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 | ||
a7c6d554 TH |
117 | static inline struct cpuset *css_cs(struct cgroup_subsys_state *css) |
118 | { | |
119 | return css ? container_of(css, struct cpuset, css) : NULL; | |
120 | } | |
121 | ||
8793d854 | 122 | /* Retrieve the cpuset for a cgroup */ |
c9e5fe66 | 123 | static inline struct cpuset *cgroup_cs(struct cgroup *cgrp) |
8793d854 | 124 | { |
a7c6d554 | 125 | return css_cs(cgroup_css(cgrp, cpuset_subsys_id)); |
8793d854 PM |
126 | } |
127 | ||
128 | /* Retrieve the cpuset for a task */ | |
129 | static inline struct cpuset *task_cs(struct task_struct *task) | |
130 | { | |
a7c6d554 | 131 | return css_cs(task_css(task, cpuset_subsys_id)); |
8793d854 | 132 | } |
8793d854 | 133 | |
c9710d80 | 134 | static inline struct cpuset *parent_cs(struct cpuset *cs) |
c431069f | 135 | { |
63876986 | 136 | return css_cs(css_parent(&cs->css)); |
c431069f TH |
137 | } |
138 | ||
b246272e DR |
139 | #ifdef CONFIG_NUMA |
140 | static inline bool task_has_mempolicy(struct task_struct *task) | |
141 | { | |
142 | return task->mempolicy; | |
143 | } | |
144 | #else | |
145 | static inline bool task_has_mempolicy(struct task_struct *task) | |
146 | { | |
147 | return false; | |
148 | } | |
149 | #endif | |
150 | ||
151 | ||
1da177e4 LT |
152 | /* bits in struct cpuset flags field */ |
153 | typedef enum { | |
efeb77b2 | 154 | CS_ONLINE, |
1da177e4 LT |
155 | CS_CPU_EXCLUSIVE, |
156 | CS_MEM_EXCLUSIVE, | |
78608366 | 157 | CS_MEM_HARDWALL, |
45b07ef3 | 158 | CS_MEMORY_MIGRATE, |
029190c5 | 159 | CS_SCHED_LOAD_BALANCE, |
825a46af PJ |
160 | CS_SPREAD_PAGE, |
161 | CS_SPREAD_SLAB, | |
1da177e4 LT |
162 | } cpuset_flagbits_t; |
163 | ||
164 | /* convenient tests for these bits */ | |
efeb77b2 TH |
165 | static inline bool is_cpuset_online(const struct cpuset *cs) |
166 | { | |
167 | return test_bit(CS_ONLINE, &cs->flags); | |
168 | } | |
169 | ||
1da177e4 LT |
170 | static inline int is_cpu_exclusive(const struct cpuset *cs) |
171 | { | |
7b5b9ef0 | 172 | return test_bit(CS_CPU_EXCLUSIVE, &cs->flags); |
1da177e4 LT |
173 | } |
174 | ||
175 | static inline int is_mem_exclusive(const struct cpuset *cs) | |
176 | { | |
7b5b9ef0 | 177 | return test_bit(CS_MEM_EXCLUSIVE, &cs->flags); |
1da177e4 LT |
178 | } |
179 | ||
78608366 PM |
180 | static inline int is_mem_hardwall(const struct cpuset *cs) |
181 | { | |
182 | return test_bit(CS_MEM_HARDWALL, &cs->flags); | |
183 | } | |
184 | ||
029190c5 PJ |
185 | static inline int is_sched_load_balance(const struct cpuset *cs) |
186 | { | |
187 | return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); | |
188 | } | |
189 | ||
45b07ef3 PJ |
190 | static inline int is_memory_migrate(const struct cpuset *cs) |
191 | { | |
7b5b9ef0 | 192 | return test_bit(CS_MEMORY_MIGRATE, &cs->flags); |
45b07ef3 PJ |
193 | } |
194 | ||
825a46af PJ |
195 | static inline int is_spread_page(const struct cpuset *cs) |
196 | { | |
197 | return test_bit(CS_SPREAD_PAGE, &cs->flags); | |
198 | } | |
199 | ||
200 | static inline int is_spread_slab(const struct cpuset *cs) | |
201 | { | |
202 | return test_bit(CS_SPREAD_SLAB, &cs->flags); | |
203 | } | |
204 | ||
1da177e4 | 205 | static struct cpuset top_cpuset = { |
efeb77b2 TH |
206 | .flags = ((1 << CS_ONLINE) | (1 << CS_CPU_EXCLUSIVE) | |
207 | (1 << CS_MEM_EXCLUSIVE)), | |
1da177e4 LT |
208 | }; |
209 | ||
ae8086ce TH |
210 | /** |
211 | * cpuset_for_each_child - traverse online children of a cpuset | |
212 | * @child_cs: loop cursor pointing to the current child | |
492eb21b | 213 | * @pos_css: used for iteration |
ae8086ce TH |
214 | * @parent_cs: target cpuset to walk children of |
215 | * | |
216 | * Walk @child_cs through the online children of @parent_cs. Must be used | |
217 | * with RCU read locked. | |
218 | */ | |
492eb21b TH |
219 | #define cpuset_for_each_child(child_cs, pos_css, parent_cs) \ |
220 | css_for_each_child((pos_css), &(parent_cs)->css) \ | |
221 | if (is_cpuset_online(((child_cs) = css_cs((pos_css))))) | |
ae8086ce | 222 | |
fc560a26 TH |
223 | /** |
224 | * cpuset_for_each_descendant_pre - pre-order walk of a cpuset's descendants | |
225 | * @des_cs: loop cursor pointing to the current descendant | |
492eb21b | 226 | * @pos_css: used for iteration |
fc560a26 TH |
227 | * @root_cs: target cpuset to walk ancestor of |
228 | * | |
229 | * Walk @des_cs through the online descendants of @root_cs. Must be used | |
492eb21b TH |
230 | * with RCU read locked. The caller may modify @pos_css by calling |
231 | * css_rightmost_descendant() to skip subtree. | |
fc560a26 | 232 | */ |
492eb21b TH |
233 | #define cpuset_for_each_descendant_pre(des_cs, pos_css, root_cs) \ |
234 | css_for_each_descendant_pre((pos_css), &(root_cs)->css) \ | |
235 | if (is_cpuset_online(((des_cs) = css_cs((pos_css))))) | |
fc560a26 | 236 | |
1da177e4 | 237 | /* |
5d21cc2d TH |
238 | * There are two global mutexes guarding cpuset structures - cpuset_mutex |
239 | * and callback_mutex. The latter may nest inside the former. We also | |
240 | * require taking task_lock() when dereferencing a task's cpuset pointer. | |
241 | * See "The task_lock() exception", at the end of this comment. | |
242 | * | |
243 | * A task must hold both mutexes to modify cpusets. If a task holds | |
244 | * cpuset_mutex, then it blocks others wanting that mutex, ensuring that it | |
245 | * is the only task able to also acquire callback_mutex and be able to | |
246 | * modify cpusets. It can perform various checks on the cpuset structure | |
247 | * first, knowing nothing will change. It can also allocate memory while | |
248 | * just holding cpuset_mutex. While it is performing these checks, various | |
249 | * callback routines can briefly acquire callback_mutex to query cpusets. | |
250 | * Once it is ready to make the changes, it takes callback_mutex, blocking | |
251 | * everyone else. | |
053199ed PJ |
252 | * |
253 | * Calls to the kernel memory allocator can not be made while holding | |
3d3f26a7 | 254 | * callback_mutex, as that would risk double tripping on callback_mutex |
053199ed PJ |
255 | * from one of the callbacks into the cpuset code from within |
256 | * __alloc_pages(). | |
257 | * | |
3d3f26a7 | 258 | * If a task is only holding callback_mutex, then it has read-only |
053199ed PJ |
259 | * access to cpusets. |
260 | * | |
58568d2a MX |
261 | * Now, the task_struct fields mems_allowed and mempolicy may be changed |
262 | * by other task, we use alloc_lock in the task_struct fields to protect | |
263 | * them. | |
053199ed | 264 | * |
3d3f26a7 | 265 | * The cpuset_common_file_read() handlers only hold callback_mutex across |
053199ed PJ |
266 | * small pieces of code, such as when reading out possibly multi-word |
267 | * cpumasks and nodemasks. | |
268 | * | |
2df167a3 PM |
269 | * Accessing a task's cpuset should be done in accordance with the |
270 | * guidelines for accessing subsystem state in kernel/cgroup.c | |
1da177e4 LT |
271 | */ |
272 | ||
5d21cc2d | 273 | static DEFINE_MUTEX(cpuset_mutex); |
3d3f26a7 | 274 | static DEFINE_MUTEX(callback_mutex); |
4247bdc6 | 275 | |
3a5a6d0c TH |
276 | /* |
277 | * CPU / memory hotplug is handled asynchronously. | |
278 | */ | |
279 | static void cpuset_hotplug_workfn(struct work_struct *work); | |
3a5a6d0c TH |
280 | static DECLARE_WORK(cpuset_hotplug_work, cpuset_hotplug_workfn); |
281 | ||
e44193d3 LZ |
282 | static DECLARE_WAIT_QUEUE_HEAD(cpuset_attach_wq); |
283 | ||
cf417141 MK |
284 | /* |
285 | * This is ugly, but preserves the userspace API for existing cpuset | |
8793d854 | 286 | * users. If someone tries to mount the "cpuset" filesystem, we |
cf417141 MK |
287 | * silently switch it to mount "cgroup" instead |
288 | */ | |
f7e83571 AV |
289 | static struct dentry *cpuset_mount(struct file_system_type *fs_type, |
290 | int flags, const char *unused_dev_name, void *data) | |
1da177e4 | 291 | { |
8793d854 | 292 | struct file_system_type *cgroup_fs = get_fs_type("cgroup"); |
f7e83571 | 293 | struct dentry *ret = ERR_PTR(-ENODEV); |
8793d854 PM |
294 | if (cgroup_fs) { |
295 | char mountopts[] = | |
296 | "cpuset,noprefix," | |
297 | "release_agent=/sbin/cpuset_release_agent"; | |
f7e83571 AV |
298 | ret = cgroup_fs->mount(cgroup_fs, flags, |
299 | unused_dev_name, mountopts); | |
8793d854 PM |
300 | put_filesystem(cgroup_fs); |
301 | } | |
302 | return ret; | |
1da177e4 LT |
303 | } |
304 | ||
305 | static struct file_system_type cpuset_fs_type = { | |
306 | .name = "cpuset", | |
f7e83571 | 307 | .mount = cpuset_mount, |
1da177e4 LT |
308 | }; |
309 | ||
1da177e4 | 310 | /* |
300ed6cb | 311 | * Return in pmask the portion of a cpusets's cpus_allowed that |
1da177e4 | 312 | * are online. If none are online, walk up the cpuset hierarchy |
40df2deb LZ |
313 | * until we find one that does have some online cpus. The top |
314 | * cpuset always has some cpus online. | |
1da177e4 LT |
315 | * |
316 | * One way or another, we guarantee to return some non-empty subset | |
5f054e31 | 317 | * of cpu_online_mask. |
1da177e4 | 318 | * |
3d3f26a7 | 319 | * Call with callback_mutex held. |
1da177e4 | 320 | */ |
c9710d80 | 321 | static void guarantee_online_cpus(struct cpuset *cs, struct cpumask *pmask) |
1da177e4 | 322 | { |
40df2deb | 323 | while (!cpumask_intersects(cs->cpus_allowed, cpu_online_mask)) |
c431069f | 324 | cs = parent_cs(cs); |
40df2deb | 325 | cpumask_and(pmask, cs->cpus_allowed, cpu_online_mask); |
1da177e4 LT |
326 | } |
327 | ||
328 | /* | |
329 | * Return in *pmask the portion of a cpusets's mems_allowed that | |
0e1e7c7a CL |
330 | * are online, with memory. If none are online with memory, walk |
331 | * up the cpuset hierarchy until we find one that does have some | |
40df2deb | 332 | * online mems. The top cpuset always has some mems online. |
1da177e4 LT |
333 | * |
334 | * One way or another, we guarantee to return some non-empty subset | |
38d7bee9 | 335 | * of node_states[N_MEMORY]. |
1da177e4 | 336 | * |
3d3f26a7 | 337 | * Call with callback_mutex held. |
1da177e4 | 338 | */ |
c9710d80 | 339 | static void guarantee_online_mems(struct cpuset *cs, nodemask_t *pmask) |
1da177e4 | 340 | { |
40df2deb | 341 | while (!nodes_intersects(cs->mems_allowed, node_states[N_MEMORY])) |
c431069f | 342 | cs = parent_cs(cs); |
40df2deb | 343 | nodes_and(*pmask, cs->mems_allowed, node_states[N_MEMORY]); |
1da177e4 LT |
344 | } |
345 | ||
f3b39d47 MX |
346 | /* |
347 | * update task's spread flag if cpuset's page/slab spread flag is set | |
348 | * | |
5d21cc2d | 349 | * Called with callback_mutex/cpuset_mutex held |
f3b39d47 MX |
350 | */ |
351 | static void cpuset_update_task_spread_flag(struct cpuset *cs, | |
352 | struct task_struct *tsk) | |
353 | { | |
354 | if (is_spread_page(cs)) | |
355 | tsk->flags |= PF_SPREAD_PAGE; | |
356 | else | |
357 | tsk->flags &= ~PF_SPREAD_PAGE; | |
358 | if (is_spread_slab(cs)) | |
359 | tsk->flags |= PF_SPREAD_SLAB; | |
360 | else | |
361 | tsk->flags &= ~PF_SPREAD_SLAB; | |
362 | } | |
363 | ||
1da177e4 LT |
364 | /* |
365 | * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q? | |
366 | * | |
367 | * One cpuset is a subset of another if all its allowed CPUs and | |
368 | * Memory Nodes are a subset of the other, and its exclusive flags | |
5d21cc2d | 369 | * are only set if the other's are set. Call holding cpuset_mutex. |
1da177e4 LT |
370 | */ |
371 | ||
372 | static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q) | |
373 | { | |
300ed6cb | 374 | return cpumask_subset(p->cpus_allowed, q->cpus_allowed) && |
1da177e4 LT |
375 | nodes_subset(p->mems_allowed, q->mems_allowed) && |
376 | is_cpu_exclusive(p) <= is_cpu_exclusive(q) && | |
377 | is_mem_exclusive(p) <= is_mem_exclusive(q); | |
378 | } | |
379 | ||
645fcc9d LZ |
380 | /** |
381 | * alloc_trial_cpuset - allocate a trial cpuset | |
382 | * @cs: the cpuset that the trial cpuset duplicates | |
383 | */ | |
c9710d80 | 384 | static struct cpuset *alloc_trial_cpuset(struct cpuset *cs) |
645fcc9d | 385 | { |
300ed6cb LZ |
386 | struct cpuset *trial; |
387 | ||
388 | trial = kmemdup(cs, sizeof(*cs), GFP_KERNEL); | |
389 | if (!trial) | |
390 | return NULL; | |
391 | ||
392 | if (!alloc_cpumask_var(&trial->cpus_allowed, GFP_KERNEL)) { | |
393 | kfree(trial); | |
394 | return NULL; | |
395 | } | |
396 | cpumask_copy(trial->cpus_allowed, cs->cpus_allowed); | |
397 | ||
398 | return trial; | |
645fcc9d LZ |
399 | } |
400 | ||
401 | /** | |
402 | * free_trial_cpuset - free the trial cpuset | |
403 | * @trial: the trial cpuset to be freed | |
404 | */ | |
405 | static void free_trial_cpuset(struct cpuset *trial) | |
406 | { | |
300ed6cb | 407 | free_cpumask_var(trial->cpus_allowed); |
645fcc9d LZ |
408 | kfree(trial); |
409 | } | |
410 | ||
1da177e4 LT |
411 | /* |
412 | * validate_change() - Used to validate that any proposed cpuset change | |
413 | * follows the structural rules for cpusets. | |
414 | * | |
415 | * If we replaced the flag and mask values of the current cpuset | |
416 | * (cur) with those values in the trial cpuset (trial), would | |
417 | * our various subset and exclusive rules still be valid? Presumes | |
5d21cc2d | 418 | * cpuset_mutex held. |
1da177e4 LT |
419 | * |
420 | * 'cur' is the address of an actual, in-use cpuset. Operations | |
421 | * such as list traversal that depend on the actual address of the | |
422 | * cpuset in the list must use cur below, not trial. | |
423 | * | |
424 | * 'trial' is the address of bulk structure copy of cur, with | |
425 | * perhaps one or more of the fields cpus_allowed, mems_allowed, | |
426 | * or flags changed to new, trial values. | |
427 | * | |
428 | * Return 0 if valid, -errno if not. | |
429 | */ | |
430 | ||
c9710d80 | 431 | static int validate_change(struct cpuset *cur, struct cpuset *trial) |
1da177e4 | 432 | { |
492eb21b | 433 | struct cgroup_subsys_state *css; |
1da177e4 | 434 | struct cpuset *c, *par; |
ae8086ce TH |
435 | int ret; |
436 | ||
437 | rcu_read_lock(); | |
1da177e4 LT |
438 | |
439 | /* Each of our child cpusets must be a subset of us */ | |
ae8086ce | 440 | ret = -EBUSY; |
492eb21b | 441 | cpuset_for_each_child(c, css, cur) |
ae8086ce TH |
442 | if (!is_cpuset_subset(c, trial)) |
443 | goto out; | |
1da177e4 LT |
444 | |
445 | /* Remaining checks don't apply to root cpuset */ | |
ae8086ce | 446 | ret = 0; |
69604067 | 447 | if (cur == &top_cpuset) |
ae8086ce | 448 | goto out; |
1da177e4 | 449 | |
c431069f | 450 | par = parent_cs(cur); |
69604067 | 451 | |
1da177e4 | 452 | /* We must be a subset of our parent cpuset */ |
ae8086ce | 453 | ret = -EACCES; |
1da177e4 | 454 | if (!is_cpuset_subset(trial, par)) |
ae8086ce | 455 | goto out; |
1da177e4 | 456 | |
2df167a3 PM |
457 | /* |
458 | * If either I or some sibling (!= me) is exclusive, we can't | |
459 | * overlap | |
460 | */ | |
ae8086ce | 461 | ret = -EINVAL; |
492eb21b | 462 | cpuset_for_each_child(c, css, par) { |
1da177e4 LT |
463 | if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) && |
464 | c != cur && | |
300ed6cb | 465 | cpumask_intersects(trial->cpus_allowed, c->cpus_allowed)) |
ae8086ce | 466 | goto out; |
1da177e4 LT |
467 | if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) && |
468 | c != cur && | |
469 | nodes_intersects(trial->mems_allowed, c->mems_allowed)) | |
ae8086ce | 470 | goto out; |
1da177e4 LT |
471 | } |
472 | ||
452477fa TH |
473 | /* |
474 | * Cpusets with tasks - existing or newly being attached - can't | |
475 | * have empty cpus_allowed or mems_allowed. | |
476 | */ | |
ae8086ce | 477 | ret = -ENOSPC; |
452477fa | 478 | if ((cgroup_task_count(cur->css.cgroup) || cur->attach_in_progress) && |
88fa523b | 479 | (cpumask_empty(trial->cpus_allowed) && |
ae8086ce TH |
480 | nodes_empty(trial->mems_allowed))) |
481 | goto out; | |
020958b6 | 482 | |
ae8086ce TH |
483 | ret = 0; |
484 | out: | |
485 | rcu_read_unlock(); | |
486 | return ret; | |
1da177e4 LT |
487 | } |
488 | ||
db7f47cf | 489 | #ifdef CONFIG_SMP |
029190c5 | 490 | /* |
cf417141 | 491 | * Helper routine for generate_sched_domains(). |
029190c5 PJ |
492 | * Do cpusets a, b have overlapping cpus_allowed masks? |
493 | */ | |
029190c5 PJ |
494 | static int cpusets_overlap(struct cpuset *a, struct cpuset *b) |
495 | { | |
300ed6cb | 496 | return cpumask_intersects(a->cpus_allowed, b->cpus_allowed); |
029190c5 PJ |
497 | } |
498 | ||
1d3504fc HS |
499 | static void |
500 | update_domain_attr(struct sched_domain_attr *dattr, struct cpuset *c) | |
501 | { | |
1d3504fc HS |
502 | if (dattr->relax_domain_level < c->relax_domain_level) |
503 | dattr->relax_domain_level = c->relax_domain_level; | |
504 | return; | |
505 | } | |
506 | ||
fc560a26 TH |
507 | static void update_domain_attr_tree(struct sched_domain_attr *dattr, |
508 | struct cpuset *root_cs) | |
f5393693 | 509 | { |
fc560a26 | 510 | struct cpuset *cp; |
492eb21b | 511 | struct cgroup_subsys_state *pos_css; |
f5393693 | 512 | |
fc560a26 | 513 | rcu_read_lock(); |
492eb21b | 514 | cpuset_for_each_descendant_pre(cp, pos_css, root_cs) { |
fc560a26 TH |
515 | /* skip the whole subtree if @cp doesn't have any CPU */ |
516 | if (cpumask_empty(cp->cpus_allowed)) { | |
492eb21b | 517 | pos_css = css_rightmost_descendant(pos_css); |
f5393693 | 518 | continue; |
fc560a26 | 519 | } |
f5393693 LJ |
520 | |
521 | if (is_sched_load_balance(cp)) | |
522 | update_domain_attr(dattr, cp); | |
f5393693 | 523 | } |
fc560a26 | 524 | rcu_read_unlock(); |
f5393693 LJ |
525 | } |
526 | ||
029190c5 | 527 | /* |
cf417141 MK |
528 | * generate_sched_domains() |
529 | * | |
530 | * This function builds a partial partition of the systems CPUs | |
531 | * A 'partial partition' is a set of non-overlapping subsets whose | |
532 | * union is a subset of that set. | |
0a0fca9d | 533 | * The output of this function needs to be passed to kernel/sched/core.c |
cf417141 MK |
534 | * partition_sched_domains() routine, which will rebuild the scheduler's |
535 | * load balancing domains (sched domains) as specified by that partial | |
536 | * partition. | |
029190c5 | 537 | * |
45ce80fb | 538 | * See "What is sched_load_balance" in Documentation/cgroups/cpusets.txt |
029190c5 PJ |
539 | * for a background explanation of this. |
540 | * | |
541 | * Does not return errors, on the theory that the callers of this | |
542 | * routine would rather not worry about failures to rebuild sched | |
543 | * domains when operating in the severe memory shortage situations | |
544 | * that could cause allocation failures below. | |
545 | * | |
5d21cc2d | 546 | * Must be called with cpuset_mutex held. |
029190c5 PJ |
547 | * |
548 | * The three key local variables below are: | |
aeed6824 | 549 | * q - a linked-list queue of cpuset pointers, used to implement a |
029190c5 PJ |
550 | * top-down scan of all cpusets. This scan loads a pointer |
551 | * to each cpuset marked is_sched_load_balance into the | |
552 | * array 'csa'. For our purposes, rebuilding the schedulers | |
553 | * sched domains, we can ignore !is_sched_load_balance cpusets. | |
554 | * csa - (for CpuSet Array) Array of pointers to all the cpusets | |
555 | * that need to be load balanced, for convenient iterative | |
556 | * access by the subsequent code that finds the best partition, | |
557 | * i.e the set of domains (subsets) of CPUs such that the | |
558 | * cpus_allowed of every cpuset marked is_sched_load_balance | |
559 | * is a subset of one of these domains, while there are as | |
560 | * many such domains as possible, each as small as possible. | |
561 | * doms - Conversion of 'csa' to an array of cpumasks, for passing to | |
0a0fca9d | 562 | * the kernel/sched/core.c routine partition_sched_domains() in a |
029190c5 PJ |
563 | * convenient format, that can be easily compared to the prior |
564 | * value to determine what partition elements (sched domains) | |
565 | * were changed (added or removed.) | |
566 | * | |
567 | * Finding the best partition (set of domains): | |
568 | * The triple nested loops below over i, j, k scan over the | |
569 | * load balanced cpusets (using the array of cpuset pointers in | |
570 | * csa[]) looking for pairs of cpusets that have overlapping | |
571 | * cpus_allowed, but which don't have the same 'pn' partition | |
572 | * number and gives them in the same partition number. It keeps | |
573 | * looping on the 'restart' label until it can no longer find | |
574 | * any such pairs. | |
575 | * | |
576 | * The union of the cpus_allowed masks from the set of | |
577 | * all cpusets having the same 'pn' value then form the one | |
578 | * element of the partition (one sched domain) to be passed to | |
579 | * partition_sched_domains(). | |
580 | */ | |
acc3f5d7 | 581 | static int generate_sched_domains(cpumask_var_t **domains, |
cf417141 | 582 | struct sched_domain_attr **attributes) |
029190c5 | 583 | { |
029190c5 PJ |
584 | struct cpuset *cp; /* scans q */ |
585 | struct cpuset **csa; /* array of all cpuset ptrs */ | |
586 | int csn; /* how many cpuset ptrs in csa so far */ | |
587 | int i, j, k; /* indices for partition finding loops */ | |
acc3f5d7 | 588 | cpumask_var_t *doms; /* resulting partition; i.e. sched domains */ |
1d3504fc | 589 | struct sched_domain_attr *dattr; /* attributes for custom domains */ |
1583715d | 590 | int ndoms = 0; /* number of sched domains in result */ |
6af866af | 591 | int nslot; /* next empty doms[] struct cpumask slot */ |
492eb21b | 592 | struct cgroup_subsys_state *pos_css; |
029190c5 | 593 | |
029190c5 | 594 | doms = NULL; |
1d3504fc | 595 | dattr = NULL; |
cf417141 | 596 | csa = NULL; |
029190c5 PJ |
597 | |
598 | /* Special case for the 99% of systems with one, full, sched domain */ | |
599 | if (is_sched_load_balance(&top_cpuset)) { | |
acc3f5d7 RR |
600 | ndoms = 1; |
601 | doms = alloc_sched_domains(ndoms); | |
029190c5 | 602 | if (!doms) |
cf417141 MK |
603 | goto done; |
604 | ||
1d3504fc HS |
605 | dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL); |
606 | if (dattr) { | |
607 | *dattr = SD_ATTR_INIT; | |
93a65575 | 608 | update_domain_attr_tree(dattr, &top_cpuset); |
1d3504fc | 609 | } |
acc3f5d7 | 610 | cpumask_copy(doms[0], top_cpuset.cpus_allowed); |
cf417141 | 611 | |
cf417141 | 612 | goto done; |
029190c5 PJ |
613 | } |
614 | ||
029190c5 PJ |
615 | csa = kmalloc(number_of_cpusets * sizeof(cp), GFP_KERNEL); |
616 | if (!csa) | |
617 | goto done; | |
618 | csn = 0; | |
619 | ||
fc560a26 | 620 | rcu_read_lock(); |
492eb21b | 621 | cpuset_for_each_descendant_pre(cp, pos_css, &top_cpuset) { |
f5393693 | 622 | /* |
fc560a26 TH |
623 | * Continue traversing beyond @cp iff @cp has some CPUs and |
624 | * isn't load balancing. The former is obvious. The | |
625 | * latter: All child cpusets contain a subset of the | |
626 | * parent's cpus, so just skip them, and then we call | |
627 | * update_domain_attr_tree() to calc relax_domain_level of | |
628 | * the corresponding sched domain. | |
f5393693 | 629 | */ |
fc560a26 TH |
630 | if (!cpumask_empty(cp->cpus_allowed) && |
631 | !is_sched_load_balance(cp)) | |
f5393693 | 632 | continue; |
489a5393 | 633 | |
fc560a26 TH |
634 | if (is_sched_load_balance(cp)) |
635 | csa[csn++] = cp; | |
636 | ||
637 | /* skip @cp's subtree */ | |
492eb21b | 638 | pos_css = css_rightmost_descendant(pos_css); |
fc560a26 TH |
639 | } |
640 | rcu_read_unlock(); | |
029190c5 PJ |
641 | |
642 | for (i = 0; i < csn; i++) | |
643 | csa[i]->pn = i; | |
644 | ndoms = csn; | |
645 | ||
646 | restart: | |
647 | /* Find the best partition (set of sched domains) */ | |
648 | for (i = 0; i < csn; i++) { | |
649 | struct cpuset *a = csa[i]; | |
650 | int apn = a->pn; | |
651 | ||
652 | for (j = 0; j < csn; j++) { | |
653 | struct cpuset *b = csa[j]; | |
654 | int bpn = b->pn; | |
655 | ||
656 | if (apn != bpn && cpusets_overlap(a, b)) { | |
657 | for (k = 0; k < csn; k++) { | |
658 | struct cpuset *c = csa[k]; | |
659 | ||
660 | if (c->pn == bpn) | |
661 | c->pn = apn; | |
662 | } | |
663 | ndoms--; /* one less element */ | |
664 | goto restart; | |
665 | } | |
666 | } | |
667 | } | |
668 | ||
cf417141 MK |
669 | /* |
670 | * Now we know how many domains to create. | |
671 | * Convert <csn, csa> to <ndoms, doms> and populate cpu masks. | |
672 | */ | |
acc3f5d7 | 673 | doms = alloc_sched_domains(ndoms); |
700018e0 | 674 | if (!doms) |
cf417141 | 675 | goto done; |
cf417141 MK |
676 | |
677 | /* | |
678 | * The rest of the code, including the scheduler, can deal with | |
679 | * dattr==NULL case. No need to abort if alloc fails. | |
680 | */ | |
1d3504fc | 681 | dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL); |
029190c5 PJ |
682 | |
683 | for (nslot = 0, i = 0; i < csn; i++) { | |
684 | struct cpuset *a = csa[i]; | |
6af866af | 685 | struct cpumask *dp; |
029190c5 PJ |
686 | int apn = a->pn; |
687 | ||
cf417141 MK |
688 | if (apn < 0) { |
689 | /* Skip completed partitions */ | |
690 | continue; | |
691 | } | |
692 | ||
acc3f5d7 | 693 | dp = doms[nslot]; |
cf417141 MK |
694 | |
695 | if (nslot == ndoms) { | |
696 | static int warnings = 10; | |
697 | if (warnings) { | |
698 | printk(KERN_WARNING | |
699 | "rebuild_sched_domains confused:" | |
700 | " nslot %d, ndoms %d, csn %d, i %d," | |
701 | " apn %d\n", | |
702 | nslot, ndoms, csn, i, apn); | |
703 | warnings--; | |
029190c5 | 704 | } |
cf417141 MK |
705 | continue; |
706 | } | |
029190c5 | 707 | |
6af866af | 708 | cpumask_clear(dp); |
cf417141 MK |
709 | if (dattr) |
710 | *(dattr + nslot) = SD_ATTR_INIT; | |
711 | for (j = i; j < csn; j++) { | |
712 | struct cpuset *b = csa[j]; | |
713 | ||
714 | if (apn == b->pn) { | |
300ed6cb | 715 | cpumask_or(dp, dp, b->cpus_allowed); |
cf417141 MK |
716 | if (dattr) |
717 | update_domain_attr_tree(dattr + nslot, b); | |
718 | ||
719 | /* Done with this partition */ | |
720 | b->pn = -1; | |
029190c5 | 721 | } |
029190c5 | 722 | } |
cf417141 | 723 | nslot++; |
029190c5 PJ |
724 | } |
725 | BUG_ON(nslot != ndoms); | |
726 | ||
cf417141 MK |
727 | done: |
728 | kfree(csa); | |
729 | ||
700018e0 LZ |
730 | /* |
731 | * Fallback to the default domain if kmalloc() failed. | |
732 | * See comments in partition_sched_domains(). | |
733 | */ | |
734 | if (doms == NULL) | |
735 | ndoms = 1; | |
736 | ||
cf417141 MK |
737 | *domains = doms; |
738 | *attributes = dattr; | |
739 | return ndoms; | |
740 | } | |
741 | ||
742 | /* | |
743 | * Rebuild scheduler domains. | |
744 | * | |
699140ba TH |
745 | * If the flag 'sched_load_balance' of any cpuset with non-empty |
746 | * 'cpus' changes, or if the 'cpus' allowed changes in any cpuset | |
747 | * which has that flag enabled, or if any cpuset with a non-empty | |
748 | * 'cpus' is removed, then call this routine to rebuild the | |
749 | * scheduler's dynamic sched domains. | |
cf417141 | 750 | * |
5d21cc2d | 751 | * Call with cpuset_mutex held. Takes get_online_cpus(). |
cf417141 | 752 | */ |
699140ba | 753 | static void rebuild_sched_domains_locked(void) |
cf417141 MK |
754 | { |
755 | struct sched_domain_attr *attr; | |
acc3f5d7 | 756 | cpumask_var_t *doms; |
cf417141 MK |
757 | int ndoms; |
758 | ||
5d21cc2d | 759 | lockdep_assert_held(&cpuset_mutex); |
86ef5c9a | 760 | get_online_cpus(); |
cf417141 | 761 | |
5b16c2a4 LZ |
762 | /* |
763 | * We have raced with CPU hotplug. Don't do anything to avoid | |
764 | * passing doms with offlined cpu to partition_sched_domains(). | |
765 | * Anyways, hotplug work item will rebuild sched domains. | |
766 | */ | |
767 | if (!cpumask_equal(top_cpuset.cpus_allowed, cpu_active_mask)) | |
768 | goto out; | |
769 | ||
cf417141 | 770 | /* Generate domain masks and attrs */ |
cf417141 | 771 | ndoms = generate_sched_domains(&doms, &attr); |
cf417141 MK |
772 | |
773 | /* Have scheduler rebuild the domains */ | |
774 | partition_sched_domains(ndoms, doms, attr); | |
5b16c2a4 | 775 | out: |
86ef5c9a | 776 | put_online_cpus(); |
cf417141 | 777 | } |
db7f47cf | 778 | #else /* !CONFIG_SMP */ |
699140ba | 779 | static void rebuild_sched_domains_locked(void) |
db7f47cf PM |
780 | { |
781 | } | |
db7f47cf | 782 | #endif /* CONFIG_SMP */ |
029190c5 | 783 | |
cf417141 MK |
784 | void rebuild_sched_domains(void) |
785 | { | |
5d21cc2d | 786 | mutex_lock(&cpuset_mutex); |
699140ba | 787 | rebuild_sched_domains_locked(); |
5d21cc2d | 788 | mutex_unlock(&cpuset_mutex); |
029190c5 PJ |
789 | } |
790 | ||
070b57fc LZ |
791 | /* |
792 | * effective_cpumask_cpuset - return nearest ancestor with non-empty cpus | |
793 | * @cs: the cpuset in interest | |
58f4790b | 794 | * |
070b57fc LZ |
795 | * A cpuset's effective cpumask is the cpumask of the nearest ancestor |
796 | * with non-empty cpus. We use effective cpumask whenever: | |
797 | * - we update tasks' cpus_allowed. (they take on the ancestor's cpumask | |
798 | * if the cpuset they reside in has no cpus) | |
799 | * - we want to retrieve task_cs(tsk)'s cpus_allowed. | |
800 | * | |
801 | * Called with cpuset_mutex held. cpuset_cpus_allowed_fallback() is an | |
802 | * exception. See comments there. | |
803 | */ | |
804 | static struct cpuset *effective_cpumask_cpuset(struct cpuset *cs) | |
805 | { | |
806 | while (cpumask_empty(cs->cpus_allowed)) | |
807 | cs = parent_cs(cs); | |
808 | return cs; | |
809 | } | |
810 | ||
811 | /* | |
812 | * effective_nodemask_cpuset - return nearest ancestor with non-empty mems | |
813 | * @cs: the cpuset in interest | |
814 | * | |
815 | * A cpuset's effective nodemask is the nodemask of the nearest ancestor | |
816 | * with non-empty memss. We use effective nodemask whenever: | |
817 | * - we update tasks' mems_allowed. (they take on the ancestor's nodemask | |
818 | * if the cpuset they reside in has no mems) | |
819 | * - we want to retrieve task_cs(tsk)'s mems_allowed. | |
820 | * | |
821 | * Called with cpuset_mutex held. | |
053199ed | 822 | */ |
070b57fc | 823 | static struct cpuset *effective_nodemask_cpuset(struct cpuset *cs) |
58f4790b | 824 | { |
070b57fc LZ |
825 | while (nodes_empty(cs->mems_allowed)) |
826 | cs = parent_cs(cs); | |
827 | return cs; | |
58f4790b | 828 | } |
053199ed | 829 | |
58f4790b CW |
830 | /** |
831 | * cpuset_change_cpumask - make a task's cpus_allowed the same as its cpuset's | |
832 | * @tsk: task to test | |
833 | * @scan: struct cgroup_scanner containing the cgroup of the task | |
834 | * | |
835 | * Called by cgroup_scan_tasks() for each task in a cgroup whose | |
836 | * cpus_allowed mask needs to be changed. | |
837 | * | |
838 | * We don't need to re-check for the cgroup/cpuset membership, since we're | |
5d21cc2d | 839 | * holding cpuset_mutex at this point. |
58f4790b | 840 | */ |
9e0c914c AB |
841 | static void cpuset_change_cpumask(struct task_struct *tsk, |
842 | struct cgroup_scanner *scan) | |
58f4790b | 843 | { |
070b57fc LZ |
844 | struct cpuset *cpus_cs; |
845 | ||
6f4b7e63 | 846 | cpus_cs = effective_cpumask_cpuset(cgroup_cs(scan->cgrp)); |
070b57fc | 847 | set_cpus_allowed_ptr(tsk, cpus_cs->cpus_allowed); |
58f4790b CW |
848 | } |
849 | ||
0b2f630a MX |
850 | /** |
851 | * update_tasks_cpumask - Update the cpumasks of tasks in the cpuset. | |
852 | * @cs: the cpuset in which each task's cpus_allowed mask needs to be changed | |
4e74339a | 853 | * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks() |
0b2f630a | 854 | * |
5d21cc2d | 855 | * Called with cpuset_mutex held |
0b2f630a MX |
856 | * |
857 | * The cgroup_scan_tasks() function will scan all the tasks in a cgroup, | |
858 | * calling callback functions for each. | |
859 | * | |
4e74339a LZ |
860 | * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0 |
861 | * if @heap != NULL. | |
0b2f630a | 862 | */ |
4e74339a | 863 | static void update_tasks_cpumask(struct cpuset *cs, struct ptr_heap *heap) |
0b2f630a MX |
864 | { |
865 | struct cgroup_scanner scan; | |
0b2f630a | 866 | |
6f4b7e63 | 867 | scan.cgrp = cs->css.cgroup; |
249cc86d | 868 | scan.test_task = NULL; |
0b2f630a | 869 | scan.process_task = cpuset_change_cpumask; |
4e74339a LZ |
870 | scan.heap = heap; |
871 | cgroup_scan_tasks(&scan); | |
0b2f630a MX |
872 | } |
873 | ||
5c5cc623 LZ |
874 | /* |
875 | * update_tasks_cpumask_hier - Update the cpumasks of tasks in the hierarchy. | |
876 | * @root_cs: the root cpuset of the hierarchy | |
877 | * @update_root: update root cpuset or not? | |
878 | * @heap: the heap used by cgroup_scan_tasks() | |
879 | * | |
880 | * This will update cpumasks of tasks in @root_cs and all other empty cpusets | |
881 | * which take on cpumask of @root_cs. | |
882 | * | |
883 | * Called with cpuset_mutex held | |
884 | */ | |
885 | static void update_tasks_cpumask_hier(struct cpuset *root_cs, | |
886 | bool update_root, struct ptr_heap *heap) | |
887 | { | |
888 | struct cpuset *cp; | |
492eb21b | 889 | struct cgroup_subsys_state *pos_css; |
5c5cc623 LZ |
890 | |
891 | if (update_root) | |
892 | update_tasks_cpumask(root_cs, heap); | |
893 | ||
894 | rcu_read_lock(); | |
492eb21b | 895 | cpuset_for_each_descendant_pre(cp, pos_css, root_cs) { |
5c5cc623 LZ |
896 | /* skip the whole subtree if @cp have some CPU */ |
897 | if (!cpumask_empty(cp->cpus_allowed)) { | |
492eb21b | 898 | pos_css = css_rightmost_descendant(pos_css); |
5c5cc623 LZ |
899 | continue; |
900 | } | |
901 | if (!css_tryget(&cp->css)) | |
902 | continue; | |
903 | rcu_read_unlock(); | |
904 | ||
905 | update_tasks_cpumask(cp, heap); | |
906 | ||
907 | rcu_read_lock(); | |
908 | css_put(&cp->css); | |
909 | } | |
910 | rcu_read_unlock(); | |
911 | } | |
912 | ||
58f4790b CW |
913 | /** |
914 | * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it | |
915 | * @cs: the cpuset to consider | |
916 | * @buf: buffer of cpu numbers written to this cpuset | |
917 | */ | |
645fcc9d LZ |
918 | static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs, |
919 | const char *buf) | |
1da177e4 | 920 | { |
4e74339a | 921 | struct ptr_heap heap; |
58f4790b CW |
922 | int retval; |
923 | int is_load_balanced; | |
1da177e4 | 924 | |
5f054e31 | 925 | /* top_cpuset.cpus_allowed tracks cpu_online_mask; it's read-only */ |
4c4d50f7 PJ |
926 | if (cs == &top_cpuset) |
927 | return -EACCES; | |
928 | ||
6f7f02e7 | 929 | /* |
c8d9c90c | 930 | * An empty cpus_allowed is ok only if the cpuset has no tasks. |
020958b6 PJ |
931 | * Since cpulist_parse() fails on an empty mask, we special case |
932 | * that parsing. The validate_change() call ensures that cpusets | |
933 | * with tasks have cpus. | |
6f7f02e7 | 934 | */ |
020958b6 | 935 | if (!*buf) { |
300ed6cb | 936 | cpumask_clear(trialcs->cpus_allowed); |
6f7f02e7 | 937 | } else { |
300ed6cb | 938 | retval = cpulist_parse(buf, trialcs->cpus_allowed); |
6f7f02e7 DR |
939 | if (retval < 0) |
940 | return retval; | |
37340746 | 941 | |
6ad4c188 | 942 | if (!cpumask_subset(trialcs->cpus_allowed, cpu_active_mask)) |
37340746 | 943 | return -EINVAL; |
6f7f02e7 | 944 | } |
029190c5 | 945 | |
8707d8b8 | 946 | /* Nothing to do if the cpus didn't change */ |
300ed6cb | 947 | if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed)) |
8707d8b8 | 948 | return 0; |
58f4790b | 949 | |
a73456f3 LZ |
950 | retval = validate_change(cs, trialcs); |
951 | if (retval < 0) | |
952 | return retval; | |
953 | ||
4e74339a LZ |
954 | retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL); |
955 | if (retval) | |
956 | return retval; | |
957 | ||
645fcc9d | 958 | is_load_balanced = is_sched_load_balance(trialcs); |
029190c5 | 959 | |
3d3f26a7 | 960 | mutex_lock(&callback_mutex); |
300ed6cb | 961 | cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed); |
3d3f26a7 | 962 | mutex_unlock(&callback_mutex); |
029190c5 | 963 | |
5c5cc623 | 964 | update_tasks_cpumask_hier(cs, true, &heap); |
4e74339a LZ |
965 | |
966 | heap_free(&heap); | |
58f4790b | 967 | |
8707d8b8 | 968 | if (is_load_balanced) |
699140ba | 969 | rebuild_sched_domains_locked(); |
85d7b949 | 970 | return 0; |
1da177e4 LT |
971 | } |
972 | ||
e4e364e8 PJ |
973 | /* |
974 | * cpuset_migrate_mm | |
975 | * | |
976 | * Migrate memory region from one set of nodes to another. | |
977 | * | |
978 | * Temporarilly set tasks mems_allowed to target nodes of migration, | |
979 | * so that the migration code can allocate pages on these nodes. | |
980 | * | |
5d21cc2d | 981 | * Call holding cpuset_mutex, so current's cpuset won't change |
c8d9c90c | 982 | * during this call, as manage_mutex holds off any cpuset_attach() |
e4e364e8 PJ |
983 | * calls. Therefore we don't need to take task_lock around the |
984 | * call to guarantee_online_mems(), as we know no one is changing | |
2df167a3 | 985 | * our task's cpuset. |
e4e364e8 | 986 | * |
e4e364e8 PJ |
987 | * While the mm_struct we are migrating is typically from some |
988 | * other task, the task_struct mems_allowed that we are hacking | |
989 | * is for our current task, which must allocate new pages for that | |
990 | * migrating memory region. | |
e4e364e8 PJ |
991 | */ |
992 | ||
993 | static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from, | |
994 | const nodemask_t *to) | |
995 | { | |
996 | struct task_struct *tsk = current; | |
070b57fc | 997 | struct cpuset *mems_cs; |
e4e364e8 | 998 | |
e4e364e8 | 999 | tsk->mems_allowed = *to; |
e4e364e8 PJ |
1000 | |
1001 | do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL); | |
1002 | ||
070b57fc LZ |
1003 | mems_cs = effective_nodemask_cpuset(task_cs(tsk)); |
1004 | guarantee_online_mems(mems_cs, &tsk->mems_allowed); | |
e4e364e8 PJ |
1005 | } |
1006 | ||
3b6766fe | 1007 | /* |
58568d2a MX |
1008 | * cpuset_change_task_nodemask - change task's mems_allowed and mempolicy |
1009 | * @tsk: the task to change | |
1010 | * @newmems: new nodes that the task will be set | |
1011 | * | |
1012 | * In order to avoid seeing no nodes if the old and new nodes are disjoint, | |
1013 | * we structure updates as setting all new allowed nodes, then clearing newly | |
1014 | * disallowed ones. | |
58568d2a MX |
1015 | */ |
1016 | static void cpuset_change_task_nodemask(struct task_struct *tsk, | |
1017 | nodemask_t *newmems) | |
1018 | { | |
b246272e | 1019 | bool need_loop; |
89e8a244 | 1020 | |
c0ff7453 MX |
1021 | /* |
1022 | * Allow tasks that have access to memory reserves because they have | |
1023 | * been OOM killed to get memory anywhere. | |
1024 | */ | |
1025 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
1026 | return; | |
1027 | if (current->flags & PF_EXITING) /* Let dying task have memory */ | |
1028 | return; | |
1029 | ||
1030 | task_lock(tsk); | |
b246272e DR |
1031 | /* |
1032 | * Determine if a loop is necessary if another thread is doing | |
1033 | * get_mems_allowed(). If at least one node remains unchanged and | |
1034 | * tsk does not have a mempolicy, then an empty nodemask will not be | |
1035 | * possible when mems_allowed is larger than a word. | |
1036 | */ | |
1037 | need_loop = task_has_mempolicy(tsk) || | |
1038 | !nodes_intersects(*newmems, tsk->mems_allowed); | |
c0ff7453 | 1039 | |
cc9a6c87 MG |
1040 | if (need_loop) |
1041 | write_seqcount_begin(&tsk->mems_allowed_seq); | |
c0ff7453 | 1042 | |
cc9a6c87 MG |
1043 | nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems); |
1044 | mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP1); | |
c0ff7453 MX |
1045 | |
1046 | mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP2); | |
58568d2a | 1047 | tsk->mems_allowed = *newmems; |
cc9a6c87 MG |
1048 | |
1049 | if (need_loop) | |
1050 | write_seqcount_end(&tsk->mems_allowed_seq); | |
1051 | ||
c0ff7453 | 1052 | task_unlock(tsk); |
58568d2a MX |
1053 | } |
1054 | ||
1055 | /* | |
1056 | * Update task's mems_allowed and rebind its mempolicy and vmas' mempolicy | |
1057 | * of it to cpuset's new mems_allowed, and migrate pages to new nodes if | |
5d21cc2d | 1058 | * memory_migrate flag is set. Called with cpuset_mutex held. |
3b6766fe LZ |
1059 | */ |
1060 | static void cpuset_change_nodemask(struct task_struct *p, | |
1061 | struct cgroup_scanner *scan) | |
1062 | { | |
6f4b7e63 | 1063 | struct cpuset *cs = cgroup_cs(scan->cgrp); |
3b6766fe | 1064 | struct mm_struct *mm; |
3b6766fe | 1065 | int migrate; |
33ad801d | 1066 | nodemask_t *newmems = scan->data; |
58568d2a | 1067 | |
33ad801d | 1068 | cpuset_change_task_nodemask(p, newmems); |
53feb297 | 1069 | |
3b6766fe LZ |
1070 | mm = get_task_mm(p); |
1071 | if (!mm) | |
1072 | return; | |
1073 | ||
3b6766fe LZ |
1074 | migrate = is_memory_migrate(cs); |
1075 | ||
1076 | mpol_rebind_mm(mm, &cs->mems_allowed); | |
1077 | if (migrate) | |
33ad801d | 1078 | cpuset_migrate_mm(mm, &cs->old_mems_allowed, newmems); |
3b6766fe LZ |
1079 | mmput(mm); |
1080 | } | |
1081 | ||
8793d854 PM |
1082 | static void *cpuset_being_rebound; |
1083 | ||
0b2f630a MX |
1084 | /** |
1085 | * update_tasks_nodemask - Update the nodemasks of tasks in the cpuset. | |
1086 | * @cs: the cpuset in which each task's mems_allowed mask needs to be changed | |
010cfac4 | 1087 | * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks() |
0b2f630a | 1088 | * |
5d21cc2d | 1089 | * Called with cpuset_mutex held |
010cfac4 LZ |
1090 | * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0 |
1091 | * if @heap != NULL. | |
0b2f630a | 1092 | */ |
33ad801d | 1093 | static void update_tasks_nodemask(struct cpuset *cs, struct ptr_heap *heap) |
1da177e4 | 1094 | { |
33ad801d | 1095 | static nodemask_t newmems; /* protected by cpuset_mutex */ |
3b6766fe | 1096 | struct cgroup_scanner scan; |
070b57fc | 1097 | struct cpuset *mems_cs = effective_nodemask_cpuset(cs); |
59dac16f | 1098 | |
846a16bf | 1099 | cpuset_being_rebound = cs; /* causes mpol_dup() rebind */ |
4225399a | 1100 | |
070b57fc | 1101 | guarantee_online_mems(mems_cs, &newmems); |
33ad801d | 1102 | |
6f4b7e63 | 1103 | scan.cgrp = cs->css.cgroup; |
3b6766fe LZ |
1104 | scan.test_task = NULL; |
1105 | scan.process_task = cpuset_change_nodemask; | |
010cfac4 | 1106 | scan.heap = heap; |
33ad801d | 1107 | scan.data = &newmems; |
4225399a PJ |
1108 | |
1109 | /* | |
3b6766fe LZ |
1110 | * The mpol_rebind_mm() call takes mmap_sem, which we couldn't |
1111 | * take while holding tasklist_lock. Forks can happen - the | |
1112 | * mpol_dup() cpuset_being_rebound check will catch such forks, | |
1113 | * and rebind their vma mempolicies too. Because we still hold | |
5d21cc2d | 1114 | * the global cpuset_mutex, we know that no other rebind effort |
3b6766fe | 1115 | * will be contending for the global variable cpuset_being_rebound. |
4225399a | 1116 | * It's ok if we rebind the same mm twice; mpol_rebind_mm() |
04c19fa6 | 1117 | * is idempotent. Also migrate pages in each mm to new nodes. |
4225399a | 1118 | */ |
010cfac4 | 1119 | cgroup_scan_tasks(&scan); |
4225399a | 1120 | |
33ad801d LZ |
1121 | /* |
1122 | * All the tasks' nodemasks have been updated, update | |
1123 | * cs->old_mems_allowed. | |
1124 | */ | |
1125 | cs->old_mems_allowed = newmems; | |
1126 | ||
2df167a3 | 1127 | /* We're done rebinding vmas to this cpuset's new mems_allowed. */ |
8793d854 | 1128 | cpuset_being_rebound = NULL; |
1da177e4 LT |
1129 | } |
1130 | ||
5c5cc623 LZ |
1131 | /* |
1132 | * update_tasks_nodemask_hier - Update the nodemasks of tasks in the hierarchy. | |
1133 | * @cs: the root cpuset of the hierarchy | |
1134 | * @update_root: update the root cpuset or not? | |
1135 | * @heap: the heap used by cgroup_scan_tasks() | |
1136 | * | |
1137 | * This will update nodemasks of tasks in @root_cs and all other empty cpusets | |
1138 | * which take on nodemask of @root_cs. | |
1139 | * | |
1140 | * Called with cpuset_mutex held | |
1141 | */ | |
1142 | static void update_tasks_nodemask_hier(struct cpuset *root_cs, | |
1143 | bool update_root, struct ptr_heap *heap) | |
1144 | { | |
1145 | struct cpuset *cp; | |
492eb21b | 1146 | struct cgroup_subsys_state *pos_css; |
5c5cc623 LZ |
1147 | |
1148 | if (update_root) | |
1149 | update_tasks_nodemask(root_cs, heap); | |
1150 | ||
1151 | rcu_read_lock(); | |
492eb21b | 1152 | cpuset_for_each_descendant_pre(cp, pos_css, root_cs) { |
5c5cc623 LZ |
1153 | /* skip the whole subtree if @cp have some CPU */ |
1154 | if (!nodes_empty(cp->mems_allowed)) { | |
492eb21b | 1155 | pos_css = css_rightmost_descendant(pos_css); |
5c5cc623 LZ |
1156 | continue; |
1157 | } | |
1158 | if (!css_tryget(&cp->css)) | |
1159 | continue; | |
1160 | rcu_read_unlock(); | |
1161 | ||
1162 | update_tasks_nodemask(cp, heap); | |
1163 | ||
1164 | rcu_read_lock(); | |
1165 | css_put(&cp->css); | |
1166 | } | |
1167 | rcu_read_unlock(); | |
1168 | } | |
1169 | ||
0b2f630a MX |
1170 | /* |
1171 | * Handle user request to change the 'mems' memory placement | |
1172 | * of a cpuset. Needs to validate the request, update the | |
58568d2a MX |
1173 | * cpusets mems_allowed, and for each task in the cpuset, |
1174 | * update mems_allowed and rebind task's mempolicy and any vma | |
1175 | * mempolicies and if the cpuset is marked 'memory_migrate', | |
1176 | * migrate the tasks pages to the new memory. | |
0b2f630a | 1177 | * |
5d21cc2d | 1178 | * Call with cpuset_mutex held. May take callback_mutex during call. |
0b2f630a MX |
1179 | * Will take tasklist_lock, scan tasklist for tasks in cpuset cs, |
1180 | * lock each such tasks mm->mmap_sem, scan its vma's and rebind | |
1181 | * their mempolicies to the cpusets new mems_allowed. | |
1182 | */ | |
645fcc9d LZ |
1183 | static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs, |
1184 | const char *buf) | |
0b2f630a | 1185 | { |
0b2f630a | 1186 | int retval; |
010cfac4 | 1187 | struct ptr_heap heap; |
0b2f630a MX |
1188 | |
1189 | /* | |
38d7bee9 | 1190 | * top_cpuset.mems_allowed tracks node_stats[N_MEMORY]; |
0b2f630a MX |
1191 | * it's read-only |
1192 | */ | |
53feb297 MX |
1193 | if (cs == &top_cpuset) { |
1194 | retval = -EACCES; | |
1195 | goto done; | |
1196 | } | |
0b2f630a | 1197 | |
0b2f630a MX |
1198 | /* |
1199 | * An empty mems_allowed is ok iff there are no tasks in the cpuset. | |
1200 | * Since nodelist_parse() fails on an empty mask, we special case | |
1201 | * that parsing. The validate_change() call ensures that cpusets | |
1202 | * with tasks have memory. | |
1203 | */ | |
1204 | if (!*buf) { | |
645fcc9d | 1205 | nodes_clear(trialcs->mems_allowed); |
0b2f630a | 1206 | } else { |
645fcc9d | 1207 | retval = nodelist_parse(buf, trialcs->mems_allowed); |
0b2f630a MX |
1208 | if (retval < 0) |
1209 | goto done; | |
1210 | ||
645fcc9d | 1211 | if (!nodes_subset(trialcs->mems_allowed, |
38d7bee9 | 1212 | node_states[N_MEMORY])) { |
53feb297 MX |
1213 | retval = -EINVAL; |
1214 | goto done; | |
1215 | } | |
0b2f630a | 1216 | } |
33ad801d LZ |
1217 | |
1218 | if (nodes_equal(cs->mems_allowed, trialcs->mems_allowed)) { | |
0b2f630a MX |
1219 | retval = 0; /* Too easy - nothing to do */ |
1220 | goto done; | |
1221 | } | |
645fcc9d | 1222 | retval = validate_change(cs, trialcs); |
0b2f630a MX |
1223 | if (retval < 0) |
1224 | goto done; | |
1225 | ||
010cfac4 LZ |
1226 | retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL); |
1227 | if (retval < 0) | |
1228 | goto done; | |
1229 | ||
0b2f630a | 1230 | mutex_lock(&callback_mutex); |
645fcc9d | 1231 | cs->mems_allowed = trialcs->mems_allowed; |
0b2f630a MX |
1232 | mutex_unlock(&callback_mutex); |
1233 | ||
5c5cc623 | 1234 | update_tasks_nodemask_hier(cs, true, &heap); |
010cfac4 LZ |
1235 | |
1236 | heap_free(&heap); | |
0b2f630a MX |
1237 | done: |
1238 | return retval; | |
1239 | } | |
1240 | ||
8793d854 PM |
1241 | int current_cpuset_is_being_rebound(void) |
1242 | { | |
1243 | return task_cs(current) == cpuset_being_rebound; | |
1244 | } | |
1245 | ||
5be7a479 | 1246 | static int update_relax_domain_level(struct cpuset *cs, s64 val) |
1d3504fc | 1247 | { |
db7f47cf | 1248 | #ifdef CONFIG_SMP |
60495e77 | 1249 | if (val < -1 || val >= sched_domain_level_max) |
30e0e178 | 1250 | return -EINVAL; |
db7f47cf | 1251 | #endif |
1d3504fc HS |
1252 | |
1253 | if (val != cs->relax_domain_level) { | |
1254 | cs->relax_domain_level = val; | |
300ed6cb LZ |
1255 | if (!cpumask_empty(cs->cpus_allowed) && |
1256 | is_sched_load_balance(cs)) | |
699140ba | 1257 | rebuild_sched_domains_locked(); |
1d3504fc HS |
1258 | } |
1259 | ||
1260 | return 0; | |
1261 | } | |
1262 | ||
950592f7 MX |
1263 | /* |
1264 | * cpuset_change_flag - make a task's spread flags the same as its cpuset's | |
1265 | * @tsk: task to be updated | |
1266 | * @scan: struct cgroup_scanner containing the cgroup of the task | |
1267 | * | |
1268 | * Called by cgroup_scan_tasks() for each task in a cgroup. | |
1269 | * | |
1270 | * We don't need to re-check for the cgroup/cpuset membership, since we're | |
5d21cc2d | 1271 | * holding cpuset_mutex at this point. |
950592f7 MX |
1272 | */ |
1273 | static void cpuset_change_flag(struct task_struct *tsk, | |
1274 | struct cgroup_scanner *scan) | |
1275 | { | |
6f4b7e63 | 1276 | cpuset_update_task_spread_flag(cgroup_cs(scan->cgrp), tsk); |
950592f7 MX |
1277 | } |
1278 | ||
1279 | /* | |
1280 | * update_tasks_flags - update the spread flags of tasks in the cpuset. | |
1281 | * @cs: the cpuset in which each task's spread flags needs to be changed | |
1282 | * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks() | |
1283 | * | |
5d21cc2d | 1284 | * Called with cpuset_mutex held |
950592f7 MX |
1285 | * |
1286 | * The cgroup_scan_tasks() function will scan all the tasks in a cgroup, | |
1287 | * calling callback functions for each. | |
1288 | * | |
1289 | * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0 | |
1290 | * if @heap != NULL. | |
1291 | */ | |
1292 | static void update_tasks_flags(struct cpuset *cs, struct ptr_heap *heap) | |
1293 | { | |
1294 | struct cgroup_scanner scan; | |
1295 | ||
6f4b7e63 | 1296 | scan.cgrp = cs->css.cgroup; |
950592f7 MX |
1297 | scan.test_task = NULL; |
1298 | scan.process_task = cpuset_change_flag; | |
1299 | scan.heap = heap; | |
1300 | cgroup_scan_tasks(&scan); | |
1301 | } | |
1302 | ||
1da177e4 LT |
1303 | /* |
1304 | * update_flag - read a 0 or a 1 in a file and update associated flag | |
78608366 PM |
1305 | * bit: the bit to update (see cpuset_flagbits_t) |
1306 | * cs: the cpuset to update | |
1307 | * turning_on: whether the flag is being set or cleared | |
053199ed | 1308 | * |
5d21cc2d | 1309 | * Call with cpuset_mutex held. |
1da177e4 LT |
1310 | */ |
1311 | ||
700fe1ab PM |
1312 | static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs, |
1313 | int turning_on) | |
1da177e4 | 1314 | { |
645fcc9d | 1315 | struct cpuset *trialcs; |
40b6a762 | 1316 | int balance_flag_changed; |
950592f7 MX |
1317 | int spread_flag_changed; |
1318 | struct ptr_heap heap; | |
1319 | int err; | |
1da177e4 | 1320 | |
645fcc9d LZ |
1321 | trialcs = alloc_trial_cpuset(cs); |
1322 | if (!trialcs) | |
1323 | return -ENOMEM; | |
1324 | ||
1da177e4 | 1325 | if (turning_on) |
645fcc9d | 1326 | set_bit(bit, &trialcs->flags); |
1da177e4 | 1327 | else |
645fcc9d | 1328 | clear_bit(bit, &trialcs->flags); |
1da177e4 | 1329 | |
645fcc9d | 1330 | err = validate_change(cs, trialcs); |
85d7b949 | 1331 | if (err < 0) |
645fcc9d | 1332 | goto out; |
029190c5 | 1333 | |
950592f7 MX |
1334 | err = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL); |
1335 | if (err < 0) | |
1336 | goto out; | |
1337 | ||
029190c5 | 1338 | balance_flag_changed = (is_sched_load_balance(cs) != |
645fcc9d | 1339 | is_sched_load_balance(trialcs)); |
029190c5 | 1340 | |
950592f7 MX |
1341 | spread_flag_changed = ((is_spread_slab(cs) != is_spread_slab(trialcs)) |
1342 | || (is_spread_page(cs) != is_spread_page(trialcs))); | |
1343 | ||
3d3f26a7 | 1344 | mutex_lock(&callback_mutex); |
645fcc9d | 1345 | cs->flags = trialcs->flags; |
3d3f26a7 | 1346 | mutex_unlock(&callback_mutex); |
85d7b949 | 1347 | |
300ed6cb | 1348 | if (!cpumask_empty(trialcs->cpus_allowed) && balance_flag_changed) |
699140ba | 1349 | rebuild_sched_domains_locked(); |
029190c5 | 1350 | |
950592f7 MX |
1351 | if (spread_flag_changed) |
1352 | update_tasks_flags(cs, &heap); | |
1353 | heap_free(&heap); | |
645fcc9d LZ |
1354 | out: |
1355 | free_trial_cpuset(trialcs); | |
1356 | return err; | |
1da177e4 LT |
1357 | } |
1358 | ||
3e0d98b9 | 1359 | /* |
80f7228b | 1360 | * Frequency meter - How fast is some event occurring? |
3e0d98b9 PJ |
1361 | * |
1362 | * These routines manage a digitally filtered, constant time based, | |
1363 | * event frequency meter. There are four routines: | |
1364 | * fmeter_init() - initialize a frequency meter. | |
1365 | * fmeter_markevent() - called each time the event happens. | |
1366 | * fmeter_getrate() - returns the recent rate of such events. | |
1367 | * fmeter_update() - internal routine used to update fmeter. | |
1368 | * | |
1369 | * A common data structure is passed to each of these routines, | |
1370 | * which is used to keep track of the state required to manage the | |
1371 | * frequency meter and its digital filter. | |
1372 | * | |
1373 | * The filter works on the number of events marked per unit time. | |
1374 | * The filter is single-pole low-pass recursive (IIR). The time unit | |
1375 | * is 1 second. Arithmetic is done using 32-bit integers scaled to | |
1376 | * simulate 3 decimal digits of precision (multiplied by 1000). | |
1377 | * | |
1378 | * With an FM_COEF of 933, and a time base of 1 second, the filter | |
1379 | * has a half-life of 10 seconds, meaning that if the events quit | |
1380 | * happening, then the rate returned from the fmeter_getrate() | |
1381 | * will be cut in half each 10 seconds, until it converges to zero. | |
1382 | * | |
1383 | * It is not worth doing a real infinitely recursive filter. If more | |
1384 | * than FM_MAXTICKS ticks have elapsed since the last filter event, | |
1385 | * just compute FM_MAXTICKS ticks worth, by which point the level | |
1386 | * will be stable. | |
1387 | * | |
1388 | * Limit the count of unprocessed events to FM_MAXCNT, so as to avoid | |
1389 | * arithmetic overflow in the fmeter_update() routine. | |
1390 | * | |
1391 | * Given the simple 32 bit integer arithmetic used, this meter works | |
1392 | * best for reporting rates between one per millisecond (msec) and | |
1393 | * one per 32 (approx) seconds. At constant rates faster than one | |
1394 | * per msec it maxes out at values just under 1,000,000. At constant | |
1395 | * rates between one per msec, and one per second it will stabilize | |
1396 | * to a value N*1000, where N is the rate of events per second. | |
1397 | * At constant rates between one per second and one per 32 seconds, | |
1398 | * it will be choppy, moving up on the seconds that have an event, | |
1399 | * and then decaying until the next event. At rates slower than | |
1400 | * about one in 32 seconds, it decays all the way back to zero between | |
1401 | * each event. | |
1402 | */ | |
1403 | ||
1404 | #define FM_COEF 933 /* coefficient for half-life of 10 secs */ | |
1405 | #define FM_MAXTICKS ((time_t)99) /* useless computing more ticks than this */ | |
1406 | #define FM_MAXCNT 1000000 /* limit cnt to avoid overflow */ | |
1407 | #define FM_SCALE 1000 /* faux fixed point scale */ | |
1408 | ||
1409 | /* Initialize a frequency meter */ | |
1410 | static void fmeter_init(struct fmeter *fmp) | |
1411 | { | |
1412 | fmp->cnt = 0; | |
1413 | fmp->val = 0; | |
1414 | fmp->time = 0; | |
1415 | spin_lock_init(&fmp->lock); | |
1416 | } | |
1417 | ||
1418 | /* Internal meter update - process cnt events and update value */ | |
1419 | static void fmeter_update(struct fmeter *fmp) | |
1420 | { | |
1421 | time_t now = get_seconds(); | |
1422 | time_t ticks = now - fmp->time; | |
1423 | ||
1424 | if (ticks == 0) | |
1425 | return; | |
1426 | ||
1427 | ticks = min(FM_MAXTICKS, ticks); | |
1428 | while (ticks-- > 0) | |
1429 | fmp->val = (FM_COEF * fmp->val) / FM_SCALE; | |
1430 | fmp->time = now; | |
1431 | ||
1432 | fmp->val += ((FM_SCALE - FM_COEF) * fmp->cnt) / FM_SCALE; | |
1433 | fmp->cnt = 0; | |
1434 | } | |
1435 | ||
1436 | /* Process any previous ticks, then bump cnt by one (times scale). */ | |
1437 | static void fmeter_markevent(struct fmeter *fmp) | |
1438 | { | |
1439 | spin_lock(&fmp->lock); | |
1440 | fmeter_update(fmp); | |
1441 | fmp->cnt = min(FM_MAXCNT, fmp->cnt + FM_SCALE); | |
1442 | spin_unlock(&fmp->lock); | |
1443 | } | |
1444 | ||
1445 | /* Process any previous ticks, then return current value. */ | |
1446 | static int fmeter_getrate(struct fmeter *fmp) | |
1447 | { | |
1448 | int val; | |
1449 | ||
1450 | spin_lock(&fmp->lock); | |
1451 | fmeter_update(fmp); | |
1452 | val = fmp->val; | |
1453 | spin_unlock(&fmp->lock); | |
1454 | return val; | |
1455 | } | |
1456 | ||
5d21cc2d | 1457 | /* Called by cgroups to determine if a cpuset is usable; cpuset_mutex held */ |
eb95419b TH |
1458 | static int cpuset_can_attach(struct cgroup_subsys_state *css, |
1459 | struct cgroup_taskset *tset) | |
f780bdb7 | 1460 | { |
eb95419b | 1461 | struct cpuset *cs = css_cs(css); |
bb9d97b6 TH |
1462 | struct task_struct *task; |
1463 | int ret; | |
1da177e4 | 1464 | |
5d21cc2d TH |
1465 | mutex_lock(&cpuset_mutex); |
1466 | ||
88fa523b LZ |
1467 | /* |
1468 | * We allow to move tasks into an empty cpuset if sane_behavior | |
1469 | * flag is set. | |
1470 | */ | |
5d21cc2d | 1471 | ret = -ENOSPC; |
eb95419b | 1472 | if (!cgroup_sane_behavior(css->cgroup) && |
88fa523b | 1473 | (cpumask_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed))) |
5d21cc2d | 1474 | goto out_unlock; |
9985b0ba | 1475 | |
eb95419b | 1476 | cgroup_taskset_for_each(task, css->cgroup, tset) { |
bb9d97b6 | 1477 | /* |
14a40ffc TH |
1478 | * Kthreads which disallow setaffinity shouldn't be moved |
1479 | * to a new cpuset; we don't want to change their cpu | |
1480 | * affinity and isolating such threads by their set of | |
1481 | * allowed nodes is unnecessary. Thus, cpusets are not | |
1482 | * applicable for such threads. This prevents checking for | |
1483 | * success of set_cpus_allowed_ptr() on all attached tasks | |
1484 | * before cpus_allowed may be changed. | |
bb9d97b6 | 1485 | */ |
5d21cc2d | 1486 | ret = -EINVAL; |
14a40ffc | 1487 | if (task->flags & PF_NO_SETAFFINITY) |
5d21cc2d TH |
1488 | goto out_unlock; |
1489 | ret = security_task_setscheduler(task); | |
1490 | if (ret) | |
1491 | goto out_unlock; | |
bb9d97b6 | 1492 | } |
f780bdb7 | 1493 | |
452477fa TH |
1494 | /* |
1495 | * Mark attach is in progress. This makes validate_change() fail | |
1496 | * changes which zero cpus/mems_allowed. | |
1497 | */ | |
1498 | cs->attach_in_progress++; | |
5d21cc2d TH |
1499 | ret = 0; |
1500 | out_unlock: | |
1501 | mutex_unlock(&cpuset_mutex); | |
1502 | return ret; | |
8793d854 | 1503 | } |
f780bdb7 | 1504 | |
eb95419b | 1505 | static void cpuset_cancel_attach(struct cgroup_subsys_state *css, |
452477fa TH |
1506 | struct cgroup_taskset *tset) |
1507 | { | |
5d21cc2d | 1508 | mutex_lock(&cpuset_mutex); |
eb95419b | 1509 | css_cs(css)->attach_in_progress--; |
5d21cc2d | 1510 | mutex_unlock(&cpuset_mutex); |
8793d854 | 1511 | } |
1da177e4 | 1512 | |
4e4c9a14 | 1513 | /* |
5d21cc2d | 1514 | * Protected by cpuset_mutex. cpus_attach is used only by cpuset_attach() |
4e4c9a14 TH |
1515 | * but we can't allocate it dynamically there. Define it global and |
1516 | * allocate from cpuset_init(). | |
1517 | */ | |
1518 | static cpumask_var_t cpus_attach; | |
1519 | ||
eb95419b TH |
1520 | static void cpuset_attach(struct cgroup_subsys_state *css, |
1521 | 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 | 1528 | struct cgroup *oldcgrp = cgroup_taskset_cur_cgroup(tset); |
eb95419b | 1529 | struct cpuset *cs = css_cs(css); |
2f7ee569 | 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 | |
eb95419b | 1544 | cgroup_taskset_for_each(task, css->cgroup, tset) { |
bb9d97b6 TH |
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 | ||
182446d0 TH |
1606 | static int cpuset_write_u64(struct cgroup_subsys_state *css, struct cftype *cft, |
1607 | u64 val) | |
700fe1ab | 1608 | { |
182446d0 | 1609 | struct cpuset *cs = css_cs(css); |
700fe1ab | 1610 | cpuset_filetype_t type = cft->private; |
5d21cc2d | 1611 | int retval = -ENODEV; |
700fe1ab | 1612 | |
5d21cc2d TH |
1613 | mutex_lock(&cpuset_mutex); |
1614 | if (!is_cpuset_online(cs)) | |
1615 | goto out_unlock; | |
700fe1ab PM |
1616 | |
1617 | switch (type) { | |
1da177e4 | 1618 | case FILE_CPU_EXCLUSIVE: |
700fe1ab | 1619 | retval = update_flag(CS_CPU_EXCLUSIVE, cs, val); |
1da177e4 LT |
1620 | break; |
1621 | case FILE_MEM_EXCLUSIVE: | |
700fe1ab | 1622 | retval = update_flag(CS_MEM_EXCLUSIVE, cs, val); |
1da177e4 | 1623 | break; |
78608366 PM |
1624 | case FILE_MEM_HARDWALL: |
1625 | retval = update_flag(CS_MEM_HARDWALL, cs, val); | |
1626 | break; | |
029190c5 | 1627 | case FILE_SCHED_LOAD_BALANCE: |
700fe1ab | 1628 | retval = update_flag(CS_SCHED_LOAD_BALANCE, cs, val); |
1d3504fc | 1629 | break; |
45b07ef3 | 1630 | case FILE_MEMORY_MIGRATE: |
700fe1ab | 1631 | retval = update_flag(CS_MEMORY_MIGRATE, cs, val); |
45b07ef3 | 1632 | break; |
3e0d98b9 | 1633 | case FILE_MEMORY_PRESSURE_ENABLED: |
700fe1ab | 1634 | cpuset_memory_pressure_enabled = !!val; |
3e0d98b9 PJ |
1635 | break; |
1636 | case FILE_MEMORY_PRESSURE: | |
1637 | retval = -EACCES; | |
1638 | break; | |
825a46af | 1639 | case FILE_SPREAD_PAGE: |
700fe1ab | 1640 | retval = update_flag(CS_SPREAD_PAGE, cs, val); |
825a46af PJ |
1641 | break; |
1642 | case FILE_SPREAD_SLAB: | |
700fe1ab | 1643 | retval = update_flag(CS_SPREAD_SLAB, cs, val); |
825a46af | 1644 | break; |
1da177e4 LT |
1645 | default: |
1646 | retval = -EINVAL; | |
700fe1ab | 1647 | break; |
1da177e4 | 1648 | } |
5d21cc2d TH |
1649 | out_unlock: |
1650 | mutex_unlock(&cpuset_mutex); | |
1da177e4 LT |
1651 | return retval; |
1652 | } | |
1653 | ||
182446d0 TH |
1654 | static int cpuset_write_s64(struct cgroup_subsys_state *css, struct cftype *cft, |
1655 | s64 val) | |
5be7a479 | 1656 | { |
182446d0 | 1657 | struct cpuset *cs = css_cs(css); |
5be7a479 | 1658 | cpuset_filetype_t type = cft->private; |
5d21cc2d | 1659 | int retval = -ENODEV; |
5be7a479 | 1660 | |
5d21cc2d TH |
1661 | mutex_lock(&cpuset_mutex); |
1662 | if (!is_cpuset_online(cs)) | |
1663 | goto out_unlock; | |
e3712395 | 1664 | |
5be7a479 PM |
1665 | switch (type) { |
1666 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1667 | retval = update_relax_domain_level(cs, val); | |
1668 | break; | |
1669 | default: | |
1670 | retval = -EINVAL; | |
1671 | break; | |
1672 | } | |
5d21cc2d TH |
1673 | out_unlock: |
1674 | mutex_unlock(&cpuset_mutex); | |
5be7a479 PM |
1675 | return retval; |
1676 | } | |
1677 | ||
e3712395 PM |
1678 | /* |
1679 | * Common handling for a write to a "cpus" or "mems" file. | |
1680 | */ | |
182446d0 TH |
1681 | static int cpuset_write_resmask(struct cgroup_subsys_state *css, |
1682 | struct cftype *cft, const char *buf) | |
e3712395 | 1683 | { |
182446d0 | 1684 | struct cpuset *cs = css_cs(css); |
645fcc9d | 1685 | struct cpuset *trialcs; |
5d21cc2d | 1686 | int retval = -ENODEV; |
e3712395 | 1687 | |
3a5a6d0c TH |
1688 | /* |
1689 | * CPU or memory hotunplug may leave @cs w/o any execution | |
1690 | * resources, in which case the hotplug code asynchronously updates | |
1691 | * configuration and transfers all tasks to the nearest ancestor | |
1692 | * which can execute. | |
1693 | * | |
1694 | * As writes to "cpus" or "mems" may restore @cs's execution | |
1695 | * resources, wait for the previously scheduled operations before | |
1696 | * proceeding, so that we don't end up keep removing tasks added | |
1697 | * after execution capability is restored. | |
1698 | */ | |
1699 | flush_work(&cpuset_hotplug_work); | |
1700 | ||
5d21cc2d TH |
1701 | mutex_lock(&cpuset_mutex); |
1702 | if (!is_cpuset_online(cs)) | |
1703 | goto out_unlock; | |
e3712395 | 1704 | |
645fcc9d | 1705 | trialcs = alloc_trial_cpuset(cs); |
b75f38d6 LZ |
1706 | if (!trialcs) { |
1707 | retval = -ENOMEM; | |
5d21cc2d | 1708 | goto out_unlock; |
b75f38d6 | 1709 | } |
645fcc9d | 1710 | |
e3712395 PM |
1711 | switch (cft->private) { |
1712 | case FILE_CPULIST: | |
645fcc9d | 1713 | retval = update_cpumask(cs, trialcs, buf); |
e3712395 PM |
1714 | break; |
1715 | case FILE_MEMLIST: | |
645fcc9d | 1716 | retval = update_nodemask(cs, trialcs, buf); |
e3712395 PM |
1717 | break; |
1718 | default: | |
1719 | retval = -EINVAL; | |
1720 | break; | |
1721 | } | |
645fcc9d LZ |
1722 | |
1723 | free_trial_cpuset(trialcs); | |
5d21cc2d TH |
1724 | out_unlock: |
1725 | mutex_unlock(&cpuset_mutex); | |
e3712395 PM |
1726 | return retval; |
1727 | } | |
1728 | ||
1da177e4 LT |
1729 | /* |
1730 | * These ascii lists should be read in a single call, by using a user | |
1731 | * buffer large enough to hold the entire map. If read in smaller | |
1732 | * chunks, there is no guarantee of atomicity. Since the display format | |
1733 | * used, list of ranges of sequential numbers, is variable length, | |
1734 | * and since these maps can change value dynamically, one could read | |
1735 | * gibberish by doing partial reads while a list was changing. | |
1736 | * A single large read to a buffer that crosses a page boundary is | |
1737 | * ok, because the result being copied to user land is not recomputed | |
1738 | * across a page fault. | |
1739 | */ | |
1740 | ||
9303e0c4 | 1741 | static size_t cpuset_sprintf_cpulist(char *page, struct cpuset *cs) |
1da177e4 | 1742 | { |
9303e0c4 | 1743 | size_t count; |
1da177e4 | 1744 | |
3d3f26a7 | 1745 | mutex_lock(&callback_mutex); |
9303e0c4 | 1746 | count = cpulist_scnprintf(page, PAGE_SIZE, cs->cpus_allowed); |
3d3f26a7 | 1747 | mutex_unlock(&callback_mutex); |
1da177e4 | 1748 | |
9303e0c4 | 1749 | return count; |
1da177e4 LT |
1750 | } |
1751 | ||
9303e0c4 | 1752 | static size_t cpuset_sprintf_memlist(char *page, struct cpuset *cs) |
1da177e4 | 1753 | { |
9303e0c4 | 1754 | size_t count; |
1da177e4 | 1755 | |
3d3f26a7 | 1756 | mutex_lock(&callback_mutex); |
9303e0c4 | 1757 | count = nodelist_scnprintf(page, PAGE_SIZE, cs->mems_allowed); |
3d3f26a7 | 1758 | mutex_unlock(&callback_mutex); |
1da177e4 | 1759 | |
9303e0c4 | 1760 | return count; |
1da177e4 LT |
1761 | } |
1762 | ||
182446d0 TH |
1763 | static ssize_t cpuset_common_file_read(struct cgroup_subsys_state *css, |
1764 | struct cftype *cft, struct file *file, | |
1765 | char __user *buf, size_t nbytes, | |
1766 | loff_t *ppos) | |
1da177e4 | 1767 | { |
182446d0 | 1768 | struct cpuset *cs = css_cs(css); |
1da177e4 LT |
1769 | cpuset_filetype_t type = cft->private; |
1770 | char *page; | |
1771 | ssize_t retval = 0; | |
1772 | char *s; | |
1da177e4 | 1773 | |
e12ba74d | 1774 | if (!(page = (char *)__get_free_page(GFP_TEMPORARY))) |
1da177e4 LT |
1775 | return -ENOMEM; |
1776 | ||
1777 | s = page; | |
1778 | ||
1779 | switch (type) { | |
1780 | case FILE_CPULIST: | |
1781 | s += cpuset_sprintf_cpulist(s, cs); | |
1782 | break; | |
1783 | case FILE_MEMLIST: | |
1784 | s += cpuset_sprintf_memlist(s, cs); | |
1785 | break; | |
1da177e4 LT |
1786 | default: |
1787 | retval = -EINVAL; | |
1788 | goto out; | |
1789 | } | |
1790 | *s++ = '\n'; | |
1da177e4 | 1791 | |
eacaa1f5 | 1792 | retval = simple_read_from_buffer(buf, nbytes, ppos, page, s - page); |
1da177e4 LT |
1793 | out: |
1794 | free_page((unsigned long)page); | |
1795 | return retval; | |
1796 | } | |
1797 | ||
182446d0 | 1798 | static u64 cpuset_read_u64(struct cgroup_subsys_state *css, struct cftype *cft) |
700fe1ab | 1799 | { |
182446d0 | 1800 | struct cpuset *cs = css_cs(css); |
700fe1ab PM |
1801 | cpuset_filetype_t type = cft->private; |
1802 | switch (type) { | |
1803 | case FILE_CPU_EXCLUSIVE: | |
1804 | return is_cpu_exclusive(cs); | |
1805 | case FILE_MEM_EXCLUSIVE: | |
1806 | return is_mem_exclusive(cs); | |
78608366 PM |
1807 | case FILE_MEM_HARDWALL: |
1808 | return is_mem_hardwall(cs); | |
700fe1ab PM |
1809 | case FILE_SCHED_LOAD_BALANCE: |
1810 | return is_sched_load_balance(cs); | |
1811 | case FILE_MEMORY_MIGRATE: | |
1812 | return is_memory_migrate(cs); | |
1813 | case FILE_MEMORY_PRESSURE_ENABLED: | |
1814 | return cpuset_memory_pressure_enabled; | |
1815 | case FILE_MEMORY_PRESSURE: | |
1816 | return fmeter_getrate(&cs->fmeter); | |
1817 | case FILE_SPREAD_PAGE: | |
1818 | return is_spread_page(cs); | |
1819 | case FILE_SPREAD_SLAB: | |
1820 | return is_spread_slab(cs); | |
1821 | default: | |
1822 | BUG(); | |
1823 | } | |
cf417141 MK |
1824 | |
1825 | /* Unreachable but makes gcc happy */ | |
1826 | return 0; | |
700fe1ab | 1827 | } |
1da177e4 | 1828 | |
182446d0 | 1829 | static s64 cpuset_read_s64(struct cgroup_subsys_state *css, struct cftype *cft) |
5be7a479 | 1830 | { |
182446d0 | 1831 | struct cpuset *cs = css_cs(css); |
5be7a479 PM |
1832 | cpuset_filetype_t type = cft->private; |
1833 | switch (type) { | |
1834 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1835 | return cs->relax_domain_level; | |
1836 | default: | |
1837 | BUG(); | |
1838 | } | |
cf417141 MK |
1839 | |
1840 | /* Unrechable but makes gcc happy */ | |
1841 | return 0; | |
5be7a479 PM |
1842 | } |
1843 | ||
1da177e4 LT |
1844 | |
1845 | /* | |
1846 | * for the common functions, 'private' gives the type of file | |
1847 | */ | |
1848 | ||
addf2c73 PM |
1849 | static struct cftype files[] = { |
1850 | { | |
1851 | .name = "cpus", | |
1852 | .read = cpuset_common_file_read, | |
e3712395 PM |
1853 | .write_string = cpuset_write_resmask, |
1854 | .max_write_len = (100U + 6 * NR_CPUS), | |
addf2c73 PM |
1855 | .private = FILE_CPULIST, |
1856 | }, | |
1857 | ||
1858 | { | |
1859 | .name = "mems", | |
1860 | .read = cpuset_common_file_read, | |
e3712395 PM |
1861 | .write_string = cpuset_write_resmask, |
1862 | .max_write_len = (100U + 6 * MAX_NUMNODES), | |
addf2c73 PM |
1863 | .private = FILE_MEMLIST, |
1864 | }, | |
1865 | ||
1866 | { | |
1867 | .name = "cpu_exclusive", | |
1868 | .read_u64 = cpuset_read_u64, | |
1869 | .write_u64 = cpuset_write_u64, | |
1870 | .private = FILE_CPU_EXCLUSIVE, | |
1871 | }, | |
1872 | ||
1873 | { | |
1874 | .name = "mem_exclusive", | |
1875 | .read_u64 = cpuset_read_u64, | |
1876 | .write_u64 = cpuset_write_u64, | |
1877 | .private = FILE_MEM_EXCLUSIVE, | |
1878 | }, | |
1879 | ||
78608366 PM |
1880 | { |
1881 | .name = "mem_hardwall", | |
1882 | .read_u64 = cpuset_read_u64, | |
1883 | .write_u64 = cpuset_write_u64, | |
1884 | .private = FILE_MEM_HARDWALL, | |
1885 | }, | |
1886 | ||
addf2c73 PM |
1887 | { |
1888 | .name = "sched_load_balance", | |
1889 | .read_u64 = cpuset_read_u64, | |
1890 | .write_u64 = cpuset_write_u64, | |
1891 | .private = FILE_SCHED_LOAD_BALANCE, | |
1892 | }, | |
1893 | ||
1894 | { | |
1895 | .name = "sched_relax_domain_level", | |
5be7a479 PM |
1896 | .read_s64 = cpuset_read_s64, |
1897 | .write_s64 = cpuset_write_s64, | |
addf2c73 PM |
1898 | .private = FILE_SCHED_RELAX_DOMAIN_LEVEL, |
1899 | }, | |
1900 | ||
1901 | { | |
1902 | .name = "memory_migrate", | |
1903 | .read_u64 = cpuset_read_u64, | |
1904 | .write_u64 = cpuset_write_u64, | |
1905 | .private = FILE_MEMORY_MIGRATE, | |
1906 | }, | |
1907 | ||
1908 | { | |
1909 | .name = "memory_pressure", | |
1910 | .read_u64 = cpuset_read_u64, | |
1911 | .write_u64 = cpuset_write_u64, | |
1912 | .private = FILE_MEMORY_PRESSURE, | |
099fca32 | 1913 | .mode = S_IRUGO, |
addf2c73 PM |
1914 | }, |
1915 | ||
1916 | { | |
1917 | .name = "memory_spread_page", | |
1918 | .read_u64 = cpuset_read_u64, | |
1919 | .write_u64 = cpuset_write_u64, | |
1920 | .private = FILE_SPREAD_PAGE, | |
1921 | }, | |
1922 | ||
1923 | { | |
1924 | .name = "memory_spread_slab", | |
1925 | .read_u64 = cpuset_read_u64, | |
1926 | .write_u64 = cpuset_write_u64, | |
1927 | .private = FILE_SPREAD_SLAB, | |
1928 | }, | |
3e0d98b9 | 1929 | |
4baf6e33 TH |
1930 | { |
1931 | .name = "memory_pressure_enabled", | |
1932 | .flags = CFTYPE_ONLY_ON_ROOT, | |
1933 | .read_u64 = cpuset_read_u64, | |
1934 | .write_u64 = cpuset_write_u64, | |
1935 | .private = FILE_MEMORY_PRESSURE_ENABLED, | |
1936 | }, | |
1da177e4 | 1937 | |
4baf6e33 TH |
1938 | { } /* terminate */ |
1939 | }; | |
1da177e4 LT |
1940 | |
1941 | /* | |
92fb9748 | 1942 | * cpuset_css_alloc - allocate a cpuset css |
c9e5fe66 | 1943 | * cgrp: control group that the new cpuset will be part of |
1da177e4 LT |
1944 | */ |
1945 | ||
eb95419b TH |
1946 | static struct cgroup_subsys_state * |
1947 | cpuset_css_alloc(struct cgroup_subsys_state *parent_css) | |
1da177e4 | 1948 | { |
c8f699bb | 1949 | struct cpuset *cs; |
1da177e4 | 1950 | |
eb95419b | 1951 | if (!parent_css) |
8793d854 | 1952 | return &top_cpuset.css; |
033fa1c5 | 1953 | |
c8f699bb | 1954 | cs = kzalloc(sizeof(*cs), GFP_KERNEL); |
1da177e4 | 1955 | if (!cs) |
8793d854 | 1956 | return ERR_PTR(-ENOMEM); |
300ed6cb LZ |
1957 | if (!alloc_cpumask_var(&cs->cpus_allowed, GFP_KERNEL)) { |
1958 | kfree(cs); | |
1959 | return ERR_PTR(-ENOMEM); | |
1960 | } | |
1da177e4 | 1961 | |
029190c5 | 1962 | set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); |
300ed6cb | 1963 | cpumask_clear(cs->cpus_allowed); |
f9a86fcb | 1964 | nodes_clear(cs->mems_allowed); |
3e0d98b9 | 1965 | fmeter_init(&cs->fmeter); |
1d3504fc | 1966 | cs->relax_domain_level = -1; |
1da177e4 | 1967 | |
c8f699bb TH |
1968 | return &cs->css; |
1969 | } | |
1970 | ||
eb95419b | 1971 | static int cpuset_css_online(struct cgroup_subsys_state *css) |
c8f699bb | 1972 | { |
eb95419b | 1973 | struct cpuset *cs = css_cs(css); |
c431069f | 1974 | struct cpuset *parent = parent_cs(cs); |
ae8086ce | 1975 | struct cpuset *tmp_cs; |
492eb21b | 1976 | struct cgroup_subsys_state *pos_css; |
c8f699bb TH |
1977 | |
1978 | if (!parent) | |
1979 | return 0; | |
1980 | ||
5d21cc2d TH |
1981 | mutex_lock(&cpuset_mutex); |
1982 | ||
efeb77b2 | 1983 | set_bit(CS_ONLINE, &cs->flags); |
c8f699bb TH |
1984 | if (is_spread_page(parent)) |
1985 | set_bit(CS_SPREAD_PAGE, &cs->flags); | |
1986 | if (is_spread_slab(parent)) | |
1987 | set_bit(CS_SPREAD_SLAB, &cs->flags); | |
1da177e4 | 1988 | |
202f72d5 | 1989 | number_of_cpusets++; |
033fa1c5 | 1990 | |
eb95419b | 1991 | if (!test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags)) |
5d21cc2d | 1992 | goto out_unlock; |
033fa1c5 TH |
1993 | |
1994 | /* | |
1995 | * Clone @parent's configuration if CGRP_CPUSET_CLONE_CHILDREN is | |
1996 | * set. This flag handling is implemented in cgroup core for | |
1997 | * histrical reasons - the flag may be specified during mount. | |
1998 | * | |
1999 | * Currently, if any sibling cpusets have exclusive cpus or mem, we | |
2000 | * refuse to clone the configuration - thereby refusing the task to | |
2001 | * be entered, and as a result refusing the sys_unshare() or | |
2002 | * clone() which initiated it. If this becomes a problem for some | |
2003 | * users who wish to allow that scenario, then this could be | |
2004 | * changed to grant parent->cpus_allowed-sibling_cpus_exclusive | |
2005 | * (and likewise for mems) to the new cgroup. | |
2006 | */ | |
ae8086ce | 2007 | rcu_read_lock(); |
492eb21b | 2008 | cpuset_for_each_child(tmp_cs, pos_css, parent) { |
ae8086ce TH |
2009 | if (is_mem_exclusive(tmp_cs) || is_cpu_exclusive(tmp_cs)) { |
2010 | rcu_read_unlock(); | |
5d21cc2d | 2011 | goto out_unlock; |
ae8086ce | 2012 | } |
033fa1c5 | 2013 | } |
ae8086ce | 2014 | rcu_read_unlock(); |
033fa1c5 TH |
2015 | |
2016 | mutex_lock(&callback_mutex); | |
2017 | cs->mems_allowed = parent->mems_allowed; | |
2018 | cpumask_copy(cs->cpus_allowed, parent->cpus_allowed); | |
2019 | mutex_unlock(&callback_mutex); | |
5d21cc2d TH |
2020 | out_unlock: |
2021 | mutex_unlock(&cpuset_mutex); | |
c8f699bb TH |
2022 | return 0; |
2023 | } | |
2024 | ||
0b9e6965 ZH |
2025 | /* |
2026 | * If the cpuset being removed has its flag 'sched_load_balance' | |
2027 | * enabled, then simulate turning sched_load_balance off, which | |
2028 | * will call rebuild_sched_domains_locked(). | |
2029 | */ | |
2030 | ||
eb95419b | 2031 | static void cpuset_css_offline(struct cgroup_subsys_state *css) |
c8f699bb | 2032 | { |
eb95419b | 2033 | struct cpuset *cs = css_cs(css); |
c8f699bb | 2034 | |
5d21cc2d | 2035 | mutex_lock(&cpuset_mutex); |
c8f699bb TH |
2036 | |
2037 | if (is_sched_load_balance(cs)) | |
2038 | update_flag(CS_SCHED_LOAD_BALANCE, cs, 0); | |
2039 | ||
2040 | number_of_cpusets--; | |
efeb77b2 | 2041 | clear_bit(CS_ONLINE, &cs->flags); |
c8f699bb | 2042 | |
5d21cc2d | 2043 | mutex_unlock(&cpuset_mutex); |
1da177e4 LT |
2044 | } |
2045 | ||
eb95419b | 2046 | static void cpuset_css_free(struct cgroup_subsys_state *css) |
1da177e4 | 2047 | { |
eb95419b | 2048 | struct cpuset *cs = css_cs(css); |
1da177e4 | 2049 | |
300ed6cb | 2050 | free_cpumask_var(cs->cpus_allowed); |
8793d854 | 2051 | kfree(cs); |
1da177e4 LT |
2052 | } |
2053 | ||
8793d854 PM |
2054 | struct cgroup_subsys cpuset_subsys = { |
2055 | .name = "cpuset", | |
92fb9748 | 2056 | .css_alloc = cpuset_css_alloc, |
c8f699bb TH |
2057 | .css_online = cpuset_css_online, |
2058 | .css_offline = cpuset_css_offline, | |
92fb9748 | 2059 | .css_free = cpuset_css_free, |
8793d854 | 2060 | .can_attach = cpuset_can_attach, |
452477fa | 2061 | .cancel_attach = cpuset_cancel_attach, |
8793d854 | 2062 | .attach = cpuset_attach, |
8793d854 | 2063 | .subsys_id = cpuset_subsys_id, |
4baf6e33 | 2064 | .base_cftypes = files, |
8793d854 PM |
2065 | .early_init = 1, |
2066 | }; | |
2067 | ||
1da177e4 LT |
2068 | /** |
2069 | * cpuset_init - initialize cpusets at system boot | |
2070 | * | |
2071 | * Description: Initialize top_cpuset and the cpuset internal file system, | |
2072 | **/ | |
2073 | ||
2074 | int __init cpuset_init(void) | |
2075 | { | |
8793d854 | 2076 | int err = 0; |
1da177e4 | 2077 | |
58568d2a MX |
2078 | if (!alloc_cpumask_var(&top_cpuset.cpus_allowed, GFP_KERNEL)) |
2079 | BUG(); | |
2080 | ||
300ed6cb | 2081 | cpumask_setall(top_cpuset.cpus_allowed); |
f9a86fcb | 2082 | nodes_setall(top_cpuset.mems_allowed); |
1da177e4 | 2083 | |
3e0d98b9 | 2084 | fmeter_init(&top_cpuset.fmeter); |
029190c5 | 2085 | set_bit(CS_SCHED_LOAD_BALANCE, &top_cpuset.flags); |
1d3504fc | 2086 | top_cpuset.relax_domain_level = -1; |
1da177e4 | 2087 | |
1da177e4 LT |
2088 | err = register_filesystem(&cpuset_fs_type); |
2089 | if (err < 0) | |
8793d854 PM |
2090 | return err; |
2091 | ||
2341d1b6 LZ |
2092 | if (!alloc_cpumask_var(&cpus_attach, GFP_KERNEL)) |
2093 | BUG(); | |
2094 | ||
202f72d5 | 2095 | number_of_cpusets = 1; |
8793d854 | 2096 | return 0; |
1da177e4 LT |
2097 | } |
2098 | ||
b1aac8bb | 2099 | /* |
cf417141 | 2100 | * If CPU and/or memory hotplug handlers, below, unplug any CPUs |
b1aac8bb PJ |
2101 | * or memory nodes, we need to walk over the cpuset hierarchy, |
2102 | * removing that CPU or node from all cpusets. If this removes the | |
956db3ca CW |
2103 | * last CPU or node from a cpuset, then move the tasks in the empty |
2104 | * cpuset to its next-highest non-empty parent. | |
b1aac8bb | 2105 | */ |
956db3ca CW |
2106 | static void remove_tasks_in_empty_cpuset(struct cpuset *cs) |
2107 | { | |
2108 | struct cpuset *parent; | |
2109 | ||
956db3ca CW |
2110 | /* |
2111 | * Find its next-highest non-empty parent, (top cpuset | |
2112 | * has online cpus, so can't be empty). | |
2113 | */ | |
c431069f | 2114 | parent = parent_cs(cs); |
300ed6cb | 2115 | while (cpumask_empty(parent->cpus_allowed) || |
b4501295 | 2116 | nodes_empty(parent->mems_allowed)) |
c431069f | 2117 | parent = parent_cs(parent); |
956db3ca | 2118 | |
8cc99345 TH |
2119 | if (cgroup_transfer_tasks(parent->css.cgroup, cs->css.cgroup)) { |
2120 | rcu_read_lock(); | |
2121 | printk(KERN_ERR "cpuset: failed to transfer tasks out of empty cpuset %s\n", | |
2122 | cgroup_name(cs->css.cgroup)); | |
2123 | rcu_read_unlock(); | |
2124 | } | |
956db3ca CW |
2125 | } |
2126 | ||
deb7aa30 | 2127 | /** |
388afd85 | 2128 | * cpuset_hotplug_update_tasks - update tasks in a cpuset for hotunplug |
deb7aa30 | 2129 | * @cs: cpuset in interest |
956db3ca | 2130 | * |
deb7aa30 TH |
2131 | * Compare @cs's cpu and mem masks against top_cpuset and if some have gone |
2132 | * offline, update @cs accordingly. If @cs ends up with no CPU or memory, | |
2133 | * all its tasks are moved to the nearest ancestor with both resources. | |
80d1fa64 | 2134 | */ |
388afd85 | 2135 | static void cpuset_hotplug_update_tasks(struct cpuset *cs) |
80d1fa64 | 2136 | { |
deb7aa30 | 2137 | static cpumask_t off_cpus; |
33ad801d | 2138 | static nodemask_t off_mems; |
5d21cc2d | 2139 | bool is_empty; |
5c5cc623 | 2140 | bool sane = cgroup_sane_behavior(cs->css.cgroup); |
80d1fa64 | 2141 | |
e44193d3 LZ |
2142 | retry: |
2143 | wait_event(cpuset_attach_wq, cs->attach_in_progress == 0); | |
80d1fa64 | 2144 | |
5d21cc2d | 2145 | mutex_lock(&cpuset_mutex); |
7ddf96b0 | 2146 | |
e44193d3 LZ |
2147 | /* |
2148 | * We have raced with task attaching. We wait until attaching | |
2149 | * is finished, so we won't attach a task to an empty cpuset. | |
2150 | */ | |
2151 | if (cs->attach_in_progress) { | |
2152 | mutex_unlock(&cpuset_mutex); | |
2153 | goto retry; | |
2154 | } | |
2155 | ||
deb7aa30 TH |
2156 | cpumask_andnot(&off_cpus, cs->cpus_allowed, top_cpuset.cpus_allowed); |
2157 | nodes_andnot(off_mems, cs->mems_allowed, top_cpuset.mems_allowed); | |
80d1fa64 | 2158 | |
5c5cc623 LZ |
2159 | mutex_lock(&callback_mutex); |
2160 | cpumask_andnot(cs->cpus_allowed, cs->cpus_allowed, &off_cpus); | |
2161 | mutex_unlock(&callback_mutex); | |
2162 | ||
2163 | /* | |
2164 | * If sane_behavior flag is set, we need to update tasks' cpumask | |
f047cecf LZ |
2165 | * for empty cpuset to take on ancestor's cpumask. Otherwise, don't |
2166 | * call update_tasks_cpumask() if the cpuset becomes empty, as | |
2167 | * the tasks in it will be migrated to an ancestor. | |
5c5cc623 LZ |
2168 | */ |
2169 | if ((sane && cpumask_empty(cs->cpus_allowed)) || | |
f047cecf | 2170 | (!cpumask_empty(&off_cpus) && !cpumask_empty(cs->cpus_allowed))) |
deb7aa30 | 2171 | update_tasks_cpumask(cs, NULL); |
80d1fa64 | 2172 | |
5c5cc623 LZ |
2173 | mutex_lock(&callback_mutex); |
2174 | nodes_andnot(cs->mems_allowed, cs->mems_allowed, off_mems); | |
2175 | mutex_unlock(&callback_mutex); | |
2176 | ||
2177 | /* | |
2178 | * If sane_behavior flag is set, we need to update tasks' nodemask | |
f047cecf LZ |
2179 | * for empty cpuset to take on ancestor's nodemask. Otherwise, don't |
2180 | * call update_tasks_nodemask() if the cpuset becomes empty, as | |
2181 | * the tasks in it will be migratd to an ancestor. | |
5c5cc623 LZ |
2182 | */ |
2183 | if ((sane && nodes_empty(cs->mems_allowed)) || | |
f047cecf | 2184 | (!nodes_empty(off_mems) && !nodes_empty(cs->mems_allowed))) |
33ad801d | 2185 | update_tasks_nodemask(cs, NULL); |
deb7aa30 | 2186 | |
5d21cc2d TH |
2187 | is_empty = cpumask_empty(cs->cpus_allowed) || |
2188 | nodes_empty(cs->mems_allowed); | |
8d033948 | 2189 | |
5d21cc2d TH |
2190 | mutex_unlock(&cpuset_mutex); |
2191 | ||
2192 | /* | |
5c5cc623 LZ |
2193 | * If sane_behavior flag is set, we'll keep tasks in empty cpusets. |
2194 | * | |
2195 | * Otherwise move tasks to the nearest ancestor with execution | |
2196 | * resources. This is full cgroup operation which will | |
5d21cc2d TH |
2197 | * also call back into cpuset. Should be done outside any lock. |
2198 | */ | |
5c5cc623 | 2199 | if (!sane && is_empty) |
5d21cc2d | 2200 | remove_tasks_in_empty_cpuset(cs); |
b1aac8bb PJ |
2201 | } |
2202 | ||
deb7aa30 | 2203 | /** |
3a5a6d0c | 2204 | * cpuset_hotplug_workfn - handle CPU/memory hotunplug for a cpuset |
956db3ca | 2205 | * |
deb7aa30 TH |
2206 | * This function is called after either CPU or memory configuration has |
2207 | * changed and updates cpuset accordingly. The top_cpuset is always | |
2208 | * synchronized to cpu_active_mask and N_MEMORY, which is necessary in | |
2209 | * order to make cpusets transparent (of no affect) on systems that are | |
2210 | * actively using CPU hotplug but making no active use of cpusets. | |
956db3ca | 2211 | * |
deb7aa30 | 2212 | * Non-root cpusets are only affected by offlining. If any CPUs or memory |
388afd85 LZ |
2213 | * nodes have been taken down, cpuset_hotplug_update_tasks() is invoked on |
2214 | * all descendants. | |
956db3ca | 2215 | * |
deb7aa30 TH |
2216 | * Note that CPU offlining during suspend is ignored. We don't modify |
2217 | * cpusets across suspend/resume cycles at all. | |
956db3ca | 2218 | */ |
3a5a6d0c | 2219 | static void cpuset_hotplug_workfn(struct work_struct *work) |
b1aac8bb | 2220 | { |
5c5cc623 LZ |
2221 | static cpumask_t new_cpus; |
2222 | static nodemask_t new_mems; | |
deb7aa30 | 2223 | bool cpus_updated, mems_updated; |
b1aac8bb | 2224 | |
5d21cc2d | 2225 | mutex_lock(&cpuset_mutex); |
956db3ca | 2226 | |
deb7aa30 TH |
2227 | /* fetch the available cpus/mems and find out which changed how */ |
2228 | cpumask_copy(&new_cpus, cpu_active_mask); | |
2229 | new_mems = node_states[N_MEMORY]; | |
7ddf96b0 | 2230 | |
deb7aa30 | 2231 | cpus_updated = !cpumask_equal(top_cpuset.cpus_allowed, &new_cpus); |
deb7aa30 | 2232 | mems_updated = !nodes_equal(top_cpuset.mems_allowed, new_mems); |
7ddf96b0 | 2233 | |
deb7aa30 TH |
2234 | /* synchronize cpus_allowed to cpu_active_mask */ |
2235 | if (cpus_updated) { | |
2236 | mutex_lock(&callback_mutex); | |
2237 | cpumask_copy(top_cpuset.cpus_allowed, &new_cpus); | |
2238 | mutex_unlock(&callback_mutex); | |
2239 | /* we don't mess with cpumasks of tasks in top_cpuset */ | |
2240 | } | |
b4501295 | 2241 | |
deb7aa30 TH |
2242 | /* synchronize mems_allowed to N_MEMORY */ |
2243 | if (mems_updated) { | |
deb7aa30 TH |
2244 | mutex_lock(&callback_mutex); |
2245 | top_cpuset.mems_allowed = new_mems; | |
2246 | mutex_unlock(&callback_mutex); | |
33ad801d | 2247 | update_tasks_nodemask(&top_cpuset, NULL); |
deb7aa30 | 2248 | } |
b4501295 | 2249 | |
388afd85 LZ |
2250 | mutex_unlock(&cpuset_mutex); |
2251 | ||
5c5cc623 LZ |
2252 | /* if cpus or mems changed, we need to propagate to descendants */ |
2253 | if (cpus_updated || mems_updated) { | |
deb7aa30 | 2254 | struct cpuset *cs; |
492eb21b | 2255 | struct cgroup_subsys_state *pos_css; |
f9b4fb8d | 2256 | |
fc560a26 | 2257 | rcu_read_lock(); |
492eb21b | 2258 | cpuset_for_each_descendant_pre(cs, pos_css, &top_cpuset) { |
388afd85 LZ |
2259 | if (!css_tryget(&cs->css)) |
2260 | continue; | |
2261 | rcu_read_unlock(); | |
7ddf96b0 | 2262 | |
388afd85 | 2263 | cpuset_hotplug_update_tasks(cs); |
b4501295 | 2264 | |
388afd85 LZ |
2265 | rcu_read_lock(); |
2266 | css_put(&cs->css); | |
2267 | } | |
2268 | rcu_read_unlock(); | |
2269 | } | |
8d033948 | 2270 | |
deb7aa30 | 2271 | /* rebuild sched domains if cpus_allowed has changed */ |
e0e80a02 LZ |
2272 | if (cpus_updated) |
2273 | rebuild_sched_domains(); | |
b1aac8bb PJ |
2274 | } |
2275 | ||
7ddf96b0 | 2276 | void cpuset_update_active_cpus(bool cpu_online) |
4c4d50f7 | 2277 | { |
3a5a6d0c TH |
2278 | /* |
2279 | * We're inside cpu hotplug critical region which usually nests | |
2280 | * inside cgroup synchronization. Bounce actual hotplug processing | |
2281 | * to a work item to avoid reverse locking order. | |
2282 | * | |
2283 | * We still need to do partition_sched_domains() synchronously; | |
2284 | * otherwise, the scheduler will get confused and put tasks to the | |
2285 | * dead CPU. Fall back to the default single domain. | |
2286 | * cpuset_hotplug_workfn() will rebuild it as necessary. | |
2287 | */ | |
2288 | partition_sched_domains(1, NULL, NULL); | |
2289 | schedule_work(&cpuset_hotplug_work); | |
4c4d50f7 | 2290 | } |
4c4d50f7 | 2291 | |
38837fc7 | 2292 | /* |
38d7bee9 LJ |
2293 | * Keep top_cpuset.mems_allowed tracking node_states[N_MEMORY]. |
2294 | * Call this routine anytime after node_states[N_MEMORY] changes. | |
a1cd2b13 | 2295 | * See cpuset_update_active_cpus() for CPU hotplug handling. |
38837fc7 | 2296 | */ |
f481891f MX |
2297 | static int cpuset_track_online_nodes(struct notifier_block *self, |
2298 | unsigned long action, void *arg) | |
38837fc7 | 2299 | { |
3a5a6d0c | 2300 | schedule_work(&cpuset_hotplug_work); |
f481891f | 2301 | return NOTIFY_OK; |
38837fc7 | 2302 | } |
d8f10cb3 AM |
2303 | |
2304 | static struct notifier_block cpuset_track_online_nodes_nb = { | |
2305 | .notifier_call = cpuset_track_online_nodes, | |
2306 | .priority = 10, /* ??! */ | |
2307 | }; | |
38837fc7 | 2308 | |
1da177e4 LT |
2309 | /** |
2310 | * cpuset_init_smp - initialize cpus_allowed | |
2311 | * | |
2312 | * Description: Finish top cpuset after cpu, node maps are initialized | |
d8f10cb3 | 2313 | */ |
1da177e4 LT |
2314 | void __init cpuset_init_smp(void) |
2315 | { | |
6ad4c188 | 2316 | cpumask_copy(top_cpuset.cpus_allowed, cpu_active_mask); |
38d7bee9 | 2317 | top_cpuset.mems_allowed = node_states[N_MEMORY]; |
33ad801d | 2318 | top_cpuset.old_mems_allowed = top_cpuset.mems_allowed; |
4c4d50f7 | 2319 | |
d8f10cb3 | 2320 | register_hotmemory_notifier(&cpuset_track_online_nodes_nb); |
1da177e4 LT |
2321 | } |
2322 | ||
2323 | /** | |
1da177e4 LT |
2324 | * cpuset_cpus_allowed - return cpus_allowed mask from a tasks cpuset. |
2325 | * @tsk: pointer to task_struct from which to obtain cpuset->cpus_allowed. | |
6af866af | 2326 | * @pmask: pointer to struct cpumask variable to receive cpus_allowed set. |
1da177e4 | 2327 | * |
300ed6cb | 2328 | * Description: Returns the cpumask_var_t cpus_allowed of the cpuset |
1da177e4 | 2329 | * attached to the specified @tsk. Guaranteed to return some non-empty |
5f054e31 | 2330 | * subset of cpu_online_mask, even if this means going outside the |
1da177e4 LT |
2331 | * tasks cpuset. |
2332 | **/ | |
2333 | ||
6af866af | 2334 | void cpuset_cpus_allowed(struct task_struct *tsk, struct cpumask *pmask) |
1da177e4 | 2335 | { |
070b57fc LZ |
2336 | struct cpuset *cpus_cs; |
2337 | ||
3d3f26a7 | 2338 | mutex_lock(&callback_mutex); |
909d75a3 | 2339 | task_lock(tsk); |
070b57fc LZ |
2340 | cpus_cs = effective_cpumask_cpuset(task_cs(tsk)); |
2341 | guarantee_online_cpus(cpus_cs, pmask); | |
909d75a3 | 2342 | task_unlock(tsk); |
897f0b3c | 2343 | mutex_unlock(&callback_mutex); |
1da177e4 LT |
2344 | } |
2345 | ||
2baab4e9 | 2346 | void cpuset_cpus_allowed_fallback(struct task_struct *tsk) |
9084bb82 | 2347 | { |
c9710d80 | 2348 | struct cpuset *cpus_cs; |
9084bb82 ON |
2349 | |
2350 | rcu_read_lock(); | |
070b57fc LZ |
2351 | cpus_cs = effective_cpumask_cpuset(task_cs(tsk)); |
2352 | do_set_cpus_allowed(tsk, cpus_cs->cpus_allowed); | |
9084bb82 ON |
2353 | rcu_read_unlock(); |
2354 | ||
2355 | /* | |
2356 | * We own tsk->cpus_allowed, nobody can change it under us. | |
2357 | * | |
2358 | * But we used cs && cs->cpus_allowed lockless and thus can | |
2359 | * race with cgroup_attach_task() or update_cpumask() and get | |
2360 | * the wrong tsk->cpus_allowed. However, both cases imply the | |
2361 | * subsequent cpuset_change_cpumask()->set_cpus_allowed_ptr() | |
2362 | * which takes task_rq_lock(). | |
2363 | * | |
2364 | * If we are called after it dropped the lock we must see all | |
2365 | * changes in tsk_cs()->cpus_allowed. Otherwise we can temporary | |
2366 | * set any mask even if it is not right from task_cs() pov, | |
2367 | * the pending set_cpus_allowed_ptr() will fix things. | |
2baab4e9 PZ |
2368 | * |
2369 | * select_fallback_rq() will fix things ups and set cpu_possible_mask | |
2370 | * if required. | |
9084bb82 | 2371 | */ |
9084bb82 ON |
2372 | } |
2373 | ||
1da177e4 LT |
2374 | void cpuset_init_current_mems_allowed(void) |
2375 | { | |
f9a86fcb | 2376 | nodes_setall(current->mems_allowed); |
1da177e4 LT |
2377 | } |
2378 | ||
909d75a3 PJ |
2379 | /** |
2380 | * cpuset_mems_allowed - return mems_allowed mask from a tasks cpuset. | |
2381 | * @tsk: pointer to task_struct from which to obtain cpuset->mems_allowed. | |
2382 | * | |
2383 | * Description: Returns the nodemask_t mems_allowed of the cpuset | |
2384 | * attached to the specified @tsk. Guaranteed to return some non-empty | |
38d7bee9 | 2385 | * subset of node_states[N_MEMORY], even if this means going outside the |
909d75a3 PJ |
2386 | * tasks cpuset. |
2387 | **/ | |
2388 | ||
2389 | nodemask_t cpuset_mems_allowed(struct task_struct *tsk) | |
2390 | { | |
070b57fc | 2391 | struct cpuset *mems_cs; |
909d75a3 PJ |
2392 | nodemask_t mask; |
2393 | ||
3d3f26a7 | 2394 | mutex_lock(&callback_mutex); |
909d75a3 | 2395 | task_lock(tsk); |
070b57fc LZ |
2396 | mems_cs = effective_nodemask_cpuset(task_cs(tsk)); |
2397 | guarantee_online_mems(mems_cs, &mask); | |
909d75a3 | 2398 | task_unlock(tsk); |
3d3f26a7 | 2399 | mutex_unlock(&callback_mutex); |
909d75a3 PJ |
2400 | |
2401 | return mask; | |
2402 | } | |
2403 | ||
d9fd8a6d | 2404 | /** |
19770b32 MG |
2405 | * cpuset_nodemask_valid_mems_allowed - check nodemask vs. curremt mems_allowed |
2406 | * @nodemask: the nodemask to be checked | |
d9fd8a6d | 2407 | * |
19770b32 | 2408 | * Are any of the nodes in the nodemask allowed in current->mems_allowed? |
1da177e4 | 2409 | */ |
19770b32 | 2410 | int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask) |
1da177e4 | 2411 | { |
19770b32 | 2412 | return nodes_intersects(*nodemask, current->mems_allowed); |
1da177e4 LT |
2413 | } |
2414 | ||
9bf2229f | 2415 | /* |
78608366 PM |
2416 | * nearest_hardwall_ancestor() - Returns the nearest mem_exclusive or |
2417 | * mem_hardwall ancestor to the specified cpuset. Call holding | |
2418 | * callback_mutex. If no ancestor is mem_exclusive or mem_hardwall | |
2419 | * (an unusual configuration), then returns the root cpuset. | |
9bf2229f | 2420 | */ |
c9710d80 | 2421 | static struct cpuset *nearest_hardwall_ancestor(struct cpuset *cs) |
9bf2229f | 2422 | { |
c431069f TH |
2423 | while (!(is_mem_exclusive(cs) || is_mem_hardwall(cs)) && parent_cs(cs)) |
2424 | cs = parent_cs(cs); | |
9bf2229f PJ |
2425 | return cs; |
2426 | } | |
2427 | ||
d9fd8a6d | 2428 | /** |
a1bc5a4e DR |
2429 | * cpuset_node_allowed_softwall - Can we allocate on a memory node? |
2430 | * @node: is this an allowed node? | |
02a0e53d | 2431 | * @gfp_mask: memory allocation flags |
d9fd8a6d | 2432 | * |
a1bc5a4e DR |
2433 | * If we're in interrupt, yes, we can always allocate. If __GFP_THISNODE is |
2434 | * set, yes, we can always allocate. If node is in our task's mems_allowed, | |
2435 | * yes. If it's not a __GFP_HARDWALL request and this node is in the nearest | |
2436 | * hardwalled cpuset ancestor to this task's cpuset, yes. If the task has been | |
2437 | * OOM killed and has access to memory reserves as specified by the TIF_MEMDIE | |
2438 | * flag, yes. | |
9bf2229f PJ |
2439 | * Otherwise, no. |
2440 | * | |
a1bc5a4e DR |
2441 | * If __GFP_HARDWALL is set, cpuset_node_allowed_softwall() reduces to |
2442 | * cpuset_node_allowed_hardwall(). Otherwise, cpuset_node_allowed_softwall() | |
2443 | * might sleep, and might allow a node from an enclosing cpuset. | |
02a0e53d | 2444 | * |
a1bc5a4e DR |
2445 | * cpuset_node_allowed_hardwall() only handles the simpler case of hardwall |
2446 | * cpusets, and never sleeps. | |
02a0e53d PJ |
2447 | * |
2448 | * The __GFP_THISNODE placement logic is really handled elsewhere, | |
2449 | * by forcibly using a zonelist starting at a specified node, and by | |
2450 | * (in get_page_from_freelist()) refusing to consider the zones for | |
2451 | * any node on the zonelist except the first. By the time any such | |
2452 | * calls get to this routine, we should just shut up and say 'yes'. | |
2453 | * | |
9bf2229f | 2454 | * GFP_USER allocations are marked with the __GFP_HARDWALL bit, |
c596d9f3 DR |
2455 | * and do not allow allocations outside the current tasks cpuset |
2456 | * unless the task has been OOM killed as is marked TIF_MEMDIE. | |
9bf2229f | 2457 | * GFP_KERNEL allocations are not so marked, so can escape to the |
78608366 | 2458 | * nearest enclosing hardwalled ancestor cpuset. |
9bf2229f | 2459 | * |
02a0e53d PJ |
2460 | * Scanning up parent cpusets requires callback_mutex. The |
2461 | * __alloc_pages() routine only calls here with __GFP_HARDWALL bit | |
2462 | * _not_ set if it's a GFP_KERNEL allocation, and all nodes in the | |
2463 | * current tasks mems_allowed came up empty on the first pass over | |
2464 | * the zonelist. So only GFP_KERNEL allocations, if all nodes in the | |
2465 | * cpuset are short of memory, might require taking the callback_mutex | |
2466 | * mutex. | |
9bf2229f | 2467 | * |
36be57ff | 2468 | * The first call here from mm/page_alloc:get_page_from_freelist() |
02a0e53d PJ |
2469 | * has __GFP_HARDWALL set in gfp_mask, enforcing hardwall cpusets, |
2470 | * so no allocation on a node outside the cpuset is allowed (unless | |
2471 | * in interrupt, of course). | |
36be57ff PJ |
2472 | * |
2473 | * The second pass through get_page_from_freelist() doesn't even call | |
2474 | * here for GFP_ATOMIC calls. For those calls, the __alloc_pages() | |
2475 | * variable 'wait' is not set, and the bit ALLOC_CPUSET is not set | |
2476 | * in alloc_flags. That logic and the checks below have the combined | |
2477 | * affect that: | |
9bf2229f PJ |
2478 | * in_interrupt - any node ok (current task context irrelevant) |
2479 | * GFP_ATOMIC - any node ok | |
c596d9f3 | 2480 | * TIF_MEMDIE - any node ok |
78608366 | 2481 | * GFP_KERNEL - any node in enclosing hardwalled cpuset ok |
9bf2229f | 2482 | * GFP_USER - only nodes in current tasks mems allowed ok. |
36be57ff PJ |
2483 | * |
2484 | * Rule: | |
a1bc5a4e | 2485 | * Don't call cpuset_node_allowed_softwall if you can't sleep, unless you |
36be57ff PJ |
2486 | * pass in the __GFP_HARDWALL flag set in gfp_flag, which disables |
2487 | * the code that might scan up ancestor cpusets and sleep. | |
02a0e53d | 2488 | */ |
a1bc5a4e | 2489 | int __cpuset_node_allowed_softwall(int node, gfp_t gfp_mask) |
1da177e4 | 2490 | { |
c9710d80 | 2491 | struct cpuset *cs; /* current cpuset ancestors */ |
29afd49b | 2492 | int allowed; /* is allocation in zone z allowed? */ |
9bf2229f | 2493 | |
9b819d20 | 2494 | if (in_interrupt() || (gfp_mask & __GFP_THISNODE)) |
9bf2229f | 2495 | return 1; |
92d1dbd2 | 2496 | might_sleep_if(!(gfp_mask & __GFP_HARDWALL)); |
9bf2229f PJ |
2497 | if (node_isset(node, current->mems_allowed)) |
2498 | return 1; | |
c596d9f3 DR |
2499 | /* |
2500 | * Allow tasks that have access to memory reserves because they have | |
2501 | * been OOM killed to get memory anywhere. | |
2502 | */ | |
2503 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
2504 | return 1; | |
9bf2229f PJ |
2505 | if (gfp_mask & __GFP_HARDWALL) /* If hardwall request, stop here */ |
2506 | return 0; | |
2507 | ||
5563e770 BP |
2508 | if (current->flags & PF_EXITING) /* Let dying task have memory */ |
2509 | return 1; | |
2510 | ||
9bf2229f | 2511 | /* Not hardwall and node outside mems_allowed: scan up cpusets */ |
3d3f26a7 | 2512 | mutex_lock(&callback_mutex); |
053199ed | 2513 | |
053199ed | 2514 | task_lock(current); |
78608366 | 2515 | cs = nearest_hardwall_ancestor(task_cs(current)); |
053199ed PJ |
2516 | task_unlock(current); |
2517 | ||
9bf2229f | 2518 | allowed = node_isset(node, cs->mems_allowed); |
3d3f26a7 | 2519 | mutex_unlock(&callback_mutex); |
9bf2229f | 2520 | return allowed; |
1da177e4 LT |
2521 | } |
2522 | ||
02a0e53d | 2523 | /* |
a1bc5a4e DR |
2524 | * cpuset_node_allowed_hardwall - Can we allocate on a memory node? |
2525 | * @node: is this an allowed node? | |
02a0e53d PJ |
2526 | * @gfp_mask: memory allocation flags |
2527 | * | |
a1bc5a4e DR |
2528 | * If we're in interrupt, yes, we can always allocate. If __GFP_THISNODE is |
2529 | * set, yes, we can always allocate. If node is in our task's mems_allowed, | |
2530 | * yes. If the task has been OOM killed and has access to memory reserves as | |
2531 | * specified by the TIF_MEMDIE flag, yes. | |
2532 | * Otherwise, no. | |
02a0e53d PJ |
2533 | * |
2534 | * The __GFP_THISNODE placement logic is really handled elsewhere, | |
2535 | * by forcibly using a zonelist starting at a specified node, and by | |
2536 | * (in get_page_from_freelist()) refusing to consider the zones for | |
2537 | * any node on the zonelist except the first. By the time any such | |
2538 | * calls get to this routine, we should just shut up and say 'yes'. | |
2539 | * | |
a1bc5a4e DR |
2540 | * Unlike the cpuset_node_allowed_softwall() variant, above, |
2541 | * this variant requires that the node be in the current task's | |
02a0e53d PJ |
2542 | * mems_allowed or that we're in interrupt. It does not scan up the |
2543 | * cpuset hierarchy for the nearest enclosing mem_exclusive cpuset. | |
2544 | * It never sleeps. | |
2545 | */ | |
a1bc5a4e | 2546 | int __cpuset_node_allowed_hardwall(int node, gfp_t gfp_mask) |
02a0e53d | 2547 | { |
02a0e53d PJ |
2548 | if (in_interrupt() || (gfp_mask & __GFP_THISNODE)) |
2549 | return 1; | |
02a0e53d PJ |
2550 | if (node_isset(node, current->mems_allowed)) |
2551 | return 1; | |
dedf8b79 DW |
2552 | /* |
2553 | * Allow tasks that have access to memory reserves because they have | |
2554 | * been OOM killed to get memory anywhere. | |
2555 | */ | |
2556 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
2557 | return 1; | |
02a0e53d PJ |
2558 | return 0; |
2559 | } | |
2560 | ||
825a46af | 2561 | /** |
6adef3eb JS |
2562 | * cpuset_mem_spread_node() - On which node to begin search for a file page |
2563 | * cpuset_slab_spread_node() - On which node to begin search for a slab page | |
825a46af PJ |
2564 | * |
2565 | * If a task is marked PF_SPREAD_PAGE or PF_SPREAD_SLAB (as for | |
2566 | * tasks in a cpuset with is_spread_page or is_spread_slab set), | |
2567 | * and if the memory allocation used cpuset_mem_spread_node() | |
2568 | * to determine on which node to start looking, as it will for | |
2569 | * certain page cache or slab cache pages such as used for file | |
2570 | * system buffers and inode caches, then instead of starting on the | |
2571 | * local node to look for a free page, rather spread the starting | |
2572 | * node around the tasks mems_allowed nodes. | |
2573 | * | |
2574 | * We don't have to worry about the returned node being offline | |
2575 | * because "it can't happen", and even if it did, it would be ok. | |
2576 | * | |
2577 | * The routines calling guarantee_online_mems() are careful to | |
2578 | * only set nodes in task->mems_allowed that are online. So it | |
2579 | * should not be possible for the following code to return an | |
2580 | * offline node. But if it did, that would be ok, as this routine | |
2581 | * is not returning the node where the allocation must be, only | |
2582 | * the node where the search should start. The zonelist passed to | |
2583 | * __alloc_pages() will include all nodes. If the slab allocator | |
2584 | * is passed an offline node, it will fall back to the local node. | |
2585 | * See kmem_cache_alloc_node(). | |
2586 | */ | |
2587 | ||
6adef3eb | 2588 | static int cpuset_spread_node(int *rotor) |
825a46af PJ |
2589 | { |
2590 | int node; | |
2591 | ||
6adef3eb | 2592 | node = next_node(*rotor, current->mems_allowed); |
825a46af PJ |
2593 | if (node == MAX_NUMNODES) |
2594 | node = first_node(current->mems_allowed); | |
6adef3eb | 2595 | *rotor = node; |
825a46af PJ |
2596 | return node; |
2597 | } | |
6adef3eb JS |
2598 | |
2599 | int cpuset_mem_spread_node(void) | |
2600 | { | |
778d3b0f MH |
2601 | if (current->cpuset_mem_spread_rotor == NUMA_NO_NODE) |
2602 | current->cpuset_mem_spread_rotor = | |
2603 | node_random(¤t->mems_allowed); | |
2604 | ||
6adef3eb JS |
2605 | return cpuset_spread_node(¤t->cpuset_mem_spread_rotor); |
2606 | } | |
2607 | ||
2608 | int cpuset_slab_spread_node(void) | |
2609 | { | |
778d3b0f MH |
2610 | if (current->cpuset_slab_spread_rotor == NUMA_NO_NODE) |
2611 | current->cpuset_slab_spread_rotor = | |
2612 | node_random(¤t->mems_allowed); | |
2613 | ||
6adef3eb JS |
2614 | return cpuset_spread_node(¤t->cpuset_slab_spread_rotor); |
2615 | } | |
2616 | ||
825a46af PJ |
2617 | EXPORT_SYMBOL_GPL(cpuset_mem_spread_node); |
2618 | ||
ef08e3b4 | 2619 | /** |
bbe373f2 DR |
2620 | * cpuset_mems_allowed_intersects - Does @tsk1's mems_allowed intersect @tsk2's? |
2621 | * @tsk1: pointer to task_struct of some task. | |
2622 | * @tsk2: pointer to task_struct of some other task. | |
2623 | * | |
2624 | * Description: Return true if @tsk1's mems_allowed intersects the | |
2625 | * mems_allowed of @tsk2. Used by the OOM killer to determine if | |
2626 | * one of the task's memory usage might impact the memory available | |
2627 | * to the other. | |
ef08e3b4 PJ |
2628 | **/ |
2629 | ||
bbe373f2 DR |
2630 | int cpuset_mems_allowed_intersects(const struct task_struct *tsk1, |
2631 | const struct task_struct *tsk2) | |
ef08e3b4 | 2632 | { |
bbe373f2 | 2633 | return nodes_intersects(tsk1->mems_allowed, tsk2->mems_allowed); |
ef08e3b4 PJ |
2634 | } |
2635 | ||
f440d98f LZ |
2636 | #define CPUSET_NODELIST_LEN (256) |
2637 | ||
75aa1994 DR |
2638 | /** |
2639 | * cpuset_print_task_mems_allowed - prints task's cpuset and mems_allowed | |
2640 | * @task: pointer to task_struct of some task. | |
2641 | * | |
2642 | * Description: Prints @task's name, cpuset name, and cached copy of its | |
2643 | * mems_allowed to the kernel log. Must hold task_lock(task) to allow | |
2644 | * dereferencing task_cs(task). | |
2645 | */ | |
2646 | void cpuset_print_task_mems_allowed(struct task_struct *tsk) | |
2647 | { | |
f440d98f LZ |
2648 | /* Statically allocated to prevent using excess stack. */ |
2649 | static char cpuset_nodelist[CPUSET_NODELIST_LEN]; | |
2650 | static DEFINE_SPINLOCK(cpuset_buffer_lock); | |
75aa1994 | 2651 | |
f440d98f | 2652 | struct cgroup *cgrp = task_cs(tsk)->css.cgroup; |
63f43f55 | 2653 | |
cfb5966b | 2654 | rcu_read_lock(); |
f440d98f | 2655 | spin_lock(&cpuset_buffer_lock); |
63f43f55 | 2656 | |
75aa1994 DR |
2657 | nodelist_scnprintf(cpuset_nodelist, CPUSET_NODELIST_LEN, |
2658 | tsk->mems_allowed); | |
2659 | printk(KERN_INFO "%s cpuset=%s mems_allowed=%s\n", | |
f440d98f LZ |
2660 | tsk->comm, cgroup_name(cgrp), cpuset_nodelist); |
2661 | ||
75aa1994 | 2662 | spin_unlock(&cpuset_buffer_lock); |
cfb5966b | 2663 | rcu_read_unlock(); |
75aa1994 DR |
2664 | } |
2665 | ||
3e0d98b9 PJ |
2666 | /* |
2667 | * Collection of memory_pressure is suppressed unless | |
2668 | * this flag is enabled by writing "1" to the special | |
2669 | * cpuset file 'memory_pressure_enabled' in the root cpuset. | |
2670 | */ | |
2671 | ||
c5b2aff8 | 2672 | int cpuset_memory_pressure_enabled __read_mostly; |
3e0d98b9 PJ |
2673 | |
2674 | /** | |
2675 | * cpuset_memory_pressure_bump - keep stats of per-cpuset reclaims. | |
2676 | * | |
2677 | * Keep a running average of the rate of synchronous (direct) | |
2678 | * page reclaim efforts initiated by tasks in each cpuset. | |
2679 | * | |
2680 | * This represents the rate at which some task in the cpuset | |
2681 | * ran low on memory on all nodes it was allowed to use, and | |
2682 | * had to enter the kernels page reclaim code in an effort to | |
2683 | * create more free memory by tossing clean pages or swapping | |
2684 | * or writing dirty pages. | |
2685 | * | |
2686 | * Display to user space in the per-cpuset read-only file | |
2687 | * "memory_pressure". Value displayed is an integer | |
2688 | * representing the recent rate of entry into the synchronous | |
2689 | * (direct) page reclaim by any task attached to the cpuset. | |
2690 | **/ | |
2691 | ||
2692 | void __cpuset_memory_pressure_bump(void) | |
2693 | { | |
3e0d98b9 | 2694 | task_lock(current); |
8793d854 | 2695 | fmeter_markevent(&task_cs(current)->fmeter); |
3e0d98b9 PJ |
2696 | task_unlock(current); |
2697 | } | |
2698 | ||
8793d854 | 2699 | #ifdef CONFIG_PROC_PID_CPUSET |
1da177e4 LT |
2700 | /* |
2701 | * proc_cpuset_show() | |
2702 | * - Print tasks cpuset path into seq_file. | |
2703 | * - Used for /proc/<pid>/cpuset. | |
053199ed PJ |
2704 | * - No need to task_lock(tsk) on this tsk->cpuset reference, as it |
2705 | * doesn't really matter if tsk->cpuset changes after we read it, | |
5d21cc2d | 2706 | * and we take cpuset_mutex, keeping cpuset_attach() from changing it |
2df167a3 | 2707 | * anyway. |
1da177e4 | 2708 | */ |
8d8b97ba | 2709 | int proc_cpuset_show(struct seq_file *m, void *unused_v) |
1da177e4 | 2710 | { |
13b41b09 | 2711 | struct pid *pid; |
1da177e4 LT |
2712 | struct task_struct *tsk; |
2713 | char *buf; | |
8793d854 | 2714 | struct cgroup_subsys_state *css; |
99f89551 | 2715 | int retval; |
1da177e4 | 2716 | |
99f89551 | 2717 | retval = -ENOMEM; |
1da177e4 LT |
2718 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
2719 | if (!buf) | |
99f89551 EB |
2720 | goto out; |
2721 | ||
2722 | retval = -ESRCH; | |
13b41b09 EB |
2723 | pid = m->private; |
2724 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
99f89551 EB |
2725 | if (!tsk) |
2726 | goto out_free; | |
1da177e4 | 2727 | |
27e89ae5 | 2728 | rcu_read_lock(); |
8af01f56 | 2729 | css = task_css(tsk, cpuset_subsys_id); |
8793d854 | 2730 | retval = cgroup_path(css->cgroup, buf, PAGE_SIZE); |
27e89ae5 | 2731 | rcu_read_unlock(); |
1da177e4 | 2732 | if (retval < 0) |
27e89ae5 | 2733 | goto out_put_task; |
1da177e4 LT |
2734 | seq_puts(m, buf); |
2735 | seq_putc(m, '\n'); | |
27e89ae5 | 2736 | out_put_task: |
99f89551 EB |
2737 | put_task_struct(tsk); |
2738 | out_free: | |
1da177e4 | 2739 | kfree(buf); |
99f89551 | 2740 | out: |
1da177e4 LT |
2741 | return retval; |
2742 | } | |
8793d854 | 2743 | #endif /* CONFIG_PROC_PID_CPUSET */ |
1da177e4 | 2744 | |
d01d4827 | 2745 | /* Display task mems_allowed in /proc/<pid>/status file. */ |
df5f8314 EB |
2746 | void cpuset_task_status_allowed(struct seq_file *m, struct task_struct *task) |
2747 | { | |
df5f8314 | 2748 | seq_printf(m, "Mems_allowed:\t"); |
30e8e136 | 2749 | seq_nodemask(m, &task->mems_allowed); |
df5f8314 | 2750 | seq_printf(m, "\n"); |
39106dcf | 2751 | seq_printf(m, "Mems_allowed_list:\t"); |
30e8e136 | 2752 | seq_nodemask_list(m, &task->mems_allowed); |
39106dcf | 2753 | seq_printf(m, "\n"); |
1da177e4 | 2754 | } |