2 * Read-Copy Update mechanism for mutual exclusion
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
18 * Copyright IBM Corporation, 2008
20 * Authors: Dipankar Sarma <dipankar@in.ibm.com>
21 * Manfred Spraul <manfred@colorfullife.com>
22 * Paul E. McKenney <paulmck@linux.vnet.ibm.com> Hierarchical version
24 * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
25 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
27 * For detailed explanation of Read-Copy Update mechanism see -
30 #include <linux/types.h>
31 #include <linux/kernel.h>
32 #include <linux/init.h>
33 #include <linux/spinlock.h>
34 #include <linux/smp.h>
35 #include <linux/rcupdate.h>
36 #include <linux/interrupt.h>
37 #include <linux/sched.h>
38 #include <linux/nmi.h>
39 #include <linux/atomic.h>
40 #include <linux/bitops.h>
41 #include <linux/module.h>
42 #include <linux/completion.h>
43 #include <linux/moduleparam.h>
44 #include <linux/percpu.h>
45 #include <linux/notifier.h>
46 #include <linux/cpu.h>
47 #include <linux/mutex.h>
48 #include <linux/time.h>
49 #include <linux/kernel_stat.h>
50 #include <linux/wait.h>
51 #include <linux/kthread.h>
52 #include <linux/prefetch.h>
55 #include <trace/events/rcu.h>
59 /* Data structures. */
61 static struct lock_class_key rcu_node_class[NUM_RCU_LVLS];
63 #define RCU_STATE_INITIALIZER(structname) { \
64 .level = { &structname##_state.node[0] }, \
66 NUM_RCU_LVL_0, /* root of hierarchy. */ \
70 NUM_RCU_LVL_4, /* == MAX_RCU_LVLS */ \
72 .signaled = RCU_GP_IDLE, \
75 .onofflock = __RAW_SPIN_LOCK_UNLOCKED(&structname##_state.onofflock), \
76 .fqslock = __RAW_SPIN_LOCK_UNLOCKED(&structname##_state.fqslock), \
78 .n_force_qs_ngp = 0, \
79 .name = #structname, \
82 struct rcu_state rcu_sched_state = RCU_STATE_INITIALIZER(rcu_sched);
83 DEFINE_PER_CPU(struct rcu_data, rcu_sched_data);
85 struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh);
86 DEFINE_PER_CPU(struct rcu_data, rcu_bh_data);
88 static struct rcu_state *rcu_state;
91 * The rcu_scheduler_active variable transitions from zero to one just
92 * before the first task is spawned. So when this variable is zero, RCU
93 * can assume that there is but one task, allowing RCU to (for example)
94 * optimized synchronize_sched() to a simple barrier(). When this variable
95 * is one, RCU must actually do all the hard work required to detect real
96 * grace periods. This variable is also used to suppress boot-time false
97 * positives from lockdep-RCU error checking.
99 int rcu_scheduler_active __read_mostly;
100 EXPORT_SYMBOL_GPL(rcu_scheduler_active);
103 * The rcu_scheduler_fully_active variable transitions from zero to one
104 * during the early_initcall() processing, which is after the scheduler
105 * is capable of creating new tasks. So RCU processing (for example,
106 * creating tasks for RCU priority boosting) must be delayed until after
107 * rcu_scheduler_fully_active transitions from zero to one. We also
108 * currently delay invocation of any RCU callbacks until after this point.
110 * It might later prove better for people registering RCU callbacks during
111 * early boot to take responsibility for these callbacks, but one step at
114 static int rcu_scheduler_fully_active __read_mostly;
116 #ifdef CONFIG_RCU_BOOST
119 * Control variables for per-CPU and per-rcu_node kthreads. These
120 * handle all flavors of RCU.
122 static DEFINE_PER_CPU(struct task_struct *, rcu_cpu_kthread_task);
123 DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
124 DEFINE_PER_CPU(int, rcu_cpu_kthread_cpu);
125 DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
126 DEFINE_PER_CPU(char, rcu_cpu_has_work);
128 #endif /* #ifdef CONFIG_RCU_BOOST */
130 static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu);
131 static void invoke_rcu_core(void);
132 static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp);
134 #define RCU_KTHREAD_PRIO 1 /* RT priority for per-CPU kthreads. */
137 * Track the rcutorture test sequence number and the update version
138 * number within a given test. The rcutorture_testseq is incremented
139 * on every rcutorture module load and unload, so has an odd value
140 * when a test is running. The rcutorture_vernum is set to zero
141 * when rcutorture starts and is incremented on each rcutorture update.
142 * These variables enable correlating rcutorture output with the
143 * RCU tracing information.
145 unsigned long rcutorture_testseq;
146 unsigned long rcutorture_vernum;
149 * Return true if an RCU grace period is in progress. The ACCESS_ONCE()s
150 * permit this function to be invoked without holding the root rcu_node
151 * structure's ->lock, but of course results can be subject to change.
153 static int rcu_gp_in_progress(struct rcu_state *rsp)
155 return ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum);
159 * Note a quiescent state. Because we do not need to know
160 * how many quiescent states passed, just if there was at least
161 * one since the start of the grace period, this just sets a flag.
163 void rcu_sched_qs(int cpu)
165 struct rcu_data *rdp = &per_cpu(rcu_sched_data, cpu);
167 rdp->passed_quiesc_completed = rdp->gpnum - 1;
169 if (rdp->passed_quiesc == 0)
170 trace_rcu_grace_period("rcu_sched", rdp->gpnum, "cpuqs");
171 rdp->passed_quiesc = 1;
174 void rcu_bh_qs(int cpu)
176 struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu);
178 rdp->passed_quiesc_completed = rdp->gpnum - 1;
180 if (rdp->passed_quiesc == 0)
181 trace_rcu_grace_period("rcu_bh", rdp->gpnum, "cpuqs");
182 rdp->passed_quiesc = 1;
186 * Note a context switch. This is a quiescent state for RCU-sched,
187 * and requires special handling for preemptible RCU.
189 void rcu_note_context_switch(int cpu)
191 trace_rcu_utilization("Start context switch");
193 rcu_preempt_note_context_switch(cpu);
194 trace_rcu_utilization("End context switch");
196 EXPORT_SYMBOL_GPL(rcu_note_context_switch);
199 DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = {
200 .dynticks_nesting = 1,
201 .dynticks = ATOMIC_INIT(1),
203 #endif /* #ifdef CONFIG_NO_HZ */
205 static int blimit = 10; /* Maximum callbacks per rcu_do_batch. */
206 static int qhimark = 10000; /* If this many pending, ignore blimit. */
207 static int qlowmark = 100; /* Once only this many pending, use blimit. */
209 module_param(blimit, int, 0);
210 module_param(qhimark, int, 0);
211 module_param(qlowmark, int, 0);
213 int rcu_cpu_stall_suppress __read_mostly;
214 module_param(rcu_cpu_stall_suppress, int, 0644);
216 static void force_quiescent_state(struct rcu_state *rsp, int relaxed);
217 static int rcu_pending(int cpu);
220 * Return the number of RCU-sched batches processed thus far for debug & stats.
222 long rcu_batches_completed_sched(void)
224 return rcu_sched_state.completed;
226 EXPORT_SYMBOL_GPL(rcu_batches_completed_sched);
229 * Return the number of RCU BH batches processed thus far for debug & stats.
231 long rcu_batches_completed_bh(void)
233 return rcu_bh_state.completed;
235 EXPORT_SYMBOL_GPL(rcu_batches_completed_bh);
238 * Force a quiescent state for RCU BH.
240 void rcu_bh_force_quiescent_state(void)
242 force_quiescent_state(&rcu_bh_state, 0);
244 EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state);
247 * Record the number of times rcutorture tests have been initiated and
248 * terminated. This information allows the debugfs tracing stats to be
249 * correlated to the rcutorture messages, even when the rcutorture module
250 * is being repeatedly loaded and unloaded. In other words, we cannot
251 * store this state in rcutorture itself.
253 void rcutorture_record_test_transition(void)
255 rcutorture_testseq++;
256 rcutorture_vernum = 0;
258 EXPORT_SYMBOL_GPL(rcutorture_record_test_transition);
261 * Record the number of writer passes through the current rcutorture test.
262 * This is also used to correlate debugfs tracing stats with the rcutorture
265 void rcutorture_record_progress(unsigned long vernum)
269 EXPORT_SYMBOL_GPL(rcutorture_record_progress);
272 * Force a quiescent state for RCU-sched.
274 void rcu_sched_force_quiescent_state(void)
276 force_quiescent_state(&rcu_sched_state, 0);
278 EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state);
281 * Does the CPU have callbacks ready to be invoked?
284 cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp)
286 return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL];
290 * Does the current CPU require a yet-as-unscheduled grace period?
293 cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp)
295 return *rdp->nxttail[RCU_DONE_TAIL] && !rcu_gp_in_progress(rsp);
299 * Return the root node of the specified rcu_state structure.
301 static struct rcu_node *rcu_get_root(struct rcu_state *rsp)
303 return &rsp->node[0];
309 * If the specified CPU is offline, tell the caller that it is in
310 * a quiescent state. Otherwise, whack it with a reschedule IPI.
311 * Grace periods can end up waiting on an offline CPU when that
312 * CPU is in the process of coming online -- it will be added to the
313 * rcu_node bitmasks before it actually makes it online. The same thing
314 * can happen while a CPU is in the process of coming online. Because this
315 * race is quite rare, we check for it after detecting that the grace
316 * period has been delayed rather than checking each and every CPU
317 * each and every time we start a new grace period.
319 static int rcu_implicit_offline_qs(struct rcu_data *rdp)
322 * If the CPU is offline, it is in a quiescent state. We can
323 * trust its state not to change because interrupts are disabled.
325 if (cpu_is_offline(rdp->cpu)) {
326 trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, "ofl");
331 /* If preemptible RCU, no point in sending reschedule IPI. */
332 if (rdp->preemptible)
335 /* The CPU is online, so send it a reschedule IPI. */
336 if (rdp->cpu != smp_processor_id())
337 smp_send_reschedule(rdp->cpu);
344 #endif /* #ifdef CONFIG_SMP */
349 * rcu_enter_nohz - inform RCU that current CPU is entering nohz
351 * Enter nohz mode, in other words, -leave- the mode in which RCU
352 * read-side critical sections can occur. (Though RCU read-side
353 * critical sections can occur in irq handlers in nohz mode, a possibility
354 * handled by rcu_irq_enter() and rcu_irq_exit()).
356 void rcu_enter_nohz(void)
359 struct rcu_dynticks *rdtp;
361 local_irq_save(flags);
362 rdtp = &__get_cpu_var(rcu_dynticks);
363 if (--rdtp->dynticks_nesting) {
364 local_irq_restore(flags);
367 trace_rcu_dyntick("Start");
368 /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */
369 smp_mb__before_atomic_inc(); /* See above. */
370 atomic_inc(&rdtp->dynticks);
371 smp_mb__after_atomic_inc(); /* Force ordering with next sojourn. */
372 WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1);
373 local_irq_restore(flags);
375 /* If the interrupt queued a callback, get out of dyntick mode. */
377 (__get_cpu_var(rcu_sched_data).nxtlist ||
378 __get_cpu_var(rcu_bh_data).nxtlist ||
379 rcu_preempt_needs_cpu(smp_processor_id())))
384 * rcu_exit_nohz - inform RCU that current CPU is leaving nohz
386 * Exit nohz mode, in other words, -enter- the mode in which RCU
387 * read-side critical sections normally occur.
389 void rcu_exit_nohz(void)
392 struct rcu_dynticks *rdtp;
394 local_irq_save(flags);
395 rdtp = &__get_cpu_var(rcu_dynticks);
396 if (rdtp->dynticks_nesting++) {
397 local_irq_restore(flags);
400 smp_mb__before_atomic_inc(); /* Force ordering w/previous sojourn. */
401 atomic_inc(&rdtp->dynticks);
402 /* CPUs seeing atomic_inc() must see later RCU read-side crit sects */
403 smp_mb__after_atomic_inc(); /* See above. */
404 WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1));
405 trace_rcu_dyntick("End");
406 local_irq_restore(flags);
410 * rcu_nmi_enter - inform RCU of entry to NMI context
412 * If the CPU was idle with dynamic ticks active, and there is no
413 * irq handler running, this updates rdtp->dynticks_nmi to let the
414 * RCU grace-period handling know that the CPU is active.
416 void rcu_nmi_enter(void)
418 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
420 if (rdtp->dynticks_nmi_nesting == 0 &&
421 (atomic_read(&rdtp->dynticks) & 0x1))
423 rdtp->dynticks_nmi_nesting++;
424 smp_mb__before_atomic_inc(); /* Force delay from prior write. */
425 atomic_inc(&rdtp->dynticks);
426 /* CPUs seeing atomic_inc() must see later RCU read-side crit sects */
427 smp_mb__after_atomic_inc(); /* See above. */
428 WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1));
432 * rcu_nmi_exit - inform RCU of exit from NMI context
434 * If the CPU was idle with dynamic ticks active, and there is no
435 * irq handler running, this updates rdtp->dynticks_nmi to let the
436 * RCU grace-period handling know that the CPU is no longer active.
438 void rcu_nmi_exit(void)
440 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
442 if (rdtp->dynticks_nmi_nesting == 0 ||
443 --rdtp->dynticks_nmi_nesting != 0)
445 /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */
446 smp_mb__before_atomic_inc(); /* See above. */
447 atomic_inc(&rdtp->dynticks);
448 smp_mb__after_atomic_inc(); /* Force delay to next write. */
449 WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1);
453 * rcu_irq_enter - inform RCU of entry to hard irq context
455 * If the CPU was idle with dynamic ticks active, this updates the
456 * rdtp->dynticks to let the RCU handling know that the CPU is active.
458 void rcu_irq_enter(void)
464 * rcu_irq_exit - inform RCU of exit from hard irq context
466 * If the CPU was idle with dynamic ticks active, update the rdp->dynticks
467 * to put let the RCU handling be aware that the CPU is going back to idle
470 void rcu_irq_exit(void)
478 * Snapshot the specified CPU's dynticks counter so that we can later
479 * credit them with an implicit quiescent state. Return 1 if this CPU
480 * is in dynticks idle mode, which is an extended quiescent state.
482 static int dyntick_save_progress_counter(struct rcu_data *rdp)
484 rdp->dynticks_snap = atomic_add_return(0, &rdp->dynticks->dynticks);
489 * Return true if the specified CPU has passed through a quiescent
490 * state by virtue of being in or having passed through an dynticks
491 * idle state since the last call to dyntick_save_progress_counter()
494 static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
499 curr = (unsigned long)atomic_add_return(0, &rdp->dynticks->dynticks);
500 snap = (unsigned long)rdp->dynticks_snap;
503 * If the CPU passed through or entered a dynticks idle phase with
504 * no active irq/NMI handlers, then we can safely pretend that the CPU
505 * already acknowledged the request to pass through a quiescent
506 * state. Either way, that CPU cannot possibly be in an RCU
507 * read-side critical section that started before the beginning
508 * of the current RCU grace period.
510 if ((curr & 0x1) == 0 || ULONG_CMP_GE(curr, snap + 2)) {
511 trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, "dti");
516 /* Go check for the CPU being offline. */
517 return rcu_implicit_offline_qs(rdp);
520 #endif /* #ifdef CONFIG_SMP */
522 #else /* #ifdef CONFIG_NO_HZ */
526 static int dyntick_save_progress_counter(struct rcu_data *rdp)
531 static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
533 return rcu_implicit_offline_qs(rdp);
536 #endif /* #ifdef CONFIG_SMP */
538 #endif /* #else #ifdef CONFIG_NO_HZ */
540 int rcu_cpu_stall_suppress __read_mostly;
542 static void record_gp_stall_check_time(struct rcu_state *rsp)
544 rsp->gp_start = jiffies;
545 rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK;
548 static void print_other_cpu_stall(struct rcu_state *rsp)
553 struct rcu_node *rnp = rcu_get_root(rsp);
555 /* Only let one CPU complain about others per time interval. */
557 raw_spin_lock_irqsave(&rnp->lock, flags);
558 delta = jiffies - rsp->jiffies_stall;
559 if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) {
560 raw_spin_unlock_irqrestore(&rnp->lock, flags);
563 rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
566 * Now rat on any tasks that got kicked up to the root rcu_node
567 * due to CPU offlining.
569 rcu_print_task_stall(rnp);
570 raw_spin_unlock_irqrestore(&rnp->lock, flags);
573 * OK, time to rat on our buddy...
574 * See Documentation/RCU/stallwarn.txt for info on how to debug
575 * RCU CPU stall warnings.
577 printk(KERN_ERR "INFO: %s detected stalls on CPUs/tasks: {",
579 rcu_for_each_leaf_node(rsp, rnp) {
580 raw_spin_lock_irqsave(&rnp->lock, flags);
581 rcu_print_task_stall(rnp);
582 raw_spin_unlock_irqrestore(&rnp->lock, flags);
583 if (rnp->qsmask == 0)
585 for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++)
586 if (rnp->qsmask & (1UL << cpu))
587 printk(" %d", rnp->grplo + cpu);
589 printk("} (detected by %d, t=%ld jiffies)\n",
590 smp_processor_id(), (long)(jiffies - rsp->gp_start));
591 trigger_all_cpu_backtrace();
593 /* If so configured, complain about tasks blocking the grace period. */
595 rcu_print_detail_task_stall(rsp);
597 force_quiescent_state(rsp, 0); /* Kick them all. */
600 static void print_cpu_stall(struct rcu_state *rsp)
603 struct rcu_node *rnp = rcu_get_root(rsp);
606 * OK, time to rat on ourselves...
607 * See Documentation/RCU/stallwarn.txt for info on how to debug
608 * RCU CPU stall warnings.
610 printk(KERN_ERR "INFO: %s detected stall on CPU %d (t=%lu jiffies)\n",
611 rsp->name, smp_processor_id(), jiffies - rsp->gp_start);
612 trigger_all_cpu_backtrace();
614 raw_spin_lock_irqsave(&rnp->lock, flags);
615 if (ULONG_CMP_GE(jiffies, rsp->jiffies_stall))
617 jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
618 raw_spin_unlock_irqrestore(&rnp->lock, flags);
620 set_need_resched(); /* kick ourselves to get things going. */
623 static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
627 struct rcu_node *rnp;
629 if (rcu_cpu_stall_suppress)
631 j = ACCESS_ONCE(jiffies);
632 js = ACCESS_ONCE(rsp->jiffies_stall);
634 if ((ACCESS_ONCE(rnp->qsmask) & rdp->grpmask) && ULONG_CMP_GE(j, js)) {
636 /* We haven't checked in, so go dump stack. */
637 print_cpu_stall(rsp);
639 } else if (rcu_gp_in_progress(rsp) &&
640 ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY)) {
642 /* They had a few time units to dump stack, so complain. */
643 print_other_cpu_stall(rsp);
647 static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr)
649 rcu_cpu_stall_suppress = 1;
654 * rcu_cpu_stall_reset - prevent further stall warnings in current grace period
656 * Set the stall-warning timeout way off into the future, thus preventing
657 * any RCU CPU stall-warning messages from appearing in the current set of
660 * The caller must disable hard irqs.
662 void rcu_cpu_stall_reset(void)
664 rcu_sched_state.jiffies_stall = jiffies + ULONG_MAX / 2;
665 rcu_bh_state.jiffies_stall = jiffies + ULONG_MAX / 2;
666 rcu_preempt_stall_reset();
669 static struct notifier_block rcu_panic_block = {
670 .notifier_call = rcu_panic,
673 static void __init check_cpu_stall_init(void)
675 atomic_notifier_chain_register(&panic_notifier_list, &rcu_panic_block);
679 * Update CPU-local rcu_data state to record the newly noticed grace period.
680 * This is used both when we started the grace period and when we notice
681 * that someone else started the grace period. The caller must hold the
682 * ->lock of the leaf rcu_node structure corresponding to the current CPU,
683 * and must have irqs disabled.
685 static void __note_new_gpnum(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp)
687 if (rdp->gpnum != rnp->gpnum) {
689 * If the current grace period is waiting for this CPU,
690 * set up to detect a quiescent state, otherwise don't
691 * go looking for one.
693 rdp->gpnum = rnp->gpnum;
694 trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpustart");
695 if (rnp->qsmask & rdp->grpmask) {
697 rdp->passed_quiesc = 0;
703 static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp)
706 struct rcu_node *rnp;
708 local_irq_save(flags);
710 if (rdp->gpnum == ACCESS_ONCE(rnp->gpnum) || /* outside lock. */
711 !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */
712 local_irq_restore(flags);
715 __note_new_gpnum(rsp, rnp, rdp);
716 raw_spin_unlock_irqrestore(&rnp->lock, flags);
720 * Did someone else start a new RCU grace period start since we last
721 * checked? Update local state appropriately if so. Must be called
722 * on the CPU corresponding to rdp.
725 check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp)
730 local_irq_save(flags);
731 if (rdp->gpnum != rsp->gpnum) {
732 note_new_gpnum(rsp, rdp);
735 local_irq_restore(flags);
740 * Advance this CPU's callbacks, but only if the current grace period
741 * has ended. This may be called only from the CPU to whom the rdp
742 * belongs. In addition, the corresponding leaf rcu_node structure's
743 * ->lock must be held by the caller, with irqs disabled.
746 __rcu_process_gp_end(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp)
748 /* Did another grace period end? */
749 if (rdp->completed != rnp->completed) {
751 /* Advance callbacks. No harm if list empty. */
752 rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL];
753 rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL];
754 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
756 /* Remember that we saw this grace-period completion. */
757 rdp->completed = rnp->completed;
758 trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpuend");
761 * If we were in an extended quiescent state, we may have
762 * missed some grace periods that others CPUs handled on
763 * our behalf. Catch up with this state to avoid noting
764 * spurious new grace periods. If another grace period
765 * has started, then rnp->gpnum will have advanced, so
766 * we will detect this later on.
768 if (ULONG_CMP_LT(rdp->gpnum, rdp->completed))
769 rdp->gpnum = rdp->completed;
772 * If RCU does not need a quiescent state from this CPU,
773 * then make sure that this CPU doesn't go looking for one.
775 if ((rnp->qsmask & rdp->grpmask) == 0)
781 * Advance this CPU's callbacks, but only if the current grace period
782 * has ended. This may be called only from the CPU to whom the rdp
786 rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp)
789 struct rcu_node *rnp;
791 local_irq_save(flags);
793 if (rdp->completed == ACCESS_ONCE(rnp->completed) || /* outside lock. */
794 !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */
795 local_irq_restore(flags);
798 __rcu_process_gp_end(rsp, rnp, rdp);
799 raw_spin_unlock_irqrestore(&rnp->lock, flags);
803 * Do per-CPU grace-period initialization for running CPU. The caller
804 * must hold the lock of the leaf rcu_node structure corresponding to
808 rcu_start_gp_per_cpu(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp)
810 /* Prior grace period ended, so advance callbacks for current CPU. */
811 __rcu_process_gp_end(rsp, rnp, rdp);
814 * Because this CPU just now started the new grace period, we know
815 * that all of its callbacks will be covered by this upcoming grace
816 * period, even the ones that were registered arbitrarily recently.
817 * Therefore, advance all outstanding callbacks to RCU_WAIT_TAIL.
819 * Other CPUs cannot be sure exactly when the grace period started.
820 * Therefore, their recently registered callbacks must pass through
821 * an additional RCU_NEXT_READY stage, so that they will be handled
822 * by the next RCU grace period.
824 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
825 rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
827 /* Set state so that this CPU will detect the next quiescent state. */
828 __note_new_gpnum(rsp, rnp, rdp);
832 * Start a new RCU grace period if warranted, re-initializing the hierarchy
833 * in preparation for detecting the next grace period. The caller must hold
834 * the root node's ->lock, which is released before return. Hard irqs must
838 rcu_start_gp(struct rcu_state *rsp, unsigned long flags)
839 __releases(rcu_get_root(rsp)->lock)
841 struct rcu_data *rdp = this_cpu_ptr(rsp->rda);
842 struct rcu_node *rnp = rcu_get_root(rsp);
844 if (!cpu_needs_another_gp(rsp, rdp) || rsp->fqs_active) {
845 if (cpu_needs_another_gp(rsp, rdp))
846 rsp->fqs_need_gp = 1;
847 if (rnp->completed == rsp->completed) {
848 raw_spin_unlock_irqrestore(&rnp->lock, flags);
851 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
854 * Propagate new ->completed value to rcu_node structures
855 * so that other CPUs don't have to wait until the start
856 * of the next grace period to process their callbacks.
858 rcu_for_each_node_breadth_first(rsp, rnp) {
859 raw_spin_lock(&rnp->lock); /* irqs already disabled. */
860 rnp->completed = rsp->completed;
861 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
863 local_irq_restore(flags);
867 /* Advance to a new grace period and initialize state. */
869 trace_rcu_grace_period(rsp->name, rsp->gpnum, "start");
870 WARN_ON_ONCE(rsp->signaled == RCU_GP_INIT);
871 rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */
872 rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
873 record_gp_stall_check_time(rsp);
875 /* Special-case the common single-level case. */
876 if (NUM_RCU_NODES == 1) {
877 rcu_preempt_check_blocked_tasks(rnp);
878 rnp->qsmask = rnp->qsmaskinit;
879 rnp->gpnum = rsp->gpnum;
880 rnp->completed = rsp->completed;
881 rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */
882 rcu_start_gp_per_cpu(rsp, rnp, rdp);
883 rcu_preempt_boost_start_gp(rnp);
884 trace_rcu_grace_period_init(rsp->name, rnp->gpnum,
885 rnp->level, rnp->grplo,
886 rnp->grphi, rnp->qsmask);
887 raw_spin_unlock_irqrestore(&rnp->lock, flags);
891 raw_spin_unlock(&rnp->lock); /* leave irqs disabled. */
894 /* Exclude any concurrent CPU-hotplug operations. */
895 raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */
898 * Set the quiescent-state-needed bits in all the rcu_node
899 * structures for all currently online CPUs in breadth-first
900 * order, starting from the root rcu_node structure. This
901 * operation relies on the layout of the hierarchy within the
902 * rsp->node[] array. Note that other CPUs will access only
903 * the leaves of the hierarchy, which still indicate that no
904 * grace period is in progress, at least until the corresponding
905 * leaf node has been initialized. In addition, we have excluded
906 * CPU-hotplug operations.
908 * Note that the grace period cannot complete until we finish
909 * the initialization process, as there will be at least one
910 * qsmask bit set in the root node until that time, namely the
911 * one corresponding to this CPU, due to the fact that we have
914 rcu_for_each_node_breadth_first(rsp, rnp) {
915 raw_spin_lock(&rnp->lock); /* irqs already disabled. */
916 rcu_preempt_check_blocked_tasks(rnp);
917 rnp->qsmask = rnp->qsmaskinit;
918 rnp->gpnum = rsp->gpnum;
919 rnp->completed = rsp->completed;
920 if (rnp == rdp->mynode)
921 rcu_start_gp_per_cpu(rsp, rnp, rdp);
922 rcu_preempt_boost_start_gp(rnp);
923 trace_rcu_grace_period_init(rsp->name, rnp->gpnum,
924 rnp->level, rnp->grplo,
925 rnp->grphi, rnp->qsmask);
926 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
929 rnp = rcu_get_root(rsp);
930 raw_spin_lock(&rnp->lock); /* irqs already disabled. */
931 rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */
932 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
933 raw_spin_unlock_irqrestore(&rsp->onofflock, flags);
937 * Report a full set of quiescent states to the specified rcu_state
938 * data structure. This involves cleaning up after the prior grace
939 * period and letting rcu_start_gp() start up the next grace period
940 * if one is needed. Note that the caller must hold rnp->lock, as
941 * required by rcu_start_gp(), which will release it.
943 static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags)
944 __releases(rcu_get_root(rsp)->lock)
946 unsigned long gp_duration;
948 WARN_ON_ONCE(!rcu_gp_in_progress(rsp));
951 * Ensure that all grace-period and pre-grace-period activity
952 * is seen before the assignment to rsp->completed.
954 smp_mb(); /* See above block comment. */
955 gp_duration = jiffies - rsp->gp_start;
956 if (gp_duration > rsp->gp_max)
957 rsp->gp_max = gp_duration;
958 rsp->completed = rsp->gpnum;
959 trace_rcu_grace_period(rsp->name, rsp->completed, "end");
960 rsp->signaled = RCU_GP_IDLE;
961 rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */
965 * Similar to rcu_report_qs_rdp(), for which it is a helper function.
966 * Allows quiescent states for a group of CPUs to be reported at one go
967 * to the specified rcu_node structure, though all the CPUs in the group
968 * must be represented by the same rcu_node structure (which need not be
969 * a leaf rcu_node structure, though it often will be). That structure's
970 * lock must be held upon entry, and it is released before return.
973 rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp,
974 struct rcu_node *rnp, unsigned long flags)
975 __releases(rnp->lock)
977 struct rcu_node *rnp_c;
979 /* Walk up the rcu_node hierarchy. */
981 if (!(rnp->qsmask & mask)) {
983 /* Our bit has already been cleared, so done. */
984 raw_spin_unlock_irqrestore(&rnp->lock, flags);
987 rnp->qsmask &= ~mask;
988 trace_rcu_quiescent_state_report(rsp->name, rnp->gpnum,
989 mask, rnp->qsmask, rnp->level,
990 rnp->grplo, rnp->grphi,
992 if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) {
994 /* Other bits still set at this level, so done. */
995 raw_spin_unlock_irqrestore(&rnp->lock, flags);
999 if (rnp->parent == NULL) {
1001 /* No more levels. Exit loop holding root lock. */
1005 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1008 raw_spin_lock_irqsave(&rnp->lock, flags);
1009 WARN_ON_ONCE(rnp_c->qsmask);
1013 * Get here if we are the last CPU to pass through a quiescent
1014 * state for this grace period. Invoke rcu_report_qs_rsp()
1015 * to clean up and start the next grace period if one is needed.
1017 rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */
1021 * Record a quiescent state for the specified CPU to that CPU's rcu_data
1022 * structure. This must be either called from the specified CPU, or
1023 * called when the specified CPU is known to be offline (and when it is
1024 * also known that no other CPU is concurrently trying to help the offline
1025 * CPU). The lastcomp argument is used to make sure we are still in the
1026 * grace period of interest. We don't want to end the current grace period
1027 * based on quiescent states detected in an earlier grace period!
1030 rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp)
1032 unsigned long flags;
1034 struct rcu_node *rnp;
1037 raw_spin_lock_irqsave(&rnp->lock, flags);
1038 if (lastcomp != rnp->completed) {
1041 * Someone beat us to it for this grace period, so leave.
1042 * The race with GP start is resolved by the fact that we
1043 * hold the leaf rcu_node lock, so that the per-CPU bits
1044 * cannot yet be initialized -- so we would simply find our
1045 * CPU's bit already cleared in rcu_report_qs_rnp() if this
1048 rdp->passed_quiesc = 0; /* try again later! */
1049 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1052 mask = rdp->grpmask;
1053 if ((rnp->qsmask & mask) == 0) {
1054 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1056 rdp->qs_pending = 0;
1059 * This GP can't end until cpu checks in, so all of our
1060 * callbacks can be processed during the next GP.
1062 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
1064 rcu_report_qs_rnp(mask, rsp, rnp, flags); /* rlses rnp->lock */
1069 * Check to see if there is a new grace period of which this CPU
1070 * is not yet aware, and if so, set up local rcu_data state for it.
1071 * Otherwise, see if this CPU has just passed through its first
1072 * quiescent state for this grace period, and record that fact if so.
1075 rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp)
1077 /* If there is now a new grace period, record and return. */
1078 if (check_for_new_grace_period(rsp, rdp))
1082 * Does this CPU still need to do its part for current grace period?
1083 * If no, return and let the other CPUs do their part as well.
1085 if (!rdp->qs_pending)
1089 * Was there a quiescent state since the beginning of the grace
1090 * period? If no, then exit and wait for the next call.
1092 if (!rdp->passed_quiesc)
1096 * Tell RCU we are done (but rcu_report_qs_rdp() will be the
1099 rcu_report_qs_rdp(rdp->cpu, rsp, rdp, rdp->passed_quiesc_completed);
1102 #ifdef CONFIG_HOTPLUG_CPU
1105 * Move a dying CPU's RCU callbacks to online CPU's callback list.
1106 * Synchronization is not required because this function executes
1107 * in stop_machine() context.
1109 static void rcu_send_cbs_to_online(struct rcu_state *rsp)
1112 /* current DYING CPU is cleared in the cpu_online_mask */
1113 int receive_cpu = cpumask_any(cpu_online_mask);
1114 struct rcu_data *rdp = this_cpu_ptr(rsp->rda);
1115 struct rcu_data *receive_rdp = per_cpu_ptr(rsp->rda, receive_cpu);
1117 if (rdp->nxtlist == NULL)
1118 return; /* irqs disabled, so comparison is stable. */
1120 *receive_rdp->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist;
1121 receive_rdp->nxttail[RCU_NEXT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
1122 receive_rdp->qlen += rdp->qlen;
1123 receive_rdp->n_cbs_adopted += rdp->qlen;
1124 rdp->n_cbs_orphaned += rdp->qlen;
1126 rdp->nxtlist = NULL;
1127 for (i = 0; i < RCU_NEXT_SIZE; i++)
1128 rdp->nxttail[i] = &rdp->nxtlist;
1133 * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy
1134 * and move all callbacks from the outgoing CPU to the current one.
1135 * There can only be one CPU hotplug operation at a time, so no other
1136 * CPU can be attempting to update rcu_cpu_kthread_task.
1138 static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp)
1140 unsigned long flags;
1142 int need_report = 0;
1143 struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
1144 struct rcu_node *rnp;
1146 rcu_stop_cpu_kthread(cpu);
1148 /* Exclude any attempts to start a new grace period. */
1149 raw_spin_lock_irqsave(&rsp->onofflock, flags);
1151 /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */
1152 rnp = rdp->mynode; /* this is the outgoing CPU's rnp. */
1153 mask = rdp->grpmask; /* rnp->grplo is constant. */
1155 raw_spin_lock(&rnp->lock); /* irqs already disabled. */
1156 rnp->qsmaskinit &= ~mask;
1157 if (rnp->qsmaskinit != 0) {
1158 if (rnp != rdp->mynode)
1159 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
1161 trace_rcu_grace_period(rsp->name,
1163 !!(rnp->qsmask & mask),
1167 if (rnp == rdp->mynode) {
1168 trace_rcu_grace_period(rsp->name,
1170 !!(rnp->qsmask & mask),
1172 need_report = rcu_preempt_offline_tasks(rsp, rnp, rdp);
1174 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
1175 mask = rnp->grpmask;
1177 } while (rnp != NULL);
1180 * We still hold the leaf rcu_node structure lock here, and
1181 * irqs are still disabled. The reason for this subterfuge is
1182 * because invoking rcu_report_unblock_qs_rnp() with ->onofflock
1183 * held leads to deadlock.
1185 raw_spin_unlock(&rsp->onofflock); /* irqs remain disabled. */
1187 if (need_report & RCU_OFL_TASKS_NORM_GP)
1188 rcu_report_unblock_qs_rnp(rnp, flags);
1190 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1191 if (need_report & RCU_OFL_TASKS_EXP_GP)
1192 rcu_report_exp_rnp(rsp, rnp);
1193 rcu_node_kthread_setaffinity(rnp, -1);
1197 * Remove the specified CPU from the RCU hierarchy and move any pending
1198 * callbacks that it might have to the current CPU. This code assumes
1199 * that at least one CPU in the system will remain running at all times.
1200 * Any attempt to offline -all- CPUs is likely to strand RCU callbacks.
1202 static void rcu_offline_cpu(int cpu)
1204 __rcu_offline_cpu(cpu, &rcu_sched_state);
1205 __rcu_offline_cpu(cpu, &rcu_bh_state);
1206 rcu_preempt_offline_cpu(cpu);
1209 #else /* #ifdef CONFIG_HOTPLUG_CPU */
1211 static void rcu_send_cbs_to_online(struct rcu_state *rsp)
1215 static void rcu_offline_cpu(int cpu)
1219 #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */
1222 * Invoke any RCU callbacks that have made it to the end of their grace
1223 * period. Thottle as specified by rdp->blimit.
1225 static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp)
1227 unsigned long flags;
1228 struct rcu_head *next, *list, **tail;
1231 /* If no callbacks are ready, just return.*/
1232 if (!cpu_has_callbacks_ready_to_invoke(rdp)) {
1233 trace_rcu_batch_start(rsp->name, 0, 0);
1234 trace_rcu_batch_end(rsp->name, 0);
1239 * Extract the list of ready callbacks, disabling to prevent
1240 * races with call_rcu() from interrupt handlers.
1242 local_irq_save(flags);
1244 trace_rcu_batch_start(rsp->name, rdp->qlen, bl);
1245 list = rdp->nxtlist;
1246 rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL];
1247 *rdp->nxttail[RCU_DONE_TAIL] = NULL;
1248 tail = rdp->nxttail[RCU_DONE_TAIL];
1249 for (count = RCU_NEXT_SIZE - 1; count >= 0; count--)
1250 if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL])
1251 rdp->nxttail[count] = &rdp->nxtlist;
1252 local_irq_restore(flags);
1254 /* Invoke callbacks. */
1259 debug_rcu_head_unqueue(list);
1260 __rcu_reclaim(rsp->name, list);
1266 local_irq_save(flags);
1267 trace_rcu_batch_end(rsp->name, count);
1269 /* Update count, and requeue any remaining callbacks. */
1271 rdp->n_cbs_invoked += count;
1273 *tail = rdp->nxtlist;
1274 rdp->nxtlist = list;
1275 for (count = 0; count < RCU_NEXT_SIZE; count++)
1276 if (&rdp->nxtlist == rdp->nxttail[count])
1277 rdp->nxttail[count] = tail;
1282 /* Reinstate batch limit if we have worked down the excess. */
1283 if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark)
1284 rdp->blimit = blimit;
1286 /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */
1287 if (rdp->qlen == 0 && rdp->qlen_last_fqs_check != 0) {
1288 rdp->qlen_last_fqs_check = 0;
1289 rdp->n_force_qs_snap = rsp->n_force_qs;
1290 } else if (rdp->qlen < rdp->qlen_last_fqs_check - qhimark)
1291 rdp->qlen_last_fqs_check = rdp->qlen;
1293 local_irq_restore(flags);
1295 /* Re-invoke RCU core processing if there are callbacks remaining. */
1296 if (cpu_has_callbacks_ready_to_invoke(rdp))
1301 * Check to see if this CPU is in a non-context-switch quiescent state
1302 * (user mode or idle loop for rcu, non-softirq execution for rcu_bh).
1303 * Also schedule RCU core processing.
1305 * This function must be called with hardirqs disabled. It is normally
1306 * invoked from the scheduling-clock interrupt. If rcu_pending returns
1307 * false, there is no point in invoking rcu_check_callbacks().
1309 void rcu_check_callbacks(int cpu, int user)
1311 trace_rcu_utilization("Start scheduler-tick");
1313 (idle_cpu(cpu) && rcu_scheduler_active &&
1314 !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) {
1317 * Get here if this CPU took its interrupt from user
1318 * mode or from the idle loop, and if this is not a
1319 * nested interrupt. In this case, the CPU is in
1320 * a quiescent state, so note it.
1322 * No memory barrier is required here because both
1323 * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local
1324 * variables that other CPUs neither access nor modify,
1325 * at least not while the corresponding CPU is online.
1331 } else if (!in_softirq()) {
1334 * Get here if this CPU did not take its interrupt from
1335 * softirq, in other words, if it is not interrupting
1336 * a rcu_bh read-side critical section. This is an _bh
1337 * critical section, so note it.
1342 rcu_preempt_check_callbacks(cpu);
1343 if (rcu_pending(cpu))
1345 trace_rcu_utilization("End scheduler-tick");
1351 * Scan the leaf rcu_node structures, processing dyntick state for any that
1352 * have not yet encountered a quiescent state, using the function specified.
1353 * Also initiate boosting for any threads blocked on the root rcu_node.
1355 * The caller must have suppressed start of new grace periods.
1357 static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *))
1361 unsigned long flags;
1363 struct rcu_node *rnp;
1365 rcu_for_each_leaf_node(rsp, rnp) {
1367 raw_spin_lock_irqsave(&rnp->lock, flags);
1368 if (!rcu_gp_in_progress(rsp)) {
1369 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1372 if (rnp->qsmask == 0) {
1373 rcu_initiate_boost(rnp, flags); /* releases rnp->lock */
1378 for (; cpu <= rnp->grphi; cpu++, bit <<= 1) {
1379 if ((rnp->qsmask & bit) != 0 &&
1380 f(per_cpu_ptr(rsp->rda, cpu)))
1385 /* rcu_report_qs_rnp() releases rnp->lock. */
1386 rcu_report_qs_rnp(mask, rsp, rnp, flags);
1389 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1391 rnp = rcu_get_root(rsp);
1392 if (rnp->qsmask == 0) {
1393 raw_spin_lock_irqsave(&rnp->lock, flags);
1394 rcu_initiate_boost(rnp, flags); /* releases rnp->lock. */
1399 * Force quiescent states on reluctant CPUs, and also detect which
1400 * CPUs are in dyntick-idle mode.
1402 static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1404 unsigned long flags;
1405 struct rcu_node *rnp = rcu_get_root(rsp);
1407 trace_rcu_utilization("Start fqs");
1408 if (!rcu_gp_in_progress(rsp)) {
1409 trace_rcu_utilization("End fqs");
1410 return; /* No grace period in progress, nothing to force. */
1412 if (!raw_spin_trylock_irqsave(&rsp->fqslock, flags)) {
1413 rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */
1414 trace_rcu_utilization("End fqs");
1415 return; /* Someone else is already on the job. */
1417 if (relaxed && ULONG_CMP_GE(rsp->jiffies_force_qs, jiffies))
1418 goto unlock_fqs_ret; /* no emergency and done recently. */
1420 raw_spin_lock(&rnp->lock); /* irqs already disabled */
1421 rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
1422 if(!rcu_gp_in_progress(rsp)) {
1423 rsp->n_force_qs_ngp++;
1424 raw_spin_unlock(&rnp->lock); /* irqs remain disabled */
1425 goto unlock_fqs_ret; /* no GP in progress, time updated. */
1427 rsp->fqs_active = 1;
1428 switch (rsp->signaled) {
1432 break; /* grace period idle or initializing, ignore. */
1434 case RCU_SAVE_DYNTICK:
1435 if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK)
1436 break; /* So gcc recognizes the dead code. */
1438 raw_spin_unlock(&rnp->lock); /* irqs remain disabled */
1440 /* Record dyntick-idle state. */
1441 force_qs_rnp(rsp, dyntick_save_progress_counter);
1442 raw_spin_lock(&rnp->lock); /* irqs already disabled */
1443 if (rcu_gp_in_progress(rsp))
1444 rsp->signaled = RCU_FORCE_QS;
1449 /* Check dyntick-idle state, send IPI to laggarts. */
1450 raw_spin_unlock(&rnp->lock); /* irqs remain disabled */
1451 force_qs_rnp(rsp, rcu_implicit_dynticks_qs);
1453 /* Leave state in case more forcing is required. */
1455 raw_spin_lock(&rnp->lock); /* irqs already disabled */
1458 rsp->fqs_active = 0;
1459 if (rsp->fqs_need_gp) {
1460 raw_spin_unlock(&rsp->fqslock); /* irqs remain disabled */
1461 rsp->fqs_need_gp = 0;
1462 rcu_start_gp(rsp, flags); /* releases rnp->lock */
1463 trace_rcu_utilization("End fqs");
1466 raw_spin_unlock(&rnp->lock); /* irqs remain disabled */
1468 raw_spin_unlock_irqrestore(&rsp->fqslock, flags);
1469 trace_rcu_utilization("End fqs");
1472 #else /* #ifdef CONFIG_SMP */
1474 static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1479 #endif /* #else #ifdef CONFIG_SMP */
1482 * This does the RCU core processing work for the specified rcu_state
1483 * and rcu_data structures. This may be called only from the CPU to
1484 * whom the rdp belongs.
1487 __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp)
1489 unsigned long flags;
1491 WARN_ON_ONCE(rdp->beenonline == 0);
1494 * If an RCU GP has gone long enough, go check for dyntick
1495 * idle CPUs and, if needed, send resched IPIs.
1497 if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies))
1498 force_quiescent_state(rsp, 1);
1501 * Advance callbacks in response to end of earlier grace
1502 * period that some other CPU ended.
1504 rcu_process_gp_end(rsp, rdp);
1506 /* Update RCU state based on any recent quiescent states. */
1507 rcu_check_quiescent_state(rsp, rdp);
1509 /* Does this CPU require a not-yet-started grace period? */
1510 if (cpu_needs_another_gp(rsp, rdp)) {
1511 raw_spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags);
1512 rcu_start_gp(rsp, flags); /* releases above lock */
1515 /* If there are callbacks ready, invoke them. */
1516 if (cpu_has_callbacks_ready_to_invoke(rdp))
1517 invoke_rcu_callbacks(rsp, rdp);
1521 * Do RCU core processing for the current CPU.
1523 static void rcu_process_callbacks(struct softirq_action *unused)
1525 trace_rcu_utilization("Start RCU core");
1526 __rcu_process_callbacks(&rcu_sched_state,
1527 &__get_cpu_var(rcu_sched_data));
1528 __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data));
1529 rcu_preempt_process_callbacks();
1531 /* If we are last CPU on way to dyntick-idle mode, accelerate it. */
1532 rcu_needs_cpu_flush();
1533 trace_rcu_utilization("End RCU core");
1537 * Schedule RCU callback invocation. If the specified type of RCU
1538 * does not support RCU priority boosting, just do a direct call,
1539 * otherwise wake up the per-CPU kernel kthread. Note that because we
1540 * are running on the current CPU with interrupts disabled, the
1541 * rcu_cpu_kthread_task cannot disappear out from under us.
1543 static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp)
1545 if (unlikely(!ACCESS_ONCE(rcu_scheduler_fully_active)))
1547 if (likely(!rsp->boost)) {
1548 rcu_do_batch(rsp, rdp);
1551 invoke_rcu_callbacks_kthread();
1554 static void invoke_rcu_core(void)
1556 raise_softirq(RCU_SOFTIRQ);
1560 __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
1561 struct rcu_state *rsp)
1563 unsigned long flags;
1564 struct rcu_data *rdp;
1566 debug_rcu_head_queue(head);
1570 smp_mb(); /* Ensure RCU update seen before callback registry. */
1573 * Opportunistically note grace-period endings and beginnings.
1574 * Note that we might see a beginning right after we see an
1575 * end, but never vice versa, since this CPU has to pass through
1576 * a quiescent state betweentimes.
1578 local_irq_save(flags);
1579 rdp = this_cpu_ptr(rsp->rda);
1581 /* Add the callback to our list. */
1582 *rdp->nxttail[RCU_NEXT_TAIL] = head;
1583 rdp->nxttail[RCU_NEXT_TAIL] = &head->next;
1586 if (__is_kfree_rcu_offset((unsigned long)func))
1587 trace_rcu_kfree_callback(rsp->name, head, (unsigned long)func,
1590 trace_rcu_callback(rsp->name, head, rdp->qlen);
1592 /* If interrupts were disabled, don't dive into RCU core. */
1593 if (irqs_disabled_flags(flags)) {
1594 local_irq_restore(flags);
1599 * Force the grace period if too many callbacks or too long waiting.
1600 * Enforce hysteresis, and don't invoke force_quiescent_state()
1601 * if some other CPU has recently done so. Also, don't bother
1602 * invoking force_quiescent_state() if the newly enqueued callback
1603 * is the only one waiting for a grace period to complete.
1605 if (unlikely(rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) {
1607 /* Are we ignoring a completed grace period? */
1608 rcu_process_gp_end(rsp, rdp);
1609 check_for_new_grace_period(rsp, rdp);
1611 /* Start a new grace period if one not already started. */
1612 if (!rcu_gp_in_progress(rsp)) {
1613 unsigned long nestflag;
1614 struct rcu_node *rnp_root = rcu_get_root(rsp);
1616 raw_spin_lock_irqsave(&rnp_root->lock, nestflag);
1617 rcu_start_gp(rsp, nestflag); /* rlses rnp_root->lock */
1619 /* Give the grace period a kick. */
1620 rdp->blimit = LONG_MAX;
1621 if (rsp->n_force_qs == rdp->n_force_qs_snap &&
1622 *rdp->nxttail[RCU_DONE_TAIL] != head)
1623 force_quiescent_state(rsp, 0);
1624 rdp->n_force_qs_snap = rsp->n_force_qs;
1625 rdp->qlen_last_fqs_check = rdp->qlen;
1627 } else if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies))
1628 force_quiescent_state(rsp, 1);
1629 local_irq_restore(flags);
1633 * Queue an RCU-sched callback for invocation after a grace period.
1635 void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1637 __call_rcu(head, func, &rcu_sched_state);
1639 EXPORT_SYMBOL_GPL(call_rcu_sched);
1642 * Queue an RCU for invocation after a quicker grace period.
1644 void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1646 __call_rcu(head, func, &rcu_bh_state);
1648 EXPORT_SYMBOL_GPL(call_rcu_bh);
1651 * synchronize_sched - wait until an rcu-sched grace period has elapsed.
1653 * Control will return to the caller some time after a full rcu-sched
1654 * grace period has elapsed, in other words after all currently executing
1655 * rcu-sched read-side critical sections have completed. These read-side
1656 * critical sections are delimited by rcu_read_lock_sched() and
1657 * rcu_read_unlock_sched(), and may be nested. Note that preempt_disable(),
1658 * local_irq_disable(), and so on may be used in place of
1659 * rcu_read_lock_sched().
1661 * This means that all preempt_disable code sequences, including NMI and
1662 * hardware-interrupt handlers, in progress on entry will have completed
1663 * before this primitive returns. However, this does not guarantee that
1664 * softirq handlers will have completed, since in some kernels, these
1665 * handlers can run in process context, and can block.
1667 * This primitive provides the guarantees made by the (now removed)
1668 * synchronize_kernel() API. In contrast, synchronize_rcu() only
1669 * guarantees that rcu_read_lock() sections will have completed.
1670 * In "classic RCU", these two guarantees happen to be one and
1671 * the same, but can differ in realtime RCU implementations.
1673 void synchronize_sched(void)
1675 if (rcu_blocking_is_gp())
1677 wait_rcu_gp(call_rcu_sched);
1679 EXPORT_SYMBOL_GPL(synchronize_sched);
1682 * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed.
1684 * Control will return to the caller some time after a full rcu_bh grace
1685 * period has elapsed, in other words after all currently executing rcu_bh
1686 * read-side critical sections have completed. RCU read-side critical
1687 * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(),
1688 * and may be nested.
1690 void synchronize_rcu_bh(void)
1692 if (rcu_blocking_is_gp())
1694 wait_rcu_gp(call_rcu_bh);
1696 EXPORT_SYMBOL_GPL(synchronize_rcu_bh);
1699 * Check to see if there is any immediate RCU-related work to be done
1700 * by the current CPU, for the specified type of RCU, returning 1 if so.
1701 * The checks are in order of increasing expense: checks that can be
1702 * carried out against CPU-local state are performed first. However,
1703 * we must check for CPU stalls first, else we might not get a chance.
1705 static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp)
1707 struct rcu_node *rnp = rdp->mynode;
1709 rdp->n_rcu_pending++;
1711 /* Check for CPU stalls, if enabled. */
1712 check_cpu_stall(rsp, rdp);
1714 /* Is the RCU core waiting for a quiescent state from this CPU? */
1715 if (rdp->qs_pending && !rdp->passed_quiesc) {
1718 * If force_quiescent_state() coming soon and this CPU
1719 * needs a quiescent state, and this is either RCU-sched
1720 * or RCU-bh, force a local reschedule.
1722 rdp->n_rp_qs_pending++;
1723 if (!rdp->preemptible &&
1724 ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs) - 1,
1727 } else if (rdp->qs_pending && rdp->passed_quiesc) {
1728 rdp->n_rp_report_qs++;
1732 /* Does this CPU have callbacks ready to invoke? */
1733 if (cpu_has_callbacks_ready_to_invoke(rdp)) {
1734 rdp->n_rp_cb_ready++;
1738 /* Has RCU gone idle with this CPU needing another grace period? */
1739 if (cpu_needs_another_gp(rsp, rdp)) {
1740 rdp->n_rp_cpu_needs_gp++;
1744 /* Has another RCU grace period completed? */
1745 if (ACCESS_ONCE(rnp->completed) != rdp->completed) { /* outside lock */
1746 rdp->n_rp_gp_completed++;
1750 /* Has a new RCU grace period started? */
1751 if (ACCESS_ONCE(rnp->gpnum) != rdp->gpnum) { /* outside lock */
1752 rdp->n_rp_gp_started++;
1756 /* Has an RCU GP gone long enough to send resched IPIs &c? */
1757 if (rcu_gp_in_progress(rsp) &&
1758 ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) {
1759 rdp->n_rp_need_fqs++;
1764 rdp->n_rp_need_nothing++;
1769 * Check to see if there is any immediate RCU-related work to be done
1770 * by the current CPU, returning 1 if so. This function is part of the
1771 * RCU implementation; it is -not- an exported member of the RCU API.
1773 static int rcu_pending(int cpu)
1775 return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) ||
1776 __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) ||
1777 rcu_preempt_pending(cpu);
1781 * Check to see if any future RCU-related work will need to be done
1782 * by the current CPU, even if none need be done immediately, returning
1785 static int rcu_needs_cpu_quick_check(int cpu)
1787 /* RCU callbacks either ready or pending? */
1788 return per_cpu(rcu_sched_data, cpu).nxtlist ||
1789 per_cpu(rcu_bh_data, cpu).nxtlist ||
1790 rcu_preempt_needs_cpu(cpu);
1793 static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL};
1794 static atomic_t rcu_barrier_cpu_count;
1795 static DEFINE_MUTEX(rcu_barrier_mutex);
1796 static struct completion rcu_barrier_completion;
1798 static void rcu_barrier_callback(struct rcu_head *notused)
1800 if (atomic_dec_and_test(&rcu_barrier_cpu_count))
1801 complete(&rcu_barrier_completion);
1805 * Called with preemption disabled, and from cross-cpu IRQ context.
1807 static void rcu_barrier_func(void *type)
1809 int cpu = smp_processor_id();
1810 struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu);
1811 void (*call_rcu_func)(struct rcu_head *head,
1812 void (*func)(struct rcu_head *head));
1814 atomic_inc(&rcu_barrier_cpu_count);
1815 call_rcu_func = type;
1816 call_rcu_func(head, rcu_barrier_callback);
1820 * Orchestrate the specified type of RCU barrier, waiting for all
1821 * RCU callbacks of the specified type to complete.
1823 static void _rcu_barrier(struct rcu_state *rsp,
1824 void (*call_rcu_func)(struct rcu_head *head,
1825 void (*func)(struct rcu_head *head)))
1827 BUG_ON(in_interrupt());
1828 /* Take mutex to serialize concurrent rcu_barrier() requests. */
1829 mutex_lock(&rcu_barrier_mutex);
1830 init_completion(&rcu_barrier_completion);
1832 * Initialize rcu_barrier_cpu_count to 1, then invoke
1833 * rcu_barrier_func() on each CPU, so that each CPU also has
1834 * incremented rcu_barrier_cpu_count. Only then is it safe to
1835 * decrement rcu_barrier_cpu_count -- otherwise the first CPU
1836 * might complete its grace period before all of the other CPUs
1837 * did their increment, causing this function to return too
1838 * early. Note that on_each_cpu() disables irqs, which prevents
1839 * any CPUs from coming online or going offline until each online
1840 * CPU has queued its RCU-barrier callback.
1842 atomic_set(&rcu_barrier_cpu_count, 1);
1843 on_each_cpu(rcu_barrier_func, (void *)call_rcu_func, 1);
1844 if (atomic_dec_and_test(&rcu_barrier_cpu_count))
1845 complete(&rcu_barrier_completion);
1846 wait_for_completion(&rcu_barrier_completion);
1847 mutex_unlock(&rcu_barrier_mutex);
1851 * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete.
1853 void rcu_barrier_bh(void)
1855 _rcu_barrier(&rcu_bh_state, call_rcu_bh);
1857 EXPORT_SYMBOL_GPL(rcu_barrier_bh);
1860 * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks.
1862 void rcu_barrier_sched(void)
1864 _rcu_barrier(&rcu_sched_state, call_rcu_sched);
1866 EXPORT_SYMBOL_GPL(rcu_barrier_sched);
1869 * Do boot-time initialization of a CPU's per-CPU RCU data.
1872 rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp)
1874 unsigned long flags;
1876 struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
1877 struct rcu_node *rnp = rcu_get_root(rsp);
1879 /* Set up local state, ensuring consistent view of global state. */
1880 raw_spin_lock_irqsave(&rnp->lock, flags);
1881 rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo);
1882 rdp->nxtlist = NULL;
1883 for (i = 0; i < RCU_NEXT_SIZE; i++)
1884 rdp->nxttail[i] = &rdp->nxtlist;
1887 rdp->dynticks = &per_cpu(rcu_dynticks, cpu);
1888 #endif /* #ifdef CONFIG_NO_HZ */
1891 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1895 * Initialize a CPU's per-CPU RCU data. Note that only one online or
1896 * offline event can be happening at a given time. Note also that we
1897 * can accept some slop in the rsp->completed access due to the fact
1898 * that this CPU cannot possibly have any RCU callbacks in flight yet.
1900 static void __cpuinit
1901 rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptible)
1903 unsigned long flags;
1905 struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
1906 struct rcu_node *rnp = rcu_get_root(rsp);
1908 /* Set up local state, ensuring consistent view of global state. */
1909 raw_spin_lock_irqsave(&rnp->lock, flags);
1910 rdp->passed_quiesc = 0; /* We could be racing with new GP, */
1911 rdp->qs_pending = 1; /* so set up to respond to current GP. */
1912 rdp->beenonline = 1; /* We have now been online. */
1913 rdp->preemptible = preemptible;
1914 rdp->qlen_last_fqs_check = 0;
1915 rdp->n_force_qs_snap = rsp->n_force_qs;
1916 rdp->blimit = blimit;
1917 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
1920 * A new grace period might start here. If so, we won't be part
1921 * of it, but that is OK, as we are currently in a quiescent state.
1924 /* Exclude any attempts to start a new GP on large systems. */
1925 raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */
1927 /* Add CPU to rcu_node bitmasks. */
1929 mask = rdp->grpmask;
1931 /* Exclude any attempts to start a new GP on small systems. */
1932 raw_spin_lock(&rnp->lock); /* irqs already disabled. */
1933 rnp->qsmaskinit |= mask;
1934 mask = rnp->grpmask;
1935 if (rnp == rdp->mynode) {
1936 rdp->gpnum = rnp->completed; /* if GP in progress... */
1937 rdp->completed = rnp->completed;
1938 rdp->passed_quiesc_completed = rnp->completed - 1;
1939 trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpuonl");
1941 raw_spin_unlock(&rnp->lock); /* irqs already disabled. */
1943 } while (rnp != NULL && !(rnp->qsmaskinit & mask));
1945 raw_spin_unlock_irqrestore(&rsp->onofflock, flags);
1948 static void __cpuinit rcu_prepare_cpu(int cpu)
1950 rcu_init_percpu_data(cpu, &rcu_sched_state, 0);
1951 rcu_init_percpu_data(cpu, &rcu_bh_state, 0);
1952 rcu_preempt_init_percpu_data(cpu);
1956 * Handle CPU online/offline notification events.
1958 static int __cpuinit rcu_cpu_notify(struct notifier_block *self,
1959 unsigned long action, void *hcpu)
1961 long cpu = (long)hcpu;
1962 struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, cpu);
1963 struct rcu_node *rnp = rdp->mynode;
1965 trace_rcu_utilization("Start CPU hotplug");
1967 case CPU_UP_PREPARE:
1968 case CPU_UP_PREPARE_FROZEN:
1969 rcu_prepare_cpu(cpu);
1970 rcu_prepare_kthreads(cpu);
1973 case CPU_DOWN_FAILED:
1974 rcu_node_kthread_setaffinity(rnp, -1);
1975 rcu_cpu_kthread_setrt(cpu, 1);
1977 case CPU_DOWN_PREPARE:
1978 rcu_node_kthread_setaffinity(rnp, cpu);
1979 rcu_cpu_kthread_setrt(cpu, 0);
1982 case CPU_DYING_FROZEN:
1984 * The whole machine is "stopped" except this CPU, so we can
1985 * touch any data without introducing corruption. We send the
1986 * dying CPU's callbacks to an arbitrarily chosen online CPU.
1988 rcu_send_cbs_to_online(&rcu_bh_state);
1989 rcu_send_cbs_to_online(&rcu_sched_state);
1990 rcu_preempt_send_cbs_to_online();
1993 case CPU_DEAD_FROZEN:
1994 case CPU_UP_CANCELED:
1995 case CPU_UP_CANCELED_FROZEN:
1996 rcu_offline_cpu(cpu);
2001 trace_rcu_utilization("End CPU hotplug");
2006 * This function is invoked towards the end of the scheduler's initialization
2007 * process. Before this is called, the idle task might contain
2008 * RCU read-side critical sections (during which time, this idle
2009 * task is booting the system). After this function is called, the
2010 * idle tasks are prohibited from containing RCU read-side critical
2011 * sections. This function also enables RCU lockdep checking.
2013 void rcu_scheduler_starting(void)
2015 WARN_ON(num_online_cpus() != 1);
2016 WARN_ON(nr_context_switches() > 0);
2017 rcu_scheduler_active = 1;
2021 * Compute the per-level fanout, either using the exact fanout specified
2022 * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT.
2024 #ifdef CONFIG_RCU_FANOUT_EXACT
2025 static void __init rcu_init_levelspread(struct rcu_state *rsp)
2029 for (i = NUM_RCU_LVLS - 1; i > 0; i--)
2030 rsp->levelspread[i] = CONFIG_RCU_FANOUT;
2031 rsp->levelspread[0] = RCU_FANOUT_LEAF;
2033 #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */
2034 static void __init rcu_init_levelspread(struct rcu_state *rsp)
2041 for (i = NUM_RCU_LVLS - 1; i >= 0; i--) {
2042 ccur = rsp->levelcnt[i];
2043 rsp->levelspread[i] = (cprv + ccur - 1) / ccur;
2047 #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */
2050 * Helper function for rcu_init() that initializes one rcu_state structure.
2052 static void __init rcu_init_one(struct rcu_state *rsp,
2053 struct rcu_data __percpu *rda)
2055 static char *buf[] = { "rcu_node_level_0",
2058 "rcu_node_level_3" }; /* Match MAX_RCU_LVLS */
2062 struct rcu_node *rnp;
2064 BUILD_BUG_ON(MAX_RCU_LVLS > ARRAY_SIZE(buf)); /* Fix buf[] init! */
2066 /* Initialize the level-tracking arrays. */
2068 for (i = 1; i < NUM_RCU_LVLS; i++)
2069 rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1];
2070 rcu_init_levelspread(rsp);
2072 /* Initialize the elements themselves, starting from the leaves. */
2074 for (i = NUM_RCU_LVLS - 1; i >= 0; i--) {
2075 cpustride *= rsp->levelspread[i];
2076 rnp = rsp->level[i];
2077 for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) {
2078 raw_spin_lock_init(&rnp->lock);
2079 lockdep_set_class_and_name(&rnp->lock,
2080 &rcu_node_class[i], buf[i]);
2083 rnp->qsmaskinit = 0;
2084 rnp->grplo = j * cpustride;
2085 rnp->grphi = (j + 1) * cpustride - 1;
2086 if (rnp->grphi >= NR_CPUS)
2087 rnp->grphi = NR_CPUS - 1;
2093 rnp->grpnum = j % rsp->levelspread[i - 1];
2094 rnp->grpmask = 1UL << rnp->grpnum;
2095 rnp->parent = rsp->level[i - 1] +
2096 j / rsp->levelspread[i - 1];
2099 INIT_LIST_HEAD(&rnp->blkd_tasks);
2104 rnp = rsp->level[NUM_RCU_LVLS - 1];
2105 for_each_possible_cpu(i) {
2106 while (i > rnp->grphi)
2108 per_cpu_ptr(rsp->rda, i)->mynode = rnp;
2109 rcu_boot_init_percpu_data(i, rsp);
2113 void __init rcu_init(void)
2117 rcu_bootup_announce();
2118 rcu_init_one(&rcu_sched_state, &rcu_sched_data);
2119 rcu_init_one(&rcu_bh_state, &rcu_bh_data);
2120 __rcu_init_preempt();
2121 open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
2124 * We don't need protection against CPU-hotplug here because
2125 * this is called early in boot, before either interrupts
2126 * or the scheduler are operational.
2128 cpu_notifier(rcu_cpu_notify, 0);
2129 for_each_online_cpu(cpu)
2130 rcu_cpu_notify(NULL, CPU_UP_PREPARE, (void *)(long)cpu);
2131 check_cpu_stall_init();
2134 #include "rcutree_plugin.h"