~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/kernel/cpuset.c

Version: ~ [ linux-5.3-rc5 ] ~ [ linux-5.2.9 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.67 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.139 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.189 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.189 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.72 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  *  kernel/cpuset.c
  3  *
  4  *  Processor and Memory placement constraints for sets of tasks.
  5  *
  6  *  Copyright (C) 2003 BULL SA.
  7  *  Copyright (C) 2004-2007 Silicon Graphics, Inc.
  8  *  Copyright (C) 2006 Google, Inc
  9  *
 10  *  Portions derived from Patrick Mochel's sysfs code.
 11  *  sysfs is Copyright (c) 2001-3 Patrick Mochel
 12  *
 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
 17  *  2008 Rework of the scheduler domains and CPU hotplug handling
 18  *       by Max Krasnyansky
 19  *
 20  *  This file is subject to the terms and conditions of the GNU General Public
 21  *  License.  See the file COPYING in the main directory of the Linux
 22  *  distribution for more details.
 23  */
 24 
 25 #include <linux/cpu.h>
 26 #include <linux/cpumask.h>
 27 #include <linux/cpuset.h>
 28 #include <linux/err.h>
 29 #include <linux/errno.h>
 30 #include <linux/file.h>
 31 #include <linux/fs.h>
 32 #include <linux/init.h>
 33 #include <linux/interrupt.h>
 34 #include <linux/kernel.h>
 35 #include <linux/kmod.h>
 36 #include <linux/list.h>
 37 #include <linux/mempolicy.h>
 38 #include <linux/mm.h>
 39 #include <linux/memory.h>
 40 #include <linux/export.h>
 41 #include <linux/mount.h>
 42 #include <linux/namei.h>
 43 #include <linux/pagemap.h>
 44 #include <linux/proc_fs.h>
 45 #include <linux/rcupdate.h>
 46 #include <linux/sched.h>
 47 #include <linux/seq_file.h>
 48 #include <linux/security.h>
 49 #include <linux/slab.h>
 50 #include <linux/spinlock.h>
 51 #include <linux/stat.h>
 52 #include <linux/string.h>
 53 #include <linux/time.h>
 54 #include <linux/backing-dev.h>
 55 #include <linux/sort.h>
 56 
 57 #include <asm/uaccess.h>
 58 #include <linux/atomic.h>
 59 #include <linux/mutex.h>
 60 #include <linux/workqueue.h>
 61 #include <linux/cgroup.h>
 62 
 63 /*
 64  * Workqueue for cpuset related tasks.
 65  *
 66  * Using kevent workqueue may cause deadlock when memory_migrate
 67  * is set. So we create a separate workqueue thread for cpuset.
 68  */
 69 static struct workqueue_struct *cpuset_wq;
 70 
 71 /*
 72  * Tracks how many cpusets are currently defined in system.
 73  * When there is only one cpuset (the root cpuset) we can
 74  * short circuit some hooks.
 75  */
 76 int number_of_cpusets __read_mostly;
 77 
 78 /* Forward declare cgroup structures */
 79 struct cgroup_subsys cpuset_subsys;
 80 struct cpuset;
 81 
 82 /* See "Frequency meter" comments, below. */
 83 
 84 struct fmeter {
 85         int cnt;                /* unprocessed events count */
 86         int val;                /* most recent output value */
 87         time_t time;            /* clock (secs) when val computed */
 88         spinlock_t lock;        /* guards read or write of above */
 89 };
 90 
 91 struct cpuset {
 92         struct cgroup_subsys_state css;
 93 
 94         unsigned long flags;            /* "unsigned long" so bitops work */
 95         cpumask_var_t cpus_allowed;     /* CPUs allowed to tasks in cpuset */
 96         nodemask_t mems_allowed;        /* Memory Nodes allowed to tasks */
 97 
 98         struct cpuset *parent;          /* my parent */
 99 
100         struct fmeter fmeter;           /* memory_pressure filter */
101 
102         /* partition number for rebuild_sched_domains() */
103         int pn;
104 
105         /* for custom sched domain */
106         int relax_domain_level;
107 
108         /* used for walking a cpuset hierarchy */
109         struct list_head stack_list;
110 };
111 
112 /* Retrieve the cpuset for a cgroup */
113 static inline struct cpuset *cgroup_cs(struct cgroup *cont)
114 {
115         return container_of(cgroup_subsys_state(cont, cpuset_subsys_id),
116                             struct cpuset, css);
117 }
118 
119 /* Retrieve the cpuset for a task */
120 static inline struct cpuset *task_cs(struct task_struct *task)
121 {
122         return container_of(task_subsys_state(task, cpuset_subsys_id),
123                             struct cpuset, css);
124 }
125 
126 #ifdef CONFIG_NUMA
127 static inline bool task_has_mempolicy(struct task_struct *task)
128 {
129         return task->mempolicy;
130 }
131 #else
132 static inline bool task_has_mempolicy(struct task_struct *task)
133 {
134         return false;
135 }
136 #endif
137 
138 
139 /* bits in struct cpuset flags field */
140 typedef enum {
141         CS_CPU_EXCLUSIVE,
142         CS_MEM_EXCLUSIVE,
143         CS_MEM_HARDWALL,
144         CS_MEMORY_MIGRATE,
145         CS_SCHED_LOAD_BALANCE,
146         CS_SPREAD_PAGE,
147         CS_SPREAD_SLAB,
148 } cpuset_flagbits_t;
149 
150 /* convenient tests for these bits */
151 static inline int is_cpu_exclusive(const struct cpuset *cs)
152 {
153         return test_bit(CS_CPU_EXCLUSIVE, &cs->flags);
154 }
155 
156 static inline int is_mem_exclusive(const struct cpuset *cs)
157 {
158         return test_bit(CS_MEM_EXCLUSIVE, &cs->flags);
159 }
160 
161 static inline int is_mem_hardwall(const struct cpuset *cs)
162 {
163         return test_bit(CS_MEM_HARDWALL, &cs->flags);
164 }
165 
166 static inline int is_sched_load_balance(const struct cpuset *cs)
167 {
168         return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
169 }
170 
171 static inline int is_memory_migrate(const struct cpuset *cs)
172 {
173         return test_bit(CS_MEMORY_MIGRATE, &cs->flags);
174 }
175 
176 static inline int is_spread_page(const struct cpuset *cs)
177 {
178         return test_bit(CS_SPREAD_PAGE, &cs->flags);
179 }
180 
181 static inline int is_spread_slab(const struct cpuset *cs)
182 {
183         return test_bit(CS_SPREAD_SLAB, &cs->flags);
184 }
185 
186 static struct cpuset top_cpuset = {
187         .flags = ((1 << CS_CPU_EXCLUSIVE) | (1 << CS_MEM_EXCLUSIVE)),
188 };
189 
190 /*
191  * There are two global mutexes guarding cpuset structures.  The first
192  * is the main control groups cgroup_mutex, accessed via
193  * cgroup_lock()/cgroup_unlock().  The second is the cpuset-specific
194  * callback_mutex, below. They can nest.  It is ok to first take
195  * cgroup_mutex, then nest callback_mutex.  We also require taking
196  * task_lock() when dereferencing a task's cpuset pointer.  See "The
197  * task_lock() exception", at the end of this comment.
198  *
199  * A task must hold both mutexes to modify cpusets.  If a task
200  * holds cgroup_mutex, then it blocks others wanting that mutex,
201  * ensuring that it is the only task able to also acquire callback_mutex
202  * and be able to modify cpusets.  It can perform various checks on
203  * the cpuset structure first, knowing nothing will change.  It can
204  * also allocate memory while just holding cgroup_mutex.  While it is
205  * performing these checks, various callback routines can briefly
206  * acquire callback_mutex to query cpusets.  Once it is ready to make
207  * the changes, it takes callback_mutex, blocking everyone else.
208  *
209  * Calls to the kernel memory allocator can not be made while holding
210  * callback_mutex, as that would risk double tripping on callback_mutex
211  * from one of the callbacks into the cpuset code from within
212  * __alloc_pages().
213  *
214  * If a task is only holding callback_mutex, then it has read-only
215  * access to cpusets.
216  *
217  * Now, the task_struct fields mems_allowed and mempolicy may be changed
218  * by other task, we use alloc_lock in the task_struct fields to protect
219  * them.
220  *
221  * The cpuset_common_file_read() handlers only hold callback_mutex across
222  * small pieces of code, such as when reading out possibly multi-word
223  * cpumasks and nodemasks.
224  *
225  * Accessing a task's cpuset should be done in accordance with the
226  * guidelines for accessing subsystem state in kernel/cgroup.c
227  */
228 
229 static DEFINE_MUTEX(callback_mutex);
230 
231 /*
232  * cpuset_buffer_lock protects both the cpuset_name and cpuset_nodelist
233  * buffers.  They are statically allocated to prevent using excess stack
234  * when calling cpuset_print_task_mems_allowed().
235  */
236 #define CPUSET_NAME_LEN         (128)
237 #define CPUSET_NODELIST_LEN     (256)
238 static char cpuset_name[CPUSET_NAME_LEN];
239 static char cpuset_nodelist[CPUSET_NODELIST_LEN];
240 static DEFINE_SPINLOCK(cpuset_buffer_lock);
241 
242 /*
243  * This is ugly, but preserves the userspace API for existing cpuset
244  * users. If someone tries to mount the "cpuset" filesystem, we
245  * silently switch it to mount "cgroup" instead
246  */
247 static struct dentry *cpuset_mount(struct file_system_type *fs_type,
248                          int flags, const char *unused_dev_name, void *data)
249 {
250         struct file_system_type *cgroup_fs = get_fs_type("cgroup");
251         struct dentry *ret = ERR_PTR(-ENODEV);
252         if (cgroup_fs) {
253                 char mountopts[] =
254                         "cpuset,noprefix,"
255                         "release_agent=/sbin/cpuset_release_agent";
256                 ret = cgroup_fs->mount(cgroup_fs, flags,
257                                            unused_dev_name, mountopts);
258                 put_filesystem(cgroup_fs);
259         }
260         return ret;
261 }
262 
263 static struct file_system_type cpuset_fs_type = {
264         .name = "cpuset",
265         .mount = cpuset_mount,
266 };
267 
268 /*
269  * Return in pmask the portion of a cpusets's cpus_allowed that
270  * are online.  If none are online, walk up the cpuset hierarchy
271  * until we find one that does have some online cpus.  If we get
272  * all the way to the top and still haven't found any online cpus,
273  * return cpu_online_mask.  Or if passed a NULL cs from an exit'ing
274  * task, return cpu_online_mask.
275  *
276  * One way or another, we guarantee to return some non-empty subset
277  * of cpu_online_mask.
278  *
279  * Call with callback_mutex held.
280  */
281 
282 static void guarantee_online_cpus(const struct cpuset *cs,
283                                   struct cpumask *pmask)
284 {
285         while (cs && !cpumask_intersects(cs->cpus_allowed, cpu_online_mask))
286                 cs = cs->parent;
287         if (cs)
288                 cpumask_and(pmask, cs->cpus_allowed, cpu_online_mask);
289         else
290                 cpumask_copy(pmask, cpu_online_mask);
291         BUG_ON(!cpumask_intersects(pmask, cpu_online_mask));
292 }
293 
294 /*
295  * Return in *pmask the portion of a cpusets's mems_allowed that
296  * are online, with memory.  If none are online with memory, walk
297  * up the cpuset hierarchy until we find one that does have some
298  * online mems.  If we get all the way to the top and still haven't
299  * found any online mems, return node_states[N_HIGH_MEMORY].
300  *
301  * One way or another, we guarantee to return some non-empty subset
302  * of node_states[N_HIGH_MEMORY].
303  *
304  * Call with callback_mutex held.
305  */
306 
307 static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask)
308 {
309         while (cs && !nodes_intersects(cs->mems_allowed,
310                                         node_states[N_HIGH_MEMORY]))
311                 cs = cs->parent;
312         if (cs)
313                 nodes_and(*pmask, cs->mems_allowed,
314                                         node_states[N_HIGH_MEMORY]);
315         else
316                 *pmask = node_states[N_HIGH_MEMORY];
317         BUG_ON(!nodes_intersects(*pmask, node_states[N_HIGH_MEMORY]));
318 }
319 
320 /*
321  * update task's spread flag if cpuset's page/slab spread flag is set
322  *
323  * Called with callback_mutex/cgroup_mutex held
324  */
325 static void cpuset_update_task_spread_flag(struct cpuset *cs,
326                                         struct task_struct *tsk)
327 {
328         if (is_spread_page(cs))
329                 task_set_spread_page(tsk);
330         else
331                 task_clear_spread_page(tsk);
332 
333         if (is_spread_slab(cs))
334                 task_set_spread_slab(tsk);
335         else
336                 task_clear_spread_slab(tsk);
337 }
338 
339 /*
340  * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q?
341  *
342  * One cpuset is a subset of another if all its allowed CPUs and
343  * Memory Nodes are a subset of the other, and its exclusive flags
344  * are only set if the other's are set.  Call holding cgroup_mutex.
345  */
346 
347 static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q)
348 {
349         return  cpumask_subset(p->cpus_allowed, q->cpus_allowed) &&
350                 nodes_subset(p->mems_allowed, q->mems_allowed) &&
351                 is_cpu_exclusive(p) <= is_cpu_exclusive(q) &&
352                 is_mem_exclusive(p) <= is_mem_exclusive(q);
353 }
354 
355 /**
356  * alloc_trial_cpuset - allocate a trial cpuset
357  * @cs: the cpuset that the trial cpuset duplicates
358  */
359 static struct cpuset *alloc_trial_cpuset(const struct cpuset *cs)
360 {
361         struct cpuset *trial;
362 
363         trial = kmemdup(cs, sizeof(*cs), GFP_KERNEL);
364         if (!trial)
365                 return NULL;
366 
367         if (!alloc_cpumask_var(&trial->cpus_allowed, GFP_KERNEL)) {
368                 kfree(trial);
369                 return NULL;
370         }
371         cpumask_copy(trial->cpus_allowed, cs->cpus_allowed);
372 
373         return trial;
374 }
375 
376 /**
377  * free_trial_cpuset - free the trial cpuset
378  * @trial: the trial cpuset to be freed
379  */
380 static void free_trial_cpuset(struct cpuset *trial)
381 {
382         free_cpumask_var(trial->cpus_allowed);
383         kfree(trial);
384 }
385 
386 /*
387  * validate_change() - Used to validate that any proposed cpuset change
388  *                     follows the structural rules for cpusets.
389  *
390  * If we replaced the flag and mask values of the current cpuset
391  * (cur) with those values in the trial cpuset (trial), would
392  * our various subset and exclusive rules still be valid?  Presumes
393  * cgroup_mutex held.
394  *
395  * 'cur' is the address of an actual, in-use cpuset.  Operations
396  * such as list traversal that depend on the actual address of the
397  * cpuset in the list must use cur below, not trial.
398  *
399  * 'trial' is the address of bulk structure copy of cur, with
400  * perhaps one or more of the fields cpus_allowed, mems_allowed,
401  * or flags changed to new, trial values.
402  *
403  * Return 0 if valid, -errno if not.
404  */
405 
406 static int validate_change(const struct cpuset *cur, const struct cpuset *trial)
407 {
408         struct cgroup *cont;
409         struct cpuset *c, *par;
410 
411         /* Each of our child cpusets must be a subset of us */
412         list_for_each_entry(cont, &cur->css.cgroup->children, sibling) {
413                 if (!is_cpuset_subset(cgroup_cs(cont), trial))
414                         return -EBUSY;
415         }
416 
417         /* Remaining checks don't apply to root cpuset */
418         if (cur == &top_cpuset)
419                 return 0;
420 
421         par = cur->parent;
422 
423         /* We must be a subset of our parent cpuset */
424         if (!is_cpuset_subset(trial, par))
425                 return -EACCES;
426 
427         /*
428          * If either I or some sibling (!= me) is exclusive, we can't
429          * overlap
430          */
431         list_for_each_entry(cont, &par->css.cgroup->children, sibling) {
432                 c = cgroup_cs(cont);
433                 if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) &&
434                     c != cur &&
435                     cpumask_intersects(trial->cpus_allowed, c->cpus_allowed))
436                         return -EINVAL;
437                 if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) &&
438                     c != cur &&
439                     nodes_intersects(trial->mems_allowed, c->mems_allowed))
440                         return -EINVAL;
441         }
442 
443         /* Cpusets with tasks can't have empty cpus_allowed or mems_allowed */
444         if (cgroup_task_count(cur->css.cgroup)) {
445                 if (cpumask_empty(trial->cpus_allowed) ||
446                     nodes_empty(trial->mems_allowed)) {
447                         return -ENOSPC;
448                 }
449         }
450 
451         return 0;
452 }
453 
454 #ifdef CONFIG_SMP
455 /*
456  * Helper routine for generate_sched_domains().
457  * Do cpusets a, b have overlapping cpus_allowed masks?
458  */
459 static int cpusets_overlap(struct cpuset *a, struct cpuset *b)
460 {
461         return cpumask_intersects(a->cpus_allowed, b->cpus_allowed);
462 }
463 
464 static void
465 update_domain_attr(struct sched_domain_attr *dattr, struct cpuset *c)
466 {
467         if (dattr->relax_domain_level < c->relax_domain_level)
468                 dattr->relax_domain_level = c->relax_domain_level;
469         return;
470 }
471 
472 static void
473 update_domain_attr_tree(struct sched_domain_attr *dattr, struct cpuset *c)
474 {
475         LIST_HEAD(q);
476 
477         list_add(&c->stack_list, &q);
478         while (!list_empty(&q)) {
479                 struct cpuset *cp;
480                 struct cgroup *cont;
481                 struct cpuset *child;
482 
483                 cp = list_first_entry(&q, struct cpuset, stack_list);
484                 list_del(q.next);
485 
486                 if (cpumask_empty(cp->cpus_allowed))
487                         continue;
488 
489                 if (is_sched_load_balance(cp))
490                         update_domain_attr(dattr, cp);
491 
492                 list_for_each_entry(cont, &cp->css.cgroup->children, sibling) {
493                         child = cgroup_cs(cont);
494                         list_add_tail(&child->stack_list, &q);
495                 }
496         }
497 }
498 
499 /*
500  * generate_sched_domains()
501  *
502  * This function builds a partial partition of the systems CPUs
503  * A 'partial partition' is a set of non-overlapping subsets whose
504  * union is a subset of that set.
505  * The output of this function needs to be passed to kernel/sched.c
506  * partition_sched_domains() routine, which will rebuild the scheduler's
507  * load balancing domains (sched domains) as specified by that partial
508  * partition.
509  *
510  * See "What is sched_load_balance" in Documentation/cgroups/cpusets.txt
511  * for a background explanation of this.
512  *
513  * Does not return errors, on the theory that the callers of this
514  * routine would rather not worry about failures to rebuild sched
515  * domains when operating in the severe memory shortage situations
516  * that could cause allocation failures below.
517  *
518  * Must be called with cgroup_lock held.
519  *
520  * The three key local variables below are:
521  *    q  - a linked-list queue of cpuset pointers, used to implement a
522  *         top-down scan of all cpusets.  This scan loads a pointer
523  *         to each cpuset marked is_sched_load_balance into the
524  *         array 'csa'.  For our purposes, rebuilding the schedulers
525  *         sched domains, we can ignore !is_sched_load_balance cpusets.
526  *  csa  - (for CpuSet Array) Array of pointers to all the cpusets
527  *         that need to be load balanced, for convenient iterative
528  *         access by the subsequent code that finds the best partition,
529  *         i.e the set of domains (subsets) of CPUs such that the
530  *         cpus_allowed of every cpuset marked is_sched_load_balance
531  *         is a subset of one of these domains, while there are as
532  *         many such domains as possible, each as small as possible.
533  * doms  - Conversion of 'csa' to an array of cpumasks, for passing to
534  *         the kernel/sched.c routine partition_sched_domains() in a
535  *         convenient format, that can be easily compared to the prior
536  *         value to determine what partition elements (sched domains)
537  *         were changed (added or removed.)
538  *
539  * Finding the best partition (set of domains):
540  *      The triple nested loops below over i, j, k scan over the
541  *      load balanced cpusets (using the array of cpuset pointers in
542  *      csa[]) looking for pairs of cpusets that have overlapping
543  *      cpus_allowed, but which don't have the same 'pn' partition
544  *      number and gives them in the same partition number.  It keeps
545  *      looping on the 'restart' label until it can no longer find
546  *      any such pairs.
547  *
548  *      The union of the cpus_allowed masks from the set of
549  *      all cpusets having the same 'pn' value then form the one
550  *      element of the partition (one sched domain) to be passed to
551  *      partition_sched_domains().
552  */
553 static int generate_sched_domains(cpumask_var_t **domains,
554                         struct sched_domain_attr **attributes)
555 {
556         LIST_HEAD(q);           /* queue of cpusets to be scanned */
557         struct cpuset *cp;      /* scans q */
558         struct cpuset **csa;    /* array of all cpuset ptrs */
559         int csn;                /* how many cpuset ptrs in csa so far */
560         int i, j, k;            /* indices for partition finding loops */
561         cpumask_var_t *doms;    /* resulting partition; i.e. sched domains */
562         struct sched_domain_attr *dattr;  /* attributes for custom domains */
563         int ndoms = 0;          /* number of sched domains in result */
564         int nslot;              /* next empty doms[] struct cpumask slot */
565 
566         doms = NULL;
567         dattr = NULL;
568         csa = NULL;
569 
570         /* Special case for the 99% of systems with one, full, sched domain */
571         if (is_sched_load_balance(&top_cpuset)) {
572                 ndoms = 1;
573                 doms = alloc_sched_domains(ndoms);
574                 if (!doms)
575                         goto done;
576 
577                 dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL);
578                 if (dattr) {
579                         *dattr = SD_ATTR_INIT;
580                         update_domain_attr_tree(dattr, &top_cpuset);
581                 }
582                 cpumask_copy(doms[0], top_cpuset.cpus_allowed);
583 
584                 goto done;
585         }
586 
587         csa = kmalloc(number_of_cpusets * sizeof(cp), GFP_KERNEL);
588         if (!csa)
589                 goto done;
590         csn = 0;
591 
592         list_add(&top_cpuset.stack_list, &q);
593         while (!list_empty(&q)) {
594                 struct cgroup *cont;
595                 struct cpuset *child;   /* scans child cpusets of cp */
596 
597                 cp = list_first_entry(&q, struct cpuset, stack_list);
598                 list_del(q.next);
599 
600                 if (cpumask_empty(cp->cpus_allowed))
601                         continue;
602 
603                 /*
604                  * All child cpusets contain a subset of the parent's cpus, so
605                  * just skip them, and then we call update_domain_attr_tree()
606                  * to calc relax_domain_level of the corresponding sched
607                  * domain.
608                  */
609                 if (is_sched_load_balance(cp)) {
610                         csa[csn++] = cp;
611                         continue;
612                 }
613 
614                 list_for_each_entry(cont, &cp->css.cgroup->children, sibling) {
615                         child = cgroup_cs(cont);
616                         list_add_tail(&child->stack_list, &q);
617                 }
618         }
619 
620         for (i = 0; i < csn; i++)
621                 csa[i]->pn = i;
622         ndoms = csn;
623 
624 restart:
625         /* Find the best partition (set of sched domains) */
626         for (i = 0; i < csn; i++) {
627                 struct cpuset *a = csa[i];
628                 int apn = a->pn;
629 
630                 for (j = 0; j < csn; j++) {
631                         struct cpuset *b = csa[j];
632                         int bpn = b->pn;
633 
634                         if (apn != bpn && cpusets_overlap(a, b)) {
635                                 for (k = 0; k < csn; k++) {
636                                         struct cpuset *c = csa[k];
637 
638                                         if (c->pn == bpn)
639                                                 c->pn = apn;
640                                 }
641                                 ndoms--;        /* one less element */
642                                 goto restart;
643                         }
644                 }
645         }
646 
647         /*
648          * Now we know how many domains to create.
649          * Convert <csn, csa> to <ndoms, doms> and populate cpu masks.
650          */
651         doms = alloc_sched_domains(ndoms);
652         if (!doms)
653                 goto done;
654 
655         /*
656          * The rest of the code, including the scheduler, can deal with
657          * dattr==NULL case. No need to abort if alloc fails.
658          */
659         dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL);
660 
661         for (nslot = 0, i = 0; i < csn; i++) {
662                 struct cpuset *a = csa[i];
663                 struct cpumask *dp;
664                 int apn = a->pn;
665 
666                 if (apn < 0) {
667                         /* Skip completed partitions */
668                         continue;
669                 }
670 
671                 dp = doms[nslot];
672 
673                 if (nslot == ndoms) {
674                         static int warnings = 10;
675                         if (warnings) {
676                                 printk(KERN_WARNING
677                                  "rebuild_sched_domains confused:"
678                                   " nslot %d, ndoms %d, csn %d, i %d,"
679                                   " apn %d\n",
680                                   nslot, ndoms, csn, i, apn);
681                                 warnings--;
682                         }
683                         continue;
684                 }
685 
686                 cpumask_clear(dp);
687                 if (dattr)
688                         *(dattr + nslot) = SD_ATTR_INIT;
689                 for (j = i; j < csn; j++) {
690                         struct cpuset *b = csa[j];
691 
692                         if (apn == b->pn) {
693                                 cpumask_or(dp, dp, b->cpus_allowed);
694                                 if (dattr)
695                                         update_domain_attr_tree(dattr + nslot, b);
696 
697                                 /* Done with this partition */
698                                 b->pn = -1;
699                         }
700                 }
701                 nslot++;
702         }
703         BUG_ON(nslot != ndoms);
704 
705 done:
706         kfree(csa);
707 
708         /*
709          * Fallback to the default domain if kmalloc() failed.
710          * See comments in partition_sched_domains().
711          */
712         if (doms == NULL)
713                 ndoms = 1;
714 
715         *domains    = doms;
716         *attributes = dattr;
717         return ndoms;
718 }
719 
720 /*
721  * Rebuild scheduler domains.
722  *
723  * Call with neither cgroup_mutex held nor within get_online_cpus().
724  * Takes both cgroup_mutex and get_online_cpus().
725  *
726  * Cannot be directly called from cpuset code handling changes
727  * to the cpuset pseudo-filesystem, because it cannot be called
728  * from code that already holds cgroup_mutex.
729  */
730 static void do_rebuild_sched_domains(struct work_struct *unused)
731 {
732         struct sched_domain_attr *attr;
733         cpumask_var_t *doms;
734         int ndoms;
735 
736         get_online_cpus();
737 
738         /* Generate domain masks and attrs */
739         cgroup_lock();
740         ndoms = generate_sched_domains(&doms, &attr);
741         cgroup_unlock();
742 
743         /* Have scheduler rebuild the domains */
744         partition_sched_domains(ndoms, doms, attr);
745 
746         put_online_cpus();
747 }
748 #else /* !CONFIG_SMP */
749 static void do_rebuild_sched_domains(struct work_struct *unused)
750 {
751 }
752 
753 static int generate_sched_domains(cpumask_var_t **domains,
754                         struct sched_domain_attr **attributes)
755 {
756         *domains = NULL;
757         return 1;
758 }
759 #endif /* CONFIG_SMP */
760 
761 static DECLARE_WORK(rebuild_sched_domains_work, do_rebuild_sched_domains);
762 
763 /*
764  * Rebuild scheduler domains, asynchronously via workqueue.
765  *
766  * If the flag 'sched_load_balance' of any cpuset with non-empty
767  * 'cpus' changes, or if the 'cpus' allowed changes in any cpuset
768  * which has that flag enabled, or if any cpuset with a non-empty
769  * 'cpus' is removed, then call this routine to rebuild the
770  * scheduler's dynamic sched domains.
771  *
772  * The rebuild_sched_domains() and partition_sched_domains()
773  * routines must nest cgroup_lock() inside get_online_cpus(),
774  * but such cpuset changes as these must nest that locking the
775  * other way, holding cgroup_lock() for much of the code.
776  *
777  * So in order to avoid an ABBA deadlock, the cpuset code handling
778  * these user changes delegates the actual sched domain rebuilding
779  * to a separate workqueue thread, which ends up processing the
780  * above do_rebuild_sched_domains() function.
781  */
782 static void async_rebuild_sched_domains(void)
783 {
784         queue_work(cpuset_wq, &rebuild_sched_domains_work);
785 }
786 
787 /*
788  * Accomplishes the same scheduler domain rebuild as the above
789  * async_rebuild_sched_domains(), however it directly calls the
790  * rebuild routine synchronously rather than calling it via an
791  * asynchronous work thread.
792  *
793  * This can only be called from code that is not holding
794  * cgroup_mutex (not nested in a cgroup_lock() call.)
795  */
796 void rebuild_sched_domains(void)
797 {
798         do_rebuild_sched_domains(NULL);
799 }
800 
801 /**
802  * cpuset_test_cpumask - test a task's cpus_allowed versus its cpuset's
803  * @tsk: task to test
804  * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner
805  *
806  * Call with cgroup_mutex held.  May take callback_mutex during call.
807  * Called for each task in a cgroup by cgroup_scan_tasks().
808  * Return nonzero if this tasks's cpus_allowed mask should be changed (in other
809  * words, if its mask is not equal to its cpuset's mask).
810  */
811 static int cpuset_test_cpumask(struct task_struct *tsk,
812                                struct cgroup_scanner *scan)
813 {
814         return !cpumask_equal(&tsk->cpus_allowed,
815                         (cgroup_cs(scan->cg))->cpus_allowed);
816 }
817 
818 /**
819  * cpuset_change_cpumask - make a task's cpus_allowed the same as its cpuset's
820  * @tsk: task to test
821  * @scan: struct cgroup_scanner containing the cgroup of the task
822  *
823  * Called by cgroup_scan_tasks() for each task in a cgroup whose
824  * cpus_allowed mask needs to be changed.
825  *
826  * We don't need to re-check for the cgroup/cpuset membership, since we're
827  * holding cgroup_lock() at this point.
828  */
829 static void cpuset_change_cpumask(struct task_struct *tsk,
830                                   struct cgroup_scanner *scan)
831 {
832         set_cpus_allowed_ptr(tsk, ((cgroup_cs(scan->cg))->cpus_allowed));
833 }
834 
835 /**
836  * update_tasks_cpumask - Update the cpumasks of tasks in the cpuset.
837  * @cs: the cpuset in which each task's cpus_allowed mask needs to be changed
838  * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks()
839  *
840  * Called with cgroup_mutex held
841  *
842  * The cgroup_scan_tasks() function will scan all the tasks in a cgroup,
843  * calling callback functions for each.
844  *
845  * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0
846  * if @heap != NULL.
847  */
848 static void update_tasks_cpumask(struct cpuset *cs, struct ptr_heap *heap)
849 {
850         struct cgroup_scanner scan;
851 
852         scan.cg = cs->css.cgroup;
853         scan.test_task = cpuset_test_cpumask;
854         scan.process_task = cpuset_change_cpumask;
855         scan.heap = heap;
856         cgroup_scan_tasks(&scan);
857 }
858 
859 /**
860  * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it
861  * @cs: the cpuset to consider
862  * @buf: buffer of cpu numbers written to this cpuset
863  */
864 static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
865                           const char *buf)
866 {
867         struct ptr_heap heap;
868         int retval;
869         int is_load_balanced;
870 
871         /* top_cpuset.cpus_allowed tracks cpu_online_mask; it's read-only */
872         if (cs == &top_cpuset)
873                 return -EACCES;
874 
875         /*
876          * An empty cpus_allowed is ok only if the cpuset has no tasks.
877          * Since cpulist_parse() fails on an empty mask, we special case
878          * that parsing.  The validate_change() call ensures that cpusets
879          * with tasks have cpus.
880          */
881         if (!*buf) {
882                 cpumask_clear(trialcs->cpus_allowed);
883         } else {
884                 retval = cpulist_parse(buf, trialcs->cpus_allowed);
885                 if (retval < 0)
886                         return retval;
887 
888                 if (!cpumask_subset(trialcs->cpus_allowed, cpu_active_mask))
889                         return -EINVAL;
890         }
891         retval = validate_change(cs, trialcs);
892         if (retval < 0)
893                 return retval;
894 
895         /* Nothing to do if the cpus didn't change */
896         if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed))
897                 return 0;
898 
899         retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL);
900         if (retval)
901                 return retval;
902 
903         is_load_balanced = is_sched_load_balance(trialcs);
904 
905         mutex_lock(&callback_mutex);
906         cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed);
907         mutex_unlock(&callback_mutex);
908 
909         /*
910          * Scan tasks in the cpuset, and update the cpumasks of any
911          * that need an update.
912          */
913         update_tasks_cpumask(cs, &heap);
914 
915         heap_free(&heap);
916 
917         if (is_load_balanced)
918                 async_rebuild_sched_domains();
919         return 0;
920 }
921 
922 /*
923  * cpuset_migrate_mm
924  *
925  *    Migrate memory region from one set of nodes to another.
926  *
927  *    Temporarilly set tasks mems_allowed to target nodes of migration,
928  *    so that the migration code can allocate pages on these nodes.
929  *
930  *    Call holding cgroup_mutex, so current's cpuset won't change
931  *    during this call, as manage_mutex holds off any cpuset_attach()
932  *    calls.  Therefore we don't need to take task_lock around the
933  *    call to guarantee_online_mems(), as we know no one is changing
934  *    our task's cpuset.
935  *
936  *    While the mm_struct we are migrating is typically from some
937  *    other task, the task_struct mems_allowed that we are hacking
938  *    is for our current task, which must allocate new pages for that
939  *    migrating memory region.
940  */
941 
942 static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from,
943                                                         const nodemask_t *to)
944 {
945         struct task_struct *tsk = current;
946 
947         tsk->mems_allowed = *to;
948 
949         do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL);
950 
951         guarantee_online_mems(task_cs(tsk),&tsk->mems_allowed);
952 }
953 
954 /*
955  * cpuset_change_task_nodemask - change task's mems_allowed and mempolicy
956  * @tsk: the task to change
957  * @newmems: new nodes that the task will be set
958  *
959  * In order to avoid seeing no nodes if the old and new nodes are disjoint,
960  * we structure updates as setting all new allowed nodes, then clearing newly
961  * disallowed ones.
962  */
963 static void cpuset_change_task_nodemask(struct task_struct *tsk,
964                                         nodemask_t *newmems)
965 {
966         bool need_loop;
967 
968         /*
969          * Allow tasks that have access to memory reserves because they have
970          * been OOM killed to get memory anywhere.
971          */
972         if (unlikely(test_thread_flag(TIF_MEMDIE)))
973                 return;
974         if (current->flags & PF_EXITING) /* Let dying task have memory */
975                 return;
976 
977         task_lock(tsk);
978         /*
979          * Determine if a loop is necessary if another thread is doing
980          * get_mems_allowed().  If at least one node remains unchanged and
981          * tsk does not have a mempolicy, then an empty nodemask will not be
982          * possible when mems_allowed is larger than a word.
983          */
984         need_loop = task_has_mempolicy(tsk) ||
985                         !nodes_intersects(*newmems, tsk->mems_allowed);
986 
987         if (need_loop) {
988                 local_irq_disable();
989                 write_seqcount_begin(&tsk->mems_allowed_seq);
990         }
991 
992         nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems);
993         mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP1);
994 
995         mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP2);
996         tsk->mems_allowed = *newmems;
997 
998         if (need_loop) {
999                 write_seqcount_end(&tsk->mems_allowed_seq);
1000                 local_irq_enable();
1001         }
1002 
1003         task_unlock(tsk);
1004 }
1005 
1006 /*
1007  * Update task's mems_allowed and rebind its mempolicy and vmas' mempolicy
1008  * of it to cpuset's new mems_allowed, and migrate pages to new nodes if
1009  * memory_migrate flag is set. Called with cgroup_mutex held.
1010  */
1011 static void cpuset_change_nodemask(struct task_struct *p,
1012                                    struct cgroup_scanner *scan)
1013 {
1014         struct mm_struct *mm;
1015         struct cpuset *cs;
1016         int migrate;
1017         const nodemask_t *oldmem = scan->data;
1018         static nodemask_t newmems;      /* protected by cgroup_mutex */
1019 
1020         cs = cgroup_cs(scan->cg);
1021         guarantee_online_mems(cs, &newmems);
1022 
1023         cpuset_change_task_nodemask(p, &newmems);
1024 
1025         mm = get_task_mm(p);
1026         if (!mm)
1027                 return;
1028 
1029         migrate = is_memory_migrate(cs);
1030 
1031         mpol_rebind_mm(mm, &cs->mems_allowed);
1032         if (migrate)
1033                 cpuset_migrate_mm(mm, oldmem, &cs->mems_allowed);
1034         mmput(mm);
1035 }
1036 
1037 static void *cpuset_being_rebound;
1038 
1039 /**
1040  * update_tasks_nodemask - Update the nodemasks of tasks in the cpuset.
1041  * @cs: the cpuset in which each task's mems_allowed mask needs to be changed
1042  * @oldmem: old mems_allowed of cpuset cs
1043  * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks()
1044  *
1045  * Called with cgroup_mutex held
1046  * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0
1047  * if @heap != NULL.
1048  */
1049 static void update_tasks_nodemask(struct cpuset *cs, const nodemask_t *oldmem,
1050                                  struct ptr_heap *heap)
1051 {
1052         struct cgroup_scanner scan;
1053 
1054         cpuset_being_rebound = cs;              /* causes mpol_dup() rebind */
1055 
1056         scan.cg = cs->css.cgroup;
1057         scan.test_task = NULL;
1058         scan.process_task = cpuset_change_nodemask;
1059         scan.heap = heap;
1060         scan.data = (nodemask_t *)oldmem;
1061 
1062         /*
1063          * The mpol_rebind_mm() call takes mmap_sem, which we couldn't
1064          * take while holding tasklist_lock.  Forks can happen - the
1065          * mpol_dup() cpuset_being_rebound check will catch such forks,
1066          * and rebind their vma mempolicies too.  Because we still hold
1067          * the global cgroup_mutex, we know that no other rebind effort
1068          * will be contending for the global variable cpuset_being_rebound.
1069          * It's ok if we rebind the same mm twice; mpol_rebind_mm()
1070          * is idempotent.  Also migrate pages in each mm to new nodes.
1071          */
1072         cgroup_scan_tasks(&scan);
1073 
1074         /* We're done rebinding vmas to this cpuset's new mems_allowed. */
1075         cpuset_being_rebound = NULL;
1076 }
1077 
1078 /*
1079  * Handle user request to change the 'mems' memory placement
1080  * of a cpuset.  Needs to validate the request, update the
1081  * cpusets mems_allowed, and for each task in the cpuset,
1082  * update mems_allowed and rebind task's mempolicy and any vma
1083  * mempolicies and if the cpuset is marked 'memory_migrate',
1084  * migrate the tasks pages to the new memory.
1085  *
1086  * Call with cgroup_mutex held.  May take callback_mutex during call.
1087  * Will take tasklist_lock, scan tasklist for tasks in cpuset cs,
1088  * lock each such tasks mm->mmap_sem, scan its vma's and rebind
1089  * their mempolicies to the cpusets new mems_allowed.
1090  */
1091 static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs,
1092                            const char *buf)
1093 {
1094         NODEMASK_ALLOC(nodemask_t, oldmem, GFP_KERNEL);
1095         int retval;
1096         struct ptr_heap heap;
1097 
1098         if (!oldmem)
1099                 return -ENOMEM;
1100 
1101         /*
1102          * top_cpuset.mems_allowed tracks node_stats[N_HIGH_MEMORY];
1103          * it's read-only
1104          */
1105         if (cs == &top_cpuset) {
1106                 retval = -EACCES;
1107                 goto done;
1108         }
1109 
1110         /*
1111          * An empty mems_allowed is ok iff there are no tasks in the cpuset.
1112          * Since nodelist_parse() fails on an empty mask, we special case
1113          * that parsing.  The validate_change() call ensures that cpusets
1114          * with tasks have memory.
1115          */
1116         if (!*buf) {
1117                 nodes_clear(trialcs->mems_allowed);
1118         } else {
1119                 retval = nodelist_parse(buf, trialcs->mems_allowed);
1120                 if (retval < 0)
1121                         goto done;
1122 
1123                 if (!nodes_subset(trialcs->mems_allowed,
1124                                 node_states[N_HIGH_MEMORY])) {
1125                         retval =  -EINVAL;
1126                         goto done;
1127                 }
1128         }
1129         *oldmem = cs->mems_allowed;
1130         if (nodes_equal(*oldmem, trialcs->mems_allowed)) {
1131                 retval = 0;             /* Too easy - nothing to do */
1132                 goto done;
1133         }
1134         retval = validate_change(cs, trialcs);
1135         if (retval < 0)
1136                 goto done;
1137 
1138         retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL);
1139         if (retval < 0)
1140                 goto done;
1141 
1142         mutex_lock(&callback_mutex);
1143         cs->mems_allowed = trialcs->mems_allowed;
1144         mutex_unlock(&callback_mutex);
1145 
1146         update_tasks_nodemask(cs, oldmem, &heap);
1147 
1148         heap_free(&heap);
1149 done:
1150         NODEMASK_FREE(oldmem);
1151         return retval;
1152 }
1153 
1154 int current_cpuset_is_being_rebound(void)
1155 {
1156         int ret;
1157 
1158         rcu_read_lock();
1159         ret = task_cs(current) == cpuset_being_rebound;
1160         rcu_read_unlock();
1161 
1162         return ret;
1163 }
1164 
1165 static int update_relax_domain_level(struct cpuset *cs, s64 val)
1166 {
1167 #ifdef CONFIG_SMP
1168         if (val < -1 || val >= sched_domain_level_max)
1169                 return -EINVAL;
1170 #endif
1171 
1172         if (val != cs->relax_domain_level) {
1173                 cs->relax_domain_level = val;
1174                 if (!cpumask_empty(cs->cpus_allowed) &&
1175                     is_sched_load_balance(cs))
1176                         async_rebuild_sched_domains();
1177         }
1178 
1179         return 0;
1180 }
1181 
1182 /*
1183  * cpuset_change_flag - make a task's spread flags the same as its cpuset's
1184  * @tsk: task to be updated
1185  * @scan: struct cgroup_scanner containing the cgroup of the task
1186  *
1187  * Called by cgroup_scan_tasks() for each task in a cgroup.
1188  *
1189  * We don't need to re-check for the cgroup/cpuset membership, since we're
1190  * holding cgroup_lock() at this point.
1191  */
1192 static void cpuset_change_flag(struct task_struct *tsk,
1193                                 struct cgroup_scanner *scan)
1194 {
1195         cpuset_update_task_spread_flag(cgroup_cs(scan->cg), tsk);
1196 }
1197 
1198 /*
1199  * update_tasks_flags - update the spread flags of tasks in the cpuset.
1200  * @cs: the cpuset in which each task's spread flags needs to be changed
1201  * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks()
1202  *
1203  * Called with cgroup_mutex held
1204  *
1205  * The cgroup_scan_tasks() function will scan all the tasks in a cgroup,
1206  * calling callback functions for each.
1207  *
1208  * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0
1209  * if @heap != NULL.
1210  */
1211 static void update_tasks_flags(struct cpuset *cs, struct ptr_heap *heap)
1212 {
1213         struct cgroup_scanner scan;
1214 
1215         scan.cg = cs->css.cgroup;
1216         scan.test_task = NULL;
1217         scan.process_task = cpuset_change_flag;
1218         scan.heap = heap;
1219         cgroup_scan_tasks(&scan);
1220 }
1221 
1222 /*
1223  * update_flag - read a 0 or a 1 in a file and update associated flag
1224  * bit:         the bit to update (see cpuset_flagbits_t)
1225  * cs:          the cpuset to update
1226  * turning_on:  whether the flag is being set or cleared
1227  *
1228  * Call with cgroup_mutex held.
1229  */
1230 
1231 static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs,
1232                        int turning_on)
1233 {
1234         struct cpuset *trialcs;
1235         int balance_flag_changed;
1236         int spread_flag_changed;
1237         struct ptr_heap heap;
1238         int err;
1239 
1240         trialcs = alloc_trial_cpuset(cs);
1241         if (!trialcs)
1242                 return -ENOMEM;
1243 
1244         if (turning_on)
1245                 set_bit(bit, &trialcs->flags);
1246         else
1247                 clear_bit(bit, &trialcs->flags);
1248 
1249         err = validate_change(cs, trialcs);
1250         if (err < 0)
1251                 goto out;
1252 
1253         err = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL);
1254         if (err < 0)
1255                 goto out;
1256 
1257         balance_flag_changed = (is_sched_load_balance(cs) !=
1258                                 is_sched_load_balance(trialcs));
1259 
1260         spread_flag_changed = ((is_spread_slab(cs) != is_spread_slab(trialcs))
1261                         || (is_spread_page(cs) != is_spread_page(trialcs)));
1262 
1263         mutex_lock(&callback_mutex);
1264         cs->flags = trialcs->flags;
1265         mutex_unlock(&callback_mutex);
1266 
1267         if (!cpumask_empty(trialcs->cpus_allowed) && balance_flag_changed)
1268                 async_rebuild_sched_domains();
1269 
1270         if (spread_flag_changed)
1271                 update_tasks_flags(cs, &heap);
1272         heap_free(&heap);
1273 out:
1274         free_trial_cpuset(trialcs);
1275         return err;
1276 }
1277 
1278 /*
1279  * Frequency meter - How fast is some event occurring?
1280  *
1281  * These routines manage a digitally filtered, constant time based,
1282  * event frequency meter.  There are four routines:
1283  *   fmeter_init() - initialize a frequency meter.
1284  *   fmeter_markevent() - called each time the event happens.
1285  *   fmeter_getrate() - returns the recent rate of such events.
1286  *   fmeter_update() - internal routine used to update fmeter.
1287  *
1288  * A common data structure is passed to each of these routines,
1289  * which is used to keep track of the state required to manage the
1290  * frequency meter and its digital filter.
1291  *
1292  * The filter works on the number of events marked per unit time.
1293  * The filter is single-pole low-pass recursive (IIR).  The time unit
1294  * is 1 second.  Arithmetic is done using 32-bit integers scaled to
1295  * simulate 3 decimal digits of precision (multiplied by 1000).
1296  *
1297  * With an FM_COEF of 933, and a time base of 1 second, the filter
1298  * has a half-life of 10 seconds, meaning that if the events quit
1299  * happening, then the rate returned from the fmeter_getrate()
1300  * will be cut in half each 10 seconds, until it converges to zero.
1301  *
1302  * It is not worth doing a real infinitely recursive filter.  If more
1303  * than FM_MAXTICKS ticks have elapsed since the last filter event,
1304  * just compute FM_MAXTICKS ticks worth, by which point the level
1305  * will be stable.
1306  *
1307  * Limit the count of unprocessed events to FM_MAXCNT, so as to avoid
1308  * arithmetic overflow in the fmeter_update() routine.
1309  *
1310  * Given the simple 32 bit integer arithmetic used, this meter works
1311  * best for reporting rates between one per millisecond (msec) and
1312  * one per 32 (approx) seconds.  At constant rates faster than one
1313  * per msec it maxes out at values just under 1,000,000.  At constant
1314  * rates between one per msec, and one per second it will stabilize
1315  * to a value N*1000, where N is the rate of events per second.
1316  * At constant rates between one per second and one per 32 seconds,
1317  * it will be choppy, moving up on the seconds that have an event,
1318  * and then decaying until the next event.  At rates slower than
1319  * about one in 32 seconds, it decays all the way back to zero between
1320  * each event.
1321  */
1322 
1323 #define FM_COEF 933             /* coefficient for half-life of 10 secs */
1324 #define FM_MAXTICKS ((time_t)99) /* useless computing more ticks than this */
1325 #define FM_MAXCNT 1000000       /* limit cnt to avoid overflow */
1326 #define FM_SCALE 1000           /* faux fixed point scale */
1327 
1328 /* Initialize a frequency meter */
1329 static void fmeter_init(struct fmeter *fmp)
1330 {
1331         fmp->cnt = 0;
1332         fmp->val = 0;
1333         fmp->time = 0;
1334         spin_lock_init(&fmp->lock);
1335 }
1336 
1337 /* Internal meter update - process cnt events and update value */
1338 static void fmeter_update(struct fmeter *fmp)
1339 {
1340         time_t now = get_seconds();
1341         time_t ticks = now - fmp->time;
1342 
1343         if (ticks == 0)
1344                 return;
1345 
1346         ticks = min(FM_MAXTICKS, ticks);
1347         while (ticks-- > 0)
1348                 fmp->val = (FM_COEF * fmp->val) / FM_SCALE;
1349         fmp->time = now;
1350 
1351         fmp->val += ((FM_SCALE - FM_COEF) * fmp->cnt) / FM_SCALE;
1352         fmp->cnt = 0;
1353 }
1354 
1355 /* Process any previous ticks, then bump cnt by one (times scale). */
1356 static void fmeter_markevent(struct fmeter *fmp)
1357 {
1358         spin_lock(&fmp->lock);
1359         fmeter_update(fmp);
1360         fmp->cnt = min(FM_MAXCNT, fmp->cnt + FM_SCALE);
1361         spin_unlock(&fmp->lock);
1362 }
1363 
1364 /* Process any previous ticks, then return current value. */
1365 static int fmeter_getrate(struct fmeter *fmp)
1366 {
1367         int val;
1368 
1369         spin_lock(&fmp->lock);
1370         fmeter_update(fmp);
1371         val = fmp->val;
1372         spin_unlock(&fmp->lock);
1373         return val;
1374 }
1375 
1376 /*
1377  * Protected by cgroup_lock. The nodemasks must be stored globally because
1378  * dynamically allocating them is not allowed in can_attach, and they must
1379  * persist until attach.
1380  */
1381 static cpumask_var_t cpus_attach;
1382 static nodemask_t cpuset_attach_nodemask_from;
1383 static nodemask_t cpuset_attach_nodemask_to;
1384 
1385 /* Called by cgroups to determine if a cpuset is usable; cgroup_mutex held */
1386 static int cpuset_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
1387 {
1388         struct cpuset *cs = cgroup_cs(cgrp);
1389         struct task_struct *task;
1390         int ret;
1391 
1392         if (cpumask_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed))
1393                 return -ENOSPC;
1394 
1395         cgroup_taskset_for_each(task, cgrp, tset) {
1396                 /*
1397                  * Kthreads bound to specific cpus cannot be moved to a new
1398                  * cpuset; we cannot change their cpu affinity and
1399                  * isolating such threads by their set of allowed nodes is
1400                  * unnecessary.  Thus, cpusets are not applicable for such
1401                  * threads.  This prevents checking for success of
1402                  * set_cpus_allowed_ptr() on all attached tasks before
1403                  * cpus_allowed may be changed.
1404                  */
1405                 if (task->flags & PF_THREAD_BOUND)
1406                         return -EINVAL;
1407                 if ((ret = security_task_setscheduler(task)))
1408                         return ret;
1409         }
1410 
1411         /* prepare for attach */
1412         if (cs == &top_cpuset)
1413                 cpumask_copy(cpus_attach, cpu_possible_mask);
1414         else
1415                 guarantee_online_cpus(cs, cpus_attach);
1416 
1417         guarantee_online_mems(cs, &cpuset_attach_nodemask_to);
1418 
1419         return 0;
1420 }
1421 
1422 static void cpuset_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
1423 {
1424         struct mm_struct *mm;
1425         struct task_struct *task;
1426         struct task_struct *leader = cgroup_taskset_first(tset);
1427         struct cgroup *oldcgrp = cgroup_taskset_cur_cgroup(tset);
1428         struct cpuset *cs = cgroup_cs(cgrp);
1429         struct cpuset *oldcs = cgroup_cs(oldcgrp);
1430 
1431         cgroup_taskset_for_each(task, cgrp, tset) {
1432                 /*
1433                  * can_attach beforehand should guarantee that this doesn't
1434                  * fail.  TODO: have a better way to handle failure here
1435                  */
1436                 WARN_ON_ONCE(set_cpus_allowed_ptr(task, cpus_attach));
1437 
1438                 cpuset_change_task_nodemask(task, &cpuset_attach_nodemask_to);
1439                 cpuset_update_task_spread_flag(cs, task);
1440         }
1441 
1442         /*
1443          * Change mm, possibly for multiple threads in a threadgroup. This is
1444          * expensive and may sleep.
1445          */
1446         cpuset_attach_nodemask_from = oldcs->mems_allowed;
1447         cpuset_attach_nodemask_to = cs->mems_allowed;
1448         mm = get_task_mm(leader);
1449         if (mm) {
1450                 mpol_rebind_mm(mm, &cpuset_attach_nodemask_to);
1451                 if (is_memory_migrate(cs))
1452                         cpuset_migrate_mm(mm, &cpuset_attach_nodemask_from,
1453                                           &cpuset_attach_nodemask_to);
1454                 mmput(mm);
1455         }
1456 }
1457 
1458 /* The various types of files and directories in a cpuset file system */
1459 
1460 typedef enum {
1461         FILE_MEMORY_MIGRATE,
1462         FILE_CPULIST,
1463         FILE_MEMLIST,
1464         FILE_CPU_EXCLUSIVE,
1465         FILE_MEM_EXCLUSIVE,
1466         FILE_MEM_HARDWALL,
1467         FILE_SCHED_LOAD_BALANCE,
1468         FILE_SCHED_RELAX_DOMAIN_LEVEL,
1469         FILE_MEMORY_PRESSURE_ENABLED,
1470         FILE_MEMORY_PRESSURE,
1471         FILE_SPREAD_PAGE,
1472         FILE_SPREAD_SLAB,
1473 } cpuset_filetype_t;
1474 
1475 static int cpuset_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val)
1476 {
1477         int retval = 0;
1478         struct cpuset *cs = cgroup_cs(cgrp);
1479         cpuset_filetype_t type = cft->private;
1480 
1481         if (!cgroup_lock_live_group(cgrp))
1482                 return -ENODEV;
1483 
1484         switch (type) {
1485         case FILE_CPU_EXCLUSIVE:
1486                 retval = update_flag(CS_CPU_EXCLUSIVE, cs, val);
1487                 break;
1488         case FILE_MEM_EXCLUSIVE:
1489                 retval = update_flag(CS_MEM_EXCLUSIVE, cs, val);
1490                 break;
1491         case FILE_MEM_HARDWALL:
1492                 retval = update_flag(CS_MEM_HARDWALL, cs, val);
1493                 break;
1494         case FILE_SCHED_LOAD_BALANCE:
1495                 retval = update_flag(CS_SCHED_LOAD_BALANCE, cs, val);
1496                 break;
1497         case FILE_MEMORY_MIGRATE:
1498                 retval = update_flag(CS_MEMORY_MIGRATE, cs, val);
1499                 break;
1500         case FILE_MEMORY_PRESSURE_ENABLED:
1501                 cpuset_memory_pressure_enabled = !!val;
1502                 break;
1503         case FILE_MEMORY_PRESSURE:
1504                 retval = -EACCES;
1505                 break;
1506         case FILE_SPREAD_PAGE:
1507                 retval = update_flag(CS_SPREAD_PAGE, cs, val);
1508                 break;
1509         case FILE_SPREAD_SLAB:
1510                 retval = update_flag(CS_SPREAD_SLAB, cs, val);
1511                 break;
1512         default:
1513                 retval = -EINVAL;
1514                 break;
1515         }
1516         cgroup_unlock();
1517         return retval;
1518 }
1519 
1520 static int cpuset_write_s64(struct cgroup *cgrp, struct cftype *cft, s64 val)
1521 {
1522         int retval = 0;
1523         struct cpuset *cs = cgroup_cs(cgrp);
1524         cpuset_filetype_t type = cft->private;
1525 
1526         if (!cgroup_lock_live_group(cgrp))
1527                 return -ENODEV;
1528 
1529         switch (type) {
1530         case FILE_SCHED_RELAX_DOMAIN_LEVEL:
1531                 retval = update_relax_domain_level(cs, val);
1532                 break;
1533         default:
1534                 retval = -EINVAL;
1535                 break;
1536         }
1537         cgroup_unlock();
1538         return retval;
1539 }
1540 
1541 /*
1542  * Common handling for a write to a "cpus" or "mems" file.
1543  */
1544 static int cpuset_write_resmask(struct cgroup *cgrp, struct cftype *cft,
1545                                 const char *buf)
1546 {
1547         int retval = 0;
1548         struct cpuset *cs = cgroup_cs(cgrp);
1549         struct cpuset *trialcs;
1550 
1551         if (!cgroup_lock_live_group(cgrp))
1552                 return -ENODEV;
1553 
1554         trialcs = alloc_trial_cpuset(cs);
1555         if (!trialcs) {
1556                 retval = -ENOMEM;
1557                 goto out;
1558         }
1559 
1560         switch (cft->private) {
1561         case FILE_CPULIST:
1562                 retval = update_cpumask(cs, trialcs, buf);
1563                 break;
1564         case FILE_MEMLIST:
1565                 retval = update_nodemask(cs, trialcs, buf);
1566                 break;
1567         default:
1568                 retval = -EINVAL;
1569                 break;
1570         }
1571 
1572         free_trial_cpuset(trialcs);
1573 out:
1574         cgroup_unlock();
1575         return retval;
1576 }
1577 
1578 /*
1579  * These ascii lists should be read in a single call, by using a user
1580  * buffer large enough to hold the entire map.  If read in smaller
1581  * chunks, there is no guarantee of atomicity.  Since the display format
1582  * used, list of ranges of sequential numbers, is variable length,
1583  * and since these maps can change value dynamically, one could read
1584  * gibberish by doing partial reads while a list was changing.
1585  * A single large read to a buffer that crosses a page boundary is
1586  * ok, because the result being copied to user land is not recomputed
1587  * across a page fault.
1588  */
1589 
1590 static size_t cpuset_sprintf_cpulist(char *page, struct cpuset *cs)
1591 {
1592         size_t count;
1593 
1594         mutex_lock(&callback_mutex);
1595         count = cpulist_scnprintf(page, PAGE_SIZE, cs->cpus_allowed);
1596         mutex_unlock(&callback_mutex);
1597 
1598         return count;
1599 }
1600 
1601 static size_t cpuset_sprintf_memlist(char *page, struct cpuset *cs)
1602 {
1603         size_t count;
1604 
1605         mutex_lock(&callback_mutex);
1606         count = nodelist_scnprintf(page, PAGE_SIZE, cs->mems_allowed);
1607         mutex_unlock(&callback_mutex);
1608 
1609         return count;
1610 }
1611 
1612 static ssize_t cpuset_common_file_read(struct cgroup *cont,
1613                                        struct cftype *cft,
1614                                        struct file *file,
1615                                        char __user *buf,
1616                                        size_t nbytes, loff_t *ppos)
1617 {
1618         struct cpuset *cs = cgroup_cs(cont);
1619         cpuset_filetype_t type = cft->private;
1620         char *page;
1621         ssize_t retval = 0;
1622         char *s;
1623 
1624         if (!(page = (char *)__get_free_page(GFP_TEMPORARY)))
1625                 return -ENOMEM;
1626 
1627         s = page;
1628 
1629         switch (type) {
1630         case FILE_CPULIST:
1631                 s += cpuset_sprintf_cpulist(s, cs);
1632                 break;
1633         case FILE_MEMLIST:
1634                 s += cpuset_sprintf_memlist(s, cs);
1635                 break;
1636         default:
1637                 retval = -EINVAL;
1638                 goto out;
1639         }
1640         *s++ = '\n';
1641 
1642         retval = simple_read_from_buffer(buf, nbytes, ppos, page, s - page);
1643 out:
1644         free_page((unsigned long)page);
1645         return retval;
1646 }
1647 
1648 static u64 cpuset_read_u64(struct cgroup *cont, struct cftype *cft)
1649 {
1650         struct cpuset *cs = cgroup_cs(cont);
1651         cpuset_filetype_t type = cft->private;
1652         switch (type) {
1653         case FILE_CPU_EXCLUSIVE:
1654                 return is_cpu_exclusive(cs);
1655         case FILE_MEM_EXCLUSIVE:
1656                 return is_mem_exclusive(cs);
1657         case FILE_MEM_HARDWALL:
1658                 return is_mem_hardwall(cs);
1659         case FILE_SCHED_LOAD_BALANCE:
1660                 return is_sched_load_balance(cs);
1661         case FILE_MEMORY_MIGRATE:
1662                 return is_memory_migrate(cs);
1663         case FILE_MEMORY_PRESSURE_ENABLED:
1664                 return cpuset_memory_pressure_enabled;
1665         case FILE_MEMORY_PRESSURE:
1666                 return fmeter_getrate(&cs->fmeter);
1667         case FILE_SPREAD_PAGE:
1668                 return is_spread_page(cs);
1669         case FILE_SPREAD_SLAB:
1670                 return is_spread_slab(cs);
1671         default:
1672                 BUG();
1673         }
1674 
1675         /* Unreachable but makes gcc happy */
1676         return 0;
1677 }
1678 
1679 static s64 cpuset_read_s64(struct cgroup *cont, struct cftype *cft)
1680 {
1681         struct cpuset *cs = cgroup_cs(cont);
1682         cpuset_filetype_t type = cft->private;
1683         switch (type) {
1684         case FILE_SCHED_RELAX_DOMAIN_LEVEL:
1685                 return cs->relax_domain_level;
1686         default:
1687                 BUG();
1688         }
1689 
1690         /* Unrechable but makes gcc happy */
1691         return 0;
1692 }
1693 
1694 
1695 /*
1696  * for the common functions, 'private' gives the type of file
1697  */
1698 
1699 static struct cftype files[] = {
1700         {
1701                 .name = "cpus",
1702                 .read = cpuset_common_file_read,
1703                 .write_string = cpuset_write_resmask,
1704                 .max_write_len = (100U + 6 * NR_CPUS),
1705                 .private = FILE_CPULIST,
1706         },
1707 
1708         {
1709                 .name = "mems",
1710                 .read = cpuset_common_file_read,
1711                 .write_string = cpuset_write_resmask,
1712                 .max_write_len = (100U + 6 * MAX_NUMNODES),
1713                 .private = FILE_MEMLIST,
1714         },
1715 
1716         {
1717                 .name = "cpu_exclusive",
1718                 .read_u64 = cpuset_read_u64,
1719                 .write_u64 = cpuset_write_u64,
1720                 .private = FILE_CPU_EXCLUSIVE,
1721         },
1722 
1723         {
1724                 .name = "mem_exclusive",
1725                 .read_u64 = cpuset_read_u64,
1726                 .write_u64 = cpuset_write_u64,
1727                 .private = FILE_MEM_EXCLUSIVE,
1728         },
1729 
1730         {
1731                 .name = "mem_hardwall",
1732                 .read_u64 = cpuset_read_u64,
1733                 .write_u64 = cpuset_write_u64,
1734                 .private = FILE_MEM_HARDWALL,
1735         },
1736 
1737         {
1738                 .name = "sched_load_balance",
1739                 .read_u64 = cpuset_read_u64,
1740                 .write_u64 = cpuset_write_u64,
1741                 .private = FILE_SCHED_LOAD_BALANCE,
1742         },
1743 
1744         {
1745                 .name = "sched_relax_domain_level",
1746                 .read_s64 = cpuset_read_s64,
1747                 .write_s64 = cpuset_write_s64,
1748                 .private = FILE_SCHED_RELAX_DOMAIN_LEVEL,
1749         },
1750 
1751         {
1752                 .name = "memory_migrate",
1753                 .read_u64 = cpuset_read_u64,
1754                 .write_u64 = cpuset_write_u64,
1755                 .private = FILE_MEMORY_MIGRATE,
1756         },
1757 
1758         {
1759                 .name = "memory_pressure",
1760                 .read_u64 = cpuset_read_u64,
1761                 .write_u64 = cpuset_write_u64,
1762                 .private = FILE_MEMORY_PRESSURE,
1763                 .mode = S_IRUGO,
1764         },
1765 
1766         {
1767                 .name = "memory_spread_page",
1768                 .read_u64 = cpuset_read_u64,
1769                 .write_u64 = cpuset_write_u64,
1770                 .private = FILE_SPREAD_PAGE,
1771         },
1772 
1773         {
1774                 .name = "memory_spread_slab",
1775                 .read_u64 = cpuset_read_u64,
1776                 .write_u64 = cpuset_write_u64,
1777                 .private = FILE_SPREAD_SLAB,
1778         },
1779 };
1780 
1781 static struct cftype cft_memory_pressure_enabled = {
1782         .name = "memory_pressure_enabled",
1783         .read_u64 = cpuset_read_u64,
1784         .write_u64 = cpuset_write_u64,
1785         .private = FILE_MEMORY_PRESSURE_ENABLED,
1786 };
1787 
1788 static int cpuset_populate(struct cgroup_subsys *ss, struct cgroup *cont)
1789 {
1790         int err;
1791 
1792         err = cgroup_add_files(cont, ss, files, ARRAY_SIZE(files));
1793         if (err)
1794                 return err;
1795         /* memory_pressure_enabled is in root cpuset only */
1796         if (!cont->parent)
1797                 err = cgroup_add_file(cont, ss,
1798                                       &cft_memory_pressure_enabled);
1799         return err;
1800 }
1801 
1802 /*
1803  * post_clone() is called during cgroup_create() when the
1804  * clone_children mount argument was specified.  The cgroup
1805  * can not yet have any tasks.
1806  *
1807  * Currently we refuse to set up the cgroup - thereby
1808  * refusing the task to be entered, and as a result refusing
1809  * the sys_unshare() or clone() which initiated it - if any
1810  * sibling cpusets have exclusive cpus or mem.
1811  *
1812  * If this becomes a problem for some users who wish to
1813  * allow that scenario, then cpuset_post_clone() could be
1814  * changed to grant parent->cpus_allowed-sibling_cpus_exclusive
1815  * (and likewise for mems) to the new cgroup. Called with cgroup_mutex
1816  * held.
1817  */
1818 static void cpuset_post_clone(struct cgroup *cgroup)
1819 {
1820         struct cgroup *parent, *child;
1821         struct cpuset *cs, *parent_cs;
1822 
1823         parent = cgroup->parent;
1824         list_for_each_entry(child, &parent->children, sibling) {
1825                 cs = cgroup_cs(child);
1826                 if (is_mem_exclusive(cs) || is_cpu_exclusive(cs))
1827                         return;
1828         }
1829         cs = cgroup_cs(cgroup);
1830         parent_cs = cgroup_cs(parent);
1831 
1832         mutex_lock(&callback_mutex);
1833         cs->mems_allowed = parent_cs->mems_allowed;
1834         cpumask_copy(cs->cpus_allowed, parent_cs->cpus_allowed);
1835         mutex_unlock(&callback_mutex);
1836         return;
1837 }
1838 
1839 /*
1840  *      cpuset_create - create a cpuset
1841  *      cont:   control group that the new cpuset will be part of
1842  */
1843 
1844 static struct cgroup_subsys_state *cpuset_create(struct cgroup *cont)
1845 {
1846         struct cpuset *cs;
1847         struct cpuset *parent;
1848 
1849         if (!cont->parent) {
1850                 return &top_cpuset.css;
1851         }
1852         parent = cgroup_cs(cont->parent);
1853         cs = kmalloc(sizeof(*cs), GFP_KERNEL);
1854         if (!cs)
1855                 return ERR_PTR(-ENOMEM);
1856         if (!alloc_cpumask_var(&cs->cpus_allowed, GFP_KERNEL)) {
1857                 kfree(cs);
1858                 return ERR_PTR(-ENOMEM);
1859         }
1860 
1861         cs->flags = 0;
1862         if (is_spread_page(parent))
1863                 set_bit(CS_SPREAD_PAGE, &cs->flags);
1864         if (is_spread_slab(parent))
1865                 set_bit(CS_SPREAD_SLAB, &cs->flags);
1866         set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
1867         cpumask_clear(cs->cpus_allowed);
1868         nodes_clear(cs->mems_allowed);
1869         fmeter_init(&cs->fmeter);
1870         cs->relax_domain_level = -1;
1871 
1872         cs->parent = parent;
1873         number_of_cpusets++;
1874         return &cs->css ;
1875 }
1876 
1877 /*
1878  * If the cpuset being removed has its flag 'sched_load_balance'
1879  * enabled, then simulate turning sched_load_balance off, which
1880  * will call async_rebuild_sched_domains().
1881  */
1882 
1883 static void cpuset_destroy(struct cgroup *cont)
1884 {
1885         struct cpuset *cs = cgroup_cs(cont);
1886 
1887         if (is_sched_load_balance(cs))
1888                 update_flag(CS_SCHED_LOAD_BALANCE, cs, 0);
1889 
1890         number_of_cpusets--;
1891         free_cpumask_var(cs->cpus_allowed);
1892         kfree(cs);
1893 }
1894 
1895 struct cgroup_subsys cpuset_subsys = {
1896         .name = "cpuset",
1897         .create = cpuset_create,
1898         .destroy = cpuset_destroy,
1899         .can_attach = cpuset_can_attach,
1900         .attach = cpuset_attach,
1901         .populate = cpuset_populate,
1902         .post_clone = cpuset_post_clone,
1903         .subsys_id = cpuset_subsys_id,
1904         .early_init = 1,
1905 };
1906 
1907 /**
1908  * cpuset_init - initialize cpusets at system boot
1909  *
1910  * Description: Initialize top_cpuset and the cpuset internal file system,
1911  **/
1912 
1913 int __init cpuset_init(void)
1914 {
1915         int err = 0;
1916 
1917         if (!alloc_cpumask_var(&top_cpuset.cpus_allowed, GFP_KERNEL))
1918                 BUG();
1919 
1920         cpumask_setall(top_cpuset.cpus_allowed);
1921         nodes_setall(top_cpuset.mems_allowed);
1922 
1923         fmeter_init(&top_cpuset.fmeter);
1924         set_bit(CS_SCHED_LOAD_BALANCE, &top_cpuset.flags);
1925         top_cpuset.relax_domain_level = -1;
1926 
1927         err = register_filesystem(&cpuset_fs_type);
1928         if (err < 0)
1929                 return err;
1930 
1931         if (!alloc_cpumask_var(&cpus_attach, GFP_KERNEL))
1932                 BUG();
1933 
1934         number_of_cpusets = 1;
1935         return 0;
1936 }
1937 
1938 /**
1939  * cpuset_do_move_task - move a given task to another cpuset
1940  * @tsk: pointer to task_struct the task to move
1941  * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner
1942  *
1943  * Called by cgroup_scan_tasks() for each task in a cgroup.
1944  * Return nonzero to stop the walk through the tasks.
1945  */
1946 static void cpuset_do_move_task(struct task_struct *tsk,
1947                                 struct cgroup_scanner *scan)
1948 {
1949         struct cgroup *new_cgroup = scan->data;
1950 
1951         cgroup_attach_task(new_cgroup, tsk);
1952 }
1953 
1954 /**
1955  * move_member_tasks_to_cpuset - move tasks from one cpuset to another
1956  * @from: cpuset in which the tasks currently reside
1957  * @to: cpuset to which the tasks will be moved
1958  *
1959  * Called with cgroup_mutex held
1960  * callback_mutex must not be held, as cpuset_attach() will take it.
1961  *
1962  * The cgroup_scan_tasks() function will scan all the tasks in a cgroup,
1963  * calling callback functions for each.
1964  */
1965 static void move_member_tasks_to_cpuset(struct cpuset *from, struct cpuset *to)
1966 {
1967         struct cgroup_scanner scan;
1968 
1969         scan.cg = from->css.cgroup;
1970         scan.test_task = NULL; /* select all tasks in cgroup */
1971         scan.process_task = cpuset_do_move_task;
1972         scan.heap = NULL;
1973         scan.data = to->css.cgroup;
1974 
1975         if (cgroup_scan_tasks(&scan))
1976                 printk(KERN_ERR "move_member_tasks_to_cpuset: "
1977                                 "cgroup_scan_tasks failed\n");
1978 }
1979 
1980 /*
1981  * If CPU and/or memory hotplug handlers, below, unplug any CPUs
1982  * or memory nodes, we need to walk over the cpuset hierarchy,
1983  * removing that CPU or node from all cpusets.  If this removes the
1984  * last CPU or node from a cpuset, then move the tasks in the empty
1985  * cpuset to its next-highest non-empty parent.
1986  *
1987  * Called with cgroup_mutex held
1988  * callback_mutex must not be held, as cpuset_attach() will take it.
1989  */
1990 static void remove_tasks_in_empty_cpuset(struct cpuset *cs)
1991 {
1992         struct cpuset *parent;
1993 
1994         /*
1995          * The cgroup's css_sets list is in use if there are tasks
1996          * in the cpuset; the list is empty if there are none;
1997          * the cs->css.refcnt seems always 0.
1998          */
1999         if (list_empty(&cs->css.cgroup->css_sets))
2000                 return;
2001 
2002         /*
2003          * Find its next-highest non-empty parent, (top cpuset
2004          * has online cpus, so can't be empty).
2005          */
2006         parent = cs->parent;
2007         while (cpumask_empty(parent->cpus_allowed) ||
2008                         nodes_empty(parent->mems_allowed))
2009                 parent = parent->parent;
2010 
2011         move_member_tasks_to_cpuset(cs, parent);
2012 }
2013 
2014 /*
2015  * Walk the specified cpuset subtree and look for empty cpusets.
2016  * The tasks of such cpuset must be moved to a parent cpuset.
2017  *
2018  * Called with cgroup_mutex held.  We take callback_mutex to modify
2019  * cpus_allowed and mems_allowed.
2020  *
2021  * This walk processes the tree from top to bottom, completing one layer
2022  * before dropping down to the next.  It always processes a node before
2023  * any of its children.
2024  *
2025  * For now, since we lack memory hot unplug, we'll never see a cpuset
2026  * that has tasks along with an empty 'mems'.  But if we did see such
2027  * a cpuset, we'd handle it just like we do if its 'cpus' was empty.
2028  */
2029 static void scan_for_empty_cpusets(struct cpuset *root)
2030 {
2031         LIST_HEAD(queue);
2032         struct cpuset *cp;      /* scans cpusets being updated */
2033         struct cpuset *child;   /* scans child cpusets of cp */
2034         struct cgroup *cont;
2035         static nodemask_t oldmems;      /* protected by cgroup_mutex */
2036 
2037         list_add_tail((struct list_head *)&root->stack_list, &queue);
2038 
2039         while (!list_empty(&queue)) {
2040                 cp = list_first_entry(&queue, struct cpuset, stack_list);
2041                 list_del(queue.next);
2042                 list_for_each_entry(cont, &cp->css.cgroup->children, sibling) {
2043                         child = cgroup_cs(cont);
2044                         list_add_tail(&child->stack_list, &queue);
2045                 }
2046 
2047                 /* Continue past cpusets with all cpus, mems online */
2048                 if (cpumask_subset(cp->cpus_allowed, cpu_active_mask) &&
2049                     nodes_subset(cp->mems_allowed, node_states[N_HIGH_MEMORY]))
2050                         continue;
2051 
2052                 oldmems = cp->mems_allowed;
2053 
2054                 /* Remove offline cpus and mems from this cpuset. */
2055                 mutex_lock(&callback_mutex);
2056                 cpumask_and(cp->cpus_allowed, cp->cpus_allowed,
2057                             cpu_active_mask);
2058                 nodes_and(cp->mems_allowed, cp->mems_allowed,
2059                                                 node_states[N_HIGH_MEMORY]);
2060                 mutex_unlock(&callback_mutex);
2061 
2062                 /* Move tasks from the empty cpuset to a parent */
2063                 if (cpumask_empty(cp->cpus_allowed) ||
2064                      nodes_empty(cp->mems_allowed))
2065                         remove_tasks_in_empty_cpuset(cp);
2066                 else {
2067                         update_tasks_cpumask(cp, NULL);
2068                         update_tasks_nodemask(cp, &oldmems, NULL);
2069                 }
2070         }
2071 }
2072 
2073 /*
2074  * The top_cpuset tracks what CPUs and Memory Nodes are online,
2075  * period.  This is necessary in order to make cpusets transparent
2076  * (of no affect) on systems that are actively using CPU hotplug
2077  * but making no active use of cpusets.
2078  *
2079  * The only exception to this is suspend/resume, where we don't
2080  * modify cpusets at all.
2081  *
2082  * This routine ensures that top_cpuset.cpus_allowed tracks
2083  * cpu_active_mask on each CPU hotplug (cpuhp) event.
2084  *
2085  * Called within get_online_cpus().  Needs to call cgroup_lock()
2086  * before calling generate_sched_domains().
2087  */
2088 void cpuset_update_active_cpus(void)
2089 {
2090         struct sched_domain_attr *attr;
2091         cpumask_var_t *doms;
2092         int ndoms;
2093 
2094         cgroup_lock();
2095         mutex_lock(&callback_mutex);
2096         cpumask_copy(top_cpuset.cpus_allowed, cpu_active_mask);
2097         mutex_unlock(&callback_mutex);
2098         scan_for_empty_cpusets(&top_cpuset);
2099         ndoms = generate_sched_domains(&doms, &attr);
2100         cgroup_unlock();
2101 
2102         /* Have scheduler rebuild the domains */
2103         partition_sched_domains(ndoms, doms, attr);
2104 }
2105 
2106 #ifdef CONFIG_MEMORY_HOTPLUG
2107 /*
2108  * Keep top_cpuset.mems_allowed tracking node_states[N_HIGH_MEMORY].
2109  * Call this routine anytime after node_states[N_HIGH_MEMORY] changes.
2110  * See also the previous routine cpuset_track_online_cpus().
2111  */
2112 static int cpuset_track_online_nodes(struct notifier_block *self,
2113                                 unsigned long action, void *arg)
2114 {
2115         static nodemask_t oldmems;      /* protected by cgroup_mutex */
2116 
2117         cgroup_lock();
2118         switch (action) {
2119         case MEM_ONLINE:
2120                 oldmems = top_cpuset.mems_allowed;
2121                 mutex_lock(&callback_mutex);
2122                 top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY];
2123                 mutex_unlock(&callback_mutex);
2124                 update_tasks_nodemask(&top_cpuset, &oldmems, NULL);
2125                 break;
2126         case MEM_OFFLINE:
2127                 /*
2128                  * needn't update top_cpuset.mems_allowed explicitly because
2129                  * scan_for_empty_cpusets() will update it.
2130                  */
2131                 scan_for_empty_cpusets(&top_cpuset);
2132                 break;
2133         default:
2134                 break;
2135         }
2136         cgroup_unlock();
2137 
2138         return NOTIFY_OK;
2139 }
2140 #endif
2141 
2142 /**
2143  * cpuset_init_smp - initialize cpus_allowed
2144  *
2145  * Description: Finish top cpuset after cpu, node maps are initialized
2146  **/
2147 
2148 void __init cpuset_init_smp(void)
2149 {
2150         cpumask_copy(top_cpuset.cpus_allowed, cpu_active_mask);
2151         top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY];
2152 
2153         hotplug_memory_notifier(cpuset_track_online_nodes, 10);
2154 
2155         cpuset_wq = create_singlethread_workqueue("cpuset");
2156         BUG_ON(!cpuset_wq);
2157 }
2158 
2159 /**
2160  * cpuset_cpus_allowed - return cpus_allowed mask from a tasks cpuset.
2161  * @tsk: pointer to task_struct from which to obtain cpuset->cpus_allowed.
2162  * @pmask: pointer to struct cpumask variable to receive cpus_allowed set.
2163  *
2164  * Description: Returns the cpumask_var_t cpus_allowed of the cpuset
2165  * attached to the specified @tsk.  Guaranteed to return some non-empty
2166  * subset of cpu_online_mask, even if this means going outside the
2167  * tasks cpuset.
2168  **/
2169 
2170 void cpuset_cpus_allowed(struct task_struct *tsk, struct cpumask *pmask)
2171 {
2172         mutex_lock(&callback_mutex);
2173         task_lock(tsk);
2174         guarantee_online_cpus(task_cs(tsk), pmask);
2175         task_unlock(tsk);
2176         mutex_unlock(&callback_mutex);
2177 }
2178 
2179 void cpuset_cpus_allowed_fallback(struct task_struct *tsk)
2180 {
2181         const struct cpuset *cs;
2182 
2183         rcu_read_lock();
2184         cs = task_cs(tsk);
2185         if (cs)
2186                 do_set_cpus_allowed(tsk, cs->cpus_allowed);
2187         rcu_read_unlock();
2188 
2189         /*
2190          * We own tsk->cpus_allowed, nobody can change it under us.
2191          *
2192          * But we used cs && cs->cpus_allowed lockless and thus can
2193          * race with cgroup_attach_task() or update_cpumask() and get
2194          * the wrong tsk->cpus_allowed. However, both cases imply the
2195          * subsequent cpuset_change_cpumask()->set_cpus_allowed_ptr()
2196          * which takes task_rq_lock().
2197          *
2198          * If we are called after it dropped the lock we must see all
2199          * changes in tsk_cs()->cpus_allowed. Otherwise we can temporary
2200          * set any mask even if it is not right from task_cs() pov,
2201          * the pending set_cpus_allowed_ptr() will fix things.
2202          *
2203          * select_fallback_rq() will fix things ups and set cpu_possible_mask
2204          * if required.
2205          */
2206 }
2207 
2208 void cpuset_init_current_mems_allowed(void)
2209 {
2210         nodes_setall(current->mems_allowed);
2211 }
2212 
2213 /**
2214  * cpuset_mems_allowed - return mems_allowed mask from a tasks cpuset.
2215  * @tsk: pointer to task_struct from which to obtain cpuset->mems_allowed.
2216  *
2217  * Description: Returns the nodemask_t mems_allowed of the cpuset
2218  * attached to the specified @tsk.  Guaranteed to return some non-empty
2219  * subset of node_states[N_HIGH_MEMORY], even if this means going outside the
2220  * tasks cpuset.
2221  **/
2222 
2223 nodemask_t cpuset_mems_allowed(struct task_struct *tsk)
2224 {
2225         nodemask_t mask;
2226 
2227         mutex_lock(&callback_mutex);
2228         task_lock(tsk);
2229         guarantee_online_mems(task_cs(tsk), &mask);
2230         task_unlock(tsk);
2231         mutex_unlock(&callback_mutex);
2232 
2233         return mask;
2234 }
2235 
2236 /**
2237  * cpuset_nodemask_valid_mems_allowed - check nodemask vs. curremt mems_allowed
2238  * @nodemask: the nodemask to be checked
2239  *
2240  * Are any of the nodes in the nodemask allowed in current->mems_allowed?
2241  */
2242 int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask)
2243 {
2244         return nodes_intersects(*nodemask, current->mems_allowed);
2245 }
2246 
2247 /*
2248  * nearest_hardwall_ancestor() - Returns the nearest mem_exclusive or
2249  * mem_hardwall ancestor to the specified cpuset.  Call holding
2250  * callback_mutex.  If no ancestor is mem_exclusive or mem_hardwall
2251  * (an unusual configuration), then returns the root cpuset.
2252  */
2253 static const struct cpuset *nearest_hardwall_ancestor(const struct cpuset *cs)
2254 {
2255         while (!(is_mem_exclusive(cs) || is_mem_hardwall(cs)) && cs->parent)
2256                 cs = cs->parent;
2257         return cs;
2258 }
2259 
2260 /**
2261  * cpuset_node_allowed_softwall - Can we allocate on a memory node?
2262  * @node: is this an allowed node?
2263  * @gfp_mask: memory allocation flags
2264  *
2265  * If we're in interrupt, yes, we can always allocate.  If __GFP_THISNODE is
2266  * set, yes, we can always allocate.  If node is in our task's mems_allowed,
2267  * yes.  If it's not a __GFP_HARDWALL request and this node is in the nearest
2268  * hardwalled cpuset ancestor to this task's cpuset, yes.  If the task has been
2269  * OOM killed and has access to memory reserves as specified by the TIF_MEMDIE
2270  * flag, yes.
2271  * Otherwise, no.
2272  *
2273  * If __GFP_HARDWALL is set, cpuset_node_allowed_softwall() reduces to
2274  * cpuset_node_allowed_hardwall().  Otherwise, cpuset_node_allowed_softwall()
2275  * might sleep, and might allow a node from an enclosing cpuset.
2276  *
2277  * cpuset_node_allowed_hardwall() only handles the simpler case of hardwall
2278  * cpusets, and never sleeps.
2279  *
2280  * The __GFP_THISNODE placement logic is really handled elsewhere,
2281  * by forcibly using a zonelist starting at a specified node, and by
2282  * (in get_page_from_freelist()) refusing to consider the zones for
2283  * any node on the zonelist except the first.  By the time any such
2284  * calls get to this routine, we should just shut up and say 'yes'.
2285  *
2286  * GFP_USER allocations are marked with the __GFP_HARDWALL bit,
2287  * and do not allow allocations outside the current tasks cpuset
2288  * unless the task has been OOM killed as is marked TIF_MEMDIE.
2289  * GFP_KERNEL allocations are not so marked, so can escape to the
2290  * nearest enclosing hardwalled ancestor cpuset.
2291  *
2292  * Scanning up parent cpusets requires callback_mutex.  The
2293  * __alloc_pages() routine only calls here with __GFP_HARDWALL bit
2294  * _not_ set if it's a GFP_KERNEL allocation, and all nodes in the
2295  * current tasks mems_allowed came up empty on the first pass over
2296  * the zonelist.  So only GFP_KERNEL allocations, if all nodes in the
2297  * cpuset are short of memory, might require taking the callback_mutex
2298  * mutex.
2299  *
2300  * The first call here from mm/page_alloc:get_page_from_freelist()
2301  * has __GFP_HARDWALL set in gfp_mask, enforcing hardwall cpusets,
2302  * so no allocation on a node outside the cpuset is allowed (unless
2303  * in interrupt, of course).
2304  *
2305  * The second pass through get_page_from_freelist() doesn't even call
2306  * here for GFP_ATOMIC calls.  For those calls, the __alloc_pages()
2307  * variable 'wait' is not set, and the bit ALLOC_CPUSET is not set
2308  * in alloc_flags.  That logic and the checks below have the combined
2309  * affect that:
2310  *      in_interrupt - any node ok (current task context irrelevant)
2311  *      GFP_ATOMIC   - any node ok
2312  *      TIF_MEMDIE   - any node ok
2313  *      GFP_KERNEL   - any node in enclosing hardwalled cpuset ok
2314  *      GFP_USER     - only nodes in current tasks mems allowed ok.
2315  *
2316  * Rule:
2317  *    Don't call cpuset_node_allowed_softwall if you can't sleep, unless you
2318  *    pass in the __GFP_HARDWALL flag set in gfp_flag, which disables
2319  *    the code that might scan up ancestor cpusets and sleep.
2320  */
2321 int __cpuset_node_allowed_softwall(int node, gfp_t gfp_mask)
2322 {
2323         const struct cpuset *cs;        /* current cpuset ancestors */
2324         int allowed;                    /* is allocation in zone z allowed? */
2325 
2326         if (in_interrupt() || (gfp_mask & __GFP_THISNODE))
2327                 return 1;
2328         might_sleep_if(!(gfp_mask & __GFP_HARDWALL));
2329         if (node_isset(node, current->mems_allowed))
2330                 return 1;
2331         /*
2332          * Allow tasks that have access to memory reserves because they have
2333          * been OOM killed to get memory anywhere.
2334          */
2335         if (unlikely(test_thread_flag(TIF_MEMDIE)))
2336                 return 1;
2337         if (gfp_mask & __GFP_HARDWALL)  /* If hardwall request, stop here */
2338                 return 0;
2339 
2340         if (current->flags & PF_EXITING) /* Let dying task have memory */
2341                 return 1;
2342 
2343         /* Not hardwall and node outside mems_allowed: scan up cpusets */
2344         mutex_lock(&callback_mutex);
2345 
2346         task_lock(current);
2347         cs = nearest_hardwall_ancestor(task_cs(current));
2348         allowed = node_isset(node, cs->mems_allowed);
2349         task_unlock(current);
2350 
2351         mutex_unlock(&callback_mutex);
2352         return allowed;
2353 }
2354 
2355 /*
2356  * cpuset_node_allowed_hardwall - Can we allocate on a memory node?
2357  * @node: is this an allowed node?
2358  * @gfp_mask: memory allocation flags
2359  *
2360  * If we're in interrupt, yes, we can always allocate.  If __GFP_THISNODE is
2361  * set, yes, we can always allocate.  If node is in our task's mems_allowed,
2362  * yes.  If the task has been OOM killed and has access to memory reserves as
2363  * specified by the TIF_MEMDIE flag, yes.
2364  * Otherwise, no.
2365  *
2366  * The __GFP_THISNODE placement logic is really handled elsewhere,
2367  * by forcibly using a zonelist starting at a specified node, and by
2368  * (in get_page_from_freelist()) refusing to consider the zones for
2369  * any node on the zonelist except the first.  By the time any such
2370  * calls get to this routine, we should just shut up and say 'yes'.
2371  *
2372  * Unlike the cpuset_node_allowed_softwall() variant, above,
2373  * this variant requires that the node be in the current task's
2374  * mems_allowed or that we're in interrupt.  It does not scan up the
2375  * cpuset hierarchy for the nearest enclosing mem_exclusive cpuset.
2376  * It never sleeps.
2377  */
2378 int __cpuset_node_allowed_hardwall(int node, gfp_t gfp_mask)
2379 {
2380         if (in_interrupt() || (gfp_mask & __GFP_THISNODE))
2381                 return 1;
2382         if (node_isset(node, current->mems_allowed))
2383                 return 1;
2384         /*
2385          * Allow tasks that have access to memory reserves because they have
2386          * been OOM killed to get memory anywhere.
2387          */
2388         if (unlikely(test_thread_flag(TIF_MEMDIE)))
2389                 return 1;
2390         return 0;
2391 }
2392 
2393 /**
2394  * cpuset_unlock - release lock on cpuset changes
2395  *
2396  * Undo the lock taken in a previous cpuset_lock() call.
2397  */
2398 
2399 void cpuset_unlock(void)
2400 {
2401         mutex_unlock(&callback_mutex);
2402 }
2403 
2404 /**
2405  * cpuset_mem_spread_node() - On which node to begin search for a file page
2406  * cpuset_slab_spread_node() - On which node to begin search for a slab page
2407  *
2408  * If a task is marked PF_SPREAD_PAGE or PF_SPREAD_SLAB (as for
2409  * tasks in a cpuset with is_spread_page or is_spread_slab set),
2410  * and if the memory allocation used cpuset_mem_spread_node()
2411  * to determine on which node to start looking, as it will for
2412  * certain page cache or slab cache pages such as used for file
2413  * system buffers and inode caches, then instead of starting on the
2414  * local node to look for a free page, rather spread the starting
2415  * node around the tasks mems_allowed nodes.
2416  *
2417  * We don't have to worry about the returned node being offline
2418  * because "it can't happen", and even if it did, it would be ok.
2419  *
2420  * The routines calling guarantee_online_mems() are careful to
2421  * only set nodes in task->mems_allowed that are online.  So it
2422  * should not be possible for the following code to return an
2423  * offline node.  But if it did, that would be ok, as this routine
2424  * is not returning the node where the allocation must be, only
2425  * the node where the search should start.  The zonelist passed to
2426  * __alloc_pages() will include all nodes.  If the slab allocator
2427  * is passed an offline node, it will fall back to the local node.
2428  * See kmem_cache_alloc_node().
2429  */
2430 
2431 static int cpuset_spread_node(int *rotor)
2432 {
2433         int node;
2434 
2435         node = next_node(*rotor, current->mems_allowed);
2436         if (node == MAX_NUMNODES)
2437                 node = first_node(current->mems_allowed);
2438         *rotor = node;
2439         return node;
2440 }
2441 
2442 int cpuset_mem_spread_node(void)
2443 {
2444         if (current->cpuset_mem_spread_rotor == NUMA_NO_NODE)
2445                 current->cpuset_mem_spread_rotor =
2446                         node_random(&current->mems_allowed);
2447 
2448         return cpuset_spread_node(&current->cpuset_mem_spread_rotor);
2449 }
2450 
2451 int cpuset_slab_spread_node(void)
2452 {
2453         if (current->cpuset_slab_spread_rotor == NUMA_NO_NODE)
2454                 current->cpuset_slab_spread_rotor =
2455                         node_random(&current->mems_allowed);
2456 
2457         return cpuset_spread_node(&current->cpuset_slab_spread_rotor);
2458 }
2459 
2460 EXPORT_SYMBOL_GPL(cpuset_mem_spread_node);
2461 
2462 /**
2463  * cpuset_mems_allowed_intersects - Does @tsk1's mems_allowed intersect @tsk2's?
2464  * @tsk1: pointer to task_struct of some task.
2465  * @tsk2: pointer to task_struct of some other task.
2466  *
2467  * Description: Return true if @tsk1's mems_allowed intersects the
2468  * mems_allowed of @tsk2.  Used by the OOM killer to determine if
2469  * one of the task's memory usage might impact the memory available
2470  * to the other.
2471  **/
2472 
2473 int cpuset_mems_allowed_intersects(const struct task_struct *tsk1,
2474                                    const struct task_struct *tsk2)
2475 {
2476         return nodes_intersects(tsk1->mems_allowed, tsk2->mems_allowed);
2477 }
2478 
2479 /**
2480  * cpuset_print_task_mems_allowed - prints task's cpuset and mems_allowed
2481  * @task: pointer to task_struct of some task.
2482  *
2483  * Description: Prints @task's name, cpuset name, and cached copy of its
2484  * mems_allowed to the kernel log.  Must hold task_lock(task) to allow
2485  * dereferencing task_cs(task).
2486  */
2487 void cpuset_print_task_mems_allowed(struct task_struct *tsk)
2488 {
2489         struct dentry *dentry;
2490 
2491         dentry = task_cs(tsk)->css.cgroup->dentry;
2492         spin_lock(&cpuset_buffer_lock);
2493 
2494         if (!dentry) {
2495                 strcpy(cpuset_name, "/");
2496         } else {
2497                 spin_lock(&dentry->d_lock);
2498                 strlcpy(cpuset_name, (const char *)dentry->d_name.name,
2499                         CPUSET_NAME_LEN);
2500                 spin_unlock(&dentry->d_lock);
2501         }
2502 
2503         nodelist_scnprintf(cpuset_nodelist, CPUSET_NODELIST_LEN,
2504                            tsk->mems_allowed);
2505         printk(KERN_INFO "%s cpuset=%s mems_allowed=%s\n",
2506                tsk->comm, cpuset_name, cpuset_nodelist);
2507         spin_unlock(&cpuset_buffer_lock);
2508 }
2509 
2510 /*
2511  * Collection of memory_pressure is suppressed unless
2512  * this flag is enabled by writing "1" to the special
2513  * cpuset file 'memory_pressure_enabled' in the root cpuset.
2514  */
2515 
2516 int cpuset_memory_pressure_enabled __read_mostly;
2517 
2518 /**
2519  * cpuset_memory_pressure_bump - keep stats of per-cpuset reclaims.
2520  *
2521  * Keep a running average of the rate of synchronous (direct)
2522  * page reclaim efforts initiated by tasks in each cpuset.
2523  *
2524  * This represents the rate at which some task in the cpuset
2525  * ran low on memory on all nodes it was allowed to use, and
2526  * had to enter the kernels page reclaim code in an effort to
2527  * create more free memory by tossing clean pages or swapping
2528  * or writing dirty pages.
2529  *
2530  * Display to user space in the per-cpuset read-only file
2531  * "memory_pressure".  Value displayed is an integer
2532  * representing the recent rate of entry into the synchronous
2533  * (direct) page reclaim by any task attached to the cpuset.
2534  **/
2535 
2536 void __cpuset_memory_pressure_bump(void)
2537 {
2538         task_lock(current);
2539         fmeter_markevent(&task_cs(current)->fmeter);
2540         task_unlock(current);
2541 }
2542 
2543 #ifdef CONFIG_PROC_PID_CPUSET
2544 /*
2545  * proc_cpuset_show()
2546  *  - Print tasks cpuset path into seq_file.
2547  *  - Used for /proc/<pid>/cpuset.
2548  *  - No need to task_lock(tsk) on this tsk->cpuset reference, as it
2549  *    doesn't really matter if tsk->cpuset changes after we read it,
2550  *    and we take cgroup_mutex, keeping cpuset_attach() from changing it
2551  *    anyway.
2552  */
2553 static int proc_cpuset_show(struct seq_file *m, void *unused_v)
2554 {
2555         struct pid *pid;
2556         struct task_struct *tsk;
2557         char *buf;
2558         struct cgroup_subsys_state *css;
2559         int retval;
2560 
2561         retval = -ENOMEM;
2562         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
2563         if (!buf)
2564                 goto out;
2565 
2566         retval = -ESRCH;
2567         pid = m->private;
2568         tsk = get_pid_task(pid, PIDTYPE_PID);
2569         if (!tsk)
2570                 goto out_free;
2571 
2572         retval = -EINVAL;
2573         cgroup_lock();
2574         css = task_subsys_state(tsk, cpuset_subsys_id);
2575         retval = cgroup_path(css->cgroup, buf, PAGE_SIZE);
2576         if (retval < 0)
2577                 goto out_unlock;
2578         seq_puts(m, buf);
2579         seq_putc(m, '\n');
2580 out_unlock:
2581         cgroup_unlock();
2582         put_task_struct(tsk);
2583 out_free:
2584         kfree(buf);
2585 out:
2586         return retval;
2587 }
2588 
2589 static int cpuset_open(struct inode *inode, struct file *file)
2590 {
2591         struct pid *pid = PROC_I(inode)->pid;
2592         return single_open(file, proc_cpuset_show, pid);
2593 }
2594 
2595 const struct file_operations proc_cpuset_operations = {
2596         .open           = cpuset_open,
2597         .read           = seq_read,
2598         .llseek         = seq_lseek,
2599         .release        = single_release,
2600 };
2601 #endif /* CONFIG_PROC_PID_CPUSET */
2602 
2603 /* Display task mems_allowed in /proc/<pid>/status file. */
2604 void cpuset_task_status_allowed(struct seq_file *m, struct task_struct *task)
2605 {
2606         seq_printf(m, "Mems_allowed:\t");
2607         seq_nodemask(m, &task->mems_allowed);
2608         seq_printf(m, "\n");
2609         seq_printf(m, "Mems_allowed_list:\t");
2610         seq_nodemask_list(m, &task->mems_allowed);
2611         seq_printf(m, "\n");
2612 }
2613 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp