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 | * |
Johannes Weiner | 1575e68 | 2015-04-14 15:44:51 -0700 | [diff] [blame] | 17 | * Native page reclaim |
| 18 | * Charge lifetime sanitation |
| 19 | * Lockless page tracking & accounting |
| 20 | * Unified hierarchy configuration model |
| 21 | * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner |
| 22 | * |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 23 | * This program is free software; you can redistribute it and/or modify |
| 24 | * it under the terms of the GNU General Public License as published by |
| 25 | * the Free Software Foundation; either version 2 of the License, or |
| 26 | * (at your option) any later version. |
| 27 | * |
| 28 | * This program is distributed in the hope that it will be useful, |
| 29 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 30 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 31 | * GNU General Public License for more details. |
| 32 | */ |
| 33 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 34 | #include <linux/page_counter.h> |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 35 | #include <linux/memcontrol.h> |
| 36 | #include <linux/cgroup.h> |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 37 | #include <linux/mm.h> |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 38 | #include <linux/hugetlb.h> |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 39 | #include <linux/pagemap.h> |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 40 | #include <linux/smp.h> |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 41 | #include <linux/page-flags.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 42 | #include <linux/backing-dev.h> |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 43 | #include <linux/bit_spinlock.h> |
| 44 | #include <linux/rcupdate.h> |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 45 | #include <linux/limits.h> |
Paul Gortmaker | b9e15ba | 2011-05-26 16:00:52 -0400 | [diff] [blame] | 46 | #include <linux/export.h> |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 47 | #include <linux/mutex.h> |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 48 | #include <linux/rbtree.h> |
Balbir Singh | b6ac57d | 2008-04-29 01:00:19 -0700 | [diff] [blame] | 49 | #include <linux/slab.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 50 | #include <linux/swap.h> |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 51 | #include <linux/swapops.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 52 | #include <linux/spinlock.h> |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 53 | #include <linux/eventfd.h> |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 54 | #include <linux/poll.h> |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 55 | #include <linux/sort.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 56 | #include <linux/fs.h> |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 57 | #include <linux/seq_file.h> |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 58 | #include <linux/vmpressure.h> |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 59 | #include <linux/mm_inline.h> |
Johannes Weiner | 5d1ea48 | 2014-12-10 15:44:55 -0800 | [diff] [blame] | 60 | #include <linux/swap_cgroup.h> |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 61 | #include <linux/cpu.h> |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 62 | #include <linux/oom.h> |
Johannes Weiner | 0056f4e | 2013-10-31 16:34:14 -0700 | [diff] [blame] | 63 | #include <linux/lockdep.h> |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 64 | #include <linux/file.h> |
Tejun Heo | b23afb9 | 2015-11-05 18:46:11 -0800 | [diff] [blame] | 65 | #include <linux/tracehook.h> |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 66 | #include "internal.h" |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 67 | #include <net/sock.h> |
Michal Hocko | 4bd2c1e | 2012-10-08 16:33:10 -0700 | [diff] [blame] | 68 | #include <net/ip.h> |
Qiang Huang | f35c3a8 | 2013-11-12 15:08:22 -0800 | [diff] [blame] | 69 | #include "slab.h" |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 70 | |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 71 | #include <asm/uaccess.h> |
| 72 | |
KOSAKI Motohiro | cc8e970 | 2010-08-09 17:19:57 -0700 | [diff] [blame] | 73 | #include <trace/events/vmscan.h> |
| 74 | |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 75 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
| 76 | EXPORT_SYMBOL(memory_cgrp_subsys); |
David Rientjes | 68ae564 | 2012-12-12 13:51:57 -0800 | [diff] [blame] | 77 | |
Johannes Weiner | 7d82860 | 2016-01-14 15:20:56 -0800 | [diff] [blame] | 78 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
| 79 | |
KAMEZAWA Hiroyuki | a181b0e | 2008-07-25 01:47:08 -0700 | [diff] [blame] | 80 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 81 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 82 | /* Socket memory accounting disabled? */ |
| 83 | static bool cgroup_memory_nosocket; |
| 84 | |
Vladimir Davydov | 04823c8 | 2016-01-20 15:02:38 -0800 | [diff] [blame] | 85 | /* Kernel memory accounting disabled? */ |
| 86 | static bool cgroup_memory_nokmem; |
| 87 | |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 88 | /* Whether the swap controller is active */ |
Andrew Morton | c255a45 | 2012-07-31 16:43:02 -0700 | [diff] [blame] | 89 | #ifdef CONFIG_MEMCG_SWAP |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 90 | int do_swap_account __read_mostly; |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 91 | #else |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 92 | #define do_swap_account 0 |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 93 | #endif |
| 94 | |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 95 | /* Whether legacy memory+swap accounting is active */ |
| 96 | static bool do_memsw_account(void) |
| 97 | { |
| 98 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account; |
| 99 | } |
| 100 | |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 101 | static const char * const mem_cgroup_stat_names[] = { |
| 102 | "cache", |
| 103 | "rss", |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 104 | "rss_huge", |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 105 | "mapped_file", |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 106 | "dirty", |
Sha Zhengju | 3ea67d0 | 2013-09-12 15:13:53 -0700 | [diff] [blame] | 107 | "writeback", |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 108 | "swap", |
| 109 | }; |
| 110 | |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 111 | static const char * const mem_cgroup_events_names[] = { |
| 112 | "pgpgin", |
| 113 | "pgpgout", |
| 114 | "pgfault", |
| 115 | "pgmajfault", |
| 116 | }; |
| 117 | |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 118 | static const char * const mem_cgroup_lru_names[] = { |
| 119 | "inactive_anon", |
| 120 | "active_anon", |
| 121 | "inactive_file", |
| 122 | "active_file", |
| 123 | "unevictable", |
| 124 | }; |
| 125 | |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 126 | #define THRESHOLDS_EVENTS_TARGET 128 |
| 127 | #define SOFTLIMIT_EVENTS_TARGET 1024 |
| 128 | #define NUMAINFO_EVENTS_TARGET 1024 |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 129 | |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 130 | /* |
| 131 | * Cgroups above their limits are maintained in a RB-Tree, independent of |
| 132 | * their hierarchy representation |
| 133 | */ |
| 134 | |
| 135 | struct mem_cgroup_tree_per_zone { |
| 136 | struct rb_root rb_root; |
| 137 | spinlock_t lock; |
| 138 | }; |
| 139 | |
| 140 | struct mem_cgroup_tree_per_node { |
| 141 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; |
| 142 | }; |
| 143 | |
| 144 | struct mem_cgroup_tree { |
| 145 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; |
| 146 | }; |
| 147 | |
| 148 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; |
| 149 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 150 | /* for OOM */ |
| 151 | struct mem_cgroup_eventfd_list { |
| 152 | struct list_head list; |
| 153 | struct eventfd_ctx *eventfd; |
| 154 | }; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 155 | |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 156 | /* |
| 157 | * cgroup_event represents events which userspace want to receive. |
| 158 | */ |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 159 | struct mem_cgroup_event { |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 160 | /* |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 161 | * memcg which the event belongs to. |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 162 | */ |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 163 | struct mem_cgroup *memcg; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 164 | /* |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 165 | * eventfd to signal userspace about the event. |
| 166 | */ |
| 167 | struct eventfd_ctx *eventfd; |
| 168 | /* |
| 169 | * Each of these stored in a list by the cgroup. |
| 170 | */ |
| 171 | struct list_head list; |
| 172 | /* |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 173 | * register_event() callback will be used to add new userspace |
| 174 | * waiter for changes related to this event. Use eventfd_signal() |
| 175 | * on eventfd to send notification to userspace. |
| 176 | */ |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 177 | int (*register_event)(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 178 | struct eventfd_ctx *eventfd, const char *args); |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 179 | /* |
| 180 | * unregister_event() callback will be called when userspace closes |
| 181 | * the eventfd or on cgroup removing. This callback must be set, |
| 182 | * if you want provide notification functionality. |
| 183 | */ |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 184 | void (*unregister_event)(struct mem_cgroup *memcg, |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 185 | struct eventfd_ctx *eventfd); |
| 186 | /* |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 187 | * All fields below needed to unregister event when |
| 188 | * userspace closes eventfd. |
| 189 | */ |
| 190 | poll_table pt; |
| 191 | wait_queue_head_t *wqh; |
| 192 | wait_queue_t wait; |
| 193 | struct work_struct remove; |
| 194 | }; |
| 195 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 196 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
| 197 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 198 | |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 199 | /* Stuffs for move charges at task migration. */ |
| 200 | /* |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 201 | * Types of charges to be moved. |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 202 | */ |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 203 | #define MOVE_ANON 0x1U |
| 204 | #define MOVE_FILE 0x2U |
| 205 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 206 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 207 | /* "mc" and its members are protected by cgroup_mutex */ |
| 208 | static struct move_charge_struct { |
Daisuke Nishimura | b1dd693 | 2010-11-24 12:57:06 -0800 | [diff] [blame] | 209 | spinlock_t lock; /* for from, to */ |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 210 | struct mm_struct *mm; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 211 | struct mem_cgroup *from; |
| 212 | struct mem_cgroup *to; |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 213 | unsigned long flags; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 214 | unsigned long precharge; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 215 | unsigned long moved_charge; |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 216 | unsigned long moved_swap; |
Daisuke Nishimura | 8033b97 | 2010-03-10 15:22:16 -0800 | [diff] [blame] | 217 | struct task_struct *moving_task; /* a task moving charges */ |
| 218 | wait_queue_head_t waitq; /* a waitq for other context */ |
| 219 | } mc = { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 220 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
Daisuke Nishimura | 8033b97 | 2010-03-10 15:22:16 -0800 | [diff] [blame] | 221 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
| 222 | }; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 223 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 224 | /* |
| 225 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft |
| 226 | * limit reclaim to prevent infinite loops, if they ever occur. |
| 227 | */ |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 228 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 229 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 230 | |
KAMEZAWA Hiroyuki | 217bc31 | 2008-02-07 00:14:17 -0800 | [diff] [blame] | 231 | enum charge_type { |
| 232 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, |
Kamezawa Hiroyuki | 41326c1 | 2012-07-31 16:41:40 -0700 | [diff] [blame] | 233 | MEM_CGROUP_CHARGE_TYPE_ANON, |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 234 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
KAMEZAWA Hiroyuki | 8a9478c | 2009-06-17 16:27:17 -0700 | [diff] [blame] | 235 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
KAMEZAWA Hiroyuki | c05555b | 2008-10-18 20:28:11 -0700 | [diff] [blame] | 236 | NR_CHARGE_TYPE, |
| 237 | }; |
| 238 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 239 | /* for encoding cft->private value on file */ |
Glauber Costa | 86ae53e | 2012-12-18 14:21:45 -0800 | [diff] [blame] | 240 | enum res_type { |
| 241 | _MEM, |
| 242 | _MEMSWAP, |
| 243 | _OOM_TYPE, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 244 | _KMEM, |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 245 | _TCP, |
Glauber Costa | 86ae53e | 2012-12-18 14:21:45 -0800 | [diff] [blame] | 246 | }; |
| 247 | |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 248 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
| 249 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 250 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 251 | /* Used for OOM nofiier */ |
| 252 | #define OOM_CONTROL (0) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 253 | |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 254 | /* Some nice accessors for the vmpressure. */ |
| 255 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) |
| 256 | { |
| 257 | if (!memcg) |
| 258 | memcg = root_mem_cgroup; |
| 259 | return &memcg->vmpressure; |
| 260 | } |
| 261 | |
| 262 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) |
| 263 | { |
| 264 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; |
| 265 | } |
| 266 | |
Michal Hocko | 7ffc0ed | 2012-10-08 16:33:13 -0700 | [diff] [blame] | 267 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
| 268 | { |
| 269 | return (memcg == root_mem_cgroup); |
| 270 | } |
| 271 | |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 272 | #ifndef CONFIG_SLOB |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 273 | /* |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 274 | * 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] | 275 | * The main reason for not using cgroup id for this: |
| 276 | * this works better in sparse environments, where we have a lot of memcgs, |
| 277 | * but only a few kmem-limited. Or also, if we have, for instance, 200 |
| 278 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a |
| 279 | * 200 entry array for that. |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 280 | * |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 281 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
| 282 | * will double each time we have to increase it. |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 283 | */ |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 284 | static DEFINE_IDA(memcg_cache_ida); |
| 285 | int memcg_nr_cache_ids; |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 286 | |
Vladimir Davydov | 05257a1 | 2015-02-12 14:59:01 -0800 | [diff] [blame] | 287 | /* Protects memcg_nr_cache_ids */ |
| 288 | static DECLARE_RWSEM(memcg_cache_ids_sem); |
| 289 | |
| 290 | void memcg_get_cache_ids(void) |
| 291 | { |
| 292 | down_read(&memcg_cache_ids_sem); |
| 293 | } |
| 294 | |
| 295 | void memcg_put_cache_ids(void) |
| 296 | { |
| 297 | up_read(&memcg_cache_ids_sem); |
| 298 | } |
| 299 | |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 300 | /* |
| 301 | * MIN_SIZE is different than 1, because we would like to avoid going through |
| 302 | * the alloc/free process all the time. In a small machine, 4 kmem-limited |
| 303 | * cgroups is a reasonable guess. In the future, it could be a parameter or |
| 304 | * tunable, but that is strictly not necessary. |
| 305 | * |
Li Zefan | b862783 | 2013-09-23 16:56:47 +0800 | [diff] [blame] | 306 | * 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] | 307 | * this constant directly from cgroup, but it is understandable that this is |
| 308 | * 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] | 309 | * 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] | 310 | * increase ours as well if it increases. |
| 311 | */ |
| 312 | #define MEMCG_CACHES_MIN_SIZE 4 |
Li Zefan | b862783 | 2013-09-23 16:56:47 +0800 | [diff] [blame] | 313 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 314 | |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 315 | /* |
| 316 | * A lot of the calls to the cache allocation functions are expected to be |
| 317 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are |
| 318 | * conditional to this static branch, we'll have to allow modules that does |
| 319 | * kmem_cache_alloc and the such to see this symbol as well |
| 320 | */ |
Johannes Weiner | ef12947 | 2016-01-14 15:21:34 -0800 | [diff] [blame] | 321 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 322 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
Glauber Costa | a8964b9 | 2012-12-18 14:22:09 -0800 | [diff] [blame] | 323 | |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 324 | #endif /* !CONFIG_SLOB */ |
Glauber Costa | a8964b9 | 2012-12-18 14:22:09 -0800 | [diff] [blame] | 325 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 326 | static struct mem_cgroup_per_zone * |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 327 | mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone) |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 328 | { |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 329 | int nid = zone_to_nid(zone); |
| 330 | int zid = zone_idx(zone); |
| 331 | |
Johannes Weiner | 54f72fe | 2013-07-08 15:59:49 -0700 | [diff] [blame] | 332 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 333 | } |
| 334 | |
Tejun Heo | ad7fa85 | 2015-05-27 20:00:02 -0400 | [diff] [blame] | 335 | /** |
| 336 | * mem_cgroup_css_from_page - css of the memcg associated with a page |
| 337 | * @page: page of interest |
| 338 | * |
| 339 | * If memcg is bound to the default hierarchy, css of the memcg associated |
| 340 | * with @page is returned. The returned css remains associated with @page |
| 341 | * until it is released. |
| 342 | * |
| 343 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup |
| 344 | * is returned. |
Tejun Heo | ad7fa85 | 2015-05-27 20:00:02 -0400 | [diff] [blame] | 345 | */ |
| 346 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) |
| 347 | { |
| 348 | struct mem_cgroup *memcg; |
| 349 | |
Tejun Heo | ad7fa85 | 2015-05-27 20:00:02 -0400 | [diff] [blame] | 350 | memcg = page->mem_cgroup; |
| 351 | |
Tejun Heo | 9e10a13 | 2015-09-18 11:56:28 -0400 | [diff] [blame] | 352 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
Tejun Heo | ad7fa85 | 2015-05-27 20:00:02 -0400 | [diff] [blame] | 353 | memcg = root_mem_cgroup; |
| 354 | |
Tejun Heo | ad7fa85 | 2015-05-27 20:00:02 -0400 | [diff] [blame] | 355 | return &memcg->css; |
| 356 | } |
| 357 | |
Vladimir Davydov | 2fc0452 | 2015-09-09 15:35:28 -0700 | [diff] [blame] | 358 | /** |
| 359 | * page_cgroup_ino - return inode number of the memcg a page is charged to |
| 360 | * @page: the page |
| 361 | * |
| 362 | * Look up the closest online ancestor of the memory cgroup @page is charged to |
| 363 | * and return its inode number or 0 if @page is not charged to any cgroup. It |
| 364 | * is safe to call this function without holding a reference to @page. |
| 365 | * |
| 366 | * Note, this function is inherently racy, because there is nothing to prevent |
| 367 | * the cgroup inode from getting torn down and potentially reallocated a moment |
| 368 | * after page_cgroup_ino() returns, so it only should be used by callers that |
| 369 | * do not care (such as procfs interfaces). |
| 370 | */ |
| 371 | ino_t page_cgroup_ino(struct page *page) |
| 372 | { |
| 373 | struct mem_cgroup *memcg; |
| 374 | unsigned long ino = 0; |
| 375 | |
| 376 | rcu_read_lock(); |
| 377 | memcg = READ_ONCE(page->mem_cgroup); |
| 378 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) |
| 379 | memcg = parent_mem_cgroup(memcg); |
| 380 | if (memcg) |
| 381 | ino = cgroup_ino(memcg->css.cgroup); |
| 382 | rcu_read_unlock(); |
| 383 | return ino; |
| 384 | } |
| 385 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 386 | static struct mem_cgroup_per_zone * |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 387 | mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 388 | { |
Johannes Weiner | 97a6c37 | 2011-03-23 16:42:27 -0700 | [diff] [blame] | 389 | int nid = page_to_nid(page); |
| 390 | int zid = page_zonenum(page); |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 391 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 392 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 393 | } |
| 394 | |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 395 | static struct mem_cgroup_tree_per_zone * |
| 396 | soft_limit_tree_node_zone(int nid, int zid) |
| 397 | { |
| 398 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; |
| 399 | } |
| 400 | |
| 401 | static struct mem_cgroup_tree_per_zone * |
| 402 | soft_limit_tree_from_page(struct page *page) |
| 403 | { |
| 404 | int nid = page_to_nid(page); |
| 405 | int zid = page_zonenum(page); |
| 406 | |
| 407 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; |
| 408 | } |
| 409 | |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 410 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz, |
| 411 | struct mem_cgroup_tree_per_zone *mctz, |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 412 | unsigned long new_usage_in_excess) |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 413 | { |
| 414 | struct rb_node **p = &mctz->rb_root.rb_node; |
| 415 | struct rb_node *parent = NULL; |
| 416 | struct mem_cgroup_per_zone *mz_node; |
| 417 | |
| 418 | if (mz->on_tree) |
| 419 | return; |
| 420 | |
| 421 | mz->usage_in_excess = new_usage_in_excess; |
| 422 | if (!mz->usage_in_excess) |
| 423 | return; |
| 424 | while (*p) { |
| 425 | parent = *p; |
| 426 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, |
| 427 | tree_node); |
| 428 | if (mz->usage_in_excess < mz_node->usage_in_excess) |
| 429 | p = &(*p)->rb_left; |
| 430 | /* |
| 431 | * We can't avoid mem cgroups that are over their soft |
| 432 | * limit by the same amount |
| 433 | */ |
| 434 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) |
| 435 | p = &(*p)->rb_right; |
| 436 | } |
| 437 | rb_link_node(&mz->tree_node, parent, p); |
| 438 | rb_insert_color(&mz->tree_node, &mctz->rb_root); |
| 439 | mz->on_tree = true; |
| 440 | } |
| 441 | |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 442 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
| 443 | struct mem_cgroup_tree_per_zone *mctz) |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 444 | { |
| 445 | if (!mz->on_tree) |
| 446 | return; |
| 447 | rb_erase(&mz->tree_node, &mctz->rb_root); |
| 448 | mz->on_tree = false; |
| 449 | } |
| 450 | |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 451 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
| 452 | struct mem_cgroup_tree_per_zone *mctz) |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 453 | { |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 454 | unsigned long flags; |
| 455 | |
| 456 | spin_lock_irqsave(&mctz->lock, flags); |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 457 | __mem_cgroup_remove_exceeded(mz, mctz); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 458 | spin_unlock_irqrestore(&mctz->lock, flags); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 459 | } |
| 460 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 461 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
| 462 | { |
| 463 | unsigned long nr_pages = page_counter_read(&memcg->memory); |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 464 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 465 | unsigned long excess = 0; |
| 466 | |
| 467 | if (nr_pages > soft_limit) |
| 468 | excess = nr_pages - soft_limit; |
| 469 | |
| 470 | return excess; |
| 471 | } |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 472 | |
| 473 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) |
| 474 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 475 | unsigned long excess; |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 476 | struct mem_cgroup_per_zone *mz; |
| 477 | struct mem_cgroup_tree_per_zone *mctz; |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 478 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 479 | mctz = soft_limit_tree_from_page(page); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 480 | /* |
| 481 | * Necessary to update all ancestors when hierarchy is used. |
| 482 | * because their event counter is not touched. |
| 483 | */ |
| 484 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 485 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 486 | excess = soft_limit_excess(memcg); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 487 | /* |
| 488 | * We have to update the tree if mz is on RB-tree or |
| 489 | * mem is over its softlimit. |
| 490 | */ |
| 491 | if (excess || mz->on_tree) { |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 492 | unsigned long flags; |
| 493 | |
| 494 | spin_lock_irqsave(&mctz->lock, flags); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 495 | /* if on-tree, remove it */ |
| 496 | if (mz->on_tree) |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 497 | __mem_cgroup_remove_exceeded(mz, mctz); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 498 | /* |
| 499 | * Insert again. mz->usage_in_excess will be updated. |
| 500 | * If excess is 0, no tree ops. |
| 501 | */ |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 502 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 503 | spin_unlock_irqrestore(&mctz->lock, flags); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 504 | } |
| 505 | } |
| 506 | } |
| 507 | |
| 508 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) |
| 509 | { |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 510 | struct mem_cgroup_tree_per_zone *mctz; |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 511 | struct mem_cgroup_per_zone *mz; |
| 512 | int nid, zid; |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 513 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 514 | for_each_node(nid) { |
| 515 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
| 516 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
| 517 | mctz = soft_limit_tree_node_zone(nid, zid); |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 518 | mem_cgroup_remove_exceeded(mz, mctz); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 519 | } |
| 520 | } |
| 521 | } |
| 522 | |
| 523 | static struct mem_cgroup_per_zone * |
| 524 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) |
| 525 | { |
| 526 | struct rb_node *rightmost = NULL; |
| 527 | struct mem_cgroup_per_zone *mz; |
| 528 | |
| 529 | retry: |
| 530 | mz = NULL; |
| 531 | rightmost = rb_last(&mctz->rb_root); |
| 532 | if (!rightmost) |
| 533 | goto done; /* Nothing to reclaim from */ |
| 534 | |
| 535 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); |
| 536 | /* |
| 537 | * Remove the node now but someone else can add it back, |
| 538 | * we will to add it back at the end of reclaim to its correct |
| 539 | * position in the tree. |
| 540 | */ |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 541 | __mem_cgroup_remove_exceeded(mz, mctz); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 542 | if (!soft_limit_excess(mz->memcg) || |
Tejun Heo | ec903c0 | 2014-05-13 12:11:01 -0400 | [diff] [blame] | 543 | !css_tryget_online(&mz->memcg->css)) |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 544 | goto retry; |
| 545 | done: |
| 546 | return mz; |
| 547 | } |
| 548 | |
| 549 | static struct mem_cgroup_per_zone * |
| 550 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) |
| 551 | { |
| 552 | struct mem_cgroup_per_zone *mz; |
| 553 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 554 | spin_lock_irq(&mctz->lock); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 555 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 556 | spin_unlock_irq(&mctz->lock); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 557 | return mz; |
| 558 | } |
| 559 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 560 | /* |
Greg Thelen | 484ebb3 | 2015-10-01 15:37:05 -0700 | [diff] [blame] | 561 | * Return page count for single (non recursive) @memcg. |
| 562 | * |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 563 | * Implementation Note: reading percpu statistics for memcg. |
| 564 | * |
| 565 | * Both of vmstat[] and percpu_counter has threshold and do periodic |
| 566 | * synchronization to implement "quick" read. There are trade-off between |
| 567 | * reading cost and precision of value. Then, we may have a chance to implement |
Greg Thelen | 484ebb3 | 2015-10-01 15:37:05 -0700 | [diff] [blame] | 568 | * a periodic synchronization of counter in memcg's counter. |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 569 | * |
| 570 | * But this _read() function is used for user interface now. The user accounts |
| 571 | * memory usage by memory cgroup and he _always_ requires exact value because |
| 572 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always |
| 573 | * have to visit all online cpus and make sum. So, for now, unnecessary |
| 574 | * synchronization is not implemented. (just implemented for cpu hotplug) |
| 575 | * |
| 576 | * If there are kernel internal actions which can make use of some not-exact |
| 577 | * value, and reading all cpu value can be performance bottleneck in some |
Greg Thelen | 484ebb3 | 2015-10-01 15:37:05 -0700 | [diff] [blame] | 578 | * common workload, threshold and synchronization as vmstat[] should be |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 579 | * implemented. |
| 580 | */ |
Greg Thelen | 484ebb3 | 2015-10-01 15:37:05 -0700 | [diff] [blame] | 581 | static unsigned long |
| 582 | mem_cgroup_read_stat(struct mem_cgroup *memcg, enum mem_cgroup_stat_index idx) |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 583 | { |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 584 | long val = 0; |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 585 | int cpu; |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 586 | |
Greg Thelen | 484ebb3 | 2015-10-01 15:37:05 -0700 | [diff] [blame] | 587 | /* Per-cpu values can be negative, use a signed accumulator */ |
Tejun Heo | 733a572 | 2015-05-22 18:23:18 -0400 | [diff] [blame] | 588 | for_each_possible_cpu(cpu) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 589 | val += per_cpu(memcg->stat->count[idx], cpu); |
Greg Thelen | 484ebb3 | 2015-10-01 15:37:05 -0700 | [diff] [blame] | 590 | /* |
| 591 | * Summing races with updates, so val may be negative. Avoid exposing |
| 592 | * transient negative values. |
| 593 | */ |
| 594 | if (val < 0) |
| 595 | val = 0; |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 596 | return val; |
| 597 | } |
| 598 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 599 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 600 | enum mem_cgroup_events_index idx) |
| 601 | { |
| 602 | unsigned long val = 0; |
| 603 | int cpu; |
| 604 | |
Tejun Heo | 733a572 | 2015-05-22 18:23:18 -0400 | [diff] [blame] | 605 | for_each_possible_cpu(cpu) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 606 | val += per_cpu(memcg->stat->events[idx], cpu); |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 607 | return val; |
| 608 | } |
| 609 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 610 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 611 | struct page *page, |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 612 | bool compound, int nr_pages) |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 613 | { |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 614 | /* |
| 615 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is |
| 616 | * counted as CACHE even if it's on ANON LRU. |
| 617 | */ |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 618 | if (PageAnon(page)) |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 619 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 620 | nr_pages); |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 621 | else |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 622 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 623 | nr_pages); |
Balaji Rao | 55e462b | 2008-05-01 04:35:12 -0700 | [diff] [blame] | 624 | |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 625 | if (compound) { |
| 626 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 627 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
| 628 | nr_pages); |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 629 | } |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 630 | |
KAMEZAWA Hiroyuki | e401f17 | 2011-01-20 14:44:23 -0800 | [diff] [blame] | 631 | /* pagein of a big page is an event. So, ignore page size */ |
| 632 | if (nr_pages > 0) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 633 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
KAMEZAWA Hiroyuki | 3751d60 | 2011-02-01 15:52:45 -0800 | [diff] [blame] | 634 | else { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 635 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
KAMEZAWA Hiroyuki | 3751d60 | 2011-02-01 15:52:45 -0800 | [diff] [blame] | 636 | nr_pages = -nr_pages; /* for event */ |
| 637 | } |
KAMEZAWA Hiroyuki | e401f17 | 2011-01-20 14:44:23 -0800 | [diff] [blame] | 638 | |
Johannes Weiner | 1311471 | 2012-05-29 15:07:07 -0700 | [diff] [blame] | 639 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 640 | } |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 641 | |
Vladimir Davydov | 0a6b76d | 2016-03-17 14:18:42 -0700 | [diff] [blame] | 642 | unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
| 643 | int nid, unsigned int lru_mask) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 644 | { |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 645 | unsigned long nr = 0; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 646 | int zid; |
| 647 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 648 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 649 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 650 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
| 651 | struct mem_cgroup_per_zone *mz; |
| 652 | enum lru_list lru; |
| 653 | |
| 654 | for_each_lru(lru) { |
| 655 | if (!(BIT(lru) & lru_mask)) |
| 656 | continue; |
| 657 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
| 658 | nr += mz->lru_size[lru]; |
| 659 | } |
| 660 | } |
| 661 | return nr; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 662 | } |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 663 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 664 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 665 | unsigned int lru_mask) |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 666 | { |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 667 | unsigned long nr = 0; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 668 | int nid; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 669 | |
Lai Jiangshan | 31aaea4 | 2012-12-12 13:51:27 -0800 | [diff] [blame] | 670 | for_each_node_state(nid, N_MEMORY) |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 671 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
| 672 | return nr; |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 673 | } |
| 674 | |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 675 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
| 676 | enum mem_cgroup_events_target target) |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 677 | { |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 678 | unsigned long val, next; |
| 679 | |
Johannes Weiner | 1311471 | 2012-05-29 15:07:07 -0700 | [diff] [blame] | 680 | val = __this_cpu_read(memcg->stat->nr_page_events); |
Steven Rostedt | 4799401 | 2011-11-02 13:38:33 -0700 | [diff] [blame] | 681 | next = __this_cpu_read(memcg->stat->targets[target]); |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 682 | /* from time_after() in jiffies.h */ |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 683 | if ((long)next - (long)val < 0) { |
| 684 | switch (target) { |
| 685 | case MEM_CGROUP_TARGET_THRESH: |
| 686 | next = val + THRESHOLDS_EVENTS_TARGET; |
| 687 | break; |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 688 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
| 689 | next = val + SOFTLIMIT_EVENTS_TARGET; |
| 690 | break; |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 691 | case MEM_CGROUP_TARGET_NUMAINFO: |
| 692 | next = val + NUMAINFO_EVENTS_TARGET; |
| 693 | break; |
| 694 | default: |
| 695 | break; |
| 696 | } |
| 697 | __this_cpu_write(memcg->stat->targets[target], next); |
| 698 | return true; |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 699 | } |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 700 | return false; |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 701 | } |
| 702 | |
| 703 | /* |
| 704 | * Check events in order. |
| 705 | * |
| 706 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 707 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 708 | { |
| 709 | /* threshold event is triggered in finer grain than soft limit */ |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 710 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
| 711 | MEM_CGROUP_TARGET_THRESH))) { |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 712 | bool do_softlimit; |
Andrew Morton | 82b3f2a | 2012-02-03 15:37:14 -0800 | [diff] [blame] | 713 | bool do_numainfo __maybe_unused; |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 714 | |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 715 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
| 716 | MEM_CGROUP_TARGET_SOFTLIMIT); |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 717 | #if MAX_NUMNODES > 1 |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 718 | do_numainfo = mem_cgroup_event_ratelimit(memcg, |
| 719 | MEM_CGROUP_TARGET_NUMAINFO); |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 720 | #endif |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 721 | mem_cgroup_threshold(memcg); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 722 | if (unlikely(do_softlimit)) |
| 723 | mem_cgroup_update_tree(memcg, page); |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 724 | #if MAX_NUMNODES > 1 |
| 725 | if (unlikely(do_numainfo)) |
| 726 | atomic_inc(&memcg->numainfo_events); |
| 727 | #endif |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 728 | } |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 729 | } |
| 730 | |
Balbir Singh | cf475ad | 2008-04-29 01:00:16 -0700 | [diff] [blame] | 731 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 732 | { |
Balbir Singh | 31a78f2 | 2008-09-28 23:09:31 +0100 | [diff] [blame] | 733 | /* |
| 734 | * mm_update_next_owner() may clear mm->owner to NULL |
| 735 | * if it races with swapoff, page migration, etc. |
| 736 | * So this can be called with p == NULL. |
| 737 | */ |
| 738 | if (unlikely(!p)) |
| 739 | return NULL; |
| 740 | |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 741 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 742 | } |
Michal Hocko | 33398cf | 2015-09-08 15:01:02 -0700 | [diff] [blame] | 743 | EXPORT_SYMBOL(mem_cgroup_from_task); |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 744 | |
Johannes Weiner | df38197 | 2014-04-07 15:37:43 -0700 | [diff] [blame] | 745 | 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] | 746 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 747 | struct mem_cgroup *memcg = NULL; |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 748 | |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 749 | rcu_read_lock(); |
| 750 | do { |
Michal Hocko | 6f6acb0 | 2014-05-22 11:54:19 -0700 | [diff] [blame] | 751 | /* |
| 752 | * Page cache insertions can happen withou an |
| 753 | * actual mm context, e.g. during disk probing |
| 754 | * on boot, loopback IO, acct() writes etc. |
| 755 | */ |
| 756 | if (unlikely(!mm)) |
Johannes Weiner | df38197 | 2014-04-07 15:37:43 -0700 | [diff] [blame] | 757 | memcg = root_mem_cgroup; |
Michal Hocko | 6f6acb0 | 2014-05-22 11:54:19 -0700 | [diff] [blame] | 758 | else { |
| 759 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
| 760 | if (unlikely(!memcg)) |
| 761 | memcg = root_mem_cgroup; |
| 762 | } |
Tejun Heo | ec903c0 | 2014-05-13 12:11:01 -0400 | [diff] [blame] | 763 | } while (!css_tryget_online(&memcg->css)); |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 764 | rcu_read_unlock(); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 765 | return memcg; |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 766 | } |
| 767 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 768 | /** |
| 769 | * mem_cgroup_iter - iterate over memory cgroup hierarchy |
| 770 | * @root: hierarchy root |
| 771 | * @prev: previously returned memcg, NULL on first invocation |
| 772 | * @reclaim: cookie for shared reclaim walks, NULL for full walks |
| 773 | * |
| 774 | * Returns references to children of the hierarchy below @root, or |
| 775 | * @root itself, or %NULL after a full round-trip. |
| 776 | * |
| 777 | * Caller must pass the return value in @prev on subsequent |
| 778 | * invocations for reference counting, or use mem_cgroup_iter_break() |
| 779 | * to cancel a hierarchy walk before the round-trip is complete. |
| 780 | * |
| 781 | * Reclaimers can specify a zone and a priority level in @reclaim to |
| 782 | * divide up the memcgs in the hierarchy among all concurrent |
| 783 | * reclaimers operating on the same zone and priority. |
| 784 | */ |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 785 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 786 | struct mem_cgroup *prev, |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 787 | struct mem_cgroup_reclaim_cookie *reclaim) |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 788 | { |
Michal Hocko | 33398cf | 2015-09-08 15:01:02 -0700 | [diff] [blame] | 789 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 790 | struct cgroup_subsys_state *css = NULL; |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 791 | struct mem_cgroup *memcg = NULL; |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 792 | struct mem_cgroup *pos = NULL; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 793 | |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 794 | if (mem_cgroup_disabled()) |
| 795 | return NULL; |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 796 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 797 | if (!root) |
| 798 | root = root_mem_cgroup; |
| 799 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 800 | if (prev && !reclaim) |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 801 | pos = prev; |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 802 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 803 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
| 804 | if (prev) |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 805 | goto out; |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 806 | return root; |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 807 | } |
| 808 | |
Michal Hocko | 542f85f | 2013-04-29 15:07:15 -0700 | [diff] [blame] | 809 | rcu_read_lock(); |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 810 | |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 811 | if (reclaim) { |
| 812 | struct mem_cgroup_per_zone *mz; |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 813 | |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 814 | mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone); |
| 815 | iter = &mz->iter[reclaim->priority]; |
Michal Hocko | 5f57816 | 2013-04-29 15:07:17 -0700 | [diff] [blame] | 816 | |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 817 | if (prev && reclaim->generation != iter->generation) |
Michal Hocko | 542f85f | 2013-04-29 15:07:15 -0700 | [diff] [blame] | 818 | goto out_unlock; |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 819 | |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 820 | while (1) { |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 821 | pos = READ_ONCE(iter->position); |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 822 | if (!pos || css_tryget(&pos->css)) |
| 823 | break; |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 824 | /* |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 825 | * css reference reached zero, so iter->position will |
| 826 | * be cleared by ->css_released. However, we should not |
| 827 | * rely on this happening soon, because ->css_released |
| 828 | * is called from a work queue, and by busy-waiting we |
| 829 | * might block it. So we clear iter->position right |
| 830 | * away. |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 831 | */ |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 832 | (void)cmpxchg(&iter->position, pos, NULL); |
| 833 | } |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 834 | } |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 835 | |
| 836 | if (pos) |
| 837 | css = &pos->css; |
| 838 | |
| 839 | for (;;) { |
| 840 | css = css_next_descendant_pre(css, &root->css); |
| 841 | if (!css) { |
| 842 | /* |
| 843 | * Reclaimers share the hierarchy walk, and a |
| 844 | * new one might jump in right at the end of |
| 845 | * the hierarchy - make sure they see at least |
| 846 | * one group and restart from the beginning. |
| 847 | */ |
| 848 | if (!prev) |
| 849 | continue; |
| 850 | break; |
| 851 | } |
| 852 | |
| 853 | /* |
| 854 | * Verify the css and acquire a reference. The root |
| 855 | * is provided by the caller, so we know it's alive |
| 856 | * and kicking, and don't take an extra reference. |
| 857 | */ |
| 858 | memcg = mem_cgroup_from_css(css); |
| 859 | |
| 860 | if (css == &root->css) |
| 861 | break; |
| 862 | |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 863 | if (css_tryget(css)) |
| 864 | break; |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 865 | |
| 866 | memcg = NULL; |
| 867 | } |
| 868 | |
| 869 | if (reclaim) { |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 870 | /* |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 871 | * The position could have already been updated by a competing |
| 872 | * thread, so check that the value hasn't changed since we read |
| 873 | * it to avoid reclaiming from the same cgroup twice. |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 874 | */ |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 875 | (void)cmpxchg(&iter->position, pos, memcg); |
| 876 | |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 877 | if (pos) |
| 878 | css_put(&pos->css); |
| 879 | |
| 880 | if (!memcg) |
| 881 | iter->generation++; |
| 882 | else if (!prev) |
| 883 | reclaim->generation = iter->generation; |
| 884 | } |
| 885 | |
Michal Hocko | 542f85f | 2013-04-29 15:07:15 -0700 | [diff] [blame] | 886 | out_unlock: |
| 887 | rcu_read_unlock(); |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 888 | out: |
Michal Hocko | c40046f | 2013-04-29 15:07:14 -0700 | [diff] [blame] | 889 | if (prev && prev != root) |
| 890 | css_put(&prev->css); |
| 891 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 892 | return memcg; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 893 | } |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 894 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 895 | /** |
| 896 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely |
| 897 | * @root: hierarchy root |
| 898 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() |
| 899 | */ |
| 900 | void mem_cgroup_iter_break(struct mem_cgroup *root, |
| 901 | struct mem_cgroup *prev) |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 902 | { |
| 903 | if (!root) |
| 904 | root = root_mem_cgroup; |
| 905 | if (prev && prev != root) |
| 906 | css_put(&prev->css); |
| 907 | } |
| 908 | |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 909 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
| 910 | { |
| 911 | struct mem_cgroup *memcg = dead_memcg; |
| 912 | struct mem_cgroup_reclaim_iter *iter; |
| 913 | struct mem_cgroup_per_zone *mz; |
| 914 | int nid, zid; |
| 915 | int i; |
| 916 | |
| 917 | while ((memcg = parent_mem_cgroup(memcg))) { |
| 918 | for_each_node(nid) { |
| 919 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
| 920 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
| 921 | for (i = 0; i <= DEF_PRIORITY; i++) { |
| 922 | iter = &mz->iter[i]; |
| 923 | cmpxchg(&iter->position, |
| 924 | dead_memcg, NULL); |
| 925 | } |
| 926 | } |
| 927 | } |
| 928 | } |
| 929 | } |
| 930 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 931 | /* |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 932 | * Iteration constructs for visiting all cgroups (under a tree). If |
| 933 | * loops are exited prematurely (break), mem_cgroup_iter_break() must |
| 934 | * be used for reference counting. |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 935 | */ |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 936 | #define for_each_mem_cgroup_tree(iter, root) \ |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 937 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 938 | iter != NULL; \ |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 939 | iter = mem_cgroup_iter(root, iter, NULL)) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 940 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 941 | #define for_each_mem_cgroup(iter) \ |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 942 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 943 | iter != NULL; \ |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 944 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 945 | |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 946 | /** |
| 947 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg |
| 948 | * @zone: zone of the wanted lruvec |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 949 | * @memcg: memcg of the wanted lruvec |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 950 | * |
| 951 | * Returns the lru list vector holding pages for the given @zone and |
| 952 | * @mem. This can be the global zone lruvec, if the memory controller |
| 953 | * is disabled. |
| 954 | */ |
| 955 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, |
| 956 | struct mem_cgroup *memcg) |
| 957 | { |
| 958 | struct mem_cgroup_per_zone *mz; |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 959 | struct lruvec *lruvec; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 960 | |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 961 | if (mem_cgroup_disabled()) { |
| 962 | lruvec = &zone->lruvec; |
| 963 | goto out; |
| 964 | } |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 965 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 966 | mz = mem_cgroup_zone_zoneinfo(memcg, zone); |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 967 | lruvec = &mz->lruvec; |
| 968 | out: |
| 969 | /* |
| 970 | * Since a node can be onlined after the mem_cgroup was created, |
| 971 | * we have to be prepared to initialize lruvec->zone here; |
| 972 | * and if offlined then reonlined, we need to reinitialize it. |
| 973 | */ |
| 974 | if (unlikely(lruvec->zone != zone)) |
| 975 | lruvec->zone = zone; |
| 976 | return lruvec; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 977 | } |
| 978 | |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 979 | /** |
Johannes Weiner | dfe0e77 | 2014-12-10 15:43:43 -0800 | [diff] [blame] | 980 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 981 | * @page: the page |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 982 | * @zone: zone of the page |
Johannes Weiner | dfe0e77 | 2014-12-10 15:43:43 -0800 | [diff] [blame] | 983 | * |
| 984 | * This function is only safe when following the LRU page isolation |
| 985 | * and putback protocol: the LRU lock must be held, and the page must |
| 986 | * either be PageLRU() or the caller must have isolated/allocated it. |
Minchan Kim | 3f58a82 | 2011-03-22 16:32:53 -0700 | [diff] [blame] | 987 | */ |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 988 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
Minchan Kim | 3f58a82 | 2011-03-22 16:32:53 -0700 | [diff] [blame] | 989 | { |
| 990 | struct mem_cgroup_per_zone *mz; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 991 | struct mem_cgroup *memcg; |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 992 | struct lruvec *lruvec; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 993 | |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 994 | if (mem_cgroup_disabled()) { |
| 995 | lruvec = &zone->lruvec; |
| 996 | goto out; |
| 997 | } |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 998 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 999 | memcg = page->mem_cgroup; |
Hugh Dickins | 7512102 | 2012-03-05 14:59:18 -0800 | [diff] [blame] | 1000 | /* |
Johannes Weiner | dfe0e77 | 2014-12-10 15:43:43 -0800 | [diff] [blame] | 1001 | * Swapcache readahead pages are added to the LRU - and |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 1002 | * possibly migrated - before they are charged. |
Hugh Dickins | 7512102 | 2012-03-05 14:59:18 -0800 | [diff] [blame] | 1003 | */ |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 1004 | if (!memcg) |
| 1005 | memcg = root_mem_cgroup; |
Hugh Dickins | 7512102 | 2012-03-05 14:59:18 -0800 | [diff] [blame] | 1006 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 1007 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 1008 | lruvec = &mz->lruvec; |
| 1009 | out: |
| 1010 | /* |
| 1011 | * Since a node can be onlined after the mem_cgroup was created, |
| 1012 | * we have to be prepared to initialize lruvec->zone here; |
| 1013 | * and if offlined then reonlined, we need to reinitialize it. |
| 1014 | */ |
| 1015 | if (unlikely(lruvec->zone != zone)) |
| 1016 | lruvec->zone = zone; |
| 1017 | return lruvec; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1018 | } |
| 1019 | |
| 1020 | /** |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1021 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
| 1022 | * @lruvec: mem_cgroup per zone lru vector |
| 1023 | * @lru: index of lru list the page is sitting on |
| 1024 | * @nr_pages: positive when adding or negative when removing |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1025 | * |
Hugh Dickins | ca70723 | 2016-05-19 17:12:35 -0700 | [diff] [blame] | 1026 | * This function must be called under lru_lock, just before a page is added |
| 1027 | * to or just after a page is removed from an lru list (that ordering being |
| 1028 | * so as to allow it to check that lru_size 0 is consistent with list_empty). |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1029 | */ |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1030 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
| 1031 | int nr_pages) |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1032 | { |
| 1033 | struct mem_cgroup_per_zone *mz; |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1034 | unsigned long *lru_size; |
Hugh Dickins | ca70723 | 2016-05-19 17:12:35 -0700 | [diff] [blame] | 1035 | long size; |
| 1036 | bool empty; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1037 | |
Hugh Dickins | 9d5e6a9 | 2016-05-19 17:12:38 -0700 | [diff] [blame] | 1038 | __update_lru_size(lruvec, lru, nr_pages); |
| 1039 | |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1040 | if (mem_cgroup_disabled()) |
| 1041 | return; |
| 1042 | |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1043 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
| 1044 | lru_size = mz->lru_size + lru; |
Hugh Dickins | ca70723 | 2016-05-19 17:12:35 -0700 | [diff] [blame] | 1045 | empty = list_empty(lruvec->lists + lru); |
| 1046 | |
| 1047 | if (nr_pages < 0) |
| 1048 | *lru_size += nr_pages; |
| 1049 | |
| 1050 | size = *lru_size; |
| 1051 | if (WARN_ONCE(size < 0 || empty != !size, |
| 1052 | "%s(%p, %d, %d): lru_size %ld but %sempty\n", |
| 1053 | __func__, lruvec, lru, nr_pages, size, empty ? "" : "not ")) { |
| 1054 | VM_BUG_ON(1); |
| 1055 | *lru_size = 0; |
| 1056 | } |
| 1057 | |
| 1058 | if (nr_pages > 0) |
| 1059 | *lru_size += nr_pages; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 1060 | } |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 1061 | |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1062 | 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] | 1063 | { |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1064 | struct mem_cgroup *task_memcg; |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 1065 | struct task_struct *p; |
David Rientjes | ffbdccf | 2013-07-03 15:01:23 -0700 | [diff] [blame] | 1066 | bool ret; |
David Rientjes | 4c4a221 | 2008-02-07 00:14:06 -0800 | [diff] [blame] | 1067 | |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 1068 | p = find_lock_task_mm(task); |
David Rientjes | de077d2 | 2012-01-12 17:18:52 -0800 | [diff] [blame] | 1069 | if (p) { |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1070 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
David Rientjes | de077d2 | 2012-01-12 17:18:52 -0800 | [diff] [blame] | 1071 | task_unlock(p); |
| 1072 | } else { |
| 1073 | /* |
| 1074 | * All threads may have already detached their mm's, but the oom |
| 1075 | * killer still needs to detect if they have already been oom |
| 1076 | * killed to prevent needlessly killing additional tasks. |
| 1077 | */ |
David Rientjes | ffbdccf | 2013-07-03 15:01:23 -0700 | [diff] [blame] | 1078 | rcu_read_lock(); |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1079 | task_memcg = mem_cgroup_from_task(task); |
| 1080 | css_get(&task_memcg->css); |
David Rientjes | ffbdccf | 2013-07-03 15:01:23 -0700 | [diff] [blame] | 1081 | rcu_read_unlock(); |
David Rientjes | de077d2 | 2012-01-12 17:18:52 -0800 | [diff] [blame] | 1082 | } |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1083 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
| 1084 | css_put(&task_memcg->css); |
David Rientjes | 4c4a221 | 2008-02-07 00:14:06 -0800 | [diff] [blame] | 1085 | return ret; |
| 1086 | } |
| 1087 | |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1088 | /** |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1089 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
Wanpeng Li | dad7557 | 2012-06-20 12:53:01 -0700 | [diff] [blame] | 1090 | * @memcg: the memory cgroup |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1091 | * |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1092 | * Returns the maximum amount of memory @mem can be charged with, in |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 1093 | * pages. |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1094 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1095 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1096 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1097 | unsigned long margin = 0; |
| 1098 | unsigned long count; |
| 1099 | unsigned long limit; |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1100 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1101 | count = page_counter_read(&memcg->memory); |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 1102 | limit = READ_ONCE(memcg->memory.limit); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1103 | if (count < limit) |
| 1104 | margin = limit - count; |
| 1105 | |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 1106 | if (do_memsw_account()) { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1107 | count = page_counter_read(&memcg->memsw); |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 1108 | limit = READ_ONCE(memcg->memsw.limit); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1109 | if (count <= limit) |
| 1110 | margin = min(margin, limit - count); |
Li RongQing | cbedbac | 2016-05-27 14:27:43 -0700 | [diff] [blame] | 1111 | else |
| 1112 | margin = 0; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1113 | } |
| 1114 | |
| 1115 | return margin; |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1116 | } |
| 1117 | |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 1118 | /* |
Qiang Huang | bdcbb65 | 2014-06-04 16:08:21 -0700 | [diff] [blame] | 1119 | * A routine for checking "mem" is under move_account() or not. |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1120 | * |
Qiang Huang | bdcbb65 | 2014-06-04 16:08:21 -0700 | [diff] [blame] | 1121 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
| 1122 | * moving cgroups. This is for waiting at high-memory pressure |
| 1123 | * caused by "move". |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1124 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1125 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1126 | { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1127 | struct mem_cgroup *from; |
| 1128 | struct mem_cgroup *to; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1129 | bool ret = false; |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1130 | /* |
| 1131 | * Unlike task_move routines, we access mc.to, mc.from not under |
| 1132 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. |
| 1133 | */ |
| 1134 | spin_lock(&mc.lock); |
| 1135 | from = mc.from; |
| 1136 | to = mc.to; |
| 1137 | if (!from) |
| 1138 | goto unlock; |
Michal Hocko | 3e92041 | 2011-07-26 16:08:29 -0700 | [diff] [blame] | 1139 | |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1140 | ret = mem_cgroup_is_descendant(from, memcg) || |
| 1141 | mem_cgroup_is_descendant(to, memcg); |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1142 | unlock: |
| 1143 | spin_unlock(&mc.lock); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1144 | return ret; |
| 1145 | } |
| 1146 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1147 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1148 | { |
| 1149 | if (mc.moving_task && current != mc.moving_task) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1150 | if (mem_cgroup_under_move(memcg)) { |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1151 | DEFINE_WAIT(wait); |
| 1152 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); |
| 1153 | /* moving charge context might have finished. */ |
| 1154 | if (mc.moving_task) |
| 1155 | schedule(); |
| 1156 | finish_wait(&mc.waitq, &wait); |
| 1157 | return true; |
| 1158 | } |
| 1159 | } |
| 1160 | return false; |
| 1161 | } |
| 1162 | |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1163 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1164 | /** |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1165 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1166 | * @memcg: The memory cgroup that went over limit |
| 1167 | * @p: Task that is going to be killed |
| 1168 | * |
| 1169 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is |
| 1170 | * enabled |
| 1171 | */ |
| 1172 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) |
| 1173 | { |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1174 | struct mem_cgroup *iter; |
| 1175 | unsigned int i; |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1176 | |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1177 | rcu_read_lock(); |
| 1178 | |
Balasubramani Vivekanandan | 2415b9f | 2015-04-14 15:48:18 -0700 | [diff] [blame] | 1179 | if (p) { |
| 1180 | pr_info("Task in "); |
| 1181 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
| 1182 | pr_cont(" killed as a result of limit of "); |
| 1183 | } else { |
| 1184 | pr_info("Memory limit reached of cgroup "); |
| 1185 | } |
| 1186 | |
Tejun Heo | e61734c | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 1187 | pr_cont_cgroup_path(memcg->css.cgroup); |
Greg Thelen | 0346dad | 2015-01-26 12:58:38 -0800 | [diff] [blame] | 1188 | pr_cont("\n"); |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1189 | |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1190 | rcu_read_unlock(); |
| 1191 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1192 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
| 1193 | K((u64)page_counter_read(&memcg->memory)), |
| 1194 | K((u64)memcg->memory.limit), memcg->memory.failcnt); |
| 1195 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", |
| 1196 | K((u64)page_counter_read(&memcg->memsw)), |
| 1197 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); |
| 1198 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", |
| 1199 | K((u64)page_counter_read(&memcg->kmem)), |
| 1200 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1201 | |
| 1202 | for_each_mem_cgroup_tree(iter, memcg) { |
Tejun Heo | e61734c | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 1203 | pr_info("Memory cgroup stats for "); |
| 1204 | pr_cont_cgroup_path(iter->css.cgroup); |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1205 | pr_cont(":"); |
| 1206 | |
| 1207 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 1208 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1209 | continue; |
Greg Thelen | 484ebb3 | 2015-10-01 15:37:05 -0700 | [diff] [blame] | 1210 | pr_cont(" %s:%luKB", mem_cgroup_stat_names[i], |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1211 | K(mem_cgroup_read_stat(iter, i))); |
| 1212 | } |
| 1213 | |
| 1214 | for (i = 0; i < NR_LRU_LISTS; i++) |
| 1215 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], |
| 1216 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); |
| 1217 | |
| 1218 | pr_cont("\n"); |
| 1219 | } |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1220 | } |
| 1221 | |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1222 | /* |
| 1223 | * This function returns the number of memcg under hierarchy tree. Returns |
| 1224 | * 1(self count) if no children. |
| 1225 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1226 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1227 | { |
| 1228 | int num = 0; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1229 | struct mem_cgroup *iter; |
| 1230 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1231 | for_each_mem_cgroup_tree(iter, memcg) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1232 | num++; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1233 | return num; |
| 1234 | } |
| 1235 | |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1236 | /* |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1237 | * Return the memory (and swap, if configured) limit for a memcg. |
| 1238 | */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1239 | static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1240 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1241 | unsigned long limit; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1242 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1243 | limit = memcg->memory.limit; |
Michal Hocko | 9a5a8f1 | 2012-11-16 14:14:49 -0800 | [diff] [blame] | 1244 | if (mem_cgroup_swappiness(memcg)) { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1245 | unsigned long memsw_limit; |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 1246 | unsigned long swap_limit; |
Michal Hocko | 9a5a8f1 | 2012-11-16 14:14:49 -0800 | [diff] [blame] | 1247 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1248 | memsw_limit = memcg->memsw.limit; |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 1249 | swap_limit = memcg->swap.limit; |
| 1250 | swap_limit = min(swap_limit, (unsigned long)total_swap_pages); |
| 1251 | limit = min(limit + swap_limit, memsw_limit); |
Michal Hocko | 9a5a8f1 | 2012-11-16 14:14:49 -0800 | [diff] [blame] | 1252 | } |
Michal Hocko | 9a5a8f1 | 2012-11-16 14:14:49 -0800 | [diff] [blame] | 1253 | return limit; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1254 | } |
| 1255 | |
Johannes Weiner | b6e6edc | 2016-03-17 14:20:28 -0700 | [diff] [blame] | 1256 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
David Rientjes | 1996546 | 2012-12-11 16:00:26 -0800 | [diff] [blame] | 1257 | int order) |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1258 | { |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 1259 | struct oom_control oc = { |
| 1260 | .zonelist = NULL, |
| 1261 | .nodemask = NULL, |
| 1262 | .gfp_mask = gfp_mask, |
| 1263 | .order = order, |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 1264 | }; |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1265 | struct mem_cgroup *iter; |
| 1266 | unsigned long chosen_points = 0; |
| 1267 | unsigned long totalpages; |
| 1268 | unsigned int points = 0; |
| 1269 | struct task_struct *chosen = NULL; |
| 1270 | |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 1271 | mutex_lock(&oom_lock); |
| 1272 | |
David Rientjes | 876aafb | 2012-07-31 16:43:48 -0700 | [diff] [blame] | 1273 | /* |
David Rientjes | 465adcf | 2013-04-29 15:08:45 -0700 | [diff] [blame] | 1274 | * If current has a pending SIGKILL or is exiting, then automatically |
| 1275 | * select it. The goal is to allow it to allocate so that it may |
| 1276 | * quickly exit and free its memory. |
David Rientjes | 876aafb | 2012-07-31 16:43:48 -0700 | [diff] [blame] | 1277 | */ |
Oleg Nesterov | d003f37 | 2014-12-12 16:56:24 -0800 | [diff] [blame] | 1278 | if (fatal_signal_pending(current) || task_will_free_mem(current)) { |
Johannes Weiner | 16e9519 | 2015-06-24 16:57:07 -0700 | [diff] [blame] | 1279 | mark_oom_victim(current); |
Michal Hocko | 3ef22df | 2016-05-19 17:13:12 -0700 | [diff] [blame] | 1280 | try_oom_reaper(current); |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 1281 | goto unlock; |
David Rientjes | 876aafb | 2012-07-31 16:43:48 -0700 | [diff] [blame] | 1282 | } |
| 1283 | |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 1284 | check_panic_on_oom(&oc, CONSTRAINT_MEMCG, memcg); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1285 | totalpages = mem_cgroup_get_limit(memcg) ? : 1; |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1286 | for_each_mem_cgroup_tree(iter, memcg) { |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 1287 | struct css_task_iter it; |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1288 | struct task_struct *task; |
| 1289 | |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 1290 | css_task_iter_start(&iter->css, &it); |
| 1291 | while ((task = css_task_iter_next(&it))) { |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 1292 | switch (oom_scan_process_thread(&oc, task, totalpages)) { |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1293 | case OOM_SCAN_SELECT: |
| 1294 | if (chosen) |
| 1295 | put_task_struct(chosen); |
| 1296 | chosen = task; |
| 1297 | chosen_points = ULONG_MAX; |
| 1298 | get_task_struct(chosen); |
| 1299 | /* fall through */ |
| 1300 | case OOM_SCAN_CONTINUE: |
| 1301 | continue; |
| 1302 | case OOM_SCAN_ABORT: |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 1303 | css_task_iter_end(&it); |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1304 | mem_cgroup_iter_break(memcg, iter); |
| 1305 | if (chosen) |
| 1306 | put_task_struct(chosen); |
Tetsuo Handa | 1ebab2d | 2016-05-26 15:16:19 -0700 | [diff] [blame] | 1307 | /* Set a dummy value to return "true". */ |
| 1308 | chosen = (void *) 1; |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 1309 | goto unlock; |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1310 | case OOM_SCAN_OK: |
| 1311 | break; |
| 1312 | }; |
| 1313 | points = oom_badness(task, memcg, NULL, totalpages); |
David Rientjes | d49ad93 | 2014-01-23 15:53:34 -0800 | [diff] [blame] | 1314 | if (!points || points < chosen_points) |
| 1315 | continue; |
| 1316 | /* Prefer thread group leaders for display purposes */ |
| 1317 | if (points == chosen_points && |
| 1318 | thread_group_leader(chosen)) |
| 1319 | continue; |
| 1320 | |
| 1321 | if (chosen) |
| 1322 | put_task_struct(chosen); |
| 1323 | chosen = task; |
| 1324 | chosen_points = points; |
| 1325 | get_task_struct(chosen); |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1326 | } |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 1327 | css_task_iter_end(&it); |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1328 | } |
| 1329 | |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 1330 | if (chosen) { |
| 1331 | points = chosen_points * 1000 / totalpages; |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 1332 | oom_kill_process(&oc, chosen, points, totalpages, memcg, |
| 1333 | "Memory cgroup out of memory"); |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 1334 | } |
| 1335 | unlock: |
| 1336 | mutex_unlock(&oom_lock); |
Johannes Weiner | b6e6edc | 2016-03-17 14:20:28 -0700 | [diff] [blame] | 1337 | return chosen; |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1338 | } |
| 1339 | |
Michele Curti | ae6e71d | 2014-12-12 16:56:35 -0800 | [diff] [blame] | 1340 | #if MAX_NUMNODES > 1 |
| 1341 | |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1342 | /** |
| 1343 | * test_mem_cgroup_node_reclaimable |
Wanpeng Li | dad7557 | 2012-06-20 12:53:01 -0700 | [diff] [blame] | 1344 | * @memcg: the target memcg |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1345 | * @nid: the node ID to be checked. |
| 1346 | * @noswap : specify true here if the user wants flle only information. |
| 1347 | * |
| 1348 | * This function returns whether the specified memcg contains any |
| 1349 | * reclaimable pages on a node. Returns true if there are any reclaimable |
| 1350 | * pages in the node. |
| 1351 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1352 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1353 | int nid, bool noswap) |
| 1354 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1355 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1356 | return true; |
| 1357 | if (noswap || !total_swap_pages) |
| 1358 | return false; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1359 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1360 | return true; |
| 1361 | return false; |
| 1362 | |
| 1363 | } |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1364 | |
| 1365 | /* |
| 1366 | * Always updating the nodemask is not very good - even if we have an empty |
| 1367 | * list or the wrong list here, we can start from some node and traverse all |
| 1368 | * nodes based on the zonelist. So update the list loosely once per 10 secs. |
| 1369 | * |
| 1370 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1371 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1372 | { |
| 1373 | int nid; |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 1374 | /* |
| 1375 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET |
| 1376 | * pagein/pageout changes since the last update. |
| 1377 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1378 | if (!atomic_read(&memcg->numainfo_events)) |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 1379 | return; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1380 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1381 | return; |
| 1382 | |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1383 | /* make a nodemask where this memcg uses memory from */ |
Lai Jiangshan | 31aaea4 | 2012-12-12 13:51:27 -0800 | [diff] [blame] | 1384 | memcg->scan_nodes = node_states[N_MEMORY]; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1385 | |
Lai Jiangshan | 31aaea4 | 2012-12-12 13:51:27 -0800 | [diff] [blame] | 1386 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1387 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1388 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
| 1389 | node_clear(nid, memcg->scan_nodes); |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1390 | } |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 1391 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1392 | atomic_set(&memcg->numainfo_events, 0); |
| 1393 | atomic_set(&memcg->numainfo_updating, 0); |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1394 | } |
| 1395 | |
| 1396 | /* |
| 1397 | * Selecting a node where we start reclaim from. Because what we need is just |
| 1398 | * reducing usage counter, start from anywhere is O,K. Considering |
| 1399 | * memory reclaim from current node, there are pros. and cons. |
| 1400 | * |
| 1401 | * Freeing memory from current node means freeing memory from a node which |
| 1402 | * we'll use or we've used. So, it may make LRU bad. And if several threads |
| 1403 | * hit limits, it will see a contention on a node. But freeing from remote |
| 1404 | * node means more costs for memory reclaim because of memory latency. |
| 1405 | * |
| 1406 | * Now, we use round-robin. Better algorithm is welcomed. |
| 1407 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1408 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1409 | { |
| 1410 | int node; |
| 1411 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1412 | mem_cgroup_may_update_nodemask(memcg); |
| 1413 | node = memcg->last_scanned_node; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1414 | |
Andrew Morton | 0edaf86 | 2016-05-19 17:10:58 -0700 | [diff] [blame] | 1415 | node = next_node_in(node, memcg->scan_nodes); |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1416 | /* |
Michal Hocko | fda3d69 | 2016-05-19 17:11:34 -0700 | [diff] [blame] | 1417 | * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages |
| 1418 | * last time it really checked all the LRUs due to rate limiting. |
| 1419 | * Fallback to the current node in that case for simplicity. |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1420 | */ |
| 1421 | if (unlikely(node == MAX_NUMNODES)) |
| 1422 | node = numa_node_id(); |
| 1423 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1424 | memcg->last_scanned_node = node; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1425 | return node; |
| 1426 | } |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1427 | #else |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1428 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1429 | { |
| 1430 | return 0; |
| 1431 | } |
| 1432 | #endif |
| 1433 | |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1434 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
| 1435 | struct zone *zone, |
| 1436 | gfp_t gfp_mask, |
| 1437 | unsigned long *total_scanned) |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1438 | { |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1439 | struct mem_cgroup *victim = NULL; |
| 1440 | int total = 0; |
| 1441 | int loop = 0; |
| 1442 | unsigned long excess; |
| 1443 | unsigned long nr_scanned; |
| 1444 | struct mem_cgroup_reclaim_cookie reclaim = { |
| 1445 | .zone = zone, |
| 1446 | .priority = 0, |
| 1447 | }; |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1448 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1449 | excess = soft_limit_excess(root_memcg); |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1450 | |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1451 | while (1) { |
| 1452 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); |
| 1453 | if (!victim) { |
| 1454 | loop++; |
| 1455 | if (loop >= 2) { |
| 1456 | /* |
| 1457 | * If we have not been able to reclaim |
| 1458 | * anything, it might because there are |
| 1459 | * no reclaimable pages under this hierarchy |
| 1460 | */ |
| 1461 | if (!total) |
| 1462 | break; |
| 1463 | /* |
| 1464 | * We want to do more targeted reclaim. |
| 1465 | * excess >> 2 is not to excessive so as to |
| 1466 | * reclaim too much, nor too less that we keep |
| 1467 | * coming back to reclaim from this cgroup |
| 1468 | */ |
| 1469 | if (total >= (excess >> 2) || |
| 1470 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) |
| 1471 | break; |
| 1472 | } |
| 1473 | continue; |
| 1474 | } |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1475 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
| 1476 | zone, &nr_scanned); |
| 1477 | *total_scanned += nr_scanned; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1478 | if (!soft_limit_excess(root_memcg)) |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1479 | break; |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1480 | } |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1481 | mem_cgroup_iter_break(root_memcg, victim); |
| 1482 | return total; |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1483 | } |
| 1484 | |
Johannes Weiner | 0056f4e | 2013-10-31 16:34:14 -0700 | [diff] [blame] | 1485 | #ifdef CONFIG_LOCKDEP |
| 1486 | static struct lockdep_map memcg_oom_lock_dep_map = { |
| 1487 | .name = "memcg_oom_lock", |
| 1488 | }; |
| 1489 | #endif |
| 1490 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1491 | static DEFINE_SPINLOCK(memcg_oom_lock); |
| 1492 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1493 | /* |
| 1494 | * Check OOM-Killer is already running under our hierarchy. |
| 1495 | * If someone is running, return false. |
| 1496 | */ |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1497 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1498 | { |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1499 | struct mem_cgroup *iter, *failed = NULL; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 1500 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1501 | spin_lock(&memcg_oom_lock); |
| 1502 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1503 | for_each_mem_cgroup_tree(iter, memcg) { |
Johannes Weiner | 23751be | 2011-08-25 15:59:16 -0700 | [diff] [blame] | 1504 | if (iter->oom_lock) { |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1505 | /* |
| 1506 | * this subtree of our hierarchy is already locked |
| 1507 | * so we cannot give a lock. |
| 1508 | */ |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1509 | failed = iter; |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1510 | mem_cgroup_iter_break(memcg, iter); |
| 1511 | break; |
Johannes Weiner | 23751be | 2011-08-25 15:59:16 -0700 | [diff] [blame] | 1512 | } else |
| 1513 | iter->oom_lock = true; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1514 | } |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1515 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1516 | if (failed) { |
| 1517 | /* |
| 1518 | * OK, we failed to lock the whole subtree so we have |
| 1519 | * to clean up what we set up to the failing subtree |
| 1520 | */ |
| 1521 | for_each_mem_cgroup_tree(iter, memcg) { |
| 1522 | if (iter == failed) { |
| 1523 | mem_cgroup_iter_break(memcg, iter); |
| 1524 | break; |
| 1525 | } |
| 1526 | iter->oom_lock = false; |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1527 | } |
Johannes Weiner | 0056f4e | 2013-10-31 16:34:14 -0700 | [diff] [blame] | 1528 | } else |
| 1529 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1530 | |
| 1531 | spin_unlock(&memcg_oom_lock); |
| 1532 | |
| 1533 | return !failed; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 1534 | } |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1535 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1536 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1537 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1538 | struct mem_cgroup *iter; |
| 1539 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1540 | spin_lock(&memcg_oom_lock); |
Johannes Weiner | 0056f4e | 2013-10-31 16:34:14 -0700 | [diff] [blame] | 1541 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1542 | for_each_mem_cgroup_tree(iter, memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1543 | iter->oom_lock = false; |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1544 | spin_unlock(&memcg_oom_lock); |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1545 | } |
| 1546 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1547 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1548 | { |
| 1549 | struct mem_cgroup *iter; |
| 1550 | |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 1551 | spin_lock(&memcg_oom_lock); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1552 | for_each_mem_cgroup_tree(iter, memcg) |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 1553 | iter->under_oom++; |
| 1554 | spin_unlock(&memcg_oom_lock); |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1555 | } |
| 1556 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1557 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1558 | { |
| 1559 | struct mem_cgroup *iter; |
| 1560 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1561 | /* |
| 1562 | * When a new child is created while the hierarchy is under oom, |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 1563 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1564 | */ |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 1565 | spin_lock(&memcg_oom_lock); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1566 | for_each_mem_cgroup_tree(iter, memcg) |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 1567 | if (iter->under_oom > 0) |
| 1568 | iter->under_oom--; |
| 1569 | spin_unlock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1570 | } |
| 1571 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1572 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
| 1573 | |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1574 | struct oom_wait_info { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 1575 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1576 | wait_queue_t wait; |
| 1577 | }; |
| 1578 | |
| 1579 | static int memcg_oom_wake_function(wait_queue_t *wait, |
| 1580 | unsigned mode, int sync, void *arg) |
| 1581 | { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 1582 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
| 1583 | struct mem_cgroup *oom_wait_memcg; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1584 | struct oom_wait_info *oom_wait_info; |
| 1585 | |
| 1586 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 1587 | oom_wait_memcg = oom_wait_info->memcg; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1588 | |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1589 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
| 1590 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1591 | return 0; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1592 | return autoremove_wake_function(wait, mode, sync, arg); |
| 1593 | } |
| 1594 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1595 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1596 | { |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 1597 | /* |
| 1598 | * For the following lockless ->under_oom test, the only required |
| 1599 | * guarantee is that it must see the state asserted by an OOM when |
| 1600 | * this function is called as a result of userland actions |
| 1601 | * triggered by the notification of the OOM. This is trivially |
| 1602 | * achieved by invoking mem_cgroup_mark_under_oom() before |
| 1603 | * triggering notification. |
| 1604 | */ |
| 1605 | if (memcg && memcg->under_oom) |
Tejun Heo | f4b90b70 | 2015-06-24 16:58:21 -0700 | [diff] [blame] | 1606 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1607 | } |
| 1608 | |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 1609 | 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] | 1610 | { |
Andrew Morton | d0db7af | 2016-06-08 15:33:47 -0700 | [diff] [blame] | 1611 | if (!current->memcg_may_oom) |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 1612 | return; |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1613 | /* |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1614 | * We are in the middle of the charge context here, so we |
| 1615 | * don't want to block when potentially sitting on a callstack |
| 1616 | * that holds all kinds of filesystem and mm locks. |
| 1617 | * |
| 1618 | * Also, the caller may handle a failed allocation gracefully |
| 1619 | * (like optional page cache readahead) and so an OOM killer |
| 1620 | * invocation might not even be necessary. |
| 1621 | * |
| 1622 | * That's why we don't do anything here except remember the |
| 1623 | * OOM context and then deal with it at the end of the page |
| 1624 | * fault when the stack is unwound, the locks are released, |
| 1625 | * and when we know whether the fault was overall successful. |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1626 | */ |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1627 | css_get(&memcg->css); |
Tejun Heo | 626ebc4 | 2015-11-05 18:46:09 -0800 | [diff] [blame] | 1628 | current->memcg_in_oom = memcg; |
| 1629 | current->memcg_oom_gfp_mask = mask; |
| 1630 | current->memcg_oom_order = order; |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1631 | } |
| 1632 | |
| 1633 | /** |
| 1634 | * mem_cgroup_oom_synchronize - complete memcg OOM handling |
| 1635 | * @handle: actually kill/wait or just clean up the OOM state |
| 1636 | * |
| 1637 | * This has to be called at the end of a page fault if the memcg OOM |
| 1638 | * handler was enabled. |
| 1639 | * |
| 1640 | * Memcg supports userspace OOM handling where failed allocations must |
| 1641 | * sleep on a waitqueue until the userspace task resolves the |
| 1642 | * situation. Sleeping directly in the charge context with all kinds |
| 1643 | * of locks held is not a good idea, instead we remember an OOM state |
| 1644 | * in the task and mem_cgroup_oom_synchronize() has to be called at |
| 1645 | * the end of the page fault to complete the OOM handling. |
| 1646 | * |
| 1647 | * Returns %true if an ongoing memcg OOM situation was detected and |
| 1648 | * completed, %false otherwise. |
| 1649 | */ |
| 1650 | bool mem_cgroup_oom_synchronize(bool handle) |
| 1651 | { |
Tejun Heo | 626ebc4 | 2015-11-05 18:46:09 -0800 | [diff] [blame] | 1652 | struct mem_cgroup *memcg = current->memcg_in_oom; |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1653 | struct oom_wait_info owait; |
| 1654 | bool locked; |
| 1655 | |
| 1656 | /* OOM is global, do not handle */ |
| 1657 | if (!memcg) |
| 1658 | return false; |
| 1659 | |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 1660 | if (!handle || oom_killer_disabled) |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1661 | goto cleanup; |
| 1662 | |
| 1663 | owait.memcg = memcg; |
| 1664 | owait.wait.flags = 0; |
| 1665 | owait.wait.func = memcg_oom_wake_function; |
| 1666 | owait.wait.private = current; |
| 1667 | INIT_LIST_HEAD(&owait.wait.task_list); |
| 1668 | |
| 1669 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1670 | mem_cgroup_mark_under_oom(memcg); |
| 1671 | |
| 1672 | locked = mem_cgroup_oom_trylock(memcg); |
| 1673 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1674 | if (locked) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1675 | mem_cgroup_oom_notify(memcg); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1676 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1677 | if (locked && !memcg->oom_kill_disable) { |
| 1678 | mem_cgroup_unmark_under_oom(memcg); |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1679 | finish_wait(&memcg_oom_waitq, &owait.wait); |
Tejun Heo | 626ebc4 | 2015-11-05 18:46:09 -0800 | [diff] [blame] | 1680 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
| 1681 | current->memcg_oom_order); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1682 | } else { |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 1683 | schedule(); |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1684 | mem_cgroup_unmark_under_oom(memcg); |
| 1685 | finish_wait(&memcg_oom_waitq, &owait.wait); |
| 1686 | } |
| 1687 | |
| 1688 | if (locked) { |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1689 | mem_cgroup_oom_unlock(memcg); |
| 1690 | /* |
| 1691 | * There is no guarantee that an OOM-lock contender |
| 1692 | * sees the wakeups triggered by the OOM kill |
| 1693 | * uncharges. Wake any sleepers explicitely. |
| 1694 | */ |
| 1695 | memcg_oom_recover(memcg); |
| 1696 | } |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1697 | cleanup: |
Tejun Heo | 626ebc4 | 2015-11-05 18:46:09 -0800 | [diff] [blame] | 1698 | current->memcg_in_oom = NULL; |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 1699 | css_put(&memcg->css); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1700 | return true; |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1701 | } |
| 1702 | |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 1703 | /** |
Johannes Weiner | 81f8c3a | 2016-03-15 14:57:04 -0700 | [diff] [blame] | 1704 | * lock_page_memcg - lock a page->mem_cgroup binding |
| 1705 | * @page: the page |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1706 | * |
Johannes Weiner | 81f8c3a | 2016-03-15 14:57:04 -0700 | [diff] [blame] | 1707 | * This function protects unlocked LRU pages from being moved to |
| 1708 | * another cgroup and stabilizes their page->mem_cgroup binding. |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1709 | */ |
Johannes Weiner | 62cccb8 | 2016-03-15 14:57:22 -0700 | [diff] [blame] | 1710 | void lock_page_memcg(struct page *page) |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1711 | { |
| 1712 | struct mem_cgroup *memcg; |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1713 | unsigned long flags; |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1714 | |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1715 | /* |
| 1716 | * The RCU lock is held throughout the transaction. The fast |
| 1717 | * path can get away without acquiring the memcg->move_lock |
| 1718 | * because page moving starts with an RCU grace period. |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1719 | */ |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 1720 | rcu_read_lock(); |
| 1721 | |
| 1722 | if (mem_cgroup_disabled()) |
Johannes Weiner | 62cccb8 | 2016-03-15 14:57:22 -0700 | [diff] [blame] | 1723 | return; |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1724 | again: |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 1725 | memcg = page->mem_cgroup; |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 1726 | if (unlikely(!memcg)) |
Johannes Weiner | 62cccb8 | 2016-03-15 14:57:22 -0700 | [diff] [blame] | 1727 | return; |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 1728 | |
Qiang Huang | bdcbb65 | 2014-06-04 16:08:21 -0700 | [diff] [blame] | 1729 | if (atomic_read(&memcg->moving_account) <= 0) |
Johannes Weiner | 62cccb8 | 2016-03-15 14:57:22 -0700 | [diff] [blame] | 1730 | return; |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1731 | |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1732 | spin_lock_irqsave(&memcg->move_lock, flags); |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 1733 | if (memcg != page->mem_cgroup) { |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1734 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1735 | goto again; |
| 1736 | } |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1737 | |
| 1738 | /* |
| 1739 | * When charge migration first begins, we can have locked and |
| 1740 | * unlocked page stat updates happening concurrently. Track |
Johannes Weiner | 81f8c3a | 2016-03-15 14:57:04 -0700 | [diff] [blame] | 1741 | * the task who has the lock for unlock_page_memcg(). |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1742 | */ |
| 1743 | memcg->move_lock_task = current; |
| 1744 | memcg->move_lock_flags = flags; |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 1745 | |
Johannes Weiner | 62cccb8 | 2016-03-15 14:57:22 -0700 | [diff] [blame] | 1746 | return; |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1747 | } |
Johannes Weiner | 81f8c3a | 2016-03-15 14:57:04 -0700 | [diff] [blame] | 1748 | EXPORT_SYMBOL(lock_page_memcg); |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1749 | |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 1750 | /** |
Johannes Weiner | 81f8c3a | 2016-03-15 14:57:04 -0700 | [diff] [blame] | 1751 | * unlock_page_memcg - unlock a page->mem_cgroup binding |
Johannes Weiner | 62cccb8 | 2016-03-15 14:57:22 -0700 | [diff] [blame] | 1752 | * @page: the page |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 1753 | */ |
Johannes Weiner | 62cccb8 | 2016-03-15 14:57:22 -0700 | [diff] [blame] | 1754 | void unlock_page_memcg(struct page *page) |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1755 | { |
Johannes Weiner | 62cccb8 | 2016-03-15 14:57:22 -0700 | [diff] [blame] | 1756 | struct mem_cgroup *memcg = page->mem_cgroup; |
| 1757 | |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1758 | if (memcg && memcg->move_lock_task == current) { |
| 1759 | unsigned long flags = memcg->move_lock_flags; |
| 1760 | |
| 1761 | memcg->move_lock_task = NULL; |
| 1762 | memcg->move_lock_flags = 0; |
| 1763 | |
| 1764 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
| 1765 | } |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1766 | |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 1767 | rcu_read_unlock(); |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1768 | } |
Johannes Weiner | 81f8c3a | 2016-03-15 14:57:04 -0700 | [diff] [blame] | 1769 | EXPORT_SYMBOL(unlock_page_memcg); |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1770 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 1771 | /* |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1772 | * size of first charge trial. "32" comes from vmscan.c's magic value. |
| 1773 | * TODO: maybe necessary to use big numbers in big irons. |
| 1774 | */ |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 1775 | #define CHARGE_BATCH 32U |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1776 | struct memcg_stock_pcp { |
| 1777 | struct mem_cgroup *cached; /* this never be root cgroup */ |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 1778 | unsigned int nr_pages; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1779 | struct work_struct work; |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 1780 | unsigned long flags; |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 1781 | #define FLUSHING_CACHED_CHARGE 0 |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1782 | }; |
| 1783 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); |
Michal Hocko | 9f50fad | 2011-08-09 11:56:26 +0200 | [diff] [blame] | 1784 | static DEFINE_MUTEX(percpu_charge_mutex); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1785 | |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 1786 | /** |
| 1787 | * consume_stock: Try to consume stocked charge on this cpu. |
| 1788 | * @memcg: memcg to consume from. |
| 1789 | * @nr_pages: how many pages to charge. |
| 1790 | * |
| 1791 | * The charges will only happen if @memcg matches the current cpu's memcg |
| 1792 | * stock, and at least @nr_pages are available in that stock. Failure to |
| 1793 | * service an allocation will refill the stock. |
| 1794 | * |
| 1795 | * returns true if successful, false otherwise. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1796 | */ |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 1797 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1798 | { |
| 1799 | struct memcg_stock_pcp *stock; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1800 | bool ret = false; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1801 | |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 1802 | if (nr_pages > CHARGE_BATCH) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1803 | return ret; |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 1804 | |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1805 | stock = &get_cpu_var(memcg_stock); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1806 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 1807 | stock->nr_pages -= nr_pages; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1808 | ret = true; |
| 1809 | } |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1810 | put_cpu_var(memcg_stock); |
| 1811 | return ret; |
| 1812 | } |
| 1813 | |
| 1814 | /* |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1815 | * Returns stocks cached in percpu and reset cached information. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1816 | */ |
| 1817 | static void drain_stock(struct memcg_stock_pcp *stock) |
| 1818 | { |
| 1819 | struct mem_cgroup *old = stock->cached; |
| 1820 | |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 1821 | if (stock->nr_pages) { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1822 | page_counter_uncharge(&old->memory, stock->nr_pages); |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 1823 | if (do_memsw_account()) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1824 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
Johannes Weiner | e8ea14c | 2014-12-10 15:42:42 -0800 | [diff] [blame] | 1825 | css_put_many(&old->css, stock->nr_pages); |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 1826 | stock->nr_pages = 0; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1827 | } |
| 1828 | stock->cached = NULL; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1829 | } |
| 1830 | |
| 1831 | /* |
| 1832 | * This must be called under preempt disabled or must be called by |
| 1833 | * a thread which is pinned to local cpu. |
| 1834 | */ |
| 1835 | static void drain_local_stock(struct work_struct *dummy) |
| 1836 | { |
Christoph Lameter | 7c8e018 | 2014-06-04 16:07:56 -0700 | [diff] [blame] | 1837 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1838 | drain_stock(stock); |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 1839 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1840 | } |
| 1841 | |
| 1842 | /* |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1843 | * Cache charges(val) to local per_cpu area. |
Greg Thelen | 320cc51 | 2010-03-15 15:27:28 +0100 | [diff] [blame] | 1844 | * This will be consumed by consume_stock() function, later. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1845 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1846 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1847 | { |
| 1848 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); |
| 1849 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1850 | if (stock->cached != memcg) { /* reset if necessary */ |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1851 | drain_stock(stock); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1852 | stock->cached = memcg; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1853 | } |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 1854 | stock->nr_pages += nr_pages; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1855 | put_cpu_var(memcg_stock); |
| 1856 | } |
| 1857 | |
| 1858 | /* |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1859 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
Johannes Weiner | 6d3d6aa | 2014-12-10 15:42:50 -0800 | [diff] [blame] | 1860 | * of the hierarchy under it. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1861 | */ |
Johannes Weiner | 6d3d6aa | 2014-12-10 15:42:50 -0800 | [diff] [blame] | 1862 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1863 | { |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 1864 | int cpu, curcpu; |
Michal Hocko | d38144b | 2011-07-26 16:08:28 -0700 | [diff] [blame] | 1865 | |
Johannes Weiner | 6d3d6aa | 2014-12-10 15:42:50 -0800 | [diff] [blame] | 1866 | /* If someone's already draining, avoid adding running more workers. */ |
| 1867 | if (!mutex_trylock(&percpu_charge_mutex)) |
| 1868 | return; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1869 | /* Notify other cpus that system-wide "drain" is running */ |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1870 | get_online_cpus(); |
Johannes Weiner | 5af12d0 | 2011-08-25 15:59:07 -0700 | [diff] [blame] | 1871 | curcpu = get_cpu(); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1872 | for_each_online_cpu(cpu) { |
| 1873 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1874 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 1875 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1876 | memcg = stock->cached; |
| 1877 | if (!memcg || !stock->nr_pages) |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 1878 | continue; |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1879 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) |
Michal Hocko | 3e92041 | 2011-07-26 16:08:29 -0700 | [diff] [blame] | 1880 | continue; |
Michal Hocko | d1a05b6 | 2011-07-26 16:08:27 -0700 | [diff] [blame] | 1881 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
| 1882 | if (cpu == curcpu) |
| 1883 | drain_local_stock(&stock->work); |
| 1884 | else |
| 1885 | schedule_work_on(cpu, &stock->work); |
| 1886 | } |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1887 | } |
Johannes Weiner | 5af12d0 | 2011-08-25 15:59:07 -0700 | [diff] [blame] | 1888 | put_cpu(); |
Andrew Morton | f894ffa | 2013-09-12 15:13:35 -0700 | [diff] [blame] | 1889 | put_online_cpus(); |
Michal Hocko | 9f50fad | 2011-08-09 11:56:26 +0200 | [diff] [blame] | 1890 | mutex_unlock(&percpu_charge_mutex); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1891 | } |
| 1892 | |
Paul Gortmaker | 0db0628 | 2013-06-19 14:53:51 -0400 | [diff] [blame] | 1893 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1894 | unsigned long action, |
| 1895 | void *hcpu) |
| 1896 | { |
| 1897 | int cpu = (unsigned long)hcpu; |
| 1898 | struct memcg_stock_pcp *stock; |
| 1899 | |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 1900 | if (action == CPU_ONLINE) |
KAMEZAWA Hiroyuki | 1489eba | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1901 | return NOTIFY_OK; |
KAMEZAWA Hiroyuki | 1489eba | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1902 | |
Kirill A. Shutemov | d833049 | 2012-04-12 12:49:11 -0700 | [diff] [blame] | 1903 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1904 | return NOTIFY_OK; |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1905 | |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1906 | stock = &per_cpu(memcg_stock, cpu); |
| 1907 | drain_stock(stock); |
| 1908 | return NOTIFY_OK; |
| 1909 | } |
| 1910 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 1911 | static void reclaim_high(struct mem_cgroup *memcg, |
| 1912 | unsigned int nr_pages, |
| 1913 | gfp_t gfp_mask) |
| 1914 | { |
| 1915 | do { |
| 1916 | if (page_counter_read(&memcg->memory) <= memcg->high) |
| 1917 | continue; |
| 1918 | mem_cgroup_events(memcg, MEMCG_HIGH, 1); |
| 1919 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); |
| 1920 | } while ((memcg = parent_mem_cgroup(memcg))); |
| 1921 | } |
| 1922 | |
| 1923 | static void high_work_func(struct work_struct *work) |
| 1924 | { |
| 1925 | struct mem_cgroup *memcg; |
| 1926 | |
| 1927 | memcg = container_of(work, struct mem_cgroup, high_work); |
| 1928 | reclaim_high(memcg, CHARGE_BATCH, GFP_KERNEL); |
| 1929 | } |
| 1930 | |
Tejun Heo | b23afb9 | 2015-11-05 18:46:11 -0800 | [diff] [blame] | 1931 | /* |
| 1932 | * Scheduled by try_charge() to be executed from the userland return path |
| 1933 | * and reclaims memory over the high limit. |
| 1934 | */ |
| 1935 | void mem_cgroup_handle_over_high(void) |
| 1936 | { |
| 1937 | unsigned int nr_pages = current->memcg_nr_pages_over_high; |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 1938 | struct mem_cgroup *memcg; |
Tejun Heo | b23afb9 | 2015-11-05 18:46:11 -0800 | [diff] [blame] | 1939 | |
| 1940 | if (likely(!nr_pages)) |
| 1941 | return; |
| 1942 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 1943 | memcg = get_mem_cgroup_from_mm(current->mm); |
| 1944 | reclaim_high(memcg, nr_pages, GFP_KERNEL); |
Tejun Heo | b23afb9 | 2015-11-05 18:46:11 -0800 | [diff] [blame] | 1945 | css_put(&memcg->css); |
| 1946 | current->memcg_nr_pages_over_high = 0; |
| 1947 | } |
| 1948 | |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 1949 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
| 1950 | unsigned int nr_pages) |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1951 | { |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 1952 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
Johannes Weiner | 9b13061 | 2014-08-06 16:05:51 -0700 | [diff] [blame] | 1953 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 1954 | struct mem_cgroup *mem_over_limit; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1955 | struct page_counter *counter; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 1956 | unsigned long nr_reclaimed; |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 1957 | bool may_swap = true; |
| 1958 | bool drained = false; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 1959 | |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 1960 | if (mem_cgroup_is_root(memcg)) |
Tejun Heo | 10d53c7 | 2015-11-05 18:46:17 -0800 | [diff] [blame] | 1961 | return 0; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 1962 | retry: |
Michal Hocko | b6b6cc7 | 2014-04-07 15:37:44 -0700 | [diff] [blame] | 1963 | if (consume_stock(memcg, nr_pages)) |
Tejun Heo | 10d53c7 | 2015-11-05 18:46:17 -0800 | [diff] [blame] | 1964 | return 0; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1965 | |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 1966 | if (!do_memsw_account() || |
Johannes Weiner | 6071ca5 | 2015-11-05 18:50:26 -0800 | [diff] [blame] | 1967 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
| 1968 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 1969 | goto done_restock; |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 1970 | if (do_memsw_account()) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1971 | page_counter_uncharge(&memcg->memsw, batch); |
| 1972 | mem_over_limit = mem_cgroup_from_counter(counter, memory); |
Johannes Weiner | 3fbe724 | 2014-10-09 15:28:54 -0700 | [diff] [blame] | 1973 | } else { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1974 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 1975 | may_swap = false; |
Johannes Weiner | 3fbe724 | 2014-10-09 15:28:54 -0700 | [diff] [blame] | 1976 | } |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 1977 | |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 1978 | if (batch > nr_pages) { |
| 1979 | batch = nr_pages; |
| 1980 | goto retry; |
| 1981 | } |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1982 | |
Johannes Weiner | 06b078f | 2014-08-06 16:05:44 -0700 | [diff] [blame] | 1983 | /* |
| 1984 | * Unlike in global OOM situations, memcg is not in a physical |
| 1985 | * memory shortage. Allow dying and OOM-killed tasks to |
| 1986 | * bypass the last charges so that they can exit quickly and |
| 1987 | * free their memory. |
| 1988 | */ |
| 1989 | if (unlikely(test_thread_flag(TIF_MEMDIE) || |
| 1990 | fatal_signal_pending(current) || |
| 1991 | current->flags & PF_EXITING)) |
Tejun Heo | 10d53c7 | 2015-11-05 18:46:17 -0800 | [diff] [blame] | 1992 | goto force; |
Johannes Weiner | 06b078f | 2014-08-06 16:05:44 -0700 | [diff] [blame] | 1993 | |
| 1994 | if (unlikely(task_in_memcg_oom(current))) |
| 1995 | goto nomem; |
| 1996 | |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 1997 | if (!gfpflags_allow_blocking(gfp_mask)) |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 1998 | goto nomem; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1999 | |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 2000 | mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1); |
| 2001 | |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2002 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
| 2003 | gfp_mask, may_swap); |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2004 | |
Johannes Weiner | 61e02c7 | 2014-08-06 16:08:16 -0700 | [diff] [blame] | 2005 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2006 | goto retry; |
Johannes Weiner | 28c34c2 | 2014-08-06 16:05:47 -0700 | [diff] [blame] | 2007 | |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2008 | if (!drained) { |
Johannes Weiner | 6d3d6aa | 2014-12-10 15:42:50 -0800 | [diff] [blame] | 2009 | drain_all_stock(mem_over_limit); |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2010 | drained = true; |
| 2011 | goto retry; |
| 2012 | } |
| 2013 | |
Johannes Weiner | 28c34c2 | 2014-08-06 16:05:47 -0700 | [diff] [blame] | 2014 | if (gfp_mask & __GFP_NORETRY) |
| 2015 | goto nomem; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2016 | /* |
| 2017 | * Even though the limit is exceeded at this point, reclaim |
| 2018 | * may have been able to free some pages. Retry the charge |
| 2019 | * before killing the task. |
| 2020 | * |
| 2021 | * Only for regular pages, though: huge pages are rather |
| 2022 | * unlikely to succeed so close to the limit, and we fall back |
| 2023 | * to regular pages anyway in case of failure. |
| 2024 | */ |
Johannes Weiner | 61e02c7 | 2014-08-06 16:08:16 -0700 | [diff] [blame] | 2025 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2026 | goto retry; |
| 2027 | /* |
| 2028 | * At task move, charge accounts can be doubly counted. So, it's |
| 2029 | * better to wait until the end of task_move if something is going on. |
| 2030 | */ |
| 2031 | if (mem_cgroup_wait_acct_move(mem_over_limit)) |
| 2032 | goto retry; |
| 2033 | |
Johannes Weiner | 9b13061 | 2014-08-06 16:05:51 -0700 | [diff] [blame] | 2034 | if (nr_retries--) |
| 2035 | goto retry; |
| 2036 | |
Johannes Weiner | 06b078f | 2014-08-06 16:05:44 -0700 | [diff] [blame] | 2037 | if (gfp_mask & __GFP_NOFAIL) |
Tejun Heo | 10d53c7 | 2015-11-05 18:46:17 -0800 | [diff] [blame] | 2038 | goto force; |
Johannes Weiner | 06b078f | 2014-08-06 16:05:44 -0700 | [diff] [blame] | 2039 | |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2040 | if (fatal_signal_pending(current)) |
Tejun Heo | 10d53c7 | 2015-11-05 18:46:17 -0800 | [diff] [blame] | 2041 | goto force; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2042 | |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 2043 | mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1); |
| 2044 | |
Jerome Marchand | 3608de0 | 2015-11-05 18:47:29 -0800 | [diff] [blame] | 2045 | mem_cgroup_oom(mem_over_limit, gfp_mask, |
| 2046 | get_order(nr_pages * PAGE_SIZE)); |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2047 | nomem: |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 2048 | if (!(gfp_mask & __GFP_NOFAIL)) |
Johannes Weiner | 3168ecb | 2013-10-31 16:34:13 -0700 | [diff] [blame] | 2049 | return -ENOMEM; |
Tejun Heo | 10d53c7 | 2015-11-05 18:46:17 -0800 | [diff] [blame] | 2050 | force: |
| 2051 | /* |
| 2052 | * The allocation either can't fail or will lead to more memory |
| 2053 | * being freed very soon. Allow memory usage go over the limit |
| 2054 | * temporarily by force charging it. |
| 2055 | */ |
| 2056 | page_counter_charge(&memcg->memory, nr_pages); |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 2057 | if (do_memsw_account()) |
Tejun Heo | 10d53c7 | 2015-11-05 18:46:17 -0800 | [diff] [blame] | 2058 | page_counter_charge(&memcg->memsw, nr_pages); |
| 2059 | css_get_many(&memcg->css, nr_pages); |
| 2060 | |
| 2061 | return 0; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2062 | |
| 2063 | done_restock: |
Johannes Weiner | e8ea14c | 2014-12-10 15:42:42 -0800 | [diff] [blame] | 2064 | css_get_many(&memcg->css, batch); |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2065 | if (batch > nr_pages) |
| 2066 | refill_stock(memcg, batch - nr_pages); |
Tejun Heo | b23afb9 | 2015-11-05 18:46:11 -0800 | [diff] [blame] | 2067 | |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 2068 | /* |
Tejun Heo | b23afb9 | 2015-11-05 18:46:11 -0800 | [diff] [blame] | 2069 | * If the hierarchy is above the normal consumption range, schedule |
| 2070 | * reclaim on returning to userland. We can perform reclaim here |
Mel Gorman | 71baba4 | 2015-11-06 16:28:28 -0800 | [diff] [blame] | 2071 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
Tejun Heo | b23afb9 | 2015-11-05 18:46:11 -0800 | [diff] [blame] | 2072 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
| 2073 | * not recorded as it most likely matches current's and won't |
| 2074 | * change in the meantime. As high limit is checked again before |
| 2075 | * reclaim, the cost of mismatch is negligible. |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 2076 | */ |
| 2077 | do { |
Tejun Heo | b23afb9 | 2015-11-05 18:46:11 -0800 | [diff] [blame] | 2078 | if (page_counter_read(&memcg->memory) > memcg->high) { |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 2079 | /* Don't bother a random interrupted task */ |
| 2080 | if (in_interrupt()) { |
| 2081 | schedule_work(&memcg->high_work); |
| 2082 | break; |
| 2083 | } |
Vladimir Davydov | 9516a18 | 2015-12-11 13:40:24 -0800 | [diff] [blame] | 2084 | current->memcg_nr_pages_over_high += batch; |
Tejun Heo | b23afb9 | 2015-11-05 18:46:11 -0800 | [diff] [blame] | 2085 | set_notify_resume(current); |
| 2086 | break; |
| 2087 | } |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 2088 | } while ((memcg = parent_mem_cgroup(memcg))); |
Tejun Heo | 10d53c7 | 2015-11-05 18:46:17 -0800 | [diff] [blame] | 2089 | |
| 2090 | return 0; |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2091 | } |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2092 | |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 2093 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
Daisuke Nishimura | a3032a2 | 2009-12-15 16:47:10 -0800 | [diff] [blame] | 2094 | { |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2095 | if (mem_cgroup_is_root(memcg)) |
| 2096 | return; |
| 2097 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2098 | page_counter_uncharge(&memcg->memory, nr_pages); |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 2099 | if (do_memsw_account()) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2100 | page_counter_uncharge(&memcg->memsw, nr_pages); |
Daisuke Nishimura | a3032a2 | 2009-12-15 16:47:10 -0800 | [diff] [blame] | 2101 | |
Johannes Weiner | e8ea14c | 2014-12-10 15:42:42 -0800 | [diff] [blame] | 2102 | css_put_many(&memcg->css, nr_pages); |
KAMEZAWA Hiroyuki | d01dd17 | 2012-05-29 15:07:03 -0700 | [diff] [blame] | 2103 | } |
| 2104 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2105 | static void lock_page_lru(struct page *page, int *isolated) |
| 2106 | { |
| 2107 | struct zone *zone = page_zone(page); |
| 2108 | |
| 2109 | spin_lock_irq(&zone->lru_lock); |
| 2110 | if (PageLRU(page)) { |
| 2111 | struct lruvec *lruvec; |
| 2112 | |
| 2113 | lruvec = mem_cgroup_page_lruvec(page, zone); |
| 2114 | ClearPageLRU(page); |
| 2115 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
| 2116 | *isolated = 1; |
| 2117 | } else |
| 2118 | *isolated = 0; |
| 2119 | } |
| 2120 | |
| 2121 | static void unlock_page_lru(struct page *page, int isolated) |
| 2122 | { |
| 2123 | struct zone *zone = page_zone(page); |
| 2124 | |
| 2125 | if (isolated) { |
| 2126 | struct lruvec *lruvec; |
| 2127 | |
| 2128 | lruvec = mem_cgroup_page_lruvec(page, zone); |
| 2129 | VM_BUG_ON_PAGE(PageLRU(page), page); |
| 2130 | SetPageLRU(page); |
| 2131 | add_page_to_lru_list(page, lruvec, page_lru(page)); |
| 2132 | } |
| 2133 | spin_unlock_irq(&zone->lru_lock); |
| 2134 | } |
| 2135 | |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 2136 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
Johannes Weiner | 6abb5a8 | 2014-08-08 14:19:33 -0700 | [diff] [blame] | 2137 | bool lrucare) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2138 | { |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2139 | int isolated; |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2140 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2141 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2142 | |
| 2143 | /* |
| 2144 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page |
| 2145 | * may already be on some other mem_cgroup's LRU. Take care of it. |
| 2146 | */ |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2147 | if (lrucare) |
| 2148 | lock_page_lru(page, &isolated); |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2149 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2150 | /* |
| 2151 | * Nobody should be changing or seriously looking at |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2152 | * page->mem_cgroup at this point: |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2153 | * |
| 2154 | * - the page is uncharged |
| 2155 | * |
| 2156 | * - the page is off-LRU |
| 2157 | * |
| 2158 | * - an anonymous fault has exclusive page access, except for |
| 2159 | * a locked page table |
| 2160 | * |
| 2161 | * - a page cache insertion, a swapin fault, or a migration |
| 2162 | * have the page locked |
| 2163 | */ |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2164 | page->mem_cgroup = memcg; |
Hugh Dickins | 3be9127 | 2008-02-07 00:14:19 -0800 | [diff] [blame] | 2165 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2166 | if (lrucare) |
| 2167 | unlock_page_lru(page, isolated); |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2168 | } |
| 2169 | |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 2170 | #ifndef CONFIG_SLOB |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2171 | static int memcg_alloc_cache_id(void) |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 2172 | { |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2173 | int id, size; |
| 2174 | int err; |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 2175 | |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 2176 | id = ida_simple_get(&memcg_cache_ida, |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2177 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
| 2178 | if (id < 0) |
| 2179 | return id; |
| 2180 | |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 2181 | if (id < memcg_nr_cache_ids) |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2182 | return id; |
| 2183 | |
| 2184 | /* |
| 2185 | * There's no space for the new id in memcg_caches arrays, |
| 2186 | * so we have to grow them. |
| 2187 | */ |
Vladimir Davydov | 05257a1 | 2015-02-12 14:59:01 -0800 | [diff] [blame] | 2188 | down_write(&memcg_cache_ids_sem); |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2189 | |
| 2190 | size = 2 * (id + 1); |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 2191 | if (size < MEMCG_CACHES_MIN_SIZE) |
| 2192 | size = MEMCG_CACHES_MIN_SIZE; |
| 2193 | else if (size > MEMCG_CACHES_MAX_SIZE) |
| 2194 | size = MEMCG_CACHES_MAX_SIZE; |
| 2195 | |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2196 | err = memcg_update_all_caches(size); |
Vladimir Davydov | 05257a1 | 2015-02-12 14:59:01 -0800 | [diff] [blame] | 2197 | if (!err) |
Vladimir Davydov | 60d3fd3 | 2015-02-12 14:59:10 -0800 | [diff] [blame] | 2198 | err = memcg_update_all_list_lrus(size); |
| 2199 | if (!err) |
Vladimir Davydov | 05257a1 | 2015-02-12 14:59:01 -0800 | [diff] [blame] | 2200 | memcg_nr_cache_ids = size; |
| 2201 | |
| 2202 | up_write(&memcg_cache_ids_sem); |
| 2203 | |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2204 | if (err) { |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 2205 | ida_simple_remove(&memcg_cache_ida, id); |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2206 | return err; |
| 2207 | } |
| 2208 | return id; |
| 2209 | } |
| 2210 | |
| 2211 | static void memcg_free_cache_id(int id) |
| 2212 | { |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 2213 | ida_simple_remove(&memcg_cache_ida, id); |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 2214 | } |
| 2215 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2216 | struct memcg_kmem_cache_create_work { |
Vladimir Davydov | 5722d09 | 2014-04-07 15:39:24 -0700 | [diff] [blame] | 2217 | struct mem_cgroup *memcg; |
| 2218 | struct kmem_cache *cachep; |
| 2219 | struct work_struct work; |
| 2220 | }; |
| 2221 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2222 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2223 | { |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2224 | struct memcg_kmem_cache_create_work *cw = |
| 2225 | container_of(w, struct memcg_kmem_cache_create_work, work); |
Vladimir Davydov | 5722d09 | 2014-04-07 15:39:24 -0700 | [diff] [blame] | 2226 | struct mem_cgroup *memcg = cw->memcg; |
| 2227 | struct kmem_cache *cachep = cw->cachep; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2228 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2229 | memcg_create_kmem_cache(memcg, cachep); |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 2230 | |
Vladimir Davydov | 5722d09 | 2014-04-07 15:39:24 -0700 | [diff] [blame] | 2231 | css_put(&memcg->css); |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2232 | kfree(cw); |
| 2233 | } |
| 2234 | |
| 2235 | /* |
| 2236 | * Enqueue the creation of a per-memcg kmem_cache. |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2237 | */ |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2238 | static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
| 2239 | struct kmem_cache *cachep) |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2240 | { |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2241 | struct memcg_kmem_cache_create_work *cw; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2242 | |
Vladimir Davydov | 776ed0f | 2014-06-04 16:10:02 -0700 | [diff] [blame] | 2243 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2244 | if (!cw) |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2245 | return; |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2246 | |
| 2247 | css_get(&memcg->css); |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2248 | |
| 2249 | cw->memcg = memcg; |
| 2250 | cw->cachep = cachep; |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2251 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2252 | |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2253 | schedule_work(&cw->work); |
| 2254 | } |
| 2255 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2256 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
| 2257 | struct kmem_cache *cachep) |
Glauber Costa | 0e9d92f | 2012-12-18 14:22:42 -0800 | [diff] [blame] | 2258 | { |
| 2259 | /* |
| 2260 | * We need to stop accounting when we kmalloc, because if the |
| 2261 | * corresponding kmalloc cache is not yet created, the first allocation |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2262 | * in __memcg_schedule_kmem_cache_create will recurse. |
Glauber Costa | 0e9d92f | 2012-12-18 14:22:42 -0800 | [diff] [blame] | 2263 | * |
| 2264 | * However, it is better to enclose the whole function. Depending on |
| 2265 | * the debugging options enabled, INIT_WORK(), for instance, can |
| 2266 | * trigger an allocation. This too, will make us recurse. Because at |
| 2267 | * this point we can't allow ourselves back into memcg_kmem_get_cache, |
| 2268 | * the safest choice is to do it like this, wrapping the whole function. |
| 2269 | */ |
Vladimir Davydov | 6f185c2 | 2014-12-12 16:55:15 -0800 | [diff] [blame] | 2270 | current->memcg_kmem_skip_account = 1; |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2271 | __memcg_schedule_kmem_cache_create(memcg, cachep); |
Vladimir Davydov | 6f185c2 | 2014-12-12 16:55:15 -0800 | [diff] [blame] | 2272 | current->memcg_kmem_skip_account = 0; |
Glauber Costa | 0e9d92f | 2012-12-18 14:22:42 -0800 | [diff] [blame] | 2273 | } |
Vladimir Davydov | c67a8a6 | 2014-06-04 16:07:39 -0700 | [diff] [blame] | 2274 | |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2275 | /* |
| 2276 | * Return the kmem_cache we're supposed to use for a slab allocation. |
| 2277 | * We try to use the current memcg's version of the cache. |
| 2278 | * |
| 2279 | * If the cache does not exist yet, if we are the first user of it, |
| 2280 | * we either create it immediately, if possible, or create it asynchronously |
| 2281 | * in a workqueue. |
| 2282 | * In the latter case, we will let the current allocation go through with |
| 2283 | * the original cache. |
| 2284 | * |
| 2285 | * Can't be called in interrupt context or from kernel threads. |
| 2286 | * This function needs to be called with rcu_read_lock() held. |
| 2287 | */ |
Vladimir Davydov | 230e9fc | 2016-01-14 15:18:15 -0800 | [diff] [blame] | 2288 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp) |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2289 | { |
| 2290 | struct mem_cgroup *memcg; |
Vladimir Davydov | 959c896 | 2014-01-23 15:52:59 -0800 | [diff] [blame] | 2291 | struct kmem_cache *memcg_cachep; |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 2292 | int kmemcg_id; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2293 | |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 2294 | VM_BUG_ON(!is_root_cache(cachep)); |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2295 | |
Vladimir Davydov | 230e9fc | 2016-01-14 15:18:15 -0800 | [diff] [blame] | 2296 | if (cachep->flags & SLAB_ACCOUNT) |
| 2297 | gfp |= __GFP_ACCOUNT; |
| 2298 | |
| 2299 | if (!(gfp & __GFP_ACCOUNT)) |
| 2300 | return cachep; |
| 2301 | |
Vladimir Davydov | 9d100c5 | 2014-12-12 16:54:53 -0800 | [diff] [blame] | 2302 | if (current->memcg_kmem_skip_account) |
Glauber Costa | 0e9d92f | 2012-12-18 14:22:42 -0800 | [diff] [blame] | 2303 | return cachep; |
| 2304 | |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2305 | memcg = get_mem_cgroup_from_mm(current->mm); |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 2306 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 2307 | if (kmemcg_id < 0) |
Li Zefan | ca0dde9 | 2013-04-29 15:08:57 -0700 | [diff] [blame] | 2308 | goto out; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2309 | |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 2310 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2311 | if (likely(memcg_cachep)) |
| 2312 | return memcg_cachep; |
Li Zefan | ca0dde9 | 2013-04-29 15:08:57 -0700 | [diff] [blame] | 2313 | |
| 2314 | /* |
| 2315 | * If we are in a safe context (can wait, and not in interrupt |
| 2316 | * context), we could be be predictable and return right away. |
| 2317 | * This would guarantee that the allocation being performed |
| 2318 | * already belongs in the new cache. |
| 2319 | * |
| 2320 | * However, there are some clashes that can arrive from locking. |
| 2321 | * For instance, because we acquire the slab_mutex while doing |
Vladimir Davydov | 776ed0f | 2014-06-04 16:10:02 -0700 | [diff] [blame] | 2322 | * memcg_create_kmem_cache, this means no further allocation |
| 2323 | * could happen with the slab_mutex held. So it's better to |
| 2324 | * defer everything. |
Li Zefan | ca0dde9 | 2013-04-29 15:08:57 -0700 | [diff] [blame] | 2325 | */ |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2326 | memcg_schedule_kmem_cache_create(memcg, cachep); |
Li Zefan | ca0dde9 | 2013-04-29 15:08:57 -0700 | [diff] [blame] | 2327 | out: |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2328 | css_put(&memcg->css); |
Li Zefan | ca0dde9 | 2013-04-29 15:08:57 -0700 | [diff] [blame] | 2329 | return cachep; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2330 | } |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2331 | |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2332 | void __memcg_kmem_put_cache(struct kmem_cache *cachep) |
| 2333 | { |
| 2334 | if (!is_root_cache(cachep)) |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 2335 | css_put(&cachep->memcg_params.memcg->css); |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2336 | } |
| 2337 | |
Vladimir Davydov | f3ccb2c4 | 2015-11-05 18:49:01 -0800 | [diff] [blame] | 2338 | int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, |
| 2339 | struct mem_cgroup *memcg) |
| 2340 | { |
| 2341 | unsigned int nr_pages = 1 << order; |
| 2342 | struct page_counter *counter; |
Johannes Weiner | 6071ca5 | 2015-11-05 18:50:26 -0800 | [diff] [blame] | 2343 | int ret; |
Vladimir Davydov | f3ccb2c4 | 2015-11-05 18:49:01 -0800 | [diff] [blame] | 2344 | |
Vladimir Davydov | f3ccb2c4 | 2015-11-05 18:49:01 -0800 | [diff] [blame] | 2345 | ret = try_charge(memcg, gfp, nr_pages); |
Johannes Weiner | 52c29b0 | 2016-01-20 15:02:35 -0800 | [diff] [blame] | 2346 | if (ret) |
Vladimir Davydov | f3ccb2c4 | 2015-11-05 18:49:01 -0800 | [diff] [blame] | 2347 | return ret; |
Johannes Weiner | 52c29b0 | 2016-01-20 15:02:35 -0800 | [diff] [blame] | 2348 | |
| 2349 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && |
| 2350 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { |
| 2351 | cancel_charge(memcg, nr_pages); |
| 2352 | return -ENOMEM; |
Vladimir Davydov | f3ccb2c4 | 2015-11-05 18:49:01 -0800 | [diff] [blame] | 2353 | } |
| 2354 | |
| 2355 | page->mem_cgroup = memcg; |
| 2356 | |
| 2357 | return 0; |
| 2358 | } |
| 2359 | |
Vladimir Davydov | d05e83a | 2015-11-05 18:48:59 -0800 | [diff] [blame] | 2360 | int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order) |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2361 | { |
| 2362 | struct mem_cgroup *memcg; |
Vladimir Davydov | fcff7d7 | 2016-03-17 14:17:29 -0700 | [diff] [blame] | 2363 | int ret = 0; |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2364 | |
Johannes Weiner | df38197 | 2014-04-07 15:37:43 -0700 | [diff] [blame] | 2365 | memcg = get_mem_cgroup_from_mm(current->mm); |
Vladimir Davydov | b6ecd2d | 2016-03-17 14:18:33 -0700 | [diff] [blame] | 2366 | if (!mem_cgroup_is_root(memcg)) |
Vladimir Davydov | fcff7d7 | 2016-03-17 14:17:29 -0700 | [diff] [blame] | 2367 | ret = __memcg_kmem_charge_memcg(page, gfp, order, memcg); |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2368 | css_put(&memcg->css); |
Vladimir Davydov | d05e83a | 2015-11-05 18:48:59 -0800 | [diff] [blame] | 2369 | return ret; |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2370 | } |
| 2371 | |
Vladimir Davydov | d05e83a | 2015-11-05 18:48:59 -0800 | [diff] [blame] | 2372 | void __memcg_kmem_uncharge(struct page *page, int order) |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2373 | { |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2374 | struct mem_cgroup *memcg = page->mem_cgroup; |
Vladimir Davydov | f3ccb2c4 | 2015-11-05 18:49:01 -0800 | [diff] [blame] | 2375 | unsigned int nr_pages = 1 << order; |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2376 | |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2377 | if (!memcg) |
| 2378 | return; |
| 2379 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 2380 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 2381 | |
Johannes Weiner | 52c29b0 | 2016-01-20 15:02:35 -0800 | [diff] [blame] | 2382 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
| 2383 | page_counter_uncharge(&memcg->kmem, nr_pages); |
| 2384 | |
Vladimir Davydov | f3ccb2c4 | 2015-11-05 18:49:01 -0800 | [diff] [blame] | 2385 | page_counter_uncharge(&memcg->memory, nr_pages); |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 2386 | if (do_memsw_account()) |
Vladimir Davydov | f3ccb2c4 | 2015-11-05 18:49:01 -0800 | [diff] [blame] | 2387 | page_counter_uncharge(&memcg->memsw, nr_pages); |
| 2388 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2389 | page->mem_cgroup = NULL; |
Vladimir Davydov | f3ccb2c4 | 2015-11-05 18:49:01 -0800 | [diff] [blame] | 2390 | css_put_many(&memcg->css, nr_pages); |
Vladimir Davydov | 60d3fd3 | 2015-02-12 14:59:10 -0800 | [diff] [blame] | 2391 | } |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 2392 | #endif /* !CONFIG_SLOB */ |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2393 | |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2394 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 2395 | |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2396 | /* |
| 2397 | * Because tail pages are not marked as "used", set it. We're under |
Kirill A. Shutemov | 3ac808f | 2016-01-15 16:53:07 -0800 | [diff] [blame] | 2398 | * zone->lru_lock and migration entries setup in all page mappings. |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2399 | */ |
KAMEZAWA Hiroyuki | e94c8a9 | 2012-01-12 17:18:20 -0800 | [diff] [blame] | 2400 | void mem_cgroup_split_huge_fixup(struct page *head) |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2401 | { |
KAMEZAWA Hiroyuki | e94c8a9 | 2012-01-12 17:18:20 -0800 | [diff] [blame] | 2402 | int i; |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2403 | |
KAMEZAWA Hiroyuki | 3d37c4a | 2011-01-25 15:07:28 -0800 | [diff] [blame] | 2404 | if (mem_cgroup_disabled()) |
| 2405 | return; |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 2406 | |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 2407 | for (i = 1; i < HPAGE_PMD_NR; i++) |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2408 | head[i].mem_cgroup = head->mem_cgroup; |
Michal Hocko | b9982f8 | 2014-12-10 15:43:51 -0800 | [diff] [blame] | 2409 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2410 | __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] | 2411 | HPAGE_PMD_NR); |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2412 | } |
Hugh Dickins | 12d2710 | 2012-01-12 17:19:52 -0800 | [diff] [blame] | 2413 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2414 | |
Andrew Morton | c255a45 | 2012-07-31 16:43:02 -0700 | [diff] [blame] | 2415 | #ifdef CONFIG_MEMCG_SWAP |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2416 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
| 2417 | bool charge) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2418 | { |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2419 | int val = (charge) ? 1 : -1; |
| 2420 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2421 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2422 | |
| 2423 | /** |
| 2424 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. |
| 2425 | * @entry: swap entry to be moved |
| 2426 | * @from: mem_cgroup which the entry is moved from |
| 2427 | * @to: mem_cgroup which the entry is moved to |
| 2428 | * |
| 2429 | * It succeeds only when the swap_cgroup's record for this entry is the same |
| 2430 | * as the mem_cgroup's id of @from. |
| 2431 | * |
| 2432 | * Returns 0 on success, -EINVAL on failure. |
| 2433 | * |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2434 | * 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] | 2435 | * both res and memsw, and called css_get(). |
| 2436 | */ |
| 2437 | static int mem_cgroup_move_swap_account(swp_entry_t entry, |
Hugh Dickins | e91cbb4 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 2438 | struct mem_cgroup *from, struct mem_cgroup *to) |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2439 | { |
| 2440 | unsigned short old_id, new_id; |
| 2441 | |
Li Zefan | 34c00c3 | 2013-09-23 16:56:01 +0800 | [diff] [blame] | 2442 | old_id = mem_cgroup_id(from); |
| 2443 | new_id = mem_cgroup_id(to); |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2444 | |
| 2445 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2446 | mem_cgroup_swap_statistics(from, false); |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2447 | mem_cgroup_swap_statistics(to, true); |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2448 | return 0; |
| 2449 | } |
| 2450 | return -EINVAL; |
| 2451 | } |
| 2452 | #else |
| 2453 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, |
Hugh Dickins | e91cbb4 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 2454 | struct mem_cgroup *from, struct mem_cgroup *to) |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2455 | { |
| 2456 | return -EINVAL; |
| 2457 | } |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2458 | #endif |
| 2459 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2460 | static DEFINE_MUTEX(memcg_limit_mutex); |
Daisuke Nishimura | f212ad7 | 2011-03-23 16:42:25 -0700 | [diff] [blame] | 2461 | |
KOSAKI Motohiro | d38d2a7 | 2009-01-06 14:39:44 -0800 | [diff] [blame] | 2462 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2463 | unsigned long limit) |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2464 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2465 | unsigned long curusage; |
| 2466 | unsigned long oldusage; |
| 2467 | bool enlarge = false; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2468 | int retry_count; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2469 | int ret; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2470 | |
| 2471 | /* |
| 2472 | * For keeping hierarchical_reclaim simple, how long we should retry |
| 2473 | * is depends on callers. We set our retry-count to be function |
| 2474 | * of # of children which we should visit in this loop. |
| 2475 | */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2476 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
| 2477 | mem_cgroup_count_children(memcg); |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2478 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2479 | oldusage = page_counter_read(&memcg->memory); |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2480 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2481 | do { |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2482 | if (signal_pending(current)) { |
| 2483 | ret = -EINTR; |
| 2484 | break; |
| 2485 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2486 | |
| 2487 | mutex_lock(&memcg_limit_mutex); |
| 2488 | if (limit > memcg->memsw.limit) { |
| 2489 | mutex_unlock(&memcg_limit_mutex); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2490 | ret = -EINVAL; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2491 | break; |
| 2492 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2493 | if (limit > memcg->memory.limit) |
| 2494 | enlarge = true; |
| 2495 | ret = page_counter_limit(&memcg->memory, limit); |
| 2496 | mutex_unlock(&memcg_limit_mutex); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2497 | |
| 2498 | if (!ret) |
| 2499 | break; |
| 2500 | |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2501 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
| 2502 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2503 | curusage = page_counter_read(&memcg->memory); |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2504 | /* Usage is reduced ? */ |
Andrew Morton | f894ffa | 2013-09-12 15:13:35 -0700 | [diff] [blame] | 2505 | if (curusage >= oldusage) |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2506 | retry_count--; |
| 2507 | else |
| 2508 | oldusage = curusage; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2509 | } while (retry_count); |
| 2510 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2511 | if (!ret && enlarge) |
| 2512 | memcg_oom_recover(memcg); |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 2513 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2514 | return ret; |
| 2515 | } |
| 2516 | |
Li Zefan | 338c843 | 2009-06-17 16:27:15 -0700 | [diff] [blame] | 2517 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2518 | unsigned long limit) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2519 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2520 | unsigned long curusage; |
| 2521 | unsigned long oldusage; |
| 2522 | bool enlarge = false; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2523 | int retry_count; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2524 | int ret; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2525 | |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2526 | /* see mem_cgroup_resize_res_limit */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2527 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
| 2528 | mem_cgroup_count_children(memcg); |
| 2529 | |
| 2530 | oldusage = page_counter_read(&memcg->memsw); |
| 2531 | |
| 2532 | do { |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2533 | if (signal_pending(current)) { |
| 2534 | ret = -EINTR; |
| 2535 | break; |
| 2536 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2537 | |
| 2538 | mutex_lock(&memcg_limit_mutex); |
| 2539 | if (limit < memcg->memory.limit) { |
| 2540 | mutex_unlock(&memcg_limit_mutex); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2541 | ret = -EINVAL; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2542 | break; |
| 2543 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2544 | if (limit > memcg->memsw.limit) |
| 2545 | enlarge = true; |
| 2546 | ret = page_counter_limit(&memcg->memsw, limit); |
| 2547 | mutex_unlock(&memcg_limit_mutex); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2548 | |
| 2549 | if (!ret) |
| 2550 | break; |
| 2551 | |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2552 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
| 2553 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2554 | curusage = page_counter_read(&memcg->memsw); |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2555 | /* Usage is reduced ? */ |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2556 | if (curusage >= oldusage) |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2557 | retry_count--; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2558 | else |
| 2559 | oldusage = curusage; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2560 | } while (retry_count); |
| 2561 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2562 | if (!ret && enlarge) |
| 2563 | memcg_oom_recover(memcg); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2564 | |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2565 | return ret; |
| 2566 | } |
| 2567 | |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2568 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
| 2569 | gfp_t gfp_mask, |
| 2570 | unsigned long *total_scanned) |
| 2571 | { |
| 2572 | unsigned long nr_reclaimed = 0; |
| 2573 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; |
| 2574 | unsigned long reclaimed; |
| 2575 | int loop = 0; |
| 2576 | struct mem_cgroup_tree_per_zone *mctz; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2577 | unsigned long excess; |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2578 | unsigned long nr_scanned; |
| 2579 | |
| 2580 | if (order > 0) |
| 2581 | return 0; |
| 2582 | |
| 2583 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); |
| 2584 | /* |
| 2585 | * This loop can run a while, specially if mem_cgroup's continuously |
| 2586 | * keep exceeding their soft limit and putting the system under |
| 2587 | * pressure |
| 2588 | */ |
| 2589 | do { |
| 2590 | if (next_mz) |
| 2591 | mz = next_mz; |
| 2592 | else |
| 2593 | mz = mem_cgroup_largest_soft_limit_node(mctz); |
| 2594 | if (!mz) |
| 2595 | break; |
| 2596 | |
| 2597 | nr_scanned = 0; |
| 2598 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, |
| 2599 | gfp_mask, &nr_scanned); |
| 2600 | nr_reclaimed += reclaimed; |
| 2601 | *total_scanned += nr_scanned; |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2602 | spin_lock_irq(&mctz->lock); |
Vladimir Davydov | bc2f2e7 | 2014-12-10 15:43:40 -0800 | [diff] [blame] | 2603 | __mem_cgroup_remove_exceeded(mz, mctz); |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2604 | |
| 2605 | /* |
| 2606 | * If we failed to reclaim anything from this memory cgroup |
| 2607 | * it is time to move on to the next cgroup |
| 2608 | */ |
| 2609 | next_mz = NULL; |
Vladimir Davydov | bc2f2e7 | 2014-12-10 15:43:40 -0800 | [diff] [blame] | 2610 | if (!reclaimed) |
| 2611 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); |
| 2612 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2613 | excess = soft_limit_excess(mz->memcg); |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2614 | /* |
| 2615 | * One school of thought says that we should not add |
| 2616 | * back the node to the tree if reclaim returns 0. |
| 2617 | * But our reclaim could return 0, simply because due |
| 2618 | * to priority we are exposing a smaller subset of |
| 2619 | * memory to reclaim from. Consider this as a longer |
| 2620 | * term TODO. |
| 2621 | */ |
| 2622 | /* If excess == 0, no tree ops */ |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 2623 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2624 | spin_unlock_irq(&mctz->lock); |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2625 | css_put(&mz->memcg->css); |
| 2626 | loop++; |
| 2627 | /* |
| 2628 | * Could not reclaim anything and there are no more |
| 2629 | * mem cgroups to try or we seem to be looping without |
| 2630 | * reclaiming anything. |
| 2631 | */ |
| 2632 | if (!nr_reclaimed && |
| 2633 | (next_mz == NULL || |
| 2634 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) |
| 2635 | break; |
| 2636 | } while (!nr_reclaimed); |
| 2637 | if (next_mz) |
| 2638 | css_put(&next_mz->memcg->css); |
| 2639 | return nr_reclaimed; |
| 2640 | } |
| 2641 | |
Tejun Heo | ea280e7 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 2642 | /* |
| 2643 | * Test whether @memcg has children, dead or alive. Note that this |
| 2644 | * function doesn't care whether @memcg has use_hierarchy enabled and |
| 2645 | * returns %true if there are child csses according to the cgroup |
| 2646 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. |
| 2647 | */ |
Glauber Costa | b5f99b5 | 2013-02-22 16:34:53 -0800 | [diff] [blame] | 2648 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
| 2649 | { |
Tejun Heo | ea280e7 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 2650 | bool ret; |
| 2651 | |
Tejun Heo | ea280e7 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 2652 | rcu_read_lock(); |
| 2653 | ret = css_next_child(NULL, &memcg->css); |
| 2654 | rcu_read_unlock(); |
| 2655 | return ret; |
Glauber Costa | b5f99b5 | 2013-02-22 16:34:53 -0800 | [diff] [blame] | 2656 | } |
| 2657 | |
| 2658 | /* |
Greg Thelen | 5103817 | 2016-05-20 16:58:18 -0700 | [diff] [blame] | 2659 | * Reclaims as many pages from the given memcg as possible. |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 2660 | * |
| 2661 | * Caller is responsible for holding css reference for memcg. |
| 2662 | */ |
| 2663 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) |
| 2664 | { |
| 2665 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 2666 | |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 2667 | /* we call try-to-free pages for make this cgroup empty */ |
| 2668 | lru_add_drain_all(); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2669 | /* try to free all pages in this cgroup */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2670 | while (nr_retries && page_counter_read(&memcg->memory)) { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2671 | int progress; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 2672 | |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 2673 | if (signal_pending(current)) |
| 2674 | return -EINTR; |
| 2675 | |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2676 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
| 2677 | GFP_KERNEL, true); |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 2678 | if (!progress) { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2679 | nr_retries--; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 2680 | /* maybe some writeback is necessary */ |
Jens Axboe | 8aa7e84 | 2009-07-09 14:52:32 +0200 | [diff] [blame] | 2681 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 2682 | } |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2683 | |
| 2684 | } |
Michal Hocko | ab5196c | 2012-10-26 13:37:32 +0200 | [diff] [blame] | 2685 | |
| 2686 | return 0; |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 2687 | } |
| 2688 | |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 2689 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
| 2690 | char *buf, size_t nbytes, |
| 2691 | loff_t off) |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 2692 | { |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 2693 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 2694 | |
Michal Hocko | d842301 | 2012-10-26 13:37:29 +0200 | [diff] [blame] | 2695 | if (mem_cgroup_is_root(memcg)) |
| 2696 | return -EINVAL; |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 2697 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 2698 | } |
| 2699 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 2700 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
| 2701 | struct cftype *cft) |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 2702 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 2703 | return mem_cgroup_from_css(css)->use_hierarchy; |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 2704 | } |
| 2705 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 2706 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
| 2707 | struct cftype *cft, u64 val) |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 2708 | { |
| 2709 | int retval = 0; |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 2710 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Tejun Heo | 5c9d535 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 2711 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 2712 | |
Glauber Costa | 567fb43 | 2012-07-31 16:43:07 -0700 | [diff] [blame] | 2713 | if (memcg->use_hierarchy == val) |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 2714 | return 0; |
Glauber Costa | 567fb43 | 2012-07-31 16:43:07 -0700 | [diff] [blame] | 2715 | |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 2716 | /* |
André Goddard Rosa | af901ca | 2009-11-14 13:09:05 -0200 | [diff] [blame] | 2717 | * 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] | 2718 | * in the child subtrees. If it is unset, then the change can |
| 2719 | * occur, provided the current cgroup has no children. |
| 2720 | * |
| 2721 | * For the root cgroup, parent_mem is NULL, we allow value to be |
| 2722 | * set if there are no children. |
| 2723 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2724 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 2725 | (val == 1 || val == 0)) { |
Tejun Heo | ea280e7 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 2726 | if (!memcg_has_children(memcg)) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2727 | memcg->use_hierarchy = val; |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 2728 | else |
| 2729 | retval = -EBUSY; |
| 2730 | } else |
| 2731 | retval = -EINVAL; |
Glauber Costa | 567fb43 | 2012-07-31 16:43:07 -0700 | [diff] [blame] | 2732 | |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 2733 | return retval; |
| 2734 | } |
| 2735 | |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 2736 | static void tree_stat(struct mem_cgroup *memcg, unsigned long *stat) |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2737 | { |
| 2738 | struct mem_cgroup *iter; |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 2739 | int i; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2740 | |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 2741 | memset(stat, 0, sizeof(*stat) * MEMCG_NR_STAT); |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2742 | |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 2743 | for_each_mem_cgroup_tree(iter, memcg) { |
| 2744 | for (i = 0; i < MEMCG_NR_STAT; i++) |
| 2745 | stat[i] += mem_cgroup_read_stat(iter, i); |
| 2746 | } |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2747 | } |
| 2748 | |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 2749 | static void tree_events(struct mem_cgroup *memcg, unsigned long *events) |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 2750 | { |
| 2751 | struct mem_cgroup *iter; |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 2752 | int i; |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 2753 | |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 2754 | memset(events, 0, sizeof(*events) * MEMCG_NR_EVENTS); |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 2755 | |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 2756 | for_each_mem_cgroup_tree(iter, memcg) { |
| 2757 | for (i = 0; i < MEMCG_NR_EVENTS; i++) |
| 2758 | events[i] += mem_cgroup_read_events(iter, i); |
| 2759 | } |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 2760 | } |
| 2761 | |
Andrew Morton | 6f64615 | 2015-11-06 16:28:58 -0800 | [diff] [blame] | 2762 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2763 | { |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 2764 | unsigned long val = 0; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2765 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2766 | if (mem_cgroup_is_root(memcg)) { |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 2767 | struct mem_cgroup *iter; |
| 2768 | |
| 2769 | for_each_mem_cgroup_tree(iter, memcg) { |
| 2770 | val += mem_cgroup_read_stat(iter, |
| 2771 | MEM_CGROUP_STAT_CACHE); |
| 2772 | val += mem_cgroup_read_stat(iter, |
| 2773 | MEM_CGROUP_STAT_RSS); |
| 2774 | if (swap) |
| 2775 | val += mem_cgroup_read_stat(iter, |
| 2776 | MEM_CGROUP_STAT_SWAP); |
| 2777 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2778 | } else { |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2779 | if (!swap) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2780 | val = page_counter_read(&memcg->memory); |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2781 | else |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2782 | val = page_counter_read(&memcg->memsw); |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2783 | } |
Michal Hocko | c12176d | 2015-11-05 18:50:29 -0800 | [diff] [blame] | 2784 | return val; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2785 | } |
| 2786 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2787 | enum { |
| 2788 | RES_USAGE, |
| 2789 | RES_LIMIT, |
| 2790 | RES_MAX_USAGE, |
| 2791 | RES_FAILCNT, |
| 2792 | RES_SOFT_LIMIT, |
| 2793 | }; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2794 | |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 2795 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 2796 | struct cftype *cft) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 2797 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 2798 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2799 | struct page_counter *counter; |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 2800 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2801 | switch (MEMFILE_TYPE(cft->private)) { |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2802 | case _MEM: |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2803 | counter = &memcg->memory; |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 2804 | break; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2805 | case _MEMSWAP: |
| 2806 | counter = &memcg->memsw; |
| 2807 | break; |
| 2808 | case _KMEM: |
| 2809 | counter = &memcg->kmem; |
| 2810 | break; |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 2811 | case _TCP: |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 2812 | counter = &memcg->tcpmem; |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 2813 | break; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2814 | default: |
| 2815 | BUG(); |
| 2816 | } |
| 2817 | |
| 2818 | switch (MEMFILE_ATTR(cft->private)) { |
| 2819 | case RES_USAGE: |
| 2820 | if (counter == &memcg->memory) |
Michal Hocko | c12176d | 2015-11-05 18:50:29 -0800 | [diff] [blame] | 2821 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2822 | if (counter == &memcg->memsw) |
Michal Hocko | c12176d | 2015-11-05 18:50:29 -0800 | [diff] [blame] | 2823 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2824 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
| 2825 | case RES_LIMIT: |
| 2826 | return (u64)counter->limit * PAGE_SIZE; |
| 2827 | case RES_MAX_USAGE: |
| 2828 | return (u64)counter->watermark * PAGE_SIZE; |
| 2829 | case RES_FAILCNT: |
| 2830 | return counter->failcnt; |
| 2831 | case RES_SOFT_LIMIT: |
| 2832 | return (u64)memcg->soft_limit * PAGE_SIZE; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2833 | default: |
| 2834 | BUG(); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2835 | } |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 2836 | } |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 2837 | |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 2838 | #ifndef CONFIG_SLOB |
Johannes Weiner | 567e9ab | 2016-01-20 15:02:24 -0800 | [diff] [blame] | 2839 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 2840 | { |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 2841 | int memcg_id; |
| 2842 | |
Vladimir Davydov | b313aee | 2016-03-17 14:18:27 -0700 | [diff] [blame] | 2843 | if (cgroup_memory_nokmem) |
| 2844 | return 0; |
| 2845 | |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 2846 | BUG_ON(memcg->kmemcg_id >= 0); |
Johannes Weiner | 567e9ab | 2016-01-20 15:02:24 -0800 | [diff] [blame] | 2847 | BUG_ON(memcg->kmem_state); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 2848 | |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2849 | memcg_id = memcg_alloc_cache_id(); |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 2850 | if (memcg_id < 0) |
| 2851 | return memcg_id; |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 2852 | |
Johannes Weiner | ef12947 | 2016-01-14 15:21:34 -0800 | [diff] [blame] | 2853 | static_branch_inc(&memcg_kmem_enabled_key); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 2854 | /* |
Johannes Weiner | 567e9ab | 2016-01-20 15:02:24 -0800 | [diff] [blame] | 2855 | * A memory cgroup is considered kmem-online as soon as it gets |
Vladimir Davydov | 900a38f | 2014-12-12 16:55:10 -0800 | [diff] [blame] | 2856 | * kmemcg_id. Setting the id after enabling static branching will |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 2857 | * guarantee no one starts accounting before all call sites are |
| 2858 | * patched. |
| 2859 | */ |
Vladimir Davydov | 900a38f | 2014-12-12 16:55:10 -0800 | [diff] [blame] | 2860 | memcg->kmemcg_id = memcg_id; |
Johannes Weiner | 567e9ab | 2016-01-20 15:02:24 -0800 | [diff] [blame] | 2861 | memcg->kmem_state = KMEM_ONLINE; |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 2862 | |
| 2863 | return 0; |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 2864 | } |
| 2865 | |
Johannes Weiner | 8e0a891 | 2016-01-20 15:02:26 -0800 | [diff] [blame] | 2866 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
| 2867 | { |
| 2868 | struct cgroup_subsys_state *css; |
| 2869 | struct mem_cgroup *parent, *child; |
| 2870 | int kmemcg_id; |
| 2871 | |
| 2872 | if (memcg->kmem_state != KMEM_ONLINE) |
| 2873 | return; |
| 2874 | /* |
| 2875 | * Clear the online state before clearing memcg_caches array |
| 2876 | * entries. The slab_mutex in memcg_deactivate_kmem_caches() |
| 2877 | * guarantees that no cache will be created for this cgroup |
| 2878 | * after we are done (see memcg_create_kmem_cache()). |
| 2879 | */ |
| 2880 | memcg->kmem_state = KMEM_ALLOCATED; |
| 2881 | |
| 2882 | memcg_deactivate_kmem_caches(memcg); |
| 2883 | |
| 2884 | kmemcg_id = memcg->kmemcg_id; |
| 2885 | BUG_ON(kmemcg_id < 0); |
| 2886 | |
| 2887 | parent = parent_mem_cgroup(memcg); |
| 2888 | if (!parent) |
| 2889 | parent = root_mem_cgroup; |
| 2890 | |
| 2891 | /* |
| 2892 | * Change kmemcg_id of this cgroup and all its descendants to the |
| 2893 | * parent's id, and then move all entries from this cgroup's list_lrus |
| 2894 | * to ones of the parent. After we have finished, all list_lrus |
| 2895 | * corresponding to this cgroup are guaranteed to remain empty. The |
| 2896 | * ordering is imposed by list_lru_node->lock taken by |
| 2897 | * memcg_drain_all_list_lrus(). |
| 2898 | */ |
Tejun Heo | 3a06bb7 | 2016-06-03 14:55:44 -0700 | [diff] [blame] | 2899 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
Johannes Weiner | 8e0a891 | 2016-01-20 15:02:26 -0800 | [diff] [blame] | 2900 | css_for_each_descendant_pre(css, &memcg->css) { |
| 2901 | child = mem_cgroup_from_css(css); |
| 2902 | BUG_ON(child->kmemcg_id != kmemcg_id); |
| 2903 | child->kmemcg_id = parent->kmemcg_id; |
| 2904 | if (!memcg->use_hierarchy) |
| 2905 | break; |
| 2906 | } |
Tejun Heo | 3a06bb7 | 2016-06-03 14:55:44 -0700 | [diff] [blame] | 2907 | rcu_read_unlock(); |
| 2908 | |
Johannes Weiner | 8e0a891 | 2016-01-20 15:02:26 -0800 | [diff] [blame] | 2909 | memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id); |
| 2910 | |
| 2911 | memcg_free_cache_id(kmemcg_id); |
| 2912 | } |
| 2913 | |
| 2914 | static void memcg_free_kmem(struct mem_cgroup *memcg) |
| 2915 | { |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 2916 | /* css_alloc() failed, offlining didn't happen */ |
| 2917 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) |
| 2918 | memcg_offline_kmem(memcg); |
| 2919 | |
Johannes Weiner | 8e0a891 | 2016-01-20 15:02:26 -0800 | [diff] [blame] | 2920 | if (memcg->kmem_state == KMEM_ALLOCATED) { |
| 2921 | memcg_destroy_kmem_caches(memcg); |
| 2922 | static_branch_dec(&memcg_kmem_enabled_key); |
| 2923 | WARN_ON(page_counter_read(&memcg->kmem)); |
| 2924 | } |
Johannes Weiner | 8e0a891 | 2016-01-20 15:02:26 -0800 | [diff] [blame] | 2925 | } |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 2926 | #else |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 2927 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 2928 | { |
| 2929 | return 0; |
| 2930 | } |
| 2931 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
| 2932 | { |
| 2933 | } |
| 2934 | static void memcg_free_kmem(struct mem_cgroup *memcg) |
| 2935 | { |
| 2936 | } |
| 2937 | #endif /* !CONFIG_SLOB */ |
| 2938 | |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 2939 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
| 2940 | unsigned long limit) |
| 2941 | { |
Vladimir Davydov | b313aee | 2016-03-17 14:18:27 -0700 | [diff] [blame] | 2942 | int ret; |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 2943 | |
| 2944 | mutex_lock(&memcg_limit_mutex); |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 2945 | ret = page_counter_limit(&memcg->kmem, limit); |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 2946 | mutex_unlock(&memcg_limit_mutex); |
| 2947 | return ret; |
| 2948 | } |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 2949 | |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 2950 | static int memcg_update_tcp_limit(struct mem_cgroup *memcg, unsigned long limit) |
| 2951 | { |
| 2952 | int ret; |
| 2953 | |
| 2954 | mutex_lock(&memcg_limit_mutex); |
| 2955 | |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 2956 | ret = page_counter_limit(&memcg->tcpmem, limit); |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 2957 | if (ret) |
| 2958 | goto out; |
| 2959 | |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 2960 | if (!memcg->tcpmem_active) { |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 2961 | /* |
| 2962 | * The active flag needs to be written after the static_key |
| 2963 | * update. This is what guarantees that the socket activation |
| 2964 | * function is the last one to run. See sock_update_memcg() for |
| 2965 | * details, and note that we don't mark any socket as belonging |
| 2966 | * to this memcg until that flag is up. |
| 2967 | * |
| 2968 | * We need to do this, because static_keys will span multiple |
| 2969 | * sites, but we can't control their order. If we mark a socket |
| 2970 | * as accounted, but the accounting functions are not patched in |
| 2971 | * yet, we'll lose accounting. |
| 2972 | * |
| 2973 | * We never race with the readers in sock_update_memcg(), |
| 2974 | * because when this value change, the code to process it is not |
| 2975 | * patched in yet. |
| 2976 | */ |
| 2977 | static_branch_inc(&memcg_sockets_enabled_key); |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 2978 | memcg->tcpmem_active = true; |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 2979 | } |
| 2980 | out: |
| 2981 | mutex_unlock(&memcg_limit_mutex); |
| 2982 | return ret; |
| 2983 | } |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 2984 | |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2985 | /* |
| 2986 | * The user of this function is... |
| 2987 | * RES_LIMIT. |
| 2988 | */ |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 2989 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
| 2990 | char *buf, size_t nbytes, loff_t off) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 2991 | { |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 2992 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2993 | unsigned long nr_pages; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2994 | int ret; |
| 2995 | |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 2996 | buf = strstrip(buf); |
Johannes Weiner | 650c5e5 | 2015-02-11 15:26:03 -0800 | [diff] [blame] | 2997 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2998 | if (ret) |
| 2999 | return ret; |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3000 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3001 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3002 | case RES_LIMIT: |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 3003 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
| 3004 | ret = -EINVAL; |
| 3005 | break; |
| 3006 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3007 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
| 3008 | case _MEM: |
| 3009 | ret = mem_cgroup_resize_limit(memcg, nr_pages); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3010 | break; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3011 | case _MEMSWAP: |
| 3012 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); |
| 3013 | break; |
| 3014 | case _KMEM: |
| 3015 | ret = memcg_update_kmem_limit(memcg, nr_pages); |
| 3016 | break; |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 3017 | case _TCP: |
| 3018 | ret = memcg_update_tcp_limit(memcg, nr_pages); |
| 3019 | break; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3020 | } |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3021 | break; |
Balbir Singh | 296c81d | 2009-09-23 15:56:36 -0700 | [diff] [blame] | 3022 | case RES_SOFT_LIMIT: |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3023 | memcg->soft_limit = nr_pages; |
| 3024 | ret = 0; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3025 | break; |
| 3026 | } |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3027 | return ret ?: nbytes; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3028 | } |
| 3029 | |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3030 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
| 3031 | size_t nbytes, loff_t off) |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 3032 | { |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3033 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3034 | struct page_counter *counter; |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 3035 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3036 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
| 3037 | case _MEM: |
| 3038 | counter = &memcg->memory; |
| 3039 | break; |
| 3040 | case _MEMSWAP: |
| 3041 | counter = &memcg->memsw; |
| 3042 | break; |
| 3043 | case _KMEM: |
| 3044 | counter = &memcg->kmem; |
| 3045 | break; |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 3046 | case _TCP: |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 3047 | counter = &memcg->tcpmem; |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 3048 | break; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3049 | default: |
| 3050 | BUG(); |
| 3051 | } |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3052 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3053 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3054 | case RES_MAX_USAGE: |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3055 | page_counter_reset_watermark(counter); |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3056 | break; |
| 3057 | case RES_FAILCNT: |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3058 | counter->failcnt = 0; |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3059 | break; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3060 | default: |
| 3061 | BUG(); |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3062 | } |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 3063 | |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3064 | return nbytes; |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 3065 | } |
| 3066 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3067 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3068 | struct cftype *cft) |
| 3069 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3070 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3071 | } |
| 3072 | |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3073 | #ifdef CONFIG_MMU |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3074 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3075 | struct cftype *cft, u64 val) |
| 3076 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3077 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3078 | |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 3079 | if (val & ~MOVE_MASK) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3080 | return -EINVAL; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3081 | |
Glauber Costa | ee5e847 | 2013-02-22 16:34:50 -0800 | [diff] [blame] | 3082 | /* |
| 3083 | * No kind of locking is needed in here, because ->can_attach() will |
| 3084 | * check this value once in the beginning of the process, and then carry |
| 3085 | * on with stale data. This means that changes to this value will only |
| 3086 | * affect task migrations starting after the change. |
| 3087 | */ |
| 3088 | memcg->move_charge_at_immigrate = val; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3089 | return 0; |
| 3090 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3091 | #else |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3092 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3093 | struct cftype *cft, u64 val) |
| 3094 | { |
| 3095 | return -ENOSYS; |
| 3096 | } |
| 3097 | #endif |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3098 | |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3099 | #ifdef CONFIG_NUMA |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3100 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3101 | { |
Greg Thelen | 25485de | 2013-11-12 15:07:40 -0800 | [diff] [blame] | 3102 | struct numa_stat { |
| 3103 | const char *name; |
| 3104 | unsigned int lru_mask; |
| 3105 | }; |
| 3106 | |
| 3107 | static const struct numa_stat stats[] = { |
| 3108 | { "total", LRU_ALL }, |
| 3109 | { "file", LRU_ALL_FILE }, |
| 3110 | { "anon", LRU_ALL_ANON }, |
| 3111 | { "unevictable", BIT(LRU_UNEVICTABLE) }, |
| 3112 | }; |
| 3113 | const struct numa_stat *stat; |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3114 | int nid; |
Greg Thelen | 25485de | 2013-11-12 15:07:40 -0800 | [diff] [blame] | 3115 | unsigned long nr; |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3116 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3117 | |
Greg Thelen | 25485de | 2013-11-12 15:07:40 -0800 | [diff] [blame] | 3118 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
| 3119 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); |
| 3120 | seq_printf(m, "%s=%lu", stat->name, nr); |
| 3121 | for_each_node_state(nid, N_MEMORY) { |
| 3122 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
| 3123 | stat->lru_mask); |
| 3124 | seq_printf(m, " N%d=%lu", nid, nr); |
| 3125 | } |
| 3126 | seq_putc(m, '\n'); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3127 | } |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3128 | |
Ying Han | 071aee1 | 2013-11-12 15:07:41 -0800 | [diff] [blame] | 3129 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
| 3130 | struct mem_cgroup *iter; |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3131 | |
Ying Han | 071aee1 | 2013-11-12 15:07:41 -0800 | [diff] [blame] | 3132 | nr = 0; |
| 3133 | for_each_mem_cgroup_tree(iter, memcg) |
| 3134 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); |
| 3135 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); |
| 3136 | for_each_node_state(nid, N_MEMORY) { |
| 3137 | nr = 0; |
| 3138 | for_each_mem_cgroup_tree(iter, memcg) |
| 3139 | nr += mem_cgroup_node_nr_lru_pages( |
| 3140 | iter, nid, stat->lru_mask); |
| 3141 | seq_printf(m, " N%d=%lu", nid, nr); |
| 3142 | } |
| 3143 | seq_putc(m, '\n'); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3144 | } |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3145 | |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3146 | return 0; |
| 3147 | } |
| 3148 | #endif /* CONFIG_NUMA */ |
| 3149 | |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3150 | static int memcg_stat_show(struct seq_file *m, void *v) |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3151 | { |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3152 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3153 | unsigned long memory, memsw; |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3154 | struct mem_cgroup *mi; |
| 3155 | unsigned int i; |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3156 | |
Greg Thelen | 0ca44b1 | 2015-02-11 15:25:58 -0800 | [diff] [blame] | 3157 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) != |
| 3158 | MEM_CGROUP_STAT_NSTATS); |
| 3159 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) != |
| 3160 | MEM_CGROUP_EVENTS_NSTATS); |
Rickard Strandqvist | 70bc068 | 2014-12-12 16:56:41 -0800 | [diff] [blame] | 3161 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
| 3162 | |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3163 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 3164 | if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account()) |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3165 | continue; |
Greg Thelen | 484ebb3 | 2015-10-01 15:37:05 -0700 | [diff] [blame] | 3166 | seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i], |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3167 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3168 | } |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 3169 | |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3170 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
| 3171 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], |
| 3172 | mem_cgroup_read_events(memcg, i)); |
| 3173 | |
| 3174 | for (i = 0; i < NR_LRU_LISTS; i++) |
| 3175 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], |
| 3176 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); |
| 3177 | |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3178 | /* Hierarchical information */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3179 | memory = memsw = PAGE_COUNTER_MAX; |
| 3180 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { |
| 3181 | memory = min(memory, mi->memory.limit); |
| 3182 | memsw = min(memsw, mi->memsw.limit); |
KAMEZAWA Hiroyuki | fee7b54 | 2009-01-07 18:08:26 -0800 | [diff] [blame] | 3183 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3184 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
| 3185 | (u64)memory * PAGE_SIZE); |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 3186 | if (do_memsw_account()) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3187 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
| 3188 | (u64)memsw * PAGE_SIZE); |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3189 | |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3190 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
Greg Thelen | 484ebb3 | 2015-10-01 15:37:05 -0700 | [diff] [blame] | 3191 | unsigned long long val = 0; |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3192 | |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 3193 | if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account()) |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3194 | continue; |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3195 | for_each_mem_cgroup_tree(mi, memcg) |
| 3196 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; |
Greg Thelen | 484ebb3 | 2015-10-01 15:37:05 -0700 | [diff] [blame] | 3197 | seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val); |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3198 | } |
| 3199 | |
| 3200 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
| 3201 | unsigned long long val = 0; |
| 3202 | |
| 3203 | for_each_mem_cgroup_tree(mi, memcg) |
| 3204 | val += mem_cgroup_read_events(mi, i); |
| 3205 | seq_printf(m, "total_%s %llu\n", |
| 3206 | mem_cgroup_events_names[i], val); |
| 3207 | } |
| 3208 | |
| 3209 | for (i = 0; i < NR_LRU_LISTS; i++) { |
| 3210 | unsigned long long val = 0; |
| 3211 | |
| 3212 | for_each_mem_cgroup_tree(mi, memcg) |
| 3213 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; |
| 3214 | 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] | 3215 | } |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3216 | |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3217 | #ifdef CONFIG_DEBUG_VM |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3218 | { |
| 3219 | int nid, zid; |
| 3220 | struct mem_cgroup_per_zone *mz; |
Hugh Dickins | 89abfab | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 3221 | struct zone_reclaim_stat *rstat; |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3222 | unsigned long recent_rotated[2] = {0, 0}; |
| 3223 | unsigned long recent_scanned[2] = {0, 0}; |
| 3224 | |
| 3225 | for_each_online_node(nid) |
| 3226 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 3227 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
Hugh Dickins | 89abfab | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 3228 | rstat = &mz->lruvec.reclaim_stat; |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3229 | |
Hugh Dickins | 89abfab | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 3230 | recent_rotated[0] += rstat->recent_rotated[0]; |
| 3231 | recent_rotated[1] += rstat->recent_rotated[1]; |
| 3232 | recent_scanned[0] += rstat->recent_scanned[0]; |
| 3233 | recent_scanned[1] += rstat->recent_scanned[1]; |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3234 | } |
Johannes Weiner | 78ccf5b | 2012-05-29 15:07:06 -0700 | [diff] [blame] | 3235 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
| 3236 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); |
| 3237 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); |
| 3238 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3239 | } |
| 3240 | #endif |
| 3241 | |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3242 | return 0; |
| 3243 | } |
| 3244 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3245 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
| 3246 | struct cftype *cft) |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3247 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3248 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3249 | |
KAMEZAWA Hiroyuki | 1f4c025 | 2011-07-26 16:08:21 -0700 | [diff] [blame] | 3250 | return mem_cgroup_swappiness(memcg); |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3251 | } |
| 3252 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3253 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
| 3254 | struct cftype *cft, u64 val) |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3255 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3256 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 3257 | |
Johannes Weiner | 3dae7fe | 2014-06-04 16:07:01 -0700 | [diff] [blame] | 3258 | if (val > 100) |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3259 | return -EINVAL; |
| 3260 | |
Linus Torvalds | 14208b0 | 2014-06-09 15:03:33 -0700 | [diff] [blame] | 3261 | if (css->parent) |
Johannes Weiner | 3dae7fe | 2014-06-04 16:07:01 -0700 | [diff] [blame] | 3262 | memcg->swappiness = val; |
| 3263 | else |
| 3264 | vm_swappiness = val; |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 3265 | |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3266 | return 0; |
| 3267 | } |
| 3268 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3269 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
| 3270 | { |
| 3271 | struct mem_cgroup_threshold_ary *t; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3272 | unsigned long usage; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3273 | int i; |
| 3274 | |
| 3275 | rcu_read_lock(); |
| 3276 | if (!swap) |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3277 | t = rcu_dereference(memcg->thresholds.primary); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3278 | else |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3279 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3280 | |
| 3281 | if (!t) |
| 3282 | goto unlock; |
| 3283 | |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3284 | usage = mem_cgroup_usage(memcg, swap); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3285 | |
| 3286 | /* |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3287 | * current_threshold points to threshold just below or equal to usage. |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3288 | * If it's not true, a threshold was crossed after last |
| 3289 | * call of __mem_cgroup_threshold(). |
| 3290 | */ |
Phil Carmody | 5407a56 | 2010-05-26 14:42:42 -0700 | [diff] [blame] | 3291 | i = t->current_threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3292 | |
| 3293 | /* |
| 3294 | * Iterate backward over array of thresholds starting from |
| 3295 | * current_threshold and check if a threshold is crossed. |
| 3296 | * If none of thresholds below usage is crossed, we read |
| 3297 | * only one element of the array here. |
| 3298 | */ |
| 3299 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) |
| 3300 | eventfd_signal(t->entries[i].eventfd, 1); |
| 3301 | |
| 3302 | /* i = current_threshold + 1 */ |
| 3303 | i++; |
| 3304 | |
| 3305 | /* |
| 3306 | * Iterate forward over array of thresholds starting from |
| 3307 | * current_threshold+1 and check if a threshold is crossed. |
| 3308 | * If none of thresholds above usage is crossed, we read |
| 3309 | * only one element of the array here. |
| 3310 | */ |
| 3311 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) |
| 3312 | eventfd_signal(t->entries[i].eventfd, 1); |
| 3313 | |
| 3314 | /* Update current_threshold */ |
Phil Carmody | 5407a56 | 2010-05-26 14:42:42 -0700 | [diff] [blame] | 3315 | t->current_threshold = i - 1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3316 | unlock: |
| 3317 | rcu_read_unlock(); |
| 3318 | } |
| 3319 | |
| 3320 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) |
| 3321 | { |
Kirill A. Shutemov | ad4ca5f | 2010-10-07 12:59:27 -0700 | [diff] [blame] | 3322 | while (memcg) { |
| 3323 | __mem_cgroup_threshold(memcg, false); |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 3324 | if (do_memsw_account()) |
Kirill A. Shutemov | ad4ca5f | 2010-10-07 12:59:27 -0700 | [diff] [blame] | 3325 | __mem_cgroup_threshold(memcg, true); |
| 3326 | |
| 3327 | memcg = parent_mem_cgroup(memcg); |
| 3328 | } |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3329 | } |
| 3330 | |
| 3331 | static int compare_thresholds(const void *a, const void *b) |
| 3332 | { |
| 3333 | const struct mem_cgroup_threshold *_a = a; |
| 3334 | const struct mem_cgroup_threshold *_b = b; |
| 3335 | |
Greg Thelen | 2bff24a | 2013-09-11 14:23:08 -0700 | [diff] [blame] | 3336 | if (_a->threshold > _b->threshold) |
| 3337 | return 1; |
| 3338 | |
| 3339 | if (_a->threshold < _b->threshold) |
| 3340 | return -1; |
| 3341 | |
| 3342 | return 0; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3343 | } |
| 3344 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3345 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3346 | { |
| 3347 | struct mem_cgroup_eventfd_list *ev; |
| 3348 | |
Michal Hocko | 2bcf2e9 | 2014-07-30 16:08:33 -0700 | [diff] [blame] | 3349 | spin_lock(&memcg_oom_lock); |
| 3350 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3351 | list_for_each_entry(ev, &memcg->oom_notify, list) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3352 | eventfd_signal(ev->eventfd, 1); |
Michal Hocko | 2bcf2e9 | 2014-07-30 16:08:33 -0700 | [diff] [blame] | 3353 | |
| 3354 | spin_unlock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3355 | return 0; |
| 3356 | } |
| 3357 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3358 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3359 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3360 | struct mem_cgroup *iter; |
| 3361 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3362 | for_each_mem_cgroup_tree(iter, memcg) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3363 | mem_cgroup_oom_notify_cb(iter); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3364 | } |
| 3365 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3366 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3367 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3368 | { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3369 | struct mem_cgroup_thresholds *thresholds; |
| 3370 | struct mem_cgroup_threshold_ary *new; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3371 | unsigned long threshold; |
| 3372 | unsigned long usage; |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3373 | int i, size, ret; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3374 | |
Johannes Weiner | 650c5e5 | 2015-02-11 15:26:03 -0800 | [diff] [blame] | 3375 | ret = page_counter_memparse(args, "-1", &threshold); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3376 | if (ret) |
| 3377 | return ret; |
| 3378 | |
| 3379 | mutex_lock(&memcg->thresholds_lock); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3380 | |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3381 | if (type == _MEM) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3382 | thresholds = &memcg->thresholds; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3383 | usage = mem_cgroup_usage(memcg, false); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3384 | } else if (type == _MEMSWAP) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3385 | thresholds = &memcg->memsw_thresholds; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3386 | usage = mem_cgroup_usage(memcg, true); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3387 | } else |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3388 | BUG(); |
| 3389 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3390 | /* Check if a threshold crossed before adding a new one */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3391 | if (thresholds->primary) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3392 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
| 3393 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3394 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3395 | |
| 3396 | /* Allocate memory for new array of thresholds */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3397 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3398 | GFP_KERNEL); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3399 | if (!new) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3400 | ret = -ENOMEM; |
| 3401 | goto unlock; |
| 3402 | } |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3403 | new->size = size; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3404 | |
| 3405 | /* Copy thresholds (if any) to new array */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3406 | if (thresholds->primary) { |
| 3407 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3408 | sizeof(struct mem_cgroup_threshold)); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3409 | } |
| 3410 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3411 | /* Add new threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3412 | new->entries[size - 1].eventfd = eventfd; |
| 3413 | new->entries[size - 1].threshold = threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3414 | |
| 3415 | /* Sort thresholds. Registering of new threshold isn't time-critical */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3416 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3417 | compare_thresholds, NULL); |
| 3418 | |
| 3419 | /* Find current threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3420 | new->current_threshold = -1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3421 | for (i = 0; i < size; i++) { |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3422 | if (new->entries[i].threshold <= usage) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3423 | /* |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3424 | * new->current_threshold will not be used until |
| 3425 | * rcu_assign_pointer(), so it's safe to increment |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3426 | * it here. |
| 3427 | */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3428 | ++new->current_threshold; |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3429 | } else |
| 3430 | break; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3431 | } |
| 3432 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3433 | /* Free old spare buffer and save old primary buffer as spare */ |
| 3434 | kfree(thresholds->spare); |
| 3435 | thresholds->spare = thresholds->primary; |
| 3436 | |
| 3437 | rcu_assign_pointer(thresholds->primary, new); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3438 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3439 | /* To be sure that nobody uses thresholds */ |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3440 | synchronize_rcu(); |
| 3441 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3442 | unlock: |
| 3443 | mutex_unlock(&memcg->thresholds_lock); |
| 3444 | |
| 3445 | return ret; |
| 3446 | } |
| 3447 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3448 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3449 | struct eventfd_ctx *eventfd, const char *args) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3450 | { |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3451 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3452 | } |
| 3453 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3454 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3455 | struct eventfd_ctx *eventfd, const char *args) |
| 3456 | { |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3457 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3458 | } |
| 3459 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3460 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3461 | struct eventfd_ctx *eventfd, enum res_type type) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3462 | { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3463 | struct mem_cgroup_thresholds *thresholds; |
| 3464 | struct mem_cgroup_threshold_ary *new; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3465 | unsigned long usage; |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3466 | int i, j, size; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3467 | |
| 3468 | mutex_lock(&memcg->thresholds_lock); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3469 | |
| 3470 | if (type == _MEM) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3471 | thresholds = &memcg->thresholds; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3472 | usage = mem_cgroup_usage(memcg, false); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3473 | } else if (type == _MEMSWAP) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3474 | thresholds = &memcg->memsw_thresholds; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3475 | usage = mem_cgroup_usage(memcg, true); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3476 | } else |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3477 | BUG(); |
| 3478 | |
Anton Vorontsov | 371528c | 2012-02-24 05:14:46 +0400 | [diff] [blame] | 3479 | if (!thresholds->primary) |
| 3480 | goto unlock; |
| 3481 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3482 | /* Check if a threshold crossed before removing */ |
| 3483 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
| 3484 | |
| 3485 | /* Calculate new number of threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3486 | size = 0; |
| 3487 | for (i = 0; i < thresholds->primary->size; i++) { |
| 3488 | if (thresholds->primary->entries[i].eventfd != eventfd) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3489 | size++; |
| 3490 | } |
| 3491 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3492 | new = thresholds->spare; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3493 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3494 | /* Set thresholds array to NULL if we don't have thresholds */ |
| 3495 | if (!size) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3496 | kfree(new); |
| 3497 | new = NULL; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3498 | goto swap_buffers; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3499 | } |
| 3500 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3501 | new->size = size; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3502 | |
| 3503 | /* Copy thresholds and find current threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3504 | new->current_threshold = -1; |
| 3505 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { |
| 3506 | if (thresholds->primary->entries[i].eventfd == eventfd) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3507 | continue; |
| 3508 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3509 | new->entries[j] = thresholds->primary->entries[i]; |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3510 | if (new->entries[j].threshold <= usage) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3511 | /* |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3512 | * new->current_threshold will not be used |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3513 | * until rcu_assign_pointer(), so it's safe to increment |
| 3514 | * it here. |
| 3515 | */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3516 | ++new->current_threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3517 | } |
| 3518 | j++; |
| 3519 | } |
| 3520 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3521 | swap_buffers: |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3522 | /* Swap primary and spare array */ |
| 3523 | thresholds->spare = thresholds->primary; |
Sha Zhengju | 8c75776 | 2012-05-10 13:01:45 -0700 | [diff] [blame] | 3524 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3525 | rcu_assign_pointer(thresholds->primary, new); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3526 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3527 | /* To be sure that nobody uses thresholds */ |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3528 | synchronize_rcu(); |
Martijn Coenen | 6611d8d | 2016-01-15 16:57:49 -0800 | [diff] [blame] | 3529 | |
| 3530 | /* If all events are unregistered, free the spare array */ |
| 3531 | if (!new) { |
| 3532 | kfree(thresholds->spare); |
| 3533 | thresholds->spare = NULL; |
| 3534 | } |
Anton Vorontsov | 371528c | 2012-02-24 05:14:46 +0400 | [diff] [blame] | 3535 | unlock: |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3536 | mutex_unlock(&memcg->thresholds_lock); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3537 | } |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3538 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3539 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3540 | struct eventfd_ctx *eventfd) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3541 | { |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3542 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3543 | } |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3544 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3545 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3546 | struct eventfd_ctx *eventfd) |
| 3547 | { |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3548 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3549 | } |
| 3550 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3551 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3552 | struct eventfd_ctx *eventfd, const char *args) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3553 | { |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3554 | struct mem_cgroup_eventfd_list *event; |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3555 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3556 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
| 3557 | if (!event) |
| 3558 | return -ENOMEM; |
| 3559 | |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 3560 | spin_lock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3561 | |
| 3562 | event->eventfd = eventfd; |
| 3563 | list_add(&event->list, &memcg->oom_notify); |
| 3564 | |
| 3565 | /* already in OOM ? */ |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 3566 | if (memcg->under_oom) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3567 | eventfd_signal(eventfd, 1); |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 3568 | spin_unlock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3569 | |
| 3570 | return 0; |
| 3571 | } |
| 3572 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3573 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3574 | struct eventfd_ctx *eventfd) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3575 | { |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3576 | struct mem_cgroup_eventfd_list *ev, *tmp; |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3577 | |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 3578 | spin_lock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3579 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3580 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3581 | if (ev->eventfd == eventfd) { |
| 3582 | list_del(&ev->list); |
| 3583 | kfree(ev); |
| 3584 | } |
| 3585 | } |
| 3586 | |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 3587 | spin_unlock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3588 | } |
| 3589 | |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3590 | 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] | 3591 | { |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3592 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3593 | |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 3594 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 3595 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3596 | return 0; |
| 3597 | } |
| 3598 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3599 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3600 | struct cftype *cft, u64 val) |
| 3601 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3602 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3603 | |
| 3604 | /* cannot set to root cgroup and only 0 and 1 are allowed */ |
Linus Torvalds | 14208b0 | 2014-06-09 15:03:33 -0700 | [diff] [blame] | 3605 | if (!css->parent || !((val == 0) || (val == 1))) |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3606 | return -EINVAL; |
| 3607 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3608 | memcg->oom_kill_disable = val; |
KAMEZAWA Hiroyuki | 4d845eb | 2010-06-29 15:05:18 -0700 | [diff] [blame] | 3609 | if (!val) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3610 | memcg_oom_recover(memcg); |
Johannes Weiner | 3dae7fe | 2014-06-04 16:07:01 -0700 | [diff] [blame] | 3611 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3612 | return 0; |
| 3613 | } |
| 3614 | |
Tejun Heo | 52ebea7 | 2015-05-22 17:13:37 -0400 | [diff] [blame] | 3615 | #ifdef CONFIG_CGROUP_WRITEBACK |
| 3616 | |
| 3617 | struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg) |
| 3618 | { |
| 3619 | return &memcg->cgwb_list; |
| 3620 | } |
| 3621 | |
Tejun Heo | 841710a | 2015-05-22 18:23:33 -0400 | [diff] [blame] | 3622 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
| 3623 | { |
| 3624 | return wb_domain_init(&memcg->cgwb_domain, gfp); |
| 3625 | } |
| 3626 | |
| 3627 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) |
| 3628 | { |
| 3629 | wb_domain_exit(&memcg->cgwb_domain); |
| 3630 | } |
| 3631 | |
Tejun Heo | 2529bb3 | 2015-05-22 18:23:34 -0400 | [diff] [blame] | 3632 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
| 3633 | { |
| 3634 | wb_domain_size_changed(&memcg->cgwb_domain); |
| 3635 | } |
| 3636 | |
Tejun Heo | 841710a | 2015-05-22 18:23:33 -0400 | [diff] [blame] | 3637 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
| 3638 | { |
| 3639 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); |
| 3640 | |
| 3641 | if (!memcg->css.parent) |
| 3642 | return NULL; |
| 3643 | |
| 3644 | return &memcg->cgwb_domain; |
| 3645 | } |
| 3646 | |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 3647 | /** |
| 3648 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg |
| 3649 | * @wb: bdi_writeback in question |
Tejun Heo | c5edf9c | 2015-09-29 13:04:26 -0400 | [diff] [blame] | 3650 | * @pfilepages: out parameter for number of file pages |
| 3651 | * @pheadroom: out parameter for number of allocatable pages according to memcg |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 3652 | * @pdirty: out parameter for number of dirty pages |
| 3653 | * @pwriteback: out parameter for number of pages under writeback |
| 3654 | * |
Tejun Heo | c5edf9c | 2015-09-29 13:04:26 -0400 | [diff] [blame] | 3655 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
| 3656 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom |
| 3657 | * is a bit more involved. |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 3658 | * |
Tejun Heo | c5edf9c | 2015-09-29 13:04:26 -0400 | [diff] [blame] | 3659 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
| 3660 | * headroom is calculated as the lowest headroom of itself and the |
| 3661 | * ancestors. Note that this doesn't consider the actual amount of |
| 3662 | * available memory in the system. The caller should further cap |
| 3663 | * *@pheadroom accordingly. |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 3664 | */ |
Tejun Heo | c5edf9c | 2015-09-29 13:04:26 -0400 | [diff] [blame] | 3665 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
| 3666 | unsigned long *pheadroom, unsigned long *pdirty, |
| 3667 | unsigned long *pwriteback) |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 3668 | { |
| 3669 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); |
| 3670 | struct mem_cgroup *parent; |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 3671 | |
| 3672 | *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY); |
| 3673 | |
| 3674 | /* this should eventually include NR_UNSTABLE_NFS */ |
| 3675 | *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK); |
Tejun Heo | c5edf9c | 2015-09-29 13:04:26 -0400 | [diff] [blame] | 3676 | *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) | |
| 3677 | (1 << LRU_ACTIVE_FILE)); |
| 3678 | *pheadroom = PAGE_COUNTER_MAX; |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 3679 | |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 3680 | while ((parent = parent_mem_cgroup(memcg))) { |
| 3681 | unsigned long ceiling = min(memcg->memory.limit, memcg->high); |
| 3682 | unsigned long used = page_counter_read(&memcg->memory); |
| 3683 | |
Tejun Heo | c5edf9c | 2015-09-29 13:04:26 -0400 | [diff] [blame] | 3684 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 3685 | memcg = parent; |
| 3686 | } |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 3687 | } |
| 3688 | |
Tejun Heo | 841710a | 2015-05-22 18:23:33 -0400 | [diff] [blame] | 3689 | #else /* CONFIG_CGROUP_WRITEBACK */ |
| 3690 | |
| 3691 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
| 3692 | { |
| 3693 | return 0; |
| 3694 | } |
| 3695 | |
| 3696 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) |
| 3697 | { |
| 3698 | } |
| 3699 | |
Tejun Heo | 2529bb3 | 2015-05-22 18:23:34 -0400 | [diff] [blame] | 3700 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
| 3701 | { |
| 3702 | } |
| 3703 | |
Tejun Heo | 52ebea7 | 2015-05-22 17:13:37 -0400 | [diff] [blame] | 3704 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
| 3705 | |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3706 | /* |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 3707 | * DO NOT USE IN NEW FILES. |
| 3708 | * |
| 3709 | * "cgroup.event_control" implementation. |
| 3710 | * |
| 3711 | * This is way over-engineered. It tries to support fully configurable |
| 3712 | * events for each user. Such level of flexibility is completely |
| 3713 | * unnecessary especially in the light of the planned unified hierarchy. |
| 3714 | * |
| 3715 | * Please deprecate this and replace with something simpler if at all |
| 3716 | * possible. |
| 3717 | */ |
| 3718 | |
| 3719 | /* |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3720 | * Unregister event and free resources. |
| 3721 | * |
| 3722 | * Gets called from workqueue. |
| 3723 | */ |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 3724 | static void memcg_event_remove(struct work_struct *work) |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3725 | { |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 3726 | struct mem_cgroup_event *event = |
| 3727 | container_of(work, struct mem_cgroup_event, remove); |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3728 | struct mem_cgroup *memcg = event->memcg; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3729 | |
| 3730 | remove_wait_queue(event->wqh, &event->wait); |
| 3731 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3732 | event->unregister_event(memcg, event->eventfd); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3733 | |
| 3734 | /* Notify userspace the event is going away. */ |
| 3735 | eventfd_signal(event->eventfd, 1); |
| 3736 | |
| 3737 | eventfd_ctx_put(event->eventfd); |
| 3738 | kfree(event); |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3739 | css_put(&memcg->css); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3740 | } |
| 3741 | |
| 3742 | /* |
| 3743 | * Gets called on POLLHUP on eventfd when user closes it. |
| 3744 | * |
| 3745 | * Called with wqh->lock held and interrupts disabled. |
| 3746 | */ |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 3747 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
| 3748 | int sync, void *key) |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3749 | { |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 3750 | struct mem_cgroup_event *event = |
| 3751 | container_of(wait, struct mem_cgroup_event, wait); |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3752 | struct mem_cgroup *memcg = event->memcg; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3753 | unsigned long flags = (unsigned long)key; |
| 3754 | |
| 3755 | if (flags & POLLHUP) { |
| 3756 | /* |
| 3757 | * If the event has been detached at cgroup removal, we |
| 3758 | * can simply return knowing the other side will cleanup |
| 3759 | * for us. |
| 3760 | * |
| 3761 | * We can't race against event freeing since the other |
| 3762 | * side will require wqh->lock via remove_wait_queue(), |
| 3763 | * which we hold. |
| 3764 | */ |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3765 | spin_lock(&memcg->event_list_lock); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3766 | if (!list_empty(&event->list)) { |
| 3767 | list_del_init(&event->list); |
| 3768 | /* |
| 3769 | * We are in atomic context, but cgroup_event_remove() |
| 3770 | * may sleep, so we have to call it in workqueue. |
| 3771 | */ |
| 3772 | schedule_work(&event->remove); |
| 3773 | } |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3774 | spin_unlock(&memcg->event_list_lock); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3775 | } |
| 3776 | |
| 3777 | return 0; |
| 3778 | } |
| 3779 | |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 3780 | static void memcg_event_ptable_queue_proc(struct file *file, |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3781 | wait_queue_head_t *wqh, poll_table *pt) |
| 3782 | { |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 3783 | struct mem_cgroup_event *event = |
| 3784 | container_of(pt, struct mem_cgroup_event, pt); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3785 | |
| 3786 | event->wqh = wqh; |
| 3787 | add_wait_queue(wqh, &event->wait); |
| 3788 | } |
| 3789 | |
| 3790 | /* |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 3791 | * DO NOT USE IN NEW FILES. |
| 3792 | * |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3793 | * Parse input and register new cgroup event handler. |
| 3794 | * |
| 3795 | * Input must be in format '<event_fd> <control_fd> <args>'. |
| 3796 | * Interpretation of args is defined by control file implementation. |
| 3797 | */ |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3798 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
| 3799 | char *buf, size_t nbytes, loff_t off) |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3800 | { |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3801 | struct cgroup_subsys_state *css = of_css(of); |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3802 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 3803 | struct mem_cgroup_event *event; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3804 | struct cgroup_subsys_state *cfile_css; |
| 3805 | unsigned int efd, cfd; |
| 3806 | struct fd efile; |
| 3807 | struct fd cfile; |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3808 | const char *name; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3809 | char *endp; |
| 3810 | int ret; |
| 3811 | |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3812 | buf = strstrip(buf); |
| 3813 | |
| 3814 | efd = simple_strtoul(buf, &endp, 10); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3815 | if (*endp != ' ') |
| 3816 | return -EINVAL; |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3817 | buf = endp + 1; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3818 | |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3819 | cfd = simple_strtoul(buf, &endp, 10); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3820 | if ((*endp != ' ') && (*endp != '\0')) |
| 3821 | return -EINVAL; |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3822 | buf = endp + 1; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3823 | |
| 3824 | event = kzalloc(sizeof(*event), GFP_KERNEL); |
| 3825 | if (!event) |
| 3826 | return -ENOMEM; |
| 3827 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3828 | event->memcg = memcg; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3829 | INIT_LIST_HEAD(&event->list); |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 3830 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
| 3831 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); |
| 3832 | INIT_WORK(&event->remove, memcg_event_remove); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3833 | |
| 3834 | efile = fdget(efd); |
| 3835 | if (!efile.file) { |
| 3836 | ret = -EBADF; |
| 3837 | goto out_kfree; |
| 3838 | } |
| 3839 | |
| 3840 | event->eventfd = eventfd_ctx_fileget(efile.file); |
| 3841 | if (IS_ERR(event->eventfd)) { |
| 3842 | ret = PTR_ERR(event->eventfd); |
| 3843 | goto out_put_efile; |
| 3844 | } |
| 3845 | |
| 3846 | cfile = fdget(cfd); |
| 3847 | if (!cfile.file) { |
| 3848 | ret = -EBADF; |
| 3849 | goto out_put_eventfd; |
| 3850 | } |
| 3851 | |
| 3852 | /* the process need read permission on control file */ |
| 3853 | /* AV: shouldn't we check that it's been opened for read instead? */ |
| 3854 | ret = inode_permission(file_inode(cfile.file), MAY_READ); |
| 3855 | if (ret < 0) |
| 3856 | goto out_put_cfile; |
| 3857 | |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3858 | /* |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3859 | * Determine the event callbacks and set them in @event. This used |
| 3860 | * to be done via struct cftype but cgroup core no longer knows |
| 3861 | * about these events. The following is crude but the whole thing |
| 3862 | * is for compatibility anyway. |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 3863 | * |
| 3864 | * DO NOT ADD NEW FILES. |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3865 | */ |
Al Viro | b583043 | 2014-10-31 01:22:04 -0400 | [diff] [blame] | 3866 | name = cfile.file->f_path.dentry->d_name.name; |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3867 | |
| 3868 | if (!strcmp(name, "memory.usage_in_bytes")) { |
| 3869 | event->register_event = mem_cgroup_usage_register_event; |
| 3870 | event->unregister_event = mem_cgroup_usage_unregister_event; |
| 3871 | } else if (!strcmp(name, "memory.oom_control")) { |
| 3872 | event->register_event = mem_cgroup_oom_register_event; |
| 3873 | event->unregister_event = mem_cgroup_oom_unregister_event; |
| 3874 | } else if (!strcmp(name, "memory.pressure_level")) { |
| 3875 | event->register_event = vmpressure_register_event; |
| 3876 | event->unregister_event = vmpressure_unregister_event; |
| 3877 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3878 | event->register_event = memsw_cgroup_usage_register_event; |
| 3879 | event->unregister_event = memsw_cgroup_usage_unregister_event; |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3880 | } else { |
| 3881 | ret = -EINVAL; |
| 3882 | goto out_put_cfile; |
| 3883 | } |
| 3884 | |
| 3885 | /* |
Tejun Heo | b5557c4 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3886 | * Verify @cfile should belong to @css. Also, remaining events are |
| 3887 | * automatically removed on cgroup destruction but the removal is |
| 3888 | * asynchronous, so take an extra ref on @css. |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3889 | */ |
Al Viro | b583043 | 2014-10-31 01:22:04 -0400 | [diff] [blame] | 3890 | 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] | 3891 | &memory_cgrp_subsys); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3892 | ret = -EINVAL; |
Tejun Heo | 5a17f54 | 2014-02-11 11:52:47 -0500 | [diff] [blame] | 3893 | if (IS_ERR(cfile_css)) |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3894 | goto out_put_cfile; |
Tejun Heo | 5a17f54 | 2014-02-11 11:52:47 -0500 | [diff] [blame] | 3895 | if (cfile_css != css) { |
| 3896 | css_put(cfile_css); |
| 3897 | goto out_put_cfile; |
| 3898 | } |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3899 | |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3900 | ret = event->register_event(memcg, event->eventfd, buf); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3901 | if (ret) |
| 3902 | goto out_put_css; |
| 3903 | |
| 3904 | efile.file->f_op->poll(efile.file, &event->pt); |
| 3905 | |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3906 | spin_lock(&memcg->event_list_lock); |
| 3907 | list_add(&event->list, &memcg->event_list); |
| 3908 | spin_unlock(&memcg->event_list_lock); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3909 | |
| 3910 | fdput(cfile); |
| 3911 | fdput(efile); |
| 3912 | |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3913 | return nbytes; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3914 | |
| 3915 | out_put_css: |
Tejun Heo | b5557c4 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3916 | css_put(css); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3917 | out_put_cfile: |
| 3918 | fdput(cfile); |
| 3919 | out_put_eventfd: |
| 3920 | eventfd_ctx_put(event->eventfd); |
| 3921 | out_put_efile: |
| 3922 | fdput(efile); |
| 3923 | out_kfree: |
| 3924 | kfree(event); |
| 3925 | |
| 3926 | return ret; |
| 3927 | } |
| 3928 | |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 3929 | static struct cftype mem_cgroup_legacy_files[] = { |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3930 | { |
Balbir Singh | 0eea103 | 2008-02-07 00:13:57 -0800 | [diff] [blame] | 3931 | .name = "usage_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3932 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 3933 | .read_u64 = mem_cgroup_read_u64, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3934 | }, |
| 3935 | { |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 3936 | .name = "max_usage_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3937 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3938 | .write = mem_cgroup_reset, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 3939 | .read_u64 = mem_cgroup_read_u64, |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 3940 | }, |
| 3941 | { |
Balbir Singh | 0eea103 | 2008-02-07 00:13:57 -0800 | [diff] [blame] | 3942 | .name = "limit_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3943 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3944 | .write = mem_cgroup_write, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 3945 | .read_u64 = mem_cgroup_read_u64, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3946 | }, |
| 3947 | { |
Balbir Singh | 296c81d | 2009-09-23 15:56:36 -0700 | [diff] [blame] | 3948 | .name = "soft_limit_in_bytes", |
| 3949 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3950 | .write = mem_cgroup_write, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 3951 | .read_u64 = mem_cgroup_read_u64, |
Balbir Singh | 296c81d | 2009-09-23 15:56:36 -0700 | [diff] [blame] | 3952 | }, |
| 3953 | { |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3954 | .name = "failcnt", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3955 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3956 | .write = mem_cgroup_reset, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 3957 | .read_u64 = mem_cgroup_read_u64, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3958 | }, |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 3959 | { |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3960 | .name = "stat", |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3961 | .seq_show = memcg_stat_show, |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3962 | }, |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3963 | { |
| 3964 | .name = "force_empty", |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3965 | .write = mem_cgroup_force_empty_write, |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3966 | }, |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3967 | { |
| 3968 | .name = "use_hierarchy", |
| 3969 | .write_u64 = mem_cgroup_hierarchy_write, |
| 3970 | .read_u64 = mem_cgroup_hierarchy_read, |
| 3971 | }, |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3972 | { |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 3973 | .name = "cgroup.event_control", /* XXX: for compat */ |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3974 | .write = memcg_write_event_control, |
Tejun Heo | 7dbdb19 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 3975 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3976 | }, |
| 3977 | { |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3978 | .name = "swappiness", |
| 3979 | .read_u64 = mem_cgroup_swappiness_read, |
| 3980 | .write_u64 = mem_cgroup_swappiness_write, |
| 3981 | }, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3982 | { |
| 3983 | .name = "move_charge_at_immigrate", |
| 3984 | .read_u64 = mem_cgroup_move_charge_read, |
| 3985 | .write_u64 = mem_cgroup_move_charge_write, |
| 3986 | }, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3987 | { |
| 3988 | .name = "oom_control", |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3989 | .seq_show = mem_cgroup_oom_control_read, |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3990 | .write_u64 = mem_cgroup_oom_control_write, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3991 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
| 3992 | }, |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 3993 | { |
| 3994 | .name = "pressure_level", |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 3995 | }, |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3996 | #ifdef CONFIG_NUMA |
| 3997 | { |
| 3998 | .name = "numa_stat", |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3999 | .seq_show = memcg_numa_stat_show, |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 4000 | }, |
| 4001 | #endif |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4002 | { |
| 4003 | .name = "kmem.limit_in_bytes", |
| 4004 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4005 | .write = mem_cgroup_write, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4006 | .read_u64 = mem_cgroup_read_u64, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4007 | }, |
| 4008 | { |
| 4009 | .name = "kmem.usage_in_bytes", |
| 4010 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4011 | .read_u64 = mem_cgroup_read_u64, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4012 | }, |
| 4013 | { |
| 4014 | .name = "kmem.failcnt", |
| 4015 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4016 | .write = mem_cgroup_reset, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4017 | .read_u64 = mem_cgroup_read_u64, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4018 | }, |
| 4019 | { |
| 4020 | .name = "kmem.max_usage_in_bytes", |
| 4021 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4022 | .write = mem_cgroup_reset, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4023 | .read_u64 = mem_cgroup_read_u64, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4024 | }, |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 4025 | #ifdef CONFIG_SLABINFO |
| 4026 | { |
| 4027 | .name = "kmem.slabinfo", |
Vladimir Davydov | b047501 | 2014-12-10 15:44:19 -0800 | [diff] [blame] | 4028 | .seq_start = slab_start, |
| 4029 | .seq_next = slab_next, |
| 4030 | .seq_stop = slab_stop, |
| 4031 | .seq_show = memcg_slab_show, |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 4032 | }, |
| 4033 | #endif |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 4034 | { |
| 4035 | .name = "kmem.tcp.limit_in_bytes", |
| 4036 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), |
| 4037 | .write = mem_cgroup_write, |
| 4038 | .read_u64 = mem_cgroup_read_u64, |
| 4039 | }, |
| 4040 | { |
| 4041 | .name = "kmem.tcp.usage_in_bytes", |
| 4042 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), |
| 4043 | .read_u64 = mem_cgroup_read_u64, |
| 4044 | }, |
| 4045 | { |
| 4046 | .name = "kmem.tcp.failcnt", |
| 4047 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), |
| 4048 | .write = mem_cgroup_reset, |
| 4049 | .read_u64 = mem_cgroup_read_u64, |
| 4050 | }, |
| 4051 | { |
| 4052 | .name = "kmem.tcp.max_usage_in_bytes", |
| 4053 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), |
| 4054 | .write = mem_cgroup_reset, |
| 4055 | .read_u64 = mem_cgroup_read_u64, |
| 4056 | }, |
Tejun Heo | 6bc1034 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4057 | { }, /* terminate */ |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4058 | }; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4059 | |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 4060 | /* |
| 4061 | * Private memory cgroup IDR |
| 4062 | * |
| 4063 | * Swap-out records and page cache shadow entries need to store memcg |
| 4064 | * references in constrained space, so we maintain an ID space that is |
| 4065 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of |
| 4066 | * memory-controlled cgroups to 64k. |
| 4067 | * |
| 4068 | * However, there usually are many references to the oflline CSS after |
| 4069 | * the cgroup has been destroyed, such as page cache or reclaimable |
| 4070 | * slab objects, that don't need to hang on to the ID. We want to keep |
| 4071 | * those dead CSS from occupying IDs, or we might quickly exhaust the |
| 4072 | * relatively small ID space and prevent the creation of new cgroups |
| 4073 | * even when there are much fewer than 64k cgroups - possibly none. |
| 4074 | * |
| 4075 | * Maintain a private 16-bit ID space for memcg, and allow the ID to |
| 4076 | * be freed and recycled when it's no longer needed, which is usually |
| 4077 | * when the CSS is offlined. |
| 4078 | * |
| 4079 | * The only exception to that are records of swapped out tmpfs/shmem |
| 4080 | * pages that need to be attributed to live ancestors on swapin. But |
| 4081 | * those references are manageable from userspace. |
| 4082 | */ |
| 4083 | |
| 4084 | static DEFINE_IDR(mem_cgroup_idr); |
| 4085 | |
| 4086 | static void mem_cgroup_id_get(struct mem_cgroup *memcg) |
| 4087 | { |
| 4088 | atomic_inc(&memcg->id.ref); |
| 4089 | } |
| 4090 | |
| 4091 | static void mem_cgroup_id_put(struct mem_cgroup *memcg) |
| 4092 | { |
| 4093 | if (atomic_dec_and_test(&memcg->id.ref)) { |
| 4094 | idr_remove(&mem_cgroup_idr, memcg->id.id); |
| 4095 | memcg->id.id = 0; |
| 4096 | |
| 4097 | /* Memcg ID pins CSS */ |
| 4098 | css_put(&memcg->css); |
| 4099 | } |
| 4100 | } |
| 4101 | |
| 4102 | /** |
| 4103 | * mem_cgroup_from_id - look up a memcg from a memcg id |
| 4104 | * @id: the memcg id to look up |
| 4105 | * |
| 4106 | * Caller must hold rcu_read_lock(). |
| 4107 | */ |
| 4108 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) |
| 4109 | { |
| 4110 | WARN_ON_ONCE(!rcu_read_lock_held()); |
| 4111 | return idr_find(&mem_cgroup_idr, id); |
| 4112 | } |
| 4113 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4114 | 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] | 4115 | { |
| 4116 | struct mem_cgroup_per_node *pn; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4117 | struct mem_cgroup_per_zone *mz; |
KAMEZAWA Hiroyuki | 41e3355 | 2008-04-08 17:41:54 -0700 | [diff] [blame] | 4118 | int zone, tmp = node; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4119 | /* |
| 4120 | * This routine is called against possible nodes. |
| 4121 | * But it's BUG to call kmalloc() against offline node. |
| 4122 | * |
| 4123 | * TODO: this routine can waste much memory for nodes which will |
| 4124 | * never be onlined. It's better to use memory hotplug callback |
| 4125 | * function. |
| 4126 | */ |
KAMEZAWA Hiroyuki | 41e3355 | 2008-04-08 17:41:54 -0700 | [diff] [blame] | 4127 | if (!node_state(node, N_NORMAL_MEMORY)) |
| 4128 | tmp = -1; |
Jesper Juhl | 17295c8 | 2011-01-13 15:47:42 -0800 | [diff] [blame] | 4129 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4130 | if (!pn) |
| 4131 | return 1; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4132 | |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4133 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
| 4134 | mz = &pn->zoneinfo[zone]; |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 4135 | lruvec_init(&mz->lruvec); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 4136 | mz->usage_in_excess = 0; |
| 4137 | mz->on_tree = false; |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4138 | mz->memcg = memcg; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4139 | } |
Johannes Weiner | 54f72fe | 2013-07-08 15:59:49 -0700 | [diff] [blame] | 4140 | memcg->nodeinfo[node] = pn; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4141 | return 0; |
| 4142 | } |
| 4143 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4144 | 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] | 4145 | { |
Johannes Weiner | 54f72fe | 2013-07-08 15:59:49 -0700 | [diff] [blame] | 4146 | kfree(memcg->nodeinfo[node]); |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4147 | } |
| 4148 | |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4149 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
| 4150 | { |
| 4151 | int node; |
| 4152 | |
| 4153 | memcg_wb_domain_exit(memcg); |
| 4154 | for_each_node(node) |
| 4155 | free_mem_cgroup_per_zone_info(memcg, node); |
| 4156 | free_percpu(memcg->stat); |
| 4157 | kfree(memcg); |
| 4158 | } |
| 4159 | |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4160 | static struct mem_cgroup *mem_cgroup_alloc(void) |
| 4161 | { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4162 | struct mem_cgroup *memcg; |
Vladimir Davydov | 8ff69e2 | 2014-01-23 15:52:52 -0800 | [diff] [blame] | 4163 | size_t size; |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4164 | int node; |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4165 | |
Vladimir Davydov | 8ff69e2 | 2014-01-23 15:52:52 -0800 | [diff] [blame] | 4166 | size = sizeof(struct mem_cgroup); |
| 4167 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4168 | |
Vladimir Davydov | 8ff69e2 | 2014-01-23 15:52:52 -0800 | [diff] [blame] | 4169 | memcg = kzalloc(size, GFP_KERNEL); |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4170 | if (!memcg) |
Dan Carpenter | e7bbcdf | 2010-03-23 13:35:12 -0700 | [diff] [blame] | 4171 | return NULL; |
| 4172 | |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 4173 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
| 4174 | 1, MEM_CGROUP_ID_MAX, |
| 4175 | GFP_KERNEL); |
| 4176 | if (memcg->id.id < 0) |
| 4177 | goto fail; |
| 4178 | |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4179 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
| 4180 | if (!memcg->stat) |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4181 | goto fail; |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 4182 | |
Bob Liu | 3ed28fa | 2012-01-12 17:19:04 -0800 | [diff] [blame] | 4183 | for_each_node(node) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4184 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4185 | goto fail; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 4186 | |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4187 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
| 4188 | goto fail; |
Balbir Singh | 28dbc4b | 2009-01-07 18:08:05 -0800 | [diff] [blame] | 4189 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 4190 | INIT_WORK(&memcg->high_work, high_work_func); |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4191 | memcg->last_scanned_node = MAX_NUMNODES; |
| 4192 | INIT_LIST_HEAD(&memcg->oom_notify); |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4193 | mutex_init(&memcg->thresholds_lock); |
| 4194 | spin_lock_init(&memcg->move_lock); |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 4195 | vmpressure_init(&memcg->vmpressure); |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4196 | INIT_LIST_HEAD(&memcg->event_list); |
| 4197 | spin_lock_init(&memcg->event_list_lock); |
Johannes Weiner | d886f4e | 2016-01-20 15:02:47 -0800 | [diff] [blame] | 4198 | memcg->socket_pressure = jiffies; |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 4199 | #ifndef CONFIG_SLOB |
Vladimir Davydov | 900a38f | 2014-12-12 16:55:10 -0800 | [diff] [blame] | 4200 | memcg->kmemcg_id = -1; |
Vladimir Davydov | 900a38f | 2014-12-12 16:55:10 -0800 | [diff] [blame] | 4201 | #endif |
Tejun Heo | 52ebea7 | 2015-05-22 17:13:37 -0400 | [diff] [blame] | 4202 | #ifdef CONFIG_CGROUP_WRITEBACK |
| 4203 | INIT_LIST_HEAD(&memcg->cgwb_list); |
| 4204 | #endif |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 4205 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4206 | return memcg; |
| 4207 | fail: |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 4208 | if (memcg->id.id > 0) |
| 4209 | idr_remove(&mem_cgroup_idr, memcg->id.id); |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4210 | mem_cgroup_free(memcg); |
| 4211 | return NULL; |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4212 | } |
| 4213 | |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4214 | static struct cgroup_subsys_state * __ref |
| 4215 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4216 | { |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4217 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
| 4218 | struct mem_cgroup *memcg; |
| 4219 | long error = -ENOMEM; |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4220 | |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4221 | memcg = mem_cgroup_alloc(); |
| 4222 | if (!memcg) |
| 4223 | return ERR_PTR(error); |
Li Zefan | 4219b2d | 2013-09-23 16:56:29 +0800 | [diff] [blame] | 4224 | |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4225 | memcg->high = PAGE_COUNTER_MAX; |
| 4226 | memcg->soft_limit = PAGE_COUNTER_MAX; |
| 4227 | if (parent) { |
| 4228 | memcg->swappiness = mem_cgroup_swappiness(parent); |
| 4229 | memcg->oom_kill_disable = parent->oom_kill_disable; |
| 4230 | } |
| 4231 | if (parent && parent->use_hierarchy) { |
| 4232 | memcg->use_hierarchy = true; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4233 | page_counter_init(&memcg->memory, &parent->memory); |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 4234 | page_counter_init(&memcg->swap, &parent->swap); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4235 | page_counter_init(&memcg->memsw, &parent->memsw); |
| 4236 | page_counter_init(&memcg->kmem, &parent->kmem); |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 4237 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4238 | } else { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4239 | page_counter_init(&memcg->memory, NULL); |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 4240 | page_counter_init(&memcg->swap, NULL); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4241 | page_counter_init(&memcg->memsw, NULL); |
| 4242 | page_counter_init(&memcg->kmem, NULL); |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 4243 | page_counter_init(&memcg->tcpmem, NULL); |
Tejun Heo | 8c7f6ed | 2012-09-13 12:20:58 -0700 | [diff] [blame] | 4244 | /* |
| 4245 | * Deeper hierachy with use_hierarchy == false doesn't make |
| 4246 | * much sense so let cgroup subsystem know about this |
| 4247 | * unfortunate state in our controller. |
| 4248 | */ |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4249 | if (parent != root_mem_cgroup) |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 4250 | memory_cgrp_subsys.broken_hierarchy = true; |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4251 | } |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 4252 | |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4253 | /* The following stuff does not apply to the root */ |
| 4254 | if (!parent) { |
| 4255 | root_mem_cgroup = memcg; |
| 4256 | return &memcg->css; |
| 4257 | } |
| 4258 | |
Vladimir Davydov | b313aee | 2016-03-17 14:18:27 -0700 | [diff] [blame] | 4259 | error = memcg_online_kmem(memcg); |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4260 | if (error) |
| 4261 | goto fail; |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 4262 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 4263 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
Johannes Weiner | ef12947 | 2016-01-14 15:21:34 -0800 | [diff] [blame] | 4264 | static_branch_inc(&memcg_sockets_enabled_key); |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 4265 | |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4266 | return &memcg->css; |
| 4267 | fail: |
| 4268 | mem_cgroup_free(memcg); |
Tejun Heo | ea3a964 | 2016-06-24 14:49:58 -0700 | [diff] [blame] | 4269 | return ERR_PTR(-ENOMEM); |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4270 | } |
| 4271 | |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 4272 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4273 | { |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 4274 | /* Online state pins memcg ID, memcg ID pins CSS */ |
| 4275 | mem_cgroup_id_get(mem_cgroup_from_css(css)); |
| 4276 | css_get(css); |
Johannes Weiner | 2f7dd7a | 2014-10-02 16:16:57 -0700 | [diff] [blame] | 4277 | return 0; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4278 | } |
| 4279 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4280 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
KAMEZAWA Hiroyuki | df878fb | 2008-02-07 00:14:28 -0800 | [diff] [blame] | 4281 | { |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4282 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4283 | struct mem_cgroup_event *event, *tmp; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4284 | |
| 4285 | /* |
| 4286 | * Unregister events and notify userspace. |
| 4287 | * Notify userspace about cgroup removing only after rmdir of cgroup |
| 4288 | * directory to avoid race between userspace and kernelspace. |
| 4289 | */ |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4290 | spin_lock(&memcg->event_list_lock); |
| 4291 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4292 | list_del_init(&event->list); |
| 4293 | schedule_work(&event->remove); |
| 4294 | } |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4295 | spin_unlock(&memcg->event_list_lock); |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 4296 | |
Johannes Weiner | 567e9ab | 2016-01-20 15:02:24 -0800 | [diff] [blame] | 4297 | memcg_offline_kmem(memcg); |
Tejun Heo | 52ebea7 | 2015-05-22 17:13:37 -0400 | [diff] [blame] | 4298 | wb_memcg_offline(memcg); |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 4299 | |
| 4300 | mem_cgroup_id_put(memcg); |
KAMEZAWA Hiroyuki | df878fb | 2008-02-07 00:14:28 -0800 | [diff] [blame] | 4301 | } |
| 4302 | |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 4303 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
| 4304 | { |
| 4305 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 4306 | |
| 4307 | invalidate_reclaim_iterators(memcg); |
| 4308 | } |
| 4309 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4310 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4311 | { |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4312 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Daisuke Nishimura | c268e99 | 2009-01-15 13:51:13 -0800 | [diff] [blame] | 4313 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 4314 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
Johannes Weiner | ef12947 | 2016-01-14 15:21:34 -0800 | [diff] [blame] | 4315 | static_branch_dec(&memcg_sockets_enabled_key); |
Johannes Weiner | 3893e30 | 2016-01-20 15:02:29 -0800 | [diff] [blame] | 4316 | |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 4317 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 4318 | static_branch_dec(&memcg_sockets_enabled_key); |
Johannes Weiner | 3893e30 | 2016-01-20 15:02:29 -0800 | [diff] [blame] | 4319 | |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4320 | vmpressure_cleanup(&memcg->vmpressure); |
| 4321 | cancel_work_sync(&memcg->high_work); |
| 4322 | mem_cgroup_remove_from_trees(memcg); |
Johannes Weiner | d886f4e | 2016-01-20 15:02:47 -0800 | [diff] [blame] | 4323 | memcg_free_kmem(memcg); |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4324 | mem_cgroup_free(memcg); |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4325 | } |
| 4326 | |
Tejun Heo | 1ced953 | 2014-07-08 18:02:57 -0400 | [diff] [blame] | 4327 | /** |
| 4328 | * mem_cgroup_css_reset - reset the states of a mem_cgroup |
| 4329 | * @css: the target css |
| 4330 | * |
| 4331 | * Reset the states of the mem_cgroup associated with @css. This is |
| 4332 | * invoked when the userland requests disabling on the default hierarchy |
| 4333 | * but the memcg is pinned through dependency. The memcg should stop |
| 4334 | * applying policies and should revert to the vanilla state as it may be |
| 4335 | * made visible again. |
| 4336 | * |
| 4337 | * The current implementation only resets the essential configurations. |
| 4338 | * This needs to be expanded to cover all the visible parts. |
| 4339 | */ |
| 4340 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) |
| 4341 | { |
| 4342 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 4343 | |
Vladimir Davydov | d334c9b | 2016-03-17 14:19:38 -0700 | [diff] [blame] | 4344 | page_counter_limit(&memcg->memory, PAGE_COUNTER_MAX); |
| 4345 | page_counter_limit(&memcg->swap, PAGE_COUNTER_MAX); |
| 4346 | page_counter_limit(&memcg->memsw, PAGE_COUNTER_MAX); |
| 4347 | page_counter_limit(&memcg->kmem, PAGE_COUNTER_MAX); |
| 4348 | page_counter_limit(&memcg->tcpmem, PAGE_COUNTER_MAX); |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 4349 | memcg->low = 0; |
| 4350 | memcg->high = PAGE_COUNTER_MAX; |
Johannes Weiner | 24d404d | 2015-01-08 14:32:35 -0800 | [diff] [blame] | 4351 | memcg->soft_limit = PAGE_COUNTER_MAX; |
Tejun Heo | 2529bb3 | 2015-05-22 18:23:34 -0400 | [diff] [blame] | 4352 | memcg_wb_domain_size_changed(memcg); |
Tejun Heo | 1ced953 | 2014-07-08 18:02:57 -0400 | [diff] [blame] | 4353 | } |
| 4354 | |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4355 | #ifdef CONFIG_MMU |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4356 | /* Handlers for move charge at task migration. */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4357 | static int mem_cgroup_do_precharge(unsigned long count) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4358 | { |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 4359 | int ret; |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 4360 | |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 4361 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
| 4362 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 4363 | if (!ret) { |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4364 | mc.precharge += count; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4365 | return ret; |
| 4366 | } |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 4367 | |
| 4368 | /* Try charges one by one with reclaim */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4369 | while (count--) { |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 4370 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1); |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 4371 | if (ret) |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 4372 | return ret; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4373 | mc.precharge++; |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 4374 | cond_resched(); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4375 | } |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 4376 | return 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4377 | } |
| 4378 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4379 | union mc_target { |
| 4380 | struct page *page; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4381 | swp_entry_t ent; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4382 | }; |
| 4383 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4384 | enum mc_target_type { |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 4385 | MC_TARGET_NONE = 0, |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4386 | MC_TARGET_PAGE, |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4387 | MC_TARGET_SWAP, |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4388 | }; |
| 4389 | |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4390 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
| 4391 | unsigned long addr, pte_t ptent) |
| 4392 | { |
| 4393 | struct page *page = vm_normal_page(vma, addr, ptent); |
| 4394 | |
| 4395 | if (!page || !page_mapped(page)) |
| 4396 | return NULL; |
| 4397 | if (PageAnon(page)) { |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 4398 | if (!(mc.flags & MOVE_ANON)) |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4399 | return NULL; |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 4400 | } else { |
| 4401 | if (!(mc.flags & MOVE_FILE)) |
| 4402 | return NULL; |
| 4403 | } |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4404 | if (!get_page_unless_zero(page)) |
| 4405 | return NULL; |
| 4406 | |
| 4407 | return page; |
| 4408 | } |
| 4409 | |
KAMEZAWA Hiroyuki | 4b91355 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 4410 | #ifdef CONFIG_SWAP |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4411 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
| 4412 | unsigned long addr, pte_t ptent, swp_entry_t *entry) |
| 4413 | { |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4414 | struct page *page = NULL; |
| 4415 | swp_entry_t ent = pte_to_swp_entry(ptent); |
| 4416 | |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 4417 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4418 | return NULL; |
KAMEZAWA Hiroyuki | 4b91355 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 4419 | /* |
| 4420 | * Because lookup_swap_cache() updates some statistics counter, |
| 4421 | * we call find_get_page() with swapper_space directly. |
| 4422 | */ |
Shaohua Li | 33806f0 | 2013-02-22 16:34:37 -0800 | [diff] [blame] | 4423 | page = find_get_page(swap_address_space(ent), ent.val); |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 4424 | if (do_memsw_account()) |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4425 | entry->val = ent.val; |
| 4426 | |
| 4427 | return page; |
| 4428 | } |
KAMEZAWA Hiroyuki | 4b91355 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 4429 | #else |
| 4430 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
| 4431 | unsigned long addr, pte_t ptent, swp_entry_t *entry) |
| 4432 | { |
| 4433 | return NULL; |
| 4434 | } |
| 4435 | #endif |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4436 | |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4437 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
| 4438 | unsigned long addr, pte_t ptent, swp_entry_t *entry) |
| 4439 | { |
| 4440 | struct page *page = NULL; |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4441 | struct address_space *mapping; |
| 4442 | pgoff_t pgoff; |
| 4443 | |
| 4444 | if (!vma->vm_file) /* anonymous vma */ |
| 4445 | return NULL; |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 4446 | if (!(mc.flags & MOVE_FILE)) |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4447 | return NULL; |
| 4448 | |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4449 | mapping = vma->vm_file->f_mapping; |
Kirill A. Shutemov | 0661a33 | 2015-02-10 14:10:04 -0800 | [diff] [blame] | 4450 | pgoff = linear_page_index(vma, addr); |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4451 | |
| 4452 | /* 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] | 4453 | #ifdef CONFIG_SWAP |
| 4454 | /* shmem/tmpfs may report page out on swap: account for that too. */ |
Johannes Weiner | 139b6a6 | 2014-05-06 12:50:05 -0700 | [diff] [blame] | 4455 | if (shmem_mapping(mapping)) { |
| 4456 | page = find_get_entry(mapping, pgoff); |
| 4457 | if (radix_tree_exceptional_entry(page)) { |
| 4458 | swp_entry_t swp = radix_to_swp_entry(page); |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 4459 | if (do_memsw_account()) |
Johannes Weiner | 139b6a6 | 2014-05-06 12:50:05 -0700 | [diff] [blame] | 4460 | *entry = swp; |
| 4461 | page = find_get_page(swap_address_space(swp), swp.val); |
| 4462 | } |
| 4463 | } else |
| 4464 | page = find_get_page(mapping, pgoff); |
| 4465 | #else |
| 4466 | page = find_get_page(mapping, pgoff); |
Hugh Dickins | aa3b189 | 2011-08-03 16:21:24 -0700 | [diff] [blame] | 4467 | #endif |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4468 | return page; |
| 4469 | } |
| 4470 | |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4471 | /** |
| 4472 | * mem_cgroup_move_account - move account of the page |
| 4473 | * @page: the page |
| 4474 | * @nr_pages: number of regular pages (>1 for huge pages) |
| 4475 | * @from: mem_cgroup which the page is moved from. |
| 4476 | * @to: mem_cgroup which the page is moved to. @from != @to. |
| 4477 | * |
Kirill A. Shutemov | 3ac808f | 2016-01-15 16:53:07 -0800 | [diff] [blame] | 4478 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4479 | * |
| 4480 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
| 4481 | * from old cgroup. |
| 4482 | */ |
| 4483 | static int mem_cgroup_move_account(struct page *page, |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 4484 | bool compound, |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4485 | struct mem_cgroup *from, |
| 4486 | struct mem_cgroup *to) |
| 4487 | { |
| 4488 | unsigned long flags; |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 4489 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4490 | int ret; |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 4491 | bool anon; |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4492 | |
| 4493 | VM_BUG_ON(from == to); |
| 4494 | VM_BUG_ON_PAGE(PageLRU(page), page); |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 4495 | VM_BUG_ON(compound && !PageTransHuge(page)); |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4496 | |
| 4497 | /* |
Johannes Weiner | 6a93ca8 | 2016-03-15 14:57:19 -0700 | [diff] [blame] | 4498 | * Prevent mem_cgroup_migrate() from looking at |
Hugh Dickins | 45637ba | 2015-11-05 18:49:40 -0800 | [diff] [blame] | 4499 | * page->mem_cgroup of its source page while we change it. |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4500 | */ |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 4501 | ret = -EBUSY; |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4502 | if (!trylock_page(page)) |
| 4503 | goto out; |
| 4504 | |
| 4505 | ret = -EINVAL; |
| 4506 | if (page->mem_cgroup != from) |
| 4507 | goto out_unlock; |
| 4508 | |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 4509 | anon = PageAnon(page); |
| 4510 | |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4511 | spin_lock_irqsave(&from->move_lock, flags); |
| 4512 | |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 4513 | if (!anon && page_mapped(page)) { |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4514 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
| 4515 | nr_pages); |
| 4516 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
| 4517 | nr_pages); |
| 4518 | } |
| 4519 | |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 4520 | /* |
| 4521 | * move_lock grabbed above and caller set from->moving_account, so |
| 4522 | * mem_cgroup_update_page_stat() will serialize updates to PageDirty. |
| 4523 | * So mapping should be stable for dirty pages. |
| 4524 | */ |
| 4525 | if (!anon && PageDirty(page)) { |
| 4526 | struct address_space *mapping = page_mapping(page); |
| 4527 | |
| 4528 | if (mapping_cap_account_dirty(mapping)) { |
| 4529 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY], |
| 4530 | nr_pages); |
| 4531 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY], |
| 4532 | nr_pages); |
| 4533 | } |
| 4534 | } |
| 4535 | |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4536 | if (PageWriteback(page)) { |
| 4537 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], |
| 4538 | nr_pages); |
| 4539 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], |
| 4540 | nr_pages); |
| 4541 | } |
| 4542 | |
| 4543 | /* |
| 4544 | * It is safe to change page->mem_cgroup here because the page |
| 4545 | * is referenced, charged, and isolated - we can't race with |
| 4546 | * uncharging, charging, migration, or LRU putback. |
| 4547 | */ |
| 4548 | |
| 4549 | /* caller should have done css_get */ |
| 4550 | page->mem_cgroup = to; |
| 4551 | spin_unlock_irqrestore(&from->move_lock, flags); |
| 4552 | |
| 4553 | ret = 0; |
| 4554 | |
| 4555 | local_irq_disable(); |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 4556 | mem_cgroup_charge_statistics(to, page, compound, nr_pages); |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4557 | memcg_check_events(to, page); |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 4558 | mem_cgroup_charge_statistics(from, page, compound, -nr_pages); |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4559 | memcg_check_events(from, page); |
| 4560 | local_irq_enable(); |
| 4561 | out_unlock: |
| 4562 | unlock_page(page); |
| 4563 | out: |
| 4564 | return ret; |
| 4565 | } |
| 4566 | |
Li RongQing | 7cf7806 | 2016-05-27 14:27:46 -0700 | [diff] [blame] | 4567 | /** |
| 4568 | * get_mctgt_type - get target type of moving charge |
| 4569 | * @vma: the vma the pte to be checked belongs |
| 4570 | * @addr: the address corresponding to the pte to be checked |
| 4571 | * @ptent: the pte to be checked |
| 4572 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
| 4573 | * |
| 4574 | * Returns |
| 4575 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. |
| 4576 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for |
| 4577 | * move charge. if @target is not NULL, the page is stored in target->page |
| 4578 | * with extra refcnt got(Callers should handle it). |
| 4579 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
| 4580 | * target for charge migration. if @target is not NULL, the entry is stored |
| 4581 | * in target->ent. |
| 4582 | * |
| 4583 | * Called with pte lock held. |
| 4584 | */ |
| 4585 | |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 4586 | 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] | 4587 | unsigned long addr, pte_t ptent, union mc_target *target) |
| 4588 | { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4589 | struct page *page = NULL; |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 4590 | enum mc_target_type ret = MC_TARGET_NONE; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4591 | swp_entry_t ent = { .val = 0 }; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4592 | |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4593 | if (pte_present(ptent)) |
| 4594 | page = mc_handle_present_pte(vma, addr, ptent); |
| 4595 | else if (is_swap_pte(ptent)) |
| 4596 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); |
Kirill A. Shutemov | 0661a33 | 2015-02-10 14:10:04 -0800 | [diff] [blame] | 4597 | else if (pte_none(ptent)) |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4598 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4599 | |
| 4600 | if (!page && !ent.val) |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 4601 | return ret; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4602 | if (page) { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4603 | /* |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 4604 | * Do only loose check w/o serialization. |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 4605 | * mem_cgroup_move_account() checks the page is valid or |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 4606 | * not under LRU exclusion. |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4607 | */ |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 4608 | if (page->mem_cgroup == mc.from) { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4609 | ret = MC_TARGET_PAGE; |
| 4610 | if (target) |
| 4611 | target->page = page; |
| 4612 | } |
| 4613 | if (!ret || !target) |
| 4614 | put_page(page); |
| 4615 | } |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4616 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
| 4617 | if (ent.val && !ret && |
Li Zefan | 34c00c3 | 2013-09-23 16:56:01 +0800 | [diff] [blame] | 4618 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
KAMEZAWA Hiroyuki | 7f0f154 | 2010-05-11 14:06:58 -0700 | [diff] [blame] | 4619 | ret = MC_TARGET_SWAP; |
| 4620 | if (target) |
| 4621 | target->ent = ent; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4622 | } |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4623 | return ret; |
| 4624 | } |
| 4625 | |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4626 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 4627 | /* |
| 4628 | * We don't consider swapping or file mapped pages because THP does not |
| 4629 | * support them for now. |
| 4630 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
| 4631 | */ |
| 4632 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, |
| 4633 | unsigned long addr, pmd_t pmd, union mc_target *target) |
| 4634 | { |
| 4635 | struct page *page = NULL; |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4636 | enum mc_target_type ret = MC_TARGET_NONE; |
| 4637 | |
| 4638 | page = pmd_page(pmd); |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 4639 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 4640 | if (!(mc.flags & MOVE_ANON)) |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4641 | return ret; |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 4642 | if (page->mem_cgroup == mc.from) { |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4643 | ret = MC_TARGET_PAGE; |
| 4644 | if (target) { |
| 4645 | get_page(page); |
| 4646 | target->page = page; |
| 4647 | } |
| 4648 | } |
| 4649 | return ret; |
| 4650 | } |
| 4651 | #else |
| 4652 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, |
| 4653 | unsigned long addr, pmd_t pmd, union mc_target *target) |
| 4654 | { |
| 4655 | return MC_TARGET_NONE; |
| 4656 | } |
| 4657 | #endif |
| 4658 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4659 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
| 4660 | unsigned long addr, unsigned long end, |
| 4661 | struct mm_walk *walk) |
| 4662 | { |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 4663 | struct vm_area_struct *vma = walk->vma; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4664 | pte_t *pte; |
| 4665 | spinlock_t *ptl; |
| 4666 | |
Kirill A. Shutemov | b6ec57f | 2016-01-21 16:40:25 -0800 | [diff] [blame] | 4667 | ptl = pmd_trans_huge_lock(pmd, vma); |
| 4668 | if (ptl) { |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4669 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
| 4670 | mc.precharge += HPAGE_PMD_NR; |
Kirill A. Shutemov | bf92915 | 2013-11-14 14:30:54 -0800 | [diff] [blame] | 4671 | spin_unlock(ptl); |
Andrea Arcangeli | 1a5a990 | 2012-03-21 16:33:42 -0700 | [diff] [blame] | 4672 | return 0; |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4673 | } |
Dave Hansen | 0331932 | 2011-03-22 16:32:56 -0700 | [diff] [blame] | 4674 | |
Andrea Arcangeli | 45f83ce | 2012-03-28 14:42:40 -0700 | [diff] [blame] | 4675 | if (pmd_trans_unstable(pmd)) |
| 4676 | return 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4677 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
| 4678 | for (; addr != end; pte++, addr += PAGE_SIZE) |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 4679 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4680 | mc.precharge++; /* increment precharge temporarily */ |
| 4681 | pte_unmap_unlock(pte - 1, ptl); |
| 4682 | cond_resched(); |
| 4683 | |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4684 | return 0; |
| 4685 | } |
| 4686 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4687 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
| 4688 | { |
| 4689 | unsigned long precharge; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4690 | |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 4691 | struct mm_walk mem_cgroup_count_precharge_walk = { |
| 4692 | .pmd_entry = mem_cgroup_count_precharge_pte_range, |
| 4693 | .mm = mm, |
| 4694 | }; |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4695 | down_read(&mm->mmap_sem); |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 4696 | walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4697 | up_read(&mm->mmap_sem); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4698 | |
| 4699 | precharge = mc.precharge; |
| 4700 | mc.precharge = 0; |
| 4701 | |
| 4702 | return precharge; |
| 4703 | } |
| 4704 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4705 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
| 4706 | { |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4707 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
| 4708 | |
| 4709 | VM_BUG_ON(mc.moving_task); |
| 4710 | mc.moving_task = current; |
| 4711 | return mem_cgroup_do_precharge(precharge); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4712 | } |
| 4713 | |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4714 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
| 4715 | static void __mem_cgroup_clear_mc(void) |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4716 | { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 4717 | struct mem_cgroup *from = mc.from; |
| 4718 | struct mem_cgroup *to = mc.to; |
| 4719 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4720 | /* we must uncharge all the leftover precharges from mc.to */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4721 | if (mc.precharge) { |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 4722 | cancel_charge(mc.to, mc.precharge); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4723 | mc.precharge = 0; |
| 4724 | } |
| 4725 | /* |
| 4726 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so |
| 4727 | * we must uncharge here. |
| 4728 | */ |
| 4729 | if (mc.moved_charge) { |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 4730 | cancel_charge(mc.from, mc.moved_charge); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4731 | mc.moved_charge = 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4732 | } |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4733 | /* we must fixup refcnts and charges */ |
| 4734 | if (mc.moved_swap) { |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4735 | /* uncharge swap account from the old cgroup */ |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 4736 | if (!mem_cgroup_is_root(mc.from)) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4737 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4738 | |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 4739 | /* |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4740 | * we charged both to->memory and to->memsw, so we |
| 4741 | * should uncharge to->memory. |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 4742 | */ |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 4743 | if (!mem_cgroup_is_root(mc.to)) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4744 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4745 | |
Johannes Weiner | e8ea14c | 2014-12-10 15:42:42 -0800 | [diff] [blame] | 4746 | css_put_many(&mc.from->css, mc.moved_swap); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4747 | |
Li Zefan | 4050377 | 2013-07-08 16:00:34 -0700 | [diff] [blame] | 4748 | /* we've already done css_get(mc.to) */ |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4749 | mc.moved_swap = 0; |
| 4750 | } |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4751 | memcg_oom_recover(from); |
| 4752 | memcg_oom_recover(to); |
| 4753 | wake_up_all(&mc.waitq); |
| 4754 | } |
| 4755 | |
| 4756 | static void mem_cgroup_clear_mc(void) |
| 4757 | { |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 4758 | struct mm_struct *mm = mc.mm; |
| 4759 | |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4760 | /* |
| 4761 | * we must clear moving_task before waking up waiters at the end of |
| 4762 | * task migration. |
| 4763 | */ |
| 4764 | mc.moving_task = NULL; |
| 4765 | __mem_cgroup_clear_mc(); |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 4766 | spin_lock(&mc.lock); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4767 | mc.from = NULL; |
| 4768 | mc.to = NULL; |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 4769 | mc.mm = NULL; |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 4770 | spin_unlock(&mc.lock); |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 4771 | |
| 4772 | mmput(mm); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4773 | } |
| 4774 | |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 4775 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4776 | { |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 4777 | struct cgroup_subsys_state *css; |
Ross Zwisler | eed67d7 | 2015-12-23 14:53:27 -0700 | [diff] [blame] | 4778 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 4779 | struct mem_cgroup *from; |
Tejun Heo | 4530edd | 2015-09-11 15:00:19 -0400 | [diff] [blame] | 4780 | struct task_struct *leader, *p; |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 4781 | struct mm_struct *mm; |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 4782 | unsigned long move_flags; |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 4783 | int ret = 0; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4784 | |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 4785 | /* charge immigration isn't supported on the default hierarchy */ |
| 4786 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 4787 | return 0; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4788 | |
Tejun Heo | 4530edd | 2015-09-11 15:00:19 -0400 | [diff] [blame] | 4789 | /* |
| 4790 | * Multi-process migrations only happen on the default hierarchy |
| 4791 | * where charge immigration is not used. Perform charge |
| 4792 | * immigration if @tset contains a leader and whine if there are |
| 4793 | * multiple. |
| 4794 | */ |
| 4795 | p = NULL; |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 4796 | cgroup_taskset_for_each_leader(leader, css, tset) { |
Tejun Heo | 4530edd | 2015-09-11 15:00:19 -0400 | [diff] [blame] | 4797 | WARN_ON_ONCE(p); |
| 4798 | p = leader; |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 4799 | memcg = mem_cgroup_from_css(css); |
Tejun Heo | 4530edd | 2015-09-11 15:00:19 -0400 | [diff] [blame] | 4800 | } |
| 4801 | if (!p) |
| 4802 | return 0; |
| 4803 | |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 4804 | /* |
| 4805 | * We are now commited to this value whatever it is. Changes in this |
| 4806 | * tunable will only affect upcoming migrations, not the current one. |
| 4807 | * So we need to save it, and keep it going. |
| 4808 | */ |
| 4809 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); |
| 4810 | if (!move_flags) |
| 4811 | return 0; |
| 4812 | |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 4813 | from = mem_cgroup_from_task(p); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4814 | |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 4815 | VM_BUG_ON(from == memcg); |
Johannes Weiner | 247b144 | 2014-12-10 15:44:11 -0800 | [diff] [blame] | 4816 | |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 4817 | mm = get_task_mm(p); |
| 4818 | if (!mm) |
| 4819 | return 0; |
| 4820 | /* We move charges only when we move a owner of the mm */ |
| 4821 | if (mm->owner == p) { |
| 4822 | VM_BUG_ON(mc.from); |
| 4823 | VM_BUG_ON(mc.to); |
| 4824 | VM_BUG_ON(mc.precharge); |
| 4825 | VM_BUG_ON(mc.moved_charge); |
| 4826 | VM_BUG_ON(mc.moved_swap); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4827 | |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 4828 | spin_lock(&mc.lock); |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 4829 | mc.mm = mm; |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 4830 | mc.from = from; |
| 4831 | mc.to = memcg; |
| 4832 | mc.flags = move_flags; |
| 4833 | spin_unlock(&mc.lock); |
| 4834 | /* We set mc.moving_task later */ |
| 4835 | |
| 4836 | ret = mem_cgroup_precharge_mc(mm); |
| 4837 | if (ret) |
| 4838 | mem_cgroup_clear_mc(); |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 4839 | } else { |
| 4840 | mmput(mm); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4841 | } |
| 4842 | return ret; |
| 4843 | } |
| 4844 | |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 4845 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4846 | { |
Johannes Weiner | 4e2f245 | 2014-12-10 15:44:08 -0800 | [diff] [blame] | 4847 | if (mc.to) |
| 4848 | mem_cgroup_clear_mc(); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4849 | } |
| 4850 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4851 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
| 4852 | unsigned long addr, unsigned long end, |
| 4853 | struct mm_walk *walk) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4854 | { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4855 | int ret = 0; |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 4856 | struct vm_area_struct *vma = walk->vma; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4857 | pte_t *pte; |
| 4858 | spinlock_t *ptl; |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4859 | enum mc_target_type target_type; |
| 4860 | union mc_target target; |
| 4861 | struct page *page; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4862 | |
Kirill A. Shutemov | b6ec57f | 2016-01-21 16:40:25 -0800 | [diff] [blame] | 4863 | ptl = pmd_trans_huge_lock(pmd, vma); |
| 4864 | if (ptl) { |
Hugh Dickins | 62ade86 | 2012-05-18 11:28:34 -0700 | [diff] [blame] | 4865 | if (mc.precharge < HPAGE_PMD_NR) { |
Kirill A. Shutemov | bf92915 | 2013-11-14 14:30:54 -0800 | [diff] [blame] | 4866 | spin_unlock(ptl); |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4867 | return 0; |
| 4868 | } |
| 4869 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); |
| 4870 | if (target_type == MC_TARGET_PAGE) { |
| 4871 | page = target.page; |
| 4872 | if (!isolate_lru_page(page)) { |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 4873 | if (!mem_cgroup_move_account(page, true, |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 4874 | mc.from, mc.to)) { |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4875 | mc.precharge -= HPAGE_PMD_NR; |
| 4876 | mc.moved_charge += HPAGE_PMD_NR; |
| 4877 | } |
| 4878 | putback_lru_page(page); |
| 4879 | } |
| 4880 | put_page(page); |
| 4881 | } |
Kirill A. Shutemov | bf92915 | 2013-11-14 14:30:54 -0800 | [diff] [blame] | 4882 | spin_unlock(ptl); |
Andrea Arcangeli | 1a5a990 | 2012-03-21 16:33:42 -0700 | [diff] [blame] | 4883 | return 0; |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4884 | } |
| 4885 | |
Andrea Arcangeli | 45f83ce | 2012-03-28 14:42:40 -0700 | [diff] [blame] | 4886 | if (pmd_trans_unstable(pmd)) |
| 4887 | return 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4888 | retry: |
| 4889 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
| 4890 | for (; addr != end; addr += PAGE_SIZE) { |
| 4891 | pte_t ptent = *(pte++); |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4892 | swp_entry_t ent; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4893 | |
| 4894 | if (!mc.precharge) |
| 4895 | break; |
| 4896 | |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 4897 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4898 | case MC_TARGET_PAGE: |
| 4899 | page = target.page; |
Kirill A. Shutemov | 53f9263 | 2016-01-15 16:53:42 -0800 | [diff] [blame] | 4900 | /* |
| 4901 | * We can have a part of the split pmd here. Moving it |
| 4902 | * can be done but it would be too convoluted so simply |
| 4903 | * ignore such a partial THP and keep it in original |
| 4904 | * memcg. There should be somebody mapping the head. |
| 4905 | */ |
| 4906 | if (PageTransCompound(page)) |
| 4907 | goto put; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4908 | if (isolate_lru_page(page)) |
| 4909 | goto put; |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 4910 | if (!mem_cgroup_move_account(page, false, |
| 4911 | mc.from, mc.to)) { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4912 | mc.precharge--; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4913 | /* we uncharge from mc.from later. */ |
| 4914 | mc.moved_charge++; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4915 | } |
| 4916 | putback_lru_page(page); |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 4917 | put: /* get_mctgt_type() gets the page */ |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4918 | put_page(page); |
| 4919 | break; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4920 | case MC_TARGET_SWAP: |
| 4921 | ent = target.ent; |
Hugh Dickins | e91cbb4 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 4922 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4923 | mc.precharge--; |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4924 | /* we fixup refcnts and charges later. */ |
| 4925 | mc.moved_swap++; |
| 4926 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4927 | break; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4928 | default: |
| 4929 | break; |
| 4930 | } |
| 4931 | } |
| 4932 | pte_unmap_unlock(pte - 1, ptl); |
| 4933 | cond_resched(); |
| 4934 | |
| 4935 | if (addr != end) { |
| 4936 | /* |
| 4937 | * We have consumed all precharges we got in can_attach(). |
| 4938 | * We try charge one by one, but don't do any additional |
| 4939 | * charges to mc.to if we have failed in charge once in attach() |
| 4940 | * phase. |
| 4941 | */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4942 | ret = mem_cgroup_do_precharge(1); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4943 | if (!ret) |
| 4944 | goto retry; |
| 4945 | } |
| 4946 | |
| 4947 | return ret; |
| 4948 | } |
| 4949 | |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 4950 | static void mem_cgroup_move_charge(void) |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4951 | { |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 4952 | struct mm_walk mem_cgroup_move_charge_walk = { |
| 4953 | .pmd_entry = mem_cgroup_move_charge_pte_range, |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 4954 | .mm = mc.mm, |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 4955 | }; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4956 | |
| 4957 | lru_add_drain_all(); |
Johannes Weiner | 312722c | 2014-12-10 15:44:25 -0800 | [diff] [blame] | 4958 | /* |
Johannes Weiner | 81f8c3a | 2016-03-15 14:57:04 -0700 | [diff] [blame] | 4959 | * Signal lock_page_memcg() to take the memcg's move_lock |
| 4960 | * while we're moving its pages to another memcg. Then wait |
| 4961 | * for already started RCU-only updates to finish. |
Johannes Weiner | 312722c | 2014-12-10 15:44:25 -0800 | [diff] [blame] | 4962 | */ |
| 4963 | atomic_inc(&mc.from->moving_account); |
| 4964 | synchronize_rcu(); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4965 | retry: |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 4966 | if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) { |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4967 | /* |
| 4968 | * Someone who are holding the mmap_sem might be waiting in |
| 4969 | * waitq. So we cancel all extra charges, wake up all waiters, |
| 4970 | * and retry. Because we cancel precharges, we might not be able |
| 4971 | * to move enough charges, but moving charge is a best-effort |
| 4972 | * feature anyway, so it wouldn't be a big problem. |
| 4973 | */ |
| 4974 | __mem_cgroup_clear_mc(); |
| 4975 | cond_resched(); |
| 4976 | goto retry; |
| 4977 | } |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 4978 | /* |
| 4979 | * When we have consumed all precharges and failed in doing |
| 4980 | * additional charge, the page walk just aborts. |
| 4981 | */ |
| 4982 | walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk); |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 4983 | up_read(&mc.mm->mmap_sem); |
Johannes Weiner | 312722c | 2014-12-10 15:44:25 -0800 | [diff] [blame] | 4984 | atomic_dec(&mc.from->moving_account); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4985 | } |
| 4986 | |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 4987 | static void mem_cgroup_move_task(void) |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 4988 | { |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 4989 | if (mc.to) { |
| 4990 | mem_cgroup_move_charge(); |
KOSAKI Motohiro | a433658 | 2011-06-15 15:08:13 -0700 | [diff] [blame] | 4991 | mem_cgroup_clear_mc(); |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 4992 | } |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 4993 | } |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 4994 | #else /* !CONFIG_MMU */ |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 4995 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 4996 | { |
| 4997 | return 0; |
| 4998 | } |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 4999 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 5000 | { |
| 5001 | } |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 5002 | static void mem_cgroup_move_task(void) |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 5003 | { |
| 5004 | } |
| 5005 | #endif |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 5006 | |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 5007 | /* |
| 5008 | * Cgroup retains root cgroups across [un]mount cycles making it necessary |
Tejun Heo | aa6ec29 | 2014-07-09 10:08:08 -0400 | [diff] [blame] | 5009 | * to verify whether we're attached to the default hierarchy on each mount |
| 5010 | * attempt. |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 5011 | */ |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5012 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 5013 | { |
| 5014 | /* |
Tejun Heo | aa6ec29 | 2014-07-09 10:08:08 -0400 | [diff] [blame] | 5015 | * use_hierarchy is forced on the default hierarchy. cgroup core |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 5016 | * guarantees that @root doesn't have any children, so turning it |
| 5017 | * on for the root memcg is enough. |
| 5018 | */ |
Tejun Heo | 9e10a13 | 2015-09-18 11:56:28 -0400 | [diff] [blame] | 5019 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
Vladimir Davydov | 7feee59 | 2015-03-12 16:26:19 -0700 | [diff] [blame] | 5020 | root_mem_cgroup->use_hierarchy = true; |
| 5021 | else |
| 5022 | root_mem_cgroup->use_hierarchy = false; |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 5023 | } |
| 5024 | |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5025 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
| 5026 | struct cftype *cft) |
| 5027 | { |
Johannes Weiner | f5fc3c5d | 2015-11-05 18:50:23 -0800 | [diff] [blame] | 5028 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 5029 | |
| 5030 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5031 | } |
| 5032 | |
| 5033 | static int memory_low_show(struct seq_file *m, void *v) |
| 5034 | { |
| 5035 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 5036 | unsigned long low = READ_ONCE(memcg->low); |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5037 | |
| 5038 | if (low == PAGE_COUNTER_MAX) |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 5039 | seq_puts(m, "max\n"); |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5040 | else |
| 5041 | seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE); |
| 5042 | |
| 5043 | return 0; |
| 5044 | } |
| 5045 | |
| 5046 | static ssize_t memory_low_write(struct kernfs_open_file *of, |
| 5047 | char *buf, size_t nbytes, loff_t off) |
| 5048 | { |
| 5049 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
| 5050 | unsigned long low; |
| 5051 | int err; |
| 5052 | |
| 5053 | buf = strstrip(buf); |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 5054 | err = page_counter_memparse(buf, "max", &low); |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5055 | if (err) |
| 5056 | return err; |
| 5057 | |
| 5058 | memcg->low = low; |
| 5059 | |
| 5060 | return nbytes; |
| 5061 | } |
| 5062 | |
| 5063 | static int memory_high_show(struct seq_file *m, void *v) |
| 5064 | { |
| 5065 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 5066 | unsigned long high = READ_ONCE(memcg->high); |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5067 | |
| 5068 | if (high == PAGE_COUNTER_MAX) |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 5069 | seq_puts(m, "max\n"); |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5070 | else |
| 5071 | seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE); |
| 5072 | |
| 5073 | return 0; |
| 5074 | } |
| 5075 | |
| 5076 | static ssize_t memory_high_write(struct kernfs_open_file *of, |
| 5077 | char *buf, size_t nbytes, loff_t off) |
| 5078 | { |
| 5079 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
Johannes Weiner | 588083b | 2016-03-17 14:20:25 -0700 | [diff] [blame] | 5080 | unsigned long nr_pages; |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5081 | unsigned long high; |
| 5082 | int err; |
| 5083 | |
| 5084 | buf = strstrip(buf); |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 5085 | err = page_counter_memparse(buf, "max", &high); |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5086 | if (err) |
| 5087 | return err; |
| 5088 | |
| 5089 | memcg->high = high; |
| 5090 | |
Johannes Weiner | 588083b | 2016-03-17 14:20:25 -0700 | [diff] [blame] | 5091 | nr_pages = page_counter_read(&memcg->memory); |
| 5092 | if (nr_pages > high) |
| 5093 | try_to_free_mem_cgroup_pages(memcg, nr_pages - high, |
| 5094 | GFP_KERNEL, true); |
| 5095 | |
Tejun Heo | 2529bb3 | 2015-05-22 18:23:34 -0400 | [diff] [blame] | 5096 | memcg_wb_domain_size_changed(memcg); |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5097 | return nbytes; |
| 5098 | } |
| 5099 | |
| 5100 | static int memory_max_show(struct seq_file *m, void *v) |
| 5101 | { |
| 5102 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 5103 | unsigned long max = READ_ONCE(memcg->memory.limit); |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5104 | |
| 5105 | if (max == PAGE_COUNTER_MAX) |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 5106 | seq_puts(m, "max\n"); |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5107 | else |
| 5108 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); |
| 5109 | |
| 5110 | return 0; |
| 5111 | } |
| 5112 | |
| 5113 | static ssize_t memory_max_write(struct kernfs_open_file *of, |
| 5114 | char *buf, size_t nbytes, loff_t off) |
| 5115 | { |
| 5116 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
Johannes Weiner | b6e6edc | 2016-03-17 14:20:28 -0700 | [diff] [blame] | 5117 | unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES; |
| 5118 | bool drained = false; |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5119 | unsigned long max; |
| 5120 | int err; |
| 5121 | |
| 5122 | buf = strstrip(buf); |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 5123 | err = page_counter_memparse(buf, "max", &max); |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5124 | if (err) |
| 5125 | return err; |
| 5126 | |
Johannes Weiner | b6e6edc | 2016-03-17 14:20:28 -0700 | [diff] [blame] | 5127 | xchg(&memcg->memory.limit, max); |
| 5128 | |
| 5129 | for (;;) { |
| 5130 | unsigned long nr_pages = page_counter_read(&memcg->memory); |
| 5131 | |
| 5132 | if (nr_pages <= max) |
| 5133 | break; |
| 5134 | |
| 5135 | if (signal_pending(current)) { |
| 5136 | err = -EINTR; |
| 5137 | break; |
| 5138 | } |
| 5139 | |
| 5140 | if (!drained) { |
| 5141 | drain_all_stock(memcg); |
| 5142 | drained = true; |
| 5143 | continue; |
| 5144 | } |
| 5145 | |
| 5146 | if (nr_reclaims) { |
| 5147 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, |
| 5148 | GFP_KERNEL, true)) |
| 5149 | nr_reclaims--; |
| 5150 | continue; |
| 5151 | } |
| 5152 | |
| 5153 | mem_cgroup_events(memcg, MEMCG_OOM, 1); |
| 5154 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
| 5155 | break; |
| 5156 | } |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5157 | |
Tejun Heo | 2529bb3 | 2015-05-22 18:23:34 -0400 | [diff] [blame] | 5158 | memcg_wb_domain_size_changed(memcg); |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5159 | return nbytes; |
| 5160 | } |
| 5161 | |
| 5162 | static int memory_events_show(struct seq_file *m, void *v) |
| 5163 | { |
| 5164 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
| 5165 | |
| 5166 | seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW)); |
| 5167 | seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH)); |
| 5168 | seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX)); |
| 5169 | seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM)); |
| 5170 | |
| 5171 | return 0; |
| 5172 | } |
| 5173 | |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 5174 | static int memory_stat_show(struct seq_file *m, void *v) |
| 5175 | { |
| 5176 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 5177 | unsigned long stat[MEMCG_NR_STAT]; |
| 5178 | unsigned long events[MEMCG_NR_EVENTS]; |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 5179 | int i; |
| 5180 | |
| 5181 | /* |
| 5182 | * Provide statistics on the state of the memory subsystem as |
| 5183 | * well as cumulative event counters that show past behavior. |
| 5184 | * |
| 5185 | * This list is ordered following a combination of these gradients: |
| 5186 | * 1) generic big picture -> specifics and details |
| 5187 | * 2) reflecting userspace activity -> reflecting kernel heuristics |
| 5188 | * |
| 5189 | * Current memory state: |
| 5190 | */ |
| 5191 | |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 5192 | tree_stat(memcg, stat); |
| 5193 | tree_events(memcg, events); |
| 5194 | |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 5195 | seq_printf(m, "anon %llu\n", |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 5196 | (u64)stat[MEM_CGROUP_STAT_RSS] * PAGE_SIZE); |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 5197 | seq_printf(m, "file %llu\n", |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 5198 | (u64)stat[MEM_CGROUP_STAT_CACHE] * PAGE_SIZE); |
Vladimir Davydov | 12580e4 | 2016-03-17 14:17:38 -0700 | [diff] [blame] | 5199 | seq_printf(m, "kernel_stack %llu\n", |
| 5200 | (u64)stat[MEMCG_KERNEL_STACK] * PAGE_SIZE); |
Vladimir Davydov | 27ee57c | 2016-03-17 14:17:35 -0700 | [diff] [blame] | 5201 | seq_printf(m, "slab %llu\n", |
| 5202 | (u64)(stat[MEMCG_SLAB_RECLAIMABLE] + |
| 5203 | stat[MEMCG_SLAB_UNRECLAIMABLE]) * PAGE_SIZE); |
Johannes Weiner | b2807f0 | 2016-01-20 15:03:22 -0800 | [diff] [blame] | 5204 | seq_printf(m, "sock %llu\n", |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 5205 | (u64)stat[MEMCG_SOCK] * PAGE_SIZE); |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 5206 | |
| 5207 | seq_printf(m, "file_mapped %llu\n", |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 5208 | (u64)stat[MEM_CGROUP_STAT_FILE_MAPPED] * PAGE_SIZE); |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 5209 | seq_printf(m, "file_dirty %llu\n", |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 5210 | (u64)stat[MEM_CGROUP_STAT_DIRTY] * PAGE_SIZE); |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 5211 | seq_printf(m, "file_writeback %llu\n", |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 5212 | (u64)stat[MEM_CGROUP_STAT_WRITEBACK] * PAGE_SIZE); |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 5213 | |
| 5214 | for (i = 0; i < NR_LRU_LISTS; i++) { |
| 5215 | struct mem_cgroup *mi; |
| 5216 | unsigned long val = 0; |
| 5217 | |
| 5218 | for_each_mem_cgroup_tree(mi, memcg) |
| 5219 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)); |
| 5220 | seq_printf(m, "%s %llu\n", |
| 5221 | mem_cgroup_lru_names[i], (u64)val * PAGE_SIZE); |
| 5222 | } |
| 5223 | |
Vladimir Davydov | 27ee57c | 2016-03-17 14:17:35 -0700 | [diff] [blame] | 5224 | seq_printf(m, "slab_reclaimable %llu\n", |
| 5225 | (u64)stat[MEMCG_SLAB_RECLAIMABLE] * PAGE_SIZE); |
| 5226 | seq_printf(m, "slab_unreclaimable %llu\n", |
| 5227 | (u64)stat[MEMCG_SLAB_UNRECLAIMABLE] * PAGE_SIZE); |
| 5228 | |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 5229 | /* Accumulated memory events */ |
| 5230 | |
| 5231 | seq_printf(m, "pgfault %lu\n", |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 5232 | events[MEM_CGROUP_EVENTS_PGFAULT]); |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 5233 | seq_printf(m, "pgmajfault %lu\n", |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 5234 | events[MEM_CGROUP_EVENTS_PGMAJFAULT]); |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 5235 | |
| 5236 | return 0; |
| 5237 | } |
| 5238 | |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5239 | static struct cftype memory_files[] = { |
| 5240 | { |
| 5241 | .name = "current", |
Johannes Weiner | f5fc3c5d | 2015-11-05 18:50:23 -0800 | [diff] [blame] | 5242 | .flags = CFTYPE_NOT_ON_ROOT, |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5243 | .read_u64 = memory_current_read, |
| 5244 | }, |
| 5245 | { |
| 5246 | .name = "low", |
| 5247 | .flags = CFTYPE_NOT_ON_ROOT, |
| 5248 | .seq_show = memory_low_show, |
| 5249 | .write = memory_low_write, |
| 5250 | }, |
| 5251 | { |
| 5252 | .name = "high", |
| 5253 | .flags = CFTYPE_NOT_ON_ROOT, |
| 5254 | .seq_show = memory_high_show, |
| 5255 | .write = memory_high_write, |
| 5256 | }, |
| 5257 | { |
| 5258 | .name = "max", |
| 5259 | .flags = CFTYPE_NOT_ON_ROOT, |
| 5260 | .seq_show = memory_max_show, |
| 5261 | .write = memory_max_write, |
| 5262 | }, |
| 5263 | { |
| 5264 | .name = "events", |
| 5265 | .flags = CFTYPE_NOT_ON_ROOT, |
Tejun Heo | 472912a | 2015-09-18 18:01:59 -0400 | [diff] [blame] | 5266 | .file_offset = offsetof(struct mem_cgroup, events_file), |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5267 | .seq_show = memory_events_show, |
| 5268 | }, |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 5269 | { |
| 5270 | .name = "stat", |
| 5271 | .flags = CFTYPE_NOT_ON_ROOT, |
| 5272 | .seq_show = memory_stat_show, |
| 5273 | }, |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5274 | { } /* terminate */ |
| 5275 | }; |
| 5276 | |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 5277 | struct cgroup_subsys memory_cgrp_subsys = { |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 5278 | .css_alloc = mem_cgroup_css_alloc, |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 5279 | .css_online = mem_cgroup_css_online, |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 5280 | .css_offline = mem_cgroup_css_offline, |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 5281 | .css_released = mem_cgroup_css_released, |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 5282 | .css_free = mem_cgroup_css_free, |
Tejun Heo | 1ced953 | 2014-07-08 18:02:57 -0400 | [diff] [blame] | 5283 | .css_reset = mem_cgroup_css_reset, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5284 | .can_attach = mem_cgroup_can_attach, |
| 5285 | .cancel_attach = mem_cgroup_cancel_attach, |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 5286 | .post_attach = mem_cgroup_move_task, |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 5287 | .bind = mem_cgroup_bind, |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5288 | .dfl_cftypes = memory_files, |
| 5289 | .legacy_cftypes = mem_cgroup_legacy_files, |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 5290 | .early_init = 0, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 5291 | }; |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 5292 | |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5293 | /** |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5294 | * mem_cgroup_low - check if memory consumption is below the normal range |
| 5295 | * @root: the highest ancestor to consider |
| 5296 | * @memcg: the memory cgroup to check |
| 5297 | * |
| 5298 | * Returns %true if memory consumption of @memcg, and that of all |
| 5299 | * configurable ancestors up to @root, is below the normal range. |
| 5300 | */ |
| 5301 | bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg) |
| 5302 | { |
| 5303 | if (mem_cgroup_disabled()) |
| 5304 | return false; |
| 5305 | |
| 5306 | /* |
| 5307 | * The toplevel group doesn't have a configurable range, so |
| 5308 | * it's never low when looked at directly, and it is not |
| 5309 | * considered an ancestor when assessing the hierarchy. |
| 5310 | */ |
| 5311 | |
| 5312 | if (memcg == root_mem_cgroup) |
| 5313 | return false; |
| 5314 | |
Michal Hocko | 4e54ded | 2015-02-27 15:51:46 -0800 | [diff] [blame] | 5315 | if (page_counter_read(&memcg->memory) >= memcg->low) |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5316 | return false; |
| 5317 | |
| 5318 | while (memcg != root) { |
| 5319 | memcg = parent_mem_cgroup(memcg); |
| 5320 | |
| 5321 | if (memcg == root_mem_cgroup) |
| 5322 | break; |
| 5323 | |
Michal Hocko | 4e54ded | 2015-02-27 15:51:46 -0800 | [diff] [blame] | 5324 | if (page_counter_read(&memcg->memory) >= memcg->low) |
Johannes Weiner | 241994e | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5325 | return false; |
| 5326 | } |
| 5327 | return true; |
| 5328 | } |
| 5329 | |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5330 | /** |
| 5331 | * mem_cgroup_try_charge - try charging a page |
| 5332 | * @page: page to charge |
| 5333 | * @mm: mm context of the victim |
| 5334 | * @gfp_mask: reclaim mode |
| 5335 | * @memcgp: charged memcg return |
| 5336 | * |
| 5337 | * Try to charge @page to the memcg that @mm belongs to, reclaiming |
| 5338 | * pages according to @gfp_mask if necessary. |
| 5339 | * |
| 5340 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. |
| 5341 | * Otherwise, an error code is returned. |
| 5342 | * |
| 5343 | * After page->mapping has been set up, the caller must finalize the |
| 5344 | * charge with mem_cgroup_commit_charge(). Or abort the transaction |
| 5345 | * with mem_cgroup_cancel_charge() in case page instantiation fails. |
| 5346 | */ |
| 5347 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 5348 | gfp_t gfp_mask, struct mem_cgroup **memcgp, |
| 5349 | bool compound) |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5350 | { |
| 5351 | struct mem_cgroup *memcg = NULL; |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 5352 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5353 | int ret = 0; |
| 5354 | |
| 5355 | if (mem_cgroup_disabled()) |
| 5356 | goto out; |
| 5357 | |
| 5358 | if (PageSwapCache(page)) { |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5359 | /* |
| 5360 | * Every swap fault against a single page tries to charge the |
| 5361 | * page, bail as early as possible. shmem_unuse() encounters |
| 5362 | * already charged pages, too. The USED bit is protected by |
| 5363 | * the page lock, which serializes swap cache removal, which |
| 5364 | * in turn serializes uncharging. |
| 5365 | */ |
Vladimir Davydov | e993d90 | 2015-09-09 15:35:35 -0700 | [diff] [blame] | 5366 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5367 | if (page->mem_cgroup) |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5368 | goto out; |
Vladimir Davydov | e993d90 | 2015-09-09 15:35:35 -0700 | [diff] [blame] | 5369 | |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 5370 | if (do_swap_account) { |
Vladimir Davydov | e993d90 | 2015-09-09 15:35:35 -0700 | [diff] [blame] | 5371 | swp_entry_t ent = { .val = page_private(page), }; |
| 5372 | unsigned short id = lookup_swap_cgroup_id(ent); |
| 5373 | |
| 5374 | rcu_read_lock(); |
| 5375 | memcg = mem_cgroup_from_id(id); |
| 5376 | if (memcg && !css_tryget_online(&memcg->css)) |
| 5377 | memcg = NULL; |
| 5378 | rcu_read_unlock(); |
| 5379 | } |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5380 | } |
| 5381 | |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5382 | if (!memcg) |
| 5383 | memcg = get_mem_cgroup_from_mm(mm); |
| 5384 | |
| 5385 | ret = try_charge(memcg, gfp_mask, nr_pages); |
| 5386 | |
| 5387 | css_put(&memcg->css); |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5388 | out: |
| 5389 | *memcgp = memcg; |
| 5390 | return ret; |
| 5391 | } |
| 5392 | |
| 5393 | /** |
| 5394 | * mem_cgroup_commit_charge - commit a page charge |
| 5395 | * @page: page to charge |
| 5396 | * @memcg: memcg to charge the page to |
| 5397 | * @lrucare: page might be on LRU already |
| 5398 | * |
| 5399 | * Finalize a charge transaction started by mem_cgroup_try_charge(), |
| 5400 | * after page->mapping has been set up. This must happen atomically |
| 5401 | * as part of the page instantiation, i.e. under the page table lock |
| 5402 | * for anonymous pages, under the page lock for page and swap cache. |
| 5403 | * |
| 5404 | * In addition, the page must not be on the LRU during the commit, to |
| 5405 | * prevent racing with task migration. If it might be, use @lrucare. |
| 5406 | * |
| 5407 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. |
| 5408 | */ |
| 5409 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 5410 | bool lrucare, bool compound) |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5411 | { |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 5412 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5413 | |
| 5414 | VM_BUG_ON_PAGE(!page->mapping, page); |
| 5415 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); |
| 5416 | |
| 5417 | if (mem_cgroup_disabled()) |
| 5418 | return; |
| 5419 | /* |
| 5420 | * Swap faults will attempt to charge the same page multiple |
| 5421 | * times. But reuse_swap_page() might have removed the page |
| 5422 | * from swapcache already, so we can't check PageSwapCache(). |
| 5423 | */ |
| 5424 | if (!memcg) |
| 5425 | return; |
| 5426 | |
Johannes Weiner | 6abb5a8 | 2014-08-08 14:19:33 -0700 | [diff] [blame] | 5427 | commit_charge(page, memcg, lrucare); |
| 5428 | |
Johannes Weiner | 6abb5a8 | 2014-08-08 14:19:33 -0700 | [diff] [blame] | 5429 | local_irq_disable(); |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 5430 | mem_cgroup_charge_statistics(memcg, page, compound, nr_pages); |
Johannes Weiner | 6abb5a8 | 2014-08-08 14:19:33 -0700 | [diff] [blame] | 5431 | memcg_check_events(memcg, page); |
| 5432 | local_irq_enable(); |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5433 | |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 5434 | if (do_memsw_account() && PageSwapCache(page)) { |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5435 | swp_entry_t entry = { .val = page_private(page) }; |
| 5436 | /* |
| 5437 | * The swap entry might not get freed for a long time, |
| 5438 | * let's not wait for it. The page already received a |
| 5439 | * memory+swap charge, drop the swap entry duplicate. |
| 5440 | */ |
| 5441 | mem_cgroup_uncharge_swap(entry); |
| 5442 | } |
| 5443 | } |
| 5444 | |
| 5445 | /** |
| 5446 | * mem_cgroup_cancel_charge - cancel a page charge |
| 5447 | * @page: page to charge |
| 5448 | * @memcg: memcg to charge the page to |
| 5449 | * |
| 5450 | * Cancel a charge transaction started by mem_cgroup_try_charge(). |
| 5451 | */ |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 5452 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg, |
| 5453 | bool compound) |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5454 | { |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 5455 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5456 | |
| 5457 | if (mem_cgroup_disabled()) |
| 5458 | return; |
| 5459 | /* |
| 5460 | * Swap faults will attempt to charge the same page multiple |
| 5461 | * times. But reuse_swap_page() might have removed the page |
| 5462 | * from swapcache already, so we can't check PageSwapCache(). |
| 5463 | */ |
| 5464 | if (!memcg) |
| 5465 | return; |
| 5466 | |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5467 | cancel_charge(memcg, nr_pages); |
| 5468 | } |
| 5469 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5470 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5471 | unsigned long nr_anon, unsigned long nr_file, |
| 5472 | unsigned long nr_huge, struct page *dummy_page) |
| 5473 | { |
Johannes Weiner | 18eca2e | 2014-12-10 15:43:57 -0800 | [diff] [blame] | 5474 | unsigned long nr_pages = nr_anon + nr_file; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5475 | unsigned long flags; |
| 5476 | |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 5477 | if (!mem_cgroup_is_root(memcg)) { |
Johannes Weiner | 18eca2e | 2014-12-10 15:43:57 -0800 | [diff] [blame] | 5478 | page_counter_uncharge(&memcg->memory, nr_pages); |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 5479 | if (do_memsw_account()) |
Johannes Weiner | 18eca2e | 2014-12-10 15:43:57 -0800 | [diff] [blame] | 5480 | page_counter_uncharge(&memcg->memsw, nr_pages); |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 5481 | memcg_oom_recover(memcg); |
| 5482 | } |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5483 | |
| 5484 | local_irq_save(flags); |
| 5485 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); |
| 5486 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); |
| 5487 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); |
| 5488 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); |
Johannes Weiner | 18eca2e | 2014-12-10 15:43:57 -0800 | [diff] [blame] | 5489 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5490 | memcg_check_events(memcg, dummy_page); |
| 5491 | local_irq_restore(flags); |
Johannes Weiner | e8ea14c | 2014-12-10 15:42:42 -0800 | [diff] [blame] | 5492 | |
| 5493 | if (!mem_cgroup_is_root(memcg)) |
Johannes Weiner | 18eca2e | 2014-12-10 15:43:57 -0800 | [diff] [blame] | 5494 | css_put_many(&memcg->css, nr_pages); |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5495 | } |
| 5496 | |
| 5497 | static void uncharge_list(struct list_head *page_list) |
| 5498 | { |
| 5499 | struct mem_cgroup *memcg = NULL; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5500 | unsigned long nr_anon = 0; |
| 5501 | unsigned long nr_file = 0; |
| 5502 | unsigned long nr_huge = 0; |
| 5503 | unsigned long pgpgout = 0; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5504 | struct list_head *next; |
| 5505 | struct page *page; |
| 5506 | |
Johannes Weiner | 8b59265 | 2016-03-17 14:20:31 -0700 | [diff] [blame] | 5507 | /* |
| 5508 | * Note that the list can be a single page->lru; hence the |
| 5509 | * do-while loop instead of a simple list_for_each_entry(). |
| 5510 | */ |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5511 | next = page_list->next; |
| 5512 | do { |
| 5513 | unsigned int nr_pages = 1; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5514 | |
| 5515 | page = list_entry(next, struct page, lru); |
| 5516 | next = page->lru.next; |
| 5517 | |
| 5518 | VM_BUG_ON_PAGE(PageLRU(page), page); |
| 5519 | VM_BUG_ON_PAGE(page_count(page), page); |
| 5520 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5521 | if (!page->mem_cgroup) |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5522 | continue; |
| 5523 | |
| 5524 | /* |
| 5525 | * Nobody should be changing or seriously looking at |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5526 | * page->mem_cgroup at this point, we have fully |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 5527 | * exclusive access to the page. |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5528 | */ |
| 5529 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5530 | if (memcg != page->mem_cgroup) { |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5531 | if (memcg) { |
Johannes Weiner | 18eca2e | 2014-12-10 15:43:57 -0800 | [diff] [blame] | 5532 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
| 5533 | nr_huge, page); |
| 5534 | pgpgout = nr_anon = nr_file = nr_huge = 0; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5535 | } |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5536 | memcg = page->mem_cgroup; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5537 | } |
| 5538 | |
| 5539 | if (PageTransHuge(page)) { |
| 5540 | nr_pages <<= compound_order(page); |
| 5541 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
| 5542 | nr_huge += nr_pages; |
| 5543 | } |
| 5544 | |
| 5545 | if (PageAnon(page)) |
| 5546 | nr_anon += nr_pages; |
| 5547 | else |
| 5548 | nr_file += nr_pages; |
| 5549 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5550 | page->mem_cgroup = NULL; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5551 | |
| 5552 | pgpgout++; |
| 5553 | } while (next != page_list); |
| 5554 | |
| 5555 | if (memcg) |
Johannes Weiner | 18eca2e | 2014-12-10 15:43:57 -0800 | [diff] [blame] | 5556 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
| 5557 | nr_huge, page); |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5558 | } |
| 5559 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5560 | /** |
| 5561 | * mem_cgroup_uncharge - uncharge a page |
| 5562 | * @page: page to uncharge |
| 5563 | * |
| 5564 | * Uncharge a page previously charged with mem_cgroup_try_charge() and |
| 5565 | * mem_cgroup_commit_charge(). |
| 5566 | */ |
| 5567 | void mem_cgroup_uncharge(struct page *page) |
| 5568 | { |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5569 | if (mem_cgroup_disabled()) |
| 5570 | return; |
| 5571 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5572 | /* Don't touch page->lru of any random page, pre-check: */ |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5573 | if (!page->mem_cgroup) |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5574 | return; |
| 5575 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5576 | INIT_LIST_HEAD(&page->lru); |
| 5577 | uncharge_list(&page->lru); |
| 5578 | } |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5579 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5580 | /** |
| 5581 | * mem_cgroup_uncharge_list - uncharge a list of page |
| 5582 | * @page_list: list of pages to uncharge |
| 5583 | * |
| 5584 | * Uncharge a list of pages previously charged with |
| 5585 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). |
| 5586 | */ |
| 5587 | void mem_cgroup_uncharge_list(struct list_head *page_list) |
| 5588 | { |
| 5589 | if (mem_cgroup_disabled()) |
| 5590 | return; |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5591 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5592 | if (!list_empty(page_list)) |
| 5593 | uncharge_list(page_list); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5594 | } |
| 5595 | |
| 5596 | /** |
Johannes Weiner | 6a93ca8 | 2016-03-15 14:57:19 -0700 | [diff] [blame] | 5597 | * mem_cgroup_migrate - charge a page's replacement |
| 5598 | * @oldpage: currently circulating page |
| 5599 | * @newpage: replacement page |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5600 | * |
Johannes Weiner | 6a93ca8 | 2016-03-15 14:57:19 -0700 | [diff] [blame] | 5601 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
| 5602 | * be uncharged upon free. |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5603 | * |
| 5604 | * Both pages must be locked, @newpage->mapping must be set up. |
| 5605 | */ |
Johannes Weiner | 6a93ca8 | 2016-03-15 14:57:19 -0700 | [diff] [blame] | 5606 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5607 | { |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 5608 | struct mem_cgroup *memcg; |
Johannes Weiner | 44b7a8d | 2016-01-20 15:03:16 -0800 | [diff] [blame] | 5609 | unsigned int nr_pages; |
| 5610 | bool compound; |
Tejun Heo | d93c413 | 2016-06-24 14:49:54 -0700 | [diff] [blame] | 5611 | unsigned long flags; |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5612 | |
| 5613 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); |
| 5614 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5615 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
Johannes Weiner | 6abb5a8 | 2014-08-08 14:19:33 -0700 | [diff] [blame] | 5616 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
| 5617 | newpage); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5618 | |
| 5619 | if (mem_cgroup_disabled()) |
| 5620 | return; |
| 5621 | |
| 5622 | /* Page cache replacement: new page already charged? */ |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5623 | if (newpage->mem_cgroup) |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5624 | return; |
| 5625 | |
Hugh Dickins | 45637ba | 2015-11-05 18:49:40 -0800 | [diff] [blame] | 5626 | /* Swapcache readahead pages can get replaced before being charged */ |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5627 | memcg = oldpage->mem_cgroup; |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 5628 | if (!memcg) |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5629 | return; |
| 5630 | |
Johannes Weiner | 44b7a8d | 2016-01-20 15:03:16 -0800 | [diff] [blame] | 5631 | /* Force-charge the new page. The old one will be freed soon */ |
| 5632 | compound = PageTransHuge(newpage); |
| 5633 | nr_pages = compound ? hpage_nr_pages(newpage) : 1; |
| 5634 | |
| 5635 | page_counter_charge(&memcg->memory, nr_pages); |
| 5636 | if (do_memsw_account()) |
| 5637 | page_counter_charge(&memcg->memsw, nr_pages); |
| 5638 | css_get_many(&memcg->css, nr_pages); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5639 | |
Johannes Weiner | 9cf7666 | 2016-03-15 14:57:58 -0700 | [diff] [blame] | 5640 | commit_charge(newpage, memcg, false); |
Johannes Weiner | 44b7a8d | 2016-01-20 15:03:16 -0800 | [diff] [blame] | 5641 | |
Tejun Heo | d93c413 | 2016-06-24 14:49:54 -0700 | [diff] [blame] | 5642 | local_irq_save(flags); |
Johannes Weiner | 44b7a8d | 2016-01-20 15:03:16 -0800 | [diff] [blame] | 5643 | mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages); |
| 5644 | memcg_check_events(memcg, newpage); |
Tejun Heo | d93c413 | 2016-06-24 14:49:54 -0700 | [diff] [blame] | 5645 | local_irq_restore(flags); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5646 | } |
| 5647 | |
Johannes Weiner | ef12947 | 2016-01-14 15:21:34 -0800 | [diff] [blame] | 5648 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 5649 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
| 5650 | |
| 5651 | void sock_update_memcg(struct sock *sk) |
| 5652 | { |
| 5653 | struct mem_cgroup *memcg; |
| 5654 | |
| 5655 | /* Socket cloning can throw us here with sk_cgrp already |
| 5656 | * filled. It won't however, necessarily happen from |
| 5657 | * process context. So the test for root memcg given |
| 5658 | * the current task's memcg won't help us in this case. |
| 5659 | * |
| 5660 | * Respecting the original socket's memcg is a better |
| 5661 | * decision in this case. |
| 5662 | */ |
| 5663 | if (sk->sk_memcg) { |
| 5664 | BUG_ON(mem_cgroup_is_root(sk->sk_memcg)); |
| 5665 | css_get(&sk->sk_memcg->css); |
| 5666 | return; |
| 5667 | } |
| 5668 | |
| 5669 | rcu_read_lock(); |
| 5670 | memcg = mem_cgroup_from_task(current); |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5671 | if (memcg == root_mem_cgroup) |
| 5672 | goto out; |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 5673 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5674 | goto out; |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5675 | if (css_tryget_online(&memcg->css)) |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 5676 | sk->sk_memcg = memcg; |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5677 | out: |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 5678 | rcu_read_unlock(); |
| 5679 | } |
| 5680 | EXPORT_SYMBOL(sock_update_memcg); |
| 5681 | |
| 5682 | void sock_release_memcg(struct sock *sk) |
| 5683 | { |
| 5684 | WARN_ON(!sk->sk_memcg); |
| 5685 | css_put(&sk->sk_memcg->css); |
| 5686 | } |
| 5687 | |
| 5688 | /** |
| 5689 | * mem_cgroup_charge_skmem - charge socket memory |
| 5690 | * @memcg: memcg to charge |
| 5691 | * @nr_pages: number of pages to charge |
| 5692 | * |
| 5693 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within |
| 5694 | * @memcg's configured limit, %false if the charge had to be forced. |
| 5695 | */ |
| 5696 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) |
| 5697 | { |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5698 | gfp_t gfp_mask = GFP_KERNEL; |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 5699 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5700 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 5701 | struct page_counter *fail; |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5702 | |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 5703 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
| 5704 | memcg->tcpmem_pressure = 0; |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5705 | return true; |
| 5706 | } |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 5707 | page_counter_charge(&memcg->tcpmem, nr_pages); |
| 5708 | memcg->tcpmem_pressure = 1; |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5709 | return false; |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 5710 | } |
Johannes Weiner | d886f4e | 2016-01-20 15:02:47 -0800 | [diff] [blame] | 5711 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5712 | /* Don't block in the packet receive path */ |
| 5713 | if (in_softirq()) |
| 5714 | gfp_mask = GFP_NOWAIT; |
| 5715 | |
Johannes Weiner | b2807f0 | 2016-01-20 15:03:22 -0800 | [diff] [blame] | 5716 | this_cpu_add(memcg->stat->count[MEMCG_SOCK], nr_pages); |
| 5717 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5718 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
| 5719 | return true; |
| 5720 | |
| 5721 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 5722 | return false; |
| 5723 | } |
| 5724 | |
| 5725 | /** |
| 5726 | * mem_cgroup_uncharge_skmem - uncharge socket memory |
| 5727 | * @memcg - memcg to uncharge |
| 5728 | * @nr_pages - number of pages to uncharge |
| 5729 | */ |
| 5730 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) |
| 5731 | { |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5732 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 5733 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5734 | return; |
| 5735 | } |
Johannes Weiner | d886f4e | 2016-01-20 15:02:47 -0800 | [diff] [blame] | 5736 | |
Johannes Weiner | b2807f0 | 2016-01-20 15:03:22 -0800 | [diff] [blame] | 5737 | this_cpu_sub(memcg->stat->count[MEMCG_SOCK], nr_pages); |
| 5738 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5739 | page_counter_uncharge(&memcg->memory, nr_pages); |
| 5740 | css_put_many(&memcg->css, nr_pages); |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 5741 | } |
| 5742 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5743 | static int __init cgroup_memory(char *s) |
| 5744 | { |
| 5745 | char *token; |
| 5746 | |
| 5747 | while ((token = strsep(&s, ",")) != NULL) { |
| 5748 | if (!*token) |
| 5749 | continue; |
| 5750 | if (!strcmp(token, "nosocket")) |
| 5751 | cgroup_memory_nosocket = true; |
Vladimir Davydov | 04823c8 | 2016-01-20 15:02:38 -0800 | [diff] [blame] | 5752 | if (!strcmp(token, "nokmem")) |
| 5753 | cgroup_memory_nokmem = true; |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5754 | } |
| 5755 | return 0; |
| 5756 | } |
| 5757 | __setup("cgroup.memory=", cgroup_memory); |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 5758 | |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 5759 | /* |
Michal Hocko | 1081312 | 2013-02-22 16:35:41 -0800 | [diff] [blame] | 5760 | * subsys_initcall() for memory controller. |
| 5761 | * |
| 5762 | * Some parts like hotcpu_notifier() have to be initialized from this context |
| 5763 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically |
| 5764 | * everything that doesn't depend on a specific mem_cgroup structure should |
| 5765 | * be initialized from here. |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 5766 | */ |
| 5767 | static int __init mem_cgroup_init(void) |
| 5768 | { |
Johannes Weiner | 95a045f | 2015-02-11 15:26:33 -0800 | [diff] [blame] | 5769 | int cpu, node; |
| 5770 | |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 5771 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
Johannes Weiner | 95a045f | 2015-02-11 15:26:33 -0800 | [diff] [blame] | 5772 | |
| 5773 | for_each_possible_cpu(cpu) |
| 5774 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, |
| 5775 | drain_local_stock); |
| 5776 | |
| 5777 | for_each_node(node) { |
| 5778 | struct mem_cgroup_tree_per_node *rtpn; |
| 5779 | int zone; |
| 5780 | |
| 5781 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, |
| 5782 | node_online(node) ? node : NUMA_NO_NODE); |
| 5783 | |
| 5784 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
| 5785 | struct mem_cgroup_tree_per_zone *rtpz; |
| 5786 | |
| 5787 | rtpz = &rtpn->rb_tree_per_zone[zone]; |
| 5788 | rtpz->rb_root = RB_ROOT; |
| 5789 | spin_lock_init(&rtpz->lock); |
| 5790 | } |
| 5791 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
| 5792 | } |
| 5793 | |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 5794 | return 0; |
| 5795 | } |
| 5796 | subsys_initcall(mem_cgroup_init); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5797 | |
| 5798 | #ifdef CONFIG_MEMCG_SWAP |
| 5799 | /** |
| 5800 | * mem_cgroup_swapout - transfer a memsw charge to swap |
| 5801 | * @page: page whose memsw charge to transfer |
| 5802 | * @entry: swap entry to move the charge to |
| 5803 | * |
| 5804 | * Transfer the memsw charge of @page to @entry. |
| 5805 | */ |
| 5806 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) |
| 5807 | { |
| 5808 | struct mem_cgroup *memcg; |
| 5809 | unsigned short oldid; |
| 5810 | |
| 5811 | VM_BUG_ON_PAGE(PageLRU(page), page); |
| 5812 | VM_BUG_ON_PAGE(page_count(page), page); |
| 5813 | |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 5814 | if (!do_memsw_account()) |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5815 | return; |
| 5816 | |
| 5817 | memcg = page->mem_cgroup; |
| 5818 | |
| 5819 | /* Readahead page, never charged */ |
| 5820 | if (!memcg) |
| 5821 | return; |
| 5822 | |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 5823 | mem_cgroup_id_get(memcg); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5824 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); |
| 5825 | VM_BUG_ON_PAGE(oldid, page); |
| 5826 | mem_cgroup_swap_statistics(memcg, true); |
| 5827 | |
| 5828 | page->mem_cgroup = NULL; |
| 5829 | |
| 5830 | if (!mem_cgroup_is_root(memcg)) |
| 5831 | page_counter_uncharge(&memcg->memory, 1); |
| 5832 | |
Sebastian Andrzej Siewior | ce9ce66 | 2015-09-04 15:47:50 -0700 | [diff] [blame] | 5833 | /* |
| 5834 | * Interrupts should be disabled here because the caller holds the |
| 5835 | * mapping->tree_lock lock which is taken with interrupts-off. It is |
| 5836 | * important here to have the interrupts disabled because it is the |
| 5837 | * only synchronisation we have for udpating the per-CPU variables. |
| 5838 | */ |
| 5839 | VM_BUG_ON(!irqs_disabled()); |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 5840 | mem_cgroup_charge_statistics(memcg, page, false, -1); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5841 | memcg_check_events(memcg, page); |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 5842 | |
| 5843 | if (!mem_cgroup_is_root(memcg)) |
| 5844 | css_put(&memcg->css); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5845 | } |
| 5846 | |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 5847 | /* |
| 5848 | * mem_cgroup_try_charge_swap - try charging a swap entry |
| 5849 | * @page: page being added to swap |
| 5850 | * @entry: swap entry to charge |
| 5851 | * |
| 5852 | * Try to charge @entry to the memcg that @page belongs to. |
| 5853 | * |
| 5854 | * Returns 0 on success, -ENOMEM on failure. |
| 5855 | */ |
| 5856 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) |
| 5857 | { |
| 5858 | struct mem_cgroup *memcg; |
| 5859 | struct page_counter *counter; |
| 5860 | unsigned short oldid; |
| 5861 | |
| 5862 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account) |
| 5863 | return 0; |
| 5864 | |
| 5865 | memcg = page->mem_cgroup; |
| 5866 | |
| 5867 | /* Readahead page, never charged */ |
| 5868 | if (!memcg) |
| 5869 | return 0; |
| 5870 | |
| 5871 | if (!mem_cgroup_is_root(memcg) && |
| 5872 | !page_counter_try_charge(&memcg->swap, 1, &counter)) |
| 5873 | return -ENOMEM; |
| 5874 | |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 5875 | mem_cgroup_id_get(memcg); |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 5876 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); |
| 5877 | VM_BUG_ON_PAGE(oldid, page); |
| 5878 | mem_cgroup_swap_statistics(memcg, true); |
| 5879 | |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 5880 | return 0; |
| 5881 | } |
| 5882 | |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5883 | /** |
| 5884 | * mem_cgroup_uncharge_swap - uncharge a swap entry |
| 5885 | * @entry: swap entry to uncharge |
| 5886 | * |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 5887 | * Drop the swap charge associated with @entry. |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5888 | */ |
| 5889 | void mem_cgroup_uncharge_swap(swp_entry_t entry) |
| 5890 | { |
| 5891 | struct mem_cgroup *memcg; |
| 5892 | unsigned short id; |
| 5893 | |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 5894 | if (!do_swap_account) |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5895 | return; |
| 5896 | |
| 5897 | id = swap_cgroup_record(entry, 0); |
| 5898 | rcu_read_lock(); |
Vladimir Davydov | adbe427 | 2015-04-15 16:13:00 -0700 | [diff] [blame] | 5899 | memcg = mem_cgroup_from_id(id); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5900 | if (memcg) { |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 5901 | if (!mem_cgroup_is_root(memcg)) { |
| 5902 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
| 5903 | page_counter_uncharge(&memcg->swap, 1); |
| 5904 | else |
| 5905 | page_counter_uncharge(&memcg->memsw, 1); |
| 5906 | } |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5907 | mem_cgroup_swap_statistics(memcg, false); |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 5908 | mem_cgroup_id_put(memcg); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5909 | } |
| 5910 | rcu_read_unlock(); |
| 5911 | } |
| 5912 | |
Vladimir Davydov | d8b3843 | 2016-01-20 15:03:07 -0800 | [diff] [blame] | 5913 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
| 5914 | { |
| 5915 | long nr_swap_pages = get_nr_swap_pages(); |
| 5916 | |
| 5917 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
| 5918 | return nr_swap_pages; |
| 5919 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) |
| 5920 | nr_swap_pages = min_t(long, nr_swap_pages, |
| 5921 | READ_ONCE(memcg->swap.limit) - |
| 5922 | page_counter_read(&memcg->swap)); |
| 5923 | return nr_swap_pages; |
| 5924 | } |
| 5925 | |
Vladimir Davydov | 5ccc5ab | 2016-01-20 15:03:10 -0800 | [diff] [blame] | 5926 | bool mem_cgroup_swap_full(struct page *page) |
| 5927 | { |
| 5928 | struct mem_cgroup *memcg; |
| 5929 | |
| 5930 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
| 5931 | |
| 5932 | if (vm_swap_full()) |
| 5933 | return true; |
| 5934 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
| 5935 | return false; |
| 5936 | |
| 5937 | memcg = page->mem_cgroup; |
| 5938 | if (!memcg) |
| 5939 | return false; |
| 5940 | |
| 5941 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) |
| 5942 | if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.limit) |
| 5943 | return true; |
| 5944 | |
| 5945 | return false; |
| 5946 | } |
| 5947 | |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5948 | /* for remember boot option*/ |
| 5949 | #ifdef CONFIG_MEMCG_SWAP_ENABLED |
| 5950 | static int really_do_swap_account __initdata = 1; |
| 5951 | #else |
| 5952 | static int really_do_swap_account __initdata; |
| 5953 | #endif |
| 5954 | |
| 5955 | static int __init enable_swap_account(char *s) |
| 5956 | { |
| 5957 | if (!strcmp(s, "1")) |
| 5958 | really_do_swap_account = 1; |
| 5959 | else if (!strcmp(s, "0")) |
| 5960 | really_do_swap_account = 0; |
| 5961 | return 1; |
| 5962 | } |
| 5963 | __setup("swapaccount=", enable_swap_account); |
| 5964 | |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 5965 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
| 5966 | struct cftype *cft) |
| 5967 | { |
| 5968 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 5969 | |
| 5970 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; |
| 5971 | } |
| 5972 | |
| 5973 | static int swap_max_show(struct seq_file *m, void *v) |
| 5974 | { |
| 5975 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
| 5976 | unsigned long max = READ_ONCE(memcg->swap.limit); |
| 5977 | |
| 5978 | if (max == PAGE_COUNTER_MAX) |
| 5979 | seq_puts(m, "max\n"); |
| 5980 | else |
| 5981 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); |
| 5982 | |
| 5983 | return 0; |
| 5984 | } |
| 5985 | |
| 5986 | static ssize_t swap_max_write(struct kernfs_open_file *of, |
| 5987 | char *buf, size_t nbytes, loff_t off) |
| 5988 | { |
| 5989 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
| 5990 | unsigned long max; |
| 5991 | int err; |
| 5992 | |
| 5993 | buf = strstrip(buf); |
| 5994 | err = page_counter_memparse(buf, "max", &max); |
| 5995 | if (err) |
| 5996 | return err; |
| 5997 | |
| 5998 | mutex_lock(&memcg_limit_mutex); |
| 5999 | err = page_counter_limit(&memcg->swap, max); |
| 6000 | mutex_unlock(&memcg_limit_mutex); |
| 6001 | if (err) |
| 6002 | return err; |
| 6003 | |
| 6004 | return nbytes; |
| 6005 | } |
| 6006 | |
| 6007 | static struct cftype swap_files[] = { |
| 6008 | { |
| 6009 | .name = "swap.current", |
| 6010 | .flags = CFTYPE_NOT_ON_ROOT, |
| 6011 | .read_u64 = swap_current_read, |
| 6012 | }, |
| 6013 | { |
| 6014 | .name = "swap.max", |
| 6015 | .flags = CFTYPE_NOT_ON_ROOT, |
| 6016 | .seq_show = swap_max_show, |
| 6017 | .write = swap_max_write, |
| 6018 | }, |
| 6019 | { } /* terminate */ |
| 6020 | }; |
| 6021 | |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 6022 | static struct cftype memsw_cgroup_files[] = { |
| 6023 | { |
| 6024 | .name = "memsw.usage_in_bytes", |
| 6025 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), |
| 6026 | .read_u64 = mem_cgroup_read_u64, |
| 6027 | }, |
| 6028 | { |
| 6029 | .name = "memsw.max_usage_in_bytes", |
| 6030 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), |
| 6031 | .write = mem_cgroup_reset, |
| 6032 | .read_u64 = mem_cgroup_read_u64, |
| 6033 | }, |
| 6034 | { |
| 6035 | .name = "memsw.limit_in_bytes", |
| 6036 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), |
| 6037 | .write = mem_cgroup_write, |
| 6038 | .read_u64 = mem_cgroup_read_u64, |
| 6039 | }, |
| 6040 | { |
| 6041 | .name = "memsw.failcnt", |
| 6042 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), |
| 6043 | .write = mem_cgroup_reset, |
| 6044 | .read_u64 = mem_cgroup_read_u64, |
| 6045 | }, |
| 6046 | { }, /* terminate */ |
| 6047 | }; |
| 6048 | |
| 6049 | static int __init mem_cgroup_swap_init(void) |
| 6050 | { |
| 6051 | if (!mem_cgroup_disabled() && really_do_swap_account) { |
| 6052 | do_swap_account = 1; |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 6053 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, |
| 6054 | swap_files)); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 6055 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
| 6056 | memsw_cgroup_files)); |
| 6057 | } |
| 6058 | return 0; |
| 6059 | } |
| 6060 | subsys_initcall(mem_cgroup_swap_init); |
| 6061 | |
| 6062 | #endif /* CONFIG_MEMCG_SWAP */ |