cpuset.c 76.8 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
/*
 * Tracks how many cpusets are currently defined in system.
 * When there is only one cpuset (the root cpuset) we can
 * short circuit some hooks.
 */
68
int number_of_cpusets __read_mostly;
69

70
/* Forward declare cgroup structures */
71 72 73
struct cgroup_subsys cpuset_subsys;
struct cpuset;

74 75 76 77 78 79 80 81 82
/* 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
83
struct cpuset {
84 85
	struct cgroup_subsys_state css;

Linus Torvalds's avatar
Linus Torvalds committed
86
	unsigned long flags;		/* "unsigned long" so bitops work */
87
	cpumask_var_t cpus_allowed;	/* CPUs allowed to tasks in cpuset */
Linus Torvalds's avatar
Linus Torvalds committed
88 89 90 91
	nodemask_t mems_allowed;	/* Memory Nodes allowed to tasks */

	struct cpuset *parent;		/* my parent */

92
	struct fmeter fmeter;		/* memory_pressure filter */
Paul Jackson's avatar
Paul Jackson committed
93

94 95 96 97 98 99
	/*
	 * Tasks are being attached to this cpuset.  Used to prevent
	 * zeroing cpus/mems_allowed between ->can_attach() and ->attach().
	 */
	int attach_in_progress;

Paul Jackson's avatar
Paul Jackson committed
100 101
	/* partition number for rebuild_sched_domains() */
	int pn;
102

103 104 105
	/* for custom sched domain */
	int relax_domain_level;

106
	struct work_struct hotplug_work;
Linus Torvalds's avatar
Linus Torvalds committed
107 108
};

109 110 111 112 113 114 115 116 117 118 119 120 121 122
/* 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);
}

123 124 125 126 127 128 129 130 131 132 133 134 135
#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
136 137
/* bits in struct cpuset flags field */
typedef enum {
Tejun Heo's avatar
Tejun Heo committed
138
	CS_ONLINE,
Linus Torvalds's avatar
Linus Torvalds committed
139 140
	CS_CPU_EXCLUSIVE,
	CS_MEM_EXCLUSIVE,
141
	CS_MEM_HARDWALL,
142
	CS_MEMORY_MIGRATE,
Paul Jackson's avatar
Paul Jackson committed
143
	CS_SCHED_LOAD_BALANCE,
144 145
	CS_SPREAD_PAGE,
	CS_SPREAD_SLAB,
Linus Torvalds's avatar
Linus Torvalds committed
146 147 148
} cpuset_flagbits_t;

/* convenient tests for these bits */
Tejun Heo's avatar
Tejun Heo committed
149 150 151 152 153
static inline bool is_cpuset_online(const struct cpuset *cs)
{
	return test_bit(CS_ONLINE, &cs->flags);
}

Linus Torvalds's avatar
Linus Torvalds committed
154 155
static inline int is_cpu_exclusive(const struct cpuset *cs)
{
156
	return test_bit(CS_CPU_EXCLUSIVE, &cs->flags);
Linus Torvalds's avatar
Linus Torvalds committed
157 158 159 160
}

static inline int is_mem_exclusive(const struct cpuset *cs)
{
161
	return test_bit(CS_MEM_EXCLUSIVE, &cs->flags);
Linus Torvalds's avatar
Linus Torvalds committed
162 163
}

164 165 166 167 168
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
169 170 171 172 173
static inline int is_sched_load_balance(const struct cpuset *cs)
{
	return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
}

174 175
static inline int is_memory_migrate(const struct cpuset *cs)
{
176
	return test_bit(CS_MEMORY_MIGRATE, &cs->flags);
177 178
}

179 180 181 182 183 184 185 186 187 188
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
189
static struct cpuset top_cpuset = {
Tejun Heo's avatar
Tejun Heo committed
190 191
	.flags = ((1 << CS_ONLINE) | (1 << CS_CPU_EXCLUSIVE) |
		  (1 << CS_MEM_EXCLUSIVE)),
Linus Torvalds's avatar
Linus Torvalds committed
192 193
};

194 195 196 197 198 199 200 201 202 203 204 205 206
/**
 * 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)))))

207 208 209 210 211 212 213 214 215 216 217 218 219 220
/**
 * cpuset_for_each_descendant_pre - pre-order walk of a cpuset's descendants
 * @des_cs: loop cursor pointing to the current descendant
 * @pos_cgrp: used for iteration
 * @root_cs: target cpuset to walk ancestor of
 *
 * Walk @des_cs through the online descendants of @root_cs.  Must be used
 * with RCU read locked.  The caller may modify @pos_cgrp by calling
 * cgroup_rightmost_descendant() to skip subtree.
 */
#define cpuset_for_each_descendant_pre(des_cs, pos_cgrp, root_cs)	\
	cgroup_for_each_descendant_pre((pos_cgrp), (root_cs)->css.cgroup) \
		if (is_cpuset_online(((des_cs) = cgroup_cs((pos_cgrp)))))

Linus Torvalds's avatar
Linus Torvalds committed
221
/*
222 223 224 225 226 227 228 229 230 231 232 233 234 235
 * There are two global mutexes guarding cpuset structures - cpuset_mutex
 * and callback_mutex.  The latter may nest inside the former.  We also
 * require taking task_lock() when dereferencing a task's cpuset pointer.
 * See "The task_lock() exception", at the end of this comment.
 *
 * A task must hold both mutexes to modify cpusets.  If a task holds
 * cpuset_mutex, then it blocks others wanting that mutex, ensuring that it
 * is the only task able to also acquire callback_mutex and be able to
 * modify cpusets.  It can perform various checks on the cpuset structure
 * first, knowing nothing will change.  It can also allocate memory while
 * just holding cpuset_mutex.  While it is performing these checks, various
 * callback routines can briefly acquire callback_mutex to query cpusets.
 * Once it is ready to make the changes, it takes callback_mutex, blocking
 * everyone else.
236 237
 *
 * Calls to the kernel memory allocator can not be made while holding
238
 * callback_mutex, as that would risk double tripping on callback_mutex
239 240 241
 * from one of the callbacks into the cpuset code from within
 * __alloc_pages().
 *
242
 * If a task is only holding callback_mutex, then it has read-only
243 244
 * access to cpusets.
 *
245 246 247
 * 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.
248
 *
249
 * The cpuset_common_file_read() handlers only hold callback_mutex across
250 251 252
 * small pieces of code, such as when reading out possibly multi-word
 * cpumasks and nodemasks.
 *
253 254
 * 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
255 256
 */

257
static DEFINE_MUTEX(cpuset_mutex);
258
static DEFINE_MUTEX(callback_mutex);
259

260 261 262 263 264 265 266 267 268 269 270
/*
 * 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);

271 272 273
/*
 * CPU / memory hotplug is handled asynchronously.
 */
274 275
static struct workqueue_struct *cpuset_propagate_hotplug_wq;

276
static void cpuset_hotplug_workfn(struct work_struct *work);
277
static void cpuset_propagate_hotplug_workfn(struct work_struct *work);
278
static void schedule_cpuset_propagate_hotplug(struct cpuset *cs);
279 280 281

static DECLARE_WORK(cpuset_hotplug_work, cpuset_hotplug_workfn);

282 283
/*
 * This is ugly, but preserves the userspace API for existing cpuset
284
 * users. If someone tries to mount the "cpuset" filesystem, we
285 286
 * silently switch it to mount "cgroup" instead
 */
Al Viro's avatar
Al Viro committed
287 288
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
289
{
290
	struct file_system_type *cgroup_fs = get_fs_type("cgroup");
Al Viro's avatar
Al Viro committed
291
	struct dentry *ret = ERR_PTR(-ENODEV);
292 293 294 295
	if (cgroup_fs) {
		char mountopts[] =
			"cpuset,noprefix,"
			"release_agent=/sbin/cpuset_release_agent";
Al Viro's avatar
Al Viro committed
296 297
		ret = cgroup_fs->mount(cgroup_fs, flags,
					   unused_dev_name, mountopts);
298 299 300
		put_filesystem(cgroup_fs);
	}
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
301 302 303 304
}

static struct file_system_type cpuset_fs_type = {
	.name = "cpuset",
Al Viro's avatar
Al Viro committed
305
	.mount = cpuset_mount,
Linus Torvalds's avatar
Linus Torvalds committed
306 307 308
};

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

322 323
static void guarantee_online_cpus(const struct cpuset *cs,
				  struct cpumask *pmask)
Linus Torvalds's avatar
Linus Torvalds committed
324
{
325
	while (cs && !cpumask_intersects(cs->cpus_allowed, cpu_online_mask))
Linus Torvalds's avatar
Linus Torvalds committed
326 327
		cs = cs->parent;
	if (cs)
328
		cpumask_and(pmask, cs->cpus_allowed, cpu_online_mask);
Linus Torvalds's avatar
Linus Torvalds committed
329
	else
330 331
		cpumask_copy(pmask, cpu_online_mask);
	BUG_ON(!cpumask_intersects(pmask, cpu_online_mask));
Linus Torvalds's avatar
Linus Torvalds committed
332 333 334 335
}

/*
 * Return in *pmask the portion of a cpusets's mems_allowed that
336 337 338
 * 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
339
 * found any online mems, return node_states[N_MEMORY].
Linus Torvalds's avatar
Linus Torvalds committed
340 341
 *
 * One way or another, we guarantee to return some non-empty subset
342
 * of node_states[N_MEMORY].
Linus Torvalds's avatar
Linus Torvalds committed
343
 *
344
 * Call with callback_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
345 346 347 348
 */

static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask)
{
349
	while (cs && !nodes_intersects(cs->mems_allowed,
350
					node_states[N_MEMORY]))
Linus Torvalds's avatar
Linus Torvalds committed
351 352
		cs = cs->parent;
	if (cs)
353
		nodes_and(*pmask, cs->mems_allowed,
354
					node_states[N_MEMORY]);
Linus Torvalds's avatar
Linus Torvalds committed
355
	else
356 357
		*pmask = node_states[N_MEMORY];
	BUG_ON(!nodes_intersects(*pmask, node_states[N_MEMORY]));
Linus Torvalds's avatar
Linus Torvalds committed
358 359
}

360 361 362
/*
 * update task's spread flag if cpuset's page/slab spread flag is set
 *
363
 * Called with callback_mutex/cpuset_mutex held
364 365 366 367 368 369 370 371 372 373 374 375 376 377
 */
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
378 379 380 381 382
/*
 * 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
383
 * are only set if the other's are set.  Call holding cpuset_mutex.
Linus Torvalds's avatar
Linus Torvalds committed
384 385 386 387
 */

static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q)
{
388
	return	cpumask_subset(p->cpus_allowed, q->cpus_allowed) &&
Linus Torvalds's avatar
Linus Torvalds committed
389 390 391 392 393
		nodes_subset(p->mems_allowed, q->mems_allowed) &&
		is_cpu_exclusive(p) <= is_cpu_exclusive(q) &&
		is_mem_exclusive(p) <= is_mem_exclusive(q);
}

394 395 396 397 398 399
/**
 * 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)
{
400 401 402 403 404 405 406 407 408 409 410 411 412
	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;
413 414 415 416 417 418 419 420
}

/**
 * free_trial_cpuset - free the trial cpuset
 * @trial: the trial cpuset to be freed
 */
static void free_trial_cpuset(struct cpuset *trial)
{
421
	free_cpumask_var(trial->cpus_allowed);
422 423 424
	kfree(trial);
}

Linus Torvalds's avatar
Linus Torvalds committed
425 426 427 428 429 430 431
/*
 * 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
432
 * cpuset_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
433 434 435 436 437 438 439 440 441 442 443 444 445 446
 *
 * '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)
{
447
	struct cgroup *cont;
Linus Torvalds's avatar
Linus Torvalds committed
448
	struct cpuset *c, *par;
449 450 451
	int ret;

	rcu_read_lock();
Linus Torvalds's avatar
Linus Torvalds committed
452 453

	/* Each of our child cpusets must be a subset of us */
454 455 456 457
	ret = -EBUSY;
	cpuset_for_each_child(c, cont, cur)
		if (!is_cpuset_subset(c, trial))
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
458 459

	/* Remaining checks don't apply to root cpuset */
460
	ret = 0;
461
	if (cur == &top_cpuset)
462
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
463

464 465
	par = cur->parent;

Linus Torvalds's avatar
Linus Torvalds committed
466
	/* We must be a subset of our parent cpuset */
467
	ret = -EACCES;
Linus Torvalds's avatar
Linus Torvalds committed
468
	if (!is_cpuset_subset(trial, par))
469
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
470

471 472 473 474
	/*
	 * If either I or some sibling (!= me) is exclusive, we can't
	 * overlap
	 */
475 476
	ret = -EINVAL;
	cpuset_for_each_child(c, cont, par) {
Linus Torvalds's avatar
Linus Torvalds committed
477 478
		if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) &&
		    c != cur &&
479
		    cpumask_intersects(trial->cpus_allowed, c->cpus_allowed))
480
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
481 482 483
		if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) &&
		    c != cur &&
		    nodes_intersects(trial->mems_allowed, c->mems_allowed))
484
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
485 486
	}

487 488 489 490
	/*
	 * Cpusets with tasks - existing or newly being attached - can't
	 * have empty cpus_allowed or mems_allowed.
	 */
491
	ret = -ENOSPC;
492
	if ((cgroup_task_count(cur->css.cgroup) || cur->attach_in_progress) &&
493 494 495
	    (cpumask_empty(trial->cpus_allowed) ||
	     nodes_empty(trial->mems_allowed)))
		goto out;
496

497 498 499 500
	ret = 0;
out:
	rcu_read_unlock();
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
501 502
}

503
#ifdef CONFIG_SMP
Paul Jackson's avatar
Paul Jackson committed
504
/*
505
 * Helper routine for generate_sched_domains().
Paul Jackson's avatar
Paul Jackson committed
506 507 508 509
 * Do cpusets a, b have overlapping cpus_allowed masks?
 */
static int cpusets_overlap(struct cpuset *a, struct cpuset *b)
{
510
	return cpumask_intersects(a->cpus_allowed, b->cpus_allowed);
Paul Jackson's avatar
Paul Jackson committed
511 512
}

513 514 515 516 517 518 519 520
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;
}

521 522
static void update_domain_attr_tree(struct sched_domain_attr *dattr,
				    struct cpuset *root_cs)
523
{
524 525
	struct cpuset *cp;
	struct cgroup *pos_cgrp;
526

527 528 529 530 531
	rcu_read_lock();
	cpuset_for_each_descendant_pre(cp, pos_cgrp, root_cs) {
		/* skip the whole subtree if @cp doesn't have any CPU */
		if (cpumask_empty(cp->cpus_allowed)) {
			pos_cgrp = cgroup_rightmost_descendant(pos_cgrp);
532
			continue;
533
		}
534 535 536 537

		if (is_sched_load_balance(cp))
			update_domain_attr(dattr, cp);
	}
538
	rcu_read_unlock();
539 540
}

Paul Jackson's avatar
Paul Jackson committed
541
/*
542 543 544 545 546 547 548 549 550
 * 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
551
 *
552
 * See "What is sched_load_balance" in Documentation/cgroups/cpusets.txt
Paul Jackson's avatar
Paul Jackson committed
553 554 555 556 557 558 559
 * 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.
 *
560
 * Must be called with cpuset_mutex held.
Paul Jackson's avatar
Paul Jackson committed
561 562
 *
 * The three key local variables below are:
563
 *    q  - a linked-list queue of cpuset pointers, used to implement a
Paul Jackson's avatar
Paul Jackson committed
564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594
 *	   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().
 */
595
static int generate_sched_domains(cpumask_var_t **domains,
596
			struct sched_domain_attr **attributes)
Paul Jackson's avatar
Paul Jackson committed
597 598 599 600 601
{
	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 */
602
	cpumask_var_t *doms;	/* resulting partition; i.e. sched domains */
603
	struct sched_domain_attr *dattr;  /* attributes for custom domains */
604
	int ndoms = 0;		/* number of sched domains in result */
605
	int nslot;		/* next empty doms[] struct cpumask slot */
606
	struct cgroup *pos_cgrp;
Paul Jackson's avatar
Paul Jackson committed
607 608

	doms = NULL;
609
	dattr = NULL;
610
	csa = NULL;
Paul Jackson's avatar
Paul Jackson committed
611 612 613

	/* Special case for the 99% of systems with one, full, sched domain */
	if (is_sched_load_balance(&top_cpuset)) {
614 615
		ndoms = 1;
		doms = alloc_sched_domains(ndoms);
Paul Jackson's avatar
Paul Jackson committed
616
		if (!doms)
617 618
			goto done;

619 620 621
		dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL);
		if (dattr) {
			*dattr = SD_ATTR_INIT;
622
			update_domain_attr_tree(dattr, &top_cpuset);
623
		}
624
		cpumask_copy(doms[0], top_cpuset.cpus_allowed);
625 626

		goto done;
Paul Jackson's avatar
Paul Jackson committed
627 628 629 630 631 632 633
	}

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

634 635
	rcu_read_lock();
	cpuset_for_each_descendant_pre(cp, pos_cgrp, &top_cpuset) {
636
		/*
637 638 639 640 641 642
		 * Continue traversing beyond @cp iff @cp has some CPUs and
		 * isn't load balancing.  The former is obvious.  The
		 * latter: 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.
643
		 */
644 645
		if (!cpumask_empty(cp->cpus_allowed) &&
		    !is_sched_load_balance(cp))
646
			continue;
647

648 649 650 651 652 653 654
		if (is_sched_load_balance(cp))
			csa[csn++] = cp;

		/* skip @cp's subtree */
		pos_cgrp = cgroup_rightmost_descendant(pos_cgrp);
	}
	rcu_read_unlock();
Paul Jackson's avatar
Paul Jackson committed
655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682

	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;
			}
		}
	}

683 684 685 686
	/*
	 * Now we know how many domains to create.
	 * Convert <csn, csa> to <ndoms, doms> and populate cpu masks.
	 */
687
	doms = alloc_sched_domains(ndoms);
688
	if (!doms)
689 690 691 692 693 694
		goto done;

	/*
	 * The rest of the code, including the scheduler, can deal with
	 * dattr==NULL case. No need to abort if alloc fails.
	 */
695
	dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL);
Paul Jackson's avatar
Paul Jackson committed
696 697 698

	for (nslot = 0, i = 0; i < csn; i++) {
		struct cpuset *a = csa[i];
699
		struct cpumask *dp;
Paul Jackson's avatar
Paul Jackson committed
700 701
		int apn = a->pn;

702 703 704 705 706
		if (apn < 0) {
			/* Skip completed partitions */
			continue;
		}

707
		dp = doms[nslot];
708 709 710 711 712 713 714 715 716 717

		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
718
			}
719 720
			continue;
		}
Paul Jackson's avatar
Paul Jackson committed
721

722
		cpumask_clear(dp);
723 724 725 726 727 728
		if (dattr)
			*(dattr + nslot) = SD_ATTR_INIT;
		for (j = i; j < csn; j++) {
			struct cpuset *b = csa[j];

			if (apn == b->pn) {
729
				cpumask_or(dp, dp, b->cpus_allowed);
730 731 732 733 734
				if (dattr)
					update_domain_attr_tree(dattr + nslot, b);

				/* Done with this partition */
				b->pn = -1;
Paul Jackson's avatar
Paul Jackson committed
735 736
			}
		}
737
		nslot++;
Paul Jackson's avatar
Paul Jackson committed
738 739 740
	}
	BUG_ON(nslot != ndoms);

741 742 743
done:
	kfree(csa);

744 745 746 747 748 749 750
	/*
	 * Fallback to the default domain if kmalloc() failed.
	 * See comments in partition_sched_domains().
	 */
	if (doms == NULL)
		ndoms = 1;

751 752 753 754 755 756 757 758
	*domains    = doms;
	*attributes = dattr;
	return ndoms;
}

/*
 * Rebuild scheduler domains.
 *
759 760 761 762 763
 * 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.
764
 *
765
 * Call with cpuset_mutex held.  Takes get_online_cpus().
766
 */
767
static void rebuild_sched_domains_locked(void)
768 769
{
	struct sched_domain_attr *attr;
770
	cpumask_var_t *doms;
771 772
	int ndoms;

773
	lockdep_assert_held(&cpuset_mutex);
774
	get_online_cpus();
775 776 777 778 779 780 781

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

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

782
	put_online_cpus();
783
}
784
#else /* !CONFIG_SMP */
785
static void rebuild_sched_domains_locked(void)
786 787 788
{
}

789
static int generate_sched_domains(cpumask_var_t **domains,
790 791 792 793 794 795
			struct sched_domain_attr **attributes)
{
	*domains = NULL;
	return 1;
}
#endif /* CONFIG_SMP */
Paul Jackson's avatar
Paul Jackson committed
796

797 798
void rebuild_sched_domains(void)
{
799
	mutex_lock(&cpuset_mutex);
800
	rebuild_sched_domains_locked();
801
	mutex_unlock(&cpuset_mutex);
Paul Jackson's avatar
Paul Jackson committed
802 803
}

804 805 806 807 808
/**
 * 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
 *
809
 * Call with cpuset_mutex held.  May take callback_mutex during call.
810 811 812
 * 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).
813
 */
814 815
static int cpuset_test_cpumask(struct task_struct *tsk,
			       struct cgroup_scanner *scan)
816
{
817
	return !cpumask_equal(&tsk->cpus_allowed,
818 819
			(cgroup_cs(scan->cg))->cpus_allowed);
}
820

821 822 823 824 825 826 827 828 829
/**
 * 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
830
 * holding cpuset_mutex at this point.
831
 */
832 833
static void cpuset_change_cpumask(struct task_struct *tsk,
				  struct cgroup_scanner *scan)
834
{
835
	set_cpus_allowed_ptr(tsk, ((cgroup_cs(scan->cg))->cpus_allowed));
836 837
}

838 839 840
/**
 * 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
841
 * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks()
842
 *
843
 * Called with cpuset_mutex held
844 845 846 847
 *
 * The cgroup_scan_tasks() function will scan all the tasks in a cgroup,
 * calling callback functions for each.
 *
848 849
 * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0
 * if @heap != NULL.
850
 */
851
static void update_tasks_cpumask(struct cpuset *cs, struct ptr_heap *heap)
852 853 854 855 856 857
{
	struct cgroup_scanner scan;

	scan.cg = cs->css.cgroup;
	scan.test_task = cpuset_test_cpumask;
	scan.process_task = cpuset_change_cpumask;
858 859
	scan.heap = heap;
	cgroup_scan_tasks(&scan);
860 861
}

862 863 864 865 866
/**
 * 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
 */
867 868
static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
			  const char *buf)
Linus Torvalds's avatar
Linus Torvalds committed
869
{
870
	struct ptr_heap heap;
871 872
	int retval;
	int is_load_balanced;
Linus Torvalds's avatar
Linus Torvalds committed
873

874
	/* top_cpuset.cpus_allowed tracks cpu_online_mask; it's read-only */
875 876 877
	if (cs == &top_cpuset)
		return -EACCES;

878
	/*
879
	 * An empty cpus_allowed is ok only if the cpuset has no tasks.
880 881 882
	 * Since cpulist_parse() fails on an empty mask, we special case
	 * that parsing.  The validate_change() call ensures that cpusets
	 * with tasks have cpus.
883
	 */
884
	if (!*buf) {
885
		cpumask_clear(trialcs->cpus_allowed);
886
	} else {
887
		retval = cpulist_parse(buf, trialcs->cpus_allowed);
888 889
		if (retval < 0)
			return retval;
890

891
		if (!cpumask_subset(trialcs->cpus_allowed, cpu_active_mask))
892
			return -EINVAL;
893
	}
894
	retval = validate_change(cs, trialcs);
895 896
	if (retval < 0)
		return retval;
Paul Jackson's avatar
Paul Jackson committed
897

Paul Menage's avatar
Paul Menage committed
898
	/* Nothing to do if the cpus didn't change */
899
	if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed))
Paul Menage's avatar
Paul Menage committed
900
		return 0;
901

902 903 904 905
	retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL);
	if (retval)
		return retval;

906
	is_load_balanced = is_sched_load_balance(trialcs);
Paul Jackson's avatar
Paul Jackson committed
907

908
	mutex_lock(&callback_mutex);
909
	cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed);
910
	mutex_unlock(&callback_mutex);
Paul Jackson's avatar
Paul Jackson committed
911

Paul Menage's avatar
Paul Menage committed
912 913
	/*
	 * Scan tasks in the cpuset, and update the cpumasks of any
914
	 * that need an update.
Paul Menage's avatar
Paul Menage committed
915
	 */
916 917 918
	update_tasks_cpumask(cs, &heap);

	heap_free(&heap);
919

Paul Menage's avatar
Paul Menage committed
920
	if (is_load_balanced)
921
		rebuild_sched_domains_locked();
922
	return 0;
Linus Torvalds's avatar
Linus Torvalds committed
923 924
}

925 926 927 928 929 930 931 932
/*
 * 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.
 *
933
 *    Call holding cpuset_mutex, so current's cpuset won't change
934
 *    during this call, as manage_mutex holds off any cpuset_attach()
935 936
 *    calls.  Therefore we don't need to take task_lock around the
 *    call to guarantee_online_mems(), as we know no one is changing
937
 *    our task's cpuset.
938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953
 *
 *    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);

954
	guarantee_online_mems(task_cs(tsk),&tsk->mems_allowed);
955 956
}

957
/*
958 959 960 961 962 963 964 965 966 967 968
 * 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)
{
969
	bool need_loop;
970

971 972 973 974 975 976 977 978 979 980
	/*
	 * 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);
981 982 983 984 985 986 987 988
	/*
	 * 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);
989

990 991
	if (need_loop)
		write_seqcount_begin(&tsk->mems_allowed_seq);
992

993 994
	nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems);
	mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP1);
995 996

	mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP2);
997
	tsk->mems_allowed = *newmems;
998 999 1000 1001

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

1002
	task_unlock(tsk);
1003 1004 1005 1006 1007
}

/*
 * 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
1008
 * memory_migrate flag is set. Called with cpuset_mutex held.
1009 1010 1011 1012 1013 1014 1015 1016
 */
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;
1017
	static nodemask_t newmems;	/* protected by cpuset_mutex */
1018 1019

	cs = cgroup_cs(scan->cg);
1020
	guarantee_online_mems(cs, &newmems);
1021

1022
	cpuset_change_task_nodemask(p, &newmems);
1023

1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035
	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);
}

1036 1037
static void *cpuset_being_rebound;

1038 1039 1040 1041
/**
 * 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
1042
 * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks()
1043
 *
1044
 * Called with cpuset_mutex held
1045 1046
 * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0
 * if @heap != NULL.
1047
 */
1048 1049
static void update_tasks_nodemask(struct cpuset *cs, const nodemask_t *oldmem,
				 struct ptr_heap *heap)
Linus Torvalds's avatar
Linus Torvalds committed
1050
{
1051
	struct cgroup_scanner scan;
1052

1053
	cpuset_being_rebound = cs;		/* causes mpol_dup() rebind */
1054

1055 1056 1057
	scan.cg = cs->css.cgroup;
	scan.test_task = NULL;
	scan.process_task = cpuset_change_nodemask;
1058
	scan.heap = heap;
1059
	scan.data = (nodemask_t *)oldmem;
1060 1061

	/*
1062 1063 1064 1065
	 * 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
1066
	 * the global cpuset_mutex, we know that no other rebind effort
1067
	 * will be contending for the global variable cpuset_being_rebound.
1068
	 * It's ok if we rebind the same mm twice; mpol_rebind_mm()
1069
	 * is idempotent.  Also migrate pages in each mm to new nodes.
1070
	 */
1071
	cgroup_scan_tasks(&scan);
1072

1073
	/* We're done rebinding vmas to this cpuset's new mems_allowed. */
1074
	cpuset_being_rebound = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
1075 1076
}

1077 1078 1079
/*
 * Handle user request to change the 'mems' memory placement
 * of a cpuset.  Needs to validate the request, update the
1080 1081 1082 1083
 * 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.
1084
 *
1085
 * Call with cpuset_mutex held.  May take callback_mutex during call.
1086 1087 1088 1089
 * 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.
 */
1090 1091
static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs,
			   const char *buf)
Miao Xie's avatar