb | 2014-02-25 15:01:44 -0800 | [diff] [blame] | 1481 | mutex_unlock(&oom_info_lock); |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1482 | } |
| 1483 | |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1484 | /* |
| 1485 | * This function returns the number of memcg under hierarchy tree. Returns |
| 1486 | * 1(self count) if no children. |
| 1487 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1488 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1489 | { |
| 1490 | int num = 0; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1491 | struct mem_cgroup *iter; |
| 1492 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1493 | for_each_mem_cgroup_tree(iter, memcg) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1494 | num++; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1495 | return num; |
| 1496 | } |
| 1497 | |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1498 | /* |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1499 | * Return the memory (and swap, if configured) limit for a memcg. |
| 1500 | */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1501 | static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1502 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1503 | unsigned long limit; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1504 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1505 | limit = memcg->memory.limit; |
Michal Hocko | 9a5a8f1 | 2012-11-16 14:14:49 -0800 | [diff] [blame] | 1506 | if (mem_cgroup_swappiness(memcg)) { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1507 | unsigned long memsw_limit; |
Michal Hocko | 9a5a8f1 | 2012-11-16 14:14:49 -0800 | [diff] [blame] | 1508 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1509 | memsw_limit = memcg->memsw.limit; |
| 1510 | limit = min(limit + total_swap_pages, memsw_limit); |
Michal Hocko | 9a5a8f1 | 2012-11-16 14:14:49 -0800 | [diff] [blame] | 1511 | } |
Michal Hocko | 9a5a8f1 | 2012-11-16 14:14:49 -0800 | [diff] [blame] | 1512 | return limit; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1513 | } |
| 1514 | |
David Rientjes | 1996546 | 2012-12-11 16:00:26 -0800 | [diff] [blame] | 1515 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
| 1516 | int order) |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1517 | { |
| 1518 | struct mem_cgroup *iter; |
| 1519 | unsigned long chosen_points = 0; |
| 1520 | unsigned long totalpages; |
| 1521 | unsigned int points = 0; |
| 1522 | struct task_struct *chosen = NULL; |
| 1523 | |
David Rientjes | 876aafb | 2012-07-31 16:43:48 -0700 | [diff] [blame] | 1524 | /* |
David Rientjes | 465adcf | 2013-04-29 15:08:45 -0700 | [diff] [blame] | 1525 | * If current has a pending SIGKILL or is exiting, then automatically |
| 1526 | * select it. The goal is to allow it to allocate so that it may |
| 1527 | * quickly exit and free its memory. |
David Rientjes | 876aafb | 2012-07-31 16:43:48 -0700 | [diff] [blame] | 1528 | */ |
Oleg Nesterov | d003f37 | 2014-12-12 16:56:24 -0800 | [diff] [blame] | 1529 | if (fatal_signal_pending(current) || task_will_free_mem(current)) { |
Michal Hocko | 49550b6 | 2015-02-11 15:26:12 -0800 | [diff] [blame] | 1530 | mark_tsk_oom_victim(current); |
David Rientjes | 876aafb | 2012-07-31 16:43:48 -0700 | [diff] [blame] | 1531 | return; |
| 1532 | } |
| 1533 | |
| 1534 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1535 | totalpages = mem_cgroup_get_limit(memcg) ? : 1; |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1536 | for_each_mem_cgroup_tree(iter, memcg) { |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 1537 | struct css_task_iter it; |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1538 | struct task_struct *task; |
| 1539 | |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 1540 | css_task_iter_start(&iter->css, &it); |
| 1541 | while ((task = css_task_iter_next(&it))) { |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1542 | switch (oom_scan_process_thread(task, totalpages, NULL, |
| 1543 | false)) { |
| 1544 | case OOM_SCAN_SELECT: |
| 1545 | if (chosen) |
| 1546 | put_task_struct(chosen); |
| 1547 | chosen = task; |
| 1548 | chosen_points = ULONG_MAX; |
| 1549 | get_task_struct(chosen); |
| 1550 | /* fall through */ |
| 1551 | case OOM_SCAN_CONTINUE: |
| 1552 | continue; |
| 1553 | case OOM_SCAN_ABORT: |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 1554 | css_task_iter_end(&it); |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1555 | mem_cgroup_iter_break(memcg, iter); |
| 1556 | if (chosen) |
| 1557 | put_task_struct(chosen); |
| 1558 | return; |
| 1559 | case OOM_SCAN_OK: |
| 1560 | break; |
| 1561 | }; |
| 1562 | points = oom_badness(task, memcg, NULL, totalpages); |
David Rientjes | d49ad93 | 2014-01-23 15:53:34 -0800 | [diff] [blame] | 1563 | if (!points || points < chosen_points) |
| 1564 | continue; |
| 1565 | /* Prefer thread group leaders for display purposes */ |
| 1566 | if (points == chosen_points && |
| 1567 | thread_group_leader(chosen)) |
| 1568 | continue; |
| 1569 | |
| 1570 | if (chosen) |
| 1571 | put_task_struct(chosen); |
| 1572 | chosen = task; |
| 1573 | chosen_points = points; |
| 1574 | get_task_struct(chosen); |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1575 | } |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 1576 | css_task_iter_end(&it); |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1577 | } |
| 1578 | |
| 1579 | if (!chosen) |
| 1580 | return; |
| 1581 | points = chosen_points * 1000 / totalpages; |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1582 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
| 1583 | NULL, "Memory cgroup out of memory"); |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1584 | } |
| 1585 | |
Michele Curti | ae6e71d | 2014-12-12 16:56:35 -0800 | [diff] [blame] | 1586 | #if MAX_NUMNODES > 1 |
| 1587 | |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1588 | /** |
| 1589 | * test_mem_cgroup_node_reclaimable |
Wanpeng Li | dad7557 | 2012-06-20 12:53:01 -0700 | [diff] [blame] | 1590 | * @memcg: the target memcg |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1591 | * @nid: the node ID to be checked. |
| 1592 | * @noswap : specify true here if the user wants flle only information. |
| 1593 | * |
| 1594 | * This function returns whether the specified memcg contains any |
| 1595 | * reclaimable pages on a node. Returns true if there are any reclaimable |
| 1596 | * pages in the node. |
| 1597 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1598 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1599 | int nid, bool noswap) |
| 1600 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1601 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1602 | return true; |
| 1603 | if (noswap || !total_swap_pages) |
| 1604 | return false; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1605 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1606 | return true; |
| 1607 | return false; |
| 1608 | |
| 1609 | } |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1610 | |
| 1611 | /* |
| 1612 | * Always updating the nodemask is not very good - even if we have an empty |
| 1613 | * list or the wrong list here, we can start from some node and traverse all |
| 1614 | * nodes based on the zonelist. So update the list loosely once per 10 secs. |
| 1615 | * |
| 1616 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1617 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1618 | { |
| 1619 | int nid; |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 1620 | /* |
| 1621 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET |
| 1622 | * pagein/pageout changes since the last update. |
| 1623 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1624 | if (!atomic_read(&memcg->numainfo_events)) |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 1625 | return; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1626 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1627 | return; |
| 1628 | |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1629 | /* make a nodemask where this memcg uses memory from */ |
Lai Jiangshan | 31aaea4 | 2012-12-12 13:51:27 -0800 | [diff] [blame] | 1630 | memcg->scan_nodes = node_states[N_MEMORY]; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1631 | |
Lai Jiangshan | 31aaea4 | 2012-12-12 13:51:27 -0800 | [diff] [blame] | 1632 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1633 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1634 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
| 1635 | node_clear(nid, memcg->scan_nodes); |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1636 | } |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 1637 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1638 | atomic_set(&memcg->numainfo_events, 0); |
| 1639 | atomic_set(&memcg->numainfo_updating, 0); |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1640 | } |
| 1641 | |
| 1642 | /* |
| 1643 | * Selecting a node where we start reclaim from. Because what we need is just |
| 1644 | * reducing usage counter, start from anywhere is O,K. Considering |
| 1645 | * memory reclaim from current node, there are pros. and cons. |
| 1646 | * |
| 1647 | * Freeing memory from current node means freeing memory from a node which |
| 1648 | * we'll use or we've used. So, it may make LRU bad. And if several threads |
| 1649 | * hit limits, it will see a contention on a node. But freeing from remote |
| 1650 | * node means more costs for memory reclaim because of memory latency. |
| 1651 | * |
| 1652 | * Now, we use round-robin. Better algorithm is welcomed. |
| 1653 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1654 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1655 | { |
| 1656 | int node; |
| 1657 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1658 | mem_cgroup_may_update_nodemask(memcg); |
| 1659 | node = memcg->last_scanned_node; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1660 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1661 | node = next_node(node, memcg->scan_nodes); |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1662 | if (node == MAX_NUMNODES) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1663 | node = first_node(memcg->scan_nodes); |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1664 | /* |
| 1665 | * We call this when we hit limit, not when pages are added to LRU. |
| 1666 | * No LRU may hold pages because all pages are UNEVICTABLE or |
| 1667 | * memcg is too small and all pages are not on LRU. In that case, |
| 1668 | * we use curret node. |
| 1669 | */ |
| 1670 | if (unlikely(node == MAX_NUMNODES)) |
| 1671 | node = numa_node_id(); |
| 1672 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1673 | memcg->last_scanned_node = node; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1674 | return node; |
| 1675 | } |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1676 | #else |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1677 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1678 | { |
| 1679 | return 0; |
| 1680 | } |
| 1681 | #endif |
| 1682 | |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1683 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
| 1684 | struct zone *zone, |
| 1685 | gfp_t gfp_mask, |
| 1686 | unsigned long *total_scanned) |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1687 | { |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1688 | struct mem_cgroup *victim = NULL; |
| 1689 | int total = 0; |
| 1690 | int loop = 0; |
| 1691 | unsigned long excess; |
| 1692 | unsigned long nr_scanned; |
| 1693 | struct mem_cgroup_reclaim_cookie reclaim = { |
| 1694 | .zone = zone, |
| 1695 | .priority = 0, |
| 1696 | }; |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1697 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1698 | excess = soft_limit_excess(root_memcg); |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1699 | |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1700 | while (1) { |
| 1701 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); |
| 1702 | if (!victim) { |
| 1703 | loop++; |
| 1704 | if (loop >= 2) { |
| 1705 | /* |
| 1706 | * If we have not been able to reclaim |
| 1707 | * anything, it might because there are |
| 1708 | * no reclaimable pages under this hierarchy |
| 1709 | */ |
| 1710 | if (!total) |
| 1711 | break; |
| 1712 | /* |
| 1713 | * We want to do more targeted reclaim. |
| 1714 | * excess >> 2 is not to excessive so as to |
| 1715 | * reclaim too much, nor too less that we keep |
| 1716 | * coming back to reclaim from this cgroup |
| 1717 | */ |
| 1718 | if (total >= (excess >> 2) || |
| 1719 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) |
| 1720 | break; |
| 1721 | } |
| 1722 | continue; |
| 1723 | } |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1724 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
| 1725 | zone, &nr_scanned); |
| 1726 | *total_scanned += nr_scanned; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1727 | if (!soft_limit_excess(root_memcg)) |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1728 | break; |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1729 | } |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1730 | mem_cgroup_iter_break(root_memcg, victim); |
| 1731 | return total; |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1732 | } |
| 1733 | |
Johannes Weiner | 0056f4e | 2013-10-31 16:34:14 -0700 | [diff] [blame] | 1734 | #ifdef CONFIG_LOCKDEP |
| 1735 | static struct lockdep_map memcg_oom_lock_dep_map = { |
| 1736 | .name = "memcg_oom_lock", |
| 1737 | }; |
| 1738 | #endif |
| 1739 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1740 | static DEFINE_SPINLOCK(memcg_oom_lock); |
| 1741 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1742 | /* |
| 1743 | * Check OOM-Killer is already running under our hierarchy. |
| 1744 | * If someone is running, return false. |
| 1745 | */ |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1746 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1747 | { |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1748 | struct mem_cgroup *iter, *failed = NULL; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 1749 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1750 | spin_lock(&memcg_oom_lock); |
| 1751 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1752 | for_each_mem_cgroup_tree(iter, memcg) { |
Johannes Weiner | 23751be | 2011-08-25 15:59:16 -0700 | [diff] [blame] | 1753 | if (iter->oom_lock) { |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1754 | /* |
| 1755 | * this subtree of our hierarchy is already locked |
| 1756 | * so we cannot give a lock. |
| 1757 | */ |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1758 | failed = iter; |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1759 | mem_cgroup_iter_break(memcg, iter); |
| 1760 | break; |
Johannes Weiner | 23751be | 2011-08-25 15:59:16 -0700 | [diff] [blame] | 1761 | } else |
| 1762 | iter->oom_lock = true; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1763 | } |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1764 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1765 | if (failed) { |
| 1766 | /* |
| 1767 | * OK, we failed to lock the whole subtree so we have |
| 1768 | * to clean up what we set up to the failing subtree |
| 1769 | */ |
| 1770 | for_each_mem_cgroup_tree(iter, memcg) { |
| 1771 | if (iter == failed) { |
| 1772 | mem_cgroup_iter_break(memcg, iter); |
| 1773 | break; |
| 1774 | } |
| 1775 | iter->oom_lock = false; |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1776 | } |
Johannes Weiner | 0056f4e | 2013-10-31 16:34:14 -0700 | [diff] [blame] | 1777 | } else |
| 1778 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1779 | |
| 1780 | spin_unlock(&memcg_oom_lock); |
| 1781 | |
| 1782 | return !failed; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 1783 | } |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1784 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1785 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1786 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1787 | struct mem_cgroup *iter; |
| 1788 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1789 | spin_lock(&memcg_oom_lock); |
Johannes Weiner | 0056f4e | 2013-10-31 16:34:14 -0700 | [diff] [blame] | 1790 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1791 | for_each_mem_cgroup_tree(iter, memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1792 | iter->oom_lock = false; |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1793 | spin_unlock(&memcg_oom_lock); |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1794 | } |
| 1795 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1796 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1797 | { |
| 1798 | struct mem_cgroup *iter; |
| 1799 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1800 | for_each_mem_cgroup_tree(iter, memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1801 | atomic_inc(&iter->under_oom); |
| 1802 | } |
| 1803 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1804 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1805 | { |
| 1806 | struct mem_cgroup *iter; |
| 1807 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1808 | /* |
| 1809 | * When a new child is created while the hierarchy is under oom, |
| 1810 | * mem_cgroup_oom_lock() may not be called. We have to use |
| 1811 | * atomic_add_unless() here. |
| 1812 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1813 | for_each_mem_cgroup_tree(iter, memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1814 | atomic_add_unless(&iter->under_oom, -1, 0); |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1815 | } |
| 1816 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1817 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
| 1818 | |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1819 | struct oom_wait_info { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 1820 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1821 | wait_queue_t wait; |
| 1822 | }; |
| 1823 | |
| 1824 | static int memcg_oom_wake_function(wait_queue_t *wait, |
| 1825 | unsigned mode, int sync, void *arg) |
| 1826 | { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 1827 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
| 1828 | struct mem_cgroup *oom_wait_memcg; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1829 | struct oom_wait_info *oom_wait_info; |
| 1830 | |
| 1831 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 1832 | oom_wait_memcg = oom_wait_info->memcg; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1833 | |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1834 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
| 1835 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1836 | return 0; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1837 | return autoremove_wake_function(wait, mode, sync, arg); |
| 1838 | } |
| 1839 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1840 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1841 | { |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 1842 | atomic_inc(&memcg->oom_wakeups); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1843 | /* for filtering, pass "memcg" as argument. */ |
| 1844 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1845 | } |
| 1846 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1847 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1848 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1849 | if (memcg && atomic_read(&memcg->under_oom)) |
| 1850 | memcg_wakeup_oom(memcg); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1851 | } |
| 1852 | |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 1853 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1854 | { |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 1855 | if (!current->memcg_oom.may_oom) |
| 1856 | return; |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1857 | /* |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1858 | * We are in the middle of the charge context here, so we |
| 1859 | * don't want to block when potentially sitting on a callstack |
| 1860 | * that holds all kinds of filesystem and mm locks. |
| 1861 | * |
| 1862 | * Also, the caller may handle a failed allocation gracefully |
| 1863 | * (like optional page cache readahead) and so an OOM killer |
| 1864 | * invocation might not even be necessary. |
| 1865 | * |
| 1866 | * That's why we don't do anything here except remember the |
| 1867 | * OOM context and then deal with it at the end of the page |
| 1868 | * fault when the stack is unwound, the locks are released, |
| 1869 | * and when we know whether the fault was overall successful. |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1870 | */ |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1871 | css_get(&memcg->css); |
| 1872 | current->memcg_oom.memcg = memcg; |
| 1873 | current->memcg_oom.gfp_mask = mask; |
| 1874 | current->memcg_oom.order = order; |
| 1875 | } |
| 1876 | |
| 1877 | /** |
| 1878 | * mem_cgroup_oom_synchronize - complete memcg OOM handling |
| 1879 | * @handle: actually kill/wait or just clean up the OOM state |
| 1880 | * |
| 1881 | * This has to be called at the end of a page fault if the memcg OOM |
| 1882 | * handler was enabled. |
| 1883 | * |
| 1884 | * Memcg supports userspace OOM handling where failed allocations must |
| 1885 | * sleep on a waitqueue until the userspace task resolves the |
| 1886 | * situation. Sleeping directly in the charge context with all kinds |
| 1887 | * of locks held is not a good idea, instead we remember an OOM state |
| 1888 | * in the task and mem_cgroup_oom_synchronize() has to be called at |
| 1889 | * the end of the page fault to complete the OOM handling. |
| 1890 | * |
| 1891 | * Returns %true if an ongoing memcg OOM situation was detected and |
| 1892 | * completed, %false otherwise. |
| 1893 | */ |
| 1894 | bool mem_cgroup_oom_synchronize(bool handle) |
| 1895 | { |
| 1896 | struct mem_cgroup *memcg = current->memcg_oom.memcg; |
| 1897 | struct oom_wait_info owait; |
| 1898 | bool locked; |
| 1899 | |
| 1900 | /* OOM is global, do not handle */ |
| 1901 | if (!memcg) |
| 1902 | return false; |
| 1903 | |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 1904 | if (!handle || oom_killer_disabled) |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1905 | goto cleanup; |
| 1906 | |
| 1907 | owait.memcg = memcg; |
| 1908 | owait.wait.flags = 0; |
| 1909 | owait.wait.func = memcg_oom_wake_function; |
| 1910 | owait.wait.private = current; |
| 1911 | INIT_LIST_HEAD(&owait.wait.task_list); |
| 1912 | |
| 1913 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1914 | mem_cgroup_mark_under_oom(memcg); |
| 1915 | |
| 1916 | locked = mem_cgroup_oom_trylock(memcg); |
| 1917 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1918 | if (locked) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1919 | mem_cgroup_oom_notify(memcg); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1920 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1921 | if (locked && !memcg->oom_kill_disable) { |
| 1922 | mem_cgroup_unmark_under_oom(memcg); |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1923 | finish_wait(&memcg_oom_waitq, &owait.wait); |
| 1924 | mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask, |
| 1925 | current->memcg_oom.order); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1926 | } else { |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 1927 | schedule(); |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1928 | mem_cgroup_unmark_under_oom(memcg); |
| 1929 | finish_wait(&memcg_oom_waitq, &owait.wait); |
| 1930 | } |
| 1931 | |
| 1932 | if (locked) { |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1933 | mem_cgroup_oom_unlock(memcg); |
| 1934 | /* |
| 1935 | * There is no guarantee that an OOM-lock contender |
| 1936 | * sees the wakeups triggered by the OOM kill |
| 1937 | * uncharges. Wake any sleepers explicitely. |
| 1938 | */ |
| 1939 | memcg_oom_recover(memcg); |
| 1940 | } |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1941 | cleanup: |
| 1942 | current->memcg_oom.memcg = NULL; |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 1943 | css_put(&memcg->css); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1944 | return true; |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1945 | } |
| 1946 | |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 1947 | /** |
| 1948 | * mem_cgroup_begin_page_stat - begin a page state statistics transaction |
| 1949 | * @page: page that is going to change accounted state |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1950 | * |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 1951 | * This function must mark the beginning of an accounted page state |
| 1952 | * change to prevent double accounting when the page is concurrently |
| 1953 | * being moved to another memcg: |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1954 | * |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1955 | * memcg = mem_cgroup_begin_page_stat(page); |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 1956 | * if (TestClearPageState(page)) |
| 1957 | * mem_cgroup_update_page_stat(memcg, state, -1); |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1958 | * mem_cgroup_end_page_stat(memcg); |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1959 | */ |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1960 | struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page) |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1961 | { |
| 1962 | struct mem_cgroup *memcg; |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1963 | unsigned long flags; |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1964 | |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1965 | /* |
| 1966 | * The RCU lock is held throughout the transaction. The fast |
| 1967 | * path can get away without acquiring the memcg->move_lock |
| 1968 | * because page moving starts with an RCU grace period. |
| 1969 | * |
| 1970 | * The RCU lock also protects the memcg from being freed when |
| 1971 | * the page state that is going to change is the only thing |
| 1972 | * preventing the page from being uncharged. |
| 1973 | * E.g. end-writeback clearing PageWriteback(), which allows |
| 1974 | * migration to go ahead and uncharge the page before the |
| 1975 | * account transaction might be complete. |
| 1976 | */ |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 1977 | rcu_read_lock(); |
| 1978 | |
| 1979 | if (mem_cgroup_disabled()) |
| 1980 | return NULL; |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1981 | again: |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 1982 | memcg = page->mem_cgroup; |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 1983 | if (unlikely(!memcg)) |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 1984 | return NULL; |
| 1985 | |
Qiang Huang | bdcbb65 | 2014-06-04 16:08:21 -0700 | [diff] [blame] | 1986 | if (atomic_read(&memcg->moving_account) <= 0) |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 1987 | return memcg; |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1988 | |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1989 | spin_lock_irqsave(&memcg->move_lock, flags); |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 1990 | if (memcg != page->mem_cgroup) { |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1991 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1992 | goto again; |
| 1993 | } |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1994 | |
| 1995 | /* |
| 1996 | * When charge migration first begins, we can have locked and |
| 1997 | * unlocked page stat updates happening concurrently. Track |
| 1998 | * the task who has the lock for mem_cgroup_end_page_stat(). |
| 1999 | */ |
| 2000 | memcg->move_lock_task = current; |
| 2001 | memcg->move_lock_flags = flags; |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 2002 | |
| 2003 | return memcg; |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2004 | } |
| 2005 | |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 2006 | /** |
| 2007 | * mem_cgroup_end_page_stat - finish a page state statistics transaction |
| 2008 | * @memcg: the memcg that was accounted against |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 2009 | */ |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 2010 | void mem_cgroup_end_page_stat(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2011 | { |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 2012 | if (memcg && memcg->move_lock_task == current) { |
| 2013 | unsigned long flags = memcg->move_lock_flags; |
| 2014 | |
| 2015 | memcg->move_lock_task = NULL; |
| 2016 | memcg->move_lock_flags = 0; |
| 2017 | |
| 2018 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
| 2019 | } |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2020 | |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 2021 | rcu_read_unlock(); |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2022 | } |
| 2023 | |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 2024 | /** |
| 2025 | * mem_cgroup_update_page_stat - update page state statistics |
| 2026 | * @memcg: memcg to account against |
| 2027 | * @idx: page state item to account |
| 2028 | * @val: number of pages (positive or negative) |
| 2029 | * |
| 2030 | * See mem_cgroup_begin_page_stat() for locking requirements. |
| 2031 | */ |
| 2032 | void mem_cgroup_update_page_stat(struct mem_cgroup *memcg, |
Sha Zhengju | 68b4876 | 2013-09-12 15:13:50 -0700 | [diff] [blame] | 2033 | enum mem_cgroup_stat_index idx, int val) |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 2034 | { |
Sha Zhengju | 658b72c | 2013-09-12 15:13:52 -0700 | [diff] [blame] | 2035 | VM_BUG_ON(!rcu_read_lock_held()); |
KAMEZAWA Hiroyuki | 26174ef | 2010-10-27 15:33:43 -0700 | [diff] [blame] | 2036 | |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 2037 | if (memcg) |
| 2038 | this_cpu_add(memcg->stat->count[idx], val); |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 2039 | } |
KAMEZAWA Hiroyuki | 26174ef | 2010-10-27 15:33:43 -0700 | [diff] [blame] | 2040 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2041 | /* |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2042 | * size of first charge trial. "32" comes from vmscan.c's magic value. |
| 2043 | * TODO: maybe necessary to use big numbers in big irons. |
| 2044 | */ |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2045 | #define CHARGE_BATCH 32U |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2046 | struct memcg_stock_pcp { |
| 2047 | struct mem_cgroup *cached; /* this never be root cgroup */ |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 2048 | unsigned int nr_pages; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2049 | struct work_struct work; |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 2050 | unsigned long flags; |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 2051 | #define FLUSHING_CACHED_CHARGE 0 |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2052 | }; |
| 2053 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); |
Michal Hocko | 9f50fad | 2011-08-09 11:56:26 +0200 | [diff] [blame] | 2054 | static DEFINE_MUTEX(percpu_charge_mutex); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2055 | |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 2056 | /** |
| 2057 | * consume_stock: Try to consume stocked charge on this cpu. |
| 2058 | * @memcg: memcg to consume from. |
| 2059 | * @nr_pages: how many pages to charge. |
| 2060 | * |
| 2061 | * The charges will only happen if @memcg matches the current cpu's memcg |
| 2062 | * stock, and at least @nr_pages are available in that stock. Failure to |
| 2063 | * service an allocation will refill the stock. |
| 2064 | * |
| 2065 | * returns true if successful, false otherwise. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2066 | */ |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 2067 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2068 | { |
| 2069 | struct memcg_stock_pcp *stock; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2070 | bool ret = false; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2071 | |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 2072 | if (nr_pages > CHARGE_BATCH) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2073 | return ret; |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 2074 | |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2075 | stock = &get_cpu_var(memcg_stock); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2076 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 2077 | stock->nr_pages -= nr_pages; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2078 | ret = true; |
| 2079 | } |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2080 | put_cpu_var(memcg_stock); |
| 2081 | return ret; |
| 2082 | } |
| 2083 | |
| 2084 | /* |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2085 | * Returns stocks cached in percpu and reset cached information. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2086 | */ |
| 2087 | static void drain_stock(struct memcg_stock_pcp *stock) |
| 2088 | { |
| 2089 | struct mem_cgroup *old = stock->cached; |
| 2090 | |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 2091 | if (stock->nr_pages) { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2092 | page_counter_uncharge(&old->memory, stock->nr_pages); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2093 | if (do_swap_account) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2094 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
Johannes Weiner | e8ea14c | 2014-12-10 15:42:42 -0800 | [diff] [blame] | 2095 | css_put_many(&old->css, stock->nr_pages); |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 2096 | stock->nr_pages = 0; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2097 | } |
| 2098 | stock->cached = NULL; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2099 | } |
| 2100 | |
| 2101 | /* |
| 2102 | * This must be called under preempt disabled or must be called by |
| 2103 | * a thread which is pinned to local cpu. |
| 2104 | */ |
| 2105 | static void drain_local_stock(struct work_struct *dummy) |
| 2106 | { |
Christoph Lameter | 7c8e018 | 2014-06-04 16:07:56 -0700 | [diff] [blame] | 2107 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2108 | drain_stock(stock); |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 2109 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2110 | } |
| 2111 | |
| 2112 | /* |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2113 | * Cache charges(val) to local per_cpu area. |
Greg Thelen | 320cc51 | 2010-03-15 15:27:28 +0100 | [diff] [blame] | 2114 | * This will be consumed by consume_stock() function, later. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2115 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2116 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2117 | { |
| 2118 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); |
| 2119 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2120 | if (stock->cached != memcg) { /* reset if necessary */ |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2121 | drain_stock(stock); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2122 | stock->cached = memcg; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2123 | } |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 2124 | stock->nr_pages += nr_pages; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2125 | put_cpu_var(memcg_stock); |
| 2126 | } |
| 2127 | |
| 2128 | /* |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2129 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
Johannes Weiner | 6d3d6aa | 2014-12-10 15:42:50 -0800 | [diff] [blame] | 2130 | * of the hierarchy under it. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2131 | */ |
Johannes Weiner | 6d3d6aa | 2014-12-10 15:42:50 -0800 | [diff] [blame] | 2132 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2133 | { |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 2134 | int cpu, curcpu; |
Michal Hocko | d38144b | 2011-07-26 16:08:28 -0700 | [diff] [blame] | 2135 | |
Johannes Weiner | 6d3d6aa | 2014-12-10 15:42:50 -0800 | [diff] [blame] | 2136 | /* If someone's already draining, avoid adding running more workers. */ |
| 2137 | if (!mutex_trylock(&percpu_charge_mutex)) |
| 2138 | return; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2139 | /* Notify other cpus that system-wide "drain" is running */ |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2140 | get_online_cpus(); |
Johannes Weiner | 5af12d0 | 2011-08-25 15:59:07 -0700 | [diff] [blame] | 2141 | curcpu = get_cpu(); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2142 | for_each_online_cpu(cpu) { |
| 2143 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2144 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 2145 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2146 | memcg = stock->cached; |
| 2147 | if (!memcg || !stock->nr_pages) |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 2148 | continue; |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 2149 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) |
Michal Hocko | 3e92041 | 2011-07-26 16:08:29 -0700 | [diff] [blame] | 2150 | continue; |
Michal Hocko | d1a05b6 | 2011-07-26 16:08:27 -0700 | [diff] [blame] | 2151 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
| 2152 | if (cpu == curcpu) |
| 2153 | drain_local_stock(&stock->work); |
| 2154 | else |
| 2155 | schedule_work_on(cpu, &stock->work); |
| 2156 | } |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2157 | } |
Johannes Weiner | 5af12d0 | 2011-08-25 15:59:07 -0700 | [diff] [blame] | 2158 | put_cpu(); |
Andrew Morton | f894ffa | 2013-09-12 15:13:35 -0700 | [diff] [blame] | 2159 | put_online_cpus(); |
Michal Hocko | 9f50fad | 2011-08-09 11:56:26 +0200 | [diff] [blame] | 2160 | mutex_unlock(&percpu_charge_mutex); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2161 | } |
| 2162 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2163 | /* |
| 2164 | * This function drains percpu counter value from DEAD cpu and |
| 2165 | * move it to local cpu. Note that this function can be preempted. |
| 2166 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2167 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2168 | { |
| 2169 | int i; |
| 2170 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2171 | spin_lock(&memcg->pcp_counter_lock); |
Johannes Weiner | 6104621 | 2012-05-29 15:07:05 -0700 | [diff] [blame] | 2172 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2173 | long x = per_cpu(memcg->stat->count[i], cpu); |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2174 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2175 | per_cpu(memcg->stat->count[i], cpu) = 0; |
| 2176 | memcg->nocpu_base.count[i] += x; |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2177 | } |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 2178 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2179 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 2180 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2181 | per_cpu(memcg->stat->events[i], cpu) = 0; |
| 2182 | memcg->nocpu_base.events[i] += x; |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 2183 | } |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2184 | spin_unlock(&memcg->pcp_counter_lock); |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2185 | } |
| 2186 | |
Paul Gortmaker | 0db0628 | 2013-06-19 14:53:51 -0400 | [diff] [blame] | 2187 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2188 | unsigned long action, |
| 2189 | void *hcpu) |
| 2190 | { |
| 2191 | int cpu = (unsigned long)hcpu; |
| 2192 | struct memcg_stock_pcp *stock; |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2193 | struct mem_cgroup *iter; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2194 | |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 2195 | if (action == CPU_ONLINE) |
KAMEZAWA Hiroyuki | 1489eba | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2196 | return NOTIFY_OK; |
KAMEZAWA Hiroyuki | 1489eba | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2197 | |
Kirill A. Shutemov | d833049 | 2012-04-12 12:49:11 -0700 | [diff] [blame] | 2198 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2199 | return NOTIFY_OK; |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2200 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 2201 | for_each_mem_cgroup(iter) |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2202 | mem_cgroup_drain_pcp_counter(iter, cpu); |
| 2203 | |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2204 | stock = &per_cpu(memcg_stock, cpu); |
| 2205 | drain_stock(stock); |
| 2206 | return NOTIFY_OK; |
| 2207 | } |
| 2208 | |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 2209 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
| 2210 | unsigned int nr_pages) |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2211 | { |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2212 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
Johannes Weiner | 9b13061 | 2014-08-06 16:05:51 -0700 | [diff] [blame] | 2213 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2214 | struct mem_cgroup *mem_over_limit; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2215 | struct page_counter *counter; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2216 | unsigned long nr_reclaimed; |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2217 | bool may_swap = true; |
| 2218 | bool drained = false; |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 2219 | int ret = 0; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 2220 | |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2221 | if (mem_cgroup_is_root(memcg)) |
| 2222 | goto done; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2223 | retry: |
Michal Hocko | b6b6cc7 | 2014-04-07 15:37:44 -0700 | [diff] [blame] | 2224 | if (consume_stock(memcg, nr_pages)) |
| 2225 | goto done; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2226 | |
Johannes Weiner | 3fbe724 | 2014-10-09 15:28:54 -0700 | [diff] [blame] | 2227 | if (!do_swap_account || |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2228 | !page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
| 2229 | if (!page_counter_try_charge(&memcg->memory, batch, &counter)) |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2230 | goto done_restock; |
Johannes Weiner | 3fbe724 | 2014-10-09 15:28:54 -0700 | [diff] [blame] | 2231 | if (do_swap_account) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2232 | page_counter_uncharge(&memcg->memsw, batch); |
| 2233 | mem_over_limit = mem_cgroup_from_counter(counter, memory); |
Johannes Weiner | 3fbe724 | 2014-10-09 15:28:54 -0700 | [diff] [blame] | 2234 | } else { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2235 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2236 | may_swap = false; |
Johannes Weiner | 3fbe724 | 2014-10-09 15:28:54 -0700 | [diff] [blame] | 2237 | } |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2238 | |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2239 | if (batch > nr_pages) { |
| 2240 | batch = nr_pages; |
| 2241 | goto retry; |
| 2242 | } |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2243 | |
Johannes Weiner | 06b078f | 2014-08-06 16:05:44 -0700 | [diff] [blame] | 2244 | /* |
| 2245 | * Unlike in global OOM situations, memcg is not in a physical |
| 2246 | * memory shortage. Allow dying and OOM-killed tasks to |
| 2247 | * bypass the last charges so that they can exit quickly and |
| 2248 | * free their memory. |
| 2249 | */ |
| 2250 | if (unlikely(test_thread_flag(TIF_MEMDIE) || |
| 2251 | fatal_signal_pending(current) || |
| 2252 | current->flags & PF_EXITING)) |
| 2253 | goto bypass; |
| 2254 | |
| 2255 | if (unlikely(task_in_memcg_oom(current))) |
| 2256 | goto nomem; |
| 2257 | |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2258 | if (!(gfp_mask & __GFP_WAIT)) |
| 2259 | goto nomem; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2260 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 2261 | mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1); |
| 2262 | |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2263 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
| 2264 | gfp_mask, may_swap); |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2265 | |
Johannes Weiner | 61e02c7 | 2014-08-06 16:08:16 -0700 | [diff] [blame] | 2266 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2267 | goto retry; |
Johannes Weiner | 28c34c2 | 2014-08-06 16:05:47 -0700 | [diff] [blame] | 2268 | |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2269 | if (!drained) { |
Johannes Weiner | 6d3d6aa | 2014-12-10 15:42:50 -0800 | [diff] [blame] | 2270 | drain_all_stock(mem_over_limit); |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2271 | drained = true; |
| 2272 | goto retry; |
| 2273 | } |
| 2274 | |
Johannes Weiner | 28c34c2 | 2014-08-06 16:05:47 -0700 | [diff] [blame] | 2275 | if (gfp_mask & __GFP_NORETRY) |
| 2276 | goto nomem; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2277 | /* |
| 2278 | * Even though the limit is exceeded at this point, reclaim |
| 2279 | * may have been able to free some pages. Retry the charge |
| 2280 | * before killing the task. |
| 2281 | * |
| 2282 | * Only for regular pages, though: huge pages are rather |
| 2283 | * unlikely to succeed so close to the limit, and we fall back |
| 2284 | * to regular pages anyway in case of failure. |
| 2285 | */ |
Johannes Weiner | 61e02c7 | 2014-08-06 16:08:16 -0700 | [diff] [blame] | 2286 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2287 | goto retry; |
| 2288 | /* |
| 2289 | * At task move, charge accounts can be doubly counted. So, it's |
| 2290 | * better to wait until the end of task_move if something is going on. |
| 2291 | */ |
| 2292 | if (mem_cgroup_wait_acct_move(mem_over_limit)) |
| 2293 | goto retry; |
| 2294 | |
Johannes Weiner | 9b13061 | 2014-08-06 16:05:51 -0700 | [diff] [blame] | 2295 | if (nr_retries--) |
| 2296 | goto retry; |
| 2297 | |
Johannes Weiner | 06b078f | 2014-08-06 16:05:44 -0700 | [diff] [blame] | 2298 | if (gfp_mask & __GFP_NOFAIL) |
| 2299 | goto bypass; |
| 2300 | |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2301 | if (fatal_signal_pending(current)) |
| 2302 | goto bypass; |
| 2303 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 2304 | mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1); |
| 2305 | |
Johannes Weiner | 61e02c7 | 2014-08-06 16:08:16 -0700 | [diff] [blame] | 2306 | mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(nr_pages)); |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2307 | nomem: |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 2308 | if (!(gfp_mask & __GFP_NOFAIL)) |
Johannes Weiner | 3168ecb | 2013-10-31 16:34:13 -0700 | [diff] [blame] | 2309 | return -ENOMEM; |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2310 | bypass: |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2311 | return -EINTR; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2312 | |
| 2313 | done_restock: |
Johannes Weiner | e8ea14c | 2014-12-10 15:42:42 -0800 | [diff] [blame] | 2314 | css_get_many(&memcg->css, batch); |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2315 | if (batch > nr_pages) |
| 2316 | refill_stock(memcg, batch - nr_pages); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 2317 | /* |
| 2318 | * If the hierarchy is above the normal consumption range, |
| 2319 | * make the charging task trim their excess contribution. |
| 2320 | */ |
| 2321 | do { |
| 2322 | if (page_counter_read(&memcg->memory) <= memcg->high) |
| 2323 | continue; |
| 2324 | mem_cgroup_events(memcg, MEMCG_HIGH, 1); |
| 2325 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); |
| 2326 | } while ((memcg = parent_mem_cgroup(memcg))); |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2327 | done: |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 2328 | return ret; |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2329 | } |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2330 | |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 2331 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
Daisuke Nishimura | a3032a2 | 2009-12-15 16:47:10 -0800 | [diff] [blame] | 2332 | { |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2333 | if (mem_cgroup_is_root(memcg)) |
| 2334 | return; |
| 2335 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2336 | page_counter_uncharge(&memcg->memory, nr_pages); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 2337 | if (do_swap_account) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2338 | page_counter_uncharge(&memcg->memsw, nr_pages); |
Daisuke Nishimura | a3032a2 | 2009-12-15 16:47:10 -0800 | [diff] [blame] | 2339 | |
Johannes Weiner | e8ea14c | 2014-12-10 15:42:42 -0800 | [diff] [blame] | 2340 | css_put_many(&memcg->css, nr_pages); |
KAMEZAWA Hiroyuki | d01dd17 | 2012-05-29 15:07:03 -0700 | [diff] [blame] | 2341 | } |
| 2342 | |
| 2343 | /* |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2344 | * A helper function to get mem_cgroup from ID. must be called under |
Tejun Heo | ec903c0 | 2014-05-13 12:11:01 -0400 | [diff] [blame] | 2345 | * rcu_read_lock(). The caller is responsible for calling |
| 2346 | * css_tryget_online() if the mem_cgroup is used for charging. (dropping |
| 2347 | * refcnt from swap can be called against removed memcg.) |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2348 | */ |
| 2349 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) |
| 2350 | { |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2351 | /* ID 0 is unused ID */ |
| 2352 | if (!id) |
| 2353 | return NULL; |
Li Zefan | 34c00c3 | 2013-09-23 16:56:01 +0800 | [diff] [blame] | 2354 | return mem_cgroup_from_id(id); |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2355 | } |
| 2356 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2357 | /* |
| 2358 | * try_get_mem_cgroup_from_page - look up page's memcg association |
| 2359 | * @page: the page |
| 2360 | * |
| 2361 | * Look up, get a css reference, and return the memcg that owns @page. |
| 2362 | * |
| 2363 | * The page must be locked to prevent racing with swap-in and page |
| 2364 | * cache charges. If coming from an unlocked page table, the caller |
| 2365 | * must ensure the page is on the LRU or this can race with charging. |
| 2366 | */ |
Wu Fengguang | e42d9d5 | 2009-12-16 12:19:59 +0100 | [diff] [blame] | 2367 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2368 | { |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 2369 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2370 | unsigned short id; |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2371 | swp_entry_t ent; |
| 2372 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 2373 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2374 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2375 | memcg = page->mem_cgroup; |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 2376 | if (memcg) { |
| 2377 | if (!css_tryget_online(&memcg->css)) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2378 | memcg = NULL; |
Wu Fengguang | e42d9d5 | 2009-12-16 12:19:59 +0100 | [diff] [blame] | 2379 | } else if (PageSwapCache(page)) { |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2380 | ent.val = page_private(page); |
Bob Liu | 9fb4b7c | 2012-01-12 17:18:48 -0800 | [diff] [blame] | 2381 | id = lookup_swap_cgroup_id(ent); |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2382 | rcu_read_lock(); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2383 | memcg = mem_cgroup_lookup(id); |
Tejun Heo | ec903c0 | 2014-05-13 12:11:01 -0400 | [diff] [blame] | 2384 | if (memcg && !css_tryget_online(&memcg->css)) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2385 | memcg = NULL; |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2386 | rcu_read_unlock(); |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2387 | } |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2388 | return memcg; |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2389 | } |
| 2390 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2391 | static void lock_page_lru(struct page *page, int *isolated) |
| 2392 | { |
| 2393 | struct zone *zone = page_zone(page); |
| 2394 | |
| 2395 | spin_lock_irq(&zone->lru_lock); |
| 2396 | if (PageLRU(page)) { |
| 2397 | struct lruvec *lruvec; |
| 2398 | |
| 2399 | lruvec = mem_cgroup_page_lruvec(page, zone); |
| 2400 | ClearPageLRU(page); |
| 2401 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
| 2402 | *isolated = 1; |
| 2403 | } else |
| 2404 | *isolated = 0; |
| 2405 | } |
| 2406 | |
| 2407 | static void unlock_page_lru(struct page *page, int isolated) |
| 2408 | { |
| 2409 | struct zone *zone = page_zone(page); |
| 2410 | |
| 2411 | if (isolated) { |
| 2412 | struct lruvec *lruvec; |
| 2413 | |
| 2414 | lruvec = mem_cgroup_page_lruvec(page, zone); |
| 2415 | VM_BUG_ON_PAGE(PageLRU(page), page); |
| 2416 | SetPageLRU(page); |
| 2417 | add_page_to_lru_list(page, lruvec, page_lru(page)); |
| 2418 | } |
| 2419 | spin_unlock_irq(&zone->lru_lock); |
| 2420 | } |
| 2421 | |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 2422 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
Johannes Weiner | 6abb5a8 | 2014-08-08 14:19:33 -0700 | [diff] [blame] | 2423 | bool lrucare) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2424 | { |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2425 | int isolated; |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2426 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2427 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2428 | |
| 2429 | /* |
| 2430 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page |
| 2431 | * may already be on some other mem_cgroup's LRU. Take care of it. |
| 2432 | */ |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2433 | if (lrucare) |
| 2434 | lock_page_lru(page, &isolated); |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2435 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2436 | /* |
| 2437 | * Nobody should be changing or seriously looking at |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2438 | * page->mem_cgroup at this point: |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2439 | * |
| 2440 | * - the page is uncharged |
| 2441 | * |
| 2442 | * - the page is off-LRU |
| 2443 | * |
| 2444 | * - an anonymous fault has exclusive page access, except for |
| 2445 | * a locked page table |
| 2446 | * |
| 2447 | * - a page cache insertion, a swapin fault, or a migration |
| 2448 | * have the page locked |
| 2449 | */ |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2450 | page->mem_cgroup = memcg; |
Hugh Dickins | 3be9127 | 2008-02-07 00:14:19 -0800 | [diff] [blame] | 2451 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2452 | if (lrucare) |
| 2453 | unlock_page_lru(page, isolated); |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2454 | } |
| 2455 | |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2456 | #ifdef CONFIG_MEMCG_KMEM |
Vladimir Davydov | dbf22eb | 2015-02-10 14:11:41 -0800 | [diff] [blame] | 2457 | int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, |
| 2458 | unsigned long nr_pages) |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 2459 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2460 | struct page_counter *counter; |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2461 | int ret = 0; |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2462 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2463 | ret = page_counter_try_charge(&memcg->kmem, nr_pages, &counter); |
| 2464 | if (ret < 0) |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2465 | return ret; |
| 2466 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2467 | ret = try_charge(memcg, gfp, nr_pages); |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2468 | if (ret == -EINTR) { |
| 2469 | /* |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 2470 | * try_charge() chose to bypass to root due to OOM kill or |
| 2471 | * fatal signal. Since our only options are to either fail |
| 2472 | * the allocation or charge it to this cgroup, do it as a |
| 2473 | * temporary condition. But we can't fail. From a kmem/slab |
| 2474 | * perspective, the cache has already been selected, by |
| 2475 | * mem_cgroup_kmem_get_cache(), so it is too late to change |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2476 | * our minds. |
| 2477 | * |
| 2478 | * This condition will only trigger if the task entered |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 2479 | * memcg_charge_kmem in a sane state, but was OOM-killed |
| 2480 | * during try_charge() above. Tasks that were already dying |
| 2481 | * when the allocation triggers should have been already |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2482 | * directed to the root cgroup in memcontrol.h |
| 2483 | */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2484 | page_counter_charge(&memcg->memory, nr_pages); |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2485 | if (do_swap_account) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2486 | page_counter_charge(&memcg->memsw, nr_pages); |
Johannes Weiner | e8ea14c | 2014-12-10 15:42:42 -0800 | [diff] [blame] | 2487 | css_get_many(&memcg->css, nr_pages); |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2488 | ret = 0; |
| 2489 | } else if (ret) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2490 | page_counter_uncharge(&memcg->kmem, nr_pages); |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2491 | |
| 2492 | return ret; |
| 2493 | } |
| 2494 | |
Vladimir Davydov | dbf22eb | 2015-02-10 14:11:41 -0800 | [diff] [blame] | 2495 | void memcg_uncharge_kmem(struct mem_cgroup *memcg, unsigned long nr_pages) |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2496 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2497 | page_counter_uncharge(&memcg->memory, nr_pages); |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2498 | if (do_swap_account) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2499 | page_counter_uncharge(&memcg->memsw, nr_pages); |
Glauber Costa | 7de3768 | 2012-12-18 14:22:07 -0800 | [diff] [blame] | 2500 | |
Johannes Weiner | 64f2199 | 2014-12-10 15:42:45 -0800 | [diff] [blame] | 2501 | page_counter_uncharge(&memcg->kmem, nr_pages); |
Glauber Costa | 7de3768 | 2012-12-18 14:22:07 -0800 | [diff] [blame] | 2502 | |
Johannes Weiner | e8ea14c | 2014-12-10 15:42:42 -0800 | [diff] [blame] | 2503 | css_put_many(&memcg->css, nr_pages); |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2504 | } |
| 2505 | |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 2506 | /* |
| 2507 | * helper for acessing a memcg's index. It will be used as an index in the |
| 2508 | * child cache array in kmem_cache, and also to derive its name. This function |
| 2509 | * will return -1 when this is not a kmem-limited memcg. |
| 2510 | */ |
| 2511 | int memcg_cache_id(struct mem_cgroup *memcg) |
| 2512 | { |
| 2513 | return memcg ? memcg->kmemcg_id : -1; |
| 2514 | } |
| 2515 | |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2516 | static int memcg_alloc_cache_id(void) |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 2517 | { |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2518 | int id, size; |
| 2519 | int err; |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 2520 | |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 2521 | id = ida_simple_get(&memcg_cache_ida, |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2522 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
| 2523 | if (id < 0) |
| 2524 | return id; |
| 2525 | |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 2526 | if (id < memcg_nr_cache_ids) |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2527 | return id; |
| 2528 | |
| 2529 | /* |
| 2530 | * There's no space for the new id in memcg_caches arrays, |
| 2531 | * so we have to grow them. |
| 2532 | */ |
Vladimir Davydov | 05257a1 | 2015-02-12 14:59:01 -0800 | [diff] [blame] | 2533 | down_write(&memcg_cache_ids_sem); |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2534 | |
| 2535 | size = 2 * (id + 1); |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 2536 | if (size < MEMCG_CACHES_MIN_SIZE) |
| 2537 | size = MEMCG_CACHES_MIN_SIZE; |
| 2538 | else if (size > MEMCG_CACHES_MAX_SIZE) |
| 2539 | size = MEMCG_CACHES_MAX_SIZE; |
| 2540 | |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2541 | err = memcg_update_all_caches(size); |
Vladimir Davydov | 05257a1 | 2015-02-12 14:59:01 -0800 | [diff] [blame] | 2542 | if (!err) |
Vladimir Davydov | 60d3fd3 | 2015-02-12 14:59:10 -0800 | [diff] [blame] | 2543 | err = memcg_update_all_list_lrus(size); |
| 2544 | if (!err) |
Vladimir Davydov | 05257a1 | 2015-02-12 14:59:01 -0800 | [diff] [blame] | 2545 | memcg_nr_cache_ids = size; |
| 2546 | |
| 2547 | up_write(&memcg_cache_ids_sem); |
| 2548 | |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2549 | if (err) { |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 2550 | ida_simple_remove(&memcg_cache_ida, id); |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2551 | return err; |
| 2552 | } |
| 2553 | return id; |
| 2554 | } |
| 2555 | |
| 2556 | static void memcg_free_cache_id(int id) |
| 2557 | { |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 2558 | ida_simple_remove(&memcg_cache_ida, id); |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 2559 | } |
| 2560 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2561 | struct memcg_kmem_cache_create_work { |
Vladimir Davydov | 5722d09 | 2014-04-07 15:39:24 -0700 | [diff] [blame] | 2562 | struct mem_cgroup *memcg; |
| 2563 | struct kmem_cache *cachep; |
| 2564 | struct work_struct work; |
| 2565 | }; |
| 2566 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2567 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2568 | { |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2569 | struct memcg_kmem_cache_create_work *cw = |
| 2570 | container_of(w, struct memcg_kmem_cache_create_work, work); |
Vladimir Davydov | 5722d09 | 2014-04-07 15:39:24 -0700 | [diff] [blame] | 2571 | struct mem_cgroup *memcg = cw->memcg; |
| 2572 | struct kmem_cache *cachep = cw->cachep; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2573 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2574 | memcg_create_kmem_cache(memcg, cachep); |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 2575 | |
Vladimir Davydov | 5722d09 | 2014-04-07 15:39:24 -0700 | [diff] [blame] | 2576 | css_put(&memcg->css); |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2577 | kfree(cw); |
| 2578 | } |
| 2579 | |
| 2580 | /* |
| 2581 | * Enqueue the creation of a per-memcg kmem_cache. |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2582 | */ |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2583 | static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
| 2584 | struct kmem_cache *cachep) |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2585 | { |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2586 | struct memcg_kmem_cache_create_work *cw; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2587 | |
Vladimir Davydov | 776ed0f | 2014-06-04 16:10:02 -0700 | [diff] [blame] | 2588 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2589 | if (!cw) |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2590 | return; |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2591 | |
| 2592 | css_get(&memcg->css); |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2593 | |
| 2594 | cw->memcg = memcg; |
| 2595 | cw->cachep = cachep; |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2596 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2597 | |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2598 | schedule_work(&cw->work); |
| 2599 | } |
| 2600 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2601 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
| 2602 | struct kmem_cache *cachep) |
Glauber Costa | 0e9d92f | 2012-12-18 14:22:42 -0800 | [diff] [blame] | 2603 | { |
| 2604 | /* |
| 2605 | * We need to stop accounting when we kmalloc, because if the |
| 2606 | * corresponding kmalloc cache is not yet created, the first allocation |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2607 | * in __memcg_schedule_kmem_cache_create will recurse. |
Glauber Costa | 0e9d92f | 2012-12-18 14:22:42 -0800 | [diff] [blame] | 2608 | * |
| 2609 | * However, it is better to enclose the whole function. Depending on |
| 2610 | * the debugging options enabled, INIT_WORK(), for instance, can |
| 2611 | * trigger an allocation. This too, will make us recurse. Because at |
| 2612 | * this point we can't allow ourselves back into memcg_kmem_get_cache, |
| 2613 | * the safest choice is to do it like this, wrapping the whole function. |
| 2614 | */ |
Vladimir Davydov | 6f185c2 | 2014-12-12 16:55:15 -0800 | [diff] [blame] | 2615 | current->memcg_kmem_skip_account = 1; |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2616 | __memcg_schedule_kmem_cache_create(memcg, cachep); |
Vladimir Davydov | 6f185c2 | 2014-12-12 16:55:15 -0800 | [diff] [blame] | 2617 | current->memcg_kmem_skip_account = 0; |
Glauber Costa | 0e9d92f | 2012-12-18 14:22:42 -0800 | [diff] [blame] | 2618 | } |
Vladimir Davydov | c67a8a6 | 2014-06-04 16:07:39 -0700 | [diff] [blame] | 2619 | |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2620 | /* |
| 2621 | * Return the kmem_cache we're supposed to use for a slab allocation. |
| 2622 | * We try to use the current memcg's version of the cache. |
| 2623 | * |
| 2624 | * If the cache does not exist yet, if we are the first user of it, |
| 2625 | * we either create it immediately, if possible, or create it asynchronously |
| 2626 | * in a workqueue. |
| 2627 | * In the latter case, we will let the current allocation go through with |
| 2628 | * the original cache. |
| 2629 | * |
| 2630 | * Can't be called in interrupt context or from kernel threads. |
| 2631 | * This function needs to be called with rcu_read_lock() held. |
| 2632 | */ |
Zhang Zhen | 056b7cc | 2014-12-12 16:55:38 -0800 | [diff] [blame] | 2633 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep) |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2634 | { |
| 2635 | struct mem_cgroup *memcg; |
Vladimir Davydov | 959c896 | 2014-01-23 15:52:59 -0800 | [diff] [blame] | 2636 | struct kmem_cache *memcg_cachep; |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 2637 | int kmemcg_id; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2638 | |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 2639 | VM_BUG_ON(!is_root_cache(cachep)); |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2640 | |
Vladimir Davydov | 9d100c5 | 2014-12-12 16:54:53 -0800 | [diff] [blame] | 2641 | if (current->memcg_kmem_skip_account) |
Glauber Costa | 0e9d92f | 2012-12-18 14:22:42 -0800 | [diff] [blame] | 2642 | return cachep; |
| 2643 | |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2644 | memcg = get_mem_cgroup_from_mm(current->mm); |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 2645 | kmemcg_id = ACCESS_ONCE(memcg->kmemcg_id); |
| 2646 | if (kmemcg_id < 0) |
Li Zefan | ca0dde9 | 2013-04-29 15:08:57 -0700 | [diff] [blame] | 2647 | goto out; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2648 | |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 2649 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2650 | if (likely(memcg_cachep)) |
| 2651 | return memcg_cachep; |
Li Zefan | ca0dde9 | 2013-04-29 15:08:57 -0700 | [diff] [blame] | 2652 | |
| 2653 | /* |
| 2654 | * If we are in a safe context (can wait, and not in interrupt |
| 2655 | * context), we could be be predictable and return right away. |
| 2656 | * This would guarantee that the allocation being performed |
| 2657 | * already belongs in the new cache. |
| 2658 | * |
| 2659 | * However, there are some clashes that can arrive from locking. |
| 2660 | * For instance, because we acquire the slab_mutex while doing |
Vladimir Davydov | 776ed0f | 2014-06-04 16:10:02 -0700 | [diff] [blame] | 2661 | * memcg_create_kmem_cache, this means no further allocation |
| 2662 | * could happen with the slab_mutex held. So it's better to |
| 2663 | * defer everything. |
Li Zefan | ca0dde9 | 2013-04-29 15:08:57 -0700 | [diff] [blame] | 2664 | */ |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2665 | memcg_schedule_kmem_cache_create(memcg, cachep); |
Li Zefan | ca0dde9 | 2013-04-29 15:08:57 -0700 | [diff] [blame] | 2666 | out: |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2667 | css_put(&memcg->css); |
Li Zefan | ca0dde9 | 2013-04-29 15:08:57 -0700 | [diff] [blame] | 2668 | return cachep; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2669 | } |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2670 | |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2671 | void __memcg_kmem_put_cache(struct kmem_cache *cachep) |
| 2672 | { |
| 2673 | if (!is_root_cache(cachep)) |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 2674 | css_put(&cachep->memcg_params.memcg->css); |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2675 | } |
| 2676 | |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2677 | /* |
| 2678 | * We need to verify if the allocation against current->mm->owner's memcg is |
| 2679 | * possible for the given order. But the page is not allocated yet, so we'll |
| 2680 | * need a further commit step to do the final arrangements. |
| 2681 | * |
| 2682 | * It is possible for the task to switch cgroups in this mean time, so at |
| 2683 | * commit time, we can't rely on task conversion any longer. We'll then use |
| 2684 | * the handle argument to return to the caller which cgroup we should commit |
| 2685 | * against. We could also return the memcg directly and avoid the pointer |
| 2686 | * passing, but a boolean return value gives better semantics considering |
| 2687 | * the compiled-out case as well. |
| 2688 | * |
| 2689 | * Returning true means the allocation is possible. |
| 2690 | */ |
| 2691 | bool |
| 2692 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) |
| 2693 | { |
| 2694 | struct mem_cgroup *memcg; |
| 2695 | int ret; |
| 2696 | |
| 2697 | *_memcg = NULL; |
Glauber Costa | 6d42c23 | 2013-07-08 16:00:00 -0700 | [diff] [blame] | 2698 | |
Johannes Weiner | df38197 | 2014-04-07 15:37:43 -0700 | [diff] [blame] | 2699 | memcg = get_mem_cgroup_from_mm(current->mm); |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2700 | |
Vladimir Davydov | cf2b8fb | 2014-10-09 15:28:59 -0700 | [diff] [blame] | 2701 | if (!memcg_kmem_is_active(memcg)) { |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2702 | css_put(&memcg->css); |
| 2703 | return true; |
| 2704 | } |
| 2705 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2706 | ret = memcg_charge_kmem(memcg, gfp, 1 << order); |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2707 | if (!ret) |
| 2708 | *_memcg = memcg; |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2709 | |
| 2710 | css_put(&memcg->css); |
| 2711 | return (ret == 0); |
| 2712 | } |
| 2713 | |
| 2714 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, |
| 2715 | int order) |
| 2716 | { |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2717 | VM_BUG_ON(mem_cgroup_is_root(memcg)); |
| 2718 | |
| 2719 | /* The page allocation failed. Revert */ |
| 2720 | if (!page) { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2721 | memcg_uncharge_kmem(memcg, 1 << order); |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2722 | return; |
| 2723 | } |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2724 | page->mem_cgroup = memcg; |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2725 | } |
| 2726 | |
| 2727 | void __memcg_kmem_uncharge_pages(struct page *page, int order) |
| 2728 | { |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2729 | struct mem_cgroup *memcg = page->mem_cgroup; |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2730 | |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2731 | if (!memcg) |
| 2732 | return; |
| 2733 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 2734 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 2735 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2736 | memcg_uncharge_kmem(memcg, 1 << order); |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2737 | page->mem_cgroup = NULL; |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2738 | } |
Vladimir Davydov | 60d3fd3 | 2015-02-12 14:59:10 -0800 | [diff] [blame] | 2739 | |
| 2740 | struct mem_cgroup *__mem_cgroup_from_kmem(void *ptr) |
| 2741 | { |
| 2742 | struct mem_cgroup *memcg = NULL; |
| 2743 | struct kmem_cache *cachep; |
| 2744 | struct page *page; |
| 2745 | |
| 2746 | page = virt_to_head_page(ptr); |
| 2747 | if (PageSlab(page)) { |
| 2748 | cachep = page->slab_cache; |
| 2749 | if (!is_root_cache(cachep)) |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 2750 | memcg = cachep->memcg_params.memcg; |
Vladimir Davydov | 60d3fd3 | 2015-02-12 14:59:10 -0800 | [diff] [blame] | 2751 | } else |
| 2752 | /* page allocated by alloc_kmem_pages */ |
| 2753 | memcg = page->mem_cgroup; |
| 2754 | |
| 2755 | return memcg; |
| 2756 | } |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2757 | #endif /* CONFIG_MEMCG_KMEM */ |
| 2758 | |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2759 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 2760 | |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2761 | /* |
| 2762 | * Because tail pages are not marked as "used", set it. We're under |
KAMEZAWA Hiroyuki | e94c8a9 | 2012-01-12 17:18:20 -0800 | [diff] [blame] | 2763 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
| 2764 | * charge/uncharge will be never happen and move_account() is done under |
| 2765 | * compound_lock(), so we don't have to take care of races. |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2766 | */ |
KAMEZAWA Hiroyuki | e94c8a9 | 2012-01-12 17:18:20 -0800 | [diff] [blame] | 2767 | void mem_cgroup_split_huge_fixup(struct page *head) |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2768 | { |
KAMEZAWA Hiroyuki | e94c8a9 | 2012-01-12 17:18:20 -0800 | [diff] [blame] | 2769 | int i; |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2770 | |
KAMEZAWA Hiroyuki | 3d37c4a | 2011-01-25 15:07:28 -0800 | [diff] [blame] | 2771 | if (mem_cgroup_disabled()) |
| 2772 | return; |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 2773 | |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 2774 | for (i = 1; i < HPAGE_PMD_NR; i++) |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2775 | head[i].mem_cgroup = head->mem_cgroup; |
Michal Hocko | b9982f8 | 2014-12-10 15:43:51 -0800 | [diff] [blame] | 2776 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2777 | __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 2778 | HPAGE_PMD_NR); |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2779 | } |
Hugh Dickins | 12d2710 | 2012-01-12 17:19:52 -0800 | [diff] [blame] | 2780 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2781 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2782 | /** |
Johannes Weiner | de3638d | 2011-03-23 16:42:28 -0700 | [diff] [blame] | 2783 | * mem_cgroup_move_account - move account of the page |
Johannes Weiner | 5564e88 | 2011-03-23 16:42:29 -0700 | [diff] [blame] | 2784 | * @page: the page |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2785 | * @nr_pages: number of regular pages (>1 for huge pages) |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2786 | * @from: mem_cgroup which the page is moved from. |
| 2787 | * @to: mem_cgroup which the page is moved to. @from != @to. |
| 2788 | * |
| 2789 | * The caller must confirm following. |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2790 | * - page is not on LRU (isolate_page() is useful.) |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2791 | * - compound_lock is held when nr_pages > 1 |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2792 | * |
KAMEZAWA Hiroyuki | 2f3479b | 2012-05-29 15:07:04 -0700 | [diff] [blame] | 2793 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
| 2794 | * from old cgroup. |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2795 | */ |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2796 | static int mem_cgroup_move_account(struct page *page, |
| 2797 | unsigned int nr_pages, |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2798 | struct mem_cgroup *from, |
KAMEZAWA Hiroyuki | 2f3479b | 2012-05-29 15:07:04 -0700 | [diff] [blame] | 2799 | struct mem_cgroup *to) |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2800 | { |
Johannes Weiner | de3638d | 2011-03-23 16:42:28 -0700 | [diff] [blame] | 2801 | unsigned long flags; |
| 2802 | int ret; |
KAMEZAWA Hiroyuki | 987eba6 | 2011-01-20 14:44:25 -0800 | [diff] [blame] | 2803 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2804 | VM_BUG_ON(from == to); |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 2805 | VM_BUG_ON_PAGE(PageLRU(page), page); |
Johannes Weiner | de3638d | 2011-03-23 16:42:28 -0700 | [diff] [blame] | 2806 | /* |
| 2807 | * The page is isolated from LRU. So, collapse function |
| 2808 | * will not handle this page. But page splitting can happen. |
| 2809 | * Do this check under compound_page_lock(). The caller should |
| 2810 | * hold it. |
| 2811 | */ |
| 2812 | ret = -EBUSY; |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2813 | if (nr_pages > 1 && !PageTransHuge(page)) |
Johannes Weiner | de3638d | 2011-03-23 16:42:28 -0700 | [diff] [blame] | 2814 | goto out; |
| 2815 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2816 | /* |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2817 | * Prevent mem_cgroup_migrate() from looking at page->mem_cgroup |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2818 | * of its source page while we change it: page migration takes |
| 2819 | * both pages off the LRU, but page cache replacement doesn't. |
| 2820 | */ |
| 2821 | if (!trylock_page(page)) |
| 2822 | goto out; |
Johannes Weiner | de3638d | 2011-03-23 16:42:28 -0700 | [diff] [blame] | 2823 | |
| 2824 | ret = -EINVAL; |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2825 | if (page->mem_cgroup != from) |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2826 | goto out_unlock; |
Johannes Weiner | de3638d | 2011-03-23 16:42:28 -0700 | [diff] [blame] | 2827 | |
Johannes Weiner | 354a478 | 2014-12-10 15:44:05 -0800 | [diff] [blame] | 2828 | spin_lock_irqsave(&from->move_lock, flags); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2829 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2830 | if (!PageAnon(page) && page_mapped(page)) { |
Johannes Weiner | 59d1d25 | 2014-04-07 15:37:40 -0700 | [diff] [blame] | 2831 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
| 2832 | nr_pages); |
| 2833 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
| 2834 | nr_pages); |
| 2835 | } |
Sha Zhengju | 3ea67d0 | 2013-09-12 15:13:53 -0700 | [diff] [blame] | 2836 | |
Johannes Weiner | 59d1d25 | 2014-04-07 15:37:40 -0700 | [diff] [blame] | 2837 | if (PageWriteback(page)) { |
| 2838 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], |
| 2839 | nr_pages); |
| 2840 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], |
| 2841 | nr_pages); |
| 2842 | } |
Sha Zhengju | 3ea67d0 | 2013-09-12 15:13:53 -0700 | [diff] [blame] | 2843 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2844 | /* |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2845 | * It is safe to change page->mem_cgroup here because the page |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2846 | * is referenced, charged, and isolated - we can't race with |
| 2847 | * uncharging, charging, migration, or LRU putback. |
| 2848 | */ |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 2849 | |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2850 | /* caller should have done css_get */ |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2851 | page->mem_cgroup = to; |
Johannes Weiner | 354a478 | 2014-12-10 15:44:05 -0800 | [diff] [blame] | 2852 | spin_unlock_irqrestore(&from->move_lock, flags); |
| 2853 | |
Johannes Weiner | de3638d | 2011-03-23 16:42:28 -0700 | [diff] [blame] | 2854 | ret = 0; |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2855 | |
| 2856 | local_irq_disable(); |
| 2857 | mem_cgroup_charge_statistics(to, page, nr_pages); |
Johannes Weiner | 5564e88 | 2011-03-23 16:42:29 -0700 | [diff] [blame] | 2858 | memcg_check_events(to, page); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2859 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
Johannes Weiner | 5564e88 | 2011-03-23 16:42:29 -0700 | [diff] [blame] | 2860 | memcg_check_events(from, page); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2861 | local_irq_enable(); |
| 2862 | out_unlock: |
| 2863 | unlock_page(page); |
Johannes Weiner | de3638d | 2011-03-23 16:42:28 -0700 | [diff] [blame] | 2864 | out: |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2865 | return ret; |
| 2866 | } |
| 2867 | |
Andrew Morton | c255a45 | 2012-07-31 16:43:02 -0700 | [diff] [blame] | 2868 | #ifdef CONFIG_MEMCG_SWAP |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2869 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
| 2870 | bool charge) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2871 | { |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2872 | int val = (charge) ? 1 : -1; |
| 2873 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2874 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2875 | |
| 2876 | /** |
| 2877 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. |
| 2878 | * @entry: swap entry to be moved |
| 2879 | * @from: mem_cgroup which the entry is moved from |
| 2880 | * @to: mem_cgroup which the entry is moved to |
| 2881 | * |
| 2882 | * It succeeds only when the swap_cgroup's record for this entry is the same |
| 2883 | * as the mem_cgroup's id of @from. |
| 2884 | * |
| 2885 | * Returns 0 on success, -EINVAL on failure. |
| 2886 | * |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2887 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2888 | * both res and memsw, and called css_get(). |
| 2889 | */ |
| 2890 | static int mem_cgroup_move_swap_account(swp_entry_t entry, |
Hugh Dickins | e91cbb4 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 2891 | struct mem_cgroup *from, struct mem_cgroup *to) |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2892 | { |
| 2893 | unsigned short old_id, new_id; |
| 2894 | |
Li Zefan | 34c00c3 | 2013-09-23 16:56:01 +0800 | [diff] [blame] | 2895 | old_id = mem_cgroup_id(from); |
| 2896 | new_id = mem_cgroup_id(to); |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2897 | |
| 2898 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2899 | mem_cgroup_swap_statistics(from, false); |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2900 | mem_cgroup_swap_statistics(to, true); |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2901 | return 0; |
| 2902 | } |
| 2903 | return -EINVAL; |
| 2904 | } |
| 2905 | #else |
| 2906 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, |
Hugh Dickins | e91cbb4 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 2907 | struct mem_cgroup *from, struct mem_cgroup *to) |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2908 | { |
| 2909 | return -EINVAL; |
| 2910 | } |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2911 | #endif |
| 2912 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2913 | static DEFINE_MUTEX(memcg_limit_mutex); |
Daisuke Nishimura | f212ad7 | 2011-03-23 16:42:25 -0700 | [diff] [blame] | 2914 | |
KOSAKI Motohiro | d38d2a7 | 2009-01-06 14:39:44 -0800 | [diff] [blame] | 2915 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2916 | unsigned long limit) |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2917 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2918 | unsigned long curusage; |
| 2919 | unsigned long oldusage; |
| 2920 | bool enlarge = false; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2921 | int retry_count; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2922 | int ret; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2923 | |
| 2924 | /* |
| 2925 | * For keeping hierarchical_reclaim simple, how long we should retry |
| 2926 | * is depends on callers. We set our retry-count to be function |
| 2927 | * of # of children which we should visit in this loop. |
| 2928 | */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2929 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
| 2930 | mem_cgroup_count_children(memcg); |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2931 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2932 | oldusage = page_counter_read(&memcg->memory); |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2933 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2934 | do { |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2935 | if (signal_pending(current)) { |
| 2936 | ret = -EINTR; |
| 2937 | break; |
| 2938 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2939 | |
| 2940 | mutex_lock(&memcg_limit_mutex); |
| 2941 | if (limit > memcg->memsw.limit) { |
| 2942 | mutex_unlock(&memcg_limit_mutex); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2943 | ret = -EINVAL; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2944 | break; |
| 2945 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2946 | if (limit > memcg->memory.limit) |
| 2947 | enlarge = true; |
| 2948 | ret = page_counter_limit(&memcg->memory, limit); |
| 2949 | mutex_unlock(&memcg_limit_mutex); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2950 | |
| 2951 | if (!ret) |
| 2952 | break; |
| 2953 | |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2954 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
| 2955 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2956 | curusage = page_counter_read(&memcg->memory); |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2957 | /* Usage is reduced ? */ |
Andrew Morton | f894ffa | 2013-09-12 15:13:35 -0700 | [diff] [blame] | 2958 | if (curusage >= oldusage) |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2959 | retry_count--; |
| 2960 | else |
| 2961 | oldusage = curusage; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2962 | } while (retry_count); |
| 2963 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2964 | if (!ret && enlarge) |
| 2965 | memcg_oom_recover(memcg); |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 2966 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2967 | return ret; |
| 2968 | } |
| 2969 | |
Li Zefan | 338c843 | 2009-06-17 16:27:15 -0700 | [diff] [blame] | 2970 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2971 | unsigned long limit) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2972 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2973 | unsigned long curusage; |
| 2974 | unsigned long oldusage; |
| 2975 | bool enlarge = false; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2976 | int retry_count; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2977 | int ret; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2978 | |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2979 | /* see mem_cgroup_resize_res_limit */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2980 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
| 2981 | mem_cgroup_count_children(memcg); |
| 2982 | |
| 2983 | oldusage = page_counter_read(&memcg->memsw); |
| 2984 | |
| 2985 | do { |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2986 | if (signal_pending(current)) { |
| 2987 | ret = -EINTR; |
| 2988 | break; |
| 2989 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2990 | |
| 2991 | mutex_lock(&memcg_limit_mutex); |
| 2992 | if (limit < memcg->memory.limit) { |
| 2993 | mutex_unlock(&memcg_limit_mutex); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2994 | ret = -EINVAL; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2995 | break; |
| 2996 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2997 | if (limit > memcg->memsw.limit) |
| 2998 | enlarge = true; |
| 2999 | ret = page_counter_limit(&memcg->memsw, limit); |
| 3000 | mutex_unlock(&memcg_limit_mutex); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3001 | |
| 3002 | if (!ret) |
| 3003 | break; |
| 3004 | |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 3005 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
| 3006 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3007 | curusage = page_counter_read(&memcg->memsw); |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3008 | /* Usage is reduced ? */ |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3009 | if (curusage >= oldusage) |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3010 | retry_count--; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3011 | else |
| 3012 | oldusage = curusage; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3013 | } while (retry_count); |
| 3014 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3015 | if (!ret && enlarge) |
| 3016 | memcg_oom_recover(memcg); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3017 | |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3018 | return ret; |
| 3019 | } |
| 3020 | |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 3021 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
| 3022 | gfp_t gfp_mask, |
| 3023 | unsigned long *total_scanned) |
| 3024 | { |
| 3025 | unsigned long nr_reclaimed = 0; |
| 3026 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; |
| 3027 | unsigned long reclaimed; |
| 3028 | int loop = 0; |
| 3029 | struct mem_cgroup_tree_per_zone *mctz; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3030 | unsigned long excess; |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 3031 | unsigned long nr_scanned; |
| 3032 | |
| 3033 | if (order > 0) |
| 3034 | return 0; |
| 3035 | |
| 3036 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); |
| 3037 | /* |
| 3038 | * This loop can run a while, specially if mem_cgroup's continuously |
| 3039 | * keep exceeding their soft limit and putting the system under |
| 3040 | * pressure |
| 3041 | */ |
| 3042 | do { |
| 3043 | if (next_mz) |
| 3044 | mz = next_mz; |
| 3045 | else |
| 3046 | mz = mem_cgroup_largest_soft_limit_node(mctz); |
| 3047 | if (!mz) |
| 3048 | break; |
| 3049 | |
| 3050 | nr_scanned = 0; |
| 3051 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, |
| 3052 | gfp_mask, &nr_scanned); |
| 3053 | nr_reclaimed += reclaimed; |
| 3054 | *total_scanned += nr_scanned; |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 3055 | spin_lock_irq(&mctz->lock); |
Vladimir Davydov | bc2f2e7 | 2014-12-10 15:43:40 -0800 | [diff] [blame] | 3056 | __mem_cgroup_remove_exceeded(mz, mctz); |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 3057 | |
| 3058 | /* |
| 3059 | * If we failed to reclaim anything from this memory cgroup |
| 3060 | * it is time to move on to the next cgroup |
| 3061 | */ |
| 3062 | next_mz = NULL; |
Vladimir Davydov | bc2f2e7 | 2014-12-10 15:43:40 -0800 | [diff] [blame] | 3063 | if (!reclaimed) |
| 3064 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); |
| 3065 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3066 | excess = soft_limit_excess(mz->memcg); |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 3067 | /* |
| 3068 | * One school of thought says that we should not add |
| 3069 | * back the node to the tree if reclaim returns 0. |
| 3070 | * But our reclaim could return 0, simply because due |
| 3071 | * to priority we are exposing a smaller subset of |
| 3072 | * memory to reclaim from. Consider this as a longer |
| 3073 | * term TODO. |
| 3074 | */ |
| 3075 | /* If excess == 0, no tree ops */ |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 3076 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 3077 | spin_unlock_irq(&mctz->lock); |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 3078 | css_put(&mz->memcg->css); |
| 3079 | loop++; |
| 3080 | /* |
| 3081 | * Could not reclaim anything and there are no more |
| 3082 | * mem cgroups to try or we seem to be looping without |
| 3083 | * reclaiming anything. |
| 3084 | */ |
| 3085 | if (!nr_reclaimed && |
| 3086 | (next_mz == NULL || |
| 3087 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) |
| 3088 | break; |
| 3089 | } while (!nr_reclaimed); |
| 3090 | if (next_mz) |
| 3091 | css_put(&next_mz->memcg->css); |
| 3092 | return nr_reclaimed; |
| 3093 | } |
| 3094 | |
Tejun Heo | ea280e7 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 3095 | /* |
| 3096 | * Test whether @memcg has children, dead or alive. Note that this |
| 3097 | * function doesn't care whether @memcg has use_hierarchy enabled and |
| 3098 | * returns %true if there are child csses according to the cgroup |
| 3099 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. |
| 3100 | */ |
Glauber Costa | b5f99b5 | 2013-02-22 16:34:53 -0800 | [diff] [blame] | 3101 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
| 3102 | { |
Tejun Heo | ea280e7 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 3103 | bool ret; |
| 3104 | |
Johannes Weiner | 696ac17 | 2013-10-31 16:34:15 -0700 | [diff] [blame] | 3105 | /* |
Tejun Heo | ea280e7 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 3106 | * The lock does not prevent addition or deletion of children, but |
| 3107 | * it prevents a new child from being initialized based on this |
| 3108 | * parent in css_online(), so it's enough to decide whether |
| 3109 | * hierarchically inherited attributes can still be changed or not. |
Johannes Weiner | 696ac17 | 2013-10-31 16:34:15 -0700 | [diff] [blame] | 3110 | */ |
Tejun Heo | ea280e7 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 3111 | lockdep_assert_held(&memcg_create_mutex); |
| 3112 | |
| 3113 | rcu_read_lock(); |
| 3114 | ret = css_next_child(NULL, &memcg->css); |
| 3115 | rcu_read_unlock(); |
| 3116 | return ret; |
Glauber Costa | b5f99b5 | 2013-02-22 16:34:53 -0800 | [diff] [blame] | 3117 | } |
| 3118 | |
| 3119 | /* |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 3120 | * Reclaims as many pages from the given memcg as possible and moves |
| 3121 | * the rest to the parent. |
| 3122 | * |
| 3123 | * Caller is responsible for holding css reference for memcg. |
| 3124 | */ |
| 3125 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) |
| 3126 | { |
| 3127 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 3128 | |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3129 | /* we call try-to-free pages for make this cgroup empty */ |
| 3130 | lru_add_drain_all(); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3131 | /* try to free all pages in this cgroup */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3132 | while (nr_retries && page_counter_read(&memcg->memory)) { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3133 | int progress; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3134 | |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 3135 | if (signal_pending(current)) |
| 3136 | return -EINTR; |
| 3137 | |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 3138 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
| 3139 | GFP_KERNEL, true); |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3140 | if (!progress) { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3141 | nr_retries--; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3142 | /* maybe some writeback is necessary */ |
Jens Axboe | 8aa7e84 | 2009-07-09 14:52:32 +0200 | [diff] [blame] | 3143 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3144 | } |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3145 | |
| 3146 | } |
Michal Hocko | ab5196c | 2012-10-26 13:37:32 +0200 | [diff] [blame] | 3147 | |
| 3148 | return 0; |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3149 | } |
| 3150 | |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3151 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
| 3152 | char *buf, size_t nbytes, |
| 3153 | loff_t off) |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3154 | { |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3155 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 3156 | |
Michal Hocko | d842301 | 2012-10-26 13:37:29 +0200 | [diff] [blame] | 3157 | if (mem_cgroup_is_root(memcg)) |
| 3158 | return -EINVAL; |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3159 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3160 | } |
| 3161 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3162 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
| 3163 | struct cftype *cft) |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3164 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3165 | return mem_cgroup_from_css(css)->use_hierarchy; |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3166 | } |
| 3167 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3168 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
| 3169 | struct cftype *cft, u64 val) |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3170 | { |
| 3171 | int retval = 0; |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3172 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Tejun Heo | 5c9d535 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 3173 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3174 | |
Glauber Costa | 0999821 | 2013-02-22 16:34:55 -0800 | [diff] [blame] | 3175 | mutex_lock(&memcg_create_mutex); |
Glauber Costa | 567fb43 | 2012-07-31 16:43:07 -0700 | [diff] [blame] | 3176 | |
| 3177 | if (memcg->use_hierarchy == val) |
| 3178 | goto out; |
| 3179 | |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3180 | /* |
André Goddard Rosa | af901ca | 2009-11-14 13:09:05 -0200 | [diff] [blame] | 3181 | * If parent's use_hierarchy is set, we can't make any modifications |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3182 | * in the child subtrees. If it is unset, then the change can |
| 3183 | * occur, provided the current cgroup has no children. |
| 3184 | * |
| 3185 | * For the root cgroup, parent_mem is NULL, we allow value to be |
| 3186 | * set if there are no children. |
| 3187 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3188 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3189 | (val == 1 || val == 0)) { |
Tejun Heo | ea280e7 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 3190 | if (!memcg_has_children(memcg)) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3191 | memcg->use_hierarchy = val; |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3192 | else |
| 3193 | retval = -EBUSY; |
| 3194 | } else |
| 3195 | retval = -EINVAL; |
Glauber Costa | 567fb43 | 2012-07-31 16:43:07 -0700 | [diff] [blame] | 3196 | |
| 3197 | out: |
Glauber Costa | 0999821 | 2013-02-22 16:34:55 -0800 | [diff] [blame] | 3198 | mutex_unlock(&memcg_create_mutex); |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3199 | |
| 3200 | return retval; |
| 3201 | } |
| 3202 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3203 | static unsigned long tree_stat(struct mem_cgroup *memcg, |
| 3204 | enum mem_cgroup_stat_index idx) |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3205 | { |
| 3206 | struct mem_cgroup *iter; |
| 3207 | long val = 0; |
| 3208 | |
| 3209 | /* Per-cpu values can be negative, use a signed accumulator */ |
| 3210 | for_each_mem_cgroup_tree(iter, memcg) |
| 3211 | val += mem_cgroup_read_stat(iter, idx); |
| 3212 | |
| 3213 | if (val < 0) /* race ? */ |
| 3214 | val = 0; |
| 3215 | return val; |
| 3216 | } |
| 3217 | |
| 3218 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
| 3219 | { |
| 3220 | u64 val; |
| 3221 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3222 | if (mem_cgroup_is_root(memcg)) { |
| 3223 | val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE); |
| 3224 | val += tree_stat(memcg, MEM_CGROUP_STAT_RSS); |
| 3225 | if (swap) |
| 3226 | val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP); |
| 3227 | } else { |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3228 | if (!swap) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3229 | val = page_counter_read(&memcg->memory); |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3230 | else |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3231 | val = page_counter_read(&memcg->memsw); |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3232 | } |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3233 | return val << PAGE_SHIFT; |
| 3234 | } |
| 3235 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3236 | enum { |
| 3237 | RES_USAGE, |
| 3238 | RES_LIMIT, |
| 3239 | RES_MAX_USAGE, |
| 3240 | RES_FAILCNT, |
| 3241 | RES_SOFT_LIMIT, |
| 3242 | }; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3243 | |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 3244 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3245 | struct cftype *cft) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3246 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3247 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3248 | struct page_counter *counter; |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3249 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3250 | switch (MEMFILE_TYPE(cft->private)) { |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3251 | case _MEM: |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3252 | counter = &memcg->memory; |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3253 | break; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3254 | case _MEMSWAP: |
| 3255 | counter = &memcg->memsw; |
| 3256 | break; |
| 3257 | case _KMEM: |
| 3258 | counter = &memcg->kmem; |
| 3259 | break; |
| 3260 | default: |
| 3261 | BUG(); |
| 3262 | } |
| 3263 | |
| 3264 | switch (MEMFILE_ATTR(cft->private)) { |
| 3265 | case RES_USAGE: |
| 3266 | if (counter == &memcg->memory) |
| 3267 | return mem_cgroup_usage(memcg, false); |
| 3268 | if (counter == &memcg->memsw) |
| 3269 | return mem_cgroup_usage(memcg, true); |
| 3270 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
| 3271 | case RES_LIMIT: |
| 3272 | return (u64)counter->limit * PAGE_SIZE; |
| 3273 | case RES_MAX_USAGE: |
| 3274 | return (u64)counter->watermark * PAGE_SIZE; |
| 3275 | case RES_FAILCNT: |
| 3276 | return counter->failcnt; |
| 3277 | case RES_SOFT_LIMIT: |
| 3278 | return (u64)memcg->soft_limit * PAGE_SIZE; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3279 | default: |
| 3280 | BUG(); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3281 | } |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3282 | } |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3283 | |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3284 | #ifdef CONFIG_MEMCG_KMEM |
Vladimir Davydov | 8c0145b | 2014-12-10 15:43:48 -0800 | [diff] [blame] | 3285 | static int memcg_activate_kmem(struct mem_cgroup *memcg, |
| 3286 | unsigned long nr_pages) |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3287 | { |
| 3288 | int err = 0; |
| 3289 | int memcg_id; |
| 3290 | |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 3291 | BUG_ON(memcg->kmemcg_id >= 0); |
Vladimir Davydov | 2788cf0 | 2015-02-12 14:59:38 -0800 | [diff] [blame] | 3292 | BUG_ON(memcg->kmem_acct_activated); |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 3293 | BUG_ON(memcg->kmem_acct_active); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3294 | |
| 3295 | /* |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3296 | * For simplicity, we won't allow this to be disabled. It also can't |
| 3297 | * be changed if the cgroup has children already, or if tasks had |
| 3298 | * already joined. |
| 3299 | * |
| 3300 | * If tasks join before we set the limit, a person looking at |
| 3301 | * kmem.usage_in_bytes will have no way to determine when it took |
| 3302 | * place, which makes the value quite meaningless. |
| 3303 | * |
| 3304 | * After it first became limited, changes in the value of the limit are |
| 3305 | * of course permitted. |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3306 | */ |
Glauber Costa | 0999821 | 2013-02-22 16:34:55 -0800 | [diff] [blame] | 3307 | mutex_lock(&memcg_create_mutex); |
Tejun Heo | ea280e7 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 3308 | if (cgroup_has_tasks(memcg->css.cgroup) || |
| 3309 | (memcg->use_hierarchy && memcg_has_children(memcg))) |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3310 | err = -EBUSY; |
Glauber Costa | 0999821 | 2013-02-22 16:34:55 -0800 | [diff] [blame] | 3311 | mutex_unlock(&memcg_create_mutex); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3312 | if (err) |
| 3313 | goto out; |
| 3314 | |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 3315 | memcg_id = memcg_alloc_cache_id(); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3316 | if (memcg_id < 0) { |
| 3317 | err = memcg_id; |
| 3318 | goto out; |
| 3319 | } |
| 3320 | |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3321 | /* |
Vladimir Davydov | 900a38f | 2014-12-12 16:55:10 -0800 | [diff] [blame] | 3322 | * We couldn't have accounted to this cgroup, because it hasn't got |
| 3323 | * activated yet, so this should succeed. |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3324 | */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3325 | err = page_counter_limit(&memcg->kmem, nr_pages); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3326 | VM_BUG_ON(err); |
| 3327 | |
| 3328 | static_key_slow_inc(&memcg_kmem_enabled_key); |
| 3329 | /* |
Vladimir Davydov | 900a38f | 2014-12-12 16:55:10 -0800 | [diff] [blame] | 3330 | * A memory cgroup is considered kmem-active as soon as it gets |
| 3331 | * kmemcg_id. Setting the id after enabling static branching will |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3332 | * guarantee no one starts accounting before all call sites are |
| 3333 | * patched. |
| 3334 | */ |
Vladimir Davydov | 900a38f | 2014-12-12 16:55:10 -0800 | [diff] [blame] | 3335 | memcg->kmemcg_id = memcg_id; |
Vladimir Davydov | 2788cf0 | 2015-02-12 14:59:38 -0800 | [diff] [blame] | 3336 | memcg->kmem_acct_activated = true; |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 3337 | memcg->kmem_acct_active = true; |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3338 | out: |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3339 | return err; |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3340 | } |
| 3341 | |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3342 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3343 | unsigned long limit) |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3344 | { |
| 3345 | int ret; |
| 3346 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3347 | mutex_lock(&memcg_limit_mutex); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3348 | if (!memcg_kmem_is_active(memcg)) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3349 | ret = memcg_activate_kmem(memcg, limit); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3350 | else |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3351 | ret = page_counter_limit(&memcg->kmem, limit); |
| 3352 | mutex_unlock(&memcg_limit_mutex); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3353 | return ret; |
| 3354 | } |
| 3355 | |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 3356 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3357 | { |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 3358 | int ret = 0; |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3359 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3360 | |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3361 | if (!parent) |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3362 | return 0; |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 3363 | |
Vladimir Davydov | 8c0145b | 2014-12-10 15:43:48 -0800 | [diff] [blame] | 3364 | mutex_lock(&memcg_limit_mutex); |
Glauber Costa | a8964b9 | 2012-12-18 14:22:09 -0800 | [diff] [blame] | 3365 | /* |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3366 | * If the parent cgroup is not kmem-active now, it cannot be activated |
| 3367 | * after this point, because it has at least one child already. |
Glauber Costa | a8964b9 | 2012-12-18 14:22:09 -0800 | [diff] [blame] | 3368 | */ |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3369 | if (memcg_kmem_is_active(parent)) |
Vladimir Davydov | 8c0145b | 2014-12-10 15:43:48 -0800 | [diff] [blame] | 3370 | ret = memcg_activate_kmem(memcg, PAGE_COUNTER_MAX); |
| 3371 | mutex_unlock(&memcg_limit_mutex); |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 3372 | return ret; |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3373 | } |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3374 | #else |
| 3375 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3376 | unsigned long limit) |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3377 | { |
| 3378 | return -EINVAL; |
| 3379 | } |
Hugh Dickins | 6d043990 | 2013-02-22 16:35:50 -0800 | [diff] [blame] | 3380 | #endif /* CONFIG_MEMCG_KMEM */ |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3381 | |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3382 | /* |
| 3383 | * The user of this function is... |
| 3384 | * RES_LIMIT. |
| 3385 | */ |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3386 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
| 3387 | char *buf, size_t nbytes, loff_t off) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3388 | { |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3389 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3390 | unsigned long nr_pages; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3391 | int ret; |
| 3392 | |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3393 | buf = strstrip(buf); |
Johannes Weiner | 650c5e5 | 2015-02-11 15:26:03 -0800 | [diff] [blame] | 3394 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3395 | if (ret) |
| 3396 | return ret; |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3397 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3398 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3399 | case RES_LIMIT: |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 3400 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
| 3401 | ret = -EINVAL; |
| 3402 | break; |
| 3403 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3404 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
| 3405 | case _MEM: |
| 3406 | ret = mem_cgroup_resize_limit(memcg, nr_pages); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3407 | break; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3408 | case _MEMSWAP: |
| 3409 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); |
| 3410 | break; |
| 3411 | case _KMEM: |
| 3412 | ret = memcg_update_kmem_limit(memcg, nr_pages); |
| 3413 | break; |
| 3414 | } |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3415 | break; |
Balbir Singh | 296c81d | 2009-09-23 15:56:36 -0700 | [diff] [blame] | 3416 | case RES_SOFT_LIMIT: |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3417 | memcg->soft_limit = nr_pages; |
| 3418 | ret = 0; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3419 | break; |
| 3420 | } |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3421 | return ret ?: nbytes; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3422 | } |
| 3423 | |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3424 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
| 3425 | size_t nbytes, loff_t off) |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 3426 | { |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3427 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3428 | struct page_counter *counter; |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 3429 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3430 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
| 3431 | case _MEM: |
| 3432 | counter = &memcg->memory; |
| 3433 | break; |
| 3434 | case _MEMSWAP: |
| 3435 | counter = &memcg->memsw; |
| 3436 | break; |
| 3437 | case _KMEM: |
| 3438 | counter = &memcg->kmem; |
| 3439 | break; |
| 3440 | default: |
| 3441 | BUG(); |
| 3442 | } |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3443 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3444 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3445 | case RES_MAX_USAGE: |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3446 | page_counter_reset_watermark(counter); |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3447 | break; |
| 3448 | case RES_FAILCNT: |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3449 | counter->failcnt = 0; |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3450 | break; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3451 | default: |
| 3452 | BUG(); |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3453 | } |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 3454 | |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3455 | return nbytes; |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 3456 | } |
| 3457 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3458 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3459 | struct cftype *cft) |
| 3460 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3461 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3462 | } |
| 3463 | |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3464 | #ifdef CONFIG_MMU |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3465 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3466 | struct cftype *cft, u64 val) |
| 3467 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3468 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3469 | |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 3470 | if (val & ~MOVE_MASK) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3471 | return -EINVAL; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3472 | |
Glauber Costa | ee5e847 | 2013-02-22 16:34:50 -0800 | [diff] [blame] | 3473 | /* |
| 3474 | * No kind of locking is needed in here, because ->can_attach() will |
| 3475 | * check this value once in the beginning of the process, and then carry |
| 3476 | * on with stale data. This means that changes to this value will only |
| 3477 | * affect task migrations starting after the change. |
| 3478 | */ |
| 3479 | memcg->move_charge_at_immigrate = val; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3480 | return 0; |
| 3481 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3482 | #else |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3483 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3484 | struct cftype *cft, u64 val) |
| 3485 | { |
| 3486 | return -ENOSYS; |
| 3487 | } |
| 3488 | #endif |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3489 | |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3490 | #ifdef CONFIG_NUMA |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3491 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3492 | { |
Greg Thelen | 25485de | 2013-11-12 15:07:40 -0800 | [diff] [blame] | 3493 | struct numa_stat { |
| 3494 | const char *name; |
| 3495 | unsigned int lru_mask; |
| 3496 | }; |
| 3497 | |
| 3498 | static const struct numa_stat stats[] = { |
| 3499 | { "total", LRU_ALL }, |
| 3500 | { "file", LRU_ALL_FILE }, |
| 3501 | { "anon", LRU_ALL_ANON }, |
| 3502 | { "unevictable", BIT(LRU_UNEVICTABLE) }, |
| 3503 | }; |
| 3504 | const struct numa_stat *stat; |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3505 | int nid; |
Greg Thelen | 25485de | 2013-11-12 15:07:40 -0800 | [diff] [blame] | 3506 | unsigned long nr; |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3507 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3508 | |
Greg Thelen | 25485de | 2013-11-12 15:07:40 -0800 | [diff] [blame] | 3509 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
| 3510 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); |
| 3511 | seq_printf(m, "%s=%lu", stat->name, nr); |
| 3512 | for_each_node_state(nid, N_MEMORY) { |
| 3513 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
| 3514 | stat->lru_mask); |
| 3515 | seq_printf(m, " N%d=%lu", nid, nr); |
| 3516 | } |
| 3517 | seq_putc(m, '\n'); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3518 | } |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3519 | |
Ying Han | 071aee1 | 2013-11-12 15:07:41 -0800 | [diff] [blame] | 3520 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
| 3521 | struct mem_cgroup *iter; |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3522 | |
Ying Han | 071aee1 | 2013-11-12 15:07:41 -0800 | [diff] [blame] | 3523 | nr = 0; |
| 3524 | for_each_mem_cgroup_tree(iter, memcg) |
| 3525 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); |
| 3526 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); |
| 3527 | for_each_node_state(nid, N_MEMORY) { |
| 3528 | nr = 0; |
| 3529 | for_each_mem_cgroup_tree(iter, memcg) |
| 3530 | nr += mem_cgroup_node_nr_lru_pages( |
| 3531 | iter, nid, stat->lru_mask); |
| 3532 | seq_printf(m, " N%d=%lu", nid, nr); |
| 3533 | } |
| 3534 | seq_putc(m, '\n'); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3535 | } |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3536 | |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3537 | return 0; |
| 3538 | } |
| 3539 | #endif /* CONFIG_NUMA */ |
| 3540 | |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3541 | static int memcg_stat_show(struct seq_file *m, void *v) |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3542 | { |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3543 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3544 | unsigned long memory, memsw; |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3545 | struct mem_cgroup *mi; |
| 3546 | unsigned int i; |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3547 | |
Greg Thelen | 0ca44b1 | 2015-02-11 15:25:58 -0800 | [diff] [blame] | 3548 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) != |
| 3549 | MEM_CGROUP_STAT_NSTATS); |
| 3550 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) != |
| 3551 | MEM_CGROUP_EVENTS_NSTATS); |
Rickard Strandqvist | 70bc068 | 2014-12-12 16:56:41 -0800 | [diff] [blame] | 3552 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
| 3553 | |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3554 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
Kamezawa Hiroyuki | bff6bb8 | 2012-07-31 16:41:38 -0700 | [diff] [blame] | 3555 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3556 | continue; |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3557 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
| 3558 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3559 | } |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 3560 | |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3561 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
| 3562 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], |
| 3563 | mem_cgroup_read_events(memcg, i)); |
| 3564 | |
| 3565 | for (i = 0; i < NR_LRU_LISTS; i++) |
| 3566 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], |
| 3567 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); |
| 3568 | |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3569 | /* Hierarchical information */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3570 | memory = memsw = PAGE_COUNTER_MAX; |
| 3571 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { |
| 3572 | memory = min(memory, mi->memory.limit); |
| 3573 | memsw = min(memsw, mi->memsw.limit); |
KAMEZAWA Hiroyuki | fee7b54 | 2009-01-07 18:08:26 -0800 | [diff] [blame] | 3574 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3575 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
| 3576 | (u64)memory * PAGE_SIZE); |
| 3577 | if (do_swap_account) |
| 3578 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
| 3579 | (u64)memsw * PAGE_SIZE); |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3580 | |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3581 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
| 3582 | long long val = 0; |
| 3583 | |
Kamezawa Hiroyuki | bff6bb8 | 2012-07-31 16:41:38 -0700 | [diff] [blame] | 3584 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3585 | continue; |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3586 | for_each_mem_cgroup_tree(mi, memcg) |
| 3587 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; |
| 3588 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); |
| 3589 | } |
| 3590 | |
| 3591 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
| 3592 | unsigned long long val = 0; |
| 3593 | |
| 3594 | for_each_mem_cgroup_tree(mi, memcg) |
| 3595 | val += mem_cgroup_read_events(mi, i); |
| 3596 | seq_printf(m, "total_%s %llu\n", |
| 3597 | mem_cgroup_events_names[i], val); |
| 3598 | } |
| 3599 | |
| 3600 | for (i = 0; i < NR_LRU_LISTS; i++) { |
| 3601 | unsigned long long val = 0; |
| 3602 | |
| 3603 | for_each_mem_cgroup_tree(mi, memcg) |
| 3604 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; |
| 3605 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3606 | } |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3607 | |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3608 | #ifdef CONFIG_DEBUG_VM |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3609 | { |
| 3610 | int nid, zid; |
| 3611 | struct mem_cgroup_per_zone *mz; |
Hugh Dickins | 89abfab | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 3612 | struct zone_reclaim_stat *rstat; |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3613 | unsigned long recent_rotated[2] = {0, 0}; |
| 3614 | unsigned long recent_scanned[2] = {0, 0}; |
| 3615 | |
| 3616 | for_each_online_node(nid) |
| 3617 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 3618 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
Hugh Dickins | 89abfab | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 3619 | rstat = &mz->lruvec.reclaim_stat; |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3620 | |
Hugh Dickins | 89abfab | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 3621 | recent_rotated[0] += rstat->recent_rotated[0]; |
| 3622 | recent_rotated[1] += rstat->recent_rotated[1]; |
| 3623 | recent_scanned[0] += rstat->recent_scanned[0]; |
| 3624 | recent_scanned[1] += rstat->recent_scanned[1]; |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3625 | } |
Johannes Weiner | 78ccf5b | 2012-05-29 15:07:06 -0700 | [diff] [blame] | 3626 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
| 3627 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); |
| 3628 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); |
| 3629 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3630 | } |
| 3631 | #endif |
| 3632 | |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3633 | return 0; |
| 3634 | } |
| 3635 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3636 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
| 3637 | struct cftype *cft) |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3638 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3639 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3640 | |
KAMEZAWA Hiroyuki | 1f4c025 | 2011-07-26 16:08:21 -0700 | [diff] [blame] | 3641 | return mem_cgroup_swappiness(memcg); |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3642 | } |
| 3643 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3644 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
| 3645 | struct cftype *cft, u64 val) |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3646 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3647 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 3648 | |
Johannes Weiner | 3dae7fe | 2014-06-04 16:07:01 -0700 | [diff] [blame] | 3649 | if (val > 100) |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3650 | return -EINVAL; |
| 3651 | |
Linus Torvalds | 14208b0 | 2014-06-09 15:03:33 -0700 | [diff] [blame] | 3652 | if (css->parent) |
Johannes Weiner | 3dae7fe | 2014-06-04 16:07:01 -0700 | [diff] [blame] | 3653 | memcg->swappiness = val; |
| 3654 | else |
| 3655 | vm_swappiness = val; |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 3656 | |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3657 | return 0; |
| 3658 | } |
| 3659 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3660 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
| 3661 | { |
| 3662 | struct mem_cgroup_threshold_ary *t; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3663 | unsigned long usage; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3664 | int i; |
| 3665 | |
| 3666 | rcu_read_lock(); |
| 3667 | if (!swap) |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3668 | t = rcu_dereference(memcg->thresholds.primary); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3669 | else |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3670 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3671 | |
| 3672 | if (!t) |
| 3673 | goto unlock; |
| 3674 | |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3675 | usage = mem_cgroup_usage(memcg, swap); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3676 | |
| 3677 | /* |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3678 | * current_threshold points to threshold just below or equal to usage. |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3679 | * If it's not true, a threshold was crossed after last |
| 3680 | * call of __mem_cgroup_threshold(). |
| 3681 | */ |
Phil Carmody | 5407a56 | 2010-05-26 14:42:42 -0700 | [diff] [blame] | 3682 | i = t->current_threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3683 | |
| 3684 | /* |
| 3685 | * Iterate backward over array of thresholds starting from |
| 3686 | * current_threshold and check if a threshold is crossed. |
| 3687 | * If none of thresholds below usage is crossed, we read |
| 3688 | * only one element of the array here. |
| 3689 | */ |
| 3690 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) |
| 3691 | eventfd_signal(t->entries[i].eventfd, 1); |
| 3692 | |
| 3693 | /* i = current_threshold + 1 */ |
| 3694 | i++; |
| 3695 | |
| 3696 | /* |
| 3697 | * Iterate forward over array of thresholds starting from |
| 3698 | * current_threshold+1 and check if a threshold is crossed. |
| 3699 | * If none of thresholds above usage is crossed, we read |
| 3700 | * only one element of the array here. |
| 3701 | */ |
| 3702 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) |
| 3703 | eventfd_signal(t->entries[i].eventfd, 1); |
| 3704 | |
| 3705 | /* Update current_threshold */ |
Phil Carmody | 5407a56 | 2010-05-26 14:42:42 -0700 | [diff] [blame] | 3706 | t->current_threshold = i - 1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3707 | unlock: |
| 3708 | rcu_read_unlock(); |
| 3709 | } |
| 3710 | |
| 3711 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) |
| 3712 | { |
Kirill A. Shutemov | ad4ca5f | 2010-10-07 12:59:27 -0700 | [diff] [blame] | 3713 | while (memcg) { |
| 3714 | __mem_cgroup_threshold(memcg, false); |
| 3715 | if (do_swap_account) |
| 3716 | __mem_cgroup_threshold(memcg, true); |
| 3717 | |
| 3718 | memcg = parent_mem_cgroup(memcg); |
| 3719 | } |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3720 | } |
| 3721 | |
| 3722 | static int compare_thresholds(const void *a, const void *b) |
| 3723 | { |
| 3724 | const struct mem_cgroup_threshold *_a = a; |
| 3725 | const struct mem_cgroup_threshold *_b = b; |
| 3726 | |
Greg Thelen | 2bff24a | 2013-09-11 14:23:08 -0700 | [diff] [blame] | 3727 | if (_a->threshold > _b->threshold) |
| 3728 | return 1; |
| 3729 | |
| 3730 | if (_a->threshold < _b->threshold) |
| 3731 | return -1; |
| 3732 | |
| 3733 | return 0; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3734 | } |
| 3735 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3736 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3737 | { |
| 3738 | struct mem_cgroup_eventfd_list *ev; |
| 3739 | |
Michal Hocko | 2bcf2e9 | 2014-07-30 16:08:33 -0700 | [diff] [blame] | 3740 | spin_lock(&memcg_oom_lock); |
| 3741 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3742 | list_for_each_entry(ev, &memcg->oom_notify, list) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3743 | eventfd_signal(ev->eventfd, 1); |
Michal Hocko | 2bcf2e9 | 2014-07-30 16:08:33 -0700 | [diff] [blame] | 3744 | |
| 3745 | spin_unlock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3746 | return 0; |
| 3747 | } |
| 3748 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3749 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3750 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3751 | struct mem_cgroup *iter; |
| 3752 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3753 | for_each_mem_cgroup_tree(iter, memcg) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3754 | mem_cgroup_oom_notify_cb(iter); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3755 | } |
| 3756 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3757 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3758 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3759 | { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3760 | struct mem_cgroup_thresholds *thresholds; |
| 3761 | struct mem_cgroup_threshold_ary *new; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3762 | unsigned long threshold; |
| 3763 | unsigned long usage; |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3764 | int i, size, ret; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3765 | |
Johannes Weiner | 650c5e5 | 2015-02-11 15:26:03 -0800 | [diff] [blame] | 3766 | ret = page_counter_memparse(args, "-1", &threshold); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3767 | if (ret) |
| 3768 | return ret; |
| 3769 | |
| 3770 | mutex_lock(&memcg->thresholds_lock); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3771 | |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3772 | if (type == _MEM) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3773 | thresholds = &memcg->thresholds; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3774 | usage = mem_cgroup_usage(memcg, false); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3775 | } else if (type == _MEMSWAP) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3776 | thresholds = &memcg->memsw_thresholds; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3777 | usage = mem_cgroup_usage(memcg, true); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3778 | } else |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3779 | BUG(); |
| 3780 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3781 | /* Check if a threshold crossed before adding a new one */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3782 | if (thresholds->primary) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3783 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
| 3784 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3785 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3786 | |
| 3787 | /* Allocate memory for new array of thresholds */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3788 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3789 | GFP_KERNEL); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3790 | if (!new) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3791 | ret = -ENOMEM; |
| 3792 | goto unlock; |
| 3793 | } |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3794 | new->size = size; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3795 | |
| 3796 | /* Copy thresholds (if any) to new array */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3797 | if (thresholds->primary) { |
| 3798 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3799 | sizeof(struct mem_cgroup_threshold)); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3800 | } |
| 3801 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3802 | /* Add new threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3803 | new->entries[size - 1].eventfd = eventfd; |
| 3804 | new->entries[size - 1].threshold = threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3805 | |
| 3806 | /* Sort thresholds. Registering of new threshold isn't time-critical */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3807 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3808 | compare_thresholds, NULL); |
| 3809 | |
| 3810 | /* Find current threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3811 | new->current_threshold = -1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3812 | for (i = 0; i < size; i++) { |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3813 | if (new->entries[i].threshold <= usage) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3814 | /* |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3815 | * new->current_threshold will not be used until |
| 3816 | * rcu_assign_pointer(), so it's safe to increment |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3817 | * it here. |
| 3818 | */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3819 | ++new->current_threshold; |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3820 | } else |
| 3821 | break; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3822 | } |
| 3823 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3824 | /* Free old spare buffer and save old primary buffer as spare */ |
| 3825 | kfree(thresholds->spare); |
| 3826 | thresholds->spare = thresholds->primary; |
| 3827 | |
| 3828 | rcu_assign_pointer(thresholds->primary, new); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3829 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3830 | /* To be sure that nobody uses thresholds */ |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3831 | synchronize_rcu(); |
| 3832 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3833 | unlock: |
| 3834 | mutex_unlock(&memcg->thresholds_lock); |
| 3835 | |
| 3836 | return ret; |
| 3837 | } |
| 3838 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3839 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3840 | struct eventfd_ctx *eventfd, const char *args) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3841 | { |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3842 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3843 | } |
| 3844 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3845 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3846 | struct eventfd_ctx *eventfd, const char *args) |
| 3847 | { |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3848 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3849 | } |
| 3850 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3851 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3852 | struct eventfd_ctx *eventfd, enum res_type type) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3853 | { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3854 | struct mem_cgroup_thresholds *thresholds; |
| 3855 | struct mem_cgroup_threshold_ary *new; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3856 | unsigned long usage; |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3857 | int i, j, size; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3858 | |
| 3859 | mutex_lock(&memcg->thresholds_lock); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3860 | |
| 3861 | if (type == _MEM) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3862 | thresholds = &memcg->thresholds; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3863 | usage = mem_cgroup_usage(memcg, false); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3864 | } else if (type == _MEMSWAP) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3865 | thresholds = &memcg->memsw_thresholds; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3866 | usage = mem_cgroup_usage(memcg, true); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3867 | } else |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3868 | BUG(); |
| 3869 | |
Anton Vorontsov | 371528c | 2012-02-24 05:14:46 +0400 | [diff] [blame] | 3870 | if (!thresholds->primary) |
| 3871 | goto unlock; |
| 3872 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3873 | /* Check if a threshold crossed before removing */ |
| 3874 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
| 3875 | |
| 3876 | /* Calculate new number of threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3877 | size = 0; |
| 3878 | for (i = 0; i < thresholds->primary->size; i++) { |
| 3879 | if (thresholds->primary->entries[i].eventfd != eventfd) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3880 | size++; |
| 3881 | } |
| 3882 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3883 | new = thresholds->spare; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3884 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3885 | /* Set thresholds array to NULL if we don't have thresholds */ |
| 3886 | if (!size) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3887 | kfree(new); |
| 3888 | new = NULL; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3889 | goto swap_buffers; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3890 | } |
| 3891 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3892 | new->size = size; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3893 | |
| 3894 | /* Copy thresholds and find current threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3895 | new->current_threshold = -1; |
| 3896 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { |
| 3897 | if (thresholds->primary->entries[i].eventfd == eventfd) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3898 | continue; |
| 3899 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3900 | new->entries[j] = thresholds->primary->entries[i]; |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3901 | if (new->entries[j].threshold <= usage) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3902 | /* |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3903 | * new->current_threshold will not be used |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3904 | * until rcu_assign_pointer(), so it's safe to increment |
| 3905 | * it here. |
| 3906 | */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3907 | ++new->current_threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3908 | } |
| 3909 | j++; |
| 3910 | } |
| 3911 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3912 | swap_buffers: |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3913 | /* Swap primary and spare array */ |
| 3914 | thresholds->spare = thresholds->primary; |
Sha Zhengju | 8c75776 | 2012-05-10 13:01:45 -0700 | [diff] [blame] | 3915 | /* If all events are unregistered, free the spare array */ |
| 3916 | if (!new) { |
| 3917 | kfree(thresholds->spare); |
| 3918 | thresholds->spare = NULL; |
| 3919 | } |
| 3920 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3921 | rcu_assign_pointer(thresholds->primary, new); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3922 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3923 | /* To be sure that nobody uses thresholds */ |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3924 | synchronize_rcu(); |
Anton Vorontsov | 371528c | 2012-02-24 05:14:46 +0400 | [diff] [blame] | 3925 | unlock: |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3926 | mutex_unlock(&memcg->thresholds_lock); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3927 | } |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3928 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3929 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3930 | struct eventfd_ctx *eventfd) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3931 | { |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3932 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3933 | } |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3934 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3935 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3936 | struct eventfd_ctx *eventfd) |
| 3937 | { |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3938 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3939 | } |
| 3940 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3941 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3942 | struct eventfd_ctx *eventfd, const char *args) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3943 | { |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3944 | struct mem_cgroup_eventfd_list *event; |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3945 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3946 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
| 3947 | if (!event) |
| 3948 | return -ENOMEM; |
| 3949 | |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 3950 | spin_lock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3951 | |
| 3952 | event->eventfd = eventfd; |
| 3953 | list_add(&event->list, &memcg->oom_notify); |
| 3954 | |
| 3955 | /* already in OOM ? */ |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 3956 | if (atomic_read(&memcg->under_oom)) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3957 | eventfd_signal(eventfd, 1); |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 3958 | spin_unlock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3959 | |
| 3960 | return 0; |
| 3961 | } |
| 3962 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3963 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3964 | struct eventfd_ctx *eventfd) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3965 | { |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3966 | struct mem_cgroup_eventfd_list *ev, *tmp; |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3967 | |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 3968 | spin_lock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3969 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3970 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3971 | if (ev->eventfd == eventfd) { |
| 3972 | list_del(&ev->list); |
| 3973 | kfree(ev); |
| 3974 | } |
| 3975 | } |
| 3976 | |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 3977 | spin_unlock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3978 | } |
| 3979 | |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3980 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3981 | { |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3982 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3983 | |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 3984 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
| 3985 | seq_printf(sf, "under_oom %d\n", (bool)atomic_read(&memcg->under_oom)); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3986 | return 0; |
| 3987 | } |
| 3988 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3989 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3990 | struct cftype *cft, u64 val) |
| 3991 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3992 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3993 | |
| 3994 | /* cannot set to root cgroup and only 0 and 1 are allowed */ |
Linus Torvalds | 14208b0 | 2014-06-09 15:03:33 -0700 | [diff] [blame] | 3995 | if (!css->parent || !((val == 0) || (val == 1))) |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3996 | return -EINVAL; |
| 3997 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3998 | memcg->oom_kill_disable = val; |
KAMEZAWA Hiroyuki | 4d845eb | 2010-06-29 15:05:18 -0700 | [diff] [blame] | 3999 | if (!val) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4000 | memcg_oom_recover(memcg); |
Johannes Weiner | 3dae7fe | 2014-06-04 16:07:01 -0700 | [diff] [blame] | 4001 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4002 | return 0; |
| 4003 | } |
| 4004 | |
Andrew Morton | c255a45 | 2012-07-31 16:43:02 -0700 | [diff] [blame] | 4005 | #ifdef CONFIG_MEMCG_KMEM |
Glauber Costa | cbe128e3 | 2012-04-09 19:36:34 -0300 | [diff] [blame] | 4006 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
Glauber Costa | e5671df | 2011-12-11 21:47:01 +0000 | [diff] [blame] | 4007 | { |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 4008 | int ret; |
| 4009 | |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 4010 | ret = memcg_propagate_kmem(memcg); |
| 4011 | if (ret) |
| 4012 | return ret; |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 4013 | |
Glauber Costa | 1d62e43 | 2012-04-09 19:36:33 -0300 | [diff] [blame] | 4014 | return mem_cgroup_sockets_init(memcg, ss); |
Michel Lespinasse | 573b400 | 2013-04-29 15:08:13 -0700 | [diff] [blame] | 4015 | } |
Glauber Costa | e5671df | 2011-12-11 21:47:01 +0000 | [diff] [blame] | 4016 | |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 4017 | static void memcg_deactivate_kmem(struct mem_cgroup *memcg) |
| 4018 | { |
Vladimir Davydov | 2788cf0 | 2015-02-12 14:59:38 -0800 | [diff] [blame] | 4019 | struct cgroup_subsys_state *css; |
| 4020 | struct mem_cgroup *parent, *child; |
| 4021 | int kmemcg_id; |
| 4022 | |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 4023 | if (!memcg->kmem_acct_active) |
| 4024 | return; |
| 4025 | |
| 4026 | /* |
| 4027 | * Clear the 'active' flag before clearing memcg_caches arrays entries. |
| 4028 | * Since we take the slab_mutex in memcg_deactivate_kmem_caches(), it |
| 4029 | * guarantees no cache will be created for this cgroup after we are |
| 4030 | * done (see memcg_create_kmem_cache()). |
| 4031 | */ |
| 4032 | memcg->kmem_acct_active = false; |
| 4033 | |
| 4034 | memcg_deactivate_kmem_caches(memcg); |
Vladimir Davydov | 2788cf0 | 2015-02-12 14:59:38 -0800 | [diff] [blame] | 4035 | |
| 4036 | kmemcg_id = memcg->kmemcg_id; |
| 4037 | BUG_ON(kmemcg_id < 0); |
| 4038 | |
| 4039 | parent = parent_mem_cgroup(memcg); |
| 4040 | if (!parent) |
| 4041 | parent = root_mem_cgroup; |
| 4042 | |
| 4043 | /* |
| 4044 | * Change kmemcg_id of this cgroup and all its descendants to the |
| 4045 | * parent's id, and then move all entries from this cgroup's list_lrus |
| 4046 | * to ones of the parent. After we have finished, all list_lrus |
| 4047 | * corresponding to this cgroup are guaranteed to remain empty. The |
| 4048 | * ordering is imposed by list_lru_node->lock taken by |
| 4049 | * memcg_drain_all_list_lrus(). |
| 4050 | */ |
| 4051 | css_for_each_descendant_pre(css, &memcg->css) { |
| 4052 | child = mem_cgroup_from_css(css); |
| 4053 | BUG_ON(child->kmemcg_id != kmemcg_id); |
| 4054 | child->kmemcg_id = parent->kmemcg_id; |
| 4055 | if (!memcg->use_hierarchy) |
| 4056 | break; |
| 4057 | } |
| 4058 | memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id); |
| 4059 | |
| 4060 | memcg_free_cache_id(kmemcg_id); |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 4061 | } |
| 4062 | |
Li Zefan | 10d5ebf | 2013-07-08 16:00:33 -0700 | [diff] [blame] | 4063 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 4064 | { |
Vladimir Davydov | f48b80a | 2015-02-12 14:59:56 -0800 | [diff] [blame] | 4065 | if (memcg->kmem_acct_activated) { |
| 4066 | memcg_destroy_kmem_caches(memcg); |
| 4067 | static_key_slow_dec(&memcg_kmem_enabled_key); |
| 4068 | WARN_ON(page_counter_read(&memcg->kmem)); |
| 4069 | } |
Glauber Costa | 1d62e43 | 2012-04-09 19:36:33 -0300 | [diff] [blame] | 4070 | mem_cgroup_sockets_destroy(memcg); |
Li Zefan | 10d5ebf | 2013-07-08 16:00:33 -0700 | [diff] [blame] | 4071 | } |
Glauber Costa | e5671df | 2011-12-11 21:47:01 +0000 | [diff] [blame] | 4072 | #else |
Glauber Costa | cbe128e3 | 2012-04-09 19:36:34 -0300 | [diff] [blame] | 4073 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
Glauber Costa | e5671df | 2011-12-11 21:47:01 +0000 | [diff] [blame] | 4074 | { |
| 4075 | return 0; |
| 4076 | } |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 4077 | |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 4078 | static void memcg_deactivate_kmem(struct mem_cgroup *memcg) |
| 4079 | { |
| 4080 | } |
| 4081 | |
Li Zefan | 10d5ebf | 2013-07-08 16:00:33 -0700 | [diff] [blame] | 4082 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
| 4083 | { |
| 4084 | } |
Glauber Costa | e5671df | 2011-12-11 21:47:01 +0000 | [diff] [blame] | 4085 | #endif |
| 4086 | |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4087 | /* |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4088 | * DO NOT USE IN NEW FILES. |
| 4089 | * |
| 4090 | * "cgroup.event_control" implementation. |
| 4091 | * |
| 4092 | * This is way over-engineered. It tries to support fully configurable |
| 4093 | * events for each user. Such level of flexibility is completely |
| 4094 | * unnecessary especially in the light of the planned unified hierarchy. |
| 4095 | * |
| 4096 | * Please deprecate this and replace with something simpler if at all |
| 4097 | * possible. |
| 4098 | */ |
| 4099 | |
| 4100 | /* |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4101 | * Unregister event and free resources. |
| 4102 | * |
| 4103 | * Gets called from workqueue. |
| 4104 | */ |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4105 | static void memcg_event_remove(struct work_struct *work) |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4106 | { |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4107 | struct mem_cgroup_event *event = |
| 4108 | container_of(work, struct mem_cgroup_event, remove); |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4109 | struct mem_cgroup *memcg = event->memcg; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4110 | |
| 4111 | remove_wait_queue(event->wqh, &event->wait); |
| 4112 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4113 | event->unregister_event(memcg, event->eventfd); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4114 | |
| 4115 | /* Notify userspace the event is going away. */ |
| 4116 | eventfd_signal(event->eventfd, 1); |
| 4117 | |
| 4118 | eventfd_ctx_put(event->eventfd); |
| 4119 | kfree(event); |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4120 | css_put(&memcg->css); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4121 | } |
| 4122 | |
| 4123 | /* |
| 4124 | * Gets called on POLLHUP on eventfd when user closes it. |
| 4125 | * |
| 4126 | * Called with wqh->lock held and interrupts disabled. |
| 4127 | */ |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4128 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
| 4129 | int sync, void *key) |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4130 | { |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4131 | struct mem_cgroup_event *event = |
| 4132 | container_of(wait, struct mem_cgroup_event, wait); |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4133 | struct mem_cgroup *memcg = event->memcg; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4134 | unsigned long flags = (unsigned long)key; |
| 4135 | |
| 4136 | if (flags & POLLHUP) { |
| 4137 | /* |
| 4138 | * If the event has been detached at cgroup removal, we |
| 4139 | * can simply return knowing the other side will cleanup |
| 4140 | * for us. |
| 4141 | * |
| 4142 | * We can't race against event freeing since the other |
| 4143 | * side will require wqh->lock via remove_wait_queue(), |
| 4144 | * which we hold. |
| 4145 | */ |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4146 | spin_lock(&memcg->event_list_lock); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4147 | if (!list_empty(&event->list)) { |
| 4148 | list_del_init(&event->list); |
| 4149 | /* |
| 4150 | * We are in atomic context, but cgroup_event_remove() |
| 4151 | * may sleep, so we have to call it in workqueue. |
| 4152 | */ |
| 4153 | schedule_work(&event->remove); |
| 4154 | } |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4155 | spin_unlock(&memcg->event_list_lock); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4156 | } |
| 4157 | |
| 4158 | return 0; |
| 4159 | } |
| 4160 | |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4161 | static void memcg_event_ptable_queue_proc(struct file *file, |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4162 | wait_queue_head_t *wqh, poll_table *pt) |
| 4163 | { |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4164 | struct mem_cgroup_event *event = |
| 4165 | container_of(pt, struct mem_cgroup_event, pt); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4166 | |
| 4167 | event->wqh = wqh; |
| 4168 | add_wait_queue(wqh, &event->wait); |
| 4169 | } |
| 4170 | |
| 4171 | /* |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4172 | * DO NOT USE IN NEW FILES. |
| 4173 | * |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4174 | * Parse input and register new cgroup event handler. |
| 4175 | * |
| 4176 | * Input must be in format '<event_fd> <control_fd> <args>'. |
| 4177 | * Interpretation of args is defined by control file implementation. |
| 4178 | */ |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4179 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
| 4180 | char *buf, size_t nbytes, loff_t off) |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4181 | { |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4182 | struct cgroup_subsys_state *css = of_css(of); |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4183 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4184 | struct mem_cgroup_event *event; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4185 | struct cgroup_subsys_state *cfile_css; |
| 4186 | unsigned int efd, cfd; |
| 4187 | struct fd efile; |
| 4188 | struct fd cfile; |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4189 | const char *name; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4190 | char *endp; |
| 4191 | int ret; |
| 4192 | |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4193 | buf = strstrip(buf); |
| 4194 | |
| 4195 | efd = simple_strtoul(buf, &endp, 10); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4196 | if (*endp != ' ') |
| 4197 | return -EINVAL; |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4198 | buf = endp + 1; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4199 | |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4200 | cfd = simple_strtoul(buf, &endp, 10); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4201 | if ((*endp != ' ') && (*endp != '\0')) |
| 4202 | return -EINVAL; |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4203 | buf = endp + 1; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4204 | |
| 4205 | event = kzalloc(sizeof(*event), GFP_KERNEL); |
| 4206 | if (!event) |
| 4207 | return -ENOMEM; |
| 4208 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4209 | event->memcg = memcg; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4210 | INIT_LIST_HEAD(&event->list); |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4211 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
| 4212 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); |
| 4213 | INIT_WORK(&event->remove, memcg_event_remove); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4214 | |
| 4215 | efile = fdget(efd); |
| 4216 | if (!efile.file) { |
| 4217 | ret = -EBADF; |
| 4218 | goto out_kfree; |
| 4219 | } |
| 4220 | |
| 4221 | event->eventfd = eventfd_ctx_fileget(efile.file); |
| 4222 | if (IS_ERR(event->eventfd)) { |
| 4223 | ret = PTR_ERR(event->eventfd); |
| 4224 | goto out_put_efile; |
| 4225 | } |
| 4226 | |
| 4227 | cfile = fdget(cfd); |
| 4228 | if (!cfile.file) { |
| 4229 | ret = -EBADF; |
| 4230 | goto out_put_eventfd; |
| 4231 | } |
| 4232 | |
| 4233 | /* the process need read permission on control file */ |
| 4234 | /* AV: shouldn't we check that it's been opened for read instead? */ |
| 4235 | ret = inode_permission(file_inode(cfile.file), MAY_READ); |
| 4236 | if (ret < 0) |
| 4237 | goto out_put_cfile; |
| 4238 | |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4239 | /* |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4240 | * Determine the event callbacks and set them in @event. This used |
| 4241 | * to be done via struct cftype but cgroup core no longer knows |
| 4242 | * about these events. The following is crude but the whole thing |
| 4243 | * is for compatibility anyway. |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4244 | * |
| 4245 | * DO NOT ADD NEW FILES. |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4246 | */ |
Al Viro | b583043 | 2014-10-31 01:22:04 -0400 | [diff] [blame] | 4247 | name = cfile.file->f_path.dentry->d_name.name; |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4248 | |
| 4249 | if (!strcmp(name, "memory.usage_in_bytes")) { |
| 4250 | event->register_event = mem_cgroup_usage_register_event; |
| 4251 | event->unregister_event = mem_cgroup_usage_unregister_event; |
| 4252 | } else if (!strcmp(name, "memory.oom_control")) { |
| 4253 | event->register_event = mem_cgroup_oom_register_event; |
| 4254 | event->unregister_event = mem_cgroup_oom_unregister_event; |
| 4255 | } else if (!strcmp(name, "memory.pressure_level")) { |
| 4256 | event->register_event = vmpressure_register_event; |
| 4257 | event->unregister_event = vmpressure_unregister_event; |
| 4258 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4259 | event->register_event = memsw_cgroup_usage_register_event; |
| 4260 | event->unregister_event = memsw_cgroup_usage_unregister_event; |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4261 | } else { |
| 4262 | ret = -EINVAL; |
| 4263 | goto out_put_cfile; |
| 4264 | } |
| 4265 | |
| 4266 | /* |
Tejun Heo | b5557c4 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4267 | * Verify @cfile should belong to @css. Also, remaining events are |
| 4268 | * automatically removed on cgroup destruction but the removal is |
| 4269 | * asynchronous, so take an extra ref on @css. |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4270 | */ |
Al Viro | b583043 | 2014-10-31 01:22:04 -0400 | [diff] [blame] | 4271 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
Tejun Heo | ec903c0 | 2014-05-13 12:11:01 -0400 | [diff] [blame] | 4272 | &memory_cgrp_subsys); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4273 | ret = -EINVAL; |
Tejun Heo | 5a17f54 | 2014-02-11 11:52:47 -0500 | [diff] [blame] | 4274 | if (IS_ERR(cfile_css)) |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4275 | goto out_put_cfile; |
Tejun Heo | 5a17f54 | 2014-02-11 11:52:47 -0500 | [diff] [blame] | 4276 | if (cfile_css != css) { |
| 4277 | css_put(cfile_css); |
| 4278 | goto out_put_cfile; |
| 4279 | } |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4280 | |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4281 | ret = event->register_event(memcg, event->eventfd, buf); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4282 | if (ret) |
| 4283 | goto out_put_css; |
| 4284 | |
| 4285 | efile.file->f_op->poll(efile.file, &event->pt); |
| 4286 | |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4287 | spin_lock(&memcg->event_list_lock); |
| 4288 | list_add(&event->list, &memcg->event_list); |
| 4289 | spin_unlock(&memcg->event_list_lock); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4290 | |
| 4291 | fdput(cfile); |
| 4292 | fdput(efile); |
| 4293 | |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4294 | return nbytes; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4295 | |
| 4296 | out_put_css: |
Tejun Heo | b5557c4 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4297 | css_put(css); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4298 | out_put_cfile: |
| 4299 | fdput(cfile); |
| 4300 | out_put_eventfd: |
| 4301 | eventfd_ctx_put(event->eventfd); |
| 4302 | out_put_efile: |
| 4303 | fdput(efile); |
| 4304 | out_kfree: |
| 4305 | kfree(event); |
| 4306 | |
| 4307 | return ret; |
| 4308 | } |
| 4309 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 4310 | static struct cftype mem_cgroup_legacy_files[] = { |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4311 | { |
Balbir Singh | 0eea103 | 2008-02-07 00:13:57 -0800 | [diff] [blame] | 4312 | .name = "usage_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4313 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4314 | .read_u64 = mem_cgroup_read_u64, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4315 | }, |
| 4316 | { |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 4317 | .name = "max_usage_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4318 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4319 | .write = mem_cgroup_reset, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4320 | .read_u64 = mem_cgroup_read_u64, |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 4321 | }, |
| 4322 | { |
Balbir Singh | 0eea103 | 2008-02-07 00:13:57 -0800 | [diff] [blame] | 4323 | .name = "limit_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4324 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4325 | .write = mem_cgroup_write, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4326 | .read_u64 = mem_cgroup_read_u64, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4327 | }, |
| 4328 | { |
Balbir Singh | 296c81d | 2009-09-23 15:56:36 -0700 | [diff] [blame] | 4329 | .name = "soft_limit_in_bytes", |
| 4330 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4331 | .write = mem_cgroup_write, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4332 | .read_u64 = mem_cgroup_read_u64, |
Balbir Singh | 296c81d | 2009-09-23 15:56:36 -0700 | [diff] [blame] | 4333 | }, |
| 4334 | { |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4335 | .name = "failcnt", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4336 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4337 | .write = mem_cgroup_reset, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4338 | .read_u64 = mem_cgroup_read_u64, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4339 | }, |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 4340 | { |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 4341 | .name = "stat", |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4342 | .seq_show = memcg_stat_show, |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 4343 | }, |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 4344 | { |
| 4345 | .name = "force_empty", |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4346 | .write = mem_cgroup_force_empty_write, |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 4347 | }, |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4348 | { |
| 4349 | .name = "use_hierarchy", |
| 4350 | .write_u64 = mem_cgroup_hierarchy_write, |
| 4351 | .read_u64 = mem_cgroup_hierarchy_read, |
| 4352 | }, |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 4353 | { |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4354 | .name = "cgroup.event_control", /* XXX: for compat */ |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4355 | .write = memcg_write_event_control, |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4356 | .flags = CFTYPE_NO_PREFIX, |
| 4357 | .mode = S_IWUGO, |
| 4358 | }, |
| 4359 | { |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 4360 | .name = "swappiness", |
| 4361 | .read_u64 = mem_cgroup_swappiness_read, |
| 4362 | .write_u64 = mem_cgroup_swappiness_write, |
| 4363 | }, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4364 | { |
| 4365 | .name = "move_charge_at_immigrate", |
| 4366 | .read_u64 = mem_cgroup_move_charge_read, |
| 4367 | .write_u64 = mem_cgroup_move_charge_write, |
| 4368 | }, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4369 | { |
| 4370 | .name = "oom_control", |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4371 | .seq_show = mem_cgroup_oom_control_read, |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4372 | .write_u64 = mem_cgroup_oom_control_write, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4373 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
| 4374 | }, |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 4375 | { |
| 4376 | .name = "pressure_level", |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 4377 | }, |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 4378 | #ifdef CONFIG_NUMA |
| 4379 | { |
| 4380 | .name = "numa_stat", |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4381 | .seq_show = memcg_numa_stat_show, |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 4382 | }, |
| 4383 | #endif |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4384 | #ifdef CONFIG_MEMCG_KMEM |
| 4385 | { |
| 4386 | .name = "kmem.limit_in_bytes", |
| 4387 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4388 | .write = mem_cgroup_write, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4389 | .read_u64 = mem_cgroup_read_u64, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4390 | }, |
| 4391 | { |
| 4392 | .name = "kmem.usage_in_bytes", |
| 4393 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4394 | .read_u64 = mem_cgroup_read_u64, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4395 | }, |
| 4396 | { |
| 4397 | .name = "kmem.failcnt", |
| 4398 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4399 | .write = mem_cgroup_reset, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4400 | .read_u64 = mem_cgroup_read_u64, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4401 | }, |
| 4402 | { |
| 4403 | .name = "kmem.max_usage_in_bytes", |
| 4404 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4405 | .write = mem_cgroup_reset, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4406 | .read_u64 = mem_cgroup_read_u64, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4407 | }, |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 4408 | #ifdef CONFIG_SLABINFO |
| 4409 | { |
| 4410 | .name = "kmem.slabinfo", |
Vladimir Davydov | b047501 | 2014-12-10 15:44:19 -0800 | [diff] [blame] | 4411 | .seq_start = slab_start, |
| 4412 | .seq_next = slab_next, |
| 4413 | .seq_stop = slab_stop, |
| 4414 | .seq_show = memcg_slab_show, |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 4415 | }, |
| 4416 | #endif |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4417 | #endif |
Tejun Heo | 6bc1034 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4418 | { }, /* terminate */ |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4419 | }; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4420 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4421 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4422 | { |
| 4423 | struct mem_cgroup_per_node *pn; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4424 | struct mem_cgroup_per_zone *mz; |
KAMEZAWA Hiroyuki | 41e3355 | 2008-04-08 17:41:54 -0700 | [diff] [blame] | 4425 | int zone, tmp = node; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4426 | /* |
| 4427 | * This routine is called against possible nodes. |
| 4428 | * But it's BUG to call kmalloc() against offline node. |
| 4429 | * |
| 4430 | * TODO: this routine can waste much memory for nodes which will |
| 4431 | * never be onlined. It's better to use memory hotplug callback |
| 4432 | * function. |
| 4433 | */ |
KAMEZAWA Hiroyuki | 41e3355 | 2008-04-08 17:41:54 -0700 | [diff] [blame] | 4434 | if (!node_state(node, N_NORMAL_MEMORY)) |
| 4435 | tmp = -1; |
Jesper Juhl | 17295c8 | 2011-01-13 15:47:42 -0800 | [diff] [blame] | 4436 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4437 | if (!pn) |
| 4438 | return 1; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4439 | |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4440 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
| 4441 | mz = &pn->zoneinfo[zone]; |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 4442 | lruvec_init(&mz->lruvec); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 4443 | mz->usage_in_excess = 0; |
| 4444 | mz->on_tree = false; |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4445 | mz->memcg = memcg; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4446 | } |
Johannes Weiner | 54f72fe | 2013-07-08 15:59:49 -0700 | [diff] [blame] | 4447 | memcg->nodeinfo[node] = pn; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4448 | return 0; |
| 4449 | } |
| 4450 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4451 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4452 | { |
Johannes Weiner | 54f72fe | 2013-07-08 15:59:49 -0700 | [diff] [blame] | 4453 | kfree(memcg->nodeinfo[node]); |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4454 | } |
| 4455 | |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4456 | static struct mem_cgroup *mem_cgroup_alloc(void) |
| 4457 | { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4458 | struct mem_cgroup *memcg; |
Vladimir Davydov | 8ff69e2 | 2014-01-23 15:52:52 -0800 | [diff] [blame] | 4459 | size_t size; |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4460 | |
Vladimir Davydov | 8ff69e2 | 2014-01-23 15:52:52 -0800 | [diff] [blame] | 4461 | size = sizeof(struct mem_cgroup); |
| 4462 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4463 | |
Vladimir Davydov | 8ff69e2 | 2014-01-23 15:52:52 -0800 | [diff] [blame] | 4464 | memcg = kzalloc(size, GFP_KERNEL); |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4465 | if (!memcg) |
Dan Carpenter | e7bbcdf | 2010-03-23 13:35:12 -0700 | [diff] [blame] | 4466 | return NULL; |
| 4467 | |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4468 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
| 4469 | if (!memcg->stat) |
Dan Carpenter | d2e61b8 | 2010-11-11 14:05:12 -0800 | [diff] [blame] | 4470 | goto out_free; |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4471 | spin_lock_init(&memcg->pcp_counter_lock); |
| 4472 | return memcg; |
Dan Carpenter | d2e61b8 | 2010-11-11 14:05:12 -0800 | [diff] [blame] | 4473 | |
| 4474 | out_free: |
Vladimir Davydov | 8ff69e2 | 2014-01-23 15:52:52 -0800 | [diff] [blame] | 4475 | kfree(memcg); |
Dan Carpenter | d2e61b8 | 2010-11-11 14:05:12 -0800 | [diff] [blame] | 4476 | return NULL; |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4477 | } |
| 4478 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4479 | /* |
Glauber Costa | c8b2a36 | 2012-12-18 14:22:13 -0800 | [diff] [blame] | 4480 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
| 4481 | * (scanning all at force_empty is too costly...) |
| 4482 | * |
| 4483 | * Instead of clearing all references at force_empty, we remember |
| 4484 | * the number of reference from swap_cgroup and free mem_cgroup when |
| 4485 | * it goes down to 0. |
| 4486 | * |
| 4487 | * Removal of cgroup itself succeeds regardless of refs from swap. |
Hugh Dickins | 59927fb | 2012-03-15 15:17:07 -0700 | [diff] [blame] | 4488 | */ |
Glauber Costa | c8b2a36 | 2012-12-18 14:22:13 -0800 | [diff] [blame] | 4489 | |
| 4490 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
Hugh Dickins | 59927fb | 2012-03-15 15:17:07 -0700 | [diff] [blame] | 4491 | { |
Glauber Costa | c8b2a36 | 2012-12-18 14:22:13 -0800 | [diff] [blame] | 4492 | int node; |
Hugh Dickins | 59927fb | 2012-03-15 15:17:07 -0700 | [diff] [blame] | 4493 | |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 4494 | mem_cgroup_remove_from_trees(memcg); |
Glauber Costa | c8b2a36 | 2012-12-18 14:22:13 -0800 | [diff] [blame] | 4495 | |
| 4496 | for_each_node(node) |
| 4497 | free_mem_cgroup_per_zone_info(memcg, node); |
| 4498 | |
| 4499 | free_percpu(memcg->stat); |
Vladimir Davydov | 8ff69e2 | 2014-01-23 15:52:52 -0800 | [diff] [blame] | 4500 | kfree(memcg); |
Hugh Dickins | 59927fb | 2012-03-15 15:17:07 -0700 | [diff] [blame] | 4501 | } |
Glauber Costa | 3afe36b | 2012-05-29 15:07:10 -0700 | [diff] [blame] | 4502 | |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 4503 | /* |
| 4504 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. |
| 4505 | */ |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 4506 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 4507 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4508 | if (!memcg->memory.parent) |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 4509 | return NULL; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4510 | return mem_cgroup_from_counter(memcg->memory.parent, memory); |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 4511 | } |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 4512 | EXPORT_SYMBOL(parent_mem_cgroup); |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4513 | |
Li Zefan | 0eb253e | 2009-01-15 13:51:25 -0800 | [diff] [blame] | 4514 | static struct cgroup_subsys_state * __ref |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4515 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4516 | { |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4517 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 4518 | long error = -ENOMEM; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4519 | int node; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4520 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4521 | memcg = mem_cgroup_alloc(); |
| 4522 | if (!memcg) |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 4523 | return ERR_PTR(error); |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 4524 | |
Bob Liu | 3ed28fa | 2012-01-12 17:19:04 -0800 | [diff] [blame] | 4525 | for_each_node(node) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4526 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4527 | goto free_out; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 4528 | |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 4529 | /* root ? */ |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4530 | if (parent_css == NULL) { |
Hillf Danton | a41c58a | 2011-12-19 17:11:57 -0800 | [diff] [blame] | 4531 | root_mem_cgroup = memcg; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4532 | page_counter_init(&memcg->memory, NULL); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 4533 | memcg->high = PAGE_COUNTER_MAX; |
Johannes Weiner | 24d404d | 2015-01-08 14:32:35 -0800 | [diff] [blame] | 4534 | memcg->soft_limit = PAGE_COUNTER_MAX; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4535 | page_counter_init(&memcg->memsw, NULL); |
| 4536 | page_counter_init(&memcg->kmem, NULL); |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4537 | } |
Balbir Singh | 28dbc4b | 2009-01-07 18:08:05 -0800 | [diff] [blame] | 4538 | |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4539 | memcg->last_scanned_node = MAX_NUMNODES; |
| 4540 | INIT_LIST_HEAD(&memcg->oom_notify); |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4541 | memcg->move_charge_at_immigrate = 0; |
| 4542 | mutex_init(&memcg->thresholds_lock); |
| 4543 | spin_lock_init(&memcg->move_lock); |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 4544 | vmpressure_init(&memcg->vmpressure); |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4545 | INIT_LIST_HEAD(&memcg->event_list); |
| 4546 | spin_lock_init(&memcg->event_list_lock); |
Vladimir Davydov | 900a38f | 2014-12-12 16:55:10 -0800 | [diff] [blame] | 4547 | #ifdef CONFIG_MEMCG_KMEM |
| 4548 | memcg->kmemcg_id = -1; |
Vladimir Davydov | 900a38f | 2014-12-12 16:55:10 -0800 | [diff] [blame] | 4549 | #endif |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4550 | |
| 4551 | return &memcg->css; |
| 4552 | |
| 4553 | free_out: |
| 4554 | __mem_cgroup_free(memcg); |
| 4555 | return ERR_PTR(error); |
| 4556 | } |
| 4557 | |
| 4558 | static int |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4559 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4560 | { |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4561 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Tejun Heo | 5c9d535 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 4562 | struct mem_cgroup *parent = mem_cgroup_from_css(css->parent); |
Johannes Weiner | 2f7dd7a | 2014-10-02 16:16:57 -0700 | [diff] [blame] | 4563 | int ret; |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4564 | |
Tejun Heo | 15a4c83 | 2014-05-04 15:09:14 -0400 | [diff] [blame] | 4565 | if (css->id > MEM_CGROUP_ID_MAX) |
Li Zefan | 4219b2d | 2013-09-23 16:56:29 +0800 | [diff] [blame] | 4566 | return -ENOSPC; |
| 4567 | |
Tejun Heo | 6387698 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4568 | if (!parent) |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4569 | return 0; |
| 4570 | |
Glauber Costa | 0999821 | 2013-02-22 16:34:55 -0800 | [diff] [blame] | 4571 | mutex_lock(&memcg_create_mutex); |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4572 | |
| 4573 | memcg->use_hierarchy = parent->use_hierarchy; |
| 4574 | memcg->oom_kill_disable = parent->oom_kill_disable; |
| 4575 | memcg->swappiness = mem_cgroup_swappiness(parent); |
| 4576 | |
| 4577 | if (parent->use_hierarchy) { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4578 | page_counter_init(&memcg->memory, &parent->memory); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 4579 | memcg->high = PAGE_COUNTER_MAX; |
Johannes Weiner | 24d404d | 2015-01-08 14:32:35 -0800 | [diff] [blame] | 4580 | memcg->soft_limit = PAGE_COUNTER_MAX; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4581 | page_counter_init(&memcg->memsw, &parent->memsw); |
| 4582 | page_counter_init(&memcg->kmem, &parent->kmem); |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 4583 | |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 4584 | /* |
Li Zefan | 8d76a97 | 2013-07-08 16:00:36 -0700 | [diff] [blame] | 4585 | * No need to take a reference to the parent because cgroup |
| 4586 | * core guarantees its existence. |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 4587 | */ |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4588 | } else { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4589 | page_counter_init(&memcg->memory, NULL); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 4590 | memcg->high = PAGE_COUNTER_MAX; |
Johannes Weiner | 24d404d | 2015-01-08 14:32:35 -0800 | [diff] [blame] | 4591 | memcg->soft_limit = PAGE_COUNTER_MAX; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4592 | page_counter_init(&memcg->memsw, NULL); |
| 4593 | page_counter_init(&memcg->kmem, NULL); |
Tejun Heo | 8c7f6ed | 2012-09-13 12:20:58 -0700 | [diff] [blame] | 4594 | /* |
| 4595 | * Deeper hierachy with use_hierarchy == false doesn't make |
| 4596 | * much sense so let cgroup subsystem know about this |
| 4597 | * unfortunate state in our controller. |
| 4598 | */ |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4599 | if (parent != root_mem_cgroup) |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 4600 | memory_cgrp_subsys.broken_hierarchy = true; |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4601 | } |
Glauber Costa | 0999821 | 2013-02-22 16:34:55 -0800 | [diff] [blame] | 4602 | mutex_unlock(&memcg_create_mutex); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 4603 | |
Johannes Weiner | 2f7dd7a | 2014-10-02 16:16:57 -0700 | [diff] [blame] | 4604 | ret = memcg_init_kmem(memcg, &memory_cgrp_subsys); |
| 4605 | if (ret) |
| 4606 | return ret; |
| 4607 | |
| 4608 | /* |
| 4609 | * Make sure the memcg is initialized: mem_cgroup_iter() |
| 4610 | * orders reading memcg->initialized against its callers |
| 4611 | * reading the memcg members. |
| 4612 | */ |
| 4613 | smp_store_release(&memcg->initialized, 1); |
| 4614 | |
| 4615 | return 0; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4616 | } |
| 4617 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4618 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
KAMEZAWA Hiroyuki | df878fb | 2008-02-07 00:14:28 -0800 | [diff] [blame] | 4619 | { |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4620 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4621 | struct mem_cgroup_event *event, *tmp; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4622 | |
| 4623 | /* |
| 4624 | * Unregister events and notify userspace. |
| 4625 | * Notify userspace about cgroup removing only after rmdir of cgroup |
| 4626 | * directory to avoid race between userspace and kernelspace. |
| 4627 | */ |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4628 | spin_lock(&memcg->event_list_lock); |
| 4629 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4630 | list_del_init(&event->list); |
| 4631 | schedule_work(&event->remove); |
| 4632 | } |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4633 | spin_unlock(&memcg->event_list_lock); |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 4634 | |
Michal Hocko | 33cb876 | 2013-07-31 13:53:51 -0700 | [diff] [blame] | 4635 | vmpressure_cleanup(&memcg->vmpressure); |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 4636 | |
| 4637 | memcg_deactivate_kmem(memcg); |
KAMEZAWA Hiroyuki | df878fb | 2008-02-07 00:14:28 -0800 | [diff] [blame] | 4638 | } |
| 4639 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4640 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4641 | { |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4642 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Daisuke Nishimura | c268e99 | 2009-01-15 13:51:13 -0800 | [diff] [blame] | 4643 | |
Li Zefan | 10d5ebf | 2013-07-08 16:00:33 -0700 | [diff] [blame] | 4644 | memcg_destroy_kmem(memcg); |
Li Zefan | 465939a | 2013-07-08 16:00:38 -0700 | [diff] [blame] | 4645 | __mem_cgroup_free(memcg); |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4646 | } |
| 4647 | |
Tejun Heo | 1ced953 | 2014-07-08 18:02:57 -0400 | [diff] [blame] | 4648 | /** |
| 4649 | * mem_cgroup_css_reset - reset the states of a mem_cgroup |
| 4650 | * @css: the target css |
| 4651 | * |
| 4652 | * Reset the states of the mem_cgroup associated with @css. This is |
| 4653 | * invoked when the userland requests disabling on the default hierarchy |
| 4654 | * but the memcg is pinned through dependency. The memcg should stop |
| 4655 | * applying policies and should revert to the vanilla state as it may be |
| 4656 | * made visible again. |
| 4657 | * |
| 4658 | * The current implementation only resets the essential configurations. |
| 4659 | * This needs to be expanded to cover all the visible parts. |
| 4660 | */ |
| 4661 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) |
| 4662 | { |
| 4663 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 4664 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4665 | mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX); |
| 4666 | mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX); |
| 4667 | memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 4668 | memcg->low = 0; |
| 4669 | memcg->high = PAGE_COUNTER_MAX; |
Johannes Weiner | 24d404d | 2015-01-08 14:32:35 -0800 | [diff] [blame] | 4670 | memcg->soft_limit = PAGE_COUNTER_MAX; |
Tejun Heo | 1ced953 | 2014-07-08 18:02:57 -0400 | [diff] [blame] | 4671 | } |
| 4672 | |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4673 | #ifdef CONFIG_MMU |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4674 | /* Handlers for move charge at task migration. */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4675 | static int mem_cgroup_do_precharge(unsigned long count) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4676 | { |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 4677 | int ret; |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 4678 | |
| 4679 | /* Try a single bulk charge without reclaim first */ |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 4680 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_WAIT, count); |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 4681 | if (!ret) { |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4682 | mc.precharge += count; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4683 | return ret; |
| 4684 | } |
Johannes Weiner | 692e7c4 | 2014-08-06 16:05:57 -0700 | [diff] [blame] | 4685 | if (ret == -EINTR) { |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 4686 | cancel_charge(root_mem_cgroup, count); |
Johannes Weiner | 692e7c4 | 2014-08-06 16:05:57 -0700 | [diff] [blame] | 4687 | return ret; |
| 4688 | } |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 4689 | |
| 4690 | /* Try charges one by one with reclaim */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4691 | while (count--) { |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 4692 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1); |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 4693 | /* |
| 4694 | * In case of failure, any residual charges against |
| 4695 | * mc.to will be dropped by mem_cgroup_clear_mc() |
Johannes Weiner | 692e7c4 | 2014-08-06 16:05:57 -0700 | [diff] [blame] | 4696 | * later on. However, cancel any charges that are |
| 4697 | * bypassed to root right away or they'll be lost. |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 4698 | */ |
Johannes Weiner | 692e7c4 | 2014-08-06 16:05:57 -0700 | [diff] [blame] | 4699 | if (ret == -EINTR) |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 4700 | cancel_charge(root_mem_cgroup, 1); |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 4701 | if (ret) |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 4702 | return ret; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4703 | mc.precharge++; |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 4704 | cond_resched(); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4705 | } |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 4706 | return 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4707 | } |
| 4708 | |
| 4709 | /** |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 4710 | * get_mctgt_type - get target type of moving charge |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4711 | * @vma: the vma the pte to be checked belongs |
| 4712 | * @addr: the address corresponding to the pte to be checked |
| 4713 | * @ptent: the pte to be checked |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4714 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4715 | * |
| 4716 | * Returns |
| 4717 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. |
| 4718 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for |
| 4719 | * move charge. if @target is not NULL, the page is stored in target->page |
| 4720 | * with extra refcnt got(Callers should handle it). |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4721 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
| 4722 | * target for charge migration. if @target is not NULL, the entry is stored |
| 4723 | * in target->ent. |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4724 | * |
| 4725 | * Called with pte lock held. |
| 4726 | */ |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4727 | union mc_target { |
| 4728 | struct page *page; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4729 | swp_entry_t ent; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4730 | }; |
| 4731 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4732 | enum mc_target_type { |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 4733 | MC_TARGET_NONE = 0, |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4734 | MC_TARGET_PAGE, |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4735 | MC_TARGET_SWAP, |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4736 | }; |
| 4737 | |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4738 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
| 4739 | unsigned long addr, pte_t ptent) |
| 4740 | { |
| 4741 | struct page *page = vm_normal_page(vma, addr, ptent); |
| 4742 | |
| 4743 | if (!page || !page_mapped(page)) |
| 4744 | return NULL; |
| 4745 | if (PageAnon(page)) { |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 4746 | if (!(mc.flags & MOVE_ANON)) |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4747 | return NULL; |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 4748 | } else { |
| 4749 | if (!(mc.flags & MOVE_FILE)) |
| 4750 | return NULL; |
| 4751 | } |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4752 | if (!get_page_unless_zero(page)) |
| 4753 | return NULL; |
| 4754 | |
| 4755 | return page; |
| 4756 | } |
| 4757 | |
KAMEZAWA Hiroyuki | 4b91355 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 4758 | #ifdef CONFIG_SWAP |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4759 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
| 4760 | unsigned long addr, pte_t ptent, swp_entry_t *entry) |
| 4761 | { |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4762 | struct page *page = NULL; |
| 4763 | swp_entry_t ent = pte_to_swp_entry(ptent); |
| 4764 | |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 4765 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4766 | return NULL; |
KAMEZAWA Hiroyuki | 4b91355 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 4767 | /* |
| 4768 | * Because lookup_swap_cache() updates some statistics counter, |
| 4769 | * we call find_get_page() with swapper_space directly. |
| 4770 | */ |
Shaohua Li | 33806f0 | 2013-02-22 16:34:37 -0800 | [diff] [blame] | 4771 | page = find_get_page(swap_address_space(ent), ent.val); |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4772 | if (do_swap_account) |
| 4773 | entry->val = ent.val; |
| 4774 | |
| 4775 | return page; |
| 4776 | } |
KAMEZAWA Hiroyuki | 4b91355 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 4777 | #else |
| 4778 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
| 4779 | unsigned long addr, pte_t ptent, swp_entry_t *entry) |
| 4780 | { |
| 4781 | return NULL; |
| 4782 | } |
| 4783 | #endif |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4784 | |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4785 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
| 4786 | unsigned long addr, pte_t ptent, swp_entry_t *entry) |
| 4787 | { |
| 4788 | struct page *page = NULL; |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4789 | struct address_space *mapping; |
| 4790 | pgoff_t pgoff; |
| 4791 | |
| 4792 | if (!vma->vm_file) /* anonymous vma */ |
| 4793 | return NULL; |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 4794 | if (!(mc.flags & MOVE_FILE)) |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4795 | return NULL; |
| 4796 | |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4797 | mapping = vma->vm_file->f_mapping; |
Kirill A. Shutemov | 0661a33 | 2015-02-10 14:10:04 -0800 | [diff] [blame] | 4798 | pgoff = linear_page_index(vma, addr); |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4799 | |
| 4800 | /* page is moved even if it's not RSS of this task(page-faulted). */ |
Hugh Dickins | aa3b189 | 2011-08-03 16:21:24 -0700 | [diff] [blame] | 4801 | #ifdef CONFIG_SWAP |
| 4802 | /* shmem/tmpfs may report page out on swap: account for that too. */ |
Johannes Weiner | 139b6a6 | 2014-05-06 12:50:05 -0700 | [diff] [blame] | 4803 | if (shmem_mapping(mapping)) { |
| 4804 | page = find_get_entry(mapping, pgoff); |
| 4805 | if (radix_tree_exceptional_entry(page)) { |
| 4806 | swp_entry_t swp = radix_to_swp_entry(page); |
| 4807 | if (do_swap_account) |
| 4808 | *entry = swp; |
| 4809 | page = find_get_page(swap_address_space(swp), swp.val); |
| 4810 | } |
| 4811 | } else |
| 4812 | page = find_get_page(mapping, pgoff); |
| 4813 | #else |
| 4814 | page = find_get_page(mapping, pgoff); |
Hugh Dickins | aa3b189 | 2011-08-03 16:21:24 -0700 | [diff] [blame] | 4815 | #endif |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4816 | return page; |
| 4817 | } |
| 4818 | |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 4819 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4820 | unsigned long addr, pte_t ptent, union mc_target *target) |
| 4821 | { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4822 | struct page *page = NULL; |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 4823 | enum mc_target_type ret = MC_TARGET_NONE; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4824 | swp_entry_t ent = { .val = 0 }; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4825 | |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4826 | if (pte_present(ptent)) |
| 4827 | page = mc_handle_present_pte(vma, addr, ptent); |
| 4828 | else if (is_swap_pte(ptent)) |
| 4829 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); |
Kirill A. Shutemov | 0661a33 | 2015-02-10 14:10:04 -0800 | [diff] [blame] | 4830 | else if (pte_none(ptent)) |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4831 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4832 | |
| 4833 | if (!page && !ent.val) |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 4834 | return ret; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4835 | if (page) { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4836 | /* |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 4837 | * Do only loose check w/o serialization. |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 4838 | * mem_cgroup_move_account() checks the page is valid or |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 4839 | * not under LRU exclusion. |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4840 | */ |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 4841 | if (page->mem_cgroup == mc.from) { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4842 | ret = MC_TARGET_PAGE; |
| 4843 | if (target) |
| 4844 | target->page = page; |
| 4845 | } |
| 4846 | if (!ret || !target) |
| 4847 | put_page(page); |
| 4848 | } |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4849 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
| 4850 | if (ent.val && !ret && |
Li Zefan | 34c00c3 | 2013-09-23 16:56:01 +0800 | [diff] [blame] | 4851 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
KAMEZAWA Hiroyuki | 7f0f154 | 2010-05-11 14:06:58 -0700 | [diff] [blame] | 4852 | ret = MC_TARGET_SWAP; |
| 4853 | if (target) |
| 4854 | target->ent = ent; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4855 | } |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4856 | return ret; |
| 4857 | } |
| 4858 | |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4859 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 4860 | /* |
| 4861 | * We don't consider swapping or file mapped pages because THP does not |
| 4862 | * support them for now. |
| 4863 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
| 4864 | */ |
| 4865 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, |
| 4866 | unsigned long addr, pmd_t pmd, union mc_target *target) |
| 4867 | { |
| 4868 | struct page *page = NULL; |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4869 | enum mc_target_type ret = MC_TARGET_NONE; |
| 4870 | |
| 4871 | page = pmd_page(pmd); |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 4872 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 4873 | if (!(mc.flags & MOVE_ANON)) |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4874 | return ret; |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 4875 | if (page->mem_cgroup == mc.from) { |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4876 | ret = MC_TARGET_PAGE; |
| 4877 | if (target) { |
| 4878 | get_page(page); |
| 4879 | target->page = page; |
| 4880 | } |
| 4881 | } |
| 4882 | return ret; |
| 4883 | } |
| 4884 | #else |
| 4885 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, |
| 4886 | unsigned long addr, pmd_t pmd, union mc_target *target) |
| 4887 | { |
| 4888 | return MC_TARGET_NONE; |
| 4889 | } |
| 4890 | #endif |
| 4891 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4892 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
| 4893 | unsigned long addr, unsigned long end, |
| 4894 | struct mm_walk *walk) |
| 4895 | { |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 4896 | struct vm_area_struct *vma = walk->vma; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4897 | pte_t *pte; |
| 4898 | spinlock_t *ptl; |
| 4899 | |
Kirill A. Shutemov | bf92915 | 2013-11-14 14:30:54 -0800 | [diff] [blame] | 4900 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4901 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
| 4902 | mc.precharge += HPAGE_PMD_NR; |
Kirill A. Shutemov | bf92915 | 2013-11-14 14:30:54 -0800 | [diff] [blame] | 4903 | spin_unlock(ptl); |
Andrea Arcangeli | 1a5a990 | 2012-03-21 16:33:42 -0700 | [diff] [blame] | 4904 | return 0; |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4905 | } |
Dave Hansen | 0331932 | 2011-03-22 16:32:56 -0700 | [diff] [blame] | 4906 | |
Andrea Arcangeli | 45f83ce | 2012-03-28 14:42:40 -0700 | [diff] [blame] | 4907 | if (pmd_trans_unstable(pmd)) |
| 4908 | return 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4909 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
| 4910 | for (; addr != end; pte++, addr += PAGE_SIZE) |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 4911 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4912 | mc.precharge++; /* increment precharge temporarily */ |
| 4913 | pte_unmap_unlock(pte - 1, ptl); |
| 4914 | cond_resched(); |
| 4915 | |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4916 | return 0; |
| 4917 | } |
| 4918 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4919 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
| 4920 | { |
| 4921 | unsigned long precharge; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4922 | |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 4923 | struct mm_walk mem_cgroup_count_precharge_walk = { |
| 4924 | .pmd_entry = mem_cgroup_count_precharge_pte_range, |
| 4925 | .mm = mm, |
| 4926 | }; |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4927 | down_read(&mm->mmap_sem); |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 4928 | walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4929 | up_read(&mm->mmap_sem); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4930 | |
| 4931 | precharge = mc.precharge; |
| 4932 | mc.precharge = 0; |
| 4933 | |
| 4934 | return precharge; |
| 4935 | } |
| 4936 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4937 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
| 4938 | { |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4939 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
| 4940 | |
| 4941 | VM_BUG_ON(mc.moving_task); |
| 4942 | mc.moving_task = current; |
| 4943 | return mem_cgroup_do_precharge(precharge); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4944 | } |
| 4945 | |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4946 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
| 4947 | static void __mem_cgroup_clear_mc(void) |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4948 | { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 4949 | struct mem_cgroup *from = mc.from; |
| 4950 | struct mem_cgroup *to = mc.to; |
| 4951 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4952 | /* we must uncharge all the leftover precharges from mc.to */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4953 | if (mc.precharge) { |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 4954 | cancel_charge(mc.to, mc.precharge); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4955 | mc.precharge = 0; |
| 4956 | } |
| 4957 | /* |
| 4958 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so |
| 4959 | * we must uncharge here. |
| 4960 | */ |
| 4961 | if (mc.moved_charge) { |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 4962 | cancel_charge(mc.from, mc.moved_charge); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4963 | mc.moved_charge = 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4964 | } |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4965 | /* we must fixup refcnts and charges */ |
| 4966 | if (mc.moved_swap) { |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4967 | /* uncharge swap account from the old cgroup */ |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 4968 | if (!mem_cgroup_is_root(mc.from)) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4969 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4970 | |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 4971 | /* |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4972 | * we charged both to->memory and to->memsw, so we |
| 4973 | * should uncharge to->memory. |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 4974 | */ |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 4975 | if (!mem_cgroup_is_root(mc.to)) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4976 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4977 | |
Johannes Weiner | e8ea14c | 2014-12-10 15:42:42 -0800 | [diff] [blame] | 4978 | css_put_many(&mc.from->css, mc.moved_swap); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4979 | |
Li Zefan | 4050377 | 2013-07-08 16:00:34 -0700 | [diff] [blame] | 4980 | /* we've already done css_get(mc.to) */ |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4981 | mc.moved_swap = 0; |
| 4982 | } |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4983 | memcg_oom_recover(from); |
| 4984 | memcg_oom_recover(to); |
| 4985 | wake_up_all(&mc.waitq); |
| 4986 | } |
| 4987 | |
| 4988 | static void mem_cgroup_clear_mc(void) |
| 4989 | { |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4990 | /* |
| 4991 | * we must clear moving_task before waking up waiters at the end of |
| 4992 | * task migration. |
| 4993 | */ |
| 4994 | mc.moving_task = NULL; |
| 4995 | __mem_cgroup_clear_mc(); |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 4996 | spin_lock(&mc.lock); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4997 | mc.from = NULL; |
| 4998 | mc.to = NULL; |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 4999 | spin_unlock(&mc.lock); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5000 | } |
| 5001 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5002 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
Li Zefan | 761b3ef | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 5003 | struct cgroup_taskset *tset) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5004 | { |
Tejun Heo | 2f7ee56 | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 5005 | struct task_struct *p = cgroup_taskset_first(tset); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5006 | int ret = 0; |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5007 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 5008 | unsigned long move_flags; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5009 | |
Glauber Costa | ee5e847 | 2013-02-22 16:34:50 -0800 | [diff] [blame] | 5010 | /* |
| 5011 | * We are now commited to this value whatever it is. Changes in this |
| 5012 | * tunable will only affect upcoming migrations, not the current one. |
| 5013 | * So we need to save it, and keep it going. |
| 5014 | */ |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 5015 | move_flags = ACCESS_ONCE(memcg->move_charge_at_immigrate); |
| 5016 | if (move_flags) { |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5017 | struct mm_struct *mm; |
| 5018 | struct mem_cgroup *from = mem_cgroup_from_task(p); |
| 5019 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5020 | VM_BUG_ON(from == memcg); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5021 | |
| 5022 | mm = get_task_mm(p); |
| 5023 | if (!mm) |
| 5024 | return 0; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5025 | /* We move charges only when we move a owner of the mm */ |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5026 | if (mm->owner == p) { |
| 5027 | VM_BUG_ON(mc.from); |
| 5028 | VM_BUG_ON(mc.to); |
| 5029 | VM_BUG_ON(mc.precharge); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5030 | VM_BUG_ON(mc.moved_charge); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 5031 | VM_BUG_ON(mc.moved_swap); |
Johannes Weiner | 247b144 | 2014-12-10 15:44:11 -0800 | [diff] [blame] | 5032 | |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 5033 | spin_lock(&mc.lock); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5034 | mc.from = from; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5035 | mc.to = memcg; |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 5036 | mc.flags = move_flags; |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 5037 | spin_unlock(&mc.lock); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5038 | /* We set mc.moving_task later */ |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5039 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5040 | ret = mem_cgroup_precharge_mc(mm); |
| 5041 | if (ret) |
| 5042 | mem_cgroup_clear_mc(); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5043 | } |
| 5044 | mmput(mm); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5045 | } |
| 5046 | return ret; |
| 5047 | } |
| 5048 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5049 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
Li Zefan | 761b3ef | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 5050 | struct cgroup_taskset *tset) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5051 | { |
Johannes Weiner | 4e2f245 | 2014-12-10 15:44:08 -0800 | [diff] [blame] | 5052 | if (mc.to) |
| 5053 | mem_cgroup_clear_mc(); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5054 | } |
| 5055 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5056 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
| 5057 | unsigned long addr, unsigned long end, |
| 5058 | struct mm_walk *walk) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5059 | { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5060 | int ret = 0; |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 5061 | struct vm_area_struct *vma = walk->vma; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5062 | pte_t *pte; |
| 5063 | spinlock_t *ptl; |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5064 | enum mc_target_type target_type; |
| 5065 | union mc_target target; |
| 5066 | struct page *page; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5067 | |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5068 | /* |
| 5069 | * We don't take compound_lock() here but no race with splitting thp |
| 5070 | * happens because: |
| 5071 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not |
| 5072 | * under splitting, which means there's no concurrent thp split, |
| 5073 | * - if another thread runs into split_huge_page() just after we |
| 5074 | * entered this if-block, the thread must wait for page table lock |
| 5075 | * to be unlocked in __split_huge_page_splitting(), where the main |
| 5076 | * part of thp split is not executed yet. |
| 5077 | */ |
Kirill A. Shutemov | bf92915 | 2013-11-14 14:30:54 -0800 | [diff] [blame] | 5078 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
Hugh Dickins | 62ade86 | 2012-05-18 11:28:34 -0700 | [diff] [blame] | 5079 | if (mc.precharge < HPAGE_PMD_NR) { |
Kirill A. Shutemov | bf92915 | 2013-11-14 14:30:54 -0800 | [diff] [blame] | 5080 | spin_unlock(ptl); |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5081 | return 0; |
| 5082 | } |
| 5083 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); |
| 5084 | if (target_type == MC_TARGET_PAGE) { |
| 5085 | page = target.page; |
| 5086 | if (!isolate_lru_page(page)) { |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5087 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5088 | mc.from, mc.to)) { |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5089 | mc.precharge -= HPAGE_PMD_NR; |
| 5090 | mc.moved_charge += HPAGE_PMD_NR; |
| 5091 | } |
| 5092 | putback_lru_page(page); |
| 5093 | } |
| 5094 | put_page(page); |
| 5095 | } |
Kirill A. Shutemov | bf92915 | 2013-11-14 14:30:54 -0800 | [diff] [blame] | 5096 | spin_unlock(ptl); |
Andrea Arcangeli | 1a5a990 | 2012-03-21 16:33:42 -0700 | [diff] [blame] | 5097 | return 0; |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5098 | } |
| 5099 | |
Andrea Arcangeli | 45f83ce | 2012-03-28 14:42:40 -0700 | [diff] [blame] | 5100 | if (pmd_trans_unstable(pmd)) |
| 5101 | return 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5102 | retry: |
| 5103 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
| 5104 | for (; addr != end; addr += PAGE_SIZE) { |
| 5105 | pte_t ptent = *(pte++); |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5106 | swp_entry_t ent; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5107 | |
| 5108 | if (!mc.precharge) |
| 5109 | break; |
| 5110 | |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 5111 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5112 | case MC_TARGET_PAGE: |
| 5113 | page = target.page; |
| 5114 | if (isolate_lru_page(page)) |
| 5115 | goto put; |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5116 | if (!mem_cgroup_move_account(page, 1, mc.from, mc.to)) { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5117 | mc.precharge--; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5118 | /* we uncharge from mc.from later. */ |
| 5119 | mc.moved_charge++; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5120 | } |
| 5121 | putback_lru_page(page); |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 5122 | put: /* get_mctgt_type() gets the page */ |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5123 | put_page(page); |
| 5124 | break; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5125 | case MC_TARGET_SWAP: |
| 5126 | ent = target.ent; |
Hugh Dickins | e91cbb4 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 5127 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5128 | mc.precharge--; |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 5129 | /* we fixup refcnts and charges later. */ |
| 5130 | mc.moved_swap++; |
| 5131 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5132 | break; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5133 | default: |
| 5134 | break; |
| 5135 | } |
| 5136 | } |
| 5137 | pte_unmap_unlock(pte - 1, ptl); |
| 5138 | cond_resched(); |
| 5139 | |
| 5140 | if (addr != end) { |
| 5141 | /* |
| 5142 | * We have consumed all precharges we got in can_attach(). |
| 5143 | * We try charge one by one, but don't do any additional |
| 5144 | * charges to mc.to if we have failed in charge once in attach() |
| 5145 | * phase. |
| 5146 | */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5147 | ret = mem_cgroup_do_precharge(1); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5148 | if (!ret) |
| 5149 | goto retry; |
| 5150 | } |
| 5151 | |
| 5152 | return ret; |
| 5153 | } |
| 5154 | |
| 5155 | static void mem_cgroup_move_charge(struct mm_struct *mm) |
| 5156 | { |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 5157 | struct mm_walk mem_cgroup_move_charge_walk = { |
| 5158 | .pmd_entry = mem_cgroup_move_charge_pte_range, |
| 5159 | .mm = mm, |
| 5160 | }; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5161 | |
| 5162 | lru_add_drain_all(); |
Johannes Weiner | 312722c | 2014-12-10 15:44:25 -0800 | [diff] [blame] | 5163 | /* |
| 5164 | * Signal mem_cgroup_begin_page_stat() to take the memcg's |
| 5165 | * move_lock while we're moving its pages to another memcg. |
| 5166 | * Then wait for already started RCU-only updates to finish. |
| 5167 | */ |
| 5168 | atomic_inc(&mc.from->moving_account); |
| 5169 | synchronize_rcu(); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5170 | retry: |
| 5171 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { |
| 5172 | /* |
| 5173 | * Someone who are holding the mmap_sem might be waiting in |
| 5174 | * waitq. So we cancel all extra charges, wake up all waiters, |
| 5175 | * and retry. Because we cancel precharges, we might not be able |
| 5176 | * to move enough charges, but moving charge is a best-effort |
| 5177 | * feature anyway, so it wouldn't be a big problem. |
| 5178 | */ |
| 5179 | __mem_cgroup_clear_mc(); |
| 5180 | cond_resched(); |
| 5181 | goto retry; |
| 5182 | } |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 5183 | /* |
| 5184 | * When we have consumed all precharges and failed in doing |
| 5185 | * additional charge, the page walk just aborts. |
| 5186 | */ |
| 5187 | walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5188 | up_read(&mm->mmap_sem); |
Johannes Weiner | 312722c | 2014-12-10 15:44:25 -0800 | [diff] [blame] | 5189 | atomic_dec(&mc.from->moving_account); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5190 | } |
| 5191 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5192 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
Li Zefan | 761b3ef | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 5193 | struct cgroup_taskset *tset) |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 5194 | { |
Tejun Heo | 2f7ee56 | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 5195 | struct task_struct *p = cgroup_taskset_first(tset); |
KOSAKI Motohiro | a433658 | 2011-06-15 15:08:13 -0700 | [diff] [blame] | 5196 | struct mm_struct *mm = get_task_mm(p); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5197 | |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5198 | if (mm) { |
KOSAKI Motohiro | a433658 | 2011-06-15 15:08:13 -0700 | [diff] [blame] | 5199 | if (mc.to) |
| 5200 | mem_cgroup_move_charge(mm); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5201 | mmput(mm); |
| 5202 | } |
KOSAKI Motohiro | a433658 | 2011-06-15 15:08:13 -0700 | [diff] [blame] | 5203 | if (mc.to) |
| 5204 | mem_cgroup_clear_mc(); |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 5205 | } |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 5206 | #else /* !CONFIG_MMU */ |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5207 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
Li Zefan | 761b3ef | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 5208 | struct cgroup_taskset *tset) |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 5209 | { |
| 5210 | return 0; |
| 5211 | } |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5212 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
Li Zefan | 761b3ef | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 5213 | struct cgroup_taskset *tset) |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 5214 | { |
| 5215 | } |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5216 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
Li Zefan | 761b3ef | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 5217 | struct cgroup_taskset *tset) |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 5218 | { |
| 5219 | } |
| 5220 | #endif |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 5221 | |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 5222 | /* |
| 5223 | * Cgroup retains root cgroups across [un]mount cycles making it necessary |
Tejun Heo | aa6ec29 | 2014-07-09 10:08:08 -0400 | [diff] [blame] | 5224 | * to verify whether we're attached to the default hierarchy on each mount |
| 5225 | * attempt. |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 5226 | */ |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5227 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 5228 | { |
| 5229 | /* |
Tejun Heo | aa6ec29 | 2014-07-09 10:08:08 -0400 | [diff] [blame] | 5230 | * use_hierarchy is forced on the default hierarchy. cgroup core |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 5231 | * guarantees that @root doesn't have any children, so turning it |
| 5232 | * on for the root memcg is enough. |
| 5233 | */ |
Tejun Heo | aa6ec29 | 2014-07-09 10:08:08 -0400 | [diff] [blame] | 5234 | if (cgroup_on_dfl(root_css->cgroup)) |
Vladimir Davydov | 7feee59 | 2015-03-12 16:26:19 -0700 | [diff] [blame] | 5235 | root_mem_cgroup->use_hierarchy = true; |
| 5236 | else |
| 5237 | root_mem_cgroup->use_hierarchy = false; |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 5238 | } |
| 5239 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5240 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
| 5241 | struct cftype *cft) |
| 5242 | { |
| 5243 | return mem_cgroup_usage(mem_cgroup_from_css(css), false); |
| 5244 | } |
| 5245 | |
| 5246 | static int memory_low_show(struct seq_file *m, void *v) |
| 5247 | { |
| 5248 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
| 5249 | unsigned long low = ACCESS_ONCE(memcg->low); |
| 5250 | |
| 5251 | if (low == PAGE_COUNTER_MAX) |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 5252 | seq_puts(m, "max\n"); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5253 | else |
| 5254 | seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE); |
| 5255 | |
| 5256 | return 0; |
| 5257 | } |
| 5258 | |
| 5259 | static ssize_t memory_low_write(struct kernfs_open_file *of, |
| 5260 | char *buf, size_t nbytes, loff_t off) |
| 5261 | { |
| 5262 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
| 5263 | unsigned long low; |
| 5264 | int err; |
| 5265 | |
| 5266 | buf = strstrip(buf); |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 5267 | err = page_counter_memparse(buf, "max", &low); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5268 | if (err) |
| 5269 | return err; |
| 5270 | |
| 5271 | memcg->low = low; |
| 5272 | |
| 5273 | return nbytes; |
| 5274 | } |
| 5275 | |
| 5276 | static int memory_high_show(struct seq_file *m, void *v) |
| 5277 | { |
| 5278 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
| 5279 | unsigned long high = ACCESS_ONCE(memcg->high); |
| 5280 | |
| 5281 | if (high == PAGE_COUNTER_MAX) |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 5282 | seq_puts(m, "max\n"); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5283 | else |
| 5284 | seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE); |
| 5285 | |
| 5286 | return 0; |
| 5287 | } |
| 5288 | |
| 5289 | static ssize_t memory_high_write(struct kernfs_open_file *of, |
| 5290 | char *buf, size_t nbytes, loff_t off) |
| 5291 | { |
| 5292 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
| 5293 | unsigned long high; |
| 5294 | int err; |
| 5295 | |
| 5296 | buf = strstrip(buf); |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 5297 | err = page_counter_memparse(buf, "max", &high); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5298 | if (err) |
| 5299 | return err; |
| 5300 | |
| 5301 | memcg->high = high; |
| 5302 | |
| 5303 | return nbytes; |
| 5304 | } |
| 5305 | |
| 5306 | static int memory_max_show(struct seq_file *m, void *v) |
| 5307 | { |
| 5308 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
| 5309 | unsigned long max = ACCESS_ONCE(memcg->memory.limit); |
| 5310 | |
| 5311 | if (max == PAGE_COUNTER_MAX) |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 5312 | seq_puts(m, "max\n"); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5313 | else |
| 5314 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); |
| 5315 | |
| 5316 | return 0; |
| 5317 | } |
| 5318 | |
| 5319 | static ssize_t memory_max_write(struct kernfs_open_file *of, |
| 5320 | char *buf, size_t nbytes, loff_t off) |
| 5321 | { |
| 5322 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
| 5323 | unsigned long max; |
| 5324 | int err; |
| 5325 | |
| 5326 | buf = strstrip(buf); |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 5327 | err = page_counter_memparse(buf, "max", &max); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5328 | if (err) |
| 5329 | return err; |
| 5330 | |
| 5331 | err = mem_cgroup_resize_limit(memcg, max); |
| 5332 | if (err) |
| 5333 | return err; |
| 5334 | |
| 5335 | return nbytes; |
| 5336 | } |
| 5337 | |
| 5338 | static int memory_events_show(struct seq_file *m, void *v) |
| 5339 | { |
| 5340 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
| 5341 | |
| 5342 | seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW)); |
| 5343 | seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH)); |
| 5344 | seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX)); |
| 5345 | seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM)); |
| 5346 | |
| 5347 | return 0; |
| 5348 | } |
| 5349 | |
| 5350 | static struct cftype memory_files[] = { |
| 5351 | { |
| 5352 | .name = "current", |
| 5353 | .read_u64 = memory_current_read, |
| 5354 | }, |
| 5355 | { |
| 5356 | .name = "low", |
| 5357 | .flags = CFTYPE_NOT_ON_ROOT, |
| 5358 | .seq_show = memory_low_show, |
| 5359 | .write = memory_low_write, |
| 5360 | }, |
| 5361 | { |
| 5362 | .name = "high", |
| 5363 | .flags = CFTYPE_NOT_ON_ROOT, |
| 5364 | .seq_show = memory_high_show, |
| 5365 | .write = memory_high_write, |
| 5366 | }, |
| 5367 | { |
| 5368 | .name = "max", |
| 5369 | .flags = CFTYPE_NOT_ON_ROOT, |
| 5370 | .seq_show = memory_max_show, |
| 5371 | .write = memory_max_write, |
| 5372 | }, |
| 5373 | { |
| 5374 | .name = "events", |
| 5375 | .flags = CFTYPE_NOT_ON_ROOT, |
| 5376 | .seq_show = memory_events_show, |
| 5377 | }, |
| 5378 | { } /* terminate */ |
| 5379 | }; |
| 5380 | |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 5381 | struct cgroup_subsys memory_cgrp_subsys = { |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 5382 | .css_alloc = mem_cgroup_css_alloc, |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 5383 | .css_online = mem_cgroup_css_online, |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 5384 | .css_offline = mem_cgroup_css_offline, |
| 5385 | .css_free = mem_cgroup_css_free, |
Tejun Heo | 1ced953 | 2014-07-08 18:02:57 -0400 | [diff] [blame] | 5386 | .css_reset = mem_cgroup_css_reset, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5387 | .can_attach = mem_cgroup_can_attach, |
| 5388 | .cancel_attach = mem_cgroup_cancel_attach, |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 5389 | .attach = mem_cgroup_move_task, |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 5390 | .bind = mem_cgroup_bind, |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5391 | .dfl_cftypes = memory_files, |
| 5392 | .legacy_cftypes = mem_cgroup_legacy_files, |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 5393 | .early_init = 0, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 5394 | }; |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 5395 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5396 | /** |
| 5397 | * mem_cgroup_events - count memory events against a cgroup |
| 5398 | * @memcg: the memory cgroup |
| 5399 | * @idx: the event index |
| 5400 | * @nr: the number of events to account for |
| 5401 | */ |
| 5402 | void mem_cgroup_events(struct mem_cgroup *memcg, |
| 5403 | enum mem_cgroup_events_index idx, |
| 5404 | unsigned int nr) |
| 5405 | { |
| 5406 | this_cpu_add(memcg->stat->events[idx], nr); |
| 5407 | } |
| 5408 | |
| 5409 | /** |
| 5410 | * mem_cgroup_low - check if memory consumption is below the normal range |
| 5411 | * @root: the highest ancestor to consider |
| 5412 | * @memcg: the memory cgroup to check |
| 5413 | * |
| 5414 | * Returns %true if memory consumption of @memcg, and that of all |
| 5415 | * configurable ancestors up to @root, is below the normal range. |
| 5416 | */ |
| 5417 | bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg) |
| 5418 | { |
| 5419 | if (mem_cgroup_disabled()) |
| 5420 | return false; |
| 5421 | |
| 5422 | /* |
| 5423 | * The toplevel group doesn't have a configurable range, so |
| 5424 | * it's never low when looked at directly, and it is not |
| 5425 | * considered an ancestor when assessing the hierarchy. |
| 5426 | */ |
| 5427 | |
| 5428 | if (memcg == root_mem_cgroup) |
| 5429 | return false; |
| 5430 | |
Michal Hocko | 4e54ded | 2015-02-27 15:51:46 -0800 | [diff] [blame] | 5431 | if (page_counter_read(&memcg->memory) >= memcg->low) |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5432 | return false; |
| 5433 | |
| 5434 | while (memcg != root) { |
| 5435 | memcg = parent_mem_cgroup(memcg); |
| 5436 | |
| 5437 | if (memcg == root_mem_cgroup) |
| 5438 | break; |
| 5439 | |
Michal Hocko | 4e54ded | 2015-02-27 15:51:46 -0800 | [diff] [blame] | 5440 | if (page_counter_read(&memcg->memory) >= memcg->low) |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5441 | return false; |
| 5442 | } |
| 5443 | return true; |
| 5444 | } |
| 5445 | |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5446 | /** |
| 5447 | * mem_cgroup_try_charge - try charging a page |
| 5448 | * @page: page to charge |
| 5449 | * @mm: mm context of the victim |
| 5450 | * @gfp_mask: reclaim mode |
| 5451 | * @memcgp: charged memcg return |
| 5452 | * |
| 5453 | * Try to charge @page to the memcg that @mm belongs to, reclaiming |
| 5454 | * pages according to @gfp_mask if necessary. |
| 5455 | * |
| 5456 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. |
| 5457 | * Otherwise, an error code is returned. |
| 5458 | * |
| 5459 | * After page->mapping has been set up, the caller must finalize the |
| 5460 | * charge with mem_cgroup_commit_charge(). Or abort the transaction |
| 5461 | * with mem_cgroup_cancel_charge() in case page instantiation fails. |
| 5462 | */ |
| 5463 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, |
| 5464 | gfp_t gfp_mask, struct mem_cgroup **memcgp) |
| 5465 | { |
| 5466 | struct mem_cgroup *memcg = NULL; |
| 5467 | unsigned int nr_pages = 1; |
| 5468 | int ret = 0; |
| 5469 | |
| 5470 | if (mem_cgroup_disabled()) |
| 5471 | goto out; |
| 5472 | |
| 5473 | if (PageSwapCache(page)) { |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5474 | /* |
| 5475 | * Every swap fault against a single page tries to charge the |
| 5476 | * page, bail as early as possible. shmem_unuse() encounters |
| 5477 | * already charged pages, too. The USED bit is protected by |
| 5478 | * the page lock, which serializes swap cache removal, which |
| 5479 | * in turn serializes uncharging. |
| 5480 | */ |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5481 | if (page->mem_cgroup) |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5482 | goto out; |
| 5483 | } |
| 5484 | |
| 5485 | if (PageTransHuge(page)) { |
| 5486 | nr_pages <<= compound_order(page); |
| 5487 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
| 5488 | } |
| 5489 | |
| 5490 | if (do_swap_account && PageSwapCache(page)) |
| 5491 | memcg = try_get_mem_cgroup_from_page(page); |
| 5492 | if (!memcg) |
| 5493 | memcg = get_mem_cgroup_from_mm(mm); |
| 5494 | |
| 5495 | ret = try_charge(memcg, gfp_mask, nr_pages); |
| 5496 | |
| 5497 | css_put(&memcg->css); |
| 5498 | |
| 5499 | if (ret == -EINTR) { |
| 5500 | memcg = root_mem_cgroup; |
| 5501 | ret = 0; |
| 5502 | } |
| 5503 | out: |
| 5504 | *memcgp = memcg; |
| 5505 | return ret; |
| 5506 | } |
| 5507 | |
| 5508 | /** |
| 5509 | * mem_cgroup_commit_charge - commit a page charge |
| 5510 | * @page: page to charge |
| 5511 | * @memcg: memcg to charge the page to |
| 5512 | * @lrucare: page might be on LRU already |
| 5513 | * |
| 5514 | * Finalize a charge transaction started by mem_cgroup_try_charge(), |
| 5515 | * after page->mapping has been set up. This must happen atomically |
| 5516 | * as part of the page instantiation, i.e. under the page table lock |
| 5517 | * for anonymous pages, under the page lock for page and swap cache. |
| 5518 | * |
| 5519 | * In addition, the page must not be on the LRU during the commit, to |
| 5520 | * prevent racing with task migration. If it might be, use @lrucare. |
| 5521 | * |
| 5522 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. |
| 5523 | */ |
| 5524 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, |
| 5525 | bool lrucare) |
| 5526 | { |
| 5527 | unsigned int nr_pages = 1; |
| 5528 | |
| 5529 | VM_BUG_ON_PAGE(!page->mapping, page); |
| 5530 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); |
| 5531 | |
| 5532 | if (mem_cgroup_disabled()) |
| 5533 | return; |
| 5534 | /* |
| 5535 | * Swap faults will attempt to charge the same page multiple |
| 5536 | * times. But reuse_swap_page() might have removed the page |
| 5537 | * from swapcache already, so we can't check PageSwapCache(). |
| 5538 | */ |
| 5539 | if (!memcg) |
| 5540 | return; |
| 5541 | |
Johannes Weiner | 6abb5a8 | 2014-08-08 14:19:33 -0700 | [diff] [blame] | 5542 | commit_charge(page, memcg, lrucare); |
| 5543 | |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5544 | if (PageTransHuge(page)) { |
| 5545 | nr_pages <<= compound_order(page); |
| 5546 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
| 5547 | } |
| 5548 | |
Johannes Weiner | 6abb5a8 | 2014-08-08 14:19:33 -0700 | [diff] [blame] | 5549 | local_irq_disable(); |
| 5550 | mem_cgroup_charge_statistics(memcg, page, nr_pages); |
| 5551 | memcg_check_events(memcg, page); |
| 5552 | local_irq_enable(); |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5553 | |
| 5554 | if (do_swap_account && PageSwapCache(page)) { |
| 5555 | swp_entry_t entry = { .val = page_private(page) }; |
| 5556 | /* |
| 5557 | * The swap entry might not get freed for a long time, |
| 5558 | * let's not wait for it. The page already received a |
| 5559 | * memory+swap charge, drop the swap entry duplicate. |
| 5560 | */ |
| 5561 | mem_cgroup_uncharge_swap(entry); |
| 5562 | } |
| 5563 | } |
| 5564 | |
| 5565 | /** |
| 5566 | * mem_cgroup_cancel_charge - cancel a page charge |
| 5567 | * @page: page to charge |
| 5568 | * @memcg: memcg to charge the page to |
| 5569 | * |
| 5570 | * Cancel a charge transaction started by mem_cgroup_try_charge(). |
| 5571 | */ |
| 5572 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg) |
| 5573 | { |
| 5574 | unsigned int nr_pages = 1; |
| 5575 | |
| 5576 | if (mem_cgroup_disabled()) |
| 5577 | return; |
| 5578 | /* |
| 5579 | * Swap faults will attempt to charge the same page multiple |
| 5580 | * times. But reuse_swap_page() might have removed the page |
| 5581 | * from swapcache already, so we can't check PageSwapCache(). |
| 5582 | */ |
| 5583 | if (!memcg) |
| 5584 | return; |
| 5585 | |
| 5586 | if (PageTransHuge(page)) { |
| 5587 | nr_pages <<= compound_order(page); |
| 5588 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
| 5589 | } |
| 5590 | |
| 5591 | cancel_charge(memcg, nr_pages); |
| 5592 | } |
| 5593 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5594 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5595 | unsigned long nr_anon, unsigned long nr_file, |
| 5596 | unsigned long nr_huge, struct page *dummy_page) |
| 5597 | { |
Johannes Weiner | 18eca2e | 2014-12-10 15:43:57 -0800 | [diff] [blame] | 5598 | unsigned long nr_pages = nr_anon + nr_file; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5599 | unsigned long flags; |
| 5600 | |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 5601 | if (!mem_cgroup_is_root(memcg)) { |
Johannes Weiner | 18eca2e | 2014-12-10 15:43:57 -0800 | [diff] [blame] | 5602 | page_counter_uncharge(&memcg->memory, nr_pages); |
| 5603 | if (do_swap_account) |
| 5604 | page_counter_uncharge(&memcg->memsw, nr_pages); |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 5605 | memcg_oom_recover(memcg); |
| 5606 | } |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5607 | |
| 5608 | local_irq_save(flags); |
| 5609 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); |
| 5610 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); |
| 5611 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); |
| 5612 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); |
Johannes Weiner | 18eca2e | 2014-12-10 15:43:57 -0800 | [diff] [blame] | 5613 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5614 | memcg_check_events(memcg, dummy_page); |
| 5615 | local_irq_restore(flags); |
Johannes Weiner | e8ea14c | 2014-12-10 15:42:42 -0800 | [diff] [blame] | 5616 | |
| 5617 | if (!mem_cgroup_is_root(memcg)) |
Johannes Weiner | 18eca2e | 2014-12-10 15:43:57 -0800 | [diff] [blame] | 5618 | css_put_many(&memcg->css, nr_pages); |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5619 | } |
| 5620 | |
| 5621 | static void uncharge_list(struct list_head *page_list) |
| 5622 | { |
| 5623 | struct mem_cgroup *memcg = NULL; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5624 | unsigned long nr_anon = 0; |
| 5625 | unsigned long nr_file = 0; |
| 5626 | unsigned long nr_huge = 0; |
| 5627 | unsigned long pgpgout = 0; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5628 | struct list_head *next; |
| 5629 | struct page *page; |
| 5630 | |
| 5631 | next = page_list->next; |
| 5632 | do { |
| 5633 | unsigned int nr_pages = 1; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5634 | |
| 5635 | page = list_entry(next, struct page, lru); |
| 5636 | next = page->lru.next; |
| 5637 | |
| 5638 | VM_BUG_ON_PAGE(PageLRU(page), page); |
| 5639 | VM_BUG_ON_PAGE(page_count(page), page); |
| 5640 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5641 | if (!page->mem_cgroup) |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5642 | continue; |
| 5643 | |
| 5644 | /* |
| 5645 | * Nobody should be changing or seriously looking at |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5646 | * page->mem_cgroup at this point, we have fully |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 5647 | * exclusive access to the page. |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5648 | */ |
| 5649 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5650 | if (memcg != page->mem_cgroup) { |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5651 | if (memcg) { |
Johannes Weiner | 18eca2e | 2014-12-10 15:43:57 -0800 | [diff] [blame] | 5652 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
| 5653 | nr_huge, page); |
| 5654 | pgpgout = nr_anon = nr_file = nr_huge = 0; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5655 | } |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5656 | memcg = page->mem_cgroup; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5657 | } |
| 5658 | |
| 5659 | if (PageTransHuge(page)) { |
| 5660 | nr_pages <<= compound_order(page); |
| 5661 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
| 5662 | nr_huge += nr_pages; |
| 5663 | } |
| 5664 | |
| 5665 | if (PageAnon(page)) |
| 5666 | nr_anon += nr_pages; |
| 5667 | else |
| 5668 | nr_file += nr_pages; |
| 5669 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5670 | page->mem_cgroup = NULL; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5671 | |
| 5672 | pgpgout++; |
| 5673 | } while (next != page_list); |
| 5674 | |
| 5675 | if (memcg) |
Johannes Weiner | 18eca2e | 2014-12-10 15:43:57 -0800 | [diff] [blame] | 5676 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
| 5677 | nr_huge, page); |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5678 | } |
| 5679 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5680 | /** |
| 5681 | * mem_cgroup_uncharge - uncharge a page |
| 5682 | * @page: page to uncharge |
| 5683 | * |
| 5684 | * Uncharge a page previously charged with mem_cgroup_try_charge() and |
| 5685 | * mem_cgroup_commit_charge(). |
| 5686 | */ |
| 5687 | void mem_cgroup_uncharge(struct page *page) |
| 5688 | { |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5689 | if (mem_cgroup_disabled()) |
| 5690 | return; |
| 5691 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5692 | /* Don't touch page->lru of any random page, pre-check: */ |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5693 | if (!page->mem_cgroup) |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5694 | return; |
| 5695 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5696 | INIT_LIST_HEAD(&page->lru); |
| 5697 | uncharge_list(&page->lru); |
| 5698 | } |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5699 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5700 | /** |
| 5701 | * mem_cgroup_uncharge_list - uncharge a list of page |
| 5702 | * @page_list: list of pages to uncharge |
| 5703 | * |
| 5704 | * Uncharge a list of pages previously charged with |
| 5705 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). |
| 5706 | */ |
| 5707 | void mem_cgroup_uncharge_list(struct list_head *page_list) |
| 5708 | { |
| 5709 | if (mem_cgroup_disabled()) |
| 5710 | return; |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5711 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5712 | if (!list_empty(page_list)) |
| 5713 | uncharge_list(page_list); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5714 | } |
| 5715 | |
| 5716 | /** |
| 5717 | * mem_cgroup_migrate - migrate a charge to another page |
| 5718 | * @oldpage: currently charged page |
| 5719 | * @newpage: page to transfer the charge to |
Michal Hocko | f5e03a4 | 2015-02-05 12:25:14 -0800 | [diff] [blame] | 5720 | * @lrucare: either or both pages might be on the LRU already |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5721 | * |
| 5722 | * Migrate the charge from @oldpage to @newpage. |
| 5723 | * |
| 5724 | * Both pages must be locked, @newpage->mapping must be set up. |
| 5725 | */ |
| 5726 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage, |
| 5727 | bool lrucare) |
| 5728 | { |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 5729 | struct mem_cgroup *memcg; |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5730 | int isolated; |
| 5731 | |
| 5732 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); |
| 5733 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); |
| 5734 | VM_BUG_ON_PAGE(!lrucare && PageLRU(oldpage), oldpage); |
| 5735 | VM_BUG_ON_PAGE(!lrucare && PageLRU(newpage), newpage); |
| 5736 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
Johannes Weiner | 6abb5a8 | 2014-08-08 14:19:33 -0700 | [diff] [blame] | 5737 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
| 5738 | newpage); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5739 | |
| 5740 | if (mem_cgroup_disabled()) |
| 5741 | return; |
| 5742 | |
| 5743 | /* Page cache replacement: new page already charged? */ |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5744 | if (newpage->mem_cgroup) |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5745 | return; |
| 5746 | |
Johannes Weiner | 7d5e324 | 2014-12-10 15:43:46 -0800 | [diff] [blame] | 5747 | /* |
| 5748 | * Swapcache readahead pages can get migrated before being |
| 5749 | * charged, and migration from compaction can happen to an |
| 5750 | * uncharged page when the PFN walker finds a page that |
| 5751 | * reclaim just put back on the LRU but has not released yet. |
| 5752 | */ |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5753 | memcg = oldpage->mem_cgroup; |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 5754 | if (!memcg) |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5755 | return; |
| 5756 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5757 | if (lrucare) |
| 5758 | lock_page_lru(oldpage, &isolated); |
| 5759 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5760 | oldpage->mem_cgroup = NULL; |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5761 | |
| 5762 | if (lrucare) |
| 5763 | unlock_page_lru(oldpage, isolated); |
| 5764 | |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 5765 | commit_charge(newpage, memcg, lrucare); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5766 | } |
| 5767 | |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 5768 | /* |
Michal Hocko | 1081312 | 2013-02-22 16:35:41 -0800 | [diff] [blame] | 5769 | * subsys_initcall() for memory controller. |
| 5770 | * |
| 5771 | * Some parts like hotcpu_notifier() have to be initialized from this context |
| 5772 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically |
| 5773 | * everything that doesn't depend on a specific mem_cgroup structure should |
| 5774 | * be initialized from here. |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 5775 | */ |
| 5776 | static int __init mem_cgroup_init(void) |
| 5777 | { |
Johannes Weiner | 95a045f | 2015-02-11 15:26:33 -0800 | [diff] [blame] | 5778 | int cpu, node; |
| 5779 | |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 5780 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
Johannes Weiner | 95a045f | 2015-02-11 15:26:33 -0800 | [diff] [blame] | 5781 | |
| 5782 | for_each_possible_cpu(cpu) |
| 5783 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, |
| 5784 | drain_local_stock); |
| 5785 | |
| 5786 | for_each_node(node) { |
| 5787 | struct mem_cgroup_tree_per_node *rtpn; |
| 5788 | int zone; |
| 5789 | |
| 5790 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, |
| 5791 | node_online(node) ? node : NUMA_NO_NODE); |
| 5792 | |
| 5793 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
| 5794 | struct mem_cgroup_tree_per_zone *rtpz; |
| 5795 | |
| 5796 | rtpz = &rtpn->rb_tree_per_zone[zone]; |
| 5797 | rtpz->rb_root = RB_ROOT; |
| 5798 | spin_lock_init(&rtpz->lock); |
| 5799 | } |
| 5800 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
| 5801 | } |
| 5802 | |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 5803 | return 0; |
| 5804 | } |
| 5805 | subsys_initcall(mem_cgroup_init); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5806 | |
| 5807 | #ifdef CONFIG_MEMCG_SWAP |
| 5808 | /** |
| 5809 | * mem_cgroup_swapout - transfer a memsw charge to swap |
| 5810 | * @page: page whose memsw charge to transfer |
| 5811 | * @entry: swap entry to move the charge to |
| 5812 | * |
| 5813 | * Transfer the memsw charge of @page to @entry. |
| 5814 | */ |
| 5815 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) |
| 5816 | { |
| 5817 | struct mem_cgroup *memcg; |
| 5818 | unsigned short oldid; |
| 5819 | |
| 5820 | VM_BUG_ON_PAGE(PageLRU(page), page); |
| 5821 | VM_BUG_ON_PAGE(page_count(page), page); |
| 5822 | |
| 5823 | if (!do_swap_account) |
| 5824 | return; |
| 5825 | |
| 5826 | memcg = page->mem_cgroup; |
| 5827 | |
| 5828 | /* Readahead page, never charged */ |
| 5829 | if (!memcg) |
| 5830 | return; |
| 5831 | |
| 5832 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); |
| 5833 | VM_BUG_ON_PAGE(oldid, page); |
| 5834 | mem_cgroup_swap_statistics(memcg, true); |
| 5835 | |
| 5836 | page->mem_cgroup = NULL; |
| 5837 | |
| 5838 | if (!mem_cgroup_is_root(memcg)) |
| 5839 | page_counter_uncharge(&memcg->memory, 1); |
| 5840 | |
| 5841 | /* XXX: caller holds IRQ-safe mapping->tree_lock */ |
| 5842 | VM_BUG_ON(!irqs_disabled()); |
| 5843 | |
| 5844 | mem_cgroup_charge_statistics(memcg, page, -1); |
| 5845 | memcg_check_events(memcg, page); |
| 5846 | } |
| 5847 | |
| 5848 | /** |
| 5849 | * mem_cgroup_uncharge_swap - uncharge a swap entry |
| 5850 | * @entry: swap entry to uncharge |
| 5851 | * |
| 5852 | * Drop the memsw charge associated with @entry. |
| 5853 | */ |
| 5854 | void mem_cgroup_uncharge_swap(swp_entry_t entry) |
| 5855 | { |
| 5856 | struct mem_cgroup *memcg; |
| 5857 | unsigned short id; |
| 5858 | |
| 5859 | if (!do_swap_account) |
| 5860 | return; |
| 5861 | |
| 5862 | id = swap_cgroup_record(entry, 0); |
| 5863 | rcu_read_lock(); |
| 5864 | memcg = mem_cgroup_lookup(id); |
| 5865 | if (memcg) { |
| 5866 | if (!mem_cgroup_is_root(memcg)) |
| 5867 | page_counter_uncharge(&memcg->memsw, 1); |
| 5868 | mem_cgroup_swap_statistics(memcg, false); |
| 5869 | css_put(&memcg->css); |
| 5870 | } |
| 5871 | rcu_read_unlock(); |
| 5872 | } |
| 5873 | |
| 5874 | /* for remember boot option*/ |
| 5875 | #ifdef CONFIG_MEMCG_SWAP_ENABLED |
| 5876 | static int really_do_swap_account __initdata = 1; |
| 5877 | #else |
| 5878 | static int really_do_swap_account __initdata; |
| 5879 | #endif |
| 5880 | |
| 5881 | static int __init enable_swap_account(char *s) |
| 5882 | { |
| 5883 | if (!strcmp(s, "1")) |
| 5884 | really_do_swap_account = 1; |
| 5885 | else if (!strcmp(s, "0")) |
| 5886 | really_do_swap_account = 0; |
| 5887 | return 1; |
| 5888 | } |
| 5889 | __setup("swapaccount=", enable_swap_account); |
| 5890 | |
| 5891 | static struct cftype memsw_cgroup_files[] = { |
| 5892 | { |
| 5893 | .name = "memsw.usage_in_bytes", |
| 5894 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), |
| 5895 | .read_u64 = mem_cgroup_read_u64, |
| 5896 | }, |
| 5897 | { |
| 5898 | .name = "memsw.max_usage_in_bytes", |
| 5899 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), |
| 5900 | .write = mem_cgroup_reset, |
| 5901 | .read_u64 = mem_cgroup_read_u64, |
| 5902 | }, |
| 5903 | { |
| 5904 | .name = "memsw.limit_in_bytes", |
| 5905 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), |
| 5906 | .write = mem_cgroup_write, |
| 5907 | .read_u64 = mem_cgroup_read_u64, |
| 5908 | }, |
| 5909 | { |
| 5910 | .name = "memsw.failcnt", |
| 5911 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), |
| 5912 | .write = mem_cgroup_reset, |
| 5913 | .read_u64 = mem_cgroup_read_u64, |
| 5914 | }, |
| 5915 | { }, /* terminate */ |
| 5916 | }; |
| 5917 | |
| 5918 | static int __init mem_cgroup_swap_init(void) |
| 5919 | { |
| 5920 | if (!mem_cgroup_disabled() && really_do_swap_account) { |
| 5921 | do_swap_account = 1; |
| 5922 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
| 5923 | memsw_cgroup_files)); |
| 5924 | } |
| 5925 | return 0; |
| 5926 | } |
| 5927 | subsys_initcall(mem_cgroup_swap_init); |
| 5928 | |
| 5929 | #endif /* CONFIG_MEMCG_SWAP */ |