4 * Processor and Memory placement constraints for sets of tasks.
6 * Copyright (C) 2003 BULL SA.
7 * Copyright (C) 2004-2007 Silicon Graphics, Inc.
8 * Copyright (C) 2006 Google, Inc
10 * Portions derived from Patrick Mochel's sysfs code.
11 * sysfs is Copyright (c) 2001-3 Patrick Mochel
13 * 2003-10-10 Written by Simon Derr.
14 * 2003-10-22 Updates by Stephen Hemminger.
15 * 2004 May-July Rework by Paul Jackson.
16 * 2006 Rework by Paul Menage to use generic cgroups
18 * This file is subject to the terms and conditions of the GNU General Public
19 * License. See the file COPYING in the main directory of the Linux
20 * distribution for more details.
23 #include <linux/cpu.h>
24 #include <linux/cpumask.h>
25 #include <linux/cpuset.h>
26 #include <linux/err.h>
27 #include <linux/errno.h>
28 #include <linux/file.h>
30 #include <linux/init.h>
31 #include <linux/interrupt.h>
32 #include <linux/kernel.h>
33 #include <linux/kmod.h>
34 #include <linux/list.h>
35 #include <linux/mempolicy.h>
37 #include <linux/module.h>
38 #include <linux/mount.h>
39 #include <linux/namei.h>
40 #include <linux/pagemap.h>
41 #include <linux/prio_heap.h>
42 #include <linux/proc_fs.h>
43 #include <linux/rcupdate.h>
44 #include <linux/sched.h>
45 #include <linux/seq_file.h>
46 #include <linux/security.h>
47 #include <linux/slab.h>
48 #include <linux/spinlock.h>
49 #include <linux/stat.h>
50 #include <linux/string.h>
51 #include <linux/time.h>
52 #include <linux/backing-dev.h>
53 #include <linux/sort.h>
55 #include <asm/uaccess.h>
56 #include <asm/atomic.h>
57 #include <linux/mutex.h>
58 #include <linux/kfifo.h>
59 #include <linux/workqueue.h>
60 #include <linux/cgroup.h>
63 * Tracks how many cpusets are currently defined in system.
64 * When there is only one cpuset (the root cpuset) we can
65 * short circuit some hooks.
67 int number_of_cpusets __read_mostly;
69 /* Retrieve the cpuset from a cgroup */
70 struct cgroup_subsys cpuset_subsys;
73 /* See "Frequency meter" comments, below. */
76 int cnt; /* unprocessed events count */
77 int val; /* most recent output value */
78 time_t time; /* clock (secs) when val computed */
79 spinlock_t lock; /* guards read or write of above */
83 struct cgroup_subsys_state css;
85 unsigned long flags; /* "unsigned long" so bitops work */
86 cpumask_t cpus_allowed; /* CPUs allowed to tasks in cpuset */
87 nodemask_t mems_allowed; /* Memory Nodes allowed to tasks */
89 struct cpuset *parent; /* my parent */
92 * Copy of global cpuset_mems_generation as of the most
93 * recent time this cpuset changed its mems_allowed.
97 struct fmeter fmeter; /* memory_pressure filter */
99 /* partition number for rebuild_sched_domains() */
102 /* used for walking a cpuset heirarchy */
103 struct list_head stack_list;
106 /* Retrieve the cpuset for a cgroup */
107 static inline struct cpuset *cgroup_cs(struct cgroup *cont)
109 return container_of(cgroup_subsys_state(cont, cpuset_subsys_id),
113 /* Retrieve the cpuset for a task */
114 static inline struct cpuset *task_cs(struct task_struct *task)
116 return container_of(task_subsys_state(task, cpuset_subsys_id),
119 struct cpuset_hotplug_scanner {
120 struct cgroup_scanner scan;
124 /* bits in struct cpuset flags field */
129 CS_SCHED_LOAD_BALANCE,
134 /* convenient tests for these bits */
135 static inline int is_cpu_exclusive(const struct cpuset *cs)
137 return test_bit(CS_CPU_EXCLUSIVE, &cs->flags);
140 static inline int is_mem_exclusive(const struct cpuset *cs)
142 return test_bit(CS_MEM_EXCLUSIVE, &cs->flags);
145 static inline int is_sched_load_balance(const struct cpuset *cs)
147 return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
150 static inline int is_memory_migrate(const struct cpuset *cs)
152 return test_bit(CS_MEMORY_MIGRATE, &cs->flags);
155 static inline int is_spread_page(const struct cpuset *cs)
157 return test_bit(CS_SPREAD_PAGE, &cs->flags);
160 static inline int is_spread_slab(const struct cpuset *cs)
162 return test_bit(CS_SPREAD_SLAB, &cs->flags);
166 * Increment this integer everytime any cpuset changes its
167 * mems_allowed value. Users of cpusets can track this generation
168 * number, and avoid having to lock and reload mems_allowed unless
169 * the cpuset they're using changes generation.
171 * A single, global generation is needed because attach_task() could
172 * reattach a task to a different cpuset, which must not have its
173 * generation numbers aliased with those of that tasks previous cpuset.
175 * Generations are needed for mems_allowed because one task cannot
176 * modify anothers memory placement. So we must enable every task,
177 * on every visit to __alloc_pages(), to efficiently check whether
178 * its current->cpuset->mems_allowed has changed, requiring an update
179 * of its current->mems_allowed.
181 * Since cpuset_mems_generation is guarded by manage_mutex,
182 * there is no need to mark it atomic.
184 static int cpuset_mems_generation;
186 static struct cpuset top_cpuset = {
187 .flags = ((1 << CS_CPU_EXCLUSIVE) | (1 << CS_MEM_EXCLUSIVE)),
188 .cpus_allowed = CPU_MASK_ALL,
189 .mems_allowed = NODE_MASK_ALL,
193 * We have two global cpuset mutexes below. They can nest.
194 * It is ok to first take manage_mutex, then nest callback_mutex. We also
195 * require taking task_lock() when dereferencing a tasks cpuset pointer.
196 * See "The task_lock() exception", at the end of this comment.
198 * A task must hold both mutexes to modify cpusets. If a task
199 * holds manage_mutex, then it blocks others wanting that mutex,
200 * ensuring that it is the only task able to also acquire callback_mutex
201 * and be able to modify cpusets. It can perform various checks on
202 * the cpuset structure first, knowing nothing will change. It can
203 * also allocate memory while just holding manage_mutex. While it is
204 * performing these checks, various callback routines can briefly
205 * acquire callback_mutex to query cpusets. Once it is ready to make
206 * the changes, it takes callback_mutex, blocking everyone else.
208 * Calls to the kernel memory allocator can not be made while holding
209 * callback_mutex, as that would risk double tripping on callback_mutex
210 * from one of the callbacks into the cpuset code from within
213 * If a task is only holding callback_mutex, then it has read-only
216 * The task_struct fields mems_allowed and mems_generation may only
217 * be accessed in the context of that task, so require no locks.
219 * Any task can increment and decrement the count field without lock.
220 * So in general, code holding manage_mutex or callback_mutex can't rely
221 * on the count field not changing. However, if the count goes to
222 * zero, then only attach_task(), which holds both mutexes, can
223 * increment it again. Because a count of zero means that no tasks
224 * are currently attached, therefore there is no way a task attached
225 * to that cpuset can fork (the other way to increment the count).
226 * So code holding manage_mutex or callback_mutex can safely assume that
227 * if the count is zero, it will stay zero. Similarly, if a task
228 * holds manage_mutex or callback_mutex on a cpuset with zero count, it
229 * knows that the cpuset won't be removed, as cpuset_rmdir() needs
230 * both of those mutexes.
232 * The cpuset_common_file_write handler for operations that modify
233 * the cpuset hierarchy holds manage_mutex across the entire operation,
234 * single threading all such cpuset modifications across the system.
236 * The cpuset_common_file_read() handlers only hold callback_mutex across
237 * small pieces of code, such as when reading out possibly multi-word
238 * cpumasks and nodemasks.
240 * The fork and exit callbacks cpuset_fork() and cpuset_exit(), don't
241 * (usually) take either mutex. These are the two most performance
242 * critical pieces of code here. The exception occurs on cpuset_exit(),
243 * when a task in a notify_on_release cpuset exits. Then manage_mutex
244 * is taken, and if the cpuset count is zero, a usermode call made
245 * to /sbin/cpuset_release_agent with the name of the cpuset (path
246 * relative to the root of cpuset file system) as the argument.
248 * A cpuset can only be deleted if both its 'count' of using tasks
249 * is zero, and its list of 'children' cpusets is empty. Since all
250 * tasks in the system use _some_ cpuset, and since there is always at
251 * least one task in the system (init), therefore, top_cpuset
252 * always has either children cpusets and/or using tasks. So we don't
253 * need a special hack to ensure that top_cpuset cannot be deleted.
255 * The above "Tale of Two Semaphores" would be complete, but for:
257 * The task_lock() exception
259 * The need for this exception arises from the action of attach_task(),
260 * which overwrites one tasks cpuset pointer with another. It does
261 * so using both mutexes, however there are several performance
262 * critical places that need to reference task->cpuset without the
263 * expense of grabbing a system global mutex. Therefore except as
264 * noted below, when dereferencing or, as in attach_task(), modifying
265 * a tasks cpuset pointer we use task_lock(), which acts on a spinlock
266 * (task->alloc_lock) already in the task_struct routinely used for
269 * P.S. One more locking exception. RCU is used to guard the
270 * update of a tasks cpuset pointer by attach_task() and the
271 * access of task->cpuset->mems_generation via that pointer in
272 * the routine cpuset_update_task_memory_state().
275 static DEFINE_MUTEX(callback_mutex);
277 /* This is ugly, but preserves the userspace API for existing cpuset
278 * users. If someone tries to mount the "cpuset" filesystem, we
279 * silently switch it to mount "cgroup" instead */
280 static int cpuset_get_sb(struct file_system_type *fs_type,
281 int flags, const char *unused_dev_name,
282 void *data, struct vfsmount *mnt)
284 struct file_system_type *cgroup_fs = get_fs_type("cgroup");
289 "release_agent=/sbin/cpuset_release_agent";
290 ret = cgroup_fs->get_sb(cgroup_fs, flags,
291 unused_dev_name, mountopts, mnt);
292 put_filesystem(cgroup_fs);
297 static struct file_system_type cpuset_fs_type = {
299 .get_sb = cpuset_get_sb,
303 * Return in *pmask the portion of a cpusets's cpus_allowed that
304 * are online. If none are online, walk up the cpuset hierarchy
305 * until we find one that does have some online cpus. If we get
306 * all the way to the top and still haven't found any online cpus,
307 * return cpu_online_map. Or if passed a NULL cs from an exit'ing
308 * task, return cpu_online_map.
310 * One way or another, we guarantee to return some non-empty subset
313 * Call with callback_mutex held.
316 static void guarantee_online_cpus(const struct cpuset *cs, cpumask_t *pmask)
318 while (cs && !cpus_intersects(cs->cpus_allowed, cpu_online_map))
321 cpus_and(*pmask, cs->cpus_allowed, cpu_online_map);
323 *pmask = cpu_online_map;
324 BUG_ON(!cpus_intersects(*pmask, cpu_online_map));
328 * Return in *pmask the portion of a cpusets's mems_allowed that
329 * are online, with memory. If none are online with memory, walk
330 * up the cpuset hierarchy until we find one that does have some
331 * online mems. If we get all the way to the top and still haven't
332 * found any online mems, return node_states[N_HIGH_MEMORY].
334 * One way or another, we guarantee to return some non-empty subset
335 * of node_states[N_HIGH_MEMORY].
337 * Call with callback_mutex held.
340 static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask)
342 while (cs && !nodes_intersects(cs->mems_allowed,
343 node_states[N_HIGH_MEMORY]))
346 nodes_and(*pmask, cs->mems_allowed,
347 node_states[N_HIGH_MEMORY]);
349 *pmask = node_states[N_HIGH_MEMORY];
350 BUG_ON(!nodes_intersects(*pmask, node_states[N_HIGH_MEMORY]));
354 * cpuset_update_task_memory_state - update task memory placement
356 * If the current tasks cpusets mems_allowed changed behind our
357 * backs, update current->mems_allowed, mems_generation and task NUMA
358 * mempolicy to the new value.
360 * Task mempolicy is updated by rebinding it relative to the
361 * current->cpuset if a task has its memory placement changed.
362 * Do not call this routine if in_interrupt().
364 * Call without callback_mutex or task_lock() held. May be
365 * called with or without manage_mutex held. Thanks in part to
366 * 'the_top_cpuset_hack', the tasks cpuset pointer will never
367 * be NULL. This routine also might acquire callback_mutex and
368 * current->mm->mmap_sem during call.
370 * Reading current->cpuset->mems_generation doesn't need task_lock
371 * to guard the current->cpuset derefence, because it is guarded
372 * from concurrent freeing of current->cpuset by attach_task(),
375 * The rcu_dereference() is technically probably not needed,
376 * as I don't actually mind if I see a new cpuset pointer but
377 * an old value of mems_generation. However this really only
378 * matters on alpha systems using cpusets heavily. If I dropped
379 * that rcu_dereference(), it would save them a memory barrier.
380 * For all other arch's, rcu_dereference is a no-op anyway, and for
381 * alpha systems not using cpusets, another planned optimization,
382 * avoiding the rcu critical section for tasks in the root cpuset
383 * which is statically allocated, so can't vanish, will make this
384 * irrelevant. Better to use RCU as intended, than to engage in
385 * some cute trick to save a memory barrier that is impossible to
386 * test, for alpha systems using cpusets heavily, which might not
389 * This routine is needed to update the per-task mems_allowed data,
390 * within the tasks context, when it is trying to allocate memory
391 * (in various mm/mempolicy.c routines) and notices that some other
392 * task has been modifying its cpuset.
395 void cpuset_update_task_memory_state(void)
397 int my_cpusets_mem_gen;
398 struct task_struct *tsk = current;
401 if (task_cs(tsk) == &top_cpuset) {
402 /* Don't need rcu for top_cpuset. It's never freed. */
403 my_cpusets_mem_gen = top_cpuset.mems_generation;
406 my_cpusets_mem_gen = task_cs(current)->mems_generation;
410 if (my_cpusets_mem_gen != tsk->cpuset_mems_generation) {
411 mutex_lock(&callback_mutex);
413 cs = task_cs(tsk); /* Maybe changed when task not locked */
414 guarantee_online_mems(cs, &tsk->mems_allowed);
415 tsk->cpuset_mems_generation = cs->mems_generation;
416 if (is_spread_page(cs))
417 tsk->flags |= PF_SPREAD_PAGE;
419 tsk->flags &= ~PF_SPREAD_PAGE;
420 if (is_spread_slab(cs))
421 tsk->flags |= PF_SPREAD_SLAB;
423 tsk->flags &= ~PF_SPREAD_SLAB;
425 mutex_unlock(&callback_mutex);
426 mpol_rebind_task(tsk, &tsk->mems_allowed);
431 * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q?
433 * One cpuset is a subset of another if all its allowed CPUs and
434 * Memory Nodes are a subset of the other, and its exclusive flags
435 * are only set if the other's are set. Call holding manage_mutex.
438 static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q)
440 return cpus_subset(p->cpus_allowed, q->cpus_allowed) &&
441 nodes_subset(p->mems_allowed, q->mems_allowed) &&
442 is_cpu_exclusive(p) <= is_cpu_exclusive(q) &&
443 is_mem_exclusive(p) <= is_mem_exclusive(q);
447 * validate_change() - Used to validate that any proposed cpuset change
448 * follows the structural rules for cpusets.
450 * If we replaced the flag and mask values of the current cpuset
451 * (cur) with those values in the trial cpuset (trial), would
452 * our various subset and exclusive rules still be valid? Presumes
455 * 'cur' is the address of an actual, in-use cpuset. Operations
456 * such as list traversal that depend on the actual address of the
457 * cpuset in the list must use cur below, not trial.
459 * 'trial' is the address of bulk structure copy of cur, with
460 * perhaps one or more of the fields cpus_allowed, mems_allowed,
461 * or flags changed to new, trial values.
463 * Return 0 if valid, -errno if not.
466 static int validate_change(const struct cpuset *cur, const struct cpuset *trial)
469 struct cpuset *c, *par;
471 /* Each of our child cpusets must be a subset of us */
472 list_for_each_entry(cont, &cur->css.cgroup->children, sibling) {
473 if (!is_cpuset_subset(cgroup_cs(cont), trial))
477 /* Remaining checks don't apply to root cpuset */
478 if (cur == &top_cpuset)
483 /* We must be a subset of our parent cpuset */
484 if (!is_cpuset_subset(trial, par))
487 /* If either I or some sibling (!= me) is exclusive, we can't overlap */
488 list_for_each_entry(cont, &par->css.cgroup->children, sibling) {
490 if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) &&
492 cpus_intersects(trial->cpus_allowed, c->cpus_allowed))
494 if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) &&
496 nodes_intersects(trial->mems_allowed, c->mems_allowed))
500 /* Cpusets with tasks can't have empty cpus_allowed or mems_allowed */
501 if (cgroup_task_count(cur->css.cgroup)) {
502 if (cpus_empty(trial->cpus_allowed) ||
503 nodes_empty(trial->mems_allowed)) {
512 * Helper routine for rebuild_sched_domains().
513 * Do cpusets a, b have overlapping cpus_allowed masks?
516 static int cpusets_overlap(struct cpuset *a, struct cpuset *b)
518 return cpus_intersects(a->cpus_allowed, b->cpus_allowed);
522 * rebuild_sched_domains()
524 * If the flag 'sched_load_balance' of any cpuset with non-empty
525 * 'cpus' changes, or if the 'cpus' allowed changes in any cpuset
526 * which has that flag enabled, or if any cpuset with a non-empty
527 * 'cpus' is removed, then call this routine to rebuild the
528 * scheduler's dynamic sched domains.
530 * This routine builds a partial partition of the systems CPUs
531 * (the set of non-overlappping cpumask_t's in the array 'part'
532 * below), and passes that partial partition to the kernel/sched.c
533 * partition_sched_domains() routine, which will rebuild the
534 * schedulers load balancing domains (sched domains) as specified
535 * by that partial partition. A 'partial partition' is a set of
536 * non-overlapping subsets whose union is a subset of that set.
538 * See "What is sched_load_balance" in Documentation/cpusets.txt
539 * for a background explanation of this.
541 * Does not return errors, on the theory that the callers of this
542 * routine would rather not worry about failures to rebuild sched
543 * domains when operating in the severe memory shortage situations
544 * that could cause allocation failures below.
546 * Call with cgroup_mutex held. May take callback_mutex during
547 * call due to the kfifo_alloc() and kmalloc() calls. May nest
548 * a call to the get_online_cpus()/put_online_cpus() pair.
549 * Must not be called holding callback_mutex, because we must not
550 * call get_online_cpus() while holding callback_mutex. Elsewhere
551 * the kernel nests callback_mutex inside get_online_cpus() calls.
552 * So the reverse nesting would risk an ABBA deadlock.
554 * The three key local variables below are:
555 * q - a kfifo queue of cpuset pointers, used to implement a
556 * top-down scan of all cpusets. This scan loads a pointer
557 * to each cpuset marked is_sched_load_balance into the
558 * array 'csa'. For our purposes, rebuilding the schedulers
559 * sched domains, we can ignore !is_sched_load_balance cpusets.
560 * csa - (for CpuSet Array) Array of pointers to all the cpusets
561 * that need to be load balanced, for convenient iterative
562 * access by the subsequent code that finds the best partition,
563 * i.e the set of domains (subsets) of CPUs such that the
564 * cpus_allowed of every cpuset marked is_sched_load_balance
565 * is a subset of one of these domains, while there are as
566 * many such domains as possible, each as small as possible.
567 * doms - Conversion of 'csa' to an array of cpumasks, for passing to
568 * the kernel/sched.c routine partition_sched_domains() in a
569 * convenient format, that can be easily compared to the prior
570 * value to determine what partition elements (sched domains)
571 * were changed (added or removed.)
573 * Finding the best partition (set of domains):
574 * The triple nested loops below over i, j, k scan over the
575 * load balanced cpusets (using the array of cpuset pointers in
576 * csa[]) looking for pairs of cpusets that have overlapping
577 * cpus_allowed, but which don't have the same 'pn' partition
578 * number and gives them in the same partition number. It keeps
579 * looping on the 'restart' label until it can no longer find
582 * The union of the cpus_allowed masks from the set of
583 * all cpusets having the same 'pn' value then form the one
584 * element of the partition (one sched domain) to be passed to
585 * partition_sched_domains().
588 static void rebuild_sched_domains(void)
590 struct kfifo *q; /* queue of cpusets to be scanned */
591 struct cpuset *cp; /* scans q */
592 struct cpuset **csa; /* array of all cpuset ptrs */
593 int csn; /* how many cpuset ptrs in csa so far */
594 int i, j, k; /* indices for partition finding loops */
595 cpumask_t *doms; /* resulting partition; i.e. sched domains */
596 int ndoms; /* number of sched domains in result */
597 int nslot; /* next empty doms[] cpumask_t slot */
603 /* Special case for the 99% of systems with one, full, sched domain */
604 if (is_sched_load_balance(&top_cpuset)) {
606 doms = kmalloc(sizeof(cpumask_t), GFP_KERNEL);
609 *doms = top_cpuset.cpus_allowed;
613 q = kfifo_alloc(number_of_cpusets * sizeof(cp), GFP_KERNEL, NULL);
616 csa = kmalloc(number_of_cpusets * sizeof(cp), GFP_KERNEL);
622 __kfifo_put(q, (void *)&cp, sizeof(cp));
623 while (__kfifo_get(q, (void *)&cp, sizeof(cp))) {
625 struct cpuset *child; /* scans child cpusets of cp */
626 if (is_sched_load_balance(cp))
628 list_for_each_entry(cont, &cp->css.cgroup->children, sibling) {
629 child = cgroup_cs(cont);
630 __kfifo_put(q, (void *)&child, sizeof(cp));
634 for (i = 0; i < csn; i++)
639 /* Find the best partition (set of sched domains) */
640 for (i = 0; i < csn; i++) {
641 struct cpuset *a = csa[i];
644 for (j = 0; j < csn; j++) {
645 struct cpuset *b = csa[j];
648 if (apn != bpn && cpusets_overlap(a, b)) {
649 for (k = 0; k < csn; k++) {
650 struct cpuset *c = csa[k];
655 ndoms--; /* one less element */
661 /* Convert <csn, csa> to <ndoms, doms> */
662 doms = kmalloc(ndoms * sizeof(cpumask_t), GFP_KERNEL);
666 for (nslot = 0, i = 0; i < csn; i++) {
667 struct cpuset *a = csa[i];
671 cpumask_t *dp = doms + nslot;
673 if (nslot == ndoms) {
674 static int warnings = 10;
677 "rebuild_sched_domains confused:"
678 " nslot %d, ndoms %d, csn %d, i %d,"
680 nslot, ndoms, csn, i, apn);
687 for (j = i; j < csn; j++) {
688 struct cpuset *b = csa[j];
691 cpus_or(*dp, *dp, b->cpus_allowed);
698 BUG_ON(nslot != ndoms);
701 /* Have scheduler rebuild sched domains */
703 partition_sched_domains(ndoms, doms);
710 /* Don't kfree(doms) -- partition_sched_domains() does that. */
713 static inline int started_after_time(struct task_struct *t1,
714 struct timespec *time,
715 struct task_struct *t2)
717 int start_diff = timespec_compare(&t1->start_time, time);
718 if (start_diff > 0) {
720 } else if (start_diff < 0) {
724 * Arbitrarily, if two processes started at the same
725 * time, we'll say that the lower pointer value
726 * started first. Note that t2 may have exited by now
727 * so this may not be a valid pointer any longer, but
728 * that's fine - it still serves to distinguish
729 * between two tasks started (effectively)
736 static inline int started_after(void *p1, void *p2)
738 struct task_struct *t1 = p1;
739 struct task_struct *t2 = p2;
740 return started_after_time(t1, &t2->start_time, t2);
744 * Call with manage_mutex held. May take callback_mutex during call.
747 static int update_cpumask(struct cpuset *cs, char *buf)
749 struct cpuset trialcs;
751 int is_load_balanced;
752 struct cgroup_iter it;
753 struct cgroup *cgrp = cs->css.cgroup;
754 struct task_struct *p, *dropped;
755 /* Never dereference latest_task, since it's not refcounted */
756 struct task_struct *latest_task = NULL;
757 struct ptr_heap heap;
758 struct timespec latest_time = { 0, 0 };
760 /* top_cpuset.cpus_allowed tracks cpu_online_map; it's read-only */
761 if (cs == &top_cpuset)
767 * An empty cpus_allowed is ok iff there are no tasks in the cpuset.
768 * Since cpulist_parse() fails on an empty mask, we special case
769 * that parsing. The validate_change() call ensures that cpusets
770 * with tasks have cpus.
774 cpus_clear(trialcs.cpus_allowed);
776 retval = cpulist_parse(buf, trialcs.cpus_allowed);
780 cpus_and(trialcs.cpus_allowed, trialcs.cpus_allowed, cpu_online_map);
781 retval = validate_change(cs, &trialcs);
785 /* Nothing to do if the cpus didn't change */
786 if (cpus_equal(cs->cpus_allowed, trialcs.cpus_allowed))
788 retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, &started_after);
792 is_load_balanced = is_sched_load_balance(&trialcs);
794 mutex_lock(&callback_mutex);
795 cs->cpus_allowed = trialcs.cpus_allowed;
796 mutex_unlock(&callback_mutex);
800 * Scan tasks in the cpuset, and update the cpumasks of any
801 * that need an update. Since we can't call set_cpus_allowed()
802 * while holding tasklist_lock, gather tasks to be processed
803 * in a heap structure. If the statically-sized heap fills up,
804 * overflow tasks that started later, and in future iterations
805 * only consider tasks that started after the latest task in
806 * the previous pass. This guarantees forward progress and
807 * that we don't miss any tasks
810 cgroup_iter_start(cgrp, &it);
811 while ((p = cgroup_iter_next(cgrp, &it))) {
812 /* Only affect tasks that don't have the right cpus_allowed */
813 if (cpus_equal(p->cpus_allowed, cs->cpus_allowed))
816 * Only process tasks that started after the last task
819 if (!started_after_time(p, &latest_time, latest_task))
821 dropped = heap_insert(&heap, p);
822 if (dropped == NULL) {
824 } else if (dropped != p) {
826 put_task_struct(dropped);
829 cgroup_iter_end(cgrp, &it);
831 for (i = 0; i < heap.size; i++) {
832 struct task_struct *p = heap.ptrs[i];
834 latest_time = p->start_time;
837 set_cpus_allowed(p, cs->cpus_allowed);
841 * If we had to process any tasks at all, scan again
842 * in case some of them were in the middle of forking
843 * children that didn't notice the new cpumask
844 * restriction. Not the most efficient way to do it,
845 * but it avoids having to take callback_mutex in the
851 if (is_load_balanced)
852 rebuild_sched_domains();
860 * Migrate memory region from one set of nodes to another.
862 * Temporarilly set tasks mems_allowed to target nodes of migration,
863 * so that the migration code can allocate pages on these nodes.
865 * Call holding manage_mutex, so our current->cpuset won't change
866 * during this call, as manage_mutex holds off any attach_task()
867 * calls. Therefore we don't need to take task_lock around the
868 * call to guarantee_online_mems(), as we know no one is changing
871 * Hold callback_mutex around the two modifications of our tasks
872 * mems_allowed to synchronize with cpuset_mems_allowed().
874 * While the mm_struct we are migrating is typically from some
875 * other task, the task_struct mems_allowed that we are hacking
876 * is for our current task, which must allocate new pages for that
877 * migrating memory region.
879 * We call cpuset_update_task_memory_state() before hacking
880 * our tasks mems_allowed, so that we are assured of being in
881 * sync with our tasks cpuset, and in particular, callbacks to
882 * cpuset_update_task_memory_state() from nested page allocations
883 * won't see any mismatch of our cpuset and task mems_generation
884 * values, so won't overwrite our hacked tasks mems_allowed
888 static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from,
889 const nodemask_t *to)
891 struct task_struct *tsk = current;
893 cpuset_update_task_memory_state();
895 mutex_lock(&callback_mutex);
896 tsk->mems_allowed = *to;
897 mutex_unlock(&callback_mutex);
899 do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL);
901 mutex_lock(&callback_mutex);
902 guarantee_online_mems(task_cs(tsk),&tsk->mems_allowed);
903 mutex_unlock(&callback_mutex);
907 * Handle user request to change the 'mems' memory placement
908 * of a cpuset. Needs to validate the request, update the
909 * cpusets mems_allowed and mems_generation, and for each
910 * task in the cpuset, rebind any vma mempolicies and if
911 * the cpuset is marked 'memory_migrate', migrate the tasks
912 * pages to the new memory.
914 * Call with manage_mutex held. May take callback_mutex during call.
915 * Will take tasklist_lock, scan tasklist for tasks in cpuset cs,
916 * lock each such tasks mm->mmap_sem, scan its vma's and rebind
917 * their mempolicies to the cpusets new mems_allowed.
920 static void *cpuset_being_rebound;
922 static int update_nodemask(struct cpuset *cs, char *buf)
924 struct cpuset trialcs;
926 struct task_struct *p;
927 struct mm_struct **mmarray;
932 struct cgroup_iter it;
935 * top_cpuset.mems_allowed tracks node_stats[N_HIGH_MEMORY];
938 if (cs == &top_cpuset)
944 * An empty mems_allowed is ok iff there are no tasks in the cpuset.
945 * Since nodelist_parse() fails on an empty mask, we special case
946 * that parsing. The validate_change() call ensures that cpusets
947 * with tasks have memory.
951 nodes_clear(trialcs.mems_allowed);
953 retval = nodelist_parse(buf, trialcs.mems_allowed);
957 nodes_and(trialcs.mems_allowed, trialcs.mems_allowed,
958 node_states[N_HIGH_MEMORY]);
959 oldmem = cs->mems_allowed;
960 if (nodes_equal(oldmem, trialcs.mems_allowed)) {
961 retval = 0; /* Too easy - nothing to do */
964 retval = validate_change(cs, &trialcs);
968 mutex_lock(&callback_mutex);
969 cs->mems_allowed = trialcs.mems_allowed;
970 cs->mems_generation = cpuset_mems_generation++;
971 mutex_unlock(&callback_mutex);
973 cpuset_being_rebound = cs; /* causes mpol_copy() rebind */
975 fudge = 10; /* spare mmarray[] slots */
976 fudge += cpus_weight(cs->cpus_allowed); /* imagine one fork-bomb/cpu */
980 * Allocate mmarray[] to hold mm reference for each task
981 * in cpuset cs. Can't kmalloc GFP_KERNEL while holding
982 * tasklist_lock. We could use GFP_ATOMIC, but with a
983 * few more lines of code, we can retry until we get a big
984 * enough mmarray[] w/o using GFP_ATOMIC.
987 ntasks = cgroup_task_count(cs->css.cgroup); /* guess */
989 mmarray = kmalloc(ntasks * sizeof(*mmarray), GFP_KERNEL);
992 read_lock(&tasklist_lock); /* block fork */
993 if (cgroup_task_count(cs->css.cgroup) <= ntasks)
994 break; /* got enough */
995 read_unlock(&tasklist_lock); /* try again */
1001 /* Load up mmarray[] with mm reference for each task in cpuset. */
1002 cgroup_iter_start(cs->css.cgroup, &it);
1003 while ((p = cgroup_iter_next(cs->css.cgroup, &it))) {
1004 struct mm_struct *mm;
1008 "Cpuset mempolicy rebind incomplete.\n");
1011 mm = get_task_mm(p);
1016 cgroup_iter_end(cs->css.cgroup, &it);
1017 read_unlock(&tasklist_lock);
1020 * Now that we've dropped the tasklist spinlock, we can
1021 * rebind the vma mempolicies of each mm in mmarray[] to their
1022 * new cpuset, and release that mm. The mpol_rebind_mm()
1023 * call takes mmap_sem, which we couldn't take while holding
1024 * tasklist_lock. Forks can happen again now - the mpol_copy()
1025 * cpuset_being_rebound check will catch such forks, and rebind
1026 * their vma mempolicies too. Because we still hold the global
1027 * cpuset manage_mutex, we know that no other rebind effort will
1028 * be contending for the global variable cpuset_being_rebound.
1029 * It's ok if we rebind the same mm twice; mpol_rebind_mm()
1030 * is idempotent. Also migrate pages in each mm to new nodes.
1032 migrate = is_memory_migrate(cs);
1033 for (i = 0; i < n; i++) {
1034 struct mm_struct *mm = mmarray[i];
1036 mpol_rebind_mm(mm, &cs->mems_allowed);
1038 cpuset_migrate_mm(mm, &oldmem, &cs->mems_allowed);
1042 /* We're done rebinding vma's to this cpusets new mems_allowed. */
1044 cpuset_being_rebound = NULL;
1050 int current_cpuset_is_being_rebound(void)
1052 return task_cs(current) == cpuset_being_rebound;
1056 * Call with manage_mutex held.
1059 static int update_memory_pressure_enabled(struct cpuset *cs, char *buf)
1061 if (simple_strtoul(buf, NULL, 10) != 0)
1062 cpuset_memory_pressure_enabled = 1;
1064 cpuset_memory_pressure_enabled = 0;
1069 * update_flag - read a 0 or a 1 in a file and update associated flag
1070 * bit: the bit to update (CS_CPU_EXCLUSIVE, CS_MEM_EXCLUSIVE,
1071 * CS_SCHED_LOAD_BALANCE,
1072 * CS_NOTIFY_ON_RELEASE, CS_MEMORY_MIGRATE,
1073 * CS_SPREAD_PAGE, CS_SPREAD_SLAB)
1074 * cs: the cpuset to update
1075 * buf: the buffer where we read the 0 or 1
1077 * Call with manage_mutex held.
1080 static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs, char *buf)
1083 struct cpuset trialcs;
1085 int cpus_nonempty, balance_flag_changed;
1087 turning_on = (simple_strtoul(buf, NULL, 10) != 0);
1091 set_bit(bit, &trialcs.flags);
1093 clear_bit(bit, &trialcs.flags);
1095 err = validate_change(cs, &trialcs);
1099 cpus_nonempty = !cpus_empty(trialcs.cpus_allowed);
1100 balance_flag_changed = (is_sched_load_balance(cs) !=
1101 is_sched_load_balance(&trialcs));
1103 mutex_lock(&callback_mutex);
1104 cs->flags = trialcs.flags;
1105 mutex_unlock(&callback_mutex);
1107 if (cpus_nonempty && balance_flag_changed)
1108 rebuild_sched_domains();
1114 * Frequency meter - How fast is some event occurring?
1116 * These routines manage a digitally filtered, constant time based,
1117 * event frequency meter. There are four routines:
1118 * fmeter_init() - initialize a frequency meter.
1119 * fmeter_markevent() - called each time the event happens.
1120 * fmeter_getrate() - returns the recent rate of such events.
1121 * fmeter_update() - internal routine used to update fmeter.
1123 * A common data structure is passed to each of these routines,
1124 * which is used to keep track of the state required to manage the
1125 * frequency meter and its digital filter.
1127 * The filter works on the number of events marked per unit time.
1128 * The filter is single-pole low-pass recursive (IIR). The time unit
1129 * is 1 second. Arithmetic is done using 32-bit integers scaled to
1130 * simulate 3 decimal digits of precision (multiplied by 1000).
1132 * With an FM_COEF of 933, and a time base of 1 second, the filter
1133 * has a half-life of 10 seconds, meaning that if the events quit
1134 * happening, then the rate returned from the fmeter_getrate()
1135 * will be cut in half each 10 seconds, until it converges to zero.
1137 * It is not worth doing a real infinitely recursive filter. If more
1138 * than FM_MAXTICKS ticks have elapsed since the last filter event,
1139 * just compute FM_MAXTICKS ticks worth, by which point the level
1142 * Limit the count of unprocessed events to FM_MAXCNT, so as to avoid
1143 * arithmetic overflow in the fmeter_update() routine.
1145 * Given the simple 32 bit integer arithmetic used, this meter works
1146 * best for reporting rates between one per millisecond (msec) and
1147 * one per 32 (approx) seconds. At constant rates faster than one
1148 * per msec it maxes out at values just under 1,000,000. At constant
1149 * rates between one per msec, and one per second it will stabilize
1150 * to a value N*1000, where N is the rate of events per second.
1151 * At constant rates between one per second and one per 32 seconds,
1152 * it will be choppy, moving up on the seconds that have an event,
1153 * and then decaying until the next event. At rates slower than
1154 * about one in 32 seconds, it decays all the way back to zero between
1158 #define FM_COEF 933 /* coefficient for half-life of 10 secs */
1159 #define FM_MAXTICKS ((time_t)99) /* useless computing more ticks than this */
1160 #define FM_MAXCNT 1000000 /* limit cnt to avoid overflow */
1161 #define FM_SCALE 1000 /* faux fixed point scale */
1163 /* Initialize a frequency meter */
1164 static void fmeter_init(struct fmeter *fmp)
1169 spin_lock_init(&fmp->lock);
1172 /* Internal meter update - process cnt events and update value */
1173 static void fmeter_update(struct fmeter *fmp)
1175 time_t now = get_seconds();
1176 time_t ticks = now - fmp->time;
1181 ticks = min(FM_MAXTICKS, ticks);
1183 fmp->val = (FM_COEF * fmp->val) / FM_SCALE;
1186 fmp->val += ((FM_SCALE - FM_COEF) * fmp->cnt) / FM_SCALE;
1190 /* Process any previous ticks, then bump cnt by one (times scale). */
1191 static void fmeter_markevent(struct fmeter *fmp)
1193 spin_lock(&fmp->lock);
1195 fmp->cnt = min(FM_MAXCNT, fmp->cnt + FM_SCALE);
1196 spin_unlock(&fmp->lock);
1199 /* Process any previous ticks, then return current value. */
1200 static int fmeter_getrate(struct fmeter *fmp)
1204 spin_lock(&fmp->lock);
1207 spin_unlock(&fmp->lock);
1211 static int cpuset_can_attach(struct cgroup_subsys *ss,
1212 struct cgroup *cont, struct task_struct *tsk)
1214 struct cpuset *cs = cgroup_cs(cont);
1216 if (cpus_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed))
1219 return security_task_setscheduler(tsk, 0, NULL);
1222 static void cpuset_attach(struct cgroup_subsys *ss,
1223 struct cgroup *cont, struct cgroup *oldcont,
1224 struct task_struct *tsk)
1227 nodemask_t from, to;
1228 struct mm_struct *mm;
1229 struct cpuset *cs = cgroup_cs(cont);
1230 struct cpuset *oldcs = cgroup_cs(oldcont);
1232 mutex_lock(&callback_mutex);
1233 guarantee_online_cpus(cs, &cpus);
1234 set_cpus_allowed(tsk, cpus);
1235 mutex_unlock(&callback_mutex);
1237 from = oldcs->mems_allowed;
1238 to = cs->mems_allowed;
1239 mm = get_task_mm(tsk);
1241 mpol_rebind_mm(mm, &to);
1242 if (is_memory_migrate(cs))
1243 cpuset_migrate_mm(mm, &from, &to);
1249 /* The various types of files and directories in a cpuset file system */
1252 FILE_MEMORY_MIGRATE,
1257 FILE_SCHED_LOAD_BALANCE,
1258 FILE_MEMORY_PRESSURE_ENABLED,
1259 FILE_MEMORY_PRESSURE,
1262 } cpuset_filetype_t;
1264 static ssize_t cpuset_common_file_write(struct cgroup *cont,
1267 const char __user *userbuf,
1268 size_t nbytes, loff_t *unused_ppos)
1270 struct cpuset *cs = cgroup_cs(cont);
1271 cpuset_filetype_t type = cft->private;
1275 /* Crude upper limit on largest legitimate cpulist user might write. */
1276 if (nbytes > 100U + 6 * max(NR_CPUS, MAX_NUMNODES))
1279 /* +1 for nul-terminator */
1280 if ((buffer = kmalloc(nbytes + 1, GFP_KERNEL)) == 0)
1283 if (copy_from_user(buffer, userbuf, nbytes)) {
1287 buffer[nbytes] = 0; /* nul-terminate */
1291 if (cgroup_is_removed(cont)) {
1298 retval = update_cpumask(cs, buffer);
1301 retval = update_nodemask(cs, buffer);
1303 case FILE_CPU_EXCLUSIVE:
1304 retval = update_flag(CS_CPU_EXCLUSIVE, cs, buffer);
1306 case FILE_MEM_EXCLUSIVE:
1307 retval = update_flag(CS_MEM_EXCLUSIVE, cs, buffer);
1309 case FILE_SCHED_LOAD_BALANCE:
1310 retval = update_flag(CS_SCHED_LOAD_BALANCE, cs, buffer);
1312 case FILE_MEMORY_MIGRATE:
1313 retval = update_flag(CS_MEMORY_MIGRATE, cs, buffer);
1315 case FILE_MEMORY_PRESSURE_ENABLED:
1316 retval = update_memory_pressure_enabled(cs, buffer);
1318 case FILE_MEMORY_PRESSURE:
1321 case FILE_SPREAD_PAGE:
1322 retval = update_flag(CS_SPREAD_PAGE, cs, buffer);
1323 cs->mems_generation = cpuset_mems_generation++;
1325 case FILE_SPREAD_SLAB:
1326 retval = update_flag(CS_SPREAD_SLAB, cs, buffer);
1327 cs->mems_generation = cpuset_mems_generation++;
1344 * These ascii lists should be read in a single call, by using a user
1345 * buffer large enough to hold the entire map. If read in smaller
1346 * chunks, there is no guarantee of atomicity. Since the display format
1347 * used, list of ranges of sequential numbers, is variable length,
1348 * and since these maps can change value dynamically, one could read
1349 * gibberish by doing partial reads while a list was changing.
1350 * A single large read to a buffer that crosses a page boundary is
1351 * ok, because the result being copied to user land is not recomputed
1352 * across a page fault.
1355 static int cpuset_sprintf_cpulist(char *page, struct cpuset *cs)
1359 mutex_lock(&callback_mutex);
1360 mask = cs->cpus_allowed;
1361 mutex_unlock(&callback_mutex);
1363 return cpulist_scnprintf(page, PAGE_SIZE, mask);
1366 static int cpuset_sprintf_memlist(char *page, struct cpuset *cs)
1370 mutex_lock(&callback_mutex);
1371 mask = cs->mems_allowed;
1372 mutex_unlock(&callback_mutex);
1374 return nodelist_scnprintf(page, PAGE_SIZE, mask);
1377 static ssize_t cpuset_common_file_read(struct cgroup *cont,
1381 size_t nbytes, loff_t *ppos)
1383 struct cpuset *cs = cgroup_cs(cont);
1384 cpuset_filetype_t type = cft->private;
1389 if (!(page = (char *)__get_free_page(GFP_TEMPORARY)))
1396 s += cpuset_sprintf_cpulist(s, cs);
1399 s += cpuset_sprintf_memlist(s, cs);
1401 case FILE_CPU_EXCLUSIVE:
1402 *s++ = is_cpu_exclusive(cs) ? '1' : '0';
1404 case FILE_MEM_EXCLUSIVE:
1405 *s++ = is_mem_exclusive(cs) ? '1' : '0';
1407 case FILE_SCHED_LOAD_BALANCE:
1408 *s++ = is_sched_load_balance(cs) ? '1' : '0';
1410 case FILE_MEMORY_MIGRATE:
1411 *s++ = is_memory_migrate(cs) ? '1' : '0';
1413 case FILE_MEMORY_PRESSURE_ENABLED:
1414 *s++ = cpuset_memory_pressure_enabled ? '1' : '0';
1416 case FILE_MEMORY_PRESSURE:
1417 s += sprintf(s, "%d", fmeter_getrate(&cs->fmeter));
1419 case FILE_SPREAD_PAGE:
1420 *s++ = is_spread_page(cs) ? '1' : '0';
1422 case FILE_SPREAD_SLAB:
1423 *s++ = is_spread_slab(cs) ? '1' : '0';
1431 retval = simple_read_from_buffer(buf, nbytes, ppos, page, s - page);
1433 free_page((unsigned long)page);
1442 * for the common functions, 'private' gives the type of file
1445 static struct cftype cft_cpus = {
1447 .read = cpuset_common_file_read,
1448 .write = cpuset_common_file_write,
1449 .private = FILE_CPULIST,
1452 static struct cftype cft_mems = {
1454 .read = cpuset_common_file_read,
1455 .write = cpuset_common_file_write,
1456 .private = FILE_MEMLIST,
1459 static struct cftype cft_cpu_exclusive = {
1460 .name = "cpu_exclusive",
1461 .read = cpuset_common_file_read,
1462 .write = cpuset_common_file_write,
1463 .private = FILE_CPU_EXCLUSIVE,
1466 static struct cftype cft_mem_exclusive = {
1467 .name = "mem_exclusive",
1468 .read = cpuset_common_file_read,
1469 .write = cpuset_common_file_write,
1470 .private = FILE_MEM_EXCLUSIVE,
1473 static struct cftype cft_sched_load_balance = {
1474 .name = "sched_load_balance",
1475 .read = cpuset_common_file_read,
1476 .write = cpuset_common_file_write,
1477 .private = FILE_SCHED_LOAD_BALANCE,
1480 static struct cftype cft_memory_migrate = {
1481 .name = "memory_migrate",
1482 .read = cpuset_common_file_read,
1483 .write = cpuset_common_file_write,
1484 .private = FILE_MEMORY_MIGRATE,
1487 static struct cftype cft_memory_pressure_enabled = {
1488 .name = "memory_pressure_enabled",
1489 .read = cpuset_common_file_read,
1490 .write = cpuset_common_file_write,
1491 .private = FILE_MEMORY_PRESSURE_ENABLED,
1494 static struct cftype cft_memory_pressure = {
1495 .name = "memory_pressure",
1496 .read = cpuset_common_file_read,
1497 .write = cpuset_common_file_write,
1498 .private = FILE_MEMORY_PRESSURE,
1501 static struct cftype cft_spread_page = {
1502 .name = "memory_spread_page",
1503 .read = cpuset_common_file_read,
1504 .write = cpuset_common_file_write,
1505 .private = FILE_SPREAD_PAGE,
1508 static struct cftype cft_spread_slab = {
1509 .name = "memory_spread_slab",
1510 .read = cpuset_common_file_read,
1511 .write = cpuset_common_file_write,
1512 .private = FILE_SPREAD_SLAB,
1515 static int cpuset_populate(struct cgroup_subsys *ss, struct cgroup *cont)
1519 if ((err = cgroup_add_file(cont, ss, &cft_cpus)) < 0)
1521 if ((err = cgroup_add_file(cont, ss, &cft_mems)) < 0)
1523 if ((err = cgroup_add_file(cont, ss, &cft_cpu_exclusive)) < 0)
1525 if ((err = cgroup_add_file(cont, ss, &cft_mem_exclusive)) < 0)
1527 if ((err = cgroup_add_file(cont, ss, &cft_memory_migrate)) < 0)
1529 if ((err = cgroup_add_file(cont, ss, &cft_sched_load_balance)) < 0)
1531 if ((err = cgroup_add_file(cont, ss, &cft_memory_pressure)) < 0)
1533 if ((err = cgroup_add_file(cont, ss, &cft_spread_page)) < 0)
1535 if ((err = cgroup_add_file(cont, ss, &cft_spread_slab)) < 0)
1537 /* memory_pressure_enabled is in root cpuset only */
1538 if (err == 0 && !cont->parent)
1539 err = cgroup_add_file(cont, ss,
1540 &cft_memory_pressure_enabled);
1545 * post_clone() is called at the end of cgroup_clone().
1546 * 'cgroup' was just created automatically as a result of
1547 * a cgroup_clone(), and the current task is about to
1548 * be moved into 'cgroup'.
1550 * Currently we refuse to set up the cgroup - thereby
1551 * refusing the task to be entered, and as a result refusing
1552 * the sys_unshare() or clone() which initiated it - if any
1553 * sibling cpusets have exclusive cpus or mem.
1555 * If this becomes a problem for some users who wish to
1556 * allow that scenario, then cpuset_post_clone() could be
1557 * changed to grant parent->cpus_allowed-sibling_cpus_exclusive
1558 * (and likewise for mems) to the new cgroup.
1560 static void cpuset_post_clone(struct cgroup_subsys *ss,
1561 struct cgroup *cgroup)
1563 struct cgroup *parent, *child;
1564 struct cpuset *cs, *parent_cs;
1566 parent = cgroup->parent;
1567 list_for_each_entry(child, &parent->children, sibling) {
1568 cs = cgroup_cs(child);
1569 if (is_mem_exclusive(cs) || is_cpu_exclusive(cs))
1572 cs = cgroup_cs(cgroup);
1573 parent_cs = cgroup_cs(parent);
1575 cs->mems_allowed = parent_cs->mems_allowed;
1576 cs->cpus_allowed = parent_cs->cpus_allowed;
1581 * cpuset_create - create a cpuset
1582 * parent: cpuset that will be parent of the new cpuset.
1583 * name: name of the new cpuset. Will be strcpy'ed.
1584 * mode: mode to set on new inode
1586 * Must be called with the mutex on the parent inode held
1589 static struct cgroup_subsys_state *cpuset_create(
1590 struct cgroup_subsys *ss,
1591 struct cgroup *cont)
1594 struct cpuset *parent;
1596 if (!cont->parent) {
1597 /* This is early initialization for the top cgroup */
1598 top_cpuset.mems_generation = cpuset_mems_generation++;
1599 return &top_cpuset.css;
1601 parent = cgroup_cs(cont->parent);
1602 cs = kmalloc(sizeof(*cs), GFP_KERNEL);
1604 return ERR_PTR(-ENOMEM);
1606 cpuset_update_task_memory_state();
1608 if (is_spread_page(parent))
1609 set_bit(CS_SPREAD_PAGE, &cs->flags);
1610 if (is_spread_slab(parent))
1611 set_bit(CS_SPREAD_SLAB, &cs->flags);
1612 set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
1613 cs->cpus_allowed = CPU_MASK_NONE;
1614 cs->mems_allowed = NODE_MASK_NONE;
1615 cs->mems_generation = cpuset_mems_generation++;
1616 fmeter_init(&cs->fmeter);
1618 cs->parent = parent;
1619 number_of_cpusets++;
1624 * Locking note on the strange update_flag() call below:
1626 * If the cpuset being removed has its flag 'sched_load_balance'
1627 * enabled, then simulate turning sched_load_balance off, which
1628 * will call rebuild_sched_domains(). The get_online_cpus()
1629 * call in rebuild_sched_domains() must not be made while holding
1630 * callback_mutex. Elsewhere the kernel nests callback_mutex inside
1631 * get_online_cpus() calls. So the reverse nesting would risk an
1635 static void cpuset_destroy(struct cgroup_subsys *ss, struct cgroup *cont)
1637 struct cpuset *cs = cgroup_cs(cont);
1639 cpuset_update_task_memory_state();
1641 if (is_sched_load_balance(cs))
1642 update_flag(CS_SCHED_LOAD_BALANCE, cs, "0");
1644 number_of_cpusets--;
1648 struct cgroup_subsys cpuset_subsys = {
1650 .create = cpuset_create,
1651 .destroy = cpuset_destroy,
1652 .can_attach = cpuset_can_attach,
1653 .attach = cpuset_attach,
1654 .populate = cpuset_populate,
1655 .post_clone = cpuset_post_clone,
1656 .subsys_id = cpuset_subsys_id,
1661 * cpuset_init_early - just enough so that the calls to
1662 * cpuset_update_task_memory_state() in early init code
1666 int __init cpuset_init_early(void)
1668 top_cpuset.mems_generation = cpuset_mems_generation++;
1674 * cpuset_init - initialize cpusets at system boot
1676 * Description: Initialize top_cpuset and the cpuset internal file system,
1679 int __init cpuset_init(void)
1683 top_cpuset.cpus_allowed = CPU_MASK_ALL;
1684 top_cpuset.mems_allowed = NODE_MASK_ALL;
1686 fmeter_init(&top_cpuset.fmeter);
1687 top_cpuset.mems_generation = cpuset_mems_generation++;
1688 set_bit(CS_SCHED_LOAD_BALANCE, &top_cpuset.flags);
1690 err = register_filesystem(&cpuset_fs_type);
1694 number_of_cpusets = 1;
1699 * cpuset_do_move_task - move a given task to another cpuset
1700 * @tsk: pointer to task_struct the task to move
1701 * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner
1703 * Called by cgroup_scan_tasks() for each task in a cgroup.
1704 * Return nonzero to stop the walk through the tasks.
1706 void cpuset_do_move_task(struct task_struct *tsk, struct cgroup_scanner *scan)
1708 struct cpuset_hotplug_scanner *chsp;
1710 chsp = container_of(scan, struct cpuset_hotplug_scanner, scan);
1711 cgroup_attach_task(chsp->to, tsk);
1715 * move_member_tasks_to_cpuset - move tasks from one cpuset to another
1716 * @from: cpuset in which the tasks currently reside
1717 * @to: cpuset to which the tasks will be moved
1719 * Called with manage_sem held
1720 * callback_mutex must not be held, as attach_task() will take it.
1722 * The cgroup_scan_tasks() function will scan all the tasks in a cgroup,
1723 * calling callback functions for each.
1725 static void move_member_tasks_to_cpuset(struct cpuset *from, struct cpuset *to)
1727 struct cpuset_hotplug_scanner scan;
1729 scan.scan.cg = from->css.cgroup;
1730 scan.scan.test_task = NULL; /* select all tasks in cgroup */
1731 scan.scan.process_task = cpuset_do_move_task;
1732 scan.scan.heap = NULL;
1733 scan.to = to->css.cgroup;
1735 if (cgroup_scan_tasks((struct cgroup_scanner *)&scan))
1736 printk(KERN_ERR "move_member_tasks_to_cpuset: "
1737 "cgroup_scan_tasks failed\n");
1741 * If common_cpu_mem_hotplug_unplug(), below, unplugs any CPUs
1742 * or memory nodes, we need to walk over the cpuset hierarchy,
1743 * removing that CPU or node from all cpusets. If this removes the
1744 * last CPU or node from a cpuset, then move the tasks in the empty
1745 * cpuset to its next-highest non-empty parent.
1747 * The parent cpuset has some superset of the 'mems' nodes that the
1748 * newly empty cpuset held, so no migration of memory is necessary.
1750 * Called with both manage_sem and callback_sem held
1752 static void remove_tasks_in_empty_cpuset(struct cpuset *cs)
1754 struct cpuset *parent;
1756 /* the cgroup's css_sets list is in use if there are tasks
1757 in the cpuset; the list is empty if there are none;
1758 the cs->css.refcnt seems always 0 */
1759 if (list_empty(&cs->css.cgroup->css_sets))
1763 * Find its next-highest non-empty parent, (top cpuset
1764 * has online cpus, so can't be empty).
1766 parent = cs->parent;
1767 while (cpus_empty(parent->cpus_allowed)) {
1769 * this empty cpuset should now be considered to
1770 * have been used, and therefore eligible for
1771 * release when empty (if it is notify_on_release)
1773 parent = parent->parent;
1776 move_member_tasks_to_cpuset(cs, parent);
1780 * Walk the specified cpuset subtree and look for empty cpusets.
1781 * The tasks of such cpuset must be moved to a parent cpuset.
1783 * Note that such a notify_on_release cpuset must have had, at some time,
1784 * member tasks or cpuset descendants and cpus and memory, before it can
1785 * be a candidate for release.
1787 * Called with manage_mutex held. We take callback_mutex to modify
1788 * cpus_allowed and mems_allowed.
1790 * This walk processes the tree from top to bottom, completing one layer
1791 * before dropping down to the next. It always processes a node before
1792 * any of its children.
1794 * For now, since we lack memory hot unplug, we'll never see a cpuset
1795 * that has tasks along with an empty 'mems'. But if we did see such
1796 * a cpuset, we'd handle it just like we do if its 'cpus' was empty.
1798 static void scan_for_empty_cpusets(const struct cpuset *root)
1800 struct cpuset *cp; /* scans cpusets being updated */
1801 struct cpuset *child; /* scans child cpusets of cp */
1802 struct list_head queue;
1803 struct cgroup *cont;
1805 INIT_LIST_HEAD(&queue);
1807 list_add_tail((struct list_head *)&root->stack_list, &queue);
1809 mutex_lock(&callback_mutex);
1810 while (!list_empty(&queue)) {
1811 cp = container_of(queue.next, struct cpuset, stack_list);
1812 list_del(queue.next);
1813 list_for_each_entry(cont, &cp->css.cgroup->children, sibling) {
1814 child = cgroup_cs(cont);
1815 list_add_tail(&child->stack_list, &queue);
1817 cont = cp->css.cgroup;
1818 /* Remove offline cpus and mems from this cpuset. */
1819 cpus_and(cp->cpus_allowed, cp->cpus_allowed, cpu_online_map);
1820 nodes_and(cp->mems_allowed, cp->mems_allowed,
1821 node_states[N_HIGH_MEMORY]);
1822 if ((cpus_empty(cp->cpus_allowed) ||
1823 nodes_empty(cp->mems_allowed))) {
1824 /* Move tasks from the empty cpuset to a parent */
1825 mutex_unlock(&callback_mutex);
1826 remove_tasks_in_empty_cpuset(cp);
1827 mutex_lock(&callback_mutex);
1830 mutex_unlock(&callback_mutex);
1835 * The cpus_allowed and mems_allowed nodemasks in the top_cpuset track
1836 * cpu_online_map and node_states[N_HIGH_MEMORY]. Force the top cpuset to
1837 * track what's online after any CPU or memory node hotplug or unplug event.
1839 * Since there are two callers of this routine, one for CPU hotplug
1840 * events and one for memory node hotplug events, we could have coded
1841 * two separate routines here. We code it as a single common routine
1842 * in order to minimize text size.
1845 static void common_cpu_mem_hotplug_unplug(void)
1849 top_cpuset.cpus_allowed = cpu_online_map;
1850 top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY];
1851 scan_for_empty_cpusets(&top_cpuset);
1857 * The top_cpuset tracks what CPUs and Memory Nodes are online,
1858 * period. This is necessary in order to make cpusets transparent
1859 * (of no affect) on systems that are actively using CPU hotplug
1860 * but making no active use of cpusets.
1862 * This routine ensures that top_cpuset.cpus_allowed tracks
1863 * cpu_online_map on each CPU hotplug (cpuhp) event.
1866 static int cpuset_handle_cpuhp(struct notifier_block *unused_nb,
1867 unsigned long phase, void *unused_cpu)
1869 if (phase == CPU_DYING || phase == CPU_DYING_FROZEN)
1872 common_cpu_mem_hotplug_unplug();
1876 #ifdef CONFIG_MEMORY_HOTPLUG
1878 * Keep top_cpuset.mems_allowed tracking node_states[N_HIGH_MEMORY].
1879 * Call this routine anytime after you change
1880 * node_states[N_HIGH_MEMORY].
1881 * See also the previous routine cpuset_handle_cpuhp().
1884 void cpuset_track_online_nodes(void)
1886 common_cpu_mem_hotplug_unplug();
1891 * cpuset_init_smp - initialize cpus_allowed
1893 * Description: Finish top cpuset after cpu, node maps are initialized
1896 void __init cpuset_init_smp(void)
1898 top_cpuset.cpus_allowed = cpu_online_map;
1899 top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY];
1901 hotcpu_notifier(cpuset_handle_cpuhp, 0);
1906 * cpuset_cpus_allowed - return cpus_allowed mask from a tasks cpuset.
1907 * @tsk: pointer to task_struct from which to obtain cpuset->cpus_allowed.
1909 * Description: Returns the cpumask_t cpus_allowed of the cpuset
1910 * attached to the specified @tsk. Guaranteed to return some non-empty
1911 * subset of cpu_online_map, even if this means going outside the
1915 cpumask_t cpuset_cpus_allowed(struct task_struct *tsk)
1919 mutex_lock(&callback_mutex);
1920 mask = cpuset_cpus_allowed_locked(tsk);
1921 mutex_unlock(&callback_mutex);
1927 * cpuset_cpus_allowed_locked - return cpus_allowed mask from a tasks cpuset.
1928 * Must be called with callback_mutex held.
1930 cpumask_t cpuset_cpus_allowed_locked(struct task_struct *tsk)
1935 guarantee_online_cpus(task_cs(tsk), &mask);
1941 void cpuset_init_current_mems_allowed(void)
1943 current->mems_allowed = NODE_MASK_ALL;
1947 * cpuset_mems_allowed - return mems_allowed mask from a tasks cpuset.
1948 * @tsk: pointer to task_struct from which to obtain cpuset->mems_allowed.
1950 * Description: Returns the nodemask_t mems_allowed of the cpuset
1951 * attached to the specified @tsk. Guaranteed to return some non-empty
1952 * subset of node_states[N_HIGH_MEMORY], even if this means going outside the
1956 nodemask_t cpuset_mems_allowed(struct task_struct *tsk)
1960 mutex_lock(&callback_mutex);
1962 guarantee_online_mems(task_cs(tsk), &mask);
1964 mutex_unlock(&callback_mutex);
1970 * cpuset_zonelist_valid_mems_allowed - check zonelist vs. curremt mems_allowed
1971 * @zl: the zonelist to be checked
1973 * Are any of the nodes on zonelist zl allowed in current->mems_allowed?
1975 int cpuset_zonelist_valid_mems_allowed(struct zonelist *zl)
1979 for (i = 0; zl->zones[i]; i++) {
1980 int nid = zone_to_nid(zl->zones[i]);
1982 if (node_isset(nid, current->mems_allowed))
1989 * nearest_exclusive_ancestor() - Returns the nearest mem_exclusive
1990 * ancestor to the specified cpuset. Call holding callback_mutex.
1991 * If no ancestor is mem_exclusive (an unusual configuration), then
1992 * returns the root cpuset.
1994 static const struct cpuset *nearest_exclusive_ancestor(const struct cpuset *cs)
1996 while (!is_mem_exclusive(cs) && cs->parent)
2002 * cpuset_zone_allowed_softwall - Can we allocate on zone z's memory node?
2003 * @z: is this zone on an allowed node?
2004 * @gfp_mask: memory allocation flags
2006 * If we're in interrupt, yes, we can always allocate. If
2007 * __GFP_THISNODE is set, yes, we can always allocate. If zone
2008 * z's node is in our tasks mems_allowed, yes. If it's not a
2009 * __GFP_HARDWALL request and this zone's nodes is in the nearest
2010 * mem_exclusive cpuset ancestor to this tasks cpuset, yes.
2011 * If the task has been OOM killed and has access to memory reserves
2012 * as specified by the TIF_MEMDIE flag, yes.
2015 * If __GFP_HARDWALL is set, cpuset_zone_allowed_softwall()
2016 * reduces to cpuset_zone_allowed_hardwall(). Otherwise,
2017 * cpuset_zone_allowed_softwall() might sleep, and might allow a zone
2018 * from an enclosing cpuset.
2020 * cpuset_zone_allowed_hardwall() only handles the simpler case of
2021 * hardwall cpusets, and never sleeps.
2023 * The __GFP_THISNODE placement logic is really handled elsewhere,
2024 * by forcibly using a zonelist starting at a specified node, and by
2025 * (in get_page_from_freelist()) refusing to consider the zones for
2026 * any node on the zonelist except the first. By the time any such
2027 * calls get to this routine, we should just shut up and say 'yes'.
2029 * GFP_USER allocations are marked with the __GFP_HARDWALL bit,
2030 * and do not allow allocations outside the current tasks cpuset
2031 * unless the task has been OOM killed as is marked TIF_MEMDIE.
2032 * GFP_KERNEL allocations are not so marked, so can escape to the
2033 * nearest enclosing mem_exclusive ancestor cpuset.
2035 * Scanning up parent cpusets requires callback_mutex. The
2036 * __alloc_pages() routine only calls here with __GFP_HARDWALL bit
2037 * _not_ set if it's a GFP_KERNEL allocation, and all nodes in the
2038 * current tasks mems_allowed came up empty on the first pass over
2039 * the zonelist. So only GFP_KERNEL allocations, if all nodes in the
2040 * cpuset are short of memory, might require taking the callback_mutex
2043 * The first call here from mm/page_alloc:get_page_from_freelist()
2044 * has __GFP_HARDWALL set in gfp_mask, enforcing hardwall cpusets,
2045 * so no allocation on a node outside the cpuset is allowed (unless
2046 * in interrupt, of course).
2048 * The second pass through get_page_from_freelist() doesn't even call
2049 * here for GFP_ATOMIC calls. For those calls, the __alloc_pages()
2050 * variable 'wait' is not set, and the bit ALLOC_CPUSET is not set
2051 * in alloc_flags. That logic and the checks below have the combined
2053 * in_interrupt - any node ok (current task context irrelevant)
2054 * GFP_ATOMIC - any node ok
2055 * TIF_MEMDIE - any node ok
2056 * GFP_KERNEL - any node in enclosing mem_exclusive cpuset ok
2057 * GFP_USER - only nodes in current tasks mems allowed ok.
2060 * Don't call cpuset_zone_allowed_softwall if you can't sleep, unless you
2061 * pass in the __GFP_HARDWALL flag set in gfp_flag, which disables
2062 * the code that might scan up ancestor cpusets and sleep.
2065 int __cpuset_zone_allowed_softwall(struct zone *z, gfp_t gfp_mask)
2067 int node; /* node that zone z is on */
2068 const struct cpuset *cs; /* current cpuset ancestors */
2069 int allowed; /* is allocation in zone z allowed? */
2071 if (in_interrupt() || (gfp_mask & __GFP_THISNODE))
2073 node = zone_to_nid(z);
2074 might_sleep_if(!(gfp_mask & __GFP_HARDWALL));
2075 if (node_isset(node, current->mems_allowed))
2078 * Allow tasks that have access to memory reserves because they have
2079 * been OOM killed to get memory anywhere.
2081 if (unlikely(test_thread_flag(TIF_MEMDIE)))
2083 if (gfp_mask & __GFP_HARDWALL) /* If hardwall request, stop here */
2086 if (current->flags & PF_EXITING) /* Let dying task have memory */
2089 /* Not hardwall and node outside mems_allowed: scan up cpusets */
2090 mutex_lock(&callback_mutex);
2093 cs = nearest_exclusive_ancestor(task_cs(current));
2094 task_unlock(current);
2096 allowed = node_isset(node, cs->mems_allowed);
2097 mutex_unlock(&callback_mutex);
2102 * cpuset_zone_allowed_hardwall - Can we allocate on zone z's memory node?
2103 * @z: is this zone on an allowed node?
2104 * @gfp_mask: memory allocation flags
2106 * If we're in interrupt, yes, we can always allocate.
2107 * If __GFP_THISNODE is set, yes, we can always allocate. If zone
2108 * z's node is in our tasks mems_allowed, yes. If the task has been
2109 * OOM killed and has access to memory reserves as specified by the
2110 * TIF_MEMDIE flag, yes. Otherwise, no.
2112 * The __GFP_THISNODE placement logic is really handled elsewhere,
2113 * by forcibly using a zonelist starting at a specified node, and by
2114 * (in get_page_from_freelist()) refusing to consider the zones for
2115 * any node on the zonelist except the first. By the time any such
2116 * calls get to this routine, we should just shut up and say 'yes'.
2118 * Unlike the cpuset_zone_allowed_softwall() variant, above,
2119 * this variant requires that the zone be in the current tasks
2120 * mems_allowed or that we're in interrupt. It does not scan up the
2121 * cpuset hierarchy for the nearest enclosing mem_exclusive cpuset.
2125 int __cpuset_zone_allowed_hardwall(struct zone *z, gfp_t gfp_mask)
2127 int node; /* node that zone z is on */
2129 if (in_interrupt() || (gfp_mask & __GFP_THISNODE))
2131 node = zone_to_nid(z);
2132 if (node_isset(node, current->mems_allowed))
2135 * Allow tasks that have access to memory reserves because they have
2136 * been OOM killed to get memory anywhere.
2138 if (unlikely(test_thread_flag(TIF_MEMDIE)))
2144 * cpuset_lock - lock out any changes to cpuset structures
2146 * The out of memory (oom) code needs to mutex_lock cpusets
2147 * from being changed while it scans the tasklist looking for a
2148 * task in an overlapping cpuset. Expose callback_mutex via this
2149 * cpuset_lock() routine, so the oom code can lock it, before
2150 * locking the task list. The tasklist_lock is a spinlock, so
2151 * must be taken inside callback_mutex.
2154 void cpuset_lock(void)
2156 mutex_lock(&callback_mutex);
2160 * cpuset_unlock - release lock on cpuset changes
2162 * Undo the lock taken in a previous cpuset_lock() call.
2165 void cpuset_unlock(void)
2167 mutex_unlock(&callback_mutex);
2171 * cpuset_mem_spread_node() - On which node to begin search for a page
2173 * If a task is marked PF_SPREAD_PAGE or PF_SPREAD_SLAB (as for
2174 * tasks in a cpuset with is_spread_page or is_spread_slab set),
2175 * and if the memory allocation used cpuset_mem_spread_node()
2176 * to determine on which node to start looking, as it will for
2177 * certain page cache or slab cache pages such as used for file
2178 * system buffers and inode caches, then instead of starting on the
2179 * local node to look for a free page, rather spread the starting
2180 * node around the tasks mems_allowed nodes.
2182 * We don't have to worry about the returned node being offline
2183 * because "it can't happen", and even if it did, it would be ok.
2185 * The routines calling guarantee_online_mems() are careful to
2186 * only set nodes in task->mems_allowed that are online. So it
2187 * should not be possible for the following code to return an
2188 * offline node. But if it did, that would be ok, as this routine
2189 * is not returning the node where the allocation must be, only
2190 * the node where the search should start. The zonelist passed to
2191 * __alloc_pages() will include all nodes. If the slab allocator
2192 * is passed an offline node, it will fall back to the local node.
2193 * See kmem_cache_alloc_node().
2196 int cpuset_mem_spread_node(void)
2200 node = next_node(current->cpuset_mem_spread_rotor, current->mems_allowed);
2201 if (node == MAX_NUMNODES)
2202 node = first_node(current->mems_allowed);
2203 current->cpuset_mem_spread_rotor = node;
2206 EXPORT_SYMBOL_GPL(cpuset_mem_spread_node);
2209 * cpuset_mems_allowed_intersects - Does @tsk1's mems_allowed intersect @tsk2's?
2210 * @tsk1: pointer to task_struct of some task.
2211 * @tsk2: pointer to task_struct of some other task.
2213 * Description: Return true if @tsk1's mems_allowed intersects the
2214 * mems_allowed of @tsk2. Used by the OOM killer to determine if
2215 * one of the task's memory usage might impact the memory available
2219 int cpuset_mems_allowed_intersects(const struct task_struct *tsk1,
2220 const struct task_struct *tsk2)
2222 return nodes_intersects(tsk1->mems_allowed, tsk2->mems_allowed);
2226 * Collection of memory_pressure is suppressed unless
2227 * this flag is enabled by writing "1" to the special
2228 * cpuset file 'memory_pressure_enabled' in the root cpuset.
2231 int cpuset_memory_pressure_enabled __read_mostly;
2234 * cpuset_memory_pressure_bump - keep stats of per-cpuset reclaims.
2236 * Keep a running average of the rate of synchronous (direct)
2237 * page reclaim efforts initiated by tasks in each cpuset.
2239 * This represents the rate at which some task in the cpuset
2240 * ran low on memory on all nodes it was allowed to use, and
2241 * had to enter the kernels page reclaim code in an effort to
2242 * create more free memory by tossing clean pages or swapping
2243 * or writing dirty pages.
2245 * Display to user space in the per-cpuset read-only file
2246 * "memory_pressure". Value displayed is an integer
2247 * representing the recent rate of entry into the synchronous
2248 * (direct) page reclaim by any task attached to the cpuset.
2251 void __cpuset_memory_pressure_bump(void)
2254 fmeter_markevent(&task_cs(current)->fmeter);
2255 task_unlock(current);
2258 #ifdef CONFIG_PROC_PID_CPUSET
2260 * proc_cpuset_show()
2261 * - Print tasks cpuset path into seq_file.
2262 * - Used for /proc/<pid>/cpuset.
2263 * - No need to task_lock(tsk) on this tsk->cpuset reference, as it
2264 * doesn't really matter if tsk->cpuset changes after we read it,
2265 * and we take manage_mutex, keeping attach_task() from changing it
2266 * anyway. No need to check that tsk->cpuset != NULL, thanks to
2267 * the_top_cpuset_hack in cpuset_exit(), which sets an exiting tasks
2268 * cpuset to top_cpuset.
2270 static int proc_cpuset_show(struct seq_file *m, void *unused_v)
2273 struct task_struct *tsk;
2275 struct cgroup_subsys_state *css;
2279 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
2285 tsk = get_pid_task(pid, PIDTYPE_PID);
2291 css = task_subsys_state(tsk, cpuset_subsys_id);
2292 retval = cgroup_path(css->cgroup, buf, PAGE_SIZE);
2299 put_task_struct(tsk);
2306 static int cpuset_open(struct inode *inode, struct file *file)
2308 struct pid *pid = PROC_I(inode)->pid;
2309 return single_open(file, proc_cpuset_show, pid);
2312 const struct file_operations proc_cpuset_operations = {
2313 .open = cpuset_open,
2315 .llseek = seq_lseek,
2316 .release = single_release,
2318 #endif /* CONFIG_PROC_PID_CPUSET */
2320 /* Display task cpus_allowed, mems_allowed in /proc/<pid>/status file. */
2321 char *cpuset_task_status_allowed(struct task_struct *task, char *buffer)
2323 buffer += sprintf(buffer, "Cpus_allowed:\t");
2324 buffer += cpumask_scnprintf(buffer, PAGE_SIZE, task->cpus_allowed);
2325 buffer += sprintf(buffer, "\n");
2326 buffer += sprintf(buffer, "Mems_allowed:\t");
2327 buffer += nodemask_scnprintf(buffer, PAGE_SIZE, task->mems_allowed);
2328 buffer += sprintf(buffer, "\n");