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 LT |
46 | #include <linux/sched.h> |
47 | #include <linux/seq_file.h> | |
22fb52dd | 48 | #include <linux/security.h> |
1da177e4 | 49 | #include <linux/slab.h> |
1da177e4 LT |
50 | #include <linux/spinlock.h> |
51 | #include <linux/stat.h> | |
52 | #include <linux/string.h> | |
53 | #include <linux/time.h> | |
54 | #include <linux/backing-dev.h> | |
55 | #include <linux/sort.h> | |
56 | ||
57 | #include <asm/uaccess.h> | |
60063497 | 58 | #include <linux/atomic.h> |
3d3f26a7 | 59 | #include <linux/mutex.h> |
956db3ca CW |
60 | #include <linux/workqueue.h> |
61 | #include <linux/cgroup.h> | |
e44193d3 | 62 | #include <linux/wait.h> |
1da177e4 | 63 | |
664eedde | 64 | struct static_key cpusets_enabled_key __read_mostly = STATIC_KEY_INIT_FALSE; |
202f72d5 | 65 | |
3e0d98b9 PJ |
66 | /* See "Frequency meter" comments, below. */ |
67 | ||
68 | struct fmeter { | |
69 | int cnt; /* unprocessed events count */ | |
70 | int val; /* most recent output value */ | |
71 | time_t time; /* clock (secs) when val computed */ | |
72 | spinlock_t lock; /* guards read or write of above */ | |
73 | }; | |
74 | ||
1da177e4 | 75 | struct cpuset { |
8793d854 PM |
76 | struct cgroup_subsys_state css; |
77 | ||
1da177e4 | 78 | unsigned long flags; /* "unsigned long" so bitops work */ |
e2b9a3d7 LZ |
79 | |
80 | /* user-configured CPUs and Memory Nodes allow to tasks */ | |
81 | cpumask_var_t cpus_allowed; | |
82 | nodemask_t mems_allowed; | |
83 | ||
84 | /* effective CPUs and Memory Nodes allow to tasks */ | |
85 | cpumask_var_t effective_cpus; | |
86 | nodemask_t effective_mems; | |
1da177e4 | 87 | |
33ad801d LZ |
88 | /* |
89 | * This is old Memory Nodes tasks took on. | |
90 | * | |
91 | * - top_cpuset.old_mems_allowed is initialized to mems_allowed. | |
92 | * - A new cpuset's old_mems_allowed is initialized when some | |
93 | * task is moved into it. | |
94 | * - old_mems_allowed is used in cpuset_migrate_mm() when we change | |
95 | * cpuset.mems_allowed and have tasks' nodemask updated, and | |
96 | * then old_mems_allowed is updated to mems_allowed. | |
97 | */ | |
98 | nodemask_t old_mems_allowed; | |
99 | ||
3e0d98b9 | 100 | struct fmeter fmeter; /* memory_pressure filter */ |
029190c5 | 101 | |
452477fa TH |
102 | /* |
103 | * Tasks are being attached to this cpuset. Used to prevent | |
104 | * zeroing cpus/mems_allowed between ->can_attach() and ->attach(). | |
105 | */ | |
106 | int attach_in_progress; | |
107 | ||
029190c5 PJ |
108 | /* partition number for rebuild_sched_domains() */ |
109 | int pn; | |
956db3ca | 110 | |
1d3504fc HS |
111 | /* for custom sched domain */ |
112 | int relax_domain_level; | |
1da177e4 LT |
113 | }; |
114 | ||
a7c6d554 | 115 | static inline struct cpuset *css_cs(struct cgroup_subsys_state *css) |
8793d854 | 116 | { |
a7c6d554 | 117 | return css ? container_of(css, struct cpuset, css) : NULL; |
8793d854 PM |
118 | } |
119 | ||
120 | /* Retrieve the cpuset for a task */ | |
121 | static inline struct cpuset *task_cs(struct task_struct *task) | |
122 | { | |
073219e9 | 123 | return css_cs(task_css(task, cpuset_cgrp_id)); |
8793d854 | 124 | } |
8793d854 | 125 | |
c9710d80 | 126 | static inline struct cpuset *parent_cs(struct cpuset *cs) |
c431069f | 127 | { |
5c9d535b | 128 | return css_cs(cs->css.parent); |
c431069f TH |
129 | } |
130 | ||
b246272e DR |
131 | #ifdef CONFIG_NUMA |
132 | static inline bool task_has_mempolicy(struct task_struct *task) | |
133 | { | |
134 | return task->mempolicy; | |
135 | } | |
136 | #else | |
137 | static inline bool task_has_mempolicy(struct task_struct *task) | |
138 | { | |
139 | return false; | |
140 | } | |
141 | #endif | |
142 | ||
143 | ||
1da177e4 LT |
144 | /* bits in struct cpuset flags field */ |
145 | typedef enum { | |
efeb77b2 | 146 | CS_ONLINE, |
1da177e4 LT |
147 | CS_CPU_EXCLUSIVE, |
148 | CS_MEM_EXCLUSIVE, | |
78608366 | 149 | CS_MEM_HARDWALL, |
45b07ef3 | 150 | CS_MEMORY_MIGRATE, |
029190c5 | 151 | CS_SCHED_LOAD_BALANCE, |
825a46af PJ |
152 | CS_SPREAD_PAGE, |
153 | CS_SPREAD_SLAB, | |
1da177e4 LT |
154 | } cpuset_flagbits_t; |
155 | ||
156 | /* convenient tests for these bits */ | |
efeb77b2 TH |
157 | static inline bool is_cpuset_online(const struct cpuset *cs) |
158 | { | |
159 | return test_bit(CS_ONLINE, &cs->flags); | |
160 | } | |
161 | ||
1da177e4 LT |
162 | static inline int is_cpu_exclusive(const struct cpuset *cs) |
163 | { | |
7b5b9ef0 | 164 | return test_bit(CS_CPU_EXCLUSIVE, &cs->flags); |
1da177e4 LT |
165 | } |
166 | ||
167 | static inline int is_mem_exclusive(const struct cpuset *cs) | |
168 | { | |
7b5b9ef0 | 169 | return test_bit(CS_MEM_EXCLUSIVE, &cs->flags); |
1da177e4 LT |
170 | } |
171 | ||
78608366 PM |
172 | static inline int is_mem_hardwall(const struct cpuset *cs) |
173 | { | |
174 | return test_bit(CS_MEM_HARDWALL, &cs->flags); | |
175 | } | |
176 | ||
029190c5 PJ |
177 | static inline int is_sched_load_balance(const struct cpuset *cs) |
178 | { | |
179 | return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); | |
180 | } | |
181 | ||
45b07ef3 PJ |
182 | static inline int is_memory_migrate(const struct cpuset *cs) |
183 | { | |
7b5b9ef0 | 184 | return test_bit(CS_MEMORY_MIGRATE, &cs->flags); |
45b07ef3 PJ |
185 | } |
186 | ||
825a46af PJ |
187 | static inline int is_spread_page(const struct cpuset *cs) |
188 | { | |
189 | return test_bit(CS_SPREAD_PAGE, &cs->flags); | |
190 | } | |
191 | ||
192 | static inline int is_spread_slab(const struct cpuset *cs) | |
193 | { | |
194 | return test_bit(CS_SPREAD_SLAB, &cs->flags); | |
195 | } | |
196 | ||
1da177e4 | 197 | static struct cpuset top_cpuset = { |
efeb77b2 TH |
198 | .flags = ((1 << CS_ONLINE) | (1 << CS_CPU_EXCLUSIVE) | |
199 | (1 << CS_MEM_EXCLUSIVE)), | |
1da177e4 LT |
200 | }; |
201 | ||
ae8086ce TH |
202 | /** |
203 | * cpuset_for_each_child - traverse online children of a cpuset | |
204 | * @child_cs: loop cursor pointing to the current child | |
492eb21b | 205 | * @pos_css: used for iteration |
ae8086ce TH |
206 | * @parent_cs: target cpuset to walk children of |
207 | * | |
208 | * Walk @child_cs through the online children of @parent_cs. Must be used | |
209 | * with RCU read locked. | |
210 | */ | |
492eb21b TH |
211 | #define cpuset_for_each_child(child_cs, pos_css, parent_cs) \ |
212 | css_for_each_child((pos_css), &(parent_cs)->css) \ | |
213 | if (is_cpuset_online(((child_cs) = css_cs((pos_css))))) | |
ae8086ce | 214 | |
fc560a26 TH |
215 | /** |
216 | * cpuset_for_each_descendant_pre - pre-order walk of a cpuset's descendants | |
217 | * @des_cs: loop cursor pointing to the current descendant | |
492eb21b | 218 | * @pos_css: used for iteration |
fc560a26 TH |
219 | * @root_cs: target cpuset to walk ancestor of |
220 | * | |
221 | * Walk @des_cs through the online descendants of @root_cs. Must be used | |
492eb21b | 222 | * with RCU read locked. The caller may modify @pos_css by calling |
bd8815a6 TH |
223 | * css_rightmost_descendant() to skip subtree. @root_cs is included in the |
224 | * iteration and the first node to be visited. | |
fc560a26 | 225 | */ |
492eb21b TH |
226 | #define cpuset_for_each_descendant_pre(des_cs, pos_css, root_cs) \ |
227 | css_for_each_descendant_pre((pos_css), &(root_cs)->css) \ | |
228 | if (is_cpuset_online(((des_cs) = css_cs((pos_css))))) | |
fc560a26 | 229 | |
1da177e4 | 230 | /* |
5d21cc2d TH |
231 | * There are two global mutexes guarding cpuset structures - cpuset_mutex |
232 | * and callback_mutex. The latter may nest inside the former. We also | |
233 | * require taking task_lock() when dereferencing a task's cpuset pointer. | |
234 | * See "The task_lock() exception", at the end of this comment. | |
235 | * | |
236 | * A task must hold both mutexes to modify cpusets. If a task holds | |
237 | * cpuset_mutex, then it blocks others wanting that mutex, ensuring that it | |
238 | * is the only task able to also acquire callback_mutex and be able to | |
239 | * modify cpusets. It can perform various checks on the cpuset structure | |
240 | * first, knowing nothing will change. It can also allocate memory while | |
241 | * just holding cpuset_mutex. While it is performing these checks, various | |
242 | * callback routines can briefly acquire callback_mutex to query cpusets. | |
243 | * Once it is ready to make the changes, it takes callback_mutex, blocking | |
244 | * everyone else. | |
053199ed PJ |
245 | * |
246 | * Calls to the kernel memory allocator can not be made while holding | |
3d3f26a7 | 247 | * callback_mutex, as that would risk double tripping on callback_mutex |
053199ed PJ |
248 | * from one of the callbacks into the cpuset code from within |
249 | * __alloc_pages(). | |
250 | * | |
3d3f26a7 | 251 | * If a task is only holding callback_mutex, then it has read-only |
053199ed PJ |
252 | * access to cpusets. |
253 | * | |
58568d2a MX |
254 | * Now, the task_struct fields mems_allowed and mempolicy may be changed |
255 | * by other task, we use alloc_lock in the task_struct fields to protect | |
256 | * them. | |
053199ed | 257 | * |
3d3f26a7 | 258 | * The cpuset_common_file_read() handlers only hold callback_mutex across |
053199ed PJ |
259 | * small pieces of code, such as when reading out possibly multi-word |
260 | * cpumasks and nodemasks. | |
261 | * | |
2df167a3 PM |
262 | * Accessing a task's cpuset should be done in accordance with the |
263 | * guidelines for accessing subsystem state in kernel/cgroup.c | |
1da177e4 LT |
264 | */ |
265 | ||
5d21cc2d | 266 | static DEFINE_MUTEX(cpuset_mutex); |
3d3f26a7 | 267 | static DEFINE_MUTEX(callback_mutex); |
4247bdc6 | 268 | |
3a5a6d0c TH |
269 | /* |
270 | * CPU / memory hotplug is handled asynchronously. | |
271 | */ | |
272 | static void cpuset_hotplug_workfn(struct work_struct *work); | |
3a5a6d0c TH |
273 | static DECLARE_WORK(cpuset_hotplug_work, cpuset_hotplug_workfn); |
274 | ||
e44193d3 LZ |
275 | static DECLARE_WAIT_QUEUE_HEAD(cpuset_attach_wq); |
276 | ||
cf417141 MK |
277 | /* |
278 | * This is ugly, but preserves the userspace API for existing cpuset | |
8793d854 | 279 | * users. If someone tries to mount the "cpuset" filesystem, we |
cf417141 MK |
280 | * silently switch it to mount "cgroup" instead |
281 | */ | |
f7e83571 AV |
282 | static struct dentry *cpuset_mount(struct file_system_type *fs_type, |
283 | int flags, const char *unused_dev_name, void *data) | |
1da177e4 | 284 | { |
8793d854 | 285 | struct file_system_type *cgroup_fs = get_fs_type("cgroup"); |
f7e83571 | 286 | struct dentry *ret = ERR_PTR(-ENODEV); |
8793d854 PM |
287 | if (cgroup_fs) { |
288 | char mountopts[] = | |
289 | "cpuset,noprefix," | |
290 | "release_agent=/sbin/cpuset_release_agent"; | |
f7e83571 AV |
291 | ret = cgroup_fs->mount(cgroup_fs, flags, |
292 | unused_dev_name, mountopts); | |
8793d854 PM |
293 | put_filesystem(cgroup_fs); |
294 | } | |
295 | return ret; | |
1da177e4 LT |
296 | } |
297 | ||
298 | static struct file_system_type cpuset_fs_type = { | |
299 | .name = "cpuset", | |
f7e83571 | 300 | .mount = cpuset_mount, |
1da177e4 LT |
301 | }; |
302 | ||
1da177e4 | 303 | /* |
300ed6cb | 304 | * Return in pmask the portion of a cpusets's cpus_allowed that |
1da177e4 | 305 | * are online. If none are online, walk up the cpuset hierarchy |
40df2deb LZ |
306 | * until we find one that does have some online cpus. The top |
307 | * cpuset always has some cpus online. | |
1da177e4 LT |
308 | * |
309 | * One way or another, we guarantee to return some non-empty subset | |
5f054e31 | 310 | * of cpu_online_mask. |
1da177e4 | 311 | * |
3d3f26a7 | 312 | * Call with callback_mutex held. |
1da177e4 | 313 | */ |
c9710d80 | 314 | static void guarantee_online_cpus(struct cpuset *cs, struct cpumask *pmask) |
1da177e4 | 315 | { |
40df2deb | 316 | while (!cpumask_intersects(cs->cpus_allowed, cpu_online_mask)) |
c431069f | 317 | cs = parent_cs(cs); |
40df2deb | 318 | cpumask_and(pmask, cs->cpus_allowed, cpu_online_mask); |
1da177e4 LT |
319 | } |
320 | ||
321 | /* | |
322 | * Return in *pmask the portion of a cpusets's mems_allowed that | |
0e1e7c7a CL |
323 | * are online, with memory. If none are online with memory, walk |
324 | * up the cpuset hierarchy until we find one that does have some | |
40df2deb | 325 | * online mems. The top cpuset always has some mems online. |
1da177e4 LT |
326 | * |
327 | * One way or another, we guarantee to return some non-empty subset | |
38d7bee9 | 328 | * of node_states[N_MEMORY]. |
1da177e4 | 329 | * |
3d3f26a7 | 330 | * Call with callback_mutex held. |
1da177e4 | 331 | */ |
c9710d80 | 332 | static void guarantee_online_mems(struct cpuset *cs, nodemask_t *pmask) |
1da177e4 | 333 | { |
40df2deb | 334 | while (!nodes_intersects(cs->mems_allowed, node_states[N_MEMORY])) |
c431069f | 335 | cs = parent_cs(cs); |
40df2deb | 336 | nodes_and(*pmask, cs->mems_allowed, node_states[N_MEMORY]); |
1da177e4 LT |
337 | } |
338 | ||
f3b39d47 MX |
339 | /* |
340 | * update task's spread flag if cpuset's page/slab spread flag is set | |
341 | * | |
5d21cc2d | 342 | * Called with callback_mutex/cpuset_mutex held |
f3b39d47 MX |
343 | */ |
344 | static void cpuset_update_task_spread_flag(struct cpuset *cs, | |
345 | struct task_struct *tsk) | |
346 | { | |
347 | if (is_spread_page(cs)) | |
348 | tsk->flags |= PF_SPREAD_PAGE; | |
349 | else | |
350 | tsk->flags &= ~PF_SPREAD_PAGE; | |
351 | if (is_spread_slab(cs)) | |
352 | tsk->flags |= PF_SPREAD_SLAB; | |
353 | else | |
354 | tsk->flags &= ~PF_SPREAD_SLAB; | |
355 | } | |
356 | ||
1da177e4 LT |
357 | /* |
358 | * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q? | |
359 | * | |
360 | * One cpuset is a subset of another if all its allowed CPUs and | |
361 | * Memory Nodes are a subset of the other, and its exclusive flags | |
5d21cc2d | 362 | * are only set if the other's are set. Call holding cpuset_mutex. |
1da177e4 LT |
363 | */ |
364 | ||
365 | static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q) | |
366 | { | |
300ed6cb | 367 | return cpumask_subset(p->cpus_allowed, q->cpus_allowed) && |
1da177e4 LT |
368 | nodes_subset(p->mems_allowed, q->mems_allowed) && |
369 | is_cpu_exclusive(p) <= is_cpu_exclusive(q) && | |
370 | is_mem_exclusive(p) <= is_mem_exclusive(q); | |
371 | } | |
372 | ||
645fcc9d LZ |
373 | /** |
374 | * alloc_trial_cpuset - allocate a trial cpuset | |
375 | * @cs: the cpuset that the trial cpuset duplicates | |
376 | */ | |
c9710d80 | 377 | static struct cpuset *alloc_trial_cpuset(struct cpuset *cs) |
645fcc9d | 378 | { |
300ed6cb LZ |
379 | struct cpuset *trial; |
380 | ||
381 | trial = kmemdup(cs, sizeof(*cs), GFP_KERNEL); | |
382 | if (!trial) | |
383 | return NULL; | |
384 | ||
e2b9a3d7 LZ |
385 | if (!alloc_cpumask_var(&trial->cpus_allowed, GFP_KERNEL)) |
386 | goto free_cs; | |
387 | if (!alloc_cpumask_var(&trial->effective_cpus, GFP_KERNEL)) | |
388 | goto free_cpus; | |
300ed6cb | 389 | |
e2b9a3d7 LZ |
390 | cpumask_copy(trial->cpus_allowed, cs->cpus_allowed); |
391 | cpumask_copy(trial->effective_cpus, cs->effective_cpus); | |
300ed6cb | 392 | return trial; |
e2b9a3d7 LZ |
393 | |
394 | free_cpus: | |
395 | free_cpumask_var(trial->cpus_allowed); | |
396 | free_cs: | |
397 | kfree(trial); | |
398 | return NULL; | |
645fcc9d LZ |
399 | } |
400 | ||
401 | /** | |
402 | * free_trial_cpuset - free the trial cpuset | |
403 | * @trial: the trial cpuset to be freed | |
404 | */ | |
405 | static void free_trial_cpuset(struct cpuset *trial) | |
406 | { | |
e2b9a3d7 | 407 | free_cpumask_var(trial->effective_cpus); |
300ed6cb | 408 | free_cpumask_var(trial->cpus_allowed); |
645fcc9d LZ |
409 | kfree(trial); |
410 | } | |
411 | ||
1da177e4 LT |
412 | /* |
413 | * validate_change() - Used to validate that any proposed cpuset change | |
414 | * follows the structural rules for cpusets. | |
415 | * | |
416 | * If we replaced the flag and mask values of the current cpuset | |
417 | * (cur) with those values in the trial cpuset (trial), would | |
418 | * our various subset and exclusive rules still be valid? Presumes | |
5d21cc2d | 419 | * cpuset_mutex held. |
1da177e4 LT |
420 | * |
421 | * 'cur' is the address of an actual, in-use cpuset. Operations | |
422 | * such as list traversal that depend on the actual address of the | |
423 | * cpuset in the list must use cur below, not trial. | |
424 | * | |
425 | * 'trial' is the address of bulk structure copy of cur, with | |
426 | * perhaps one or more of the fields cpus_allowed, mems_allowed, | |
427 | * or flags changed to new, trial values. | |
428 | * | |
429 | * Return 0 if valid, -errno if not. | |
430 | */ | |
431 | ||
c9710d80 | 432 | static int validate_change(struct cpuset *cur, struct cpuset *trial) |
1da177e4 | 433 | { |
492eb21b | 434 | struct cgroup_subsys_state *css; |
1da177e4 | 435 | struct cpuset *c, *par; |
ae8086ce TH |
436 | int ret; |
437 | ||
438 | rcu_read_lock(); | |
1da177e4 LT |
439 | |
440 | /* Each of our child cpusets must be a subset of us */ | |
ae8086ce | 441 | ret = -EBUSY; |
492eb21b | 442 | cpuset_for_each_child(c, css, cur) |
ae8086ce TH |
443 | if (!is_cpuset_subset(c, trial)) |
444 | goto out; | |
1da177e4 LT |
445 | |
446 | /* Remaining checks don't apply to root cpuset */ | |
ae8086ce | 447 | ret = 0; |
69604067 | 448 | if (cur == &top_cpuset) |
ae8086ce | 449 | goto out; |
1da177e4 | 450 | |
c431069f | 451 | par = parent_cs(cur); |
69604067 | 452 | |
1da177e4 | 453 | /* We must be a subset of our parent cpuset */ |
ae8086ce | 454 | ret = -EACCES; |
1da177e4 | 455 | if (!is_cpuset_subset(trial, par)) |
ae8086ce | 456 | goto out; |
1da177e4 | 457 | |
2df167a3 PM |
458 | /* |
459 | * If either I or some sibling (!= me) is exclusive, we can't | |
460 | * overlap | |
461 | */ | |
ae8086ce | 462 | ret = -EINVAL; |
492eb21b | 463 | cpuset_for_each_child(c, css, par) { |
1da177e4 LT |
464 | if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) && |
465 | c != cur && | |
300ed6cb | 466 | cpumask_intersects(trial->cpus_allowed, c->cpus_allowed)) |
ae8086ce | 467 | goto out; |
1da177e4 LT |
468 | if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) && |
469 | c != cur && | |
470 | nodes_intersects(trial->mems_allowed, c->mems_allowed)) | |
ae8086ce | 471 | goto out; |
1da177e4 LT |
472 | } |
473 | ||
452477fa TH |
474 | /* |
475 | * Cpusets with tasks - existing or newly being attached - can't | |
1c09b195 | 476 | * be changed to have empty cpus_allowed or mems_allowed. |
452477fa | 477 | */ |
ae8086ce | 478 | ret = -ENOSPC; |
07bc356e | 479 | if ((cgroup_has_tasks(cur->css.cgroup) || cur->attach_in_progress)) { |
1c09b195 LZ |
480 | if (!cpumask_empty(cur->cpus_allowed) && |
481 | cpumask_empty(trial->cpus_allowed)) | |
482 | goto out; | |
483 | if (!nodes_empty(cur->mems_allowed) && | |
484 | nodes_empty(trial->mems_allowed)) | |
485 | goto out; | |
486 | } | |
020958b6 | 487 | |
ae8086ce TH |
488 | ret = 0; |
489 | out: | |
490 | rcu_read_unlock(); | |
491 | return ret; | |
1da177e4 LT |
492 | } |
493 | ||
db7f47cf | 494 | #ifdef CONFIG_SMP |
029190c5 | 495 | /* |
cf417141 | 496 | * Helper routine for generate_sched_domains(). |
029190c5 PJ |
497 | * Do cpusets a, b have overlapping cpus_allowed masks? |
498 | */ | |
029190c5 PJ |
499 | static int cpusets_overlap(struct cpuset *a, struct cpuset *b) |
500 | { | |
300ed6cb | 501 | return cpumask_intersects(a->cpus_allowed, b->cpus_allowed); |
029190c5 PJ |
502 | } |
503 | ||
1d3504fc HS |
504 | static void |
505 | update_domain_attr(struct sched_domain_attr *dattr, struct cpuset *c) | |
506 | { | |
1d3504fc HS |
507 | if (dattr->relax_domain_level < c->relax_domain_level) |
508 | dattr->relax_domain_level = c->relax_domain_level; | |
509 | return; | |
510 | } | |
511 | ||
fc560a26 TH |
512 | static void update_domain_attr_tree(struct sched_domain_attr *dattr, |
513 | struct cpuset *root_cs) | |
f5393693 | 514 | { |
fc560a26 | 515 | struct cpuset *cp; |
492eb21b | 516 | struct cgroup_subsys_state *pos_css; |
f5393693 | 517 | |
fc560a26 | 518 | rcu_read_lock(); |
492eb21b | 519 | cpuset_for_each_descendant_pre(cp, pos_css, root_cs) { |
bd8815a6 TH |
520 | if (cp == root_cs) |
521 | continue; | |
522 | ||
fc560a26 TH |
523 | /* skip the whole subtree if @cp doesn't have any CPU */ |
524 | if (cpumask_empty(cp->cpus_allowed)) { | |
492eb21b | 525 | pos_css = css_rightmost_descendant(pos_css); |
f5393693 | 526 | continue; |
fc560a26 | 527 | } |
f5393693 LJ |
528 | |
529 | if (is_sched_load_balance(cp)) | |
530 | update_domain_attr(dattr, cp); | |
f5393693 | 531 | } |
fc560a26 | 532 | rcu_read_unlock(); |
f5393693 LJ |
533 | } |
534 | ||
029190c5 | 535 | /* |
cf417141 MK |
536 | * generate_sched_domains() |
537 | * | |
538 | * This function builds a partial partition of the systems CPUs | |
539 | * A 'partial partition' is a set of non-overlapping subsets whose | |
540 | * union is a subset of that set. | |
0a0fca9d | 541 | * The output of this function needs to be passed to kernel/sched/core.c |
cf417141 MK |
542 | * partition_sched_domains() routine, which will rebuild the scheduler's |
543 | * load balancing domains (sched domains) as specified by that partial | |
544 | * partition. | |
029190c5 | 545 | * |
45ce80fb | 546 | * See "What is sched_load_balance" in Documentation/cgroups/cpusets.txt |
029190c5 PJ |
547 | * for a background explanation of this. |
548 | * | |
549 | * Does not return errors, on the theory that the callers of this | |
550 | * routine would rather not worry about failures to rebuild sched | |
551 | * domains when operating in the severe memory shortage situations | |
552 | * that could cause allocation failures below. | |
553 | * | |
5d21cc2d | 554 | * Must be called with cpuset_mutex held. |
029190c5 PJ |
555 | * |
556 | * The three key local variables below are: | |
aeed6824 | 557 | * q - a linked-list queue of cpuset pointers, used to implement a |
029190c5 PJ |
558 | * top-down scan of all cpusets. This scan loads a pointer |
559 | * to each cpuset marked is_sched_load_balance into the | |
560 | * array 'csa'. For our purposes, rebuilding the schedulers | |
561 | * sched domains, we can ignore !is_sched_load_balance cpusets. | |
562 | * csa - (for CpuSet Array) Array of pointers to all the cpusets | |
563 | * that need to be load balanced, for convenient iterative | |
564 | * access by the subsequent code that finds the best partition, | |
565 | * i.e the set of domains (subsets) of CPUs such that the | |
566 | * cpus_allowed of every cpuset marked is_sched_load_balance | |
567 | * is a subset of one of these domains, while there are as | |
568 | * many such domains as possible, each as small as possible. | |
569 | * doms - Conversion of 'csa' to an array of cpumasks, for passing to | |
0a0fca9d | 570 | * the kernel/sched/core.c routine partition_sched_domains() in a |
029190c5 PJ |
571 | * convenient format, that can be easily compared to the prior |
572 | * value to determine what partition elements (sched domains) | |
573 | * were changed (added or removed.) | |
574 | * | |
575 | * Finding the best partition (set of domains): | |
576 | * The triple nested loops below over i, j, k scan over the | |
577 | * load balanced cpusets (using the array of cpuset pointers in | |
578 | * csa[]) looking for pairs of cpusets that have overlapping | |
579 | * cpus_allowed, but which don't have the same 'pn' partition | |
580 | * number and gives them in the same partition number. It keeps | |
581 | * looping on the 'restart' label until it can no longer find | |
582 | * any such pairs. | |
583 | * | |
584 | * The union of the cpus_allowed masks from the set of | |
585 | * all cpusets having the same 'pn' value then form the one | |
586 | * element of the partition (one sched domain) to be passed to | |
587 | * partition_sched_domains(). | |
588 | */ | |
acc3f5d7 | 589 | static int generate_sched_domains(cpumask_var_t **domains, |
cf417141 | 590 | struct sched_domain_attr **attributes) |
029190c5 | 591 | { |
029190c5 PJ |
592 | struct cpuset *cp; /* scans q */ |
593 | struct cpuset **csa; /* array of all cpuset ptrs */ | |
594 | int csn; /* how many cpuset ptrs in csa so far */ | |
595 | int i, j, k; /* indices for partition finding loops */ | |
acc3f5d7 | 596 | cpumask_var_t *doms; /* resulting partition; i.e. sched domains */ |
1d3504fc | 597 | struct sched_domain_attr *dattr; /* attributes for custom domains */ |
1583715d | 598 | int ndoms = 0; /* number of sched domains in result */ |
6af866af | 599 | int nslot; /* next empty doms[] struct cpumask slot */ |
492eb21b | 600 | struct cgroup_subsys_state *pos_css; |
029190c5 | 601 | |
029190c5 | 602 | doms = NULL; |
1d3504fc | 603 | dattr = NULL; |
cf417141 | 604 | csa = NULL; |
029190c5 PJ |
605 | |
606 | /* Special case for the 99% of systems with one, full, sched domain */ | |
607 | if (is_sched_load_balance(&top_cpuset)) { | |
acc3f5d7 RR |
608 | ndoms = 1; |
609 | doms = alloc_sched_domains(ndoms); | |
029190c5 | 610 | if (!doms) |
cf417141 MK |
611 | goto done; |
612 | ||
1d3504fc HS |
613 | dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL); |
614 | if (dattr) { | |
615 | *dattr = SD_ATTR_INIT; | |
93a65575 | 616 | update_domain_attr_tree(dattr, &top_cpuset); |
1d3504fc | 617 | } |
acc3f5d7 | 618 | cpumask_copy(doms[0], top_cpuset.cpus_allowed); |
cf417141 | 619 | |
cf417141 | 620 | goto done; |
029190c5 PJ |
621 | } |
622 | ||
664eedde | 623 | csa = kmalloc(nr_cpusets() * sizeof(cp), GFP_KERNEL); |
029190c5 PJ |
624 | if (!csa) |
625 | goto done; | |
626 | csn = 0; | |
627 | ||
fc560a26 | 628 | rcu_read_lock(); |
492eb21b | 629 | cpuset_for_each_descendant_pre(cp, pos_css, &top_cpuset) { |
bd8815a6 TH |
630 | if (cp == &top_cpuset) |
631 | continue; | |
f5393693 | 632 | /* |
fc560a26 TH |
633 | * Continue traversing beyond @cp iff @cp has some CPUs and |
634 | * isn't load balancing. The former is obvious. The | |
635 | * latter: All child cpusets contain a subset of the | |
636 | * parent's cpus, so just skip them, and then we call | |
637 | * update_domain_attr_tree() to calc relax_domain_level of | |
638 | * the corresponding sched domain. | |
f5393693 | 639 | */ |
fc560a26 TH |
640 | if (!cpumask_empty(cp->cpus_allowed) && |
641 | !is_sched_load_balance(cp)) | |
f5393693 | 642 | continue; |
489a5393 | 643 | |
fc560a26 TH |
644 | if (is_sched_load_balance(cp)) |
645 | csa[csn++] = cp; | |
646 | ||
647 | /* skip @cp's subtree */ | |
492eb21b | 648 | pos_css = css_rightmost_descendant(pos_css); |
fc560a26 TH |
649 | } |
650 | rcu_read_unlock(); | |
029190c5 PJ |
651 | |
652 | for (i = 0; i < csn; i++) | |
653 | csa[i]->pn = i; | |
654 | ndoms = csn; | |
655 | ||
656 | restart: | |
657 | /* Find the best partition (set of sched domains) */ | |
658 | for (i = 0; i < csn; i++) { | |
659 | struct cpuset *a = csa[i]; | |
660 | int apn = a->pn; | |
661 | ||
662 | for (j = 0; j < csn; j++) { | |
663 | struct cpuset *b = csa[j]; | |
664 | int bpn = b->pn; | |
665 | ||
666 | if (apn != bpn && cpusets_overlap(a, b)) { | |
667 | for (k = 0; k < csn; k++) { | |
668 | struct cpuset *c = csa[k]; | |
669 | ||
670 | if (c->pn == bpn) | |
671 | c->pn = apn; | |
672 | } | |
673 | ndoms--; /* one less element */ | |
674 | goto restart; | |
675 | } | |
676 | } | |
677 | } | |
678 | ||
cf417141 MK |
679 | /* |
680 | * Now we know how many domains to create. | |
681 | * Convert <csn, csa> to <ndoms, doms> and populate cpu masks. | |
682 | */ | |
acc3f5d7 | 683 | doms = alloc_sched_domains(ndoms); |
700018e0 | 684 | if (!doms) |
cf417141 | 685 | goto done; |
cf417141 MK |
686 | |
687 | /* | |
688 | * The rest of the code, including the scheduler, can deal with | |
689 | * dattr==NULL case. No need to abort if alloc fails. | |
690 | */ | |
1d3504fc | 691 | dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL); |
029190c5 PJ |
692 | |
693 | for (nslot = 0, i = 0; i < csn; i++) { | |
694 | struct cpuset *a = csa[i]; | |
6af866af | 695 | struct cpumask *dp; |
029190c5 PJ |
696 | int apn = a->pn; |
697 | ||
cf417141 MK |
698 | if (apn < 0) { |
699 | /* Skip completed partitions */ | |
700 | continue; | |
701 | } | |
702 | ||
acc3f5d7 | 703 | dp = doms[nslot]; |
cf417141 MK |
704 | |
705 | if (nslot == ndoms) { | |
706 | static int warnings = 10; | |
707 | if (warnings) { | |
12d3089c FF |
708 | pr_warn("rebuild_sched_domains confused: nslot %d, ndoms %d, csn %d, i %d, apn %d\n", |
709 | nslot, ndoms, csn, i, apn); | |
cf417141 | 710 | warnings--; |
029190c5 | 711 | } |
cf417141 MK |
712 | continue; |
713 | } | |
029190c5 | 714 | |
6af866af | 715 | cpumask_clear(dp); |
cf417141 MK |
716 | if (dattr) |
717 | *(dattr + nslot) = SD_ATTR_INIT; | |
718 | for (j = i; j < csn; j++) { | |
719 | struct cpuset *b = csa[j]; | |
720 | ||
721 | if (apn == b->pn) { | |
300ed6cb | 722 | cpumask_or(dp, dp, b->cpus_allowed); |
cf417141 MK |
723 | if (dattr) |
724 | update_domain_attr_tree(dattr + nslot, b); | |
725 | ||
726 | /* Done with this partition */ | |
727 | b->pn = -1; | |
029190c5 | 728 | } |
029190c5 | 729 | } |
cf417141 | 730 | nslot++; |
029190c5 PJ |
731 | } |
732 | BUG_ON(nslot != ndoms); | |
733 | ||
cf417141 MK |
734 | done: |
735 | kfree(csa); | |
736 | ||
700018e0 LZ |
737 | /* |
738 | * Fallback to the default domain if kmalloc() failed. | |
739 | * See comments in partition_sched_domains(). | |
740 | */ | |
741 | if (doms == NULL) | |
742 | ndoms = 1; | |
743 | ||
cf417141 MK |
744 | *domains = doms; |
745 | *attributes = dattr; | |
746 | return ndoms; | |
747 | } | |
748 | ||
749 | /* | |
750 | * Rebuild scheduler domains. | |
751 | * | |
699140ba TH |
752 | * If the flag 'sched_load_balance' of any cpuset with non-empty |
753 | * 'cpus' changes, or if the 'cpus' allowed changes in any cpuset | |
754 | * which has that flag enabled, or if any cpuset with a non-empty | |
755 | * 'cpus' is removed, then call this routine to rebuild the | |
756 | * scheduler's dynamic sched domains. | |
cf417141 | 757 | * |
5d21cc2d | 758 | * Call with cpuset_mutex held. Takes get_online_cpus(). |
cf417141 | 759 | */ |
699140ba | 760 | static void rebuild_sched_domains_locked(void) |
cf417141 MK |
761 | { |
762 | struct sched_domain_attr *attr; | |
acc3f5d7 | 763 | cpumask_var_t *doms; |
cf417141 MK |
764 | int ndoms; |
765 | ||
5d21cc2d | 766 | lockdep_assert_held(&cpuset_mutex); |
86ef5c9a | 767 | get_online_cpus(); |
cf417141 | 768 | |
5b16c2a4 LZ |
769 | /* |
770 | * We have raced with CPU hotplug. Don't do anything to avoid | |
771 | * passing doms with offlined cpu to partition_sched_domains(). | |
772 | * Anyways, hotplug work item will rebuild sched domains. | |
773 | */ | |
774 | if (!cpumask_equal(top_cpuset.cpus_allowed, cpu_active_mask)) | |
775 | goto out; | |
776 | ||
cf417141 | 777 | /* Generate domain masks and attrs */ |
cf417141 | 778 | ndoms = generate_sched_domains(&doms, &attr); |
cf417141 MK |
779 | |
780 | /* Have scheduler rebuild the domains */ | |
781 | partition_sched_domains(ndoms, doms, attr); | |
5b16c2a4 | 782 | out: |
86ef5c9a | 783 | put_online_cpus(); |
cf417141 | 784 | } |
db7f47cf | 785 | #else /* !CONFIG_SMP */ |
699140ba | 786 | static void rebuild_sched_domains_locked(void) |
db7f47cf PM |
787 | { |
788 | } | |
db7f47cf | 789 | #endif /* CONFIG_SMP */ |
029190c5 | 790 | |
cf417141 MK |
791 | void rebuild_sched_domains(void) |
792 | { | |
5d21cc2d | 793 | mutex_lock(&cpuset_mutex); |
699140ba | 794 | rebuild_sched_domains_locked(); |
5d21cc2d | 795 | mutex_unlock(&cpuset_mutex); |
029190c5 PJ |
796 | } |
797 | ||
070b57fc LZ |
798 | /* |
799 | * effective_cpumask_cpuset - return nearest ancestor with non-empty cpus | |
800 | * @cs: the cpuset in interest | |
58f4790b | 801 | * |
070b57fc LZ |
802 | * A cpuset's effective cpumask is the cpumask of the nearest ancestor |
803 | * with non-empty cpus. We use effective cpumask whenever: | |
804 | * - we update tasks' cpus_allowed. (they take on the ancestor's cpumask | |
805 | * if the cpuset they reside in has no cpus) | |
806 | * - we want to retrieve task_cs(tsk)'s cpus_allowed. | |
807 | * | |
808 | * Called with cpuset_mutex held. cpuset_cpus_allowed_fallback() is an | |
809 | * exception. See comments there. | |
810 | */ | |
811 | static struct cpuset *effective_cpumask_cpuset(struct cpuset *cs) | |
812 | { | |
813 | while (cpumask_empty(cs->cpus_allowed)) | |
814 | cs = parent_cs(cs); | |
815 | return cs; | |
816 | } | |
817 | ||
818 | /* | |
819 | * effective_nodemask_cpuset - return nearest ancestor with non-empty mems | |
820 | * @cs: the cpuset in interest | |
821 | * | |
822 | * A cpuset's effective nodemask is the nodemask of the nearest ancestor | |
823 | * with non-empty memss. We use effective nodemask whenever: | |
824 | * - we update tasks' mems_allowed. (they take on the ancestor's nodemask | |
825 | * if the cpuset they reside in has no mems) | |
826 | * - we want to retrieve task_cs(tsk)'s mems_allowed. | |
827 | * | |
828 | * Called with cpuset_mutex held. | |
053199ed | 829 | */ |
070b57fc | 830 | static struct cpuset *effective_nodemask_cpuset(struct cpuset *cs) |
58f4790b | 831 | { |
070b57fc LZ |
832 | while (nodes_empty(cs->mems_allowed)) |
833 | cs = parent_cs(cs); | |
834 | return cs; | |
58f4790b | 835 | } |
053199ed | 836 | |
0b2f630a MX |
837 | /** |
838 | * update_tasks_cpumask - Update the cpumasks of tasks in the cpuset. | |
839 | * @cs: the cpuset in which each task's cpus_allowed mask needs to be changed | |
0b2f630a | 840 | * |
d66393e5 TH |
841 | * Iterate through each task of @cs updating its cpus_allowed to the |
842 | * effective cpuset's. As this function is called with cpuset_mutex held, | |
843 | * cpuset membership stays stable. | |
0b2f630a | 844 | */ |
d66393e5 | 845 | static void update_tasks_cpumask(struct cpuset *cs) |
0b2f630a | 846 | { |
d66393e5 TH |
847 | struct cpuset *cpus_cs = effective_cpumask_cpuset(cs); |
848 | struct css_task_iter it; | |
849 | struct task_struct *task; | |
850 | ||
851 | css_task_iter_start(&cs->css, &it); | |
852 | while ((task = css_task_iter_next(&it))) | |
853 | set_cpus_allowed_ptr(task, cpus_cs->cpus_allowed); | |
854 | css_task_iter_end(&it); | |
0b2f630a MX |
855 | } |
856 | ||
5c5cc623 | 857 | /* |
734d4513 LZ |
858 | * update_cpumasks_hier - Update effective cpumasks and tasks in the subtree |
859 | * @cs: the cpuset to consider | |
860 | * @new_cpus: temp variable for calculating new effective_cpus | |
861 | * | |
862 | * When congifured cpumask is changed, the effective cpumasks of this cpuset | |
863 | * and all its descendants need to be updated. | |
5c5cc623 | 864 | * |
734d4513 | 865 | * On legacy hierachy, effective_cpus will be the same with cpu_allowed. |
5c5cc623 LZ |
866 | * |
867 | * Called with cpuset_mutex held | |
868 | */ | |
734d4513 | 869 | static void update_cpumasks_hier(struct cpuset *cs, struct cpumask *new_cpus) |
5c5cc623 LZ |
870 | { |
871 | struct cpuset *cp; | |
492eb21b | 872 | struct cgroup_subsys_state *pos_css; |
5c5cc623 LZ |
873 | |
874 | rcu_read_lock(); | |
734d4513 LZ |
875 | cpuset_for_each_descendant_pre(cp, pos_css, cs) { |
876 | struct cpuset *parent = parent_cs(cp); | |
877 | ||
878 | cpumask_and(new_cpus, cp->cpus_allowed, parent->effective_cpus); | |
879 | ||
880 | /* Skip the whole subtree if the cpumask remains the same. */ | |
881 | if (cpumask_equal(new_cpus, cp->effective_cpus)) { | |
882 | pos_css = css_rightmost_descendant(pos_css); | |
883 | continue; | |
5c5cc623 | 884 | } |
734d4513 | 885 | |
ec903c0c | 886 | if (!css_tryget_online(&cp->css)) |
5c5cc623 LZ |
887 | continue; |
888 | rcu_read_unlock(); | |
889 | ||
734d4513 LZ |
890 | mutex_lock(&callback_mutex); |
891 | cpumask_copy(cp->effective_cpus, new_cpus); | |
892 | mutex_unlock(&callback_mutex); | |
893 | ||
894 | WARN_ON(!cgroup_on_dfl(cp->css.cgroup) && | |
895 | !cpumask_equal(cp->cpus_allowed, cp->effective_cpus)); | |
896 | ||
d66393e5 | 897 | update_tasks_cpumask(cp); |
5c5cc623 LZ |
898 | |
899 | rcu_read_lock(); | |
900 | css_put(&cp->css); | |
901 | } | |
902 | rcu_read_unlock(); | |
903 | } | |
904 | ||
58f4790b CW |
905 | /** |
906 | * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it | |
907 | * @cs: the cpuset to consider | |
fc34ac1d | 908 | * @trialcs: trial cpuset |
58f4790b CW |
909 | * @buf: buffer of cpu numbers written to this cpuset |
910 | */ | |
645fcc9d LZ |
911 | static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs, |
912 | const char *buf) | |
1da177e4 | 913 | { |
58f4790b CW |
914 | int retval; |
915 | int is_load_balanced; | |
1da177e4 | 916 | |
5f054e31 | 917 | /* top_cpuset.cpus_allowed tracks cpu_online_mask; it's read-only */ |
4c4d50f7 PJ |
918 | if (cs == &top_cpuset) |
919 | return -EACCES; | |
920 | ||
6f7f02e7 | 921 | /* |
c8d9c90c | 922 | * An empty cpus_allowed is ok only if the cpuset has no tasks. |
020958b6 PJ |
923 | * Since cpulist_parse() fails on an empty mask, we special case |
924 | * that parsing. The validate_change() call ensures that cpusets | |
925 | * with tasks have cpus. | |
6f7f02e7 | 926 | */ |
020958b6 | 927 | if (!*buf) { |
300ed6cb | 928 | cpumask_clear(trialcs->cpus_allowed); |
6f7f02e7 | 929 | } else { |
300ed6cb | 930 | retval = cpulist_parse(buf, trialcs->cpus_allowed); |
6f7f02e7 DR |
931 | if (retval < 0) |
932 | return retval; | |
37340746 | 933 | |
6ad4c188 | 934 | if (!cpumask_subset(trialcs->cpus_allowed, cpu_active_mask)) |
37340746 | 935 | return -EINVAL; |
6f7f02e7 | 936 | } |
029190c5 | 937 | |
8707d8b8 | 938 | /* Nothing to do if the cpus didn't change */ |
300ed6cb | 939 | if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed)) |
8707d8b8 | 940 | return 0; |
58f4790b | 941 | |
a73456f3 LZ |
942 | retval = validate_change(cs, trialcs); |
943 | if (retval < 0) | |
944 | return retval; | |
945 | ||
645fcc9d | 946 | is_load_balanced = is_sched_load_balance(trialcs); |
029190c5 | 947 | |
3d3f26a7 | 948 | mutex_lock(&callback_mutex); |
300ed6cb | 949 | cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed); |
3d3f26a7 | 950 | mutex_unlock(&callback_mutex); |
029190c5 | 951 | |
734d4513 LZ |
952 | /* use trialcs->cpus_allowed as a temp variable */ |
953 | update_cpumasks_hier(cs, trialcs->cpus_allowed); | |
58f4790b | 954 | |
8707d8b8 | 955 | if (is_load_balanced) |
699140ba | 956 | rebuild_sched_domains_locked(); |
85d7b949 | 957 | return 0; |
1da177e4 LT |
958 | } |
959 | ||
e4e364e8 PJ |
960 | /* |
961 | * cpuset_migrate_mm | |
962 | * | |
963 | * Migrate memory region from one set of nodes to another. | |
964 | * | |
965 | * Temporarilly set tasks mems_allowed to target nodes of migration, | |
966 | * so that the migration code can allocate pages on these nodes. | |
967 | * | |
e4e364e8 PJ |
968 | * While the mm_struct we are migrating is typically from some |
969 | * other task, the task_struct mems_allowed that we are hacking | |
970 | * is for our current task, which must allocate new pages for that | |
971 | * migrating memory region. | |
e4e364e8 PJ |
972 | */ |
973 | ||
974 | static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from, | |
975 | const nodemask_t *to) | |
976 | { | |
977 | struct task_struct *tsk = current; | |
070b57fc | 978 | struct cpuset *mems_cs; |
e4e364e8 | 979 | |
e4e364e8 | 980 | tsk->mems_allowed = *to; |
e4e364e8 PJ |
981 | |
982 | do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL); | |
983 | ||
47295830 | 984 | rcu_read_lock(); |
070b57fc LZ |
985 | mems_cs = effective_nodemask_cpuset(task_cs(tsk)); |
986 | guarantee_online_mems(mems_cs, &tsk->mems_allowed); | |
47295830 | 987 | rcu_read_unlock(); |
e4e364e8 PJ |
988 | } |
989 | ||
3b6766fe | 990 | /* |
58568d2a MX |
991 | * cpuset_change_task_nodemask - change task's mems_allowed and mempolicy |
992 | * @tsk: the task to change | |
993 | * @newmems: new nodes that the task will be set | |
994 | * | |
995 | * In order to avoid seeing no nodes if the old and new nodes are disjoint, | |
996 | * we structure updates as setting all new allowed nodes, then clearing newly | |
997 | * disallowed ones. | |
58568d2a MX |
998 | */ |
999 | static void cpuset_change_task_nodemask(struct task_struct *tsk, | |
1000 | nodemask_t *newmems) | |
1001 | { | |
b246272e | 1002 | bool need_loop; |
89e8a244 | 1003 | |
c0ff7453 MX |
1004 | /* |
1005 | * Allow tasks that have access to memory reserves because they have | |
1006 | * been OOM killed to get memory anywhere. | |
1007 | */ | |
1008 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
1009 | return; | |
1010 | if (current->flags & PF_EXITING) /* Let dying task have memory */ | |
1011 | return; | |
1012 | ||
1013 | task_lock(tsk); | |
b246272e DR |
1014 | /* |
1015 | * Determine if a loop is necessary if another thread is doing | |
d26914d1 | 1016 | * read_mems_allowed_begin(). If at least one node remains unchanged and |
b246272e DR |
1017 | * tsk does not have a mempolicy, then an empty nodemask will not be |
1018 | * possible when mems_allowed is larger than a word. | |
1019 | */ | |
1020 | need_loop = task_has_mempolicy(tsk) || | |
1021 | !nodes_intersects(*newmems, tsk->mems_allowed); | |
c0ff7453 | 1022 | |
0fc0287c PZ |
1023 | if (need_loop) { |
1024 | local_irq_disable(); | |
cc9a6c87 | 1025 | write_seqcount_begin(&tsk->mems_allowed_seq); |
0fc0287c | 1026 | } |
c0ff7453 | 1027 | |
cc9a6c87 MG |
1028 | nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems); |
1029 | mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP1); | |
c0ff7453 MX |
1030 | |
1031 | mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP2); | |
58568d2a | 1032 | tsk->mems_allowed = *newmems; |
cc9a6c87 | 1033 | |
0fc0287c | 1034 | if (need_loop) { |
cc9a6c87 | 1035 | write_seqcount_end(&tsk->mems_allowed_seq); |
0fc0287c PZ |
1036 | local_irq_enable(); |
1037 | } | |
cc9a6c87 | 1038 | |
c0ff7453 | 1039 | task_unlock(tsk); |
58568d2a MX |
1040 | } |
1041 | ||
8793d854 PM |
1042 | static void *cpuset_being_rebound; |
1043 | ||
0b2f630a MX |
1044 | /** |
1045 | * update_tasks_nodemask - Update the nodemasks of tasks in the cpuset. | |
1046 | * @cs: the cpuset in which each task's mems_allowed mask needs to be changed | |
0b2f630a | 1047 | * |
d66393e5 TH |
1048 | * Iterate through each task of @cs updating its mems_allowed to the |
1049 | * effective cpuset's. As this function is called with cpuset_mutex held, | |
1050 | * cpuset membership stays stable. | |
0b2f630a | 1051 | */ |
d66393e5 | 1052 | static void update_tasks_nodemask(struct cpuset *cs) |
1da177e4 | 1053 | { |
33ad801d | 1054 | static nodemask_t newmems; /* protected by cpuset_mutex */ |
070b57fc | 1055 | struct cpuset *mems_cs = effective_nodemask_cpuset(cs); |
d66393e5 TH |
1056 | struct css_task_iter it; |
1057 | struct task_struct *task; | |
59dac16f | 1058 | |
846a16bf | 1059 | cpuset_being_rebound = cs; /* causes mpol_dup() rebind */ |
4225399a | 1060 | |
070b57fc | 1061 | guarantee_online_mems(mems_cs, &newmems); |
33ad801d | 1062 | |
4225399a | 1063 | /* |
3b6766fe LZ |
1064 | * The mpol_rebind_mm() call takes mmap_sem, which we couldn't |
1065 | * take while holding tasklist_lock. Forks can happen - the | |
1066 | * mpol_dup() cpuset_being_rebound check will catch such forks, | |
1067 | * and rebind their vma mempolicies too. Because we still hold | |
5d21cc2d | 1068 | * the global cpuset_mutex, we know that no other rebind effort |
3b6766fe | 1069 | * will be contending for the global variable cpuset_being_rebound. |
4225399a | 1070 | * It's ok if we rebind the same mm twice; mpol_rebind_mm() |
04c19fa6 | 1071 | * is idempotent. Also migrate pages in each mm to new nodes. |
4225399a | 1072 | */ |
d66393e5 TH |
1073 | css_task_iter_start(&cs->css, &it); |
1074 | while ((task = css_task_iter_next(&it))) { | |
1075 | struct mm_struct *mm; | |
1076 | bool migrate; | |
1077 | ||
1078 | cpuset_change_task_nodemask(task, &newmems); | |
1079 | ||
1080 | mm = get_task_mm(task); | |
1081 | if (!mm) | |
1082 | continue; | |
1083 | ||
1084 | migrate = is_memory_migrate(cs); | |
1085 | ||
1086 | mpol_rebind_mm(mm, &cs->mems_allowed); | |
1087 | if (migrate) | |
1088 | cpuset_migrate_mm(mm, &cs->old_mems_allowed, &newmems); | |
1089 | mmput(mm); | |
1090 | } | |
1091 | css_task_iter_end(&it); | |
4225399a | 1092 | |
33ad801d LZ |
1093 | /* |
1094 | * All the tasks' nodemasks have been updated, update | |
1095 | * cs->old_mems_allowed. | |
1096 | */ | |
1097 | cs->old_mems_allowed = newmems; | |
1098 | ||
2df167a3 | 1099 | /* We're done rebinding vmas to this cpuset's new mems_allowed. */ |
8793d854 | 1100 | cpuset_being_rebound = NULL; |
1da177e4 LT |
1101 | } |
1102 | ||
5c5cc623 | 1103 | /* |
734d4513 LZ |
1104 | * update_nodemasks_hier - Update effective nodemasks and tasks in the subtree |
1105 | * @cs: the cpuset to consider | |
1106 | * @new_mems: a temp variable for calculating new effective_mems | |
1107 | * | |
1108 | * When configured nodemask is changed, the effective nodemasks of this cpuset | |
1109 | * and all its descendants need to be updated. | |
5c5cc623 | 1110 | * |
734d4513 | 1111 | * On legacy hiearchy, effective_mems will be the same with mems_allowed. |
5c5cc623 LZ |
1112 | * |
1113 | * Called with cpuset_mutex held | |
1114 | */ | |
734d4513 | 1115 | static void update_nodemasks_hier(struct cpuset *cs, nodemask_t *new_mems) |
5c5cc623 LZ |
1116 | { |
1117 | struct cpuset *cp; | |
492eb21b | 1118 | struct cgroup_subsys_state *pos_css; |
5c5cc623 LZ |
1119 | |
1120 | rcu_read_lock(); | |
734d4513 LZ |
1121 | cpuset_for_each_descendant_pre(cp, pos_css, cs) { |
1122 | struct cpuset *parent = parent_cs(cp); | |
1123 | ||
1124 | nodes_and(*new_mems, cp->mems_allowed, parent->effective_mems); | |
1125 | ||
1126 | /* Skip the whole subtree if the nodemask remains the same. */ | |
1127 | if (nodes_equal(*new_mems, cp->effective_mems)) { | |
1128 | pos_css = css_rightmost_descendant(pos_css); | |
1129 | continue; | |
5c5cc623 | 1130 | } |
734d4513 | 1131 | |
ec903c0c | 1132 | if (!css_tryget_online(&cp->css)) |
5c5cc623 LZ |
1133 | continue; |
1134 | rcu_read_unlock(); | |
1135 | ||
734d4513 LZ |
1136 | mutex_lock(&callback_mutex); |
1137 | cp->effective_mems = *new_mems; | |
1138 | mutex_unlock(&callback_mutex); | |
1139 | ||
1140 | WARN_ON(!cgroup_on_dfl(cp->css.cgroup) && | |
1141 | nodes_equal(cp->mems_allowed, cp->effective_mems)); | |
1142 | ||
d66393e5 | 1143 | update_tasks_nodemask(cp); |
5c5cc623 LZ |
1144 | |
1145 | rcu_read_lock(); | |
1146 | css_put(&cp->css); | |
1147 | } | |
1148 | rcu_read_unlock(); | |
1149 | } | |
1150 | ||
0b2f630a MX |
1151 | /* |
1152 | * Handle user request to change the 'mems' memory placement | |
1153 | * of a cpuset. Needs to validate the request, update the | |
58568d2a MX |
1154 | * cpusets mems_allowed, and for each task in the cpuset, |
1155 | * update mems_allowed and rebind task's mempolicy and any vma | |
1156 | * mempolicies and if the cpuset is marked 'memory_migrate', | |
1157 | * migrate the tasks pages to the new memory. | |
0b2f630a | 1158 | * |
5d21cc2d | 1159 | * Call with cpuset_mutex held. May take callback_mutex during call. |
0b2f630a MX |
1160 | * Will take tasklist_lock, scan tasklist for tasks in cpuset cs, |
1161 | * lock each such tasks mm->mmap_sem, scan its vma's and rebind | |
1162 | * their mempolicies to the cpusets new mems_allowed. | |
1163 | */ | |
645fcc9d LZ |
1164 | static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs, |
1165 | const char *buf) | |
0b2f630a | 1166 | { |
0b2f630a MX |
1167 | int retval; |
1168 | ||
1169 | /* | |
38d7bee9 | 1170 | * top_cpuset.mems_allowed tracks node_stats[N_MEMORY]; |
0b2f630a MX |
1171 | * it's read-only |
1172 | */ | |
53feb297 MX |
1173 | if (cs == &top_cpuset) { |
1174 | retval = -EACCES; | |
1175 | goto done; | |
1176 | } | |
0b2f630a | 1177 | |
0b2f630a MX |
1178 | /* |
1179 | * An empty mems_allowed is ok iff there are no tasks in the cpuset. | |
1180 | * Since nodelist_parse() fails on an empty mask, we special case | |
1181 | * that parsing. The validate_change() call ensures that cpusets | |
1182 | * with tasks have memory. | |
1183 | */ | |
1184 | if (!*buf) { | |
645fcc9d | 1185 | nodes_clear(trialcs->mems_allowed); |
0b2f630a | 1186 | } else { |
645fcc9d | 1187 | retval = nodelist_parse(buf, trialcs->mems_allowed); |
0b2f630a MX |
1188 | if (retval < 0) |
1189 | goto done; | |
1190 | ||
645fcc9d | 1191 | if (!nodes_subset(trialcs->mems_allowed, |
38d7bee9 | 1192 | node_states[N_MEMORY])) { |
53feb297 MX |
1193 | retval = -EINVAL; |
1194 | goto done; | |
1195 | } | |
0b2f630a | 1196 | } |
33ad801d LZ |
1197 | |
1198 | if (nodes_equal(cs->mems_allowed, trialcs->mems_allowed)) { | |
0b2f630a MX |
1199 | retval = 0; /* Too easy - nothing to do */ |
1200 | goto done; | |
1201 | } | |
645fcc9d | 1202 | retval = validate_change(cs, trialcs); |
0b2f630a MX |
1203 | if (retval < 0) |
1204 | goto done; | |
1205 | ||
1206 | mutex_lock(&callback_mutex); | |
645fcc9d | 1207 | cs->mems_allowed = trialcs->mems_allowed; |
0b2f630a MX |
1208 | mutex_unlock(&callback_mutex); |
1209 | ||
734d4513 LZ |
1210 | /* use trialcs->mems_allowed as a temp variable */ |
1211 | update_nodemasks_hier(cs, &cs->mems_allowed); | |
0b2f630a MX |
1212 | done: |
1213 | return retval; | |
1214 | } | |
1215 | ||
8793d854 PM |
1216 | int current_cpuset_is_being_rebound(void) |
1217 | { | |
1218 | return task_cs(current) == cpuset_being_rebound; | |
1219 | } | |
1220 | ||
5be7a479 | 1221 | static int update_relax_domain_level(struct cpuset *cs, s64 val) |
1d3504fc | 1222 | { |
db7f47cf | 1223 | #ifdef CONFIG_SMP |
60495e77 | 1224 | if (val < -1 || val >= sched_domain_level_max) |
30e0e178 | 1225 | return -EINVAL; |
db7f47cf | 1226 | #endif |
1d3504fc HS |
1227 | |
1228 | if (val != cs->relax_domain_level) { | |
1229 | cs->relax_domain_level = val; | |
300ed6cb LZ |
1230 | if (!cpumask_empty(cs->cpus_allowed) && |
1231 | is_sched_load_balance(cs)) | |
699140ba | 1232 | rebuild_sched_domains_locked(); |
1d3504fc HS |
1233 | } |
1234 | ||
1235 | return 0; | |
1236 | } | |
1237 | ||
72ec7029 | 1238 | /** |
950592f7 MX |
1239 | * update_tasks_flags - update the spread flags of tasks in the cpuset. |
1240 | * @cs: the cpuset in which each task's spread flags needs to be changed | |
950592f7 | 1241 | * |
d66393e5 TH |
1242 | * Iterate through each task of @cs updating its spread flags. As this |
1243 | * function is called with cpuset_mutex held, cpuset membership stays | |
1244 | * stable. | |
950592f7 | 1245 | */ |
d66393e5 | 1246 | static void update_tasks_flags(struct cpuset *cs) |
950592f7 | 1247 | { |
d66393e5 TH |
1248 | struct css_task_iter it; |
1249 | struct task_struct *task; | |
1250 | ||
1251 | css_task_iter_start(&cs->css, &it); | |
1252 | while ((task = css_task_iter_next(&it))) | |
1253 | cpuset_update_task_spread_flag(cs, task); | |
1254 | css_task_iter_end(&it); | |
950592f7 MX |
1255 | } |
1256 | ||
1da177e4 LT |
1257 | /* |
1258 | * update_flag - read a 0 or a 1 in a file and update associated flag | |
78608366 PM |
1259 | * bit: the bit to update (see cpuset_flagbits_t) |
1260 | * cs: the cpuset to update | |
1261 | * turning_on: whether the flag is being set or cleared | |
053199ed | 1262 | * |
5d21cc2d | 1263 | * Call with cpuset_mutex held. |
1da177e4 LT |
1264 | */ |
1265 | ||
700fe1ab PM |
1266 | static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs, |
1267 | int turning_on) | |
1da177e4 | 1268 | { |
645fcc9d | 1269 | struct cpuset *trialcs; |
40b6a762 | 1270 | int balance_flag_changed; |
950592f7 | 1271 | int spread_flag_changed; |
950592f7 | 1272 | int err; |
1da177e4 | 1273 | |
645fcc9d LZ |
1274 | trialcs = alloc_trial_cpuset(cs); |
1275 | if (!trialcs) | |
1276 | return -ENOMEM; | |
1277 | ||
1da177e4 | 1278 | if (turning_on) |
645fcc9d | 1279 | set_bit(bit, &trialcs->flags); |
1da177e4 | 1280 | else |
645fcc9d | 1281 | clear_bit(bit, &trialcs->flags); |
1da177e4 | 1282 | |
645fcc9d | 1283 | err = validate_change(cs, trialcs); |
85d7b949 | 1284 | if (err < 0) |
645fcc9d | 1285 | goto out; |
029190c5 | 1286 | |
029190c5 | 1287 | balance_flag_changed = (is_sched_load_balance(cs) != |
645fcc9d | 1288 | is_sched_load_balance(trialcs)); |
029190c5 | 1289 | |
950592f7 MX |
1290 | spread_flag_changed = ((is_spread_slab(cs) != is_spread_slab(trialcs)) |
1291 | || (is_spread_page(cs) != is_spread_page(trialcs))); | |
1292 | ||
3d3f26a7 | 1293 | mutex_lock(&callback_mutex); |
645fcc9d | 1294 | cs->flags = trialcs->flags; |
3d3f26a7 | 1295 | mutex_unlock(&callback_mutex); |
85d7b949 | 1296 | |
300ed6cb | 1297 | if (!cpumask_empty(trialcs->cpus_allowed) && balance_flag_changed) |
699140ba | 1298 | rebuild_sched_domains_locked(); |
029190c5 | 1299 | |
950592f7 | 1300 | if (spread_flag_changed) |
d66393e5 | 1301 | update_tasks_flags(cs); |
645fcc9d LZ |
1302 | out: |
1303 | free_trial_cpuset(trialcs); | |
1304 | return err; | |
1da177e4 LT |
1305 | } |
1306 | ||
3e0d98b9 | 1307 | /* |
80f7228b | 1308 | * Frequency meter - How fast is some event occurring? |
3e0d98b9 PJ |
1309 | * |
1310 | * These routines manage a digitally filtered, constant time based, | |
1311 | * event frequency meter. There are four routines: | |
1312 | * fmeter_init() - initialize a frequency meter. | |
1313 | * fmeter_markevent() - called each time the event happens. | |
1314 | * fmeter_getrate() - returns the recent rate of such events. | |
1315 | * fmeter_update() - internal routine used to update fmeter. | |
1316 | * | |
1317 | * A common data structure is passed to each of these routines, | |
1318 | * which is used to keep track of the state required to manage the | |
1319 | * frequency meter and its digital filter. | |
1320 | * | |
1321 | * The filter works on the number of events marked per unit time. | |
1322 | * The filter is single-pole low-pass recursive (IIR). The time unit | |
1323 | * is 1 second. Arithmetic is done using 32-bit integers scaled to | |
1324 | * simulate 3 decimal digits of precision (multiplied by 1000). | |
1325 | * | |
1326 | * With an FM_COEF of 933, and a time base of 1 second, the filter | |
1327 | * has a half-life of 10 seconds, meaning that if the events quit | |
1328 | * happening, then the rate returned from the fmeter_getrate() | |
1329 | * will be cut in half each 10 seconds, until it converges to zero. | |
1330 | * | |
1331 | * It is not worth doing a real infinitely recursive filter. If more | |
1332 | * than FM_MAXTICKS ticks have elapsed since the last filter event, | |
1333 | * just compute FM_MAXTICKS ticks worth, by which point the level | |
1334 | * will be stable. | |
1335 | * | |
1336 | * Limit the count of unprocessed events to FM_MAXCNT, so as to avoid | |
1337 | * arithmetic overflow in the fmeter_update() routine. | |
1338 | * | |
1339 | * Given the simple 32 bit integer arithmetic used, this meter works | |
1340 | * best for reporting rates between one per millisecond (msec) and | |
1341 | * one per 32 (approx) seconds. At constant rates faster than one | |
1342 | * per msec it maxes out at values just under 1,000,000. At constant | |
1343 | * rates between one per msec, and one per second it will stabilize | |
1344 | * to a value N*1000, where N is the rate of events per second. | |
1345 | * At constant rates between one per second and one per 32 seconds, | |
1346 | * it will be choppy, moving up on the seconds that have an event, | |
1347 | * and then decaying until the next event. At rates slower than | |
1348 | * about one in 32 seconds, it decays all the way back to zero between | |
1349 | * each event. | |
1350 | */ | |
1351 | ||
1352 | #define FM_COEF 933 /* coefficient for half-life of 10 secs */ | |
1353 | #define FM_MAXTICKS ((time_t)99) /* useless computing more ticks than this */ | |
1354 | #define FM_MAXCNT 1000000 /* limit cnt to avoid overflow */ | |
1355 | #define FM_SCALE 1000 /* faux fixed point scale */ | |
1356 | ||
1357 | /* Initialize a frequency meter */ | |
1358 | static void fmeter_init(struct fmeter *fmp) | |
1359 | { | |
1360 | fmp->cnt = 0; | |
1361 | fmp->val = 0; | |
1362 | fmp->time = 0; | |
1363 | spin_lock_init(&fmp->lock); | |
1364 | } | |
1365 | ||
1366 | /* Internal meter update - process cnt events and update value */ | |
1367 | static void fmeter_update(struct fmeter *fmp) | |
1368 | { | |
1369 | time_t now = get_seconds(); | |
1370 | time_t ticks = now - fmp->time; | |
1371 | ||
1372 | if (ticks == 0) | |
1373 | return; | |
1374 | ||
1375 | ticks = min(FM_MAXTICKS, ticks); | |
1376 | while (ticks-- > 0) | |
1377 | fmp->val = (FM_COEF * fmp->val) / FM_SCALE; | |
1378 | fmp->time = now; | |
1379 | ||
1380 | fmp->val += ((FM_SCALE - FM_COEF) * fmp->cnt) / FM_SCALE; | |
1381 | fmp->cnt = 0; | |
1382 | } | |
1383 | ||
1384 | /* Process any previous ticks, then bump cnt by one (times scale). */ | |
1385 | static void fmeter_markevent(struct fmeter *fmp) | |
1386 | { | |
1387 | spin_lock(&fmp->lock); | |
1388 | fmeter_update(fmp); | |
1389 | fmp->cnt = min(FM_MAXCNT, fmp->cnt + FM_SCALE); | |
1390 | spin_unlock(&fmp->lock); | |
1391 | } | |
1392 | ||
1393 | /* Process any previous ticks, then return current value. */ | |
1394 | static int fmeter_getrate(struct fmeter *fmp) | |
1395 | { | |
1396 | int val; | |
1397 | ||
1398 | spin_lock(&fmp->lock); | |
1399 | fmeter_update(fmp); | |
1400 | val = fmp->val; | |
1401 | spin_unlock(&fmp->lock); | |
1402 | return val; | |
1403 | } | |
1404 | ||
57fce0a6 TH |
1405 | static struct cpuset *cpuset_attach_old_cs; |
1406 | ||
5d21cc2d | 1407 | /* Called by cgroups to determine if a cpuset is usable; cpuset_mutex held */ |
eb95419b TH |
1408 | static int cpuset_can_attach(struct cgroup_subsys_state *css, |
1409 | struct cgroup_taskset *tset) | |
f780bdb7 | 1410 | { |
eb95419b | 1411 | struct cpuset *cs = css_cs(css); |
bb9d97b6 TH |
1412 | struct task_struct *task; |
1413 | int ret; | |
1da177e4 | 1414 | |
57fce0a6 TH |
1415 | /* used later by cpuset_attach() */ |
1416 | cpuset_attach_old_cs = task_cs(cgroup_taskset_first(tset)); | |
1417 | ||
5d21cc2d TH |
1418 | mutex_lock(&cpuset_mutex); |
1419 | ||
aa6ec29b | 1420 | /* allow moving tasks into an empty cpuset if on default hierarchy */ |
5d21cc2d | 1421 | ret = -ENOSPC; |
aa6ec29b | 1422 | if (!cgroup_on_dfl(css->cgroup) && |
88fa523b | 1423 | (cpumask_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed))) |
5d21cc2d | 1424 | goto out_unlock; |
9985b0ba | 1425 | |
924f0d9a | 1426 | cgroup_taskset_for_each(task, tset) { |
bb9d97b6 | 1427 | /* |
14a40ffc TH |
1428 | * Kthreads which disallow setaffinity shouldn't be moved |
1429 | * to a new cpuset; we don't want to change their cpu | |
1430 | * affinity and isolating such threads by their set of | |
1431 | * allowed nodes is unnecessary. Thus, cpusets are not | |
1432 | * applicable for such threads. This prevents checking for | |
1433 | * success of set_cpus_allowed_ptr() on all attached tasks | |
1434 | * before cpus_allowed may be changed. | |
bb9d97b6 | 1435 | */ |
5d21cc2d | 1436 | ret = -EINVAL; |
14a40ffc | 1437 | if (task->flags & PF_NO_SETAFFINITY) |
5d21cc2d TH |
1438 | goto out_unlock; |
1439 | ret = security_task_setscheduler(task); | |
1440 | if (ret) | |
1441 | goto out_unlock; | |
bb9d97b6 | 1442 | } |
f780bdb7 | 1443 | |
452477fa TH |
1444 | /* |
1445 | * Mark attach is in progress. This makes validate_change() fail | |
1446 | * changes which zero cpus/mems_allowed. | |
1447 | */ | |
1448 | cs->attach_in_progress++; | |
5d21cc2d TH |
1449 | ret = 0; |
1450 | out_unlock: | |
1451 | mutex_unlock(&cpuset_mutex); | |
1452 | return ret; | |
8793d854 | 1453 | } |
f780bdb7 | 1454 | |
eb95419b | 1455 | static void cpuset_cancel_attach(struct cgroup_subsys_state *css, |
452477fa TH |
1456 | struct cgroup_taskset *tset) |
1457 | { | |
5d21cc2d | 1458 | mutex_lock(&cpuset_mutex); |
eb95419b | 1459 | css_cs(css)->attach_in_progress--; |
5d21cc2d | 1460 | mutex_unlock(&cpuset_mutex); |
8793d854 | 1461 | } |
1da177e4 | 1462 | |
4e4c9a14 | 1463 | /* |
5d21cc2d | 1464 | * Protected by cpuset_mutex. cpus_attach is used only by cpuset_attach() |
4e4c9a14 TH |
1465 | * but we can't allocate it dynamically there. Define it global and |
1466 | * allocate from cpuset_init(). | |
1467 | */ | |
1468 | static cpumask_var_t cpus_attach; | |
1469 | ||
eb95419b TH |
1470 | static void cpuset_attach(struct cgroup_subsys_state *css, |
1471 | struct cgroup_taskset *tset) | |
8793d854 | 1472 | { |
67bd2c59 | 1473 | /* static buf protected by cpuset_mutex */ |
4e4c9a14 | 1474 | static nodemask_t cpuset_attach_nodemask_to; |
8793d854 | 1475 | struct mm_struct *mm; |
bb9d97b6 TH |
1476 | struct task_struct *task; |
1477 | struct task_struct *leader = cgroup_taskset_first(tset); | |
eb95419b | 1478 | struct cpuset *cs = css_cs(css); |
57fce0a6 | 1479 | struct cpuset *oldcs = cpuset_attach_old_cs; |
070b57fc LZ |
1480 | struct cpuset *cpus_cs = effective_cpumask_cpuset(cs); |
1481 | struct cpuset *mems_cs = effective_nodemask_cpuset(cs); | |
22fb52dd | 1482 | |
5d21cc2d TH |
1483 | mutex_lock(&cpuset_mutex); |
1484 | ||
4e4c9a14 TH |
1485 | /* prepare for attach */ |
1486 | if (cs == &top_cpuset) | |
1487 | cpumask_copy(cpus_attach, cpu_possible_mask); | |
1488 | else | |
070b57fc | 1489 | guarantee_online_cpus(cpus_cs, cpus_attach); |
4e4c9a14 | 1490 | |
070b57fc | 1491 | guarantee_online_mems(mems_cs, &cpuset_attach_nodemask_to); |
4e4c9a14 | 1492 | |
924f0d9a | 1493 | cgroup_taskset_for_each(task, tset) { |
bb9d97b6 TH |
1494 | /* |
1495 | * can_attach beforehand should guarantee that this doesn't | |
1496 | * fail. TODO: have a better way to handle failure here | |
1497 | */ | |
1498 | WARN_ON_ONCE(set_cpus_allowed_ptr(task, cpus_attach)); | |
1499 | ||
1500 | cpuset_change_task_nodemask(task, &cpuset_attach_nodemask_to); | |
1501 | cpuset_update_task_spread_flag(cs, task); | |
1502 | } | |
22fb52dd | 1503 | |
f780bdb7 BB |
1504 | /* |
1505 | * Change mm, possibly for multiple threads in a threadgroup. This is | |
1506 | * expensive and may sleep. | |
1507 | */ | |
f780bdb7 | 1508 | cpuset_attach_nodemask_to = cs->mems_allowed; |
bb9d97b6 | 1509 | mm = get_task_mm(leader); |
4225399a | 1510 | if (mm) { |
070b57fc LZ |
1511 | struct cpuset *mems_oldcs = effective_nodemask_cpuset(oldcs); |
1512 | ||
f780bdb7 | 1513 | mpol_rebind_mm(mm, &cpuset_attach_nodemask_to); |
f047cecf LZ |
1514 | |
1515 | /* | |
1516 | * old_mems_allowed is the same with mems_allowed here, except | |
1517 | * if this task is being moved automatically due to hotplug. | |
1518 | * In that case @mems_allowed has been updated and is empty, | |
1519 | * so @old_mems_allowed is the right nodesets that we migrate | |
1520 | * mm from. | |
1521 | */ | |
1522 | if (is_memory_migrate(cs)) { | |
1523 | cpuset_migrate_mm(mm, &mems_oldcs->old_mems_allowed, | |
f780bdb7 | 1524 | &cpuset_attach_nodemask_to); |
f047cecf | 1525 | } |
4225399a PJ |
1526 | mmput(mm); |
1527 | } | |
452477fa | 1528 | |
33ad801d | 1529 | cs->old_mems_allowed = cpuset_attach_nodemask_to; |
02bb5863 | 1530 | |
452477fa | 1531 | cs->attach_in_progress--; |
e44193d3 LZ |
1532 | if (!cs->attach_in_progress) |
1533 | wake_up(&cpuset_attach_wq); | |
5d21cc2d TH |
1534 | |
1535 | mutex_unlock(&cpuset_mutex); | |
1da177e4 LT |
1536 | } |
1537 | ||
1538 | /* The various types of files and directories in a cpuset file system */ | |
1539 | ||
1540 | typedef enum { | |
45b07ef3 | 1541 | FILE_MEMORY_MIGRATE, |
1da177e4 LT |
1542 | FILE_CPULIST, |
1543 | FILE_MEMLIST, | |
1544 | FILE_CPU_EXCLUSIVE, | |
1545 | FILE_MEM_EXCLUSIVE, | |
78608366 | 1546 | FILE_MEM_HARDWALL, |
029190c5 | 1547 | FILE_SCHED_LOAD_BALANCE, |
1d3504fc | 1548 | FILE_SCHED_RELAX_DOMAIN_LEVEL, |
3e0d98b9 PJ |
1549 | FILE_MEMORY_PRESSURE_ENABLED, |
1550 | FILE_MEMORY_PRESSURE, | |
825a46af PJ |
1551 | FILE_SPREAD_PAGE, |
1552 | FILE_SPREAD_SLAB, | |
1da177e4 LT |
1553 | } cpuset_filetype_t; |
1554 | ||
182446d0 TH |
1555 | static int cpuset_write_u64(struct cgroup_subsys_state *css, struct cftype *cft, |
1556 | u64 val) | |
700fe1ab | 1557 | { |
182446d0 | 1558 | struct cpuset *cs = css_cs(css); |
700fe1ab | 1559 | cpuset_filetype_t type = cft->private; |
a903f086 | 1560 | int retval = 0; |
700fe1ab | 1561 | |
5d21cc2d | 1562 | mutex_lock(&cpuset_mutex); |
a903f086 LZ |
1563 | if (!is_cpuset_online(cs)) { |
1564 | retval = -ENODEV; | |
5d21cc2d | 1565 | goto out_unlock; |
a903f086 | 1566 | } |
700fe1ab PM |
1567 | |
1568 | switch (type) { | |
1da177e4 | 1569 | case FILE_CPU_EXCLUSIVE: |
700fe1ab | 1570 | retval = update_flag(CS_CPU_EXCLUSIVE, cs, val); |
1da177e4 LT |
1571 | break; |
1572 | case FILE_MEM_EXCLUSIVE: | |
700fe1ab | 1573 | retval = update_flag(CS_MEM_EXCLUSIVE, cs, val); |
1da177e4 | 1574 | break; |
78608366 PM |
1575 | case FILE_MEM_HARDWALL: |
1576 | retval = update_flag(CS_MEM_HARDWALL, cs, val); | |
1577 | break; | |
029190c5 | 1578 | case FILE_SCHED_LOAD_BALANCE: |
700fe1ab | 1579 | retval = update_flag(CS_SCHED_LOAD_BALANCE, cs, val); |
1d3504fc | 1580 | break; |
45b07ef3 | 1581 | case FILE_MEMORY_MIGRATE: |
700fe1ab | 1582 | retval = update_flag(CS_MEMORY_MIGRATE, cs, val); |
45b07ef3 | 1583 | break; |
3e0d98b9 | 1584 | case FILE_MEMORY_PRESSURE_ENABLED: |
700fe1ab | 1585 | cpuset_memory_pressure_enabled = !!val; |
3e0d98b9 PJ |
1586 | break; |
1587 | case FILE_MEMORY_PRESSURE: | |
1588 | retval = -EACCES; | |
1589 | break; | |
825a46af | 1590 | case FILE_SPREAD_PAGE: |
700fe1ab | 1591 | retval = update_flag(CS_SPREAD_PAGE, cs, val); |
825a46af PJ |
1592 | break; |
1593 | case FILE_SPREAD_SLAB: | |
700fe1ab | 1594 | retval = update_flag(CS_SPREAD_SLAB, cs, val); |
825a46af | 1595 | break; |
1da177e4 LT |
1596 | default: |
1597 | retval = -EINVAL; | |
700fe1ab | 1598 | break; |
1da177e4 | 1599 | } |
5d21cc2d TH |
1600 | out_unlock: |
1601 | mutex_unlock(&cpuset_mutex); | |
1da177e4 LT |
1602 | return retval; |
1603 | } | |
1604 | ||
182446d0 TH |
1605 | static int cpuset_write_s64(struct cgroup_subsys_state *css, struct cftype *cft, |
1606 | s64 val) | |
5be7a479 | 1607 | { |
182446d0 | 1608 | struct cpuset *cs = css_cs(css); |
5be7a479 | 1609 | cpuset_filetype_t type = cft->private; |
5d21cc2d | 1610 | int retval = -ENODEV; |
5be7a479 | 1611 | |
5d21cc2d TH |
1612 | mutex_lock(&cpuset_mutex); |
1613 | if (!is_cpuset_online(cs)) | |
1614 | goto out_unlock; | |
e3712395 | 1615 | |
5be7a479 PM |
1616 | switch (type) { |
1617 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1618 | retval = update_relax_domain_level(cs, val); | |
1619 | break; | |
1620 | default: | |
1621 | retval = -EINVAL; | |
1622 | break; | |
1623 | } | |
5d21cc2d TH |
1624 | out_unlock: |
1625 | mutex_unlock(&cpuset_mutex); | |
5be7a479 PM |
1626 | return retval; |
1627 | } | |
1628 | ||
e3712395 PM |
1629 | /* |
1630 | * Common handling for a write to a "cpus" or "mems" file. | |
1631 | */ | |
451af504 TH |
1632 | static ssize_t cpuset_write_resmask(struct kernfs_open_file *of, |
1633 | char *buf, size_t nbytes, loff_t off) | |
e3712395 | 1634 | { |
451af504 | 1635 | struct cpuset *cs = css_cs(of_css(of)); |
645fcc9d | 1636 | struct cpuset *trialcs; |
5d21cc2d | 1637 | int retval = -ENODEV; |
e3712395 | 1638 | |
451af504 TH |
1639 | buf = strstrip(buf); |
1640 | ||
3a5a6d0c TH |
1641 | /* |
1642 | * CPU or memory hotunplug may leave @cs w/o any execution | |
1643 | * resources, in which case the hotplug code asynchronously updates | |
1644 | * configuration and transfers all tasks to the nearest ancestor | |
1645 | * which can execute. | |
1646 | * | |
1647 | * As writes to "cpus" or "mems" may restore @cs's execution | |
1648 | * resources, wait for the previously scheduled operations before | |
1649 | * proceeding, so that we don't end up keep removing tasks added | |
1650 | * after execution capability is restored. | |
1651 | */ | |
1652 | flush_work(&cpuset_hotplug_work); | |
1653 | ||
5d21cc2d TH |
1654 | mutex_lock(&cpuset_mutex); |
1655 | if (!is_cpuset_online(cs)) | |
1656 | goto out_unlock; | |
e3712395 | 1657 | |
645fcc9d | 1658 | trialcs = alloc_trial_cpuset(cs); |
b75f38d6 LZ |
1659 | if (!trialcs) { |
1660 | retval = -ENOMEM; | |
5d21cc2d | 1661 | goto out_unlock; |
b75f38d6 | 1662 | } |
645fcc9d | 1663 | |
451af504 | 1664 | switch (of_cft(of)->private) { |
e3712395 | 1665 | case FILE_CPULIST: |
645fcc9d | 1666 | retval = update_cpumask(cs, trialcs, buf); |
e3712395 PM |
1667 | break; |
1668 | case FILE_MEMLIST: | |
645fcc9d | 1669 | retval = update_nodemask(cs, trialcs, buf); |
e3712395 PM |
1670 | break; |
1671 | default: | |
1672 | retval = -EINVAL; | |
1673 | break; | |
1674 | } | |
645fcc9d LZ |
1675 | |
1676 | free_trial_cpuset(trialcs); | |
5d21cc2d TH |
1677 | out_unlock: |
1678 | mutex_unlock(&cpuset_mutex); | |
451af504 | 1679 | return retval ?: nbytes; |
e3712395 PM |
1680 | } |
1681 | ||
1da177e4 LT |
1682 | /* |
1683 | * These ascii lists should be read in a single call, by using a user | |
1684 | * buffer large enough to hold the entire map. If read in smaller | |
1685 | * chunks, there is no guarantee of atomicity. Since the display format | |
1686 | * used, list of ranges of sequential numbers, is variable length, | |
1687 | * and since these maps can change value dynamically, one could read | |
1688 | * gibberish by doing partial reads while a list was changing. | |
1da177e4 | 1689 | */ |
2da8ca82 | 1690 | static int cpuset_common_seq_show(struct seq_file *sf, void *v) |
1da177e4 | 1691 | { |
2da8ca82 TH |
1692 | struct cpuset *cs = css_cs(seq_css(sf)); |
1693 | cpuset_filetype_t type = seq_cft(sf)->private; | |
51ffe411 TH |
1694 | ssize_t count; |
1695 | char *buf, *s; | |
1696 | int ret = 0; | |
1da177e4 | 1697 | |
51ffe411 TH |
1698 | count = seq_get_buf(sf, &buf); |
1699 | s = buf; | |
1da177e4 | 1700 | |
51ffe411 | 1701 | mutex_lock(&callback_mutex); |
1da177e4 LT |
1702 | |
1703 | switch (type) { | |
1704 | case FILE_CPULIST: | |
51ffe411 | 1705 | s += cpulist_scnprintf(s, count, cs->cpus_allowed); |
1da177e4 LT |
1706 | break; |
1707 | case FILE_MEMLIST: | |
51ffe411 | 1708 | s += nodelist_scnprintf(s, count, cs->mems_allowed); |
1da177e4 | 1709 | break; |
1da177e4 | 1710 | default: |
51ffe411 TH |
1711 | ret = -EINVAL; |
1712 | goto out_unlock; | |
1da177e4 | 1713 | } |
1da177e4 | 1714 | |
51ffe411 TH |
1715 | if (s < buf + count - 1) { |
1716 | *s++ = '\n'; | |
1717 | seq_commit(sf, s - buf); | |
1718 | } else { | |
1719 | seq_commit(sf, -1); | |
1720 | } | |
1721 | out_unlock: | |
1722 | mutex_unlock(&callback_mutex); | |
1723 | return ret; | |
1da177e4 LT |
1724 | } |
1725 | ||
182446d0 | 1726 | static u64 cpuset_read_u64(struct cgroup_subsys_state *css, struct cftype *cft) |
700fe1ab | 1727 | { |
182446d0 | 1728 | struct cpuset *cs = css_cs(css); |
700fe1ab PM |
1729 | cpuset_filetype_t type = cft->private; |
1730 | switch (type) { | |
1731 | case FILE_CPU_EXCLUSIVE: | |
1732 | return is_cpu_exclusive(cs); | |
1733 | case FILE_MEM_EXCLUSIVE: | |
1734 | return is_mem_exclusive(cs); | |
78608366 PM |
1735 | case FILE_MEM_HARDWALL: |
1736 | return is_mem_hardwall(cs); | |
700fe1ab PM |
1737 | case FILE_SCHED_LOAD_BALANCE: |
1738 | return is_sched_load_balance(cs); | |
1739 | case FILE_MEMORY_MIGRATE: | |
1740 | return is_memory_migrate(cs); | |
1741 | case FILE_MEMORY_PRESSURE_ENABLED: | |
1742 | return cpuset_memory_pressure_enabled; | |
1743 | case FILE_MEMORY_PRESSURE: | |
1744 | return fmeter_getrate(&cs->fmeter); | |
1745 | case FILE_SPREAD_PAGE: | |
1746 | return is_spread_page(cs); | |
1747 | case FILE_SPREAD_SLAB: | |
1748 | return is_spread_slab(cs); | |
1749 | default: | |
1750 | BUG(); | |
1751 | } | |
cf417141 MK |
1752 | |
1753 | /* Unreachable but makes gcc happy */ | |
1754 | return 0; | |
700fe1ab | 1755 | } |
1da177e4 | 1756 | |
182446d0 | 1757 | static s64 cpuset_read_s64(struct cgroup_subsys_state *css, struct cftype *cft) |
5be7a479 | 1758 | { |
182446d0 | 1759 | struct cpuset *cs = css_cs(css); |
5be7a479 PM |
1760 | cpuset_filetype_t type = cft->private; |
1761 | switch (type) { | |
1762 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1763 | return cs->relax_domain_level; | |
1764 | default: | |
1765 | BUG(); | |
1766 | } | |
cf417141 MK |
1767 | |
1768 | /* Unrechable but makes gcc happy */ | |
1769 | return 0; | |
5be7a479 PM |
1770 | } |
1771 | ||
1da177e4 LT |
1772 | |
1773 | /* | |
1774 | * for the common functions, 'private' gives the type of file | |
1775 | */ | |
1776 | ||
addf2c73 PM |
1777 | static struct cftype files[] = { |
1778 | { | |
1779 | .name = "cpus", | |
2da8ca82 | 1780 | .seq_show = cpuset_common_seq_show, |
451af504 | 1781 | .write = cpuset_write_resmask, |
e3712395 | 1782 | .max_write_len = (100U + 6 * NR_CPUS), |
addf2c73 PM |
1783 | .private = FILE_CPULIST, |
1784 | }, | |
1785 | ||
1786 | { | |
1787 | .name = "mems", | |
2da8ca82 | 1788 | .seq_show = cpuset_common_seq_show, |
451af504 | 1789 | .write = cpuset_write_resmask, |
e3712395 | 1790 | .max_write_len = (100U + 6 * MAX_NUMNODES), |
addf2c73 PM |
1791 | .private = FILE_MEMLIST, |
1792 | }, | |
1793 | ||
1794 | { | |
1795 | .name = "cpu_exclusive", | |
1796 | .read_u64 = cpuset_read_u64, | |
1797 | .write_u64 = cpuset_write_u64, | |
1798 | .private = FILE_CPU_EXCLUSIVE, | |
1799 | }, | |
1800 | ||
1801 | { | |
1802 | .name = "mem_exclusive", | |
1803 | .read_u64 = cpuset_read_u64, | |
1804 | .write_u64 = cpuset_write_u64, | |
1805 | .private = FILE_MEM_EXCLUSIVE, | |
1806 | }, | |
1807 | ||
78608366 PM |
1808 | { |
1809 | .name = "mem_hardwall", | |
1810 | .read_u64 = cpuset_read_u64, | |
1811 | .write_u64 = cpuset_write_u64, | |
1812 | .private = FILE_MEM_HARDWALL, | |
1813 | }, | |
1814 | ||
addf2c73 PM |
1815 | { |
1816 | .name = "sched_load_balance", | |
1817 | .read_u64 = cpuset_read_u64, | |
1818 | .write_u64 = cpuset_write_u64, | |
1819 | .private = FILE_SCHED_LOAD_BALANCE, | |
1820 | }, | |
1821 | ||
1822 | { | |
1823 | .name = "sched_relax_domain_level", | |
5be7a479 PM |
1824 | .read_s64 = cpuset_read_s64, |
1825 | .write_s64 = cpuset_write_s64, | |
addf2c73 PM |
1826 | .private = FILE_SCHED_RELAX_DOMAIN_LEVEL, |
1827 | }, | |
1828 | ||
1829 | { | |
1830 | .name = "memory_migrate", | |
1831 | .read_u64 = cpuset_read_u64, | |
1832 | .write_u64 = cpuset_write_u64, | |
1833 | .private = FILE_MEMORY_MIGRATE, | |
1834 | }, | |
1835 | ||
1836 | { | |
1837 | .name = "memory_pressure", | |
1838 | .read_u64 = cpuset_read_u64, | |
1839 | .write_u64 = cpuset_write_u64, | |
1840 | .private = FILE_MEMORY_PRESSURE, | |
099fca32 | 1841 | .mode = S_IRUGO, |
addf2c73 PM |
1842 | }, |
1843 | ||
1844 | { | |
1845 | .name = "memory_spread_page", | |
1846 | .read_u64 = cpuset_read_u64, | |
1847 | .write_u64 = cpuset_write_u64, | |
1848 | .private = FILE_SPREAD_PAGE, | |
1849 | }, | |
1850 | ||
1851 | { | |
1852 | .name = "memory_spread_slab", | |
1853 | .read_u64 = cpuset_read_u64, | |
1854 | .write_u64 = cpuset_write_u64, | |
1855 | .private = FILE_SPREAD_SLAB, | |
1856 | }, | |
3e0d98b9 | 1857 | |
4baf6e33 TH |
1858 | { |
1859 | .name = "memory_pressure_enabled", | |
1860 | .flags = CFTYPE_ONLY_ON_ROOT, | |
1861 | .read_u64 = cpuset_read_u64, | |
1862 | .write_u64 = cpuset_write_u64, | |
1863 | .private = FILE_MEMORY_PRESSURE_ENABLED, | |
1864 | }, | |
1da177e4 | 1865 | |
4baf6e33 TH |
1866 | { } /* terminate */ |
1867 | }; | |
1da177e4 LT |
1868 | |
1869 | /* | |
92fb9748 | 1870 | * cpuset_css_alloc - allocate a cpuset css |
c9e5fe66 | 1871 | * cgrp: control group that the new cpuset will be part of |
1da177e4 LT |
1872 | */ |
1873 | ||
eb95419b TH |
1874 | static struct cgroup_subsys_state * |
1875 | cpuset_css_alloc(struct cgroup_subsys_state *parent_css) | |
1da177e4 | 1876 | { |
c8f699bb | 1877 | struct cpuset *cs; |
1da177e4 | 1878 | |
eb95419b | 1879 | if (!parent_css) |
8793d854 | 1880 | return &top_cpuset.css; |
033fa1c5 | 1881 | |
c8f699bb | 1882 | cs = kzalloc(sizeof(*cs), GFP_KERNEL); |
1da177e4 | 1883 | if (!cs) |
8793d854 | 1884 | return ERR_PTR(-ENOMEM); |
e2b9a3d7 LZ |
1885 | if (!alloc_cpumask_var(&cs->cpus_allowed, GFP_KERNEL)) |
1886 | goto free_cs; | |
1887 | if (!alloc_cpumask_var(&cs->effective_cpus, GFP_KERNEL)) | |
1888 | goto free_cpus; | |
1da177e4 | 1889 | |
029190c5 | 1890 | set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); |
300ed6cb | 1891 | cpumask_clear(cs->cpus_allowed); |
f9a86fcb | 1892 | nodes_clear(cs->mems_allowed); |
e2b9a3d7 LZ |
1893 | cpumask_clear(cs->effective_cpus); |
1894 | nodes_clear(cs->effective_mems); | |
3e0d98b9 | 1895 | fmeter_init(&cs->fmeter); |
1d3504fc | 1896 | cs->relax_domain_level = -1; |
1da177e4 | 1897 | |
c8f699bb | 1898 | return &cs->css; |
e2b9a3d7 LZ |
1899 | |
1900 | free_cpus: | |
1901 | free_cpumask_var(cs->cpus_allowed); | |
1902 | free_cs: | |
1903 | kfree(cs); | |
1904 | return ERR_PTR(-ENOMEM); | |
c8f699bb TH |
1905 | } |
1906 | ||
eb95419b | 1907 | static int cpuset_css_online(struct cgroup_subsys_state *css) |
c8f699bb | 1908 | { |
eb95419b | 1909 | struct cpuset *cs = css_cs(css); |
c431069f | 1910 | struct cpuset *parent = parent_cs(cs); |
ae8086ce | 1911 | struct cpuset *tmp_cs; |
492eb21b | 1912 | struct cgroup_subsys_state *pos_css; |
c8f699bb TH |
1913 | |
1914 | if (!parent) | |
1915 | return 0; | |
1916 | ||
5d21cc2d TH |
1917 | mutex_lock(&cpuset_mutex); |
1918 | ||
efeb77b2 | 1919 | set_bit(CS_ONLINE, &cs->flags); |
c8f699bb TH |
1920 | if (is_spread_page(parent)) |
1921 | set_bit(CS_SPREAD_PAGE, &cs->flags); | |
1922 | if (is_spread_slab(parent)) | |
1923 | set_bit(CS_SPREAD_SLAB, &cs->flags); | |
1da177e4 | 1924 | |
664eedde | 1925 | cpuset_inc(); |
033fa1c5 | 1926 | |
e2b9a3d7 LZ |
1927 | mutex_lock(&callback_mutex); |
1928 | if (cgroup_on_dfl(cs->css.cgroup)) { | |
1929 | cpumask_copy(cs->effective_cpus, parent->effective_cpus); | |
1930 | cs->effective_mems = parent->effective_mems; | |
1931 | } | |
1932 | mutex_unlock(&callback_mutex); | |
1933 | ||
eb95419b | 1934 | if (!test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags)) |
5d21cc2d | 1935 | goto out_unlock; |
033fa1c5 TH |
1936 | |
1937 | /* | |
1938 | * Clone @parent's configuration if CGRP_CPUSET_CLONE_CHILDREN is | |
1939 | * set. This flag handling is implemented in cgroup core for | |
1940 | * histrical reasons - the flag may be specified during mount. | |
1941 | * | |
1942 | * Currently, if any sibling cpusets have exclusive cpus or mem, we | |
1943 | * refuse to clone the configuration - thereby refusing the task to | |
1944 | * be entered, and as a result refusing the sys_unshare() or | |
1945 | * clone() which initiated it. If this becomes a problem for some | |
1946 | * users who wish to allow that scenario, then this could be | |
1947 | * changed to grant parent->cpus_allowed-sibling_cpus_exclusive | |
1948 | * (and likewise for mems) to the new cgroup. | |
1949 | */ | |
ae8086ce | 1950 | rcu_read_lock(); |
492eb21b | 1951 | cpuset_for_each_child(tmp_cs, pos_css, parent) { |
ae8086ce TH |
1952 | if (is_mem_exclusive(tmp_cs) || is_cpu_exclusive(tmp_cs)) { |
1953 | rcu_read_unlock(); | |
5d21cc2d | 1954 | goto out_unlock; |
ae8086ce | 1955 | } |
033fa1c5 | 1956 | } |
ae8086ce | 1957 | rcu_read_unlock(); |
033fa1c5 TH |
1958 | |
1959 | mutex_lock(&callback_mutex); | |
1960 | cs->mems_allowed = parent->mems_allowed; | |
1961 | cpumask_copy(cs->cpus_allowed, parent->cpus_allowed); | |
1962 | mutex_unlock(&callback_mutex); | |
5d21cc2d TH |
1963 | out_unlock: |
1964 | mutex_unlock(&cpuset_mutex); | |
c8f699bb TH |
1965 | return 0; |
1966 | } | |
1967 | ||
0b9e6965 ZH |
1968 | /* |
1969 | * If the cpuset being removed has its flag 'sched_load_balance' | |
1970 | * enabled, then simulate turning sched_load_balance off, which | |
1971 | * will call rebuild_sched_domains_locked(). | |
1972 | */ | |
1973 | ||
eb95419b | 1974 | static void cpuset_css_offline(struct cgroup_subsys_state *css) |
c8f699bb | 1975 | { |
eb95419b | 1976 | struct cpuset *cs = css_cs(css); |
c8f699bb | 1977 | |
5d21cc2d | 1978 | mutex_lock(&cpuset_mutex); |
c8f699bb TH |
1979 | |
1980 | if (is_sched_load_balance(cs)) | |
1981 | update_flag(CS_SCHED_LOAD_BALANCE, cs, 0); | |
1982 | ||
664eedde | 1983 | cpuset_dec(); |
efeb77b2 | 1984 | clear_bit(CS_ONLINE, &cs->flags); |
c8f699bb | 1985 | |
5d21cc2d | 1986 | mutex_unlock(&cpuset_mutex); |
1da177e4 LT |
1987 | } |
1988 | ||
eb95419b | 1989 | static void cpuset_css_free(struct cgroup_subsys_state *css) |
1da177e4 | 1990 | { |
eb95419b | 1991 | struct cpuset *cs = css_cs(css); |
1da177e4 | 1992 | |
e2b9a3d7 | 1993 | free_cpumask_var(cs->effective_cpus); |
300ed6cb | 1994 | free_cpumask_var(cs->cpus_allowed); |
8793d854 | 1995 | kfree(cs); |
1da177e4 LT |
1996 | } |
1997 | ||
073219e9 | 1998 | struct cgroup_subsys cpuset_cgrp_subsys = { |
92fb9748 | 1999 | .css_alloc = cpuset_css_alloc, |
c8f699bb TH |
2000 | .css_online = cpuset_css_online, |
2001 | .css_offline = cpuset_css_offline, | |
92fb9748 | 2002 | .css_free = cpuset_css_free, |
8793d854 | 2003 | .can_attach = cpuset_can_attach, |
452477fa | 2004 | .cancel_attach = cpuset_cancel_attach, |
8793d854 | 2005 | .attach = cpuset_attach, |
4baf6e33 | 2006 | .base_cftypes = files, |
8793d854 PM |
2007 | .early_init = 1, |
2008 | }; | |
2009 | ||
1da177e4 LT |
2010 | /** |
2011 | * cpuset_init - initialize cpusets at system boot | |
2012 | * | |
2013 | * Description: Initialize top_cpuset and the cpuset internal file system, | |
2014 | **/ | |
2015 | ||
2016 | int __init cpuset_init(void) | |
2017 | { | |
8793d854 | 2018 | int err = 0; |
1da177e4 | 2019 | |
58568d2a MX |
2020 | if (!alloc_cpumask_var(&top_cpuset.cpus_allowed, GFP_KERNEL)) |
2021 | BUG(); | |
e2b9a3d7 LZ |
2022 | if (!alloc_cpumask_var(&top_cpuset.effective_cpus, GFP_KERNEL)) |
2023 | BUG(); | |
58568d2a | 2024 | |
300ed6cb | 2025 | cpumask_setall(top_cpuset.cpus_allowed); |
f9a86fcb | 2026 | nodes_setall(top_cpuset.mems_allowed); |
e2b9a3d7 LZ |
2027 | cpumask_setall(top_cpuset.effective_cpus); |
2028 | nodes_setall(top_cpuset.effective_mems); | |
1da177e4 | 2029 | |
3e0d98b9 | 2030 | fmeter_init(&top_cpuset.fmeter); |
029190c5 | 2031 | set_bit(CS_SCHED_LOAD_BALANCE, &top_cpuset.flags); |
1d3504fc | 2032 | top_cpuset.relax_domain_level = -1; |
1da177e4 | 2033 | |
1da177e4 LT |
2034 | err = register_filesystem(&cpuset_fs_type); |
2035 | if (err < 0) | |
8793d854 PM |
2036 | return err; |
2037 | ||
2341d1b6 LZ |
2038 | if (!alloc_cpumask_var(&cpus_attach, GFP_KERNEL)) |
2039 | BUG(); | |
2040 | ||
8793d854 | 2041 | return 0; |
1da177e4 LT |
2042 | } |
2043 | ||
b1aac8bb | 2044 | /* |
cf417141 | 2045 | * If CPU and/or memory hotplug handlers, below, unplug any CPUs |
b1aac8bb PJ |
2046 | * or memory nodes, we need to walk over the cpuset hierarchy, |
2047 | * removing that CPU or node from all cpusets. If this removes the | |
956db3ca CW |
2048 | * last CPU or node from a cpuset, then move the tasks in the empty |
2049 | * cpuset to its next-highest non-empty parent. | |
b1aac8bb | 2050 | */ |
956db3ca CW |
2051 | static void remove_tasks_in_empty_cpuset(struct cpuset *cs) |
2052 | { | |
2053 | struct cpuset *parent; | |
2054 | ||
956db3ca CW |
2055 | /* |
2056 | * Find its next-highest non-empty parent, (top cpuset | |
2057 | * has online cpus, so can't be empty). | |
2058 | */ | |
c431069f | 2059 | parent = parent_cs(cs); |
300ed6cb | 2060 | while (cpumask_empty(parent->cpus_allowed) || |
b4501295 | 2061 | nodes_empty(parent->mems_allowed)) |
c431069f | 2062 | parent = parent_cs(parent); |
956db3ca | 2063 | |
8cc99345 | 2064 | if (cgroup_transfer_tasks(parent->css.cgroup, cs->css.cgroup)) { |
12d3089c | 2065 | pr_err("cpuset: failed to transfer tasks out of empty cpuset "); |
e61734c5 TH |
2066 | pr_cont_cgroup_name(cs->css.cgroup); |
2067 | pr_cont("\n"); | |
8cc99345 | 2068 | } |
956db3ca CW |
2069 | } |
2070 | ||
deb7aa30 | 2071 | /** |
388afd85 | 2072 | * cpuset_hotplug_update_tasks - update tasks in a cpuset for hotunplug |
deb7aa30 | 2073 | * @cs: cpuset in interest |
956db3ca | 2074 | * |
deb7aa30 TH |
2075 | * Compare @cs's cpu and mem masks against top_cpuset and if some have gone |
2076 | * offline, update @cs accordingly. If @cs ends up with no CPU or memory, | |
2077 | * all its tasks are moved to the nearest ancestor with both resources. | |
80d1fa64 | 2078 | */ |
388afd85 | 2079 | static void cpuset_hotplug_update_tasks(struct cpuset *cs) |
80d1fa64 | 2080 | { |
deb7aa30 | 2081 | static cpumask_t off_cpus; |
33ad801d | 2082 | static nodemask_t off_mems; |
5d21cc2d | 2083 | bool is_empty; |
aa6ec29b | 2084 | bool on_dfl = cgroup_on_dfl(cs->css.cgroup); |
80d1fa64 | 2085 | |
e44193d3 LZ |
2086 | retry: |
2087 | wait_event(cpuset_attach_wq, cs->attach_in_progress == 0); | |
80d1fa64 | 2088 | |
5d21cc2d | 2089 | mutex_lock(&cpuset_mutex); |
7ddf96b0 | 2090 | |
e44193d3 LZ |
2091 | /* |
2092 | * We have raced with task attaching. We wait until attaching | |
2093 | * is finished, so we won't attach a task to an empty cpuset. | |
2094 | */ | |
2095 | if (cs->attach_in_progress) { | |
2096 | mutex_unlock(&cpuset_mutex); | |
2097 | goto retry; | |
2098 | } | |
2099 | ||
deb7aa30 TH |
2100 | cpumask_andnot(&off_cpus, cs->cpus_allowed, top_cpuset.cpus_allowed); |
2101 | nodes_andnot(off_mems, cs->mems_allowed, top_cpuset.mems_allowed); | |
80d1fa64 | 2102 | |
5c5cc623 LZ |
2103 | mutex_lock(&callback_mutex); |
2104 | cpumask_andnot(cs->cpus_allowed, cs->cpus_allowed, &off_cpus); | |
1344ab9c | 2105 | cpumask_andnot(cs->effective_cpus, cs->effective_cpus, &off_cpus); |
5c5cc623 LZ |
2106 | mutex_unlock(&callback_mutex); |
2107 | ||
2108 | /* | |
aa6ec29b TH |
2109 | * If on_dfl, we need to update tasks' cpumask for empty cpuset to |
2110 | * take on ancestor's cpumask. Otherwise, don't call | |
2111 | * update_tasks_cpumask() if the cpuset becomes empty, as the tasks | |
2112 | * in it will be migrated to an ancestor. | |
5c5cc623 | 2113 | */ |
aa6ec29b | 2114 | if ((on_dfl && cpumask_empty(cs->cpus_allowed)) || |
f047cecf | 2115 | (!cpumask_empty(&off_cpus) && !cpumask_empty(cs->cpus_allowed))) |
d66393e5 | 2116 | update_tasks_cpumask(cs); |
80d1fa64 | 2117 | |
5c5cc623 LZ |
2118 | mutex_lock(&callback_mutex); |
2119 | nodes_andnot(cs->mems_allowed, cs->mems_allowed, off_mems); | |
1344ab9c | 2120 | nodes_andnot(cs->effective_mems, cs->effective_mems, off_mems); |
5c5cc623 LZ |
2121 | mutex_unlock(&callback_mutex); |
2122 | ||
2123 | /* | |
aa6ec29b TH |
2124 | * If on_dfl, we need to update tasks' nodemask for empty cpuset to |
2125 | * take on ancestor's nodemask. Otherwise, don't call | |
2126 | * update_tasks_nodemask() if the cpuset becomes empty, as the | |
2127 | * tasks in it will be migratd to an ancestor. | |
5c5cc623 | 2128 | */ |
aa6ec29b | 2129 | if ((on_dfl && nodes_empty(cs->mems_allowed)) || |
f047cecf | 2130 | (!nodes_empty(off_mems) && !nodes_empty(cs->mems_allowed))) |
d66393e5 | 2131 | update_tasks_nodemask(cs); |
deb7aa30 | 2132 | |
5d21cc2d TH |
2133 | is_empty = cpumask_empty(cs->cpus_allowed) || |
2134 | nodes_empty(cs->mems_allowed); | |
8d033948 | 2135 | |
5d21cc2d TH |
2136 | mutex_unlock(&cpuset_mutex); |
2137 | ||
2138 | /* | |
aa6ec29b | 2139 | * If on_dfl, we'll keep tasks in empty cpusets. |
5c5cc623 LZ |
2140 | * |
2141 | * Otherwise move tasks to the nearest ancestor with execution | |
2142 | * resources. This is full cgroup operation which will | |
5d21cc2d TH |
2143 | * also call back into cpuset. Should be done outside any lock. |
2144 | */ | |
aa6ec29b | 2145 | if (!on_dfl && is_empty) |
5d21cc2d | 2146 | remove_tasks_in_empty_cpuset(cs); |
b1aac8bb PJ |
2147 | } |
2148 | ||
deb7aa30 | 2149 | /** |
3a5a6d0c | 2150 | * cpuset_hotplug_workfn - handle CPU/memory hotunplug for a cpuset |
956db3ca | 2151 | * |
deb7aa30 TH |
2152 | * This function is called after either CPU or memory configuration has |
2153 | * changed and updates cpuset accordingly. The top_cpuset is always | |
2154 | * synchronized to cpu_active_mask and N_MEMORY, which is necessary in | |
2155 | * order to make cpusets transparent (of no affect) on systems that are | |
2156 | * actively using CPU hotplug but making no active use of cpusets. | |
956db3ca | 2157 | * |
deb7aa30 | 2158 | * Non-root cpusets are only affected by offlining. If any CPUs or memory |
388afd85 LZ |
2159 | * nodes have been taken down, cpuset_hotplug_update_tasks() is invoked on |
2160 | * all descendants. | |
956db3ca | 2161 | * |
deb7aa30 TH |
2162 | * Note that CPU offlining during suspend is ignored. We don't modify |
2163 | * cpusets across suspend/resume cycles at all. | |
956db3ca | 2164 | */ |
3a5a6d0c | 2165 | static void cpuset_hotplug_workfn(struct work_struct *work) |
b1aac8bb | 2166 | { |
5c5cc623 LZ |
2167 | static cpumask_t new_cpus; |
2168 | static nodemask_t new_mems; | |
deb7aa30 | 2169 | bool cpus_updated, mems_updated; |
b1aac8bb | 2170 | |
5d21cc2d | 2171 | mutex_lock(&cpuset_mutex); |
956db3ca | 2172 | |
deb7aa30 TH |
2173 | /* fetch the available cpus/mems and find out which changed how */ |
2174 | cpumask_copy(&new_cpus, cpu_active_mask); | |
2175 | new_mems = node_states[N_MEMORY]; | |
7ddf96b0 | 2176 | |
deb7aa30 | 2177 | cpus_updated = !cpumask_equal(top_cpuset.cpus_allowed, &new_cpus); |
deb7aa30 | 2178 | mems_updated = !nodes_equal(top_cpuset.mems_allowed, new_mems); |
7ddf96b0 | 2179 | |
deb7aa30 TH |
2180 | /* synchronize cpus_allowed to cpu_active_mask */ |
2181 | if (cpus_updated) { | |
2182 | mutex_lock(&callback_mutex); | |
2183 | cpumask_copy(top_cpuset.cpus_allowed, &new_cpus); | |
1344ab9c | 2184 | cpumask_copy(top_cpuset.effective_cpus, &new_cpus); |
deb7aa30 TH |
2185 | mutex_unlock(&callback_mutex); |
2186 | /* we don't mess with cpumasks of tasks in top_cpuset */ | |
2187 | } | |
b4501295 | 2188 | |
deb7aa30 TH |
2189 | /* synchronize mems_allowed to N_MEMORY */ |
2190 | if (mems_updated) { | |
deb7aa30 TH |
2191 | mutex_lock(&callback_mutex); |
2192 | top_cpuset.mems_allowed = new_mems; | |
1344ab9c | 2193 | top_cpuset.effective_mems = new_mems; |
deb7aa30 | 2194 | mutex_unlock(&callback_mutex); |
d66393e5 | 2195 | update_tasks_nodemask(&top_cpuset); |
deb7aa30 | 2196 | } |
b4501295 | 2197 | |
388afd85 LZ |
2198 | mutex_unlock(&cpuset_mutex); |
2199 | ||
5c5cc623 LZ |
2200 | /* if cpus or mems changed, we need to propagate to descendants */ |
2201 | if (cpus_updated || mems_updated) { | |
deb7aa30 | 2202 | struct cpuset *cs; |
492eb21b | 2203 | struct cgroup_subsys_state *pos_css; |
f9b4fb8d | 2204 | |
fc560a26 | 2205 | rcu_read_lock(); |
492eb21b | 2206 | cpuset_for_each_descendant_pre(cs, pos_css, &top_cpuset) { |
ec903c0c | 2207 | if (cs == &top_cpuset || !css_tryget_online(&cs->css)) |
388afd85 LZ |
2208 | continue; |
2209 | rcu_read_unlock(); | |
7ddf96b0 | 2210 | |
388afd85 | 2211 | cpuset_hotplug_update_tasks(cs); |
b4501295 | 2212 | |
388afd85 LZ |
2213 | rcu_read_lock(); |
2214 | css_put(&cs->css); | |
2215 | } | |
2216 | rcu_read_unlock(); | |
2217 | } | |
8d033948 | 2218 | |
deb7aa30 | 2219 | /* rebuild sched domains if cpus_allowed has changed */ |
e0e80a02 LZ |
2220 | if (cpus_updated) |
2221 | rebuild_sched_domains(); | |
b1aac8bb PJ |
2222 | } |
2223 | ||
7ddf96b0 | 2224 | void cpuset_update_active_cpus(bool cpu_online) |
4c4d50f7 | 2225 | { |
3a5a6d0c TH |
2226 | /* |
2227 | * We're inside cpu hotplug critical region which usually nests | |
2228 | * inside cgroup synchronization. Bounce actual hotplug processing | |
2229 | * to a work item to avoid reverse locking order. | |
2230 | * | |
2231 | * We still need to do partition_sched_domains() synchronously; | |
2232 | * otherwise, the scheduler will get confused and put tasks to the | |
2233 | * dead CPU. Fall back to the default single domain. | |
2234 | * cpuset_hotplug_workfn() will rebuild it as necessary. | |
2235 | */ | |
2236 | partition_sched_domains(1, NULL, NULL); | |
2237 | schedule_work(&cpuset_hotplug_work); | |
4c4d50f7 | 2238 | } |
4c4d50f7 | 2239 | |
38837fc7 | 2240 | /* |
38d7bee9 LJ |
2241 | * Keep top_cpuset.mems_allowed tracking node_states[N_MEMORY]. |
2242 | * Call this routine anytime after node_states[N_MEMORY] changes. | |
a1cd2b13 | 2243 | * See cpuset_update_active_cpus() for CPU hotplug handling. |
38837fc7 | 2244 | */ |
f481891f MX |
2245 | static int cpuset_track_online_nodes(struct notifier_block *self, |
2246 | unsigned long action, void *arg) | |
38837fc7 | 2247 | { |
3a5a6d0c | 2248 | schedule_work(&cpuset_hotplug_work); |
f481891f | 2249 | return NOTIFY_OK; |
38837fc7 | 2250 | } |
d8f10cb3 AM |
2251 | |
2252 | static struct notifier_block cpuset_track_online_nodes_nb = { | |
2253 | .notifier_call = cpuset_track_online_nodes, | |
2254 | .priority = 10, /* ??! */ | |
2255 | }; | |
38837fc7 | 2256 | |
1da177e4 LT |
2257 | /** |
2258 | * cpuset_init_smp - initialize cpus_allowed | |
2259 | * | |
2260 | * Description: Finish top cpuset after cpu, node maps are initialized | |
d8f10cb3 | 2261 | */ |
1da177e4 LT |
2262 | void __init cpuset_init_smp(void) |
2263 | { | |
6ad4c188 | 2264 | cpumask_copy(top_cpuset.cpus_allowed, cpu_active_mask); |
38d7bee9 | 2265 | top_cpuset.mems_allowed = node_states[N_MEMORY]; |
33ad801d | 2266 | top_cpuset.old_mems_allowed = top_cpuset.mems_allowed; |
4c4d50f7 | 2267 | |
e2b9a3d7 LZ |
2268 | cpumask_copy(top_cpuset.effective_cpus, cpu_active_mask); |
2269 | top_cpuset.effective_mems = node_states[N_MEMORY]; | |
2270 | ||
d8f10cb3 | 2271 | register_hotmemory_notifier(&cpuset_track_online_nodes_nb); |
1da177e4 LT |
2272 | } |
2273 | ||
2274 | /** | |
1da177e4 LT |
2275 | * cpuset_cpus_allowed - return cpus_allowed mask from a tasks cpuset. |
2276 | * @tsk: pointer to task_struct from which to obtain cpuset->cpus_allowed. | |
6af866af | 2277 | * @pmask: pointer to struct cpumask variable to receive cpus_allowed set. |
1da177e4 | 2278 | * |
300ed6cb | 2279 | * Description: Returns the cpumask_var_t cpus_allowed of the cpuset |
1da177e4 | 2280 | * attached to the specified @tsk. Guaranteed to return some non-empty |
5f054e31 | 2281 | * subset of cpu_online_mask, even if this means going outside the |
1da177e4 LT |
2282 | * tasks cpuset. |
2283 | **/ | |
2284 | ||
6af866af | 2285 | void cpuset_cpus_allowed(struct task_struct *tsk, struct cpumask *pmask) |
1da177e4 | 2286 | { |
070b57fc LZ |
2287 | struct cpuset *cpus_cs; |
2288 | ||
3d3f26a7 | 2289 | mutex_lock(&callback_mutex); |
b8dadcb5 | 2290 | rcu_read_lock(); |
070b57fc LZ |
2291 | cpus_cs = effective_cpumask_cpuset(task_cs(tsk)); |
2292 | guarantee_online_cpus(cpus_cs, pmask); | |
b8dadcb5 | 2293 | rcu_read_unlock(); |
897f0b3c | 2294 | mutex_unlock(&callback_mutex); |
1da177e4 LT |
2295 | } |
2296 | ||
2baab4e9 | 2297 | void cpuset_cpus_allowed_fallback(struct task_struct *tsk) |
9084bb82 | 2298 | { |
c9710d80 | 2299 | struct cpuset *cpus_cs; |
9084bb82 ON |
2300 | |
2301 | rcu_read_lock(); | |
070b57fc LZ |
2302 | cpus_cs = effective_cpumask_cpuset(task_cs(tsk)); |
2303 | do_set_cpus_allowed(tsk, cpus_cs->cpus_allowed); | |
9084bb82 ON |
2304 | rcu_read_unlock(); |
2305 | ||
2306 | /* | |
2307 | * We own tsk->cpus_allowed, nobody can change it under us. | |
2308 | * | |
2309 | * But we used cs && cs->cpus_allowed lockless and thus can | |
2310 | * race with cgroup_attach_task() or update_cpumask() and get | |
2311 | * the wrong tsk->cpus_allowed. However, both cases imply the | |
2312 | * subsequent cpuset_change_cpumask()->set_cpus_allowed_ptr() | |
2313 | * which takes task_rq_lock(). | |
2314 | * | |
2315 | * If we are called after it dropped the lock we must see all | |
2316 | * changes in tsk_cs()->cpus_allowed. Otherwise we can temporary | |
2317 | * set any mask even if it is not right from task_cs() pov, | |
2318 | * the pending set_cpus_allowed_ptr() will fix things. | |
2baab4e9 PZ |
2319 | * |
2320 | * select_fallback_rq() will fix things ups and set cpu_possible_mask | |
2321 | * if required. | |
9084bb82 | 2322 | */ |
9084bb82 ON |
2323 | } |
2324 | ||
1da177e4 LT |
2325 | void cpuset_init_current_mems_allowed(void) |
2326 | { | |
f9a86fcb | 2327 | nodes_setall(current->mems_allowed); |
1da177e4 LT |
2328 | } |
2329 | ||
909d75a3 PJ |
2330 | /** |
2331 | * cpuset_mems_allowed - return mems_allowed mask from a tasks cpuset. | |
2332 | * @tsk: pointer to task_struct from which to obtain cpuset->mems_allowed. | |
2333 | * | |
2334 | * Description: Returns the nodemask_t mems_allowed of the cpuset | |
2335 | * attached to the specified @tsk. Guaranteed to return some non-empty | |
38d7bee9 | 2336 | * subset of node_states[N_MEMORY], even if this means going outside the |
909d75a3 PJ |
2337 | * tasks cpuset. |
2338 | **/ | |
2339 | ||
2340 | nodemask_t cpuset_mems_allowed(struct task_struct *tsk) | |
2341 | { | |
070b57fc | 2342 | struct cpuset *mems_cs; |
909d75a3 PJ |
2343 | nodemask_t mask; |
2344 | ||
3d3f26a7 | 2345 | mutex_lock(&callback_mutex); |
b8dadcb5 | 2346 | rcu_read_lock(); |
070b57fc LZ |
2347 | mems_cs = effective_nodemask_cpuset(task_cs(tsk)); |
2348 | guarantee_online_mems(mems_cs, &mask); | |
b8dadcb5 | 2349 | rcu_read_unlock(); |
3d3f26a7 | 2350 | mutex_unlock(&callback_mutex); |
909d75a3 PJ |
2351 | |
2352 | return mask; | |
2353 | } | |
2354 | ||
d9fd8a6d | 2355 | /** |
19770b32 MG |
2356 | * cpuset_nodemask_valid_mems_allowed - check nodemask vs. curremt mems_allowed |
2357 | * @nodemask: the nodemask to be checked | |
d9fd8a6d | 2358 | * |
19770b32 | 2359 | * Are any of the nodes in the nodemask allowed in current->mems_allowed? |
1da177e4 | 2360 | */ |
19770b32 | 2361 | int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask) |
1da177e4 | 2362 | { |
19770b32 | 2363 | return nodes_intersects(*nodemask, current->mems_allowed); |
1da177e4 LT |
2364 | } |
2365 | ||
9bf2229f | 2366 | /* |
78608366 PM |
2367 | * nearest_hardwall_ancestor() - Returns the nearest mem_exclusive or |
2368 | * mem_hardwall ancestor to the specified cpuset. Call holding | |
2369 | * callback_mutex. If no ancestor is mem_exclusive or mem_hardwall | |
2370 | * (an unusual configuration), then returns the root cpuset. | |
9bf2229f | 2371 | */ |
c9710d80 | 2372 | static struct cpuset *nearest_hardwall_ancestor(struct cpuset *cs) |
9bf2229f | 2373 | { |
c431069f TH |
2374 | while (!(is_mem_exclusive(cs) || is_mem_hardwall(cs)) && parent_cs(cs)) |
2375 | cs = parent_cs(cs); | |
9bf2229f PJ |
2376 | return cs; |
2377 | } | |
2378 | ||
d9fd8a6d | 2379 | /** |
a1bc5a4e DR |
2380 | * cpuset_node_allowed_softwall - Can we allocate on a memory node? |
2381 | * @node: is this an allowed node? | |
02a0e53d | 2382 | * @gfp_mask: memory allocation flags |
d9fd8a6d | 2383 | * |
a1bc5a4e DR |
2384 | * If we're in interrupt, yes, we can always allocate. If __GFP_THISNODE is |
2385 | * set, yes, we can always allocate. If node is in our task's mems_allowed, | |
2386 | * yes. If it's not a __GFP_HARDWALL request and this node is in the nearest | |
2387 | * hardwalled cpuset ancestor to this task's cpuset, yes. If the task has been | |
2388 | * OOM killed and has access to memory reserves as specified by the TIF_MEMDIE | |
2389 | * flag, yes. | |
9bf2229f PJ |
2390 | * Otherwise, no. |
2391 | * | |
a1bc5a4e DR |
2392 | * If __GFP_HARDWALL is set, cpuset_node_allowed_softwall() reduces to |
2393 | * cpuset_node_allowed_hardwall(). Otherwise, cpuset_node_allowed_softwall() | |
2394 | * might sleep, and might allow a node from an enclosing cpuset. | |
02a0e53d | 2395 | * |
a1bc5a4e DR |
2396 | * cpuset_node_allowed_hardwall() only handles the simpler case of hardwall |
2397 | * cpusets, and never sleeps. | |
02a0e53d PJ |
2398 | * |
2399 | * The __GFP_THISNODE placement logic is really handled elsewhere, | |
2400 | * by forcibly using a zonelist starting at a specified node, and by | |
2401 | * (in get_page_from_freelist()) refusing to consider the zones for | |
2402 | * any node on the zonelist except the first. By the time any such | |
2403 | * calls get to this routine, we should just shut up and say 'yes'. | |
2404 | * | |
9bf2229f | 2405 | * GFP_USER allocations are marked with the __GFP_HARDWALL bit, |
c596d9f3 DR |
2406 | * and do not allow allocations outside the current tasks cpuset |
2407 | * unless the task has been OOM killed as is marked TIF_MEMDIE. | |
9bf2229f | 2408 | * GFP_KERNEL allocations are not so marked, so can escape to the |
78608366 | 2409 | * nearest enclosing hardwalled ancestor cpuset. |
9bf2229f | 2410 | * |
02a0e53d PJ |
2411 | * Scanning up parent cpusets requires callback_mutex. The |
2412 | * __alloc_pages() routine only calls here with __GFP_HARDWALL bit | |
2413 | * _not_ set if it's a GFP_KERNEL allocation, and all nodes in the | |
2414 | * current tasks mems_allowed came up empty on the first pass over | |
2415 | * the zonelist. So only GFP_KERNEL allocations, if all nodes in the | |
2416 | * cpuset are short of memory, might require taking the callback_mutex | |
2417 | * mutex. | |
9bf2229f | 2418 | * |
36be57ff | 2419 | * The first call here from mm/page_alloc:get_page_from_freelist() |
02a0e53d PJ |
2420 | * has __GFP_HARDWALL set in gfp_mask, enforcing hardwall cpusets, |
2421 | * so no allocation on a node outside the cpuset is allowed (unless | |
2422 | * in interrupt, of course). | |
36be57ff PJ |
2423 | * |
2424 | * The second pass through get_page_from_freelist() doesn't even call | |
2425 | * here for GFP_ATOMIC calls. For those calls, the __alloc_pages() | |
2426 | * variable 'wait' is not set, and the bit ALLOC_CPUSET is not set | |
2427 | * in alloc_flags. That logic and the checks below have the combined | |
2428 | * affect that: | |
9bf2229f PJ |
2429 | * in_interrupt - any node ok (current task context irrelevant) |
2430 | * GFP_ATOMIC - any node ok | |
c596d9f3 | 2431 | * TIF_MEMDIE - any node ok |
78608366 | 2432 | * GFP_KERNEL - any node in enclosing hardwalled cpuset ok |
9bf2229f | 2433 | * GFP_USER - only nodes in current tasks mems allowed ok. |
36be57ff PJ |
2434 | * |
2435 | * Rule: | |
a1bc5a4e | 2436 | * Don't call cpuset_node_allowed_softwall if you can't sleep, unless you |
36be57ff PJ |
2437 | * pass in the __GFP_HARDWALL flag set in gfp_flag, which disables |
2438 | * the code that might scan up ancestor cpusets and sleep. | |
02a0e53d | 2439 | */ |
a1bc5a4e | 2440 | int __cpuset_node_allowed_softwall(int node, gfp_t gfp_mask) |
1da177e4 | 2441 | { |
c9710d80 | 2442 | struct cpuset *cs; /* current cpuset ancestors */ |
29afd49b | 2443 | int allowed; /* is allocation in zone z allowed? */ |
9bf2229f | 2444 | |
9b819d20 | 2445 | if (in_interrupt() || (gfp_mask & __GFP_THISNODE)) |
9bf2229f | 2446 | return 1; |
92d1dbd2 | 2447 | might_sleep_if(!(gfp_mask & __GFP_HARDWALL)); |
9bf2229f PJ |
2448 | if (node_isset(node, current->mems_allowed)) |
2449 | return 1; | |
c596d9f3 DR |
2450 | /* |
2451 | * Allow tasks that have access to memory reserves because they have | |
2452 | * been OOM killed to get memory anywhere. | |
2453 | */ | |
2454 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
2455 | return 1; | |
9bf2229f PJ |
2456 | if (gfp_mask & __GFP_HARDWALL) /* If hardwall request, stop here */ |
2457 | return 0; | |
2458 | ||
5563e770 BP |
2459 | if (current->flags & PF_EXITING) /* Let dying task have memory */ |
2460 | return 1; | |
2461 | ||
9bf2229f | 2462 | /* Not hardwall and node outside mems_allowed: scan up cpusets */ |
3d3f26a7 | 2463 | mutex_lock(&callback_mutex); |
053199ed | 2464 | |
b8dadcb5 | 2465 | rcu_read_lock(); |
78608366 | 2466 | cs = nearest_hardwall_ancestor(task_cs(current)); |
99afb0fd | 2467 | allowed = node_isset(node, cs->mems_allowed); |
b8dadcb5 | 2468 | rcu_read_unlock(); |
053199ed | 2469 | |
3d3f26a7 | 2470 | mutex_unlock(&callback_mutex); |
9bf2229f | 2471 | return allowed; |
1da177e4 LT |
2472 | } |
2473 | ||
02a0e53d | 2474 | /* |
a1bc5a4e DR |
2475 | * cpuset_node_allowed_hardwall - Can we allocate on a memory node? |
2476 | * @node: is this an allowed node? | |
02a0e53d PJ |
2477 | * @gfp_mask: memory allocation flags |
2478 | * | |
a1bc5a4e DR |
2479 | * If we're in interrupt, yes, we can always allocate. If __GFP_THISNODE is |
2480 | * set, yes, we can always allocate. If node is in our task's mems_allowed, | |
2481 | * yes. If the task has been OOM killed and has access to memory reserves as | |
2482 | * specified by the TIF_MEMDIE flag, yes. | |
2483 | * Otherwise, no. | |
02a0e53d PJ |
2484 | * |
2485 | * The __GFP_THISNODE placement logic is really handled elsewhere, | |
2486 | * by forcibly using a zonelist starting at a specified node, and by | |
2487 | * (in get_page_from_freelist()) refusing to consider the zones for | |
2488 | * any node on the zonelist except the first. By the time any such | |
2489 | * calls get to this routine, we should just shut up and say 'yes'. | |
2490 | * | |
a1bc5a4e DR |
2491 | * Unlike the cpuset_node_allowed_softwall() variant, above, |
2492 | * this variant requires that the node be in the current task's | |
02a0e53d PJ |
2493 | * mems_allowed or that we're in interrupt. It does not scan up the |
2494 | * cpuset hierarchy for the nearest enclosing mem_exclusive cpuset. | |
2495 | * It never sleeps. | |
2496 | */ | |
a1bc5a4e | 2497 | int __cpuset_node_allowed_hardwall(int node, gfp_t gfp_mask) |
02a0e53d | 2498 | { |
02a0e53d PJ |
2499 | if (in_interrupt() || (gfp_mask & __GFP_THISNODE)) |
2500 | return 1; | |
02a0e53d PJ |
2501 | if (node_isset(node, current->mems_allowed)) |
2502 | return 1; | |
dedf8b79 DW |
2503 | /* |
2504 | * Allow tasks that have access to memory reserves because they have | |
2505 | * been OOM killed to get memory anywhere. | |
2506 | */ | |
2507 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
2508 | return 1; | |
02a0e53d PJ |
2509 | return 0; |
2510 | } | |
2511 | ||
825a46af | 2512 | /** |
6adef3eb JS |
2513 | * cpuset_mem_spread_node() - On which node to begin search for a file page |
2514 | * cpuset_slab_spread_node() - On which node to begin search for a slab page | |
825a46af PJ |
2515 | * |
2516 | * If a task is marked PF_SPREAD_PAGE or PF_SPREAD_SLAB (as for | |
2517 | * tasks in a cpuset with is_spread_page or is_spread_slab set), | |
2518 | * and if the memory allocation used cpuset_mem_spread_node() | |
2519 | * to determine on which node to start looking, as it will for | |
2520 | * certain page cache or slab cache pages such as used for file | |
2521 | * system buffers and inode caches, then instead of starting on the | |
2522 | * local node to look for a free page, rather spread the starting | |
2523 | * node around the tasks mems_allowed nodes. | |
2524 | * | |
2525 | * We don't have to worry about the returned node being offline | |
2526 | * because "it can't happen", and even if it did, it would be ok. | |
2527 | * | |
2528 | * The routines calling guarantee_online_mems() are careful to | |
2529 | * only set nodes in task->mems_allowed that are online. So it | |
2530 | * should not be possible for the following code to return an | |
2531 | * offline node. But if it did, that would be ok, as this routine | |
2532 | * is not returning the node where the allocation must be, only | |
2533 | * the node where the search should start. The zonelist passed to | |
2534 | * __alloc_pages() will include all nodes. If the slab allocator | |
2535 | * is passed an offline node, it will fall back to the local node. | |
2536 | * See kmem_cache_alloc_node(). | |
2537 | */ | |
2538 | ||
6adef3eb | 2539 | static int cpuset_spread_node(int *rotor) |
825a46af PJ |
2540 | { |
2541 | int node; | |
2542 | ||
6adef3eb | 2543 | node = next_node(*rotor, current->mems_allowed); |
825a46af PJ |
2544 | if (node == MAX_NUMNODES) |
2545 | node = first_node(current->mems_allowed); | |
6adef3eb | 2546 | *rotor = node; |
825a46af PJ |
2547 | return node; |
2548 | } | |
6adef3eb JS |
2549 | |
2550 | int cpuset_mem_spread_node(void) | |
2551 | { | |
778d3b0f MH |
2552 | if (current->cpuset_mem_spread_rotor == NUMA_NO_NODE) |
2553 | current->cpuset_mem_spread_rotor = | |
2554 | node_random(¤t->mems_allowed); | |
2555 | ||
6adef3eb JS |
2556 | return cpuset_spread_node(¤t->cpuset_mem_spread_rotor); |
2557 | } | |
2558 | ||
2559 | int cpuset_slab_spread_node(void) | |
2560 | { | |
778d3b0f MH |
2561 | if (current->cpuset_slab_spread_rotor == NUMA_NO_NODE) |
2562 | current->cpuset_slab_spread_rotor = | |
2563 | node_random(¤t->mems_allowed); | |
2564 | ||
6adef3eb JS |
2565 | return cpuset_spread_node(¤t->cpuset_slab_spread_rotor); |
2566 | } | |
2567 | ||
825a46af PJ |
2568 | EXPORT_SYMBOL_GPL(cpuset_mem_spread_node); |
2569 | ||
ef08e3b4 | 2570 | /** |
bbe373f2 DR |
2571 | * cpuset_mems_allowed_intersects - Does @tsk1's mems_allowed intersect @tsk2's? |
2572 | * @tsk1: pointer to task_struct of some task. | |
2573 | * @tsk2: pointer to task_struct of some other task. | |
2574 | * | |
2575 | * Description: Return true if @tsk1's mems_allowed intersects the | |
2576 | * mems_allowed of @tsk2. Used by the OOM killer to determine if | |
2577 | * one of the task's memory usage might impact the memory available | |
2578 | * to the other. | |
ef08e3b4 PJ |
2579 | **/ |
2580 | ||
bbe373f2 DR |
2581 | int cpuset_mems_allowed_intersects(const struct task_struct *tsk1, |
2582 | const struct task_struct *tsk2) | |
ef08e3b4 | 2583 | { |
bbe373f2 | 2584 | return nodes_intersects(tsk1->mems_allowed, tsk2->mems_allowed); |
ef08e3b4 PJ |
2585 | } |
2586 | ||
f440d98f LZ |
2587 | #define CPUSET_NODELIST_LEN (256) |
2588 | ||
75aa1994 DR |
2589 | /** |
2590 | * cpuset_print_task_mems_allowed - prints task's cpuset and mems_allowed | |
fc34ac1d | 2591 | * @tsk: pointer to task_struct of some task. |
75aa1994 DR |
2592 | * |
2593 | * Description: Prints @task's name, cpuset name, and cached copy of its | |
b8dadcb5 | 2594 | * mems_allowed to the kernel log. |
75aa1994 DR |
2595 | */ |
2596 | void cpuset_print_task_mems_allowed(struct task_struct *tsk) | |
2597 | { | |
f440d98f LZ |
2598 | /* Statically allocated to prevent using excess stack. */ |
2599 | static char cpuset_nodelist[CPUSET_NODELIST_LEN]; | |
2600 | static DEFINE_SPINLOCK(cpuset_buffer_lock); | |
b8dadcb5 | 2601 | struct cgroup *cgrp; |
75aa1994 | 2602 | |
f440d98f | 2603 | spin_lock(&cpuset_buffer_lock); |
b8dadcb5 | 2604 | rcu_read_lock(); |
63f43f55 | 2605 | |
b8dadcb5 | 2606 | cgrp = task_cs(tsk)->css.cgroup; |
75aa1994 DR |
2607 | nodelist_scnprintf(cpuset_nodelist, CPUSET_NODELIST_LEN, |
2608 | tsk->mems_allowed); | |
12d3089c | 2609 | pr_info("%s cpuset=", tsk->comm); |
e61734c5 TH |
2610 | pr_cont_cgroup_name(cgrp); |
2611 | pr_cont(" mems_allowed=%s\n", cpuset_nodelist); | |
f440d98f | 2612 | |
cfb5966b | 2613 | rcu_read_unlock(); |
75aa1994 DR |
2614 | spin_unlock(&cpuset_buffer_lock); |
2615 | } | |
2616 | ||
3e0d98b9 PJ |
2617 | /* |
2618 | * Collection of memory_pressure is suppressed unless | |
2619 | * this flag is enabled by writing "1" to the special | |
2620 | * cpuset file 'memory_pressure_enabled' in the root cpuset. | |
2621 | */ | |
2622 | ||
c5b2aff8 | 2623 | int cpuset_memory_pressure_enabled __read_mostly; |
3e0d98b9 PJ |
2624 | |
2625 | /** | |
2626 | * cpuset_memory_pressure_bump - keep stats of per-cpuset reclaims. | |
2627 | * | |
2628 | * Keep a running average of the rate of synchronous (direct) | |
2629 | * page reclaim efforts initiated by tasks in each cpuset. | |
2630 | * | |
2631 | * This represents the rate at which some task in the cpuset | |
2632 | * ran low on memory on all nodes it was allowed to use, and | |
2633 | * had to enter the kernels page reclaim code in an effort to | |
2634 | * create more free memory by tossing clean pages or swapping | |
2635 | * or writing dirty pages. | |
2636 | * | |
2637 | * Display to user space in the per-cpuset read-only file | |
2638 | * "memory_pressure". Value displayed is an integer | |
2639 | * representing the recent rate of entry into the synchronous | |
2640 | * (direct) page reclaim by any task attached to the cpuset. | |
2641 | **/ | |
2642 | ||
2643 | void __cpuset_memory_pressure_bump(void) | |
2644 | { | |
b8dadcb5 | 2645 | rcu_read_lock(); |
8793d854 | 2646 | fmeter_markevent(&task_cs(current)->fmeter); |
b8dadcb5 | 2647 | rcu_read_unlock(); |
3e0d98b9 PJ |
2648 | } |
2649 | ||
8793d854 | 2650 | #ifdef CONFIG_PROC_PID_CPUSET |
1da177e4 LT |
2651 | /* |
2652 | * proc_cpuset_show() | |
2653 | * - Print tasks cpuset path into seq_file. | |
2654 | * - Used for /proc/<pid>/cpuset. | |
053199ed PJ |
2655 | * - No need to task_lock(tsk) on this tsk->cpuset reference, as it |
2656 | * doesn't really matter if tsk->cpuset changes after we read it, | |
5d21cc2d | 2657 | * and we take cpuset_mutex, keeping cpuset_attach() from changing it |
2df167a3 | 2658 | * anyway. |
1da177e4 | 2659 | */ |
8d8b97ba | 2660 | int proc_cpuset_show(struct seq_file *m, void *unused_v) |
1da177e4 | 2661 | { |
13b41b09 | 2662 | struct pid *pid; |
1da177e4 | 2663 | struct task_struct *tsk; |
e61734c5 | 2664 | char *buf, *p; |
8793d854 | 2665 | struct cgroup_subsys_state *css; |
99f89551 | 2666 | int retval; |
1da177e4 | 2667 | |
99f89551 | 2668 | retval = -ENOMEM; |
e61734c5 | 2669 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
1da177e4 | 2670 | if (!buf) |
99f89551 EB |
2671 | goto out; |
2672 | ||
2673 | retval = -ESRCH; | |
13b41b09 EB |
2674 | pid = m->private; |
2675 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
99f89551 EB |
2676 | if (!tsk) |
2677 | goto out_free; | |
1da177e4 | 2678 | |
e61734c5 | 2679 | retval = -ENAMETOOLONG; |
27e89ae5 | 2680 | rcu_read_lock(); |
073219e9 | 2681 | css = task_css(tsk, cpuset_cgrp_id); |
e61734c5 | 2682 | p = cgroup_path(css->cgroup, buf, PATH_MAX); |
27e89ae5 | 2683 | rcu_read_unlock(); |
e61734c5 | 2684 | if (!p) |
27e89ae5 | 2685 | goto out_put_task; |
e61734c5 | 2686 | seq_puts(m, p); |
1da177e4 | 2687 | seq_putc(m, '\n'); |
e61734c5 | 2688 | retval = 0; |
27e89ae5 | 2689 | out_put_task: |
99f89551 EB |
2690 | put_task_struct(tsk); |
2691 | out_free: | |
1da177e4 | 2692 | kfree(buf); |
99f89551 | 2693 | out: |
1da177e4 LT |
2694 | return retval; |
2695 | } | |
8793d854 | 2696 | #endif /* CONFIG_PROC_PID_CPUSET */ |
1da177e4 | 2697 | |
d01d4827 | 2698 | /* Display task mems_allowed in /proc/<pid>/status file. */ |
df5f8314 EB |
2699 | void cpuset_task_status_allowed(struct seq_file *m, struct task_struct *task) |
2700 | { | |
fc34ac1d | 2701 | seq_puts(m, "Mems_allowed:\t"); |
30e8e136 | 2702 | seq_nodemask(m, &task->mems_allowed); |
fc34ac1d FF |
2703 | seq_puts(m, "\n"); |
2704 | seq_puts(m, "Mems_allowed_list:\t"); | |
30e8e136 | 2705 | seq_nodemask_list(m, &task->mems_allowed); |
fc34ac1d | 2706 | seq_puts(m, "\n"); |
1da177e4 | 2707 | } |