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 | |
202f72d5 PJ |
64 | /* |
65 | * Tracks how many cpusets are currently defined in system. | |
66 | * When there is only one cpuset (the root cpuset) we can | |
67 | * short circuit some hooks. | |
68 | */ | |
7edc5962 | 69 | int number_of_cpusets __read_mostly; |
202f72d5 | 70 | |
2df167a3 | 71 | /* Forward declare cgroup structures */ |
8793d854 | 72 | struct cgroup_subsys cpuset_subsys; |
8793d854 | 73 | |
3e0d98b9 PJ |
74 | /* See "Frequency meter" comments, below. */ |
75 | ||
76 | struct fmeter { | |
77 | int cnt; /* unprocessed events count */ | |
78 | int val; /* most recent output value */ | |
79 | time_t time; /* clock (secs) when val computed */ | |
80 | spinlock_t lock; /* guards read or write of above */ | |
81 | }; | |
82 | ||
1da177e4 | 83 | struct cpuset { |
8793d854 PM |
84 | struct cgroup_subsys_state css; |
85 | ||
1da177e4 | 86 | unsigned long flags; /* "unsigned long" so bitops work */ |
300ed6cb | 87 | cpumask_var_t cpus_allowed; /* CPUs allowed to tasks in cpuset */ |
1da177e4 LT |
88 | nodemask_t mems_allowed; /* Memory Nodes allowed to tasks */ |
89 | ||
33ad801d LZ |
90 | /* |
91 | * This is old Memory Nodes tasks took on. | |
92 | * | |
93 | * - top_cpuset.old_mems_allowed is initialized to mems_allowed. | |
94 | * - A new cpuset's old_mems_allowed is initialized when some | |
95 | * task is moved into it. | |
96 | * - old_mems_allowed is used in cpuset_migrate_mm() when we change | |
97 | * cpuset.mems_allowed and have tasks' nodemask updated, and | |
98 | * then old_mems_allowed is updated to mems_allowed. | |
99 | */ | |
100 | nodemask_t old_mems_allowed; | |
101 | ||
3e0d98b9 | 102 | struct fmeter fmeter; /* memory_pressure filter */ |
029190c5 | 103 | |
452477fa TH |
104 | /* |
105 | * Tasks are being attached to this cpuset. Used to prevent | |
106 | * zeroing cpus/mems_allowed between ->can_attach() and ->attach(). | |
107 | */ | |
108 | int attach_in_progress; | |
109 | ||
029190c5 PJ |
110 | /* partition number for rebuild_sched_domains() */ |
111 | int pn; | |
956db3ca | 112 | |
1d3504fc HS |
113 | /* for custom sched domain */ |
114 | int relax_domain_level; | |
1da177e4 LT |
115 | }; |
116 | ||
8793d854 | 117 | /* Retrieve the cpuset for a cgroup */ |
c9e5fe66 | 118 | static inline struct cpuset *cgroup_cs(struct cgroup *cgrp) |
8793d854 | 119 | { |
8af01f56 | 120 | return container_of(cgroup_css(cgrp, cpuset_subsys_id), |
8793d854 PM |
121 | struct cpuset, css); |
122 | } | |
123 | ||
124 | /* Retrieve the cpuset for a task */ | |
125 | static inline struct cpuset *task_cs(struct task_struct *task) | |
126 | { | |
8af01f56 | 127 | return container_of(task_css(task, cpuset_subsys_id), |
8793d854 PM |
128 | struct cpuset, css); |
129 | } | |
8793d854 | 130 | |
c431069f TH |
131 | static inline struct cpuset *parent_cs(const struct cpuset *cs) |
132 | { | |
133 | struct cgroup *pcgrp = cs->css.cgroup->parent; | |
134 | ||
135 | if (pcgrp) | |
136 | return cgroup_cs(pcgrp); | |
137 | return NULL; | |
138 | } | |
139 | ||
b246272e DR |
140 | #ifdef CONFIG_NUMA |
141 | static inline bool task_has_mempolicy(struct task_struct *task) | |
142 | { | |
143 | return task->mempolicy; | |
144 | } | |
145 | #else | |
146 | static inline bool task_has_mempolicy(struct task_struct *task) | |
147 | { | |
148 | return false; | |
149 | } | |
150 | #endif | |
151 | ||
152 | ||
1da177e4 LT |
153 | /* bits in struct cpuset flags field */ |
154 | typedef enum { | |
efeb77b2 | 155 | CS_ONLINE, |
1da177e4 LT |
156 | CS_CPU_EXCLUSIVE, |
157 | CS_MEM_EXCLUSIVE, | |
78608366 | 158 | CS_MEM_HARDWALL, |
45b07ef3 | 159 | CS_MEMORY_MIGRATE, |
029190c5 | 160 | CS_SCHED_LOAD_BALANCE, |
825a46af PJ |
161 | CS_SPREAD_PAGE, |
162 | CS_SPREAD_SLAB, | |
1da177e4 LT |
163 | } cpuset_flagbits_t; |
164 | ||
165 | /* convenient tests for these bits */ | |
efeb77b2 TH |
166 | static inline bool is_cpuset_online(const struct cpuset *cs) |
167 | { | |
168 | return test_bit(CS_ONLINE, &cs->flags); | |
169 | } | |
170 | ||
1da177e4 LT |
171 | static inline int is_cpu_exclusive(const struct cpuset *cs) |
172 | { | |
7b5b9ef0 | 173 | return test_bit(CS_CPU_EXCLUSIVE, &cs->flags); |
1da177e4 LT |
174 | } |
175 | ||
176 | static inline int is_mem_exclusive(const struct cpuset *cs) | |
177 | { | |
7b5b9ef0 | 178 | return test_bit(CS_MEM_EXCLUSIVE, &cs->flags); |
1da177e4 LT |
179 | } |
180 | ||
78608366 PM |
181 | static inline int is_mem_hardwall(const struct cpuset *cs) |
182 | { | |
183 | return test_bit(CS_MEM_HARDWALL, &cs->flags); | |
184 | } | |
185 | ||
029190c5 PJ |
186 | static inline int is_sched_load_balance(const struct cpuset *cs) |
187 | { | |
188 | return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); | |
189 | } | |
190 | ||
45b07ef3 PJ |
191 | static inline int is_memory_migrate(const struct cpuset *cs) |
192 | { | |
7b5b9ef0 | 193 | return test_bit(CS_MEMORY_MIGRATE, &cs->flags); |
45b07ef3 PJ |
194 | } |
195 | ||
825a46af PJ |
196 | static inline int is_spread_page(const struct cpuset *cs) |
197 | { | |
198 | return test_bit(CS_SPREAD_PAGE, &cs->flags); | |
199 | } | |
200 | ||
201 | static inline int is_spread_slab(const struct cpuset *cs) | |
202 | { | |
203 | return test_bit(CS_SPREAD_SLAB, &cs->flags); | |
204 | } | |
205 | ||
1da177e4 | 206 | static struct cpuset top_cpuset = { |
efeb77b2 TH |
207 | .flags = ((1 << CS_ONLINE) | (1 << CS_CPU_EXCLUSIVE) | |
208 | (1 << CS_MEM_EXCLUSIVE)), | |
1da177e4 LT |
209 | }; |
210 | ||
ae8086ce TH |
211 | /** |
212 | * cpuset_for_each_child - traverse online children of a cpuset | |
213 | * @child_cs: loop cursor pointing to the current child | |
214 | * @pos_cgrp: used for iteration | |
215 | * @parent_cs: target cpuset to walk children of | |
216 | * | |
217 | * Walk @child_cs through the online children of @parent_cs. Must be used | |
218 | * with RCU read locked. | |
219 | */ | |
220 | #define cpuset_for_each_child(child_cs, pos_cgrp, parent_cs) \ | |
221 | cgroup_for_each_child((pos_cgrp), (parent_cs)->css.cgroup) \ | |
222 | if (is_cpuset_online(((child_cs) = cgroup_cs((pos_cgrp))))) | |
223 | ||
fc560a26 TH |
224 | /** |
225 | * cpuset_for_each_descendant_pre - pre-order walk of a cpuset's descendants | |
226 | * @des_cs: loop cursor pointing to the current descendant | |
227 | * @pos_cgrp: used for iteration | |
228 | * @root_cs: target cpuset to walk ancestor of | |
229 | * | |
230 | * Walk @des_cs through the online descendants of @root_cs. Must be used | |
231 | * with RCU read locked. The caller may modify @pos_cgrp by calling | |
232 | * cgroup_rightmost_descendant() to skip subtree. | |
233 | */ | |
234 | #define cpuset_for_each_descendant_pre(des_cs, pos_cgrp, root_cs) \ | |
235 | cgroup_for_each_descendant_pre((pos_cgrp), (root_cs)->css.cgroup) \ | |
236 | if (is_cpuset_online(((des_cs) = cgroup_cs((pos_cgrp))))) | |
237 | ||
1da177e4 | 238 | /* |
5d21cc2d TH |
239 | * There are two global mutexes guarding cpuset structures - cpuset_mutex |
240 | * and callback_mutex. The latter may nest inside the former. We also | |
241 | * require taking task_lock() when dereferencing a task's cpuset pointer. | |
242 | * See "The task_lock() exception", at the end of this comment. | |
243 | * | |
244 | * A task must hold both mutexes to modify cpusets. If a task holds | |
245 | * cpuset_mutex, then it blocks others wanting that mutex, ensuring that it | |
246 | * is the only task able to also acquire callback_mutex and be able to | |
247 | * modify cpusets. It can perform various checks on the cpuset structure | |
248 | * first, knowing nothing will change. It can also allocate memory while | |
249 | * just holding cpuset_mutex. While it is performing these checks, various | |
250 | * callback routines can briefly acquire callback_mutex to query cpusets. | |
251 | * Once it is ready to make the changes, it takes callback_mutex, blocking | |
252 | * everyone else. | |
053199ed PJ |
253 | * |
254 | * Calls to the kernel memory allocator can not be made while holding | |
3d3f26a7 | 255 | * callback_mutex, as that would risk double tripping on callback_mutex |
053199ed PJ |
256 | * from one of the callbacks into the cpuset code from within |
257 | * __alloc_pages(). | |
258 | * | |
3d3f26a7 | 259 | * If a task is only holding callback_mutex, then it has read-only |
053199ed PJ |
260 | * access to cpusets. |
261 | * | |
58568d2a MX |
262 | * Now, the task_struct fields mems_allowed and mempolicy may be changed |
263 | * by other task, we use alloc_lock in the task_struct fields to protect | |
264 | * them. | |
053199ed | 265 | * |
3d3f26a7 | 266 | * The cpuset_common_file_read() handlers only hold callback_mutex across |
053199ed PJ |
267 | * small pieces of code, such as when reading out possibly multi-word |
268 | * cpumasks and nodemasks. | |
269 | * | |
2df167a3 PM |
270 | * Accessing a task's cpuset should be done in accordance with the |
271 | * guidelines for accessing subsystem state in kernel/cgroup.c | |
1da177e4 LT |
272 | */ |
273 | ||
5d21cc2d | 274 | static DEFINE_MUTEX(cpuset_mutex); |
3d3f26a7 | 275 | static DEFINE_MUTEX(callback_mutex); |
4247bdc6 | 276 | |
3a5a6d0c TH |
277 | /* |
278 | * CPU / memory hotplug is handled asynchronously. | |
279 | */ | |
280 | static void cpuset_hotplug_workfn(struct work_struct *work); | |
3a5a6d0c TH |
281 | static DECLARE_WORK(cpuset_hotplug_work, cpuset_hotplug_workfn); |
282 | ||
e44193d3 LZ |
283 | static DECLARE_WAIT_QUEUE_HEAD(cpuset_attach_wq); |
284 | ||
cf417141 MK |
285 | /* |
286 | * This is ugly, but preserves the userspace API for existing cpuset | |
8793d854 | 287 | * users. If someone tries to mount the "cpuset" filesystem, we |
cf417141 MK |
288 | * silently switch it to mount "cgroup" instead |
289 | */ | |
f7e83571 AV |
290 | static struct dentry *cpuset_mount(struct file_system_type *fs_type, |
291 | int flags, const char *unused_dev_name, void *data) | |
1da177e4 | 292 | { |
8793d854 | 293 | struct file_system_type *cgroup_fs = get_fs_type("cgroup"); |
f7e83571 | 294 | struct dentry *ret = ERR_PTR(-ENODEV); |
8793d854 PM |
295 | if (cgroup_fs) { |
296 | char mountopts[] = | |
297 | "cpuset,noprefix," | |
298 | "release_agent=/sbin/cpuset_release_agent"; | |
f7e83571 AV |
299 | ret = cgroup_fs->mount(cgroup_fs, flags, |
300 | unused_dev_name, mountopts); | |
8793d854 PM |
301 | put_filesystem(cgroup_fs); |
302 | } | |
303 | return ret; | |
1da177e4 LT |
304 | } |
305 | ||
306 | static struct file_system_type cpuset_fs_type = { | |
307 | .name = "cpuset", | |
f7e83571 | 308 | .mount = cpuset_mount, |
1da177e4 LT |
309 | }; |
310 | ||
1da177e4 | 311 | /* |
300ed6cb | 312 | * Return in pmask the portion of a cpusets's cpus_allowed that |
1da177e4 | 313 | * are online. If none are online, walk up the cpuset hierarchy |
40df2deb LZ |
314 | * until we find one that does have some online cpus. The top |
315 | * cpuset always has some cpus online. | |
1da177e4 LT |
316 | * |
317 | * One way or another, we guarantee to return some non-empty subset | |
5f054e31 | 318 | * of cpu_online_mask. |
1da177e4 | 319 | * |
3d3f26a7 | 320 | * Call with callback_mutex held. |
1da177e4 | 321 | */ |
6af866af LZ |
322 | static void guarantee_online_cpus(const struct cpuset *cs, |
323 | struct cpumask *pmask) | |
1da177e4 | 324 | { |
40df2deb | 325 | while (!cpumask_intersects(cs->cpus_allowed, cpu_online_mask)) |
c431069f | 326 | cs = parent_cs(cs); |
40df2deb | 327 | cpumask_and(pmask, cs->cpus_allowed, cpu_online_mask); |
1da177e4 LT |
328 | } |
329 | ||
330 | /* | |
331 | * Return in *pmask the portion of a cpusets's mems_allowed that | |
0e1e7c7a CL |
332 | * are online, with memory. If none are online with memory, walk |
333 | * up the cpuset hierarchy until we find one that does have some | |
40df2deb | 334 | * online mems. The top cpuset always has some mems online. |
1da177e4 LT |
335 | * |
336 | * One way or another, we guarantee to return some non-empty subset | |
38d7bee9 | 337 | * of node_states[N_MEMORY]. |
1da177e4 | 338 | * |
3d3f26a7 | 339 | * Call with callback_mutex held. |
1da177e4 | 340 | */ |
1da177e4 LT |
341 | static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask) |
342 | { | |
40df2deb | 343 | while (!nodes_intersects(cs->mems_allowed, node_states[N_MEMORY])) |
c431069f | 344 | cs = parent_cs(cs); |
40df2deb | 345 | nodes_and(*pmask, cs->mems_allowed, node_states[N_MEMORY]); |
1da177e4 LT |
346 | } |
347 | ||
f3b39d47 MX |
348 | /* |
349 | * update task's spread flag if cpuset's page/slab spread flag is set | |
350 | * | |
5d21cc2d | 351 | * Called with callback_mutex/cpuset_mutex held |
f3b39d47 MX |
352 | */ |
353 | static void cpuset_update_task_spread_flag(struct cpuset *cs, | |
354 | struct task_struct *tsk) | |
355 | { | |
356 | if (is_spread_page(cs)) | |
357 | tsk->flags |= PF_SPREAD_PAGE; | |
358 | else | |
359 | tsk->flags &= ~PF_SPREAD_PAGE; | |
360 | if (is_spread_slab(cs)) | |
361 | tsk->flags |= PF_SPREAD_SLAB; | |
362 | else | |
363 | tsk->flags &= ~PF_SPREAD_SLAB; | |
364 | } | |
365 | ||
1da177e4 LT |
366 | /* |
367 | * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q? | |
368 | * | |
369 | * One cpuset is a subset of another if all its allowed CPUs and | |
370 | * Memory Nodes are a subset of the other, and its exclusive flags | |
5d21cc2d | 371 | * are only set if the other's are set. Call holding cpuset_mutex. |
1da177e4 LT |
372 | */ |
373 | ||
374 | static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q) | |
375 | { | |
300ed6cb | 376 | return cpumask_subset(p->cpus_allowed, q->cpus_allowed) && |
1da177e4 LT |
377 | nodes_subset(p->mems_allowed, q->mems_allowed) && |
378 | is_cpu_exclusive(p) <= is_cpu_exclusive(q) && | |
379 | is_mem_exclusive(p) <= is_mem_exclusive(q); | |
380 | } | |
381 | ||
645fcc9d LZ |
382 | /** |
383 | * alloc_trial_cpuset - allocate a trial cpuset | |
384 | * @cs: the cpuset that the trial cpuset duplicates | |
385 | */ | |
386 | static struct cpuset *alloc_trial_cpuset(const struct cpuset *cs) | |
387 | { | |
300ed6cb LZ |
388 | struct cpuset *trial; |
389 | ||
390 | trial = kmemdup(cs, sizeof(*cs), GFP_KERNEL); | |
391 | if (!trial) | |
392 | return NULL; | |
393 | ||
394 | if (!alloc_cpumask_var(&trial->cpus_allowed, GFP_KERNEL)) { | |
395 | kfree(trial); | |
396 | return NULL; | |
397 | } | |
398 | cpumask_copy(trial->cpus_allowed, cs->cpus_allowed); | |
399 | ||
400 | return trial; | |
645fcc9d LZ |
401 | } |
402 | ||
403 | /** | |
404 | * free_trial_cpuset - free the trial cpuset | |
405 | * @trial: the trial cpuset to be freed | |
406 | */ | |
407 | static void free_trial_cpuset(struct cpuset *trial) | |
408 | { | |
300ed6cb | 409 | free_cpumask_var(trial->cpus_allowed); |
645fcc9d LZ |
410 | kfree(trial); |
411 | } | |
412 | ||
1da177e4 LT |
413 | /* |
414 | * validate_change() - Used to validate that any proposed cpuset change | |
415 | * follows the structural rules for cpusets. | |
416 | * | |
417 | * If we replaced the flag and mask values of the current cpuset | |
418 | * (cur) with those values in the trial cpuset (trial), would | |
419 | * our various subset and exclusive rules still be valid? Presumes | |
5d21cc2d | 420 | * cpuset_mutex held. |
1da177e4 LT |
421 | * |
422 | * 'cur' is the address of an actual, in-use cpuset. Operations | |
423 | * such as list traversal that depend on the actual address of the | |
424 | * cpuset in the list must use cur below, not trial. | |
425 | * | |
426 | * 'trial' is the address of bulk structure copy of cur, with | |
427 | * perhaps one or more of the fields cpus_allowed, mems_allowed, | |
428 | * or flags changed to new, trial values. | |
429 | * | |
430 | * Return 0 if valid, -errno if not. | |
431 | */ | |
432 | ||
433 | static int validate_change(const struct cpuset *cur, const struct cpuset *trial) | |
434 | { | |
c9e5fe66 | 435 | struct cgroup *cgrp; |
1da177e4 | 436 | struct cpuset *c, *par; |
ae8086ce TH |
437 | int ret; |
438 | ||
439 | rcu_read_lock(); | |
1da177e4 LT |
440 | |
441 | /* Each of our child cpusets must be a subset of us */ | |
ae8086ce | 442 | ret = -EBUSY; |
c9e5fe66 | 443 | cpuset_for_each_child(c, cgrp, cur) |
ae8086ce TH |
444 | if (!is_cpuset_subset(c, trial)) |
445 | goto out; | |
1da177e4 LT |
446 | |
447 | /* Remaining checks don't apply to root cpuset */ | |
ae8086ce | 448 | ret = 0; |
69604067 | 449 | if (cur == &top_cpuset) |
ae8086ce | 450 | goto out; |
1da177e4 | 451 | |
c431069f | 452 | par = parent_cs(cur); |
69604067 | 453 | |
1da177e4 | 454 | /* We must be a subset of our parent cpuset */ |
ae8086ce | 455 | ret = -EACCES; |
1da177e4 | 456 | if (!is_cpuset_subset(trial, par)) |
ae8086ce | 457 | goto out; |
1da177e4 | 458 | |
2df167a3 PM |
459 | /* |
460 | * If either I or some sibling (!= me) is exclusive, we can't | |
461 | * overlap | |
462 | */ | |
ae8086ce | 463 | ret = -EINVAL; |
c9e5fe66 | 464 | cpuset_for_each_child(c, cgrp, par) { |
1da177e4 LT |
465 | if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) && |
466 | c != cur && | |
300ed6cb | 467 | cpumask_intersects(trial->cpus_allowed, c->cpus_allowed)) |
ae8086ce | 468 | goto out; |
1da177e4 LT |
469 | if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) && |
470 | c != cur && | |
471 | nodes_intersects(trial->mems_allowed, c->mems_allowed)) | |
ae8086ce | 472 | goto out; |
1da177e4 LT |
473 | } |
474 | ||
452477fa TH |
475 | /* |
476 | * Cpusets with tasks - existing or newly being attached - can't | |
477 | * have empty cpus_allowed or mems_allowed. | |
478 | */ | |
ae8086ce | 479 | ret = -ENOSPC; |
452477fa | 480 | if ((cgroup_task_count(cur->css.cgroup) || cur->attach_in_progress) && |
88fa523b | 481 | (cpumask_empty(trial->cpus_allowed) && |
ae8086ce TH |
482 | nodes_empty(trial->mems_allowed))) |
483 | goto out; | |
020958b6 | 484 | |
ae8086ce TH |
485 | ret = 0; |
486 | out: | |
487 | rcu_read_unlock(); | |
488 | return ret; | |
1da177e4 LT |
489 | } |
490 | ||
db7f47cf | 491 | #ifdef CONFIG_SMP |
029190c5 | 492 | /* |
cf417141 | 493 | * Helper routine for generate_sched_domains(). |
029190c5 PJ |
494 | * Do cpusets a, b have overlapping cpus_allowed masks? |
495 | */ | |
029190c5 PJ |
496 | static int cpusets_overlap(struct cpuset *a, struct cpuset *b) |
497 | { | |
300ed6cb | 498 | return cpumask_intersects(a->cpus_allowed, b->cpus_allowed); |
029190c5 PJ |
499 | } |
500 | ||
1d3504fc HS |
501 | static void |
502 | update_domain_attr(struct sched_domain_attr *dattr, struct cpuset *c) | |
503 | { | |
1d3504fc HS |
504 | if (dattr->relax_domain_level < c->relax_domain_level) |
505 | dattr->relax_domain_level = c->relax_domain_level; | |
506 | return; | |
507 | } | |
508 | ||
fc560a26 TH |
509 | static void update_domain_attr_tree(struct sched_domain_attr *dattr, |
510 | struct cpuset *root_cs) | |
f5393693 | 511 | { |
fc560a26 TH |
512 | struct cpuset *cp; |
513 | struct cgroup *pos_cgrp; | |
f5393693 | 514 | |
fc560a26 TH |
515 | rcu_read_lock(); |
516 | cpuset_for_each_descendant_pre(cp, pos_cgrp, root_cs) { | |
517 | /* skip the whole subtree if @cp doesn't have any CPU */ | |
518 | if (cpumask_empty(cp->cpus_allowed)) { | |
519 | pos_cgrp = cgroup_rightmost_descendant(pos_cgrp); | |
f5393693 | 520 | continue; |
fc560a26 | 521 | } |
f5393693 LJ |
522 | |
523 | if (is_sched_load_balance(cp)) | |
524 | update_domain_attr(dattr, cp); | |
f5393693 | 525 | } |
fc560a26 | 526 | rcu_read_unlock(); |
f5393693 LJ |
527 | } |
528 | ||
029190c5 | 529 | /* |
cf417141 MK |
530 | * generate_sched_domains() |
531 | * | |
532 | * This function builds a partial partition of the systems CPUs | |
533 | * A 'partial partition' is a set of non-overlapping subsets whose | |
534 | * union is a subset of that set. | |
0a0fca9d | 535 | * The output of this function needs to be passed to kernel/sched/core.c |
cf417141 MK |
536 | * partition_sched_domains() routine, which will rebuild the scheduler's |
537 | * load balancing domains (sched domains) as specified by that partial | |
538 | * partition. | |
029190c5 | 539 | * |
45ce80fb | 540 | * See "What is sched_load_balance" in Documentation/cgroups/cpusets.txt |
029190c5 PJ |
541 | * for a background explanation of this. |
542 | * | |
543 | * Does not return errors, on the theory that the callers of this | |
544 | * routine would rather not worry about failures to rebuild sched | |
545 | * domains when operating in the severe memory shortage situations | |
546 | * that could cause allocation failures below. | |
547 | * | |
5d21cc2d | 548 | * Must be called with cpuset_mutex held. |
029190c5 PJ |
549 | * |
550 | * The three key local variables below are: | |
aeed6824 | 551 | * q - a linked-list queue of cpuset pointers, used to implement a |
029190c5 PJ |
552 | * top-down scan of all cpusets. This scan loads a pointer |
553 | * to each cpuset marked is_sched_load_balance into the | |
554 | * array 'csa'. For our purposes, rebuilding the schedulers | |
555 | * sched domains, we can ignore !is_sched_load_balance cpusets. | |
556 | * csa - (for CpuSet Array) Array of pointers to all the cpusets | |
557 | * that need to be load balanced, for convenient iterative | |
558 | * access by the subsequent code that finds the best partition, | |
559 | * i.e the set of domains (subsets) of CPUs such that the | |
560 | * cpus_allowed of every cpuset marked is_sched_load_balance | |
561 | * is a subset of one of these domains, while there are as | |
562 | * many such domains as possible, each as small as possible. | |
563 | * doms - Conversion of 'csa' to an array of cpumasks, for passing to | |
0a0fca9d | 564 | * the kernel/sched/core.c routine partition_sched_domains() in a |
029190c5 PJ |
565 | * convenient format, that can be easily compared to the prior |
566 | * value to determine what partition elements (sched domains) | |
567 | * were changed (added or removed.) | |
568 | * | |
569 | * Finding the best partition (set of domains): | |
570 | * The triple nested loops below over i, j, k scan over the | |
571 | * load balanced cpusets (using the array of cpuset pointers in | |
572 | * csa[]) looking for pairs of cpusets that have overlapping | |
573 | * cpus_allowed, but which don't have the same 'pn' partition | |
574 | * number and gives them in the same partition number. It keeps | |
575 | * looping on the 'restart' label until it can no longer find | |
576 | * any such pairs. | |
577 | * | |
578 | * The union of the cpus_allowed masks from the set of | |
579 | * all cpusets having the same 'pn' value then form the one | |
580 | * element of the partition (one sched domain) to be passed to | |
581 | * partition_sched_domains(). | |
582 | */ | |
acc3f5d7 | 583 | static int generate_sched_domains(cpumask_var_t **domains, |
cf417141 | 584 | struct sched_domain_attr **attributes) |
029190c5 | 585 | { |
029190c5 PJ |
586 | struct cpuset *cp; /* scans q */ |
587 | struct cpuset **csa; /* array of all cpuset ptrs */ | |
588 | int csn; /* how many cpuset ptrs in csa so far */ | |
589 | int i, j, k; /* indices for partition finding loops */ | |
acc3f5d7 | 590 | cpumask_var_t *doms; /* resulting partition; i.e. sched domains */ |
1d3504fc | 591 | struct sched_domain_attr *dattr; /* attributes for custom domains */ |
1583715d | 592 | int ndoms = 0; /* number of sched domains in result */ |
6af866af | 593 | int nslot; /* next empty doms[] struct cpumask slot */ |
fc560a26 | 594 | struct cgroup *pos_cgrp; |
029190c5 | 595 | |
029190c5 | 596 | doms = NULL; |
1d3504fc | 597 | dattr = NULL; |
cf417141 | 598 | csa = NULL; |
029190c5 PJ |
599 | |
600 | /* Special case for the 99% of systems with one, full, sched domain */ | |
601 | if (is_sched_load_balance(&top_cpuset)) { | |
acc3f5d7 RR |
602 | ndoms = 1; |
603 | doms = alloc_sched_domains(ndoms); | |
029190c5 | 604 | if (!doms) |
cf417141 MK |
605 | goto done; |
606 | ||
1d3504fc HS |
607 | dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL); |
608 | if (dattr) { | |
609 | *dattr = SD_ATTR_INIT; | |
93a65575 | 610 | update_domain_attr_tree(dattr, &top_cpuset); |
1d3504fc | 611 | } |
acc3f5d7 | 612 | cpumask_copy(doms[0], top_cpuset.cpus_allowed); |
cf417141 | 613 | |
cf417141 | 614 | goto done; |
029190c5 PJ |
615 | } |
616 | ||
029190c5 PJ |
617 | csa = kmalloc(number_of_cpusets * sizeof(cp), GFP_KERNEL); |
618 | if (!csa) | |
619 | goto done; | |
620 | csn = 0; | |
621 | ||
fc560a26 TH |
622 | rcu_read_lock(); |
623 | cpuset_for_each_descendant_pre(cp, pos_cgrp, &top_cpuset) { | |
f5393693 | 624 | /* |
fc560a26 TH |
625 | * Continue traversing beyond @cp iff @cp has some CPUs and |
626 | * isn't load balancing. The former is obvious. The | |
627 | * latter: All child cpusets contain a subset of the | |
628 | * parent's cpus, so just skip them, and then we call | |
629 | * update_domain_attr_tree() to calc relax_domain_level of | |
630 | * the corresponding sched domain. | |
f5393693 | 631 | */ |
fc560a26 TH |
632 | if (!cpumask_empty(cp->cpus_allowed) && |
633 | !is_sched_load_balance(cp)) | |
f5393693 | 634 | continue; |
489a5393 | 635 | |
fc560a26 TH |
636 | if (is_sched_load_balance(cp)) |
637 | csa[csn++] = cp; | |
638 | ||
639 | /* skip @cp's subtree */ | |
640 | pos_cgrp = cgroup_rightmost_descendant(pos_cgrp); | |
641 | } | |
642 | rcu_read_unlock(); | |
029190c5 PJ |
643 | |
644 | for (i = 0; i < csn; i++) | |
645 | csa[i]->pn = i; | |
646 | ndoms = csn; | |
647 | ||
648 | restart: | |
649 | /* Find the best partition (set of sched domains) */ | |
650 | for (i = 0; i < csn; i++) { | |
651 | struct cpuset *a = csa[i]; | |
652 | int apn = a->pn; | |
653 | ||
654 | for (j = 0; j < csn; j++) { | |
655 | struct cpuset *b = csa[j]; | |
656 | int bpn = b->pn; | |
657 | ||
658 | if (apn != bpn && cpusets_overlap(a, b)) { | |
659 | for (k = 0; k < csn; k++) { | |
660 | struct cpuset *c = csa[k]; | |
661 | ||
662 | if (c->pn == bpn) | |
663 | c->pn = apn; | |
664 | } | |
665 | ndoms--; /* one less element */ | |
666 | goto restart; | |
667 | } | |
668 | } | |
669 | } | |
670 | ||
cf417141 MK |
671 | /* |
672 | * Now we know how many domains to create. | |
673 | * Convert <csn, csa> to <ndoms, doms> and populate cpu masks. | |
674 | */ | |
acc3f5d7 | 675 | doms = alloc_sched_domains(ndoms); |
700018e0 | 676 | if (!doms) |
cf417141 | 677 | goto done; |
cf417141 MK |
678 | |
679 | /* | |
680 | * The rest of the code, including the scheduler, can deal with | |
681 | * dattr==NULL case. No need to abort if alloc fails. | |
682 | */ | |
1d3504fc | 683 | dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL); |
029190c5 PJ |
684 | |
685 | for (nslot = 0, i = 0; i < csn; i++) { | |
686 | struct cpuset *a = csa[i]; | |
6af866af | 687 | struct cpumask *dp; |
029190c5 PJ |
688 | int apn = a->pn; |
689 | ||
cf417141 MK |
690 | if (apn < 0) { |
691 | /* Skip completed partitions */ | |
692 | continue; | |
693 | } | |
694 | ||
acc3f5d7 | 695 | dp = doms[nslot]; |
cf417141 MK |
696 | |
697 | if (nslot == ndoms) { | |
698 | static int warnings = 10; | |
699 | if (warnings) { | |
700 | printk(KERN_WARNING | |
701 | "rebuild_sched_domains confused:" | |
702 | " nslot %d, ndoms %d, csn %d, i %d," | |
703 | " apn %d\n", | |
704 | nslot, ndoms, csn, i, apn); | |
705 | warnings--; | |
029190c5 | 706 | } |
cf417141 MK |
707 | continue; |
708 | } | |
029190c5 | 709 | |
6af866af | 710 | cpumask_clear(dp); |
cf417141 MK |
711 | if (dattr) |
712 | *(dattr + nslot) = SD_ATTR_INIT; | |
713 | for (j = i; j < csn; j++) { | |
714 | struct cpuset *b = csa[j]; | |
715 | ||
716 | if (apn == b->pn) { | |
300ed6cb | 717 | cpumask_or(dp, dp, b->cpus_allowed); |
cf417141 MK |
718 | if (dattr) |
719 | update_domain_attr_tree(dattr + nslot, b); | |
720 | ||
721 | /* Done with this partition */ | |
722 | b->pn = -1; | |
029190c5 | 723 | } |
029190c5 | 724 | } |
cf417141 | 725 | nslot++; |
029190c5 PJ |
726 | } |
727 | BUG_ON(nslot != ndoms); | |
728 | ||
cf417141 MK |
729 | done: |
730 | kfree(csa); | |
731 | ||
700018e0 LZ |
732 | /* |
733 | * Fallback to the default domain if kmalloc() failed. | |
734 | * See comments in partition_sched_domains(). | |
735 | */ | |
736 | if (doms == NULL) | |
737 | ndoms = 1; | |
738 | ||
cf417141 MK |
739 | *domains = doms; |
740 | *attributes = dattr; | |
741 | return ndoms; | |
742 | } | |
743 | ||
744 | /* | |
745 | * Rebuild scheduler domains. | |
746 | * | |
699140ba TH |
747 | * If the flag 'sched_load_balance' of any cpuset with non-empty |
748 | * 'cpus' changes, or if the 'cpus' allowed changes in any cpuset | |
749 | * which has that flag enabled, or if any cpuset with a non-empty | |
750 | * 'cpus' is removed, then call this routine to rebuild the | |
751 | * scheduler's dynamic sched domains. | |
cf417141 | 752 | * |
5d21cc2d | 753 | * Call with cpuset_mutex held. Takes get_online_cpus(). |
cf417141 | 754 | */ |
699140ba | 755 | static void rebuild_sched_domains_locked(void) |
cf417141 MK |
756 | { |
757 | struct sched_domain_attr *attr; | |
acc3f5d7 | 758 | cpumask_var_t *doms; |
cf417141 MK |
759 | int ndoms; |
760 | ||
5d21cc2d | 761 | lockdep_assert_held(&cpuset_mutex); |
86ef5c9a | 762 | get_online_cpus(); |
cf417141 | 763 | |
5b16c2a4 LZ |
764 | /* |
765 | * We have raced with CPU hotplug. Don't do anything to avoid | |
766 | * passing doms with offlined cpu to partition_sched_domains(). | |
767 | * Anyways, hotplug work item will rebuild sched domains. | |
768 | */ | |
769 | if (!cpumask_equal(top_cpuset.cpus_allowed, cpu_active_mask)) | |
770 | goto out; | |
771 | ||
cf417141 | 772 | /* Generate domain masks and attrs */ |
cf417141 | 773 | ndoms = generate_sched_domains(&doms, &attr); |
cf417141 MK |
774 | |
775 | /* Have scheduler rebuild the domains */ | |
776 | partition_sched_domains(ndoms, doms, attr); | |
5b16c2a4 | 777 | out: |
86ef5c9a | 778 | put_online_cpus(); |
cf417141 | 779 | } |
db7f47cf | 780 | #else /* !CONFIG_SMP */ |
699140ba | 781 | static void rebuild_sched_domains_locked(void) |
db7f47cf PM |
782 | { |
783 | } | |
db7f47cf | 784 | #endif /* CONFIG_SMP */ |
029190c5 | 785 | |
cf417141 MK |
786 | void rebuild_sched_domains(void) |
787 | { | |
5d21cc2d | 788 | mutex_lock(&cpuset_mutex); |
699140ba | 789 | rebuild_sched_domains_locked(); |
5d21cc2d | 790 | mutex_unlock(&cpuset_mutex); |
029190c5 PJ |
791 | } |
792 | ||
070b57fc LZ |
793 | /* |
794 | * effective_cpumask_cpuset - return nearest ancestor with non-empty cpus | |
795 | * @cs: the cpuset in interest | |
58f4790b | 796 | * |
070b57fc LZ |
797 | * A cpuset's effective cpumask is the cpumask of the nearest ancestor |
798 | * with non-empty cpus. We use effective cpumask whenever: | |
799 | * - we update tasks' cpus_allowed. (they take on the ancestor's cpumask | |
800 | * if the cpuset they reside in has no cpus) | |
801 | * - we want to retrieve task_cs(tsk)'s cpus_allowed. | |
802 | * | |
803 | * Called with cpuset_mutex held. cpuset_cpus_allowed_fallback() is an | |
804 | * exception. See comments there. | |
805 | */ | |
806 | static struct cpuset *effective_cpumask_cpuset(struct cpuset *cs) | |
807 | { | |
808 | while (cpumask_empty(cs->cpus_allowed)) | |
809 | cs = parent_cs(cs); | |
810 | return cs; | |
811 | } | |
812 | ||
813 | /* | |
814 | * effective_nodemask_cpuset - return nearest ancestor with non-empty mems | |
815 | * @cs: the cpuset in interest | |
816 | * | |
817 | * A cpuset's effective nodemask is the nodemask of the nearest ancestor | |
818 | * with non-empty memss. We use effective nodemask whenever: | |
819 | * - we update tasks' mems_allowed. (they take on the ancestor's nodemask | |
820 | * if the cpuset they reside in has no mems) | |
821 | * - we want to retrieve task_cs(tsk)'s mems_allowed. | |
822 | * | |
823 | * Called with cpuset_mutex held. | |
053199ed | 824 | */ |
070b57fc | 825 | static struct cpuset *effective_nodemask_cpuset(struct cpuset *cs) |
58f4790b | 826 | { |
070b57fc LZ |
827 | while (nodes_empty(cs->mems_allowed)) |
828 | cs = parent_cs(cs); | |
829 | return cs; | |
58f4790b | 830 | } |
053199ed | 831 | |
58f4790b CW |
832 | /** |
833 | * cpuset_change_cpumask - make a task's cpus_allowed the same as its cpuset's | |
834 | * @tsk: task to test | |
835 | * @scan: struct cgroup_scanner containing the cgroup of the task | |
836 | * | |
837 | * Called by cgroup_scan_tasks() for each task in a cgroup whose | |
838 | * cpus_allowed mask needs to be changed. | |
839 | * | |
840 | * We don't need to re-check for the cgroup/cpuset membership, since we're | |
5d21cc2d | 841 | * holding cpuset_mutex at this point. |
58f4790b | 842 | */ |
9e0c914c AB |
843 | static void cpuset_change_cpumask(struct task_struct *tsk, |
844 | struct cgroup_scanner *scan) | |
58f4790b | 845 | { |
070b57fc LZ |
846 | struct cpuset *cpus_cs; |
847 | ||
6f4b7e63 | 848 | cpus_cs = effective_cpumask_cpuset(cgroup_cs(scan->cgrp)); |
070b57fc | 849 | set_cpus_allowed_ptr(tsk, cpus_cs->cpus_allowed); |
58f4790b CW |
850 | } |
851 | ||
0b2f630a MX |
852 | /** |
853 | * update_tasks_cpumask - Update the cpumasks of tasks in the cpuset. | |
854 | * @cs: the cpuset in which each task's cpus_allowed mask needs to be changed | |
4e74339a | 855 | * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks() |
0b2f630a | 856 | * |
5d21cc2d | 857 | * Called with cpuset_mutex held |
0b2f630a MX |
858 | * |
859 | * The cgroup_scan_tasks() function will scan all the tasks in a cgroup, | |
860 | * calling callback functions for each. | |
861 | * | |
4e74339a LZ |
862 | * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0 |
863 | * if @heap != NULL. | |
0b2f630a | 864 | */ |
4e74339a | 865 | static void update_tasks_cpumask(struct cpuset *cs, struct ptr_heap *heap) |
0b2f630a MX |
866 | { |
867 | struct cgroup_scanner scan; | |
0b2f630a | 868 | |
6f4b7e63 | 869 | scan.cgrp = cs->css.cgroup; |
249cc86d | 870 | scan.test_task = NULL; |
0b2f630a | 871 | scan.process_task = cpuset_change_cpumask; |
4e74339a LZ |
872 | scan.heap = heap; |
873 | cgroup_scan_tasks(&scan); | |
0b2f630a MX |
874 | } |
875 | ||
5c5cc623 LZ |
876 | /* |
877 | * update_tasks_cpumask_hier - Update the cpumasks of tasks in the hierarchy. | |
878 | * @root_cs: the root cpuset of the hierarchy | |
879 | * @update_root: update root cpuset or not? | |
880 | * @heap: the heap used by cgroup_scan_tasks() | |
881 | * | |
882 | * This will update cpumasks of tasks in @root_cs and all other empty cpusets | |
883 | * which take on cpumask of @root_cs. | |
884 | * | |
885 | * Called with cpuset_mutex held | |
886 | */ | |
887 | static void update_tasks_cpumask_hier(struct cpuset *root_cs, | |
888 | bool update_root, struct ptr_heap *heap) | |
889 | { | |
890 | struct cpuset *cp; | |
891 | struct cgroup *pos_cgrp; | |
892 | ||
893 | if (update_root) | |
894 | update_tasks_cpumask(root_cs, heap); | |
895 | ||
896 | rcu_read_lock(); | |
897 | cpuset_for_each_descendant_pre(cp, pos_cgrp, root_cs) { | |
898 | /* skip the whole subtree if @cp have some CPU */ | |
899 | if (!cpumask_empty(cp->cpus_allowed)) { | |
900 | pos_cgrp = cgroup_rightmost_descendant(pos_cgrp); | |
901 | continue; | |
902 | } | |
903 | if (!css_tryget(&cp->css)) | |
904 | continue; | |
905 | rcu_read_unlock(); | |
906 | ||
907 | update_tasks_cpumask(cp, heap); | |
908 | ||
909 | rcu_read_lock(); | |
910 | css_put(&cp->css); | |
911 | } | |
912 | rcu_read_unlock(); | |
913 | } | |
914 | ||
58f4790b CW |
915 | /** |
916 | * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it | |
917 | * @cs: the cpuset to consider | |
918 | * @buf: buffer of cpu numbers written to this cpuset | |
919 | */ | |
645fcc9d LZ |
920 | static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs, |
921 | const char *buf) | |
1da177e4 | 922 | { |
4e74339a | 923 | struct ptr_heap heap; |
58f4790b CW |
924 | int retval; |
925 | int is_load_balanced; | |
1da177e4 | 926 | |
5f054e31 | 927 | /* top_cpuset.cpus_allowed tracks cpu_online_mask; it's read-only */ |
4c4d50f7 PJ |
928 | if (cs == &top_cpuset) |
929 | return -EACCES; | |
930 | ||
6f7f02e7 | 931 | /* |
c8d9c90c | 932 | * An empty cpus_allowed is ok only if the cpuset has no tasks. |
020958b6 PJ |
933 | * Since cpulist_parse() fails on an empty mask, we special case |
934 | * that parsing. The validate_change() call ensures that cpusets | |
935 | * with tasks have cpus. | |
6f7f02e7 | 936 | */ |
020958b6 | 937 | if (!*buf) { |
300ed6cb | 938 | cpumask_clear(trialcs->cpus_allowed); |
6f7f02e7 | 939 | } else { |
300ed6cb | 940 | retval = cpulist_parse(buf, trialcs->cpus_allowed); |
6f7f02e7 DR |
941 | if (retval < 0) |
942 | return retval; | |
37340746 | 943 | |
6ad4c188 | 944 | if (!cpumask_subset(trialcs->cpus_allowed, cpu_active_mask)) |
37340746 | 945 | return -EINVAL; |
6f7f02e7 | 946 | } |
029190c5 | 947 | |
8707d8b8 | 948 | /* Nothing to do if the cpus didn't change */ |
300ed6cb | 949 | if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed)) |
8707d8b8 | 950 | return 0; |
58f4790b | 951 | |
a73456f3 LZ |
952 | retval = validate_change(cs, trialcs); |
953 | if (retval < 0) | |
954 | return retval; | |
955 | ||
4e74339a LZ |
956 | retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL); |
957 | if (retval) | |
958 | return retval; | |
959 | ||
645fcc9d | 960 | is_load_balanced = is_sched_load_balance(trialcs); |
029190c5 | 961 | |
3d3f26a7 | 962 | mutex_lock(&callback_mutex); |
300ed6cb | 963 | cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed); |
3d3f26a7 | 964 | mutex_unlock(&callback_mutex); |
029190c5 | 965 | |
5c5cc623 | 966 | update_tasks_cpumask_hier(cs, true, &heap); |
4e74339a LZ |
967 | |
968 | heap_free(&heap); | |
58f4790b | 969 | |
8707d8b8 | 970 | if (is_load_balanced) |
699140ba | 971 | rebuild_sched_domains_locked(); |
85d7b949 | 972 | return 0; |
1da177e4 LT |
973 | } |
974 | ||
e4e364e8 PJ |
975 | /* |
976 | * cpuset_migrate_mm | |
977 | * | |
978 | * Migrate memory region from one set of nodes to another. | |
979 | * | |
980 | * Temporarilly set tasks mems_allowed to target nodes of migration, | |
981 | * so that the migration code can allocate pages on these nodes. | |
982 | * | |
5d21cc2d | 983 | * Call holding cpuset_mutex, so current's cpuset won't change |
c8d9c90c | 984 | * during this call, as manage_mutex holds off any cpuset_attach() |
e4e364e8 PJ |
985 | * calls. Therefore we don't need to take task_lock around the |
986 | * call to guarantee_online_mems(), as we know no one is changing | |
2df167a3 | 987 | * our task's cpuset. |
e4e364e8 | 988 | * |
e4e364e8 PJ |
989 | * While the mm_struct we are migrating is typically from some |
990 | * other task, the task_struct mems_allowed that we are hacking | |
991 | * is for our current task, which must allocate new pages for that | |
992 | * migrating memory region. | |
e4e364e8 PJ |
993 | */ |
994 | ||
995 | static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from, | |
996 | const nodemask_t *to) | |
997 | { | |
998 | struct task_struct *tsk = current; | |
070b57fc | 999 | struct cpuset *mems_cs; |
e4e364e8 | 1000 | |
e4e364e8 | 1001 | tsk->mems_allowed = *to; |
e4e364e8 PJ |
1002 | |
1003 | do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL); | |
1004 | ||
070b57fc LZ |
1005 | mems_cs = effective_nodemask_cpuset(task_cs(tsk)); |
1006 | guarantee_online_mems(mems_cs, &tsk->mems_allowed); | |
e4e364e8 PJ |
1007 | } |
1008 | ||
3b6766fe | 1009 | /* |
58568d2a MX |
1010 | * cpuset_change_task_nodemask - change task's mems_allowed and mempolicy |
1011 | * @tsk: the task to change | |
1012 | * @newmems: new nodes that the task will be set | |
1013 | * | |
1014 | * In order to avoid seeing no nodes if the old and new nodes are disjoint, | |
1015 | * we structure updates as setting all new allowed nodes, then clearing newly | |
1016 | * disallowed ones. | |
58568d2a MX |
1017 | */ |
1018 | static void cpuset_change_task_nodemask(struct task_struct *tsk, | |
1019 | nodemask_t *newmems) | |
1020 | { | |
b246272e | 1021 | bool need_loop; |
89e8a244 | 1022 | |
c0ff7453 MX |
1023 | /* |
1024 | * Allow tasks that have access to memory reserves because they have | |
1025 | * been OOM killed to get memory anywhere. | |
1026 | */ | |
1027 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
1028 | return; | |
1029 | if (current->flags & PF_EXITING) /* Let dying task have memory */ | |
1030 | return; | |
1031 | ||
1032 | task_lock(tsk); | |
b246272e DR |
1033 | /* |
1034 | * Determine if a loop is necessary if another thread is doing | |
1035 | * get_mems_allowed(). If at least one node remains unchanged and | |
1036 | * tsk does not have a mempolicy, then an empty nodemask will not be | |
1037 | * possible when mems_allowed is larger than a word. | |
1038 | */ | |
1039 | need_loop = task_has_mempolicy(tsk) || | |
1040 | !nodes_intersects(*newmems, tsk->mems_allowed); | |
c0ff7453 | 1041 | |
cc9a6c87 MG |
1042 | if (need_loop) |
1043 | write_seqcount_begin(&tsk->mems_allowed_seq); | |
c0ff7453 | 1044 | |
cc9a6c87 MG |
1045 | nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems); |
1046 | mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP1); | |
c0ff7453 MX |
1047 | |
1048 | mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP2); | |
58568d2a | 1049 | tsk->mems_allowed = *newmems; |
cc9a6c87 MG |
1050 | |
1051 | if (need_loop) | |
1052 | write_seqcount_end(&tsk->mems_allowed_seq); | |
1053 | ||
c0ff7453 | 1054 | task_unlock(tsk); |
58568d2a MX |
1055 | } |
1056 | ||
1057 | /* | |
1058 | * Update task's mems_allowed and rebind its mempolicy and vmas' mempolicy | |
1059 | * of it to cpuset's new mems_allowed, and migrate pages to new nodes if | |
5d21cc2d | 1060 | * memory_migrate flag is set. Called with cpuset_mutex held. |
3b6766fe LZ |
1061 | */ |
1062 | static void cpuset_change_nodemask(struct task_struct *p, | |
1063 | struct cgroup_scanner *scan) | |
1064 | { | |
6f4b7e63 | 1065 | struct cpuset *cs = cgroup_cs(scan->cgrp); |
3b6766fe | 1066 | struct mm_struct *mm; |
3b6766fe | 1067 | int migrate; |
33ad801d | 1068 | nodemask_t *newmems = scan->data; |
58568d2a | 1069 | |
33ad801d | 1070 | cpuset_change_task_nodemask(p, newmems); |
53feb297 | 1071 | |
3b6766fe LZ |
1072 | mm = get_task_mm(p); |
1073 | if (!mm) | |
1074 | return; | |
1075 | ||
3b6766fe LZ |
1076 | migrate = is_memory_migrate(cs); |
1077 | ||
1078 | mpol_rebind_mm(mm, &cs->mems_allowed); | |
1079 | if (migrate) | |
33ad801d | 1080 | cpuset_migrate_mm(mm, &cs->old_mems_allowed, newmems); |
3b6766fe LZ |
1081 | mmput(mm); |
1082 | } | |
1083 | ||
8793d854 PM |
1084 | static void *cpuset_being_rebound; |
1085 | ||
0b2f630a MX |
1086 | /** |
1087 | * update_tasks_nodemask - Update the nodemasks of tasks in the cpuset. | |
1088 | * @cs: the cpuset in which each task's mems_allowed mask needs to be changed | |
010cfac4 | 1089 | * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks() |
0b2f630a | 1090 | * |
5d21cc2d | 1091 | * Called with cpuset_mutex held |
010cfac4 LZ |
1092 | * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0 |
1093 | * if @heap != NULL. | |
0b2f630a | 1094 | */ |
33ad801d | 1095 | static void update_tasks_nodemask(struct cpuset *cs, struct ptr_heap *heap) |
1da177e4 | 1096 | { |
33ad801d | 1097 | static nodemask_t newmems; /* protected by cpuset_mutex */ |
3b6766fe | 1098 | struct cgroup_scanner scan; |
070b57fc | 1099 | struct cpuset *mems_cs = effective_nodemask_cpuset(cs); |
59dac16f | 1100 | |
846a16bf | 1101 | cpuset_being_rebound = cs; /* causes mpol_dup() rebind */ |
4225399a | 1102 | |
070b57fc | 1103 | guarantee_online_mems(mems_cs, &newmems); |
33ad801d | 1104 | |
6f4b7e63 | 1105 | scan.cgrp = cs->css.cgroup; |
3b6766fe LZ |
1106 | scan.test_task = NULL; |
1107 | scan.process_task = cpuset_change_nodemask; | |
010cfac4 | 1108 | scan.heap = heap; |
33ad801d | 1109 | scan.data = &newmems; |
4225399a PJ |
1110 | |
1111 | /* | |
3b6766fe LZ |
1112 | * The mpol_rebind_mm() call takes mmap_sem, which we couldn't |
1113 | * take while holding tasklist_lock. Forks can happen - the | |
1114 | * mpol_dup() cpuset_being_rebound check will catch such forks, | |
1115 | * and rebind their vma mempolicies too. Because we still hold | |
5d21cc2d | 1116 | * the global cpuset_mutex, we know that no other rebind effort |
3b6766fe | 1117 | * will be contending for the global variable cpuset_being_rebound. |
4225399a | 1118 | * It's ok if we rebind the same mm twice; mpol_rebind_mm() |
04c19fa6 | 1119 | * is idempotent. Also migrate pages in each mm to new nodes. |
4225399a | 1120 | */ |
010cfac4 | 1121 | cgroup_scan_tasks(&scan); |
4225399a | 1122 | |
33ad801d LZ |
1123 | /* |
1124 | * All the tasks' nodemasks have been updated, update | |
1125 | * cs->old_mems_allowed. | |
1126 | */ | |
1127 | cs->old_mems_allowed = newmems; | |
1128 | ||
2df167a3 | 1129 | /* We're done rebinding vmas to this cpuset's new mems_allowed. */ |
8793d854 | 1130 | cpuset_being_rebound = NULL; |
1da177e4 LT |
1131 | } |
1132 | ||
5c5cc623 LZ |
1133 | /* |
1134 | * update_tasks_nodemask_hier - Update the nodemasks of tasks in the hierarchy. | |
1135 | * @cs: the root cpuset of the hierarchy | |
1136 | * @update_root: update the root cpuset or not? | |
1137 | * @heap: the heap used by cgroup_scan_tasks() | |
1138 | * | |
1139 | * This will update nodemasks of tasks in @root_cs and all other empty cpusets | |
1140 | * which take on nodemask of @root_cs. | |
1141 | * | |
1142 | * Called with cpuset_mutex held | |
1143 | */ | |
1144 | static void update_tasks_nodemask_hier(struct cpuset *root_cs, | |
1145 | bool update_root, struct ptr_heap *heap) | |
1146 | { | |
1147 | struct cpuset *cp; | |
1148 | struct cgroup *pos_cgrp; | |
1149 | ||
1150 | if (update_root) | |
1151 | update_tasks_nodemask(root_cs, heap); | |
1152 | ||
1153 | rcu_read_lock(); | |
1154 | cpuset_for_each_descendant_pre(cp, pos_cgrp, root_cs) { | |
1155 | /* skip the whole subtree if @cp have some CPU */ | |
1156 | if (!nodes_empty(cp->mems_allowed)) { | |
1157 | pos_cgrp = cgroup_rightmost_descendant(pos_cgrp); | |
1158 | continue; | |
1159 | } | |
1160 | if (!css_tryget(&cp->css)) | |
1161 | continue; | |
1162 | rcu_read_unlock(); | |
1163 | ||
1164 | update_tasks_nodemask(cp, heap); | |
1165 | ||
1166 | rcu_read_lock(); | |
1167 | css_put(&cp->css); | |
1168 | } | |
1169 | rcu_read_unlock(); | |
1170 | } | |
1171 | ||
0b2f630a MX |
1172 | /* |
1173 | * Handle user request to change the 'mems' memory placement | |
1174 | * of a cpuset. Needs to validate the request, update the | |
58568d2a MX |
1175 | * cpusets mems_allowed, and for each task in the cpuset, |
1176 | * update mems_allowed and rebind task's mempolicy and any vma | |
1177 | * mempolicies and if the cpuset is marked 'memory_migrate', | |
1178 | * migrate the tasks pages to the new memory. | |
0b2f630a | 1179 | * |
5d21cc2d | 1180 | * Call with cpuset_mutex held. May take callback_mutex during call. |
0b2f630a MX |
1181 | * Will take tasklist_lock, scan tasklist for tasks in cpuset cs, |
1182 | * lock each such tasks mm->mmap_sem, scan its vma's and rebind | |
1183 | * their mempolicies to the cpusets new mems_allowed. | |
1184 | */ | |
645fcc9d LZ |
1185 | static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs, |
1186 | const char *buf) | |
0b2f630a | 1187 | { |
0b2f630a | 1188 | int retval; |
010cfac4 | 1189 | struct ptr_heap heap; |
0b2f630a MX |
1190 | |
1191 | /* | |
38d7bee9 | 1192 | * top_cpuset.mems_allowed tracks node_stats[N_MEMORY]; |
0b2f630a MX |
1193 | * it's read-only |
1194 | */ | |
53feb297 MX |
1195 | if (cs == &top_cpuset) { |
1196 | retval = -EACCES; | |
1197 | goto done; | |
1198 | } | |
0b2f630a | 1199 | |
0b2f630a MX |
1200 | /* |
1201 | * An empty mems_allowed is ok iff there are no tasks in the cpuset. | |
1202 | * Since nodelist_parse() fails on an empty mask, we special case | |
1203 | * that parsing. The validate_change() call ensures that cpusets | |
1204 | * with tasks have memory. | |
1205 | */ | |
1206 | if (!*buf) { | |
645fcc9d | 1207 | nodes_clear(trialcs->mems_allowed); |
0b2f630a | 1208 | } else { |
645fcc9d | 1209 | retval = nodelist_parse(buf, trialcs->mems_allowed); |
0b2f630a MX |
1210 | if (retval < 0) |
1211 | goto done; | |
1212 | ||
645fcc9d | 1213 | if (!nodes_subset(trialcs->mems_allowed, |
38d7bee9 | 1214 | node_states[N_MEMORY])) { |
53feb297 MX |
1215 | retval = -EINVAL; |
1216 | goto done; | |
1217 | } | |
0b2f630a | 1218 | } |
33ad801d LZ |
1219 | |
1220 | if (nodes_equal(cs->mems_allowed, trialcs->mems_allowed)) { | |
0b2f630a MX |
1221 | retval = 0; /* Too easy - nothing to do */ |
1222 | goto done; | |
1223 | } | |
645fcc9d | 1224 | retval = validate_change(cs, trialcs); |
0b2f630a MX |
1225 | if (retval < 0) |
1226 | goto done; | |
1227 | ||
010cfac4 LZ |
1228 | retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL); |
1229 | if (retval < 0) | |
1230 | goto done; | |
1231 | ||
0b2f630a | 1232 | mutex_lock(&callback_mutex); |
645fcc9d | 1233 | cs->mems_allowed = trialcs->mems_allowed; |
0b2f630a MX |
1234 | mutex_unlock(&callback_mutex); |
1235 | ||
5c5cc623 | 1236 | update_tasks_nodemask_hier(cs, true, &heap); |
010cfac4 LZ |
1237 | |
1238 | heap_free(&heap); | |
0b2f630a MX |
1239 | done: |
1240 | return retval; | |
1241 | } | |
1242 | ||
8793d854 PM |
1243 | int current_cpuset_is_being_rebound(void) |
1244 | { | |
1245 | return task_cs(current) == cpuset_being_rebound; | |
1246 | } | |
1247 | ||
5be7a479 | 1248 | static int update_relax_domain_level(struct cpuset *cs, s64 val) |
1d3504fc | 1249 | { |
db7f47cf | 1250 | #ifdef CONFIG_SMP |
60495e77 | 1251 | if (val < -1 || val >= sched_domain_level_max) |
30e0e178 | 1252 | return -EINVAL; |
db7f47cf | 1253 | #endif |
1d3504fc HS |
1254 | |
1255 | if (val != cs->relax_domain_level) { | |
1256 | cs->relax_domain_level = val; | |
300ed6cb LZ |
1257 | if (!cpumask_empty(cs->cpus_allowed) && |
1258 | is_sched_load_balance(cs)) | |
699140ba | 1259 | rebuild_sched_domains_locked(); |
1d3504fc HS |
1260 | } |
1261 | ||
1262 | return 0; | |
1263 | } | |
1264 | ||
950592f7 MX |
1265 | /* |
1266 | * cpuset_change_flag - make a task's spread flags the same as its cpuset's | |
1267 | * @tsk: task to be updated | |
1268 | * @scan: struct cgroup_scanner containing the cgroup of the task | |
1269 | * | |
1270 | * Called by cgroup_scan_tasks() for each task in a cgroup. | |
1271 | * | |
1272 | * We don't need to re-check for the cgroup/cpuset membership, since we're | |
5d21cc2d | 1273 | * holding cpuset_mutex at this point. |
950592f7 MX |
1274 | */ |
1275 | static void cpuset_change_flag(struct task_struct *tsk, | |
1276 | struct cgroup_scanner *scan) | |
1277 | { | |
6f4b7e63 | 1278 | cpuset_update_task_spread_flag(cgroup_cs(scan->cgrp), tsk); |
950592f7 MX |
1279 | } |
1280 | ||
1281 | /* | |
1282 | * update_tasks_flags - update the spread flags of tasks in the cpuset. | |
1283 | * @cs: the cpuset in which each task's spread flags needs to be changed | |
1284 | * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks() | |
1285 | * | |
5d21cc2d | 1286 | * Called with cpuset_mutex held |
950592f7 MX |
1287 | * |
1288 | * The cgroup_scan_tasks() function will scan all the tasks in a cgroup, | |
1289 | * calling callback functions for each. | |
1290 | * | |
1291 | * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0 | |
1292 | * if @heap != NULL. | |
1293 | */ | |
1294 | static void update_tasks_flags(struct cpuset *cs, struct ptr_heap *heap) | |
1295 | { | |
1296 | struct cgroup_scanner scan; | |
1297 | ||
6f4b7e63 | 1298 | scan.cgrp = cs->css.cgroup; |
950592f7 MX |
1299 | scan.test_task = NULL; |
1300 | scan.process_task = cpuset_change_flag; | |
1301 | scan.heap = heap; | |
1302 | cgroup_scan_tasks(&scan); | |
1303 | } | |
1304 | ||
1da177e4 LT |
1305 | /* |
1306 | * update_flag - read a 0 or a 1 in a file and update associated flag | |
78608366 PM |
1307 | * bit: the bit to update (see cpuset_flagbits_t) |
1308 | * cs: the cpuset to update | |
1309 | * turning_on: whether the flag is being set or cleared | |
053199ed | 1310 | * |
5d21cc2d | 1311 | * Call with cpuset_mutex held. |
1da177e4 LT |
1312 | */ |
1313 | ||
700fe1ab PM |
1314 | static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs, |
1315 | int turning_on) | |
1da177e4 | 1316 | { |
645fcc9d | 1317 | struct cpuset *trialcs; |
40b6a762 | 1318 | int balance_flag_changed; |
950592f7 MX |
1319 | int spread_flag_changed; |
1320 | struct ptr_heap heap; | |
1321 | int err; | |
1da177e4 | 1322 | |
645fcc9d LZ |
1323 | trialcs = alloc_trial_cpuset(cs); |
1324 | if (!trialcs) | |
1325 | return -ENOMEM; | |
1326 | ||
1da177e4 | 1327 | if (turning_on) |
645fcc9d | 1328 | set_bit(bit, &trialcs->flags); |
1da177e4 | 1329 | else |
645fcc9d | 1330 | clear_bit(bit, &trialcs->flags); |
1da177e4 | 1331 | |
645fcc9d | 1332 | err = validate_change(cs, trialcs); |
85d7b949 | 1333 | if (err < 0) |
645fcc9d | 1334 | goto out; |
029190c5 | 1335 | |
950592f7 MX |
1336 | err = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL); |
1337 | if (err < 0) | |
1338 | goto out; | |
1339 | ||
029190c5 | 1340 | balance_flag_changed = (is_sched_load_balance(cs) != |
645fcc9d | 1341 | is_sched_load_balance(trialcs)); |
029190c5 | 1342 | |
950592f7 MX |
1343 | spread_flag_changed = ((is_spread_slab(cs) != is_spread_slab(trialcs)) |
1344 | || (is_spread_page(cs) != is_spread_page(trialcs))); | |
1345 | ||
3d3f26a7 | 1346 | mutex_lock(&callback_mutex); |
645fcc9d | 1347 | cs->flags = trialcs->flags; |
3d3f26a7 | 1348 | mutex_unlock(&callback_mutex); |
85d7b949 | 1349 | |
300ed6cb | 1350 | if (!cpumask_empty(trialcs->cpus_allowed) && balance_flag_changed) |
699140ba | 1351 | rebuild_sched_domains_locked(); |
029190c5 | 1352 | |
950592f7 MX |
1353 | if (spread_flag_changed) |
1354 | update_tasks_flags(cs, &heap); | |
1355 | heap_free(&heap); | |
645fcc9d LZ |
1356 | out: |
1357 | free_trial_cpuset(trialcs); | |
1358 | return err; | |
1da177e4 LT |
1359 | } |
1360 | ||
3e0d98b9 | 1361 | /* |
80f7228b | 1362 | * Frequency meter - How fast is some event occurring? |
3e0d98b9 PJ |
1363 | * |
1364 | * These routines manage a digitally filtered, constant time based, | |
1365 | * event frequency meter. There are four routines: | |
1366 | * fmeter_init() - initialize a frequency meter. | |
1367 | * fmeter_markevent() - called each time the event happens. | |
1368 | * fmeter_getrate() - returns the recent rate of such events. | |
1369 | * fmeter_update() - internal routine used to update fmeter. | |
1370 | * | |
1371 | * A common data structure is passed to each of these routines, | |
1372 | * which is used to keep track of the state required to manage the | |
1373 | * frequency meter and its digital filter. | |
1374 | * | |
1375 | * The filter works on the number of events marked per unit time. | |
1376 | * The filter is single-pole low-pass recursive (IIR). The time unit | |
1377 | * is 1 second. Arithmetic is done using 32-bit integers scaled to | |
1378 | * simulate 3 decimal digits of precision (multiplied by 1000). | |
1379 | * | |
1380 | * With an FM_COEF of 933, and a time base of 1 second, the filter | |
1381 | * has a half-life of 10 seconds, meaning that if the events quit | |
1382 | * happening, then the rate returned from the fmeter_getrate() | |
1383 | * will be cut in half each 10 seconds, until it converges to zero. | |
1384 | * | |
1385 | * It is not worth doing a real infinitely recursive filter. If more | |
1386 | * than FM_MAXTICKS ticks have elapsed since the last filter event, | |
1387 | * just compute FM_MAXTICKS ticks worth, by which point the level | |
1388 | * will be stable. | |
1389 | * | |
1390 | * Limit the count of unprocessed events to FM_MAXCNT, so as to avoid | |
1391 | * arithmetic overflow in the fmeter_update() routine. | |
1392 | * | |
1393 | * Given the simple 32 bit integer arithmetic used, this meter works | |
1394 | * best for reporting rates between one per millisecond (msec) and | |
1395 | * one per 32 (approx) seconds. At constant rates faster than one | |
1396 | * per msec it maxes out at values just under 1,000,000. At constant | |
1397 | * rates between one per msec, and one per second it will stabilize | |
1398 | * to a value N*1000, where N is the rate of events per second. | |
1399 | * At constant rates between one per second and one per 32 seconds, | |
1400 | * it will be choppy, moving up on the seconds that have an event, | |
1401 | * and then decaying until the next event. At rates slower than | |
1402 | * about one in 32 seconds, it decays all the way back to zero between | |
1403 | * each event. | |
1404 | */ | |
1405 | ||
1406 | #define FM_COEF 933 /* coefficient for half-life of 10 secs */ | |
1407 | #define FM_MAXTICKS ((time_t)99) /* useless computing more ticks than this */ | |
1408 | #define FM_MAXCNT 1000000 /* limit cnt to avoid overflow */ | |
1409 | #define FM_SCALE 1000 /* faux fixed point scale */ | |
1410 | ||
1411 | /* Initialize a frequency meter */ | |
1412 | static void fmeter_init(struct fmeter *fmp) | |
1413 | { | |
1414 | fmp->cnt = 0; | |
1415 | fmp->val = 0; | |
1416 | fmp->time = 0; | |
1417 | spin_lock_init(&fmp->lock); | |
1418 | } | |
1419 | ||
1420 | /* Internal meter update - process cnt events and update value */ | |
1421 | static void fmeter_update(struct fmeter *fmp) | |
1422 | { | |
1423 | time_t now = get_seconds(); | |
1424 | time_t ticks = now - fmp->time; | |
1425 | ||
1426 | if (ticks == 0) | |
1427 | return; | |
1428 | ||
1429 | ticks = min(FM_MAXTICKS, ticks); | |
1430 | while (ticks-- > 0) | |
1431 | fmp->val = (FM_COEF * fmp->val) / FM_SCALE; | |
1432 | fmp->time = now; | |
1433 | ||
1434 | fmp->val += ((FM_SCALE - FM_COEF) * fmp->cnt) / FM_SCALE; | |
1435 | fmp->cnt = 0; | |
1436 | } | |
1437 | ||
1438 | /* Process any previous ticks, then bump cnt by one (times scale). */ | |
1439 | static void fmeter_markevent(struct fmeter *fmp) | |
1440 | { | |
1441 | spin_lock(&fmp->lock); | |
1442 | fmeter_update(fmp); | |
1443 | fmp->cnt = min(FM_MAXCNT, fmp->cnt + FM_SCALE); | |
1444 | spin_unlock(&fmp->lock); | |
1445 | } | |
1446 | ||
1447 | /* Process any previous ticks, then return current value. */ | |
1448 | static int fmeter_getrate(struct fmeter *fmp) | |
1449 | { | |
1450 | int val; | |
1451 | ||
1452 | spin_lock(&fmp->lock); | |
1453 | fmeter_update(fmp); | |
1454 | val = fmp->val; | |
1455 | spin_unlock(&fmp->lock); | |
1456 | return val; | |
1457 | } | |
1458 | ||
5d21cc2d | 1459 | /* Called by cgroups to determine if a cpuset is usable; cpuset_mutex held */ |
761b3ef5 | 1460 | static int cpuset_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset) |
f780bdb7 | 1461 | { |
2f7ee569 | 1462 | struct cpuset *cs = cgroup_cs(cgrp); |
bb9d97b6 TH |
1463 | struct task_struct *task; |
1464 | int ret; | |
1da177e4 | 1465 | |
5d21cc2d TH |
1466 | mutex_lock(&cpuset_mutex); |
1467 | ||
88fa523b LZ |
1468 | /* |
1469 | * We allow to move tasks into an empty cpuset if sane_behavior | |
1470 | * flag is set. | |
1471 | */ | |
5d21cc2d | 1472 | ret = -ENOSPC; |
88fa523b LZ |
1473 | if (!cgroup_sane_behavior(cgrp) && |
1474 | (cpumask_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed))) | |
5d21cc2d | 1475 | goto out_unlock; |
9985b0ba | 1476 | |
bb9d97b6 TH |
1477 | cgroup_taskset_for_each(task, cgrp, tset) { |
1478 | /* | |
14a40ffc TH |
1479 | * Kthreads which disallow setaffinity shouldn't be moved |
1480 | * to a new cpuset; we don't want to change their cpu | |
1481 | * affinity and isolating such threads by their set of | |
1482 | * allowed nodes is unnecessary. Thus, cpusets are not | |
1483 | * applicable for such threads. This prevents checking for | |
1484 | * success of set_cpus_allowed_ptr() on all attached tasks | |
1485 | * before cpus_allowed may be changed. | |
bb9d97b6 | 1486 | */ |
5d21cc2d | 1487 | ret = -EINVAL; |
14a40ffc | 1488 | if (task->flags & PF_NO_SETAFFINITY) |
5d21cc2d TH |
1489 | goto out_unlock; |
1490 | ret = security_task_setscheduler(task); | |
1491 | if (ret) | |
1492 | goto out_unlock; | |
bb9d97b6 | 1493 | } |
f780bdb7 | 1494 | |
452477fa TH |
1495 | /* |
1496 | * Mark attach is in progress. This makes validate_change() fail | |
1497 | * changes which zero cpus/mems_allowed. | |
1498 | */ | |
1499 | cs->attach_in_progress++; | |
5d21cc2d TH |
1500 | ret = 0; |
1501 | out_unlock: | |
1502 | mutex_unlock(&cpuset_mutex); | |
1503 | return ret; | |
8793d854 | 1504 | } |
f780bdb7 | 1505 | |
452477fa TH |
1506 | static void cpuset_cancel_attach(struct cgroup *cgrp, |
1507 | struct cgroup_taskset *tset) | |
1508 | { | |
5d21cc2d | 1509 | mutex_lock(&cpuset_mutex); |
452477fa | 1510 | cgroup_cs(cgrp)->attach_in_progress--; |
5d21cc2d | 1511 | mutex_unlock(&cpuset_mutex); |
8793d854 | 1512 | } |
1da177e4 | 1513 | |
4e4c9a14 | 1514 | /* |
5d21cc2d | 1515 | * Protected by cpuset_mutex. cpus_attach is used only by cpuset_attach() |
4e4c9a14 TH |
1516 | * but we can't allocate it dynamically there. Define it global and |
1517 | * allocate from cpuset_init(). | |
1518 | */ | |
1519 | static cpumask_var_t cpus_attach; | |
1520 | ||
761b3ef5 | 1521 | static void cpuset_attach(struct cgroup *cgrp, struct cgroup_taskset *tset) |
8793d854 | 1522 | { |
67bd2c59 | 1523 | /* static buf protected by cpuset_mutex */ |
4e4c9a14 | 1524 | static nodemask_t cpuset_attach_nodemask_to; |
8793d854 | 1525 | struct mm_struct *mm; |
bb9d97b6 TH |
1526 | struct task_struct *task; |
1527 | struct task_struct *leader = cgroup_taskset_first(tset); | |
2f7ee569 TH |
1528 | struct cgroup *oldcgrp = cgroup_taskset_cur_cgroup(tset); |
1529 | struct cpuset *cs = cgroup_cs(cgrp); | |
1530 | struct cpuset *oldcs = cgroup_cs(oldcgrp); | |
070b57fc LZ |
1531 | struct cpuset *cpus_cs = effective_cpumask_cpuset(cs); |
1532 | struct cpuset *mems_cs = effective_nodemask_cpuset(cs); | |
22fb52dd | 1533 | |
5d21cc2d TH |
1534 | mutex_lock(&cpuset_mutex); |
1535 | ||
4e4c9a14 TH |
1536 | /* prepare for attach */ |
1537 | if (cs == &top_cpuset) | |
1538 | cpumask_copy(cpus_attach, cpu_possible_mask); | |
1539 | else | |
070b57fc | 1540 | guarantee_online_cpus(cpus_cs, cpus_attach); |
4e4c9a14 | 1541 | |
070b57fc | 1542 | guarantee_online_mems(mems_cs, &cpuset_attach_nodemask_to); |
4e4c9a14 | 1543 | |
bb9d97b6 TH |
1544 | cgroup_taskset_for_each(task, cgrp, tset) { |
1545 | /* | |
1546 | * can_attach beforehand should guarantee that this doesn't | |
1547 | * fail. TODO: have a better way to handle failure here | |
1548 | */ | |
1549 | WARN_ON_ONCE(set_cpus_allowed_ptr(task, cpus_attach)); | |
1550 | ||
1551 | cpuset_change_task_nodemask(task, &cpuset_attach_nodemask_to); | |
1552 | cpuset_update_task_spread_flag(cs, task); | |
1553 | } | |
22fb52dd | 1554 | |
f780bdb7 BB |
1555 | /* |
1556 | * Change mm, possibly for multiple threads in a threadgroup. This is | |
1557 | * expensive and may sleep. | |
1558 | */ | |
f780bdb7 | 1559 | cpuset_attach_nodemask_to = cs->mems_allowed; |
bb9d97b6 | 1560 | mm = get_task_mm(leader); |
4225399a | 1561 | if (mm) { |
070b57fc LZ |
1562 | struct cpuset *mems_oldcs = effective_nodemask_cpuset(oldcs); |
1563 | ||
f780bdb7 | 1564 | mpol_rebind_mm(mm, &cpuset_attach_nodemask_to); |
f047cecf LZ |
1565 | |
1566 | /* | |
1567 | * old_mems_allowed is the same with mems_allowed here, except | |
1568 | * if this task is being moved automatically due to hotplug. | |
1569 | * In that case @mems_allowed has been updated and is empty, | |
1570 | * so @old_mems_allowed is the right nodesets that we migrate | |
1571 | * mm from. | |
1572 | */ | |
1573 | if (is_memory_migrate(cs)) { | |
1574 | cpuset_migrate_mm(mm, &mems_oldcs->old_mems_allowed, | |
f780bdb7 | 1575 | &cpuset_attach_nodemask_to); |
f047cecf | 1576 | } |
4225399a PJ |
1577 | mmput(mm); |
1578 | } | |
452477fa | 1579 | |
33ad801d | 1580 | cs->old_mems_allowed = cpuset_attach_nodemask_to; |
02bb5863 | 1581 | |
452477fa | 1582 | cs->attach_in_progress--; |
e44193d3 LZ |
1583 | if (!cs->attach_in_progress) |
1584 | wake_up(&cpuset_attach_wq); | |
5d21cc2d TH |
1585 | |
1586 | mutex_unlock(&cpuset_mutex); | |
1da177e4 LT |
1587 | } |
1588 | ||
1589 | /* The various types of files and directories in a cpuset file system */ | |
1590 | ||
1591 | typedef enum { | |
45b07ef3 | 1592 | FILE_MEMORY_MIGRATE, |
1da177e4 LT |
1593 | FILE_CPULIST, |
1594 | FILE_MEMLIST, | |
1595 | FILE_CPU_EXCLUSIVE, | |
1596 | FILE_MEM_EXCLUSIVE, | |
78608366 | 1597 | FILE_MEM_HARDWALL, |
029190c5 | 1598 | FILE_SCHED_LOAD_BALANCE, |
1d3504fc | 1599 | FILE_SCHED_RELAX_DOMAIN_LEVEL, |
3e0d98b9 PJ |
1600 | FILE_MEMORY_PRESSURE_ENABLED, |
1601 | FILE_MEMORY_PRESSURE, | |
825a46af PJ |
1602 | FILE_SPREAD_PAGE, |
1603 | FILE_SPREAD_SLAB, | |
1da177e4 LT |
1604 | } cpuset_filetype_t; |
1605 | ||
700fe1ab PM |
1606 | static int cpuset_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val) |
1607 | { | |
700fe1ab PM |
1608 | struct cpuset *cs = cgroup_cs(cgrp); |
1609 | cpuset_filetype_t type = cft->private; | |
5d21cc2d | 1610 | int retval = -ENODEV; |
700fe1ab | 1611 | |
5d21cc2d TH |
1612 | mutex_lock(&cpuset_mutex); |
1613 | if (!is_cpuset_online(cs)) | |
1614 | goto out_unlock; | |
700fe1ab PM |
1615 | |
1616 | switch (type) { | |
1da177e4 | 1617 | case FILE_CPU_EXCLUSIVE: |
700fe1ab | 1618 | retval = update_flag(CS_CPU_EXCLUSIVE, cs, val); |
1da177e4 LT |
1619 | break; |
1620 | case FILE_MEM_EXCLUSIVE: | |
700fe1ab | 1621 | retval = update_flag(CS_MEM_EXCLUSIVE, cs, val); |
1da177e4 | 1622 | break; |
78608366 PM |
1623 | case FILE_MEM_HARDWALL: |
1624 | retval = update_flag(CS_MEM_HARDWALL, cs, val); | |
1625 | break; | |
029190c5 | 1626 | case FILE_SCHED_LOAD_BALANCE: |
700fe1ab | 1627 | retval = update_flag(CS_SCHED_LOAD_BALANCE, cs, val); |
1d3504fc | 1628 | break; |
45b07ef3 | 1629 | case FILE_MEMORY_MIGRATE: |
700fe1ab | 1630 | retval = update_flag(CS_MEMORY_MIGRATE, cs, val); |
45b07ef3 | 1631 | break; |
3e0d98b9 | 1632 | case FILE_MEMORY_PRESSURE_ENABLED: |
700fe1ab | 1633 | cpuset_memory_pressure_enabled = !!val; |
3e0d98b9 PJ |
1634 | break; |
1635 | case FILE_MEMORY_PRESSURE: | |
1636 | retval = -EACCES; | |
1637 | break; | |
825a46af | 1638 | case FILE_SPREAD_PAGE: |
700fe1ab | 1639 | retval = update_flag(CS_SPREAD_PAGE, cs, val); |
825a46af PJ |
1640 | break; |
1641 | case FILE_SPREAD_SLAB: | |
700fe1ab | 1642 | retval = update_flag(CS_SPREAD_SLAB, cs, val); |
825a46af | 1643 | break; |
1da177e4 LT |
1644 | default: |
1645 | retval = -EINVAL; | |
700fe1ab | 1646 | break; |
1da177e4 | 1647 | } |
5d21cc2d TH |
1648 | out_unlock: |
1649 | mutex_unlock(&cpuset_mutex); | |
1da177e4 LT |
1650 | return retval; |
1651 | } | |
1652 | ||
5be7a479 PM |
1653 | static int cpuset_write_s64(struct cgroup *cgrp, struct cftype *cft, s64 val) |
1654 | { | |
5be7a479 PM |
1655 | struct cpuset *cs = cgroup_cs(cgrp); |
1656 | cpuset_filetype_t type = cft->private; | |
5d21cc2d | 1657 | int retval = -ENODEV; |
5be7a479 | 1658 | |
5d21cc2d TH |
1659 | mutex_lock(&cpuset_mutex); |
1660 | if (!is_cpuset_online(cs)) | |
1661 | goto out_unlock; | |
e3712395 | 1662 | |
5be7a479 PM |
1663 | switch (type) { |
1664 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1665 | retval = update_relax_domain_level(cs, val); | |
1666 | break; | |
1667 | default: | |
1668 | retval = -EINVAL; | |
1669 | break; | |
1670 | } | |
5d21cc2d TH |
1671 | out_unlock: |
1672 | mutex_unlock(&cpuset_mutex); | |
5be7a479 PM |
1673 | return retval; |
1674 | } | |
1675 | ||
e3712395 PM |
1676 | /* |
1677 | * Common handling for a write to a "cpus" or "mems" file. | |
1678 | */ | |
1679 | static int cpuset_write_resmask(struct cgroup *cgrp, struct cftype *cft, | |
1680 | const char *buf) | |
1681 | { | |
645fcc9d LZ |
1682 | struct cpuset *cs = cgroup_cs(cgrp); |
1683 | struct cpuset *trialcs; | |
5d21cc2d | 1684 | int retval = -ENODEV; |
e3712395 | 1685 | |
3a5a6d0c TH |
1686 | /* |
1687 | * CPU or memory hotunplug may leave @cs w/o any execution | |
1688 | * resources, in which case the hotplug code asynchronously updates | |
1689 | * configuration and transfers all tasks to the nearest ancestor | |
1690 | * which can execute. | |
1691 | * | |
1692 | * As writes to "cpus" or "mems" may restore @cs's execution | |
1693 | * resources, wait for the previously scheduled operations before | |
1694 | * proceeding, so that we don't end up keep removing tasks added | |
1695 | * after execution capability is restored. | |
1696 | */ | |
1697 | flush_work(&cpuset_hotplug_work); | |
1698 | ||
5d21cc2d TH |
1699 | mutex_lock(&cpuset_mutex); |
1700 | if (!is_cpuset_online(cs)) | |
1701 | goto out_unlock; | |
e3712395 | 1702 | |
645fcc9d | 1703 | trialcs = alloc_trial_cpuset(cs); |
b75f38d6 LZ |
1704 | if (!trialcs) { |
1705 | retval = -ENOMEM; | |
5d21cc2d | 1706 | goto out_unlock; |
b75f38d6 | 1707 | } |
645fcc9d | 1708 | |
e3712395 PM |
1709 | switch (cft->private) { |
1710 | case FILE_CPULIST: | |
645fcc9d | 1711 | retval = update_cpumask(cs, trialcs, buf); |
e3712395 PM |
1712 | break; |
1713 | case FILE_MEMLIST: | |
645fcc9d | 1714 | retval = update_nodemask(cs, trialcs, buf); |
e3712395 PM |
1715 | break; |
1716 | default: | |
1717 | retval = -EINVAL; | |
1718 | break; | |
1719 | } | |
645fcc9d LZ |
1720 | |
1721 | free_trial_cpuset(trialcs); | |
5d21cc2d TH |
1722 | out_unlock: |
1723 | mutex_unlock(&cpuset_mutex); | |
e3712395 PM |
1724 | return retval; |
1725 | } | |
1726 | ||
1da177e4 LT |
1727 | /* |
1728 | * These ascii lists should be read in a single call, by using a user | |
1729 | * buffer large enough to hold the entire map. If read in smaller | |
1730 | * chunks, there is no guarantee of atomicity. Since the display format | |
1731 | * used, list of ranges of sequential numbers, is variable length, | |
1732 | * and since these maps can change value dynamically, one could read | |
1733 | * gibberish by doing partial reads while a list was changing. | |
1734 | * A single large read to a buffer that crosses a page boundary is | |
1735 | * ok, because the result being copied to user land is not recomputed | |
1736 | * across a page fault. | |
1737 | */ | |
1738 | ||
9303e0c4 | 1739 | static size_t cpuset_sprintf_cpulist(char *page, struct cpuset *cs) |
1da177e4 | 1740 | { |
9303e0c4 | 1741 | size_t count; |
1da177e4 | 1742 | |
3d3f26a7 | 1743 | mutex_lock(&callback_mutex); |
9303e0c4 | 1744 | count = cpulist_scnprintf(page, PAGE_SIZE, cs->cpus_allowed); |
3d3f26a7 | 1745 | mutex_unlock(&callback_mutex); |
1da177e4 | 1746 | |
9303e0c4 | 1747 | return count; |
1da177e4 LT |
1748 | } |
1749 | ||
9303e0c4 | 1750 | static size_t cpuset_sprintf_memlist(char *page, struct cpuset *cs) |
1da177e4 | 1751 | { |
9303e0c4 | 1752 | size_t count; |
1da177e4 | 1753 | |
3d3f26a7 | 1754 | mutex_lock(&callback_mutex); |
9303e0c4 | 1755 | count = nodelist_scnprintf(page, PAGE_SIZE, cs->mems_allowed); |
3d3f26a7 | 1756 | mutex_unlock(&callback_mutex); |
1da177e4 | 1757 | |
9303e0c4 | 1758 | return count; |
1da177e4 LT |
1759 | } |
1760 | ||
c9e5fe66 | 1761 | static ssize_t cpuset_common_file_read(struct cgroup *cgrp, |
8793d854 PM |
1762 | struct cftype *cft, |
1763 | struct file *file, | |
1764 | char __user *buf, | |
1765 | size_t nbytes, loff_t *ppos) | |
1da177e4 | 1766 | { |
c9e5fe66 | 1767 | struct cpuset *cs = cgroup_cs(cgrp); |
1da177e4 LT |
1768 | cpuset_filetype_t type = cft->private; |
1769 | char *page; | |
1770 | ssize_t retval = 0; | |
1771 | char *s; | |
1da177e4 | 1772 | |
e12ba74d | 1773 | if (!(page = (char *)__get_free_page(GFP_TEMPORARY))) |
1da177e4 LT |
1774 | return -ENOMEM; |
1775 | ||
1776 | s = page; | |
1777 | ||
1778 | switch (type) { | |
1779 | case FILE_CPULIST: | |
1780 | s += cpuset_sprintf_cpulist(s, cs); | |
1781 | break; | |
1782 | case FILE_MEMLIST: | |
1783 | s += cpuset_sprintf_memlist(s, cs); | |
1784 | break; | |
1da177e4 LT |
1785 | default: |
1786 | retval = -EINVAL; | |
1787 | goto out; | |
1788 | } | |
1789 | *s++ = '\n'; | |
1da177e4 | 1790 | |
eacaa1f5 | 1791 | retval = simple_read_from_buffer(buf, nbytes, ppos, page, s - page); |
1da177e4 LT |
1792 | out: |
1793 | free_page((unsigned long)page); | |
1794 | return retval; | |
1795 | } | |
1796 | ||
c9e5fe66 | 1797 | static u64 cpuset_read_u64(struct cgroup *cgrp, struct cftype *cft) |
700fe1ab | 1798 | { |
c9e5fe66 | 1799 | struct cpuset *cs = cgroup_cs(cgrp); |
700fe1ab PM |
1800 | cpuset_filetype_t type = cft->private; |
1801 | switch (type) { | |
1802 | case FILE_CPU_EXCLUSIVE: | |
1803 | return is_cpu_exclusive(cs); | |
1804 | case FILE_MEM_EXCLUSIVE: | |
1805 | return is_mem_exclusive(cs); | |
78608366 PM |
1806 | case FILE_MEM_HARDWALL: |
1807 | return is_mem_hardwall(cs); | |
700fe1ab PM |
1808 | case FILE_SCHED_LOAD_BALANCE: |
1809 | return is_sched_load_balance(cs); | |
1810 | case FILE_MEMORY_MIGRATE: | |
1811 | return is_memory_migrate(cs); | |
1812 | case FILE_MEMORY_PRESSURE_ENABLED: | |
1813 | return cpuset_memory_pressure_enabled; | |
1814 | case FILE_MEMORY_PRESSURE: | |
1815 | return fmeter_getrate(&cs->fmeter); | |
1816 | case FILE_SPREAD_PAGE: | |
1817 | return is_spread_page(cs); | |
1818 | case FILE_SPREAD_SLAB: | |
1819 | return is_spread_slab(cs); | |
1820 | default: | |
1821 | BUG(); | |
1822 | } | |
cf417141 MK |
1823 | |
1824 | /* Unreachable but makes gcc happy */ | |
1825 | return 0; | |
700fe1ab | 1826 | } |
1da177e4 | 1827 | |
c9e5fe66 | 1828 | static s64 cpuset_read_s64(struct cgroup *cgrp, struct cftype *cft) |
5be7a479 | 1829 | { |
c9e5fe66 | 1830 | struct cpuset *cs = cgroup_cs(cgrp); |
5be7a479 PM |
1831 | cpuset_filetype_t type = cft->private; |
1832 | switch (type) { | |
1833 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1834 | return cs->relax_domain_level; | |
1835 | default: | |
1836 | BUG(); | |
1837 | } | |
cf417141 MK |
1838 | |
1839 | /* Unrechable but makes gcc happy */ | |
1840 | return 0; | |
5be7a479 PM |
1841 | } |
1842 | ||
1da177e4 LT |
1843 | |
1844 | /* | |
1845 | * for the common functions, 'private' gives the type of file | |
1846 | */ | |
1847 | ||
addf2c73 PM |
1848 | static struct cftype files[] = { |
1849 | { | |
1850 | .name = "cpus", | |
1851 | .read = cpuset_common_file_read, | |
e3712395 PM |
1852 | .write_string = cpuset_write_resmask, |
1853 | .max_write_len = (100U + 6 * NR_CPUS), | |
addf2c73 PM |
1854 | .private = FILE_CPULIST, |
1855 | }, | |
1856 | ||
1857 | { | |
1858 | .name = "mems", | |
1859 | .read = cpuset_common_file_read, | |
e3712395 PM |
1860 | .write_string = cpuset_write_resmask, |
1861 | .max_write_len = (100U + 6 * MAX_NUMNODES), | |
addf2c73 PM |
1862 | .private = FILE_MEMLIST, |
1863 | }, | |
1864 | ||
1865 | { | |
1866 | .name = "cpu_exclusive", | |
1867 | .read_u64 = cpuset_read_u64, | |
1868 | .write_u64 = cpuset_write_u64, | |
1869 | .private = FILE_CPU_EXCLUSIVE, | |
1870 | }, | |
1871 | ||
1872 | { | |
1873 | .name = "mem_exclusive", | |
1874 | .read_u64 = cpuset_read_u64, | |
1875 | .write_u64 = cpuset_write_u64, | |
1876 | .private = FILE_MEM_EXCLUSIVE, | |
1877 | }, | |
1878 | ||
78608366 PM |
1879 | { |
1880 | .name = "mem_hardwall", | |
1881 | .read_u64 = cpuset_read_u64, | |
1882 | .write_u64 = cpuset_write_u64, | |
1883 | .private = FILE_MEM_HARDWALL, | |
1884 | }, | |
1885 | ||
addf2c73 PM |
1886 | { |
1887 | .name = "sched_load_balance", | |
1888 | .read_u64 = cpuset_read_u64, | |
1889 | .write_u64 = cpuset_write_u64, | |
1890 | .private = FILE_SCHED_LOAD_BALANCE, | |
1891 | }, | |
1892 | ||
1893 | { | |
1894 | .name = "sched_relax_domain_level", | |
5be7a479 PM |
1895 | .read_s64 = cpuset_read_s64, |
1896 | .write_s64 = cpuset_write_s64, | |
addf2c73 PM |
1897 | .private = FILE_SCHED_RELAX_DOMAIN_LEVEL, |
1898 | }, | |
1899 | ||
1900 | { | |
1901 | .name = "memory_migrate", | |
1902 | .read_u64 = cpuset_read_u64, | |
1903 | .write_u64 = cpuset_write_u64, | |
1904 | .private = FILE_MEMORY_MIGRATE, | |
1905 | }, | |
1906 | ||
1907 | { | |
1908 | .name = "memory_pressure", | |
1909 | .read_u64 = cpuset_read_u64, | |
1910 | .write_u64 = cpuset_write_u64, | |
1911 | .private = FILE_MEMORY_PRESSURE, | |
099fca32 | 1912 | .mode = S_IRUGO, |
addf2c73 PM |
1913 | }, |
1914 | ||
1915 | { | |
1916 | .name = "memory_spread_page", | |
1917 | .read_u64 = cpuset_read_u64, | |
1918 | .write_u64 = cpuset_write_u64, | |
1919 | .private = FILE_SPREAD_PAGE, | |
1920 | }, | |
1921 | ||
1922 | { | |
1923 | .name = "memory_spread_slab", | |
1924 | .read_u64 = cpuset_read_u64, | |
1925 | .write_u64 = cpuset_write_u64, | |
1926 | .private = FILE_SPREAD_SLAB, | |
1927 | }, | |
3e0d98b9 | 1928 | |
4baf6e33 TH |
1929 | { |
1930 | .name = "memory_pressure_enabled", | |
1931 | .flags = CFTYPE_ONLY_ON_ROOT, | |
1932 | .read_u64 = cpuset_read_u64, | |
1933 | .write_u64 = cpuset_write_u64, | |
1934 | .private = FILE_MEMORY_PRESSURE_ENABLED, | |
1935 | }, | |
1da177e4 | 1936 | |
4baf6e33 TH |
1937 | { } /* terminate */ |
1938 | }; | |
1da177e4 LT |
1939 | |
1940 | /* | |
92fb9748 | 1941 | * cpuset_css_alloc - allocate a cpuset css |
c9e5fe66 | 1942 | * cgrp: control group that the new cpuset will be part of |
1da177e4 LT |
1943 | */ |
1944 | ||
c9e5fe66 | 1945 | static struct cgroup_subsys_state *cpuset_css_alloc(struct cgroup *cgrp) |
1da177e4 | 1946 | { |
c8f699bb | 1947 | struct cpuset *cs; |
1da177e4 | 1948 | |
c9e5fe66 | 1949 | if (!cgrp->parent) |
8793d854 | 1950 | return &top_cpuset.css; |
033fa1c5 | 1951 | |
c8f699bb | 1952 | cs = kzalloc(sizeof(*cs), GFP_KERNEL); |
1da177e4 | 1953 | if (!cs) |
8793d854 | 1954 | return ERR_PTR(-ENOMEM); |
300ed6cb LZ |
1955 | if (!alloc_cpumask_var(&cs->cpus_allowed, GFP_KERNEL)) { |
1956 | kfree(cs); | |
1957 | return ERR_PTR(-ENOMEM); | |
1958 | } | |
1da177e4 | 1959 | |
029190c5 | 1960 | set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); |
300ed6cb | 1961 | cpumask_clear(cs->cpus_allowed); |
f9a86fcb | 1962 | nodes_clear(cs->mems_allowed); |
3e0d98b9 | 1963 | fmeter_init(&cs->fmeter); |
1d3504fc | 1964 | cs->relax_domain_level = -1; |
1da177e4 | 1965 | |
c8f699bb TH |
1966 | return &cs->css; |
1967 | } | |
1968 | ||
1969 | static int cpuset_css_online(struct cgroup *cgrp) | |
1970 | { | |
1971 | struct cpuset *cs = cgroup_cs(cgrp); | |
c431069f | 1972 | struct cpuset *parent = parent_cs(cs); |
ae8086ce | 1973 | struct cpuset *tmp_cs; |
6f4b7e63 | 1974 | struct cgroup *pos_cgrp; |
c8f699bb TH |
1975 | |
1976 | if (!parent) | |
1977 | return 0; | |
1978 | ||
5d21cc2d TH |
1979 | mutex_lock(&cpuset_mutex); |
1980 | ||
efeb77b2 | 1981 | set_bit(CS_ONLINE, &cs->flags); |
c8f699bb TH |
1982 | if (is_spread_page(parent)) |
1983 | set_bit(CS_SPREAD_PAGE, &cs->flags); | |
1984 | if (is_spread_slab(parent)) | |
1985 | set_bit(CS_SPREAD_SLAB, &cs->flags); | |
1da177e4 | 1986 | |
202f72d5 | 1987 | number_of_cpusets++; |
033fa1c5 | 1988 | |
c8f699bb | 1989 | if (!test_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags)) |
5d21cc2d | 1990 | goto out_unlock; |
033fa1c5 TH |
1991 | |
1992 | /* | |
1993 | * Clone @parent's configuration if CGRP_CPUSET_CLONE_CHILDREN is | |
1994 | * set. This flag handling is implemented in cgroup core for | |
1995 | * histrical reasons - the flag may be specified during mount. | |
1996 | * | |
1997 | * Currently, if any sibling cpusets have exclusive cpus or mem, we | |
1998 | * refuse to clone the configuration - thereby refusing the task to | |
1999 | * be entered, and as a result refusing the sys_unshare() or | |
2000 | * clone() which initiated it. If this becomes a problem for some | |
2001 | * users who wish to allow that scenario, then this could be | |
2002 | * changed to grant parent->cpus_allowed-sibling_cpus_exclusive | |
2003 | * (and likewise for mems) to the new cgroup. | |
2004 | */ | |
ae8086ce | 2005 | rcu_read_lock(); |
6f4b7e63 | 2006 | cpuset_for_each_child(tmp_cs, pos_cgrp, parent) { |
ae8086ce TH |
2007 | if (is_mem_exclusive(tmp_cs) || is_cpu_exclusive(tmp_cs)) { |
2008 | rcu_read_unlock(); | |
5d21cc2d | 2009 | goto out_unlock; |
ae8086ce | 2010 | } |
033fa1c5 | 2011 | } |
ae8086ce | 2012 | rcu_read_unlock(); |
033fa1c5 TH |
2013 | |
2014 | mutex_lock(&callback_mutex); | |
2015 | cs->mems_allowed = parent->mems_allowed; | |
2016 | cpumask_copy(cs->cpus_allowed, parent->cpus_allowed); | |
2017 | mutex_unlock(&callback_mutex); | |
5d21cc2d TH |
2018 | out_unlock: |
2019 | mutex_unlock(&cpuset_mutex); | |
c8f699bb TH |
2020 | return 0; |
2021 | } | |
2022 | ||
0b9e6965 ZH |
2023 | /* |
2024 | * If the cpuset being removed has its flag 'sched_load_balance' | |
2025 | * enabled, then simulate turning sched_load_balance off, which | |
2026 | * will call rebuild_sched_domains_locked(). | |
2027 | */ | |
2028 | ||
c8f699bb TH |
2029 | static void cpuset_css_offline(struct cgroup *cgrp) |
2030 | { | |
2031 | struct cpuset *cs = cgroup_cs(cgrp); | |
2032 | ||
5d21cc2d | 2033 | mutex_lock(&cpuset_mutex); |
c8f699bb TH |
2034 | |
2035 | if (is_sched_load_balance(cs)) | |
2036 | update_flag(CS_SCHED_LOAD_BALANCE, cs, 0); | |
2037 | ||
2038 | number_of_cpusets--; | |
efeb77b2 | 2039 | clear_bit(CS_ONLINE, &cs->flags); |
c8f699bb | 2040 | |
5d21cc2d | 2041 | mutex_unlock(&cpuset_mutex); |
1da177e4 LT |
2042 | } |
2043 | ||
c9e5fe66 | 2044 | static void cpuset_css_free(struct cgroup *cgrp) |
1da177e4 | 2045 | { |
c9e5fe66 | 2046 | struct cpuset *cs = cgroup_cs(cgrp); |
1da177e4 | 2047 | |
300ed6cb | 2048 | free_cpumask_var(cs->cpus_allowed); |
8793d854 | 2049 | kfree(cs); |
1da177e4 LT |
2050 | } |
2051 | ||
8793d854 PM |
2052 | struct cgroup_subsys cpuset_subsys = { |
2053 | .name = "cpuset", | |
92fb9748 | 2054 | .css_alloc = cpuset_css_alloc, |
c8f699bb TH |
2055 | .css_online = cpuset_css_online, |
2056 | .css_offline = cpuset_css_offline, | |
92fb9748 | 2057 | .css_free = cpuset_css_free, |
8793d854 | 2058 | .can_attach = cpuset_can_attach, |
452477fa | 2059 | .cancel_attach = cpuset_cancel_attach, |
8793d854 | 2060 | .attach = cpuset_attach, |
8793d854 | 2061 | .subsys_id = cpuset_subsys_id, |
4baf6e33 | 2062 | .base_cftypes = files, |
8793d854 PM |
2063 | .early_init = 1, |
2064 | }; | |
2065 | ||
1da177e4 LT |
2066 | /** |
2067 | * cpuset_init - initialize cpusets at system boot | |
2068 | * | |
2069 | * Description: Initialize top_cpuset and the cpuset internal file system, | |
2070 | **/ | |
2071 | ||
2072 | int __init cpuset_init(void) | |
2073 | { | |
8793d854 | 2074 | int err = 0; |
1da177e4 | 2075 | |
58568d2a MX |
2076 | if (!alloc_cpumask_var(&top_cpuset.cpus_allowed, GFP_KERNEL)) |
2077 | BUG(); | |
2078 | ||
300ed6cb | 2079 | cpumask_setall(top_cpuset.cpus_allowed); |
f9a86fcb | 2080 | nodes_setall(top_cpuset.mems_allowed); |
1da177e4 | 2081 | |
3e0d98b9 | 2082 | fmeter_init(&top_cpuset.fmeter); |
029190c5 | 2083 | set_bit(CS_SCHED_LOAD_BALANCE, &top_cpuset.flags); |
1d3504fc | 2084 | top_cpuset.relax_domain_level = -1; |
1da177e4 | 2085 | |
1da177e4 LT |
2086 | err = register_filesystem(&cpuset_fs_type); |
2087 | if (err < 0) | |
8793d854 PM |
2088 | return err; |
2089 | ||
2341d1b6 LZ |
2090 | if (!alloc_cpumask_var(&cpus_attach, GFP_KERNEL)) |
2091 | BUG(); | |
2092 | ||
202f72d5 | 2093 | number_of_cpusets = 1; |
8793d854 | 2094 | return 0; |
1da177e4 LT |
2095 | } |
2096 | ||
b1aac8bb | 2097 | /* |
cf417141 | 2098 | * If CPU and/or memory hotplug handlers, below, unplug any CPUs |
b1aac8bb PJ |
2099 | * or memory nodes, we need to walk over the cpuset hierarchy, |
2100 | * removing that CPU or node from all cpusets. If this removes the | |
956db3ca CW |
2101 | * last CPU or node from a cpuset, then move the tasks in the empty |
2102 | * cpuset to its next-highest non-empty parent. | |
b1aac8bb | 2103 | */ |
956db3ca CW |
2104 | static void remove_tasks_in_empty_cpuset(struct cpuset *cs) |
2105 | { | |
2106 | struct cpuset *parent; | |
2107 | ||
956db3ca CW |
2108 | /* |
2109 | * Find its next-highest non-empty parent, (top cpuset | |
2110 | * has online cpus, so can't be empty). | |
2111 | */ | |
c431069f | 2112 | parent = parent_cs(cs); |
300ed6cb | 2113 | while (cpumask_empty(parent->cpus_allowed) || |
b4501295 | 2114 | nodes_empty(parent->mems_allowed)) |
c431069f | 2115 | parent = parent_cs(parent); |
956db3ca | 2116 | |
8cc99345 TH |
2117 | if (cgroup_transfer_tasks(parent->css.cgroup, cs->css.cgroup)) { |
2118 | rcu_read_lock(); | |
2119 | printk(KERN_ERR "cpuset: failed to transfer tasks out of empty cpuset %s\n", | |
2120 | cgroup_name(cs->css.cgroup)); | |
2121 | rcu_read_unlock(); | |
2122 | } | |
956db3ca CW |
2123 | } |
2124 | ||
deb7aa30 | 2125 | /** |
388afd85 | 2126 | * cpuset_hotplug_update_tasks - update tasks in a cpuset for hotunplug |
deb7aa30 | 2127 | * @cs: cpuset in interest |
956db3ca | 2128 | * |
deb7aa30 TH |
2129 | * Compare @cs's cpu and mem masks against top_cpuset and if some have gone |
2130 | * offline, update @cs accordingly. If @cs ends up with no CPU or memory, | |
2131 | * all its tasks are moved to the nearest ancestor with both resources. | |
80d1fa64 | 2132 | */ |
388afd85 | 2133 | static void cpuset_hotplug_update_tasks(struct cpuset *cs) |
80d1fa64 | 2134 | { |
deb7aa30 | 2135 | static cpumask_t off_cpus; |
33ad801d | 2136 | static nodemask_t off_mems; |
5d21cc2d | 2137 | bool is_empty; |
5c5cc623 | 2138 | bool sane = cgroup_sane_behavior(cs->css.cgroup); |
80d1fa64 | 2139 | |
e44193d3 LZ |
2140 | retry: |
2141 | wait_event(cpuset_attach_wq, cs->attach_in_progress == 0); | |
80d1fa64 | 2142 | |
5d21cc2d | 2143 | mutex_lock(&cpuset_mutex); |
7ddf96b0 | 2144 | |
e44193d3 LZ |
2145 | /* |
2146 | * We have raced with task attaching. We wait until attaching | |
2147 | * is finished, so we won't attach a task to an empty cpuset. | |
2148 | */ | |
2149 | if (cs->attach_in_progress) { | |
2150 | mutex_unlock(&cpuset_mutex); | |
2151 | goto retry; | |
2152 | } | |
2153 | ||
deb7aa30 TH |
2154 | cpumask_andnot(&off_cpus, cs->cpus_allowed, top_cpuset.cpus_allowed); |
2155 | nodes_andnot(off_mems, cs->mems_allowed, top_cpuset.mems_allowed); | |
80d1fa64 | 2156 | |
5c5cc623 LZ |
2157 | mutex_lock(&callback_mutex); |
2158 | cpumask_andnot(cs->cpus_allowed, cs->cpus_allowed, &off_cpus); | |
2159 | mutex_unlock(&callback_mutex); | |
2160 | ||
2161 | /* | |
2162 | * If sane_behavior flag is set, we need to update tasks' cpumask | |
f047cecf LZ |
2163 | * for empty cpuset to take on ancestor's cpumask. Otherwise, don't |
2164 | * call update_tasks_cpumask() if the cpuset becomes empty, as | |
2165 | * the tasks in it will be migrated to an ancestor. | |
5c5cc623 LZ |
2166 | */ |
2167 | if ((sane && cpumask_empty(cs->cpus_allowed)) || | |
f047cecf | 2168 | (!cpumask_empty(&off_cpus) && !cpumask_empty(cs->cpus_allowed))) |
deb7aa30 | 2169 | update_tasks_cpumask(cs, NULL); |
80d1fa64 | 2170 | |
5c5cc623 LZ |
2171 | mutex_lock(&callback_mutex); |
2172 | nodes_andnot(cs->mems_allowed, cs->mems_allowed, off_mems); | |
2173 | mutex_unlock(&callback_mutex); | |
2174 | ||
2175 | /* | |
2176 | * If sane_behavior flag is set, we need to update tasks' nodemask | |
f047cecf LZ |
2177 | * for empty cpuset to take on ancestor's nodemask. Otherwise, don't |
2178 | * call update_tasks_nodemask() if the cpuset becomes empty, as | |
2179 | * the tasks in it will be migratd to an ancestor. | |
5c5cc623 LZ |
2180 | */ |
2181 | if ((sane && nodes_empty(cs->mems_allowed)) || | |
f047cecf | 2182 | (!nodes_empty(off_mems) && !nodes_empty(cs->mems_allowed))) |
33ad801d | 2183 | update_tasks_nodemask(cs, NULL); |
deb7aa30 | 2184 | |
5d21cc2d TH |
2185 | is_empty = cpumask_empty(cs->cpus_allowed) || |
2186 | nodes_empty(cs->mems_allowed); | |
8d033948 | 2187 | |
5d21cc2d TH |
2188 | mutex_unlock(&cpuset_mutex); |
2189 | ||
2190 | /* | |
5c5cc623 LZ |
2191 | * If sane_behavior flag is set, we'll keep tasks in empty cpusets. |
2192 | * | |
2193 | * Otherwise move tasks to the nearest ancestor with execution | |
2194 | * resources. This is full cgroup operation which will | |
5d21cc2d TH |
2195 | * also call back into cpuset. Should be done outside any lock. |
2196 | */ | |
5c5cc623 | 2197 | if (!sane && is_empty) |
5d21cc2d | 2198 | remove_tasks_in_empty_cpuset(cs); |
b1aac8bb PJ |
2199 | } |
2200 | ||
deb7aa30 | 2201 | /** |
3a5a6d0c | 2202 | * cpuset_hotplug_workfn - handle CPU/memory hotunplug for a cpuset |
956db3ca | 2203 | * |
deb7aa30 TH |
2204 | * This function is called after either CPU or memory configuration has |
2205 | * changed and updates cpuset accordingly. The top_cpuset is always | |
2206 | * synchronized to cpu_active_mask and N_MEMORY, which is necessary in | |
2207 | * order to make cpusets transparent (of no affect) on systems that are | |
2208 | * actively using CPU hotplug but making no active use of cpusets. | |
956db3ca | 2209 | * |
deb7aa30 | 2210 | * Non-root cpusets are only affected by offlining. If any CPUs or memory |
388afd85 LZ |
2211 | * nodes have been taken down, cpuset_hotplug_update_tasks() is invoked on |
2212 | * all descendants. | |
956db3ca | 2213 | * |
deb7aa30 TH |
2214 | * Note that CPU offlining during suspend is ignored. We don't modify |
2215 | * cpusets across suspend/resume cycles at all. | |
956db3ca | 2216 | */ |
3a5a6d0c | 2217 | static void cpuset_hotplug_workfn(struct work_struct *work) |
b1aac8bb | 2218 | { |
5c5cc623 LZ |
2219 | static cpumask_t new_cpus; |
2220 | static nodemask_t new_mems; | |
deb7aa30 | 2221 | bool cpus_updated, mems_updated; |
b1aac8bb | 2222 | |
5d21cc2d | 2223 | mutex_lock(&cpuset_mutex); |
956db3ca | 2224 | |
deb7aa30 TH |
2225 | /* fetch the available cpus/mems and find out which changed how */ |
2226 | cpumask_copy(&new_cpus, cpu_active_mask); | |
2227 | new_mems = node_states[N_MEMORY]; | |
7ddf96b0 | 2228 | |
deb7aa30 | 2229 | cpus_updated = !cpumask_equal(top_cpuset.cpus_allowed, &new_cpus); |
deb7aa30 | 2230 | mems_updated = !nodes_equal(top_cpuset.mems_allowed, new_mems); |
7ddf96b0 | 2231 | |
deb7aa30 TH |
2232 | /* synchronize cpus_allowed to cpu_active_mask */ |
2233 | if (cpus_updated) { | |
2234 | mutex_lock(&callback_mutex); | |
2235 | cpumask_copy(top_cpuset.cpus_allowed, &new_cpus); | |
2236 | mutex_unlock(&callback_mutex); | |
2237 | /* we don't mess with cpumasks of tasks in top_cpuset */ | |
2238 | } | |
b4501295 | 2239 | |
deb7aa30 TH |
2240 | /* synchronize mems_allowed to N_MEMORY */ |
2241 | if (mems_updated) { | |
deb7aa30 TH |
2242 | mutex_lock(&callback_mutex); |
2243 | top_cpuset.mems_allowed = new_mems; | |
2244 | mutex_unlock(&callback_mutex); | |
33ad801d | 2245 | update_tasks_nodemask(&top_cpuset, NULL); |
deb7aa30 | 2246 | } |
b4501295 | 2247 | |
388afd85 LZ |
2248 | mutex_unlock(&cpuset_mutex); |
2249 | ||
5c5cc623 LZ |
2250 | /* if cpus or mems changed, we need to propagate to descendants */ |
2251 | if (cpus_updated || mems_updated) { | |
deb7aa30 | 2252 | struct cpuset *cs; |
fc560a26 | 2253 | struct cgroup *pos_cgrp; |
f9b4fb8d | 2254 | |
fc560a26 | 2255 | rcu_read_lock(); |
388afd85 LZ |
2256 | cpuset_for_each_descendant_pre(cs, pos_cgrp, &top_cpuset) { |
2257 | if (!css_tryget(&cs->css)) | |
2258 | continue; | |
2259 | rcu_read_unlock(); | |
7ddf96b0 | 2260 | |
388afd85 | 2261 | cpuset_hotplug_update_tasks(cs); |
b4501295 | 2262 | |
388afd85 LZ |
2263 | rcu_read_lock(); |
2264 | css_put(&cs->css); | |
2265 | } | |
2266 | rcu_read_unlock(); | |
2267 | } | |
8d033948 | 2268 | |
deb7aa30 | 2269 | /* rebuild sched domains if cpus_allowed has changed */ |
e0e80a02 LZ |
2270 | if (cpus_updated) |
2271 | rebuild_sched_domains(); | |
b1aac8bb PJ |
2272 | } |
2273 | ||
7ddf96b0 | 2274 | void cpuset_update_active_cpus(bool cpu_online) |
4c4d50f7 | 2275 | { |
3a5a6d0c TH |
2276 | /* |
2277 | * We're inside cpu hotplug critical region which usually nests | |
2278 | * inside cgroup synchronization. Bounce actual hotplug processing | |
2279 | * to a work item to avoid reverse locking order. | |
2280 | * | |
2281 | * We still need to do partition_sched_domains() synchronously; | |
2282 | * otherwise, the scheduler will get confused and put tasks to the | |
2283 | * dead CPU. Fall back to the default single domain. | |
2284 | * cpuset_hotplug_workfn() will rebuild it as necessary. | |
2285 | */ | |
2286 | partition_sched_domains(1, NULL, NULL); | |
2287 | schedule_work(&cpuset_hotplug_work); | |
4c4d50f7 | 2288 | } |
4c4d50f7 | 2289 | |
38837fc7 | 2290 | /* |
38d7bee9 LJ |
2291 | * Keep top_cpuset.mems_allowed tracking node_states[N_MEMORY]. |
2292 | * Call this routine anytime after node_states[N_MEMORY] changes. | |
a1cd2b13 | 2293 | * See cpuset_update_active_cpus() for CPU hotplug handling. |
38837fc7 | 2294 | */ |
f481891f MX |
2295 | static int cpuset_track_online_nodes(struct notifier_block *self, |
2296 | unsigned long action, void *arg) | |
38837fc7 | 2297 | { |
3a5a6d0c | 2298 | schedule_work(&cpuset_hotplug_work); |
f481891f | 2299 | return NOTIFY_OK; |
38837fc7 | 2300 | } |
d8f10cb3 AM |
2301 | |
2302 | static struct notifier_block cpuset_track_online_nodes_nb = { | |
2303 | .notifier_call = cpuset_track_online_nodes, | |
2304 | .priority = 10, /* ??! */ | |
2305 | }; | |
38837fc7 | 2306 | |
1da177e4 LT |
2307 | /** |
2308 | * cpuset_init_smp - initialize cpus_allowed | |
2309 | * | |
2310 | * Description: Finish top cpuset after cpu, node maps are initialized | |
d8f10cb3 | 2311 | */ |
1da177e4 LT |
2312 | void __init cpuset_init_smp(void) |
2313 | { | |
6ad4c188 | 2314 | cpumask_copy(top_cpuset.cpus_allowed, cpu_active_mask); |
38d7bee9 | 2315 | top_cpuset.mems_allowed = node_states[N_MEMORY]; |
33ad801d | 2316 | top_cpuset.old_mems_allowed = top_cpuset.mems_allowed; |
4c4d50f7 | 2317 | |
d8f10cb3 | 2318 | register_hotmemory_notifier(&cpuset_track_online_nodes_nb); |
1da177e4 LT |
2319 | } |
2320 | ||
2321 | /** | |
1da177e4 LT |
2322 | * cpuset_cpus_allowed - return cpus_allowed mask from a tasks cpuset. |
2323 | * @tsk: pointer to task_struct from which to obtain cpuset->cpus_allowed. | |
6af866af | 2324 | * @pmask: pointer to struct cpumask variable to receive cpus_allowed set. |
1da177e4 | 2325 | * |
300ed6cb | 2326 | * Description: Returns the cpumask_var_t cpus_allowed of the cpuset |
1da177e4 | 2327 | * attached to the specified @tsk. Guaranteed to return some non-empty |
5f054e31 | 2328 | * subset of cpu_online_mask, even if this means going outside the |
1da177e4 LT |
2329 | * tasks cpuset. |
2330 | **/ | |
2331 | ||
6af866af | 2332 | void cpuset_cpus_allowed(struct task_struct *tsk, struct cpumask *pmask) |
1da177e4 | 2333 | { |
070b57fc LZ |
2334 | struct cpuset *cpus_cs; |
2335 | ||
3d3f26a7 | 2336 | mutex_lock(&callback_mutex); |
909d75a3 | 2337 | task_lock(tsk); |
070b57fc LZ |
2338 | cpus_cs = effective_cpumask_cpuset(task_cs(tsk)); |
2339 | guarantee_online_cpus(cpus_cs, pmask); | |
909d75a3 | 2340 | task_unlock(tsk); |
897f0b3c | 2341 | mutex_unlock(&callback_mutex); |
1da177e4 LT |
2342 | } |
2343 | ||
2baab4e9 | 2344 | void cpuset_cpus_allowed_fallback(struct task_struct *tsk) |
9084bb82 | 2345 | { |
070b57fc | 2346 | const struct cpuset *cpus_cs; |
9084bb82 ON |
2347 | |
2348 | rcu_read_lock(); | |
070b57fc LZ |
2349 | cpus_cs = effective_cpumask_cpuset(task_cs(tsk)); |
2350 | do_set_cpus_allowed(tsk, cpus_cs->cpus_allowed); | |
9084bb82 ON |
2351 | rcu_read_unlock(); |
2352 | ||
2353 | /* | |
2354 | * We own tsk->cpus_allowed, nobody can change it under us. | |
2355 | * | |
2356 | * But we used cs && cs->cpus_allowed lockless and thus can | |
2357 | * race with cgroup_attach_task() or update_cpumask() and get | |
2358 | * the wrong tsk->cpus_allowed. However, both cases imply the | |
2359 | * subsequent cpuset_change_cpumask()->set_cpus_allowed_ptr() | |
2360 | * which takes task_rq_lock(). | |
2361 | * | |
2362 | * If we are called after it dropped the lock we must see all | |
2363 | * changes in tsk_cs()->cpus_allowed. Otherwise we can temporary | |
2364 | * set any mask even if it is not right from task_cs() pov, | |
2365 | * the pending set_cpus_allowed_ptr() will fix things. | |
2baab4e9 PZ |
2366 | * |
2367 | * select_fallback_rq() will fix things ups and set cpu_possible_mask | |
2368 | * if required. | |
9084bb82 | 2369 | */ |
9084bb82 ON |
2370 | } |
2371 | ||
1da177e4 LT |
2372 | void cpuset_init_current_mems_allowed(void) |
2373 | { | |
f9a86fcb | 2374 | nodes_setall(current->mems_allowed); |
1da177e4 LT |
2375 | } |
2376 | ||
909d75a3 PJ |
2377 | /** |
2378 | * cpuset_mems_allowed - return mems_allowed mask from a tasks cpuset. | |
2379 | * @tsk: pointer to task_struct from which to obtain cpuset->mems_allowed. | |
2380 | * | |
2381 | * Description: Returns the nodemask_t mems_allowed of the cpuset | |
2382 | * attached to the specified @tsk. Guaranteed to return some non-empty | |
38d7bee9 | 2383 | * subset of node_states[N_MEMORY], even if this means going outside the |
909d75a3 PJ |
2384 | * tasks cpuset. |
2385 | **/ | |
2386 | ||
2387 | nodemask_t cpuset_mems_allowed(struct task_struct *tsk) | |
2388 | { | |
070b57fc | 2389 | struct cpuset *mems_cs; |
909d75a3 PJ |
2390 | nodemask_t mask; |
2391 | ||
3d3f26a7 | 2392 | mutex_lock(&callback_mutex); |
909d75a3 | 2393 | task_lock(tsk); |
070b57fc LZ |
2394 | mems_cs = effective_nodemask_cpuset(task_cs(tsk)); |
2395 | guarantee_online_mems(mems_cs, &mask); | |
909d75a3 | 2396 | task_unlock(tsk); |
3d3f26a7 | 2397 | mutex_unlock(&callback_mutex); |
909d75a3 PJ |
2398 | |
2399 | return mask; | |
2400 | } | |
2401 | ||
d9fd8a6d | 2402 | /** |
19770b32 MG |
2403 | * cpuset_nodemask_valid_mems_allowed - check nodemask vs. curremt mems_allowed |
2404 | * @nodemask: the nodemask to be checked | |
d9fd8a6d | 2405 | * |
19770b32 | 2406 | * Are any of the nodes in the nodemask allowed in current->mems_allowed? |
1da177e4 | 2407 | */ |
19770b32 | 2408 | int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask) |
1da177e4 | 2409 | { |
19770b32 | 2410 | return nodes_intersects(*nodemask, current->mems_allowed); |
1da177e4 LT |
2411 | } |
2412 | ||
9bf2229f | 2413 | /* |
78608366 PM |
2414 | * nearest_hardwall_ancestor() - Returns the nearest mem_exclusive or |
2415 | * mem_hardwall ancestor to the specified cpuset. Call holding | |
2416 | * callback_mutex. If no ancestor is mem_exclusive or mem_hardwall | |
2417 | * (an unusual configuration), then returns the root cpuset. | |
9bf2229f | 2418 | */ |
78608366 | 2419 | static const struct cpuset *nearest_hardwall_ancestor(const struct cpuset *cs) |
9bf2229f | 2420 | { |
c431069f TH |
2421 | while (!(is_mem_exclusive(cs) || is_mem_hardwall(cs)) && parent_cs(cs)) |
2422 | cs = parent_cs(cs); | |
9bf2229f PJ |
2423 | return cs; |
2424 | } | |
2425 | ||
d9fd8a6d | 2426 | /** |
a1bc5a4e DR |
2427 | * cpuset_node_allowed_softwall - Can we allocate on a memory node? |
2428 | * @node: is this an allowed node? | |
02a0e53d | 2429 | * @gfp_mask: memory allocation flags |
d9fd8a6d | 2430 | * |
a1bc5a4e DR |
2431 | * If we're in interrupt, yes, we can always allocate. If __GFP_THISNODE is |
2432 | * set, yes, we can always allocate. If node is in our task's mems_allowed, | |
2433 | * yes. If it's not a __GFP_HARDWALL request and this node is in the nearest | |
2434 | * hardwalled cpuset ancestor to this task's cpuset, yes. If the task has been | |
2435 | * OOM killed and has access to memory reserves as specified by the TIF_MEMDIE | |
2436 | * flag, yes. | |
9bf2229f PJ |
2437 | * Otherwise, no. |
2438 | * | |
a1bc5a4e DR |
2439 | * If __GFP_HARDWALL is set, cpuset_node_allowed_softwall() reduces to |
2440 | * cpuset_node_allowed_hardwall(). Otherwise, cpuset_node_allowed_softwall() | |
2441 | * might sleep, and might allow a node from an enclosing cpuset. | |
02a0e53d | 2442 | * |
a1bc5a4e DR |
2443 | * cpuset_node_allowed_hardwall() only handles the simpler case of hardwall |
2444 | * cpusets, and never sleeps. | |
02a0e53d PJ |
2445 | * |
2446 | * The __GFP_THISNODE placement logic is really handled elsewhere, | |
2447 | * by forcibly using a zonelist starting at a specified node, and by | |
2448 | * (in get_page_from_freelist()) refusing to consider the zones for | |
2449 | * any node on the zonelist except the first. By the time any such | |
2450 | * calls get to this routine, we should just shut up and say 'yes'. | |
2451 | * | |
9bf2229f | 2452 | * GFP_USER allocations are marked with the __GFP_HARDWALL bit, |
c596d9f3 DR |
2453 | * and do not allow allocations outside the current tasks cpuset |
2454 | * unless the task has been OOM killed as is marked TIF_MEMDIE. | |
9bf2229f | 2455 | * GFP_KERNEL allocations are not so marked, so can escape to the |
78608366 | 2456 | * nearest enclosing hardwalled ancestor cpuset. |
9bf2229f | 2457 | * |
02a0e53d PJ |
2458 | * Scanning up parent cpusets requires callback_mutex. The |
2459 | * __alloc_pages() routine only calls here with __GFP_HARDWALL bit | |
2460 | * _not_ set if it's a GFP_KERNEL allocation, and all nodes in the | |
2461 | * current tasks mems_allowed came up empty on the first pass over | |
2462 | * the zonelist. So only GFP_KERNEL allocations, if all nodes in the | |
2463 | * cpuset are short of memory, might require taking the callback_mutex | |
2464 | * mutex. | |
9bf2229f | 2465 | * |
36be57ff | 2466 | * The first call here from mm/page_alloc:get_page_from_freelist() |
02a0e53d PJ |
2467 | * has __GFP_HARDWALL set in gfp_mask, enforcing hardwall cpusets, |
2468 | * so no allocation on a node outside the cpuset is allowed (unless | |
2469 | * in interrupt, of course). | |
36be57ff PJ |
2470 | * |
2471 | * The second pass through get_page_from_freelist() doesn't even call | |
2472 | * here for GFP_ATOMIC calls. For those calls, the __alloc_pages() | |
2473 | * variable 'wait' is not set, and the bit ALLOC_CPUSET is not set | |
2474 | * in alloc_flags. That logic and the checks below have the combined | |
2475 | * affect that: | |
9bf2229f PJ |
2476 | * in_interrupt - any node ok (current task context irrelevant) |
2477 | * GFP_ATOMIC - any node ok | |
c596d9f3 | 2478 | * TIF_MEMDIE - any node ok |
78608366 | 2479 | * GFP_KERNEL - any node in enclosing hardwalled cpuset ok |
9bf2229f | 2480 | * GFP_USER - only nodes in current tasks mems allowed ok. |
36be57ff PJ |
2481 | * |
2482 | * Rule: | |
a1bc5a4e | 2483 | * Don't call cpuset_node_allowed_softwall if you can't sleep, unless you |
36be57ff PJ |
2484 | * pass in the __GFP_HARDWALL flag set in gfp_flag, which disables |
2485 | * the code that might scan up ancestor cpusets and sleep. | |
02a0e53d | 2486 | */ |
a1bc5a4e | 2487 | int __cpuset_node_allowed_softwall(int node, gfp_t gfp_mask) |
1da177e4 | 2488 | { |
9bf2229f | 2489 | const struct cpuset *cs; /* current cpuset ancestors */ |
29afd49b | 2490 | int allowed; /* is allocation in zone z allowed? */ |
9bf2229f | 2491 | |
9b819d20 | 2492 | if (in_interrupt() || (gfp_mask & __GFP_THISNODE)) |
9bf2229f | 2493 | return 1; |
92d1dbd2 | 2494 | might_sleep_if(!(gfp_mask & __GFP_HARDWALL)); |
9bf2229f PJ |
2495 | if (node_isset(node, current->mems_allowed)) |
2496 | return 1; | |
c596d9f3 DR |
2497 | /* |
2498 | * Allow tasks that have access to memory reserves because they have | |
2499 | * been OOM killed to get memory anywhere. | |
2500 | */ | |
2501 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
2502 | return 1; | |
9bf2229f PJ |
2503 | if (gfp_mask & __GFP_HARDWALL) /* If hardwall request, stop here */ |
2504 | return 0; | |
2505 | ||
5563e770 BP |
2506 | if (current->flags & PF_EXITING) /* Let dying task have memory */ |
2507 | return 1; | |
2508 | ||
9bf2229f | 2509 | /* Not hardwall and node outside mems_allowed: scan up cpusets */ |
3d3f26a7 | 2510 | mutex_lock(&callback_mutex); |
053199ed | 2511 | |
053199ed | 2512 | task_lock(current); |
78608366 | 2513 | cs = nearest_hardwall_ancestor(task_cs(current)); |
053199ed PJ |
2514 | task_unlock(current); |
2515 | ||
9bf2229f | 2516 | allowed = node_isset(node, cs->mems_allowed); |
3d3f26a7 | 2517 | mutex_unlock(&callback_mutex); |
9bf2229f | 2518 | return allowed; |
1da177e4 LT |
2519 | } |
2520 | ||
02a0e53d | 2521 | /* |
a1bc5a4e DR |
2522 | * cpuset_node_allowed_hardwall - Can we allocate on a memory node? |
2523 | * @node: is this an allowed node? | |
02a0e53d PJ |
2524 | * @gfp_mask: memory allocation flags |
2525 | * | |
a1bc5a4e DR |
2526 | * If we're in interrupt, yes, we can always allocate. If __GFP_THISNODE is |
2527 | * set, yes, we can always allocate. If node is in our task's mems_allowed, | |
2528 | * yes. If the task has been OOM killed and has access to memory reserves as | |
2529 | * specified by the TIF_MEMDIE flag, yes. | |
2530 | * Otherwise, no. | |
02a0e53d PJ |
2531 | * |
2532 | * The __GFP_THISNODE placement logic is really handled elsewhere, | |
2533 | * by forcibly using a zonelist starting at a specified node, and by | |
2534 | * (in get_page_from_freelist()) refusing to consider the zones for | |
2535 | * any node on the zonelist except the first. By the time any such | |
2536 | * calls get to this routine, we should just shut up and say 'yes'. | |
2537 | * | |
a1bc5a4e DR |
2538 | * Unlike the cpuset_node_allowed_softwall() variant, above, |
2539 | * this variant requires that the node be in the current task's | |
02a0e53d PJ |
2540 | * mems_allowed or that we're in interrupt. It does not scan up the |
2541 | * cpuset hierarchy for the nearest enclosing mem_exclusive cpuset. | |
2542 | * It never sleeps. | |
2543 | */ | |
a1bc5a4e | 2544 | int __cpuset_node_allowed_hardwall(int node, gfp_t gfp_mask) |
02a0e53d | 2545 | { |
02a0e53d PJ |
2546 | if (in_interrupt() || (gfp_mask & __GFP_THISNODE)) |
2547 | return 1; | |
02a0e53d PJ |
2548 | if (node_isset(node, current->mems_allowed)) |
2549 | return 1; | |
dedf8b79 DW |
2550 | /* |
2551 | * Allow tasks that have access to memory reserves because they have | |
2552 | * been OOM killed to get memory anywhere. | |
2553 | */ | |
2554 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
2555 | return 1; | |
02a0e53d PJ |
2556 | return 0; |
2557 | } | |
2558 | ||
825a46af | 2559 | /** |
6adef3eb JS |
2560 | * cpuset_mem_spread_node() - On which node to begin search for a file page |
2561 | * cpuset_slab_spread_node() - On which node to begin search for a slab page | |
825a46af PJ |
2562 | * |
2563 | * If a task is marked PF_SPREAD_PAGE or PF_SPREAD_SLAB (as for | |
2564 | * tasks in a cpuset with is_spread_page or is_spread_slab set), | |
2565 | * and if the memory allocation used cpuset_mem_spread_node() | |
2566 | * to determine on which node to start looking, as it will for | |
2567 | * certain page cache or slab cache pages such as used for file | |
2568 | * system buffers and inode caches, then instead of starting on the | |
2569 | * local node to look for a free page, rather spread the starting | |
2570 | * node around the tasks mems_allowed nodes. | |
2571 | * | |
2572 | * We don't have to worry about the returned node being offline | |
2573 | * because "it can't happen", and even if it did, it would be ok. | |
2574 | * | |
2575 | * The routines calling guarantee_online_mems() are careful to | |
2576 | * only set nodes in task->mems_allowed that are online. So it | |
2577 | * should not be possible for the following code to return an | |
2578 | * offline node. But if it did, that would be ok, as this routine | |
2579 | * is not returning the node where the allocation must be, only | |
2580 | * the node where the search should start. The zonelist passed to | |
2581 | * __alloc_pages() will include all nodes. If the slab allocator | |
2582 | * is passed an offline node, it will fall back to the local node. | |
2583 | * See kmem_cache_alloc_node(). | |
2584 | */ | |
2585 | ||
6adef3eb | 2586 | static int cpuset_spread_node(int *rotor) |
825a46af PJ |
2587 | { |
2588 | int node; | |
2589 | ||
6adef3eb | 2590 | node = next_node(*rotor, current->mems_allowed); |
825a46af PJ |
2591 | if (node == MAX_NUMNODES) |
2592 | node = first_node(current->mems_allowed); | |
6adef3eb | 2593 | *rotor = node; |
825a46af PJ |
2594 | return node; |
2595 | } | |
6adef3eb JS |
2596 | |
2597 | int cpuset_mem_spread_node(void) | |
2598 | { | |
778d3b0f MH |
2599 | if (current->cpuset_mem_spread_rotor == NUMA_NO_NODE) |
2600 | current->cpuset_mem_spread_rotor = | |
2601 | node_random(¤t->mems_allowed); | |
2602 | ||
6adef3eb JS |
2603 | return cpuset_spread_node(¤t->cpuset_mem_spread_rotor); |
2604 | } | |
2605 | ||
2606 | int cpuset_slab_spread_node(void) | |
2607 | { | |
778d3b0f MH |
2608 | if (current->cpuset_slab_spread_rotor == NUMA_NO_NODE) |
2609 | current->cpuset_slab_spread_rotor = | |
2610 | node_random(¤t->mems_allowed); | |
2611 | ||
6adef3eb JS |
2612 | return cpuset_spread_node(¤t->cpuset_slab_spread_rotor); |
2613 | } | |
2614 | ||
825a46af PJ |
2615 | EXPORT_SYMBOL_GPL(cpuset_mem_spread_node); |
2616 | ||
ef08e3b4 | 2617 | /** |
bbe373f2 DR |
2618 | * cpuset_mems_allowed_intersects - Does @tsk1's mems_allowed intersect @tsk2's? |
2619 | * @tsk1: pointer to task_struct of some task. | |
2620 | * @tsk2: pointer to task_struct of some other task. | |
2621 | * | |
2622 | * Description: Return true if @tsk1's mems_allowed intersects the | |
2623 | * mems_allowed of @tsk2. Used by the OOM killer to determine if | |
2624 | * one of the task's memory usage might impact the memory available | |
2625 | * to the other. | |
ef08e3b4 PJ |
2626 | **/ |
2627 | ||
bbe373f2 DR |
2628 | int cpuset_mems_allowed_intersects(const struct task_struct *tsk1, |
2629 | const struct task_struct *tsk2) | |
ef08e3b4 | 2630 | { |
bbe373f2 | 2631 | return nodes_intersects(tsk1->mems_allowed, tsk2->mems_allowed); |
ef08e3b4 PJ |
2632 | } |
2633 | ||
f440d98f LZ |
2634 | #define CPUSET_NODELIST_LEN (256) |
2635 | ||
75aa1994 DR |
2636 | /** |
2637 | * cpuset_print_task_mems_allowed - prints task's cpuset and mems_allowed | |
2638 | * @task: pointer to task_struct of some task. | |
2639 | * | |
2640 | * Description: Prints @task's name, cpuset name, and cached copy of its | |
2641 | * mems_allowed to the kernel log. Must hold task_lock(task) to allow | |
2642 | * dereferencing task_cs(task). | |
2643 | */ | |
2644 | void cpuset_print_task_mems_allowed(struct task_struct *tsk) | |
2645 | { | |
f440d98f LZ |
2646 | /* Statically allocated to prevent using excess stack. */ |
2647 | static char cpuset_nodelist[CPUSET_NODELIST_LEN]; | |
2648 | static DEFINE_SPINLOCK(cpuset_buffer_lock); | |
75aa1994 | 2649 | |
f440d98f | 2650 | struct cgroup *cgrp = task_cs(tsk)->css.cgroup; |
63f43f55 | 2651 | |
cfb5966b | 2652 | rcu_read_lock(); |
f440d98f | 2653 | spin_lock(&cpuset_buffer_lock); |
63f43f55 | 2654 | |
75aa1994 DR |
2655 | nodelist_scnprintf(cpuset_nodelist, CPUSET_NODELIST_LEN, |
2656 | tsk->mems_allowed); | |
2657 | printk(KERN_INFO "%s cpuset=%s mems_allowed=%s\n", | |
f440d98f LZ |
2658 | tsk->comm, cgroup_name(cgrp), cpuset_nodelist); |
2659 | ||
75aa1994 | 2660 | spin_unlock(&cpuset_buffer_lock); |
cfb5966b | 2661 | rcu_read_unlock(); |
75aa1994 DR |
2662 | } |
2663 | ||
3e0d98b9 PJ |
2664 | /* |
2665 | * Collection of memory_pressure is suppressed unless | |
2666 | * this flag is enabled by writing "1" to the special | |
2667 | * cpuset file 'memory_pressure_enabled' in the root cpuset. | |
2668 | */ | |
2669 | ||
c5b2aff8 | 2670 | int cpuset_memory_pressure_enabled __read_mostly; |
3e0d98b9 PJ |
2671 | |
2672 | /** | |
2673 | * cpuset_memory_pressure_bump - keep stats of per-cpuset reclaims. | |
2674 | * | |
2675 | * Keep a running average of the rate of synchronous (direct) | |
2676 | * page reclaim efforts initiated by tasks in each cpuset. | |
2677 | * | |
2678 | * This represents the rate at which some task in the cpuset | |
2679 | * ran low on memory on all nodes it was allowed to use, and | |
2680 | * had to enter the kernels page reclaim code in an effort to | |
2681 | * create more free memory by tossing clean pages or swapping | |
2682 | * or writing dirty pages. | |
2683 | * | |
2684 | * Display to user space in the per-cpuset read-only file | |
2685 | * "memory_pressure". Value displayed is an integer | |
2686 | * representing the recent rate of entry into the synchronous | |
2687 | * (direct) page reclaim by any task attached to the cpuset. | |
2688 | **/ | |
2689 | ||
2690 | void __cpuset_memory_pressure_bump(void) | |
2691 | { | |
3e0d98b9 | 2692 | task_lock(current); |
8793d854 | 2693 | fmeter_markevent(&task_cs(current)->fmeter); |
3e0d98b9 PJ |
2694 | task_unlock(current); |
2695 | } | |
2696 | ||
8793d854 | 2697 | #ifdef CONFIG_PROC_PID_CPUSET |
1da177e4 LT |
2698 | /* |
2699 | * proc_cpuset_show() | |
2700 | * - Print tasks cpuset path into seq_file. | |
2701 | * - Used for /proc/<pid>/cpuset. | |
053199ed PJ |
2702 | * - No need to task_lock(tsk) on this tsk->cpuset reference, as it |
2703 | * doesn't really matter if tsk->cpuset changes after we read it, | |
5d21cc2d | 2704 | * and we take cpuset_mutex, keeping cpuset_attach() from changing it |
2df167a3 | 2705 | * anyway. |
1da177e4 | 2706 | */ |
8d8b97ba | 2707 | int proc_cpuset_show(struct seq_file *m, void *unused_v) |
1da177e4 | 2708 | { |
13b41b09 | 2709 | struct pid *pid; |
1da177e4 LT |
2710 | struct task_struct *tsk; |
2711 | char *buf; | |
8793d854 | 2712 | struct cgroup_subsys_state *css; |
99f89551 | 2713 | int retval; |
1da177e4 | 2714 | |
99f89551 | 2715 | retval = -ENOMEM; |
1da177e4 LT |
2716 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
2717 | if (!buf) | |
99f89551 EB |
2718 | goto out; |
2719 | ||
2720 | retval = -ESRCH; | |
13b41b09 EB |
2721 | pid = m->private; |
2722 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
99f89551 EB |
2723 | if (!tsk) |
2724 | goto out_free; | |
1da177e4 | 2725 | |
27e89ae5 | 2726 | rcu_read_lock(); |
8af01f56 | 2727 | css = task_css(tsk, cpuset_subsys_id); |
8793d854 | 2728 | retval = cgroup_path(css->cgroup, buf, PAGE_SIZE); |
27e89ae5 | 2729 | rcu_read_unlock(); |
1da177e4 | 2730 | if (retval < 0) |
27e89ae5 | 2731 | goto out_put_task; |
1da177e4 LT |
2732 | seq_puts(m, buf); |
2733 | seq_putc(m, '\n'); | |
27e89ae5 | 2734 | out_put_task: |
99f89551 EB |
2735 | put_task_struct(tsk); |
2736 | out_free: | |
1da177e4 | 2737 | kfree(buf); |
99f89551 | 2738 | out: |
1da177e4 LT |
2739 | return retval; |
2740 | } | |
8793d854 | 2741 | #endif /* CONFIG_PROC_PID_CPUSET */ |
1da177e4 | 2742 | |
d01d4827 | 2743 | /* Display task mems_allowed in /proc/<pid>/status file. */ |
df5f8314 EB |
2744 | void cpuset_task_status_allowed(struct seq_file *m, struct task_struct *task) |
2745 | { | |
df5f8314 | 2746 | seq_printf(m, "Mems_allowed:\t"); |
30e8e136 | 2747 | seq_nodemask(m, &task->mems_allowed); |
df5f8314 | 2748 | seq_printf(m, "\n"); |
39106dcf | 2749 | seq_printf(m, "Mems_allowed_list:\t"); |
30e8e136 | 2750 | seq_nodemask_list(m, &task->mems_allowed); |
39106dcf | 2751 | seq_printf(m, "\n"); |
1da177e4 | 2752 | } |