cpuset.c 77.1 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
	/* used for walking a cpuset hierarchy */
107
	struct list_head stack_list;
108 109

	struct work_struct hotplug_work;
Linus Torvalds's avatar
Linus Torvalds committed
110 111
};

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

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

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


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

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

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

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

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

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

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

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

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

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

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

Linus Torvalds's avatar
Linus Torvalds committed
210
/*
211 212 213 214 215 216 217 218 219 220 221 222 223 224
 * 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.
225 226
 *
 * Calls to the kernel memory allocator can not be made while holding
227
 * callback_mutex, as that would risk double tripping on callback_mutex
228 229 230
 * from one of the callbacks into the cpuset code from within
 * __alloc_pages().
 *
231
 * If a task is only holding callback_mutex, then it has read-only
232 233
 * access to cpusets.
 *
234 235 236
 * 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.
237
 *
238
 * The cpuset_common_file_read() handlers only hold callback_mutex across
239 240 241
 * small pieces of code, such as when reading out possibly multi-word
 * cpumasks and nodemasks.
 *
242 243
 * 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
244 245
 */

246
static DEFINE_MUTEX(cpuset_mutex);
247
static DEFINE_MUTEX(callback_mutex);
248

249 250 251 252 253 254 255 256 257 258 259
/*
 * 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);

260 261 262
/*
 * CPU / memory hotplug is handled asynchronously.
 */
263 264
static struct workqueue_struct *cpuset_propagate_hotplug_wq;

265
static void cpuset_hotplug_workfn(struct work_struct *work);
266
static void cpuset_propagate_hotplug_workfn(struct work_struct *work);
267
static void schedule_cpuset_propagate_hotplug(struct cpuset *cs);
268 269 270

static DECLARE_WORK(cpuset_hotplug_work, cpuset_hotplug_workfn);

271 272
/*
 * This is ugly, but preserves the userspace API for existing cpuset
273
 * users. If someone tries to mount the "cpuset" filesystem, we
274 275
 * silently switch it to mount "cgroup" instead
 */
Al Viro's avatar
Al Viro committed
276 277
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
278
{
279
	struct file_system_type *cgroup_fs = get_fs_type("cgroup");
Al Viro's avatar
Al Viro committed
280
	struct dentry *ret = ERR_PTR(-ENODEV);
281 282 283 284
	if (cgroup_fs) {
		char mountopts[] =
			"cpuset,noprefix,"
			"release_agent=/sbin/cpuset_release_agent";
Al Viro's avatar
Al Viro committed
285 286
		ret = cgroup_fs->mount(cgroup_fs, flags,
					   unused_dev_name, mountopts);
287 288 289
		put_filesystem(cgroup_fs);
	}
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
290 291 292 293
}

static struct file_system_type cpuset_fs_type = {
	.name = "cpuset",
Al Viro's avatar
Al Viro committed
294
	.mount = cpuset_mount,
Linus Torvalds's avatar
Linus Torvalds committed
295 296 297
};

/*
298
 * Return in pmask the portion of a cpusets's cpus_allowed that
Linus Torvalds's avatar
Linus Torvalds committed
299 300 301
 * 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,
302 303
 * 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
304 305
 *
 * One way or another, we guarantee to return some non-empty subset
306
 * of cpu_online_mask.
Linus Torvalds's avatar
Linus Torvalds committed
307
 *
308
 * Call with callback_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
309 310
 */

311 312
static void guarantee_online_cpus(const struct cpuset *cs,
				  struct cpumask *pmask)
Linus Torvalds's avatar
Linus Torvalds committed
313
{
314
	while (cs && !cpumask_intersects(cs->cpus_allowed, cpu_online_mask))
Linus Torvalds's avatar
Linus Torvalds committed
315 316
		cs = cs->parent;
	if (cs)
317
		cpumask_and(pmask, cs->cpus_allowed, cpu_online_mask);
Linus Torvalds's avatar
Linus Torvalds committed
318
	else
319 320
		cpumask_copy(pmask, cpu_online_mask);
	BUG_ON(!cpumask_intersects(pmask, cpu_online_mask));
Linus Torvalds's avatar
Linus Torvalds committed
321 322 323 324
}

/*
 * Return in *pmask the portion of a cpusets's mems_allowed that
325 326 327
 * 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
328
 * found any online mems, return node_states[N_MEMORY].
Linus Torvalds's avatar
Linus Torvalds committed
329 330
 *
 * One way or another, we guarantee to return some non-empty subset
331
 * of node_states[N_MEMORY].
Linus Torvalds's avatar
Linus Torvalds committed
332
 *
333
 * Call with callback_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
334 335 336 337
 */

static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask)
{
338
	while (cs && !nodes_intersects(cs->mems_allowed,
339
					node_states[N_MEMORY]))
Linus Torvalds's avatar
Linus Torvalds committed
340 341
		cs = cs->parent;
	if (cs)
342
		nodes_and(*pmask, cs->mems_allowed,
343
					node_states[N_MEMORY]);
Linus Torvalds's avatar
Linus Torvalds committed
344
	else
345 346
		*pmask = node_states[N_MEMORY];
	BUG_ON(!nodes_intersects(*pmask, node_states[N_MEMORY]));
Linus Torvalds's avatar
Linus Torvalds committed
347 348
}

349 350 351
/*
 * update task's spread flag if cpuset's page/slab spread flag is set
 *
352
 * Called with callback_mutex/cpuset_mutex held
353 354 355 356 357 358 359 360 361 362 363 364 365 366
 */
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
367 368 369 370 371
/*
 * 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
372
 * are only set if the other's are set.  Call holding cpuset_mutex.
Linus Torvalds's avatar
Linus Torvalds committed
373 374 375 376
 */

static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q)
{
377
	return	cpumask_subset(p->cpus_allowed, q->cpus_allowed) &&
Linus Torvalds's avatar
Linus Torvalds committed
378 379 380 381 382
		nodes_subset(p->mems_allowed, q->mems_allowed) &&
		is_cpu_exclusive(p) <= is_cpu_exclusive(q) &&
		is_mem_exclusive(p) <= is_mem_exclusive(q);
}

383 384 385 386 387 388
/**
 * 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)
{
389 390 391 392 393 394 395 396 397 398 399 400 401
	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;
402 403 404 405 406 407 408 409
}

/**
 * free_trial_cpuset - free the trial cpuset
 * @trial: the trial cpuset to be freed
 */
static void free_trial_cpuset(struct cpuset *trial)
{
410
	free_cpumask_var(trial->cpus_allowed);
411 412 413
	kfree(trial);
}

Linus Torvalds's avatar
Linus Torvalds committed
414 415 416 417 418 419 420
/*
 * 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
421
 * cpuset_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
422 423 424 425 426 427 428 429 430 431 432 433 434 435
 *
 * '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)
{
436
	struct cgroup *cont;
Linus Torvalds's avatar
Linus Torvalds committed
437
	struct cpuset *c, *par;
438 439 440
	int ret;

	rcu_read_lock();
Linus Torvalds's avatar
Linus Torvalds committed
441 442

	/* Each of our child cpusets must be a subset of us */
443 444 445 446
	ret = -EBUSY;
	cpuset_for_each_child(c, cont, cur)
		if (!is_cpuset_subset(c, trial))
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
447 448

	/* Remaining checks don't apply to root cpuset */
449
	ret = 0;
450
	if (cur == &top_cpuset)
451
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
452

453 454
	par = cur->parent;

Linus Torvalds's avatar
Linus Torvalds committed
455
	/* We must be a subset of our parent cpuset */
456
	ret = -EACCES;
Linus Torvalds's avatar
Linus Torvalds committed
457
	if (!is_cpuset_subset(trial, par))
458
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
459

460 461 462 463
	/*
	 * If either I or some sibling (!= me) is exclusive, we can't
	 * overlap
	 */
464 465
	ret = -EINVAL;
	cpuset_for_each_child(c, cont, par) {
Linus Torvalds's avatar
Linus Torvalds committed
466 467
		if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) &&
		    c != cur &&
468
		    cpumask_intersects(trial->cpus_allowed, c->cpus_allowed))
469
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
470 471 472
		if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) &&
		    c != cur &&
		    nodes_intersects(trial->mems_allowed, c->mems_allowed))
473
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
474 475
	}

476 477 478 479
	/*
	 * Cpusets with tasks - existing or newly being attached - can't
	 * have empty cpus_allowed or mems_allowed.
	 */
480
	ret = -ENOSPC;
481
	if ((cgroup_task_count(cur->css.cgroup) || cur->attach_in_progress) &&
482 483 484
	    (cpumask_empty(trial->cpus_allowed) ||
	     nodes_empty(trial->mems_allowed)))
		goto out;
485

486 487 488 489
	ret = 0;
out:
	rcu_read_unlock();
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
490 491
}

492
#ifdef CONFIG_SMP
Paul Jackson's avatar
Paul Jackson committed
493
/*
494
 * Helper routine for generate_sched_domains().
Paul Jackson's avatar
Paul Jackson committed
495 496 497 498
 * Do cpusets a, b have overlapping cpus_allowed masks?
 */
static int cpusets_overlap(struct cpuset *a, struct cpuset *b)
{
499
	return cpumask_intersects(a->cpus_allowed, b->cpus_allowed);
Paul Jackson's avatar
Paul Jackson committed
500 501
}

502 503 504 505 506 507 508 509
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;
}

510 511 512 513 514 515 516 517 518 519 520 521 522 523
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);

524
		if (cpumask_empty(cp->cpus_allowed))
525 526 527 528 529
			continue;

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

530 531
		rcu_read_lock();
		cpuset_for_each_child(child, cont, cp)
532
			list_add_tail(&child->stack_list, &q);
533
		rcu_read_unlock();
534 535 536
	}
}

Paul Jackson's avatar
Paul Jackson committed
537
/*
538 539 540 541 542 543 544 545 546
 * 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
547
 *
548
 * See "What is sched_load_balance" in Documentation/cgroups/cpusets.txt
Paul Jackson's avatar
Paul Jackson committed
549 550 551 552 553 554 555
 * 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.
 *
556
 * Must be called with cpuset_mutex held.
Paul Jackson's avatar
Paul Jackson committed
557 558
 *
 * The three key local variables below are:
559
 *    q  - a linked-list queue of cpuset pointers, used to implement a
Paul Jackson's avatar
Paul Jackson committed
560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590
 *	   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().
 */
591
static int generate_sched_domains(cpumask_var_t **domains,
592
			struct sched_domain_attr **attributes)
Paul Jackson's avatar
Paul Jackson committed
593
{
594
	LIST_HEAD(q);		/* queue of cpusets to be scanned */
Paul Jackson's avatar
Paul Jackson committed
595 596 597 598
	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 */
599
	cpumask_var_t *doms;	/* resulting partition; i.e. sched domains */
600
	struct sched_domain_attr *dattr;  /* attributes for custom domains */
601
	int ndoms = 0;		/* number of sched domains in result */
602
	int nslot;		/* next empty doms[] struct cpumask slot */
Paul Jackson's avatar
Paul Jackson committed
603 604

	doms = NULL;
605
	dattr = NULL;
606
	csa = NULL;
Paul Jackson's avatar
Paul Jackson committed
607 608 609

	/* Special case for the 99% of systems with one, full, sched domain */
	if (is_sched_load_balance(&top_cpuset)) {
610 611
		ndoms = 1;
		doms = alloc_sched_domains(ndoms);
Paul Jackson's avatar
Paul Jackson committed
612
		if (!doms)
613 614
			goto done;

615 616 617
		dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL);
		if (dattr) {
			*dattr = SD_ATTR_INIT;
618
			update_domain_attr_tree(dattr, &top_cpuset);
619
		}
620
		cpumask_copy(doms[0], top_cpuset.cpus_allowed);
621 622

		goto done;
Paul Jackson's avatar
Paul Jackson committed
623 624 625 626 627 628 629
	}

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

630 631
	list_add(&top_cpuset.stack_list, &q);
	while (!list_empty(&q)) {
Paul Jackson's avatar
Paul Jackson committed
632 633
		struct cgroup *cont;
		struct cpuset *child;   /* scans child cpusets of cp */
634

635 636 637
		cp = list_first_entry(&q, struct cpuset, stack_list);
		list_del(q.next);

638
		if (cpumask_empty(cp->cpus_allowed))
639 640
			continue;

641 642 643 644 645 646 647
		/*
		 * 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
648
			csa[csn++] = cp;
649 650
			continue;
		}
651

652 653
		rcu_read_lock();
		cpuset_for_each_child(child, cont, cp)
654
			list_add_tail(&child->stack_list, &q);
655
		rcu_read_unlock();
Paul Jackson's avatar
Paul Jackson committed
656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684
  	}

	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;
			}
		}
	}

685 686 687 688
	/*
	 * Now we know how many domains to create.
	 * Convert <csn, csa> to <ndoms, doms> and populate cpu masks.
	 */
689
	doms = alloc_sched_domains(ndoms);
690
	if (!doms)
691 692 693 694 695 696
		goto done;

	/*
	 * The rest of the code, including the scheduler, can deal with
	 * dattr==NULL case. No need to abort if alloc fails.
	 */
697
	dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL);
Paul Jackson's avatar
Paul Jackson committed
698 699 700

	for (nslot = 0, i = 0; i < csn; i++) {
		struct cpuset *a = csa[i];
701
		struct cpumask *dp;
Paul Jackson's avatar
Paul Jackson committed
702 703
		int apn = a->pn;

704 705 706 707 708
		if (apn < 0) {
			/* Skip completed partitions */
			continue;
		}

709
		dp = doms[nslot];
710 711 712 713 714 715 716 717 718 719

		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
720
			}
721 722
			continue;
		}
Paul Jackson's avatar
Paul Jackson committed
723

724
		cpumask_clear(dp);
725 726 727 728 729 730
		if (dattr)
			*(dattr + nslot) = SD_ATTR_INIT;
		for (j = i; j < csn; j++) {
			struct cpuset *b = csa[j];

			if (apn == b->pn) {
731
				cpumask_or(dp, dp, b->cpus_allowed);
732 733 734 735 736
				if (dattr)
					update_domain_attr_tree(dattr + nslot, b);

				/* Done with this partition */
				b->pn = -1;
Paul Jackson's avatar
Paul Jackson committed
737 738
			}
		}
739
		nslot++;
Paul Jackson's avatar
Paul Jackson committed
740 741 742
	}
	BUG_ON(nslot != ndoms);

743 744 745
done:
	kfree(csa);

746 747 748 749 750 751 752
	/*
	 * Fallback to the default domain if kmalloc() failed.
	 * See comments in partition_sched_domains().
	 */
	if (doms == NULL)
		ndoms = 1;

753 754 755 756 757 758 759 760
	*domains    = doms;
	*attributes = dattr;
	return ndoms;
}

/*
 * Rebuild scheduler domains.
 *
761 762 763 764 765
 * 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.
766
 *
767
 * Call with cpuset_mutex held.  Takes get_online_cpus().
768
 */
769
static void rebuild_sched_domains_locked(void)
770 771
{
	struct sched_domain_attr *attr;
772
	cpumask_var_t *doms;
773 774
	int ndoms;

775
	lockdep_assert_held(&cpuset_mutex);
776
	get_online_cpus();
777 778 779 780 781 782 783

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

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

784
	put_online_cpus();
785
}
786
#else /* !CONFIG_SMP */
787
static void rebuild_sched_domains_locked(void)
788 789 790
{
}

791
static int generate_sched_domains(cpumask_var_t **domains,
792 793 794 795 796 797
			struct sched_domain_attr **attributes)
{
	*domains = NULL;
	return 1;
}
#endif /* CONFIG_SMP */
Paul Jackson's avatar
Paul Jackson committed
798

799 800
void rebuild_sched_domains(void)
{
801
	mutex_lock(&cpuset_mutex);
802
	rebuild_sched_domains_locked();
803
	mutex_unlock(&cpuset_mutex);
Paul Jackson's avatar
Paul Jackson committed
804 805
}

806 807 808 809 810
/**
 * 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
 *
811
 * Call with cpuset_mutex held.  May take callback_mutex during call.
812 813 814
 * 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).
815
 */
816 817
static int cpuset_test_cpumask(struct task_struct *tsk,
			       struct cgroup_scanner *scan)
818
{
819
	return !cpumask_equal(&tsk->cpus_allowed,
820 821
			(cgroup_cs(scan->cg))->cpus_allowed);
}
822

823 824 825 826 827 828 829 830 831
/**
 * 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
832
 * holding cpuset_mutex at this point.
833
 */
834 835
static void cpuset_change_cpumask(struct task_struct *tsk,
				  struct cgroup_scanner *scan)
836
{
837
	set_cpus_allowed_ptr(tsk, ((cgroup_cs(scan->cg))->cpus_allowed));
838 839
}

840 841 842
/**
 * 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
843
 * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks()
844
 *
845
 * Called with cpuset_mutex held
846 847 848 849
 *
 * The cgroup_scan_tasks() function will scan all the tasks in a cgroup,
 * calling callback functions for each.
 *
850 851
 * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0
 * if @heap != NULL.
852
 */
853
static void update_tasks_cpumask(struct cpuset *cs, struct ptr_heap *heap)
854 855 856 857 858 859
{
	struct cgroup_scanner scan;

	scan.cg = cs->css.cgroup;
	scan.test_task = cpuset_test_cpumask;
	scan.process_task = cpuset_change_cpumask;
860 861
	scan.heap = heap;
	cgroup_scan_tasks(&scan);
862 863
}

864 865 866 867 868
/**
 * 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
 */
869 870
static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
			  const char *buf)
Linus Torvalds's avatar
Linus Torvalds committed
871
{
872
	struct ptr_heap heap;
873 874
	int retval;
	int is_load_balanced;
Linus Torvalds's avatar
Linus Torvalds committed
875

876
	/* top_cpuset.cpus_allowed tracks cpu_online_mask; it's read-only */
877 878 879
	if (cs == &top_cpuset)
		return -EACCES;

880
	/*
881
	 * An empty cpus_allowed is ok only if the cpuset has no tasks.
882 883 884
	 * Since cpulist_parse() fails on an empty mask, we special case
	 * that parsing.  The validate_change() call ensures that cpusets
	 * with tasks have cpus.
885
	 */
886
	if (!*buf) {
887
		cpumask_clear(trialcs->cpus_allowed);
888
	} else {
889
		retval = cpulist_parse(buf, trialcs->cpus_allowed);
890 891
		if (retval < 0)
			return retval;
892

893
		if (!cpumask_subset(trialcs->cpus_allowed, cpu_active_mask))
894
			return -EINVAL;
895
	}
896
	retval = validate_change(cs, trialcs);
897 898
	if (retval < 0)
		return retval;
Paul Jackson's avatar
Paul Jackson committed
899

Paul Menage's avatar
Paul Menage committed
900
	/* Nothing to do if the cpus didn't change */
901
	if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed))
Paul Menage's avatar
Paul Menage committed
902
		return 0;
903

904 905 906 907
	retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL);
	if (retval)
		return retval;

908
	is_load_balanced = is_sched_load_balance(trialcs);
Paul Jackson's avatar
Paul Jackson committed
909

910
	mutex_lock(&callback_mutex);
911
	cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed);
912
	mutex_unlock(&callback_mutex);
Paul Jackson's avatar
Paul Jackson committed
913

Paul Menage's avatar
Paul Menage committed
914 915
	/*
	 * Scan tasks in the cpuset, and update the cpumasks of any
916
	 * that need an update.
Paul Menage's avatar
Paul Menage committed
917
	 */
918 919 920
	update_tasks_cpumask(cs, &heap);

	heap_free(&heap);
921

Paul Menage's avatar
Paul Menage committed
922
	if (is_load_balanced)
923
		rebuild_sched_domains_locked();
924
	return 0;
Linus Torvalds's avatar
Linus Torvalds committed
925 926
}

927 928 929 930 931 932 933 934
/*
 * 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.
 *
935
 *    Call holding cpuset_mutex, so current's cpuset won't change
936
 *    during this call, as manage_mutex holds off any cpuset_attach()
937 938
 *    calls.  Therefore we don't need to take task_lock around the
 *    call to guarantee_online_mems(), as we know no one is changing
939
 *    our task's cpuset.
940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955
 *
 *    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);

956
	guarantee_online_mems(task_cs(tsk),&tsk->mems_allowed);
957 958
}

959
/*
960 961 962 963 964 965 966 967 968 969 970
 * 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)
{
971
	bool need_loop;
972

973 974 975 976 977 978 979 980 981 982
	/*
	 * 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);
983 984 985 986 987 988 989 990
	/*
	 * 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);
991

992 993
	if (need_loop)
		write_seqcount_begin(&tsk->mems_allowed_seq);
994

995 996
	nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems);
	mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP1);
997 998

	mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP2);
999
	tsk->mems_allowed = *newmems;
1000 1001 1002 1003

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

1004
	task_unlock(tsk);
1005 1006 1007 1008 1009
}

/*
 * 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
1010
 * memory_migrate flag is set. Called with cpuset_mutex held.
1011 1012 1013 1014 1015 1016 1017 1018
 */
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;
1019
	static nodemask_t newmems;	/* protected by cpuset_mutex */
1020 1021

	cs = cgroup_cs(scan->cg);
1022
	guarantee_online_mems(cs, &newmems);
1023

1024
	cpuset_change_task_nodemask(p, &newmems);
1025

1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037
	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);
}

1038 1039
static void *cpuset_being_rebound;

1040 1041 1042 1043
/**
 * 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
1044
 * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks()
1045
 *
1046
 * Called with cpuset_mutex held
1047 1048
 * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0
 * if @heap != NULL.
1049
 */
1050 1051
static void update_tasks_nodemask(struct cpuset *cs, const nodemask_t *oldmem,
				 struct ptr_heap *heap)
Linus Torvalds's avatar
Linus Torvalds committed
1052
{
1053
	struct cgroup_scanner scan;
1054

1055
	cpuset_being_rebound = cs;		/* causes mpol_dup() rebind */
1056

1057 1058 1059
	scan.cg = cs->css.cgroup;
	scan.test_task = NULL;
	scan.process_task = cpuset_change_nodemask;
1060
	scan.heap = heap;
1061
	scan.data = (nodemask_t *)oldmem;
1062 1063

	/*
1064 1065 1066 1067
	 * 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
1068
	 * the global cpuset_mutex, we know that no other rebind effort
1069
	 * will be contending for the global variable cpuset_being_rebound.
1070
	 * It's ok if we rebind the same mm twice; mpol_rebind_mm()
1071
	 * is idempotent.  Also migrate pages in each mm to new nodes.
1072
	 */
1073
	cgroup_scan_tasks(&scan);
1074

1075
	/* We're done rebinding vmas to this cpuset's new mems_allowed. */
1076
	cpuset_being_rebound = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
1077 1078
}

1079 1080 1081
/*
 * Handle user request to change the 'mems' memory placement
 * of a cpuset.  Needs to validate the request, update the
1082 1083 1084 1085
 * 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.
1086
 *
Tejun Heo's avatar