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 <asm/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>
52 #ifdef CONFIG_DEBUG_LOCK_ALLOC
53 static struct lock_class_key rcu_lock_key
;
54 struct lockdep_map rcu_lock_map
=
55 STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key
);
56 EXPORT_SYMBOL_GPL(rcu_lock_map
);
59 /* Data structures. */
61 #define RCU_STATE_INITIALIZER(name) { \
62 .level = { &name.node[0] }, \
64 NUM_RCU_LVL_0, /* root of hierarchy. */ \
67 NUM_RCU_LVL_3, /* == MAX_RCU_LVLS */ \
69 .signaled = RCU_SIGNAL_INIT, \
72 .onofflock = __SPIN_LOCK_UNLOCKED(&name.onofflock), \
73 .fqslock = __SPIN_LOCK_UNLOCKED(&name.fqslock), \
75 .n_force_qs_ngp = 0, \
78 struct rcu_state rcu_sched_state
= RCU_STATE_INITIALIZER(rcu_sched_state
);
79 DEFINE_PER_CPU(struct rcu_data
, rcu_sched_data
);
81 struct rcu_state rcu_bh_state
= RCU_STATE_INITIALIZER(rcu_bh_state
);
82 DEFINE_PER_CPU(struct rcu_data
, rcu_bh_data
);
84 extern long rcu_batches_completed_sched(void);
85 static struct rcu_node
*rcu_get_root(struct rcu_state
*rsp
);
86 static void cpu_quiet_msk(unsigned long mask
, struct rcu_state
*rsp
,
87 struct rcu_node
*rnp
, unsigned long flags
);
88 static void cpu_quiet_msk_finish(struct rcu_state
*rsp
, unsigned long flags
);
89 #ifdef CONFIG_HOTPLUG_CPU
90 static void __rcu_offline_cpu(int cpu
, struct rcu_state
*rsp
);
91 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
92 static void __rcu_process_callbacks(struct rcu_state
*rsp
,
93 struct rcu_data
*rdp
);
94 static void __call_rcu(struct rcu_head
*head
,
95 void (*func
)(struct rcu_head
*rcu
),
96 struct rcu_state
*rsp
);
97 static int __rcu_pending(struct rcu_state
*rsp
, struct rcu_data
*rdp
);
98 static void __cpuinit
rcu_init_percpu_data(int cpu
, struct rcu_state
*rsp
,
101 #include "rcutree_plugin.h"
104 * Note a quiescent state. Because we do not need to know
105 * how many quiescent states passed, just if there was at least
106 * one since the start of the grace period, this just sets a flag.
108 void rcu_sched_qs(int cpu
)
111 struct rcu_data
*rdp
;
113 local_irq_save(flags
);
114 rdp
= &per_cpu(rcu_sched_data
, cpu
);
115 rdp
->passed_quiesc
= 1;
116 rdp
->passed_quiesc_completed
= rdp
->completed
;
118 local_irq_restore(flags
);
121 void rcu_bh_qs(int cpu
)
124 struct rcu_data
*rdp
;
126 local_irq_save(flags
);
127 rdp
= &per_cpu(rcu_bh_data
, cpu
);
128 rdp
->passed_quiesc
= 1;
129 rdp
->passed_quiesc_completed
= rdp
->completed
;
130 local_irq_restore(flags
);
134 DEFINE_PER_CPU(struct rcu_dynticks
, rcu_dynticks
) = {
135 .dynticks_nesting
= 1,
138 #endif /* #ifdef CONFIG_NO_HZ */
140 static int blimit
= 10; /* Maximum callbacks per softirq. */
141 static int qhimark
= 10000; /* If this many pending, ignore blimit. */
142 static int qlowmark
= 100; /* Once only this many pending, use blimit. */
144 static void force_quiescent_state(struct rcu_state
*rsp
, int relaxed
);
145 static int rcu_pending(int cpu
);
148 * Return the number of RCU-sched batches processed thus far for debug & stats.
150 long rcu_batches_completed_sched(void)
152 return rcu_sched_state
.completed
;
154 EXPORT_SYMBOL_GPL(rcu_batches_completed_sched
);
157 * Return the number of RCU BH batches processed thus far for debug & stats.
159 long rcu_batches_completed_bh(void)
161 return rcu_bh_state
.completed
;
163 EXPORT_SYMBOL_GPL(rcu_batches_completed_bh
);
166 * Does the CPU have callbacks ready to be invoked?
169 cpu_has_callbacks_ready_to_invoke(struct rcu_data
*rdp
)
171 return &rdp
->nxtlist
!= rdp
->nxttail
[RCU_DONE_TAIL
];
175 * Does the current CPU require a yet-as-unscheduled grace period?
178 cpu_needs_another_gp(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
180 /* ACCESS_ONCE() because we are accessing outside of lock. */
181 return *rdp
->nxttail
[RCU_DONE_TAIL
] &&
182 ACCESS_ONCE(rsp
->completed
) == ACCESS_ONCE(rsp
->gpnum
);
186 * Return the root node of the specified rcu_state structure.
188 static struct rcu_node
*rcu_get_root(struct rcu_state
*rsp
)
190 return &rsp
->node
[0];
196 * If the specified CPU is offline, tell the caller that it is in
197 * a quiescent state. Otherwise, whack it with a reschedule IPI.
198 * Grace periods can end up waiting on an offline CPU when that
199 * CPU is in the process of coming online -- it will be added to the
200 * rcu_node bitmasks before it actually makes it online. The same thing
201 * can happen while a CPU is in the process of coming online. Because this
202 * race is quite rare, we check for it after detecting that the grace
203 * period has been delayed rather than checking each and every CPU
204 * each and every time we start a new grace period.
206 static int rcu_implicit_offline_qs(struct rcu_data
*rdp
)
209 * If the CPU is offline, it is in a quiescent state. We can
210 * trust its state not to change because interrupts are disabled.
212 if (cpu_is_offline(rdp
->cpu
)) {
217 /* If preemptable RCU, no point in sending reschedule IPI. */
218 if (rdp
->preemptable
)
221 /* The CPU is online, so send it a reschedule IPI. */
222 if (rdp
->cpu
!= smp_processor_id())
223 smp_send_reschedule(rdp
->cpu
);
230 #endif /* #ifdef CONFIG_SMP */
235 * rcu_enter_nohz - inform RCU that current CPU is entering nohz
237 * Enter nohz mode, in other words, -leave- the mode in which RCU
238 * read-side critical sections can occur. (Though RCU read-side
239 * critical sections can occur in irq handlers in nohz mode, a possibility
240 * handled by rcu_irq_enter() and rcu_irq_exit()).
242 void rcu_enter_nohz(void)
245 struct rcu_dynticks
*rdtp
;
247 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
248 local_irq_save(flags
);
249 rdtp
= &__get_cpu_var(rcu_dynticks
);
251 rdtp
->dynticks_nesting
--;
252 WARN_ON_ONCE(rdtp
->dynticks
& 0x1);
253 local_irq_restore(flags
);
257 * rcu_exit_nohz - inform RCU that current CPU is leaving nohz
259 * Exit nohz mode, in other words, -enter- the mode in which RCU
260 * read-side critical sections normally occur.
262 void rcu_exit_nohz(void)
265 struct rcu_dynticks
*rdtp
;
267 local_irq_save(flags
);
268 rdtp
= &__get_cpu_var(rcu_dynticks
);
270 rdtp
->dynticks_nesting
++;
271 WARN_ON_ONCE(!(rdtp
->dynticks
& 0x1));
272 local_irq_restore(flags
);
273 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
277 * rcu_nmi_enter - inform RCU of entry to NMI context
279 * If the CPU was idle with dynamic ticks active, and there is no
280 * irq handler running, this updates rdtp->dynticks_nmi to let the
281 * RCU grace-period handling know that the CPU is active.
283 void rcu_nmi_enter(void)
285 struct rcu_dynticks
*rdtp
= &__get_cpu_var(rcu_dynticks
);
287 if (rdtp
->dynticks
& 0x1)
289 rdtp
->dynticks_nmi
++;
290 WARN_ON_ONCE(!(rdtp
->dynticks_nmi
& 0x1));
291 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
295 * rcu_nmi_exit - inform RCU of exit from NMI context
297 * If the CPU was idle with dynamic ticks active, and there is no
298 * irq handler running, this updates rdtp->dynticks_nmi to let the
299 * RCU grace-period handling know that the CPU is no longer active.
301 void rcu_nmi_exit(void)
303 struct rcu_dynticks
*rdtp
= &__get_cpu_var(rcu_dynticks
);
305 if (rdtp
->dynticks
& 0x1)
307 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
308 rdtp
->dynticks_nmi
++;
309 WARN_ON_ONCE(rdtp
->dynticks_nmi
& 0x1);
313 * rcu_irq_enter - inform RCU of entry to hard irq context
315 * If the CPU was idle with dynamic ticks active, this updates the
316 * rdtp->dynticks to let the RCU handling know that the CPU is active.
318 void rcu_irq_enter(void)
320 struct rcu_dynticks
*rdtp
= &__get_cpu_var(rcu_dynticks
);
322 if (rdtp
->dynticks_nesting
++)
325 WARN_ON_ONCE(!(rdtp
->dynticks
& 0x1));
326 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
330 * rcu_irq_exit - inform RCU of exit from hard irq context
332 * If the CPU was idle with dynamic ticks active, update the rdp->dynticks
333 * to put let the RCU handling be aware that the CPU is going back to idle
336 void rcu_irq_exit(void)
338 struct rcu_dynticks
*rdtp
= &__get_cpu_var(rcu_dynticks
);
340 if (--rdtp
->dynticks_nesting
)
342 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
344 WARN_ON_ONCE(rdtp
->dynticks
& 0x1);
346 /* If the interrupt queued a callback, get out of dyntick mode. */
347 if (__get_cpu_var(rcu_sched_data
).nxtlist
||
348 __get_cpu_var(rcu_bh_data
).nxtlist
)
353 * Record the specified "completed" value, which is later used to validate
354 * dynticks counter manipulations. Specify "rsp->completed - 1" to
355 * unconditionally invalidate any future dynticks manipulations (which is
356 * useful at the beginning of a grace period).
358 static void dyntick_record_completed(struct rcu_state
*rsp
, long comp
)
360 rsp
->dynticks_completed
= comp
;
366 * Recall the previously recorded value of the completion for dynticks.
368 static long dyntick_recall_completed(struct rcu_state
*rsp
)
370 return rsp
->dynticks_completed
;
374 * Snapshot the specified CPU's dynticks counter so that we can later
375 * credit them with an implicit quiescent state. Return 1 if this CPU
376 * is already in a quiescent state courtesy of dynticks idle mode.
378 static int dyntick_save_progress_counter(struct rcu_data
*rdp
)
384 snap
= rdp
->dynticks
->dynticks
;
385 snap_nmi
= rdp
->dynticks
->dynticks_nmi
;
386 smp_mb(); /* Order sampling of snap with end of grace period. */
387 rdp
->dynticks_snap
= snap
;
388 rdp
->dynticks_nmi_snap
= snap_nmi
;
389 ret
= ((snap
& 0x1) == 0) && ((snap_nmi
& 0x1) == 0);
396 * Return true if the specified CPU has passed through a quiescent
397 * state by virtue of being in or having passed through an dynticks
398 * idle state since the last call to dyntick_save_progress_counter()
401 static int rcu_implicit_dynticks_qs(struct rcu_data
*rdp
)
408 curr
= rdp
->dynticks
->dynticks
;
409 snap
= rdp
->dynticks_snap
;
410 curr_nmi
= rdp
->dynticks
->dynticks_nmi
;
411 snap_nmi
= rdp
->dynticks_nmi_snap
;
412 smp_mb(); /* force ordering with cpu entering/leaving dynticks. */
415 * If the CPU passed through or entered a dynticks idle phase with
416 * no active irq/NMI handlers, then we can safely pretend that the CPU
417 * already acknowledged the request to pass through a quiescent
418 * state. Either way, that CPU cannot possibly be in an RCU
419 * read-side critical section that started before the beginning
420 * of the current RCU grace period.
422 if ((curr
!= snap
|| (curr
& 0x1) == 0) &&
423 (curr_nmi
!= snap_nmi
|| (curr_nmi
& 0x1) == 0)) {
428 /* Go check for the CPU being offline. */
429 return rcu_implicit_offline_qs(rdp
);
432 #endif /* #ifdef CONFIG_SMP */
434 #else /* #ifdef CONFIG_NO_HZ */
436 static void dyntick_record_completed(struct rcu_state
*rsp
, long comp
)
443 * If there are no dynticks, then the only way that a CPU can passively
444 * be in a quiescent state is to be offline. Unlike dynticks idle, which
445 * is a point in time during the prior (already finished) grace period,
446 * an offline CPU is always in a quiescent state, and thus can be
447 * unconditionally applied. So just return the current value of completed.
449 static long dyntick_recall_completed(struct rcu_state
*rsp
)
451 return rsp
->completed
;
454 static int dyntick_save_progress_counter(struct rcu_data
*rdp
)
459 static int rcu_implicit_dynticks_qs(struct rcu_data
*rdp
)
461 return rcu_implicit_offline_qs(rdp
);
464 #endif /* #ifdef CONFIG_SMP */
466 #endif /* #else #ifdef CONFIG_NO_HZ */
468 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
470 static void record_gp_stall_check_time(struct rcu_state
*rsp
)
472 rsp
->gp_start
= jiffies
;
473 rsp
->jiffies_stall
= jiffies
+ RCU_SECONDS_TILL_STALL_CHECK
;
476 static void print_other_cpu_stall(struct rcu_state
*rsp
)
481 struct rcu_node
*rnp
= rcu_get_root(rsp
);
482 struct rcu_node
*rnp_cur
= rsp
->level
[NUM_RCU_LVLS
- 1];
483 struct rcu_node
*rnp_end
= &rsp
->node
[NUM_RCU_NODES
];
485 /* Only let one CPU complain about others per time interval. */
487 spin_lock_irqsave(&rnp
->lock
, flags
);
488 delta
= jiffies
- rsp
->jiffies_stall
;
489 if (delta
< RCU_STALL_RAT_DELAY
|| rsp
->gpnum
== rsp
->completed
) {
490 spin_unlock_irqrestore(&rnp
->lock
, flags
);
493 rsp
->jiffies_stall
= jiffies
+ RCU_SECONDS_TILL_STALL_RECHECK
;
494 spin_unlock_irqrestore(&rnp
->lock
, flags
);
496 /* OK, time to rat on our buddy... */
498 printk(KERN_ERR
"INFO: RCU detected CPU stalls:");
499 for (; rnp_cur
< rnp_end
; rnp_cur
++) {
500 rcu_print_task_stall(rnp
);
501 if (rnp_cur
->qsmask
== 0)
503 for (cpu
= 0; cpu
<= rnp_cur
->grphi
- rnp_cur
->grplo
; cpu
++)
504 if (rnp_cur
->qsmask
& (1UL << cpu
))
505 printk(" %d", rnp_cur
->grplo
+ cpu
);
507 printk(" (detected by %d, t=%ld jiffies)\n",
508 smp_processor_id(), (long)(jiffies
- rsp
->gp_start
));
509 trigger_all_cpu_backtrace();
511 force_quiescent_state(rsp
, 0); /* Kick them all. */
514 static void print_cpu_stall(struct rcu_state
*rsp
)
517 struct rcu_node
*rnp
= rcu_get_root(rsp
);
519 printk(KERN_ERR
"INFO: RCU detected CPU %d stall (t=%lu jiffies)\n",
520 smp_processor_id(), jiffies
- rsp
->gp_start
);
521 trigger_all_cpu_backtrace();
523 spin_lock_irqsave(&rnp
->lock
, flags
);
524 if ((long)(jiffies
- rsp
->jiffies_stall
) >= 0)
526 jiffies
+ RCU_SECONDS_TILL_STALL_RECHECK
;
527 spin_unlock_irqrestore(&rnp
->lock
, flags
);
529 set_need_resched(); /* kick ourselves to get things going. */
532 static void check_cpu_stall(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
535 struct rcu_node
*rnp
;
537 delta
= jiffies
- rsp
->jiffies_stall
;
539 if ((rnp
->qsmask
& rdp
->grpmask
) && delta
>= 0) {
541 /* We haven't checked in, so go dump stack. */
542 print_cpu_stall(rsp
);
544 } else if (rsp
->gpnum
!= rsp
->completed
&&
545 delta
>= RCU_STALL_RAT_DELAY
) {
547 /* They had two time units to dump stack, so complain. */
548 print_other_cpu_stall(rsp
);
552 #else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
554 static void record_gp_stall_check_time(struct rcu_state
*rsp
)
558 static void check_cpu_stall(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
562 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
565 * Update CPU-local rcu_data state to record the newly noticed grace period.
566 * This is used both when we started the grace period and when we notice
567 * that someone else started the grace period.
569 static void note_new_gpnum(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
572 rdp
->passed_quiesc
= 0;
573 rdp
->gpnum
= rsp
->gpnum
;
577 * Did someone else start a new RCU grace period start since we last
578 * checked? Update local state appropriately if so. Must be called
579 * on the CPU corresponding to rdp.
582 check_for_new_grace_period(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
587 local_irq_save(flags
);
588 if (rdp
->gpnum
!= rsp
->gpnum
) {
589 note_new_gpnum(rsp
, rdp
);
592 local_irq_restore(flags
);
597 * Start a new RCU grace period if warranted, re-initializing the hierarchy
598 * in preparation for detecting the next grace period. The caller must hold
599 * the root node's ->lock, which is released before return. Hard irqs must
603 rcu_start_gp(struct rcu_state
*rsp
, unsigned long flags
)
604 __releases(rcu_get_root(rsp
)->lock
)
606 struct rcu_data
*rdp
= rsp
->rda
[smp_processor_id()];
607 struct rcu_node
*rnp
= rcu_get_root(rsp
);
608 struct rcu_node
*rnp_cur
;
609 struct rcu_node
*rnp_end
;
611 if (!cpu_needs_another_gp(rsp
, rdp
)) {
612 spin_unlock_irqrestore(&rnp
->lock
, flags
);
616 /* Advance to a new grace period and initialize state. */
618 rsp
->signaled
= RCU_GP_INIT
; /* Hold off force_quiescent_state. */
619 rsp
->jiffies_force_qs
= jiffies
+ RCU_JIFFIES_TILL_FORCE_QS
;
620 record_gp_stall_check_time(rsp
);
621 dyntick_record_completed(rsp
, rsp
->completed
- 1);
622 note_new_gpnum(rsp
, rdp
);
625 * Because we are first, we know that all our callbacks will
626 * be covered by this upcoming grace period, even the ones
627 * that were registered arbitrarily recently.
629 rdp
->nxttail
[RCU_NEXT_READY_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
630 rdp
->nxttail
[RCU_WAIT_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
632 /* Special-case the common single-level case. */
633 if (NUM_RCU_NODES
== 1) {
634 rnp
->qsmask
= rnp
->qsmaskinit
;
635 rnp
->gpnum
= rsp
->gpnum
;
636 rsp
->signaled
= RCU_SIGNAL_INIT
; /* force_quiescent_state OK. */
637 spin_unlock_irqrestore(&rnp
->lock
, flags
);
641 spin_unlock(&rnp
->lock
); /* leave irqs disabled. */
644 /* Exclude any concurrent CPU-hotplug operations. */
645 spin_lock(&rsp
->onofflock
); /* irqs already disabled. */
648 * Set the quiescent-state-needed bits in all the non-leaf RCU
649 * nodes for all currently online CPUs. This operation relies
650 * on the layout of the hierarchy within the rsp->node[] array.
651 * Note that other CPUs will access only the leaves of the
652 * hierarchy, which still indicate that no grace period is in
653 * progress. In addition, we have excluded CPU-hotplug operations.
655 * We therefore do not need to hold any locks. Any required
656 * memory barriers will be supplied by the locks guarding the
657 * leaf rcu_nodes in the hierarchy.
660 rnp_end
= rsp
->level
[NUM_RCU_LVLS
- 1];
661 for (rnp_cur
= &rsp
->node
[0]; rnp_cur
< rnp_end
; rnp_cur
++) {
662 rnp_cur
->qsmask
= rnp_cur
->qsmaskinit
;
663 rnp
->gpnum
= rsp
->gpnum
;
667 * Now set up the leaf nodes. Here we must be careful. First,
668 * we need to hold the lock in order to exclude other CPUs, which
669 * might be contending for the leaf nodes' locks. Second, as
670 * soon as we initialize a given leaf node, its CPUs might run
671 * up the rest of the hierarchy. We must therefore acquire locks
672 * for each node that we touch during this stage. (But we still
673 * are excluding CPU-hotplug operations.)
675 * Note that the grace period cannot complete until we finish
676 * the initialization process, as there will be at least one
677 * qsmask bit set in the root node until that time, namely the
678 * one corresponding to this CPU.
680 rnp_end
= &rsp
->node
[NUM_RCU_NODES
];
681 rnp_cur
= rsp
->level
[NUM_RCU_LVLS
- 1];
682 for (; rnp_cur
< rnp_end
; rnp_cur
++) {
683 spin_lock(&rnp_cur
->lock
); /* irqs already disabled. */
684 rnp_cur
->qsmask
= rnp_cur
->qsmaskinit
;
685 rnp
->gpnum
= rsp
->gpnum
;
686 spin_unlock(&rnp_cur
->lock
); /* irqs already disabled. */
689 rsp
->signaled
= RCU_SIGNAL_INIT
; /* force_quiescent_state now OK. */
690 spin_unlock_irqrestore(&rsp
->onofflock
, flags
);
694 * Advance this CPU's callbacks, but only if the current grace period
695 * has ended. This may be called only from the CPU to whom the rdp
699 rcu_process_gp_end(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
704 local_irq_save(flags
);
705 completed_snap
= ACCESS_ONCE(rsp
->completed
); /* outside of lock. */
707 /* Did another grace period end? */
708 if (rdp
->completed
!= completed_snap
) {
710 /* Advance callbacks. No harm if list empty. */
711 rdp
->nxttail
[RCU_DONE_TAIL
] = rdp
->nxttail
[RCU_WAIT_TAIL
];
712 rdp
->nxttail
[RCU_WAIT_TAIL
] = rdp
->nxttail
[RCU_NEXT_READY_TAIL
];
713 rdp
->nxttail
[RCU_NEXT_READY_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
715 /* Remember that we saw this grace-period completion. */
716 rdp
->completed
= completed_snap
;
718 local_irq_restore(flags
);
722 * Clean up after the prior grace period and let rcu_start_gp() start up
723 * the next grace period if one is needed. Note that the caller must
724 * hold rnp->lock, as required by rcu_start_gp(), which will release it.
726 static void cpu_quiet_msk_finish(struct rcu_state
*rsp
, unsigned long flags
)
727 __releases(rnp
->lock
)
729 rsp
->completed
= rsp
->gpnum
;
730 rcu_process_gp_end(rsp
, rsp
->rda
[smp_processor_id()]);
731 rcu_start_gp(rsp
, flags
); /* releases root node's rnp->lock. */
735 * Similar to cpu_quiet(), for which it is a helper function. Allows
736 * a group of CPUs to be quieted at one go, though all the CPUs in the
737 * group must be represented by the same leaf rcu_node structure.
738 * That structure's lock must be held upon entry, and it is released
742 cpu_quiet_msk(unsigned long mask
, struct rcu_state
*rsp
, struct rcu_node
*rnp
,
744 __releases(rnp
->lock
)
746 /* Walk up the rcu_node hierarchy. */
748 if (!(rnp
->qsmask
& mask
)) {
750 /* Our bit has already been cleared, so done. */
751 spin_unlock_irqrestore(&rnp
->lock
, flags
);
754 rnp
->qsmask
&= ~mask
;
755 if (rnp
->qsmask
!= 0 || rcu_preempted_readers(rnp
)) {
757 /* Other bits still set at this level, so done. */
758 spin_unlock_irqrestore(&rnp
->lock
, flags
);
762 if (rnp
->parent
== NULL
) {
764 /* No more levels. Exit loop holding root lock. */
768 spin_unlock_irqrestore(&rnp
->lock
, flags
);
770 spin_lock_irqsave(&rnp
->lock
, flags
);
774 * Get here if we are the last CPU to pass through a quiescent
775 * state for this grace period. Invoke cpu_quiet_msk_finish()
776 * to clean up and start the next grace period if one is needed.
778 cpu_quiet_msk_finish(rsp
, flags
); /* releases rnp->lock. */
782 * Record a quiescent state for the specified CPU, which must either be
783 * the current CPU or an offline CPU. The lastcomp argument is used to
784 * make sure we are still in the grace period of interest. We don't want
785 * to end the current grace period based on quiescent states detected in
786 * an earlier grace period!
789 cpu_quiet(int cpu
, struct rcu_state
*rsp
, struct rcu_data
*rdp
, long lastcomp
)
793 struct rcu_node
*rnp
;
796 spin_lock_irqsave(&rnp
->lock
, flags
);
797 if (lastcomp
!= ACCESS_ONCE(rsp
->completed
)) {
800 * Someone beat us to it for this grace period, so leave.
801 * The race with GP start is resolved by the fact that we
802 * hold the leaf rcu_node lock, so that the per-CPU bits
803 * cannot yet be initialized -- so we would simply find our
804 * CPU's bit already cleared in cpu_quiet_msk() if this race
807 rdp
->passed_quiesc
= 0; /* try again later! */
808 spin_unlock_irqrestore(&rnp
->lock
, flags
);
812 if ((rnp
->qsmask
& mask
) == 0) {
813 spin_unlock_irqrestore(&rnp
->lock
, flags
);
818 * This GP can't end until cpu checks in, so all of our
819 * callbacks can be processed during the next GP.
821 rdp
= rsp
->rda
[smp_processor_id()];
822 rdp
->nxttail
[RCU_NEXT_READY_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
824 cpu_quiet_msk(mask
, rsp
, rnp
, flags
); /* releases rnp->lock */
829 * Check to see if there is a new grace period of which this CPU
830 * is not yet aware, and if so, set up local rcu_data state for it.
831 * Otherwise, see if this CPU has just passed through its first
832 * quiescent state for this grace period, and record that fact if so.
835 rcu_check_quiescent_state(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
837 /* If there is now a new grace period, record and return. */
838 if (check_for_new_grace_period(rsp
, rdp
))
842 * Does this CPU still need to do its part for current grace period?
843 * If no, return and let the other CPUs do their part as well.
845 if (!rdp
->qs_pending
)
849 * Was there a quiescent state since the beginning of the grace
850 * period? If no, then exit and wait for the next call.
852 if (!rdp
->passed_quiesc
)
855 /* Tell RCU we are done (but cpu_quiet() will be the judge of that). */
856 cpu_quiet(rdp
->cpu
, rsp
, rdp
, rdp
->passed_quiesc_completed
);
859 #ifdef CONFIG_HOTPLUG_CPU
862 * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy
863 * and move all callbacks from the outgoing CPU to the current one.
865 static void __rcu_offline_cpu(int cpu
, struct rcu_state
*rsp
)
871 struct rcu_data
*rdp
= rsp
->rda
[cpu
];
872 struct rcu_data
*rdp_me
;
873 struct rcu_node
*rnp
;
875 /* Exclude any attempts to start a new grace period. */
876 spin_lock_irqsave(&rsp
->onofflock
, flags
);
878 /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */
880 mask
= rdp
->grpmask
; /* rnp->grplo is constant. */
882 spin_lock(&rnp
->lock
); /* irqs already disabled. */
883 rnp
->qsmaskinit
&= ~mask
;
884 if (rnp
->qsmaskinit
!= 0) {
885 spin_unlock(&rnp
->lock
); /* irqs remain disabled. */
888 rcu_preempt_offline_tasks(rsp
, rnp
);
890 spin_unlock(&rnp
->lock
); /* irqs remain disabled. */
892 } while (rnp
!= NULL
);
893 lastcomp
= rsp
->completed
;
895 spin_unlock(&rsp
->onofflock
); /* irqs remain disabled. */
897 /* Being offline is a quiescent state, so go record it. */
898 cpu_quiet(cpu
, rsp
, rdp
, lastcomp
);
901 * Move callbacks from the outgoing CPU to the running CPU.
902 * Note that the outgoing CPU is now quiscent, so it is now
903 * (uncharacteristically) safe to access its rcu_data structure.
904 * Note also that we must carefully retain the order of the
905 * outgoing CPU's callbacks in order for rcu_barrier() to work
906 * correctly. Finally, note that we start all the callbacks
907 * afresh, even those that have passed through a grace period
908 * and are therefore ready to invoke. The theory is that hotplug
909 * events are rare, and that if they are frequent enough to
910 * indefinitely delay callbacks, you have far worse things to
913 rdp_me
= rsp
->rda
[smp_processor_id()];
914 if (rdp
->nxtlist
!= NULL
) {
915 *rdp_me
->nxttail
[RCU_NEXT_TAIL
] = rdp
->nxtlist
;
916 rdp_me
->nxttail
[RCU_NEXT_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
918 for (i
= 0; i
< RCU_NEXT_SIZE
; i
++)
919 rdp
->nxttail
[i
] = &rdp
->nxtlist
;
920 rdp_me
->qlen
+= rdp
->qlen
;
923 local_irq_restore(flags
);
927 * Remove the specified CPU from the RCU hierarchy and move any pending
928 * callbacks that it might have to the current CPU. This code assumes
929 * that at least one CPU in the system will remain running at all times.
930 * Any attempt to offline -all- CPUs is likely to strand RCU callbacks.
932 static void rcu_offline_cpu(int cpu
)
934 __rcu_offline_cpu(cpu
, &rcu_sched_state
);
935 __rcu_offline_cpu(cpu
, &rcu_bh_state
);
936 rcu_preempt_offline_cpu(cpu
);
939 #else /* #ifdef CONFIG_HOTPLUG_CPU */
941 static void rcu_offline_cpu(int cpu
)
945 #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */
948 * Invoke any RCU callbacks that have made it to the end of their grace
949 * period. Thottle as specified by rdp->blimit.
951 static void rcu_do_batch(struct rcu_data
*rdp
)
954 struct rcu_head
*next
, *list
, **tail
;
957 /* If no callbacks are ready, just return.*/
958 if (!cpu_has_callbacks_ready_to_invoke(rdp
))
962 * Extract the list of ready callbacks, disabling to prevent
963 * races with call_rcu() from interrupt handlers.
965 local_irq_save(flags
);
967 rdp
->nxtlist
= *rdp
->nxttail
[RCU_DONE_TAIL
];
968 *rdp
->nxttail
[RCU_DONE_TAIL
] = NULL
;
969 tail
= rdp
->nxttail
[RCU_DONE_TAIL
];
970 for (count
= RCU_NEXT_SIZE
- 1; count
>= 0; count
--)
971 if (rdp
->nxttail
[count
] == rdp
->nxttail
[RCU_DONE_TAIL
])
972 rdp
->nxttail
[count
] = &rdp
->nxtlist
;
973 local_irq_restore(flags
);
975 /* Invoke callbacks. */
982 if (++count
>= rdp
->blimit
)
986 local_irq_save(flags
);
988 /* Update count, and requeue any remaining callbacks. */
991 *tail
= rdp
->nxtlist
;
993 for (count
= 0; count
< RCU_NEXT_SIZE
; count
++)
994 if (&rdp
->nxtlist
== rdp
->nxttail
[count
])
995 rdp
->nxttail
[count
] = tail
;
1000 /* Reinstate batch limit if we have worked down the excess. */
1001 if (rdp
->blimit
== LONG_MAX
&& rdp
->qlen
<= qlowmark
)
1002 rdp
->blimit
= blimit
;
1004 local_irq_restore(flags
);
1006 /* Re-raise the RCU softirq if there are callbacks remaining. */
1007 if (cpu_has_callbacks_ready_to_invoke(rdp
))
1008 raise_softirq(RCU_SOFTIRQ
);
1012 * Check to see if this CPU is in a non-context-switch quiescent state
1013 * (user mode or idle loop for rcu, non-softirq execution for rcu_bh).
1014 * Also schedule the RCU softirq handler.
1016 * This function must be called with hardirqs disabled. It is normally
1017 * invoked from the scheduling-clock interrupt. If rcu_pending returns
1018 * false, there is no point in invoking rcu_check_callbacks().
1020 void rcu_check_callbacks(int cpu
, int user
)
1022 if (!rcu_pending(cpu
))
1023 return; /* if nothing for RCU to do. */
1025 (idle_cpu(cpu
) && rcu_scheduler_active
&&
1026 !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT
))) {
1029 * Get here if this CPU took its interrupt from user
1030 * mode or from the idle loop, and if this is not a
1031 * nested interrupt. In this case, the CPU is in
1032 * a quiescent state, so note it.
1034 * No memory barrier is required here because both
1035 * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local
1036 * variables that other CPUs neither access nor modify,
1037 * at least not while the corresponding CPU is online.
1043 } else if (!in_softirq()) {
1046 * Get here if this CPU did not take its interrupt from
1047 * softirq, in other words, if it is not interrupting
1048 * a rcu_bh read-side critical section. This is an _bh
1049 * critical section, so note it.
1054 rcu_preempt_check_callbacks(cpu
);
1055 raise_softirq(RCU_SOFTIRQ
);
1061 * Scan the leaf rcu_node structures, processing dyntick state for any that
1062 * have not yet encountered a quiescent state, using the function specified.
1063 * Returns 1 if the current grace period ends while scanning (possibly
1064 * because we made it end).
1066 static int rcu_process_dyntick(struct rcu_state
*rsp
, long lastcomp
,
1067 int (*f
)(struct rcu_data
*))
1071 unsigned long flags
;
1073 struct rcu_node
*rnp_cur
= rsp
->level
[NUM_RCU_LVLS
- 1];
1074 struct rcu_node
*rnp_end
= &rsp
->node
[NUM_RCU_NODES
];
1076 for (; rnp_cur
< rnp_end
; rnp_cur
++) {
1078 spin_lock_irqsave(&rnp_cur
->lock
, flags
);
1079 if (rsp
->completed
!= lastcomp
) {
1080 spin_unlock_irqrestore(&rnp_cur
->lock
, flags
);
1083 if (rnp_cur
->qsmask
== 0) {
1084 spin_unlock_irqrestore(&rnp_cur
->lock
, flags
);
1087 cpu
= rnp_cur
->grplo
;
1089 for (; cpu
<= rnp_cur
->grphi
; cpu
++, bit
<<= 1) {
1090 if ((rnp_cur
->qsmask
& bit
) != 0 && f(rsp
->rda
[cpu
]))
1093 if (mask
!= 0 && rsp
->completed
== lastcomp
) {
1095 /* cpu_quiet_msk() releases rnp_cur->lock. */
1096 cpu_quiet_msk(mask
, rsp
, rnp_cur
, flags
);
1099 spin_unlock_irqrestore(&rnp_cur
->lock
, flags
);
1105 * Force quiescent states on reluctant CPUs, and also detect which
1106 * CPUs are in dyntick-idle mode.
1108 static void force_quiescent_state(struct rcu_state
*rsp
, int relaxed
)
1110 unsigned long flags
;
1112 struct rcu_node
*rnp
= rcu_get_root(rsp
);
1115 if (ACCESS_ONCE(rsp
->completed
) == ACCESS_ONCE(rsp
->gpnum
))
1116 return; /* No grace period in progress, nothing to force. */
1117 if (!spin_trylock_irqsave(&rsp
->fqslock
, flags
)) {
1118 rsp
->n_force_qs_lh
++; /* Inexact, can lose counts. Tough! */
1119 return; /* Someone else is already on the job. */
1122 (long)(rsp
->jiffies_force_qs
- jiffies
) >= 0)
1123 goto unlock_ret
; /* no emergency and done recently. */
1125 spin_lock(&rnp
->lock
);
1126 lastcomp
= rsp
->completed
;
1127 signaled
= rsp
->signaled
;
1128 rsp
->jiffies_force_qs
= jiffies
+ RCU_JIFFIES_TILL_FORCE_QS
;
1129 if (lastcomp
== rsp
->gpnum
) {
1130 rsp
->n_force_qs_ngp
++;
1131 spin_unlock(&rnp
->lock
);
1132 goto unlock_ret
; /* no GP in progress, time updated. */
1134 spin_unlock(&rnp
->lock
);
1138 break; /* grace period still initializing, ignore. */
1140 case RCU_SAVE_DYNTICK
:
1142 if (RCU_SIGNAL_INIT
!= RCU_SAVE_DYNTICK
)
1143 break; /* So gcc recognizes the dead code. */
1145 /* Record dyntick-idle state. */
1146 if (rcu_process_dyntick(rsp
, lastcomp
,
1147 dyntick_save_progress_counter
))
1150 /* Update state, record completion counter. */
1151 spin_lock(&rnp
->lock
);
1152 if (lastcomp
== rsp
->completed
) {
1153 rsp
->signaled
= RCU_FORCE_QS
;
1154 dyntick_record_completed(rsp
, lastcomp
);
1156 spin_unlock(&rnp
->lock
);
1161 /* Check dyntick-idle state, send IPI to laggarts. */
1162 if (rcu_process_dyntick(rsp
, dyntick_recall_completed(rsp
),
1163 rcu_implicit_dynticks_qs
))
1166 /* Leave state in case more forcing is required. */
1171 spin_unlock_irqrestore(&rsp
->fqslock
, flags
);
1174 #else /* #ifdef CONFIG_SMP */
1176 static void force_quiescent_state(struct rcu_state
*rsp
, int relaxed
)
1181 #endif /* #else #ifdef CONFIG_SMP */
1184 * This does the RCU processing work from softirq context for the
1185 * specified rcu_state and rcu_data structures. This may be called
1186 * only from the CPU to whom the rdp belongs.
1189 __rcu_process_callbacks(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
1191 unsigned long flags
;
1193 WARN_ON_ONCE(rdp
->beenonline
== 0);
1196 * If an RCU GP has gone long enough, go check for dyntick
1197 * idle CPUs and, if needed, send resched IPIs.
1199 if ((long)(ACCESS_ONCE(rsp
->jiffies_force_qs
) - jiffies
) < 0)
1200 force_quiescent_state(rsp
, 1);
1203 * Advance callbacks in response to end of earlier grace
1204 * period that some other CPU ended.
1206 rcu_process_gp_end(rsp
, rdp
);
1208 /* Update RCU state based on any recent quiescent states. */
1209 rcu_check_quiescent_state(rsp
, rdp
);
1211 /* Does this CPU require a not-yet-started grace period? */
1212 if (cpu_needs_another_gp(rsp
, rdp
)) {
1213 spin_lock_irqsave(&rcu_get_root(rsp
)->lock
, flags
);
1214 rcu_start_gp(rsp
, flags
); /* releases above lock */
1217 /* If there are callbacks ready, invoke them. */
1222 * Do softirq processing for the current CPU.
1224 static void rcu_process_callbacks(struct softirq_action
*unused
)
1227 * Memory references from any prior RCU read-side critical sections
1228 * executed by the interrupted code must be seen before any RCU
1229 * grace-period manipulations below.
1231 smp_mb(); /* See above block comment. */
1233 __rcu_process_callbacks(&rcu_sched_state
,
1234 &__get_cpu_var(rcu_sched_data
));
1235 __rcu_process_callbacks(&rcu_bh_state
, &__get_cpu_var(rcu_bh_data
));
1236 rcu_preempt_process_callbacks();
1239 * Memory references from any later RCU read-side critical sections
1240 * executed by the interrupted code must be seen after any RCU
1241 * grace-period manipulations above.
1243 smp_mb(); /* See above block comment. */
1247 __call_rcu(struct rcu_head
*head
, void (*func
)(struct rcu_head
*rcu
),
1248 struct rcu_state
*rsp
)
1250 unsigned long flags
;
1251 struct rcu_data
*rdp
;
1256 smp_mb(); /* Ensure RCU update seen before callback registry. */
1259 * Opportunistically note grace-period endings and beginnings.
1260 * Note that we might see a beginning right after we see an
1261 * end, but never vice versa, since this CPU has to pass through
1262 * a quiescent state betweentimes.
1264 local_irq_save(flags
);
1265 rdp
= rsp
->rda
[smp_processor_id()];
1266 rcu_process_gp_end(rsp
, rdp
);
1267 check_for_new_grace_period(rsp
, rdp
);
1269 /* Add the callback to our list. */
1270 *rdp
->nxttail
[RCU_NEXT_TAIL
] = head
;
1271 rdp
->nxttail
[RCU_NEXT_TAIL
] = &head
->next
;
1273 /* Start a new grace period if one not already started. */
1274 if (ACCESS_ONCE(rsp
->completed
) == ACCESS_ONCE(rsp
->gpnum
)) {
1275 unsigned long nestflag
;
1276 struct rcu_node
*rnp_root
= rcu_get_root(rsp
);
1278 spin_lock_irqsave(&rnp_root
->lock
, nestflag
);
1279 rcu_start_gp(rsp
, nestflag
); /* releases rnp_root->lock. */
1282 /* Force the grace period if too many callbacks or too long waiting. */
1283 if (unlikely(++rdp
->qlen
> qhimark
)) {
1284 rdp
->blimit
= LONG_MAX
;
1285 force_quiescent_state(rsp
, 0);
1286 } else if ((long)(ACCESS_ONCE(rsp
->jiffies_force_qs
) - jiffies
) < 0)
1287 force_quiescent_state(rsp
, 1);
1288 local_irq_restore(flags
);
1292 * Queue an RCU-sched callback for invocation after a grace period.
1294 void call_rcu_sched(struct rcu_head
*head
, void (*func
)(struct rcu_head
*rcu
))
1296 __call_rcu(head
, func
, &rcu_sched_state
);
1298 EXPORT_SYMBOL_GPL(call_rcu_sched
);
1301 * Queue an RCU for invocation after a quicker grace period.
1303 void call_rcu_bh(struct rcu_head
*head
, void (*func
)(struct rcu_head
*rcu
))
1305 __call_rcu(head
, func
, &rcu_bh_state
);
1307 EXPORT_SYMBOL_GPL(call_rcu_bh
);
1310 * Check to see if there is any immediate RCU-related work to be done
1311 * by the current CPU, for the specified type of RCU, returning 1 if so.
1312 * The checks are in order of increasing expense: checks that can be
1313 * carried out against CPU-local state are performed first. However,
1314 * we must check for CPU stalls first, else we might not get a chance.
1316 static int __rcu_pending(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
1318 rdp
->n_rcu_pending
++;
1320 /* Check for CPU stalls, if enabled. */
1321 check_cpu_stall(rsp
, rdp
);
1323 /* Is the RCU core waiting for a quiescent state from this CPU? */
1324 if (rdp
->qs_pending
) {
1325 rdp
->n_rp_qs_pending
++;
1329 /* Does this CPU have callbacks ready to invoke? */
1330 if (cpu_has_callbacks_ready_to_invoke(rdp
)) {
1331 rdp
->n_rp_cb_ready
++;
1335 /* Has RCU gone idle with this CPU needing another grace period? */
1336 if (cpu_needs_another_gp(rsp
, rdp
)) {
1337 rdp
->n_rp_cpu_needs_gp
++;
1341 /* Has another RCU grace period completed? */
1342 if (ACCESS_ONCE(rsp
->completed
) != rdp
->completed
) { /* outside lock */
1343 rdp
->n_rp_gp_completed
++;
1347 /* Has a new RCU grace period started? */
1348 if (ACCESS_ONCE(rsp
->gpnum
) != rdp
->gpnum
) { /* outside lock */
1349 rdp
->n_rp_gp_started
++;
1353 /* Has an RCU GP gone long enough to send resched IPIs &c? */
1354 if (ACCESS_ONCE(rsp
->completed
) != ACCESS_ONCE(rsp
->gpnum
) &&
1355 ((long)(ACCESS_ONCE(rsp
->jiffies_force_qs
) - jiffies
) < 0)) {
1356 rdp
->n_rp_need_fqs
++;
1361 rdp
->n_rp_need_nothing
++;
1366 * Check to see if there is any immediate RCU-related work to be done
1367 * by the current CPU, returning 1 if so. This function is part of the
1368 * RCU implementation; it is -not- an exported member of the RCU API.
1370 static int rcu_pending(int cpu
)
1372 return __rcu_pending(&rcu_sched_state
, &per_cpu(rcu_sched_data
, cpu
)) ||
1373 __rcu_pending(&rcu_bh_state
, &per_cpu(rcu_bh_data
, cpu
)) ||
1374 rcu_preempt_pending(cpu
);
1378 * Check to see if any future RCU-related work will need to be done
1379 * by the current CPU, even if none need be done immediately, returning
1380 * 1 if so. This function is part of the RCU implementation; it is -not-
1381 * an exported member of the RCU API.
1383 int rcu_needs_cpu(int cpu
)
1385 /* RCU callbacks either ready or pending? */
1386 return per_cpu(rcu_sched_data
, cpu
).nxtlist
||
1387 per_cpu(rcu_bh_data
, cpu
).nxtlist
||
1388 rcu_preempt_needs_cpu(cpu
);
1392 * Do boot-time initialization of a CPU's per-CPU RCU data.
1395 rcu_boot_init_percpu_data(int cpu
, struct rcu_state
*rsp
)
1397 unsigned long flags
;
1399 struct rcu_data
*rdp
= rsp
->rda
[cpu
];
1400 struct rcu_node
*rnp
= rcu_get_root(rsp
);
1402 /* Set up local state, ensuring consistent view of global state. */
1403 spin_lock_irqsave(&rnp
->lock
, flags
);
1404 rdp
->grpmask
= 1UL << (cpu
- rdp
->mynode
->grplo
);
1405 rdp
->nxtlist
= NULL
;
1406 for (i
= 0; i
< RCU_NEXT_SIZE
; i
++)
1407 rdp
->nxttail
[i
] = &rdp
->nxtlist
;
1410 rdp
->dynticks
= &per_cpu(rcu_dynticks
, cpu
);
1411 #endif /* #ifdef CONFIG_NO_HZ */
1413 spin_unlock_irqrestore(&rnp
->lock
, flags
);
1417 * Initialize a CPU's per-CPU RCU data. Note that only one online or
1418 * offline event can be happening at a given time. Note also that we
1419 * can accept some slop in the rsp->completed access due to the fact
1420 * that this CPU cannot possibly have any RCU callbacks in flight yet.
1422 static void __cpuinit
1423 rcu_init_percpu_data(int cpu
, struct rcu_state
*rsp
, int preemptable
)
1425 unsigned long flags
;
1428 struct rcu_data
*rdp
= rsp
->rda
[cpu
];
1429 struct rcu_node
*rnp
= rcu_get_root(rsp
);
1431 /* Set up local state, ensuring consistent view of global state. */
1432 spin_lock_irqsave(&rnp
->lock
, flags
);
1433 lastcomp
= rsp
->completed
;
1434 rdp
->completed
= lastcomp
;
1435 rdp
->gpnum
= lastcomp
;
1436 rdp
->passed_quiesc
= 0; /* We could be racing with new GP, */
1437 rdp
->qs_pending
= 1; /* so set up to respond to current GP. */
1438 rdp
->beenonline
= 1; /* We have now been online. */
1439 rdp
->preemptable
= preemptable
;
1440 rdp
->passed_quiesc_completed
= lastcomp
- 1;
1441 rdp
->blimit
= blimit
;
1442 spin_unlock(&rnp
->lock
); /* irqs remain disabled. */
1445 * A new grace period might start here. If so, we won't be part
1446 * of it, but that is OK, as we are currently in a quiescent state.
1449 /* Exclude any attempts to start a new GP on large systems. */
1450 spin_lock(&rsp
->onofflock
); /* irqs already disabled. */
1452 /* Add CPU to rcu_node bitmasks. */
1454 mask
= rdp
->grpmask
;
1456 /* Exclude any attempts to start a new GP on small systems. */
1457 spin_lock(&rnp
->lock
); /* irqs already disabled. */
1458 rnp
->qsmaskinit
|= mask
;
1459 mask
= rnp
->grpmask
;
1460 spin_unlock(&rnp
->lock
); /* irqs already disabled. */
1462 } while (rnp
!= NULL
&& !(rnp
->qsmaskinit
& mask
));
1464 spin_unlock(&rsp
->onofflock
); /* irqs remain disabled. */
1467 * A new grace period might start here. If so, we will be part of
1468 * it, and its gpnum will be greater than ours, so we will
1469 * participate. It is also possible for the gpnum to have been
1470 * incremented before this function was called, and the bitmasks
1471 * to not be filled out until now, in which case we will also
1472 * participate due to our gpnum being behind.
1475 /* Since it is coming online, the CPU is in a quiescent state. */
1476 cpu_quiet(cpu
, rsp
, rdp
, lastcomp
);
1477 local_irq_restore(flags
);
1480 static void __cpuinit
rcu_online_cpu(int cpu
)
1482 rcu_init_percpu_data(cpu
, &rcu_sched_state
, 0);
1483 rcu_init_percpu_data(cpu
, &rcu_bh_state
, 0);
1484 rcu_preempt_init_percpu_data(cpu
);
1488 * Handle CPU online/offline notification events.
1490 int __cpuinit
rcu_cpu_notify(struct notifier_block
*self
,
1491 unsigned long action
, void *hcpu
)
1493 long cpu
= (long)hcpu
;
1496 case CPU_UP_PREPARE
:
1497 case CPU_UP_PREPARE_FROZEN
:
1498 rcu_online_cpu(cpu
);
1501 case CPU_DEAD_FROZEN
:
1502 case CPU_UP_CANCELED
:
1503 case CPU_UP_CANCELED_FROZEN
:
1504 rcu_offline_cpu(cpu
);
1513 * Compute the per-level fanout, either using the exact fanout specified
1514 * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT.
1516 #ifdef CONFIG_RCU_FANOUT_EXACT
1517 static void __init
rcu_init_levelspread(struct rcu_state
*rsp
)
1521 for (i
= NUM_RCU_LVLS
- 1; i
>= 0; i
--)
1522 rsp
->levelspread
[i
] = CONFIG_RCU_FANOUT
;
1524 #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */
1525 static void __init
rcu_init_levelspread(struct rcu_state
*rsp
)
1532 for (i
= NUM_RCU_LVLS
- 1; i
>= 0; i
--) {
1533 ccur
= rsp
->levelcnt
[i
];
1534 rsp
->levelspread
[i
] = (cprv
+ ccur
- 1) / ccur
;
1538 #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */
1541 * Helper function for rcu_init() that initializes one rcu_state structure.
1543 static void __init
rcu_init_one(struct rcu_state
*rsp
)
1548 struct rcu_node
*rnp
;
1550 /* Initialize the level-tracking arrays. */
1552 for (i
= 1; i
< NUM_RCU_LVLS
; i
++)
1553 rsp
->level
[i
] = rsp
->level
[i
- 1] + rsp
->levelcnt
[i
- 1];
1554 rcu_init_levelspread(rsp
);
1556 /* Initialize the elements themselves, starting from the leaves. */
1558 for (i
= NUM_RCU_LVLS
- 1; i
>= 0; i
--) {
1559 cpustride
*= rsp
->levelspread
[i
];
1560 rnp
= rsp
->level
[i
];
1561 for (j
= 0; j
< rsp
->levelcnt
[i
]; j
++, rnp
++) {
1562 spin_lock_init(&rnp
->lock
);
1565 rnp
->qsmaskinit
= 0;
1566 rnp
->grplo
= j
* cpustride
;
1567 rnp
->grphi
= (j
+ 1) * cpustride
- 1;
1568 if (rnp
->grphi
>= NR_CPUS
)
1569 rnp
->grphi
= NR_CPUS
- 1;
1575 rnp
->grpnum
= j
% rsp
->levelspread
[i
- 1];
1576 rnp
->grpmask
= 1UL << rnp
->grpnum
;
1577 rnp
->parent
= rsp
->level
[i
- 1] +
1578 j
/ rsp
->levelspread
[i
- 1];
1581 INIT_LIST_HEAD(&rnp
->blocked_tasks
[0]);
1582 INIT_LIST_HEAD(&rnp
->blocked_tasks
[1]);
1588 * Helper macro for __rcu_init() and __rcu_init_preempt(). To be used
1589 * nowhere else! Assigns leaf node pointers into each CPU's rcu_data
1592 #define RCU_INIT_FLAVOR(rsp, rcu_data) \
1594 rcu_init_one(rsp); \
1595 rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \
1597 for_each_possible_cpu(i) { \
1598 if (i > rnp[j].grphi) \
1600 per_cpu(rcu_data, i).mynode = &rnp[j]; \
1601 (rsp)->rda[i] = &per_cpu(rcu_data, i); \
1602 rcu_boot_init_percpu_data(i, rsp); \
1606 #ifdef CONFIG_TREE_PREEMPT_RCU
1608 void __init
__rcu_init_preempt(void)
1610 int i
; /* All used by RCU_INIT_FLAVOR(). */
1612 struct rcu_node
*rnp
;
1614 RCU_INIT_FLAVOR(&rcu_preempt_state
, rcu_preempt_data
);
1617 #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
1619 void __init
__rcu_init_preempt(void)
1623 #endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */
1625 void __init
__rcu_init(void)
1627 int i
; /* All used by RCU_INIT_FLAVOR(). */
1629 struct rcu_node
*rnp
;
1631 rcu_bootup_announce();
1632 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
1633 printk(KERN_INFO
"RCU-based detection of stalled CPUs is enabled.\n");
1634 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
1635 RCU_INIT_FLAVOR(&rcu_sched_state
, rcu_sched_data
);
1636 RCU_INIT_FLAVOR(&rcu_bh_state
, rcu_bh_data
);
1637 __rcu_init_preempt();
1638 open_softirq(RCU_SOFTIRQ
, rcu_process_callbacks
);
1641 module_param(blimit
, int, 0);
1642 module_param(qhimark
, int, 0);
1643 module_param(qlowmark
, int, 0);