cpuset.c 75.6 KB
Newer Older
Linus Torvalds's avatar
Linus Torvalds committed
1 2 3 4 5 6
/*
 *  kernel/cpuset.c
 *
 *  Processor and Memory placement constraints for sets of tasks.
 *
 *  Copyright (C) 2003 BULL SA.
Paul Jackson's avatar
Paul Jackson committed
7
 *  Copyright (C) 2004-2007 Silicon Graphics, Inc.
8
 *  Copyright (C) 2006 Google, Inc
Linus Torvalds's avatar
Linus Torvalds committed
9 10 11 12
 *
 *  Portions derived from Patrick Mochel's sysfs code.
 *  sysfs is Copyright (c) 2001-3 Patrick Mochel
 *
13
 *  2003-10-10 Written by Simon Derr.
Linus Torvalds's avatar
Linus Torvalds committed
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 18
 *  2008 Rework of the scheduler domains and CPU hotplug handling
 *       by Max Krasnyansky
Linus Torvalds's avatar
Linus Torvalds committed
19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36
 *
 *  This file is subject to the terms and conditions of the GNU General Public
 *  License.  See the file COPYING in the main directory of the Linux
 *  distribution for more details.
 */

#include <linux/cpu.h>
#include <linux/cpumask.h>
#include <linux/cpuset.h>
#include <linux/err.h>
#include <linux/errno.h>
#include <linux/file.h>
#include <linux/fs.h>
#include <linux/init.h>
#include <linux/interrupt.h>
#include <linux/kernel.h>
#include <linux/kmod.h>
#include <linux/list.h>
37
#include <linux/mempolicy.h>
Linus Torvalds's avatar
Linus Torvalds committed
38
#include <linux/mm.h>
39
#include <linux/memory.h>
40
#include <linux/export.h>
Linus Torvalds's avatar
Linus Torvalds committed
41 42 43 44
#include <linux/mount.h>
#include <linux/namei.h>
#include <linux/pagemap.h>
#include <linux/proc_fs.h>
45
#include <linux/rcupdate.h>
Linus Torvalds's avatar
Linus Torvalds committed
46 47
#include <linux/sched.h>
#include <linux/seq_file.h>
48
#include <linux/security.h>
Linus Torvalds's avatar
Linus Torvalds committed
49 50 51 52 53 54 55 56 57
#include <linux/slab.h>
#include <linux/spinlock.h>
#include <linux/stat.h>
#include <linux/string.h>
#include <linux/time.h>
#include <linux/backing-dev.h>
#include <linux/sort.h>

#include <asm/uaccess.h>
Arun Sharma's avatar
Arun Sharma committed
58
#include <linux/atomic.h>
59
#include <linux/mutex.h>
60 61
#include <linux/workqueue.h>
#include <linux/cgroup.h>
Linus Torvalds's avatar
Linus Torvalds committed
62

63 64 65 66 67 68 69 70
/*
 * Workqueue for cpuset related tasks.
 *
 * Using kevent workqueue may cause deadlock when memory_migrate
 * is set. So we create a separate workqueue thread for cpuset.
 */
static struct workqueue_struct *cpuset_wq;

71 72 73 74 75
/*
 * Tracks how many cpusets are currently defined in system.
 * When there is only one cpuset (the root cpuset) we can
 * short circuit some hooks.
 */
76
int number_of_cpusets __read_mostly;
77

78
/* Forward declare cgroup structures */
79 80 81
struct cgroup_subsys cpuset_subsys;
struct cpuset;

82 83 84 85 86 87 88 89 90
/* See "Frequency meter" comments, below. */

struct fmeter {
	int cnt;		/* unprocessed events count */
	int val;		/* most recent output value */
	time_t time;		/* clock (secs) when val computed */
	spinlock_t lock;	/* guards read or write of above */
};

Linus Torvalds's avatar
Linus Torvalds committed
91
struct cpuset {
92 93
	struct cgroup_subsys_state css;

Linus Torvalds's avatar
Linus Torvalds committed
94
	unsigned long flags;		/* "unsigned long" so bitops work */
95
	cpumask_var_t cpus_allowed;	/* CPUs allowed to tasks in cpuset */
Linus Torvalds's avatar
Linus Torvalds committed
96 97 98 99
	nodemask_t mems_allowed;	/* Memory Nodes allowed to tasks */

	struct cpuset *parent;		/* my parent */

100
	struct fmeter fmeter;		/* memory_pressure filter */
Paul Jackson's avatar
Paul Jackson committed
101 102 103

	/* partition number for rebuild_sched_domains() */
	int pn;
104

105 106 107
	/* for custom sched domain */
	int relax_domain_level;

108
	/* used for walking a cpuset hierarchy */
109
	struct list_head stack_list;
Linus Torvalds's avatar
Linus Torvalds committed
110 111
};

112 113 114 115 116 117 118 119 120 121 122 123 124 125
/* Retrieve the cpuset for a cgroup */
static inline struct cpuset *cgroup_cs(struct cgroup *cont)
{
	return container_of(cgroup_subsys_state(cont, cpuset_subsys_id),
			    struct cpuset, css);
}

/* Retrieve the cpuset for a task */
static inline struct cpuset *task_cs(struct task_struct *task)
{
	return container_of(task_subsys_state(task, cpuset_subsys_id),
			    struct cpuset, css);
}

126 127 128 129 130 131 132 133 134 135 136 137 138
#ifdef CONFIG_NUMA
static inline bool task_has_mempolicy(struct task_struct *task)
{
	return task->mempolicy;
}
#else
static inline bool task_has_mempolicy(struct task_struct *task)
{
	return false;
}
#endif


Linus Torvalds's avatar
Linus Torvalds committed
139 140
/* bits in struct cpuset flags field */
typedef enum {
Tejun Heo's avatar
Tejun Heo committed
141
	CS_ONLINE,
Linus Torvalds's avatar
Linus Torvalds committed
142 143
	CS_CPU_EXCLUSIVE,
	CS_MEM_EXCLUSIVE,
144
	CS_MEM_HARDWALL,
145
	CS_MEMORY_MIGRATE,
Paul Jackson's avatar
Paul Jackson committed
146
	CS_SCHED_LOAD_BALANCE,
147 148
	CS_SPREAD_PAGE,
	CS_SPREAD_SLAB,
Linus Torvalds's avatar
Linus Torvalds committed
149 150 151
} cpuset_flagbits_t;

/* convenient tests for these bits */
Tejun Heo's avatar
Tejun Heo committed
152 153 154 155 156
static inline bool is_cpuset_online(const struct cpuset *cs)
{
	return test_bit(CS_ONLINE, &cs->flags);
}

Linus Torvalds's avatar
Linus Torvalds committed
157 158
static inline int is_cpu_exclusive(const struct cpuset *cs)
{
159
	return test_bit(CS_CPU_EXCLUSIVE, &cs->flags);
Linus Torvalds's avatar
Linus Torvalds committed
160 161 162 163
}

static inline int is_mem_exclusive(const struct cpuset *cs)
{
164
	return test_bit(CS_MEM_EXCLUSIVE, &cs->flags);
Linus Torvalds's avatar
Linus Torvalds committed
165 166
}

167 168 169 170 171
static inline int is_mem_hardwall(const struct cpuset *cs)
{
	return test_bit(CS_MEM_HARDWALL, &cs->flags);
}

Paul Jackson's avatar
Paul Jackson committed
172 173 174 175 176
static inline int is_sched_load_balance(const struct cpuset *cs)
{
	return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
}

177 178
static inline int is_memory_migrate(const struct cpuset *cs)
{
179
	return test_bit(CS_MEMORY_MIGRATE, &cs->flags);
180 181
}

182 183 184 185 186 187 188 189 190 191
static inline int is_spread_page(const struct cpuset *cs)
{
	return test_bit(CS_SPREAD_PAGE, &cs->flags);
}

static inline int is_spread_slab(const struct cpuset *cs)
{
	return test_bit(CS_SPREAD_SLAB, &cs->flags);
}

Linus Torvalds's avatar
Linus Torvalds committed
192
static struct cpuset top_cpuset = {
Tejun Heo's avatar
Tejun Heo committed
193 194
	.flags = ((1 << CS_ONLINE) | (1 << CS_CPU_EXCLUSIVE) |
		  (1 << CS_MEM_EXCLUSIVE)),
Linus Torvalds's avatar
Linus Torvalds committed
195 196
};

197 198 199 200 201 202 203 204 205 206 207 208 209
/**
 * cpuset_for_each_child - traverse online children of a cpuset
 * @child_cs: loop cursor pointing to the current child
 * @pos_cgrp: used for iteration
 * @parent_cs: target cpuset to walk children of
 *
 * Walk @child_cs through the online children of @parent_cs.  Must be used
 * with RCU read locked.
 */
#define cpuset_for_each_child(child_cs, pos_cgrp, parent_cs)		\
	cgroup_for_each_child((pos_cgrp), (parent_cs)->css.cgroup)	\
		if (is_cpuset_online(((child_cs) = cgroup_cs((pos_cgrp)))))

Linus Torvalds's avatar
Linus Torvalds committed
210
/*
211 212 213 214 215 216 217
 * There are two global mutexes guarding cpuset structures.  The first
 * is the main control groups cgroup_mutex, accessed via
 * cgroup_lock()/cgroup_unlock().  The second is the cpuset-specific
 * callback_mutex, below. They can nest.  It is ok to first take
 * cgroup_mutex, then nest callback_mutex.  We also require taking
 * task_lock() when dereferencing a task's cpuset pointer.  See "The
 * task_lock() exception", at the end of this comment.
218
 *
219
 * A task must hold both mutexes to modify cpusets.  If a task
220
 * holds cgroup_mutex, then it blocks others wanting that mutex,
221
 * ensuring that it is the only task able to also acquire callback_mutex
222 223
 * and be able to modify cpusets.  It can perform various checks on
 * the cpuset structure first, knowing nothing will change.  It can
224
 * also allocate memory while just holding cgroup_mutex.  While it is
225
 * performing these checks, various callback routines can briefly
226 227
 * acquire callback_mutex to query cpusets.  Once it is ready to make
 * the changes, it takes callback_mutex, blocking everyone else.
228 229
 *
 * Calls to the kernel memory allocator can not be made while holding
230
 * callback_mutex, as that would risk double tripping on callback_mutex
231 232 233
 * from one of the callbacks into the cpuset code from within
 * __alloc_pages().
 *
234
 * If a task is only holding callback_mutex, then it has read-only
235 236
 * access to cpusets.
 *
237 238 239
 * Now, the task_struct fields mems_allowed and mempolicy may be changed
 * by other task, we use alloc_lock in the task_struct fields to protect
 * them.
240
 *
241
 * The cpuset_common_file_read() handlers only hold callback_mutex across
242 243 244
 * small pieces of code, such as when reading out possibly multi-word
 * cpumasks and nodemasks.
 *
245 246
 * Accessing a task's cpuset should be done in accordance with the
 * guidelines for accessing subsystem state in kernel/cgroup.c
Linus Torvalds's avatar
Linus Torvalds committed
247 248
 */

249
static DEFINE_MUTEX(callback_mutex);
250

251 252 253 254 255 256 257 258 259 260 261
/*
 * cpuset_buffer_lock protects both the cpuset_name and cpuset_nodelist
 * buffers.  They are statically allocated to prevent using excess stack
 * when calling cpuset_print_task_mems_allowed().
 */
#define CPUSET_NAME_LEN		(128)
#define	CPUSET_NODELIST_LEN	(256)
static char cpuset_name[CPUSET_NAME_LEN];
static char cpuset_nodelist[CPUSET_NODELIST_LEN];
static DEFINE_SPINLOCK(cpuset_buffer_lock);

262 263 264 265 266 267 268
/*
 * CPU / memory hotplug is handled asynchronously.
 */
static void cpuset_hotplug_workfn(struct work_struct *work);

static DECLARE_WORK(cpuset_hotplug_work, cpuset_hotplug_workfn);

269 270
/*
 * This is ugly, but preserves the userspace API for existing cpuset
271
 * users. If someone tries to mount the "cpuset" filesystem, we
272 273
 * silently switch it to mount "cgroup" instead
 */
Al Viro's avatar
Al Viro committed
274 275
static struct dentry *cpuset_mount(struct file_system_type *fs_type,
			 int flags, const char *unused_dev_name, void *data)
Linus Torvalds's avatar
Linus Torvalds committed
276
{
277
	struct file_system_type *cgroup_fs = get_fs_type("cgroup");
Al Viro's avatar
Al Viro committed
278
	struct dentry *ret = ERR_PTR(-ENODEV);
279 280 281 282
	if (cgroup_fs) {
		char mountopts[] =
			"cpuset,noprefix,"
			"release_agent=/sbin/cpuset_release_agent";
Al Viro's avatar
Al Viro committed
283 284
		ret = cgroup_fs->mount(cgroup_fs, flags,
					   unused_dev_name, mountopts);
285 286 287
		put_filesystem(cgroup_fs);
	}
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
288 289 290 291
}

static struct file_system_type cpuset_fs_type = {
	.name = "cpuset",
Al Viro's avatar
Al Viro committed
292
	.mount = cpuset_mount,
Linus Torvalds's avatar
Linus Torvalds committed
293 294 295
};

/*
296
 * Return in pmask the portion of a cpusets's cpus_allowed that
Linus Torvalds's avatar
Linus Torvalds committed
297 298 299
 * are online.  If none are online, walk up the cpuset hierarchy
 * until we find one that does have some online cpus.  If we get
 * all the way to the top and still haven't found any online cpus,
300 301
 * return cpu_online_mask.  Or if passed a NULL cs from an exit'ing
 * task, return cpu_online_mask.
Linus Torvalds's avatar
Linus Torvalds committed
302 303
 *
 * One way or another, we guarantee to return some non-empty subset
304
 * of cpu_online_mask.
Linus Torvalds's avatar
Linus Torvalds committed
305
 *
306
 * Call with callback_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
307 308
 */

309 310
static void guarantee_online_cpus(const struct cpuset *cs,
				  struct cpumask *pmask)
Linus Torvalds's avatar
Linus Torvalds committed
311
{
312
	while (cs && !cpumask_intersects(cs->cpus_allowed, cpu_online_mask))
Linus Torvalds's avatar
Linus Torvalds committed
313 314
		cs = cs->parent;
	if (cs)
315
		cpumask_and(pmask, cs->cpus_allowed, cpu_online_mask);
Linus Torvalds's avatar
Linus Torvalds committed
316
	else
317 318
		cpumask_copy(pmask, cpu_online_mask);
	BUG_ON(!cpumask_intersects(pmask, cpu_online_mask));
Linus Torvalds's avatar
Linus Torvalds committed
319 320 321 322
}

/*
 * Return in *pmask the portion of a cpusets's mems_allowed that
323 324 325
 * are online, with memory.  If none are online with memory, walk
 * up the cpuset hierarchy until we find one that does have some
 * online mems.  If we get all the way to the top and still haven't
326
 * found any online mems, return node_states[N_MEMORY].
Linus Torvalds's avatar
Linus Torvalds committed
327 328
 *
 * One way or another, we guarantee to return some non-empty subset
329
 * of node_states[N_MEMORY].
Linus Torvalds's avatar
Linus Torvalds committed
330
 *
331
 * Call with callback_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
332 333 334 335
 */

static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask)
{
336
	while (cs && !nodes_intersects(cs->mems_allowed,
337
					node_states[N_MEMORY]))
Linus Torvalds's avatar
Linus Torvalds committed
338 339
		cs = cs->parent;
	if (cs)
340
		nodes_and(*pmask, cs->mems_allowed,
341
					node_states[N_MEMORY]);
Linus Torvalds's avatar
Linus Torvalds committed
342
	else
343 344
		*pmask = node_states[N_MEMORY];
	BUG_ON(!nodes_intersects(*pmask, node_states[N_MEMORY]));
Linus Torvalds's avatar
Linus Torvalds committed
345 346
}

347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364
/*
 * update task's spread flag if cpuset's page/slab spread flag is set
 *
 * Called with callback_mutex/cgroup_mutex held
 */
static void cpuset_update_task_spread_flag(struct cpuset *cs,
					struct task_struct *tsk)
{
	if (is_spread_page(cs))
		tsk->flags |= PF_SPREAD_PAGE;
	else
		tsk->flags &= ~PF_SPREAD_PAGE;
	if (is_spread_slab(cs))
		tsk->flags |= PF_SPREAD_SLAB;
	else
		tsk->flags &= ~PF_SPREAD_SLAB;
}

Linus Torvalds's avatar
Linus Torvalds committed
365 366 367 368 369
/*
 * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q?
 *
 * One cpuset is a subset of another if all its allowed CPUs and
 * Memory Nodes are a subset of the other, and its exclusive flags
370
 * are only set if the other's are set.  Call holding cgroup_mutex.
Linus Torvalds's avatar
Linus Torvalds committed
371 372 373 374
 */

static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q)
{
375
	return	cpumask_subset(p->cpus_allowed, q->cpus_allowed) &&
Linus Torvalds's avatar
Linus Torvalds committed
376 377 378 379 380
		nodes_subset(p->mems_allowed, q->mems_allowed) &&
		is_cpu_exclusive(p) <= is_cpu_exclusive(q) &&
		is_mem_exclusive(p) <= is_mem_exclusive(q);
}

381 382 383 384 385 386
/**
 * alloc_trial_cpuset - allocate a trial cpuset
 * @cs: the cpuset that the trial cpuset duplicates
 */
static struct cpuset *alloc_trial_cpuset(const struct cpuset *cs)
{
387 388 389 390 391 392 393 394 395 396 397 398 399
	struct cpuset *trial;

	trial = kmemdup(cs, sizeof(*cs), GFP_KERNEL);
	if (!trial)
		return NULL;

	if (!alloc_cpumask_var(&trial->cpus_allowed, GFP_KERNEL)) {
		kfree(trial);
		return NULL;
	}
	cpumask_copy(trial->cpus_allowed, cs->cpus_allowed);

	return trial;
400 401 402 403 404 405 406 407
}

/**
 * free_trial_cpuset - free the trial cpuset
 * @trial: the trial cpuset to be freed
 */
static void free_trial_cpuset(struct cpuset *trial)
{
408
	free_cpumask_var(trial->cpus_allowed);
409 410 411
	kfree(trial);
}

Linus Torvalds's avatar
Linus Torvalds committed
412 413 414 415 416 417 418
/*
 * validate_change() - Used to validate that any proposed cpuset change
 *		       follows the structural rules for cpusets.
 *
 * If we replaced the flag and mask values of the current cpuset
 * (cur) with those values in the trial cpuset (trial), would
 * our various subset and exclusive rules still be valid?  Presumes
419
 * cgroup_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
420 421 422 423 424 425 426 427 428 429 430 431 432 433
 *
 * 'cur' is the address of an actual, in-use cpuset.  Operations
 * such as list traversal that depend on the actual address of the
 * cpuset in the list must use cur below, not trial.
 *
 * 'trial' is the address of bulk structure copy of cur, with
 * perhaps one or more of the fields cpus_allowed, mems_allowed,
 * or flags changed to new, trial values.
 *
 * Return 0 if valid, -errno if not.
 */

static int validate_change(const struct cpuset *cur, const struct cpuset *trial)
{
434
	struct cgroup *cont;
Linus Torvalds's avatar
Linus Torvalds committed
435
	struct cpuset *c, *par;
436 437 438
	int ret;

	rcu_read_lock();
Linus Torvalds's avatar
Linus Torvalds committed
439 440

	/* Each of our child cpusets must be a subset of us */
441 442 443 444
	ret = -EBUSY;
	cpuset_for_each_child(c, cont, cur)
		if (!is_cpuset_subset(c, trial))
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
445 446

	/* Remaining checks don't apply to root cpuset */
447
	ret = 0;
448
	if (cur == &top_cpuset)
449
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
450

451 452
	par = cur->parent;

Linus Torvalds's avatar
Linus Torvalds committed
453
	/* We must be a subset of our parent cpuset */
454
	ret = -EACCES;
Linus Torvalds's avatar
Linus Torvalds committed
455
	if (!is_cpuset_subset(trial, par))
456
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
457

458 459 460 461
	/*
	 * If either I or some sibling (!= me) is exclusive, we can't
	 * overlap
	 */
462 463
	ret = -EINVAL;
	cpuset_for_each_child(c, cont, par) {
Linus Torvalds's avatar
Linus Torvalds committed
464 465
		if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) &&
		    c != cur &&
466
		    cpumask_intersects(trial->cpus_allowed, c->cpus_allowed))
467
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
468 469 470
		if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) &&
		    c != cur &&
		    nodes_intersects(trial->mems_allowed, c->mems_allowed))
471
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
472 473
	}

474
	/* Cpusets with tasks can't have empty cpus_allowed or mems_allowed */
475 476 477 478 479
	ret = -ENOSPC;
	if (cgroup_task_count(cur->css.cgroup) &&
	    (cpumask_empty(trial->cpus_allowed) ||
	     nodes_empty(trial->mems_allowed)))
		goto out;
480

481 482 483 484
	ret = 0;
out:
	rcu_read_unlock();
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
485 486
}

487
#ifdef CONFIG_SMP
Paul Jackson's avatar
Paul Jackson committed
488
/*
489
 * Helper routine for generate_sched_domains().
Paul Jackson's avatar
Paul Jackson committed
490 491 492 493
 * Do cpusets a, b have overlapping cpus_allowed masks?
 */
static int cpusets_overlap(struct cpuset *a, struct cpuset *b)
{
494
	return cpumask_intersects(a->cpus_allowed, b->cpus_allowed);
Paul Jackson's avatar
Paul Jackson committed
495 496
}

497 498 499 500 501 502 503 504
static void
update_domain_attr(struct sched_domain_attr *dattr, struct cpuset *c)
{
	if (dattr->relax_domain_level < c->relax_domain_level)
		dattr->relax_domain_level = c->relax_domain_level;
	return;
}

505 506 507 508 509 510 511 512 513 514 515 516 517 518
static void
update_domain_attr_tree(struct sched_domain_attr *dattr, struct cpuset *c)
{
	LIST_HEAD(q);

	list_add(&c->stack_list, &q);
	while (!list_empty(&q)) {
		struct cpuset *cp;
		struct cgroup *cont;
		struct cpuset *child;

		cp = list_first_entry(&q, struct cpuset, stack_list);
		list_del(q.next);

519
		if (cpumask_empty(cp->cpus_allowed))
520 521 522 523 524
			continue;

		if (is_sched_load_balance(cp))
			update_domain_attr(dattr, cp);

525 526
		rcu_read_lock();
		cpuset_for_each_child(child, cont, cp)
527
			list_add_tail(&child->stack_list, &q);
528
		rcu_read_unlock();
529 530 531
	}
}

Paul Jackson's avatar
Paul Jackson committed
532
/*
533 534 535 536 537 538 539 540 541
 * generate_sched_domains()
 *
 * This function builds a partial partition of the systems CPUs
 * A 'partial partition' is a set of non-overlapping subsets whose
 * union is a subset of that set.
 * The output of this function needs to be passed to kernel/sched.c
 * partition_sched_domains() routine, which will rebuild the scheduler's
 * load balancing domains (sched domains) as specified by that partial
 * partition.
Paul Jackson's avatar
Paul Jackson committed
542
 *
543
 * See "What is sched_load_balance" in Documentation/cgroups/cpusets.txt
Paul Jackson's avatar
Paul Jackson committed
544 545 546 547 548 549 550
 * for a background explanation of this.
 *
 * Does not return errors, on the theory that the callers of this
 * routine would rather not worry about failures to rebuild sched
 * domains when operating in the severe memory shortage situations
 * that could cause allocation failures below.
 *
551
 * Must be called with cgroup_lock held.
Paul Jackson's avatar
Paul Jackson committed
552 553
 *
 * The three key local variables below are:
554
 *    q  - a linked-list queue of cpuset pointers, used to implement a
Paul Jackson's avatar
Paul Jackson committed
555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585
 *	   top-down scan of all cpusets.  This scan loads a pointer
 *	   to each cpuset marked is_sched_load_balance into the
 *	   array 'csa'.  For our purposes, rebuilding the schedulers
 *	   sched domains, we can ignore !is_sched_load_balance cpusets.
 *  csa  - (for CpuSet Array) Array of pointers to all the cpusets
 *	   that need to be load balanced, for convenient iterative
 *	   access by the subsequent code that finds the best partition,
 *	   i.e the set of domains (subsets) of CPUs such that the
 *	   cpus_allowed of every cpuset marked is_sched_load_balance
 *	   is a subset of one of these domains, while there are as
 *	   many such domains as possible, each as small as possible.
 * doms  - Conversion of 'csa' to an array of cpumasks, for passing to
 *	   the kernel/sched.c routine partition_sched_domains() in a
 *	   convenient format, that can be easily compared to the prior
 *	   value to determine what partition elements (sched domains)
 *	   were changed (added or removed.)
 *
 * Finding the best partition (set of domains):
 *	The triple nested loops below over i, j, k scan over the
 *	load balanced cpusets (using the array of cpuset pointers in
 *	csa[]) looking for pairs of cpusets that have overlapping
 *	cpus_allowed, but which don't have the same 'pn' partition
 *	number and gives them in the same partition number.  It keeps
 *	looping on the 'restart' label until it can no longer find
 *	any such pairs.
 *
 *	The union of the cpus_allowed masks from the set of
 *	all cpusets having the same 'pn' value then form the one
 *	element of the partition (one sched domain) to be passed to
 *	partition_sched_domains().
 */
586
static int generate_sched_domains(cpumask_var_t **domains,
587
			struct sched_domain_attr **attributes)
Paul Jackson's avatar
Paul Jackson committed
588
{
589
	LIST_HEAD(q);		/* queue of cpusets to be scanned */
Paul Jackson's avatar
Paul Jackson committed
590 591 592 593
	struct cpuset *cp;	/* scans q */
	struct cpuset **csa;	/* array of all cpuset ptrs */
	int csn;		/* how many cpuset ptrs in csa so far */
	int i, j, k;		/* indices for partition finding loops */
594
	cpumask_var_t *doms;	/* resulting partition; i.e. sched domains */
595
	struct sched_domain_attr *dattr;  /* attributes for custom domains */
596
	int ndoms = 0;		/* number of sched domains in result */
597
	int nslot;		/* next empty doms[] struct cpumask slot */
Paul Jackson's avatar
Paul Jackson committed
598 599

	doms = NULL;
600
	dattr = NULL;
601
	csa = NULL;
Paul Jackson's avatar
Paul Jackson committed
602 603 604

	/* Special case for the 99% of systems with one, full, sched domain */
	if (is_sched_load_balance(&top_cpuset)) {
605 606
		ndoms = 1;
		doms = alloc_sched_domains(ndoms);
Paul Jackson's avatar
Paul Jackson committed
607
		if (!doms)
608 609
			goto done;

610 611 612
		dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL);
		if (dattr) {
			*dattr = SD_ATTR_INIT;
613
			update_domain_attr_tree(dattr, &top_cpuset);
614
		}
615
		cpumask_copy(doms[0], top_cpuset.cpus_allowed);
616 617

		goto done;
Paul Jackson's avatar
Paul Jackson committed
618 619 620 621 622 623 624
	}

	csa = kmalloc(number_of_cpusets * sizeof(cp), GFP_KERNEL);
	if (!csa)
		goto done;
	csn = 0;

625 626
	list_add(&top_cpuset.stack_list, &q);
	while (!list_empty(&q)) {
Paul Jackson's avatar
Paul Jackson committed
627 628
		struct cgroup *cont;
		struct cpuset *child;   /* scans child cpusets of cp */
629

630 631 632
		cp = list_first_entry(&q, struct cpuset, stack_list);
		list_del(q.next);

633
		if (cpumask_empty(cp->cpus_allowed))
634 635
			continue;

636 637 638 639 640 641 642
		/*
		 * All child cpusets contain a subset of the parent's cpus, so
		 * just skip them, and then we call update_domain_attr_tree()
		 * to calc relax_domain_level of the corresponding sched
		 * domain.
		 */
		if (is_sched_load_balance(cp)) {
Paul Jackson's avatar
Paul Jackson committed
643
			csa[csn++] = cp;
644 645
			continue;
		}
646

647 648
		rcu_read_lock();
		cpuset_for_each_child(child, cont, cp)
649
			list_add_tail(&child->stack_list, &q);
650
		rcu_read_unlock();
Paul Jackson's avatar
Paul Jackson committed
651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679
  	}

	for (i = 0; i < csn; i++)
		csa[i]->pn = i;
	ndoms = csn;

restart:
	/* Find the best partition (set of sched domains) */
	for (i = 0; i < csn; i++) {
		struct cpuset *a = csa[i];
		int apn = a->pn;

		for (j = 0; j < csn; j++) {
			struct cpuset *b = csa[j];
			int bpn = b->pn;

			if (apn != bpn && cpusets_overlap(a, b)) {
				for (k = 0; k < csn; k++) {
					struct cpuset *c = csa[k];

					if (c->pn == bpn)
						c->pn = apn;
				}
				ndoms--;	/* one less element */
				goto restart;
			}
		}
	}

680 681 682 683
	/*
	 * Now we know how many domains to create.
	 * Convert <csn, csa> to <ndoms, doms> and populate cpu masks.
	 */
684
	doms = alloc_sched_domains(ndoms);
685
	if (!doms)
686 687 688 689 690 691
		goto done;

	/*
	 * The rest of the code, including the scheduler, can deal with
	 * dattr==NULL case. No need to abort if alloc fails.
	 */
692
	dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL);
Paul Jackson's avatar
Paul Jackson committed
693 694 695

	for (nslot = 0, i = 0; i < csn; i++) {
		struct cpuset *a = csa[i];
696
		struct cpumask *dp;
Paul Jackson's avatar
Paul Jackson committed
697 698
		int apn = a->pn;

699 700 701 702 703
		if (apn < 0) {
			/* Skip completed partitions */
			continue;
		}

704
		dp = doms[nslot];
705 706 707 708 709 710 711 712 713 714

		if (nslot == ndoms) {
			static int warnings = 10;
			if (warnings) {
				printk(KERN_WARNING
				 "rebuild_sched_domains confused:"
				  " nslot %d, ndoms %d, csn %d, i %d,"
				  " apn %d\n",
				  nslot, ndoms, csn, i, apn);
				warnings--;
Paul Jackson's avatar
Paul Jackson committed
715
			}
716 717
			continue;
		}
Paul Jackson's avatar
Paul Jackson committed
718

719
		cpumask_clear(dp);
720 721 722 723 724 725
		if (dattr)
			*(dattr + nslot) = SD_ATTR_INIT;
		for (j = i; j < csn; j++) {
			struct cpuset *b = csa[j];

			if (apn == b->pn) {
726
				cpumask_or(dp, dp, b->cpus_allowed);
727 728 729 730 731
				if (dattr)
					update_domain_attr_tree(dattr + nslot, b);

				/* Done with this partition */
				b->pn = -1;
Paul Jackson's avatar
Paul Jackson committed
732 733
			}
		}
734
		nslot++;
Paul Jackson's avatar
Paul Jackson committed
735 736 737
	}
	BUG_ON(nslot != ndoms);

738 739 740
done:
	kfree(csa);

741 742 743 744 745 746 747
	/*
	 * Fallback to the default domain if kmalloc() failed.
	 * See comments in partition_sched_domains().
	 */
	if (doms == NULL)
		ndoms = 1;

748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765
	*domains    = doms;
	*attributes = dattr;
	return ndoms;
}

/*
 * Rebuild scheduler domains.
 *
 * Call with neither cgroup_mutex held nor within get_online_cpus().
 * Takes both cgroup_mutex and get_online_cpus().
 *
 * Cannot be directly called from cpuset code handling changes
 * to the cpuset pseudo-filesystem, because it cannot be called
 * from code that already holds cgroup_mutex.
 */
static void do_rebuild_sched_domains(struct work_struct *unused)
{
	struct sched_domain_attr *attr;
766
	cpumask_var_t *doms;
767 768
	int ndoms;

769
	get_online_cpus();
770 771 772 773 774 775 776 777 778

	/* Generate domain masks and attrs */
	cgroup_lock();
	ndoms = generate_sched_domains(&doms, &attr);
	cgroup_unlock();

	/* Have scheduler rebuild the domains */
	partition_sched_domains(ndoms, doms, attr);

779
	put_online_cpus();
780
}
781 782 783 784 785
#else /* !CONFIG_SMP */
static void do_rebuild_sched_domains(struct work_struct *unused)
{
}

786
static int generate_sched_domains(cpumask_var_t **domains,
787 788 789 790 791 792
			struct sched_domain_attr **attributes)
{
	*domains = NULL;
	return 1;
}
#endif /* CONFIG_SMP */
Paul Jackson's avatar
Paul Jackson committed
793

794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816
static DECLARE_WORK(rebuild_sched_domains_work, do_rebuild_sched_domains);

/*
 * Rebuild scheduler domains, asynchronously via workqueue.
 *
 * If the flag 'sched_load_balance' of any cpuset with non-empty
 * 'cpus' changes, or if the 'cpus' allowed changes in any cpuset
 * which has that flag enabled, or if any cpuset with a non-empty
 * 'cpus' is removed, then call this routine to rebuild the
 * scheduler's dynamic sched domains.
 *
 * The rebuild_sched_domains() and partition_sched_domains()
 * routines must nest cgroup_lock() inside get_online_cpus(),
 * but such cpuset changes as these must nest that locking the
 * other way, holding cgroup_lock() for much of the code.
 *
 * So in order to avoid an ABBA deadlock, the cpuset code handling
 * these user changes delegates the actual sched domain rebuilding
 * to a separate workqueue thread, which ends up processing the
 * above do_rebuild_sched_domains() function.
 */
static void async_rebuild_sched_domains(void)
{
817
	queue_work(cpuset_wq, &rebuild_sched_domains_work);
818 819 820 821 822 823 824 825 826 827 828 829 830 831
}

/*
 * Accomplishes the same scheduler domain rebuild as the above
 * async_rebuild_sched_domains(), however it directly calls the
 * rebuild routine synchronously rather than calling it via an
 * asynchronous work thread.
 *
 * This can only be called from code that is not holding
 * cgroup_mutex (not nested in a cgroup_lock() call.)
 */
void rebuild_sched_domains(void)
{
	do_rebuild_sched_domains(NULL);
Paul Jackson's avatar
Paul Jackson committed
832 833
}

834 835 836 837 838
/**
 * cpuset_test_cpumask - test a task's cpus_allowed versus its cpuset's
 * @tsk: task to test
 * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner
 *
839
 * Call with cgroup_mutex held.  May take callback_mutex during call.
840 841 842
 * Called for each task in a cgroup by cgroup_scan_tasks().
 * Return nonzero if this tasks's cpus_allowed mask should be changed (in other
 * words, if its mask is not equal to its cpuset's mask).
843
 */
844 845
static int cpuset_test_cpumask(struct task_struct *tsk,
			       struct cgroup_scanner *scan)
846
{
847
	return !cpumask_equal(&tsk->cpus_allowed,
848 849
			(cgroup_cs(scan->cg))->cpus_allowed);
}
850

851 852 853 854 855 856 857 858 859 860 861
/**
 * cpuset_change_cpumask - make a task's cpus_allowed the same as its cpuset's
 * @tsk: task to test
 * @scan: struct cgroup_scanner containing the cgroup of the task
 *
 * Called by cgroup_scan_tasks() for each task in a cgroup whose
 * cpus_allowed mask needs to be changed.
 *
 * We don't need to re-check for the cgroup/cpuset membership, since we're
 * holding cgroup_lock() at this point.
 */
862 863
static void cpuset_change_cpumask(struct task_struct *tsk,
				  struct cgroup_scanner *scan)
864
{
865
	set_cpus_allowed_ptr(tsk, ((cgroup_cs(scan->cg))->cpus_allowed));
866 867
}

868 869 870
/**
 * update_tasks_cpumask - Update the cpumasks of tasks in the cpuset.
 * @cs: the cpuset in which each task's cpus_allowed mask needs to be changed
871
 * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks()
872 873 874 875 876 877
 *
 * Called with cgroup_mutex held
 *
 * The cgroup_scan_tasks() function will scan all the tasks in a cgroup,
 * calling callback functions for each.
 *
878 879
 * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0
 * if @heap != NULL.
880
 */
881
static void update_tasks_cpumask(struct cpuset *cs, struct ptr_heap *heap)
882 883 884 885 886 887
{
	struct cgroup_scanner scan;

	scan.cg = cs->css.cgroup;
	scan.test_task = cpuset_test_cpumask;
	scan.process_task = cpuset_change_cpumask;
888 889
	scan.heap = heap;
	cgroup_scan_tasks(&scan);
890 891
}

892 893 894 895 896
/**
 * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it
 * @cs: the cpuset to consider
 * @buf: buffer of cpu numbers written to this cpuset
 */
897 898
static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
			  const char *buf)
Linus Torvalds's avatar
Linus Torvalds committed
899
{
900
	struct ptr_heap heap;
901 902
	int retval;
	int is_load_balanced;
Linus Torvalds's avatar
Linus Torvalds committed
903

904
	/* top_cpuset.cpus_allowed tracks cpu_online_mask; it's read-only */
905 906 907
	if (cs == &top_cpuset)
		return -EACCES;

908
	/*
909
	 * An empty cpus_allowed is ok only if the cpuset has no tasks.
910 911 912
	 * Since cpulist_parse() fails on an empty mask, we special case
	 * that parsing.  The validate_change() call ensures that cpusets
	 * with tasks have cpus.
913
	 */
914
	if (!*buf) {
915
		cpumask_clear(trialcs->cpus_allowed);
916
	} else {
917
		retval = cpulist_parse(buf, trialcs->cpus_allowed);
918 919
		if (retval < 0)
			return retval;
920

921
		if (!cpumask_subset(trialcs->cpus_allowed, cpu_active_mask))
922
			return -EINVAL;
923
	}
924
	retval = validate_change(cs, trialcs);
925 926
	if (retval < 0)
		return retval;
Paul Jackson's avatar
Paul Jackson committed
927

Paul Menage's avatar
Paul Menage committed
928
	/* Nothing to do if the cpus didn't change */
929
	if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed))
Paul Menage's avatar
Paul Menage committed
930
		return 0;
931

932 933 934 935
	retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL);
	if (retval)
		return retval;

936
	is_load_balanced = is_sched_load_balance(trialcs);
Paul Jackson's avatar
Paul Jackson committed
937

938
	mutex_lock(&callback_mutex);
939
	cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed);
940
	mutex_unlock(&callback_mutex);
Paul Jackson's avatar
Paul Jackson committed
941

Paul Menage's avatar
Paul Menage committed
942 943
	/*
	 * Scan tasks in the cpuset, and update the cpumasks of any
944
	 * that need an update.
Paul Menage's avatar
Paul Menage committed
945
	 */
946 947 948
	update_tasks_cpumask(cs, &heap);

	heap_free(&heap);
949

Paul Menage's avatar
Paul Menage committed
950
	if (is_load_balanced)
951
		async_rebuild_sched_domains();
952
	return 0;
Linus Torvalds's avatar
Linus Torvalds committed
953 954
}

955 956 957 958 959 960 961 962
/*
 * cpuset_migrate_mm
 *
 *    Migrate memory region from one set of nodes to another.
 *
 *    Temporarilly set tasks mems_allowed to target nodes of migration,
 *    so that the migration code can allocate pages on these nodes.
 *
963
 *    Call holding cgroup_mutex, so current's cpuset won't change
964
 *    during this call, as manage_mutex holds off any cpuset_attach()
965 966
 *    calls.  Therefore we don't need to take task_lock around the
 *    call to guarantee_online_mems(), as we know no one is changing
967
 *    our task's cpuset.
968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983
 *
 *    While the mm_struct we are migrating is typically from some
 *    other task, the task_struct mems_allowed that we are hacking
 *    is for our current task, which must allocate new pages for that
 *    migrating memory region.
 */

static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from,
							const nodemask_t *to)
{
	struct task_struct *tsk = current;

	tsk->mems_allowed = *to;

	do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL);

984
	guarantee_online_mems(task_cs(tsk),&tsk->mems_allowed);
985 986
}

987
/*
988 989 990 991 992 993 994 995 996 997 998
 * cpuset_change_task_nodemask - change task's mems_allowed and mempolicy
 * @tsk: the task to change
 * @newmems: new nodes that the task will be set
 *
 * In order to avoid seeing no nodes if the old and new nodes are disjoint,
 * we structure updates as setting all new allowed nodes, then clearing newly
 * disallowed ones.
 */
static void cpuset_change_task_nodemask(struct task_struct *tsk,
					nodemask_t *newmems)
{
999
	bool need_loop;
1000

1001 1002 1003 1004 1005 1006 1007 1008 1009 1010
	/*
	 * Allow tasks that have access to memory reserves because they have
	 * been OOM killed to get memory anywhere.
	 */
	if (unlikely(test_thread_flag(TIF_MEMDIE)))
		return;
	if (current->flags & PF_EXITING) /* Let dying task have memory */
		return;

	task_lock(tsk);
1011 1012 1013 1014 1015 1016 1017 1018
	/*
	 * Determine if a loop is necessary if another thread is doing
	 * get_mems_allowed().  If at least one node remains unchanged and
	 * tsk does not have a mempolicy, then an empty nodemask will not be
	 * possible when mems_allowed is larger than a word.
	 */
	need_loop = task_has_mempolicy(tsk) ||
			!nodes_intersects(*newmems, tsk->mems_allowed);
1019

1020 1021
	if (need_loop)
		write_seqcount_begin(&tsk->mems_allowed_seq);
1022

1023 1024
	nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems);
	mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP1);
1025 1026

	mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP2);
1027
	tsk->mems_allowed = *newmems;
1028 1029 1030 1031

	if (need_loop)
		write_seqcount_end(&tsk->mems_allowed_seq);

1032
	task_unlock(tsk);
1033 1034 1035 1036 1037 1038
}

/*
 * Update task's mems_allowed and rebind its mempolicy and vmas' mempolicy
 * of it to cpuset's new mems_allowed, and migrate pages to new nodes if
 * memory_migrate flag is set. Called with cgroup_mutex held.
1039 1040 1041 1042 1043 1044 1045 1046
 */
static void cpuset_change_nodemask(struct task_struct *p,
				   struct cgroup_scanner *scan)
{
	struct mm_struct *mm;
	struct cpuset *cs;
	int migrate;
	const nodemask_t *oldmem = scan->data;
1047
	static nodemask_t newmems;	/* protected by cgroup_mutex */
1048 1049

	cs = cgroup_cs(scan->cg);
1050
	guarantee_online_mems(cs, &newmems);
1051

1052
	cpuset_change_task_nodemask(p, &newmems);
1053

1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065
	mm = get_task_mm(p);
	if (!mm)
		return;

	migrate = is_memory_migrate(cs);

	mpol_rebind_mm(mm, &cs->mems_allowed);
	if (migrate)
		cpuset_migrate_mm(mm, oldmem, &cs->mems_allowed);
	mmput(mm);
}

1066 1067
static void *cpuset_being_rebound;

1068 1069 1070 1071
/**
 * update_tasks_nodemask - Update the nodemasks of tasks in the cpuset.
 * @cs: the cpuset in which each task's mems_allowed mask needs to be changed
 * @oldmem: old mems_allowed of cpuset cs
1072
 * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks()
1073 1074
 *
 * Called with cgroup_mutex held
1075 1076
 * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0
 * if @heap != NULL.
1077
 */
1078 1079
static void update_tasks_nodemask(struct cpuset *cs, const nodemask_t *oldmem,
				 struct ptr_heap *heap)
Linus Torvalds's avatar
Linus Torvalds committed
1080
{
1081
	struct cgroup_scanner scan;
1082

1083
	cpuset_being_rebound = cs;		/* causes mpol_dup() rebind */
1084

1085 1086 1087
	scan.cg = cs->css.cgroup;
	scan.test_task = NULL;
	scan.process_task = cpuset_change_nodemask;
1088
	scan.heap = heap;
1089
	scan.data = (nodemask_t *)oldmem;
1090 1091

	/*
1092 1093 1094 1095 1096 1097
	 * The mpol_rebind_mm() call takes mmap_sem, which we couldn't
	 * take while holding tasklist_lock.  Forks can happen - the
	 * mpol_dup() cpuset_being_rebound check will catch such forks,
	 * and rebind their vma mempolicies too.  Because we still hold
	 * the global cgroup_mutex, we know that no other rebind effort
	 * will be contending for the global variable cpuset_being_rebound.
1098
	 * It's ok if we rebind the same mm twice; mpol_rebind_mm()
1099
	 * is idempotent.  Also migrate pages in each mm to new nodes.
1100
	 */
1101
	cgroup_scan_tasks(&scan);
1102

1103
	/* We're done rebinding vmas to this cpuset's new mems_allowed. */
1104
	cpuset_being_rebound = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
1105 1106
}

1107 1108 1109
/*
 * Handle user request to change the 'mems' memory placement
 * of a cpuset.  Needs to validate the request, update the
1110 1111 1112 1113
 * cpusets mems_allowed, and for each task in the cpuset,
 * update mems_allowed and rebind task's mempolicy and any vma
 * mempolicies and if the cpuset is marked 'memory_migrate',
 * migrate the tasks pages to the new memory.
1114 1115 1116 1117 1118 1119
 *
 * Call with cgroup_mutex held.  May take callback_mutex during call.
 * Will take tasklist_lock, scan tasklist for tasks in cpuset cs,
 * lock each such tasks mm->mmap_sem, scan its vma's and rebind
 * their mempolicies to the cpusets new mems_allowed.
 */
1120 1121
static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs,
			   const char *buf)
1122
{
1123
	NODEMASK_ALLOC(nodemask_t, oldmem, GFP_KERNEL);
1124
	int retval;
1125
	struct ptr_heap heap;
1126

1127 1128 1129
	if (!oldmem)
		return -ENOMEM;

1130
	/*
1131
	 * top_cpuset.mems_allowed tracks node_stats[N_MEMORY];
1132 1133
	 * it's read-only
	 */