]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/cpuset.c | |
3 | * | |
4 | * Processor and Memory placement constraints for sets of tasks. | |
5 | * | |
6 | * Copyright (C) 2003 BULL SA. | |
029190c5 | 7 | * Copyright (C) 2004-2007 Silicon Graphics, Inc. |
8793d854 | 8 | * Copyright (C) 2006 Google, Inc |
1da177e4 LT |
9 | * |
10 | * Portions derived from Patrick Mochel's sysfs code. | |
11 | * sysfs is Copyright (c) 2001-3 Patrick Mochel | |
1da177e4 | 12 | * |
825a46af | 13 | * 2003-10-10 Written by Simon Derr. |
1da177e4 | 14 | * 2003-10-22 Updates by Stephen Hemminger. |
825a46af | 15 | * 2004 May-July Rework by Paul Jackson. |
8793d854 | 16 | * 2006 Rework by Paul Menage to use generic cgroups |
cf417141 MK |
17 | * 2008 Rework of the scheduler domains and CPU hotplug handling |
18 | * by Max Krasnyansky | |
1da177e4 LT |
19 | * |
20 | * This file is subject to the terms and conditions of the GNU General Public | |
21 | * License. See the file COPYING in the main directory of the Linux | |
22 | * distribution for more details. | |
23 | */ | |
24 | ||
1da177e4 LT |
25 | #include <linux/cpu.h> |
26 | #include <linux/cpumask.h> | |
27 | #include <linux/cpuset.h> | |
28 | #include <linux/err.h> | |
29 | #include <linux/errno.h> | |
30 | #include <linux/file.h> | |
31 | #include <linux/fs.h> | |
32 | #include <linux/init.h> | |
33 | #include <linux/interrupt.h> | |
34 | #include <linux/kernel.h> | |
35 | #include <linux/kmod.h> | |
36 | #include <linux/list.h> | |
68860ec1 | 37 | #include <linux/mempolicy.h> |
1da177e4 | 38 | #include <linux/mm.h> |
f481891f | 39 | #include <linux/memory.h> |
1da177e4 LT |
40 | #include <linux/module.h> |
41 | #include <linux/mount.h> | |
42 | #include <linux/namei.h> | |
43 | #include <linux/pagemap.h> | |
44 | #include <linux/proc_fs.h> | |
6b9c2603 | 45 | #include <linux/rcupdate.h> |
1da177e4 LT |
46 | #include <linux/sched.h> |
47 | #include <linux/seq_file.h> | |
22fb52dd | 48 | #include <linux/security.h> |
1da177e4 | 49 | #include <linux/slab.h> |
1da177e4 LT |
50 | #include <linux/spinlock.h> |
51 | #include <linux/stat.h> | |
52 | #include <linux/string.h> | |
53 | #include <linux/time.h> | |
54 | #include <linux/backing-dev.h> | |
55 | #include <linux/sort.h> | |
56 | ||
57 | #include <asm/uaccess.h> | |
58 | #include <asm/atomic.h> | |
3d3f26a7 | 59 | #include <linux/mutex.h> |
956db3ca CW |
60 | #include <linux/workqueue.h> |
61 | #include <linux/cgroup.h> | |
1da177e4 | 62 | |
f90d4118 MX |
63 | /* |
64 | * Workqueue for cpuset related tasks. | |
65 | * | |
66 | * Using kevent workqueue may cause deadlock when memory_migrate | |
67 | * is set. So we create a separate workqueue thread for cpuset. | |
68 | */ | |
69 | static struct workqueue_struct *cpuset_wq; | |
70 | ||
202f72d5 PJ |
71 | /* |
72 | * Tracks how many cpusets are currently defined in system. | |
73 | * When there is only one cpuset (the root cpuset) we can | |
74 | * short circuit some hooks. | |
75 | */ | |
7edc5962 | 76 | int number_of_cpusets __read_mostly; |
202f72d5 | 77 | |
2df167a3 | 78 | /* Forward declare cgroup structures */ |
8793d854 PM |
79 | struct cgroup_subsys cpuset_subsys; |
80 | struct cpuset; | |
81 | ||
3e0d98b9 PJ |
82 | /* See "Frequency meter" comments, below. */ |
83 | ||
84 | struct fmeter { | |
85 | int cnt; /* unprocessed events count */ | |
86 | int val; /* most recent output value */ | |
87 | time_t time; /* clock (secs) when val computed */ | |
88 | spinlock_t lock; /* guards read or write of above */ | |
89 | }; | |
90 | ||
1da177e4 | 91 | struct cpuset { |
8793d854 PM |
92 | struct cgroup_subsys_state css; |
93 | ||
1da177e4 | 94 | unsigned long flags; /* "unsigned long" so bitops work */ |
300ed6cb | 95 | cpumask_var_t cpus_allowed; /* CPUs allowed to tasks in cpuset */ |
1da177e4 LT |
96 | nodemask_t mems_allowed; /* Memory Nodes allowed to tasks */ |
97 | ||
1da177e4 | 98 | struct cpuset *parent; /* my parent */ |
1da177e4 LT |
99 | |
100 | /* | |
101 | * Copy of global cpuset_mems_generation as of the most | |
102 | * recent time this cpuset changed its mems_allowed. | |
103 | */ | |
3e0d98b9 PJ |
104 | int mems_generation; |
105 | ||
106 | struct fmeter fmeter; /* memory_pressure filter */ | |
029190c5 PJ |
107 | |
108 | /* partition number for rebuild_sched_domains() */ | |
109 | int pn; | |
956db3ca | 110 | |
1d3504fc HS |
111 | /* for custom sched domain */ |
112 | int relax_domain_level; | |
113 | ||
956db3ca CW |
114 | /* used for walking a cpuset heirarchy */ |
115 | struct list_head stack_list; | |
1da177e4 LT |
116 | }; |
117 | ||
8793d854 PM |
118 | /* Retrieve the cpuset for a cgroup */ |
119 | static inline struct cpuset *cgroup_cs(struct cgroup *cont) | |
120 | { | |
121 | return container_of(cgroup_subsys_state(cont, cpuset_subsys_id), | |
122 | struct cpuset, css); | |
123 | } | |
124 | ||
125 | /* Retrieve the cpuset for a task */ | |
126 | static inline struct cpuset *task_cs(struct task_struct *task) | |
127 | { | |
128 | return container_of(task_subsys_state(task, cpuset_subsys_id), | |
129 | struct cpuset, css); | |
130 | } | |
8793d854 | 131 | |
1da177e4 LT |
132 | /* bits in struct cpuset flags field */ |
133 | typedef enum { | |
134 | CS_CPU_EXCLUSIVE, | |
135 | CS_MEM_EXCLUSIVE, | |
78608366 | 136 | CS_MEM_HARDWALL, |
45b07ef3 | 137 | CS_MEMORY_MIGRATE, |
029190c5 | 138 | CS_SCHED_LOAD_BALANCE, |
825a46af PJ |
139 | CS_SPREAD_PAGE, |
140 | CS_SPREAD_SLAB, | |
1da177e4 LT |
141 | } cpuset_flagbits_t; |
142 | ||
143 | /* convenient tests for these bits */ | |
144 | static inline int is_cpu_exclusive(const struct cpuset *cs) | |
145 | { | |
7b5b9ef0 | 146 | return test_bit(CS_CPU_EXCLUSIVE, &cs->flags); |
1da177e4 LT |
147 | } |
148 | ||
149 | static inline int is_mem_exclusive(const struct cpuset *cs) | |
150 | { | |
7b5b9ef0 | 151 | return test_bit(CS_MEM_EXCLUSIVE, &cs->flags); |
1da177e4 LT |
152 | } |
153 | ||
78608366 PM |
154 | static inline int is_mem_hardwall(const struct cpuset *cs) |
155 | { | |
156 | return test_bit(CS_MEM_HARDWALL, &cs->flags); | |
157 | } | |
158 | ||
029190c5 PJ |
159 | static inline int is_sched_load_balance(const struct cpuset *cs) |
160 | { | |
161 | return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); | |
162 | } | |
163 | ||
45b07ef3 PJ |
164 | static inline int is_memory_migrate(const struct cpuset *cs) |
165 | { | |
7b5b9ef0 | 166 | return test_bit(CS_MEMORY_MIGRATE, &cs->flags); |
45b07ef3 PJ |
167 | } |
168 | ||
825a46af PJ |
169 | static inline int is_spread_page(const struct cpuset *cs) |
170 | { | |
171 | return test_bit(CS_SPREAD_PAGE, &cs->flags); | |
172 | } | |
173 | ||
174 | static inline int is_spread_slab(const struct cpuset *cs) | |
175 | { | |
176 | return test_bit(CS_SPREAD_SLAB, &cs->flags); | |
177 | } | |
178 | ||
1da177e4 | 179 | /* |
151a4420 | 180 | * Increment this integer everytime any cpuset changes its |
1da177e4 LT |
181 | * mems_allowed value. Users of cpusets can track this generation |
182 | * number, and avoid having to lock and reload mems_allowed unless | |
183 | * the cpuset they're using changes generation. | |
184 | * | |
2df167a3 | 185 | * A single, global generation is needed because cpuset_attach_task() could |
1da177e4 LT |
186 | * reattach a task to a different cpuset, which must not have its |
187 | * generation numbers aliased with those of that tasks previous cpuset. | |
188 | * | |
189 | * Generations are needed for mems_allowed because one task cannot | |
2df167a3 | 190 | * modify another's memory placement. So we must enable every task, |
1da177e4 LT |
191 | * on every visit to __alloc_pages(), to efficiently check whether |
192 | * its current->cpuset->mems_allowed has changed, requiring an update | |
193 | * of its current->mems_allowed. | |
151a4420 | 194 | * |
2df167a3 | 195 | * Since writes to cpuset_mems_generation are guarded by the cgroup lock |
151a4420 | 196 | * there is no need to mark it atomic. |
1da177e4 | 197 | */ |
151a4420 | 198 | static int cpuset_mems_generation; |
1da177e4 LT |
199 | |
200 | static struct cpuset top_cpuset = { | |
201 | .flags = ((1 << CS_CPU_EXCLUSIVE) | (1 << CS_MEM_EXCLUSIVE)), | |
1da177e4 LT |
202 | }; |
203 | ||
1da177e4 | 204 | /* |
2df167a3 PM |
205 | * There are two global mutexes guarding cpuset structures. The first |
206 | * is the main control groups cgroup_mutex, accessed via | |
207 | * cgroup_lock()/cgroup_unlock(). The second is the cpuset-specific | |
208 | * callback_mutex, below. They can nest. It is ok to first take | |
209 | * cgroup_mutex, then nest callback_mutex. We also require taking | |
210 | * task_lock() when dereferencing a task's cpuset pointer. See "The | |
211 | * task_lock() exception", at the end of this comment. | |
053199ed | 212 | * |
3d3f26a7 | 213 | * A task must hold both mutexes to modify cpusets. If a task |
2df167a3 | 214 | * holds cgroup_mutex, then it blocks others wanting that mutex, |
3d3f26a7 | 215 | * ensuring that it is the only task able to also acquire callback_mutex |
053199ed PJ |
216 | * and be able to modify cpusets. It can perform various checks on |
217 | * the cpuset structure first, knowing nothing will change. It can | |
2df167a3 | 218 | * also allocate memory while just holding cgroup_mutex. While it is |
053199ed | 219 | * performing these checks, various callback routines can briefly |
3d3f26a7 IM |
220 | * acquire callback_mutex to query cpusets. Once it is ready to make |
221 | * the changes, it takes callback_mutex, blocking everyone else. | |
053199ed PJ |
222 | * |
223 | * Calls to the kernel memory allocator can not be made while holding | |
3d3f26a7 | 224 | * callback_mutex, as that would risk double tripping on callback_mutex |
053199ed PJ |
225 | * from one of the callbacks into the cpuset code from within |
226 | * __alloc_pages(). | |
227 | * | |
3d3f26a7 | 228 | * If a task is only holding callback_mutex, then it has read-only |
053199ed PJ |
229 | * access to cpusets. |
230 | * | |
231 | * The task_struct fields mems_allowed and mems_generation may only | |
232 | * be accessed in the context of that task, so require no locks. | |
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 | |
75aa1994 DR |
244 | /* |
245 | * cpuset_buffer_lock protects both the cpuset_name and cpuset_nodelist | |
246 | * buffers. They are statically allocated to prevent using excess stack | |
247 | * when calling cpuset_print_task_mems_allowed(). | |
248 | */ | |
249 | #define CPUSET_NAME_LEN (128) | |
250 | #define CPUSET_NODELIST_LEN (256) | |
251 | static char cpuset_name[CPUSET_NAME_LEN]; | |
252 | static char cpuset_nodelist[CPUSET_NODELIST_LEN]; | |
253 | static DEFINE_SPINLOCK(cpuset_buffer_lock); | |
254 | ||
cf417141 MK |
255 | /* |
256 | * This is ugly, but preserves the userspace API for existing cpuset | |
8793d854 | 257 | * users. If someone tries to mount the "cpuset" filesystem, we |
cf417141 MK |
258 | * silently switch it to mount "cgroup" instead |
259 | */ | |
454e2398 DH |
260 | static int cpuset_get_sb(struct file_system_type *fs_type, |
261 | int flags, const char *unused_dev_name, | |
262 | void *data, struct vfsmount *mnt) | |
1da177e4 | 263 | { |
8793d854 PM |
264 | struct file_system_type *cgroup_fs = get_fs_type("cgroup"); |
265 | int ret = -ENODEV; | |
266 | if (cgroup_fs) { | |
267 | char mountopts[] = | |
268 | "cpuset,noprefix," | |
269 | "release_agent=/sbin/cpuset_release_agent"; | |
270 | ret = cgroup_fs->get_sb(cgroup_fs, flags, | |
271 | unused_dev_name, mountopts, mnt); | |
272 | put_filesystem(cgroup_fs); | |
273 | } | |
274 | return ret; | |
1da177e4 LT |
275 | } |
276 | ||
277 | static struct file_system_type cpuset_fs_type = { | |
278 | .name = "cpuset", | |
279 | .get_sb = cpuset_get_sb, | |
1da177e4 LT |
280 | }; |
281 | ||
1da177e4 | 282 | /* |
300ed6cb | 283 | * Return in pmask the portion of a cpusets's cpus_allowed that |
1da177e4 LT |
284 | * are online. If none are online, walk up the cpuset hierarchy |
285 | * until we find one that does have some online cpus. If we get | |
286 | * all the way to the top and still haven't found any online cpus, | |
287 | * return cpu_online_map. Or if passed a NULL cs from an exit'ing | |
288 | * task, return cpu_online_map. | |
289 | * | |
290 | * One way or another, we guarantee to return some non-empty subset | |
291 | * of cpu_online_map. | |
292 | * | |
3d3f26a7 | 293 | * Call with callback_mutex held. |
1da177e4 LT |
294 | */ |
295 | ||
6af866af LZ |
296 | static void guarantee_online_cpus(const struct cpuset *cs, |
297 | struct cpumask *pmask) | |
1da177e4 | 298 | { |
300ed6cb | 299 | while (cs && !cpumask_intersects(cs->cpus_allowed, cpu_online_mask)) |
1da177e4 LT |
300 | cs = cs->parent; |
301 | if (cs) | |
300ed6cb | 302 | cpumask_and(pmask, cs->cpus_allowed, cpu_online_mask); |
1da177e4 | 303 | else |
300ed6cb LZ |
304 | cpumask_copy(pmask, cpu_online_mask); |
305 | BUG_ON(!cpumask_intersects(pmask, cpu_online_mask)); | |
1da177e4 LT |
306 | } |
307 | ||
308 | /* | |
309 | * Return in *pmask the portion of a cpusets's mems_allowed that | |
0e1e7c7a CL |
310 | * are online, with memory. If none are online with memory, walk |
311 | * up the cpuset hierarchy until we find one that does have some | |
312 | * online mems. If we get all the way to the top and still haven't | |
313 | * found any online mems, return node_states[N_HIGH_MEMORY]. | |
1da177e4 LT |
314 | * |
315 | * One way or another, we guarantee to return some non-empty subset | |
0e1e7c7a | 316 | * of node_states[N_HIGH_MEMORY]. |
1da177e4 | 317 | * |
3d3f26a7 | 318 | * Call with callback_mutex held. |
1da177e4 LT |
319 | */ |
320 | ||
321 | static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask) | |
322 | { | |
0e1e7c7a CL |
323 | while (cs && !nodes_intersects(cs->mems_allowed, |
324 | node_states[N_HIGH_MEMORY])) | |
1da177e4 LT |
325 | cs = cs->parent; |
326 | if (cs) | |
0e1e7c7a CL |
327 | nodes_and(*pmask, cs->mems_allowed, |
328 | node_states[N_HIGH_MEMORY]); | |
1da177e4 | 329 | else |
0e1e7c7a CL |
330 | *pmask = node_states[N_HIGH_MEMORY]; |
331 | BUG_ON(!nodes_intersects(*pmask, node_states[N_HIGH_MEMORY])); | |
1da177e4 LT |
332 | } |
333 | ||
f3b39d47 MX |
334 | /* |
335 | * update task's spread flag if cpuset's page/slab spread flag is set | |
336 | * | |
337 | * Called with callback_mutex/cgroup_mutex held | |
338 | */ | |
339 | static void cpuset_update_task_spread_flag(struct cpuset *cs, | |
340 | struct task_struct *tsk) | |
341 | { | |
342 | if (is_spread_page(cs)) | |
343 | tsk->flags |= PF_SPREAD_PAGE; | |
344 | else | |
345 | tsk->flags &= ~PF_SPREAD_PAGE; | |
346 | if (is_spread_slab(cs)) | |
347 | tsk->flags |= PF_SPREAD_SLAB; | |
348 | else | |
349 | tsk->flags &= ~PF_SPREAD_SLAB; | |
350 | } | |
351 | ||
cf2a473c PJ |
352 | /** |
353 | * cpuset_update_task_memory_state - update task memory placement | |
354 | * | |
355 | * If the current tasks cpusets mems_allowed changed behind our | |
356 | * backs, update current->mems_allowed, mems_generation and task NUMA | |
357 | * mempolicy to the new value. | |
053199ed | 358 | * |
cf2a473c PJ |
359 | * Task mempolicy is updated by rebinding it relative to the |
360 | * current->cpuset if a task has its memory placement changed. | |
361 | * Do not call this routine if in_interrupt(). | |
362 | * | |
4a01c8d5 | 363 | * Call without callback_mutex or task_lock() held. May be |
2df167a3 PM |
364 | * called with or without cgroup_mutex held. Thanks in part to |
365 | * 'the_top_cpuset_hack', the task's cpuset pointer will never | |
41f7f60d DR |
366 | * be NULL. This routine also might acquire callback_mutex during |
367 | * call. | |
053199ed | 368 | * |
6b9c2603 PJ |
369 | * Reading current->cpuset->mems_generation doesn't need task_lock |
370 | * to guard the current->cpuset derefence, because it is guarded | |
2df167a3 | 371 | * from concurrent freeing of current->cpuset using RCU. |
6b9c2603 PJ |
372 | * |
373 | * The rcu_dereference() is technically probably not needed, | |
374 | * as I don't actually mind if I see a new cpuset pointer but | |
375 | * an old value of mems_generation. However this really only | |
376 | * matters on alpha systems using cpusets heavily. If I dropped | |
377 | * that rcu_dereference(), it would save them a memory barrier. | |
378 | * For all other arch's, rcu_dereference is a no-op anyway, and for | |
379 | * alpha systems not using cpusets, another planned optimization, | |
380 | * avoiding the rcu critical section for tasks in the root cpuset | |
381 | * which is statically allocated, so can't vanish, will make this | |
382 | * irrelevant. Better to use RCU as intended, than to engage in | |
383 | * some cute trick to save a memory barrier that is impossible to | |
384 | * test, for alpha systems using cpusets heavily, which might not | |
385 | * even exist. | |
053199ed PJ |
386 | * |
387 | * This routine is needed to update the per-task mems_allowed data, | |
388 | * within the tasks context, when it is trying to allocate memory | |
389 | * (in various mm/mempolicy.c routines) and notices that some other | |
390 | * task has been modifying its cpuset. | |
1da177e4 LT |
391 | */ |
392 | ||
fe85a998 | 393 | void cpuset_update_task_memory_state(void) |
1da177e4 | 394 | { |
053199ed | 395 | int my_cpusets_mem_gen; |
cf2a473c | 396 | struct task_struct *tsk = current; |
6b9c2603 | 397 | struct cpuset *cs; |
053199ed | 398 | |
13337714 LJ |
399 | rcu_read_lock(); |
400 | my_cpusets_mem_gen = task_cs(tsk)->mems_generation; | |
401 | rcu_read_unlock(); | |
1da177e4 | 402 | |
cf2a473c | 403 | if (my_cpusets_mem_gen != tsk->cpuset_mems_generation) { |
3d3f26a7 | 404 | mutex_lock(&callback_mutex); |
cf2a473c | 405 | task_lock(tsk); |
8793d854 | 406 | cs = task_cs(tsk); /* Maybe changed when task not locked */ |
cf2a473c PJ |
407 | guarantee_online_mems(cs, &tsk->mems_allowed); |
408 | tsk->cpuset_mems_generation = cs->mems_generation; | |
f3b39d47 | 409 | cpuset_update_task_spread_flag(cs, tsk); |
cf2a473c | 410 | task_unlock(tsk); |
3d3f26a7 | 411 | mutex_unlock(&callback_mutex); |
74cb2155 | 412 | mpol_rebind_task(tsk, &tsk->mems_allowed); |
1da177e4 LT |
413 | } |
414 | } | |
415 | ||
416 | /* | |
417 | * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q? | |
418 | * | |
419 | * One cpuset is a subset of another if all its allowed CPUs and | |
420 | * Memory Nodes are a subset of the other, and its exclusive flags | |
2df167a3 | 421 | * are only set if the other's are set. Call holding cgroup_mutex. |
1da177e4 LT |
422 | */ |
423 | ||
424 | static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q) | |
425 | { | |
300ed6cb | 426 | return cpumask_subset(p->cpus_allowed, q->cpus_allowed) && |
1da177e4 LT |
427 | nodes_subset(p->mems_allowed, q->mems_allowed) && |
428 | is_cpu_exclusive(p) <= is_cpu_exclusive(q) && | |
429 | is_mem_exclusive(p) <= is_mem_exclusive(q); | |
430 | } | |
431 | ||
645fcc9d LZ |
432 | /** |
433 | * alloc_trial_cpuset - allocate a trial cpuset | |
434 | * @cs: the cpuset that the trial cpuset duplicates | |
435 | */ | |
436 | static struct cpuset *alloc_trial_cpuset(const struct cpuset *cs) | |
437 | { | |
300ed6cb LZ |
438 | struct cpuset *trial; |
439 | ||
440 | trial = kmemdup(cs, sizeof(*cs), GFP_KERNEL); | |
441 | if (!trial) | |
442 | return NULL; | |
443 | ||
444 | if (!alloc_cpumask_var(&trial->cpus_allowed, GFP_KERNEL)) { | |
445 | kfree(trial); | |
446 | return NULL; | |
447 | } | |
448 | cpumask_copy(trial->cpus_allowed, cs->cpus_allowed); | |
449 | ||
450 | return trial; | |
645fcc9d LZ |
451 | } |
452 | ||
453 | /** | |
454 | * free_trial_cpuset - free the trial cpuset | |
455 | * @trial: the trial cpuset to be freed | |
456 | */ | |
457 | static void free_trial_cpuset(struct cpuset *trial) | |
458 | { | |
300ed6cb | 459 | free_cpumask_var(trial->cpus_allowed); |
645fcc9d LZ |
460 | kfree(trial); |
461 | } | |
462 | ||
1da177e4 LT |
463 | /* |
464 | * validate_change() - Used to validate that any proposed cpuset change | |
465 | * follows the structural rules for cpusets. | |
466 | * | |
467 | * If we replaced the flag and mask values of the current cpuset | |
468 | * (cur) with those values in the trial cpuset (trial), would | |
469 | * our various subset and exclusive rules still be valid? Presumes | |
2df167a3 | 470 | * cgroup_mutex held. |
1da177e4 LT |
471 | * |
472 | * 'cur' is the address of an actual, in-use cpuset. Operations | |
473 | * such as list traversal that depend on the actual address of the | |
474 | * cpuset in the list must use cur below, not trial. | |
475 | * | |
476 | * 'trial' is the address of bulk structure copy of cur, with | |
477 | * perhaps one or more of the fields cpus_allowed, mems_allowed, | |
478 | * or flags changed to new, trial values. | |
479 | * | |
480 | * Return 0 if valid, -errno if not. | |
481 | */ | |
482 | ||
483 | static int validate_change(const struct cpuset *cur, const struct cpuset *trial) | |
484 | { | |
8793d854 | 485 | struct cgroup *cont; |
1da177e4 LT |
486 | struct cpuset *c, *par; |
487 | ||
488 | /* Each of our child cpusets must be a subset of us */ | |
8793d854 PM |
489 | list_for_each_entry(cont, &cur->css.cgroup->children, sibling) { |
490 | if (!is_cpuset_subset(cgroup_cs(cont), trial)) | |
1da177e4 LT |
491 | return -EBUSY; |
492 | } | |
493 | ||
494 | /* Remaining checks don't apply to root cpuset */ | |
69604067 | 495 | if (cur == &top_cpuset) |
1da177e4 LT |
496 | return 0; |
497 | ||
69604067 PJ |
498 | par = cur->parent; |
499 | ||
1da177e4 LT |
500 | /* We must be a subset of our parent cpuset */ |
501 | if (!is_cpuset_subset(trial, par)) | |
502 | return -EACCES; | |
503 | ||
2df167a3 PM |
504 | /* |
505 | * If either I or some sibling (!= me) is exclusive, we can't | |
506 | * overlap | |
507 | */ | |
8793d854 PM |
508 | list_for_each_entry(cont, &par->css.cgroup->children, sibling) { |
509 | c = cgroup_cs(cont); | |
1da177e4 LT |
510 | if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) && |
511 | c != cur && | |
300ed6cb | 512 | cpumask_intersects(trial->cpus_allowed, c->cpus_allowed)) |
1da177e4 LT |
513 | return -EINVAL; |
514 | if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) && | |
515 | c != cur && | |
516 | nodes_intersects(trial->mems_allowed, c->mems_allowed)) | |
517 | return -EINVAL; | |
518 | } | |
519 | ||
020958b6 PJ |
520 | /* Cpusets with tasks can't have empty cpus_allowed or mems_allowed */ |
521 | if (cgroup_task_count(cur->css.cgroup)) { | |
300ed6cb | 522 | if (cpumask_empty(trial->cpus_allowed) || |
020958b6 PJ |
523 | nodes_empty(trial->mems_allowed)) { |
524 | return -ENOSPC; | |
525 | } | |
526 | } | |
527 | ||
1da177e4 LT |
528 | return 0; |
529 | } | |
530 | ||
db7f47cf | 531 | #ifdef CONFIG_SMP |
029190c5 | 532 | /* |
cf417141 | 533 | * Helper routine for generate_sched_domains(). |
029190c5 PJ |
534 | * Do cpusets a, b have overlapping cpus_allowed masks? |
535 | */ | |
029190c5 PJ |
536 | static int cpusets_overlap(struct cpuset *a, struct cpuset *b) |
537 | { | |
300ed6cb | 538 | return cpumask_intersects(a->cpus_allowed, b->cpus_allowed); |
029190c5 PJ |
539 | } |
540 | ||
1d3504fc HS |
541 | static void |
542 | update_domain_attr(struct sched_domain_attr *dattr, struct cpuset *c) | |
543 | { | |
1d3504fc HS |
544 | if (dattr->relax_domain_level < c->relax_domain_level) |
545 | dattr->relax_domain_level = c->relax_domain_level; | |
546 | return; | |
547 | } | |
548 | ||
f5393693 LJ |
549 | static void |
550 | update_domain_attr_tree(struct sched_domain_attr *dattr, struct cpuset *c) | |
551 | { | |
552 | LIST_HEAD(q); | |
553 | ||
554 | list_add(&c->stack_list, &q); | |
555 | while (!list_empty(&q)) { | |
556 | struct cpuset *cp; | |
557 | struct cgroup *cont; | |
558 | struct cpuset *child; | |
559 | ||
560 | cp = list_first_entry(&q, struct cpuset, stack_list); | |
561 | list_del(q.next); | |
562 | ||
300ed6cb | 563 | if (cpumask_empty(cp->cpus_allowed)) |
f5393693 LJ |
564 | continue; |
565 | ||
566 | if (is_sched_load_balance(cp)) | |
567 | update_domain_attr(dattr, cp); | |
568 | ||
569 | list_for_each_entry(cont, &cp->css.cgroup->children, sibling) { | |
570 | child = cgroup_cs(cont); | |
571 | list_add_tail(&child->stack_list, &q); | |
572 | } | |
573 | } | |
574 | } | |
575 | ||
029190c5 | 576 | /* |
cf417141 MK |
577 | * generate_sched_domains() |
578 | * | |
579 | * This function builds a partial partition of the systems CPUs | |
580 | * A 'partial partition' is a set of non-overlapping subsets whose | |
581 | * union is a subset of that set. | |
582 | * The output of this function needs to be passed to kernel/sched.c | |
583 | * partition_sched_domains() routine, which will rebuild the scheduler's | |
584 | * load balancing domains (sched domains) as specified by that partial | |
585 | * partition. | |
029190c5 | 586 | * |
45ce80fb | 587 | * See "What is sched_load_balance" in Documentation/cgroups/cpusets.txt |
029190c5 PJ |
588 | * for a background explanation of this. |
589 | * | |
590 | * Does not return errors, on the theory that the callers of this | |
591 | * routine would rather not worry about failures to rebuild sched | |
592 | * domains when operating in the severe memory shortage situations | |
593 | * that could cause allocation failures below. | |
594 | * | |
cf417141 | 595 | * Must be called with cgroup_lock held. |
029190c5 PJ |
596 | * |
597 | * The three key local variables below are: | |
aeed6824 | 598 | * q - a linked-list queue of cpuset pointers, used to implement a |
029190c5 PJ |
599 | * top-down scan of all cpusets. This scan loads a pointer |
600 | * to each cpuset marked is_sched_load_balance into the | |
601 | * array 'csa'. For our purposes, rebuilding the schedulers | |
602 | * sched domains, we can ignore !is_sched_load_balance cpusets. | |
603 | * csa - (for CpuSet Array) Array of pointers to all the cpusets | |
604 | * that need to be load balanced, for convenient iterative | |
605 | * access by the subsequent code that finds the best partition, | |
606 | * i.e the set of domains (subsets) of CPUs such that the | |
607 | * cpus_allowed of every cpuset marked is_sched_load_balance | |
608 | * is a subset of one of these domains, while there are as | |
609 | * many such domains as possible, each as small as possible. | |
610 | * doms - Conversion of 'csa' to an array of cpumasks, for passing to | |
611 | * the kernel/sched.c routine partition_sched_domains() in a | |
612 | * convenient format, that can be easily compared to the prior | |
613 | * value to determine what partition elements (sched domains) | |
614 | * were changed (added or removed.) | |
615 | * | |
616 | * Finding the best partition (set of domains): | |
617 | * The triple nested loops below over i, j, k scan over the | |
618 | * load balanced cpusets (using the array of cpuset pointers in | |
619 | * csa[]) looking for pairs of cpusets that have overlapping | |
620 | * cpus_allowed, but which don't have the same 'pn' partition | |
621 | * number and gives them in the same partition number. It keeps | |
622 | * looping on the 'restart' label until it can no longer find | |
623 | * any such pairs. | |
624 | * | |
625 | * The union of the cpus_allowed masks from the set of | |
626 | * all cpusets having the same 'pn' value then form the one | |
627 | * element of the partition (one sched domain) to be passed to | |
628 | * partition_sched_domains(). | |
629 | */ | |
6af866af LZ |
630 | /* FIXME: see the FIXME in partition_sched_domains() */ |
631 | static int generate_sched_domains(struct cpumask **domains, | |
cf417141 | 632 | struct sched_domain_attr **attributes) |
029190c5 | 633 | { |
cf417141 | 634 | LIST_HEAD(q); /* queue of cpusets to be scanned */ |
029190c5 PJ |
635 | struct cpuset *cp; /* scans q */ |
636 | struct cpuset **csa; /* array of all cpuset ptrs */ | |
637 | int csn; /* how many cpuset ptrs in csa so far */ | |
638 | int i, j, k; /* indices for partition finding loops */ | |
6af866af | 639 | struct cpumask *doms; /* resulting partition; i.e. sched domains */ |
1d3504fc | 640 | struct sched_domain_attr *dattr; /* attributes for custom domains */ |
1583715d | 641 | int ndoms = 0; /* number of sched domains in result */ |
6af866af | 642 | int nslot; /* next empty doms[] struct cpumask slot */ |
029190c5 | 643 | |
029190c5 | 644 | doms = NULL; |
1d3504fc | 645 | dattr = NULL; |
cf417141 | 646 | csa = NULL; |
029190c5 PJ |
647 | |
648 | /* Special case for the 99% of systems with one, full, sched domain */ | |
649 | if (is_sched_load_balance(&top_cpuset)) { | |
6af866af | 650 | doms = kmalloc(cpumask_size(), GFP_KERNEL); |
029190c5 | 651 | if (!doms) |
cf417141 MK |
652 | goto done; |
653 | ||
1d3504fc HS |
654 | dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL); |
655 | if (dattr) { | |
656 | *dattr = SD_ATTR_INIT; | |
93a65575 | 657 | update_domain_attr_tree(dattr, &top_cpuset); |
1d3504fc | 658 | } |
300ed6cb | 659 | cpumask_copy(doms, top_cpuset.cpus_allowed); |
cf417141 MK |
660 | |
661 | ndoms = 1; | |
662 | goto done; | |
029190c5 PJ |
663 | } |
664 | ||
029190c5 PJ |
665 | csa = kmalloc(number_of_cpusets * sizeof(cp), GFP_KERNEL); |
666 | if (!csa) | |
667 | goto done; | |
668 | csn = 0; | |
669 | ||
aeed6824 LZ |
670 | list_add(&top_cpuset.stack_list, &q); |
671 | while (!list_empty(&q)) { | |
029190c5 PJ |
672 | struct cgroup *cont; |
673 | struct cpuset *child; /* scans child cpusets of cp */ | |
489a5393 | 674 | |
aeed6824 LZ |
675 | cp = list_first_entry(&q, struct cpuset, stack_list); |
676 | list_del(q.next); | |
677 | ||
300ed6cb | 678 | if (cpumask_empty(cp->cpus_allowed)) |
489a5393 LJ |
679 | continue; |
680 | ||
f5393693 LJ |
681 | /* |
682 | * All child cpusets contain a subset of the parent's cpus, so | |
683 | * just skip them, and then we call update_domain_attr_tree() | |
684 | * to calc relax_domain_level of the corresponding sched | |
685 | * domain. | |
686 | */ | |
687 | if (is_sched_load_balance(cp)) { | |
029190c5 | 688 | csa[csn++] = cp; |
f5393693 LJ |
689 | continue; |
690 | } | |
489a5393 | 691 | |
029190c5 PJ |
692 | list_for_each_entry(cont, &cp->css.cgroup->children, sibling) { |
693 | child = cgroup_cs(cont); | |
aeed6824 | 694 | list_add_tail(&child->stack_list, &q); |
029190c5 PJ |
695 | } |
696 | } | |
697 | ||
698 | for (i = 0; i < csn; i++) | |
699 | csa[i]->pn = i; | |
700 | ndoms = csn; | |
701 | ||
702 | restart: | |
703 | /* Find the best partition (set of sched domains) */ | |
704 | for (i = 0; i < csn; i++) { | |
705 | struct cpuset *a = csa[i]; | |
706 | int apn = a->pn; | |
707 | ||
708 | for (j = 0; j < csn; j++) { | |
709 | struct cpuset *b = csa[j]; | |
710 | int bpn = b->pn; | |
711 | ||
712 | if (apn != bpn && cpusets_overlap(a, b)) { | |
713 | for (k = 0; k < csn; k++) { | |
714 | struct cpuset *c = csa[k]; | |
715 | ||
716 | if (c->pn == bpn) | |
717 | c->pn = apn; | |
718 | } | |
719 | ndoms--; /* one less element */ | |
720 | goto restart; | |
721 | } | |
722 | } | |
723 | } | |
724 | ||
cf417141 MK |
725 | /* |
726 | * Now we know how many domains to create. | |
727 | * Convert <csn, csa> to <ndoms, doms> and populate cpu masks. | |
728 | */ | |
6af866af | 729 | doms = kmalloc(ndoms * cpumask_size(), GFP_KERNEL); |
700018e0 | 730 | if (!doms) |
cf417141 | 731 | goto done; |
cf417141 MK |
732 | |
733 | /* | |
734 | * The rest of the code, including the scheduler, can deal with | |
735 | * dattr==NULL case. No need to abort if alloc fails. | |
736 | */ | |
1d3504fc | 737 | dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL); |
029190c5 PJ |
738 | |
739 | for (nslot = 0, i = 0; i < csn; i++) { | |
740 | struct cpuset *a = csa[i]; | |
6af866af | 741 | struct cpumask *dp; |
029190c5 PJ |
742 | int apn = a->pn; |
743 | ||
cf417141 MK |
744 | if (apn < 0) { |
745 | /* Skip completed partitions */ | |
746 | continue; | |
747 | } | |
748 | ||
749 | dp = doms + nslot; | |
750 | ||
751 | if (nslot == ndoms) { | |
752 | static int warnings = 10; | |
753 | if (warnings) { | |
754 | printk(KERN_WARNING | |
755 | "rebuild_sched_domains confused:" | |
756 | " nslot %d, ndoms %d, csn %d, i %d," | |
757 | " apn %d\n", | |
758 | nslot, ndoms, csn, i, apn); | |
759 | warnings--; | |
029190c5 | 760 | } |
cf417141 MK |
761 | continue; |
762 | } | |
029190c5 | 763 | |
6af866af | 764 | cpumask_clear(dp); |
cf417141 MK |
765 | if (dattr) |
766 | *(dattr + nslot) = SD_ATTR_INIT; | |
767 | for (j = i; j < csn; j++) { | |
768 | struct cpuset *b = csa[j]; | |
769 | ||
770 | if (apn == b->pn) { | |
300ed6cb | 771 | cpumask_or(dp, dp, b->cpus_allowed); |
cf417141 MK |
772 | if (dattr) |
773 | update_domain_attr_tree(dattr + nslot, b); | |
774 | ||
775 | /* Done with this partition */ | |
776 | b->pn = -1; | |
029190c5 | 777 | } |
029190c5 | 778 | } |
cf417141 | 779 | nslot++; |
029190c5 PJ |
780 | } |
781 | BUG_ON(nslot != ndoms); | |
782 | ||
cf417141 MK |
783 | done: |
784 | kfree(csa); | |
785 | ||
700018e0 LZ |
786 | /* |
787 | * Fallback to the default domain if kmalloc() failed. | |
788 | * See comments in partition_sched_domains(). | |
789 | */ | |
790 | if (doms == NULL) | |
791 | ndoms = 1; | |
792 | ||
cf417141 MK |
793 | *domains = doms; |
794 | *attributes = dattr; | |
795 | return ndoms; | |
796 | } | |
797 | ||
798 | /* | |
799 | * Rebuild scheduler domains. | |
800 | * | |
801 | * Call with neither cgroup_mutex held nor within get_online_cpus(). | |
802 | * Takes both cgroup_mutex and get_online_cpus(). | |
803 | * | |
804 | * Cannot be directly called from cpuset code handling changes | |
805 | * to the cpuset pseudo-filesystem, because it cannot be called | |
806 | * from code that already holds cgroup_mutex. | |
807 | */ | |
808 | static void do_rebuild_sched_domains(struct work_struct *unused) | |
809 | { | |
810 | struct sched_domain_attr *attr; | |
6af866af | 811 | struct cpumask *doms; |
cf417141 MK |
812 | int ndoms; |
813 | ||
86ef5c9a | 814 | get_online_cpus(); |
cf417141 MK |
815 | |
816 | /* Generate domain masks and attrs */ | |
817 | cgroup_lock(); | |
818 | ndoms = generate_sched_domains(&doms, &attr); | |
819 | cgroup_unlock(); | |
820 | ||
821 | /* Have scheduler rebuild the domains */ | |
822 | partition_sched_domains(ndoms, doms, attr); | |
823 | ||
86ef5c9a | 824 | put_online_cpus(); |
cf417141 | 825 | } |
db7f47cf PM |
826 | #else /* !CONFIG_SMP */ |
827 | static void do_rebuild_sched_domains(struct work_struct *unused) | |
828 | { | |
829 | } | |
830 | ||
831 | static int generate_sched_domains(struct cpumask **domains, | |
832 | struct sched_domain_attr **attributes) | |
833 | { | |
834 | *domains = NULL; | |
835 | return 1; | |
836 | } | |
837 | #endif /* CONFIG_SMP */ | |
029190c5 | 838 | |
cf417141 MK |
839 | static DECLARE_WORK(rebuild_sched_domains_work, do_rebuild_sched_domains); |
840 | ||
841 | /* | |
842 | * Rebuild scheduler domains, asynchronously via workqueue. | |
843 | * | |
844 | * If the flag 'sched_load_balance' of any cpuset with non-empty | |
845 | * 'cpus' changes, or if the 'cpus' allowed changes in any cpuset | |
846 | * which has that flag enabled, or if any cpuset with a non-empty | |
847 | * 'cpus' is removed, then call this routine to rebuild the | |
848 | * scheduler's dynamic sched domains. | |
849 | * | |
850 | * The rebuild_sched_domains() and partition_sched_domains() | |
851 | * routines must nest cgroup_lock() inside get_online_cpus(), | |
852 | * but such cpuset changes as these must nest that locking the | |
853 | * other way, holding cgroup_lock() for much of the code. | |
854 | * | |
855 | * So in order to avoid an ABBA deadlock, the cpuset code handling | |
856 | * these user changes delegates the actual sched domain rebuilding | |
857 | * to a separate workqueue thread, which ends up processing the | |
858 | * above do_rebuild_sched_domains() function. | |
859 | */ | |
860 | static void async_rebuild_sched_domains(void) | |
861 | { | |
f90d4118 | 862 | queue_work(cpuset_wq, &rebuild_sched_domains_work); |
cf417141 MK |
863 | } |
864 | ||
865 | /* | |
866 | * Accomplishes the same scheduler domain rebuild as the above | |
867 | * async_rebuild_sched_domains(), however it directly calls the | |
868 | * rebuild routine synchronously rather than calling it via an | |
869 | * asynchronous work thread. | |
870 | * | |
871 | * This can only be called from code that is not holding | |
872 | * cgroup_mutex (not nested in a cgroup_lock() call.) | |
873 | */ | |
874 | void rebuild_sched_domains(void) | |
875 | { | |
876 | do_rebuild_sched_domains(NULL); | |
029190c5 PJ |
877 | } |
878 | ||
58f4790b CW |
879 | /** |
880 | * cpuset_test_cpumask - test a task's cpus_allowed versus its cpuset's | |
881 | * @tsk: task to test | |
882 | * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner | |
883 | * | |
2df167a3 | 884 | * Call with cgroup_mutex held. May take callback_mutex during call. |
58f4790b CW |
885 | * Called for each task in a cgroup by cgroup_scan_tasks(). |
886 | * Return nonzero if this tasks's cpus_allowed mask should be changed (in other | |
887 | * words, if its mask is not equal to its cpuset's mask). | |
053199ed | 888 | */ |
9e0c914c AB |
889 | static int cpuset_test_cpumask(struct task_struct *tsk, |
890 | struct cgroup_scanner *scan) | |
58f4790b | 891 | { |
300ed6cb | 892 | return !cpumask_equal(&tsk->cpus_allowed, |
58f4790b CW |
893 | (cgroup_cs(scan->cg))->cpus_allowed); |
894 | } | |
053199ed | 895 | |
58f4790b CW |
896 | /** |
897 | * cpuset_change_cpumask - make a task's cpus_allowed the same as its cpuset's | |
898 | * @tsk: task to test | |
899 | * @scan: struct cgroup_scanner containing the cgroup of the task | |
900 | * | |
901 | * Called by cgroup_scan_tasks() for each task in a cgroup whose | |
902 | * cpus_allowed mask needs to be changed. | |
903 | * | |
904 | * We don't need to re-check for the cgroup/cpuset membership, since we're | |
905 | * holding cgroup_lock() at this point. | |
906 | */ | |
9e0c914c AB |
907 | static void cpuset_change_cpumask(struct task_struct *tsk, |
908 | struct cgroup_scanner *scan) | |
58f4790b | 909 | { |
300ed6cb | 910 | set_cpus_allowed_ptr(tsk, ((cgroup_cs(scan->cg))->cpus_allowed)); |
58f4790b CW |
911 | } |
912 | ||
0b2f630a MX |
913 | /** |
914 | * update_tasks_cpumask - Update the cpumasks of tasks in the cpuset. | |
915 | * @cs: the cpuset in which each task's cpus_allowed mask needs to be changed | |
4e74339a | 916 | * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks() |
0b2f630a MX |
917 | * |
918 | * Called with cgroup_mutex held | |
919 | * | |
920 | * The cgroup_scan_tasks() function will scan all the tasks in a cgroup, | |
921 | * calling callback functions for each. | |
922 | * | |
4e74339a LZ |
923 | * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0 |
924 | * if @heap != NULL. | |
0b2f630a | 925 | */ |
4e74339a | 926 | static void update_tasks_cpumask(struct cpuset *cs, struct ptr_heap *heap) |
0b2f630a MX |
927 | { |
928 | struct cgroup_scanner scan; | |
0b2f630a MX |
929 | |
930 | scan.cg = cs->css.cgroup; | |
931 | scan.test_task = cpuset_test_cpumask; | |
932 | scan.process_task = cpuset_change_cpumask; | |
4e74339a LZ |
933 | scan.heap = heap; |
934 | cgroup_scan_tasks(&scan); | |
0b2f630a MX |
935 | } |
936 | ||
58f4790b CW |
937 | /** |
938 | * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it | |
939 | * @cs: the cpuset to consider | |
940 | * @buf: buffer of cpu numbers written to this cpuset | |
941 | */ | |
645fcc9d LZ |
942 | static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs, |
943 | const char *buf) | |
1da177e4 | 944 | { |
4e74339a | 945 | struct ptr_heap heap; |
58f4790b CW |
946 | int retval; |
947 | int is_load_balanced; | |
1da177e4 | 948 | |
4c4d50f7 PJ |
949 | /* top_cpuset.cpus_allowed tracks cpu_online_map; it's read-only */ |
950 | if (cs == &top_cpuset) | |
951 | return -EACCES; | |
952 | ||
6f7f02e7 | 953 | /* |
c8d9c90c | 954 | * An empty cpus_allowed is ok only if the cpuset has no tasks. |
020958b6 PJ |
955 | * Since cpulist_parse() fails on an empty mask, we special case |
956 | * that parsing. The validate_change() call ensures that cpusets | |
957 | * with tasks have cpus. | |
6f7f02e7 | 958 | */ |
020958b6 | 959 | if (!*buf) { |
300ed6cb | 960 | cpumask_clear(trialcs->cpus_allowed); |
6f7f02e7 | 961 | } else { |
300ed6cb | 962 | retval = cpulist_parse(buf, trialcs->cpus_allowed); |
6f7f02e7 DR |
963 | if (retval < 0) |
964 | return retval; | |
37340746 | 965 | |
300ed6cb | 966 | if (!cpumask_subset(trialcs->cpus_allowed, cpu_online_mask)) |
37340746 | 967 | return -EINVAL; |
6f7f02e7 | 968 | } |
645fcc9d | 969 | retval = validate_change(cs, trialcs); |
85d7b949 DG |
970 | if (retval < 0) |
971 | return retval; | |
029190c5 | 972 | |
8707d8b8 | 973 | /* Nothing to do if the cpus didn't change */ |
300ed6cb | 974 | if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed)) |
8707d8b8 | 975 | return 0; |
58f4790b | 976 | |
4e74339a LZ |
977 | retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL); |
978 | if (retval) | |
979 | return retval; | |
980 | ||
645fcc9d | 981 | is_load_balanced = is_sched_load_balance(trialcs); |
029190c5 | 982 | |
3d3f26a7 | 983 | mutex_lock(&callback_mutex); |
300ed6cb | 984 | cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed); |
3d3f26a7 | 985 | mutex_unlock(&callback_mutex); |
029190c5 | 986 | |
8707d8b8 PM |
987 | /* |
988 | * Scan tasks in the cpuset, and update the cpumasks of any | |
58f4790b | 989 | * that need an update. |
8707d8b8 | 990 | */ |
4e74339a LZ |
991 | update_tasks_cpumask(cs, &heap); |
992 | ||
993 | heap_free(&heap); | |
58f4790b | 994 | |
8707d8b8 | 995 | if (is_load_balanced) |
cf417141 | 996 | async_rebuild_sched_domains(); |
85d7b949 | 997 | return 0; |
1da177e4 LT |
998 | } |
999 | ||
e4e364e8 PJ |
1000 | /* |
1001 | * cpuset_migrate_mm | |
1002 | * | |
1003 | * Migrate memory region from one set of nodes to another. | |
1004 | * | |
1005 | * Temporarilly set tasks mems_allowed to target nodes of migration, | |
1006 | * so that the migration code can allocate pages on these nodes. | |
1007 | * | |
2df167a3 | 1008 | * Call holding cgroup_mutex, so current's cpuset won't change |
c8d9c90c | 1009 | * during this call, as manage_mutex holds off any cpuset_attach() |
e4e364e8 PJ |
1010 | * calls. Therefore we don't need to take task_lock around the |
1011 | * call to guarantee_online_mems(), as we know no one is changing | |
2df167a3 | 1012 | * our task's cpuset. |
e4e364e8 PJ |
1013 | * |
1014 | * Hold callback_mutex around the two modifications of our tasks | |
1015 | * mems_allowed to synchronize with cpuset_mems_allowed(). | |
1016 | * | |
1017 | * While the mm_struct we are migrating is typically from some | |
1018 | * other task, the task_struct mems_allowed that we are hacking | |
1019 | * is for our current task, which must allocate new pages for that | |
1020 | * migrating memory region. | |
1021 | * | |
1022 | * We call cpuset_update_task_memory_state() before hacking | |
1023 | * our tasks mems_allowed, so that we are assured of being in | |
1024 | * sync with our tasks cpuset, and in particular, callbacks to | |
1025 | * cpuset_update_task_memory_state() from nested page allocations | |
1026 | * won't see any mismatch of our cpuset and task mems_generation | |
1027 | * values, so won't overwrite our hacked tasks mems_allowed | |
1028 | * nodemask. | |
1029 | */ | |
1030 | ||
1031 | static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from, | |
1032 | const nodemask_t *to) | |
1033 | { | |
1034 | struct task_struct *tsk = current; | |
1035 | ||
1036 | cpuset_update_task_memory_state(); | |
1037 | ||
1038 | mutex_lock(&callback_mutex); | |
1039 | tsk->mems_allowed = *to; | |
1040 | mutex_unlock(&callback_mutex); | |
1041 | ||
1042 | do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL); | |
1043 | ||
1044 | mutex_lock(&callback_mutex); | |
8793d854 | 1045 | guarantee_online_mems(task_cs(tsk),&tsk->mems_allowed); |
e4e364e8 PJ |
1046 | mutex_unlock(&callback_mutex); |
1047 | } | |
1048 | ||
3b6766fe LZ |
1049 | /* |
1050 | * Rebind task's vmas to cpuset's new mems_allowed, and migrate pages to new | |
1051 | * nodes if memory_migrate flag is set. Called with cgroup_mutex held. | |
1052 | */ | |
1053 | static void cpuset_change_nodemask(struct task_struct *p, | |
1054 | struct cgroup_scanner *scan) | |
1055 | { | |
1056 | struct mm_struct *mm; | |
1057 | struct cpuset *cs; | |
1058 | int migrate; | |
1059 | const nodemask_t *oldmem = scan->data; | |
1060 | ||
1061 | mm = get_task_mm(p); | |
1062 | if (!mm) | |
1063 | return; | |
1064 | ||
1065 | cs = cgroup_cs(scan->cg); | |
1066 | migrate = is_memory_migrate(cs); | |
1067 | ||
1068 | mpol_rebind_mm(mm, &cs->mems_allowed); | |
1069 | if (migrate) | |
1070 | cpuset_migrate_mm(mm, oldmem, &cs->mems_allowed); | |
1071 | mmput(mm); | |
1072 | } | |
1073 | ||
8793d854 PM |
1074 | static void *cpuset_being_rebound; |
1075 | ||
0b2f630a MX |
1076 | /** |
1077 | * update_tasks_nodemask - Update the nodemasks of tasks in the cpuset. | |
1078 | * @cs: the cpuset in which each task's mems_allowed mask needs to be changed | |
1079 | * @oldmem: old mems_allowed of cpuset cs | |
010cfac4 | 1080 | * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks() |
0b2f630a MX |
1081 | * |
1082 | * Called with cgroup_mutex held | |
010cfac4 LZ |
1083 | * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0 |
1084 | * if @heap != NULL. | |
0b2f630a | 1085 | */ |
010cfac4 LZ |
1086 | static void update_tasks_nodemask(struct cpuset *cs, const nodemask_t *oldmem, |
1087 | struct ptr_heap *heap) | |
1da177e4 | 1088 | { |
3b6766fe | 1089 | struct cgroup_scanner scan; |
59dac16f | 1090 | |
846a16bf | 1091 | cpuset_being_rebound = cs; /* causes mpol_dup() rebind */ |
4225399a | 1092 | |
3b6766fe LZ |
1093 | scan.cg = cs->css.cgroup; |
1094 | scan.test_task = NULL; | |
1095 | scan.process_task = cpuset_change_nodemask; | |
010cfac4 | 1096 | scan.heap = heap; |
3b6766fe | 1097 | scan.data = (nodemask_t *)oldmem; |
4225399a PJ |
1098 | |
1099 | /* | |
3b6766fe LZ |
1100 | * The mpol_rebind_mm() call takes mmap_sem, which we couldn't |
1101 | * take while holding tasklist_lock. Forks can happen - the | |
1102 | * mpol_dup() cpuset_being_rebound check will catch such forks, | |
1103 | * and rebind their vma mempolicies too. Because we still hold | |
1104 | * the global cgroup_mutex, we know that no other rebind effort | |
1105 | * will be contending for the global variable cpuset_being_rebound. | |
4225399a | 1106 | * It's ok if we rebind the same mm twice; mpol_rebind_mm() |
04c19fa6 | 1107 | * is idempotent. Also migrate pages in each mm to new nodes. |
4225399a | 1108 | */ |
010cfac4 | 1109 | cgroup_scan_tasks(&scan); |
4225399a | 1110 | |
2df167a3 | 1111 | /* We're done rebinding vmas to this cpuset's new mems_allowed. */ |
8793d854 | 1112 | cpuset_being_rebound = NULL; |
1da177e4 LT |
1113 | } |
1114 | ||
0b2f630a MX |
1115 | /* |
1116 | * Handle user request to change the 'mems' memory placement | |
1117 | * of a cpuset. Needs to validate the request, update the | |
1118 | * cpusets mems_allowed and mems_generation, and for each | |
1119 | * task in the cpuset, rebind any vma mempolicies and if | |
1120 | * the cpuset is marked 'memory_migrate', migrate the tasks | |
1121 | * pages to the new memory. | |
1122 | * | |
1123 | * Call with cgroup_mutex held. May take callback_mutex during call. | |
1124 | * Will take tasklist_lock, scan tasklist for tasks in cpuset cs, | |
1125 | * lock each such tasks mm->mmap_sem, scan its vma's and rebind | |
1126 | * their mempolicies to the cpusets new mems_allowed. | |
1127 | */ | |
645fcc9d LZ |
1128 | static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs, |
1129 | const char *buf) | |
0b2f630a | 1130 | { |
0b2f630a MX |
1131 | nodemask_t oldmem; |
1132 | int retval; | |
010cfac4 | 1133 | struct ptr_heap heap; |
0b2f630a MX |
1134 | |
1135 | /* | |
1136 | * top_cpuset.mems_allowed tracks node_stats[N_HIGH_MEMORY]; | |
1137 | * it's read-only | |
1138 | */ | |
1139 | if (cs == &top_cpuset) | |
1140 | return -EACCES; | |
1141 | ||
0b2f630a MX |
1142 | /* |
1143 | * An empty mems_allowed is ok iff there are no tasks in the cpuset. | |
1144 | * Since nodelist_parse() fails on an empty mask, we special case | |
1145 | * that parsing. The validate_change() call ensures that cpusets | |
1146 | * with tasks have memory. | |
1147 | */ | |
1148 | if (!*buf) { | |
645fcc9d | 1149 | nodes_clear(trialcs->mems_allowed); |
0b2f630a | 1150 | } else { |
645fcc9d | 1151 | retval = nodelist_parse(buf, trialcs->mems_allowed); |
0b2f630a MX |
1152 | if (retval < 0) |
1153 | goto done; | |
1154 | ||
645fcc9d | 1155 | if (!nodes_subset(trialcs->mems_allowed, |
0b2f630a MX |
1156 | node_states[N_HIGH_MEMORY])) |
1157 | return -EINVAL; | |
1158 | } | |
1159 | oldmem = cs->mems_allowed; | |
645fcc9d | 1160 | if (nodes_equal(oldmem, trialcs->mems_allowed)) { |
0b2f630a MX |
1161 | retval = 0; /* Too easy - nothing to do */ |
1162 | goto done; | |
1163 | } | |
645fcc9d | 1164 | retval = validate_change(cs, trialcs); |
0b2f630a MX |
1165 | if (retval < 0) |
1166 | goto done; | |
1167 | ||
010cfac4 LZ |
1168 | retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL); |
1169 | if (retval < 0) | |
1170 | goto done; | |
1171 | ||
0b2f630a | 1172 | mutex_lock(&callback_mutex); |
645fcc9d | 1173 | cs->mems_allowed = trialcs->mems_allowed; |
0b2f630a MX |
1174 | cs->mems_generation = cpuset_mems_generation++; |
1175 | mutex_unlock(&callback_mutex); | |
1176 | ||
010cfac4 LZ |
1177 | update_tasks_nodemask(cs, &oldmem, &heap); |
1178 | ||
1179 | heap_free(&heap); | |
0b2f630a MX |
1180 | done: |
1181 | return retval; | |
1182 | } | |
1183 | ||
8793d854 PM |
1184 | int current_cpuset_is_being_rebound(void) |
1185 | { | |
1186 | return task_cs(current) == cpuset_being_rebound; | |
1187 | } | |
1188 | ||
5be7a479 | 1189 | static int update_relax_domain_level(struct cpuset *cs, s64 val) |
1d3504fc | 1190 | { |
db7f47cf | 1191 | #ifdef CONFIG_SMP |
30e0e178 LZ |
1192 | if (val < -1 || val >= SD_LV_MAX) |
1193 | return -EINVAL; | |
db7f47cf | 1194 | #endif |
1d3504fc HS |
1195 | |
1196 | if (val != cs->relax_domain_level) { | |
1197 | cs->relax_domain_level = val; | |
300ed6cb LZ |
1198 | if (!cpumask_empty(cs->cpus_allowed) && |
1199 | is_sched_load_balance(cs)) | |
cf417141 | 1200 | async_rebuild_sched_domains(); |
1d3504fc HS |
1201 | } |
1202 | ||
1203 | return 0; | |
1204 | } | |
1205 | ||
1da177e4 LT |
1206 | /* |
1207 | * update_flag - read a 0 or a 1 in a file and update associated flag | |
78608366 PM |
1208 | * bit: the bit to update (see cpuset_flagbits_t) |
1209 | * cs: the cpuset to update | |
1210 | * turning_on: whether the flag is being set or cleared | |
053199ed | 1211 | * |
2df167a3 | 1212 | * Call with cgroup_mutex held. |
1da177e4 LT |
1213 | */ |
1214 | ||
700fe1ab PM |
1215 | static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs, |
1216 | int turning_on) | |
1da177e4 | 1217 | { |
645fcc9d | 1218 | struct cpuset *trialcs; |
607717a6 | 1219 | int err; |
40b6a762 | 1220 | int balance_flag_changed; |
1da177e4 | 1221 | |
645fcc9d LZ |
1222 | trialcs = alloc_trial_cpuset(cs); |
1223 | if (!trialcs) | |
1224 | return -ENOMEM; | |
1225 | ||
1da177e4 | 1226 | if (turning_on) |
645fcc9d | 1227 | set_bit(bit, &trialcs->flags); |
1da177e4 | 1228 | else |
645fcc9d | 1229 | clear_bit(bit, &trialcs->flags); |
1da177e4 | 1230 | |
645fcc9d | 1231 | err = validate_change(cs, trialcs); |
85d7b949 | 1232 | if (err < 0) |
645fcc9d | 1233 | goto out; |
029190c5 | 1234 | |
029190c5 | 1235 | balance_flag_changed = (is_sched_load_balance(cs) != |
645fcc9d | 1236 | is_sched_load_balance(trialcs)); |
029190c5 | 1237 | |
3d3f26a7 | 1238 | mutex_lock(&callback_mutex); |
645fcc9d | 1239 | cs->flags = trialcs->flags; |
3d3f26a7 | 1240 | mutex_unlock(&callback_mutex); |
85d7b949 | 1241 | |
300ed6cb | 1242 | if (!cpumask_empty(trialcs->cpus_allowed) && balance_flag_changed) |
cf417141 | 1243 | async_rebuild_sched_domains(); |
029190c5 | 1244 | |
645fcc9d LZ |
1245 | out: |
1246 | free_trial_cpuset(trialcs); | |
1247 | return err; | |
1da177e4 LT |
1248 | } |
1249 | ||
3e0d98b9 | 1250 | /* |
80f7228b | 1251 | * Frequency meter - How fast is some event occurring? |
3e0d98b9 PJ |
1252 | * |
1253 | * These routines manage a digitally filtered, constant time based, | |
1254 | * event frequency meter. There are four routines: | |
1255 | * fmeter_init() - initialize a frequency meter. | |
1256 | * fmeter_markevent() - called each time the event happens. | |
1257 | * fmeter_getrate() - returns the recent rate of such events. | |
1258 | * fmeter_update() - internal routine used to update fmeter. | |
1259 | * | |
1260 | * A common data structure is passed to each of these routines, | |
1261 | * which is used to keep track of the state required to manage the | |
1262 | * frequency meter and its digital filter. | |
1263 | * | |
1264 | * The filter works on the number of events marked per unit time. | |
1265 | * The filter is single-pole low-pass recursive (IIR). The time unit | |
1266 | * is 1 second. Arithmetic is done using 32-bit integers scaled to | |
1267 | * simulate 3 decimal digits of precision (multiplied by 1000). | |
1268 | * | |
1269 | * With an FM_COEF of 933, and a time base of 1 second, the filter | |
1270 | * has a half-life of 10 seconds, meaning that if the events quit | |
1271 | * happening, then the rate returned from the fmeter_getrate() | |
1272 | * will be cut in half each 10 seconds, until it converges to zero. | |
1273 | * | |
1274 | * It is not worth doing a real infinitely recursive filter. If more | |
1275 | * than FM_MAXTICKS ticks have elapsed since the last filter event, | |
1276 | * just compute FM_MAXTICKS ticks worth, by which point the level | |
1277 | * will be stable. | |
1278 | * | |
1279 | * Limit the count of unprocessed events to FM_MAXCNT, so as to avoid | |
1280 | * arithmetic overflow in the fmeter_update() routine. | |
1281 | * | |
1282 | * Given the simple 32 bit integer arithmetic used, this meter works | |
1283 | * best for reporting rates between one per millisecond (msec) and | |
1284 | * one per 32 (approx) seconds. At constant rates faster than one | |
1285 | * per msec it maxes out at values just under 1,000,000. At constant | |
1286 | * rates between one per msec, and one per second it will stabilize | |
1287 | * to a value N*1000, where N is the rate of events per second. | |
1288 | * At constant rates between one per second and one per 32 seconds, | |
1289 | * it will be choppy, moving up on the seconds that have an event, | |
1290 | * and then decaying until the next event. At rates slower than | |
1291 | * about one in 32 seconds, it decays all the way back to zero between | |
1292 | * each event. | |
1293 | */ | |
1294 | ||
1295 | #define FM_COEF 933 /* coefficient for half-life of 10 secs */ | |
1296 | #define FM_MAXTICKS ((time_t)99) /* useless computing more ticks than this */ | |
1297 | #define FM_MAXCNT 1000000 /* limit cnt to avoid overflow */ | |
1298 | #define FM_SCALE 1000 /* faux fixed point scale */ | |
1299 | ||
1300 | /* Initialize a frequency meter */ | |
1301 | static void fmeter_init(struct fmeter *fmp) | |
1302 | { | |
1303 | fmp->cnt = 0; | |
1304 | fmp->val = 0; | |
1305 | fmp->time = 0; | |
1306 | spin_lock_init(&fmp->lock); | |
1307 | } | |
1308 | ||
1309 | /* Internal meter update - process cnt events and update value */ | |
1310 | static void fmeter_update(struct fmeter *fmp) | |
1311 | { | |
1312 | time_t now = get_seconds(); | |
1313 | time_t ticks = now - fmp->time; | |
1314 | ||
1315 | if (ticks == 0) | |
1316 | return; | |
1317 | ||
1318 | ticks = min(FM_MAXTICKS, ticks); | |
1319 | while (ticks-- > 0) | |
1320 | fmp->val = (FM_COEF * fmp->val) / FM_SCALE; | |
1321 | fmp->time = now; | |
1322 | ||
1323 | fmp->val += ((FM_SCALE - FM_COEF) * fmp->cnt) / FM_SCALE; | |
1324 | fmp->cnt = 0; | |
1325 | } | |
1326 | ||
1327 | /* Process any previous ticks, then bump cnt by one (times scale). */ | |
1328 | static void fmeter_markevent(struct fmeter *fmp) | |
1329 | { | |
1330 | spin_lock(&fmp->lock); | |
1331 | fmeter_update(fmp); | |
1332 | fmp->cnt = min(FM_MAXCNT, fmp->cnt + FM_SCALE); | |
1333 | spin_unlock(&fmp->lock); | |
1334 | } | |
1335 | ||
1336 | /* Process any previous ticks, then return current value. */ | |
1337 | static int fmeter_getrate(struct fmeter *fmp) | |
1338 | { | |
1339 | int val; | |
1340 | ||
1341 | spin_lock(&fmp->lock); | |
1342 | fmeter_update(fmp); | |
1343 | val = fmp->val; | |
1344 | spin_unlock(&fmp->lock); | |
1345 | return val; | |
1346 | } | |
1347 | ||
2341d1b6 LZ |
1348 | /* Protected by cgroup_lock */ |
1349 | static cpumask_var_t cpus_attach; | |
1350 | ||
2df167a3 | 1351 | /* Called by cgroups to determine if a cpuset is usable; cgroup_mutex held */ |
8793d854 PM |
1352 | static int cpuset_can_attach(struct cgroup_subsys *ss, |
1353 | struct cgroup *cont, struct task_struct *tsk) | |
1da177e4 | 1354 | { |
8793d854 | 1355 | struct cpuset *cs = cgroup_cs(cont); |
1da177e4 | 1356 | |
300ed6cb | 1357 | if (cpumask_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed)) |
1da177e4 | 1358 | return -ENOSPC; |
9985b0ba | 1359 | |
6d7b2f5f DR |
1360 | /* |
1361 | * Kthreads bound to specific cpus cannot be moved to a new cpuset; we | |
1362 | * cannot change their cpu affinity and isolating such threads by their | |
1363 | * set of allowed nodes is unnecessary. Thus, cpusets are not | |
1364 | * applicable for such threads. This prevents checking for success of | |
1365 | * set_cpus_allowed_ptr() on all attached tasks before cpus_allowed may | |
1366 | * be changed. | |
1367 | */ | |
1368 | if (tsk->flags & PF_THREAD_BOUND) | |
1369 | return -EINVAL; | |
1da177e4 | 1370 | |
6d7b2f5f | 1371 | return security_task_setscheduler(tsk, 0, NULL); |
8793d854 | 1372 | } |
1da177e4 | 1373 | |
8793d854 PM |
1374 | static void cpuset_attach(struct cgroup_subsys *ss, |
1375 | struct cgroup *cont, struct cgroup *oldcont, | |
1376 | struct task_struct *tsk) | |
1377 | { | |
8793d854 PM |
1378 | nodemask_t from, to; |
1379 | struct mm_struct *mm; | |
1380 | struct cpuset *cs = cgroup_cs(cont); | |
1381 | struct cpuset *oldcs = cgroup_cs(oldcont); | |
9985b0ba | 1382 | int err; |
22fb52dd | 1383 | |
f5813d94 | 1384 | if (cs == &top_cpuset) { |
2341d1b6 | 1385 | cpumask_copy(cpus_attach, cpu_possible_mask); |
f5813d94 MX |
1386 | } else { |
1387 | mutex_lock(&callback_mutex); | |
2341d1b6 | 1388 | guarantee_online_cpus(cs, cpus_attach); |
f5813d94 MX |
1389 | mutex_unlock(&callback_mutex); |
1390 | } | |
2341d1b6 | 1391 | err = set_cpus_allowed_ptr(tsk, cpus_attach); |
9985b0ba DR |
1392 | if (err) |
1393 | return; | |
1da177e4 | 1394 | |
45b07ef3 PJ |
1395 | from = oldcs->mems_allowed; |
1396 | to = cs->mems_allowed; | |
4225399a PJ |
1397 | mm = get_task_mm(tsk); |
1398 | if (mm) { | |
1399 | mpol_rebind_mm(mm, &to); | |
2741a559 | 1400 | if (is_memory_migrate(cs)) |
e4e364e8 | 1401 | cpuset_migrate_mm(mm, &from, &to); |
4225399a PJ |
1402 | mmput(mm); |
1403 | } | |
1da177e4 LT |
1404 | } |
1405 | ||
1406 | /* The various types of files and directories in a cpuset file system */ | |
1407 | ||
1408 | typedef enum { | |
45b07ef3 | 1409 | FILE_MEMORY_MIGRATE, |
1da177e4 LT |
1410 | FILE_CPULIST, |
1411 | FILE_MEMLIST, | |
1412 | FILE_CPU_EXCLUSIVE, | |
1413 | FILE_MEM_EXCLUSIVE, | |
78608366 | 1414 | FILE_MEM_HARDWALL, |
029190c5 | 1415 | FILE_SCHED_LOAD_BALANCE, |
1d3504fc | 1416 | FILE_SCHED_RELAX_DOMAIN_LEVEL, |
3e0d98b9 PJ |
1417 | FILE_MEMORY_PRESSURE_ENABLED, |
1418 | FILE_MEMORY_PRESSURE, | |
825a46af PJ |
1419 | FILE_SPREAD_PAGE, |
1420 | FILE_SPREAD_SLAB, | |
1da177e4 LT |
1421 | } cpuset_filetype_t; |
1422 | ||
700fe1ab PM |
1423 | static int cpuset_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val) |
1424 | { | |
1425 | int retval = 0; | |
1426 | struct cpuset *cs = cgroup_cs(cgrp); | |
1427 | cpuset_filetype_t type = cft->private; | |
1428 | ||
e3712395 | 1429 | if (!cgroup_lock_live_group(cgrp)) |
700fe1ab | 1430 | return -ENODEV; |
700fe1ab PM |
1431 | |
1432 | switch (type) { | |
1da177e4 | 1433 | case FILE_CPU_EXCLUSIVE: |
700fe1ab | 1434 | retval = update_flag(CS_CPU_EXCLUSIVE, cs, val); |
1da177e4 LT |
1435 | break; |
1436 | case FILE_MEM_EXCLUSIVE: | |
700fe1ab | 1437 | retval = update_flag(CS_MEM_EXCLUSIVE, cs, val); |
1da177e4 | 1438 | break; |
78608366 PM |
1439 | case FILE_MEM_HARDWALL: |
1440 | retval = update_flag(CS_MEM_HARDWALL, cs, val); | |
1441 | break; | |
029190c5 | 1442 | case FILE_SCHED_LOAD_BALANCE: |
700fe1ab | 1443 | retval = update_flag(CS_SCHED_LOAD_BALANCE, cs, val); |
1d3504fc | 1444 | break; |
45b07ef3 | 1445 | case FILE_MEMORY_MIGRATE: |
700fe1ab | 1446 | retval = update_flag(CS_MEMORY_MIGRATE, cs, val); |
45b07ef3 | 1447 | break; |
3e0d98b9 | 1448 | case FILE_MEMORY_PRESSURE_ENABLED: |
700fe1ab | 1449 | cpuset_memory_pressure_enabled = !!val; |
3e0d98b9 PJ |
1450 | break; |
1451 | case FILE_MEMORY_PRESSURE: | |
1452 | retval = -EACCES; | |
1453 | break; | |
825a46af | 1454 | case FILE_SPREAD_PAGE: |
700fe1ab | 1455 | retval = update_flag(CS_SPREAD_PAGE, cs, val); |
151a4420 | 1456 | cs->mems_generation = cpuset_mems_generation++; |
825a46af PJ |
1457 | break; |
1458 | case FILE_SPREAD_SLAB: | |
700fe1ab | 1459 | retval = update_flag(CS_SPREAD_SLAB, cs, val); |
151a4420 | 1460 | cs->mems_generation = cpuset_mems_generation++; |
825a46af | 1461 | break; |
1da177e4 LT |
1462 | default: |
1463 | retval = -EINVAL; | |
700fe1ab | 1464 | break; |
1da177e4 | 1465 | } |
8793d854 | 1466 | cgroup_unlock(); |
1da177e4 LT |
1467 | return retval; |
1468 | } | |
1469 | ||
5be7a479 PM |
1470 | static int cpuset_write_s64(struct cgroup *cgrp, struct cftype *cft, s64 val) |
1471 | { | |
1472 | int retval = 0; | |
1473 | struct cpuset *cs = cgroup_cs(cgrp); | |
1474 | cpuset_filetype_t type = cft->private; | |
1475 | ||
e3712395 | 1476 | if (!cgroup_lock_live_group(cgrp)) |
5be7a479 | 1477 | return -ENODEV; |
e3712395 | 1478 | |
5be7a479 PM |
1479 | switch (type) { |
1480 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1481 | retval = update_relax_domain_level(cs, val); | |
1482 | break; | |
1483 | default: | |
1484 | retval = -EINVAL; | |
1485 | break; | |
1486 | } | |
1487 | cgroup_unlock(); | |
1488 | return retval; | |
1489 | } | |
1490 | ||
e3712395 PM |
1491 | /* |
1492 | * Common handling for a write to a "cpus" or "mems" file. | |
1493 | */ | |
1494 | static int cpuset_write_resmask(struct cgroup *cgrp, struct cftype *cft, | |
1495 | const char *buf) | |
1496 | { | |
1497 | int retval = 0; | |
645fcc9d LZ |
1498 | struct cpuset *cs = cgroup_cs(cgrp); |
1499 | struct cpuset *trialcs; | |
e3712395 PM |
1500 | |
1501 | if (!cgroup_lock_live_group(cgrp)) | |
1502 | return -ENODEV; | |
1503 | ||
645fcc9d LZ |
1504 | trialcs = alloc_trial_cpuset(cs); |
1505 | if (!trialcs) | |
1506 | return -ENOMEM; | |
1507 | ||
e3712395 PM |
1508 | switch (cft->private) { |
1509 | case FILE_CPULIST: | |
645fcc9d | 1510 | retval = update_cpumask(cs, trialcs, buf); |
e3712395 PM |
1511 | break; |
1512 | case FILE_MEMLIST: | |
645fcc9d | 1513 | retval = update_nodemask(cs, trialcs, buf); |
e3712395 PM |
1514 | break; |
1515 | default: | |
1516 | retval = -EINVAL; | |
1517 | break; | |
1518 | } | |
645fcc9d LZ |
1519 | |
1520 | free_trial_cpuset(trialcs); | |
e3712395 PM |
1521 | cgroup_unlock(); |
1522 | return retval; | |
1523 | } | |
1524 | ||
1da177e4 LT |
1525 | /* |
1526 | * These ascii lists should be read in a single call, by using a user | |
1527 | * buffer large enough to hold the entire map. If read in smaller | |
1528 | * chunks, there is no guarantee of atomicity. Since the display format | |
1529 | * used, list of ranges of sequential numbers, is variable length, | |
1530 | * and since these maps can change value dynamically, one could read | |
1531 | * gibberish by doing partial reads while a list was changing. | |
1532 | * A single large read to a buffer that crosses a page boundary is | |
1533 | * ok, because the result being copied to user land is not recomputed | |
1534 | * across a page fault. | |
1535 | */ | |
1536 | ||
1537 | static int cpuset_sprintf_cpulist(char *page, struct cpuset *cs) | |
1538 | { | |
5a7625df | 1539 | int ret; |
1da177e4 | 1540 | |
3d3f26a7 | 1541 | mutex_lock(&callback_mutex); |
300ed6cb | 1542 | ret = cpulist_scnprintf(page, PAGE_SIZE, cs->cpus_allowed); |
3d3f26a7 | 1543 | mutex_unlock(&callback_mutex); |
1da177e4 | 1544 | |
5a7625df | 1545 | return ret; |
1da177e4 LT |
1546 | } |
1547 | ||
1548 | static int cpuset_sprintf_memlist(char *page, struct cpuset *cs) | |
1549 | { | |
1550 | nodemask_t mask; | |
1551 | ||
3d3f26a7 | 1552 | mutex_lock(&callback_mutex); |
1da177e4 | 1553 | mask = cs->mems_allowed; |
3d3f26a7 | 1554 | mutex_unlock(&callback_mutex); |
1da177e4 LT |
1555 | |
1556 | return nodelist_scnprintf(page, PAGE_SIZE, mask); | |
1557 | } | |
1558 | ||
8793d854 PM |
1559 | static ssize_t cpuset_common_file_read(struct cgroup *cont, |
1560 | struct cftype *cft, | |
1561 | struct file *file, | |
1562 | char __user *buf, | |
1563 | size_t nbytes, loff_t *ppos) | |
1da177e4 | 1564 | { |
8793d854 | 1565 | struct cpuset *cs = cgroup_cs(cont); |
1da177e4 LT |
1566 | cpuset_filetype_t type = cft->private; |
1567 | char *page; | |
1568 | ssize_t retval = 0; | |
1569 | char *s; | |
1da177e4 | 1570 | |
e12ba74d | 1571 | if (!(page = (char *)__get_free_page(GFP_TEMPORARY))) |
1da177e4 LT |
1572 | return -ENOMEM; |
1573 | ||
1574 | s = page; | |
1575 | ||
1576 | switch (type) { | |
1577 | case FILE_CPULIST: | |
1578 | s += cpuset_sprintf_cpulist(s, cs); | |
1579 | break; | |
1580 | case FILE_MEMLIST: | |
1581 | s += cpuset_sprintf_memlist(s, cs); | |
1582 | break; | |
1da177e4 LT |
1583 | default: |
1584 | retval = -EINVAL; | |
1585 | goto out; | |
1586 | } | |
1587 | *s++ = '\n'; | |
1da177e4 | 1588 | |
eacaa1f5 | 1589 | retval = simple_read_from_buffer(buf, nbytes, ppos, page, s - page); |
1da177e4 LT |
1590 | out: |
1591 | free_page((unsigned long)page); | |
1592 | return retval; | |
1593 | } | |
1594 | ||
700fe1ab PM |
1595 | static u64 cpuset_read_u64(struct cgroup *cont, struct cftype *cft) |
1596 | { | |
1597 | struct cpuset *cs = cgroup_cs(cont); | |
1598 | cpuset_filetype_t type = cft->private; | |
1599 | switch (type) { | |
1600 | case FILE_CPU_EXCLUSIVE: | |
1601 | return is_cpu_exclusive(cs); | |
1602 | case FILE_MEM_EXCLUSIVE: | |
1603 | return is_mem_exclusive(cs); | |
78608366 PM |
1604 | case FILE_MEM_HARDWALL: |
1605 | return is_mem_hardwall(cs); | |
700fe1ab PM |
1606 | case FILE_SCHED_LOAD_BALANCE: |
1607 | return is_sched_load_balance(cs); | |
1608 | case FILE_MEMORY_MIGRATE: | |
1609 | return is_memory_migrate(cs); | |
1610 | case FILE_MEMORY_PRESSURE_ENABLED: | |
1611 | return cpuset_memory_pressure_enabled; | |
1612 | case FILE_MEMORY_PRESSURE: | |
1613 | return fmeter_getrate(&cs->fmeter); | |
1614 | case FILE_SPREAD_PAGE: | |
1615 | return is_spread_page(cs); | |
1616 | case FILE_SPREAD_SLAB: | |
1617 | return is_spread_slab(cs); | |
1618 | default: | |
1619 | BUG(); | |
1620 | } | |
cf417141 MK |
1621 | |
1622 | /* Unreachable but makes gcc happy */ | |
1623 | return 0; | |
700fe1ab | 1624 | } |
1da177e4 | 1625 | |
5be7a479 PM |
1626 | static s64 cpuset_read_s64(struct cgroup *cont, struct cftype *cft) |
1627 | { | |
1628 | struct cpuset *cs = cgroup_cs(cont); | |
1629 | cpuset_filetype_t type = cft->private; | |
1630 | switch (type) { | |
1631 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1632 | return cs->relax_domain_level; | |
1633 | default: | |
1634 | BUG(); | |
1635 | } | |
cf417141 MK |
1636 | |
1637 | /* Unrechable but makes gcc happy */ | |
1638 | return 0; | |
5be7a479 PM |
1639 | } |
1640 | ||
1da177e4 LT |
1641 | |
1642 | /* | |
1643 | * for the common functions, 'private' gives the type of file | |
1644 | */ | |
1645 | ||
addf2c73 PM |
1646 | static struct cftype files[] = { |
1647 | { | |
1648 | .name = "cpus", | |
1649 | .read = cpuset_common_file_read, | |
e3712395 PM |
1650 | .write_string = cpuset_write_resmask, |
1651 | .max_write_len = (100U + 6 * NR_CPUS), | |
addf2c73 PM |
1652 | .private = FILE_CPULIST, |
1653 | }, | |
1654 | ||
1655 | { | |
1656 | .name = "mems", | |
1657 | .read = cpuset_common_file_read, | |
e3712395 PM |
1658 | .write_string = cpuset_write_resmask, |
1659 | .max_write_len = (100U + 6 * MAX_NUMNODES), | |
addf2c73 PM |
1660 | .private = FILE_MEMLIST, |
1661 | }, | |
1662 | ||
1663 | { | |
1664 | .name = "cpu_exclusive", | |
1665 | .read_u64 = cpuset_read_u64, | |
1666 | .write_u64 = cpuset_write_u64, | |
1667 | .private = FILE_CPU_EXCLUSIVE, | |
1668 | }, | |
1669 | ||
1670 | { | |
1671 | .name = "mem_exclusive", | |
1672 | .read_u64 = cpuset_read_u64, | |
1673 | .write_u64 = cpuset_write_u64, | |
1674 | .private = FILE_MEM_EXCLUSIVE, | |
1675 | }, | |
1676 | ||
78608366 PM |
1677 | { |
1678 | .name = "mem_hardwall", | |
1679 | .read_u64 = cpuset_read_u64, | |
1680 | .write_u64 = cpuset_write_u64, | |
1681 | .private = FILE_MEM_HARDWALL, | |
1682 | }, | |
1683 | ||
addf2c73 PM |
1684 | { |
1685 | .name = "sched_load_balance", | |
1686 | .read_u64 = cpuset_read_u64, | |
1687 | .write_u64 = cpuset_write_u64, | |
1688 | .private = FILE_SCHED_LOAD_BALANCE, | |
1689 | }, | |
1690 | ||
1691 | { | |
1692 | .name = "sched_relax_domain_level", | |
5be7a479 PM |
1693 | .read_s64 = cpuset_read_s64, |
1694 | .write_s64 = cpuset_write_s64, | |
addf2c73 PM |
1695 | .private = FILE_SCHED_RELAX_DOMAIN_LEVEL, |
1696 | }, | |
1697 | ||
1698 | { | |
1699 | .name = "memory_migrate", | |
1700 | .read_u64 = cpuset_read_u64, | |
1701 | .write_u64 = cpuset_write_u64, | |
1702 | .private = FILE_MEMORY_MIGRATE, | |
1703 | }, | |
1704 | ||
1705 | { | |
1706 | .name = "memory_pressure", | |
1707 | .read_u64 = cpuset_read_u64, | |
1708 | .write_u64 = cpuset_write_u64, | |
1709 | .private = FILE_MEMORY_PRESSURE, | |
099fca32 | 1710 | .mode = S_IRUGO, |
addf2c73 PM |
1711 | }, |
1712 | ||
1713 | { | |
1714 | .name = "memory_spread_page", | |
1715 | .read_u64 = cpuset_read_u64, | |
1716 | .write_u64 = cpuset_write_u64, | |
1717 | .private = FILE_SPREAD_PAGE, | |
1718 | }, | |
1719 | ||
1720 | { | |
1721 | .name = "memory_spread_slab", | |
1722 | .read_u64 = cpuset_read_u64, | |
1723 | .write_u64 = cpuset_write_u64, | |
1724 | .private = FILE_SPREAD_SLAB, | |
1725 | }, | |
45b07ef3 PJ |
1726 | }; |
1727 | ||
3e0d98b9 PJ |
1728 | static struct cftype cft_memory_pressure_enabled = { |
1729 | .name = "memory_pressure_enabled", | |
700fe1ab PM |
1730 | .read_u64 = cpuset_read_u64, |
1731 | .write_u64 = cpuset_write_u64, | |
3e0d98b9 PJ |
1732 | .private = FILE_MEMORY_PRESSURE_ENABLED, |
1733 | }; | |
1734 | ||
8793d854 | 1735 | static int cpuset_populate(struct cgroup_subsys *ss, struct cgroup *cont) |
1da177e4 LT |
1736 | { |
1737 | int err; | |
1738 | ||
addf2c73 PM |
1739 | err = cgroup_add_files(cont, ss, files, ARRAY_SIZE(files)); |
1740 | if (err) | |
1da177e4 | 1741 | return err; |
8793d854 | 1742 | /* memory_pressure_enabled is in root cpuset only */ |
addf2c73 | 1743 | if (!cont->parent) |
8793d854 | 1744 | err = cgroup_add_file(cont, ss, |
addf2c73 PM |
1745 | &cft_memory_pressure_enabled); |
1746 | return err; | |
1da177e4 LT |
1747 | } |
1748 | ||
8793d854 PM |
1749 | /* |
1750 | * post_clone() is called at the end of cgroup_clone(). | |
1751 | * 'cgroup' was just created automatically as a result of | |
1752 | * a cgroup_clone(), and the current task is about to | |
1753 | * be moved into 'cgroup'. | |
1754 | * | |
1755 | * Currently we refuse to set up the cgroup - thereby | |
1756 | * refusing the task to be entered, and as a result refusing | |
1757 | * the sys_unshare() or clone() which initiated it - if any | |
1758 | * sibling cpusets have exclusive cpus or mem. | |
1759 | * | |
1760 | * If this becomes a problem for some users who wish to | |
1761 | * allow that scenario, then cpuset_post_clone() could be | |
1762 | * changed to grant parent->cpus_allowed-sibling_cpus_exclusive | |
2df167a3 PM |
1763 | * (and likewise for mems) to the new cgroup. Called with cgroup_mutex |
1764 | * held. | |
8793d854 PM |
1765 | */ |
1766 | static void cpuset_post_clone(struct cgroup_subsys *ss, | |
1767 | struct cgroup *cgroup) | |
1768 | { | |
1769 | struct cgroup *parent, *child; | |
1770 | struct cpuset *cs, *parent_cs; | |
1771 | ||
1772 | parent = cgroup->parent; | |
1773 | list_for_each_entry(child, &parent->children, sibling) { | |
1774 | cs = cgroup_cs(child); | |
1775 | if (is_mem_exclusive(cs) || is_cpu_exclusive(cs)) | |
1776 | return; | |
1777 | } | |
1778 | cs = cgroup_cs(cgroup); | |
1779 | parent_cs = cgroup_cs(parent); | |
1780 | ||
1781 | cs->mems_allowed = parent_cs->mems_allowed; | |
300ed6cb | 1782 | cpumask_copy(cs->cpus_allowed, parent_cs->cpus_allowed); |
8793d854 PM |
1783 | return; |
1784 | } | |
1785 | ||
1da177e4 LT |
1786 | /* |
1787 | * cpuset_create - create a cpuset | |
2df167a3 PM |
1788 | * ss: cpuset cgroup subsystem |
1789 | * cont: control group that the new cpuset will be part of | |
1da177e4 LT |
1790 | */ |
1791 | ||
8793d854 PM |
1792 | static struct cgroup_subsys_state *cpuset_create( |
1793 | struct cgroup_subsys *ss, | |
1794 | struct cgroup *cont) | |
1da177e4 LT |
1795 | { |
1796 | struct cpuset *cs; | |
8793d854 | 1797 | struct cpuset *parent; |
1da177e4 | 1798 | |
8793d854 PM |
1799 | if (!cont->parent) { |
1800 | /* This is early initialization for the top cgroup */ | |
1801 | top_cpuset.mems_generation = cpuset_mems_generation++; | |
1802 | return &top_cpuset.css; | |
1803 | } | |
1804 | parent = cgroup_cs(cont->parent); | |
1da177e4 LT |
1805 | cs = kmalloc(sizeof(*cs), GFP_KERNEL); |
1806 | if (!cs) | |
8793d854 | 1807 | return ERR_PTR(-ENOMEM); |
300ed6cb LZ |
1808 | if (!alloc_cpumask_var(&cs->cpus_allowed, GFP_KERNEL)) { |
1809 | kfree(cs); | |
1810 | return ERR_PTR(-ENOMEM); | |
1811 | } | |
1da177e4 | 1812 | |
cf2a473c | 1813 | cpuset_update_task_memory_state(); |
1da177e4 | 1814 | cs->flags = 0; |
825a46af PJ |
1815 | if (is_spread_page(parent)) |
1816 | set_bit(CS_SPREAD_PAGE, &cs->flags); | |
1817 | if (is_spread_slab(parent)) | |
1818 | set_bit(CS_SPREAD_SLAB, &cs->flags); | |
029190c5 | 1819 | set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); |
300ed6cb | 1820 | cpumask_clear(cs->cpus_allowed); |
f9a86fcb | 1821 | nodes_clear(cs->mems_allowed); |
151a4420 | 1822 | cs->mems_generation = cpuset_mems_generation++; |
3e0d98b9 | 1823 | fmeter_init(&cs->fmeter); |
1d3504fc | 1824 | cs->relax_domain_level = -1; |
1da177e4 LT |
1825 | |
1826 | cs->parent = parent; | |
202f72d5 | 1827 | number_of_cpusets++; |
8793d854 | 1828 | return &cs->css ; |
1da177e4 LT |
1829 | } |
1830 | ||
029190c5 | 1831 | /* |
029190c5 PJ |
1832 | * If the cpuset being removed has its flag 'sched_load_balance' |
1833 | * enabled, then simulate turning sched_load_balance off, which | |
cf417141 | 1834 | * will call async_rebuild_sched_domains(). |
029190c5 PJ |
1835 | */ |
1836 | ||
8793d854 | 1837 | static void cpuset_destroy(struct cgroup_subsys *ss, struct cgroup *cont) |
1da177e4 | 1838 | { |
8793d854 | 1839 | struct cpuset *cs = cgroup_cs(cont); |
1da177e4 | 1840 | |
cf2a473c | 1841 | cpuset_update_task_memory_state(); |
029190c5 PJ |
1842 | |
1843 | if (is_sched_load_balance(cs)) | |
700fe1ab | 1844 | update_flag(CS_SCHED_LOAD_BALANCE, cs, 0); |
029190c5 | 1845 | |
202f72d5 | 1846 | number_of_cpusets--; |
300ed6cb | 1847 | free_cpumask_var(cs->cpus_allowed); |
8793d854 | 1848 | kfree(cs); |
1da177e4 LT |
1849 | } |
1850 | ||
8793d854 PM |
1851 | struct cgroup_subsys cpuset_subsys = { |
1852 | .name = "cpuset", | |
1853 | .create = cpuset_create, | |
cf417141 | 1854 | .destroy = cpuset_destroy, |
8793d854 PM |
1855 | .can_attach = cpuset_can_attach, |
1856 | .attach = cpuset_attach, | |
1857 | .populate = cpuset_populate, | |
1858 | .post_clone = cpuset_post_clone, | |
1859 | .subsys_id = cpuset_subsys_id, | |
1860 | .early_init = 1, | |
1861 | }; | |
1862 | ||
c417f024 PJ |
1863 | /* |
1864 | * cpuset_init_early - just enough so that the calls to | |
1865 | * cpuset_update_task_memory_state() in early init code | |
1866 | * are harmless. | |
1867 | */ | |
1868 | ||
1869 | int __init cpuset_init_early(void) | |
1870 | { | |
38c7fed2 | 1871 | alloc_cpumask_var(&top_cpuset.cpus_allowed, GFP_NOWAIT); |
300ed6cb | 1872 | |
8793d854 | 1873 | top_cpuset.mems_generation = cpuset_mems_generation++; |
c417f024 PJ |
1874 | return 0; |
1875 | } | |
1876 | ||
8793d854 | 1877 | |
1da177e4 LT |
1878 | /** |
1879 | * cpuset_init - initialize cpusets at system boot | |
1880 | * | |
1881 | * Description: Initialize top_cpuset and the cpuset internal file system, | |
1882 | **/ | |
1883 | ||
1884 | int __init cpuset_init(void) | |
1885 | { | |
8793d854 | 1886 | int err = 0; |
1da177e4 | 1887 | |
300ed6cb | 1888 | cpumask_setall(top_cpuset.cpus_allowed); |
f9a86fcb | 1889 | nodes_setall(top_cpuset.mems_allowed); |
1da177e4 | 1890 | |
3e0d98b9 | 1891 | fmeter_init(&top_cpuset.fmeter); |
151a4420 | 1892 | top_cpuset.mems_generation = cpuset_mems_generation++; |
029190c5 | 1893 | set_bit(CS_SCHED_LOAD_BALANCE, &top_cpuset.flags); |
1d3504fc | 1894 | top_cpuset.relax_domain_level = -1; |
1da177e4 | 1895 | |
1da177e4 LT |
1896 | err = register_filesystem(&cpuset_fs_type); |
1897 | if (err < 0) | |
8793d854 PM |
1898 | return err; |
1899 | ||
2341d1b6 LZ |
1900 | if (!alloc_cpumask_var(&cpus_attach, GFP_KERNEL)) |
1901 | BUG(); | |
1902 | ||
202f72d5 | 1903 | number_of_cpusets = 1; |
8793d854 | 1904 | return 0; |
1da177e4 LT |
1905 | } |
1906 | ||
956db3ca CW |
1907 | /** |
1908 | * cpuset_do_move_task - move a given task to another cpuset | |
1909 | * @tsk: pointer to task_struct the task to move | |
1910 | * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner | |
1911 | * | |
1912 | * Called by cgroup_scan_tasks() for each task in a cgroup. | |
1913 | * Return nonzero to stop the walk through the tasks. | |
1914 | */ | |
9e0c914c AB |
1915 | static void cpuset_do_move_task(struct task_struct *tsk, |
1916 | struct cgroup_scanner *scan) | |
956db3ca | 1917 | { |
7f81b1ae | 1918 | struct cgroup *new_cgroup = scan->data; |
956db3ca | 1919 | |
7f81b1ae | 1920 | cgroup_attach_task(new_cgroup, tsk); |
956db3ca CW |
1921 | } |
1922 | ||
1923 | /** | |
1924 | * move_member_tasks_to_cpuset - move tasks from one cpuset to another | |
1925 | * @from: cpuset in which the tasks currently reside | |
1926 | * @to: cpuset to which the tasks will be moved | |
1927 | * | |
c8d9c90c PJ |
1928 | * Called with cgroup_mutex held |
1929 | * callback_mutex must not be held, as cpuset_attach() will take it. | |
956db3ca CW |
1930 | * |
1931 | * The cgroup_scan_tasks() function will scan all the tasks in a cgroup, | |
1932 | * calling callback functions for each. | |
1933 | */ | |
1934 | static void move_member_tasks_to_cpuset(struct cpuset *from, struct cpuset *to) | |
1935 | { | |
7f81b1ae | 1936 | struct cgroup_scanner scan; |
956db3ca | 1937 | |
7f81b1ae LZ |
1938 | scan.cg = from->css.cgroup; |
1939 | scan.test_task = NULL; /* select all tasks in cgroup */ | |
1940 | scan.process_task = cpuset_do_move_task; | |
1941 | scan.heap = NULL; | |
1942 | scan.data = to->css.cgroup; | |
956db3ca | 1943 | |
7f81b1ae | 1944 | if (cgroup_scan_tasks(&scan)) |
956db3ca CW |
1945 | printk(KERN_ERR "move_member_tasks_to_cpuset: " |
1946 | "cgroup_scan_tasks failed\n"); | |
1947 | } | |
1948 | ||
b1aac8bb | 1949 | /* |
cf417141 | 1950 | * If CPU and/or memory hotplug handlers, below, unplug any CPUs |
b1aac8bb PJ |
1951 | * or memory nodes, we need to walk over the cpuset hierarchy, |
1952 | * removing that CPU or node from all cpusets. If this removes the | |
956db3ca CW |
1953 | * last CPU or node from a cpuset, then move the tasks in the empty |
1954 | * cpuset to its next-highest non-empty parent. | |
b1aac8bb | 1955 | * |
c8d9c90c PJ |
1956 | * Called with cgroup_mutex held |
1957 | * callback_mutex must not be held, as cpuset_attach() will take it. | |
b1aac8bb | 1958 | */ |
956db3ca CW |
1959 | static void remove_tasks_in_empty_cpuset(struct cpuset *cs) |
1960 | { | |
1961 | struct cpuset *parent; | |
1962 | ||
c8d9c90c PJ |
1963 | /* |
1964 | * The cgroup's css_sets list is in use if there are tasks | |
1965 | * in the cpuset; the list is empty if there are none; | |
1966 | * the cs->css.refcnt seems always 0. | |
1967 | */ | |
956db3ca CW |
1968 | if (list_empty(&cs->css.cgroup->css_sets)) |
1969 | return; | |
b1aac8bb | 1970 | |
956db3ca CW |
1971 | /* |
1972 | * Find its next-highest non-empty parent, (top cpuset | |
1973 | * has online cpus, so can't be empty). | |
1974 | */ | |
1975 | parent = cs->parent; | |
300ed6cb | 1976 | while (cpumask_empty(parent->cpus_allowed) || |
b4501295 | 1977 | nodes_empty(parent->mems_allowed)) |
956db3ca | 1978 | parent = parent->parent; |
956db3ca CW |
1979 | |
1980 | move_member_tasks_to_cpuset(cs, parent); | |
1981 | } | |
1982 | ||
1983 | /* | |
1984 | * Walk the specified cpuset subtree and look for empty cpusets. | |
1985 | * The tasks of such cpuset must be moved to a parent cpuset. | |
1986 | * | |
2df167a3 | 1987 | * Called with cgroup_mutex held. We take callback_mutex to modify |
956db3ca CW |
1988 | * cpus_allowed and mems_allowed. |
1989 | * | |
1990 | * This walk processes the tree from top to bottom, completing one layer | |
1991 | * before dropping down to the next. It always processes a node before | |
1992 | * any of its children. | |
1993 | * | |
1994 | * For now, since we lack memory hot unplug, we'll never see a cpuset | |
1995 | * that has tasks along with an empty 'mems'. But if we did see such | |
1996 | * a cpuset, we'd handle it just like we do if its 'cpus' was empty. | |
1997 | */ | |
d294eb83 | 1998 | static void scan_for_empty_cpusets(struct cpuset *root) |
b1aac8bb | 1999 | { |
8d1e6266 | 2000 | LIST_HEAD(queue); |
956db3ca CW |
2001 | struct cpuset *cp; /* scans cpusets being updated */ |
2002 | struct cpuset *child; /* scans child cpusets of cp */ | |
8793d854 | 2003 | struct cgroup *cont; |
f9b4fb8d | 2004 | nodemask_t oldmems; |
b1aac8bb | 2005 | |
956db3ca CW |
2006 | list_add_tail((struct list_head *)&root->stack_list, &queue); |
2007 | ||
956db3ca | 2008 | while (!list_empty(&queue)) { |
8d1e6266 | 2009 | cp = list_first_entry(&queue, struct cpuset, stack_list); |
956db3ca CW |
2010 | list_del(queue.next); |
2011 | list_for_each_entry(cont, &cp->css.cgroup->children, sibling) { | |
2012 | child = cgroup_cs(cont); | |
2013 | list_add_tail(&child->stack_list, &queue); | |
2014 | } | |
b4501295 PJ |
2015 | |
2016 | /* Continue past cpusets with all cpus, mems online */ | |
300ed6cb | 2017 | if (cpumask_subset(cp->cpus_allowed, cpu_online_mask) && |
b4501295 PJ |
2018 | nodes_subset(cp->mems_allowed, node_states[N_HIGH_MEMORY])) |
2019 | continue; | |
2020 | ||
f9b4fb8d MX |
2021 | oldmems = cp->mems_allowed; |
2022 | ||
956db3ca | 2023 | /* Remove offline cpus and mems from this cpuset. */ |
b4501295 | 2024 | mutex_lock(&callback_mutex); |
300ed6cb LZ |
2025 | cpumask_and(cp->cpus_allowed, cp->cpus_allowed, |
2026 | cpu_online_mask); | |
956db3ca CW |
2027 | nodes_and(cp->mems_allowed, cp->mems_allowed, |
2028 | node_states[N_HIGH_MEMORY]); | |
b4501295 PJ |
2029 | mutex_unlock(&callback_mutex); |
2030 | ||
2031 | /* Move tasks from the empty cpuset to a parent */ | |
300ed6cb | 2032 | if (cpumask_empty(cp->cpus_allowed) || |
b4501295 | 2033 | nodes_empty(cp->mems_allowed)) |
956db3ca | 2034 | remove_tasks_in_empty_cpuset(cp); |
f9b4fb8d | 2035 | else { |
4e74339a | 2036 | update_tasks_cpumask(cp, NULL); |
010cfac4 | 2037 | update_tasks_nodemask(cp, &oldmems, NULL); |
f9b4fb8d | 2038 | } |
b1aac8bb PJ |
2039 | } |
2040 | } | |
2041 | ||
4c4d50f7 PJ |
2042 | /* |
2043 | * The top_cpuset tracks what CPUs and Memory Nodes are online, | |
2044 | * period. This is necessary in order to make cpusets transparent | |
2045 | * (of no affect) on systems that are actively using CPU hotplug | |
2046 | * but making no active use of cpusets. | |
2047 | * | |
38837fc7 PJ |
2048 | * This routine ensures that top_cpuset.cpus_allowed tracks |
2049 | * cpu_online_map on each CPU hotplug (cpuhp) event. | |
cf417141 MK |
2050 | * |
2051 | * Called within get_online_cpus(). Needs to call cgroup_lock() | |
2052 | * before calling generate_sched_domains(). | |
4c4d50f7 | 2053 | */ |
cf417141 | 2054 | static int cpuset_track_online_cpus(struct notifier_block *unused_nb, |
029190c5 | 2055 | unsigned long phase, void *unused_cpu) |
4c4d50f7 | 2056 | { |
cf417141 | 2057 | struct sched_domain_attr *attr; |
6af866af | 2058 | struct cpumask *doms; |
cf417141 MK |
2059 | int ndoms; |
2060 | ||
3e84050c | 2061 | switch (phase) { |
3e84050c DA |
2062 | case CPU_ONLINE: |
2063 | case CPU_ONLINE_FROZEN: | |
2064 | case CPU_DEAD: | |
2065 | case CPU_DEAD_FROZEN: | |
3e84050c | 2066 | break; |
cf417141 | 2067 | |
3e84050c | 2068 | default: |
ac076758 | 2069 | return NOTIFY_DONE; |
3e84050c | 2070 | } |
ac076758 | 2071 | |
cf417141 | 2072 | cgroup_lock(); |
0b4217b3 | 2073 | mutex_lock(&callback_mutex); |
300ed6cb | 2074 | cpumask_copy(top_cpuset.cpus_allowed, cpu_online_mask); |
0b4217b3 | 2075 | mutex_unlock(&callback_mutex); |
cf417141 MK |
2076 | scan_for_empty_cpusets(&top_cpuset); |
2077 | ndoms = generate_sched_domains(&doms, &attr); | |
2078 | cgroup_unlock(); | |
2079 | ||
2080 | /* Have scheduler rebuild the domains */ | |
2081 | partition_sched_domains(ndoms, doms, attr); | |
2082 | ||
3e84050c | 2083 | return NOTIFY_OK; |
4c4d50f7 | 2084 | } |
4c4d50f7 | 2085 | |
b1aac8bb | 2086 | #ifdef CONFIG_MEMORY_HOTPLUG |
38837fc7 | 2087 | /* |
0e1e7c7a | 2088 | * Keep top_cpuset.mems_allowed tracking node_states[N_HIGH_MEMORY]. |
cf417141 MK |
2089 | * Call this routine anytime after node_states[N_HIGH_MEMORY] changes. |
2090 | * See also the previous routine cpuset_track_online_cpus(). | |
38837fc7 | 2091 | */ |
f481891f MX |
2092 | static int cpuset_track_online_nodes(struct notifier_block *self, |
2093 | unsigned long action, void *arg) | |
38837fc7 | 2094 | { |
cf417141 | 2095 | cgroup_lock(); |
f481891f MX |
2096 | switch (action) { |
2097 | case MEM_ONLINE: | |
f481891f | 2098 | case MEM_OFFLINE: |
0b4217b3 | 2099 | mutex_lock(&callback_mutex); |
f481891f | 2100 | top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY]; |
0b4217b3 LZ |
2101 | mutex_unlock(&callback_mutex); |
2102 | if (action == MEM_OFFLINE) | |
2103 | scan_for_empty_cpusets(&top_cpuset); | |
f481891f MX |
2104 | break; |
2105 | default: | |
2106 | break; | |
2107 | } | |
cf417141 | 2108 | cgroup_unlock(); |
f481891f | 2109 | return NOTIFY_OK; |
38837fc7 PJ |
2110 | } |
2111 | #endif | |
2112 | ||
1da177e4 LT |
2113 | /** |
2114 | * cpuset_init_smp - initialize cpus_allowed | |
2115 | * | |
2116 | * Description: Finish top cpuset after cpu, node maps are initialized | |
2117 | **/ | |
2118 | ||
2119 | void __init cpuset_init_smp(void) | |
2120 | { | |
300ed6cb | 2121 | cpumask_copy(top_cpuset.cpus_allowed, cpu_online_mask); |
0e1e7c7a | 2122 | top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY]; |
4c4d50f7 | 2123 | |
cf417141 | 2124 | hotcpu_notifier(cpuset_track_online_cpus, 0); |
f481891f | 2125 | hotplug_memory_notifier(cpuset_track_online_nodes, 10); |
f90d4118 MX |
2126 | |
2127 | cpuset_wq = create_singlethread_workqueue("cpuset"); | |
2128 | BUG_ON(!cpuset_wq); | |
1da177e4 LT |
2129 | } |
2130 | ||
2131 | /** | |
1da177e4 LT |
2132 | * cpuset_cpus_allowed - return cpus_allowed mask from a tasks cpuset. |
2133 | * @tsk: pointer to task_struct from which to obtain cpuset->cpus_allowed. | |
6af866af | 2134 | * @pmask: pointer to struct cpumask variable to receive cpus_allowed set. |
1da177e4 | 2135 | * |
300ed6cb | 2136 | * Description: Returns the cpumask_var_t cpus_allowed of the cpuset |
1da177e4 LT |
2137 | * attached to the specified @tsk. Guaranteed to return some non-empty |
2138 | * subset of cpu_online_map, even if this means going outside the | |
2139 | * tasks cpuset. | |
2140 | **/ | |
2141 | ||
6af866af | 2142 | void cpuset_cpus_allowed(struct task_struct *tsk, struct cpumask *pmask) |
1da177e4 | 2143 | { |
3d3f26a7 | 2144 | mutex_lock(&callback_mutex); |
f9a86fcb | 2145 | cpuset_cpus_allowed_locked(tsk, pmask); |
470fd646 | 2146 | mutex_unlock(&callback_mutex); |
470fd646 CW |
2147 | } |
2148 | ||
2149 | /** | |
2150 | * cpuset_cpus_allowed_locked - return cpus_allowed mask from a tasks cpuset. | |
2df167a3 | 2151 | * Must be called with callback_mutex held. |
470fd646 | 2152 | **/ |
6af866af | 2153 | void cpuset_cpus_allowed_locked(struct task_struct *tsk, struct cpumask *pmask) |
470fd646 | 2154 | { |
909d75a3 | 2155 | task_lock(tsk); |
f9a86fcb | 2156 | guarantee_online_cpus(task_cs(tsk), pmask); |
909d75a3 | 2157 | task_unlock(tsk); |
1da177e4 LT |
2158 | } |
2159 | ||
2160 | void cpuset_init_current_mems_allowed(void) | |
2161 | { | |
f9a86fcb | 2162 | nodes_setall(current->mems_allowed); |
1da177e4 LT |
2163 | } |
2164 | ||
909d75a3 PJ |
2165 | /** |
2166 | * cpuset_mems_allowed - return mems_allowed mask from a tasks cpuset. | |
2167 | * @tsk: pointer to task_struct from which to obtain cpuset->mems_allowed. | |
2168 | * | |
2169 | * Description: Returns the nodemask_t mems_allowed of the cpuset | |
2170 | * attached to the specified @tsk. Guaranteed to return some non-empty | |
0e1e7c7a | 2171 | * subset of node_states[N_HIGH_MEMORY], even if this means going outside the |
909d75a3 PJ |
2172 | * tasks cpuset. |
2173 | **/ | |
2174 | ||
2175 | nodemask_t cpuset_mems_allowed(struct task_struct *tsk) | |
2176 | { | |
2177 | nodemask_t mask; | |
2178 | ||
3d3f26a7 | 2179 | mutex_lock(&callback_mutex); |
909d75a3 | 2180 | task_lock(tsk); |
8793d854 | 2181 | guarantee_online_mems(task_cs(tsk), &mask); |
909d75a3 | 2182 | task_unlock(tsk); |
3d3f26a7 | 2183 | mutex_unlock(&callback_mutex); |
909d75a3 PJ |
2184 | |
2185 | return mask; | |
2186 | } | |
2187 | ||
d9fd8a6d | 2188 | /** |
19770b32 MG |
2189 | * cpuset_nodemask_valid_mems_allowed - check nodemask vs. curremt mems_allowed |
2190 | * @nodemask: the nodemask to be checked | |
d9fd8a6d | 2191 | * |
19770b32 | 2192 | * Are any of the nodes in the nodemask allowed in current->mems_allowed? |
1da177e4 | 2193 | */ |
19770b32 | 2194 | int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask) |
1da177e4 | 2195 | { |
19770b32 | 2196 | return nodes_intersects(*nodemask, current->mems_allowed); |
1da177e4 LT |
2197 | } |
2198 | ||
9bf2229f | 2199 | /* |
78608366 PM |
2200 | * nearest_hardwall_ancestor() - Returns the nearest mem_exclusive or |
2201 | * mem_hardwall ancestor to the specified cpuset. Call holding | |
2202 | * callback_mutex. If no ancestor is mem_exclusive or mem_hardwall | |
2203 | * (an unusual configuration), then returns the root cpuset. | |
9bf2229f | 2204 | */ |
78608366 | 2205 | static const struct cpuset *nearest_hardwall_ancestor(const struct cpuset *cs) |
9bf2229f | 2206 | { |
78608366 | 2207 | while (!(is_mem_exclusive(cs) || is_mem_hardwall(cs)) && cs->parent) |
9bf2229f PJ |
2208 | cs = cs->parent; |
2209 | return cs; | |
2210 | } | |
2211 | ||
d9fd8a6d | 2212 | /** |
a1bc5a4e DR |
2213 | * cpuset_node_allowed_softwall - Can we allocate on a memory node? |
2214 | * @node: is this an allowed node? | |
02a0e53d | 2215 | * @gfp_mask: memory allocation flags |
d9fd8a6d | 2216 | * |
a1bc5a4e DR |
2217 | * If we're in interrupt, yes, we can always allocate. If __GFP_THISNODE is |
2218 | * set, yes, we can always allocate. If node is in our task's mems_allowed, | |
2219 | * yes. If it's not a __GFP_HARDWALL request and this node is in the nearest | |
2220 | * hardwalled cpuset ancestor to this task's cpuset, yes. If the task has been | |
2221 | * OOM killed and has access to memory reserves as specified by the TIF_MEMDIE | |
2222 | * flag, yes. | |
9bf2229f PJ |
2223 | * Otherwise, no. |
2224 | * | |
a1bc5a4e DR |
2225 | * If __GFP_HARDWALL is set, cpuset_node_allowed_softwall() reduces to |
2226 | * cpuset_node_allowed_hardwall(). Otherwise, cpuset_node_allowed_softwall() | |
2227 | * might sleep, and might allow a node from an enclosing cpuset. | |
02a0e53d | 2228 | * |
a1bc5a4e DR |
2229 | * cpuset_node_allowed_hardwall() only handles the simpler case of hardwall |
2230 | * cpusets, and never sleeps. | |
02a0e53d PJ |
2231 | * |
2232 | * The __GFP_THISNODE placement logic is really handled elsewhere, | |
2233 | * by forcibly using a zonelist starting at a specified node, and by | |
2234 | * (in get_page_from_freelist()) refusing to consider the zones for | |
2235 | * any node on the zonelist except the first. By the time any such | |
2236 | * calls get to this routine, we should just shut up and say 'yes'. | |
2237 | * | |
9bf2229f | 2238 | * GFP_USER allocations are marked with the __GFP_HARDWALL bit, |
c596d9f3 DR |
2239 | * and do not allow allocations outside the current tasks cpuset |
2240 | * unless the task has been OOM killed as is marked TIF_MEMDIE. | |
9bf2229f | 2241 | * GFP_KERNEL allocations are not so marked, so can escape to the |
78608366 | 2242 | * nearest enclosing hardwalled ancestor cpuset. |
9bf2229f | 2243 | * |
02a0e53d PJ |
2244 | * Scanning up parent cpusets requires callback_mutex. The |
2245 | * __alloc_pages() routine only calls here with __GFP_HARDWALL bit | |
2246 | * _not_ set if it's a GFP_KERNEL allocation, and all nodes in the | |
2247 | * current tasks mems_allowed came up empty on the first pass over | |
2248 | * the zonelist. So only GFP_KERNEL allocations, if all nodes in the | |
2249 | * cpuset are short of memory, might require taking the callback_mutex | |
2250 | * mutex. | |
9bf2229f | 2251 | * |
36be57ff | 2252 | * The first call here from mm/page_alloc:get_page_from_freelist() |
02a0e53d PJ |
2253 | * has __GFP_HARDWALL set in gfp_mask, enforcing hardwall cpusets, |
2254 | * so no allocation on a node outside the cpuset is allowed (unless | |
2255 | * in interrupt, of course). | |
36be57ff PJ |
2256 | * |
2257 | * The second pass through get_page_from_freelist() doesn't even call | |
2258 | * here for GFP_ATOMIC calls. For those calls, the __alloc_pages() | |
2259 | * variable 'wait' is not set, and the bit ALLOC_CPUSET is not set | |
2260 | * in alloc_flags. That logic and the checks below have the combined | |
2261 | * affect that: | |
9bf2229f PJ |
2262 | * in_interrupt - any node ok (current task context irrelevant) |
2263 | * GFP_ATOMIC - any node ok | |
c596d9f3 | 2264 | * TIF_MEMDIE - any node ok |
78608366 | 2265 | * GFP_KERNEL - any node in enclosing hardwalled cpuset ok |
9bf2229f | 2266 | * GFP_USER - only nodes in current tasks mems allowed ok. |
36be57ff PJ |
2267 | * |
2268 | * Rule: | |
a1bc5a4e | 2269 | * Don't call cpuset_node_allowed_softwall if you can't sleep, unless you |
36be57ff PJ |
2270 | * pass in the __GFP_HARDWALL flag set in gfp_flag, which disables |
2271 | * the code that might scan up ancestor cpusets and sleep. | |
02a0e53d | 2272 | */ |
a1bc5a4e | 2273 | int __cpuset_node_allowed_softwall(int node, gfp_t gfp_mask) |
1da177e4 | 2274 | { |
9bf2229f | 2275 | const struct cpuset *cs; /* current cpuset ancestors */ |
29afd49b | 2276 | int allowed; /* is allocation in zone z allowed? */ |
9bf2229f | 2277 | |
9b819d20 | 2278 | if (in_interrupt() || (gfp_mask & __GFP_THISNODE)) |
9bf2229f | 2279 | return 1; |
92d1dbd2 | 2280 | might_sleep_if(!(gfp_mask & __GFP_HARDWALL)); |
9bf2229f PJ |
2281 | if (node_isset(node, current->mems_allowed)) |
2282 | return 1; | |
c596d9f3 DR |
2283 | /* |
2284 | * Allow tasks that have access to memory reserves because they have | |
2285 | * been OOM killed to get memory anywhere. | |
2286 | */ | |
2287 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
2288 | return 1; | |
9bf2229f PJ |
2289 | if (gfp_mask & __GFP_HARDWALL) /* If hardwall request, stop here */ |
2290 | return 0; | |
2291 | ||
5563e770 BP |
2292 | if (current->flags & PF_EXITING) /* Let dying task have memory */ |
2293 | return 1; | |
2294 | ||
9bf2229f | 2295 | /* Not hardwall and node outside mems_allowed: scan up cpusets */ |
3d3f26a7 | 2296 | mutex_lock(&callback_mutex); |
053199ed | 2297 | |
053199ed | 2298 | task_lock(current); |
78608366 | 2299 | cs = nearest_hardwall_ancestor(task_cs(current)); |
053199ed PJ |
2300 | task_unlock(current); |
2301 | ||
9bf2229f | 2302 | allowed = node_isset(node, cs->mems_allowed); |
3d3f26a7 | 2303 | mutex_unlock(&callback_mutex); |
9bf2229f | 2304 | return allowed; |
1da177e4 LT |
2305 | } |
2306 | ||
02a0e53d | 2307 | /* |
a1bc5a4e DR |
2308 | * cpuset_node_allowed_hardwall - Can we allocate on a memory node? |
2309 | * @node: is this an allowed node? | |
02a0e53d PJ |
2310 | * @gfp_mask: memory allocation flags |
2311 | * | |
a1bc5a4e DR |
2312 | * If we're in interrupt, yes, we can always allocate. If __GFP_THISNODE is |
2313 | * set, yes, we can always allocate. If node is in our task's mems_allowed, | |
2314 | * yes. If the task has been OOM killed and has access to memory reserves as | |
2315 | * specified by the TIF_MEMDIE flag, yes. | |
2316 | * Otherwise, no. | |
02a0e53d PJ |
2317 | * |
2318 | * The __GFP_THISNODE placement logic is really handled elsewhere, | |
2319 | * by forcibly using a zonelist starting at a specified node, and by | |
2320 | * (in get_page_from_freelist()) refusing to consider the zones for | |
2321 | * any node on the zonelist except the first. By the time any such | |
2322 | * calls get to this routine, we should just shut up and say 'yes'. | |
2323 | * | |
a1bc5a4e DR |
2324 | * Unlike the cpuset_node_allowed_softwall() variant, above, |
2325 | * this variant requires that the node be in the current task's | |
02a0e53d PJ |
2326 | * mems_allowed or that we're in interrupt. It does not scan up the |
2327 | * cpuset hierarchy for the nearest enclosing mem_exclusive cpuset. | |
2328 | * It never sleeps. | |
2329 | */ | |
a1bc5a4e | 2330 | int __cpuset_node_allowed_hardwall(int node, gfp_t gfp_mask) |
02a0e53d | 2331 | { |
02a0e53d PJ |
2332 | if (in_interrupt() || (gfp_mask & __GFP_THISNODE)) |
2333 | return 1; | |
02a0e53d PJ |
2334 | if (node_isset(node, current->mems_allowed)) |
2335 | return 1; | |
dedf8b79 DW |
2336 | /* |
2337 | * Allow tasks that have access to memory reserves because they have | |
2338 | * been OOM killed to get memory anywhere. | |
2339 | */ | |
2340 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
2341 | return 1; | |
02a0e53d PJ |
2342 | return 0; |
2343 | } | |
2344 | ||
505970b9 PJ |
2345 | /** |
2346 | * cpuset_lock - lock out any changes to cpuset structures | |
2347 | * | |
3d3f26a7 | 2348 | * The out of memory (oom) code needs to mutex_lock cpusets |
505970b9 | 2349 | * from being changed while it scans the tasklist looking for a |
3d3f26a7 | 2350 | * task in an overlapping cpuset. Expose callback_mutex via this |
505970b9 PJ |
2351 | * cpuset_lock() routine, so the oom code can lock it, before |
2352 | * locking the task list. The tasklist_lock is a spinlock, so | |
3d3f26a7 | 2353 | * must be taken inside callback_mutex. |
505970b9 PJ |
2354 | */ |
2355 | ||
2356 | void cpuset_lock(void) | |
2357 | { | |
3d3f26a7 | 2358 | mutex_lock(&callback_mutex); |
505970b9 PJ |
2359 | } |
2360 | ||
2361 | /** | |
2362 | * cpuset_unlock - release lock on cpuset changes | |
2363 | * | |
2364 | * Undo the lock taken in a previous cpuset_lock() call. | |
2365 | */ | |
2366 | ||
2367 | void cpuset_unlock(void) | |
2368 | { | |
3d3f26a7 | 2369 | mutex_unlock(&callback_mutex); |
505970b9 PJ |
2370 | } |
2371 | ||
825a46af PJ |
2372 | /** |
2373 | * cpuset_mem_spread_node() - On which node to begin search for a page | |
2374 | * | |
2375 | * If a task is marked PF_SPREAD_PAGE or PF_SPREAD_SLAB (as for | |
2376 | * tasks in a cpuset with is_spread_page or is_spread_slab set), | |
2377 | * and if the memory allocation used cpuset_mem_spread_node() | |
2378 | * to determine on which node to start looking, as it will for | |
2379 | * certain page cache or slab cache pages such as used for file | |
2380 | * system buffers and inode caches, then instead of starting on the | |
2381 | * local node to look for a free page, rather spread the starting | |
2382 | * node around the tasks mems_allowed nodes. | |
2383 | * | |
2384 | * We don't have to worry about the returned node being offline | |
2385 | * because "it can't happen", and even if it did, it would be ok. | |
2386 | * | |
2387 | * The routines calling guarantee_online_mems() are careful to | |
2388 | * only set nodes in task->mems_allowed that are online. So it | |
2389 | * should not be possible for the following code to return an | |
2390 | * offline node. But if it did, that would be ok, as this routine | |
2391 | * is not returning the node where the allocation must be, only | |
2392 | * the node where the search should start. The zonelist passed to | |
2393 | * __alloc_pages() will include all nodes. If the slab allocator | |
2394 | * is passed an offline node, it will fall back to the local node. | |
2395 | * See kmem_cache_alloc_node(). | |
2396 | */ | |
2397 | ||
2398 | int cpuset_mem_spread_node(void) | |
2399 | { | |
2400 | int node; | |
2401 | ||
2402 | node = next_node(current->cpuset_mem_spread_rotor, current->mems_allowed); | |
2403 | if (node == MAX_NUMNODES) | |
2404 | node = first_node(current->mems_allowed); | |
2405 | current->cpuset_mem_spread_rotor = node; | |
2406 | return node; | |
2407 | } | |
2408 | EXPORT_SYMBOL_GPL(cpuset_mem_spread_node); | |
2409 | ||
ef08e3b4 | 2410 | /** |
bbe373f2 DR |
2411 | * cpuset_mems_allowed_intersects - Does @tsk1's mems_allowed intersect @tsk2's? |
2412 | * @tsk1: pointer to task_struct of some task. | |
2413 | * @tsk2: pointer to task_struct of some other task. | |
2414 | * | |
2415 | * Description: Return true if @tsk1's mems_allowed intersects the | |
2416 | * mems_allowed of @tsk2. Used by the OOM killer to determine if | |
2417 | * one of the task's memory usage might impact the memory available | |
2418 | * to the other. | |
ef08e3b4 PJ |
2419 | **/ |
2420 | ||
bbe373f2 DR |
2421 | int cpuset_mems_allowed_intersects(const struct task_struct *tsk1, |
2422 | const struct task_struct *tsk2) | |
ef08e3b4 | 2423 | { |
bbe373f2 | 2424 | return nodes_intersects(tsk1->mems_allowed, tsk2->mems_allowed); |
ef08e3b4 PJ |
2425 | } |
2426 | ||
75aa1994 DR |
2427 | /** |
2428 | * cpuset_print_task_mems_allowed - prints task's cpuset and mems_allowed | |
2429 | * @task: pointer to task_struct of some task. | |
2430 | * | |
2431 | * Description: Prints @task's name, cpuset name, and cached copy of its | |
2432 | * mems_allowed to the kernel log. Must hold task_lock(task) to allow | |
2433 | * dereferencing task_cs(task). | |
2434 | */ | |
2435 | void cpuset_print_task_mems_allowed(struct task_struct *tsk) | |
2436 | { | |
2437 | struct dentry *dentry; | |
2438 | ||
2439 | dentry = task_cs(tsk)->css.cgroup->dentry; | |
2440 | spin_lock(&cpuset_buffer_lock); | |
2441 | snprintf(cpuset_name, CPUSET_NAME_LEN, | |
2442 | dentry ? (const char *)dentry->d_name.name : "/"); | |
2443 | nodelist_scnprintf(cpuset_nodelist, CPUSET_NODELIST_LEN, | |
2444 | tsk->mems_allowed); | |
2445 | printk(KERN_INFO "%s cpuset=%s mems_allowed=%s\n", | |
2446 | tsk->comm, cpuset_name, cpuset_nodelist); | |
2447 | spin_unlock(&cpuset_buffer_lock); | |
2448 | } | |
2449 | ||
3e0d98b9 PJ |
2450 | /* |
2451 | * Collection of memory_pressure is suppressed unless | |
2452 | * this flag is enabled by writing "1" to the special | |
2453 | * cpuset file 'memory_pressure_enabled' in the root cpuset. | |
2454 | */ | |
2455 | ||
c5b2aff8 | 2456 | int cpuset_memory_pressure_enabled __read_mostly; |
3e0d98b9 PJ |
2457 | |
2458 | /** | |
2459 | * cpuset_memory_pressure_bump - keep stats of per-cpuset reclaims. | |
2460 | * | |
2461 | * Keep a running average of the rate of synchronous (direct) | |
2462 | * page reclaim efforts initiated by tasks in each cpuset. | |
2463 | * | |
2464 | * This represents the rate at which some task in the cpuset | |
2465 | * ran low on memory on all nodes it was allowed to use, and | |
2466 | * had to enter the kernels page reclaim code in an effort to | |
2467 | * create more free memory by tossing clean pages or swapping | |
2468 | * or writing dirty pages. | |
2469 | * | |
2470 | * Display to user space in the per-cpuset read-only file | |
2471 | * "memory_pressure". Value displayed is an integer | |
2472 | * representing the recent rate of entry into the synchronous | |
2473 | * (direct) page reclaim by any task attached to the cpuset. | |
2474 | **/ | |
2475 | ||
2476 | void __cpuset_memory_pressure_bump(void) | |
2477 | { | |
3e0d98b9 | 2478 | task_lock(current); |
8793d854 | 2479 | fmeter_markevent(&task_cs(current)->fmeter); |
3e0d98b9 PJ |
2480 | task_unlock(current); |
2481 | } | |
2482 | ||
8793d854 | 2483 | #ifdef CONFIG_PROC_PID_CPUSET |
1da177e4 LT |
2484 | /* |
2485 | * proc_cpuset_show() | |
2486 | * - Print tasks cpuset path into seq_file. | |
2487 | * - Used for /proc/<pid>/cpuset. | |
053199ed PJ |
2488 | * - No need to task_lock(tsk) on this tsk->cpuset reference, as it |
2489 | * doesn't really matter if tsk->cpuset changes after we read it, | |
c8d9c90c | 2490 | * and we take cgroup_mutex, keeping cpuset_attach() from changing it |
2df167a3 | 2491 | * anyway. |
1da177e4 | 2492 | */ |
029190c5 | 2493 | static int proc_cpuset_show(struct seq_file *m, void *unused_v) |
1da177e4 | 2494 | { |
13b41b09 | 2495 | struct pid *pid; |
1da177e4 LT |
2496 | struct task_struct *tsk; |
2497 | char *buf; | |
8793d854 | 2498 | struct cgroup_subsys_state *css; |
99f89551 | 2499 | int retval; |
1da177e4 | 2500 | |
99f89551 | 2501 | retval = -ENOMEM; |
1da177e4 LT |
2502 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
2503 | if (!buf) | |
99f89551 EB |
2504 | goto out; |
2505 | ||
2506 | retval = -ESRCH; | |
13b41b09 EB |
2507 | pid = m->private; |
2508 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
99f89551 EB |
2509 | if (!tsk) |
2510 | goto out_free; | |
1da177e4 | 2511 | |
99f89551 | 2512 | retval = -EINVAL; |
8793d854 PM |
2513 | cgroup_lock(); |
2514 | css = task_subsys_state(tsk, cpuset_subsys_id); | |
2515 | retval = cgroup_path(css->cgroup, buf, PAGE_SIZE); | |
1da177e4 | 2516 | if (retval < 0) |
99f89551 | 2517 | goto out_unlock; |
1da177e4 LT |
2518 | seq_puts(m, buf); |
2519 | seq_putc(m, '\n'); | |
99f89551 | 2520 | out_unlock: |
8793d854 | 2521 | cgroup_unlock(); |
99f89551 EB |
2522 | put_task_struct(tsk); |
2523 | out_free: | |
1da177e4 | 2524 | kfree(buf); |
99f89551 | 2525 | out: |
1da177e4 LT |
2526 | return retval; |
2527 | } | |
2528 | ||
2529 | static int cpuset_open(struct inode *inode, struct file *file) | |
2530 | { | |
13b41b09 EB |
2531 | struct pid *pid = PROC_I(inode)->pid; |
2532 | return single_open(file, proc_cpuset_show, pid); | |
1da177e4 LT |
2533 | } |
2534 | ||
9a32144e | 2535 | const struct file_operations proc_cpuset_operations = { |
1da177e4 LT |
2536 | .open = cpuset_open, |
2537 | .read = seq_read, | |
2538 | .llseek = seq_lseek, | |
2539 | .release = single_release, | |
2540 | }; | |
8793d854 | 2541 | #endif /* CONFIG_PROC_PID_CPUSET */ |
1da177e4 LT |
2542 | |
2543 | /* Display task cpus_allowed, mems_allowed in /proc/<pid>/status file. */ | |
df5f8314 EB |
2544 | void cpuset_task_status_allowed(struct seq_file *m, struct task_struct *task) |
2545 | { | |
2546 | seq_printf(m, "Cpus_allowed:\t"); | |
30e8e136 | 2547 | seq_cpumask(m, &task->cpus_allowed); |
df5f8314 | 2548 | seq_printf(m, "\n"); |
39106dcf | 2549 | seq_printf(m, "Cpus_allowed_list:\t"); |
30e8e136 | 2550 | seq_cpumask_list(m, &task->cpus_allowed); |
39106dcf | 2551 | seq_printf(m, "\n"); |
df5f8314 | 2552 | seq_printf(m, "Mems_allowed:\t"); |
30e8e136 | 2553 | seq_nodemask(m, &task->mems_allowed); |
df5f8314 | 2554 | seq_printf(m, "\n"); |
39106dcf | 2555 | seq_printf(m, "Mems_allowed_list:\t"); |
30e8e136 | 2556 | seq_nodemask_list(m, &task->mems_allowed); |
39106dcf | 2557 | seq_printf(m, "\n"); |
1da177e4 | 2558 | } |