tree.c 107.6 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
/*
 * 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
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
 *
 * 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 62 63
#include <trace/events/rcu.h>

#include "rcu.h"
64

65 66 67 68 69 70
MODULE_ALIAS("rcutree");
#ifdef MODULE_PARAM_PREFIX
#undef MODULE_PARAM_PREFIX
#endif
#define MODULE_PARAM_PREFIX "rcutree."

71 72
/* Data structures. */

73
static struct lock_class_key rcu_node_class[RCU_NUM_LVLS];
74
static struct lock_class_key rcu_fqs_class[RCU_NUM_LVLS];
75

76 77 78 79 80 81 82 83
/*
 * 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.
 */
84
#define RCU_STATE_INITIALIZER(sname, sabbr, cr) \
85 86
static char sname##_varname[] = #sname; \
static const char *tp_##sname##_varname __used __tracepoint_string = sname##_varname; \
87
struct rcu_state sname##_state = { \
88
	.level = { &sname##_state.node[0] }, \
89
	.call = cr, \
90
	.fqs_state = RCU_GP_IDLE, \
P
Paul E. McKenney 已提交
91 92
	.gpnum = 0UL - 300UL, \
	.completed = 0UL - 300UL, \
93
	.orphan_lock = __RAW_SPIN_LOCK_UNLOCKED(&sname##_state.orphan_lock), \
94 95
	.orphan_nxttail = &sname##_state.orphan_nxtlist, \
	.orphan_donetail = &sname##_state.orphan_donelist, \
96
	.barrier_mutex = __MUTEX_INITIALIZER(sname##_state.barrier_mutex), \
97
	.onoff_mutex = __MUTEX_INITIALIZER(sname##_state.onoff_mutex), \
98
	.name = sname##_varname, \
99
	.abbr = sabbr, \
100 101
}; \
DEFINE_PER_CPU(struct rcu_data, sname##_data)
102

103 104
RCU_STATE_INITIALIZER(rcu_sched, 's', call_rcu_sched);
RCU_STATE_INITIALIZER(rcu_bh, 'b', call_rcu_bh);
105

106
static struct rcu_state *rcu_state;
107
LIST_HEAD(rcu_struct_flavors);
108

109 110
/* Increase (but not decrease) the CONFIG_RCU_FANOUT_LEAF at boot time. */
static int rcu_fanout_leaf = CONFIG_RCU_FANOUT_LEAF;
111
module_param(rcu_fanout_leaf, int, 0444);
112 113 114 115 116 117 118 119 120 121
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. */

122 123 124 125
/*
 * 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)
126
 * optimize synchronize_sched() to a simple barrier().  When this variable
127 128 129 130
 * 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.
 */
131 132 133
int rcu_scheduler_active __read_mostly;
EXPORT_SYMBOL_GPL(rcu_scheduler_active);

134 135 136 137 138 139 140 141 142 143 144 145 146 147
/*
 * 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;

148 149
#ifdef CONFIG_RCU_BOOST

150 151 152 153 154
/*
 * 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);
155
DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
156
DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
157
DEFINE_PER_CPU(char, rcu_cpu_has_work);
158

159 160
#endif /* #ifdef CONFIG_RCU_BOOST */

T
Thomas Gleixner 已提交
161
static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu);
162 163
static void invoke_rcu_core(void);
static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp);
164

165 166 167 168 169 170 171 172 173 174 175 176
/*
 * 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;

177 178 179 180 181 182 183 184 185 186
/*
 * 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);
}

187
/*
188
 * Note a quiescent state.  Because we do not need to know
189
 * how many quiescent states passed, just if there was at least
190
 * one since the start of the grace period, this just sets a flag.
191
 * The caller must have disabled preemption.
192
 */
193
void rcu_sched_qs(int cpu)
194
{
195
	struct rcu_data *rdp = &per_cpu(rcu_sched_data, cpu);
196

197
	if (rdp->passed_quiesce == 0)
198
		trace_rcu_grace_period(TPS("rcu_sched"), rdp->gpnum, TPS("cpuqs"));
199
	rdp->passed_quiesce = 1;
200 201
}

202
void rcu_bh_qs(int cpu)
203
{
204
	struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu);
205

206
	if (rdp->passed_quiesce == 0)
207
		trace_rcu_grace_period(TPS("rcu_bh"), rdp->gpnum, TPS("cpuqs"));
208
	rdp->passed_quiesce = 1;
209
}
210

211 212 213
/*
 * Note a context switch.  This is a quiescent state for RCU-sched,
 * and requires special handling for preemptible RCU.
214
 * The caller must have disabled preemption.
215 216 217
 */
void rcu_note_context_switch(int cpu)
{
218
	trace_rcu_utilization(TPS("Start context switch"));
219
	rcu_sched_qs(cpu);
220
	rcu_preempt_note_context_switch(cpu);
221
	trace_rcu_utilization(TPS("End context switch"));
222
}
223
EXPORT_SYMBOL_GPL(rcu_note_context_switch);
224

225
static DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = {
226
	.dynticks_nesting = DYNTICK_TASK_EXIT_IDLE,
227
	.dynticks = ATOMIC_INIT(1),
228 229 230 231
#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 */
232
};
233

E
Eric Dumazet 已提交
234 235 236
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. */
237

E
Eric Dumazet 已提交
238 239 240
module_param(blimit, long, 0444);
module_param(qhimark, long, 0444);
module_param(qlowmark, long, 0444);
241

242 243
static ulong jiffies_till_first_fqs = ULONG_MAX;
static ulong jiffies_till_next_fqs = ULONG_MAX;
244 245 246 247

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

248 249
static void rcu_start_gp_advanced(struct rcu_state *rsp, struct rcu_node *rnp,
				  struct rcu_data *rdp);
250 251 252 253
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);
254
static void force_quiescent_state(struct rcu_state *rsp);
255
static int rcu_pending(int cpu);
256 257

/*
258
 * Return the number of RCU-sched batches processed thus far for debug & stats.
259
 */
260
long rcu_batches_completed_sched(void)
261
{
262
	return rcu_sched_state.completed;
263
}
264
EXPORT_SYMBOL_GPL(rcu_batches_completed_sched);
265 266 267 268 269 270 271 272 273 274

/*
 * 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);

275 276 277 278 279
/*
 * Force a quiescent state for RCU BH.
 */
void rcu_bh_force_quiescent_state(void)
{
280
	force_quiescent_state(&rcu_bh_state);
281 282 283
}
EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state);

284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308
/*
 * 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);

309 310 311 312 313
/*
 * Force a quiescent state for RCU-sched.
 */
void rcu_sched_force_quiescent_state(void)
{
314
	force_quiescent_state(&rcu_sched_state);
315 316 317
}
EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state);

318 319 320 321 322 323
/*
 * 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
324 325
	return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL] &&
	       rdp->nxttail[RCU_DONE_TAIL] != NULL;
326 327 328
}

/*
329 330 331
 * Does the current CPU require a not-yet-started grace period?
 * The caller must have disabled interrupts to prevent races with
 * normal callback registry.
332 333 334 335
 */
static int
cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp)
{
336
	int i;
P
Paul E. McKenney 已提交
337

338 339
	if (rcu_gp_in_progress(rsp))
		return 0;  /* No, a grace period is already in progress. */
340
	if (rcu_nocb_needs_gp(rsp))
341
		return 1;  /* Yes, a no-CBs CPU needs one. */
342 343 344 345 346 347 348 349 350 351
	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. */
352 353 354 355 356 357 358 359 360 361
}

/*
 * 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];
}

362
/*
363
 * rcu_eqs_enter_common - current CPU is moving towards extended quiescent state
364 365 366 367 368
 *
 * 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.
 */
369 370
static void rcu_eqs_enter_common(struct rcu_dynticks *rdtp, long long oldval,
				bool user)
371
{
372 373 374
	struct rcu_state *rsp;
	struct rcu_data *rdp;

375
	trace_rcu_dyntick(TPS("Start"), oldval, rdtp->dynticks_nesting);
376
	if (!user && !is_idle_task(current)) {
377 378
		struct task_struct *idle __maybe_unused =
			idle_task(smp_processor_id());
379

380
		trace_rcu_dyntick(TPS("Error on entry: not idle task"), oldval, 0);
381
		ftrace_dump(DUMP_ORIG);
382 383 384
		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! */
385
	}
386 387 388 389
	for_each_rcu_flavor(rsp) {
		rdp = this_cpu_ptr(rsp->rda);
		do_nocb_deferred_wakeup(rdp);
	}
390
	rcu_prepare_for_idle(smp_processor_id());
391 392 393 394 395
	/* 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);
396 397

	/*
398
	 * It is illegal to enter an extended quiescent state while
399 400 401 402 403 404 405 406
	 * 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.");
407
}
408

409 410 411
/*
 * Enter an RCU extended quiescent state, which can be either the
 * idle loop or adaptive-tickless usermode execution.
412
 */
413
static void rcu_eqs_enter(bool user)
414
{
415
	long long oldval;
416 417
	struct rcu_dynticks *rdtp;

418
	rdtp = this_cpu_ptr(&rcu_dynticks);
419
	oldval = rdtp->dynticks_nesting;
420
	WARN_ON_ONCE((oldval & DYNTICK_TASK_NEST_MASK) == 0);
421
	if ((oldval & DYNTICK_TASK_NEST_MASK) == DYNTICK_TASK_NEST_VALUE) {
422
		rdtp->dynticks_nesting = 0;
423 424
		rcu_eqs_enter_common(rdtp, oldval, user);
	} else {
425
		rdtp->dynticks_nesting -= DYNTICK_TASK_NEST_VALUE;
426
	}
427
}
428 429 430 431 432 433 434 435 436 437 438 439 440 441 442

/**
 * 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)
{
443 444 445
	unsigned long flags;

	local_irq_save(flags);
446
	rcu_eqs_enter(false);
447
	rcu_sysidle_enter(this_cpu_ptr(&rcu_dynticks), 0);
448
	local_irq_restore(flags);
449
}
450
EXPORT_SYMBOL_GPL(rcu_idle_enter);
451

452
#ifdef CONFIG_RCU_USER_QS
453 454 455 456 457 458 459 460 461 462
/**
 * 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)
{
463
	rcu_eqs_enter(1);
464
}
465
#endif /* CONFIG_RCU_USER_QS */
466

467 468 469 470 471 472
/**
 * 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.
473
 *
474 475 476 477 478 479 480 481
 * 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.
482
 */
483
void rcu_irq_exit(void)
484 485
{
	unsigned long flags;
486
	long long oldval;
487 488 489
	struct rcu_dynticks *rdtp;

	local_irq_save(flags);
490
	rdtp = this_cpu_ptr(&rcu_dynticks);
491
	oldval = rdtp->dynticks_nesting;
492 493
	rdtp->dynticks_nesting--;
	WARN_ON_ONCE(rdtp->dynticks_nesting < 0);
494
	if (rdtp->dynticks_nesting)
495
		trace_rcu_dyntick(TPS("--="), oldval, rdtp->dynticks_nesting);
496
	else
497
		rcu_eqs_enter_common(rdtp, oldval, true);
498
	rcu_sysidle_enter(rdtp, 1);
499 500 501 502
	local_irq_restore(flags);
}

/*
503
 * rcu_eqs_exit_common - current CPU moving away from extended quiescent state
504 505 506 507 508
 *
 * 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.
 */
509 510
static void rcu_eqs_exit_common(struct rcu_dynticks *rdtp, long long oldval,
			       int user)
511
{
512 513 514 515 516
	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));
517
	rcu_cleanup_after_idle(smp_processor_id());
518
	trace_rcu_dyntick(TPS("End"), oldval, rdtp->dynticks_nesting);
519
	if (!user && !is_idle_task(current)) {
520 521
		struct task_struct *idle __maybe_unused =
			idle_task(smp_processor_id());
522

523
		trace_rcu_dyntick(TPS("Error on exit: not idle task"),
524
				  oldval, rdtp->dynticks_nesting);
525
		ftrace_dump(DUMP_ORIG);
526 527 528
		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! */
529 530 531
	}
}

532 533 534
/*
 * Exit an RCU extended quiescent state, which can be either the
 * idle loop or adaptive-tickless usermode execution.
535
 */
536
static void rcu_eqs_exit(bool user)
537 538 539 540
{
	struct rcu_dynticks *rdtp;
	long long oldval;

541
	rdtp = this_cpu_ptr(&rcu_dynticks);
542
	oldval = rdtp->dynticks_nesting;
543
	WARN_ON_ONCE(oldval < 0);
544
	if (oldval & DYNTICK_TASK_NEST_MASK) {
545
		rdtp->dynticks_nesting += DYNTICK_TASK_NEST_VALUE;
546
	} else {
547
		rdtp->dynticks_nesting = DYNTICK_TASK_EXIT_IDLE;
548 549
		rcu_eqs_exit_common(rdtp, oldval, user);
	}
550
}
551 552 553 554 555 556 557 558 559 560 561 562 563 564

/**
 * 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)
{
565 566 567
	unsigned long flags;

	local_irq_save(flags);
568
	rcu_eqs_exit(false);
569
	rcu_sysidle_exit(this_cpu_ptr(&rcu_dynticks), 0);
570
	local_irq_restore(flags);
571
}
572
EXPORT_SYMBOL_GPL(rcu_idle_exit);
573

574
#ifdef CONFIG_RCU_USER_QS
575 576 577 578 579 580 581 582
/**
 * 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)
{
583
	rcu_eqs_exit(1);
584
}
585
#endif /* CONFIG_RCU_USER_QS */
586

587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612
/**
 * 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);
613
	rdtp = this_cpu_ptr(&rcu_dynticks);
614 615 616
	oldval = rdtp->dynticks_nesting;
	rdtp->dynticks_nesting++;
	WARN_ON_ONCE(rdtp->dynticks_nesting == 0);
617
	if (oldval)
618
		trace_rcu_dyntick(TPS("++="), oldval, rdtp->dynticks_nesting);
619
	else
620
		rcu_eqs_exit_common(rdtp, oldval, true);
621
	rcu_sysidle_exit(rdtp, 1);
622 623 624 625 626 627 628 629 630 631 632 633
	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)
{
634
	struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
635

636 637
	if (rdtp->dynticks_nmi_nesting == 0 &&
	    (atomic_read(&rdtp->dynticks) & 0x1))
638
		return;
639 640 641 642 643 644
	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));
645 646 647 648 649 650 651 652 653 654 655
}

/**
 * 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)
{
656
	struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
657

658 659
	if (rdtp->dynticks_nmi_nesting == 0 ||
	    --rdtp->dynticks_nmi_nesting != 0)
660
		return;
661 662 663 664 665
	/* 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);
666 667 668
}

/**
669 670 671 672 673 674 675
 * __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.
 */
676
bool notrace __rcu_is_watching(void)
677 678 679 680 681 682
{
	return atomic_read(this_cpu_ptr(&rcu_dynticks.dynticks)) & 0x1;
}

/**
 * rcu_is_watching - see if RCU thinks that the current CPU is idle
683
 *
684
 * If the current CPU is in its idle loop and is neither in an interrupt
685
 * or NMI handler, return true.
686
 */
687
bool notrace rcu_is_watching(void)
688
{
689 690 691
	int ret;

	preempt_disable();
692
	ret = __rcu_is_watching();
693 694
	preempt_enable();
	return ret;
695
}
696
EXPORT_SYMBOL_GPL(rcu_is_watching);
697

698
#if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU)
699 700 701 702 703 704 705

/*
 * 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
706 707 708 709 710 711 712 713 714 715 716
 * 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.
717 718 719 720 721 722
 *
 * Disable checking if in an NMI handler because we cannot safely report
 * errors from NMI handlers anyway.
 */
bool rcu_lockdep_current_cpu_online(void)
{
723 724
	struct rcu_data *rdp;
	struct rcu_node *rnp;
725 726 727 728 729
	bool ret;

	if (in_nmi())
		return 1;
	preempt_disable();
730
	rdp = this_cpu_ptr(&rcu_sched_data);
731 732
	rnp = rdp->mynode;
	ret = (rdp->grpmask & rnp->qsmaskinit) ||
733 734 735 736 737 738
	      !rcu_scheduler_fully_active;
	preempt_enable();
	return ret;
}
EXPORT_SYMBOL_GPL(rcu_lockdep_current_cpu_online);

739
#endif /* #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) */
740

741
/**
742
 * rcu_is_cpu_rrupt_from_idle - see if idle or immediately interrupted from idle
743
 *
744 745 746
 * 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.
747
 */
748
static int rcu_is_cpu_rrupt_from_idle(void)
749
{
750
	return __this_cpu_read(rcu_dynticks.dynticks_nesting) <= 1;
751 752 753 754 755
}

/*
 * Snapshot the specified CPU's dynticks counter so that we can later
 * credit them with an implicit quiescent state.  Return 1 if this CPU
756
 * is in dynticks idle mode, which is an extended quiescent state.
757
 */
758 759
static int dyntick_save_progress_counter(struct rcu_data *rdp,
					 bool *isidle, unsigned long *maxj)
760
{
761
	rdp->dynticks_snap = atomic_add_return(0, &rdp->dynticks->dynticks);
762
	rcu_sysidle_check_cpu(rdp, isidle, maxj);
763
	return (rdp->dynticks_snap & 0x1) == 0;
764 765
}

766 767 768 769 770 771
/*
 * 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);

772 773 774 775
/*
 * 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()
776
 * for this same CPU, or by virtue of having been offline.
777
 */
778 779
static int rcu_implicit_dynticks_qs(struct rcu_data *rdp,
				    bool *isidle, unsigned long *maxj)
780
{
781 782
	unsigned int curr;
	unsigned int snap;
783

784 785
	curr = (unsigned int)atomic_add_return(0, &rdp->dynticks->dynticks);
	snap = (unsigned int)rdp->dynticks_snap;
786 787 788 789 790 791 792 793 794

	/*
	 * 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.
	 */
795
	if ((curr & 0x1) == 0 || UINT_CMP_GE(curr, snap + 2)) {
796
		trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("dti"));
797 798 799 800
		rdp->dynticks_fqs++;
		return 1;
	}

801 802 803 804 805 806 807 808 809 810 811 812 813 814 815
	/*
	 * 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)) {
816
		trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("ofl"));
817 818 819
		rdp->offline_fqs++;
		return 1;
	}
820 821 822 823 824 825 826 827 828 829

	/*
	 * 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);

830 831 832 833 834 835 836 837 838 839 840 841 842 843 844
	/*
	 * 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 &&
	    ULONG_CMP_GE(ACCESS_ONCE(jiffies), rdp->rsp->jiffies_resched)) {
		rdp->rsp->jiffies_resched += 5;
		resched_cpu(rdp->cpu);
	}

845
	return 0;
846 847 848 849
}

static void record_gp_stall_check_time(struct rcu_state *rsp)
{
850
	unsigned long j = ACCESS_ONCE(jiffies);
851
	unsigned long j1;
852 853 854

	rsp->gp_start = j;
	smp_wmb(); /* Record start time before stall time. */
855 856 857
	j1 = rcu_jiffies_till_stall_check();
	rsp->jiffies_stall = j + j1;
	rsp->jiffies_resched = j + j1 / 2;
858 859
}

860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882
/*
 * 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);
	}
}

883 884 885 886 887
static void print_other_cpu_stall(struct rcu_state *rsp)
{
	int cpu;
	long delta;
	unsigned long flags;
888
	int ndetected = 0;
889
	struct rcu_node *rnp = rcu_get_root(rsp);
890
	long totqlen = 0;
891 892 893

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

P
Paul E. McKenney 已提交
894
	raw_spin_lock_irqsave(&rnp->lock, flags);
895
	delta = jiffies - rsp->jiffies_stall;
896
	if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) {
P
Paul E. McKenney 已提交
897
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
898 899
		return;
	}
900
	rsp->jiffies_stall = jiffies + 3 * rcu_jiffies_till_stall_check() + 3;
P
Paul E. McKenney 已提交
901
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
902

903 904 905 906 907
	/*
	 * OK, time to rat on our buddy...
	 * See Documentation/RCU/stallwarn.txt for info on how to debug
	 * RCU CPU stall warnings.
	 */
908
	pr_err("INFO: %s detected stalls on CPUs/tasks:",
909
	       rsp->name);
910
	print_cpu_stall_info_begin();
911
	rcu_for_each_leaf_node(rsp, rnp) {
912
		raw_spin_lock_irqsave(&rnp->lock, flags);
913
		ndetected += rcu_print_task_stall(rnp);
914 915 916 917 918 919 920 921
		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++;
				}
		}
922
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
923
	}
924 925 926 927 928 929 930

	/*
	 * 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);
931
	ndetected += rcu_print_task_stall(rnp);
932 933 934
	raw_spin_unlock_irqrestore(&rnp->lock, flags);

	print_cpu_stall_info_end();
935 936 937
	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",
938
	       smp_processor_id(), (long)(jiffies - rsp->gp_start),
939
	       rsp->gpnum, rsp->completed, totqlen);
940
	if (ndetected == 0)
941
		pr_err("INFO: Stall ended before state dump start\n");
942
	else if (!trigger_all_cpu_backtrace())
943
		rcu_dump_cpu_stacks(rsp);
944

945
	/* Complain about tasks blocking the grace period. */
946 947 948

	rcu_print_detail_task_stall(rsp);

949
	force_quiescent_state(rsp);  /* Kick them all. */
950 951
}

952 953 954 955 956 957
/*
 * 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);

958 959
static void print_cpu_stall(struct rcu_state *rsp)
{
960
	int cpu;
961 962
	unsigned long flags;
	struct rcu_node *rnp = rcu_get_root(rsp);
963
	long totqlen = 0;
964

965 966 967 968 969
	/*
	 * OK, time to rat on ourselves...
	 * See Documentation/RCU/stallwarn.txt for info on how to debug
	 * RCU CPU stall warnings.
	 */
970
	pr_err("INFO: %s self-detected stall on CPU", rsp->name);
971 972 973
	print_cpu_stall_info_begin();
	print_cpu_stall_info(rsp, smp_processor_id());
	print_cpu_stall_info_end();
974 975 976 977
	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);
978 979
	if (!trigger_all_cpu_backtrace())
		dump_stack();
980

P
Paul E. McKenney 已提交
981
	raw_spin_lock_irqsave(&rnp->lock, flags);
982
	if (ULONG_CMP_GE(jiffies, rsp->jiffies_stall))
983
		rsp->jiffies_stall = jiffies +
984
				     3 * rcu_jiffies_till_stall_check() + 3;
P
Paul E. McKenney 已提交
985
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
986

987 988 989 990 991 992 993 994
	/*
	 * 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());
995 996 997 998
}

static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
{
999 1000 1001
	unsigned long completed;
	unsigned long gpnum;
	unsigned long gps;
1002 1003
	unsigned long j;
	unsigned long js;
1004 1005
	struct rcu_node *rnp;

1006
	if (rcu_cpu_stall_suppress || !rcu_gp_in_progress(rsp))
1007
		return;
1008
	j = ACCESS_ONCE(jiffies);
1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028

	/*
	 * 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... */
1029
	js = ACCESS_ONCE(rsp->jiffies_stall);
1030 1031 1032 1033 1034 1035 1036 1037
	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. */
1038
	rnp = rdp->mynode;
1039
	if (rcu_gp_in_progress(rsp) &&
1040
	    (ACCESS_ONCE(rnp->qsmask) & rdp->grpmask)) {
1041 1042 1043 1044

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

1045 1046
	} else if (rcu_gp_in_progress(rsp) &&
		   ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY)) {
1047

1048
		/* They had a few time units to dump stack, so complain. */
1049 1050 1051 1052
		print_other_cpu_stall(rsp);
	}
}

1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063
/**
 * 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)
{
1064 1065 1066 1067
	struct rcu_state *rsp;

	for_each_rcu_flavor(rsp)
		rsp->jiffies_stall = jiffies + ULONG_MAX / 2;
1068 1069
}

1070 1071 1072 1073 1074 1075 1076
/*
 * Initialize the specified rcu_data structure's callback list to empty.
 */
static void init_callback_list(struct rcu_data *rdp)
{
	int i;

1077 1078
	if (init_nocb_callback_list(rdp))
		return;
1079 1080 1081 1082 1083
	rdp->nxtlist = NULL;
	for (i = 0; i < RCU_NEXT_SIZE; i++)
		rdp->nxttail[i] = &rdp->nxtlist;
}

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 1111 1112
/*
 * 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;
}

1113 1114 1115 1116 1117
/*
 * 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,
1118
				unsigned long c, const char *s)
1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143
{
	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);
1144
	trace_rcu_future_gp(rnp, rdp, c, TPS("Startleaf"));
1145
	if (rnp->need_future_gp[c & 0x1]) {
1146
		trace_rcu_future_gp(rnp, rdp, c, TPS("Prestartleaf"));
1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159
		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]++;
1160
		trace_rcu_future_gp(rnp, rdp, c, TPS("Startedleaf"));
1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187
		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).
	 */
	if (rnp != rnp_root)
		raw_spin_lock(&rnp_root->lock);

	/*
	 * 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
	__note_gp_changes(rsp, rnp, rdp);
1391 1392 1393
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
}

1394
/*
1395
 * Initialize a new grace period.  Return 0 if no grace period required.
1396
 */
1397
static int rcu_gp_init(struct rcu_state *rsp)
1398 1399
{
	struct rcu_data *rdp;
1400
	struct rcu_node *rnp = rcu_get_root(rsp);
1401

1402
	rcu_bind_gp_kthread();
1403
	raw_spin_lock_irq(&rnp->lock);
1404 1405 1406 1407 1408
	if (rsp->gp_flags == 0) {
		/* Spurious wakeup, tell caller to go back to sleep.  */
		raw_spin_unlock_irq(&rnp->lock);
		return 0;
	}
1409
	rsp->gp_flags = 0; /* Clear all flags: New grace period. */
1410

1411 1412 1413 1414 1415
	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.
		 */
1416 1417 1418 1419 1420
		raw_spin_unlock_irq(&rnp->lock);
		return 0;
	}

	/* Advance to a new grace period and initialize state. */
1421 1422
	record_gp_stall_check_time(rsp);
	smp_wmb(); /* Record GP times before starting GP. */
1423
	rsp->gpnum++;
1424
	trace_rcu_grace_period(rsp->name, rsp->gpnum, TPS("start"));
1425 1426 1427
	raw_spin_unlock_irq(&rnp->lock);

	/* Exclude any concurrent CPU-hotplug operations. */
1428
	mutex_lock(&rsp->onoff_mutex);
1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443

	/*
	 * 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) {
1444 1445
		raw_spin_lock_irq(&rnp->lock);
		rdp = this_cpu_ptr(rsp->rda);
1446 1447
		rcu_preempt_check_blocked_tasks(rnp);
		rnp->qsmask = rnp->qsmaskinit;
1448
		ACCESS_ONCE(rnp->gpnum) = rsp->gpnum;
1449
		WARN_ON_ONCE(rnp->completed != rsp->completed);
1450
		ACCESS_ONCE(rnp->completed) = rsp->completed;
1451
		if (rnp == rdp->mynode)
1452
			__note_gp_changes(rsp, rnp, rdp);
1453 1454 1455 1456 1457
		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);
1458
#ifdef CONFIG_PROVE_RCU_DELAY
1459
		if ((prandom_u32() % (rcu_num_nodes + 1)) == 0 &&
1460
		    system_state == SYSTEM_RUNNING)
1461
			udelay(200);
1462
#endif /* #ifdef CONFIG_PROVE_RCU_DELAY */
1463 1464
		cond_resched();
	}
1465

1466
	mutex_unlock(&rsp->onoff_mutex);
1467 1468
	return 1;
}
1469

1470 1471 1472
/*
 * Do one round of quiescent-state forcing.
 */
1473
static int rcu_gp_fqs(struct rcu_state *rsp, int fqs_state_in)
1474 1475
{
	int fqs_state = fqs_state_in;
1476 1477
	bool isidle = false;
	unsigned long maxj;
1478 1479 1480 1481 1482
	struct rcu_node *rnp = rcu_get_root(rsp);

	rsp->n_force_qs++;
	if (fqs_state == RCU_SAVE_DYNTICK) {
		/* Collect dyntick-idle snapshots. */
1483 1484 1485 1486
		if (is_sysidle_rcu_state(rsp)) {
			isidle = 1;
			maxj = jiffies - ULONG_MAX / 4;
		}
1487 1488
		force_qs_rnp(rsp, dyntick_save_progress_counter,
			     &isidle, &maxj);
1489
		rcu_sysidle_report_gp(rsp, isidle, maxj);
1490 1491 1492
		fqs_state = RCU_FORCE_QS;
	} else {
		/* Handle dyntick-idle and offline CPUs. */
1493
		isidle = 0;
1494
		force_qs_rnp(rsp, rcu_implicit_dynticks_qs, &isidle, &maxj);
1495 1496 1497 1498 1499 1500 1501 1502 1503 1504
	}
	/* Clear flag to prevent immediate re-entry. */
	if (ACCESS_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) {
		raw_spin_lock_irq(&rnp->lock);
		rsp->gp_flags &= ~RCU_GP_FLAG_FQS;
		raw_spin_unlock_irq(&rnp->lock);
	}
	return fqs_state;
}

1505 1506 1507
/*
 * Clean up after the old grace period.
 */
1508
static void rcu_gp_cleanup(struct rcu_state *rsp)
1509 1510
{
	unsigned long gp_duration;
1511
	int nocb = 0;
1512 1513
	struct rcu_data *rdp;
	struct rcu_node *rnp = rcu_get_root(rsp);
1514

1515 1516 1517 1518
	raw_spin_lock_irq(&rnp->lock);
	gp_duration = jiffies - rsp->gp_start;
	if (gp_duration > rsp->gp_max)
		rsp->gp_max = gp_duration;
1519

1520 1521 1522 1523 1524 1525 1526 1527
	/*
	 * 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.
	 */
1528
	raw_spin_unlock_irq(&rnp->lock);
1529

1530 1531 1532 1533 1534 1535 1536 1537 1538 1539
	/*
	 * 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) {
1540
		raw_spin_lock_irq(&rnp->lock);
1541
		ACCESS_ONCE(rnp->completed) = rsp->gpnum;
1542 1543
		rdp = this_cpu_ptr(rsp->rda);
		if (rnp == rdp->mynode)
1544
			__note_gp_changes(rsp, rnp, rdp);
1545
		/* smp_mb() provided by prior unlock-lock pair. */
1546
		nocb += rcu_future_gp_cleanup(rsp, rnp);
1547 1548
		raw_spin_unlock_irq(&rnp->lock);
		cond_resched();
1549
	}
1550 1551
	rnp = rcu_get_root(rsp);
	raw_spin_lock_irq(&rnp->lock);
1552
	rcu_nocb_gp_set(rnp, nocb);
1553 1554

	rsp->completed = rsp->gpnum; /* Declare grace period done. */
1555
	trace_rcu_grace_period(rsp->name, rsp->completed, TPS("end"));
1556
	rsp->fqs_state = RCU_GP_IDLE;
1557
	rdp = this_cpu_ptr(rsp->rda);
1558
	rcu_advance_cbs(rsp, rnp, rdp);  /* Reduce false positives below. */
1559
	if (cpu_needs_another_gp(rsp, rdp)) {
1560
		rsp->gp_flags = RCU_GP_FLAG_INIT;
1561 1562 1563 1564
		trace_rcu_grace_period(rsp->name,
				       ACCESS_ONCE(rsp->gpnum),
				       TPS("newreq"));
	}
1565 1566 1567 1568 1569 1570 1571 1572
	raw_spin_unlock_irq(&rnp->lock);
}

/*
 * Body of kthread that handles grace periods.
 */
static int __noreturn rcu_gp_kthread(void *arg)
{
1573
	int fqs_state;
1574
	int gf;
1575
	unsigned long j;
1576
	int ret;
1577 1578 1579 1580 1581 1582 1583
	struct rcu_state *rsp = arg;
	struct rcu_node *rnp = rcu_get_root(rsp);

	for (;;) {

		/* Handle grace-period start. */
		for (;;) {
1584 1585 1586
			trace_rcu_grace_period(rsp->name,
					       ACCESS_ONCE(rsp->gpnum),
					       TPS("reqwait"));
1587
			wait_event_interruptible(rsp->gp_wq,
1588
						 ACCESS_ONCE(rsp->gp_flags) &
1589
						 RCU_GP_FLAG_INIT);
1590
			/* Locking provides needed memory barrier. */
1591
			if (rcu_gp_init(rsp))
1592 1593 1594
				break;
			cond_resched();
			flush_signals(current);
1595 1596 1597
			trace_rcu_grace_period(rsp->name,
					       ACCESS_ONCE(rsp->gpnum),
					       TPS("reqwaitsig"));
1598
		}
1599

1600 1601
		/* Handle quiescent-state forcing. */
		fqs_state = RCU_SAVE_DYNTICK;
1602 1603 1604 1605 1606
		j = jiffies_till_first_fqs;
		if (j > HZ) {
			j = HZ;
			jiffies_till_first_fqs = HZ;
		}
1607
		ret = 0;
1608
		for (;;) {
1609 1610
			if (!ret)
				rsp->jiffies_force_qs = jiffies + j;
1611 1612 1613
			trace_rcu_grace_period(rsp->name,
					       ACCESS_ONCE(rsp->gpnum),
					       TPS("fqswait"));
1614
			ret = wait_event_interruptible_timeout(rsp->gp_wq,
1615 1616
					((gf = ACCESS_ONCE(rsp->gp_flags)) &
					 RCU_GP_FLAG_FQS) ||
1617 1618
					(!ACCESS_ONCE(rnp->qsmask) &&
					 !rcu_preempt_blocked_readers_cgp(rnp)),
1619
					j);
1620
			/* Locking provides needed memory barriers. */
1621
			/* If grace period done, leave loop. */
1622
			if (!ACCESS_ONCE(rnp->qsmask) &&
1623
			    !rcu_preempt_blocked_readers_cgp(rnp))
1624
				break;
1625
			/* If time for quiescent-state forcing, do it. */
1626 1627
			if (ULONG_CMP_GE(jiffies, rsp->jiffies_force_qs) ||
			    (gf & RCU_GP_FLAG_FQS)) {
1628 1629 1630
				trace_rcu_grace_period(rsp->name,
						       ACCESS_ONCE(rsp->gpnum),
						       TPS("fqsstart"));
1631
				fqs_state = rcu_gp_fqs(rsp, fqs_state);
1632 1633 1634
				trace_rcu_grace_period(rsp->name,
						       ACCESS_ONCE(rsp->gpnum),
						       TPS("fqsend"));
1635 1636 1637 1638 1639
				cond_resched();
			} else {
				/* Deal with stray signal. */
				cond_resched();
				flush_signals(current);
1640 1641 1642
				trace_rcu_grace_period(rsp->name,
						       ACCESS_ONCE(rsp->gpnum),
						       TPS("fqswaitsig"));
1643
			}
1644 1645 1646 1647 1648 1649 1650 1651
			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;
			}
1652
		}
1653 1654 1655

		/* Handle grace-period end. */
		rcu_gp_cleanup(rsp);
1656 1657 1658
	}
}

1659 1660 1661 1662 1663 1664 1665 1666
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);
}

1667 1668 1669
/*
 * Start a new RCU grace period if warranted, re-initializing the hierarchy
 * in preparation for detecting the next grace period.  The caller must hold
1670
 * the root node's ->lock and hard irqs must be disabled.
1671 1672 1673 1674
 *
 * 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.
1675 1676
 */
static void
1677 1678
rcu_start_gp_advanced(struct rcu_state *rsp, struct rcu_node *rnp,
		      struct rcu_data *rdp)
1679
{
1680
	if (!rsp->gp_kthread || !cpu_needs_another_gp(rsp, rdp)) {
1681
		/*
1682
		 * Either we have not yet spawned the grace-period
1683 1684
		 * task, this CPU does not need another grace period,
		 * or a grace period is already in progress.
1685
		 * Either way, don't start a new grace period.
1686 1687 1688
		 */
		return;
	}
1689
	rsp->gp_flags = RCU_GP_FLAG_INIT;
1690 1691
	trace_rcu_grace_period(rsp->name, ACCESS_ONCE(rsp->gpnum),
			       TPS("newreq"));
1692

1693 1694
	/*
	 * We can't do wakeups while holding the rnp->lock, as that
1695 1696 1697
	 * could cause possible deadlocks with the rq->lock. Defer
	 * the wakeup to interrupt context.  And don't bother waking
	 * up the running kthread.
1698
	 */
1699 1700
	if (current != rsp->gp_kthread)
		irq_work_queue(&rsp->wakeup_work);
1701 1702
}

1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727
/*
 * 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);
}

1728
/*
P
Paul E. McKenney 已提交
1729 1730 1731
 * 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
1732 1733
 * if one is needed.  Note that the caller must hold rnp->lock, which
 * is released before return.
1734
 */
P
Paul E. McKenney 已提交
1735
static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags)
1736
	__releases(rcu_get_root(rsp)->lock)
1737
{
1738
	WARN_ON_ONCE(!rcu_gp_in_progress(rsp));
1739 1740
	raw_spin_unlock_irqrestore(&rcu_get_root(rsp)->lock, flags);
	wake_up(&rsp->gp_wq);  /* Memory barrier implied by wake_up() path. */
1741 1742
}

1743
/*
P
Paul E. McKenney 已提交
1744 1745 1746 1747 1748 1749
 * 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.
1750 1751
 */
static void
P
Paul E. McKenney 已提交
1752 1753
rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp,
		  struct rcu_node *rnp, unsigned long flags)
1754 1755
	__releases(rnp->lock)
{
1756 1757
	struct rcu_node *rnp_c;

1758 1759 1760 1761 1762
	/* 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
1763
			raw_spin_unlock_irqrestore(&rnp->lock, flags);
1764 1765 1766
			return;
		}
		rnp->qsmask &= ~mask;
1767 1768 1769 1770
		trace_rcu_quiescent_state_report(rsp->name, rnp->gpnum,
						 mask, rnp->qsmask, rnp->level,
						 rnp->grplo, rnp->grphi,
						 !!rnp->gp_tasks);
1771
		if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) {
1772 1773

			/* Other bits still set at this level, so done. */
P
Paul E. McKenney 已提交
1774
			raw_spin_unlock_irqrestore(&rnp->lock, flags);
1775 1776 1777 1778 1779 1780 1781 1782 1783
			return;
		}
		mask = rnp->grpmask;
		if (rnp->parent == NULL) {

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

			break;
		}
P
Paul E. McKenney 已提交
1784
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
1785
		rnp_c = rnp;
1786
		rnp = rnp->parent;
P
Paul E. McKenney 已提交
1787
		raw_spin_lock_irqsave(&rnp->lock, flags);
1788
		WARN_ON_ONCE(rnp_c->qsmask);
1789 1790 1791 1792
	}

	/*
	 * Get here if we are the last CPU to pass through a quiescent
P
Paul E. McKenney 已提交
1793
	 * state for this grace period.  Invoke rcu_report_qs_rsp()
1794
	 * to clean up and start the next grace period if one is needed.
1795
	 */
P
Paul E. McKenney 已提交
1796
	rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */
1797 1798 1799
}

/*
P
Paul E. McKenney 已提交
1800 1801 1802 1803 1804 1805 1806
 * 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!
1807 1808
 */
static void
1809
rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp)
1810 1811 1812 1813 1814 1815
{
	unsigned long flags;
	unsigned long mask;
	struct rcu_node *rnp;

	rnp = rdp->mynode;
P
Paul E. McKenney 已提交
1816
	raw_spin_lock_irqsave(&rnp->lock, flags);
1817 1818
	if (rdp->passed_quiesce == 0 || rdp->gpnum != rnp->gpnum ||
	    rnp->completed == rnp->gpnum) {
1819 1820

		/*
1821 1822 1823 1824
		 * 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.
1825
		 */
1826
		rdp->passed_quiesce = 0;	/* need qs for new gp. */
P
Paul E. McKenney 已提交
1827
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
1828 1829 1830 1831
		return;
	}
	mask = rdp->grpmask;
	if ((rnp->qsmask & mask) == 0) {
P
Paul E. McKenney 已提交
1832
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
1833 1834 1835 1836 1837 1838 1839
	} 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.
		 */
1840
		rcu_accelerate_cbs(rsp, rnp, rdp);
1841

P
Paul E. McKenney 已提交
1842
		rcu_report_qs_rnp(mask, rsp, rnp, flags); /* rlses rnp->lock */
1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854
	}
}

/*
 * 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)
{
1855 1856
	/* Check for grace-period ends and beginnings. */
	note_gp_changes(rsp, rdp);
1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868

	/*
	 * 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.
	 */
1869
	if (!rdp->passed_quiesce)
1870 1871
		return;

P
Paul E. McKenney 已提交
1872 1873 1874 1875
	/*
	 * Tell RCU we are done (but rcu_report_qs_rdp() will be the
	 * judge of that).
	 */
1876
	rcu_report_qs_rdp(rdp->cpu, rsp, rdp);
1877 1878 1879 1880
}

#ifdef CONFIG_HOTPLUG_CPU

1881
/*
1882 1883
 * Send the specified CPU's RCU callbacks to the orphanage.  The
 * specified CPU must be offline, and the caller must hold the
1884
 * ->orphan_lock.
1885
 */
1886 1887 1888
static void
rcu_send_cbs_to_orphanage(int cpu, struct rcu_state *rsp,
			  struct rcu_node *rnp, struct rcu_data *rdp)
1889
{
P
Paul E. McKenney 已提交
1890
	/* No-CBs CPUs do not have orphanable callbacks. */
1891
	if (rcu_is_nocb_cpu(rdp->cpu))
P
Paul E. McKenney 已提交
1892 1893
		return;

1894 1895
	/*
	 * Orphan the callbacks.  First adjust the counts.  This is safe
1896 1897
	 * because _rcu_barrier() excludes CPU-hotplug operations, so it
	 * cannot be running now.  Thus no memory barrier is required.
1898
	 */
1899
	if (rdp->nxtlist != NULL) {
1900 1901 1902
		rsp->qlen_lazy += rdp->qlen_lazy;
		rsp->qlen += rdp->qlen;
		rdp->n_cbs_orphaned += rdp->qlen;
1903
		rdp->qlen_lazy = 0;
1904
		ACCESS_ONCE(rdp->qlen) = 0;
1905 1906 1907
	}

	/*
1908 1909 1910 1911 1912 1913 1914
	 * 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.
1915
	 */
1916 1917 1918 1919
	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;
1920 1921 1922
	}

	/*
1923 1924 1925
	 * 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.
1926
	 */
1927
	if (rdp->nxtlist != NULL) {
1928 1929
		*rsp->orphan_donetail = rdp->nxtlist;
		rsp->orphan_donetail = rdp->nxttail[RCU_DONE_TAIL];
1930
	}
1931

1932
	/* Finally, initialize the rcu_data structure's list to empty.  */
1933
	init_callback_list(rdp);
1934 1935 1936 1937
}

/*
 * Adopt the RCU callbacks from the specified rcu_state structure's
1938
 * orphanage.  The caller must hold the ->orphan_lock.
1939
 */
1940
static void rcu_adopt_orphan_cbs(struct rcu_state *rsp, unsigned long flags)
1941 1942 1943 1944
{
	int i;
	struct rcu_data *rdp = __this_cpu_ptr(rsp->rda);

P
Paul E. McKenney 已提交
1945
	/* No-CBs CPUs are handled specially. */
1946
	if (rcu_nocb_adopt_orphan_cbs(rsp, rdp, flags))
P
Paul E. McKenney 已提交
1947 1948
		return;

1949 1950 1951 1952
	/* Do the accounting first. */
	rdp->qlen_lazy += rsp->qlen_lazy;
	rdp->qlen += rsp->qlen;
	rdp->n_cbs_adopted += rsp->qlen;
1953 1954
	if (rsp->qlen_lazy != rsp->qlen)
		rcu_idle_count_callbacks_posted();
1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 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
	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);
1994 1995
	trace_rcu_grace_period(rsp->name,
			       rnp->gpnum + 1 - !!(rnp->qsmask & mask),
1996
			       TPS("cpuofl"));
1997 1998 1999
}

/*
2000
 * The CPU has been completely removed, and some other CPU is reporting
2001 2002
 * this fact from process context.  Do the remainder of the cleanup,
 * including orphaning the outgoing CPU's RCU callbacks, and also
2003 2004
 * 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.
2005
 */
2006
static void rcu_cleanup_dead_cpu(int cpu, struct rcu_state *rsp)
2007
{
2008 2009 2010
	unsigned long flags;
	unsigned long mask;
	int need_report = 0;
2011
	struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
2012
	struct rcu_node *rnp = rdp->mynode;  /* Outgoing CPU's rdp & rnp. */
2013

2014
	/* Adjust any no-longer-needed kthreads. */
T
Thomas Gleixner 已提交
2015
	rcu_boost_kthread_setaffinity(rnp, -1);
2016

2017
	/* Remove the dead CPU from the bitmasks in the rcu_node hierarchy. */
2018 2019

	/* Exclude any attempts to start a new grace period. */
2020
	mutex_lock(&rsp->onoff_mutex);
2021
	raw_spin_lock_irqsave(&rsp->orphan_lock, flags);
2022

2023 2024
	/* Orphan the dead CPU's callbacks, and adopt them if appropriate. */
	rcu_send_cbs_to_orphanage(cpu, rsp, rnp, rdp);
2025
	rcu_adopt_orphan_cbs(rsp, flags);
2026

2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047
	/* 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. */
		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
2048
	 * because invoking rcu_report_unblock_qs_rnp() with ->orphan_lock
2049 2050
	 * held leads to deadlock.
	 */
2051
	raw_spin_unlock(&rsp->orphan_lock); /* irqs remain disabled. */
2052 2053 2054 2055 2056 2057 2058
	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);
2059 2060 2061
	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);
2062 2063 2064
	init_callback_list(rdp);
	/* Disallow further callbacks on this CPU. */
	rdp->nxttail[RCU_NEXT_TAIL] = NULL;
2065
	mutex_unlock(&rsp->onoff_mutex);
2066 2067 2068 2069
}

#else /* #ifdef CONFIG_HOTPLUG_CPU */

2070
static void rcu_cleanup_dying_cpu(struct rcu_state *rsp)
2071 2072 2073
{
}

2074
static void rcu_cleanup_dead_cpu(int cpu, struct rcu_state *rsp)
2075 2076 2077 2078 2079 2080 2081 2082 2083
{
}

#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.
 */
2084
static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp)
2085 2086 2087
{
	unsigned long flags;
	struct rcu_head *next, *list, **tail;
E
Eric Dumazet 已提交
2088 2089
	long bl, count, count_lazy;
	int i;
2090

2091
	/* If no callbacks are ready, just return. */
2092
	if (!cpu_has_callbacks_ready_to_invoke(rdp)) {
2093
		trace_rcu_batch_start(rsp->name, rdp->qlen_lazy, rdp->qlen, 0);
2094 2095 2096
		trace_rcu_batch_end(rsp->name, 0, !!ACCESS_ONCE(rdp->nxtlist),
				    need_resched(), is_idle_task(current),
				    rcu_is_callbacks_kthread());
2097
		return;
2098
	}
2099 2100 2101 2102 2103 2104

	/*
	 * Extract the list of ready callbacks, disabling to prevent
	 * races with call_rcu() from interrupt handlers.
	 */
	local_irq_save(flags);
2105
	WARN_ON_ONCE(cpu_is_offline(smp_processor_id()));
2106
	bl = rdp->blimit;
2107
	trace_rcu_batch_start(rsp->name, rdp->qlen_lazy, rdp->qlen, bl);
2108 2109 2110 2111
	list = rdp->nxtlist;
	rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL];
	*rdp->nxttail[RCU_DONE_TAIL] = NULL;
	tail = rdp->nxttail[RCU_DONE_TAIL];
2112 2113 2114
	for (i = RCU_NEXT_SIZE - 1; i >= 0; i--)
		if (rdp->nxttail[i] == rdp->nxttail[RCU_DONE_TAIL])
			rdp->nxttail[i] = &rdp->nxtlist;
2115 2116 2117
	local_irq_restore(flags);

	/* Invoke callbacks. */
2118
	count = count_lazy = 0;
2119 2120 2121
	while (list) {
		next = list->next;
		prefetch(next);
2122
		debug_rcu_head_unqueue(list);
2123 2124
		if (__rcu_reclaim(rsp->name, list))
			count_lazy++;
2125
		list = next;
2126 2127 2128 2129
		/* Stop only if limit reached and CPU has something to do. */
		if (++count >= bl &&
		    (need_resched() ||
		     (!is_idle_task(current) && !rcu_is_callbacks_kthread())))
2130 2131 2132 2133
			break;
	}

	local_irq_save(flags);
2134 2135 2136
	trace_rcu_batch_end(rsp->name, count, !!list, need_resched(),
			    is_idle_task(current),
			    rcu_is_callbacks_kthread());
2137 2138 2139 2140 2141

	/* Update count, and requeue any remaining callbacks. */
	if (list != NULL) {
		*tail = rdp->nxtlist;
		rdp->nxtlist = list;
2142 2143 2144
		for (i = 0; i < RCU_NEXT_SIZE; i++)
			if (&rdp->nxtlist == rdp->nxttail[i])
				rdp->nxttail[i] = tail;
2145 2146 2147
			else
				break;
	}
2148 2149
	smp_mb(); /* List handling before counting for rcu_barrier(). */
	rdp->qlen_lazy -= count_lazy;
2150
	ACCESS_ONCE(rdp->qlen) -= count;
2151
	rdp->n_cbs_invoked += count;
2152 2153 2154 2155 2156

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

2157 2158 2159 2160 2161 2162
	/* 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;
2163
	WARN_ON_ONCE((rdp->nxtlist == NULL) != (rdp->qlen == 0));
2164

2165 2166
	local_irq_restore(flags);

2167
	/* Re-invoke RCU core processing if there are callbacks remaining. */
2168
	if (cpu_has_callbacks_ready_to_invoke(rdp))
2169
		invoke_rcu_core();
2170 2171 2172 2173 2174
}

/*
 * 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).
2175
 * Also schedule RCU core processing.
2176
 *
2177
 * This function must be called from hardirq context.  It is normally
2178 2179 2180 2181 2182
 * 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)
{
2183
	trace_rcu_utilization(TPS("Start scheduler-tick"));
2184
	increment_cpu_stall_ticks();
2185
	if (user || rcu_is_cpu_rrupt_from_idle()) {
2186 2187 2188 2189 2190

		/*
		 * 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
2191
		 * a quiescent state, so note it.
2192 2193
		 *
		 * No memory barrier is required here because both
2194 2195 2196
		 * 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.
2197 2198
		 */

2199 2200
		rcu_sched_qs(cpu);
		rcu_bh_qs(cpu);
2201 2202 2203 2204 2205 2206 2207

	} 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
2208
		 * critical section, so note it.
2209 2210
		 */

2211
		rcu_bh_qs(cpu);
2212
	}
2213
	rcu_preempt_check_callbacks(cpu);
2214
	if (rcu_pending(cpu))
2215
		invoke_rcu_core();
2216
	trace_rcu_utilization(TPS("End scheduler-tick"));
2217 2218 2219 2220 2221
}

/*
 * Scan the leaf rcu_node structures, processing dyntick state for any that
 * have not yet encountered a quiescent state, using the function specified.
2222 2223
 * Also initiate boosting for any threads blocked on the root rcu_node.
 *
2224
 * The caller must have suppressed start of new grace periods.
2225
 */
2226 2227 2228 2229
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)
2230 2231 2232 2233 2234
{
	unsigned long bit;
	int cpu;
	unsigned long flags;
	unsigned long mask;
2235
	struct rcu_node *rnp;
2236

2237
	rcu_for_each_leaf_node(rsp, rnp) {
2238
		cond_resched();
2239
		mask = 0;
P
Paul E. McKenney 已提交
2240
		raw_spin_lock_irqsave(&rnp->lock, flags);
2241
		if (!rcu_gp_in_progress(rsp)) {
P
Paul E. McKenney 已提交
2242
			raw_spin_unlock_irqrestore(&rnp->lock, flags);
2243
			return;
2244
		}
2245
		if (rnp->qsmask == 0) {
2246
			rcu_initiate_boost(rnp, flags); /* releases rnp->lock */
2247 2248
			continue;
		}
2249
		cpu = rnp->grplo;
2250
		bit = 1;
2251
		for (; cpu <= rnp->grphi; cpu++, bit <<= 1) {
2252 2253 2254 2255 2256 2257
			if ((rnp->qsmask & bit) != 0) {
				if ((rnp->qsmaskinit & bit) != 0)
					*isidle = 0;
				if (f(per_cpu_ptr(rsp->rda, cpu), isidle, maxj))
					mask |= bit;
			}
2258
		}
2259
		if (mask != 0) {
2260

P
Paul E. McKenney 已提交
2261 2262
			/* rcu_report_qs_rnp() releases rnp->lock. */
			rcu_report_qs_rnp(mask, rsp, rnp, flags);
2263 2264
			continue;
		}
P
Paul E. McKenney 已提交
2265
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
2266
	}
2267
	rnp = rcu_get_root(rsp);
2268 2269 2270 2271
	if (rnp->qsmask == 0) {
		raw_spin_lock_irqsave(&rnp->lock, flags);
		rcu_initiate_boost(rnp, flags); /* releases rnp->lock. */
	}
2272 2273 2274 2275 2276 2277
}

/*
 * Force quiescent states on reluctant CPUs, and also detect which
 * CPUs are in dyntick-idle mode.
 */
2278
static void force_quiescent_state(struct rcu_state *rsp)
2279 2280
{
	unsigned long flags;
2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298
	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) {
			rsp->n_force_qs_lh++;
			return;
		}
		rnp_old = rnp;
	}
	/* rnp_old == rcu_get_root(rsp), rnp == NULL. */
2299

2300 2301 2302 2303 2304 2305
	/* Reached the root of the rcu_node tree, acquire lock. */
	raw_spin_lock_irqsave(&rnp_old->lock, flags);
	raw_spin_unlock(&rnp_old->fqslock);
	if (ACCESS_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) {
		rsp->n_force_qs_lh++;
		raw_spin_unlock_irqrestore(&rnp_old->lock, flags);
2306
		return;  /* Someone beat us to it. */
2307
	}
2308
	rsp->gp_flags |= RCU_GP_FLAG_FQS;
2309
	raw_spin_unlock_irqrestore(&rnp_old->lock, flags);
2310
	wake_up(&rsp->gp_wq);  /* Memory barrier implied by wake_up() path. */
2311 2312 2313
}

/*
2314 2315 2316
 * 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.
2317 2318
 */
static void
2319
__rcu_process_callbacks(struct rcu_state *rsp)
2320 2321
{
	unsigned long flags;
2322
	struct rcu_data *rdp = __this_cpu_ptr(rsp->rda);
2323

2324 2325
	WARN_ON_ONCE(rdp->beenonline == 0);

2326 2327 2328 2329
	/* 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? */
2330
	local_irq_save(flags);
2331
	if (cpu_needs_another_gp(rsp, rdp)) {
2332
		raw_spin_lock(&rcu_get_root(rsp)->lock); /* irqs disabled. */
2333 2334
		rcu_start_gp(rsp);
		raw_spin_unlock_irqrestore(&rcu_get_root(rsp)->lock, flags);
2335 2336
	} else {
		local_irq_restore(flags);
2337 2338 2339
	}

	/* If there are callbacks ready, invoke them. */
2340
	if (cpu_has_callbacks_ready_to_invoke(rdp))
2341
		invoke_rcu_callbacks(rsp, rdp);
2342 2343 2344

	/* Do any needed deferred wakeups of rcuo kthreads. */
	do_nocb_deferred_wakeup(rdp);
2345 2346
}

2347
/*
2348
 * Do RCU core processing for the current CPU.
2349
 */
2350
static void rcu_process_callbacks(struct softirq_action *unused)
2351
{
2352 2353
	struct rcu_state *rsp;

2354 2355
	if (cpu_is_offline(smp_processor_id()))
		return;
2356
	trace_rcu_utilization(TPS("Start RCU core"));
2357 2358
	for_each_rcu_flavor(rsp)
		__rcu_process_callbacks(rsp);
2359
	trace_rcu_utilization(TPS("End RCU core"));
2360 2361
}

2362
/*
2363 2364 2365 2366 2367
 * 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.
2368
 */
2369
static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp)
2370
{
2371 2372
	if (unlikely(!ACCESS_ONCE(rcu_scheduler_fully_active)))
		return;
2373 2374
	if (likely(!rsp->boost)) {
		rcu_do_batch(rsp, rdp);
2375 2376
		return;
	}
2377
	invoke_rcu_callbacks_kthread();
2378 2379
}

2380
static void invoke_rcu_core(void)
2381
{
2382 2383
	if (cpu_online(smp_processor_id()))
		raise_softirq(RCU_SOFTIRQ);
2384 2385
}

2386 2387 2388 2389 2390
/*
 * 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)
2391
{
2392 2393 2394 2395
	/*
	 * If called from an extended quiescent state, invoke the RCU
	 * core in order to force a re-evaluation of RCU's idleness.
	 */
2396
	if (!rcu_is_watching() && cpu_online(smp_processor_id()))
2397 2398
		invoke_rcu_core();

2399
	/* If interrupts were disabled or CPU offline, don't invoke RCU core. */
2400
	if (irqs_disabled_flags(flags) || cpu_is_offline(smp_processor_id()))
2401
		return;
2402

2403 2404 2405 2406 2407 2408 2409
	/*
	 * 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.
	 */
2410
	if (unlikely(rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) {
2411 2412

		/* Are we ignoring a completed grace period? */
2413
		note_gp_changes(rsp, rdp);
2414 2415 2416 2417 2418

		/* 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);

2419 2420 2421
			raw_spin_lock(&rnp_root->lock);
			rcu_start_gp(rsp);
			raw_spin_unlock(&rnp_root->lock);
2422 2423 2424 2425 2426
		} 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)
2427
				force_quiescent_state(rsp);
2428 2429 2430
			rdp->n_force_qs_snap = rsp->n_force_qs;
			rdp->qlen_last_fqs_check = rdp->qlen;
		}
2431
	}
2432 2433
}

2434 2435 2436 2437 2438 2439 2440
/*
 * RCU callback function to leak a callback.
 */
static void rcu_leak_callback(struct rcu_head *rhp)
{
}

P
Paul E. McKenney 已提交
2441 2442 2443 2444 2445 2446
/*
 * 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.
 */
2447 2448
static void
__call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
P
Paul E. McKenney 已提交
2449
	   struct rcu_state *rsp, int cpu, bool lazy)
2450 2451 2452 2453
{
	unsigned long flags;
	struct rcu_data *rdp;

2454
	WARN_ON_ONCE((unsigned long)head & 0x3); /* Misaligned rcu_head! */
2455 2456 2457 2458 2459 2460
	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;
	}
2461 2462 2463 2464 2465 2466 2467 2468 2469 2470
	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);
2471
	rdp = this_cpu_ptr(rsp->rda);
2472 2473

	/* Add the callback to our list. */
P
Paul E. McKenney 已提交
2474 2475 2476 2477 2478
	if (unlikely(rdp->nxttail[RCU_NEXT_TAIL] == NULL) || cpu != -1) {
		int offline;

		if (cpu != -1)
			rdp = per_cpu_ptr(rsp->rda, cpu);
2479
		offline = !__call_rcu_nocb(rdp, head, lazy, flags);
P
Paul E. McKenney 已提交
2480
		WARN_ON_ONCE(offline);
2481 2482 2483 2484
		/* _call_rcu() is illegal on offline CPU; leak the callback. */
		local_irq_restore(flags);
		return;
	}
2485
	ACCESS_ONCE(rdp->qlen)++;
2486 2487
	if (lazy)
		rdp->qlen_lazy++;
2488 2489
	else
		rcu_idle_count_callbacks_posted();
2490 2491 2492
	smp_mb();  /* Count before adding callback for rcu_barrier(). */
	*rdp->nxttail[RCU_NEXT_TAIL] = head;
	rdp->nxttail[RCU_NEXT_TAIL] = &head->next;
2493

2494 2495
	if (__is_kfree_rcu_offset((unsigned long)func))
		trace_rcu_kfree_callback(rsp->name, head, (unsigned long)func,
2496
					 rdp->qlen_lazy, rdp->qlen);
2497
	else
2498
		trace_rcu_callback(rsp->name, head, rdp->qlen_lazy, rdp->qlen);
2499

2500 2501
	/* Go handle any RCU core processing required. */
	__call_rcu_core(rsp, rdp, head, flags);
2502 2503 2504 2505
	local_irq_restore(flags);
}

/*
2506
 * Queue an RCU-sched callback for invocation after a grace period.
2507
 */
2508
void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
2509
{
P
Paul E. McKenney 已提交
2510
	__call_rcu(head, func, &rcu_sched_state, -1, 0);
2511
}
2512
EXPORT_SYMBOL_GPL(call_rcu_sched);
2513 2514

/*
2515
 * Queue an RCU callback for invocation after a quicker grace period.
2516 2517 2518
 */
void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
{
P
Paul E. McKenney 已提交
2519
	__call_rcu(head, func, &rcu_bh_state, -1, 0);
2520 2521 2522
}
EXPORT_SYMBOL_GPL(call_rcu_bh);

2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533
/*
 * 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)
{
2534 2535
	int ret;

2536
	might_sleep();  /* Check for RCU read-side critical section. */
2537 2538 2539 2540
	preempt_disable();
	ret = num_online_cpus() <= 1;
	preempt_enable();
	return ret;
2541 2542
}

2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554
/**
 * 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
2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576
 * 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).
2577 2578 2579 2580 2581 2582 2583 2584 2585
 *
 * 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)
{
2586 2587 2588 2589
	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");
2590 2591
	if (rcu_blocking_is_gp())
		return;
2592 2593 2594 2595
	if (rcu_expedited)
		synchronize_sched_expedited();
	else
		wait_rcu_gp(call_rcu_sched);
2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606
}
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.
2607 2608 2609
 *
 * See the description of synchronize_sched() for more detailed information
 * on memory ordering guarantees.
2610 2611 2612
 */
void synchronize_rcu_bh(void)
{
2613 2614 2615 2616
	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");
2617 2618
	if (rcu_blocking_is_gp())
		return;
2619 2620 2621 2622
	if (rcu_expedited)
		synchronize_rcu_bh_expedited();
	else
		wait_rcu_gp(call_rcu_bh);
2623 2624 2625
}
EXPORT_SYMBOL_GPL(synchronize_rcu_bh);

2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642
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;
}

2643 2644 2645 2646 2647 2648 2649 2650 2651 2652
/**
 * 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.
2653
 *
2654 2655 2656 2657
 * 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.
2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681
 *
 * 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)
{
2682 2683
	long firstsnap, s, snap;
	int trycount = 0;
2684
	struct rcu_state *rsp = &rcu_sched_state;
2685

2686 2687 2688 2689 2690 2691 2692 2693
	/*
	 * 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.
	 */
2694 2695
	if (ULONG_CMP_GE((ulong)atomic_long_read(&rsp->expedited_start),
			 (ulong)atomic_long_read(&rsp->expedited_done) +
2696 2697
			 ULONG_MAX / 8)) {
		synchronize_sched();
2698
		atomic_long_inc(&rsp->expedited_wrap);
2699 2700
		return;
	}
2701

2702 2703 2704 2705
	/*
	 * Take a ticket.  Note that atomic_inc_return() implies a
	 * full memory barrier.
	 */
2706
	snap = atomic_long_inc_return(&rsp->expedited_start);
2707
	firstsnap = snap;
2708
	get_online_cpus();
2709
	WARN_ON_ONCE(cpu_is_offline(raw_smp_processor_id()));
2710 2711 2712 2713 2714 2715 2716 2717 2718

	/*
	 * 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();
2719
		atomic_long_inc(&rsp->expedited_tryfail);
2720

2721
		/* Check to see if someone else did our work for us. */
2722
		s = atomic_long_read(&rsp->expedited_done);
2723
		if (ULONG_CMP_GE((ulong)s, (ulong)firstsnap)) {
2724 2725 2726
			/* ensure test happens before caller kfree */
			smp_mb__before_atomic_inc(); /* ^^^ */
			atomic_long_inc(&rsp->expedited_workdone1);
2727 2728
			return;
		}
2729 2730

		/* No joy, try again later.  Or just synchronize_sched(). */
2731
		if (trycount++ < 10) {
2732
			udelay(trycount * num_online_cpus());
2733
		} else {
2734
			wait_rcu_gp(call_rcu_sched);
2735
			atomic_long_inc(&rsp->expedited_normal);
2736 2737 2738
			return;
		}

2739
		/* Recheck to see if someone else did our work for us. */
2740
		s = atomic_long_read(&rsp->expedited_done);
2741
		if (ULONG_CMP_GE((ulong)s, (ulong)firstsnap)) {
2742 2743 2744
			/* ensure test happens before caller kfree */
			smp_mb__before_atomic_inc(); /* ^^^ */
			atomic_long_inc(&rsp->expedited_workdone2);
2745 2746 2747 2748 2749
			return;
		}

		/*
		 * Refetching sync_sched_expedited_started allows later
2750 2751 2752 2753
		 * 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.
2754 2755
		 */
		get_online_cpus();
2756
		snap = atomic_long_read(&rsp->expedited_start);
2757 2758
		smp_mb(); /* ensure read is before try_stop_cpus(). */
	}
2759
	atomic_long_inc(&rsp->expedited_stoppedcpus);
2760 2761 2762 2763 2764

	/*
	 * 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
2765
	 * than we did already did their update.
2766 2767
	 */
	do {
2768
		atomic_long_inc(&rsp->expedited_done_tries);
2769
		s = atomic_long_read(&rsp->expedited_done);
2770
		if (ULONG_CMP_GE((ulong)s, (ulong)snap)) {
2771 2772 2773
			/* ensure test happens before caller kfree */
			smp_mb__before_atomic_inc(); /* ^^^ */
			atomic_long_inc(&rsp->expedited_done_lost);
2774 2775
			break;
		}
2776
	} while (atomic_long_cmpxchg(&rsp->expedited_done, s, snap) != s);
2777
	atomic_long_inc(&rsp->expedited_done_exit);
2778 2779 2780 2781 2782

	put_online_cpus();
}
EXPORT_SYMBOL_GPL(synchronize_sched_expedited);

2783 2784 2785 2786 2787 2788 2789 2790 2791
/*
 * 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)
{
2792 2793
	struct rcu_node *rnp = rdp->mynode;

2794 2795 2796 2797 2798 2799
	rdp->n_rcu_pending++;

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

	/* Is the RCU core waiting for a quiescent state from this CPU? */
2800 2801
	if (rcu_scheduler_fully_active &&
	    rdp->qs_pending && !rdp->passed_quiesce) {
2802
		rdp->n_rp_qs_pending++;
2803
	} else if (rdp->qs_pending && rdp->passed_quiesce) {
2804
		rdp->n_rp_report_qs++;
2805
		return 1;
2806
	}
2807 2808

	/* Does this CPU have callbacks ready to invoke? */
2809 2810
	if (cpu_has_callbacks_ready_to_invoke(rdp)) {
		rdp->n_rp_cb_ready++;
2811
		return 1;
2812
	}
2813 2814

	/* Has RCU gone idle with this CPU needing another grace period? */
2815 2816
	if (cpu_needs_another_gp(rsp, rdp)) {
		rdp->n_rp_cpu_needs_gp++;
2817
		return 1;
2818
	}
2819 2820

	/* Has another RCU grace period completed?  */
2821
	if (ACCESS_ONCE(rnp->completed) != rdp->completed) { /* outside lock */
2822
		rdp->n_rp_gp_completed++;
2823
		return 1;
2824
	}
2825 2826

	/* Has a new RCU grace period started? */
2827
	if (ACCESS_ONCE(rnp->gpnum) != rdp->gpnum) { /* outside lock */
2828
		rdp->n_rp_gp_started++;
2829
		return 1;
2830
	}
2831

2832 2833 2834 2835 2836 2837
	/* Does this CPU need a deferred NOCB wakeup? */
	if (rcu_nocb_need_deferred_wakeup(rdp)) {
		rdp->n_rp_nocb_defer_wakeup++;
		return 1;
	}

2838
	/* nothing to do */
2839
	rdp->n_rp_need_nothing++;
2840 2841 2842 2843 2844 2845 2846 2847
	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.
 */
2848
static int rcu_pending(int cpu)
2849
{
2850 2851 2852 2853 2854 2855
	struct rcu_state *rsp;

	for_each_rcu_flavor(rsp)
		if (__rcu_pending(rsp, per_cpu_ptr(rsp->rda, cpu)))
			return 1;
	return 0;
2856 2857 2858
}

/*
2859 2860 2861
 * 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.)
2862
 */
2863
static int rcu_cpu_has_callbacks(int cpu, bool *all_lazy)
2864
{
2865 2866 2867
	bool al = true;
	bool hc = false;
	struct rcu_data *rdp;
2868 2869
	struct rcu_state *rsp;

2870 2871
	for_each_rcu_flavor(rsp) {
		rdp = per_cpu_ptr(rsp->rda, cpu);
2872 2873 2874 2875
		if (!rdp->nxtlist)
			continue;
		hc = true;
		if (rdp->qlen != rdp->qlen_lazy || !all_lazy) {
2876
			al = false;
2877 2878
			break;
		}
2879 2880 2881 2882
	}
	if (all_lazy)
		*all_lazy = al;
	return hc;
2883 2884
}

2885 2886 2887 2888
/*
 * Helper function for _rcu_barrier() tracing.  If tracing is disabled,
 * the compiler is expected to optimize this away.
 */
2889
static void _rcu_barrier_trace(struct rcu_state *rsp, const char *s,
2890 2891 2892 2893 2894 2895
			       int cpu, unsigned long done)
{
	trace_rcu_barrier(rsp->name, s, cpu,
			  atomic_read(&rsp->barrier_cpu_count), done);
}

2896 2897 2898 2899
/*
 * RCU callback function for _rcu_barrier().  If we are last, wake
 * up the task executing _rcu_barrier().
 */
2900
static void rcu_barrier_callback(struct rcu_head *rhp)
2901
{
2902 2903 2904
	struct rcu_data *rdp = container_of(rhp, struct rcu_data, barrier_head);
	struct rcu_state *rsp = rdp->rsp;

2905 2906
	if (atomic_dec_and_test(&rsp->barrier_cpu_count)) {
		_rcu_barrier_trace(rsp, "LastCB", -1, rsp->n_barrier_done);
2907
		complete(&rsp->barrier_completion);
2908 2909 2910
	} else {
		_rcu_barrier_trace(rsp, "CB", -1, rsp->n_barrier_done);
	}
2911 2912 2913 2914 2915 2916 2917
}

/*
 * Called with preemption disabled, and from cross-cpu IRQ context.
 */
static void rcu_barrier_func(void *type)
{
2918
	struct rcu_state *rsp = type;
2919
	struct rcu_data *rdp = __this_cpu_ptr(rsp->rda);
2920

2921
	_rcu_barrier_trace(rsp, "IRQ", -1, rsp->n_barrier_done);
2922
	atomic_inc(&rsp->barrier_cpu_count);
2923
	rsp->call(&rdp->barrier_head, rcu_barrier_callback);
2924 2925 2926 2927 2928 2929
}

/*
 * Orchestrate the specified type of RCU barrier, waiting for all
 * RCU callbacks of the specified type to complete.
 */
2930
static void _rcu_barrier(struct rcu_state *rsp)
2931
{
2932 2933
	int cpu;
	struct rcu_data *rdp;
2934 2935
	unsigned long snap = ACCESS_ONCE(rsp->n_barrier_done);
	unsigned long snap_done;
2936

2937
	_rcu_barrier_trace(rsp, "Begin", -1, snap);
2938

2939
	/* Take mutex to serialize concurrent rcu_barrier() requests. */
2940
	mutex_lock(&rsp->barrier_mutex);
2941

2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953
	/*
	 * 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.
	 */
2954
	snap_done = rsp->n_barrier_done;
2955
	_rcu_barrier_trace(rsp, "Check", -1, snap_done);
2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967

	/*
	 * 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)) {
2968
		_rcu_barrier_trace(rsp, "EarlyExit", -1, snap_done);
2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980
		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);
2981
	_rcu_barrier_trace(rsp, "Inc1", -1, rsp->n_barrier_done);
2982
	smp_mb(); /* Order ->n_barrier_done increment with below mechanism. */
2983

2984
	/*
2985 2986
	 * 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
2987 2988
	 * (or preemption of this task).  Exclude CPU-hotplug operations
	 * to ensure that no offline CPU has callbacks queued.
2989
	 */
2990
	init_completion(&rsp->barrier_completion);
2991
	atomic_set(&rsp->barrier_cpu_count, 1);
2992
	get_online_cpus();
2993 2994

	/*
2995 2996 2997
	 * 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.
2998
	 */
P
Paul E. McKenney 已提交
2999
	for_each_possible_cpu(cpu) {
3000
		if (!cpu_online(cpu) && !rcu_is_nocb_cpu(cpu))
P
Paul E. McKenney 已提交
3001
			continue;
3002
		rdp = per_cpu_ptr(rsp->rda, cpu);
3003
		if (rcu_is_nocb_cpu(cpu)) {
P
Paul E. McKenney 已提交
3004 3005 3006 3007 3008 3009
			_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)) {
3010 3011
			_rcu_barrier_trace(rsp, "OnlineQ", cpu,
					   rsp->n_barrier_done);
3012
			smp_call_function_single(cpu, rcu_barrier_func, rsp, 1);
3013
		} else {
3014 3015
			_rcu_barrier_trace(rsp, "OnlineNQ", cpu,
					   rsp->n_barrier_done);
3016 3017
		}
	}
3018
	put_online_cpus();
3019 3020 3021 3022 3023

	/*
	 * Now that we have an rcu_barrier_callback() callback on each
	 * CPU, and thus each counted, remove the initial count.
	 */
3024
	if (atomic_dec_and_test(&rsp->barrier_cpu_count))
3025
		complete(&rsp->barrier_completion);
3026

3027 3028 3029 3030
	/* 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);
3031
	_rcu_barrier_trace(rsp, "Inc2", -1, rsp->n_barrier_done);
3032 3033
	smp_mb(); /* Keep increment before caller's subsequent code. */

3034
	/* Wait for all rcu_barrier_callback() callbacks to be invoked. */
3035
	wait_for_completion(&rsp->barrier_completion);
3036 3037

	/* Other rcu_barrier() invocations can now safely proceed. */
3038
	mutex_unlock(&rsp->barrier_mutex);
3039 3040 3041 3042 3043 3044 3045
}

/**
 * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete.
 */
void rcu_barrier_bh(void)
{
3046
	_rcu_barrier(&rcu_bh_state);
3047 3048 3049 3050 3051 3052 3053 3054
}
EXPORT_SYMBOL_GPL(rcu_barrier_bh);

/**
 * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks.
 */
void rcu_barrier_sched(void)
{
3055
	_rcu_barrier(&rcu_sched_state);
3056 3057 3058
}
EXPORT_SYMBOL_GPL(rcu_barrier_sched);

3059
/*
3060
 * Do boot-time initialization of a CPU's per-CPU RCU data.
3061
 */
3062 3063
static void __init
rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp)
3064 3065
{
	unsigned long flags;
3066
	struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
3067 3068 3069
	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
3070
	raw_spin_lock_irqsave(&rnp->lock, flags);
3071
	rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo);
3072
	init_callback_list(rdp);
3073
	rdp->qlen_lazy = 0;
3074
	ACCESS_ONCE(rdp->qlen) = 0;
3075
	rdp->dynticks = &per_cpu(rcu_dynticks, cpu);
3076
	WARN_ON_ONCE(rdp->dynticks->dynticks_nesting != DYNTICK_TASK_EXIT_IDLE);
3077
	WARN_ON_ONCE(atomic_read(&rdp->dynticks->dynticks) != 1);
3078
	rdp->cpu = cpu;
3079
	rdp->rsp = rsp;
P
Paul E. McKenney 已提交
3080
	rcu_boot_init_nocb_percpu_data(rdp);
P
Paul E. McKenney 已提交
3081
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
3082 3083 3084 3085 3086 3087 3088
}

/*
 * 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.
3089
 */
3090
static void
P
Paul E. McKenney 已提交
3091
rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptible)
3092 3093 3094
{
	unsigned long flags;
	unsigned long mask;
3095
	struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
3096 3097
	struct rcu_node *rnp = rcu_get_root(rsp);

3098 3099 3100
	/* Exclude new grace periods. */
	mutex_lock(&rsp->onoff_mutex);

3101
	/* Set up local state, ensuring consistent view of global state. */
P
Paul E. McKenney 已提交
3102
	raw_spin_lock_irqsave(&rnp->lock, flags);
3103
	rdp->beenonline = 1;	 /* We have now been online. */
P
Paul E. McKenney 已提交
3104
	rdp->preemptible = preemptible;
3105 3106
	rdp->qlen_last_fqs_check = 0;
	rdp->n_force_qs_snap = rsp->n_force_qs;
3107
	rdp->blimit = blimit;
3108
	init_callback_list(rdp);  /* Re-enable callbacks on this CPU. */
3109
	rdp->dynticks->dynticks_nesting = DYNTICK_TASK_EXIT_IDLE;
3110
	rcu_sysidle_init_percpu_data(rdp->dynticks);
3111 3112
	atomic_set(&rdp->dynticks->dynticks,
		   (atomic_read(&rdp->dynticks->dynticks) & ~0x1) + 1);
P
Paul E. McKenney 已提交
3113
	raw_spin_unlock(&rnp->lock);		/* irqs remain disabled. */
3114 3115 3116 3117 3118 3119

	/* 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
3120
		raw_spin_lock(&rnp->lock);	/* irqs already disabled. */
3121 3122
		rnp->qsmaskinit |= mask;
		mask = rnp->grpmask;
3123
		if (rnp == rdp->mynode) {
3124 3125 3126 3127 3128 3129
			/*
			 * 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;
3130
			rdp->completed = rnp->completed;
3131 3132
			rdp->passed_quiesce = 0;
			rdp->qs_pending = 0;
3133
			trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpuonl"));
3134
		}
P
Paul E. McKenney 已提交
3135
		raw_spin_unlock(&rnp->lock); /* irqs already disabled. */
3136 3137
		rnp = rnp->parent;
	} while (rnp != NULL && !(rnp->qsmaskinit & mask));
3138
	local_irq_restore(flags);
3139

3140
	mutex_unlock(&rsp->onoff_mutex);
3141 3142
}

3143
static void rcu_prepare_cpu(int cpu)
3144
{
3145 3146 3147 3148 3149
	struct rcu_state *rsp;

	for_each_rcu_flavor(rsp)
		rcu_init_percpu_data(cpu, rsp,
				     strcmp(rsp->name, "rcu_preempt") == 0);
3150 3151 3152
}

/*
3153
 * Handle CPU online/offline notification events.
3154
 */
3155
static int rcu_cpu_notify(struct notifier_block *self,
3156
				    unsigned long action, void *hcpu)
3157 3158
{
	long cpu = (long)hcpu;
3159
	struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, cpu);
3160
	struct rcu_node *rnp = rdp->mynode;
3161
	struct rcu_state *rsp;
3162

3163
	trace_rcu_utilization(TPS("Start CPU hotplug"));
3164 3165 3166
	switch (action) {
	case CPU_UP_PREPARE:
	case CPU_UP_PREPARE_FROZEN:
P
Peter Zijlstra 已提交
3167 3168
		rcu_prepare_cpu(cpu);
		rcu_prepare_kthreads(cpu);
3169 3170
		break;
	case CPU_ONLINE:
3171
	case CPU_DOWN_FAILED:
T
Thomas Gleixner 已提交
3172
		rcu_boost_kthread_setaffinity(rnp, -1);
3173 3174
		break;
	case CPU_DOWN_PREPARE:
3175
		rcu_boost_kthread_setaffinity(rnp, cpu);
3176
		break;
3177 3178
	case CPU_DYING:
	case CPU_DYING_FROZEN:
3179 3180
		for_each_rcu_flavor(rsp)
			rcu_cleanup_dying_cpu(rsp);
3181
		break;
3182 3183 3184 3185
	case CPU_DEAD:
	case CPU_DEAD_FROZEN:
	case CPU_UP_CANCELED:
	case CPU_UP_CANCELED_FROZEN:
3186 3187
		for_each_rcu_flavor(rsp)
			rcu_cleanup_dead_cpu(cpu, rsp);
3188 3189 3190 3191
		break;
	default:
		break;
	}
3192
	trace_rcu_utilization(TPS("End CPU hotplug"));
3193
	return NOTIFY_OK;
3194 3195
}

3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214
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;
}

3215 3216 3217 3218 3219 3220 3221 3222 3223 3224 3225
/*
 * 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) {
3226
		t = kthread_run(rcu_gp_kthread, rsp, "%s", rsp->name);
3227 3228 3229 3230 3231
		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
3232
		rcu_spawn_nocb_kthreads(rsp);
3233 3234 3235 3236 3237
	}
	return 0;
}
early_initcall(rcu_spawn_gp_kthread);

3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249 3250 3251 3252
/*
 * 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;
}

3253 3254 3255 3256 3257 3258 3259 3260 3261
/*
 * 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;

3262
	for (i = rcu_num_lvls - 1; i > 0; i--)
3263
		rsp->levelspread[i] = CONFIG_RCU_FANOUT;
3264
	rsp->levelspread[0] = rcu_fanout_leaf;
3265 3266 3267 3268 3269 3270 3271 3272
}
#else /* #ifdef CONFIG_RCU_FANOUT_EXACT */
static void __init rcu_init_levelspread(struct rcu_state *rsp)
{
	int ccur;
	int cprv;
	int i;

3273
	cprv = nr_cpu_ids;
3274
	for (i = rcu_num_lvls - 1; i >= 0; i--) {
3275 3276 3277 3278 3279 3280 3281 3282 3283 3284
		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.
 */
3285 3286
static void __init rcu_init_one(struct rcu_state *rsp,
		struct rcu_data __percpu *rda)
3287
{
3288 3289 3290 3291 3292 3293 3294 3295
	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 */
3296 3297 3298 3299 3300
	int cpustride = 1;
	int i;
	int j;
	struct rcu_node *rnp;

3301 3302
	BUILD_BUG_ON(MAX_RCU_LVLS > ARRAY_SIZE(buf));  /* Fix buf[] init! */

3303 3304 3305 3306
	/* 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");

3307 3308
	/* Initialize the level-tracking arrays. */

3309 3310 3311
	for (i = 0; i < rcu_num_lvls; i++)
		rsp->levelcnt[i] = num_rcu_lvl[i];
	for (i = 1; i < rcu_num_lvls; i++)
3312 3313 3314 3315 3316
		rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1];
	rcu_init_levelspread(rsp);

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

3317
	for (i = rcu_num_lvls - 1; i >= 0; i--) {
3318 3319 3320
		cpustride *= rsp->levelspread[i];
		rnp = rsp->level[i];
		for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) {
P
Paul E. McKenney 已提交
3321
			raw_spin_lock_init(&rnp->lock);
3322 3323
			lockdep_set_class_and_name(&rnp->lock,
						   &rcu_node_class[i], buf[i]);
3324 3325 3326
			raw_spin_lock_init(&rnp->fqslock);
			lockdep_set_class_and_name(&rnp->fqslock,
						   &rcu_fqs_class[i], fqs[i]);
3327 3328
			rnp->gpnum = rsp->gpnum;
			rnp->completed = rsp->completed;
3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345
			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;
3346
			INIT_LIST_HEAD(&rnp->blkd_tasks);
3347
			rcu_init_one_nocb(rnp);
3348 3349
		}
	}
3350

3351
	rsp->rda = rda;
3352
	init_waitqueue_head(&rsp->gp_wq);
3353
	init_irq_work(&rsp->wakeup_work, rsp_wakeup);
3354
	rnp = rsp->level[rcu_num_lvls - 1];
3355
	for_each_possible_cpu(i) {
3356
		while (i > rnp->grphi)
3357
			rnp++;
3358
		per_cpu_ptr(rsp->rda, i)->mynode = rnp;
3359 3360
		rcu_boot_init_percpu_data(i, rsp);
	}
3361
	list_add(&rsp->flavors, &rcu_struct_flavors);
3362 3363
}

3364 3365
/*
 * Compute the rcu_node tree geometry from kernel parameters.  This cannot
3366
 * replace the definitions in tree.h because those are needed to size
3367 3368 3369 3370
 * the ->node array in the rcu_state structure.
 */
static void __init rcu_init_geometry(void)
{
3371
	ulong d;
3372 3373
	int i;
	int j;
3374
	int n = nr_cpu_ids;
3375 3376
	int rcu_capacity[MAX_RCU_LVLS + 1];

3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389
	/*
	 * 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;

3390
	/* If the compile-time values are accurate, just leave. */
3391 3392
	if (rcu_fanout_leaf == CONFIG_RCU_FANOUT_LEAF &&
	    nr_cpu_ids == NR_CPUS)
3393
		return;
3394 3395
	pr_info("RCU: Adjusting geometry for rcu_fanout_leaf=%d, nr_cpu_ids=%d\n",
		rcu_fanout_leaf, nr_cpu_ids);
3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440

	/*
	 * 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;
}

3441
void __init rcu_init(void)
3442
{
P
Paul E. McKenney 已提交
3443
	int cpu;
3444

3445
	rcu_bootup_announce();
3446
	rcu_init_geometry();
3447
	rcu_init_one(&rcu_bh_state, &rcu_bh_data);
3448
	rcu_init_one(&rcu_sched_state, &rcu_sched_data);
3449
	__rcu_init_preempt();
J
Jiang Fang 已提交
3450
	open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
3451 3452 3453 3454 3455 3456 3457

	/*
	 * 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);
3458
	pm_notifier(rcu_pm_notify, 0);
P
Paul E. McKenney 已提交
3459 3460
	for_each_online_cpu(cpu)
		rcu_cpu_notify(NULL, CPU_UP_PREPARE, (void *)(long)cpu);
3461 3462
}

3463
#include "tree_plugin.h"