tree.c 109.8 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
#define RCU_STATE_INITIALIZER(sname, sabbr, cr) \
83 84
static char sname##_varname[] = #sname; \
static const char *tp_##sname##_varname __used __tracepoint_string = sname##_varname; \
85
struct rcu_state sname##_state = { \
86
	.level = { &sname##_state.node[0] }, \
87
	.call = cr, \
88
	.fqs_state = RCU_GP_IDLE, \
P
Paul E. McKenney 已提交
89 90
	.gpnum = 0UL - 300UL, \
	.completed = 0UL - 300UL, \
91
	.orphan_lock = __RAW_SPIN_LOCK_UNLOCKED(&sname##_state.orphan_lock), \
92 93
	.orphan_nxttail = &sname##_state.orphan_nxtlist, \
	.orphan_donetail = &sname##_state.orphan_donelist, \
94
	.barrier_mutex = __MUTEX_INITIALIZER(sname##_state.barrier_mutex), \
95
	.onoff_mutex = __MUTEX_INITIALIZER(sname##_state.onoff_mutex), \
96
	.name = sname##_varname, \
97
	.abbr = sabbr, \
98 99
}; \
DEFINE_PER_CPU(struct rcu_data, sname##_data)
100

101 102
RCU_STATE_INITIALIZER(rcu_sched, 's', call_rcu_sched);
RCU_STATE_INITIALIZER(rcu_bh, 'b', call_rcu_bh);
103

104
static struct rcu_state *rcu_state;
105
LIST_HEAD(rcu_struct_flavors);
106

107 108
/* Increase (but not decrease) the CONFIG_RCU_FANOUT_LEAF at boot time. */
static int rcu_fanout_leaf = CONFIG_RCU_FANOUT_LEAF;
109
module_param(rcu_fanout_leaf, int, 0444);
110 111 112 113 114 115 116 117 118 119
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. */

120 121 122 123
/*
 * 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)
124
 * optimize synchronize_sched() to a simple barrier().  When this variable
125 126 127 128
 * 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.
 */
129 130 131
int rcu_scheduler_active __read_mostly;
EXPORT_SYMBOL_GPL(rcu_scheduler_active);

132 133 134 135 136 137 138 139 140 141 142 143 144 145
/*
 * 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;

146 147
#ifdef CONFIG_RCU_BOOST

148 149 150 151 152
/*
 * Control variables for per-CPU and per-rcu_node kthreads.  These
 * handle all flavors of RCU.
 */
static DEFINE_PER_CPU(struct task_struct *, rcu_cpu_kthread_task);
153
DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
154
DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
155
DEFINE_PER_CPU(char, rcu_cpu_has_work);
156

157 158
#endif /* #ifdef CONFIG_RCU_BOOST */

T
Thomas Gleixner 已提交
159
static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu);
160 161
static void invoke_rcu_core(void);
static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp);
162

163 164 165 166 167 168 169 170 171 172 173 174
/*
 * 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;

175 176 177 178 179 180 181 182 183 184
/*
 * 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);
}

185
/*
186
 * Note a quiescent state.  Because we do not need to know
187
 * how many quiescent states passed, just if there was at least
188
 * one since the start of the grace period, this just sets a flag.
189
 * The caller must have disabled preemption.
190
 */
191
void rcu_sched_qs(int cpu)
192
{
193
	struct rcu_data *rdp = &per_cpu(rcu_sched_data, cpu);
194

195
	if (rdp->passed_quiesce == 0)
196
		trace_rcu_grace_period(TPS("rcu_sched"), rdp->gpnum, TPS("cpuqs"));
197
	rdp->passed_quiesce = 1;
198 199
}

200
void rcu_bh_qs(int cpu)
201
{
202
	struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu);
203

204
	if (rdp->passed_quiesce == 0)
205
		trace_rcu_grace_period(TPS("rcu_bh"), rdp->gpnum, TPS("cpuqs"));
206
	rdp->passed_quiesce = 1;
207
}
208

209 210 211
/*
 * Note a context switch.  This is a quiescent state for RCU-sched,
 * and requires special handling for preemptible RCU.
212
 * The caller must have disabled preemption.
213 214 215
 */
void rcu_note_context_switch(int cpu)
{
216
	trace_rcu_utilization(TPS("Start context switch"));
217
	rcu_sched_qs(cpu);
218
	rcu_preempt_note_context_switch(cpu);
219
	trace_rcu_utilization(TPS("End context switch"));
220
}
221
EXPORT_SYMBOL_GPL(rcu_note_context_switch);
222

223
static DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = {
224
	.dynticks_nesting = DYNTICK_TASK_EXIT_IDLE,
225
	.dynticks = ATOMIC_INIT(1),
226 227 228 229
#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 */
230
};
231

E
Eric Dumazet 已提交
232 233 234
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. */
235

E
Eric Dumazet 已提交
236 237 238
module_param(blimit, long, 0444);
module_param(qhimark, long, 0444);
module_param(qlowmark, long, 0444);
239

240 241
static ulong jiffies_till_first_fqs = ULONG_MAX;
static ulong jiffies_till_next_fqs = ULONG_MAX;
242 243 244 245

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

246 247
static void rcu_start_gp_advanced(struct rcu_state *rsp, struct rcu_node *rnp,
				  struct rcu_data *rdp);
248 249 250 251
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);
252
static void force_quiescent_state(struct rcu_state *rsp);
253
static int rcu_pending(int cpu);
254 255

/*
256
 * Return the number of RCU-sched batches processed thus far for debug & stats.
257
 */
258
long rcu_batches_completed_sched(void)
259
{
260
	return rcu_sched_state.completed;
261
}
262
EXPORT_SYMBOL_GPL(rcu_batches_completed_sched);
263 264 265 266 267 268 269 270 271 272

/*
 * 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);

273 274 275 276 277
/*
 * Force a quiescent state for RCU BH.
 */
void rcu_bh_force_quiescent_state(void)
{
278
	force_quiescent_state(&rcu_bh_state);
279 280 281
}
EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state);

282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306
/*
 * 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);

/*
 * 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);

307 308 309 310 311
/*
 * Force a quiescent state for RCU-sched.
 */
void rcu_sched_force_quiescent_state(void)
{
312
	force_quiescent_state(&rcu_sched_state);
313 314 315
}
EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state);

316 317 318 319 320 321
/*
 * 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
322 323
	return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL] &&
	       rdp->nxttail[RCU_DONE_TAIL] != NULL;
324 325 326
}

/*
327 328 329
 * Does the current CPU require a not-yet-started grace period?
 * The caller must have disabled interrupts to prevent races with
 * normal callback registry.
330 331 332 333
 */
static int
cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp)
{
334
	int i;
P
Paul E. McKenney 已提交
335

336 337
	if (rcu_gp_in_progress(rsp))
		return 0;  /* No, a grace period is already in progress. */
338
	if (rcu_nocb_needs_gp(rsp))
339
		return 1;  /* Yes, a no-CBs CPU needs one. */
340 341 342 343 344 345 346 347 348 349
	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. */
350 351 352 353 354 355 356 357 358 359
}

/*
 * 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];
}

360
/*
361
 * rcu_eqs_enter_common - current CPU is moving towards extended quiescent state
362 363 364 365 366
 *
 * 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.
 */
367 368
static void rcu_eqs_enter_common(struct rcu_dynticks *rdtp, long long oldval,
				bool user)
369
{
370 371 372
	struct rcu_state *rsp;
	struct rcu_data *rdp;

373
	trace_rcu_dyntick(TPS("Start"), oldval, rdtp->dynticks_nesting);
374
	if (!user && !is_idle_task(current)) {
375 376
		struct task_struct *idle __maybe_unused =
			idle_task(smp_processor_id());
377

378
		trace_rcu_dyntick(TPS("Error on entry: not idle task"), oldval, 0);
379
		ftrace_dump(DUMP_ORIG);
380 381 382
		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! */
383
	}
384 385 386 387
	for_each_rcu_flavor(rsp) {
		rdp = this_cpu_ptr(rsp->rda);
		do_nocb_deferred_wakeup(rdp);
	}
388
	rcu_prepare_for_idle(smp_processor_id());
389 390 391 392 393
	/* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */
	smp_mb__before_atomic_inc();  /* See above. */
	atomic_inc(&rdtp->dynticks);
	smp_mb__after_atomic_inc();  /* Force ordering with next sojourn. */
	WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1);
394 395

	/*
396
	 * It is illegal to enter an extended quiescent state while
397 398 399 400 401 402 403 404
	 * 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.");
405
}
406

407 408 409
/*
 * Enter an RCU extended quiescent state, which can be either the
 * idle loop or adaptive-tickless usermode execution.
410
 */
411
static void rcu_eqs_enter(bool user)
412
{
413
	long long oldval;
414 415
	struct rcu_dynticks *rdtp;

416
	rdtp = this_cpu_ptr(&rcu_dynticks);
417
	oldval = rdtp->dynticks_nesting;
418
	WARN_ON_ONCE((oldval & DYNTICK_TASK_NEST_MASK) == 0);
419
	if ((oldval & DYNTICK_TASK_NEST_MASK) == DYNTICK_TASK_NEST_VALUE) {
420
		rdtp->dynticks_nesting = 0;
421 422
		rcu_eqs_enter_common(rdtp, oldval, user);
	} else {
423
		rdtp->dynticks_nesting -= DYNTICK_TASK_NEST_VALUE;
424
	}
425
}
426 427 428 429 430 431 432 433 434 435 436 437 438 439 440

/**
 * 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)
{
441 442 443
	unsigned long flags;

	local_irq_save(flags);
444
	rcu_eqs_enter(false);
445
	rcu_sysidle_enter(this_cpu_ptr(&rcu_dynticks), 0);
446
	local_irq_restore(flags);
447
}
448
EXPORT_SYMBOL_GPL(rcu_idle_enter);
449

450
#ifdef CONFIG_RCU_USER_QS
451 452 453 454 455 456 457 458 459 460
/**
 * 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)
{
461
	rcu_eqs_enter(1);
462
}
463
#endif /* CONFIG_RCU_USER_QS */
464

465 466 467 468 469 470
/**
 * 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.
471
 *
472 473 474 475 476 477 478 479
 * 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.
480
 */
481
void rcu_irq_exit(void)
482 483
{
	unsigned long flags;
484
	long long oldval;
485 486 487
	struct rcu_dynticks *rdtp;

	local_irq_save(flags);
488
	rdtp = this_cpu_ptr(&rcu_dynticks);
489
	oldval = rdtp->dynticks_nesting;
490 491
	rdtp->dynticks_nesting--;
	WARN_ON_ONCE(rdtp->dynticks_nesting < 0);
492
	if (rdtp->dynticks_nesting)
493
		trace_rcu_dyntick(TPS("--="), oldval, rdtp->dynticks_nesting);
494
	else
495
		rcu_eqs_enter_common(rdtp, oldval, true);
496
	rcu_sysidle_enter(rdtp, 1);
497 498 499 500
	local_irq_restore(flags);
}

/*
501
 * rcu_eqs_exit_common - current CPU moving away from extended quiescent state
502 503 504 505 506
 *
 * 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.
 */
507 508
static void rcu_eqs_exit_common(struct rcu_dynticks *rdtp, long long oldval,
			       int user)
509
{
510 511 512 513 514
	smp_mb__before_atomic_inc();  /* Force ordering w/previous sojourn. */
	atomic_inc(&rdtp->dynticks);
	/* CPUs seeing atomic_inc() must see later RCU read-side crit sects */
	smp_mb__after_atomic_inc();  /* See above. */
	WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1));
515
	rcu_cleanup_after_idle(smp_processor_id());
516
	trace_rcu_dyntick(TPS("End"), oldval, rdtp->dynticks_nesting);
517
	if (!user && !is_idle_task(current)) {
518 519
		struct task_struct *idle __maybe_unused =
			idle_task(smp_processor_id());
520

521
		trace_rcu_dyntick(TPS("Error on exit: not idle task"),
522
				  oldval, rdtp->dynticks_nesting);
523
		ftrace_dump(DUMP_ORIG);
524 525 526
		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! */
527 528 529
	}
}

530 531 532
/*
 * Exit an RCU extended quiescent state, which can be either the
 * idle loop or adaptive-tickless usermode execution.
533
 */
534
static void rcu_eqs_exit(bool user)
535 536 537 538
{
	struct rcu_dynticks *rdtp;
	long long oldval;

539
	rdtp = this_cpu_ptr(&rcu_dynticks);
540
	oldval = rdtp->dynticks_nesting;
541
	WARN_ON_ONCE(oldval < 0);
542
	if (oldval & DYNTICK_TASK_NEST_MASK) {
543
		rdtp->dynticks_nesting += DYNTICK_TASK_NEST_VALUE;
544
	} else {
545
		rdtp->dynticks_nesting = DYNTICK_TASK_EXIT_IDLE;
546 547
		rcu_eqs_exit_common(rdtp, oldval, user);
	}
548
}
549 550 551 552 553 554 555 556 557 558 559 560 561 562

/**
 * 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)
{
563 564 565
	unsigned long flags;

	local_irq_save(flags);
566
	rcu_eqs_exit(false);
567
	rcu_sysidle_exit(this_cpu_ptr(&rcu_dynticks), 0);
568
	local_irq_restore(flags);
569
}
570
EXPORT_SYMBOL_GPL(rcu_idle_exit);
571

572
#ifdef CONFIG_RCU_USER_QS
573 574 575 576 577 578 579 580
/**
 * 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)
{
581
	rcu_eqs_exit(1);
582
}
583
#endif /* CONFIG_RCU_USER_QS */
584

585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610
/**
 * 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);
611
	rdtp = this_cpu_ptr(&rcu_dynticks);
612 613 614
	oldval = rdtp->dynticks_nesting;
	rdtp->dynticks_nesting++;
	WARN_ON_ONCE(rdtp->dynticks_nesting == 0);
615
	if (oldval)
616
		trace_rcu_dyntick(TPS("++="), oldval, rdtp->dynticks_nesting);
617
	else
618
		rcu_eqs_exit_common(rdtp, oldval, true);
619
	rcu_sysidle_exit(rdtp, 1);
620 621 622 623 624 625 626 627 628 629 630 631
	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)
{
632
	struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
633

634 635
	if (rdtp->dynticks_nmi_nesting == 0 &&
	    (atomic_read(&rdtp->dynticks) & 0x1))
636
		return;
637 638 639 640 641 642
	rdtp->dynticks_nmi_nesting++;
	smp_mb__before_atomic_inc();  /* Force delay from prior write. */
	atomic_inc(&rdtp->dynticks);
	/* CPUs seeing atomic_inc() must see later RCU read-side crit sects */
	smp_mb__after_atomic_inc();  /* See above. */
	WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1));
643 644 645 646 647 648 649 650 651 652 653
}

/**
 * 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)
{
654
	struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
655

656 657
	if (rdtp->dynticks_nmi_nesting == 0 ||
	    --rdtp->dynticks_nmi_nesting != 0)
658
		return;
659 660 661 662 663
	/* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */
	smp_mb__before_atomic_inc();  /* See above. */
	atomic_inc(&rdtp->dynticks);
	smp_mb__after_atomic_inc();  /* Force delay to next write. */
	WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1);
664 665 666
}

/**
667 668 669 670 671 672 673
 * __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.
 */
674
bool notrace __rcu_is_watching(void)
675 676 677 678 679 680
{
	return atomic_read(this_cpu_ptr(&rcu_dynticks.dynticks)) & 0x1;
}

/**
 * rcu_is_watching - see if RCU thinks that the current CPU is idle
681
 *
682
 * If the current CPU is in its idle loop and is neither in an interrupt
683
 * or NMI handler, return true.
684
 */
685
bool notrace rcu_is_watching(void)
686
{
687 688 689
	int ret;

	preempt_disable();
690
	ret = __rcu_is_watching();
691 692
	preempt_enable();
	return ret;
693
}
694
EXPORT_SYMBOL_GPL(rcu_is_watching);
695

696
#if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU)
697 698 699 700 701 702 703

/*
 * 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
704 705 706 707 708 709 710 711 712 713 714
 * 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.
715 716 717 718 719 720
 *
 * Disable checking if in an NMI handler because we cannot safely report
 * errors from NMI handlers anyway.
 */
bool rcu_lockdep_current_cpu_online(void)
{
721 722
	struct rcu_data *rdp;
	struct rcu_node *rnp;
723 724 725
	bool ret;

	if (in_nmi())
F
Fengguang Wu 已提交
726
		return true;
727
	preempt_disable();
728
	rdp = this_cpu_ptr(&rcu_sched_data);
729 730
	rnp = rdp->mynode;
	ret = (rdp->grpmask & rnp->qsmaskinit) ||
731 732 733 734 735 736
	      !rcu_scheduler_fully_active;
	preempt_enable();
	return ret;
}
EXPORT_SYMBOL_GPL(rcu_lockdep_current_cpu_online);

737
#endif /* #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) */
738

739
/**
740
 * rcu_is_cpu_rrupt_from_idle - see if idle or immediately interrupted from idle
741
 *
742 743 744
 * 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.
745
 */
746
static int rcu_is_cpu_rrupt_from_idle(void)
747
{
748
	return __this_cpu_read(rcu_dynticks.dynticks_nesting) <= 1;
749 750 751 752 753
}

/*
 * Snapshot the specified CPU's dynticks counter so that we can later
 * credit them with an implicit quiescent state.  Return 1 if this CPU
754
 * is in dynticks idle mode, which is an extended quiescent state.
755
 */
756 757
static int dyntick_save_progress_counter(struct rcu_data *rdp,
					 bool *isidle, unsigned long *maxj)
758
{
759
	rdp->dynticks_snap = atomic_add_return(0, &rdp->dynticks->dynticks);
760
	rcu_sysidle_check_cpu(rdp, isidle, maxj);
761
	return (rdp->dynticks_snap & 0x1) == 0;
762 763
}

764 765 766 767 768 769
/*
 * 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);

770 771 772 773
/*
 * 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()
774
 * for this same CPU, or by virtue of having been offline.
775
 */
776 777
static int rcu_implicit_dynticks_qs(struct rcu_data *rdp,
				    bool *isidle, unsigned long *maxj)
778
{
779 780
	unsigned int curr;
	unsigned int snap;
781

782 783
	curr = (unsigned int)atomic_add_return(0, &rdp->dynticks->dynticks);
	snap = (unsigned int)rdp->dynticks_snap;
784 785 786 787 788 789 790 791 792

	/*
	 * 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.
	 */
793
	if ((curr & 0x1) == 0 || UINT_CMP_GE(curr, snap + 2)) {
794
		trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("dti"));
795 796 797 798
		rdp->dynticks_fqs++;
		return 1;
	}

799 800 801 802 803 804 805 806 807 808 809 810 811 812 813
	/*
	 * 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)) {
814
		trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("ofl"));
815 816 817
		rdp->offline_fqs++;
		return 1;
	}
818 819 820 821 822 823 824 825 826 827

	/*
	 * There is a possibility that a CPU in adaptive-ticks state
	 * might run in the kernel with the scheduling-clock tick disabled
	 * for an extended time period.  Invoke rcu_kick_nohz_cpu() to
	 * force the CPU to restart the scheduling-clock tick in this
	 * CPU is in this state.
	 */
	rcu_kick_nohz_cpu(rdp->cpu);

828 829 830 831 832 833 834 835 836 837
	/*
	 * Alternatively, the CPU might be running in the kernel
	 * for an extended period of time without a quiescent state.
	 * Attempt to force the CPU through the scheduler to gain the
	 * needed quiescent state, but only if the grace period has gone
	 * on for an uncommonly long time.  If there are many stuck CPUs,
	 * we will beat on the first one until it gets unstuck, then move
	 * to the next.  Only do this for the primary flavor of RCU.
	 */
	if (rdp->rsp == rcu_state &&
838
	    ULONG_CMP_GE(jiffies, rdp->rsp->jiffies_resched)) {
839 840 841 842
		rdp->rsp->jiffies_resched += 5;
		resched_cpu(rdp->cpu);
	}

843
	return 0;
844 845 846 847
}

static void record_gp_stall_check_time(struct rcu_state *rsp)
{
848
	unsigned long j = jiffies;
849
	unsigned long j1;
850 851 852

	rsp->gp_start = j;
	smp_wmb(); /* Record start time before stall time. */
853 854 855
	j1 = rcu_jiffies_till_stall_check();
	rsp->jiffies_stall = j + j1;
	rsp->jiffies_resched = j + j1 / 2;
856 857
}

858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880
/*
 * Dump stacks of all tasks running on stalled CPUs.  This is a fallback
 * for architectures that do not implement trigger_all_cpu_backtrace().
 * The NMI-triggered stack traces are more accurate because they are
 * printed by the target CPU.
 */
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);
	}
}

881 882 883 884 885
static void print_other_cpu_stall(struct rcu_state *rsp)
{
	int cpu;
	long delta;
	unsigned long flags;
886
	int ndetected = 0;
887
	struct rcu_node *rnp = rcu_get_root(rsp);
888
	long totqlen = 0;
889 890 891

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

P
Paul E. McKenney 已提交
892
	raw_spin_lock_irqsave(&rnp->lock, flags);
893
	delta = jiffies - rsp->jiffies_stall;
894
	if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) {
P
Paul E. McKenney 已提交
895
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
896 897
		return;
	}
898
	rsp->jiffies_stall = jiffies + 3 * rcu_jiffies_till_stall_check() + 3;
P
Paul E. McKenney 已提交
899
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
900

901 902 903 904 905
	/*
	 * OK, time to rat on our buddy...
	 * See Documentation/RCU/stallwarn.txt for info on how to debug
	 * RCU CPU stall warnings.
	 */
906
	pr_err("INFO: %s detected stalls on CPUs/tasks:",
907
	       rsp->name);
908
	print_cpu_stall_info_begin();
909
	rcu_for_each_leaf_node(rsp, rnp) {
910
		raw_spin_lock_irqsave(&rnp->lock, flags);
911
		ndetected += rcu_print_task_stall(rnp);
912 913 914 915 916 917 918 919
		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++;
				}
		}
920
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
921
	}
922 923 924 925 926 927 928

	/*
	 * 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);
929
	ndetected += rcu_print_task_stall(rnp);
930 931 932
	raw_spin_unlock_irqrestore(&rnp->lock, flags);

	print_cpu_stall_info_end();
933 934 935
	for_each_possible_cpu(cpu)
		totqlen += per_cpu_ptr(rsp->rda, cpu)->qlen;
	pr_cont("(detected by %d, t=%ld jiffies, g=%lu, c=%lu, q=%lu)\n",
936
	       smp_processor_id(), (long)(jiffies - rsp->gp_start),
937
	       rsp->gpnum, rsp->completed, totqlen);
938
	if (ndetected == 0)
939
		pr_err("INFO: Stall ended before state dump start\n");
940
	else if (!trigger_all_cpu_backtrace())
941
		rcu_dump_cpu_stacks(rsp);
942

943
	/* Complain about tasks blocking the grace period. */
944 945 946

	rcu_print_detail_task_stall(rsp);

947
	force_quiescent_state(rsp);  /* Kick them all. */
948 949
}

950 951 952 953 954 955
/*
 * 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);

956 957
static void print_cpu_stall(struct rcu_state *rsp)
{
958
	int cpu;
959 960
	unsigned long flags;
	struct rcu_node *rnp = rcu_get_root(rsp);
961
	long totqlen = 0;
962

963 964 965 966 967
	/*
	 * OK, time to rat on ourselves...
	 * See Documentation/RCU/stallwarn.txt for info on how to debug
	 * RCU CPU stall warnings.
	 */
968
	pr_err("INFO: %s self-detected stall on CPU", rsp->name);
969 970 971
	print_cpu_stall_info_begin();
	print_cpu_stall_info(rsp, smp_processor_id());
	print_cpu_stall_info_end();
972 973 974 975
	for_each_possible_cpu(cpu)
		totqlen += per_cpu_ptr(rsp->rda, cpu)->qlen;
	pr_cont(" (t=%lu jiffies g=%lu c=%lu q=%lu)\n",
		jiffies - rsp->gp_start, rsp->gpnum, rsp->completed, totqlen);
976 977
	if (!trigger_all_cpu_backtrace())
		dump_stack();
978

P
Paul E. McKenney 已提交
979
	raw_spin_lock_irqsave(&rnp->lock, flags);
980
	if (ULONG_CMP_GE(jiffies, rsp->jiffies_stall))
981
		rsp->jiffies_stall = jiffies +
982
				     3 * rcu_jiffies_till_stall_check() + 3;
P
Paul E. McKenney 已提交
983
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
984

985 986 987 988 989 990 991 992
	/*
	 * 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());
993 994 995 996
}

static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
{
997 998 999
	unsigned long completed;
	unsigned long gpnum;
	unsigned long gps;
1000 1001
	unsigned long j;
	unsigned long js;
1002 1003
	struct rcu_node *rnp;

1004
	if (rcu_cpu_stall_suppress || !rcu_gp_in_progress(rsp))
1005
		return;
1006
	j = jiffies;
1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026

	/*
	 * 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... */
1027
	js = ACCESS_ONCE(rsp->jiffies_stall);
1028 1029 1030 1031 1032 1033 1034 1035
	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. */
1036
	rnp = rdp->mynode;
1037
	if (rcu_gp_in_progress(rsp) &&
1038
	    (ACCESS_ONCE(rnp->qsmask) & rdp->grpmask)) {
1039 1040 1041 1042

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

1043 1044
	} else if (rcu_gp_in_progress(rsp) &&
		   ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY)) {
1045

1046
		/* They had a few time units to dump stack, so complain. */
1047 1048 1049 1050
		print_other_cpu_stall(rsp);
	}
}

1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061
/**
 * 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)
{
1062 1063 1064 1065
	struct rcu_state *rsp;

	for_each_rcu_flavor(rsp)
		rsp->jiffies_stall = jiffies + ULONG_MAX / 2;
1066 1067
}

1068 1069 1070 1071 1072 1073 1074
/*
 * Initialize the specified rcu_data structure's callback list to empty.
 */
static void init_callback_list(struct rcu_data *rdp)
{
	int i;

1075 1076
	if (init_nocb_callback_list(rdp))
		return;
1077 1078 1079 1080 1081
	rdp->nxtlist = NULL;
	for (i = 0; i < RCU_NEXT_SIZE; i++)
		rdp->nxttail[i] = &rdp->nxtlist;
}

1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110
/*
 * 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;
}

1111 1112 1113 1114 1115
/*
 * 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,
1116
				unsigned long c, const char *s)
1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141
{
	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
 * rcu_node structure's ->need_future_gp field.
 *
 * The caller must hold the specified rcu_node structure's ->lock.
 */
static unsigned long __maybe_unused
rcu_start_future_gp(struct rcu_node *rnp, struct rcu_data *rdp)
{
	unsigned long c;
	int i;
	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);
1142
	trace_rcu_future_gp(rnp, rdp, c, TPS("Startleaf"));
1143
	if (rnp->need_future_gp[c & 0x1]) {
1144
		trace_rcu_future_gp(rnp, rdp, c, TPS("Prestartleaf"));
1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157
		return c;
	}

	/*
	 * 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
	 * need to explicitly start one.
	 */
	if (rnp->gpnum != rnp->completed ||
	    ACCESS_ONCE(rnp->gpnum) != ACCESS_ONCE(rnp->completed)) {
		rnp->need_future_gp[c & 0x1]++;
1158
		trace_rcu_future_gp(rnp, rdp, c, TPS("Startedleaf"));
1159 1160 1161 1162 1163 1164 1165 1166
		return c;
	}

	/*
	 * 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).
	 */
1167
	if (rnp != rnp_root) {
1168
		raw_spin_lock(&rnp_root->lock);
1169 1170
		smp_mb__after_unlock_lock();
	}
1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187

	/*
	 * 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]) {
1188
		trace_rcu_future_gp(rnp, rdp, c, TPS("Prestartedroot"));
1189 1190 1191 1192 1193 1194 1195 1196
		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) {
1197
		trace_rcu_future_gp(rnp, rdp, c, TPS("Startedleafroot"));
1198
	} else {
1199
		trace_rcu_future_gp(rnp, rdp, c, TPS("Startedroot"));
1200
		rcu_start_gp_advanced(rdp->rsp, rnp_root, rdp);
1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222
	}
unlock_out:
	if (rnp != rnp_root)
		raw_spin_unlock(&rnp_root->lock);
	return c;
}

/*
 * 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];
1223 1224
	trace_rcu_future_gp(rnp, rdp, c,
			    needmore ? TPS("CleanupMore") : TPS("Cleanup"));
1225 1226 1227
	return needmore;
}

1228 1229 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 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286
/*
 * 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
 * not hurt to call it repeatedly.
 *
 * The caller must hold rnp->lock with interrupts disabled.
 */
static void rcu_accelerate_cbs(struct rcu_state *rsp, struct rcu_node *rnp,
			       struct rcu_data *rdp)
{
	unsigned long c;
	int i;

	/* If the CPU has no callbacks, nothing to do. */
	if (!rdp->nxttail[RCU_NEXT_TAIL] || !*rdp->nxttail[RCU_DONE_TAIL])
		return;

	/*
	 * 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)
		return;

	/*
	 * 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;
	}
1287 1288
	/* Record any needed additional grace periods. */
	rcu_start_future_gp(rnp, rdp);
1289 1290 1291

	/* Trace depending on how much we were able to accelerate. */
	if (!*rdp->nxttail[RCU_WAIT_TAIL])
1292
		trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("AccWaitCB"));
1293
	else
1294
		trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("AccReadyCB"));
1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339
}

/*
 * 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...
 *
 * The caller must hold rnp->lock with interrupts disabled.
 */
static void rcu_advance_cbs(struct rcu_state *rsp, struct rcu_node *rnp,
			    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])
		return;

	/*
	 * 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. */
	rcu_accelerate_cbs(rsp, rnp, rdp);
}

1340
/*
1341 1342 1343
 * 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.
1344
 */
1345
static void __note_gp_changes(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp)
1346
{
1347
	/* Handle the ends of any preceding grace periods first. */
1348
	if (rdp->completed == rnp->completed) {
1349

1350
		/* No grace period end, so just accelerate recent callbacks. */
1351
		rcu_accelerate_cbs(rsp, rnp, rdp);
1352

1353 1354 1355 1356
	} else {

		/* Advance callbacks. */
		rcu_advance_cbs(rsp, rnp, rdp);
1357 1358 1359

		/* Remember that we saw this grace-period completion. */
		rdp->completed = rnp->completed;
1360
		trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpuend"));
1361
	}
1362

1363 1364 1365 1366 1367 1368 1369
	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;
1370
		trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpustart"));
1371 1372 1373 1374 1375 1376
		rdp->passed_quiesce = 0;
		rdp->qs_pending = !!(rnp->qsmask & rdp->grpmask);
		zero_cpu_stall_ticks(rdp);
	}
}

1377
static void note_gp_changes(struct rcu_state *rsp, struct rcu_data *rdp)
1378 1379 1380 1381 1382 1383
{
	unsigned long flags;
	struct rcu_node *rnp;

	local_irq_save(flags);
	rnp = rdp->mynode;
1384 1385
	if ((rdp->gpnum == ACCESS_ONCE(rnp->gpnum) &&
	     rdp->completed == ACCESS_ONCE(rnp->completed)) || /* w/out lock. */
1386 1387 1388 1389
	    !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */
		local_irq_restore(flags);
		return;
	}
1390
	smp_mb__after_unlock_lock();
1391
	__note_gp_changes(rsp, rnp, rdp);
1392 1393 1394
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
}

1395
/*
1396
 * Initialize a new grace period.  Return 0 if no grace period required.
1397
 */
1398
static int rcu_gp_init(struct rcu_state *rsp)
1399 1400
{
	struct rcu_data *rdp;
1401
	struct rcu_node *rnp = rcu_get_root(rsp);
1402

1403
	rcu_bind_gp_kthread();
1404
	raw_spin_lock_irq(&rnp->lock);
1405
	smp_mb__after_unlock_lock();
1406 1407 1408 1409 1410
	if (rsp->gp_flags == 0) {
		/* Spurious wakeup, tell caller to go back to sleep.  */
		raw_spin_unlock_irq(&rnp->lock);
		return 0;
	}
1411
	rsp->gp_flags = 0; /* Clear all flags: New grace period. */
1412

1413 1414 1415 1416 1417
	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.
		 */
1418 1419 1420 1421 1422
		raw_spin_unlock_irq(&rnp->lock);
		return 0;
	}

	/* Advance to a new grace period and initialize state. */
1423
	record_gp_stall_check_time(rsp);
1424 1425
	/* Record GP times before starting GP, hence smp_store_release(). */
	smp_store_release(&rsp->gpnum, rsp->gpnum + 1);
1426
	trace_rcu_grace_period(rsp->name, rsp->gpnum, TPS("start"));
1427 1428 1429
	raw_spin_unlock_irq(&rnp->lock);

	/* Exclude any concurrent CPU-hotplug operations. */
1430
	mutex_lock(&rsp->onoff_mutex);
1431
	smp_mb__after_unlock_lock(); /* ->gpnum increment before GP! */
1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446

	/*
	 * 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) {
1447
		raw_spin_lock_irq(&rnp->lock);
1448
		smp_mb__after_unlock_lock();
1449
		rdp = this_cpu_ptr(rsp->rda);
1450 1451
		rcu_preempt_check_blocked_tasks(rnp);
		rnp->qsmask = rnp->qsmaskinit;
1452
		ACCESS_ONCE(rnp->gpnum) = rsp->gpnum;
1453
		WARN_ON_ONCE(rnp->completed != rsp->completed);
1454
		ACCESS_ONCE(rnp->completed) = rsp->completed;
1455
		if (rnp == rdp->mynode)
1456
			__note_gp_changes(rsp, rnp, rdp);
1457 1458 1459 1460 1461
		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);
1462
#ifdef CONFIG_PROVE_RCU_DELAY
1463
		if ((prandom_u32() % (rcu_num_nodes + 1)) == 0 &&
1464
		    system_state == SYSTEM_RUNNING)
1465
			udelay(200);
1466
#endif /* #ifdef CONFIG_PROVE_RCU_DELAY */
1467 1468
		cond_resched();
	}
1469

1470
	mutex_unlock(&rsp->onoff_mutex);
1471 1472
	return 1;
}
1473

1474 1475 1476
/*
 * Do one round of quiescent-state forcing.
 */
1477
static int rcu_gp_fqs(struct rcu_state *rsp, int fqs_state_in)
1478 1479
{
	int fqs_state = fqs_state_in;
1480 1481
	bool isidle = false;
	unsigned long maxj;
1482 1483 1484 1485 1486
	struct rcu_node *rnp = rcu_get_root(rsp);

	rsp->n_force_qs++;
	if (fqs_state == RCU_SAVE_DYNTICK) {
		/* Collect dyntick-idle snapshots. */
1487 1488 1489 1490
		if (is_sysidle_rcu_state(rsp)) {
			isidle = 1;
			maxj = jiffies - ULONG_MAX / 4;
		}
1491 1492
		force_qs_rnp(rsp, dyntick_save_progress_counter,
			     &isidle, &maxj);
1493
		rcu_sysidle_report_gp(rsp, isidle, maxj);
1494 1495 1496
		fqs_state = RCU_FORCE_QS;
	} else {
		/* Handle dyntick-idle and offline CPUs. */
1497
		isidle = 0;
1498
		force_qs_rnp(rsp, rcu_implicit_dynticks_qs, &isidle, &maxj);
1499 1500 1501 1502
	}
	/* Clear flag to prevent immediate re-entry. */
	if (ACCESS_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) {
		raw_spin_lock_irq(&rnp->lock);
1503
		smp_mb__after_unlock_lock();
1504 1505 1506 1507 1508 1509
		rsp->gp_flags &= ~RCU_GP_FLAG_FQS;
		raw_spin_unlock_irq(&rnp->lock);
	}
	return fqs_state;
}

1510 1511 1512
/*
 * Clean up after the old grace period.
 */
1513
static void rcu_gp_cleanup(struct rcu_state *rsp)
1514 1515
{
	unsigned long gp_duration;
1516
	int nocb = 0;
1517 1518
	struct rcu_data *rdp;
	struct rcu_node *rnp = rcu_get_root(rsp);
1519

1520
	raw_spin_lock_irq(&rnp->lock);
1521
	smp_mb__after_unlock_lock();
1522 1523 1524
	gp_duration = jiffies - rsp->gp_start;
	if (gp_duration > rsp->gp_max)
		rsp->gp_max = gp_duration;
1525

1526 1527 1528 1529 1530 1531 1532 1533
	/*
	 * 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.
	 */
1534
	raw_spin_unlock_irq(&rnp->lock);
1535

1536 1537 1538 1539 1540 1541 1542 1543 1544 1545
	/*
	 * 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) {
1546
		raw_spin_lock_irq(&rnp->lock);
1547
		smp_mb__after_unlock_lock();
1548
		ACCESS_ONCE(rnp->completed) = rsp->gpnum;
1549 1550
		rdp = this_cpu_ptr(rsp->rda);
		if (rnp == rdp->mynode)
1551
			__note_gp_changes(rsp, rnp, rdp);
1552
		/* smp_mb() provided by prior unlock-lock pair. */
1553
		nocb += rcu_future_gp_cleanup(rsp, rnp);
1554 1555
		raw_spin_unlock_irq(&rnp->lock);
		cond_resched();
1556
	}
1557 1558
	rnp = rcu_get_root(rsp);
	raw_spin_lock_irq(&rnp->lock);
1559
	smp_mb__after_unlock_lock(); /* Order GP before ->completed update. */
1560
	rcu_nocb_gp_set(rnp, nocb);
1561

1562 1563
	/* Declare grace period done. */
	ACCESS_ONCE(rsp->completed) = rsp->gpnum;
1564
	trace_rcu_grace_period(rsp->name, rsp->completed, TPS("end"));
1565
	rsp->fqs_state = RCU_GP_IDLE;
1566
	rdp = this_cpu_ptr(rsp->rda);
1567
	rcu_advance_cbs(rsp, rnp, rdp);  /* Reduce false positives below. */
1568
	if (cpu_needs_another_gp(rsp, rdp)) {
1569
		rsp->gp_flags = RCU_GP_FLAG_INIT;
1570 1571 1572 1573
		trace_rcu_grace_period(rsp->name,
				       ACCESS_ONCE(rsp->gpnum),
				       TPS("newreq"));
	}
1574 1575 1576 1577 1578 1579 1580 1581
	raw_spin_unlock_irq(&rnp->lock);
}

/*
 * Body of kthread that handles grace periods.
 */
static int __noreturn rcu_gp_kthread(void *arg)
{
1582
	int fqs_state;
1583
	int gf;
1584
	unsigned long j;
1585
	int ret;
1586 1587 1588 1589 1590 1591 1592
	struct rcu_state *rsp = arg;
	struct rcu_node *rnp = rcu_get_root(rsp);

	for (;;) {

		/* Handle grace-period start. */
		for (;;) {
1593 1594 1595
			trace_rcu_grace_period(rsp->name,
					       ACCESS_ONCE(rsp->gpnum),
					       TPS("reqwait"));
1596
			wait_event_interruptible(rsp->gp_wq,
1597
						 ACCESS_ONCE(rsp->gp_flags) &
1598
						 RCU_GP_FLAG_INIT);
1599
			/* Locking provides needed memory barrier. */
1600
			if (rcu_gp_init(rsp))
1601 1602 1603
				break;
			cond_resched();
			flush_signals(current);
1604 1605 1606
			trace_rcu_grace_period(rsp->name,
					       ACCESS_ONCE(rsp->gpnum),
					       TPS("reqwaitsig"));
1607
		}
1608

1609 1610
		/* Handle quiescent-state forcing. */
		fqs_state = RCU_SAVE_DYNTICK;
1611 1612 1613 1614 1615
		j = jiffies_till_first_fqs;
		if (j > HZ) {
			j = HZ;
			jiffies_till_first_fqs = HZ;
		}
1616
		ret = 0;
1617
		for (;;) {
1618 1619
			if (!ret)
				rsp->jiffies_force_qs = jiffies + j;
1620 1621 1622
			trace_rcu_grace_period(rsp->name,
					       ACCESS_ONCE(rsp->gpnum),
					       TPS("fqswait"));
1623
			ret = wait_event_interruptible_timeout(rsp->gp_wq,
1624 1625
					((gf = ACCESS_ONCE(rsp->gp_flags)) &
					 RCU_GP_FLAG_FQS) ||
1626 1627
					(!ACCESS_ONCE(rnp->qsmask) &&
					 !rcu_preempt_blocked_readers_cgp(rnp)),
1628
					j);
1629
			/* Locking provides needed memory barriers. */
1630
			/* If grace period done, leave loop. */
1631
			if (!ACCESS_ONCE(rnp->qsmask) &&
1632
			    !rcu_preempt_blocked_readers_cgp(rnp))
1633
				break;
1634
			/* If time for quiescent-state forcing, do it. */
1635 1636
			if (ULONG_CMP_GE(jiffies, rsp->jiffies_force_qs) ||
			    (gf & RCU_GP_FLAG_FQS)) {
1637 1638 1639
				trace_rcu_grace_period(rsp->name,
						       ACCESS_ONCE(rsp->gpnum),
						       TPS("fqsstart"));
1640
				fqs_state = rcu_gp_fqs(rsp, fqs_state);
1641 1642 1643
				trace_rcu_grace_period(rsp->name,
						       ACCESS_ONCE(rsp->gpnum),
						       TPS("fqsend"));
1644 1645 1646 1647 1648
				cond_resched();
			} else {
				/* Deal with stray signal. */
				cond_resched();
				flush_signals(current);
1649 1650 1651
				trace_rcu_grace_period(rsp->name,
						       ACCESS_ONCE(rsp->gpnum),
						       TPS("fqswaitsig"));
1652
			}
1653 1654 1655 1656 1657 1658 1659 1660
			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;
			}
1661
		}
1662 1663 1664

		/* Handle grace-period end. */
		rcu_gp_cleanup(rsp);
1665 1666 1667
	}
}

1668 1669 1670 1671 1672 1673 1674 1675
static void rsp_wakeup(struct irq_work *work)
{
	struct rcu_state *rsp = container_of(work, struct rcu_state, wakeup_work);

	/* Wake up rcu_gp_kthread() to start the grace period. */
	wake_up(&rsp->gp_wq);
}

1676 1677 1678
/*
 * Start a new RCU grace period if warranted, re-initializing the hierarchy
 * in preparation for detecting the next grace period.  The caller must hold
1679
 * the root node's ->lock and hard irqs must be disabled.
1680 1681 1682 1683
 *
 * 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.
1684 1685
 */
static void
1686 1687
rcu_start_gp_advanced(struct rcu_state *rsp, struct rcu_node *rnp,
		      struct rcu_data *rdp)
1688
{
1689
	if (!rsp->gp_kthread || !cpu_needs_another_gp(rsp, rdp)) {
1690
		/*
1691
		 * Either we have not yet spawned the grace-period
1692 1693
		 * task, this CPU does not need another grace period,
		 * or a grace period is already in progress.
1694
		 * Either way, don't start a new grace period.
1695 1696 1697
		 */
		return;
	}
1698
	rsp->gp_flags = RCU_GP_FLAG_INIT;
1699 1700
	trace_rcu_grace_period(rsp->name, ACCESS_ONCE(rsp->gpnum),
			       TPS("newreq"));
1701

1702 1703
	/*
	 * We can't do wakeups while holding the rnp->lock, as that
1704 1705 1706
	 * could cause possible deadlocks with the rq->lock. Defer
	 * the wakeup to interrupt context.  And don't bother waking
	 * up the running kthread.
1707
	 */
1708 1709
	if (current != rsp->gp_kthread)
		irq_work_queue(&rsp->wakeup_work);
1710 1711
}

1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736
/*
 * 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.
 */
static void
rcu_start_gp(struct rcu_state *rsp)
{
	struct rcu_data *rdp = this_cpu_ptr(rsp->rda);
	struct rcu_node *rnp = rcu_get_root(rsp);

	/*
	 * 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!
	 */
	rcu_advance_cbs(rsp, rnp, rdp);
	rcu_start_gp_advanced(rsp, rnp, rdp);
}

1737
/*
P
Paul E. McKenney 已提交
1738 1739 1740
 * 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
1741 1742
 * if one is needed.  Note that the caller must hold rnp->lock, which
 * is released before return.
1743
 */
P
Paul E. McKenney 已提交
1744
static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags)
1745
	__releases(rcu_get_root(rsp)->lock)
1746
{
1747
	WARN_ON_ONCE(!rcu_gp_in_progress(rsp));
1748 1749
	raw_spin_unlock_irqrestore(&rcu_get_root(rsp)->lock, flags);
	wake_up(&rsp->gp_wq);  /* Memory barrier implied by wake_up() path. */
1750 1751
}

1752
/*
P
Paul E. McKenney 已提交
1753 1754 1755 1756 1757 1758
 * 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.
1759 1760
 */
static void
P
Paul E. McKenney 已提交
1761 1762
rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp,
		  struct rcu_node *rnp, unsigned long flags)
1763 1764
	__releases(rnp->lock)
{
1765 1766
	struct rcu_node *rnp_c;

1767 1768 1769 1770 1771
	/* 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
1772
			raw_spin_unlock_irqrestore(&rnp->lock, flags);
1773 1774 1775
			return;
		}
		rnp->qsmask &= ~mask;
1776 1777 1778 1779
		trace_rcu_quiescent_state_report(rsp->name, rnp->gpnum,
						 mask, rnp->qsmask, rnp->level,
						 rnp->grplo, rnp->grphi,
						 !!rnp->gp_tasks);
1780
		if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) {
1781 1782

			/* Other bits still set at this level, so done. */
P
Paul E. McKenney 已提交
1783
			raw_spin_unlock_irqrestore(&rnp->lock, flags);
1784 1785 1786 1787 1788 1789 1790 1791 1792
			return;
		}
		mask = rnp->grpmask;
		if (rnp->parent == NULL) {

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

			break;
		}
P
Paul E. McKenney 已提交
1793
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
1794
		rnp_c = rnp;
1795
		rnp = rnp->parent;
P
Paul E. McKenney 已提交
1796
		raw_spin_lock_irqsave(&rnp->lock, flags);
1797
		smp_mb__after_unlock_lock();
1798
		WARN_ON_ONCE(rnp_c->qsmask);
1799 1800 1801 1802
	}

	/*
	 * Get here if we are the last CPU to pass through a quiescent
P
Paul E. McKenney 已提交
1803
	 * state for this grace period.  Invoke rcu_report_qs_rsp()
1804
	 * to clean up and start the next grace period if one is needed.
1805
	 */
P
Paul E. McKenney 已提交
1806
	rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */
1807 1808 1809
}

/*
P
Paul E. McKenney 已提交
1810 1811 1812 1813 1814 1815 1816
 * 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!
1817 1818
 */
static void
1819
rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp)
1820 1821 1822 1823 1824 1825
{
	unsigned long flags;
	unsigned long mask;
	struct rcu_node *rnp;

	rnp = rdp->mynode;
P
Paul E. McKenney 已提交
1826
	raw_spin_lock_irqsave(&rnp->lock, flags);
1827
	smp_mb__after_unlock_lock();
1828 1829
	if (rdp->passed_quiesce == 0 || rdp->gpnum != rnp->gpnum ||
	    rnp->completed == rnp->gpnum) {
1830 1831

		/*
1832 1833 1834 1835
		 * 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.
1836
		 */
1837
		rdp->passed_quiesce = 0;	/* need qs for new gp. */
P
Paul E. McKenney 已提交
1838
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
1839 1840 1841 1842
		return;
	}
	mask = rdp->grpmask;
	if ((rnp->qsmask & mask) == 0) {
P
Paul E. McKenney 已提交
1843
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
1844 1845 1846 1847 1848 1849 1850
	} 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.
		 */
1851
		rcu_accelerate_cbs(rsp, rnp, rdp);
1852

P
Paul E. McKenney 已提交
1853
		rcu_report_qs_rnp(mask, rsp, rnp, flags); /* rlses rnp->lock */
1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865
	}
}

/*
 * 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)
{
1866 1867
	/* Check for grace-period ends and beginnings. */
	note_gp_changes(rsp, rdp);
1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879

	/*
	 * 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.
	 */
1880
	if (!rdp->passed_quiesce)
1881 1882
		return;

P
Paul E. McKenney 已提交
1883 1884 1885 1886
	/*
	 * Tell RCU we are done (but rcu_report_qs_rdp() will be the
	 * judge of that).
	 */
1887
	rcu_report_qs_rdp(rdp->cpu, rsp, rdp);
1888 1889 1890 1891
}

#ifdef CONFIG_HOTPLUG_CPU

1892
/*
1893 1894
 * Send the specified CPU's RCU callbacks to the orphanage.  The
 * specified CPU must be offline, and the caller must hold the
1895
 * ->orphan_lock.
1896
 */
1897 1898 1899
static void
rcu_send_cbs_to_orphanage(int cpu, struct rcu_state *rsp,
			  struct rcu_node *rnp, struct rcu_data *rdp)
1900
{
P
Paul E. McKenney 已提交
1901
	/* No-CBs CPUs do not have orphanable callbacks. */
1902
	if (rcu_is_nocb_cpu(rdp->cpu))
P
Paul E. McKenney 已提交
1903 1904
		return;

1905 1906
	/*
	 * Orphan the callbacks.  First adjust the counts.  This is safe
1907 1908
	 * because _rcu_barrier() excludes CPU-hotplug operations, so it
	 * cannot be running now.  Thus no memory barrier is required.
1909
	 */
1910
	if (rdp->nxtlist != NULL) {
1911 1912 1913
		rsp->qlen_lazy += rdp->qlen_lazy;
		rsp->qlen += rdp->qlen;
		rdp->n_cbs_orphaned += rdp->qlen;
1914
		rdp->qlen_lazy = 0;
1915
		ACCESS_ONCE(rdp->qlen) = 0;
1916 1917 1918
	}

	/*
1919 1920 1921 1922 1923 1924 1925
	 * 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.
1926
	 */
1927 1928 1929 1930
	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;
1931 1932 1933
	}

	/*
1934 1935 1936
	 * 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.
1937
	 */
1938
	if (rdp->nxtlist != NULL) {
1939 1940
		*rsp->orphan_donetail = rdp->nxtlist;
		rsp->orphan_donetail = rdp->nxttail[RCU_DONE_TAIL];
1941
	}
1942

1943
	/* Finally, initialize the rcu_data structure's list to empty.  */
1944
	init_callback_list(rdp);
1945 1946 1947 1948
}

/*
 * Adopt the RCU callbacks from the specified rcu_state structure's
1949
 * orphanage.  The caller must hold the ->orphan_lock.
1950
 */
1951
static void rcu_adopt_orphan_cbs(struct rcu_state *rsp, unsigned long flags)
1952 1953 1954 1955
{
	int i;
	struct rcu_data *rdp = __this_cpu_ptr(rsp->rda);

P
Paul E. McKenney 已提交
1956
	/* No-CBs CPUs are handled specially. */
1957
	if (rcu_nocb_adopt_orphan_cbs(rsp, rdp, flags))
P
Paul E. McKenney 已提交
1958 1959
		return;

1960 1961 1962 1963
	/* Do the accounting first. */
	rdp->qlen_lazy += rsp->qlen_lazy;
	rdp->qlen += rsp->qlen;
	rdp->n_cbs_adopted += rsp->qlen;
1964 1965
	if (rsp->qlen_lazy != rsp->qlen)
		rcu_idle_count_callbacks_posted();
1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004
	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);
2005 2006
	trace_rcu_grace_period(rsp->name,
			       rnp->gpnum + 1 - !!(rnp->qsmask & mask),
2007
			       TPS("cpuofl"));
2008 2009 2010
}

/*
2011
 * The CPU has been completely removed, and some other CPU is reporting
2012 2013
 * this fact from process context.  Do the remainder of the cleanup,
 * including orphaning the outgoing CPU's RCU callbacks, and also
2014 2015
 * 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.
2016
 */
2017
static void rcu_cleanup_dead_cpu(int cpu, struct rcu_state *rsp)
2018
{
2019 2020 2021
	unsigned long flags;
	unsigned long mask;
	int need_report = 0;
2022
	struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
2023
	struct rcu_node *rnp = rdp->mynode;  /* Outgoing CPU's rdp & rnp. */
2024

2025
	/* Adjust any no-longer-needed kthreads. */
T
Thomas Gleixner 已提交
2026
	rcu_boost_kthread_setaffinity(rnp, -1);
2027

2028
	/* Remove the dead CPU from the bitmasks in the rcu_node hierarchy. */
2029 2030

	/* Exclude any attempts to start a new grace period. */
2031
	mutex_lock(&rsp->onoff_mutex);
2032
	raw_spin_lock_irqsave(&rsp->orphan_lock, flags);
2033

2034 2035
	/* Orphan the dead CPU's callbacks, and adopt them if appropriate. */
	rcu_send_cbs_to_orphanage(cpu, rsp, rnp, rdp);
2036
	rcu_adopt_orphan_cbs(rsp, flags);
2037

2038 2039 2040 2041
	/* 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. */
2042
		smp_mb__after_unlock_lock();
2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059
		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
2060
	 * because invoking rcu_report_unblock_qs_rnp() with ->orphan_lock
2061 2062
	 * held leads to deadlock.
	 */
2063
	raw_spin_unlock(&rsp->orphan_lock); /* irqs remain disabled. */
2064 2065 2066 2067 2068 2069 2070
	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);
2071 2072 2073
	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);
2074 2075 2076
	init_callback_list(rdp);
	/* Disallow further callbacks on this CPU. */
	rdp->nxttail[RCU_NEXT_TAIL] = NULL;
2077
	mutex_unlock(&rsp->onoff_mutex);
2078 2079 2080 2081
}

#else /* #ifdef CONFIG_HOTPLUG_CPU */

2082
static void rcu_cleanup_dying_cpu(struct rcu_state *rsp)
2083 2084 2085
{
}

2086
static void rcu_cleanup_dead_cpu(int cpu, struct rcu_state *rsp)
2087 2088 2089 2090 2091 2092 2093 2094 2095
{
}

#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.
 */
2096
static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp)
2097 2098 2099
{
	unsigned long flags;
	struct rcu_head *next, *list, **tail;
E
Eric Dumazet 已提交
2100 2101
	long bl, count, count_lazy;
	int i;
2102

2103
	/* If no callbacks are ready, just return. */
2104
	if (!cpu_has_callbacks_ready_to_invoke(rdp)) {
2105
		trace_rcu_batch_start(rsp->name, rdp->qlen_lazy, rdp->qlen, 0);
2106 2107 2108
		trace_rcu_batch_end(rsp->name, 0, !!ACCESS_ONCE(rdp->nxtlist),
				    need_resched(), is_idle_task(current),
				    rcu_is_callbacks_kthread());
2109
		return;
2110
	}
2111 2112 2113 2114 2115 2116

	/*
	 * Extract the list of ready callbacks, disabling to prevent
	 * races with call_rcu() from interrupt handlers.
	 */
	local_irq_save(flags);
2117
	WARN_ON_ONCE(cpu_is_offline(smp_processor_id()));
2118
	bl = rdp->blimit;
2119
	trace_rcu_batch_start(rsp->name, rdp->qlen_lazy, rdp->qlen, bl);
2120 2121 2122 2123
	list = rdp->nxtlist;
	rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL];
	*rdp->nxttail[RCU_DONE_TAIL] = NULL;
	tail = rdp->nxttail[RCU_DONE_TAIL];
2124 2125 2126
	for (i = RCU_NEXT_SIZE - 1; i >= 0; i--)
		if (rdp->nxttail[i] == rdp->nxttail[RCU_DONE_TAIL])
			rdp->nxttail[i] = &rdp->nxtlist;
2127 2128 2129
	local_irq_restore(flags);

	/* Invoke callbacks. */
2130
	count = count_lazy = 0;
2131 2132 2133
	while (list) {
		next = list->next;
		prefetch(next);
2134
		debug_rcu_head_unqueue(list);
2135 2136
		if (__rcu_reclaim(rsp->name, list))
			count_lazy++;
2137
		list = next;
2138 2139 2140 2141
		/* Stop only if limit reached and CPU has something to do. */
		if (++count >= bl &&
		    (need_resched() ||
		     (!is_idle_task(current) && !rcu_is_callbacks_kthread())))
2142 2143 2144 2145
			break;
	}

	local_irq_save(flags);
2146 2147 2148
	trace_rcu_batch_end(rsp->name, count, !!list, need_resched(),
			    is_idle_task(current),
			    rcu_is_callbacks_kthread());
2149 2150 2151 2152 2153

	/* Update count, and requeue any remaining callbacks. */
	if (list != NULL) {
		*tail = rdp->nxtlist;
		rdp->nxtlist = list;
2154 2155 2156
		for (i = 0; i < RCU_NEXT_SIZE; i++)
			if (&rdp->nxtlist == rdp->nxttail[i])
				rdp->nxttail[i] = tail;
2157 2158 2159
			else
				break;
	}
2160 2161
	smp_mb(); /* List handling before counting for rcu_barrier(). */
	rdp->qlen_lazy -= count_lazy;
2162
	ACCESS_ONCE(rdp->qlen) -= count;
2163
	rdp->n_cbs_invoked += count;
2164 2165 2166 2167 2168

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

2169 2170 2171 2172 2173 2174
	/* 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;
2175
	WARN_ON_ONCE((rdp->nxtlist == NULL) != (rdp->qlen == 0));
2176

2177 2178
	local_irq_restore(flags);

2179
	/* Re-invoke RCU core processing if there are callbacks remaining. */
2180
	if (cpu_has_callbacks_ready_to_invoke(rdp))
2181
		invoke_rcu_core();
2182 2183 2184 2185 2186
}

/*
 * 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).
2187
 * Also schedule RCU core processing.
2188
 *
2189
 * This function must be called from hardirq context.  It is normally
2190 2191 2192 2193 2194
 * 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)
{
2195
	trace_rcu_utilization(TPS("Start scheduler-tick"));
2196
	increment_cpu_stall_ticks();
2197
	if (user || rcu_is_cpu_rrupt_from_idle()) {
2198 2199 2200 2201 2202

		/*
		 * 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
2203
		 * a quiescent state, so note it.
2204 2205
		 *
		 * No memory barrier is required here because both
2206 2207 2208
		 * 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.
2209 2210
		 */

2211 2212
		rcu_sched_qs(cpu);
		rcu_bh_qs(cpu);
2213 2214 2215 2216 2217 2218 2219

	} 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
2220
		 * critical section, so note it.
2221 2222
		 */

2223
		rcu_bh_qs(cpu);
2224
	}
2225
	rcu_preempt_check_callbacks(cpu);
2226
	if (rcu_pending(cpu))
2227
		invoke_rcu_core();
2228
	trace_rcu_utilization(TPS("End scheduler-tick"));
2229 2230 2231 2232 2233
}

/*
 * Scan the leaf rcu_node structures, processing dyntick state for any that
 * have not yet encountered a quiescent state, using the function specified.
2234 2235
 * Also initiate boosting for any threads blocked on the root rcu_node.
 *
2236
 * The caller must have suppressed start of new grace periods.
2237
 */
2238 2239 2240 2241
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)
2242 2243 2244 2245 2246
{
	unsigned long bit;
	int cpu;
	unsigned long flags;
	unsigned long mask;
2247
	struct rcu_node *rnp;
2248

2249
	rcu_for_each_leaf_node(rsp, rnp) {
2250
		cond_resched();
2251
		mask = 0;
P
Paul E. McKenney 已提交
2252
		raw_spin_lock_irqsave(&rnp->lock, flags);
2253
		smp_mb__after_unlock_lock();
2254
		if (!rcu_gp_in_progress(rsp)) {
P
Paul E. McKenney 已提交
2255
			raw_spin_unlock_irqrestore(&rnp->lock, flags);
2256
			return;
2257
		}
2258
		if (rnp->qsmask == 0) {
2259
			rcu_initiate_boost(rnp, flags); /* releases rnp->lock */
2260 2261
			continue;
		}
2262
		cpu = rnp->grplo;
2263
		bit = 1;
2264
		for (; cpu <= rnp->grphi; cpu++, bit <<= 1) {
2265 2266 2267 2268 2269 2270
			if ((rnp->qsmask & bit) != 0) {
				if ((rnp->qsmaskinit & bit) != 0)
					*isidle = 0;
				if (f(per_cpu_ptr(rsp->rda, cpu), isidle, maxj))
					mask |= bit;
			}
2271
		}
2272
		if (mask != 0) {
2273

P
Paul E. McKenney 已提交
2274 2275
			/* rcu_report_qs_rnp() releases rnp->lock. */
			rcu_report_qs_rnp(mask, rsp, rnp, flags);
2276 2277
			continue;
		}
P
Paul E. McKenney 已提交
2278
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
2279
	}
2280
	rnp = rcu_get_root(rsp);
2281 2282
	if (rnp->qsmask == 0) {
		raw_spin_lock_irqsave(&rnp->lock, flags);
2283
		smp_mb__after_unlock_lock();
2284 2285
		rcu_initiate_boost(rnp, flags); /* releases rnp->lock. */
	}
2286 2287 2288 2289 2290 2291
}

/*
 * Force quiescent states on reluctant CPUs, and also detect which
 * CPUs are in dyntick-idle mode.
 */
2292
static void force_quiescent_state(struct rcu_state *rsp)
2293 2294
{
	unsigned long flags;
2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306
	bool ret;
	struct rcu_node *rnp;
	struct rcu_node *rnp_old = NULL;

	/* Funnel through hierarchy to reduce memory contention. */
	rnp = per_cpu_ptr(rsp->rda, raw_smp_processor_id())->mynode;
	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) {
2307
			ACCESS_ONCE(rsp->n_force_qs_lh)++;
2308 2309 2310 2311 2312
			return;
		}
		rnp_old = rnp;
	}
	/* rnp_old == rcu_get_root(rsp), rnp == NULL. */
2313

2314 2315
	/* Reached the root of the rcu_node tree, acquire lock. */
	raw_spin_lock_irqsave(&rnp_old->lock, flags);
2316
	smp_mb__after_unlock_lock();
2317 2318
	raw_spin_unlock(&rnp_old->fqslock);
	if (ACCESS_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) {
2319
		ACCESS_ONCE(rsp->n_force_qs_lh)++;
2320
		raw_spin_unlock_irqrestore(&rnp_old->lock, flags);
2321
		return;  /* Someone beat us to it. */
2322
	}
2323
	rsp->gp_flags |= RCU_GP_FLAG_FQS;
2324
	raw_spin_unlock_irqrestore(&rnp_old->lock, flags);
2325
	wake_up(&rsp->gp_wq);  /* Memory barrier implied by wake_up() path. */
2326 2327 2328
}

/*
2329 2330 2331
 * 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.
2332 2333
 */
static void
2334
__rcu_process_callbacks(struct rcu_state *rsp)
2335 2336
{
	unsigned long flags;
2337
	struct rcu_data *rdp = __this_cpu_ptr(rsp->rda);
2338

2339 2340
	WARN_ON_ONCE(rdp->beenonline == 0);

2341 2342 2343 2344
	/* 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? */
2345
	local_irq_save(flags);
2346
	if (cpu_needs_another_gp(rsp, rdp)) {
2347
		raw_spin_lock(&rcu_get_root(rsp)->lock); /* irqs disabled. */
2348 2349
		rcu_start_gp(rsp);
		raw_spin_unlock_irqrestore(&rcu_get_root(rsp)->lock, flags);
2350 2351
	} else {
		local_irq_restore(flags);
2352 2353 2354
	}

	/* If there are callbacks ready, invoke them. */
2355
	if (cpu_has_callbacks_ready_to_invoke(rdp))
2356
		invoke_rcu_callbacks(rsp, rdp);
2357 2358 2359

	/* Do any needed deferred wakeups of rcuo kthreads. */
	do_nocb_deferred_wakeup(rdp);
2360 2361
}

2362
/*
2363
 * Do RCU core processing for the current CPU.
2364
 */
2365
static void rcu_process_callbacks(struct softirq_action *unused)
2366
{
2367 2368
	struct rcu_state *rsp;

2369 2370
	if (cpu_is_offline(smp_processor_id()))
		return;
2371
	trace_rcu_utilization(TPS("Start RCU core"));
2372 2373
	for_each_rcu_flavor(rsp)
		__rcu_process_callbacks(rsp);
2374
	trace_rcu_utilization(TPS("End RCU core"));
2375 2376
}

2377
/*
2378 2379 2380 2381 2382
 * 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.
2383
 */
2384
static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp)
2385
{
2386 2387
	if (unlikely(!ACCESS_ONCE(rcu_scheduler_fully_active)))
		return;
2388 2389
	if (likely(!rsp->boost)) {
		rcu_do_batch(rsp, rdp);
2390 2391
		return;
	}
2392
	invoke_rcu_callbacks_kthread();
2393 2394
}

2395
static void invoke_rcu_core(void)
2396
{
2397 2398
	if (cpu_online(smp_processor_id()))
		raise_softirq(RCU_SOFTIRQ);
2399 2400
}

2401 2402 2403 2404 2405
/*
 * 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)
2406
{
2407 2408 2409 2410
	/*
	 * If called from an extended quiescent state, invoke the RCU
	 * core in order to force a re-evaluation of RCU's idleness.
	 */
2411
	if (!rcu_is_watching() && cpu_online(smp_processor_id()))
2412 2413
		invoke_rcu_core();

2414
	/* If interrupts were disabled or CPU offline, don't invoke RCU core. */
2415
	if (irqs_disabled_flags(flags) || cpu_is_offline(smp_processor_id()))
2416
		return;
2417

2418 2419 2420 2421 2422 2423 2424
	/*
	 * 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.
	 */
2425
	if (unlikely(rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) {
2426 2427

		/* Are we ignoring a completed grace period? */
2428
		note_gp_changes(rsp, rdp);
2429 2430 2431 2432 2433

		/* 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);

2434
			raw_spin_lock(&rnp_root->lock);
2435
			smp_mb__after_unlock_lock();
2436 2437
			rcu_start_gp(rsp);
			raw_spin_unlock(&rnp_root->lock);
2438 2439 2440 2441 2442
		} 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)
2443
				force_quiescent_state(rsp);
2444 2445 2446
			rdp->n_force_qs_snap = rsp->n_force_qs;
			rdp->qlen_last_fqs_check = rdp->qlen;
		}
2447
	}
2448 2449
}

2450 2451 2452 2453 2454 2455 2456
/*
 * RCU callback function to leak a callback.
 */
static void rcu_leak_callback(struct rcu_head *rhp)
{
}

P
Paul E. McKenney 已提交
2457 2458 2459 2460 2461 2462
/*
 * 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.
 */
2463 2464
static void
__call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
P
Paul E. McKenney 已提交
2465
	   struct rcu_state *rsp, int cpu, bool lazy)
2466 2467 2468 2469
{
	unsigned long flags;
	struct rcu_data *rdp;

2470
	WARN_ON_ONCE((unsigned long)head & 0x3); /* Misaligned rcu_head! */
2471 2472 2473 2474 2475 2476
	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;
	}
2477 2478 2479 2480 2481 2482 2483 2484 2485 2486
	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);
2487
	rdp = this_cpu_ptr(rsp->rda);
2488 2489

	/* Add the callback to our list. */
P
Paul E. McKenney 已提交
2490 2491 2492 2493 2494
	if (unlikely(rdp->nxttail[RCU_NEXT_TAIL] == NULL) || cpu != -1) {
		int offline;

		if (cpu != -1)
			rdp = per_cpu_ptr(rsp->rda, cpu);
2495
		offline = !__call_rcu_nocb(rdp, head, lazy, flags);
P
Paul E. McKenney 已提交
2496
		WARN_ON_ONCE(offline);
2497 2498 2499 2500
		/* _call_rcu() is illegal on offline CPU; leak the callback. */
		local_irq_restore(flags);
		return;
	}
2501
	ACCESS_ONCE(rdp->qlen)++;
2502 2503
	if (lazy)
		rdp->qlen_lazy++;
2504 2505
	else
		rcu_idle_count_callbacks_posted();
2506 2507 2508
	smp_mb();  /* Count before adding callback for rcu_barrier(). */
	*rdp->nxttail[RCU_NEXT_TAIL] = head;
	rdp->nxttail[RCU_NEXT_TAIL] = &head->next;
2509

2510 2511
	if (__is_kfree_rcu_offset((unsigned long)func))
		trace_rcu_kfree_callback(rsp->name, head, (unsigned long)func,
2512
					 rdp->qlen_lazy, rdp->qlen);
2513
	else
2514
		trace_rcu_callback(rsp->name, head, rdp->qlen_lazy, rdp->qlen);
2515

2516 2517
	/* Go handle any RCU core processing required. */
	__call_rcu_core(rsp, rdp, head, flags);
2518 2519 2520 2521
	local_irq_restore(flags);
}

/*
2522
 * Queue an RCU-sched callback for invocation after a grace period.
2523
 */
2524
void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
2525
{
P
Paul E. McKenney 已提交
2526
	__call_rcu(head, func, &rcu_sched_state, -1, 0);
2527
}
2528
EXPORT_SYMBOL_GPL(call_rcu_sched);
2529 2530

/*
2531
 * Queue an RCU callback for invocation after a quicker grace period.
2532 2533 2534
 */
void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
{
P
Paul E. McKenney 已提交
2535
	__call_rcu(head, func, &rcu_bh_state, -1, 0);
2536 2537 2538
}
EXPORT_SYMBOL_GPL(call_rcu_bh);

2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549
/*
 * 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)
{
2550 2551
	int ret;

2552
	might_sleep();  /* Check for RCU read-side critical section. */
2553 2554 2555 2556
	preempt_disable();
	ret = num_online_cpus() <= 1;
	preempt_enable();
	return ret;
2557 2558
}

2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570
/**
 * 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
2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592
 * 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).
2593 2594 2595 2596 2597 2598 2599 2600 2601
 *
 * 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)
{
2602 2603 2604 2605
	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");
2606 2607
	if (rcu_blocking_is_gp())
		return;
2608 2609 2610 2611
	if (rcu_expedited)
		synchronize_sched_expedited();
	else
		wait_rcu_gp(call_rcu_sched);
2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622
}
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.
2623 2624 2625
 *
 * See the description of synchronize_sched() for more detailed information
 * on memory ordering guarantees.
2626 2627 2628
 */
void synchronize_rcu_bh(void)
{
2629 2630 2631 2632
	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");
2633 2634
	if (rcu_blocking_is_gp())
		return;
2635 2636 2637 2638
	if (rcu_expedited)
		synchronize_rcu_bh_expedited();
	else
		wait_rcu_gp(call_rcu_bh);
2639 2640 2641
}
EXPORT_SYMBOL_GPL(synchronize_rcu_bh);

2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693
/**
 * 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().
	 */
	return smp_load_acquire(&rcu_state->gpnum);
}
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.
	 */
	newstate = smp_load_acquire(&rcu_state->completed);
	if (ULONG_CMP_GE(oldstate, newstate))
		synchronize_rcu();
}
EXPORT_SYMBOL_GPL(cond_synchronize_rcu);

2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710
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;
}

2711 2712 2713 2714 2715 2716 2717 2718 2719 2720
/**
 * 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.
2721
 *
2722 2723 2724 2725
 * Note that it is illegal to call this function while holding any lock
 * that is acquired by a CPU-hotplug notifier.  And yes, it is also illegal
 * to call this function from a CPU-hotplug notifier.  Failing to observe
 * these restriction will result in deadlock.
2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749
 *
 * 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)
{
2750 2751
	long firstsnap, s, snap;
	int trycount = 0;
2752
	struct rcu_state *rsp = &rcu_sched_state;
2753

2754 2755 2756 2757 2758 2759 2760 2761
	/*
	 * 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.
	 */
2762 2763
	if (ULONG_CMP_GE((ulong)atomic_long_read(&rsp->expedited_start),
			 (ulong)atomic_long_read(&rsp->expedited_done) +
2764 2765
			 ULONG_MAX / 8)) {
		synchronize_sched();
2766
		atomic_long_inc(&rsp->expedited_wrap);
2767 2768
		return;
	}
2769

2770 2771 2772 2773
	/*
	 * Take a ticket.  Note that atomic_inc_return() implies a
	 * full memory barrier.
	 */
2774
	snap = atomic_long_inc_return(&rsp->expedited_start);
2775
	firstsnap = snap;
2776
	get_online_cpus();
2777
	WARN_ON_ONCE(cpu_is_offline(raw_smp_processor_id()));
2778 2779 2780 2781 2782 2783 2784 2785 2786

	/*
	 * 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();
2787
		atomic_long_inc(&rsp->expedited_tryfail);
2788

2789
		/* Check to see if someone else did our work for us. */
2790
		s = atomic_long_read(&rsp->expedited_done);
2791
		if (ULONG_CMP_GE((ulong)s, (ulong)firstsnap)) {
2792 2793 2794
			/* ensure test happens before caller kfree */
			smp_mb__before_atomic_inc(); /* ^^^ */
			atomic_long_inc(&rsp->expedited_workdone1);
2795 2796
			return;
		}
2797 2798

		/* No joy, try again later.  Or just synchronize_sched(). */
2799
		if (trycount++ < 10) {
2800
			udelay(trycount * num_online_cpus());
2801
		} else {
2802
			wait_rcu_gp(call_rcu_sched);
2803
			atomic_long_inc(&rsp->expedited_normal);
2804 2805 2806
			return;
		}

2807
		/* Recheck to see if someone else did our work for us. */
2808
		s = atomic_long_read(&rsp->expedited_done);
2809
		if (ULONG_CMP_GE((ulong)s, (ulong)firstsnap)) {
2810 2811 2812
			/* ensure test happens before caller kfree */
			smp_mb__before_atomic_inc(); /* ^^^ */
			atomic_long_inc(&rsp->expedited_workdone2);
2813 2814 2815 2816 2817
			return;
		}

		/*
		 * Refetching sync_sched_expedited_started allows later
2818 2819 2820 2821
		 * 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.
2822 2823
		 */
		get_online_cpus();
2824
		snap = atomic_long_read(&rsp->expedited_start);
2825 2826
		smp_mb(); /* ensure read is before try_stop_cpus(). */
	}
2827
	atomic_long_inc(&rsp->expedited_stoppedcpus);
2828 2829 2830 2831 2832

	/*
	 * 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
2833
	 * than we did already did their update.
2834 2835
	 */
	do {
2836
		atomic_long_inc(&rsp->expedited_done_tries);
2837
		s = atomic_long_read(&rsp->expedited_done);
2838
		if (ULONG_CMP_GE((ulong)s, (ulong)snap)) {
2839 2840 2841
			/* ensure test happens before caller kfree */
			smp_mb__before_atomic_inc(); /* ^^^ */
			atomic_long_inc(&rsp->expedited_done_lost);
2842 2843
			break;
		}
2844
	} while (atomic_long_cmpxchg(&rsp->expedited_done, s, snap) != s);
2845
	atomic_long_inc(&rsp->expedited_done_exit);
2846 2847 2848 2849 2850

	put_online_cpus();
}
EXPORT_SYMBOL_GPL(synchronize_sched_expedited);

2851 2852 2853 2854 2855 2856 2857 2858 2859
/*
 * 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)
{
2860 2861
	struct rcu_node *rnp = rdp->mynode;

2862 2863 2864 2865 2866
	rdp->n_rcu_pending++;

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

2867 2868 2869 2870
	/* Is this CPU a NO_HZ_FULL CPU that should ignore RCU? */
	if (rcu_nohz_full_cpu(rsp))
		return 0;

2871
	/* Is the RCU core waiting for a quiescent state from this CPU? */
2872 2873
	if (rcu_scheduler_fully_active &&
	    rdp->qs_pending && !rdp->passed_quiesce) {
2874
		rdp->n_rp_qs_pending++;
2875
	} else if (rdp->qs_pending && rdp->passed_quiesce) {
2876
		rdp->n_rp_report_qs++;
2877
		return 1;
2878
	}
2879 2880

	/* Does this CPU have callbacks ready to invoke? */
2881 2882
	if (cpu_has_callbacks_ready_to_invoke(rdp)) {
		rdp->n_rp_cb_ready++;
2883
		return 1;
2884
	}
2885 2886

	/* Has RCU gone idle with this CPU needing another grace period? */
2887 2888
	if (cpu_needs_another_gp(rsp, rdp)) {
		rdp->n_rp_cpu_needs_gp++;
2889
		return 1;
2890
	}
2891 2892

	/* Has another RCU grace period completed?  */
2893
	if (ACCESS_ONCE(rnp->completed) != rdp->completed) { /* outside lock */
2894
		rdp->n_rp_gp_completed++;
2895
		return 1;
2896
	}
2897 2898

	/* Has a new RCU grace period started? */
2899
	if (ACCESS_ONCE(rnp->gpnum) != rdp->gpnum) { /* outside lock */
2900
		rdp->n_rp_gp_started++;
2901
		return 1;
2902
	}
2903

2904 2905 2906 2907 2908 2909
	/* Does this CPU need a deferred NOCB wakeup? */
	if (rcu_nocb_need_deferred_wakeup(rdp)) {
		rdp->n_rp_nocb_defer_wakeup++;
		return 1;
	}

2910
	/* nothing to do */
2911
	rdp->n_rp_need_nothing++;
2912 2913 2914 2915 2916 2917 2918 2919
	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.
 */
2920
static int rcu_pending(int cpu)
2921
{
2922 2923 2924 2925 2926 2927
	struct rcu_state *rsp;

	for_each_rcu_flavor(rsp)
		if (__rcu_pending(rsp, per_cpu_ptr(rsp->rda, cpu)))
			return 1;
	return 0;
2928 2929 2930
}

/*
2931 2932 2933
 * 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.)
2934
 */
2935
static int __maybe_unused rcu_cpu_has_callbacks(int cpu, bool *all_lazy)
2936
{
2937 2938 2939
	bool al = true;
	bool hc = false;
	struct rcu_data *rdp;
2940 2941
	struct rcu_state *rsp;

2942 2943
	for_each_rcu_flavor(rsp) {
		rdp = per_cpu_ptr(rsp->rda, cpu);
2944 2945 2946 2947
		if (!rdp->nxtlist)
			continue;
		hc = true;
		if (rdp->qlen != rdp->qlen_lazy || !all_lazy) {
2948
			al = false;
2949 2950
			break;
		}
2951 2952 2953 2954
	}
	if (all_lazy)
		*all_lazy = al;
	return hc;
2955 2956
}

2957 2958 2959 2960
/*
 * Helper function for _rcu_barrier() tracing.  If tracing is disabled,
 * the compiler is expected to optimize this away.
 */
2961
static void _rcu_barrier_trace(struct rcu_state *rsp, const char *s,
2962 2963 2964 2965 2966 2967
			       int cpu, unsigned long done)
{
	trace_rcu_barrier(rsp->name, s, cpu,
			  atomic_read(&rsp->barrier_cpu_count), done);
}

2968 2969 2970 2971
/*
 * RCU callback function for _rcu_barrier().  If we are last, wake
 * up the task executing _rcu_barrier().
 */
2972
static void rcu_barrier_callback(struct rcu_head *rhp)
2973
{
2974 2975 2976
	struct rcu_data *rdp = container_of(rhp, struct rcu_data, barrier_head);
	struct rcu_state *rsp = rdp->rsp;

2977 2978
	if (atomic_dec_and_test(&rsp->barrier_cpu_count)) {
		_rcu_barrier_trace(rsp, "LastCB", -1, rsp->n_barrier_done);
2979
		complete(&rsp->barrier_completion);
2980 2981 2982
	} else {
		_rcu_barrier_trace(rsp, "CB", -1, rsp->n_barrier_done);
	}
2983 2984 2985 2986 2987 2988 2989
}

/*
 * Called with preemption disabled, and from cross-cpu IRQ context.
 */
static void rcu_barrier_func(void *type)
{
2990
	struct rcu_state *rsp = type;
2991
	struct rcu_data *rdp = __this_cpu_ptr(rsp->rda);
2992

2993
	_rcu_barrier_trace(rsp, "IRQ", -1, rsp->n_barrier_done);
2994
	atomic_inc(&rsp->barrier_cpu_count);
2995
	rsp->call(&rdp->barrier_head, rcu_barrier_callback);
2996 2997 2998 2999 3000 3001
}

/*
 * Orchestrate the specified type of RCU barrier, waiting for all
 * RCU callbacks of the specified type to complete.
 */
3002
static void _rcu_barrier(struct rcu_state *rsp)
3003
{
3004 3005
	int cpu;
	struct rcu_data *rdp;
3006 3007
	unsigned long snap = ACCESS_ONCE(rsp->n_barrier_done);
	unsigned long snap_done;
3008

3009
	_rcu_barrier_trace(rsp, "Begin", -1, snap);
3010

3011
	/* Take mutex to serialize concurrent rcu_barrier() requests. */
3012
	mutex_lock(&rsp->barrier_mutex);
3013

3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025
	/*
	 * 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.
	 */
3026
	snap_done = rsp->n_barrier_done;
3027
	_rcu_barrier_trace(rsp, "Check", -1, snap_done);
3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039

	/*
	 * 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)) {
3040
		_rcu_barrier_trace(rsp, "EarlyExit", -1, snap_done);
3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052
		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.
	 */
	ACCESS_ONCE(rsp->n_barrier_done)++;
	WARN_ON_ONCE((rsp->n_barrier_done & 0x1) != 1);
3053
	_rcu_barrier_trace(rsp, "Inc1", -1, rsp->n_barrier_done);
3054
	smp_mb(); /* Order ->n_barrier_done increment with below mechanism. */
3055

3056
	/*
3057 3058
	 * 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
3059 3060
	 * (or preemption of this task).  Exclude CPU-hotplug operations
	 * to ensure that no offline CPU has callbacks queued.
3061
	 */
3062
	init_completion(&rsp->barrier_completion);
3063
	atomic_set(&rsp->barrier_cpu_count, 1);
3064
	get_online_cpus();
3065 3066

	/*
3067 3068 3069
	 * 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.
3070
	 */
P
Paul E. McKenney 已提交
3071
	for_each_possible_cpu(cpu) {
3072
		if (!cpu_online(cpu) && !rcu_is_nocb_cpu(cpu))
P
Paul E. McKenney 已提交
3073
			continue;
3074
		rdp = per_cpu_ptr(rsp->rda, cpu);
3075
		if (rcu_is_nocb_cpu(cpu)) {
P
Paul E. McKenney 已提交
3076 3077 3078 3079 3080 3081
			_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);
		} else if (ACCESS_ONCE(rdp->qlen)) {
3082 3083
			_rcu_barrier_trace(rsp, "OnlineQ", cpu,
					   rsp->n_barrier_done);
3084
			smp_call_function_single(cpu, rcu_barrier_func, rsp, 1);
3085
		} else {
3086 3087
			_rcu_barrier_trace(rsp, "OnlineNQ", cpu,
					   rsp->n_barrier_done);
3088 3089
		}
	}
3090
	put_online_cpus();
3091 3092 3093 3094 3095

	/*
	 * Now that we have an rcu_barrier_callback() callback on each
	 * CPU, and thus each counted, remove the initial count.
	 */
3096
	if (atomic_dec_and_test(&rsp->barrier_cpu_count))
3097
		complete(&rsp->barrier_completion);
3098

3099 3100 3101 3102
	/* Increment ->n_barrier_done to prevent duplicate work. */
	smp_mb(); /* Keep increment after above mechanism. */
	ACCESS_ONCE(rsp->n_barrier_done)++;
	WARN_ON_ONCE((rsp->n_barrier_done & 0x1) != 0);
3103
	_rcu_barrier_trace(rsp, "Inc2", -1, rsp->n_barrier_done);
3104 3105
	smp_mb(); /* Keep increment before caller's subsequent code. */

3106
	/* Wait for all rcu_barrier_callback() callbacks to be invoked. */
3107
	wait_for_completion(&rsp->barrier_completion);
3108 3109

	/* Other rcu_barrier() invocations can now safely proceed. */
3110
	mutex_unlock(&rsp->barrier_mutex);
3111 3112 3113 3114 3115 3116 3117
}

/**
 * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete.
 */
void rcu_barrier_bh(void)
{
3118
	_rcu_barrier(&rcu_bh_state);
3119 3120 3121 3122 3123 3124 3125 3126
}
EXPORT_SYMBOL_GPL(rcu_barrier_bh);

/**
 * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks.
 */
void rcu_barrier_sched(void)
{
3127
	_rcu_barrier(&rcu_sched_state);
3128 3129 3130
}
EXPORT_SYMBOL_GPL(rcu_barrier_sched);

3131
/*
3132
 * Do boot-time initialization of a CPU's per-CPU RCU data.
3133
 */
3134 3135
static void __init
rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp)
3136 3137
{
	unsigned long flags;
3138
	struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
3139 3140 3141
	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
3142
	raw_spin_lock_irqsave(&rnp->lock, flags);
3143
	rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo);
3144
	init_callback_list(rdp);
3145
	rdp->qlen_lazy = 0;
3146
	ACCESS_ONCE(rdp->qlen) = 0;
3147
	rdp->dynticks = &per_cpu(rcu_dynticks, cpu);
3148
	WARN_ON_ONCE(rdp->dynticks->dynticks_nesting != DYNTICK_TASK_EXIT_IDLE);
3149
	WARN_ON_ONCE(atomic_read(&rdp->dynticks->dynticks) != 1);
3150
	rdp->cpu = cpu;
3151
	rdp->rsp = rsp;
P
Paul E. McKenney 已提交
3152
	rcu_boot_init_nocb_percpu_data(rdp);
P
Paul E. McKenney 已提交
3153
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
3154 3155 3156 3157 3158 3159 3160
}

/*
 * 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.
3161
 */
3162
static void
P
Paul E. McKenney 已提交
3163
rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptible)
3164 3165 3166
{
	unsigned long flags;
	unsigned long mask;
3167
	struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
3168 3169
	struct rcu_node *rnp = rcu_get_root(rsp);

3170 3171 3172
	/* Exclude new grace periods. */
	mutex_lock(&rsp->onoff_mutex);

3173
	/* Set up local state, ensuring consistent view of global state. */
P
Paul E. McKenney 已提交
3174
	raw_spin_lock_irqsave(&rnp->lock, flags);
3175
	rdp->beenonline = 1;	 /* We have now been online. */
P
Paul E. McKenney 已提交
3176
	rdp->preemptible = preemptible;
3177 3178
	rdp->qlen_last_fqs_check = 0;
	rdp->n_force_qs_snap = rsp->n_force_qs;
3179
	rdp->blimit = blimit;
3180
	init_callback_list(rdp);  /* Re-enable callbacks on this CPU. */
3181
	rdp->dynticks->dynticks_nesting = DYNTICK_TASK_EXIT_IDLE;
3182
	rcu_sysidle_init_percpu_data(rdp->dynticks);
3183 3184
	atomic_set(&rdp->dynticks->dynticks,
		   (atomic_read(&rdp->dynticks->dynticks) & ~0x1) + 1);
P
Paul E. McKenney 已提交
3185
	raw_spin_unlock(&rnp->lock);		/* irqs remain disabled. */
3186 3187 3188 3189 3190 3191

	/* 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
3192
		raw_spin_lock(&rnp->lock);	/* irqs already disabled. */
3193 3194
		rnp->qsmaskinit |= mask;
		mask = rnp->grpmask;
3195
		if (rnp == rdp->mynode) {
3196 3197 3198 3199 3200 3201
			/*
			 * 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;
3202
			rdp->completed = rnp->completed;
3203 3204
			rdp->passed_quiesce = 0;
			rdp->qs_pending = 0;
3205
			trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpuonl"));
3206
		}
P
Paul E. McKenney 已提交
3207
		raw_spin_unlock(&rnp->lock); /* irqs already disabled. */
3208 3209
		rnp = rnp->parent;
	} while (rnp != NULL && !(rnp->qsmaskinit & mask));
3210
	local_irq_restore(flags);
3211

3212
	mutex_unlock(&rsp->onoff_mutex);
3213 3214
}

3215
static void rcu_prepare_cpu(int cpu)
3216
{
3217 3218 3219 3220 3221
	struct rcu_state *rsp;

	for_each_rcu_flavor(rsp)
		rcu_init_percpu_data(cpu, rsp,
				     strcmp(rsp->name, "rcu_preempt") == 0);
3222 3223 3224
}

/*
3225
 * Handle CPU online/offline notification events.
3226
 */
3227
static int rcu_cpu_notify(struct notifier_block *self,
3228
				    unsigned long action, void *hcpu)
3229 3230
{
	long cpu = (long)hcpu;
3231
	struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, cpu);
3232
	struct rcu_node *rnp = rdp->mynode;
3233
	struct rcu_state *rsp;
3234

3235
	trace_rcu_utilization(TPS("Start CPU hotplug"));
3236 3237 3238
	switch (action) {
	case CPU_UP_PREPARE:
	case CPU_UP_PREPARE_FROZEN:
P
Peter Zijlstra 已提交
3239 3240
		rcu_prepare_cpu(cpu);
		rcu_prepare_kthreads(cpu);
3241 3242
		break;
	case CPU_ONLINE:
3243
	case CPU_DOWN_FAILED:
T
Thomas Gleixner 已提交
3244
		rcu_boost_kthread_setaffinity(rnp, -1);
3245 3246
		break;
	case CPU_DOWN_PREPARE:
3247
		rcu_boost_kthread_setaffinity(rnp, cpu);
3248
		break;
3249 3250
	case CPU_DYING:
	case CPU_DYING_FROZEN:
3251 3252
		for_each_rcu_flavor(rsp)
			rcu_cleanup_dying_cpu(rsp);
3253
		break;
3254 3255 3256 3257
	case CPU_DEAD:
	case CPU_DEAD_FROZEN:
	case CPU_UP_CANCELED:
	case CPU_UP_CANCELED_FROZEN:
3258 3259
		for_each_rcu_flavor(rsp)
			rcu_cleanup_dead_cpu(cpu, rsp);
3260 3261 3262 3263
		break;
	default:
		break;
	}
3264
	trace_rcu_utilization(TPS("End CPU hotplug"));
3265
	return NOTIFY_OK;
3266 3267
}

3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286
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;
}

3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297
/*
 * Spawn the kthread that handles this RCU flavor's grace periods.
 */
static int __init rcu_spawn_gp_kthread(void)
{
	unsigned long flags;
	struct rcu_node *rnp;
	struct rcu_state *rsp;
	struct task_struct *t;

	for_each_rcu_flavor(rsp) {
3298
		t = kthread_run(rcu_gp_kthread, rsp, "%s", rsp->name);
3299 3300 3301 3302 3303
		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);
P
Paul E. McKenney 已提交
3304
		rcu_spawn_nocb_kthreads(rsp);
3305 3306 3307 3308 3309
	}
	return 0;
}
early_initcall(rcu_spawn_gp_kthread);

3310 3311 3312 3313 3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324
/*
 * 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;
}

3325 3326 3327 3328 3329 3330 3331 3332 3333
/*
 * 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;

3334 3335
	rsp->levelspread[rcu_num_lvls - 1] = rcu_fanout_leaf;
	for (i = rcu_num_lvls - 2; i >= 0; i--)
3336 3337 3338 3339 3340 3341 3342 3343 3344
		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;

3345
	cprv = nr_cpu_ids;
3346
	for (i = rcu_num_lvls - 1; i >= 0; i--) {
3347 3348 3349 3350 3351 3352 3353 3354 3355 3356
		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.
 */
3357 3358
static void __init rcu_init_one(struct rcu_state *rsp,
		struct rcu_data __percpu *rda)
3359
{
3360 3361 3362 3363 3364 3365 3366 3367
	static char *buf[] = { "rcu_node_0",
			       "rcu_node_1",
			       "rcu_node_2",
			       "rcu_node_3" };  /* Match MAX_RCU_LVLS */
	static char *fqs[] = { "rcu_node_fqs_0",
			       "rcu_node_fqs_1",
			       "rcu_node_fqs_2",
			       "rcu_node_fqs_3" };  /* Match MAX_RCU_LVLS */
3368 3369 3370 3371 3372
	int cpustride = 1;
	int i;
	int j;
	struct rcu_node *rnp;

3373 3374
	BUILD_BUG_ON(MAX_RCU_LVLS > ARRAY_SIZE(buf));  /* Fix buf[] init! */

3375 3376 3377 3378
	/* 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");

3379 3380
	/* Initialize the level-tracking arrays. */

3381 3382 3383
	for (i = 0; i < rcu_num_lvls; i++)
		rsp->levelcnt[i] = num_rcu_lvl[i];
	for (i = 1; i < rcu_num_lvls; i++)
3384 3385 3386 3387 3388
		rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1];
	rcu_init_levelspread(rsp);

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

3389
	for (i = rcu_num_lvls - 1; i >= 0; i--) {
3390 3391 3392
		cpustride *= rsp->levelspread[i];
		rnp = rsp->level[i];
		for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) {
P
Paul E. McKenney 已提交
3393
			raw_spin_lock_init(&rnp->lock);
3394 3395
			lockdep_set_class_and_name(&rnp->lock,
						   &rcu_node_class[i], buf[i]);
3396 3397 3398
			raw_spin_lock_init(&rnp->fqslock);
			lockdep_set_class_and_name(&rnp->fqslock,
						   &rcu_fqs_class[i], fqs[i]);
3399 3400
			rnp->gpnum = rsp->gpnum;
			rnp->completed = rsp->completed;
3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417
			rnp->qsmask = 0;
			rnp->qsmaskinit = 0;
			rnp->grplo = j * cpustride;
			rnp->grphi = (j + 1) * cpustride - 1;
			if (rnp->grphi >= NR_CPUS)
				rnp->grphi = NR_CPUS - 1;
			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;
3418
			INIT_LIST_HEAD(&rnp->blkd_tasks);
3419
			rcu_init_one_nocb(rnp);
3420 3421
		}
	}
3422

3423
	rsp->rda = rda;
3424
	init_waitqueue_head(&rsp->gp_wq);
3425
	init_irq_work(&rsp->wakeup_work, rsp_wakeup);
3426
	rnp = rsp->level[rcu_num_lvls - 1];
3427
	for_each_possible_cpu(i) {
3428
		while (i > rnp->grphi)
3429
			rnp++;
3430
		per_cpu_ptr(rsp->rda, i)->mynode = rnp;
3431 3432
		rcu_boot_init_percpu_data(i, rsp);
	}
3433
	list_add(&rsp->flavors, &rcu_struct_flavors);
3434 3435
}

3436 3437
/*
 * Compute the rcu_node tree geometry from kernel parameters.  This cannot
3438
 * replace the definitions in tree.h because those are needed to size
3439 3440 3441 3442
 * the ->node array in the rcu_state structure.
 */
static void __init rcu_init_geometry(void)
{
3443
	ulong d;
3444 3445
	int i;
	int j;
3446
	int n = nr_cpu_ids;
3447 3448
	int rcu_capacity[MAX_RCU_LVLS + 1];

3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459 3460 3461
	/*
	 * 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;

3462
	/* If the compile-time values are accurate, just leave. */
3463 3464
	if (rcu_fanout_leaf == CONFIG_RCU_FANOUT_LEAF &&
	    nr_cpu_ids == NR_CPUS)
3465
		return;
3466 3467
	pr_info("RCU: Adjusting geometry for rcu_fanout_leaf=%d, nr_cpu_ids=%d\n",
		rcu_fanout_leaf, nr_cpu_ids);
3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512

	/*
	 * 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;
}

3513
void __init rcu_init(void)
3514
{
P
Paul E. McKenney 已提交
3515
	int cpu;
3516

3517
	rcu_bootup_announce();
3518
	rcu_init_geometry();
3519
	rcu_init_one(&rcu_bh_state, &rcu_bh_data);
3520
	rcu_init_one(&rcu_sched_state, &rcu_sched_data);
3521
	__rcu_init_preempt();
J
Jiang Fang 已提交
3522
	open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
3523 3524 3525 3526 3527 3528 3529

	/*
	 * 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);
3530
	pm_notifier(rcu_pm_notify, 0);
P
Paul E. McKenney 已提交
3531 3532
	for_each_online_cpu(cpu)
		rcu_cpu_notify(NULL, CPU_UP_PREPARE, (void *)(long)cpu);
3533 3534
}

3535
#include "tree_plugin.h"