cpuset.c 75.3 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

	/* partition number for rebuild_sched_domains() */
	int pn;
96

97 98 99
	/* for custom sched domain */
	int relax_domain_level;

100
	/* used for walking a cpuset hierarchy */
101
	struct list_head stack_list;
102 103

	struct work_struct hotplug_work;
Linus Torvalds's avatar
Linus Torvalds committed
104 105
};

106 107 108 109 110 111 112 113 114 115 116 117 118 119
/* 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);
}

120 121 122 123 124 125 126 127 128 129 130 131 132
#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
133 134
/* bits in struct cpuset flags field */
typedef enum {
Tejun Heo's avatar
Tejun Heo committed
135
	CS_ONLINE,
Linus Torvalds's avatar
Linus Torvalds committed
136 137
	CS_CPU_EXCLUSIVE,
	CS_MEM_EXCLUSIVE,
138
	CS_MEM_HARDWALL,
139
	CS_MEMORY_MIGRATE,
Paul Jackson's avatar
Paul Jackson committed
140
	CS_SCHED_LOAD_BALANCE,
141 142
	CS_SPREAD_PAGE,
	CS_SPREAD_SLAB,
Linus Torvalds's avatar
Linus Torvalds committed
143 144 145
} cpuset_flagbits_t;

/* convenient tests for these bits */
Tejun Heo's avatar
Tejun Heo committed
146 147 148 149 150
static inline bool is_cpuset_online(const struct cpuset *cs)
{
	return test_bit(CS_ONLINE, &cs->flags);
}

Linus Torvalds's avatar
Linus Torvalds committed
151 152
static inline int is_cpu_exclusive(const struct cpuset *cs)
{
153
	return test_bit(CS_CPU_EXCLUSIVE, &cs->flags);
Linus Torvalds's avatar
Linus Torvalds committed
154 155 156 157
}

static inline int is_mem_exclusive(const struct cpuset *cs)
{
158
	return test_bit(CS_MEM_EXCLUSIVE, &cs->flags);
Linus Torvalds's avatar
Linus Torvalds committed
159 160
}

161 162 163 164 165
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
166 167 168 169 170
static inline int is_sched_load_balance(const struct cpuset *cs)
{
	return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
}

171 172
static inline int is_memory_migrate(const struct cpuset *cs)
{
173
	return test_bit(CS_MEMORY_MIGRATE, &cs->flags);
174 175
}

176 177 178 179 180 181 182 183 184 185
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
186
static struct cpuset top_cpuset = {
Tejun Heo's avatar
Tejun Heo committed
187 188
	.flags = ((1 << CS_ONLINE) | (1 << CS_CPU_EXCLUSIVE) |
		  (1 << CS_MEM_EXCLUSIVE)),
Linus Torvalds's avatar
Linus Torvalds committed
189 190
};

191 192 193 194 195 196 197 198 199 200 201 202 203
/**
 * 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
204
/*
205 206 207 208 209 210 211
 * 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.
212
 *
213
 * A task must hold both mutexes to modify cpusets.  If a task
214
 * holds cgroup_mutex, then it blocks others wanting that mutex,
215
 * ensuring that it is the only task able to also acquire callback_mutex
216 217
 * and be able to modify cpusets.  It can perform various checks on
 * the cpuset structure first, knowing nothing will change.  It can
218
 * also allocate memory while just holding cgroup_mutex.  While it is
219
 * performing these checks, various callback routines can briefly
220 221
 * acquire callback_mutex to query cpusets.  Once it is ready to make
 * the changes, it takes callback_mutex, blocking everyone else.
222 223
 *
 * Calls to the kernel memory allocator can not be made while holding
224
 * callback_mutex, as that would risk double tripping on callback_mutex
225 226 227
 * from one of the callbacks into the cpuset code from within
 * __alloc_pages().
 *
228
 * If a task is only holding callback_mutex, then it has read-only
229 230
 * access to cpusets.
 *
231 232 233
 * 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.
234
 *
235
 * The cpuset_common_file_read() handlers only hold callback_mutex across
236 237 238
 * small pieces of code, such as when reading out possibly multi-word
 * cpumasks and nodemasks.
 *
239 240
 * 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
241 242
 */

243
static DEFINE_MUTEX(callback_mutex);
244

245 246 247 248 249 250 251 252 253 254 255
/*
 * 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);

256 257 258
/*
 * CPU / memory hotplug is handled asynchronously.
 */
259 260
static struct workqueue_struct *cpuset_propagate_hotplug_wq;

261
static void cpuset_hotplug_workfn(struct work_struct *work);
262
static void cpuset_propagate_hotplug_workfn(struct work_struct *work);
263 264 265

static DECLARE_WORK(cpuset_hotplug_work, cpuset_hotplug_workfn);

266 267
/*
 * This is ugly, but preserves the userspace API for existing cpuset
268
 * users. If someone tries to mount the "cpuset" filesystem, we
269 270
 * silently switch it to mount "cgroup" instead
 */
Al Viro's avatar
Al Viro committed
271 272
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
273
{
274
	struct file_system_type *cgroup_fs = get_fs_type("cgroup");
Al Viro's avatar
Al Viro committed
275
	struct dentry *ret = ERR_PTR(-ENODEV);
276 277 278 279
	if (cgroup_fs) {
		char mountopts[] =
			"cpuset,noprefix,"
			"release_agent=/sbin/cpuset_release_agent";
Al Viro's avatar
Al Viro committed
280 281
		ret = cgroup_fs->mount(cgroup_fs, flags,
					   unused_dev_name, mountopts);
282 283 284
		put_filesystem(cgroup_fs);
	}
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
285 286 287 288
}

static struct file_system_type cpuset_fs_type = {
	.name = "cpuset",
Al Viro's avatar
Al Viro committed
289
	.mount = cpuset_mount,
Linus Torvalds's avatar
Linus Torvalds committed
290 291 292
};

/*
293
 * Return in pmask the portion of a cpusets's cpus_allowed that
Linus Torvalds's avatar
Linus Torvalds committed
294 295 296
 * 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,
297 298
 * 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
299 300
 *
 * One way or another, we guarantee to return some non-empty subset
301
 * of cpu_online_mask.
Linus Torvalds's avatar
Linus Torvalds committed
302
 *
303
 * Call with callback_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
304 305
 */

306 307
static void guarantee_online_cpus(const struct cpuset *cs,
				  struct cpumask *pmask)
Linus Torvalds's avatar
Linus Torvalds committed
308
{
309
	while (cs && !cpumask_intersects(cs->cpus_allowed, cpu_online_mask))
Linus Torvalds's avatar
Linus Torvalds committed
310 311
		cs = cs->parent;
	if (cs)
312
		cpumask_and(pmask, cs->cpus_allowed, cpu_online_mask);
Linus Torvalds's avatar
Linus Torvalds committed
313
	else
314 315
		cpumask_copy(pmask, cpu_online_mask);
	BUG_ON(!cpumask_intersects(pmask, cpu_online_mask));
Linus Torvalds's avatar
Linus Torvalds committed
316 317 318 319
}

/*
 * Return in *pmask the portion of a cpusets's mems_allowed that
320 321 322
 * 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
323
 * found any online mems, return node_states[N_MEMORY].
Linus Torvalds's avatar
Linus Torvalds committed
324 325
 *
 * One way or another, we guarantee to return some non-empty subset
326
 * of node_states[N_MEMORY].
Linus Torvalds's avatar
Linus Torvalds committed
327
 *
328
 * Call with callback_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
329 330 331 332
 */

static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask)
{
333
	while (cs && !nodes_intersects(cs->mems_allowed,
334
					node_states[N_MEMORY]))
Linus Torvalds's avatar
Linus Torvalds committed
335 336
		cs = cs->parent;
	if (cs)
337
		nodes_and(*pmask, cs->mems_allowed,
338
					node_states[N_MEMORY]);
Linus Torvalds's avatar
Linus Torvalds committed
339
	else
340 341
		*pmask = node_states[N_MEMORY];
	BUG_ON(!nodes_intersects(*pmask, node_states[N_MEMORY]));
Linus Torvalds's avatar
Linus Torvalds committed
342 343
}

344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361
/*
 * 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
362 363 364 365 366
/*
 * 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
367
 * are only set if the other's are set.  Call holding cgroup_mutex.
Linus Torvalds's avatar
Linus Torvalds committed
368 369 370 371
 */

static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q)
{
372
	return	cpumask_subset(p->cpus_allowed, q->cpus_allowed) &&
Linus Torvalds's avatar
Linus Torvalds committed
373 374 375 376 377
		nodes_subset(p->mems_allowed, q->mems_allowed) &&
		is_cpu_exclusive(p) <= is_cpu_exclusive(q) &&
		is_mem_exclusive(p) <= is_mem_exclusive(q);
}

378 379 380 381 382 383
/**
 * 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)
{
384 385 386 387 388 389 390 391 392 393 394 395 396
	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;
397 398 399 400 401 402 403 404
}

/**
 * free_trial_cpuset - free the trial cpuset
 * @trial: the trial cpuset to be freed
 */
static void free_trial_cpuset(struct cpuset *trial)
{
405
	free_cpumask_var(trial->cpus_allowed);
406 407 408
	kfree(trial);
}

Linus Torvalds's avatar
Linus Torvalds committed
409 410 411 412 413 414 415
/*
 * 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
416
 * cgroup_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
417 418 419 420 421 422 423 424 425 426 427 428 429 430
 *
 * '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)
{
431
	struct cgroup *cont;
Linus Torvalds's avatar
Linus Torvalds committed
432
	struct cpuset *c, *par;
433 434 435
	int ret;

	rcu_read_lock();
Linus Torvalds's avatar
Linus Torvalds committed
436 437

	/* Each of our child cpusets must be a subset of us */
438 439 440 441
	ret = -EBUSY;
	cpuset_for_each_child(c, cont, cur)
		if (!is_cpuset_subset(c, trial))
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
442 443

	/* Remaining checks don't apply to root cpuset */
444
	ret = 0;
445
	if (cur == &top_cpuset)
446
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
447

448 449
	par = cur->parent;

Linus Torvalds's avatar
Linus Torvalds committed
450
	/* We must be a subset of our parent cpuset */
451
	ret = -EACCES;
Linus Torvalds's avatar
Linus Torvalds committed
452
	if (!is_cpuset_subset(trial, par))
453
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
454

455 456 457 458
	/*
	 * If either I or some sibling (!= me) is exclusive, we can't
	 * overlap
	 */
459 460
	ret = -EINVAL;
	cpuset_for_each_child(c, cont, par) {
Linus Torvalds's avatar
Linus Torvalds committed
461 462
		if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) &&
		    c != cur &&
463
		    cpumask_intersects(trial->cpus_allowed, c->cpus_allowed))
464
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
465 466 467
		if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) &&
		    c != cur &&
		    nodes_intersects(trial->mems_allowed, c->mems_allowed))
468
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
469 470
	}

471
	/* Cpusets with tasks can't have empty cpus_allowed or mems_allowed */
472 473 474 475 476
	ret = -ENOSPC;
	if (cgroup_task_count(cur->css.cgroup) &&
	    (cpumask_empty(trial->cpus_allowed) ||
	     nodes_empty(trial->mems_allowed)))
		goto out;
477

478 479 480 481
	ret = 0;
out:
	rcu_read_unlock();
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
482 483
}

484
#ifdef CONFIG_SMP
Paul Jackson's avatar
Paul Jackson committed
485
/*
486
 * Helper routine for generate_sched_domains().
Paul Jackson's avatar
Paul Jackson committed
487 488 489 490
 * Do cpusets a, b have overlapping cpus_allowed masks?
 */
static int cpusets_overlap(struct cpuset *a, struct cpuset *b)
{
491
	return cpumask_intersects(a->cpus_allowed, b->cpus_allowed);
Paul Jackson's avatar
Paul Jackson committed
492 493
}

494 495 496 497 498 499 500 501
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;
}

502 503 504 505 506 507 508 509 510 511 512 513 514 515
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);

516
		if (cpumask_empty(cp->cpus_allowed))
517 518 519 520 521
			continue;

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

522 523
		rcu_read_lock();
		cpuset_for_each_child(child, cont, cp)
524
			list_add_tail(&child->stack_list, &q);
525
		rcu_read_unlock();
526 527 528
	}
}

Paul Jackson's avatar
Paul Jackson committed
529
/*
530 531 532 533 534 535 536 537 538
 * 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
539
 *
540
 * See "What is sched_load_balance" in Documentation/cgroups/cpusets.txt
Paul Jackson's avatar
Paul Jackson committed
541 542 543 544 545 546 547
 * 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.
 *
548
 * Must be called with cgroup_lock held.
Paul Jackson's avatar
Paul Jackson committed
549 550
 *
 * The three key local variables below are:
551
 *    q  - a linked-list queue of cpuset pointers, used to implement a
Paul Jackson's avatar
Paul Jackson committed
552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582
 *	   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().
 */
583
static int generate_sched_domains(cpumask_var_t **domains,
584
			struct sched_domain_attr **attributes)
Paul Jackson's avatar
Paul Jackson committed
585
{
586
	LIST_HEAD(q);		/* queue of cpusets to be scanned */
Paul Jackson's avatar
Paul Jackson committed
587 588 589 590
	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 */
591
	cpumask_var_t *doms;	/* resulting partition; i.e. sched domains */
592
	struct sched_domain_attr *dattr;  /* attributes for custom domains */
593
	int ndoms = 0;		/* number of sched domains in result */
594
	int nslot;		/* next empty doms[] struct cpumask slot */
Paul Jackson's avatar
Paul Jackson committed
595 596

	doms = NULL;
597
	dattr = NULL;
598
	csa = NULL;
Paul Jackson's avatar
Paul Jackson committed
599 600 601

	/* Special case for the 99% of systems with one, full, sched domain */
	if (is_sched_load_balance(&top_cpuset)) {
602 603
		ndoms = 1;
		doms = alloc_sched_domains(ndoms);
Paul Jackson's avatar
Paul Jackson committed
604
		if (!doms)
605 606
			goto done;

607 608 609
		dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL);
		if (dattr) {
			*dattr = SD_ATTR_INIT;
610
			update_domain_attr_tree(dattr, &top_cpuset);
611
		}
612
		cpumask_copy(doms[0], top_cpuset.cpus_allowed);
613 614

		goto done;
Paul Jackson's avatar
Paul Jackson committed
615 616 617 618 619 620 621
	}

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

622 623
	list_add(&top_cpuset.stack_list, &q);
	while (!list_empty(&q)) {
Paul Jackson's avatar
Paul Jackson committed
624 625
		struct cgroup *cont;
		struct cpuset *child;   /* scans child cpusets of cp */
626

627 628 629
		cp = list_first_entry(&q, struct cpuset, stack_list);
		list_del(q.next);

630
		if (cpumask_empty(cp->cpus_allowed))
631 632
			continue;

633 634 635 636 637 638 639
		/*
		 * 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
640
			csa[csn++] = cp;
641 642
			continue;
		}
643

644 645
		rcu_read_lock();
		cpuset_for_each_child(child, cont, cp)
646
			list_add_tail(&child->stack_list, &q);
647
		rcu_read_unlock();
Paul Jackson's avatar
Paul Jackson committed
648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676
  	}

	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;
			}
		}
	}

677 678 679 680
	/*
	 * Now we know how many domains to create.
	 * Convert <csn, csa> to <ndoms, doms> and populate cpu masks.
	 */
681
	doms = alloc_sched_domains(ndoms);
682
	if (!doms)
683 684 685 686 687 688
		goto done;

	/*
	 * The rest of the code, including the scheduler, can deal with
	 * dattr==NULL case. No need to abort if alloc fails.
	 */
689
	dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL);
Paul Jackson's avatar
Paul Jackson committed
690 691 692

	for (nslot = 0, i = 0; i < csn; i++) {
		struct cpuset *a = csa[i];
693
		struct cpumask *dp;
Paul Jackson's avatar
Paul Jackson committed
694 695
		int apn = a->pn;

696 697 698 699 700
		if (apn < 0) {
			/* Skip completed partitions */
			continue;
		}

701
		dp = doms[nslot];
702 703 704 705 706 707 708 709 710 711

		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
712
			}
713 714
			continue;
		}
Paul Jackson's avatar
Paul Jackson committed
715

716
		cpumask_clear(dp);
717 718 719 720 721 722
		if (dattr)
			*(dattr + nslot) = SD_ATTR_INIT;
		for (j = i; j < csn; j++) {
			struct cpuset *b = csa[j];

			if (apn == b->pn) {
723
				cpumask_or(dp, dp, b->cpus_allowed);
724 725 726 727 728
				if (dattr)
					update_domain_attr_tree(dattr + nslot, b);

				/* Done with this partition */
				b->pn = -1;
Paul Jackson's avatar
Paul Jackson committed
729 730
			}
		}
731
		nslot++;
Paul Jackson's avatar
Paul Jackson committed
732 733 734
	}
	BUG_ON(nslot != ndoms);

735 736 737
done:
	kfree(csa);

738 739 740 741 742 743 744
	/*
	 * Fallback to the default domain if kmalloc() failed.
	 * See comments in partition_sched_domains().
	 */
	if (doms == NULL)
		ndoms = 1;

745 746 747 748 749 750 751 752
	*domains    = doms;
	*attributes = dattr;
	return ndoms;
}

/*
 * Rebuild scheduler domains.
 *
753 754 755 756 757
 * 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.
758
 *
759
 * Call with cgroup_mutex held.  Takes get_online_cpus().
760
 */
761
static void rebuild_sched_domains_locked(void)
762 763
{
	struct sched_domain_attr *attr;
764
	cpumask_var_t *doms;
765 766
	int ndoms;

767
	WARN_ON_ONCE(!cgroup_lock_is_held());
768
	get_online_cpus();
769 770 771 772 773 774 775

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

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

776
	put_online_cpus();
777
}
778
#else /* !CONFIG_SMP */
779
static void rebuild_sched_domains_locked(void)
780 781 782
{
}

783
static int generate_sched_domains(cpumask_var_t **domains,
784 785 786 787 788 789
			struct sched_domain_attr **attributes)
{
	*domains = NULL;
	return 1;
}
#endif /* CONFIG_SMP */
Paul Jackson's avatar
Paul Jackson committed
790

791 792
void rebuild_sched_domains(void)
{
793 794 795
	cgroup_lock();
	rebuild_sched_domains_locked();
	cgroup_unlock();
Paul Jackson's avatar
Paul Jackson committed
796 797
}

798 799 800 801 802
/**
 * 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
 *
803
 * Call with cgroup_mutex held.  May take callback_mutex during call.
804 805 806
 * 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).
807
 */
808 809
static int cpuset_test_cpumask(struct task_struct *tsk,
			       struct cgroup_scanner *scan)
810
{
811
	return !cpumask_equal(&tsk->cpus_allowed,
812 813
			(cgroup_cs(scan->cg))->cpus_allowed);
}
814

815 816 817 818 819 820 821 822 823 824 825
/**
 * 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.
 */
826 827
static void cpuset_change_cpumask(struct task_struct *tsk,
				  struct cgroup_scanner *scan)
828
{
829
	set_cpus_allowed_ptr(tsk, ((cgroup_cs(scan->cg))->cpus_allowed));
830 831
}

832 833 834
/**
 * 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
835
 * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks()
836 837 838 839 840 841
 *
 * Called with cgroup_mutex held
 *
 * The cgroup_scan_tasks() function will scan all the tasks in a cgroup,
 * calling callback functions for each.
 *
842 843
 * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0
 * if @heap != NULL.
844
 */
845
static void update_tasks_cpumask(struct cpuset *cs, struct ptr_heap *heap)
846 847 848 849 850 851
{
	struct cgroup_scanner scan;

	scan.cg = cs->css.cgroup;
	scan.test_task = cpuset_test_cpumask;
	scan.process_task = cpuset_change_cpumask;
852 853
	scan.heap = heap;
	cgroup_scan_tasks(&scan);
854 855
}

856 857 858 859 860
/**
 * 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
 */
861 862
static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
			  const char *buf)
Linus Torvalds's avatar
Linus Torvalds committed
863
{
864
	struct ptr_heap heap;
865 866
	int retval;
	int is_load_balanced;
Linus Torvalds's avatar
Linus Torvalds committed
867

868
	/* top_cpuset.cpus_allowed tracks cpu_online_mask; it's read-only */
869 870 871
	if (cs == &top_cpuset)
		return -EACCES;

872
	/*
873
	 * An empty cpus_allowed is ok only if the cpuset has no tasks.
874 875 876
	 * Since cpulist_parse() fails on an empty mask, we special case
	 * that parsing.  The validate_change() call ensures that cpusets
	 * with tasks have cpus.
877
	 */
878
	if (!*buf) {
879
		cpumask_clear(trialcs->cpus_allowed);
880
	} else {
881
		retval = cpulist_parse(buf, trialcs->cpus_allowed);
882 883
		if (retval < 0)
			return retval;
884

885
		if (!cpumask_subset(trialcs->cpus_allowed, cpu_active_mask))
886
			return -EINVAL;
887
	}
888
	retval = validate_change(cs, trialcs);
889 890
	if (retval < 0)
		return retval;
Paul Jackson's avatar
Paul Jackson committed
891

Paul Menage's avatar
Paul Menage committed
892
	/* Nothing to do if the cpus didn't change */
893
	if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed))
Paul Menage's avatar
Paul Menage committed
894
		return 0;
895

896 897 898 899
	retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL);
	if (retval)
		return retval;

900
	is_load_balanced = is_sched_load_balance(trialcs);
Paul Jackson's avatar
Paul Jackson committed
901

902
	mutex_lock(&callback_mutex);
903
	cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed);
904
	mutex_unlock(&callback_mutex);
Paul Jackson's avatar
Paul Jackson committed
905

Paul Menage's avatar
Paul Menage committed
906 907
	/*
	 * Scan tasks in the cpuset, and update the cpumasks of any
908
	 * that need an update.
Paul Menage's avatar
Paul Menage committed
909
	 */
910 911 912
	update_tasks_cpumask(cs, &heap);

	heap_free(&heap);
913

Paul Menage's avatar
Paul Menage committed
914
	if (is_load_balanced)
915
		rebuild_sched_domains_locked();
916
	return 0;
Linus Torvalds's avatar
Linus Torvalds committed
917 918
}

919 920 921 922 923 924 925 926
/*
 * 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.
 *
927
 *    Call holding cgroup_mutex, so current's cpuset won't change
928
 *    during this call, as manage_mutex holds off any cpuset_attach()
929 930
 *    calls.  Therefore we don't need to take task_lock around the
 *    call to guarantee_online_mems(), as we know no one is changing
931
 *    our task's cpuset.
932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947
 *
 *    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);

948
	guarantee_online_mems(task_cs(tsk),&tsk->mems_allowed);
949 950
}

951
/*
952 953 954 955 956 957 958 959 960 961 962
 * 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)
{
963
	bool need_loop;
964

965 966 967 968 969 970 971 972 973 974
	/*
	 * 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);
975 976 977 978 979 980 981 982
	/*
	 * 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);
983

984 985
	if (need_loop)
		write_seqcount_begin(&tsk->mems_allowed_seq);
986

987 988
	nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems);
	mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP1);
989 990

	mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP2);
991
	tsk->mems_allowed = *newmems;
992 993 994 995

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

996
	task_unlock(tsk);
997 998 999 1000 1001 1002
}

/*
 * 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.
1003 1004 1005 1006 1007 1008 1009 1010
 */
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;
1011
	static nodemask_t newmems;	/* protected by cgroup_mutex */
1012 1013

	cs = cgroup_cs(scan->cg);
1014
	guarantee_online_mems(cs, &newmems);
1015

1016
	cpuset_change_task_nodemask(p, &newmems);
1017

1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029
	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);
}

1030 1031
static void *cpuset_being_rebound;

1032 1033 1034 1035
/**
 * 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
1036
 * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks()
1037 1038
 *
 * Called with cgroup_mutex held
1039 1040
 * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0
 * if @heap != NULL.
1041
 */
1042 1043
static void update_tasks_nodemask(struct cpuset *cs, const nodemask_t *oldmem,
				 struct ptr_heap *heap)
Linus Torvalds's avatar
Linus Torvalds committed
1044
{
1045
	struct cgroup_scanner scan;
1046

1047
	cpuset_being_rebound = cs;		/* causes mpol_dup() rebind */
1048

1049 1050 1051
	scan.cg = cs->css.cgroup;
	scan.test_task = NULL;
	scan.process_task = cpuset_change_nodemask;
1052
	scan.heap = heap;
1053
	scan.data = (nodemask_t *)oldmem;
1054 1055

	/*
1056 1057 1058 1059 1060 1061
	 * 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.
1062
	 * It's ok if we rebind the same mm twice; mpol_rebind_mm()
1063
	 * is idempotent.  Also migrate pages in each mm to new nodes.
1064
	 */
1065
	cgroup_scan_tasks(&scan);
1066

1067
	/* We're done rebinding vmas to this cpuset's new mems_allowed. */
1068
	cpuset_being_rebound = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
1069 1070
}

1071 1072 1073
/*
 * Handle user request to change the 'mems' memory placement
 * of a cpuset.  Needs to validate the request, update the
1074 1075 1076 1077
 * 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.
1078 1079 1080 1081 1082 1083
 *
 * 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.
 */
1084 1085
static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs,
			   const char *buf)
1086
{
1087
	NODEMASK_ALLOC(nodemask_t, oldmem, GFP_KERNEL);
1088
	int retval;
1089
	struct ptr_heap heap;
1090

1091 1092 1093
	if (!oldmem)
		return -ENOMEM;

1094
	/*