]>
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 |
1da177e4 LT |
17 | * |
18 | * This file is subject to the terms and conditions of the GNU General Public | |
19 | * License. See the file COPYING in the main directory of the Linux | |
20 | * distribution for more details. | |
21 | */ | |
22 | ||
1da177e4 LT |
23 | #include <linux/cpu.h> |
24 | #include <linux/cpumask.h> | |
25 | #include <linux/cpuset.h> | |
26 | #include <linux/err.h> | |
27 | #include <linux/errno.h> | |
28 | #include <linux/file.h> | |
29 | #include <linux/fs.h> | |
30 | #include <linux/init.h> | |
31 | #include <linux/interrupt.h> | |
32 | #include <linux/kernel.h> | |
33 | #include <linux/kmod.h> | |
34 | #include <linux/list.h> | |
68860ec1 | 35 | #include <linux/mempolicy.h> |
1da177e4 LT |
36 | #include <linux/mm.h> |
37 | #include <linux/module.h> | |
38 | #include <linux/mount.h> | |
39 | #include <linux/namei.h> | |
40 | #include <linux/pagemap.h> | |
41 | #include <linux/proc_fs.h> | |
6b9c2603 | 42 | #include <linux/rcupdate.h> |
1da177e4 LT |
43 | #include <linux/sched.h> |
44 | #include <linux/seq_file.h> | |
22fb52dd | 45 | #include <linux/security.h> |
1da177e4 | 46 | #include <linux/slab.h> |
1da177e4 LT |
47 | #include <linux/spinlock.h> |
48 | #include <linux/stat.h> | |
49 | #include <linux/string.h> | |
50 | #include <linux/time.h> | |
51 | #include <linux/backing-dev.h> | |
52 | #include <linux/sort.h> | |
53 | ||
54 | #include <asm/uaccess.h> | |
55 | #include <asm/atomic.h> | |
3d3f26a7 | 56 | #include <linux/mutex.h> |
029190c5 | 57 | #include <linux/kfifo.h> |
956db3ca CW |
58 | #include <linux/workqueue.h> |
59 | #include <linux/cgroup.h> | |
1da177e4 | 60 | |
202f72d5 PJ |
61 | /* |
62 | * Tracks how many cpusets are currently defined in system. | |
63 | * When there is only one cpuset (the root cpuset) we can | |
64 | * short circuit some hooks. | |
65 | */ | |
7edc5962 | 66 | int number_of_cpusets __read_mostly; |
202f72d5 | 67 | |
2df167a3 | 68 | /* Forward declare cgroup structures */ |
8793d854 PM |
69 | struct cgroup_subsys cpuset_subsys; |
70 | struct cpuset; | |
71 | ||
3e0d98b9 PJ |
72 | /* See "Frequency meter" comments, below. */ |
73 | ||
74 | struct fmeter { | |
75 | int cnt; /* unprocessed events count */ | |
76 | int val; /* most recent output value */ | |
77 | time_t time; /* clock (secs) when val computed */ | |
78 | spinlock_t lock; /* guards read or write of above */ | |
79 | }; | |
80 | ||
1da177e4 | 81 | struct cpuset { |
8793d854 PM |
82 | struct cgroup_subsys_state css; |
83 | ||
1da177e4 LT |
84 | unsigned long flags; /* "unsigned long" so bitops work */ |
85 | cpumask_t cpus_allowed; /* CPUs allowed to tasks in cpuset */ | |
86 | nodemask_t mems_allowed; /* Memory Nodes allowed to tasks */ | |
87 | ||
1da177e4 | 88 | struct cpuset *parent; /* my parent */ |
1da177e4 LT |
89 | |
90 | /* | |
91 | * Copy of global cpuset_mems_generation as of the most | |
92 | * recent time this cpuset changed its mems_allowed. | |
93 | */ | |
3e0d98b9 PJ |
94 | int mems_generation; |
95 | ||
96 | struct fmeter fmeter; /* memory_pressure filter */ | |
029190c5 PJ |
97 | |
98 | /* partition number for rebuild_sched_domains() */ | |
99 | int pn; | |
956db3ca | 100 | |
1d3504fc HS |
101 | /* for custom sched domain */ |
102 | int relax_domain_level; | |
103 | ||
956db3ca CW |
104 | /* used for walking a cpuset heirarchy */ |
105 | struct list_head stack_list; | |
1da177e4 LT |
106 | }; |
107 | ||
8793d854 PM |
108 | /* Retrieve the cpuset for a cgroup */ |
109 | static inline struct cpuset *cgroup_cs(struct cgroup *cont) | |
110 | { | |
111 | return container_of(cgroup_subsys_state(cont, cpuset_subsys_id), | |
112 | struct cpuset, css); | |
113 | } | |
114 | ||
115 | /* Retrieve the cpuset for a task */ | |
116 | static inline struct cpuset *task_cs(struct task_struct *task) | |
117 | { | |
118 | return container_of(task_subsys_state(task, cpuset_subsys_id), | |
119 | struct cpuset, css); | |
120 | } | |
956db3ca CW |
121 | struct cpuset_hotplug_scanner { |
122 | struct cgroup_scanner scan; | |
123 | struct cgroup *to; | |
124 | }; | |
8793d854 | 125 | |
1da177e4 LT |
126 | /* bits in struct cpuset flags field */ |
127 | typedef enum { | |
128 | CS_CPU_EXCLUSIVE, | |
129 | CS_MEM_EXCLUSIVE, | |
78608366 | 130 | CS_MEM_HARDWALL, |
45b07ef3 | 131 | CS_MEMORY_MIGRATE, |
029190c5 | 132 | CS_SCHED_LOAD_BALANCE, |
825a46af PJ |
133 | CS_SPREAD_PAGE, |
134 | CS_SPREAD_SLAB, | |
1da177e4 LT |
135 | } cpuset_flagbits_t; |
136 | ||
137 | /* convenient tests for these bits */ | |
138 | static inline int is_cpu_exclusive(const struct cpuset *cs) | |
139 | { | |
7b5b9ef0 | 140 | return test_bit(CS_CPU_EXCLUSIVE, &cs->flags); |
1da177e4 LT |
141 | } |
142 | ||
143 | static inline int is_mem_exclusive(const struct cpuset *cs) | |
144 | { | |
7b5b9ef0 | 145 | return test_bit(CS_MEM_EXCLUSIVE, &cs->flags); |
1da177e4 LT |
146 | } |
147 | ||
78608366 PM |
148 | static inline int is_mem_hardwall(const struct cpuset *cs) |
149 | { | |
150 | return test_bit(CS_MEM_HARDWALL, &cs->flags); | |
151 | } | |
152 | ||
029190c5 PJ |
153 | static inline int is_sched_load_balance(const struct cpuset *cs) |
154 | { | |
155 | return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); | |
156 | } | |
157 | ||
45b07ef3 PJ |
158 | static inline int is_memory_migrate(const struct cpuset *cs) |
159 | { | |
7b5b9ef0 | 160 | return test_bit(CS_MEMORY_MIGRATE, &cs->flags); |
45b07ef3 PJ |
161 | } |
162 | ||
825a46af PJ |
163 | static inline int is_spread_page(const struct cpuset *cs) |
164 | { | |
165 | return test_bit(CS_SPREAD_PAGE, &cs->flags); | |
166 | } | |
167 | ||
168 | static inline int is_spread_slab(const struct cpuset *cs) | |
169 | { | |
170 | return test_bit(CS_SPREAD_SLAB, &cs->flags); | |
171 | } | |
172 | ||
1da177e4 | 173 | /* |
151a4420 | 174 | * Increment this integer everytime any cpuset changes its |
1da177e4 LT |
175 | * mems_allowed value. Users of cpusets can track this generation |
176 | * number, and avoid having to lock and reload mems_allowed unless | |
177 | * the cpuset they're using changes generation. | |
178 | * | |
2df167a3 | 179 | * A single, global generation is needed because cpuset_attach_task() could |
1da177e4 LT |
180 | * reattach a task to a different cpuset, which must not have its |
181 | * generation numbers aliased with those of that tasks previous cpuset. | |
182 | * | |
183 | * Generations are needed for mems_allowed because one task cannot | |
2df167a3 | 184 | * modify another's memory placement. So we must enable every task, |
1da177e4 LT |
185 | * on every visit to __alloc_pages(), to efficiently check whether |
186 | * its current->cpuset->mems_allowed has changed, requiring an update | |
187 | * of its current->mems_allowed. | |
151a4420 | 188 | * |
2df167a3 | 189 | * Since writes to cpuset_mems_generation are guarded by the cgroup lock |
151a4420 | 190 | * there is no need to mark it atomic. |
1da177e4 | 191 | */ |
151a4420 | 192 | static int cpuset_mems_generation; |
1da177e4 LT |
193 | |
194 | static struct cpuset top_cpuset = { | |
195 | .flags = ((1 << CS_CPU_EXCLUSIVE) | (1 << CS_MEM_EXCLUSIVE)), | |
196 | .cpus_allowed = CPU_MASK_ALL, | |
197 | .mems_allowed = NODE_MASK_ALL, | |
1da177e4 LT |
198 | }; |
199 | ||
1da177e4 | 200 | /* |
2df167a3 PM |
201 | * There are two global mutexes guarding cpuset structures. The first |
202 | * is the main control groups cgroup_mutex, accessed via | |
203 | * cgroup_lock()/cgroup_unlock(). The second is the cpuset-specific | |
204 | * callback_mutex, below. They can nest. It is ok to first take | |
205 | * cgroup_mutex, then nest callback_mutex. We also require taking | |
206 | * task_lock() when dereferencing a task's cpuset pointer. See "The | |
207 | * task_lock() exception", at the end of this comment. | |
053199ed | 208 | * |
3d3f26a7 | 209 | * A task must hold both mutexes to modify cpusets. If a task |
2df167a3 | 210 | * holds cgroup_mutex, then it blocks others wanting that mutex, |
3d3f26a7 | 211 | * ensuring that it is the only task able to also acquire callback_mutex |
053199ed PJ |
212 | * and be able to modify cpusets. It can perform various checks on |
213 | * the cpuset structure first, knowing nothing will change. It can | |
2df167a3 | 214 | * also allocate memory while just holding cgroup_mutex. While it is |
053199ed | 215 | * performing these checks, various callback routines can briefly |
3d3f26a7 IM |
216 | * acquire callback_mutex to query cpusets. Once it is ready to make |
217 | * the changes, it takes callback_mutex, blocking everyone else. | |
053199ed PJ |
218 | * |
219 | * Calls to the kernel memory allocator can not be made while holding | |
3d3f26a7 | 220 | * callback_mutex, as that would risk double tripping on callback_mutex |
053199ed PJ |
221 | * from one of the callbacks into the cpuset code from within |
222 | * __alloc_pages(). | |
223 | * | |
3d3f26a7 | 224 | * If a task is only holding callback_mutex, then it has read-only |
053199ed PJ |
225 | * access to cpusets. |
226 | * | |
227 | * The task_struct fields mems_allowed and mems_generation may only | |
228 | * be accessed in the context of that task, so require no locks. | |
229 | * | |
053199ed | 230 | * The cpuset_common_file_write handler for operations that modify |
2df167a3 | 231 | * the cpuset hierarchy holds cgroup_mutex across the entire operation, |
053199ed PJ |
232 | * single threading all such cpuset modifications across the system. |
233 | * | |
3d3f26a7 | 234 | * The cpuset_common_file_read() handlers only hold callback_mutex across |
053199ed PJ |
235 | * small pieces of code, such as when reading out possibly multi-word |
236 | * cpumasks and nodemasks. | |
237 | * | |
2df167a3 PM |
238 | * Accessing a task's cpuset should be done in accordance with the |
239 | * guidelines for accessing subsystem state in kernel/cgroup.c | |
1da177e4 LT |
240 | */ |
241 | ||
3d3f26a7 | 242 | static DEFINE_MUTEX(callback_mutex); |
4247bdc6 | 243 | |
8793d854 PM |
244 | /* This is ugly, but preserves the userspace API for existing cpuset |
245 | * users. If someone tries to mount the "cpuset" filesystem, we | |
246 | * silently switch it to mount "cgroup" instead */ | |
454e2398 DH |
247 | static int cpuset_get_sb(struct file_system_type *fs_type, |
248 | int flags, const char *unused_dev_name, | |
249 | void *data, struct vfsmount *mnt) | |
1da177e4 | 250 | { |
8793d854 PM |
251 | struct file_system_type *cgroup_fs = get_fs_type("cgroup"); |
252 | int ret = -ENODEV; | |
253 | if (cgroup_fs) { | |
254 | char mountopts[] = | |
255 | "cpuset,noprefix," | |
256 | "release_agent=/sbin/cpuset_release_agent"; | |
257 | ret = cgroup_fs->get_sb(cgroup_fs, flags, | |
258 | unused_dev_name, mountopts, mnt); | |
259 | put_filesystem(cgroup_fs); | |
260 | } | |
261 | return ret; | |
1da177e4 LT |
262 | } |
263 | ||
264 | static struct file_system_type cpuset_fs_type = { | |
265 | .name = "cpuset", | |
266 | .get_sb = cpuset_get_sb, | |
1da177e4 LT |
267 | }; |
268 | ||
1da177e4 LT |
269 | /* |
270 | * Return in *pmask the portion of a cpusets's cpus_allowed that | |
271 | * are online. If none are online, walk up the cpuset hierarchy | |
272 | * until we find one that does have some online cpus. If we get | |
273 | * all the way to the top and still haven't found any online cpus, | |
274 | * return cpu_online_map. Or if passed a NULL cs from an exit'ing | |
275 | * task, return cpu_online_map. | |
276 | * | |
277 | * One way or another, we guarantee to return some non-empty subset | |
278 | * of cpu_online_map. | |
279 | * | |
3d3f26a7 | 280 | * Call with callback_mutex held. |
1da177e4 LT |
281 | */ |
282 | ||
283 | static void guarantee_online_cpus(const struct cpuset *cs, cpumask_t *pmask) | |
284 | { | |
285 | while (cs && !cpus_intersects(cs->cpus_allowed, cpu_online_map)) | |
286 | cs = cs->parent; | |
287 | if (cs) | |
288 | cpus_and(*pmask, cs->cpus_allowed, cpu_online_map); | |
289 | else | |
290 | *pmask = cpu_online_map; | |
291 | BUG_ON(!cpus_intersects(*pmask, cpu_online_map)); | |
292 | } | |
293 | ||
294 | /* | |
295 | * Return in *pmask the portion of a cpusets's mems_allowed that | |
0e1e7c7a CL |
296 | * are online, with memory. If none are online with memory, walk |
297 | * up the cpuset hierarchy until we find one that does have some | |
298 | * online mems. If we get all the way to the top and still haven't | |
299 | * found any online mems, return node_states[N_HIGH_MEMORY]. | |
1da177e4 LT |
300 | * |
301 | * One way or another, we guarantee to return some non-empty subset | |
0e1e7c7a | 302 | * of node_states[N_HIGH_MEMORY]. |
1da177e4 | 303 | * |
3d3f26a7 | 304 | * Call with callback_mutex held. |
1da177e4 LT |
305 | */ |
306 | ||
307 | static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask) | |
308 | { | |
0e1e7c7a CL |
309 | while (cs && !nodes_intersects(cs->mems_allowed, |
310 | node_states[N_HIGH_MEMORY])) | |
1da177e4 LT |
311 | cs = cs->parent; |
312 | if (cs) | |
0e1e7c7a CL |
313 | nodes_and(*pmask, cs->mems_allowed, |
314 | node_states[N_HIGH_MEMORY]); | |
1da177e4 | 315 | else |
0e1e7c7a CL |
316 | *pmask = node_states[N_HIGH_MEMORY]; |
317 | BUG_ON(!nodes_intersects(*pmask, node_states[N_HIGH_MEMORY])); | |
1da177e4 LT |
318 | } |
319 | ||
cf2a473c PJ |
320 | /** |
321 | * cpuset_update_task_memory_state - update task memory placement | |
322 | * | |
323 | * If the current tasks cpusets mems_allowed changed behind our | |
324 | * backs, update current->mems_allowed, mems_generation and task NUMA | |
325 | * mempolicy to the new value. | |
053199ed | 326 | * |
cf2a473c PJ |
327 | * Task mempolicy is updated by rebinding it relative to the |
328 | * current->cpuset if a task has its memory placement changed. | |
329 | * Do not call this routine if in_interrupt(). | |
330 | * | |
4a01c8d5 | 331 | * Call without callback_mutex or task_lock() held. May be |
2df167a3 PM |
332 | * called with or without cgroup_mutex held. Thanks in part to |
333 | * 'the_top_cpuset_hack', the task's cpuset pointer will never | |
41f7f60d DR |
334 | * be NULL. This routine also might acquire callback_mutex during |
335 | * call. | |
053199ed | 336 | * |
6b9c2603 PJ |
337 | * Reading current->cpuset->mems_generation doesn't need task_lock |
338 | * to guard the current->cpuset derefence, because it is guarded | |
2df167a3 | 339 | * from concurrent freeing of current->cpuset using RCU. |
6b9c2603 PJ |
340 | * |
341 | * The rcu_dereference() is technically probably not needed, | |
342 | * as I don't actually mind if I see a new cpuset pointer but | |
343 | * an old value of mems_generation. However this really only | |
344 | * matters on alpha systems using cpusets heavily. If I dropped | |
345 | * that rcu_dereference(), it would save them a memory barrier. | |
346 | * For all other arch's, rcu_dereference is a no-op anyway, and for | |
347 | * alpha systems not using cpusets, another planned optimization, | |
348 | * avoiding the rcu critical section for tasks in the root cpuset | |
349 | * which is statically allocated, so can't vanish, will make this | |
350 | * irrelevant. Better to use RCU as intended, than to engage in | |
351 | * some cute trick to save a memory barrier that is impossible to | |
352 | * test, for alpha systems using cpusets heavily, which might not | |
353 | * even exist. | |
053199ed PJ |
354 | * |
355 | * This routine is needed to update the per-task mems_allowed data, | |
356 | * within the tasks context, when it is trying to allocate memory | |
357 | * (in various mm/mempolicy.c routines) and notices that some other | |
358 | * task has been modifying its cpuset. | |
1da177e4 LT |
359 | */ |
360 | ||
fe85a998 | 361 | void cpuset_update_task_memory_state(void) |
1da177e4 | 362 | { |
053199ed | 363 | int my_cpusets_mem_gen; |
cf2a473c | 364 | struct task_struct *tsk = current; |
6b9c2603 | 365 | struct cpuset *cs; |
053199ed | 366 | |
8793d854 | 367 | if (task_cs(tsk) == &top_cpuset) { |
03a285f5 PJ |
368 | /* Don't need rcu for top_cpuset. It's never freed. */ |
369 | my_cpusets_mem_gen = top_cpuset.mems_generation; | |
370 | } else { | |
371 | rcu_read_lock(); | |
8793d854 | 372 | my_cpusets_mem_gen = task_cs(current)->mems_generation; |
03a285f5 PJ |
373 | rcu_read_unlock(); |
374 | } | |
1da177e4 | 375 | |
cf2a473c | 376 | if (my_cpusets_mem_gen != tsk->cpuset_mems_generation) { |
3d3f26a7 | 377 | mutex_lock(&callback_mutex); |
cf2a473c | 378 | task_lock(tsk); |
8793d854 | 379 | cs = task_cs(tsk); /* Maybe changed when task not locked */ |
cf2a473c PJ |
380 | guarantee_online_mems(cs, &tsk->mems_allowed); |
381 | tsk->cpuset_mems_generation = cs->mems_generation; | |
825a46af PJ |
382 | if (is_spread_page(cs)) |
383 | tsk->flags |= PF_SPREAD_PAGE; | |
384 | else | |
385 | tsk->flags &= ~PF_SPREAD_PAGE; | |
386 | if (is_spread_slab(cs)) | |
387 | tsk->flags |= PF_SPREAD_SLAB; | |
388 | else | |
389 | tsk->flags &= ~PF_SPREAD_SLAB; | |
cf2a473c | 390 | task_unlock(tsk); |
3d3f26a7 | 391 | mutex_unlock(&callback_mutex); |
74cb2155 | 392 | mpol_rebind_task(tsk, &tsk->mems_allowed); |
1da177e4 LT |
393 | } |
394 | } | |
395 | ||
396 | /* | |
397 | * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q? | |
398 | * | |
399 | * One cpuset is a subset of another if all its allowed CPUs and | |
400 | * Memory Nodes are a subset of the other, and its exclusive flags | |
2df167a3 | 401 | * are only set if the other's are set. Call holding cgroup_mutex. |
1da177e4 LT |
402 | */ |
403 | ||
404 | static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q) | |
405 | { | |
406 | return cpus_subset(p->cpus_allowed, q->cpus_allowed) && | |
407 | nodes_subset(p->mems_allowed, q->mems_allowed) && | |
408 | is_cpu_exclusive(p) <= is_cpu_exclusive(q) && | |
409 | is_mem_exclusive(p) <= is_mem_exclusive(q); | |
410 | } | |
411 | ||
412 | /* | |
413 | * validate_change() - Used to validate that any proposed cpuset change | |
414 | * follows the structural rules for cpusets. | |
415 | * | |
416 | * If we replaced the flag and mask values of the current cpuset | |
417 | * (cur) with those values in the trial cpuset (trial), would | |
418 | * our various subset and exclusive rules still be valid? Presumes | |
2df167a3 | 419 | * cgroup_mutex held. |
1da177e4 LT |
420 | * |
421 | * 'cur' is the address of an actual, in-use cpuset. Operations | |
422 | * such as list traversal that depend on the actual address of the | |
423 | * cpuset in the list must use cur below, not trial. | |
424 | * | |
425 | * 'trial' is the address of bulk structure copy of cur, with | |
426 | * perhaps one or more of the fields cpus_allowed, mems_allowed, | |
427 | * or flags changed to new, trial values. | |
428 | * | |
429 | * Return 0 if valid, -errno if not. | |
430 | */ | |
431 | ||
432 | static int validate_change(const struct cpuset *cur, const struct cpuset *trial) | |
433 | { | |
8793d854 | 434 | struct cgroup *cont; |
1da177e4 LT |
435 | struct cpuset *c, *par; |
436 | ||
437 | /* Each of our child cpusets must be a subset of us */ | |
8793d854 PM |
438 | list_for_each_entry(cont, &cur->css.cgroup->children, sibling) { |
439 | if (!is_cpuset_subset(cgroup_cs(cont), trial)) | |
1da177e4 LT |
440 | return -EBUSY; |
441 | } | |
442 | ||
443 | /* Remaining checks don't apply to root cpuset */ | |
69604067 | 444 | if (cur == &top_cpuset) |
1da177e4 LT |
445 | return 0; |
446 | ||
69604067 PJ |
447 | par = cur->parent; |
448 | ||
1da177e4 LT |
449 | /* We must be a subset of our parent cpuset */ |
450 | if (!is_cpuset_subset(trial, par)) | |
451 | return -EACCES; | |
452 | ||
2df167a3 PM |
453 | /* |
454 | * If either I or some sibling (!= me) is exclusive, we can't | |
455 | * overlap | |
456 | */ | |
8793d854 PM |
457 | list_for_each_entry(cont, &par->css.cgroup->children, sibling) { |
458 | c = cgroup_cs(cont); | |
1da177e4 LT |
459 | if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) && |
460 | c != cur && | |
461 | cpus_intersects(trial->cpus_allowed, c->cpus_allowed)) | |
462 | return -EINVAL; | |
463 | if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) && | |
464 | c != cur && | |
465 | nodes_intersects(trial->mems_allowed, c->mems_allowed)) | |
466 | return -EINVAL; | |
467 | } | |
468 | ||
020958b6 PJ |
469 | /* Cpusets with tasks can't have empty cpus_allowed or mems_allowed */ |
470 | if (cgroup_task_count(cur->css.cgroup)) { | |
471 | if (cpus_empty(trial->cpus_allowed) || | |
472 | nodes_empty(trial->mems_allowed)) { | |
473 | return -ENOSPC; | |
474 | } | |
475 | } | |
476 | ||
1da177e4 LT |
477 | return 0; |
478 | } | |
479 | ||
029190c5 PJ |
480 | /* |
481 | * Helper routine for rebuild_sched_domains(). | |
482 | * Do cpusets a, b have overlapping cpus_allowed masks? | |
483 | */ | |
484 | ||
485 | static int cpusets_overlap(struct cpuset *a, struct cpuset *b) | |
486 | { | |
487 | return cpus_intersects(a->cpus_allowed, b->cpus_allowed); | |
488 | } | |
489 | ||
1d3504fc HS |
490 | static void |
491 | update_domain_attr(struct sched_domain_attr *dattr, struct cpuset *c) | |
492 | { | |
493 | if (!dattr) | |
494 | return; | |
495 | if (dattr->relax_domain_level < c->relax_domain_level) | |
496 | dattr->relax_domain_level = c->relax_domain_level; | |
497 | return; | |
498 | } | |
499 | ||
029190c5 PJ |
500 | /* |
501 | * rebuild_sched_domains() | |
502 | * | |
503 | * If the flag 'sched_load_balance' of any cpuset with non-empty | |
504 | * 'cpus' changes, or if the 'cpus' allowed changes in any cpuset | |
505 | * which has that flag enabled, or if any cpuset with a non-empty | |
506 | * 'cpus' is removed, then call this routine to rebuild the | |
507 | * scheduler's dynamic sched domains. | |
508 | * | |
509 | * This routine builds a partial partition of the systems CPUs | |
510 | * (the set of non-overlappping cpumask_t's in the array 'part' | |
511 | * below), and passes that partial partition to the kernel/sched.c | |
512 | * partition_sched_domains() routine, which will rebuild the | |
513 | * schedulers load balancing domains (sched domains) as specified | |
514 | * by that partial partition. A 'partial partition' is a set of | |
515 | * non-overlapping subsets whose union is a subset of that set. | |
516 | * | |
517 | * See "What is sched_load_balance" in Documentation/cpusets.txt | |
518 | * for a background explanation of this. | |
519 | * | |
520 | * Does not return errors, on the theory that the callers of this | |
521 | * routine would rather not worry about failures to rebuild sched | |
522 | * domains when operating in the severe memory shortage situations | |
523 | * that could cause allocation failures below. | |
524 | * | |
525 | * Call with cgroup_mutex held. May take callback_mutex during | |
526 | * call due to the kfifo_alloc() and kmalloc() calls. May nest | |
86ef5c9a | 527 | * a call to the get_online_cpus()/put_online_cpus() pair. |
029190c5 | 528 | * Must not be called holding callback_mutex, because we must not |
86ef5c9a GS |
529 | * call get_online_cpus() while holding callback_mutex. Elsewhere |
530 | * the kernel nests callback_mutex inside get_online_cpus() calls. | |
029190c5 PJ |
531 | * So the reverse nesting would risk an ABBA deadlock. |
532 | * | |
533 | * The three key local variables below are: | |
534 | * q - a kfifo queue of cpuset pointers, used to implement a | |
535 | * top-down scan of all cpusets. This scan loads a pointer | |
536 | * to each cpuset marked is_sched_load_balance into the | |
537 | * array 'csa'. For our purposes, rebuilding the schedulers | |
538 | * sched domains, we can ignore !is_sched_load_balance cpusets. | |
539 | * csa - (for CpuSet Array) Array of pointers to all the cpusets | |
540 | * that need to be load balanced, for convenient iterative | |
541 | * access by the subsequent code that finds the best partition, | |
542 | * i.e the set of domains (subsets) of CPUs such that the | |
543 | * cpus_allowed of every cpuset marked is_sched_load_balance | |
544 | * is a subset of one of these domains, while there are as | |
545 | * many such domains as possible, each as small as possible. | |
546 | * doms - Conversion of 'csa' to an array of cpumasks, for passing to | |
547 | * the kernel/sched.c routine partition_sched_domains() in a | |
548 | * convenient format, that can be easily compared to the prior | |
549 | * value to determine what partition elements (sched domains) | |
550 | * were changed (added or removed.) | |
551 | * | |
552 | * Finding the best partition (set of domains): | |
553 | * The triple nested loops below over i, j, k scan over the | |
554 | * load balanced cpusets (using the array of cpuset pointers in | |
555 | * csa[]) looking for pairs of cpusets that have overlapping | |
556 | * cpus_allowed, but which don't have the same 'pn' partition | |
557 | * number and gives them in the same partition number. It keeps | |
558 | * looping on the 'restart' label until it can no longer find | |
559 | * any such pairs. | |
560 | * | |
561 | * The union of the cpus_allowed masks from the set of | |
562 | * all cpusets having the same 'pn' value then form the one | |
563 | * element of the partition (one sched domain) to be passed to | |
564 | * partition_sched_domains(). | |
565 | */ | |
566 | ||
e761b772 | 567 | void rebuild_sched_domains(void) |
029190c5 PJ |
568 | { |
569 | struct kfifo *q; /* queue of cpusets to be scanned */ | |
570 | struct cpuset *cp; /* scans q */ | |
571 | struct cpuset **csa; /* array of all cpuset ptrs */ | |
572 | int csn; /* how many cpuset ptrs in csa so far */ | |
573 | int i, j, k; /* indices for partition finding loops */ | |
574 | cpumask_t *doms; /* resulting partition; i.e. sched domains */ | |
1d3504fc | 575 | struct sched_domain_attr *dattr; /* attributes for custom domains */ |
029190c5 PJ |
576 | int ndoms; /* number of sched domains in result */ |
577 | int nslot; /* next empty doms[] cpumask_t slot */ | |
578 | ||
579 | q = NULL; | |
580 | csa = NULL; | |
581 | doms = NULL; | |
1d3504fc | 582 | dattr = NULL; |
029190c5 PJ |
583 | |
584 | /* Special case for the 99% of systems with one, full, sched domain */ | |
585 | if (is_sched_load_balance(&top_cpuset)) { | |
586 | ndoms = 1; | |
587 | doms = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
588 | if (!doms) | |
589 | goto rebuild; | |
1d3504fc HS |
590 | dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL); |
591 | if (dattr) { | |
592 | *dattr = SD_ATTR_INIT; | |
593 | update_domain_attr(dattr, &top_cpuset); | |
594 | } | |
029190c5 PJ |
595 | *doms = top_cpuset.cpus_allowed; |
596 | goto rebuild; | |
597 | } | |
598 | ||
599 | q = kfifo_alloc(number_of_cpusets * sizeof(cp), GFP_KERNEL, NULL); | |
600 | if (IS_ERR(q)) | |
601 | goto done; | |
602 | csa = kmalloc(number_of_cpusets * sizeof(cp), GFP_KERNEL); | |
603 | if (!csa) | |
604 | goto done; | |
605 | csn = 0; | |
606 | ||
607 | cp = &top_cpuset; | |
608 | __kfifo_put(q, (void *)&cp, sizeof(cp)); | |
609 | while (__kfifo_get(q, (void *)&cp, sizeof(cp))) { | |
610 | struct cgroup *cont; | |
611 | struct cpuset *child; /* scans child cpusets of cp */ | |
612 | if (is_sched_load_balance(cp)) | |
613 | csa[csn++] = cp; | |
614 | list_for_each_entry(cont, &cp->css.cgroup->children, sibling) { | |
615 | child = cgroup_cs(cont); | |
616 | __kfifo_put(q, (void *)&child, sizeof(cp)); | |
617 | } | |
618 | } | |
619 | ||
620 | for (i = 0; i < csn; i++) | |
621 | csa[i]->pn = i; | |
622 | ndoms = csn; | |
623 | ||
624 | restart: | |
625 | /* Find the best partition (set of sched domains) */ | |
626 | for (i = 0; i < csn; i++) { | |
627 | struct cpuset *a = csa[i]; | |
628 | int apn = a->pn; | |
629 | ||
630 | for (j = 0; j < csn; j++) { | |
631 | struct cpuset *b = csa[j]; | |
632 | int bpn = b->pn; | |
633 | ||
634 | if (apn != bpn && cpusets_overlap(a, b)) { | |
635 | for (k = 0; k < csn; k++) { | |
636 | struct cpuset *c = csa[k]; | |
637 | ||
638 | if (c->pn == bpn) | |
639 | c->pn = apn; | |
640 | } | |
641 | ndoms--; /* one less element */ | |
642 | goto restart; | |
643 | } | |
644 | } | |
645 | } | |
646 | ||
647 | /* Convert <csn, csa> to <ndoms, doms> */ | |
648 | doms = kmalloc(ndoms * sizeof(cpumask_t), GFP_KERNEL); | |
649 | if (!doms) | |
650 | goto rebuild; | |
1d3504fc | 651 | dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL); |
029190c5 PJ |
652 | |
653 | for (nslot = 0, i = 0; i < csn; i++) { | |
654 | struct cpuset *a = csa[i]; | |
655 | int apn = a->pn; | |
656 | ||
657 | if (apn >= 0) { | |
658 | cpumask_t *dp = doms + nslot; | |
659 | ||
660 | if (nslot == ndoms) { | |
661 | static int warnings = 10; | |
662 | if (warnings) { | |
663 | printk(KERN_WARNING | |
664 | "rebuild_sched_domains confused:" | |
665 | " nslot %d, ndoms %d, csn %d, i %d," | |
666 | " apn %d\n", | |
667 | nslot, ndoms, csn, i, apn); | |
668 | warnings--; | |
669 | } | |
670 | continue; | |
671 | } | |
672 | ||
673 | cpus_clear(*dp); | |
1d3504fc HS |
674 | if (dattr) |
675 | *(dattr + nslot) = SD_ATTR_INIT; | |
029190c5 PJ |
676 | for (j = i; j < csn; j++) { |
677 | struct cpuset *b = csa[j]; | |
678 | ||
679 | if (apn == b->pn) { | |
680 | cpus_or(*dp, *dp, b->cpus_allowed); | |
681 | b->pn = -1; | |
91cd4d6e MX |
682 | if (dattr) |
683 | update_domain_attr(dattr | |
684 | + nslot, b); | |
029190c5 PJ |
685 | } |
686 | } | |
687 | nslot++; | |
688 | } | |
689 | } | |
690 | BUG_ON(nslot != ndoms); | |
691 | ||
692 | rebuild: | |
693 | /* Have scheduler rebuild sched domains */ | |
86ef5c9a | 694 | get_online_cpus(); |
1d3504fc | 695 | partition_sched_domains(ndoms, doms, dattr); |
86ef5c9a | 696 | put_online_cpus(); |
029190c5 PJ |
697 | |
698 | done: | |
699 | if (q && !IS_ERR(q)) | |
700 | kfifo_free(q); | |
701 | kfree(csa); | |
702 | /* Don't kfree(doms) -- partition_sched_domains() does that. */ | |
1d3504fc | 703 | /* Don't kfree(dattr) -- partition_sched_domains() does that. */ |
029190c5 PJ |
704 | } |
705 | ||
8707d8b8 PM |
706 | static inline int started_after_time(struct task_struct *t1, |
707 | struct timespec *time, | |
708 | struct task_struct *t2) | |
709 | { | |
710 | int start_diff = timespec_compare(&t1->start_time, time); | |
711 | if (start_diff > 0) { | |
712 | return 1; | |
713 | } else if (start_diff < 0) { | |
714 | return 0; | |
715 | } else { | |
716 | /* | |
717 | * Arbitrarily, if two processes started at the same | |
718 | * time, we'll say that the lower pointer value | |
719 | * started first. Note that t2 may have exited by now | |
720 | * so this may not be a valid pointer any longer, but | |
721 | * that's fine - it still serves to distinguish | |
722 | * between two tasks started (effectively) | |
723 | * simultaneously. | |
724 | */ | |
725 | return t1 > t2; | |
726 | } | |
727 | } | |
728 | ||
729 | static inline int started_after(void *p1, void *p2) | |
730 | { | |
731 | struct task_struct *t1 = p1; | |
732 | struct task_struct *t2 = p2; | |
733 | return started_after_time(t1, &t2->start_time, t2); | |
734 | } | |
735 | ||
58f4790b CW |
736 | /** |
737 | * cpuset_test_cpumask - test a task's cpus_allowed versus its cpuset's | |
738 | * @tsk: task to test | |
739 | * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner | |
740 | * | |
2df167a3 | 741 | * Call with cgroup_mutex held. May take callback_mutex during call. |
58f4790b CW |
742 | * Called for each task in a cgroup by cgroup_scan_tasks(). |
743 | * Return nonzero if this tasks's cpus_allowed mask should be changed (in other | |
744 | * words, if its mask is not equal to its cpuset's mask). | |
053199ed | 745 | */ |
9e0c914c AB |
746 | static int cpuset_test_cpumask(struct task_struct *tsk, |
747 | struct cgroup_scanner *scan) | |
58f4790b CW |
748 | { |
749 | return !cpus_equal(tsk->cpus_allowed, | |
750 | (cgroup_cs(scan->cg))->cpus_allowed); | |
751 | } | |
053199ed | 752 | |
58f4790b CW |
753 | /** |
754 | * cpuset_change_cpumask - make a task's cpus_allowed the same as its cpuset's | |
755 | * @tsk: task to test | |
756 | * @scan: struct cgroup_scanner containing the cgroup of the task | |
757 | * | |
758 | * Called by cgroup_scan_tasks() for each task in a cgroup whose | |
759 | * cpus_allowed mask needs to be changed. | |
760 | * | |
761 | * We don't need to re-check for the cgroup/cpuset membership, since we're | |
762 | * holding cgroup_lock() at this point. | |
763 | */ | |
9e0c914c AB |
764 | static void cpuset_change_cpumask(struct task_struct *tsk, |
765 | struct cgroup_scanner *scan) | |
58f4790b | 766 | { |
f9a86fcb | 767 | set_cpus_allowed_ptr(tsk, &((cgroup_cs(scan->cg))->cpus_allowed)); |
58f4790b CW |
768 | } |
769 | ||
770 | /** | |
771 | * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it | |
772 | * @cs: the cpuset to consider | |
773 | * @buf: buffer of cpu numbers written to this cpuset | |
774 | */ | |
1da177e4 LT |
775 | static int update_cpumask(struct cpuset *cs, char *buf) |
776 | { | |
777 | struct cpuset trialcs; | |
58f4790b | 778 | struct cgroup_scanner scan; |
8707d8b8 | 779 | struct ptr_heap heap; |
58f4790b CW |
780 | int retval; |
781 | int is_load_balanced; | |
1da177e4 | 782 | |
4c4d50f7 PJ |
783 | /* top_cpuset.cpus_allowed tracks cpu_online_map; it's read-only */ |
784 | if (cs == &top_cpuset) | |
785 | return -EACCES; | |
786 | ||
1da177e4 | 787 | trialcs = *cs; |
6f7f02e7 DR |
788 | |
789 | /* | |
c8d9c90c | 790 | * An empty cpus_allowed is ok only if the cpuset has no tasks. |
020958b6 PJ |
791 | * Since cpulist_parse() fails on an empty mask, we special case |
792 | * that parsing. The validate_change() call ensures that cpusets | |
793 | * with tasks have cpus. | |
6f7f02e7 | 794 | */ |
020958b6 PJ |
795 | buf = strstrip(buf); |
796 | if (!*buf) { | |
6f7f02e7 DR |
797 | cpus_clear(trialcs.cpus_allowed); |
798 | } else { | |
799 | retval = cpulist_parse(buf, trialcs.cpus_allowed); | |
800 | if (retval < 0) | |
801 | return retval; | |
37340746 LJ |
802 | |
803 | if (!cpus_subset(trialcs.cpus_allowed, cpu_online_map)) | |
804 | return -EINVAL; | |
6f7f02e7 | 805 | } |
1da177e4 | 806 | retval = validate_change(cs, &trialcs); |
85d7b949 DG |
807 | if (retval < 0) |
808 | return retval; | |
029190c5 | 809 | |
8707d8b8 PM |
810 | /* Nothing to do if the cpus didn't change */ |
811 | if (cpus_equal(cs->cpus_allowed, trialcs.cpus_allowed)) | |
812 | return 0; | |
58f4790b | 813 | |
8707d8b8 PM |
814 | retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, &started_after); |
815 | if (retval) | |
816 | return retval; | |
817 | ||
029190c5 PJ |
818 | is_load_balanced = is_sched_load_balance(&trialcs); |
819 | ||
3d3f26a7 | 820 | mutex_lock(&callback_mutex); |
85d7b949 | 821 | cs->cpus_allowed = trialcs.cpus_allowed; |
3d3f26a7 | 822 | mutex_unlock(&callback_mutex); |
029190c5 | 823 | |
8707d8b8 PM |
824 | /* |
825 | * Scan tasks in the cpuset, and update the cpumasks of any | |
58f4790b | 826 | * that need an update. |
8707d8b8 | 827 | */ |
58f4790b CW |
828 | scan.cg = cs->css.cgroup; |
829 | scan.test_task = cpuset_test_cpumask; | |
830 | scan.process_task = cpuset_change_cpumask; | |
831 | scan.heap = &heap; | |
832 | cgroup_scan_tasks(&scan); | |
8707d8b8 | 833 | heap_free(&heap); |
58f4790b | 834 | |
8707d8b8 | 835 | if (is_load_balanced) |
029190c5 | 836 | rebuild_sched_domains(); |
85d7b949 | 837 | return 0; |
1da177e4 LT |
838 | } |
839 | ||
e4e364e8 PJ |
840 | /* |
841 | * cpuset_migrate_mm | |
842 | * | |
843 | * Migrate memory region from one set of nodes to another. | |
844 | * | |
845 | * Temporarilly set tasks mems_allowed to target nodes of migration, | |
846 | * so that the migration code can allocate pages on these nodes. | |
847 | * | |
2df167a3 | 848 | * Call holding cgroup_mutex, so current's cpuset won't change |
c8d9c90c | 849 | * during this call, as manage_mutex holds off any cpuset_attach() |
e4e364e8 PJ |
850 | * calls. Therefore we don't need to take task_lock around the |
851 | * call to guarantee_online_mems(), as we know no one is changing | |
2df167a3 | 852 | * our task's cpuset. |
e4e364e8 PJ |
853 | * |
854 | * Hold callback_mutex around the two modifications of our tasks | |
855 | * mems_allowed to synchronize with cpuset_mems_allowed(). | |
856 | * | |
857 | * While the mm_struct we are migrating is typically from some | |
858 | * other task, the task_struct mems_allowed that we are hacking | |
859 | * is for our current task, which must allocate new pages for that | |
860 | * migrating memory region. | |
861 | * | |
862 | * We call cpuset_update_task_memory_state() before hacking | |
863 | * our tasks mems_allowed, so that we are assured of being in | |
864 | * sync with our tasks cpuset, and in particular, callbacks to | |
865 | * cpuset_update_task_memory_state() from nested page allocations | |
866 | * won't see any mismatch of our cpuset and task mems_generation | |
867 | * values, so won't overwrite our hacked tasks mems_allowed | |
868 | * nodemask. | |
869 | */ | |
870 | ||
871 | static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from, | |
872 | const nodemask_t *to) | |
873 | { | |
874 | struct task_struct *tsk = current; | |
875 | ||
876 | cpuset_update_task_memory_state(); | |
877 | ||
878 | mutex_lock(&callback_mutex); | |
879 | tsk->mems_allowed = *to; | |
880 | mutex_unlock(&callback_mutex); | |
881 | ||
882 | do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL); | |
883 | ||
884 | mutex_lock(&callback_mutex); | |
8793d854 | 885 | guarantee_online_mems(task_cs(tsk),&tsk->mems_allowed); |
e4e364e8 PJ |
886 | mutex_unlock(&callback_mutex); |
887 | } | |
888 | ||
053199ed | 889 | /* |
4225399a PJ |
890 | * Handle user request to change the 'mems' memory placement |
891 | * of a cpuset. Needs to validate the request, update the | |
892 | * cpusets mems_allowed and mems_generation, and for each | |
04c19fa6 PJ |
893 | * task in the cpuset, rebind any vma mempolicies and if |
894 | * the cpuset is marked 'memory_migrate', migrate the tasks | |
895 | * pages to the new memory. | |
4225399a | 896 | * |
2df167a3 | 897 | * Call with cgroup_mutex held. May take callback_mutex during call. |
4225399a PJ |
898 | * Will take tasklist_lock, scan tasklist for tasks in cpuset cs, |
899 | * lock each such tasks mm->mmap_sem, scan its vma's and rebind | |
900 | * their mempolicies to the cpusets new mems_allowed. | |
053199ed PJ |
901 | */ |
902 | ||
8793d854 PM |
903 | static void *cpuset_being_rebound; |
904 | ||
1da177e4 LT |
905 | static int update_nodemask(struct cpuset *cs, char *buf) |
906 | { | |
907 | struct cpuset trialcs; | |
04c19fa6 | 908 | nodemask_t oldmem; |
8793d854 | 909 | struct task_struct *p; |
4225399a PJ |
910 | struct mm_struct **mmarray; |
911 | int i, n, ntasks; | |
04c19fa6 | 912 | int migrate; |
4225399a | 913 | int fudge; |
1da177e4 | 914 | int retval; |
8793d854 | 915 | struct cgroup_iter it; |
1da177e4 | 916 | |
0e1e7c7a CL |
917 | /* |
918 | * top_cpuset.mems_allowed tracks node_stats[N_HIGH_MEMORY]; | |
919 | * it's read-only | |
920 | */ | |
38837fc7 PJ |
921 | if (cs == &top_cpuset) |
922 | return -EACCES; | |
923 | ||
1da177e4 | 924 | trialcs = *cs; |
6f7f02e7 DR |
925 | |
926 | /* | |
020958b6 PJ |
927 | * An empty mems_allowed is ok iff there are no tasks in the cpuset. |
928 | * Since nodelist_parse() fails on an empty mask, we special case | |
929 | * that parsing. The validate_change() call ensures that cpusets | |
930 | * with tasks have memory. | |
6f7f02e7 | 931 | */ |
020958b6 PJ |
932 | buf = strstrip(buf); |
933 | if (!*buf) { | |
6f7f02e7 DR |
934 | nodes_clear(trialcs.mems_allowed); |
935 | } else { | |
936 | retval = nodelist_parse(buf, trialcs.mems_allowed); | |
937 | if (retval < 0) | |
938 | goto done; | |
37340746 LJ |
939 | |
940 | if (!nodes_subset(trialcs.mems_allowed, | |
941 | node_states[N_HIGH_MEMORY])) | |
942 | return -EINVAL; | |
6f7f02e7 | 943 | } |
04c19fa6 PJ |
944 | oldmem = cs->mems_allowed; |
945 | if (nodes_equal(oldmem, trialcs.mems_allowed)) { | |
946 | retval = 0; /* Too easy - nothing to do */ | |
947 | goto done; | |
948 | } | |
59dac16f PJ |
949 | retval = validate_change(cs, &trialcs); |
950 | if (retval < 0) | |
951 | goto done; | |
952 | ||
3d3f26a7 | 953 | mutex_lock(&callback_mutex); |
59dac16f | 954 | cs->mems_allowed = trialcs.mems_allowed; |
151a4420 | 955 | cs->mems_generation = cpuset_mems_generation++; |
3d3f26a7 | 956 | mutex_unlock(&callback_mutex); |
59dac16f | 957 | |
846a16bf | 958 | cpuset_being_rebound = cs; /* causes mpol_dup() rebind */ |
4225399a PJ |
959 | |
960 | fudge = 10; /* spare mmarray[] slots */ | |
961 | fudge += cpus_weight(cs->cpus_allowed); /* imagine one fork-bomb/cpu */ | |
962 | retval = -ENOMEM; | |
963 | ||
964 | /* | |
965 | * Allocate mmarray[] to hold mm reference for each task | |
966 | * in cpuset cs. Can't kmalloc GFP_KERNEL while holding | |
967 | * tasklist_lock. We could use GFP_ATOMIC, but with a | |
968 | * few more lines of code, we can retry until we get a big | |
969 | * enough mmarray[] w/o using GFP_ATOMIC. | |
970 | */ | |
971 | while (1) { | |
8793d854 | 972 | ntasks = cgroup_task_count(cs->css.cgroup); /* guess */ |
4225399a PJ |
973 | ntasks += fudge; |
974 | mmarray = kmalloc(ntasks * sizeof(*mmarray), GFP_KERNEL); | |
975 | if (!mmarray) | |
976 | goto done; | |
c2aef333 | 977 | read_lock(&tasklist_lock); /* block fork */ |
8793d854 | 978 | if (cgroup_task_count(cs->css.cgroup) <= ntasks) |
4225399a | 979 | break; /* got enough */ |
c2aef333 | 980 | read_unlock(&tasklist_lock); /* try again */ |
4225399a PJ |
981 | kfree(mmarray); |
982 | } | |
983 | ||
984 | n = 0; | |
985 | ||
986 | /* Load up mmarray[] with mm reference for each task in cpuset. */ | |
8793d854 PM |
987 | cgroup_iter_start(cs->css.cgroup, &it); |
988 | while ((p = cgroup_iter_next(cs->css.cgroup, &it))) { | |
4225399a PJ |
989 | struct mm_struct *mm; |
990 | ||
991 | if (n >= ntasks) { | |
992 | printk(KERN_WARNING | |
993 | "Cpuset mempolicy rebind incomplete.\n"); | |
8793d854 | 994 | break; |
4225399a | 995 | } |
4225399a PJ |
996 | mm = get_task_mm(p); |
997 | if (!mm) | |
998 | continue; | |
999 | mmarray[n++] = mm; | |
8793d854 PM |
1000 | } |
1001 | cgroup_iter_end(cs->css.cgroup, &it); | |
c2aef333 | 1002 | read_unlock(&tasklist_lock); |
4225399a PJ |
1003 | |
1004 | /* | |
1005 | * Now that we've dropped the tasklist spinlock, we can | |
1006 | * rebind the vma mempolicies of each mm in mmarray[] to their | |
1007 | * new cpuset, and release that mm. The mpol_rebind_mm() | |
1008 | * call takes mmap_sem, which we couldn't take while holding | |
846a16bf | 1009 | * tasklist_lock. Forks can happen again now - the mpol_dup() |
4225399a PJ |
1010 | * cpuset_being_rebound check will catch such forks, and rebind |
1011 | * their vma mempolicies too. Because we still hold the global | |
2df167a3 | 1012 | * cgroup_mutex, we know that no other rebind effort will |
4225399a PJ |
1013 | * be contending for the global variable cpuset_being_rebound. |
1014 | * It's ok if we rebind the same mm twice; mpol_rebind_mm() | |
04c19fa6 | 1015 | * is idempotent. Also migrate pages in each mm to new nodes. |
4225399a | 1016 | */ |
04c19fa6 | 1017 | migrate = is_memory_migrate(cs); |
4225399a PJ |
1018 | for (i = 0; i < n; i++) { |
1019 | struct mm_struct *mm = mmarray[i]; | |
1020 | ||
1021 | mpol_rebind_mm(mm, &cs->mems_allowed); | |
e4e364e8 PJ |
1022 | if (migrate) |
1023 | cpuset_migrate_mm(mm, &oldmem, &cs->mems_allowed); | |
4225399a PJ |
1024 | mmput(mm); |
1025 | } | |
1026 | ||
2df167a3 | 1027 | /* We're done rebinding vmas to this cpuset's new mems_allowed. */ |
4225399a | 1028 | kfree(mmarray); |
8793d854 | 1029 | cpuset_being_rebound = NULL; |
4225399a | 1030 | retval = 0; |
59dac16f | 1031 | done: |
1da177e4 LT |
1032 | return retval; |
1033 | } | |
1034 | ||
8793d854 PM |
1035 | int current_cpuset_is_being_rebound(void) |
1036 | { | |
1037 | return task_cs(current) == cpuset_being_rebound; | |
1038 | } | |
1039 | ||
5be7a479 | 1040 | static int update_relax_domain_level(struct cpuset *cs, s64 val) |
1d3504fc | 1041 | { |
30e0e178 LZ |
1042 | if (val < -1 || val >= SD_LV_MAX) |
1043 | return -EINVAL; | |
1d3504fc HS |
1044 | |
1045 | if (val != cs->relax_domain_level) { | |
1046 | cs->relax_domain_level = val; | |
1047 | rebuild_sched_domains(); | |
1048 | } | |
1049 | ||
1050 | return 0; | |
1051 | } | |
1052 | ||
1da177e4 LT |
1053 | /* |
1054 | * update_flag - read a 0 or a 1 in a file and update associated flag | |
78608366 PM |
1055 | * bit: the bit to update (see cpuset_flagbits_t) |
1056 | * cs: the cpuset to update | |
1057 | * turning_on: whether the flag is being set or cleared | |
053199ed | 1058 | * |
2df167a3 | 1059 | * Call with cgroup_mutex held. |
1da177e4 LT |
1060 | */ |
1061 | ||
700fe1ab PM |
1062 | static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs, |
1063 | int turning_on) | |
1da177e4 | 1064 | { |
1da177e4 | 1065 | struct cpuset trialcs; |
607717a6 | 1066 | int err; |
029190c5 | 1067 | int cpus_nonempty, balance_flag_changed; |
1da177e4 | 1068 | |
1da177e4 LT |
1069 | trialcs = *cs; |
1070 | if (turning_on) | |
1071 | set_bit(bit, &trialcs.flags); | |
1072 | else | |
1073 | clear_bit(bit, &trialcs.flags); | |
1074 | ||
1075 | err = validate_change(cs, &trialcs); | |
85d7b949 DG |
1076 | if (err < 0) |
1077 | return err; | |
029190c5 PJ |
1078 | |
1079 | cpus_nonempty = !cpus_empty(trialcs.cpus_allowed); | |
1080 | balance_flag_changed = (is_sched_load_balance(cs) != | |
1081 | is_sched_load_balance(&trialcs)); | |
1082 | ||
3d3f26a7 | 1083 | mutex_lock(&callback_mutex); |
69604067 | 1084 | cs->flags = trialcs.flags; |
3d3f26a7 | 1085 | mutex_unlock(&callback_mutex); |
85d7b949 | 1086 | |
029190c5 PJ |
1087 | if (cpus_nonempty && balance_flag_changed) |
1088 | rebuild_sched_domains(); | |
1089 | ||
85d7b949 | 1090 | return 0; |
1da177e4 LT |
1091 | } |
1092 | ||
3e0d98b9 | 1093 | /* |
80f7228b | 1094 | * Frequency meter - How fast is some event occurring? |
3e0d98b9 PJ |
1095 | * |
1096 | * These routines manage a digitally filtered, constant time based, | |
1097 | * event frequency meter. There are four routines: | |
1098 | * fmeter_init() - initialize a frequency meter. | |
1099 | * fmeter_markevent() - called each time the event happens. | |
1100 | * fmeter_getrate() - returns the recent rate of such events. | |
1101 | * fmeter_update() - internal routine used to update fmeter. | |
1102 | * | |
1103 | * A common data structure is passed to each of these routines, | |
1104 | * which is used to keep track of the state required to manage the | |
1105 | * frequency meter and its digital filter. | |
1106 | * | |
1107 | * The filter works on the number of events marked per unit time. | |
1108 | * The filter is single-pole low-pass recursive (IIR). The time unit | |
1109 | * is 1 second. Arithmetic is done using 32-bit integers scaled to | |
1110 | * simulate 3 decimal digits of precision (multiplied by 1000). | |
1111 | * | |
1112 | * With an FM_COEF of 933, and a time base of 1 second, the filter | |
1113 | * has a half-life of 10 seconds, meaning that if the events quit | |
1114 | * happening, then the rate returned from the fmeter_getrate() | |
1115 | * will be cut in half each 10 seconds, until it converges to zero. | |
1116 | * | |
1117 | * It is not worth doing a real infinitely recursive filter. If more | |
1118 | * than FM_MAXTICKS ticks have elapsed since the last filter event, | |
1119 | * just compute FM_MAXTICKS ticks worth, by which point the level | |
1120 | * will be stable. | |
1121 | * | |
1122 | * Limit the count of unprocessed events to FM_MAXCNT, so as to avoid | |
1123 | * arithmetic overflow in the fmeter_update() routine. | |
1124 | * | |
1125 | * Given the simple 32 bit integer arithmetic used, this meter works | |
1126 | * best for reporting rates between one per millisecond (msec) and | |
1127 | * one per 32 (approx) seconds. At constant rates faster than one | |
1128 | * per msec it maxes out at values just under 1,000,000. At constant | |
1129 | * rates between one per msec, and one per second it will stabilize | |
1130 | * to a value N*1000, where N is the rate of events per second. | |
1131 | * At constant rates between one per second and one per 32 seconds, | |
1132 | * it will be choppy, moving up on the seconds that have an event, | |
1133 | * and then decaying until the next event. At rates slower than | |
1134 | * about one in 32 seconds, it decays all the way back to zero between | |
1135 | * each event. | |
1136 | */ | |
1137 | ||
1138 | #define FM_COEF 933 /* coefficient for half-life of 10 secs */ | |
1139 | #define FM_MAXTICKS ((time_t)99) /* useless computing more ticks than this */ | |
1140 | #define FM_MAXCNT 1000000 /* limit cnt to avoid overflow */ | |
1141 | #define FM_SCALE 1000 /* faux fixed point scale */ | |
1142 | ||
1143 | /* Initialize a frequency meter */ | |
1144 | static void fmeter_init(struct fmeter *fmp) | |
1145 | { | |
1146 | fmp->cnt = 0; | |
1147 | fmp->val = 0; | |
1148 | fmp->time = 0; | |
1149 | spin_lock_init(&fmp->lock); | |
1150 | } | |
1151 | ||
1152 | /* Internal meter update - process cnt events and update value */ | |
1153 | static void fmeter_update(struct fmeter *fmp) | |
1154 | { | |
1155 | time_t now = get_seconds(); | |
1156 | time_t ticks = now - fmp->time; | |
1157 | ||
1158 | if (ticks == 0) | |
1159 | return; | |
1160 | ||
1161 | ticks = min(FM_MAXTICKS, ticks); | |
1162 | while (ticks-- > 0) | |
1163 | fmp->val = (FM_COEF * fmp->val) / FM_SCALE; | |
1164 | fmp->time = now; | |
1165 | ||
1166 | fmp->val += ((FM_SCALE - FM_COEF) * fmp->cnt) / FM_SCALE; | |
1167 | fmp->cnt = 0; | |
1168 | } | |
1169 | ||
1170 | /* Process any previous ticks, then bump cnt by one (times scale). */ | |
1171 | static void fmeter_markevent(struct fmeter *fmp) | |
1172 | { | |
1173 | spin_lock(&fmp->lock); | |
1174 | fmeter_update(fmp); | |
1175 | fmp->cnt = min(FM_MAXCNT, fmp->cnt + FM_SCALE); | |
1176 | spin_unlock(&fmp->lock); | |
1177 | } | |
1178 | ||
1179 | /* Process any previous ticks, then return current value. */ | |
1180 | static int fmeter_getrate(struct fmeter *fmp) | |
1181 | { | |
1182 | int val; | |
1183 | ||
1184 | spin_lock(&fmp->lock); | |
1185 | fmeter_update(fmp); | |
1186 | val = fmp->val; | |
1187 | spin_unlock(&fmp->lock); | |
1188 | return val; | |
1189 | } | |
1190 | ||
2df167a3 | 1191 | /* Called by cgroups to determine if a cpuset is usable; cgroup_mutex held */ |
8793d854 PM |
1192 | static int cpuset_can_attach(struct cgroup_subsys *ss, |
1193 | struct cgroup *cont, struct task_struct *tsk) | |
1da177e4 | 1194 | { |
8793d854 | 1195 | struct cpuset *cs = cgroup_cs(cont); |
1da177e4 | 1196 | |
1da177e4 LT |
1197 | if (cpus_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed)) |
1198 | return -ENOSPC; | |
9985b0ba DR |
1199 | if (tsk->flags & PF_THREAD_BOUND) { |
1200 | cpumask_t mask; | |
1201 | ||
1202 | mutex_lock(&callback_mutex); | |
1203 | mask = cs->cpus_allowed; | |
1204 | mutex_unlock(&callback_mutex); | |
1205 | if (!cpus_equal(tsk->cpus_allowed, mask)) | |
1206 | return -EINVAL; | |
1207 | } | |
1da177e4 | 1208 | |
8793d854 PM |
1209 | return security_task_setscheduler(tsk, 0, NULL); |
1210 | } | |
1da177e4 | 1211 | |
8793d854 PM |
1212 | static void cpuset_attach(struct cgroup_subsys *ss, |
1213 | struct cgroup *cont, struct cgroup *oldcont, | |
1214 | struct task_struct *tsk) | |
1215 | { | |
1216 | cpumask_t cpus; | |
1217 | nodemask_t from, to; | |
1218 | struct mm_struct *mm; | |
1219 | struct cpuset *cs = cgroup_cs(cont); | |
1220 | struct cpuset *oldcs = cgroup_cs(oldcont); | |
9985b0ba | 1221 | int err; |
22fb52dd | 1222 | |
3d3f26a7 | 1223 | mutex_lock(&callback_mutex); |
1da177e4 | 1224 | guarantee_online_cpus(cs, &cpus); |
9985b0ba | 1225 | err = set_cpus_allowed_ptr(tsk, &cpus); |
8793d854 | 1226 | mutex_unlock(&callback_mutex); |
9985b0ba DR |
1227 | if (err) |
1228 | return; | |
1da177e4 | 1229 | |
45b07ef3 PJ |
1230 | from = oldcs->mems_allowed; |
1231 | to = cs->mems_allowed; | |
4225399a PJ |
1232 | mm = get_task_mm(tsk); |
1233 | if (mm) { | |
1234 | mpol_rebind_mm(mm, &to); | |
2741a559 | 1235 | if (is_memory_migrate(cs)) |
e4e364e8 | 1236 | cpuset_migrate_mm(mm, &from, &to); |
4225399a PJ |
1237 | mmput(mm); |
1238 | } | |
1239 | ||
1da177e4 LT |
1240 | } |
1241 | ||
1242 | /* The various types of files and directories in a cpuset file system */ | |
1243 | ||
1244 | typedef enum { | |
45b07ef3 | 1245 | FILE_MEMORY_MIGRATE, |
1da177e4 LT |
1246 | FILE_CPULIST, |
1247 | FILE_MEMLIST, | |
1248 | FILE_CPU_EXCLUSIVE, | |
1249 | FILE_MEM_EXCLUSIVE, | |
78608366 | 1250 | FILE_MEM_HARDWALL, |
029190c5 | 1251 | FILE_SCHED_LOAD_BALANCE, |
1d3504fc | 1252 | FILE_SCHED_RELAX_DOMAIN_LEVEL, |
3e0d98b9 PJ |
1253 | FILE_MEMORY_PRESSURE_ENABLED, |
1254 | FILE_MEMORY_PRESSURE, | |
825a46af PJ |
1255 | FILE_SPREAD_PAGE, |
1256 | FILE_SPREAD_SLAB, | |
1da177e4 LT |
1257 | } cpuset_filetype_t; |
1258 | ||
8793d854 PM |
1259 | static ssize_t cpuset_common_file_write(struct cgroup *cont, |
1260 | struct cftype *cft, | |
1261 | struct file *file, | |
d3ed11c3 | 1262 | const char __user *userbuf, |
1da177e4 LT |
1263 | size_t nbytes, loff_t *unused_ppos) |
1264 | { | |
8793d854 | 1265 | struct cpuset *cs = cgroup_cs(cont); |
1da177e4 LT |
1266 | cpuset_filetype_t type = cft->private; |
1267 | char *buffer; | |
1268 | int retval = 0; | |
1269 | ||
1270 | /* Crude upper limit on largest legitimate cpulist user might write. */ | |
029190c5 | 1271 | if (nbytes > 100U + 6 * max(NR_CPUS, MAX_NUMNODES)) |
1da177e4 LT |
1272 | return -E2BIG; |
1273 | ||
1274 | /* +1 for nul-terminator */ | |
b331d259 HH |
1275 | buffer = kmalloc(nbytes + 1, GFP_KERNEL); |
1276 | if (!buffer) | |
1da177e4 LT |
1277 | return -ENOMEM; |
1278 | ||
1279 | if (copy_from_user(buffer, userbuf, nbytes)) { | |
1280 | retval = -EFAULT; | |
1281 | goto out1; | |
1282 | } | |
1283 | buffer[nbytes] = 0; /* nul-terminate */ | |
1284 | ||
8793d854 | 1285 | cgroup_lock(); |
1da177e4 | 1286 | |
8793d854 | 1287 | if (cgroup_is_removed(cont)) { |
1da177e4 LT |
1288 | retval = -ENODEV; |
1289 | goto out2; | |
1290 | } | |
1291 | ||
1292 | switch (type) { | |
1293 | case FILE_CPULIST: | |
1294 | retval = update_cpumask(cs, buffer); | |
1295 | break; | |
1296 | case FILE_MEMLIST: | |
1297 | retval = update_nodemask(cs, buffer); | |
1298 | break; | |
700fe1ab PM |
1299 | default: |
1300 | retval = -EINVAL; | |
1301 | goto out2; | |
1302 | } | |
1303 | ||
1304 | if (retval == 0) | |
1305 | retval = nbytes; | |
1306 | out2: | |
1307 | cgroup_unlock(); | |
1308 | out1: | |
1309 | kfree(buffer); | |
1310 | return retval; | |
1311 | } | |
1312 | ||
1313 | static int cpuset_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val) | |
1314 | { | |
1315 | int retval = 0; | |
1316 | struct cpuset *cs = cgroup_cs(cgrp); | |
1317 | cpuset_filetype_t type = cft->private; | |
1318 | ||
1319 | cgroup_lock(); | |
1320 | ||
1321 | if (cgroup_is_removed(cgrp)) { | |
1322 | cgroup_unlock(); | |
1323 | return -ENODEV; | |
1324 | } | |
1325 | ||
1326 | switch (type) { | |
1da177e4 | 1327 | case FILE_CPU_EXCLUSIVE: |
700fe1ab | 1328 | retval = update_flag(CS_CPU_EXCLUSIVE, cs, val); |
1da177e4 LT |
1329 | break; |
1330 | case FILE_MEM_EXCLUSIVE: | |
700fe1ab | 1331 | retval = update_flag(CS_MEM_EXCLUSIVE, cs, val); |
1da177e4 | 1332 | break; |
78608366 PM |
1333 | case FILE_MEM_HARDWALL: |
1334 | retval = update_flag(CS_MEM_HARDWALL, cs, val); | |
1335 | break; | |
029190c5 | 1336 | case FILE_SCHED_LOAD_BALANCE: |
700fe1ab | 1337 | retval = update_flag(CS_SCHED_LOAD_BALANCE, cs, val); |
1d3504fc | 1338 | break; |
45b07ef3 | 1339 | case FILE_MEMORY_MIGRATE: |
700fe1ab | 1340 | retval = update_flag(CS_MEMORY_MIGRATE, cs, val); |
45b07ef3 | 1341 | break; |
3e0d98b9 | 1342 | case FILE_MEMORY_PRESSURE_ENABLED: |
700fe1ab | 1343 | cpuset_memory_pressure_enabled = !!val; |
3e0d98b9 PJ |
1344 | break; |
1345 | case FILE_MEMORY_PRESSURE: | |
1346 | retval = -EACCES; | |
1347 | break; | |
825a46af | 1348 | case FILE_SPREAD_PAGE: |
700fe1ab | 1349 | retval = update_flag(CS_SPREAD_PAGE, cs, val); |
151a4420 | 1350 | cs->mems_generation = cpuset_mems_generation++; |
825a46af PJ |
1351 | break; |
1352 | case FILE_SPREAD_SLAB: | |
700fe1ab | 1353 | retval = update_flag(CS_SPREAD_SLAB, cs, val); |
151a4420 | 1354 | cs->mems_generation = cpuset_mems_generation++; |
825a46af | 1355 | break; |
1da177e4 LT |
1356 | default: |
1357 | retval = -EINVAL; | |
700fe1ab | 1358 | break; |
1da177e4 | 1359 | } |
8793d854 | 1360 | cgroup_unlock(); |
1da177e4 LT |
1361 | return retval; |
1362 | } | |
1363 | ||
5be7a479 PM |
1364 | static int cpuset_write_s64(struct cgroup *cgrp, struct cftype *cft, s64 val) |
1365 | { | |
1366 | int retval = 0; | |
1367 | struct cpuset *cs = cgroup_cs(cgrp); | |
1368 | cpuset_filetype_t type = cft->private; | |
1369 | ||
1370 | cgroup_lock(); | |
1371 | ||
1372 | if (cgroup_is_removed(cgrp)) { | |
1373 | cgroup_unlock(); | |
1374 | return -ENODEV; | |
1375 | } | |
1376 | switch (type) { | |
1377 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1378 | retval = update_relax_domain_level(cs, val); | |
1379 | break; | |
1380 | default: | |
1381 | retval = -EINVAL; | |
1382 | break; | |
1383 | } | |
1384 | cgroup_unlock(); | |
1385 | return retval; | |
1386 | } | |
1387 | ||
1da177e4 LT |
1388 | /* |
1389 | * These ascii lists should be read in a single call, by using a user | |
1390 | * buffer large enough to hold the entire map. If read in smaller | |
1391 | * chunks, there is no guarantee of atomicity. Since the display format | |
1392 | * used, list of ranges of sequential numbers, is variable length, | |
1393 | * and since these maps can change value dynamically, one could read | |
1394 | * gibberish by doing partial reads while a list was changing. | |
1395 | * A single large read to a buffer that crosses a page boundary is | |
1396 | * ok, because the result being copied to user land is not recomputed | |
1397 | * across a page fault. | |
1398 | */ | |
1399 | ||
1400 | static int cpuset_sprintf_cpulist(char *page, struct cpuset *cs) | |
1401 | { | |
1402 | cpumask_t mask; | |
1403 | ||
3d3f26a7 | 1404 | mutex_lock(&callback_mutex); |
1da177e4 | 1405 | mask = cs->cpus_allowed; |
3d3f26a7 | 1406 | mutex_unlock(&callback_mutex); |
1da177e4 LT |
1407 | |
1408 | return cpulist_scnprintf(page, PAGE_SIZE, mask); | |
1409 | } | |
1410 | ||
1411 | static int cpuset_sprintf_memlist(char *page, struct cpuset *cs) | |
1412 | { | |
1413 | nodemask_t mask; | |
1414 | ||
3d3f26a7 | 1415 | mutex_lock(&callback_mutex); |
1da177e4 | 1416 | mask = cs->mems_allowed; |
3d3f26a7 | 1417 | mutex_unlock(&callback_mutex); |
1da177e4 LT |
1418 | |
1419 | return nodelist_scnprintf(page, PAGE_SIZE, mask); | |
1420 | } | |
1421 | ||
8793d854 PM |
1422 | static ssize_t cpuset_common_file_read(struct cgroup *cont, |
1423 | struct cftype *cft, | |
1424 | struct file *file, | |
1425 | char __user *buf, | |
1426 | size_t nbytes, loff_t *ppos) | |
1da177e4 | 1427 | { |
8793d854 | 1428 | struct cpuset *cs = cgroup_cs(cont); |
1da177e4 LT |
1429 | cpuset_filetype_t type = cft->private; |
1430 | char *page; | |
1431 | ssize_t retval = 0; | |
1432 | char *s; | |
1da177e4 | 1433 | |
e12ba74d | 1434 | if (!(page = (char *)__get_free_page(GFP_TEMPORARY))) |
1da177e4 LT |
1435 | return -ENOMEM; |
1436 | ||
1437 | s = page; | |
1438 | ||
1439 | switch (type) { | |
1440 | case FILE_CPULIST: | |
1441 | s += cpuset_sprintf_cpulist(s, cs); | |
1442 | break; | |
1443 | case FILE_MEMLIST: | |
1444 | s += cpuset_sprintf_memlist(s, cs); | |
1445 | break; | |
1da177e4 LT |
1446 | default: |
1447 | retval = -EINVAL; | |
1448 | goto out; | |
1449 | } | |
1450 | *s++ = '\n'; | |
1da177e4 | 1451 | |
eacaa1f5 | 1452 | retval = simple_read_from_buffer(buf, nbytes, ppos, page, s - page); |
1da177e4 LT |
1453 | out: |
1454 | free_page((unsigned long)page); | |
1455 | return retval; | |
1456 | } | |
1457 | ||
700fe1ab PM |
1458 | static u64 cpuset_read_u64(struct cgroup *cont, struct cftype *cft) |
1459 | { | |
1460 | struct cpuset *cs = cgroup_cs(cont); | |
1461 | cpuset_filetype_t type = cft->private; | |
1462 | switch (type) { | |
1463 | case FILE_CPU_EXCLUSIVE: | |
1464 | return is_cpu_exclusive(cs); | |
1465 | case FILE_MEM_EXCLUSIVE: | |
1466 | return is_mem_exclusive(cs); | |
78608366 PM |
1467 | case FILE_MEM_HARDWALL: |
1468 | return is_mem_hardwall(cs); | |
700fe1ab PM |
1469 | case FILE_SCHED_LOAD_BALANCE: |
1470 | return is_sched_load_balance(cs); | |
1471 | case FILE_MEMORY_MIGRATE: | |
1472 | return is_memory_migrate(cs); | |
1473 | case FILE_MEMORY_PRESSURE_ENABLED: | |
1474 | return cpuset_memory_pressure_enabled; | |
1475 | case FILE_MEMORY_PRESSURE: | |
1476 | return fmeter_getrate(&cs->fmeter); | |
1477 | case FILE_SPREAD_PAGE: | |
1478 | return is_spread_page(cs); | |
1479 | case FILE_SPREAD_SLAB: | |
1480 | return is_spread_slab(cs); | |
1481 | default: | |
1482 | BUG(); | |
1483 | } | |
1484 | } | |
1da177e4 | 1485 | |
5be7a479 PM |
1486 | static s64 cpuset_read_s64(struct cgroup *cont, struct cftype *cft) |
1487 | { | |
1488 | struct cpuset *cs = cgroup_cs(cont); | |
1489 | cpuset_filetype_t type = cft->private; | |
1490 | switch (type) { | |
1491 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1492 | return cs->relax_domain_level; | |
1493 | default: | |
1494 | BUG(); | |
1495 | } | |
1496 | } | |
1497 | ||
1da177e4 LT |
1498 | |
1499 | /* | |
1500 | * for the common functions, 'private' gives the type of file | |
1501 | */ | |
1502 | ||
addf2c73 PM |
1503 | static struct cftype files[] = { |
1504 | { | |
1505 | .name = "cpus", | |
1506 | .read = cpuset_common_file_read, | |
1507 | .write = cpuset_common_file_write, | |
1508 | .private = FILE_CPULIST, | |
1509 | }, | |
1510 | ||
1511 | { | |
1512 | .name = "mems", | |
1513 | .read = cpuset_common_file_read, | |
1514 | .write = cpuset_common_file_write, | |
1515 | .private = FILE_MEMLIST, | |
1516 | }, | |
1517 | ||
1518 | { | |
1519 | .name = "cpu_exclusive", | |
1520 | .read_u64 = cpuset_read_u64, | |
1521 | .write_u64 = cpuset_write_u64, | |
1522 | .private = FILE_CPU_EXCLUSIVE, | |
1523 | }, | |
1524 | ||
1525 | { | |
1526 | .name = "mem_exclusive", | |
1527 | .read_u64 = cpuset_read_u64, | |
1528 | .write_u64 = cpuset_write_u64, | |
1529 | .private = FILE_MEM_EXCLUSIVE, | |
1530 | }, | |
1531 | ||
78608366 PM |
1532 | { |
1533 | .name = "mem_hardwall", | |
1534 | .read_u64 = cpuset_read_u64, | |
1535 | .write_u64 = cpuset_write_u64, | |
1536 | .private = FILE_MEM_HARDWALL, | |
1537 | }, | |
1538 | ||
addf2c73 PM |
1539 | { |
1540 | .name = "sched_load_balance", | |
1541 | .read_u64 = cpuset_read_u64, | |
1542 | .write_u64 = cpuset_write_u64, | |
1543 | .private = FILE_SCHED_LOAD_BALANCE, | |
1544 | }, | |
1545 | ||
1546 | { | |
1547 | .name = "sched_relax_domain_level", | |
5be7a479 PM |
1548 | .read_s64 = cpuset_read_s64, |
1549 | .write_s64 = cpuset_write_s64, | |
addf2c73 PM |
1550 | .private = FILE_SCHED_RELAX_DOMAIN_LEVEL, |
1551 | }, | |
1552 | ||
1553 | { | |
1554 | .name = "memory_migrate", | |
1555 | .read_u64 = cpuset_read_u64, | |
1556 | .write_u64 = cpuset_write_u64, | |
1557 | .private = FILE_MEMORY_MIGRATE, | |
1558 | }, | |
1559 | ||
1560 | { | |
1561 | .name = "memory_pressure", | |
1562 | .read_u64 = cpuset_read_u64, | |
1563 | .write_u64 = cpuset_write_u64, | |
1564 | .private = FILE_MEMORY_PRESSURE, | |
1565 | }, | |
1566 | ||
1567 | { | |
1568 | .name = "memory_spread_page", | |
1569 | .read_u64 = cpuset_read_u64, | |
1570 | .write_u64 = cpuset_write_u64, | |
1571 | .private = FILE_SPREAD_PAGE, | |
1572 | }, | |
1573 | ||
1574 | { | |
1575 | .name = "memory_spread_slab", | |
1576 | .read_u64 = cpuset_read_u64, | |
1577 | .write_u64 = cpuset_write_u64, | |
1578 | .private = FILE_SPREAD_SLAB, | |
1579 | }, | |
45b07ef3 PJ |
1580 | }; |
1581 | ||
3e0d98b9 PJ |
1582 | static struct cftype cft_memory_pressure_enabled = { |
1583 | .name = "memory_pressure_enabled", | |
700fe1ab PM |
1584 | .read_u64 = cpuset_read_u64, |
1585 | .write_u64 = cpuset_write_u64, | |
3e0d98b9 PJ |
1586 | .private = FILE_MEMORY_PRESSURE_ENABLED, |
1587 | }; | |
1588 | ||
8793d854 | 1589 | static int cpuset_populate(struct cgroup_subsys *ss, struct cgroup *cont) |
1da177e4 LT |
1590 | { |
1591 | int err; | |
1592 | ||
addf2c73 PM |
1593 | err = cgroup_add_files(cont, ss, files, ARRAY_SIZE(files)); |
1594 | if (err) | |
1da177e4 | 1595 | return err; |
8793d854 | 1596 | /* memory_pressure_enabled is in root cpuset only */ |
addf2c73 | 1597 | if (!cont->parent) |
8793d854 | 1598 | err = cgroup_add_file(cont, ss, |
addf2c73 PM |
1599 | &cft_memory_pressure_enabled); |
1600 | return err; | |
1da177e4 LT |
1601 | } |
1602 | ||
8793d854 PM |
1603 | /* |
1604 | * post_clone() is called at the end of cgroup_clone(). | |
1605 | * 'cgroup' was just created automatically as a result of | |
1606 | * a cgroup_clone(), and the current task is about to | |
1607 | * be moved into 'cgroup'. | |
1608 | * | |
1609 | * Currently we refuse to set up the cgroup - thereby | |
1610 | * refusing the task to be entered, and as a result refusing | |
1611 | * the sys_unshare() or clone() which initiated it - if any | |
1612 | * sibling cpusets have exclusive cpus or mem. | |
1613 | * | |
1614 | * If this becomes a problem for some users who wish to | |
1615 | * allow that scenario, then cpuset_post_clone() could be | |
1616 | * changed to grant parent->cpus_allowed-sibling_cpus_exclusive | |
2df167a3 PM |
1617 | * (and likewise for mems) to the new cgroup. Called with cgroup_mutex |
1618 | * held. | |
8793d854 PM |
1619 | */ |
1620 | static void cpuset_post_clone(struct cgroup_subsys *ss, | |
1621 | struct cgroup *cgroup) | |
1622 | { | |
1623 | struct cgroup *parent, *child; | |
1624 | struct cpuset *cs, *parent_cs; | |
1625 | ||
1626 | parent = cgroup->parent; | |
1627 | list_for_each_entry(child, &parent->children, sibling) { | |
1628 | cs = cgroup_cs(child); | |
1629 | if (is_mem_exclusive(cs) || is_cpu_exclusive(cs)) | |
1630 | return; | |
1631 | } | |
1632 | cs = cgroup_cs(cgroup); | |
1633 | parent_cs = cgroup_cs(parent); | |
1634 | ||
1635 | cs->mems_allowed = parent_cs->mems_allowed; | |
1636 | cs->cpus_allowed = parent_cs->cpus_allowed; | |
1637 | return; | |
1638 | } | |
1639 | ||
1da177e4 LT |
1640 | /* |
1641 | * cpuset_create - create a cpuset | |
2df167a3 PM |
1642 | * ss: cpuset cgroup subsystem |
1643 | * cont: control group that the new cpuset will be part of | |
1da177e4 LT |
1644 | */ |
1645 | ||
8793d854 PM |
1646 | static struct cgroup_subsys_state *cpuset_create( |
1647 | struct cgroup_subsys *ss, | |
1648 | struct cgroup *cont) | |
1da177e4 LT |
1649 | { |
1650 | struct cpuset *cs; | |
8793d854 | 1651 | struct cpuset *parent; |
1da177e4 | 1652 | |
8793d854 PM |
1653 | if (!cont->parent) { |
1654 | /* This is early initialization for the top cgroup */ | |
1655 | top_cpuset.mems_generation = cpuset_mems_generation++; | |
1656 | return &top_cpuset.css; | |
1657 | } | |
1658 | parent = cgroup_cs(cont->parent); | |
1da177e4 LT |
1659 | cs = kmalloc(sizeof(*cs), GFP_KERNEL); |
1660 | if (!cs) | |
8793d854 | 1661 | return ERR_PTR(-ENOMEM); |
1da177e4 | 1662 | |
cf2a473c | 1663 | cpuset_update_task_memory_state(); |
1da177e4 | 1664 | cs->flags = 0; |
825a46af PJ |
1665 | if (is_spread_page(parent)) |
1666 | set_bit(CS_SPREAD_PAGE, &cs->flags); | |
1667 | if (is_spread_slab(parent)) | |
1668 | set_bit(CS_SPREAD_SLAB, &cs->flags); | |
029190c5 | 1669 | set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); |
f9a86fcb MT |
1670 | cpus_clear(cs->cpus_allowed); |
1671 | nodes_clear(cs->mems_allowed); | |
151a4420 | 1672 | cs->mems_generation = cpuset_mems_generation++; |
3e0d98b9 | 1673 | fmeter_init(&cs->fmeter); |
1d3504fc | 1674 | cs->relax_domain_level = -1; |
1da177e4 LT |
1675 | |
1676 | cs->parent = parent; | |
202f72d5 | 1677 | number_of_cpusets++; |
8793d854 | 1678 | return &cs->css ; |
1da177e4 LT |
1679 | } |
1680 | ||
029190c5 PJ |
1681 | /* |
1682 | * Locking note on the strange update_flag() call below: | |
1683 | * | |
1684 | * If the cpuset being removed has its flag 'sched_load_balance' | |
1685 | * enabled, then simulate turning sched_load_balance off, which | |
86ef5c9a | 1686 | * will call rebuild_sched_domains(). The get_online_cpus() |
029190c5 PJ |
1687 | * call in rebuild_sched_domains() must not be made while holding |
1688 | * callback_mutex. Elsewhere the kernel nests callback_mutex inside | |
86ef5c9a | 1689 | * get_online_cpus() calls. So the reverse nesting would risk an |
029190c5 PJ |
1690 | * ABBA deadlock. |
1691 | */ | |
1692 | ||
8793d854 | 1693 | static void cpuset_destroy(struct cgroup_subsys *ss, struct cgroup *cont) |
1da177e4 | 1694 | { |
8793d854 | 1695 | struct cpuset *cs = cgroup_cs(cont); |
1da177e4 | 1696 | |
cf2a473c | 1697 | cpuset_update_task_memory_state(); |
029190c5 PJ |
1698 | |
1699 | if (is_sched_load_balance(cs)) | |
700fe1ab | 1700 | update_flag(CS_SCHED_LOAD_BALANCE, cs, 0); |
029190c5 | 1701 | |
202f72d5 | 1702 | number_of_cpusets--; |
8793d854 | 1703 | kfree(cs); |
1da177e4 LT |
1704 | } |
1705 | ||
8793d854 PM |
1706 | struct cgroup_subsys cpuset_subsys = { |
1707 | .name = "cpuset", | |
1708 | .create = cpuset_create, | |
1709 | .destroy = cpuset_destroy, | |
1710 | .can_attach = cpuset_can_attach, | |
1711 | .attach = cpuset_attach, | |
1712 | .populate = cpuset_populate, | |
1713 | .post_clone = cpuset_post_clone, | |
1714 | .subsys_id = cpuset_subsys_id, | |
1715 | .early_init = 1, | |
1716 | }; | |
1717 | ||
c417f024 PJ |
1718 | /* |
1719 | * cpuset_init_early - just enough so that the calls to | |
1720 | * cpuset_update_task_memory_state() in early init code | |
1721 | * are harmless. | |
1722 | */ | |
1723 | ||
1724 | int __init cpuset_init_early(void) | |
1725 | { | |
8793d854 | 1726 | top_cpuset.mems_generation = cpuset_mems_generation++; |
c417f024 PJ |
1727 | return 0; |
1728 | } | |
1729 | ||
8793d854 | 1730 | |
1da177e4 LT |
1731 | /** |
1732 | * cpuset_init - initialize cpusets at system boot | |
1733 | * | |
1734 | * Description: Initialize top_cpuset and the cpuset internal file system, | |
1735 | **/ | |
1736 | ||
1737 | int __init cpuset_init(void) | |
1738 | { | |
8793d854 | 1739 | int err = 0; |
1da177e4 | 1740 | |
f9a86fcb MT |
1741 | cpus_setall(top_cpuset.cpus_allowed); |
1742 | nodes_setall(top_cpuset.mems_allowed); | |
1da177e4 | 1743 | |
3e0d98b9 | 1744 | fmeter_init(&top_cpuset.fmeter); |
151a4420 | 1745 | top_cpuset.mems_generation = cpuset_mems_generation++; |
029190c5 | 1746 | set_bit(CS_SCHED_LOAD_BALANCE, &top_cpuset.flags); |
1d3504fc | 1747 | top_cpuset.relax_domain_level = -1; |
1da177e4 | 1748 | |
1da177e4 LT |
1749 | err = register_filesystem(&cpuset_fs_type); |
1750 | if (err < 0) | |
8793d854 PM |
1751 | return err; |
1752 | ||
202f72d5 | 1753 | number_of_cpusets = 1; |
8793d854 | 1754 | return 0; |
1da177e4 LT |
1755 | } |
1756 | ||
956db3ca CW |
1757 | /** |
1758 | * cpuset_do_move_task - move a given task to another cpuset | |
1759 | * @tsk: pointer to task_struct the task to move | |
1760 | * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner | |
1761 | * | |
1762 | * Called by cgroup_scan_tasks() for each task in a cgroup. | |
1763 | * Return nonzero to stop the walk through the tasks. | |
1764 | */ | |
9e0c914c AB |
1765 | static void cpuset_do_move_task(struct task_struct *tsk, |
1766 | struct cgroup_scanner *scan) | |
956db3ca CW |
1767 | { |
1768 | struct cpuset_hotplug_scanner *chsp; | |
1769 | ||
1770 | chsp = container_of(scan, struct cpuset_hotplug_scanner, scan); | |
1771 | cgroup_attach_task(chsp->to, tsk); | |
1772 | } | |
1773 | ||
1774 | /** | |
1775 | * move_member_tasks_to_cpuset - move tasks from one cpuset to another | |
1776 | * @from: cpuset in which the tasks currently reside | |
1777 | * @to: cpuset to which the tasks will be moved | |
1778 | * | |
c8d9c90c PJ |
1779 | * Called with cgroup_mutex held |
1780 | * callback_mutex must not be held, as cpuset_attach() will take it. | |
956db3ca CW |
1781 | * |
1782 | * The cgroup_scan_tasks() function will scan all the tasks in a cgroup, | |
1783 | * calling callback functions for each. | |
1784 | */ | |
1785 | static void move_member_tasks_to_cpuset(struct cpuset *from, struct cpuset *to) | |
1786 | { | |
1787 | struct cpuset_hotplug_scanner scan; | |
1788 | ||
1789 | scan.scan.cg = from->css.cgroup; | |
1790 | scan.scan.test_task = NULL; /* select all tasks in cgroup */ | |
1791 | scan.scan.process_task = cpuset_do_move_task; | |
1792 | scan.scan.heap = NULL; | |
1793 | scan.to = to->css.cgroup; | |
1794 | ||
1795 | if (cgroup_scan_tasks((struct cgroup_scanner *)&scan)) | |
1796 | printk(KERN_ERR "move_member_tasks_to_cpuset: " | |
1797 | "cgroup_scan_tasks failed\n"); | |
1798 | } | |
1799 | ||
b1aac8bb PJ |
1800 | /* |
1801 | * If common_cpu_mem_hotplug_unplug(), below, unplugs any CPUs | |
1802 | * or memory nodes, we need to walk over the cpuset hierarchy, | |
1803 | * removing that CPU or node from all cpusets. If this removes the | |
956db3ca CW |
1804 | * last CPU or node from a cpuset, then move the tasks in the empty |
1805 | * cpuset to its next-highest non-empty parent. | |
b1aac8bb | 1806 | * |
c8d9c90c PJ |
1807 | * Called with cgroup_mutex held |
1808 | * callback_mutex must not be held, as cpuset_attach() will take it. | |
b1aac8bb | 1809 | */ |
956db3ca CW |
1810 | static void remove_tasks_in_empty_cpuset(struct cpuset *cs) |
1811 | { | |
1812 | struct cpuset *parent; | |
1813 | ||
c8d9c90c PJ |
1814 | /* |
1815 | * The cgroup's css_sets list is in use if there are tasks | |
1816 | * in the cpuset; the list is empty if there are none; | |
1817 | * the cs->css.refcnt seems always 0. | |
1818 | */ | |
956db3ca CW |
1819 | if (list_empty(&cs->css.cgroup->css_sets)) |
1820 | return; | |
b1aac8bb | 1821 | |
956db3ca CW |
1822 | /* |
1823 | * Find its next-highest non-empty parent, (top cpuset | |
1824 | * has online cpus, so can't be empty). | |
1825 | */ | |
1826 | parent = cs->parent; | |
b4501295 PJ |
1827 | while (cpus_empty(parent->cpus_allowed) || |
1828 | nodes_empty(parent->mems_allowed)) | |
956db3ca | 1829 | parent = parent->parent; |
956db3ca CW |
1830 | |
1831 | move_member_tasks_to_cpuset(cs, parent); | |
1832 | } | |
1833 | ||
1834 | /* | |
1835 | * Walk the specified cpuset subtree and look for empty cpusets. | |
1836 | * The tasks of such cpuset must be moved to a parent cpuset. | |
1837 | * | |
2df167a3 | 1838 | * Called with cgroup_mutex held. We take callback_mutex to modify |
956db3ca CW |
1839 | * cpus_allowed and mems_allowed. |
1840 | * | |
1841 | * This walk processes the tree from top to bottom, completing one layer | |
1842 | * before dropping down to the next. It always processes a node before | |
1843 | * any of its children. | |
1844 | * | |
1845 | * For now, since we lack memory hot unplug, we'll never see a cpuset | |
1846 | * that has tasks along with an empty 'mems'. But if we did see such | |
1847 | * a cpuset, we'd handle it just like we do if its 'cpus' was empty. | |
1848 | */ | |
1849 | static void scan_for_empty_cpusets(const struct cpuset *root) | |
b1aac8bb | 1850 | { |
956db3ca CW |
1851 | struct cpuset *cp; /* scans cpusets being updated */ |
1852 | struct cpuset *child; /* scans child cpusets of cp */ | |
1853 | struct list_head queue; | |
8793d854 | 1854 | struct cgroup *cont; |
b1aac8bb | 1855 | |
956db3ca CW |
1856 | INIT_LIST_HEAD(&queue); |
1857 | ||
1858 | list_add_tail((struct list_head *)&root->stack_list, &queue); | |
1859 | ||
956db3ca CW |
1860 | while (!list_empty(&queue)) { |
1861 | cp = container_of(queue.next, struct cpuset, stack_list); | |
1862 | list_del(queue.next); | |
1863 | list_for_each_entry(cont, &cp->css.cgroup->children, sibling) { | |
1864 | child = cgroup_cs(cont); | |
1865 | list_add_tail(&child->stack_list, &queue); | |
1866 | } | |
1867 | cont = cp->css.cgroup; | |
b4501295 PJ |
1868 | |
1869 | /* Continue past cpusets with all cpus, mems online */ | |
1870 | if (cpus_subset(cp->cpus_allowed, cpu_online_map) && | |
1871 | nodes_subset(cp->mems_allowed, node_states[N_HIGH_MEMORY])) | |
1872 | continue; | |
1873 | ||
956db3ca | 1874 | /* Remove offline cpus and mems from this cpuset. */ |
b4501295 | 1875 | mutex_lock(&callback_mutex); |
956db3ca CW |
1876 | cpus_and(cp->cpus_allowed, cp->cpus_allowed, cpu_online_map); |
1877 | nodes_and(cp->mems_allowed, cp->mems_allowed, | |
1878 | node_states[N_HIGH_MEMORY]); | |
b4501295 PJ |
1879 | mutex_unlock(&callback_mutex); |
1880 | ||
1881 | /* Move tasks from the empty cpuset to a parent */ | |
c8d9c90c | 1882 | if (cpus_empty(cp->cpus_allowed) || |
b4501295 | 1883 | nodes_empty(cp->mems_allowed)) |
956db3ca | 1884 | remove_tasks_in_empty_cpuset(cp); |
b1aac8bb PJ |
1885 | } |
1886 | } | |
1887 | ||
1888 | /* | |
1889 | * The cpus_allowed and mems_allowed nodemasks in the top_cpuset track | |
0e1e7c7a | 1890 | * cpu_online_map and node_states[N_HIGH_MEMORY]. Force the top cpuset to |
956db3ca | 1891 | * track what's online after any CPU or memory node hotplug or unplug event. |
b1aac8bb PJ |
1892 | * |
1893 | * Since there are two callers of this routine, one for CPU hotplug | |
1894 | * events and one for memory node hotplug events, we could have coded | |
1895 | * two separate routines here. We code it as a single common routine | |
1896 | * in order to minimize text size. | |
1897 | */ | |
1898 | ||
3e84050c | 1899 | static void common_cpu_mem_hotplug_unplug(int rebuild_sd) |
b1aac8bb | 1900 | { |
8793d854 | 1901 | cgroup_lock(); |
b1aac8bb | 1902 | |
b1aac8bb | 1903 | top_cpuset.cpus_allowed = cpu_online_map; |
0e1e7c7a | 1904 | top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY]; |
956db3ca | 1905 | scan_for_empty_cpusets(&top_cpuset); |
b1aac8bb | 1906 | |
5c8e1ed1 MK |
1907 | /* |
1908 | * Scheduler destroys domains on hotplug events. | |
1909 | * Rebuild them based on the current settings. | |
1910 | */ | |
3e84050c DA |
1911 | if (rebuild_sd) |
1912 | rebuild_sched_domains(); | |
5c8e1ed1 | 1913 | |
8793d854 | 1914 | cgroup_unlock(); |
b1aac8bb | 1915 | } |
b1aac8bb | 1916 | |
4c4d50f7 PJ |
1917 | /* |
1918 | * The top_cpuset tracks what CPUs and Memory Nodes are online, | |
1919 | * period. This is necessary in order to make cpusets transparent | |
1920 | * (of no affect) on systems that are actively using CPU hotplug | |
1921 | * but making no active use of cpusets. | |
1922 | * | |
38837fc7 PJ |
1923 | * This routine ensures that top_cpuset.cpus_allowed tracks |
1924 | * cpu_online_map on each CPU hotplug (cpuhp) event. | |
4c4d50f7 PJ |
1925 | */ |
1926 | ||
029190c5 PJ |
1927 | static int cpuset_handle_cpuhp(struct notifier_block *unused_nb, |
1928 | unsigned long phase, void *unused_cpu) | |
4c4d50f7 | 1929 | { |
3e84050c DA |
1930 | switch (phase) { |
1931 | case CPU_UP_CANCELED: | |
1932 | case CPU_UP_CANCELED_FROZEN: | |
1933 | case CPU_DOWN_FAILED: | |
1934 | case CPU_DOWN_FAILED_FROZEN: | |
1935 | case CPU_ONLINE: | |
1936 | case CPU_ONLINE_FROZEN: | |
1937 | case CPU_DEAD: | |
1938 | case CPU_DEAD_FROZEN: | |
1939 | common_cpu_mem_hotplug_unplug(1); | |
1940 | break; | |
1941 | default: | |
ac076758 | 1942 | return NOTIFY_DONE; |
3e84050c | 1943 | } |
ac076758 | 1944 | |
3e84050c | 1945 | return NOTIFY_OK; |
4c4d50f7 | 1946 | } |
4c4d50f7 | 1947 | |
b1aac8bb | 1948 | #ifdef CONFIG_MEMORY_HOTPLUG |
38837fc7 | 1949 | /* |
0e1e7c7a CL |
1950 | * Keep top_cpuset.mems_allowed tracking node_states[N_HIGH_MEMORY]. |
1951 | * Call this routine anytime after you change | |
1952 | * node_states[N_HIGH_MEMORY]. | |
38837fc7 PJ |
1953 | * See also the previous routine cpuset_handle_cpuhp(). |
1954 | */ | |
1955 | ||
1af98928 | 1956 | void cpuset_track_online_nodes(void) |
38837fc7 | 1957 | { |
3e84050c | 1958 | common_cpu_mem_hotplug_unplug(0); |
38837fc7 PJ |
1959 | } |
1960 | #endif | |
1961 | ||
1da177e4 LT |
1962 | /** |
1963 | * cpuset_init_smp - initialize cpus_allowed | |
1964 | * | |
1965 | * Description: Finish top cpuset after cpu, node maps are initialized | |
1966 | **/ | |
1967 | ||
1968 | void __init cpuset_init_smp(void) | |
1969 | { | |
1970 | top_cpuset.cpus_allowed = cpu_online_map; | |
0e1e7c7a | 1971 | top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY]; |
4c4d50f7 PJ |
1972 | |
1973 | hotcpu_notifier(cpuset_handle_cpuhp, 0); | |
1da177e4 LT |
1974 | } |
1975 | ||
1976 | /** | |
3077a260 | 1977 | |
1da177e4 LT |
1978 | * cpuset_cpus_allowed - return cpus_allowed mask from a tasks cpuset. |
1979 | * @tsk: pointer to task_struct from which to obtain cpuset->cpus_allowed. | |
f9a86fcb | 1980 | * @pmask: pointer to cpumask_t variable to receive cpus_allowed set. |
1da177e4 LT |
1981 | * |
1982 | * Description: Returns the cpumask_t cpus_allowed of the cpuset | |
1983 | * attached to the specified @tsk. Guaranteed to return some non-empty | |
1984 | * subset of cpu_online_map, even if this means going outside the | |
1985 | * tasks cpuset. | |
1986 | **/ | |
1987 | ||
f9a86fcb | 1988 | void cpuset_cpus_allowed(struct task_struct *tsk, cpumask_t *pmask) |
1da177e4 | 1989 | { |
3d3f26a7 | 1990 | mutex_lock(&callback_mutex); |
f9a86fcb | 1991 | cpuset_cpus_allowed_locked(tsk, pmask); |
470fd646 | 1992 | mutex_unlock(&callback_mutex); |
470fd646 CW |
1993 | } |
1994 | ||
1995 | /** | |
1996 | * cpuset_cpus_allowed_locked - return cpus_allowed mask from a tasks cpuset. | |
2df167a3 | 1997 | * Must be called with callback_mutex held. |
470fd646 | 1998 | **/ |
f9a86fcb | 1999 | void cpuset_cpus_allowed_locked(struct task_struct *tsk, cpumask_t *pmask) |
470fd646 | 2000 | { |
909d75a3 | 2001 | task_lock(tsk); |
f9a86fcb | 2002 | guarantee_online_cpus(task_cs(tsk), pmask); |
909d75a3 | 2003 | task_unlock(tsk); |
1da177e4 LT |
2004 | } |
2005 | ||
2006 | void cpuset_init_current_mems_allowed(void) | |
2007 | { | |
f9a86fcb | 2008 | nodes_setall(current->mems_allowed); |
1da177e4 LT |
2009 | } |
2010 | ||
909d75a3 PJ |
2011 | /** |
2012 | * cpuset_mems_allowed - return mems_allowed mask from a tasks cpuset. | |
2013 | * @tsk: pointer to task_struct from which to obtain cpuset->mems_allowed. | |
2014 | * | |
2015 | * Description: Returns the nodemask_t mems_allowed of the cpuset | |
2016 | * attached to the specified @tsk. Guaranteed to return some non-empty | |
0e1e7c7a | 2017 | * subset of node_states[N_HIGH_MEMORY], even if this means going outside the |
909d75a3 PJ |
2018 | * tasks cpuset. |
2019 | **/ | |
2020 | ||
2021 | nodemask_t cpuset_mems_allowed(struct task_struct *tsk) | |
2022 | { | |
2023 | nodemask_t mask; | |
2024 | ||
3d3f26a7 | 2025 | mutex_lock(&callback_mutex); |
909d75a3 | 2026 | task_lock(tsk); |
8793d854 | 2027 | guarantee_online_mems(task_cs(tsk), &mask); |
909d75a3 | 2028 | task_unlock(tsk); |
3d3f26a7 | 2029 | mutex_unlock(&callback_mutex); |
909d75a3 PJ |
2030 | |
2031 | return mask; | |
2032 | } | |
2033 | ||
d9fd8a6d | 2034 | /** |
19770b32 MG |
2035 | * cpuset_nodemask_valid_mems_allowed - check nodemask vs. curremt mems_allowed |
2036 | * @nodemask: the nodemask to be checked | |
d9fd8a6d | 2037 | * |
19770b32 | 2038 | * Are any of the nodes in the nodemask allowed in current->mems_allowed? |
1da177e4 | 2039 | */ |
19770b32 | 2040 | int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask) |
1da177e4 | 2041 | { |
19770b32 | 2042 | return nodes_intersects(*nodemask, current->mems_allowed); |
1da177e4 LT |
2043 | } |
2044 | ||
9bf2229f | 2045 | /* |
78608366 PM |
2046 | * nearest_hardwall_ancestor() - Returns the nearest mem_exclusive or |
2047 | * mem_hardwall ancestor to the specified cpuset. Call holding | |
2048 | * callback_mutex. If no ancestor is mem_exclusive or mem_hardwall | |
2049 | * (an unusual configuration), then returns the root cpuset. | |
9bf2229f | 2050 | */ |
78608366 | 2051 | static const struct cpuset *nearest_hardwall_ancestor(const struct cpuset *cs) |
9bf2229f | 2052 | { |
78608366 | 2053 | while (!(is_mem_exclusive(cs) || is_mem_hardwall(cs)) && cs->parent) |
9bf2229f PJ |
2054 | cs = cs->parent; |
2055 | return cs; | |
2056 | } | |
2057 | ||
d9fd8a6d | 2058 | /** |
02a0e53d | 2059 | * cpuset_zone_allowed_softwall - Can we allocate on zone z's memory node? |
9bf2229f | 2060 | * @z: is this zone on an allowed node? |
02a0e53d | 2061 | * @gfp_mask: memory allocation flags |
d9fd8a6d | 2062 | * |
02a0e53d PJ |
2063 | * If we're in interrupt, yes, we can always allocate. If |
2064 | * __GFP_THISNODE is set, yes, we can always allocate. If zone | |
9bf2229f PJ |
2065 | * z's node is in our tasks mems_allowed, yes. If it's not a |
2066 | * __GFP_HARDWALL request and this zone's nodes is in the nearest | |
78608366 | 2067 | * hardwalled cpuset ancestor to this tasks cpuset, yes. |
c596d9f3 DR |
2068 | * If the task has been OOM killed and has access to memory reserves |
2069 | * as specified by the TIF_MEMDIE flag, yes. | |
9bf2229f PJ |
2070 | * Otherwise, no. |
2071 | * | |
02a0e53d PJ |
2072 | * If __GFP_HARDWALL is set, cpuset_zone_allowed_softwall() |
2073 | * reduces to cpuset_zone_allowed_hardwall(). Otherwise, | |
2074 | * cpuset_zone_allowed_softwall() might sleep, and might allow a zone | |
2075 | * from an enclosing cpuset. | |
2076 | * | |
2077 | * cpuset_zone_allowed_hardwall() only handles the simpler case of | |
2078 | * hardwall cpusets, and never sleeps. | |
2079 | * | |
2080 | * The __GFP_THISNODE placement logic is really handled elsewhere, | |
2081 | * by forcibly using a zonelist starting at a specified node, and by | |
2082 | * (in get_page_from_freelist()) refusing to consider the zones for | |
2083 | * any node on the zonelist except the first. By the time any such | |
2084 | * calls get to this routine, we should just shut up and say 'yes'. | |
2085 | * | |
9bf2229f | 2086 | * GFP_USER allocations are marked with the __GFP_HARDWALL bit, |
c596d9f3 DR |
2087 | * and do not allow allocations outside the current tasks cpuset |
2088 | * unless the task has been OOM killed as is marked TIF_MEMDIE. | |
9bf2229f | 2089 | * GFP_KERNEL allocations are not so marked, so can escape to the |
78608366 | 2090 | * nearest enclosing hardwalled ancestor cpuset. |
9bf2229f | 2091 | * |
02a0e53d PJ |
2092 | * Scanning up parent cpusets requires callback_mutex. The |
2093 | * __alloc_pages() routine only calls here with __GFP_HARDWALL bit | |
2094 | * _not_ set if it's a GFP_KERNEL allocation, and all nodes in the | |
2095 | * current tasks mems_allowed came up empty on the first pass over | |
2096 | * the zonelist. So only GFP_KERNEL allocations, if all nodes in the | |
2097 | * cpuset are short of memory, might require taking the callback_mutex | |
2098 | * mutex. | |
9bf2229f | 2099 | * |
36be57ff | 2100 | * The first call here from mm/page_alloc:get_page_from_freelist() |
02a0e53d PJ |
2101 | * has __GFP_HARDWALL set in gfp_mask, enforcing hardwall cpusets, |
2102 | * so no allocation on a node outside the cpuset is allowed (unless | |
2103 | * in interrupt, of course). | |
36be57ff PJ |
2104 | * |
2105 | * The second pass through get_page_from_freelist() doesn't even call | |
2106 | * here for GFP_ATOMIC calls. For those calls, the __alloc_pages() | |
2107 | * variable 'wait' is not set, and the bit ALLOC_CPUSET is not set | |
2108 | * in alloc_flags. That logic and the checks below have the combined | |
2109 | * affect that: | |
9bf2229f PJ |
2110 | * in_interrupt - any node ok (current task context irrelevant) |
2111 | * GFP_ATOMIC - any node ok | |
c596d9f3 | 2112 | * TIF_MEMDIE - any node ok |
78608366 | 2113 | * GFP_KERNEL - any node in enclosing hardwalled cpuset ok |
9bf2229f | 2114 | * GFP_USER - only nodes in current tasks mems allowed ok. |
36be57ff PJ |
2115 | * |
2116 | * Rule: | |
02a0e53d | 2117 | * Don't call cpuset_zone_allowed_softwall if you can't sleep, unless you |
36be57ff PJ |
2118 | * pass in the __GFP_HARDWALL flag set in gfp_flag, which disables |
2119 | * the code that might scan up ancestor cpusets and sleep. | |
02a0e53d | 2120 | */ |
9bf2229f | 2121 | |
02a0e53d | 2122 | int __cpuset_zone_allowed_softwall(struct zone *z, gfp_t gfp_mask) |
1da177e4 | 2123 | { |
9bf2229f PJ |
2124 | int node; /* node that zone z is on */ |
2125 | const struct cpuset *cs; /* current cpuset ancestors */ | |
29afd49b | 2126 | int allowed; /* is allocation in zone z allowed? */ |
9bf2229f | 2127 | |
9b819d20 | 2128 | if (in_interrupt() || (gfp_mask & __GFP_THISNODE)) |
9bf2229f | 2129 | return 1; |
89fa3024 | 2130 | node = zone_to_nid(z); |
92d1dbd2 | 2131 | might_sleep_if(!(gfp_mask & __GFP_HARDWALL)); |
9bf2229f PJ |
2132 | if (node_isset(node, current->mems_allowed)) |
2133 | return 1; | |
c596d9f3 DR |
2134 | /* |
2135 | * Allow tasks that have access to memory reserves because they have | |
2136 | * been OOM killed to get memory anywhere. | |
2137 | */ | |
2138 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
2139 | return 1; | |
9bf2229f PJ |
2140 | if (gfp_mask & __GFP_HARDWALL) /* If hardwall request, stop here */ |
2141 | return 0; | |
2142 | ||
5563e770 BP |
2143 | if (current->flags & PF_EXITING) /* Let dying task have memory */ |
2144 | return 1; | |
2145 | ||
9bf2229f | 2146 | /* Not hardwall and node outside mems_allowed: scan up cpusets */ |
3d3f26a7 | 2147 | mutex_lock(&callback_mutex); |
053199ed | 2148 | |
053199ed | 2149 | task_lock(current); |
78608366 | 2150 | cs = nearest_hardwall_ancestor(task_cs(current)); |
053199ed PJ |
2151 | task_unlock(current); |
2152 | ||
9bf2229f | 2153 | allowed = node_isset(node, cs->mems_allowed); |
3d3f26a7 | 2154 | mutex_unlock(&callback_mutex); |
9bf2229f | 2155 | return allowed; |
1da177e4 LT |
2156 | } |
2157 | ||
02a0e53d PJ |
2158 | /* |
2159 | * cpuset_zone_allowed_hardwall - Can we allocate on zone z's memory node? | |
2160 | * @z: is this zone on an allowed node? | |
2161 | * @gfp_mask: memory allocation flags | |
2162 | * | |
2163 | * If we're in interrupt, yes, we can always allocate. | |
2164 | * If __GFP_THISNODE is set, yes, we can always allocate. If zone | |
c596d9f3 DR |
2165 | * z's node is in our tasks mems_allowed, yes. If the task has been |
2166 | * OOM killed and has access to memory reserves as specified by the | |
2167 | * TIF_MEMDIE flag, yes. Otherwise, no. | |
02a0e53d PJ |
2168 | * |
2169 | * The __GFP_THISNODE placement logic is really handled elsewhere, | |
2170 | * by forcibly using a zonelist starting at a specified node, and by | |
2171 | * (in get_page_from_freelist()) refusing to consider the zones for | |
2172 | * any node on the zonelist except the first. By the time any such | |
2173 | * calls get to this routine, we should just shut up and say 'yes'. | |
2174 | * | |
2175 | * Unlike the cpuset_zone_allowed_softwall() variant, above, | |
2176 | * this variant requires that the zone be in the current tasks | |
2177 | * mems_allowed or that we're in interrupt. It does not scan up the | |
2178 | * cpuset hierarchy for the nearest enclosing mem_exclusive cpuset. | |
2179 | * It never sleeps. | |
2180 | */ | |
2181 | ||
2182 | int __cpuset_zone_allowed_hardwall(struct zone *z, gfp_t gfp_mask) | |
2183 | { | |
2184 | int node; /* node that zone z is on */ | |
2185 | ||
2186 | if (in_interrupt() || (gfp_mask & __GFP_THISNODE)) | |
2187 | return 1; | |
2188 | node = zone_to_nid(z); | |
2189 | if (node_isset(node, current->mems_allowed)) | |
2190 | return 1; | |
dedf8b79 DW |
2191 | /* |
2192 | * Allow tasks that have access to memory reserves because they have | |
2193 | * been OOM killed to get memory anywhere. | |
2194 | */ | |
2195 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
2196 | return 1; | |
02a0e53d PJ |
2197 | return 0; |
2198 | } | |
2199 | ||
505970b9 PJ |
2200 | /** |
2201 | * cpuset_lock - lock out any changes to cpuset structures | |
2202 | * | |
3d3f26a7 | 2203 | * The out of memory (oom) code needs to mutex_lock cpusets |
505970b9 | 2204 | * from being changed while it scans the tasklist looking for a |
3d3f26a7 | 2205 | * task in an overlapping cpuset. Expose callback_mutex via this |
505970b9 PJ |
2206 | * cpuset_lock() routine, so the oom code can lock it, before |
2207 | * locking the task list. The tasklist_lock is a spinlock, so | |
3d3f26a7 | 2208 | * must be taken inside callback_mutex. |
505970b9 PJ |
2209 | */ |
2210 | ||
2211 | void cpuset_lock(void) | |
2212 | { | |
3d3f26a7 | 2213 | mutex_lock(&callback_mutex); |
505970b9 PJ |
2214 | } |
2215 | ||
2216 | /** | |
2217 | * cpuset_unlock - release lock on cpuset changes | |
2218 | * | |
2219 | * Undo the lock taken in a previous cpuset_lock() call. | |
2220 | */ | |
2221 | ||
2222 | void cpuset_unlock(void) | |
2223 | { | |
3d3f26a7 | 2224 | mutex_unlock(&callback_mutex); |
505970b9 PJ |
2225 | } |
2226 | ||
825a46af PJ |
2227 | /** |
2228 | * cpuset_mem_spread_node() - On which node to begin search for a page | |
2229 | * | |
2230 | * If a task is marked PF_SPREAD_PAGE or PF_SPREAD_SLAB (as for | |
2231 | * tasks in a cpuset with is_spread_page or is_spread_slab set), | |
2232 | * and if the memory allocation used cpuset_mem_spread_node() | |
2233 | * to determine on which node to start looking, as it will for | |
2234 | * certain page cache or slab cache pages such as used for file | |
2235 | * system buffers and inode caches, then instead of starting on the | |
2236 | * local node to look for a free page, rather spread the starting | |
2237 | * node around the tasks mems_allowed nodes. | |
2238 | * | |
2239 | * We don't have to worry about the returned node being offline | |
2240 | * because "it can't happen", and even if it did, it would be ok. | |
2241 | * | |
2242 | * The routines calling guarantee_online_mems() are careful to | |
2243 | * only set nodes in task->mems_allowed that are online. So it | |
2244 | * should not be possible for the following code to return an | |
2245 | * offline node. But if it did, that would be ok, as this routine | |
2246 | * is not returning the node where the allocation must be, only | |
2247 | * the node where the search should start. The zonelist passed to | |
2248 | * __alloc_pages() will include all nodes. If the slab allocator | |
2249 | * is passed an offline node, it will fall back to the local node. | |
2250 | * See kmem_cache_alloc_node(). | |
2251 | */ | |
2252 | ||
2253 | int cpuset_mem_spread_node(void) | |
2254 | { | |
2255 | int node; | |
2256 | ||
2257 | node = next_node(current->cpuset_mem_spread_rotor, current->mems_allowed); | |
2258 | if (node == MAX_NUMNODES) | |
2259 | node = first_node(current->mems_allowed); | |
2260 | current->cpuset_mem_spread_rotor = node; | |
2261 | return node; | |
2262 | } | |
2263 | EXPORT_SYMBOL_GPL(cpuset_mem_spread_node); | |
2264 | ||
ef08e3b4 | 2265 | /** |
bbe373f2 DR |
2266 | * cpuset_mems_allowed_intersects - Does @tsk1's mems_allowed intersect @tsk2's? |
2267 | * @tsk1: pointer to task_struct of some task. | |
2268 | * @tsk2: pointer to task_struct of some other task. | |
2269 | * | |
2270 | * Description: Return true if @tsk1's mems_allowed intersects the | |
2271 | * mems_allowed of @tsk2. Used by the OOM killer to determine if | |
2272 | * one of the task's memory usage might impact the memory available | |
2273 | * to the other. | |
ef08e3b4 PJ |
2274 | **/ |
2275 | ||
bbe373f2 DR |
2276 | int cpuset_mems_allowed_intersects(const struct task_struct *tsk1, |
2277 | const struct task_struct *tsk2) | |
ef08e3b4 | 2278 | { |
bbe373f2 | 2279 | return nodes_intersects(tsk1->mems_allowed, tsk2->mems_allowed); |
ef08e3b4 PJ |
2280 | } |
2281 | ||
3e0d98b9 PJ |
2282 | /* |
2283 | * Collection of memory_pressure is suppressed unless | |
2284 | * this flag is enabled by writing "1" to the special | |
2285 | * cpuset file 'memory_pressure_enabled' in the root cpuset. | |
2286 | */ | |
2287 | ||
c5b2aff8 | 2288 | int cpuset_memory_pressure_enabled __read_mostly; |
3e0d98b9 PJ |
2289 | |
2290 | /** | |
2291 | * cpuset_memory_pressure_bump - keep stats of per-cpuset reclaims. | |
2292 | * | |
2293 | * Keep a running average of the rate of synchronous (direct) | |
2294 | * page reclaim efforts initiated by tasks in each cpuset. | |
2295 | * | |
2296 | * This represents the rate at which some task in the cpuset | |
2297 | * ran low on memory on all nodes it was allowed to use, and | |
2298 | * had to enter the kernels page reclaim code in an effort to | |
2299 | * create more free memory by tossing clean pages or swapping | |
2300 | * or writing dirty pages. | |
2301 | * | |
2302 | * Display to user space in the per-cpuset read-only file | |
2303 | * "memory_pressure". Value displayed is an integer | |
2304 | * representing the recent rate of entry into the synchronous | |
2305 | * (direct) page reclaim by any task attached to the cpuset. | |
2306 | **/ | |
2307 | ||
2308 | void __cpuset_memory_pressure_bump(void) | |
2309 | { | |
3e0d98b9 | 2310 | task_lock(current); |
8793d854 | 2311 | fmeter_markevent(&task_cs(current)->fmeter); |
3e0d98b9 PJ |
2312 | task_unlock(current); |
2313 | } | |
2314 | ||
8793d854 | 2315 | #ifdef CONFIG_PROC_PID_CPUSET |
1da177e4 LT |
2316 | /* |
2317 | * proc_cpuset_show() | |
2318 | * - Print tasks cpuset path into seq_file. | |
2319 | * - Used for /proc/<pid>/cpuset. | |
053199ed PJ |
2320 | * - No need to task_lock(tsk) on this tsk->cpuset reference, as it |
2321 | * doesn't really matter if tsk->cpuset changes after we read it, | |
c8d9c90c | 2322 | * and we take cgroup_mutex, keeping cpuset_attach() from changing it |
2df167a3 | 2323 | * anyway. |
1da177e4 | 2324 | */ |
029190c5 | 2325 | static int proc_cpuset_show(struct seq_file *m, void *unused_v) |
1da177e4 | 2326 | { |
13b41b09 | 2327 | struct pid *pid; |
1da177e4 LT |
2328 | struct task_struct *tsk; |
2329 | char *buf; | |
8793d854 | 2330 | struct cgroup_subsys_state *css; |
99f89551 | 2331 | int retval; |
1da177e4 | 2332 | |
99f89551 | 2333 | retval = -ENOMEM; |
1da177e4 LT |
2334 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
2335 | if (!buf) | |
99f89551 EB |
2336 | goto out; |
2337 | ||
2338 | retval = -ESRCH; | |
13b41b09 EB |
2339 | pid = m->private; |
2340 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
99f89551 EB |
2341 | if (!tsk) |
2342 | goto out_free; | |
1da177e4 | 2343 | |
99f89551 | 2344 | retval = -EINVAL; |
8793d854 PM |
2345 | cgroup_lock(); |
2346 | css = task_subsys_state(tsk, cpuset_subsys_id); | |
2347 | retval = cgroup_path(css->cgroup, buf, PAGE_SIZE); | |
1da177e4 | 2348 | if (retval < 0) |
99f89551 | 2349 | goto out_unlock; |
1da177e4 LT |
2350 | seq_puts(m, buf); |
2351 | seq_putc(m, '\n'); | |
99f89551 | 2352 | out_unlock: |
8793d854 | 2353 | cgroup_unlock(); |
99f89551 EB |
2354 | put_task_struct(tsk); |
2355 | out_free: | |
1da177e4 | 2356 | kfree(buf); |
99f89551 | 2357 | out: |
1da177e4 LT |
2358 | return retval; |
2359 | } | |
2360 | ||
2361 | static int cpuset_open(struct inode *inode, struct file *file) | |
2362 | { | |
13b41b09 EB |
2363 | struct pid *pid = PROC_I(inode)->pid; |
2364 | return single_open(file, proc_cpuset_show, pid); | |
1da177e4 LT |
2365 | } |
2366 | ||
9a32144e | 2367 | const struct file_operations proc_cpuset_operations = { |
1da177e4 LT |
2368 | .open = cpuset_open, |
2369 | .read = seq_read, | |
2370 | .llseek = seq_lseek, | |
2371 | .release = single_release, | |
2372 | }; | |
8793d854 | 2373 | #endif /* CONFIG_PROC_PID_CPUSET */ |
1da177e4 LT |
2374 | |
2375 | /* Display task cpus_allowed, mems_allowed in /proc/<pid>/status file. */ | |
df5f8314 EB |
2376 | void cpuset_task_status_allowed(struct seq_file *m, struct task_struct *task) |
2377 | { | |
2378 | seq_printf(m, "Cpus_allowed:\t"); | |
2379 | m->count += cpumask_scnprintf(m->buf + m->count, m->size - m->count, | |
2380 | task->cpus_allowed); | |
2381 | seq_printf(m, "\n"); | |
39106dcf MT |
2382 | seq_printf(m, "Cpus_allowed_list:\t"); |
2383 | m->count += cpulist_scnprintf(m->buf + m->count, m->size - m->count, | |
2384 | task->cpus_allowed); | |
2385 | seq_printf(m, "\n"); | |
df5f8314 EB |
2386 | seq_printf(m, "Mems_allowed:\t"); |
2387 | m->count += nodemask_scnprintf(m->buf + m->count, m->size - m->count, | |
2388 | task->mems_allowed); | |
2389 | seq_printf(m, "\n"); | |
39106dcf MT |
2390 | seq_printf(m, "Mems_allowed_list:\t"); |
2391 | m->count += nodelist_scnprintf(m->buf + m->count, m->size - m->count, | |
2392 | task->mems_allowed); | |
2393 | seq_printf(m, "\n"); | |
1da177e4 | 2394 | } |