cpuset.c 74 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
} cpuset_flagbits_t;

151 152 153 154 155 156
/* the type of hotplug event */
enum hotplug_event {
	CPUSET_CPU_OFFLINE,
	CPUSET_MEM_OFFLINE,
};

Linus Torvalds's avatar
Linus Torvalds committed
157
/* convenient tests for these bits */
Tejun Heo's avatar
Tejun Heo committed
158 159 160 161 162
static inline bool is_cpuset_online(const struct cpuset *cs)
{
	return test_bit(CS_ONLINE, &cs->flags);
}

Linus Torvalds's avatar
Linus Torvalds committed
163 164
static inline int is_cpu_exclusive(const struct cpuset *cs)
{
165
	return test_bit(CS_CPU_EXCLUSIVE, &cs->flags);
Linus Torvalds's avatar
Linus Torvalds committed
166 167 168 169
}

static inline int is_mem_exclusive(const struct cpuset *cs)
{
170
	return test_bit(CS_MEM_EXCLUSIVE, &cs->flags);
Linus Torvalds's avatar
Linus Torvalds committed
171 172
}

173 174 175 176 177
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
178 179 180 181 182
static inline int is_sched_load_balance(const struct cpuset *cs)
{
	return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
}

183 184
static inline int is_memory_migrate(const struct cpuset *cs)
{
185
	return test_bit(CS_MEMORY_MIGRATE, &cs->flags);
186 187
}

188 189 190 191 192 193 194 195 196 197
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
198
static struct cpuset top_cpuset = {
Tejun Heo's avatar
Tejun Heo committed
199 200
	.flags = ((1 << CS_ONLINE) | (1 << CS_CPU_EXCLUSIVE) |
		  (1 << CS_MEM_EXCLUSIVE)),
Linus Torvalds's avatar
Linus Torvalds committed
201 202 203
};

/*
204 205 206 207 208 209 210
 * 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.
211
 *
212
 * A task must hold both mutexes to modify cpusets.  If a task
213
 * holds cgroup_mutex, then it blocks others wanting that mutex,
214
 * ensuring that it is the only task able to also acquire callback_mutex
215 216
 * and be able to modify cpusets.  It can perform various checks on
 * the cpuset structure first, knowing nothing will change.  It can
217
 * also allocate memory while just holding cgroup_mutex.  While it is
218
 * performing these checks, various callback routines can briefly
219 220
 * acquire callback_mutex to query cpusets.  Once it is ready to make
 * the changes, it takes callback_mutex, blocking everyone else.
221 222
 *
 * Calls to the kernel memory allocator can not be made while holding
223
 * callback_mutex, as that would risk double tripping on callback_mutex
224 225 226
 * from one of the callbacks into the cpuset code from within
 * __alloc_pages().
 *
227
 * If a task is only holding callback_mutex, then it has read-only
228 229
 * access to cpusets.
 *
230 231 232
 * 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.
233
 *
234
 * The cpuset_common_file_read() handlers only hold callback_mutex across
235 236 237
 * small pieces of code, such as when reading out possibly multi-word
 * cpumasks and nodemasks.
 *
238 239
 * 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
240 241
 */

242
static DEFINE_MUTEX(callback_mutex);
243

244 245 246 247 248 249 250 251 252 253 254
/*
 * 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);

255 256
/*
 * This is ugly, but preserves the userspace API for existing cpuset
257
 * users. If someone tries to mount the "cpuset" filesystem, we
258 259
 * silently switch it to mount "cgroup" instead
 */
Al Viro's avatar
Al Viro committed
260 261
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
262
{
263
	struct file_system_type *cgroup_fs = get_fs_type("cgroup");
Al Viro's avatar
Al Viro committed
264
	struct dentry *ret = ERR_PTR(-ENODEV);
265 266 267 268
	if (cgroup_fs) {
		char mountopts[] =
			"cpuset,noprefix,"
			"release_agent=/sbin/cpuset_release_agent";
Al Viro's avatar
Al Viro committed
269 270
		ret = cgroup_fs->mount(cgroup_fs, flags,
					   unused_dev_name, mountopts);
271 272 273
		put_filesystem(cgroup_fs);
	}
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
274 275 276 277
}

static struct file_system_type cpuset_fs_type = {
	.name = "cpuset",
Al Viro's avatar
Al Viro committed
278
	.mount = cpuset_mount,
Linus Torvalds's avatar
Linus Torvalds committed
279 280 281
};

/*
282
 * Return in pmask the portion of a cpusets's cpus_allowed that
Linus Torvalds's avatar
Linus Torvalds committed
283 284 285
 * 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,
286 287
 * 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
288 289
 *
 * One way or another, we guarantee to return some non-empty subset
290
 * of cpu_online_mask.
Linus Torvalds's avatar
Linus Torvalds committed
291
 *
292
 * Call with callback_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
293 294
 */

295 296
static void guarantee_online_cpus(const struct cpuset *cs,
				  struct cpumask *pmask)
Linus Torvalds's avatar
Linus Torvalds committed
297
{
298
	while (cs && !cpumask_intersects(cs->cpus_allowed, cpu_online_mask))
Linus Torvalds's avatar
Linus Torvalds committed
299 300
		cs = cs->parent;
	if (cs)
301
		cpumask_and(pmask, cs->cpus_allowed, cpu_online_mask);
Linus Torvalds's avatar
Linus Torvalds committed
302
	else
303 304
		cpumask_copy(pmask, cpu_online_mask);
	BUG_ON(!cpumask_intersects(pmask, cpu_online_mask));
Linus Torvalds's avatar
Linus Torvalds committed
305 306 307 308
}

/*
 * Return in *pmask the portion of a cpusets's mems_allowed that
309 310 311
 * 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
312
 * found any online mems, return node_states[N_MEMORY].
Linus Torvalds's avatar
Linus Torvalds committed
313 314
 *
 * One way or another, we guarantee to return some non-empty subset
315
 * of node_states[N_MEMORY].
Linus Torvalds's avatar
Linus Torvalds committed
316
 *
317
 * Call with callback_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
318 319 320 321
 */

static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask)
{
322
	while (cs && !nodes_intersects(cs->mems_allowed,
323
					node_states[N_MEMORY]))
Linus Torvalds's avatar
Linus Torvalds committed
324 325
		cs = cs->parent;
	if (cs)
326
		nodes_and(*pmask, cs->mems_allowed,
327
					node_states[N_MEMORY]);
Linus Torvalds's avatar
Linus Torvalds committed
328
	else
329 330
		*pmask = node_states[N_MEMORY];
	BUG_ON(!nodes_intersects(*pmask, node_states[N_MEMORY]));
Linus Torvalds's avatar
Linus Torvalds committed
331 332
}

333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350
/*
 * 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
351 352 353 354 355
/*
 * 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
356
 * are only set if the other's are set.  Call holding cgroup_mutex.
Linus Torvalds's avatar
Linus Torvalds committed
357 358 359 360
 */

static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q)
{
361
	return	cpumask_subset(p->cpus_allowed, q->cpus_allowed) &&
Linus Torvalds's avatar
Linus Torvalds committed
362 363 364 365 366
		nodes_subset(p->mems_allowed, q->mems_allowed) &&
		is_cpu_exclusive(p) <= is_cpu_exclusive(q) &&
		is_mem_exclusive(p) <= is_mem_exclusive(q);
}

367 368 369 370 371 372
/**
 * 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)
{
373 374 375 376 377 378 379 380 381 382 383 384 385
	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;
386 387 388 389 390 391 392 393
}

/**
 * free_trial_cpuset - free the trial cpuset
 * @trial: the trial cpuset to be freed
 */
static void free_trial_cpuset(struct cpuset *trial)
{
394
	free_cpumask_var(trial->cpus_allowed);
395 396 397
	kfree(trial);
}

Linus Torvalds's avatar
Linus Torvalds committed
398 399 400 401 402 403 404
/*
 * 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
405
 * cgroup_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
406 407 408 409 410 411 412 413 414 415 416 417 418 419
 *
 * '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)
{
420
	struct cgroup *cont;
Linus Torvalds's avatar
Linus Torvalds committed
421 422 423
	struct cpuset *c, *par;

	/* Each of our child cpusets must be a subset of us */
424 425
	list_for_each_entry(cont, &cur->css.cgroup->children, sibling) {
		if (!is_cpuset_subset(cgroup_cs(cont), trial))
Linus Torvalds's avatar
Linus Torvalds committed
426 427 428 429
			return -EBUSY;
	}

	/* Remaining checks don't apply to root cpuset */
430
	if (cur == &top_cpuset)
Linus Torvalds's avatar
Linus Torvalds committed
431 432
		return 0;

433 434
	par = cur->parent;

Linus Torvalds's avatar
Linus Torvalds committed
435 436 437 438
	/* We must be a subset of our parent cpuset */
	if (!is_cpuset_subset(trial, par))
		return -EACCES;

439 440 441 442
	/*
	 * If either I or some sibling (!= me) is exclusive, we can't
	 * overlap
	 */
443 444
	list_for_each_entry(cont, &par->css.cgroup->children, sibling) {
		c = cgroup_cs(cont);
Linus Torvalds's avatar
Linus Torvalds committed
445 446
		if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) &&
		    c != cur &&
447
		    cpumask_intersects(trial->cpus_allowed, c->cpus_allowed))
Linus Torvalds's avatar
Linus Torvalds committed
448 449 450 451 452 453 454
			return -EINVAL;
		if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) &&
		    c != cur &&
		    nodes_intersects(trial->mems_allowed, c->mems_allowed))
			return -EINVAL;
	}

455 456
	/* Cpusets with tasks can't have empty cpus_allowed or mems_allowed */
	if (cgroup_task_count(cur->css.cgroup)) {
457
		if (cpumask_empty(trial->cpus_allowed) ||
458 459 460 461 462
		    nodes_empty(trial->mems_allowed)) {
			return -ENOSPC;
		}
	}

Linus Torvalds's avatar
Linus Torvalds committed
463 464 465
	return 0;
}

466
#ifdef CONFIG_SMP
Paul Jackson's avatar
Paul Jackson committed
467
/*
468
 * Helper routine for generate_sched_domains().
Paul Jackson's avatar
Paul Jackson committed
469 470 471 472
 * Do cpusets a, b have overlapping cpus_allowed masks?
 */
static int cpusets_overlap(struct cpuset *a, struct cpuset *b)
{
473
	return cpumask_intersects(a->cpus_allowed, b->cpus_allowed);
Paul Jackson's avatar
Paul Jackson committed
474 475
}

476 477 478 479 480 481 482 483
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;
}

484 485 486 487 488 489 490 491 492 493 494 495 496 497
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);

498
		if (cpumask_empty(cp->cpus_allowed))
499 500 501 502 503 504 505 506 507 508 509 510
			continue;

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

		list_for_each_entry(cont, &cp->css.cgroup->children, sibling) {
			child = cgroup_cs(cont);
			list_add_tail(&child->stack_list, &q);
		}
	}
}

Paul Jackson's avatar
Paul Jackson committed
511
/*
512 513 514 515 516 517 518 519 520
 * 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
521
 *
522
 * See "What is sched_load_balance" in Documentation/cgroups/cpusets.txt
Paul Jackson's avatar
Paul Jackson committed
523 524 525 526 527 528 529
 * 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.
 *
530
 * Must be called with cgroup_lock held.
Paul Jackson's avatar
Paul Jackson committed
531 532
 *
 * The three key local variables below are:
533
 *    q  - a linked-list queue of cpuset pointers, used to implement a
Paul Jackson's avatar
Paul Jackson committed
534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564
 *	   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().
 */
565
static int generate_sched_domains(cpumask_var_t **domains,
566
			struct sched_domain_attr **attributes)
Paul Jackson's avatar
Paul Jackson committed
567
{
568
	LIST_HEAD(q);		/* queue of cpusets to be scanned */
Paul Jackson's avatar
Paul Jackson committed
569 570 571 572
	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 */
573
	cpumask_var_t *doms;	/* resulting partition; i.e. sched domains */
574
	struct sched_domain_attr *dattr;  /* attributes for custom domains */
575
	int ndoms = 0;		/* number of sched domains in result */
576
	int nslot;		/* next empty doms[] struct cpumask slot */
Paul Jackson's avatar
Paul Jackson committed
577 578

	doms = NULL;
579
	dattr = NULL;
580
	csa = NULL;
Paul Jackson's avatar
Paul Jackson committed
581 582 583

	/* Special case for the 99% of systems with one, full, sched domain */
	if (is_sched_load_balance(&top_cpuset)) {
584 585
		ndoms = 1;
		doms = alloc_sched_domains(ndoms);
Paul Jackson's avatar
Paul Jackson committed
586
		if (!doms)
587 588
			goto done;

589 590 591
		dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL);
		if (dattr) {
			*dattr = SD_ATTR_INIT;
592
			update_domain_attr_tree(dattr, &top_cpuset);
593
		}
594
		cpumask_copy(doms[0], top_cpuset.cpus_allowed);
595 596

		goto done;
Paul Jackson's avatar
Paul Jackson committed
597 598 599 600 601 602 603
	}

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

604 605
	list_add(&top_cpuset.stack_list, &q);
	while (!list_empty(&q)) {
Paul Jackson's avatar
Paul Jackson committed
606 607
		struct cgroup *cont;
		struct cpuset *child;   /* scans child cpusets of cp */
608

609 610 611
		cp = list_first_entry(&q, struct cpuset, stack_list);
		list_del(q.next);

612
		if (cpumask_empty(cp->cpus_allowed))
613 614
			continue;

615 616 617 618 619 620 621
		/*
		 * 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
622
			csa[csn++] = cp;
623 624
			continue;
		}
625

Paul Jackson's avatar
Paul Jackson committed
626 627
		list_for_each_entry(cont, &cp->css.cgroup->children, sibling) {
			child = cgroup_cs(cont);
628
			list_add_tail(&child->stack_list, &q);
Paul Jackson's avatar
Paul Jackson committed
629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658
		}
  	}

	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;
			}
		}
	}

659 660 661 662
	/*
	 * Now we know how many domains to create.
	 * Convert <csn, csa> to <ndoms, doms> and populate cpu masks.
	 */
663
	doms = alloc_sched_domains(ndoms);
664
	if (!doms)
665 666 667 668 669 670
		goto done;

	/*
	 * The rest of the code, including the scheduler, can deal with
	 * dattr==NULL case. No need to abort if alloc fails.
	 */
671
	dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL);
Paul Jackson's avatar
Paul Jackson committed
672 673 674

	for (nslot = 0, i = 0; i < csn; i++) {
		struct cpuset *a = csa[i];
675
		struct cpumask *dp;
Paul Jackson's avatar
Paul Jackson committed
676 677
		int apn = a->pn;

678 679 680 681 682
		if (apn < 0) {
			/* Skip completed partitions */
			continue;
		}

683
		dp = doms[nslot];
684 685 686 687 688 689 690 691 692 693

		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
694
			}
695 696
			continue;
		}
Paul Jackson's avatar
Paul Jackson committed
697

698
		cpumask_clear(dp);
699 700 701 702 703 704
		if (dattr)
			*(dattr + nslot) = SD_ATTR_INIT;
		for (j = i; j < csn; j++) {
			struct cpuset *b = csa[j];

			if (apn == b->pn) {
705
				cpumask_or(dp, dp, b->cpus_allowed);
706 707 708 709 710
				if (dattr)
					update_domain_attr_tree(dattr + nslot, b);

				/* Done with this partition */
				b->pn = -1;
Paul Jackson's avatar
Paul Jackson committed
711 712
			}
		}
713
		nslot++;
Paul Jackson's avatar
Paul Jackson committed
714 715 716
	}
	BUG_ON(nslot != ndoms);

717 718 719
done:
	kfree(csa);

720 721 722 723 724 725 726
	/*
	 * Fallback to the default domain if kmalloc() failed.
	 * See comments in partition_sched_domains().
	 */
	if (doms == NULL)
		ndoms = 1;

727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744
	*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;
745
	cpumask_var_t *doms;
746 747
	int ndoms;

748
	get_online_cpus();
749 750 751 752 753 754 755 756 757

	/* 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);

758
	put_online_cpus();
759
}
760 761 762 763 764
#else /* !CONFIG_SMP */
static void do_rebuild_sched_domains(struct work_struct *unused)
{
}

765
static int generate_sched_domains(cpumask_var_t **domains,
766 767 768 769 770 771
			struct sched_domain_attr **attributes)
{
	*domains = NULL;
	return 1;
}
#endif /* CONFIG_SMP */
Paul Jackson's avatar
Paul Jackson committed
772

773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795
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)
{
796
	queue_work(cpuset_wq, &rebuild_sched_domains_work);
797 798 799 800 801 802 803 804 805 806 807 808 809 810
}

/*
 * 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
811 812
}

813 814 815 816 817
/**
 * 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
 *
818
 * Call with cgroup_mutex held.  May take callback_mutex during call.
819 820 821
 * 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).
822
 */
823 824
static int cpuset_test_cpumask(struct task_struct *tsk,
			       struct cgroup_scanner *scan)
825
{
826
	return !cpumask_equal(&tsk->cpus_allowed,
827 828
			(cgroup_cs(scan->cg))->cpus_allowed);
}
829

830 831 832 833 834 835 836 837 838 839 840
/**
 * 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.
 */
841 842
static void cpuset_change_cpumask(struct task_struct *tsk,
				  struct cgroup_scanner *scan)
843
{
844
	set_cpus_allowed_ptr(tsk, ((cgroup_cs(scan->cg))->cpus_allowed));
845 846
}

847 848 849
/**
 * 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
850
 * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks()
851 852 853 854 855 856
 *
 * Called with cgroup_mutex held
 *
 * The cgroup_scan_tasks() function will scan all the tasks in a cgroup,
 * calling callback functions for each.
 *
857 858
 * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0
 * if @heap != NULL.
859
 */
860
static void update_tasks_cpumask(struct cpuset *cs, struct ptr_heap *heap)
861 862 863 864 865 866
{
	struct cgroup_scanner scan;

	scan.cg = cs->css.cgroup;
	scan.test_task = cpuset_test_cpumask;
	scan.process_task = cpuset_change_cpumask;
867 868
	scan.heap = heap;
	cgroup_scan_tasks(&scan);
869 870
}

871 872 873 874 875
/**
 * 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
 */
876 877
static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
			  const char *buf)
Linus Torvalds's avatar
Linus Torvalds committed
878
{
879
	struct ptr_heap heap;
880 881
	int retval;
	int is_load_balanced;
Linus Torvalds's avatar
Linus Torvalds committed
882

883
	/* top_cpuset.cpus_allowed tracks cpu_online_mask; it's read-only */
884 885 886
	if (cs == &top_cpuset)
		return -EACCES;

887
	/*
888
	 * An empty cpus_allowed is ok only if the cpuset has no tasks.
889 890 891
	 * Since cpulist_parse() fails on an empty mask, we special case
	 * that parsing.  The validate_change() call ensures that cpusets
	 * with tasks have cpus.
892
	 */
893
	if (!*buf) {
894
		cpumask_clear(trialcs->cpus_allowed);
895
	} else {
896
		retval = cpulist_parse(buf, trialcs->cpus_allowed);
897 898
		if (retval < 0)
			return retval;
899

900
		if (!cpumask_subset(trialcs->cpus_allowed, cpu_active_mask))
901
			return -EINVAL;
902
	}
903
	retval = validate_change(cs, trialcs);
904 905
	if (retval < 0)
		return retval;
Paul Jackson's avatar
Paul Jackson committed
906

Paul Menage's avatar
Paul Menage committed
907
	/* Nothing to do if the cpus didn't change */
908
	if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed))
Paul Menage's avatar
Paul Menage committed
909
		return 0;
910

911 912 913 914
	retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL);
	if (retval)
		return retval;

915
	is_load_balanced = is_sched_load_balance(trialcs);
Paul Jackson's avatar
Paul Jackson committed
916

917
	mutex_lock(&callback_mutex);
918
	cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed);
919
	mutex_unlock(&callback_mutex);
Paul Jackson's avatar
Paul Jackson committed
920

Paul Menage's avatar
Paul Menage committed
921 922
	/*
	 * Scan tasks in the cpuset, and update the cpumasks of any
923
	 * that need an update.
Paul Menage's avatar
Paul Menage committed
924
	 */
925 926 927
	update_tasks_cpumask(cs, &heap);

	heap_free(&heap);
928

Paul Menage's avatar
Paul Menage committed
929
	if (is_load_balanced)
930
		async_rebuild_sched_domains();
931
	return 0;
Linus Torvalds's avatar
Linus Torvalds committed
932 933
}

934 935 936 937 938 939 940 941
/*
 * 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.
 *
942
 *    Call holding cgroup_mutex, so current's cpuset won't change
943
 *    during this call, as manage_mutex holds off any cpuset_attach()
944 945
 *    calls.  Therefore we don't need to take task_lock around the
 *    call to guarantee_online_mems(), as we know no one is changing
946
 *    our task's cpuset.
947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962
 *
 *    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);

963
	guarantee_online_mems(task_cs(tsk),&tsk->mems_allowed);
964 965
}

966
/*
967 968 969 970 971 972 973 974 975 976 977
 * 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)
{
978
	bool need_loop;
979

980 981 982 983 984 985 986 987 988 989
	/*
	 * 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);
990 991 992 993 994 995 996 997
	/*
	 * 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);
998

999 1000
	if (need_loop)
		write_seqcount_begin(&tsk->mems_allowed_seq);
1001

1002 1003
	nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems);
	mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP1);
1004 1005

	mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP2);
1006
	tsk->mems_allowed = *newmems;
1007 1008 1009 1010

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

1011
	task_unlock(tsk);
1012 1013 1014 1015 1016 1017
}

/*
 * 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.
1018 1019 1020 1021 1022 1023 1024 1025
 */
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;
1026
	static nodemask_t newmems;	/* protected by cgroup_mutex */
1027 1028

	cs = cgroup_cs(scan->cg);
1029
	guarantee_online_mems(cs, &newmems);
1030

1031
	cpuset_change_task_nodemask(p, &newmems);
1032

1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044
	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);
}

1045 1046
static void *cpuset_being_rebound;

1047 1048 1049 1050
/**
 * 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
1051
 * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks()
1052 1053
 *
 * Called with cgroup_mutex held
1054 1055
 * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0
 * if @heap != NULL.
1056
 */
1057 1058
static void update_tasks_nodemask(struct cpuset *cs, const nodemask_t *oldmem,
				 struct ptr_heap *heap)
Linus Torvalds's avatar
Linus Torvalds committed
1059
{
1060
	struct cgroup_scanner scan;
1061

1062
	cpuset_being_rebound = cs;		/* causes mpol_dup() rebind */
1063

1064 1065 1066
	scan.cg = cs->css.cgroup;
	scan.test_task = NULL;
	scan.process_task = cpuset_change_nodemask;
1067
	scan.heap = heap;
1068
	scan.data = (nodemask_t *)oldmem;
1069 1070

	/*
1071 1072 1073 1074 1075 1076
	 * 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.
1077
	 * It's ok if we rebind the same mm twice; mpol_rebind_mm()
1078
	 * is idempotent.  Also migrate pages in each mm to new nodes.
1079
	 */
1080
	cgroup_scan_tasks(&scan);
1081

1082
	/* We're done rebinding vmas to this cpuset's new mems_allowed. */
1083
	cpuset_being_rebound = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
1084 1085
}

1086 1087 1088
/*
 * Handle user request to change the 'mems' memory placement
 * of a cpuset.  Needs to validate the request, update the
1089 1090 1091 1092
 * 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.
1093 1094 1095 1096 1097 1098
 *
 * 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.
 */
1099 1100
static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs,
			   const char *buf)
1101
{
1102
	NODEMASK_ALLOC(nodemask_t, oldmem, GFP_KERNEL);
1103
	int retval;
1104
	struct ptr_heap heap;
1105

1106 1107 1108
	if (!oldmem)
		return -ENOMEM;

1109
	/*
1110
	 * top_cpuset.mems_allowed tracks node_stats[N_MEMORY];
1111 1112
	 * it's read-only
	 */
1113 1114 1115 1116
	if (cs == &top_cpuset) {
		retval = -EACCES;
		goto done;
	}
1117 1118 1119 1120 1121 1122 1123 1124

	/*
	 * An empty mems_allowed is ok iff there are no tasks in the cpuset.
	 * Since nodelist_parse() fails on an empty mask, we special case
	 * that parsing.  The validate_change() call ensures that cpusets
	 * with tasks have memory.
	 */
	if (!*buf) {
1125
		nodes_clear(trialcs->mems_allowed);
1126
	} else {
1127
		retval = nodelist_parse(buf, trialcs->mems_allowed);
1128 1129 1130
		if (retval < 0)
			goto done;

1131
		if (!nodes_subset(trialcs->mems_allowed,
1132
				node_states[N_MEMORY])) {
1133 1134 1135
			retval =  -EINVAL;
			goto done;
		}
1136
	}
1137 1138
	*oldmem = cs->mems_allowed;
	if (nodes_equal(*oldmem, trialcs->mems_allowed)) {
1139 1140 1141
		retval = 0;		/* Too easy - nothing to do */
		goto done;
	}
1142
	retval = validate_change(cs, trialcs);