tree.c 116.3 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14
/*
 * Read-Copy Update mechanism for mutual exclusion
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
15 16
 * along with this program; if not, you can access it online at
 * http://www.gnu.org/licenses/gpl-2.0.html.
17 18 19 20 21 22 23 24 25 26 27
 *
 * Copyright IBM Corporation, 2008
 *
 * Authors: Dipankar Sarma <dipankar@in.ibm.com>
 *	    Manfred Spraul <manfred@colorfullife.com>
 *	    Paul E. McKenney <paulmck@linux.vnet.ibm.com> Hierarchical version
 *
 * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
 *
 * For detailed explanation of Read-Copy Update mechanism see -
28
 *	Documentation/RCU
29 30 31 32 33 34 35 36 37
 */
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/init.h>
#include <linux/spinlock.h>
#include <linux/smp.h>
#include <linux/rcupdate.h>
#include <linux/interrupt.h>
#include <linux/sched.h>
38
#include <linux/nmi.h>
39
#include <linux/atomic.h>
40
#include <linux/bitops.h>
41
#include <linux/export.h>
42 43
#include <linux/completion.h>
#include <linux/moduleparam.h>
44
#include <linux/module.h>
45 46 47 48 49
#include <linux/percpu.h>
#include <linux/notifier.h>
#include <linux/cpu.h>
#include <linux/mutex.h>
#include <linux/time.h>
50
#include <linux/kernel_stat.h>
51 52
#include <linux/wait.h>
#include <linux/kthread.h>
53
#include <linux/prefetch.h>
54 55
#include <linux/delay.h>
#include <linux/stop_machine.h>
56
#include <linux/random.h>
57
#include <linux/ftrace_event.h>
58
#include <linux/suspend.h>
59

60
#include "tree.h"
61
#include "rcu.h"
62

63 64 65 66 67 68
MODULE_ALIAS("rcutree");
#ifdef MODULE_PARAM_PREFIX
#undef MODULE_PARAM_PREFIX
#endif
#define MODULE_PARAM_PREFIX "rcutree."

69 70
/* Data structures. */

71
static struct lock_class_key rcu_node_class[RCU_NUM_LVLS];
72
static struct lock_class_key rcu_fqs_class[RCU_NUM_LVLS];
73

74 75 76 77 78 79 80 81
/*
 * In order to export the rcu_state name to the tracing tools, it
 * needs to be added in the __tracepoint_string section.
 * This requires defining a separate variable tp_<sname>_varname
 * that points to the string being used, and this will allow
 * the tracing userspace tools to be able to decipher the string
 * address to the matching string.
 */
82 83
#ifdef CONFIG_TRACING
# define DEFINE_RCU_TPS(sname) \
84
static char sname##_varname[] = #sname; \
85 86 87 88 89 90 91 92 93
static const char *tp_##sname##_varname __used __tracepoint_string = sname##_varname;
# define RCU_STATE_NAME(sname) sname##_varname
#else
# define DEFINE_RCU_TPS(sname)
# define RCU_STATE_NAME(sname) __stringify(sname)
#endif

#define RCU_STATE_INITIALIZER(sname, sabbr, cr) \
DEFINE_RCU_TPS(sname) \
94
struct rcu_state sname##_state = { \
95
	.level = { &sname##_state.node[0] }, \
96
	.call = cr, \
97
	.fqs_state = RCU_GP_IDLE, \
P
Paul E. McKenney 已提交
98 99
	.gpnum = 0UL - 300UL, \
	.completed = 0UL - 300UL, \
100
	.orphan_lock = __RAW_SPIN_LOCK_UNLOCKED(&sname##_state.orphan_lock), \
101 102
	.orphan_nxttail = &sname##_state.orphan_nxtlist, \
	.orphan_donetail = &sname##_state.orphan_donelist, \
103
	.barrier_mutex = __MUTEX_INITIALIZER(sname##_state.barrier_mutex), \
104
	.onoff_mutex = __MUTEX_INITIALIZER(sname##_state.onoff_mutex), \
105
	.name = RCU_STATE_NAME(sname), \
106
	.abbr = sabbr, \
107 108
}; \
DEFINE_PER_CPU(struct rcu_data, sname##_data)
109

110 111
RCU_STATE_INITIALIZER(rcu_sched, 's', call_rcu_sched);
RCU_STATE_INITIALIZER(rcu_bh, 'b', call_rcu_bh);
112

113
static struct rcu_state *rcu_state_p;
114
LIST_HEAD(rcu_struct_flavors);
115

116 117
/* Increase (but not decrease) the CONFIG_RCU_FANOUT_LEAF at boot time. */
static int rcu_fanout_leaf = CONFIG_RCU_FANOUT_LEAF;
118
module_param(rcu_fanout_leaf, int, 0444);
119 120 121 122 123 124 125 126 127 128
int rcu_num_lvls __read_mostly = RCU_NUM_LVLS;
static int num_rcu_lvl[] = {  /* Number of rcu_nodes at specified level. */
	NUM_RCU_LVL_0,
	NUM_RCU_LVL_1,
	NUM_RCU_LVL_2,
	NUM_RCU_LVL_3,
	NUM_RCU_LVL_4,
};
int rcu_num_nodes __read_mostly = NUM_RCU_NODES; /* Total # rcu_nodes in use. */

129 130 131 132
/*
 * The rcu_scheduler_active variable transitions from zero to one just
 * before the first task is spawned.  So when this variable is zero, RCU
 * can assume that there is but one task, allowing RCU to (for example)
133
 * optimize synchronize_sched() to a simple barrier().  When this variable
134 135 136 137
 * is one, RCU must actually do all the hard work required to detect real
 * grace periods.  This variable is also used to suppress boot-time false
 * positives from lockdep-RCU error checking.
 */
138 139 140
int rcu_scheduler_active __read_mostly;
EXPORT_SYMBOL_GPL(rcu_scheduler_active);

141 142 143 144 145 146 147 148 149 150 151 152 153 154
/*
 * The rcu_scheduler_fully_active variable transitions from zero to one
 * during the early_initcall() processing, which is after the scheduler
 * is capable of creating new tasks.  So RCU processing (for example,
 * creating tasks for RCU priority boosting) must be delayed until after
 * rcu_scheduler_fully_active transitions from zero to one.  We also
 * currently delay invocation of any RCU callbacks until after this point.
 *
 * It might later prove better for people registering RCU callbacks during
 * early boot to take responsibility for these callbacks, but one step at
 * a time.
 */
static int rcu_scheduler_fully_active __read_mostly;

T
Thomas Gleixner 已提交
155
static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu);
156 157
static void invoke_rcu_core(void);
static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp);
158

159 160 161 162 163 164 165 166 167 168 169 170
/*
 * Track the rcutorture test sequence number and the update version
 * number within a given test.  The rcutorture_testseq is incremented
 * on every rcutorture module load and unload, so has an odd value
 * when a test is running.  The rcutorture_vernum is set to zero
 * when rcutorture starts and is incremented on each rcutorture update.
 * These variables enable correlating rcutorture output with the
 * RCU tracing information.
 */
unsigned long rcutorture_testseq;
unsigned long rcutorture_vernum;

171 172 173 174 175 176 177 178 179 180
/*
 * Return true if an RCU grace period is in progress.  The ACCESS_ONCE()s
 * permit this function to be invoked without holding the root rcu_node
 * structure's ->lock, but of course results can be subject to change.
 */
static int rcu_gp_in_progress(struct rcu_state *rsp)
{
	return ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum);
}

181
/*
182
 * Note a quiescent state.  Because we do not need to know
183
 * how many quiescent states passed, just if there was at least
184
 * one since the start of the grace period, this just sets a flag.
185
 * The caller must have disabled preemption.
186
 */
187
void rcu_sched_qs(void)
188
{
189 190 191 192 193 194
	if (!__this_cpu_read(rcu_sched_data.passed_quiesce)) {
		trace_rcu_grace_period(TPS("rcu_sched"),
				       __this_cpu_read(rcu_sched_data.gpnum),
				       TPS("cpuqs"));
		__this_cpu_write(rcu_sched_data.passed_quiesce, 1);
	}
195 196
}

197
void rcu_bh_qs(void)
198
{
199 200 201 202 203 204
	if (!__this_cpu_read(rcu_bh_data.passed_quiesce)) {
		trace_rcu_grace_period(TPS("rcu_bh"),
				       __this_cpu_read(rcu_bh_data.gpnum),
				       TPS("cpuqs"));
		__this_cpu_write(rcu_bh_data.passed_quiesce, 1);
	}
205
}
206

207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270
static DEFINE_PER_CPU(int, rcu_sched_qs_mask);

static DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = {
	.dynticks_nesting = DYNTICK_TASK_EXIT_IDLE,
	.dynticks = ATOMIC_INIT(1),
#ifdef CONFIG_NO_HZ_FULL_SYSIDLE
	.dynticks_idle_nesting = DYNTICK_TASK_NEST_VALUE,
	.dynticks_idle = ATOMIC_INIT(1),
#endif /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
};

/*
 * Let the RCU core know that this CPU has gone through the scheduler,
 * which is a quiescent state.  This is called when the need for a
 * quiescent state is urgent, so we burn an atomic operation and full
 * memory barriers to let the RCU core know about it, regardless of what
 * this CPU might (or might not) do in the near future.
 *
 * We inform the RCU core by emulating a zero-duration dyntick-idle
 * period, which we in turn do by incrementing the ->dynticks counter
 * by two.
 */
static void rcu_momentary_dyntick_idle(void)
{
	unsigned long flags;
	struct rcu_data *rdp;
	struct rcu_dynticks *rdtp;
	int resched_mask;
	struct rcu_state *rsp;

	local_irq_save(flags);

	/*
	 * Yes, we can lose flag-setting operations.  This is OK, because
	 * the flag will be set again after some delay.
	 */
	resched_mask = raw_cpu_read(rcu_sched_qs_mask);
	raw_cpu_write(rcu_sched_qs_mask, 0);

	/* Find the flavor that needs a quiescent state. */
	for_each_rcu_flavor(rsp) {
		rdp = raw_cpu_ptr(rsp->rda);
		if (!(resched_mask & rsp->flavor_mask))
			continue;
		smp_mb(); /* rcu_sched_qs_mask before cond_resched_completed. */
		if (ACCESS_ONCE(rdp->mynode->completed) !=
		    ACCESS_ONCE(rdp->cond_resched_completed))
			continue;

		/*
		 * Pretend to be momentarily idle for the quiescent state.
		 * This allows the grace-period kthread to record the
		 * quiescent state, with no need for this CPU to do anything
		 * further.
		 */
		rdtp = this_cpu_ptr(&rcu_dynticks);
		smp_mb__before_atomic(); /* Earlier stuff before QS. */
		atomic_add(2, &rdtp->dynticks);  /* QS. */
		smp_mb__after_atomic(); /* Later stuff after QS. */
		break;
	}
	local_irq_restore(flags);
}

271 272 273
/*
 * Note a context switch.  This is a quiescent state for RCU-sched,
 * and requires special handling for preemptible RCU.
274
 * The caller must have disabled preemption.
275 276 277
 */
void rcu_note_context_switch(int cpu)
{
278
	trace_rcu_utilization(TPS("Start context switch"));
279
	rcu_sched_qs();
280
	rcu_preempt_note_context_switch(cpu);
281 282
	if (unlikely(raw_cpu_read(rcu_sched_qs_mask)))
		rcu_momentary_dyntick_idle();
283
	trace_rcu_utilization(TPS("End context switch"));
284
}
285
EXPORT_SYMBOL_GPL(rcu_note_context_switch);
286

E
Eric Dumazet 已提交
287 288 289
static long blimit = 10;	/* Maximum callbacks per rcu_do_batch. */
static long qhimark = 10000;	/* If this many pending, ignore blimit. */
static long qlowmark = 100;	/* Once only this many pending, use blimit. */
290

E
Eric Dumazet 已提交
291 292 293
module_param(blimit, long, 0444);
module_param(qhimark, long, 0444);
module_param(qlowmark, long, 0444);
294

295 296
static ulong jiffies_till_first_fqs = ULONG_MAX;
static ulong jiffies_till_next_fqs = ULONG_MAX;
297 298 299 300

module_param(jiffies_till_first_fqs, ulong, 0644);
module_param(jiffies_till_next_fqs, ulong, 0644);

301 302 303 304 305 306 307
/*
 * How long the grace period must be before we start recruiting
 * quiescent-state help from rcu_note_context_switch().
 */
static ulong jiffies_till_sched_qs = HZ / 20;
module_param(jiffies_till_sched_qs, ulong, 0644);

308
static bool rcu_start_gp_advanced(struct rcu_state *rsp, struct rcu_node *rnp,
309
				  struct rcu_data *rdp);
310 311 312 313
static void force_qs_rnp(struct rcu_state *rsp,
			 int (*f)(struct rcu_data *rsp, bool *isidle,
				  unsigned long *maxj),
			 bool *isidle, unsigned long *maxj);
314
static void force_quiescent_state(struct rcu_state *rsp);
315
static int rcu_pending(int cpu);
316 317

/*
318
 * Return the number of RCU-sched batches processed thus far for debug & stats.
319
 */
320
long rcu_batches_completed_sched(void)
321
{
322
	return rcu_sched_state.completed;
323
}
324
EXPORT_SYMBOL_GPL(rcu_batches_completed_sched);
325 326 327 328 329 330 331 332 333 334

/*
 * Return the number of RCU BH batches processed thus far for debug & stats.
 */
long rcu_batches_completed_bh(void)
{
	return rcu_bh_state.completed;
}
EXPORT_SYMBOL_GPL(rcu_batches_completed_bh);

335 336 337 338 339
/*
 * Force a quiescent state.
 */
void rcu_force_quiescent_state(void)
{
340
	force_quiescent_state(rcu_state_p);
341 342 343
}
EXPORT_SYMBOL_GPL(rcu_force_quiescent_state);

344 345 346 347 348
/*
 * Force a quiescent state for RCU BH.
 */
void rcu_bh_force_quiescent_state(void)
{
349
	force_quiescent_state(&rcu_bh_state);
350 351 352
}
EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state);

353 354 355 356 357 358 359 360 361 362 363 364 365 366 367
/*
 * Show the state of the grace-period kthreads.
 */
void show_rcu_gp_kthreads(void)
{
	struct rcu_state *rsp;

	for_each_rcu_flavor(rsp) {
		pr_info("%s: wait state: %d ->state: %#lx\n",
			rsp->name, rsp->gp_state, rsp->gp_kthread->state);
		/* sched_show_task(rsp->gp_kthread); */
	}
}
EXPORT_SYMBOL_GPL(show_rcu_gp_kthreads);

368 369 370 371 372 373 374 375 376 377 378 379 380 381
/*
 * Record the number of times rcutorture tests have been initiated and
 * terminated.  This information allows the debugfs tracing stats to be
 * correlated to the rcutorture messages, even when the rcutorture module
 * is being repeatedly loaded and unloaded.  In other words, we cannot
 * store this state in rcutorture itself.
 */
void rcutorture_record_test_transition(void)
{
	rcutorture_testseq++;
	rcutorture_vernum = 0;
}
EXPORT_SYMBOL_GPL(rcutorture_record_test_transition);

382 383 384 385 386 387 388 389 390 391
/*
 * Send along grace-period-related data for rcutorture diagnostics.
 */
void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags,
			    unsigned long *gpnum, unsigned long *completed)
{
	struct rcu_state *rsp = NULL;

	switch (test_type) {
	case RCU_FLAVOR:
392
		rsp = rcu_state_p;
393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414
		break;
	case RCU_BH_FLAVOR:
		rsp = &rcu_bh_state;
		break;
	case RCU_SCHED_FLAVOR:
		rsp = &rcu_sched_state;
		break;
	default:
		break;
	}
	if (rsp != NULL) {
		*flags = ACCESS_ONCE(rsp->gp_flags);
		*gpnum = ACCESS_ONCE(rsp->gpnum);
		*completed = ACCESS_ONCE(rsp->completed);
		return;
	}
	*flags = 0;
	*gpnum = 0;
	*completed = 0;
}
EXPORT_SYMBOL_GPL(rcutorture_get_gp_data);

415 416 417 418 419 420 421 422 423 424 425
/*
 * Record the number of writer passes through the current rcutorture test.
 * This is also used to correlate debugfs tracing stats with the rcutorture
 * messages.
 */
void rcutorture_record_progress(unsigned long vernum)
{
	rcutorture_vernum++;
}
EXPORT_SYMBOL_GPL(rcutorture_record_progress);

426 427 428 429 430
/*
 * Force a quiescent state for RCU-sched.
 */
void rcu_sched_force_quiescent_state(void)
{
431
	force_quiescent_state(&rcu_sched_state);
432 433 434
}
EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state);

435 436 437 438 439 440
/*
 * Does the CPU have callbacks ready to be invoked?
 */
static int
cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp)
{
P
Paul E. McKenney 已提交
441 442
	return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL] &&
	       rdp->nxttail[RCU_DONE_TAIL] != NULL;
443 444
}

445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466
/*
 * Return the root node of the specified rcu_state structure.
 */
static struct rcu_node *rcu_get_root(struct rcu_state *rsp)
{
	return &rsp->node[0];
}

/*
 * Is there any need for future grace periods?
 * Interrupts must be disabled.  If the caller does not hold the root
 * rnp_node structure's ->lock, the results are advisory only.
 */
static int rcu_future_needs_gp(struct rcu_state *rsp)
{
	struct rcu_node *rnp = rcu_get_root(rsp);
	int idx = (ACCESS_ONCE(rnp->completed) + 1) & 0x1;
	int *fp = &rnp->need_future_gp[idx];

	return ACCESS_ONCE(*fp);
}

467
/*
468 469 470
 * Does the current CPU require a not-yet-started grace period?
 * The caller must have disabled interrupts to prevent races with
 * normal callback registry.
471 472 473 474
 */
static int
cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp)
{
475
	int i;
P
Paul E. McKenney 已提交
476

477 478
	if (rcu_gp_in_progress(rsp))
		return 0;  /* No, a grace period is already in progress. */
479
	if (rcu_future_needs_gp(rsp))
480
		return 1;  /* Yes, a no-CBs CPU needs one. */
481 482 483 484 485 486 487 488 489 490
	if (!rdp->nxttail[RCU_NEXT_TAIL])
		return 0;  /* No, this is a no-CBs (or offline) CPU. */
	if (*rdp->nxttail[RCU_NEXT_READY_TAIL])
		return 1;  /* Yes, this CPU has newly registered callbacks. */
	for (i = RCU_WAIT_TAIL; i < RCU_NEXT_TAIL; i++)
		if (rdp->nxttail[i - 1] != rdp->nxttail[i] &&
		    ULONG_CMP_LT(ACCESS_ONCE(rsp->completed),
				 rdp->nxtcompleted[i]))
			return 1;  /* Yes, CBs for future grace period. */
	return 0; /* No grace period needed. */
491 492
}

493
/*
494
 * rcu_eqs_enter_common - current CPU is moving towards extended quiescent state
495 496 497 498 499
 *
 * If the new value of the ->dynticks_nesting counter now is zero,
 * we really have entered idle, and must do the appropriate accounting.
 * The caller must have disabled interrupts.
 */
500 501
static void rcu_eqs_enter_common(struct rcu_dynticks *rdtp, long long oldval,
				bool user)
502
{
503 504 505
	struct rcu_state *rsp;
	struct rcu_data *rdp;

506
	trace_rcu_dyntick(TPS("Start"), oldval, rdtp->dynticks_nesting);
507
	if (!user && !is_idle_task(current)) {
508 509
		struct task_struct *idle __maybe_unused =
			idle_task(smp_processor_id());
510

511
		trace_rcu_dyntick(TPS("Error on entry: not idle task"), oldval, 0);
512
		ftrace_dump(DUMP_ORIG);
513 514 515
		WARN_ONCE(1, "Current pid: %d comm: %s / Idle pid: %d comm: %s",
			  current->pid, current->comm,
			  idle->pid, idle->comm); /* must be idle task! */
516
	}
517 518 519 520
	for_each_rcu_flavor(rsp) {
		rdp = this_cpu_ptr(rsp->rda);
		do_nocb_deferred_wakeup(rdp);
	}
521
	rcu_prepare_for_idle(smp_processor_id());
522
	/* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */
523
	smp_mb__before_atomic();  /* See above. */
524
	atomic_inc(&rdtp->dynticks);
525
	smp_mb__after_atomic();  /* Force ordering with next sojourn. */
526
	WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1);
527
	rcu_dynticks_task_enter();
528 529

	/*
530
	 * It is illegal to enter an extended quiescent state while
531 532 533 534 535 536 537 538
	 * in an RCU read-side critical section.
	 */
	rcu_lockdep_assert(!lock_is_held(&rcu_lock_map),
			   "Illegal idle entry in RCU read-side critical section.");
	rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map),
			   "Illegal idle entry in RCU-bh read-side critical section.");
	rcu_lockdep_assert(!lock_is_held(&rcu_sched_lock_map),
			   "Illegal idle entry in RCU-sched read-side critical section.");
539
}
540

541 542 543
/*
 * Enter an RCU extended quiescent state, which can be either the
 * idle loop or adaptive-tickless usermode execution.
544
 */
545
static void rcu_eqs_enter(bool user)
546
{
547
	long long oldval;
548 549
	struct rcu_dynticks *rdtp;

550
	rdtp = this_cpu_ptr(&rcu_dynticks);
551
	oldval = rdtp->dynticks_nesting;
552
	WARN_ON_ONCE((oldval & DYNTICK_TASK_NEST_MASK) == 0);
553
	if ((oldval & DYNTICK_TASK_NEST_MASK) == DYNTICK_TASK_NEST_VALUE) {
554
		rdtp->dynticks_nesting = 0;
555 556
		rcu_eqs_enter_common(rdtp, oldval, user);
	} else {
557
		rdtp->dynticks_nesting -= DYNTICK_TASK_NEST_VALUE;
558
	}
559
}
560 561 562 563 564 565 566 567 568 569 570 571 572 573 574

/**
 * rcu_idle_enter - inform RCU that current CPU is entering idle
 *
 * Enter idle mode, in other words, -leave- the mode in which RCU
 * read-side critical sections can occur.  (Though RCU read-side
 * critical sections can occur in irq handlers in idle, a possibility
 * handled by irq_enter() and irq_exit().)
 *
 * We crowbar the ->dynticks_nesting field to zero to allow for
 * the possibility of usermode upcalls having messed up our count
 * of interrupt nesting level during the prior busy period.
 */
void rcu_idle_enter(void)
{
575 576 577
	unsigned long flags;

	local_irq_save(flags);
578
	rcu_eqs_enter(false);
579
	rcu_sysidle_enter(this_cpu_ptr(&rcu_dynticks), 0);
580
	local_irq_restore(flags);
581
}
582
EXPORT_SYMBOL_GPL(rcu_idle_enter);
583

584
#ifdef CONFIG_RCU_USER_QS
585 586 587 588 589 590 591 592 593 594
/**
 * rcu_user_enter - inform RCU that we are resuming userspace.
 *
 * Enter RCU idle mode right before resuming userspace.  No use of RCU
 * is permitted between this call and rcu_user_exit(). This way the
 * CPU doesn't need to maintain the tick for RCU maintenance purposes
 * when the CPU runs in userspace.
 */
void rcu_user_enter(void)
{
595
	rcu_eqs_enter(1);
596
}
597
#endif /* CONFIG_RCU_USER_QS */
598

599 600 601 602 603 604
/**
 * rcu_irq_exit - inform RCU that current CPU is exiting irq towards idle
 *
 * Exit from an interrupt handler, which might possibly result in entering
 * idle mode, in other words, leaving the mode in which read-side critical
 * sections can occur.
605
 *
606 607 608 609 610 611 612 613
 * This code assumes that the idle loop never does anything that might
 * result in unbalanced calls to irq_enter() and irq_exit().  If your
 * architecture violates this assumption, RCU will give you what you
 * deserve, good and hard.  But very infrequently and irreproducibly.
 *
 * Use things like work queues to work around this limitation.
 *
 * You have been warned.
614
 */
615
void rcu_irq_exit(void)
616 617
{
	unsigned long flags;
618
	long long oldval;
619 620 621
	struct rcu_dynticks *rdtp;

	local_irq_save(flags);
622
	rdtp = this_cpu_ptr(&rcu_dynticks);
623
	oldval = rdtp->dynticks_nesting;
624 625
	rdtp->dynticks_nesting--;
	WARN_ON_ONCE(rdtp->dynticks_nesting < 0);
626
	if (rdtp->dynticks_nesting)
627
		trace_rcu_dyntick(TPS("--="), oldval, rdtp->dynticks_nesting);
628
	else
629
		rcu_eqs_enter_common(rdtp, oldval, true);
630
	rcu_sysidle_enter(rdtp, 1);
631 632 633 634
	local_irq_restore(flags);
}

/*
635
 * rcu_eqs_exit_common - current CPU moving away from extended quiescent state
636 637 638 639 640
 *
 * If the new value of the ->dynticks_nesting counter was previously zero,
 * we really have exited idle, and must do the appropriate accounting.
 * The caller must have disabled interrupts.
 */
641 642
static void rcu_eqs_exit_common(struct rcu_dynticks *rdtp, long long oldval,
			       int user)
643
{
644
	rcu_dynticks_task_exit();
645
	smp_mb__before_atomic();  /* Force ordering w/previous sojourn. */
646 647
	atomic_inc(&rdtp->dynticks);
	/* CPUs seeing atomic_inc() must see later RCU read-side crit sects */
648
	smp_mb__after_atomic();  /* See above. */
649
	WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1));
650
	rcu_cleanup_after_idle(smp_processor_id());
651
	trace_rcu_dyntick(TPS("End"), oldval, rdtp->dynticks_nesting);
652
	if (!user && !is_idle_task(current)) {
653 654
		struct task_struct *idle __maybe_unused =
			idle_task(smp_processor_id());
655

656
		trace_rcu_dyntick(TPS("Error on exit: not idle task"),
657
				  oldval, rdtp->dynticks_nesting);
658
		ftrace_dump(DUMP_ORIG);
659 660 661
		WARN_ONCE(1, "Current pid: %d comm: %s / Idle pid: %d comm: %s",
			  current->pid, current->comm,
			  idle->pid, idle->comm); /* must be idle task! */
662 663 664
	}
}

665 666 667
/*
 * Exit an RCU extended quiescent state, which can be either the
 * idle loop or adaptive-tickless usermode execution.
668
 */
669
static void rcu_eqs_exit(bool user)
670 671 672 673
{
	struct rcu_dynticks *rdtp;
	long long oldval;

674
	rdtp = this_cpu_ptr(&rcu_dynticks);
675
	oldval = rdtp->dynticks_nesting;
676
	WARN_ON_ONCE(oldval < 0);
677
	if (oldval & DYNTICK_TASK_NEST_MASK) {
678
		rdtp->dynticks_nesting += DYNTICK_TASK_NEST_VALUE;
679
	} else {
680
		rdtp->dynticks_nesting = DYNTICK_TASK_EXIT_IDLE;
681 682
		rcu_eqs_exit_common(rdtp, oldval, user);
	}
683
}
684 685 686 687 688 689 690 691 692 693 694 695 696 697

/**
 * rcu_idle_exit - inform RCU that current CPU is leaving idle
 *
 * Exit idle mode, in other words, -enter- the mode in which RCU
 * read-side critical sections can occur.
 *
 * We crowbar the ->dynticks_nesting field to DYNTICK_TASK_NEST to
 * allow for the possibility of usermode upcalls messing up our count
 * of interrupt nesting level during the busy period that is just
 * now starting.
 */
void rcu_idle_exit(void)
{
698 699 700
	unsigned long flags;

	local_irq_save(flags);
701
	rcu_eqs_exit(false);
702
	rcu_sysidle_exit(this_cpu_ptr(&rcu_dynticks), 0);
703
	local_irq_restore(flags);
704
}
705
EXPORT_SYMBOL_GPL(rcu_idle_exit);
706

707
#ifdef CONFIG_RCU_USER_QS
708 709 710 711 712 713 714 715
/**
 * rcu_user_exit - inform RCU that we are exiting userspace.
 *
 * Exit RCU idle mode while entering the kernel because it can
 * run a RCU read side critical section anytime.
 */
void rcu_user_exit(void)
{
716
	rcu_eqs_exit(1);
717
}
718
#endif /* CONFIG_RCU_USER_QS */
719

720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745
/**
 * rcu_irq_enter - inform RCU that current CPU is entering irq away from idle
 *
 * Enter an interrupt handler, which might possibly result in exiting
 * idle mode, in other words, entering the mode in which read-side critical
 * sections can occur.
 *
 * Note that the Linux kernel is fully capable of entering an interrupt
 * handler that it never exits, for example when doing upcalls to
 * user mode!  This code assumes that the idle loop never does upcalls to
 * user mode.  If your architecture does do upcalls from the idle loop (or
 * does anything else that results in unbalanced calls to the irq_enter()
 * and irq_exit() functions), RCU will give you what you deserve, good
 * and hard.  But very infrequently and irreproducibly.
 *
 * Use things like work queues to work around this limitation.
 *
 * You have been warned.
 */
void rcu_irq_enter(void)
{
	unsigned long flags;
	struct rcu_dynticks *rdtp;
	long long oldval;

	local_irq_save(flags);
746
	rdtp = this_cpu_ptr(&rcu_dynticks);
747 748 749
	oldval = rdtp->dynticks_nesting;
	rdtp->dynticks_nesting++;
	WARN_ON_ONCE(rdtp->dynticks_nesting == 0);
750
	if (oldval)
751
		trace_rcu_dyntick(TPS("++="), oldval, rdtp->dynticks_nesting);
752
	else
753
		rcu_eqs_exit_common(rdtp, oldval, true);
754
	rcu_sysidle_exit(rdtp, 1);
755 756 757 758 759 760 761 762 763 764 765 766
	local_irq_restore(flags);
}

/**
 * rcu_nmi_enter - inform RCU of entry to NMI context
 *
 * If the CPU was idle with dynamic ticks active, and there is no
 * irq handler running, this updates rdtp->dynticks_nmi to let the
 * RCU grace-period handling know that the CPU is active.
 */
void rcu_nmi_enter(void)
{
767
	struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
768

769 770
	if (rdtp->dynticks_nmi_nesting == 0 &&
	    (atomic_read(&rdtp->dynticks) & 0x1))
771
		return;
772
	rdtp->dynticks_nmi_nesting++;
773
	smp_mb__before_atomic();  /* Force delay from prior write. */
774 775
	atomic_inc(&rdtp->dynticks);
	/* CPUs seeing atomic_inc() must see later RCU read-side crit sects */
776
	smp_mb__after_atomic();  /* See above. */
777
	WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1));
778 779 780 781 782 783 784 785 786 787 788
}

/**
 * rcu_nmi_exit - inform RCU of exit from NMI context
 *
 * If the CPU was idle with dynamic ticks active, and there is no
 * irq handler running, this updates rdtp->dynticks_nmi to let the
 * RCU grace-period handling know that the CPU is no longer active.
 */
void rcu_nmi_exit(void)
{
789
	struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
790

791 792
	if (rdtp->dynticks_nmi_nesting == 0 ||
	    --rdtp->dynticks_nmi_nesting != 0)
793
		return;
794
	/* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */
795
	smp_mb__before_atomic();  /* See above. */
796
	atomic_inc(&rdtp->dynticks);
797
	smp_mb__after_atomic();  /* Force delay to next write. */
798
	WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1);
799 800 801
}

/**
802 803 804 805 806 807 808
 * __rcu_is_watching - are RCU read-side critical sections safe?
 *
 * Return true if RCU is watching the running CPU, which means that
 * this CPU can safely enter RCU read-side critical sections.  Unlike
 * rcu_is_watching(), the caller of __rcu_is_watching() must have at
 * least disabled preemption.
 */
809
bool notrace __rcu_is_watching(void)
810 811 812 813 814 815
{
	return atomic_read(this_cpu_ptr(&rcu_dynticks.dynticks)) & 0x1;
}

/**
 * rcu_is_watching - see if RCU thinks that the current CPU is idle
816
 *
817
 * If the current CPU is in its idle loop and is neither in an interrupt
818
 * or NMI handler, return true.
819
 */
820
bool notrace rcu_is_watching(void)
821
{
822
	bool ret;
823 824

	preempt_disable();
825
	ret = __rcu_is_watching();
826 827
	preempt_enable();
	return ret;
828
}
829
EXPORT_SYMBOL_GPL(rcu_is_watching);
830

831
#if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU)
832 833 834 835 836 837 838

/*
 * Is the current CPU online?  Disable preemption to avoid false positives
 * that could otherwise happen due to the current CPU number being sampled,
 * this task being preempted, its old CPU being taken offline, resuming
 * on some other CPU, then determining that its old CPU is now offline.
 * It is OK to use RCU on an offline processor during initial boot, hence
839 840 841 842 843 844 845 846 847 848 849
 * the check for rcu_scheduler_fully_active.  Note also that it is OK
 * for a CPU coming online to use RCU for one jiffy prior to marking itself
 * online in the cpu_online_mask.  Similarly, it is OK for a CPU going
 * offline to continue to use RCU for one jiffy after marking itself
 * offline in the cpu_online_mask.  This leniency is necessary given the
 * non-atomic nature of the online and offline processing, for example,
 * the fact that a CPU enters the scheduler after completing the CPU_DYING
 * notifiers.
 *
 * This is also why RCU internally marks CPUs online during the
 * CPU_UP_PREPARE phase and offline during the CPU_DEAD phase.
850 851 852 853 854 855
 *
 * Disable checking if in an NMI handler because we cannot safely report
 * errors from NMI handlers anyway.
 */
bool rcu_lockdep_current_cpu_online(void)
{
856 857
	struct rcu_data *rdp;
	struct rcu_node *rnp;
858 859 860
	bool ret;

	if (in_nmi())
F
Fengguang Wu 已提交
861
		return true;
862
	preempt_disable();
863
	rdp = this_cpu_ptr(&rcu_sched_data);
864 865
	rnp = rdp->mynode;
	ret = (rdp->grpmask & rnp->qsmaskinit) ||
866 867 868 869 870 871
	      !rcu_scheduler_fully_active;
	preempt_enable();
	return ret;
}
EXPORT_SYMBOL_GPL(rcu_lockdep_current_cpu_online);

872
#endif /* #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) */
873

874
/**
875
 * rcu_is_cpu_rrupt_from_idle - see if idle or immediately interrupted from idle
876
 *
877 878 879
 * If the current CPU is idle or running at a first-level (not nested)
 * interrupt from idle, return true.  The caller must have at least
 * disabled preemption.
880
 */
881
static int rcu_is_cpu_rrupt_from_idle(void)
882
{
883
	return __this_cpu_read(rcu_dynticks.dynticks_nesting) <= 1;
884 885 886 887 888
}

/*
 * Snapshot the specified CPU's dynticks counter so that we can later
 * credit them with an implicit quiescent state.  Return 1 if this CPU
889
 * is in dynticks idle mode, which is an extended quiescent state.
890
 */
891 892
static int dyntick_save_progress_counter(struct rcu_data *rdp,
					 bool *isidle, unsigned long *maxj)
893
{
894
	rdp->dynticks_snap = atomic_add_return(0, &rdp->dynticks->dynticks);
895
	rcu_sysidle_check_cpu(rdp, isidle, maxj);
896 897 898 899 900 901
	if ((rdp->dynticks_snap & 0x1) == 0) {
		trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("dti"));
		return 1;
	} else {
		return 0;
	}
902 903
}

904 905 906 907 908 909
/*
 * This function really isn't for public consumption, but RCU is special in
 * that context switches can allow the state machine to make progress.
 */
extern void resched_cpu(int cpu);

910 911 912 913
/*
 * Return true if the specified CPU has passed through a quiescent
 * state by virtue of being in or having passed through an dynticks
 * idle state since the last call to dyntick_save_progress_counter()
914
 * for this same CPU, or by virtue of having been offline.
915
 */
916 917
static int rcu_implicit_dynticks_qs(struct rcu_data *rdp,
				    bool *isidle, unsigned long *maxj)
918
{
919
	unsigned int curr;
920
	int *rcrmp;
921
	unsigned int snap;
922

923 924
	curr = (unsigned int)atomic_add_return(0, &rdp->dynticks->dynticks);
	snap = (unsigned int)rdp->dynticks_snap;
925 926 927 928 929 930 931 932 933

	/*
	 * If the CPU passed through or entered a dynticks idle phase with
	 * no active irq/NMI handlers, then we can safely pretend that the CPU
	 * already acknowledged the request to pass through a quiescent
	 * state.  Either way, that CPU cannot possibly be in an RCU
	 * read-side critical section that started before the beginning
	 * of the current RCU grace period.
	 */
934
	if ((curr & 0x1) == 0 || UINT_CMP_GE(curr, snap + 2)) {
935
		trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("dti"));
936 937 938 939
		rdp->dynticks_fqs++;
		return 1;
	}

940 941 942 943 944 945 946 947 948 949 950 951 952 953 954
	/*
	 * Check for the CPU being offline, but only if the grace period
	 * is old enough.  We don't need to worry about the CPU changing
	 * state: If we see it offline even once, it has been through a
	 * quiescent state.
	 *
	 * The reason for insisting that the grace period be at least
	 * one jiffy old is that CPUs that are not quite online and that
	 * have just gone offline can still execute RCU read-side critical
	 * sections.
	 */
	if (ULONG_CMP_GE(rdp->rsp->gp_start + 2, jiffies))
		return 0;  /* Grace period is not old enough. */
	barrier();
	if (cpu_is_offline(rdp->cpu)) {
955
		trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("ofl"));
956 957 958
		rdp->offline_fqs++;
		return 1;
	}
959 960

	/*
961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979
	 * A CPU running for an extended time within the kernel can
	 * delay RCU grace periods.  When the CPU is in NO_HZ_FULL mode,
	 * even context-switching back and forth between a pair of
	 * in-kernel CPU-bound tasks cannot advance grace periods.
	 * So if the grace period is old enough, make the CPU pay attention.
	 * Note that the unsynchronized assignments to the per-CPU
	 * rcu_sched_qs_mask variable are safe.  Yes, setting of
	 * bits can be lost, but they will be set again on the next
	 * force-quiescent-state pass.  So lost bit sets do not result
	 * in incorrect behavior, merely in a grace period lasting
	 * a few jiffies longer than it might otherwise.  Because
	 * there are at most four threads involved, and because the
	 * updates are only once every few jiffies, the probability of
	 * lossage (and thus of slight grace-period extension) is
	 * quite low.
	 *
	 * Note that if the jiffies_till_sched_qs boot/sysfs parameter
	 * is set too high, we override with half of the RCU CPU stall
	 * warning delay.
980
	 */
981 982 983
	rcrmp = &per_cpu(rcu_sched_qs_mask, rdp->cpu);
	if (ULONG_CMP_GE(jiffies,
			 rdp->rsp->gp_start + jiffies_till_sched_qs) ||
984
	    ULONG_CMP_GE(jiffies, rdp->rsp->jiffies_resched)) {
985 986 987 988 989 990 991 992 993 994 995 996 997
		if (!(ACCESS_ONCE(*rcrmp) & rdp->rsp->flavor_mask)) {
			ACCESS_ONCE(rdp->cond_resched_completed) =
				ACCESS_ONCE(rdp->mynode->completed);
			smp_mb(); /* ->cond_resched_completed before *rcrmp. */
			ACCESS_ONCE(*rcrmp) =
				ACCESS_ONCE(*rcrmp) + rdp->rsp->flavor_mask;
			resched_cpu(rdp->cpu);  /* Force CPU into scheduler. */
			rdp->rsp->jiffies_resched += 5; /* Enable beating. */
		} else if (ULONG_CMP_GE(jiffies, rdp->rsp->jiffies_resched)) {
			/* Time to beat on that CPU again! */
			resched_cpu(rdp->cpu);  /* Force CPU into scheduler. */
			rdp->rsp->jiffies_resched += 5; /* Re-enable beating. */
		}
998 999
	}

1000
	return 0;
1001 1002 1003 1004
}

static void record_gp_stall_check_time(struct rcu_state *rsp)
{
1005
	unsigned long j = jiffies;
1006
	unsigned long j1;
1007 1008 1009

	rsp->gp_start = j;
	smp_wmb(); /* Record start time before stall time. */
1010
	j1 = rcu_jiffies_till_stall_check();
1011
	ACCESS_ONCE(rsp->jiffies_stall) = j + j1;
1012
	rsp->jiffies_resched = j + j1 / 2;
1013 1014
}

1015
/*
1016
 * Dump stacks of all tasks running on stalled CPUs.
1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034
 */
static void rcu_dump_cpu_stacks(struct rcu_state *rsp)
{
	int cpu;
	unsigned long flags;
	struct rcu_node *rnp;

	rcu_for_each_leaf_node(rsp, rnp) {
		raw_spin_lock_irqsave(&rnp->lock, flags);
		if (rnp->qsmask != 0) {
			for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++)
				if (rnp->qsmask & (1UL << cpu))
					dump_cpu_task(rnp->grplo + cpu);
		}
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
	}
}

1035 1036 1037 1038 1039
static void print_other_cpu_stall(struct rcu_state *rsp)
{
	int cpu;
	long delta;
	unsigned long flags;
1040
	int ndetected = 0;
1041
	struct rcu_node *rnp = rcu_get_root(rsp);
1042
	long totqlen = 0;
1043 1044 1045

	/* Only let one CPU complain about others per time interval. */

P
Paul E. McKenney 已提交
1046
	raw_spin_lock_irqsave(&rnp->lock, flags);
1047
	delta = jiffies - ACCESS_ONCE(rsp->jiffies_stall);
1048
	if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) {
P
Paul E. McKenney 已提交
1049
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
1050 1051
		return;
	}
1052
	ACCESS_ONCE(rsp->jiffies_stall) = jiffies + 3 * rcu_jiffies_till_stall_check() + 3;
P
Paul E. McKenney 已提交
1053
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
1054

1055 1056 1057 1058 1059
	/*
	 * OK, time to rat on our buddy...
	 * See Documentation/RCU/stallwarn.txt for info on how to debug
	 * RCU CPU stall warnings.
	 */
1060
	pr_err("INFO: %s detected stalls on CPUs/tasks:",
1061
	       rsp->name);
1062
	print_cpu_stall_info_begin();
1063
	rcu_for_each_leaf_node(rsp, rnp) {
1064
		raw_spin_lock_irqsave(&rnp->lock, flags);
1065
		ndetected += rcu_print_task_stall(rnp);
1066 1067 1068 1069 1070 1071 1072 1073
		if (rnp->qsmask != 0) {
			for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++)
				if (rnp->qsmask & (1UL << cpu)) {
					print_cpu_stall_info(rsp,
							     rnp->grplo + cpu);
					ndetected++;
				}
		}
1074
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
1075
	}
1076 1077 1078 1079 1080 1081 1082

	/*
	 * Now rat on any tasks that got kicked up to the root rcu_node
	 * due to CPU offlining.
	 */
	rnp = rcu_get_root(rsp);
	raw_spin_lock_irqsave(&rnp->lock, flags);
1083
	ndetected += rcu_print_task_stall(rnp);
1084 1085 1086
	raw_spin_unlock_irqrestore(&rnp->lock, flags);

	print_cpu_stall_info_end();
1087 1088
	for_each_possible_cpu(cpu)
		totqlen += per_cpu_ptr(rsp->rda, cpu)->qlen;
1089
	pr_cont("(detected by %d, t=%ld jiffies, g=%ld, c=%ld, q=%lu)\n",
1090
	       smp_processor_id(), (long)(jiffies - rsp->gp_start),
1091
	       (long)rsp->gpnum, (long)rsp->completed, totqlen);
1092
	if (ndetected == 0)
1093
		pr_err("INFO: Stall ended before state dump start\n");
1094
	else
1095
		rcu_dump_cpu_stacks(rsp);
1096

1097
	/* Complain about tasks blocking the grace period. */
1098 1099 1100

	rcu_print_detail_task_stall(rsp);

1101
	force_quiescent_state(rsp);  /* Kick them all. */
1102 1103 1104 1105
}

static void print_cpu_stall(struct rcu_state *rsp)
{
1106
	int cpu;
1107 1108
	unsigned long flags;
	struct rcu_node *rnp = rcu_get_root(rsp);
1109
	long totqlen = 0;
1110

1111 1112 1113 1114 1115
	/*
	 * OK, time to rat on ourselves...
	 * See Documentation/RCU/stallwarn.txt for info on how to debug
	 * RCU CPU stall warnings.
	 */
1116
	pr_err("INFO: %s self-detected stall on CPU", rsp->name);
1117 1118 1119
	print_cpu_stall_info_begin();
	print_cpu_stall_info(rsp, smp_processor_id());
	print_cpu_stall_info_end();
1120 1121
	for_each_possible_cpu(cpu)
		totqlen += per_cpu_ptr(rsp->rda, cpu)->qlen;
1122 1123 1124
	pr_cont(" (t=%lu jiffies g=%ld c=%ld q=%lu)\n",
		jiffies - rsp->gp_start,
		(long)rsp->gpnum, (long)rsp->completed, totqlen);
1125
	rcu_dump_cpu_stacks(rsp);
1126

P
Paul E. McKenney 已提交
1127
	raw_spin_lock_irqsave(&rnp->lock, flags);
1128 1129
	if (ULONG_CMP_GE(jiffies, ACCESS_ONCE(rsp->jiffies_stall)))
		ACCESS_ONCE(rsp->jiffies_stall) = jiffies +
1130
				     3 * rcu_jiffies_till_stall_check() + 3;
P
Paul E. McKenney 已提交
1131
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
1132

1133 1134 1135 1136 1137 1138 1139 1140
	/*
	 * Attempt to revive the RCU machinery by forcing a context switch.
	 *
	 * A context switch would normally allow the RCU state machine to make
	 * progress and it could be we're stuck in kernel space without context
	 * switches for an entirely unreasonable amount of time.
	 */
	resched_cpu(smp_processor_id());
1141 1142 1143 1144
}

static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
{
1145 1146 1147
	unsigned long completed;
	unsigned long gpnum;
	unsigned long gps;
1148 1149
	unsigned long j;
	unsigned long js;
1150 1151
	struct rcu_node *rnp;

1152
	if (rcu_cpu_stall_suppress || !rcu_gp_in_progress(rsp))
1153
		return;
1154
	j = jiffies;
1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174

	/*
	 * Lots of memory barriers to reject false positives.
	 *
	 * The idea is to pick up rsp->gpnum, then rsp->jiffies_stall,
	 * then rsp->gp_start, and finally rsp->completed.  These values
	 * are updated in the opposite order with memory barriers (or
	 * equivalent) during grace-period initialization and cleanup.
	 * Now, a false positive can occur if we get an new value of
	 * rsp->gp_start and a old value of rsp->jiffies_stall.  But given
	 * the memory barriers, the only way that this can happen is if one
	 * grace period ends and another starts between these two fetches.
	 * Detect this by comparing rsp->completed with the previous fetch
	 * from rsp->gpnum.
	 *
	 * Given this check, comparisons of jiffies, rsp->jiffies_stall,
	 * and rsp->gp_start suffice to forestall false positives.
	 */
	gpnum = ACCESS_ONCE(rsp->gpnum);
	smp_rmb(); /* Pick up ->gpnum first... */
1175
	js = ACCESS_ONCE(rsp->jiffies_stall);
1176 1177 1178 1179 1180 1181 1182 1183
	smp_rmb(); /* ...then ->jiffies_stall before the rest... */
	gps = ACCESS_ONCE(rsp->gp_start);
	smp_rmb(); /* ...and finally ->gp_start before ->completed. */
	completed = ACCESS_ONCE(rsp->completed);
	if (ULONG_CMP_GE(completed, gpnum) ||
	    ULONG_CMP_LT(j, js) ||
	    ULONG_CMP_GE(gps, js))
		return; /* No stall or GP completed since entering function. */
1184
	rnp = rdp->mynode;
1185
	if (rcu_gp_in_progress(rsp) &&
1186
	    (ACCESS_ONCE(rnp->qsmask) & rdp->grpmask)) {
1187 1188 1189 1190

		/* We haven't checked in, so go dump stack. */
		print_cpu_stall(rsp);

1191 1192
	} else if (rcu_gp_in_progress(rsp) &&
		   ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY)) {
1193

1194
		/* They had a few time units to dump stack, so complain. */
1195 1196 1197 1198
		print_other_cpu_stall(rsp);
	}
}

1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209
/**
 * rcu_cpu_stall_reset - prevent further stall warnings in current grace period
 *
 * Set the stall-warning timeout way off into the future, thus preventing
 * any RCU CPU stall-warning messages from appearing in the current set of
 * RCU grace periods.
 *
 * The caller must disable hard irqs.
 */
void rcu_cpu_stall_reset(void)
{
1210 1211 1212
	struct rcu_state *rsp;

	for_each_rcu_flavor(rsp)
1213
		ACCESS_ONCE(rsp->jiffies_stall) = jiffies + ULONG_MAX / 2;
1214 1215
}

1216 1217 1218 1219 1220 1221 1222
/*
 * Initialize the specified rcu_data structure's callback list to empty.
 */
static void init_callback_list(struct rcu_data *rdp)
{
	int i;

1223 1224
	if (init_nocb_callback_list(rdp))
		return;
1225 1226 1227 1228 1229
	rdp->nxtlist = NULL;
	for (i = 0; i < RCU_NEXT_SIZE; i++)
		rdp->nxttail[i] = &rdp->nxtlist;
}

1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258
/*
 * Determine the value that ->completed will have at the end of the
 * next subsequent grace period.  This is used to tag callbacks so that
 * a CPU can invoke callbacks in a timely fashion even if that CPU has
 * been dyntick-idle for an extended period with callbacks under the
 * influence of RCU_FAST_NO_HZ.
 *
 * The caller must hold rnp->lock with interrupts disabled.
 */
static unsigned long rcu_cbs_completed(struct rcu_state *rsp,
				       struct rcu_node *rnp)
{
	/*
	 * If RCU is idle, we just wait for the next grace period.
	 * But we can only be sure that RCU is idle if we are looking
	 * at the root rcu_node structure -- otherwise, a new grace
	 * period might have started, but just not yet gotten around
	 * to initializing the current non-root rcu_node structure.
	 */
	if (rcu_get_root(rsp) == rnp && rnp->gpnum == rnp->completed)
		return rnp->completed + 1;

	/*
	 * Otherwise, wait for a possible partial grace period and
	 * then the subsequent full grace period.
	 */
	return rnp->completed + 2;
}

1259 1260 1261 1262 1263
/*
 * Trace-event helper function for rcu_start_future_gp() and
 * rcu_nocb_wait_gp().
 */
static void trace_rcu_future_gp(struct rcu_node *rnp, struct rcu_data *rdp,
1264
				unsigned long c, const char *s)
1265 1266 1267 1268 1269 1270 1271 1272 1273
{
	trace_rcu_future_grace_period(rdp->rsp->name, rnp->gpnum,
				      rnp->completed, c, rnp->level,
				      rnp->grplo, rnp->grphi, s);
}

/*
 * Start some future grace period, as needed to handle newly arrived
 * callbacks.  The required future grace periods are recorded in each
1274 1275
 * rcu_node structure's ->need_future_gp field.  Returns true if there
 * is reason to awaken the grace-period kthread.
1276 1277 1278
 *
 * The caller must hold the specified rcu_node structure's ->lock.
 */
1279 1280 1281
static bool __maybe_unused
rcu_start_future_gp(struct rcu_node *rnp, struct rcu_data *rdp,
		    unsigned long *c_out)
1282 1283 1284
{
	unsigned long c;
	int i;
1285
	bool ret = false;
1286 1287 1288 1289 1290 1291 1292
	struct rcu_node *rnp_root = rcu_get_root(rdp->rsp);

	/*
	 * Pick up grace-period number for new callbacks.  If this
	 * grace period is already marked as needed, return to the caller.
	 */
	c = rcu_cbs_completed(rdp->rsp, rnp);
1293
	trace_rcu_future_gp(rnp, rdp, c, TPS("Startleaf"));
1294
	if (rnp->need_future_gp[c & 0x1]) {
1295
		trace_rcu_future_gp(rnp, rdp, c, TPS("Prestartleaf"));
1296
		goto out;
1297 1298 1299 1300 1301 1302 1303
	}

	/*
	 * If either this rcu_node structure or the root rcu_node structure
	 * believe that a grace period is in progress, then we must wait
	 * for the one following, which is in "c".  Because our request
	 * will be noticed at the end of the current grace period, we don't
1304 1305 1306 1307 1308 1309 1310
	 * need to explicitly start one.  We only do the lockless check
	 * of rnp_root's fields if the current rcu_node structure thinks
	 * there is no grace period in flight, and because we hold rnp->lock,
	 * the only possible change is when rnp_root's two fields are
	 * equal, in which case rnp_root->gpnum might be concurrently
	 * incremented.  But that is OK, as it will just result in our
	 * doing some extra useless work.
1311 1312
	 */
	if (rnp->gpnum != rnp->completed ||
1313
	    ACCESS_ONCE(rnp_root->gpnum) != ACCESS_ONCE(rnp_root->completed)) {
1314
		rnp->need_future_gp[c & 0x1]++;
1315
		trace_rcu_future_gp(rnp, rdp, c, TPS("Startedleaf"));
1316
		goto out;
1317 1318 1319 1320 1321 1322 1323
	}

	/*
	 * There might be no grace period in progress.  If we don't already
	 * hold it, acquire the root rcu_node structure's lock in order to
	 * start one (if needed).
	 */
1324
	if (rnp != rnp_root) {
1325
		raw_spin_lock(&rnp_root->lock);
1326 1327
		smp_mb__after_unlock_lock();
	}
1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344

	/*
	 * Get a new grace-period number.  If there really is no grace
	 * period in progress, it will be smaller than the one we obtained
	 * earlier.  Adjust callbacks as needed.  Note that even no-CBs
	 * CPUs have a ->nxtcompleted[] array, so no no-CBs checks needed.
	 */
	c = rcu_cbs_completed(rdp->rsp, rnp_root);
	for (i = RCU_DONE_TAIL; i < RCU_NEXT_TAIL; i++)
		if (ULONG_CMP_LT(c, rdp->nxtcompleted[i]))
			rdp->nxtcompleted[i] = c;

	/*
	 * If the needed for the required grace period is already
	 * recorded, trace and leave.
	 */
	if (rnp_root->need_future_gp[c & 0x1]) {
1345
		trace_rcu_future_gp(rnp, rdp, c, TPS("Prestartedroot"));
1346 1347 1348 1349 1350 1351 1352 1353
		goto unlock_out;
	}

	/* Record the need for the future grace period. */
	rnp_root->need_future_gp[c & 0x1]++;

	/* If a grace period is not already in progress, start one. */
	if (rnp_root->gpnum != rnp_root->completed) {
1354
		trace_rcu_future_gp(rnp, rdp, c, TPS("Startedleafroot"));
1355
	} else {
1356
		trace_rcu_future_gp(rnp, rdp, c, TPS("Startedroot"));
1357
		ret = rcu_start_gp_advanced(rdp->rsp, rnp_root, rdp);
1358 1359 1360 1361
	}
unlock_out:
	if (rnp != rnp_root)
		raw_spin_unlock(&rnp_root->lock);
1362 1363 1364 1365
out:
	if (c_out != NULL)
		*c_out = c;
	return ret;
1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382
}

/*
 * Clean up any old requests for the just-ended grace period.  Also return
 * whether any additional grace periods have been requested.  Also invoke
 * rcu_nocb_gp_cleanup() in order to wake up any no-callbacks kthreads
 * waiting for this grace period to complete.
 */
static int rcu_future_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp)
{
	int c = rnp->completed;
	int needmore;
	struct rcu_data *rdp = this_cpu_ptr(rsp->rda);

	rcu_nocb_gp_cleanup(rsp, rnp);
	rnp->need_future_gp[c & 0x1] = 0;
	needmore = rnp->need_future_gp[(c + 1) & 0x1];
1383 1384
	trace_rcu_future_gp(rnp, rdp, c,
			    needmore ? TPS("CleanupMore") : TPS("Cleanup"));
1385 1386 1387
	return needmore;
}

1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403
/*
 * Awaken the grace-period kthread for the specified flavor of RCU.
 * Don't do a self-awaken, and don't bother awakening when there is
 * nothing for the grace-period kthread to do (as in several CPUs
 * raced to awaken, and we lost), and finally don't try to awaken
 * a kthread that has not yet been created.
 */
static void rcu_gp_kthread_wake(struct rcu_state *rsp)
{
	if (current == rsp->gp_kthread ||
	    !ACCESS_ONCE(rsp->gp_flags) ||
	    !rsp->gp_kthread)
		return;
	wake_up(&rsp->gp_wq);
}

1404 1405 1406 1407 1408 1409 1410
/*
 * If there is room, assign a ->completed number to any callbacks on
 * this CPU that have not already been assigned.  Also accelerate any
 * callbacks that were previously assigned a ->completed number that has
 * since proven to be too conservative, which can happen if callbacks get
 * assigned a ->completed number while RCU is idle, but with reference to
 * a non-root rcu_node structure.  This function is idempotent, so it does
1411 1412
 * not hurt to call it repeatedly.  Returns an flag saying that we should
 * awaken the RCU grace-period kthread.
1413 1414 1415
 *
 * The caller must hold rnp->lock with interrupts disabled.
 */
1416
static bool rcu_accelerate_cbs(struct rcu_state *rsp, struct rcu_node *rnp,
1417 1418 1419 1420
			       struct rcu_data *rdp)
{
	unsigned long c;
	int i;
1421
	bool ret;
1422 1423 1424

	/* If the CPU has no callbacks, nothing to do. */
	if (!rdp->nxttail[RCU_NEXT_TAIL] || !*rdp->nxttail[RCU_DONE_TAIL])
1425
		return false;
1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453

	/*
	 * Starting from the sublist containing the callbacks most
	 * recently assigned a ->completed number and working down, find the
	 * first sublist that is not assignable to an upcoming grace period.
	 * Such a sublist has something in it (first two tests) and has
	 * a ->completed number assigned that will complete sooner than
	 * the ->completed number for newly arrived callbacks (last test).
	 *
	 * The key point is that any later sublist can be assigned the
	 * same ->completed number as the newly arrived callbacks, which
	 * means that the callbacks in any of these later sublist can be
	 * grouped into a single sublist, whether or not they have already
	 * been assigned a ->completed number.
	 */
	c = rcu_cbs_completed(rsp, rnp);
	for (i = RCU_NEXT_TAIL - 1; i > RCU_DONE_TAIL; i--)
		if (rdp->nxttail[i] != rdp->nxttail[i - 1] &&
		    !ULONG_CMP_GE(rdp->nxtcompleted[i], c))
			break;

	/*
	 * If there are no sublist for unassigned callbacks, leave.
	 * At the same time, advance "i" one sublist, so that "i" will
	 * index into the sublist where all the remaining callbacks should
	 * be grouped into.
	 */
	if (++i >= RCU_NEXT_TAIL)
1454
		return false;
1455 1456 1457 1458 1459 1460 1461 1462 1463 1464

	/*
	 * Assign all subsequent callbacks' ->completed number to the next
	 * full grace period and group them all in the sublist initially
	 * indexed by "i".
	 */
	for (; i <= RCU_NEXT_TAIL; i++) {
		rdp->nxttail[i] = rdp->nxttail[RCU_NEXT_TAIL];
		rdp->nxtcompleted[i] = c;
	}
1465
	/* Record any needed additional grace periods. */
1466
	ret = rcu_start_future_gp(rnp, rdp, NULL);
1467 1468 1469

	/* Trace depending on how much we were able to accelerate. */
	if (!*rdp->nxttail[RCU_WAIT_TAIL])
1470
		trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("AccWaitCB"));
1471
	else
1472
		trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("AccReadyCB"));
1473
	return ret;
1474 1475 1476 1477 1478 1479 1480 1481
}

/*
 * Move any callbacks whose grace period has completed to the
 * RCU_DONE_TAIL sublist, then compact the remaining sublists and
 * assign ->completed numbers to any callbacks in the RCU_NEXT_TAIL
 * sublist.  This function is idempotent, so it does not hurt to
 * invoke it repeatedly.  As long as it is not invoked -too- often...
1482
 * Returns true if the RCU grace-period kthread needs to be awakened.
1483 1484 1485
 *
 * The caller must hold rnp->lock with interrupts disabled.
 */
1486
static bool rcu_advance_cbs(struct rcu_state *rsp, struct rcu_node *rnp,
1487 1488 1489 1490 1491 1492
			    struct rcu_data *rdp)
{
	int i, j;

	/* If the CPU has no callbacks, nothing to do. */
	if (!rdp->nxttail[RCU_NEXT_TAIL] || !*rdp->nxttail[RCU_DONE_TAIL])
1493
		return false;
1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516

	/*
	 * Find all callbacks whose ->completed numbers indicate that they
	 * are ready to invoke, and put them into the RCU_DONE_TAIL sublist.
	 */
	for (i = RCU_WAIT_TAIL; i < RCU_NEXT_TAIL; i++) {
		if (ULONG_CMP_LT(rnp->completed, rdp->nxtcompleted[i]))
			break;
		rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[i];
	}
	/* Clean up any sublist tail pointers that were misordered above. */
	for (j = RCU_WAIT_TAIL; j < i; j++)
		rdp->nxttail[j] = rdp->nxttail[RCU_DONE_TAIL];

	/* Copy down callbacks to fill in empty sublists. */
	for (j = RCU_WAIT_TAIL; i < RCU_NEXT_TAIL; i++, j++) {
		if (rdp->nxttail[j] == rdp->nxttail[RCU_NEXT_TAIL])
			break;
		rdp->nxttail[j] = rdp->nxttail[i];
		rdp->nxtcompleted[j] = rdp->nxtcompleted[i];
	}

	/* Classify any remaining callbacks. */
1517
	return rcu_accelerate_cbs(rsp, rnp, rdp);
1518 1519
}

1520
/*
1521 1522 1523
 * Update CPU-local rcu_data state to record the beginnings and ends of
 * grace periods.  The caller must hold the ->lock of the leaf rcu_node
 * structure corresponding to the current CPU, and must have irqs disabled.
1524
 * Returns true if the grace-period kthread needs to be awakened.
1525
 */
1526 1527
static bool __note_gp_changes(struct rcu_state *rsp, struct rcu_node *rnp,
			      struct rcu_data *rdp)
1528
{
1529 1530
	bool ret;

1531
	/* Handle the ends of any preceding grace periods first. */
1532
	if (rdp->completed == rnp->completed) {
1533

1534
		/* No grace period end, so just accelerate recent callbacks. */
1535
		ret = rcu_accelerate_cbs(rsp, rnp, rdp);
1536

1537 1538 1539
	} else {

		/* Advance callbacks. */
1540
		ret = rcu_advance_cbs(rsp, rnp, rdp);
1541 1542 1543

		/* Remember that we saw this grace-period completion. */
		rdp->completed = rnp->completed;
1544
		trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpuend"));
1545
	}
1546

1547 1548 1549 1550 1551 1552 1553
	if (rdp->gpnum != rnp->gpnum) {
		/*
		 * If the current grace period is waiting for this CPU,
		 * set up to detect a quiescent state, otherwise don't
		 * go looking for one.
		 */
		rdp->gpnum = rnp->gpnum;
1554
		trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpustart"));
1555 1556 1557 1558
		rdp->passed_quiesce = 0;
		rdp->qs_pending = !!(rnp->qsmask & rdp->grpmask);
		zero_cpu_stall_ticks(rdp);
	}
1559
	return ret;
1560 1561
}

1562
static void note_gp_changes(struct rcu_state *rsp, struct rcu_data *rdp)
1563 1564
{
	unsigned long flags;
1565
	bool needwake;
1566 1567 1568 1569
	struct rcu_node *rnp;

	local_irq_save(flags);
	rnp = rdp->mynode;
1570 1571
	if ((rdp->gpnum == ACCESS_ONCE(rnp->gpnum) &&
	     rdp->completed == ACCESS_ONCE(rnp->completed)) || /* w/out lock. */
1572 1573 1574 1575
	    !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */
		local_irq_restore(flags);
		return;
	}
1576
	smp_mb__after_unlock_lock();
1577
	needwake = __note_gp_changes(rsp, rnp, rdp);
1578
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
1579 1580
	if (needwake)
		rcu_gp_kthread_wake(rsp);
1581 1582
}

1583
/*
1584
 * Initialize a new grace period.  Return 0 if no grace period required.
1585
 */
1586
static int rcu_gp_init(struct rcu_state *rsp)
1587 1588
{
	struct rcu_data *rdp;
1589
	struct rcu_node *rnp = rcu_get_root(rsp);
1590

1591
	rcu_bind_gp_kthread();
1592
	raw_spin_lock_irq(&rnp->lock);
1593
	smp_mb__after_unlock_lock();
1594
	if (!ACCESS_ONCE(rsp->gp_flags)) {
1595 1596 1597 1598
		/* Spurious wakeup, tell caller to go back to sleep.  */
		raw_spin_unlock_irq(&rnp->lock);
		return 0;
	}
1599
	ACCESS_ONCE(rsp->gp_flags) = 0; /* Clear all flags: New grace period. */
1600

1601 1602 1603 1604 1605
	if (WARN_ON_ONCE(rcu_gp_in_progress(rsp))) {
		/*
		 * Grace period already in progress, don't start another.
		 * Not supposed to be able to happen.
		 */
1606 1607 1608 1609 1610
		raw_spin_unlock_irq(&rnp->lock);
		return 0;
	}

	/* Advance to a new grace period and initialize state. */
1611
	record_gp_stall_check_time(rsp);
1612 1613
	/* Record GP times before starting GP, hence smp_store_release(). */
	smp_store_release(&rsp->gpnum, rsp->gpnum + 1);
1614
	trace_rcu_grace_period(rsp->name, rsp->gpnum, TPS("start"));
1615 1616 1617
	raw_spin_unlock_irq(&rnp->lock);

	/* Exclude any concurrent CPU-hotplug operations. */
1618
	mutex_lock(&rsp->onoff_mutex);
1619
	smp_mb__after_unlock_lock(); /* ->gpnum increment before GP! */
1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634

	/*
	 * Set the quiescent-state-needed bits in all the rcu_node
	 * structures for all currently online CPUs in breadth-first order,
	 * starting from the root rcu_node structure, relying on the layout
	 * of the tree within the rsp->node[] array.  Note that other CPUs
	 * will access only the leaves of the hierarchy, thus seeing that no
	 * grace period is in progress, at least until the corresponding
	 * leaf node has been initialized.  In addition, we have excluded
	 * CPU-hotplug operations.
	 *
	 * The grace period cannot complete until the initialization
	 * process finishes, because this kthread handles both.
	 */
	rcu_for_each_node_breadth_first(rsp, rnp) {
1635
		raw_spin_lock_irq(&rnp->lock);
1636
		smp_mb__after_unlock_lock();
1637
		rdp = this_cpu_ptr(rsp->rda);
1638 1639
		rcu_preempt_check_blocked_tasks(rnp);
		rnp->qsmask = rnp->qsmaskinit;
1640
		ACCESS_ONCE(rnp->gpnum) = rsp->gpnum;
1641
		WARN_ON_ONCE(rnp->completed != rsp->completed);
1642
		ACCESS_ONCE(rnp->completed) = rsp->completed;
1643
		if (rnp == rdp->mynode)
1644
			(void)__note_gp_changes(rsp, rnp, rdp);
1645 1646 1647 1648 1649
		rcu_preempt_boost_start_gp(rnp);
		trace_rcu_grace_period_init(rsp->name, rnp->gpnum,
					    rnp->level, rnp->grplo,
					    rnp->grphi, rnp->qsmask);
		raw_spin_unlock_irq(&rnp->lock);
1650
		cond_resched_rcu_qs();
1651
	}
1652

1653
	mutex_unlock(&rsp->onoff_mutex);
1654 1655
	return 1;
}
1656

1657 1658 1659
/*
 * Do one round of quiescent-state forcing.
 */
1660
static int rcu_gp_fqs(struct rcu_state *rsp, int fqs_state_in)
1661 1662
{
	int fqs_state = fqs_state_in;
1663 1664
	bool isidle = false;
	unsigned long maxj;
1665 1666 1667 1668 1669
	struct rcu_node *rnp = rcu_get_root(rsp);

	rsp->n_force_qs++;
	if (fqs_state == RCU_SAVE_DYNTICK) {
		/* Collect dyntick-idle snapshots. */
1670
		if (is_sysidle_rcu_state(rsp)) {
1671
			isidle = true;
1672 1673
			maxj = jiffies - ULONG_MAX / 4;
		}
1674 1675
		force_qs_rnp(rsp, dyntick_save_progress_counter,
			     &isidle, &maxj);
1676
		rcu_sysidle_report_gp(rsp, isidle, maxj);
1677 1678 1679
		fqs_state = RCU_FORCE_QS;
	} else {
		/* Handle dyntick-idle and offline CPUs. */
1680
		isidle = false;
1681
		force_qs_rnp(rsp, rcu_implicit_dynticks_qs, &isidle, &maxj);
1682 1683 1684 1685
	}
	/* Clear flag to prevent immediate re-entry. */
	if (ACCESS_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) {
		raw_spin_lock_irq(&rnp->lock);
1686
		smp_mb__after_unlock_lock();
1687 1688
		ACCESS_ONCE(rsp->gp_flags) =
			ACCESS_ONCE(rsp->gp_flags) & ~RCU_GP_FLAG_FQS;
1689 1690 1691 1692 1693
		raw_spin_unlock_irq(&rnp->lock);
	}
	return fqs_state;
}

1694 1695 1696
/*
 * Clean up after the old grace period.
 */
1697
static void rcu_gp_cleanup(struct rcu_state *rsp)
1698 1699
{
	unsigned long gp_duration;
1700
	bool needgp = false;
1701
	int nocb = 0;
1702 1703
	struct rcu_data *rdp;
	struct rcu_node *rnp = rcu_get_root(rsp);
1704

1705
	raw_spin_lock_irq(&rnp->lock);
1706
	smp_mb__after_unlock_lock();
1707 1708 1709
	gp_duration = jiffies - rsp->gp_start;
	if (gp_duration > rsp->gp_max)
		rsp->gp_max = gp_duration;
1710

1711 1712 1713 1714 1715 1716 1717 1718
	/*
	 * We know the grace period is complete, but to everyone else
	 * it appears to still be ongoing.  But it is also the case
	 * that to everyone else it looks like there is nothing that
	 * they can do to advance the grace period.  It is therefore
	 * safe for us to drop the lock in order to mark the grace
	 * period as completed in all of the rcu_node structures.
	 */
1719
	raw_spin_unlock_irq(&rnp->lock);
1720

1721 1722 1723 1724 1725 1726 1727 1728 1729 1730
	/*
	 * Propagate new ->completed value to rcu_node structures so
	 * that other CPUs don't have to wait until the start of the next
	 * grace period to process their callbacks.  This also avoids
	 * some nasty RCU grace-period initialization races by forcing
	 * the end of the current grace period to be completely recorded in
	 * all of the rcu_node structures before the beginning of the next
	 * grace period is recorded in any of the rcu_node structures.
	 */
	rcu_for_each_node_breadth_first(rsp, rnp) {
1731
		raw_spin_lock_irq(&rnp->lock);
1732
		smp_mb__after_unlock_lock();
1733
		ACCESS_ONCE(rnp->completed) = rsp->gpnum;
1734 1735
		rdp = this_cpu_ptr(rsp->rda);
		if (rnp == rdp->mynode)
1736
			needgp = __note_gp_changes(rsp, rnp, rdp) || needgp;
1737
		/* smp_mb() provided by prior unlock-lock pair. */
1738
		nocb += rcu_future_gp_cleanup(rsp, rnp);
1739
		raw_spin_unlock_irq(&rnp->lock);
1740
		cond_resched_rcu_qs();
1741
	}
1742 1743
	rnp = rcu_get_root(rsp);
	raw_spin_lock_irq(&rnp->lock);
1744
	smp_mb__after_unlock_lock(); /* Order GP before ->completed update. */
1745
	rcu_nocb_gp_set(rnp, nocb);
1746

1747 1748
	/* Declare grace period done. */
	ACCESS_ONCE(rsp->completed) = rsp->gpnum;
1749
	trace_rcu_grace_period(rsp->name, rsp->completed, TPS("end"));
1750
	rsp->fqs_state = RCU_GP_IDLE;
1751
	rdp = this_cpu_ptr(rsp->rda);
1752 1753 1754
	/* Advance CBs to reduce false positives below. */
	needgp = rcu_advance_cbs(rsp, rnp, rdp) || needgp;
	if (needgp || cpu_needs_another_gp(rsp, rdp)) {
1755
		ACCESS_ONCE(rsp->gp_flags) = RCU_GP_FLAG_INIT;
1756 1757 1758 1759
		trace_rcu_grace_period(rsp->name,
				       ACCESS_ONCE(rsp->gpnum),
				       TPS("newreq"));
	}
1760 1761 1762 1763 1764 1765 1766 1767
	raw_spin_unlock_irq(&rnp->lock);
}

/*
 * Body of kthread that handles grace periods.
 */
static int __noreturn rcu_gp_kthread(void *arg)
{
1768
	int fqs_state;
1769
	int gf;
1770
	unsigned long j;
1771
	int ret;
1772 1773 1774 1775 1776 1777 1778
	struct rcu_state *rsp = arg;
	struct rcu_node *rnp = rcu_get_root(rsp);

	for (;;) {

		/* Handle grace-period start. */
		for (;;) {
1779 1780 1781
			trace_rcu_grace_period(rsp->name,
					       ACCESS_ONCE(rsp->gpnum),
					       TPS("reqwait"));
1782
			rsp->gp_state = RCU_GP_WAIT_GPS;
1783
			wait_event_interruptible(rsp->gp_wq,
1784
						 ACCESS_ONCE(rsp->gp_flags) &
1785
						 RCU_GP_FLAG_INIT);
1786
			/* Locking provides needed memory barrier. */
1787
			if (rcu_gp_init(rsp))
1788
				break;
1789
			cond_resched_rcu_qs();
1790
			WARN_ON(signal_pending(current));
1791 1792 1793
			trace_rcu_grace_period(rsp->name,
					       ACCESS_ONCE(rsp->gpnum),
					       TPS("reqwaitsig"));
1794
		}
1795

1796 1797
		/* Handle quiescent-state forcing. */
		fqs_state = RCU_SAVE_DYNTICK;
1798 1799 1800 1801 1802
		j = jiffies_till_first_fqs;
		if (j > HZ) {
			j = HZ;
			jiffies_till_first_fqs = HZ;
		}
1803
		ret = 0;
1804
		for (;;) {
1805 1806
			if (!ret)
				rsp->jiffies_force_qs = jiffies + j;
1807 1808 1809
			trace_rcu_grace_period(rsp->name,
					       ACCESS_ONCE(rsp->gpnum),
					       TPS("fqswait"));
1810
			rsp->gp_state = RCU_GP_WAIT_FQS;
1811
			ret = wait_event_interruptible_timeout(rsp->gp_wq,
1812 1813
					((gf = ACCESS_ONCE(rsp->gp_flags)) &
					 RCU_GP_FLAG_FQS) ||
1814 1815
					(!ACCESS_ONCE(rnp->qsmask) &&
					 !rcu_preempt_blocked_readers_cgp(rnp)),
1816
					j);
1817
			/* Locking provides needed memory barriers. */
1818
			/* If grace period done, leave loop. */
1819
			if (!ACCESS_ONCE(rnp->qsmask) &&
1820
			    !rcu_preempt_blocked_readers_cgp(rnp))
1821
				break;
1822
			/* If time for quiescent-state forcing, do it. */
1823 1824
			if (ULONG_CMP_GE(jiffies, rsp->jiffies_force_qs) ||
			    (gf & RCU_GP_FLAG_FQS)) {
1825 1826 1827
				trace_rcu_grace_period(rsp->name,
						       ACCESS_ONCE(rsp->gpnum),
						       TPS("fqsstart"));
1828
				fqs_state = rcu_gp_fqs(rsp, fqs_state);
1829 1830 1831
				trace_rcu_grace_period(rsp->name,
						       ACCESS_ONCE(rsp->gpnum),
						       TPS("fqsend"));
1832
				cond_resched_rcu_qs();
1833 1834
			} else {
				/* Deal with stray signal. */
1835
				cond_resched_rcu_qs();
1836
				WARN_ON(signal_pending(current));
1837 1838 1839
				trace_rcu_grace_period(rsp->name,
						       ACCESS_ONCE(rsp->gpnum),
						       TPS("fqswaitsig"));
1840
			}
1841 1842 1843 1844 1845 1846 1847 1848
			j = jiffies_till_next_fqs;
			if (j > HZ) {
				j = HZ;
				jiffies_till_next_fqs = HZ;
			} else if (j < 1) {
				j = 1;
				jiffies_till_next_fqs = 1;
			}
1849
		}
1850 1851 1852

		/* Handle grace-period end. */
		rcu_gp_cleanup(rsp);
1853 1854 1855
	}
}

1856 1857 1858
/*
 * Start a new RCU grace period if warranted, re-initializing the hierarchy
 * in preparation for detecting the next grace period.  The caller must hold
1859
 * the root node's ->lock and hard irqs must be disabled.
1860 1861 1862 1863
 *
 * Note that it is legal for a dying CPU (which is marked as offline) to
 * invoke this function.  This can happen when the dying CPU reports its
 * quiescent state.
1864 1865
 *
 * Returns true if the grace-period kthread must be awakened.
1866
 */
1867
static bool
1868 1869
rcu_start_gp_advanced(struct rcu_state *rsp, struct rcu_node *rnp,
		      struct rcu_data *rdp)
1870
{
1871
	if (!rsp->gp_kthread || !cpu_needs_another_gp(rsp, rdp)) {
1872
		/*
1873
		 * Either we have not yet spawned the grace-period
1874 1875
		 * task, this CPU does not need another grace period,
		 * or a grace period is already in progress.
1876
		 * Either way, don't start a new grace period.
1877
		 */
1878
		return false;
1879
	}
1880
	ACCESS_ONCE(rsp->gp_flags) = RCU_GP_FLAG_INIT;
1881 1882
	trace_rcu_grace_period(rsp->name, ACCESS_ONCE(rsp->gpnum),
			       TPS("newreq"));
1883

1884 1885
	/*
	 * We can't do wakeups while holding the rnp->lock, as that
1886
	 * could cause possible deadlocks with the rq->lock. Defer
1887
	 * the wakeup to our caller.
1888
	 */
1889
	return true;
1890 1891
}

1892 1893 1894 1895 1896 1897
/*
 * Similar to rcu_start_gp_advanced(), but also advance the calling CPU's
 * callbacks.  Note that rcu_start_gp_advanced() cannot do this because it
 * is invoked indirectly from rcu_advance_cbs(), which would result in
 * endless recursion -- or would do so if it wasn't for the self-deadlock
 * that is encountered beforehand.
1898 1899
 *
 * Returns true if the grace-period kthread needs to be awakened.
1900
 */
1901
static bool rcu_start_gp(struct rcu_state *rsp)
1902 1903 1904
{
	struct rcu_data *rdp = this_cpu_ptr(rsp->rda);
	struct rcu_node *rnp = rcu_get_root(rsp);
1905
	bool ret = false;
1906 1907 1908 1909 1910 1911 1912 1913 1914

	/*
	 * If there is no grace period in progress right now, any
	 * callbacks we have up to this point will be satisfied by the
	 * next grace period.  Also, advancing the callbacks reduces the
	 * probability of false positives from cpu_needs_another_gp()
	 * resulting in pointless grace periods.  So, advance callbacks
	 * then start the grace period!
	 */
1915 1916 1917
	ret = rcu_advance_cbs(rsp, rnp, rdp) || ret;
	ret = rcu_start_gp_advanced(rsp, rnp, rdp) || ret;
	return ret;
1918 1919
}

1920
/*
P
Paul E. McKenney 已提交
1921 1922 1923
 * Report a full set of quiescent states to the specified rcu_state
 * data structure.  This involves cleaning up after the prior grace
 * period and letting rcu_start_gp() start up the next grace period
1924 1925
 * if one is needed.  Note that the caller must hold rnp->lock, which
 * is released before return.
1926
 */
P
Paul E. McKenney 已提交
1927
static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags)
1928
	__releases(rcu_get_root(rsp)->lock)
1929
{
1930
	WARN_ON_ONCE(!rcu_gp_in_progress(rsp));
1931
	raw_spin_unlock_irqrestore(&rcu_get_root(rsp)->lock, flags);
1932
	rcu_gp_kthread_wake(rsp);
1933 1934
}

1935
/*
P
Paul E. McKenney 已提交
1936 1937 1938 1939 1940 1941
 * Similar to rcu_report_qs_rdp(), for which it is a helper function.
 * Allows quiescent states for a group of CPUs to be reported at one go
 * to the specified rcu_node structure, though all the CPUs in the group
 * must be represented by the same rcu_node structure (which need not be
 * a leaf rcu_node structure, though it often will be).  That structure's
 * lock must be held upon entry, and it is released before return.
1942 1943
 */
static void
P
Paul E. McKenney 已提交
1944 1945
rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp,
		  struct rcu_node *rnp, unsigned long flags)
1946 1947
	__releases(rnp->lock)
{
1948 1949
	struct rcu_node *rnp_c;

1950 1951 1952 1953 1954
	/* Walk up the rcu_node hierarchy. */
	for (;;) {
		if (!(rnp->qsmask & mask)) {

			/* Our bit has already been cleared, so done. */
P
Paul E. McKenney 已提交
1955
			raw_spin_unlock_irqrestore(&rnp->lock, flags);
1956 1957 1958
			return;
		}
		rnp->qsmask &= ~mask;
1959 1960 1961 1962
		trace_rcu_quiescent_state_report(rsp->name, rnp->gpnum,
						 mask, rnp->qsmask, rnp->level,
						 rnp->grplo, rnp->grphi,
						 !!rnp->gp_tasks);
1963
		if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) {
1964 1965

			/* Other bits still set at this level, so done. */
P
Paul E. McKenney 已提交
1966
			raw_spin_unlock_irqrestore(&rnp->lock, flags);
1967 1968 1969 1970 1971 1972 1973 1974 1975
			return;
		}
		mask = rnp->grpmask;
		if (rnp->parent == NULL) {

			/* No more levels.  Exit loop holding root lock. */

			break;
		}
P
Paul E. McKenney 已提交
1976
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
1977
		rnp_c = rnp;
1978
		rnp = rnp->parent;
P
Paul E. McKenney 已提交
1979
		raw_spin_lock_irqsave(&rnp->lock, flags);
1980
		smp_mb__after_unlock_lock();
1981
		WARN_ON_ONCE(rnp_c->qsmask);
1982 1983 1984 1985
	}

	/*
	 * Get here if we are the last CPU to pass through a quiescent
P
Paul E. McKenney 已提交
1986
	 * state for this grace period.  Invoke rcu_report_qs_rsp()
1987
	 * to clean up and start the next grace period if one is needed.
1988
	 */
P
Paul E. McKenney 已提交
1989
	rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */
1990 1991 1992
}

/*
P
Paul E. McKenney 已提交
1993 1994 1995 1996 1997 1998 1999
 * Record a quiescent state for the specified CPU to that CPU's rcu_data
 * structure.  This must be either called from the specified CPU, or
 * called when the specified CPU is known to be offline (and when it is
 * also known that no other CPU is concurrently trying to help the offline
 * CPU).  The lastcomp argument is used to make sure we are still in the
 * grace period of interest.  We don't want to end the current grace period
 * based on quiescent states detected in an earlier grace period!
2000 2001
 */
static void
2002
rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp)
2003 2004 2005
{
	unsigned long flags;
	unsigned long mask;
2006
	bool needwake;
2007 2008 2009
	struct rcu_node *rnp;

	rnp = rdp->mynode;
P
Paul E. McKenney 已提交
2010
	raw_spin_lock_irqsave(&rnp->lock, flags);
2011
	smp_mb__after_unlock_lock();
2012 2013
	if (rdp->passed_quiesce == 0 || rdp->gpnum != rnp->gpnum ||
	    rnp->completed == rnp->gpnum) {
2014 2015

		/*
2016 2017 2018 2019
		 * The grace period in which this quiescent state was
		 * recorded has ended, so don't report it upwards.
		 * We will instead need a new quiescent state that lies
		 * within the current grace period.
2020
		 */
2021
		rdp->passed_quiesce = 0;	/* need qs for new gp. */
P
Paul E. McKenney 已提交
2022
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
2023 2024 2025 2026
		return;
	}
	mask = rdp->grpmask;
	if ((rnp->qsmask & mask) == 0) {
P
Paul E. McKenney 已提交
2027
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
2028 2029 2030 2031 2032 2033 2034
	} else {
		rdp->qs_pending = 0;

		/*
		 * This GP can't end until cpu checks in, so all of our
		 * callbacks can be processed during the next GP.
		 */
2035
		needwake = rcu_accelerate_cbs(rsp, rnp, rdp);
2036

P
Paul E. McKenney 已提交
2037
		rcu_report_qs_rnp(mask, rsp, rnp, flags); /* rlses rnp->lock */
2038 2039
		if (needwake)
			rcu_gp_kthread_wake(rsp);
2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051
	}
}

/*
 * Check to see if there is a new grace period of which this CPU
 * is not yet aware, and if so, set up local rcu_data state for it.
 * Otherwise, see if this CPU has just passed through its first
 * quiescent state for this grace period, and record that fact if so.
 */
static void
rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp)
{
2052 2053
	/* Check for grace-period ends and beginnings. */
	note_gp_changes(rsp, rdp);
2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065

	/*
	 * Does this CPU still need to do its part for current grace period?
	 * If no, return and let the other CPUs do their part as well.
	 */
	if (!rdp->qs_pending)
		return;

	/*
	 * Was there a quiescent state since the beginning of the grace
	 * period? If no, then exit and wait for the next call.
	 */
2066
	if (!rdp->passed_quiesce)
2067 2068
		return;

P
Paul E. McKenney 已提交
2069 2070 2071 2072
	/*
	 * Tell RCU we are done (but rcu_report_qs_rdp() will be the
	 * judge of that).
	 */
2073
	rcu_report_qs_rdp(rdp->cpu, rsp, rdp);
2074 2075 2076 2077
}

#ifdef CONFIG_HOTPLUG_CPU

2078
/*
2079 2080
 * Send the specified CPU's RCU callbacks to the orphanage.  The
 * specified CPU must be offline, and the caller must hold the
2081
 * ->orphan_lock.
2082
 */
2083 2084 2085
static void
rcu_send_cbs_to_orphanage(int cpu, struct rcu_state *rsp,
			  struct rcu_node *rnp, struct rcu_data *rdp)
2086
{
P
Paul E. McKenney 已提交
2087
	/* No-CBs CPUs do not have orphanable callbacks. */
2088
	if (rcu_is_nocb_cpu(rdp->cpu))
P
Paul E. McKenney 已提交
2089 2090
		return;

2091 2092
	/*
	 * Orphan the callbacks.  First adjust the counts.  This is safe
2093 2094
	 * because _rcu_barrier() excludes CPU-hotplug operations, so it
	 * cannot be running now.  Thus no memory barrier is required.
2095
	 */
2096
	if (rdp->nxtlist != NULL) {
2097 2098 2099
		rsp->qlen_lazy += rdp->qlen_lazy;
		rsp->qlen += rdp->qlen;
		rdp->n_cbs_orphaned += rdp->qlen;
2100
		rdp->qlen_lazy = 0;
2101
		ACCESS_ONCE(rdp->qlen) = 0;
2102 2103 2104
	}

	/*
2105 2106 2107 2108 2109 2110 2111
	 * Next, move those callbacks still needing a grace period to
	 * the orphanage, where some other CPU will pick them up.
	 * Some of the callbacks might have gone partway through a grace
	 * period, but that is too bad.  They get to start over because we
	 * cannot assume that grace periods are synchronized across CPUs.
	 * We don't bother updating the ->nxttail[] array yet, instead
	 * we just reset the whole thing later on.
2112
	 */
2113 2114 2115 2116
	if (*rdp->nxttail[RCU_DONE_TAIL] != NULL) {
		*rsp->orphan_nxttail = *rdp->nxttail[RCU_DONE_TAIL];
		rsp->orphan_nxttail = rdp->nxttail[RCU_NEXT_TAIL];
		*rdp->nxttail[RCU_DONE_TAIL] = NULL;
2117 2118 2119
	}

	/*
2120 2121 2122
	 * Then move the ready-to-invoke callbacks to the orphanage,
	 * where some other CPU will pick them up.  These will not be
	 * required to pass though another grace period: They are done.
2123
	 */
2124
	if (rdp->nxtlist != NULL) {
2125 2126
		*rsp->orphan_donetail = rdp->nxtlist;
		rsp->orphan_donetail = rdp->nxttail[RCU_DONE_TAIL];
2127
	}
2128

2129
	/* Finally, initialize the rcu_data structure's list to empty.  */
2130
	init_callback_list(rdp);
2131 2132 2133 2134
}

/*
 * Adopt the RCU callbacks from the specified rcu_state structure's
2135
 * orphanage.  The caller must hold the ->orphan_lock.
2136
 */
2137
static void rcu_adopt_orphan_cbs(struct rcu_state *rsp, unsigned long flags)
2138 2139
{
	int i;
2140
	struct rcu_data *rdp = raw_cpu_ptr(rsp->rda);
2141

P
Paul E. McKenney 已提交
2142
	/* No-CBs CPUs are handled specially. */
2143
	if (rcu_nocb_adopt_orphan_cbs(rsp, rdp, flags))
P
Paul E. McKenney 已提交
2144 2145
		return;

2146 2147 2148 2149
	/* Do the accounting first. */
	rdp->qlen_lazy += rsp->qlen_lazy;
	rdp->qlen += rsp->qlen;
	rdp->n_cbs_adopted += rsp->qlen;
2150 2151
	if (rsp->qlen_lazy != rsp->qlen)
		rcu_idle_count_callbacks_posted();
2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190
	rsp->qlen_lazy = 0;
	rsp->qlen = 0;

	/*
	 * We do not need a memory barrier here because the only way we
	 * can get here if there is an rcu_barrier() in flight is if
	 * we are the task doing the rcu_barrier().
	 */

	/* First adopt the ready-to-invoke callbacks. */
	if (rsp->orphan_donelist != NULL) {
		*rsp->orphan_donetail = *rdp->nxttail[RCU_DONE_TAIL];
		*rdp->nxttail[RCU_DONE_TAIL] = rsp->orphan_donelist;
		for (i = RCU_NEXT_SIZE - 1; i >= RCU_DONE_TAIL; i--)
			if (rdp->nxttail[i] == rdp->nxttail[RCU_DONE_TAIL])
				rdp->nxttail[i] = rsp->orphan_donetail;
		rsp->orphan_donelist = NULL;
		rsp->orphan_donetail = &rsp->orphan_donelist;
	}

	/* And then adopt the callbacks that still need a grace period. */
	if (rsp->orphan_nxtlist != NULL) {
		*rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_nxtlist;
		rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_nxttail;
		rsp->orphan_nxtlist = NULL;
		rsp->orphan_nxttail = &rsp->orphan_nxtlist;
	}
}

/*
 * Trace the fact that this CPU is going offline.
 */
static void rcu_cleanup_dying_cpu(struct rcu_state *rsp)
{
	RCU_TRACE(unsigned long mask);
	RCU_TRACE(struct rcu_data *rdp = this_cpu_ptr(rsp->rda));
	RCU_TRACE(struct rcu_node *rnp = rdp->mynode);

	RCU_TRACE(mask = rdp->grpmask);
2191 2192
	trace_rcu_grace_period(rsp->name,
			       rnp->gpnum + 1 - !!(rnp->qsmask & mask),
2193
			       TPS("cpuofl"));
2194 2195 2196
}

/*
2197
 * The CPU has been completely removed, and some other CPU is reporting
2198 2199
 * this fact from process context.  Do the remainder of the cleanup,
 * including orphaning the outgoing CPU's RCU callbacks, and also
2200 2201
 * adopting them.  There can only be one CPU hotplug operation at a time,
 * so no other CPU can be attempting to update rcu_cpu_kthread_task.
2202
 */
2203
static void rcu_cleanup_dead_cpu(int cpu, struct rcu_state *rsp)
2204
{
2205 2206 2207
	unsigned long flags;
	unsigned long mask;
	int need_report = 0;
2208
	struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
2209
	struct rcu_node *rnp = rdp->mynode;  /* Outgoing CPU's rdp & rnp. */
2210

2211
	/* Adjust any no-longer-needed kthreads. */
T
Thomas Gleixner 已提交
2212
	rcu_boost_kthread_setaffinity(rnp, -1);
2213 2214

	/* Exclude any attempts to start a new grace period. */
2215
	mutex_lock(&rsp->onoff_mutex);
2216
	raw_spin_lock_irqsave(&rsp->orphan_lock, flags);
2217

2218 2219
	/* Orphan the dead CPU's callbacks, and adopt them if appropriate. */
	rcu_send_cbs_to_orphanage(cpu, rsp, rnp, rdp);
2220
	rcu_adopt_orphan_cbs(rsp, flags);
2221

2222 2223 2224 2225
	/* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */
	mask = rdp->grpmask;	/* rnp->grplo is constant. */
	do {
		raw_spin_lock(&rnp->lock);	/* irqs already disabled. */
2226
		smp_mb__after_unlock_lock();
2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243
		rnp->qsmaskinit &= ~mask;
		if (rnp->qsmaskinit != 0) {
			if (rnp != rdp->mynode)
				raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
			break;
		}
		if (rnp == rdp->mynode)
			need_report = rcu_preempt_offline_tasks(rsp, rnp, rdp);
		else
			raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
		mask = rnp->grpmask;
		rnp = rnp->parent;
	} while (rnp != NULL);

	/*
	 * We still hold the leaf rcu_node structure lock here, and
	 * irqs are still disabled.  The reason for this subterfuge is
2244
	 * because invoking rcu_report_unblock_qs_rnp() with ->orphan_lock
2245 2246
	 * held leads to deadlock.
	 */
2247
	raw_spin_unlock(&rsp->orphan_lock); /* irqs remain disabled. */
2248 2249 2250 2251 2252 2253 2254
	rnp = rdp->mynode;
	if (need_report & RCU_OFL_TASKS_NORM_GP)
		rcu_report_unblock_qs_rnp(rnp, flags);
	else
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
	if (need_report & RCU_OFL_TASKS_EXP_GP)
		rcu_report_exp_rnp(rsp, rnp, true);
2255 2256 2257
	WARN_ONCE(rdp->qlen != 0 || rdp->nxtlist != NULL,
		  "rcu_cleanup_dead_cpu: Callbacks on offline CPU %d: qlen=%lu, nxtlist=%p\n",
		  cpu, rdp->qlen, rdp->nxtlist);
2258 2259 2260
	init_callback_list(rdp);
	/* Disallow further callbacks on this CPU. */
	rdp->nxttail[RCU_NEXT_TAIL] = NULL;
2261
	mutex_unlock(&rsp->onoff_mutex);
2262 2263 2264 2265
}

#else /* #ifdef CONFIG_HOTPLUG_CPU */

2266
static void rcu_cleanup_dying_cpu(struct rcu_state *rsp)
2267 2268 2269
{
}

2270
static void rcu_cleanup_dead_cpu(int cpu, struct rcu_state *rsp)
2271 2272 2273 2274 2275 2276 2277 2278 2279
{
}

#endif /* #else #ifdef CONFIG_HOTPLUG_CPU */

/*
 * Invoke any RCU callbacks that have made it to the end of their grace
 * period.  Thottle as specified by rdp->blimit.
 */
2280
static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp)
2281 2282 2283
{
	unsigned long flags;
	struct rcu_head *next, *list, **tail;
E
Eric Dumazet 已提交
2284 2285
	long bl, count, count_lazy;
	int i;
2286

2287
	/* If no callbacks are ready, just return. */
2288
	if (!cpu_has_callbacks_ready_to_invoke(rdp)) {
2289
		trace_rcu_batch_start(rsp->name, rdp->qlen_lazy, rdp->qlen, 0);
2290 2291 2292
		trace_rcu_batch_end(rsp->name, 0, !!ACCESS_ONCE(rdp->nxtlist),
				    need_resched(), is_idle_task(current),
				    rcu_is_callbacks_kthread());
2293
		return;
2294
	}
2295 2296 2297 2298 2299 2300

	/*
	 * Extract the list of ready callbacks, disabling to prevent
	 * races with call_rcu() from interrupt handlers.
	 */
	local_irq_save(flags);
2301
	WARN_ON_ONCE(cpu_is_offline(smp_processor_id()));
2302
	bl = rdp->blimit;
2303
	trace_rcu_batch_start(rsp->name, rdp->qlen_lazy, rdp->qlen, bl);
2304 2305 2306 2307
	list = rdp->nxtlist;
	rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL];
	*rdp->nxttail[RCU_DONE_TAIL] = NULL;
	tail = rdp->nxttail[RCU_DONE_TAIL];
2308 2309 2310
	for (i = RCU_NEXT_SIZE - 1; i >= 0; i--)
		if (rdp->nxttail[i] == rdp->nxttail[RCU_DONE_TAIL])
			rdp->nxttail[i] = &rdp->nxtlist;
2311 2312 2313
	local_irq_restore(flags);

	/* Invoke callbacks. */
2314
	count = count_lazy = 0;
2315 2316 2317
	while (list) {
		next = list->next;
		prefetch(next);
2318
		debug_rcu_head_unqueue(list);
2319 2320
		if (__rcu_reclaim(rsp->name, list))
			count_lazy++;
2321
		list = next;
2322 2323 2324 2325
		/* Stop only if limit reached and CPU has something to do. */
		if (++count >= bl &&
		    (need_resched() ||
		     (!is_idle_task(current) && !rcu_is_callbacks_kthread())))
2326 2327 2328 2329
			break;
	}

	local_irq_save(flags);
2330 2331 2332
	trace_rcu_batch_end(rsp->name, count, !!list, need_resched(),
			    is_idle_task(current),
			    rcu_is_callbacks_kthread());
2333 2334 2335 2336 2337

	/* Update count, and requeue any remaining callbacks. */
	if (list != NULL) {
		*tail = rdp->nxtlist;
		rdp->nxtlist = list;
2338 2339 2340
		for (i = 0; i < RCU_NEXT_SIZE; i++)
			if (&rdp->nxtlist == rdp->nxttail[i])
				rdp->nxttail[i] = tail;
2341 2342 2343
			else
				break;
	}
2344 2345
	smp_mb(); /* List handling before counting for rcu_barrier(). */
	rdp->qlen_lazy -= count_lazy;
2346
	ACCESS_ONCE(rdp->qlen) = rdp->qlen - count;
2347
	rdp->n_cbs_invoked += count;
2348 2349 2350 2351 2352

	/* Reinstate batch limit if we have worked down the excess. */
	if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark)
		rdp->blimit = blimit;

2353 2354 2355 2356 2357 2358
	/* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */
	if (rdp->qlen == 0 && rdp->qlen_last_fqs_check != 0) {
		rdp->qlen_last_fqs_check = 0;
		rdp->n_force_qs_snap = rsp->n_force_qs;
	} else if (rdp->qlen < rdp->qlen_last_fqs_check - qhimark)
		rdp->qlen_last_fqs_check = rdp->qlen;
2359
	WARN_ON_ONCE((rdp->nxtlist == NULL) != (rdp->qlen == 0));
2360

2361 2362
	local_irq_restore(flags);

2363
	/* Re-invoke RCU core processing if there are callbacks remaining. */
2364
	if (cpu_has_callbacks_ready_to_invoke(rdp))
2365
		invoke_rcu_core();
2366 2367 2368 2369 2370
}

/*
 * Check to see if this CPU is in a non-context-switch quiescent state
 * (user mode or idle loop for rcu, non-softirq execution for rcu_bh).
2371
 * Also schedule RCU core processing.
2372
 *
2373
 * This function must be called from hardirq context.  It is normally
2374 2375 2376 2377 2378
 * invoked from the scheduling-clock interrupt.  If rcu_pending returns
 * false, there is no point in invoking rcu_check_callbacks().
 */
void rcu_check_callbacks(int cpu, int user)
{
2379
	trace_rcu_utilization(TPS("Start scheduler-tick"));
2380
	increment_cpu_stall_ticks();
2381
	if (user || rcu_is_cpu_rrupt_from_idle()) {
2382 2383 2384 2385 2386

		/*
		 * Get here if this CPU took its interrupt from user
		 * mode or from the idle loop, and if this is not a
		 * nested interrupt.  In this case, the CPU is in
2387
		 * a quiescent state, so note it.
2388 2389
		 *
		 * No memory barrier is required here because both
2390 2391 2392
		 * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local
		 * variables that other CPUs neither access nor modify,
		 * at least not while the corresponding CPU is online.
2393 2394
		 */

2395 2396
		rcu_sched_qs();
		rcu_bh_qs();
2397 2398 2399 2400 2401 2402 2403

	} else if (!in_softirq()) {

		/*
		 * Get here if this CPU did not take its interrupt from
		 * softirq, in other words, if it is not interrupting
		 * a rcu_bh read-side critical section.  This is an _bh
2404
		 * critical section, so note it.
2405 2406
		 */

2407
		rcu_bh_qs();
2408
	}
2409
	rcu_preempt_check_callbacks(cpu);
2410
	if (rcu_pending(cpu))
2411
		invoke_rcu_core();
P
Paul E. McKenney 已提交
2412 2413
	if (user)
		rcu_note_voluntary_context_switch(current);
2414
	trace_rcu_utilization(TPS("End scheduler-tick"));
2415 2416 2417 2418 2419
}

/*
 * Scan the leaf rcu_node structures, processing dyntick state for any that
 * have not yet encountered a quiescent state, using the function specified.
2420 2421
 * Also initiate boosting for any threads blocked on the root rcu_node.
 *
2422
 * The caller must have suppressed start of new grace periods.
2423
 */
2424 2425 2426 2427
static void force_qs_rnp(struct rcu_state *rsp,
			 int (*f)(struct rcu_data *rsp, bool *isidle,
				  unsigned long *maxj),
			 bool *isidle, unsigned long *maxj)
2428 2429 2430 2431 2432
{
	unsigned long bit;
	int cpu;
	unsigned long flags;
	unsigned long mask;
2433
	struct rcu_node *rnp;
2434

2435
	rcu_for_each_leaf_node(rsp, rnp) {
2436
		cond_resched_rcu_qs();
2437
		mask = 0;
P
Paul E. McKenney 已提交
2438
		raw_spin_lock_irqsave(&rnp->lock, flags);
2439
		smp_mb__after_unlock_lock();
2440
		if (!rcu_gp_in_progress(rsp)) {
P
Paul E. McKenney 已提交
2441
			raw_spin_unlock_irqrestore(&rnp->lock, flags);
2442
			return;
2443
		}
2444
		if (rnp->qsmask == 0) {
2445
			rcu_initiate_boost(rnp, flags); /* releases rnp->lock */
2446 2447
			continue;
		}
2448
		cpu = rnp->grplo;
2449
		bit = 1;
2450
		for (; cpu <= rnp->grphi; cpu++, bit <<= 1) {
2451 2452
			if ((rnp->qsmask & bit) != 0) {
				if ((rnp->qsmaskinit & bit) != 0)
2453
					*isidle = false;
2454 2455 2456
				if (f(per_cpu_ptr(rsp->rda, cpu), isidle, maxj))
					mask |= bit;
			}
2457
		}
2458
		if (mask != 0) {
2459

P
Paul E. McKenney 已提交
2460 2461
			/* rcu_report_qs_rnp() releases rnp->lock. */
			rcu_report_qs_rnp(mask, rsp, rnp, flags);
2462 2463
			continue;
		}
P
Paul E. McKenney 已提交
2464
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
2465
	}
2466
	rnp = rcu_get_root(rsp);
2467 2468
	if (rnp->qsmask == 0) {
		raw_spin_lock_irqsave(&rnp->lock, flags);
2469
		smp_mb__after_unlock_lock();
2470 2471
		rcu_initiate_boost(rnp, flags); /* releases rnp->lock. */
	}
2472 2473 2474 2475 2476 2477
}

/*
 * Force quiescent states on reluctant CPUs, and also detect which
 * CPUs are in dyntick-idle mode.
 */
2478
static void force_quiescent_state(struct rcu_state *rsp)
2479 2480
{
	unsigned long flags;
2481 2482 2483 2484 2485
	bool ret;
	struct rcu_node *rnp;
	struct rcu_node *rnp_old = NULL;

	/* Funnel through hierarchy to reduce memory contention. */
2486
	rnp = __this_cpu_read(rsp->rda->mynode);
2487 2488 2489 2490 2491 2492
	for (; rnp != NULL; rnp = rnp->parent) {
		ret = (ACCESS_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) ||
		      !raw_spin_trylock(&rnp->fqslock);
		if (rnp_old != NULL)
			raw_spin_unlock(&rnp_old->fqslock);
		if (ret) {
2493
			rsp->n_force_qs_lh++;
2494 2495 2496 2497 2498
			return;
		}
		rnp_old = rnp;
	}
	/* rnp_old == rcu_get_root(rsp), rnp == NULL. */
2499

2500 2501
	/* Reached the root of the rcu_node tree, acquire lock. */
	raw_spin_lock_irqsave(&rnp_old->lock, flags);
2502
	smp_mb__after_unlock_lock();
2503 2504
	raw_spin_unlock(&rnp_old->fqslock);
	if (ACCESS_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) {
2505
		rsp->n_force_qs_lh++;
2506
		raw_spin_unlock_irqrestore(&rnp_old->lock, flags);
2507
		return;  /* Someone beat us to it. */
2508
	}
2509 2510
	ACCESS_ONCE(rsp->gp_flags) =
		ACCESS_ONCE(rsp->gp_flags) | RCU_GP_FLAG_FQS;
2511
	raw_spin_unlock_irqrestore(&rnp_old->lock, flags);
2512
	rcu_gp_kthread_wake(rsp);
2513 2514 2515
}

/*
2516 2517 2518
 * This does the RCU core processing work for the specified rcu_state
 * and rcu_data structures.  This may be called only from the CPU to
 * whom the rdp belongs.
2519 2520
 */
static void
2521
__rcu_process_callbacks(struct rcu_state *rsp)
2522 2523
{
	unsigned long flags;
2524
	bool needwake;
2525
	struct rcu_data *rdp = raw_cpu_ptr(rsp->rda);
2526

2527 2528
	WARN_ON_ONCE(rdp->beenonline == 0);

2529 2530 2531 2532
	/* Update RCU state based on any recent quiescent states. */
	rcu_check_quiescent_state(rsp, rdp);

	/* Does this CPU require a not-yet-started grace period? */
2533
	local_irq_save(flags);
2534
	if (cpu_needs_another_gp(rsp, rdp)) {
2535
		raw_spin_lock(&rcu_get_root(rsp)->lock); /* irqs disabled. */
2536
		needwake = rcu_start_gp(rsp);
2537
		raw_spin_unlock_irqrestore(&rcu_get_root(rsp)->lock, flags);
2538 2539
		if (needwake)
			rcu_gp_kthread_wake(rsp);
2540 2541
	} else {
		local_irq_restore(flags);
2542 2543 2544
	}

	/* If there are callbacks ready, invoke them. */
2545
	if (cpu_has_callbacks_ready_to_invoke(rdp))
2546
		invoke_rcu_callbacks(rsp, rdp);
2547 2548 2549

	/* Do any needed deferred wakeups of rcuo kthreads. */
	do_nocb_deferred_wakeup(rdp);
2550 2551
}

2552
/*
2553
 * Do RCU core processing for the current CPU.
2554
 */
2555
static void rcu_process_callbacks(struct softirq_action *unused)
2556
{
2557 2558
	struct rcu_state *rsp;

2559 2560
	if (cpu_is_offline(smp_processor_id()))
		return;
2561
	trace_rcu_utilization(TPS("Start RCU core"));
2562 2563
	for_each_rcu_flavor(rsp)
		__rcu_process_callbacks(rsp);
2564
	trace_rcu_utilization(TPS("End RCU core"));
2565 2566
}

2567
/*
2568 2569 2570 2571 2572
 * Schedule RCU callback invocation.  If the specified type of RCU
 * does not support RCU priority boosting, just do a direct call,
 * otherwise wake up the per-CPU kernel kthread.  Note that because we
 * are running on the current CPU with interrupts disabled, the
 * rcu_cpu_kthread_task cannot disappear out from under us.
2573
 */
2574
static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp)
2575
{
2576 2577
	if (unlikely(!ACCESS_ONCE(rcu_scheduler_fully_active)))
		return;
2578 2579
	if (likely(!rsp->boost)) {
		rcu_do_batch(rsp, rdp);
2580 2581
		return;
	}
2582
	invoke_rcu_callbacks_kthread();
2583 2584
}

2585
static void invoke_rcu_core(void)
2586
{
2587 2588
	if (cpu_online(smp_processor_id()))
		raise_softirq(RCU_SOFTIRQ);
2589 2590
}

2591 2592 2593 2594 2595
/*
 * Handle any core-RCU processing required by a call_rcu() invocation.
 */
static void __call_rcu_core(struct rcu_state *rsp, struct rcu_data *rdp,
			    struct rcu_head *head, unsigned long flags)
2596
{
2597 2598
	bool needwake;

2599 2600 2601 2602
	/*
	 * If called from an extended quiescent state, invoke the RCU
	 * core in order to force a re-evaluation of RCU's idleness.
	 */
2603
	if (!rcu_is_watching() && cpu_online(smp_processor_id()))
2604 2605
		invoke_rcu_core();

2606
	/* If interrupts were disabled or CPU offline, don't invoke RCU core. */
2607
	if (irqs_disabled_flags(flags) || cpu_is_offline(smp_processor_id()))
2608
		return;
2609

2610 2611 2612 2613 2614 2615 2616
	/*
	 * Force the grace period if too many callbacks or too long waiting.
	 * Enforce hysteresis, and don't invoke force_quiescent_state()
	 * if some other CPU has recently done so.  Also, don't bother
	 * invoking force_quiescent_state() if the newly enqueued callback
	 * is the only one waiting for a grace period to complete.
	 */
2617
	if (unlikely(rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) {
2618 2619

		/* Are we ignoring a completed grace period? */
2620
		note_gp_changes(rsp, rdp);
2621 2622 2623 2624 2625

		/* Start a new grace period if one not already started. */
		if (!rcu_gp_in_progress(rsp)) {
			struct rcu_node *rnp_root = rcu_get_root(rsp);

2626
			raw_spin_lock(&rnp_root->lock);
2627
			smp_mb__after_unlock_lock();
2628
			needwake = rcu_start_gp(rsp);
2629
			raw_spin_unlock(&rnp_root->lock);
2630 2631
			if (needwake)
				rcu_gp_kthread_wake(rsp);
2632 2633 2634 2635 2636
		} else {
			/* Give the grace period a kick. */
			rdp->blimit = LONG_MAX;
			if (rsp->n_force_qs == rdp->n_force_qs_snap &&
			    *rdp->nxttail[RCU_DONE_TAIL] != head)
2637
				force_quiescent_state(rsp);
2638 2639 2640
			rdp->n_force_qs_snap = rsp->n_force_qs;
			rdp->qlen_last_fqs_check = rdp->qlen;
		}
2641
	}
2642 2643
}

2644 2645 2646 2647 2648 2649 2650
/*
 * RCU callback function to leak a callback.
 */
static void rcu_leak_callback(struct rcu_head *rhp)
{
}

P
Paul E. McKenney 已提交
2651 2652 2653 2654 2655 2656
/*
 * Helper function for call_rcu() and friends.  The cpu argument will
 * normally be -1, indicating "currently running CPU".  It may specify
 * a CPU only if that CPU is a no-CBs CPU.  Currently, only _rcu_barrier()
 * is expected to specify a CPU.
 */
2657 2658
static void
__call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
P
Paul E. McKenney 已提交
2659
	   struct rcu_state *rsp, int cpu, bool lazy)
2660 2661 2662 2663
{
	unsigned long flags;
	struct rcu_data *rdp;

2664
	WARN_ON_ONCE((unsigned long)head & 0x1); /* Misaligned rcu_head! */
2665 2666 2667 2668 2669 2670
	if (debug_rcu_head_queue(head)) {
		/* Probable double call_rcu(), so leak the callback. */
		ACCESS_ONCE(head->func) = rcu_leak_callback;
		WARN_ONCE(1, "__call_rcu(): Leaked duplicate callback\n");
		return;
	}
2671 2672 2673 2674 2675 2676 2677 2678 2679 2680
	head->func = func;
	head->next = NULL;

	/*
	 * Opportunistically note grace-period endings and beginnings.
	 * Note that we might see a beginning right after we see an
	 * end, but never vice versa, since this CPU has to pass through
	 * a quiescent state betweentimes.
	 */
	local_irq_save(flags);
2681
	rdp = this_cpu_ptr(rsp->rda);
2682 2683

	/* Add the callback to our list. */
P
Paul E. McKenney 已提交
2684 2685 2686 2687 2688
	if (unlikely(rdp->nxttail[RCU_NEXT_TAIL] == NULL) || cpu != -1) {
		int offline;

		if (cpu != -1)
			rdp = per_cpu_ptr(rsp->rda, cpu);
2689
		offline = !__call_rcu_nocb(rdp, head, lazy, flags);
P
Paul E. McKenney 已提交
2690
		WARN_ON_ONCE(offline);
2691 2692 2693 2694
		/* _call_rcu() is illegal on offline CPU; leak the callback. */
		local_irq_restore(flags);
		return;
	}
2695
	ACCESS_ONCE(rdp->qlen) = rdp->qlen + 1;
2696 2697
	if (lazy)
		rdp->qlen_lazy++;
2698 2699
	else
		rcu_idle_count_callbacks_posted();
2700 2701 2702
	smp_mb();  /* Count before adding callback for rcu_barrier(). */
	*rdp->nxttail[RCU_NEXT_TAIL] = head;
	rdp->nxttail[RCU_NEXT_TAIL] = &head->next;
2703

2704 2705
	if (__is_kfree_rcu_offset((unsigned long)func))
		trace_rcu_kfree_callback(rsp->name, head, (unsigned long)func,
2706
					 rdp->qlen_lazy, rdp->qlen);
2707
	else
2708
		trace_rcu_callback(rsp->name, head, rdp->qlen_lazy, rdp->qlen);
2709

2710 2711
	/* Go handle any RCU core processing required. */
	__call_rcu_core(rsp, rdp, head, flags);
2712 2713 2714 2715
	local_irq_restore(flags);
}

/*
2716
 * Queue an RCU-sched callback for invocation after a grace period.
2717
 */
2718
void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
2719
{
P
Paul E. McKenney 已提交
2720
	__call_rcu(head, func, &rcu_sched_state, -1, 0);
2721
}
2722
EXPORT_SYMBOL_GPL(call_rcu_sched);
2723 2724

/*
2725
 * Queue an RCU callback for invocation after a quicker grace period.
2726 2727 2728
 */
void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
{
P
Paul E. McKenney 已提交
2729
	__call_rcu(head, func, &rcu_bh_state, -1, 0);
2730 2731 2732
}
EXPORT_SYMBOL_GPL(call_rcu_bh);

2733 2734 2735 2736 2737 2738 2739 2740 2741 2742
/*
 * Queue an RCU callback for lazy invocation after a grace period.
 * This will likely be later named something like "call_rcu_lazy()",
 * but this change will require some way of tagging the lazy RCU
 * callbacks in the list of pending callbacks. Until then, this
 * function may only be called from __kfree_rcu().
 */
void kfree_call_rcu(struct rcu_head *head,
		    void (*func)(struct rcu_head *rcu))
{
2743
	__call_rcu(head, func, rcu_state_p, -1, 1);
2744 2745 2746
}
EXPORT_SYMBOL_GPL(kfree_call_rcu);

2747 2748 2749 2750 2751 2752 2753 2754 2755 2756 2757
/*
 * Because a context switch is a grace period for RCU-sched and RCU-bh,
 * any blocking grace-period wait automatically implies a grace period
 * if there is only one CPU online at any point time during execution
 * of either synchronize_sched() or synchronize_rcu_bh().  It is OK to
 * occasionally incorrectly indicate that there are multiple CPUs online
 * when there was in fact only one the whole time, as this just adds
 * some overhead: RCU still operates correctly.
 */
static inline int rcu_blocking_is_gp(void)
{
2758 2759
	int ret;

2760
	might_sleep();  /* Check for RCU read-side critical section. */
2761 2762 2763 2764
	preempt_disable();
	ret = num_online_cpus() <= 1;
	preempt_enable();
	return ret;
2765 2766
}

2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778
/**
 * synchronize_sched - wait until an rcu-sched grace period has elapsed.
 *
 * Control will return to the caller some time after a full rcu-sched
 * grace period has elapsed, in other words after all currently executing
 * rcu-sched read-side critical sections have completed.   These read-side
 * critical sections are delimited by rcu_read_lock_sched() and
 * rcu_read_unlock_sched(), and may be nested.  Note that preempt_disable(),
 * local_irq_disable(), and so on may be used in place of
 * rcu_read_lock_sched().
 *
 * This means that all preempt_disable code sequences, including NMI and
2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800
 * non-threaded hardware-interrupt handlers, in progress on entry will
 * have completed before this primitive returns.  However, this does not
 * guarantee that softirq handlers will have completed, since in some
 * kernels, these handlers can run in process context, and can block.
 *
 * Note that this guarantee implies further memory-ordering guarantees.
 * On systems with more than one CPU, when synchronize_sched() returns,
 * each CPU is guaranteed to have executed a full memory barrier since the
 * end of its last RCU-sched read-side critical section whose beginning
 * preceded the call to synchronize_sched().  In addition, each CPU having
 * an RCU read-side critical section that extends beyond the return from
 * synchronize_sched() is guaranteed to have executed a full memory barrier
 * after the beginning of synchronize_sched() and before the beginning of
 * that RCU read-side critical section.  Note that these guarantees include
 * CPUs that are offline, idle, or executing in user mode, as well as CPUs
 * that are executing in the kernel.
 *
 * Furthermore, if CPU A invoked synchronize_sched(), which returned
 * to its caller on CPU B, then both CPU A and CPU B are guaranteed
 * to have executed a full memory barrier during the execution of
 * synchronize_sched() -- even if CPU A and CPU B are the same CPU (but
 * again only if the system has more than one CPU).
2801 2802 2803 2804 2805 2806 2807 2808 2809
 *
 * This primitive provides the guarantees made by the (now removed)
 * synchronize_kernel() API.  In contrast, synchronize_rcu() only
 * guarantees that rcu_read_lock() sections will have completed.
 * In "classic RCU", these two guarantees happen to be one and
 * the same, but can differ in realtime RCU implementations.
 */
void synchronize_sched(void)
{
2810 2811 2812 2813
	rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map) &&
			   !lock_is_held(&rcu_lock_map) &&
			   !lock_is_held(&rcu_sched_lock_map),
			   "Illegal synchronize_sched() in RCU-sched read-side critical section");
2814 2815
	if (rcu_blocking_is_gp())
		return;
2816 2817 2818 2819
	if (rcu_expedited)
		synchronize_sched_expedited();
	else
		wait_rcu_gp(call_rcu_sched);
2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830
}
EXPORT_SYMBOL_GPL(synchronize_sched);

/**
 * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed.
 *
 * Control will return to the caller some time after a full rcu_bh grace
 * period has elapsed, in other words after all currently executing rcu_bh
 * read-side critical sections have completed.  RCU read-side critical
 * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(),
 * and may be nested.
2831 2832 2833
 *
 * See the description of synchronize_sched() for more detailed information
 * on memory ordering guarantees.
2834 2835 2836
 */
void synchronize_rcu_bh(void)
{
2837 2838 2839 2840
	rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map) &&
			   !lock_is_held(&rcu_lock_map) &&
			   !lock_is_held(&rcu_sched_lock_map),
			   "Illegal synchronize_rcu_bh() in RCU-bh read-side critical section");
2841 2842
	if (rcu_blocking_is_gp())
		return;
2843 2844 2845 2846
	if (rcu_expedited)
		synchronize_rcu_bh_expedited();
	else
		wait_rcu_gp(call_rcu_bh);
2847 2848 2849
}
EXPORT_SYMBOL_GPL(synchronize_rcu_bh);

2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869
/**
 * get_state_synchronize_rcu - Snapshot current RCU state
 *
 * Returns a cookie that is used by a later call to cond_synchronize_rcu()
 * to determine whether or not a full grace period has elapsed in the
 * meantime.
 */
unsigned long get_state_synchronize_rcu(void)
{
	/*
	 * Any prior manipulation of RCU-protected data must happen
	 * before the load from ->gpnum.
	 */
	smp_mb();  /* ^^^ */

	/*
	 * Make sure this load happens before the purportedly
	 * time-consuming work between get_state_synchronize_rcu()
	 * and cond_synchronize_rcu().
	 */
2870
	return smp_load_acquire(&rcu_state_p->gpnum);
2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895
}
EXPORT_SYMBOL_GPL(get_state_synchronize_rcu);

/**
 * cond_synchronize_rcu - Conditionally wait for an RCU grace period
 *
 * @oldstate: return value from earlier call to get_state_synchronize_rcu()
 *
 * If a full RCU grace period has elapsed since the earlier call to
 * get_state_synchronize_rcu(), just return.  Otherwise, invoke
 * synchronize_rcu() to wait for a full grace period.
 *
 * Yes, this function does not take counter wrap into account.  But
 * counter wrap is harmless.  If the counter wraps, we have waited for
 * more than 2 billion grace periods (and way more on a 64-bit system!),
 * so waiting for one additional grace period should be just fine.
 */
void cond_synchronize_rcu(unsigned long oldstate)
{
	unsigned long newstate;

	/*
	 * Ensure that this load happens before any RCU-destructive
	 * actions the caller might carry out after we return.
	 */
2896
	newstate = smp_load_acquire(&rcu_state_p->completed);
2897 2898 2899 2900 2901
	if (ULONG_CMP_GE(oldstate, newstate))
		synchronize_rcu();
}
EXPORT_SYMBOL_GPL(cond_synchronize_rcu);

2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918
static int synchronize_sched_expedited_cpu_stop(void *data)
{
	/*
	 * There must be a full memory barrier on each affected CPU
	 * between the time that try_stop_cpus() is called and the
	 * time that it returns.
	 *
	 * In the current initial implementation of cpu_stop, the
	 * above condition is already met when the control reaches
	 * this point and the following smp_mb() is not strictly
	 * necessary.  Do smp_mb() anyway for documentation and
	 * robustness against future implementation changes.
	 */
	smp_mb(); /* See above comment block. */
	return 0;
}

2919 2920 2921 2922 2923 2924 2925 2926 2927 2928
/**
 * synchronize_sched_expedited - Brute-force RCU-sched grace period
 *
 * Wait for an RCU-sched grace period to elapse, but use a "big hammer"
 * approach to force the grace period to end quickly.  This consumes
 * significant time on all CPUs and is unfriendly to real-time workloads,
 * so is thus not recommended for any sort of common-case code.  In fact,
 * if you are using synchronize_sched_expedited() in a loop, please
 * restructure your code to batch your updates, and then use a single
 * synchronize_sched() instead.
2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952
 *
 * This implementation can be thought of as an application of ticket
 * locking to RCU, with sync_sched_expedited_started and
 * sync_sched_expedited_done taking on the roles of the halves
 * of the ticket-lock word.  Each task atomically increments
 * sync_sched_expedited_started upon entry, snapshotting the old value,
 * then attempts to stop all the CPUs.  If this succeeds, then each
 * CPU will have executed a context switch, resulting in an RCU-sched
 * grace period.  We are then done, so we use atomic_cmpxchg() to
 * update sync_sched_expedited_done to match our snapshot -- but
 * only if someone else has not already advanced past our snapshot.
 *
 * On the other hand, if try_stop_cpus() fails, we check the value
 * of sync_sched_expedited_done.  If it has advanced past our
 * initial snapshot, then someone else must have forced a grace period
 * some time after we took our snapshot.  In this case, our work is
 * done for us, and we can simply return.  Otherwise, we try again,
 * but keep our initial snapshot for purposes of checking for someone
 * doing our work for us.
 *
 * If we fail too many times in a row, we fall back to synchronize_sched().
 */
void synchronize_sched_expedited(void)
{
2953 2954
	long firstsnap, s, snap;
	int trycount = 0;
2955
	struct rcu_state *rsp = &rcu_sched_state;
2956

2957 2958 2959 2960 2961 2962 2963 2964
	/*
	 * If we are in danger of counter wrap, just do synchronize_sched().
	 * By allowing sync_sched_expedited_started to advance no more than
	 * ULONG_MAX/8 ahead of sync_sched_expedited_done, we are ensuring
	 * that more than 3.5 billion CPUs would be required to force a
	 * counter wrap on a 32-bit system.  Quite a few more CPUs would of
	 * course be required on a 64-bit system.
	 */
2965 2966
	if (ULONG_CMP_GE((ulong)atomic_long_read(&rsp->expedited_start),
			 (ulong)atomic_long_read(&rsp->expedited_done) +
2967 2968
			 ULONG_MAX / 8)) {
		synchronize_sched();
2969
		atomic_long_inc(&rsp->expedited_wrap);
2970 2971
		return;
	}
2972

2973 2974 2975 2976
	/*
	 * Take a ticket.  Note that atomic_inc_return() implies a
	 * full memory barrier.
	 */
2977
	snap = atomic_long_inc_return(&rsp->expedited_start);
2978
	firstsnap = snap;
2979 2980 2981 2982 2983 2984
	if (!try_get_online_cpus()) {
		/* CPU hotplug operation in flight, fall back to normal GP. */
		wait_rcu_gp(call_rcu_sched);
		atomic_long_inc(&rsp->expedited_normal);
		return;
	}
2985
	WARN_ON_ONCE(cpu_is_offline(raw_smp_processor_id()));
2986 2987 2988 2989 2990 2991 2992 2993 2994

	/*
	 * Each pass through the following loop attempts to force a
	 * context switch on each CPU.
	 */
	while (try_stop_cpus(cpu_online_mask,
			     synchronize_sched_expedited_cpu_stop,
			     NULL) == -EAGAIN) {
		put_online_cpus();
2995
		atomic_long_inc(&rsp->expedited_tryfail);
2996

2997
		/* Check to see if someone else did our work for us. */
2998
		s = atomic_long_read(&rsp->expedited_done);
2999
		if (ULONG_CMP_GE((ulong)s, (ulong)firstsnap)) {
3000
			/* ensure test happens before caller kfree */
3001
			smp_mb__before_atomic(); /* ^^^ */
3002
			atomic_long_inc(&rsp->expedited_workdone1);
3003 3004
			return;
		}
3005 3006

		/* No joy, try again later.  Or just synchronize_sched(). */
3007
		if (trycount++ < 10) {
3008
			udelay(trycount * num_online_cpus());
3009
		} else {
3010
			wait_rcu_gp(call_rcu_sched);
3011
			atomic_long_inc(&rsp->expedited_normal);
3012 3013 3014
			return;
		}

3015
		/* Recheck to see if someone else did our work for us. */
3016
		s = atomic_long_read(&rsp->expedited_done);
3017
		if (ULONG_CMP_GE((ulong)s, (ulong)firstsnap)) {
3018
			/* ensure test happens before caller kfree */
3019
			smp_mb__before_atomic(); /* ^^^ */
3020
			atomic_long_inc(&rsp->expedited_workdone2);
3021 3022 3023 3024 3025
			return;
		}

		/*
		 * Refetching sync_sched_expedited_started allows later
3026 3027 3028 3029
		 * callers to piggyback on our grace period.  We retry
		 * after they started, so our grace period works for them,
		 * and they started after our first try, so their grace
		 * period works for us.
3030
		 */
3031 3032 3033 3034 3035 3036
		if (!try_get_online_cpus()) {
			/* CPU hotplug operation in flight, use normal GP. */
			wait_rcu_gp(call_rcu_sched);
			atomic_long_inc(&rsp->expedited_normal);
			return;
		}
3037
		snap = atomic_long_read(&rsp->expedited_start);
3038 3039
		smp_mb(); /* ensure read is before try_stop_cpus(). */
	}
3040
	atomic_long_inc(&rsp->expedited_stoppedcpus);
3041 3042 3043 3044 3045

	/*
	 * Everyone up to our most recent fetch is covered by our grace
	 * period.  Update the counter, but only if our work is still
	 * relevant -- which it won't be if someone who started later
3046
	 * than we did already did their update.
3047 3048
	 */
	do {
3049
		atomic_long_inc(&rsp->expedited_done_tries);
3050
		s = atomic_long_read(&rsp->expedited_done);
3051
		if (ULONG_CMP_GE((ulong)s, (ulong)snap)) {
3052
			/* ensure test happens before caller kfree */
3053
			smp_mb__before_atomic(); /* ^^^ */
3054
			atomic_long_inc(&rsp->expedited_done_lost);
3055 3056
			break;
		}
3057
	} while (atomic_long_cmpxchg(&rsp->expedited_done, s, snap) != s);
3058
	atomic_long_inc(&rsp->expedited_done_exit);
3059 3060 3061 3062 3063

	put_online_cpus();
}
EXPORT_SYMBOL_GPL(synchronize_sched_expedited);

3064 3065 3066 3067 3068 3069 3070 3071 3072
/*
 * Check to see if there is any immediate RCU-related work to be done
 * by the current CPU, for the specified type of RCU, returning 1 if so.
 * The checks are in order of increasing expense: checks that can be
 * carried out against CPU-local state are performed first.  However,
 * we must check for CPU stalls first, else we might not get a chance.
 */
static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp)
{
3073 3074
	struct rcu_node *rnp = rdp->mynode;

3075 3076 3077 3078 3079
	rdp->n_rcu_pending++;

	/* Check for CPU stalls, if enabled. */
	check_cpu_stall(rsp, rdp);

3080 3081 3082 3083
	/* Is this CPU a NO_HZ_FULL CPU that should ignore RCU? */
	if (rcu_nohz_full_cpu(rsp))
		return 0;

3084
	/* Is the RCU core waiting for a quiescent state from this CPU? */
3085 3086
	if (rcu_scheduler_fully_active &&
	    rdp->qs_pending && !rdp->passed_quiesce) {
3087
		rdp->n_rp_qs_pending++;
3088
	} else if (rdp->qs_pending && rdp->passed_quiesce) {
3089
		rdp->n_rp_report_qs++;
3090
		return 1;
3091
	}
3092 3093

	/* Does this CPU have callbacks ready to invoke? */
3094 3095
	if (cpu_has_callbacks_ready_to_invoke(rdp)) {
		rdp->n_rp_cb_ready++;
3096
		return 1;
3097
	}
3098 3099

	/* Has RCU gone idle with this CPU needing another grace period? */
3100 3101
	if (cpu_needs_another_gp(rsp, rdp)) {
		rdp->n_rp_cpu_needs_gp++;
3102
		return 1;
3103
	}
3104 3105

	/* Has another RCU grace period completed?  */
3106
	if (ACCESS_ONCE(rnp->completed) != rdp->completed) { /* outside lock */
3107
		rdp->n_rp_gp_completed++;
3108
		return 1;
3109
	}
3110 3111

	/* Has a new RCU grace period started? */
3112
	if (ACCESS_ONCE(rnp->gpnum) != rdp->gpnum) { /* outside lock */
3113
		rdp->n_rp_gp_started++;
3114
		return 1;
3115
	}
3116

3117 3118 3119 3120 3121 3122
	/* Does this CPU need a deferred NOCB wakeup? */
	if (rcu_nocb_need_deferred_wakeup(rdp)) {
		rdp->n_rp_nocb_defer_wakeup++;
		return 1;
	}

3123
	/* nothing to do */
3124
	rdp->n_rp_need_nothing++;
3125 3126 3127 3128 3129 3130 3131 3132
	return 0;
}

/*
 * Check to see if there is any immediate RCU-related work to be done
 * by the current CPU, returning 1 if so.  This function is part of the
 * RCU implementation; it is -not- an exported member of the RCU API.
 */
3133
static int rcu_pending(int cpu)
3134
{
3135 3136 3137 3138 3139 3140
	struct rcu_state *rsp;

	for_each_rcu_flavor(rsp)
		if (__rcu_pending(rsp, per_cpu_ptr(rsp->rda, cpu)))
			return 1;
	return 0;
3141 3142 3143
}

/*
3144 3145 3146
 * Return true if the specified CPU has any callback.  If all_lazy is
 * non-NULL, store an indication of whether all callbacks are lazy.
 * (If there are no callbacks, all of them are deemed to be lazy.)
3147
 */
3148
static int __maybe_unused rcu_cpu_has_callbacks(int cpu, bool *all_lazy)
3149
{
3150 3151 3152
	bool al = true;
	bool hc = false;
	struct rcu_data *rdp;
3153 3154
	struct rcu_state *rsp;

3155 3156
	for_each_rcu_flavor(rsp) {
		rdp = per_cpu_ptr(rsp->rda, cpu);
3157 3158 3159 3160
		if (!rdp->nxtlist)
			continue;
		hc = true;
		if (rdp->qlen != rdp->qlen_lazy || !all_lazy) {
3161
			al = false;
3162 3163
			break;
		}
3164 3165 3166 3167
	}
	if (all_lazy)
		*all_lazy = al;
	return hc;
3168 3169
}

3170 3171 3172 3173
/*
 * Helper function for _rcu_barrier() tracing.  If tracing is disabled,
 * the compiler is expected to optimize this away.
 */
3174
static void _rcu_barrier_trace(struct rcu_state *rsp, const char *s,
3175 3176 3177 3178 3179 3180
			       int cpu, unsigned long done)
{
	trace_rcu_barrier(rsp->name, s, cpu,
			  atomic_read(&rsp->barrier_cpu_count), done);
}

3181 3182 3183 3184
/*
 * RCU callback function for _rcu_barrier().  If we are last, wake
 * up the task executing _rcu_barrier().
 */
3185
static void rcu_barrier_callback(struct rcu_head *rhp)
3186
{
3187 3188 3189
	struct rcu_data *rdp = container_of(rhp, struct rcu_data, barrier_head);
	struct rcu_state *rsp = rdp->rsp;

3190 3191
	if (atomic_dec_and_test(&rsp->barrier_cpu_count)) {
		_rcu_barrier_trace(rsp, "LastCB", -1, rsp->n_barrier_done);
3192
		complete(&rsp->barrier_completion);
3193 3194 3195
	} else {
		_rcu_barrier_trace(rsp, "CB", -1, rsp->n_barrier_done);
	}
3196 3197 3198 3199 3200 3201 3202
}

/*
 * Called with preemption disabled, and from cross-cpu IRQ context.
 */
static void rcu_barrier_func(void *type)
{
3203
	struct rcu_state *rsp = type;
3204
	struct rcu_data *rdp = raw_cpu_ptr(rsp->rda);
3205

3206
	_rcu_barrier_trace(rsp, "IRQ", -1, rsp->n_barrier_done);
3207
	atomic_inc(&rsp->barrier_cpu_count);
3208
	rsp->call(&rdp->barrier_head, rcu_barrier_callback);
3209 3210 3211 3212 3213 3214
}

/*
 * Orchestrate the specified type of RCU barrier, waiting for all
 * RCU callbacks of the specified type to complete.
 */
3215
static void _rcu_barrier(struct rcu_state *rsp)
3216
{
3217 3218
	int cpu;
	struct rcu_data *rdp;
3219 3220
	unsigned long snap = ACCESS_ONCE(rsp->n_barrier_done);
	unsigned long snap_done;
3221

3222
	_rcu_barrier_trace(rsp, "Begin", -1, snap);
3223

3224
	/* Take mutex to serialize concurrent rcu_barrier() requests. */
3225
	mutex_lock(&rsp->barrier_mutex);
3226

3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238
	/*
	 * Ensure that all prior references, including to ->n_barrier_done,
	 * are ordered before the _rcu_barrier() machinery.
	 */
	smp_mb();  /* See above block comment. */

	/*
	 * Recheck ->n_barrier_done to see if others did our work for us.
	 * This means checking ->n_barrier_done for an even-to-odd-to-even
	 * transition.  The "if" expression below therefore rounds the old
	 * value up to the next even number and adds two before comparing.
	 */
3239
	snap_done = rsp->n_barrier_done;
3240
	_rcu_barrier_trace(rsp, "Check", -1, snap_done);
3241 3242 3243 3244 3245 3246 3247 3248 3249 3250 3251 3252

	/*
	 * If the value in snap is odd, we needed to wait for the current
	 * rcu_barrier() to complete, then wait for the next one, in other
	 * words, we need the value of snap_done to be three larger than
	 * the value of snap.  On the other hand, if the value in snap is
	 * even, we only had to wait for the next rcu_barrier() to complete,
	 * in other words, we need the value of snap_done to be only two
	 * greater than the value of snap.  The "(snap + 3) & ~0x1" computes
	 * this for us (thank you, Linus!).
	 */
	if (ULONG_CMP_GE(snap_done, (snap + 3) & ~0x1)) {
3253
		_rcu_barrier_trace(rsp, "EarlyExit", -1, snap_done);
3254 3255 3256 3257 3258 3259 3260 3261 3262 3263
		smp_mb(); /* caller's subsequent code after above check. */
		mutex_unlock(&rsp->barrier_mutex);
		return;
	}

	/*
	 * Increment ->n_barrier_done to avoid duplicate work.  Use
	 * ACCESS_ONCE() to prevent the compiler from speculating
	 * the increment to precede the early-exit check.
	 */
3264
	ACCESS_ONCE(rsp->n_barrier_done) = rsp->n_barrier_done + 1;
3265
	WARN_ON_ONCE((rsp->n_barrier_done & 0x1) != 1);
3266
	_rcu_barrier_trace(rsp, "Inc1", -1, rsp->n_barrier_done);
3267
	smp_mb(); /* Order ->n_barrier_done increment with below mechanism. */
3268

3269
	/*
3270 3271
	 * Initialize the count to one rather than to zero in order to
	 * avoid a too-soon return to zero in case of a short grace period
3272 3273
	 * (or preemption of this task).  Exclude CPU-hotplug operations
	 * to ensure that no offline CPU has callbacks queued.
3274
	 */
3275
	init_completion(&rsp->barrier_completion);
3276
	atomic_set(&rsp->barrier_cpu_count, 1);
3277
	get_online_cpus();
3278 3279

	/*
3280 3281 3282
	 * Force each CPU with callbacks to register a new callback.
	 * When that callback is invoked, we will know that all of the
	 * corresponding CPU's preceding callbacks have been invoked.
3283
	 */
P
Paul E. McKenney 已提交
3284
	for_each_possible_cpu(cpu) {
3285
		if (!cpu_online(cpu) && !rcu_is_nocb_cpu(cpu))
P
Paul E. McKenney 已提交
3286
			continue;
3287
		rdp = per_cpu_ptr(rsp->rda, cpu);
3288
		if (rcu_is_nocb_cpu(cpu)) {
3289 3290 3291 3292 3293 3294 3295 3296 3297 3298
			if (!rcu_nocb_cpu_needs_barrier(rsp, cpu)) {
				_rcu_barrier_trace(rsp, "OfflineNoCB", cpu,
						   rsp->n_barrier_done);
			} else {
				_rcu_barrier_trace(rsp, "OnlineNoCB", cpu,
						   rsp->n_barrier_done);
				atomic_inc(&rsp->barrier_cpu_count);
				__call_rcu(&rdp->barrier_head,
					   rcu_barrier_callback, rsp, cpu, 0);
			}
P
Paul E. McKenney 已提交
3299
		} else if (ACCESS_ONCE(rdp->qlen)) {
3300 3301
			_rcu_barrier_trace(rsp, "OnlineQ", cpu,
					   rsp->n_barrier_done);
3302
			smp_call_function_single(cpu, rcu_barrier_func, rsp, 1);
3303
		} else {
3304 3305
			_rcu_barrier_trace(rsp, "OnlineNQ", cpu,
					   rsp->n_barrier_done);
3306 3307
		}
	}
3308
	put_online_cpus();
3309 3310 3311 3312 3313

	/*
	 * Now that we have an rcu_barrier_callback() callback on each
	 * CPU, and thus each counted, remove the initial count.
	 */
3314
	if (atomic_dec_and_test(&rsp->barrier_cpu_count))
3315
		complete(&rsp->barrier_completion);
3316

3317 3318
	/* Increment ->n_barrier_done to prevent duplicate work. */
	smp_mb(); /* Keep increment after above mechanism. */
3319
	ACCESS_ONCE(rsp->n_barrier_done) = rsp->n_barrier_done + 1;
3320
	WARN_ON_ONCE((rsp->n_barrier_done & 0x1) != 0);
3321
	_rcu_barrier_trace(rsp, "Inc2", -1, rsp->n_barrier_done);
3322 3323
	smp_mb(); /* Keep increment before caller's subsequent code. */

3324
	/* Wait for all rcu_barrier_callback() callbacks to be invoked. */
3325
	wait_for_completion(&rsp->barrier_completion);
3326 3327

	/* Other rcu_barrier() invocations can now safely proceed. */
3328
	mutex_unlock(&rsp->barrier_mutex);
3329 3330 3331 3332 3333 3334 3335
}

/**
 * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete.
 */
void rcu_barrier_bh(void)
{
3336
	_rcu_barrier(&rcu_bh_state);
3337 3338 3339 3340 3341 3342 3343 3344
}
EXPORT_SYMBOL_GPL(rcu_barrier_bh);

/**
 * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks.
 */
void rcu_barrier_sched(void)
{
3345
	_rcu_barrier(&rcu_sched_state);
3346 3347 3348
}
EXPORT_SYMBOL_GPL(rcu_barrier_sched);

3349
/*
3350
 * Do boot-time initialization of a CPU's per-CPU RCU data.
3351
 */
3352 3353
static void __init
rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp)
3354 3355
{
	unsigned long flags;
3356
	struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
3357 3358 3359
	struct rcu_node *rnp = rcu_get_root(rsp);

	/* Set up local state, ensuring consistent view of global state. */
P
Paul E. McKenney 已提交
3360
	raw_spin_lock_irqsave(&rnp->lock, flags);
3361
	rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo);
3362
	init_callback_list(rdp);
3363
	rdp->qlen_lazy = 0;
3364
	ACCESS_ONCE(rdp->qlen) = 0;
3365
	rdp->dynticks = &per_cpu(rcu_dynticks, cpu);
3366
	WARN_ON_ONCE(rdp->dynticks->dynticks_nesting != DYNTICK_TASK_EXIT_IDLE);
3367
	WARN_ON_ONCE(atomic_read(&rdp->dynticks->dynticks) != 1);
3368
	rdp->cpu = cpu;
3369
	rdp->rsp = rsp;
P
Paul E. McKenney 已提交
3370
	rcu_boot_init_nocb_percpu_data(rdp);
P
Paul E. McKenney 已提交
3371
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
3372 3373 3374 3375 3376 3377 3378
}

/*
 * Initialize a CPU's per-CPU RCU data.  Note that only one online or
 * offline event can be happening at a given time.  Note also that we
 * can accept some slop in the rsp->completed access due to the fact
 * that this CPU cannot possibly have any RCU callbacks in flight yet.
3379
 */
3380
static void
3381
rcu_init_percpu_data(int cpu, struct rcu_state *rsp)
3382 3383 3384
{
	unsigned long flags;
	unsigned long mask;
3385
	struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
3386 3387
	struct rcu_node *rnp = rcu_get_root(rsp);

3388 3389 3390
	/* Exclude new grace periods. */
	mutex_lock(&rsp->onoff_mutex);

3391
	/* Set up local state, ensuring consistent view of global state. */
P
Paul E. McKenney 已提交
3392
	raw_spin_lock_irqsave(&rnp->lock, flags);
3393
	rdp->beenonline = 1;	 /* We have now been online. */
3394 3395
	rdp->qlen_last_fqs_check = 0;
	rdp->n_force_qs_snap = rsp->n_force_qs;
3396
	rdp->blimit = blimit;
3397
	init_callback_list(rdp);  /* Re-enable callbacks on this CPU. */
3398
	rdp->dynticks->dynticks_nesting = DYNTICK_TASK_EXIT_IDLE;
3399
	rcu_sysidle_init_percpu_data(rdp->dynticks);
3400 3401
	atomic_set(&rdp->dynticks->dynticks,
		   (atomic_read(&rdp->dynticks->dynticks) & ~0x1) + 1);
P
Paul E. McKenney 已提交
3402
	raw_spin_unlock(&rnp->lock);		/* irqs remain disabled. */
3403 3404 3405 3406 3407 3408

	/* Add CPU to rcu_node bitmasks. */
	rnp = rdp->mynode;
	mask = rdp->grpmask;
	do {
		/* Exclude any attempts to start a new GP on small systems. */
P
Paul E. McKenney 已提交
3409
		raw_spin_lock(&rnp->lock);	/* irqs already disabled. */
3410 3411
		rnp->qsmaskinit |= mask;
		mask = rnp->grpmask;
3412
		if (rnp == rdp->mynode) {
3413 3414 3415 3416 3417 3418
			/*
			 * If there is a grace period in progress, we will
			 * set up to wait for it next time we run the
			 * RCU core code.
			 */
			rdp->gpnum = rnp->completed;
3419
			rdp->completed = rnp->completed;
3420 3421
			rdp->passed_quiesce = 0;
			rdp->qs_pending = 0;
3422
			trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpuonl"));
3423
		}
P
Paul E. McKenney 已提交
3424
		raw_spin_unlock(&rnp->lock); /* irqs already disabled. */
3425 3426
		rnp = rnp->parent;
	} while (rnp != NULL && !(rnp->qsmaskinit & mask));
3427
	local_irq_restore(flags);
3428

3429
	mutex_unlock(&rsp->onoff_mutex);
3430 3431
}

3432
static void rcu_prepare_cpu(int cpu)
3433
{
3434 3435 3436
	struct rcu_state *rsp;

	for_each_rcu_flavor(rsp)
3437
		rcu_init_percpu_data(cpu, rsp);
3438 3439 3440
}

/*
3441
 * Handle CPU online/offline notification events.
3442
 */
3443
static int rcu_cpu_notify(struct notifier_block *self,
3444
				    unsigned long action, void *hcpu)
3445 3446
{
	long cpu = (long)hcpu;
3447
	struct rcu_data *rdp = per_cpu_ptr(rcu_state_p->rda, cpu);
3448
	struct rcu_node *rnp = rdp->mynode;
3449
	struct rcu_state *rsp;
3450

3451
	trace_rcu_utilization(TPS("Start CPU hotplug"));
3452 3453 3454
	switch (action) {
	case CPU_UP_PREPARE:
	case CPU_UP_PREPARE_FROZEN:
P
Peter Zijlstra 已提交
3455 3456
		rcu_prepare_cpu(cpu);
		rcu_prepare_kthreads(cpu);
3457
		rcu_spawn_all_nocb_kthreads(cpu);
3458 3459
		break;
	case CPU_ONLINE:
3460
	case CPU_DOWN_FAILED:
T
Thomas Gleixner 已提交
3461
		rcu_boost_kthread_setaffinity(rnp, -1);
3462 3463
		break;
	case CPU_DOWN_PREPARE:
3464
		rcu_boost_kthread_setaffinity(rnp, cpu);
3465
		break;
3466 3467
	case CPU_DYING:
	case CPU_DYING_FROZEN:
3468 3469
		for_each_rcu_flavor(rsp)
			rcu_cleanup_dying_cpu(rsp);
3470
		break;
3471 3472 3473 3474
	case CPU_DEAD:
	case CPU_DEAD_FROZEN:
	case CPU_UP_CANCELED:
	case CPU_UP_CANCELED_FROZEN:
3475 3476
		for_each_rcu_flavor(rsp)
			rcu_cleanup_dead_cpu(cpu, rsp);
3477 3478 3479 3480
		break;
	default:
		break;
	}
3481
	trace_rcu_utilization(TPS("End CPU hotplug"));
3482
	return NOTIFY_OK;
3483 3484
}

3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503
static int rcu_pm_notify(struct notifier_block *self,
			 unsigned long action, void *hcpu)
{
	switch (action) {
	case PM_HIBERNATION_PREPARE:
	case PM_SUSPEND_PREPARE:
		if (nr_cpu_ids <= 256) /* Expediting bad for large systems. */
			rcu_expedited = 1;
		break;
	case PM_POST_HIBERNATION:
	case PM_POST_SUSPEND:
		rcu_expedited = 0;
		break;
	default:
		break;
	}
	return NOTIFY_OK;
}

3504
/*
3505
 * Spawn the kthreads that handle each RCU flavor's grace periods.
3506 3507 3508 3509 3510 3511 3512 3513
 */
static int __init rcu_spawn_gp_kthread(void)
{
	unsigned long flags;
	struct rcu_node *rnp;
	struct rcu_state *rsp;
	struct task_struct *t;

3514
	rcu_scheduler_fully_active = 1;
3515
	for_each_rcu_flavor(rsp) {
3516
		t = kthread_run(rcu_gp_kthread, rsp, "%s", rsp->name);
3517 3518 3519 3520 3521 3522
		BUG_ON(IS_ERR(t));
		rnp = rcu_get_root(rsp);
		raw_spin_lock_irqsave(&rnp->lock, flags);
		rsp->gp_kthread = t;
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
	}
3523
	rcu_spawn_nocb_kthreads();
3524
	rcu_spawn_boost_kthreads();
3525 3526 3527 3528
	return 0;
}
early_initcall(rcu_spawn_gp_kthread);

3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543
/*
 * This function is invoked towards the end of the scheduler's initialization
 * process.  Before this is called, the idle task might contain
 * RCU read-side critical sections (during which time, this idle
 * task is booting the system).  After this function is called, the
 * idle tasks are prohibited from containing RCU read-side critical
 * sections.  This function also enables RCU lockdep checking.
 */
void rcu_scheduler_starting(void)
{
	WARN_ON(num_online_cpus() != 1);
	WARN_ON(nr_context_switches() > 0);
	rcu_scheduler_active = 1;
}

3544 3545 3546 3547 3548 3549 3550 3551 3552
/*
 * Compute the per-level fanout, either using the exact fanout specified
 * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT.
 */
#ifdef CONFIG_RCU_FANOUT_EXACT
static void __init rcu_init_levelspread(struct rcu_state *rsp)
{
	int i;

3553 3554
	rsp->levelspread[rcu_num_lvls - 1] = rcu_fanout_leaf;
	for (i = rcu_num_lvls - 2; i >= 0; i--)
3555 3556 3557 3558 3559 3560 3561 3562 3563
		rsp->levelspread[i] = CONFIG_RCU_FANOUT;
}
#else /* #ifdef CONFIG_RCU_FANOUT_EXACT */
static void __init rcu_init_levelspread(struct rcu_state *rsp)
{
	int ccur;
	int cprv;
	int i;

3564
	cprv = nr_cpu_ids;
3565
	for (i = rcu_num_lvls - 1; i >= 0; i--) {
3566 3567 3568 3569 3570 3571 3572 3573 3574 3575
		ccur = rsp->levelcnt[i];
		rsp->levelspread[i] = (cprv + ccur - 1) / ccur;
		cprv = ccur;
	}
}
#endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */

/*
 * Helper function for rcu_init() that initializes one rcu_state structure.
 */
3576 3577
static void __init rcu_init_one(struct rcu_state *rsp,
		struct rcu_data __percpu *rda)
3578
{
3579 3580 3581 3582 3583 3584 3585 3586 3587 3588
	static const char * const buf[] = {
		"rcu_node_0",
		"rcu_node_1",
		"rcu_node_2",
		"rcu_node_3" };  /* Match MAX_RCU_LVLS */
	static const char * const fqs[] = {
		"rcu_node_fqs_0",
		"rcu_node_fqs_1",
		"rcu_node_fqs_2",
		"rcu_node_fqs_3" };  /* Match MAX_RCU_LVLS */
3589
	static u8 fl_mask = 0x1;
3590 3591 3592 3593 3594
	int cpustride = 1;
	int i;
	int j;
	struct rcu_node *rnp;

3595 3596
	BUILD_BUG_ON(MAX_RCU_LVLS > ARRAY_SIZE(buf));  /* Fix buf[] init! */

3597 3598 3599 3600
	/* Silence gcc 4.8 warning about array index out of range. */
	if (rcu_num_lvls > RCU_NUM_LVLS)
		panic("rcu_init_one: rcu_num_lvls overflow");

3601 3602
	/* Initialize the level-tracking arrays. */

3603 3604 3605
	for (i = 0; i < rcu_num_lvls; i++)
		rsp->levelcnt[i] = num_rcu_lvl[i];
	for (i = 1; i < rcu_num_lvls; i++)
3606 3607
		rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1];
	rcu_init_levelspread(rsp);
3608 3609
	rsp->flavor_mask = fl_mask;
	fl_mask <<= 1;
3610 3611 3612

	/* Initialize the elements themselves, starting from the leaves. */

3613
	for (i = rcu_num_lvls - 1; i >= 0; i--) {
3614 3615 3616
		cpustride *= rsp->levelspread[i];
		rnp = rsp->level[i];
		for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) {
P
Paul E. McKenney 已提交
3617
			raw_spin_lock_init(&rnp->lock);
3618 3619
			lockdep_set_class_and_name(&rnp->lock,
						   &rcu_node_class[i], buf[i]);
3620 3621 3622
			raw_spin_lock_init(&rnp->fqslock);
			lockdep_set_class_and_name(&rnp->fqslock,
						   &rcu_fqs_class[i], fqs[i]);
3623 3624
			rnp->gpnum = rsp->gpnum;
			rnp->completed = rsp->completed;
3625 3626 3627 3628
			rnp->qsmask = 0;
			rnp->qsmaskinit = 0;
			rnp->grplo = j * cpustride;
			rnp->grphi = (j + 1) * cpustride - 1;
3629 3630
			if (rnp->grphi >= nr_cpu_ids)
				rnp->grphi = nr_cpu_ids - 1;
3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641
			if (i == 0) {
				rnp->grpnum = 0;
				rnp->grpmask = 0;
				rnp->parent = NULL;
			} else {
				rnp->grpnum = j % rsp->levelspread[i - 1];
				rnp->grpmask = 1UL << rnp->grpnum;
				rnp->parent = rsp->level[i - 1] +
					      j / rsp->levelspread[i - 1];
			}
			rnp->level = i;
3642
			INIT_LIST_HEAD(&rnp->blkd_tasks);
3643
			rcu_init_one_nocb(rnp);
3644 3645
		}
	}
3646

3647
	rsp->rda = rda;
3648
	init_waitqueue_head(&rsp->gp_wq);
3649
	rnp = rsp->level[rcu_num_lvls - 1];
3650
	for_each_possible_cpu(i) {
3651
		while (i > rnp->grphi)
3652
			rnp++;
3653
		per_cpu_ptr(rsp->rda, i)->mynode = rnp;
3654 3655
		rcu_boot_init_percpu_data(i, rsp);
	}
3656
	list_add(&rsp->flavors, &rcu_struct_flavors);
3657 3658
}

3659 3660
/*
 * Compute the rcu_node tree geometry from kernel parameters.  This cannot
3661
 * replace the definitions in tree.h because those are needed to size
3662 3663 3664 3665
 * the ->node array in the rcu_state structure.
 */
static void __init rcu_init_geometry(void)
{
3666
	ulong d;
3667 3668
	int i;
	int j;
3669
	int n = nr_cpu_ids;
3670 3671
	int rcu_capacity[MAX_RCU_LVLS + 1];

3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683 3684
	/*
	 * Initialize any unspecified boot parameters.
	 * The default values of jiffies_till_first_fqs and
	 * jiffies_till_next_fqs are set to the RCU_JIFFIES_TILL_FORCE_QS
	 * value, which is a function of HZ, then adding one for each
	 * RCU_JIFFIES_FQS_DIV CPUs that might be on the system.
	 */
	d = RCU_JIFFIES_TILL_FORCE_QS + nr_cpu_ids / RCU_JIFFIES_FQS_DIV;
	if (jiffies_till_first_fqs == ULONG_MAX)
		jiffies_till_first_fqs = d;
	if (jiffies_till_next_fqs == ULONG_MAX)
		jiffies_till_next_fqs = d;

3685
	/* If the compile-time values are accurate, just leave. */
3686 3687
	if (rcu_fanout_leaf == CONFIG_RCU_FANOUT_LEAF &&
	    nr_cpu_ids == NR_CPUS)
3688
		return;
3689 3690
	pr_info("RCU: Adjusting geometry for rcu_fanout_leaf=%d, nr_cpu_ids=%d\n",
		rcu_fanout_leaf, nr_cpu_ids);
3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735

	/*
	 * Compute number of nodes that can be handled an rcu_node tree
	 * with the given number of levels.  Setting rcu_capacity[0] makes
	 * some of the arithmetic easier.
	 */
	rcu_capacity[0] = 1;
	rcu_capacity[1] = rcu_fanout_leaf;
	for (i = 2; i <= MAX_RCU_LVLS; i++)
		rcu_capacity[i] = rcu_capacity[i - 1] * CONFIG_RCU_FANOUT;

	/*
	 * The boot-time rcu_fanout_leaf parameter is only permitted
	 * to increase the leaf-level fanout, not decrease it.  Of course,
	 * the leaf-level fanout cannot exceed the number of bits in
	 * the rcu_node masks.  Finally, the tree must be able to accommodate
	 * the configured number of CPUs.  Complain and fall back to the
	 * compile-time values if these limits are exceeded.
	 */
	if (rcu_fanout_leaf < CONFIG_RCU_FANOUT_LEAF ||
	    rcu_fanout_leaf > sizeof(unsigned long) * 8 ||
	    n > rcu_capacity[MAX_RCU_LVLS]) {
		WARN_ON(1);
		return;
	}

	/* Calculate the number of rcu_nodes at each level of the tree. */
	for (i = 1; i <= MAX_RCU_LVLS; i++)
		if (n <= rcu_capacity[i]) {
			for (j = 0; j <= i; j++)
				num_rcu_lvl[j] =
					DIV_ROUND_UP(n, rcu_capacity[i - j]);
			rcu_num_lvls = i;
			for (j = i + 1; j <= MAX_RCU_LVLS; j++)
				num_rcu_lvl[j] = 0;
			break;
		}

	/* Calculate the total number of rcu_node structures. */
	rcu_num_nodes = 0;
	for (i = 0; i <= MAX_RCU_LVLS; i++)
		rcu_num_nodes += num_rcu_lvl[i];
	rcu_num_nodes -= n;
}

3736
void __init rcu_init(void)
3737
{
P
Paul E. McKenney 已提交
3738
	int cpu;
3739

3740
	rcu_bootup_announce();
3741
	rcu_init_geometry();
3742
	rcu_init_one(&rcu_bh_state, &rcu_bh_data);
3743
	rcu_init_one(&rcu_sched_state, &rcu_sched_data);
3744
	__rcu_init_preempt();
J
Jiang Fang 已提交
3745
	open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
3746 3747 3748 3749 3750 3751 3752

	/*
	 * We don't need protection against CPU-hotplug here because
	 * this is called early in boot, before either interrupts
	 * or the scheduler are operational.
	 */
	cpu_notifier(rcu_cpu_notify, 0);
3753
	pm_notifier(rcu_pm_notify, 0);
P
Paul E. McKenney 已提交
3754 3755
	for_each_online_cpu(cpu)
		rcu_cpu_notify(NULL, CPU_UP_PREPARE, (void *)(long)cpu);
3756 3757
}

3758
#include "tree_plugin.h"