Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * kernel/cpuset.c |
| 3 | * |
| 4 | * Processor and Memory placement constraints for sets of tasks. |
| 5 | * |
| 6 | * Copyright (C) 2003 BULL SA. |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 7 | * Copyright (C) 2004-2007 Silicon Graphics, Inc. |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 8 | * Copyright (C) 2006 Google, Inc |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 9 | * |
| 10 | * Portions derived from Patrick Mochel's sysfs code. |
| 11 | * sysfs is Copyright (c) 2001-3 Patrick Mochel |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 12 | * |
Paul Jackson | 825a46a | 2006-03-24 03:16:03 -0800 | [diff] [blame] | 13 | * 2003-10-10 Written by Simon Derr. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 14 | * 2003-10-22 Updates by Stephen Hemminger. |
Paul Jackson | 825a46a | 2006-03-24 03:16:03 -0800 | [diff] [blame] | 15 | * 2004 May-July Rework by Paul Jackson. |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 16 | * 2006 Rework by Paul Menage to use generic cgroups |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 17 | * 2008 Rework of the scheduler domains and CPU hotplug handling |
| 18 | * by Max Krasnyansky |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 19 | * |
| 20 | * This file is subject to the terms and conditions of the GNU General Public |
| 21 | * License. See the file COPYING in the main directory of the Linux |
| 22 | * distribution for more details. |
| 23 | */ |
| 24 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 25 | #include <linux/cpu.h> |
| 26 | #include <linux/cpumask.h> |
| 27 | #include <linux/cpuset.h> |
| 28 | #include <linux/err.h> |
| 29 | #include <linux/errno.h> |
| 30 | #include <linux/file.h> |
| 31 | #include <linux/fs.h> |
| 32 | #include <linux/init.h> |
| 33 | #include <linux/interrupt.h> |
| 34 | #include <linux/kernel.h> |
| 35 | #include <linux/kmod.h> |
| 36 | #include <linux/list.h> |
Paul Jackson | 68860ec | 2005-10-30 15:02:36 -0800 | [diff] [blame] | 37 | #include <linux/mempolicy.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 38 | #include <linux/mm.h> |
Miao Xie | f481891 | 2008-11-19 15:36:30 -0800 | [diff] [blame] | 39 | #include <linux/memory.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 40 | #include <linux/module.h> |
| 41 | #include <linux/mount.h> |
| 42 | #include <linux/namei.h> |
| 43 | #include <linux/pagemap.h> |
| 44 | #include <linux/proc_fs.h> |
Paul Jackson | 6b9c260 | 2006-01-08 01:02:02 -0800 | [diff] [blame] | 45 | #include <linux/rcupdate.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 46 | #include <linux/sched.h> |
| 47 | #include <linux/seq_file.h> |
David Quigley | 22fb52d | 2006-06-23 02:04:00 -0700 | [diff] [blame] | 48 | #include <linux/security.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 49 | #include <linux/slab.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 50 | #include <linux/spinlock.h> |
| 51 | #include <linux/stat.h> |
| 52 | #include <linux/string.h> |
| 53 | #include <linux/time.h> |
| 54 | #include <linux/backing-dev.h> |
| 55 | #include <linux/sort.h> |
| 56 | |
| 57 | #include <asm/uaccess.h> |
| 58 | #include <asm/atomic.h> |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 59 | #include <linux/mutex.h> |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 60 | #include <linux/workqueue.h> |
| 61 | #include <linux/cgroup.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 62 | |
Paul Jackson | 202f72d | 2006-01-08 01:01:57 -0800 | [diff] [blame] | 63 | /* |
Miao Xie | f90d411 | 2009-01-16 10:24:10 +0800 | [diff] [blame] | 64 | * Workqueue for cpuset related tasks. |
| 65 | * |
| 66 | * Using kevent workqueue may cause deadlock when memory_migrate |
| 67 | * is set. So we create a separate workqueue thread for cpuset. |
| 68 | */ |
| 69 | static struct workqueue_struct *cpuset_wq; |
| 70 | |
| 71 | /* |
Paul Jackson | 202f72d | 2006-01-08 01:01:57 -0800 | [diff] [blame] | 72 | * Tracks how many cpusets are currently defined in system. |
| 73 | * When there is only one cpuset (the root cpuset) we can |
| 74 | * short circuit some hooks. |
| 75 | */ |
Paul Jackson | 7edc596 | 2006-01-08 01:02:03 -0800 | [diff] [blame] | 76 | int number_of_cpusets __read_mostly; |
Paul Jackson | 202f72d | 2006-01-08 01:01:57 -0800 | [diff] [blame] | 77 | |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 78 | /* Forward declare cgroup structures */ |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 79 | struct cgroup_subsys cpuset_subsys; |
| 80 | struct cpuset; |
| 81 | |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 82 | /* See "Frequency meter" comments, below. */ |
| 83 | |
| 84 | struct fmeter { |
| 85 | int cnt; /* unprocessed events count */ |
| 86 | int val; /* most recent output value */ |
| 87 | time_t time; /* clock (secs) when val computed */ |
| 88 | spinlock_t lock; /* guards read or write of above */ |
| 89 | }; |
| 90 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 91 | struct cpuset { |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 92 | struct cgroup_subsys_state css; |
| 93 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 94 | unsigned long flags; /* "unsigned long" so bitops work */ |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 95 | cpumask_var_t cpus_allowed; /* CPUs allowed to tasks in cpuset */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 96 | nodemask_t mems_allowed; /* Memory Nodes allowed to tasks */ |
| 97 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 98 | struct cpuset *parent; /* my parent */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 99 | |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 100 | struct fmeter fmeter; /* memory_pressure filter */ |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 101 | |
| 102 | /* partition number for rebuild_sched_domains() */ |
| 103 | int pn; |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 104 | |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 105 | /* for custom sched domain */ |
| 106 | int relax_domain_level; |
| 107 | |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 108 | /* used for walking a cpuset heirarchy */ |
| 109 | struct list_head stack_list; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 110 | }; |
| 111 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 112 | /* Retrieve the cpuset for a cgroup */ |
| 113 | static inline struct cpuset *cgroup_cs(struct cgroup *cont) |
| 114 | { |
| 115 | return container_of(cgroup_subsys_state(cont, cpuset_subsys_id), |
| 116 | struct cpuset, css); |
| 117 | } |
| 118 | |
| 119 | /* Retrieve the cpuset for a task */ |
| 120 | static inline struct cpuset *task_cs(struct task_struct *task) |
| 121 | { |
| 122 | return container_of(task_subsys_state(task, cpuset_subsys_id), |
| 123 | struct cpuset, css); |
| 124 | } |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 125 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 126 | /* bits in struct cpuset flags field */ |
| 127 | typedef enum { |
| 128 | CS_CPU_EXCLUSIVE, |
| 129 | CS_MEM_EXCLUSIVE, |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 130 | CS_MEM_HARDWALL, |
Paul Jackson | 45b07ef | 2006-01-08 01:00:56 -0800 | [diff] [blame] | 131 | CS_MEMORY_MIGRATE, |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 132 | CS_SCHED_LOAD_BALANCE, |
Paul Jackson | 825a46a | 2006-03-24 03:16:03 -0800 | [diff] [blame] | 133 | CS_SPREAD_PAGE, |
| 134 | CS_SPREAD_SLAB, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 135 | } cpuset_flagbits_t; |
| 136 | |
| 137 | /* convenient tests for these bits */ |
| 138 | static inline int is_cpu_exclusive(const struct cpuset *cs) |
| 139 | { |
Paul Jackson | 7b5b9ef | 2006-03-24 03:16:00 -0800 | [diff] [blame] | 140 | return test_bit(CS_CPU_EXCLUSIVE, &cs->flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 141 | } |
| 142 | |
| 143 | static inline int is_mem_exclusive(const struct cpuset *cs) |
| 144 | { |
Paul Jackson | 7b5b9ef | 2006-03-24 03:16:00 -0800 | [diff] [blame] | 145 | return test_bit(CS_MEM_EXCLUSIVE, &cs->flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 146 | } |
| 147 | |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 148 | static inline int is_mem_hardwall(const struct cpuset *cs) |
| 149 | { |
| 150 | return test_bit(CS_MEM_HARDWALL, &cs->flags); |
| 151 | } |
| 152 | |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 153 | static inline int is_sched_load_balance(const struct cpuset *cs) |
| 154 | { |
| 155 | return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); |
| 156 | } |
| 157 | |
Paul Jackson | 45b07ef | 2006-01-08 01:00:56 -0800 | [diff] [blame] | 158 | static inline int is_memory_migrate(const struct cpuset *cs) |
| 159 | { |
Paul Jackson | 7b5b9ef | 2006-03-24 03:16:00 -0800 | [diff] [blame] | 160 | return test_bit(CS_MEMORY_MIGRATE, &cs->flags); |
Paul Jackson | 45b07ef | 2006-01-08 01:00:56 -0800 | [diff] [blame] | 161 | } |
| 162 | |
Paul Jackson | 825a46a | 2006-03-24 03:16:03 -0800 | [diff] [blame] | 163 | static inline int is_spread_page(const struct cpuset *cs) |
| 164 | { |
| 165 | return test_bit(CS_SPREAD_PAGE, &cs->flags); |
| 166 | } |
| 167 | |
| 168 | static inline int is_spread_slab(const struct cpuset *cs) |
| 169 | { |
| 170 | return test_bit(CS_SPREAD_SLAB, &cs->flags); |
| 171 | } |
| 172 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 173 | static struct cpuset top_cpuset = { |
| 174 | .flags = ((1 << CS_CPU_EXCLUSIVE) | (1 << CS_MEM_EXCLUSIVE)), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 175 | }; |
| 176 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 177 | /* |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 178 | * There are two global mutexes guarding cpuset structures. The first |
| 179 | * is the main control groups cgroup_mutex, accessed via |
| 180 | * cgroup_lock()/cgroup_unlock(). The second is the cpuset-specific |
| 181 | * callback_mutex, below. They can nest. It is ok to first take |
| 182 | * cgroup_mutex, then nest callback_mutex. We also require taking |
| 183 | * task_lock() when dereferencing a task's cpuset pointer. See "The |
| 184 | * task_lock() exception", at the end of this comment. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 185 | * |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 186 | * A task must hold both mutexes to modify cpusets. If a task |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 187 | * holds cgroup_mutex, then it blocks others wanting that mutex, |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 188 | * ensuring that it is the only task able to also acquire callback_mutex |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 189 | * and be able to modify cpusets. It can perform various checks on |
| 190 | * the cpuset structure first, knowing nothing will change. It can |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 191 | * also allocate memory while just holding cgroup_mutex. While it is |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 192 | * performing these checks, various callback routines can briefly |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 193 | * acquire callback_mutex to query cpusets. Once it is ready to make |
| 194 | * the changes, it takes callback_mutex, blocking everyone else. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 195 | * |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 196 | * Calls to the kernel memory allocator can not be made while holding |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 197 | * callback_mutex, as that would risk double tripping on callback_mutex |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 198 | * from one of the callbacks into the cpuset code from within |
| 199 | * __alloc_pages(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 200 | * |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 201 | * If a task is only holding callback_mutex, then it has read-only |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 202 | * access to cpusets. |
| 203 | * |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 204 | * Now, the task_struct fields mems_allowed and mempolicy may be changed |
| 205 | * by other task, we use alloc_lock in the task_struct fields to protect |
| 206 | * them. |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 207 | * |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 208 | * The cpuset_common_file_read() handlers only hold callback_mutex across |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 209 | * small pieces of code, such as when reading out possibly multi-word |
| 210 | * cpumasks and nodemasks. |
| 211 | * |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 212 | * Accessing a task's cpuset should be done in accordance with the |
| 213 | * guidelines for accessing subsystem state in kernel/cgroup.c |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 214 | */ |
| 215 | |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 216 | static DEFINE_MUTEX(callback_mutex); |
Paul Jackson | 4247bdc | 2005-09-10 00:26:06 -0700 | [diff] [blame] | 217 | |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 218 | /* |
David Rientjes | 75aa199 | 2009-01-06 14:39:01 -0800 | [diff] [blame] | 219 | * cpuset_buffer_lock protects both the cpuset_name and cpuset_nodelist |
| 220 | * buffers. They are statically allocated to prevent using excess stack |
| 221 | * when calling cpuset_print_task_mems_allowed(). |
| 222 | */ |
| 223 | #define CPUSET_NAME_LEN (128) |
| 224 | #define CPUSET_NODELIST_LEN (256) |
| 225 | static char cpuset_name[CPUSET_NAME_LEN]; |
| 226 | static char cpuset_nodelist[CPUSET_NODELIST_LEN]; |
| 227 | static DEFINE_SPINLOCK(cpuset_buffer_lock); |
| 228 | |
| 229 | /* |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 230 | * This is ugly, but preserves the userspace API for existing cpuset |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 231 | * users. If someone tries to mount the "cpuset" filesystem, we |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 232 | * silently switch it to mount "cgroup" instead |
| 233 | */ |
David Howells | 454e239 | 2006-06-23 02:02:57 -0700 | [diff] [blame] | 234 | static int cpuset_get_sb(struct file_system_type *fs_type, |
| 235 | int flags, const char *unused_dev_name, |
| 236 | void *data, struct vfsmount *mnt) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 237 | { |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 238 | struct file_system_type *cgroup_fs = get_fs_type("cgroup"); |
| 239 | int ret = -ENODEV; |
| 240 | if (cgroup_fs) { |
| 241 | char mountopts[] = |
| 242 | "cpuset,noprefix," |
| 243 | "release_agent=/sbin/cpuset_release_agent"; |
| 244 | ret = cgroup_fs->get_sb(cgroup_fs, flags, |
| 245 | unused_dev_name, mountopts, mnt); |
| 246 | put_filesystem(cgroup_fs); |
| 247 | } |
| 248 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 249 | } |
| 250 | |
| 251 | static struct file_system_type cpuset_fs_type = { |
| 252 | .name = "cpuset", |
| 253 | .get_sb = cpuset_get_sb, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 254 | }; |
| 255 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 256 | /* |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 257 | * Return in pmask the portion of a cpusets's cpus_allowed that |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 258 | * are online. If none are online, walk up the cpuset hierarchy |
| 259 | * until we find one that does have some online cpus. If we get |
| 260 | * all the way to the top and still haven't found any online cpus, |
| 261 | * return cpu_online_map. Or if passed a NULL cs from an exit'ing |
| 262 | * task, return cpu_online_map. |
| 263 | * |
| 264 | * One way or another, we guarantee to return some non-empty subset |
| 265 | * of cpu_online_map. |
| 266 | * |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 267 | * Call with callback_mutex held. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 268 | */ |
| 269 | |
Li Zefan | 6af866a | 2009-01-07 18:08:45 -0800 | [diff] [blame] | 270 | static void guarantee_online_cpus(const struct cpuset *cs, |
| 271 | struct cpumask *pmask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 272 | { |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 273 | while (cs && !cpumask_intersects(cs->cpus_allowed, cpu_online_mask)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 274 | cs = cs->parent; |
| 275 | if (cs) |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 276 | cpumask_and(pmask, cs->cpus_allowed, cpu_online_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 277 | else |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 278 | cpumask_copy(pmask, cpu_online_mask); |
| 279 | BUG_ON(!cpumask_intersects(pmask, cpu_online_mask)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 280 | } |
| 281 | |
| 282 | /* |
| 283 | * Return in *pmask the portion of a cpusets's mems_allowed that |
Christoph Lameter | 0e1e7c7 | 2007-10-16 01:25:38 -0700 | [diff] [blame] | 284 | * are online, with memory. If none are online with memory, walk |
| 285 | * up the cpuset hierarchy until we find one that does have some |
| 286 | * online mems. If we get all the way to the top and still haven't |
| 287 | * found any online mems, return node_states[N_HIGH_MEMORY]. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 288 | * |
| 289 | * One way or another, we guarantee to return some non-empty subset |
Christoph Lameter | 0e1e7c7 | 2007-10-16 01:25:38 -0700 | [diff] [blame] | 290 | * of node_states[N_HIGH_MEMORY]. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 291 | * |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 292 | * Call with callback_mutex held. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 293 | */ |
| 294 | |
| 295 | static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask) |
| 296 | { |
Christoph Lameter | 0e1e7c7 | 2007-10-16 01:25:38 -0700 | [diff] [blame] | 297 | while (cs && !nodes_intersects(cs->mems_allowed, |
| 298 | node_states[N_HIGH_MEMORY])) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 299 | cs = cs->parent; |
| 300 | if (cs) |
Christoph Lameter | 0e1e7c7 | 2007-10-16 01:25:38 -0700 | [diff] [blame] | 301 | nodes_and(*pmask, cs->mems_allowed, |
| 302 | node_states[N_HIGH_MEMORY]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 303 | else |
Christoph Lameter | 0e1e7c7 | 2007-10-16 01:25:38 -0700 | [diff] [blame] | 304 | *pmask = node_states[N_HIGH_MEMORY]; |
| 305 | BUG_ON(!nodes_intersects(*pmask, node_states[N_HIGH_MEMORY])); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 306 | } |
| 307 | |
Miao Xie | f3b39d4 | 2009-06-16 15:31:46 -0700 | [diff] [blame] | 308 | /* |
| 309 | * update task's spread flag if cpuset's page/slab spread flag is set |
| 310 | * |
| 311 | * Called with callback_mutex/cgroup_mutex held |
| 312 | */ |
| 313 | static void cpuset_update_task_spread_flag(struct cpuset *cs, |
| 314 | struct task_struct *tsk) |
| 315 | { |
| 316 | if (is_spread_page(cs)) |
| 317 | tsk->flags |= PF_SPREAD_PAGE; |
| 318 | else |
| 319 | tsk->flags &= ~PF_SPREAD_PAGE; |
| 320 | if (is_spread_slab(cs)) |
| 321 | tsk->flags |= PF_SPREAD_SLAB; |
| 322 | else |
| 323 | tsk->flags &= ~PF_SPREAD_SLAB; |
| 324 | } |
| 325 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 326 | /* |
| 327 | * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q? |
| 328 | * |
| 329 | * One cpuset is a subset of another if all its allowed CPUs and |
| 330 | * Memory Nodes are a subset of the other, and its exclusive flags |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 331 | * are only set if the other's are set. Call holding cgroup_mutex. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 332 | */ |
| 333 | |
| 334 | static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q) |
| 335 | { |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 336 | return cpumask_subset(p->cpus_allowed, q->cpus_allowed) && |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 337 | nodes_subset(p->mems_allowed, q->mems_allowed) && |
| 338 | is_cpu_exclusive(p) <= is_cpu_exclusive(q) && |
| 339 | is_mem_exclusive(p) <= is_mem_exclusive(q); |
| 340 | } |
| 341 | |
Li Zefan | 645fcc9 | 2009-01-07 18:08:43 -0800 | [diff] [blame] | 342 | /** |
| 343 | * alloc_trial_cpuset - allocate a trial cpuset |
| 344 | * @cs: the cpuset that the trial cpuset duplicates |
| 345 | */ |
| 346 | static struct cpuset *alloc_trial_cpuset(const struct cpuset *cs) |
| 347 | { |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 348 | struct cpuset *trial; |
| 349 | |
| 350 | trial = kmemdup(cs, sizeof(*cs), GFP_KERNEL); |
| 351 | if (!trial) |
| 352 | return NULL; |
| 353 | |
| 354 | if (!alloc_cpumask_var(&trial->cpus_allowed, GFP_KERNEL)) { |
| 355 | kfree(trial); |
| 356 | return NULL; |
| 357 | } |
| 358 | cpumask_copy(trial->cpus_allowed, cs->cpus_allowed); |
| 359 | |
| 360 | return trial; |
Li Zefan | 645fcc9 | 2009-01-07 18:08:43 -0800 | [diff] [blame] | 361 | } |
| 362 | |
| 363 | /** |
| 364 | * free_trial_cpuset - free the trial cpuset |
| 365 | * @trial: the trial cpuset to be freed |
| 366 | */ |
| 367 | static void free_trial_cpuset(struct cpuset *trial) |
| 368 | { |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 369 | free_cpumask_var(trial->cpus_allowed); |
Li Zefan | 645fcc9 | 2009-01-07 18:08:43 -0800 | [diff] [blame] | 370 | kfree(trial); |
| 371 | } |
| 372 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 373 | /* |
| 374 | * validate_change() - Used to validate that any proposed cpuset change |
| 375 | * follows the structural rules for cpusets. |
| 376 | * |
| 377 | * If we replaced the flag and mask values of the current cpuset |
| 378 | * (cur) with those values in the trial cpuset (trial), would |
| 379 | * our various subset and exclusive rules still be valid? Presumes |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 380 | * cgroup_mutex held. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 381 | * |
| 382 | * 'cur' is the address of an actual, in-use cpuset. Operations |
| 383 | * such as list traversal that depend on the actual address of the |
| 384 | * cpuset in the list must use cur below, not trial. |
| 385 | * |
| 386 | * 'trial' is the address of bulk structure copy of cur, with |
| 387 | * perhaps one or more of the fields cpus_allowed, mems_allowed, |
| 388 | * or flags changed to new, trial values. |
| 389 | * |
| 390 | * Return 0 if valid, -errno if not. |
| 391 | */ |
| 392 | |
| 393 | static int validate_change(const struct cpuset *cur, const struct cpuset *trial) |
| 394 | { |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 395 | struct cgroup *cont; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 396 | struct cpuset *c, *par; |
| 397 | |
| 398 | /* Each of our child cpusets must be a subset of us */ |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 399 | list_for_each_entry(cont, &cur->css.cgroup->children, sibling) { |
| 400 | if (!is_cpuset_subset(cgroup_cs(cont), trial)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 401 | return -EBUSY; |
| 402 | } |
| 403 | |
| 404 | /* Remaining checks don't apply to root cpuset */ |
Paul Jackson | 6960406 | 2006-12-06 20:36:15 -0800 | [diff] [blame] | 405 | if (cur == &top_cpuset) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 406 | return 0; |
| 407 | |
Paul Jackson | 6960406 | 2006-12-06 20:36:15 -0800 | [diff] [blame] | 408 | par = cur->parent; |
| 409 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 410 | /* We must be a subset of our parent cpuset */ |
| 411 | if (!is_cpuset_subset(trial, par)) |
| 412 | return -EACCES; |
| 413 | |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 414 | /* |
| 415 | * If either I or some sibling (!= me) is exclusive, we can't |
| 416 | * overlap |
| 417 | */ |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 418 | list_for_each_entry(cont, &par->css.cgroup->children, sibling) { |
| 419 | c = cgroup_cs(cont); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 420 | if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) && |
| 421 | c != cur && |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 422 | cpumask_intersects(trial->cpus_allowed, c->cpus_allowed)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 423 | return -EINVAL; |
| 424 | if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) && |
| 425 | c != cur && |
| 426 | nodes_intersects(trial->mems_allowed, c->mems_allowed)) |
| 427 | return -EINVAL; |
| 428 | } |
| 429 | |
Paul Jackson | 020958b | 2007-10-18 23:40:21 -0700 | [diff] [blame] | 430 | /* Cpusets with tasks can't have empty cpus_allowed or mems_allowed */ |
| 431 | if (cgroup_task_count(cur->css.cgroup)) { |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 432 | if (cpumask_empty(trial->cpus_allowed) || |
Paul Jackson | 020958b | 2007-10-18 23:40:21 -0700 | [diff] [blame] | 433 | nodes_empty(trial->mems_allowed)) { |
| 434 | return -ENOSPC; |
| 435 | } |
| 436 | } |
| 437 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 438 | return 0; |
| 439 | } |
| 440 | |
Paul Menage | db7f47c | 2009-04-02 16:57:55 -0700 | [diff] [blame] | 441 | #ifdef CONFIG_SMP |
Dinakar Guniguntala | 85d7b94 | 2005-06-25 14:57:34 -0700 | [diff] [blame] | 442 | /* |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 443 | * Helper routine for generate_sched_domains(). |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 444 | * Do cpusets a, b have overlapping cpus_allowed masks? |
| 445 | */ |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 446 | static int cpusets_overlap(struct cpuset *a, struct cpuset *b) |
| 447 | { |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 448 | return cpumask_intersects(a->cpus_allowed, b->cpus_allowed); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 449 | } |
| 450 | |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 451 | static void |
| 452 | update_domain_attr(struct sched_domain_attr *dattr, struct cpuset *c) |
| 453 | { |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 454 | if (dattr->relax_domain_level < c->relax_domain_level) |
| 455 | dattr->relax_domain_level = c->relax_domain_level; |
| 456 | return; |
| 457 | } |
| 458 | |
Lai Jiangshan | f539369 | 2008-07-29 22:33:22 -0700 | [diff] [blame] | 459 | static void |
| 460 | update_domain_attr_tree(struct sched_domain_attr *dattr, struct cpuset *c) |
| 461 | { |
| 462 | LIST_HEAD(q); |
| 463 | |
| 464 | list_add(&c->stack_list, &q); |
| 465 | while (!list_empty(&q)) { |
| 466 | struct cpuset *cp; |
| 467 | struct cgroup *cont; |
| 468 | struct cpuset *child; |
| 469 | |
| 470 | cp = list_first_entry(&q, struct cpuset, stack_list); |
| 471 | list_del(q.next); |
| 472 | |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 473 | if (cpumask_empty(cp->cpus_allowed)) |
Lai Jiangshan | f539369 | 2008-07-29 22:33:22 -0700 | [diff] [blame] | 474 | continue; |
| 475 | |
| 476 | if (is_sched_load_balance(cp)) |
| 477 | update_domain_attr(dattr, cp); |
| 478 | |
| 479 | list_for_each_entry(cont, &cp->css.cgroup->children, sibling) { |
| 480 | child = cgroup_cs(cont); |
| 481 | list_add_tail(&child->stack_list, &q); |
| 482 | } |
| 483 | } |
| 484 | } |
| 485 | |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 486 | /* |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 487 | * generate_sched_domains() |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 488 | * |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 489 | * This function builds a partial partition of the systems CPUs |
| 490 | * A 'partial partition' is a set of non-overlapping subsets whose |
| 491 | * union is a subset of that set. |
| 492 | * The output of this function needs to be passed to kernel/sched.c |
| 493 | * partition_sched_domains() routine, which will rebuild the scheduler's |
| 494 | * load balancing domains (sched domains) as specified by that partial |
| 495 | * partition. |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 496 | * |
Li Zefan | 45ce80f | 2009-01-15 13:50:59 -0800 | [diff] [blame] | 497 | * See "What is sched_load_balance" in Documentation/cgroups/cpusets.txt |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 498 | * for a background explanation of this. |
| 499 | * |
| 500 | * Does not return errors, on the theory that the callers of this |
| 501 | * routine would rather not worry about failures to rebuild sched |
| 502 | * domains when operating in the severe memory shortage situations |
| 503 | * that could cause allocation failures below. |
| 504 | * |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 505 | * Must be called with cgroup_lock held. |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 506 | * |
| 507 | * The three key local variables below are: |
Li Zefan | aeed682 | 2008-07-29 22:33:24 -0700 | [diff] [blame] | 508 | * q - a linked-list queue of cpuset pointers, used to implement a |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 509 | * top-down scan of all cpusets. This scan loads a pointer |
| 510 | * to each cpuset marked is_sched_load_balance into the |
| 511 | * array 'csa'. For our purposes, rebuilding the schedulers |
| 512 | * sched domains, we can ignore !is_sched_load_balance cpusets. |
| 513 | * csa - (for CpuSet Array) Array of pointers to all the cpusets |
| 514 | * that need to be load balanced, for convenient iterative |
| 515 | * access by the subsequent code that finds the best partition, |
| 516 | * i.e the set of domains (subsets) of CPUs such that the |
| 517 | * cpus_allowed of every cpuset marked is_sched_load_balance |
| 518 | * is a subset of one of these domains, while there are as |
| 519 | * many such domains as possible, each as small as possible. |
| 520 | * doms - Conversion of 'csa' to an array of cpumasks, for passing to |
| 521 | * the kernel/sched.c routine partition_sched_domains() in a |
| 522 | * convenient format, that can be easily compared to the prior |
| 523 | * value to determine what partition elements (sched domains) |
| 524 | * were changed (added or removed.) |
| 525 | * |
| 526 | * Finding the best partition (set of domains): |
| 527 | * The triple nested loops below over i, j, k scan over the |
| 528 | * load balanced cpusets (using the array of cpuset pointers in |
| 529 | * csa[]) looking for pairs of cpusets that have overlapping |
| 530 | * cpus_allowed, but which don't have the same 'pn' partition |
| 531 | * number and gives them in the same partition number. It keeps |
| 532 | * looping on the 'restart' label until it can no longer find |
| 533 | * any such pairs. |
| 534 | * |
| 535 | * The union of the cpus_allowed masks from the set of |
| 536 | * all cpusets having the same 'pn' value then form the one |
| 537 | * element of the partition (one sched domain) to be passed to |
| 538 | * partition_sched_domains(). |
| 539 | */ |
Li Zefan | 6af866a | 2009-01-07 18:08:45 -0800 | [diff] [blame] | 540 | /* FIXME: see the FIXME in partition_sched_domains() */ |
| 541 | static int generate_sched_domains(struct cpumask **domains, |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 542 | struct sched_domain_attr **attributes) |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 543 | { |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 544 | LIST_HEAD(q); /* queue of cpusets to be scanned */ |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 545 | struct cpuset *cp; /* scans q */ |
| 546 | struct cpuset **csa; /* array of all cpuset ptrs */ |
| 547 | int csn; /* how many cpuset ptrs in csa so far */ |
| 548 | int i, j, k; /* indices for partition finding loops */ |
Li Zefan | 6af866a | 2009-01-07 18:08:45 -0800 | [diff] [blame] | 549 | struct cpumask *doms; /* resulting partition; i.e. sched domains */ |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 550 | struct sched_domain_attr *dattr; /* attributes for custom domains */ |
Ingo Molnar | 1583715 | 2008-11-25 10:27:49 +0100 | [diff] [blame] | 551 | int ndoms = 0; /* number of sched domains in result */ |
Li Zefan | 6af866a | 2009-01-07 18:08:45 -0800 | [diff] [blame] | 552 | int nslot; /* next empty doms[] struct cpumask slot */ |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 553 | |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 554 | doms = NULL; |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 555 | dattr = NULL; |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 556 | csa = NULL; |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 557 | |
| 558 | /* Special case for the 99% of systems with one, full, sched domain */ |
| 559 | if (is_sched_load_balance(&top_cpuset)) { |
Li Zefan | 6af866a | 2009-01-07 18:08:45 -0800 | [diff] [blame] | 560 | doms = kmalloc(cpumask_size(), GFP_KERNEL); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 561 | if (!doms) |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 562 | goto done; |
| 563 | |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 564 | dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL); |
| 565 | if (dattr) { |
| 566 | *dattr = SD_ATTR_INIT; |
Li Zefan | 93a6557 | 2008-07-29 22:33:23 -0700 | [diff] [blame] | 567 | update_domain_attr_tree(dattr, &top_cpuset); |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 568 | } |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 569 | cpumask_copy(doms, top_cpuset.cpus_allowed); |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 570 | |
| 571 | ndoms = 1; |
| 572 | goto done; |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 573 | } |
| 574 | |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 575 | csa = kmalloc(number_of_cpusets * sizeof(cp), GFP_KERNEL); |
| 576 | if (!csa) |
| 577 | goto done; |
| 578 | csn = 0; |
| 579 | |
Li Zefan | aeed682 | 2008-07-29 22:33:24 -0700 | [diff] [blame] | 580 | list_add(&top_cpuset.stack_list, &q); |
| 581 | while (!list_empty(&q)) { |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 582 | struct cgroup *cont; |
| 583 | struct cpuset *child; /* scans child cpusets of cp */ |
Lai Jiangshan | 489a539 | 2008-07-25 01:47:23 -0700 | [diff] [blame] | 584 | |
Li Zefan | aeed682 | 2008-07-29 22:33:24 -0700 | [diff] [blame] | 585 | cp = list_first_entry(&q, struct cpuset, stack_list); |
| 586 | list_del(q.next); |
| 587 | |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 588 | if (cpumask_empty(cp->cpus_allowed)) |
Lai Jiangshan | 489a539 | 2008-07-25 01:47:23 -0700 | [diff] [blame] | 589 | continue; |
| 590 | |
Lai Jiangshan | f539369 | 2008-07-29 22:33:22 -0700 | [diff] [blame] | 591 | /* |
| 592 | * All child cpusets contain a subset of the parent's cpus, so |
| 593 | * just skip them, and then we call update_domain_attr_tree() |
| 594 | * to calc relax_domain_level of the corresponding sched |
| 595 | * domain. |
| 596 | */ |
| 597 | if (is_sched_load_balance(cp)) { |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 598 | csa[csn++] = cp; |
Lai Jiangshan | f539369 | 2008-07-29 22:33:22 -0700 | [diff] [blame] | 599 | continue; |
| 600 | } |
Lai Jiangshan | 489a539 | 2008-07-25 01:47:23 -0700 | [diff] [blame] | 601 | |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 602 | list_for_each_entry(cont, &cp->css.cgroup->children, sibling) { |
| 603 | child = cgroup_cs(cont); |
Li Zefan | aeed682 | 2008-07-29 22:33:24 -0700 | [diff] [blame] | 604 | list_add_tail(&child->stack_list, &q); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 605 | } |
| 606 | } |
| 607 | |
| 608 | for (i = 0; i < csn; i++) |
| 609 | csa[i]->pn = i; |
| 610 | ndoms = csn; |
| 611 | |
| 612 | restart: |
| 613 | /* Find the best partition (set of sched domains) */ |
| 614 | for (i = 0; i < csn; i++) { |
| 615 | struct cpuset *a = csa[i]; |
| 616 | int apn = a->pn; |
| 617 | |
| 618 | for (j = 0; j < csn; j++) { |
| 619 | struct cpuset *b = csa[j]; |
| 620 | int bpn = b->pn; |
| 621 | |
| 622 | if (apn != bpn && cpusets_overlap(a, b)) { |
| 623 | for (k = 0; k < csn; k++) { |
| 624 | struct cpuset *c = csa[k]; |
| 625 | |
| 626 | if (c->pn == bpn) |
| 627 | c->pn = apn; |
| 628 | } |
| 629 | ndoms--; /* one less element */ |
| 630 | goto restart; |
| 631 | } |
| 632 | } |
| 633 | } |
| 634 | |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 635 | /* |
| 636 | * Now we know how many domains to create. |
| 637 | * Convert <csn, csa> to <ndoms, doms> and populate cpu masks. |
| 638 | */ |
Li Zefan | 6af866a | 2009-01-07 18:08:45 -0800 | [diff] [blame] | 639 | doms = kmalloc(ndoms * cpumask_size(), GFP_KERNEL); |
Li Zefan | 700018e | 2008-11-18 14:02:03 +0800 | [diff] [blame] | 640 | if (!doms) |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 641 | goto done; |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 642 | |
| 643 | /* |
| 644 | * The rest of the code, including the scheduler, can deal with |
| 645 | * dattr==NULL case. No need to abort if alloc fails. |
| 646 | */ |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 647 | dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 648 | |
| 649 | for (nslot = 0, i = 0; i < csn; i++) { |
| 650 | struct cpuset *a = csa[i]; |
Li Zefan | 6af866a | 2009-01-07 18:08:45 -0800 | [diff] [blame] | 651 | struct cpumask *dp; |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 652 | int apn = a->pn; |
| 653 | |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 654 | if (apn < 0) { |
| 655 | /* Skip completed partitions */ |
| 656 | continue; |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 657 | } |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 658 | |
| 659 | dp = doms + nslot; |
| 660 | |
| 661 | if (nslot == ndoms) { |
| 662 | static int warnings = 10; |
| 663 | if (warnings) { |
| 664 | printk(KERN_WARNING |
| 665 | "rebuild_sched_domains confused:" |
| 666 | " nslot %d, ndoms %d, csn %d, i %d," |
| 667 | " apn %d\n", |
| 668 | nslot, ndoms, csn, i, apn); |
| 669 | warnings--; |
| 670 | } |
| 671 | continue; |
| 672 | } |
| 673 | |
Li Zefan | 6af866a | 2009-01-07 18:08:45 -0800 | [diff] [blame] | 674 | cpumask_clear(dp); |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 675 | if (dattr) |
| 676 | *(dattr + nslot) = SD_ATTR_INIT; |
| 677 | for (j = i; j < csn; j++) { |
| 678 | struct cpuset *b = csa[j]; |
| 679 | |
| 680 | if (apn == b->pn) { |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 681 | cpumask_or(dp, dp, b->cpus_allowed); |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 682 | if (dattr) |
| 683 | update_domain_attr_tree(dattr + nslot, b); |
| 684 | |
| 685 | /* Done with this partition */ |
| 686 | b->pn = -1; |
| 687 | } |
| 688 | } |
| 689 | nslot++; |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 690 | } |
| 691 | BUG_ON(nslot != ndoms); |
| 692 | |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 693 | done: |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 694 | kfree(csa); |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 695 | |
Li Zefan | 700018e | 2008-11-18 14:02:03 +0800 | [diff] [blame] | 696 | /* |
| 697 | * Fallback to the default domain if kmalloc() failed. |
| 698 | * See comments in partition_sched_domains(). |
| 699 | */ |
| 700 | if (doms == NULL) |
| 701 | ndoms = 1; |
| 702 | |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 703 | *domains = doms; |
| 704 | *attributes = dattr; |
| 705 | return ndoms; |
| 706 | } |
| 707 | |
| 708 | /* |
| 709 | * Rebuild scheduler domains. |
| 710 | * |
| 711 | * Call with neither cgroup_mutex held nor within get_online_cpus(). |
| 712 | * Takes both cgroup_mutex and get_online_cpus(). |
| 713 | * |
| 714 | * Cannot be directly called from cpuset code handling changes |
| 715 | * to the cpuset pseudo-filesystem, because it cannot be called |
| 716 | * from code that already holds cgroup_mutex. |
| 717 | */ |
| 718 | static void do_rebuild_sched_domains(struct work_struct *unused) |
| 719 | { |
| 720 | struct sched_domain_attr *attr; |
Li Zefan | 6af866a | 2009-01-07 18:08:45 -0800 | [diff] [blame] | 721 | struct cpumask *doms; |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 722 | int ndoms; |
| 723 | |
| 724 | get_online_cpus(); |
| 725 | |
| 726 | /* Generate domain masks and attrs */ |
| 727 | cgroup_lock(); |
| 728 | ndoms = generate_sched_domains(&doms, &attr); |
| 729 | cgroup_unlock(); |
| 730 | |
| 731 | /* Have scheduler rebuild the domains */ |
| 732 | partition_sched_domains(ndoms, doms, attr); |
| 733 | |
| 734 | put_online_cpus(); |
| 735 | } |
Paul Menage | db7f47c | 2009-04-02 16:57:55 -0700 | [diff] [blame] | 736 | #else /* !CONFIG_SMP */ |
| 737 | static void do_rebuild_sched_domains(struct work_struct *unused) |
| 738 | { |
| 739 | } |
| 740 | |
| 741 | static int generate_sched_domains(struct cpumask **domains, |
| 742 | struct sched_domain_attr **attributes) |
| 743 | { |
| 744 | *domains = NULL; |
| 745 | return 1; |
| 746 | } |
| 747 | #endif /* CONFIG_SMP */ |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 748 | |
| 749 | static DECLARE_WORK(rebuild_sched_domains_work, do_rebuild_sched_domains); |
| 750 | |
| 751 | /* |
| 752 | * Rebuild scheduler domains, asynchronously via workqueue. |
| 753 | * |
| 754 | * If the flag 'sched_load_balance' of any cpuset with non-empty |
| 755 | * 'cpus' changes, or if the 'cpus' allowed changes in any cpuset |
| 756 | * which has that flag enabled, or if any cpuset with a non-empty |
| 757 | * 'cpus' is removed, then call this routine to rebuild the |
| 758 | * scheduler's dynamic sched domains. |
| 759 | * |
| 760 | * The rebuild_sched_domains() and partition_sched_domains() |
| 761 | * routines must nest cgroup_lock() inside get_online_cpus(), |
| 762 | * but such cpuset changes as these must nest that locking the |
| 763 | * other way, holding cgroup_lock() for much of the code. |
| 764 | * |
| 765 | * So in order to avoid an ABBA deadlock, the cpuset code handling |
| 766 | * these user changes delegates the actual sched domain rebuilding |
| 767 | * to a separate workqueue thread, which ends up processing the |
| 768 | * above do_rebuild_sched_domains() function. |
| 769 | */ |
| 770 | static void async_rebuild_sched_domains(void) |
| 771 | { |
Miao Xie | f90d411 | 2009-01-16 10:24:10 +0800 | [diff] [blame] | 772 | queue_work(cpuset_wq, &rebuild_sched_domains_work); |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 773 | } |
| 774 | |
| 775 | /* |
| 776 | * Accomplishes the same scheduler domain rebuild as the above |
| 777 | * async_rebuild_sched_domains(), however it directly calls the |
| 778 | * rebuild routine synchronously rather than calling it via an |
| 779 | * asynchronous work thread. |
| 780 | * |
| 781 | * This can only be called from code that is not holding |
| 782 | * cgroup_mutex (not nested in a cgroup_lock() call.) |
| 783 | */ |
| 784 | void rebuild_sched_domains(void) |
| 785 | { |
| 786 | do_rebuild_sched_domains(NULL); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 787 | } |
| 788 | |
Cliff Wickman | 58f4790 | 2008-02-07 00:14:44 -0800 | [diff] [blame] | 789 | /** |
| 790 | * cpuset_test_cpumask - test a task's cpus_allowed versus its cpuset's |
| 791 | * @tsk: task to test |
| 792 | * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner |
| 793 | * |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 794 | * Call with cgroup_mutex held. May take callback_mutex during call. |
Cliff Wickman | 58f4790 | 2008-02-07 00:14:44 -0800 | [diff] [blame] | 795 | * Called for each task in a cgroup by cgroup_scan_tasks(). |
| 796 | * Return nonzero if this tasks's cpus_allowed mask should be changed (in other |
| 797 | * words, if its mask is not equal to its cpuset's mask). |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 798 | */ |
Adrian Bunk | 9e0c914 | 2008-04-29 01:00:25 -0700 | [diff] [blame] | 799 | static int cpuset_test_cpumask(struct task_struct *tsk, |
| 800 | struct cgroup_scanner *scan) |
Cliff Wickman | 58f4790 | 2008-02-07 00:14:44 -0800 | [diff] [blame] | 801 | { |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 802 | return !cpumask_equal(&tsk->cpus_allowed, |
Cliff Wickman | 58f4790 | 2008-02-07 00:14:44 -0800 | [diff] [blame] | 803 | (cgroup_cs(scan->cg))->cpus_allowed); |
| 804 | } |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 805 | |
Cliff Wickman | 58f4790 | 2008-02-07 00:14:44 -0800 | [diff] [blame] | 806 | /** |
| 807 | * cpuset_change_cpumask - make a task's cpus_allowed the same as its cpuset's |
| 808 | * @tsk: task to test |
| 809 | * @scan: struct cgroup_scanner containing the cgroup of the task |
| 810 | * |
| 811 | * Called by cgroup_scan_tasks() for each task in a cgroup whose |
| 812 | * cpus_allowed mask needs to be changed. |
| 813 | * |
| 814 | * We don't need to re-check for the cgroup/cpuset membership, since we're |
| 815 | * holding cgroup_lock() at this point. |
| 816 | */ |
Adrian Bunk | 9e0c914 | 2008-04-29 01:00:25 -0700 | [diff] [blame] | 817 | static void cpuset_change_cpumask(struct task_struct *tsk, |
| 818 | struct cgroup_scanner *scan) |
Cliff Wickman | 58f4790 | 2008-02-07 00:14:44 -0800 | [diff] [blame] | 819 | { |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 820 | set_cpus_allowed_ptr(tsk, ((cgroup_cs(scan->cg))->cpus_allowed)); |
Cliff Wickman | 58f4790 | 2008-02-07 00:14:44 -0800 | [diff] [blame] | 821 | } |
| 822 | |
| 823 | /** |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 824 | * update_tasks_cpumask - Update the cpumasks of tasks in the cpuset. |
| 825 | * @cs: the cpuset in which each task's cpus_allowed mask needs to be changed |
Li Zefan | 4e74339 | 2008-09-13 02:33:08 -0700 | [diff] [blame] | 826 | * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks() |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 827 | * |
| 828 | * Called with cgroup_mutex held |
| 829 | * |
| 830 | * The cgroup_scan_tasks() function will scan all the tasks in a cgroup, |
| 831 | * calling callback functions for each. |
| 832 | * |
Li Zefan | 4e74339 | 2008-09-13 02:33:08 -0700 | [diff] [blame] | 833 | * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0 |
| 834 | * if @heap != NULL. |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 835 | */ |
Li Zefan | 4e74339 | 2008-09-13 02:33:08 -0700 | [diff] [blame] | 836 | static void update_tasks_cpumask(struct cpuset *cs, struct ptr_heap *heap) |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 837 | { |
| 838 | struct cgroup_scanner scan; |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 839 | |
| 840 | scan.cg = cs->css.cgroup; |
| 841 | scan.test_task = cpuset_test_cpumask; |
| 842 | scan.process_task = cpuset_change_cpumask; |
Li Zefan | 4e74339 | 2008-09-13 02:33:08 -0700 | [diff] [blame] | 843 | scan.heap = heap; |
| 844 | cgroup_scan_tasks(&scan); |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 845 | } |
| 846 | |
| 847 | /** |
Cliff Wickman | 58f4790 | 2008-02-07 00:14:44 -0800 | [diff] [blame] | 848 | * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it |
| 849 | * @cs: the cpuset to consider |
| 850 | * @buf: buffer of cpu numbers written to this cpuset |
| 851 | */ |
Li Zefan | 645fcc9 | 2009-01-07 18:08:43 -0800 | [diff] [blame] | 852 | static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs, |
| 853 | const char *buf) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 854 | { |
Li Zefan | 4e74339 | 2008-09-13 02:33:08 -0700 | [diff] [blame] | 855 | struct ptr_heap heap; |
Cliff Wickman | 58f4790 | 2008-02-07 00:14:44 -0800 | [diff] [blame] | 856 | int retval; |
| 857 | int is_load_balanced; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 858 | |
Paul Jackson | 4c4d50f | 2006-08-27 01:23:51 -0700 | [diff] [blame] | 859 | /* top_cpuset.cpus_allowed tracks cpu_online_map; it's read-only */ |
| 860 | if (cs == &top_cpuset) |
| 861 | return -EACCES; |
| 862 | |
David Rientjes | 6f7f02e | 2007-05-08 00:31:43 -0700 | [diff] [blame] | 863 | /* |
Paul Jackson | c8d9c90 | 2008-02-07 00:14:46 -0800 | [diff] [blame] | 864 | * An empty cpus_allowed is ok only if the cpuset has no tasks. |
Paul Jackson | 020958b | 2007-10-18 23:40:21 -0700 | [diff] [blame] | 865 | * Since cpulist_parse() fails on an empty mask, we special case |
| 866 | * that parsing. The validate_change() call ensures that cpusets |
| 867 | * with tasks have cpus. |
David Rientjes | 6f7f02e | 2007-05-08 00:31:43 -0700 | [diff] [blame] | 868 | */ |
Paul Jackson | 020958b | 2007-10-18 23:40:21 -0700 | [diff] [blame] | 869 | if (!*buf) { |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 870 | cpumask_clear(trialcs->cpus_allowed); |
David Rientjes | 6f7f02e | 2007-05-08 00:31:43 -0700 | [diff] [blame] | 871 | } else { |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 872 | retval = cpulist_parse(buf, trialcs->cpus_allowed); |
David Rientjes | 6f7f02e | 2007-05-08 00:31:43 -0700 | [diff] [blame] | 873 | if (retval < 0) |
| 874 | return retval; |
Lai Jiangshan | 3734074 | 2008-06-05 22:46:32 -0700 | [diff] [blame] | 875 | |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 876 | if (!cpumask_subset(trialcs->cpus_allowed, cpu_online_mask)) |
Lai Jiangshan | 3734074 | 2008-06-05 22:46:32 -0700 | [diff] [blame] | 877 | return -EINVAL; |
David Rientjes | 6f7f02e | 2007-05-08 00:31:43 -0700 | [diff] [blame] | 878 | } |
Li Zefan | 645fcc9 | 2009-01-07 18:08:43 -0800 | [diff] [blame] | 879 | retval = validate_change(cs, trialcs); |
Dinakar Guniguntala | 85d7b94 | 2005-06-25 14:57:34 -0700 | [diff] [blame] | 880 | if (retval < 0) |
| 881 | return retval; |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 882 | |
Paul Menage | 8707d8b | 2007-10-18 23:40:22 -0700 | [diff] [blame] | 883 | /* Nothing to do if the cpus didn't change */ |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 884 | if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed)) |
Paul Menage | 8707d8b | 2007-10-18 23:40:22 -0700 | [diff] [blame] | 885 | return 0; |
Cliff Wickman | 58f4790 | 2008-02-07 00:14:44 -0800 | [diff] [blame] | 886 | |
Li Zefan | 4e74339 | 2008-09-13 02:33:08 -0700 | [diff] [blame] | 887 | retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL); |
| 888 | if (retval) |
| 889 | return retval; |
| 890 | |
Li Zefan | 645fcc9 | 2009-01-07 18:08:43 -0800 | [diff] [blame] | 891 | is_load_balanced = is_sched_load_balance(trialcs); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 892 | |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 893 | mutex_lock(&callback_mutex); |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 894 | cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed); |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 895 | mutex_unlock(&callback_mutex); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 896 | |
Paul Menage | 8707d8b | 2007-10-18 23:40:22 -0700 | [diff] [blame] | 897 | /* |
| 898 | * Scan tasks in the cpuset, and update the cpumasks of any |
Cliff Wickman | 58f4790 | 2008-02-07 00:14:44 -0800 | [diff] [blame] | 899 | * that need an update. |
Paul Menage | 8707d8b | 2007-10-18 23:40:22 -0700 | [diff] [blame] | 900 | */ |
Li Zefan | 4e74339 | 2008-09-13 02:33:08 -0700 | [diff] [blame] | 901 | update_tasks_cpumask(cs, &heap); |
| 902 | |
| 903 | heap_free(&heap); |
Cliff Wickman | 58f4790 | 2008-02-07 00:14:44 -0800 | [diff] [blame] | 904 | |
Paul Menage | 8707d8b | 2007-10-18 23:40:22 -0700 | [diff] [blame] | 905 | if (is_load_balanced) |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 906 | async_rebuild_sched_domains(); |
Dinakar Guniguntala | 85d7b94 | 2005-06-25 14:57:34 -0700 | [diff] [blame] | 907 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 908 | } |
| 909 | |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 910 | /* |
Paul Jackson | e4e364e | 2006-03-31 02:30:52 -0800 | [diff] [blame] | 911 | * cpuset_migrate_mm |
| 912 | * |
| 913 | * Migrate memory region from one set of nodes to another. |
| 914 | * |
| 915 | * Temporarilly set tasks mems_allowed to target nodes of migration, |
| 916 | * so that the migration code can allocate pages on these nodes. |
| 917 | * |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 918 | * Call holding cgroup_mutex, so current's cpuset won't change |
Paul Jackson | c8d9c90 | 2008-02-07 00:14:46 -0800 | [diff] [blame] | 919 | * during this call, as manage_mutex holds off any cpuset_attach() |
Paul Jackson | e4e364e | 2006-03-31 02:30:52 -0800 | [diff] [blame] | 920 | * calls. Therefore we don't need to take task_lock around the |
| 921 | * call to guarantee_online_mems(), as we know no one is changing |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 922 | * our task's cpuset. |
Paul Jackson | e4e364e | 2006-03-31 02:30:52 -0800 | [diff] [blame] | 923 | * |
| 924 | * Hold callback_mutex around the two modifications of our tasks |
| 925 | * mems_allowed to synchronize with cpuset_mems_allowed(). |
| 926 | * |
| 927 | * While the mm_struct we are migrating is typically from some |
| 928 | * other task, the task_struct mems_allowed that we are hacking |
| 929 | * is for our current task, which must allocate new pages for that |
| 930 | * migrating memory region. |
Paul Jackson | e4e364e | 2006-03-31 02:30:52 -0800 | [diff] [blame] | 931 | */ |
| 932 | |
| 933 | static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from, |
| 934 | const nodemask_t *to) |
| 935 | { |
| 936 | struct task_struct *tsk = current; |
| 937 | |
Paul Jackson | e4e364e | 2006-03-31 02:30:52 -0800 | [diff] [blame] | 938 | tsk->mems_allowed = *to; |
Paul Jackson | e4e364e | 2006-03-31 02:30:52 -0800 | [diff] [blame] | 939 | |
| 940 | do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL); |
| 941 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 942 | guarantee_online_mems(task_cs(tsk),&tsk->mems_allowed); |
Paul Jackson | e4e364e | 2006-03-31 02:30:52 -0800 | [diff] [blame] | 943 | } |
| 944 | |
Li Zefan | 3b6766f | 2009-04-02 16:57:51 -0700 | [diff] [blame] | 945 | /* |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 946 | * cpuset_change_task_nodemask - change task's mems_allowed and mempolicy |
| 947 | * @tsk: the task to change |
| 948 | * @newmems: new nodes that the task will be set |
| 949 | * |
| 950 | * In order to avoid seeing no nodes if the old and new nodes are disjoint, |
| 951 | * we structure updates as setting all new allowed nodes, then clearing newly |
| 952 | * disallowed ones. |
| 953 | * |
| 954 | * Called with task's alloc_lock held |
| 955 | */ |
| 956 | static void cpuset_change_task_nodemask(struct task_struct *tsk, |
| 957 | nodemask_t *newmems) |
| 958 | { |
| 959 | nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems); |
| 960 | mpol_rebind_task(tsk, &tsk->mems_allowed); |
| 961 | mpol_rebind_task(tsk, newmems); |
| 962 | tsk->mems_allowed = *newmems; |
| 963 | } |
| 964 | |
| 965 | /* |
| 966 | * Update task's mems_allowed and rebind its mempolicy and vmas' mempolicy |
| 967 | * of it to cpuset's new mems_allowed, and migrate pages to new nodes if |
| 968 | * memory_migrate flag is set. Called with cgroup_mutex held. |
Li Zefan | 3b6766f | 2009-04-02 16:57:51 -0700 | [diff] [blame] | 969 | */ |
| 970 | static void cpuset_change_nodemask(struct task_struct *p, |
| 971 | struct cgroup_scanner *scan) |
| 972 | { |
| 973 | struct mm_struct *mm; |
| 974 | struct cpuset *cs; |
| 975 | int migrate; |
| 976 | const nodemask_t *oldmem = scan->data; |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 977 | nodemask_t newmems; |
| 978 | |
| 979 | cs = cgroup_cs(scan->cg); |
| 980 | guarantee_online_mems(cs, &newmems); |
| 981 | |
| 982 | task_lock(p); |
| 983 | cpuset_change_task_nodemask(p, &newmems); |
| 984 | task_unlock(p); |
Li Zefan | 3b6766f | 2009-04-02 16:57:51 -0700 | [diff] [blame] | 985 | |
| 986 | mm = get_task_mm(p); |
| 987 | if (!mm) |
| 988 | return; |
| 989 | |
Li Zefan | 3b6766f | 2009-04-02 16:57:51 -0700 | [diff] [blame] | 990 | migrate = is_memory_migrate(cs); |
| 991 | |
| 992 | mpol_rebind_mm(mm, &cs->mems_allowed); |
| 993 | if (migrate) |
| 994 | cpuset_migrate_mm(mm, oldmem, &cs->mems_allowed); |
| 995 | mmput(mm); |
| 996 | } |
| 997 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 998 | static void *cpuset_being_rebound; |
| 999 | |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 1000 | /** |
| 1001 | * update_tasks_nodemask - Update the nodemasks of tasks in the cpuset. |
| 1002 | * @cs: the cpuset in which each task's mems_allowed mask needs to be changed |
| 1003 | * @oldmem: old mems_allowed of cpuset cs |
Li Zefan | 010cfac | 2009-04-02 16:57:52 -0700 | [diff] [blame] | 1004 | * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks() |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 1005 | * |
| 1006 | * Called with cgroup_mutex held |
Li Zefan | 010cfac | 2009-04-02 16:57:52 -0700 | [diff] [blame] | 1007 | * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0 |
| 1008 | * if @heap != NULL. |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 1009 | */ |
Li Zefan | 010cfac | 2009-04-02 16:57:52 -0700 | [diff] [blame] | 1010 | static void update_tasks_nodemask(struct cpuset *cs, const nodemask_t *oldmem, |
| 1011 | struct ptr_heap *heap) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1012 | { |
Li Zefan | 3b6766f | 2009-04-02 16:57:51 -0700 | [diff] [blame] | 1013 | struct cgroup_scanner scan; |
Paul Jackson | 59dac16 | 2006-01-08 01:01:52 -0800 | [diff] [blame] | 1014 | |
Lee Schermerhorn | 846a16b | 2008-04-28 02:13:09 -0700 | [diff] [blame] | 1015 | cpuset_being_rebound = cs; /* causes mpol_dup() rebind */ |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 1016 | |
Li Zefan | 3b6766f | 2009-04-02 16:57:51 -0700 | [diff] [blame] | 1017 | scan.cg = cs->css.cgroup; |
| 1018 | scan.test_task = NULL; |
| 1019 | scan.process_task = cpuset_change_nodemask; |
Li Zefan | 010cfac | 2009-04-02 16:57:52 -0700 | [diff] [blame] | 1020 | scan.heap = heap; |
Li Zefan | 3b6766f | 2009-04-02 16:57:51 -0700 | [diff] [blame] | 1021 | scan.data = (nodemask_t *)oldmem; |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 1022 | |
| 1023 | /* |
Li Zefan | 3b6766f | 2009-04-02 16:57:51 -0700 | [diff] [blame] | 1024 | * The mpol_rebind_mm() call takes mmap_sem, which we couldn't |
| 1025 | * take while holding tasklist_lock. Forks can happen - the |
| 1026 | * mpol_dup() cpuset_being_rebound check will catch such forks, |
| 1027 | * and rebind their vma mempolicies too. Because we still hold |
| 1028 | * the global cgroup_mutex, we know that no other rebind effort |
| 1029 | * will be contending for the global variable cpuset_being_rebound. |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 1030 | * It's ok if we rebind the same mm twice; mpol_rebind_mm() |
Paul Jackson | 04c19fa | 2006-01-08 01:02:00 -0800 | [diff] [blame] | 1031 | * is idempotent. Also migrate pages in each mm to new nodes. |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 1032 | */ |
Li Zefan | 010cfac | 2009-04-02 16:57:52 -0700 | [diff] [blame] | 1033 | cgroup_scan_tasks(&scan); |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 1034 | |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 1035 | /* We're done rebinding vmas to this cpuset's new mems_allowed. */ |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1036 | cpuset_being_rebound = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1037 | } |
| 1038 | |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 1039 | /* |
| 1040 | * Handle user request to change the 'mems' memory placement |
| 1041 | * of a cpuset. Needs to validate the request, update the |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 1042 | * cpusets mems_allowed, and for each task in the cpuset, |
| 1043 | * update mems_allowed and rebind task's mempolicy and any vma |
| 1044 | * mempolicies and if the cpuset is marked 'memory_migrate', |
| 1045 | * migrate the tasks pages to the new memory. |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 1046 | * |
| 1047 | * Call with cgroup_mutex held. May take callback_mutex during call. |
| 1048 | * Will take tasklist_lock, scan tasklist for tasks in cpuset cs, |
| 1049 | * lock each such tasks mm->mmap_sem, scan its vma's and rebind |
| 1050 | * their mempolicies to the cpusets new mems_allowed. |
| 1051 | */ |
Li Zefan | 645fcc9 | 2009-01-07 18:08:43 -0800 | [diff] [blame] | 1052 | static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs, |
| 1053 | const char *buf) |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 1054 | { |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 1055 | nodemask_t oldmem; |
| 1056 | int retval; |
Li Zefan | 010cfac | 2009-04-02 16:57:52 -0700 | [diff] [blame] | 1057 | struct ptr_heap heap; |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 1058 | |
| 1059 | /* |
| 1060 | * top_cpuset.mems_allowed tracks node_stats[N_HIGH_MEMORY]; |
| 1061 | * it's read-only |
| 1062 | */ |
| 1063 | if (cs == &top_cpuset) |
| 1064 | return -EACCES; |
| 1065 | |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 1066 | /* |
| 1067 | * An empty mems_allowed is ok iff there are no tasks in the cpuset. |
| 1068 | * Since nodelist_parse() fails on an empty mask, we special case |
| 1069 | * that parsing. The validate_change() call ensures that cpusets |
| 1070 | * with tasks have memory. |
| 1071 | */ |
| 1072 | if (!*buf) { |
Li Zefan | 645fcc9 | 2009-01-07 18:08:43 -0800 | [diff] [blame] | 1073 | nodes_clear(trialcs->mems_allowed); |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 1074 | } else { |
Li Zefan | 645fcc9 | 2009-01-07 18:08:43 -0800 | [diff] [blame] | 1075 | retval = nodelist_parse(buf, trialcs->mems_allowed); |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 1076 | if (retval < 0) |
| 1077 | goto done; |
| 1078 | |
Li Zefan | 645fcc9 | 2009-01-07 18:08:43 -0800 | [diff] [blame] | 1079 | if (!nodes_subset(trialcs->mems_allowed, |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 1080 | node_states[N_HIGH_MEMORY])) |
| 1081 | return -EINVAL; |
| 1082 | } |
| 1083 | oldmem = cs->mems_allowed; |
Li Zefan | 645fcc9 | 2009-01-07 18:08:43 -0800 | [diff] [blame] | 1084 | if (nodes_equal(oldmem, trialcs->mems_allowed)) { |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 1085 | retval = 0; /* Too easy - nothing to do */ |
| 1086 | goto done; |
| 1087 | } |
Li Zefan | 645fcc9 | 2009-01-07 18:08:43 -0800 | [diff] [blame] | 1088 | retval = validate_change(cs, trialcs); |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 1089 | if (retval < 0) |
| 1090 | goto done; |
| 1091 | |
Li Zefan | 010cfac | 2009-04-02 16:57:52 -0700 | [diff] [blame] | 1092 | retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL); |
| 1093 | if (retval < 0) |
| 1094 | goto done; |
| 1095 | |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 1096 | mutex_lock(&callback_mutex); |
Li Zefan | 645fcc9 | 2009-01-07 18:08:43 -0800 | [diff] [blame] | 1097 | cs->mems_allowed = trialcs->mems_allowed; |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 1098 | mutex_unlock(&callback_mutex); |
| 1099 | |
Li Zefan | 010cfac | 2009-04-02 16:57:52 -0700 | [diff] [blame] | 1100 | update_tasks_nodemask(cs, &oldmem, &heap); |
| 1101 | |
| 1102 | heap_free(&heap); |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 1103 | done: |
| 1104 | return retval; |
| 1105 | } |
| 1106 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1107 | int current_cpuset_is_being_rebound(void) |
| 1108 | { |
| 1109 | return task_cs(current) == cpuset_being_rebound; |
| 1110 | } |
| 1111 | |
Paul Menage | 5be7a47 | 2008-05-06 20:42:41 -0700 | [diff] [blame] | 1112 | static int update_relax_domain_level(struct cpuset *cs, s64 val) |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 1113 | { |
Paul Menage | db7f47c | 2009-04-02 16:57:55 -0700 | [diff] [blame] | 1114 | #ifdef CONFIG_SMP |
Li Zefan | 30e0e17 | 2008-05-13 10:27:17 +0800 | [diff] [blame] | 1115 | if (val < -1 || val >= SD_LV_MAX) |
| 1116 | return -EINVAL; |
Paul Menage | db7f47c | 2009-04-02 16:57:55 -0700 | [diff] [blame] | 1117 | #endif |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 1118 | |
| 1119 | if (val != cs->relax_domain_level) { |
| 1120 | cs->relax_domain_level = val; |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 1121 | if (!cpumask_empty(cs->cpus_allowed) && |
| 1122 | is_sched_load_balance(cs)) |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 1123 | async_rebuild_sched_domains(); |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 1124 | } |
| 1125 | |
| 1126 | return 0; |
| 1127 | } |
| 1128 | |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 1129 | /* |
Miao Xie | 950592f | 2009-06-16 15:31:47 -0700 | [diff] [blame] | 1130 | * cpuset_change_flag - make a task's spread flags the same as its cpuset's |
| 1131 | * @tsk: task to be updated |
| 1132 | * @scan: struct cgroup_scanner containing the cgroup of the task |
| 1133 | * |
| 1134 | * Called by cgroup_scan_tasks() for each task in a cgroup. |
| 1135 | * |
| 1136 | * We don't need to re-check for the cgroup/cpuset membership, since we're |
| 1137 | * holding cgroup_lock() at this point. |
| 1138 | */ |
| 1139 | static void cpuset_change_flag(struct task_struct *tsk, |
| 1140 | struct cgroup_scanner *scan) |
| 1141 | { |
| 1142 | cpuset_update_task_spread_flag(cgroup_cs(scan->cg), tsk); |
| 1143 | } |
| 1144 | |
| 1145 | /* |
| 1146 | * update_tasks_flags - update the spread flags of tasks in the cpuset. |
| 1147 | * @cs: the cpuset in which each task's spread flags needs to be changed |
| 1148 | * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks() |
| 1149 | * |
| 1150 | * Called with cgroup_mutex held |
| 1151 | * |
| 1152 | * The cgroup_scan_tasks() function will scan all the tasks in a cgroup, |
| 1153 | * calling callback functions for each. |
| 1154 | * |
| 1155 | * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0 |
| 1156 | * if @heap != NULL. |
| 1157 | */ |
| 1158 | static void update_tasks_flags(struct cpuset *cs, struct ptr_heap *heap) |
| 1159 | { |
| 1160 | struct cgroup_scanner scan; |
| 1161 | |
| 1162 | scan.cg = cs->css.cgroup; |
| 1163 | scan.test_task = NULL; |
| 1164 | scan.process_task = cpuset_change_flag; |
| 1165 | scan.heap = heap; |
| 1166 | cgroup_scan_tasks(&scan); |
| 1167 | } |
| 1168 | |
| 1169 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1170 | * update_flag - read a 0 or a 1 in a file and update associated flag |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1171 | * bit: the bit to update (see cpuset_flagbits_t) |
| 1172 | * cs: the cpuset to update |
| 1173 | * turning_on: whether the flag is being set or cleared |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 1174 | * |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 1175 | * Call with cgroup_mutex held. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1176 | */ |
| 1177 | |
Paul Menage | 700fe1a | 2008-04-29 01:00:00 -0700 | [diff] [blame] | 1178 | static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs, |
| 1179 | int turning_on) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1180 | { |
Li Zefan | 645fcc9 | 2009-01-07 18:08:43 -0800 | [diff] [blame] | 1181 | struct cpuset *trialcs; |
Rakib Mullick | 40b6a76 | 2008-10-18 20:28:18 -0700 | [diff] [blame] | 1182 | int balance_flag_changed; |
Miao Xie | 950592f | 2009-06-16 15:31:47 -0700 | [diff] [blame] | 1183 | int spread_flag_changed; |
| 1184 | struct ptr_heap heap; |
| 1185 | int err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1186 | |
Li Zefan | 645fcc9 | 2009-01-07 18:08:43 -0800 | [diff] [blame] | 1187 | trialcs = alloc_trial_cpuset(cs); |
| 1188 | if (!trialcs) |
| 1189 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1190 | |
Li Zefan | 645fcc9 | 2009-01-07 18:08:43 -0800 | [diff] [blame] | 1191 | if (turning_on) |
| 1192 | set_bit(bit, &trialcs->flags); |
| 1193 | else |
| 1194 | clear_bit(bit, &trialcs->flags); |
| 1195 | |
| 1196 | err = validate_change(cs, trialcs); |
Dinakar Guniguntala | 85d7b94 | 2005-06-25 14:57:34 -0700 | [diff] [blame] | 1197 | if (err < 0) |
Li Zefan | 645fcc9 | 2009-01-07 18:08:43 -0800 | [diff] [blame] | 1198 | goto out; |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 1199 | |
Miao Xie | 950592f | 2009-06-16 15:31:47 -0700 | [diff] [blame] | 1200 | err = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL); |
| 1201 | if (err < 0) |
| 1202 | goto out; |
| 1203 | |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 1204 | balance_flag_changed = (is_sched_load_balance(cs) != |
Li Zefan | 645fcc9 | 2009-01-07 18:08:43 -0800 | [diff] [blame] | 1205 | is_sched_load_balance(trialcs)); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 1206 | |
Miao Xie | 950592f | 2009-06-16 15:31:47 -0700 | [diff] [blame] | 1207 | spread_flag_changed = ((is_spread_slab(cs) != is_spread_slab(trialcs)) |
| 1208 | || (is_spread_page(cs) != is_spread_page(trialcs))); |
| 1209 | |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 1210 | mutex_lock(&callback_mutex); |
Li Zefan | 645fcc9 | 2009-01-07 18:08:43 -0800 | [diff] [blame] | 1211 | cs->flags = trialcs->flags; |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 1212 | mutex_unlock(&callback_mutex); |
Dinakar Guniguntala | 85d7b94 | 2005-06-25 14:57:34 -0700 | [diff] [blame] | 1213 | |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 1214 | if (!cpumask_empty(trialcs->cpus_allowed) && balance_flag_changed) |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 1215 | async_rebuild_sched_domains(); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 1216 | |
Miao Xie | 950592f | 2009-06-16 15:31:47 -0700 | [diff] [blame] | 1217 | if (spread_flag_changed) |
| 1218 | update_tasks_flags(cs, &heap); |
| 1219 | heap_free(&heap); |
Li Zefan | 645fcc9 | 2009-01-07 18:08:43 -0800 | [diff] [blame] | 1220 | out: |
| 1221 | free_trial_cpuset(trialcs); |
| 1222 | return err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1223 | } |
| 1224 | |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 1225 | /* |
Adrian Bunk | 80f7228 | 2006-06-30 18:27:16 +0200 | [diff] [blame] | 1226 | * Frequency meter - How fast is some event occurring? |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 1227 | * |
| 1228 | * These routines manage a digitally filtered, constant time based, |
| 1229 | * event frequency meter. There are four routines: |
| 1230 | * fmeter_init() - initialize a frequency meter. |
| 1231 | * fmeter_markevent() - called each time the event happens. |
| 1232 | * fmeter_getrate() - returns the recent rate of such events. |
| 1233 | * fmeter_update() - internal routine used to update fmeter. |
| 1234 | * |
| 1235 | * A common data structure is passed to each of these routines, |
| 1236 | * which is used to keep track of the state required to manage the |
| 1237 | * frequency meter and its digital filter. |
| 1238 | * |
| 1239 | * The filter works on the number of events marked per unit time. |
| 1240 | * The filter is single-pole low-pass recursive (IIR). The time unit |
| 1241 | * is 1 second. Arithmetic is done using 32-bit integers scaled to |
| 1242 | * simulate 3 decimal digits of precision (multiplied by 1000). |
| 1243 | * |
| 1244 | * With an FM_COEF of 933, and a time base of 1 second, the filter |
| 1245 | * has a half-life of 10 seconds, meaning that if the events quit |
| 1246 | * happening, then the rate returned from the fmeter_getrate() |
| 1247 | * will be cut in half each 10 seconds, until it converges to zero. |
| 1248 | * |
| 1249 | * It is not worth doing a real infinitely recursive filter. If more |
| 1250 | * than FM_MAXTICKS ticks have elapsed since the last filter event, |
| 1251 | * just compute FM_MAXTICKS ticks worth, by which point the level |
| 1252 | * will be stable. |
| 1253 | * |
| 1254 | * Limit the count of unprocessed events to FM_MAXCNT, so as to avoid |
| 1255 | * arithmetic overflow in the fmeter_update() routine. |
| 1256 | * |
| 1257 | * Given the simple 32 bit integer arithmetic used, this meter works |
| 1258 | * best for reporting rates between one per millisecond (msec) and |
| 1259 | * one per 32 (approx) seconds. At constant rates faster than one |
| 1260 | * per msec it maxes out at values just under 1,000,000. At constant |
| 1261 | * rates between one per msec, and one per second it will stabilize |
| 1262 | * to a value N*1000, where N is the rate of events per second. |
| 1263 | * At constant rates between one per second and one per 32 seconds, |
| 1264 | * it will be choppy, moving up on the seconds that have an event, |
| 1265 | * and then decaying until the next event. At rates slower than |
| 1266 | * about one in 32 seconds, it decays all the way back to zero between |
| 1267 | * each event. |
| 1268 | */ |
| 1269 | |
| 1270 | #define FM_COEF 933 /* coefficient for half-life of 10 secs */ |
| 1271 | #define FM_MAXTICKS ((time_t)99) /* useless computing more ticks than this */ |
| 1272 | #define FM_MAXCNT 1000000 /* limit cnt to avoid overflow */ |
| 1273 | #define FM_SCALE 1000 /* faux fixed point scale */ |
| 1274 | |
| 1275 | /* Initialize a frequency meter */ |
| 1276 | static void fmeter_init(struct fmeter *fmp) |
| 1277 | { |
| 1278 | fmp->cnt = 0; |
| 1279 | fmp->val = 0; |
| 1280 | fmp->time = 0; |
| 1281 | spin_lock_init(&fmp->lock); |
| 1282 | } |
| 1283 | |
| 1284 | /* Internal meter update - process cnt events and update value */ |
| 1285 | static void fmeter_update(struct fmeter *fmp) |
| 1286 | { |
| 1287 | time_t now = get_seconds(); |
| 1288 | time_t ticks = now - fmp->time; |
| 1289 | |
| 1290 | if (ticks == 0) |
| 1291 | return; |
| 1292 | |
| 1293 | ticks = min(FM_MAXTICKS, ticks); |
| 1294 | while (ticks-- > 0) |
| 1295 | fmp->val = (FM_COEF * fmp->val) / FM_SCALE; |
| 1296 | fmp->time = now; |
| 1297 | |
| 1298 | fmp->val += ((FM_SCALE - FM_COEF) * fmp->cnt) / FM_SCALE; |
| 1299 | fmp->cnt = 0; |
| 1300 | } |
| 1301 | |
| 1302 | /* Process any previous ticks, then bump cnt by one (times scale). */ |
| 1303 | static void fmeter_markevent(struct fmeter *fmp) |
| 1304 | { |
| 1305 | spin_lock(&fmp->lock); |
| 1306 | fmeter_update(fmp); |
| 1307 | fmp->cnt = min(FM_MAXCNT, fmp->cnt + FM_SCALE); |
| 1308 | spin_unlock(&fmp->lock); |
| 1309 | } |
| 1310 | |
| 1311 | /* Process any previous ticks, then return current value. */ |
| 1312 | static int fmeter_getrate(struct fmeter *fmp) |
| 1313 | { |
| 1314 | int val; |
| 1315 | |
| 1316 | spin_lock(&fmp->lock); |
| 1317 | fmeter_update(fmp); |
| 1318 | val = fmp->val; |
| 1319 | spin_unlock(&fmp->lock); |
| 1320 | return val; |
| 1321 | } |
| 1322 | |
Li Zefan | 2341d1b | 2009-01-07 18:08:42 -0800 | [diff] [blame] | 1323 | /* Protected by cgroup_lock */ |
| 1324 | static cpumask_var_t cpus_attach; |
| 1325 | |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 1326 | /* Called by cgroups to determine if a cpuset is usable; cgroup_mutex held */ |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1327 | static int cpuset_can_attach(struct cgroup_subsys *ss, |
| 1328 | struct cgroup *cont, struct task_struct *tsk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1329 | { |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1330 | struct cpuset *cs = cgroup_cs(cont); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1331 | |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 1332 | if (cpumask_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1333 | return -ENOSPC; |
David Rientjes | 9985b0b | 2008-06-05 12:57:11 -0700 | [diff] [blame] | 1334 | |
David Rientjes | 6d7b2f5 | 2009-04-02 16:57:57 -0700 | [diff] [blame] | 1335 | /* |
| 1336 | * Kthreads bound to specific cpus cannot be moved to a new cpuset; we |
| 1337 | * cannot change their cpu affinity and isolating such threads by their |
| 1338 | * set of allowed nodes is unnecessary. Thus, cpusets are not |
| 1339 | * applicable for such threads. This prevents checking for success of |
| 1340 | * set_cpus_allowed_ptr() on all attached tasks before cpus_allowed may |
| 1341 | * be changed. |
| 1342 | */ |
| 1343 | if (tsk->flags & PF_THREAD_BOUND) |
| 1344 | return -EINVAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1345 | |
David Rientjes | 6d7b2f5 | 2009-04-02 16:57:57 -0700 | [diff] [blame] | 1346 | return security_task_setscheduler(tsk, 0, NULL); |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1347 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1348 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1349 | static void cpuset_attach(struct cgroup_subsys *ss, |
| 1350 | struct cgroup *cont, struct cgroup *oldcont, |
| 1351 | struct task_struct *tsk) |
| 1352 | { |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1353 | nodemask_t from, to; |
| 1354 | struct mm_struct *mm; |
| 1355 | struct cpuset *cs = cgroup_cs(cont); |
| 1356 | struct cpuset *oldcs = cgroup_cs(oldcont); |
David Rientjes | 9985b0b | 2008-06-05 12:57:11 -0700 | [diff] [blame] | 1357 | int err; |
David Quigley | 22fb52d | 2006-06-23 02:04:00 -0700 | [diff] [blame] | 1358 | |
Miao Xie | f5813d9 | 2009-01-07 18:08:40 -0800 | [diff] [blame] | 1359 | if (cs == &top_cpuset) { |
Li Zefan | 2341d1b | 2009-01-07 18:08:42 -0800 | [diff] [blame] | 1360 | cpumask_copy(cpus_attach, cpu_possible_mask); |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 1361 | to = node_possible_map; |
Miao Xie | f5813d9 | 2009-01-07 18:08:40 -0800 | [diff] [blame] | 1362 | } else { |
Li Zefan | 2341d1b | 2009-01-07 18:08:42 -0800 | [diff] [blame] | 1363 | guarantee_online_cpus(cs, cpus_attach); |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 1364 | guarantee_online_mems(cs, &to); |
Miao Xie | f5813d9 | 2009-01-07 18:08:40 -0800 | [diff] [blame] | 1365 | } |
Li Zefan | 2341d1b | 2009-01-07 18:08:42 -0800 | [diff] [blame] | 1366 | err = set_cpus_allowed_ptr(tsk, cpus_attach); |
David Rientjes | 9985b0b | 2008-06-05 12:57:11 -0700 | [diff] [blame] | 1367 | if (err) |
| 1368 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1369 | |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 1370 | task_lock(tsk); |
| 1371 | cpuset_change_task_nodemask(tsk, &to); |
| 1372 | task_unlock(tsk); |
Miao Xie | 950592f | 2009-06-16 15:31:47 -0700 | [diff] [blame] | 1373 | cpuset_update_task_spread_flag(cs, tsk); |
| 1374 | |
Paul Jackson | 45b07ef | 2006-01-08 01:00:56 -0800 | [diff] [blame] | 1375 | from = oldcs->mems_allowed; |
| 1376 | to = cs->mems_allowed; |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 1377 | mm = get_task_mm(tsk); |
| 1378 | if (mm) { |
| 1379 | mpol_rebind_mm(mm, &to); |
Paul Jackson | 2741a55 | 2006-03-31 02:30:51 -0800 | [diff] [blame] | 1380 | if (is_memory_migrate(cs)) |
Paul Jackson | e4e364e | 2006-03-31 02:30:52 -0800 | [diff] [blame] | 1381 | cpuset_migrate_mm(mm, &from, &to); |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 1382 | mmput(mm); |
| 1383 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1384 | } |
| 1385 | |
| 1386 | /* The various types of files and directories in a cpuset file system */ |
| 1387 | |
| 1388 | typedef enum { |
Paul Jackson | 45b07ef | 2006-01-08 01:00:56 -0800 | [diff] [blame] | 1389 | FILE_MEMORY_MIGRATE, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1390 | FILE_CPULIST, |
| 1391 | FILE_MEMLIST, |
| 1392 | FILE_CPU_EXCLUSIVE, |
| 1393 | FILE_MEM_EXCLUSIVE, |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1394 | FILE_MEM_HARDWALL, |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 1395 | FILE_SCHED_LOAD_BALANCE, |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 1396 | FILE_SCHED_RELAX_DOMAIN_LEVEL, |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 1397 | FILE_MEMORY_PRESSURE_ENABLED, |
| 1398 | FILE_MEMORY_PRESSURE, |
Paul Jackson | 825a46a | 2006-03-24 03:16:03 -0800 | [diff] [blame] | 1399 | FILE_SPREAD_PAGE, |
| 1400 | FILE_SPREAD_SLAB, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1401 | } cpuset_filetype_t; |
| 1402 | |
Paul Menage | 700fe1a | 2008-04-29 01:00:00 -0700 | [diff] [blame] | 1403 | static int cpuset_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val) |
| 1404 | { |
| 1405 | int retval = 0; |
| 1406 | struct cpuset *cs = cgroup_cs(cgrp); |
| 1407 | cpuset_filetype_t type = cft->private; |
| 1408 | |
Paul Menage | e371239 | 2008-07-25 01:47:02 -0700 | [diff] [blame] | 1409 | if (!cgroup_lock_live_group(cgrp)) |
Paul Menage | 700fe1a | 2008-04-29 01:00:00 -0700 | [diff] [blame] | 1410 | return -ENODEV; |
Paul Menage | 700fe1a | 2008-04-29 01:00:00 -0700 | [diff] [blame] | 1411 | |
| 1412 | switch (type) { |
| 1413 | case FILE_CPU_EXCLUSIVE: |
| 1414 | retval = update_flag(CS_CPU_EXCLUSIVE, cs, val); |
| 1415 | break; |
| 1416 | case FILE_MEM_EXCLUSIVE: |
| 1417 | retval = update_flag(CS_MEM_EXCLUSIVE, cs, val); |
| 1418 | break; |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1419 | case FILE_MEM_HARDWALL: |
| 1420 | retval = update_flag(CS_MEM_HARDWALL, cs, val); |
| 1421 | break; |
Paul Menage | 700fe1a | 2008-04-29 01:00:00 -0700 | [diff] [blame] | 1422 | case FILE_SCHED_LOAD_BALANCE: |
| 1423 | retval = update_flag(CS_SCHED_LOAD_BALANCE, cs, val); |
| 1424 | break; |
| 1425 | case FILE_MEMORY_MIGRATE: |
| 1426 | retval = update_flag(CS_MEMORY_MIGRATE, cs, val); |
| 1427 | break; |
| 1428 | case FILE_MEMORY_PRESSURE_ENABLED: |
| 1429 | cpuset_memory_pressure_enabled = !!val; |
| 1430 | break; |
| 1431 | case FILE_MEMORY_PRESSURE: |
| 1432 | retval = -EACCES; |
| 1433 | break; |
| 1434 | case FILE_SPREAD_PAGE: |
| 1435 | retval = update_flag(CS_SPREAD_PAGE, cs, val); |
Paul Menage | 700fe1a | 2008-04-29 01:00:00 -0700 | [diff] [blame] | 1436 | break; |
| 1437 | case FILE_SPREAD_SLAB: |
| 1438 | retval = update_flag(CS_SPREAD_SLAB, cs, val); |
Paul Menage | 700fe1a | 2008-04-29 01:00:00 -0700 | [diff] [blame] | 1439 | break; |
| 1440 | default: |
| 1441 | retval = -EINVAL; |
| 1442 | break; |
| 1443 | } |
| 1444 | cgroup_unlock(); |
| 1445 | return retval; |
| 1446 | } |
| 1447 | |
Paul Menage | 5be7a47 | 2008-05-06 20:42:41 -0700 | [diff] [blame] | 1448 | static int cpuset_write_s64(struct cgroup *cgrp, struct cftype *cft, s64 val) |
| 1449 | { |
| 1450 | int retval = 0; |
| 1451 | struct cpuset *cs = cgroup_cs(cgrp); |
| 1452 | cpuset_filetype_t type = cft->private; |
| 1453 | |
Paul Menage | e371239 | 2008-07-25 01:47:02 -0700 | [diff] [blame] | 1454 | if (!cgroup_lock_live_group(cgrp)) |
Paul Menage | 5be7a47 | 2008-05-06 20:42:41 -0700 | [diff] [blame] | 1455 | return -ENODEV; |
Paul Menage | e371239 | 2008-07-25 01:47:02 -0700 | [diff] [blame] | 1456 | |
Paul Menage | 5be7a47 | 2008-05-06 20:42:41 -0700 | [diff] [blame] | 1457 | switch (type) { |
| 1458 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: |
| 1459 | retval = update_relax_domain_level(cs, val); |
| 1460 | break; |
| 1461 | default: |
| 1462 | retval = -EINVAL; |
| 1463 | break; |
| 1464 | } |
| 1465 | cgroup_unlock(); |
| 1466 | return retval; |
| 1467 | } |
| 1468 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1469 | /* |
Paul Menage | e371239 | 2008-07-25 01:47:02 -0700 | [diff] [blame] | 1470 | * Common handling for a write to a "cpus" or "mems" file. |
| 1471 | */ |
| 1472 | static int cpuset_write_resmask(struct cgroup *cgrp, struct cftype *cft, |
| 1473 | const char *buf) |
| 1474 | { |
| 1475 | int retval = 0; |
Li Zefan | 645fcc9 | 2009-01-07 18:08:43 -0800 | [diff] [blame] | 1476 | struct cpuset *cs = cgroup_cs(cgrp); |
| 1477 | struct cpuset *trialcs; |
Paul Menage | e371239 | 2008-07-25 01:47:02 -0700 | [diff] [blame] | 1478 | |
| 1479 | if (!cgroup_lock_live_group(cgrp)) |
| 1480 | return -ENODEV; |
| 1481 | |
Li Zefan | 645fcc9 | 2009-01-07 18:08:43 -0800 | [diff] [blame] | 1482 | trialcs = alloc_trial_cpuset(cs); |
| 1483 | if (!trialcs) |
| 1484 | return -ENOMEM; |
| 1485 | |
Paul Menage | e371239 | 2008-07-25 01:47:02 -0700 | [diff] [blame] | 1486 | switch (cft->private) { |
| 1487 | case FILE_CPULIST: |
Li Zefan | 645fcc9 | 2009-01-07 18:08:43 -0800 | [diff] [blame] | 1488 | retval = update_cpumask(cs, trialcs, buf); |
Paul Menage | e371239 | 2008-07-25 01:47:02 -0700 | [diff] [blame] | 1489 | break; |
| 1490 | case FILE_MEMLIST: |
Li Zefan | 645fcc9 | 2009-01-07 18:08:43 -0800 | [diff] [blame] | 1491 | retval = update_nodemask(cs, trialcs, buf); |
Paul Menage | e371239 | 2008-07-25 01:47:02 -0700 | [diff] [blame] | 1492 | break; |
| 1493 | default: |
| 1494 | retval = -EINVAL; |
| 1495 | break; |
| 1496 | } |
Li Zefan | 645fcc9 | 2009-01-07 18:08:43 -0800 | [diff] [blame] | 1497 | |
| 1498 | free_trial_cpuset(trialcs); |
Paul Menage | e371239 | 2008-07-25 01:47:02 -0700 | [diff] [blame] | 1499 | cgroup_unlock(); |
| 1500 | return retval; |
| 1501 | } |
| 1502 | |
| 1503 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1504 | * These ascii lists should be read in a single call, by using a user |
| 1505 | * buffer large enough to hold the entire map. If read in smaller |
| 1506 | * chunks, there is no guarantee of atomicity. Since the display format |
| 1507 | * used, list of ranges of sequential numbers, is variable length, |
| 1508 | * and since these maps can change value dynamically, one could read |
| 1509 | * gibberish by doing partial reads while a list was changing. |
| 1510 | * A single large read to a buffer that crosses a page boundary is |
| 1511 | * ok, because the result being copied to user land is not recomputed |
| 1512 | * across a page fault. |
| 1513 | */ |
| 1514 | |
| 1515 | static int cpuset_sprintf_cpulist(char *page, struct cpuset *cs) |
| 1516 | { |
Li Zefan | 5a7625d | 2009-01-07 18:08:41 -0800 | [diff] [blame] | 1517 | int ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1518 | |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 1519 | mutex_lock(&callback_mutex); |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 1520 | ret = cpulist_scnprintf(page, PAGE_SIZE, cs->cpus_allowed); |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 1521 | mutex_unlock(&callback_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1522 | |
Li Zefan | 5a7625d | 2009-01-07 18:08:41 -0800 | [diff] [blame] | 1523 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1524 | } |
| 1525 | |
| 1526 | static int cpuset_sprintf_memlist(char *page, struct cpuset *cs) |
| 1527 | { |
| 1528 | nodemask_t mask; |
| 1529 | |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 1530 | mutex_lock(&callback_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1531 | mask = cs->mems_allowed; |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 1532 | mutex_unlock(&callback_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1533 | |
| 1534 | return nodelist_scnprintf(page, PAGE_SIZE, mask); |
| 1535 | } |
| 1536 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1537 | static ssize_t cpuset_common_file_read(struct cgroup *cont, |
| 1538 | struct cftype *cft, |
| 1539 | struct file *file, |
| 1540 | char __user *buf, |
| 1541 | size_t nbytes, loff_t *ppos) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1542 | { |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1543 | struct cpuset *cs = cgroup_cs(cont); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1544 | cpuset_filetype_t type = cft->private; |
| 1545 | char *page; |
| 1546 | ssize_t retval = 0; |
| 1547 | char *s; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1548 | |
Mel Gorman | e12ba74 | 2007-10-16 01:25:52 -0700 | [diff] [blame] | 1549 | if (!(page = (char *)__get_free_page(GFP_TEMPORARY))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1550 | return -ENOMEM; |
| 1551 | |
| 1552 | s = page; |
| 1553 | |
| 1554 | switch (type) { |
| 1555 | case FILE_CPULIST: |
| 1556 | s += cpuset_sprintf_cpulist(s, cs); |
| 1557 | break; |
| 1558 | case FILE_MEMLIST: |
| 1559 | s += cpuset_sprintf_memlist(s, cs); |
| 1560 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1561 | default: |
| 1562 | retval = -EINVAL; |
| 1563 | goto out; |
| 1564 | } |
| 1565 | *s++ = '\n'; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1566 | |
Al Viro | eacaa1f | 2005-09-30 03:26:43 +0100 | [diff] [blame] | 1567 | retval = simple_read_from_buffer(buf, nbytes, ppos, page, s - page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1568 | out: |
| 1569 | free_page((unsigned long)page); |
| 1570 | return retval; |
| 1571 | } |
| 1572 | |
Paul Menage | 700fe1a | 2008-04-29 01:00:00 -0700 | [diff] [blame] | 1573 | static u64 cpuset_read_u64(struct cgroup *cont, struct cftype *cft) |
| 1574 | { |
| 1575 | struct cpuset *cs = cgroup_cs(cont); |
| 1576 | cpuset_filetype_t type = cft->private; |
| 1577 | switch (type) { |
| 1578 | case FILE_CPU_EXCLUSIVE: |
| 1579 | return is_cpu_exclusive(cs); |
| 1580 | case FILE_MEM_EXCLUSIVE: |
| 1581 | return is_mem_exclusive(cs); |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1582 | case FILE_MEM_HARDWALL: |
| 1583 | return is_mem_hardwall(cs); |
Paul Menage | 700fe1a | 2008-04-29 01:00:00 -0700 | [diff] [blame] | 1584 | case FILE_SCHED_LOAD_BALANCE: |
| 1585 | return is_sched_load_balance(cs); |
| 1586 | case FILE_MEMORY_MIGRATE: |
| 1587 | return is_memory_migrate(cs); |
| 1588 | case FILE_MEMORY_PRESSURE_ENABLED: |
| 1589 | return cpuset_memory_pressure_enabled; |
| 1590 | case FILE_MEMORY_PRESSURE: |
| 1591 | return fmeter_getrate(&cs->fmeter); |
| 1592 | case FILE_SPREAD_PAGE: |
| 1593 | return is_spread_page(cs); |
| 1594 | case FILE_SPREAD_SLAB: |
| 1595 | return is_spread_slab(cs); |
| 1596 | default: |
| 1597 | BUG(); |
| 1598 | } |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 1599 | |
| 1600 | /* Unreachable but makes gcc happy */ |
| 1601 | return 0; |
Paul Menage | 700fe1a | 2008-04-29 01:00:00 -0700 | [diff] [blame] | 1602 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1603 | |
Paul Menage | 5be7a47 | 2008-05-06 20:42:41 -0700 | [diff] [blame] | 1604 | static s64 cpuset_read_s64(struct cgroup *cont, struct cftype *cft) |
| 1605 | { |
| 1606 | struct cpuset *cs = cgroup_cs(cont); |
| 1607 | cpuset_filetype_t type = cft->private; |
| 1608 | switch (type) { |
| 1609 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: |
| 1610 | return cs->relax_domain_level; |
| 1611 | default: |
| 1612 | BUG(); |
| 1613 | } |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 1614 | |
| 1615 | /* Unrechable but makes gcc happy */ |
| 1616 | return 0; |
Paul Menage | 5be7a47 | 2008-05-06 20:42:41 -0700 | [diff] [blame] | 1617 | } |
| 1618 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1619 | |
| 1620 | /* |
| 1621 | * for the common functions, 'private' gives the type of file |
| 1622 | */ |
| 1623 | |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1624 | static struct cftype files[] = { |
| 1625 | { |
| 1626 | .name = "cpus", |
| 1627 | .read = cpuset_common_file_read, |
Paul Menage | e371239 | 2008-07-25 01:47:02 -0700 | [diff] [blame] | 1628 | .write_string = cpuset_write_resmask, |
| 1629 | .max_write_len = (100U + 6 * NR_CPUS), |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1630 | .private = FILE_CPULIST, |
| 1631 | }, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1632 | |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1633 | { |
| 1634 | .name = "mems", |
| 1635 | .read = cpuset_common_file_read, |
Paul Menage | e371239 | 2008-07-25 01:47:02 -0700 | [diff] [blame] | 1636 | .write_string = cpuset_write_resmask, |
| 1637 | .max_write_len = (100U + 6 * MAX_NUMNODES), |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1638 | .private = FILE_MEMLIST, |
| 1639 | }, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1640 | |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1641 | { |
| 1642 | .name = "cpu_exclusive", |
| 1643 | .read_u64 = cpuset_read_u64, |
| 1644 | .write_u64 = cpuset_write_u64, |
| 1645 | .private = FILE_CPU_EXCLUSIVE, |
| 1646 | }, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1647 | |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1648 | { |
| 1649 | .name = "mem_exclusive", |
| 1650 | .read_u64 = cpuset_read_u64, |
| 1651 | .write_u64 = cpuset_write_u64, |
| 1652 | .private = FILE_MEM_EXCLUSIVE, |
| 1653 | }, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1654 | |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1655 | { |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1656 | .name = "mem_hardwall", |
| 1657 | .read_u64 = cpuset_read_u64, |
| 1658 | .write_u64 = cpuset_write_u64, |
| 1659 | .private = FILE_MEM_HARDWALL, |
| 1660 | }, |
| 1661 | |
| 1662 | { |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1663 | .name = "sched_load_balance", |
| 1664 | .read_u64 = cpuset_read_u64, |
| 1665 | .write_u64 = cpuset_write_u64, |
| 1666 | .private = FILE_SCHED_LOAD_BALANCE, |
| 1667 | }, |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 1668 | |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1669 | { |
| 1670 | .name = "sched_relax_domain_level", |
Paul Menage | 5be7a47 | 2008-05-06 20:42:41 -0700 | [diff] [blame] | 1671 | .read_s64 = cpuset_read_s64, |
| 1672 | .write_s64 = cpuset_write_s64, |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1673 | .private = FILE_SCHED_RELAX_DOMAIN_LEVEL, |
| 1674 | }, |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 1675 | |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1676 | { |
| 1677 | .name = "memory_migrate", |
| 1678 | .read_u64 = cpuset_read_u64, |
| 1679 | .write_u64 = cpuset_write_u64, |
| 1680 | .private = FILE_MEMORY_MIGRATE, |
| 1681 | }, |
| 1682 | |
| 1683 | { |
| 1684 | .name = "memory_pressure", |
| 1685 | .read_u64 = cpuset_read_u64, |
| 1686 | .write_u64 = cpuset_write_u64, |
| 1687 | .private = FILE_MEMORY_PRESSURE, |
Li Zefan | 099fca3 | 2009-04-02 16:57:29 -0700 | [diff] [blame] | 1688 | .mode = S_IRUGO, |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1689 | }, |
| 1690 | |
| 1691 | { |
| 1692 | .name = "memory_spread_page", |
| 1693 | .read_u64 = cpuset_read_u64, |
| 1694 | .write_u64 = cpuset_write_u64, |
| 1695 | .private = FILE_SPREAD_PAGE, |
| 1696 | }, |
| 1697 | |
| 1698 | { |
| 1699 | .name = "memory_spread_slab", |
| 1700 | .read_u64 = cpuset_read_u64, |
| 1701 | .write_u64 = cpuset_write_u64, |
| 1702 | .private = FILE_SPREAD_SLAB, |
| 1703 | }, |
Paul Jackson | 45b07ef | 2006-01-08 01:00:56 -0800 | [diff] [blame] | 1704 | }; |
| 1705 | |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 1706 | static struct cftype cft_memory_pressure_enabled = { |
| 1707 | .name = "memory_pressure_enabled", |
Paul Menage | 700fe1a | 2008-04-29 01:00:00 -0700 | [diff] [blame] | 1708 | .read_u64 = cpuset_read_u64, |
| 1709 | .write_u64 = cpuset_write_u64, |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 1710 | .private = FILE_MEMORY_PRESSURE_ENABLED, |
| 1711 | }; |
| 1712 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1713 | static int cpuset_populate(struct cgroup_subsys *ss, struct cgroup *cont) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1714 | { |
| 1715 | int err; |
| 1716 | |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1717 | err = cgroup_add_files(cont, ss, files, ARRAY_SIZE(files)); |
| 1718 | if (err) |
Paul Jackson | 825a46a | 2006-03-24 03:16:03 -0800 | [diff] [blame] | 1719 | return err; |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1720 | /* memory_pressure_enabled is in root cpuset only */ |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1721 | if (!cont->parent) |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1722 | err = cgroup_add_file(cont, ss, |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1723 | &cft_memory_pressure_enabled); |
| 1724 | return err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1725 | } |
| 1726 | |
| 1727 | /* |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1728 | * post_clone() is called at the end of cgroup_clone(). |
| 1729 | * 'cgroup' was just created automatically as a result of |
| 1730 | * a cgroup_clone(), and the current task is about to |
| 1731 | * be moved into 'cgroup'. |
| 1732 | * |
| 1733 | * Currently we refuse to set up the cgroup - thereby |
| 1734 | * refusing the task to be entered, and as a result refusing |
| 1735 | * the sys_unshare() or clone() which initiated it - if any |
| 1736 | * sibling cpusets have exclusive cpus or mem. |
| 1737 | * |
| 1738 | * If this becomes a problem for some users who wish to |
| 1739 | * allow that scenario, then cpuset_post_clone() could be |
| 1740 | * changed to grant parent->cpus_allowed-sibling_cpus_exclusive |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 1741 | * (and likewise for mems) to the new cgroup. Called with cgroup_mutex |
| 1742 | * held. |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1743 | */ |
| 1744 | static void cpuset_post_clone(struct cgroup_subsys *ss, |
| 1745 | struct cgroup *cgroup) |
| 1746 | { |
| 1747 | struct cgroup *parent, *child; |
| 1748 | struct cpuset *cs, *parent_cs; |
| 1749 | |
| 1750 | parent = cgroup->parent; |
| 1751 | list_for_each_entry(child, &parent->children, sibling) { |
| 1752 | cs = cgroup_cs(child); |
| 1753 | if (is_mem_exclusive(cs) || is_cpu_exclusive(cs)) |
| 1754 | return; |
| 1755 | } |
| 1756 | cs = cgroup_cs(cgroup); |
| 1757 | parent_cs = cgroup_cs(parent); |
| 1758 | |
| 1759 | cs->mems_allowed = parent_cs->mems_allowed; |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 1760 | cpumask_copy(cs->cpus_allowed, parent_cs->cpus_allowed); |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1761 | return; |
| 1762 | } |
| 1763 | |
| 1764 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1765 | * cpuset_create - create a cpuset |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 1766 | * ss: cpuset cgroup subsystem |
| 1767 | * cont: control group that the new cpuset will be part of |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1768 | */ |
| 1769 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1770 | static struct cgroup_subsys_state *cpuset_create( |
| 1771 | struct cgroup_subsys *ss, |
| 1772 | struct cgroup *cont) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1773 | { |
| 1774 | struct cpuset *cs; |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1775 | struct cpuset *parent; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1776 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1777 | if (!cont->parent) { |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1778 | return &top_cpuset.css; |
| 1779 | } |
| 1780 | parent = cgroup_cs(cont->parent); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1781 | cs = kmalloc(sizeof(*cs), GFP_KERNEL); |
| 1782 | if (!cs) |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1783 | return ERR_PTR(-ENOMEM); |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 1784 | if (!alloc_cpumask_var(&cs->cpus_allowed, GFP_KERNEL)) { |
| 1785 | kfree(cs); |
| 1786 | return ERR_PTR(-ENOMEM); |
| 1787 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1788 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1789 | cs->flags = 0; |
Paul Jackson | 825a46a | 2006-03-24 03:16:03 -0800 | [diff] [blame] | 1790 | if (is_spread_page(parent)) |
| 1791 | set_bit(CS_SPREAD_PAGE, &cs->flags); |
| 1792 | if (is_spread_slab(parent)) |
| 1793 | set_bit(CS_SPREAD_SLAB, &cs->flags); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 1794 | set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 1795 | cpumask_clear(cs->cpus_allowed); |
Mike Travis | f9a86fc | 2008-04-04 18:11:07 -0700 | [diff] [blame] | 1796 | nodes_clear(cs->mems_allowed); |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 1797 | fmeter_init(&cs->fmeter); |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 1798 | cs->relax_domain_level = -1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1799 | |
| 1800 | cs->parent = parent; |
Paul Jackson | 202f72d | 2006-01-08 01:01:57 -0800 | [diff] [blame] | 1801 | number_of_cpusets++; |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1802 | return &cs->css ; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1803 | } |
| 1804 | |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 1805 | /* |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 1806 | * If the cpuset being removed has its flag 'sched_load_balance' |
| 1807 | * enabled, then simulate turning sched_load_balance off, which |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 1808 | * will call async_rebuild_sched_domains(). |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 1809 | */ |
| 1810 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1811 | static void cpuset_destroy(struct cgroup_subsys *ss, struct cgroup *cont) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1812 | { |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1813 | struct cpuset *cs = cgroup_cs(cont); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1814 | |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 1815 | if (is_sched_load_balance(cs)) |
Paul Menage | 700fe1a | 2008-04-29 01:00:00 -0700 | [diff] [blame] | 1816 | update_flag(CS_SCHED_LOAD_BALANCE, cs, 0); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 1817 | |
Paul Jackson | 202f72d | 2006-01-08 01:01:57 -0800 | [diff] [blame] | 1818 | number_of_cpusets--; |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 1819 | free_cpumask_var(cs->cpus_allowed); |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1820 | kfree(cs); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1821 | } |
| 1822 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1823 | struct cgroup_subsys cpuset_subsys = { |
| 1824 | .name = "cpuset", |
| 1825 | .create = cpuset_create, |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 1826 | .destroy = cpuset_destroy, |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1827 | .can_attach = cpuset_can_attach, |
| 1828 | .attach = cpuset_attach, |
| 1829 | .populate = cpuset_populate, |
| 1830 | .post_clone = cpuset_post_clone, |
| 1831 | .subsys_id = cpuset_subsys_id, |
| 1832 | .early_init = 1, |
| 1833 | }; |
| 1834 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1835 | /** |
| 1836 | * cpuset_init - initialize cpusets at system boot |
| 1837 | * |
| 1838 | * Description: Initialize top_cpuset and the cpuset internal file system, |
| 1839 | **/ |
| 1840 | |
| 1841 | int __init cpuset_init(void) |
| 1842 | { |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1843 | int err = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1844 | |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 1845 | if (!alloc_cpumask_var(&top_cpuset.cpus_allowed, GFP_KERNEL)) |
| 1846 | BUG(); |
| 1847 | |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 1848 | cpumask_setall(top_cpuset.cpus_allowed); |
Mike Travis | f9a86fc | 2008-04-04 18:11:07 -0700 | [diff] [blame] | 1849 | nodes_setall(top_cpuset.mems_allowed); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1850 | |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 1851 | fmeter_init(&top_cpuset.fmeter); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 1852 | set_bit(CS_SCHED_LOAD_BALANCE, &top_cpuset.flags); |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 1853 | top_cpuset.relax_domain_level = -1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1854 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1855 | err = register_filesystem(&cpuset_fs_type); |
| 1856 | if (err < 0) |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1857 | return err; |
| 1858 | |
Li Zefan | 2341d1b | 2009-01-07 18:08:42 -0800 | [diff] [blame] | 1859 | if (!alloc_cpumask_var(&cpus_attach, GFP_KERNEL)) |
| 1860 | BUG(); |
| 1861 | |
Paul Jackson | 202f72d | 2006-01-08 01:01:57 -0800 | [diff] [blame] | 1862 | number_of_cpusets = 1; |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1863 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1864 | } |
| 1865 | |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1866 | /** |
| 1867 | * cpuset_do_move_task - move a given task to another cpuset |
| 1868 | * @tsk: pointer to task_struct the task to move |
| 1869 | * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner |
| 1870 | * |
| 1871 | * Called by cgroup_scan_tasks() for each task in a cgroup. |
| 1872 | * Return nonzero to stop the walk through the tasks. |
| 1873 | */ |
Adrian Bunk | 9e0c914 | 2008-04-29 01:00:25 -0700 | [diff] [blame] | 1874 | static void cpuset_do_move_task(struct task_struct *tsk, |
| 1875 | struct cgroup_scanner *scan) |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1876 | { |
Li Zefan | 7f81b1a | 2009-04-02 16:57:53 -0700 | [diff] [blame] | 1877 | struct cgroup *new_cgroup = scan->data; |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1878 | |
Li Zefan | 7f81b1a | 2009-04-02 16:57:53 -0700 | [diff] [blame] | 1879 | cgroup_attach_task(new_cgroup, tsk); |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1880 | } |
| 1881 | |
| 1882 | /** |
| 1883 | * move_member_tasks_to_cpuset - move tasks from one cpuset to another |
| 1884 | * @from: cpuset in which the tasks currently reside |
| 1885 | * @to: cpuset to which the tasks will be moved |
| 1886 | * |
Paul Jackson | c8d9c90 | 2008-02-07 00:14:46 -0800 | [diff] [blame] | 1887 | * Called with cgroup_mutex held |
| 1888 | * callback_mutex must not be held, as cpuset_attach() will take it. |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1889 | * |
| 1890 | * The cgroup_scan_tasks() function will scan all the tasks in a cgroup, |
| 1891 | * calling callback functions for each. |
| 1892 | */ |
| 1893 | static void move_member_tasks_to_cpuset(struct cpuset *from, struct cpuset *to) |
| 1894 | { |
Li Zefan | 7f81b1a | 2009-04-02 16:57:53 -0700 | [diff] [blame] | 1895 | struct cgroup_scanner scan; |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1896 | |
Li Zefan | 7f81b1a | 2009-04-02 16:57:53 -0700 | [diff] [blame] | 1897 | scan.cg = from->css.cgroup; |
| 1898 | scan.test_task = NULL; /* select all tasks in cgroup */ |
| 1899 | scan.process_task = cpuset_do_move_task; |
| 1900 | scan.heap = NULL; |
| 1901 | scan.data = to->css.cgroup; |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1902 | |
Li Zefan | 7f81b1a | 2009-04-02 16:57:53 -0700 | [diff] [blame] | 1903 | if (cgroup_scan_tasks(&scan)) |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1904 | printk(KERN_ERR "move_member_tasks_to_cpuset: " |
| 1905 | "cgroup_scan_tasks failed\n"); |
| 1906 | } |
| 1907 | |
Paul Jackson | b1aac8b | 2006-09-29 02:01:17 -0700 | [diff] [blame] | 1908 | /* |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 1909 | * If CPU and/or memory hotplug handlers, below, unplug any CPUs |
Paul Jackson | b1aac8b | 2006-09-29 02:01:17 -0700 | [diff] [blame] | 1910 | * or memory nodes, we need to walk over the cpuset hierarchy, |
| 1911 | * removing that CPU or node from all cpusets. If this removes the |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1912 | * last CPU or node from a cpuset, then move the tasks in the empty |
| 1913 | * cpuset to its next-highest non-empty parent. |
Paul Jackson | b1aac8b | 2006-09-29 02:01:17 -0700 | [diff] [blame] | 1914 | * |
Paul Jackson | c8d9c90 | 2008-02-07 00:14:46 -0800 | [diff] [blame] | 1915 | * Called with cgroup_mutex held |
| 1916 | * callback_mutex must not be held, as cpuset_attach() will take it. |
Paul Jackson | b1aac8b | 2006-09-29 02:01:17 -0700 | [diff] [blame] | 1917 | */ |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1918 | static void remove_tasks_in_empty_cpuset(struct cpuset *cs) |
Paul Jackson | b1aac8b | 2006-09-29 02:01:17 -0700 | [diff] [blame] | 1919 | { |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1920 | struct cpuset *parent; |
Paul Jackson | b1aac8b | 2006-09-29 02:01:17 -0700 | [diff] [blame] | 1921 | |
Paul Jackson | c8d9c90 | 2008-02-07 00:14:46 -0800 | [diff] [blame] | 1922 | /* |
| 1923 | * The cgroup's css_sets list is in use if there are tasks |
| 1924 | * in the cpuset; the list is empty if there are none; |
| 1925 | * the cs->css.refcnt seems always 0. |
| 1926 | */ |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1927 | if (list_empty(&cs->css.cgroup->css_sets)) |
| 1928 | return; |
| 1929 | |
| 1930 | /* |
| 1931 | * Find its next-highest non-empty parent, (top cpuset |
| 1932 | * has online cpus, so can't be empty). |
| 1933 | */ |
| 1934 | parent = cs->parent; |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 1935 | while (cpumask_empty(parent->cpus_allowed) || |
Paul Jackson | b450129 | 2008-02-07 00:14:47 -0800 | [diff] [blame] | 1936 | nodes_empty(parent->mems_allowed)) |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1937 | parent = parent->parent; |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1938 | |
| 1939 | move_member_tasks_to_cpuset(cs, parent); |
| 1940 | } |
| 1941 | |
| 1942 | /* |
| 1943 | * Walk the specified cpuset subtree and look for empty cpusets. |
| 1944 | * The tasks of such cpuset must be moved to a parent cpuset. |
| 1945 | * |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 1946 | * Called with cgroup_mutex held. We take callback_mutex to modify |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1947 | * cpus_allowed and mems_allowed. |
| 1948 | * |
| 1949 | * This walk processes the tree from top to bottom, completing one layer |
| 1950 | * before dropping down to the next. It always processes a node before |
| 1951 | * any of its children. |
| 1952 | * |
| 1953 | * For now, since we lack memory hot unplug, we'll never see a cpuset |
| 1954 | * that has tasks along with an empty 'mems'. But if we did see such |
| 1955 | * a cpuset, we'd handle it just like we do if its 'cpus' was empty. |
| 1956 | */ |
Frederic Weisbecker | d294eb8 | 2008-10-03 12:10:10 +0200 | [diff] [blame] | 1957 | static void scan_for_empty_cpusets(struct cpuset *root) |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1958 | { |
Li Zefan | 8d1e626 | 2008-07-29 22:33:21 -0700 | [diff] [blame] | 1959 | LIST_HEAD(queue); |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1960 | struct cpuset *cp; /* scans cpusets being updated */ |
| 1961 | struct cpuset *child; /* scans child cpusets of cp */ |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1962 | struct cgroup *cont; |
Miao Xie | f9b4fb8 | 2008-07-25 01:47:22 -0700 | [diff] [blame] | 1963 | nodemask_t oldmems; |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1964 | |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1965 | list_add_tail((struct list_head *)&root->stack_list, &queue); |
| 1966 | |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1967 | while (!list_empty(&queue)) { |
Li Zefan | 8d1e626 | 2008-07-29 22:33:21 -0700 | [diff] [blame] | 1968 | cp = list_first_entry(&queue, struct cpuset, stack_list); |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1969 | list_del(queue.next); |
| 1970 | list_for_each_entry(cont, &cp->css.cgroup->children, sibling) { |
| 1971 | child = cgroup_cs(cont); |
| 1972 | list_add_tail(&child->stack_list, &queue); |
| 1973 | } |
Paul Jackson | b450129 | 2008-02-07 00:14:47 -0800 | [diff] [blame] | 1974 | |
| 1975 | /* Continue past cpusets with all cpus, mems online */ |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 1976 | if (cpumask_subset(cp->cpus_allowed, cpu_online_mask) && |
Paul Jackson | b450129 | 2008-02-07 00:14:47 -0800 | [diff] [blame] | 1977 | nodes_subset(cp->mems_allowed, node_states[N_HIGH_MEMORY])) |
| 1978 | continue; |
| 1979 | |
Miao Xie | f9b4fb8 | 2008-07-25 01:47:22 -0700 | [diff] [blame] | 1980 | oldmems = cp->mems_allowed; |
| 1981 | |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1982 | /* Remove offline cpus and mems from this cpuset. */ |
Paul Jackson | b450129 | 2008-02-07 00:14:47 -0800 | [diff] [blame] | 1983 | mutex_lock(&callback_mutex); |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 1984 | cpumask_and(cp->cpus_allowed, cp->cpus_allowed, |
| 1985 | cpu_online_mask); |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1986 | nodes_and(cp->mems_allowed, cp->mems_allowed, |
| 1987 | node_states[N_HIGH_MEMORY]); |
Paul Jackson | b450129 | 2008-02-07 00:14:47 -0800 | [diff] [blame] | 1988 | mutex_unlock(&callback_mutex); |
| 1989 | |
| 1990 | /* Move tasks from the empty cpuset to a parent */ |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 1991 | if (cpumask_empty(cp->cpus_allowed) || |
Paul Jackson | b450129 | 2008-02-07 00:14:47 -0800 | [diff] [blame] | 1992 | nodes_empty(cp->mems_allowed)) |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1993 | remove_tasks_in_empty_cpuset(cp); |
Miao Xie | f9b4fb8 | 2008-07-25 01:47:22 -0700 | [diff] [blame] | 1994 | else { |
Li Zefan | 4e74339 | 2008-09-13 02:33:08 -0700 | [diff] [blame] | 1995 | update_tasks_cpumask(cp, NULL); |
Li Zefan | 010cfac | 2009-04-02 16:57:52 -0700 | [diff] [blame] | 1996 | update_tasks_nodemask(cp, &oldmems, NULL); |
Miao Xie | f9b4fb8 | 2008-07-25 01:47:22 -0700 | [diff] [blame] | 1997 | } |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1998 | } |
Paul Jackson | b1aac8b | 2006-09-29 02:01:17 -0700 | [diff] [blame] | 1999 | } |
| 2000 | |
| 2001 | /* |
Paul Jackson | 4c4d50f | 2006-08-27 01:23:51 -0700 | [diff] [blame] | 2002 | * The top_cpuset tracks what CPUs and Memory Nodes are online, |
| 2003 | * period. This is necessary in order to make cpusets transparent |
| 2004 | * (of no affect) on systems that are actively using CPU hotplug |
| 2005 | * but making no active use of cpusets. |
| 2006 | * |
Paul Jackson | 38837fc | 2006-09-29 02:01:16 -0700 | [diff] [blame] | 2007 | * This routine ensures that top_cpuset.cpus_allowed tracks |
| 2008 | * cpu_online_map on each CPU hotplug (cpuhp) event. |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 2009 | * |
| 2010 | * Called within get_online_cpus(). Needs to call cgroup_lock() |
| 2011 | * before calling generate_sched_domains(). |
Paul Jackson | 4c4d50f | 2006-08-27 01:23:51 -0700 | [diff] [blame] | 2012 | */ |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 2013 | static int cpuset_track_online_cpus(struct notifier_block *unused_nb, |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 2014 | unsigned long phase, void *unused_cpu) |
Paul Jackson | 4c4d50f | 2006-08-27 01:23:51 -0700 | [diff] [blame] | 2015 | { |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 2016 | struct sched_domain_attr *attr; |
Li Zefan | 6af866a | 2009-01-07 18:08:45 -0800 | [diff] [blame] | 2017 | struct cpumask *doms; |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 2018 | int ndoms; |
| 2019 | |
Dmitry Adamushko | 3e84050 | 2008-07-13 02:10:29 +0200 | [diff] [blame] | 2020 | switch (phase) { |
Dmitry Adamushko | 3e84050 | 2008-07-13 02:10:29 +0200 | [diff] [blame] | 2021 | case CPU_ONLINE: |
| 2022 | case CPU_ONLINE_FROZEN: |
| 2023 | case CPU_DEAD: |
| 2024 | case CPU_DEAD_FROZEN: |
Dmitry Adamushko | 3e84050 | 2008-07-13 02:10:29 +0200 | [diff] [blame] | 2025 | break; |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 2026 | |
Dmitry Adamushko | 3e84050 | 2008-07-13 02:10:29 +0200 | [diff] [blame] | 2027 | default: |
Avi Kivity | ac07675 | 2007-05-24 12:33:15 +0300 | [diff] [blame] | 2028 | return NOTIFY_DONE; |
Dmitry Adamushko | 3e84050 | 2008-07-13 02:10:29 +0200 | [diff] [blame] | 2029 | } |
Avi Kivity | ac07675 | 2007-05-24 12:33:15 +0300 | [diff] [blame] | 2030 | |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 2031 | cgroup_lock(); |
Li Zefan | 0b4217b | 2009-04-02 16:57:49 -0700 | [diff] [blame] | 2032 | mutex_lock(&callback_mutex); |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 2033 | cpumask_copy(top_cpuset.cpus_allowed, cpu_online_mask); |
Li Zefan | 0b4217b | 2009-04-02 16:57:49 -0700 | [diff] [blame] | 2034 | mutex_unlock(&callback_mutex); |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 2035 | scan_for_empty_cpusets(&top_cpuset); |
| 2036 | ndoms = generate_sched_domains(&doms, &attr); |
| 2037 | cgroup_unlock(); |
| 2038 | |
| 2039 | /* Have scheduler rebuild the domains */ |
| 2040 | partition_sched_domains(ndoms, doms, attr); |
| 2041 | |
Dmitry Adamushko | 3e84050 | 2008-07-13 02:10:29 +0200 | [diff] [blame] | 2042 | return NOTIFY_OK; |
Paul Jackson | 4c4d50f | 2006-08-27 01:23:51 -0700 | [diff] [blame] | 2043 | } |
Paul Jackson | 4c4d50f | 2006-08-27 01:23:51 -0700 | [diff] [blame] | 2044 | |
Paul Jackson | b1aac8b | 2006-09-29 02:01:17 -0700 | [diff] [blame] | 2045 | #ifdef CONFIG_MEMORY_HOTPLUG |
Paul Jackson | 38837fc | 2006-09-29 02:01:16 -0700 | [diff] [blame] | 2046 | /* |
Christoph Lameter | 0e1e7c7 | 2007-10-16 01:25:38 -0700 | [diff] [blame] | 2047 | * Keep top_cpuset.mems_allowed tracking node_states[N_HIGH_MEMORY]. |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 2048 | * Call this routine anytime after node_states[N_HIGH_MEMORY] changes. |
| 2049 | * See also the previous routine cpuset_track_online_cpus(). |
Paul Jackson | 38837fc | 2006-09-29 02:01:16 -0700 | [diff] [blame] | 2050 | */ |
Miao Xie | f481891 | 2008-11-19 15:36:30 -0800 | [diff] [blame] | 2051 | static int cpuset_track_online_nodes(struct notifier_block *self, |
| 2052 | unsigned long action, void *arg) |
Paul Jackson | 38837fc | 2006-09-29 02:01:16 -0700 | [diff] [blame] | 2053 | { |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 2054 | cgroup_lock(); |
Miao Xie | f481891 | 2008-11-19 15:36:30 -0800 | [diff] [blame] | 2055 | switch (action) { |
| 2056 | case MEM_ONLINE: |
Miao Xie | f481891 | 2008-11-19 15:36:30 -0800 | [diff] [blame] | 2057 | case MEM_OFFLINE: |
Li Zefan | 0b4217b | 2009-04-02 16:57:49 -0700 | [diff] [blame] | 2058 | mutex_lock(&callback_mutex); |
Miao Xie | f481891 | 2008-11-19 15:36:30 -0800 | [diff] [blame] | 2059 | top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY]; |
Li Zefan | 0b4217b | 2009-04-02 16:57:49 -0700 | [diff] [blame] | 2060 | mutex_unlock(&callback_mutex); |
| 2061 | if (action == MEM_OFFLINE) |
| 2062 | scan_for_empty_cpusets(&top_cpuset); |
Miao Xie | f481891 | 2008-11-19 15:36:30 -0800 | [diff] [blame] | 2063 | break; |
| 2064 | default: |
| 2065 | break; |
| 2066 | } |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 2067 | cgroup_unlock(); |
Miao Xie | f481891 | 2008-11-19 15:36:30 -0800 | [diff] [blame] | 2068 | return NOTIFY_OK; |
Paul Jackson | 38837fc | 2006-09-29 02:01:16 -0700 | [diff] [blame] | 2069 | } |
| 2070 | #endif |
| 2071 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2072 | /** |
| 2073 | * cpuset_init_smp - initialize cpus_allowed |
| 2074 | * |
| 2075 | * Description: Finish top cpuset after cpu, node maps are initialized |
| 2076 | **/ |
| 2077 | |
| 2078 | void __init cpuset_init_smp(void) |
| 2079 | { |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 2080 | cpumask_copy(top_cpuset.cpus_allowed, cpu_online_mask); |
Christoph Lameter | 0e1e7c7 | 2007-10-16 01:25:38 -0700 | [diff] [blame] | 2081 | top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY]; |
Paul Jackson | 4c4d50f | 2006-08-27 01:23:51 -0700 | [diff] [blame] | 2082 | |
Max Krasnyansky | cf41714 | 2008-08-11 14:33:53 -0700 | [diff] [blame] | 2083 | hotcpu_notifier(cpuset_track_online_cpus, 0); |
Miao Xie | f481891 | 2008-11-19 15:36:30 -0800 | [diff] [blame] | 2084 | hotplug_memory_notifier(cpuset_track_online_nodes, 10); |
Miao Xie | f90d411 | 2009-01-16 10:24:10 +0800 | [diff] [blame] | 2085 | |
| 2086 | cpuset_wq = create_singlethread_workqueue("cpuset"); |
| 2087 | BUG_ON(!cpuset_wq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2088 | } |
| 2089 | |
| 2090 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2091 | * cpuset_cpus_allowed - return cpus_allowed mask from a tasks cpuset. |
| 2092 | * @tsk: pointer to task_struct from which to obtain cpuset->cpus_allowed. |
Li Zefan | 6af866a | 2009-01-07 18:08:45 -0800 | [diff] [blame] | 2093 | * @pmask: pointer to struct cpumask variable to receive cpus_allowed set. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2094 | * |
Li Zefan | 300ed6c | 2009-01-07 18:08:44 -0800 | [diff] [blame] | 2095 | * Description: Returns the cpumask_var_t cpus_allowed of the cpuset |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2096 | * attached to the specified @tsk. Guaranteed to return some non-empty |
| 2097 | * subset of cpu_online_map, even if this means going outside the |
| 2098 | * tasks cpuset. |
| 2099 | **/ |
| 2100 | |
Li Zefan | 6af866a | 2009-01-07 18:08:45 -0800 | [diff] [blame] | 2101 | void cpuset_cpus_allowed(struct task_struct *tsk, struct cpumask *pmask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2102 | { |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 2103 | mutex_lock(&callback_mutex); |
Mike Travis | f9a86fc | 2008-04-04 18:11:07 -0700 | [diff] [blame] | 2104 | cpuset_cpus_allowed_locked(tsk, pmask); |
Cliff Wickman | 470fd64 | 2007-10-18 23:40:46 -0700 | [diff] [blame] | 2105 | mutex_unlock(&callback_mutex); |
Cliff Wickman | 470fd64 | 2007-10-18 23:40:46 -0700 | [diff] [blame] | 2106 | } |
| 2107 | |
| 2108 | /** |
| 2109 | * cpuset_cpus_allowed_locked - return cpus_allowed mask from a tasks cpuset. |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 2110 | * Must be called with callback_mutex held. |
Cliff Wickman | 470fd64 | 2007-10-18 23:40:46 -0700 | [diff] [blame] | 2111 | **/ |
Li Zefan | 6af866a | 2009-01-07 18:08:45 -0800 | [diff] [blame] | 2112 | void cpuset_cpus_allowed_locked(struct task_struct *tsk, struct cpumask *pmask) |
Cliff Wickman | 470fd64 | 2007-10-18 23:40:46 -0700 | [diff] [blame] | 2113 | { |
Paul Jackson | 909d75a | 2006-01-08 01:01:55 -0800 | [diff] [blame] | 2114 | task_lock(tsk); |
Mike Travis | f9a86fc | 2008-04-04 18:11:07 -0700 | [diff] [blame] | 2115 | guarantee_online_cpus(task_cs(tsk), pmask); |
Paul Jackson | 909d75a | 2006-01-08 01:01:55 -0800 | [diff] [blame] | 2116 | task_unlock(tsk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2117 | } |
| 2118 | |
| 2119 | void cpuset_init_current_mems_allowed(void) |
| 2120 | { |
Mike Travis | f9a86fc | 2008-04-04 18:11:07 -0700 | [diff] [blame] | 2121 | nodes_setall(current->mems_allowed); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2122 | } |
| 2123 | |
Randy Dunlap | d9fd8a6 | 2005-07-27 11:45:11 -0700 | [diff] [blame] | 2124 | /** |
Paul Jackson | 909d75a | 2006-01-08 01:01:55 -0800 | [diff] [blame] | 2125 | * cpuset_mems_allowed - return mems_allowed mask from a tasks cpuset. |
| 2126 | * @tsk: pointer to task_struct from which to obtain cpuset->mems_allowed. |
| 2127 | * |
| 2128 | * Description: Returns the nodemask_t mems_allowed of the cpuset |
| 2129 | * attached to the specified @tsk. Guaranteed to return some non-empty |
Christoph Lameter | 0e1e7c7 | 2007-10-16 01:25:38 -0700 | [diff] [blame] | 2130 | * subset of node_states[N_HIGH_MEMORY], even if this means going outside the |
Paul Jackson | 909d75a | 2006-01-08 01:01:55 -0800 | [diff] [blame] | 2131 | * tasks cpuset. |
| 2132 | **/ |
| 2133 | |
| 2134 | nodemask_t cpuset_mems_allowed(struct task_struct *tsk) |
| 2135 | { |
| 2136 | nodemask_t mask; |
| 2137 | |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 2138 | mutex_lock(&callback_mutex); |
Paul Jackson | 909d75a | 2006-01-08 01:01:55 -0800 | [diff] [blame] | 2139 | task_lock(tsk); |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 2140 | guarantee_online_mems(task_cs(tsk), &mask); |
Paul Jackson | 909d75a | 2006-01-08 01:01:55 -0800 | [diff] [blame] | 2141 | task_unlock(tsk); |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 2142 | mutex_unlock(&callback_mutex); |
Paul Jackson | 909d75a | 2006-01-08 01:01:55 -0800 | [diff] [blame] | 2143 | |
| 2144 | return mask; |
| 2145 | } |
| 2146 | |
| 2147 | /** |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 2148 | * cpuset_nodemask_valid_mems_allowed - check nodemask vs. curremt mems_allowed |
| 2149 | * @nodemask: the nodemask to be checked |
Randy Dunlap | d9fd8a6 | 2005-07-27 11:45:11 -0700 | [diff] [blame] | 2150 | * |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 2151 | * Are any of the nodes in the nodemask allowed in current->mems_allowed? |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2152 | */ |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 2153 | int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2154 | { |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 2155 | return nodes_intersects(*nodemask, current->mems_allowed); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2156 | } |
| 2157 | |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2158 | /* |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 2159 | * nearest_hardwall_ancestor() - Returns the nearest mem_exclusive or |
| 2160 | * mem_hardwall ancestor to the specified cpuset. Call holding |
| 2161 | * callback_mutex. If no ancestor is mem_exclusive or mem_hardwall |
| 2162 | * (an unusual configuration), then returns the root cpuset. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2163 | */ |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 2164 | static const struct cpuset *nearest_hardwall_ancestor(const struct cpuset *cs) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2165 | { |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 2166 | while (!(is_mem_exclusive(cs) || is_mem_hardwall(cs)) && cs->parent) |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2167 | cs = cs->parent; |
| 2168 | return cs; |
| 2169 | } |
| 2170 | |
| 2171 | /** |
David Rientjes | a1bc5a4 | 2009-04-02 16:57:54 -0700 | [diff] [blame] | 2172 | * cpuset_node_allowed_softwall - Can we allocate on a memory node? |
| 2173 | * @node: is this an allowed node? |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2174 | * @gfp_mask: memory allocation flags |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2175 | * |
David Rientjes | a1bc5a4 | 2009-04-02 16:57:54 -0700 | [diff] [blame] | 2176 | * If we're in interrupt, yes, we can always allocate. If __GFP_THISNODE is |
| 2177 | * set, yes, we can always allocate. If node is in our task's mems_allowed, |
| 2178 | * yes. If it's not a __GFP_HARDWALL request and this node is in the nearest |
| 2179 | * hardwalled cpuset ancestor to this task's cpuset, yes. If the task has been |
| 2180 | * OOM killed and has access to memory reserves as specified by the TIF_MEMDIE |
| 2181 | * flag, yes. |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2182 | * Otherwise, no. |
| 2183 | * |
David Rientjes | a1bc5a4 | 2009-04-02 16:57:54 -0700 | [diff] [blame] | 2184 | * If __GFP_HARDWALL is set, cpuset_node_allowed_softwall() reduces to |
| 2185 | * cpuset_node_allowed_hardwall(). Otherwise, cpuset_node_allowed_softwall() |
| 2186 | * might sleep, and might allow a node from an enclosing cpuset. |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2187 | * |
David Rientjes | a1bc5a4 | 2009-04-02 16:57:54 -0700 | [diff] [blame] | 2188 | * cpuset_node_allowed_hardwall() only handles the simpler case of hardwall |
| 2189 | * cpusets, and never sleeps. |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2190 | * |
| 2191 | * The __GFP_THISNODE placement logic is really handled elsewhere, |
| 2192 | * by forcibly using a zonelist starting at a specified node, and by |
| 2193 | * (in get_page_from_freelist()) refusing to consider the zones for |
| 2194 | * any node on the zonelist except the first. By the time any such |
| 2195 | * calls get to this routine, we should just shut up and say 'yes'. |
| 2196 | * |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2197 | * GFP_USER allocations are marked with the __GFP_HARDWALL bit, |
David Rientjes | c596d9f | 2007-05-06 14:49:32 -0700 | [diff] [blame] | 2198 | * and do not allow allocations outside the current tasks cpuset |
| 2199 | * unless the task has been OOM killed as is marked TIF_MEMDIE. |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2200 | * GFP_KERNEL allocations are not so marked, so can escape to the |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 2201 | * nearest enclosing hardwalled ancestor cpuset. |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2202 | * |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2203 | * Scanning up parent cpusets requires callback_mutex. The |
| 2204 | * __alloc_pages() routine only calls here with __GFP_HARDWALL bit |
| 2205 | * _not_ set if it's a GFP_KERNEL allocation, and all nodes in the |
| 2206 | * current tasks mems_allowed came up empty on the first pass over |
| 2207 | * the zonelist. So only GFP_KERNEL allocations, if all nodes in the |
| 2208 | * cpuset are short of memory, might require taking the callback_mutex |
| 2209 | * mutex. |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2210 | * |
Paul Jackson | 36be57f | 2006-05-20 15:00:10 -0700 | [diff] [blame] | 2211 | * The first call here from mm/page_alloc:get_page_from_freelist() |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2212 | * has __GFP_HARDWALL set in gfp_mask, enforcing hardwall cpusets, |
| 2213 | * so no allocation on a node outside the cpuset is allowed (unless |
| 2214 | * in interrupt, of course). |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2215 | * |
Paul Jackson | 36be57f | 2006-05-20 15:00:10 -0700 | [diff] [blame] | 2216 | * The second pass through get_page_from_freelist() doesn't even call |
| 2217 | * here for GFP_ATOMIC calls. For those calls, the __alloc_pages() |
| 2218 | * variable 'wait' is not set, and the bit ALLOC_CPUSET is not set |
| 2219 | * in alloc_flags. That logic and the checks below have the combined |
| 2220 | * affect that: |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2221 | * in_interrupt - any node ok (current task context irrelevant) |
| 2222 | * GFP_ATOMIC - any node ok |
David Rientjes | c596d9f | 2007-05-06 14:49:32 -0700 | [diff] [blame] | 2223 | * TIF_MEMDIE - any node ok |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 2224 | * GFP_KERNEL - any node in enclosing hardwalled cpuset ok |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2225 | * GFP_USER - only nodes in current tasks mems allowed ok. |
Paul Jackson | 36be57f | 2006-05-20 15:00:10 -0700 | [diff] [blame] | 2226 | * |
| 2227 | * Rule: |
David Rientjes | a1bc5a4 | 2009-04-02 16:57:54 -0700 | [diff] [blame] | 2228 | * Don't call cpuset_node_allowed_softwall if you can't sleep, unless you |
Paul Jackson | 36be57f | 2006-05-20 15:00:10 -0700 | [diff] [blame] | 2229 | * pass in the __GFP_HARDWALL flag set in gfp_flag, which disables |
| 2230 | * the code that might scan up ancestor cpusets and sleep. |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2231 | */ |
David Rientjes | a1bc5a4 | 2009-04-02 16:57:54 -0700 | [diff] [blame] | 2232 | int __cpuset_node_allowed_softwall(int node, gfp_t gfp_mask) |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2233 | { |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2234 | const struct cpuset *cs; /* current cpuset ancestors */ |
Paul Jackson | 29afd49 | 2006-03-24 03:16:12 -0800 | [diff] [blame] | 2235 | int allowed; /* is allocation in zone z allowed? */ |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2236 | |
Christoph Lameter | 9b819d2 | 2006-09-25 23:31:40 -0700 | [diff] [blame] | 2237 | if (in_interrupt() || (gfp_mask & __GFP_THISNODE)) |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2238 | return 1; |
Paul Jackson | 92d1dbd | 2006-05-20 15:00:11 -0700 | [diff] [blame] | 2239 | might_sleep_if(!(gfp_mask & __GFP_HARDWALL)); |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2240 | if (node_isset(node, current->mems_allowed)) |
| 2241 | return 1; |
David Rientjes | c596d9f | 2007-05-06 14:49:32 -0700 | [diff] [blame] | 2242 | /* |
| 2243 | * Allow tasks that have access to memory reserves because they have |
| 2244 | * been OOM killed to get memory anywhere. |
| 2245 | */ |
| 2246 | if (unlikely(test_thread_flag(TIF_MEMDIE))) |
| 2247 | return 1; |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2248 | if (gfp_mask & __GFP_HARDWALL) /* If hardwall request, stop here */ |
| 2249 | return 0; |
| 2250 | |
Bob Picco | 5563e77 | 2005-11-13 16:06:35 -0800 | [diff] [blame] | 2251 | if (current->flags & PF_EXITING) /* Let dying task have memory */ |
| 2252 | return 1; |
| 2253 | |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2254 | /* Not hardwall and node outside mems_allowed: scan up cpusets */ |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 2255 | mutex_lock(&callback_mutex); |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 2256 | |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 2257 | task_lock(current); |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 2258 | cs = nearest_hardwall_ancestor(task_cs(current)); |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 2259 | task_unlock(current); |
| 2260 | |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2261 | allowed = node_isset(node, cs->mems_allowed); |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 2262 | mutex_unlock(&callback_mutex); |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2263 | return allowed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2264 | } |
| 2265 | |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2266 | /* |
David Rientjes | a1bc5a4 | 2009-04-02 16:57:54 -0700 | [diff] [blame] | 2267 | * cpuset_node_allowed_hardwall - Can we allocate on a memory node? |
| 2268 | * @node: is this an allowed node? |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2269 | * @gfp_mask: memory allocation flags |
| 2270 | * |
David Rientjes | a1bc5a4 | 2009-04-02 16:57:54 -0700 | [diff] [blame] | 2271 | * If we're in interrupt, yes, we can always allocate. If __GFP_THISNODE is |
| 2272 | * set, yes, we can always allocate. If node is in our task's mems_allowed, |
| 2273 | * yes. If the task has been OOM killed and has access to memory reserves as |
| 2274 | * specified by the TIF_MEMDIE flag, yes. |
| 2275 | * Otherwise, no. |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2276 | * |
| 2277 | * The __GFP_THISNODE placement logic is really handled elsewhere, |
| 2278 | * by forcibly using a zonelist starting at a specified node, and by |
| 2279 | * (in get_page_from_freelist()) refusing to consider the zones for |
| 2280 | * any node on the zonelist except the first. By the time any such |
| 2281 | * calls get to this routine, we should just shut up and say 'yes'. |
| 2282 | * |
David Rientjes | a1bc5a4 | 2009-04-02 16:57:54 -0700 | [diff] [blame] | 2283 | * Unlike the cpuset_node_allowed_softwall() variant, above, |
| 2284 | * this variant requires that the node be in the current task's |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2285 | * mems_allowed or that we're in interrupt. It does not scan up the |
| 2286 | * cpuset hierarchy for the nearest enclosing mem_exclusive cpuset. |
| 2287 | * It never sleeps. |
| 2288 | */ |
David Rientjes | a1bc5a4 | 2009-04-02 16:57:54 -0700 | [diff] [blame] | 2289 | int __cpuset_node_allowed_hardwall(int node, gfp_t gfp_mask) |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2290 | { |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2291 | if (in_interrupt() || (gfp_mask & __GFP_THISNODE)) |
| 2292 | return 1; |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2293 | if (node_isset(node, current->mems_allowed)) |
| 2294 | return 1; |
Daniel Walker | dedf8b7 | 2007-10-18 03:06:04 -0700 | [diff] [blame] | 2295 | /* |
| 2296 | * Allow tasks that have access to memory reserves because they have |
| 2297 | * been OOM killed to get memory anywhere. |
| 2298 | */ |
| 2299 | if (unlikely(test_thread_flag(TIF_MEMDIE))) |
| 2300 | return 1; |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2301 | return 0; |
| 2302 | } |
| 2303 | |
Paul Jackson | ef08e3b | 2005-09-06 15:18:13 -0700 | [diff] [blame] | 2304 | /** |
Paul Jackson | 505970b | 2006-01-14 13:21:06 -0800 | [diff] [blame] | 2305 | * cpuset_lock - lock out any changes to cpuset structures |
| 2306 | * |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 2307 | * The out of memory (oom) code needs to mutex_lock cpusets |
Paul Jackson | 505970b | 2006-01-14 13:21:06 -0800 | [diff] [blame] | 2308 | * from being changed while it scans the tasklist looking for a |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 2309 | * task in an overlapping cpuset. Expose callback_mutex via this |
Paul Jackson | 505970b | 2006-01-14 13:21:06 -0800 | [diff] [blame] | 2310 | * cpuset_lock() routine, so the oom code can lock it, before |
| 2311 | * locking the task list. The tasklist_lock is a spinlock, so |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 2312 | * must be taken inside callback_mutex. |
Paul Jackson | 505970b | 2006-01-14 13:21:06 -0800 | [diff] [blame] | 2313 | */ |
| 2314 | |
| 2315 | void cpuset_lock(void) |
| 2316 | { |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 2317 | mutex_lock(&callback_mutex); |
Paul Jackson | 505970b | 2006-01-14 13:21:06 -0800 | [diff] [blame] | 2318 | } |
| 2319 | |
| 2320 | /** |
| 2321 | * cpuset_unlock - release lock on cpuset changes |
| 2322 | * |
| 2323 | * Undo the lock taken in a previous cpuset_lock() call. |
| 2324 | */ |
| 2325 | |
| 2326 | void cpuset_unlock(void) |
| 2327 | { |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 2328 | mutex_unlock(&callback_mutex); |
Paul Jackson | 505970b | 2006-01-14 13:21:06 -0800 | [diff] [blame] | 2329 | } |
| 2330 | |
| 2331 | /** |
Paul Jackson | 825a46a | 2006-03-24 03:16:03 -0800 | [diff] [blame] | 2332 | * cpuset_mem_spread_node() - On which node to begin search for a page |
| 2333 | * |
| 2334 | * If a task is marked PF_SPREAD_PAGE or PF_SPREAD_SLAB (as for |
| 2335 | * tasks in a cpuset with is_spread_page or is_spread_slab set), |
| 2336 | * and if the memory allocation used cpuset_mem_spread_node() |
| 2337 | * to determine on which node to start looking, as it will for |
| 2338 | * certain page cache or slab cache pages such as used for file |
| 2339 | * system buffers and inode caches, then instead of starting on the |
| 2340 | * local node to look for a free page, rather spread the starting |
| 2341 | * node around the tasks mems_allowed nodes. |
| 2342 | * |
| 2343 | * We don't have to worry about the returned node being offline |
| 2344 | * because "it can't happen", and even if it did, it would be ok. |
| 2345 | * |
| 2346 | * The routines calling guarantee_online_mems() are careful to |
| 2347 | * only set nodes in task->mems_allowed that are online. So it |
| 2348 | * should not be possible for the following code to return an |
| 2349 | * offline node. But if it did, that would be ok, as this routine |
| 2350 | * is not returning the node where the allocation must be, only |
| 2351 | * the node where the search should start. The zonelist passed to |
| 2352 | * __alloc_pages() will include all nodes. If the slab allocator |
| 2353 | * is passed an offline node, it will fall back to the local node. |
| 2354 | * See kmem_cache_alloc_node(). |
| 2355 | */ |
| 2356 | |
| 2357 | int cpuset_mem_spread_node(void) |
| 2358 | { |
| 2359 | int node; |
| 2360 | |
| 2361 | node = next_node(current->cpuset_mem_spread_rotor, current->mems_allowed); |
| 2362 | if (node == MAX_NUMNODES) |
| 2363 | node = first_node(current->mems_allowed); |
| 2364 | current->cpuset_mem_spread_rotor = node; |
| 2365 | return node; |
| 2366 | } |
| 2367 | EXPORT_SYMBOL_GPL(cpuset_mem_spread_node); |
| 2368 | |
| 2369 | /** |
David Rientjes | bbe373f | 2007-10-16 23:25:58 -0700 | [diff] [blame] | 2370 | * cpuset_mems_allowed_intersects - Does @tsk1's mems_allowed intersect @tsk2's? |
| 2371 | * @tsk1: pointer to task_struct of some task. |
| 2372 | * @tsk2: pointer to task_struct of some other task. |
Paul Jackson | ef08e3b | 2005-09-06 15:18:13 -0700 | [diff] [blame] | 2373 | * |
David Rientjes | bbe373f | 2007-10-16 23:25:58 -0700 | [diff] [blame] | 2374 | * Description: Return true if @tsk1's mems_allowed intersects the |
| 2375 | * mems_allowed of @tsk2. Used by the OOM killer to determine if |
| 2376 | * one of the task's memory usage might impact the memory available |
| 2377 | * to the other. |
Paul Jackson | ef08e3b | 2005-09-06 15:18:13 -0700 | [diff] [blame] | 2378 | **/ |
| 2379 | |
David Rientjes | bbe373f | 2007-10-16 23:25:58 -0700 | [diff] [blame] | 2380 | int cpuset_mems_allowed_intersects(const struct task_struct *tsk1, |
| 2381 | const struct task_struct *tsk2) |
Paul Jackson | ef08e3b | 2005-09-06 15:18:13 -0700 | [diff] [blame] | 2382 | { |
David Rientjes | bbe373f | 2007-10-16 23:25:58 -0700 | [diff] [blame] | 2383 | return nodes_intersects(tsk1->mems_allowed, tsk2->mems_allowed); |
Paul Jackson | ef08e3b | 2005-09-06 15:18:13 -0700 | [diff] [blame] | 2384 | } |
| 2385 | |
David Rientjes | 75aa199 | 2009-01-06 14:39:01 -0800 | [diff] [blame] | 2386 | /** |
| 2387 | * cpuset_print_task_mems_allowed - prints task's cpuset and mems_allowed |
| 2388 | * @task: pointer to task_struct of some task. |
| 2389 | * |
| 2390 | * Description: Prints @task's name, cpuset name, and cached copy of its |
| 2391 | * mems_allowed to the kernel log. Must hold task_lock(task) to allow |
| 2392 | * dereferencing task_cs(task). |
| 2393 | */ |
| 2394 | void cpuset_print_task_mems_allowed(struct task_struct *tsk) |
| 2395 | { |
| 2396 | struct dentry *dentry; |
| 2397 | |
| 2398 | dentry = task_cs(tsk)->css.cgroup->dentry; |
| 2399 | spin_lock(&cpuset_buffer_lock); |
| 2400 | snprintf(cpuset_name, CPUSET_NAME_LEN, |
| 2401 | dentry ? (const char *)dentry->d_name.name : "/"); |
| 2402 | nodelist_scnprintf(cpuset_nodelist, CPUSET_NODELIST_LEN, |
| 2403 | tsk->mems_allowed); |
| 2404 | printk(KERN_INFO "%s cpuset=%s mems_allowed=%s\n", |
| 2405 | tsk->comm, cpuset_name, cpuset_nodelist); |
| 2406 | spin_unlock(&cpuset_buffer_lock); |
| 2407 | } |
| 2408 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2409 | /* |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 2410 | * Collection of memory_pressure is suppressed unless |
| 2411 | * this flag is enabled by writing "1" to the special |
| 2412 | * cpuset file 'memory_pressure_enabled' in the root cpuset. |
| 2413 | */ |
| 2414 | |
Paul Jackson | c5b2aff | 2006-01-08 01:01:51 -0800 | [diff] [blame] | 2415 | int cpuset_memory_pressure_enabled __read_mostly; |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 2416 | |
| 2417 | /** |
| 2418 | * cpuset_memory_pressure_bump - keep stats of per-cpuset reclaims. |
| 2419 | * |
| 2420 | * Keep a running average of the rate of synchronous (direct) |
| 2421 | * page reclaim efforts initiated by tasks in each cpuset. |
| 2422 | * |
| 2423 | * This represents the rate at which some task in the cpuset |
| 2424 | * ran low on memory on all nodes it was allowed to use, and |
| 2425 | * had to enter the kernels page reclaim code in an effort to |
| 2426 | * create more free memory by tossing clean pages or swapping |
| 2427 | * or writing dirty pages. |
| 2428 | * |
| 2429 | * Display to user space in the per-cpuset read-only file |
| 2430 | * "memory_pressure". Value displayed is an integer |
| 2431 | * representing the recent rate of entry into the synchronous |
| 2432 | * (direct) page reclaim by any task attached to the cpuset. |
| 2433 | **/ |
| 2434 | |
| 2435 | void __cpuset_memory_pressure_bump(void) |
| 2436 | { |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 2437 | task_lock(current); |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 2438 | fmeter_markevent(&task_cs(current)->fmeter); |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 2439 | task_unlock(current); |
| 2440 | } |
| 2441 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 2442 | #ifdef CONFIG_PROC_PID_CPUSET |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 2443 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2444 | * proc_cpuset_show() |
| 2445 | * - Print tasks cpuset path into seq_file. |
| 2446 | * - Used for /proc/<pid>/cpuset. |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 2447 | * - No need to task_lock(tsk) on this tsk->cpuset reference, as it |
| 2448 | * doesn't really matter if tsk->cpuset changes after we read it, |
Paul Jackson | c8d9c90 | 2008-02-07 00:14:46 -0800 | [diff] [blame] | 2449 | * and we take cgroup_mutex, keeping cpuset_attach() from changing it |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 2450 | * anyway. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2451 | */ |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 2452 | static int proc_cpuset_show(struct seq_file *m, void *unused_v) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2453 | { |
Eric W. Biederman | 13b41b0 | 2006-06-26 00:25:56 -0700 | [diff] [blame] | 2454 | struct pid *pid; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2455 | struct task_struct *tsk; |
| 2456 | char *buf; |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 2457 | struct cgroup_subsys_state *css; |
Eric W. Biederman | 99f8955 | 2006-06-26 00:25:55 -0700 | [diff] [blame] | 2458 | int retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2459 | |
Eric W. Biederman | 99f8955 | 2006-06-26 00:25:55 -0700 | [diff] [blame] | 2460 | retval = -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2461 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
| 2462 | if (!buf) |
Eric W. Biederman | 99f8955 | 2006-06-26 00:25:55 -0700 | [diff] [blame] | 2463 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2464 | |
Eric W. Biederman | 99f8955 | 2006-06-26 00:25:55 -0700 | [diff] [blame] | 2465 | retval = -ESRCH; |
Eric W. Biederman | 13b41b0 | 2006-06-26 00:25:56 -0700 | [diff] [blame] | 2466 | pid = m->private; |
| 2467 | tsk = get_pid_task(pid, PIDTYPE_PID); |
Eric W. Biederman | 99f8955 | 2006-06-26 00:25:55 -0700 | [diff] [blame] | 2468 | if (!tsk) |
| 2469 | goto out_free; |
| 2470 | |
| 2471 | retval = -EINVAL; |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 2472 | cgroup_lock(); |
| 2473 | css = task_subsys_state(tsk, cpuset_subsys_id); |
| 2474 | retval = cgroup_path(css->cgroup, buf, PAGE_SIZE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2475 | if (retval < 0) |
Eric W. Biederman | 99f8955 | 2006-06-26 00:25:55 -0700 | [diff] [blame] | 2476 | goto out_unlock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2477 | seq_puts(m, buf); |
| 2478 | seq_putc(m, '\n'); |
Eric W. Biederman | 99f8955 | 2006-06-26 00:25:55 -0700 | [diff] [blame] | 2479 | out_unlock: |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 2480 | cgroup_unlock(); |
Eric W. Biederman | 99f8955 | 2006-06-26 00:25:55 -0700 | [diff] [blame] | 2481 | put_task_struct(tsk); |
| 2482 | out_free: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2483 | kfree(buf); |
Eric W. Biederman | 99f8955 | 2006-06-26 00:25:55 -0700 | [diff] [blame] | 2484 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2485 | return retval; |
| 2486 | } |
| 2487 | |
| 2488 | static int cpuset_open(struct inode *inode, struct file *file) |
| 2489 | { |
Eric W. Biederman | 13b41b0 | 2006-06-26 00:25:56 -0700 | [diff] [blame] | 2490 | struct pid *pid = PROC_I(inode)->pid; |
| 2491 | return single_open(file, proc_cpuset_show, pid); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2492 | } |
| 2493 | |
Arjan van de Ven | 9a32144 | 2007-02-12 00:55:35 -0800 | [diff] [blame] | 2494 | const struct file_operations proc_cpuset_operations = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2495 | .open = cpuset_open, |
| 2496 | .read = seq_read, |
| 2497 | .llseek = seq_lseek, |
| 2498 | .release = single_release, |
| 2499 | }; |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 2500 | #endif /* CONFIG_PROC_PID_CPUSET */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2501 | |
| 2502 | /* Display task cpus_allowed, mems_allowed in /proc/<pid>/status file. */ |
Eric W. Biederman | df5f831 | 2008-02-08 04:18:33 -0800 | [diff] [blame] | 2503 | void cpuset_task_status_allowed(struct seq_file *m, struct task_struct *task) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2504 | { |
Eric W. Biederman | df5f831 | 2008-02-08 04:18:33 -0800 | [diff] [blame] | 2505 | seq_printf(m, "Cpus_allowed:\t"); |
Lai Jiangshan | 30e8e13 | 2008-10-18 20:28:20 -0700 | [diff] [blame] | 2506 | seq_cpumask(m, &task->cpus_allowed); |
Eric W. Biederman | df5f831 | 2008-02-08 04:18:33 -0800 | [diff] [blame] | 2507 | seq_printf(m, "\n"); |
Mike Travis | 39106dc | 2008-04-08 11:43:03 -0700 | [diff] [blame] | 2508 | seq_printf(m, "Cpus_allowed_list:\t"); |
Lai Jiangshan | 30e8e13 | 2008-10-18 20:28:20 -0700 | [diff] [blame] | 2509 | seq_cpumask_list(m, &task->cpus_allowed); |
Mike Travis | 39106dc | 2008-04-08 11:43:03 -0700 | [diff] [blame] | 2510 | seq_printf(m, "\n"); |
Eric W. Biederman | df5f831 | 2008-02-08 04:18:33 -0800 | [diff] [blame] | 2511 | seq_printf(m, "Mems_allowed:\t"); |
Lai Jiangshan | 30e8e13 | 2008-10-18 20:28:20 -0700 | [diff] [blame] | 2512 | seq_nodemask(m, &task->mems_allowed); |
Eric W. Biederman | df5f831 | 2008-02-08 04:18:33 -0800 | [diff] [blame] | 2513 | seq_printf(m, "\n"); |
Mike Travis | 39106dc | 2008-04-08 11:43:03 -0700 | [diff] [blame] | 2514 | seq_printf(m, "Mems_allowed_list:\t"); |
Lai Jiangshan | 30e8e13 | 2008-10-18 20:28:20 -0700 | [diff] [blame] | 2515 | seq_nodemask_list(m, &task->mems_allowed); |
Mike Travis | 39106dc | 2008-04-08 11:43:03 -0700 | [diff] [blame] | 2516 | seq_printf(m, "\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2517 | } |