linux/kernel/cpuset.c
<<
>>
Prefs
   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 */
  69static 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 */
  76int number_of_cpusets __read_mostly;
  77
  78/* Forward declare cgroup structures */
  79struct cgroup_subsys cpuset_subsys;
  80struct cpuset;
  81
  82/* See "Frequency meter" comments, below. */
  83
  84struct 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
  91struct 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 */
 113static 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 */
 120static 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
 127static inline bool task_has_mempolicy(struct task_struct *task)
 128{
 129        return task->mempolicy;
 130}
 131#else
 132static 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 */
 140typedef 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/* the type of hotplug event */
 151enum hotplug_event {
 152        CPUSET_CPU_OFFLINE,
 153        CPUSET_MEM_OFFLINE,
 154};
 155
 156/* convenient tests for these bits */
 157static inline int is_cpu_exclusive(const struct cpuset *cs)
 158{
 159        return test_bit(CS_CPU_EXCLUSIVE, &cs->flags);
 160}
 161
 162static inline int is_mem_exclusive(const struct cpuset *cs)
 163{
 164        return test_bit(CS_MEM_EXCLUSIVE, &cs->flags);
 165}
 166
 167static inline int is_mem_hardwall(const struct cpuset *cs)
 168{
 169        return test_bit(CS_MEM_HARDWALL, &cs->flags);
 170}
 171
 172static inline int is_sched_load_balance(const struct cpuset *cs)
 173{
 174        return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
 175}
 176
 177static inline int is_memory_migrate(const struct cpuset *cs)
 178{
 179        return test_bit(CS_MEMORY_MIGRATE, &cs->flags);
 180}
 181
 182static inline int is_spread_page(const struct cpuset *cs)
 183{
 184        return test_bit(CS_SPREAD_PAGE, &cs->flags);
 185}
 186
 187static inline int is_spread_slab(const struct cpuset *cs)
 188{
 189        return test_bit(CS_SPREAD_SLAB, &cs->flags);
 190}
 191
 192static struct cpuset top_cpuset = {
 193        .flags = ((1 << CS_CPU_EXCLUSIVE) | (1 << CS_MEM_EXCLUSIVE)),
 194};
 195
 196/*
 197 * There are two global mutexes guarding cpuset structures.  The first
 198 * is the main control groups cgroup_mutex, accessed via
 199 * cgroup_lock()/cgroup_unlock().  The second is the cpuset-specific
 200 * callback_mutex, below. They can nest.  It is ok to first take
 201 * cgroup_mutex, then nest callback_mutex.  We also require taking
 202 * task_lock() when dereferencing a task's cpuset pointer.  See "The
 203 * task_lock() exception", at the end of this comment.
 204 *
 205 * A task must hold both mutexes to modify cpusets.  If a task
 206 * holds cgroup_mutex, then it blocks others wanting that mutex,
 207 * ensuring that it is the only task able to also acquire callback_mutex
 208 * and be able to modify cpusets.  It can perform various checks on
 209 * the cpuset structure first, knowing nothing will change.  It can
 210 * also allocate memory while just holding cgroup_mutex.  While it is
 211 * performing these checks, various callback routines can briefly
 212 * acquire callback_mutex to query cpusets.  Once it is ready to make
 213 * the changes, it takes callback_mutex, blocking everyone else.
 214 *
 215 * Calls to the kernel memory allocator can not be made while holding
 216 * callback_mutex, as that would risk double tripping on callback_mutex
 217 * from one of the callbacks into the cpuset code from within
 218 * __alloc_pages().
 219 *
 220 * If a task is only holding callback_mutex, then it has read-only
 221 * access to cpusets.
 222 *
 223 * Now, the task_struct fields mems_allowed and mempolicy may be changed
 224 * by other task, we use alloc_lock in the task_struct fields to protect
 225 * them.
 226 *
 227 * The cpuset_common_file_read() handlers only hold callback_mutex across
 228 * small pieces of code, such as when reading out possibly multi-word
 229 * cpumasks and nodemasks.
 230 *
 231 * Accessing a task's cpuset should be done in accordance with the
 232 * guidelines for accessing subsystem state in kernel/cgroup.c
 233 */
 234
 235static DEFINE_MUTEX(callback_mutex);
 236
 237/*
 238 * cpuset_buffer_lock protects both the cpuset_name and cpuset_nodelist
 239 * buffers.  They are statically allocated to prevent using excess stack
 240 * when calling cpuset_print_task_mems_allowed().
 241 */
 242#define CPUSET_NAME_LEN         (128)
 243#define CPUSET_NODELIST_LEN     (256)
 244static char cpuset_name[CPUSET_NAME_LEN];
 245static char cpuset_nodelist[CPUSET_NODELIST_LEN];
 246static DEFINE_SPINLOCK(cpuset_buffer_lock);
 247
 248/*
 249 * This is ugly, but preserves the userspace API for existing cpuset
 250 * users. If someone tries to mount the "cpuset" filesystem, we
 251 * silently switch it to mount "cgroup" instead
 252 */
 253static struct dentry *cpuset_mount(struct file_system_type *fs_type,
 254                         int flags, const char *unused_dev_name, void *data)
 255{
 256        struct file_system_type *cgroup_fs = get_fs_type("cgroup");
 257        struct dentry *ret = ERR_PTR(-ENODEV);
 258        if (cgroup_fs) {
 259                char mountopts[] =
 260                        "cpuset,noprefix,"
 261                        "release_agent=/sbin/cpuset_release_agent";
 262                ret = cgroup_fs->mount(cgroup_fs, flags,
 263                                           unused_dev_name, mountopts);
 264                put_filesystem(cgroup_fs);
 265        }
 266        return ret;
 267}
 268
 269static struct file_system_type cpuset_fs_type = {
 270        .name = "cpuset",
 271        .mount = cpuset_mount,
 272};
 273
 274/*
 275 * Return in pmask the portion of a cpusets's cpus_allowed that
 276 * are online.  If none are online, walk up the cpuset hierarchy
 277 * until we find one that does have some online cpus.  If we get
 278 * all the way to the top and still haven't found any online cpus,
 279 * return cpu_online_mask.  Or if passed a NULL cs from an exit'ing
 280 * task, return cpu_online_mask.
 281 *
 282 * One way or another, we guarantee to return some non-empty subset
 283 * of cpu_online_mask.
 284 *
 285 * Call with callback_mutex held.
 286 */
 287
 288static void guarantee_online_cpus(const struct cpuset *cs,
 289                                  struct cpumask *pmask)
 290{
 291        while (cs && !cpumask_intersects(cs->cpus_allowed, cpu_online_mask))
 292                cs = cs->parent;
 293        if (cs)
 294                cpumask_and(pmask, cs->cpus_allowed, cpu_online_mask);
 295        else
 296                cpumask_copy(pmask, cpu_online_mask);
 297        BUG_ON(!cpumask_intersects(pmask, cpu_online_mask));
 298}
 299
 300/*
 301 * Return in *pmask the portion of a cpusets's mems_allowed that
 302 * are online, with memory.  If none are online with memory, walk
 303 * up the cpuset hierarchy until we find one that does have some
 304 * online mems.  If we get all the way to the top and still haven't
 305 * found any online mems, return node_states[N_MEMORY].
 306 *
 307 * One way or another, we guarantee to return some non-empty subset
 308 * of node_states[N_MEMORY].
 309 *
 310 * Call with callback_mutex held.
 311 */
 312
 313static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask)
 314{
 315        while (cs && !nodes_intersects(cs->mems_allowed,
 316                                        node_states[N_MEMORY]))
 317                cs = cs->parent;
 318        if (cs)
 319                nodes_and(*pmask, cs->mems_allowed,
 320                                        node_states[N_MEMORY]);
 321        else
 322                *pmask = node_states[N_MEMORY];
 323        BUG_ON(!nodes_intersects(*pmask, node_states[N_MEMORY]));
 324}
 325
 326/*
 327 * update task's spread flag if cpuset's page/slab spread flag is set
 328 *
 329 * Called with callback_mutex/cgroup_mutex held
 330 */
 331static void cpuset_update_task_spread_flag(struct cpuset *cs,
 332                                        struct task_struct *tsk)
 333{
 334        if (is_spread_page(cs))
 335                tsk->flags |= PF_SPREAD_PAGE;
 336        else
 337                tsk->flags &= ~PF_SPREAD_PAGE;
 338        if (is_spread_slab(cs))
 339                tsk->flags |= PF_SPREAD_SLAB;
 340        else
 341                tsk->flags &= ~PF_SPREAD_SLAB;
 342}
 343
 344/*
 345 * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q?
 346 *
 347 * One cpuset is a subset of another if all its allowed CPUs and
 348 * Memory Nodes are a subset of the other, and its exclusive flags
 349 * are only set if the other's are set.  Call holding cgroup_mutex.
 350 */
 351
 352static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q)
 353{
 354        return  cpumask_subset(p->cpus_allowed, q->cpus_allowed) &&
 355                nodes_subset(p->mems_allowed, q->mems_allowed) &&
 356                is_cpu_exclusive(p) <= is_cpu_exclusive(q) &&
 357                is_mem_exclusive(p) <= is_mem_exclusive(q);
 358}
 359
 360/**
 361 * alloc_trial_cpuset - allocate a trial cpuset
 362 * @cs: the cpuset that the trial cpuset duplicates
 363 */
 364static struct cpuset *alloc_trial_cpuset(const struct cpuset *cs)
 365{
 366        struct cpuset *trial;
 367
 368        trial = kmemdup(cs, sizeof(*cs), GFP_KERNEL);
 369        if (!trial)
 370                return NULL;
 371
 372        if (!alloc_cpumask_var(&trial->cpus_allowed, GFP_KERNEL)) {
 373                kfree(trial);
 374                return NULL;
 375        }
 376        cpumask_copy(trial->cpus_allowed, cs->cpus_allowed);
 377
 378        return trial;
 379}
 380
 381/**
 382 * free_trial_cpuset - free the trial cpuset
 383 * @trial: the trial cpuset to be freed
 384 */
 385static void free_trial_cpuset(struct cpuset *trial)
 386{
 387        free_cpumask_var(trial->cpus_allowed);
 388        kfree(trial);
 389}
 390
 391/*
 392 * validate_change() - Used to validate that any proposed cpuset change
 393 *                     follows the structural rules for cpusets.
 394 *
 395 * If we replaced the flag and mask values of the current cpuset
 396 * (cur) with those values in the trial cpuset (trial), would
 397 * our various subset and exclusive rules still be valid?  Presumes
 398 * cgroup_mutex held.
 399 *
 400 * 'cur' is the address of an actual, in-use cpuset.  Operations
 401 * such as list traversal that depend on the actual address of the
 402 * cpuset in the list must use cur below, not trial.
 403 *
 404 * 'trial' is the address of bulk structure copy of cur, with
 405 * perhaps one or more of the fields cpus_allowed, mems_allowed,
 406 * or flags changed to new, trial values.
 407 *
 408 * Return 0 if valid, -errno if not.
 409 */
 410
 411static int validate_change(const struct cpuset *cur, const struct cpuset *trial)
 412{
 413        struct cgroup *cont;
 414        struct cpuset *c, *par;
 415
 416        /* Each of our child cpusets must be a subset of us */
 417        list_for_each_entry(cont, &cur->css.cgroup->children, sibling) {
 418                if (!is_cpuset_subset(cgroup_cs(cont), trial))
 419                        return -EBUSY;
 420        }
 421
 422        /* Remaining checks don't apply to root cpuset */
 423        if (cur == &top_cpuset)
 424                return 0;
 425
 426        par = cur->parent;
 427
 428        /* We must be a subset of our parent cpuset */
 429        if (!is_cpuset_subset(trial, par))
 430                return -EACCES;
 431
 432        /*
 433         * If either I or some sibling (!= me) is exclusive, we can't
 434         * overlap
 435         */
 436        list_for_each_entry(cont, &par->css.cgroup->children, sibling) {
 437                c = cgroup_cs(cont);
 438                if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) &&
 439                    c != cur &&
 440                    cpumask_intersects(trial->cpus_allowed, c->cpus_allowed))
 441                        return -EINVAL;
 442                if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) &&
 443                    c != cur &&
 444                    nodes_intersects(trial->mems_allowed, c->mems_allowed))
 445                        return -EINVAL;
 446        }
 447
 448        /* Cpusets with tasks can't have empty cpus_allowed or mems_allowed */
 449        if (cgroup_task_count(cur->css.cgroup)) {
 450                if (cpumask_empty(trial->cpus_allowed) ||
 451                    nodes_empty(trial->mems_allowed)) {
 452                        return -ENOSPC;
 453                }
 454        }
 455
 456        return 0;
 457}
 458
 459#ifdef CONFIG_SMP
 460/*
 461 * Helper routine for generate_sched_domains().
 462 * Do cpusets a, b have overlapping cpus_allowed masks?
 463 */
 464static int cpusets_overlap(struct cpuset *a, struct cpuset *b)
 465{
 466        return cpumask_intersects(a->cpus_allowed, b->cpus_allowed);
 467}
 468
 469static void
 470update_domain_attr(struct sched_domain_attr *dattr, struct cpuset *c)
 471{
 472        if (dattr->relax_domain_level < c->relax_domain_level)
 473                dattr->relax_domain_level = c->relax_domain_level;
 474        return;
 475}
 476
 477static void
 478update_domain_attr_tree(struct sched_domain_attr *dattr, struct cpuset *c)
 479{
 480        LIST_HEAD(q);
 481
 482        list_add(&c->stack_list, &q);
 483        while (!list_empty(&q)) {
 484                struct cpuset *cp;
 485                struct cgroup *cont;
 486                struct cpuset *child;
 487
 488                cp = list_first_entry(&q, struct cpuset, stack_list);
 489                list_del(q.next);
 490
 491                if (cpumask_empty(cp->cpus_allowed))
 492                        continue;
 493
 494                if (is_sched_load_balance(cp))
 495                        update_domain_attr(dattr, cp);
 496
 497                list_for_each_entry(cont, &cp->css.cgroup->children, sibling) {
 498                        child = cgroup_cs(cont);
 499                        list_add_tail(&child->stack_list, &q);
 500                }
 501        }
 502}
 503
 504/*
 505 * generate_sched_domains()
 506 *
 507 * This function builds a partial partition of the systems CPUs
 508 * A 'partial partition' is a set of non-overlapping subsets whose
 509 * union is a subset of that set.
 510 * The output of this function needs to be passed to kernel/sched.c
 511 * partition_sched_domains() routine, which will rebuild the scheduler's
 512 * load balancing domains (sched domains) as specified by that partial
 513 * partition.
 514 *
 515 * See "What is sched_load_balance" in Documentation/cgroups/cpusets.txt
 516 * for a background explanation of this.
 517 *
 518 * Does not return errors, on the theory that the callers of this
 519 * routine would rather not worry about failures to rebuild sched
 520 * domains when operating in the severe memory shortage situations
 521 * that could cause allocation failures below.
 522 *
 523 * Must be called with cgroup_lock held.
 524 *
 525 * The three key local variables below are:
 526 *    q  - a linked-list queue of cpuset pointers, used to implement a
 527 *         top-down scan of all cpusets.  This scan loads a pointer
 528 *         to each cpuset marked is_sched_load_balance into the
 529 *         array 'csa'.  For our purposes, rebuilding the schedulers
 530 *         sched domains, we can ignore !is_sched_load_balance cpusets.
 531 *  csa  - (for CpuSet Array) Array of pointers to all the cpusets
 532 *         that need to be load balanced, for convenient iterative
 533 *         access by the subsequent code that finds the best partition,
 534 *         i.e the set of domains (subsets) of CPUs such that the
 535 *         cpus_allowed of every cpuset marked is_sched_load_balance
 536 *         is a subset of one of these domains, while there are as
 537 *         many such domains as possible, each as small as possible.
 538 * doms  - Conversion of 'csa' to an array of cpumasks, for passing to
 539 *         the kernel/sched.c routine partition_sched_domains() in a
 540 *         convenient format, that can be easily compared to the prior
 541 *         value to determine what partition elements (sched domains)
 542 *         were changed (added or removed.)
 543 *
 544 * Finding the best partition (set of domains):
 545 *      The triple nested loops below over i, j, k scan over the
 546 *      load balanced cpusets (using the array of cpuset pointers in
 547 *      csa[]) looking for pairs of cpusets that have overlapping
 548 *      cpus_allowed, but which don't have the same 'pn' partition
 549 *      number and gives them in the same partition number.  It keeps
 550 *      looping on the 'restart' label until it can no longer find
 551 *      any such pairs.
 552 *
 553 *      The union of the cpus_allowed masks from the set of
 554 *      all cpusets having the same 'pn' value then form the one
 555 *      element of the partition (one sched domain) to be passed to
 556 *      partition_sched_domains().
 557 */
 558static int generate_sched_domains(cpumask_var_t **domains,
 559                        struct sched_domain_attr **attributes)
 560{
 561        LIST_HEAD(q);           /* queue of cpusets to be scanned */
 562        struct cpuset *cp;      /* scans q */
 563        struct cpuset **csa;    /* array of all cpuset ptrs */
 564        int csn;                /* how many cpuset ptrs in csa so far */
 565        int i, j, k;            /* indices for partition finding loops */
 566        cpumask_var_t *doms;    /* resulting partition; i.e. sched domains */
 567        struct sched_domain_attr *dattr;  /* attributes for custom domains */
 568        int ndoms = 0;          /* number of sched domains in result */
 569        int nslot;              /* next empty doms[] struct cpumask slot */
 570
 571        doms = NULL;
 572        dattr = NULL;
 573        csa = NULL;
 574
 575        /* Special case for the 99% of systems with one, full, sched domain */
 576        if (is_sched_load_balance(&top_cpuset)) {
 577                ndoms = 1;
 578                doms = alloc_sched_domains(ndoms);
 579                if (!doms)
 580                        goto done;
 581
 582                dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL);
 583                if (dattr) {
 584                        *dattr = SD_ATTR_INIT;
 585                        update_domain_attr_tree(dattr, &top_cpuset);
 586                }
 587                cpumask_copy(doms[0], top_cpuset.cpus_allowed);
 588
 589                goto done;
 590        }
 591
 592        csa = kmalloc(number_of_cpusets * sizeof(cp), GFP_KERNEL);
 593        if (!csa)
 594                goto done;
 595        csn = 0;
 596
 597        list_add(&top_cpuset.stack_list, &q);
 598        while (!list_empty(&q)) {
 599                struct cgroup *cont;
 600                struct cpuset *child;   /* scans child cpusets of cp */
 601
 602                cp = list_first_entry(&q, struct cpuset, stack_list);
 603                list_del(q.next);
 604
 605                if (cpumask_empty(cp->cpus_allowed))
 606                        continue;
 607
 608                /*
 609                 * All child cpusets contain a subset of the parent's cpus, so
 610                 * just skip them, and then we call update_domain_attr_tree()
 611                 * to calc relax_domain_level of the corresponding sched
 612                 * domain.
 613                 */
 614                if (is_sched_load_balance(cp)) {
 615                        csa[csn++] = cp;
 616                        continue;
 617                }
 618
 619                list_for_each_entry(cont, &cp->css.cgroup->children, sibling) {
 620                        child = cgroup_cs(cont);
 621                        list_add_tail(&child->stack_list, &q);
 622                }
 623        }
 624
 625        for (i = 0; i < csn; i++)
 626                csa[i]->pn = i;
 627        ndoms = csn;
 628
 629restart:
 630        /* Find the best partition (set of sched domains) */
 631        for (i = 0; i < csn; i++) {
 632                struct cpuset *a = csa[i];
 633                int apn = a->pn;
 634
 635                for (j = 0; j < csn; j++) {
 636                        struct cpuset *b = csa[j];
 637                        int bpn = b->pn;
 638
 639                        if (apn != bpn && cpusets_overlap(a, b)) {
 640                                for (k = 0; k < csn; k++) {
 641                                        struct cpuset *c = csa[k];
 642
 643                                        if (c->pn == bpn)
 644                                                c->pn = apn;
 645                                }
 646                                ndoms--;        /* one less element */
 647                                goto restart;
 648                        }
 649                }
 650        }
 651
 652        /*
 653         * Now we know how many domains to create.
 654         * Convert <csn, csa> to <ndoms, doms> and populate cpu masks.
 655         */
 656        doms = alloc_sched_domains(ndoms);
 657        if (!doms)
 658                goto done;
 659
 660        /*
 661         * The rest of the code, including the scheduler, can deal with
 662         * dattr==NULL case. No need to abort if alloc fails.
 663         */
 664        dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL);
 665
 666        for (nslot = 0, i = 0; i < csn; i++) {
 667                struct cpuset *a = csa[i];
 668                struct cpumask *dp;
 669                int apn = a->pn;
 670
 671                if (apn < 0) {
 672                        /* Skip completed partitions */
 673                        continue;
 674                }
 675
 676                dp = doms[nslot];
 677
 678                if (nslot == ndoms) {
 679                        static int warnings = 10;
 680                        if (warnings) {
 681                                printk(KERN_WARNING
 682                                 "rebuild_sched_domains confused:"
 683                                  " nslot %d, ndoms %d, csn %d, i %d,"
 684                                  " apn %d\n",
 685                                  nslot, ndoms, csn, i, apn);
 686                                warnings--;
 687                        }
 688                        continue;
 689                }
 690
 691                cpumask_clear(dp);
 692                if (dattr)
 693                        *(dattr + nslot) = SD_ATTR_INIT;
 694                for (j = i; j < csn; j++) {
 695                        struct cpuset *b = csa[j];
 696
 697                        if (apn == b->pn) {
 698                                cpumask_or(dp, dp, b->cpus_allowed);
 699                                if (dattr)
 700                                        update_domain_attr_tree(dattr + nslot, b);
 701
 702                                /* Done with this partition */
 703                                b->pn = -1;
 704                        }
 705                }
 706                nslot++;
 707        }
 708        BUG_ON(nslot != ndoms);
 709
 710done:
 711        kfree(csa);
 712
 713        /*
 714         * Fallback to the default domain if kmalloc() failed.
 715         * See comments in partition_sched_domains().
 716         */
 717        if (doms == NULL)
 718                ndoms = 1;
 719
 720        *domains    = doms;
 721        *attributes = dattr;
 722        return ndoms;
 723}
 724
 725/*
 726 * Rebuild scheduler domains.
 727 *
 728 * Call with neither cgroup_mutex held nor within get_online_cpus().
 729 * Takes both cgroup_mutex and get_online_cpus().
 730 *
 731 * Cannot be directly called from cpuset code handling changes
 732 * to the cpuset pseudo-filesystem, because it cannot be called
 733 * from code that already holds cgroup_mutex.
 734 */
 735static void do_rebuild_sched_domains(struct work_struct *unused)
 736{
 737        struct sched_domain_attr *attr;
 738        cpumask_var_t *doms;
 739        int ndoms;
 740
 741        get_online_cpus();
 742
 743        /* Generate domain masks and attrs */
 744        cgroup_lock();
 745        ndoms = generate_sched_domains(&doms, &attr);
 746        cgroup_unlock();
 747
 748        /* Have scheduler rebuild the domains */
 749        partition_sched_domains(ndoms, doms, attr);
 750
 751        put_online_cpus();
 752}
 753#else /* !CONFIG_SMP */
 754static void do_rebuild_sched_domains(struct work_struct *unused)
 755{
 756}
 757
 758static int generate_sched_domains(cpumask_var_t **domains,
 759                        struct sched_domain_attr **attributes)
 760{
 761        *domains = NULL;
 762        return 1;
 763}
 764#endif /* CONFIG_SMP */
 765
 766static DECLARE_WORK(rebuild_sched_domains_work, do_rebuild_sched_domains);
 767
 768/*
 769 * Rebuild scheduler domains, asynchronously via workqueue.
 770 *
 771 * If the flag 'sched_load_balance' of any cpuset with non-empty
 772 * 'cpus' changes, or if the 'cpus' allowed changes in any cpuset
 773 * which has that flag enabled, or if any cpuset with a non-empty
 774 * 'cpus' is removed, then call this routine to rebuild the
 775 * scheduler's dynamic sched domains.
 776 *
 777 * The rebuild_sched_domains() and partition_sched_domains()
 778 * routines must nest cgroup_lock() inside get_online_cpus(),
 779 * but such cpuset changes as these must nest that locking the
 780 * other way, holding cgroup_lock() for much of the code.
 781 *
 782 * So in order to avoid an ABBA deadlock, the cpuset code handling
 783 * these user changes delegates the actual sched domain rebuilding
 784 * to a separate workqueue thread, which ends up processing the
 785 * above do_rebuild_sched_domains() function.
 786 */
 787static void async_rebuild_sched_domains(void)
 788{
 789        queue_work(cpuset_wq, &rebuild_sched_domains_work);
 790}
 791
 792/*
 793 * Accomplishes the same scheduler domain rebuild as the above
 794 * async_rebuild_sched_domains(), however it directly calls the
 795 * rebuild routine synchronously rather than calling it via an
 796 * asynchronous work thread.
 797 *
 798 * This can only be called from code that is not holding
 799 * cgroup_mutex (not nested in a cgroup_lock() call.)
 800 */
 801void rebuild_sched_domains(void)
 802{
 803        do_rebuild_sched_domains(NULL);
 804}
 805
 806/**
 807 * cpuset_test_cpumask - test a task's cpus_allowed versus its cpuset's
 808 * @tsk: task to test
 809 * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner
 810 *
 811 * Call with cgroup_mutex held.  May take callback_mutex during call.
 812 * Called for each task in a cgroup by cgroup_scan_tasks().
 813 * Return nonzero if this tasks's cpus_allowed mask should be changed (in other
 814 * words, if its mask is not equal to its cpuset's mask).
 815 */
 816static int cpuset_test_cpumask(struct task_struct *tsk,
 817                               struct cgroup_scanner *scan)
 818{
 819        return !cpumask_equal(&tsk->cpus_allowed,
 820                        (cgroup_cs(scan->cg))->cpus_allowed);
 821}
 822
 823/**
 824 * cpuset_change_cpumask - make a task's cpus_allowed the same as its cpuset's
 825 * @tsk: task to test
 826 * @scan: struct cgroup_scanner containing the cgroup of the task
 827 *
 828 * Called by cgroup_scan_tasks() for each task in a cgroup whose
 829 * cpus_allowed mask needs to be changed.
 830 *
 831 * We don't need to re-check for the cgroup/cpuset membership, since we're
 832 * holding cgroup_lock() at this point.
 833 */
 834static void cpuset_change_cpumask(struct task_struct *tsk,
 835                                  struct cgroup_scanner *scan)
 836{
 837        set_cpus_allowed_ptr(tsk, ((cgroup_cs(scan->cg))->cpus_allowed));
 838}
 839
 840/**
 841 * update_tasks_cpumask - Update the cpumasks of tasks in the cpuset.
 842 * @cs: the cpuset in which each task's cpus_allowed mask needs to be changed
 843 * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks()
 844 *
 845 * Called with cgroup_mutex held
 846 *
 847 * The cgroup_scan_tasks() function will scan all the tasks in a cgroup,
 848 * calling callback functions for each.
 849 *
 850 * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0
 851 * if @heap != NULL.
 852 */
 853static void update_tasks_cpumask(struct cpuset *cs, struct ptr_heap *heap)
 854{
 855        struct cgroup_scanner scan;
 856
 857        scan.cg = cs->css.cgroup;
 858        scan.test_task = cpuset_test_cpumask;
 859        scan.process_task = cpuset_change_cpumask;
 860        scan.heap = heap;
 861        cgroup_scan_tasks(&scan);
 862}
 863
 864/**
 865 * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it
 866 * @cs: the cpuset to consider
 867 * @buf: buffer of cpu numbers written to this cpuset
 868 */
 869static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
 870                          const char *buf)
 871{
 872        struct ptr_heap heap;
 873        int retval;
 874        int is_load_balanced;
 875
 876        /* top_cpuset.cpus_allowed tracks cpu_online_mask; it's read-only */
 877        if (cs == &top_cpuset)
 878                return -EACCES;
 879
 880        /*
 881         * An empty cpus_allowed is ok only if the cpuset has no tasks.
 882         * Since cpulist_parse() fails on an empty mask, we special case
 883         * that parsing.  The validate_change() call ensures that cpusets
 884         * with tasks have cpus.
 885         */
 886        if (!*buf) {
 887                cpumask_clear(trialcs->cpus_allowed);
 888        } else {
 889                retval = cpulist_parse(buf, trialcs->cpus_allowed);
 890                if (retval < 0)
 891                        return retval;
 892
 893                if (!cpumask_subset(trialcs->cpus_allowed, cpu_active_mask))
 894                        return -EINVAL;
 895        }
 896        retval = validate_change(cs, trialcs);
 897        if (retval < 0)
 898                return retval;
 899
 900        /* Nothing to do if the cpus didn't change */
 901        if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed))
 902                return 0;
 903
 904        retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL);
 905        if (retval)
 906                return retval;
 907
 908        is_load_balanced = is_sched_load_balance(trialcs);
 909
 910        mutex_lock(&callback_mutex);
 911        cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed);
 912        mutex_unlock(&callback_mutex);
 913
 914        /*
 915         * Scan tasks in the cpuset, and update the cpumasks of any
 916         * that need an update.
 917         */
 918        update_tasks_cpumask(cs, &heap);
 919
 920        heap_free(&heap);
 921
 922        if (is_load_balanced)
 923                async_rebuild_sched_domains();
 924        return 0;
 925}
 926
 927/*
 928 * cpuset_migrate_mm
 929 *
 930 *    Migrate memory region from one set of nodes to another.
 931 *
 932 *    Temporarilly set tasks mems_allowed to target nodes of migration,
 933 *    so that the migration code can allocate pages on these nodes.
 934 *
 935 *    Call holding cgroup_mutex, so current's cpuset won't change
 936 *    during this call, as manage_mutex holds off any cpuset_attach()
 937 *    calls.  Therefore we don't need to take task_lock around the
 938 *    call to guarantee_online_mems(), as we know no one is changing
 939 *    our task's cpuset.
 940 *
 941 *    While the mm_struct we are migrating is typically from some
 942 *    other task, the task_struct mems_allowed that we are hacking
 943 *    is for our current task, which must allocate new pages for that
 944 *    migrating memory region.
 945 */
 946
 947static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from,
 948                                                        const nodemask_t *to)
 949{
 950        struct task_struct *tsk = current;
 951
 952        tsk->mems_allowed = *to;
 953
 954        do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL);
 955
 956        guarantee_online_mems(task_cs(tsk),&tsk->mems_allowed);
 957}
 958
 959/*
 960 * cpuset_change_task_nodemask - change task's mems_allowed and mempolicy
 961 * @tsk: the task to change
 962 * @newmems: new nodes that the task will be set
 963 *
 964 * In order to avoid seeing no nodes if the old and new nodes are disjoint,
 965 * we structure updates as setting all new allowed nodes, then clearing newly
 966 * disallowed ones.
 967 */
 968static void cpuset_change_task_nodemask(struct task_struct *tsk,
 969                                        nodemask_t *newmems)
 970{
 971        bool need_loop;
 972
 973        /*
 974         * Allow tasks that have access to memory reserves because they have
 975         * been OOM killed to get memory anywhere.
 976         */
 977        if (unlikely(test_thread_flag(TIF_MEMDIE)))
 978                return;
 979        if (current->flags & PF_EXITING) /* Let dying task have memory */
 980                return;
 981
 982        task_lock(tsk);
 983        /*
 984         * Determine if a loop is necessary if another thread is doing
 985         * get_mems_allowed().  If at least one node remains unchanged and
 986         * tsk does not have a mempolicy, then an empty nodemask will not be
 987         * possible when mems_allowed is larger than a word.
 988         */
 989        need_loop = task_has_mempolicy(tsk) ||
 990                        !nodes_intersects(*newmems, tsk->mems_allowed);
 991
 992        if (need_loop)
 993                write_seqcount_begin(&tsk->mems_allowed_seq);
 994
 995        nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems);
 996        mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP1);
 997
 998        mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP2);
 999        tsk->mems_allowed = *newmems;
1000
1001        if (need_loop)
1002                write_seqcount_end(&tsk->mems_allowed_seq);
1003
1004        task_unlock(tsk);
1005}
1006
1007/*
1008 * Update task's mems_allowed and rebind its mempolicy and vmas' mempolicy
1009 * of it to cpuset's new mems_allowed, and migrate pages to new nodes if
1010 * memory_migrate flag is set. Called with cgroup_mutex held.
1011 */
1012static void cpuset_change_nodemask(struct task_struct *p,
1013                                   struct cgroup_scanner *scan)
1014{
1015        struct mm_struct *mm;
1016        struct cpuset *cs;
1017        int migrate;
1018        const nodemask_t *oldmem = scan->data;
1019        static nodemask_t newmems;      /* protected by cgroup_mutex */
1020
1021        cs = cgroup_cs(scan->cg);
1022        guarantee_online_mems(cs, &newmems);
1023
1024        cpuset_change_task_nodemask(p, &newmems);
1025
1026        mm = get_task_mm(p);
1027        if (!mm)
1028                return;
1029
1030        migrate = is_memory_migrate(cs);
1031
1032        mpol_rebind_mm(mm, &cs->mems_allowed);
1033        if (migrate)
1034                cpuset_migrate_mm(mm, oldmem, &cs->mems_allowed);
1035        mmput(mm);
1036}
1037
1038static void *cpuset_being_rebound;
1039
1040/**
1041 * update_tasks_nodemask - Update the nodemasks of tasks in the cpuset.
1042 * @cs: the cpuset in which each task's mems_allowed mask needs to be changed
1043 * @oldmem: old mems_allowed of cpuset cs
1044 * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks()
1045 *
1046 * Called with cgroup_mutex held
1047 * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0
1048 * if @heap != NULL.
1049 */
1050static void update_tasks_nodemask(struct cpuset *cs, const nodemask_t *oldmem,
1051                                 struct ptr_heap *heap)
1052{
1053        struct cgroup_scanner scan;
1054
1055        cpuset_being_rebound = cs;              /* causes mpol_dup() rebind */
1056
1057        scan.cg = cs->css.cgroup;
1058        scan.test_task = NULL;
1059        scan.process_task = cpuset_change_nodemask;
1060        scan.heap = heap;
1061        scan.data = (nodemask_t *)oldmem;
1062
1063        /*
1064         * The mpol_rebind_mm() call takes mmap_sem, which we couldn't
1065         * take while holding tasklist_lock.  Forks can happen - the
1066         * mpol_dup() cpuset_being_rebound check will catch such forks,
1067         * and rebind their vma mempolicies too.  Because we still hold
1068         * the global cgroup_mutex, we know that no other rebind effort
1069         * will be contending for the global variable cpuset_being_rebound.
1070         * It's ok if we rebind the same mm twice; mpol_rebind_mm()
1071         * is idempotent.  Also migrate pages in each mm to new nodes.
1072         */
1073        cgroup_scan_tasks(&scan);
1074
1075        /* We're done rebinding vmas to this cpuset's new mems_allowed. */
1076        cpuset_being_rebound = NULL;
1077}
1078
1079/*
1080 * Handle user request to change the 'mems' memory placement
1081 * of a cpuset.  Needs to validate the request, update the
1082 * cpusets mems_allowed, and for each task in the cpuset,
1083 * update mems_allowed and rebind task's mempolicy and any vma
1084 * mempolicies and if the cpuset is marked 'memory_migrate',
1085 * migrate the tasks pages to the new memory.
1086 *
1087 * Call with cgroup_mutex held.  May take callback_mutex during call.
1088 * Will take tasklist_lock, scan tasklist for tasks in cpuset cs,
1089 * lock each such tasks mm->mmap_sem, scan its vma's and rebind
1090 * their mempolicies to the cpusets new mems_allowed.
1091 */
1092static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs,
1093                           const char *buf)
1094{
1095        NODEMASK_ALLOC(nodemask_t, oldmem, GFP_KERNEL);
1096        int retval;
1097        struct ptr_heap heap;
1098
1099        if (!oldmem)
1100                return -ENOMEM;
1101
1102        /*
1103         * top_cpuset.mems_allowed tracks node_stats[N_MEMORY];
1104         * it's read-only
1105         */
1106        if (cs == &top_cpuset) {
1107                retval = -EACCES;
1108                goto done;
1109        }
1110
1111        /*
1112         * An empty mems_allowed is ok iff there are no tasks in the cpuset.
1113         * Since nodelist_parse() fails on an empty mask, we special case
1114         * that parsing.  The validate_change() call ensures that cpusets
1115         * with tasks have memory.
1116         */
1117        if (!*buf) {
1118                nodes_clear(trialcs->mems_allowed);
1119        } else {
1120                retval = nodelist_parse(buf, trialcs->mems_allowed);
1121                if (retval < 0)
1122                        goto done;
1123
1124                if (!nodes_subset(trialcs->mems_allowed,
1125                                node_states[N_MEMORY])) {
1126                        retval =  -EINVAL;
1127                        goto done;
1128                }
1129        }
1130        *oldmem = cs->mems_allowed;
1131        if (nodes_equal(*oldmem, trialcs->mems_allowed)) {
1132                retval = 0;             /* Too easy - nothing to do */
1133                goto done;
1134        }
1135        retval = validate_change(cs, trialcs);
1136        if (retval < 0)
1137                goto done;
1138
1139        retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL);
1140        if (retval < 0)
1141                goto done;
1142
1143        mutex_lock(&callback_mutex);
1144        cs->mems_allowed = trialcs->mems_allowed;
1145        mutex_unlock(&callback_mutex);
1146
1147        update_tasks_nodemask(cs, oldmem, &heap);
1148
1149        heap_free(&heap);
1150done:
1151        NODEMASK_FREE(oldmem);
1152        return retval;
1153}
1154
1155int current_cpuset_is_being_rebound(void)
1156{
1157        return task_cs(current) == cpuset_being_rebound;
1158}
1159
1160static int update_relax_domain_level(struct cpuset *cs, s64 val)
1161{
1162#ifdef CONFIG_SMP
1163        if (val < -1 || val >= sched_domain_level_max)
1164                return -EINVAL;
1165#endif
1166
1167        if (val != cs->relax_domain_level) {
1168                cs->relax_domain_level = val;
1169                if (!cpumask_empty(cs->cpus_allowed) &&
1170                    is_sched_load_balance(cs))
1171                        async_rebuild_sched_domains();
1172        }
1173
1174        return 0;
1175}
1176
1177/*
1178 * cpuset_change_flag - make a task's spread flags the same as its cpuset's
1179 * @tsk: task to be updated
1180 * @scan: struct cgroup_scanner containing the cgroup of the task
1181 *
1182 * Called by cgroup_scan_tasks() for each task in a cgroup.
1183 *
1184 * We don't need to re-check for the cgroup/cpuset membership, since we're
1185 * holding cgroup_lock() at this point.
1186 */
1187static void cpuset_change_flag(struct task_struct *tsk,
1188                                struct cgroup_scanner *scan)
1189{
1190        cpuset_update_task_spread_flag(cgroup_cs(scan->cg), tsk);
1191}
1192
1193/*
1194 * update_tasks_flags - update the spread flags of tasks in the cpuset.
1195 * @cs: the cpuset in which each task's spread flags needs to be changed
1196 * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks()
1197 *
1198 * Called with cgroup_mutex held
1199 *
1200 * The cgroup_scan_tasks() function will scan all the tasks in a cgroup,
1201 * calling callback functions for each.
1202 *
1203 * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0
1204 * if @heap != NULL.
1205 */
1206static void update_tasks_flags(struct cpuset *cs, struct ptr_heap *heap)
1207{
1208        struct cgroup_scanner scan;
1209
1210        scan.cg = cs->css.cgroup;
1211        scan.test_task = NULL;
1212        scan.process_task = cpuset_change_flag;
1213        scan.heap = heap;
1214        cgroup_scan_tasks(&scan);
1215}
1216
1217/*
1218 * update_flag - read a 0 or a 1 in a file and update associated flag
1219 * bit:         the bit to update (see cpuset_flagbits_t)
1220 * cs:          the cpuset to update
1221 * turning_on:  whether the flag is being set or cleared
1222 *
1223 * Call with cgroup_mutex held.
1224 */
1225
1226static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs,
1227                       int turning_on)
1228{
1229        struct cpuset *trialcs;
1230        int balance_flag_changed;
1231        int spread_flag_changed;
1232        struct ptr_heap heap;
1233        int err;
1234
1235        trialcs = alloc_trial_cpuset(cs);
1236        if (!trialcs)
1237                return -ENOMEM;
1238
1239        if (turning_on)
1240                set_bit(bit, &trialcs->flags);
1241        else
1242                clear_bit(bit, &trialcs->flags);
1243
1244        err = validate_change(cs, trialcs);
1245        if (err < 0)
1246                goto out;
1247
1248        err = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL);
1249        if (err < 0)
1250                goto out;
1251
1252        balance_flag_changed = (is_sched_load_balance(cs) !=
1253                                is_sched_load_balance(trialcs));
1254
1255        spread_flag_changed = ((is_spread_slab(cs) != is_spread_slab(trialcs))
1256                        || (is_spread_page(cs) != is_spread_page(trialcs)));
1257
1258        mutex_lock(&callback_mutex);
1259        cs->flags = trialcs->flags;
1260        mutex_unlock(&callback_mutex);
1261
1262        if (!cpumask_empty(trialcs->cpus_allowed) && balance_flag_changed)
1263                async_rebuild_sched_domains();
1264
1265        if (spread_flag_changed)
1266                update_tasks_flags(cs, &heap);
1267        heap_free(&heap);
1268out:
1269        free_trial_cpuset(trialcs);
1270        return err;
1271}
1272
1273/*
1274 * Frequency meter - How fast is some event occurring?
1275 *
1276 * These routines manage a digitally filtered, constant time based,
1277 * event frequency meter.  There are four routines:
1278 *   fmeter_init() - initialize a frequency meter.
1279 *   fmeter_markevent() - called each time the event happens.
1280 *   fmeter_getrate() - returns the recent rate of such events.
1281 *   fmeter_update() - internal routine used to update fmeter.
1282 *
1283 * A common data structure is passed to each of these routines,
1284 * which is used to keep track of the state required to manage the
1285 * frequency meter and its digital filter.
1286 *
1287 * The filter works on the number of events marked per unit time.
1288 * The filter is single-pole low-pass recursive (IIR).  The time unit
1289 * is 1 second.  Arithmetic is done using 32-bit integers scaled to
1290 * simulate 3 decimal digits of precision (multiplied by 1000).
1291 *
1292 * With an FM_COEF of 933, and a time base of 1 second, the filter
1293 * has a half-life of 10 seconds, meaning that if the events quit
1294 * happening, then the rate returned from the fmeter_getrate()
1295 * will be cut in half each 10 seconds, until it converges to zero.
1296 *
1297 * It is not worth doing a real infinitely recursive filter.  If more
1298 * than FM_MAXTICKS ticks have elapsed since the last filter event,
1299 * just compute FM_MAXTICKS ticks worth, by which point the level
1300 * will be stable.
1301 *
1302 * Limit the count of unprocessed events to FM_MAXCNT, so as to avoid
1303 * arithmetic overflow in the fmeter_update() routine.
1304 *
1305 * Given the simple 32 bit integer arithmetic used, this meter works
1306 * best for reporting rates between one per millisecond (msec) and
1307 * one per 32 (approx) seconds.  At constant rates faster than one
1308 * per msec it maxes out at values just under 1,000,000.  At constant
1309 * rates between one per msec, and one per second it will stabilize
1310 * to a value N*1000, where N is the rate of events per second.
1311 * At constant rates between one per second and one per 32 seconds,
1312 * it will be choppy, moving up on the seconds that have an event,
1313 * and then decaying until the next event.  At rates slower than
1314 * about one in 32 seconds, it decays all the way back to zero between
1315 * each event.
1316 */
1317
1318#define FM_COEF 933             /* coefficient for half-life of 10 secs */
1319#define FM_MAXTICKS ((time_t)99) /* useless computing more ticks than this */
1320#define FM_MAXCNT 1000000       /* limit cnt to avoid overflow */
1321#define FM_SCALE 1000           /* faux fixed point scale */
1322
1323/* Initialize a frequency meter */
1324static void fmeter_init(struct fmeter *fmp)
1325{
1326        fmp->cnt = 0;
1327        fmp->val = 0;
1328        fmp->time = 0;
1329        spin_lock_init(&fmp->lock);
1330}
1331
1332/* Internal meter update - process cnt events and update value */
1333static void fmeter_update(struct fmeter *fmp)
1334{
1335        time_t now = get_seconds();
1336        time_t ticks = now - fmp->time;
1337
1338        if (ticks == 0)
1339                return;
1340
1341        ticks = min(FM_MAXTICKS, ticks);
1342        while (ticks-- > 0)
1343                fmp->val = (FM_COEF * fmp->val) / FM_SCALE;
1344        fmp->time = now;
1345
1346        fmp->val += ((FM_SCALE - FM_COEF) * fmp->cnt) / FM_SCALE;
1347        fmp->cnt = 0;
1348}
1349
1350/* Process any previous ticks, then bump cnt by one (times scale). */
1351static void fmeter_markevent(struct fmeter *fmp)
1352{
1353        spin_lock(&fmp->lock);
1354        fmeter_update(fmp);
1355        fmp->cnt = min(FM_MAXCNT, fmp->cnt + FM_SCALE);
1356        spin_unlock(&fmp->lock);
1357}
1358
1359/* Process any previous ticks, then return current value. */
1360static int fmeter_getrate(struct fmeter *fmp)
1361{
1362        int val;
1363
1364        spin_lock(&fmp->lock);
1365        fmeter_update(fmp);
1366        val = fmp->val;
1367        spin_unlock(&fmp->lock);
1368        return val;
1369}
1370
1371/*
1372 * Protected by cgroup_lock. The nodemasks must be stored globally because
1373 * dynamically allocating them is not allowed in can_attach, and they must
1374 * persist until attach.
1375 */
1376static cpumask_var_t cpus_attach;
1377static nodemask_t cpuset_attach_nodemask_from;
1378static nodemask_t cpuset_attach_nodemask_to;
1379
1380/* Called by cgroups to determine if a cpuset is usable; cgroup_mutex held */
1381static int cpuset_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
1382{
1383        struct cpuset *cs = cgroup_cs(cgrp);
1384        struct task_struct *task;
1385        int ret;
1386
1387        if (cpumask_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed))
1388                return -ENOSPC;
1389
1390        cgroup_taskset_for_each(task, cgrp, tset) {
1391                /*
1392                 * Kthreads bound to specific cpus cannot be moved to a new
1393                 * cpuset; we cannot change their cpu affinity and
1394                 * isolating such threads by their set of allowed nodes is
1395                 * unnecessary.  Thus, cpusets are not applicable for such
1396                 * threads.  This prevents checking for success of
1397                 * set_cpus_allowed_ptr() on all attached tasks before
1398                 * cpus_allowed may be changed.
1399                 */
1400                if (task->flags & PF_THREAD_BOUND)
1401                        return -EINVAL;
1402                if ((ret = security_task_setscheduler(task)))
1403                        return ret;
1404        }
1405
1406        /* prepare for attach */
1407        if (cs == &top_cpuset)
1408                cpumask_copy(cpus_attach, cpu_possible_mask);
1409        else
1410                guarantee_online_cpus(cs, cpus_attach);
1411
1412        guarantee_online_mems(cs, &cpuset_attach_nodemask_to);
1413
1414        return 0;
1415}
1416
1417static void cpuset_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
1418{
1419        struct mm_struct *mm;
1420        struct task_struct *task;
1421        struct task_struct *leader = cgroup_taskset_first(tset);
1422        struct cgroup *oldcgrp = cgroup_taskset_cur_cgroup(tset);
1423        struct cpuset *cs = cgroup_cs(cgrp);
1424        struct cpuset *oldcs = cgroup_cs(oldcgrp);
1425
1426        cgroup_taskset_for_each(task, cgrp, tset) {
1427                /*
1428                 * can_attach beforehand should guarantee that this doesn't
1429                 * fail.  TODO: have a better way to handle failure here
1430                 */
1431                WARN_ON_ONCE(set_cpus_allowed_ptr(task, cpus_attach));
1432
1433                cpuset_change_task_nodemask(task, &cpuset_attach_nodemask_to);
1434                cpuset_update_task_spread_flag(cs, task);
1435        }
1436
1437        /*
1438         * Change mm, possibly for multiple threads in a threadgroup. This is
1439         * expensive and may sleep.
1440         */
1441        cpuset_attach_nodemask_from = oldcs->mems_allowed;
1442        cpuset_attach_nodemask_to = cs->mems_allowed;
1443        mm = get_task_mm(leader);
1444        if (mm) {
1445                mpol_rebind_mm(mm, &cpuset_attach_nodemask_to);
1446                if (is_memory_migrate(cs))
1447                        cpuset_migrate_mm(mm, &cpuset_attach_nodemask_from,
1448                                          &cpuset_attach_nodemask_to);
1449                mmput(mm);
1450        }
1451}
1452
1453/* The various types of files and directories in a cpuset file system */
1454
1455typedef enum {
1456        FILE_MEMORY_MIGRATE,
1457        FILE_CPULIST,
1458        FILE_MEMLIST,
1459        FILE_CPU_EXCLUSIVE,
1460        FILE_MEM_EXCLUSIVE,
1461        FILE_MEM_HARDWALL,
1462        FILE_SCHED_LOAD_BALANCE,
1463        FILE_SCHED_RELAX_DOMAIN_LEVEL,
1464        FILE_MEMORY_PRESSURE_ENABLED,
1465        FILE_MEMORY_PRESSURE,
1466        FILE_SPREAD_PAGE,
1467        FILE_SPREAD_SLAB,
1468} cpuset_filetype_t;
1469
1470static int cpuset_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val)
1471{
1472        int retval = 0;
1473        struct cpuset *cs = cgroup_cs(cgrp);
1474        cpuset_filetype_t type = cft->private;
1475
1476        if (!cgroup_lock_live_group(cgrp))
1477                return -ENODEV;
1478
1479        switch (type) {
1480        case FILE_CPU_EXCLUSIVE:
1481                retval = update_flag(CS_CPU_EXCLUSIVE, cs, val);
1482                break;
1483        case FILE_MEM_EXCLUSIVE:
1484                retval = update_flag(CS_MEM_EXCLUSIVE, cs, val);
1485                break;
1486        case FILE_MEM_HARDWALL:
1487                retval = update_flag(CS_MEM_HARDWALL, cs, val);
1488                break;
1489        case FILE_SCHED_LOAD_BALANCE:
1490                retval = update_flag(CS_SCHED_LOAD_BALANCE, cs, val);
1491                break;
1492        case FILE_MEMORY_MIGRATE:
1493                retval = update_flag(CS_MEMORY_MIGRATE, cs, val);
1494                break;
1495        case FILE_MEMORY_PRESSURE_ENABLED:
1496                cpuset_memory_pressure_enabled = !!val;
1497                break;
1498        case FILE_MEMORY_PRESSURE:
1499                retval = -EACCES;
1500                break;
1501        case FILE_SPREAD_PAGE:
1502                retval = update_flag(CS_SPREAD_PAGE, cs, val);
1503                break;
1504        case FILE_SPREAD_SLAB:
1505                retval = update_flag(CS_SPREAD_SLAB, cs, val);
1506                break;
1507        default:
1508                retval = -EINVAL;
1509                break;
1510        }
1511        cgroup_unlock();
1512        return retval;
1513}
1514
1515static int cpuset_write_s64(struct cgroup *cgrp, struct cftype *cft, s64 val)
1516{
1517        int retval = 0;
1518        struct cpuset *cs = cgroup_cs(cgrp);
1519        cpuset_filetype_t type = cft->private;
1520
1521        if (!cgroup_lock_live_group(cgrp))
1522                return -ENODEV;
1523
1524        switch (type) {
1525        case FILE_SCHED_RELAX_DOMAIN_LEVEL:
1526                retval = update_relax_domain_level(cs, val);
1527                break;
1528        default:
1529                retval = -EINVAL;
1530                break;
1531        }
1532        cgroup_unlock();
1533        return retval;
1534}
1535
1536/*
1537 * Common handling for a write to a "cpus" or "mems" file.
1538 */
1539static int cpuset_write_resmask(struct cgroup *cgrp, struct cftype *cft,
1540                                const char *buf)
1541{
1542        int retval = 0;
1543        struct cpuset *cs = cgroup_cs(cgrp);
1544        struct cpuset *trialcs;
1545
1546        if (!cgroup_lock_live_group(cgrp))
1547                return -ENODEV;
1548
1549        trialcs = alloc_trial_cpuset(cs);
1550        if (!trialcs) {
1551                retval = -ENOMEM;
1552                goto out;
1553        }
1554
1555        switch (cft->private) {
1556        case FILE_CPULIST:
1557                retval = update_cpumask(cs, trialcs, buf);
1558                break;
1559        case FILE_MEMLIST:
1560                retval = update_nodemask(cs, trialcs, buf);
1561                break;
1562        default:
1563                retval = -EINVAL;
1564                break;
1565        }
1566
1567        free_trial_cpuset(trialcs);
1568out:
1569        cgroup_unlock();
1570        return retval;
1571}
1572
1573/*
1574 * These ascii lists should be read in a single call, by using a user
1575 * buffer large enough to hold the entire map.  If read in smaller
1576 * chunks, there is no guarantee of atomicity.  Since the display format
1577 * used, list of ranges of sequential numbers, is variable length,
1578 * and since these maps can change value dynamically, one could read
1579 * gibberish by doing partial reads while a list was changing.
1580 * A single large read to a buffer that crosses a page boundary is
1581 * ok, because the result being copied to user land is not recomputed
1582 * across a page fault.
1583 */
1584
1585static size_t cpuset_sprintf_cpulist(char *page, struct cpuset *cs)
1586{
1587        size_t count;
1588
1589        mutex_lock(&callback_mutex);
1590        count = cpulist_scnprintf(page, PAGE_SIZE, cs->cpus_allowed);
1591        mutex_unlock(&callback_mutex);
1592
1593        return count;
1594}
1595
1596static size_t cpuset_sprintf_memlist(char *page, struct cpuset *cs)
1597{
1598        size_t count;
1599
1600        mutex_lock(&callback_mutex);
1601        count = nodelist_scnprintf(page, PAGE_SIZE, cs->mems_allowed);
1602        mutex_unlock(&callback_mutex);
1603
1604        return count;
1605}
1606
1607static ssize_t cpuset_common_file_read(struct cgroup *cont,
1608                                       struct cftype *cft,
1609                                       struct file *file,
1610                                       char __user *buf,
1611                                       size_t nbytes, loff_t *ppos)
1612{
1613        struct cpuset *cs = cgroup_cs(cont);
1614        cpuset_filetype_t type = cft->private;
1615        char *page;
1616        ssize_t retval = 0;
1617        char *s;
1618
1619        if (!(page = (char *)__get_free_page(GFP_TEMPORARY)))
1620                return -ENOMEM;
1621
1622        s = page;
1623
1624        switch (type) {
1625        case FILE_CPULIST:
1626                s += cpuset_sprintf_cpulist(s, cs);
1627                break;
1628        case FILE_MEMLIST:
1629                s += cpuset_sprintf_memlist(s, cs);
1630                break;
1631        default:
1632                retval = -EINVAL;
1633                goto out;
1634        }
1635        *s++ = '\n';
1636
1637        retval = simple_read_from_buffer(buf, nbytes, ppos, page, s - page);
1638out:
1639        free_page((unsigned long)page);
1640        return retval;
1641}
1642
1643static u64 cpuset_read_u64(struct cgroup *cont, struct cftype *cft)
1644{
1645        struct cpuset *cs = cgroup_cs(cont);
1646        cpuset_filetype_t type = cft->private;
1647        switch (type) {
1648        case FILE_CPU_EXCLUSIVE:
1649                return is_cpu_exclusive(cs);
1650        case FILE_MEM_EXCLUSIVE:
1651                return is_mem_exclusive(cs);
1652        case FILE_MEM_HARDWALL:
1653                return is_mem_hardwall(cs);
1654        case FILE_SCHED_LOAD_BALANCE:
1655                return is_sched_load_balance(cs);
1656        case FILE_MEMORY_MIGRATE:
1657                return is_memory_migrate(cs);
1658        case FILE_MEMORY_PRESSURE_ENABLED:
1659                return cpuset_memory_pressure_enabled;
1660        case FILE_MEMORY_PRESSURE:
1661                return fmeter_getrate(&cs->fmeter);
1662        case FILE_SPREAD_PAGE:
1663                return is_spread_page(cs);
1664        case FILE_SPREAD_SLAB:
1665                return is_spread_slab(cs);
1666        default:
1667                BUG();
1668        }
1669
1670        /* Unreachable but makes gcc happy */
1671        return 0;
1672}
1673
1674static s64 cpuset_read_s64(struct cgroup *cont, struct cftype *cft)
1675{
1676        struct cpuset *cs = cgroup_cs(cont);
1677        cpuset_filetype_t type = cft->private;
1678        switch (type) {
1679        case FILE_SCHED_RELAX_DOMAIN_LEVEL:
1680                return cs->relax_domain_level;
1681        default:
1682                BUG();
1683        }
1684
1685        /* Unrechable but makes gcc happy */
1686        return 0;
1687}
1688
1689
1690/*
1691 * for the common functions, 'private' gives the type of file
1692 */
1693
1694static struct cftype files[] = {
1695        {
1696                .name = "cpus",
1697                .read = cpuset_common_file_read,
1698                .write_string = cpuset_write_resmask,
1699                .max_write_len = (100U + 6 * NR_CPUS),
1700                .private = FILE_CPULIST,
1701        },
1702
1703        {
1704                .name = "mems",
1705                .read = cpuset_common_file_read,
1706                .write_string = cpuset_write_resmask,
1707                .max_write_len = (100U + 6 * MAX_NUMNODES),
1708                .private = FILE_MEMLIST,
1709        },
1710
1711        {
1712                .name = "cpu_exclusive",
1713                .read_u64 = cpuset_read_u64,
1714                .write_u64 = cpuset_write_u64,
1715                .private = FILE_CPU_EXCLUSIVE,
1716        },
1717
1718        {
1719                .name = "mem_exclusive",
1720                .read_u64 = cpuset_read_u64,
1721                .write_u64 = cpuset_write_u64,
1722                .private = FILE_MEM_EXCLUSIVE,
1723        },
1724
1725        {
1726                .name = "mem_hardwall",
1727                .read_u64 = cpuset_read_u64,
1728                .write_u64 = cpuset_write_u64,
1729                .private = FILE_MEM_HARDWALL,
1730        },
1731
1732        {
1733                .name = "sched_load_balance",
1734                .read_u64 = cpuset_read_u64,
1735                .write_u64 = cpuset_write_u64,
1736                .private = FILE_SCHED_LOAD_BALANCE,
1737        },
1738
1739        {
1740                .name = "sched_relax_domain_level",
1741                .read_s64 = cpuset_read_s64,
1742                .write_s64 = cpuset_write_s64,
1743                .private = FILE_SCHED_RELAX_DOMAIN_LEVEL,
1744        },
1745
1746        {
1747                .name = "memory_migrate",
1748                .read_u64 = cpuset_read_u64,
1749                .write_u64 = cpuset_write_u64,
1750                .private = FILE_MEMORY_MIGRATE,
1751        },
1752
1753        {
1754                .name = "memory_pressure",
1755                .read_u64 = cpuset_read_u64,
1756                .write_u64 = cpuset_write_u64,
1757                .private = FILE_MEMORY_PRESSURE,
1758                .mode = S_IRUGO,
1759        },
1760
1761        {
1762                .name = "memory_spread_page",
1763                .read_u64 = cpuset_read_u64,
1764                .write_u64 = cpuset_write_u64,
1765                .private = FILE_SPREAD_PAGE,
1766        },
1767
1768        {
1769                .name = "memory_spread_slab",
1770                .read_u64 = cpuset_read_u64,
1771                .write_u64 = cpuset_write_u64,
1772                .private = FILE_SPREAD_SLAB,
1773        },
1774
1775        {
1776                .name = "memory_pressure_enabled",
1777                .flags = CFTYPE_ONLY_ON_ROOT,
1778                .read_u64 = cpuset_read_u64,
1779                .write_u64 = cpuset_write_u64,
1780                .private = FILE_MEMORY_PRESSURE_ENABLED,
1781        },
1782
1783        { }     /* terminate */
1784};
1785
1786/*
1787 *      cpuset_css_alloc - allocate a cpuset css
1788 *      cont:   control group that the new cpuset will be part of
1789 */
1790
1791static struct cgroup_subsys_state *cpuset_css_alloc(struct cgroup *cont)
1792{
1793        struct cgroup *parent_cg = cont->parent;
1794        struct cgroup *tmp_cg;
1795        struct cpuset *parent, *cs;
1796
1797        if (!parent_cg)
1798                return &top_cpuset.css;
1799        parent = cgroup_cs(parent_cg);
1800
1801        cs = kmalloc(sizeof(*cs), GFP_KERNEL);
1802        if (!cs)
1803                return ERR_PTR(-ENOMEM);
1804        if (!alloc_cpumask_var(&cs->cpus_allowed, GFP_KERNEL)) {
1805                kfree(cs);
1806                return ERR_PTR(-ENOMEM);
1807        }
1808
1809        cs->flags = 0;
1810        if (is_spread_page(parent))
1811                set_bit(CS_SPREAD_PAGE, &cs->flags);
1812        if (is_spread_slab(parent))
1813                set_bit(CS_SPREAD_SLAB, &cs->flags);
1814        set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
1815        cpumask_clear(cs->cpus_allowed);
1816        nodes_clear(cs->mems_allowed);
1817        fmeter_init(&cs->fmeter);
1818        cs->relax_domain_level = -1;
1819
1820        cs->parent = parent;
1821        number_of_cpusets++;
1822
1823        if (!test_bit(CGRP_CPUSET_CLONE_CHILDREN, &cont->flags))
1824                goto skip_clone;
1825
1826        /*
1827         * Clone @parent's configuration if CGRP_CPUSET_CLONE_CHILDREN is
1828         * set.  This flag handling is implemented in cgroup core for
1829         * histrical reasons - the flag may be specified during mount.
1830         *
1831         * Currently, if any sibling cpusets have exclusive cpus or mem, we
1832         * refuse to clone the configuration - thereby refusing the task to
1833         * be entered, and as a result refusing the sys_unshare() or
1834         * clone() which initiated it.  If this becomes a problem for some
1835         * users who wish to allow that scenario, then this could be
1836         * changed to grant parent->cpus_allowed-sibling_cpus_exclusive
1837         * (and likewise for mems) to the new cgroup.
1838         */
1839        list_for_each_entry(tmp_cg, &parent_cg->children, sibling) {
1840                struct cpuset *tmp_cs = cgroup_cs(tmp_cg);
1841
1842                if (is_mem_exclusive(tmp_cs) || is_cpu_exclusive(tmp_cs))
1843                        goto skip_clone;
1844        }
1845
1846        mutex_lock(&callback_mutex);
1847        cs->mems_allowed = parent->mems_allowed;
1848        cpumask_copy(cs->cpus_allowed, parent->cpus_allowed);
1849        mutex_unlock(&callback_mutex);
1850skip_clone:
1851        return &cs->css;
1852}
1853
1854/*
1855 * If the cpuset being removed has its flag 'sched_load_balance'
1856 * enabled, then simulate turning sched_load_balance off, which
1857 * will call async_rebuild_sched_domains().
1858 */
1859
1860static void cpuset_css_free(struct cgroup *cont)
1861{
1862        struct cpuset *cs = cgroup_cs(cont);
1863
1864        if (is_sched_load_balance(cs))
1865                update_flag(CS_SCHED_LOAD_BALANCE, cs, 0);
1866
1867        number_of_cpusets--;
1868        free_cpumask_var(cs->cpus_allowed);
1869        kfree(cs);
1870}
1871
1872struct cgroup_subsys cpuset_subsys = {
1873        .name = "cpuset",
1874        .css_alloc = cpuset_css_alloc,
1875        .css_free = cpuset_css_free,
1876        .can_attach = cpuset_can_attach,
1877        .attach = cpuset_attach,
1878        .subsys_id = cpuset_subsys_id,
1879        .base_cftypes = files,
1880        .early_init = 1,
1881};
1882
1883/**
1884 * cpuset_init - initialize cpusets at system boot
1885 *
1886 * Description: Initialize top_cpuset and the cpuset internal file system,
1887 **/
1888
1889int __init cpuset_init(void)
1890{
1891        int err = 0;
1892
1893        if (!alloc_cpumask_var(&top_cpuset.cpus_allowed, GFP_KERNEL))
1894                BUG();
1895
1896        cpumask_setall(top_cpuset.cpus_allowed);
1897        nodes_setall(top_cpuset.mems_allowed);
1898
1899        fmeter_init(&top_cpuset.fmeter);
1900        set_bit(CS_SCHED_LOAD_BALANCE, &top_cpuset.flags);
1901        top_cpuset.relax_domain_level = -1;
1902
1903        err = register_filesystem(&cpuset_fs_type);
1904        if (err < 0)
1905                return err;
1906
1907        if (!alloc_cpumask_var(&cpus_attach, GFP_KERNEL))
1908                BUG();
1909
1910        number_of_cpusets = 1;
1911        return 0;
1912}
1913
1914/**
1915 * cpuset_do_move_task - move a given task to another cpuset
1916 * @tsk: pointer to task_struct the task to move
1917 * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner
1918 *
1919 * Called by cgroup_scan_tasks() for each task in a cgroup.
1920 * Return nonzero to stop the walk through the tasks.
1921 */
1922static void cpuset_do_move_task(struct task_struct *tsk,
1923                                struct cgroup_scanner *scan)
1924{
1925        struct cgroup *new_cgroup = scan->data;
1926
1927        cgroup_attach_task(new_cgroup, tsk);
1928}
1929
1930/**
1931 * move_member_tasks_to_cpuset - move tasks from one cpuset to another
1932 * @from: cpuset in which the tasks currently reside
1933 * @to: cpuset to which the tasks will be moved
1934 *
1935 * Called with cgroup_mutex held
1936 * callback_mutex must not be held, as cpuset_attach() will take it.
1937 *
1938 * The cgroup_scan_tasks() function will scan all the tasks in a cgroup,
1939 * calling callback functions for each.
1940 */
1941static void move_member_tasks_to_cpuset(struct cpuset *from, struct cpuset *to)
1942{
1943        struct cgroup_scanner scan;
1944
1945        scan.cg = from->css.cgroup;
1946        scan.test_task = NULL; /* select all tasks in cgroup */
1947        scan.process_task = cpuset_do_move_task;
1948        scan.heap = NULL;
1949        scan.data = to->css.cgroup;
1950
1951        if (cgroup_scan_tasks(&scan))
1952                printk(KERN_ERR "move_member_tasks_to_cpuset: "
1953                                "cgroup_scan_tasks failed\n");
1954}
1955
1956/*
1957 * If CPU and/or memory hotplug handlers, below, unplug any CPUs
1958 * or memory nodes, we need to walk over the cpuset hierarchy,
1959 * removing that CPU or node from all cpusets.  If this removes the
1960 * last CPU or node from a cpuset, then move the tasks in the empty
1961 * cpuset to its next-highest non-empty parent.
1962 *
1963 * Called with cgroup_mutex held
1964 * callback_mutex must not be held, as cpuset_attach() will take it.
1965 */
1966static void remove_tasks_in_empty_cpuset(struct cpuset *cs)
1967{
1968        struct cpuset *parent;
1969
1970        /*
1971         * The cgroup's css_sets list is in use if there are tasks
1972         * in the cpuset; the list is empty if there are none;
1973         * the cs->css.refcnt seems always 0.
1974         */
1975        if (list_empty(&cs->css.cgroup->css_sets))
1976                return;
1977
1978        /*
1979         * Find its next-highest non-empty parent, (top cpuset
1980         * has online cpus, so can't be empty).
1981         */
1982        parent = cs->parent;
1983        while (cpumask_empty(parent->cpus_allowed) ||
1984                        nodes_empty(parent->mems_allowed))
1985                parent = parent->parent;
1986
1987        move_member_tasks_to_cpuset(cs, parent);
1988}
1989
1990/*
1991 * Helper function to traverse cpusets.
1992 * It can be used to walk the cpuset tree from top to bottom, completing
1993 * one layer before dropping down to the next (thus always processing a
1994 * node before any of its children).
1995 */
1996static struct cpuset *cpuset_next(struct list_head *queue)
1997{
1998        struct cpuset *cp;
1999        struct cpuset *child;   /* scans child cpusets of cp */
2000        struct cgroup *cont;
2001
2002        if (list_empty(queue))
2003                return NULL;
2004
2005        cp = list_first_entry(queue, struct cpuset, stack_list);
2006        list_del(queue->next);
2007        list_for_each_entry(cont, &cp->css.cgroup->children, sibling) {
2008                child = cgroup_cs(cont);
2009                list_add_tail(&child->stack_list, queue);
2010        }
2011
2012        return cp;
2013}
2014
2015
2016/*
2017 * Walk the specified cpuset subtree upon a hotplug operation (CPU/Memory
2018 * online/offline) and update the cpusets accordingly.
2019 * For regular CPU/Mem hotplug, look for empty cpusets; the tasks of such
2020 * cpuset must be moved to a parent cpuset.
2021 *
2022 * Called with cgroup_mutex held.  We take callback_mutex to modify
2023 * cpus_allowed and mems_allowed.
2024 *
2025 * This walk processes the tree from top to bottom, completing one layer
2026 * before dropping down to the next.  It always processes a node before
2027 * any of its children.
2028 *
2029 * In the case of memory hot-unplug, it will remove nodes from N_MEMORY
2030 * if all present pages from a node are offlined.
2031 */
2032static void
2033scan_cpusets_upon_hotplug(struct cpuset *root, enum hotplug_event event)
2034{
2035        LIST_HEAD(queue);
2036        struct cpuset *cp;              /* scans cpusets being updated */
2037        static nodemask_t oldmems;      /* protected by cgroup_mutex */
2038
2039        list_add_tail((struct list_head *)&root->stack_list, &queue);
2040
2041        switch (event) {
2042        case CPUSET_CPU_OFFLINE:
2043                while ((cp = cpuset_next(&queue)) != NULL) {
2044
2045                        /* Continue past cpusets with all cpus online */
2046                        if (cpumask_subset(cp->cpus_allowed, cpu_active_mask))
2047                                continue;
2048
2049                        /* Remove offline cpus from this cpuset. */
2050                        mutex_lock(&callback_mutex);
2051                        cpumask_and(cp->cpus_allowed, cp->cpus_allowed,
2052                                                        cpu_active_mask);
2053                        mutex_unlock(&callback_mutex);
2054
2055                        /* Move tasks from the empty cpuset to a parent */
2056                        if (cpumask_empty(cp->cpus_allowed))
2057                                remove_tasks_in_empty_cpuset(cp);
2058                        else
2059                                update_tasks_cpumask(cp, NULL);
2060                }
2061                break;
2062
2063        case CPUSET_MEM_OFFLINE:
2064                while ((cp = cpuset_next(&queue)) != NULL) {
2065
2066                        /* Continue past cpusets with all mems online */
2067                        if (nodes_subset(cp->mems_allowed,
2068                                        node_states[N_MEMORY]))
2069                                continue;
2070
2071                        oldmems = cp->mems_allowed;
2072
2073                        /* Remove offline mems from this cpuset. */
2074                        mutex_lock(&callback_mutex);
2075                        nodes_and(cp->mems_allowed, cp->mems_allowed,
2076                                                node_states[N_MEMORY]);
2077                        mutex_unlock(&callback_mutex);
2078
2079                        /* Move tasks from the empty cpuset to a parent */
2080                        if (nodes_empty(cp->mems_allowed))
2081                                remove_tasks_in_empty_cpuset(cp);
2082                        else
2083                                update_tasks_nodemask(cp, &oldmems, NULL);
2084                }
2085        }
2086}
2087
2088/*
2089 * The top_cpuset tracks what CPUs and Memory Nodes are online,
2090 * period.  This is necessary in order to make cpusets transparent
2091 * (of no affect) on systems that are actively using CPU hotplug
2092 * but making no active use of cpusets.
2093 *
2094 * The only exception to this is suspend/resume, where we don't
2095 * modify cpusets at all.
2096 *
2097 * This routine ensures that top_cpuset.cpus_allowed tracks
2098 * cpu_active_mask on each CPU hotplug (cpuhp) event.
2099 *
2100 * Called within get_online_cpus().  Needs to call cgroup_lock()
2101 * before calling generate_sched_domains().
2102 *
2103 * @cpu_online: Indicates whether this is a CPU online event (true) or
2104 * a CPU offline event (false).
2105 */
2106void cpuset_update_active_cpus(bool cpu_online)
2107{
2108        struct sched_domain_attr *attr;
2109        cpumask_var_t *doms;
2110        int ndoms;
2111
2112        cgroup_lock();
2113        mutex_lock(&callback_mutex);
2114        cpumask_copy(top_cpuset.cpus_allowed, cpu_active_mask);
2115        mutex_unlock(&callback_mutex);
2116
2117        if (!cpu_online)
2118                scan_cpusets_upon_hotplug(&top_cpuset, CPUSET_CPU_OFFLINE);
2119
2120        ndoms = generate_sched_domains(&doms, &attr);
2121        cgroup_unlock();
2122
2123        /* Have scheduler rebuild the domains */
2124        partition_sched_domains(ndoms, doms, attr);
2125}
2126
2127#ifdef CONFIG_MEMORY_HOTPLUG
2128/*
2129 * Keep top_cpuset.mems_allowed tracking node_states[N_MEMORY].
2130 * Call this routine anytime after node_states[N_MEMORY] changes.
2131 * See cpuset_update_active_cpus() for CPU hotplug handling.
2132 */
2133static int cpuset_track_online_nodes(struct notifier_block *self,
2134                                unsigned long action, void *arg)
2135{
2136        static nodemask_t oldmems;      /* protected by cgroup_mutex */
2137
2138        cgroup_lock();
2139        switch (action) {
2140        case MEM_ONLINE:
2141                oldmems = top_cpuset.mems_allowed;
2142                mutex_lock(&callback_mutex);
2143                top_cpuset.mems_allowed = node_states[N_MEMORY];
2144                mutex_unlock(&callback_mutex);
2145                update_tasks_nodemask(&top_cpuset, &oldmems, NULL);
2146                break;
2147        case MEM_OFFLINE:
2148                /*
2149                 * needn't update top_cpuset.mems_allowed explicitly because
2150                 * scan_cpusets_upon_hotplug() will update it.
2151                 */
2152                scan_cpusets_upon_hotplug(&top_cpuset, CPUSET_MEM_OFFLINE);
2153                break;
2154        default:
2155                break;
2156        }
2157        cgroup_unlock();
2158
2159        return NOTIFY_OK;
2160}
2161#endif
2162
2163/**
2164 * cpuset_init_smp - initialize cpus_allowed
2165 *
2166 * Description: Finish top cpuset after cpu, node maps are initialized
2167 **/
2168
2169void __init cpuset_init_smp(void)
2170{
2171        cpumask_copy(top_cpuset.cpus_allowed, cpu_active_mask);
2172        top_cpuset.mems_allowed = node_states[N_MEMORY];
2173
2174        hotplug_memory_notifier(cpuset_track_online_nodes, 10);
2175
2176        cpuset_wq = create_singlethread_workqueue("cpuset");
2177        BUG_ON(!cpuset_wq);
2178}
2179
2180/**
2181 * cpuset_cpus_allowed - return cpus_allowed mask from a tasks cpuset.
2182 * @tsk: pointer to task_struct from which to obtain cpuset->cpus_allowed.
2183 * @pmask: pointer to struct cpumask variable to receive cpus_allowed set.
2184 *
2185 * Description: Returns the cpumask_var_t cpus_allowed of the cpuset
2186 * attached to the specified @tsk.  Guaranteed to return some non-empty
2187 * subset of cpu_online_mask, even if this means going outside the
2188 * tasks cpuset.
2189 **/
2190
2191void cpuset_cpus_allowed(struct task_struct *tsk, struct cpumask *pmask)
2192{
2193        mutex_lock(&callback_mutex);
2194        task_lock(tsk);
2195        guarantee_online_cpus(task_cs(tsk), pmask);
2196        task_unlock(tsk);
2197        mutex_unlock(&callback_mutex);
2198}
2199
2200void cpuset_cpus_allowed_fallback(struct task_struct *tsk)
2201{
2202        const struct cpuset *cs;
2203
2204        rcu_read_lock();
2205        cs = task_cs(tsk);
2206        if (cs)
2207                do_set_cpus_allowed(tsk, cs->cpus_allowed);
2208        rcu_read_unlock();
2209
2210        /*
2211         * We own tsk->cpus_allowed, nobody can change it under us.
2212         *
2213         * But we used cs && cs->cpus_allowed lockless and thus can
2214         * race with cgroup_attach_task() or update_cpumask() and get
2215         * the wrong tsk->cpus_allowed. However, both cases imply the
2216         * subsequent cpuset_change_cpumask()->set_cpus_allowed_ptr()
2217         * which takes task_rq_lock().
2218         *
2219         * If we are called after it dropped the lock we must see all
2220         * changes in tsk_cs()->cpus_allowed. Otherwise we can temporary
2221         * set any mask even if it is not right from task_cs() pov,
2222         * the pending set_cpus_allowed_ptr() will fix things.
2223         *
2224         * select_fallback_rq() will fix things ups and set cpu_possible_mask
2225         * if required.
2226         */
2227}
2228
2229void cpuset_init_current_mems_allowed(void)
2230{
2231        nodes_setall(current->mems_allowed);
2232}
2233
2234/**
2235 * cpuset_mems_allowed - return mems_allowed mask from a tasks cpuset.
2236 * @tsk: pointer to task_struct from which to obtain cpuset->mems_allowed.
2237 *
2238 * Description: Returns the nodemask_t mems_allowed of the cpuset
2239 * attached to the specified @tsk.  Guaranteed to return some non-empty
2240 * subset of node_states[N_MEMORY], even if this means going outside the
2241 * tasks cpuset.
2242 **/
2243
2244nodemask_t cpuset_mems_allowed(struct task_struct *tsk)
2245{
2246        nodemask_t mask;
2247
2248        mutex_lock(&callback_mutex);
2249        task_lock(tsk);
2250        guarantee_online_mems(task_cs(tsk), &mask);
2251        task_unlock(tsk);
2252        mutex_unlock(&callback_mutex);
2253
2254        return mask;
2255}
2256
2257/**
2258 * cpuset_nodemask_valid_mems_allowed - check nodemask vs. curremt mems_allowed
2259 * @nodemask: the nodemask to be checked
2260 *
2261 * Are any of the nodes in the nodemask allowed in current->mems_allowed?
2262 */
2263int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask)
2264{
2265        return nodes_intersects(*nodemask, current->mems_allowed);
2266}
2267
2268/*
2269 * nearest_hardwall_ancestor() - Returns the nearest mem_exclusive or
2270 * mem_hardwall ancestor to the specified cpuset.  Call holding
2271 * callback_mutex.  If no ancestor is mem_exclusive or mem_hardwall
2272 * (an unusual configuration), then returns the root cpuset.
2273 */
2274static const struct cpuset *nearest_hardwall_ancestor(const struct cpuset *cs)
2275{
2276        while (!(is_mem_exclusive(cs) || is_mem_hardwall(cs)) && cs->parent)
2277                cs = cs->parent;
2278        return cs;
2279}
2280
2281/**
2282 * cpuset_node_allowed_softwall - Can we allocate on a memory node?
2283 * @node: is this an allowed node?
2284 * @gfp_mask: memory allocation flags
2285 *
2286 * If we're in interrupt, yes, we can always allocate.  If __GFP_THISNODE is
2287 * set, yes, we can always allocate.  If node is in our task's mems_allowed,
2288 * yes.  If it's not a __GFP_HARDWALL request and this node is in the nearest
2289 * hardwalled cpuset ancestor to this task's cpuset, yes.  If the task has been
2290 * OOM killed and has access to memory reserves as specified by the TIF_MEMDIE
2291 * flag, yes.
2292 * Otherwise, no.
2293 *
2294 * If __GFP_HARDWALL is set, cpuset_node_allowed_softwall() reduces to
2295 * cpuset_node_allowed_hardwall().  Otherwise, cpuset_node_allowed_softwall()
2296 * might sleep, and might allow a node from an enclosing cpuset.
2297 *
2298 * cpuset_node_allowed_hardwall() only handles the simpler case of hardwall
2299 * cpusets, and never sleeps.
2300 *
2301 * The __GFP_THISNODE placement logic is really handled elsewhere,
2302 * by forcibly using a zonelist starting at a specified node, and by
2303 * (in get_page_from_freelist()) refusing to consider the zones for
2304 * any node on the zonelist except the first.  By the time any such
2305 * calls get to this routine, we should just shut up and say 'yes'.
2306 *
2307 * GFP_USER allocations are marked with the __GFP_HARDWALL bit,
2308 * and do not allow allocations outside the current tasks cpuset
2309 * unless the task has been OOM killed as is marked TIF_MEMDIE.
2310 * GFP_KERNEL allocations are not so marked, so can escape to the
2311 * nearest enclosing hardwalled ancestor cpuset.
2312 *
2313 * Scanning up parent cpusets requires callback_mutex.  The
2314 * __alloc_pages() routine only calls here with __GFP_HARDWALL bit
2315 * _not_ set if it's a GFP_KERNEL allocation, and all nodes in the
2316 * current tasks mems_allowed came up empty on the first pass over
2317 * the zonelist.  So only GFP_KERNEL allocations, if all nodes in the
2318 * cpuset are short of memory, might require taking the callback_mutex
2319 * mutex.
2320 *
2321 * The first call here from mm/page_alloc:get_page_from_freelist()
2322 * has __GFP_HARDWALL set in gfp_mask, enforcing hardwall cpusets,
2323 * so no allocation on a node outside the cpuset is allowed (unless
2324 * in interrupt, of course).
2325 *
2326 * The second pass through get_page_from_freelist() doesn't even call
2327 * here for GFP_ATOMIC calls.  For those calls, the __alloc_pages()
2328 * variable 'wait' is not set, and the bit ALLOC_CPUSET is not set
2329 * in alloc_flags.  That logic and the checks below have the combined
2330 * affect that:
2331 *      in_interrupt - any node ok (current task context irrelevant)
2332 *      GFP_ATOMIC   - any node ok
2333 *      TIF_MEMDIE   - any node ok
2334 *      GFP_KERNEL   - any node in enclosing hardwalled cpuset ok
2335 *      GFP_USER     - only nodes in current tasks mems allowed ok.
2336 *
2337 * Rule:
2338 *    Don't call cpuset_node_allowed_softwall if you can't sleep, unless you
2339 *    pass in the __GFP_HARDWALL flag set in gfp_flag, which disables
2340 *    the code that might scan up ancestor cpusets and sleep.
2341 */
2342int __cpuset_node_allowed_softwall(int node, gfp_t gfp_mask)
2343{
2344        const struct cpuset *cs;        /* current cpuset ancestors */
2345        int allowed;                    /* is allocation in zone z allowed? */
2346
2347        if (in_interrupt() || (gfp_mask & __GFP_THISNODE))
2348                return 1;
2349        might_sleep_if(!(gfp_mask & __GFP_HARDWALL));
2350        if (node_isset(node, current->mems_allowed))
2351                return 1;
2352        /*
2353         * Allow tasks that have access to memory reserves because they have
2354         * been OOM killed to get memory anywhere.
2355         */
2356        if (unlikely(test_thread_flag(TIF_MEMDIE)))
2357                return 1;
2358        if (gfp_mask & __GFP_HARDWALL)  /* If hardwall request, stop here */
2359                return 0;
2360
2361        if (current->flags & PF_EXITING) /* Let dying task have memory */
2362                return 1;
2363
2364        /* Not hardwall and node outside mems_allowed: scan up cpusets */
2365        mutex_lock(&callback_mutex);
2366
2367        task_lock(current);
2368        cs = nearest_hardwall_ancestor(task_cs(current));
2369        task_unlock(current);
2370
2371        allowed = node_isset(node, cs->mems_allowed);
2372        mutex_unlock(&callback_mutex);
2373        return allowed;
2374}
2375
2376/*
2377 * cpuset_node_allowed_hardwall - Can we allocate on a memory node?
2378 * @node: is this an allowed node?
2379 * @gfp_mask: memory allocation flags
2380 *
2381 * If we're in interrupt, yes, we can always allocate.  If __GFP_THISNODE is
2382 * set, yes, we can always allocate.  If node is in our task's mems_allowed,
2383 * yes.  If the task has been OOM killed and has access to memory reserves as
2384 * specified by the TIF_MEMDIE flag, yes.
2385 * Otherwise, no.
2386 *
2387 * The __GFP_THISNODE placement logic is really handled elsewhere,
2388 * by forcibly using a zonelist starting at a specified node, and by
2389 * (in get_page_from_freelist()) refusing to consider the zones for
2390 * any node on the zonelist except the first.  By the time any such
2391 * calls get to this routine, we should just shut up and say 'yes'.
2392 *
2393 * Unlike the cpuset_node_allowed_softwall() variant, above,
2394 * this variant requires that the node be in the current task's
2395 * mems_allowed or that we're in interrupt.  It does not scan up the
2396 * cpuset hierarchy for the nearest enclosing mem_exclusive cpuset.
2397 * It never sleeps.
2398 */
2399int __cpuset_node_allowed_hardwall(int node, gfp_t gfp_mask)
2400{
2401        if (in_interrupt() || (gfp_mask & __GFP_THISNODE))
2402                return 1;
2403        if (node_isset(node, current->mems_allowed))
2404                return 1;
2405        /*
2406         * Allow tasks that have access to memory reserves because they have
2407         * been OOM killed to get memory anywhere.
2408         */
2409        if (unlikely(test_thread_flag(TIF_MEMDIE)))
2410                return 1;
2411        return 0;
2412}
2413
2414/**
2415 * cpuset_unlock - release lock on cpuset changes
2416 *
2417 * Undo the lock taken in a previous cpuset_lock() call.
2418 */
2419
2420void cpuset_unlock(void)
2421{
2422        mutex_unlock(&callback_mutex);
2423}
2424
2425/**
2426 * cpuset_mem_spread_node() - On which node to begin search for a file page
2427 * cpuset_slab_spread_node() - On which node to begin search for a slab page
2428 *
2429 * If a task is marked PF_SPREAD_PAGE or PF_SPREAD_SLAB (as for
2430 * tasks in a cpuset with is_spread_page or is_spread_slab set),
2431 * and if the memory allocation used cpuset_mem_spread_node()
2432 * to determine on which node to start looking, as it will for
2433 * certain page cache or slab cache pages such as used for file
2434 * system buffers and inode caches, then instead of starting on the
2435 * local node to look for a free page, rather spread the starting
2436 * node around the tasks mems_allowed nodes.
2437 *
2438 * We don't have to worry about the returned node being offline
2439 * because "it can't happen", and even if it did, it would be ok.
2440 *
2441 * The routines calling guarantee_online_mems() are careful to
2442 * only set nodes in task->mems_allowed that are online.  So it
2443 * should not be possible for the following code to return an
2444 * offline node.  But if it did, that would be ok, as this routine
2445 * is not returning the node where the allocation must be, only
2446 * the node where the search should start.  The zonelist passed to
2447 * __alloc_pages() will include all nodes.  If the slab allocator
2448 * is passed an offline node, it will fall back to the local node.
2449 * See kmem_cache_alloc_node().
2450 */
2451
2452static int cpuset_spread_node(int *rotor)
2453{
2454        int node;
2455
2456        node = next_node(*rotor, current->mems_allowed);
2457        if (node == MAX_NUMNODES)
2458                node = first_node(current->mems_allowed);
2459        *rotor = node;
2460        return node;
2461}
2462
2463int cpuset_mem_spread_node(void)
2464{
2465        if (current->cpuset_mem_spread_rotor == NUMA_NO_NODE)
2466                current->cpuset_mem_spread_rotor =
2467                        node_random(&current->mems_allowed);
2468
2469        return cpuset_spread_node(&current->cpuset_mem_spread_rotor);
2470}
2471
2472int cpuset_slab_spread_node(void)
2473{
2474        if (current->cpuset_slab_spread_rotor == NUMA_NO_NODE)
2475                current->cpuset_slab_spread_rotor =
2476                        node_random(&current->mems_allowed);
2477
2478        return cpuset_spread_node(&current->cpuset_slab_spread_rotor);
2479}
2480
2481EXPORT_SYMBOL_GPL(cpuset_mem_spread_node);
2482
2483/**
2484 * cpuset_mems_allowed_intersects - Does @tsk1's mems_allowed intersect @tsk2's?
2485 * @tsk1: pointer to task_struct of some task.
2486 * @tsk2: pointer to task_struct of some other task.
2487 *
2488 * Description: Return true if @tsk1's mems_allowed intersects the
2489 * mems_allowed of @tsk2.  Used by the OOM killer to determine if
2490 * one of the task's memory usage might impact the memory available
2491 * to the other.
2492 **/
2493
2494int cpuset_mems_allowed_intersects(const struct task_struct *tsk1,
2495                                   const struct task_struct *tsk2)
2496{
2497        return nodes_intersects(tsk1->mems_allowed, tsk2->mems_allowed);
2498}
2499
2500/**
2501 * cpuset_print_task_mems_allowed - prints task's cpuset and mems_allowed
2502 * @task: pointer to task_struct of some task.
2503 *
2504 * Description: Prints @task's name, cpuset name, and cached copy of its
2505 * mems_allowed to the kernel log.  Must hold task_lock(task) to allow
2506 * dereferencing task_cs(task).
2507 */
2508void cpuset_print_task_mems_allowed(struct task_struct *tsk)
2509{
2510        struct dentry *dentry;
2511
2512        dentry = task_cs(tsk)->css.cgroup->dentry;
2513        spin_lock(&cpuset_buffer_lock);
2514        snprintf(cpuset_name, CPUSET_NAME_LEN,
2515                 dentry ? (const char *)dentry->d_name.name : "/");
2516        nodelist_scnprintf(cpuset_nodelist, CPUSET_NODELIST_LEN,
2517                           tsk->mems_allowed);
2518        printk(KERN_INFO "%s cpuset=%s mems_allowed=%s\n",
2519               tsk->comm, cpuset_name, cpuset_nodelist);
2520        spin_unlock(&cpuset_buffer_lock);
2521}
2522
2523/*
2524 * Collection of memory_pressure is suppressed unless
2525 * this flag is enabled by writing "1" to the special
2526 * cpuset file 'memory_pressure_enabled' in the root cpuset.
2527 */
2528
2529int cpuset_memory_pressure_enabled __read_mostly;
2530
2531/**
2532 * cpuset_memory_pressure_bump - keep stats of per-cpuset reclaims.
2533 *
2534 * Keep a running average of the rate of synchronous (direct)
2535 * page reclaim efforts initiated by tasks in each cpuset.
2536 *
2537 * This represents the rate at which some task in the cpuset
2538 * ran low on memory on all nodes it was allowed to use, and
2539 * had to enter the kernels page reclaim code in an effort to
2540 * create more free memory by tossing clean pages or swapping
2541 * or writing dirty pages.
2542 *
2543 * Display to user space in the per-cpuset read-only file
2544 * "memory_pressure".  Value displayed is an integer
2545 * representing the recent rate of entry into the synchronous
2546 * (direct) page reclaim by any task attached to the cpuset.
2547 **/
2548
2549void __cpuset_memory_pressure_bump(void)
2550{
2551        task_lock(current);
2552        fmeter_markevent(&task_cs(current)->fmeter);
2553        task_unlock(current);
2554}
2555
2556#ifdef CONFIG_PROC_PID_CPUSET
2557/*
2558 * proc_cpuset_show()
2559 *  - Print tasks cpuset path into seq_file.
2560 *  - Used for /proc/<pid>/cpuset.
2561 *  - No need to task_lock(tsk) on this tsk->cpuset reference, as it
2562 *    doesn't really matter if tsk->cpuset changes after we read it,
2563 *    and we take cgroup_mutex, keeping cpuset_attach() from changing it
2564 *    anyway.
2565 */
2566static int proc_cpuset_show(struct seq_file *m, void *unused_v)
2567{
2568        struct pid *pid;
2569        struct task_struct *tsk;
2570        char *buf;
2571        struct cgroup_subsys_state *css;
2572        int retval;
2573
2574        retval = -ENOMEM;
2575        buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
2576        if (!buf)
2577                goto out;
2578
2579        retval = -ESRCH;
2580        pid = m->private;
2581        tsk = get_pid_task(pid, PIDTYPE_PID);
2582        if (!tsk)
2583                goto out_free;
2584
2585        retval = -EINVAL;
2586        cgroup_lock();
2587        css = task_subsys_state(tsk, cpuset_subsys_id);
2588        retval = cgroup_path(css->cgroup, buf, PAGE_SIZE);
2589        if (retval < 0)
2590                goto out_unlock;
2591        seq_puts(m, buf);
2592        seq_putc(m, '\n');
2593out_unlock:
2594        cgroup_unlock();
2595        put_task_struct(tsk);
2596out_free:
2597        kfree(buf);
2598out:
2599        return retval;
2600}
2601
2602static int cpuset_open(struct inode *inode, struct file *file)
2603{
2604        struct pid *pid = PROC_I(inode)->pid;
2605        return single_open(file, proc_cpuset_show, pid);
2606}
2607
2608const struct file_operations proc_cpuset_operations = {
2609        .open           = cpuset_open,
2610        .read           = seq_read,
2611        .llseek         = seq_lseek,
2612        .release        = single_release,
2613};
2614#endif /* CONFIG_PROC_PID_CPUSET */
2615
2616/* Display task mems_allowed in /proc/<pid>/status file. */
2617void cpuset_task_status_allowed(struct seq_file *m, struct task_struct *task)
2618{
2619        seq_printf(m, "Mems_allowed:\t");
2620        seq_nodemask(m, &task->mems_allowed);
2621        seq_printf(m, "\n");
2622        seq_printf(m, "Mems_allowed_list:\t");
2623        seq_nodemask_list(m, &task->mems_allowed);
2624        seq_printf(m, "\n");
2625}
2626
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.