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/proc_fs.h>
42 #include <linux/rcupdate.h>
43 #include <linux/sched.h>
44 #include <linux/seq_file.h>
45 #include <linux/security.h>
46 #include <linux/slab.h>
47 #include <linux/spinlock.h>
48 #include <linux/stat.h>
49 #include <linux/string.h>
50 #include <linux/time.h>
51 #include <linux/backing-dev.h>
52 #include <linux/sort.h>
54 #include <asm/uaccess.h>
55 #include <asm/atomic.h>
56 #include <linux/mutex.h>
57 #include <linux/kfifo.h>
58 #include <linux/workqueue.h>
59 #include <linux/cgroup.h>
62 * Tracks how many cpusets are currently defined in system.
63 * When there is only one cpuset (the root cpuset) we can
64 * short circuit some hooks.
66 int number_of_cpusets __read_mostly;
68 /* Forward declare cgroup structures */
69 struct cgroup_subsys cpuset_subsys;
72 /* See "Frequency meter" comments, below. */
75 int cnt; /* unprocessed events count */
76 int val; /* most recent output value */
77 time_t time; /* clock (secs) when val computed */
78 spinlock_t lock; /* guards read or write of above */
82 struct cgroup_subsys_state css;
84 unsigned long flags; /* "unsigned long" so bitops work */
85 cpumask_t cpus_allowed; /* CPUs allowed to tasks in cpuset */
86 nodemask_t mems_allowed; /* Memory Nodes allowed to tasks */
88 struct cpuset *parent; /* my parent */
91 * Copy of global cpuset_mems_generation as of the most
92 * recent time this cpuset changed its mems_allowed.
96 struct fmeter fmeter; /* memory_pressure filter */
98 /* partition number for rebuild_sched_domains() */
101 /* for custom sched domain */
102 int relax_domain_level;
104 /* used for walking a cpuset heirarchy */
105 struct list_head stack_list;
108 /* Retrieve the cpuset for a cgroup */
109 static inline struct cpuset *cgroup_cs(struct cgroup *cont)
111 return container_of(cgroup_subsys_state(cont, cpuset_subsys_id),
115 /* Retrieve the cpuset for a task */
116 static inline struct cpuset *task_cs(struct task_struct *task)
118 return container_of(task_subsys_state(task, cpuset_subsys_id),
121 struct cpuset_hotplug_scanner {
122 struct cgroup_scanner scan;
126 /* bits in struct cpuset flags field */
132 CS_SCHED_LOAD_BALANCE,
137 /* convenient tests for these bits */
138 static inline int is_cpu_exclusive(const struct cpuset *cs)
140 return test_bit(CS_CPU_EXCLUSIVE, &cs->flags);
143 static inline int is_mem_exclusive(const struct cpuset *cs)
145 return test_bit(CS_MEM_EXCLUSIVE, &cs->flags);
148 static inline int is_mem_hardwall(const struct cpuset *cs)
150 return test_bit(CS_MEM_HARDWALL, &cs->flags);
153 static inline int is_sched_load_balance(const struct cpuset *cs)
155 return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
158 static inline int is_memory_migrate(const struct cpuset *cs)
160 return test_bit(CS_MEMORY_MIGRATE, &cs->flags);
163 static inline int is_spread_page(const struct cpuset *cs)
165 return test_bit(CS_SPREAD_PAGE, &cs->flags);
168 static inline int is_spread_slab(const struct cpuset *cs)
170 return test_bit(CS_SPREAD_SLAB, &cs->flags);
174 * Increment this integer everytime any cpuset changes its
175 * mems_allowed value. Users of cpusets can track this generation
176 * number, and avoid having to lock and reload mems_allowed unless
177 * the cpuset they're using changes generation.
179 * A single, global generation is needed because cpuset_attach_task() could
180 * reattach a task to a different cpuset, which must not have its
181 * generation numbers aliased with those of that tasks previous cpuset.
183 * Generations are needed for mems_allowed because one task cannot
184 * modify another's memory placement. So we must enable every task,
185 * on every visit to __alloc_pages(), to efficiently check whether
186 * its current->cpuset->mems_allowed has changed, requiring an update
187 * of its current->mems_allowed.
189 * Since writes to cpuset_mems_generation are guarded by the cgroup lock
190 * there is no need to mark it atomic.
192 static int cpuset_mems_generation;
194 static struct cpuset top_cpuset = {
195 .flags = ((1 << CS_CPU_EXCLUSIVE) | (1 << CS_MEM_EXCLUSIVE)),
196 .cpus_allowed = CPU_MASK_ALL,
197 .mems_allowed = NODE_MASK_ALL,
201 * There are two global mutexes guarding cpuset structures. The first
202 * is the main control groups cgroup_mutex, accessed via
203 * cgroup_lock()/cgroup_unlock(). The second is the cpuset-specific
204 * callback_mutex, below. They can nest. It is ok to first take
205 * cgroup_mutex, then nest callback_mutex. We also require taking
206 * task_lock() when dereferencing a task's cpuset pointer. See "The
207 * task_lock() exception", at the end of this comment.
209 * A task must hold both mutexes to modify cpusets. If a task
210 * holds cgroup_mutex, then it blocks others wanting that mutex,
211 * ensuring that it is the only task able to also acquire callback_mutex
212 * and be able to modify cpusets. It can perform various checks on
213 * the cpuset structure first, knowing nothing will change. It can
214 * also allocate memory while just holding cgroup_mutex. While it is
215 * performing these checks, various callback routines can briefly
216 * acquire callback_mutex to query cpusets. Once it is ready to make
217 * the changes, it takes callback_mutex, blocking everyone else.
219 * Calls to the kernel memory allocator can not be made while holding
220 * callback_mutex, as that would risk double tripping on callback_mutex
221 * from one of the callbacks into the cpuset code from within
224 * If a task is only holding callback_mutex, then it has read-only
227 * The task_struct fields mems_allowed and mems_generation may only
228 * be accessed in the context of that task, so require no locks.
230 * The cpuset_common_file_read() handlers only hold callback_mutex across
231 * small pieces of code, such as when reading out possibly multi-word
232 * cpumasks and nodemasks.
234 * Accessing a task's cpuset should be done in accordance with the
235 * guidelines for accessing subsystem state in kernel/cgroup.c
238 static DEFINE_MUTEX(callback_mutex);
240 /* This is ugly, but preserves the userspace API for existing cpuset
241 * users. If someone tries to mount the "cpuset" filesystem, we
242 * silently switch it to mount "cgroup" instead */
243 static int cpuset_get_sb(struct file_system_type *fs_type,
244 int flags, const char *unused_dev_name,
245 void *data, struct vfsmount *mnt)
247 struct file_system_type *cgroup_fs = get_fs_type("cgroup");
252 "release_agent=/sbin/cpuset_release_agent";
253 ret = cgroup_fs->get_sb(cgroup_fs, flags,
254 unused_dev_name, mountopts, mnt);
255 put_filesystem(cgroup_fs);
260 static struct file_system_type cpuset_fs_type = {
262 .get_sb = cpuset_get_sb,
266 * Return in *pmask the portion of a cpusets's cpus_allowed that
267 * are online. If none are online, walk up the cpuset hierarchy
268 * until we find one that does have some online cpus. If we get
269 * all the way to the top and still haven't found any online cpus,
270 * return cpu_online_map. Or if passed a NULL cs from an exit'ing
271 * task, return cpu_online_map.
273 * One way or another, we guarantee to return some non-empty subset
276 * Call with callback_mutex held.
279 static void guarantee_online_cpus(const struct cpuset *cs, cpumask_t *pmask)
281 while (cs && !cpus_intersects(cs->cpus_allowed, cpu_online_map))
284 cpus_and(*pmask, cs->cpus_allowed, cpu_online_map);
286 *pmask = cpu_online_map;
287 BUG_ON(!cpus_intersects(*pmask, cpu_online_map));
291 * Return in *pmask the portion of a cpusets's mems_allowed that
292 * are online, with memory. If none are online with memory, walk
293 * up the cpuset hierarchy until we find one that does have some
294 * online mems. If we get all the way to the top and still haven't
295 * found any online mems, return node_states[N_HIGH_MEMORY].
297 * One way or another, we guarantee to return some non-empty subset
298 * of node_states[N_HIGH_MEMORY].
300 * Call with callback_mutex held.
303 static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask)
305 while (cs && !nodes_intersects(cs->mems_allowed,
306 node_states[N_HIGH_MEMORY]))
309 nodes_and(*pmask, cs->mems_allowed,
310 node_states[N_HIGH_MEMORY]);
312 *pmask = node_states[N_HIGH_MEMORY];
313 BUG_ON(!nodes_intersects(*pmask, node_states[N_HIGH_MEMORY]));
317 * cpuset_update_task_memory_state - update task memory placement
319 * If the current tasks cpusets mems_allowed changed behind our
320 * backs, update current->mems_allowed, mems_generation and task NUMA
321 * mempolicy to the new value.
323 * Task mempolicy is updated by rebinding it relative to the
324 * current->cpuset if a task has its memory placement changed.
325 * Do not call this routine if in_interrupt().
327 * Call without callback_mutex or task_lock() held. May be
328 * called with or without cgroup_mutex held. Thanks in part to
329 * 'the_top_cpuset_hack', the task's cpuset pointer will never
330 * be NULL. This routine also might acquire callback_mutex during
333 * Reading current->cpuset->mems_generation doesn't need task_lock
334 * to guard the current->cpuset derefence, because it is guarded
335 * from concurrent freeing of current->cpuset using RCU.
337 * The rcu_dereference() is technically probably not needed,
338 * as I don't actually mind if I see a new cpuset pointer but
339 * an old value of mems_generation. However this really only
340 * matters on alpha systems using cpusets heavily. If I dropped
341 * that rcu_dereference(), it would save them a memory barrier.
342 * For all other arch's, rcu_dereference is a no-op anyway, and for
343 * alpha systems not using cpusets, another planned optimization,
344 * avoiding the rcu critical section for tasks in the root cpuset
345 * which is statically allocated, so can't vanish, will make this
346 * irrelevant. Better to use RCU as intended, than to engage in
347 * some cute trick to save a memory barrier that is impossible to
348 * test, for alpha systems using cpusets heavily, which might not
351 * This routine is needed to update the per-task mems_allowed data,
352 * within the tasks context, when it is trying to allocate memory
353 * (in various mm/mempolicy.c routines) and notices that some other
354 * task has been modifying its cpuset.
357 void cpuset_update_task_memory_state(void)
359 int my_cpusets_mem_gen;
360 struct task_struct *tsk = current;
363 if (task_cs(tsk) == &top_cpuset) {
364 /* Don't need rcu for top_cpuset. It's never freed. */
365 my_cpusets_mem_gen = top_cpuset.mems_generation;
368 my_cpusets_mem_gen = task_cs(tsk)->mems_generation;
372 if (my_cpusets_mem_gen != tsk->cpuset_mems_generation) {
373 mutex_lock(&callback_mutex);
375 cs = task_cs(tsk); /* Maybe changed when task not locked */
376 guarantee_online_mems(cs, &tsk->mems_allowed);
377 tsk->cpuset_mems_generation = cs->mems_generation;
378 if (is_spread_page(cs))
379 tsk->flags |= PF_SPREAD_PAGE;
381 tsk->flags &= ~PF_SPREAD_PAGE;
382 if (is_spread_slab(cs))
383 tsk->flags |= PF_SPREAD_SLAB;
385 tsk->flags &= ~PF_SPREAD_SLAB;
387 mutex_unlock(&callback_mutex);
388 mpol_rebind_task(tsk, &tsk->mems_allowed);
393 * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q?
395 * One cpuset is a subset of another if all its allowed CPUs and
396 * Memory Nodes are a subset of the other, and its exclusive flags
397 * are only set if the other's are set. Call holding cgroup_mutex.
400 static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q)
402 return cpus_subset(p->cpus_allowed, q->cpus_allowed) &&
403 nodes_subset(p->mems_allowed, q->mems_allowed) &&
404 is_cpu_exclusive(p) <= is_cpu_exclusive(q) &&
405 is_mem_exclusive(p) <= is_mem_exclusive(q);
409 * validate_change() - Used to validate that any proposed cpuset change
410 * follows the structural rules for cpusets.
412 * If we replaced the flag and mask values of the current cpuset
413 * (cur) with those values in the trial cpuset (trial), would
414 * our various subset and exclusive rules still be valid? Presumes
417 * 'cur' is the address of an actual, in-use cpuset. Operations
418 * such as list traversal that depend on the actual address of the
419 * cpuset in the list must use cur below, not trial.
421 * 'trial' is the address of bulk structure copy of cur, with
422 * perhaps one or more of the fields cpus_allowed, mems_allowed,
423 * or flags changed to new, trial values.
425 * Return 0 if valid, -errno if not.
428 static int validate_change(const struct cpuset *cur, const struct cpuset *trial)
431 struct cpuset *c, *par;
433 /* Each of our child cpusets must be a subset of us */
434 list_for_each_entry(cont, &cur->css.cgroup->children, sibling) {
435 if (!is_cpuset_subset(cgroup_cs(cont), trial))
439 /* Remaining checks don't apply to root cpuset */
440 if (cur == &top_cpuset)
445 /* We must be a subset of our parent cpuset */
446 if (!is_cpuset_subset(trial, par))
450 * If either I or some sibling (!= me) is exclusive, we can't
453 list_for_each_entry(cont, &par->css.cgroup->children, sibling) {
455 if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) &&
457 cpus_intersects(trial->cpus_allowed, c->cpus_allowed))
459 if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) &&
461 nodes_intersects(trial->mems_allowed, c->mems_allowed))
465 /* Cpusets with tasks can't have empty cpus_allowed or mems_allowed */
466 if (cgroup_task_count(cur->css.cgroup)) {
467 if (cpus_empty(trial->cpus_allowed) ||
468 nodes_empty(trial->mems_allowed)) {
477 * Helper routine for rebuild_sched_domains().
478 * Do cpusets a, b have overlapping cpus_allowed masks?
481 static int cpusets_overlap(struct cpuset *a, struct cpuset *b)
483 return cpus_intersects(a->cpus_allowed, b->cpus_allowed);
487 update_domain_attr(struct sched_domain_attr *dattr, struct cpuset *c)
489 if (dattr->relax_domain_level < c->relax_domain_level)
490 dattr->relax_domain_level = c->relax_domain_level;
495 update_domain_attr_tree(struct sched_domain_attr *dattr, struct cpuset *c)
499 list_add(&c->stack_list, &q);
500 while (!list_empty(&q)) {
503 struct cpuset *child;
505 cp = list_first_entry(&q, struct cpuset, stack_list);
508 if (cpus_empty(cp->cpus_allowed))
511 if (is_sched_load_balance(cp))
512 update_domain_attr(dattr, cp);
514 list_for_each_entry(cont, &cp->css.cgroup->children, sibling) {
515 child = cgroup_cs(cont);
516 list_add_tail(&child->stack_list, &q);
522 * rebuild_sched_domains()
524 * This routine will be called to rebuild the scheduler's dynamic
526 * - if the flag 'sched_load_balance' of any cpuset with non-empty
528 * - or if the 'cpus' allowed changes in any cpuset which has that
530 * - or if the 'sched_relax_domain_level' of any cpuset which has
531 * that flag enabled and with non-empty 'cpus' changes,
532 * - or if any cpuset with non-empty 'cpus' is removed,
533 * - or if a cpu gets offlined.
535 * This routine builds a partial partition of the systems CPUs
536 * (the set of non-overlappping cpumask_t's in the array 'part'
537 * below), and passes that partial partition to the kernel/sched.c
538 * partition_sched_domains() routine, which will rebuild the
539 * schedulers load balancing domains (sched domains) as specified
540 * by that partial partition. A 'partial partition' is a set of
541 * non-overlapping subsets whose union is a subset of that set.
543 * See "What is sched_load_balance" in Documentation/cpusets.txt
544 * for a background explanation of this.
546 * Does not return errors, on the theory that the callers of this
547 * routine would rather not worry about failures to rebuild sched
548 * domains when operating in the severe memory shortage situations
549 * that could cause allocation failures below.
551 * Call with cgroup_mutex held. May take callback_mutex during
552 * call due to the kfifo_alloc() and kmalloc() calls. May nest
553 * a call to the get_online_cpus()/put_online_cpus() pair.
554 * Must not be called holding callback_mutex, because we must not
555 * call get_online_cpus() while holding callback_mutex. Elsewhere
556 * the kernel nests callback_mutex inside get_online_cpus() calls.
557 * So the reverse nesting would risk an ABBA deadlock.
559 * The three key local variables below are:
560 * q - a kfifo queue of cpuset pointers, used to implement a
561 * top-down scan of all cpusets. This scan loads a pointer
562 * to each cpuset marked is_sched_load_balance into the
563 * array 'csa'. For our purposes, rebuilding the schedulers
564 * sched domains, we can ignore !is_sched_load_balance cpusets.
565 * csa - (for CpuSet Array) Array of pointers to all the cpusets
566 * that need to be load balanced, for convenient iterative
567 * access by the subsequent code that finds the best partition,
568 * i.e the set of domains (subsets) of CPUs such that the
569 * cpus_allowed of every cpuset marked is_sched_load_balance
570 * is a subset of one of these domains, while there are as
571 * many such domains as possible, each as small as possible.
572 * doms - Conversion of 'csa' to an array of cpumasks, for passing to
573 * the kernel/sched.c routine partition_sched_domains() in a
574 * convenient format, that can be easily compared to the prior
575 * value to determine what partition elements (sched domains)
576 * were changed (added or removed.)
578 * Finding the best partition (set of domains):
579 * The triple nested loops below over i, j, k scan over the
580 * load balanced cpusets (using the array of cpuset pointers in
581 * csa[]) looking for pairs of cpusets that have overlapping
582 * cpus_allowed, but which don't have the same 'pn' partition
583 * number and gives them in the same partition number. It keeps
584 * looping on the 'restart' label until it can no longer find
587 * The union of the cpus_allowed masks from the set of
588 * all cpusets having the same 'pn' value then form the one
589 * element of the partition (one sched domain) to be passed to
590 * partition_sched_domains().
593 void rebuild_sched_domains(void)
595 struct kfifo *q; /* queue of cpusets to be scanned */
596 struct cpuset *cp; /* scans q */
597 struct cpuset **csa; /* array of all cpuset ptrs */
598 int csn; /* how many cpuset ptrs in csa so far */
599 int i, j, k; /* indices for partition finding loops */
600 cpumask_t *doms; /* resulting partition; i.e. sched domains */
601 struct sched_domain_attr *dattr; /* attributes for custom domains */
602 int ndoms; /* number of sched domains in result */
603 int nslot; /* next empty doms[] cpumask_t slot */
610 /* Special case for the 99% of systems with one, full, sched domain */
611 if (is_sched_load_balance(&top_cpuset)) {
613 doms = kmalloc(sizeof(cpumask_t), GFP_KERNEL);
616 dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL);
618 *dattr = SD_ATTR_INIT;
619 update_domain_attr(dattr, &top_cpuset);
621 *doms = top_cpuset.cpus_allowed;
625 q = kfifo_alloc(number_of_cpusets * sizeof(cp), GFP_KERNEL, NULL);
628 csa = kmalloc(number_of_cpusets * sizeof(cp), GFP_KERNEL);
634 __kfifo_put(q, (void *)&cp, sizeof(cp));
635 while (__kfifo_get(q, (void *)&cp, sizeof(cp))) {
637 struct cpuset *child; /* scans child cpusets of cp */
639 if (cpus_empty(cp->cpus_allowed))
643 * All child cpusets contain a subset of the parent's cpus, so
644 * just skip them, and then we call update_domain_attr_tree()
645 * to calc relax_domain_level of the corresponding sched
648 if (is_sched_load_balance(cp)) {
653 list_for_each_entry(cont, &cp->css.cgroup->children, sibling) {
654 child = cgroup_cs(cont);
655 __kfifo_put(q, (void *)&child, sizeof(cp));
659 for (i = 0; i < csn; i++)
664 /* Find the best partition (set of sched domains) */
665 for (i = 0; i < csn; i++) {
666 struct cpuset *a = csa[i];
669 for (j = 0; j < csn; j++) {
670 struct cpuset *b = csa[j];
673 if (apn != bpn && cpusets_overlap(a, b)) {
674 for (k = 0; k < csn; k++) {
675 struct cpuset *c = csa[k];
680 ndoms--; /* one less element */
686 /* Convert <csn, csa> to <ndoms, doms> */
687 doms = kmalloc(ndoms * sizeof(cpumask_t), GFP_KERNEL);
690 dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL);
692 for (nslot = 0, i = 0; i < csn; i++) {
693 struct cpuset *a = csa[i];
697 cpumask_t *dp = doms + nslot;
699 if (nslot == ndoms) {
700 static int warnings = 10;
703 "rebuild_sched_domains confused:"
704 " nslot %d, ndoms %d, csn %d, i %d,"
706 nslot, ndoms, csn, i, apn);
714 *(dattr + nslot) = SD_ATTR_INIT;
715 for (j = i; j < csn; j++) {
716 struct cpuset *b = csa[j];
719 cpus_or(*dp, *dp, b->cpus_allowed);
722 update_domain_attr_tree(dattr
729 BUG_ON(nslot != ndoms);
732 /* Have scheduler rebuild sched domains */
734 partition_sched_domains(ndoms, doms, dattr);
741 /* Don't kfree(doms) -- partition_sched_domains() does that. */
742 /* Don't kfree(dattr) -- partition_sched_domains() does that. */
746 * cpuset_test_cpumask - test a task's cpus_allowed versus its cpuset's
748 * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner
750 * Call with cgroup_mutex held. May take callback_mutex during call.
751 * Called for each task in a cgroup by cgroup_scan_tasks().
752 * Return nonzero if this tasks's cpus_allowed mask should be changed (in other
753 * words, if its mask is not equal to its cpuset's mask).
755 static int cpuset_test_cpumask(struct task_struct *tsk,
756 struct cgroup_scanner *scan)
758 return !cpus_equal(tsk->cpus_allowed,
759 (cgroup_cs(scan->cg))->cpus_allowed);
763 * cpuset_change_cpumask - make a task's cpus_allowed the same as its cpuset's
765 * @scan: struct cgroup_scanner containing the cgroup of the task
767 * Called by cgroup_scan_tasks() for each task in a cgroup whose
768 * cpus_allowed mask needs to be changed.
770 * We don't need to re-check for the cgroup/cpuset membership, since we're
771 * holding cgroup_lock() at this point.
773 static void cpuset_change_cpumask(struct task_struct *tsk,
774 struct cgroup_scanner *scan)
776 set_cpus_allowed_ptr(tsk, &((cgroup_cs(scan->cg))->cpus_allowed));
780 * update_tasks_cpumask - Update the cpumasks of tasks in the cpuset.
781 * @cs: the cpuset in which each task's cpus_allowed mask needs to be changed
783 * Called with cgroup_mutex held
785 * The cgroup_scan_tasks() function will scan all the tasks in a cgroup,
786 * calling callback functions for each.
788 * Return 0 if successful, -errno if not.
790 static int update_tasks_cpumask(struct cpuset *cs)
792 struct cgroup_scanner scan;
793 struct ptr_heap heap;
797 * cgroup_scan_tasks() will initialize heap->gt for us.
798 * heap_init() is still needed here for we should not change
799 * cs->cpus_allowed when heap_init() fails.
801 retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL);
805 scan.cg = cs->css.cgroup;
806 scan.test_task = cpuset_test_cpumask;
807 scan.process_task = cpuset_change_cpumask;
809 retval = cgroup_scan_tasks(&scan);
816 * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it
817 * @cs: the cpuset to consider
818 * @buf: buffer of cpu numbers written to this cpuset
820 static int update_cpumask(struct cpuset *cs, const char *buf)
822 struct cpuset trialcs;
824 int is_load_balanced;
826 /* top_cpuset.cpus_allowed tracks cpu_online_map; it's read-only */
827 if (cs == &top_cpuset)
833 * An empty cpus_allowed is ok only if the cpuset has no tasks.
834 * Since cpulist_parse() fails on an empty mask, we special case
835 * that parsing. The validate_change() call ensures that cpusets
836 * with tasks have cpus.
839 cpus_clear(trialcs.cpus_allowed);
841 retval = cpulist_parse(buf, trialcs.cpus_allowed);
845 if (!cpus_subset(trialcs.cpus_allowed, cpu_online_map))
848 retval = validate_change(cs, &trialcs);
852 /* Nothing to do if the cpus didn't change */
853 if (cpus_equal(cs->cpus_allowed, trialcs.cpus_allowed))
856 is_load_balanced = is_sched_load_balance(&trialcs);
858 mutex_lock(&callback_mutex);
859 cs->cpus_allowed = trialcs.cpus_allowed;
860 mutex_unlock(&callback_mutex);
863 * Scan tasks in the cpuset, and update the cpumasks of any
864 * that need an update.
866 retval = update_tasks_cpumask(cs);
870 if (is_load_balanced)
871 rebuild_sched_domains();
878 * Migrate memory region from one set of nodes to another.
880 * Temporarilly set tasks mems_allowed to target nodes of migration,
881 * so that the migration code can allocate pages on these nodes.
883 * Call holding cgroup_mutex, so current's cpuset won't change
884 * during this call, as manage_mutex holds off any cpuset_attach()
885 * calls. Therefore we don't need to take task_lock around the
886 * call to guarantee_online_mems(), as we know no one is changing
889 * Hold callback_mutex around the two modifications of our tasks
890 * mems_allowed to synchronize with cpuset_mems_allowed().
892 * While the mm_struct we are migrating is typically from some
893 * other task, the task_struct mems_allowed that we are hacking
894 * is for our current task, which must allocate new pages for that
895 * migrating memory region.
897 * We call cpuset_update_task_memory_state() before hacking
898 * our tasks mems_allowed, so that we are assured of being in
899 * sync with our tasks cpuset, and in particular, callbacks to
900 * cpuset_update_task_memory_state() from nested page allocations
901 * won't see any mismatch of our cpuset and task mems_generation
902 * values, so won't overwrite our hacked tasks mems_allowed
906 static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from,
907 const nodemask_t *to)
909 struct task_struct *tsk = current;
911 cpuset_update_task_memory_state();
913 mutex_lock(&callback_mutex);
914 tsk->mems_allowed = *to;
915 mutex_unlock(&callback_mutex);
917 do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL);
919 mutex_lock(&callback_mutex);
920 guarantee_online_mems(task_cs(tsk),&tsk->mems_allowed);
921 mutex_unlock(&callback_mutex);
924 static void *cpuset_being_rebound;
927 * update_tasks_nodemask - Update the nodemasks of tasks in the cpuset.
928 * @cs: the cpuset in which each task's mems_allowed mask needs to be changed
929 * @oldmem: old mems_allowed of cpuset cs
931 * Called with cgroup_mutex held
932 * Return 0 if successful, -errno if not.
934 static int update_tasks_nodemask(struct cpuset *cs, const nodemask_t *oldmem)
936 struct task_struct *p;
937 struct mm_struct **mmarray;
941 struct cgroup_iter it;
944 cpuset_being_rebound = cs; /* causes mpol_dup() rebind */
946 fudge = 10; /* spare mmarray[] slots */
947 fudge += cpus_weight(cs->cpus_allowed); /* imagine one fork-bomb/cpu */
951 * Allocate mmarray[] to hold mm reference for each task
952 * in cpuset cs. Can't kmalloc GFP_KERNEL while holding
953 * tasklist_lock. We could use GFP_ATOMIC, but with a
954 * few more lines of code, we can retry until we get a big
955 * enough mmarray[] w/o using GFP_ATOMIC.
958 ntasks = cgroup_task_count(cs->css.cgroup); /* guess */
960 mmarray = kmalloc(ntasks * sizeof(*mmarray), GFP_KERNEL);
963 read_lock(&tasklist_lock); /* block fork */
964 if (cgroup_task_count(cs->css.cgroup) <= ntasks)
965 break; /* got enough */
966 read_unlock(&tasklist_lock); /* try again */
972 /* Load up mmarray[] with mm reference for each task in cpuset. */
973 cgroup_iter_start(cs->css.cgroup, &it);
974 while ((p = cgroup_iter_next(cs->css.cgroup, &it))) {
975 struct mm_struct *mm;
979 "Cpuset mempolicy rebind incomplete.\n");
987 cgroup_iter_end(cs->css.cgroup, &it);
988 read_unlock(&tasklist_lock);
991 * Now that we've dropped the tasklist spinlock, we can
992 * rebind the vma mempolicies of each mm in mmarray[] to their
993 * new cpuset, and release that mm. The mpol_rebind_mm()
994 * call takes mmap_sem, which we couldn't take while holding
995 * tasklist_lock. Forks can happen again now - the mpol_dup()
996 * cpuset_being_rebound check will catch such forks, and rebind
997 * their vma mempolicies too. Because we still hold the global
998 * cgroup_mutex, we know that no other rebind effort will
999 * be contending for the global variable cpuset_being_rebound.
1000 * It's ok if we rebind the same mm twice; mpol_rebind_mm()
1001 * is idempotent. Also migrate pages in each mm to new nodes.
1003 migrate = is_memory_migrate(cs);
1004 for (i = 0; i < n; i++) {
1005 struct mm_struct *mm = mmarray[i];
1007 mpol_rebind_mm(mm, &cs->mems_allowed);
1009 cpuset_migrate_mm(mm, oldmem, &cs->mems_allowed);
1013 /* We're done rebinding vmas to this cpuset's new mems_allowed. */
1015 cpuset_being_rebound = NULL;
1022 * Handle user request to change the 'mems' memory placement
1023 * of a cpuset. Needs to validate the request, update the
1024 * cpusets mems_allowed and mems_generation, and for each
1025 * task in the cpuset, rebind any vma mempolicies and if
1026 * the cpuset is marked 'memory_migrate', migrate the tasks
1027 * pages to the new memory.
1029 * Call with cgroup_mutex held. May take callback_mutex during call.
1030 * Will take tasklist_lock, scan tasklist for tasks in cpuset cs,
1031 * lock each such tasks mm->mmap_sem, scan its vma's and rebind
1032 * their mempolicies to the cpusets new mems_allowed.
1034 static int update_nodemask(struct cpuset *cs, const char *buf)
1036 struct cpuset trialcs;
1041 * top_cpuset.mems_allowed tracks node_stats[N_HIGH_MEMORY];
1044 if (cs == &top_cpuset)
1050 * An empty mems_allowed is ok iff there are no tasks in the cpuset.
1051 * Since nodelist_parse() fails on an empty mask, we special case
1052 * that parsing. The validate_change() call ensures that cpusets
1053 * with tasks have memory.
1056 nodes_clear(trialcs.mems_allowed);
1058 retval = nodelist_parse(buf, trialcs.mems_allowed);
1062 if (!nodes_subset(trialcs.mems_allowed,
1063 node_states[N_HIGH_MEMORY]))
1066 oldmem = cs->mems_allowed;
1067 if (nodes_equal(oldmem, trialcs.mems_allowed)) {
1068 retval = 0; /* Too easy - nothing to do */
1071 retval = validate_change(cs, &trialcs);
1075 mutex_lock(&callback_mutex);
1076 cs->mems_allowed = trialcs.mems_allowed;
1077 cs->mems_generation = cpuset_mems_generation++;
1078 mutex_unlock(&callback_mutex);
1080 retval = update_tasks_nodemask(cs, &oldmem);
1085 int current_cpuset_is_being_rebound(void)
1087 return task_cs(current) == cpuset_being_rebound;
1090 static int update_relax_domain_level(struct cpuset *cs, s64 val)
1092 if (val < -1 || val >= SD_LV_MAX)
1095 if (val != cs->relax_domain_level) {
1096 cs->relax_domain_level = val;
1097 if (!cpus_empty(cs->cpus_allowed) && is_sched_load_balance(cs))
1098 rebuild_sched_domains();
1105 * update_flag - read a 0 or a 1 in a file and update associated flag
1106 * bit: the bit to update (see cpuset_flagbits_t)
1107 * cs: the cpuset to update
1108 * turning_on: whether the flag is being set or cleared
1110 * Call with cgroup_mutex held.
1113 static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs,
1116 struct cpuset trialcs;
1118 int cpus_nonempty, balance_flag_changed;
1122 set_bit(bit, &trialcs.flags);
1124 clear_bit(bit, &trialcs.flags);
1126 err = validate_change(cs, &trialcs);
1130 cpus_nonempty = !cpus_empty(trialcs.cpus_allowed);
1131 balance_flag_changed = (is_sched_load_balance(cs) !=
1132 is_sched_load_balance(&trialcs));
1134 mutex_lock(&callback_mutex);
1135 cs->flags = trialcs.flags;
1136 mutex_unlock(&callback_mutex);
1138 if (cpus_nonempty && balance_flag_changed)
1139 rebuild_sched_domains();
1145 * Frequency meter - How fast is some event occurring?
1147 * These routines manage a digitally filtered, constant time based,
1148 * event frequency meter. There are four routines:
1149 * fmeter_init() - initialize a frequency meter.
1150 * fmeter_markevent() - called each time the event happens.
1151 * fmeter_getrate() - returns the recent rate of such events.
1152 * fmeter_update() - internal routine used to update fmeter.
1154 * A common data structure is passed to each of these routines,
1155 * which is used to keep track of the state required to manage the
1156 * frequency meter and its digital filter.
1158 * The filter works on the number of events marked per unit time.
1159 * The filter is single-pole low-pass recursive (IIR). The time unit
1160 * is 1 second. Arithmetic is done using 32-bit integers scaled to
1161 * simulate 3 decimal digits of precision (multiplied by 1000).
1163 * With an FM_COEF of 933, and a time base of 1 second, the filter
1164 * has a half-life of 10 seconds, meaning that if the events quit
1165 * happening, then the rate returned from the fmeter_getrate()
1166 * will be cut in half each 10 seconds, until it converges to zero.
1168 * It is not worth doing a real infinitely recursive filter. If more
1169 * than FM_MAXTICKS ticks have elapsed since the last filter event,
1170 * just compute FM_MAXTICKS ticks worth, by which point the level
1173 * Limit the count of unprocessed events to FM_MAXCNT, so as to avoid
1174 * arithmetic overflow in the fmeter_update() routine.
1176 * Given the simple 32 bit integer arithmetic used, this meter works
1177 * best for reporting rates between one per millisecond (msec) and
1178 * one per 32 (approx) seconds. At constant rates faster than one
1179 * per msec it maxes out at values just under 1,000,000. At constant
1180 * rates between one per msec, and one per second it will stabilize
1181 * to a value N*1000, where N is the rate of events per second.
1182 * At constant rates between one per second and one per 32 seconds,
1183 * it will be choppy, moving up on the seconds that have an event,
1184 * and then decaying until the next event. At rates slower than
1185 * about one in 32 seconds, it decays all the way back to zero between
1189 #define FM_COEF 933 /* coefficient for half-life of 10 secs */
1190 #define FM_MAXTICKS ((time_t)99) /* useless computing more ticks than this */
1191 #define FM_MAXCNT 1000000 /* limit cnt to avoid overflow */
1192 #define FM_SCALE 1000 /* faux fixed point scale */
1194 /* Initialize a frequency meter */
1195 static void fmeter_init(struct fmeter *fmp)
1200 spin_lock_init(&fmp->lock);
1203 /* Internal meter update - process cnt events and update value */
1204 static void fmeter_update(struct fmeter *fmp)
1206 time_t now = get_seconds();
1207 time_t ticks = now - fmp->time;
1212 ticks = min(FM_MAXTICKS, ticks);
1214 fmp->val = (FM_COEF * fmp->val) / FM_SCALE;
1217 fmp->val += ((FM_SCALE - FM_COEF) * fmp->cnt) / FM_SCALE;
1221 /* Process any previous ticks, then bump cnt by one (times scale). */
1222 static void fmeter_markevent(struct fmeter *fmp)
1224 spin_lock(&fmp->lock);
1226 fmp->cnt = min(FM_MAXCNT, fmp->cnt + FM_SCALE);
1227 spin_unlock(&fmp->lock);
1230 /* Process any previous ticks, then return current value. */
1231 static int fmeter_getrate(struct fmeter *fmp)
1235 spin_lock(&fmp->lock);
1238 spin_unlock(&fmp->lock);
1242 /* Called by cgroups to determine if a cpuset is usable; cgroup_mutex held */
1243 static int cpuset_can_attach(struct cgroup_subsys *ss,
1244 struct cgroup *cont, struct task_struct *tsk)
1246 struct cpuset *cs = cgroup_cs(cont);
1248 if (cpus_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed))
1250 if (tsk->flags & PF_THREAD_BOUND) {
1253 mutex_lock(&callback_mutex);
1254 mask = cs->cpus_allowed;
1255 mutex_unlock(&callback_mutex);
1256 if (!cpus_equal(tsk->cpus_allowed, mask))
1260 return security_task_setscheduler(tsk, 0, NULL);
1263 static void cpuset_attach(struct cgroup_subsys *ss,
1264 struct cgroup *cont, struct cgroup *oldcont,
1265 struct task_struct *tsk)
1268 nodemask_t from, to;
1269 struct mm_struct *mm;
1270 struct cpuset *cs = cgroup_cs(cont);
1271 struct cpuset *oldcs = cgroup_cs(oldcont);
1274 mutex_lock(&callback_mutex);
1275 guarantee_online_cpus(cs, &cpus);
1276 err = set_cpus_allowed_ptr(tsk, &cpus);
1277 mutex_unlock(&callback_mutex);
1281 from = oldcs->mems_allowed;
1282 to = cs->mems_allowed;
1283 mm = get_task_mm(tsk);
1285 mpol_rebind_mm(mm, &to);
1286 if (is_memory_migrate(cs))
1287 cpuset_migrate_mm(mm, &from, &to);
1293 /* The various types of files and directories in a cpuset file system */
1296 FILE_MEMORY_MIGRATE,
1302 FILE_SCHED_LOAD_BALANCE,
1303 FILE_SCHED_RELAX_DOMAIN_LEVEL,
1304 FILE_MEMORY_PRESSURE_ENABLED,
1305 FILE_MEMORY_PRESSURE,
1308 } cpuset_filetype_t;
1310 static int cpuset_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val)
1313 struct cpuset *cs = cgroup_cs(cgrp);
1314 cpuset_filetype_t type = cft->private;
1316 if (!cgroup_lock_live_group(cgrp))
1320 case FILE_CPU_EXCLUSIVE:
1321 retval = update_flag(CS_CPU_EXCLUSIVE, cs, val);
1323 case FILE_MEM_EXCLUSIVE:
1324 retval = update_flag(CS_MEM_EXCLUSIVE, cs, val);
1326 case FILE_MEM_HARDWALL:
1327 retval = update_flag(CS_MEM_HARDWALL, cs, val);
1329 case FILE_SCHED_LOAD_BALANCE:
1330 retval = update_flag(CS_SCHED_LOAD_BALANCE, cs, val);
1332 case FILE_MEMORY_MIGRATE:
1333 retval = update_flag(CS_MEMORY_MIGRATE, cs, val);
1335 case FILE_MEMORY_PRESSURE_ENABLED:
1336 cpuset_memory_pressure_enabled = !!val;
1338 case FILE_MEMORY_PRESSURE:
1341 case FILE_SPREAD_PAGE:
1342 retval = update_flag(CS_SPREAD_PAGE, cs, val);
1343 cs->mems_generation = cpuset_mems_generation++;
1345 case FILE_SPREAD_SLAB:
1346 retval = update_flag(CS_SPREAD_SLAB, cs, val);
1347 cs->mems_generation = cpuset_mems_generation++;
1357 static int cpuset_write_s64(struct cgroup *cgrp, struct cftype *cft, s64 val)
1360 struct cpuset *cs = cgroup_cs(cgrp);
1361 cpuset_filetype_t type = cft->private;
1363 if (!cgroup_lock_live_group(cgrp))
1367 case FILE_SCHED_RELAX_DOMAIN_LEVEL:
1368 retval = update_relax_domain_level(cs, val);
1379 * Common handling for a write to a "cpus" or "mems" file.
1381 static int cpuset_write_resmask(struct cgroup *cgrp, struct cftype *cft,
1386 if (!cgroup_lock_live_group(cgrp))
1389 switch (cft->private) {
1391 retval = update_cpumask(cgroup_cs(cgrp), buf);
1394 retval = update_nodemask(cgroup_cs(cgrp), buf);
1405 * These ascii lists should be read in a single call, by using a user
1406 * buffer large enough to hold the entire map. If read in smaller
1407 * chunks, there is no guarantee of atomicity. Since the display format
1408 * used, list of ranges of sequential numbers, is variable length,
1409 * and since these maps can change value dynamically, one could read
1410 * gibberish by doing partial reads while a list was changing.
1411 * A single large read to a buffer that crosses a page boundary is
1412 * ok, because the result being copied to user land is not recomputed
1413 * across a page fault.
1416 static int cpuset_sprintf_cpulist(char *page, struct cpuset *cs)
1420 mutex_lock(&callback_mutex);
1421 mask = cs->cpus_allowed;
1422 mutex_unlock(&callback_mutex);
1424 return cpulist_scnprintf(page, PAGE_SIZE, mask);
1427 static int cpuset_sprintf_memlist(char *page, struct cpuset *cs)
1431 mutex_lock(&callback_mutex);
1432 mask = cs->mems_allowed;
1433 mutex_unlock(&callback_mutex);
1435 return nodelist_scnprintf(page, PAGE_SIZE, mask);
1438 static ssize_t cpuset_common_file_read(struct cgroup *cont,
1442 size_t nbytes, loff_t *ppos)
1444 struct cpuset *cs = cgroup_cs(cont);
1445 cpuset_filetype_t type = cft->private;
1450 if (!(page = (char *)__get_free_page(GFP_TEMPORARY)))
1457 s += cpuset_sprintf_cpulist(s, cs);
1460 s += cpuset_sprintf_memlist(s, cs);
1468 retval = simple_read_from_buffer(buf, nbytes, ppos, page, s - page);
1470 free_page((unsigned long)page);
1474 static u64 cpuset_read_u64(struct cgroup *cont, struct cftype *cft)
1476 struct cpuset *cs = cgroup_cs(cont);
1477 cpuset_filetype_t type = cft->private;
1479 case FILE_CPU_EXCLUSIVE:
1480 return is_cpu_exclusive(cs);
1481 case FILE_MEM_EXCLUSIVE:
1482 return is_mem_exclusive(cs);
1483 case FILE_MEM_HARDWALL:
1484 return is_mem_hardwall(cs);
1485 case FILE_SCHED_LOAD_BALANCE:
1486 return is_sched_load_balance(cs);
1487 case FILE_MEMORY_MIGRATE:
1488 return is_memory_migrate(cs);
1489 case FILE_MEMORY_PRESSURE_ENABLED:
1490 return cpuset_memory_pressure_enabled;
1491 case FILE_MEMORY_PRESSURE:
1492 return fmeter_getrate(&cs->fmeter);
1493 case FILE_SPREAD_PAGE:
1494 return is_spread_page(cs);
1495 case FILE_SPREAD_SLAB:
1496 return is_spread_slab(cs);
1502 static s64 cpuset_read_s64(struct cgroup *cont, struct cftype *cft)
1504 struct cpuset *cs = cgroup_cs(cont);
1505 cpuset_filetype_t type = cft->private;
1507 case FILE_SCHED_RELAX_DOMAIN_LEVEL:
1508 return cs->relax_domain_level;
1516 * for the common functions, 'private' gives the type of file
1519 static struct cftype files[] = {
1522 .read = cpuset_common_file_read,
1523 .write_string = cpuset_write_resmask,
1524 .max_write_len = (100U + 6 * NR_CPUS),
1525 .private = FILE_CPULIST,
1530 .read = cpuset_common_file_read,
1531 .write_string = cpuset_write_resmask,
1532 .max_write_len = (100U + 6 * MAX_NUMNODES),
1533 .private = FILE_MEMLIST,
1537 .name = "cpu_exclusive",
1538 .read_u64 = cpuset_read_u64,
1539 .write_u64 = cpuset_write_u64,
1540 .private = FILE_CPU_EXCLUSIVE,
1544 .name = "mem_exclusive",
1545 .read_u64 = cpuset_read_u64,
1546 .write_u64 = cpuset_write_u64,
1547 .private = FILE_MEM_EXCLUSIVE,
1551 .name = "mem_hardwall",
1552 .read_u64 = cpuset_read_u64,
1553 .write_u64 = cpuset_write_u64,
1554 .private = FILE_MEM_HARDWALL,
1558 .name = "sched_load_balance",
1559 .read_u64 = cpuset_read_u64,
1560 .write_u64 = cpuset_write_u64,
1561 .private = FILE_SCHED_LOAD_BALANCE,
1565 .name = "sched_relax_domain_level",
1566 .read_s64 = cpuset_read_s64,
1567 .write_s64 = cpuset_write_s64,
1568 .private = FILE_SCHED_RELAX_DOMAIN_LEVEL,
1572 .name = "memory_migrate",
1573 .read_u64 = cpuset_read_u64,
1574 .write_u64 = cpuset_write_u64,
1575 .private = FILE_MEMORY_MIGRATE,
1579 .name = "memory_pressure",
1580 .read_u64 = cpuset_read_u64,
1581 .write_u64 = cpuset_write_u64,
1582 .private = FILE_MEMORY_PRESSURE,
1586 .name = "memory_spread_page",
1587 .read_u64 = cpuset_read_u64,
1588 .write_u64 = cpuset_write_u64,
1589 .private = FILE_SPREAD_PAGE,
1593 .name = "memory_spread_slab",
1594 .read_u64 = cpuset_read_u64,
1595 .write_u64 = cpuset_write_u64,
1596 .private = FILE_SPREAD_SLAB,
1600 static struct cftype cft_memory_pressure_enabled = {
1601 .name = "memory_pressure_enabled",
1602 .read_u64 = cpuset_read_u64,
1603 .write_u64 = cpuset_write_u64,
1604 .private = FILE_MEMORY_PRESSURE_ENABLED,
1607 static int cpuset_populate(struct cgroup_subsys *ss, struct cgroup *cont)
1611 err = cgroup_add_files(cont, ss, files, ARRAY_SIZE(files));
1614 /* memory_pressure_enabled is in root cpuset only */
1616 err = cgroup_add_file(cont, ss,
1617 &cft_memory_pressure_enabled);
1622 * post_clone() is called at the end of cgroup_clone().
1623 * 'cgroup' was just created automatically as a result of
1624 * a cgroup_clone(), and the current task is about to
1625 * be moved into 'cgroup'.
1627 * Currently we refuse to set up the cgroup - thereby
1628 * refusing the task to be entered, and as a result refusing
1629 * the sys_unshare() or clone() which initiated it - if any
1630 * sibling cpusets have exclusive cpus or mem.
1632 * If this becomes a problem for some users who wish to
1633 * allow that scenario, then cpuset_post_clone() could be
1634 * changed to grant parent->cpus_allowed-sibling_cpus_exclusive
1635 * (and likewise for mems) to the new cgroup. Called with cgroup_mutex
1638 static void cpuset_post_clone(struct cgroup_subsys *ss,
1639 struct cgroup *cgroup)
1641 struct cgroup *parent, *child;
1642 struct cpuset *cs, *parent_cs;
1644 parent = cgroup->parent;
1645 list_for_each_entry(child, &parent->children, sibling) {
1646 cs = cgroup_cs(child);
1647 if (is_mem_exclusive(cs) || is_cpu_exclusive(cs))
1650 cs = cgroup_cs(cgroup);
1651 parent_cs = cgroup_cs(parent);
1653 cs->mems_allowed = parent_cs->mems_allowed;
1654 cs->cpus_allowed = parent_cs->cpus_allowed;
1659 * cpuset_create - create a cpuset
1660 * ss: cpuset cgroup subsystem
1661 * cont: control group that the new cpuset will be part of
1664 static struct cgroup_subsys_state *cpuset_create(
1665 struct cgroup_subsys *ss,
1666 struct cgroup *cont)
1669 struct cpuset *parent;
1671 if (!cont->parent) {
1672 /* This is early initialization for the top cgroup */
1673 top_cpuset.mems_generation = cpuset_mems_generation++;
1674 return &top_cpuset.css;
1676 parent = cgroup_cs(cont->parent);
1677 cs = kmalloc(sizeof(*cs), GFP_KERNEL);
1679 return ERR_PTR(-ENOMEM);
1681 cpuset_update_task_memory_state();
1683 if (is_spread_page(parent))
1684 set_bit(CS_SPREAD_PAGE, &cs->flags);
1685 if (is_spread_slab(parent))
1686 set_bit(CS_SPREAD_SLAB, &cs->flags);
1687 set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
1688 cpus_clear(cs->cpus_allowed);
1689 nodes_clear(cs->mems_allowed);
1690 cs->mems_generation = cpuset_mems_generation++;
1691 fmeter_init(&cs->fmeter);
1692 cs->relax_domain_level = -1;
1694 cs->parent = parent;
1695 number_of_cpusets++;
1700 * Locking note on the strange update_flag() call below:
1702 * If the cpuset being removed has its flag 'sched_load_balance'
1703 * enabled, then simulate turning sched_load_balance off, which
1704 * will call rebuild_sched_domains(). The get_online_cpus()
1705 * call in rebuild_sched_domains() must not be made while holding
1706 * callback_mutex. Elsewhere the kernel nests callback_mutex inside
1707 * get_online_cpus() calls. So the reverse nesting would risk an
1711 static void cpuset_destroy(struct cgroup_subsys *ss, struct cgroup *cont)
1713 struct cpuset *cs = cgroup_cs(cont);
1715 cpuset_update_task_memory_state();
1717 if (is_sched_load_balance(cs))
1718 update_flag(CS_SCHED_LOAD_BALANCE, cs, 0);
1720 number_of_cpusets--;
1724 struct cgroup_subsys cpuset_subsys = {
1726 .create = cpuset_create,
1727 .destroy = cpuset_destroy,
1728 .can_attach = cpuset_can_attach,
1729 .attach = cpuset_attach,
1730 .populate = cpuset_populate,
1731 .post_clone = cpuset_post_clone,
1732 .subsys_id = cpuset_subsys_id,
1737 * cpuset_init_early - just enough so that the calls to
1738 * cpuset_update_task_memory_state() in early init code
1742 int __init cpuset_init_early(void)
1744 top_cpuset.mems_generation = cpuset_mems_generation++;
1750 * cpuset_init - initialize cpusets at system boot
1752 * Description: Initialize top_cpuset and the cpuset internal file system,
1755 int __init cpuset_init(void)
1759 cpus_setall(top_cpuset.cpus_allowed);
1760 nodes_setall(top_cpuset.mems_allowed);
1762 fmeter_init(&top_cpuset.fmeter);
1763 top_cpuset.mems_generation = cpuset_mems_generation++;
1764 set_bit(CS_SCHED_LOAD_BALANCE, &top_cpuset.flags);
1765 top_cpuset.relax_domain_level = -1;
1767 err = register_filesystem(&cpuset_fs_type);
1771 number_of_cpusets = 1;
1776 * cpuset_do_move_task - move a given task to another cpuset
1777 * @tsk: pointer to task_struct the task to move
1778 * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner
1780 * Called by cgroup_scan_tasks() for each task in a cgroup.
1781 * Return nonzero to stop the walk through the tasks.
1783 static void cpuset_do_move_task(struct task_struct *tsk,
1784 struct cgroup_scanner *scan)
1786 struct cpuset_hotplug_scanner *chsp;
1788 chsp = container_of(scan, struct cpuset_hotplug_scanner, scan);
1789 cgroup_attach_task(chsp->to, tsk);
1793 * move_member_tasks_to_cpuset - move tasks from one cpuset to another
1794 * @from: cpuset in which the tasks currently reside
1795 * @to: cpuset to which the tasks will be moved
1797 * Called with cgroup_mutex held
1798 * callback_mutex must not be held, as cpuset_attach() will take it.
1800 * The cgroup_scan_tasks() function will scan all the tasks in a cgroup,
1801 * calling callback functions for each.
1803 static void move_member_tasks_to_cpuset(struct cpuset *from, struct cpuset *to)
1805 struct cpuset_hotplug_scanner scan;
1807 scan.scan.cg = from->css.cgroup;
1808 scan.scan.test_task = NULL; /* select all tasks in cgroup */
1809 scan.scan.process_task = cpuset_do_move_task;
1810 scan.scan.heap = NULL;
1811 scan.to = to->css.cgroup;
1813 if (cgroup_scan_tasks(&scan.scan))
1814 printk(KERN_ERR "move_member_tasks_to_cpuset: "
1815 "cgroup_scan_tasks failed\n");
1819 * If common_cpu_mem_hotplug_unplug(), below, unplugs any CPUs
1820 * or memory nodes, we need to walk over the cpuset hierarchy,
1821 * removing that CPU or node from all cpusets. If this removes the
1822 * last CPU or node from a cpuset, then move the tasks in the empty
1823 * cpuset to its next-highest non-empty parent.
1825 * Called with cgroup_mutex held
1826 * callback_mutex must not be held, as cpuset_attach() will take it.
1828 static void remove_tasks_in_empty_cpuset(struct cpuset *cs)
1830 struct cpuset *parent;
1833 * The cgroup's css_sets list is in use if there are tasks
1834 * in the cpuset; the list is empty if there are none;
1835 * the cs->css.refcnt seems always 0.
1837 if (list_empty(&cs->css.cgroup->css_sets))
1841 * Find its next-highest non-empty parent, (top cpuset
1842 * has online cpus, so can't be empty).
1844 parent = cs->parent;
1845 while (cpus_empty(parent->cpus_allowed) ||
1846 nodes_empty(parent->mems_allowed))
1847 parent = parent->parent;
1849 move_member_tasks_to_cpuset(cs, parent);
1853 * Walk the specified cpuset subtree and look for empty cpusets.
1854 * The tasks of such cpuset must be moved to a parent cpuset.
1856 * Called with cgroup_mutex held. We take callback_mutex to modify
1857 * cpus_allowed and mems_allowed.
1859 * This walk processes the tree from top to bottom, completing one layer
1860 * before dropping down to the next. It always processes a node before
1861 * any of its children.
1863 * For now, since we lack memory hot unplug, we'll never see a cpuset
1864 * that has tasks along with an empty 'mems'. But if we did see such
1865 * a cpuset, we'd handle it just like we do if its 'cpus' was empty.
1867 static void scan_for_empty_cpusets(const struct cpuset *root)
1870 struct cpuset *cp; /* scans cpusets being updated */
1871 struct cpuset *child; /* scans child cpusets of cp */
1872 struct cgroup *cont;
1875 list_add_tail((struct list_head *)&root->stack_list, &queue);
1877 while (!list_empty(&queue)) {
1878 cp = list_first_entry(&queue, struct cpuset, stack_list);
1879 list_del(queue.next);
1880 list_for_each_entry(cont, &cp->css.cgroup->children, sibling) {
1881 child = cgroup_cs(cont);
1882 list_add_tail(&child->stack_list, &queue);
1885 /* Continue past cpusets with all cpus, mems online */
1886 if (cpus_subset(cp->cpus_allowed, cpu_online_map) &&
1887 nodes_subset(cp->mems_allowed, node_states[N_HIGH_MEMORY]))
1890 oldmems = cp->mems_allowed;
1892 /* Remove offline cpus and mems from this cpuset. */
1893 mutex_lock(&callback_mutex);
1894 cpus_and(cp->cpus_allowed, cp->cpus_allowed, cpu_online_map);
1895 nodes_and(cp->mems_allowed, cp->mems_allowed,
1896 node_states[N_HIGH_MEMORY]);
1897 mutex_unlock(&callback_mutex);
1899 /* Move tasks from the empty cpuset to a parent */
1900 if (cpus_empty(cp->cpus_allowed) ||
1901 nodes_empty(cp->mems_allowed))
1902 remove_tasks_in_empty_cpuset(cp);
1904 update_tasks_cpumask(cp);
1905 update_tasks_nodemask(cp, &oldmems);
1911 * The cpus_allowed and mems_allowed nodemasks in the top_cpuset track
1912 * cpu_online_map and node_states[N_HIGH_MEMORY]. Force the top cpuset to
1913 * track what's online after any CPU or memory node hotplug or unplug event.
1915 * Since there are two callers of this routine, one for CPU hotplug
1916 * events and one for memory node hotplug events, we could have coded
1917 * two separate routines here. We code it as a single common routine
1918 * in order to minimize text size.
1921 static void common_cpu_mem_hotplug_unplug(int rebuild_sd)
1925 top_cpuset.cpus_allowed = cpu_online_map;
1926 top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY];
1927 scan_for_empty_cpusets(&top_cpuset);
1930 * Scheduler destroys domains on hotplug events.
1931 * Rebuild them based on the current settings.
1934 rebuild_sched_domains();
1940 * The top_cpuset tracks what CPUs and Memory Nodes are online,
1941 * period. This is necessary in order to make cpusets transparent
1942 * (of no affect) on systems that are actively using CPU hotplug
1943 * but making no active use of cpusets.
1945 * This routine ensures that top_cpuset.cpus_allowed tracks
1946 * cpu_online_map on each CPU hotplug (cpuhp) event.
1949 static int cpuset_handle_cpuhp(struct notifier_block *unused_nb,
1950 unsigned long phase, void *unused_cpu)
1953 case CPU_UP_CANCELED:
1954 case CPU_UP_CANCELED_FROZEN:
1955 case CPU_DOWN_FAILED:
1956 case CPU_DOWN_FAILED_FROZEN:
1958 case CPU_ONLINE_FROZEN:
1960 case CPU_DEAD_FROZEN:
1961 common_cpu_mem_hotplug_unplug(1);
1970 #ifdef CONFIG_MEMORY_HOTPLUG
1972 * Keep top_cpuset.mems_allowed tracking node_states[N_HIGH_MEMORY].
1973 * Call this routine anytime after you change
1974 * node_states[N_HIGH_MEMORY].
1975 * See also the previous routine cpuset_handle_cpuhp().
1978 void cpuset_track_online_nodes(void)
1980 common_cpu_mem_hotplug_unplug(0);
1985 * cpuset_init_smp - initialize cpus_allowed
1987 * Description: Finish top cpuset after cpu, node maps are initialized
1990 void __init cpuset_init_smp(void)
1992 top_cpuset.cpus_allowed = cpu_online_map;
1993 top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY];
1995 hotcpu_notifier(cpuset_handle_cpuhp, 0);
1999 * cpuset_cpus_allowed - return cpus_allowed mask from a tasks cpuset.
2000 * @tsk: pointer to task_struct from which to obtain cpuset->cpus_allowed.
2001 * @pmask: pointer to cpumask_t variable to receive cpus_allowed set.
2003 * Description: Returns the cpumask_t cpus_allowed of the cpuset
2004 * attached to the specified @tsk. Guaranteed to return some non-empty
2005 * subset of cpu_online_map, even if this means going outside the
2009 void cpuset_cpus_allowed(struct task_struct *tsk, cpumask_t *pmask)
2011 mutex_lock(&callback_mutex);
2012 cpuset_cpus_allowed_locked(tsk, pmask);
2013 mutex_unlock(&callback_mutex);
2017 * cpuset_cpus_allowed_locked - return cpus_allowed mask from a tasks cpuset.
2018 * Must be called with callback_mutex held.
2020 void cpuset_cpus_allowed_locked(struct task_struct *tsk, cpumask_t *pmask)
2023 guarantee_online_cpus(task_cs(tsk), pmask);
2027 void cpuset_init_current_mems_allowed(void)
2029 nodes_setall(current->mems_allowed);
2033 * cpuset_mems_allowed - return mems_allowed mask from a tasks cpuset.
2034 * @tsk: pointer to task_struct from which to obtain cpuset->mems_allowed.
2036 * Description: Returns the nodemask_t mems_allowed of the cpuset
2037 * attached to the specified @tsk. Guaranteed to return some non-empty
2038 * subset of node_states[N_HIGH_MEMORY], even if this means going outside the
2042 nodemask_t cpuset_mems_allowed(struct task_struct *tsk)
2046 mutex_lock(&callback_mutex);
2048 guarantee_online_mems(task_cs(tsk), &mask);
2050 mutex_unlock(&callback_mutex);
2056 * cpuset_nodemask_valid_mems_allowed - check nodemask vs. curremt mems_allowed
2057 * @nodemask: the nodemask to be checked
2059 * Are any of the nodes in the nodemask allowed in current->mems_allowed?
2061 int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask)
2063 return nodes_intersects(*nodemask, current->mems_allowed);
2067 * nearest_hardwall_ancestor() - Returns the nearest mem_exclusive or
2068 * mem_hardwall ancestor to the specified cpuset. Call holding
2069 * callback_mutex. If no ancestor is mem_exclusive or mem_hardwall
2070 * (an unusual configuration), then returns the root cpuset.
2072 static const struct cpuset *nearest_hardwall_ancestor(const struct cpuset *cs)
2074 while (!(is_mem_exclusive(cs) || is_mem_hardwall(cs)) && cs->parent)
2080 * cpuset_zone_allowed_softwall - Can we allocate on zone z's memory node?
2081 * @z: is this zone on an allowed node?
2082 * @gfp_mask: memory allocation flags
2084 * If we're in interrupt, yes, we can always allocate. If
2085 * __GFP_THISNODE is set, yes, we can always allocate. If zone
2086 * z's node is in our tasks mems_allowed, yes. If it's not a
2087 * __GFP_HARDWALL request and this zone's nodes is in the nearest
2088 * hardwalled cpuset ancestor to this tasks cpuset, yes.
2089 * If the task has been OOM killed and has access to memory reserves
2090 * as specified by the TIF_MEMDIE flag, yes.
2093 * If __GFP_HARDWALL is set, cpuset_zone_allowed_softwall()
2094 * reduces to cpuset_zone_allowed_hardwall(). Otherwise,
2095 * cpuset_zone_allowed_softwall() might sleep, and might allow a zone
2096 * from an enclosing cpuset.
2098 * cpuset_zone_allowed_hardwall() only handles the simpler case of
2099 * hardwall cpusets, and never sleeps.
2101 * The __GFP_THISNODE placement logic is really handled elsewhere,
2102 * by forcibly using a zonelist starting at a specified node, and by
2103 * (in get_page_from_freelist()) refusing to consider the zones for
2104 * any node on the zonelist except the first. By the time any such
2105 * calls get to this routine, we should just shut up and say 'yes'.
2107 * GFP_USER allocations are marked with the __GFP_HARDWALL bit,
2108 * and do not allow allocations outside the current tasks cpuset
2109 * unless the task has been OOM killed as is marked TIF_MEMDIE.
2110 * GFP_KERNEL allocations are not so marked, so can escape to the
2111 * nearest enclosing hardwalled ancestor cpuset.
2113 * Scanning up parent cpusets requires callback_mutex. The
2114 * __alloc_pages() routine only calls here with __GFP_HARDWALL bit
2115 * _not_ set if it's a GFP_KERNEL allocation, and all nodes in the
2116 * current tasks mems_allowed came up empty on the first pass over
2117 * the zonelist. So only GFP_KERNEL allocations, if all nodes in the
2118 * cpuset are short of memory, might require taking the callback_mutex
2121 * The first call here from mm/page_alloc:get_page_from_freelist()
2122 * has __GFP_HARDWALL set in gfp_mask, enforcing hardwall cpusets,
2123 * so no allocation on a node outside the cpuset is allowed (unless
2124 * in interrupt, of course).
2126 * The second pass through get_page_from_freelist() doesn't even call
2127 * here for GFP_ATOMIC calls. For those calls, the __alloc_pages()
2128 * variable 'wait' is not set, and the bit ALLOC_CPUSET is not set
2129 * in alloc_flags. That logic and the checks below have the combined
2131 * in_interrupt - any node ok (current task context irrelevant)
2132 * GFP_ATOMIC - any node ok
2133 * TIF_MEMDIE - any node ok
2134 * GFP_KERNEL - any node in enclosing hardwalled cpuset ok
2135 * GFP_USER - only nodes in current tasks mems allowed ok.
2138 * Don't call cpuset_zone_allowed_softwall if you can't sleep, unless you
2139 * pass in the __GFP_HARDWALL flag set in gfp_flag, which disables
2140 * the code that might scan up ancestor cpusets and sleep.
2143 int __cpuset_zone_allowed_softwall(struct zone *z, gfp_t gfp_mask)
2145 int node; /* node that zone z is on */
2146 const struct cpuset *cs; /* current cpuset ancestors */
2147 int allowed; /* is allocation in zone z allowed? */
2149 if (in_interrupt() || (gfp_mask & __GFP_THISNODE))
2151 node = zone_to_nid(z);
2152 might_sleep_if(!(gfp_mask & __GFP_HARDWALL));
2153 if (node_isset(node, current->mems_allowed))
2156 * Allow tasks that have access to memory reserves because they have
2157 * been OOM killed to get memory anywhere.
2159 if (unlikely(test_thread_flag(TIF_MEMDIE)))
2161 if (gfp_mask & __GFP_HARDWALL) /* If hardwall request, stop here */
2164 if (current->flags & PF_EXITING) /* Let dying task have memory */
2167 /* Not hardwall and node outside mems_allowed: scan up cpusets */
2168 mutex_lock(&callback_mutex);
2171 cs = nearest_hardwall_ancestor(task_cs(current));
2172 task_unlock(current);
2174 allowed = node_isset(node, cs->mems_allowed);
2175 mutex_unlock(&callback_mutex);
2180 * cpuset_zone_allowed_hardwall - Can we allocate on zone z's memory node?
2181 * @z: is this zone on an allowed node?
2182 * @gfp_mask: memory allocation flags
2184 * If we're in interrupt, yes, we can always allocate.
2185 * If __GFP_THISNODE is set, yes, we can always allocate. If zone
2186 * z's node is in our tasks mems_allowed, yes. If the task has been
2187 * OOM killed and has access to memory reserves as specified by the
2188 * TIF_MEMDIE flag, yes. Otherwise, no.
2190 * The __GFP_THISNODE placement logic is really handled elsewhere,
2191 * by forcibly using a zonelist starting at a specified node, and by
2192 * (in get_page_from_freelist()) refusing to consider the zones for
2193 * any node on the zonelist except the first. By the time any such
2194 * calls get to this routine, we should just shut up and say 'yes'.
2196 * Unlike the cpuset_zone_allowed_softwall() variant, above,
2197 * this variant requires that the zone be in the current tasks
2198 * mems_allowed or that we're in interrupt. It does not scan up the
2199 * cpuset hierarchy for the nearest enclosing mem_exclusive cpuset.
2203 int __cpuset_zone_allowed_hardwall(struct zone *z, gfp_t gfp_mask)
2205 int node; /* node that zone z is on */
2207 if (in_interrupt() || (gfp_mask & __GFP_THISNODE))
2209 node = zone_to_nid(z);
2210 if (node_isset(node, current->mems_allowed))
2213 * Allow tasks that have access to memory reserves because they have
2214 * been OOM killed to get memory anywhere.
2216 if (unlikely(test_thread_flag(TIF_MEMDIE)))
2222 * cpuset_lock - lock out any changes to cpuset structures
2224 * The out of memory (oom) code needs to mutex_lock cpusets
2225 * from being changed while it scans the tasklist looking for a
2226 * task in an overlapping cpuset. Expose callback_mutex via this
2227 * cpuset_lock() routine, so the oom code can lock it, before
2228 * locking the task list. The tasklist_lock is a spinlock, so
2229 * must be taken inside callback_mutex.
2232 void cpuset_lock(void)
2234 mutex_lock(&callback_mutex);
2238 * cpuset_unlock - release lock on cpuset changes
2240 * Undo the lock taken in a previous cpuset_lock() call.
2243 void cpuset_unlock(void)
2245 mutex_unlock(&callback_mutex);
2249 * cpuset_mem_spread_node() - On which node to begin search for a page
2251 * If a task is marked PF_SPREAD_PAGE or PF_SPREAD_SLAB (as for
2252 * tasks in a cpuset with is_spread_page or is_spread_slab set),
2253 * and if the memory allocation used cpuset_mem_spread_node()
2254 * to determine on which node to start looking, as it will for
2255 * certain page cache or slab cache pages such as used for file
2256 * system buffers and inode caches, then instead of starting on the
2257 * local node to look for a free page, rather spread the starting
2258 * node around the tasks mems_allowed nodes.
2260 * We don't have to worry about the returned node being offline
2261 * because "it can't happen", and even if it did, it would be ok.
2263 * The routines calling guarantee_online_mems() are careful to
2264 * only set nodes in task->mems_allowed that are online. So it
2265 * should not be possible for the following code to return an
2266 * offline node. But if it did, that would be ok, as this routine
2267 * is not returning the node where the allocation must be, only
2268 * the node where the search should start. The zonelist passed to
2269 * __alloc_pages() will include all nodes. If the slab allocator
2270 * is passed an offline node, it will fall back to the local node.
2271 * See kmem_cache_alloc_node().
2274 int cpuset_mem_spread_node(void)
2278 node = next_node(current->cpuset_mem_spread_rotor, current->mems_allowed);
2279 if (node == MAX_NUMNODES)
2280 node = first_node(current->mems_allowed);
2281 current->cpuset_mem_spread_rotor = node;
2284 EXPORT_SYMBOL_GPL(cpuset_mem_spread_node);
2287 * cpuset_mems_allowed_intersects - Does @tsk1's mems_allowed intersect @tsk2's?
2288 * @tsk1: pointer to task_struct of some task.
2289 * @tsk2: pointer to task_struct of some other task.
2291 * Description: Return true if @tsk1's mems_allowed intersects the
2292 * mems_allowed of @tsk2. Used by the OOM killer to determine if
2293 * one of the task's memory usage might impact the memory available
2297 int cpuset_mems_allowed_intersects(const struct task_struct *tsk1,
2298 const struct task_struct *tsk2)
2300 return nodes_intersects(tsk1->mems_allowed, tsk2->mems_allowed);
2304 * Collection of memory_pressure is suppressed unless
2305 * this flag is enabled by writing "1" to the special
2306 * cpuset file 'memory_pressure_enabled' in the root cpuset.
2309 int cpuset_memory_pressure_enabled __read_mostly;
2312 * cpuset_memory_pressure_bump - keep stats of per-cpuset reclaims.
2314 * Keep a running average of the rate of synchronous (direct)
2315 * page reclaim efforts initiated by tasks in each cpuset.
2317 * This represents the rate at which some task in the cpuset
2318 * ran low on memory on all nodes it was allowed to use, and
2319 * had to enter the kernels page reclaim code in an effort to
2320 * create more free memory by tossing clean pages or swapping
2321 * or writing dirty pages.
2323 * Display to user space in the per-cpuset read-only file
2324 * "memory_pressure". Value displayed is an integer
2325 * representing the recent rate of entry into the synchronous
2326 * (direct) page reclaim by any task attached to the cpuset.
2329 void __cpuset_memory_pressure_bump(void)
2332 fmeter_markevent(&task_cs(current)->fmeter);
2333 task_unlock(current);
2336 #ifdef CONFIG_PROC_PID_CPUSET
2338 * proc_cpuset_show()
2339 * - Print tasks cpuset path into seq_file.
2340 * - Used for /proc/<pid>/cpuset.
2341 * - No need to task_lock(tsk) on this tsk->cpuset reference, as it
2342 * doesn't really matter if tsk->cpuset changes after we read it,
2343 * and we take cgroup_mutex, keeping cpuset_attach() from changing it
2346 static int proc_cpuset_show(struct seq_file *m, void *unused_v)
2349 struct task_struct *tsk;
2351 struct cgroup_subsys_state *css;
2355 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
2361 tsk = get_pid_task(pid, PIDTYPE_PID);
2367 css = task_subsys_state(tsk, cpuset_subsys_id);
2368 retval = cgroup_path(css->cgroup, buf, PAGE_SIZE);
2375 put_task_struct(tsk);
2382 static int cpuset_open(struct inode *inode, struct file *file)
2384 struct pid *pid = PROC_I(inode)->pid;
2385 return single_open(file, proc_cpuset_show, pid);
2388 const struct file_operations proc_cpuset_operations = {
2389 .open = cpuset_open,
2391 .llseek = seq_lseek,
2392 .release = single_release,
2394 #endif /* CONFIG_PROC_PID_CPUSET */
2396 /* Display task cpus_allowed, mems_allowed in /proc/<pid>/status file. */
2397 void cpuset_task_status_allowed(struct seq_file *m, struct task_struct *task)
2399 seq_printf(m, "Cpus_allowed:\t");
2400 m->count += cpumask_scnprintf(m->buf + m->count, m->size - m->count,
2401 task->cpus_allowed);
2402 seq_printf(m, "\n");
2403 seq_printf(m, "Cpus_allowed_list:\t");
2404 m->count += cpulist_scnprintf(m->buf + m->count, m->size - m->count,
2405 task->cpus_allowed);
2406 seq_printf(m, "\n");
2407 seq_printf(m, "Mems_allowed:\t");
2408 m->count += nodemask_scnprintf(m->buf + m->count, m->size - m->count,
2409 task->mems_allowed);
2410 seq_printf(m, "\n");
2411 seq_printf(m, "Mems_allowed_list:\t");
2412 m->count += nodelist_scnprintf(m->buf + m->count, m->size - m->count,
2413 task->mems_allowed);
2414 seq_printf(m, "\n");