]>
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> | |
59 | ||
7c0f6ba6 | 60 | #include <linux/uaccess.h> |
60063497 | 61 | #include <linux/atomic.h> |
3d3f26a7 | 62 | #include <linux/mutex.h> |
956db3ca | 63 | #include <linux/cgroup.h> |
e44193d3 | 64 | #include <linux/wait.h> |
1da177e4 | 65 | |
89affbf5 | 66 | DEFINE_STATIC_KEY_FALSE(cpusets_pre_enable_key); |
002f2906 | 67 | DEFINE_STATIC_KEY_FALSE(cpusets_enabled_key); |
202f72d5 | 68 | |
3e0d98b9 PJ |
69 | /* See "Frequency meter" comments, below. */ |
70 | ||
71 | struct fmeter { | |
72 | int cnt; /* unprocessed events count */ | |
73 | int val; /* most recent output value */ | |
d2b43658 | 74 | time64_t time; /* clock (secs) when val computed */ |
3e0d98b9 PJ |
75 | spinlock_t lock; /* guards read or write of above */ |
76 | }; | |
77 | ||
1da177e4 | 78 | struct cpuset { |
8793d854 PM |
79 | struct cgroup_subsys_state css; |
80 | ||
1da177e4 | 81 | unsigned long flags; /* "unsigned long" so bitops work */ |
e2b9a3d7 | 82 | |
7e88291b LZ |
83 | /* |
84 | * On default hierarchy: | |
85 | * | |
86 | * The user-configured masks can only be changed by writing to | |
87 | * cpuset.cpus and cpuset.mems, and won't be limited by the | |
88 | * parent masks. | |
89 | * | |
90 | * The effective masks is the real masks that apply to the tasks | |
91 | * in the cpuset. They may be changed if the configured masks are | |
92 | * changed or hotplug happens. | |
93 | * | |
94 | * effective_mask == configured_mask & parent's effective_mask, | |
95 | * and if it ends up empty, it will inherit the parent's mask. | |
96 | * | |
97 | * | |
98 | * On legacy hierachy: | |
99 | * | |
100 | * The user-configured masks are always the same with effective masks. | |
101 | */ | |
102 | ||
e2b9a3d7 LZ |
103 | /* user-configured CPUs and Memory Nodes allow to tasks */ |
104 | cpumask_var_t cpus_allowed; | |
105 | nodemask_t mems_allowed; | |
106 | ||
107 | /* effective CPUs and Memory Nodes allow to tasks */ | |
108 | cpumask_var_t effective_cpus; | |
109 | nodemask_t effective_mems; | |
1da177e4 | 110 | |
33ad801d LZ |
111 | /* |
112 | * This is old Memory Nodes tasks took on. | |
113 | * | |
114 | * - top_cpuset.old_mems_allowed is initialized to mems_allowed. | |
115 | * - A new cpuset's old_mems_allowed is initialized when some | |
116 | * task is moved into it. | |
117 | * - old_mems_allowed is used in cpuset_migrate_mm() when we change | |
118 | * cpuset.mems_allowed and have tasks' nodemask updated, and | |
119 | * then old_mems_allowed is updated to mems_allowed. | |
120 | */ | |
121 | nodemask_t old_mems_allowed; | |
122 | ||
3e0d98b9 | 123 | struct fmeter fmeter; /* memory_pressure filter */ |
029190c5 | 124 | |
452477fa TH |
125 | /* |
126 | * Tasks are being attached to this cpuset. Used to prevent | |
127 | * zeroing cpus/mems_allowed between ->can_attach() and ->attach(). | |
128 | */ | |
129 | int attach_in_progress; | |
130 | ||
029190c5 PJ |
131 | /* partition number for rebuild_sched_domains() */ |
132 | int pn; | |
956db3ca | 133 | |
1d3504fc HS |
134 | /* for custom sched domain */ |
135 | int relax_domain_level; | |
1da177e4 LT |
136 | }; |
137 | ||
a7c6d554 | 138 | static inline struct cpuset *css_cs(struct cgroup_subsys_state *css) |
8793d854 | 139 | { |
a7c6d554 | 140 | return css ? container_of(css, struct cpuset, css) : NULL; |
8793d854 PM |
141 | } |
142 | ||
143 | /* Retrieve the cpuset for a task */ | |
144 | static inline struct cpuset *task_cs(struct task_struct *task) | |
145 | { | |
073219e9 | 146 | return css_cs(task_css(task, cpuset_cgrp_id)); |
8793d854 | 147 | } |
8793d854 | 148 | |
c9710d80 | 149 | static inline struct cpuset *parent_cs(struct cpuset *cs) |
c431069f | 150 | { |
5c9d535b | 151 | return css_cs(cs->css.parent); |
c431069f TH |
152 | } |
153 | ||
b246272e DR |
154 | #ifdef CONFIG_NUMA |
155 | static inline bool task_has_mempolicy(struct task_struct *task) | |
156 | { | |
157 | return task->mempolicy; | |
158 | } | |
159 | #else | |
160 | static inline bool task_has_mempolicy(struct task_struct *task) | |
161 | { | |
162 | return false; | |
163 | } | |
164 | #endif | |
165 | ||
166 | ||
1da177e4 LT |
167 | /* bits in struct cpuset flags field */ |
168 | typedef enum { | |
efeb77b2 | 169 | CS_ONLINE, |
1da177e4 LT |
170 | CS_CPU_EXCLUSIVE, |
171 | CS_MEM_EXCLUSIVE, | |
78608366 | 172 | CS_MEM_HARDWALL, |
45b07ef3 | 173 | CS_MEMORY_MIGRATE, |
029190c5 | 174 | CS_SCHED_LOAD_BALANCE, |
825a46af PJ |
175 | CS_SPREAD_PAGE, |
176 | CS_SPREAD_SLAB, | |
1da177e4 LT |
177 | } cpuset_flagbits_t; |
178 | ||
179 | /* convenient tests for these bits */ | |
41c25707 | 180 | static inline bool is_cpuset_online(struct cpuset *cs) |
efeb77b2 | 181 | { |
41c25707 | 182 | return test_bit(CS_ONLINE, &cs->flags) && !css_is_dying(&cs->css); |
efeb77b2 TH |
183 | } |
184 | ||
1da177e4 LT |
185 | static inline int is_cpu_exclusive(const struct cpuset *cs) |
186 | { | |
7b5b9ef0 | 187 | return test_bit(CS_CPU_EXCLUSIVE, &cs->flags); |
1da177e4 LT |
188 | } |
189 | ||
190 | static inline int is_mem_exclusive(const struct cpuset *cs) | |
191 | { | |
7b5b9ef0 | 192 | return test_bit(CS_MEM_EXCLUSIVE, &cs->flags); |
1da177e4 LT |
193 | } |
194 | ||
78608366 PM |
195 | static inline int is_mem_hardwall(const struct cpuset *cs) |
196 | { | |
197 | return test_bit(CS_MEM_HARDWALL, &cs->flags); | |
198 | } | |
199 | ||
029190c5 PJ |
200 | static inline int is_sched_load_balance(const struct cpuset *cs) |
201 | { | |
202 | return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); | |
203 | } | |
204 | ||
45b07ef3 PJ |
205 | static inline int is_memory_migrate(const struct cpuset *cs) |
206 | { | |
7b5b9ef0 | 207 | return test_bit(CS_MEMORY_MIGRATE, &cs->flags); |
45b07ef3 PJ |
208 | } |
209 | ||
825a46af PJ |
210 | static inline int is_spread_page(const struct cpuset *cs) |
211 | { | |
212 | return test_bit(CS_SPREAD_PAGE, &cs->flags); | |
213 | } | |
214 | ||
215 | static inline int is_spread_slab(const struct cpuset *cs) | |
216 | { | |
217 | return test_bit(CS_SPREAD_SLAB, &cs->flags); | |
218 | } | |
219 | ||
1da177e4 | 220 | static struct cpuset top_cpuset = { |
efeb77b2 TH |
221 | .flags = ((1 << CS_ONLINE) | (1 << CS_CPU_EXCLUSIVE) | |
222 | (1 << CS_MEM_EXCLUSIVE)), | |
1da177e4 LT |
223 | }; |
224 | ||
ae8086ce TH |
225 | /** |
226 | * cpuset_for_each_child - traverse online children of a cpuset | |
227 | * @child_cs: loop cursor pointing to the current child | |
492eb21b | 228 | * @pos_css: used for iteration |
ae8086ce TH |
229 | * @parent_cs: target cpuset to walk children of |
230 | * | |
231 | * Walk @child_cs through the online children of @parent_cs. Must be used | |
232 | * with RCU read locked. | |
233 | */ | |
492eb21b TH |
234 | #define cpuset_for_each_child(child_cs, pos_css, parent_cs) \ |
235 | css_for_each_child((pos_css), &(parent_cs)->css) \ | |
236 | if (is_cpuset_online(((child_cs) = css_cs((pos_css))))) | |
ae8086ce | 237 | |
fc560a26 TH |
238 | /** |
239 | * cpuset_for_each_descendant_pre - pre-order walk of a cpuset's descendants | |
240 | * @des_cs: loop cursor pointing to the current descendant | |
492eb21b | 241 | * @pos_css: used for iteration |
fc560a26 TH |
242 | * @root_cs: target cpuset to walk ancestor of |
243 | * | |
244 | * Walk @des_cs through the online descendants of @root_cs. Must be used | |
492eb21b | 245 | * with RCU read locked. The caller may modify @pos_css by calling |
bd8815a6 TH |
246 | * css_rightmost_descendant() to skip subtree. @root_cs is included in the |
247 | * iteration and the first node to be visited. | |
fc560a26 | 248 | */ |
492eb21b TH |
249 | #define cpuset_for_each_descendant_pre(des_cs, pos_css, root_cs) \ |
250 | css_for_each_descendant_pre((pos_css), &(root_cs)->css) \ | |
251 | if (is_cpuset_online(((des_cs) = css_cs((pos_css))))) | |
fc560a26 | 252 | |
1da177e4 | 253 | /* |
8447a0fe VD |
254 | * There are two global locks guarding cpuset structures - cpuset_mutex and |
255 | * callback_lock. We also require taking task_lock() when dereferencing a | |
256 | * task's cpuset pointer. See "The task_lock() exception", at the end of this | |
257 | * comment. | |
5d21cc2d | 258 | * |
8447a0fe | 259 | * A task must hold both locks to modify cpusets. If a task holds |
5d21cc2d | 260 | * cpuset_mutex, then it blocks others wanting that mutex, ensuring that it |
8447a0fe | 261 | * is the only task able to also acquire callback_lock and be able to |
5d21cc2d TH |
262 | * modify cpusets. It can perform various checks on the cpuset structure |
263 | * first, knowing nothing will change. It can also allocate memory while | |
264 | * just holding cpuset_mutex. While it is performing these checks, various | |
8447a0fe VD |
265 | * callback routines can briefly acquire callback_lock to query cpusets. |
266 | * Once it is ready to make the changes, it takes callback_lock, blocking | |
5d21cc2d | 267 | * everyone else. |
053199ed PJ |
268 | * |
269 | * Calls to the kernel memory allocator can not be made while holding | |
8447a0fe | 270 | * callback_lock, as that would risk double tripping on callback_lock |
053199ed PJ |
271 | * from one of the callbacks into the cpuset code from within |
272 | * __alloc_pages(). | |
273 | * | |
8447a0fe | 274 | * If a task is only holding callback_lock, then it has read-only |
053199ed PJ |
275 | * access to cpusets. |
276 | * | |
58568d2a MX |
277 | * Now, the task_struct fields mems_allowed and mempolicy may be changed |
278 | * by other task, we use alloc_lock in the task_struct fields to protect | |
279 | * them. | |
053199ed | 280 | * |
8447a0fe | 281 | * The cpuset_common_file_read() handlers only hold callback_lock across |
053199ed PJ |
282 | * small pieces of code, such as when reading out possibly multi-word |
283 | * cpumasks and nodemasks. | |
284 | * | |
2df167a3 PM |
285 | * Accessing a task's cpuset should be done in accordance with the |
286 | * guidelines for accessing subsystem state in kernel/cgroup.c | |
1da177e4 LT |
287 | */ |
288 | ||
5d21cc2d | 289 | static DEFINE_MUTEX(cpuset_mutex); |
8447a0fe | 290 | static DEFINE_SPINLOCK(callback_lock); |
4247bdc6 | 291 | |
e93ad19d TH |
292 | static struct workqueue_struct *cpuset_migrate_mm_wq; |
293 | ||
3a5a6d0c TH |
294 | /* |
295 | * CPU / memory hotplug is handled asynchronously. | |
296 | */ | |
297 | static void cpuset_hotplug_workfn(struct work_struct *work); | |
3a5a6d0c TH |
298 | static DECLARE_WORK(cpuset_hotplug_work, cpuset_hotplug_workfn); |
299 | ||
e44193d3 LZ |
300 | static DECLARE_WAIT_QUEUE_HEAD(cpuset_attach_wq); |
301 | ||
cf417141 MK |
302 | /* |
303 | * This is ugly, but preserves the userspace API for existing cpuset | |
8793d854 | 304 | * users. If someone tries to mount the "cpuset" filesystem, we |
cf417141 MK |
305 | * silently switch it to mount "cgroup" instead |
306 | */ | |
f7e83571 AV |
307 | static struct dentry *cpuset_mount(struct file_system_type *fs_type, |
308 | int flags, const char *unused_dev_name, void *data) | |
1da177e4 | 309 | { |
8793d854 | 310 | struct file_system_type *cgroup_fs = get_fs_type("cgroup"); |
f7e83571 | 311 | struct dentry *ret = ERR_PTR(-ENODEV); |
8793d854 PM |
312 | if (cgroup_fs) { |
313 | char mountopts[] = | |
314 | "cpuset,noprefix," | |
315 | "release_agent=/sbin/cpuset_release_agent"; | |
f7e83571 AV |
316 | ret = cgroup_fs->mount(cgroup_fs, flags, |
317 | unused_dev_name, mountopts); | |
8793d854 PM |
318 | put_filesystem(cgroup_fs); |
319 | } | |
320 | return ret; | |
1da177e4 LT |
321 | } |
322 | ||
323 | static struct file_system_type cpuset_fs_type = { | |
324 | .name = "cpuset", | |
f7e83571 | 325 | .mount = cpuset_mount, |
1da177e4 LT |
326 | }; |
327 | ||
1da177e4 | 328 | /* |
300ed6cb | 329 | * Return in pmask the portion of a cpusets's cpus_allowed that |
1da177e4 | 330 | * are online. If none are online, walk up the cpuset hierarchy |
28b89b9e | 331 | * until we find one that does have some online cpus. |
1da177e4 LT |
332 | * |
333 | * One way or another, we guarantee to return some non-empty subset | |
5f054e31 | 334 | * of cpu_online_mask. |
1da177e4 | 335 | * |
8447a0fe | 336 | * Call with callback_lock or cpuset_mutex held. |
1da177e4 | 337 | */ |
c9710d80 | 338 | static void guarantee_online_cpus(struct cpuset *cs, struct cpumask *pmask) |
1da177e4 | 339 | { |
28b89b9e | 340 | while (!cpumask_intersects(cs->effective_cpus, cpu_online_mask)) { |
c431069f | 341 | cs = parent_cs(cs); |
28b89b9e JP |
342 | if (unlikely(!cs)) { |
343 | /* | |
344 | * The top cpuset doesn't have any online cpu as a | |
345 | * consequence of a race between cpuset_hotplug_work | |
346 | * and cpu hotplug notifier. But we know the top | |
347 | * cpuset's effective_cpus is on its way to to be | |
348 | * identical to cpu_online_mask. | |
349 | */ | |
350 | cpumask_copy(pmask, cpu_online_mask); | |
351 | return; | |
352 | } | |
353 | } | |
ae1c8023 | 354 | cpumask_and(pmask, cs->effective_cpus, cpu_online_mask); |
1da177e4 LT |
355 | } |
356 | ||
357 | /* | |
358 | * Return in *pmask the portion of a cpusets's mems_allowed that | |
0e1e7c7a CL |
359 | * are online, with memory. If none are online with memory, walk |
360 | * up the cpuset hierarchy until we find one that does have some | |
40df2deb | 361 | * online mems. The top cpuset always has some mems online. |
1da177e4 LT |
362 | * |
363 | * One way or another, we guarantee to return some non-empty subset | |
38d7bee9 | 364 | * of node_states[N_MEMORY]. |
1da177e4 | 365 | * |
8447a0fe | 366 | * Call with callback_lock or cpuset_mutex held. |
1da177e4 | 367 | */ |
c9710d80 | 368 | static void guarantee_online_mems(struct cpuset *cs, nodemask_t *pmask) |
1da177e4 | 369 | { |
ae1c8023 | 370 | while (!nodes_intersects(cs->effective_mems, node_states[N_MEMORY])) |
c431069f | 371 | cs = parent_cs(cs); |
ae1c8023 | 372 | nodes_and(*pmask, cs->effective_mems, node_states[N_MEMORY]); |
1da177e4 LT |
373 | } |
374 | ||
f3b39d47 MX |
375 | /* |
376 | * update task's spread flag if cpuset's page/slab spread flag is set | |
377 | * | |
8447a0fe | 378 | * Call with callback_lock or cpuset_mutex held. |
f3b39d47 MX |
379 | */ |
380 | static void cpuset_update_task_spread_flag(struct cpuset *cs, | |
381 | struct task_struct *tsk) | |
382 | { | |
383 | if (is_spread_page(cs)) | |
2ad654bc | 384 | task_set_spread_page(tsk); |
f3b39d47 | 385 | else |
2ad654bc ZL |
386 | task_clear_spread_page(tsk); |
387 | ||
f3b39d47 | 388 | if (is_spread_slab(cs)) |
2ad654bc | 389 | task_set_spread_slab(tsk); |
f3b39d47 | 390 | else |
2ad654bc | 391 | task_clear_spread_slab(tsk); |
f3b39d47 MX |
392 | } |
393 | ||
1da177e4 LT |
394 | /* |
395 | * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q? | |
396 | * | |
397 | * One cpuset is a subset of another if all its allowed CPUs and | |
398 | * Memory Nodes are a subset of the other, and its exclusive flags | |
5d21cc2d | 399 | * are only set if the other's are set. Call holding cpuset_mutex. |
1da177e4 LT |
400 | */ |
401 | ||
402 | static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q) | |
403 | { | |
300ed6cb | 404 | return cpumask_subset(p->cpus_allowed, q->cpus_allowed) && |
1da177e4 LT |
405 | nodes_subset(p->mems_allowed, q->mems_allowed) && |
406 | is_cpu_exclusive(p) <= is_cpu_exclusive(q) && | |
407 | is_mem_exclusive(p) <= is_mem_exclusive(q); | |
408 | } | |
409 | ||
645fcc9d LZ |
410 | /** |
411 | * alloc_trial_cpuset - allocate a trial cpuset | |
412 | * @cs: the cpuset that the trial cpuset duplicates | |
413 | */ | |
c9710d80 | 414 | static struct cpuset *alloc_trial_cpuset(struct cpuset *cs) |
645fcc9d | 415 | { |
300ed6cb LZ |
416 | struct cpuset *trial; |
417 | ||
418 | trial = kmemdup(cs, sizeof(*cs), GFP_KERNEL); | |
419 | if (!trial) | |
420 | return NULL; | |
421 | ||
e2b9a3d7 LZ |
422 | if (!alloc_cpumask_var(&trial->cpus_allowed, GFP_KERNEL)) |
423 | goto free_cs; | |
424 | if (!alloc_cpumask_var(&trial->effective_cpus, GFP_KERNEL)) | |
425 | goto free_cpus; | |
300ed6cb | 426 | |
e2b9a3d7 LZ |
427 | cpumask_copy(trial->cpus_allowed, cs->cpus_allowed); |
428 | cpumask_copy(trial->effective_cpus, cs->effective_cpus); | |
300ed6cb | 429 | return trial; |
e2b9a3d7 LZ |
430 | |
431 | free_cpus: | |
432 | free_cpumask_var(trial->cpus_allowed); | |
433 | free_cs: | |
434 | kfree(trial); | |
435 | return NULL; | |
645fcc9d LZ |
436 | } |
437 | ||
438 | /** | |
439 | * free_trial_cpuset - free the trial cpuset | |
440 | * @trial: the trial cpuset to be freed | |
441 | */ | |
442 | static void free_trial_cpuset(struct cpuset *trial) | |
443 | { | |
e2b9a3d7 | 444 | free_cpumask_var(trial->effective_cpus); |
300ed6cb | 445 | free_cpumask_var(trial->cpus_allowed); |
645fcc9d LZ |
446 | kfree(trial); |
447 | } | |
448 | ||
1da177e4 LT |
449 | /* |
450 | * validate_change() - Used to validate that any proposed cpuset change | |
451 | * follows the structural rules for cpusets. | |
452 | * | |
453 | * If we replaced the flag and mask values of the current cpuset | |
454 | * (cur) with those values in the trial cpuset (trial), would | |
455 | * our various subset and exclusive rules still be valid? Presumes | |
5d21cc2d | 456 | * cpuset_mutex held. |
1da177e4 LT |
457 | * |
458 | * 'cur' is the address of an actual, in-use cpuset. Operations | |
459 | * such as list traversal that depend on the actual address of the | |
460 | * cpuset in the list must use cur below, not trial. | |
461 | * | |
462 | * 'trial' is the address of bulk structure copy of cur, with | |
463 | * perhaps one or more of the fields cpus_allowed, mems_allowed, | |
464 | * or flags changed to new, trial values. | |
465 | * | |
466 | * Return 0 if valid, -errno if not. | |
467 | */ | |
468 | ||
c9710d80 | 469 | static int validate_change(struct cpuset *cur, struct cpuset *trial) |
1da177e4 | 470 | { |
492eb21b | 471 | struct cgroup_subsys_state *css; |
1da177e4 | 472 | struct cpuset *c, *par; |
ae8086ce TH |
473 | int ret; |
474 | ||
475 | rcu_read_lock(); | |
1da177e4 LT |
476 | |
477 | /* Each of our child cpusets must be a subset of us */ | |
ae8086ce | 478 | ret = -EBUSY; |
492eb21b | 479 | cpuset_for_each_child(c, css, cur) |
ae8086ce TH |
480 | if (!is_cpuset_subset(c, trial)) |
481 | goto out; | |
1da177e4 LT |
482 | |
483 | /* Remaining checks don't apply to root cpuset */ | |
ae8086ce | 484 | ret = 0; |
69604067 | 485 | if (cur == &top_cpuset) |
ae8086ce | 486 | goto out; |
1da177e4 | 487 | |
c431069f | 488 | par = parent_cs(cur); |
69604067 | 489 | |
7e88291b | 490 | /* On legacy hiearchy, we must be a subset of our parent cpuset. */ |
ae8086ce | 491 | ret = -EACCES; |
9e10a130 TH |
492 | if (!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) && |
493 | !is_cpuset_subset(trial, par)) | |
ae8086ce | 494 | goto out; |
1da177e4 | 495 | |
2df167a3 PM |
496 | /* |
497 | * If either I or some sibling (!= me) is exclusive, we can't | |
498 | * overlap | |
499 | */ | |
ae8086ce | 500 | ret = -EINVAL; |
492eb21b | 501 | cpuset_for_each_child(c, css, par) { |
1da177e4 LT |
502 | if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) && |
503 | c != cur && | |
300ed6cb | 504 | cpumask_intersects(trial->cpus_allowed, c->cpus_allowed)) |
ae8086ce | 505 | goto out; |
1da177e4 LT |
506 | if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) && |
507 | c != cur && | |
508 | nodes_intersects(trial->mems_allowed, c->mems_allowed)) | |
ae8086ce | 509 | goto out; |
1da177e4 LT |
510 | } |
511 | ||
452477fa TH |
512 | /* |
513 | * Cpusets with tasks - existing or newly being attached - can't | |
1c09b195 | 514 | * be changed to have empty cpus_allowed or mems_allowed. |
452477fa | 515 | */ |
ae8086ce | 516 | ret = -ENOSPC; |
27bd4dbb | 517 | if ((cgroup_is_populated(cur->css.cgroup) || cur->attach_in_progress)) { |
1c09b195 LZ |
518 | if (!cpumask_empty(cur->cpus_allowed) && |
519 | cpumask_empty(trial->cpus_allowed)) | |
520 | goto out; | |
521 | if (!nodes_empty(cur->mems_allowed) && | |
522 | nodes_empty(trial->mems_allowed)) | |
523 | goto out; | |
524 | } | |
020958b6 | 525 | |
f82f8042 JL |
526 | /* |
527 | * We can't shrink if we won't have enough room for SCHED_DEADLINE | |
528 | * tasks. | |
529 | */ | |
530 | ret = -EBUSY; | |
531 | if (is_cpu_exclusive(cur) && | |
532 | !cpuset_cpumask_can_shrink(cur->cpus_allowed, | |
533 | trial->cpus_allowed)) | |
534 | goto out; | |
535 | ||
ae8086ce TH |
536 | ret = 0; |
537 | out: | |
538 | rcu_read_unlock(); | |
539 | return ret; | |
1da177e4 LT |
540 | } |
541 | ||
db7f47cf | 542 | #ifdef CONFIG_SMP |
029190c5 | 543 | /* |
cf417141 | 544 | * Helper routine for generate_sched_domains(). |
8b5f1c52 | 545 | * Do cpusets a, b have overlapping effective cpus_allowed masks? |
029190c5 | 546 | */ |
029190c5 PJ |
547 | static int cpusets_overlap(struct cpuset *a, struct cpuset *b) |
548 | { | |
8b5f1c52 | 549 | return cpumask_intersects(a->effective_cpus, b->effective_cpus); |
029190c5 PJ |
550 | } |
551 | ||
1d3504fc HS |
552 | static void |
553 | update_domain_attr(struct sched_domain_attr *dattr, struct cpuset *c) | |
554 | { | |
1d3504fc HS |
555 | if (dattr->relax_domain_level < c->relax_domain_level) |
556 | dattr->relax_domain_level = c->relax_domain_level; | |
557 | return; | |
558 | } | |
559 | ||
fc560a26 TH |
560 | static void update_domain_attr_tree(struct sched_domain_attr *dattr, |
561 | struct cpuset *root_cs) | |
f5393693 | 562 | { |
fc560a26 | 563 | struct cpuset *cp; |
492eb21b | 564 | struct cgroup_subsys_state *pos_css; |
f5393693 | 565 | |
fc560a26 | 566 | rcu_read_lock(); |
492eb21b | 567 | cpuset_for_each_descendant_pre(cp, pos_css, root_cs) { |
fc560a26 TH |
568 | /* skip the whole subtree if @cp doesn't have any CPU */ |
569 | if (cpumask_empty(cp->cpus_allowed)) { | |
492eb21b | 570 | pos_css = css_rightmost_descendant(pos_css); |
f5393693 | 571 | continue; |
fc560a26 | 572 | } |
f5393693 LJ |
573 | |
574 | if (is_sched_load_balance(cp)) | |
575 | update_domain_attr(dattr, cp); | |
f5393693 | 576 | } |
fc560a26 | 577 | rcu_read_unlock(); |
f5393693 LJ |
578 | } |
579 | ||
029190c5 | 580 | /* |
cf417141 MK |
581 | * generate_sched_domains() |
582 | * | |
583 | * This function builds a partial partition of the systems CPUs | |
584 | * A 'partial partition' is a set of non-overlapping subsets whose | |
585 | * union is a subset of that set. | |
0a0fca9d | 586 | * The output of this function needs to be passed to kernel/sched/core.c |
cf417141 MK |
587 | * partition_sched_domains() routine, which will rebuild the scheduler's |
588 | * load balancing domains (sched domains) as specified by that partial | |
589 | * partition. | |
029190c5 | 590 | * |
45ce80fb | 591 | * See "What is sched_load_balance" in Documentation/cgroups/cpusets.txt |
029190c5 PJ |
592 | * for a background explanation of this. |
593 | * | |
594 | * Does not return errors, on the theory that the callers of this | |
595 | * routine would rather not worry about failures to rebuild sched | |
596 | * domains when operating in the severe memory shortage situations | |
597 | * that could cause allocation failures below. | |
598 | * | |
5d21cc2d | 599 | * Must be called with cpuset_mutex held. |
029190c5 PJ |
600 | * |
601 | * The three key local variables below are: | |
aeed6824 | 602 | * q - a linked-list queue of cpuset pointers, used to implement a |
029190c5 PJ |
603 | * top-down scan of all cpusets. This scan loads a pointer |
604 | * to each cpuset marked is_sched_load_balance into the | |
605 | * array 'csa'. For our purposes, rebuilding the schedulers | |
606 | * sched domains, we can ignore !is_sched_load_balance cpusets. | |
607 | * csa - (for CpuSet Array) Array of pointers to all the cpusets | |
608 | * that need to be load balanced, for convenient iterative | |
609 | * access by the subsequent code that finds the best partition, | |
610 | * i.e the set of domains (subsets) of CPUs such that the | |
611 | * cpus_allowed of every cpuset marked is_sched_load_balance | |
612 | * is a subset of one of these domains, while there are as | |
613 | * many such domains as possible, each as small as possible. | |
614 | * doms - Conversion of 'csa' to an array of cpumasks, for passing to | |
0a0fca9d | 615 | * the kernel/sched/core.c routine partition_sched_domains() in a |
029190c5 PJ |
616 | * convenient format, that can be easily compared to the prior |
617 | * value to determine what partition elements (sched domains) | |
618 | * were changed (added or removed.) | |
619 | * | |
620 | * Finding the best partition (set of domains): | |
621 | * The triple nested loops below over i, j, k scan over the | |
622 | * load balanced cpusets (using the array of cpuset pointers in | |
623 | * csa[]) looking for pairs of cpusets that have overlapping | |
624 | * cpus_allowed, but which don't have the same 'pn' partition | |
625 | * number and gives them in the same partition number. It keeps | |
626 | * looping on the 'restart' label until it can no longer find | |
627 | * any such pairs. | |
628 | * | |
629 | * The union of the cpus_allowed masks from the set of | |
630 | * all cpusets having the same 'pn' value then form the one | |
631 | * element of the partition (one sched domain) to be passed to | |
632 | * partition_sched_domains(). | |
633 | */ | |
acc3f5d7 | 634 | static int generate_sched_domains(cpumask_var_t **domains, |
cf417141 | 635 | struct sched_domain_attr **attributes) |
029190c5 | 636 | { |
029190c5 PJ |
637 | struct cpuset *cp; /* scans q */ |
638 | struct cpuset **csa; /* array of all cpuset ptrs */ | |
639 | int csn; /* how many cpuset ptrs in csa so far */ | |
640 | int i, j, k; /* indices for partition finding loops */ | |
acc3f5d7 | 641 | cpumask_var_t *doms; /* resulting partition; i.e. sched domains */ |
47b8ea71 | 642 | cpumask_var_t non_isolated_cpus; /* load balanced CPUs */ |
1d3504fc | 643 | struct sched_domain_attr *dattr; /* attributes for custom domains */ |
1583715d | 644 | int ndoms = 0; /* number of sched domains in result */ |
6af866af | 645 | int nslot; /* next empty doms[] struct cpumask slot */ |
492eb21b | 646 | struct cgroup_subsys_state *pos_css; |
029190c5 | 647 | |
029190c5 | 648 | doms = NULL; |
1d3504fc | 649 | dattr = NULL; |
cf417141 | 650 | csa = NULL; |
029190c5 | 651 | |
47b8ea71 RR |
652 | if (!alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL)) |
653 | goto done; | |
654 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); | |
655 | ||
029190c5 PJ |
656 | /* Special case for the 99% of systems with one, full, sched domain */ |
657 | if (is_sched_load_balance(&top_cpuset)) { | |
acc3f5d7 RR |
658 | ndoms = 1; |
659 | doms = alloc_sched_domains(ndoms); | |
029190c5 | 660 | if (!doms) |
cf417141 MK |
661 | goto done; |
662 | ||
1d3504fc HS |
663 | dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL); |
664 | if (dattr) { | |
665 | *dattr = SD_ATTR_INIT; | |
93a65575 | 666 | update_domain_attr_tree(dattr, &top_cpuset); |
1d3504fc | 667 | } |
47b8ea71 RR |
668 | cpumask_and(doms[0], top_cpuset.effective_cpus, |
669 | non_isolated_cpus); | |
cf417141 | 670 | |
cf417141 | 671 | goto done; |
029190c5 PJ |
672 | } |
673 | ||
664eedde | 674 | csa = kmalloc(nr_cpusets() * sizeof(cp), GFP_KERNEL); |
029190c5 PJ |
675 | if (!csa) |
676 | goto done; | |
677 | csn = 0; | |
678 | ||
fc560a26 | 679 | rcu_read_lock(); |
492eb21b | 680 | cpuset_for_each_descendant_pre(cp, pos_css, &top_cpuset) { |
bd8815a6 TH |
681 | if (cp == &top_cpuset) |
682 | continue; | |
f5393693 | 683 | /* |
fc560a26 TH |
684 | * Continue traversing beyond @cp iff @cp has some CPUs and |
685 | * isn't load balancing. The former is obvious. The | |
686 | * latter: All child cpusets contain a subset of the | |
687 | * parent's cpus, so just skip them, and then we call | |
688 | * update_domain_attr_tree() to calc relax_domain_level of | |
689 | * the corresponding sched domain. | |
f5393693 | 690 | */ |
fc560a26 | 691 | if (!cpumask_empty(cp->cpus_allowed) && |
47b8ea71 RR |
692 | !(is_sched_load_balance(cp) && |
693 | cpumask_intersects(cp->cpus_allowed, non_isolated_cpus))) | |
f5393693 | 694 | continue; |
489a5393 | 695 | |
fc560a26 TH |
696 | if (is_sched_load_balance(cp)) |
697 | csa[csn++] = cp; | |
698 | ||
699 | /* skip @cp's subtree */ | |
492eb21b | 700 | pos_css = css_rightmost_descendant(pos_css); |
fc560a26 TH |
701 | } |
702 | rcu_read_unlock(); | |
029190c5 PJ |
703 | |
704 | for (i = 0; i < csn; i++) | |
705 | csa[i]->pn = i; | |
706 | ndoms = csn; | |
707 | ||
708 | restart: | |
709 | /* Find the best partition (set of sched domains) */ | |
710 | for (i = 0; i < csn; i++) { | |
711 | struct cpuset *a = csa[i]; | |
712 | int apn = a->pn; | |
713 | ||
714 | for (j = 0; j < csn; j++) { | |
715 | struct cpuset *b = csa[j]; | |
716 | int bpn = b->pn; | |
717 | ||
718 | if (apn != bpn && cpusets_overlap(a, b)) { | |
719 | for (k = 0; k < csn; k++) { | |
720 | struct cpuset *c = csa[k]; | |
721 | ||
722 | if (c->pn == bpn) | |
723 | c->pn = apn; | |
724 | } | |
725 | ndoms--; /* one less element */ | |
726 | goto restart; | |
727 | } | |
728 | } | |
729 | } | |
730 | ||
cf417141 MK |
731 | /* |
732 | * Now we know how many domains to create. | |
733 | * Convert <csn, csa> to <ndoms, doms> and populate cpu masks. | |
734 | */ | |
acc3f5d7 | 735 | doms = alloc_sched_domains(ndoms); |
700018e0 | 736 | if (!doms) |
cf417141 | 737 | goto done; |
cf417141 MK |
738 | |
739 | /* | |
740 | * The rest of the code, including the scheduler, can deal with | |
741 | * dattr==NULL case. No need to abort if alloc fails. | |
742 | */ | |
1d3504fc | 743 | dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL); |
029190c5 PJ |
744 | |
745 | for (nslot = 0, i = 0; i < csn; i++) { | |
746 | struct cpuset *a = csa[i]; | |
6af866af | 747 | struct cpumask *dp; |
029190c5 PJ |
748 | int apn = a->pn; |
749 | ||
cf417141 MK |
750 | if (apn < 0) { |
751 | /* Skip completed partitions */ | |
752 | continue; | |
753 | } | |
754 | ||
acc3f5d7 | 755 | dp = doms[nslot]; |
cf417141 MK |
756 | |
757 | if (nslot == ndoms) { | |
758 | static int warnings = 10; | |
759 | if (warnings) { | |
12d3089c FF |
760 | pr_warn("rebuild_sched_domains confused: nslot %d, ndoms %d, csn %d, i %d, apn %d\n", |
761 | nslot, ndoms, csn, i, apn); | |
cf417141 | 762 | warnings--; |
029190c5 | 763 | } |
cf417141 MK |
764 | continue; |
765 | } | |
029190c5 | 766 | |
6af866af | 767 | cpumask_clear(dp); |
cf417141 MK |
768 | if (dattr) |
769 | *(dattr + nslot) = SD_ATTR_INIT; | |
770 | for (j = i; j < csn; j++) { | |
771 | struct cpuset *b = csa[j]; | |
772 | ||
773 | if (apn == b->pn) { | |
8b5f1c52 | 774 | cpumask_or(dp, dp, b->effective_cpus); |
47b8ea71 | 775 | cpumask_and(dp, dp, non_isolated_cpus); |
cf417141 MK |
776 | if (dattr) |
777 | update_domain_attr_tree(dattr + nslot, b); | |
778 | ||
779 | /* Done with this partition */ | |
780 | b->pn = -1; | |
029190c5 | 781 | } |
029190c5 | 782 | } |
cf417141 | 783 | nslot++; |
029190c5 PJ |
784 | } |
785 | BUG_ON(nslot != ndoms); | |
786 | ||
cf417141 | 787 | done: |
47b8ea71 | 788 | free_cpumask_var(non_isolated_cpus); |
cf417141 MK |
789 | kfree(csa); |
790 | ||
700018e0 LZ |
791 | /* |
792 | * Fallback to the default domain if kmalloc() failed. | |
793 | * See comments in partition_sched_domains(). | |
794 | */ | |
795 | if (doms == NULL) | |
796 | ndoms = 1; | |
797 | ||
cf417141 MK |
798 | *domains = doms; |
799 | *attributes = dattr; | |
800 | return ndoms; | |
801 | } | |
802 | ||
803 | /* | |
804 | * Rebuild scheduler domains. | |
805 | * | |
699140ba TH |
806 | * If the flag 'sched_load_balance' of any cpuset with non-empty |
807 | * 'cpus' changes, or if the 'cpus' allowed changes in any cpuset | |
808 | * which has that flag enabled, or if any cpuset with a non-empty | |
809 | * 'cpus' is removed, then call this routine to rebuild the | |
810 | * scheduler's dynamic sched domains. | |
cf417141 | 811 | * |
5d21cc2d | 812 | * Call with cpuset_mutex held. Takes get_online_cpus(). |
cf417141 | 813 | */ |
699140ba | 814 | static void rebuild_sched_domains_locked(void) |
cf417141 MK |
815 | { |
816 | struct sched_domain_attr *attr; | |
acc3f5d7 | 817 | cpumask_var_t *doms; |
cf417141 MK |
818 | int ndoms; |
819 | ||
5d21cc2d | 820 | lockdep_assert_held(&cpuset_mutex); |
86ef5c9a | 821 | get_online_cpus(); |
cf417141 | 822 | |
5b16c2a4 LZ |
823 | /* |
824 | * We have raced with CPU hotplug. Don't do anything to avoid | |
825 | * passing doms with offlined cpu to partition_sched_domains(). | |
826 | * Anyways, hotplug work item will rebuild sched domains. | |
827 | */ | |
8b5f1c52 | 828 | if (!cpumask_equal(top_cpuset.effective_cpus, cpu_active_mask)) |
5b16c2a4 LZ |
829 | goto out; |
830 | ||
cf417141 | 831 | /* Generate domain masks and attrs */ |
cf417141 | 832 | ndoms = generate_sched_domains(&doms, &attr); |
cf417141 MK |
833 | |
834 | /* Have scheduler rebuild the domains */ | |
835 | partition_sched_domains(ndoms, doms, attr); | |
5b16c2a4 | 836 | out: |
86ef5c9a | 837 | put_online_cpus(); |
cf417141 | 838 | } |
db7f47cf | 839 | #else /* !CONFIG_SMP */ |
699140ba | 840 | static void rebuild_sched_domains_locked(void) |
db7f47cf PM |
841 | { |
842 | } | |
db7f47cf | 843 | #endif /* CONFIG_SMP */ |
029190c5 | 844 | |
cf417141 MK |
845 | void rebuild_sched_domains(void) |
846 | { | |
5d21cc2d | 847 | mutex_lock(&cpuset_mutex); |
699140ba | 848 | rebuild_sched_domains_locked(); |
5d21cc2d | 849 | mutex_unlock(&cpuset_mutex); |
029190c5 PJ |
850 | } |
851 | ||
0b2f630a MX |
852 | /** |
853 | * update_tasks_cpumask - Update the cpumasks of tasks in the cpuset. | |
854 | * @cs: the cpuset in which each task's cpus_allowed mask needs to be changed | |
0b2f630a | 855 | * |
d66393e5 TH |
856 | * Iterate through each task of @cs updating its cpus_allowed to the |
857 | * effective cpuset's. As this function is called with cpuset_mutex held, | |
858 | * cpuset membership stays stable. | |
0b2f630a | 859 | */ |
d66393e5 | 860 | static void update_tasks_cpumask(struct cpuset *cs) |
0b2f630a | 861 | { |
d66393e5 TH |
862 | struct css_task_iter it; |
863 | struct task_struct *task; | |
864 | ||
865 | css_task_iter_start(&cs->css, &it); | |
866 | while ((task = css_task_iter_next(&it))) | |
ae1c8023 | 867 | set_cpus_allowed_ptr(task, cs->effective_cpus); |
d66393e5 | 868 | css_task_iter_end(&it); |
0b2f630a MX |
869 | } |
870 | ||
5c5cc623 | 871 | /* |
734d4513 LZ |
872 | * update_cpumasks_hier - Update effective cpumasks and tasks in the subtree |
873 | * @cs: the cpuset to consider | |
874 | * @new_cpus: temp variable for calculating new effective_cpus | |
875 | * | |
876 | * When congifured cpumask is changed, the effective cpumasks of this cpuset | |
877 | * and all its descendants need to be updated. | |
5c5cc623 | 878 | * |
734d4513 | 879 | * On legacy hierachy, effective_cpus will be the same with cpu_allowed. |
5c5cc623 LZ |
880 | * |
881 | * Called with cpuset_mutex held | |
882 | */ | |
734d4513 | 883 | static void update_cpumasks_hier(struct cpuset *cs, struct cpumask *new_cpus) |
5c5cc623 LZ |
884 | { |
885 | struct cpuset *cp; | |
492eb21b | 886 | struct cgroup_subsys_state *pos_css; |
8b5f1c52 | 887 | bool need_rebuild_sched_domains = false; |
5c5cc623 LZ |
888 | |
889 | rcu_read_lock(); | |
734d4513 LZ |
890 | cpuset_for_each_descendant_pre(cp, pos_css, cs) { |
891 | struct cpuset *parent = parent_cs(cp); | |
892 | ||
893 | cpumask_and(new_cpus, cp->cpus_allowed, parent->effective_cpus); | |
894 | ||
554b0d1c LZ |
895 | /* |
896 | * If it becomes empty, inherit the effective mask of the | |
897 | * parent, which is guaranteed to have some CPUs. | |
898 | */ | |
9e10a130 TH |
899 | if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys) && |
900 | cpumask_empty(new_cpus)) | |
554b0d1c LZ |
901 | cpumask_copy(new_cpus, parent->effective_cpus); |
902 | ||
734d4513 LZ |
903 | /* Skip the whole subtree if the cpumask remains the same. */ |
904 | if (cpumask_equal(new_cpus, cp->effective_cpus)) { | |
905 | pos_css = css_rightmost_descendant(pos_css); | |
906 | continue; | |
5c5cc623 | 907 | } |
734d4513 | 908 | |
ec903c0c | 909 | if (!css_tryget_online(&cp->css)) |
5c5cc623 LZ |
910 | continue; |
911 | rcu_read_unlock(); | |
912 | ||
8447a0fe | 913 | spin_lock_irq(&callback_lock); |
734d4513 | 914 | cpumask_copy(cp->effective_cpus, new_cpus); |
8447a0fe | 915 | spin_unlock_irq(&callback_lock); |
734d4513 | 916 | |
9e10a130 | 917 | WARN_ON(!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) && |
734d4513 LZ |
918 | !cpumask_equal(cp->cpus_allowed, cp->effective_cpus)); |
919 | ||
d66393e5 | 920 | update_tasks_cpumask(cp); |
5c5cc623 | 921 | |
8b5f1c52 LZ |
922 | /* |
923 | * If the effective cpumask of any non-empty cpuset is changed, | |
924 | * we need to rebuild sched domains. | |
925 | */ | |
926 | if (!cpumask_empty(cp->cpus_allowed) && | |
927 | is_sched_load_balance(cp)) | |
928 | need_rebuild_sched_domains = true; | |
929 | ||
5c5cc623 LZ |
930 | rcu_read_lock(); |
931 | css_put(&cp->css); | |
932 | } | |
933 | rcu_read_unlock(); | |
8b5f1c52 LZ |
934 | |
935 | if (need_rebuild_sched_domains) | |
936 | rebuild_sched_domains_locked(); | |
5c5cc623 LZ |
937 | } |
938 | ||
58f4790b CW |
939 | /** |
940 | * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it | |
941 | * @cs: the cpuset to consider | |
fc34ac1d | 942 | * @trialcs: trial cpuset |
58f4790b CW |
943 | * @buf: buffer of cpu numbers written to this cpuset |
944 | */ | |
645fcc9d LZ |
945 | static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs, |
946 | const char *buf) | |
1da177e4 | 947 | { |
58f4790b | 948 | int retval; |
1da177e4 | 949 | |
5f054e31 | 950 | /* top_cpuset.cpus_allowed tracks cpu_online_mask; it's read-only */ |
4c4d50f7 PJ |
951 | if (cs == &top_cpuset) |
952 | return -EACCES; | |
953 | ||
6f7f02e7 | 954 | /* |
c8d9c90c | 955 | * An empty cpus_allowed is ok only if the cpuset has no tasks. |
020958b6 PJ |
956 | * Since cpulist_parse() fails on an empty mask, we special case |
957 | * that parsing. The validate_change() call ensures that cpusets | |
958 | * with tasks have cpus. | |
6f7f02e7 | 959 | */ |
020958b6 | 960 | if (!*buf) { |
300ed6cb | 961 | cpumask_clear(trialcs->cpus_allowed); |
6f7f02e7 | 962 | } else { |
300ed6cb | 963 | retval = cpulist_parse(buf, trialcs->cpus_allowed); |
6f7f02e7 DR |
964 | if (retval < 0) |
965 | return retval; | |
37340746 | 966 | |
5d8ba82c LZ |
967 | if (!cpumask_subset(trialcs->cpus_allowed, |
968 | top_cpuset.cpus_allowed)) | |
37340746 | 969 | return -EINVAL; |
6f7f02e7 | 970 | } |
029190c5 | 971 | |
8707d8b8 | 972 | /* Nothing to do if the cpus didn't change */ |
300ed6cb | 973 | if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed)) |
8707d8b8 | 974 | return 0; |
58f4790b | 975 | |
a73456f3 LZ |
976 | retval = validate_change(cs, trialcs); |
977 | if (retval < 0) | |
978 | return retval; | |
979 | ||
8447a0fe | 980 | spin_lock_irq(&callback_lock); |
300ed6cb | 981 | cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed); |
8447a0fe | 982 | spin_unlock_irq(&callback_lock); |
029190c5 | 983 | |
734d4513 LZ |
984 | /* use trialcs->cpus_allowed as a temp variable */ |
985 | update_cpumasks_hier(cs, trialcs->cpus_allowed); | |
85d7b949 | 986 | return 0; |
1da177e4 LT |
987 | } |
988 | ||
e4e364e8 | 989 | /* |
e93ad19d TH |
990 | * Migrate memory region from one set of nodes to another. This is |
991 | * performed asynchronously as it can be called from process migration path | |
992 | * holding locks involved in process management. All mm migrations are | |
993 | * performed in the queued order and can be waited for by flushing | |
994 | * cpuset_migrate_mm_wq. | |
e4e364e8 PJ |
995 | */ |
996 | ||
e93ad19d TH |
997 | struct cpuset_migrate_mm_work { |
998 | struct work_struct work; | |
999 | struct mm_struct *mm; | |
1000 | nodemask_t from; | |
1001 | nodemask_t to; | |
1002 | }; | |
1003 | ||
1004 | static void cpuset_migrate_mm_workfn(struct work_struct *work) | |
1005 | { | |
1006 | struct cpuset_migrate_mm_work *mwork = | |
1007 | container_of(work, struct cpuset_migrate_mm_work, work); | |
1008 | ||
1009 | /* on a wq worker, no need to worry about %current's mems_allowed */ | |
1010 | do_migrate_pages(mwork->mm, &mwork->from, &mwork->to, MPOL_MF_MOVE_ALL); | |
1011 | mmput(mwork->mm); | |
1012 | kfree(mwork); | |
1013 | } | |
1014 | ||
e4e364e8 PJ |
1015 | static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from, |
1016 | const nodemask_t *to) | |
1017 | { | |
e93ad19d | 1018 | struct cpuset_migrate_mm_work *mwork; |
e4e364e8 | 1019 | |
e93ad19d TH |
1020 | mwork = kzalloc(sizeof(*mwork), GFP_KERNEL); |
1021 | if (mwork) { | |
1022 | mwork->mm = mm; | |
1023 | mwork->from = *from; | |
1024 | mwork->to = *to; | |
1025 | INIT_WORK(&mwork->work, cpuset_migrate_mm_workfn); | |
1026 | queue_work(cpuset_migrate_mm_wq, &mwork->work); | |
1027 | } else { | |
1028 | mmput(mm); | |
1029 | } | |
1030 | } | |
e4e364e8 | 1031 | |
5cf1cacb | 1032 | static void cpuset_post_attach(void) |
e93ad19d TH |
1033 | { |
1034 | flush_workqueue(cpuset_migrate_mm_wq); | |
e4e364e8 PJ |
1035 | } |
1036 | ||
3b6766fe | 1037 | /* |
58568d2a MX |
1038 | * cpuset_change_task_nodemask - change task's mems_allowed and mempolicy |
1039 | * @tsk: the task to change | |
1040 | * @newmems: new nodes that the task will be set | |
1041 | * | |
5f155f27 VB |
1042 | * We use the mems_allowed_seq seqlock to safely update both tsk->mems_allowed |
1043 | * and rebind an eventual tasks' mempolicy. If the task is allocating in | |
1044 | * parallel, it might temporarily see an empty intersection, which results in | |
1045 | * a seqlock check and retry before OOM or allocation failure. | |
58568d2a MX |
1046 | */ |
1047 | static void cpuset_change_task_nodemask(struct task_struct *tsk, | |
1048 | nodemask_t *newmems) | |
1049 | { | |
c0ff7453 | 1050 | task_lock(tsk); |
c0ff7453 | 1051 | |
5f155f27 VB |
1052 | local_irq_disable(); |
1053 | write_seqcount_begin(&tsk->mems_allowed_seq); | |
c0ff7453 | 1054 | |
cc9a6c87 | 1055 | nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems); |
213980c0 | 1056 | mpol_rebind_task(tsk, newmems); |
58568d2a | 1057 | tsk->mems_allowed = *newmems; |
cc9a6c87 | 1058 | |
5f155f27 VB |
1059 | write_seqcount_end(&tsk->mems_allowed_seq); |
1060 | local_irq_enable(); | |
cc9a6c87 | 1061 | |
c0ff7453 | 1062 | task_unlock(tsk); |
58568d2a MX |
1063 | } |
1064 | ||
8793d854 PM |
1065 | static void *cpuset_being_rebound; |
1066 | ||
0b2f630a MX |
1067 | /** |
1068 | * update_tasks_nodemask - Update the nodemasks of tasks in the cpuset. | |
1069 | * @cs: the cpuset in which each task's mems_allowed mask needs to be changed | |
0b2f630a | 1070 | * |
d66393e5 TH |
1071 | * Iterate through each task of @cs updating its mems_allowed to the |
1072 | * effective cpuset's. As this function is called with cpuset_mutex held, | |
1073 | * cpuset membership stays stable. | |
0b2f630a | 1074 | */ |
d66393e5 | 1075 | static void update_tasks_nodemask(struct cpuset *cs) |
1da177e4 | 1076 | { |
33ad801d | 1077 | static nodemask_t newmems; /* protected by cpuset_mutex */ |
d66393e5 TH |
1078 | struct css_task_iter it; |
1079 | struct task_struct *task; | |
59dac16f | 1080 | |
846a16bf | 1081 | cpuset_being_rebound = cs; /* causes mpol_dup() rebind */ |
4225399a | 1082 | |
ae1c8023 | 1083 | guarantee_online_mems(cs, &newmems); |
33ad801d | 1084 | |
4225399a | 1085 | /* |
3b6766fe LZ |
1086 | * The mpol_rebind_mm() call takes mmap_sem, which we couldn't |
1087 | * take while holding tasklist_lock. Forks can happen - the | |
1088 | * mpol_dup() cpuset_being_rebound check will catch such forks, | |
1089 | * and rebind their vma mempolicies too. Because we still hold | |
5d21cc2d | 1090 | * the global cpuset_mutex, we know that no other rebind effort |
3b6766fe | 1091 | * will be contending for the global variable cpuset_being_rebound. |
4225399a | 1092 | * It's ok if we rebind the same mm twice; mpol_rebind_mm() |
04c19fa6 | 1093 | * is idempotent. Also migrate pages in each mm to new nodes. |
4225399a | 1094 | */ |
d66393e5 TH |
1095 | css_task_iter_start(&cs->css, &it); |
1096 | while ((task = css_task_iter_next(&it))) { | |
1097 | struct mm_struct *mm; | |
1098 | bool migrate; | |
1099 | ||
1100 | cpuset_change_task_nodemask(task, &newmems); | |
1101 | ||
1102 | mm = get_task_mm(task); | |
1103 | if (!mm) | |
1104 | continue; | |
1105 | ||
1106 | migrate = is_memory_migrate(cs); | |
1107 | ||
1108 | mpol_rebind_mm(mm, &cs->mems_allowed); | |
1109 | if (migrate) | |
1110 | cpuset_migrate_mm(mm, &cs->old_mems_allowed, &newmems); | |
e93ad19d TH |
1111 | else |
1112 | mmput(mm); | |
d66393e5 TH |
1113 | } |
1114 | css_task_iter_end(&it); | |
4225399a | 1115 | |
33ad801d LZ |
1116 | /* |
1117 | * All the tasks' nodemasks have been updated, update | |
1118 | * cs->old_mems_allowed. | |
1119 | */ | |
1120 | cs->old_mems_allowed = newmems; | |
1121 | ||
2df167a3 | 1122 | /* We're done rebinding vmas to this cpuset's new mems_allowed. */ |
8793d854 | 1123 | cpuset_being_rebound = NULL; |
1da177e4 LT |
1124 | } |
1125 | ||
5c5cc623 | 1126 | /* |
734d4513 LZ |
1127 | * update_nodemasks_hier - Update effective nodemasks and tasks in the subtree |
1128 | * @cs: the cpuset to consider | |
1129 | * @new_mems: a temp variable for calculating new effective_mems | |
5c5cc623 | 1130 | * |
734d4513 LZ |
1131 | * When configured nodemask is changed, the effective nodemasks of this cpuset |
1132 | * and all its descendants need to be updated. | |
5c5cc623 | 1133 | * |
734d4513 | 1134 | * On legacy hiearchy, effective_mems will be the same with mems_allowed. |
5c5cc623 LZ |
1135 | * |
1136 | * Called with cpuset_mutex held | |
1137 | */ | |
734d4513 | 1138 | static void update_nodemasks_hier(struct cpuset *cs, nodemask_t *new_mems) |
5c5cc623 LZ |
1139 | { |
1140 | struct cpuset *cp; | |
492eb21b | 1141 | struct cgroup_subsys_state *pos_css; |
5c5cc623 LZ |
1142 | |
1143 | rcu_read_lock(); | |
734d4513 LZ |
1144 | cpuset_for_each_descendant_pre(cp, pos_css, cs) { |
1145 | struct cpuset *parent = parent_cs(cp); | |
1146 | ||
1147 | nodes_and(*new_mems, cp->mems_allowed, parent->effective_mems); | |
1148 | ||
554b0d1c LZ |
1149 | /* |
1150 | * If it becomes empty, inherit the effective mask of the | |
1151 | * parent, which is guaranteed to have some MEMs. | |
1152 | */ | |
9e10a130 TH |
1153 | if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys) && |
1154 | nodes_empty(*new_mems)) | |
554b0d1c LZ |
1155 | *new_mems = parent->effective_mems; |
1156 | ||
734d4513 LZ |
1157 | /* Skip the whole subtree if the nodemask remains the same. */ |
1158 | if (nodes_equal(*new_mems, cp->effective_mems)) { | |
1159 | pos_css = css_rightmost_descendant(pos_css); | |
1160 | continue; | |
5c5cc623 | 1161 | } |
734d4513 | 1162 | |
ec903c0c | 1163 | if (!css_tryget_online(&cp->css)) |
5c5cc623 LZ |
1164 | continue; |
1165 | rcu_read_unlock(); | |
1166 | ||
8447a0fe | 1167 | spin_lock_irq(&callback_lock); |
734d4513 | 1168 | cp->effective_mems = *new_mems; |
8447a0fe | 1169 | spin_unlock_irq(&callback_lock); |
734d4513 | 1170 | |
9e10a130 | 1171 | WARN_ON(!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) && |
a1381268 | 1172 | !nodes_equal(cp->mems_allowed, cp->effective_mems)); |
734d4513 | 1173 | |
d66393e5 | 1174 | update_tasks_nodemask(cp); |
5c5cc623 LZ |
1175 | |
1176 | rcu_read_lock(); | |
1177 | css_put(&cp->css); | |
1178 | } | |
1179 | rcu_read_unlock(); | |
1180 | } | |
1181 | ||
0b2f630a MX |
1182 | /* |
1183 | * Handle user request to change the 'mems' memory placement | |
1184 | * of a cpuset. Needs to validate the request, update the | |
58568d2a MX |
1185 | * cpusets mems_allowed, and for each task in the cpuset, |
1186 | * update mems_allowed and rebind task's mempolicy and any vma | |
1187 | * mempolicies and if the cpuset is marked 'memory_migrate', | |
1188 | * migrate the tasks pages to the new memory. | |
0b2f630a | 1189 | * |
8447a0fe | 1190 | * Call with cpuset_mutex held. May take callback_lock during call. |
0b2f630a MX |
1191 | * Will take tasklist_lock, scan tasklist for tasks in cpuset cs, |
1192 | * lock each such tasks mm->mmap_sem, scan its vma's and rebind | |
1193 | * their mempolicies to the cpusets new mems_allowed. | |
1194 | */ | |
645fcc9d LZ |
1195 | static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs, |
1196 | const char *buf) | |
0b2f630a | 1197 | { |
0b2f630a MX |
1198 | int retval; |
1199 | ||
1200 | /* | |
38d7bee9 | 1201 | * top_cpuset.mems_allowed tracks node_stats[N_MEMORY]; |
0b2f630a MX |
1202 | * it's read-only |
1203 | */ | |
53feb297 MX |
1204 | if (cs == &top_cpuset) { |
1205 | retval = -EACCES; | |
1206 | goto done; | |
1207 | } | |
0b2f630a | 1208 | |
0b2f630a MX |
1209 | /* |
1210 | * An empty mems_allowed is ok iff there are no tasks in the cpuset. | |
1211 | * Since nodelist_parse() fails on an empty mask, we special case | |
1212 | * that parsing. The validate_change() call ensures that cpusets | |
1213 | * with tasks have memory. | |
1214 | */ | |
1215 | if (!*buf) { | |
645fcc9d | 1216 | nodes_clear(trialcs->mems_allowed); |
0b2f630a | 1217 | } else { |
645fcc9d | 1218 | retval = nodelist_parse(buf, trialcs->mems_allowed); |
0b2f630a MX |
1219 | if (retval < 0) |
1220 | goto done; | |
1221 | ||
645fcc9d | 1222 | if (!nodes_subset(trialcs->mems_allowed, |
5d8ba82c LZ |
1223 | top_cpuset.mems_allowed)) { |
1224 | retval = -EINVAL; | |
53feb297 MX |
1225 | goto done; |
1226 | } | |
0b2f630a | 1227 | } |
33ad801d LZ |
1228 | |
1229 | if (nodes_equal(cs->mems_allowed, trialcs->mems_allowed)) { | |
0b2f630a MX |
1230 | retval = 0; /* Too easy - nothing to do */ |
1231 | goto done; | |
1232 | } | |
645fcc9d | 1233 | retval = validate_change(cs, trialcs); |
0b2f630a MX |
1234 | if (retval < 0) |
1235 | goto done; | |
1236 | ||
8447a0fe | 1237 | spin_lock_irq(&callback_lock); |
645fcc9d | 1238 | cs->mems_allowed = trialcs->mems_allowed; |
8447a0fe | 1239 | spin_unlock_irq(&callback_lock); |
0b2f630a | 1240 | |
734d4513 | 1241 | /* use trialcs->mems_allowed as a temp variable */ |
24ee3cf8 | 1242 | update_nodemasks_hier(cs, &trialcs->mems_allowed); |
0b2f630a MX |
1243 | done: |
1244 | return retval; | |
1245 | } | |
1246 | ||
8793d854 PM |
1247 | int current_cpuset_is_being_rebound(void) |
1248 | { | |
391acf97 GZ |
1249 | int ret; |
1250 | ||
1251 | rcu_read_lock(); | |
1252 | ret = task_cs(current) == cpuset_being_rebound; | |
1253 | rcu_read_unlock(); | |
1254 | ||
1255 | return ret; | |
8793d854 PM |
1256 | } |
1257 | ||
5be7a479 | 1258 | static int update_relax_domain_level(struct cpuset *cs, s64 val) |
1d3504fc | 1259 | { |
db7f47cf | 1260 | #ifdef CONFIG_SMP |
60495e77 | 1261 | if (val < -1 || val >= sched_domain_level_max) |
30e0e178 | 1262 | return -EINVAL; |
db7f47cf | 1263 | #endif |
1d3504fc HS |
1264 | |
1265 | if (val != cs->relax_domain_level) { | |
1266 | cs->relax_domain_level = val; | |
300ed6cb LZ |
1267 | if (!cpumask_empty(cs->cpus_allowed) && |
1268 | is_sched_load_balance(cs)) | |
699140ba | 1269 | rebuild_sched_domains_locked(); |
1d3504fc HS |
1270 | } |
1271 | ||
1272 | return 0; | |
1273 | } | |
1274 | ||
72ec7029 | 1275 | /** |
950592f7 MX |
1276 | * update_tasks_flags - update the spread flags of tasks in the cpuset. |
1277 | * @cs: the cpuset in which each task's spread flags needs to be changed | |
950592f7 | 1278 | * |
d66393e5 TH |
1279 | * Iterate through each task of @cs updating its spread flags. As this |
1280 | * function is called with cpuset_mutex held, cpuset membership stays | |
1281 | * stable. | |
950592f7 | 1282 | */ |
d66393e5 | 1283 | static void update_tasks_flags(struct cpuset *cs) |
950592f7 | 1284 | { |
d66393e5 TH |
1285 | struct css_task_iter it; |
1286 | struct task_struct *task; | |
1287 | ||
1288 | css_task_iter_start(&cs->css, &it); | |
1289 | while ((task = css_task_iter_next(&it))) | |
1290 | cpuset_update_task_spread_flag(cs, task); | |
1291 | css_task_iter_end(&it); | |
950592f7 MX |
1292 | } |
1293 | ||
1da177e4 LT |
1294 | /* |
1295 | * update_flag - read a 0 or a 1 in a file and update associated flag | |
78608366 PM |
1296 | * bit: the bit to update (see cpuset_flagbits_t) |
1297 | * cs: the cpuset to update | |
1298 | * turning_on: whether the flag is being set or cleared | |
053199ed | 1299 | * |
5d21cc2d | 1300 | * Call with cpuset_mutex held. |
1da177e4 LT |
1301 | */ |
1302 | ||
700fe1ab PM |
1303 | static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs, |
1304 | int turning_on) | |
1da177e4 | 1305 | { |
645fcc9d | 1306 | struct cpuset *trialcs; |
40b6a762 | 1307 | int balance_flag_changed; |
950592f7 | 1308 | int spread_flag_changed; |
950592f7 | 1309 | int err; |
1da177e4 | 1310 | |
645fcc9d LZ |
1311 | trialcs = alloc_trial_cpuset(cs); |
1312 | if (!trialcs) | |
1313 | return -ENOMEM; | |
1314 | ||
1da177e4 | 1315 | if (turning_on) |
645fcc9d | 1316 | set_bit(bit, &trialcs->flags); |
1da177e4 | 1317 | else |
645fcc9d | 1318 | clear_bit(bit, &trialcs->flags); |
1da177e4 | 1319 | |
645fcc9d | 1320 | err = validate_change(cs, trialcs); |
85d7b949 | 1321 | if (err < 0) |
645fcc9d | 1322 | goto out; |
029190c5 | 1323 | |
029190c5 | 1324 | balance_flag_changed = (is_sched_load_balance(cs) != |
645fcc9d | 1325 | is_sched_load_balance(trialcs)); |
029190c5 | 1326 | |
950592f7 MX |
1327 | spread_flag_changed = ((is_spread_slab(cs) != is_spread_slab(trialcs)) |
1328 | || (is_spread_page(cs) != is_spread_page(trialcs))); | |
1329 | ||
8447a0fe | 1330 | spin_lock_irq(&callback_lock); |
645fcc9d | 1331 | cs->flags = trialcs->flags; |
8447a0fe | 1332 | spin_unlock_irq(&callback_lock); |
85d7b949 | 1333 | |
300ed6cb | 1334 | if (!cpumask_empty(trialcs->cpus_allowed) && balance_flag_changed) |
699140ba | 1335 | rebuild_sched_domains_locked(); |
029190c5 | 1336 | |
950592f7 | 1337 | if (spread_flag_changed) |
d66393e5 | 1338 | update_tasks_flags(cs); |
645fcc9d LZ |
1339 | out: |
1340 | free_trial_cpuset(trialcs); | |
1341 | return err; | |
1da177e4 LT |
1342 | } |
1343 | ||
3e0d98b9 | 1344 | /* |
80f7228b | 1345 | * Frequency meter - How fast is some event occurring? |
3e0d98b9 PJ |
1346 | * |
1347 | * These routines manage a digitally filtered, constant time based, | |
1348 | * event frequency meter. There are four routines: | |
1349 | * fmeter_init() - initialize a frequency meter. | |
1350 | * fmeter_markevent() - called each time the event happens. | |
1351 | * fmeter_getrate() - returns the recent rate of such events. | |
1352 | * fmeter_update() - internal routine used to update fmeter. | |
1353 | * | |
1354 | * A common data structure is passed to each of these routines, | |
1355 | * which is used to keep track of the state required to manage the | |
1356 | * frequency meter and its digital filter. | |
1357 | * | |
1358 | * The filter works on the number of events marked per unit time. | |
1359 | * The filter is single-pole low-pass recursive (IIR). The time unit | |
1360 | * is 1 second. Arithmetic is done using 32-bit integers scaled to | |
1361 | * simulate 3 decimal digits of precision (multiplied by 1000). | |
1362 | * | |
1363 | * With an FM_COEF of 933, and a time base of 1 second, the filter | |
1364 | * has a half-life of 10 seconds, meaning that if the events quit | |
1365 | * happening, then the rate returned from the fmeter_getrate() | |
1366 | * will be cut in half each 10 seconds, until it converges to zero. | |
1367 | * | |
1368 | * It is not worth doing a real infinitely recursive filter. If more | |
1369 | * than FM_MAXTICKS ticks have elapsed since the last filter event, | |
1370 | * just compute FM_MAXTICKS ticks worth, by which point the level | |
1371 | * will be stable. | |
1372 | * | |
1373 | * Limit the count of unprocessed events to FM_MAXCNT, so as to avoid | |
1374 | * arithmetic overflow in the fmeter_update() routine. | |
1375 | * | |
1376 | * Given the simple 32 bit integer arithmetic used, this meter works | |
1377 | * best for reporting rates between one per millisecond (msec) and | |
1378 | * one per 32 (approx) seconds. At constant rates faster than one | |
1379 | * per msec it maxes out at values just under 1,000,000. At constant | |
1380 | * rates between one per msec, and one per second it will stabilize | |
1381 | * to a value N*1000, where N is the rate of events per second. | |
1382 | * At constant rates between one per second and one per 32 seconds, | |
1383 | * it will be choppy, moving up on the seconds that have an event, | |
1384 | * and then decaying until the next event. At rates slower than | |
1385 | * about one in 32 seconds, it decays all the way back to zero between | |
1386 | * each event. | |
1387 | */ | |
1388 | ||
1389 | #define FM_COEF 933 /* coefficient for half-life of 10 secs */ | |
d2b43658 | 1390 | #define FM_MAXTICKS ((u32)99) /* useless computing more ticks than this */ |
3e0d98b9 PJ |
1391 | #define FM_MAXCNT 1000000 /* limit cnt to avoid overflow */ |
1392 | #define FM_SCALE 1000 /* faux fixed point scale */ | |
1393 | ||
1394 | /* Initialize a frequency meter */ | |
1395 | static void fmeter_init(struct fmeter *fmp) | |
1396 | { | |
1397 | fmp->cnt = 0; | |
1398 | fmp->val = 0; | |
1399 | fmp->time = 0; | |
1400 | spin_lock_init(&fmp->lock); | |
1401 | } | |
1402 | ||
1403 | /* Internal meter update - process cnt events and update value */ | |
1404 | static void fmeter_update(struct fmeter *fmp) | |
1405 | { | |
d2b43658 AB |
1406 | time64_t now; |
1407 | u32 ticks; | |
1408 | ||
1409 | now = ktime_get_seconds(); | |
1410 | ticks = now - fmp->time; | |
3e0d98b9 PJ |
1411 | |
1412 | if (ticks == 0) | |
1413 | return; | |
1414 | ||
1415 | ticks = min(FM_MAXTICKS, ticks); | |
1416 | while (ticks-- > 0) | |
1417 | fmp->val = (FM_COEF * fmp->val) / FM_SCALE; | |
1418 | fmp->time = now; | |
1419 | ||
1420 | fmp->val += ((FM_SCALE - FM_COEF) * fmp->cnt) / FM_SCALE; | |
1421 | fmp->cnt = 0; | |
1422 | } | |
1423 | ||
1424 | /* Process any previous ticks, then bump cnt by one (times scale). */ | |
1425 | static void fmeter_markevent(struct fmeter *fmp) | |
1426 | { | |
1427 | spin_lock(&fmp->lock); | |
1428 | fmeter_update(fmp); | |
1429 | fmp->cnt = min(FM_MAXCNT, fmp->cnt + FM_SCALE); | |
1430 | spin_unlock(&fmp->lock); | |
1431 | } | |
1432 | ||
1433 | /* Process any previous ticks, then return current value. */ | |
1434 | static int fmeter_getrate(struct fmeter *fmp) | |
1435 | { | |
1436 | int val; | |
1437 | ||
1438 | spin_lock(&fmp->lock); | |
1439 | fmeter_update(fmp); | |
1440 | val = fmp->val; | |
1441 | spin_unlock(&fmp->lock); | |
1442 | return val; | |
1443 | } | |
1444 | ||
57fce0a6 TH |
1445 | static struct cpuset *cpuset_attach_old_cs; |
1446 | ||
5d21cc2d | 1447 | /* Called by cgroups to determine if a cpuset is usable; cpuset_mutex held */ |
1f7dd3e5 | 1448 | static int cpuset_can_attach(struct cgroup_taskset *tset) |
f780bdb7 | 1449 | { |
1f7dd3e5 TH |
1450 | struct cgroup_subsys_state *css; |
1451 | struct cpuset *cs; | |
bb9d97b6 TH |
1452 | struct task_struct *task; |
1453 | int ret; | |
1da177e4 | 1454 | |
57fce0a6 | 1455 | /* used later by cpuset_attach() */ |
1f7dd3e5 TH |
1456 | cpuset_attach_old_cs = task_cs(cgroup_taskset_first(tset, &css)); |
1457 | cs = css_cs(css); | |
57fce0a6 | 1458 | |
5d21cc2d TH |
1459 | mutex_lock(&cpuset_mutex); |
1460 | ||
aa6ec29b | 1461 | /* allow moving tasks into an empty cpuset if on default hierarchy */ |
5d21cc2d | 1462 | ret = -ENOSPC; |
9e10a130 | 1463 | if (!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) && |
88fa523b | 1464 | (cpumask_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed))) |
5d21cc2d | 1465 | goto out_unlock; |
9985b0ba | 1466 | |
1f7dd3e5 | 1467 | cgroup_taskset_for_each(task, css, tset) { |
7f51412a JL |
1468 | ret = task_can_attach(task, cs->cpus_allowed); |
1469 | if (ret) | |
5d21cc2d TH |
1470 | goto out_unlock; |
1471 | ret = security_task_setscheduler(task); | |
1472 | if (ret) | |
1473 | goto out_unlock; | |
bb9d97b6 | 1474 | } |
f780bdb7 | 1475 | |
452477fa TH |
1476 | /* |
1477 | * Mark attach is in progress. This makes validate_change() fail | |
1478 | * changes which zero cpus/mems_allowed. | |
1479 | */ | |
1480 | cs->attach_in_progress++; | |
5d21cc2d TH |
1481 | ret = 0; |
1482 | out_unlock: | |
1483 | mutex_unlock(&cpuset_mutex); | |
1484 | return ret; | |
8793d854 | 1485 | } |
f780bdb7 | 1486 | |
1f7dd3e5 | 1487 | static void cpuset_cancel_attach(struct cgroup_taskset *tset) |
452477fa | 1488 | { |
1f7dd3e5 TH |
1489 | struct cgroup_subsys_state *css; |
1490 | struct cpuset *cs; | |
1491 | ||
1492 | cgroup_taskset_first(tset, &css); | |
1493 | cs = css_cs(css); | |
1494 | ||
5d21cc2d | 1495 | mutex_lock(&cpuset_mutex); |
eb95419b | 1496 | css_cs(css)->attach_in_progress--; |
5d21cc2d | 1497 | mutex_unlock(&cpuset_mutex); |
8793d854 | 1498 | } |
1da177e4 | 1499 | |
4e4c9a14 | 1500 | /* |
5d21cc2d | 1501 | * Protected by cpuset_mutex. cpus_attach is used only by cpuset_attach() |
4e4c9a14 TH |
1502 | * but we can't allocate it dynamically there. Define it global and |
1503 | * allocate from cpuset_init(). | |
1504 | */ | |
1505 | static cpumask_var_t cpus_attach; | |
1506 | ||
1f7dd3e5 | 1507 | static void cpuset_attach(struct cgroup_taskset *tset) |
8793d854 | 1508 | { |
67bd2c59 | 1509 | /* static buf protected by cpuset_mutex */ |
4e4c9a14 | 1510 | static nodemask_t cpuset_attach_nodemask_to; |
bb9d97b6 | 1511 | struct task_struct *task; |
4530eddb | 1512 | struct task_struct *leader; |
1f7dd3e5 TH |
1513 | struct cgroup_subsys_state *css; |
1514 | struct cpuset *cs; | |
57fce0a6 | 1515 | struct cpuset *oldcs = cpuset_attach_old_cs; |
22fb52dd | 1516 | |
1f7dd3e5 TH |
1517 | cgroup_taskset_first(tset, &css); |
1518 | cs = css_cs(css); | |
1519 | ||
5d21cc2d TH |
1520 | mutex_lock(&cpuset_mutex); |
1521 | ||
4e4c9a14 TH |
1522 | /* prepare for attach */ |
1523 | if (cs == &top_cpuset) | |
1524 | cpumask_copy(cpus_attach, cpu_possible_mask); | |
1525 | else | |
ae1c8023 | 1526 | guarantee_online_cpus(cs, cpus_attach); |
4e4c9a14 | 1527 | |
ae1c8023 | 1528 | guarantee_online_mems(cs, &cpuset_attach_nodemask_to); |
4e4c9a14 | 1529 | |
1f7dd3e5 | 1530 | cgroup_taskset_for_each(task, css, tset) { |
bb9d97b6 TH |
1531 | /* |
1532 | * can_attach beforehand should guarantee that this doesn't | |
1533 | * fail. TODO: have a better way to handle failure here | |
1534 | */ | |
1535 | WARN_ON_ONCE(set_cpus_allowed_ptr(task, cpus_attach)); | |
1536 | ||
1537 | cpuset_change_task_nodemask(task, &cpuset_attach_nodemask_to); | |
1538 | cpuset_update_task_spread_flag(cs, task); | |
1539 | } | |
22fb52dd | 1540 | |
f780bdb7 | 1541 | /* |
4530eddb TH |
1542 | * Change mm for all threadgroup leaders. This is expensive and may |
1543 | * sleep and should be moved outside migration path proper. | |
f780bdb7 | 1544 | */ |
ae1c8023 | 1545 | cpuset_attach_nodemask_to = cs->effective_mems; |
1f7dd3e5 | 1546 | cgroup_taskset_for_each_leader(leader, css, tset) { |
3df9ca0a TH |
1547 | struct mm_struct *mm = get_task_mm(leader); |
1548 | ||
1549 | if (mm) { | |
1550 | mpol_rebind_mm(mm, &cpuset_attach_nodemask_to); | |
1551 | ||
1552 | /* | |
1553 | * old_mems_allowed is the same with mems_allowed | |
1554 | * here, except if this task is being moved | |
1555 | * automatically due to hotplug. In that case | |
1556 | * @mems_allowed has been updated and is empty, so | |
1557 | * @old_mems_allowed is the right nodesets that we | |
1558 | * migrate mm from. | |
1559 | */ | |
e93ad19d | 1560 | if (is_memory_migrate(cs)) |
3df9ca0a TH |
1561 | cpuset_migrate_mm(mm, &oldcs->old_mems_allowed, |
1562 | &cpuset_attach_nodemask_to); | |
e93ad19d TH |
1563 | else |
1564 | mmput(mm); | |
f047cecf | 1565 | } |
4225399a | 1566 | } |
452477fa | 1567 | |
33ad801d | 1568 | cs->old_mems_allowed = cpuset_attach_nodemask_to; |
02bb5863 | 1569 | |
452477fa | 1570 | cs->attach_in_progress--; |
e44193d3 LZ |
1571 | if (!cs->attach_in_progress) |
1572 | wake_up(&cpuset_attach_wq); | |
5d21cc2d TH |
1573 | |
1574 | mutex_unlock(&cpuset_mutex); | |
1da177e4 LT |
1575 | } |
1576 | ||
1577 | /* The various types of files and directories in a cpuset file system */ | |
1578 | ||
1579 | typedef enum { | |
45b07ef3 | 1580 | FILE_MEMORY_MIGRATE, |
1da177e4 LT |
1581 | FILE_CPULIST, |
1582 | FILE_MEMLIST, | |
afd1a8b3 LZ |
1583 | FILE_EFFECTIVE_CPULIST, |
1584 | FILE_EFFECTIVE_MEMLIST, | |
1da177e4 LT |
1585 | FILE_CPU_EXCLUSIVE, |
1586 | FILE_MEM_EXCLUSIVE, | |
78608366 | 1587 | FILE_MEM_HARDWALL, |
029190c5 | 1588 | FILE_SCHED_LOAD_BALANCE, |
1d3504fc | 1589 | FILE_SCHED_RELAX_DOMAIN_LEVEL, |
3e0d98b9 PJ |
1590 | FILE_MEMORY_PRESSURE_ENABLED, |
1591 | FILE_MEMORY_PRESSURE, | |
825a46af PJ |
1592 | FILE_SPREAD_PAGE, |
1593 | FILE_SPREAD_SLAB, | |
1da177e4 LT |
1594 | } cpuset_filetype_t; |
1595 | ||
182446d0 TH |
1596 | static int cpuset_write_u64(struct cgroup_subsys_state *css, struct cftype *cft, |
1597 | u64 val) | |
700fe1ab | 1598 | { |
182446d0 | 1599 | struct cpuset *cs = css_cs(css); |
700fe1ab | 1600 | cpuset_filetype_t type = cft->private; |
a903f086 | 1601 | int retval = 0; |
700fe1ab | 1602 | |
5d21cc2d | 1603 | mutex_lock(&cpuset_mutex); |
a903f086 LZ |
1604 | if (!is_cpuset_online(cs)) { |
1605 | retval = -ENODEV; | |
5d21cc2d | 1606 | goto out_unlock; |
a903f086 | 1607 | } |
700fe1ab PM |
1608 | |
1609 | switch (type) { | |
1da177e4 | 1610 | case FILE_CPU_EXCLUSIVE: |
700fe1ab | 1611 | retval = update_flag(CS_CPU_EXCLUSIVE, cs, val); |
1da177e4 LT |
1612 | break; |
1613 | case FILE_MEM_EXCLUSIVE: | |
700fe1ab | 1614 | retval = update_flag(CS_MEM_EXCLUSIVE, cs, val); |
1da177e4 | 1615 | break; |
78608366 PM |
1616 | case FILE_MEM_HARDWALL: |
1617 | retval = update_flag(CS_MEM_HARDWALL, cs, val); | |
1618 | break; | |
029190c5 | 1619 | case FILE_SCHED_LOAD_BALANCE: |
700fe1ab | 1620 | retval = update_flag(CS_SCHED_LOAD_BALANCE, cs, val); |
1d3504fc | 1621 | break; |
45b07ef3 | 1622 | case FILE_MEMORY_MIGRATE: |
700fe1ab | 1623 | retval = update_flag(CS_MEMORY_MIGRATE, cs, val); |
45b07ef3 | 1624 | break; |
3e0d98b9 | 1625 | case FILE_MEMORY_PRESSURE_ENABLED: |
700fe1ab | 1626 | cpuset_memory_pressure_enabled = !!val; |
3e0d98b9 | 1627 | break; |
825a46af | 1628 | case FILE_SPREAD_PAGE: |
700fe1ab | 1629 | retval = update_flag(CS_SPREAD_PAGE, cs, val); |
825a46af PJ |
1630 | break; |
1631 | case FILE_SPREAD_SLAB: | |
700fe1ab | 1632 | retval = update_flag(CS_SPREAD_SLAB, cs, val); |
825a46af | 1633 | break; |
1da177e4 LT |
1634 | default: |
1635 | retval = -EINVAL; | |
700fe1ab | 1636 | break; |
1da177e4 | 1637 | } |
5d21cc2d TH |
1638 | out_unlock: |
1639 | mutex_unlock(&cpuset_mutex); | |
1da177e4 LT |
1640 | return retval; |
1641 | } | |
1642 | ||
182446d0 TH |
1643 | static int cpuset_write_s64(struct cgroup_subsys_state *css, struct cftype *cft, |
1644 | s64 val) | |
5be7a479 | 1645 | { |
182446d0 | 1646 | struct cpuset *cs = css_cs(css); |
5be7a479 | 1647 | cpuset_filetype_t type = cft->private; |
5d21cc2d | 1648 | int retval = -ENODEV; |
5be7a479 | 1649 | |
5d21cc2d TH |
1650 | mutex_lock(&cpuset_mutex); |
1651 | if (!is_cpuset_online(cs)) | |
1652 | goto out_unlock; | |
e3712395 | 1653 | |
5be7a479 PM |
1654 | switch (type) { |
1655 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1656 | retval = update_relax_domain_level(cs, val); | |
1657 | break; | |
1658 | default: | |
1659 | retval = -EINVAL; | |
1660 | break; | |
1661 | } | |
5d21cc2d TH |
1662 | out_unlock: |
1663 | mutex_unlock(&cpuset_mutex); | |
5be7a479 PM |
1664 | return retval; |
1665 | } | |
1666 | ||
e3712395 PM |
1667 | /* |
1668 | * Common handling for a write to a "cpus" or "mems" file. | |
1669 | */ | |
451af504 TH |
1670 | static ssize_t cpuset_write_resmask(struct kernfs_open_file *of, |
1671 | char *buf, size_t nbytes, loff_t off) | |
e3712395 | 1672 | { |
451af504 | 1673 | struct cpuset *cs = css_cs(of_css(of)); |
645fcc9d | 1674 | struct cpuset *trialcs; |
5d21cc2d | 1675 | int retval = -ENODEV; |
e3712395 | 1676 | |
451af504 TH |
1677 | buf = strstrip(buf); |
1678 | ||
3a5a6d0c TH |
1679 | /* |
1680 | * CPU or memory hotunplug may leave @cs w/o any execution | |
1681 | * resources, in which case the hotplug code asynchronously updates | |
1682 | * configuration and transfers all tasks to the nearest ancestor | |
1683 | * which can execute. | |
1684 | * | |
1685 | * As writes to "cpus" or "mems" may restore @cs's execution | |
1686 | * resources, wait for the previously scheduled operations before | |
1687 | * proceeding, so that we don't end up keep removing tasks added | |
1688 | * after execution capability is restored. | |
76bb5ab8 TH |
1689 | * |
1690 | * cpuset_hotplug_work calls back into cgroup core via | |
1691 | * cgroup_transfer_tasks() and waiting for it from a cgroupfs | |
1692 | * operation like this one can lead to a deadlock through kernfs | |
1693 | * active_ref protection. Let's break the protection. Losing the | |
1694 | * protection is okay as we check whether @cs is online after | |
1695 | * grabbing cpuset_mutex anyway. This only happens on the legacy | |
1696 | * hierarchies. | |
3a5a6d0c | 1697 | */ |
76bb5ab8 TH |
1698 | css_get(&cs->css); |
1699 | kernfs_break_active_protection(of->kn); | |
3a5a6d0c TH |
1700 | flush_work(&cpuset_hotplug_work); |
1701 | ||
5d21cc2d TH |
1702 | mutex_lock(&cpuset_mutex); |
1703 | if (!is_cpuset_online(cs)) | |
1704 | goto out_unlock; | |
e3712395 | 1705 | |
645fcc9d | 1706 | trialcs = alloc_trial_cpuset(cs); |
b75f38d6 LZ |
1707 | if (!trialcs) { |
1708 | retval = -ENOMEM; | |
5d21cc2d | 1709 | goto out_unlock; |
b75f38d6 | 1710 | } |
645fcc9d | 1711 | |
451af504 | 1712 | switch (of_cft(of)->private) { |
e3712395 | 1713 | case FILE_CPULIST: |
645fcc9d | 1714 | retval = update_cpumask(cs, trialcs, buf); |
e3712395 PM |
1715 | break; |
1716 | case FILE_MEMLIST: | |
645fcc9d | 1717 | retval = update_nodemask(cs, trialcs, buf); |
e3712395 PM |
1718 | break; |
1719 | default: | |
1720 | retval = -EINVAL; | |
1721 | break; | |
1722 | } | |
645fcc9d LZ |
1723 | |
1724 | free_trial_cpuset(trialcs); | |
5d21cc2d TH |
1725 | out_unlock: |
1726 | mutex_unlock(&cpuset_mutex); | |
76bb5ab8 TH |
1727 | kernfs_unbreak_active_protection(of->kn); |
1728 | css_put(&cs->css); | |
e93ad19d | 1729 | flush_workqueue(cpuset_migrate_mm_wq); |
451af504 | 1730 | return retval ?: nbytes; |
e3712395 PM |
1731 | } |
1732 | ||
1da177e4 LT |
1733 | /* |
1734 | * These ascii lists should be read in a single call, by using a user | |
1735 | * buffer large enough to hold the entire map. If read in smaller | |
1736 | * chunks, there is no guarantee of atomicity. Since the display format | |
1737 | * used, list of ranges of sequential numbers, is variable length, | |
1738 | * and since these maps can change value dynamically, one could read | |
1739 | * gibberish by doing partial reads while a list was changing. | |
1da177e4 | 1740 | */ |
2da8ca82 | 1741 | static int cpuset_common_seq_show(struct seq_file *sf, void *v) |
1da177e4 | 1742 | { |
2da8ca82 TH |
1743 | struct cpuset *cs = css_cs(seq_css(sf)); |
1744 | cpuset_filetype_t type = seq_cft(sf)->private; | |
51ffe411 | 1745 | int ret = 0; |
1da177e4 | 1746 | |
8447a0fe | 1747 | spin_lock_irq(&callback_lock); |
1da177e4 LT |
1748 | |
1749 | switch (type) { | |
1750 | case FILE_CPULIST: | |
e8e6d97c | 1751 | seq_printf(sf, "%*pbl\n", cpumask_pr_args(cs->cpus_allowed)); |
1da177e4 LT |
1752 | break; |
1753 | case FILE_MEMLIST: | |
e8e6d97c | 1754 | seq_printf(sf, "%*pbl\n", nodemask_pr_args(&cs->mems_allowed)); |
1da177e4 | 1755 | break; |
afd1a8b3 | 1756 | case FILE_EFFECTIVE_CPULIST: |
e8e6d97c | 1757 | seq_printf(sf, "%*pbl\n", cpumask_pr_args(cs->effective_cpus)); |
afd1a8b3 LZ |
1758 | break; |
1759 | case FILE_EFFECTIVE_MEMLIST: | |
e8e6d97c | 1760 | seq_printf(sf, "%*pbl\n", nodemask_pr_args(&cs->effective_mems)); |
afd1a8b3 | 1761 | break; |
1da177e4 | 1762 | default: |
51ffe411 | 1763 | ret = -EINVAL; |
1da177e4 | 1764 | } |
1da177e4 | 1765 | |
8447a0fe | 1766 | spin_unlock_irq(&callback_lock); |
51ffe411 | 1767 | return ret; |
1da177e4 LT |
1768 | } |
1769 | ||
182446d0 | 1770 | static u64 cpuset_read_u64(struct cgroup_subsys_state *css, struct cftype *cft) |
700fe1ab | 1771 | { |
182446d0 | 1772 | struct cpuset *cs = css_cs(css); |
700fe1ab PM |
1773 | cpuset_filetype_t type = cft->private; |
1774 | switch (type) { | |
1775 | case FILE_CPU_EXCLUSIVE: | |
1776 | return is_cpu_exclusive(cs); | |
1777 | case FILE_MEM_EXCLUSIVE: | |
1778 | return is_mem_exclusive(cs); | |
78608366 PM |
1779 | case FILE_MEM_HARDWALL: |
1780 | return is_mem_hardwall(cs); | |
700fe1ab PM |
1781 | case FILE_SCHED_LOAD_BALANCE: |
1782 | return is_sched_load_balance(cs); | |
1783 | case FILE_MEMORY_MIGRATE: | |
1784 | return is_memory_migrate(cs); | |
1785 | case FILE_MEMORY_PRESSURE_ENABLED: | |
1786 | return cpuset_memory_pressure_enabled; | |
1787 | case FILE_MEMORY_PRESSURE: | |
1788 | return fmeter_getrate(&cs->fmeter); | |
1789 | case FILE_SPREAD_PAGE: | |
1790 | return is_spread_page(cs); | |
1791 | case FILE_SPREAD_SLAB: | |
1792 | return is_spread_slab(cs); | |
1793 | default: | |
1794 | BUG(); | |
1795 | } | |
cf417141 MK |
1796 | |
1797 | /* Unreachable but makes gcc happy */ | |
1798 | return 0; | |
700fe1ab | 1799 | } |
1da177e4 | 1800 | |
182446d0 | 1801 | static s64 cpuset_read_s64(struct cgroup_subsys_state *css, struct cftype *cft) |
5be7a479 | 1802 | { |
182446d0 | 1803 | struct cpuset *cs = css_cs(css); |
5be7a479 PM |
1804 | cpuset_filetype_t type = cft->private; |
1805 | switch (type) { | |
1806 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1807 | return cs->relax_domain_level; | |
1808 | default: | |
1809 | BUG(); | |
1810 | } | |
cf417141 MK |
1811 | |
1812 | /* Unrechable but makes gcc happy */ | |
1813 | return 0; | |
5be7a479 PM |
1814 | } |
1815 | ||
1da177e4 LT |
1816 | |
1817 | /* | |
1818 | * for the common functions, 'private' gives the type of file | |
1819 | */ | |
1820 | ||
addf2c73 PM |
1821 | static struct cftype files[] = { |
1822 | { | |
1823 | .name = "cpus", | |
2da8ca82 | 1824 | .seq_show = cpuset_common_seq_show, |
451af504 | 1825 | .write = cpuset_write_resmask, |
e3712395 | 1826 | .max_write_len = (100U + 6 * NR_CPUS), |
addf2c73 PM |
1827 | .private = FILE_CPULIST, |
1828 | }, | |
1829 | ||
1830 | { | |
1831 | .name = "mems", | |
2da8ca82 | 1832 | .seq_show = cpuset_common_seq_show, |
451af504 | 1833 | .write = cpuset_write_resmask, |
e3712395 | 1834 | .max_write_len = (100U + 6 * MAX_NUMNODES), |
addf2c73 PM |
1835 | .private = FILE_MEMLIST, |
1836 | }, | |
1837 | ||
afd1a8b3 LZ |
1838 | { |
1839 | .name = "effective_cpus", | |
1840 | .seq_show = cpuset_common_seq_show, | |
1841 | .private = FILE_EFFECTIVE_CPULIST, | |
1842 | }, | |
1843 | ||
1844 | { | |
1845 | .name = "effective_mems", | |
1846 | .seq_show = cpuset_common_seq_show, | |
1847 | .private = FILE_EFFECTIVE_MEMLIST, | |
1848 | }, | |
1849 | ||
addf2c73 PM |
1850 | { |
1851 | .name = "cpu_exclusive", | |
1852 | .read_u64 = cpuset_read_u64, | |
1853 | .write_u64 = cpuset_write_u64, | |
1854 | .private = FILE_CPU_EXCLUSIVE, | |
1855 | }, | |
1856 | ||
1857 | { | |
1858 | .name = "mem_exclusive", | |
1859 | .read_u64 = cpuset_read_u64, | |
1860 | .write_u64 = cpuset_write_u64, | |
1861 | .private = FILE_MEM_EXCLUSIVE, | |
1862 | }, | |
1863 | ||
78608366 PM |
1864 | { |
1865 | .name = "mem_hardwall", | |
1866 | .read_u64 = cpuset_read_u64, | |
1867 | .write_u64 = cpuset_write_u64, | |
1868 | .private = FILE_MEM_HARDWALL, | |
1869 | }, | |
1870 | ||
addf2c73 PM |
1871 | { |
1872 | .name = "sched_load_balance", | |
1873 | .read_u64 = cpuset_read_u64, | |
1874 | .write_u64 = cpuset_write_u64, | |
1875 | .private = FILE_SCHED_LOAD_BALANCE, | |
1876 | }, | |
1877 | ||
1878 | { | |
1879 | .name = "sched_relax_domain_level", | |
5be7a479 PM |
1880 | .read_s64 = cpuset_read_s64, |
1881 | .write_s64 = cpuset_write_s64, | |
addf2c73 PM |
1882 | .private = FILE_SCHED_RELAX_DOMAIN_LEVEL, |
1883 | }, | |
1884 | ||
1885 | { | |
1886 | .name = "memory_migrate", | |
1887 | .read_u64 = cpuset_read_u64, | |
1888 | .write_u64 = cpuset_write_u64, | |
1889 | .private = FILE_MEMORY_MIGRATE, | |
1890 | }, | |
1891 | ||
1892 | { | |
1893 | .name = "memory_pressure", | |
1894 | .read_u64 = cpuset_read_u64, | |
addf2c73 PM |
1895 | }, |
1896 | ||
1897 | { | |
1898 | .name = "memory_spread_page", | |
1899 | .read_u64 = cpuset_read_u64, | |
1900 | .write_u64 = cpuset_write_u64, | |
1901 | .private = FILE_SPREAD_PAGE, | |
1902 | }, | |
1903 | ||
1904 | { | |
1905 | .name = "memory_spread_slab", | |
1906 | .read_u64 = cpuset_read_u64, | |
1907 | .write_u64 = cpuset_write_u64, | |
1908 | .private = FILE_SPREAD_SLAB, | |
1909 | }, | |
3e0d98b9 | 1910 | |
4baf6e33 TH |
1911 | { |
1912 | .name = "memory_pressure_enabled", | |
1913 | .flags = CFTYPE_ONLY_ON_ROOT, | |
1914 | .read_u64 = cpuset_read_u64, | |
1915 | .write_u64 = cpuset_write_u64, | |
1916 | .private = FILE_MEMORY_PRESSURE_ENABLED, | |
1917 | }, | |
1da177e4 | 1918 | |
4baf6e33 TH |
1919 | { } /* terminate */ |
1920 | }; | |
1da177e4 LT |
1921 | |
1922 | /* | |
92fb9748 | 1923 | * cpuset_css_alloc - allocate a cpuset css |
c9e5fe66 | 1924 | * cgrp: control group that the new cpuset will be part of |
1da177e4 LT |
1925 | */ |
1926 | ||
eb95419b TH |
1927 | static struct cgroup_subsys_state * |
1928 | cpuset_css_alloc(struct cgroup_subsys_state *parent_css) | |
1da177e4 | 1929 | { |
c8f699bb | 1930 | struct cpuset *cs; |
1da177e4 | 1931 | |
eb95419b | 1932 | if (!parent_css) |
8793d854 | 1933 | return &top_cpuset.css; |
033fa1c5 | 1934 | |
c8f699bb | 1935 | cs = kzalloc(sizeof(*cs), GFP_KERNEL); |
1da177e4 | 1936 | if (!cs) |
8793d854 | 1937 | return ERR_PTR(-ENOMEM); |
e2b9a3d7 LZ |
1938 | if (!alloc_cpumask_var(&cs->cpus_allowed, GFP_KERNEL)) |
1939 | goto free_cs; | |
1940 | if (!alloc_cpumask_var(&cs->effective_cpus, GFP_KERNEL)) | |
1941 | goto free_cpus; | |
1da177e4 | 1942 | |
029190c5 | 1943 | set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); |
300ed6cb | 1944 | cpumask_clear(cs->cpus_allowed); |
f9a86fcb | 1945 | nodes_clear(cs->mems_allowed); |
e2b9a3d7 LZ |
1946 | cpumask_clear(cs->effective_cpus); |
1947 | nodes_clear(cs->effective_mems); | |
3e0d98b9 | 1948 | fmeter_init(&cs->fmeter); |
1d3504fc | 1949 | cs->relax_domain_level = -1; |
1da177e4 | 1950 | |
c8f699bb | 1951 | return &cs->css; |
e2b9a3d7 LZ |
1952 | |
1953 | free_cpus: | |
1954 | free_cpumask_var(cs->cpus_allowed); | |
1955 | free_cs: | |
1956 | kfree(cs); | |
1957 | return ERR_PTR(-ENOMEM); | |
c8f699bb TH |
1958 | } |
1959 | ||
eb95419b | 1960 | static int cpuset_css_online(struct cgroup_subsys_state *css) |
c8f699bb | 1961 | { |
eb95419b | 1962 | struct cpuset *cs = css_cs(css); |
c431069f | 1963 | struct cpuset *parent = parent_cs(cs); |
ae8086ce | 1964 | struct cpuset *tmp_cs; |
492eb21b | 1965 | struct cgroup_subsys_state *pos_css; |
c8f699bb TH |
1966 | |
1967 | if (!parent) | |
1968 | return 0; | |
1969 | ||
5d21cc2d TH |
1970 | mutex_lock(&cpuset_mutex); |
1971 | ||
efeb77b2 | 1972 | set_bit(CS_ONLINE, &cs->flags); |
c8f699bb TH |
1973 | if (is_spread_page(parent)) |
1974 | set_bit(CS_SPREAD_PAGE, &cs->flags); | |
1975 | if (is_spread_slab(parent)) | |
1976 | set_bit(CS_SPREAD_SLAB, &cs->flags); | |
1da177e4 | 1977 | |
664eedde | 1978 | cpuset_inc(); |
033fa1c5 | 1979 | |
8447a0fe | 1980 | spin_lock_irq(&callback_lock); |
9e10a130 | 1981 | if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys)) { |
e2b9a3d7 LZ |
1982 | cpumask_copy(cs->effective_cpus, parent->effective_cpus); |
1983 | cs->effective_mems = parent->effective_mems; | |
1984 | } | |
8447a0fe | 1985 | spin_unlock_irq(&callback_lock); |
e2b9a3d7 | 1986 | |
eb95419b | 1987 | if (!test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags)) |
5d21cc2d | 1988 | goto out_unlock; |
033fa1c5 TH |
1989 | |
1990 | /* | |
1991 | * Clone @parent's configuration if CGRP_CPUSET_CLONE_CHILDREN is | |
1992 | * set. This flag handling is implemented in cgroup core for | |
1993 | * histrical reasons - the flag may be specified during mount. | |
1994 | * | |
1995 | * Currently, if any sibling cpusets have exclusive cpus or mem, we | |
1996 | * refuse to clone the configuration - thereby refusing the task to | |
1997 | * be entered, and as a result refusing the sys_unshare() or | |
1998 | * clone() which initiated it. If this becomes a problem for some | |
1999 | * users who wish to allow that scenario, then this could be | |
2000 | * changed to grant parent->cpus_allowed-sibling_cpus_exclusive | |
2001 | * (and likewise for mems) to the new cgroup. | |
2002 | */ | |
ae8086ce | 2003 | rcu_read_lock(); |
492eb21b | 2004 | cpuset_for_each_child(tmp_cs, pos_css, parent) { |
ae8086ce TH |
2005 | if (is_mem_exclusive(tmp_cs) || is_cpu_exclusive(tmp_cs)) { |
2006 | rcu_read_unlock(); | |
5d21cc2d | 2007 | goto out_unlock; |
ae8086ce | 2008 | } |
033fa1c5 | 2009 | } |
ae8086ce | 2010 | rcu_read_unlock(); |
033fa1c5 | 2011 | |
8447a0fe | 2012 | spin_lock_irq(&callback_lock); |
033fa1c5 | 2013 | cs->mems_allowed = parent->mems_allowed; |
790317e1 | 2014 | cs->effective_mems = parent->mems_allowed; |
033fa1c5 | 2015 | cpumask_copy(cs->cpus_allowed, parent->cpus_allowed); |
790317e1 | 2016 | cpumask_copy(cs->effective_cpus, parent->cpus_allowed); |
cea74465 | 2017 | spin_unlock_irq(&callback_lock); |
5d21cc2d TH |
2018 | out_unlock: |
2019 | mutex_unlock(&cpuset_mutex); | |
c8f699bb TH |
2020 | return 0; |
2021 | } | |
2022 | ||
0b9e6965 ZH |
2023 | /* |
2024 | * If the cpuset being removed has its flag 'sched_load_balance' | |
2025 | * enabled, then simulate turning sched_load_balance off, which | |
2026 | * will call rebuild_sched_domains_locked(). | |
2027 | */ | |
2028 | ||
eb95419b | 2029 | static void cpuset_css_offline(struct cgroup_subsys_state *css) |
c8f699bb | 2030 | { |
eb95419b | 2031 | struct cpuset *cs = css_cs(css); |
c8f699bb | 2032 | |
5d21cc2d | 2033 | mutex_lock(&cpuset_mutex); |
c8f699bb TH |
2034 | |
2035 | if (is_sched_load_balance(cs)) | |
2036 | update_flag(CS_SCHED_LOAD_BALANCE, cs, 0); | |
2037 | ||
664eedde | 2038 | cpuset_dec(); |
efeb77b2 | 2039 | clear_bit(CS_ONLINE, &cs->flags); |
c8f699bb | 2040 | |
5d21cc2d | 2041 | mutex_unlock(&cpuset_mutex); |
1da177e4 LT |
2042 | } |
2043 | ||
eb95419b | 2044 | static void cpuset_css_free(struct cgroup_subsys_state *css) |
1da177e4 | 2045 | { |
eb95419b | 2046 | struct cpuset *cs = css_cs(css); |
1da177e4 | 2047 | |
e2b9a3d7 | 2048 | free_cpumask_var(cs->effective_cpus); |
300ed6cb | 2049 | free_cpumask_var(cs->cpus_allowed); |
8793d854 | 2050 | kfree(cs); |
1da177e4 LT |
2051 | } |
2052 | ||
39bd0d15 LZ |
2053 | static void cpuset_bind(struct cgroup_subsys_state *root_css) |
2054 | { | |
2055 | mutex_lock(&cpuset_mutex); | |
8447a0fe | 2056 | spin_lock_irq(&callback_lock); |
39bd0d15 | 2057 | |
9e10a130 | 2058 | if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys)) { |
39bd0d15 LZ |
2059 | cpumask_copy(top_cpuset.cpus_allowed, cpu_possible_mask); |
2060 | top_cpuset.mems_allowed = node_possible_map; | |
2061 | } else { | |
2062 | cpumask_copy(top_cpuset.cpus_allowed, | |
2063 | top_cpuset.effective_cpus); | |
2064 | top_cpuset.mems_allowed = top_cpuset.effective_mems; | |
2065 | } | |
2066 | ||
8447a0fe | 2067 | spin_unlock_irq(&callback_lock); |
39bd0d15 LZ |
2068 | mutex_unlock(&cpuset_mutex); |
2069 | } | |
2070 | ||
06f4e948 ZL |
2071 | /* |
2072 | * Make sure the new task conform to the current state of its parent, | |
2073 | * which could have been changed by cpuset just after it inherits the | |
2074 | * state from the parent and before it sits on the cgroup's task list. | |
2075 | */ | |
8a15b817 | 2076 | static void cpuset_fork(struct task_struct *task) |
06f4e948 ZL |
2077 | { |
2078 | if (task_css_is_root(task, cpuset_cgrp_id)) | |
2079 | return; | |
2080 | ||
2081 | set_cpus_allowed_ptr(task, ¤t->cpus_allowed); | |
2082 | task->mems_allowed = current->mems_allowed; | |
2083 | } | |
2084 | ||
073219e9 | 2085 | struct cgroup_subsys cpuset_cgrp_subsys = { |
39bd0d15 LZ |
2086 | .css_alloc = cpuset_css_alloc, |
2087 | .css_online = cpuset_css_online, | |
2088 | .css_offline = cpuset_css_offline, | |
2089 | .css_free = cpuset_css_free, | |
2090 | .can_attach = cpuset_can_attach, | |
2091 | .cancel_attach = cpuset_cancel_attach, | |
2092 | .attach = cpuset_attach, | |
5cf1cacb | 2093 | .post_attach = cpuset_post_attach, |
39bd0d15 | 2094 | .bind = cpuset_bind, |
06f4e948 | 2095 | .fork = cpuset_fork, |
5577964e | 2096 | .legacy_cftypes = files, |
b38e42e9 | 2097 | .early_init = true, |
8793d854 PM |
2098 | }; |
2099 | ||
1da177e4 LT |
2100 | /** |
2101 | * cpuset_init - initialize cpusets at system boot | |
2102 | * | |
2103 | * Description: Initialize top_cpuset and the cpuset internal file system, | |
2104 | **/ | |
2105 | ||
2106 | int __init cpuset_init(void) | |
2107 | { | |
8793d854 | 2108 | int err = 0; |
1da177e4 | 2109 | |
75fa8e5d NMG |
2110 | BUG_ON(!alloc_cpumask_var(&top_cpuset.cpus_allowed, GFP_KERNEL)); |
2111 | BUG_ON(!alloc_cpumask_var(&top_cpuset.effective_cpus, GFP_KERNEL)); | |
58568d2a | 2112 | |
300ed6cb | 2113 | cpumask_setall(top_cpuset.cpus_allowed); |
f9a86fcb | 2114 | nodes_setall(top_cpuset.mems_allowed); |
e2b9a3d7 LZ |
2115 | cpumask_setall(top_cpuset.effective_cpus); |
2116 | nodes_setall(top_cpuset.effective_mems); | |
1da177e4 | 2117 | |
3e0d98b9 | 2118 | fmeter_init(&top_cpuset.fmeter); |
029190c5 | 2119 | set_bit(CS_SCHED_LOAD_BALANCE, &top_cpuset.flags); |
1d3504fc | 2120 | top_cpuset.relax_domain_level = -1; |
1da177e4 | 2121 | |
1da177e4 LT |
2122 | err = register_filesystem(&cpuset_fs_type); |
2123 | if (err < 0) | |
8793d854 PM |
2124 | return err; |
2125 | ||
75fa8e5d | 2126 | BUG_ON(!alloc_cpumask_var(&cpus_attach, GFP_KERNEL)); |
2341d1b6 | 2127 | |
8793d854 | 2128 | return 0; |
1da177e4 LT |
2129 | } |
2130 | ||
b1aac8bb | 2131 | /* |
cf417141 | 2132 | * If CPU and/or memory hotplug handlers, below, unplug any CPUs |
b1aac8bb PJ |
2133 | * or memory nodes, we need to walk over the cpuset hierarchy, |
2134 | * removing that CPU or node from all cpusets. If this removes the | |
956db3ca CW |
2135 | * last CPU or node from a cpuset, then move the tasks in the empty |
2136 | * cpuset to its next-highest non-empty parent. | |
b1aac8bb | 2137 | */ |
956db3ca CW |
2138 | static void remove_tasks_in_empty_cpuset(struct cpuset *cs) |
2139 | { | |
2140 | struct cpuset *parent; | |
2141 | ||
956db3ca CW |
2142 | /* |
2143 | * Find its next-highest non-empty parent, (top cpuset | |
2144 | * has online cpus, so can't be empty). | |
2145 | */ | |
c431069f | 2146 | parent = parent_cs(cs); |
300ed6cb | 2147 | while (cpumask_empty(parent->cpus_allowed) || |
b4501295 | 2148 | nodes_empty(parent->mems_allowed)) |
c431069f | 2149 | parent = parent_cs(parent); |
956db3ca | 2150 | |
8cc99345 | 2151 | if (cgroup_transfer_tasks(parent->css.cgroup, cs->css.cgroup)) { |
12d3089c | 2152 | pr_err("cpuset: failed to transfer tasks out of empty cpuset "); |
e61734c5 TH |
2153 | pr_cont_cgroup_name(cs->css.cgroup); |
2154 | pr_cont("\n"); | |
8cc99345 | 2155 | } |
956db3ca CW |
2156 | } |
2157 | ||
be4c9dd7 LZ |
2158 | static void |
2159 | hotplug_update_tasks_legacy(struct cpuset *cs, | |
2160 | struct cpumask *new_cpus, nodemask_t *new_mems, | |
2161 | bool cpus_updated, bool mems_updated) | |
390a36aa LZ |
2162 | { |
2163 | bool is_empty; | |
2164 | ||
8447a0fe | 2165 | spin_lock_irq(&callback_lock); |
be4c9dd7 LZ |
2166 | cpumask_copy(cs->cpus_allowed, new_cpus); |
2167 | cpumask_copy(cs->effective_cpus, new_cpus); | |
2168 | cs->mems_allowed = *new_mems; | |
2169 | cs->effective_mems = *new_mems; | |
8447a0fe | 2170 | spin_unlock_irq(&callback_lock); |
390a36aa LZ |
2171 | |
2172 | /* | |
2173 | * Don't call update_tasks_cpumask() if the cpuset becomes empty, | |
2174 | * as the tasks will be migratecd to an ancestor. | |
2175 | */ | |
be4c9dd7 | 2176 | if (cpus_updated && !cpumask_empty(cs->cpus_allowed)) |
390a36aa | 2177 | update_tasks_cpumask(cs); |
be4c9dd7 | 2178 | if (mems_updated && !nodes_empty(cs->mems_allowed)) |
390a36aa LZ |
2179 | update_tasks_nodemask(cs); |
2180 | ||
2181 | is_empty = cpumask_empty(cs->cpus_allowed) || | |
2182 | nodes_empty(cs->mems_allowed); | |
2183 | ||
2184 | mutex_unlock(&cpuset_mutex); | |
2185 | ||
2186 | /* | |
2187 | * Move tasks to the nearest ancestor with execution resources, | |
2188 | * This is full cgroup operation which will also call back into | |
2189 | * cpuset. Should be done outside any lock. | |
2190 | */ | |
2191 | if (is_empty) | |
2192 | remove_tasks_in_empty_cpuset(cs); | |
2193 | ||
2194 | mutex_lock(&cpuset_mutex); | |
2195 | } | |
2196 | ||
be4c9dd7 LZ |
2197 | static void |
2198 | hotplug_update_tasks(struct cpuset *cs, | |
2199 | struct cpumask *new_cpus, nodemask_t *new_mems, | |
2200 | bool cpus_updated, bool mems_updated) | |
390a36aa | 2201 | { |
be4c9dd7 LZ |
2202 | if (cpumask_empty(new_cpus)) |
2203 | cpumask_copy(new_cpus, parent_cs(cs)->effective_cpus); | |
2204 | if (nodes_empty(*new_mems)) | |
2205 | *new_mems = parent_cs(cs)->effective_mems; | |
2206 | ||
8447a0fe | 2207 | spin_lock_irq(&callback_lock); |
be4c9dd7 LZ |
2208 | cpumask_copy(cs->effective_cpus, new_cpus); |
2209 | cs->effective_mems = *new_mems; | |
8447a0fe | 2210 | spin_unlock_irq(&callback_lock); |
390a36aa | 2211 | |
be4c9dd7 | 2212 | if (cpus_updated) |
390a36aa | 2213 | update_tasks_cpumask(cs); |
be4c9dd7 | 2214 | if (mems_updated) |
390a36aa LZ |
2215 | update_tasks_nodemask(cs); |
2216 | } | |
2217 | ||
deb7aa30 | 2218 | /** |
388afd85 | 2219 | * cpuset_hotplug_update_tasks - update tasks in a cpuset for hotunplug |
deb7aa30 | 2220 | * @cs: cpuset in interest |
956db3ca | 2221 | * |
deb7aa30 TH |
2222 | * Compare @cs's cpu and mem masks against top_cpuset and if some have gone |
2223 | * offline, update @cs accordingly. If @cs ends up with no CPU or memory, | |
2224 | * all its tasks are moved to the nearest ancestor with both resources. | |
80d1fa64 | 2225 | */ |
388afd85 | 2226 | static void cpuset_hotplug_update_tasks(struct cpuset *cs) |
80d1fa64 | 2227 | { |
be4c9dd7 LZ |
2228 | static cpumask_t new_cpus; |
2229 | static nodemask_t new_mems; | |
2230 | bool cpus_updated; | |
2231 | bool mems_updated; | |
e44193d3 LZ |
2232 | retry: |
2233 | wait_event(cpuset_attach_wq, cs->attach_in_progress == 0); | |
80d1fa64 | 2234 | |
5d21cc2d | 2235 | mutex_lock(&cpuset_mutex); |
7ddf96b0 | 2236 | |
e44193d3 LZ |
2237 | /* |
2238 | * We have raced with task attaching. We wait until attaching | |
2239 | * is finished, so we won't attach a task to an empty cpuset. | |
2240 | */ | |
2241 | if (cs->attach_in_progress) { | |
2242 | mutex_unlock(&cpuset_mutex); | |
2243 | goto retry; | |
2244 | } | |
2245 | ||
be4c9dd7 LZ |
2246 | cpumask_and(&new_cpus, cs->cpus_allowed, parent_cs(cs)->effective_cpus); |
2247 | nodes_and(new_mems, cs->mems_allowed, parent_cs(cs)->effective_mems); | |
80d1fa64 | 2248 | |
be4c9dd7 LZ |
2249 | cpus_updated = !cpumask_equal(&new_cpus, cs->effective_cpus); |
2250 | mems_updated = !nodes_equal(new_mems, cs->effective_mems); | |
deb7aa30 | 2251 | |
9e10a130 | 2252 | if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys)) |
be4c9dd7 LZ |
2253 | hotplug_update_tasks(cs, &new_cpus, &new_mems, |
2254 | cpus_updated, mems_updated); | |
390a36aa | 2255 | else |
be4c9dd7 LZ |
2256 | hotplug_update_tasks_legacy(cs, &new_cpus, &new_mems, |
2257 | cpus_updated, mems_updated); | |
8d033948 | 2258 | |
5d21cc2d | 2259 | mutex_unlock(&cpuset_mutex); |
b1aac8bb PJ |
2260 | } |
2261 | ||
deb7aa30 | 2262 | /** |
3a5a6d0c | 2263 | * cpuset_hotplug_workfn - handle CPU/memory hotunplug for a cpuset |
956db3ca | 2264 | * |
deb7aa30 TH |
2265 | * This function is called after either CPU or memory configuration has |
2266 | * changed and updates cpuset accordingly. The top_cpuset is always | |
2267 | * synchronized to cpu_active_mask and N_MEMORY, which is necessary in | |
2268 | * order to make cpusets transparent (of no affect) on systems that are | |
2269 | * actively using CPU hotplug but making no active use of cpusets. | |
956db3ca | 2270 | * |
deb7aa30 | 2271 | * Non-root cpusets are only affected by offlining. If any CPUs or memory |
388afd85 LZ |
2272 | * nodes have been taken down, cpuset_hotplug_update_tasks() is invoked on |
2273 | * all descendants. | |
956db3ca | 2274 | * |
deb7aa30 TH |
2275 | * Note that CPU offlining during suspend is ignored. We don't modify |
2276 | * cpusets across suspend/resume cycles at all. | |
956db3ca | 2277 | */ |
3a5a6d0c | 2278 | static void cpuset_hotplug_workfn(struct work_struct *work) |
b1aac8bb | 2279 | { |
5c5cc623 LZ |
2280 | static cpumask_t new_cpus; |
2281 | static nodemask_t new_mems; | |
deb7aa30 | 2282 | bool cpus_updated, mems_updated; |
9e10a130 | 2283 | bool on_dfl = cgroup_subsys_on_dfl(cpuset_cgrp_subsys); |
b1aac8bb | 2284 | |
5d21cc2d | 2285 | mutex_lock(&cpuset_mutex); |
956db3ca | 2286 | |
deb7aa30 TH |
2287 | /* fetch the available cpus/mems and find out which changed how */ |
2288 | cpumask_copy(&new_cpus, cpu_active_mask); | |
2289 | new_mems = node_states[N_MEMORY]; | |
7ddf96b0 | 2290 | |
7e88291b LZ |
2291 | cpus_updated = !cpumask_equal(top_cpuset.effective_cpus, &new_cpus); |
2292 | mems_updated = !nodes_equal(top_cpuset.effective_mems, new_mems); | |
7ddf96b0 | 2293 | |
deb7aa30 TH |
2294 | /* synchronize cpus_allowed to cpu_active_mask */ |
2295 | if (cpus_updated) { | |
8447a0fe | 2296 | spin_lock_irq(&callback_lock); |
7e88291b LZ |
2297 | if (!on_dfl) |
2298 | cpumask_copy(top_cpuset.cpus_allowed, &new_cpus); | |
1344ab9c | 2299 | cpumask_copy(top_cpuset.effective_cpus, &new_cpus); |
8447a0fe | 2300 | spin_unlock_irq(&callback_lock); |
deb7aa30 TH |
2301 | /* we don't mess with cpumasks of tasks in top_cpuset */ |
2302 | } | |
b4501295 | 2303 | |
deb7aa30 TH |
2304 | /* synchronize mems_allowed to N_MEMORY */ |
2305 | if (mems_updated) { | |
8447a0fe | 2306 | spin_lock_irq(&callback_lock); |
7e88291b LZ |
2307 | if (!on_dfl) |
2308 | top_cpuset.mems_allowed = new_mems; | |
1344ab9c | 2309 | top_cpuset.effective_mems = new_mems; |
8447a0fe | 2310 | spin_unlock_irq(&callback_lock); |
d66393e5 | 2311 | update_tasks_nodemask(&top_cpuset); |
deb7aa30 | 2312 | } |
b4501295 | 2313 | |
388afd85 LZ |
2314 | mutex_unlock(&cpuset_mutex); |
2315 | ||
5c5cc623 LZ |
2316 | /* if cpus or mems changed, we need to propagate to descendants */ |
2317 | if (cpus_updated || mems_updated) { | |
deb7aa30 | 2318 | struct cpuset *cs; |
492eb21b | 2319 | struct cgroup_subsys_state *pos_css; |
f9b4fb8d | 2320 | |
fc560a26 | 2321 | rcu_read_lock(); |
492eb21b | 2322 | cpuset_for_each_descendant_pre(cs, pos_css, &top_cpuset) { |
ec903c0c | 2323 | if (cs == &top_cpuset || !css_tryget_online(&cs->css)) |
388afd85 LZ |
2324 | continue; |
2325 | rcu_read_unlock(); | |
7ddf96b0 | 2326 | |
388afd85 | 2327 | cpuset_hotplug_update_tasks(cs); |
b4501295 | 2328 | |
388afd85 LZ |
2329 | rcu_read_lock(); |
2330 | css_put(&cs->css); | |
2331 | } | |
2332 | rcu_read_unlock(); | |
2333 | } | |
8d033948 | 2334 | |
deb7aa30 | 2335 | /* rebuild sched domains if cpus_allowed has changed */ |
e0e80a02 LZ |
2336 | if (cpus_updated) |
2337 | rebuild_sched_domains(); | |
b1aac8bb PJ |
2338 | } |
2339 | ||
30e03acd | 2340 | void cpuset_update_active_cpus(void) |
4c4d50f7 | 2341 | { |
3a5a6d0c TH |
2342 | /* |
2343 | * We're inside cpu hotplug critical region which usually nests | |
2344 | * inside cgroup synchronization. Bounce actual hotplug processing | |
2345 | * to a work item to avoid reverse locking order. | |
3a5a6d0c | 2346 | */ |
3a5a6d0c | 2347 | schedule_work(&cpuset_hotplug_work); |
4c4d50f7 | 2348 | } |
4c4d50f7 | 2349 | |
38837fc7 | 2350 | /* |
38d7bee9 LJ |
2351 | * Keep top_cpuset.mems_allowed tracking node_states[N_MEMORY]. |
2352 | * Call this routine anytime after node_states[N_MEMORY] changes. | |
a1cd2b13 | 2353 | * See cpuset_update_active_cpus() for CPU hotplug handling. |
38837fc7 | 2354 | */ |
f481891f MX |
2355 | static int cpuset_track_online_nodes(struct notifier_block *self, |
2356 | unsigned long action, void *arg) | |
38837fc7 | 2357 | { |
3a5a6d0c | 2358 | schedule_work(&cpuset_hotplug_work); |
f481891f | 2359 | return NOTIFY_OK; |
38837fc7 | 2360 | } |
d8f10cb3 AM |
2361 | |
2362 | static struct notifier_block cpuset_track_online_nodes_nb = { | |
2363 | .notifier_call = cpuset_track_online_nodes, | |
2364 | .priority = 10, /* ??! */ | |
2365 | }; | |
38837fc7 | 2366 | |
1da177e4 LT |
2367 | /** |
2368 | * cpuset_init_smp - initialize cpus_allowed | |
2369 | * | |
2370 | * Description: Finish top cpuset after cpu, node maps are initialized | |
d8f10cb3 | 2371 | */ |
1da177e4 LT |
2372 | void __init cpuset_init_smp(void) |
2373 | { | |
6ad4c188 | 2374 | cpumask_copy(top_cpuset.cpus_allowed, cpu_active_mask); |
38d7bee9 | 2375 | top_cpuset.mems_allowed = node_states[N_MEMORY]; |
33ad801d | 2376 | top_cpuset.old_mems_allowed = top_cpuset.mems_allowed; |
4c4d50f7 | 2377 | |
e2b9a3d7 LZ |
2378 | cpumask_copy(top_cpuset.effective_cpus, cpu_active_mask); |
2379 | top_cpuset.effective_mems = node_states[N_MEMORY]; | |
2380 | ||
d8f10cb3 | 2381 | register_hotmemory_notifier(&cpuset_track_online_nodes_nb); |
e93ad19d TH |
2382 | |
2383 | cpuset_migrate_mm_wq = alloc_ordered_workqueue("cpuset_migrate_mm", 0); | |
2384 | BUG_ON(!cpuset_migrate_mm_wq); | |
1da177e4 LT |
2385 | } |
2386 | ||
2387 | /** | |
1da177e4 LT |
2388 | * cpuset_cpus_allowed - return cpus_allowed mask from a tasks cpuset. |
2389 | * @tsk: pointer to task_struct from which to obtain cpuset->cpus_allowed. | |
6af866af | 2390 | * @pmask: pointer to struct cpumask variable to receive cpus_allowed set. |
1da177e4 | 2391 | * |
300ed6cb | 2392 | * Description: Returns the cpumask_var_t cpus_allowed of the cpuset |
1da177e4 | 2393 | * attached to the specified @tsk. Guaranteed to return some non-empty |
5f054e31 | 2394 | * subset of cpu_online_mask, even if this means going outside the |
1da177e4 LT |
2395 | * tasks cpuset. |
2396 | **/ | |
2397 | ||
6af866af | 2398 | void cpuset_cpus_allowed(struct task_struct *tsk, struct cpumask *pmask) |
1da177e4 | 2399 | { |
8447a0fe VD |
2400 | unsigned long flags; |
2401 | ||
2402 | spin_lock_irqsave(&callback_lock, flags); | |
b8dadcb5 | 2403 | rcu_read_lock(); |
ae1c8023 | 2404 | guarantee_online_cpus(task_cs(tsk), pmask); |
b8dadcb5 | 2405 | rcu_read_unlock(); |
8447a0fe | 2406 | spin_unlock_irqrestore(&callback_lock, flags); |
1da177e4 LT |
2407 | } |
2408 | ||
2baab4e9 | 2409 | void cpuset_cpus_allowed_fallback(struct task_struct *tsk) |
9084bb82 | 2410 | { |
9084bb82 | 2411 | rcu_read_lock(); |
ae1c8023 | 2412 | do_set_cpus_allowed(tsk, task_cs(tsk)->effective_cpus); |
9084bb82 ON |
2413 | rcu_read_unlock(); |
2414 | ||
2415 | /* | |
2416 | * We own tsk->cpus_allowed, nobody can change it under us. | |
2417 | * | |
2418 | * But we used cs && cs->cpus_allowed lockless and thus can | |
2419 | * race with cgroup_attach_task() or update_cpumask() and get | |
2420 | * the wrong tsk->cpus_allowed. However, both cases imply the | |
2421 | * subsequent cpuset_change_cpumask()->set_cpus_allowed_ptr() | |
2422 | * which takes task_rq_lock(). | |
2423 | * | |
2424 | * If we are called after it dropped the lock we must see all | |
2425 | * changes in tsk_cs()->cpus_allowed. Otherwise we can temporary | |
2426 | * set any mask even if it is not right from task_cs() pov, | |
2427 | * the pending set_cpus_allowed_ptr() will fix things. | |
2baab4e9 PZ |
2428 | * |
2429 | * select_fallback_rq() will fix things ups and set cpu_possible_mask | |
2430 | * if required. | |
9084bb82 | 2431 | */ |
9084bb82 ON |
2432 | } |
2433 | ||
8f4ab07f | 2434 | void __init cpuset_init_current_mems_allowed(void) |
1da177e4 | 2435 | { |
f9a86fcb | 2436 | nodes_setall(current->mems_allowed); |
1da177e4 LT |
2437 | } |
2438 | ||
909d75a3 PJ |
2439 | /** |
2440 | * cpuset_mems_allowed - return mems_allowed mask from a tasks cpuset. | |
2441 | * @tsk: pointer to task_struct from which to obtain cpuset->mems_allowed. | |
2442 | * | |
2443 | * Description: Returns the nodemask_t mems_allowed of the cpuset | |
2444 | * attached to the specified @tsk. Guaranteed to return some non-empty | |
38d7bee9 | 2445 | * subset of node_states[N_MEMORY], even if this means going outside the |
909d75a3 PJ |
2446 | * tasks cpuset. |
2447 | **/ | |
2448 | ||
2449 | nodemask_t cpuset_mems_allowed(struct task_struct *tsk) | |
2450 | { | |
2451 | nodemask_t mask; | |
8447a0fe | 2452 | unsigned long flags; |
909d75a3 | 2453 | |
8447a0fe | 2454 | spin_lock_irqsave(&callback_lock, flags); |
b8dadcb5 | 2455 | rcu_read_lock(); |
ae1c8023 | 2456 | guarantee_online_mems(task_cs(tsk), &mask); |
b8dadcb5 | 2457 | rcu_read_unlock(); |
8447a0fe | 2458 | spin_unlock_irqrestore(&callback_lock, flags); |
909d75a3 PJ |
2459 | |
2460 | return mask; | |
2461 | } | |
2462 | ||
d9fd8a6d | 2463 | /** |
19770b32 MG |
2464 | * cpuset_nodemask_valid_mems_allowed - check nodemask vs. curremt mems_allowed |
2465 | * @nodemask: the nodemask to be checked | |
d9fd8a6d | 2466 | * |
19770b32 | 2467 | * Are any of the nodes in the nodemask allowed in current->mems_allowed? |
1da177e4 | 2468 | */ |
19770b32 | 2469 | int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask) |
1da177e4 | 2470 | { |
19770b32 | 2471 | return nodes_intersects(*nodemask, current->mems_allowed); |
1da177e4 LT |
2472 | } |
2473 | ||
9bf2229f | 2474 | /* |
78608366 PM |
2475 | * nearest_hardwall_ancestor() - Returns the nearest mem_exclusive or |
2476 | * mem_hardwall ancestor to the specified cpuset. Call holding | |
8447a0fe | 2477 | * callback_lock. If no ancestor is mem_exclusive or mem_hardwall |
78608366 | 2478 | * (an unusual configuration), then returns the root cpuset. |
9bf2229f | 2479 | */ |
c9710d80 | 2480 | static struct cpuset *nearest_hardwall_ancestor(struct cpuset *cs) |
9bf2229f | 2481 | { |
c431069f TH |
2482 | while (!(is_mem_exclusive(cs) || is_mem_hardwall(cs)) && parent_cs(cs)) |
2483 | cs = parent_cs(cs); | |
9bf2229f PJ |
2484 | return cs; |
2485 | } | |
2486 | ||
d9fd8a6d | 2487 | /** |
344736f2 | 2488 | * cpuset_node_allowed - Can we allocate on a memory node? |
a1bc5a4e | 2489 | * @node: is this an allowed node? |
02a0e53d | 2490 | * @gfp_mask: memory allocation flags |
d9fd8a6d | 2491 | * |
6e276d2a DR |
2492 | * If we're in interrupt, yes, we can always allocate. If @node is set in |
2493 | * current's mems_allowed, yes. If it's not a __GFP_HARDWALL request and this | |
2494 | * node is set in the nearest hardwalled cpuset ancestor to current's cpuset, | |
2495 | * yes. If current has access to memory reserves due to TIF_MEMDIE, yes. | |
9bf2229f PJ |
2496 | * Otherwise, no. |
2497 | * | |
2498 | * GFP_USER allocations are marked with the __GFP_HARDWALL bit, | |
c596d9f3 DR |
2499 | * and do not allow allocations outside the current tasks cpuset |
2500 | * unless the task has been OOM killed as is marked TIF_MEMDIE. | |
9bf2229f | 2501 | * GFP_KERNEL allocations are not so marked, so can escape to the |
78608366 | 2502 | * nearest enclosing hardwalled ancestor cpuset. |
9bf2229f | 2503 | * |
8447a0fe | 2504 | * Scanning up parent cpusets requires callback_lock. The |
02a0e53d PJ |
2505 | * __alloc_pages() routine only calls here with __GFP_HARDWALL bit |
2506 | * _not_ set if it's a GFP_KERNEL allocation, and all nodes in the | |
2507 | * current tasks mems_allowed came up empty on the first pass over | |
2508 | * the zonelist. So only GFP_KERNEL allocations, if all nodes in the | |
8447a0fe | 2509 | * cpuset are short of memory, might require taking the callback_lock. |
9bf2229f | 2510 | * |
36be57ff | 2511 | * The first call here from mm/page_alloc:get_page_from_freelist() |
02a0e53d PJ |
2512 | * has __GFP_HARDWALL set in gfp_mask, enforcing hardwall cpusets, |
2513 | * so no allocation on a node outside the cpuset is allowed (unless | |
2514 | * in interrupt, of course). | |
36be57ff PJ |
2515 | * |
2516 | * The second pass through get_page_from_freelist() doesn't even call | |
2517 | * here for GFP_ATOMIC calls. For those calls, the __alloc_pages() | |
2518 | * variable 'wait' is not set, and the bit ALLOC_CPUSET is not set | |
2519 | * in alloc_flags. That logic and the checks below have the combined | |
2520 | * affect that: | |
9bf2229f PJ |
2521 | * in_interrupt - any node ok (current task context irrelevant) |
2522 | * GFP_ATOMIC - any node ok | |
c596d9f3 | 2523 | * TIF_MEMDIE - any node ok |
78608366 | 2524 | * GFP_KERNEL - any node in enclosing hardwalled cpuset ok |
9bf2229f | 2525 | * GFP_USER - only nodes in current tasks mems allowed ok. |
02a0e53d | 2526 | */ |
002f2906 | 2527 | bool __cpuset_node_allowed(int node, gfp_t gfp_mask) |
1da177e4 | 2528 | { |
c9710d80 | 2529 | struct cpuset *cs; /* current cpuset ancestors */ |
29afd49b | 2530 | int allowed; /* is allocation in zone z allowed? */ |
8447a0fe | 2531 | unsigned long flags; |
9bf2229f | 2532 | |
6e276d2a | 2533 | if (in_interrupt()) |
002f2906 | 2534 | return true; |
9bf2229f | 2535 | if (node_isset(node, current->mems_allowed)) |
002f2906 | 2536 | return true; |
c596d9f3 DR |
2537 | /* |
2538 | * Allow tasks that have access to memory reserves because they have | |
2539 | * been OOM killed to get memory anywhere. | |
2540 | */ | |
2541 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
002f2906 | 2542 | return true; |
9bf2229f | 2543 | if (gfp_mask & __GFP_HARDWALL) /* If hardwall request, stop here */ |
002f2906 | 2544 | return false; |
9bf2229f | 2545 | |
5563e770 | 2546 | if (current->flags & PF_EXITING) /* Let dying task have memory */ |
002f2906 | 2547 | return true; |
5563e770 | 2548 | |
9bf2229f | 2549 | /* Not hardwall and node outside mems_allowed: scan up cpusets */ |
8447a0fe | 2550 | spin_lock_irqsave(&callback_lock, flags); |
053199ed | 2551 | |
b8dadcb5 | 2552 | rcu_read_lock(); |
78608366 | 2553 | cs = nearest_hardwall_ancestor(task_cs(current)); |
99afb0fd | 2554 | allowed = node_isset(node, cs->mems_allowed); |
b8dadcb5 | 2555 | rcu_read_unlock(); |
053199ed | 2556 | |
8447a0fe | 2557 | spin_unlock_irqrestore(&callback_lock, flags); |
9bf2229f | 2558 | return allowed; |
1da177e4 LT |
2559 | } |
2560 | ||
825a46af | 2561 | /** |
6adef3eb JS |
2562 | * cpuset_mem_spread_node() - On which node to begin search for a file page |
2563 | * cpuset_slab_spread_node() - On which node to begin search for a slab page | |
825a46af PJ |
2564 | * |
2565 | * If a task is marked PF_SPREAD_PAGE or PF_SPREAD_SLAB (as for | |
2566 | * tasks in a cpuset with is_spread_page or is_spread_slab set), | |
2567 | * and if the memory allocation used cpuset_mem_spread_node() | |
2568 | * to determine on which node to start looking, as it will for | |
2569 | * certain page cache or slab cache pages such as used for file | |
2570 | * system buffers and inode caches, then instead of starting on the | |
2571 | * local node to look for a free page, rather spread the starting | |
2572 | * node around the tasks mems_allowed nodes. | |
2573 | * | |
2574 | * We don't have to worry about the returned node being offline | |
2575 | * because "it can't happen", and even if it did, it would be ok. | |
2576 | * | |
2577 | * The routines calling guarantee_online_mems() are careful to | |
2578 | * only set nodes in task->mems_allowed that are online. So it | |
2579 | * should not be possible for the following code to return an | |
2580 | * offline node. But if it did, that would be ok, as this routine | |
2581 | * is not returning the node where the allocation must be, only | |
2582 | * the node where the search should start. The zonelist passed to | |
2583 | * __alloc_pages() will include all nodes. If the slab allocator | |
2584 | * is passed an offline node, it will fall back to the local node. | |
2585 | * See kmem_cache_alloc_node(). | |
2586 | */ | |
2587 | ||
6adef3eb | 2588 | static int cpuset_spread_node(int *rotor) |
825a46af | 2589 | { |
0edaf86c | 2590 | return *rotor = next_node_in(*rotor, current->mems_allowed); |
825a46af | 2591 | } |
6adef3eb JS |
2592 | |
2593 | int cpuset_mem_spread_node(void) | |
2594 | { | |
778d3b0f MH |
2595 | if (current->cpuset_mem_spread_rotor == NUMA_NO_NODE) |
2596 | current->cpuset_mem_spread_rotor = | |
2597 | node_random(¤t->mems_allowed); | |
2598 | ||
6adef3eb JS |
2599 | return cpuset_spread_node(¤t->cpuset_mem_spread_rotor); |
2600 | } | |
2601 | ||
2602 | int cpuset_slab_spread_node(void) | |
2603 | { | |
778d3b0f MH |
2604 | if (current->cpuset_slab_spread_rotor == NUMA_NO_NODE) |
2605 | current->cpuset_slab_spread_rotor = | |
2606 | node_random(¤t->mems_allowed); | |
2607 | ||
6adef3eb JS |
2608 | return cpuset_spread_node(¤t->cpuset_slab_spread_rotor); |
2609 | } | |
2610 | ||
825a46af PJ |
2611 | EXPORT_SYMBOL_GPL(cpuset_mem_spread_node); |
2612 | ||
ef08e3b4 | 2613 | /** |
bbe373f2 DR |
2614 | * cpuset_mems_allowed_intersects - Does @tsk1's mems_allowed intersect @tsk2's? |
2615 | * @tsk1: pointer to task_struct of some task. | |
2616 | * @tsk2: pointer to task_struct of some other task. | |
2617 | * | |
2618 | * Description: Return true if @tsk1's mems_allowed intersects the | |
2619 | * mems_allowed of @tsk2. Used by the OOM killer to determine if | |
2620 | * one of the task's memory usage might impact the memory available | |
2621 | * to the other. | |
ef08e3b4 PJ |
2622 | **/ |
2623 | ||
bbe373f2 DR |
2624 | int cpuset_mems_allowed_intersects(const struct task_struct *tsk1, |
2625 | const struct task_struct *tsk2) | |
ef08e3b4 | 2626 | { |
bbe373f2 | 2627 | return nodes_intersects(tsk1->mems_allowed, tsk2->mems_allowed); |
ef08e3b4 PJ |
2628 | } |
2629 | ||
75aa1994 | 2630 | /** |
da39da3a | 2631 | * cpuset_print_current_mems_allowed - prints current's cpuset and mems_allowed |
75aa1994 | 2632 | * |
da39da3a | 2633 | * Description: Prints current's name, cpuset name, and cached copy of its |
b8dadcb5 | 2634 | * mems_allowed to the kernel log. |
75aa1994 | 2635 | */ |
da39da3a | 2636 | void cpuset_print_current_mems_allowed(void) |
75aa1994 | 2637 | { |
b8dadcb5 | 2638 | struct cgroup *cgrp; |
75aa1994 | 2639 | |
b8dadcb5 | 2640 | rcu_read_lock(); |
63f43f55 | 2641 | |
da39da3a DR |
2642 | cgrp = task_cs(current)->css.cgroup; |
2643 | pr_info("%s cpuset=", current->comm); | |
e61734c5 | 2644 | pr_cont_cgroup_name(cgrp); |
da39da3a DR |
2645 | pr_cont(" mems_allowed=%*pbl\n", |
2646 | nodemask_pr_args(¤t->mems_allowed)); | |
f440d98f | 2647 | |
cfb5966b | 2648 | rcu_read_unlock(); |
75aa1994 DR |
2649 | } |
2650 | ||
3e0d98b9 PJ |
2651 | /* |
2652 | * Collection of memory_pressure is suppressed unless | |
2653 | * this flag is enabled by writing "1" to the special | |
2654 | * cpuset file 'memory_pressure_enabled' in the root cpuset. | |
2655 | */ | |
2656 | ||
c5b2aff8 | 2657 | int cpuset_memory_pressure_enabled __read_mostly; |
3e0d98b9 PJ |
2658 | |
2659 | /** | |
2660 | * cpuset_memory_pressure_bump - keep stats of per-cpuset reclaims. | |
2661 | * | |
2662 | * Keep a running average of the rate of synchronous (direct) | |
2663 | * page reclaim efforts initiated by tasks in each cpuset. | |
2664 | * | |
2665 | * This represents the rate at which some task in the cpuset | |
2666 | * ran low on memory on all nodes it was allowed to use, and | |
2667 | * had to enter the kernels page reclaim code in an effort to | |
2668 | * create more free memory by tossing clean pages or swapping | |
2669 | * or writing dirty pages. | |
2670 | * | |
2671 | * Display to user space in the per-cpuset read-only file | |
2672 | * "memory_pressure". Value displayed is an integer | |
2673 | * representing the recent rate of entry into the synchronous | |
2674 | * (direct) page reclaim by any task attached to the cpuset. | |
2675 | **/ | |
2676 | ||
2677 | void __cpuset_memory_pressure_bump(void) | |
2678 | { | |
b8dadcb5 | 2679 | rcu_read_lock(); |
8793d854 | 2680 | fmeter_markevent(&task_cs(current)->fmeter); |
b8dadcb5 | 2681 | rcu_read_unlock(); |
3e0d98b9 PJ |
2682 | } |
2683 | ||
8793d854 | 2684 | #ifdef CONFIG_PROC_PID_CPUSET |
1da177e4 LT |
2685 | /* |
2686 | * proc_cpuset_show() | |
2687 | * - Print tasks cpuset path into seq_file. | |
2688 | * - Used for /proc/<pid>/cpuset. | |
053199ed PJ |
2689 | * - No need to task_lock(tsk) on this tsk->cpuset reference, as it |
2690 | * doesn't really matter if tsk->cpuset changes after we read it, | |
5d21cc2d | 2691 | * and we take cpuset_mutex, keeping cpuset_attach() from changing it |
2df167a3 | 2692 | * anyway. |
1da177e4 | 2693 | */ |
52de4779 ZL |
2694 | int proc_cpuset_show(struct seq_file *m, struct pid_namespace *ns, |
2695 | struct pid *pid, struct task_struct *tsk) | |
1da177e4 | 2696 | { |
4c737b41 | 2697 | char *buf; |
8793d854 | 2698 | struct cgroup_subsys_state *css; |
99f89551 | 2699 | int retval; |
1da177e4 | 2700 | |
99f89551 | 2701 | retval = -ENOMEM; |
e61734c5 | 2702 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
1da177e4 | 2703 | if (!buf) |
99f89551 EB |
2704 | goto out; |
2705 | ||
a79a908f | 2706 | css = task_get_css(tsk, cpuset_cgrp_id); |
4c737b41 TH |
2707 | retval = cgroup_path_ns(css->cgroup, buf, PATH_MAX, |
2708 | current->nsproxy->cgroup_ns); | |
a79a908f | 2709 | css_put(css); |
4c737b41 | 2710 | if (retval >= PATH_MAX) |
679a5e3f TH |
2711 | retval = -ENAMETOOLONG; |
2712 | if (retval < 0) | |
52de4779 | 2713 | goto out_free; |
4c737b41 | 2714 | seq_puts(m, buf); |
1da177e4 | 2715 | seq_putc(m, '\n'); |
e61734c5 | 2716 | retval = 0; |
99f89551 | 2717 | out_free: |
1da177e4 | 2718 | kfree(buf); |
99f89551 | 2719 | out: |
1da177e4 LT |
2720 | return retval; |
2721 | } | |
8793d854 | 2722 | #endif /* CONFIG_PROC_PID_CPUSET */ |
1da177e4 | 2723 | |
d01d4827 | 2724 | /* Display task mems_allowed in /proc/<pid>/status file. */ |
df5f8314 EB |
2725 | void cpuset_task_status_allowed(struct seq_file *m, struct task_struct *task) |
2726 | { | |
e8e6d97c TH |
2727 | seq_printf(m, "Mems_allowed:\t%*pb\n", |
2728 | nodemask_pr_args(&task->mems_allowed)); | |
2729 | seq_printf(m, "Mems_allowed_list:\t%*pbl\n", | |
2730 | nodemask_pr_args(&task->mems_allowed)); | |
1da177e4 | 2731 | } |