Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/cpuset.c | |
3 | * | |
4 | * Processor and Memory placement constraints for sets of tasks. | |
5 | * | |
6 | * Copyright (C) 2003 BULL SA. | |
029190c5 | 7 | * Copyright (C) 2004-2007 Silicon Graphics, Inc. |
8793d854 | 8 | * Copyright (C) 2006 Google, Inc |
1da177e4 LT |
9 | * |
10 | * Portions derived from Patrick Mochel's sysfs code. | |
11 | * sysfs is Copyright (c) 2001-3 Patrick Mochel | |
1da177e4 | 12 | * |
825a46af | 13 | * 2003-10-10 Written by Simon Derr. |
1da177e4 | 14 | * 2003-10-22 Updates by Stephen Hemminger. |
825a46af | 15 | * 2004 May-July Rework by Paul Jackson. |
8793d854 | 16 | * 2006 Rework by Paul Menage to use generic cgroups |
cf417141 MK |
17 | * 2008 Rework of the scheduler domains and CPU hotplug handling |
18 | * by Max Krasnyansky | |
1da177e4 LT |
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 | ||
1da177e4 LT |
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> | |
68860ec1 | 37 | #include <linux/mempolicy.h> |
1da177e4 | 38 | #include <linux/mm.h> |
f481891f | 39 | #include <linux/memory.h> |
9984de1a | 40 | #include <linux/export.h> |
1da177e4 LT |
41 | #include <linux/mount.h> |
42 | #include <linux/namei.h> | |
43 | #include <linux/pagemap.h> | |
44 | #include <linux/proc_fs.h> | |
6b9c2603 | 45 | #include <linux/rcupdate.h> |
1da177e4 | 46 | #include <linux/sched.h> |
6e84f315 | 47 | #include <linux/sched/mm.h> |
f719ff9b | 48 | #include <linux/sched/task.h> |
1da177e4 | 49 | #include <linux/seq_file.h> |
22fb52dd | 50 | #include <linux/security.h> |
1da177e4 | 51 | #include <linux/slab.h> |
1da177e4 LT |
52 | #include <linux/spinlock.h> |
53 | #include <linux/stat.h> | |
54 | #include <linux/string.h> | |
55 | #include <linux/time.h> | |
d2b43658 | 56 | #include <linux/time64.h> |
1da177e4 LT |
57 | #include <linux/backing-dev.h> |
58 | #include <linux/sort.h> | |
da99ecf1 | 59 | #include <linux/oom.h> |
edb93821 | 60 | #include <linux/sched/isolation.h> |
7c0f6ba6 | 61 | #include <linux/uaccess.h> |
60063497 | 62 | #include <linux/atomic.h> |
3d3f26a7 | 63 | #include <linux/mutex.h> |
956db3ca | 64 | #include <linux/cgroup.h> |
e44193d3 | 65 | #include <linux/wait.h> |
1da177e4 | 66 | |
89affbf5 | 67 | DEFINE_STATIC_KEY_FALSE(cpusets_pre_enable_key); |
002f2906 | 68 | DEFINE_STATIC_KEY_FALSE(cpusets_enabled_key); |
202f72d5 | 69 | |
3e0d98b9 PJ |
70 | /* See "Frequency meter" comments, below. */ |
71 | ||
72 | struct fmeter { | |
73 | int cnt; /* unprocessed events count */ | |
74 | int val; /* most recent output value */ | |
d2b43658 | 75 | time64_t time; /* clock (secs) when val computed */ |
3e0d98b9 PJ |
76 | spinlock_t lock; /* guards read or write of above */ |
77 | }; | |
78 | ||
1da177e4 | 79 | struct cpuset { |
8793d854 PM |
80 | struct cgroup_subsys_state css; |
81 | ||
1da177e4 | 82 | unsigned long flags; /* "unsigned long" so bitops work */ |
e2b9a3d7 | 83 | |
7e88291b LZ |
84 | /* |
85 | * On default hierarchy: | |
86 | * | |
87 | * The user-configured masks can only be changed by writing to | |
88 | * cpuset.cpus and cpuset.mems, and won't be limited by the | |
89 | * parent masks. | |
90 | * | |
91 | * The effective masks is the real masks that apply to the tasks | |
92 | * in the cpuset. They may be changed if the configured masks are | |
93 | * changed or hotplug happens. | |
94 | * | |
95 | * effective_mask == configured_mask & parent's effective_mask, | |
96 | * and if it ends up empty, it will inherit the parent's mask. | |
97 | * | |
98 | * | |
99 | * On legacy hierachy: | |
100 | * | |
101 | * The user-configured masks are always the same with effective masks. | |
102 | */ | |
103 | ||
e2b9a3d7 LZ |
104 | /* user-configured CPUs and Memory Nodes allow to tasks */ |
105 | cpumask_var_t cpus_allowed; | |
106 | nodemask_t mems_allowed; | |
107 | ||
108 | /* effective CPUs and Memory Nodes allow to tasks */ | |
109 | cpumask_var_t effective_cpus; | |
110 | nodemask_t effective_mems; | |
1da177e4 | 111 | |
33ad801d LZ |
112 | /* |
113 | * This is old Memory Nodes tasks took on. | |
114 | * | |
115 | * - top_cpuset.old_mems_allowed is initialized to mems_allowed. | |
116 | * - A new cpuset's old_mems_allowed is initialized when some | |
117 | * task is moved into it. | |
118 | * - old_mems_allowed is used in cpuset_migrate_mm() when we change | |
119 | * cpuset.mems_allowed and have tasks' nodemask updated, and | |
120 | * then old_mems_allowed is updated to mems_allowed. | |
121 | */ | |
122 | nodemask_t old_mems_allowed; | |
123 | ||
3e0d98b9 | 124 | struct fmeter fmeter; /* memory_pressure filter */ |
029190c5 | 125 | |
452477fa TH |
126 | /* |
127 | * Tasks are being attached to this cpuset. Used to prevent | |
128 | * zeroing cpus/mems_allowed between ->can_attach() and ->attach(). | |
129 | */ | |
130 | int attach_in_progress; | |
131 | ||
029190c5 PJ |
132 | /* partition number for rebuild_sched_domains() */ |
133 | int pn; | |
956db3ca | 134 | |
1d3504fc HS |
135 | /* for custom sched domain */ |
136 | int relax_domain_level; | |
1da177e4 LT |
137 | }; |
138 | ||
a7c6d554 | 139 | static inline struct cpuset *css_cs(struct cgroup_subsys_state *css) |
8793d854 | 140 | { |
a7c6d554 | 141 | return css ? container_of(css, struct cpuset, css) : NULL; |
8793d854 PM |
142 | } |
143 | ||
144 | /* Retrieve the cpuset for a task */ | |
145 | static inline struct cpuset *task_cs(struct task_struct *task) | |
146 | { | |
073219e9 | 147 | return css_cs(task_css(task, cpuset_cgrp_id)); |
8793d854 | 148 | } |
8793d854 | 149 | |
c9710d80 | 150 | static inline struct cpuset *parent_cs(struct cpuset *cs) |
c431069f | 151 | { |
5c9d535b | 152 | return css_cs(cs->css.parent); |
c431069f TH |
153 | } |
154 | ||
b246272e DR |
155 | #ifdef CONFIG_NUMA |
156 | static inline bool task_has_mempolicy(struct task_struct *task) | |
157 | { | |
158 | return task->mempolicy; | |
159 | } | |
160 | #else | |
161 | static inline bool task_has_mempolicy(struct task_struct *task) | |
162 | { | |
163 | return false; | |
164 | } | |
165 | #endif | |
166 | ||
167 | ||
1da177e4 LT |
168 | /* bits in struct cpuset flags field */ |
169 | typedef enum { | |
efeb77b2 | 170 | CS_ONLINE, |
1da177e4 LT |
171 | CS_CPU_EXCLUSIVE, |
172 | CS_MEM_EXCLUSIVE, | |
78608366 | 173 | CS_MEM_HARDWALL, |
45b07ef3 | 174 | CS_MEMORY_MIGRATE, |
029190c5 | 175 | CS_SCHED_LOAD_BALANCE, |
825a46af PJ |
176 | CS_SPREAD_PAGE, |
177 | CS_SPREAD_SLAB, | |
1da177e4 LT |
178 | } cpuset_flagbits_t; |
179 | ||
180 | /* convenient tests for these bits */ | |
41c25707 | 181 | static inline bool is_cpuset_online(struct cpuset *cs) |
efeb77b2 | 182 | { |
41c25707 | 183 | return test_bit(CS_ONLINE, &cs->flags) && !css_is_dying(&cs->css); |
efeb77b2 TH |
184 | } |
185 | ||
1da177e4 LT |
186 | static inline int is_cpu_exclusive(const struct cpuset *cs) |
187 | { | |
7b5b9ef0 | 188 | return test_bit(CS_CPU_EXCLUSIVE, &cs->flags); |
1da177e4 LT |
189 | } |
190 | ||
191 | static inline int is_mem_exclusive(const struct cpuset *cs) | |
192 | { | |
7b5b9ef0 | 193 | return test_bit(CS_MEM_EXCLUSIVE, &cs->flags); |
1da177e4 LT |
194 | } |
195 | ||
78608366 PM |
196 | static inline int is_mem_hardwall(const struct cpuset *cs) |
197 | { | |
198 | return test_bit(CS_MEM_HARDWALL, &cs->flags); | |
199 | } | |
200 | ||
029190c5 PJ |
201 | static inline int is_sched_load_balance(const struct cpuset *cs) |
202 | { | |
203 | return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); | |
204 | } | |
205 | ||
45b07ef3 PJ |
206 | static inline int is_memory_migrate(const struct cpuset *cs) |
207 | { | |
7b5b9ef0 | 208 | return test_bit(CS_MEMORY_MIGRATE, &cs->flags); |
45b07ef3 PJ |
209 | } |
210 | ||
825a46af PJ |
211 | static inline int is_spread_page(const struct cpuset *cs) |
212 | { | |
213 | return test_bit(CS_SPREAD_PAGE, &cs->flags); | |
214 | } | |
215 | ||
216 | static inline int is_spread_slab(const struct cpuset *cs) | |
217 | { | |
218 | return test_bit(CS_SPREAD_SLAB, &cs->flags); | |
219 | } | |
220 | ||
1da177e4 | 221 | static struct cpuset top_cpuset = { |
efeb77b2 TH |
222 | .flags = ((1 << CS_ONLINE) | (1 << CS_CPU_EXCLUSIVE) | |
223 | (1 << CS_MEM_EXCLUSIVE)), | |
1da177e4 LT |
224 | }; |
225 | ||
ae8086ce TH |
226 | /** |
227 | * cpuset_for_each_child - traverse online children of a cpuset | |
228 | * @child_cs: loop cursor pointing to the current child | |
492eb21b | 229 | * @pos_css: used for iteration |
ae8086ce TH |
230 | * @parent_cs: target cpuset to walk children of |
231 | * | |
232 | * Walk @child_cs through the online children of @parent_cs. Must be used | |
233 | * with RCU read locked. | |
234 | */ | |
492eb21b TH |
235 | #define cpuset_for_each_child(child_cs, pos_css, parent_cs) \ |
236 | css_for_each_child((pos_css), &(parent_cs)->css) \ | |
237 | if (is_cpuset_online(((child_cs) = css_cs((pos_css))))) | |
ae8086ce | 238 | |
fc560a26 TH |
239 | /** |
240 | * cpuset_for_each_descendant_pre - pre-order walk of a cpuset's descendants | |
241 | * @des_cs: loop cursor pointing to the current descendant | |
492eb21b | 242 | * @pos_css: used for iteration |
fc560a26 TH |
243 | * @root_cs: target cpuset to walk ancestor of |
244 | * | |
245 | * Walk @des_cs through the online descendants of @root_cs. Must be used | |
492eb21b | 246 | * with RCU read locked. The caller may modify @pos_css by calling |
bd8815a6 TH |
247 | * css_rightmost_descendant() to skip subtree. @root_cs is included in the |
248 | * iteration and the first node to be visited. | |
fc560a26 | 249 | */ |
492eb21b TH |
250 | #define cpuset_for_each_descendant_pre(des_cs, pos_css, root_cs) \ |
251 | css_for_each_descendant_pre((pos_css), &(root_cs)->css) \ | |
252 | if (is_cpuset_online(((des_cs) = css_cs((pos_css))))) | |
fc560a26 | 253 | |
1da177e4 | 254 | /* |
8447a0fe VD |
255 | * There are two global locks guarding cpuset structures - cpuset_mutex and |
256 | * callback_lock. We also require taking task_lock() when dereferencing a | |
257 | * task's cpuset pointer. See "The task_lock() exception", at the end of this | |
258 | * comment. | |
5d21cc2d | 259 | * |
8447a0fe | 260 | * A task must hold both locks to modify cpusets. If a task holds |
5d21cc2d | 261 | * cpuset_mutex, then it blocks others wanting that mutex, ensuring that it |
8447a0fe | 262 | * is the only task able to also acquire callback_lock and be able to |
5d21cc2d TH |
263 | * modify cpusets. It can perform various checks on the cpuset structure |
264 | * first, knowing nothing will change. It can also allocate memory while | |
265 | * just holding cpuset_mutex. While it is performing these checks, various | |
8447a0fe VD |
266 | * callback routines can briefly acquire callback_lock to query cpusets. |
267 | * Once it is ready to make the changes, it takes callback_lock, blocking | |
5d21cc2d | 268 | * everyone else. |
053199ed PJ |
269 | * |
270 | * Calls to the kernel memory allocator can not be made while holding | |
8447a0fe | 271 | * callback_lock, as that would risk double tripping on callback_lock |
053199ed PJ |
272 | * from one of the callbacks into the cpuset code from within |
273 | * __alloc_pages(). | |
274 | * | |
8447a0fe | 275 | * If a task is only holding callback_lock, then it has read-only |
053199ed PJ |
276 | * access to cpusets. |
277 | * | |
58568d2a MX |
278 | * Now, the task_struct fields mems_allowed and mempolicy may be changed |
279 | * by other task, we use alloc_lock in the task_struct fields to protect | |
280 | * them. | |
053199ed | 281 | * |
8447a0fe | 282 | * The cpuset_common_file_read() handlers only hold callback_lock across |
053199ed PJ |
283 | * small pieces of code, such as when reading out possibly multi-word |
284 | * cpumasks and nodemasks. | |
285 | * | |
2df167a3 PM |
286 | * Accessing a task's cpuset should be done in accordance with the |
287 | * guidelines for accessing subsystem state in kernel/cgroup.c | |
1da177e4 LT |
288 | */ |
289 | ||
5d21cc2d | 290 | static DEFINE_MUTEX(cpuset_mutex); |
8447a0fe | 291 | static DEFINE_SPINLOCK(callback_lock); |
4247bdc6 | 292 | |
e93ad19d TH |
293 | static struct workqueue_struct *cpuset_migrate_mm_wq; |
294 | ||
3a5a6d0c TH |
295 | /* |
296 | * CPU / memory hotplug is handled asynchronously. | |
297 | */ | |
298 | static void cpuset_hotplug_workfn(struct work_struct *work); | |
3a5a6d0c TH |
299 | static DECLARE_WORK(cpuset_hotplug_work, cpuset_hotplug_workfn); |
300 | ||
e44193d3 LZ |
301 | static DECLARE_WAIT_QUEUE_HEAD(cpuset_attach_wq); |
302 | ||
b8d1b8ee WL |
303 | /* |
304 | * Cgroup v2 behavior is used when on default hierarchy or the | |
305 | * cgroup_v2_mode flag is set. | |
306 | */ | |
307 | static inline bool is_in_v2_mode(void) | |
308 | { | |
309 | return cgroup_subsys_on_dfl(cpuset_cgrp_subsys) || | |
310 | (cpuset_cgrp_subsys.root->flags & CGRP_ROOT_CPUSET_V2_MODE); | |
311 | } | |
312 | ||
cf417141 MK |
313 | /* |
314 | * This is ugly, but preserves the userspace API for existing cpuset | |
8793d854 | 315 | * users. If someone tries to mount the "cpuset" filesystem, we |
cf417141 MK |
316 | * silently switch it to mount "cgroup" instead |
317 | */ | |
f7e83571 AV |
318 | static struct dentry *cpuset_mount(struct file_system_type *fs_type, |
319 | int flags, const char *unused_dev_name, void *data) | |
1da177e4 | 320 | { |
8793d854 | 321 | struct file_system_type *cgroup_fs = get_fs_type("cgroup"); |
f7e83571 | 322 | struct dentry *ret = ERR_PTR(-ENODEV); |
8793d854 PM |
323 | if (cgroup_fs) { |
324 | char mountopts[] = | |
325 | "cpuset,noprefix," | |
326 | "release_agent=/sbin/cpuset_release_agent"; | |
f7e83571 AV |
327 | ret = cgroup_fs->mount(cgroup_fs, flags, |
328 | unused_dev_name, mountopts); | |
8793d854 PM |
329 | put_filesystem(cgroup_fs); |
330 | } | |
331 | return ret; | |
1da177e4 LT |
332 | } |
333 | ||
334 | static struct file_system_type cpuset_fs_type = { | |
335 | .name = "cpuset", | |
f7e83571 | 336 | .mount = cpuset_mount, |
1da177e4 LT |
337 | }; |
338 | ||
1da177e4 | 339 | /* |
300ed6cb | 340 | * Return in pmask the portion of a cpusets's cpus_allowed that |
1da177e4 | 341 | * are online. If none are online, walk up the cpuset hierarchy |
28b89b9e | 342 | * until we find one that does have some online cpus. |
1da177e4 LT |
343 | * |
344 | * One way or another, we guarantee to return some non-empty subset | |
5f054e31 | 345 | * of cpu_online_mask. |
1da177e4 | 346 | * |
8447a0fe | 347 | * Call with callback_lock or cpuset_mutex held. |
1da177e4 | 348 | */ |
c9710d80 | 349 | static void guarantee_online_cpus(struct cpuset *cs, struct cpumask *pmask) |
1da177e4 | 350 | { |
28b89b9e | 351 | while (!cpumask_intersects(cs->effective_cpus, cpu_online_mask)) { |
c431069f | 352 | cs = parent_cs(cs); |
28b89b9e JP |
353 | if (unlikely(!cs)) { |
354 | /* | |
355 | * The top cpuset doesn't have any online cpu as a | |
356 | * consequence of a race between cpuset_hotplug_work | |
357 | * and cpu hotplug notifier. But we know the top | |
358 | * cpuset's effective_cpus is on its way to to be | |
359 | * identical to cpu_online_mask. | |
360 | */ | |
361 | cpumask_copy(pmask, cpu_online_mask); | |
362 | return; | |
363 | } | |
364 | } | |
ae1c8023 | 365 | cpumask_and(pmask, cs->effective_cpus, cpu_online_mask); |
1da177e4 LT |
366 | } |
367 | ||
368 | /* | |
369 | * Return in *pmask the portion of a cpusets's mems_allowed that | |
0e1e7c7a CL |
370 | * are online, with memory. If none are online with memory, walk |
371 | * up the cpuset hierarchy until we find one that does have some | |
40df2deb | 372 | * online mems. The top cpuset always has some mems online. |
1da177e4 LT |
373 | * |
374 | * One way or another, we guarantee to return some non-empty subset | |
38d7bee9 | 375 | * of node_states[N_MEMORY]. |
1da177e4 | 376 | * |
8447a0fe | 377 | * Call with callback_lock or cpuset_mutex held. |
1da177e4 | 378 | */ |
c9710d80 | 379 | static void guarantee_online_mems(struct cpuset *cs, nodemask_t *pmask) |
1da177e4 | 380 | { |
ae1c8023 | 381 | while (!nodes_intersects(cs->effective_mems, node_states[N_MEMORY])) |
c431069f | 382 | cs = parent_cs(cs); |
ae1c8023 | 383 | nodes_and(*pmask, cs->effective_mems, node_states[N_MEMORY]); |
1da177e4 LT |
384 | } |
385 | ||
f3b39d47 MX |
386 | /* |
387 | * update task's spread flag if cpuset's page/slab spread flag is set | |
388 | * | |
8447a0fe | 389 | * Call with callback_lock or cpuset_mutex held. |
f3b39d47 MX |
390 | */ |
391 | static void cpuset_update_task_spread_flag(struct cpuset *cs, | |
392 | struct task_struct *tsk) | |
393 | { | |
394 | if (is_spread_page(cs)) | |
2ad654bc | 395 | task_set_spread_page(tsk); |
f3b39d47 | 396 | else |
2ad654bc ZL |
397 | task_clear_spread_page(tsk); |
398 | ||
f3b39d47 | 399 | if (is_spread_slab(cs)) |
2ad654bc | 400 | task_set_spread_slab(tsk); |
f3b39d47 | 401 | else |
2ad654bc | 402 | task_clear_spread_slab(tsk); |
f3b39d47 MX |
403 | } |
404 | ||
1da177e4 LT |
405 | /* |
406 | * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q? | |
407 | * | |
408 | * One cpuset is a subset of another if all its allowed CPUs and | |
409 | * Memory Nodes are a subset of the other, and its exclusive flags | |
5d21cc2d | 410 | * are only set if the other's are set. Call holding cpuset_mutex. |
1da177e4 LT |
411 | */ |
412 | ||
413 | static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q) | |
414 | { | |
300ed6cb | 415 | return cpumask_subset(p->cpus_allowed, q->cpus_allowed) && |
1da177e4 LT |
416 | nodes_subset(p->mems_allowed, q->mems_allowed) && |
417 | is_cpu_exclusive(p) <= is_cpu_exclusive(q) && | |
418 | is_mem_exclusive(p) <= is_mem_exclusive(q); | |
419 | } | |
420 | ||
645fcc9d LZ |
421 | /** |
422 | * alloc_trial_cpuset - allocate a trial cpuset | |
423 | * @cs: the cpuset that the trial cpuset duplicates | |
424 | */ | |
c9710d80 | 425 | static struct cpuset *alloc_trial_cpuset(struct cpuset *cs) |
645fcc9d | 426 | { |
300ed6cb LZ |
427 | struct cpuset *trial; |
428 | ||
429 | trial = kmemdup(cs, sizeof(*cs), GFP_KERNEL); | |
430 | if (!trial) | |
431 | return NULL; | |
432 | ||
e2b9a3d7 LZ |
433 | if (!alloc_cpumask_var(&trial->cpus_allowed, GFP_KERNEL)) |
434 | goto free_cs; | |
435 | if (!alloc_cpumask_var(&trial->effective_cpus, GFP_KERNEL)) | |
436 | goto free_cpus; | |
300ed6cb | 437 | |
e2b9a3d7 LZ |
438 | cpumask_copy(trial->cpus_allowed, cs->cpus_allowed); |
439 | cpumask_copy(trial->effective_cpus, cs->effective_cpus); | |
300ed6cb | 440 | return trial; |
e2b9a3d7 LZ |
441 | |
442 | free_cpus: | |
443 | free_cpumask_var(trial->cpus_allowed); | |
444 | free_cs: | |
445 | kfree(trial); | |
446 | return NULL; | |
645fcc9d LZ |
447 | } |
448 | ||
449 | /** | |
450 | * free_trial_cpuset - free the trial cpuset | |
451 | * @trial: the trial cpuset to be freed | |
452 | */ | |
453 | static void free_trial_cpuset(struct cpuset *trial) | |
454 | { | |
e2b9a3d7 | 455 | free_cpumask_var(trial->effective_cpus); |
300ed6cb | 456 | free_cpumask_var(trial->cpus_allowed); |
645fcc9d LZ |
457 | kfree(trial); |
458 | } | |
459 | ||
1da177e4 LT |
460 | /* |
461 | * validate_change() - Used to validate that any proposed cpuset change | |
462 | * follows the structural rules for cpusets. | |
463 | * | |
464 | * If we replaced the flag and mask values of the current cpuset | |
465 | * (cur) with those values in the trial cpuset (trial), would | |
466 | * our various subset and exclusive rules still be valid? Presumes | |
5d21cc2d | 467 | * cpuset_mutex held. |
1da177e4 LT |
468 | * |
469 | * 'cur' is the address of an actual, in-use cpuset. Operations | |
470 | * such as list traversal that depend on the actual address of the | |
471 | * cpuset in the list must use cur below, not trial. | |
472 | * | |
473 | * 'trial' is the address of bulk structure copy of cur, with | |
474 | * perhaps one or more of the fields cpus_allowed, mems_allowed, | |
475 | * or flags changed to new, trial values. | |
476 | * | |
477 | * Return 0 if valid, -errno if not. | |
478 | */ | |
479 | ||
c9710d80 | 480 | static int validate_change(struct cpuset *cur, struct cpuset *trial) |
1da177e4 | 481 | { |
492eb21b | 482 | struct cgroup_subsys_state *css; |
1da177e4 | 483 | struct cpuset *c, *par; |
ae8086ce TH |
484 | int ret; |
485 | ||
486 | rcu_read_lock(); | |
1da177e4 LT |
487 | |
488 | /* Each of our child cpusets must be a subset of us */ | |
ae8086ce | 489 | ret = -EBUSY; |
492eb21b | 490 | cpuset_for_each_child(c, css, cur) |
ae8086ce TH |
491 | if (!is_cpuset_subset(c, trial)) |
492 | goto out; | |
1da177e4 LT |
493 | |
494 | /* Remaining checks don't apply to root cpuset */ | |
ae8086ce | 495 | ret = 0; |
69604067 | 496 | if (cur == &top_cpuset) |
ae8086ce | 497 | goto out; |
1da177e4 | 498 | |
c431069f | 499 | par = parent_cs(cur); |
69604067 | 500 | |
7e88291b | 501 | /* On legacy hiearchy, we must be a subset of our parent cpuset. */ |
ae8086ce | 502 | ret = -EACCES; |
b8d1b8ee | 503 | if (!is_in_v2_mode() && !is_cpuset_subset(trial, par)) |
ae8086ce | 504 | goto out; |
1da177e4 | 505 | |
2df167a3 PM |
506 | /* |
507 | * If either I or some sibling (!= me) is exclusive, we can't | |
508 | * overlap | |
509 | */ | |
ae8086ce | 510 | ret = -EINVAL; |
492eb21b | 511 | cpuset_for_each_child(c, css, par) { |
1da177e4 LT |
512 | if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) && |
513 | c != cur && | |
300ed6cb | 514 | cpumask_intersects(trial->cpus_allowed, c->cpus_allowed)) |
ae8086ce | 515 | goto out; |
1da177e4 LT |
516 | if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) && |
517 | c != cur && | |
518 | nodes_intersects(trial->mems_allowed, c->mems_allowed)) | |
ae8086ce | 519 | goto out; |
1da177e4 LT |
520 | } |
521 | ||
452477fa TH |
522 | /* |
523 | * Cpusets with tasks - existing or newly being attached - can't | |
1c09b195 | 524 | * be changed to have empty cpus_allowed or mems_allowed. |
452477fa | 525 | */ |
ae8086ce | 526 | ret = -ENOSPC; |
27bd4dbb | 527 | if ((cgroup_is_populated(cur->css.cgroup) || cur->attach_in_progress)) { |
1c09b195 LZ |
528 | if (!cpumask_empty(cur->cpus_allowed) && |
529 | cpumask_empty(trial->cpus_allowed)) | |
530 | goto out; | |
531 | if (!nodes_empty(cur->mems_allowed) && | |
532 | nodes_empty(trial->mems_allowed)) | |
533 | goto out; | |
534 | } | |
020958b6 | 535 | |
f82f8042 JL |
536 | /* |
537 | * We can't shrink if we won't have enough room for SCHED_DEADLINE | |
538 | * tasks. | |
539 | */ | |
540 | ret = -EBUSY; | |
541 | if (is_cpu_exclusive(cur) && | |
542 | !cpuset_cpumask_can_shrink(cur->cpus_allowed, | |
543 | trial->cpus_allowed)) | |
544 | goto out; | |
545 | ||
ae8086ce TH |
546 | ret = 0; |
547 | out: | |
548 | rcu_read_unlock(); | |
549 | return ret; | |
1da177e4 LT |
550 | } |
551 | ||
db7f47cf | 552 | #ifdef CONFIG_SMP |
029190c5 | 553 | /* |
cf417141 | 554 | * Helper routine for generate_sched_domains(). |
8b5f1c52 | 555 | * Do cpusets a, b have overlapping effective cpus_allowed masks? |
029190c5 | 556 | */ |
029190c5 PJ |
557 | static int cpusets_overlap(struct cpuset *a, struct cpuset *b) |
558 | { | |
8b5f1c52 | 559 | return cpumask_intersects(a->effective_cpus, b->effective_cpus); |
029190c5 PJ |
560 | } |
561 | ||
1d3504fc HS |
562 | static void |
563 | update_domain_attr(struct sched_domain_attr *dattr, struct cpuset *c) | |
564 | { | |
1d3504fc HS |
565 | if (dattr->relax_domain_level < c->relax_domain_level) |
566 | dattr->relax_domain_level = c->relax_domain_level; | |
567 | return; | |
568 | } | |
569 | ||
fc560a26 TH |
570 | static void update_domain_attr_tree(struct sched_domain_attr *dattr, |
571 | struct cpuset *root_cs) | |
f5393693 | 572 | { |
fc560a26 | 573 | struct cpuset *cp; |
492eb21b | 574 | struct cgroup_subsys_state *pos_css; |
f5393693 | 575 | |
fc560a26 | 576 | rcu_read_lock(); |
492eb21b | 577 | cpuset_for_each_descendant_pre(cp, pos_css, root_cs) { |
fc560a26 TH |
578 | /* skip the whole subtree if @cp doesn't have any CPU */ |
579 | if (cpumask_empty(cp->cpus_allowed)) { | |
492eb21b | 580 | pos_css = css_rightmost_descendant(pos_css); |
f5393693 | 581 | continue; |
fc560a26 | 582 | } |
f5393693 LJ |
583 | |
584 | if (is_sched_load_balance(cp)) | |
585 | update_domain_attr(dattr, cp); | |
f5393693 | 586 | } |
fc560a26 | 587 | rcu_read_unlock(); |
f5393693 LJ |
588 | } |
589 | ||
be040bea PB |
590 | /* Must be called with cpuset_mutex held. */ |
591 | static inline int nr_cpusets(void) | |
592 | { | |
593 | /* jump label reference count + the top-level cpuset */ | |
594 | return static_key_count(&cpusets_enabled_key.key) + 1; | |
595 | } | |
596 | ||
029190c5 | 597 | /* |
cf417141 MK |
598 | * generate_sched_domains() |
599 | * | |
600 | * This function builds a partial partition of the systems CPUs | |
601 | * A 'partial partition' is a set of non-overlapping subsets whose | |
602 | * union is a subset of that set. | |
0a0fca9d | 603 | * The output of this function needs to be passed to kernel/sched/core.c |
cf417141 MK |
604 | * partition_sched_domains() routine, which will rebuild the scheduler's |
605 | * load balancing domains (sched domains) as specified by that partial | |
606 | * partition. | |
029190c5 | 607 | * |
45ce80fb | 608 | * See "What is sched_load_balance" in Documentation/cgroups/cpusets.txt |
029190c5 PJ |
609 | * for a background explanation of this. |
610 | * | |
611 | * Does not return errors, on the theory that the callers of this | |
612 | * routine would rather not worry about failures to rebuild sched | |
613 | * domains when operating in the severe memory shortage situations | |
614 | * that could cause allocation failures below. | |
615 | * | |
5d21cc2d | 616 | * Must be called with cpuset_mutex held. |
029190c5 PJ |
617 | * |
618 | * The three key local variables below are: | |
aeed6824 | 619 | * q - a linked-list queue of cpuset pointers, used to implement a |
029190c5 PJ |
620 | * top-down scan of all cpusets. This scan loads a pointer |
621 | * to each cpuset marked is_sched_load_balance into the | |
622 | * array 'csa'. For our purposes, rebuilding the schedulers | |
623 | * sched domains, we can ignore !is_sched_load_balance cpusets. | |
624 | * csa - (for CpuSet Array) Array of pointers to all the cpusets | |
625 | * that need to be load balanced, for convenient iterative | |
626 | * access by the subsequent code that finds the best partition, | |
627 | * i.e the set of domains (subsets) of CPUs such that the | |
628 | * cpus_allowed of every cpuset marked is_sched_load_balance | |
629 | * is a subset of one of these domains, while there are as | |
630 | * many such domains as possible, each as small as possible. | |
631 | * doms - Conversion of 'csa' to an array of cpumasks, for passing to | |
0a0fca9d | 632 | * the kernel/sched/core.c routine partition_sched_domains() in a |
029190c5 PJ |
633 | * convenient format, that can be easily compared to the prior |
634 | * value to determine what partition elements (sched domains) | |
635 | * were changed (added or removed.) | |
636 | * | |
637 | * Finding the best partition (set of domains): | |
638 | * The triple nested loops below over i, j, k scan over the | |
639 | * load balanced cpusets (using the array of cpuset pointers in | |
640 | * csa[]) looking for pairs of cpusets that have overlapping | |
641 | * cpus_allowed, but which don't have the same 'pn' partition | |
642 | * number and gives them in the same partition number. It keeps | |
643 | * looping on the 'restart' label until it can no longer find | |
644 | * any such pairs. | |
645 | * | |
646 | * The union of the cpus_allowed masks from the set of | |
647 | * all cpusets having the same 'pn' value then form the one | |
648 | * element of the partition (one sched domain) to be passed to | |
649 | * partition_sched_domains(). | |
650 | */ | |
acc3f5d7 | 651 | static int generate_sched_domains(cpumask_var_t **domains, |
cf417141 | 652 | struct sched_domain_attr **attributes) |
029190c5 | 653 | { |
029190c5 PJ |
654 | struct cpuset *cp; /* scans q */ |
655 | struct cpuset **csa; /* array of all cpuset ptrs */ | |
656 | int csn; /* how many cpuset ptrs in csa so far */ | |
657 | int i, j, k; /* indices for partition finding loops */ | |
acc3f5d7 | 658 | cpumask_var_t *doms; /* resulting partition; i.e. sched domains */ |
1d3504fc | 659 | struct sched_domain_attr *dattr; /* attributes for custom domains */ |
1583715d | 660 | int ndoms = 0; /* number of sched domains in result */ |
6af866af | 661 | int nslot; /* next empty doms[] struct cpumask slot */ |
492eb21b | 662 | struct cgroup_subsys_state *pos_css; |
029190c5 | 663 | |
029190c5 | 664 | doms = NULL; |
1d3504fc | 665 | dattr = NULL; |
cf417141 | 666 | csa = NULL; |
029190c5 PJ |
667 | |
668 | /* Special case for the 99% of systems with one, full, sched domain */ | |
669 | if (is_sched_load_balance(&top_cpuset)) { | |
acc3f5d7 RR |
670 | ndoms = 1; |
671 | doms = alloc_sched_domains(ndoms); | |
029190c5 | 672 | if (!doms) |
cf417141 MK |
673 | goto done; |
674 | ||
1d3504fc HS |
675 | dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL); |
676 | if (dattr) { | |
677 | *dattr = SD_ATTR_INIT; | |
93a65575 | 678 | update_domain_attr_tree(dattr, &top_cpuset); |
1d3504fc | 679 | } |
47b8ea71 | 680 | cpumask_and(doms[0], top_cpuset.effective_cpus, |
edb93821 | 681 | housekeeping_cpumask(HK_FLAG_DOMAIN)); |
cf417141 | 682 | |
cf417141 | 683 | goto done; |
029190c5 PJ |
684 | } |
685 | ||
6da2ec56 | 686 | csa = kmalloc_array(nr_cpusets(), sizeof(cp), GFP_KERNEL); |
029190c5 PJ |
687 | if (!csa) |
688 | goto done; | |
689 | csn = 0; | |
690 | ||
fc560a26 | 691 | rcu_read_lock(); |
492eb21b | 692 | cpuset_for_each_descendant_pre(cp, pos_css, &top_cpuset) { |
bd8815a6 TH |
693 | if (cp == &top_cpuset) |
694 | continue; | |
f5393693 | 695 | /* |
fc560a26 TH |
696 | * Continue traversing beyond @cp iff @cp has some CPUs and |
697 | * isn't load balancing. The former is obvious. The | |
698 | * latter: All child cpusets contain a subset of the | |
699 | * parent's cpus, so just skip them, and then we call | |
700 | * update_domain_attr_tree() to calc relax_domain_level of | |
701 | * the corresponding sched domain. | |
f5393693 | 702 | */ |
fc560a26 | 703 | if (!cpumask_empty(cp->cpus_allowed) && |
47b8ea71 | 704 | !(is_sched_load_balance(cp) && |
edb93821 FW |
705 | cpumask_intersects(cp->cpus_allowed, |
706 | housekeeping_cpumask(HK_FLAG_DOMAIN)))) | |
f5393693 | 707 | continue; |
489a5393 | 708 | |
fc560a26 TH |
709 | if (is_sched_load_balance(cp)) |
710 | csa[csn++] = cp; | |
711 | ||
712 | /* skip @cp's subtree */ | |
492eb21b | 713 | pos_css = css_rightmost_descendant(pos_css); |
fc560a26 TH |
714 | } |
715 | rcu_read_unlock(); | |
029190c5 PJ |
716 | |
717 | for (i = 0; i < csn; i++) | |
718 | csa[i]->pn = i; | |
719 | ndoms = csn; | |
720 | ||
721 | restart: | |
722 | /* Find the best partition (set of sched domains) */ | |
723 | for (i = 0; i < csn; i++) { | |
724 | struct cpuset *a = csa[i]; | |
725 | int apn = a->pn; | |
726 | ||
727 | for (j = 0; j < csn; j++) { | |
728 | struct cpuset *b = csa[j]; | |
729 | int bpn = b->pn; | |
730 | ||
731 | if (apn != bpn && cpusets_overlap(a, b)) { | |
732 | for (k = 0; k < csn; k++) { | |
733 | struct cpuset *c = csa[k]; | |
734 | ||
735 | if (c->pn == bpn) | |
736 | c->pn = apn; | |
737 | } | |
738 | ndoms--; /* one less element */ | |
739 | goto restart; | |
740 | } | |
741 | } | |
742 | } | |
743 | ||
cf417141 MK |
744 | /* |
745 | * Now we know how many domains to create. | |
746 | * Convert <csn, csa> to <ndoms, doms> and populate cpu masks. | |
747 | */ | |
acc3f5d7 | 748 | doms = alloc_sched_domains(ndoms); |
700018e0 | 749 | if (!doms) |
cf417141 | 750 | goto done; |
cf417141 MK |
751 | |
752 | /* | |
753 | * The rest of the code, including the scheduler, can deal with | |
754 | * dattr==NULL case. No need to abort if alloc fails. | |
755 | */ | |
6da2ec56 KC |
756 | dattr = kmalloc_array(ndoms, sizeof(struct sched_domain_attr), |
757 | GFP_KERNEL); | |
029190c5 PJ |
758 | |
759 | for (nslot = 0, i = 0; i < csn; i++) { | |
760 | struct cpuset *a = csa[i]; | |
6af866af | 761 | struct cpumask *dp; |
029190c5 PJ |
762 | int apn = a->pn; |
763 | ||
cf417141 MK |
764 | if (apn < 0) { |
765 | /* Skip completed partitions */ | |
766 | continue; | |
767 | } | |
768 | ||
acc3f5d7 | 769 | dp = doms[nslot]; |
cf417141 MK |
770 | |
771 | if (nslot == ndoms) { | |
772 | static int warnings = 10; | |
773 | if (warnings) { | |
12d3089c FF |
774 | pr_warn("rebuild_sched_domains confused: nslot %d, ndoms %d, csn %d, i %d, apn %d\n", |
775 | nslot, ndoms, csn, i, apn); | |
cf417141 | 776 | warnings--; |
029190c5 | 777 | } |
cf417141 MK |
778 | continue; |
779 | } | |
029190c5 | 780 | |
6af866af | 781 | cpumask_clear(dp); |
cf417141 MK |
782 | if (dattr) |
783 | *(dattr + nslot) = SD_ATTR_INIT; | |
784 | for (j = i; j < csn; j++) { | |
785 | struct cpuset *b = csa[j]; | |
786 | ||
787 | if (apn == b->pn) { | |
8b5f1c52 | 788 | cpumask_or(dp, dp, b->effective_cpus); |
edb93821 | 789 | cpumask_and(dp, dp, housekeeping_cpumask(HK_FLAG_DOMAIN)); |
cf417141 MK |
790 | if (dattr) |
791 | update_domain_attr_tree(dattr + nslot, b); | |
792 | ||
793 | /* Done with this partition */ | |
794 | b->pn = -1; | |
029190c5 | 795 | } |
029190c5 | 796 | } |
cf417141 | 797 | nslot++; |
029190c5 PJ |
798 | } |
799 | BUG_ON(nslot != ndoms); | |
800 | ||
cf417141 MK |
801 | done: |
802 | kfree(csa); | |
803 | ||
700018e0 LZ |
804 | /* |
805 | * Fallback to the default domain if kmalloc() failed. | |
806 | * See comments in partition_sched_domains(). | |
807 | */ | |
808 | if (doms == NULL) | |
809 | ndoms = 1; | |
810 | ||
cf417141 MK |
811 | *domains = doms; |
812 | *attributes = dattr; | |
813 | return ndoms; | |
814 | } | |
815 | ||
816 | /* | |
817 | * Rebuild scheduler domains. | |
818 | * | |
699140ba TH |
819 | * If the flag 'sched_load_balance' of any cpuset with non-empty |
820 | * 'cpus' changes, or if the 'cpus' allowed changes in any cpuset | |
821 | * which has that flag enabled, or if any cpuset with a non-empty | |
822 | * 'cpus' is removed, then call this routine to rebuild the | |
823 | * scheduler's dynamic sched domains. | |
cf417141 | 824 | * |
5d21cc2d | 825 | * Call with cpuset_mutex held. Takes get_online_cpus(). |
cf417141 | 826 | */ |
699140ba | 827 | static void rebuild_sched_domains_locked(void) |
cf417141 MK |
828 | { |
829 | struct sched_domain_attr *attr; | |
acc3f5d7 | 830 | cpumask_var_t *doms; |
cf417141 MK |
831 | int ndoms; |
832 | ||
5d21cc2d | 833 | lockdep_assert_held(&cpuset_mutex); |
86ef5c9a | 834 | get_online_cpus(); |
cf417141 | 835 | |
5b16c2a4 LZ |
836 | /* |
837 | * We have raced with CPU hotplug. Don't do anything to avoid | |
838 | * passing doms with offlined cpu to partition_sched_domains(). | |
839 | * Anyways, hotplug work item will rebuild sched domains. | |
840 | */ | |
8b5f1c52 | 841 | if (!cpumask_equal(top_cpuset.effective_cpus, cpu_active_mask)) |
5b16c2a4 LZ |
842 | goto out; |
843 | ||
cf417141 | 844 | /* Generate domain masks and attrs */ |
cf417141 | 845 | ndoms = generate_sched_domains(&doms, &attr); |
cf417141 MK |
846 | |
847 | /* Have scheduler rebuild the domains */ | |
848 | partition_sched_domains(ndoms, doms, attr); | |
5b16c2a4 | 849 | out: |
86ef5c9a | 850 | put_online_cpus(); |
cf417141 | 851 | } |
db7f47cf | 852 | #else /* !CONFIG_SMP */ |
699140ba | 853 | static void rebuild_sched_domains_locked(void) |
db7f47cf PM |
854 | { |
855 | } | |
db7f47cf | 856 | #endif /* CONFIG_SMP */ |
029190c5 | 857 | |
cf417141 MK |
858 | void rebuild_sched_domains(void) |
859 | { | |
5d21cc2d | 860 | mutex_lock(&cpuset_mutex); |
699140ba | 861 | rebuild_sched_domains_locked(); |
5d21cc2d | 862 | mutex_unlock(&cpuset_mutex); |
029190c5 PJ |
863 | } |
864 | ||
0b2f630a MX |
865 | /** |
866 | * update_tasks_cpumask - Update the cpumasks of tasks in the cpuset. | |
867 | * @cs: the cpuset in which each task's cpus_allowed mask needs to be changed | |
0b2f630a | 868 | * |
d66393e5 TH |
869 | * Iterate through each task of @cs updating its cpus_allowed to the |
870 | * effective cpuset's. As this function is called with cpuset_mutex held, | |
871 | * cpuset membership stays stable. | |
0b2f630a | 872 | */ |
d66393e5 | 873 | static void update_tasks_cpumask(struct cpuset *cs) |
0b2f630a | 874 | { |
d66393e5 TH |
875 | struct css_task_iter it; |
876 | struct task_struct *task; | |
877 | ||
bc2fb7ed | 878 | css_task_iter_start(&cs->css, 0, &it); |
d66393e5 | 879 | while ((task = css_task_iter_next(&it))) |
ae1c8023 | 880 | set_cpus_allowed_ptr(task, cs->effective_cpus); |
d66393e5 | 881 | css_task_iter_end(&it); |
0b2f630a MX |
882 | } |
883 | ||
5c5cc623 | 884 | /* |
734d4513 LZ |
885 | * update_cpumasks_hier - Update effective cpumasks and tasks in the subtree |
886 | * @cs: the cpuset to consider | |
887 | * @new_cpus: temp variable for calculating new effective_cpus | |
888 | * | |
889 | * When congifured cpumask is changed, the effective cpumasks of this cpuset | |
890 | * and all its descendants need to be updated. | |
5c5cc623 | 891 | * |
734d4513 | 892 | * On legacy hierachy, effective_cpus will be the same with cpu_allowed. |
5c5cc623 LZ |
893 | * |
894 | * Called with cpuset_mutex held | |
895 | */ | |
734d4513 | 896 | static void update_cpumasks_hier(struct cpuset *cs, struct cpumask *new_cpus) |
5c5cc623 LZ |
897 | { |
898 | struct cpuset *cp; | |
492eb21b | 899 | struct cgroup_subsys_state *pos_css; |
8b5f1c52 | 900 | bool need_rebuild_sched_domains = false; |
5c5cc623 LZ |
901 | |
902 | rcu_read_lock(); | |
734d4513 LZ |
903 | cpuset_for_each_descendant_pre(cp, pos_css, cs) { |
904 | struct cpuset *parent = parent_cs(cp); | |
905 | ||
906 | cpumask_and(new_cpus, cp->cpus_allowed, parent->effective_cpus); | |
907 | ||
554b0d1c LZ |
908 | /* |
909 | * If it becomes empty, inherit the effective mask of the | |
910 | * parent, which is guaranteed to have some CPUs. | |
911 | */ | |
b8d1b8ee | 912 | if (is_in_v2_mode() && cpumask_empty(new_cpus)) |
554b0d1c LZ |
913 | cpumask_copy(new_cpus, parent->effective_cpus); |
914 | ||
734d4513 LZ |
915 | /* Skip the whole subtree if the cpumask remains the same. */ |
916 | if (cpumask_equal(new_cpus, cp->effective_cpus)) { | |
917 | pos_css = css_rightmost_descendant(pos_css); | |
918 | continue; | |
5c5cc623 | 919 | } |
734d4513 | 920 | |
ec903c0c | 921 | if (!css_tryget_online(&cp->css)) |
5c5cc623 LZ |
922 | continue; |
923 | rcu_read_unlock(); | |
924 | ||
8447a0fe | 925 | spin_lock_irq(&callback_lock); |
734d4513 | 926 | cpumask_copy(cp->effective_cpus, new_cpus); |
8447a0fe | 927 | spin_unlock_irq(&callback_lock); |
734d4513 | 928 | |
b8d1b8ee | 929 | WARN_ON(!is_in_v2_mode() && |
734d4513 LZ |
930 | !cpumask_equal(cp->cpus_allowed, cp->effective_cpus)); |
931 | ||
d66393e5 | 932 | update_tasks_cpumask(cp); |
5c5cc623 | 933 | |
8b5f1c52 LZ |
934 | /* |
935 | * If the effective cpumask of any non-empty cpuset is changed, | |
936 | * we need to rebuild sched domains. | |
937 | */ | |
938 | if (!cpumask_empty(cp->cpus_allowed) && | |
939 | is_sched_load_balance(cp)) | |
940 | need_rebuild_sched_domains = true; | |
941 | ||
5c5cc623 LZ |
942 | rcu_read_lock(); |
943 | css_put(&cp->css); | |
944 | } | |
945 | rcu_read_unlock(); | |
8b5f1c52 LZ |
946 | |
947 | if (need_rebuild_sched_domains) | |
948 | rebuild_sched_domains_locked(); | |
5c5cc623 LZ |
949 | } |
950 | ||
58f4790b CW |
951 | /** |
952 | * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it | |
953 | * @cs: the cpuset to consider | |
fc34ac1d | 954 | * @trialcs: trial cpuset |
58f4790b CW |
955 | * @buf: buffer of cpu numbers written to this cpuset |
956 | */ | |
645fcc9d LZ |
957 | static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs, |
958 | const char *buf) | |
1da177e4 | 959 | { |
58f4790b | 960 | int retval; |
1da177e4 | 961 | |
5f054e31 | 962 | /* top_cpuset.cpus_allowed tracks cpu_online_mask; it's read-only */ |
4c4d50f7 PJ |
963 | if (cs == &top_cpuset) |
964 | return -EACCES; | |
965 | ||
6f7f02e7 | 966 | /* |
c8d9c90c | 967 | * An empty cpus_allowed is ok only if the cpuset has no tasks. |
020958b6 PJ |
968 | * Since cpulist_parse() fails on an empty mask, we special case |
969 | * that parsing. The validate_change() call ensures that cpusets | |
970 | * with tasks have cpus. | |
6f7f02e7 | 971 | */ |
020958b6 | 972 | if (!*buf) { |
300ed6cb | 973 | cpumask_clear(trialcs->cpus_allowed); |
6f7f02e7 | 974 | } else { |
300ed6cb | 975 | retval = cpulist_parse(buf, trialcs->cpus_allowed); |
6f7f02e7 DR |
976 | if (retval < 0) |
977 | return retval; | |
37340746 | 978 | |
5d8ba82c LZ |
979 | if (!cpumask_subset(trialcs->cpus_allowed, |
980 | top_cpuset.cpus_allowed)) | |
37340746 | 981 | return -EINVAL; |
6f7f02e7 | 982 | } |
029190c5 | 983 | |
8707d8b8 | 984 | /* Nothing to do if the cpus didn't change */ |
300ed6cb | 985 | if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed)) |
8707d8b8 | 986 | return 0; |
58f4790b | 987 | |
a73456f3 LZ |
988 | retval = validate_change(cs, trialcs); |
989 | if (retval < 0) | |
990 | return retval; | |
991 | ||
8447a0fe | 992 | spin_lock_irq(&callback_lock); |
300ed6cb | 993 | cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed); |
8447a0fe | 994 | spin_unlock_irq(&callback_lock); |
029190c5 | 995 | |
734d4513 LZ |
996 | /* use trialcs->cpus_allowed as a temp variable */ |
997 | update_cpumasks_hier(cs, trialcs->cpus_allowed); | |
85d7b949 | 998 | return 0; |
1da177e4 LT |
999 | } |
1000 | ||
e4e364e8 | 1001 | /* |
e93ad19d TH |
1002 | * Migrate memory region from one set of nodes to another. This is |
1003 | * performed asynchronously as it can be called from process migration path | |
1004 | * holding locks involved in process management. All mm migrations are | |
1005 | * performed in the queued order and can be waited for by flushing | |
1006 | * cpuset_migrate_mm_wq. | |
e4e364e8 PJ |
1007 | */ |
1008 | ||
e93ad19d TH |
1009 | struct cpuset_migrate_mm_work { |
1010 | struct work_struct work; | |
1011 | struct mm_struct *mm; | |
1012 | nodemask_t from; | |
1013 | nodemask_t to; | |
1014 | }; | |
1015 | ||
1016 | static void cpuset_migrate_mm_workfn(struct work_struct *work) | |
1017 | { | |
1018 | struct cpuset_migrate_mm_work *mwork = | |
1019 | container_of(work, struct cpuset_migrate_mm_work, work); | |
1020 | ||
1021 | /* on a wq worker, no need to worry about %current's mems_allowed */ | |
1022 | do_migrate_pages(mwork->mm, &mwork->from, &mwork->to, MPOL_MF_MOVE_ALL); | |
1023 | mmput(mwork->mm); | |
1024 | kfree(mwork); | |
1025 | } | |
1026 | ||
e4e364e8 PJ |
1027 | static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from, |
1028 | const nodemask_t *to) | |
1029 | { | |
e93ad19d | 1030 | struct cpuset_migrate_mm_work *mwork; |
e4e364e8 | 1031 | |
e93ad19d TH |
1032 | mwork = kzalloc(sizeof(*mwork), GFP_KERNEL); |
1033 | if (mwork) { | |
1034 | mwork->mm = mm; | |
1035 | mwork->from = *from; | |
1036 | mwork->to = *to; | |
1037 | INIT_WORK(&mwork->work, cpuset_migrate_mm_workfn); | |
1038 | queue_work(cpuset_migrate_mm_wq, &mwork->work); | |
1039 | } else { | |
1040 | mmput(mm); | |
1041 | } | |
1042 | } | |
e4e364e8 | 1043 | |
5cf1cacb | 1044 | static void cpuset_post_attach(void) |
e93ad19d TH |
1045 | { |
1046 | flush_workqueue(cpuset_migrate_mm_wq); | |
e4e364e8 PJ |
1047 | } |
1048 | ||
3b6766fe | 1049 | /* |
58568d2a MX |
1050 | * cpuset_change_task_nodemask - change task's mems_allowed and mempolicy |
1051 | * @tsk: the task to change | |
1052 | * @newmems: new nodes that the task will be set | |
1053 | * | |
5f155f27 VB |
1054 | * We use the mems_allowed_seq seqlock to safely update both tsk->mems_allowed |
1055 | * and rebind an eventual tasks' mempolicy. If the task is allocating in | |
1056 | * parallel, it might temporarily see an empty intersection, which results in | |
1057 | * a seqlock check and retry before OOM or allocation failure. | |
58568d2a MX |
1058 | */ |
1059 | static void cpuset_change_task_nodemask(struct task_struct *tsk, | |
1060 | nodemask_t *newmems) | |
1061 | { | |
c0ff7453 | 1062 | task_lock(tsk); |
c0ff7453 | 1063 | |
5f155f27 VB |
1064 | local_irq_disable(); |
1065 | write_seqcount_begin(&tsk->mems_allowed_seq); | |
c0ff7453 | 1066 | |
cc9a6c87 | 1067 | nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems); |
213980c0 | 1068 | mpol_rebind_task(tsk, newmems); |
58568d2a | 1069 | tsk->mems_allowed = *newmems; |
cc9a6c87 | 1070 | |
5f155f27 VB |
1071 | write_seqcount_end(&tsk->mems_allowed_seq); |
1072 | local_irq_enable(); | |
cc9a6c87 | 1073 | |
c0ff7453 | 1074 | task_unlock(tsk); |
58568d2a MX |
1075 | } |
1076 | ||
8793d854 PM |
1077 | static void *cpuset_being_rebound; |
1078 | ||
0b2f630a MX |
1079 | /** |
1080 | * update_tasks_nodemask - Update the nodemasks of tasks in the cpuset. | |
1081 | * @cs: the cpuset in which each task's mems_allowed mask needs to be changed | |
0b2f630a | 1082 | * |
d66393e5 TH |
1083 | * Iterate through each task of @cs updating its mems_allowed to the |
1084 | * effective cpuset's. As this function is called with cpuset_mutex held, | |
1085 | * cpuset membership stays stable. | |
0b2f630a | 1086 | */ |
d66393e5 | 1087 | static void update_tasks_nodemask(struct cpuset *cs) |
1da177e4 | 1088 | { |
33ad801d | 1089 | static nodemask_t newmems; /* protected by cpuset_mutex */ |
d66393e5 TH |
1090 | struct css_task_iter it; |
1091 | struct task_struct *task; | |
59dac16f | 1092 | |
846a16bf | 1093 | cpuset_being_rebound = cs; /* causes mpol_dup() rebind */ |
4225399a | 1094 | |
ae1c8023 | 1095 | guarantee_online_mems(cs, &newmems); |
33ad801d | 1096 | |
4225399a | 1097 | /* |
3b6766fe LZ |
1098 | * The mpol_rebind_mm() call takes mmap_sem, which we couldn't |
1099 | * take while holding tasklist_lock. Forks can happen - the | |
1100 | * mpol_dup() cpuset_being_rebound check will catch such forks, | |
1101 | * and rebind their vma mempolicies too. Because we still hold | |
5d21cc2d | 1102 | * the global cpuset_mutex, we know that no other rebind effort |
3b6766fe | 1103 | * will be contending for the global variable cpuset_being_rebound. |
4225399a | 1104 | * It's ok if we rebind the same mm twice; mpol_rebind_mm() |
04c19fa6 | 1105 | * is idempotent. Also migrate pages in each mm to new nodes. |
4225399a | 1106 | */ |
bc2fb7ed | 1107 | css_task_iter_start(&cs->css, 0, &it); |
d66393e5 TH |
1108 | while ((task = css_task_iter_next(&it))) { |
1109 | struct mm_struct *mm; | |
1110 | bool migrate; | |
1111 | ||
1112 | cpuset_change_task_nodemask(task, &newmems); | |
1113 | ||
1114 | mm = get_task_mm(task); | |
1115 | if (!mm) | |
1116 | continue; | |
1117 | ||
1118 | migrate = is_memory_migrate(cs); | |
1119 | ||
1120 | mpol_rebind_mm(mm, &cs->mems_allowed); | |
1121 | if (migrate) | |
1122 | cpuset_migrate_mm(mm, &cs->old_mems_allowed, &newmems); | |
e93ad19d TH |
1123 | else |
1124 | mmput(mm); | |
d66393e5 TH |
1125 | } |
1126 | css_task_iter_end(&it); | |
4225399a | 1127 | |
33ad801d LZ |
1128 | /* |
1129 | * All the tasks' nodemasks have been updated, update | |
1130 | * cs->old_mems_allowed. | |
1131 | */ | |
1132 | cs->old_mems_allowed = newmems; | |
1133 | ||
2df167a3 | 1134 | /* We're done rebinding vmas to this cpuset's new mems_allowed. */ |
8793d854 | 1135 | cpuset_being_rebound = NULL; |
1da177e4 LT |
1136 | } |
1137 | ||
5c5cc623 | 1138 | /* |
734d4513 LZ |
1139 | * update_nodemasks_hier - Update effective nodemasks and tasks in the subtree |
1140 | * @cs: the cpuset to consider | |
1141 | * @new_mems: a temp variable for calculating new effective_mems | |
5c5cc623 | 1142 | * |
734d4513 LZ |
1143 | * When configured nodemask is changed, the effective nodemasks of this cpuset |
1144 | * and all its descendants need to be updated. | |
5c5cc623 | 1145 | * |
734d4513 | 1146 | * On legacy hiearchy, effective_mems will be the same with mems_allowed. |
5c5cc623 LZ |
1147 | * |
1148 | * Called with cpuset_mutex held | |
1149 | */ | |
734d4513 | 1150 | static void update_nodemasks_hier(struct cpuset *cs, nodemask_t *new_mems) |
5c5cc623 LZ |
1151 | { |
1152 | struct cpuset *cp; | |
492eb21b | 1153 | struct cgroup_subsys_state *pos_css; |
5c5cc623 LZ |
1154 | |
1155 | rcu_read_lock(); | |
734d4513 LZ |
1156 | cpuset_for_each_descendant_pre(cp, pos_css, cs) { |
1157 | struct cpuset *parent = parent_cs(cp); | |
1158 | ||
1159 | nodes_and(*new_mems, cp->mems_allowed, parent->effective_mems); | |
1160 | ||
554b0d1c LZ |
1161 | /* |
1162 | * If it becomes empty, inherit the effective mask of the | |
1163 | * parent, which is guaranteed to have some MEMs. | |
1164 | */ | |
b8d1b8ee | 1165 | if (is_in_v2_mode() && nodes_empty(*new_mems)) |
554b0d1c LZ |
1166 | *new_mems = parent->effective_mems; |
1167 | ||
734d4513 LZ |
1168 | /* Skip the whole subtree if the nodemask remains the same. */ |
1169 | if (nodes_equal(*new_mems, cp->effective_mems)) { | |
1170 | pos_css = css_rightmost_descendant(pos_css); | |
1171 | continue; | |
5c5cc623 | 1172 | } |
734d4513 | 1173 | |
ec903c0c | 1174 | if (!css_tryget_online(&cp->css)) |
5c5cc623 LZ |
1175 | continue; |
1176 | rcu_read_unlock(); | |
1177 | ||
8447a0fe | 1178 | spin_lock_irq(&callback_lock); |
734d4513 | 1179 | cp->effective_mems = *new_mems; |
8447a0fe | 1180 | spin_unlock_irq(&callback_lock); |
734d4513 | 1181 | |
b8d1b8ee | 1182 | WARN_ON(!is_in_v2_mode() && |
a1381268 | 1183 | !nodes_equal(cp->mems_allowed, cp->effective_mems)); |
734d4513 | 1184 | |
d66393e5 | 1185 | update_tasks_nodemask(cp); |
5c5cc623 LZ |
1186 | |
1187 | rcu_read_lock(); | |
1188 | css_put(&cp->css); | |
1189 | } | |
1190 | rcu_read_unlock(); | |
1191 | } | |
1192 | ||
0b2f630a MX |
1193 | /* |
1194 | * Handle user request to change the 'mems' memory placement | |
1195 | * of a cpuset. Needs to validate the request, update the | |
58568d2a MX |
1196 | * cpusets mems_allowed, and for each task in the cpuset, |
1197 | * update mems_allowed and rebind task's mempolicy and any vma | |
1198 | * mempolicies and if the cpuset is marked 'memory_migrate', | |
1199 | * migrate the tasks pages to the new memory. | |
0b2f630a | 1200 | * |
8447a0fe | 1201 | * Call with cpuset_mutex held. May take callback_lock during call. |
0b2f630a MX |
1202 | * Will take tasklist_lock, scan tasklist for tasks in cpuset cs, |
1203 | * lock each such tasks mm->mmap_sem, scan its vma's and rebind | |
1204 | * their mempolicies to the cpusets new mems_allowed. | |
1205 | */ | |
645fcc9d LZ |
1206 | static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs, |
1207 | const char *buf) | |
0b2f630a | 1208 | { |
0b2f630a MX |
1209 | int retval; |
1210 | ||
1211 | /* | |
38d7bee9 | 1212 | * top_cpuset.mems_allowed tracks node_stats[N_MEMORY]; |
0b2f630a MX |
1213 | * it's read-only |
1214 | */ | |
53feb297 MX |
1215 | if (cs == &top_cpuset) { |
1216 | retval = -EACCES; | |
1217 | goto done; | |
1218 | } | |
0b2f630a | 1219 | |
0b2f630a MX |
1220 | /* |
1221 | * An empty mems_allowed is ok iff there are no tasks in the cpuset. | |
1222 | * Since nodelist_parse() fails on an empty mask, we special case | |
1223 | * that parsing. The validate_change() call ensures that cpusets | |
1224 | * with tasks have memory. | |
1225 | */ | |
1226 | if (!*buf) { | |
645fcc9d | 1227 | nodes_clear(trialcs->mems_allowed); |
0b2f630a | 1228 | } else { |
645fcc9d | 1229 | retval = nodelist_parse(buf, trialcs->mems_allowed); |
0b2f630a MX |
1230 | if (retval < 0) |
1231 | goto done; | |
1232 | ||
645fcc9d | 1233 | if (!nodes_subset(trialcs->mems_allowed, |
5d8ba82c LZ |
1234 | top_cpuset.mems_allowed)) { |
1235 | retval = -EINVAL; | |
53feb297 MX |
1236 | goto done; |
1237 | } | |
0b2f630a | 1238 | } |
33ad801d LZ |
1239 | |
1240 | if (nodes_equal(cs->mems_allowed, trialcs->mems_allowed)) { | |
0b2f630a MX |
1241 | retval = 0; /* Too easy - nothing to do */ |
1242 | goto done; | |
1243 | } | |
645fcc9d | 1244 | retval = validate_change(cs, trialcs); |
0b2f630a MX |
1245 | if (retval < 0) |
1246 | goto done; | |
1247 | ||
8447a0fe | 1248 | spin_lock_irq(&callback_lock); |
645fcc9d | 1249 | cs->mems_allowed = trialcs->mems_allowed; |
8447a0fe | 1250 | spin_unlock_irq(&callback_lock); |
0b2f630a | 1251 | |
734d4513 | 1252 | /* use trialcs->mems_allowed as a temp variable */ |
24ee3cf8 | 1253 | update_nodemasks_hier(cs, &trialcs->mems_allowed); |
0b2f630a MX |
1254 | done: |
1255 | return retval; | |
1256 | } | |
1257 | ||
77ef80c6 | 1258 | bool current_cpuset_is_being_rebound(void) |
8793d854 | 1259 | { |
77ef80c6 | 1260 | bool ret; |
391acf97 GZ |
1261 | |
1262 | rcu_read_lock(); | |
1263 | ret = task_cs(current) == cpuset_being_rebound; | |
1264 | rcu_read_unlock(); | |
1265 | ||
1266 | return ret; | |
8793d854 PM |
1267 | } |
1268 | ||
5be7a479 | 1269 | static int update_relax_domain_level(struct cpuset *cs, s64 val) |
1d3504fc | 1270 | { |
db7f47cf | 1271 | #ifdef CONFIG_SMP |
60495e77 | 1272 | if (val < -1 || val >= sched_domain_level_max) |
30e0e178 | 1273 | return -EINVAL; |
db7f47cf | 1274 | #endif |
1d3504fc HS |
1275 | |
1276 | if (val != cs->relax_domain_level) { | |
1277 | cs->relax_domain_level = val; | |
300ed6cb LZ |
1278 | if (!cpumask_empty(cs->cpus_allowed) && |
1279 | is_sched_load_balance(cs)) | |
699140ba | 1280 | rebuild_sched_domains_locked(); |
1d3504fc HS |
1281 | } |
1282 | ||
1283 | return 0; | |
1284 | } | |
1285 | ||
72ec7029 | 1286 | /** |
950592f7 MX |
1287 | * update_tasks_flags - update the spread flags of tasks in the cpuset. |
1288 | * @cs: the cpuset in which each task's spread flags needs to be changed | |
950592f7 | 1289 | * |
d66393e5 TH |
1290 | * Iterate through each task of @cs updating its spread flags. As this |
1291 | * function is called with cpuset_mutex held, cpuset membership stays | |
1292 | * stable. | |
950592f7 | 1293 | */ |
d66393e5 | 1294 | static void update_tasks_flags(struct cpuset *cs) |
950592f7 | 1295 | { |
d66393e5 TH |
1296 | struct css_task_iter it; |
1297 | struct task_struct *task; | |
1298 | ||
bc2fb7ed | 1299 | css_task_iter_start(&cs->css, 0, &it); |
d66393e5 TH |
1300 | while ((task = css_task_iter_next(&it))) |
1301 | cpuset_update_task_spread_flag(cs, task); | |
1302 | css_task_iter_end(&it); | |
950592f7 MX |
1303 | } |
1304 | ||
1da177e4 LT |
1305 | /* |
1306 | * update_flag - read a 0 or a 1 in a file and update associated flag | |
78608366 PM |
1307 | * bit: the bit to update (see cpuset_flagbits_t) |
1308 | * cs: the cpuset to update | |
1309 | * turning_on: whether the flag is being set or cleared | |
053199ed | 1310 | * |
5d21cc2d | 1311 | * Call with cpuset_mutex held. |
1da177e4 LT |
1312 | */ |
1313 | ||
700fe1ab PM |
1314 | static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs, |
1315 | int turning_on) | |
1da177e4 | 1316 | { |
645fcc9d | 1317 | struct cpuset *trialcs; |
40b6a762 | 1318 | int balance_flag_changed; |
950592f7 | 1319 | int spread_flag_changed; |
950592f7 | 1320 | int err; |
1da177e4 | 1321 | |
645fcc9d LZ |
1322 | trialcs = alloc_trial_cpuset(cs); |
1323 | if (!trialcs) | |
1324 | return -ENOMEM; | |
1325 | ||
1da177e4 | 1326 | if (turning_on) |
645fcc9d | 1327 | set_bit(bit, &trialcs->flags); |
1da177e4 | 1328 | else |
645fcc9d | 1329 | clear_bit(bit, &trialcs->flags); |
1da177e4 | 1330 | |
645fcc9d | 1331 | err = validate_change(cs, trialcs); |
85d7b949 | 1332 | if (err < 0) |
645fcc9d | 1333 | goto out; |
029190c5 | 1334 | |
029190c5 | 1335 | balance_flag_changed = (is_sched_load_balance(cs) != |
645fcc9d | 1336 | is_sched_load_balance(trialcs)); |
029190c5 | 1337 | |
950592f7 MX |
1338 | spread_flag_changed = ((is_spread_slab(cs) != is_spread_slab(trialcs)) |
1339 | || (is_spread_page(cs) != is_spread_page(trialcs))); | |
1340 | ||
8447a0fe | 1341 | spin_lock_irq(&callback_lock); |
645fcc9d | 1342 | cs->flags = trialcs->flags; |
8447a0fe | 1343 | spin_unlock_irq(&callback_lock); |
85d7b949 | 1344 | |
300ed6cb | 1345 | if (!cpumask_empty(trialcs->cpus_allowed) && balance_flag_changed) |
699140ba | 1346 | rebuild_sched_domains_locked(); |
029190c5 | 1347 | |
950592f7 | 1348 | if (spread_flag_changed) |
d66393e5 | 1349 | update_tasks_flags(cs); |
645fcc9d LZ |
1350 | out: |
1351 | free_trial_cpuset(trialcs); | |
1352 | return err; | |
1da177e4 LT |
1353 | } |
1354 | ||
3e0d98b9 | 1355 | /* |
80f7228b | 1356 | * Frequency meter - How fast is some event occurring? |
3e0d98b9 PJ |
1357 | * |
1358 | * These routines manage a digitally filtered, constant time based, | |
1359 | * event frequency meter. There are four routines: | |
1360 | * fmeter_init() - initialize a frequency meter. | |
1361 | * fmeter_markevent() - called each time the event happens. | |
1362 | * fmeter_getrate() - returns the recent rate of such events. | |
1363 | * fmeter_update() - internal routine used to update fmeter. | |
1364 | * | |
1365 | * A common data structure is passed to each of these routines, | |
1366 | * which is used to keep track of the state required to manage the | |
1367 | * frequency meter and its digital filter. | |
1368 | * | |
1369 | * The filter works on the number of events marked per unit time. | |
1370 | * The filter is single-pole low-pass recursive (IIR). The time unit | |
1371 | * is 1 second. Arithmetic is done using 32-bit integers scaled to | |
1372 | * simulate 3 decimal digits of precision (multiplied by 1000). | |
1373 | * | |
1374 | * With an FM_COEF of 933, and a time base of 1 second, the filter | |
1375 | * has a half-life of 10 seconds, meaning that if the events quit | |
1376 | * happening, then the rate returned from the fmeter_getrate() | |
1377 | * will be cut in half each 10 seconds, until it converges to zero. | |
1378 | * | |
1379 | * It is not worth doing a real infinitely recursive filter. If more | |
1380 | * than FM_MAXTICKS ticks have elapsed since the last filter event, | |
1381 | * just compute FM_MAXTICKS ticks worth, by which point the level | |
1382 | * will be stable. | |
1383 | * | |
1384 | * Limit the count of unprocessed events to FM_MAXCNT, so as to avoid | |
1385 | * arithmetic overflow in the fmeter_update() routine. | |
1386 | * | |
1387 | * Given the simple 32 bit integer arithmetic used, this meter works | |
1388 | * best for reporting rates between one per millisecond (msec) and | |
1389 | * one per 32 (approx) seconds. At constant rates faster than one | |
1390 | * per msec it maxes out at values just under 1,000,000. At constant | |
1391 | * rates between one per msec, and one per second it will stabilize | |
1392 | * to a value N*1000, where N is the rate of events per second. | |
1393 | * At constant rates between one per second and one per 32 seconds, | |
1394 | * it will be choppy, moving up on the seconds that have an event, | |
1395 | * and then decaying until the next event. At rates slower than | |
1396 | * about one in 32 seconds, it decays all the way back to zero between | |
1397 | * each event. | |
1398 | */ | |
1399 | ||
1400 | #define FM_COEF 933 /* coefficient for half-life of 10 secs */ | |
d2b43658 | 1401 | #define FM_MAXTICKS ((u32)99) /* useless computing more ticks than this */ |
3e0d98b9 PJ |
1402 | #define FM_MAXCNT 1000000 /* limit cnt to avoid overflow */ |
1403 | #define FM_SCALE 1000 /* faux fixed point scale */ | |
1404 | ||
1405 | /* Initialize a frequency meter */ | |
1406 | static void fmeter_init(struct fmeter *fmp) | |
1407 | { | |
1408 | fmp->cnt = 0; | |
1409 | fmp->val = 0; | |
1410 | fmp->time = 0; | |
1411 | spin_lock_init(&fmp->lock); | |
1412 | } | |
1413 | ||
1414 | /* Internal meter update - process cnt events and update value */ | |
1415 | static void fmeter_update(struct fmeter *fmp) | |
1416 | { | |
d2b43658 AB |
1417 | time64_t now; |
1418 | u32 ticks; | |
1419 | ||
1420 | now = ktime_get_seconds(); | |
1421 | ticks = now - fmp->time; | |
3e0d98b9 PJ |
1422 | |
1423 | if (ticks == 0) | |
1424 | return; | |
1425 | ||
1426 | ticks = min(FM_MAXTICKS, ticks); | |
1427 | while (ticks-- > 0) | |
1428 | fmp->val = (FM_COEF * fmp->val) / FM_SCALE; | |
1429 | fmp->time = now; | |
1430 | ||
1431 | fmp->val += ((FM_SCALE - FM_COEF) * fmp->cnt) / FM_SCALE; | |
1432 | fmp->cnt = 0; | |
1433 | } | |
1434 | ||
1435 | /* Process any previous ticks, then bump cnt by one (times scale). */ | |
1436 | static void fmeter_markevent(struct fmeter *fmp) | |
1437 | { | |
1438 | spin_lock(&fmp->lock); | |
1439 | fmeter_update(fmp); | |
1440 | fmp->cnt = min(FM_MAXCNT, fmp->cnt + FM_SCALE); | |
1441 | spin_unlock(&fmp->lock); | |
1442 | } | |
1443 | ||
1444 | /* Process any previous ticks, then return current value. */ | |
1445 | static int fmeter_getrate(struct fmeter *fmp) | |
1446 | { | |
1447 | int val; | |
1448 | ||
1449 | spin_lock(&fmp->lock); | |
1450 | fmeter_update(fmp); | |
1451 | val = fmp->val; | |
1452 | spin_unlock(&fmp->lock); | |
1453 | return val; | |
1454 | } | |
1455 | ||
57fce0a6 TH |
1456 | static struct cpuset *cpuset_attach_old_cs; |
1457 | ||
5d21cc2d | 1458 | /* Called by cgroups to determine if a cpuset is usable; cpuset_mutex held */ |
1f7dd3e5 | 1459 | static int cpuset_can_attach(struct cgroup_taskset *tset) |
f780bdb7 | 1460 | { |
1f7dd3e5 TH |
1461 | struct cgroup_subsys_state *css; |
1462 | struct cpuset *cs; | |
bb9d97b6 TH |
1463 | struct task_struct *task; |
1464 | int ret; | |
1da177e4 | 1465 | |
57fce0a6 | 1466 | /* used later by cpuset_attach() */ |
1f7dd3e5 TH |
1467 | cpuset_attach_old_cs = task_cs(cgroup_taskset_first(tset, &css)); |
1468 | cs = css_cs(css); | |
57fce0a6 | 1469 | |
5d21cc2d TH |
1470 | mutex_lock(&cpuset_mutex); |
1471 | ||
aa6ec29b | 1472 | /* allow moving tasks into an empty cpuset if on default hierarchy */ |
5d21cc2d | 1473 | ret = -ENOSPC; |
b8d1b8ee | 1474 | if (!is_in_v2_mode() && |
88fa523b | 1475 | (cpumask_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed))) |
5d21cc2d | 1476 | goto out_unlock; |
9985b0ba | 1477 | |
1f7dd3e5 | 1478 | cgroup_taskset_for_each(task, css, tset) { |
7f51412a JL |
1479 | ret = task_can_attach(task, cs->cpus_allowed); |
1480 | if (ret) | |
5d21cc2d TH |
1481 | goto out_unlock; |
1482 | ret = security_task_setscheduler(task); | |
1483 | if (ret) | |
1484 | goto out_unlock; | |
bb9d97b6 | 1485 | } |
f780bdb7 | 1486 | |
452477fa TH |
1487 | /* |
1488 | * Mark attach is in progress. This makes validate_change() fail | |
1489 | * changes which zero cpus/mems_allowed. | |
1490 | */ | |
1491 | cs->attach_in_progress++; | |
5d21cc2d TH |
1492 | ret = 0; |
1493 | out_unlock: | |
1494 | mutex_unlock(&cpuset_mutex); | |
1495 | return ret; | |
8793d854 | 1496 | } |
f780bdb7 | 1497 | |
1f7dd3e5 | 1498 | static void cpuset_cancel_attach(struct cgroup_taskset *tset) |
452477fa | 1499 | { |
1f7dd3e5 TH |
1500 | struct cgroup_subsys_state *css; |
1501 | struct cpuset *cs; | |
1502 | ||
1503 | cgroup_taskset_first(tset, &css); | |
1504 | cs = css_cs(css); | |
1505 | ||
5d21cc2d | 1506 | mutex_lock(&cpuset_mutex); |
eb95419b | 1507 | css_cs(css)->attach_in_progress--; |
5d21cc2d | 1508 | mutex_unlock(&cpuset_mutex); |
8793d854 | 1509 | } |
1da177e4 | 1510 | |
4e4c9a14 | 1511 | /* |
5d21cc2d | 1512 | * Protected by cpuset_mutex. cpus_attach is used only by cpuset_attach() |
4e4c9a14 TH |
1513 | * but we can't allocate it dynamically there. Define it global and |
1514 | * allocate from cpuset_init(). | |
1515 | */ | |
1516 | static cpumask_var_t cpus_attach; | |
1517 | ||
1f7dd3e5 | 1518 | static void cpuset_attach(struct cgroup_taskset *tset) |
8793d854 | 1519 | { |
67bd2c59 | 1520 | /* static buf protected by cpuset_mutex */ |
4e4c9a14 | 1521 | static nodemask_t cpuset_attach_nodemask_to; |
bb9d97b6 | 1522 | struct task_struct *task; |
4530eddb | 1523 | struct task_struct *leader; |
1f7dd3e5 TH |
1524 | struct cgroup_subsys_state *css; |
1525 | struct cpuset *cs; | |
57fce0a6 | 1526 | struct cpuset *oldcs = cpuset_attach_old_cs; |
22fb52dd | 1527 | |
1f7dd3e5 TH |
1528 | cgroup_taskset_first(tset, &css); |
1529 | cs = css_cs(css); | |
1530 | ||
5d21cc2d TH |
1531 | mutex_lock(&cpuset_mutex); |
1532 | ||
4e4c9a14 TH |
1533 | /* prepare for attach */ |
1534 | if (cs == &top_cpuset) | |
1535 | cpumask_copy(cpus_attach, cpu_possible_mask); | |
1536 | else | |
ae1c8023 | 1537 | guarantee_online_cpus(cs, cpus_attach); |
4e4c9a14 | 1538 | |
ae1c8023 | 1539 | guarantee_online_mems(cs, &cpuset_attach_nodemask_to); |
4e4c9a14 | 1540 | |
1f7dd3e5 | 1541 | cgroup_taskset_for_each(task, css, tset) { |
bb9d97b6 TH |
1542 | /* |
1543 | * can_attach beforehand should guarantee that this doesn't | |
1544 | * fail. TODO: have a better way to handle failure here | |
1545 | */ | |
1546 | WARN_ON_ONCE(set_cpus_allowed_ptr(task, cpus_attach)); | |
1547 | ||
1548 | cpuset_change_task_nodemask(task, &cpuset_attach_nodemask_to); | |
1549 | cpuset_update_task_spread_flag(cs, task); | |
1550 | } | |
22fb52dd | 1551 | |
f780bdb7 | 1552 | /* |
4530eddb TH |
1553 | * Change mm for all threadgroup leaders. This is expensive and may |
1554 | * sleep and should be moved outside migration path proper. | |
f780bdb7 | 1555 | */ |
ae1c8023 | 1556 | cpuset_attach_nodemask_to = cs->effective_mems; |
1f7dd3e5 | 1557 | cgroup_taskset_for_each_leader(leader, css, tset) { |
3df9ca0a TH |
1558 | struct mm_struct *mm = get_task_mm(leader); |
1559 | ||
1560 | if (mm) { | |
1561 | mpol_rebind_mm(mm, &cpuset_attach_nodemask_to); | |
1562 | ||
1563 | /* | |
1564 | * old_mems_allowed is the same with mems_allowed | |
1565 | * here, except if this task is being moved | |
1566 | * automatically due to hotplug. In that case | |
1567 | * @mems_allowed has been updated and is empty, so | |
1568 | * @old_mems_allowed is the right nodesets that we | |
1569 | * migrate mm from. | |
1570 | */ | |
e93ad19d | 1571 | if (is_memory_migrate(cs)) |
3df9ca0a TH |
1572 | cpuset_migrate_mm(mm, &oldcs->old_mems_allowed, |
1573 | &cpuset_attach_nodemask_to); | |
e93ad19d TH |
1574 | else |
1575 | mmput(mm); | |
f047cecf | 1576 | } |
4225399a | 1577 | } |
452477fa | 1578 | |
33ad801d | 1579 | cs->old_mems_allowed = cpuset_attach_nodemask_to; |
02bb5863 | 1580 | |
452477fa | 1581 | cs->attach_in_progress--; |
e44193d3 LZ |
1582 | if (!cs->attach_in_progress) |
1583 | wake_up(&cpuset_attach_wq); | |
5d21cc2d TH |
1584 | |
1585 | mutex_unlock(&cpuset_mutex); | |
1da177e4 LT |
1586 | } |
1587 | ||
1588 | /* The various types of files and directories in a cpuset file system */ | |
1589 | ||
1590 | typedef enum { | |
45b07ef3 | 1591 | FILE_MEMORY_MIGRATE, |
1da177e4 LT |
1592 | FILE_CPULIST, |
1593 | FILE_MEMLIST, | |
afd1a8b3 LZ |
1594 | FILE_EFFECTIVE_CPULIST, |
1595 | FILE_EFFECTIVE_MEMLIST, | |
1da177e4 LT |
1596 | FILE_CPU_EXCLUSIVE, |
1597 | FILE_MEM_EXCLUSIVE, | |
78608366 | 1598 | FILE_MEM_HARDWALL, |
029190c5 | 1599 | FILE_SCHED_LOAD_BALANCE, |
1d3504fc | 1600 | FILE_SCHED_RELAX_DOMAIN_LEVEL, |
3e0d98b9 PJ |
1601 | FILE_MEMORY_PRESSURE_ENABLED, |
1602 | FILE_MEMORY_PRESSURE, | |
825a46af PJ |
1603 | FILE_SPREAD_PAGE, |
1604 | FILE_SPREAD_SLAB, | |
1da177e4 LT |
1605 | } cpuset_filetype_t; |
1606 | ||
182446d0 TH |
1607 | static int cpuset_write_u64(struct cgroup_subsys_state *css, struct cftype *cft, |
1608 | u64 val) | |
700fe1ab | 1609 | { |
182446d0 | 1610 | struct cpuset *cs = css_cs(css); |
700fe1ab | 1611 | cpuset_filetype_t type = cft->private; |
a903f086 | 1612 | int retval = 0; |
700fe1ab | 1613 | |
5d21cc2d | 1614 | mutex_lock(&cpuset_mutex); |
a903f086 LZ |
1615 | if (!is_cpuset_online(cs)) { |
1616 | retval = -ENODEV; | |
5d21cc2d | 1617 | goto out_unlock; |
a903f086 | 1618 | } |
700fe1ab PM |
1619 | |
1620 | switch (type) { | |
1da177e4 | 1621 | case FILE_CPU_EXCLUSIVE: |
700fe1ab | 1622 | retval = update_flag(CS_CPU_EXCLUSIVE, cs, val); |
1da177e4 LT |
1623 | break; |
1624 | case FILE_MEM_EXCLUSIVE: | |
700fe1ab | 1625 | retval = update_flag(CS_MEM_EXCLUSIVE, cs, val); |
1da177e4 | 1626 | break; |
78608366 PM |
1627 | case FILE_MEM_HARDWALL: |
1628 | retval = update_flag(CS_MEM_HARDWALL, cs, val); | |
1629 | break; | |
029190c5 | 1630 | case FILE_SCHED_LOAD_BALANCE: |
700fe1ab | 1631 | retval = update_flag(CS_SCHED_LOAD_BALANCE, cs, val); |
1d3504fc | 1632 | break; |
45b07ef3 | 1633 | case FILE_MEMORY_MIGRATE: |
700fe1ab | 1634 | retval = update_flag(CS_MEMORY_MIGRATE, cs, val); |
45b07ef3 | 1635 | break; |
3e0d98b9 | 1636 | case FILE_MEMORY_PRESSURE_ENABLED: |
700fe1ab | 1637 | cpuset_memory_pressure_enabled = !!val; |
3e0d98b9 | 1638 | break; |
825a46af | 1639 | case FILE_SPREAD_PAGE: |
700fe1ab | 1640 | retval = update_flag(CS_SPREAD_PAGE, cs, val); |
825a46af PJ |
1641 | break; |
1642 | case FILE_SPREAD_SLAB: | |
700fe1ab | 1643 | retval = update_flag(CS_SPREAD_SLAB, cs, val); |
825a46af | 1644 | break; |
1da177e4 LT |
1645 | default: |
1646 | retval = -EINVAL; | |
700fe1ab | 1647 | break; |
1da177e4 | 1648 | } |
5d21cc2d TH |
1649 | out_unlock: |
1650 | mutex_unlock(&cpuset_mutex); | |
1da177e4 LT |
1651 | return retval; |
1652 | } | |
1653 | ||
182446d0 TH |
1654 | static int cpuset_write_s64(struct cgroup_subsys_state *css, struct cftype *cft, |
1655 | s64 val) | |
5be7a479 | 1656 | { |
182446d0 | 1657 | struct cpuset *cs = css_cs(css); |
5be7a479 | 1658 | cpuset_filetype_t type = cft->private; |
5d21cc2d | 1659 | int retval = -ENODEV; |
5be7a479 | 1660 | |
5d21cc2d TH |
1661 | mutex_lock(&cpuset_mutex); |
1662 | if (!is_cpuset_online(cs)) | |
1663 | goto out_unlock; | |
e3712395 | 1664 | |
5be7a479 PM |
1665 | switch (type) { |
1666 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1667 | retval = update_relax_domain_level(cs, val); | |
1668 | break; | |
1669 | default: | |
1670 | retval = -EINVAL; | |
1671 | break; | |
1672 | } | |
5d21cc2d TH |
1673 | out_unlock: |
1674 | mutex_unlock(&cpuset_mutex); | |
5be7a479 PM |
1675 | return retval; |
1676 | } | |
1677 | ||
e3712395 PM |
1678 | /* |
1679 | * Common handling for a write to a "cpus" or "mems" file. | |
1680 | */ | |
451af504 TH |
1681 | static ssize_t cpuset_write_resmask(struct kernfs_open_file *of, |
1682 | char *buf, size_t nbytes, loff_t off) | |
e3712395 | 1683 | { |
451af504 | 1684 | struct cpuset *cs = css_cs(of_css(of)); |
645fcc9d | 1685 | struct cpuset *trialcs; |
5d21cc2d | 1686 | int retval = -ENODEV; |
e3712395 | 1687 | |
451af504 TH |
1688 | buf = strstrip(buf); |
1689 | ||
3a5a6d0c TH |
1690 | /* |
1691 | * CPU or memory hotunplug may leave @cs w/o any execution | |
1692 | * resources, in which case the hotplug code asynchronously updates | |
1693 | * configuration and transfers all tasks to the nearest ancestor | |
1694 | * which can execute. | |
1695 | * | |
1696 | * As writes to "cpus" or "mems" may restore @cs's execution | |
1697 | * resources, wait for the previously scheduled operations before | |
1698 | * proceeding, so that we don't end up keep removing tasks added | |
1699 | * after execution capability is restored. | |
76bb5ab8 TH |
1700 | * |
1701 | * cpuset_hotplug_work calls back into cgroup core via | |
1702 | * cgroup_transfer_tasks() and waiting for it from a cgroupfs | |
1703 | * operation like this one can lead to a deadlock through kernfs | |
1704 | * active_ref protection. Let's break the protection. Losing the | |
1705 | * protection is okay as we check whether @cs is online after | |
1706 | * grabbing cpuset_mutex anyway. This only happens on the legacy | |
1707 | * hierarchies. | |
3a5a6d0c | 1708 | */ |
76bb5ab8 TH |
1709 | css_get(&cs->css); |
1710 | kernfs_break_active_protection(of->kn); | |
3a5a6d0c TH |
1711 | flush_work(&cpuset_hotplug_work); |
1712 | ||
5d21cc2d TH |
1713 | mutex_lock(&cpuset_mutex); |
1714 | if (!is_cpuset_online(cs)) | |
1715 | goto out_unlock; | |
e3712395 | 1716 | |
645fcc9d | 1717 | trialcs = alloc_trial_cpuset(cs); |
b75f38d6 LZ |
1718 | if (!trialcs) { |
1719 | retval = -ENOMEM; | |
5d21cc2d | 1720 | goto out_unlock; |
b75f38d6 | 1721 | } |
645fcc9d | 1722 | |
451af504 | 1723 | switch (of_cft(of)->private) { |
e3712395 | 1724 | case FILE_CPULIST: |
645fcc9d | 1725 | retval = update_cpumask(cs, trialcs, buf); |
e3712395 PM |
1726 | break; |
1727 | case FILE_MEMLIST: | |
645fcc9d | 1728 | retval = update_nodemask(cs, trialcs, buf); |
e3712395 PM |
1729 | break; |
1730 | default: | |
1731 | retval = -EINVAL; | |
1732 | break; | |
1733 | } | |
645fcc9d LZ |
1734 | |
1735 | free_trial_cpuset(trialcs); | |
5d21cc2d TH |
1736 | out_unlock: |
1737 | mutex_unlock(&cpuset_mutex); | |
76bb5ab8 TH |
1738 | kernfs_unbreak_active_protection(of->kn); |
1739 | css_put(&cs->css); | |
e93ad19d | 1740 | flush_workqueue(cpuset_migrate_mm_wq); |
451af504 | 1741 | return retval ?: nbytes; |
e3712395 PM |
1742 | } |
1743 | ||
1da177e4 LT |
1744 | /* |
1745 | * These ascii lists should be read in a single call, by using a user | |
1746 | * buffer large enough to hold the entire map. If read in smaller | |
1747 | * chunks, there is no guarantee of atomicity. Since the display format | |
1748 | * used, list of ranges of sequential numbers, is variable length, | |
1749 | * and since these maps can change value dynamically, one could read | |
1750 | * gibberish by doing partial reads while a list was changing. | |
1da177e4 | 1751 | */ |
2da8ca82 | 1752 | static int cpuset_common_seq_show(struct seq_file *sf, void *v) |
1da177e4 | 1753 | { |
2da8ca82 TH |
1754 | struct cpuset *cs = css_cs(seq_css(sf)); |
1755 | cpuset_filetype_t type = seq_cft(sf)->private; | |
51ffe411 | 1756 | int ret = 0; |
1da177e4 | 1757 | |
8447a0fe | 1758 | spin_lock_irq(&callback_lock); |
1da177e4 LT |
1759 | |
1760 | switch (type) { | |
1761 | case FILE_CPULIST: | |
e8e6d97c | 1762 | seq_printf(sf, "%*pbl\n", cpumask_pr_args(cs->cpus_allowed)); |
1da177e4 LT |
1763 | break; |
1764 | case FILE_MEMLIST: | |
e8e6d97c | 1765 | seq_printf(sf, "%*pbl\n", nodemask_pr_args(&cs->mems_allowed)); |
1da177e4 | 1766 | break; |
afd1a8b3 | 1767 | case FILE_EFFECTIVE_CPULIST: |
e8e6d97c | 1768 | seq_printf(sf, "%*pbl\n", cpumask_pr_args(cs->effective_cpus)); |
afd1a8b3 LZ |
1769 | break; |
1770 | case FILE_EFFECTIVE_MEMLIST: | |
e8e6d97c | 1771 | seq_printf(sf, "%*pbl\n", nodemask_pr_args(&cs->effective_mems)); |
afd1a8b3 | 1772 | break; |
1da177e4 | 1773 | default: |
51ffe411 | 1774 | ret = -EINVAL; |
1da177e4 | 1775 | } |
1da177e4 | 1776 | |
8447a0fe | 1777 | spin_unlock_irq(&callback_lock); |
51ffe411 | 1778 | return ret; |
1da177e4 LT |
1779 | } |
1780 | ||
182446d0 | 1781 | static u64 cpuset_read_u64(struct cgroup_subsys_state *css, struct cftype *cft) |
700fe1ab | 1782 | { |
182446d0 | 1783 | struct cpuset *cs = css_cs(css); |
700fe1ab PM |
1784 | cpuset_filetype_t type = cft->private; |
1785 | switch (type) { | |
1786 | case FILE_CPU_EXCLUSIVE: | |
1787 | return is_cpu_exclusive(cs); | |
1788 | case FILE_MEM_EXCLUSIVE: | |
1789 | return is_mem_exclusive(cs); | |
78608366 PM |
1790 | case FILE_MEM_HARDWALL: |
1791 | return is_mem_hardwall(cs); | |
700fe1ab PM |
1792 | case FILE_SCHED_LOAD_BALANCE: |
1793 | return is_sched_load_balance(cs); | |
1794 | case FILE_MEMORY_MIGRATE: | |
1795 | return is_memory_migrate(cs); | |
1796 | case FILE_MEMORY_PRESSURE_ENABLED: | |
1797 | return cpuset_memory_pressure_enabled; | |
1798 | case FILE_MEMORY_PRESSURE: | |
1799 | return fmeter_getrate(&cs->fmeter); | |
1800 | case FILE_SPREAD_PAGE: | |
1801 | return is_spread_page(cs); | |
1802 | case FILE_SPREAD_SLAB: | |
1803 | return is_spread_slab(cs); | |
1804 | default: | |
1805 | BUG(); | |
1806 | } | |
cf417141 MK |
1807 | |
1808 | /* Unreachable but makes gcc happy */ | |
1809 | return 0; | |
700fe1ab | 1810 | } |
1da177e4 | 1811 | |
182446d0 | 1812 | static s64 cpuset_read_s64(struct cgroup_subsys_state *css, struct cftype *cft) |
5be7a479 | 1813 | { |
182446d0 | 1814 | struct cpuset *cs = css_cs(css); |
5be7a479 PM |
1815 | cpuset_filetype_t type = cft->private; |
1816 | switch (type) { | |
1817 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1818 | return cs->relax_domain_level; | |
1819 | default: | |
1820 | BUG(); | |
1821 | } | |
cf417141 MK |
1822 | |
1823 | /* Unrechable but makes gcc happy */ | |
1824 | return 0; | |
5be7a479 PM |
1825 | } |
1826 | ||
1da177e4 LT |
1827 | |
1828 | /* | |
1829 | * for the common functions, 'private' gives the type of file | |
1830 | */ | |
1831 | ||
addf2c73 PM |
1832 | static struct cftype files[] = { |
1833 | { | |
1834 | .name = "cpus", | |
2da8ca82 | 1835 | .seq_show = cpuset_common_seq_show, |
451af504 | 1836 | .write = cpuset_write_resmask, |
e3712395 | 1837 | .max_write_len = (100U + 6 * NR_CPUS), |
addf2c73 PM |
1838 | .private = FILE_CPULIST, |
1839 | }, | |
1840 | ||
1841 | { | |
1842 | .name = "mems", | |
2da8ca82 | 1843 | .seq_show = cpuset_common_seq_show, |
451af504 | 1844 | .write = cpuset_write_resmask, |
e3712395 | 1845 | .max_write_len = (100U + 6 * MAX_NUMNODES), |
addf2c73 PM |
1846 | .private = FILE_MEMLIST, |
1847 | }, | |
1848 | ||
afd1a8b3 LZ |
1849 | { |
1850 | .name = "effective_cpus", | |
1851 | .seq_show = cpuset_common_seq_show, | |
1852 | .private = FILE_EFFECTIVE_CPULIST, | |
1853 | }, | |
1854 | ||
1855 | { | |
1856 | .name = "effective_mems", | |
1857 | .seq_show = cpuset_common_seq_show, | |
1858 | .private = FILE_EFFECTIVE_MEMLIST, | |
1859 | }, | |
1860 | ||
addf2c73 PM |
1861 | { |
1862 | .name = "cpu_exclusive", | |
1863 | .read_u64 = cpuset_read_u64, | |
1864 | .write_u64 = cpuset_write_u64, | |
1865 | .private = FILE_CPU_EXCLUSIVE, | |
1866 | }, | |
1867 | ||
1868 | { | |
1869 | .name = "mem_exclusive", | |
1870 | .read_u64 = cpuset_read_u64, | |
1871 | .write_u64 = cpuset_write_u64, | |
1872 | .private = FILE_MEM_EXCLUSIVE, | |
1873 | }, | |
1874 | ||
78608366 PM |
1875 | { |
1876 | .name = "mem_hardwall", | |
1877 | .read_u64 = cpuset_read_u64, | |
1878 | .write_u64 = cpuset_write_u64, | |
1879 | .private = FILE_MEM_HARDWALL, | |
1880 | }, | |
1881 | ||
addf2c73 PM |
1882 | { |
1883 | .name = "sched_load_balance", | |
1884 | .read_u64 = cpuset_read_u64, | |
1885 | .write_u64 = cpuset_write_u64, | |
1886 | .private = FILE_SCHED_LOAD_BALANCE, | |
1887 | }, | |
1888 | ||
1889 | { | |
1890 | .name = "sched_relax_domain_level", | |
5be7a479 PM |
1891 | .read_s64 = cpuset_read_s64, |
1892 | .write_s64 = cpuset_write_s64, | |
addf2c73 PM |
1893 | .private = FILE_SCHED_RELAX_DOMAIN_LEVEL, |
1894 | }, | |
1895 | ||
1896 | { | |
1897 | .name = "memory_migrate", | |
1898 | .read_u64 = cpuset_read_u64, | |
1899 | .write_u64 = cpuset_write_u64, | |
1900 | .private = FILE_MEMORY_MIGRATE, | |
1901 | }, | |
1902 | ||
1903 | { | |
1904 | .name = "memory_pressure", | |
1905 | .read_u64 = cpuset_read_u64, | |
1c08c22c | 1906 | .private = FILE_MEMORY_PRESSURE, |
addf2c73 PM |
1907 | }, |
1908 | ||
1909 | { | |
1910 | .name = "memory_spread_page", | |
1911 | .read_u64 = cpuset_read_u64, | |
1912 | .write_u64 = cpuset_write_u64, | |
1913 | .private = FILE_SPREAD_PAGE, | |
1914 | }, | |
1915 | ||
1916 | { | |
1917 | .name = "memory_spread_slab", | |
1918 | .read_u64 = cpuset_read_u64, | |
1919 | .write_u64 = cpuset_write_u64, | |
1920 | .private = FILE_SPREAD_SLAB, | |
1921 | }, | |
3e0d98b9 | 1922 | |
4baf6e33 TH |
1923 | { |
1924 | .name = "memory_pressure_enabled", | |
1925 | .flags = CFTYPE_ONLY_ON_ROOT, | |
1926 | .read_u64 = cpuset_read_u64, | |
1927 | .write_u64 = cpuset_write_u64, | |
1928 | .private = FILE_MEMORY_PRESSURE_ENABLED, | |
1929 | }, | |
1da177e4 | 1930 | |
4baf6e33 TH |
1931 | { } /* terminate */ |
1932 | }; | |
1da177e4 LT |
1933 | |
1934 | /* | |
92fb9748 | 1935 | * cpuset_css_alloc - allocate a cpuset css |
c9e5fe66 | 1936 | * cgrp: control group that the new cpuset will be part of |
1da177e4 LT |
1937 | */ |
1938 | ||
eb95419b TH |
1939 | static struct cgroup_subsys_state * |
1940 | cpuset_css_alloc(struct cgroup_subsys_state *parent_css) | |
1da177e4 | 1941 | { |
c8f699bb | 1942 | struct cpuset *cs; |
1da177e4 | 1943 | |
eb95419b | 1944 | if (!parent_css) |
8793d854 | 1945 | return &top_cpuset.css; |
033fa1c5 | 1946 | |
c8f699bb | 1947 | cs = kzalloc(sizeof(*cs), GFP_KERNEL); |
1da177e4 | 1948 | if (!cs) |
8793d854 | 1949 | return ERR_PTR(-ENOMEM); |
e2b9a3d7 LZ |
1950 | if (!alloc_cpumask_var(&cs->cpus_allowed, GFP_KERNEL)) |
1951 | goto free_cs; | |
1952 | if (!alloc_cpumask_var(&cs->effective_cpus, GFP_KERNEL)) | |
1953 | goto free_cpus; | |
1da177e4 | 1954 | |
029190c5 | 1955 | set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); |
300ed6cb | 1956 | cpumask_clear(cs->cpus_allowed); |
f9a86fcb | 1957 | nodes_clear(cs->mems_allowed); |
e2b9a3d7 LZ |
1958 | cpumask_clear(cs->effective_cpus); |
1959 | nodes_clear(cs->effective_mems); | |
3e0d98b9 | 1960 | fmeter_init(&cs->fmeter); |
1d3504fc | 1961 | cs->relax_domain_level = -1; |
1da177e4 | 1962 | |
c8f699bb | 1963 | return &cs->css; |
e2b9a3d7 LZ |
1964 | |
1965 | free_cpus: | |
1966 | free_cpumask_var(cs->cpus_allowed); | |
1967 | free_cs: | |
1968 | kfree(cs); | |
1969 | return ERR_PTR(-ENOMEM); | |
c8f699bb TH |
1970 | } |
1971 | ||
eb95419b | 1972 | static int cpuset_css_online(struct cgroup_subsys_state *css) |
c8f699bb | 1973 | { |
eb95419b | 1974 | struct cpuset *cs = css_cs(css); |
c431069f | 1975 | struct cpuset *parent = parent_cs(cs); |
ae8086ce | 1976 | struct cpuset *tmp_cs; |
492eb21b | 1977 | struct cgroup_subsys_state *pos_css; |
c8f699bb TH |
1978 | |
1979 | if (!parent) | |
1980 | return 0; | |
1981 | ||
5d21cc2d TH |
1982 | mutex_lock(&cpuset_mutex); |
1983 | ||
efeb77b2 | 1984 | set_bit(CS_ONLINE, &cs->flags); |
c8f699bb TH |
1985 | if (is_spread_page(parent)) |
1986 | set_bit(CS_SPREAD_PAGE, &cs->flags); | |
1987 | if (is_spread_slab(parent)) | |
1988 | set_bit(CS_SPREAD_SLAB, &cs->flags); | |
1da177e4 | 1989 | |
664eedde | 1990 | cpuset_inc(); |
033fa1c5 | 1991 | |
8447a0fe | 1992 | spin_lock_irq(&callback_lock); |
b8d1b8ee | 1993 | if (is_in_v2_mode()) { |
e2b9a3d7 LZ |
1994 | cpumask_copy(cs->effective_cpus, parent->effective_cpus); |
1995 | cs->effective_mems = parent->effective_mems; | |
1996 | } | |
8447a0fe | 1997 | spin_unlock_irq(&callback_lock); |
e2b9a3d7 | 1998 | |
eb95419b | 1999 | if (!test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags)) |
5d21cc2d | 2000 | goto out_unlock; |
033fa1c5 TH |
2001 | |
2002 | /* | |
2003 | * Clone @parent's configuration if CGRP_CPUSET_CLONE_CHILDREN is | |
2004 | * set. This flag handling is implemented in cgroup core for | |
2005 | * histrical reasons - the flag may be specified during mount. | |
2006 | * | |
2007 | * Currently, if any sibling cpusets have exclusive cpus or mem, we | |
2008 | * refuse to clone the configuration - thereby refusing the task to | |
2009 | * be entered, and as a result refusing the sys_unshare() or | |
2010 | * clone() which initiated it. If this becomes a problem for some | |
2011 | * users who wish to allow that scenario, then this could be | |
2012 | * changed to grant parent->cpus_allowed-sibling_cpus_exclusive | |
2013 | * (and likewise for mems) to the new cgroup. | |
2014 | */ | |
ae8086ce | 2015 | rcu_read_lock(); |
492eb21b | 2016 | cpuset_for_each_child(tmp_cs, pos_css, parent) { |
ae8086ce TH |
2017 | if (is_mem_exclusive(tmp_cs) || is_cpu_exclusive(tmp_cs)) { |
2018 | rcu_read_unlock(); | |
5d21cc2d | 2019 | goto out_unlock; |
ae8086ce | 2020 | } |
033fa1c5 | 2021 | } |
ae8086ce | 2022 | rcu_read_unlock(); |
033fa1c5 | 2023 | |
8447a0fe | 2024 | spin_lock_irq(&callback_lock); |
033fa1c5 | 2025 | cs->mems_allowed = parent->mems_allowed; |
790317e1 | 2026 | cs->effective_mems = parent->mems_allowed; |
033fa1c5 | 2027 | cpumask_copy(cs->cpus_allowed, parent->cpus_allowed); |
790317e1 | 2028 | cpumask_copy(cs->effective_cpus, parent->cpus_allowed); |
cea74465 | 2029 | spin_unlock_irq(&callback_lock); |
5d21cc2d TH |
2030 | out_unlock: |
2031 | mutex_unlock(&cpuset_mutex); | |
c8f699bb TH |
2032 | return 0; |
2033 | } | |
2034 | ||
0b9e6965 ZH |
2035 | /* |
2036 | * If the cpuset being removed has its flag 'sched_load_balance' | |
2037 | * enabled, then simulate turning sched_load_balance off, which | |
2038 | * will call rebuild_sched_domains_locked(). | |
2039 | */ | |
2040 | ||
eb95419b | 2041 | static void cpuset_css_offline(struct cgroup_subsys_state *css) |
c8f699bb | 2042 | { |
eb95419b | 2043 | struct cpuset *cs = css_cs(css); |
c8f699bb | 2044 | |
5d21cc2d | 2045 | mutex_lock(&cpuset_mutex); |
c8f699bb TH |
2046 | |
2047 | if (is_sched_load_balance(cs)) | |
2048 | update_flag(CS_SCHED_LOAD_BALANCE, cs, 0); | |
2049 | ||
664eedde | 2050 | cpuset_dec(); |
efeb77b2 | 2051 | clear_bit(CS_ONLINE, &cs->flags); |
c8f699bb | 2052 | |
5d21cc2d | 2053 | mutex_unlock(&cpuset_mutex); |
1da177e4 LT |
2054 | } |
2055 | ||
eb95419b | 2056 | static void cpuset_css_free(struct cgroup_subsys_state *css) |
1da177e4 | 2057 | { |
eb95419b | 2058 | struct cpuset *cs = css_cs(css); |
1da177e4 | 2059 | |
e2b9a3d7 | 2060 | free_cpumask_var(cs->effective_cpus); |
300ed6cb | 2061 | free_cpumask_var(cs->cpus_allowed); |
8793d854 | 2062 | kfree(cs); |
1da177e4 LT |
2063 | } |
2064 | ||
39bd0d15 LZ |
2065 | static void cpuset_bind(struct cgroup_subsys_state *root_css) |
2066 | { | |
2067 | mutex_lock(&cpuset_mutex); | |
8447a0fe | 2068 | spin_lock_irq(&callback_lock); |
39bd0d15 | 2069 | |
b8d1b8ee | 2070 | if (is_in_v2_mode()) { |
39bd0d15 LZ |
2071 | cpumask_copy(top_cpuset.cpus_allowed, cpu_possible_mask); |
2072 | top_cpuset.mems_allowed = node_possible_map; | |
2073 | } else { | |
2074 | cpumask_copy(top_cpuset.cpus_allowed, | |
2075 | top_cpuset.effective_cpus); | |
2076 | top_cpuset.mems_allowed = top_cpuset.effective_mems; | |
2077 | } | |
2078 | ||
8447a0fe | 2079 | spin_unlock_irq(&callback_lock); |
39bd0d15 LZ |
2080 | mutex_unlock(&cpuset_mutex); |
2081 | } | |
2082 | ||
06f4e948 ZL |
2083 | /* |
2084 | * Make sure the new task conform to the current state of its parent, | |
2085 | * which could have been changed by cpuset just after it inherits the | |
2086 | * state from the parent and before it sits on the cgroup's task list. | |
2087 | */ | |
8a15b817 | 2088 | static void cpuset_fork(struct task_struct *task) |
06f4e948 ZL |
2089 | { |
2090 | if (task_css_is_root(task, cpuset_cgrp_id)) | |
2091 | return; | |
2092 | ||
2093 | set_cpus_allowed_ptr(task, ¤t->cpus_allowed); | |
2094 | task->mems_allowed = current->mems_allowed; | |
2095 | } | |
2096 | ||
073219e9 | 2097 | struct cgroup_subsys cpuset_cgrp_subsys = { |
39bd0d15 LZ |
2098 | .css_alloc = cpuset_css_alloc, |
2099 | .css_online = cpuset_css_online, | |
2100 | .css_offline = cpuset_css_offline, | |
2101 | .css_free = cpuset_css_free, | |
2102 | .can_attach = cpuset_can_attach, | |
2103 | .cancel_attach = cpuset_cancel_attach, | |
2104 | .attach = cpuset_attach, | |
5cf1cacb | 2105 | .post_attach = cpuset_post_attach, |
39bd0d15 | 2106 | .bind = cpuset_bind, |
06f4e948 | 2107 | .fork = cpuset_fork, |
5577964e | 2108 | .legacy_cftypes = files, |
b38e42e9 | 2109 | .early_init = true, |
8793d854 PM |
2110 | }; |
2111 | ||
1da177e4 LT |
2112 | /** |
2113 | * cpuset_init - initialize cpusets at system boot | |
2114 | * | |
2115 | * Description: Initialize top_cpuset and the cpuset internal file system, | |
2116 | **/ | |
2117 | ||
2118 | int __init cpuset_init(void) | |
2119 | { | |
8793d854 | 2120 | int err = 0; |
1da177e4 | 2121 | |
75fa8e5d NMG |
2122 | BUG_ON(!alloc_cpumask_var(&top_cpuset.cpus_allowed, GFP_KERNEL)); |
2123 | BUG_ON(!alloc_cpumask_var(&top_cpuset.effective_cpus, GFP_KERNEL)); | |
58568d2a | 2124 | |
300ed6cb | 2125 | cpumask_setall(top_cpuset.cpus_allowed); |
f9a86fcb | 2126 | nodes_setall(top_cpuset.mems_allowed); |
e2b9a3d7 LZ |
2127 | cpumask_setall(top_cpuset.effective_cpus); |
2128 | nodes_setall(top_cpuset.effective_mems); | |
1da177e4 | 2129 | |
3e0d98b9 | 2130 | fmeter_init(&top_cpuset.fmeter); |
029190c5 | 2131 | set_bit(CS_SCHED_LOAD_BALANCE, &top_cpuset.flags); |
1d3504fc | 2132 | top_cpuset.relax_domain_level = -1; |
1da177e4 | 2133 | |
1da177e4 LT |
2134 | err = register_filesystem(&cpuset_fs_type); |
2135 | if (err < 0) | |
8793d854 PM |
2136 | return err; |
2137 | ||
75fa8e5d | 2138 | BUG_ON(!alloc_cpumask_var(&cpus_attach, GFP_KERNEL)); |
2341d1b6 | 2139 | |
8793d854 | 2140 | return 0; |
1da177e4 LT |
2141 | } |
2142 | ||
b1aac8bb | 2143 | /* |
cf417141 | 2144 | * If CPU and/or memory hotplug handlers, below, unplug any CPUs |
b1aac8bb PJ |
2145 | * or memory nodes, we need to walk over the cpuset hierarchy, |
2146 | * removing that CPU or node from all cpusets. If this removes the | |
956db3ca CW |
2147 | * last CPU or node from a cpuset, then move the tasks in the empty |
2148 | * cpuset to its next-highest non-empty parent. | |
b1aac8bb | 2149 | */ |
956db3ca CW |
2150 | static void remove_tasks_in_empty_cpuset(struct cpuset *cs) |
2151 | { | |
2152 | struct cpuset *parent; | |
2153 | ||
956db3ca CW |
2154 | /* |
2155 | * Find its next-highest non-empty parent, (top cpuset | |
2156 | * has online cpus, so can't be empty). | |
2157 | */ | |
c431069f | 2158 | parent = parent_cs(cs); |
300ed6cb | 2159 | while (cpumask_empty(parent->cpus_allowed) || |
b4501295 | 2160 | nodes_empty(parent->mems_allowed)) |
c431069f | 2161 | parent = parent_cs(parent); |
956db3ca | 2162 | |
8cc99345 | 2163 | if (cgroup_transfer_tasks(parent->css.cgroup, cs->css.cgroup)) { |
12d3089c | 2164 | pr_err("cpuset: failed to transfer tasks out of empty cpuset "); |
e61734c5 TH |
2165 | pr_cont_cgroup_name(cs->css.cgroup); |
2166 | pr_cont("\n"); | |
8cc99345 | 2167 | } |
956db3ca CW |
2168 | } |
2169 | ||
be4c9dd7 LZ |
2170 | static void |
2171 | hotplug_update_tasks_legacy(struct cpuset *cs, | |
2172 | struct cpumask *new_cpus, nodemask_t *new_mems, | |
2173 | bool cpus_updated, bool mems_updated) | |
390a36aa LZ |
2174 | { |
2175 | bool is_empty; | |
2176 | ||
8447a0fe | 2177 | spin_lock_irq(&callback_lock); |
be4c9dd7 LZ |
2178 | cpumask_copy(cs->cpus_allowed, new_cpus); |
2179 | cpumask_copy(cs->effective_cpus, new_cpus); | |
2180 | cs->mems_allowed = *new_mems; | |
2181 | cs->effective_mems = *new_mems; | |
8447a0fe | 2182 | spin_unlock_irq(&callback_lock); |
390a36aa LZ |
2183 | |
2184 | /* | |
2185 | * Don't call update_tasks_cpumask() if the cpuset becomes empty, | |
2186 | * as the tasks will be migratecd to an ancestor. | |
2187 | */ | |
be4c9dd7 | 2188 | if (cpus_updated && !cpumask_empty(cs->cpus_allowed)) |
390a36aa | 2189 | update_tasks_cpumask(cs); |
be4c9dd7 | 2190 | if (mems_updated && !nodes_empty(cs->mems_allowed)) |
390a36aa LZ |
2191 | update_tasks_nodemask(cs); |
2192 | ||
2193 | is_empty = cpumask_empty(cs->cpus_allowed) || | |
2194 | nodes_empty(cs->mems_allowed); | |
2195 | ||
2196 | mutex_unlock(&cpuset_mutex); | |
2197 | ||
2198 | /* | |
2199 | * Move tasks to the nearest ancestor with execution resources, | |
2200 | * This is full cgroup operation which will also call back into | |
2201 | * cpuset. Should be done outside any lock. | |
2202 | */ | |
2203 | if (is_empty) | |
2204 | remove_tasks_in_empty_cpuset(cs); | |
2205 | ||
2206 | mutex_lock(&cpuset_mutex); | |
2207 | } | |
2208 | ||
be4c9dd7 LZ |
2209 | static void |
2210 | hotplug_update_tasks(struct cpuset *cs, | |
2211 | struct cpumask *new_cpus, nodemask_t *new_mems, | |
2212 | bool cpus_updated, bool mems_updated) | |
390a36aa | 2213 | { |
be4c9dd7 LZ |
2214 | if (cpumask_empty(new_cpus)) |
2215 | cpumask_copy(new_cpus, parent_cs(cs)->effective_cpus); | |
2216 | if (nodes_empty(*new_mems)) | |
2217 | *new_mems = parent_cs(cs)->effective_mems; | |
2218 | ||
8447a0fe | 2219 | spin_lock_irq(&callback_lock); |
be4c9dd7 LZ |
2220 | cpumask_copy(cs->effective_cpus, new_cpus); |
2221 | cs->effective_mems = *new_mems; | |
8447a0fe | 2222 | spin_unlock_irq(&callback_lock); |
390a36aa | 2223 | |
be4c9dd7 | 2224 | if (cpus_updated) |
390a36aa | 2225 | update_tasks_cpumask(cs); |
be4c9dd7 | 2226 | if (mems_updated) |
390a36aa LZ |
2227 | update_tasks_nodemask(cs); |
2228 | } | |
2229 | ||
deb7aa30 | 2230 | /** |
388afd85 | 2231 | * cpuset_hotplug_update_tasks - update tasks in a cpuset for hotunplug |
deb7aa30 | 2232 | * @cs: cpuset in interest |
956db3ca | 2233 | * |
deb7aa30 TH |
2234 | * Compare @cs's cpu and mem masks against top_cpuset and if some have gone |
2235 | * offline, update @cs accordingly. If @cs ends up with no CPU or memory, | |
2236 | * all its tasks are moved to the nearest ancestor with both resources. | |
80d1fa64 | 2237 | */ |
388afd85 | 2238 | static void cpuset_hotplug_update_tasks(struct cpuset *cs) |
80d1fa64 | 2239 | { |
be4c9dd7 LZ |
2240 | static cpumask_t new_cpus; |
2241 | static nodemask_t new_mems; | |
2242 | bool cpus_updated; | |
2243 | bool mems_updated; | |
e44193d3 LZ |
2244 | retry: |
2245 | wait_event(cpuset_attach_wq, cs->attach_in_progress == 0); | |
80d1fa64 | 2246 | |
5d21cc2d | 2247 | mutex_lock(&cpuset_mutex); |
7ddf96b0 | 2248 | |
e44193d3 LZ |
2249 | /* |
2250 | * We have raced with task attaching. We wait until attaching | |
2251 | * is finished, so we won't attach a task to an empty cpuset. | |
2252 | */ | |
2253 | if (cs->attach_in_progress) { | |
2254 | mutex_unlock(&cpuset_mutex); | |
2255 | goto retry; | |
2256 | } | |
2257 | ||
be4c9dd7 LZ |
2258 | cpumask_and(&new_cpus, cs->cpus_allowed, parent_cs(cs)->effective_cpus); |
2259 | nodes_and(new_mems, cs->mems_allowed, parent_cs(cs)->effective_mems); | |
80d1fa64 | 2260 | |
be4c9dd7 LZ |
2261 | cpus_updated = !cpumask_equal(&new_cpus, cs->effective_cpus); |
2262 | mems_updated = !nodes_equal(new_mems, cs->effective_mems); | |
deb7aa30 | 2263 | |
b8d1b8ee | 2264 | if (is_in_v2_mode()) |
be4c9dd7 LZ |
2265 | hotplug_update_tasks(cs, &new_cpus, &new_mems, |
2266 | cpus_updated, mems_updated); | |
390a36aa | 2267 | else |
be4c9dd7 LZ |
2268 | hotplug_update_tasks_legacy(cs, &new_cpus, &new_mems, |
2269 | cpus_updated, mems_updated); | |
8d033948 | 2270 | |
5d21cc2d | 2271 | mutex_unlock(&cpuset_mutex); |
b1aac8bb PJ |
2272 | } |
2273 | ||
50e76632 PZ |
2274 | static bool force_rebuild; |
2275 | ||
2276 | void cpuset_force_rebuild(void) | |
2277 | { | |
2278 | force_rebuild = true; | |
2279 | } | |
2280 | ||
deb7aa30 | 2281 | /** |
3a5a6d0c | 2282 | * cpuset_hotplug_workfn - handle CPU/memory hotunplug for a cpuset |
956db3ca | 2283 | * |
deb7aa30 TH |
2284 | * This function is called after either CPU or memory configuration has |
2285 | * changed and updates cpuset accordingly. The top_cpuset is always | |
2286 | * synchronized to cpu_active_mask and N_MEMORY, which is necessary in | |
2287 | * order to make cpusets transparent (of no affect) on systems that are | |
2288 | * actively using CPU hotplug but making no active use of cpusets. | |
956db3ca | 2289 | * |
deb7aa30 | 2290 | * Non-root cpusets are only affected by offlining. If any CPUs or memory |
388afd85 LZ |
2291 | * nodes have been taken down, cpuset_hotplug_update_tasks() is invoked on |
2292 | * all descendants. | |
956db3ca | 2293 | * |
deb7aa30 TH |
2294 | * Note that CPU offlining during suspend is ignored. We don't modify |
2295 | * cpusets across suspend/resume cycles at all. | |
956db3ca | 2296 | */ |
3a5a6d0c | 2297 | static void cpuset_hotplug_workfn(struct work_struct *work) |
b1aac8bb | 2298 | { |
5c5cc623 LZ |
2299 | static cpumask_t new_cpus; |
2300 | static nodemask_t new_mems; | |
deb7aa30 | 2301 | bool cpus_updated, mems_updated; |
b8d1b8ee | 2302 | bool on_dfl = is_in_v2_mode(); |
b1aac8bb | 2303 | |
5d21cc2d | 2304 | mutex_lock(&cpuset_mutex); |
956db3ca | 2305 | |
deb7aa30 TH |
2306 | /* fetch the available cpus/mems and find out which changed how */ |
2307 | cpumask_copy(&new_cpus, cpu_active_mask); | |
2308 | new_mems = node_states[N_MEMORY]; | |
7ddf96b0 | 2309 | |
7e88291b LZ |
2310 | cpus_updated = !cpumask_equal(top_cpuset.effective_cpus, &new_cpus); |
2311 | mems_updated = !nodes_equal(top_cpuset.effective_mems, new_mems); | |
7ddf96b0 | 2312 | |
deb7aa30 TH |
2313 | /* synchronize cpus_allowed to cpu_active_mask */ |
2314 | if (cpus_updated) { | |
8447a0fe | 2315 | spin_lock_irq(&callback_lock); |
7e88291b LZ |
2316 | if (!on_dfl) |
2317 | cpumask_copy(top_cpuset.cpus_allowed, &new_cpus); | |
1344ab9c | 2318 | cpumask_copy(top_cpuset.effective_cpus, &new_cpus); |
8447a0fe | 2319 | spin_unlock_irq(&callback_lock); |
deb7aa30 TH |
2320 | /* we don't mess with cpumasks of tasks in top_cpuset */ |
2321 | } | |
b4501295 | 2322 | |
deb7aa30 TH |
2323 | /* synchronize mems_allowed to N_MEMORY */ |
2324 | if (mems_updated) { | |
8447a0fe | 2325 | spin_lock_irq(&callback_lock); |
7e88291b LZ |
2326 | if (!on_dfl) |
2327 | top_cpuset.mems_allowed = new_mems; | |
1344ab9c | 2328 | top_cpuset.effective_mems = new_mems; |
8447a0fe | 2329 | spin_unlock_irq(&callback_lock); |
d66393e5 | 2330 | update_tasks_nodemask(&top_cpuset); |
deb7aa30 | 2331 | } |
b4501295 | 2332 | |
388afd85 LZ |
2333 | mutex_unlock(&cpuset_mutex); |
2334 | ||
5c5cc623 LZ |
2335 | /* if cpus or mems changed, we need to propagate to descendants */ |
2336 | if (cpus_updated || mems_updated) { | |
deb7aa30 | 2337 | struct cpuset *cs; |
492eb21b | 2338 | struct cgroup_subsys_state *pos_css; |
f9b4fb8d | 2339 | |
fc560a26 | 2340 | rcu_read_lock(); |
492eb21b | 2341 | cpuset_for_each_descendant_pre(cs, pos_css, &top_cpuset) { |
ec903c0c | 2342 | if (cs == &top_cpuset || !css_tryget_online(&cs->css)) |
388afd85 LZ |
2343 | continue; |
2344 | rcu_read_unlock(); | |
7ddf96b0 | 2345 | |
388afd85 | 2346 | cpuset_hotplug_update_tasks(cs); |
b4501295 | 2347 | |
388afd85 LZ |
2348 | rcu_read_lock(); |
2349 | css_put(&cs->css); | |
2350 | } | |
2351 | rcu_read_unlock(); | |
2352 | } | |
8d033948 | 2353 | |
deb7aa30 | 2354 | /* rebuild sched domains if cpus_allowed has changed */ |
50e76632 PZ |
2355 | if (cpus_updated || force_rebuild) { |
2356 | force_rebuild = false; | |
e0e80a02 | 2357 | rebuild_sched_domains(); |
50e76632 | 2358 | } |
b1aac8bb PJ |
2359 | } |
2360 | ||
30e03acd | 2361 | void cpuset_update_active_cpus(void) |
4c4d50f7 | 2362 | { |
3a5a6d0c TH |
2363 | /* |
2364 | * We're inside cpu hotplug critical region which usually nests | |
2365 | * inside cgroup synchronization. Bounce actual hotplug processing | |
2366 | * to a work item to avoid reverse locking order. | |
3a5a6d0c | 2367 | */ |
3a5a6d0c | 2368 | schedule_work(&cpuset_hotplug_work); |
4c4d50f7 | 2369 | } |
4c4d50f7 | 2370 | |
50e76632 PZ |
2371 | void cpuset_wait_for_hotplug(void) |
2372 | { | |
2373 | flush_work(&cpuset_hotplug_work); | |
2374 | } | |
2375 | ||
38837fc7 | 2376 | /* |
38d7bee9 LJ |
2377 | * Keep top_cpuset.mems_allowed tracking node_states[N_MEMORY]. |
2378 | * Call this routine anytime after node_states[N_MEMORY] changes. | |
a1cd2b13 | 2379 | * See cpuset_update_active_cpus() for CPU hotplug handling. |
38837fc7 | 2380 | */ |
f481891f MX |
2381 | static int cpuset_track_online_nodes(struct notifier_block *self, |
2382 | unsigned long action, void *arg) | |
38837fc7 | 2383 | { |
3a5a6d0c | 2384 | schedule_work(&cpuset_hotplug_work); |
f481891f | 2385 | return NOTIFY_OK; |
38837fc7 | 2386 | } |
d8f10cb3 AM |
2387 | |
2388 | static struct notifier_block cpuset_track_online_nodes_nb = { | |
2389 | .notifier_call = cpuset_track_online_nodes, | |
2390 | .priority = 10, /* ??! */ | |
2391 | }; | |
38837fc7 | 2392 | |
1da177e4 LT |
2393 | /** |
2394 | * cpuset_init_smp - initialize cpus_allowed | |
2395 | * | |
2396 | * Description: Finish top cpuset after cpu, node maps are initialized | |
d8f10cb3 | 2397 | */ |
1da177e4 LT |
2398 | void __init cpuset_init_smp(void) |
2399 | { | |
6ad4c188 | 2400 | cpumask_copy(top_cpuset.cpus_allowed, cpu_active_mask); |
38d7bee9 | 2401 | top_cpuset.mems_allowed = node_states[N_MEMORY]; |
33ad801d | 2402 | top_cpuset.old_mems_allowed = top_cpuset.mems_allowed; |
4c4d50f7 | 2403 | |
e2b9a3d7 LZ |
2404 | cpumask_copy(top_cpuset.effective_cpus, cpu_active_mask); |
2405 | top_cpuset.effective_mems = node_states[N_MEMORY]; | |
2406 | ||
d8f10cb3 | 2407 | register_hotmemory_notifier(&cpuset_track_online_nodes_nb); |
e93ad19d TH |
2408 | |
2409 | cpuset_migrate_mm_wq = alloc_ordered_workqueue("cpuset_migrate_mm", 0); | |
2410 | BUG_ON(!cpuset_migrate_mm_wq); | |
1da177e4 LT |
2411 | } |
2412 | ||
2413 | /** | |
1da177e4 LT |
2414 | * cpuset_cpus_allowed - return cpus_allowed mask from a tasks cpuset. |
2415 | * @tsk: pointer to task_struct from which to obtain cpuset->cpus_allowed. | |
6af866af | 2416 | * @pmask: pointer to struct cpumask variable to receive cpus_allowed set. |
1da177e4 | 2417 | * |
300ed6cb | 2418 | * Description: Returns the cpumask_var_t cpus_allowed of the cpuset |
1da177e4 | 2419 | * attached to the specified @tsk. Guaranteed to return some non-empty |
5f054e31 | 2420 | * subset of cpu_online_mask, even if this means going outside the |
1da177e4 LT |
2421 | * tasks cpuset. |
2422 | **/ | |
2423 | ||
6af866af | 2424 | void cpuset_cpus_allowed(struct task_struct *tsk, struct cpumask *pmask) |
1da177e4 | 2425 | { |
8447a0fe VD |
2426 | unsigned long flags; |
2427 | ||
2428 | spin_lock_irqsave(&callback_lock, flags); | |
b8dadcb5 | 2429 | rcu_read_lock(); |
ae1c8023 | 2430 | guarantee_online_cpus(task_cs(tsk), pmask); |
b8dadcb5 | 2431 | rcu_read_unlock(); |
8447a0fe | 2432 | spin_unlock_irqrestore(&callback_lock, flags); |
1da177e4 LT |
2433 | } |
2434 | ||
2baab4e9 | 2435 | void cpuset_cpus_allowed_fallback(struct task_struct *tsk) |
9084bb82 | 2436 | { |
9084bb82 | 2437 | rcu_read_lock(); |
ae1c8023 | 2438 | do_set_cpus_allowed(tsk, task_cs(tsk)->effective_cpus); |
9084bb82 ON |
2439 | rcu_read_unlock(); |
2440 | ||
2441 | /* | |
2442 | * We own tsk->cpus_allowed, nobody can change it under us. | |
2443 | * | |
2444 | * But we used cs && cs->cpus_allowed lockless and thus can | |
2445 | * race with cgroup_attach_task() or update_cpumask() and get | |
2446 | * the wrong tsk->cpus_allowed. However, both cases imply the | |
2447 | * subsequent cpuset_change_cpumask()->set_cpus_allowed_ptr() | |
2448 | * which takes task_rq_lock(). | |
2449 | * | |
2450 | * If we are called after it dropped the lock we must see all | |
2451 | * changes in tsk_cs()->cpus_allowed. Otherwise we can temporary | |
2452 | * set any mask even if it is not right from task_cs() pov, | |
2453 | * the pending set_cpus_allowed_ptr() will fix things. | |
2baab4e9 PZ |
2454 | * |
2455 | * select_fallback_rq() will fix things ups and set cpu_possible_mask | |
2456 | * if required. | |
9084bb82 | 2457 | */ |
9084bb82 ON |
2458 | } |
2459 | ||
8f4ab07f | 2460 | void __init cpuset_init_current_mems_allowed(void) |
1da177e4 | 2461 | { |
f9a86fcb | 2462 | nodes_setall(current->mems_allowed); |
1da177e4 LT |
2463 | } |
2464 | ||
909d75a3 PJ |
2465 | /** |
2466 | * cpuset_mems_allowed - return mems_allowed mask from a tasks cpuset. | |
2467 | * @tsk: pointer to task_struct from which to obtain cpuset->mems_allowed. | |
2468 | * | |
2469 | * Description: Returns the nodemask_t mems_allowed of the cpuset | |
2470 | * attached to the specified @tsk. Guaranteed to return some non-empty | |
38d7bee9 | 2471 | * subset of node_states[N_MEMORY], even if this means going outside the |
909d75a3 PJ |
2472 | * tasks cpuset. |
2473 | **/ | |
2474 | ||
2475 | nodemask_t cpuset_mems_allowed(struct task_struct *tsk) | |
2476 | { | |
2477 | nodemask_t mask; | |
8447a0fe | 2478 | unsigned long flags; |
909d75a3 | 2479 | |
8447a0fe | 2480 | spin_lock_irqsave(&callback_lock, flags); |
b8dadcb5 | 2481 | rcu_read_lock(); |
ae1c8023 | 2482 | guarantee_online_mems(task_cs(tsk), &mask); |
b8dadcb5 | 2483 | rcu_read_unlock(); |
8447a0fe | 2484 | spin_unlock_irqrestore(&callback_lock, flags); |
909d75a3 PJ |
2485 | |
2486 | return mask; | |
2487 | } | |
2488 | ||
d9fd8a6d | 2489 | /** |
19770b32 MG |
2490 | * cpuset_nodemask_valid_mems_allowed - check nodemask vs. curremt mems_allowed |
2491 | * @nodemask: the nodemask to be checked | |
d9fd8a6d | 2492 | * |
19770b32 | 2493 | * Are any of the nodes in the nodemask allowed in current->mems_allowed? |
1da177e4 | 2494 | */ |
19770b32 | 2495 | int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask) |
1da177e4 | 2496 | { |
19770b32 | 2497 | return nodes_intersects(*nodemask, current->mems_allowed); |
1da177e4 LT |
2498 | } |
2499 | ||
9bf2229f | 2500 | /* |
78608366 PM |
2501 | * nearest_hardwall_ancestor() - Returns the nearest mem_exclusive or |
2502 | * mem_hardwall ancestor to the specified cpuset. Call holding | |
8447a0fe | 2503 | * callback_lock. If no ancestor is mem_exclusive or mem_hardwall |
78608366 | 2504 | * (an unusual configuration), then returns the root cpuset. |
9bf2229f | 2505 | */ |
c9710d80 | 2506 | static struct cpuset *nearest_hardwall_ancestor(struct cpuset *cs) |
9bf2229f | 2507 | { |
c431069f TH |
2508 | while (!(is_mem_exclusive(cs) || is_mem_hardwall(cs)) && parent_cs(cs)) |
2509 | cs = parent_cs(cs); | |
9bf2229f PJ |
2510 | return cs; |
2511 | } | |
2512 | ||
d9fd8a6d | 2513 | /** |
344736f2 | 2514 | * cpuset_node_allowed - Can we allocate on a memory node? |
a1bc5a4e | 2515 | * @node: is this an allowed node? |
02a0e53d | 2516 | * @gfp_mask: memory allocation flags |
d9fd8a6d | 2517 | * |
6e276d2a DR |
2518 | * If we're in interrupt, yes, we can always allocate. If @node is set in |
2519 | * current's mems_allowed, yes. If it's not a __GFP_HARDWALL request and this | |
2520 | * node is set in the nearest hardwalled cpuset ancestor to current's cpuset, | |
da99ecf1 | 2521 | * yes. If current has access to memory reserves as an oom victim, yes. |
9bf2229f PJ |
2522 | * Otherwise, no. |
2523 | * | |
2524 | * GFP_USER allocations are marked with the __GFP_HARDWALL bit, | |
c596d9f3 | 2525 | * and do not allow allocations outside the current tasks cpuset |
da99ecf1 | 2526 | * unless the task has been OOM killed. |
9bf2229f | 2527 | * GFP_KERNEL allocations are not so marked, so can escape to the |
78608366 | 2528 | * nearest enclosing hardwalled ancestor cpuset. |
9bf2229f | 2529 | * |
8447a0fe | 2530 | * Scanning up parent cpusets requires callback_lock. The |
02a0e53d PJ |
2531 | * __alloc_pages() routine only calls here with __GFP_HARDWALL bit |
2532 | * _not_ set if it's a GFP_KERNEL allocation, and all nodes in the | |
2533 | * current tasks mems_allowed came up empty on the first pass over | |
2534 | * the zonelist. So only GFP_KERNEL allocations, if all nodes in the | |
8447a0fe | 2535 | * cpuset are short of memory, might require taking the callback_lock. |
9bf2229f | 2536 | * |
36be57ff | 2537 | * The first call here from mm/page_alloc:get_page_from_freelist() |
02a0e53d PJ |
2538 | * has __GFP_HARDWALL set in gfp_mask, enforcing hardwall cpusets, |
2539 | * so no allocation on a node outside the cpuset is allowed (unless | |
2540 | * in interrupt, of course). | |
36be57ff PJ |
2541 | * |
2542 | * The second pass through get_page_from_freelist() doesn't even call | |
2543 | * here for GFP_ATOMIC calls. For those calls, the __alloc_pages() | |
2544 | * variable 'wait' is not set, and the bit ALLOC_CPUSET is not set | |
2545 | * in alloc_flags. That logic and the checks below have the combined | |
2546 | * affect that: | |
9bf2229f PJ |
2547 | * in_interrupt - any node ok (current task context irrelevant) |
2548 | * GFP_ATOMIC - any node ok | |
da99ecf1 | 2549 | * tsk_is_oom_victim - any node ok |
78608366 | 2550 | * GFP_KERNEL - any node in enclosing hardwalled cpuset ok |
9bf2229f | 2551 | * GFP_USER - only nodes in current tasks mems allowed ok. |
02a0e53d | 2552 | */ |
002f2906 | 2553 | bool __cpuset_node_allowed(int node, gfp_t gfp_mask) |
1da177e4 | 2554 | { |
c9710d80 | 2555 | struct cpuset *cs; /* current cpuset ancestors */ |
29afd49b | 2556 | int allowed; /* is allocation in zone z allowed? */ |
8447a0fe | 2557 | unsigned long flags; |
9bf2229f | 2558 | |
6e276d2a | 2559 | if (in_interrupt()) |
002f2906 | 2560 | return true; |
9bf2229f | 2561 | if (node_isset(node, current->mems_allowed)) |
002f2906 | 2562 | return true; |
c596d9f3 DR |
2563 | /* |
2564 | * Allow tasks that have access to memory reserves because they have | |
2565 | * been OOM killed to get memory anywhere. | |
2566 | */ | |
da99ecf1 | 2567 | if (unlikely(tsk_is_oom_victim(current))) |
002f2906 | 2568 | return true; |
9bf2229f | 2569 | if (gfp_mask & __GFP_HARDWALL) /* If hardwall request, stop here */ |
002f2906 | 2570 | return false; |
9bf2229f | 2571 | |
5563e770 | 2572 | if (current->flags & PF_EXITING) /* Let dying task have memory */ |
002f2906 | 2573 | return true; |
5563e770 | 2574 | |
9bf2229f | 2575 | /* Not hardwall and node outside mems_allowed: scan up cpusets */ |
8447a0fe | 2576 | spin_lock_irqsave(&callback_lock, flags); |
053199ed | 2577 | |
b8dadcb5 | 2578 | rcu_read_lock(); |
78608366 | 2579 | cs = nearest_hardwall_ancestor(task_cs(current)); |
99afb0fd | 2580 | allowed = node_isset(node, cs->mems_allowed); |
b8dadcb5 | 2581 | rcu_read_unlock(); |
053199ed | 2582 | |
8447a0fe | 2583 | spin_unlock_irqrestore(&callback_lock, flags); |
9bf2229f | 2584 | return allowed; |
1da177e4 LT |
2585 | } |
2586 | ||
825a46af | 2587 | /** |
6adef3eb JS |
2588 | * cpuset_mem_spread_node() - On which node to begin search for a file page |
2589 | * cpuset_slab_spread_node() - On which node to begin search for a slab page | |
825a46af PJ |
2590 | * |
2591 | * If a task is marked PF_SPREAD_PAGE or PF_SPREAD_SLAB (as for | |
2592 | * tasks in a cpuset with is_spread_page or is_spread_slab set), | |
2593 | * and if the memory allocation used cpuset_mem_spread_node() | |
2594 | * to determine on which node to start looking, as it will for | |
2595 | * certain page cache or slab cache pages such as used for file | |
2596 | * system buffers and inode caches, then instead of starting on the | |
2597 | * local node to look for a free page, rather spread the starting | |
2598 | * node around the tasks mems_allowed nodes. | |
2599 | * | |
2600 | * We don't have to worry about the returned node being offline | |
2601 | * because "it can't happen", and even if it did, it would be ok. | |
2602 | * | |
2603 | * The routines calling guarantee_online_mems() are careful to | |
2604 | * only set nodes in task->mems_allowed that are online. So it | |
2605 | * should not be possible for the following code to return an | |
2606 | * offline node. But if it did, that would be ok, as this routine | |
2607 | * is not returning the node where the allocation must be, only | |
2608 | * the node where the search should start. The zonelist passed to | |
2609 | * __alloc_pages() will include all nodes. If the slab allocator | |
2610 | * is passed an offline node, it will fall back to the local node. | |
2611 | * See kmem_cache_alloc_node(). | |
2612 | */ | |
2613 | ||
6adef3eb | 2614 | static int cpuset_spread_node(int *rotor) |
825a46af | 2615 | { |
0edaf86c | 2616 | return *rotor = next_node_in(*rotor, current->mems_allowed); |
825a46af | 2617 | } |
6adef3eb JS |
2618 | |
2619 | int cpuset_mem_spread_node(void) | |
2620 | { | |
778d3b0f MH |
2621 | if (current->cpuset_mem_spread_rotor == NUMA_NO_NODE) |
2622 | current->cpuset_mem_spread_rotor = | |
2623 | node_random(¤t->mems_allowed); | |
2624 | ||
6adef3eb JS |
2625 | return cpuset_spread_node(¤t->cpuset_mem_spread_rotor); |
2626 | } | |
2627 | ||
2628 | int cpuset_slab_spread_node(void) | |
2629 | { | |
778d3b0f MH |
2630 | if (current->cpuset_slab_spread_rotor == NUMA_NO_NODE) |
2631 | current->cpuset_slab_spread_rotor = | |
2632 | node_random(¤t->mems_allowed); | |
2633 | ||
6adef3eb JS |
2634 | return cpuset_spread_node(¤t->cpuset_slab_spread_rotor); |
2635 | } | |
2636 | ||
825a46af PJ |
2637 | EXPORT_SYMBOL_GPL(cpuset_mem_spread_node); |
2638 | ||
ef08e3b4 | 2639 | /** |
bbe373f2 DR |
2640 | * cpuset_mems_allowed_intersects - Does @tsk1's mems_allowed intersect @tsk2's? |
2641 | * @tsk1: pointer to task_struct of some task. | |
2642 | * @tsk2: pointer to task_struct of some other task. | |
2643 | * | |
2644 | * Description: Return true if @tsk1's mems_allowed intersects the | |
2645 | * mems_allowed of @tsk2. Used by the OOM killer to determine if | |
2646 | * one of the task's memory usage might impact the memory available | |
2647 | * to the other. | |
ef08e3b4 PJ |
2648 | **/ |
2649 | ||
bbe373f2 DR |
2650 | int cpuset_mems_allowed_intersects(const struct task_struct *tsk1, |
2651 | const struct task_struct *tsk2) | |
ef08e3b4 | 2652 | { |
bbe373f2 | 2653 | return nodes_intersects(tsk1->mems_allowed, tsk2->mems_allowed); |
ef08e3b4 PJ |
2654 | } |
2655 | ||
75aa1994 | 2656 | /** |
da39da3a | 2657 | * cpuset_print_current_mems_allowed - prints current's cpuset and mems_allowed |
75aa1994 | 2658 | * |
da39da3a | 2659 | * Description: Prints current's name, cpuset name, and cached copy of its |
b8dadcb5 | 2660 | * mems_allowed to the kernel log. |
75aa1994 | 2661 | */ |
da39da3a | 2662 | void cpuset_print_current_mems_allowed(void) |
75aa1994 | 2663 | { |
b8dadcb5 | 2664 | struct cgroup *cgrp; |
75aa1994 | 2665 | |
b8dadcb5 | 2666 | rcu_read_lock(); |
63f43f55 | 2667 | |
da39da3a DR |
2668 | cgrp = task_cs(current)->css.cgroup; |
2669 | pr_info("%s cpuset=", current->comm); | |
e61734c5 | 2670 | pr_cont_cgroup_name(cgrp); |
da39da3a DR |
2671 | pr_cont(" mems_allowed=%*pbl\n", |
2672 | nodemask_pr_args(¤t->mems_allowed)); | |
f440d98f | 2673 | |
cfb5966b | 2674 | rcu_read_unlock(); |
75aa1994 DR |
2675 | } |
2676 | ||
3e0d98b9 PJ |
2677 | /* |
2678 | * Collection of memory_pressure is suppressed unless | |
2679 | * this flag is enabled by writing "1" to the special | |
2680 | * cpuset file 'memory_pressure_enabled' in the root cpuset. | |
2681 | */ | |
2682 | ||
c5b2aff8 | 2683 | int cpuset_memory_pressure_enabled __read_mostly; |
3e0d98b9 PJ |
2684 | |
2685 | /** | |
2686 | * cpuset_memory_pressure_bump - keep stats of per-cpuset reclaims. | |
2687 | * | |
2688 | * Keep a running average of the rate of synchronous (direct) | |
2689 | * page reclaim efforts initiated by tasks in each cpuset. | |
2690 | * | |
2691 | * This represents the rate at which some task in the cpuset | |
2692 | * ran low on memory on all nodes it was allowed to use, and | |
2693 | * had to enter the kernels page reclaim code in an effort to | |
2694 | * create more free memory by tossing clean pages or swapping | |
2695 | * or writing dirty pages. | |
2696 | * | |
2697 | * Display to user space in the per-cpuset read-only file | |
2698 | * "memory_pressure". Value displayed is an integer | |
2699 | * representing the recent rate of entry into the synchronous | |
2700 | * (direct) page reclaim by any task attached to the cpuset. | |
2701 | **/ | |
2702 | ||
2703 | void __cpuset_memory_pressure_bump(void) | |
2704 | { | |
b8dadcb5 | 2705 | rcu_read_lock(); |
8793d854 | 2706 | fmeter_markevent(&task_cs(current)->fmeter); |
b8dadcb5 | 2707 | rcu_read_unlock(); |
3e0d98b9 PJ |
2708 | } |
2709 | ||
8793d854 | 2710 | #ifdef CONFIG_PROC_PID_CPUSET |
1da177e4 LT |
2711 | /* |
2712 | * proc_cpuset_show() | |
2713 | * - Print tasks cpuset path into seq_file. | |
2714 | * - Used for /proc/<pid>/cpuset. | |
053199ed PJ |
2715 | * - No need to task_lock(tsk) on this tsk->cpuset reference, as it |
2716 | * doesn't really matter if tsk->cpuset changes after we read it, | |
5d21cc2d | 2717 | * and we take cpuset_mutex, keeping cpuset_attach() from changing it |
2df167a3 | 2718 | * anyway. |
1da177e4 | 2719 | */ |
52de4779 ZL |
2720 | int proc_cpuset_show(struct seq_file *m, struct pid_namespace *ns, |
2721 | struct pid *pid, struct task_struct *tsk) | |
1da177e4 | 2722 | { |
4c737b41 | 2723 | char *buf; |
8793d854 | 2724 | struct cgroup_subsys_state *css; |
99f89551 | 2725 | int retval; |
1da177e4 | 2726 | |
99f89551 | 2727 | retval = -ENOMEM; |
e61734c5 | 2728 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
1da177e4 | 2729 | if (!buf) |
99f89551 EB |
2730 | goto out; |
2731 | ||
a79a908f | 2732 | css = task_get_css(tsk, cpuset_cgrp_id); |
4c737b41 TH |
2733 | retval = cgroup_path_ns(css->cgroup, buf, PATH_MAX, |
2734 | current->nsproxy->cgroup_ns); | |
a79a908f | 2735 | css_put(css); |
4c737b41 | 2736 | if (retval >= PATH_MAX) |
679a5e3f TH |
2737 | retval = -ENAMETOOLONG; |
2738 | if (retval < 0) | |
52de4779 | 2739 | goto out_free; |
4c737b41 | 2740 | seq_puts(m, buf); |
1da177e4 | 2741 | seq_putc(m, '\n'); |
e61734c5 | 2742 | retval = 0; |
99f89551 | 2743 | out_free: |
1da177e4 | 2744 | kfree(buf); |
99f89551 | 2745 | out: |
1da177e4 LT |
2746 | return retval; |
2747 | } | |
8793d854 | 2748 | #endif /* CONFIG_PROC_PID_CPUSET */ |
1da177e4 | 2749 | |
d01d4827 | 2750 | /* Display task mems_allowed in /proc/<pid>/status file. */ |
df5f8314 EB |
2751 | void cpuset_task_status_allowed(struct seq_file *m, struct task_struct *task) |
2752 | { | |
e8e6d97c TH |
2753 | seq_printf(m, "Mems_allowed:\t%*pb\n", |
2754 | nodemask_pr_args(&task->mems_allowed)); | |
2755 | seq_printf(m, "Mems_allowed_list:\t%*pbl\n", | |
2756 | nodemask_pr_args(&task->mems_allowed)); | |
1da177e4 | 2757 | } |